1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2011 NetApp, Inc. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND 17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19 * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE 20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26 * SUCH DAMAGE. 27 * 28 * $FreeBSD$ 29 */ 30 31 #include <sys/cdefs.h> 32 __FBSDID("$FreeBSD$"); 33 34 #include <sys/param.h> 35 #include <sys/systm.h> 36 #include <sys/kernel.h> 37 #include <sys/malloc.h> 38 #include <sys/module.h> 39 #include <sys/bus.h> 40 #include <sys/pciio.h> 41 #include <sys/rman.h> 42 #include <sys/smp.h> 43 #include <sys/sysctl.h> 44 45 #include <dev/pci/pcivar.h> 46 #include <dev/pci/pcireg.h> 47 48 #include <machine/resource.h> 49 50 #include <machine/vmm.h> 51 #include <machine/vmm_dev.h> 52 53 #include "vmm_lapic.h" 54 #include "vmm_ktr.h" 55 56 #include "iommu.h" 57 #include "ppt.h" 58 59 /* XXX locking */ 60 61 #define MAX_MSIMSGS 32 62 63 /* 64 * If the MSI-X table is located in the middle of a BAR then that MMIO 65 * region gets split into two segments - one segment above the MSI-X table 66 * and the other segment below the MSI-X table - with a hole in place of 67 * the MSI-X table so accesses to it can be trapped and emulated. 68 * 69 * So, allocate a MMIO segment for each BAR register + 1 additional segment. 70 */ 71 #define MAX_MMIOSEGS ((PCIR_MAX_BAR_0 + 1) + 1) 72 73 MALLOC_DEFINE(M_PPTMSIX, "pptmsix", "Passthru MSI-X resources"); 74 75 struct pptintr_arg { /* pptintr(pptintr_arg) */ 76 struct pptdev *pptdev; 77 uint64_t addr; 78 uint64_t msg_data; 79 }; 80 81 struct pptseg { 82 vm_paddr_t gpa; 83 size_t len; 84 int wired; 85 }; 86 87 struct pptdev { 88 device_t dev; 89 struct vm *vm; /* owner of this device */ 90 TAILQ_ENTRY(pptdev) next; 91 struct pptseg mmio[MAX_MMIOSEGS]; 92 struct { 93 int num_msgs; /* guest state */ 94 95 int startrid; /* host state */ 96 struct resource *res[MAX_MSIMSGS]; 97 void *cookie[MAX_MSIMSGS]; 98 struct pptintr_arg arg[MAX_MSIMSGS]; 99 } msi; 100 101 struct { 102 int num_msgs; 103 int startrid; 104 int msix_table_rid; 105 int msix_pba_rid; 106 struct resource *msix_table_res; 107 struct resource *msix_pba_res; 108 struct resource **res; 109 void **cookie; 110 struct pptintr_arg *arg; 111 } msix; 112 }; 113 114 SYSCTL_DECL(_hw_vmm); 115 SYSCTL_NODE(_hw_vmm, OID_AUTO, ppt, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 116 "bhyve passthru devices"); 117 118 static int num_pptdevs; 119 SYSCTL_INT(_hw_vmm_ppt, OID_AUTO, devices, CTLFLAG_RD, &num_pptdevs, 0, 120 "number of pci passthru devices"); 121 122 static TAILQ_HEAD(, pptdev) pptdev_list = TAILQ_HEAD_INITIALIZER(pptdev_list); 123 124 static int 125 ppt_probe(device_t dev) 126 { 127 int bus, slot, func; 128 struct pci_devinfo *dinfo; 129 130 dinfo = (struct pci_devinfo *)device_get_ivars(dev); 131 132 bus = pci_get_bus(dev); 133 slot = pci_get_slot(dev); 134 func = pci_get_function(dev); 135 136 /* 137 * To qualify as a pci passthrough device a device must: 138 * - be allowed by administrator to be used in this role 139 * - be an endpoint device 140 */ 141 if ((dinfo->cfg.hdrtype & PCIM_HDRTYPE) != PCIM_HDRTYPE_NORMAL) 142 return (ENXIO); 143 else if (vmm_is_pptdev(bus, slot, func)) 144 return (0); 145 else 146 /* 147 * Returning BUS_PROBE_NOWILDCARD here matches devices that the 148 * SR-IOV infrastructure specified as "ppt" passthrough devices. 149 * All normal devices that did not have "ppt" specified as their 150 * driver will not be matched by this. 151 */ 152 return (BUS_PROBE_NOWILDCARD); 153 } 154 155 static int 156 ppt_attach(device_t dev) 157 { 158 struct pptdev *ppt; 159 160 ppt = device_get_softc(dev); 161 162 iommu_remove_device(iommu_host_domain(), pci_get_rid(dev)); 163 num_pptdevs++; 164 TAILQ_INSERT_TAIL(&pptdev_list, ppt, next); 165 ppt->dev = dev; 166 167 if (bootverbose) 168 device_printf(dev, "attached\n"); 169 170 return (0); 171 } 172 173 static int 174 ppt_detach(device_t dev) 175 { 176 struct pptdev *ppt; 177 178 ppt = device_get_softc(dev); 179 180 if (ppt->vm != NULL) 181 return (EBUSY); 182 num_pptdevs--; 183 TAILQ_REMOVE(&pptdev_list, ppt, next); 184 pci_disable_busmaster(dev); 185 iommu_add_device(iommu_host_domain(), pci_get_rid(dev)); 186 187 return (0); 188 } 189 190 static device_method_t ppt_methods[] = { 191 /* Device interface */ 192 DEVMETHOD(device_probe, ppt_probe), 193 DEVMETHOD(device_attach, ppt_attach), 194 DEVMETHOD(device_detach, ppt_detach), 195 {0, 0} 196 }; 197 198 static devclass_t ppt_devclass; 199 DEFINE_CLASS_0(ppt, ppt_driver, ppt_methods, sizeof(struct pptdev)); 200 DRIVER_MODULE(ppt, pci, ppt_driver, ppt_devclass, NULL, NULL); 201 202 static int 203 ppt_find(struct vm *vm, int bus, int slot, int func, struct pptdev **pptp) 204 { 205 device_t dev; 206 struct pptdev *ppt; 207 int b, s, f; 208 209 TAILQ_FOREACH(ppt, &pptdev_list, next) { 210 dev = ppt->dev; 211 b = pci_get_bus(dev); 212 s = pci_get_slot(dev); 213 f = pci_get_function(dev); 214 if (bus == b && slot == s && func == f) 215 break; 216 } 217 218 if (ppt == NULL) 219 return (ENOENT); 220 if (ppt->vm != vm) /* Make sure we own this device */ 221 return (EBUSY); 222 *pptp = ppt; 223 return (0); 224 } 225 226 static void 227 ppt_unmap_all_mmio(struct vm *vm, struct pptdev *ppt) 228 { 229 int i; 230 struct pptseg *seg; 231 232 for (i = 0; i < MAX_MMIOSEGS; i++) { 233 seg = &ppt->mmio[i]; 234 if (seg->len == 0) 235 continue; 236 (void)vm_unmap_mmio(vm, seg->gpa, seg->len); 237 bzero(seg, sizeof(struct pptseg)); 238 } 239 } 240 241 static void 242 ppt_teardown_msi(struct pptdev *ppt) 243 { 244 int i, rid; 245 void *cookie; 246 struct resource *res; 247 248 if (ppt->msi.num_msgs == 0) 249 return; 250 251 for (i = 0; i < ppt->msi.num_msgs; i++) { 252 rid = ppt->msi.startrid + i; 253 res = ppt->msi.res[i]; 254 cookie = ppt->msi.cookie[i]; 255 256 if (cookie != NULL) 257 bus_teardown_intr(ppt->dev, res, cookie); 258 259 if (res != NULL) 260 bus_release_resource(ppt->dev, SYS_RES_IRQ, rid, res); 261 262 ppt->msi.res[i] = NULL; 263 ppt->msi.cookie[i] = NULL; 264 } 265 266 if (ppt->msi.startrid == 1) 267 pci_release_msi(ppt->dev); 268 269 ppt->msi.num_msgs = 0; 270 } 271 272 static void 273 ppt_teardown_msix_intr(struct pptdev *ppt, int idx) 274 { 275 int rid; 276 struct resource *res; 277 void *cookie; 278 279 rid = ppt->msix.startrid + idx; 280 res = ppt->msix.res[idx]; 281 cookie = ppt->msix.cookie[idx]; 282 283 if (cookie != NULL) 284 bus_teardown_intr(ppt->dev, res, cookie); 285 286 if (res != NULL) 287 bus_release_resource(ppt->dev, SYS_RES_IRQ, rid, res); 288 289 ppt->msix.res[idx] = NULL; 290 ppt->msix.cookie[idx] = NULL; 291 } 292 293 static void 294 ppt_teardown_msix(struct pptdev *ppt) 295 { 296 int i; 297 298 if (ppt->msix.num_msgs == 0) 299 return; 300 301 for (i = 0; i < ppt->msix.num_msgs; i++) 302 ppt_teardown_msix_intr(ppt, i); 303 304 free(ppt->msix.res, M_PPTMSIX); 305 free(ppt->msix.cookie, M_PPTMSIX); 306 free(ppt->msix.arg, M_PPTMSIX); 307 308 pci_release_msi(ppt->dev); 309 310 if (ppt->msix.msix_table_res) { 311 bus_release_resource(ppt->dev, SYS_RES_MEMORY, 312 ppt->msix.msix_table_rid, 313 ppt->msix.msix_table_res); 314 ppt->msix.msix_table_res = NULL; 315 ppt->msix.msix_table_rid = 0; 316 } 317 if (ppt->msix.msix_pba_res) { 318 bus_release_resource(ppt->dev, SYS_RES_MEMORY, 319 ppt->msix.msix_pba_rid, 320 ppt->msix.msix_pba_res); 321 ppt->msix.msix_pba_res = NULL; 322 ppt->msix.msix_pba_rid = 0; 323 } 324 325 ppt->msix.num_msgs = 0; 326 } 327 328 int 329 ppt_avail_devices(void) 330 { 331 332 return (num_pptdevs); 333 } 334 335 int 336 ppt_assigned_devices(struct vm *vm) 337 { 338 struct pptdev *ppt; 339 int num; 340 341 num = 0; 342 TAILQ_FOREACH(ppt, &pptdev_list, next) { 343 if (ppt->vm == vm) 344 num++; 345 } 346 return (num); 347 } 348 349 bool 350 ppt_is_mmio(struct vm *vm, vm_paddr_t gpa) 351 { 352 int i; 353 struct pptdev *ppt; 354 struct pptseg *seg; 355 356 TAILQ_FOREACH(ppt, &pptdev_list, next) { 357 if (ppt->vm != vm) 358 continue; 359 360 for (i = 0; i < MAX_MMIOSEGS; i++) { 361 seg = &ppt->mmio[i]; 362 if (seg->len == 0) 363 continue; 364 if (gpa >= seg->gpa && gpa < seg->gpa + seg->len) 365 return (true); 366 } 367 } 368 369 return (false); 370 } 371 372 static void 373 ppt_pci_reset(device_t dev) 374 { 375 376 if (pcie_flr(dev, 377 max(pcie_get_max_completion_timeout(dev) / 1000, 10), true)) 378 return; 379 380 pci_power_reset(dev); 381 } 382 383 int 384 ppt_assign_device(struct vm *vm, int bus, int slot, int func) 385 { 386 struct pptdev *ppt; 387 int error; 388 389 /* Passing NULL requires the device to be unowned. */ 390 error = ppt_find(NULL, bus, slot, func, &ppt); 391 if (error) 392 return (error); 393 394 pci_save_state(ppt->dev); 395 ppt_pci_reset(ppt->dev); 396 pci_restore_state(ppt->dev); 397 ppt->vm = vm; 398 iommu_add_device(vm_iommu_domain(vm), pci_get_rid(ppt->dev)); 399 return (0); 400 } 401 402 int 403 ppt_unassign_device(struct vm *vm, int bus, int slot, int func) 404 { 405 struct pptdev *ppt; 406 int error; 407 408 error = ppt_find(vm, bus, slot, func, &ppt); 409 if (error) 410 return (error); 411 412 pci_save_state(ppt->dev); 413 ppt_pci_reset(ppt->dev); 414 pci_restore_state(ppt->dev); 415 ppt_unmap_all_mmio(vm, ppt); 416 ppt_teardown_msi(ppt); 417 ppt_teardown_msix(ppt); 418 iommu_remove_device(vm_iommu_domain(vm), pci_get_rid(ppt->dev)); 419 ppt->vm = NULL; 420 return (0); 421 } 422 423 int 424 ppt_unassign_all(struct vm *vm) 425 { 426 struct pptdev *ppt; 427 int bus, slot, func; 428 device_t dev; 429 430 TAILQ_FOREACH(ppt, &pptdev_list, next) { 431 if (ppt->vm == vm) { 432 dev = ppt->dev; 433 bus = pci_get_bus(dev); 434 slot = pci_get_slot(dev); 435 func = pci_get_function(dev); 436 vm_unassign_pptdev(vm, bus, slot, func); 437 } 438 } 439 440 return (0); 441 } 442 443 int 444 ppt_map_mmio(struct vm *vm, int bus, int slot, int func, 445 vm_paddr_t gpa, size_t len, vm_paddr_t hpa) 446 { 447 int i, error; 448 struct pptseg *seg; 449 struct pptdev *ppt; 450 451 error = ppt_find(vm, bus, slot, func, &ppt); 452 if (error) 453 return (error); 454 455 for (i = 0; i < MAX_MMIOSEGS; i++) { 456 seg = &ppt->mmio[i]; 457 if (seg->len == 0) { 458 error = vm_map_mmio(vm, gpa, len, hpa); 459 if (error == 0) { 460 seg->gpa = gpa; 461 seg->len = len; 462 } 463 return (error); 464 } 465 } 466 return (ENOSPC); 467 } 468 469 int 470 ppt_unmap_mmio(struct vm *vm, int bus, int slot, int func, 471 vm_paddr_t gpa, size_t len) 472 { 473 int i, error; 474 struct pptseg *seg; 475 struct pptdev *ppt; 476 477 error = ppt_find(vm, bus, slot, func, &ppt); 478 if (error) 479 return (error); 480 481 for (i = 0; i < MAX_MMIOSEGS; i++) { 482 seg = &ppt->mmio[i]; 483 if (seg->gpa == gpa && seg->len == len) { 484 error = vm_unmap_mmio(vm, seg->gpa, seg->len); 485 if (error == 0) { 486 seg->gpa = 0; 487 seg->len = 0; 488 } 489 return (error); 490 } 491 } 492 return (ENOENT); 493 } 494 495 static int 496 pptintr(void *arg) 497 { 498 struct pptdev *ppt; 499 struct pptintr_arg *pptarg; 500 501 pptarg = arg; 502 ppt = pptarg->pptdev; 503 504 if (ppt->vm != NULL) 505 lapic_intr_msi(ppt->vm, pptarg->addr, pptarg->msg_data); 506 else { 507 /* 508 * XXX 509 * This is not expected to happen - panic? 510 */ 511 } 512 513 /* 514 * For legacy interrupts give other filters a chance in case 515 * the interrupt was not generated by the passthrough device. 516 */ 517 if (ppt->msi.startrid == 0) 518 return (FILTER_STRAY); 519 else 520 return (FILTER_HANDLED); 521 } 522 523 int 524 ppt_setup_msi(struct vm *vm, int vcpu, int bus, int slot, int func, 525 uint64_t addr, uint64_t msg, int numvec) 526 { 527 int i, rid, flags; 528 int msi_count, startrid, error, tmp; 529 struct pptdev *ppt; 530 531 if (numvec < 0 || numvec > MAX_MSIMSGS) 532 return (EINVAL); 533 534 error = ppt_find(vm, bus, slot, func, &ppt); 535 if (error) 536 return (error); 537 538 /* Reject attempts to enable MSI while MSI-X is active. */ 539 if (ppt->msix.num_msgs != 0 && numvec != 0) 540 return (EBUSY); 541 542 /* Free any allocated resources */ 543 ppt_teardown_msi(ppt); 544 545 if (numvec == 0) /* nothing more to do */ 546 return (0); 547 548 flags = RF_ACTIVE; 549 msi_count = pci_msi_count(ppt->dev); 550 if (msi_count == 0) { 551 startrid = 0; /* legacy interrupt */ 552 msi_count = 1; 553 flags |= RF_SHAREABLE; 554 } else 555 startrid = 1; /* MSI */ 556 557 /* 558 * The device must be capable of supporting the number of vectors 559 * the guest wants to allocate. 560 */ 561 if (numvec > msi_count) 562 return (EINVAL); 563 564 /* 565 * Make sure that we can allocate all the MSI vectors that are needed 566 * by the guest. 567 */ 568 if (startrid == 1) { 569 tmp = numvec; 570 error = pci_alloc_msi(ppt->dev, &tmp); 571 if (error) 572 return (error); 573 else if (tmp != numvec) { 574 pci_release_msi(ppt->dev); 575 return (ENOSPC); 576 } else { 577 /* success */ 578 } 579 } 580 581 ppt->msi.startrid = startrid; 582 583 /* 584 * Allocate the irq resource and attach it to the interrupt handler. 585 */ 586 for (i = 0; i < numvec; i++) { 587 ppt->msi.num_msgs = i + 1; 588 ppt->msi.cookie[i] = NULL; 589 590 rid = startrid + i; 591 ppt->msi.res[i] = bus_alloc_resource_any(ppt->dev, SYS_RES_IRQ, 592 &rid, flags); 593 if (ppt->msi.res[i] == NULL) 594 break; 595 596 ppt->msi.arg[i].pptdev = ppt; 597 ppt->msi.arg[i].addr = addr; 598 ppt->msi.arg[i].msg_data = msg + i; 599 600 error = bus_setup_intr(ppt->dev, ppt->msi.res[i], 601 INTR_TYPE_NET | INTR_MPSAFE, 602 pptintr, NULL, &ppt->msi.arg[i], 603 &ppt->msi.cookie[i]); 604 if (error != 0) 605 break; 606 } 607 608 if (i < numvec) { 609 ppt_teardown_msi(ppt); 610 return (ENXIO); 611 } 612 613 return (0); 614 } 615 616 int 617 ppt_setup_msix(struct vm *vm, int vcpu, int bus, int slot, int func, 618 int idx, uint64_t addr, uint64_t msg, uint32_t vector_control) 619 { 620 struct pptdev *ppt; 621 struct pci_devinfo *dinfo; 622 int numvec, alloced, rid, error; 623 size_t res_size, cookie_size, arg_size; 624 625 error = ppt_find(vm, bus, slot, func, &ppt); 626 if (error) 627 return (error); 628 629 /* Reject attempts to enable MSI-X while MSI is active. */ 630 if (ppt->msi.num_msgs != 0) 631 return (EBUSY); 632 633 dinfo = device_get_ivars(ppt->dev); 634 if (!dinfo) 635 return (ENXIO); 636 637 /* 638 * First-time configuration: 639 * Allocate the MSI-X table 640 * Allocate the IRQ resources 641 * Set up some variables in ppt->msix 642 */ 643 if (ppt->msix.num_msgs == 0) { 644 numvec = pci_msix_count(ppt->dev); 645 if (numvec <= 0) 646 return (EINVAL); 647 648 ppt->msix.startrid = 1; 649 ppt->msix.num_msgs = numvec; 650 651 res_size = numvec * sizeof(ppt->msix.res[0]); 652 cookie_size = numvec * sizeof(ppt->msix.cookie[0]); 653 arg_size = numvec * sizeof(ppt->msix.arg[0]); 654 655 ppt->msix.res = malloc(res_size, M_PPTMSIX, M_WAITOK | M_ZERO); 656 ppt->msix.cookie = malloc(cookie_size, M_PPTMSIX, 657 M_WAITOK | M_ZERO); 658 ppt->msix.arg = malloc(arg_size, M_PPTMSIX, M_WAITOK | M_ZERO); 659 660 rid = dinfo->cfg.msix.msix_table_bar; 661 ppt->msix.msix_table_res = bus_alloc_resource_any(ppt->dev, 662 SYS_RES_MEMORY, &rid, RF_ACTIVE); 663 664 if (ppt->msix.msix_table_res == NULL) { 665 ppt_teardown_msix(ppt); 666 return (ENOSPC); 667 } 668 ppt->msix.msix_table_rid = rid; 669 670 if (dinfo->cfg.msix.msix_table_bar != 671 dinfo->cfg.msix.msix_pba_bar) { 672 rid = dinfo->cfg.msix.msix_pba_bar; 673 ppt->msix.msix_pba_res = bus_alloc_resource_any( 674 ppt->dev, SYS_RES_MEMORY, &rid, RF_ACTIVE); 675 676 if (ppt->msix.msix_pba_res == NULL) { 677 ppt_teardown_msix(ppt); 678 return (ENOSPC); 679 } 680 ppt->msix.msix_pba_rid = rid; 681 } 682 683 alloced = numvec; 684 error = pci_alloc_msix(ppt->dev, &alloced); 685 if (error || alloced != numvec) { 686 ppt_teardown_msix(ppt); 687 return (error == 0 ? ENOSPC: error); 688 } 689 } 690 691 if ((vector_control & PCIM_MSIX_VCTRL_MASK) == 0) { 692 /* Tear down the IRQ if it's already set up */ 693 ppt_teardown_msix_intr(ppt, idx); 694 695 /* Allocate the IRQ resource */ 696 ppt->msix.cookie[idx] = NULL; 697 rid = ppt->msix.startrid + idx; 698 ppt->msix.res[idx] = bus_alloc_resource_any(ppt->dev, SYS_RES_IRQ, 699 &rid, RF_ACTIVE); 700 if (ppt->msix.res[idx] == NULL) 701 return (ENXIO); 702 703 ppt->msix.arg[idx].pptdev = ppt; 704 ppt->msix.arg[idx].addr = addr; 705 ppt->msix.arg[idx].msg_data = msg; 706 707 /* Setup the MSI-X interrupt */ 708 error = bus_setup_intr(ppt->dev, ppt->msix.res[idx], 709 INTR_TYPE_NET | INTR_MPSAFE, 710 pptintr, NULL, &ppt->msix.arg[idx], 711 &ppt->msix.cookie[idx]); 712 713 if (error != 0) { 714 bus_release_resource(ppt->dev, SYS_RES_IRQ, rid, ppt->msix.res[idx]); 715 ppt->msix.cookie[idx] = NULL; 716 ppt->msix.res[idx] = NULL; 717 return (ENXIO); 718 } 719 } else { 720 /* Masked, tear it down if it's already been set up */ 721 ppt_teardown_msix_intr(ppt, idx); 722 } 723 724 return (0); 725 } 726 727 int 728 ppt_disable_msix(struct vm *vm, int bus, int slot, int func) 729 { 730 struct pptdev *ppt; 731 int error; 732 733 error = ppt_find(vm, bus, slot, func, &ppt); 734 if (error) 735 return (error); 736 737 ppt_teardown_msix(ppt); 738 return (0); 739 } 740