1 /*- 2 * Copyright (c) 2016-2017 Microsoft Corp. 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 */ 26 27 #include <sys/cdefs.h> 28 __FBSDID("$FreeBSD$"); 29 30 #ifdef NEW_PCIB 31 32 #include <sys/param.h> 33 #include <sys/systm.h> 34 #include <sys/types.h> 35 #include <sys/malloc.h> 36 #include <sys/module.h> 37 #include <sys/kernel.h> 38 #include <sys/queue.h> 39 #include <sys/lock.h> 40 #include <sys/sx.h> 41 #include <sys/smp.h> 42 #include <sys/sysctl.h> 43 #include <sys/bus.h> 44 #include <sys/rman.h> 45 #include <sys/mutex.h> 46 #include <sys/errno.h> 47 48 #include <vm/vm.h> 49 #include <vm/vm_param.h> 50 #include <vm/vm_kern.h> 51 #include <vm/pmap.h> 52 53 #include <machine/atomic.h> 54 #include <machine/bus.h> 55 #include <machine/frame.h> 56 #include <machine/pci_cfgreg.h> 57 #include <machine/resource.h> 58 59 #include <sys/pciio.h> 60 #include <dev/pci/pcireg.h> 61 #include <dev/pci/pcivar.h> 62 #include <dev/pci/pci_private.h> 63 #include <dev/pci/pcib_private.h> 64 #include "pcib_if.h" 65 66 #include <machine/intr_machdep.h> 67 #include <x86/apicreg.h> 68 69 #include <dev/hyperv/include/hyperv.h> 70 #include <dev/hyperv/include/hyperv_busdma.h> 71 #include <dev/hyperv/include/vmbus_xact.h> 72 #include <dev/hyperv/vmbus/vmbus_reg.h> 73 #include <dev/hyperv/vmbus/vmbus_chanvar.h> 74 75 #include "vmbus_if.h" 76 77 #if __FreeBSD_version < 1100000 78 typedef u_long rman_res_t; 79 #define RM_MAX_END (~(rman_res_t)0) 80 #endif 81 82 struct completion { 83 unsigned int done; 84 struct mtx lock; 85 }; 86 87 static void 88 init_completion(struct completion *c) 89 { 90 memset(c, 0, sizeof(*c)); 91 mtx_init(&c->lock, "hvcmpl", NULL, MTX_DEF); 92 c->done = 0; 93 } 94 95 static void 96 free_completion(struct completion *c) 97 { 98 mtx_destroy(&c->lock); 99 } 100 101 static void 102 complete(struct completion *c) 103 { 104 mtx_lock(&c->lock); 105 c->done++; 106 mtx_unlock(&c->lock); 107 wakeup(c); 108 } 109 110 static void 111 wait_for_completion(struct completion *c) 112 { 113 mtx_lock(&c->lock); 114 while (c->done == 0) 115 mtx_sleep(c, &c->lock, 0, "hvwfc", 0); 116 c->done--; 117 mtx_unlock(&c->lock); 118 } 119 120 /* 121 * Return: 0 if completed, a non-zero value if timed out. 122 */ 123 static int 124 wait_for_completion_timeout(struct completion *c, int timeout) 125 { 126 int ret; 127 128 mtx_lock(&c->lock); 129 130 if (c->done == 0) 131 mtx_sleep(c, &c->lock, 0, "hvwfc", timeout); 132 133 if (c->done > 0) { 134 c->done--; 135 ret = 0; 136 } else { 137 ret = 1; 138 } 139 140 mtx_unlock(&c->lock); 141 142 return (ret); 143 } 144 145 #define PCI_MAKE_VERSION(major, minor) ((uint32_t)(((major) << 16) | (major))) 146 147 enum { 148 PCI_PROTOCOL_VERSION_1_1 = PCI_MAKE_VERSION(1, 1), 149 PCI_PROTOCOL_VERSION_CURRENT = PCI_PROTOCOL_VERSION_1_1 150 }; 151 152 #define PCI_CONFIG_MMIO_LENGTH 0x2000 153 #define CFG_PAGE_OFFSET 0x1000 154 #define CFG_PAGE_SIZE (PCI_CONFIG_MMIO_LENGTH - CFG_PAGE_OFFSET) 155 156 /* 157 * Message Types 158 */ 159 160 enum pci_message_type { 161 /* 162 * Version 1.1 163 */ 164 PCI_MESSAGE_BASE = 0x42490000, 165 PCI_BUS_RELATIONS = PCI_MESSAGE_BASE + 0, 166 PCI_QUERY_BUS_RELATIONS = PCI_MESSAGE_BASE + 1, 167 PCI_POWER_STATE_CHANGE = PCI_MESSAGE_BASE + 4, 168 PCI_QUERY_RESOURCE_REQUIREMENTS = PCI_MESSAGE_BASE + 5, 169 PCI_QUERY_RESOURCE_RESOURCES = PCI_MESSAGE_BASE + 6, 170 PCI_BUS_D0ENTRY = PCI_MESSAGE_BASE + 7, 171 PCI_BUS_D0EXIT = PCI_MESSAGE_BASE + 8, 172 PCI_READ_BLOCK = PCI_MESSAGE_BASE + 9, 173 PCI_WRITE_BLOCK = PCI_MESSAGE_BASE + 0xA, 174 PCI_EJECT = PCI_MESSAGE_BASE + 0xB, 175 PCI_QUERY_STOP = PCI_MESSAGE_BASE + 0xC, 176 PCI_REENABLE = PCI_MESSAGE_BASE + 0xD, 177 PCI_QUERY_STOP_FAILED = PCI_MESSAGE_BASE + 0xE, 178 PCI_EJECTION_COMPLETE = PCI_MESSAGE_BASE + 0xF, 179 PCI_RESOURCES_ASSIGNED = PCI_MESSAGE_BASE + 0x10, 180 PCI_RESOURCES_RELEASED = PCI_MESSAGE_BASE + 0x11, 181 PCI_INVALIDATE_BLOCK = PCI_MESSAGE_BASE + 0x12, 182 PCI_QUERY_PROTOCOL_VERSION = PCI_MESSAGE_BASE + 0x13, 183 PCI_CREATE_INTERRUPT_MESSAGE = PCI_MESSAGE_BASE + 0x14, 184 PCI_DELETE_INTERRUPT_MESSAGE = PCI_MESSAGE_BASE + 0x15, 185 PCI_MESSAGE_MAXIMUM 186 }; 187 188 /* 189 * Structures defining the virtual PCI Express protocol. 190 */ 191 192 union pci_version { 193 struct { 194 uint16_t minor_version; 195 uint16_t major_version; 196 } parts; 197 uint32_t version; 198 } __packed; 199 200 /* 201 * This representation is the one used in Windows, which is 202 * what is expected when sending this back and forth with 203 * the Hyper-V parent partition. 204 */ 205 union win_slot_encoding { 206 struct { 207 uint32_t slot:5; 208 uint32_t func:3; 209 uint32_t reserved:24; 210 } bits; 211 uint32_t val; 212 } __packed; 213 214 struct pci_func_desc { 215 uint16_t v_id; /* vendor ID */ 216 uint16_t d_id; /* device ID */ 217 uint8_t rev; 218 uint8_t prog_intf; 219 uint8_t subclass; 220 uint8_t base_class; 221 uint32_t subsystem_id; 222 union win_slot_encoding wslot; 223 uint32_t ser; /* serial number */ 224 } __packed; 225 226 struct hv_msi_desc { 227 uint8_t vector; 228 uint8_t delivery_mode; 229 uint16_t vector_count; 230 uint32_t reserved; 231 uint64_t cpu_mask; 232 } __packed; 233 234 struct tran_int_desc { 235 uint16_t reserved; 236 uint16_t vector_count; 237 uint32_t data; 238 uint64_t address; 239 } __packed; 240 241 struct pci_message { 242 uint32_t type; 243 } __packed; 244 245 struct pci_child_message { 246 struct pci_message message_type; 247 union win_slot_encoding wslot; 248 } __packed; 249 250 struct pci_incoming_message { 251 struct vmbus_chanpkt_hdr hdr; 252 struct pci_message message_type; 253 } __packed; 254 255 struct pci_response { 256 struct vmbus_chanpkt_hdr hdr; 257 int32_t status; /* negative values are failures */ 258 } __packed; 259 260 struct pci_packet { 261 void (*completion_func)(void *context, struct pci_response *resp, 262 int resp_packet_size); 263 void *compl_ctxt; 264 265 struct pci_message message[0]; 266 }; 267 268 /* 269 * Specific message types supporting the PCI protocol. 270 */ 271 272 struct pci_version_request { 273 struct pci_message message_type; 274 uint32_t protocol_version; 275 uint32_t is_last_attempt:1; 276 uint32_t reservedz:31; 277 } __packed; 278 279 struct pci_bus_d0_entry { 280 struct pci_message message_type; 281 uint32_t reserved; 282 uint64_t mmio_base; 283 } __packed; 284 285 struct pci_bus_relations { 286 struct pci_incoming_message incoming; 287 uint32_t device_count; 288 struct pci_func_desc func[0]; 289 } __packed; 290 291 #define MAX_NUM_BARS (PCIR_MAX_BAR_0 + 1) 292 struct pci_q_res_req_response { 293 struct vmbus_chanpkt_hdr hdr; 294 int32_t status; /* negative values are failures */ 295 uint32_t probed_bar[MAX_NUM_BARS]; 296 } __packed; 297 298 struct pci_resources_assigned { 299 struct pci_message message_type; 300 union win_slot_encoding wslot; 301 uint8_t memory_range[0x14][MAX_NUM_BARS]; /* unused here */ 302 uint32_t msi_descriptors; 303 uint32_t reserved[4]; 304 } __packed; 305 306 struct pci_create_interrupt { 307 struct pci_message message_type; 308 union win_slot_encoding wslot; 309 struct hv_msi_desc int_desc; 310 } __packed; 311 312 struct pci_create_int_response { 313 struct pci_response response; 314 uint32_t reserved; 315 struct tran_int_desc int_desc; 316 } __packed; 317 318 struct pci_delete_interrupt { 319 struct pci_message message_type; 320 union win_slot_encoding wslot; 321 struct tran_int_desc int_desc; 322 } __packed; 323 324 struct pci_dev_incoming { 325 struct pci_incoming_message incoming; 326 union win_slot_encoding wslot; 327 } __packed; 328 329 struct pci_eject_response { 330 struct pci_message message_type; 331 union win_slot_encoding wslot; 332 uint32_t status; 333 } __packed; 334 335 /* 336 * Driver specific state. 337 */ 338 339 enum hv_pcibus_state { 340 hv_pcibus_init = 0, 341 hv_pcibus_installed, 342 }; 343 344 struct hv_pcibus { 345 device_t pcib; 346 device_t pci_bus; 347 struct vmbus_pcib_softc *sc; 348 349 uint16_t pci_domain; 350 351 enum hv_pcibus_state state; 352 353 struct resource *cfg_res; 354 355 struct completion query_completion, *query_comp; 356 357 struct mtx config_lock; /* Avoid two threads writing index page */ 358 struct mtx device_list_lock; /* Protect lists below */ 359 TAILQ_HEAD(, hv_pci_dev) children; 360 TAILQ_HEAD(, hv_dr_state) dr_list; 361 362 volatile int detaching; 363 }; 364 365 struct hv_pci_dev { 366 TAILQ_ENTRY(hv_pci_dev) link; 367 368 struct pci_func_desc desc; 369 370 bool reported_missing; 371 372 struct hv_pcibus *hbus; 373 struct task eject_task; 374 375 TAILQ_HEAD(, hv_irq_desc) irq_desc_list; 376 377 /* 378 * What would be observed if one wrote 0xFFFFFFFF to a BAR and then 379 * read it back, for each of the BAR offsets within config space. 380 */ 381 uint32_t probed_bar[MAX_NUM_BARS]; 382 }; 383 384 /* 385 * Tracks "Device Relations" messages from the host, which must be both 386 * processed in order. 387 */ 388 struct hv_dr_work { 389 struct task task; 390 struct hv_pcibus *bus; 391 }; 392 393 struct hv_dr_state { 394 TAILQ_ENTRY(hv_dr_state) link; 395 uint32_t device_count; 396 struct pci_func_desc func[0]; 397 }; 398 399 struct hv_irq_desc { 400 TAILQ_ENTRY(hv_irq_desc) link; 401 struct tran_int_desc desc; 402 int irq; 403 }; 404 405 #define PCI_DEVFN(slot, func) ((((slot) & 0x1f) << 3) | ((func) & 0x07)) 406 #define PCI_SLOT(devfn) (((devfn) >> 3) & 0x1f) 407 #define PCI_FUNC(devfn) ((devfn) & 0x07) 408 409 static uint32_t 410 devfn_to_wslot(unsigned int devfn) 411 { 412 union win_slot_encoding wslot; 413 414 wslot.val = 0; 415 wslot.bits.slot = PCI_SLOT(devfn); 416 wslot.bits.func = PCI_FUNC(devfn); 417 418 return (wslot.val); 419 } 420 421 static unsigned int 422 wslot_to_devfn(uint32_t wslot) 423 { 424 union win_slot_encoding encoding; 425 unsigned int slot; 426 unsigned int func; 427 428 encoding.val = wslot; 429 430 slot = encoding.bits.slot; 431 func = encoding.bits.func; 432 433 return (PCI_DEVFN(slot, func)); 434 } 435 436 struct vmbus_pcib_softc { 437 struct vmbus_channel *chan; 438 void *rx_buf; 439 440 struct taskqueue *taskq; 441 442 struct hv_pcibus *hbus; 443 }; 444 445 /* {44C4F61D-4444-4400-9D52-802E27EDE19F} */ 446 static const struct hyperv_guid g_pass_through_dev_type = { 447 .hv_guid = {0x1D, 0xF6, 0xC4, 0x44, 0x44, 0x44, 0x00, 0x44, 448 0x9D, 0x52, 0x80, 0x2E, 0x27, 0xED, 0xE1, 0x9F} 449 }; 450 451 struct hv_pci_compl { 452 struct completion host_event; 453 int32_t completion_status; 454 }; 455 456 struct q_res_req_compl { 457 struct completion host_event; 458 struct hv_pci_dev *hpdev; 459 }; 460 461 struct compose_comp_ctxt { 462 struct hv_pci_compl comp_pkt; 463 struct tran_int_desc int_desc; 464 }; 465 466 /* 467 * It is possible the device is revoked during initialization. 468 * Check if this happens during wait. 469 * Return: 0 if response arrived, ENODEV if device revoked. 470 */ 471 static int 472 wait_for_response(struct hv_pcibus *hbus, struct completion *c) 473 { 474 do { 475 if (vmbus_chan_is_revoked(hbus->sc->chan)) { 476 device_printf(hbus->pcib, 477 "The device is revoked.\n"); 478 return (ENODEV); 479 } 480 } while (wait_for_completion_timeout(c, hz /10) != 0); 481 482 return 0; 483 } 484 485 static void 486 hv_pci_generic_compl(void *context, struct pci_response *resp, 487 int resp_packet_size) 488 { 489 struct hv_pci_compl *comp_pkt = context; 490 491 if (resp_packet_size >= sizeof(struct pci_response)) 492 comp_pkt->completion_status = resp->status; 493 else 494 comp_pkt->completion_status = -1; 495 496 complete(&comp_pkt->host_event); 497 } 498 499 static void 500 q_resource_requirements(void *context, struct pci_response *resp, 501 int resp_packet_size) 502 { 503 struct q_res_req_compl *completion = context; 504 struct pci_q_res_req_response *q_res_req = 505 (struct pci_q_res_req_response *)resp; 506 int i; 507 508 if (resp->status < 0) { 509 printf("vmbus_pcib: failed to query resource requirements\n"); 510 } else { 511 for (i = 0; i < MAX_NUM_BARS; i++) 512 completion->hpdev->probed_bar[i] = 513 q_res_req->probed_bar[i]; 514 } 515 516 complete(&completion->host_event); 517 } 518 519 static void 520 hv_pci_compose_compl(void *context, struct pci_response *resp, 521 int resp_packet_size) 522 { 523 struct compose_comp_ctxt *comp_pkt = context; 524 struct pci_create_int_response *int_resp = 525 (struct pci_create_int_response *)resp; 526 527 comp_pkt->comp_pkt.completion_status = resp->status; 528 comp_pkt->int_desc = int_resp->int_desc; 529 complete(&comp_pkt->comp_pkt.host_event); 530 } 531 532 static void 533 hv_int_desc_free(struct hv_pci_dev *hpdev, struct hv_irq_desc *hid) 534 { 535 struct pci_delete_interrupt *int_pkt; 536 struct { 537 struct pci_packet pkt; 538 uint8_t buffer[sizeof(struct pci_delete_interrupt)]; 539 } ctxt; 540 541 memset(&ctxt, 0, sizeof(ctxt)); 542 int_pkt = (struct pci_delete_interrupt *)&ctxt.pkt.message; 543 int_pkt->message_type.type = PCI_DELETE_INTERRUPT_MESSAGE; 544 int_pkt->wslot.val = hpdev->desc.wslot.val; 545 int_pkt->int_desc = hid->desc; 546 547 vmbus_chan_send(hpdev->hbus->sc->chan, VMBUS_CHANPKT_TYPE_INBAND, 0, 548 int_pkt, sizeof(*int_pkt), 0); 549 550 free(hid, M_DEVBUF); 551 } 552 553 static void 554 hv_pci_delete_device(struct hv_pci_dev *hpdev) 555 { 556 struct hv_pcibus *hbus = hpdev->hbus; 557 struct hv_irq_desc *hid, *tmp_hid; 558 device_t pci_dev; 559 int devfn; 560 561 devfn = wslot_to_devfn(hpdev->desc.wslot.val); 562 563 bus_topo_lock(); 564 565 pci_dev = pci_find_dbsf(hbus->pci_domain, 566 0, PCI_SLOT(devfn), PCI_FUNC(devfn)); 567 if (pci_dev) 568 device_delete_child(hbus->pci_bus, pci_dev); 569 570 bus_topo_unlock(); 571 572 mtx_lock(&hbus->device_list_lock); 573 TAILQ_REMOVE(&hbus->children, hpdev, link); 574 mtx_unlock(&hbus->device_list_lock); 575 576 TAILQ_FOREACH_SAFE(hid, &hpdev->irq_desc_list, link, tmp_hid) 577 hv_int_desc_free(hpdev, hid); 578 579 free(hpdev, M_DEVBUF); 580 } 581 582 static struct hv_pci_dev * 583 new_pcichild_device(struct hv_pcibus *hbus, struct pci_func_desc *desc) 584 { 585 struct hv_pci_dev *hpdev; 586 struct pci_child_message *res_req; 587 struct q_res_req_compl comp_pkt; 588 struct { 589 struct pci_packet pkt; 590 uint8_t buffer[sizeof(struct pci_child_message)]; 591 } ctxt; 592 int ret; 593 594 hpdev = malloc(sizeof(*hpdev), M_DEVBUF, M_WAITOK | M_ZERO); 595 hpdev->hbus = hbus; 596 597 TAILQ_INIT(&hpdev->irq_desc_list); 598 599 init_completion(&comp_pkt.host_event); 600 comp_pkt.hpdev = hpdev; 601 602 ctxt.pkt.compl_ctxt = &comp_pkt; 603 ctxt.pkt.completion_func = q_resource_requirements; 604 605 res_req = (struct pci_child_message *)&ctxt.pkt.message; 606 res_req->message_type.type = PCI_QUERY_RESOURCE_REQUIREMENTS; 607 res_req->wslot.val = desc->wslot.val; 608 609 ret = vmbus_chan_send(hbus->sc->chan, 610 VMBUS_CHANPKT_TYPE_INBAND, VMBUS_CHANPKT_FLAG_RC, 611 res_req, sizeof(*res_req), (uint64_t)(uintptr_t)&ctxt.pkt); 612 if (ret) 613 goto err; 614 615 if (wait_for_response(hbus, &comp_pkt.host_event)) 616 goto err; 617 618 free_completion(&comp_pkt.host_event); 619 620 hpdev->desc = *desc; 621 622 mtx_lock(&hbus->device_list_lock); 623 if (TAILQ_EMPTY(&hbus->children)) 624 hbus->pci_domain = desc->ser & 0xFFFF; 625 TAILQ_INSERT_TAIL(&hbus->children, hpdev, link); 626 mtx_unlock(&hbus->device_list_lock); 627 return (hpdev); 628 err: 629 free_completion(&comp_pkt.host_event); 630 free(hpdev, M_DEVBUF); 631 return (NULL); 632 } 633 634 #if __FreeBSD_version < 1100000 635 636 /* Old versions don't have BUS_RESCAN(). Let's copy it from FreeBSD 11. */ 637 638 static struct pci_devinfo * 639 pci_identify_function(device_t pcib, device_t dev, int domain, int busno, 640 int slot, int func, size_t dinfo_size) 641 { 642 struct pci_devinfo *dinfo; 643 644 dinfo = pci_read_device(pcib, domain, busno, slot, func, dinfo_size); 645 if (dinfo != NULL) 646 pci_add_child(dev, dinfo); 647 648 return (dinfo); 649 } 650 651 static int 652 pci_rescan(device_t dev) 653 { 654 #define REG(n, w) PCIB_READ_CONFIG(pcib, busno, s, f, n, w) 655 device_t pcib = device_get_parent(dev); 656 struct pci_softc *sc; 657 device_t child, *devlist, *unchanged; 658 int devcount, error, i, j, maxslots, oldcount; 659 int busno, domain, s, f, pcifunchigh; 660 uint8_t hdrtype; 661 662 /* No need to check for ARI on a rescan. */ 663 error = device_get_children(dev, &devlist, &devcount); 664 if (error) 665 return (error); 666 if (devcount != 0) { 667 unchanged = malloc(devcount * sizeof(device_t), M_TEMP, 668 M_NOWAIT | M_ZERO); 669 if (unchanged == NULL) { 670 free(devlist, M_TEMP); 671 return (ENOMEM); 672 } 673 } else 674 unchanged = NULL; 675 676 sc = device_get_softc(dev); 677 domain = pcib_get_domain(dev); 678 busno = pcib_get_bus(dev); 679 maxslots = PCIB_MAXSLOTS(pcib); 680 for (s = 0; s <= maxslots; s++) { 681 /* If function 0 is not present, skip to the next slot. */ 682 f = 0; 683 if (REG(PCIR_VENDOR, 2) == 0xffff) 684 continue; 685 pcifunchigh = 0; 686 hdrtype = REG(PCIR_HDRTYPE, 1); 687 if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE) 688 continue; 689 if (hdrtype & PCIM_MFDEV) 690 pcifunchigh = PCIB_MAXFUNCS(pcib); 691 for (f = 0; f <= pcifunchigh; f++) { 692 if (REG(PCIR_VENDOR, 2) == 0xffff) 693 continue; 694 695 /* 696 * Found a valid function. Check if a 697 * device_t for this device already exists. 698 */ 699 for (i = 0; i < devcount; i++) { 700 child = devlist[i]; 701 if (child == NULL) 702 continue; 703 if (pci_get_slot(child) == s && 704 pci_get_function(child) == f) { 705 unchanged[i] = child; 706 goto next_func; 707 } 708 } 709 710 pci_identify_function(pcib, dev, domain, busno, s, f, 711 sizeof(struct pci_devinfo)); 712 next_func:; 713 } 714 } 715 716 /* Remove devices that are no longer present. */ 717 for (i = 0; i < devcount; i++) { 718 if (unchanged[i] != NULL) 719 continue; 720 device_delete_child(dev, devlist[i]); 721 } 722 723 free(devlist, M_TEMP); 724 oldcount = devcount; 725 726 /* Try to attach the devices just added. */ 727 error = device_get_children(dev, &devlist, &devcount); 728 if (error) { 729 free(unchanged, M_TEMP); 730 return (error); 731 } 732 733 for (i = 0; i < devcount; i++) { 734 for (j = 0; j < oldcount; j++) { 735 if (devlist[i] == unchanged[j]) 736 goto next_device; 737 } 738 739 device_probe_and_attach(devlist[i]); 740 next_device:; 741 } 742 743 free(unchanged, M_TEMP); 744 free(devlist, M_TEMP); 745 return (0); 746 #undef REG 747 } 748 749 #else 750 751 static int 752 pci_rescan(device_t dev) 753 { 754 return (BUS_RESCAN(dev)); 755 } 756 757 #endif 758 759 static void 760 pci_devices_present_work(void *arg, int pending __unused) 761 { 762 struct hv_dr_work *dr_wrk = arg; 763 struct hv_dr_state *dr = NULL; 764 struct hv_pcibus *hbus; 765 uint32_t child_no; 766 bool found; 767 struct pci_func_desc *new_desc; 768 struct hv_pci_dev *hpdev, *tmp_hpdev; 769 struct completion *query_comp; 770 bool need_rescan = false; 771 772 hbus = dr_wrk->bus; 773 free(dr_wrk, M_DEVBUF); 774 775 /* Pull this off the queue and process it if it was the last one. */ 776 mtx_lock(&hbus->device_list_lock); 777 while (!TAILQ_EMPTY(&hbus->dr_list)) { 778 dr = TAILQ_FIRST(&hbus->dr_list); 779 TAILQ_REMOVE(&hbus->dr_list, dr, link); 780 781 /* Throw this away if the list still has stuff in it. */ 782 if (!TAILQ_EMPTY(&hbus->dr_list)) { 783 free(dr, M_DEVBUF); 784 continue; 785 } 786 } 787 mtx_unlock(&hbus->device_list_lock); 788 789 if (!dr) 790 return; 791 792 /* First, mark all existing children as reported missing. */ 793 mtx_lock(&hbus->device_list_lock); 794 TAILQ_FOREACH(hpdev, &hbus->children, link) 795 hpdev->reported_missing = true; 796 mtx_unlock(&hbus->device_list_lock); 797 798 /* Next, add back any reported devices. */ 799 for (child_no = 0; child_no < dr->device_count; child_no++) { 800 found = false; 801 new_desc = &dr->func[child_no]; 802 803 mtx_lock(&hbus->device_list_lock); 804 TAILQ_FOREACH(hpdev, &hbus->children, link) { 805 if ((hpdev->desc.wslot.val == 806 new_desc->wslot.val) && 807 (hpdev->desc.v_id == new_desc->v_id) && 808 (hpdev->desc.d_id == new_desc->d_id) && 809 (hpdev->desc.ser == new_desc->ser)) { 810 hpdev->reported_missing = false; 811 found = true; 812 break; 813 } 814 } 815 mtx_unlock(&hbus->device_list_lock); 816 817 if (!found) { 818 if (!need_rescan) 819 need_rescan = true; 820 821 hpdev = new_pcichild_device(hbus, new_desc); 822 if (!hpdev) 823 printf("vmbus_pcib: failed to add a child\n"); 824 } 825 } 826 827 /* Remove missing device(s), if any */ 828 TAILQ_FOREACH_SAFE(hpdev, &hbus->children, link, tmp_hpdev) { 829 if (hpdev->reported_missing) 830 hv_pci_delete_device(hpdev); 831 } 832 833 /* Rescan the bus to find any new device, if necessary. */ 834 if (hbus->state == hv_pcibus_installed && need_rescan) 835 pci_rescan(hbus->pci_bus); 836 837 /* Wake up hv_pci_query_relations(), if it's waiting. */ 838 query_comp = hbus->query_comp; 839 if (query_comp) { 840 hbus->query_comp = NULL; 841 complete(query_comp); 842 } 843 844 free(dr, M_DEVBUF); 845 } 846 847 static struct hv_pci_dev * 848 get_pcichild_wslot(struct hv_pcibus *hbus, uint32_t wslot) 849 { 850 struct hv_pci_dev *hpdev, *ret = NULL; 851 852 mtx_lock(&hbus->device_list_lock); 853 TAILQ_FOREACH(hpdev, &hbus->children, link) { 854 if (hpdev->desc.wslot.val == wslot) { 855 ret = hpdev; 856 break; 857 } 858 } 859 mtx_unlock(&hbus->device_list_lock); 860 861 return (ret); 862 } 863 864 static void 865 hv_pci_devices_present(struct hv_pcibus *hbus, 866 struct pci_bus_relations *relations) 867 { 868 struct hv_dr_state *dr; 869 struct hv_dr_work *dr_wrk; 870 unsigned long dr_size; 871 872 if (hbus->detaching && relations->device_count > 0) 873 return; 874 875 dr_size = offsetof(struct hv_dr_state, func) + 876 (sizeof(struct pci_func_desc) * relations->device_count); 877 dr = malloc(dr_size, M_DEVBUF, M_WAITOK | M_ZERO); 878 879 dr->device_count = relations->device_count; 880 if (dr->device_count != 0) 881 memcpy(dr->func, relations->func, 882 sizeof(struct pci_func_desc) * dr->device_count); 883 884 mtx_lock(&hbus->device_list_lock); 885 TAILQ_INSERT_TAIL(&hbus->dr_list, dr, link); 886 mtx_unlock(&hbus->device_list_lock); 887 888 dr_wrk = malloc(sizeof(*dr_wrk), M_DEVBUF, M_WAITOK | M_ZERO); 889 dr_wrk->bus = hbus; 890 TASK_INIT(&dr_wrk->task, 0, pci_devices_present_work, dr_wrk); 891 taskqueue_enqueue(hbus->sc->taskq, &dr_wrk->task); 892 } 893 894 static void 895 hv_eject_device_work(void *arg, int pending __unused) 896 { 897 struct hv_pci_dev *hpdev = arg; 898 union win_slot_encoding wslot = hpdev->desc.wslot; 899 struct hv_pcibus *hbus = hpdev->hbus; 900 struct pci_eject_response *eject_pkt; 901 struct { 902 struct pci_packet pkt; 903 uint8_t buffer[sizeof(struct pci_eject_response)]; 904 } ctxt; 905 906 hv_pci_delete_device(hpdev); 907 908 memset(&ctxt, 0, sizeof(ctxt)); 909 eject_pkt = (struct pci_eject_response *)&ctxt.pkt.message; 910 eject_pkt->message_type.type = PCI_EJECTION_COMPLETE; 911 eject_pkt->wslot.val = wslot.val; 912 vmbus_chan_send(hbus->sc->chan, VMBUS_CHANPKT_TYPE_INBAND, 0, 913 eject_pkt, sizeof(*eject_pkt), 0); 914 } 915 916 static void 917 hv_pci_eject_device(struct hv_pci_dev *hpdev) 918 { 919 struct hv_pcibus *hbus = hpdev->hbus; 920 struct taskqueue *taskq; 921 922 if (hbus->detaching) 923 return; 924 925 /* 926 * Push this task into the same taskqueue on which 927 * vmbus_pcib_attach() runs, so we're sure this task can't run 928 * concurrently with vmbus_pcib_attach(). 929 */ 930 TASK_INIT(&hpdev->eject_task, 0, hv_eject_device_work, hpdev); 931 taskq = vmbus_chan_mgmt_tq(hbus->sc->chan); 932 taskqueue_enqueue(taskq, &hpdev->eject_task); 933 } 934 935 #define PCIB_PACKET_SIZE 0x100 936 937 static void 938 vmbus_pcib_on_channel_callback(struct vmbus_channel *chan, void *arg) 939 { 940 struct vmbus_pcib_softc *sc = arg; 941 struct hv_pcibus *hbus = sc->hbus; 942 943 void *buffer; 944 int bufferlen = PCIB_PACKET_SIZE; 945 946 struct pci_packet *comp_packet; 947 struct pci_response *response; 948 struct pci_incoming_message *new_msg; 949 struct pci_bus_relations *bus_rel; 950 struct pci_dev_incoming *dev_msg; 951 struct hv_pci_dev *hpdev; 952 953 buffer = sc->rx_buf; 954 do { 955 struct vmbus_chanpkt_hdr *pkt = buffer; 956 uint32_t bytes_rxed; 957 int ret; 958 959 bytes_rxed = bufferlen; 960 ret = vmbus_chan_recv_pkt(chan, pkt, &bytes_rxed); 961 962 if (ret == ENOBUFS) { 963 /* Handle large packet */ 964 if (bufferlen > PCIB_PACKET_SIZE) { 965 free(buffer, M_DEVBUF); 966 buffer = NULL; 967 } 968 969 /* alloc new buffer */ 970 buffer = malloc(bytes_rxed, M_DEVBUF, M_WAITOK | M_ZERO); 971 bufferlen = bytes_rxed; 972 973 continue; 974 } 975 976 if (ret != 0) { 977 /* ignore EIO or EAGAIN */ 978 break; 979 } 980 981 if (bytes_rxed <= sizeof(struct pci_response)) 982 continue; 983 984 switch (pkt->cph_type) { 985 case VMBUS_CHANPKT_TYPE_COMP: 986 comp_packet = 987 (struct pci_packet *)(uintptr_t)pkt->cph_xactid; 988 response = (struct pci_response *)pkt; 989 comp_packet->completion_func(comp_packet->compl_ctxt, 990 response, bytes_rxed); 991 break; 992 case VMBUS_CHANPKT_TYPE_INBAND: 993 new_msg = (struct pci_incoming_message *)buffer; 994 995 switch (new_msg->message_type.type) { 996 case PCI_BUS_RELATIONS: 997 bus_rel = (struct pci_bus_relations *)buffer; 998 999 if (bus_rel->device_count == 0) 1000 break; 1001 1002 if (bytes_rxed < 1003 offsetof(struct pci_bus_relations, func) + 1004 (sizeof(struct pci_func_desc) * 1005 (bus_rel->device_count))) 1006 break; 1007 1008 hv_pci_devices_present(hbus, bus_rel); 1009 break; 1010 1011 case PCI_EJECT: 1012 dev_msg = (struct pci_dev_incoming *)buffer; 1013 hpdev = get_pcichild_wslot(hbus, 1014 dev_msg->wslot.val); 1015 1016 if (hpdev) 1017 hv_pci_eject_device(hpdev); 1018 1019 break; 1020 default: 1021 printf("vmbus_pcib: Unknown msg type 0x%x\n", 1022 new_msg->message_type.type); 1023 break; 1024 } 1025 break; 1026 default: 1027 printf("vmbus_pcib: Unknown VMBus msg type %hd\n", 1028 pkt->cph_type); 1029 break; 1030 } 1031 } while (1); 1032 1033 if (bufferlen > PCIB_PACKET_SIZE) 1034 free(buffer, M_DEVBUF); 1035 } 1036 1037 static int 1038 hv_pci_protocol_negotiation(struct hv_pcibus *hbus) 1039 { 1040 struct pci_version_request *version_req; 1041 struct hv_pci_compl comp_pkt; 1042 struct { 1043 struct pci_packet pkt; 1044 uint8_t buffer[sizeof(struct pci_version_request)]; 1045 } ctxt; 1046 int ret; 1047 1048 init_completion(&comp_pkt.host_event); 1049 1050 ctxt.pkt.completion_func = hv_pci_generic_compl; 1051 ctxt.pkt.compl_ctxt = &comp_pkt; 1052 version_req = (struct pci_version_request *)&ctxt.pkt.message; 1053 version_req->message_type.type = PCI_QUERY_PROTOCOL_VERSION; 1054 version_req->protocol_version = PCI_PROTOCOL_VERSION_CURRENT; 1055 version_req->is_last_attempt = 1; 1056 1057 ret = vmbus_chan_send(hbus->sc->chan, VMBUS_CHANPKT_TYPE_INBAND, 1058 VMBUS_CHANPKT_FLAG_RC, version_req, sizeof(*version_req), 1059 (uint64_t)(uintptr_t)&ctxt.pkt); 1060 if (!ret) 1061 ret = wait_for_response(hbus, &comp_pkt.host_event); 1062 1063 if (ret) { 1064 device_printf(hbus->pcib, 1065 "vmbus_pcib failed to request version: %d\n", 1066 ret); 1067 goto out; 1068 } 1069 1070 if (comp_pkt.completion_status < 0) { 1071 device_printf(hbus->pcib, 1072 "vmbus_pcib version negotiation failed: %x\n", 1073 comp_pkt.completion_status); 1074 ret = EPROTO; 1075 } else { 1076 ret = 0; 1077 } 1078 out: 1079 free_completion(&comp_pkt.host_event); 1080 return (ret); 1081 } 1082 1083 /* Ask the host to send along the list of child devices */ 1084 static int 1085 hv_pci_query_relations(struct hv_pcibus *hbus) 1086 { 1087 struct pci_message message; 1088 int ret; 1089 1090 message.type = PCI_QUERY_BUS_RELATIONS; 1091 ret = vmbus_chan_send(hbus->sc->chan, VMBUS_CHANPKT_TYPE_INBAND, 0, 1092 &message, sizeof(message), 0); 1093 return (ret); 1094 } 1095 1096 static int 1097 hv_pci_enter_d0(struct hv_pcibus *hbus) 1098 { 1099 struct pci_bus_d0_entry *d0_entry; 1100 struct hv_pci_compl comp_pkt; 1101 struct { 1102 struct pci_packet pkt; 1103 uint8_t buffer[sizeof(struct pci_bus_d0_entry)]; 1104 } ctxt; 1105 int ret; 1106 1107 /* 1108 * Tell the host that the bus is ready to use, and moved into the 1109 * powered-on state. This includes telling the host which region 1110 * of memory-mapped I/O space has been chosen for configuration space 1111 * access. 1112 */ 1113 init_completion(&comp_pkt.host_event); 1114 1115 ctxt.pkt.completion_func = hv_pci_generic_compl; 1116 ctxt.pkt.compl_ctxt = &comp_pkt; 1117 1118 d0_entry = (struct pci_bus_d0_entry *)&ctxt.pkt.message; 1119 memset(d0_entry, 0, sizeof(*d0_entry)); 1120 d0_entry->message_type.type = PCI_BUS_D0ENTRY; 1121 d0_entry->mmio_base = rman_get_start(hbus->cfg_res); 1122 1123 ret = vmbus_chan_send(hbus->sc->chan, VMBUS_CHANPKT_TYPE_INBAND, 1124 VMBUS_CHANPKT_FLAG_RC, d0_entry, sizeof(*d0_entry), 1125 (uint64_t)(uintptr_t)&ctxt.pkt); 1126 if (!ret) 1127 ret = wait_for_response(hbus, &comp_pkt.host_event); 1128 1129 if (ret) 1130 goto out; 1131 1132 if (comp_pkt.completion_status < 0) { 1133 device_printf(hbus->pcib, "vmbus_pcib failed to enable D0\n"); 1134 ret = EPROTO; 1135 } else { 1136 ret = 0; 1137 } 1138 1139 out: 1140 free_completion(&comp_pkt.host_event); 1141 return (ret); 1142 } 1143 1144 /* 1145 * It looks this is only needed by Windows VM, but let's send the message too 1146 * just to make the host happy. 1147 */ 1148 static int 1149 hv_send_resources_allocated(struct hv_pcibus *hbus) 1150 { 1151 struct pci_resources_assigned *res_assigned; 1152 struct hv_pci_compl comp_pkt; 1153 struct hv_pci_dev *hpdev; 1154 struct pci_packet *pkt; 1155 uint32_t wslot; 1156 int ret = 0; 1157 1158 pkt = malloc(sizeof(*pkt) + sizeof(*res_assigned), 1159 M_DEVBUF, M_WAITOK | M_ZERO); 1160 1161 for (wslot = 0; wslot < 256; wslot++) { 1162 hpdev = get_pcichild_wslot(hbus, wslot); 1163 if (!hpdev) 1164 continue; 1165 1166 init_completion(&comp_pkt.host_event); 1167 1168 memset(pkt, 0, sizeof(*pkt) + sizeof(*res_assigned)); 1169 pkt->completion_func = hv_pci_generic_compl; 1170 pkt->compl_ctxt = &comp_pkt; 1171 1172 res_assigned = (struct pci_resources_assigned *)&pkt->message; 1173 res_assigned->message_type.type = PCI_RESOURCES_ASSIGNED; 1174 res_assigned->wslot.val = hpdev->desc.wslot.val; 1175 1176 ret = vmbus_chan_send(hbus->sc->chan, 1177 VMBUS_CHANPKT_TYPE_INBAND, VMBUS_CHANPKT_FLAG_RC, 1178 &pkt->message, sizeof(*res_assigned), 1179 (uint64_t)(uintptr_t)pkt); 1180 if (!ret) 1181 ret = wait_for_response(hbus, &comp_pkt.host_event); 1182 1183 free_completion(&comp_pkt.host_event); 1184 1185 if (ret) 1186 break; 1187 1188 if (comp_pkt.completion_status < 0) { 1189 ret = EPROTO; 1190 device_printf(hbus->pcib, 1191 "failed to send PCI_RESOURCES_ASSIGNED\n"); 1192 break; 1193 } 1194 } 1195 1196 free(pkt, M_DEVBUF); 1197 return (ret); 1198 } 1199 1200 static int 1201 hv_send_resources_released(struct hv_pcibus *hbus) 1202 { 1203 struct pci_child_message pkt; 1204 struct hv_pci_dev *hpdev; 1205 uint32_t wslot; 1206 int ret; 1207 1208 for (wslot = 0; wslot < 256; wslot++) { 1209 hpdev = get_pcichild_wslot(hbus, wslot); 1210 if (!hpdev) 1211 continue; 1212 1213 pkt.message_type.type = PCI_RESOURCES_RELEASED; 1214 pkt.wslot.val = hpdev->desc.wslot.val; 1215 1216 ret = vmbus_chan_send(hbus->sc->chan, 1217 VMBUS_CHANPKT_TYPE_INBAND, 0, &pkt, sizeof(pkt), 0); 1218 if (ret) 1219 return (ret); 1220 } 1221 1222 return (0); 1223 } 1224 1225 #define hv_cfg_read(x, s) \ 1226 static inline uint##x##_t hv_cfg_read_##s(struct hv_pcibus *bus, \ 1227 bus_size_t offset) \ 1228 { \ 1229 return (bus_read_##s(bus->cfg_res, offset)); \ 1230 } 1231 1232 #define hv_cfg_write(x, s) \ 1233 static inline void hv_cfg_write_##s(struct hv_pcibus *bus, \ 1234 bus_size_t offset, uint##x##_t val) \ 1235 { \ 1236 return (bus_write_##s(bus->cfg_res, offset, val)); \ 1237 } 1238 1239 hv_cfg_read(8, 1) 1240 hv_cfg_read(16, 2) 1241 hv_cfg_read(32, 4) 1242 1243 hv_cfg_write(8, 1) 1244 hv_cfg_write(16, 2) 1245 hv_cfg_write(32, 4) 1246 1247 static void 1248 _hv_pcifront_read_config(struct hv_pci_dev *hpdev, int where, int size, 1249 uint32_t *val) 1250 { 1251 struct hv_pcibus *hbus = hpdev->hbus; 1252 bus_size_t addr = CFG_PAGE_OFFSET + where; 1253 1254 /* 1255 * If the attempt is to read the IDs or the ROM BAR, simulate that. 1256 */ 1257 if (where + size <= PCIR_COMMAND) { 1258 memcpy(val, ((uint8_t *)&hpdev->desc.v_id) + where, size); 1259 } else if (where >= PCIR_REVID && where + size <= 1260 PCIR_CACHELNSZ) { 1261 memcpy(val, ((uint8_t *)&hpdev->desc.rev) + where - 1262 PCIR_REVID, size); 1263 } else if (where >= PCIR_SUBVEND_0 && where + size <= 1264 PCIR_BIOS) { 1265 memcpy(val, (uint8_t *)&hpdev->desc.subsystem_id + where - 1266 PCIR_SUBVEND_0, size); 1267 } else if (where >= PCIR_BIOS && where + size <= 1268 PCIR_CAP_PTR) { 1269 /* ROM BARs are unimplemented */ 1270 *val = 0; 1271 } else if ((where >= PCIR_INTLINE && where + size <= 1272 PCIR_INTPIN) ||(where == PCIR_INTPIN && size == 1)) { 1273 /* 1274 * Interrupt Line and Interrupt PIN are hard-wired to zero 1275 * because this front-end only supports message-signaled 1276 * interrupts. 1277 */ 1278 *val = 0; 1279 } else if (where + size <= CFG_PAGE_SIZE) { 1280 mtx_lock(&hbus->config_lock); 1281 1282 /* Choose the function to be read. */ 1283 hv_cfg_write_4(hbus, 0, hpdev->desc.wslot.val); 1284 1285 /* Make sure the function was chosen before we start reading.*/ 1286 mb(); 1287 1288 /* Read from that function's config space. */ 1289 switch (size) { 1290 case 1: 1291 *((uint8_t *)val) = hv_cfg_read_1(hbus, addr); 1292 break; 1293 case 2: 1294 *((uint16_t *)val) = hv_cfg_read_2(hbus, addr); 1295 break; 1296 default: 1297 *((uint32_t *)val) = hv_cfg_read_4(hbus, addr); 1298 break; 1299 } 1300 /* 1301 * Make sure the write was done before we release the lock, 1302 * allowing consecutive reads/writes. 1303 */ 1304 mb(); 1305 1306 mtx_unlock(&hbus->config_lock); 1307 } else { 1308 /* Invalid config read: it's unlikely to reach here. */ 1309 memset(val, 0, size); 1310 } 1311 } 1312 1313 static void 1314 _hv_pcifront_write_config(struct hv_pci_dev *hpdev, int where, int size, 1315 uint32_t val) 1316 { 1317 struct hv_pcibus *hbus = hpdev->hbus; 1318 bus_size_t addr = CFG_PAGE_OFFSET + where; 1319 1320 /* SSIDs and ROM BARs are read-only */ 1321 if (where >= PCIR_SUBVEND_0 && where + size <= PCIR_CAP_PTR) 1322 return; 1323 1324 if (where >= PCIR_COMMAND && where + size <= CFG_PAGE_SIZE) { 1325 mtx_lock(&hbus->config_lock); 1326 1327 /* Choose the function to be written. */ 1328 hv_cfg_write_4(hbus, 0, hpdev->desc.wslot.val); 1329 1330 /* Make sure the function was chosen before we start writing.*/ 1331 wmb(); 1332 1333 /* Write to that function's config space. */ 1334 switch (size) { 1335 case 1: 1336 hv_cfg_write_1(hbus, addr, (uint8_t)val); 1337 break; 1338 case 2: 1339 hv_cfg_write_2(hbus, addr, (uint16_t)val); 1340 break; 1341 default: 1342 hv_cfg_write_4(hbus, addr, (uint32_t)val); 1343 break; 1344 } 1345 1346 /* 1347 * Make sure the write was done before we release the lock, 1348 * allowing consecutive reads/writes. 1349 */ 1350 mb(); 1351 1352 mtx_unlock(&hbus->config_lock); 1353 } else { 1354 /* Invalid config write: it's unlikely to reach here. */ 1355 return; 1356 } 1357 } 1358 1359 /* 1360 * The vPCI in some Hyper-V releases do not initialize the last 4 1361 * bit of BAR registers. This could result weird problems causing PCI 1362 * code fail to configure BAR correctly. 1363 * 1364 * Just write all 1's to those BARs whose probed values are not zero. 1365 * This seems to make the Hyper-V vPCI and pci_write_bar() to cooperate 1366 * correctly. 1367 */ 1368 1369 static void 1370 vmbus_pcib_prepopulate_bars(struct hv_pcibus *hbus) 1371 { 1372 struct hv_pci_dev *hpdev; 1373 int i; 1374 1375 mtx_lock(&hbus->device_list_lock); 1376 TAILQ_FOREACH(hpdev, &hbus->children, link) { 1377 for (i = 0; i < 6; i++) { 1378 /* Ignore empty bar */ 1379 if (hpdev->probed_bar[i] == 0) 1380 continue; 1381 1382 uint32_t bar_val = 0; 1383 1384 _hv_pcifront_read_config(hpdev, PCIR_BAR(i), 1385 4, &bar_val); 1386 1387 if (hpdev->probed_bar[i] != bar_val) { 1388 if (bootverbose) 1389 printf("vmbus_pcib: initialize bar %d " 1390 "by writing all 1s\n", i); 1391 1392 _hv_pcifront_write_config(hpdev, PCIR_BAR(i), 1393 4, 0xffffffff); 1394 1395 /* Now write the original value back */ 1396 _hv_pcifront_write_config(hpdev, PCIR_BAR(i), 1397 4, bar_val); 1398 } 1399 } 1400 } 1401 mtx_unlock(&hbus->device_list_lock); 1402 } 1403 1404 static void 1405 vmbus_pcib_set_detaching(void *arg, int pending __unused) 1406 { 1407 struct hv_pcibus *hbus = arg; 1408 1409 atomic_set_int(&hbus->detaching, 1); 1410 } 1411 1412 static void 1413 vmbus_pcib_pre_detach(struct hv_pcibus *hbus) 1414 { 1415 struct task task; 1416 1417 TASK_INIT(&task, 0, vmbus_pcib_set_detaching, hbus); 1418 1419 /* 1420 * Make sure the channel callback won't push any possible new 1421 * PCI_BUS_RELATIONS and PCI_EJECT tasks to sc->taskq. 1422 */ 1423 vmbus_chan_run_task(hbus->sc->chan, &task); 1424 1425 taskqueue_drain_all(hbus->sc->taskq); 1426 } 1427 1428 1429 /* 1430 * Standard probe entry point. 1431 * 1432 */ 1433 static int 1434 vmbus_pcib_probe(device_t dev) 1435 { 1436 if (VMBUS_PROBE_GUID(device_get_parent(dev), dev, 1437 &g_pass_through_dev_type) == 0) { 1438 device_set_desc(dev, "Hyper-V PCI Express Pass Through"); 1439 return (BUS_PROBE_DEFAULT); 1440 } 1441 return (ENXIO); 1442 } 1443 1444 /* 1445 * Standard attach entry point. 1446 * 1447 */ 1448 static int 1449 vmbus_pcib_attach(device_t dev) 1450 { 1451 const int pci_ring_size = (4 * PAGE_SIZE); 1452 const struct hyperv_guid *inst_guid; 1453 struct vmbus_channel *channel; 1454 struct vmbus_pcib_softc *sc; 1455 struct hv_pcibus *hbus; 1456 int rid = 0; 1457 int ret; 1458 1459 hbus = malloc(sizeof(*hbus), M_DEVBUF, M_WAITOK | M_ZERO); 1460 hbus->pcib = dev; 1461 1462 channel = vmbus_get_channel(dev); 1463 inst_guid = vmbus_chan_guid_inst(channel); 1464 hbus->pci_domain = inst_guid->hv_guid[9] | 1465 (inst_guid->hv_guid[8] << 8); 1466 1467 mtx_init(&hbus->config_lock, "hbcfg", NULL, MTX_DEF); 1468 mtx_init(&hbus->device_list_lock, "hbdl", NULL, MTX_DEF); 1469 TAILQ_INIT(&hbus->children); 1470 TAILQ_INIT(&hbus->dr_list); 1471 1472 hbus->cfg_res = bus_alloc_resource(dev, SYS_RES_MEMORY, &rid, 1473 0, RM_MAX_END, PCI_CONFIG_MMIO_LENGTH, 1474 RF_ACTIVE | rman_make_alignment_flags(PAGE_SIZE)); 1475 1476 if (!hbus->cfg_res) { 1477 device_printf(dev, "failed to get resource for cfg window\n"); 1478 ret = ENXIO; 1479 goto free_bus; 1480 } 1481 1482 sc = device_get_softc(dev); 1483 sc->chan = channel; 1484 sc->rx_buf = malloc(PCIB_PACKET_SIZE, M_DEVBUF, M_WAITOK | M_ZERO); 1485 sc->hbus = hbus; 1486 1487 /* 1488 * The taskq is used to handle PCI_BUS_RELATIONS and PCI_EJECT 1489 * messages. NB: we can't handle the messages in the channel callback 1490 * directly, because the message handlers need to send new messages 1491 * to the host and waits for the host's completion messages, which 1492 * must also be handled by the channel callback. 1493 */ 1494 sc->taskq = taskqueue_create("vmbus_pcib_tq", M_WAITOK, 1495 taskqueue_thread_enqueue, &sc->taskq); 1496 taskqueue_start_threads(&sc->taskq, 1, PI_NET, "vmbus_pcib_tq"); 1497 1498 hbus->sc = sc; 1499 1500 init_completion(&hbus->query_completion); 1501 hbus->query_comp = &hbus->query_completion; 1502 1503 ret = vmbus_chan_open(sc->chan, pci_ring_size, pci_ring_size, 1504 NULL, 0, vmbus_pcib_on_channel_callback, sc); 1505 if (ret) 1506 goto free_res; 1507 1508 ret = hv_pci_protocol_negotiation(hbus); 1509 if (ret) 1510 goto vmbus_close; 1511 1512 ret = hv_pci_query_relations(hbus); 1513 if (!ret) 1514 ret = wait_for_response(hbus, hbus->query_comp); 1515 1516 if (ret) 1517 goto vmbus_close; 1518 1519 ret = hv_pci_enter_d0(hbus); 1520 if (ret) 1521 goto vmbus_close; 1522 1523 ret = hv_send_resources_allocated(hbus); 1524 if (ret) 1525 goto vmbus_close; 1526 1527 vmbus_pcib_prepopulate_bars(hbus); 1528 1529 hbus->pci_bus = device_add_child(dev, "pci", -1); 1530 if (!hbus->pci_bus) { 1531 device_printf(dev, "failed to create pci bus\n"); 1532 ret = ENXIO; 1533 goto vmbus_close; 1534 } 1535 1536 bus_generic_attach(dev); 1537 1538 hbus->state = hv_pcibus_installed; 1539 1540 return (0); 1541 1542 vmbus_close: 1543 vmbus_pcib_pre_detach(hbus); 1544 vmbus_chan_close(sc->chan); 1545 free_res: 1546 taskqueue_free(sc->taskq); 1547 free_completion(&hbus->query_completion); 1548 free(sc->rx_buf, M_DEVBUF); 1549 bus_release_resource(dev, SYS_RES_MEMORY, 0, hbus->cfg_res); 1550 free_bus: 1551 mtx_destroy(&hbus->device_list_lock); 1552 mtx_destroy(&hbus->config_lock); 1553 free(hbus, M_DEVBUF); 1554 return (ret); 1555 } 1556 1557 /* 1558 * Standard detach entry point 1559 */ 1560 static int 1561 vmbus_pcib_detach(device_t dev) 1562 { 1563 struct vmbus_pcib_softc *sc = device_get_softc(dev); 1564 struct hv_pcibus *hbus = sc->hbus; 1565 struct pci_message teardown_packet; 1566 struct pci_bus_relations relations; 1567 int ret; 1568 1569 vmbus_pcib_pre_detach(hbus); 1570 1571 if (hbus->state == hv_pcibus_installed) 1572 bus_generic_detach(dev); 1573 1574 /* Delete any children which might still exist. */ 1575 memset(&relations, 0, sizeof(relations)); 1576 hv_pci_devices_present(hbus, &relations); 1577 1578 ret = hv_send_resources_released(hbus); 1579 if (ret) 1580 device_printf(dev, "failed to send PCI_RESOURCES_RELEASED\n"); 1581 1582 teardown_packet.type = PCI_BUS_D0EXIT; 1583 ret = vmbus_chan_send(sc->chan, VMBUS_CHANPKT_TYPE_INBAND, 0, 1584 &teardown_packet, sizeof(struct pci_message), 0); 1585 if (ret) 1586 device_printf(dev, "failed to send PCI_BUS_D0EXIT\n"); 1587 1588 taskqueue_drain_all(hbus->sc->taskq); 1589 vmbus_chan_close(sc->chan); 1590 taskqueue_free(sc->taskq); 1591 1592 free_completion(&hbus->query_completion); 1593 free(sc->rx_buf, M_DEVBUF); 1594 bus_release_resource(dev, SYS_RES_MEMORY, 0, hbus->cfg_res); 1595 1596 mtx_destroy(&hbus->device_list_lock); 1597 mtx_destroy(&hbus->config_lock); 1598 free(hbus, M_DEVBUF); 1599 1600 return (0); 1601 } 1602 1603 static int 1604 vmbus_pcib_read_ivar(device_t dev, device_t child, int which, uintptr_t *val) 1605 { 1606 struct vmbus_pcib_softc *sc = device_get_softc(dev); 1607 1608 switch (which) { 1609 case PCIB_IVAR_DOMAIN: 1610 *val = sc->hbus->pci_domain; 1611 return (0); 1612 1613 case PCIB_IVAR_BUS: 1614 /* There is only bus 0. */ 1615 *val = 0; 1616 return (0); 1617 } 1618 return (ENOENT); 1619 } 1620 1621 static int 1622 vmbus_pcib_write_ivar(device_t dev, device_t child, int which, uintptr_t val) 1623 { 1624 return (ENOENT); 1625 } 1626 1627 static struct resource * 1628 vmbus_pcib_alloc_resource(device_t dev, device_t child, int type, int *rid, 1629 rman_res_t start, rman_res_t end, rman_res_t count, u_int flags) 1630 { 1631 unsigned int bar_no; 1632 struct hv_pci_dev *hpdev; 1633 struct vmbus_pcib_softc *sc = device_get_softc(dev); 1634 struct resource *res; 1635 unsigned int devfn; 1636 1637 if (type == PCI_RES_BUS) 1638 return (pci_domain_alloc_bus(sc->hbus->pci_domain, child, rid, 1639 start, end, count, flags)); 1640 1641 /* Devices with port I/O BAR are not supported. */ 1642 if (type == SYS_RES_IOPORT) 1643 return (NULL); 1644 1645 if (type == SYS_RES_MEMORY) { 1646 devfn = PCI_DEVFN(pci_get_slot(child), 1647 pci_get_function(child)); 1648 hpdev = get_pcichild_wslot(sc->hbus, devfn_to_wslot(devfn)); 1649 if (!hpdev) 1650 return (NULL); 1651 1652 bar_no = PCI_RID2BAR(*rid); 1653 if (bar_no >= MAX_NUM_BARS) 1654 return (NULL); 1655 1656 /* Make sure a 32-bit BAR gets a 32-bit address */ 1657 if (!(hpdev->probed_bar[bar_no] & PCIM_BAR_MEM_64)) 1658 end = ulmin(end, 0xFFFFFFFF); 1659 } 1660 1661 res = bus_generic_alloc_resource(dev, child, type, rid, 1662 start, end, count, flags); 1663 /* 1664 * If this is a request for a specific range, assume it is 1665 * correct and pass it up to the parent. 1666 */ 1667 if (res == NULL && start + count - 1 == end) 1668 res = bus_generic_alloc_resource(dev, child, type, rid, 1669 start, end, count, flags); 1670 return (res); 1671 } 1672 1673 static int 1674 vmbus_pcib_release_resource(device_t dev, device_t child, int type, int rid, 1675 struct resource *r) 1676 { 1677 struct vmbus_pcib_softc *sc = device_get_softc(dev); 1678 1679 if (type == PCI_RES_BUS) 1680 return (pci_domain_release_bus(sc->hbus->pci_domain, child, 1681 rid, r)); 1682 1683 if (type == SYS_RES_IOPORT) 1684 return (EINVAL); 1685 1686 return (bus_generic_release_resource(dev, child, type, rid, r)); 1687 } 1688 1689 #if __FreeBSD_version >= 1100000 1690 static int 1691 vmbus_pcib_get_cpus(device_t pcib, device_t dev, enum cpu_sets op, 1692 size_t setsize, cpuset_t *cpuset) 1693 { 1694 return (bus_get_cpus(pcib, op, setsize, cpuset)); 1695 } 1696 #endif 1697 1698 static uint32_t 1699 vmbus_pcib_read_config(device_t dev, u_int bus, u_int slot, u_int func, 1700 u_int reg, int bytes) 1701 { 1702 struct vmbus_pcib_softc *sc = device_get_softc(dev); 1703 struct hv_pci_dev *hpdev; 1704 unsigned int devfn = PCI_DEVFN(slot, func); 1705 uint32_t data = 0; 1706 1707 KASSERT(bus == 0, ("bus should be 0, but is %u", bus)); 1708 1709 hpdev = get_pcichild_wslot(sc->hbus, devfn_to_wslot(devfn)); 1710 if (!hpdev) 1711 return (~0); 1712 1713 _hv_pcifront_read_config(hpdev, reg, bytes, &data); 1714 1715 return (data); 1716 } 1717 1718 static void 1719 vmbus_pcib_write_config(device_t dev, u_int bus, u_int slot, u_int func, 1720 u_int reg, uint32_t data, int bytes) 1721 { 1722 struct vmbus_pcib_softc *sc = device_get_softc(dev); 1723 struct hv_pci_dev *hpdev; 1724 unsigned int devfn = PCI_DEVFN(slot, func); 1725 1726 KASSERT(bus == 0, ("bus should be 0, but is %u", bus)); 1727 1728 hpdev = get_pcichild_wslot(sc->hbus, devfn_to_wslot(devfn)); 1729 if (!hpdev) 1730 return; 1731 1732 _hv_pcifront_write_config(hpdev, reg, bytes, data); 1733 } 1734 1735 static int 1736 vmbus_pcib_route_intr(device_t pcib, device_t dev, int pin) 1737 { 1738 /* We only support MSI/MSI-X and don't support INTx interrupt. */ 1739 return (PCI_INVALID_IRQ); 1740 } 1741 1742 static int 1743 vmbus_pcib_alloc_msi(device_t pcib, device_t dev, int count, 1744 int maxcount, int *irqs) 1745 { 1746 return (PCIB_ALLOC_MSI(device_get_parent(pcib), dev, count, maxcount, 1747 irqs)); 1748 } 1749 1750 static int 1751 vmbus_pcib_release_msi(device_t pcib, device_t dev, int count, int *irqs) 1752 { 1753 return (PCIB_RELEASE_MSI(device_get_parent(pcib), dev, count, irqs)); 1754 } 1755 1756 static int 1757 vmbus_pcib_alloc_msix(device_t pcib, device_t dev, int *irq) 1758 { 1759 return (PCIB_ALLOC_MSIX(device_get_parent(pcib), dev, irq)); 1760 } 1761 1762 static int 1763 vmbus_pcib_release_msix(device_t pcib, device_t dev, int irq) 1764 { 1765 return (PCIB_RELEASE_MSIX(device_get_parent(pcib), dev, irq)); 1766 } 1767 1768 #define MSI_INTEL_ADDR_DEST 0x000ff000 1769 #define MSI_INTEL_DATA_INTVEC IOART_INTVEC /* Interrupt vector. */ 1770 #define MSI_INTEL_DATA_DELFIXED IOART_DELFIXED 1771 1772 static int 1773 vmbus_pcib_map_msi(device_t pcib, device_t child, int irq, 1774 uint64_t *addr, uint32_t *data) 1775 { 1776 unsigned int devfn; 1777 struct hv_pci_dev *hpdev; 1778 1779 uint64_t v_addr; 1780 uint32_t v_data; 1781 struct hv_irq_desc *hid, *tmp_hid; 1782 unsigned int cpu, vcpu_id; 1783 unsigned int vector; 1784 1785 struct vmbus_pcib_softc *sc = device_get_softc(pcib); 1786 struct pci_create_interrupt *int_pkt; 1787 struct compose_comp_ctxt comp; 1788 struct { 1789 struct pci_packet pkt; 1790 uint8_t buffer[sizeof(struct pci_create_interrupt)]; 1791 } ctxt; 1792 1793 int ret; 1794 1795 devfn = PCI_DEVFN(pci_get_slot(child), pci_get_function(child)); 1796 hpdev = get_pcichild_wslot(sc->hbus, devfn_to_wslot(devfn)); 1797 if (!hpdev) 1798 return (ENOENT); 1799 1800 ret = PCIB_MAP_MSI(device_get_parent(pcib), child, irq, 1801 &v_addr, &v_data); 1802 if (ret) 1803 return (ret); 1804 1805 TAILQ_FOREACH_SAFE(hid, &hpdev->irq_desc_list, link, tmp_hid) { 1806 if (hid->irq == irq) { 1807 TAILQ_REMOVE(&hpdev->irq_desc_list, hid, link); 1808 hv_int_desc_free(hpdev, hid); 1809 break; 1810 } 1811 } 1812 1813 cpu = (v_addr & MSI_INTEL_ADDR_DEST) >> 12; 1814 vcpu_id = VMBUS_GET_VCPU_ID(device_get_parent(pcib), pcib, cpu); 1815 vector = v_data & MSI_INTEL_DATA_INTVEC; 1816 1817 init_completion(&comp.comp_pkt.host_event); 1818 1819 memset(&ctxt, 0, sizeof(ctxt)); 1820 ctxt.pkt.completion_func = hv_pci_compose_compl; 1821 ctxt.pkt.compl_ctxt = ∁ 1822 1823 int_pkt = (struct pci_create_interrupt *)&ctxt.pkt.message; 1824 int_pkt->message_type.type = PCI_CREATE_INTERRUPT_MESSAGE; 1825 int_pkt->wslot.val = hpdev->desc.wslot.val; 1826 int_pkt->int_desc.vector = vector; 1827 int_pkt->int_desc.vector_count = 1; 1828 int_pkt->int_desc.delivery_mode = MSI_INTEL_DATA_DELFIXED; 1829 int_pkt->int_desc.cpu_mask = 1ULL << vcpu_id; 1830 1831 ret = vmbus_chan_send(sc->chan, VMBUS_CHANPKT_TYPE_INBAND, 1832 VMBUS_CHANPKT_FLAG_RC, int_pkt, sizeof(*int_pkt), 1833 (uint64_t)(uintptr_t)&ctxt.pkt); 1834 if (ret) { 1835 free_completion(&comp.comp_pkt.host_event); 1836 return (ret); 1837 } 1838 1839 wait_for_completion(&comp.comp_pkt.host_event); 1840 free_completion(&comp.comp_pkt.host_event); 1841 1842 if (comp.comp_pkt.completion_status < 0) 1843 return (EPROTO); 1844 1845 *addr = comp.int_desc.address; 1846 *data = comp.int_desc.data; 1847 1848 hid = malloc(sizeof(struct hv_irq_desc), M_DEVBUF, M_WAITOK | M_ZERO); 1849 hid->irq = irq; 1850 hid->desc = comp.int_desc; 1851 TAILQ_INSERT_TAIL(&hpdev->irq_desc_list, hid, link); 1852 1853 return (0); 1854 } 1855 1856 static device_method_t vmbus_pcib_methods[] = { 1857 /* Device interface */ 1858 DEVMETHOD(device_probe, vmbus_pcib_probe), 1859 DEVMETHOD(device_attach, vmbus_pcib_attach), 1860 DEVMETHOD(device_detach, vmbus_pcib_detach), 1861 DEVMETHOD(device_shutdown, bus_generic_shutdown), 1862 DEVMETHOD(device_suspend, bus_generic_suspend), 1863 DEVMETHOD(device_resume, bus_generic_resume), 1864 1865 /* Bus interface */ 1866 DEVMETHOD(bus_read_ivar, vmbus_pcib_read_ivar), 1867 DEVMETHOD(bus_write_ivar, vmbus_pcib_write_ivar), 1868 DEVMETHOD(bus_alloc_resource, vmbus_pcib_alloc_resource), 1869 DEVMETHOD(bus_release_resource, vmbus_pcib_release_resource), 1870 DEVMETHOD(bus_activate_resource, bus_generic_activate_resource), 1871 DEVMETHOD(bus_deactivate_resource, bus_generic_deactivate_resource), 1872 DEVMETHOD(bus_setup_intr, bus_generic_setup_intr), 1873 DEVMETHOD(bus_teardown_intr, bus_generic_teardown_intr), 1874 #if __FreeBSD_version >= 1100000 1875 DEVMETHOD(bus_get_cpus, vmbus_pcib_get_cpus), 1876 #endif 1877 1878 /* pcib interface */ 1879 DEVMETHOD(pcib_maxslots, pcib_maxslots), 1880 DEVMETHOD(pcib_read_config, vmbus_pcib_read_config), 1881 DEVMETHOD(pcib_write_config, vmbus_pcib_write_config), 1882 DEVMETHOD(pcib_route_interrupt, vmbus_pcib_route_intr), 1883 DEVMETHOD(pcib_alloc_msi, vmbus_pcib_alloc_msi), 1884 DEVMETHOD(pcib_release_msi, vmbus_pcib_release_msi), 1885 DEVMETHOD(pcib_alloc_msix, vmbus_pcib_alloc_msix), 1886 DEVMETHOD(pcib_release_msix, vmbus_pcib_release_msix), 1887 DEVMETHOD(pcib_map_msi, vmbus_pcib_map_msi), 1888 DEVMETHOD(pcib_request_feature, pcib_request_feature_allow), 1889 1890 DEVMETHOD_END 1891 }; 1892 1893 DEFINE_CLASS_0(pcib, vmbus_pcib_driver, vmbus_pcib_methods, 1894 sizeof(struct vmbus_pcib_softc)); 1895 DRIVER_MODULE(vmbus_pcib, vmbus, vmbus_pcib_driver, 0, 0); 1896 MODULE_DEPEND(vmbus_pcib, vmbus, 1, 1, 1); 1897 MODULE_DEPEND(vmbus_pcib, pci, 1, 1, 1); 1898 1899 #endif /* NEW_PCIB */ 1900