net/hyperv: Add support for jumbo frame up to 64KB
[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 #include <linux/if_ether.h>
32
33 #include "hyperv_net.h"
34
35
36 static struct netvsc_device *alloc_net_device(struct hv_device *device)
37 {
38         struct netvsc_device *net_device;
39         struct net_device *ndev = hv_get_drvdata(device);
40
41         net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
42         if (!net_device)
43                 return NULL;
44
45         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
54 static 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
65 static 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
78 get_in_err:
79         return net_device;
80 }
81
82
83 static 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                 free_pages((unsigned long)net_device->recv_buf,
140                         get_order(net_device->recv_buf_size));
141                 net_device->recv_buf = NULL;
142         }
143
144         if (net_device->recv_section) {
145                 net_device->recv_section_cnt = 0;
146                 kfree(net_device->recv_section);
147                 net_device->recv_section = NULL;
148         }
149
150         return ret;
151 }
152
153 static int netvsc_init_recv_buf(struct hv_device *device)
154 {
155         int ret = 0;
156         int t;
157         struct netvsc_device *net_device;
158         struct nvsp_message *init_packet;
159         struct net_device *ndev;
160
161         net_device = get_outbound_net_device(device);
162         if (!net_device)
163                 return -ENODEV;
164         ndev = net_device->ndev;
165
166         net_device->recv_buf =
167                 (void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO,
168                                 get_order(net_device->recv_buf_size));
169         if (!net_device->recv_buf) {
170                 netdev_err(ndev, "unable to allocate receive "
171                         "buffer of size %d\n", net_device->recv_buf_size);
172                 ret = -ENOMEM;
173                 goto cleanup;
174         }
175
176         /*
177          * Establish the gpadl handle for this buffer on this
178          * channel.  Note: This call uses the vmbus connection rather
179          * than the channel to establish the gpadl handle.
180          */
181         ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
182                                     net_device->recv_buf_size,
183                                     &net_device->recv_buf_gpadl_handle);
184         if (ret != 0) {
185                 netdev_err(ndev,
186                         "unable to establish receive buffer's gpadl\n");
187                 goto cleanup;
188         }
189
190
191         /* Notify the NetVsp of the gpadl handle */
192         init_packet = &net_device->channel_init_pkt;
193
194         memset(init_packet, 0, sizeof(struct nvsp_message));
195
196         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
197         init_packet->msg.v1_msg.send_recv_buf.
198                 gpadl_handle = net_device->recv_buf_gpadl_handle;
199         init_packet->msg.v1_msg.
200                 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
201
202         /* Send the gpadl notification request */
203         ret = vmbus_sendpacket(device->channel, init_packet,
204                                sizeof(struct nvsp_message),
205                                (unsigned long)init_packet,
206                                VM_PKT_DATA_INBAND,
207                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
208         if (ret != 0) {
209                 netdev_err(ndev,
210                         "unable to send receive buffer's gpadl to netvsp\n");
211                 goto cleanup;
212         }
213
214         t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
215         BUG_ON(t == 0);
216
217
218         /* Check the response */
219         if (init_packet->msg.v1_msg.
220             send_recv_buf_complete.status != NVSP_STAT_SUCCESS) {
221                 netdev_err(ndev, "Unable to complete receive buffer "
222                            "initialization with NetVsp - status %d\n",
223                            init_packet->msg.v1_msg.
224                            send_recv_buf_complete.status);
225                 ret = -EINVAL;
226                 goto cleanup;
227         }
228
229         /* Parse the response */
230
231         net_device->recv_section_cnt = init_packet->msg.
232                 v1_msg.send_recv_buf_complete.num_sections;
233
234         net_device->recv_section = kmemdup(
235                 init_packet->msg.v1_msg.send_recv_buf_complete.sections,
236                 net_device->recv_section_cnt *
237                 sizeof(struct nvsp_1_receive_buffer_section),
238                 GFP_KERNEL);
239         if (net_device->recv_section == NULL) {
240                 ret = -EINVAL;
241                 goto cleanup;
242         }
243
244         /*
245          * For 1st release, there should only be 1 section that represents the
246          * entire receive buffer
247          */
248         if (net_device->recv_section_cnt != 1 ||
249             net_device->recv_section->offset != 0) {
250                 ret = -EINVAL;
251                 goto cleanup;
252         }
253
254         goto exit;
255
256 cleanup:
257         netvsc_destroy_recv_buf(net_device);
258
259 exit:
260         return ret;
261 }
262
263
264 /* Negotiate NVSP protocol version */
265 static int negotiate_nvsp_ver(struct hv_device *device,
266                               struct netvsc_device *net_device,
267                               struct nvsp_message *init_packet,
268                               u32 nvsp_ver)
269 {
270         int ret, t;
271
272         memset(init_packet, 0, sizeof(struct nvsp_message));
273         init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
274         init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
275         init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
276
277         /* Send the init request */
278         ret = vmbus_sendpacket(device->channel, init_packet,
279                                sizeof(struct nvsp_message),
280                                (unsigned long)init_packet,
281                                VM_PKT_DATA_INBAND,
282                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
283
284         if (ret != 0)
285                 return ret;
286
287         t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
288
289         if (t == 0)
290                 return -ETIMEDOUT;
291
292         if (init_packet->msg.init_msg.init_complete.status !=
293             NVSP_STAT_SUCCESS)
294                 return -EINVAL;
295
296         if (nvsp_ver != NVSP_PROTOCOL_VERSION_2)
297                 return 0;
298
299         /* NVSPv2 only: Send NDIS config */
300         memset(init_packet, 0, sizeof(struct nvsp_message));
301         init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
302         init_packet->msg.v2_msg.send_ndis_config.mtu = net_device->ndev->mtu;
303
304         ret = vmbus_sendpacket(device->channel, init_packet,
305                                 sizeof(struct nvsp_message),
306                                 (unsigned long)init_packet,
307                                 VM_PKT_DATA_INBAND, 0);
308
309         return ret;
310 }
311
312 static int netvsc_connect_vsp(struct hv_device *device)
313 {
314         int ret;
315         struct netvsc_device *net_device;
316         struct nvsp_message *init_packet;
317         int ndis_version;
318         struct net_device *ndev;
319
320         net_device = get_outbound_net_device(device);
321         if (!net_device)
322                 return -ENODEV;
323         ndev = net_device->ndev;
324
325         init_packet = &net_device->channel_init_pkt;
326
327         /* Negotiate the latest NVSP protocol supported */
328         if (negotiate_nvsp_ver(device, net_device, init_packet,
329                                NVSP_PROTOCOL_VERSION_2) == 0) {
330                 net_device->nvsp_version = NVSP_PROTOCOL_VERSION_2;
331         } else if (negotiate_nvsp_ver(device, net_device, init_packet,
332                                     NVSP_PROTOCOL_VERSION_1) == 0) {
333                 net_device->nvsp_version = NVSP_PROTOCOL_VERSION_1;
334         } else {
335                 ret = -EPROTO;
336                 goto cleanup;
337         }
338
339         pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
340
341         /* Send the ndis version */
342         memset(init_packet, 0, sizeof(struct nvsp_message));
343
344         ndis_version = 0x00050000;
345
346         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
347         init_packet->msg.v1_msg.
348                 send_ndis_ver.ndis_major_ver =
349                                 (ndis_version & 0xFFFF0000) >> 16;
350         init_packet->msg.v1_msg.
351                 send_ndis_ver.ndis_minor_ver =
352                                 ndis_version & 0xFFFF;
353
354         /* Send the init request */
355         ret = vmbus_sendpacket(device->channel, init_packet,
356                                 sizeof(struct nvsp_message),
357                                 (unsigned long)init_packet,
358                                 VM_PKT_DATA_INBAND, 0);
359         if (ret != 0)
360                 goto cleanup;
361
362         /* Post the big receive buffer to NetVSP */
363         ret = netvsc_init_recv_buf(device);
364
365 cleanup:
366         return ret;
367 }
368
369 static void netvsc_disconnect_vsp(struct netvsc_device *net_device)
370 {
371         netvsc_destroy_recv_buf(net_device);
372 }
373
374 /*
375  * netvsc_device_remove - Callback when the root bus device is removed
376  */
377 int netvsc_device_remove(struct hv_device *device)
378 {
379         struct netvsc_device *net_device;
380         struct hv_netvsc_packet *netvsc_packet, *pos;
381         unsigned long flags;
382
383         net_device = hv_get_drvdata(device);
384         spin_lock_irqsave(&device->channel->inbound_lock, flags);
385         net_device->destroy = true;
386         spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
387
388         /* Wait for all send completions */
389         while (atomic_read(&net_device->num_outstanding_sends)) {
390                 dev_info(&device->device,
391                         "waiting for %d requests to complete...\n",
392                         atomic_read(&net_device->num_outstanding_sends));
393                 udelay(100);
394         }
395
396         netvsc_disconnect_vsp(net_device);
397
398         /*
399          * Since we have already drained, we don't need to busy wait
400          * as was done in final_release_stor_device()
401          * Note that we cannot set the ext pointer to NULL until
402          * we have drained - to drain the outgoing packets, we need to
403          * allow incoming packets.
404          */
405
406         spin_lock_irqsave(&device->channel->inbound_lock, flags);
407         hv_set_drvdata(device, NULL);
408         spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
409
410         /*
411          * At this point, no one should be accessing net_device
412          * except in here
413          */
414         dev_notice(&device->device, "net device safe to remove\n");
415
416         /* Now, we can close the channel safely */
417         vmbus_close(device->channel);
418
419         /* Release all resources */
420         list_for_each_entry_safe(netvsc_packet, pos,
421                                  &net_device->recv_pkt_list, list_ent) {
422                 list_del(&netvsc_packet->list_ent);
423                 kfree(netvsc_packet);
424         }
425
426         kfree(net_device);
427         return 0;
428 }
429
430 static void netvsc_send_completion(struct hv_device *device,
431                                    struct vmpacket_descriptor *packet)
432 {
433         struct netvsc_device *net_device;
434         struct nvsp_message *nvsp_packet;
435         struct hv_netvsc_packet *nvsc_packet;
436         struct net_device *ndev;
437
438         net_device = get_inbound_net_device(device);
439         if (!net_device)
440                 return;
441         ndev = net_device->ndev;
442
443         nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
444                         (packet->offset8 << 3));
445
446         if ((nvsp_packet->hdr.msg_type == NVSP_MSG_TYPE_INIT_COMPLETE) ||
447             (nvsp_packet->hdr.msg_type ==
448              NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE) ||
449             (nvsp_packet->hdr.msg_type ==
450              NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE)) {
451                 /* Copy the response back */
452                 memcpy(&net_device->channel_init_pkt, nvsp_packet,
453                        sizeof(struct nvsp_message));
454                 complete(&net_device->channel_init_wait);
455         } else if (nvsp_packet->hdr.msg_type ==
456                    NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE) {
457                 /* Get the send context */
458                 nvsc_packet = (struct hv_netvsc_packet *)(unsigned long)
459                         packet->trans_id;
460
461                 /* Notify the layer above us */
462                 nvsc_packet->completion.send.send_completion(
463                         nvsc_packet->completion.send.send_completion_ctx);
464
465                 atomic_dec(&net_device->num_outstanding_sends);
466
467                 if (netif_queue_stopped(ndev) && !net_device->start_remove)
468                         netif_wake_queue(ndev);
469         } else {
470                 netdev_err(ndev, "Unknown send completion packet type- "
471                            "%d received!!\n", nvsp_packet->hdr.msg_type);
472         }
473
474 }
475
476 int netvsc_send(struct hv_device *device,
477                         struct hv_netvsc_packet *packet)
478 {
479         struct netvsc_device *net_device;
480         int ret = 0;
481         struct nvsp_message sendMessage;
482         struct net_device *ndev;
483
484         net_device = get_outbound_net_device(device);
485         if (!net_device)
486                 return -ENODEV;
487         ndev = net_device->ndev;
488
489         sendMessage.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
490         if (packet->is_data_pkt) {
491                 /* 0 is RMC_DATA; */
492                 sendMessage.msg.v1_msg.send_rndis_pkt.channel_type = 0;
493         } else {
494                 /* 1 is RMC_CONTROL; */
495                 sendMessage.msg.v1_msg.send_rndis_pkt.channel_type = 1;
496         }
497
498         /* Not using send buffer section */
499         sendMessage.msg.v1_msg.send_rndis_pkt.send_buf_section_index =
500                 0xFFFFFFFF;
501         sendMessage.msg.v1_msg.send_rndis_pkt.send_buf_section_size = 0;
502
503         if (packet->page_buf_cnt) {
504                 ret = vmbus_sendpacket_pagebuffer(device->channel,
505                                                   packet->page_buf,
506                                                   packet->page_buf_cnt,
507                                                   &sendMessage,
508                                                   sizeof(struct nvsp_message),
509                                                   (unsigned long)packet);
510         } else {
511                 ret = vmbus_sendpacket(device->channel, &sendMessage,
512                                 sizeof(struct nvsp_message),
513                                 (unsigned long)packet,
514                                 VM_PKT_DATA_INBAND,
515                                 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
516
517         }
518
519         if (ret == 0) {
520                 atomic_inc(&net_device->num_outstanding_sends);
521         } else if (ret == -EAGAIN) {
522                 netif_stop_queue(ndev);
523                 if (atomic_read(&net_device->num_outstanding_sends) < 1)
524                         netif_wake_queue(ndev);
525         } else {
526                 netdev_err(ndev, "Unable to send packet %p ret %d\n",
527                            packet, ret);
528         }
529
530         return ret;
531 }
532
533 static void netvsc_send_recv_completion(struct hv_device *device,
534                                         u64 transaction_id)
535 {
536         struct nvsp_message recvcompMessage;
537         int retries = 0;
538         int ret;
539         struct net_device *ndev;
540         struct netvsc_device *net_device = hv_get_drvdata(device);
541
542         ndev = net_device->ndev;
543
544         recvcompMessage.hdr.msg_type =
545                                 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE;
546
547         /* FIXME: Pass in the status */
548         recvcompMessage.msg.v1_msg.send_rndis_pkt_complete.status =
549                 NVSP_STAT_SUCCESS;
550
551 retry_send_cmplt:
552         /* Send the completion */
553         ret = vmbus_sendpacket(device->channel, &recvcompMessage,
554                                sizeof(struct nvsp_message), transaction_id,
555                                VM_PKT_COMP, 0);
556         if (ret == 0) {
557                 /* success */
558                 /* no-op */
559         } else if (ret == -EAGAIN) {
560                 /* no more room...wait a bit and attempt to retry 3 times */
561                 retries++;
562                 netdev_err(ndev, "unable to send receive completion pkt"
563                         " (tid %llx)...retrying %d\n", transaction_id, retries);
564
565                 if (retries < 4) {
566                         udelay(100);
567                         goto retry_send_cmplt;
568                 } else {
569                         netdev_err(ndev, "unable to send receive "
570                                 "completion pkt (tid %llx)...give up retrying\n",
571                                 transaction_id);
572                 }
573         } else {
574                 netdev_err(ndev, "unable to send receive "
575                         "completion pkt - %llx\n", transaction_id);
576         }
577 }
578
579 /* Send a receive completion packet to RNDIS device (ie NetVsp) */
580 static void netvsc_receive_completion(void *context)
581 {
582         struct hv_netvsc_packet *packet = context;
583         struct hv_device *device = (struct hv_device *)packet->device;
584         struct netvsc_device *net_device;
585         u64 transaction_id = 0;
586         bool fsend_receive_comp = false;
587         unsigned long flags;
588         struct net_device *ndev;
589
590         /*
591          * Even though it seems logical to do a GetOutboundNetDevice() here to
592          * send out receive completion, we are using GetInboundNetDevice()
593          * since we may have disable outbound traffic already.
594          */
595         net_device = get_inbound_net_device(device);
596         if (!net_device)
597                 return;
598         ndev = net_device->ndev;
599
600         /* Overloading use of the lock. */
601         spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
602
603         packet->xfer_page_pkt->count--;
604
605         /*
606          * Last one in the line that represent 1 xfer page packet.
607          * Return the xfer page packet itself to the freelist
608          */
609         if (packet->xfer_page_pkt->count == 0) {
610                 fsend_receive_comp = true;
611                 transaction_id = packet->completion.recv.recv_completion_tid;
612                 list_add_tail(&packet->xfer_page_pkt->list_ent,
613                               &net_device->recv_pkt_list);
614
615         }
616
617         /* Put the packet back */
618         list_add_tail(&packet->list_ent, &net_device->recv_pkt_list);
619         spin_unlock_irqrestore(&net_device->recv_pkt_list_lock, flags);
620
621         /* Send a receive completion for the xfer page packet */
622         if (fsend_receive_comp)
623                 netvsc_send_recv_completion(device, transaction_id);
624
625 }
626
627 static void netvsc_receive(struct hv_device *device,
628                             struct vmpacket_descriptor *packet)
629 {
630         struct netvsc_device *net_device;
631         struct vmtransfer_page_packet_header *vmxferpage_packet;
632         struct nvsp_message *nvsp_packet;
633         struct hv_netvsc_packet *netvsc_packet = NULL;
634         /* struct netvsc_driver *netvscDriver; */
635         struct xferpage_packet *xferpage_packet = NULL;
636         int i;
637         int count = 0;
638         unsigned long flags;
639         struct net_device *ndev;
640
641         LIST_HEAD(listHead);
642
643         net_device = get_inbound_net_device(device);
644         if (!net_device)
645                 return;
646         ndev = net_device->ndev;
647
648         /*
649          * All inbound packets other than send completion should be xfer page
650          * packet
651          */
652         if (packet->type != VM_PKT_DATA_USING_XFER_PAGES) {
653                 netdev_err(ndev, "Unknown packet type received - %d\n",
654                            packet->type);
655                 return;
656         }
657
658         nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
659                         (packet->offset8 << 3));
660
661         /* Make sure this is a valid nvsp packet */
662         if (nvsp_packet->hdr.msg_type !=
663             NVSP_MSG1_TYPE_SEND_RNDIS_PKT) {
664                 netdev_err(ndev, "Unknown nvsp packet type received-"
665                         " %d\n", nvsp_packet->hdr.msg_type);
666                 return;
667         }
668
669         vmxferpage_packet = (struct vmtransfer_page_packet_header *)packet;
670
671         if (vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID) {
672                 netdev_err(ndev, "Invalid xfer page set id - "
673                            "expecting %x got %x\n", NETVSC_RECEIVE_BUFFER_ID,
674                            vmxferpage_packet->xfer_pageset_id);
675                 return;
676         }
677
678         /*
679          * Grab free packets (range count + 1) to represent this xfer
680          * page packet. +1 to represent the xfer page packet itself.
681          * We grab it here so that we know exactly how many we can
682          * fulfil
683          */
684         spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
685         while (!list_empty(&net_device->recv_pkt_list)) {
686                 list_move_tail(net_device->recv_pkt_list.next, &listHead);
687                 if (++count == vmxferpage_packet->range_cnt + 1)
688                         break;
689         }
690         spin_unlock_irqrestore(&net_device->recv_pkt_list_lock, flags);
691
692         /*
693          * We need at least 2 netvsc pkts (1 to represent the xfer
694          * page and at least 1 for the range) i.e. we can handled
695          * some of the xfer page packet ranges...
696          */
697         if (count < 2) {
698                 netdev_err(ndev, "Got only %d netvsc pkt...needed "
699                         "%d pkts. Dropping this xfer page packet completely!\n",
700                         count, vmxferpage_packet->range_cnt + 1);
701
702                 /* Return it to the freelist */
703                 spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
704                 for (i = count; i != 0; i--) {
705                         list_move_tail(listHead.next,
706                                        &net_device->recv_pkt_list);
707                 }
708                 spin_unlock_irqrestore(&net_device->recv_pkt_list_lock,
709                                        flags);
710
711                 netvsc_send_recv_completion(device,
712                                             vmxferpage_packet->d.trans_id);
713
714                 return;
715         }
716
717         /* Remove the 1st packet to represent the xfer page packet itself */
718         xferpage_packet = (struct xferpage_packet *)listHead.next;
719         list_del(&xferpage_packet->list_ent);
720
721         /* This is how much we can satisfy */
722         xferpage_packet->count = count - 1;
723
724         if (xferpage_packet->count != vmxferpage_packet->range_cnt) {
725                 netdev_err(ndev, "Needed %d netvsc pkts to satisfy "
726                         "this xfer page...got %d\n",
727                         vmxferpage_packet->range_cnt, xferpage_packet->count);
728         }
729
730         /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
731         for (i = 0; i < (count - 1); i++) {
732                 netvsc_packet = (struct hv_netvsc_packet *)listHead.next;
733                 list_del(&netvsc_packet->list_ent);
734
735                 /* Initialize the netvsc packet */
736                 netvsc_packet->xfer_page_pkt = xferpage_packet;
737                 netvsc_packet->completion.recv.recv_completion =
738                                         netvsc_receive_completion;
739                 netvsc_packet->completion.recv.recv_completion_ctx =
740                                         netvsc_packet;
741                 netvsc_packet->device = device;
742                 /* Save this so that we can send it back */
743                 netvsc_packet->completion.recv.recv_completion_tid =
744                                         vmxferpage_packet->d.trans_id;
745
746                 netvsc_packet->data = (void *)((unsigned long)net_device->
747                         recv_buf + vmxferpage_packet->ranges[i].byte_offset);
748                 netvsc_packet->total_data_buflen =
749                                         vmxferpage_packet->ranges[i].byte_count;
750
751                 /* Pass it to the upper layer */
752                 rndis_filter_receive(device, netvsc_packet);
753
754                 netvsc_receive_completion(netvsc_packet->
755                                 completion.recv.recv_completion_ctx);
756         }
757
758 }
759
760 static void netvsc_channel_cb(void *context)
761 {
762         int ret;
763         struct hv_device *device = context;
764         struct netvsc_device *net_device;
765         u32 bytes_recvd;
766         u64 request_id;
767         unsigned char *packet;
768         struct vmpacket_descriptor *desc;
769         unsigned char *buffer;
770         int bufferlen = NETVSC_PACKET_SIZE;
771         struct net_device *ndev;
772
773         packet = kzalloc(NETVSC_PACKET_SIZE * sizeof(unsigned char),
774                          GFP_ATOMIC);
775         if (!packet)
776                 return;
777         buffer = packet;
778
779         net_device = get_inbound_net_device(device);
780         if (!net_device)
781                 goto out;
782         ndev = net_device->ndev;
783
784         do {
785                 ret = vmbus_recvpacket_raw(device->channel, buffer, bufferlen,
786                                            &bytes_recvd, &request_id);
787                 if (ret == 0) {
788                         if (bytes_recvd > 0) {
789                                 desc = (struct vmpacket_descriptor *)buffer;
790                                 switch (desc->type) {
791                                 case VM_PKT_COMP:
792                                         netvsc_send_completion(device, desc);
793                                         break;
794
795                                 case VM_PKT_DATA_USING_XFER_PAGES:
796                                         netvsc_receive(device, desc);
797                                         break;
798
799                                 default:
800                                         netdev_err(ndev,
801                                                    "unhandled packet type %d, "
802                                                    "tid %llx len %d\n",
803                                                    desc->type, request_id,
804                                                    bytes_recvd);
805                                         break;
806                                 }
807
808                                 /* reset */
809                                 if (bufferlen > NETVSC_PACKET_SIZE) {
810                                         kfree(buffer);
811                                         buffer = packet;
812                                         bufferlen = NETVSC_PACKET_SIZE;
813                                 }
814                         } else {
815                                 /* reset */
816                                 if (bufferlen > NETVSC_PACKET_SIZE) {
817                                         kfree(buffer);
818                                         buffer = packet;
819                                         bufferlen = NETVSC_PACKET_SIZE;
820                                 }
821
822                                 break;
823                         }
824                 } else if (ret == -ENOBUFS) {
825                         /* Handle large packet */
826                         buffer = kmalloc(bytes_recvd, GFP_ATOMIC);
827                         if (buffer == NULL) {
828                                 /* Try again next time around */
829                                 netdev_err(ndev,
830                                            "unable to allocate buffer of size "
831                                            "(%d)!!\n", bytes_recvd);
832                                 break;
833                         }
834
835                         bufferlen = bytes_recvd;
836                 }
837         } while (1);
838
839 out:
840         kfree(buffer);
841         return;
842 }
843
844 /*
845  * netvsc_device_add - Callback when the device belonging to this
846  * driver is added
847  */
848 int netvsc_device_add(struct hv_device *device, void *additional_info)
849 {
850         int ret = 0;
851         int i;
852         int ring_size =
853         ((struct netvsc_device_info *)additional_info)->ring_size;
854         struct netvsc_device *net_device;
855         struct hv_netvsc_packet *packet, *pos;
856         struct net_device *ndev;
857
858         net_device = alloc_net_device(device);
859         if (!net_device) {
860                 ret = -ENOMEM;
861                 goto cleanup;
862         }
863
864         /*
865          * Coming into this function, struct net_device * is
866          * registered as the driver private data.
867          * In alloc_net_device(), we register struct netvsc_device *
868          * as the driver private data and stash away struct net_device *
869          * in struct netvsc_device *.
870          */
871         ndev = net_device->ndev;
872
873         /* Initialize the NetVSC channel extension */
874         net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
875         spin_lock_init(&net_device->recv_pkt_list_lock);
876
877         INIT_LIST_HEAD(&net_device->recv_pkt_list);
878
879         for (i = 0; i < NETVSC_RECEIVE_PACKETLIST_COUNT; i++) {
880                 packet = kzalloc(sizeof(struct hv_netvsc_packet) +
881                                  (NETVSC_RECEIVE_SG_COUNT *
882                                   sizeof(struct hv_page_buffer)), GFP_KERNEL);
883                 if (!packet)
884                         break;
885
886                 list_add_tail(&packet->list_ent,
887                               &net_device->recv_pkt_list);
888         }
889         init_completion(&net_device->channel_init_wait);
890
891         /* Open the channel */
892         ret = vmbus_open(device->channel, ring_size * PAGE_SIZE,
893                          ring_size * PAGE_SIZE, NULL, 0,
894                          netvsc_channel_cb, device);
895
896         if (ret != 0) {
897                 netdev_err(ndev, "unable to open channel: %d\n", ret);
898                 goto cleanup;
899         }
900
901         /* Channel is opened */
902         pr_info("hv_netvsc channel opened successfully\n");
903
904         /* Connect with the NetVsp */
905         ret = netvsc_connect_vsp(device);
906         if (ret != 0) {
907                 netdev_err(ndev,
908                         "unable to connect to NetVSP - %d\n", ret);
909                 goto close;
910         }
911
912         return ret;
913
914 close:
915         /* Now, we can close the channel safely */
916         vmbus_close(device->channel);
917
918 cleanup:
919
920         if (net_device) {
921                 list_for_each_entry_safe(packet, pos,
922                                          &net_device->recv_pkt_list,
923                                          list_ent) {
924                         list_del(&packet->list_ent);
925                         kfree(packet);
926                 }
927
928                 kfree(net_device);
929         }
930
931         return ret;
932 }