Merge tag 'kbuild-fixes-v6.10' of git://git.kernel.org/pub/scm/linux/kernel/git/masah...
[linux-2.6-block.git] / drivers / net / hyperv / netvsc.c
CommitLineData
9952f691 1// SPDX-License-Identifier: GPL-2.0-only
fceaf24a 2/*
fceaf24a
HJ
3 * Copyright (c) 2009, Microsoft Corporation.
4 *
fceaf24a 5 * Authors:
d0e94d17 6 * Haiyang Zhang <haiyangz@microsoft.com>
fceaf24a 7 * Hank Janssen <hjanssen@microsoft.com>
fceaf24a 8 */
eb335bc4
HJ
9#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10
5654e932 11#include <linux/kernel.h>
0c3b7b2f
S
12#include <linux/sched.h>
13#include <linux/wait.h>
0ffa63b0 14#include <linux/mm.h>
b4362c9c 15#include <linux/delay.h>
21a80820 16#include <linux/io.h>
5a0e3ad6 17#include <linux/slab.h>
d9871158 18#include <linux/netdevice.h>
f157e78d 19#include <linux/if_ether.h>
d6472302 20#include <linux/vmalloc.h>
9749fed5 21#include <linux/rtnetlink.h>
43bf99ce 22#include <linux/prefetch.h>
1cb9d3b6 23#include <linux/filter.h>
9749fed5 24
c25aaf81 25#include <asm/sync_bitops.h>
96854bbd 26#include <asm/mshyperv.h>
3f335ea2 27
5ca7252a 28#include "hyperv_net.h"
ec966381 29#include "netvsc_trace.h"
fceaf24a 30
84bf9cef
KS
31/*
32 * Switch the data path from the synthetic interface to the VF
33 * interface.
34 */
d0922bf7 35int netvsc_switch_datapath(struct net_device *ndev, bool vf)
84bf9cef 36{
3d541ac5
VK
37 struct net_device_context *net_device_ctx = netdev_priv(ndev);
38 struct hv_device *dev = net_device_ctx->device_ctx;
79e8cbe7 39 struct netvsc_device *nv_dev = rtnl_dereference(net_device_ctx->nvdev);
0a1275ca 40 struct nvsp_message *init_pkt = &nv_dev->channel_init_pkt;
d0922bf7 41 int ret, retry = 0;
84bf9cef 42
8b31f8c9
LL
43 /* Block sending traffic to VF if it's about to be gone */
44 if (!vf)
45 net_device_ctx->data_path_is_vf = vf;
46
84bf9cef
KS
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
d0922bf7 56again:
ec966381
SH
57 trace_nvsp_send(ndev, init_pkt);
58
d0922bf7 59 ret = vmbus_sendpacket(dev->channel, init_pkt,
84bf9cef 60 sizeof(struct nvsp_message),
d0922bf7 61 (unsigned long)init_pkt, VM_PKT_DATA_INBAND,
8b31f8c9 62 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
d0922bf7
HZ
63
64 /* If failed to switch to/from VF, let data_path_is_vf stay false,
65 * so we use synthetic path to send data.
66 */
67 if (ret) {
68 if (ret != -EAGAIN) {
69 netdev_err(ndev,
70 "Unable to send sw datapath msg, err: %d\n",
71 ret);
72 return ret;
73 }
74
75 if (retry++ < RETRY_MAX) {
76 usleep_range(RETRY_US_LO, RETRY_US_HI);
77 goto again;
78 } else {
79 netdev_err(
80 ndev,
81 "Retry failed to send sw datapath msg, err: %d\n",
82 ret);
83 return ret;
84 }
85 }
86
8b31f8c9
LL
87 wait_for_completion(&nv_dev->channel_init_wait);
88 net_device_ctx->data_path_is_vf = vf;
d0922bf7
HZ
89
90 return 0;
84bf9cef
KS
91}
92
3ffe64f1
SH
93/* Worker to setup sub channels on initial setup
94 * Initial hotplug event occurs in softirq context
95 * and can't wait for channels.
96 */
97static void netvsc_subchan_work(struct work_struct *w)
98{
99 struct netvsc_device *nvdev =
100 container_of(w, struct netvsc_device, subchan_work);
101 struct rndis_device *rdev;
102 int i, ret;
103
104 /* Avoid deadlock with device removal already under RTNL */
105 if (!rtnl_trylock()) {
106 schedule_work(w);
107 return;
108 }
109
110 rdev = nvdev->extension;
111 if (rdev) {
17d91256 112 ret = rndis_set_subchannel(rdev->ndev, nvdev, NULL);
3ffe64f1
SH
113 if (ret == 0) {
114 netif_device_attach(rdev->ndev);
115 } else {
116 /* fallback to only primary channel */
117 for (i = 1; i < nvdev->num_chn; i++)
118 netif_napi_del(&nvdev->chan_table[i].napi);
119
120 nvdev->max_chn = 1;
121 nvdev->num_chn = 1;
122 }
123 }
124
125 rtnl_unlock();
126}
127
88098834 128static struct netvsc_device *alloc_net_device(void)
fceaf24a 129{
85799a37 130 struct netvsc_device *net_device;
fceaf24a 131
85799a37
HZ
132 net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
133 if (!net_device)
fceaf24a
HJ
134 return NULL;
135
dc5cd894 136 init_waitqueue_head(&net_device->wait_drain);
c38b9c71 137 net_device->destroy = false;
f6f13c12 138 net_device->tx_disable = true;
0da6edbd 139
7c3877f2
HZ
140 net_device->max_pkt = RNDIS_MAX_PKT_DEFAULT;
141 net_device->pkt_align = RNDIS_PKT_ALIGN_DEFAULT;
8b532797 142
fd612602 143 init_completion(&net_device->channel_init_wait);
732e4985 144 init_waitqueue_head(&net_device->subchan_open);
3ffe64f1 145 INIT_WORK(&net_device->subchan_work, netvsc_subchan_work);
7c3877f2 146
85799a37 147 return net_device;
fceaf24a
HJ
148}
149
545a8e79 150static void free_netvsc_device(struct rcu_head *head)
f90251c8 151{
545a8e79 152 struct netvsc_device *nvdev
153 = container_of(head, struct netvsc_device, rcu);
c0b558e5
HZ
154 int i;
155
02400fce 156 kfree(nvdev->extension);
bbf9ac34
RE
157
158 if (!nvdev->recv_buf_gpadl_handle.decrypted)
159 vfree(nvdev->recv_buf);
160 if (!nvdev->send_buf_gpadl_handle.decrypted)
161 vfree(nvdev->send_buf);
e9268a94 162 bitmap_free(nvdev->send_section_map);
02400fce 163
351e1581
HZ
164 for (i = 0; i < VRSS_CHANNEL_MAX; i++) {
165 xdp_rxq_info_unreg(&nvdev->chan_table[i].xdp_rxq);
0ba35fe9 166 kfree(nvdev->chan_table[i].recv_buf);
7426b1a5 167 vfree(nvdev->chan_table[i].mrc.slots);
351e1581 168 }
c0b558e5 169
f90251c8
HZ
170 kfree(nvdev);
171}
172
545a8e79 173static void free_netvsc_device_rcu(struct netvsc_device *nvdev)
174{
175 call_rcu(&nvdev->rcu, free_netvsc_device);
176}
fceaf24a 177
7992894c 178static void netvsc_revoke_recv_buf(struct hv_device *device,
3f076eff
MG
179 struct netvsc_device *net_device,
180 struct net_device *ndev)
ec91cd09 181{
7992894c 182 struct nvsp_message *revoke_packet;
7a2a0a84 183 int ret;
ec91cd09
HZ
184
185 /*
186 * If we got a section count, it means we received a
187 * SendReceiveBufferComplete msg (ie sent
188 * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
189 * to send a revoke msg here
190 */
191 if (net_device->recv_section_cnt) {
192 /* Send the revoke receive buffer */
193 revoke_packet = &net_device->revoke_packet;
194 memset(revoke_packet, 0, sizeof(struct nvsp_message));
195
196 revoke_packet->hdr.msg_type =
197 NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
198 revoke_packet->msg.v1_msg.
199 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
200
ec966381
SH
201 trace_nvsp_send(ndev, revoke_packet);
202
3d541ac5 203 ret = vmbus_sendpacket(device->channel,
ec91cd09
HZ
204 revoke_packet,
205 sizeof(struct nvsp_message),
4d18fcc9 206 VMBUS_RQST_ID_NO_RESPONSE,
ec91cd09 207 VM_PKT_DATA_INBAND, 0);
73e64fa4
S
208 /* If the failure is because the channel is rescinded;
209 * ignore the failure since we cannot send on a rescinded
210 * channel. This would allow us to properly cleanup
211 * even when the channel is rescinded.
212 */
213 if (device->channel->rescind)
214 ret = 0;
ec91cd09
HZ
215 /*
216 * If we failed here, we might as well return and
217 * have a leak rather than continue and a bugchk
218 */
219 if (ret != 0) {
d9871158 220 netdev_err(ndev, "unable to send "
c909ebbd 221 "revoke receive buffer to netvsp\n");
7a2a0a84 222 return;
ec91cd09 223 }
8b532797 224 net_device->recv_section_cnt = 0;
ec91cd09 225 }
7992894c
MG
226}
227
228static void netvsc_revoke_send_buf(struct hv_device *device,
3f076eff
MG
229 struct netvsc_device *net_device,
230 struct net_device *ndev)
7992894c 231{
7992894c
MG
232 struct nvsp_message *revoke_packet;
233 int ret;
ec91cd09 234
c25aaf81
KS
235 /* Deal with the send buffer we may have setup.
236 * If we got a send section size, it means we received a
c51ed182
HZ
237 * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
238 * NVSP_MSG1_TYPE_SEND_SEND_BUF msg) therefore, we need
c25aaf81
KS
239 * to send a revoke msg here
240 */
8b532797 241 if (net_device->send_section_cnt) {
c25aaf81
KS
242 /* Send the revoke receive buffer */
243 revoke_packet = &net_device->revoke_packet;
244 memset(revoke_packet, 0, sizeof(struct nvsp_message));
245
246 revoke_packet->hdr.msg_type =
247 NVSP_MSG1_TYPE_REVOKE_SEND_BUF;
c51ed182
HZ
248 revoke_packet->msg.v1_msg.revoke_send_buf.id =
249 NETVSC_SEND_BUFFER_ID;
c25aaf81 250
ec966381
SH
251 trace_nvsp_send(ndev, revoke_packet);
252
3d541ac5 253 ret = vmbus_sendpacket(device->channel,
c25aaf81
KS
254 revoke_packet,
255 sizeof(struct nvsp_message),
4d18fcc9 256 VMBUS_RQST_ID_NO_RESPONSE,
c25aaf81 257 VM_PKT_DATA_INBAND, 0);
73e64fa4
S
258
259 /* If the failure is because the channel is rescinded;
260 * ignore the failure since we cannot send on a rescinded
261 * channel. This would allow us to properly cleanup
262 * even when the channel is rescinded.
263 */
264 if (device->channel->rescind)
265 ret = 0;
266
c25aaf81
KS
267 /* If we failed here, we might as well return and
268 * have a leak rather than continue and a bugchk
269 */
270 if (ret != 0) {
271 netdev_err(ndev, "unable to send "
272 "revoke send buffer to netvsp\n");
7a2a0a84 273 return;
c25aaf81 274 }
8b532797 275 net_device->send_section_cnt = 0;
c25aaf81 276 }
0cf73780
VK
277}
278
7992894c 279static void netvsc_teardown_recv_gpadl(struct hv_device *device,
3f076eff
MG
280 struct netvsc_device *net_device,
281 struct net_device *ndev)
0cf73780 282{
0cf73780
VK
283 int ret;
284
d4dccf35 285 if (net_device->recv_buf_gpadl_handle.gpadl_handle) {
0cf73780 286 ret = vmbus_teardown_gpadl(device->channel,
d4dccf35 287 &net_device->recv_buf_gpadl_handle);
0cf73780
VK
288
289 /* If we failed here, we might as well return and have a leak
290 * rather than continue and a bugchk
291 */
292 if (ret != 0) {
293 netdev_err(ndev,
294 "unable to teardown receive buffer's gpadl\n");
295 return;
296 }
0cf73780 297 }
7992894c
MG
298}
299
300static void netvsc_teardown_send_gpadl(struct hv_device *device,
3f076eff
MG
301 struct netvsc_device *net_device,
302 struct net_device *ndev)
7992894c 303{
7992894c 304 int ret;
0cf73780 305
d4dccf35 306 if (net_device->send_buf_gpadl_handle.gpadl_handle) {
3d541ac5 307 ret = vmbus_teardown_gpadl(device->channel,
d4dccf35 308 &net_device->send_buf_gpadl_handle);
c25aaf81
KS
309
310 /* If we failed here, we might as well return and have a leak
311 * rather than continue and a bugchk
312 */
313 if (ret != 0) {
314 netdev_err(ndev,
315 "unable to teardown send buffer's gpadl\n");
7a2a0a84 316 return;
c25aaf81 317 }
c25aaf81 318 }
ec91cd09
HZ
319}
320
7426b1a5 321int netvsc_alloc_recv_comp_ring(struct netvsc_device *net_device, u32 q_idx)
322{
323 struct netvsc_channel *nvchan = &net_device->chan_table[q_idx];
324 int node = cpu_to_node(nvchan->channel->target_cpu);
325 size_t size;
326
327 size = net_device->recv_completion_cnt * sizeof(struct recv_comp_data);
328 nvchan->mrc.slots = vzalloc_node(size, node);
329 if (!nvchan->mrc.slots)
330 nvchan->mrc.slots = vzalloc(size);
331
332 return nvchan->mrc.slots ? 0 : -ENOMEM;
333}
334
95790837 335static int netvsc_init_buf(struct hv_device *device,
8b532797 336 struct netvsc_device *net_device,
337 const struct netvsc_device_info *device_info)
fceaf24a 338{
7426b1a5 339 struct nvsp_1_message_send_receive_buffer_complete *resp;
95833370 340 struct net_device *ndev = hv_get_drvdata(device);
341 struct nvsp_message *init_packet;
8b532797 342 unsigned int buf_size;
0102eeed 343 int i, ret = 0;
0a726c2b 344
8b532797 345 /* Get receive buffer area. */
0ab09bef 346 buf_size = device_info->recv_sections * device_info->recv_section_size;
8b532797 347 buf_size = roundup(buf_size, PAGE_SIZE);
348
11b2b653
HZ
349 /* Legacy hosts only allow smaller receive buffer */
350 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
351 buf_size = min_t(unsigned int, buf_size,
352 NETVSC_RECEIVE_BUFFER_SIZE_LEGACY);
353
8b532797 354 net_device->recv_buf = vzalloc(buf_size);
53d21fdb 355 if (!net_device->recv_buf) {
8b532797 356 netdev_err(ndev,
357 "unable to allocate receive buffer of size %u\n",
358 buf_size);
927bc33c 359 ret = -ENOMEM;
0c3b7b2f 360 goto cleanup;
fceaf24a 361 }
fceaf24a 362
c5d24bdd
HZ
363 net_device->recv_buf_size = buf_size;
364
454f18a9
BP
365 /*
366 * Establish the gpadl handle for this buffer on this
367 * channel. Note: This call uses the vmbus connection rather
368 * than the channel to establish the gpadl handle.
369 */
53d21fdb 370 ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
8b532797 371 buf_size,
53d21fdb 372 &net_device->recv_buf_gpadl_handle);
21a80820 373 if (ret != 0) {
d9871158 374 netdev_err(ndev,
c909ebbd 375 "unable to establish receive buffer's gpadl\n");
0c3b7b2f 376 goto cleanup;
fceaf24a
HJ
377 }
378
454f18a9 379 /* Notify the NetVsp of the gpadl handle */
53d21fdb 380 init_packet = &net_device->channel_init_pkt;
85799a37 381 memset(init_packet, 0, sizeof(struct nvsp_message));
53d21fdb
HZ
382 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
383 init_packet->msg.v1_msg.send_recv_buf.
d4dccf35 384 gpadl_handle = net_device->recv_buf_gpadl_handle.gpadl_handle;
53d21fdb
HZ
385 init_packet->msg.v1_msg.
386 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
fceaf24a 387
ec966381
SH
388 trace_nvsp_send(ndev, init_packet);
389
454f18a9 390 /* Send the gpadl notification request */
85799a37 391 ret = vmbus_sendpacket(device->channel, init_packet,
5a4df290 392 sizeof(struct nvsp_message),
85799a37 393 (unsigned long)init_packet,
415f2287 394 VM_PKT_DATA_INBAND,
5a4df290 395 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
21a80820 396 if (ret != 0) {
d9871158 397 netdev_err(ndev,
c909ebbd 398 "unable to send receive buffer's gpadl to netvsp\n");
0c3b7b2f 399 goto cleanup;
fceaf24a
HJ
400 }
401
5362855a 402 wait_for_completion(&net_device->channel_init_wait);
fceaf24a 403
454f18a9 404 /* Check the response */
7426b1a5 405 resp = &init_packet->msg.v1_msg.send_recv_buf_complete;
406 if (resp->status != NVSP_STAT_SUCCESS) {
407 netdev_err(ndev,
408 "Unable to complete receive buffer initialization with NetVsp - status %d\n",
409 resp->status);
927bc33c 410 ret = -EINVAL;
0c3b7b2f 411 goto cleanup;
fceaf24a
HJ
412 }
413
454f18a9 414 /* Parse the response */
7426b1a5 415 netdev_dbg(ndev, "Receive sections: %u sub_allocs: size %u count: %u\n",
416 resp->num_sections, resp->sections[0].sub_alloc_size,
417 resp->sections[0].num_sub_allocs);
fceaf24a 418
8b532797 419 /* There should only be one section for the entire receive buffer */
420 if (resp->num_sections != 1 || resp->sections[0].offset != 0) {
927bc33c 421 ret = -EINVAL;
0c3b7b2f 422 goto cleanup;
fceaf24a
HJ
423 }
424
8b532797 425 net_device->recv_section_size = resp->sections[0].sub_alloc_size;
426 net_device->recv_section_cnt = resp->sections[0].num_sub_allocs;
427
44144185
AB
428 /* Ensure buffer will not overflow */
429 if (net_device->recv_section_size < NETVSC_MTU_MIN || (u64)net_device->recv_section_size *
430 (u64)net_device->recv_section_cnt > (u64)buf_size) {
431 netdev_err(ndev, "invalid recv_section_size %u\n",
432 net_device->recv_section_size);
433 ret = -EINVAL;
434 goto cleanup;
435 }
436
0102eeed
APM
437 for (i = 0; i < VRSS_CHANNEL_MAX; i++) {
438 struct netvsc_channel *nvchan = &net_device->chan_table[i];
439
440 nvchan->recv_buf = kzalloc(net_device->recv_section_size, GFP_KERNEL);
441 if (nvchan->recv_buf == NULL) {
442 ret = -ENOMEM;
443 goto cleanup;
444 }
445 }
446
f87238d3
HZ
447 /* Setup receive completion ring.
448 * Add 1 to the recv_section_cnt because at least one entry in a
449 * ring buffer has to be empty.
450 */
451 net_device->recv_completion_cnt = net_device->recv_section_cnt + 1;
7426b1a5 452 ret = netvsc_alloc_recv_comp_ring(net_device, 0);
453 if (ret)
454 goto cleanup;
455
456 /* Now setup the send buffer. */
0ab09bef 457 buf_size = device_info->send_sections * device_info->send_section_size;
8b532797 458 buf_size = round_up(buf_size, PAGE_SIZE);
459
460 net_device->send_buf = vzalloc(buf_size);
c25aaf81 461 if (!net_device->send_buf) {
8b532797 462 netdev_err(ndev, "unable to allocate send buffer of size %u\n",
463 buf_size);
c25aaf81
KS
464 ret = -ENOMEM;
465 goto cleanup;
466 }
d4dccf35 467 net_device->send_buf_size = buf_size;
c25aaf81
KS
468
469 /* Establish the gpadl handle for this buffer on this
470 * channel. Note: This call uses the vmbus connection rather
471 * than the channel to establish the gpadl handle.
472 */
473 ret = vmbus_establish_gpadl(device->channel, net_device->send_buf,
8b532797 474 buf_size,
c25aaf81
KS
475 &net_device->send_buf_gpadl_handle);
476 if (ret != 0) {
477 netdev_err(ndev,
478 "unable to establish send buffer's gpadl\n");
479 goto cleanup;
480 }
481
482 /* Notify the NetVsp of the gpadl handle */
483 init_packet = &net_device->channel_init_pkt;
484 memset(init_packet, 0, sizeof(struct nvsp_message));
485 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_SEND_BUF;
c51ed182 486 init_packet->msg.v1_msg.send_send_buf.gpadl_handle =
d4dccf35 487 net_device->send_buf_gpadl_handle.gpadl_handle;
c51ed182 488 init_packet->msg.v1_msg.send_send_buf.id = NETVSC_SEND_BUFFER_ID;
c25aaf81 489
ec966381
SH
490 trace_nvsp_send(ndev, init_packet);
491
c25aaf81
KS
492 /* Send the gpadl notification request */
493 ret = vmbus_sendpacket(device->channel, init_packet,
494 sizeof(struct nvsp_message),
495 (unsigned long)init_packet,
496 VM_PKT_DATA_INBAND,
497 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
498 if (ret != 0) {
499 netdev_err(ndev,
500 "unable to send send buffer's gpadl to netvsp\n");
501 goto cleanup;
502 }
503
5362855a 504 wait_for_completion(&net_device->channel_init_wait);
c25aaf81
KS
505
506 /* Check the response */
507 if (init_packet->msg.v1_msg.
508 send_send_buf_complete.status != NVSP_STAT_SUCCESS) {
509 netdev_err(ndev, "Unable to complete send buffer "
510 "initialization with NetVsp - status %d\n",
511 init_packet->msg.v1_msg.
c51ed182 512 send_send_buf_complete.status);
c25aaf81
KS
513 ret = -EINVAL;
514 goto cleanup;
515 }
516
517 /* Parse the response */
518 net_device->send_section_size = init_packet->msg.
519 v1_msg.send_send_buf_complete.section_size;
44144185
AB
520 if (net_device->send_section_size < NETVSC_MTU_MIN) {
521 netdev_err(ndev, "invalid send_section_size %u\n",
522 net_device->send_section_size);
523 ret = -EINVAL;
524 goto cleanup;
525 }
c25aaf81 526
8b532797 527 /* Section count is simply the size divided by the section size. */
528 net_device->send_section_cnt = buf_size / net_device->send_section_size;
c25aaf81 529
93ba2222
VK
530 netdev_dbg(ndev, "Send section size: %d, Section count:%d\n",
531 net_device->send_section_size, net_device->send_section_cnt);
c25aaf81
KS
532
533 /* Setup state for managing the send buffer. */
e9268a94
CJ
534 net_device->send_section_map = bitmap_zalloc(net_device->send_section_cnt,
535 GFP_KERNEL);
536 if (!net_device->send_section_map) {
dd1d3f8f 537 ret = -ENOMEM;
c25aaf81 538 goto cleanup;
dd1d3f8f 539 }
c25aaf81 540
0c3b7b2f 541 goto exit;
fceaf24a 542
0c3b7b2f 543cleanup:
3f076eff
MG
544 netvsc_revoke_recv_buf(device, net_device, ndev);
545 netvsc_revoke_send_buf(device, net_device, ndev);
546 netvsc_teardown_recv_gpadl(device, net_device, ndev);
547 netvsc_teardown_send_gpadl(device, net_device, ndev);
fceaf24a 548
0c3b7b2f 549exit:
fceaf24a
HJ
550 return ret;
551}
552
f157e78d
HZ
553/* Negotiate NVSP protocol version */
554static int negotiate_nvsp_ver(struct hv_device *device,
555 struct netvsc_device *net_device,
556 struct nvsp_message *init_packet,
557 u32 nvsp_ver)
fceaf24a 558{
0a1275ca 559 struct net_device *ndev = hv_get_drvdata(device);
7390fe9c 560 int ret;
fceaf24a 561
85799a37 562 memset(init_packet, 0, sizeof(struct nvsp_message));
53d21fdb 563 init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
f157e78d
HZ
564 init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
565 init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
ec966381
SH
566 trace_nvsp_send(ndev, init_packet);
567
454f18a9 568 /* Send the init request */
85799a37 569 ret = vmbus_sendpacket(device->channel, init_packet,
5a4df290 570 sizeof(struct nvsp_message),
85799a37 571 (unsigned long)init_packet,
415f2287 572 VM_PKT_DATA_INBAND,
5a4df290 573 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
21a80820 574
b8a3d52b 575 if (ret != 0)
f157e78d 576 return ret;
fceaf24a 577
5362855a 578 wait_for_completion(&net_device->channel_init_wait);
fceaf24a 579
53d21fdb 580 if (init_packet->msg.init_msg.init_complete.status !=
f157e78d
HZ
581 NVSP_STAT_SUCCESS)
582 return -EINVAL;
fceaf24a 583
a1eabb01 584 if (nvsp_ver == NVSP_PROTOCOL_VERSION_1)
f157e78d
HZ
585 return 0;
586
71790a27 587 /* NVSPv2 or later: Send NDIS config */
f157e78d
HZ
588 memset(init_packet, 0, sizeof(struct nvsp_message));
589 init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
0a1275ca 590 init_packet->msg.v2_msg.send_ndis_config.mtu = ndev->mtu + ETH_HLEN;
1f5f3a75 591 init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
f157e78d 592
7f5d5af0 593 if (nvsp_ver >= NVSP_PROTOCOL_VERSION_5) {
96854bbd
APM
594 if (hv_is_isolation_supported())
595 netdev_info(ndev, "SR-IOV not advertised by guests on the host supporting isolation\n");
596 else
597 init_packet->msg.v2_msg.send_ndis_config.capability.sriov = 1;
71790a27 598
7f5d5af0
HZ
599 /* Teaming bit is needed to receive link speed updates */
600 init_packet->msg.v2_msg.send_ndis_config.capability.teaming = 1;
601 }
602
c8e4eff4
HZ
603 if (nvsp_ver >= NVSP_PROTOCOL_VERSION_61)
604 init_packet->msg.v2_msg.send_ndis_config.capability.rsc = 1;
605
ec966381
SH
606 trace_nvsp_send(ndev, init_packet);
607
f157e78d
HZ
608 ret = vmbus_sendpacket(device->channel, init_packet,
609 sizeof(struct nvsp_message),
4d18fcc9 610 VMBUS_RQST_ID_NO_RESPONSE,
f157e78d
HZ
611 VM_PKT_DATA_INBAND, 0);
612
613 return ret;
614}
615
95790837 616static int netvsc_connect_vsp(struct hv_device *device,
8b532797 617 struct netvsc_device *net_device,
618 const struct netvsc_device_info *device_info)
f157e78d 619{
ec966381 620 struct net_device *ndev = hv_get_drvdata(device);
1b17ca04 621 static const u32 ver_list[] = {
e5a78fad 622 NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
0dcec221
HZ
623 NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5,
624 NVSP_PROTOCOL_VERSION_6, NVSP_PROTOCOL_VERSION_61
95790837 625 };
626 struct nvsp_message *init_packet;
627 int ndis_version, i, ret;
f157e78d
HZ
628
629 init_packet = &net_device->channel_init_pkt;
630
631 /* Negotiate the latest NVSP protocol supported */
e5a78fad 632 for (i = ARRAY_SIZE(ver_list) - 1; i >= 0; i--)
a1eabb01
HZ
633 if (negotiate_nvsp_ver(device, net_device, init_packet,
634 ver_list[i]) == 0) {
635 net_device->nvsp_version = ver_list[i];
636 break;
637 }
638
639 if (i < 0) {
0f48c72c 640 ret = -EPROTO;
0c3b7b2f 641 goto cleanup;
fceaf24a 642 }
f157e78d 643
96854bbd
APM
644 if (hv_is_isolation_supported() && net_device->nvsp_version < NVSP_PROTOCOL_VERSION_61) {
645 netdev_err(ndev, "Invalid NVSP version 0x%x (expected >= 0x%x) from the host supporting isolation\n",
646 net_device->nvsp_version, NVSP_PROTOCOL_VERSION_61);
647 ret = -EPROTO;
648 goto cleanup;
649 }
650
f157e78d
HZ
651 pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
652
454f18a9 653 /* Send the ndis version */
85799a37 654 memset(init_packet, 0, sizeof(struct nvsp_message));
fceaf24a 655
a1eabb01 656 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
1f73db49 657 ndis_version = 0x00060001;
a1eabb01
HZ
658 else
659 ndis_version = 0x0006001e;
fceaf24a 660
53d21fdb
HZ
661 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
662 init_packet->msg.v1_msg.
663 send_ndis_ver.ndis_major_ver =
85799a37 664 (ndis_version & 0xFFFF0000) >> 16;
53d21fdb
HZ
665 init_packet->msg.v1_msg.
666 send_ndis_ver.ndis_minor_ver =
85799a37 667 ndis_version & 0xFFFF;
fceaf24a 668
ec966381
SH
669 trace_nvsp_send(ndev, init_packet);
670
454f18a9 671 /* Send the init request */
85799a37 672 ret = vmbus_sendpacket(device->channel, init_packet,
0c3b7b2f 673 sizeof(struct nvsp_message),
4d18fcc9 674 VMBUS_RQST_ID_NO_RESPONSE,
0c3b7b2f 675 VM_PKT_DATA_INBAND, 0);
0f48c72c 676 if (ret != 0)
0c3b7b2f 677 goto cleanup;
454f18a9 678
99d3016d 679
8b532797 680 ret = netvsc_init_buf(device, net_device, device_info);
fceaf24a 681
0c3b7b2f 682cleanup:
fceaf24a
HJ
683 return ret;
684}
685
3e189519 686/*
5a71ae30 687 * netvsc_device_remove - Callback when the root bus device is removed
21a80820 688 */
e08f3ea5 689void netvsc_device_remove(struct hv_device *device)
fceaf24a 690{
3d541ac5
VK
691 struct net_device *ndev = hv_get_drvdata(device);
692 struct net_device_context *net_device_ctx = netdev_priv(ndev);
79e8cbe7 693 struct netvsc_device *net_device
694 = rtnl_dereference(net_device_ctx->nvdev);
15a863bf 695 int i;
fceaf24a 696
a56d99d7
MG
697 /*
698 * Revoke receive buffer. If host is pre-Win2016 then tear down
699 * receive buffer GPADL. Do the same for send buffer.
700 */
3f076eff 701 netvsc_revoke_recv_buf(device, net_device, ndev);
a56d99d7 702 if (vmbus_proto_version < VERSION_WIN10)
3f076eff 703 netvsc_teardown_recv_gpadl(device, net_device, ndev);
a56d99d7 704
3f076eff 705 netvsc_revoke_send_buf(device, net_device, ndev);
a56d99d7 706 if (vmbus_proto_version < VERSION_WIN10)
3f076eff 707 netvsc_teardown_send_gpadl(device, net_device, ndev);
9d88f33a 708
545a8e79 709 RCU_INIT_POINTER(net_device_ctx->nvdev, NULL);
3852409b 710
ac504767
APM
711 /* Disable NAPI and disassociate its context from the device. */
712 for (i = 0; i < net_device->num_chn; i++) {
713 /* See also vmbus_reset_channel_cb(). */
e0526ec5
SC
714 /* only disable enabled NAPI channel */
715 if (i < ndev->real_num_rx_queues)
716 napi_disable(&net_device->chan_table[i].napi);
717
8348e046 718 netif_napi_del(&net_device->chan_table[i].napi);
ac504767 719 }
8348e046 720
86c921af
S
721 /*
722 * At this point, no one should be accessing net_device
723 * except in here
724 */
93ba2222 725 netdev_dbg(ndev, "net device safe to remove\n");
fceaf24a 726
454f18a9 727 /* Now, we can close the channel safely */
85799a37 728 vmbus_close(device->channel);
fceaf24a 729
a56d99d7
MG
730 /*
731 * If host is Win2016 or higher then we do the GPADL tear down
732 * here after VMBus is closed.
733 */
7992894c 734 if (vmbus_proto_version >= VERSION_WIN10) {
3f076eff
MG
735 netvsc_teardown_recv_gpadl(device, net_device, ndev);
736 netvsc_teardown_send_gpadl(device, net_device, ndev);
7992894c 737 }
15a863bf 738
454f18a9 739 /* Release all resources */
545a8e79 740 free_netvsc_device_rcu(net_device);
fceaf24a
HJ
741}
742
33be96e4
HZ
743#define RING_AVAIL_PERCENT_HIWATER 20
744#define RING_AVAIL_PERCENT_LOWATER 10
745
c25aaf81
KS
746static inline void netvsc_free_send_slot(struct netvsc_device *net_device,
747 u32 index)
748{
749 sync_change_bit(index, net_device->send_section_map);
750}
751
c347b927
SH
752static void netvsc_send_tx_complete(struct net_device *ndev,
753 struct netvsc_device *net_device,
754 struct vmbus_channel *channel,
f9645430 755 const struct vmpacket_descriptor *desc,
756 int budget)
bc304dd3 757{
09af87d1 758 struct net_device_context *ndev_ctx = netdev_priv(ndev);
4d18fcc9 759 struct sk_buff *skb;
bc304dd3
SH
760 u16 q_idx = 0;
761 int queue_sends;
4d18fcc9
AB
762 u64 cmd_rqst;
763
26894cd9 764 cmd_rqst = channel->request_addr_callback(channel, desc->trans_id);
4d18fcc9 765 if (cmd_rqst == VMBUS_RQST_ERROR) {
26894cd9 766 netdev_err(ndev, "Invalid transaction ID %llx\n", desc->trans_id);
4d18fcc9
AB
767 return;
768 }
769
770 skb = (struct sk_buff *)(unsigned long)cmd_rqst;
bc304dd3
SH
771
772 /* Notify the layer above us */
773 if (likely(skb)) {
846da38d 774 struct hv_netvsc_packet *packet
bc304dd3 775 = (struct hv_netvsc_packet *)skb->cb;
793e3955 776 u32 send_index = packet->send_buf_index;
1cb9d3b6 777 struct netvsc_stats_tx *tx_stats;
bc304dd3
SH
778
779 if (send_index != NETVSC_INVALID_INDEX)
780 netvsc_free_send_slot(net_device, send_index);
793e3955 781 q_idx = packet->q_idx;
bc304dd3 782
6c80f3fc 783 tx_stats = &net_device->chan_table[q_idx].tx_stats;
793e3955 784
785 u64_stats_update_begin(&tx_stats->syncp);
786 tx_stats->packets += packet->total_packets;
787 tx_stats->bytes += packet->total_bytes;
788 u64_stats_update_end(&tx_stats->syncp);
789
846da38d 790 netvsc_dma_unmap(ndev_ctx->device_ctx, packet);
f9645430 791 napi_consume_skb(skb, budget);
bc304dd3
SH
792 }
793
b8b835a8 794 queue_sends =
795 atomic_dec_return(&net_device->chan_table[q_idx].queue_sends);
bc304dd3 796
7b2ee50c
SH
797 if (unlikely(net_device->destroy)) {
798 if (queue_sends == 0)
799 wake_up(&net_device->wait_drain);
800 } else {
801 struct netdev_queue *txq = netdev_get_tx_queue(ndev, q_idx);
802
1b704c4a 803 if (netif_tx_queue_stopped(txq) && !net_device->tx_disable &&
6b1f8376
LL
804 (hv_get_avail_to_write_percent(&channel->outbound) >
805 RING_AVAIL_PERCENT_HIWATER || queue_sends < 1)) {
7b2ee50c
SH
806 netif_tx_wake_queue(txq);
807 ndev_ctx->eth_stats.wake_queue++;
808 }
09af87d1 809 }
bc304dd3
SH
810}
811
c347b927
SH
812static void netvsc_send_completion(struct net_device *ndev,
813 struct netvsc_device *net_device,
25b85ee8 814 struct vmbus_channel *incoming_channel,
f9645430 815 const struct vmpacket_descriptor *desc,
816 int budget)
fceaf24a 817{
8b31f8c9 818 const struct nvsp_message *nvsp_packet;
44144185 819 u32 msglen = hv_pkt_datalen(desc);
8b31f8c9
LL
820 struct nvsp_message *pkt_rqst;
821 u64 cmd_rqst;
dca5161f 822 u32 status;
8b31f8c9
LL
823
824 /* First check if this is a VMBUS completion without data payload */
825 if (!msglen) {
bf5fd8ca 826 cmd_rqst = incoming_channel->request_addr_callback(incoming_channel,
26894cd9 827 desc->trans_id);
8b31f8c9 828 if (cmd_rqst == VMBUS_RQST_ERROR) {
26894cd9 829 netdev_err(ndev, "Invalid transaction ID %llx\n", desc->trans_id);
8b31f8c9
LL
830 return;
831 }
832
833 pkt_rqst = (struct nvsp_message *)(uintptr_t)cmd_rqst;
834 switch (pkt_rqst->hdr.msg_type) {
835 case NVSP_MSG4_TYPE_SWITCH_DATA_PATH:
836 complete(&net_device->channel_init_wait);
837 break;
838
839 default:
840 netdev_err(ndev, "Unexpected VMBUS completion!!\n");
841 }
842 return;
843 }
44144185
AB
844
845 /* Ensure packet is big enough to read header fields */
846 if (msglen < sizeof(struct nvsp_message_header)) {
847 netdev_err(ndev, "nvsp_message length too small: %u\n", msglen);
848 return;
849 }
fceaf24a 850
8b31f8c9 851 nvsp_packet = hv_pkt_data(desc);
bc304dd3
SH
852 switch (nvsp_packet->hdr.msg_type) {
853 case NVSP_MSG_TYPE_INIT_COMPLETE:
44144185
AB
854 if (msglen < sizeof(struct nvsp_message_header) +
855 sizeof(struct nvsp_message_init_complete)) {
856 netdev_err(ndev, "nvsp_msg length too small: %u\n",
857 msglen);
858 return;
859 }
9bae5b05 860 break;
44144185 861
bc304dd3 862 case NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE:
44144185
AB
863 if (msglen < sizeof(struct nvsp_message_header) +
864 sizeof(struct nvsp_1_message_send_receive_buffer_complete)) {
865 netdev_err(ndev, "nvsp_msg1 length too small: %u\n",
866 msglen);
867 return;
868 }
9bae5b05 869 break;
44144185 870
bc304dd3 871 case NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE:
44144185
AB
872 if (msglen < sizeof(struct nvsp_message_header) +
873 sizeof(struct nvsp_1_message_send_send_buffer_complete)) {
874 netdev_err(ndev, "nvsp_msg1 length too small: %u\n",
875 msglen);
876 return;
877 }
9bae5b05 878 break;
44144185 879
bc304dd3 880 case NVSP_MSG5_TYPE_SUBCHANNEL:
44144185
AB
881 if (msglen < sizeof(struct nvsp_message_header) +
882 sizeof(struct nvsp_5_subchannel_complete)) {
883 netdev_err(ndev, "nvsp_msg5 length too small: %u\n",
884 msglen);
885 return;
886 }
bc304dd3
SH
887 break;
888
889 case NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE:
dca5161f
MK
890 if (msglen < sizeof(struct nvsp_message_header) +
891 sizeof(struct nvsp_1_message_send_rndis_packet_complete)) {
892 if (net_ratelimit())
893 netdev_err(ndev, "nvsp_rndis_pkt_complete length too small: %u\n",
894 msglen);
895 return;
896 }
897
898 /* If status indicates an error, output a message so we know
899 * there's a problem. But process the completion anyway so the
900 * resources are released.
901 */
902 status = nvsp_packet->msg.v1_msg.send_rndis_pkt_complete.status;
903 if (status != NVSP_STAT_SUCCESS && net_ratelimit())
904 netdev_err(ndev, "nvsp_rndis_pkt_complete error status: %x\n",
905 status);
906
c347b927
SH
907 netvsc_send_tx_complete(ndev, net_device, incoming_channel,
908 desc, budget);
9bae5b05 909 return;
fceaf24a 910
bc304dd3
SH
911 default:
912 netdev_err(ndev,
913 "Unknown send completion type %d received!!\n",
914 nvsp_packet->hdr.msg_type);
9bae5b05 915 return;
fceaf24a 916 }
9bae5b05
SS
917
918 /* Copy the response back */
919 memcpy(&net_device->channel_init_pkt, nvsp_packet,
920 sizeof(struct nvsp_message));
921 complete(&net_device->channel_init_wait);
fceaf24a
HJ
922}
923
c25aaf81
KS
924static u32 netvsc_get_next_send_section(struct netvsc_device *net_device)
925{
b58a1858 926 unsigned long *map_addr = net_device->send_section_map;
927 unsigned int i;
928
fdfb70d2 929 for_each_clear_bit(i, map_addr, net_device->send_section_cnt) {
b58a1858 930 if (sync_test_and_set_bit(i, map_addr) == 0)
931 return i;
c25aaf81 932 }
b58a1858 933
934 return NETVSC_INVALID_INDEX;
c25aaf81
KS
935}
936
26a11262
SH
937static void netvsc_copy_to_send_buf(struct netvsc_device *net_device,
938 unsigned int section_index,
939 u32 pend_size,
940 struct hv_netvsc_packet *packet,
941 struct rndis_message *rndis_msg,
942 struct hv_page_buffer *pb,
cfd8afd9 943 bool xmit_more)
c25aaf81
KS
944{
945 char *start = net_device->send_buf;
7c3877f2
HZ
946 char *dest = start + (section_index * net_device->send_section_size)
947 + pend_size;
c25aaf81 948 int i;
7c3877f2 949 u32 padding = 0;
aa0a34be
HZ
950 u32 page_count = packet->cp_partial ? packet->rmsg_pgcnt :
951 packet->page_buf_cnt;
b85e06f7 952 u32 remain;
7c3877f2
HZ
953
954 /* Add padding */
b85e06f7 955 remain = packet->total_data_buflen & (net_device->pkt_align - 1);
cfd8afd9 956 if (xmit_more && remain) {
7c3877f2 957 padding = net_device->pkt_align - remain;
24476760 958 rndis_msg->msg_len += padding;
7c3877f2
HZ
959 packet->total_data_buflen += padding;
960 }
c25aaf81 961
aa0a34be 962 for (i = 0; i < page_count; i++) {
11d8620e 963 char *src = phys_to_virt(pb[i].pfn << HV_HYP_PAGE_SHIFT);
02b6de01 964 u32 offset = pb[i].offset;
965 u32 len = pb[i].len;
c25aaf81
KS
966
967 memcpy(dest, (src + offset), len);
c25aaf81
KS
968 dest += len;
969 }
7c3877f2 970
26a11262 971 if (padding)
7c3877f2 972 memset(dest, 0, padding);
c25aaf81
KS
973}
974
846da38d
TL
975void netvsc_dma_unmap(struct hv_device *hv_dev,
976 struct hv_netvsc_packet *packet)
977{
846da38d
TL
978 int i;
979
980 if (!hv_is_isolation_supported())
981 return;
982
983 if (!packet->dma_range)
984 return;
985
99f1c460 986 for (i = 0; i < packet->page_buf_cnt; i++)
846da38d
TL
987 dma_unmap_single(&hv_dev->device, packet->dma_range[i].dma,
988 packet->dma_range[i].mapping_size,
989 DMA_TO_DEVICE);
990
991 kfree(packet->dma_range);
992}
993
994/* netvsc_dma_map - Map swiotlb bounce buffer with data page of
995 * packet sent by vmbus_sendpacket_pagebuffer() in the Isolation
996 * VM.
997 *
998 * In isolation VM, netvsc send buffer has been marked visible to
999 * host and so the data copied to send buffer doesn't need to use
1000 * bounce buffer. The data pages handled by vmbus_sendpacket_pagebuffer()
1001 * may not be copied to send buffer and so these pages need to be
1002 * mapped with swiotlb bounce buffer. netvsc_dma_map() is to do
1003 * that. The pfns in the struct hv_page_buffer need to be converted
1004 * to bounce buffer's pfn. The loop here is necessary because the
1005 * entries in the page buffer array are not necessarily full
1006 * pages of data. Each entry in the array has a separate offset and
1007 * len that may be non-zero, even for entries in the middle of the
1008 * array. And the entries are not physically contiguous. So each
1009 * entry must be individually mapped rather than as a contiguous unit.
1010 * So not use dma_map_sg() here.
1011 */
1012static int netvsc_dma_map(struct hv_device *hv_dev,
1013 struct hv_netvsc_packet *packet,
1014 struct hv_page_buffer *pb)
1015{
99f1c460 1016 u32 page_count = packet->page_buf_cnt;
846da38d
TL
1017 dma_addr_t dma;
1018 int i;
1019
1020 if (!hv_is_isolation_supported())
1021 return 0;
1022
1023 packet->dma_range = kcalloc(page_count,
1024 sizeof(*packet->dma_range),
c6aa9d3b 1025 GFP_ATOMIC);
846da38d
TL
1026 if (!packet->dma_range)
1027 return -ENOMEM;
1028
1029 for (i = 0; i < page_count; i++) {
1030 char *src = phys_to_virt((pb[i].pfn << HV_HYP_PAGE_SHIFT)
1031 + pb[i].offset);
1032 u32 len = pb[i].len;
1033
1034 dma = dma_map_single(&hv_dev->device, src, len,
1035 DMA_TO_DEVICE);
1036 if (dma_mapping_error(&hv_dev->device, dma)) {
1037 kfree(packet->dma_range);
1038 return -ENOMEM;
1039 }
1040
1041 /* pb[].offset and pb[].len are not changed during dma mapping
1042 * and so not reassign.
1043 */
1044 packet->dma_range[i].dma = dma;
1045 packet->dma_range[i].mapping_size = len;
1046 pb[i].pfn = dma >> HV_HYP_PAGE_SHIFT;
1047 }
1048
1049 return 0;
1050}
1051
3a8963ac 1052static inline int netvsc_send_pkt(
0a1275ca 1053 struct hv_device *device,
7c3877f2 1054 struct hv_netvsc_packet *packet,
a9f2e2d6 1055 struct netvsc_device *net_device,
02b6de01 1056 struct hv_page_buffer *pb,
3a3d9a0a 1057 struct sk_buff *skb)
fceaf24a 1058{
7c3877f2 1059 struct nvsp_message nvmsg;
ec966381 1060 struct nvsp_1_message_send_rndis_packet *rpkt =
956a25c9
JP
1061 &nvmsg.msg.v1_msg.send_rndis_pkt;
1062 struct netvsc_channel * const nvchan =
1063 &net_device->chan_table[packet->q_idx];
b8b835a8 1064 struct vmbus_channel *out_channel = nvchan->channel;
0a1275ca 1065 struct net_device *ndev = hv_get_drvdata(device);
09af87d1 1066 struct net_device_context *ndev_ctx = netdev_priv(ndev);
b8b835a8 1067 struct netdev_queue *txq = netdev_get_tx_queue(ndev, packet->q_idx);
7c3877f2
HZ
1068 u64 req_id;
1069 int ret;
6b1f8376 1070 u32 ring_avail = hv_get_avail_to_write_percent(&out_channel->outbound);
c25aaf81 1071
505e3f00 1072 memset(&nvmsg, 0, sizeof(struct nvsp_message));
7c3877f2 1073 nvmsg.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
956a25c9
JP
1074 if (skb)
1075 rpkt->channel_type = 0; /* 0 is RMC_DATA */
1076 else
1077 rpkt->channel_type = 1; /* 1 is RMC_CONTROL */
fceaf24a 1078
956a25c9 1079 rpkt->send_buf_section_index = packet->send_buf_index;
7c3877f2 1080 if (packet->send_buf_index == NETVSC_INVALID_INDEX)
956a25c9 1081 rpkt->send_buf_section_size = 0;
7c3877f2 1082 else
956a25c9 1083 rpkt->send_buf_section_size = packet->total_data_buflen;
21a80820 1084
3a3d9a0a 1085 req_id = (ulong)skb;
f1ea3cd7 1086
c3582a2c
HZ
1087 if (out_channel->rescind)
1088 return -ENODEV;
1089
ec966381
SH
1090 trace_nvsp_send_pkt(ndev, out_channel, rpkt);
1091
846da38d 1092 packet->dma_range = NULL;
72a2f5bd 1093 if (packet->page_buf_cnt) {
02b6de01 1094 if (packet->cp_partial)
1095 pb += packet->rmsg_pgcnt;
1096
846da38d
TL
1097 ret = netvsc_dma_map(ndev_ctx->device_ctx, packet, pb);
1098 if (ret) {
1099 ret = -EAGAIN;
1100 goto exit;
1101 }
1102
5a668d8c 1103 ret = vmbus_sendpacket_pagebuffer(out_channel,
1104 pb, packet->page_buf_cnt,
1105 &nvmsg, sizeof(nvmsg),
1106 req_id);
846da38d
TL
1107
1108 if (ret)
1109 netvsc_dma_unmap(ndev_ctx->device_ctx, packet);
21a80820 1110 } else {
5dd0fb9b 1111 ret = vmbus_sendpacket(out_channel,
1112 &nvmsg, sizeof(nvmsg),
1113 req_id, VM_PKT_DATA_INBAND,
1114 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
fceaf24a
HJ
1115 }
1116
846da38d 1117exit:
1d06825b 1118 if (ret == 0) {
b8b835a8 1119 atomic_inc_return(&nvchan->queue_sends);
5b54dac8 1120
09af87d1 1121 if (ring_avail < RING_AVAIL_PERCENT_LOWATER) {
b8b835a8 1122 netif_tx_stop_queue(txq);
09af87d1
SX
1123 ndev_ctx->eth_stats.stop_queue++;
1124 }
1d06825b 1125 } else if (ret == -EAGAIN) {
b8b835a8 1126 netif_tx_stop_queue(txq);
09af87d1 1127 ndev_ctx->eth_stats.stop_queue++;
1d06825b 1128 } else {
4a2176c6 1129 netdev_err(ndev,
1130 "Unable to send packet pages %u len %u, ret %d\n",
1131 packet->page_buf_cnt, packet->total_data_buflen,
1132 ret);
1d06825b 1133 }
fceaf24a 1134
93aa4792
HZ
1135 if (netif_tx_queue_stopped(txq) &&
1136 atomic_read(&nvchan->queue_sends) < 1 &&
1137 !net_device->tx_disable) {
1138 netif_tx_wake_queue(txq);
1139 ndev_ctx->eth_stats.wake_queue++;
1140 if (ret == -EAGAIN)
1141 ret = -ENOSPC;
1142 }
1143
7c3877f2
HZ
1144 return ret;
1145}
1146
c85e4924
HZ
1147/* Move packet out of multi send data (msd), and clear msd */
1148static inline void move_pkt_msd(struct hv_netvsc_packet **msd_send,
1149 struct sk_buff **msd_skb,
1150 struct multi_send_data *msdp)
1151{
1152 *msd_skb = msdp->skb;
1153 *msd_send = msdp->pkt;
1154 msdp->skb = NULL;
1155 msdp->pkt = NULL;
1156 msdp->count = 0;
1157}
1158
2a926f79 1159/* RCU already held by caller */
bd49fea7
SR
1160/* Batching/bouncing logic is designed to attempt to optimize
1161 * performance.
1162 *
1163 * For small, non-LSO packets we copy the packet to a send buffer
1164 * which is pre-registered with the Hyper-V side. This enables the
1165 * hypervisor to avoid remapping the aperture to access the packet
1166 * descriptor and data.
1167 *
1168 * If we already started using a buffer and the netdev is transmitting
1169 * a burst of packets, keep on copying into the buffer until it is
1170 * full or we are done collecting a burst. If there is an existing
1171 * buffer with space for the RNDIS descriptor but not the packet, copy
1172 * the RNDIS descriptor to the buffer, keeping the packet in place.
1173 *
1174 * If we do batching and send more than one packet using a single
1175 * NetVSC message, free the SKBs of the packets copied, except for the
1176 * last packet. This is done to streamline the handling of the case
1177 * where the last packet only had the RNDIS descriptor copied to the
1178 * send buffer, with the data pointers included in the NetVSC message.
1179 */
cfd8afd9 1180int netvsc_send(struct net_device *ndev,
24476760 1181 struct hv_netvsc_packet *packet,
a9f2e2d6 1182 struct rndis_message *rndis_msg,
02b6de01 1183 struct hv_page_buffer *pb,
351e1581
HZ
1184 struct sk_buff *skb,
1185 bool xdp_tx)
7c3877f2 1186{
cfd8afd9 1187 struct net_device_context *ndev_ctx = netdev_priv(ndev);
3962981f 1188 struct netvsc_device *net_device
867047c4 1189 = rcu_dereference_bh(ndev_ctx->nvdev);
2a926f79 1190 struct hv_device *device = ndev_ctx->device_ctx;
6c4c137e 1191 int ret = 0;
b8b835a8 1192 struct netvsc_channel *nvchan;
7c3877f2
HZ
1193 u32 pktlen = packet->total_data_buflen, msd_len = 0;
1194 unsigned int section_index = NETVSC_INVALID_INDEX;
7c3877f2
HZ
1195 struct multi_send_data *msdp;
1196 struct hv_netvsc_packet *msd_send = NULL, *cur_send = NULL;
c85e4924 1197 struct sk_buff *msd_skb = NULL;
cfd8afd9 1198 bool try_batch, xmit_more;
7c3877f2 1199
592b4fe8 1200 /* If device is rescinded, return error and packet will get dropped. */
2a926f79 1201 if (unlikely(!net_device || net_device->destroy))
7c3877f2
HZ
1202 return -ENODEV;
1203
b8b835a8 1204 nvchan = &net_device->chan_table[packet->q_idx];
7c3877f2 1205 packet->send_buf_index = NETVSC_INVALID_INDEX;
aa0a34be 1206 packet->cp_partial = false;
7c3877f2 1207
351e1581
HZ
1208 /* Send a control message or XDP packet directly without accessing
1209 * msd (Multi-Send Data) field which may be changed during data packet
1210 * processing.
cf8190e4 1211 */
351e1581 1212 if (!skb || xdp_tx)
12f69661 1213 return netvsc_send_pkt(device, packet, net_device, pb, skb);
cf8190e4 1214
7c3877f2 1215 /* batch packets in send buffer if possible */
b8b835a8 1216 msdp = &nvchan->msd;
7c3877f2
HZ
1217 if (msdp->pkt)
1218 msd_len = msdp->pkt->total_data_buflen;
1219
ebc1dcf6 1220 try_batch = msd_len > 0 && msdp->count < net_device->max_pkt;
aa0a34be 1221 if (try_batch && msd_len + pktlen + net_device->pkt_align <
7c3877f2
HZ
1222 net_device->send_section_size) {
1223 section_index = msdp->pkt->send_buf_index;
1224
aa0a34be
HZ
1225 } else if (try_batch && msd_len + packet->rmsg_size <
1226 net_device->send_section_size) {
1227 section_index = msdp->pkt->send_buf_index;
1228 packet->cp_partial = true;
1229
ebc1dcf6 1230 } else if (pktlen + net_device->pkt_align <
7c3877f2
HZ
1231 net_device->send_section_size) {
1232 section_index = netvsc_get_next_send_section(net_device);
cad5c197 1233 if (unlikely(section_index == NETVSC_INVALID_INDEX)) {
1234 ++ndev_ctx->eth_stats.tx_send_full;
1235 } else {
c85e4924
HZ
1236 move_pkt_msd(&msd_send, &msd_skb, msdp);
1237 msd_len = 0;
7c3877f2
HZ
1238 }
1239 }
1240
cfd8afd9
SH
1241 /* Keep aggregating only if stack says more data is coming
1242 * and not doing mixed modes send and not flow blocked
1243 */
6b16f9ee 1244 xmit_more = netdev_xmit_more() &&
cfd8afd9
SH
1245 !packet->cp_partial &&
1246 !netif_xmit_stopped(netdev_get_tx_queue(ndev, packet->q_idx));
1247
7c3877f2
HZ
1248 if (section_index != NETVSC_INVALID_INDEX) {
1249 netvsc_copy_to_send_buf(net_device,
1250 section_index, msd_len,
cfd8afd9 1251 packet, rndis_msg, pb, xmit_more);
b08cc791 1252
7c3877f2 1253 packet->send_buf_index = section_index;
aa0a34be
HZ
1254
1255 if (packet->cp_partial) {
1256 packet->page_buf_cnt -= packet->rmsg_pgcnt;
1257 packet->total_data_buflen = msd_len + packet->rmsg_size;
1258 } else {
1259 packet->page_buf_cnt = 0;
1260 packet->total_data_buflen += msd_len;
aa0a34be 1261 }
7c3877f2 1262
793e3955 1263 if (msdp->pkt) {
1264 packet->total_packets += msdp->pkt->total_packets;
1265 packet->total_bytes += msdp->pkt->total_bytes;
1266 }
1267
c85e4924 1268 if (msdp->skb)
17db4bce 1269 dev_consume_skb_any(msdp->skb);
ee90b812 1270
cfd8afd9 1271 if (xmit_more) {
c85e4924 1272 msdp->skb = skb;
7c3877f2
HZ
1273 msdp->pkt = packet;
1274 msdp->count++;
1275 } else {
1276 cur_send = packet;
c85e4924 1277 msdp->skb = NULL;
7c3877f2
HZ
1278 msdp->pkt = NULL;
1279 msdp->count = 0;
1280 }
1281 } else {
c85e4924 1282 move_pkt_msd(&msd_send, &msd_skb, msdp);
7c3877f2
HZ
1283 cur_send = packet;
1284 }
1285
7c3877f2 1286 if (msd_send) {
6c4c137e
SH
1287 int m_ret = netvsc_send_pkt(device, msd_send, net_device,
1288 NULL, msd_skb);
7c3877f2
HZ
1289
1290 if (m_ret != 0) {
1291 netvsc_free_send_slot(net_device,
1292 msd_send->send_buf_index);
c85e4924 1293 dev_kfree_skb_any(msd_skb);
7c3877f2
HZ
1294 }
1295 }
1296
1297 if (cur_send)
0a1275ca 1298 ret = netvsc_send_pkt(device, cur_send, net_device, pb, skb);
7c3877f2 1299
7aab5159
JS
1300 if (ret != 0 && section_index != NETVSC_INVALID_INDEX)
1301 netvsc_free_send_slot(net_device, section_index);
d953ca4d 1302
fceaf24a
HJ
1303 return ret;
1304}
1305
7426b1a5 1306/* Send pending recv completions */
cad5c197 1307static int send_recv_completions(struct net_device *ndev,
1308 struct netvsc_device *nvdev,
1309 struct netvsc_channel *nvchan)
5fa9d3c5 1310{
7426b1a5 1311 struct multi_recv_comp *mrc = &nvchan->mrc;
1312 struct recv_comp_msg {
1313 struct nvsp_message_header hdr;
1314 u32 status;
1315 } __packed;
1316 struct recv_comp_msg msg = {
1317 .hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE,
1318 };
5fa9d3c5
HZ
1319 int ret;
1320
7426b1a5 1321 while (mrc->first != mrc->next) {
1322 const struct recv_comp_data *rcd
1323 = mrc->slots + mrc->first;
c0b558e5 1324
7426b1a5 1325 msg.status = rcd->status;
1326 ret = vmbus_sendpacket(nvchan->channel, &msg, sizeof(msg),
1327 rcd->tid, VM_PKT_COMP, 0);
cad5c197 1328 if (unlikely(ret)) {
1329 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1330
1331 ++ndev_ctx->eth_stats.rx_comp_busy;
7426b1a5 1332 return ret;
cad5c197 1333 }
c0b558e5 1334
7426b1a5 1335 if (++mrc->first == nvdev->recv_completion_cnt)
1336 mrc->first = 0;
1337 }
c0b558e5 1338
7426b1a5 1339 /* receive completion ring has been emptied */
1340 if (unlikely(nvdev->destroy))
1341 wake_up(&nvdev->wait_drain);
c0b558e5 1342
7426b1a5 1343 return 0;
c0b558e5
HZ
1344}
1345
7426b1a5 1346/* Count how many receive completions are outstanding */
1347static void recv_comp_slot_avail(const struct netvsc_device *nvdev,
1348 const struct multi_recv_comp *mrc,
1349 u32 *filled, u32 *avail)
c0b558e5 1350{
7426b1a5 1351 u32 count = nvdev->recv_completion_cnt;
c0b558e5 1352
7426b1a5 1353 if (mrc->next >= mrc->first)
1354 *filled = mrc->next - mrc->first;
1355 else
1356 *filled = (count - mrc->first) + mrc->next;
c0b558e5 1357
7426b1a5 1358 *avail = count - *filled - 1;
c0b558e5
HZ
1359}
1360
7426b1a5 1361/* Add receive complete to ring to send to host. */
1362static void enq_receive_complete(struct net_device *ndev,
1363 struct netvsc_device *nvdev, u16 q_idx,
1364 u64 tid, u32 status)
c0b558e5 1365{
7426b1a5 1366 struct netvsc_channel *nvchan = &nvdev->chan_table[q_idx];
1367 struct multi_recv_comp *mrc = &nvchan->mrc;
c0b558e5 1368 struct recv_comp_data *rcd;
7426b1a5 1369 u32 filled, avail;
c0b558e5 1370
7426b1a5 1371 recv_comp_slot_avail(nvdev, mrc, &filled, &avail);
c0b558e5 1372
7426b1a5 1373 if (unlikely(filled > NAPI_POLL_WEIGHT)) {
cad5c197 1374 send_recv_completions(ndev, nvdev, nvchan);
7426b1a5 1375 recv_comp_slot_avail(nvdev, mrc, &filled, &avail);
5fa9d3c5 1376 }
c0b558e5 1377
7426b1a5 1378 if (unlikely(!avail)) {
1379 netdev_err(ndev, "Recv_comp full buf q:%hd, tid:%llx\n",
1380 q_idx, tid);
1381 return;
1382 }
c0b558e5 1383
7426b1a5 1384 rcd = mrc->slots + mrc->next;
1385 rcd->tid = tid;
1386 rcd->status = status;
c0b558e5 1387
7426b1a5 1388 if (++mrc->next == nvdev->recv_completion_cnt)
1389 mrc->next = 0;
c0b558e5
HZ
1390}
1391
15a863bf 1392static int netvsc_receive(struct net_device *ndev,
7426b1a5 1393 struct netvsc_device *net_device,
c8e4eff4 1394 struct netvsc_channel *nvchan,
44144185 1395 const struct vmpacket_descriptor *desc)
fceaf24a 1396{
c347b927 1397 struct net_device_context *net_device_ctx = netdev_priv(ndev);
c8e4eff4 1398 struct vmbus_channel *channel = nvchan->channel;
f3dd3f47 1399 const struct vmtransfer_page_packet_header *vmxferpage_packet
1400 = container_of(desc, const struct vmtransfer_page_packet_header, d);
44144185
AB
1401 const struct nvsp_message *nvsp = hv_pkt_data(desc);
1402 u32 msglen = hv_pkt_datalen(desc);
15a863bf 1403 u16 q_idx = channel->offermsg.offer.sub_channel_index;
dc54a08c 1404 char *recv_buf = net_device->recv_buf;
4baab261 1405 u32 status = NVSP_STAT_SUCCESS;
45326342
HZ
1406 int i;
1407 int count = 0;
779b4d17 1408
44144185
AB
1409 /* Ensure packet is big enough to read header fields */
1410 if (msglen < sizeof(struct nvsp_message_header)) {
1411 netif_err(net_device_ctx, rx_err, ndev,
1412 "invalid nvsp header, length too small: %u\n",
1413 msglen);
1414 return 0;
1415 }
1416
454f18a9 1417 /* Make sure this is a valid nvsp packet */
dc54a08c 1418 if (unlikely(nvsp->hdr.msg_type != NVSP_MSG1_TYPE_SEND_RNDIS_PKT)) {
1419 netif_err(net_device_ctx, rx_err, ndev,
1420 "Unknown nvsp packet type received %u\n",
1421 nvsp->hdr.msg_type);
15a863bf 1422 return 0;
fceaf24a
HJ
1423 }
1424
44144185
AB
1425 /* Validate xfer page pkt header */
1426 if ((desc->offset8 << 3) < sizeof(struct vmtransfer_page_packet_header)) {
1427 netif_err(net_device_ctx, rx_err, ndev,
1428 "Invalid xfer page pkt, offset too small: %u\n",
1429 desc->offset8 << 3);
1430 return 0;
1431 }
1432
dc54a08c 1433 if (unlikely(vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID)) {
1434 netif_err(net_device_ctx, rx_err, ndev,
1435 "Invalid xfer page set id - expecting %x got %x\n",
1436 NETVSC_RECEIVE_BUFFER_ID,
1437 vmxferpage_packet->xfer_pageset_id);
15a863bf 1438 return 0;
fceaf24a
HJ
1439 }
1440
4baab261 1441 count = vmxferpage_packet->range_cnt;
fceaf24a 1442
44144185
AB
1443 /* Check count for a valid value */
1444 if (NETVSC_XFER_HEADER_SIZE(count) > desc->offset8 << 3) {
1445 netif_err(net_device_ctx, rx_err, ndev,
1446 "Range count is not valid: %d\n",
1447 count);
1448 return 0;
1449 }
1450
454f18a9 1451 /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
4baab261 1452 for (i = 0; i < count; i++) {
c5d24bdd 1453 u32 offset = vmxferpage_packet->ranges[i].byte_offset;
dc54a08c 1454 u32 buflen = vmxferpage_packet->ranges[i].byte_count;
c5d24bdd 1455 void *data;
5c71dadb 1456 int ret;
fceaf24a 1457
44144185
AB
1458 if (unlikely(offset > net_device->recv_buf_size ||
1459 buflen > net_device->recv_buf_size - offset)) {
c8e4eff4 1460 nvchan->rsc.cnt = 0;
c5d24bdd
HZ
1461 status = NVSP_STAT_FAIL;
1462 netif_err(net_device_ctx, rx_err, ndev,
1463 "Packet offset:%u + len:%u too big\n",
1464 offset, buflen);
1465
1466 continue;
1467 }
1468
0ba35fe9
APM
1469 /* We're going to copy (sections of) the packet into nvchan->recv_buf;
1470 * make sure that nvchan->recv_buf is large enough to hold the packet.
1471 */
1472 if (unlikely(buflen > net_device->recv_section_size)) {
1473 nvchan->rsc.cnt = 0;
1474 status = NVSP_STAT_FAIL;
1475 netif_err(net_device_ctx, rx_err, ndev,
1476 "Packet too big: buflen=%u recv_section_size=%u\n",
1477 buflen, net_device->recv_section_size);
1478
1479 continue;
1480 }
1481
c5d24bdd
HZ
1482 data = recv_buf + offset;
1483
c8e4eff4
HZ
1484 nvchan->rsc.is_last = (i == count - 1);
1485
ec966381
SH
1486 trace_rndis_recv(ndev, q_idx, data);
1487
454f18a9 1488 /* Pass it to the upper layer */
5c71dadb 1489 ret = rndis_filter_receive(ndev, net_device,
c8e4eff4 1490 nvchan, data, buflen);
5c71dadb 1491
12bc8dfb
APM
1492 if (unlikely(ret != NVSP_STAT_SUCCESS)) {
1493 /* Drop incomplete packet */
1494 nvchan->rsc.cnt = 0;
5c71dadb 1495 status = NVSP_STAT_FAIL;
12bc8dfb 1496 }
fceaf24a
HJ
1497 }
1498
7426b1a5 1499 enq_receive_complete(ndev, net_device, q_idx,
1500 vmxferpage_packet->d.trans_id, status);
15a863bf 1501
15a863bf 1502 return count;
fceaf24a
HJ
1503}
1504
c347b927 1505static void netvsc_send_table(struct net_device *ndev,
171c1fd9 1506 struct netvsc_device *nvscdev,
71f21959
HZ
1507 const struct nvsp_message *nvmsg,
1508 u32 msglen)
5b54dac8 1509{
7ce10124 1510 struct net_device_context *net_device_ctx = netdev_priv(ndev);
71f21959 1511 u32 count, offset, *tab;
c347b927 1512 int i;
5b54dac8 1513
44144185
AB
1514 /* Ensure packet is big enough to read send_table fields */
1515 if (msglen < sizeof(struct nvsp_message_header) +
1516 sizeof(struct nvsp_5_send_indirect_table)) {
1517 netdev_err(ndev, "nvsp_v5_msg length too small: %u\n", msglen);
1518 return;
1519 }
1520
5b54dac8 1521 count = nvmsg->msg.v5_msg.send_table.count;
71f21959
HZ
1522 offset = nvmsg->msg.v5_msg.send_table.offset;
1523
5b54dac8
HZ
1524 if (count != VRSS_SEND_TAB_SIZE) {
1525 netdev_err(ndev, "Received wrong send-table size:%u\n", count);
1526 return;
1527 }
1528
171c1fd9
HZ
1529 /* If negotiated version <= NVSP_PROTOCOL_VERSION_6, the offset may be
1530 * wrong due to a host bug. So fix the offset here.
1531 */
1532 if (nvscdev->nvsp_version <= NVSP_PROTOCOL_VERSION_6 &&
1533 msglen >= sizeof(struct nvsp_message_header) +
1534 sizeof(union nvsp_6_message_uber) + count * sizeof(u32))
1535 offset = sizeof(struct nvsp_message_header) +
1536 sizeof(union nvsp_6_message_uber);
1537
1538 /* Boundary check for all versions */
505e3f00 1539 if (msglen < count * sizeof(u32) || offset > msglen - count * sizeof(u32)) {
71f21959
HZ
1540 netdev_err(ndev, "Received send-table offset too big:%u\n",
1541 offset);
1542 return;
1543 }
1544
1545 tab = (void *)nvmsg + offset;
5b54dac8
HZ
1546
1547 for (i = 0; i < count; i++)
39e91cfb 1548 net_device_ctx->tx_table[i] = tab[i];
5b54dac8
HZ
1549}
1550
c347b927 1551static void netvsc_send_vf(struct net_device *ndev,
44144185
AB
1552 const struct nvsp_message *nvmsg,
1553 u32 msglen)
71790a27 1554{
c347b927
SH
1555 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1556
44144185
AB
1557 /* Ensure packet is big enough to read its fields */
1558 if (msglen < sizeof(struct nvsp_message_header) +
1559 sizeof(struct nvsp_4_send_vf_association)) {
1560 netdev_err(ndev, "nvsp_v4_msg length too small: %u\n", msglen);
1561 return;
1562 }
1563
f9a7da91
VK
1564 net_device_ctx->vf_alloc = nvmsg->msg.v4_msg.vf_assoc.allocated;
1565 net_device_ctx->vf_serial = nvmsg->msg.v4_msg.vf_assoc.serial;
365e1ece
GK
1566
1567 if (net_device_ctx->vf_alloc)
1568 complete(&net_device_ctx->vf_add);
1569
00d7ddba
SH
1570 netdev_info(ndev, "VF slot %u %s\n",
1571 net_device_ctx->vf_serial,
1572 net_device_ctx->vf_alloc ? "added" : "removed");
71790a27
HZ
1573}
1574
71f21959 1575static void netvsc_receive_inband(struct net_device *ndev,
171c1fd9 1576 struct netvsc_device *nvscdev,
44144185 1577 const struct vmpacket_descriptor *desc)
71790a27 1578{
44144185
AB
1579 const struct nvsp_message *nvmsg = hv_pkt_data(desc);
1580 u32 msglen = hv_pkt_datalen(desc);
1581
1582 /* Ensure packet is big enough to read header fields */
1583 if (msglen < sizeof(struct nvsp_message_header)) {
1584 netdev_err(ndev, "inband nvsp_message length too small: %u\n", msglen);
1585 return;
1586 }
1587
71790a27
HZ
1588 switch (nvmsg->hdr.msg_type) {
1589 case NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE:
171c1fd9 1590 netvsc_send_table(ndev, nvscdev, nvmsg, msglen);
71790a27
HZ
1591 break;
1592
1593 case NVSP_MSG4_TYPE_SEND_VF_ASSOCIATION:
96854bbd
APM
1594 if (hv_is_isolation_supported())
1595 netdev_err(ndev, "Ignore VF_ASSOCIATION msg from the host supporting isolation\n");
1596 else
1597 netvsc_send_vf(ndev, nvmsg, msglen);
71790a27
HZ
1598 break;
1599 }
1600}
1601
15a863bf 1602static int netvsc_process_raw_pkt(struct hv_device *device,
c8e4eff4 1603 struct netvsc_channel *nvchan,
15a863bf 1604 struct netvsc_device *net_device,
1605 struct net_device *ndev,
f9645430 1606 const struct vmpacket_descriptor *desc,
1607 int budget)
99a50bb1 1608{
c8e4eff4 1609 struct vmbus_channel *channel = nvchan->channel;
c347b927 1610 const struct nvsp_message *nvmsg = hv_pkt_data(desc);
99a50bb1 1611
ec966381
SH
1612 trace_nvsp_recv(ndev, channel, nvmsg);
1613
99a50bb1
S
1614 switch (desc->type) {
1615 case VM_PKT_COMP:
44144185 1616 netvsc_send_completion(ndev, net_device, channel, desc, budget);
99a50bb1
S
1617 break;
1618
1619 case VM_PKT_DATA_USING_XFER_PAGES:
44144185 1620 return netvsc_receive(ndev, net_device, nvchan, desc);
99a50bb1
S
1621
1622 case VM_PKT_DATA_INBAND:
44144185 1623 netvsc_receive_inband(ndev, net_device, desc);
99a50bb1
S
1624 break;
1625
1626 default:
1627 netdev_err(ndev, "unhandled packet type %d, tid %llx\n",
f4f1c23d 1628 desc->type, desc->trans_id);
99a50bb1
S
1629 break;
1630 }
15a863bf 1631
1632 return 0;
1633}
1634
1635static struct hv_device *netvsc_channel_to_device(struct vmbus_channel *channel)
1636{
1637 struct vmbus_channel *primary = channel->primary_channel;
1638
1639 return primary ? primary->device_obj : channel->device_obj;
1640}
1641
262b7f14 1642/* Network processing softirq
1643 * Process data in incoming ring buffer from host
1644 * Stops when ring is empty or budget is met or exceeded.
1645 */
15a863bf 1646int netvsc_poll(struct napi_struct *napi, int budget)
1647{
1648 struct netvsc_channel *nvchan
1649 = container_of(napi, struct netvsc_channel, napi);
35fbbccf 1650 struct netvsc_device *net_device = nvchan->net_device;
15a863bf 1651 struct vmbus_channel *channel = nvchan->channel;
1652 struct hv_device *device = netvsc_channel_to_device(channel);
15a863bf 1653 struct net_device *ndev = hv_get_drvdata(device);
15a863bf 1654 int work_done = 0;
6b81b193 1655 int ret;
15a863bf 1656
f4f1c23d 1657 /* If starting a new interval */
1658 if (!nvchan->desc)
1659 nvchan->desc = hv_pkt_iter_first(channel);
15a863bf 1660
1cb9d3b6
HZ
1661 nvchan->xdp_flush = false;
1662
f4f1c23d 1663 while (nvchan->desc && work_done < budget) {
c8e4eff4 1664 work_done += netvsc_process_raw_pkt(device, nvchan, net_device,
f9645430 1665 ndev, nvchan->desc, budget);
f4f1c23d 1666 nvchan->desc = hv_pkt_iter_next(channel, nvchan->desc);
15a863bf 1667 }
15a863bf 1668
1cb9d3b6
HZ
1669 if (nvchan->xdp_flush)
1670 xdp_do_flush();
1671
6b81b193
HZ
1672 /* Send any pending receive completions */
1673 ret = send_recv_completions(ndev, net_device, nvchan);
1674
1675 /* If it did not exhaust NAPI budget this time
1676 * and not doing busy poll
f4e40363 1677 * then re-enable host interrupts
6b81b193
HZ
1678 * and reschedule if ring is not empty
1679 * or sending receive completion failed.
262b7f14 1680 */
6b81b193 1681 if (work_done < budget &&
15a863bf 1682 napi_complete_done(napi, work_done) &&
6b81b193 1683 (ret || hv_end_read(&channel->inbound)) &&
d64e38ae 1684 napi_schedule_prep(napi)) {
7426b1a5 1685 hv_begin_read(&channel->inbound);
d64e38ae 1686 __napi_schedule(napi);
7426b1a5 1687 }
f4f1c23d 1688
1689 /* Driver may overshoot since multiple packets per descriptor */
1690 return min(work_done, budget);
99a50bb1
S
1691}
1692
262b7f14 1693/* Call back when data is available in host ring buffer.
1694 * Processing is deferred until network softirq (NAPI)
1695 */
5b54dac8 1696void netvsc_channel_cb(void *context)
fceaf24a 1697{
6de38af6 1698 struct netvsc_channel *nvchan = context;
43bf99ce 1699 struct vmbus_channel *channel = nvchan->channel;
1700 struct hv_ring_buffer_info *rbi = &channel->inbound;
1701
1702 /* preload first vmpacket descriptor */
1703 prefetch(hv_get_ring_buffer(rbi) + rbi->priv_read_index);
0b307ebd 1704
f4f1c23d 1705 if (napi_schedule_prep(&nvchan->napi)) {
52d3b494 1706 /* disable interrupts from host */
43bf99ce 1707 hv_begin_read(rbi);
0d6dd357 1708
68633eda 1709 __napi_schedule_irqoff(&nvchan->napi);
f4f1c23d 1710 }
fceaf24a 1711}
af24ce42 1712
b637e023
HZ
1713/*
1714 * netvsc_device_add - Callback when the device belonging to this
1715 * driver is added
1716 */
9749fed5 1717struct netvsc_device *netvsc_device_add(struct hv_device *device,
1718 const struct netvsc_device_info *device_info)
b637e023 1719{
88098834 1720 int i, ret = 0;
b637e023 1721 struct netvsc_device *net_device;
88098834
VK
1722 struct net_device *ndev = hv_get_drvdata(device);
1723 struct net_device_context *net_device_ctx = netdev_priv(ndev);
b637e023 1724
88098834 1725 net_device = alloc_net_device();
b1c84927 1726 if (!net_device)
9749fed5 1727 return ERR_PTR(-ENOMEM);
b637e023 1728
6b0cbe31
HZ
1729 for (i = 0; i < VRSS_SEND_TAB_SIZE; i++)
1730 net_device_ctx->tx_table[i] = 0;
1731
15a863bf 1732 /* Because the device uses NAPI, all the interrupt batching and
1733 * control is done via Net softirq, not the channel handling
1734 */
1735 set_channel_read_mode(device->channel, HV_CALL_ISR);
1736
bffb1842
S
1737 /* If we're reopening the device we may have multiple queues, fill the
1738 * chn_table with the default channel to use it before subchannels are
1739 * opened.
1740 * Initialize the channel state before we open;
1741 * we can be interrupted as soon as we open the channel.
1742 */
1743
1744 for (i = 0; i < VRSS_CHANNEL_MAX; i++) {
1745 struct netvsc_channel *nvchan = &net_device->chan_table[i];
1746
1747 nvchan->channel = device->channel;
35fbbccf 1748 nvchan->net_device = net_device;
4a0dee1f
FF
1749 u64_stats_init(&nvchan->tx_stats.syncp);
1750 u64_stats_init(&nvchan->rx_stats.syncp);
351e1581 1751
b02e5a0e 1752 ret = xdp_rxq_info_reg(&nvchan->xdp_rxq, ndev, i, 0);
351e1581
HZ
1753
1754 if (ret) {
1755 netdev_err(ndev, "xdp_rxq_info_reg fail: %d\n", ret);
1756 goto cleanup2;
1757 }
1758
1759 ret = xdp_rxq_info_reg_mem_model(&nvchan->xdp_rxq,
1760 MEM_TYPE_PAGE_SHARED, NULL);
1761
1762 if (ret) {
1763 netdev_err(ndev, "xdp reg_mem_model fail: %d\n", ret);
1764 goto cleanup2;
1765 }
bffb1842
S
1766 }
1767
2be0f264 1768 /* Enable NAPI handler before init callbacks */
b48b89f9 1769 netif_napi_add(ndev, &net_device->chan_table[0].napi, netvsc_poll);
2be0f264 1770
b637e023 1771 /* Open the channel */
bf5fd8ca
APM
1772 device->channel->next_request_id_callback = vmbus_next_request_id;
1773 device->channel->request_addr_callback = vmbus_request_addr;
4d18fcc9 1774 device->channel->rqstor_size = netvsc_rqstor_size(netvsc_ring_bytes);
adae1e93
AB
1775 device->channel->max_pkt_size = NETVSC_MAX_PKT_SIZE;
1776
a7f99d0f
SH
1777 ret = vmbus_open(device->channel, netvsc_ring_bytes,
1778 netvsc_ring_bytes, NULL, 0,
1779 netvsc_channel_cb, net_device->chan_table);
b637e023
HZ
1780
1781 if (ret != 0) {
d9871158 1782 netdev_err(ndev, "unable to open channel: %d\n", ret);
b637e023
HZ
1783 goto cleanup;
1784 }
1785
1786 /* Channel is opened */
93ba2222 1787 netdev_dbg(ndev, "hv_netvsc channel opened successfully\n");
b637e023 1788
15a863bf 1789 napi_enable(&net_device->chan_table[0].napi);
88098834 1790
b637e023 1791 /* Connect with the NetVsp */
8b532797 1792 ret = netvsc_connect_vsp(device, net_device, device_info);
b637e023 1793 if (ret != 0) {
d9871158 1794 netdev_err(ndev,
c909ebbd 1795 "unable to connect to NetVSP - %d\n", ret);
b637e023
HZ
1796 goto close;
1797 }
1798
12f69661
SH
1799 /* Writing nvdev pointer unlocks netvsc_send(), make sure chn_table is
1800 * populated.
1801 */
1802 rcu_assign_pointer(net_device_ctx->nvdev, net_device);
1803
9749fed5 1804 return net_device;
b637e023
HZ
1805
1806close:
49393347 1807 RCU_INIT_POINTER(net_device_ctx->nvdev, NULL);
1808 napi_disable(&net_device->chan_table[0].napi);
15a863bf 1809
b637e023
HZ
1810 /* Now, we can close the channel safely */
1811 vmbus_close(device->channel);
1812
1813cleanup:
fcfb4a00 1814 netif_napi_del(&net_device->chan_table[0].napi);
351e1581
HZ
1815
1816cleanup2:
545a8e79 1817 free_netvsc_device(&net_device->rcu);
b637e023 1818
9749fed5 1819 return ERR_PTR(ret);
b637e023 1820}