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 DEFINE_CLASS_0(ppt, ppt_driver, ppt_methods, sizeof(struct pptdev)); 199 DRIVER_MODULE(ppt, pci, ppt_driver, NULL, NULL); 200 201 static int 202 ppt_find(struct vm *vm, int bus, int slot, int func, struct pptdev **pptp) 203 { 204 device_t dev; 205 struct pptdev *ppt; 206 int b, s, f; 207 208 TAILQ_FOREACH(ppt, &pptdev_list, next) { 209 dev = ppt->dev; 210 b = pci_get_bus(dev); 211 s = pci_get_slot(dev); 212 f = pci_get_function(dev); 213 if (bus == b && slot == s && func == f) 214 break; 215 } 216 217 if (ppt == NULL) 218 return (ENOENT); 219 if (ppt->vm != vm) /* Make sure we own this device */ 220 return (EBUSY); 221 *pptp = ppt; 222 return (0); 223 } 224 225 static void 226 ppt_unmap_all_mmio(struct vm *vm, struct pptdev *ppt) 227 { 228 int i; 229 struct pptseg *seg; 230 231 for (i = 0; i < MAX_MMIOSEGS; i++) { 232 seg = &ppt->mmio[i]; 233 if (seg->len == 0) 234 continue; 235 (void)vm_unmap_mmio(vm, seg->gpa, seg->len); 236 bzero(seg, sizeof(struct pptseg)); 237 } 238 } 239 240 static void 241 ppt_teardown_msi(struct pptdev *ppt) 242 { 243 int i, rid; 244 void *cookie; 245 struct resource *res; 246 247 if (ppt->msi.num_msgs == 0) 248 return; 249 250 for (i = 0; i < ppt->msi.num_msgs; i++) { 251 rid = ppt->msi.startrid + i; 252 res = ppt->msi.res[i]; 253 cookie = ppt->msi.cookie[i]; 254 255 if (cookie != NULL) 256 bus_teardown_intr(ppt->dev, res, cookie); 257 258 if (res != NULL) 259 bus_release_resource(ppt->dev, SYS_RES_IRQ, rid, res); 260 261 ppt->msi.res[i] = NULL; 262 ppt->msi.cookie[i] = NULL; 263 } 264 265 if (ppt->msi.startrid == 1) 266 pci_release_msi(ppt->dev); 267 268 ppt->msi.num_msgs = 0; 269 } 270 271 static void 272 ppt_teardown_msix_intr(struct pptdev *ppt, int idx) 273 { 274 int rid; 275 struct resource *res; 276 void *cookie; 277 278 rid = ppt->msix.startrid + idx; 279 res = ppt->msix.res[idx]; 280 cookie = ppt->msix.cookie[idx]; 281 282 if (cookie != NULL) 283 bus_teardown_intr(ppt->dev, res, cookie); 284 285 if (res != NULL) 286 bus_release_resource(ppt->dev, SYS_RES_IRQ, rid, res); 287 288 ppt->msix.res[idx] = NULL; 289 ppt->msix.cookie[idx] = NULL; 290 } 291 292 static void 293 ppt_teardown_msix(struct pptdev *ppt) 294 { 295 int i; 296 297 if (ppt->msix.num_msgs == 0) 298 return; 299 300 for (i = 0; i < ppt->msix.num_msgs; i++) 301 ppt_teardown_msix_intr(ppt, i); 302 303 free(ppt->msix.res, M_PPTMSIX); 304 free(ppt->msix.cookie, M_PPTMSIX); 305 free(ppt->msix.arg, M_PPTMSIX); 306 307 pci_release_msi(ppt->dev); 308 309 if (ppt->msix.msix_table_res) { 310 bus_release_resource(ppt->dev, SYS_RES_MEMORY, 311 ppt->msix.msix_table_rid, 312 ppt->msix.msix_table_res); 313 ppt->msix.msix_table_res = NULL; 314 ppt->msix.msix_table_rid = 0; 315 } 316 if (ppt->msix.msix_pba_res) { 317 bus_release_resource(ppt->dev, SYS_RES_MEMORY, 318 ppt->msix.msix_pba_rid, 319 ppt->msix.msix_pba_res); 320 ppt->msix.msix_pba_res = NULL; 321 ppt->msix.msix_pba_rid = 0; 322 } 323 324 ppt->msix.num_msgs = 0; 325 } 326 327 int 328 ppt_avail_devices(void) 329 { 330 331 return (num_pptdevs); 332 } 333 334 int 335 ppt_assigned_devices(struct vm *vm) 336 { 337 struct pptdev *ppt; 338 int num; 339 340 num = 0; 341 TAILQ_FOREACH(ppt, &pptdev_list, next) { 342 if (ppt->vm == vm) 343 num++; 344 } 345 return (num); 346 } 347 348 bool 349 ppt_is_mmio(struct vm *vm, vm_paddr_t gpa) 350 { 351 int i; 352 struct pptdev *ppt; 353 struct pptseg *seg; 354 355 TAILQ_FOREACH(ppt, &pptdev_list, next) { 356 if (ppt->vm != vm) 357 continue; 358 359 for (i = 0; i < MAX_MMIOSEGS; i++) { 360 seg = &ppt->mmio[i]; 361 if (seg->len == 0) 362 continue; 363 if (gpa >= seg->gpa && gpa < seg->gpa + seg->len) 364 return (true); 365 } 366 } 367 368 return (false); 369 } 370 371 static void 372 ppt_pci_reset(device_t dev) 373 { 374 375 if (pcie_flr(dev, 376 max(pcie_get_max_completion_timeout(dev) / 1000, 10), true)) 377 return; 378 379 pci_power_reset(dev); 380 } 381 382 int 383 ppt_assign_device(struct vm *vm, int bus, int slot, int func) 384 { 385 struct pptdev *ppt; 386 int error; 387 388 /* Passing NULL requires the device to be unowned. */ 389 error = ppt_find(NULL, bus, slot, func, &ppt); 390 if (error) 391 return (error); 392 393 pci_save_state(ppt->dev); 394 ppt_pci_reset(ppt->dev); 395 pci_restore_state(ppt->dev); 396 ppt->vm = vm; 397 iommu_add_device(vm_iommu_domain(vm), pci_get_rid(ppt->dev)); 398 return (0); 399 } 400 401 int 402 ppt_unassign_device(struct vm *vm, int bus, int slot, int func) 403 { 404 struct pptdev *ppt; 405 int error; 406 407 error = ppt_find(vm, bus, slot, func, &ppt); 408 if (error) 409 return (error); 410 411 pci_save_state(ppt->dev); 412 ppt_pci_reset(ppt->dev); 413 pci_restore_state(ppt->dev); 414 ppt_unmap_all_mmio(vm, ppt); 415 ppt_teardown_msi(ppt); 416 ppt_teardown_msix(ppt); 417 iommu_remove_device(vm_iommu_domain(vm), pci_get_rid(ppt->dev)); 418 ppt->vm = NULL; 419 return (0); 420 } 421 422 int 423 ppt_unassign_all(struct vm *vm) 424 { 425 struct pptdev *ppt; 426 int bus, slot, func; 427 device_t dev; 428 429 TAILQ_FOREACH(ppt, &pptdev_list, next) { 430 if (ppt->vm == vm) { 431 dev = ppt->dev; 432 bus = pci_get_bus(dev); 433 slot = pci_get_slot(dev); 434 func = pci_get_function(dev); 435 vm_unassign_pptdev(vm, bus, slot, func); 436 } 437 } 438 439 return (0); 440 } 441 442 int 443 ppt_map_mmio(struct vm *vm, int bus, int slot, int func, 444 vm_paddr_t gpa, size_t len, vm_paddr_t hpa) 445 { 446 int i, error; 447 struct pptseg *seg; 448 struct pptdev *ppt; 449 450 error = ppt_find(vm, bus, slot, func, &ppt); 451 if (error) 452 return (error); 453 454 for (i = 0; i < MAX_MMIOSEGS; i++) { 455 seg = &ppt->mmio[i]; 456 if (seg->len == 0) { 457 error = vm_map_mmio(vm, gpa, len, hpa); 458 if (error == 0) { 459 seg->gpa = gpa; 460 seg->len = len; 461 } 462 return (error); 463 } 464 } 465 return (ENOSPC); 466 } 467 468 int 469 ppt_unmap_mmio(struct vm *vm, int bus, int slot, int func, 470 vm_paddr_t gpa, size_t len) 471 { 472 int i, error; 473 struct pptseg *seg; 474 struct pptdev *ppt; 475 476 error = ppt_find(vm, bus, slot, func, &ppt); 477 if (error) 478 return (error); 479 480 for (i = 0; i < MAX_MMIOSEGS; i++) { 481 seg = &ppt->mmio[i]; 482 if (seg->gpa == gpa && seg->len == len) { 483 error = vm_unmap_mmio(vm, seg->gpa, seg->len); 484 if (error == 0) { 485 seg->gpa = 0; 486 seg->len = 0; 487 } 488 return (error); 489 } 490 } 491 return (ENOENT); 492 } 493 494 static int 495 pptintr(void *arg) 496 { 497 struct pptdev *ppt; 498 struct pptintr_arg *pptarg; 499 500 pptarg = arg; 501 ppt = pptarg->pptdev; 502 503 if (ppt->vm != NULL) 504 lapic_intr_msi(ppt->vm, pptarg->addr, pptarg->msg_data); 505 else { 506 /* 507 * XXX 508 * This is not expected to happen - panic? 509 */ 510 } 511 512 /* 513 * For legacy interrupts give other filters a chance in case 514 * the interrupt was not generated by the passthrough device. 515 */ 516 if (ppt->msi.startrid == 0) 517 return (FILTER_STRAY); 518 else 519 return (FILTER_HANDLED); 520 } 521 522 int 523 ppt_setup_msi(struct vm *vm, int vcpu, int bus, int slot, int func, 524 uint64_t addr, uint64_t msg, int numvec) 525 { 526 int i, rid, flags; 527 int msi_count, startrid, error, tmp; 528 struct pptdev *ppt; 529 530 if (numvec < 0 || numvec > MAX_MSIMSGS) 531 return (EINVAL); 532 533 error = ppt_find(vm, bus, slot, func, &ppt); 534 if (error) 535 return (error); 536 537 /* Reject attempts to enable MSI while MSI-X is active. */ 538 if (ppt->msix.num_msgs != 0 && numvec != 0) 539 return (EBUSY); 540 541 /* Free any allocated resources */ 542 ppt_teardown_msi(ppt); 543 544 if (numvec == 0) /* nothing more to do */ 545 return (0); 546 547 flags = RF_ACTIVE; 548 msi_count = pci_msi_count(ppt->dev); 549 if (msi_count == 0) { 550 startrid = 0; /* legacy interrupt */ 551 msi_count = 1; 552 flags |= RF_SHAREABLE; 553 } else 554 startrid = 1; /* MSI */ 555 556 /* 557 * The device must be capable of supporting the number of vectors 558 * the guest wants to allocate. 559 */ 560 if (numvec > msi_count) 561 return (EINVAL); 562 563 /* 564 * Make sure that we can allocate all the MSI vectors that are needed 565 * by the guest. 566 */ 567 if (startrid == 1) { 568 tmp = numvec; 569 error = pci_alloc_msi(ppt->dev, &tmp); 570 if (error) 571 return (error); 572 else if (tmp != numvec) { 573 pci_release_msi(ppt->dev); 574 return (ENOSPC); 575 } else { 576 /* success */ 577 } 578 } 579 580 ppt->msi.startrid = startrid; 581 582 /* 583 * Allocate the irq resource and attach it to the interrupt handler. 584 */ 585 for (i = 0; i < numvec; i++) { 586 ppt->msi.num_msgs = i + 1; 587 ppt->msi.cookie[i] = NULL; 588 589 rid = startrid + i; 590 ppt->msi.res[i] = bus_alloc_resource_any(ppt->dev, SYS_RES_IRQ, 591 &rid, flags); 592 if (ppt->msi.res[i] == NULL) 593 break; 594 595 ppt->msi.arg[i].pptdev = ppt; 596 ppt->msi.arg[i].addr = addr; 597 ppt->msi.arg[i].msg_data = msg + i; 598 599 error = bus_setup_intr(ppt->dev, ppt->msi.res[i], 600 INTR_TYPE_NET | INTR_MPSAFE, 601 pptintr, NULL, &ppt->msi.arg[i], 602 &ppt->msi.cookie[i]); 603 if (error != 0) 604 break; 605 } 606 607 if (i < numvec) { 608 ppt_teardown_msi(ppt); 609 return (ENXIO); 610 } 611 612 return (0); 613 } 614 615 int 616 ppt_setup_msix(struct vm *vm, int vcpu, int bus, int slot, int func, 617 int idx, uint64_t addr, uint64_t msg, uint32_t vector_control) 618 { 619 struct pptdev *ppt; 620 struct pci_devinfo *dinfo; 621 int numvec, alloced, rid, error; 622 size_t res_size, cookie_size, arg_size; 623 624 error = ppt_find(vm, bus, slot, func, &ppt); 625 if (error) 626 return (error); 627 628 /* Reject attempts to enable MSI-X while MSI is active. */ 629 if (ppt->msi.num_msgs != 0) 630 return (EBUSY); 631 632 dinfo = device_get_ivars(ppt->dev); 633 if (!dinfo) 634 return (ENXIO); 635 636 /* 637 * First-time configuration: 638 * Allocate the MSI-X table 639 * Allocate the IRQ resources 640 * Set up some variables in ppt->msix 641 */ 642 if (ppt->msix.num_msgs == 0) { 643 numvec = pci_msix_count(ppt->dev); 644 if (numvec <= 0) 645 return (EINVAL); 646 647 ppt->msix.startrid = 1; 648 ppt->msix.num_msgs = numvec; 649 650 res_size = numvec * sizeof(ppt->msix.res[0]); 651 cookie_size = numvec * sizeof(ppt->msix.cookie[0]); 652 arg_size = numvec * sizeof(ppt->msix.arg[0]); 653 654 ppt->msix.res = malloc(res_size, M_PPTMSIX, M_WAITOK | M_ZERO); 655 ppt->msix.cookie = malloc(cookie_size, M_PPTMSIX, 656 M_WAITOK | M_ZERO); 657 ppt->msix.arg = malloc(arg_size, M_PPTMSIX, M_WAITOK | M_ZERO); 658 659 rid = dinfo->cfg.msix.msix_table_bar; 660 ppt->msix.msix_table_res = bus_alloc_resource_any(ppt->dev, 661 SYS_RES_MEMORY, &rid, RF_ACTIVE); 662 663 if (ppt->msix.msix_table_res == NULL) { 664 ppt_teardown_msix(ppt); 665 return (ENOSPC); 666 } 667 ppt->msix.msix_table_rid = rid; 668 669 if (dinfo->cfg.msix.msix_table_bar != 670 dinfo->cfg.msix.msix_pba_bar) { 671 rid = dinfo->cfg.msix.msix_pba_bar; 672 ppt->msix.msix_pba_res = bus_alloc_resource_any( 673 ppt->dev, SYS_RES_MEMORY, &rid, RF_ACTIVE); 674 675 if (ppt->msix.msix_pba_res == NULL) { 676 ppt_teardown_msix(ppt); 677 return (ENOSPC); 678 } 679 ppt->msix.msix_pba_rid = rid; 680 } 681 682 alloced = numvec; 683 error = pci_alloc_msix(ppt->dev, &alloced); 684 if (error || alloced != numvec) { 685 ppt_teardown_msix(ppt); 686 return (error == 0 ? ENOSPC: error); 687 } 688 } 689 690 if ((vector_control & PCIM_MSIX_VCTRL_MASK) == 0) { 691 /* Tear down the IRQ if it's already set up */ 692 ppt_teardown_msix_intr(ppt, idx); 693 694 /* Allocate the IRQ resource */ 695 ppt->msix.cookie[idx] = NULL; 696 rid = ppt->msix.startrid + idx; 697 ppt->msix.res[idx] = bus_alloc_resource_any(ppt->dev, SYS_RES_IRQ, 698 &rid, RF_ACTIVE); 699 if (ppt->msix.res[idx] == NULL) 700 return (ENXIO); 701 702 ppt->msix.arg[idx].pptdev = ppt; 703 ppt->msix.arg[idx].addr = addr; 704 ppt->msix.arg[idx].msg_data = msg; 705 706 /* Setup the MSI-X interrupt */ 707 error = bus_setup_intr(ppt->dev, ppt->msix.res[idx], 708 INTR_TYPE_NET | INTR_MPSAFE, 709 pptintr, NULL, &ppt->msix.arg[idx], 710 &ppt->msix.cookie[idx]); 711 712 if (error != 0) { 713 bus_release_resource(ppt->dev, SYS_RES_IRQ, rid, ppt->msix.res[idx]); 714 ppt->msix.cookie[idx] = NULL; 715 ppt->msix.res[idx] = NULL; 716 return (ENXIO); 717 } 718 } else { 719 /* Masked, tear it down if it's already been set up */ 720 ppt_teardown_msix_intr(ppt, idx); 721 } 722 723 return (0); 724 } 725 726 int 727 ppt_disable_msix(struct vm *vm, int bus, int slot, int func) 728 { 729 struct pptdev *ppt; 730 int error; 731 732 error = ppt_find(vm, bus, slot, func, &ppt); 733 if (error) 734 return (error); 735 736 ppt_teardown_msix(ppt); 737 return (0); 738 } 739