1 /*- 2 * Copyright (c) 2011 NetApp, Inc. 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 NETAPP, INC ``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 NETAPP, INC 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 * $FreeBSD$ 27 */ 28 29 #include <sys/cdefs.h> 30 __FBSDID("$FreeBSD$"); 31 32 #include <sys/param.h> 33 #include <sys/systm.h> 34 #include <sys/kernel.h> 35 #include <sys/malloc.h> 36 #include <sys/module.h> 37 #include <sys/bus.h> 38 #include <sys/pciio.h> 39 #include <sys/rman.h> 40 #include <sys/smp.h> 41 #include <sys/sysctl.h> 42 43 #include <dev/pci/pcivar.h> 44 #include <dev/pci/pcireg.h> 45 46 #include <machine/resource.h> 47 48 #include <machine/vmm.h> 49 #include <machine/vmm_dev.h> 50 51 #include "vmm_lapic.h" 52 #include "vmm_ktr.h" 53 54 #include "iommu.h" 55 #include "ppt.h" 56 57 /* XXX locking */ 58 59 #define MAX_MSIMSGS 32 60 61 /* 62 * If the MSI-X table is located in the middle of a BAR then that MMIO 63 * region gets split into two segments - one segment above the MSI-X table 64 * and the other segment below the MSI-X table - with a hole in place of 65 * the MSI-X table so accesses to it can be trapped and emulated. 66 * 67 * So, allocate a MMIO segment for each BAR register + 1 additional segment. 68 */ 69 #define MAX_MMIOSEGS ((PCIR_MAX_BAR_0 + 1) + 1) 70 71 MALLOC_DEFINE(M_PPTMSIX, "pptmsix", "Passthru MSI-X resources"); 72 73 struct pptintr_arg { /* pptintr(pptintr_arg) */ 74 struct pptdev *pptdev; 75 uint64_t addr; 76 uint64_t msg_data; 77 }; 78 79 struct pptseg { 80 vm_paddr_t gpa; 81 size_t len; 82 int wired; 83 }; 84 85 struct pptdev { 86 device_t dev; 87 struct vm *vm; /* owner of this device */ 88 TAILQ_ENTRY(pptdev) next; 89 struct pptseg mmio[MAX_MMIOSEGS]; 90 struct { 91 int num_msgs; /* guest state */ 92 93 int startrid; /* host state */ 94 struct resource *res[MAX_MSIMSGS]; 95 void *cookie[MAX_MSIMSGS]; 96 struct pptintr_arg arg[MAX_MSIMSGS]; 97 } msi; 98 99 struct { 100 int num_msgs; 101 int startrid; 102 int msix_table_rid; 103 struct resource *msix_table_res; 104 struct resource **res; 105 void **cookie; 106 struct pptintr_arg *arg; 107 } msix; 108 }; 109 110 SYSCTL_DECL(_hw_vmm); 111 SYSCTL_NODE(_hw_vmm, OID_AUTO, ppt, CTLFLAG_RW, 0, "bhyve passthru devices"); 112 113 static int num_pptdevs; 114 SYSCTL_INT(_hw_vmm_ppt, OID_AUTO, devices, CTLFLAG_RD, &num_pptdevs, 0, 115 "number of pci passthru devices"); 116 117 static TAILQ_HEAD(, pptdev) pptdev_list = TAILQ_HEAD_INITIALIZER(pptdev_list); 118 119 static int 120 ppt_probe(device_t dev) 121 { 122 int bus, slot, func; 123 struct pci_devinfo *dinfo; 124 125 dinfo = (struct pci_devinfo *)device_get_ivars(dev); 126 127 bus = pci_get_bus(dev); 128 slot = pci_get_slot(dev); 129 func = pci_get_function(dev); 130 131 /* 132 * To qualify as a pci passthrough device a device must: 133 * - be allowed by administrator to be used in this role 134 * - be an endpoint device 135 */ 136 if ((dinfo->cfg.hdrtype & PCIM_HDRTYPE) != PCIM_HDRTYPE_NORMAL) 137 return (ENXIO); 138 else if (vmm_is_pptdev(bus, slot, func)) 139 return (0); 140 else 141 /* 142 * Returning BUS_PROBE_NOWILDCARD here matches devices that the 143 * SR-IOV infrastructure specified as "ppt" passthrough devices. 144 * All normal devices that did not have "ppt" specified as their 145 * driver will not be matched by this. 146 */ 147 return (BUS_PROBE_NOWILDCARD); 148 } 149 150 static int 151 ppt_attach(device_t dev) 152 { 153 struct pptdev *ppt; 154 155 ppt = device_get_softc(dev); 156 157 num_pptdevs++; 158 TAILQ_INSERT_TAIL(&pptdev_list, ppt, next); 159 ppt->dev = dev; 160 161 if (bootverbose) 162 device_printf(dev, "attached\n"); 163 164 return (0); 165 } 166 167 static int 168 ppt_detach(device_t dev) 169 { 170 struct pptdev *ppt; 171 172 ppt = device_get_softc(dev); 173 174 if (ppt->vm != NULL) 175 return (EBUSY); 176 num_pptdevs--; 177 TAILQ_REMOVE(&pptdev_list, ppt, next); 178 179 return (0); 180 } 181 182 static device_method_t ppt_methods[] = { 183 /* Device interface */ 184 DEVMETHOD(device_probe, ppt_probe), 185 DEVMETHOD(device_attach, ppt_attach), 186 DEVMETHOD(device_detach, ppt_detach), 187 {0, 0} 188 }; 189 190 static devclass_t ppt_devclass; 191 DEFINE_CLASS_0(ppt, ppt_driver, ppt_methods, sizeof(struct pptdev)); 192 DRIVER_MODULE(ppt, pci, ppt_driver, ppt_devclass, NULL, NULL); 193 194 static struct pptdev * 195 ppt_find(int bus, int slot, int func) 196 { 197 device_t dev; 198 struct pptdev *ppt; 199 int b, s, f; 200 201 TAILQ_FOREACH(ppt, &pptdev_list, next) { 202 dev = ppt->dev; 203 b = pci_get_bus(dev); 204 s = pci_get_slot(dev); 205 f = pci_get_function(dev); 206 if (bus == b && slot == s && func == f) 207 return (ppt); 208 } 209 return (NULL); 210 } 211 212 static void 213 ppt_unmap_mmio(struct vm *vm, struct pptdev *ppt) 214 { 215 int i; 216 struct pptseg *seg; 217 218 for (i = 0; i < MAX_MMIOSEGS; i++) { 219 seg = &ppt->mmio[i]; 220 if (seg->len == 0) 221 continue; 222 (void)vm_unmap_mmio(vm, seg->gpa, seg->len); 223 bzero(seg, sizeof(struct pptseg)); 224 } 225 } 226 227 static void 228 ppt_teardown_msi(struct pptdev *ppt) 229 { 230 int i, rid; 231 void *cookie; 232 struct resource *res; 233 234 if (ppt->msi.num_msgs == 0) 235 return; 236 237 for (i = 0; i < ppt->msi.num_msgs; i++) { 238 rid = ppt->msi.startrid + i; 239 res = ppt->msi.res[i]; 240 cookie = ppt->msi.cookie[i]; 241 242 if (cookie != NULL) 243 bus_teardown_intr(ppt->dev, res, cookie); 244 245 if (res != NULL) 246 bus_release_resource(ppt->dev, SYS_RES_IRQ, rid, res); 247 248 ppt->msi.res[i] = NULL; 249 ppt->msi.cookie[i] = NULL; 250 } 251 252 if (ppt->msi.startrid == 1) 253 pci_release_msi(ppt->dev); 254 255 ppt->msi.num_msgs = 0; 256 } 257 258 static void 259 ppt_teardown_msix_intr(struct pptdev *ppt, int idx) 260 { 261 int rid; 262 struct resource *res; 263 void *cookie; 264 265 rid = ppt->msix.startrid + idx; 266 res = ppt->msix.res[idx]; 267 cookie = ppt->msix.cookie[idx]; 268 269 if (cookie != NULL) 270 bus_teardown_intr(ppt->dev, res, cookie); 271 272 if (res != NULL) 273 bus_release_resource(ppt->dev, SYS_RES_IRQ, rid, res); 274 275 ppt->msix.res[idx] = NULL; 276 ppt->msix.cookie[idx] = NULL; 277 } 278 279 static void 280 ppt_teardown_msix(struct pptdev *ppt) 281 { 282 int i; 283 284 if (ppt->msix.num_msgs == 0) 285 return; 286 287 for (i = 0; i < ppt->msix.num_msgs; i++) 288 ppt_teardown_msix_intr(ppt, i); 289 290 if (ppt->msix.msix_table_res) { 291 bus_release_resource(ppt->dev, SYS_RES_MEMORY, 292 ppt->msix.msix_table_rid, 293 ppt->msix.msix_table_res); 294 ppt->msix.msix_table_res = NULL; 295 ppt->msix.msix_table_rid = 0; 296 } 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 ppt->msix.num_msgs = 0; 305 } 306 307 int 308 ppt_avail_devices(void) 309 { 310 311 return (num_pptdevs); 312 } 313 314 int 315 ppt_assigned_devices(struct vm *vm) 316 { 317 struct pptdev *ppt; 318 int num; 319 320 num = 0; 321 TAILQ_FOREACH(ppt, &pptdev_list, next) { 322 if (ppt->vm == vm) 323 num++; 324 } 325 return (num); 326 } 327 328 boolean_t 329 ppt_is_mmio(struct vm *vm, vm_paddr_t gpa) 330 { 331 int i; 332 struct pptdev *ppt; 333 struct pptseg *seg; 334 335 TAILQ_FOREACH(ppt, &pptdev_list, next) { 336 if (ppt->vm != vm) 337 continue; 338 339 for (i = 0; i < MAX_MMIOSEGS; i++) { 340 seg = &ppt->mmio[i]; 341 if (seg->len == 0) 342 continue; 343 if (gpa >= seg->gpa && gpa < seg->gpa + seg->len) 344 return (TRUE); 345 } 346 } 347 348 return (FALSE); 349 } 350 351 int 352 ppt_assign_device(struct vm *vm, int bus, int slot, int func) 353 { 354 struct pptdev *ppt; 355 356 ppt = ppt_find(bus, slot, func); 357 if (ppt != NULL) { 358 /* 359 * If this device is owned by a different VM then we 360 * cannot change its owner. 361 */ 362 if (ppt->vm != NULL && ppt->vm != vm) 363 return (EBUSY); 364 365 ppt->vm = vm; 366 iommu_remove_device(iommu_host_domain(), pci_get_rid(ppt->dev)); 367 iommu_add_device(vm_iommu_domain(vm), pci_get_rid(ppt->dev)); 368 return (0); 369 } 370 return (ENOENT); 371 } 372 373 int 374 ppt_unassign_device(struct vm *vm, int bus, int slot, int func) 375 { 376 struct pptdev *ppt; 377 378 ppt = ppt_find(bus, slot, func); 379 if (ppt != NULL) { 380 /* 381 * If this device is not owned by this 'vm' then bail out. 382 */ 383 if (ppt->vm != vm) 384 return (EBUSY); 385 ppt_unmap_mmio(vm, ppt); 386 ppt_teardown_msi(ppt); 387 ppt_teardown_msix(ppt); 388 iommu_remove_device(vm_iommu_domain(vm), pci_get_rid(ppt->dev)); 389 iommu_add_device(iommu_host_domain(), pci_get_rid(ppt->dev)); 390 ppt->vm = NULL; 391 return (0); 392 } 393 return (ENOENT); 394 } 395 396 int 397 ppt_unassign_all(struct vm *vm) 398 { 399 struct pptdev *ppt; 400 int bus, slot, func; 401 device_t dev; 402 403 TAILQ_FOREACH(ppt, &pptdev_list, next) { 404 if (ppt->vm == vm) { 405 dev = ppt->dev; 406 bus = pci_get_bus(dev); 407 slot = pci_get_slot(dev); 408 func = pci_get_function(dev); 409 vm_unassign_pptdev(vm, bus, slot, func); 410 } 411 } 412 413 return (0); 414 } 415 416 int 417 ppt_map_mmio(struct vm *vm, int bus, int slot, int func, 418 vm_paddr_t gpa, size_t len, vm_paddr_t hpa) 419 { 420 int i, error; 421 struct pptseg *seg; 422 struct pptdev *ppt; 423 424 ppt = ppt_find(bus, slot, func); 425 if (ppt != NULL) { 426 if (ppt->vm != vm) 427 return (EBUSY); 428 429 for (i = 0; i < MAX_MMIOSEGS; i++) { 430 seg = &ppt->mmio[i]; 431 if (seg->len == 0) { 432 error = vm_map_mmio(vm, gpa, len, hpa); 433 if (error == 0) { 434 seg->gpa = gpa; 435 seg->len = len; 436 } 437 return (error); 438 } 439 } 440 return (ENOSPC); 441 } 442 return (ENOENT); 443 } 444 445 static int 446 pptintr(void *arg) 447 { 448 struct pptdev *ppt; 449 struct pptintr_arg *pptarg; 450 451 pptarg = arg; 452 ppt = pptarg->pptdev; 453 454 if (ppt->vm != NULL) 455 lapic_intr_msi(ppt->vm, pptarg->addr, pptarg->msg_data); 456 else { 457 /* 458 * XXX 459 * This is not expected to happen - panic? 460 */ 461 } 462 463 /* 464 * For legacy interrupts give other filters a chance in case 465 * the interrupt was not generated by the passthrough device. 466 */ 467 if (ppt->msi.startrid == 0) 468 return (FILTER_STRAY); 469 else 470 return (FILTER_HANDLED); 471 } 472 473 int 474 ppt_setup_msi(struct vm *vm, int vcpu, int bus, int slot, int func, 475 uint64_t addr, uint64_t msg, int numvec) 476 { 477 int i, rid, flags; 478 int msi_count, startrid, error, tmp; 479 struct pptdev *ppt; 480 481 if (numvec < 0 || numvec > MAX_MSIMSGS) 482 return (EINVAL); 483 484 ppt = ppt_find(bus, slot, func); 485 if (ppt == NULL) 486 return (ENOENT); 487 if (ppt->vm != vm) /* Make sure we own this device */ 488 return (EBUSY); 489 490 /* Free any allocated resources */ 491 ppt_teardown_msi(ppt); 492 493 if (numvec == 0) /* nothing more to do */ 494 return (0); 495 496 flags = RF_ACTIVE; 497 msi_count = pci_msi_count(ppt->dev); 498 if (msi_count == 0) { 499 startrid = 0; /* legacy interrupt */ 500 msi_count = 1; 501 flags |= RF_SHAREABLE; 502 } else 503 startrid = 1; /* MSI */ 504 505 /* 506 * The device must be capable of supporting the number of vectors 507 * the guest wants to allocate. 508 */ 509 if (numvec > msi_count) 510 return (EINVAL); 511 512 /* 513 * Make sure that we can allocate all the MSI vectors that are needed 514 * by the guest. 515 */ 516 if (startrid == 1) { 517 tmp = numvec; 518 error = pci_alloc_msi(ppt->dev, &tmp); 519 if (error) 520 return (error); 521 else if (tmp != numvec) { 522 pci_release_msi(ppt->dev); 523 return (ENOSPC); 524 } else { 525 /* success */ 526 } 527 } 528 529 ppt->msi.startrid = startrid; 530 531 /* 532 * Allocate the irq resource and attach it to the interrupt handler. 533 */ 534 for (i = 0; i < numvec; i++) { 535 ppt->msi.num_msgs = i + 1; 536 ppt->msi.cookie[i] = NULL; 537 538 rid = startrid + i; 539 ppt->msi.res[i] = bus_alloc_resource_any(ppt->dev, SYS_RES_IRQ, 540 &rid, flags); 541 if (ppt->msi.res[i] == NULL) 542 break; 543 544 ppt->msi.arg[i].pptdev = ppt; 545 ppt->msi.arg[i].addr = addr; 546 ppt->msi.arg[i].msg_data = msg + i; 547 548 error = bus_setup_intr(ppt->dev, ppt->msi.res[i], 549 INTR_TYPE_NET | INTR_MPSAFE, 550 pptintr, NULL, &ppt->msi.arg[i], 551 &ppt->msi.cookie[i]); 552 if (error != 0) 553 break; 554 } 555 556 if (i < numvec) { 557 ppt_teardown_msi(ppt); 558 return (ENXIO); 559 } 560 561 return (0); 562 } 563 564 int 565 ppt_setup_msix(struct vm *vm, int vcpu, int bus, int slot, int func, 566 int idx, uint64_t addr, uint64_t msg, uint32_t vector_control) 567 { 568 struct pptdev *ppt; 569 struct pci_devinfo *dinfo; 570 int numvec, alloced, rid, error; 571 size_t res_size, cookie_size, arg_size; 572 573 ppt = ppt_find(bus, slot, func); 574 if (ppt == NULL) 575 return (ENOENT); 576 if (ppt->vm != vm) /* Make sure we own this device */ 577 return (EBUSY); 578 579 dinfo = device_get_ivars(ppt->dev); 580 if (!dinfo) 581 return (ENXIO); 582 583 /* 584 * First-time configuration: 585 * Allocate the MSI-X table 586 * Allocate the IRQ resources 587 * Set up some variables in ppt->msix 588 */ 589 if (ppt->msix.num_msgs == 0) { 590 numvec = pci_msix_count(ppt->dev); 591 if (numvec <= 0) 592 return (EINVAL); 593 594 ppt->msix.startrid = 1; 595 ppt->msix.num_msgs = numvec; 596 597 res_size = numvec * sizeof(ppt->msix.res[0]); 598 cookie_size = numvec * sizeof(ppt->msix.cookie[0]); 599 arg_size = numvec * sizeof(ppt->msix.arg[0]); 600 601 ppt->msix.res = malloc(res_size, M_PPTMSIX, M_WAITOK | M_ZERO); 602 ppt->msix.cookie = malloc(cookie_size, M_PPTMSIX, 603 M_WAITOK | M_ZERO); 604 ppt->msix.arg = malloc(arg_size, M_PPTMSIX, M_WAITOK | M_ZERO); 605 606 rid = dinfo->cfg.msix.msix_table_bar; 607 ppt->msix.msix_table_res = bus_alloc_resource_any(ppt->dev, 608 SYS_RES_MEMORY, &rid, RF_ACTIVE); 609 610 if (ppt->msix.msix_table_res == NULL) { 611 ppt_teardown_msix(ppt); 612 return (ENOSPC); 613 } 614 ppt->msix.msix_table_rid = rid; 615 616 alloced = numvec; 617 error = pci_alloc_msix(ppt->dev, &alloced); 618 if (error || alloced != numvec) { 619 ppt_teardown_msix(ppt); 620 return (error == 0 ? ENOSPC: error); 621 } 622 } 623 624 if ((vector_control & PCIM_MSIX_VCTRL_MASK) == 0) { 625 /* Tear down the IRQ if it's already set up */ 626 ppt_teardown_msix_intr(ppt, idx); 627 628 /* Allocate the IRQ resource */ 629 ppt->msix.cookie[idx] = NULL; 630 rid = ppt->msix.startrid + idx; 631 ppt->msix.res[idx] = bus_alloc_resource_any(ppt->dev, SYS_RES_IRQ, 632 &rid, RF_ACTIVE); 633 if (ppt->msix.res[idx] == NULL) 634 return (ENXIO); 635 636 ppt->msix.arg[idx].pptdev = ppt; 637 ppt->msix.arg[idx].addr = addr; 638 ppt->msix.arg[idx].msg_data = msg; 639 640 /* Setup the MSI-X interrupt */ 641 error = bus_setup_intr(ppt->dev, ppt->msix.res[idx], 642 INTR_TYPE_NET | INTR_MPSAFE, 643 pptintr, NULL, &ppt->msix.arg[idx], 644 &ppt->msix.cookie[idx]); 645 646 if (error != 0) { 647 bus_teardown_intr(ppt->dev, ppt->msix.res[idx], ppt->msix.cookie[idx]); 648 bus_release_resource(ppt->dev, SYS_RES_IRQ, rid, ppt->msix.res[idx]); 649 ppt->msix.cookie[idx] = NULL; 650 ppt->msix.res[idx] = NULL; 651 return (ENXIO); 652 } 653 } else { 654 /* Masked, tear it down if it's already been set up */ 655 ppt_teardown_msix_intr(ppt, idx); 656 } 657 658 return (0); 659 } 660