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