GNU Linux-libre 4.19.264-gnu1
[releases.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 <linux/rtnetlink.h>
33 #include <linux/prefetch.h>
34
35 #include <asm/sync_bitops.h>
36
37 #include "hyperv_net.h"
38 #include "netvsc_trace.h"
39
40 /*
41  * Switch the data path from the synthetic interface to the VF
42  * interface.
43  */
44 void netvsc_switch_datapath(struct net_device *ndev, bool vf)
45 {
46         struct net_device_context *net_device_ctx = netdev_priv(ndev);
47         struct hv_device *dev = net_device_ctx->device_ctx;
48         struct netvsc_device *nv_dev = rtnl_dereference(net_device_ctx->nvdev);
49         struct nvsp_message *init_pkt = &nv_dev->channel_init_pkt;
50
51         memset(init_pkt, 0, sizeof(struct nvsp_message));
52         init_pkt->hdr.msg_type = NVSP_MSG4_TYPE_SWITCH_DATA_PATH;
53         if (vf)
54                 init_pkt->msg.v4_msg.active_dp.active_datapath =
55                         NVSP_DATAPATH_VF;
56         else
57                 init_pkt->msg.v4_msg.active_dp.active_datapath =
58                         NVSP_DATAPATH_SYNTHETIC;
59
60         trace_nvsp_send(ndev, init_pkt);
61
62         vmbus_sendpacket(dev->channel, init_pkt,
63                                sizeof(struct nvsp_message),
64                                (unsigned long)init_pkt,
65                                VM_PKT_DATA_INBAND, 0);
66 }
67
68 /* Worker to setup sub channels on initial setup
69  * Initial hotplug event occurs in softirq context
70  * and can't wait for channels.
71  */
72 static void netvsc_subchan_work(struct work_struct *w)
73 {
74         struct netvsc_device *nvdev =
75                 container_of(w, struct netvsc_device, subchan_work);
76         struct rndis_device *rdev;
77         int i, ret;
78
79         /* Avoid deadlock with device removal already under RTNL */
80         if (!rtnl_trylock()) {
81                 schedule_work(w);
82                 return;
83         }
84
85         rdev = nvdev->extension;
86         if (rdev) {
87                 ret = rndis_set_subchannel(rdev->ndev, nvdev, NULL);
88                 if (ret == 0) {
89                         netif_device_attach(rdev->ndev);
90                 } else {
91                         /* fallback to only primary channel */
92                         for (i = 1; i < nvdev->num_chn; i++)
93                                 netif_napi_del(&nvdev->chan_table[i].napi);
94
95                         nvdev->max_chn = 1;
96                         nvdev->num_chn = 1;
97                 }
98         }
99
100         rtnl_unlock();
101 }
102
103 static struct netvsc_device *alloc_net_device(void)
104 {
105         struct netvsc_device *net_device;
106
107         net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
108         if (!net_device)
109                 return NULL;
110
111         init_waitqueue_head(&net_device->wait_drain);
112         net_device->destroy = false;
113         net_device->tx_disable = true;
114
115         net_device->max_pkt = RNDIS_MAX_PKT_DEFAULT;
116         net_device->pkt_align = RNDIS_PKT_ALIGN_DEFAULT;
117
118         init_completion(&net_device->channel_init_wait);
119         init_waitqueue_head(&net_device->subchan_open);
120         INIT_WORK(&net_device->subchan_work, netvsc_subchan_work);
121
122         return net_device;
123 }
124
125 static void free_netvsc_device(struct rcu_head *head)
126 {
127         struct netvsc_device *nvdev
128                 = container_of(head, struct netvsc_device, rcu);
129         int i;
130
131         kfree(nvdev->extension);
132         vfree(nvdev->recv_buf);
133         vfree(nvdev->send_buf);
134         kfree(nvdev->send_section_map);
135
136         for (i = 0; i < VRSS_CHANNEL_MAX; i++)
137                 vfree(nvdev->chan_table[i].mrc.slots);
138
139         kfree(nvdev);
140 }
141
142 static void free_netvsc_device_rcu(struct netvsc_device *nvdev)
143 {
144         call_rcu(&nvdev->rcu, free_netvsc_device);
145 }
146
147 static void netvsc_revoke_recv_buf(struct hv_device *device,
148                                    struct netvsc_device *net_device,
149                                    struct net_device *ndev)
150 {
151         struct nvsp_message *revoke_packet;
152         int ret;
153
154         /*
155          * If we got a section count, it means we received a
156          * SendReceiveBufferComplete msg (ie sent
157          * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
158          * to send a revoke msg here
159          */
160         if (net_device->recv_section_cnt) {
161                 /* Send the revoke receive buffer */
162                 revoke_packet = &net_device->revoke_packet;
163                 memset(revoke_packet, 0, sizeof(struct nvsp_message));
164
165                 revoke_packet->hdr.msg_type =
166                         NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
167                 revoke_packet->msg.v1_msg.
168                 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
169
170                 trace_nvsp_send(ndev, revoke_packet);
171
172                 ret = vmbus_sendpacket(device->channel,
173                                        revoke_packet,
174                                        sizeof(struct nvsp_message),
175                                        (unsigned long)revoke_packet,
176                                        VM_PKT_DATA_INBAND, 0);
177                 /* If the failure is because the channel is rescinded;
178                  * ignore the failure since we cannot send on a rescinded
179                  * channel. This would allow us to properly cleanup
180                  * even when the channel is rescinded.
181                  */
182                 if (device->channel->rescind)
183                         ret = 0;
184                 /*
185                  * If we failed here, we might as well return and
186                  * have a leak rather than continue and a bugchk
187                  */
188                 if (ret != 0) {
189                         netdev_err(ndev, "unable to send "
190                                 "revoke receive buffer to netvsp\n");
191                         return;
192                 }
193                 net_device->recv_section_cnt = 0;
194         }
195 }
196
197 static void netvsc_revoke_send_buf(struct hv_device *device,
198                                    struct netvsc_device *net_device,
199                                    struct net_device *ndev)
200 {
201         struct nvsp_message *revoke_packet;
202         int ret;
203
204         /* Deal with the send buffer we may have setup.
205          * If we got a  send section size, it means we received a
206          * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
207          * NVSP_MSG1_TYPE_SEND_SEND_BUF msg) therefore, we need
208          * to send a revoke msg here
209          */
210         if (net_device->send_section_cnt) {
211                 /* Send the revoke receive buffer */
212                 revoke_packet = &net_device->revoke_packet;
213                 memset(revoke_packet, 0, sizeof(struct nvsp_message));
214
215                 revoke_packet->hdr.msg_type =
216                         NVSP_MSG1_TYPE_REVOKE_SEND_BUF;
217                 revoke_packet->msg.v1_msg.revoke_send_buf.id =
218                         NETVSC_SEND_BUFFER_ID;
219
220                 trace_nvsp_send(ndev, revoke_packet);
221
222                 ret = vmbus_sendpacket(device->channel,
223                                        revoke_packet,
224                                        sizeof(struct nvsp_message),
225                                        (unsigned long)revoke_packet,
226                                        VM_PKT_DATA_INBAND, 0);
227
228                 /* If the failure is because the channel is rescinded;
229                  * ignore the failure since we cannot send on a rescinded
230                  * channel. This would allow us to properly cleanup
231                  * even when the channel is rescinded.
232                  */
233                 if (device->channel->rescind)
234                         ret = 0;
235
236                 /* If we failed here, we might as well return and
237                  * have a leak rather than continue and a bugchk
238                  */
239                 if (ret != 0) {
240                         netdev_err(ndev, "unable to send "
241                                    "revoke send buffer to netvsp\n");
242                         return;
243                 }
244                 net_device->send_section_cnt = 0;
245         }
246 }
247
248 static void netvsc_teardown_recv_gpadl(struct hv_device *device,
249                                        struct netvsc_device *net_device,
250                                        struct net_device *ndev)
251 {
252         int ret;
253
254         if (net_device->recv_buf_gpadl_handle) {
255                 ret = vmbus_teardown_gpadl(device->channel,
256                                            net_device->recv_buf_gpadl_handle);
257
258                 /* If we failed here, we might as well return and have a leak
259                  * rather than continue and a bugchk
260                  */
261                 if (ret != 0) {
262                         netdev_err(ndev,
263                                    "unable to teardown receive buffer's gpadl\n");
264                         return;
265                 }
266                 net_device->recv_buf_gpadl_handle = 0;
267         }
268 }
269
270 static void netvsc_teardown_send_gpadl(struct hv_device *device,
271                                        struct netvsc_device *net_device,
272                                        struct net_device *ndev)
273 {
274         int ret;
275
276         if (net_device->send_buf_gpadl_handle) {
277                 ret = vmbus_teardown_gpadl(device->channel,
278                                            net_device->send_buf_gpadl_handle);
279
280                 /* If we failed here, we might as well return and have a leak
281                  * rather than continue and a bugchk
282                  */
283                 if (ret != 0) {
284                         netdev_err(ndev,
285                                    "unable to teardown send buffer's gpadl\n");
286                         return;
287                 }
288                 net_device->send_buf_gpadl_handle = 0;
289         }
290 }
291
292 int netvsc_alloc_recv_comp_ring(struct netvsc_device *net_device, u32 q_idx)
293 {
294         struct netvsc_channel *nvchan = &net_device->chan_table[q_idx];
295         int node = cpu_to_node(nvchan->channel->target_cpu);
296         size_t size;
297
298         size = net_device->recv_completion_cnt * sizeof(struct recv_comp_data);
299         nvchan->mrc.slots = vzalloc_node(size, node);
300         if (!nvchan->mrc.slots)
301                 nvchan->mrc.slots = vzalloc(size);
302
303         return nvchan->mrc.slots ? 0 : -ENOMEM;
304 }
305
306 static int netvsc_init_buf(struct hv_device *device,
307                            struct netvsc_device *net_device,
308                            const struct netvsc_device_info *device_info)
309 {
310         struct nvsp_1_message_send_receive_buffer_complete *resp;
311         struct net_device *ndev = hv_get_drvdata(device);
312         struct nvsp_message *init_packet;
313         unsigned int buf_size;
314         size_t map_words;
315         int ret = 0;
316
317         /* Get receive buffer area. */
318         buf_size = device_info->recv_sections * device_info->recv_section_size;
319         buf_size = roundup(buf_size, PAGE_SIZE);
320
321         /* Legacy hosts only allow smaller receive buffer */
322         if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
323                 buf_size = min_t(unsigned int, buf_size,
324                                  NETVSC_RECEIVE_BUFFER_SIZE_LEGACY);
325
326         net_device->recv_buf = vzalloc(buf_size);
327         if (!net_device->recv_buf) {
328                 netdev_err(ndev,
329                            "unable to allocate receive buffer of size %u\n",
330                            buf_size);
331                 ret = -ENOMEM;
332                 goto cleanup;
333         }
334
335         net_device->recv_buf_size = buf_size;
336
337         /*
338          * Establish the gpadl handle for this buffer on this
339          * channel.  Note: This call uses the vmbus connection rather
340          * than the channel to establish the gpadl handle.
341          */
342         ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
343                                     buf_size,
344                                     &net_device->recv_buf_gpadl_handle);
345         if (ret != 0) {
346                 netdev_err(ndev,
347                         "unable to establish receive buffer's gpadl\n");
348                 goto cleanup;
349         }
350
351         /* Notify the NetVsp of the gpadl handle */
352         init_packet = &net_device->channel_init_pkt;
353         memset(init_packet, 0, sizeof(struct nvsp_message));
354         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
355         init_packet->msg.v1_msg.send_recv_buf.
356                 gpadl_handle = net_device->recv_buf_gpadl_handle;
357         init_packet->msg.v1_msg.
358                 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
359
360         trace_nvsp_send(ndev, init_packet);
361
362         /* Send the gpadl notification request */
363         ret = vmbus_sendpacket(device->channel, init_packet,
364                                sizeof(struct nvsp_message),
365                                (unsigned long)init_packet,
366                                VM_PKT_DATA_INBAND,
367                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
368         if (ret != 0) {
369                 netdev_err(ndev,
370                         "unable to send receive buffer's gpadl to netvsp\n");
371                 goto cleanup;
372         }
373
374         wait_for_completion(&net_device->channel_init_wait);
375
376         /* Check the response */
377         resp = &init_packet->msg.v1_msg.send_recv_buf_complete;
378         if (resp->status != NVSP_STAT_SUCCESS) {
379                 netdev_err(ndev,
380                            "Unable to complete receive buffer initialization with NetVsp - status %d\n",
381                            resp->status);
382                 ret = -EINVAL;
383                 goto cleanup;
384         }
385
386         /* Parse the response */
387         netdev_dbg(ndev, "Receive sections: %u sub_allocs: size %u count: %u\n",
388                    resp->num_sections, resp->sections[0].sub_alloc_size,
389                    resp->sections[0].num_sub_allocs);
390
391         /* There should only be one section for the entire receive buffer */
392         if (resp->num_sections != 1 || resp->sections[0].offset != 0) {
393                 ret = -EINVAL;
394                 goto cleanup;
395         }
396
397         net_device->recv_section_size = resp->sections[0].sub_alloc_size;
398         net_device->recv_section_cnt = resp->sections[0].num_sub_allocs;
399
400         /* Setup receive completion ring */
401         net_device->recv_completion_cnt
402                 = round_up(net_device->recv_section_cnt + 1,
403                            PAGE_SIZE / sizeof(u64));
404         ret = netvsc_alloc_recv_comp_ring(net_device, 0);
405         if (ret)
406                 goto cleanup;
407
408         /* Now setup the send buffer. */
409         buf_size = device_info->send_sections * device_info->send_section_size;
410         buf_size = round_up(buf_size, PAGE_SIZE);
411
412         net_device->send_buf = vzalloc(buf_size);
413         if (!net_device->send_buf) {
414                 netdev_err(ndev, "unable to allocate send buffer of size %u\n",
415                            buf_size);
416                 ret = -ENOMEM;
417                 goto cleanup;
418         }
419
420         /* Establish the gpadl handle for this buffer on this
421          * channel.  Note: This call uses the vmbus connection rather
422          * than the channel to establish the gpadl handle.
423          */
424         ret = vmbus_establish_gpadl(device->channel, net_device->send_buf,
425                                     buf_size,
426                                     &net_device->send_buf_gpadl_handle);
427         if (ret != 0) {
428                 netdev_err(ndev,
429                            "unable to establish send buffer's gpadl\n");
430                 goto cleanup;
431         }
432
433         /* Notify the NetVsp of the gpadl handle */
434         init_packet = &net_device->channel_init_pkt;
435         memset(init_packet, 0, sizeof(struct nvsp_message));
436         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_SEND_BUF;
437         init_packet->msg.v1_msg.send_send_buf.gpadl_handle =
438                 net_device->send_buf_gpadl_handle;
439         init_packet->msg.v1_msg.send_send_buf.id = NETVSC_SEND_BUFFER_ID;
440
441         trace_nvsp_send(ndev, init_packet);
442
443         /* Send the gpadl notification request */
444         ret = vmbus_sendpacket(device->channel, init_packet,
445                                sizeof(struct nvsp_message),
446                                (unsigned long)init_packet,
447                                VM_PKT_DATA_INBAND,
448                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
449         if (ret != 0) {
450                 netdev_err(ndev,
451                            "unable to send send buffer's gpadl to netvsp\n");
452                 goto cleanup;
453         }
454
455         wait_for_completion(&net_device->channel_init_wait);
456
457         /* Check the response */
458         if (init_packet->msg.v1_msg.
459             send_send_buf_complete.status != NVSP_STAT_SUCCESS) {
460                 netdev_err(ndev, "Unable to complete send buffer "
461                            "initialization with NetVsp - status %d\n",
462                            init_packet->msg.v1_msg.
463                            send_send_buf_complete.status);
464                 ret = -EINVAL;
465                 goto cleanup;
466         }
467
468         /* Parse the response */
469         net_device->send_section_size = init_packet->msg.
470                                 v1_msg.send_send_buf_complete.section_size;
471
472         /* Section count is simply the size divided by the section size. */
473         net_device->send_section_cnt = buf_size / net_device->send_section_size;
474
475         netdev_dbg(ndev, "Send section size: %d, Section count:%d\n",
476                    net_device->send_section_size, net_device->send_section_cnt);
477
478         /* Setup state for managing the send buffer. */
479         map_words = DIV_ROUND_UP(net_device->send_section_cnt, BITS_PER_LONG);
480
481         net_device->send_section_map = kcalloc(map_words, sizeof(ulong), GFP_KERNEL);
482         if (net_device->send_section_map == NULL) {
483                 ret = -ENOMEM;
484                 goto cleanup;
485         }
486
487         goto exit;
488
489 cleanup:
490         netvsc_revoke_recv_buf(device, net_device, ndev);
491         netvsc_revoke_send_buf(device, net_device, ndev);
492         netvsc_teardown_recv_gpadl(device, net_device, ndev);
493         netvsc_teardown_send_gpadl(device, net_device, ndev);
494
495 exit:
496         return ret;
497 }
498
499 /* Negotiate NVSP protocol version */
500 static int negotiate_nvsp_ver(struct hv_device *device,
501                               struct netvsc_device *net_device,
502                               struct nvsp_message *init_packet,
503                               u32 nvsp_ver)
504 {
505         struct net_device *ndev = hv_get_drvdata(device);
506         int ret;
507
508         memset(init_packet, 0, sizeof(struct nvsp_message));
509         init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
510         init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
511         init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
512         trace_nvsp_send(ndev, init_packet);
513
514         /* Send the init request */
515         ret = vmbus_sendpacket(device->channel, init_packet,
516                                sizeof(struct nvsp_message),
517                                (unsigned long)init_packet,
518                                VM_PKT_DATA_INBAND,
519                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
520
521         if (ret != 0)
522                 return ret;
523
524         wait_for_completion(&net_device->channel_init_wait);
525
526         if (init_packet->msg.init_msg.init_complete.status !=
527             NVSP_STAT_SUCCESS)
528                 return -EINVAL;
529
530         if (nvsp_ver == NVSP_PROTOCOL_VERSION_1)
531                 return 0;
532
533         /* NVSPv2 or later: Send NDIS config */
534         memset(init_packet, 0, sizeof(struct nvsp_message));
535         init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
536         init_packet->msg.v2_msg.send_ndis_config.mtu = ndev->mtu + ETH_HLEN;
537         init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
538
539         if (nvsp_ver >= NVSP_PROTOCOL_VERSION_5) {
540                 init_packet->msg.v2_msg.send_ndis_config.capability.sriov = 1;
541
542                 /* Teaming bit is needed to receive link speed updates */
543                 init_packet->msg.v2_msg.send_ndis_config.capability.teaming = 1;
544         }
545
546         trace_nvsp_send(ndev, init_packet);
547
548         ret = vmbus_sendpacket(device->channel, init_packet,
549                                 sizeof(struct nvsp_message),
550                                 (unsigned long)init_packet,
551                                 VM_PKT_DATA_INBAND, 0);
552
553         return ret;
554 }
555
556 static int netvsc_connect_vsp(struct hv_device *device,
557                               struct netvsc_device *net_device,
558                               const struct netvsc_device_info *device_info)
559 {
560         struct net_device *ndev = hv_get_drvdata(device);
561         static const u32 ver_list[] = {
562                 NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
563                 NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5,
564                 NVSP_PROTOCOL_VERSION_6, NVSP_PROTOCOL_VERSION_61
565         };
566         struct nvsp_message *init_packet;
567         int ndis_version, i, ret;
568
569         init_packet = &net_device->channel_init_pkt;
570
571         /* Negotiate the latest NVSP protocol supported */
572         for (i = ARRAY_SIZE(ver_list) - 1; i >= 0; i--)
573                 if (negotiate_nvsp_ver(device, net_device, init_packet,
574                                        ver_list[i])  == 0) {
575                         net_device->nvsp_version = ver_list[i];
576                         break;
577                 }
578
579         if (i < 0) {
580                 ret = -EPROTO;
581                 goto cleanup;
582         }
583
584         pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
585
586         /* Send the ndis version */
587         memset(init_packet, 0, sizeof(struct nvsp_message));
588
589         if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
590                 ndis_version = 0x00060001;
591         else
592                 ndis_version = 0x0006001e;
593
594         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
595         init_packet->msg.v1_msg.
596                 send_ndis_ver.ndis_major_ver =
597                                 (ndis_version & 0xFFFF0000) >> 16;
598         init_packet->msg.v1_msg.
599                 send_ndis_ver.ndis_minor_ver =
600                                 ndis_version & 0xFFFF;
601
602         trace_nvsp_send(ndev, init_packet);
603
604         /* Send the init request */
605         ret = vmbus_sendpacket(device->channel, init_packet,
606                                 sizeof(struct nvsp_message),
607                                 (unsigned long)init_packet,
608                                 VM_PKT_DATA_INBAND, 0);
609         if (ret != 0)
610                 goto cleanup;
611
612
613         ret = netvsc_init_buf(device, net_device, device_info);
614
615 cleanup:
616         return ret;
617 }
618
619 /*
620  * netvsc_device_remove - Callback when the root bus device is removed
621  */
622 void netvsc_device_remove(struct hv_device *device)
623 {
624         struct net_device *ndev = hv_get_drvdata(device);
625         struct net_device_context *net_device_ctx = netdev_priv(ndev);
626         struct netvsc_device *net_device
627                 = rtnl_dereference(net_device_ctx->nvdev);
628         int i;
629
630         /*
631          * Revoke receive buffer. If host is pre-Win2016 then tear down
632          * receive buffer GPADL. Do the same for send buffer.
633          */
634         netvsc_revoke_recv_buf(device, net_device, ndev);
635         if (vmbus_proto_version < VERSION_WIN10)
636                 netvsc_teardown_recv_gpadl(device, net_device, ndev);
637
638         netvsc_revoke_send_buf(device, net_device, ndev);
639         if (vmbus_proto_version < VERSION_WIN10)
640                 netvsc_teardown_send_gpadl(device, net_device, ndev);
641
642         RCU_INIT_POINTER(net_device_ctx->nvdev, NULL);
643
644         /* And disassociate NAPI context from device */
645         for (i = 0; i < net_device->num_chn; i++)
646                 netif_napi_del(&net_device->chan_table[i].napi);
647
648         /*
649          * At this point, no one should be accessing net_device
650          * except in here
651          */
652         netdev_dbg(ndev, "net device safe to remove\n");
653
654         /* Now, we can close the channel safely */
655         vmbus_close(device->channel);
656
657         /*
658          * If host is Win2016 or higher then we do the GPADL tear down
659          * here after VMBus is closed.
660         */
661         if (vmbus_proto_version >= VERSION_WIN10) {
662                 netvsc_teardown_recv_gpadl(device, net_device, ndev);
663                 netvsc_teardown_send_gpadl(device, net_device, ndev);
664         }
665
666         /* Release all resources */
667         free_netvsc_device_rcu(net_device);
668 }
669
670 #define RING_AVAIL_PERCENT_HIWATER 20
671 #define RING_AVAIL_PERCENT_LOWATER 10
672
673 static inline void netvsc_free_send_slot(struct netvsc_device *net_device,
674                                          u32 index)
675 {
676         sync_change_bit(index, net_device->send_section_map);
677 }
678
679 static void netvsc_send_tx_complete(struct net_device *ndev,
680                                     struct netvsc_device *net_device,
681                                     struct vmbus_channel *channel,
682                                     const struct vmpacket_descriptor *desc,
683                                     int budget)
684 {
685         struct sk_buff *skb = (struct sk_buff *)(unsigned long)desc->trans_id;
686         struct net_device_context *ndev_ctx = netdev_priv(ndev);
687         u16 q_idx = 0;
688         int queue_sends;
689
690         /* Notify the layer above us */
691         if (likely(skb)) {
692                 const struct hv_netvsc_packet *packet
693                         = (struct hv_netvsc_packet *)skb->cb;
694                 u32 send_index = packet->send_buf_index;
695                 struct netvsc_stats *tx_stats;
696
697                 if (send_index != NETVSC_INVALID_INDEX)
698                         netvsc_free_send_slot(net_device, send_index);
699                 q_idx = packet->q_idx;
700
701                 tx_stats = &net_device->chan_table[q_idx].tx_stats;
702
703                 u64_stats_update_begin(&tx_stats->syncp);
704                 tx_stats->packets += packet->total_packets;
705                 tx_stats->bytes += packet->total_bytes;
706                 u64_stats_update_end(&tx_stats->syncp);
707
708                 napi_consume_skb(skb, budget);
709         }
710
711         queue_sends =
712                 atomic_dec_return(&net_device->chan_table[q_idx].queue_sends);
713
714         if (unlikely(net_device->destroy)) {
715                 if (queue_sends == 0)
716                         wake_up(&net_device->wait_drain);
717         } else {
718                 struct netdev_queue *txq = netdev_get_tx_queue(ndev, q_idx);
719
720                 if (netif_tx_queue_stopped(txq) && !net_device->tx_disable &&
721                     (hv_get_avail_to_write_percent(&channel->outbound) >
722                      RING_AVAIL_PERCENT_HIWATER || queue_sends < 1)) {
723                         netif_tx_wake_queue(txq);
724                         ndev_ctx->eth_stats.wake_queue++;
725                 }
726         }
727 }
728
729 static void netvsc_send_completion(struct net_device *ndev,
730                                    struct netvsc_device *net_device,
731                                    struct vmbus_channel *incoming_channel,
732                                    const struct vmpacket_descriptor *desc,
733                                    int budget)
734 {
735         const struct nvsp_message *nvsp_packet = hv_pkt_data(desc);
736
737         switch (nvsp_packet->hdr.msg_type) {
738         case NVSP_MSG_TYPE_INIT_COMPLETE:
739         case NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE:
740         case NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE:
741         case NVSP_MSG5_TYPE_SUBCHANNEL:
742                 /* Copy the response back */
743                 memcpy(&net_device->channel_init_pkt, nvsp_packet,
744                        sizeof(struct nvsp_message));
745                 complete(&net_device->channel_init_wait);
746                 break;
747
748         case NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE:
749                 netvsc_send_tx_complete(ndev, net_device, incoming_channel,
750                                         desc, budget);
751                 break;
752
753         default:
754                 netdev_err(ndev,
755                            "Unknown send completion type %d received!!\n",
756                            nvsp_packet->hdr.msg_type);
757         }
758 }
759
760 static u32 netvsc_get_next_send_section(struct netvsc_device *net_device)
761 {
762         unsigned long *map_addr = net_device->send_section_map;
763         unsigned int i;
764
765         for_each_clear_bit(i, map_addr, net_device->send_section_cnt) {
766                 if (sync_test_and_set_bit(i, map_addr) == 0)
767                         return i;
768         }
769
770         return NETVSC_INVALID_INDEX;
771 }
772
773 static void netvsc_copy_to_send_buf(struct netvsc_device *net_device,
774                                     unsigned int section_index,
775                                     u32 pend_size,
776                                     struct hv_netvsc_packet *packet,
777                                     struct rndis_message *rndis_msg,
778                                     struct hv_page_buffer *pb,
779                                     bool xmit_more)
780 {
781         char *start = net_device->send_buf;
782         char *dest = start + (section_index * net_device->send_section_size)
783                      + pend_size;
784         int i;
785         u32 padding = 0;
786         u32 page_count = packet->cp_partial ? packet->rmsg_pgcnt :
787                 packet->page_buf_cnt;
788         u32 remain;
789
790         /* Add padding */
791         remain = packet->total_data_buflen & (net_device->pkt_align - 1);
792         if (xmit_more && remain) {
793                 padding = net_device->pkt_align - remain;
794                 rndis_msg->msg_len += padding;
795                 packet->total_data_buflen += padding;
796         }
797
798         for (i = 0; i < page_count; i++) {
799                 char *src = phys_to_virt(pb[i].pfn << PAGE_SHIFT);
800                 u32 offset = pb[i].offset;
801                 u32 len = pb[i].len;
802
803                 memcpy(dest, (src + offset), len);
804                 dest += len;
805         }
806
807         if (padding)
808                 memset(dest, 0, padding);
809 }
810
811 static inline int netvsc_send_pkt(
812         struct hv_device *device,
813         struct hv_netvsc_packet *packet,
814         struct netvsc_device *net_device,
815         struct hv_page_buffer *pb,
816         struct sk_buff *skb)
817 {
818         struct nvsp_message nvmsg;
819         struct nvsp_1_message_send_rndis_packet *rpkt =
820                 &nvmsg.msg.v1_msg.send_rndis_pkt;
821         struct netvsc_channel * const nvchan =
822                 &net_device->chan_table[packet->q_idx];
823         struct vmbus_channel *out_channel = nvchan->channel;
824         struct net_device *ndev = hv_get_drvdata(device);
825         struct net_device_context *ndev_ctx = netdev_priv(ndev);
826         struct netdev_queue *txq = netdev_get_tx_queue(ndev, packet->q_idx);
827         u64 req_id;
828         int ret;
829         u32 ring_avail = hv_get_avail_to_write_percent(&out_channel->outbound);
830
831         nvmsg.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
832         if (skb)
833                 rpkt->channel_type = 0;         /* 0 is RMC_DATA */
834         else
835                 rpkt->channel_type = 1;         /* 1 is RMC_CONTROL */
836
837         rpkt->send_buf_section_index = packet->send_buf_index;
838         if (packet->send_buf_index == NETVSC_INVALID_INDEX)
839                 rpkt->send_buf_section_size = 0;
840         else
841                 rpkt->send_buf_section_size = packet->total_data_buflen;
842
843         req_id = (ulong)skb;
844
845         if (out_channel->rescind)
846                 return -ENODEV;
847
848         trace_nvsp_send_pkt(ndev, out_channel, rpkt);
849
850         if (packet->page_buf_cnt) {
851                 if (packet->cp_partial)
852                         pb += packet->rmsg_pgcnt;
853
854                 ret = vmbus_sendpacket_pagebuffer(out_channel,
855                                                   pb, packet->page_buf_cnt,
856                                                   &nvmsg, sizeof(nvmsg),
857                                                   req_id);
858         } else {
859                 ret = vmbus_sendpacket(out_channel,
860                                        &nvmsg, sizeof(nvmsg),
861                                        req_id, VM_PKT_DATA_INBAND,
862                                        VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
863         }
864
865         if (ret == 0) {
866                 atomic_inc_return(&nvchan->queue_sends);
867
868                 if (ring_avail < RING_AVAIL_PERCENT_LOWATER) {
869                         netif_tx_stop_queue(txq);
870                         ndev_ctx->eth_stats.stop_queue++;
871                 }
872         } else if (ret == -EAGAIN) {
873                 netif_tx_stop_queue(txq);
874                 ndev_ctx->eth_stats.stop_queue++;
875         } else {
876                 netdev_err(ndev,
877                            "Unable to send packet pages %u len %u, ret %d\n",
878                            packet->page_buf_cnt, packet->total_data_buflen,
879                            ret);
880         }
881
882         if (netif_tx_queue_stopped(txq) &&
883             atomic_read(&nvchan->queue_sends) < 1 &&
884             !net_device->tx_disable) {
885                 netif_tx_wake_queue(txq);
886                 ndev_ctx->eth_stats.wake_queue++;
887                 if (ret == -EAGAIN)
888                         ret = -ENOSPC;
889         }
890
891         return ret;
892 }
893
894 /* Move packet out of multi send data (msd), and clear msd */
895 static inline void move_pkt_msd(struct hv_netvsc_packet **msd_send,
896                                 struct sk_buff **msd_skb,
897                                 struct multi_send_data *msdp)
898 {
899         *msd_skb = msdp->skb;
900         *msd_send = msdp->pkt;
901         msdp->skb = NULL;
902         msdp->pkt = NULL;
903         msdp->count = 0;
904 }
905
906 /* RCU already held by caller */
907 int netvsc_send(struct net_device *ndev,
908                 struct hv_netvsc_packet *packet,
909                 struct rndis_message *rndis_msg,
910                 struct hv_page_buffer *pb,
911                 struct sk_buff *skb)
912 {
913         struct net_device_context *ndev_ctx = netdev_priv(ndev);
914         struct netvsc_device *net_device
915                 = rcu_dereference_bh(ndev_ctx->nvdev);
916         struct hv_device *device = ndev_ctx->device_ctx;
917         int ret = 0;
918         struct netvsc_channel *nvchan;
919         u32 pktlen = packet->total_data_buflen, msd_len = 0;
920         unsigned int section_index = NETVSC_INVALID_INDEX;
921         struct multi_send_data *msdp;
922         struct hv_netvsc_packet *msd_send = NULL, *cur_send = NULL;
923         struct sk_buff *msd_skb = NULL;
924         bool try_batch, xmit_more;
925
926         /* If device is rescinded, return error and packet will get dropped. */
927         if (unlikely(!net_device || net_device->destroy))
928                 return -ENODEV;
929
930         nvchan = &net_device->chan_table[packet->q_idx];
931         packet->send_buf_index = NETVSC_INVALID_INDEX;
932         packet->cp_partial = false;
933
934         /* Send control message directly without accessing msd (Multi-Send
935          * Data) field which may be changed during data packet processing.
936          */
937         if (!skb)
938                 return netvsc_send_pkt(device, packet, net_device, pb, skb);
939
940         /* batch packets in send buffer if possible */
941         msdp = &nvchan->msd;
942         if (msdp->pkt)
943                 msd_len = msdp->pkt->total_data_buflen;
944
945         try_batch =  msd_len > 0 && msdp->count < net_device->max_pkt;
946         if (try_batch && msd_len + pktlen + net_device->pkt_align <
947             net_device->send_section_size) {
948                 section_index = msdp->pkt->send_buf_index;
949
950         } else if (try_batch && msd_len + packet->rmsg_size <
951                    net_device->send_section_size) {
952                 section_index = msdp->pkt->send_buf_index;
953                 packet->cp_partial = true;
954
955         } else if (pktlen + net_device->pkt_align <
956                    net_device->send_section_size) {
957                 section_index = netvsc_get_next_send_section(net_device);
958                 if (unlikely(section_index == NETVSC_INVALID_INDEX)) {
959                         ++ndev_ctx->eth_stats.tx_send_full;
960                 } else {
961                         move_pkt_msd(&msd_send, &msd_skb, msdp);
962                         msd_len = 0;
963                 }
964         }
965
966         /* Keep aggregating only if stack says more data is coming
967          * and not doing mixed modes send and not flow blocked
968          */
969         xmit_more = skb->xmit_more &&
970                 !packet->cp_partial &&
971                 !netif_xmit_stopped(netdev_get_tx_queue(ndev, packet->q_idx));
972
973         if (section_index != NETVSC_INVALID_INDEX) {
974                 netvsc_copy_to_send_buf(net_device,
975                                         section_index, msd_len,
976                                         packet, rndis_msg, pb, xmit_more);
977
978                 packet->send_buf_index = section_index;
979
980                 if (packet->cp_partial) {
981                         packet->page_buf_cnt -= packet->rmsg_pgcnt;
982                         packet->total_data_buflen = msd_len + packet->rmsg_size;
983                 } else {
984                         packet->page_buf_cnt = 0;
985                         packet->total_data_buflen += msd_len;
986                 }
987
988                 if (msdp->pkt) {
989                         packet->total_packets += msdp->pkt->total_packets;
990                         packet->total_bytes += msdp->pkt->total_bytes;
991                 }
992
993                 if (msdp->skb)
994                         dev_consume_skb_any(msdp->skb);
995
996                 if (xmit_more) {
997                         msdp->skb = skb;
998                         msdp->pkt = packet;
999                         msdp->count++;
1000                 } else {
1001                         cur_send = packet;
1002                         msdp->skb = NULL;
1003                         msdp->pkt = NULL;
1004                         msdp->count = 0;
1005                 }
1006         } else {
1007                 move_pkt_msd(&msd_send, &msd_skb, msdp);
1008                 cur_send = packet;
1009         }
1010
1011         if (msd_send) {
1012                 int m_ret = netvsc_send_pkt(device, msd_send, net_device,
1013                                             NULL, msd_skb);
1014
1015                 if (m_ret != 0) {
1016                         netvsc_free_send_slot(net_device,
1017                                               msd_send->send_buf_index);
1018                         dev_kfree_skb_any(msd_skb);
1019                 }
1020         }
1021
1022         if (cur_send)
1023                 ret = netvsc_send_pkt(device, cur_send, net_device, pb, skb);
1024
1025         if (ret != 0 && section_index != NETVSC_INVALID_INDEX)
1026                 netvsc_free_send_slot(net_device, section_index);
1027
1028         return ret;
1029 }
1030
1031 /* Send pending recv completions */
1032 static int send_recv_completions(struct net_device *ndev,
1033                                  struct netvsc_device *nvdev,
1034                                  struct netvsc_channel *nvchan)
1035 {
1036         struct multi_recv_comp *mrc = &nvchan->mrc;
1037         struct recv_comp_msg {
1038                 struct nvsp_message_header hdr;
1039                 u32 status;
1040         }  __packed;
1041         struct recv_comp_msg msg = {
1042                 .hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE,
1043         };
1044         int ret;
1045
1046         while (mrc->first != mrc->next) {
1047                 const struct recv_comp_data *rcd
1048                         = mrc->slots + mrc->first;
1049
1050                 msg.status = rcd->status;
1051                 ret = vmbus_sendpacket(nvchan->channel, &msg, sizeof(msg),
1052                                        rcd->tid, VM_PKT_COMP, 0);
1053                 if (unlikely(ret)) {
1054                         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1055
1056                         ++ndev_ctx->eth_stats.rx_comp_busy;
1057                         return ret;
1058                 }
1059
1060                 if (++mrc->first == nvdev->recv_completion_cnt)
1061                         mrc->first = 0;
1062         }
1063
1064         /* receive completion ring has been emptied */
1065         if (unlikely(nvdev->destroy))
1066                 wake_up(&nvdev->wait_drain);
1067
1068         return 0;
1069 }
1070
1071 /* Count how many receive completions are outstanding */
1072 static void recv_comp_slot_avail(const struct netvsc_device *nvdev,
1073                                  const struct multi_recv_comp *mrc,
1074                                  u32 *filled, u32 *avail)
1075 {
1076         u32 count = nvdev->recv_completion_cnt;
1077
1078         if (mrc->next >= mrc->first)
1079                 *filled = mrc->next - mrc->first;
1080         else
1081                 *filled = (count - mrc->first) + mrc->next;
1082
1083         *avail = count - *filled - 1;
1084 }
1085
1086 /* Add receive complete to ring to send to host. */
1087 static void enq_receive_complete(struct net_device *ndev,
1088                                  struct netvsc_device *nvdev, u16 q_idx,
1089                                  u64 tid, u32 status)
1090 {
1091         struct netvsc_channel *nvchan = &nvdev->chan_table[q_idx];
1092         struct multi_recv_comp *mrc = &nvchan->mrc;
1093         struct recv_comp_data *rcd;
1094         u32 filled, avail;
1095
1096         recv_comp_slot_avail(nvdev, mrc, &filled, &avail);
1097
1098         if (unlikely(filled > NAPI_POLL_WEIGHT)) {
1099                 send_recv_completions(ndev, nvdev, nvchan);
1100                 recv_comp_slot_avail(nvdev, mrc, &filled, &avail);
1101         }
1102
1103         if (unlikely(!avail)) {
1104                 netdev_err(ndev, "Recv_comp full buf q:%hd, tid:%llx\n",
1105                            q_idx, tid);
1106                 return;
1107         }
1108
1109         rcd = mrc->slots + mrc->next;
1110         rcd->tid = tid;
1111         rcd->status = status;
1112
1113         if (++mrc->next == nvdev->recv_completion_cnt)
1114                 mrc->next = 0;
1115 }
1116
1117 static int netvsc_receive(struct net_device *ndev,
1118                           struct netvsc_device *net_device,
1119                           struct vmbus_channel *channel,
1120                           const struct vmpacket_descriptor *desc,
1121                           const struct nvsp_message *nvsp)
1122 {
1123         struct net_device_context *net_device_ctx = netdev_priv(ndev);
1124         const struct vmtransfer_page_packet_header *vmxferpage_packet
1125                 = container_of(desc, const struct vmtransfer_page_packet_header, d);
1126         u16 q_idx = channel->offermsg.offer.sub_channel_index;
1127         char *recv_buf = net_device->recv_buf;
1128         u32 status = NVSP_STAT_SUCCESS;
1129         int i;
1130         int count = 0;
1131
1132         /* Make sure this is a valid nvsp packet */
1133         if (unlikely(nvsp->hdr.msg_type != NVSP_MSG1_TYPE_SEND_RNDIS_PKT)) {
1134                 netif_err(net_device_ctx, rx_err, ndev,
1135                           "Unknown nvsp packet type received %u\n",
1136                           nvsp->hdr.msg_type);
1137                 return 0;
1138         }
1139
1140         if (unlikely(vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID)) {
1141                 netif_err(net_device_ctx, rx_err, ndev,
1142                           "Invalid xfer page set id - expecting %x got %x\n",
1143                           NETVSC_RECEIVE_BUFFER_ID,
1144                           vmxferpage_packet->xfer_pageset_id);
1145                 return 0;
1146         }
1147
1148         count = vmxferpage_packet->range_cnt;
1149
1150         /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
1151         for (i = 0; i < count; i++) {
1152                 u32 offset = vmxferpage_packet->ranges[i].byte_offset;
1153                 u32 buflen = vmxferpage_packet->ranges[i].byte_count;
1154                 void *data;
1155                 int ret;
1156
1157                 if (unlikely(offset + buflen > net_device->recv_buf_size)) {
1158                         status = NVSP_STAT_FAIL;
1159                         netif_err(net_device_ctx, rx_err, ndev,
1160                                   "Packet offset:%u + len:%u too big\n",
1161                                   offset, buflen);
1162
1163                         continue;
1164                 }
1165
1166                 data = recv_buf + offset;
1167
1168                 trace_rndis_recv(ndev, q_idx, data);
1169
1170                 /* Pass it to the upper layer */
1171                 ret = rndis_filter_receive(ndev, net_device,
1172                                            channel, data, buflen);
1173
1174                 if (unlikely(ret != NVSP_STAT_SUCCESS))
1175                         status = NVSP_STAT_FAIL;
1176         }
1177
1178         enq_receive_complete(ndev, net_device, q_idx,
1179                              vmxferpage_packet->d.trans_id, status);
1180
1181         return count;
1182 }
1183
1184 static void netvsc_send_table(struct net_device *ndev,
1185                               struct netvsc_device *nvscdev,
1186                               const struct nvsp_message *nvmsg,
1187                               u32 msglen)
1188 {
1189         struct net_device_context *net_device_ctx = netdev_priv(ndev);
1190         u32 count, offset, *tab;
1191         int i;
1192
1193         count = nvmsg->msg.v5_msg.send_table.count;
1194         offset = nvmsg->msg.v5_msg.send_table.offset;
1195
1196         if (count != VRSS_SEND_TAB_SIZE) {
1197                 netdev_err(ndev, "Received wrong send-table size:%u\n", count);
1198                 return;
1199         }
1200
1201         /* If negotiated version <= NVSP_PROTOCOL_VERSION_6, the offset may be
1202          * wrong due to a host bug. So fix the offset here.
1203          */
1204         if (nvscdev->nvsp_version <= NVSP_PROTOCOL_VERSION_6 &&
1205             msglen >= sizeof(struct nvsp_message_header) +
1206             sizeof(union nvsp_6_message_uber) + count * sizeof(u32))
1207                 offset = sizeof(struct nvsp_message_header) +
1208                          sizeof(union nvsp_6_message_uber);
1209
1210         /* Boundary check for all versions */
1211         if (offset > msglen - count * sizeof(u32)) {
1212                 netdev_err(ndev, "Received send-table offset too big:%u\n",
1213                            offset);
1214                 return;
1215         }
1216
1217         tab = (void *)nvmsg + offset;
1218
1219         for (i = 0; i < count; i++)
1220                 net_device_ctx->tx_table[i] = tab[i];
1221 }
1222
1223 static void netvsc_send_vf(struct net_device *ndev,
1224                            const struct nvsp_message *nvmsg)
1225 {
1226         struct net_device_context *net_device_ctx = netdev_priv(ndev);
1227
1228         net_device_ctx->vf_alloc = nvmsg->msg.v4_msg.vf_assoc.allocated;
1229         net_device_ctx->vf_serial = nvmsg->msg.v4_msg.vf_assoc.serial;
1230
1231         if (net_device_ctx->vf_alloc)
1232                 complete(&net_device_ctx->vf_add);
1233
1234         netdev_info(ndev, "VF slot %u %s\n",
1235                     net_device_ctx->vf_serial,
1236                     net_device_ctx->vf_alloc ? "added" : "removed");
1237 }
1238
1239 static void netvsc_receive_inband(struct net_device *ndev,
1240                                   struct netvsc_device *nvscdev,
1241                                   const struct nvsp_message *nvmsg,
1242                                   u32 msglen)
1243 {
1244         switch (nvmsg->hdr.msg_type) {
1245         case NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE:
1246                 netvsc_send_table(ndev, nvscdev, nvmsg, msglen);
1247                 break;
1248
1249         case NVSP_MSG4_TYPE_SEND_VF_ASSOCIATION:
1250                 netvsc_send_vf(ndev, nvmsg);
1251                 break;
1252         }
1253 }
1254
1255 static int netvsc_process_raw_pkt(struct hv_device *device,
1256                                   struct vmbus_channel *channel,
1257                                   struct netvsc_device *net_device,
1258                                   struct net_device *ndev,
1259                                   const struct vmpacket_descriptor *desc,
1260                                   int budget)
1261 {
1262         const struct nvsp_message *nvmsg = hv_pkt_data(desc);
1263         u32 msglen = hv_pkt_datalen(desc);
1264
1265         trace_nvsp_recv(ndev, channel, nvmsg);
1266
1267         switch (desc->type) {
1268         case VM_PKT_COMP:
1269                 netvsc_send_completion(ndev, net_device, channel,
1270                                        desc, budget);
1271                 break;
1272
1273         case VM_PKT_DATA_USING_XFER_PAGES:
1274                 return netvsc_receive(ndev, net_device, channel,
1275                                       desc, nvmsg);
1276                 break;
1277
1278         case VM_PKT_DATA_INBAND:
1279                 netvsc_receive_inband(ndev, net_device, nvmsg, msglen);
1280                 break;
1281
1282         default:
1283                 netdev_err(ndev, "unhandled packet type %d, tid %llx\n",
1284                            desc->type, desc->trans_id);
1285                 break;
1286         }
1287
1288         return 0;
1289 }
1290
1291 static struct hv_device *netvsc_channel_to_device(struct vmbus_channel *channel)
1292 {
1293         struct vmbus_channel *primary = channel->primary_channel;
1294
1295         return primary ? primary->device_obj : channel->device_obj;
1296 }
1297
1298 /* Network processing softirq
1299  * Process data in incoming ring buffer from host
1300  * Stops when ring is empty or budget is met or exceeded.
1301  */
1302 int netvsc_poll(struct napi_struct *napi, int budget)
1303 {
1304         struct netvsc_channel *nvchan
1305                 = container_of(napi, struct netvsc_channel, napi);
1306         struct netvsc_device *net_device = nvchan->net_device;
1307         struct vmbus_channel *channel = nvchan->channel;
1308         struct hv_device *device = netvsc_channel_to_device(channel);
1309         struct net_device *ndev = hv_get_drvdata(device);
1310         int work_done = 0;
1311         int ret;
1312
1313         /* If starting a new interval */
1314         if (!nvchan->desc)
1315                 nvchan->desc = hv_pkt_iter_first(channel);
1316
1317         while (nvchan->desc && work_done < budget) {
1318                 work_done += netvsc_process_raw_pkt(device, channel, net_device,
1319                                                     ndev, nvchan->desc, budget);
1320                 nvchan->desc = hv_pkt_iter_next(channel, nvchan->desc);
1321         }
1322
1323         /* Send any pending receive completions */
1324         ret = send_recv_completions(ndev, net_device, nvchan);
1325
1326         /* If it did not exhaust NAPI budget this time
1327          *  and not doing busy poll
1328          * then re-enable host interrupts
1329          *  and reschedule if ring is not empty
1330          *   or sending receive completion failed.
1331          */
1332         if (work_done < budget &&
1333             napi_complete_done(napi, work_done) &&
1334             (ret || hv_end_read(&channel->inbound)) &&
1335             napi_schedule_prep(napi)) {
1336                 hv_begin_read(&channel->inbound);
1337                 __napi_schedule(napi);
1338         }
1339
1340         /* Driver may overshoot since multiple packets per descriptor */
1341         return min(work_done, budget);
1342 }
1343
1344 /* Call back when data is available in host ring buffer.
1345  * Processing is deferred until network softirq (NAPI)
1346  */
1347 void netvsc_channel_cb(void *context)
1348 {
1349         struct netvsc_channel *nvchan = context;
1350         struct vmbus_channel *channel = nvchan->channel;
1351         struct hv_ring_buffer_info *rbi = &channel->inbound;
1352
1353         /* preload first vmpacket descriptor */
1354         prefetch(hv_get_ring_buffer(rbi) + rbi->priv_read_index);
1355
1356         if (napi_schedule_prep(&nvchan->napi)) {
1357                 /* disable interupts from host */
1358                 hv_begin_read(rbi);
1359
1360                 __napi_schedule_irqoff(&nvchan->napi);
1361         }
1362 }
1363
1364 /*
1365  * netvsc_device_add - Callback when the device belonging to this
1366  * driver is added
1367  */
1368 struct netvsc_device *netvsc_device_add(struct hv_device *device,
1369                                 const struct netvsc_device_info *device_info)
1370 {
1371         int i, ret = 0;
1372         struct netvsc_device *net_device;
1373         struct net_device *ndev = hv_get_drvdata(device);
1374         struct net_device_context *net_device_ctx = netdev_priv(ndev);
1375
1376         net_device = alloc_net_device();
1377         if (!net_device)
1378                 return ERR_PTR(-ENOMEM);
1379
1380         for (i = 0; i < VRSS_SEND_TAB_SIZE; i++)
1381                 net_device_ctx->tx_table[i] = 0;
1382
1383         /* Because the device uses NAPI, all the interrupt batching and
1384          * control is done via Net softirq, not the channel handling
1385          */
1386         set_channel_read_mode(device->channel, HV_CALL_ISR);
1387
1388         /* If we're reopening the device we may have multiple queues, fill the
1389          * chn_table with the default channel to use it before subchannels are
1390          * opened.
1391          * Initialize the channel state before we open;
1392          * we can be interrupted as soon as we open the channel.
1393          */
1394
1395         for (i = 0; i < VRSS_CHANNEL_MAX; i++) {
1396                 struct netvsc_channel *nvchan = &net_device->chan_table[i];
1397
1398                 nvchan->channel = device->channel;
1399                 nvchan->net_device = net_device;
1400                 u64_stats_init(&nvchan->tx_stats.syncp);
1401                 u64_stats_init(&nvchan->rx_stats.syncp);
1402         }
1403
1404         /* Enable NAPI handler before init callbacks */
1405         netif_napi_add(ndev, &net_device->chan_table[0].napi,
1406                        netvsc_poll, NAPI_POLL_WEIGHT);
1407
1408         /* Open the channel */
1409         ret = vmbus_open(device->channel, netvsc_ring_bytes,
1410                          netvsc_ring_bytes,  NULL, 0,
1411                          netvsc_channel_cb, net_device->chan_table);
1412
1413         if (ret != 0) {
1414                 netdev_err(ndev, "unable to open channel: %d\n", ret);
1415                 goto cleanup;
1416         }
1417
1418         /* Channel is opened */
1419         netdev_dbg(ndev, "hv_netvsc channel opened successfully\n");
1420
1421         napi_enable(&net_device->chan_table[0].napi);
1422
1423         /* Connect with the NetVsp */
1424         ret = netvsc_connect_vsp(device, net_device, device_info);
1425         if (ret != 0) {
1426                 netdev_err(ndev,
1427                         "unable to connect to NetVSP - %d\n", ret);
1428                 goto close;
1429         }
1430
1431         /* Writing nvdev pointer unlocks netvsc_send(), make sure chn_table is
1432          * populated.
1433          */
1434         rcu_assign_pointer(net_device_ctx->nvdev, net_device);
1435
1436         return net_device;
1437
1438 close:
1439         RCU_INIT_POINTER(net_device_ctx->nvdev, NULL);
1440         napi_disable(&net_device->chan_table[0].napi);
1441
1442         /* Now, we can close the channel safely */
1443         vmbus_close(device->channel);
1444
1445 cleanup:
1446         netif_napi_del(&net_device->chan_table[0].napi);
1447         free_netvsc_device(&net_device->rcu);
1448
1449         return ERR_PTR(ret);
1450 }