Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (c) 2009, Microsoft Corporation.
4 *
5 * Authors:
6 * Haiyang Zhang <haiyangz@microsoft.com>
7 * Hank Janssen <hjanssen@microsoft.com>
8 */
9#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10
11#include <linux/kernel.h>
12#include <linux/sched.h>
13#include <linux/wait.h>
14#include <linux/mm.h>
15#include <linux/delay.h>
16#include <linux/io.h>
17#include <linux/slab.h>
18#include <linux/netdevice.h>
19#include <linux/if_ether.h>
20#include <linux/vmalloc.h>
21#include <linux/rtnetlink.h>
22#include <linux/prefetch.h>
23
24#include <asm/sync_bitops.h>
25
26#include "hyperv_net.h"
27#include "netvsc_trace.h"
28
29/*
30 * Switch the data path from the synthetic interface to the VF
31 * interface.
32 */
33void netvsc_switch_datapath(struct net_device *ndev, bool vf)
34{
35 struct net_device_context *net_device_ctx = netdev_priv(ndev);
36 struct hv_device *dev = net_device_ctx->device_ctx;
37 struct netvsc_device *nv_dev = rtnl_dereference(net_device_ctx->nvdev);
38 struct nvsp_message *init_pkt = &nv_dev->channel_init_pkt;
39
40 memset(init_pkt, 0, sizeof(struct nvsp_message));
41 init_pkt->hdr.msg_type = NVSP_MSG4_TYPE_SWITCH_DATA_PATH;
42 if (vf)
43 init_pkt->msg.v4_msg.active_dp.active_datapath =
44 NVSP_DATAPATH_VF;
45 else
46 init_pkt->msg.v4_msg.active_dp.active_datapath =
47 NVSP_DATAPATH_SYNTHETIC;
48
49 trace_nvsp_send(ndev, init_pkt);
50
51 vmbus_sendpacket(dev->channel, init_pkt,
52 sizeof(struct nvsp_message),
53 (unsigned long)init_pkt,
54 VM_PKT_DATA_INBAND, 0);
55}
56
57/* Worker to setup sub channels on initial setup
58 * Initial hotplug event occurs in softirq context
59 * and can't wait for channels.
60 */
61static void netvsc_subchan_work(struct work_struct *w)
62{
63 struct netvsc_device *nvdev =
64 container_of(w, struct netvsc_device, subchan_work);
65 struct rndis_device *rdev;
66 int i, ret;
67
68 /* Avoid deadlock with device removal already under RTNL */
69 if (!rtnl_trylock()) {
70 schedule_work(w);
71 return;
72 }
73
74 rdev = nvdev->extension;
75 if (rdev) {
76 ret = rndis_set_subchannel(rdev->ndev, nvdev, NULL);
77 if (ret == 0) {
78 netif_device_attach(rdev->ndev);
79 } else {
80 /* fallback to only primary channel */
81 for (i = 1; i < nvdev->num_chn; i++)
82 netif_napi_del(&nvdev->chan_table[i].napi);
83
84 nvdev->max_chn = 1;
85 nvdev->num_chn = 1;
86 }
87 }
88
89 rtnl_unlock();
90}
91
92static struct netvsc_device *alloc_net_device(void)
93{
94 struct netvsc_device *net_device;
95
96 net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
97 if (!net_device)
98 return NULL;
99
100 init_waitqueue_head(&net_device->wait_drain);
101 net_device->destroy = false;
102 net_device->tx_disable = false;
103
104 net_device->max_pkt = RNDIS_MAX_PKT_DEFAULT;
105 net_device->pkt_align = RNDIS_PKT_ALIGN_DEFAULT;
106
107 init_completion(&net_device->channel_init_wait);
108 init_waitqueue_head(&net_device->subchan_open);
109 INIT_WORK(&net_device->subchan_work, netvsc_subchan_work);
110
111 return net_device;
112}
113
114static void free_netvsc_device(struct rcu_head *head)
115{
116 struct netvsc_device *nvdev
117 = container_of(head, struct netvsc_device, rcu);
118 int i;
119
120 kfree(nvdev->extension);
121 vfree(nvdev->recv_buf);
122 vfree(nvdev->send_buf);
123 kfree(nvdev->send_section_map);
124
125 for (i = 0; i < VRSS_CHANNEL_MAX; i++)
126 vfree(nvdev->chan_table[i].mrc.slots);
127
128 kfree(nvdev);
129}
130
131static void free_netvsc_device_rcu(struct netvsc_device *nvdev)
132{
133 call_rcu(&nvdev->rcu, free_netvsc_device);
134}
135
136static void netvsc_revoke_recv_buf(struct hv_device *device,
137 struct netvsc_device *net_device,
138 struct net_device *ndev)
139{
140 struct nvsp_message *revoke_packet;
141 int ret;
142
143 /*
144 * If we got a section count, it means we received a
145 * SendReceiveBufferComplete msg (ie sent
146 * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
147 * to send a revoke msg here
148 */
149 if (net_device->recv_section_cnt) {
150 /* Send the revoke receive buffer */
151 revoke_packet = &net_device->revoke_packet;
152 memset(revoke_packet, 0, sizeof(struct nvsp_message));
153
154 revoke_packet->hdr.msg_type =
155 NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
156 revoke_packet->msg.v1_msg.
157 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
158
159 trace_nvsp_send(ndev, revoke_packet);
160
161 ret = vmbus_sendpacket(device->channel,
162 revoke_packet,
163 sizeof(struct nvsp_message),
164 (unsigned long)revoke_packet,
165 VM_PKT_DATA_INBAND, 0);
166 /* If the failure is because the channel is rescinded;
167 * ignore the failure since we cannot send on a rescinded
168 * channel. This would allow us to properly cleanup
169 * even when the channel is rescinded.
170 */
171 if (device->channel->rescind)
172 ret = 0;
173 /*
174 * If we failed here, we might as well return and
175 * have a leak rather than continue and a bugchk
176 */
177 if (ret != 0) {
178 netdev_err(ndev, "unable to send "
179 "revoke receive buffer to netvsp\n");
180 return;
181 }
182 net_device->recv_section_cnt = 0;
183 }
184}
185
186static void netvsc_revoke_send_buf(struct hv_device *device,
187 struct netvsc_device *net_device,
188 struct net_device *ndev)
189{
190 struct nvsp_message *revoke_packet;
191 int ret;
192
193 /* Deal with the send buffer we may have setup.
194 * If we got a send section size, it means we received a
195 * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
196 * NVSP_MSG1_TYPE_SEND_SEND_BUF msg) therefore, we need
197 * to send a revoke msg here
198 */
199 if (net_device->send_section_cnt) {
200 /* Send the revoke receive buffer */
201 revoke_packet = &net_device->revoke_packet;
202 memset(revoke_packet, 0, sizeof(struct nvsp_message));
203
204 revoke_packet->hdr.msg_type =
205 NVSP_MSG1_TYPE_REVOKE_SEND_BUF;
206 revoke_packet->msg.v1_msg.revoke_send_buf.id =
207 NETVSC_SEND_BUFFER_ID;
208
209 trace_nvsp_send(ndev, revoke_packet);
210
211 ret = vmbus_sendpacket(device->channel,
212 revoke_packet,
213 sizeof(struct nvsp_message),
214 (unsigned long)revoke_packet,
215 VM_PKT_DATA_INBAND, 0);
216
217 /* If the failure is because the channel is rescinded;
218 * ignore the failure since we cannot send on a rescinded
219 * channel. This would allow us to properly cleanup
220 * even when the channel is rescinded.
221 */
222 if (device->channel->rescind)
223 ret = 0;
224
225 /* If we failed here, we might as well return and
226 * have a leak rather than continue and a bugchk
227 */
228 if (ret != 0) {
229 netdev_err(ndev, "unable to send "
230 "revoke send buffer to netvsp\n");
231 return;
232 }
233 net_device->send_section_cnt = 0;
234 }
235}
236
237static void netvsc_teardown_recv_gpadl(struct hv_device *device,
238 struct netvsc_device *net_device,
239 struct net_device *ndev)
240{
241 int ret;
242
243 if (net_device->recv_buf_gpadl_handle) {
244 ret = vmbus_teardown_gpadl(device->channel,
245 net_device->recv_buf_gpadl_handle);
246
247 /* If we failed here, we might as well return and have a leak
248 * rather than continue and a bugchk
249 */
250 if (ret != 0) {
251 netdev_err(ndev,
252 "unable to teardown receive buffer's gpadl\n");
253 return;
254 }
255 net_device->recv_buf_gpadl_handle = 0;
256 }
257}
258
259static void netvsc_teardown_send_gpadl(struct hv_device *device,
260 struct netvsc_device *net_device,
261 struct net_device *ndev)
262{
263 int ret;
264
265 if (net_device->send_buf_gpadl_handle) {
266 ret = vmbus_teardown_gpadl(device->channel,
267 net_device->send_buf_gpadl_handle);
268
269 /* If we failed here, we might as well return and have a leak
270 * rather than continue and a bugchk
271 */
272 if (ret != 0) {
273 netdev_err(ndev,
274 "unable to teardown send buffer's gpadl\n");
275 return;
276 }
277 net_device->send_buf_gpadl_handle = 0;
278 }
279}
280
281int netvsc_alloc_recv_comp_ring(struct netvsc_device *net_device, u32 q_idx)
282{
283 struct netvsc_channel *nvchan = &net_device->chan_table[q_idx];
284 int node = cpu_to_node(nvchan->channel->target_cpu);
285 size_t size;
286
287 size = net_device->recv_completion_cnt * sizeof(struct recv_comp_data);
288 nvchan->mrc.slots = vzalloc_node(size, node);
289 if (!nvchan->mrc.slots)
290 nvchan->mrc.slots = vzalloc(size);
291
292 return nvchan->mrc.slots ? 0 : -ENOMEM;
293}
294
295static int netvsc_init_buf(struct hv_device *device,
296 struct netvsc_device *net_device,
297 const struct netvsc_device_info *device_info)
298{
299 struct nvsp_1_message_send_receive_buffer_complete *resp;
300 struct net_device *ndev = hv_get_drvdata(device);
301 struct nvsp_message *init_packet;
302 unsigned int buf_size;
303 size_t map_words;
304 int ret = 0;
305
306 /* Get receive buffer area. */
307 buf_size = device_info->recv_sections * device_info->recv_section_size;
308 buf_size = roundup(buf_size, PAGE_SIZE);
309
310 /* Legacy hosts only allow smaller receive buffer */
311 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
312 buf_size = min_t(unsigned int, buf_size,
313 NETVSC_RECEIVE_BUFFER_SIZE_LEGACY);
314
315 net_device->recv_buf = vzalloc(buf_size);
316 if (!net_device->recv_buf) {
317 netdev_err(ndev,
318 "unable to allocate receive buffer of size %u\n",
319 buf_size);
320 ret = -ENOMEM;
321 goto cleanup;
322 }
323
324 net_device->recv_buf_size = buf_size;
325
326 /*
327 * Establish the gpadl handle for this buffer on this
328 * channel. Note: This call uses the vmbus connection rather
329 * than the channel to establish the gpadl handle.
330 */
331 ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
332 buf_size,
333 &net_device->recv_buf_gpadl_handle);
334 if (ret != 0) {
335 netdev_err(ndev,
336 "unable to establish receive buffer's gpadl\n");
337 goto cleanup;
338 }
339
340 /* Notify the NetVsp of the gpadl handle */
341 init_packet = &net_device->channel_init_pkt;
342 memset(init_packet, 0, sizeof(struct nvsp_message));
343 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
344 init_packet->msg.v1_msg.send_recv_buf.
345 gpadl_handle = net_device->recv_buf_gpadl_handle;
346 init_packet->msg.v1_msg.
347 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
348
349 trace_nvsp_send(ndev, init_packet);
350
351 /* Send the gpadl notification request */
352 ret = vmbus_sendpacket(device->channel, init_packet,
353 sizeof(struct nvsp_message),
354 (unsigned long)init_packet,
355 VM_PKT_DATA_INBAND,
356 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
357 if (ret != 0) {
358 netdev_err(ndev,
359 "unable to send receive buffer's gpadl to netvsp\n");
360 goto cleanup;
361 }
362
363 wait_for_completion(&net_device->channel_init_wait);
364
365 /* Check the response */
366 resp = &init_packet->msg.v1_msg.send_recv_buf_complete;
367 if (resp->status != NVSP_STAT_SUCCESS) {
368 netdev_err(ndev,
369 "Unable to complete receive buffer initialization with NetVsp - status %d\n",
370 resp->status);
371 ret = -EINVAL;
372 goto cleanup;
373 }
374
375 /* Parse the response */
376 netdev_dbg(ndev, "Receive sections: %u sub_allocs: size %u count: %u\n",
377 resp->num_sections, resp->sections[0].sub_alloc_size,
378 resp->sections[0].num_sub_allocs);
379
380 /* There should only be one section for the entire receive buffer */
381 if (resp->num_sections != 1 || resp->sections[0].offset != 0) {
382 ret = -EINVAL;
383 goto cleanup;
384 }
385
386 net_device->recv_section_size = resp->sections[0].sub_alloc_size;
387 net_device->recv_section_cnt = resp->sections[0].num_sub_allocs;
388
389 /* Setup receive completion ring */
390 net_device->recv_completion_cnt
391 = round_up(net_device->recv_section_cnt + 1,
392 PAGE_SIZE / sizeof(u64));
393 ret = netvsc_alloc_recv_comp_ring(net_device, 0);
394 if (ret)
395 goto cleanup;
396
397 /* Now setup the send buffer. */
398 buf_size = device_info->send_sections * device_info->send_section_size;
399 buf_size = round_up(buf_size, PAGE_SIZE);
400
401 net_device->send_buf = vzalloc(buf_size);
402 if (!net_device->send_buf) {
403 netdev_err(ndev, "unable to allocate send buffer of size %u\n",
404 buf_size);
405 ret = -ENOMEM;
406 goto cleanup;
407 }
408
409 /* Establish the gpadl handle for this buffer on this
410 * channel. Note: This call uses the vmbus connection rather
411 * than the channel to establish the gpadl handle.
412 */
413 ret = vmbus_establish_gpadl(device->channel, net_device->send_buf,
414 buf_size,
415 &net_device->send_buf_gpadl_handle);
416 if (ret != 0) {
417 netdev_err(ndev,
418 "unable to establish send buffer's gpadl\n");
419 goto cleanup;
420 }
421
422 /* Notify the NetVsp of the gpadl handle */
423 init_packet = &net_device->channel_init_pkt;
424 memset(init_packet, 0, sizeof(struct nvsp_message));
425 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_SEND_BUF;
426 init_packet->msg.v1_msg.send_send_buf.gpadl_handle =
427 net_device->send_buf_gpadl_handle;
428 init_packet->msg.v1_msg.send_send_buf.id = NETVSC_SEND_BUFFER_ID;
429
430 trace_nvsp_send(ndev, init_packet);
431
432 /* Send the gpadl notification request */
433 ret = vmbus_sendpacket(device->channel, init_packet,
434 sizeof(struct nvsp_message),
435 (unsigned long)init_packet,
436 VM_PKT_DATA_INBAND,
437 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
438 if (ret != 0) {
439 netdev_err(ndev,
440 "unable to send send buffer's gpadl to netvsp\n");
441 goto cleanup;
442 }
443
444 wait_for_completion(&net_device->channel_init_wait);
445
446 /* Check the response */
447 if (init_packet->msg.v1_msg.
448 send_send_buf_complete.status != NVSP_STAT_SUCCESS) {
449 netdev_err(ndev, "Unable to complete send buffer "
450 "initialization with NetVsp - status %d\n",
451 init_packet->msg.v1_msg.
452 send_send_buf_complete.status);
453 ret = -EINVAL;
454 goto cleanup;
455 }
456
457 /* Parse the response */
458 net_device->send_section_size = init_packet->msg.
459 v1_msg.send_send_buf_complete.section_size;
460
461 /* Section count is simply the size divided by the section size. */
462 net_device->send_section_cnt = buf_size / net_device->send_section_size;
463
464 netdev_dbg(ndev, "Send section size: %d, Section count:%d\n",
465 net_device->send_section_size, net_device->send_section_cnt);
466
467 /* Setup state for managing the send buffer. */
468 map_words = DIV_ROUND_UP(net_device->send_section_cnt, BITS_PER_LONG);
469
470 net_device->send_section_map = kcalloc(map_words, sizeof(ulong), GFP_KERNEL);
471 if (net_device->send_section_map == NULL) {
472 ret = -ENOMEM;
473 goto cleanup;
474 }
475
476 goto exit;
477
478cleanup:
479 netvsc_revoke_recv_buf(device, net_device, ndev);
480 netvsc_revoke_send_buf(device, net_device, ndev);
481 netvsc_teardown_recv_gpadl(device, net_device, ndev);
482 netvsc_teardown_send_gpadl(device, net_device, ndev);
483
484exit:
485 return ret;
486}
487
488/* Negotiate NVSP protocol version */
489static int negotiate_nvsp_ver(struct hv_device *device,
490 struct netvsc_device *net_device,
491 struct nvsp_message *init_packet,
492 u32 nvsp_ver)
493{
494 struct net_device *ndev = hv_get_drvdata(device);
495 int ret;
496
497 memset(init_packet, 0, sizeof(struct nvsp_message));
498 init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
499 init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
500 init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
501 trace_nvsp_send(ndev, init_packet);
502
503 /* Send the init request */
504 ret = vmbus_sendpacket(device->channel, init_packet,
505 sizeof(struct nvsp_message),
506 (unsigned long)init_packet,
507 VM_PKT_DATA_INBAND,
508 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
509
510 if (ret != 0)
511 return ret;
512
513 wait_for_completion(&net_device->channel_init_wait);
514
515 if (init_packet->msg.init_msg.init_complete.status !=
516 NVSP_STAT_SUCCESS)
517 return -EINVAL;
518
519 if (nvsp_ver == NVSP_PROTOCOL_VERSION_1)
520 return 0;
521
522 /* NVSPv2 or later: Send NDIS config */
523 memset(init_packet, 0, sizeof(struct nvsp_message));
524 init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
525 init_packet->msg.v2_msg.send_ndis_config.mtu = ndev->mtu + ETH_HLEN;
526 init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
527
528 if (nvsp_ver >= NVSP_PROTOCOL_VERSION_5) {
529 init_packet->msg.v2_msg.send_ndis_config.capability.sriov = 1;
530
531 /* Teaming bit is needed to receive link speed updates */
532 init_packet->msg.v2_msg.send_ndis_config.capability.teaming = 1;
533 }
534
535 if (nvsp_ver >= NVSP_PROTOCOL_VERSION_61)
536 init_packet->msg.v2_msg.send_ndis_config.capability.rsc = 1;
537
538 trace_nvsp_send(ndev, init_packet);
539
540 ret = vmbus_sendpacket(device->channel, init_packet,
541 sizeof(struct nvsp_message),
542 (unsigned long)init_packet,
543 VM_PKT_DATA_INBAND, 0);
544
545 return ret;
546}
547
548static int netvsc_connect_vsp(struct hv_device *device,
549 struct netvsc_device *net_device,
550 const struct netvsc_device_info *device_info)
551{
552 struct net_device *ndev = hv_get_drvdata(device);
553 static const u32 ver_list[] = {
554 NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
555 NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5,
556 NVSP_PROTOCOL_VERSION_6, NVSP_PROTOCOL_VERSION_61
557 };
558 struct nvsp_message *init_packet;
559 int ndis_version, i, ret;
560
561 init_packet = &net_device->channel_init_pkt;
562
563 /* Negotiate the latest NVSP protocol supported */
564 for (i = ARRAY_SIZE(ver_list) - 1; i >= 0; i--)
565 if (negotiate_nvsp_ver(device, net_device, init_packet,
566 ver_list[i]) == 0) {
567 net_device->nvsp_version = ver_list[i];
568 break;
569 }
570
571 if (i < 0) {
572 ret = -EPROTO;
573 goto cleanup;
574 }
575
576 pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
577
578 /* Send the ndis version */
579 memset(init_packet, 0, sizeof(struct nvsp_message));
580
581 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
582 ndis_version = 0x00060001;
583 else
584 ndis_version = 0x0006001e;
585
586 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
587 init_packet->msg.v1_msg.
588 send_ndis_ver.ndis_major_ver =
589 (ndis_version & 0xFFFF0000) >> 16;
590 init_packet->msg.v1_msg.
591 send_ndis_ver.ndis_minor_ver =
592 ndis_version & 0xFFFF;
593
594 trace_nvsp_send(ndev, init_packet);
595
596 /* Send the init request */
597 ret = vmbus_sendpacket(device->channel, init_packet,
598 sizeof(struct nvsp_message),
599 (unsigned long)init_packet,
600 VM_PKT_DATA_INBAND, 0);
601 if (ret != 0)
602 goto cleanup;
603
604
605 ret = netvsc_init_buf(device, net_device, device_info);
606
607cleanup:
608 return ret;
609}
610
611/*
612 * netvsc_device_remove - Callback when the root bus device is removed
613 */
614void netvsc_device_remove(struct hv_device *device)
615{
616 struct net_device *ndev = hv_get_drvdata(device);
617 struct net_device_context *net_device_ctx = netdev_priv(ndev);
618 struct netvsc_device *net_device
619 = rtnl_dereference(net_device_ctx->nvdev);
620 int i;
621
622 /*
623 * Revoke receive buffer. If host is pre-Win2016 then tear down
624 * receive buffer GPADL. Do the same for send buffer.
625 */
626 netvsc_revoke_recv_buf(device, net_device, ndev);
627 if (vmbus_proto_version < VERSION_WIN10)
628 netvsc_teardown_recv_gpadl(device, net_device, ndev);
629
630 netvsc_revoke_send_buf(device, net_device, ndev);
631 if (vmbus_proto_version < VERSION_WIN10)
632 netvsc_teardown_send_gpadl(device, net_device, ndev);
633
634 RCU_INIT_POINTER(net_device_ctx->nvdev, NULL);
635
636 /* And disassociate NAPI context from device */
637 for (i = 0; i < net_device->num_chn; i++)
638 netif_napi_del(&net_device->chan_table[i].napi);
639
640 /*
641 * At this point, no one should be accessing net_device
642 * except in here
643 */
644 netdev_dbg(ndev, "net device safe to remove\n");
645
646 /* Now, we can close the channel safely */
647 vmbus_close(device->channel);
648
649 /*
650 * If host is Win2016 or higher then we do the GPADL tear down
651 * here after VMBus is closed.
652 */
653 if (vmbus_proto_version >= VERSION_WIN10) {
654 netvsc_teardown_recv_gpadl(device, net_device, ndev);
655 netvsc_teardown_send_gpadl(device, net_device, ndev);
656 }
657
658 /* Release all resources */
659 free_netvsc_device_rcu(net_device);
660}
661
662#define RING_AVAIL_PERCENT_HIWATER 20
663#define RING_AVAIL_PERCENT_LOWATER 10
664
665static inline void netvsc_free_send_slot(struct netvsc_device *net_device,
666 u32 index)
667{
668 sync_change_bit(index, net_device->send_section_map);
669}
670
671static void netvsc_send_tx_complete(struct net_device *ndev,
672 struct netvsc_device *net_device,
673 struct vmbus_channel *channel,
674 const struct vmpacket_descriptor *desc,
675 int budget)
676{
677 struct sk_buff *skb = (struct sk_buff *)(unsigned long)desc->trans_id;
678 struct net_device_context *ndev_ctx = netdev_priv(ndev);
679 u16 q_idx = 0;
680 int queue_sends;
681
682 /* Notify the layer above us */
683 if (likely(skb)) {
684 const struct hv_netvsc_packet *packet
685 = (struct hv_netvsc_packet *)skb->cb;
686 u32 send_index = packet->send_buf_index;
687 struct netvsc_stats *tx_stats;
688
689 if (send_index != NETVSC_INVALID_INDEX)
690 netvsc_free_send_slot(net_device, send_index);
691 q_idx = packet->q_idx;
692
693 tx_stats = &net_device->chan_table[q_idx].tx_stats;
694
695 u64_stats_update_begin(&tx_stats->syncp);
696 tx_stats->packets += packet->total_packets;
697 tx_stats->bytes += packet->total_bytes;
698 u64_stats_update_end(&tx_stats->syncp);
699
700 napi_consume_skb(skb, budget);
701 }
702
703 queue_sends =
704 atomic_dec_return(&net_device->chan_table[q_idx].queue_sends);
705
706 if (unlikely(net_device->destroy)) {
707 if (queue_sends == 0)
708 wake_up(&net_device->wait_drain);
709 } else {
710 struct netdev_queue *txq = netdev_get_tx_queue(ndev, q_idx);
711
712 if (netif_tx_queue_stopped(txq) && !net_device->tx_disable &&
713 (hv_get_avail_to_write_percent(&channel->outbound) >
714 RING_AVAIL_PERCENT_HIWATER || queue_sends < 1)) {
715 netif_tx_wake_queue(txq);
716 ndev_ctx->eth_stats.wake_queue++;
717 }
718 }
719}
720
721static void netvsc_send_completion(struct net_device *ndev,
722 struct netvsc_device *net_device,
723 struct vmbus_channel *incoming_channel,
724 const struct vmpacket_descriptor *desc,
725 int budget)
726{
727 const struct nvsp_message *nvsp_packet = hv_pkt_data(desc);
728
729 switch (nvsp_packet->hdr.msg_type) {
730 case NVSP_MSG_TYPE_INIT_COMPLETE:
731 case NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE:
732 case NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE:
733 case NVSP_MSG5_TYPE_SUBCHANNEL:
734 /* Copy the response back */
735 memcpy(&net_device->channel_init_pkt, nvsp_packet,
736 sizeof(struct nvsp_message));
737 complete(&net_device->channel_init_wait);
738 break;
739
740 case NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE:
741 netvsc_send_tx_complete(ndev, net_device, incoming_channel,
742 desc, budget);
743 break;
744
745 default:
746 netdev_err(ndev,
747 "Unknown send completion type %d received!!\n",
748 nvsp_packet->hdr.msg_type);
749 }
750}
751
752static u32 netvsc_get_next_send_section(struct netvsc_device *net_device)
753{
754 unsigned long *map_addr = net_device->send_section_map;
755 unsigned int i;
756
757 for_each_clear_bit(i, map_addr, net_device->send_section_cnt) {
758 if (sync_test_and_set_bit(i, map_addr) == 0)
759 return i;
760 }
761
762 return NETVSC_INVALID_INDEX;
763}
764
765static void netvsc_copy_to_send_buf(struct netvsc_device *net_device,
766 unsigned int section_index,
767 u32 pend_size,
768 struct hv_netvsc_packet *packet,
769 struct rndis_message *rndis_msg,
770 struct hv_page_buffer *pb,
771 bool xmit_more)
772{
773 char *start = net_device->send_buf;
774 char *dest = start + (section_index * net_device->send_section_size)
775 + pend_size;
776 int i;
777 u32 padding = 0;
778 u32 page_count = packet->cp_partial ? packet->rmsg_pgcnt :
779 packet->page_buf_cnt;
780 u32 remain;
781
782 /* Add padding */
783 remain = packet->total_data_buflen & (net_device->pkt_align - 1);
784 if (xmit_more && remain) {
785 padding = net_device->pkt_align - remain;
786 rndis_msg->msg_len += padding;
787 packet->total_data_buflen += padding;
788 }
789
790 for (i = 0; i < page_count; i++) {
791 char *src = phys_to_virt(pb[i].pfn << PAGE_SHIFT);
792 u32 offset = pb[i].offset;
793 u32 len = pb[i].len;
794
795 memcpy(dest, (src + offset), len);
796 dest += len;
797 }
798
799 if (padding)
800 memset(dest, 0, padding);
801}
802
803static inline int netvsc_send_pkt(
804 struct hv_device *device,
805 struct hv_netvsc_packet *packet,
806 struct netvsc_device *net_device,
807 struct hv_page_buffer *pb,
808 struct sk_buff *skb)
809{
810 struct nvsp_message nvmsg;
811 struct nvsp_1_message_send_rndis_packet *rpkt =
812 &nvmsg.msg.v1_msg.send_rndis_pkt;
813 struct netvsc_channel * const nvchan =
814 &net_device->chan_table[packet->q_idx];
815 struct vmbus_channel *out_channel = nvchan->channel;
816 struct net_device *ndev = hv_get_drvdata(device);
817 struct net_device_context *ndev_ctx = netdev_priv(ndev);
818 struct netdev_queue *txq = netdev_get_tx_queue(ndev, packet->q_idx);
819 u64 req_id;
820 int ret;
821 u32 ring_avail = hv_get_avail_to_write_percent(&out_channel->outbound);
822
823 nvmsg.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
824 if (skb)
825 rpkt->channel_type = 0; /* 0 is RMC_DATA */
826 else
827 rpkt->channel_type = 1; /* 1 is RMC_CONTROL */
828
829 rpkt->send_buf_section_index = packet->send_buf_index;
830 if (packet->send_buf_index == NETVSC_INVALID_INDEX)
831 rpkt->send_buf_section_size = 0;
832 else
833 rpkt->send_buf_section_size = packet->total_data_buflen;
834
835 req_id = (ulong)skb;
836
837 if (out_channel->rescind)
838 return -ENODEV;
839
840 trace_nvsp_send_pkt(ndev, out_channel, rpkt);
841
842 if (packet->page_buf_cnt) {
843 if (packet->cp_partial)
844 pb += packet->rmsg_pgcnt;
845
846 ret = vmbus_sendpacket_pagebuffer(out_channel,
847 pb, packet->page_buf_cnt,
848 &nvmsg, sizeof(nvmsg),
849 req_id);
850 } else {
851 ret = vmbus_sendpacket(out_channel,
852 &nvmsg, sizeof(nvmsg),
853 req_id, VM_PKT_DATA_INBAND,
854 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
855 }
856
857 if (ret == 0) {
858 atomic_inc_return(&nvchan->queue_sends);
859
860 if (ring_avail < RING_AVAIL_PERCENT_LOWATER) {
861 netif_tx_stop_queue(txq);
862 ndev_ctx->eth_stats.stop_queue++;
863 }
864 } else if (ret == -EAGAIN) {
865 netif_tx_stop_queue(txq);
866 ndev_ctx->eth_stats.stop_queue++;
867 } else {
868 netdev_err(ndev,
869 "Unable to send packet pages %u len %u, ret %d\n",
870 packet->page_buf_cnt, packet->total_data_buflen,
871 ret);
872 }
873
874 if (netif_tx_queue_stopped(txq) &&
875 atomic_read(&nvchan->queue_sends) < 1 &&
876 !net_device->tx_disable) {
877 netif_tx_wake_queue(txq);
878 ndev_ctx->eth_stats.wake_queue++;
879 if (ret == -EAGAIN)
880 ret = -ENOSPC;
881 }
882
883 return ret;
884}
885
886/* Move packet out of multi send data (msd), and clear msd */
887static inline void move_pkt_msd(struct hv_netvsc_packet **msd_send,
888 struct sk_buff **msd_skb,
889 struct multi_send_data *msdp)
890{
891 *msd_skb = msdp->skb;
892 *msd_send = msdp->pkt;
893 msdp->skb = NULL;
894 msdp->pkt = NULL;
895 msdp->count = 0;
896}
897
898/* RCU already held by caller */
899int netvsc_send(struct net_device *ndev,
900 struct hv_netvsc_packet *packet,
901 struct rndis_message *rndis_msg,
902 struct hv_page_buffer *pb,
903 struct sk_buff *skb)
904{
905 struct net_device_context *ndev_ctx = netdev_priv(ndev);
906 struct netvsc_device *net_device
907 = rcu_dereference_bh(ndev_ctx->nvdev);
908 struct hv_device *device = ndev_ctx->device_ctx;
909 int ret = 0;
910 struct netvsc_channel *nvchan;
911 u32 pktlen = packet->total_data_buflen, msd_len = 0;
912 unsigned int section_index = NETVSC_INVALID_INDEX;
913 struct multi_send_data *msdp;
914 struct hv_netvsc_packet *msd_send = NULL, *cur_send = NULL;
915 struct sk_buff *msd_skb = NULL;
916 bool try_batch, xmit_more;
917
918 /* If device is rescinded, return error and packet will get dropped. */
919 if (unlikely(!net_device || net_device->destroy))
920 return -ENODEV;
921
922 nvchan = &net_device->chan_table[packet->q_idx];
923 packet->send_buf_index = NETVSC_INVALID_INDEX;
924 packet->cp_partial = false;
925
926 /* Send control message directly without accessing msd (Multi-Send
927 * Data) field which may be changed during data packet processing.
928 */
929 if (!skb)
930 return netvsc_send_pkt(device, packet, net_device, pb, skb);
931
932 /* batch packets in send buffer if possible */
933 msdp = &nvchan->msd;
934 if (msdp->pkt)
935 msd_len = msdp->pkt->total_data_buflen;
936
937 try_batch = msd_len > 0 && msdp->count < net_device->max_pkt;
938 if (try_batch && msd_len + pktlen + net_device->pkt_align <
939 net_device->send_section_size) {
940 section_index = msdp->pkt->send_buf_index;
941
942 } else if (try_batch && msd_len + packet->rmsg_size <
943 net_device->send_section_size) {
944 section_index = msdp->pkt->send_buf_index;
945 packet->cp_partial = true;
946
947 } else if (pktlen + net_device->pkt_align <
948 net_device->send_section_size) {
949 section_index = netvsc_get_next_send_section(net_device);
950 if (unlikely(section_index == NETVSC_INVALID_INDEX)) {
951 ++ndev_ctx->eth_stats.tx_send_full;
952 } else {
953 move_pkt_msd(&msd_send, &msd_skb, msdp);
954 msd_len = 0;
955 }
956 }
957
958 /* Keep aggregating only if stack says more data is coming
959 * and not doing mixed modes send and not flow blocked
960 */
961 xmit_more = netdev_xmit_more() &&
962 !packet->cp_partial &&
963 !netif_xmit_stopped(netdev_get_tx_queue(ndev, packet->q_idx));
964
965 if (section_index != NETVSC_INVALID_INDEX) {
966 netvsc_copy_to_send_buf(net_device,
967 section_index, msd_len,
968 packet, rndis_msg, pb, xmit_more);
969
970 packet->send_buf_index = section_index;
971
972 if (packet->cp_partial) {
973 packet->page_buf_cnt -= packet->rmsg_pgcnt;
974 packet->total_data_buflen = msd_len + packet->rmsg_size;
975 } else {
976 packet->page_buf_cnt = 0;
977 packet->total_data_buflen += msd_len;
978 }
979
980 if (msdp->pkt) {
981 packet->total_packets += msdp->pkt->total_packets;
982 packet->total_bytes += msdp->pkt->total_bytes;
983 }
984
985 if (msdp->skb)
986 dev_consume_skb_any(msdp->skb);
987
988 if (xmit_more) {
989 msdp->skb = skb;
990 msdp->pkt = packet;
991 msdp->count++;
992 } else {
993 cur_send = packet;
994 msdp->skb = NULL;
995 msdp->pkt = NULL;
996 msdp->count = 0;
997 }
998 } else {
999 move_pkt_msd(&msd_send, &msd_skb, msdp);
1000 cur_send = packet;
1001 }
1002
1003 if (msd_send) {
1004 int m_ret = netvsc_send_pkt(device, msd_send, net_device,
1005 NULL, msd_skb);
1006
1007 if (m_ret != 0) {
1008 netvsc_free_send_slot(net_device,
1009 msd_send->send_buf_index);
1010 dev_kfree_skb_any(msd_skb);
1011 }
1012 }
1013
1014 if (cur_send)
1015 ret = netvsc_send_pkt(device, cur_send, net_device, pb, skb);
1016
1017 if (ret != 0 && section_index != NETVSC_INVALID_INDEX)
1018 netvsc_free_send_slot(net_device, section_index);
1019
1020 return ret;
1021}
1022
1023/* Send pending recv completions */
1024static int send_recv_completions(struct net_device *ndev,
1025 struct netvsc_device *nvdev,
1026 struct netvsc_channel *nvchan)
1027{
1028 struct multi_recv_comp *mrc = &nvchan->mrc;
1029 struct recv_comp_msg {
1030 struct nvsp_message_header hdr;
1031 u32 status;
1032 } __packed;
1033 struct recv_comp_msg msg = {
1034 .hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE,
1035 };
1036 int ret;
1037
1038 while (mrc->first != mrc->next) {
1039 const struct recv_comp_data *rcd
1040 = mrc->slots + mrc->first;
1041
1042 msg.status = rcd->status;
1043 ret = vmbus_sendpacket(nvchan->channel, &msg, sizeof(msg),
1044 rcd->tid, VM_PKT_COMP, 0);
1045 if (unlikely(ret)) {
1046 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1047
1048 ++ndev_ctx->eth_stats.rx_comp_busy;
1049 return ret;
1050 }
1051
1052 if (++mrc->first == nvdev->recv_completion_cnt)
1053 mrc->first = 0;
1054 }
1055
1056 /* receive completion ring has been emptied */
1057 if (unlikely(nvdev->destroy))
1058 wake_up(&nvdev->wait_drain);
1059
1060 return 0;
1061}
1062
1063/* Count how many receive completions are outstanding */
1064static void recv_comp_slot_avail(const struct netvsc_device *nvdev,
1065 const struct multi_recv_comp *mrc,
1066 u32 *filled, u32 *avail)
1067{
1068 u32 count = nvdev->recv_completion_cnt;
1069
1070 if (mrc->next >= mrc->first)
1071 *filled = mrc->next - mrc->first;
1072 else
1073 *filled = (count - mrc->first) + mrc->next;
1074
1075 *avail = count - *filled - 1;
1076}
1077
1078/* Add receive complete to ring to send to host. */
1079static void enq_receive_complete(struct net_device *ndev,
1080 struct netvsc_device *nvdev, u16 q_idx,
1081 u64 tid, u32 status)
1082{
1083 struct netvsc_channel *nvchan = &nvdev->chan_table[q_idx];
1084 struct multi_recv_comp *mrc = &nvchan->mrc;
1085 struct recv_comp_data *rcd;
1086 u32 filled, avail;
1087
1088 recv_comp_slot_avail(nvdev, mrc, &filled, &avail);
1089
1090 if (unlikely(filled > NAPI_POLL_WEIGHT)) {
1091 send_recv_completions(ndev, nvdev, nvchan);
1092 recv_comp_slot_avail(nvdev, mrc, &filled, &avail);
1093 }
1094
1095 if (unlikely(!avail)) {
1096 netdev_err(ndev, "Recv_comp full buf q:%hd, tid:%llx\n",
1097 q_idx, tid);
1098 return;
1099 }
1100
1101 rcd = mrc->slots + mrc->next;
1102 rcd->tid = tid;
1103 rcd->status = status;
1104
1105 if (++mrc->next == nvdev->recv_completion_cnt)
1106 mrc->next = 0;
1107}
1108
1109static int netvsc_receive(struct net_device *ndev,
1110 struct netvsc_device *net_device,
1111 struct netvsc_channel *nvchan,
1112 const struct vmpacket_descriptor *desc,
1113 const struct nvsp_message *nvsp)
1114{
1115 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1116 struct vmbus_channel *channel = nvchan->channel;
1117 const struct vmtransfer_page_packet_header *vmxferpage_packet
1118 = container_of(desc, const struct vmtransfer_page_packet_header, d);
1119 u16 q_idx = channel->offermsg.offer.sub_channel_index;
1120 char *recv_buf = net_device->recv_buf;
1121 u32 status = NVSP_STAT_SUCCESS;
1122 int i;
1123 int count = 0;
1124
1125 /* Make sure this is a valid nvsp packet */
1126 if (unlikely(nvsp->hdr.msg_type != NVSP_MSG1_TYPE_SEND_RNDIS_PKT)) {
1127 netif_err(net_device_ctx, rx_err, ndev,
1128 "Unknown nvsp packet type received %u\n",
1129 nvsp->hdr.msg_type);
1130 return 0;
1131 }
1132
1133 if (unlikely(vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID)) {
1134 netif_err(net_device_ctx, rx_err, ndev,
1135 "Invalid xfer page set id - expecting %x got %x\n",
1136 NETVSC_RECEIVE_BUFFER_ID,
1137 vmxferpage_packet->xfer_pageset_id);
1138 return 0;
1139 }
1140
1141 count = vmxferpage_packet->range_cnt;
1142
1143 /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
1144 for (i = 0; i < count; i++) {
1145 u32 offset = vmxferpage_packet->ranges[i].byte_offset;
1146 u32 buflen = vmxferpage_packet->ranges[i].byte_count;
1147 void *data;
1148 int ret;
1149
1150 if (unlikely(offset + buflen > net_device->recv_buf_size)) {
1151 nvchan->rsc.cnt = 0;
1152 status = NVSP_STAT_FAIL;
1153 netif_err(net_device_ctx, rx_err, ndev,
1154 "Packet offset:%u + len:%u too big\n",
1155 offset, buflen);
1156
1157 continue;
1158 }
1159
1160 data = recv_buf + offset;
1161
1162 nvchan->rsc.is_last = (i == count - 1);
1163
1164 trace_rndis_recv(ndev, q_idx, data);
1165
1166 /* Pass it to the upper layer */
1167 ret = rndis_filter_receive(ndev, net_device,
1168 nvchan, data, buflen);
1169
1170 if (unlikely(ret != NVSP_STAT_SUCCESS))
1171 status = NVSP_STAT_FAIL;
1172 }
1173
1174 enq_receive_complete(ndev, net_device, q_idx,
1175 vmxferpage_packet->d.trans_id, status);
1176
1177 return count;
1178}
1179
1180static void netvsc_send_table(struct net_device *ndev,
1181 struct netvsc_device *nvscdev,
1182 const struct nvsp_message *nvmsg,
1183 u32 msglen)
1184{
1185 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1186 u32 count, offset, *tab;
1187 int i;
1188
1189 count = nvmsg->msg.v5_msg.send_table.count;
1190 offset = nvmsg->msg.v5_msg.send_table.offset;
1191
1192 if (count != VRSS_SEND_TAB_SIZE) {
1193 netdev_err(ndev, "Received wrong send-table size:%u\n", count);
1194 return;
1195 }
1196
1197 /* If negotiated version <= NVSP_PROTOCOL_VERSION_6, the offset may be
1198 * wrong due to a host bug. So fix the offset here.
1199 */
1200 if (nvscdev->nvsp_version <= NVSP_PROTOCOL_VERSION_6 &&
1201 msglen >= sizeof(struct nvsp_message_header) +
1202 sizeof(union nvsp_6_message_uber) + count * sizeof(u32))
1203 offset = sizeof(struct nvsp_message_header) +
1204 sizeof(union nvsp_6_message_uber);
1205
1206 /* Boundary check for all versions */
1207 if (offset > msglen - count * sizeof(u32)) {
1208 netdev_err(ndev, "Received send-table offset too big:%u\n",
1209 offset);
1210 return;
1211 }
1212
1213 tab = (void *)nvmsg + offset;
1214
1215 for (i = 0; i < count; i++)
1216 net_device_ctx->tx_table[i] = tab[i];
1217}
1218
1219static void netvsc_send_vf(struct net_device *ndev,
1220 const struct nvsp_message *nvmsg)
1221{
1222 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1223
1224 net_device_ctx->vf_alloc = nvmsg->msg.v4_msg.vf_assoc.allocated;
1225 net_device_ctx->vf_serial = nvmsg->msg.v4_msg.vf_assoc.serial;
1226 netdev_info(ndev, "VF slot %u %s\n",
1227 net_device_ctx->vf_serial,
1228 net_device_ctx->vf_alloc ? "added" : "removed");
1229}
1230
1231static void netvsc_receive_inband(struct net_device *ndev,
1232 struct netvsc_device *nvscdev,
1233 const struct nvsp_message *nvmsg,
1234 u32 msglen)
1235{
1236 switch (nvmsg->hdr.msg_type) {
1237 case NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE:
1238 netvsc_send_table(ndev, nvscdev, nvmsg, msglen);
1239 break;
1240
1241 case NVSP_MSG4_TYPE_SEND_VF_ASSOCIATION:
1242 netvsc_send_vf(ndev, nvmsg);
1243 break;
1244 }
1245}
1246
1247static int netvsc_process_raw_pkt(struct hv_device *device,
1248 struct netvsc_channel *nvchan,
1249 struct netvsc_device *net_device,
1250 struct net_device *ndev,
1251 const struct vmpacket_descriptor *desc,
1252 int budget)
1253{
1254 struct vmbus_channel *channel = nvchan->channel;
1255 const struct nvsp_message *nvmsg = hv_pkt_data(desc);
1256 u32 msglen = hv_pkt_datalen(desc);
1257
1258 trace_nvsp_recv(ndev, channel, nvmsg);
1259
1260 switch (desc->type) {
1261 case VM_PKT_COMP:
1262 netvsc_send_completion(ndev, net_device, channel,
1263 desc, budget);
1264 break;
1265
1266 case VM_PKT_DATA_USING_XFER_PAGES:
1267 return netvsc_receive(ndev, net_device, nvchan,
1268 desc, nvmsg);
1269 break;
1270
1271 case VM_PKT_DATA_INBAND:
1272 netvsc_receive_inband(ndev, net_device, nvmsg, msglen);
1273 break;
1274
1275 default:
1276 netdev_err(ndev, "unhandled packet type %d, tid %llx\n",
1277 desc->type, desc->trans_id);
1278 break;
1279 }
1280
1281 return 0;
1282}
1283
1284static struct hv_device *netvsc_channel_to_device(struct vmbus_channel *channel)
1285{
1286 struct vmbus_channel *primary = channel->primary_channel;
1287
1288 return primary ? primary->device_obj : channel->device_obj;
1289}
1290
1291/* Network processing softirq
1292 * Process data in incoming ring buffer from host
1293 * Stops when ring is empty or budget is met or exceeded.
1294 */
1295int netvsc_poll(struct napi_struct *napi, int budget)
1296{
1297 struct netvsc_channel *nvchan
1298 = container_of(napi, struct netvsc_channel, napi);
1299 struct netvsc_device *net_device = nvchan->net_device;
1300 struct vmbus_channel *channel = nvchan->channel;
1301 struct hv_device *device = netvsc_channel_to_device(channel);
1302 struct net_device *ndev = hv_get_drvdata(device);
1303 int work_done = 0;
1304 int ret;
1305
1306 /* If starting a new interval */
1307 if (!nvchan->desc)
1308 nvchan->desc = hv_pkt_iter_first(channel);
1309
1310 while (nvchan->desc && work_done < budget) {
1311 work_done += netvsc_process_raw_pkt(device, nvchan, net_device,
1312 ndev, nvchan->desc, budget);
1313 nvchan->desc = hv_pkt_iter_next(channel, nvchan->desc);
1314 }
1315
1316 /* Send any pending receive completions */
1317 ret = send_recv_completions(ndev, net_device, nvchan);
1318
1319 /* If it did not exhaust NAPI budget this time
1320 * and not doing busy poll
1321 * then re-enable host interrupts
1322 * and reschedule if ring is not empty
1323 * or sending receive completion failed.
1324 */
1325 if (work_done < budget &&
1326 napi_complete_done(napi, work_done) &&
1327 (ret || hv_end_read(&channel->inbound)) &&
1328 napi_schedule_prep(napi)) {
1329 hv_begin_read(&channel->inbound);
1330 __napi_schedule(napi);
1331 }
1332
1333 /* Driver may overshoot since multiple packets per descriptor */
1334 return min(work_done, budget);
1335}
1336
1337/* Call back when data is available in host ring buffer.
1338 * Processing is deferred until network softirq (NAPI)
1339 */
1340void netvsc_channel_cb(void *context)
1341{
1342 struct netvsc_channel *nvchan = context;
1343 struct vmbus_channel *channel = nvchan->channel;
1344 struct hv_ring_buffer_info *rbi = &channel->inbound;
1345
1346 /* preload first vmpacket descriptor */
1347 prefetch(hv_get_ring_buffer(rbi) + rbi->priv_read_index);
1348
1349 if (napi_schedule_prep(&nvchan->napi)) {
1350 /* disable interrupts from host */
1351 hv_begin_read(rbi);
1352
1353 __napi_schedule_irqoff(&nvchan->napi);
1354 }
1355}
1356
1357/*
1358 * netvsc_device_add - Callback when the device belonging to this
1359 * driver is added
1360 */
1361struct netvsc_device *netvsc_device_add(struct hv_device *device,
1362 const struct netvsc_device_info *device_info)
1363{
1364 int i, ret = 0;
1365 struct netvsc_device *net_device;
1366 struct net_device *ndev = hv_get_drvdata(device);
1367 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1368
1369 net_device = alloc_net_device();
1370 if (!net_device)
1371 return ERR_PTR(-ENOMEM);
1372
1373 for (i = 0; i < VRSS_SEND_TAB_SIZE; i++)
1374 net_device_ctx->tx_table[i] = 0;
1375
1376 /* Because the device uses NAPI, all the interrupt batching and
1377 * control is done via Net softirq, not the channel handling
1378 */
1379 set_channel_read_mode(device->channel, HV_CALL_ISR);
1380
1381 /* If we're reopening the device we may have multiple queues, fill the
1382 * chn_table with the default channel to use it before subchannels are
1383 * opened.
1384 * Initialize the channel state before we open;
1385 * we can be interrupted as soon as we open the channel.
1386 */
1387
1388 for (i = 0; i < VRSS_CHANNEL_MAX; i++) {
1389 struct netvsc_channel *nvchan = &net_device->chan_table[i];
1390
1391 nvchan->channel = device->channel;
1392 nvchan->net_device = net_device;
1393 u64_stats_init(&nvchan->tx_stats.syncp);
1394 u64_stats_init(&nvchan->rx_stats.syncp);
1395 }
1396
1397 /* Enable NAPI handler before init callbacks */
1398 netif_napi_add(ndev, &net_device->chan_table[0].napi,
1399 netvsc_poll, NAPI_POLL_WEIGHT);
1400
1401 /* Open the channel */
1402 ret = vmbus_open(device->channel, netvsc_ring_bytes,
1403 netvsc_ring_bytes, NULL, 0,
1404 netvsc_channel_cb, net_device->chan_table);
1405
1406 if (ret != 0) {
1407 netdev_err(ndev, "unable to open channel: %d\n", ret);
1408 goto cleanup;
1409 }
1410
1411 /* Channel is opened */
1412 netdev_dbg(ndev, "hv_netvsc channel opened successfully\n");
1413
1414 napi_enable(&net_device->chan_table[0].napi);
1415
1416 /* Connect with the NetVsp */
1417 ret = netvsc_connect_vsp(device, net_device, device_info);
1418 if (ret != 0) {
1419 netdev_err(ndev,
1420 "unable to connect to NetVSP - %d\n", ret);
1421 goto close;
1422 }
1423
1424 /* Writing nvdev pointer unlocks netvsc_send(), make sure chn_table is
1425 * populated.
1426 */
1427 rcu_assign_pointer(net_device_ctx->nvdev, net_device);
1428
1429 return net_device;
1430
1431close:
1432 RCU_INIT_POINTER(net_device_ctx->nvdev, NULL);
1433 napi_disable(&net_device->chan_table[0].napi);
1434
1435 /* Now, we can close the channel safely */
1436 vmbus_close(device->channel);
1437
1438cleanup:
1439 netif_napi_del(&net_device->chan_table[0].napi);
1440 free_netvsc_device(&net_device->rcu);
1441
1442 return ERR_PTR(ret);
1443}
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 * Switch the data path from the synthetic interface to the VF
38 * interface.
39 */
40void netvsc_switch_datapath(struct net_device *ndev, bool vf)
41{
42 struct net_device_context *net_device_ctx = netdev_priv(ndev);
43 struct hv_device *dev = net_device_ctx->device_ctx;
44 struct netvsc_device *nv_dev = net_device_ctx->nvdev;
45 struct nvsp_message *init_pkt = &nv_dev->channel_init_pkt;
46
47 memset(init_pkt, 0, sizeof(struct nvsp_message));
48 init_pkt->hdr.msg_type = NVSP_MSG4_TYPE_SWITCH_DATA_PATH;
49 if (vf)
50 init_pkt->msg.v4_msg.active_dp.active_datapath =
51 NVSP_DATAPATH_VF;
52 else
53 init_pkt->msg.v4_msg.active_dp.active_datapath =
54 NVSP_DATAPATH_SYNTHETIC;
55
56 vmbus_sendpacket(dev->channel, init_pkt,
57 sizeof(struct nvsp_message),
58 (unsigned long)init_pkt,
59 VM_PKT_DATA_INBAND, 0);
60}
61
62static struct netvsc_device *alloc_net_device(void)
63{
64 struct netvsc_device *net_device;
65
66 net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
67 if (!net_device)
68 return NULL;
69
70 net_device->cb_buffer = kzalloc(NETVSC_PACKET_SIZE, GFP_KERNEL);
71 if (!net_device->cb_buffer) {
72 kfree(net_device);
73 return NULL;
74 }
75
76 net_device->mrc[0].buf = vzalloc(NETVSC_RECVSLOT_MAX *
77 sizeof(struct recv_comp_data));
78
79 init_waitqueue_head(&net_device->wait_drain);
80 net_device->destroy = false;
81 atomic_set(&net_device->open_cnt, 0);
82 net_device->max_pkt = RNDIS_MAX_PKT_DEFAULT;
83 net_device->pkt_align = RNDIS_PKT_ALIGN_DEFAULT;
84 init_completion(&net_device->channel_init_wait);
85
86 return net_device;
87}
88
89static void free_netvsc_device(struct netvsc_device *nvdev)
90{
91 int i;
92
93 for (i = 0; i < VRSS_CHANNEL_MAX; i++)
94 vfree(nvdev->mrc[i].buf);
95
96 kfree(nvdev->cb_buffer);
97 kfree(nvdev);
98}
99
100static struct netvsc_device *get_outbound_net_device(struct hv_device *device)
101{
102 struct netvsc_device *net_device = hv_device_to_netvsc_device(device);
103
104 if (net_device && net_device->destroy)
105 net_device = NULL;
106
107 return net_device;
108}
109
110static struct netvsc_device *get_inbound_net_device(struct hv_device *device)
111{
112 struct netvsc_device *net_device = hv_device_to_netvsc_device(device);
113
114 if (!net_device)
115 goto get_in_err;
116
117 if (net_device->destroy &&
118 atomic_read(&net_device->num_outstanding_sends) == 0 &&
119 atomic_read(&net_device->num_outstanding_recvs) == 0)
120 net_device = NULL;
121
122get_in_err:
123 return net_device;
124}
125
126static void netvsc_destroy_buf(struct hv_device *device)
127{
128 struct nvsp_message *revoke_packet;
129 struct net_device *ndev = hv_get_drvdata(device);
130 struct netvsc_device *net_device = net_device_to_netvsc_device(ndev);
131 int ret;
132
133 /*
134 * If we got a section count, it means we received a
135 * SendReceiveBufferComplete msg (ie sent
136 * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
137 * to send a revoke msg here
138 */
139 if (net_device->recv_section_cnt) {
140 /* Send the revoke receive buffer */
141 revoke_packet = &net_device->revoke_packet;
142 memset(revoke_packet, 0, sizeof(struct nvsp_message));
143
144 revoke_packet->hdr.msg_type =
145 NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
146 revoke_packet->msg.v1_msg.
147 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
148
149 ret = vmbus_sendpacket(device->channel,
150 revoke_packet,
151 sizeof(struct nvsp_message),
152 (unsigned long)revoke_packet,
153 VM_PKT_DATA_INBAND, 0);
154 /*
155 * If we failed here, we might as well return and
156 * have a leak rather than continue and a bugchk
157 */
158 if (ret != 0) {
159 netdev_err(ndev, "unable to send "
160 "revoke receive buffer to netvsp\n");
161 return;
162 }
163 }
164
165 /* Teardown the gpadl on the vsp end */
166 if (net_device->recv_buf_gpadl_handle) {
167 ret = vmbus_teardown_gpadl(device->channel,
168 net_device->recv_buf_gpadl_handle);
169
170 /* If we failed here, we might as well return and have a leak
171 * rather than continue and a bugchk
172 */
173 if (ret != 0) {
174 netdev_err(ndev,
175 "unable to teardown receive buffer's gpadl\n");
176 return;
177 }
178 net_device->recv_buf_gpadl_handle = 0;
179 }
180
181 if (net_device->recv_buf) {
182 /* Free up the receive buffer */
183 vfree(net_device->recv_buf);
184 net_device->recv_buf = NULL;
185 }
186
187 if (net_device->recv_section) {
188 net_device->recv_section_cnt = 0;
189 kfree(net_device->recv_section);
190 net_device->recv_section = NULL;
191 }
192
193 /* Deal with the send buffer we may have setup.
194 * If we got a send section size, it means we received a
195 * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
196 * NVSP_MSG1_TYPE_SEND_SEND_BUF msg) therefore, we need
197 * to send a revoke msg here
198 */
199 if (net_device->send_section_size) {
200 /* Send the revoke receive buffer */
201 revoke_packet = &net_device->revoke_packet;
202 memset(revoke_packet, 0, sizeof(struct nvsp_message));
203
204 revoke_packet->hdr.msg_type =
205 NVSP_MSG1_TYPE_REVOKE_SEND_BUF;
206 revoke_packet->msg.v1_msg.revoke_send_buf.id =
207 NETVSC_SEND_BUFFER_ID;
208
209 ret = vmbus_sendpacket(device->channel,
210 revoke_packet,
211 sizeof(struct nvsp_message),
212 (unsigned long)revoke_packet,
213 VM_PKT_DATA_INBAND, 0);
214 /* If we failed here, we might as well return and
215 * have a leak rather than continue and a bugchk
216 */
217 if (ret != 0) {
218 netdev_err(ndev, "unable to send "
219 "revoke send buffer to netvsp\n");
220 return;
221 }
222 }
223 /* Teardown the gpadl on the vsp end */
224 if (net_device->send_buf_gpadl_handle) {
225 ret = vmbus_teardown_gpadl(device->channel,
226 net_device->send_buf_gpadl_handle);
227
228 /* If we failed here, we might as well return and have a leak
229 * rather than continue and a bugchk
230 */
231 if (ret != 0) {
232 netdev_err(ndev,
233 "unable to teardown send buffer's gpadl\n");
234 return;
235 }
236 net_device->send_buf_gpadl_handle = 0;
237 }
238 if (net_device->send_buf) {
239 /* Free up the send buffer */
240 vfree(net_device->send_buf);
241 net_device->send_buf = NULL;
242 }
243 kfree(net_device->send_section_map);
244}
245
246static int netvsc_init_buf(struct hv_device *device)
247{
248 int ret = 0;
249 struct netvsc_device *net_device;
250 struct nvsp_message *init_packet;
251 struct net_device *ndev;
252 int node;
253
254 net_device = get_outbound_net_device(device);
255 if (!net_device)
256 return -ENODEV;
257 ndev = hv_get_drvdata(device);
258
259 node = cpu_to_node(device->channel->target_cpu);
260 net_device->recv_buf = vzalloc_node(net_device->recv_buf_size, node);
261 if (!net_device->recv_buf)
262 net_device->recv_buf = vzalloc(net_device->recv_buf_size);
263
264 if (!net_device->recv_buf) {
265 netdev_err(ndev, "unable to allocate receive "
266 "buffer of size %d\n", net_device->recv_buf_size);
267 ret = -ENOMEM;
268 goto cleanup;
269 }
270
271 /*
272 * Establish the gpadl handle for this buffer on this
273 * channel. Note: This call uses the vmbus connection rather
274 * than the channel to establish the gpadl handle.
275 */
276 ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
277 net_device->recv_buf_size,
278 &net_device->recv_buf_gpadl_handle);
279 if (ret != 0) {
280 netdev_err(ndev,
281 "unable to establish receive buffer's gpadl\n");
282 goto cleanup;
283 }
284
285 /* Notify the NetVsp of the gpadl handle */
286 init_packet = &net_device->channel_init_pkt;
287
288 memset(init_packet, 0, sizeof(struct nvsp_message));
289
290 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
291 init_packet->msg.v1_msg.send_recv_buf.
292 gpadl_handle = net_device->recv_buf_gpadl_handle;
293 init_packet->msg.v1_msg.
294 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
295
296 /* Send the gpadl notification request */
297 ret = vmbus_sendpacket(device->channel, init_packet,
298 sizeof(struct nvsp_message),
299 (unsigned long)init_packet,
300 VM_PKT_DATA_INBAND,
301 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
302 if (ret != 0) {
303 netdev_err(ndev,
304 "unable to send receive buffer's gpadl to netvsp\n");
305 goto cleanup;
306 }
307
308 wait_for_completion(&net_device->channel_init_wait);
309
310 /* Check the response */
311 if (init_packet->msg.v1_msg.
312 send_recv_buf_complete.status != NVSP_STAT_SUCCESS) {
313 netdev_err(ndev, "Unable to complete receive buffer "
314 "initialization with NetVsp - status %d\n",
315 init_packet->msg.v1_msg.
316 send_recv_buf_complete.status);
317 ret = -EINVAL;
318 goto cleanup;
319 }
320
321 /* Parse the response */
322
323 net_device->recv_section_cnt = init_packet->msg.
324 v1_msg.send_recv_buf_complete.num_sections;
325
326 net_device->recv_section = kmemdup(
327 init_packet->msg.v1_msg.send_recv_buf_complete.sections,
328 net_device->recv_section_cnt *
329 sizeof(struct nvsp_1_receive_buffer_section),
330 GFP_KERNEL);
331 if (net_device->recv_section == NULL) {
332 ret = -EINVAL;
333 goto cleanup;
334 }
335
336 /*
337 * For 1st release, there should only be 1 section that represents the
338 * entire receive buffer
339 */
340 if (net_device->recv_section_cnt != 1 ||
341 net_device->recv_section->offset != 0) {
342 ret = -EINVAL;
343 goto cleanup;
344 }
345
346 /* Now setup the send buffer.
347 */
348 net_device->send_buf = vzalloc_node(net_device->send_buf_size, node);
349 if (!net_device->send_buf)
350 net_device->send_buf = vzalloc(net_device->send_buf_size);
351 if (!net_device->send_buf) {
352 netdev_err(ndev, "unable to allocate send "
353 "buffer of size %d\n", net_device->send_buf_size);
354 ret = -ENOMEM;
355 goto cleanup;
356 }
357
358 /* Establish the gpadl handle for this buffer on this
359 * channel. Note: This call uses the vmbus connection rather
360 * than the channel to establish the gpadl handle.
361 */
362 ret = vmbus_establish_gpadl(device->channel, net_device->send_buf,
363 net_device->send_buf_size,
364 &net_device->send_buf_gpadl_handle);
365 if (ret != 0) {
366 netdev_err(ndev,
367 "unable to establish send buffer's gpadl\n");
368 goto cleanup;
369 }
370
371 /* Notify the NetVsp of the gpadl handle */
372 init_packet = &net_device->channel_init_pkt;
373 memset(init_packet, 0, sizeof(struct nvsp_message));
374 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_SEND_BUF;
375 init_packet->msg.v1_msg.send_send_buf.gpadl_handle =
376 net_device->send_buf_gpadl_handle;
377 init_packet->msg.v1_msg.send_send_buf.id = NETVSC_SEND_BUFFER_ID;
378
379 /* Send the gpadl notification request */
380 ret = vmbus_sendpacket(device->channel, init_packet,
381 sizeof(struct nvsp_message),
382 (unsigned long)init_packet,
383 VM_PKT_DATA_INBAND,
384 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
385 if (ret != 0) {
386 netdev_err(ndev,
387 "unable to send send buffer's gpadl to netvsp\n");
388 goto cleanup;
389 }
390
391 wait_for_completion(&net_device->channel_init_wait);
392
393 /* Check the response */
394 if (init_packet->msg.v1_msg.
395 send_send_buf_complete.status != NVSP_STAT_SUCCESS) {
396 netdev_err(ndev, "Unable to complete send buffer "
397 "initialization with NetVsp - status %d\n",
398 init_packet->msg.v1_msg.
399 send_send_buf_complete.status);
400 ret = -EINVAL;
401 goto cleanup;
402 }
403
404 /* Parse the response */
405 net_device->send_section_size = init_packet->msg.
406 v1_msg.send_send_buf_complete.section_size;
407
408 /* Section count is simply the size divided by the section size.
409 */
410 net_device->send_section_cnt =
411 net_device->send_buf_size / net_device->send_section_size;
412
413 netdev_dbg(ndev, "Send section size: %d, Section count:%d\n",
414 net_device->send_section_size, net_device->send_section_cnt);
415
416 /* Setup state for managing the send buffer. */
417 net_device->map_words = DIV_ROUND_UP(net_device->send_section_cnt,
418 BITS_PER_LONG);
419
420 net_device->send_section_map = kcalloc(net_device->map_words,
421 sizeof(ulong), GFP_KERNEL);
422 if (net_device->send_section_map == NULL) {
423 ret = -ENOMEM;
424 goto cleanup;
425 }
426
427 goto exit;
428
429cleanup:
430 netvsc_destroy_buf(device);
431
432exit:
433 return ret;
434}
435
436/* Negotiate NVSP protocol version */
437static int negotiate_nvsp_ver(struct hv_device *device,
438 struct netvsc_device *net_device,
439 struct nvsp_message *init_packet,
440 u32 nvsp_ver)
441{
442 struct net_device *ndev = hv_get_drvdata(device);
443 int ret;
444
445 memset(init_packet, 0, sizeof(struct nvsp_message));
446 init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
447 init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
448 init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
449
450 /* Send the init request */
451 ret = vmbus_sendpacket(device->channel, init_packet,
452 sizeof(struct nvsp_message),
453 (unsigned long)init_packet,
454 VM_PKT_DATA_INBAND,
455 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
456
457 if (ret != 0)
458 return ret;
459
460 wait_for_completion(&net_device->channel_init_wait);
461
462 if (init_packet->msg.init_msg.init_complete.status !=
463 NVSP_STAT_SUCCESS)
464 return -EINVAL;
465
466 if (nvsp_ver == NVSP_PROTOCOL_VERSION_1)
467 return 0;
468
469 /* NVSPv2 or later: Send NDIS config */
470 memset(init_packet, 0, sizeof(struct nvsp_message));
471 init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
472 init_packet->msg.v2_msg.send_ndis_config.mtu = ndev->mtu + ETH_HLEN;
473 init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
474
475 if (nvsp_ver >= NVSP_PROTOCOL_VERSION_5) {
476 init_packet->msg.v2_msg.send_ndis_config.capability.sriov = 1;
477
478 /* Teaming bit is needed to receive link speed updates */
479 init_packet->msg.v2_msg.send_ndis_config.capability.teaming = 1;
480 }
481
482 ret = vmbus_sendpacket(device->channel, init_packet,
483 sizeof(struct nvsp_message),
484 (unsigned long)init_packet,
485 VM_PKT_DATA_INBAND, 0);
486
487 return ret;
488}
489
490static int netvsc_connect_vsp(struct hv_device *device)
491{
492 int ret;
493 struct netvsc_device *net_device;
494 struct nvsp_message *init_packet;
495 int ndis_version;
496 const u32 ver_list[] = {
497 NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
498 NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5 };
499 int i;
500
501 net_device = get_outbound_net_device(device);
502 if (!net_device)
503 return -ENODEV;
504
505 init_packet = &net_device->channel_init_pkt;
506
507 /* Negotiate the latest NVSP protocol supported */
508 for (i = ARRAY_SIZE(ver_list) - 1; i >= 0; i--)
509 if (negotiate_nvsp_ver(device, net_device, init_packet,
510 ver_list[i]) == 0) {
511 net_device->nvsp_version = ver_list[i];
512 break;
513 }
514
515 if (i < 0) {
516 ret = -EPROTO;
517 goto cleanup;
518 }
519
520 pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
521
522 /* Send the ndis version */
523 memset(init_packet, 0, sizeof(struct nvsp_message));
524
525 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
526 ndis_version = 0x00060001;
527 else
528 ndis_version = 0x0006001e;
529
530 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
531 init_packet->msg.v1_msg.
532 send_ndis_ver.ndis_major_ver =
533 (ndis_version & 0xFFFF0000) >> 16;
534 init_packet->msg.v1_msg.
535 send_ndis_ver.ndis_minor_ver =
536 ndis_version & 0xFFFF;
537
538 /* Send the init request */
539 ret = vmbus_sendpacket(device->channel, init_packet,
540 sizeof(struct nvsp_message),
541 (unsigned long)init_packet,
542 VM_PKT_DATA_INBAND, 0);
543 if (ret != 0)
544 goto cleanup;
545
546 /* Post the big receive buffer to NetVSP */
547 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
548 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
549 else
550 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
551 net_device->send_buf_size = NETVSC_SEND_BUFFER_SIZE;
552
553 ret = netvsc_init_buf(device);
554
555cleanup:
556 return ret;
557}
558
559static void netvsc_disconnect_vsp(struct hv_device *device)
560{
561 netvsc_destroy_buf(device);
562}
563
564/*
565 * netvsc_device_remove - Callback when the root bus device is removed
566 */
567void netvsc_device_remove(struct hv_device *device)
568{
569 struct net_device *ndev = hv_get_drvdata(device);
570 struct net_device_context *net_device_ctx = netdev_priv(ndev);
571 struct netvsc_device *net_device = net_device_ctx->nvdev;
572
573 netvsc_disconnect_vsp(device);
574
575 net_device_ctx->nvdev = NULL;
576
577 /*
578 * At this point, no one should be accessing net_device
579 * except in here
580 */
581 netdev_dbg(ndev, "net device safe to remove\n");
582
583 /* Now, we can close the channel safely */
584 vmbus_close(device->channel);
585
586 /* Release all resources */
587 vfree(net_device->sub_cb_buf);
588 free_netvsc_device(net_device);
589}
590
591#define RING_AVAIL_PERCENT_HIWATER 20
592#define RING_AVAIL_PERCENT_LOWATER 10
593
594/*
595 * Get the percentage of available bytes to write in the ring.
596 * The return value is in range from 0 to 100.
597 */
598static inline u32 hv_ringbuf_avail_percent(
599 struct hv_ring_buffer_info *ring_info)
600{
601 u32 avail_read, avail_write;
602
603 hv_get_ringbuffer_availbytes(ring_info, &avail_read, &avail_write);
604
605 return avail_write * 100 / ring_info->ring_datasize;
606}
607
608static inline void netvsc_free_send_slot(struct netvsc_device *net_device,
609 u32 index)
610{
611 sync_change_bit(index, net_device->send_section_map);
612}
613
614static void netvsc_send_tx_complete(struct netvsc_device *net_device,
615 struct vmbus_channel *incoming_channel,
616 struct hv_device *device,
617 struct vmpacket_descriptor *packet)
618{
619 struct sk_buff *skb = (struct sk_buff *)(unsigned long)packet->trans_id;
620 struct net_device *ndev = hv_get_drvdata(device);
621 struct net_device_context *net_device_ctx = netdev_priv(ndev);
622 struct vmbus_channel *channel = device->channel;
623 int num_outstanding_sends;
624 u16 q_idx = 0;
625 int queue_sends;
626
627 /* Notify the layer above us */
628 if (likely(skb)) {
629 struct hv_netvsc_packet *nvsc_packet
630 = (struct hv_netvsc_packet *)skb->cb;
631 u32 send_index = nvsc_packet->send_buf_index;
632
633 if (send_index != NETVSC_INVALID_INDEX)
634 netvsc_free_send_slot(net_device, send_index);
635 q_idx = nvsc_packet->q_idx;
636 channel = incoming_channel;
637
638 dev_consume_skb_any(skb);
639 }
640
641 num_outstanding_sends =
642 atomic_dec_return(&net_device->num_outstanding_sends);
643 queue_sends = atomic_dec_return(&net_device->queue_sends[q_idx]);
644
645 if (net_device->destroy && num_outstanding_sends == 0)
646 wake_up(&net_device->wait_drain);
647
648 if (netif_tx_queue_stopped(netdev_get_tx_queue(ndev, q_idx)) &&
649 !net_device_ctx->start_remove &&
650 (hv_ringbuf_avail_percent(&channel->outbound) > RING_AVAIL_PERCENT_HIWATER ||
651 queue_sends < 1))
652 netif_tx_wake_queue(netdev_get_tx_queue(ndev, q_idx));
653}
654
655static void netvsc_send_completion(struct netvsc_device *net_device,
656 struct vmbus_channel *incoming_channel,
657 struct hv_device *device,
658 struct vmpacket_descriptor *packet)
659{
660 struct nvsp_message *nvsp_packet;
661 struct net_device *ndev = hv_get_drvdata(device);
662
663 nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
664 (packet->offset8 << 3));
665
666 switch (nvsp_packet->hdr.msg_type) {
667 case NVSP_MSG_TYPE_INIT_COMPLETE:
668 case NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE:
669 case NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE:
670 case NVSP_MSG5_TYPE_SUBCHANNEL:
671 /* Copy the response back */
672 memcpy(&net_device->channel_init_pkt, nvsp_packet,
673 sizeof(struct nvsp_message));
674 complete(&net_device->channel_init_wait);
675 break;
676
677 case NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE:
678 netvsc_send_tx_complete(net_device, incoming_channel,
679 device, packet);
680 break;
681
682 default:
683 netdev_err(ndev,
684 "Unknown send completion type %d received!!\n",
685 nvsp_packet->hdr.msg_type);
686 }
687}
688
689static u32 netvsc_get_next_send_section(struct netvsc_device *net_device)
690{
691 unsigned long index;
692 u32 max_words = net_device->map_words;
693 unsigned long *map_addr = (unsigned long *)net_device->send_section_map;
694 u32 section_cnt = net_device->send_section_cnt;
695 int ret_val = NETVSC_INVALID_INDEX;
696 int i;
697 int prev_val;
698
699 for (i = 0; i < max_words; i++) {
700 if (!~(map_addr[i]))
701 continue;
702 index = ffz(map_addr[i]);
703 prev_val = sync_test_and_set_bit(index, &map_addr[i]);
704 if (prev_val)
705 continue;
706 if ((index + (i * BITS_PER_LONG)) >= section_cnt)
707 break;
708 ret_val = (index + (i * BITS_PER_LONG));
709 break;
710 }
711 return ret_val;
712}
713
714static u32 netvsc_copy_to_send_buf(struct netvsc_device *net_device,
715 unsigned int section_index,
716 u32 pend_size,
717 struct hv_netvsc_packet *packet,
718 struct rndis_message *rndis_msg,
719 struct hv_page_buffer **pb,
720 struct sk_buff *skb)
721{
722 char *start = net_device->send_buf;
723 char *dest = start + (section_index * net_device->send_section_size)
724 + pend_size;
725 int i;
726 bool is_data_pkt = (skb != NULL) ? true : false;
727 bool xmit_more = (skb != NULL) ? skb->xmit_more : false;
728 u32 msg_size = 0;
729 u32 padding = 0;
730 u32 remain = packet->total_data_buflen % net_device->pkt_align;
731 u32 page_count = packet->cp_partial ? packet->rmsg_pgcnt :
732 packet->page_buf_cnt;
733
734 /* Add padding */
735 if (is_data_pkt && xmit_more && remain &&
736 !packet->cp_partial) {
737 padding = net_device->pkt_align - remain;
738 rndis_msg->msg_len += padding;
739 packet->total_data_buflen += padding;
740 }
741
742 for (i = 0; i < page_count; i++) {
743 char *src = phys_to_virt((*pb)[i].pfn << PAGE_SHIFT);
744 u32 offset = (*pb)[i].offset;
745 u32 len = (*pb)[i].len;
746
747 memcpy(dest, (src + offset), len);
748 msg_size += len;
749 dest += len;
750 }
751
752 if (padding) {
753 memset(dest, 0, padding);
754 msg_size += padding;
755 }
756
757 return msg_size;
758}
759
760static inline int netvsc_send_pkt(
761 struct hv_device *device,
762 struct hv_netvsc_packet *packet,
763 struct netvsc_device *net_device,
764 struct hv_page_buffer **pb,
765 struct sk_buff *skb)
766{
767 struct nvsp_message nvmsg;
768 u16 q_idx = packet->q_idx;
769 struct vmbus_channel *out_channel = net_device->chn_table[q_idx];
770 struct net_device *ndev = hv_get_drvdata(device);
771 u64 req_id;
772 int ret;
773 struct hv_page_buffer *pgbuf;
774 u32 ring_avail = hv_ringbuf_avail_percent(&out_channel->outbound);
775 bool xmit_more = (skb != NULL) ? skb->xmit_more : false;
776
777 nvmsg.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
778 if (skb != NULL) {
779 /* 0 is RMC_DATA; */
780 nvmsg.msg.v1_msg.send_rndis_pkt.channel_type = 0;
781 } else {
782 /* 1 is RMC_CONTROL; */
783 nvmsg.msg.v1_msg.send_rndis_pkt.channel_type = 1;
784 }
785
786 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_index =
787 packet->send_buf_index;
788 if (packet->send_buf_index == NETVSC_INVALID_INDEX)
789 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_size = 0;
790 else
791 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_size =
792 packet->total_data_buflen;
793
794 req_id = (ulong)skb;
795
796 if (out_channel->rescind)
797 return -ENODEV;
798
799 /*
800 * It is possible that once we successfully place this packet
801 * on the ringbuffer, we may stop the queue. In that case, we want
802 * to notify the host independent of the xmit_more flag. We don't
803 * need to be precise here; in the worst case we may signal the host
804 * unnecessarily.
805 */
806 if (ring_avail < (RING_AVAIL_PERCENT_LOWATER + 1))
807 xmit_more = false;
808
809 if (packet->page_buf_cnt) {
810 pgbuf = packet->cp_partial ? (*pb) +
811 packet->rmsg_pgcnt : (*pb);
812 ret = vmbus_sendpacket_pagebuffer_ctl(out_channel,
813 pgbuf,
814 packet->page_buf_cnt,
815 &nvmsg,
816 sizeof(struct nvsp_message),
817 req_id,
818 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED,
819 !xmit_more);
820 } else {
821 ret = vmbus_sendpacket_ctl(out_channel, &nvmsg,
822 sizeof(struct nvsp_message),
823 req_id,
824 VM_PKT_DATA_INBAND,
825 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED,
826 !xmit_more);
827 }
828
829 if (ret == 0) {
830 atomic_inc(&net_device->num_outstanding_sends);
831 atomic_inc(&net_device->queue_sends[q_idx]);
832
833 if (ring_avail < RING_AVAIL_PERCENT_LOWATER) {
834 netif_tx_stop_queue(netdev_get_tx_queue(ndev, q_idx));
835
836 if (atomic_read(&net_device->
837 queue_sends[q_idx]) < 1)
838 netif_tx_wake_queue(netdev_get_tx_queue(
839 ndev, q_idx));
840 }
841 } else if (ret == -EAGAIN) {
842 netif_tx_stop_queue(netdev_get_tx_queue(
843 ndev, q_idx));
844 if (atomic_read(&net_device->queue_sends[q_idx]) < 1) {
845 netif_tx_wake_queue(netdev_get_tx_queue(
846 ndev, q_idx));
847 ret = -ENOSPC;
848 }
849 } else {
850 netdev_err(ndev, "Unable to send packet %p ret %d\n",
851 packet, ret);
852 }
853
854 return ret;
855}
856
857/* Move packet out of multi send data (msd), and clear msd */
858static inline void move_pkt_msd(struct hv_netvsc_packet **msd_send,
859 struct sk_buff **msd_skb,
860 struct multi_send_data *msdp)
861{
862 *msd_skb = msdp->skb;
863 *msd_send = msdp->pkt;
864 msdp->skb = NULL;
865 msdp->pkt = NULL;
866 msdp->count = 0;
867}
868
869int netvsc_send(struct hv_device *device,
870 struct hv_netvsc_packet *packet,
871 struct rndis_message *rndis_msg,
872 struct hv_page_buffer **pb,
873 struct sk_buff *skb)
874{
875 struct netvsc_device *net_device;
876 int ret = 0;
877 struct vmbus_channel *out_channel;
878 u16 q_idx = packet->q_idx;
879 u32 pktlen = packet->total_data_buflen, msd_len = 0;
880 unsigned int section_index = NETVSC_INVALID_INDEX;
881 struct multi_send_data *msdp;
882 struct hv_netvsc_packet *msd_send = NULL, *cur_send = NULL;
883 struct sk_buff *msd_skb = NULL;
884 bool try_batch;
885 bool xmit_more = (skb != NULL) ? skb->xmit_more : false;
886
887 net_device = get_outbound_net_device(device);
888 if (!net_device)
889 return -ENODEV;
890
891 /* We may race with netvsc_connect_vsp()/netvsc_init_buf() and get
892 * here before the negotiation with the host is finished and
893 * send_section_map may not be allocated yet.
894 */
895 if (!net_device->send_section_map)
896 return -EAGAIN;
897
898 out_channel = net_device->chn_table[q_idx];
899
900 packet->send_buf_index = NETVSC_INVALID_INDEX;
901 packet->cp_partial = false;
902
903 /* Send control message directly without accessing msd (Multi-Send
904 * Data) field which may be changed during data packet processing.
905 */
906 if (!skb) {
907 cur_send = packet;
908 goto send_now;
909 }
910
911 msdp = &net_device->msd[q_idx];
912
913 /* batch packets in send buffer if possible */
914 if (msdp->pkt)
915 msd_len = msdp->pkt->total_data_buflen;
916
917 try_batch = (skb != NULL) && msd_len > 0 && msdp->count <
918 net_device->max_pkt;
919
920 if (try_batch && msd_len + pktlen + net_device->pkt_align <
921 net_device->send_section_size) {
922 section_index = msdp->pkt->send_buf_index;
923
924 } else if (try_batch && msd_len + packet->rmsg_size <
925 net_device->send_section_size) {
926 section_index = msdp->pkt->send_buf_index;
927 packet->cp_partial = true;
928
929 } else if ((skb != NULL) && pktlen + net_device->pkt_align <
930 net_device->send_section_size) {
931 section_index = netvsc_get_next_send_section(net_device);
932 if (section_index != NETVSC_INVALID_INDEX) {
933 move_pkt_msd(&msd_send, &msd_skb, msdp);
934 msd_len = 0;
935 }
936 }
937
938 if (section_index != NETVSC_INVALID_INDEX) {
939 netvsc_copy_to_send_buf(net_device,
940 section_index, msd_len,
941 packet, rndis_msg, pb, skb);
942
943 packet->send_buf_index = section_index;
944
945 if (packet->cp_partial) {
946 packet->page_buf_cnt -= packet->rmsg_pgcnt;
947 packet->total_data_buflen = msd_len + packet->rmsg_size;
948 } else {
949 packet->page_buf_cnt = 0;
950 packet->total_data_buflen += msd_len;
951 }
952
953 if (msdp->skb)
954 dev_consume_skb_any(msdp->skb);
955
956 if (xmit_more && !packet->cp_partial) {
957 msdp->skb = skb;
958 msdp->pkt = packet;
959 msdp->count++;
960 } else {
961 cur_send = packet;
962 msdp->skb = NULL;
963 msdp->pkt = NULL;
964 msdp->count = 0;
965 }
966 } else {
967 move_pkt_msd(&msd_send, &msd_skb, msdp);
968 cur_send = packet;
969 }
970
971 if (msd_send) {
972 int m_ret = netvsc_send_pkt(device, msd_send, net_device,
973 NULL, msd_skb);
974
975 if (m_ret != 0) {
976 netvsc_free_send_slot(net_device,
977 msd_send->send_buf_index);
978 dev_kfree_skb_any(msd_skb);
979 }
980 }
981
982send_now:
983 if (cur_send)
984 ret = netvsc_send_pkt(device, cur_send, net_device, pb, skb);
985
986 if (ret != 0 && section_index != NETVSC_INVALID_INDEX)
987 netvsc_free_send_slot(net_device, section_index);
988
989 return ret;
990}
991
992static int netvsc_send_recv_completion(struct vmbus_channel *channel,
993 u64 transaction_id, u32 status)
994{
995 struct nvsp_message recvcompMessage;
996 int ret;
997
998 recvcompMessage.hdr.msg_type =
999 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE;
1000
1001 recvcompMessage.msg.v1_msg.send_rndis_pkt_complete.status = status;
1002
1003 /* Send the completion */
1004 ret = vmbus_sendpacket(channel, &recvcompMessage,
1005 sizeof(struct nvsp_message_header) + sizeof(u32),
1006 transaction_id, VM_PKT_COMP, 0);
1007
1008 return ret;
1009}
1010
1011static inline void count_recv_comp_slot(struct netvsc_device *nvdev, u16 q_idx,
1012 u32 *filled, u32 *avail)
1013{
1014 u32 first = nvdev->mrc[q_idx].first;
1015 u32 next = nvdev->mrc[q_idx].next;
1016
1017 *filled = (first > next) ? NETVSC_RECVSLOT_MAX - first + next :
1018 next - first;
1019
1020 *avail = NETVSC_RECVSLOT_MAX - *filled - 1;
1021}
1022
1023/* Read the first filled slot, no change to index */
1024static inline struct recv_comp_data *read_recv_comp_slot(struct netvsc_device
1025 *nvdev, u16 q_idx)
1026{
1027 u32 filled, avail;
1028
1029 if (!nvdev->mrc[q_idx].buf)
1030 return NULL;
1031
1032 count_recv_comp_slot(nvdev, q_idx, &filled, &avail);
1033 if (!filled)
1034 return NULL;
1035
1036 return nvdev->mrc[q_idx].buf + nvdev->mrc[q_idx].first *
1037 sizeof(struct recv_comp_data);
1038}
1039
1040/* Put the first filled slot back to available pool */
1041static inline void put_recv_comp_slot(struct netvsc_device *nvdev, u16 q_idx)
1042{
1043 int num_recv;
1044
1045 nvdev->mrc[q_idx].first = (nvdev->mrc[q_idx].first + 1) %
1046 NETVSC_RECVSLOT_MAX;
1047
1048 num_recv = atomic_dec_return(&nvdev->num_outstanding_recvs);
1049
1050 if (nvdev->destroy && num_recv == 0)
1051 wake_up(&nvdev->wait_drain);
1052}
1053
1054/* Check and send pending recv completions */
1055static void netvsc_chk_recv_comp(struct netvsc_device *nvdev,
1056 struct vmbus_channel *channel, u16 q_idx)
1057{
1058 struct recv_comp_data *rcd;
1059 int ret;
1060
1061 while (true) {
1062 rcd = read_recv_comp_slot(nvdev, q_idx);
1063 if (!rcd)
1064 break;
1065
1066 ret = netvsc_send_recv_completion(channel, rcd->tid,
1067 rcd->status);
1068 if (ret)
1069 break;
1070
1071 put_recv_comp_slot(nvdev, q_idx);
1072 }
1073}
1074
1075#define NETVSC_RCD_WATERMARK 80
1076
1077/* Get next available slot */
1078static inline struct recv_comp_data *get_recv_comp_slot(
1079 struct netvsc_device *nvdev, struct vmbus_channel *channel, u16 q_idx)
1080{
1081 u32 filled, avail, next;
1082 struct recv_comp_data *rcd;
1083
1084 if (!nvdev->recv_section)
1085 return NULL;
1086
1087 if (!nvdev->mrc[q_idx].buf)
1088 return NULL;
1089
1090 if (atomic_read(&nvdev->num_outstanding_recvs) >
1091 nvdev->recv_section->num_sub_allocs * NETVSC_RCD_WATERMARK / 100)
1092 netvsc_chk_recv_comp(nvdev, channel, q_idx);
1093
1094 count_recv_comp_slot(nvdev, q_idx, &filled, &avail);
1095 if (!avail)
1096 return NULL;
1097
1098 next = nvdev->mrc[q_idx].next;
1099 rcd = nvdev->mrc[q_idx].buf + next * sizeof(struct recv_comp_data);
1100 nvdev->mrc[q_idx].next = (next + 1) % NETVSC_RECVSLOT_MAX;
1101
1102 atomic_inc(&nvdev->num_outstanding_recvs);
1103
1104 return rcd;
1105}
1106
1107static void netvsc_receive(struct netvsc_device *net_device,
1108 struct vmbus_channel *channel,
1109 struct hv_device *device,
1110 struct vmpacket_descriptor *packet)
1111{
1112 struct vmtransfer_page_packet_header *vmxferpage_packet;
1113 struct nvsp_message *nvsp_packet;
1114 struct hv_netvsc_packet nv_pkt;
1115 struct hv_netvsc_packet *netvsc_packet = &nv_pkt;
1116 u32 status = NVSP_STAT_SUCCESS;
1117 int i;
1118 int count = 0;
1119 struct net_device *ndev = hv_get_drvdata(device);
1120 void *data;
1121 int ret;
1122 struct recv_comp_data *rcd;
1123 u16 q_idx = channel->offermsg.offer.sub_channel_index;
1124
1125 /*
1126 * All inbound packets other than send completion should be xfer page
1127 * packet
1128 */
1129 if (packet->type != VM_PKT_DATA_USING_XFER_PAGES) {
1130 netdev_err(ndev, "Unknown packet type received - %d\n",
1131 packet->type);
1132 return;
1133 }
1134
1135 nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
1136 (packet->offset8 << 3));
1137
1138 /* Make sure this is a valid nvsp packet */
1139 if (nvsp_packet->hdr.msg_type !=
1140 NVSP_MSG1_TYPE_SEND_RNDIS_PKT) {
1141 netdev_err(ndev, "Unknown nvsp packet type received-"
1142 " %d\n", nvsp_packet->hdr.msg_type);
1143 return;
1144 }
1145
1146 vmxferpage_packet = (struct vmtransfer_page_packet_header *)packet;
1147
1148 if (vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID) {
1149 netdev_err(ndev, "Invalid xfer page set id - "
1150 "expecting %x got %x\n", NETVSC_RECEIVE_BUFFER_ID,
1151 vmxferpage_packet->xfer_pageset_id);
1152 return;
1153 }
1154
1155 count = vmxferpage_packet->range_cnt;
1156
1157 /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
1158 for (i = 0; i < count; i++) {
1159 /* Initialize the netvsc packet */
1160 data = (void *)((unsigned long)net_device->
1161 recv_buf + vmxferpage_packet->ranges[i].byte_offset);
1162 netvsc_packet->total_data_buflen =
1163 vmxferpage_packet->ranges[i].byte_count;
1164
1165 /* Pass it to the upper layer */
1166 status = rndis_filter_receive(device, netvsc_packet, &data,
1167 channel);
1168 }
1169
1170 if (!net_device->mrc[q_idx].buf) {
1171 ret = netvsc_send_recv_completion(channel,
1172 vmxferpage_packet->d.trans_id,
1173 status);
1174 if (ret)
1175 netdev_err(ndev, "Recv_comp q:%hd, tid:%llx, err:%d\n",
1176 q_idx, vmxferpage_packet->d.trans_id, ret);
1177 return;
1178 }
1179
1180 rcd = get_recv_comp_slot(net_device, channel, q_idx);
1181
1182 if (!rcd) {
1183 netdev_err(ndev, "Recv_comp full buf q:%hd, tid:%llx\n",
1184 q_idx, vmxferpage_packet->d.trans_id);
1185 return;
1186 }
1187
1188 rcd->tid = vmxferpage_packet->d.trans_id;
1189 rcd->status = status;
1190}
1191
1192static void netvsc_send_table(struct hv_device *hdev,
1193 struct nvsp_message *nvmsg)
1194{
1195 struct netvsc_device *nvscdev;
1196 struct net_device *ndev = hv_get_drvdata(hdev);
1197 int i;
1198 u32 count, *tab;
1199
1200 nvscdev = get_outbound_net_device(hdev);
1201 if (!nvscdev)
1202 return;
1203
1204 count = nvmsg->msg.v5_msg.send_table.count;
1205 if (count != VRSS_SEND_TAB_SIZE) {
1206 netdev_err(ndev, "Received wrong send-table size:%u\n", count);
1207 return;
1208 }
1209
1210 tab = (u32 *)((unsigned long)&nvmsg->msg.v5_msg.send_table +
1211 nvmsg->msg.v5_msg.send_table.offset);
1212
1213 for (i = 0; i < count; i++)
1214 nvscdev->send_table[i] = tab[i];
1215}
1216
1217static void netvsc_send_vf(struct net_device_context *net_device_ctx,
1218 struct nvsp_message *nvmsg)
1219{
1220 net_device_ctx->vf_alloc = nvmsg->msg.v4_msg.vf_assoc.allocated;
1221 net_device_ctx->vf_serial = nvmsg->msg.v4_msg.vf_assoc.serial;
1222}
1223
1224static inline void netvsc_receive_inband(struct hv_device *hdev,
1225 struct net_device_context *net_device_ctx,
1226 struct nvsp_message *nvmsg)
1227{
1228 switch (nvmsg->hdr.msg_type) {
1229 case NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE:
1230 netvsc_send_table(hdev, nvmsg);
1231 break;
1232
1233 case NVSP_MSG4_TYPE_SEND_VF_ASSOCIATION:
1234 netvsc_send_vf(net_device_ctx, nvmsg);
1235 break;
1236 }
1237}
1238
1239static void netvsc_process_raw_pkt(struct hv_device *device,
1240 struct vmbus_channel *channel,
1241 struct netvsc_device *net_device,
1242 struct net_device *ndev,
1243 u64 request_id,
1244 struct vmpacket_descriptor *desc)
1245{
1246 struct nvsp_message *nvmsg;
1247 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1248
1249 nvmsg = (struct nvsp_message *)((unsigned long)
1250 desc + (desc->offset8 << 3));
1251
1252 switch (desc->type) {
1253 case VM_PKT_COMP:
1254 netvsc_send_completion(net_device, channel, device, desc);
1255 break;
1256
1257 case VM_PKT_DATA_USING_XFER_PAGES:
1258 netvsc_receive(net_device, channel, device, desc);
1259 break;
1260
1261 case VM_PKT_DATA_INBAND:
1262 netvsc_receive_inband(device, net_device_ctx, nvmsg);
1263 break;
1264
1265 default:
1266 netdev_err(ndev, "unhandled packet type %d, tid %llx\n",
1267 desc->type, request_id);
1268 break;
1269 }
1270}
1271
1272void netvsc_channel_cb(void *context)
1273{
1274 int ret;
1275 struct vmbus_channel *channel = (struct vmbus_channel *)context;
1276 u16 q_idx = channel->offermsg.offer.sub_channel_index;
1277 struct hv_device *device;
1278 struct netvsc_device *net_device;
1279 u32 bytes_recvd;
1280 u64 request_id;
1281 struct vmpacket_descriptor *desc;
1282 unsigned char *buffer;
1283 int bufferlen = NETVSC_PACKET_SIZE;
1284 struct net_device *ndev;
1285 bool need_to_commit = false;
1286
1287 if (channel->primary_channel != NULL)
1288 device = channel->primary_channel->device_obj;
1289 else
1290 device = channel->device_obj;
1291
1292 net_device = get_inbound_net_device(device);
1293 if (!net_device)
1294 return;
1295 ndev = hv_get_drvdata(device);
1296 buffer = get_per_channel_state(channel);
1297
1298 /* commit_rd_index() -> hv_signal_on_read() needs this. */
1299 init_cached_read_index(channel);
1300
1301 do {
1302 desc = get_next_pkt_raw(channel);
1303 if (desc != NULL) {
1304 netvsc_process_raw_pkt(device,
1305 channel,
1306 net_device,
1307 ndev,
1308 desc->trans_id,
1309 desc);
1310
1311 put_pkt_raw(channel, desc);
1312 need_to_commit = true;
1313 continue;
1314 }
1315 if (need_to_commit) {
1316 need_to_commit = false;
1317 commit_rd_index(channel);
1318 }
1319
1320 ret = vmbus_recvpacket_raw(channel, buffer, bufferlen,
1321 &bytes_recvd, &request_id);
1322 if (ret == 0) {
1323 if (bytes_recvd > 0) {
1324 desc = (struct vmpacket_descriptor *)buffer;
1325 netvsc_process_raw_pkt(device,
1326 channel,
1327 net_device,
1328 ndev,
1329 request_id,
1330 desc);
1331 } else {
1332 /*
1333 * We are done for this pass.
1334 */
1335 break;
1336 }
1337
1338 } else if (ret == -ENOBUFS) {
1339 if (bufferlen > NETVSC_PACKET_SIZE)
1340 kfree(buffer);
1341 /* Handle large packet */
1342 buffer = kmalloc(bytes_recvd, GFP_ATOMIC);
1343 if (buffer == NULL) {
1344 /* Try again next time around */
1345 netdev_err(ndev,
1346 "unable to allocate buffer of size "
1347 "(%d)!!\n", bytes_recvd);
1348 break;
1349 }
1350
1351 bufferlen = bytes_recvd;
1352 }
1353
1354 init_cached_read_index(channel);
1355
1356 } while (1);
1357
1358 if (bufferlen > NETVSC_PACKET_SIZE)
1359 kfree(buffer);
1360
1361 netvsc_chk_recv_comp(net_device, channel, q_idx);
1362}
1363
1364/*
1365 * netvsc_device_add - Callback when the device belonging to this
1366 * driver is added
1367 */
1368int netvsc_device_add(struct hv_device *device, void *additional_info)
1369{
1370 int i, ret = 0;
1371 int ring_size =
1372 ((struct netvsc_device_info *)additional_info)->ring_size;
1373 struct netvsc_device *net_device;
1374 struct net_device *ndev = hv_get_drvdata(device);
1375 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1376
1377 net_device = alloc_net_device();
1378 if (!net_device)
1379 return -ENOMEM;
1380
1381 net_device->ring_size = ring_size;
1382
1383 set_per_channel_state(device->channel, net_device->cb_buffer);
1384
1385 /* Open the channel */
1386 ret = vmbus_open(device->channel, ring_size * PAGE_SIZE,
1387 ring_size * PAGE_SIZE, NULL, 0,
1388 netvsc_channel_cb, device->channel);
1389
1390 if (ret != 0) {
1391 netdev_err(ndev, "unable to open channel: %d\n", ret);
1392 goto cleanup;
1393 }
1394
1395 /* Channel is opened */
1396 netdev_dbg(ndev, "hv_netvsc channel opened successfully\n");
1397
1398 /* If we're reopening the device we may have multiple queues, fill the
1399 * chn_table with the default channel to use it before subchannels are
1400 * opened.
1401 */
1402 for (i = 0; i < VRSS_CHANNEL_MAX; i++)
1403 net_device->chn_table[i] = device->channel;
1404
1405 /* Writing nvdev pointer unlocks netvsc_send(), make sure chn_table is
1406 * populated.
1407 */
1408 wmb();
1409
1410 net_device_ctx->nvdev = net_device;
1411
1412 /* Connect with the NetVsp */
1413 ret = netvsc_connect_vsp(device);
1414 if (ret != 0) {
1415 netdev_err(ndev,
1416 "unable to connect to NetVSP - %d\n", ret);
1417 goto close;
1418 }
1419
1420 return ret;
1421
1422close:
1423 /* Now, we can close the channel safely */
1424 vmbus_close(device->channel);
1425
1426cleanup:
1427 free_netvsc_device(net_device);
1428
1429 return ret;
1430}