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, write to the Free Software Foundation, Inc., 59 Temple
15 * Place - Suite 330, Boston, MA 02111-1307 USA.
16 *
17 * Authors:
18 * Haiyang Zhang <haiyangz@microsoft.com>
19 * Hank Janssen <hjanssen@microsoft.com>
20 */
21#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
22
23#include <linux/kernel.h>
24#include <linux/sched.h>
25#include <linux/wait.h>
26#include <linux/mm.h>
27#include <linux/delay.h>
28#include <linux/io.h>
29#include <linux/slab.h>
30#include <linux/netdevice.h>
31#include <linux/if_ether.h>
32
33#include "hyperv_net.h"
34
35
36static struct netvsc_device *alloc_net_device(struct hv_device *device)
37{
38 struct netvsc_device *net_device;
39 struct net_device *ndev = hv_get_drvdata(device);
40
41 net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
42 if (!net_device)
43 return NULL;
44
45 init_waitqueue_head(&net_device->wait_drain);
46 net_device->start_remove = false;
47 net_device->destroy = false;
48 net_device->dev = device;
49 net_device->ndev = ndev;
50
51 hv_set_drvdata(device, net_device);
52 return net_device;
53}
54
55static struct netvsc_device *get_outbound_net_device(struct hv_device *device)
56{
57 struct netvsc_device *net_device;
58
59 net_device = hv_get_drvdata(device);
60 if (net_device && net_device->destroy)
61 net_device = NULL;
62
63 return net_device;
64}
65
66static struct netvsc_device *get_inbound_net_device(struct hv_device *device)
67{
68 struct netvsc_device *net_device;
69
70 net_device = hv_get_drvdata(device);
71
72 if (!net_device)
73 goto get_in_err;
74
75 if (net_device->destroy &&
76 atomic_read(&net_device->num_outstanding_sends) == 0)
77 net_device = NULL;
78
79get_in_err:
80 return net_device;
81}
82
83
84static int netvsc_destroy_recv_buf(struct netvsc_device *net_device)
85{
86 struct nvsp_message *revoke_packet;
87 int ret = 0;
88 struct net_device *ndev = net_device->ndev;
89
90 /*
91 * If we got a section count, it means we received a
92 * SendReceiveBufferComplete msg (ie sent
93 * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
94 * to send a revoke msg here
95 */
96 if (net_device->recv_section_cnt) {
97 /* Send the revoke receive buffer */
98 revoke_packet = &net_device->revoke_packet;
99 memset(revoke_packet, 0, sizeof(struct nvsp_message));
100
101 revoke_packet->hdr.msg_type =
102 NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
103 revoke_packet->msg.v1_msg.
104 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
105
106 ret = vmbus_sendpacket(net_device->dev->channel,
107 revoke_packet,
108 sizeof(struct nvsp_message),
109 (unsigned long)revoke_packet,
110 VM_PKT_DATA_INBAND, 0);
111 /*
112 * If we failed here, we might as well return and
113 * have a leak rather than continue and a bugchk
114 */
115 if (ret != 0) {
116 netdev_err(ndev, "unable to send "
117 "revoke receive buffer to netvsp\n");
118 return ret;
119 }
120 }
121
122 /* Teardown the gpadl on the vsp end */
123 if (net_device->recv_buf_gpadl_handle) {
124 ret = vmbus_teardown_gpadl(net_device->dev->channel,
125 net_device->recv_buf_gpadl_handle);
126
127 /* If we failed here, we might as well return and have a leak
128 * rather than continue and a bugchk
129 */
130 if (ret != 0) {
131 netdev_err(ndev,
132 "unable to teardown receive buffer's gpadl\n");
133 return ret;
134 }
135 net_device->recv_buf_gpadl_handle = 0;
136 }
137
138 if (net_device->recv_buf) {
139 /* Free up the receive buffer */
140 free_pages((unsigned long)net_device->recv_buf,
141 get_order(net_device->recv_buf_size));
142 net_device->recv_buf = NULL;
143 }
144
145 if (net_device->recv_section) {
146 net_device->recv_section_cnt = 0;
147 kfree(net_device->recv_section);
148 net_device->recv_section = NULL;
149 }
150
151 return ret;
152}
153
154static int netvsc_init_recv_buf(struct hv_device *device)
155{
156 int ret = 0;
157 int t;
158 struct netvsc_device *net_device;
159 struct nvsp_message *init_packet;
160 struct net_device *ndev;
161
162 net_device = get_outbound_net_device(device);
163 if (!net_device)
164 return -ENODEV;
165 ndev = net_device->ndev;
166
167 net_device->recv_buf =
168 (void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO,
169 get_order(net_device->recv_buf_size));
170 if (!net_device->recv_buf) {
171 netdev_err(ndev, "unable to allocate receive "
172 "buffer of size %d\n", net_device->recv_buf_size);
173 ret = -ENOMEM;
174 goto cleanup;
175 }
176
177 /*
178 * Establish the gpadl handle for this buffer on this
179 * channel. Note: This call uses the vmbus connection rather
180 * than the channel to establish the gpadl handle.
181 */
182 ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
183 net_device->recv_buf_size,
184 &net_device->recv_buf_gpadl_handle);
185 if (ret != 0) {
186 netdev_err(ndev,
187 "unable to establish receive buffer's gpadl\n");
188 goto cleanup;
189 }
190
191
192 /* Notify the NetVsp of the gpadl handle */
193 init_packet = &net_device->channel_init_pkt;
194
195 memset(init_packet, 0, sizeof(struct nvsp_message));
196
197 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
198 init_packet->msg.v1_msg.send_recv_buf.
199 gpadl_handle = net_device->recv_buf_gpadl_handle;
200 init_packet->msg.v1_msg.
201 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
202
203 /* Send the gpadl notification request */
204 ret = vmbus_sendpacket(device->channel, init_packet,
205 sizeof(struct nvsp_message),
206 (unsigned long)init_packet,
207 VM_PKT_DATA_INBAND,
208 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
209 if (ret != 0) {
210 netdev_err(ndev,
211 "unable to send receive buffer's gpadl to netvsp\n");
212 goto cleanup;
213 }
214
215 t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
216 BUG_ON(t == 0);
217
218
219 /* Check the response */
220 if (init_packet->msg.v1_msg.
221 send_recv_buf_complete.status != NVSP_STAT_SUCCESS) {
222 netdev_err(ndev, "Unable to complete receive buffer "
223 "initialization with NetVsp - status %d\n",
224 init_packet->msg.v1_msg.
225 send_recv_buf_complete.status);
226 ret = -EINVAL;
227 goto cleanup;
228 }
229
230 /* Parse the response */
231
232 net_device->recv_section_cnt = init_packet->msg.
233 v1_msg.send_recv_buf_complete.num_sections;
234
235 net_device->recv_section = kmemdup(
236 init_packet->msg.v1_msg.send_recv_buf_complete.sections,
237 net_device->recv_section_cnt *
238 sizeof(struct nvsp_1_receive_buffer_section),
239 GFP_KERNEL);
240 if (net_device->recv_section == NULL) {
241 ret = -EINVAL;
242 goto cleanup;
243 }
244
245 /*
246 * For 1st release, there should only be 1 section that represents the
247 * entire receive buffer
248 */
249 if (net_device->recv_section_cnt != 1 ||
250 net_device->recv_section->offset != 0) {
251 ret = -EINVAL;
252 goto cleanup;
253 }
254
255 goto exit;
256
257cleanup:
258 netvsc_destroy_recv_buf(net_device);
259
260exit:
261 return ret;
262}
263
264
265/* Negotiate NVSP protocol version */
266static int negotiate_nvsp_ver(struct hv_device *device,
267 struct netvsc_device *net_device,
268 struct nvsp_message *init_packet,
269 u32 nvsp_ver)
270{
271 int ret, t;
272
273 memset(init_packet, 0, sizeof(struct nvsp_message));
274 init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
275 init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
276 init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
277
278 /* Send the init request */
279 ret = vmbus_sendpacket(device->channel, init_packet,
280 sizeof(struct nvsp_message),
281 (unsigned long)init_packet,
282 VM_PKT_DATA_INBAND,
283 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
284
285 if (ret != 0)
286 return ret;
287
288 t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
289
290 if (t == 0)
291 return -ETIMEDOUT;
292
293 if (init_packet->msg.init_msg.init_complete.status !=
294 NVSP_STAT_SUCCESS)
295 return -EINVAL;
296
297 if (nvsp_ver != NVSP_PROTOCOL_VERSION_2)
298 return 0;
299
300 /* NVSPv2 only: Send NDIS config */
301 memset(init_packet, 0, sizeof(struct nvsp_message));
302 init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
303 init_packet->msg.v2_msg.send_ndis_config.mtu = net_device->ndev->mtu;
304 init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
305
306 ret = vmbus_sendpacket(device->channel, init_packet,
307 sizeof(struct nvsp_message),
308 (unsigned long)init_packet,
309 VM_PKT_DATA_INBAND, 0);
310
311 return ret;
312}
313
314static int netvsc_connect_vsp(struct hv_device *device)
315{
316 int ret;
317 struct netvsc_device *net_device;
318 struct nvsp_message *init_packet;
319 int ndis_version;
320 struct net_device *ndev;
321
322 net_device = get_outbound_net_device(device);
323 if (!net_device)
324 return -ENODEV;
325 ndev = net_device->ndev;
326
327 init_packet = &net_device->channel_init_pkt;
328
329 /* Negotiate the latest NVSP protocol supported */
330 if (negotiate_nvsp_ver(device, net_device, init_packet,
331 NVSP_PROTOCOL_VERSION_2) == 0) {
332 net_device->nvsp_version = NVSP_PROTOCOL_VERSION_2;
333 } else if (negotiate_nvsp_ver(device, net_device, init_packet,
334 NVSP_PROTOCOL_VERSION_1) == 0) {
335 net_device->nvsp_version = NVSP_PROTOCOL_VERSION_1;
336 } else {
337 ret = -EPROTO;
338 goto cleanup;
339 }
340
341 pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
342
343 /* Send the ndis version */
344 memset(init_packet, 0, sizeof(struct nvsp_message));
345
346 ndis_version = 0x00050001;
347
348 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
349 init_packet->msg.v1_msg.
350 send_ndis_ver.ndis_major_ver =
351 (ndis_version & 0xFFFF0000) >> 16;
352 init_packet->msg.v1_msg.
353 send_ndis_ver.ndis_minor_ver =
354 ndis_version & 0xFFFF;
355
356 /* Send the init request */
357 ret = vmbus_sendpacket(device->channel, init_packet,
358 sizeof(struct nvsp_message),
359 (unsigned long)init_packet,
360 VM_PKT_DATA_INBAND, 0);
361 if (ret != 0)
362 goto cleanup;
363
364 /* Post the big receive buffer to NetVSP */
365 ret = netvsc_init_recv_buf(device);
366
367cleanup:
368 return ret;
369}
370
371static void netvsc_disconnect_vsp(struct netvsc_device *net_device)
372{
373 netvsc_destroy_recv_buf(net_device);
374}
375
376/*
377 * netvsc_device_remove - Callback when the root bus device is removed
378 */
379int netvsc_device_remove(struct hv_device *device)
380{
381 struct netvsc_device *net_device;
382 struct hv_netvsc_packet *netvsc_packet, *pos;
383 unsigned long flags;
384
385 net_device = hv_get_drvdata(device);
386 spin_lock_irqsave(&device->channel->inbound_lock, flags);
387 net_device->destroy = true;
388 spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
389
390 /* Wait for all send completions */
391 wait_event(net_device->wait_drain,
392 atomic_read(&net_device->num_outstanding_sends) == 0);
393
394 netvsc_disconnect_vsp(net_device);
395
396 /*
397 * Since we have already drained, we don't need to busy wait
398 * as was done in final_release_stor_device()
399 * Note that we cannot set the ext pointer to NULL until
400 * we have drained - to drain the outgoing packets, we need to
401 * allow incoming packets.
402 */
403
404 spin_lock_irqsave(&device->channel->inbound_lock, flags);
405 hv_set_drvdata(device, NULL);
406 spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
407
408 /*
409 * At this point, no one should be accessing net_device
410 * except in here
411 */
412 dev_notice(&device->device, "net device safe to remove\n");
413
414 /* Now, we can close the channel safely */
415 vmbus_close(device->channel);
416
417 /* Release all resources */
418 list_for_each_entry_safe(netvsc_packet, pos,
419 &net_device->recv_pkt_list, list_ent) {
420 list_del(&netvsc_packet->list_ent);
421 kfree(netvsc_packet);
422 }
423
424 kfree(net_device);
425 return 0;
426}
427
428
429#define RING_AVAIL_PERCENT_HIWATER 20
430#define RING_AVAIL_PERCENT_LOWATER 10
431
432/*
433 * Get the percentage of available bytes to write in the ring.
434 * The return value is in range from 0 to 100.
435 */
436static inline u32 hv_ringbuf_avail_percent(
437 struct hv_ring_buffer_info *ring_info)
438{
439 u32 avail_read, avail_write;
440
441 hv_get_ringbuffer_availbytes(ring_info, &avail_read, &avail_write);
442
443 return avail_write * 100 / ring_info->ring_datasize;
444}
445
446static void netvsc_send_completion(struct hv_device *device,
447 struct vmpacket_descriptor *packet)
448{
449 struct netvsc_device *net_device;
450 struct nvsp_message *nvsp_packet;
451 struct hv_netvsc_packet *nvsc_packet;
452 struct net_device *ndev;
453
454 net_device = get_inbound_net_device(device);
455 if (!net_device)
456 return;
457 ndev = net_device->ndev;
458
459 nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
460 (packet->offset8 << 3));
461
462 if ((nvsp_packet->hdr.msg_type == NVSP_MSG_TYPE_INIT_COMPLETE) ||
463 (nvsp_packet->hdr.msg_type ==
464 NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE) ||
465 (nvsp_packet->hdr.msg_type ==
466 NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE)) {
467 /* Copy the response back */
468 memcpy(&net_device->channel_init_pkt, nvsp_packet,
469 sizeof(struct nvsp_message));
470 complete(&net_device->channel_init_wait);
471 } else if (nvsp_packet->hdr.msg_type ==
472 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE) {
473 int num_outstanding_sends;
474
475 /* Get the send context */
476 nvsc_packet = (struct hv_netvsc_packet *)(unsigned long)
477 packet->trans_id;
478
479 /* Notify the layer above us */
480 nvsc_packet->completion.send.send_completion(
481 nvsc_packet->completion.send.send_completion_ctx);
482
483 num_outstanding_sends =
484 atomic_dec_return(&net_device->num_outstanding_sends);
485
486 if (net_device->destroy && num_outstanding_sends == 0)
487 wake_up(&net_device->wait_drain);
488
489 if (netif_queue_stopped(ndev) && !net_device->start_remove &&
490 (hv_ringbuf_avail_percent(&device->channel->outbound)
491 > RING_AVAIL_PERCENT_HIWATER ||
492 num_outstanding_sends < 1))
493 netif_wake_queue(ndev);
494 } else {
495 netdev_err(ndev, "Unknown send completion packet type- "
496 "%d received!!\n", nvsp_packet->hdr.msg_type);
497 }
498
499}
500
501int netvsc_send(struct hv_device *device,
502 struct hv_netvsc_packet *packet)
503{
504 struct netvsc_device *net_device;
505 int ret = 0;
506 struct nvsp_message sendMessage;
507 struct net_device *ndev;
508
509 net_device = get_outbound_net_device(device);
510 if (!net_device)
511 return -ENODEV;
512 ndev = net_device->ndev;
513
514 sendMessage.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
515 if (packet->is_data_pkt) {
516 /* 0 is RMC_DATA; */
517 sendMessage.msg.v1_msg.send_rndis_pkt.channel_type = 0;
518 } else {
519 /* 1 is RMC_CONTROL; */
520 sendMessage.msg.v1_msg.send_rndis_pkt.channel_type = 1;
521 }
522
523 /* Not using send buffer section */
524 sendMessage.msg.v1_msg.send_rndis_pkt.send_buf_section_index =
525 0xFFFFFFFF;
526 sendMessage.msg.v1_msg.send_rndis_pkt.send_buf_section_size = 0;
527
528 if (packet->page_buf_cnt) {
529 ret = vmbus_sendpacket_pagebuffer(device->channel,
530 packet->page_buf,
531 packet->page_buf_cnt,
532 &sendMessage,
533 sizeof(struct nvsp_message),
534 (unsigned long)packet);
535 } else {
536 ret = vmbus_sendpacket(device->channel, &sendMessage,
537 sizeof(struct nvsp_message),
538 (unsigned long)packet,
539 VM_PKT_DATA_INBAND,
540 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
541
542 }
543
544 if (ret == 0) {
545 atomic_inc(&net_device->num_outstanding_sends);
546 if (hv_ringbuf_avail_percent(&device->channel->outbound) <
547 RING_AVAIL_PERCENT_LOWATER) {
548 netif_stop_queue(ndev);
549 if (atomic_read(&net_device->
550 num_outstanding_sends) < 1)
551 netif_wake_queue(ndev);
552 }
553 } else if (ret == -EAGAIN) {
554 netif_stop_queue(ndev);
555 if (atomic_read(&net_device->num_outstanding_sends) < 1) {
556 netif_wake_queue(ndev);
557 ret = -ENOSPC;
558 }
559 } else {
560 netdev_err(ndev, "Unable to send packet %p ret %d\n",
561 packet, ret);
562 }
563
564 return ret;
565}
566
567static void netvsc_send_recv_completion(struct hv_device *device,
568 u64 transaction_id)
569{
570 struct nvsp_message recvcompMessage;
571 int retries = 0;
572 int ret;
573 struct net_device *ndev;
574 struct netvsc_device *net_device = hv_get_drvdata(device);
575
576 ndev = net_device->ndev;
577
578 recvcompMessage.hdr.msg_type =
579 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE;
580
581 /* FIXME: Pass in the status */
582 recvcompMessage.msg.v1_msg.send_rndis_pkt_complete.status =
583 NVSP_STAT_SUCCESS;
584
585retry_send_cmplt:
586 /* Send the completion */
587 ret = vmbus_sendpacket(device->channel, &recvcompMessage,
588 sizeof(struct nvsp_message), transaction_id,
589 VM_PKT_COMP, 0);
590 if (ret == 0) {
591 /* success */
592 /* no-op */
593 } else if (ret == -EAGAIN) {
594 /* no more room...wait a bit and attempt to retry 3 times */
595 retries++;
596 netdev_err(ndev, "unable to send receive completion pkt"
597 " (tid %llx)...retrying %d\n", transaction_id, retries);
598
599 if (retries < 4) {
600 udelay(100);
601 goto retry_send_cmplt;
602 } else {
603 netdev_err(ndev, "unable to send receive "
604 "completion pkt (tid %llx)...give up retrying\n",
605 transaction_id);
606 }
607 } else {
608 netdev_err(ndev, "unable to send receive "
609 "completion pkt - %llx\n", transaction_id);
610 }
611}
612
613/* Send a receive completion packet to RNDIS device (ie NetVsp) */
614static void netvsc_receive_completion(void *context)
615{
616 struct hv_netvsc_packet *packet = context;
617 struct hv_device *device = (struct hv_device *)packet->device;
618 struct netvsc_device *net_device;
619 u64 transaction_id = 0;
620 bool fsend_receive_comp = false;
621 unsigned long flags;
622 struct net_device *ndev;
623
624 /*
625 * Even though it seems logical to do a GetOutboundNetDevice() here to
626 * send out receive completion, we are using GetInboundNetDevice()
627 * since we may have disable outbound traffic already.
628 */
629 net_device = get_inbound_net_device(device);
630 if (!net_device)
631 return;
632 ndev = net_device->ndev;
633
634 /* Overloading use of the lock. */
635 spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
636
637 packet->xfer_page_pkt->count--;
638
639 /*
640 * Last one in the line that represent 1 xfer page packet.
641 * Return the xfer page packet itself to the freelist
642 */
643 if (packet->xfer_page_pkt->count == 0) {
644 fsend_receive_comp = true;
645 transaction_id = packet->completion.recv.recv_completion_tid;
646 list_add_tail(&packet->xfer_page_pkt->list_ent,
647 &net_device->recv_pkt_list);
648
649 }
650
651 /* Put the packet back */
652 list_add_tail(&packet->list_ent, &net_device->recv_pkt_list);
653 spin_unlock_irqrestore(&net_device->recv_pkt_list_lock, flags);
654
655 /* Send a receive completion for the xfer page packet */
656 if (fsend_receive_comp)
657 netvsc_send_recv_completion(device, transaction_id);
658
659}
660
661static void netvsc_receive(struct hv_device *device,
662 struct vmpacket_descriptor *packet)
663{
664 struct netvsc_device *net_device;
665 struct vmtransfer_page_packet_header *vmxferpage_packet;
666 struct nvsp_message *nvsp_packet;
667 struct hv_netvsc_packet *netvsc_packet = NULL;
668 /* struct netvsc_driver *netvscDriver; */
669 struct xferpage_packet *xferpage_packet = NULL;
670 int i;
671 int count = 0;
672 unsigned long flags;
673 struct net_device *ndev;
674
675 LIST_HEAD(listHead);
676
677 net_device = get_inbound_net_device(device);
678 if (!net_device)
679 return;
680 ndev = net_device->ndev;
681
682 /*
683 * All inbound packets other than send completion should be xfer page
684 * packet
685 */
686 if (packet->type != VM_PKT_DATA_USING_XFER_PAGES) {
687 netdev_err(ndev, "Unknown packet type received - %d\n",
688 packet->type);
689 return;
690 }
691
692 nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
693 (packet->offset8 << 3));
694
695 /* Make sure this is a valid nvsp packet */
696 if (nvsp_packet->hdr.msg_type !=
697 NVSP_MSG1_TYPE_SEND_RNDIS_PKT) {
698 netdev_err(ndev, "Unknown nvsp packet type received-"
699 " %d\n", nvsp_packet->hdr.msg_type);
700 return;
701 }
702
703 vmxferpage_packet = (struct vmtransfer_page_packet_header *)packet;
704
705 if (vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID) {
706 netdev_err(ndev, "Invalid xfer page set id - "
707 "expecting %x got %x\n", NETVSC_RECEIVE_BUFFER_ID,
708 vmxferpage_packet->xfer_pageset_id);
709 return;
710 }
711
712 /*
713 * Grab free packets (range count + 1) to represent this xfer
714 * page packet. +1 to represent the xfer page packet itself.
715 * We grab it here so that we know exactly how many we can
716 * fulfil
717 */
718 spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
719 while (!list_empty(&net_device->recv_pkt_list)) {
720 list_move_tail(net_device->recv_pkt_list.next, &listHead);
721 if (++count == vmxferpage_packet->range_cnt + 1)
722 break;
723 }
724 spin_unlock_irqrestore(&net_device->recv_pkt_list_lock, flags);
725
726 /*
727 * We need at least 2 netvsc pkts (1 to represent the xfer
728 * page and at least 1 for the range) i.e. we can handled
729 * some of the xfer page packet ranges...
730 */
731 if (count < 2) {
732 netdev_err(ndev, "Got only %d netvsc pkt...needed "
733 "%d pkts. Dropping this xfer page packet completely!\n",
734 count, vmxferpage_packet->range_cnt + 1);
735
736 /* Return it to the freelist */
737 spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
738 for (i = count; i != 0; i--) {
739 list_move_tail(listHead.next,
740 &net_device->recv_pkt_list);
741 }
742 spin_unlock_irqrestore(&net_device->recv_pkt_list_lock,
743 flags);
744
745 netvsc_send_recv_completion(device,
746 vmxferpage_packet->d.trans_id);
747
748 return;
749 }
750
751 /* Remove the 1st packet to represent the xfer page packet itself */
752 xferpage_packet = (struct xferpage_packet *)listHead.next;
753 list_del(&xferpage_packet->list_ent);
754
755 /* This is how much we can satisfy */
756 xferpage_packet->count = count - 1;
757
758 if (xferpage_packet->count != vmxferpage_packet->range_cnt) {
759 netdev_err(ndev, "Needed %d netvsc pkts to satisfy "
760 "this xfer page...got %d\n",
761 vmxferpage_packet->range_cnt, xferpage_packet->count);
762 }
763
764 /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
765 for (i = 0; i < (count - 1); i++) {
766 netvsc_packet = (struct hv_netvsc_packet *)listHead.next;
767 list_del(&netvsc_packet->list_ent);
768
769 /* Initialize the netvsc packet */
770 netvsc_packet->xfer_page_pkt = xferpage_packet;
771 netvsc_packet->completion.recv.recv_completion =
772 netvsc_receive_completion;
773 netvsc_packet->completion.recv.recv_completion_ctx =
774 netvsc_packet;
775 netvsc_packet->device = device;
776 /* Save this so that we can send it back */
777 netvsc_packet->completion.recv.recv_completion_tid =
778 vmxferpage_packet->d.trans_id;
779
780 netvsc_packet->data = (void *)((unsigned long)net_device->
781 recv_buf + vmxferpage_packet->ranges[i].byte_offset);
782 netvsc_packet->total_data_buflen =
783 vmxferpage_packet->ranges[i].byte_count;
784
785 /* Pass it to the upper layer */
786 rndis_filter_receive(device, netvsc_packet);
787
788 netvsc_receive_completion(netvsc_packet->
789 completion.recv.recv_completion_ctx);
790 }
791
792}
793
794static void netvsc_channel_cb(void *context)
795{
796 int ret;
797 struct hv_device *device = context;
798 struct netvsc_device *net_device;
799 u32 bytes_recvd;
800 u64 request_id;
801 unsigned char *packet;
802 struct vmpacket_descriptor *desc;
803 unsigned char *buffer;
804 int bufferlen = NETVSC_PACKET_SIZE;
805 struct net_device *ndev;
806
807 packet = kzalloc(NETVSC_PACKET_SIZE * sizeof(unsigned char),
808 GFP_ATOMIC);
809 if (!packet)
810 return;
811 buffer = packet;
812
813 net_device = get_inbound_net_device(device);
814 if (!net_device)
815 goto out;
816 ndev = net_device->ndev;
817
818 do {
819 ret = vmbus_recvpacket_raw(device->channel, buffer, bufferlen,
820 &bytes_recvd, &request_id);
821 if (ret == 0) {
822 if (bytes_recvd > 0) {
823 desc = (struct vmpacket_descriptor *)buffer;
824 switch (desc->type) {
825 case VM_PKT_COMP:
826 netvsc_send_completion(device, desc);
827 break;
828
829 case VM_PKT_DATA_USING_XFER_PAGES:
830 netvsc_receive(device, desc);
831 break;
832
833 default:
834 netdev_err(ndev,
835 "unhandled packet type %d, "
836 "tid %llx len %d\n",
837 desc->type, request_id,
838 bytes_recvd);
839 break;
840 }
841
842 /* reset */
843 if (bufferlen > NETVSC_PACKET_SIZE) {
844 kfree(buffer);
845 buffer = packet;
846 bufferlen = NETVSC_PACKET_SIZE;
847 }
848 } else {
849 /* reset */
850 if (bufferlen > NETVSC_PACKET_SIZE) {
851 kfree(buffer);
852 buffer = packet;
853 bufferlen = NETVSC_PACKET_SIZE;
854 }
855
856 break;
857 }
858 } else if (ret == -ENOBUFS) {
859 /* Handle large packet */
860 buffer = kmalloc(bytes_recvd, GFP_ATOMIC);
861 if (buffer == NULL) {
862 /* Try again next time around */
863 netdev_err(ndev,
864 "unable to allocate buffer of size "
865 "(%d)!!\n", bytes_recvd);
866 break;
867 }
868
869 bufferlen = bytes_recvd;
870 }
871 } while (1);
872
873out:
874 kfree(buffer);
875 return;
876}
877
878/*
879 * netvsc_device_add - Callback when the device belonging to this
880 * driver is added
881 */
882int netvsc_device_add(struct hv_device *device, void *additional_info)
883{
884 int ret = 0;
885 int i;
886 int ring_size =
887 ((struct netvsc_device_info *)additional_info)->ring_size;
888 struct netvsc_device *net_device;
889 struct hv_netvsc_packet *packet, *pos;
890 struct net_device *ndev;
891
892 net_device = alloc_net_device(device);
893 if (!net_device) {
894 ret = -ENOMEM;
895 goto cleanup;
896 }
897
898 /*
899 * Coming into this function, struct net_device * is
900 * registered as the driver private data.
901 * In alloc_net_device(), we register struct netvsc_device *
902 * as the driver private data and stash away struct net_device *
903 * in struct netvsc_device *.
904 */
905 ndev = net_device->ndev;
906
907 /* Initialize the NetVSC channel extension */
908 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
909 spin_lock_init(&net_device->recv_pkt_list_lock);
910
911 INIT_LIST_HEAD(&net_device->recv_pkt_list);
912
913 for (i = 0; i < NETVSC_RECEIVE_PACKETLIST_COUNT; i++) {
914 packet = kzalloc(sizeof(struct hv_netvsc_packet) +
915 (NETVSC_RECEIVE_SG_COUNT *
916 sizeof(struct hv_page_buffer)), GFP_KERNEL);
917 if (!packet)
918 break;
919
920 list_add_tail(&packet->list_ent,
921 &net_device->recv_pkt_list);
922 }
923 init_completion(&net_device->channel_init_wait);
924
925 /* Open the channel */
926 ret = vmbus_open(device->channel, ring_size * PAGE_SIZE,
927 ring_size * PAGE_SIZE, NULL, 0,
928 netvsc_channel_cb, device);
929
930 if (ret != 0) {
931 netdev_err(ndev, "unable to open channel: %d\n", ret);
932 goto cleanup;
933 }
934
935 /* Channel is opened */
936 pr_info("hv_netvsc channel opened successfully\n");
937
938 /* Connect with the NetVsp */
939 ret = netvsc_connect_vsp(device);
940 if (ret != 0) {
941 netdev_err(ndev,
942 "unable to connect to NetVSP - %d\n", ret);
943 goto close;
944 }
945
946 return ret;
947
948close:
949 /* Now, we can close the channel safely */
950 vmbus_close(device->channel);
951
952cleanup:
953
954 if (net_device) {
955 list_for_each_entry_safe(packet, pos,
956 &net_device->recv_pkt_list,
957 list_ent) {
958 list_del(&packet->list_ent);
959 kfree(packet);
960 }
961
962 kfree(net_device);
963 }
964
965 return ret;
966}