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/kernel.h> 36 #include <sys/systm.h> 37 #include <sys/kmem.h> 38 39 #include <dev/pci/pcireg.h> 40 41 #include <machine/vmparam.h> 42 #include <sys/vmm_vm.h> 43 44 #include <contrib/dev/acpica/include/acpi.h> 45 46 #include <sys/sunndi.h> 47 48 #include "io/iommu.h" 49 50 /* 51 * Documented in the "Intel Virtualization Technology for Directed I/O", 52 * Architecture Spec, September 2008. 53 */ 54 55 #define VTD_DRHD_INCLUDE_PCI_ALL(Flags) (((Flags) >> 0) & 0x1) 56 57 /* Section 10.4 "Register Descriptions" */ 58 struct vtdmap { 59 volatile uint32_t version; 60 volatile uint32_t res0; 61 volatile uint64_t cap; 62 volatile uint64_t ext_cap; 63 volatile uint32_t gcr; 64 volatile uint32_t gsr; 65 volatile uint64_t rta; 66 volatile uint64_t ccr; 67 }; 68 69 #define VTD_CAP_SAGAW(cap) (((cap) >> 8) & 0x1F) 70 #define VTD_CAP_ND(cap) ((cap) & 0x7) 71 #define VTD_CAP_CM(cap) (((cap) >> 7) & 0x1) 72 #define VTD_CAP_SPS(cap) (((cap) >> 34) & 0xF) 73 #define VTD_CAP_RWBF(cap) (((cap) >> 4) & 0x1) 74 75 #define VTD_ECAP_DI(ecap) (((ecap) >> 2) & 0x1) 76 #define VTD_ECAP_COHERENCY(ecap) ((ecap) & 0x1) 77 #define VTD_ECAP_IRO(ecap) (((ecap) >> 8) & 0x3FF) 78 79 #define VTD_GCR_WBF (1 << 27) 80 #define VTD_GCR_SRTP (1 << 30) 81 #define VTD_GCR_TE (1U << 31) 82 83 #define VTD_GSR_WBFS (1 << 27) 84 #define VTD_GSR_RTPS (1 << 30) 85 #define VTD_GSR_TES (1U << 31) 86 87 #define VTD_CCR_ICC (1UL << 63) /* invalidate context cache */ 88 #define VTD_CCR_CIRG_GLOBAL (1UL << 61) /* global invalidation */ 89 90 #define VTD_IIR_IVT (1UL << 63) /* invalidation IOTLB */ 91 #define VTD_IIR_IIRG_GLOBAL (1ULL << 60) /* global IOTLB invalidation */ 92 #define VTD_IIR_IIRG_DOMAIN (2ULL << 60) /* domain IOTLB invalidation */ 93 #define VTD_IIR_IIRG_PAGE (3ULL << 60) /* page IOTLB invalidation */ 94 #define VTD_IIR_DRAIN_READS (1ULL << 49) /* drain pending DMA reads */ 95 #define VTD_IIR_DRAIN_WRITES (1ULL << 48) /* drain pending DMA writes */ 96 #define VTD_IIR_DOMAIN_P 32 97 98 #define VTD_ROOT_PRESENT 0x1 99 #define VTD_CTX_PRESENT 0x1 100 #define VTD_CTX_TT_ALL (1UL << 2) 101 102 #define VTD_PTE_RD (1UL << 0) 103 #define VTD_PTE_WR (1UL << 1) 104 #define VTD_PTE_SUPERPAGE (1UL << 7) 105 #define VTD_PTE_ADDR_M (0x000FFFFFFFFFF000UL) 106 107 #define VTD_RID2IDX(rid) (((rid) & 0xff) * 2) 108 109 struct domain { 110 uint64_t *ptp; /* first level page table page */ 111 int pt_levels; /* number of page table levels */ 112 int addrwidth; /* 'AW' field in context entry */ 113 int spsmask; /* supported super page sizes */ 114 uint_t id; /* domain id */ 115 vm_paddr_t maxaddr; /* highest address to be mapped */ 116 SLIST_ENTRY(domain) next; 117 }; 118 119 static SLIST_HEAD(, domain) domhead; 120 121 #define DRHD_MAX_UNITS 8 122 static ACPI_DMAR_HARDWARE_UNIT *drhds[DRHD_MAX_UNITS]; 123 static int drhd_num; 124 static struct vtdmap *vtdmaps[DRHD_MAX_UNITS]; 125 static int max_domains; 126 typedef int (*drhd_ident_func_t)(void); 127 #ifndef __FreeBSD__ 128 static dev_info_t *vtddips[DRHD_MAX_UNITS]; 129 #endif 130 131 static uint64_t root_table[PAGE_SIZE / sizeof (uint64_t)] __aligned(4096); 132 static uint64_t ctx_tables[256][PAGE_SIZE / sizeof (uint64_t)] __aligned(4096); 133 134 static int 135 vtd_max_domains(struct vtdmap *vtdmap) 136 { 137 int nd; 138 139 nd = VTD_CAP_ND(vtdmap->cap); 140 141 switch (nd) { 142 case 0: 143 return (16); 144 case 1: 145 return (64); 146 case 2: 147 return (256); 148 case 3: 149 return (1024); 150 case 4: 151 return (4 * 1024); 152 case 5: 153 return (16 * 1024); 154 case 6: 155 return (64 * 1024); 156 default: 157 panic("vtd_max_domains: invalid value of nd (0x%0x)", nd); 158 } 159 } 160 161 static uint_t 162 domain_id(void) 163 { 164 uint_t id; 165 struct domain *dom; 166 167 /* Skip domain id 0 - it is reserved when Caching Mode field is set */ 168 for (id = 1; id < max_domains; id++) { 169 SLIST_FOREACH(dom, &domhead, next) { 170 if (dom->id == id) 171 break; 172 } 173 if (dom == NULL) 174 break; /* found it */ 175 } 176 177 if (id >= max_domains) 178 panic("domain ids exhausted"); 179 180 return (id); 181 } 182 183 static struct vtdmap * 184 vtd_device_scope(uint16_t rid) 185 { 186 int i, remaining, pathrem; 187 char *end, *pathend; 188 struct vtdmap *vtdmap; 189 ACPI_DMAR_HARDWARE_UNIT *drhd; 190 ACPI_DMAR_DEVICE_SCOPE *device_scope; 191 ACPI_DMAR_PCI_PATH *path; 192 193 for (i = 0; i < drhd_num; i++) { 194 drhd = drhds[i]; 195 196 if (VTD_DRHD_INCLUDE_PCI_ALL(drhd->Flags)) { 197 /* 198 * From Intel VT-d arch spec, version 3.0: 199 * If a DRHD structure with INCLUDE_PCI_ALL flag Set is 200 * reported for a Segment, it must be enumerated by BIOS 201 * after all other DRHD structures for the same Segment. 202 */ 203 vtdmap = vtdmaps[i]; 204 return (vtdmap); 205 } 206 207 end = (char *)drhd + drhd->Header.Length; 208 remaining = drhd->Header.Length - 209 sizeof (ACPI_DMAR_HARDWARE_UNIT); 210 while (remaining > sizeof (ACPI_DMAR_DEVICE_SCOPE)) { 211 device_scope = 212 (ACPI_DMAR_DEVICE_SCOPE *)(end - remaining); 213 remaining -= device_scope->Length; 214 215 switch (device_scope->EntryType) { 216 /* 0x01 and 0x02 are PCI device entries */ 217 case 0x01: 218 case 0x02: 219 break; 220 default: 221 continue; 222 } 223 224 if (PCI_RID2BUS(rid) != device_scope->Bus) 225 continue; 226 227 pathend = (char *)device_scope + device_scope->Length; 228 pathrem = device_scope->Length - 229 sizeof (ACPI_DMAR_DEVICE_SCOPE); 230 while (pathrem >= sizeof (ACPI_DMAR_PCI_PATH)) { 231 path = (ACPI_DMAR_PCI_PATH *) 232 (pathend - pathrem); 233 pathrem -= sizeof (ACPI_DMAR_PCI_PATH); 234 235 if (PCI_RID2SLOT(rid) != path->Device) 236 continue; 237 if (PCI_RID2FUNC(rid) != path->Function) 238 continue; 239 240 vtdmap = vtdmaps[i]; 241 return (vtdmap); 242 } 243 } 244 } 245 246 /* No matching scope */ 247 return (NULL); 248 } 249 250 static void 251 vtd_wbflush(struct vtdmap *vtdmap) 252 { 253 254 if (VTD_ECAP_COHERENCY(vtdmap->ext_cap) == 0) 255 invalidate_cache_all(); 256 257 if (VTD_CAP_RWBF(vtdmap->cap)) { 258 vtdmap->gcr = VTD_GCR_WBF; 259 while ((vtdmap->gsr & VTD_GSR_WBFS) != 0) 260 ; 261 } 262 } 263 264 static void 265 vtd_ctx_global_invalidate(struct vtdmap *vtdmap) 266 { 267 268 vtdmap->ccr = VTD_CCR_ICC | VTD_CCR_CIRG_GLOBAL; 269 while ((vtdmap->ccr & VTD_CCR_ICC) != 0) 270 ; 271 } 272 273 static void 274 vtd_iotlb_global_invalidate(struct vtdmap *vtdmap) 275 { 276 int offset; 277 volatile uint64_t *iotlb_reg, val; 278 279 vtd_wbflush(vtdmap); 280 281 offset = VTD_ECAP_IRO(vtdmap->ext_cap) * 16; 282 iotlb_reg = (volatile uint64_t *)((caddr_t)vtdmap + offset + 8); 283 284 *iotlb_reg = VTD_IIR_IVT | VTD_IIR_IIRG_GLOBAL | 285 VTD_IIR_DRAIN_READS | VTD_IIR_DRAIN_WRITES; 286 287 while (1) { 288 val = *iotlb_reg; 289 if ((val & VTD_IIR_IVT) == 0) 290 break; 291 } 292 } 293 294 static void 295 vtd_translation_enable(struct vtdmap *vtdmap) 296 { 297 298 vtdmap->gcr = VTD_GCR_TE; 299 while ((vtdmap->gsr & VTD_GSR_TES) == 0) 300 ; 301 } 302 303 static void 304 vtd_translation_disable(struct vtdmap *vtdmap) 305 { 306 307 vtdmap->gcr = 0; 308 while ((vtdmap->gsr & VTD_GSR_TES) != 0) 309 ; 310 } 311 312 static void * 313 vtd_map(dev_info_t *dip) 314 { 315 caddr_t regs; 316 ddi_acc_handle_t hdl; 317 int error; 318 319 static ddi_device_acc_attr_t regs_attr = { 320 DDI_DEVICE_ATTR_V0, 321 DDI_NEVERSWAP_ACC, 322 DDI_STRICTORDER_ACC, 323 }; 324 325 error = ddi_regs_map_setup(dip, 0, ®s, 0, PAGE_SIZE, ®s_attr, 326 &hdl); 327 328 if (error != DDI_SUCCESS) 329 return (NULL); 330 331 ddi_set_driver_private(dip, hdl); 332 333 return (regs); 334 } 335 336 static void 337 vtd_unmap(dev_info_t *dip) 338 { 339 ddi_acc_handle_t hdl = ddi_get_driver_private(dip); 340 341 if (hdl != NULL) 342 ddi_regs_map_free(&hdl); 343 } 344 345 #ifndef __FreeBSD__ 346 /* 347 * This lives in vtd_sol.c for license reasons. 348 */ 349 extern dev_info_t *vtd_get_dip(ACPI_DMAR_HARDWARE_UNIT *, int); 350 #endif 351 352 static int 353 vtd_init(void) 354 { 355 int i, units, remaining, tmp; 356 struct vtdmap *vtdmap; 357 vm_paddr_t ctx_paddr; 358 char *end; 359 #ifdef __FreeBSD__ 360 char envname[32]; 361 unsigned long mapaddr; 362 #endif 363 ACPI_STATUS status; 364 ACPI_TABLE_DMAR *dmar; 365 ACPI_DMAR_HEADER *hdr; 366 ACPI_DMAR_HARDWARE_UNIT *drhd; 367 368 #ifdef __FreeBSD__ 369 /* 370 * Allow the user to override the ACPI DMAR table by specifying the 371 * physical address of each remapping unit. 372 * 373 * The following example specifies two remapping units at 374 * physical addresses 0xfed90000 and 0xfeda0000 respectively. 375 * set vtd.regmap.0.addr=0xfed90000 376 * set vtd.regmap.1.addr=0xfeda0000 377 */ 378 for (units = 0; units < DRHD_MAX_UNITS; units++) { 379 snprintf(envname, sizeof (envname), "vtd.regmap.%d.addr", 380 units); 381 if (getenv_ulong(envname, &mapaddr) == 0) 382 break; 383 vtdmaps[units] = (struct vtdmap *)PHYS_TO_DMAP(mapaddr); 384 } 385 386 if (units > 0) 387 goto skip_dmar; 388 #else 389 units = 0; 390 #endif 391 /* Search for DMAR table. */ 392 status = AcpiGetTable(ACPI_SIG_DMAR, 0, (ACPI_TABLE_HEADER **)&dmar); 393 if (ACPI_FAILURE(status)) 394 return (ENXIO); 395 396 end = (char *)dmar + dmar->Header.Length; 397 remaining = dmar->Header.Length - sizeof (ACPI_TABLE_DMAR); 398 while (remaining > sizeof (ACPI_DMAR_HEADER)) { 399 hdr = (ACPI_DMAR_HEADER *)(end - remaining); 400 if (hdr->Length > remaining) 401 break; 402 /* 403 * From Intel VT-d arch spec, version 1.3: 404 * BIOS implementations must report mapping structures 405 * in numerical order, i.e. All remapping structures of 406 * type 0 (DRHD) enumerated before remapping structures of 407 * type 1 (RMRR) and so forth. 408 */ 409 if (hdr->Type != ACPI_DMAR_TYPE_HARDWARE_UNIT) 410 break; 411 412 drhd = (ACPI_DMAR_HARDWARE_UNIT *)hdr; 413 drhds[units] = drhd; 414 #ifdef __FreeBSD__ 415 vtdmaps[units] = (struct vtdmap *)PHYS_TO_DMAP(drhd->Address); 416 #else 417 vtddips[units] = vtd_get_dip(drhd, units); 418 vtdmaps[units] = (struct vtdmap *)vtd_map(vtddips[units]); 419 if (vtdmaps[units] == NULL) 420 goto fail; 421 #endif 422 if (++units >= DRHD_MAX_UNITS) 423 break; 424 remaining -= hdr->Length; 425 } 426 427 if (units <= 0) 428 return (ENXIO); 429 430 #ifdef __FreeBSD__ 431 skip_dmar: 432 #endif 433 drhd_num = units; 434 435 max_domains = 64 * 1024; /* maximum valid value */ 436 for (i = 0; i < drhd_num; i++) { 437 vtdmap = vtdmaps[i]; 438 439 if (VTD_CAP_CM(vtdmap->cap) != 0) 440 panic("vtd_init: invalid caching mode"); 441 442 /* take most compatible (minimum) value */ 443 if ((tmp = vtd_max_domains(vtdmap)) < max_domains) 444 max_domains = tmp; 445 } 446 447 /* 448 * Set up the root-table to point to the context-entry tables 449 */ 450 for (i = 0; i < 256; i++) { 451 ctx_paddr = vtophys(ctx_tables[i]); 452 if (ctx_paddr & PAGE_MASK) 453 panic("ctx table (0x%0lx) not page aligned", ctx_paddr); 454 455 root_table[i * 2] = ctx_paddr | VTD_ROOT_PRESENT; 456 } 457 458 return (0); 459 460 #ifndef __FreeBSD__ 461 fail: 462 for (i = 0; i <= units; i++) 463 vtd_unmap(vtddips[i]); 464 return (ENXIO); 465 #endif 466 } 467 468 static void 469 vtd_cleanup(void) 470 { 471 #ifndef __FreeBSD__ 472 int i; 473 474 KASSERT(SLIST_EMPTY(&domhead), ("domain list not empty")); 475 476 bzero(root_table, sizeof (root_table)); 477 478 for (i = 0; i <= drhd_num; i++) { 479 vtdmaps[i] = NULL; 480 /* 481 * Unmap the vtd registers. Note that the devinfo nodes 482 * themselves aren't removed, they are considered system state 483 * and can be reused when the module is reloaded. 484 */ 485 if (vtddips[i] != NULL) 486 vtd_unmap(vtddips[i]); 487 } 488 #endif 489 } 490 491 static void 492 vtd_enable(void) 493 { 494 int i; 495 struct vtdmap *vtdmap; 496 497 for (i = 0; i < drhd_num; i++) { 498 vtdmap = vtdmaps[i]; 499 vtd_wbflush(vtdmap); 500 501 /* Update the root table address */ 502 vtdmap->rta = vtophys(root_table); 503 vtdmap->gcr = VTD_GCR_SRTP; 504 while ((vtdmap->gsr & VTD_GSR_RTPS) == 0) 505 ; 506 507 vtd_ctx_global_invalidate(vtdmap); 508 vtd_iotlb_global_invalidate(vtdmap); 509 510 vtd_translation_enable(vtdmap); 511 } 512 } 513 514 static void 515 vtd_disable(void) 516 { 517 int i; 518 struct vtdmap *vtdmap; 519 520 for (i = 0; i < drhd_num; i++) { 521 vtdmap = vtdmaps[i]; 522 vtd_translation_disable(vtdmap); 523 } 524 } 525 526 static void 527 vtd_add_device(void *arg, uint16_t rid) 528 { 529 int idx; 530 uint64_t *ctxp; 531 struct domain *dom = arg; 532 vm_paddr_t pt_paddr; 533 struct vtdmap *vtdmap; 534 uint8_t bus; 535 536 bus = PCI_RID2BUS(rid); 537 ctxp = ctx_tables[bus]; 538 pt_paddr = vtophys(dom->ptp); 539 idx = VTD_RID2IDX(rid); 540 541 if (ctxp[idx] & VTD_CTX_PRESENT) { 542 panic("vtd_add_device: device %x is already owned by " 543 "domain %d", rid, (uint16_t)(ctxp[idx + 1] >> 8)); 544 } 545 546 if ((vtdmap = vtd_device_scope(rid)) == NULL) 547 panic("vtd_add_device: device %x is not in scope for " 548 "any DMA remapping unit", rid); 549 550 /* 551 * Order is important. The 'present' bit is set only after all fields 552 * of the context pointer are initialized. 553 */ 554 ctxp[idx + 1] = dom->addrwidth | (dom->id << 8); 555 556 if (VTD_ECAP_DI(vtdmap->ext_cap)) 557 ctxp[idx] = VTD_CTX_TT_ALL; 558 else 559 ctxp[idx] = 0; 560 561 ctxp[idx] |= pt_paddr | VTD_CTX_PRESENT; 562 563 /* 564 * 'Not Present' entries are not cached in either the Context Cache 565 * or in the IOTLB, so there is no need to invalidate either of them. 566 */ 567 } 568 569 static void 570 vtd_remove_device(void *arg, uint16_t rid) 571 { 572 int i, idx; 573 uint64_t *ctxp; 574 struct vtdmap *vtdmap; 575 uint8_t bus; 576 577 bus = PCI_RID2BUS(rid); 578 ctxp = ctx_tables[bus]; 579 idx = VTD_RID2IDX(rid); 580 581 /* 582 * Order is important. The 'present' bit is must be cleared first. 583 */ 584 ctxp[idx] = 0; 585 ctxp[idx + 1] = 0; 586 587 /* 588 * Invalidate the Context Cache and the IOTLB. 589 * 590 * XXX use device-selective invalidation for Context Cache 591 * XXX use domain-selective invalidation for IOTLB 592 */ 593 for (i = 0; i < drhd_num; i++) { 594 vtdmap = vtdmaps[i]; 595 vtd_ctx_global_invalidate(vtdmap); 596 vtd_iotlb_global_invalidate(vtdmap); 597 } 598 } 599 600 #define CREATE_MAPPING 0 601 #define REMOVE_MAPPING 1 602 603 static uint64_t 604 vtd_update_mapping(void *arg, vm_paddr_t gpa, vm_paddr_t hpa, uint64_t len, 605 int remove) 606 { 607 struct domain *dom; 608 int i, spshift, ptpshift, ptpindex, nlevels; 609 uint64_t spsize, *ptp; 610 611 dom = arg; 612 ptpindex = 0; 613 ptpshift = 0; 614 615 KASSERT(gpa + len > gpa, ("%s: invalid gpa range %lx/%lx", __func__, 616 gpa, len)); 617 KASSERT(gpa + len <= dom->maxaddr, ("%s: gpa range %lx/%lx beyond " 618 "domain maxaddr %lx", __func__, gpa, len, dom->maxaddr)); 619 620 if (gpa & PAGE_MASK) 621 panic("vtd_create_mapping: unaligned gpa 0x%0lx", gpa); 622 623 if (hpa & PAGE_MASK) 624 panic("vtd_create_mapping: unaligned hpa 0x%0lx", hpa); 625 626 if (len & PAGE_MASK) 627 panic("vtd_create_mapping: unaligned len 0x%0lx", len); 628 629 /* 630 * Compute the size of the mapping that we can accommodate. 631 * 632 * This is based on three factors: 633 * - supported super page size 634 * - alignment of the region starting at 'gpa' and 'hpa' 635 * - length of the region 'len' 636 */ 637 spshift = 48; 638 for (i = 3; i >= 0; i--) { 639 spsize = 1UL << spshift; 640 if ((dom->spsmask & (1 << i)) != 0 && 641 (gpa & (spsize - 1)) == 0 && 642 (hpa & (spsize - 1)) == 0 && 643 (len >= spsize)) { 644 break; 645 } 646 spshift -= 9; 647 } 648 649 ptp = dom->ptp; 650 nlevels = dom->pt_levels; 651 while (--nlevels >= 0) { 652 ptpshift = 12 + nlevels * 9; 653 ptpindex = (gpa >> ptpshift) & 0x1FF; 654 655 /* We have reached the leaf mapping */ 656 if (spshift >= ptpshift) { 657 break; 658 } 659 660 /* 661 * We are working on a non-leaf page table page. 662 * 663 * Create a downstream page table page if necessary and point 664 * to it from the current page table. 665 */ 666 if (ptp[ptpindex] == 0) { 667 void *nlp = vmm_ptp_alloc(); 668 ptp[ptpindex] = vtophys(nlp)| VTD_PTE_RD | VTD_PTE_WR; 669 } 670 671 ptp = (uint64_t *)PHYS_TO_DMAP(ptp[ptpindex] & VTD_PTE_ADDR_M); 672 } 673 674 if ((gpa & ((1UL << ptpshift) - 1)) != 0) 675 panic("gpa 0x%lx and ptpshift %d mismatch", gpa, ptpshift); 676 677 /* 678 * Update the 'gpa' -> 'hpa' mapping 679 */ 680 if (remove) { 681 ptp[ptpindex] = 0; 682 } else { 683 ptp[ptpindex] = hpa | VTD_PTE_RD | VTD_PTE_WR; 684 685 if (nlevels > 0) 686 ptp[ptpindex] |= VTD_PTE_SUPERPAGE; 687 } 688 689 return (1UL << ptpshift); 690 } 691 692 static uint64_t 693 vtd_create_mapping(void *arg, vm_paddr_t gpa, vm_paddr_t hpa, uint64_t len) 694 { 695 696 return (vtd_update_mapping(arg, gpa, hpa, len, CREATE_MAPPING)); 697 } 698 699 static uint64_t 700 vtd_remove_mapping(void *arg, vm_paddr_t gpa, uint64_t len) 701 { 702 703 return (vtd_update_mapping(arg, gpa, 0, len, REMOVE_MAPPING)); 704 } 705 706 static void 707 vtd_invalidate_tlb(void *dom) 708 { 709 int i; 710 struct vtdmap *vtdmap; 711 712 /* 713 * Invalidate the IOTLB. 714 * XXX use domain-selective invalidation for IOTLB 715 */ 716 for (i = 0; i < drhd_num; i++) { 717 vtdmap = vtdmaps[i]; 718 vtd_iotlb_global_invalidate(vtdmap); 719 } 720 } 721 722 static void * 723 vtd_create_domain(vm_paddr_t maxaddr) 724 { 725 struct domain *dom; 726 vm_paddr_t addr; 727 int tmp, i, gaw, agaw, sagaw, res, pt_levels, addrwidth; 728 struct vtdmap *vtdmap; 729 730 if (drhd_num <= 0) 731 panic("vtd_create_domain: no dma remapping hardware available"); 732 733 /* 734 * Calculate AGAW. 735 * Section 3.4.2 "Adjusted Guest Address Width", Architecture Spec. 736 */ 737 addr = 0; 738 for (gaw = 0; addr < maxaddr; gaw++) 739 addr = 1ULL << gaw; 740 741 res = (gaw - 12) % 9; 742 if (res == 0) 743 agaw = gaw; 744 else 745 agaw = gaw + 9 - res; 746 747 if (agaw > 64) 748 agaw = 64; 749 750 /* 751 * Select the smallest Supported AGAW and the corresponding number 752 * of page table levels. 753 */ 754 pt_levels = 2; 755 sagaw = 30; 756 addrwidth = 0; 757 758 tmp = ~0; 759 for (i = 0; i < drhd_num; i++) { 760 vtdmap = vtdmaps[i]; 761 /* take most compatible value */ 762 tmp &= VTD_CAP_SAGAW(vtdmap->cap); 763 } 764 765 for (i = 0; i < 5; i++) { 766 if ((tmp & (1 << i)) != 0 && sagaw >= agaw) 767 break; 768 pt_levels++; 769 addrwidth++; 770 sagaw += 9; 771 if (sagaw > 64) 772 sagaw = 64; 773 } 774 775 if (i >= 5) { 776 panic("vtd_create_domain: SAGAW 0x%x does not support AGAW %d", 777 tmp, agaw); 778 } 779 780 dom = kmem_zalloc(sizeof (struct domain), KM_SLEEP); 781 dom->pt_levels = pt_levels; 782 dom->addrwidth = addrwidth; 783 dom->id = domain_id(); 784 dom->maxaddr = maxaddr; 785 dom->ptp = vmm_ptp_alloc(); 786 if ((uintptr_t)dom->ptp & PAGE_MASK) 787 panic("vtd_create_domain: ptp (%p) not page aligned", dom->ptp); 788 789 #ifdef __FreeBSD__ 790 #ifdef notyet 791 /* 792 * XXX superpage mappings for the iommu do not work correctly. 793 * 794 * By default all physical memory is mapped into the host_domain. 795 * When a VM is allocated wired memory the pages belonging to it 796 * are removed from the host_domain and added to the vm's domain. 797 * 798 * If the page being removed was mapped using a superpage mapping 799 * in the host_domain then we need to demote the mapping before 800 * removing the page. 801 * 802 * There is not any code to deal with the demotion at the moment 803 * so we disable superpage mappings altogether. 804 */ 805 dom->spsmask = ~0; 806 for (i = 0; i < drhd_num; i++) { 807 vtdmap = vtdmaps[i]; 808 /* take most compatible value */ 809 dom->spsmask &= VTD_CAP_SPS(vtdmap->cap); 810 } 811 #endif 812 #else 813 /* 814 * On illumos we decidedly do not remove memory mapped to a VM's domain 815 * from the host_domain, so we don't have to deal with page demotion and 816 * can just use large pages. 817 * 818 * Since VM memory is currently allocated as 4k pages and mapped into 819 * the VM domain page by page, the use of large pages is essentially 820 * limited to the host_domain. 821 */ 822 dom->spsmask = VTD_CAP_SPS(vtdmap->cap); 823 #endif 824 825 SLIST_INSERT_HEAD(&domhead, dom, next); 826 827 return (dom); 828 } 829 830 static void 831 vtd_free_ptp(uint64_t *ptp, int level) 832 { 833 int i; 834 uint64_t *nlp; 835 836 if (level > 1) { 837 for (i = 0; i < 512; i++) { 838 if ((ptp[i] & (VTD_PTE_RD | VTD_PTE_WR)) == 0) 839 continue; 840 if ((ptp[i] & VTD_PTE_SUPERPAGE) != 0) 841 continue; 842 nlp = (uint64_t *)PHYS_TO_DMAP(ptp[i] & VTD_PTE_ADDR_M); 843 vtd_free_ptp(nlp, level - 1); 844 } 845 } 846 847 vmm_ptp_free(ptp); 848 } 849 850 static void 851 vtd_destroy_domain(void *arg) 852 { 853 struct domain *dom; 854 855 dom = arg; 856 857 SLIST_REMOVE(&domhead, dom, domain, next); 858 vtd_free_ptp(dom->ptp, dom->pt_levels); 859 kmem_free(dom, sizeof (*dom)); 860 } 861 862 const struct iommu_ops iommu_ops_intel = { 863 .init = vtd_init, 864 .cleanup = vtd_cleanup, 865 .enable = vtd_enable, 866 .disable = vtd_disable, 867 .create_domain = vtd_create_domain, 868 .destroy_domain = vtd_destroy_domain, 869 .create_mapping = vtd_create_mapping, 870 .remove_mapping = vtd_remove_mapping, 871 .add_device = vtd_add_device, 872 .remove_device = vtd_remove_device, 873 .invalidate_tlb = vtd_invalidate_tlb, 874 }; 875