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