These changes are the raw update to linux-4.4.6-rt14. Kernel sources
[kvmfornfv.git] / kernel / 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         int i;
42
43         net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
44         if (!net_device)
45                 return NULL;
46
47         net_device->cb_buffer = kzalloc(NETVSC_PACKET_SIZE, GFP_KERNEL);
48         if (!net_device->cb_buffer) {
49                 kfree(net_device);
50                 return NULL;
51         }
52
53         init_waitqueue_head(&net_device->wait_drain);
54         net_device->start_remove = false;
55         net_device->destroy = false;
56         net_device->dev = device;
57         net_device->ndev = ndev;
58         net_device->max_pkt = RNDIS_MAX_PKT_DEFAULT;
59         net_device->pkt_align = RNDIS_PKT_ALIGN_DEFAULT;
60
61         for (i = 0; i < num_online_cpus(); i++)
62                 spin_lock_init(&net_device->msd[i].lock);
63
64         hv_set_drvdata(device, net_device);
65         return net_device;
66 }
67
68 static void free_netvsc_device(struct netvsc_device *nvdev)
69 {
70         kfree(nvdev->cb_buffer);
71         kfree(nvdev);
72 }
73
74 static struct netvsc_device *get_outbound_net_device(struct hv_device *device)
75 {
76         struct netvsc_device *net_device;
77
78         net_device = hv_get_drvdata(device);
79         if (net_device && net_device->destroy)
80                 net_device = NULL;
81
82         return net_device;
83 }
84
85 static struct netvsc_device *get_inbound_net_device(struct hv_device *device)
86 {
87         struct netvsc_device *net_device;
88
89         net_device = hv_get_drvdata(device);
90
91         if (!net_device)
92                 goto get_in_err;
93
94         if (net_device->destroy &&
95                 atomic_read(&net_device->num_outstanding_sends) == 0)
96                 net_device = NULL;
97
98 get_in_err:
99         return net_device;
100 }
101
102
103 static int netvsc_destroy_buf(struct netvsc_device *net_device)
104 {
105         struct nvsp_message *revoke_packet;
106         int ret = 0;
107         struct net_device *ndev = net_device->ndev;
108
109         /*
110          * If we got a section count, it means we received a
111          * SendReceiveBufferComplete msg (ie sent
112          * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
113          * to send a revoke msg here
114          */
115         if (net_device->recv_section_cnt) {
116                 /* Send the revoke receive buffer */
117                 revoke_packet = &net_device->revoke_packet;
118                 memset(revoke_packet, 0, sizeof(struct nvsp_message));
119
120                 revoke_packet->hdr.msg_type =
121                         NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
122                 revoke_packet->msg.v1_msg.
123                 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
124
125                 ret = vmbus_sendpacket(net_device->dev->channel,
126                                        revoke_packet,
127                                        sizeof(struct nvsp_message),
128                                        (unsigned long)revoke_packet,
129                                        VM_PKT_DATA_INBAND, 0);
130                 /*
131                  * If we failed here, we might as well return and
132                  * have a leak rather than continue and a bugchk
133                  */
134                 if (ret != 0) {
135                         netdev_err(ndev, "unable to send "
136                                 "revoke receive buffer to netvsp\n");
137                         return ret;
138                 }
139         }
140
141         /* Teardown the gpadl on the vsp end */
142         if (net_device->recv_buf_gpadl_handle) {
143                 ret = vmbus_teardown_gpadl(net_device->dev->channel,
144                            net_device->recv_buf_gpadl_handle);
145
146                 /* If we failed here, we might as well return and have a leak
147                  * rather than continue and a bugchk
148                  */
149                 if (ret != 0) {
150                         netdev_err(ndev,
151                                    "unable to teardown receive buffer's gpadl\n");
152                         return ret;
153                 }
154                 net_device->recv_buf_gpadl_handle = 0;
155         }
156
157         if (net_device->recv_buf) {
158                 /* Free up the receive buffer */
159                 vfree(net_device->recv_buf);
160                 net_device->recv_buf = NULL;
161         }
162
163         if (net_device->recv_section) {
164                 net_device->recv_section_cnt = 0;
165                 kfree(net_device->recv_section);
166                 net_device->recv_section = NULL;
167         }
168
169         /* Deal with the send buffer we may have setup.
170          * If we got a  send section size, it means we received a
171          * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
172          * NVSP_MSG1_TYPE_SEND_SEND_BUF msg) therefore, we need
173          * to send a revoke msg here
174          */
175         if (net_device->send_section_size) {
176                 /* Send the revoke receive buffer */
177                 revoke_packet = &net_device->revoke_packet;
178                 memset(revoke_packet, 0, sizeof(struct nvsp_message));
179
180                 revoke_packet->hdr.msg_type =
181                         NVSP_MSG1_TYPE_REVOKE_SEND_BUF;
182                 revoke_packet->msg.v1_msg.revoke_send_buf.id =
183                         NETVSC_SEND_BUFFER_ID;
184
185                 ret = vmbus_sendpacket(net_device->dev->channel,
186                                        revoke_packet,
187                                        sizeof(struct nvsp_message),
188                                        (unsigned long)revoke_packet,
189                                        VM_PKT_DATA_INBAND, 0);
190                 /* If we failed here, we might as well return and
191                  * have a leak rather than continue and a bugchk
192                  */
193                 if (ret != 0) {
194                         netdev_err(ndev, "unable to send "
195                                    "revoke send buffer to netvsp\n");
196                         return ret;
197                 }
198         }
199         /* Teardown the gpadl on the vsp end */
200         if (net_device->send_buf_gpadl_handle) {
201                 ret = vmbus_teardown_gpadl(net_device->dev->channel,
202                                            net_device->send_buf_gpadl_handle);
203
204                 /* If we failed here, we might as well return and have a leak
205                  * rather than continue and a bugchk
206                  */
207                 if (ret != 0) {
208                         netdev_err(ndev,
209                                    "unable to teardown send buffer's gpadl\n");
210                         return ret;
211                 }
212                 net_device->send_buf_gpadl_handle = 0;
213         }
214         if (net_device->send_buf) {
215                 /* Free up the send buffer */
216                 vfree(net_device->send_buf);
217                 net_device->send_buf = NULL;
218         }
219         kfree(net_device->send_section_map);
220
221         return ret;
222 }
223
224 static int netvsc_init_buf(struct hv_device *device)
225 {
226         int ret = 0;
227         unsigned long t;
228         struct netvsc_device *net_device;
229         struct nvsp_message *init_packet;
230         struct net_device *ndev;
231         int node;
232
233         net_device = get_outbound_net_device(device);
234         if (!net_device)
235                 return -ENODEV;
236         ndev = net_device->ndev;
237
238         node = cpu_to_node(device->channel->target_cpu);
239         net_device->recv_buf = vzalloc_node(net_device->recv_buf_size, node);
240         if (!net_device->recv_buf)
241                 net_device->recv_buf = vzalloc(net_device->recv_buf_size);
242
243         if (!net_device->recv_buf) {
244                 netdev_err(ndev, "unable to allocate receive "
245                         "buffer of size %d\n", net_device->recv_buf_size);
246                 ret = -ENOMEM;
247                 goto cleanup;
248         }
249
250         /*
251          * Establish the gpadl handle for this buffer on this
252          * channel.  Note: This call uses the vmbus connection rather
253          * than the channel to establish the gpadl handle.
254          */
255         ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
256                                     net_device->recv_buf_size,
257                                     &net_device->recv_buf_gpadl_handle);
258         if (ret != 0) {
259                 netdev_err(ndev,
260                         "unable to establish receive buffer's gpadl\n");
261                 goto cleanup;
262         }
263
264
265         /* Notify the NetVsp of the gpadl handle */
266         init_packet = &net_device->channel_init_pkt;
267
268         memset(init_packet, 0, sizeof(struct nvsp_message));
269
270         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
271         init_packet->msg.v1_msg.send_recv_buf.
272                 gpadl_handle = net_device->recv_buf_gpadl_handle;
273         init_packet->msg.v1_msg.
274                 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
275
276         /* Send the gpadl notification request */
277         ret = vmbus_sendpacket(device->channel, init_packet,
278                                sizeof(struct nvsp_message),
279                                (unsigned long)init_packet,
280                                VM_PKT_DATA_INBAND,
281                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
282         if (ret != 0) {
283                 netdev_err(ndev,
284                         "unable to send receive buffer's gpadl to netvsp\n");
285                 goto cleanup;
286         }
287
288         t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
289         BUG_ON(t == 0);
290
291
292         /* Check the response */
293         if (init_packet->msg.v1_msg.
294             send_recv_buf_complete.status != NVSP_STAT_SUCCESS) {
295                 netdev_err(ndev, "Unable to complete receive buffer "
296                            "initialization with NetVsp - status %d\n",
297                            init_packet->msg.v1_msg.
298                            send_recv_buf_complete.status);
299                 ret = -EINVAL;
300                 goto cleanup;
301         }
302
303         /* Parse the response */
304
305         net_device->recv_section_cnt = init_packet->msg.
306                 v1_msg.send_recv_buf_complete.num_sections;
307
308         net_device->recv_section = kmemdup(
309                 init_packet->msg.v1_msg.send_recv_buf_complete.sections,
310                 net_device->recv_section_cnt *
311                 sizeof(struct nvsp_1_receive_buffer_section),
312                 GFP_KERNEL);
313         if (net_device->recv_section == NULL) {
314                 ret = -EINVAL;
315                 goto cleanup;
316         }
317
318         /*
319          * For 1st release, there should only be 1 section that represents the
320          * entire receive buffer
321          */
322         if (net_device->recv_section_cnt != 1 ||
323             net_device->recv_section->offset != 0) {
324                 ret = -EINVAL;
325                 goto cleanup;
326         }
327
328         /* Now setup the send buffer.
329          */
330         net_device->send_buf = vzalloc_node(net_device->send_buf_size, node);
331         if (!net_device->send_buf)
332                 net_device->send_buf = vzalloc(net_device->send_buf_size);
333         if (!net_device->send_buf) {
334                 netdev_err(ndev, "unable to allocate send "
335                            "buffer of size %d\n", net_device->send_buf_size);
336                 ret = -ENOMEM;
337                 goto cleanup;
338         }
339
340         /* Establish the gpadl handle for this buffer on this
341          * channel.  Note: This call uses the vmbus connection rather
342          * than the channel to establish the gpadl handle.
343          */
344         ret = vmbus_establish_gpadl(device->channel, net_device->send_buf,
345                                     net_device->send_buf_size,
346                                     &net_device->send_buf_gpadl_handle);
347         if (ret != 0) {
348                 netdev_err(ndev,
349                            "unable to establish send buffer's gpadl\n");
350                 goto cleanup;
351         }
352
353         /* Notify the NetVsp of the gpadl handle */
354         init_packet = &net_device->channel_init_pkt;
355         memset(init_packet, 0, sizeof(struct nvsp_message));
356         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_SEND_BUF;
357         init_packet->msg.v1_msg.send_send_buf.gpadl_handle =
358                 net_device->send_buf_gpadl_handle;
359         init_packet->msg.v1_msg.send_send_buf.id = NETVSC_SEND_BUFFER_ID;
360
361         /* Send the gpadl notification request */
362         ret = vmbus_sendpacket(device->channel, init_packet,
363                                sizeof(struct nvsp_message),
364                                (unsigned long)init_packet,
365                                VM_PKT_DATA_INBAND,
366                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
367         if (ret != 0) {
368                 netdev_err(ndev,
369                            "unable to send send buffer's gpadl to netvsp\n");
370                 goto cleanup;
371         }
372
373         t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
374         BUG_ON(t == 0);
375
376         /* Check the response */
377         if (init_packet->msg.v1_msg.
378             send_send_buf_complete.status != NVSP_STAT_SUCCESS) {
379                 netdev_err(ndev, "Unable to complete send buffer "
380                            "initialization with NetVsp - status %d\n",
381                            init_packet->msg.v1_msg.
382                            send_send_buf_complete.status);
383                 ret = -EINVAL;
384                 goto cleanup;
385         }
386
387         /* Parse the response */
388         net_device->send_section_size = init_packet->msg.
389                                 v1_msg.send_send_buf_complete.section_size;
390
391         /* Section count is simply the size divided by the section size.
392          */
393         net_device->send_section_cnt =
394                 net_device->send_buf_size/net_device->send_section_size;
395
396         dev_info(&device->device, "Send section size: %d, Section count:%d\n",
397                  net_device->send_section_size, net_device->send_section_cnt);
398
399         /* Setup state for managing the send buffer. */
400         net_device->map_words = DIV_ROUND_UP(net_device->send_section_cnt,
401                                              BITS_PER_LONG);
402
403         net_device->send_section_map =
404                 kzalloc(net_device->map_words * sizeof(ulong), GFP_KERNEL);
405         if (net_device->send_section_map == NULL) {
406                 ret = -ENOMEM;
407                 goto cleanup;
408         }
409
410         goto exit;
411
412 cleanup:
413         netvsc_destroy_buf(net_device);
414
415 exit:
416         return ret;
417 }
418
419
420 /* Negotiate NVSP protocol version */
421 static int negotiate_nvsp_ver(struct hv_device *device,
422                               struct netvsc_device *net_device,
423                               struct nvsp_message *init_packet,
424                               u32 nvsp_ver)
425 {
426         int ret;
427         unsigned long t;
428
429         memset(init_packet, 0, sizeof(struct nvsp_message));
430         init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
431         init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
432         init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
433
434         /* Send the init request */
435         ret = vmbus_sendpacket(device->channel, init_packet,
436                                sizeof(struct nvsp_message),
437                                (unsigned long)init_packet,
438                                VM_PKT_DATA_INBAND,
439                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
440
441         if (ret != 0)
442                 return ret;
443
444         t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
445
446         if (t == 0)
447                 return -ETIMEDOUT;
448
449         if (init_packet->msg.init_msg.init_complete.status !=
450             NVSP_STAT_SUCCESS)
451                 return -EINVAL;
452
453         if (nvsp_ver == NVSP_PROTOCOL_VERSION_1)
454                 return 0;
455
456         /* NVSPv2 or later: Send NDIS config */
457         memset(init_packet, 0, sizeof(struct nvsp_message));
458         init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
459         init_packet->msg.v2_msg.send_ndis_config.mtu = net_device->ndev->mtu +
460                                                        ETH_HLEN;
461         init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
462
463         if (nvsp_ver >= NVSP_PROTOCOL_VERSION_5)
464                 init_packet->msg.v2_msg.send_ndis_config.capability.sriov = 1;
465
466         ret = vmbus_sendpacket(device->channel, init_packet,
467                                 sizeof(struct nvsp_message),
468                                 (unsigned long)init_packet,
469                                 VM_PKT_DATA_INBAND, 0);
470
471         return ret;
472 }
473
474 static int netvsc_connect_vsp(struct hv_device *device)
475 {
476         int ret;
477         struct netvsc_device *net_device;
478         struct nvsp_message *init_packet;
479         int ndis_version;
480         struct net_device *ndev;
481         u32 ver_list[] = { NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
482                 NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5 };
483         int i, num_ver = 4; /* number of different NVSP versions */
484
485         net_device = get_outbound_net_device(device);
486         if (!net_device)
487                 return -ENODEV;
488         ndev = net_device->ndev;
489
490         init_packet = &net_device->channel_init_pkt;
491
492         /* Negotiate the latest NVSP protocol supported */
493         for (i = num_ver - 1; i >= 0; i--)
494                 if (negotiate_nvsp_ver(device, net_device, init_packet,
495                                        ver_list[i])  == 0) {
496                         net_device->nvsp_version = ver_list[i];
497                         break;
498                 }
499
500         if (i < 0) {
501                 ret = -EPROTO;
502                 goto cleanup;
503         }
504
505         pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
506
507         /* Send the ndis version */
508         memset(init_packet, 0, sizeof(struct nvsp_message));
509
510         if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
511                 ndis_version = 0x00060001;
512         else
513                 ndis_version = 0x0006001e;
514
515         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
516         init_packet->msg.v1_msg.
517                 send_ndis_ver.ndis_major_ver =
518                                 (ndis_version & 0xFFFF0000) >> 16;
519         init_packet->msg.v1_msg.
520                 send_ndis_ver.ndis_minor_ver =
521                                 ndis_version & 0xFFFF;
522
523         /* Send the init request */
524         ret = vmbus_sendpacket(device->channel, init_packet,
525                                 sizeof(struct nvsp_message),
526                                 (unsigned long)init_packet,
527                                 VM_PKT_DATA_INBAND, 0);
528         if (ret != 0)
529                 goto cleanup;
530
531         /* Post the big receive buffer to NetVSP */
532         if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
533                 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
534         else
535                 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
536         net_device->send_buf_size = NETVSC_SEND_BUFFER_SIZE;
537
538         ret = netvsc_init_buf(device);
539
540 cleanup:
541         return ret;
542 }
543
544 static void netvsc_disconnect_vsp(struct netvsc_device *net_device)
545 {
546         netvsc_destroy_buf(net_device);
547 }
548
549 /*
550  * netvsc_device_remove - Callback when the root bus device is removed
551  */
552 int netvsc_device_remove(struct hv_device *device)
553 {
554         struct netvsc_device *net_device;
555         unsigned long flags;
556
557         net_device = hv_get_drvdata(device);
558
559         netvsc_disconnect_vsp(net_device);
560
561         /*
562          * Since we have already drained, we don't need to busy wait
563          * as was done in final_release_stor_device()
564          * Note that we cannot set the ext pointer to NULL until
565          * we have drained - to drain the outgoing packets, we need to
566          * allow incoming packets.
567          */
568
569         spin_lock_irqsave(&device->channel->inbound_lock, flags);
570         hv_set_drvdata(device, NULL);
571         spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
572
573         /*
574          * At this point, no one should be accessing net_device
575          * except in here
576          */
577         dev_notice(&device->device, "net device safe to remove\n");
578
579         /* Now, we can close the channel safely */
580         vmbus_close(device->channel);
581
582         /* Release all resources */
583         vfree(net_device->sub_cb_buf);
584         free_netvsc_device(net_device);
585         return 0;
586 }
587
588
589 #define RING_AVAIL_PERCENT_HIWATER 20
590 #define RING_AVAIL_PERCENT_LOWATER 10
591
592 /*
593  * Get the percentage of available bytes to write in the ring.
594  * The return value is in range from 0 to 100.
595  */
596 static inline u32 hv_ringbuf_avail_percent(
597                 struct hv_ring_buffer_info *ring_info)
598 {
599         u32 avail_read, avail_write;
600
601         hv_get_ringbuffer_availbytes(ring_info, &avail_read, &avail_write);
602
603         return avail_write * 100 / ring_info->ring_datasize;
604 }
605
606 static inline void netvsc_free_send_slot(struct netvsc_device *net_device,
607                                          u32 index)
608 {
609         sync_change_bit(index, net_device->send_section_map);
610 }
611
612 static void netvsc_send_completion(struct netvsc_device *net_device,
613                                    struct hv_device *device,
614                                    struct vmpacket_descriptor *packet)
615 {
616         struct nvsp_message *nvsp_packet;
617         struct hv_netvsc_packet *nvsc_packet;
618         struct net_device *ndev;
619         u32 send_index;
620
621         ndev = net_device->ndev;
622
623         nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
624                         (packet->offset8 << 3));
625
626         if ((nvsp_packet->hdr.msg_type == NVSP_MSG_TYPE_INIT_COMPLETE) ||
627             (nvsp_packet->hdr.msg_type ==
628              NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE) ||
629             (nvsp_packet->hdr.msg_type ==
630              NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE) ||
631             (nvsp_packet->hdr.msg_type ==
632              NVSP_MSG5_TYPE_SUBCHANNEL)) {
633                 /* Copy the response back */
634                 memcpy(&net_device->channel_init_pkt, nvsp_packet,
635                        sizeof(struct nvsp_message));
636                 complete(&net_device->channel_init_wait);
637         } else if (nvsp_packet->hdr.msg_type ==
638                    NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE) {
639                 int num_outstanding_sends;
640                 u16 q_idx = 0;
641                 struct vmbus_channel *channel = device->channel;
642                 int queue_sends;
643
644                 /* Get the send context */
645                 nvsc_packet = (struct hv_netvsc_packet *)(unsigned long)
646                         packet->trans_id;
647
648                 /* Notify the layer above us */
649                 if (nvsc_packet) {
650                         send_index = nvsc_packet->send_buf_index;
651                         if (send_index != NETVSC_INVALID_INDEX)
652                                 netvsc_free_send_slot(net_device, send_index);
653                         q_idx = nvsc_packet->q_idx;
654                         channel = nvsc_packet->channel;
655                         nvsc_packet->send_completion(nvsc_packet->
656                                                      send_completion_ctx);
657                 }
658
659                 num_outstanding_sends =
660                         atomic_dec_return(&net_device->num_outstanding_sends);
661                 queue_sends = atomic_dec_return(&net_device->
662                                                 queue_sends[q_idx]);
663
664                 if (net_device->destroy && num_outstanding_sends == 0)
665                         wake_up(&net_device->wait_drain);
666
667                 if (netif_tx_queue_stopped(netdev_get_tx_queue(ndev, q_idx)) &&
668                     !net_device->start_remove &&
669                     (hv_ringbuf_avail_percent(&channel->outbound) >
670                      RING_AVAIL_PERCENT_HIWATER || queue_sends < 1))
671                                 netif_tx_wake_queue(netdev_get_tx_queue(
672                                                     ndev, q_idx));
673         } else {
674                 netdev_err(ndev, "Unknown send completion packet type- "
675                            "%d received!!\n", nvsp_packet->hdr.msg_type);
676         }
677
678 }
679
680 static u32 netvsc_get_next_send_section(struct netvsc_device *net_device)
681 {
682         unsigned long index;
683         u32 max_words = net_device->map_words;
684         unsigned long *map_addr = (unsigned long *)net_device->send_section_map;
685         u32 section_cnt = net_device->send_section_cnt;
686         int ret_val = NETVSC_INVALID_INDEX;
687         int i;
688         int prev_val;
689
690         for (i = 0; i < max_words; i++) {
691                 if (!~(map_addr[i]))
692                         continue;
693                 index = ffz(map_addr[i]);
694                 prev_val = sync_test_and_set_bit(index, &map_addr[i]);
695                 if (prev_val)
696                         continue;
697                 if ((index + (i * BITS_PER_LONG)) >= section_cnt)
698                         break;
699                 ret_val = (index + (i * BITS_PER_LONG));
700                 break;
701         }
702         return ret_val;
703 }
704
705 static u32 netvsc_copy_to_send_buf(struct netvsc_device *net_device,
706                                    unsigned int section_index,
707                                    u32 pend_size,
708                                    struct hv_netvsc_packet *packet)
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         u32 msg_size = 0;
715         u32 padding = 0;
716         u32 remain = packet->total_data_buflen % net_device->pkt_align;
717         u32 page_count = packet->cp_partial ? packet->rmsg_pgcnt :
718                 packet->page_buf_cnt;
719
720         /* Add padding */
721         if (packet->is_data_pkt && packet->xmit_more && remain &&
722             !packet->cp_partial) {
723                 padding = net_device->pkt_align - remain;
724                 packet->rndis_msg->msg_len += padding;
725                 packet->total_data_buflen += padding;
726         }
727
728         for (i = 0; i < page_count; i++) {
729                 char *src = phys_to_virt(packet->page_buf[i].pfn << PAGE_SHIFT);
730                 u32 offset = packet->page_buf[i].offset;
731                 u32 len = packet->page_buf[i].len;
732
733                 memcpy(dest, (src + offset), len);
734                 msg_size += len;
735                 dest += len;
736         }
737
738         if (padding) {
739                 memset(dest, 0, padding);
740                 msg_size += padding;
741         }
742
743         return msg_size;
744 }
745
746 static inline int netvsc_send_pkt(
747         struct hv_netvsc_packet *packet,
748         struct netvsc_device *net_device)
749 {
750         struct nvsp_message nvmsg;
751         struct vmbus_channel *out_channel = packet->channel;
752         u16 q_idx = packet->q_idx;
753         struct net_device *ndev = net_device->ndev;
754         u64 req_id;
755         int ret;
756         struct hv_page_buffer *pgbuf;
757         u32 ring_avail = hv_ringbuf_avail_percent(&out_channel->outbound);
758
759         nvmsg.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
760         if (packet->is_data_pkt) {
761                 /* 0 is RMC_DATA; */
762                 nvmsg.msg.v1_msg.send_rndis_pkt.channel_type = 0;
763         } else {
764                 /* 1 is RMC_CONTROL; */
765                 nvmsg.msg.v1_msg.send_rndis_pkt.channel_type = 1;
766         }
767
768         nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_index =
769                 packet->send_buf_index;
770         if (packet->send_buf_index == NETVSC_INVALID_INDEX)
771                 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_size = 0;
772         else
773                 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_size =
774                         packet->total_data_buflen;
775
776         if (packet->send_completion)
777                 req_id = (ulong)packet;
778         else
779                 req_id = 0;
780
781         if (out_channel->rescind)
782                 return -ENODEV;
783
784         /*
785          * It is possible that once we successfully place this packet
786          * on the ringbuffer, we may stop the queue. In that case, we want
787          * to notify the host independent of the xmit_more flag. We don't
788          * need to be precise here; in the worst case we may signal the host
789          * unnecessarily.
790          */
791         if (ring_avail < (RING_AVAIL_PERCENT_LOWATER + 1))
792                 packet->xmit_more = false;
793
794         if (packet->page_buf_cnt) {
795                 pgbuf = packet->cp_partial ? packet->page_buf +
796                         packet->rmsg_pgcnt : packet->page_buf;
797                 ret = vmbus_sendpacket_pagebuffer_ctl(out_channel,
798                                                       pgbuf,
799                                                       packet->page_buf_cnt,
800                                                       &nvmsg,
801                                                       sizeof(struct nvsp_message),
802                                                       req_id,
803                                                       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED,
804                                                       !packet->xmit_more);
805         } else {
806                 ret = vmbus_sendpacket_ctl(out_channel, &nvmsg,
807                                            sizeof(struct nvsp_message),
808                                            req_id,
809                                            VM_PKT_DATA_INBAND,
810                                            VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED,
811                                            !packet->xmit_more);
812         }
813
814         if (ret == 0) {
815                 atomic_inc(&net_device->num_outstanding_sends);
816                 atomic_inc(&net_device->queue_sends[q_idx]);
817
818                 if (ring_avail < RING_AVAIL_PERCENT_LOWATER) {
819                         netif_tx_stop_queue(netdev_get_tx_queue(ndev, q_idx));
820
821                         if (atomic_read(&net_device->
822                                 queue_sends[q_idx]) < 1)
823                                 netif_tx_wake_queue(netdev_get_tx_queue(
824                                                     ndev, q_idx));
825                 }
826         } else if (ret == -EAGAIN) {
827                 netif_tx_stop_queue(netdev_get_tx_queue(
828                                     ndev, q_idx));
829                 if (atomic_read(&net_device->queue_sends[q_idx]) < 1) {
830                         netif_tx_wake_queue(netdev_get_tx_queue(
831                                             ndev, q_idx));
832                         ret = -ENOSPC;
833                 }
834         } else {
835                 netdev_err(ndev, "Unable to send packet %p ret %d\n",
836                            packet, ret);
837         }
838
839         return ret;
840 }
841
842 int netvsc_send(struct hv_device *device,
843                 struct hv_netvsc_packet *packet)
844 {
845         struct netvsc_device *net_device;
846         int ret = 0, m_ret = 0;
847         struct vmbus_channel *out_channel;
848         u16 q_idx = packet->q_idx;
849         u32 pktlen = packet->total_data_buflen, msd_len = 0;
850         unsigned int section_index = NETVSC_INVALID_INDEX;
851         unsigned long flag;
852         struct multi_send_data *msdp;
853         struct hv_netvsc_packet *msd_send = NULL, *cur_send = NULL;
854         bool try_batch;
855
856         net_device = get_outbound_net_device(device);
857         if (!net_device)
858                 return -ENODEV;
859
860         out_channel = net_device->chn_table[q_idx];
861         if (!out_channel) {
862                 out_channel = device->channel;
863                 q_idx = 0;
864                 packet->q_idx = 0;
865         }
866         packet->channel = out_channel;
867         packet->send_buf_index = NETVSC_INVALID_INDEX;
868         packet->cp_partial = false;
869
870         msdp = &net_device->msd[q_idx];
871
872         /* batch packets in send buffer if possible */
873         spin_lock_irqsave(&msdp->lock, flag);
874         if (msdp->pkt)
875                 msd_len = msdp->pkt->total_data_buflen;
876
877         try_batch = packet->is_data_pkt && msd_len > 0 && msdp->count <
878                     net_device->max_pkt;
879
880         if (try_batch && msd_len + pktlen + net_device->pkt_align <
881             net_device->send_section_size) {
882                 section_index = msdp->pkt->send_buf_index;
883
884         } else if (try_batch && msd_len + packet->rmsg_size <
885                    net_device->send_section_size) {
886                 section_index = msdp->pkt->send_buf_index;
887                 packet->cp_partial = true;
888
889         } else if (packet->is_data_pkt && pktlen + net_device->pkt_align <
890                    net_device->send_section_size) {
891                 section_index = netvsc_get_next_send_section(net_device);
892                 if (section_index != NETVSC_INVALID_INDEX) {
893                                 msd_send = msdp->pkt;
894                                 msdp->pkt = NULL;
895                                 msdp->count = 0;
896                                 msd_len = 0;
897                 }
898         }
899
900         if (section_index != NETVSC_INVALID_INDEX) {
901                 netvsc_copy_to_send_buf(net_device,
902                                         section_index, msd_len,
903                                         packet);
904
905                 packet->send_buf_index = section_index;
906
907                 if (packet->cp_partial) {
908                         packet->page_buf_cnt -= packet->rmsg_pgcnt;
909                         packet->total_data_buflen = msd_len + packet->rmsg_size;
910                 } else {
911                         packet->page_buf_cnt = 0;
912                         packet->total_data_buflen += msd_len;
913                 }
914
915                 if (msdp->pkt)
916                         netvsc_xmit_completion(msdp->pkt);
917
918                 if (packet->xmit_more && !packet->cp_partial) {
919                         msdp->pkt = packet;
920                         msdp->count++;
921                 } else {
922                         cur_send = packet;
923                         msdp->pkt = NULL;
924                         msdp->count = 0;
925                 }
926         } else {
927                 msd_send = msdp->pkt;
928                 msdp->pkt = NULL;
929                 msdp->count = 0;
930                 cur_send = packet;
931         }
932
933         spin_unlock_irqrestore(&msdp->lock, flag);
934
935         if (msd_send) {
936                 m_ret = netvsc_send_pkt(msd_send, net_device);
937
938                 if (m_ret != 0) {
939                         netvsc_free_send_slot(net_device,
940                                               msd_send->send_buf_index);
941                         netvsc_xmit_completion(msd_send);
942                 }
943         }
944
945         if (cur_send)
946                 ret = netvsc_send_pkt(cur_send, net_device);
947
948         if (ret != 0 && section_index != NETVSC_INVALID_INDEX)
949                 netvsc_free_send_slot(net_device, section_index);
950
951         return ret;
952 }
953
954 static void netvsc_send_recv_completion(struct hv_device *device,
955                                         struct vmbus_channel *channel,
956                                         struct netvsc_device *net_device,
957                                         u64 transaction_id, u32 status)
958 {
959         struct nvsp_message recvcompMessage;
960         int retries = 0;
961         int ret;
962         struct net_device *ndev;
963
964         ndev = net_device->ndev;
965
966         recvcompMessage.hdr.msg_type =
967                                 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE;
968
969         recvcompMessage.msg.v1_msg.send_rndis_pkt_complete.status = status;
970
971 retry_send_cmplt:
972         /* Send the completion */
973         ret = vmbus_sendpacket(channel, &recvcompMessage,
974                                sizeof(struct nvsp_message), transaction_id,
975                                VM_PKT_COMP, 0);
976         if (ret == 0) {
977                 /* success */
978                 /* no-op */
979         } else if (ret == -EAGAIN) {
980                 /* no more room...wait a bit and attempt to retry 3 times */
981                 retries++;
982                 netdev_err(ndev, "unable to send receive completion pkt"
983                         " (tid %llx)...retrying %d\n", transaction_id, retries);
984
985                 if (retries < 4) {
986                         udelay(100);
987                         goto retry_send_cmplt;
988                 } else {
989                         netdev_err(ndev, "unable to send receive "
990                                 "completion pkt (tid %llx)...give up retrying\n",
991                                 transaction_id);
992                 }
993         } else {
994                 netdev_err(ndev, "unable to send receive "
995                         "completion pkt - %llx\n", transaction_id);
996         }
997 }
998
999 static void netvsc_receive(struct netvsc_device *net_device,
1000                         struct vmbus_channel *channel,
1001                         struct hv_device *device,
1002                         struct vmpacket_descriptor *packet)
1003 {
1004         struct vmtransfer_page_packet_header *vmxferpage_packet;
1005         struct nvsp_message *nvsp_packet;
1006         struct hv_netvsc_packet nv_pkt;
1007         struct hv_netvsc_packet *netvsc_packet = &nv_pkt;
1008         u32 status = NVSP_STAT_SUCCESS;
1009         int i;
1010         int count = 0;
1011         struct net_device *ndev;
1012
1013         ndev = net_device->ndev;
1014
1015         /*
1016          * All inbound packets other than send completion should be xfer page
1017          * packet
1018          */
1019         if (packet->type != VM_PKT_DATA_USING_XFER_PAGES) {
1020                 netdev_err(ndev, "Unknown packet type received - %d\n",
1021                            packet->type);
1022                 return;
1023         }
1024
1025         nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
1026                         (packet->offset8 << 3));
1027
1028         /* Make sure this is a valid nvsp packet */
1029         if (nvsp_packet->hdr.msg_type !=
1030             NVSP_MSG1_TYPE_SEND_RNDIS_PKT) {
1031                 netdev_err(ndev, "Unknown nvsp packet type received-"
1032                         " %d\n", nvsp_packet->hdr.msg_type);
1033                 return;
1034         }
1035
1036         vmxferpage_packet = (struct vmtransfer_page_packet_header *)packet;
1037
1038         if (vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID) {
1039                 netdev_err(ndev, "Invalid xfer page set id - "
1040                            "expecting %x got %x\n", NETVSC_RECEIVE_BUFFER_ID,
1041                            vmxferpage_packet->xfer_pageset_id);
1042                 return;
1043         }
1044
1045         count = vmxferpage_packet->range_cnt;
1046         netvsc_packet->channel = channel;
1047
1048         /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
1049         for (i = 0; i < count; i++) {
1050                 /* Initialize the netvsc packet */
1051                 netvsc_packet->status = NVSP_STAT_SUCCESS;
1052                 netvsc_packet->data = (void *)((unsigned long)net_device->
1053                         recv_buf + vmxferpage_packet->ranges[i].byte_offset);
1054                 netvsc_packet->total_data_buflen =
1055                                         vmxferpage_packet->ranges[i].byte_count;
1056
1057                 /* Pass it to the upper layer */
1058                 rndis_filter_receive(device, netvsc_packet);
1059
1060                 if (netvsc_packet->status != NVSP_STAT_SUCCESS)
1061                         status = NVSP_STAT_FAIL;
1062         }
1063
1064         netvsc_send_recv_completion(device, channel, net_device,
1065                                     vmxferpage_packet->d.trans_id, status);
1066 }
1067
1068
1069 static void netvsc_send_table(struct hv_device *hdev,
1070                               struct nvsp_message *nvmsg)
1071 {
1072         struct netvsc_device *nvscdev;
1073         struct net_device *ndev;
1074         int i;
1075         u32 count, *tab;
1076
1077         nvscdev = get_outbound_net_device(hdev);
1078         if (!nvscdev)
1079                 return;
1080         ndev = nvscdev->ndev;
1081
1082         count = nvmsg->msg.v5_msg.send_table.count;
1083         if (count != VRSS_SEND_TAB_SIZE) {
1084                 netdev_err(ndev, "Received wrong send-table size:%u\n", count);
1085                 return;
1086         }
1087
1088         tab = (u32 *)((unsigned long)&nvmsg->msg.v5_msg.send_table +
1089                       nvmsg->msg.v5_msg.send_table.offset);
1090
1091         for (i = 0; i < count; i++)
1092                 nvscdev->send_table[i] = tab[i];
1093 }
1094
1095 static void netvsc_send_vf(struct netvsc_device *nvdev,
1096                            struct nvsp_message *nvmsg)
1097 {
1098         nvdev->vf_alloc = nvmsg->msg.v4_msg.vf_assoc.allocated;
1099         nvdev->vf_serial = nvmsg->msg.v4_msg.vf_assoc.serial;
1100 }
1101
1102 static inline void netvsc_receive_inband(struct hv_device *hdev,
1103                                          struct netvsc_device *nvdev,
1104                                          struct nvsp_message *nvmsg)
1105 {
1106         switch (nvmsg->hdr.msg_type) {
1107         case NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE:
1108                 netvsc_send_table(hdev, nvmsg);
1109                 break;
1110
1111         case NVSP_MSG4_TYPE_SEND_VF_ASSOCIATION:
1112                 netvsc_send_vf(nvdev, nvmsg);
1113                 break;
1114         }
1115 }
1116
1117 void netvsc_channel_cb(void *context)
1118 {
1119         int ret;
1120         struct vmbus_channel *channel = (struct vmbus_channel *)context;
1121         struct hv_device *device;
1122         struct netvsc_device *net_device;
1123         u32 bytes_recvd;
1124         u64 request_id;
1125         struct vmpacket_descriptor *desc;
1126         unsigned char *buffer;
1127         int bufferlen = NETVSC_PACKET_SIZE;
1128         struct net_device *ndev;
1129         struct nvsp_message *nvmsg;
1130
1131         if (channel->primary_channel != NULL)
1132                 device = channel->primary_channel->device_obj;
1133         else
1134                 device = channel->device_obj;
1135
1136         net_device = get_inbound_net_device(device);
1137         if (!net_device)
1138                 return;
1139         ndev = net_device->ndev;
1140         buffer = get_per_channel_state(channel);
1141
1142         do {
1143                 ret = vmbus_recvpacket_raw(channel, buffer, bufferlen,
1144                                            &bytes_recvd, &request_id);
1145                 if (ret == 0) {
1146                         if (bytes_recvd > 0) {
1147                                 desc = (struct vmpacket_descriptor *)buffer;
1148                                 nvmsg = (struct nvsp_message *)((unsigned long)
1149                                          desc + (desc->offset8 << 3));
1150                                 switch (desc->type) {
1151                                 case VM_PKT_COMP:
1152                                         netvsc_send_completion(net_device,
1153                                                                 device, desc);
1154                                         break;
1155
1156                                 case VM_PKT_DATA_USING_XFER_PAGES:
1157                                         netvsc_receive(net_device, channel,
1158                                                        device, desc);
1159                                         break;
1160
1161                                 case VM_PKT_DATA_INBAND:
1162                                         netvsc_receive_inband(device,
1163                                                               net_device,
1164                                                               nvmsg);
1165                                         break;
1166
1167                                 default:
1168                                         netdev_err(ndev,
1169                                                    "unhandled packet type %d, "
1170                                                    "tid %llx len %d\n",
1171                                                    desc->type, request_id,
1172                                                    bytes_recvd);
1173                                         break;
1174                                 }
1175
1176                         } else {
1177                                 /*
1178                                  * We are done for this pass.
1179                                  */
1180                                 break;
1181                         }
1182
1183                 } else if (ret == -ENOBUFS) {
1184                         if (bufferlen > NETVSC_PACKET_SIZE)
1185                                 kfree(buffer);
1186                         /* Handle large packet */
1187                         buffer = kmalloc(bytes_recvd, GFP_ATOMIC);
1188                         if (buffer == NULL) {
1189                                 /* Try again next time around */
1190                                 netdev_err(ndev,
1191                                            "unable to allocate buffer of size "
1192                                            "(%d)!!\n", bytes_recvd);
1193                                 break;
1194                         }
1195
1196                         bufferlen = bytes_recvd;
1197                 }
1198         } while (1);
1199
1200         if (bufferlen > NETVSC_PACKET_SIZE)
1201                 kfree(buffer);
1202         return;
1203 }
1204
1205 /*
1206  * netvsc_device_add - Callback when the device belonging to this
1207  * driver is added
1208  */
1209 int netvsc_device_add(struct hv_device *device, void *additional_info)
1210 {
1211         int ret = 0;
1212         int ring_size =
1213         ((struct netvsc_device_info *)additional_info)->ring_size;
1214         struct netvsc_device *net_device;
1215         struct net_device *ndev;
1216
1217         net_device = alloc_net_device(device);
1218         if (!net_device)
1219                 return -ENOMEM;
1220
1221         net_device->ring_size = ring_size;
1222
1223         /*
1224          * Coming into this function, struct net_device * is
1225          * registered as the driver private data.
1226          * In alloc_net_device(), we register struct netvsc_device *
1227          * as the driver private data and stash away struct net_device *
1228          * in struct netvsc_device *.
1229          */
1230         ndev = net_device->ndev;
1231
1232         /* Add netvsc_device context to netvsc_device */
1233         net_device->nd_ctx = netdev_priv(ndev);
1234
1235         /* Initialize the NetVSC channel extension */
1236         init_completion(&net_device->channel_init_wait);
1237
1238         set_per_channel_state(device->channel, net_device->cb_buffer);
1239
1240         /* Open the channel */
1241         ret = vmbus_open(device->channel, ring_size * PAGE_SIZE,
1242                          ring_size * PAGE_SIZE, NULL, 0,
1243                          netvsc_channel_cb, device->channel);
1244
1245         if (ret != 0) {
1246                 netdev_err(ndev, "unable to open channel: %d\n", ret);
1247                 goto cleanup;
1248         }
1249
1250         /* Channel is opened */
1251         pr_info("hv_netvsc channel opened successfully\n");
1252
1253         net_device->chn_table[0] = device->channel;
1254
1255         /* Connect with the NetVsp */
1256         ret = netvsc_connect_vsp(device);
1257         if (ret != 0) {
1258                 netdev_err(ndev,
1259                         "unable to connect to NetVSP - %d\n", ret);
1260                 goto close;
1261         }
1262
1263         return ret;
1264
1265 close:
1266         /* Now, we can close the channel safely */
1267         vmbus_close(device->channel);
1268
1269 cleanup:
1270         free_netvsc_device(net_device);
1271
1272         return ret;
1273 }