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