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 */ 40 void 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 62 63 static struct netvsc_device *alloc_net_device(void) 64 { 65 struct netvsc_device *net_device; 66 67 net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL); 68 if (!net_device) 69 return NULL; 70 71 net_device->cb_buffer = kzalloc(NETVSC_PACKET_SIZE, GFP_KERNEL); 72 if (!net_device->cb_buffer) { 73 kfree(net_device); 74 return NULL; 75 } 76 77 init_waitqueue_head(&net_device->wait_drain); 78 net_device->destroy = false; 79 atomic_set(&net_device->open_cnt, 0); 80 atomic_set(&net_device->vf_use_cnt, 0); 81 net_device->max_pkt = RNDIS_MAX_PKT_DEFAULT; 82 net_device->pkt_align = RNDIS_PKT_ALIGN_DEFAULT; 83 84 net_device->vf_netdev = NULL; 85 net_device->vf_inject = false; 86 87 return net_device; 88 } 89 90 static void free_netvsc_device(struct netvsc_device *nvdev) 91 { 92 kfree(nvdev->cb_buffer); 93 kfree(nvdev); 94 } 95 96 static struct netvsc_device *get_outbound_net_device(struct hv_device *device) 97 { 98 struct netvsc_device *net_device = hv_device_to_netvsc_device(device); 99 100 if (net_device && net_device->destroy) 101 net_device = NULL; 102 103 return net_device; 104 } 105 106 static struct netvsc_device *get_inbound_net_device(struct hv_device *device) 107 { 108 struct netvsc_device *net_device = hv_device_to_netvsc_device(device); 109 110 if (!net_device) 111 goto get_in_err; 112 113 if (net_device->destroy && 114 atomic_read(&net_device->num_outstanding_sends) == 0) 115 net_device = NULL; 116 117 get_in_err: 118 return net_device; 119 } 120 121 122 static int netvsc_destroy_buf(struct hv_device *device) 123 { 124 struct nvsp_message *revoke_packet; 125 int ret = 0; 126 struct net_device *ndev = hv_get_drvdata(device); 127 struct netvsc_device *net_device = net_device_to_netvsc_device(ndev); 128 129 /* 130 * If we got a section count, it means we received a 131 * SendReceiveBufferComplete msg (ie sent 132 * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need 133 * to send a revoke msg here 134 */ 135 if (net_device->recv_section_cnt) { 136 /* Send the revoke receive buffer */ 137 revoke_packet = &net_device->revoke_packet; 138 memset(revoke_packet, 0, sizeof(struct nvsp_message)); 139 140 revoke_packet->hdr.msg_type = 141 NVSP_MSG1_TYPE_REVOKE_RECV_BUF; 142 revoke_packet->msg.v1_msg. 143 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID; 144 145 ret = vmbus_sendpacket(device->channel, 146 revoke_packet, 147 sizeof(struct nvsp_message), 148 (unsigned long)revoke_packet, 149 VM_PKT_DATA_INBAND, 0); 150 /* 151 * If we failed here, we might as well return and 152 * have a leak rather than continue and a bugchk 153 */ 154 if (ret != 0) { 155 netdev_err(ndev, "unable to send " 156 "revoke receive buffer to netvsp\n"); 157 return ret; 158 } 159 } 160 161 /* Teardown the gpadl on the vsp end */ 162 if (net_device->recv_buf_gpadl_handle) { 163 ret = vmbus_teardown_gpadl(device->channel, 164 net_device->recv_buf_gpadl_handle); 165 166 /* If we failed here, we might as well return and have a leak 167 * rather than continue and a bugchk 168 */ 169 if (ret != 0) { 170 netdev_err(ndev, 171 "unable to teardown receive buffer's gpadl\n"); 172 return ret; 173 } 174 net_device->recv_buf_gpadl_handle = 0; 175 } 176 177 if (net_device->recv_buf) { 178 /* Free up the receive buffer */ 179 vfree(net_device->recv_buf); 180 net_device->recv_buf = NULL; 181 } 182 183 if (net_device->recv_section) { 184 net_device->recv_section_cnt = 0; 185 kfree(net_device->recv_section); 186 net_device->recv_section = NULL; 187 } 188 189 /* Deal with the send buffer we may have setup. 190 * If we got a send section size, it means we received a 191 * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent 192 * NVSP_MSG1_TYPE_SEND_SEND_BUF msg) therefore, we need 193 * to send a revoke msg here 194 */ 195 if (net_device->send_section_size) { 196 /* Send the revoke receive buffer */ 197 revoke_packet = &net_device->revoke_packet; 198 memset(revoke_packet, 0, sizeof(struct nvsp_message)); 199 200 revoke_packet->hdr.msg_type = 201 NVSP_MSG1_TYPE_REVOKE_SEND_BUF; 202 revoke_packet->msg.v1_msg.revoke_send_buf.id = 203 NETVSC_SEND_BUFFER_ID; 204 205 ret = vmbus_sendpacket(device->channel, 206 revoke_packet, 207 sizeof(struct nvsp_message), 208 (unsigned long)revoke_packet, 209 VM_PKT_DATA_INBAND, 0); 210 /* If we failed here, we might as well return and 211 * have a leak rather than continue and a bugchk 212 */ 213 if (ret != 0) { 214 netdev_err(ndev, "unable to send " 215 "revoke send buffer to netvsp\n"); 216 return ret; 217 } 218 } 219 /* Teardown the gpadl on the vsp end */ 220 if (net_device->send_buf_gpadl_handle) { 221 ret = vmbus_teardown_gpadl(device->channel, 222 net_device->send_buf_gpadl_handle); 223 224 /* If we failed here, we might as well return and have a leak 225 * rather than continue and a bugchk 226 */ 227 if (ret != 0) { 228 netdev_err(ndev, 229 "unable to teardown send buffer's gpadl\n"); 230 return ret; 231 } 232 net_device->send_buf_gpadl_handle = 0; 233 } 234 if (net_device->send_buf) { 235 /* Free up the send buffer */ 236 vfree(net_device->send_buf); 237 net_device->send_buf = NULL; 238 } 239 kfree(net_device->send_section_map); 240 241 return ret; 242 } 243 244 static int netvsc_init_buf(struct hv_device *device) 245 { 246 int ret = 0; 247 struct netvsc_device *net_device; 248 struct nvsp_message *init_packet; 249 struct net_device *ndev; 250 int node; 251 252 net_device = get_outbound_net_device(device); 253 if (!net_device) 254 return -ENODEV; 255 ndev = hv_get_drvdata(device); 256 257 node = cpu_to_node(device->channel->target_cpu); 258 net_device->recv_buf = vzalloc_node(net_device->recv_buf_size, node); 259 if (!net_device->recv_buf) 260 net_device->recv_buf = vzalloc(net_device->recv_buf_size); 261 262 if (!net_device->recv_buf) { 263 netdev_err(ndev, "unable to allocate receive " 264 "buffer of size %d\n", net_device->recv_buf_size); 265 ret = -ENOMEM; 266 goto cleanup; 267 } 268 269 /* 270 * Establish the gpadl handle for this buffer on this 271 * channel. Note: This call uses the vmbus connection rather 272 * than the channel to establish the gpadl handle. 273 */ 274 ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf, 275 net_device->recv_buf_size, 276 &net_device->recv_buf_gpadl_handle); 277 if (ret != 0) { 278 netdev_err(ndev, 279 "unable to establish receive buffer's gpadl\n"); 280 goto cleanup; 281 } 282 283 284 /* Notify the NetVsp of the gpadl handle */ 285 init_packet = &net_device->channel_init_pkt; 286 287 memset(init_packet, 0, sizeof(struct nvsp_message)); 288 289 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF; 290 init_packet->msg.v1_msg.send_recv_buf. 291 gpadl_handle = net_device->recv_buf_gpadl_handle; 292 init_packet->msg.v1_msg. 293 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID; 294 295 /* Send the gpadl notification request */ 296 ret = vmbus_sendpacket(device->channel, init_packet, 297 sizeof(struct nvsp_message), 298 (unsigned long)init_packet, 299 VM_PKT_DATA_INBAND, 300 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); 301 if (ret != 0) { 302 netdev_err(ndev, 303 "unable to send receive buffer's gpadl to netvsp\n"); 304 goto cleanup; 305 } 306 307 wait_for_completion(&net_device->channel_init_wait); 308 309 /* Check the response */ 310 if (init_packet->msg.v1_msg. 311 send_recv_buf_complete.status != NVSP_STAT_SUCCESS) { 312 netdev_err(ndev, "Unable to complete receive buffer " 313 "initialization with NetVsp - status %d\n", 314 init_packet->msg.v1_msg. 315 send_recv_buf_complete.status); 316 ret = -EINVAL; 317 goto cleanup; 318 } 319 320 /* Parse the response */ 321 322 net_device->recv_section_cnt = init_packet->msg. 323 v1_msg.send_recv_buf_complete.num_sections; 324 325 net_device->recv_section = kmemdup( 326 init_packet->msg.v1_msg.send_recv_buf_complete.sections, 327 net_device->recv_section_cnt * 328 sizeof(struct nvsp_1_receive_buffer_section), 329 GFP_KERNEL); 330 if (net_device->recv_section == NULL) { 331 ret = -EINVAL; 332 goto cleanup; 333 } 334 335 /* 336 * For 1st release, there should only be 1 section that represents the 337 * entire receive buffer 338 */ 339 if (net_device->recv_section_cnt != 1 || 340 net_device->recv_section->offset != 0) { 341 ret = -EINVAL; 342 goto cleanup; 343 } 344 345 /* Now setup the send buffer. 346 */ 347 net_device->send_buf = vzalloc_node(net_device->send_buf_size, node); 348 if (!net_device->send_buf) 349 net_device->send_buf = vzalloc(net_device->send_buf_size); 350 if (!net_device->send_buf) { 351 netdev_err(ndev, "unable to allocate send " 352 "buffer of size %d\n", net_device->send_buf_size); 353 ret = -ENOMEM; 354 goto cleanup; 355 } 356 357 /* Establish the gpadl handle for this buffer on this 358 * channel. Note: This call uses the vmbus connection rather 359 * than the channel to establish the gpadl handle. 360 */ 361 ret = vmbus_establish_gpadl(device->channel, net_device->send_buf, 362 net_device->send_buf_size, 363 &net_device->send_buf_gpadl_handle); 364 if (ret != 0) { 365 netdev_err(ndev, 366 "unable to establish send buffer's gpadl\n"); 367 goto cleanup; 368 } 369 370 /* Notify the NetVsp of the gpadl handle */ 371 init_packet = &net_device->channel_init_pkt; 372 memset(init_packet, 0, sizeof(struct nvsp_message)); 373 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_SEND_BUF; 374 init_packet->msg.v1_msg.send_send_buf.gpadl_handle = 375 net_device->send_buf_gpadl_handle; 376 init_packet->msg.v1_msg.send_send_buf.id = NETVSC_SEND_BUFFER_ID; 377 378 /* Send the gpadl notification request */ 379 ret = vmbus_sendpacket(device->channel, init_packet, 380 sizeof(struct nvsp_message), 381 (unsigned long)init_packet, 382 VM_PKT_DATA_INBAND, 383 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); 384 if (ret != 0) { 385 netdev_err(ndev, 386 "unable to send send buffer's gpadl to netvsp\n"); 387 goto cleanup; 388 } 389 390 wait_for_completion(&net_device->channel_init_wait); 391 392 /* Check the response */ 393 if (init_packet->msg.v1_msg. 394 send_send_buf_complete.status != NVSP_STAT_SUCCESS) { 395 netdev_err(ndev, "Unable to complete send buffer " 396 "initialization with NetVsp - status %d\n", 397 init_packet->msg.v1_msg. 398 send_send_buf_complete.status); 399 ret = -EINVAL; 400 goto cleanup; 401 } 402 403 /* Parse the response */ 404 net_device->send_section_size = init_packet->msg. 405 v1_msg.send_send_buf_complete.section_size; 406 407 /* Section count is simply the size divided by the section size. 408 */ 409 net_device->send_section_cnt = 410 net_device->send_buf_size/net_device->send_section_size; 411 412 dev_info(&device->device, "Send section size: %d, Section count:%d\n", 413 net_device->send_section_size, net_device->send_section_cnt); 414 415 /* Setup state for managing the send buffer. */ 416 net_device->map_words = DIV_ROUND_UP(net_device->send_section_cnt, 417 BITS_PER_LONG); 418 419 net_device->send_section_map = 420 kzalloc(net_device->map_words * sizeof(ulong), GFP_KERNEL); 421 if (net_device->send_section_map == NULL) { 422 ret = -ENOMEM; 423 goto cleanup; 424 } 425 426 goto exit; 427 428 cleanup: 429 netvsc_destroy_buf(device); 430 431 exit: 432 return ret; 433 } 434 435 436 /* Negotiate NVSP protocol version */ 437 static 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 ret = vmbus_sendpacket(device->channel, init_packet, 479 sizeof(struct nvsp_message), 480 (unsigned long)init_packet, 481 VM_PKT_DATA_INBAND, 0); 482 483 return ret; 484 } 485 486 static int netvsc_connect_vsp(struct hv_device *device) 487 { 488 int ret; 489 struct netvsc_device *net_device; 490 struct nvsp_message *init_packet; 491 int ndis_version; 492 u32 ver_list[] = { NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2, 493 NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5 }; 494 int i, num_ver = 4; /* number of different NVSP versions */ 495 496 net_device = get_outbound_net_device(device); 497 if (!net_device) 498 return -ENODEV; 499 500 init_packet = &net_device->channel_init_pkt; 501 502 /* Negotiate the latest NVSP protocol supported */ 503 for (i = num_ver - 1; i >= 0; i--) 504 if (negotiate_nvsp_ver(device, net_device, init_packet, 505 ver_list[i]) == 0) { 506 net_device->nvsp_version = ver_list[i]; 507 break; 508 } 509 510 if (i < 0) { 511 ret = -EPROTO; 512 goto cleanup; 513 } 514 515 pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version); 516 517 /* Send the ndis version */ 518 memset(init_packet, 0, sizeof(struct nvsp_message)); 519 520 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4) 521 ndis_version = 0x00060001; 522 else 523 ndis_version = 0x0006001e; 524 525 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER; 526 init_packet->msg.v1_msg. 527 send_ndis_ver.ndis_major_ver = 528 (ndis_version & 0xFFFF0000) >> 16; 529 init_packet->msg.v1_msg. 530 send_ndis_ver.ndis_minor_ver = 531 ndis_version & 0xFFFF; 532 533 /* Send the init request */ 534 ret = vmbus_sendpacket(device->channel, init_packet, 535 sizeof(struct nvsp_message), 536 (unsigned long)init_packet, 537 VM_PKT_DATA_INBAND, 0); 538 if (ret != 0) 539 goto cleanup; 540 541 /* Post the big receive buffer to NetVSP */ 542 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2) 543 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY; 544 else 545 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE; 546 net_device->send_buf_size = NETVSC_SEND_BUFFER_SIZE; 547 548 ret = netvsc_init_buf(device); 549 550 cleanup: 551 return ret; 552 } 553 554 static void netvsc_disconnect_vsp(struct hv_device *device) 555 { 556 netvsc_destroy_buf(device); 557 } 558 559 /* 560 * netvsc_device_remove - Callback when the root bus device is removed 561 */ 562 int netvsc_device_remove(struct hv_device *device) 563 { 564 struct net_device *ndev = hv_get_drvdata(device); 565 struct net_device_context *net_device_ctx = netdev_priv(ndev); 566 struct netvsc_device *net_device = net_device_ctx->nvdev; 567 568 netvsc_disconnect_vsp(device); 569 570 net_device_ctx->nvdev = NULL; 571 572 /* 573 * At this point, no one should be accessing net_device 574 * except in here 575 */ 576 dev_notice(&device->device, "net device safe to remove\n"); 577 578 /* Now, we can close the channel safely */ 579 vmbus_close(device->channel); 580 581 /* Release all resources */ 582 vfree(net_device->sub_cb_buf); 583 free_netvsc_device(net_device); 584 return 0; 585 } 586 587 588 #define RING_AVAIL_PERCENT_HIWATER 20 589 #define RING_AVAIL_PERCENT_LOWATER 10 590 591 /* 592 * Get the percentage of available bytes to write in the ring. 593 * The return value is in range from 0 to 100. 594 */ 595 static inline u32 hv_ringbuf_avail_percent( 596 struct hv_ring_buffer_info *ring_info) 597 { 598 u32 avail_read, avail_write; 599 600 hv_get_ringbuffer_availbytes(ring_info, &avail_read, &avail_write); 601 602 return avail_write * 100 / ring_info->ring_datasize; 603 } 604 605 static inline void netvsc_free_send_slot(struct netvsc_device *net_device, 606 u32 index) 607 { 608 sync_change_bit(index, net_device->send_section_map); 609 } 610 611 static void netvsc_send_completion(struct netvsc_device *net_device, 612 struct vmbus_channel *incoming_channel, 613 struct hv_device *device, 614 struct vmpacket_descriptor *packet) 615 { 616 struct nvsp_message *nvsp_packet; 617 struct hv_netvsc_packet *nvsc_packet; 618 struct net_device *ndev = hv_get_drvdata(device); 619 struct net_device_context *net_device_ctx = netdev_priv(ndev); 620 u32 send_index; 621 struct sk_buff *skb; 622 623 nvsp_packet = (struct nvsp_message *)((unsigned long)packet + 624 (packet->offset8 << 3)); 625 626 if ((nvsp_packet->hdr.msg_type == NVSP_MSG_TYPE_INIT_COMPLETE) || 627 (nvsp_packet->hdr.msg_type == 628 NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE) || 629 (nvsp_packet->hdr.msg_type == 630 NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE) || 631 (nvsp_packet->hdr.msg_type == 632 NVSP_MSG5_TYPE_SUBCHANNEL)) { 633 /* Copy the response back */ 634 memcpy(&net_device->channel_init_pkt, nvsp_packet, 635 sizeof(struct nvsp_message)); 636 complete(&net_device->channel_init_wait); 637 } else if (nvsp_packet->hdr.msg_type == 638 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE) { 639 int num_outstanding_sends; 640 u16 q_idx = 0; 641 struct vmbus_channel *channel = device->channel; 642 int queue_sends; 643 644 /* Get the send context */ 645 skb = (struct sk_buff *)(unsigned long)packet->trans_id; 646 647 /* Notify the layer above us */ 648 if (skb) { 649 nvsc_packet = (struct hv_netvsc_packet *) skb->cb; 650 send_index = nvsc_packet->send_buf_index; 651 if (send_index != NETVSC_INVALID_INDEX) 652 netvsc_free_send_slot(net_device, send_index); 653 q_idx = nvsc_packet->q_idx; 654 channel = incoming_channel; 655 dev_kfree_skb_any(skb); 656 } 657 658 num_outstanding_sends = 659 atomic_dec_return(&net_device->num_outstanding_sends); 660 queue_sends = atomic_dec_return(&net_device-> 661 queue_sends[q_idx]); 662 663 if (net_device->destroy && num_outstanding_sends == 0) 664 wake_up(&net_device->wait_drain); 665 666 if (netif_tx_queue_stopped(netdev_get_tx_queue(ndev, q_idx)) && 667 !net_device_ctx->start_remove && 668 (hv_ringbuf_avail_percent(&channel->outbound) > 669 RING_AVAIL_PERCENT_HIWATER || queue_sends < 1)) 670 netif_tx_wake_queue(netdev_get_tx_queue( 671 ndev, q_idx)); 672 } else { 673 netdev_err(ndev, "Unknown send completion packet type- " 674 "%d received!!\n", nvsp_packet->hdr.msg_type); 675 } 676 677 } 678 679 static u32 netvsc_get_next_send_section(struct netvsc_device *net_device) 680 { 681 unsigned long index; 682 u32 max_words = net_device->map_words; 683 unsigned long *map_addr = (unsigned long *)net_device->send_section_map; 684 u32 section_cnt = net_device->send_section_cnt; 685 int ret_val = NETVSC_INVALID_INDEX; 686 int i; 687 int prev_val; 688 689 for (i = 0; i < max_words; i++) { 690 if (!~(map_addr[i])) 691 continue; 692 index = ffz(map_addr[i]); 693 prev_val = sync_test_and_set_bit(index, &map_addr[i]); 694 if (prev_val) 695 continue; 696 if ((index + (i * BITS_PER_LONG)) >= section_cnt) 697 break; 698 ret_val = (index + (i * BITS_PER_LONG)); 699 break; 700 } 701 return ret_val; 702 } 703 704 static u32 netvsc_copy_to_send_buf(struct netvsc_device *net_device, 705 unsigned int section_index, 706 u32 pend_size, 707 struct hv_netvsc_packet *packet, 708 struct rndis_message *rndis_msg, 709 struct hv_page_buffer **pb, 710 struct sk_buff *skb) 711 { 712 char *start = net_device->send_buf; 713 char *dest = start + (section_index * net_device->send_section_size) 714 + pend_size; 715 int i; 716 bool is_data_pkt = (skb != NULL) ? true : false; 717 bool xmit_more = (skb != NULL) ? skb->xmit_more : false; 718 u32 msg_size = 0; 719 u32 padding = 0; 720 u32 remain = packet->total_data_buflen % net_device->pkt_align; 721 u32 page_count = packet->cp_partial ? packet->rmsg_pgcnt : 722 packet->page_buf_cnt; 723 724 /* Add padding */ 725 if (is_data_pkt && xmit_more && remain && 726 !packet->cp_partial) { 727 padding = net_device->pkt_align - remain; 728 rndis_msg->msg_len += padding; 729 packet->total_data_buflen += padding; 730 } 731 732 for (i = 0; i < page_count; i++) { 733 char *src = phys_to_virt((*pb)[i].pfn << PAGE_SHIFT); 734 u32 offset = (*pb)[i].offset; 735 u32 len = (*pb)[i].len; 736 737 memcpy(dest, (src + offset), len); 738 msg_size += len; 739 dest += len; 740 } 741 742 if (padding) { 743 memset(dest, 0, padding); 744 msg_size += padding; 745 } 746 747 return msg_size; 748 } 749 750 static inline int netvsc_send_pkt( 751 struct hv_device *device, 752 struct hv_netvsc_packet *packet, 753 struct netvsc_device *net_device, 754 struct hv_page_buffer **pb, 755 struct sk_buff *skb) 756 { 757 struct nvsp_message nvmsg; 758 u16 q_idx = packet->q_idx; 759 struct vmbus_channel *out_channel = net_device->chn_table[q_idx]; 760 struct net_device *ndev = hv_get_drvdata(device); 761 u64 req_id; 762 int ret; 763 struct hv_page_buffer *pgbuf; 764 u32 ring_avail = hv_ringbuf_avail_percent(&out_channel->outbound); 765 bool xmit_more = (skb != NULL) ? skb->xmit_more : false; 766 767 nvmsg.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT; 768 if (skb != NULL) { 769 /* 0 is RMC_DATA; */ 770 nvmsg.msg.v1_msg.send_rndis_pkt.channel_type = 0; 771 } else { 772 /* 1 is RMC_CONTROL; */ 773 nvmsg.msg.v1_msg.send_rndis_pkt.channel_type = 1; 774 } 775 776 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_index = 777 packet->send_buf_index; 778 if (packet->send_buf_index == NETVSC_INVALID_INDEX) 779 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_size = 0; 780 else 781 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_size = 782 packet->total_data_buflen; 783 784 req_id = (ulong)skb; 785 786 if (out_channel->rescind) 787 return -ENODEV; 788 789 /* 790 * It is possible that once we successfully place this packet 791 * on the ringbuffer, we may stop the queue. In that case, we want 792 * to notify the host independent of the xmit_more flag. We don't 793 * need to be precise here; in the worst case we may signal the host 794 * unnecessarily. 795 */ 796 if (ring_avail < (RING_AVAIL_PERCENT_LOWATER + 1)) 797 xmit_more = false; 798 799 if (packet->page_buf_cnt) { 800 pgbuf = packet->cp_partial ? (*pb) + 801 packet->rmsg_pgcnt : (*pb); 802 ret = vmbus_sendpacket_pagebuffer_ctl(out_channel, 803 pgbuf, 804 packet->page_buf_cnt, 805 &nvmsg, 806 sizeof(struct nvsp_message), 807 req_id, 808 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED, 809 !xmit_more); 810 } else { 811 ret = vmbus_sendpacket_ctl(out_channel, &nvmsg, 812 sizeof(struct nvsp_message), 813 req_id, 814 VM_PKT_DATA_INBAND, 815 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED, 816 !xmit_more); 817 } 818 819 if (ret == 0) { 820 atomic_inc(&net_device->num_outstanding_sends); 821 atomic_inc(&net_device->queue_sends[q_idx]); 822 823 if (ring_avail < RING_AVAIL_PERCENT_LOWATER) { 824 netif_tx_stop_queue(netdev_get_tx_queue(ndev, q_idx)); 825 826 if (atomic_read(&net_device-> 827 queue_sends[q_idx]) < 1) 828 netif_tx_wake_queue(netdev_get_tx_queue( 829 ndev, q_idx)); 830 } 831 } else if (ret == -EAGAIN) { 832 netif_tx_stop_queue(netdev_get_tx_queue( 833 ndev, q_idx)); 834 if (atomic_read(&net_device->queue_sends[q_idx]) < 1) { 835 netif_tx_wake_queue(netdev_get_tx_queue( 836 ndev, q_idx)); 837 ret = -ENOSPC; 838 } 839 } else { 840 netdev_err(ndev, "Unable to send packet %p ret %d\n", 841 packet, ret); 842 } 843 844 return ret; 845 } 846 847 /* Move packet out of multi send data (msd), and clear msd */ 848 static inline void move_pkt_msd(struct hv_netvsc_packet **msd_send, 849 struct sk_buff **msd_skb, 850 struct multi_send_data *msdp) 851 { 852 *msd_skb = msdp->skb; 853 *msd_send = msdp->pkt; 854 msdp->skb = NULL; 855 msdp->pkt = NULL; 856 msdp->count = 0; 857 } 858 859 int netvsc_send(struct hv_device *device, 860 struct hv_netvsc_packet *packet, 861 struct rndis_message *rndis_msg, 862 struct hv_page_buffer **pb, 863 struct sk_buff *skb) 864 { 865 struct netvsc_device *net_device; 866 int ret = 0, m_ret = 0; 867 struct vmbus_channel *out_channel; 868 u16 q_idx = packet->q_idx; 869 u32 pktlen = packet->total_data_buflen, msd_len = 0; 870 unsigned int section_index = NETVSC_INVALID_INDEX; 871 struct multi_send_data *msdp; 872 struct hv_netvsc_packet *msd_send = NULL, *cur_send = NULL; 873 struct sk_buff *msd_skb = NULL; 874 bool try_batch; 875 bool xmit_more = (skb != NULL) ? skb->xmit_more : false; 876 877 net_device = get_outbound_net_device(device); 878 if (!net_device) 879 return -ENODEV; 880 881 out_channel = net_device->chn_table[q_idx]; 882 883 packet->send_buf_index = NETVSC_INVALID_INDEX; 884 packet->cp_partial = false; 885 886 /* Send control message directly without accessing msd (Multi-Send 887 * Data) field which may be changed during data packet processing. 888 */ 889 if (!skb) { 890 cur_send = packet; 891 goto send_now; 892 } 893 894 msdp = &net_device->msd[q_idx]; 895 896 /* batch packets in send buffer if possible */ 897 if (msdp->pkt) 898 msd_len = msdp->pkt->total_data_buflen; 899 900 try_batch = (skb != NULL) && msd_len > 0 && msdp->count < 901 net_device->max_pkt; 902 903 if (try_batch && msd_len + pktlen + net_device->pkt_align < 904 net_device->send_section_size) { 905 section_index = msdp->pkt->send_buf_index; 906 907 } else if (try_batch && msd_len + packet->rmsg_size < 908 net_device->send_section_size) { 909 section_index = msdp->pkt->send_buf_index; 910 packet->cp_partial = true; 911 912 } else if ((skb != NULL) && pktlen + net_device->pkt_align < 913 net_device->send_section_size) { 914 section_index = netvsc_get_next_send_section(net_device); 915 if (section_index != NETVSC_INVALID_INDEX) { 916 move_pkt_msd(&msd_send, &msd_skb, msdp); 917 msd_len = 0; 918 } 919 } 920 921 if (section_index != NETVSC_INVALID_INDEX) { 922 netvsc_copy_to_send_buf(net_device, 923 section_index, msd_len, 924 packet, rndis_msg, pb, skb); 925 926 packet->send_buf_index = section_index; 927 928 if (packet->cp_partial) { 929 packet->page_buf_cnt -= packet->rmsg_pgcnt; 930 packet->total_data_buflen = msd_len + packet->rmsg_size; 931 } else { 932 packet->page_buf_cnt = 0; 933 packet->total_data_buflen += msd_len; 934 } 935 936 if (msdp->skb) 937 dev_kfree_skb_any(msdp->skb); 938 939 if (xmit_more && !packet->cp_partial) { 940 msdp->skb = skb; 941 msdp->pkt = packet; 942 msdp->count++; 943 } else { 944 cur_send = packet; 945 msdp->skb = NULL; 946 msdp->pkt = NULL; 947 msdp->count = 0; 948 } 949 } else { 950 move_pkt_msd(&msd_send, &msd_skb, msdp); 951 cur_send = packet; 952 } 953 954 if (msd_send) { 955 m_ret = netvsc_send_pkt(device, msd_send, net_device, 956 NULL, msd_skb); 957 958 if (m_ret != 0) { 959 netvsc_free_send_slot(net_device, 960 msd_send->send_buf_index); 961 dev_kfree_skb_any(msd_skb); 962 } 963 } 964 965 send_now: 966 if (cur_send) 967 ret = netvsc_send_pkt(device, cur_send, net_device, pb, skb); 968 969 if (ret != 0 && section_index != NETVSC_INVALID_INDEX) 970 netvsc_free_send_slot(net_device, section_index); 971 972 return ret; 973 } 974 975 static void netvsc_send_recv_completion(struct hv_device *device, 976 struct vmbus_channel *channel, 977 struct netvsc_device *net_device, 978 u64 transaction_id, u32 status) 979 { 980 struct nvsp_message recvcompMessage; 981 int retries = 0; 982 int ret; 983 struct net_device *ndev = hv_get_drvdata(device); 984 985 recvcompMessage.hdr.msg_type = 986 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE; 987 988 recvcompMessage.msg.v1_msg.send_rndis_pkt_complete.status = status; 989 990 retry_send_cmplt: 991 /* Send the completion */ 992 ret = vmbus_sendpacket(channel, &recvcompMessage, 993 sizeof(struct nvsp_message), transaction_id, 994 VM_PKT_COMP, 0); 995 if (ret == 0) { 996 /* success */ 997 /* no-op */ 998 } else if (ret == -EAGAIN) { 999 /* no more room...wait a bit and attempt to retry 3 times */ 1000 retries++; 1001 netdev_err(ndev, "unable to send receive completion pkt" 1002 " (tid %llx)...retrying %d\n", transaction_id, retries); 1003 1004 if (retries < 4) { 1005 udelay(100); 1006 goto retry_send_cmplt; 1007 } else { 1008 netdev_err(ndev, "unable to send receive " 1009 "completion pkt (tid %llx)...give up retrying\n", 1010 transaction_id); 1011 } 1012 } else { 1013 netdev_err(ndev, "unable to send receive " 1014 "completion pkt - %llx\n", transaction_id); 1015 } 1016 } 1017 1018 static void netvsc_receive(struct netvsc_device *net_device, 1019 struct vmbus_channel *channel, 1020 struct hv_device *device, 1021 struct vmpacket_descriptor *packet) 1022 { 1023 struct vmtransfer_page_packet_header *vmxferpage_packet; 1024 struct nvsp_message *nvsp_packet; 1025 struct hv_netvsc_packet nv_pkt; 1026 struct hv_netvsc_packet *netvsc_packet = &nv_pkt; 1027 u32 status = NVSP_STAT_SUCCESS; 1028 int i; 1029 int count = 0; 1030 struct net_device *ndev = hv_get_drvdata(device); 1031 void *data; 1032 1033 /* 1034 * All inbound packets other than send completion should be xfer page 1035 * packet 1036 */ 1037 if (packet->type != VM_PKT_DATA_USING_XFER_PAGES) { 1038 netdev_err(ndev, "Unknown packet type received - %d\n", 1039 packet->type); 1040 return; 1041 } 1042 1043 nvsp_packet = (struct nvsp_message *)((unsigned long)packet + 1044 (packet->offset8 << 3)); 1045 1046 /* Make sure this is a valid nvsp packet */ 1047 if (nvsp_packet->hdr.msg_type != 1048 NVSP_MSG1_TYPE_SEND_RNDIS_PKT) { 1049 netdev_err(ndev, "Unknown nvsp packet type received-" 1050 " %d\n", nvsp_packet->hdr.msg_type); 1051 return; 1052 } 1053 1054 vmxferpage_packet = (struct vmtransfer_page_packet_header *)packet; 1055 1056 if (vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID) { 1057 netdev_err(ndev, "Invalid xfer page set id - " 1058 "expecting %x got %x\n", NETVSC_RECEIVE_BUFFER_ID, 1059 vmxferpage_packet->xfer_pageset_id); 1060 return; 1061 } 1062 1063 count = vmxferpage_packet->range_cnt; 1064 1065 /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */ 1066 for (i = 0; i < count; i++) { 1067 /* Initialize the netvsc packet */ 1068 data = (void *)((unsigned long)net_device-> 1069 recv_buf + vmxferpage_packet->ranges[i].byte_offset); 1070 netvsc_packet->total_data_buflen = 1071 vmxferpage_packet->ranges[i].byte_count; 1072 1073 /* Pass it to the upper layer */ 1074 status = rndis_filter_receive(device, netvsc_packet, &data, 1075 channel); 1076 1077 } 1078 1079 netvsc_send_recv_completion(device, channel, net_device, 1080 vmxferpage_packet->d.trans_id, status); 1081 } 1082 1083 1084 static void netvsc_send_table(struct hv_device *hdev, 1085 struct nvsp_message *nvmsg) 1086 { 1087 struct netvsc_device *nvscdev; 1088 struct net_device *ndev = hv_get_drvdata(hdev); 1089 int i; 1090 u32 count, *tab; 1091 1092 nvscdev = get_outbound_net_device(hdev); 1093 if (!nvscdev) 1094 return; 1095 1096 count = nvmsg->msg.v5_msg.send_table.count; 1097 if (count != VRSS_SEND_TAB_SIZE) { 1098 netdev_err(ndev, "Received wrong send-table size:%u\n", count); 1099 return; 1100 } 1101 1102 tab = (u32 *)((unsigned long)&nvmsg->msg.v5_msg.send_table + 1103 nvmsg->msg.v5_msg.send_table.offset); 1104 1105 for (i = 0; i < count; i++) 1106 nvscdev->send_table[i] = tab[i]; 1107 } 1108 1109 static void netvsc_send_vf(struct netvsc_device *nvdev, 1110 struct nvsp_message *nvmsg) 1111 { 1112 nvdev->vf_alloc = nvmsg->msg.v4_msg.vf_assoc.allocated; 1113 nvdev->vf_serial = nvmsg->msg.v4_msg.vf_assoc.serial; 1114 } 1115 1116 static inline void netvsc_receive_inband(struct hv_device *hdev, 1117 struct netvsc_device *nvdev, 1118 struct nvsp_message *nvmsg) 1119 { 1120 switch (nvmsg->hdr.msg_type) { 1121 case NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE: 1122 netvsc_send_table(hdev, nvmsg); 1123 break; 1124 1125 case NVSP_MSG4_TYPE_SEND_VF_ASSOCIATION: 1126 netvsc_send_vf(nvdev, nvmsg); 1127 break; 1128 } 1129 } 1130 1131 static void netvsc_process_raw_pkt(struct hv_device *device, 1132 struct vmbus_channel *channel, 1133 struct netvsc_device *net_device, 1134 struct net_device *ndev, 1135 u64 request_id, 1136 struct vmpacket_descriptor *desc) 1137 { 1138 struct nvsp_message *nvmsg; 1139 1140 nvmsg = (struct nvsp_message *)((unsigned long) 1141 desc + (desc->offset8 << 3)); 1142 1143 switch (desc->type) { 1144 case VM_PKT_COMP: 1145 netvsc_send_completion(net_device, channel, device, desc); 1146 break; 1147 1148 case VM_PKT_DATA_USING_XFER_PAGES: 1149 netvsc_receive(net_device, channel, device, desc); 1150 break; 1151 1152 case VM_PKT_DATA_INBAND: 1153 netvsc_receive_inband(device, net_device, nvmsg); 1154 break; 1155 1156 default: 1157 netdev_err(ndev, "unhandled packet type %d, tid %llx\n", 1158 desc->type, request_id); 1159 break; 1160 } 1161 } 1162 1163 1164 void netvsc_channel_cb(void *context) 1165 { 1166 int ret; 1167 struct vmbus_channel *channel = (struct vmbus_channel *)context; 1168 struct hv_device *device; 1169 struct netvsc_device *net_device; 1170 u32 bytes_recvd; 1171 u64 request_id; 1172 struct vmpacket_descriptor *desc; 1173 unsigned char *buffer; 1174 int bufferlen = NETVSC_PACKET_SIZE; 1175 struct net_device *ndev; 1176 bool need_to_commit = false; 1177 1178 if (channel->primary_channel != NULL) 1179 device = channel->primary_channel->device_obj; 1180 else 1181 device = channel->device_obj; 1182 1183 net_device = get_inbound_net_device(device); 1184 if (!net_device) 1185 return; 1186 ndev = hv_get_drvdata(device); 1187 buffer = get_per_channel_state(channel); 1188 1189 do { 1190 desc = get_next_pkt_raw(channel); 1191 if (desc != NULL) { 1192 netvsc_process_raw_pkt(device, 1193 channel, 1194 net_device, 1195 ndev, 1196 desc->trans_id, 1197 desc); 1198 1199 put_pkt_raw(channel, desc); 1200 need_to_commit = true; 1201 continue; 1202 } 1203 if (need_to_commit) { 1204 need_to_commit = false; 1205 commit_rd_index(channel); 1206 } 1207 1208 ret = vmbus_recvpacket_raw(channel, buffer, bufferlen, 1209 &bytes_recvd, &request_id); 1210 if (ret == 0) { 1211 if (bytes_recvd > 0) { 1212 desc = (struct vmpacket_descriptor *)buffer; 1213 netvsc_process_raw_pkt(device, 1214 channel, 1215 net_device, 1216 ndev, 1217 request_id, 1218 desc); 1219 1220 1221 } else { 1222 /* 1223 * We are done for this pass. 1224 */ 1225 break; 1226 } 1227 1228 } else if (ret == -ENOBUFS) { 1229 if (bufferlen > NETVSC_PACKET_SIZE) 1230 kfree(buffer); 1231 /* Handle large packet */ 1232 buffer = kmalloc(bytes_recvd, GFP_ATOMIC); 1233 if (buffer == NULL) { 1234 /* Try again next time around */ 1235 netdev_err(ndev, 1236 "unable to allocate buffer of size " 1237 "(%d)!!\n", bytes_recvd); 1238 break; 1239 } 1240 1241 bufferlen = bytes_recvd; 1242 } 1243 } while (1); 1244 1245 if (bufferlen > NETVSC_PACKET_SIZE) 1246 kfree(buffer); 1247 return; 1248 } 1249 1250 /* 1251 * netvsc_device_add - Callback when the device belonging to this 1252 * driver is added 1253 */ 1254 int netvsc_device_add(struct hv_device *device, void *additional_info) 1255 { 1256 int i, ret = 0; 1257 int ring_size = 1258 ((struct netvsc_device_info *)additional_info)->ring_size; 1259 struct netvsc_device *net_device; 1260 struct net_device *ndev = hv_get_drvdata(device); 1261 struct net_device_context *net_device_ctx = netdev_priv(ndev); 1262 1263 net_device = alloc_net_device(); 1264 if (!net_device) 1265 return -ENOMEM; 1266 1267 net_device->ring_size = ring_size; 1268 1269 /* Initialize the NetVSC channel extension */ 1270 init_completion(&net_device->channel_init_wait); 1271 1272 set_per_channel_state(device->channel, net_device->cb_buffer); 1273 1274 /* Open the channel */ 1275 ret = vmbus_open(device->channel, ring_size * PAGE_SIZE, 1276 ring_size * PAGE_SIZE, NULL, 0, 1277 netvsc_channel_cb, device->channel); 1278 1279 if (ret != 0) { 1280 netdev_err(ndev, "unable to open channel: %d\n", ret); 1281 goto cleanup; 1282 } 1283 1284 /* Channel is opened */ 1285 pr_info("hv_netvsc channel opened successfully\n"); 1286 1287 /* If we're reopening the device we may have multiple queues, fill the 1288 * chn_table with the default channel to use it before subchannels are 1289 * opened. 1290 */ 1291 for (i = 0; i < VRSS_CHANNEL_MAX; i++) 1292 net_device->chn_table[i] = device->channel; 1293 1294 /* Writing nvdev pointer unlocks netvsc_send(), make sure chn_table is 1295 * populated. 1296 */ 1297 wmb(); 1298 1299 net_device_ctx->nvdev = net_device; 1300 1301 /* Connect with the NetVsp */ 1302 ret = netvsc_connect_vsp(device); 1303 if (ret != 0) { 1304 netdev_err(ndev, 1305 "unable to connect to NetVSP - %d\n", ret); 1306 goto close; 1307 } 1308 1309 return ret; 1310 1311 close: 1312 /* Now, we can close the channel safely */ 1313 vmbus_close(device->channel); 1314 1315 cleanup: 1316 free_netvsc_device(net_device); 1317 1318 return ret; 1319 } 1320