1 /* 2 * QEMU emulation of an Intel IOMMU (VT-d) 3 * (DMA Remapping device) 4 * 5 * Copyright (C) 2013 Knut Omang, Oracle <knut.omang@oracle.com> 6 * Copyright (C) 2014 Le Tan, <tamlokveer@gmail.com> 7 * 8 * This program is free software; you can redistribute it and/or modify 9 * it under the terms of the GNU General Public License as published by 10 * the Free Software Foundation; either version 2 of the License, or 11 * (at your option) any later version. 12 13 * This program is distributed in the hope that it will be useful, 14 * but WITHOUT ANY WARRANTY; without even the implied warranty of 15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 16 * GNU General Public License for more details. 17 18 * You should have received a copy of the GNU General Public License along 19 * with this program; if not, see <http://www.gnu.org/licenses/>. 20 */ 21 22 #include "qemu/osdep.h" 23 #include "qemu/error-report.h" 24 #include "qapi/error.h" 25 #include "hw/sysbus.h" 26 #include "exec/address-spaces.h" 27 #include "intel_iommu_internal.h" 28 #include "hw/pci/pci.h" 29 #include "hw/pci/pci_bus.h" 30 #include "hw/i386/pc.h" 31 #include "hw/i386/apic-msidef.h" 32 #include "hw/boards.h" 33 #include "hw/i386/x86-iommu.h" 34 #include "hw/pci-host/q35.h" 35 #include "sysemu/kvm.h" 36 #include "hw/i386/apic_internal.h" 37 #include "kvm_i386.h" 38 39 /*#define DEBUG_INTEL_IOMMU*/ 40 #ifdef DEBUG_INTEL_IOMMU 41 enum { 42 DEBUG_GENERAL, DEBUG_CSR, DEBUG_INV, DEBUG_MMU, DEBUG_FLOG, 43 DEBUG_CACHE, DEBUG_IR, 44 }; 45 #define VTD_DBGBIT(x) (1 << DEBUG_##x) 46 static int vtd_dbgflags = VTD_DBGBIT(GENERAL) | VTD_DBGBIT(CSR); 47 48 #define VTD_DPRINTF(what, fmt, ...) do { \ 49 if (vtd_dbgflags & VTD_DBGBIT(what)) { \ 50 fprintf(stderr, "(vtd)%s: " fmt "\n", __func__, \ 51 ## __VA_ARGS__); } \ 52 } while (0) 53 #else 54 #define VTD_DPRINTF(what, fmt, ...) do {} while (0) 55 #endif 56 57 static void vtd_define_quad(IntelIOMMUState *s, hwaddr addr, uint64_t val, 58 uint64_t wmask, uint64_t w1cmask) 59 { 60 stq_le_p(&s->csr[addr], val); 61 stq_le_p(&s->wmask[addr], wmask); 62 stq_le_p(&s->w1cmask[addr], w1cmask); 63 } 64 65 static void vtd_define_quad_wo(IntelIOMMUState *s, hwaddr addr, uint64_t mask) 66 { 67 stq_le_p(&s->womask[addr], mask); 68 } 69 70 static void vtd_define_long(IntelIOMMUState *s, hwaddr addr, uint32_t val, 71 uint32_t wmask, uint32_t w1cmask) 72 { 73 stl_le_p(&s->csr[addr], val); 74 stl_le_p(&s->wmask[addr], wmask); 75 stl_le_p(&s->w1cmask[addr], w1cmask); 76 } 77 78 static void vtd_define_long_wo(IntelIOMMUState *s, hwaddr addr, uint32_t mask) 79 { 80 stl_le_p(&s->womask[addr], mask); 81 } 82 83 /* "External" get/set operations */ 84 static void vtd_set_quad(IntelIOMMUState *s, hwaddr addr, uint64_t val) 85 { 86 uint64_t oldval = ldq_le_p(&s->csr[addr]); 87 uint64_t wmask = ldq_le_p(&s->wmask[addr]); 88 uint64_t w1cmask = ldq_le_p(&s->w1cmask[addr]); 89 stq_le_p(&s->csr[addr], 90 ((oldval & ~wmask) | (val & wmask)) & ~(w1cmask & val)); 91 } 92 93 static void vtd_set_long(IntelIOMMUState *s, hwaddr addr, uint32_t val) 94 { 95 uint32_t oldval = ldl_le_p(&s->csr[addr]); 96 uint32_t wmask = ldl_le_p(&s->wmask[addr]); 97 uint32_t w1cmask = ldl_le_p(&s->w1cmask[addr]); 98 stl_le_p(&s->csr[addr], 99 ((oldval & ~wmask) | (val & wmask)) & ~(w1cmask & val)); 100 } 101 102 static uint64_t vtd_get_quad(IntelIOMMUState *s, hwaddr addr) 103 { 104 uint64_t val = ldq_le_p(&s->csr[addr]); 105 uint64_t womask = ldq_le_p(&s->womask[addr]); 106 return val & ~womask; 107 } 108 109 static uint32_t vtd_get_long(IntelIOMMUState *s, hwaddr addr) 110 { 111 uint32_t val = ldl_le_p(&s->csr[addr]); 112 uint32_t womask = ldl_le_p(&s->womask[addr]); 113 return val & ~womask; 114 } 115 116 /* "Internal" get/set operations */ 117 static uint64_t vtd_get_quad_raw(IntelIOMMUState *s, hwaddr addr) 118 { 119 return ldq_le_p(&s->csr[addr]); 120 } 121 122 static uint32_t vtd_get_long_raw(IntelIOMMUState *s, hwaddr addr) 123 { 124 return ldl_le_p(&s->csr[addr]); 125 } 126 127 static void vtd_set_quad_raw(IntelIOMMUState *s, hwaddr addr, uint64_t val) 128 { 129 stq_le_p(&s->csr[addr], val); 130 } 131 132 static uint32_t vtd_set_clear_mask_long(IntelIOMMUState *s, hwaddr addr, 133 uint32_t clear, uint32_t mask) 134 { 135 uint32_t new_val = (ldl_le_p(&s->csr[addr]) & ~clear) | mask; 136 stl_le_p(&s->csr[addr], new_val); 137 return new_val; 138 } 139 140 static uint64_t vtd_set_clear_mask_quad(IntelIOMMUState *s, hwaddr addr, 141 uint64_t clear, uint64_t mask) 142 { 143 uint64_t new_val = (ldq_le_p(&s->csr[addr]) & ~clear) | mask; 144 stq_le_p(&s->csr[addr], new_val); 145 return new_val; 146 } 147 148 /* GHashTable functions */ 149 static gboolean vtd_uint64_equal(gconstpointer v1, gconstpointer v2) 150 { 151 return *((const uint64_t *)v1) == *((const uint64_t *)v2); 152 } 153 154 static guint vtd_uint64_hash(gconstpointer v) 155 { 156 return (guint)*(const uint64_t *)v; 157 } 158 159 static gboolean vtd_hash_remove_by_domain(gpointer key, gpointer value, 160 gpointer user_data) 161 { 162 VTDIOTLBEntry *entry = (VTDIOTLBEntry *)value; 163 uint16_t domain_id = *(uint16_t *)user_data; 164 return entry->domain_id == domain_id; 165 } 166 167 /* The shift of an addr for a certain level of paging structure */ 168 static inline uint32_t vtd_slpt_level_shift(uint32_t level) 169 { 170 return VTD_PAGE_SHIFT_4K + (level - 1) * VTD_SL_LEVEL_BITS; 171 } 172 173 static inline uint64_t vtd_slpt_level_page_mask(uint32_t level) 174 { 175 return ~((1ULL << vtd_slpt_level_shift(level)) - 1); 176 } 177 178 static gboolean vtd_hash_remove_by_page(gpointer key, gpointer value, 179 gpointer user_data) 180 { 181 VTDIOTLBEntry *entry = (VTDIOTLBEntry *)value; 182 VTDIOTLBPageInvInfo *info = (VTDIOTLBPageInvInfo *)user_data; 183 uint64_t gfn = (info->addr >> VTD_PAGE_SHIFT_4K) & info->mask; 184 uint64_t gfn_tlb = (info->addr & entry->mask) >> VTD_PAGE_SHIFT_4K; 185 return (entry->domain_id == info->domain_id) && 186 (((entry->gfn & info->mask) == gfn) || 187 (entry->gfn == gfn_tlb)); 188 } 189 190 /* Reset all the gen of VTDAddressSpace to zero and set the gen of 191 * IntelIOMMUState to 1. 192 */ 193 static void vtd_reset_context_cache(IntelIOMMUState *s) 194 { 195 VTDAddressSpace *vtd_as; 196 VTDBus *vtd_bus; 197 GHashTableIter bus_it; 198 uint32_t devfn_it; 199 200 g_hash_table_iter_init(&bus_it, s->vtd_as_by_busptr); 201 202 VTD_DPRINTF(CACHE, "global context_cache_gen=1"); 203 while (g_hash_table_iter_next (&bus_it, NULL, (void**)&vtd_bus)) { 204 for (devfn_it = 0; devfn_it < X86_IOMMU_PCI_DEVFN_MAX; ++devfn_it) { 205 vtd_as = vtd_bus->dev_as[devfn_it]; 206 if (!vtd_as) { 207 continue; 208 } 209 vtd_as->context_cache_entry.context_cache_gen = 0; 210 } 211 } 212 s->context_cache_gen = 1; 213 } 214 215 static void vtd_reset_iotlb(IntelIOMMUState *s) 216 { 217 assert(s->iotlb); 218 g_hash_table_remove_all(s->iotlb); 219 } 220 221 static uint64_t vtd_get_iotlb_key(uint64_t gfn, uint16_t source_id, 222 uint32_t level) 223 { 224 return gfn | ((uint64_t)(source_id) << VTD_IOTLB_SID_SHIFT) | 225 ((uint64_t)(level) << VTD_IOTLB_LVL_SHIFT); 226 } 227 228 static uint64_t vtd_get_iotlb_gfn(hwaddr addr, uint32_t level) 229 { 230 return (addr & vtd_slpt_level_page_mask(level)) >> VTD_PAGE_SHIFT_4K; 231 } 232 233 static VTDIOTLBEntry *vtd_lookup_iotlb(IntelIOMMUState *s, uint16_t source_id, 234 hwaddr addr) 235 { 236 VTDIOTLBEntry *entry; 237 uint64_t key; 238 int level; 239 240 for (level = VTD_SL_PT_LEVEL; level < VTD_SL_PML4_LEVEL; level++) { 241 key = vtd_get_iotlb_key(vtd_get_iotlb_gfn(addr, level), 242 source_id, level); 243 entry = g_hash_table_lookup(s->iotlb, &key); 244 if (entry) { 245 goto out; 246 } 247 } 248 249 out: 250 return entry; 251 } 252 253 static void vtd_update_iotlb(IntelIOMMUState *s, uint16_t source_id, 254 uint16_t domain_id, hwaddr addr, uint64_t slpte, 255 bool read_flags, bool write_flags, 256 uint32_t level) 257 { 258 VTDIOTLBEntry *entry = g_malloc(sizeof(*entry)); 259 uint64_t *key = g_malloc(sizeof(*key)); 260 uint64_t gfn = vtd_get_iotlb_gfn(addr, level); 261 262 VTD_DPRINTF(CACHE, "update iotlb sid 0x%"PRIx16 " gpa 0x%"PRIx64 263 " slpte 0x%"PRIx64 " did 0x%"PRIx16, source_id, addr, slpte, 264 domain_id); 265 if (g_hash_table_size(s->iotlb) >= VTD_IOTLB_MAX_SIZE) { 266 VTD_DPRINTF(CACHE, "iotlb exceeds size limit, forced to reset"); 267 vtd_reset_iotlb(s); 268 } 269 270 entry->gfn = gfn; 271 entry->domain_id = domain_id; 272 entry->slpte = slpte; 273 entry->read_flags = read_flags; 274 entry->write_flags = write_flags; 275 entry->mask = vtd_slpt_level_page_mask(level); 276 *key = vtd_get_iotlb_key(gfn, source_id, level); 277 g_hash_table_replace(s->iotlb, key, entry); 278 } 279 280 /* Given the reg addr of both the message data and address, generate an 281 * interrupt via MSI. 282 */ 283 static void vtd_generate_interrupt(IntelIOMMUState *s, hwaddr mesg_addr_reg, 284 hwaddr mesg_data_reg) 285 { 286 MSIMessage msi; 287 288 assert(mesg_data_reg < DMAR_REG_SIZE); 289 assert(mesg_addr_reg < DMAR_REG_SIZE); 290 291 msi.address = vtd_get_long_raw(s, mesg_addr_reg); 292 msi.data = vtd_get_long_raw(s, mesg_data_reg); 293 294 VTD_DPRINTF(FLOG, "msi: addr 0x%"PRIx64 " data 0x%"PRIx32, 295 msi.address, msi.data); 296 apic_get_class()->send_msi(&msi); 297 } 298 299 /* Generate a fault event to software via MSI if conditions are met. 300 * Notice that the value of FSTS_REG being passed to it should be the one 301 * before any update. 302 */ 303 static void vtd_generate_fault_event(IntelIOMMUState *s, uint32_t pre_fsts) 304 { 305 if (pre_fsts & VTD_FSTS_PPF || pre_fsts & VTD_FSTS_PFO || 306 pre_fsts & VTD_FSTS_IQE) { 307 VTD_DPRINTF(FLOG, "there are previous interrupt conditions " 308 "to be serviced by software, fault event is not generated " 309 "(FSTS_REG 0x%"PRIx32 ")", pre_fsts); 310 return; 311 } 312 vtd_set_clear_mask_long(s, DMAR_FECTL_REG, 0, VTD_FECTL_IP); 313 if (vtd_get_long_raw(s, DMAR_FECTL_REG) & VTD_FECTL_IM) { 314 VTD_DPRINTF(FLOG, "Interrupt Mask set, fault event is not generated"); 315 } else { 316 vtd_generate_interrupt(s, DMAR_FEADDR_REG, DMAR_FEDATA_REG); 317 vtd_set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0); 318 } 319 } 320 321 /* Check if the Fault (F) field of the Fault Recording Register referenced by 322 * @index is Set. 323 */ 324 static bool vtd_is_frcd_set(IntelIOMMUState *s, uint16_t index) 325 { 326 /* Each reg is 128-bit */ 327 hwaddr addr = DMAR_FRCD_REG_OFFSET + (((uint64_t)index) << 4); 328 addr += 8; /* Access the high 64-bit half */ 329 330 assert(index < DMAR_FRCD_REG_NR); 331 332 return vtd_get_quad_raw(s, addr) & VTD_FRCD_F; 333 } 334 335 /* Update the PPF field of Fault Status Register. 336 * Should be called whenever change the F field of any fault recording 337 * registers. 338 */ 339 static void vtd_update_fsts_ppf(IntelIOMMUState *s) 340 { 341 uint32_t i; 342 uint32_t ppf_mask = 0; 343 344 for (i = 0; i < DMAR_FRCD_REG_NR; i++) { 345 if (vtd_is_frcd_set(s, i)) { 346 ppf_mask = VTD_FSTS_PPF; 347 break; 348 } 349 } 350 vtd_set_clear_mask_long(s, DMAR_FSTS_REG, VTD_FSTS_PPF, ppf_mask); 351 VTD_DPRINTF(FLOG, "set PPF of FSTS_REG to %d", ppf_mask ? 1 : 0); 352 } 353 354 static void vtd_set_frcd_and_update_ppf(IntelIOMMUState *s, uint16_t index) 355 { 356 /* Each reg is 128-bit */ 357 hwaddr addr = DMAR_FRCD_REG_OFFSET + (((uint64_t)index) << 4); 358 addr += 8; /* Access the high 64-bit half */ 359 360 assert(index < DMAR_FRCD_REG_NR); 361 362 vtd_set_clear_mask_quad(s, addr, 0, VTD_FRCD_F); 363 vtd_update_fsts_ppf(s); 364 } 365 366 /* Must not update F field now, should be done later */ 367 static void vtd_record_frcd(IntelIOMMUState *s, uint16_t index, 368 uint16_t source_id, hwaddr addr, 369 VTDFaultReason fault, bool is_write) 370 { 371 uint64_t hi = 0, lo; 372 hwaddr frcd_reg_addr = DMAR_FRCD_REG_OFFSET + (((uint64_t)index) << 4); 373 374 assert(index < DMAR_FRCD_REG_NR); 375 376 lo = VTD_FRCD_FI(addr); 377 hi = VTD_FRCD_SID(source_id) | VTD_FRCD_FR(fault); 378 if (!is_write) { 379 hi |= VTD_FRCD_T; 380 } 381 vtd_set_quad_raw(s, frcd_reg_addr, lo); 382 vtd_set_quad_raw(s, frcd_reg_addr + 8, hi); 383 VTD_DPRINTF(FLOG, "record to FRCD_REG #%"PRIu16 ": hi 0x%"PRIx64 384 ", lo 0x%"PRIx64, index, hi, lo); 385 } 386 387 /* Try to collapse multiple pending faults from the same requester */ 388 static bool vtd_try_collapse_fault(IntelIOMMUState *s, uint16_t source_id) 389 { 390 uint32_t i; 391 uint64_t frcd_reg; 392 hwaddr addr = DMAR_FRCD_REG_OFFSET + 8; /* The high 64-bit half */ 393 394 for (i = 0; i < DMAR_FRCD_REG_NR; i++) { 395 frcd_reg = vtd_get_quad_raw(s, addr); 396 VTD_DPRINTF(FLOG, "frcd_reg #%d 0x%"PRIx64, i, frcd_reg); 397 if ((frcd_reg & VTD_FRCD_F) && 398 ((frcd_reg & VTD_FRCD_SID_MASK) == source_id)) { 399 return true; 400 } 401 addr += 16; /* 128-bit for each */ 402 } 403 return false; 404 } 405 406 /* Log and report an DMAR (address translation) fault to software */ 407 static void vtd_report_dmar_fault(IntelIOMMUState *s, uint16_t source_id, 408 hwaddr addr, VTDFaultReason fault, 409 bool is_write) 410 { 411 uint32_t fsts_reg = vtd_get_long_raw(s, DMAR_FSTS_REG); 412 413 assert(fault < VTD_FR_MAX); 414 415 if (fault == VTD_FR_RESERVED_ERR) { 416 /* This is not a normal fault reason case. Drop it. */ 417 return; 418 } 419 VTD_DPRINTF(FLOG, "sid 0x%"PRIx16 ", fault %d, addr 0x%"PRIx64 420 ", is_write %d", source_id, fault, addr, is_write); 421 if (fsts_reg & VTD_FSTS_PFO) { 422 VTD_DPRINTF(FLOG, "new fault is not recorded due to " 423 "Primary Fault Overflow"); 424 return; 425 } 426 if (vtd_try_collapse_fault(s, source_id)) { 427 VTD_DPRINTF(FLOG, "new fault is not recorded due to " 428 "compression of faults"); 429 return; 430 } 431 if (vtd_is_frcd_set(s, s->next_frcd_reg)) { 432 VTD_DPRINTF(FLOG, "Primary Fault Overflow and " 433 "new fault is not recorded, set PFO field"); 434 vtd_set_clear_mask_long(s, DMAR_FSTS_REG, 0, VTD_FSTS_PFO); 435 return; 436 } 437 438 vtd_record_frcd(s, s->next_frcd_reg, source_id, addr, fault, is_write); 439 440 if (fsts_reg & VTD_FSTS_PPF) { 441 VTD_DPRINTF(FLOG, "there are pending faults already, " 442 "fault event is not generated"); 443 vtd_set_frcd_and_update_ppf(s, s->next_frcd_reg); 444 s->next_frcd_reg++; 445 if (s->next_frcd_reg == DMAR_FRCD_REG_NR) { 446 s->next_frcd_reg = 0; 447 } 448 } else { 449 vtd_set_clear_mask_long(s, DMAR_FSTS_REG, VTD_FSTS_FRI_MASK, 450 VTD_FSTS_FRI(s->next_frcd_reg)); 451 vtd_set_frcd_and_update_ppf(s, s->next_frcd_reg); /* Will set PPF */ 452 s->next_frcd_reg++; 453 if (s->next_frcd_reg == DMAR_FRCD_REG_NR) { 454 s->next_frcd_reg = 0; 455 } 456 /* This case actually cause the PPF to be Set. 457 * So generate fault event (interrupt). 458 */ 459 vtd_generate_fault_event(s, fsts_reg); 460 } 461 } 462 463 /* Handle Invalidation Queue Errors of queued invalidation interface error 464 * conditions. 465 */ 466 static void vtd_handle_inv_queue_error(IntelIOMMUState *s) 467 { 468 uint32_t fsts_reg = vtd_get_long_raw(s, DMAR_FSTS_REG); 469 470 vtd_set_clear_mask_long(s, DMAR_FSTS_REG, 0, VTD_FSTS_IQE); 471 vtd_generate_fault_event(s, fsts_reg); 472 } 473 474 /* Set the IWC field and try to generate an invalidation completion interrupt */ 475 static void vtd_generate_completion_event(IntelIOMMUState *s) 476 { 477 VTD_DPRINTF(INV, "completes an invalidation wait command with " 478 "Interrupt Flag"); 479 if (vtd_get_long_raw(s, DMAR_ICS_REG) & VTD_ICS_IWC) { 480 VTD_DPRINTF(INV, "there is a previous interrupt condition to be " 481 "serviced by software, " 482 "new invalidation event is not generated"); 483 return; 484 } 485 vtd_set_clear_mask_long(s, DMAR_ICS_REG, 0, VTD_ICS_IWC); 486 vtd_set_clear_mask_long(s, DMAR_IECTL_REG, 0, VTD_IECTL_IP); 487 if (vtd_get_long_raw(s, DMAR_IECTL_REG) & VTD_IECTL_IM) { 488 VTD_DPRINTF(INV, "IM filed in IECTL_REG is set, new invalidation " 489 "event is not generated"); 490 return; 491 } else { 492 /* Generate the interrupt event */ 493 vtd_generate_interrupt(s, DMAR_IEADDR_REG, DMAR_IEDATA_REG); 494 vtd_set_clear_mask_long(s, DMAR_IECTL_REG, VTD_IECTL_IP, 0); 495 } 496 } 497 498 static inline bool vtd_root_entry_present(VTDRootEntry *root) 499 { 500 return root->val & VTD_ROOT_ENTRY_P; 501 } 502 503 static int vtd_get_root_entry(IntelIOMMUState *s, uint8_t index, 504 VTDRootEntry *re) 505 { 506 dma_addr_t addr; 507 508 addr = s->root + index * sizeof(*re); 509 if (dma_memory_read(&address_space_memory, addr, re, sizeof(*re))) { 510 VTD_DPRINTF(GENERAL, "error: fail to access root-entry at 0x%"PRIx64 511 " + %"PRIu8, s->root, index); 512 re->val = 0; 513 return -VTD_FR_ROOT_TABLE_INV; 514 } 515 re->val = le64_to_cpu(re->val); 516 return 0; 517 } 518 519 static inline bool vtd_context_entry_present(VTDContextEntry *context) 520 { 521 return context->lo & VTD_CONTEXT_ENTRY_P; 522 } 523 524 static int vtd_get_context_entry_from_root(VTDRootEntry *root, uint8_t index, 525 VTDContextEntry *ce) 526 { 527 dma_addr_t addr; 528 529 if (!vtd_root_entry_present(root)) { 530 VTD_DPRINTF(GENERAL, "error: root-entry is not present"); 531 return -VTD_FR_ROOT_ENTRY_P; 532 } 533 addr = (root->val & VTD_ROOT_ENTRY_CTP) + index * sizeof(*ce); 534 if (dma_memory_read(&address_space_memory, addr, ce, sizeof(*ce))) { 535 VTD_DPRINTF(GENERAL, "error: fail to access context-entry at 0x%"PRIx64 536 " + %"PRIu8, 537 (uint64_t)(root->val & VTD_ROOT_ENTRY_CTP), index); 538 return -VTD_FR_CONTEXT_TABLE_INV; 539 } 540 ce->lo = le64_to_cpu(ce->lo); 541 ce->hi = le64_to_cpu(ce->hi); 542 return 0; 543 } 544 545 static inline dma_addr_t vtd_get_slpt_base_from_context(VTDContextEntry *ce) 546 { 547 return ce->lo & VTD_CONTEXT_ENTRY_SLPTPTR; 548 } 549 550 static inline uint64_t vtd_get_slpte_addr(uint64_t slpte) 551 { 552 return slpte & VTD_SL_PT_BASE_ADDR_MASK; 553 } 554 555 /* Whether the pte indicates the address of the page frame */ 556 static inline bool vtd_is_last_slpte(uint64_t slpte, uint32_t level) 557 { 558 return level == VTD_SL_PT_LEVEL || (slpte & VTD_SL_PT_PAGE_SIZE_MASK); 559 } 560 561 /* Get the content of a spte located in @base_addr[@index] */ 562 static uint64_t vtd_get_slpte(dma_addr_t base_addr, uint32_t index) 563 { 564 uint64_t slpte; 565 566 assert(index < VTD_SL_PT_ENTRY_NR); 567 568 if (dma_memory_read(&address_space_memory, 569 base_addr + index * sizeof(slpte), &slpte, 570 sizeof(slpte))) { 571 slpte = (uint64_t)-1; 572 return slpte; 573 } 574 slpte = le64_to_cpu(slpte); 575 return slpte; 576 } 577 578 /* Given a gpa and the level of paging structure, return the offset of current 579 * level. 580 */ 581 static inline uint32_t vtd_gpa_level_offset(uint64_t gpa, uint32_t level) 582 { 583 return (gpa >> vtd_slpt_level_shift(level)) & 584 ((1ULL << VTD_SL_LEVEL_BITS) - 1); 585 } 586 587 /* Check Capability Register to see if the @level of page-table is supported */ 588 static inline bool vtd_is_level_supported(IntelIOMMUState *s, uint32_t level) 589 { 590 return VTD_CAP_SAGAW_MASK & s->cap & 591 (1ULL << (level - 2 + VTD_CAP_SAGAW_SHIFT)); 592 } 593 594 /* Get the page-table level that hardware should use for the second-level 595 * page-table walk from the Address Width field of context-entry. 596 */ 597 static inline uint32_t vtd_get_level_from_context_entry(VTDContextEntry *ce) 598 { 599 return 2 + (ce->hi & VTD_CONTEXT_ENTRY_AW); 600 } 601 602 static inline uint32_t vtd_get_agaw_from_context_entry(VTDContextEntry *ce) 603 { 604 return 30 + (ce->hi & VTD_CONTEXT_ENTRY_AW) * 9; 605 } 606 607 static const uint64_t vtd_paging_entry_rsvd_field[] = { 608 [0] = ~0ULL, 609 /* For not large page */ 610 [1] = 0x800ULL | ~(VTD_HAW_MASK | VTD_SL_IGN_COM), 611 [2] = 0x800ULL | ~(VTD_HAW_MASK | VTD_SL_IGN_COM), 612 [3] = 0x800ULL | ~(VTD_HAW_MASK | VTD_SL_IGN_COM), 613 [4] = 0x880ULL | ~(VTD_HAW_MASK | VTD_SL_IGN_COM), 614 /* For large page */ 615 [5] = 0x800ULL | ~(VTD_HAW_MASK | VTD_SL_IGN_COM), 616 [6] = 0x1ff800ULL | ~(VTD_HAW_MASK | VTD_SL_IGN_COM), 617 [7] = 0x3ffff800ULL | ~(VTD_HAW_MASK | VTD_SL_IGN_COM), 618 [8] = 0x880ULL | ~(VTD_HAW_MASK | VTD_SL_IGN_COM), 619 }; 620 621 static bool vtd_slpte_nonzero_rsvd(uint64_t slpte, uint32_t level) 622 { 623 if (slpte & VTD_SL_PT_PAGE_SIZE_MASK) { 624 /* Maybe large page */ 625 return slpte & vtd_paging_entry_rsvd_field[level + 4]; 626 } else { 627 return slpte & vtd_paging_entry_rsvd_field[level]; 628 } 629 } 630 631 /* Given the @gpa, get relevant @slptep. @slpte_level will be the last level 632 * of the translation, can be used for deciding the size of large page. 633 */ 634 static int vtd_gpa_to_slpte(VTDContextEntry *ce, uint64_t gpa, bool is_write, 635 uint64_t *slptep, uint32_t *slpte_level, 636 bool *reads, bool *writes) 637 { 638 dma_addr_t addr = vtd_get_slpt_base_from_context(ce); 639 uint32_t level = vtd_get_level_from_context_entry(ce); 640 uint32_t offset; 641 uint64_t slpte; 642 uint32_t ce_agaw = vtd_get_agaw_from_context_entry(ce); 643 uint64_t access_right_check; 644 645 /* Check if @gpa is above 2^X-1, where X is the minimum of MGAW in CAP_REG 646 * and AW in context-entry. 647 */ 648 if (gpa & ~((1ULL << MIN(ce_agaw, VTD_MGAW)) - 1)) { 649 VTD_DPRINTF(GENERAL, "error: gpa 0x%"PRIx64 " exceeds limits", gpa); 650 return -VTD_FR_ADDR_BEYOND_MGAW; 651 } 652 653 /* FIXME: what is the Atomics request here? */ 654 access_right_check = is_write ? VTD_SL_W : VTD_SL_R; 655 656 while (true) { 657 offset = vtd_gpa_level_offset(gpa, level); 658 slpte = vtd_get_slpte(addr, offset); 659 660 if (slpte == (uint64_t)-1) { 661 VTD_DPRINTF(GENERAL, "error: fail to access second-level paging " 662 "entry at level %"PRIu32 " for gpa 0x%"PRIx64, 663 level, gpa); 664 if (level == vtd_get_level_from_context_entry(ce)) { 665 /* Invalid programming of context-entry */ 666 return -VTD_FR_CONTEXT_ENTRY_INV; 667 } else { 668 return -VTD_FR_PAGING_ENTRY_INV; 669 } 670 } 671 *reads = (*reads) && (slpte & VTD_SL_R); 672 *writes = (*writes) && (slpte & VTD_SL_W); 673 if (!(slpte & access_right_check)) { 674 VTD_DPRINTF(GENERAL, "error: lack of %s permission for " 675 "gpa 0x%"PRIx64 " slpte 0x%"PRIx64, 676 (is_write ? "write" : "read"), gpa, slpte); 677 return is_write ? -VTD_FR_WRITE : -VTD_FR_READ; 678 } 679 if (vtd_slpte_nonzero_rsvd(slpte, level)) { 680 VTD_DPRINTF(GENERAL, "error: non-zero reserved field in second " 681 "level paging entry level %"PRIu32 " slpte 0x%"PRIx64, 682 level, slpte); 683 return -VTD_FR_PAGING_ENTRY_RSVD; 684 } 685 686 if (vtd_is_last_slpte(slpte, level)) { 687 *slptep = slpte; 688 *slpte_level = level; 689 return 0; 690 } 691 addr = vtd_get_slpte_addr(slpte); 692 level--; 693 } 694 } 695 696 /* Map a device to its corresponding domain (context-entry) */ 697 static int vtd_dev_to_context_entry(IntelIOMMUState *s, uint8_t bus_num, 698 uint8_t devfn, VTDContextEntry *ce) 699 { 700 VTDRootEntry re; 701 int ret_fr; 702 703 ret_fr = vtd_get_root_entry(s, bus_num, &re); 704 if (ret_fr) { 705 return ret_fr; 706 } 707 708 if (!vtd_root_entry_present(&re)) { 709 VTD_DPRINTF(GENERAL, "error: root-entry #%"PRIu8 " is not present", 710 bus_num); 711 return -VTD_FR_ROOT_ENTRY_P; 712 } else if (re.rsvd || (re.val & VTD_ROOT_ENTRY_RSVD)) { 713 VTD_DPRINTF(GENERAL, "error: non-zero reserved field in root-entry " 714 "hi 0x%"PRIx64 " lo 0x%"PRIx64, re.rsvd, re.val); 715 return -VTD_FR_ROOT_ENTRY_RSVD; 716 } 717 718 ret_fr = vtd_get_context_entry_from_root(&re, devfn, ce); 719 if (ret_fr) { 720 return ret_fr; 721 } 722 723 if (!vtd_context_entry_present(ce)) { 724 VTD_DPRINTF(GENERAL, 725 "error: context-entry #%"PRIu8 "(bus #%"PRIu8 ") " 726 "is not present", devfn, bus_num); 727 return -VTD_FR_CONTEXT_ENTRY_P; 728 } else if ((ce->hi & VTD_CONTEXT_ENTRY_RSVD_HI) || 729 (ce->lo & VTD_CONTEXT_ENTRY_RSVD_LO)) { 730 VTD_DPRINTF(GENERAL, 731 "error: non-zero reserved field in context-entry " 732 "hi 0x%"PRIx64 " lo 0x%"PRIx64, ce->hi, ce->lo); 733 return -VTD_FR_CONTEXT_ENTRY_RSVD; 734 } 735 /* Check if the programming of context-entry is valid */ 736 if (!vtd_is_level_supported(s, vtd_get_level_from_context_entry(ce))) { 737 VTD_DPRINTF(GENERAL, "error: unsupported Address Width value in " 738 "context-entry hi 0x%"PRIx64 " lo 0x%"PRIx64, 739 ce->hi, ce->lo); 740 return -VTD_FR_CONTEXT_ENTRY_INV; 741 } else { 742 switch (ce->lo & VTD_CONTEXT_ENTRY_TT) { 743 case VTD_CONTEXT_TT_MULTI_LEVEL: 744 /* fall through */ 745 case VTD_CONTEXT_TT_DEV_IOTLB: 746 break; 747 default: 748 VTD_DPRINTF(GENERAL, "error: unsupported Translation Type in " 749 "context-entry hi 0x%"PRIx64 " lo 0x%"PRIx64, 750 ce->hi, ce->lo); 751 return -VTD_FR_CONTEXT_ENTRY_INV; 752 } 753 } 754 return 0; 755 } 756 757 static inline uint16_t vtd_make_source_id(uint8_t bus_num, uint8_t devfn) 758 { 759 return ((bus_num & 0xffUL) << 8) | (devfn & 0xffUL); 760 } 761 762 static const bool vtd_qualified_faults[] = { 763 [VTD_FR_RESERVED] = false, 764 [VTD_FR_ROOT_ENTRY_P] = false, 765 [VTD_FR_CONTEXT_ENTRY_P] = true, 766 [VTD_FR_CONTEXT_ENTRY_INV] = true, 767 [VTD_FR_ADDR_BEYOND_MGAW] = true, 768 [VTD_FR_WRITE] = true, 769 [VTD_FR_READ] = true, 770 [VTD_FR_PAGING_ENTRY_INV] = true, 771 [VTD_FR_ROOT_TABLE_INV] = false, 772 [VTD_FR_CONTEXT_TABLE_INV] = false, 773 [VTD_FR_ROOT_ENTRY_RSVD] = false, 774 [VTD_FR_PAGING_ENTRY_RSVD] = true, 775 [VTD_FR_CONTEXT_ENTRY_TT] = true, 776 [VTD_FR_RESERVED_ERR] = false, 777 [VTD_FR_MAX] = false, 778 }; 779 780 /* To see if a fault condition is "qualified", which is reported to software 781 * only if the FPD field in the context-entry used to process the faulting 782 * request is 0. 783 */ 784 static inline bool vtd_is_qualified_fault(VTDFaultReason fault) 785 { 786 return vtd_qualified_faults[fault]; 787 } 788 789 static inline bool vtd_is_interrupt_addr(hwaddr addr) 790 { 791 return VTD_INTERRUPT_ADDR_FIRST <= addr && addr <= VTD_INTERRUPT_ADDR_LAST; 792 } 793 794 /* Map dev to context-entry then do a paging-structures walk to do a iommu 795 * translation. 796 * 797 * Called from RCU critical section. 798 * 799 * @bus_num: The bus number 800 * @devfn: The devfn, which is the combined of device and function number 801 * @is_write: The access is a write operation 802 * @entry: IOMMUTLBEntry that contain the addr to be translated and result 803 */ 804 static void vtd_do_iommu_translate(VTDAddressSpace *vtd_as, PCIBus *bus, 805 uint8_t devfn, hwaddr addr, bool is_write, 806 IOMMUTLBEntry *entry) 807 { 808 IntelIOMMUState *s = vtd_as->iommu_state; 809 VTDContextEntry ce; 810 uint8_t bus_num = pci_bus_num(bus); 811 VTDContextCacheEntry *cc_entry = &vtd_as->context_cache_entry; 812 uint64_t slpte, page_mask; 813 uint32_t level; 814 uint16_t source_id = vtd_make_source_id(bus_num, devfn); 815 int ret_fr; 816 bool is_fpd_set = false; 817 bool reads = true; 818 bool writes = true; 819 VTDIOTLBEntry *iotlb_entry; 820 821 /* Check if the request is in interrupt address range */ 822 if (vtd_is_interrupt_addr(addr)) { 823 if (is_write) { 824 /* FIXME: since we don't know the length of the access here, we 825 * treat Non-DWORD length write requests without PASID as 826 * interrupt requests, too. Withoud interrupt remapping support, 827 * we just use 1:1 mapping. 828 */ 829 VTD_DPRINTF(MMU, "write request to interrupt address " 830 "gpa 0x%"PRIx64, addr); 831 entry->iova = addr & VTD_PAGE_MASK_4K; 832 entry->translated_addr = addr & VTD_PAGE_MASK_4K; 833 entry->addr_mask = ~VTD_PAGE_MASK_4K; 834 entry->perm = IOMMU_WO; 835 return; 836 } else { 837 VTD_DPRINTF(GENERAL, "error: read request from interrupt address " 838 "gpa 0x%"PRIx64, addr); 839 vtd_report_dmar_fault(s, source_id, addr, VTD_FR_READ, is_write); 840 return; 841 } 842 } 843 /* Try to fetch slpte form IOTLB */ 844 iotlb_entry = vtd_lookup_iotlb(s, source_id, addr); 845 if (iotlb_entry) { 846 VTD_DPRINTF(CACHE, "hit iotlb sid 0x%"PRIx16 " gpa 0x%"PRIx64 847 " slpte 0x%"PRIx64 " did 0x%"PRIx16, source_id, addr, 848 iotlb_entry->slpte, iotlb_entry->domain_id); 849 slpte = iotlb_entry->slpte; 850 reads = iotlb_entry->read_flags; 851 writes = iotlb_entry->write_flags; 852 page_mask = iotlb_entry->mask; 853 goto out; 854 } 855 /* Try to fetch context-entry from cache first */ 856 if (cc_entry->context_cache_gen == s->context_cache_gen) { 857 VTD_DPRINTF(CACHE, "hit context-cache bus %d devfn %d " 858 "(hi %"PRIx64 " lo %"PRIx64 " gen %"PRIu32 ")", 859 bus_num, devfn, cc_entry->context_entry.hi, 860 cc_entry->context_entry.lo, cc_entry->context_cache_gen); 861 ce = cc_entry->context_entry; 862 is_fpd_set = ce.lo & VTD_CONTEXT_ENTRY_FPD; 863 } else { 864 ret_fr = vtd_dev_to_context_entry(s, bus_num, devfn, &ce); 865 is_fpd_set = ce.lo & VTD_CONTEXT_ENTRY_FPD; 866 if (ret_fr) { 867 ret_fr = -ret_fr; 868 if (is_fpd_set && vtd_is_qualified_fault(ret_fr)) { 869 VTD_DPRINTF(FLOG, "fault processing is disabled for DMA " 870 "requests through this context-entry " 871 "(with FPD Set)"); 872 } else { 873 vtd_report_dmar_fault(s, source_id, addr, ret_fr, is_write); 874 } 875 return; 876 } 877 /* Update context-cache */ 878 VTD_DPRINTF(CACHE, "update context-cache bus %d devfn %d " 879 "(hi %"PRIx64 " lo %"PRIx64 " gen %"PRIu32 "->%"PRIu32 ")", 880 bus_num, devfn, ce.hi, ce.lo, 881 cc_entry->context_cache_gen, s->context_cache_gen); 882 cc_entry->context_entry = ce; 883 cc_entry->context_cache_gen = s->context_cache_gen; 884 } 885 886 ret_fr = vtd_gpa_to_slpte(&ce, addr, is_write, &slpte, &level, 887 &reads, &writes); 888 if (ret_fr) { 889 ret_fr = -ret_fr; 890 if (is_fpd_set && vtd_is_qualified_fault(ret_fr)) { 891 VTD_DPRINTF(FLOG, "fault processing is disabled for DMA requests " 892 "through this context-entry (with FPD Set)"); 893 } else { 894 vtd_report_dmar_fault(s, source_id, addr, ret_fr, is_write); 895 } 896 return; 897 } 898 899 page_mask = vtd_slpt_level_page_mask(level); 900 vtd_update_iotlb(s, source_id, VTD_CONTEXT_ENTRY_DID(ce.hi), addr, slpte, 901 reads, writes, level); 902 out: 903 entry->iova = addr & page_mask; 904 entry->translated_addr = vtd_get_slpte_addr(slpte) & page_mask; 905 entry->addr_mask = ~page_mask; 906 entry->perm = (writes ? 2 : 0) + (reads ? 1 : 0); 907 } 908 909 static void vtd_root_table_setup(IntelIOMMUState *s) 910 { 911 s->root = vtd_get_quad_raw(s, DMAR_RTADDR_REG); 912 s->root_extended = s->root & VTD_RTADDR_RTT; 913 s->root &= VTD_RTADDR_ADDR_MASK; 914 915 VTD_DPRINTF(CSR, "root_table addr 0x%"PRIx64 " %s", s->root, 916 (s->root_extended ? "(extended)" : "")); 917 } 918 919 static void vtd_iec_notify_all(IntelIOMMUState *s, bool global, 920 uint32_t index, uint32_t mask) 921 { 922 x86_iommu_iec_notify_all(X86_IOMMU_DEVICE(s), global, index, mask); 923 } 924 925 static void vtd_interrupt_remap_table_setup(IntelIOMMUState *s) 926 { 927 uint64_t value = 0; 928 value = vtd_get_quad_raw(s, DMAR_IRTA_REG); 929 s->intr_size = 1UL << ((value & VTD_IRTA_SIZE_MASK) + 1); 930 s->intr_root = value & VTD_IRTA_ADDR_MASK; 931 s->intr_eime = value & VTD_IRTA_EIME; 932 933 /* Notify global invalidation */ 934 vtd_iec_notify_all(s, true, 0, 0); 935 936 VTD_DPRINTF(CSR, "int remap table addr 0x%"PRIx64 " size %"PRIu32, 937 s->intr_root, s->intr_size); 938 } 939 940 static void vtd_context_global_invalidate(IntelIOMMUState *s) 941 { 942 s->context_cache_gen++; 943 if (s->context_cache_gen == VTD_CONTEXT_CACHE_GEN_MAX) { 944 vtd_reset_context_cache(s); 945 } 946 } 947 948 949 /* Find the VTD address space currently associated with a given bus number, 950 */ 951 static VTDBus *vtd_find_as_from_bus_num(IntelIOMMUState *s, uint8_t bus_num) 952 { 953 VTDBus *vtd_bus = s->vtd_as_by_bus_num[bus_num]; 954 if (!vtd_bus) { 955 /* Iterate over the registered buses to find the one 956 * which currently hold this bus number, and update the bus_num lookup table: 957 */ 958 GHashTableIter iter; 959 960 g_hash_table_iter_init(&iter, s->vtd_as_by_busptr); 961 while (g_hash_table_iter_next (&iter, NULL, (void**)&vtd_bus)) { 962 if (pci_bus_num(vtd_bus->bus) == bus_num) { 963 s->vtd_as_by_bus_num[bus_num] = vtd_bus; 964 return vtd_bus; 965 } 966 } 967 } 968 return vtd_bus; 969 } 970 971 /* Do a context-cache device-selective invalidation. 972 * @func_mask: FM field after shifting 973 */ 974 static void vtd_context_device_invalidate(IntelIOMMUState *s, 975 uint16_t source_id, 976 uint16_t func_mask) 977 { 978 uint16_t mask; 979 VTDBus *vtd_bus; 980 VTDAddressSpace *vtd_as; 981 uint16_t devfn; 982 uint16_t devfn_it; 983 984 switch (func_mask & 3) { 985 case 0: 986 mask = 0; /* No bits in the SID field masked */ 987 break; 988 case 1: 989 mask = 4; /* Mask bit 2 in the SID field */ 990 break; 991 case 2: 992 mask = 6; /* Mask bit 2:1 in the SID field */ 993 break; 994 case 3: 995 mask = 7; /* Mask bit 2:0 in the SID field */ 996 break; 997 } 998 mask = ~mask; 999 VTD_DPRINTF(INV, "device-selective invalidation source 0x%"PRIx16 1000 " mask %"PRIu16, source_id, mask); 1001 vtd_bus = vtd_find_as_from_bus_num(s, VTD_SID_TO_BUS(source_id)); 1002 if (vtd_bus) { 1003 devfn = VTD_SID_TO_DEVFN(source_id); 1004 for (devfn_it = 0; devfn_it < X86_IOMMU_PCI_DEVFN_MAX; ++devfn_it) { 1005 vtd_as = vtd_bus->dev_as[devfn_it]; 1006 if (vtd_as && ((devfn_it & mask) == (devfn & mask))) { 1007 VTD_DPRINTF(INV, "invalidate context-cahce of devfn 0x%"PRIx16, 1008 devfn_it); 1009 vtd_as->context_cache_entry.context_cache_gen = 0; 1010 } 1011 } 1012 } 1013 } 1014 1015 /* Context-cache invalidation 1016 * Returns the Context Actual Invalidation Granularity. 1017 * @val: the content of the CCMD_REG 1018 */ 1019 static uint64_t vtd_context_cache_invalidate(IntelIOMMUState *s, uint64_t val) 1020 { 1021 uint64_t caig; 1022 uint64_t type = val & VTD_CCMD_CIRG_MASK; 1023 1024 switch (type) { 1025 case VTD_CCMD_DOMAIN_INVL: 1026 VTD_DPRINTF(INV, "domain-selective invalidation domain 0x%"PRIx16, 1027 (uint16_t)VTD_CCMD_DID(val)); 1028 /* Fall through */ 1029 case VTD_CCMD_GLOBAL_INVL: 1030 VTD_DPRINTF(INV, "global invalidation"); 1031 caig = VTD_CCMD_GLOBAL_INVL_A; 1032 vtd_context_global_invalidate(s); 1033 break; 1034 1035 case VTD_CCMD_DEVICE_INVL: 1036 caig = VTD_CCMD_DEVICE_INVL_A; 1037 vtd_context_device_invalidate(s, VTD_CCMD_SID(val), VTD_CCMD_FM(val)); 1038 break; 1039 1040 default: 1041 VTD_DPRINTF(GENERAL, "error: invalid granularity"); 1042 caig = 0; 1043 } 1044 return caig; 1045 } 1046 1047 static void vtd_iotlb_global_invalidate(IntelIOMMUState *s) 1048 { 1049 vtd_reset_iotlb(s); 1050 } 1051 1052 static void vtd_iotlb_domain_invalidate(IntelIOMMUState *s, uint16_t domain_id) 1053 { 1054 g_hash_table_foreach_remove(s->iotlb, vtd_hash_remove_by_domain, 1055 &domain_id); 1056 } 1057 1058 static void vtd_iotlb_page_invalidate(IntelIOMMUState *s, uint16_t domain_id, 1059 hwaddr addr, uint8_t am) 1060 { 1061 VTDIOTLBPageInvInfo info; 1062 1063 assert(am <= VTD_MAMV); 1064 info.domain_id = domain_id; 1065 info.addr = addr; 1066 info.mask = ~((1 << am) - 1); 1067 g_hash_table_foreach_remove(s->iotlb, vtd_hash_remove_by_page, &info); 1068 } 1069 1070 /* Flush IOTLB 1071 * Returns the IOTLB Actual Invalidation Granularity. 1072 * @val: the content of the IOTLB_REG 1073 */ 1074 static uint64_t vtd_iotlb_flush(IntelIOMMUState *s, uint64_t val) 1075 { 1076 uint64_t iaig; 1077 uint64_t type = val & VTD_TLB_FLUSH_GRANU_MASK; 1078 uint16_t domain_id; 1079 hwaddr addr; 1080 uint8_t am; 1081 1082 switch (type) { 1083 case VTD_TLB_GLOBAL_FLUSH: 1084 VTD_DPRINTF(INV, "global invalidation"); 1085 iaig = VTD_TLB_GLOBAL_FLUSH_A; 1086 vtd_iotlb_global_invalidate(s); 1087 break; 1088 1089 case VTD_TLB_DSI_FLUSH: 1090 domain_id = VTD_TLB_DID(val); 1091 VTD_DPRINTF(INV, "domain-selective invalidation domain 0x%"PRIx16, 1092 domain_id); 1093 iaig = VTD_TLB_DSI_FLUSH_A; 1094 vtd_iotlb_domain_invalidate(s, domain_id); 1095 break; 1096 1097 case VTD_TLB_PSI_FLUSH: 1098 domain_id = VTD_TLB_DID(val); 1099 addr = vtd_get_quad_raw(s, DMAR_IVA_REG); 1100 am = VTD_IVA_AM(addr); 1101 addr = VTD_IVA_ADDR(addr); 1102 VTD_DPRINTF(INV, "page-selective invalidation domain 0x%"PRIx16 1103 " addr 0x%"PRIx64 " mask %"PRIu8, domain_id, addr, am); 1104 if (am > VTD_MAMV) { 1105 VTD_DPRINTF(GENERAL, "error: supported max address mask value is " 1106 "%"PRIu8, (uint8_t)VTD_MAMV); 1107 iaig = 0; 1108 break; 1109 } 1110 iaig = VTD_TLB_PSI_FLUSH_A; 1111 vtd_iotlb_page_invalidate(s, domain_id, addr, am); 1112 break; 1113 1114 default: 1115 VTD_DPRINTF(GENERAL, "error: invalid granularity"); 1116 iaig = 0; 1117 } 1118 return iaig; 1119 } 1120 1121 static inline bool vtd_queued_inv_enable_check(IntelIOMMUState *s) 1122 { 1123 return s->iq_tail == 0; 1124 } 1125 1126 static inline bool vtd_queued_inv_disable_check(IntelIOMMUState *s) 1127 { 1128 return s->qi_enabled && (s->iq_tail == s->iq_head) && 1129 (s->iq_last_desc_type == VTD_INV_DESC_WAIT); 1130 } 1131 1132 static void vtd_handle_gcmd_qie(IntelIOMMUState *s, bool en) 1133 { 1134 uint64_t iqa_val = vtd_get_quad_raw(s, DMAR_IQA_REG); 1135 1136 VTD_DPRINTF(INV, "Queued Invalidation Enable %s", (en ? "on" : "off")); 1137 if (en) { 1138 if (vtd_queued_inv_enable_check(s)) { 1139 s->iq = iqa_val & VTD_IQA_IQA_MASK; 1140 /* 2^(x+8) entries */ 1141 s->iq_size = 1UL << ((iqa_val & VTD_IQA_QS) + 8); 1142 s->qi_enabled = true; 1143 VTD_DPRINTF(INV, "DMAR_IQA_REG 0x%"PRIx64, iqa_val); 1144 VTD_DPRINTF(INV, "Invalidation Queue addr 0x%"PRIx64 " size %d", 1145 s->iq, s->iq_size); 1146 /* Ok - report back to driver */ 1147 vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_QIES); 1148 } else { 1149 VTD_DPRINTF(GENERAL, "error: can't enable Queued Invalidation: " 1150 "tail %"PRIu16, s->iq_tail); 1151 } 1152 } else { 1153 if (vtd_queued_inv_disable_check(s)) { 1154 /* disable Queued Invalidation */ 1155 vtd_set_quad_raw(s, DMAR_IQH_REG, 0); 1156 s->iq_head = 0; 1157 s->qi_enabled = false; 1158 /* Ok - report back to driver */ 1159 vtd_set_clear_mask_long(s, DMAR_GSTS_REG, VTD_GSTS_QIES, 0); 1160 } else { 1161 VTD_DPRINTF(GENERAL, "error: can't disable Queued Invalidation: " 1162 "head %"PRIu16 ", tail %"PRIu16 1163 ", last_descriptor %"PRIu8, 1164 s->iq_head, s->iq_tail, s->iq_last_desc_type); 1165 } 1166 } 1167 } 1168 1169 /* Set Root Table Pointer */ 1170 static void vtd_handle_gcmd_srtp(IntelIOMMUState *s) 1171 { 1172 VTD_DPRINTF(CSR, "set Root Table Pointer"); 1173 1174 vtd_root_table_setup(s); 1175 /* Ok - report back to driver */ 1176 vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_RTPS); 1177 } 1178 1179 /* Set Interrupt Remap Table Pointer */ 1180 static void vtd_handle_gcmd_sirtp(IntelIOMMUState *s) 1181 { 1182 VTD_DPRINTF(CSR, "set Interrupt Remap Table Pointer"); 1183 1184 vtd_interrupt_remap_table_setup(s); 1185 /* Ok - report back to driver */ 1186 vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_IRTPS); 1187 } 1188 1189 /* Handle Translation Enable/Disable */ 1190 static void vtd_handle_gcmd_te(IntelIOMMUState *s, bool en) 1191 { 1192 VTD_DPRINTF(CSR, "Translation Enable %s", (en ? "on" : "off")); 1193 1194 if (en) { 1195 s->dmar_enabled = true; 1196 /* Ok - report back to driver */ 1197 vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_TES); 1198 } else { 1199 s->dmar_enabled = false; 1200 1201 /* Clear the index of Fault Recording Register */ 1202 s->next_frcd_reg = 0; 1203 /* Ok - report back to driver */ 1204 vtd_set_clear_mask_long(s, DMAR_GSTS_REG, VTD_GSTS_TES, 0); 1205 } 1206 } 1207 1208 /* Handle Interrupt Remap Enable/Disable */ 1209 static void vtd_handle_gcmd_ire(IntelIOMMUState *s, bool en) 1210 { 1211 VTD_DPRINTF(CSR, "Interrupt Remap Enable %s", (en ? "on" : "off")); 1212 1213 if (en) { 1214 s->intr_enabled = true; 1215 /* Ok - report back to driver */ 1216 vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_IRES); 1217 } else { 1218 s->intr_enabled = false; 1219 /* Ok - report back to driver */ 1220 vtd_set_clear_mask_long(s, DMAR_GSTS_REG, VTD_GSTS_IRES, 0); 1221 } 1222 } 1223 1224 /* Handle write to Global Command Register */ 1225 static void vtd_handle_gcmd_write(IntelIOMMUState *s) 1226 { 1227 uint32_t status = vtd_get_long_raw(s, DMAR_GSTS_REG); 1228 uint32_t val = vtd_get_long_raw(s, DMAR_GCMD_REG); 1229 uint32_t changed = status ^ val; 1230 1231 VTD_DPRINTF(CSR, "value 0x%"PRIx32 " status 0x%"PRIx32, val, status); 1232 if (changed & VTD_GCMD_TE) { 1233 /* Translation enable/disable */ 1234 vtd_handle_gcmd_te(s, val & VTD_GCMD_TE); 1235 } 1236 if (val & VTD_GCMD_SRTP) { 1237 /* Set/update the root-table pointer */ 1238 vtd_handle_gcmd_srtp(s); 1239 } 1240 if (changed & VTD_GCMD_QIE) { 1241 /* Queued Invalidation Enable */ 1242 vtd_handle_gcmd_qie(s, val & VTD_GCMD_QIE); 1243 } 1244 if (val & VTD_GCMD_SIRTP) { 1245 /* Set/update the interrupt remapping root-table pointer */ 1246 vtd_handle_gcmd_sirtp(s); 1247 } 1248 if (changed & VTD_GCMD_IRE) { 1249 /* Interrupt remap enable/disable */ 1250 vtd_handle_gcmd_ire(s, val & VTD_GCMD_IRE); 1251 } 1252 } 1253 1254 /* Handle write to Context Command Register */ 1255 static void vtd_handle_ccmd_write(IntelIOMMUState *s) 1256 { 1257 uint64_t ret; 1258 uint64_t val = vtd_get_quad_raw(s, DMAR_CCMD_REG); 1259 1260 /* Context-cache invalidation request */ 1261 if (val & VTD_CCMD_ICC) { 1262 if (s->qi_enabled) { 1263 VTD_DPRINTF(GENERAL, "error: Queued Invalidation enabled, " 1264 "should not use register-based invalidation"); 1265 return; 1266 } 1267 ret = vtd_context_cache_invalidate(s, val); 1268 /* Invalidation completed. Change something to show */ 1269 vtd_set_clear_mask_quad(s, DMAR_CCMD_REG, VTD_CCMD_ICC, 0ULL); 1270 ret = vtd_set_clear_mask_quad(s, DMAR_CCMD_REG, VTD_CCMD_CAIG_MASK, 1271 ret); 1272 VTD_DPRINTF(INV, "CCMD_REG write-back val: 0x%"PRIx64, ret); 1273 } 1274 } 1275 1276 /* Handle write to IOTLB Invalidation Register */ 1277 static void vtd_handle_iotlb_write(IntelIOMMUState *s) 1278 { 1279 uint64_t ret; 1280 uint64_t val = vtd_get_quad_raw(s, DMAR_IOTLB_REG); 1281 1282 /* IOTLB invalidation request */ 1283 if (val & VTD_TLB_IVT) { 1284 if (s->qi_enabled) { 1285 VTD_DPRINTF(GENERAL, "error: Queued Invalidation enabled, " 1286 "should not use register-based invalidation"); 1287 return; 1288 } 1289 ret = vtd_iotlb_flush(s, val); 1290 /* Invalidation completed. Change something to show */ 1291 vtd_set_clear_mask_quad(s, DMAR_IOTLB_REG, VTD_TLB_IVT, 0ULL); 1292 ret = vtd_set_clear_mask_quad(s, DMAR_IOTLB_REG, 1293 VTD_TLB_FLUSH_GRANU_MASK_A, ret); 1294 VTD_DPRINTF(INV, "IOTLB_REG write-back val: 0x%"PRIx64, ret); 1295 } 1296 } 1297 1298 /* Fetch an Invalidation Descriptor from the Invalidation Queue */ 1299 static bool vtd_get_inv_desc(dma_addr_t base_addr, uint32_t offset, 1300 VTDInvDesc *inv_desc) 1301 { 1302 dma_addr_t addr = base_addr + offset * sizeof(*inv_desc); 1303 if (dma_memory_read(&address_space_memory, addr, inv_desc, 1304 sizeof(*inv_desc))) { 1305 VTD_DPRINTF(GENERAL, "error: fail to fetch Invalidation Descriptor " 1306 "base_addr 0x%"PRIx64 " offset %"PRIu32, base_addr, offset); 1307 inv_desc->lo = 0; 1308 inv_desc->hi = 0; 1309 1310 return false; 1311 } 1312 inv_desc->lo = le64_to_cpu(inv_desc->lo); 1313 inv_desc->hi = le64_to_cpu(inv_desc->hi); 1314 return true; 1315 } 1316 1317 static bool vtd_process_wait_desc(IntelIOMMUState *s, VTDInvDesc *inv_desc) 1318 { 1319 if ((inv_desc->hi & VTD_INV_DESC_WAIT_RSVD_HI) || 1320 (inv_desc->lo & VTD_INV_DESC_WAIT_RSVD_LO)) { 1321 VTD_DPRINTF(GENERAL, "error: non-zero reserved field in Invalidation " 1322 "Wait Descriptor hi 0x%"PRIx64 " lo 0x%"PRIx64, 1323 inv_desc->hi, inv_desc->lo); 1324 return false; 1325 } 1326 if (inv_desc->lo & VTD_INV_DESC_WAIT_SW) { 1327 /* Status Write */ 1328 uint32_t status_data = (uint32_t)(inv_desc->lo >> 1329 VTD_INV_DESC_WAIT_DATA_SHIFT); 1330 1331 assert(!(inv_desc->lo & VTD_INV_DESC_WAIT_IF)); 1332 1333 /* FIXME: need to be masked with HAW? */ 1334 dma_addr_t status_addr = inv_desc->hi; 1335 VTD_DPRINTF(INV, "status data 0x%x, status addr 0x%"PRIx64, 1336 status_data, status_addr); 1337 status_data = cpu_to_le32(status_data); 1338 if (dma_memory_write(&address_space_memory, status_addr, &status_data, 1339 sizeof(status_data))) { 1340 VTD_DPRINTF(GENERAL, "error: fail to perform a coherent write"); 1341 return false; 1342 } 1343 } else if (inv_desc->lo & VTD_INV_DESC_WAIT_IF) { 1344 /* Interrupt flag */ 1345 VTD_DPRINTF(INV, "Invalidation Wait Descriptor interrupt completion"); 1346 vtd_generate_completion_event(s); 1347 } else { 1348 VTD_DPRINTF(GENERAL, "error: invalid Invalidation Wait Descriptor: " 1349 "hi 0x%"PRIx64 " lo 0x%"PRIx64, inv_desc->hi, inv_desc->lo); 1350 return false; 1351 } 1352 return true; 1353 } 1354 1355 static bool vtd_process_context_cache_desc(IntelIOMMUState *s, 1356 VTDInvDesc *inv_desc) 1357 { 1358 if ((inv_desc->lo & VTD_INV_DESC_CC_RSVD) || inv_desc->hi) { 1359 VTD_DPRINTF(GENERAL, "error: non-zero reserved field in Context-cache " 1360 "Invalidate Descriptor"); 1361 return false; 1362 } 1363 switch (inv_desc->lo & VTD_INV_DESC_CC_G) { 1364 case VTD_INV_DESC_CC_DOMAIN: 1365 VTD_DPRINTF(INV, "domain-selective invalidation domain 0x%"PRIx16, 1366 (uint16_t)VTD_INV_DESC_CC_DID(inv_desc->lo)); 1367 /* Fall through */ 1368 case VTD_INV_DESC_CC_GLOBAL: 1369 VTD_DPRINTF(INV, "global invalidation"); 1370 vtd_context_global_invalidate(s); 1371 break; 1372 1373 case VTD_INV_DESC_CC_DEVICE: 1374 vtd_context_device_invalidate(s, VTD_INV_DESC_CC_SID(inv_desc->lo), 1375 VTD_INV_DESC_CC_FM(inv_desc->lo)); 1376 break; 1377 1378 default: 1379 VTD_DPRINTF(GENERAL, "error: invalid granularity in Context-cache " 1380 "Invalidate Descriptor hi 0x%"PRIx64 " lo 0x%"PRIx64, 1381 inv_desc->hi, inv_desc->lo); 1382 return false; 1383 } 1384 return true; 1385 } 1386 1387 static bool vtd_process_iotlb_desc(IntelIOMMUState *s, VTDInvDesc *inv_desc) 1388 { 1389 uint16_t domain_id; 1390 uint8_t am; 1391 hwaddr addr; 1392 1393 if ((inv_desc->lo & VTD_INV_DESC_IOTLB_RSVD_LO) || 1394 (inv_desc->hi & VTD_INV_DESC_IOTLB_RSVD_HI)) { 1395 VTD_DPRINTF(GENERAL, "error: non-zero reserved field in IOTLB " 1396 "Invalidate Descriptor hi 0x%"PRIx64 " lo 0x%"PRIx64, 1397 inv_desc->hi, inv_desc->lo); 1398 return false; 1399 } 1400 1401 switch (inv_desc->lo & VTD_INV_DESC_IOTLB_G) { 1402 case VTD_INV_DESC_IOTLB_GLOBAL: 1403 VTD_DPRINTF(INV, "global invalidation"); 1404 vtd_iotlb_global_invalidate(s); 1405 break; 1406 1407 case VTD_INV_DESC_IOTLB_DOMAIN: 1408 domain_id = VTD_INV_DESC_IOTLB_DID(inv_desc->lo); 1409 VTD_DPRINTF(INV, "domain-selective invalidation domain 0x%"PRIx16, 1410 domain_id); 1411 vtd_iotlb_domain_invalidate(s, domain_id); 1412 break; 1413 1414 case VTD_INV_DESC_IOTLB_PAGE: 1415 domain_id = VTD_INV_DESC_IOTLB_DID(inv_desc->lo); 1416 addr = VTD_INV_DESC_IOTLB_ADDR(inv_desc->hi); 1417 am = VTD_INV_DESC_IOTLB_AM(inv_desc->hi); 1418 VTD_DPRINTF(INV, "page-selective invalidation domain 0x%"PRIx16 1419 " addr 0x%"PRIx64 " mask %"PRIu8, domain_id, addr, am); 1420 if (am > VTD_MAMV) { 1421 VTD_DPRINTF(GENERAL, "error: supported max address mask value is " 1422 "%"PRIu8, (uint8_t)VTD_MAMV); 1423 return false; 1424 } 1425 vtd_iotlb_page_invalidate(s, domain_id, addr, am); 1426 break; 1427 1428 default: 1429 VTD_DPRINTF(GENERAL, "error: invalid granularity in IOTLB Invalidate " 1430 "Descriptor hi 0x%"PRIx64 " lo 0x%"PRIx64, 1431 inv_desc->hi, inv_desc->lo); 1432 return false; 1433 } 1434 return true; 1435 } 1436 1437 static bool vtd_process_inv_iec_desc(IntelIOMMUState *s, 1438 VTDInvDesc *inv_desc) 1439 { 1440 VTD_DPRINTF(INV, "inv ir glob %d index %d mask %d", 1441 inv_desc->iec.granularity, 1442 inv_desc->iec.index, 1443 inv_desc->iec.index_mask); 1444 1445 vtd_iec_notify_all(s, !inv_desc->iec.granularity, 1446 inv_desc->iec.index, 1447 inv_desc->iec.index_mask); 1448 return true; 1449 } 1450 1451 static bool vtd_process_device_iotlb_desc(IntelIOMMUState *s, 1452 VTDInvDesc *inv_desc) 1453 { 1454 VTDAddressSpace *vtd_dev_as; 1455 IOMMUTLBEntry entry; 1456 struct VTDBus *vtd_bus; 1457 hwaddr addr; 1458 uint64_t sz; 1459 uint16_t sid; 1460 uint8_t devfn; 1461 bool size; 1462 uint8_t bus_num; 1463 1464 addr = VTD_INV_DESC_DEVICE_IOTLB_ADDR(inv_desc->hi); 1465 sid = VTD_INV_DESC_DEVICE_IOTLB_SID(inv_desc->lo); 1466 devfn = sid & 0xff; 1467 bus_num = sid >> 8; 1468 size = VTD_INV_DESC_DEVICE_IOTLB_SIZE(inv_desc->hi); 1469 1470 if ((inv_desc->lo & VTD_INV_DESC_DEVICE_IOTLB_RSVD_LO) || 1471 (inv_desc->hi & VTD_INV_DESC_DEVICE_IOTLB_RSVD_HI)) { 1472 VTD_DPRINTF(GENERAL, "error: non-zero reserved field in Device " 1473 "IOTLB Invalidate Descriptor hi 0x%"PRIx64 " lo 0x%"PRIx64, 1474 inv_desc->hi, inv_desc->lo); 1475 return false; 1476 } 1477 1478 vtd_bus = vtd_find_as_from_bus_num(s, bus_num); 1479 if (!vtd_bus) { 1480 goto done; 1481 } 1482 1483 vtd_dev_as = vtd_bus->dev_as[devfn]; 1484 if (!vtd_dev_as) { 1485 goto done; 1486 } 1487 1488 if (size) { 1489 sz = 1 << (ctz64(~(addr | (VTD_PAGE_MASK_4K - 1))) + 1); 1490 addr &= ~(sz - 1); 1491 } else { 1492 sz = VTD_PAGE_SIZE; 1493 } 1494 1495 entry.target_as = &vtd_dev_as->as; 1496 entry.addr_mask = sz - 1; 1497 entry.iova = addr; 1498 entry.perm = IOMMU_NONE; 1499 entry.translated_addr = 0; 1500 memory_region_notify_iommu(entry.target_as->root, entry); 1501 1502 done: 1503 return true; 1504 } 1505 1506 static bool vtd_process_inv_desc(IntelIOMMUState *s) 1507 { 1508 VTDInvDesc inv_desc; 1509 uint8_t desc_type; 1510 1511 VTD_DPRINTF(INV, "iq head %"PRIu16, s->iq_head); 1512 if (!vtd_get_inv_desc(s->iq, s->iq_head, &inv_desc)) { 1513 s->iq_last_desc_type = VTD_INV_DESC_NONE; 1514 return false; 1515 } 1516 desc_type = inv_desc.lo & VTD_INV_DESC_TYPE; 1517 /* FIXME: should update at first or at last? */ 1518 s->iq_last_desc_type = desc_type; 1519 1520 switch (desc_type) { 1521 case VTD_INV_DESC_CC: 1522 VTD_DPRINTF(INV, "Context-cache Invalidate Descriptor hi 0x%"PRIx64 1523 " lo 0x%"PRIx64, inv_desc.hi, inv_desc.lo); 1524 if (!vtd_process_context_cache_desc(s, &inv_desc)) { 1525 return false; 1526 } 1527 break; 1528 1529 case VTD_INV_DESC_IOTLB: 1530 VTD_DPRINTF(INV, "IOTLB Invalidate Descriptor hi 0x%"PRIx64 1531 " lo 0x%"PRIx64, inv_desc.hi, inv_desc.lo); 1532 if (!vtd_process_iotlb_desc(s, &inv_desc)) { 1533 return false; 1534 } 1535 break; 1536 1537 case VTD_INV_DESC_WAIT: 1538 VTD_DPRINTF(INV, "Invalidation Wait Descriptor hi 0x%"PRIx64 1539 " lo 0x%"PRIx64, inv_desc.hi, inv_desc.lo); 1540 if (!vtd_process_wait_desc(s, &inv_desc)) { 1541 return false; 1542 } 1543 break; 1544 1545 case VTD_INV_DESC_IEC: 1546 VTD_DPRINTF(INV, "Invalidation Interrupt Entry Cache " 1547 "Descriptor hi 0x%"PRIx64 " lo 0x%"PRIx64, 1548 inv_desc.hi, inv_desc.lo); 1549 if (!vtd_process_inv_iec_desc(s, &inv_desc)) { 1550 return false; 1551 } 1552 break; 1553 1554 case VTD_INV_DESC_DEVICE: 1555 VTD_DPRINTF(INV, "Device IOTLB Invalidation Descriptor hi 0x%"PRIx64 1556 " lo 0x%"PRIx64, inv_desc.hi, inv_desc.lo); 1557 if (!vtd_process_device_iotlb_desc(s, &inv_desc)) { 1558 return false; 1559 } 1560 break; 1561 1562 default: 1563 VTD_DPRINTF(GENERAL, "error: unkonw Invalidation Descriptor type " 1564 "hi 0x%"PRIx64 " lo 0x%"PRIx64 " type %"PRIu8, 1565 inv_desc.hi, inv_desc.lo, desc_type); 1566 return false; 1567 } 1568 s->iq_head++; 1569 if (s->iq_head == s->iq_size) { 1570 s->iq_head = 0; 1571 } 1572 return true; 1573 } 1574 1575 /* Try to fetch and process more Invalidation Descriptors */ 1576 static void vtd_fetch_inv_desc(IntelIOMMUState *s) 1577 { 1578 VTD_DPRINTF(INV, "fetch Invalidation Descriptors"); 1579 if (s->iq_tail >= s->iq_size) { 1580 /* Detects an invalid Tail pointer */ 1581 VTD_DPRINTF(GENERAL, "error: iq_tail is %"PRIu16 1582 " while iq_size is %"PRIu16, s->iq_tail, s->iq_size); 1583 vtd_handle_inv_queue_error(s); 1584 return; 1585 } 1586 while (s->iq_head != s->iq_tail) { 1587 if (!vtd_process_inv_desc(s)) { 1588 /* Invalidation Queue Errors */ 1589 vtd_handle_inv_queue_error(s); 1590 break; 1591 } 1592 /* Must update the IQH_REG in time */ 1593 vtd_set_quad_raw(s, DMAR_IQH_REG, 1594 (((uint64_t)(s->iq_head)) << VTD_IQH_QH_SHIFT) & 1595 VTD_IQH_QH_MASK); 1596 } 1597 } 1598 1599 /* Handle write to Invalidation Queue Tail Register */ 1600 static void vtd_handle_iqt_write(IntelIOMMUState *s) 1601 { 1602 uint64_t val = vtd_get_quad_raw(s, DMAR_IQT_REG); 1603 1604 s->iq_tail = VTD_IQT_QT(val); 1605 VTD_DPRINTF(INV, "set iq tail %"PRIu16, s->iq_tail); 1606 if (s->qi_enabled && !(vtd_get_long_raw(s, DMAR_FSTS_REG) & VTD_FSTS_IQE)) { 1607 /* Process Invalidation Queue here */ 1608 vtd_fetch_inv_desc(s); 1609 } 1610 } 1611 1612 static void vtd_handle_fsts_write(IntelIOMMUState *s) 1613 { 1614 uint32_t fsts_reg = vtd_get_long_raw(s, DMAR_FSTS_REG); 1615 uint32_t fectl_reg = vtd_get_long_raw(s, DMAR_FECTL_REG); 1616 uint32_t status_fields = VTD_FSTS_PFO | VTD_FSTS_PPF | VTD_FSTS_IQE; 1617 1618 if ((fectl_reg & VTD_FECTL_IP) && !(fsts_reg & status_fields)) { 1619 vtd_set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0); 1620 VTD_DPRINTF(FLOG, "all pending interrupt conditions serviced, clear " 1621 "IP field of FECTL_REG"); 1622 } 1623 /* FIXME: when IQE is Clear, should we try to fetch some Invalidation 1624 * Descriptors if there are any when Queued Invalidation is enabled? 1625 */ 1626 } 1627 1628 static void vtd_handle_fectl_write(IntelIOMMUState *s) 1629 { 1630 uint32_t fectl_reg; 1631 /* FIXME: when software clears the IM field, check the IP field. But do we 1632 * need to compare the old value and the new value to conclude that 1633 * software clears the IM field? Or just check if the IM field is zero? 1634 */ 1635 fectl_reg = vtd_get_long_raw(s, DMAR_FECTL_REG); 1636 if ((fectl_reg & VTD_FECTL_IP) && !(fectl_reg & VTD_FECTL_IM)) { 1637 vtd_generate_interrupt(s, DMAR_FEADDR_REG, DMAR_FEDATA_REG); 1638 vtd_set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0); 1639 VTD_DPRINTF(FLOG, "IM field is cleared, generate " 1640 "fault event interrupt"); 1641 } 1642 } 1643 1644 static void vtd_handle_ics_write(IntelIOMMUState *s) 1645 { 1646 uint32_t ics_reg = vtd_get_long_raw(s, DMAR_ICS_REG); 1647 uint32_t iectl_reg = vtd_get_long_raw(s, DMAR_IECTL_REG); 1648 1649 if ((iectl_reg & VTD_IECTL_IP) && !(ics_reg & VTD_ICS_IWC)) { 1650 vtd_set_clear_mask_long(s, DMAR_IECTL_REG, VTD_IECTL_IP, 0); 1651 VTD_DPRINTF(INV, "pending completion interrupt condition serviced, " 1652 "clear IP field of IECTL_REG"); 1653 } 1654 } 1655 1656 static void vtd_handle_iectl_write(IntelIOMMUState *s) 1657 { 1658 uint32_t iectl_reg; 1659 /* FIXME: when software clears the IM field, check the IP field. But do we 1660 * need to compare the old value and the new value to conclude that 1661 * software clears the IM field? Or just check if the IM field is zero? 1662 */ 1663 iectl_reg = vtd_get_long_raw(s, DMAR_IECTL_REG); 1664 if ((iectl_reg & VTD_IECTL_IP) && !(iectl_reg & VTD_IECTL_IM)) { 1665 vtd_generate_interrupt(s, DMAR_IEADDR_REG, DMAR_IEDATA_REG); 1666 vtd_set_clear_mask_long(s, DMAR_IECTL_REG, VTD_IECTL_IP, 0); 1667 VTD_DPRINTF(INV, "IM field is cleared, generate " 1668 "invalidation event interrupt"); 1669 } 1670 } 1671 1672 static uint64_t vtd_mem_read(void *opaque, hwaddr addr, unsigned size) 1673 { 1674 IntelIOMMUState *s = opaque; 1675 uint64_t val; 1676 1677 if (addr + size > DMAR_REG_SIZE) { 1678 VTD_DPRINTF(GENERAL, "error: addr outside region: max 0x%"PRIx64 1679 ", got 0x%"PRIx64 " %d", 1680 (uint64_t)DMAR_REG_SIZE, addr, size); 1681 return (uint64_t)-1; 1682 } 1683 1684 switch (addr) { 1685 /* Root Table Address Register, 64-bit */ 1686 case DMAR_RTADDR_REG: 1687 if (size == 4) { 1688 val = s->root & ((1ULL << 32) - 1); 1689 } else { 1690 val = s->root; 1691 } 1692 break; 1693 1694 case DMAR_RTADDR_REG_HI: 1695 assert(size == 4); 1696 val = s->root >> 32; 1697 break; 1698 1699 /* Invalidation Queue Address Register, 64-bit */ 1700 case DMAR_IQA_REG: 1701 val = s->iq | (vtd_get_quad(s, DMAR_IQA_REG) & VTD_IQA_QS); 1702 if (size == 4) { 1703 val = val & ((1ULL << 32) - 1); 1704 } 1705 break; 1706 1707 case DMAR_IQA_REG_HI: 1708 assert(size == 4); 1709 val = s->iq >> 32; 1710 break; 1711 1712 default: 1713 if (size == 4) { 1714 val = vtd_get_long(s, addr); 1715 } else { 1716 val = vtd_get_quad(s, addr); 1717 } 1718 } 1719 VTD_DPRINTF(CSR, "addr 0x%"PRIx64 " size %d val 0x%"PRIx64, 1720 addr, size, val); 1721 return val; 1722 } 1723 1724 static void vtd_mem_write(void *opaque, hwaddr addr, 1725 uint64_t val, unsigned size) 1726 { 1727 IntelIOMMUState *s = opaque; 1728 1729 if (addr + size > DMAR_REG_SIZE) { 1730 VTD_DPRINTF(GENERAL, "error: addr outside region: max 0x%"PRIx64 1731 ", got 0x%"PRIx64 " %d", 1732 (uint64_t)DMAR_REG_SIZE, addr, size); 1733 return; 1734 } 1735 1736 switch (addr) { 1737 /* Global Command Register, 32-bit */ 1738 case DMAR_GCMD_REG: 1739 VTD_DPRINTF(CSR, "DMAR_GCMD_REG write addr 0x%"PRIx64 1740 ", size %d, val 0x%"PRIx64, addr, size, val); 1741 vtd_set_long(s, addr, val); 1742 vtd_handle_gcmd_write(s); 1743 break; 1744 1745 /* Context Command Register, 64-bit */ 1746 case DMAR_CCMD_REG: 1747 VTD_DPRINTF(CSR, "DMAR_CCMD_REG write addr 0x%"PRIx64 1748 ", size %d, val 0x%"PRIx64, addr, size, val); 1749 if (size == 4) { 1750 vtd_set_long(s, addr, val); 1751 } else { 1752 vtd_set_quad(s, addr, val); 1753 vtd_handle_ccmd_write(s); 1754 } 1755 break; 1756 1757 case DMAR_CCMD_REG_HI: 1758 VTD_DPRINTF(CSR, "DMAR_CCMD_REG_HI write addr 0x%"PRIx64 1759 ", size %d, val 0x%"PRIx64, addr, size, val); 1760 assert(size == 4); 1761 vtd_set_long(s, addr, val); 1762 vtd_handle_ccmd_write(s); 1763 break; 1764 1765 /* IOTLB Invalidation Register, 64-bit */ 1766 case DMAR_IOTLB_REG: 1767 VTD_DPRINTF(INV, "DMAR_IOTLB_REG write addr 0x%"PRIx64 1768 ", size %d, val 0x%"PRIx64, addr, size, val); 1769 if (size == 4) { 1770 vtd_set_long(s, addr, val); 1771 } else { 1772 vtd_set_quad(s, addr, val); 1773 vtd_handle_iotlb_write(s); 1774 } 1775 break; 1776 1777 case DMAR_IOTLB_REG_HI: 1778 VTD_DPRINTF(INV, "DMAR_IOTLB_REG_HI write addr 0x%"PRIx64 1779 ", size %d, val 0x%"PRIx64, addr, size, val); 1780 assert(size == 4); 1781 vtd_set_long(s, addr, val); 1782 vtd_handle_iotlb_write(s); 1783 break; 1784 1785 /* Invalidate Address Register, 64-bit */ 1786 case DMAR_IVA_REG: 1787 VTD_DPRINTF(INV, "DMAR_IVA_REG write addr 0x%"PRIx64 1788 ", size %d, val 0x%"PRIx64, addr, size, val); 1789 if (size == 4) { 1790 vtd_set_long(s, addr, val); 1791 } else { 1792 vtd_set_quad(s, addr, val); 1793 } 1794 break; 1795 1796 case DMAR_IVA_REG_HI: 1797 VTD_DPRINTF(INV, "DMAR_IVA_REG_HI write addr 0x%"PRIx64 1798 ", size %d, val 0x%"PRIx64, addr, size, val); 1799 assert(size == 4); 1800 vtd_set_long(s, addr, val); 1801 break; 1802 1803 /* Fault Status Register, 32-bit */ 1804 case DMAR_FSTS_REG: 1805 VTD_DPRINTF(FLOG, "DMAR_FSTS_REG write addr 0x%"PRIx64 1806 ", size %d, val 0x%"PRIx64, addr, size, val); 1807 assert(size == 4); 1808 vtd_set_long(s, addr, val); 1809 vtd_handle_fsts_write(s); 1810 break; 1811 1812 /* Fault Event Control Register, 32-bit */ 1813 case DMAR_FECTL_REG: 1814 VTD_DPRINTF(FLOG, "DMAR_FECTL_REG write addr 0x%"PRIx64 1815 ", size %d, val 0x%"PRIx64, addr, size, val); 1816 assert(size == 4); 1817 vtd_set_long(s, addr, val); 1818 vtd_handle_fectl_write(s); 1819 break; 1820 1821 /* Fault Event Data Register, 32-bit */ 1822 case DMAR_FEDATA_REG: 1823 VTD_DPRINTF(FLOG, "DMAR_FEDATA_REG write addr 0x%"PRIx64 1824 ", size %d, val 0x%"PRIx64, addr, size, val); 1825 assert(size == 4); 1826 vtd_set_long(s, addr, val); 1827 break; 1828 1829 /* Fault Event Address Register, 32-bit */ 1830 case DMAR_FEADDR_REG: 1831 VTD_DPRINTF(FLOG, "DMAR_FEADDR_REG write addr 0x%"PRIx64 1832 ", size %d, val 0x%"PRIx64, addr, size, val); 1833 assert(size == 4); 1834 vtd_set_long(s, addr, val); 1835 break; 1836 1837 /* Fault Event Upper Address Register, 32-bit */ 1838 case DMAR_FEUADDR_REG: 1839 VTD_DPRINTF(FLOG, "DMAR_FEUADDR_REG write addr 0x%"PRIx64 1840 ", size %d, val 0x%"PRIx64, addr, size, val); 1841 assert(size == 4); 1842 vtd_set_long(s, addr, val); 1843 break; 1844 1845 /* Protected Memory Enable Register, 32-bit */ 1846 case DMAR_PMEN_REG: 1847 VTD_DPRINTF(CSR, "DMAR_PMEN_REG write addr 0x%"PRIx64 1848 ", size %d, val 0x%"PRIx64, addr, size, val); 1849 assert(size == 4); 1850 vtd_set_long(s, addr, val); 1851 break; 1852 1853 /* Root Table Address Register, 64-bit */ 1854 case DMAR_RTADDR_REG: 1855 VTD_DPRINTF(CSR, "DMAR_RTADDR_REG write addr 0x%"PRIx64 1856 ", size %d, val 0x%"PRIx64, addr, size, val); 1857 if (size == 4) { 1858 vtd_set_long(s, addr, val); 1859 } else { 1860 vtd_set_quad(s, addr, val); 1861 } 1862 break; 1863 1864 case DMAR_RTADDR_REG_HI: 1865 VTD_DPRINTF(CSR, "DMAR_RTADDR_REG_HI write addr 0x%"PRIx64 1866 ", size %d, val 0x%"PRIx64, addr, size, val); 1867 assert(size == 4); 1868 vtd_set_long(s, addr, val); 1869 break; 1870 1871 /* Invalidation Queue Tail Register, 64-bit */ 1872 case DMAR_IQT_REG: 1873 VTD_DPRINTF(INV, "DMAR_IQT_REG write addr 0x%"PRIx64 1874 ", size %d, val 0x%"PRIx64, addr, size, val); 1875 if (size == 4) { 1876 vtd_set_long(s, addr, val); 1877 } else { 1878 vtd_set_quad(s, addr, val); 1879 } 1880 vtd_handle_iqt_write(s); 1881 break; 1882 1883 case DMAR_IQT_REG_HI: 1884 VTD_DPRINTF(INV, "DMAR_IQT_REG_HI write addr 0x%"PRIx64 1885 ", size %d, val 0x%"PRIx64, addr, size, val); 1886 assert(size == 4); 1887 vtd_set_long(s, addr, val); 1888 /* 19:63 of IQT_REG is RsvdZ, do nothing here */ 1889 break; 1890 1891 /* Invalidation Queue Address Register, 64-bit */ 1892 case DMAR_IQA_REG: 1893 VTD_DPRINTF(INV, "DMAR_IQA_REG write addr 0x%"PRIx64 1894 ", size %d, val 0x%"PRIx64, addr, size, val); 1895 if (size == 4) { 1896 vtd_set_long(s, addr, val); 1897 } else { 1898 vtd_set_quad(s, addr, val); 1899 } 1900 break; 1901 1902 case DMAR_IQA_REG_HI: 1903 VTD_DPRINTF(INV, "DMAR_IQA_REG_HI write addr 0x%"PRIx64 1904 ", size %d, val 0x%"PRIx64, addr, size, val); 1905 assert(size == 4); 1906 vtd_set_long(s, addr, val); 1907 break; 1908 1909 /* Invalidation Completion Status Register, 32-bit */ 1910 case DMAR_ICS_REG: 1911 VTD_DPRINTF(INV, "DMAR_ICS_REG write addr 0x%"PRIx64 1912 ", size %d, val 0x%"PRIx64, addr, size, val); 1913 assert(size == 4); 1914 vtd_set_long(s, addr, val); 1915 vtd_handle_ics_write(s); 1916 break; 1917 1918 /* Invalidation Event Control Register, 32-bit */ 1919 case DMAR_IECTL_REG: 1920 VTD_DPRINTF(INV, "DMAR_IECTL_REG write addr 0x%"PRIx64 1921 ", size %d, val 0x%"PRIx64, addr, size, val); 1922 assert(size == 4); 1923 vtd_set_long(s, addr, val); 1924 vtd_handle_iectl_write(s); 1925 break; 1926 1927 /* Invalidation Event Data Register, 32-bit */ 1928 case DMAR_IEDATA_REG: 1929 VTD_DPRINTF(INV, "DMAR_IEDATA_REG write addr 0x%"PRIx64 1930 ", size %d, val 0x%"PRIx64, addr, size, val); 1931 assert(size == 4); 1932 vtd_set_long(s, addr, val); 1933 break; 1934 1935 /* Invalidation Event Address Register, 32-bit */ 1936 case DMAR_IEADDR_REG: 1937 VTD_DPRINTF(INV, "DMAR_IEADDR_REG write addr 0x%"PRIx64 1938 ", size %d, val 0x%"PRIx64, addr, size, val); 1939 assert(size == 4); 1940 vtd_set_long(s, addr, val); 1941 break; 1942 1943 /* Invalidation Event Upper Address Register, 32-bit */ 1944 case DMAR_IEUADDR_REG: 1945 VTD_DPRINTF(INV, "DMAR_IEUADDR_REG write addr 0x%"PRIx64 1946 ", size %d, val 0x%"PRIx64, addr, size, val); 1947 assert(size == 4); 1948 vtd_set_long(s, addr, val); 1949 break; 1950 1951 /* Fault Recording Registers, 128-bit */ 1952 case DMAR_FRCD_REG_0_0: 1953 VTD_DPRINTF(FLOG, "DMAR_FRCD_REG_0_0 write addr 0x%"PRIx64 1954 ", size %d, val 0x%"PRIx64, addr, size, val); 1955 if (size == 4) { 1956 vtd_set_long(s, addr, val); 1957 } else { 1958 vtd_set_quad(s, addr, val); 1959 } 1960 break; 1961 1962 case DMAR_FRCD_REG_0_1: 1963 VTD_DPRINTF(FLOG, "DMAR_FRCD_REG_0_1 write addr 0x%"PRIx64 1964 ", size %d, val 0x%"PRIx64, addr, size, val); 1965 assert(size == 4); 1966 vtd_set_long(s, addr, val); 1967 break; 1968 1969 case DMAR_FRCD_REG_0_2: 1970 VTD_DPRINTF(FLOG, "DMAR_FRCD_REG_0_2 write addr 0x%"PRIx64 1971 ", size %d, val 0x%"PRIx64, addr, size, val); 1972 if (size == 4) { 1973 vtd_set_long(s, addr, val); 1974 } else { 1975 vtd_set_quad(s, addr, val); 1976 /* May clear bit 127 (Fault), update PPF */ 1977 vtd_update_fsts_ppf(s); 1978 } 1979 break; 1980 1981 case DMAR_FRCD_REG_0_3: 1982 VTD_DPRINTF(FLOG, "DMAR_FRCD_REG_0_3 write addr 0x%"PRIx64 1983 ", size %d, val 0x%"PRIx64, addr, size, val); 1984 assert(size == 4); 1985 vtd_set_long(s, addr, val); 1986 /* May clear bit 127 (Fault), update PPF */ 1987 vtd_update_fsts_ppf(s); 1988 break; 1989 1990 case DMAR_IRTA_REG: 1991 VTD_DPRINTF(IR, "DMAR_IRTA_REG write addr 0x%"PRIx64 1992 ", size %d, val 0x%"PRIx64, addr, size, val); 1993 if (size == 4) { 1994 vtd_set_long(s, addr, val); 1995 } else { 1996 vtd_set_quad(s, addr, val); 1997 } 1998 break; 1999 2000 case DMAR_IRTA_REG_HI: 2001 VTD_DPRINTF(IR, "DMAR_IRTA_REG_HI write addr 0x%"PRIx64 2002 ", size %d, val 0x%"PRIx64, addr, size, val); 2003 assert(size == 4); 2004 vtd_set_long(s, addr, val); 2005 break; 2006 2007 default: 2008 VTD_DPRINTF(GENERAL, "error: unhandled reg write addr 0x%"PRIx64 2009 ", size %d, val 0x%"PRIx64, addr, size, val); 2010 if (size == 4) { 2011 vtd_set_long(s, addr, val); 2012 } else { 2013 vtd_set_quad(s, addr, val); 2014 } 2015 } 2016 } 2017 2018 static IOMMUTLBEntry vtd_iommu_translate(MemoryRegion *iommu, hwaddr addr, 2019 bool is_write) 2020 { 2021 VTDAddressSpace *vtd_as = container_of(iommu, VTDAddressSpace, iommu); 2022 IntelIOMMUState *s = vtd_as->iommu_state; 2023 IOMMUTLBEntry ret = { 2024 .target_as = &address_space_memory, 2025 .iova = addr, 2026 .translated_addr = 0, 2027 .addr_mask = ~(hwaddr)0, 2028 .perm = IOMMU_NONE, 2029 }; 2030 2031 if (!s->dmar_enabled) { 2032 /* DMAR disabled, passthrough, use 4k-page*/ 2033 ret.iova = addr & VTD_PAGE_MASK_4K; 2034 ret.translated_addr = addr & VTD_PAGE_MASK_4K; 2035 ret.addr_mask = ~VTD_PAGE_MASK_4K; 2036 ret.perm = IOMMU_RW; 2037 return ret; 2038 } 2039 2040 vtd_do_iommu_translate(vtd_as, vtd_as->bus, vtd_as->devfn, addr, 2041 is_write, &ret); 2042 VTD_DPRINTF(MMU, 2043 "bus %"PRIu8 " slot %"PRIu8 " func %"PRIu8 " devfn %"PRIu8 2044 " gpa 0x%"PRIx64 " hpa 0x%"PRIx64, pci_bus_num(vtd_as->bus), 2045 VTD_PCI_SLOT(vtd_as->devfn), VTD_PCI_FUNC(vtd_as->devfn), 2046 vtd_as->devfn, addr, ret.translated_addr); 2047 return ret; 2048 } 2049 2050 static void vtd_iommu_notify_flag_changed(MemoryRegion *iommu, 2051 IOMMUNotifierFlag old, 2052 IOMMUNotifierFlag new) 2053 { 2054 VTDAddressSpace *vtd_as = container_of(iommu, VTDAddressSpace, iommu); 2055 2056 if (new & IOMMU_NOTIFIER_MAP) { 2057 error_report("Device at bus %s addr %02x.%d requires iommu " 2058 "notifier which is currently not supported by " 2059 "intel-iommu emulation", 2060 vtd_as->bus->qbus.name, PCI_SLOT(vtd_as->devfn), 2061 PCI_FUNC(vtd_as->devfn)); 2062 exit(1); 2063 } 2064 } 2065 2066 static const VMStateDescription vtd_vmstate = { 2067 .name = "iommu-intel", 2068 .version_id = 1, 2069 .minimum_version_id = 1, 2070 .priority = MIG_PRI_IOMMU, 2071 .fields = (VMStateField[]) { 2072 VMSTATE_UINT64(root, IntelIOMMUState), 2073 VMSTATE_UINT64(intr_root, IntelIOMMUState), 2074 VMSTATE_UINT64(iq, IntelIOMMUState), 2075 VMSTATE_UINT32(intr_size, IntelIOMMUState), 2076 VMSTATE_UINT16(iq_head, IntelIOMMUState), 2077 VMSTATE_UINT16(iq_tail, IntelIOMMUState), 2078 VMSTATE_UINT16(iq_size, IntelIOMMUState), 2079 VMSTATE_UINT16(next_frcd_reg, IntelIOMMUState), 2080 VMSTATE_UINT8_ARRAY(csr, IntelIOMMUState, DMAR_REG_SIZE), 2081 VMSTATE_UINT8(iq_last_desc_type, IntelIOMMUState), 2082 VMSTATE_BOOL(root_extended, IntelIOMMUState), 2083 VMSTATE_BOOL(dmar_enabled, IntelIOMMUState), 2084 VMSTATE_BOOL(qi_enabled, IntelIOMMUState), 2085 VMSTATE_BOOL(intr_enabled, IntelIOMMUState), 2086 VMSTATE_BOOL(intr_eime, IntelIOMMUState), 2087 VMSTATE_END_OF_LIST() 2088 } 2089 }; 2090 2091 static const MemoryRegionOps vtd_mem_ops = { 2092 .read = vtd_mem_read, 2093 .write = vtd_mem_write, 2094 .endianness = DEVICE_LITTLE_ENDIAN, 2095 .impl = { 2096 .min_access_size = 4, 2097 .max_access_size = 8, 2098 }, 2099 .valid = { 2100 .min_access_size = 4, 2101 .max_access_size = 8, 2102 }, 2103 }; 2104 2105 static Property vtd_properties[] = { 2106 DEFINE_PROP_UINT32("version", IntelIOMMUState, version, 0), 2107 DEFINE_PROP_ON_OFF_AUTO("eim", IntelIOMMUState, intr_eim, 2108 ON_OFF_AUTO_AUTO), 2109 DEFINE_PROP_BOOL("x-buggy-eim", IntelIOMMUState, buggy_eim, false), 2110 DEFINE_PROP_END_OF_LIST(), 2111 }; 2112 2113 /* Read IRTE entry with specific index */ 2114 static int vtd_irte_get(IntelIOMMUState *iommu, uint16_t index, 2115 VTD_IR_TableEntry *entry, uint16_t sid) 2116 { 2117 static const uint16_t vtd_svt_mask[VTD_SQ_MAX] = \ 2118 {0xffff, 0xfffb, 0xfff9, 0xfff8}; 2119 dma_addr_t addr = 0x00; 2120 uint16_t mask, source_id; 2121 uint8_t bus, bus_max, bus_min; 2122 2123 addr = iommu->intr_root + index * sizeof(*entry); 2124 if (dma_memory_read(&address_space_memory, addr, entry, 2125 sizeof(*entry))) { 2126 VTD_DPRINTF(GENERAL, "error: fail to access IR root at 0x%"PRIx64 2127 " + %"PRIu16, iommu->intr_root, index); 2128 return -VTD_FR_IR_ROOT_INVAL; 2129 } 2130 2131 if (!entry->irte.present) { 2132 VTD_DPRINTF(GENERAL, "error: present flag not set in IRTE" 2133 " entry index %u value 0x%"PRIx64 " 0x%"PRIx64, 2134 index, le64_to_cpu(entry->data[1]), 2135 le64_to_cpu(entry->data[0])); 2136 return -VTD_FR_IR_ENTRY_P; 2137 } 2138 2139 if (entry->irte.__reserved_0 || entry->irte.__reserved_1 || 2140 entry->irte.__reserved_2) { 2141 VTD_DPRINTF(GENERAL, "error: IRTE entry index %"PRIu16 2142 " reserved fields non-zero: 0x%"PRIx64 " 0x%"PRIx64, 2143 index, le64_to_cpu(entry->data[1]), 2144 le64_to_cpu(entry->data[0])); 2145 return -VTD_FR_IR_IRTE_RSVD; 2146 } 2147 2148 if (sid != X86_IOMMU_SID_INVALID) { 2149 /* Validate IRTE SID */ 2150 source_id = le32_to_cpu(entry->irte.source_id); 2151 switch (entry->irte.sid_vtype) { 2152 case VTD_SVT_NONE: 2153 VTD_DPRINTF(IR, "No SID validation for IRTE index %d", index); 2154 break; 2155 2156 case VTD_SVT_ALL: 2157 mask = vtd_svt_mask[entry->irte.sid_q]; 2158 if ((source_id & mask) != (sid & mask)) { 2159 VTD_DPRINTF(GENERAL, "SID validation for IRTE index " 2160 "%d failed (reqid 0x%04x sid 0x%04x)", index, 2161 sid, source_id); 2162 return -VTD_FR_IR_SID_ERR; 2163 } 2164 break; 2165 2166 case VTD_SVT_BUS: 2167 bus_max = source_id >> 8; 2168 bus_min = source_id & 0xff; 2169 bus = sid >> 8; 2170 if (bus > bus_max || bus < bus_min) { 2171 VTD_DPRINTF(GENERAL, "SID validation for IRTE index %d " 2172 "failed (bus %d outside %d-%d)", index, bus, 2173 bus_min, bus_max); 2174 return -VTD_FR_IR_SID_ERR; 2175 } 2176 break; 2177 2178 default: 2179 VTD_DPRINTF(GENERAL, "Invalid SVT bits (0x%x) in IRTE index " 2180 "%d", entry->irte.sid_vtype, index); 2181 /* Take this as verification failure. */ 2182 return -VTD_FR_IR_SID_ERR; 2183 break; 2184 } 2185 } 2186 2187 return 0; 2188 } 2189 2190 /* Fetch IRQ information of specific IR index */ 2191 static int vtd_remap_irq_get(IntelIOMMUState *iommu, uint16_t index, 2192 VTDIrq *irq, uint16_t sid) 2193 { 2194 VTD_IR_TableEntry irte = {}; 2195 int ret = 0; 2196 2197 ret = vtd_irte_get(iommu, index, &irte, sid); 2198 if (ret) { 2199 return ret; 2200 } 2201 2202 irq->trigger_mode = irte.irte.trigger_mode; 2203 irq->vector = irte.irte.vector; 2204 irq->delivery_mode = irte.irte.delivery_mode; 2205 irq->dest = le32_to_cpu(irte.irte.dest_id); 2206 if (!iommu->intr_eime) { 2207 #define VTD_IR_APIC_DEST_MASK (0xff00ULL) 2208 #define VTD_IR_APIC_DEST_SHIFT (8) 2209 irq->dest = (irq->dest & VTD_IR_APIC_DEST_MASK) >> 2210 VTD_IR_APIC_DEST_SHIFT; 2211 } 2212 irq->dest_mode = irte.irte.dest_mode; 2213 irq->redir_hint = irte.irte.redir_hint; 2214 2215 VTD_DPRINTF(IR, "remapping interrupt index %d: trig:%u,vec:%u," 2216 "deliver:%u,dest:%u,dest_mode:%u", index, 2217 irq->trigger_mode, irq->vector, irq->delivery_mode, 2218 irq->dest, irq->dest_mode); 2219 2220 return 0; 2221 } 2222 2223 /* Generate one MSI message from VTDIrq info */ 2224 static void vtd_generate_msi_message(VTDIrq *irq, MSIMessage *msg_out) 2225 { 2226 VTD_MSIMessage msg = {}; 2227 2228 /* Generate address bits */ 2229 msg.dest_mode = irq->dest_mode; 2230 msg.redir_hint = irq->redir_hint; 2231 msg.dest = irq->dest; 2232 msg.__addr_hi = irq->dest & 0xffffff00; 2233 msg.__addr_head = cpu_to_le32(0xfee); 2234 /* Keep this from original MSI address bits */ 2235 msg.__not_used = irq->msi_addr_last_bits; 2236 2237 /* Generate data bits */ 2238 msg.vector = irq->vector; 2239 msg.delivery_mode = irq->delivery_mode; 2240 msg.level = 1; 2241 msg.trigger_mode = irq->trigger_mode; 2242 2243 msg_out->address = msg.msi_addr; 2244 msg_out->data = msg.msi_data; 2245 } 2246 2247 /* Interrupt remapping for MSI/MSI-X entry */ 2248 static int vtd_interrupt_remap_msi(IntelIOMMUState *iommu, 2249 MSIMessage *origin, 2250 MSIMessage *translated, 2251 uint16_t sid) 2252 { 2253 int ret = 0; 2254 VTD_IR_MSIAddress addr; 2255 uint16_t index; 2256 VTDIrq irq = {}; 2257 2258 assert(origin && translated); 2259 2260 if (!iommu || !iommu->intr_enabled) { 2261 goto do_not_translate; 2262 } 2263 2264 if (origin->address & VTD_MSI_ADDR_HI_MASK) { 2265 VTD_DPRINTF(GENERAL, "error: MSI addr high 32 bits nonzero" 2266 " during interrupt remapping: 0x%"PRIx32, 2267 (uint32_t)((origin->address & VTD_MSI_ADDR_HI_MASK) >> \ 2268 VTD_MSI_ADDR_HI_SHIFT)); 2269 return -VTD_FR_IR_REQ_RSVD; 2270 } 2271 2272 addr.data = origin->address & VTD_MSI_ADDR_LO_MASK; 2273 if (addr.addr.__head != 0xfee) { 2274 VTD_DPRINTF(GENERAL, "error: MSI addr low 32 bits invalid: " 2275 "0x%"PRIx32, addr.data); 2276 return -VTD_FR_IR_REQ_RSVD; 2277 } 2278 2279 /* This is compatible mode. */ 2280 if (addr.addr.int_mode != VTD_IR_INT_FORMAT_REMAP) { 2281 goto do_not_translate; 2282 } 2283 2284 index = addr.addr.index_h << 15 | le16_to_cpu(addr.addr.index_l); 2285 2286 #define VTD_IR_MSI_DATA_SUBHANDLE (0x0000ffff) 2287 #define VTD_IR_MSI_DATA_RESERVED (0xffff0000) 2288 2289 if (addr.addr.sub_valid) { 2290 /* See VT-d spec 5.1.2.2 and 5.1.3 on subhandle */ 2291 index += origin->data & VTD_IR_MSI_DATA_SUBHANDLE; 2292 } 2293 2294 ret = vtd_remap_irq_get(iommu, index, &irq, sid); 2295 if (ret) { 2296 return ret; 2297 } 2298 2299 if (addr.addr.sub_valid) { 2300 VTD_DPRINTF(IR, "received MSI interrupt"); 2301 if (origin->data & VTD_IR_MSI_DATA_RESERVED) { 2302 VTD_DPRINTF(GENERAL, "error: MSI data bits non-zero for " 2303 "interrupt remappable entry: 0x%"PRIx32, 2304 origin->data); 2305 return -VTD_FR_IR_REQ_RSVD; 2306 } 2307 } else { 2308 uint8_t vector = origin->data & 0xff; 2309 uint8_t trigger_mode = (origin->data >> MSI_DATA_TRIGGER_SHIFT) & 0x1; 2310 2311 VTD_DPRINTF(IR, "received IOAPIC interrupt"); 2312 /* IOAPIC entry vector should be aligned with IRTE vector 2313 * (see vt-d spec 5.1.5.1). */ 2314 if (vector != irq.vector) { 2315 VTD_DPRINTF(GENERAL, "IOAPIC vector inconsistent: " 2316 "entry: %d, IRTE: %d, index: %d", 2317 vector, irq.vector, index); 2318 } 2319 2320 /* The Trigger Mode field must match the Trigger Mode in the IRTE. 2321 * (see vt-d spec 5.1.5.1). */ 2322 if (trigger_mode != irq.trigger_mode) { 2323 VTD_DPRINTF(GENERAL, "IOAPIC trigger mode inconsistent: " 2324 "entry: %u, IRTE: %u, index: %d", 2325 trigger_mode, irq.trigger_mode, index); 2326 } 2327 2328 } 2329 2330 /* 2331 * We'd better keep the last two bits, assuming that guest OS 2332 * might modify it. Keep it does not hurt after all. 2333 */ 2334 irq.msi_addr_last_bits = addr.addr.__not_care; 2335 2336 /* Translate VTDIrq to MSI message */ 2337 vtd_generate_msi_message(&irq, translated); 2338 2339 VTD_DPRINTF(IR, "mapping MSI 0x%"PRIx64":0x%"PRIx32 " -> " 2340 "0x%"PRIx64":0x%"PRIx32, origin->address, origin->data, 2341 translated->address, translated->data); 2342 return 0; 2343 2344 do_not_translate: 2345 memcpy(translated, origin, sizeof(*origin)); 2346 return 0; 2347 } 2348 2349 static int vtd_int_remap(X86IOMMUState *iommu, MSIMessage *src, 2350 MSIMessage *dst, uint16_t sid) 2351 { 2352 return vtd_interrupt_remap_msi(INTEL_IOMMU_DEVICE(iommu), 2353 src, dst, sid); 2354 } 2355 2356 static MemTxResult vtd_mem_ir_read(void *opaque, hwaddr addr, 2357 uint64_t *data, unsigned size, 2358 MemTxAttrs attrs) 2359 { 2360 return MEMTX_OK; 2361 } 2362 2363 static MemTxResult vtd_mem_ir_write(void *opaque, hwaddr addr, 2364 uint64_t value, unsigned size, 2365 MemTxAttrs attrs) 2366 { 2367 int ret = 0; 2368 MSIMessage from = {}, to = {}; 2369 uint16_t sid = X86_IOMMU_SID_INVALID; 2370 2371 from.address = (uint64_t) addr + VTD_INTERRUPT_ADDR_FIRST; 2372 from.data = (uint32_t) value; 2373 2374 if (!attrs.unspecified) { 2375 /* We have explicit Source ID */ 2376 sid = attrs.requester_id; 2377 } 2378 2379 ret = vtd_interrupt_remap_msi(opaque, &from, &to, sid); 2380 if (ret) { 2381 /* TODO: report error */ 2382 VTD_DPRINTF(GENERAL, "int remap fail for addr 0x%"PRIx64 2383 " data 0x%"PRIx32, from.address, from.data); 2384 /* Drop this interrupt */ 2385 return MEMTX_ERROR; 2386 } 2387 2388 VTD_DPRINTF(IR, "delivering MSI 0x%"PRIx64":0x%"PRIx32 2389 " for device sid 0x%04x", 2390 to.address, to.data, sid); 2391 2392 apic_get_class()->send_msi(&to); 2393 2394 return MEMTX_OK; 2395 } 2396 2397 static const MemoryRegionOps vtd_mem_ir_ops = { 2398 .read_with_attrs = vtd_mem_ir_read, 2399 .write_with_attrs = vtd_mem_ir_write, 2400 .endianness = DEVICE_LITTLE_ENDIAN, 2401 .impl = { 2402 .min_access_size = 4, 2403 .max_access_size = 4, 2404 }, 2405 .valid = { 2406 .min_access_size = 4, 2407 .max_access_size = 4, 2408 }, 2409 }; 2410 2411 VTDAddressSpace *vtd_find_add_as(IntelIOMMUState *s, PCIBus *bus, int devfn) 2412 { 2413 uintptr_t key = (uintptr_t)bus; 2414 VTDBus *vtd_bus = g_hash_table_lookup(s->vtd_as_by_busptr, &key); 2415 VTDAddressSpace *vtd_dev_as; 2416 char name[128]; 2417 2418 if (!vtd_bus) { 2419 uintptr_t *new_key = g_malloc(sizeof(*new_key)); 2420 *new_key = (uintptr_t)bus; 2421 /* No corresponding free() */ 2422 vtd_bus = g_malloc0(sizeof(VTDBus) + sizeof(VTDAddressSpace *) * \ 2423 X86_IOMMU_PCI_DEVFN_MAX); 2424 vtd_bus->bus = bus; 2425 g_hash_table_insert(s->vtd_as_by_busptr, new_key, vtd_bus); 2426 } 2427 2428 vtd_dev_as = vtd_bus->dev_as[devfn]; 2429 2430 if (!vtd_dev_as) { 2431 snprintf(name, sizeof(name), "intel_iommu_devfn_%d", devfn); 2432 vtd_bus->dev_as[devfn] = vtd_dev_as = g_malloc0(sizeof(VTDAddressSpace)); 2433 2434 vtd_dev_as->bus = bus; 2435 vtd_dev_as->devfn = (uint8_t)devfn; 2436 vtd_dev_as->iommu_state = s; 2437 vtd_dev_as->context_cache_entry.context_cache_gen = 0; 2438 memory_region_init_iommu(&vtd_dev_as->iommu, OBJECT(s), 2439 &s->iommu_ops, "intel_iommu", UINT64_MAX); 2440 memory_region_init_io(&vtd_dev_as->iommu_ir, OBJECT(s), 2441 &vtd_mem_ir_ops, s, "intel_iommu_ir", 2442 VTD_INTERRUPT_ADDR_SIZE); 2443 memory_region_add_subregion(&vtd_dev_as->iommu, VTD_INTERRUPT_ADDR_FIRST, 2444 &vtd_dev_as->iommu_ir); 2445 address_space_init(&vtd_dev_as->as, 2446 &vtd_dev_as->iommu, name); 2447 } 2448 return vtd_dev_as; 2449 } 2450 2451 /* Do the initialization. It will also be called when reset, so pay 2452 * attention when adding new initialization stuff. 2453 */ 2454 static void vtd_init(IntelIOMMUState *s) 2455 { 2456 X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s); 2457 2458 memset(s->csr, 0, DMAR_REG_SIZE); 2459 memset(s->wmask, 0, DMAR_REG_SIZE); 2460 memset(s->w1cmask, 0, DMAR_REG_SIZE); 2461 memset(s->womask, 0, DMAR_REG_SIZE); 2462 2463 s->iommu_ops.translate = vtd_iommu_translate; 2464 s->iommu_ops.notify_flag_changed = vtd_iommu_notify_flag_changed; 2465 s->root = 0; 2466 s->root_extended = false; 2467 s->dmar_enabled = false; 2468 s->iq_head = 0; 2469 s->iq_tail = 0; 2470 s->iq = 0; 2471 s->iq_size = 0; 2472 s->qi_enabled = false; 2473 s->iq_last_desc_type = VTD_INV_DESC_NONE; 2474 s->next_frcd_reg = 0; 2475 s->cap = VTD_CAP_FRO | VTD_CAP_NFR | VTD_CAP_ND | VTD_CAP_MGAW | 2476 VTD_CAP_SAGAW | VTD_CAP_MAMV | VTD_CAP_PSI | VTD_CAP_SLLPS; 2477 s->ecap = VTD_ECAP_QI | VTD_ECAP_IRO; 2478 2479 if (x86_iommu->intr_supported) { 2480 s->ecap |= VTD_ECAP_IR | VTD_ECAP_MHMV; 2481 if (s->intr_eim == ON_OFF_AUTO_ON) { 2482 s->ecap |= VTD_ECAP_EIM; 2483 } 2484 assert(s->intr_eim != ON_OFF_AUTO_AUTO); 2485 } 2486 2487 if (x86_iommu->dt_supported) { 2488 s->ecap |= VTD_ECAP_DT; 2489 } 2490 2491 vtd_reset_context_cache(s); 2492 vtd_reset_iotlb(s); 2493 2494 /* Define registers with default values and bit semantics */ 2495 vtd_define_long(s, DMAR_VER_REG, 0x10UL, 0, 0); 2496 vtd_define_quad(s, DMAR_CAP_REG, s->cap, 0, 0); 2497 vtd_define_quad(s, DMAR_ECAP_REG, s->ecap, 0, 0); 2498 vtd_define_long(s, DMAR_GCMD_REG, 0, 0xff800000UL, 0); 2499 vtd_define_long_wo(s, DMAR_GCMD_REG, 0xff800000UL); 2500 vtd_define_long(s, DMAR_GSTS_REG, 0, 0, 0); 2501 vtd_define_quad(s, DMAR_RTADDR_REG, 0, 0xfffffffffffff000ULL, 0); 2502 vtd_define_quad(s, DMAR_CCMD_REG, 0, 0xe0000003ffffffffULL, 0); 2503 vtd_define_quad_wo(s, DMAR_CCMD_REG, 0x3ffff0000ULL); 2504 2505 /* Advanced Fault Logging not supported */ 2506 vtd_define_long(s, DMAR_FSTS_REG, 0, 0, 0x11UL); 2507 vtd_define_long(s, DMAR_FECTL_REG, 0x80000000UL, 0x80000000UL, 0); 2508 vtd_define_long(s, DMAR_FEDATA_REG, 0, 0x0000ffffUL, 0); 2509 vtd_define_long(s, DMAR_FEADDR_REG, 0, 0xfffffffcUL, 0); 2510 2511 /* Treated as RsvdZ when EIM in ECAP_REG is not supported 2512 * vtd_define_long(s, DMAR_FEUADDR_REG, 0, 0xffffffffUL, 0); 2513 */ 2514 vtd_define_long(s, DMAR_FEUADDR_REG, 0, 0, 0); 2515 2516 /* Treated as RO for implementations that PLMR and PHMR fields reported 2517 * as Clear in the CAP_REG. 2518 * vtd_define_long(s, DMAR_PMEN_REG, 0, 0x80000000UL, 0); 2519 */ 2520 vtd_define_long(s, DMAR_PMEN_REG, 0, 0, 0); 2521 2522 vtd_define_quad(s, DMAR_IQH_REG, 0, 0, 0); 2523 vtd_define_quad(s, DMAR_IQT_REG, 0, 0x7fff0ULL, 0); 2524 vtd_define_quad(s, DMAR_IQA_REG, 0, 0xfffffffffffff007ULL, 0); 2525 vtd_define_long(s, DMAR_ICS_REG, 0, 0, 0x1UL); 2526 vtd_define_long(s, DMAR_IECTL_REG, 0x80000000UL, 0x80000000UL, 0); 2527 vtd_define_long(s, DMAR_IEDATA_REG, 0, 0xffffffffUL, 0); 2528 vtd_define_long(s, DMAR_IEADDR_REG, 0, 0xfffffffcUL, 0); 2529 /* Treadted as RsvdZ when EIM in ECAP_REG is not supported */ 2530 vtd_define_long(s, DMAR_IEUADDR_REG, 0, 0, 0); 2531 2532 /* IOTLB registers */ 2533 vtd_define_quad(s, DMAR_IOTLB_REG, 0, 0Xb003ffff00000000ULL, 0); 2534 vtd_define_quad(s, DMAR_IVA_REG, 0, 0xfffffffffffff07fULL, 0); 2535 vtd_define_quad_wo(s, DMAR_IVA_REG, 0xfffffffffffff07fULL); 2536 2537 /* Fault Recording Registers, 128-bit */ 2538 vtd_define_quad(s, DMAR_FRCD_REG_0_0, 0, 0, 0); 2539 vtd_define_quad(s, DMAR_FRCD_REG_0_2, 0, 0, 0x8000000000000000ULL); 2540 2541 /* 2542 * Interrupt remapping registers. 2543 */ 2544 vtd_define_quad(s, DMAR_IRTA_REG, 0, 0xfffffffffffff80fULL, 0); 2545 } 2546 2547 /* Should not reset address_spaces when reset because devices will still use 2548 * the address space they got at first (won't ask the bus again). 2549 */ 2550 static void vtd_reset(DeviceState *dev) 2551 { 2552 IntelIOMMUState *s = INTEL_IOMMU_DEVICE(dev); 2553 2554 VTD_DPRINTF(GENERAL, ""); 2555 vtd_init(s); 2556 } 2557 2558 static AddressSpace *vtd_host_dma_iommu(PCIBus *bus, void *opaque, int devfn) 2559 { 2560 IntelIOMMUState *s = opaque; 2561 VTDAddressSpace *vtd_as; 2562 2563 assert(0 <= devfn && devfn < X86_IOMMU_PCI_DEVFN_MAX); 2564 2565 vtd_as = vtd_find_add_as(s, bus, devfn); 2566 return &vtd_as->as; 2567 } 2568 2569 static bool vtd_decide_config(IntelIOMMUState *s, Error **errp) 2570 { 2571 X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s); 2572 2573 /* Currently Intel IOMMU IR only support "kernel-irqchip={off|split}" */ 2574 if (x86_iommu->intr_supported && kvm_irqchip_in_kernel() && 2575 !kvm_irqchip_is_split()) { 2576 error_setg(errp, "Intel Interrupt Remapping cannot work with " 2577 "kernel-irqchip=on, please use 'split|off'."); 2578 return false; 2579 } 2580 if (s->intr_eim == ON_OFF_AUTO_ON && !x86_iommu->intr_supported) { 2581 error_setg(errp, "eim=on cannot be selected without intremap=on"); 2582 return false; 2583 } 2584 2585 if (s->intr_eim == ON_OFF_AUTO_AUTO) { 2586 s->intr_eim = (kvm_irqchip_in_kernel() || s->buggy_eim) 2587 && x86_iommu->intr_supported ? 2588 ON_OFF_AUTO_ON : ON_OFF_AUTO_OFF; 2589 } 2590 if (s->intr_eim == ON_OFF_AUTO_ON && !s->buggy_eim) { 2591 if (!kvm_irqchip_in_kernel()) { 2592 error_setg(errp, "eim=on requires accel=kvm,kernel-irqchip=split"); 2593 return false; 2594 } 2595 if (!kvm_enable_x2apic()) { 2596 error_setg(errp, "eim=on requires support on the KVM side" 2597 "(X2APIC_API, first shipped in v4.7)"); 2598 return false; 2599 } 2600 } 2601 2602 return true; 2603 } 2604 2605 static void vtd_realize(DeviceState *dev, Error **errp) 2606 { 2607 PCMachineState *pcms = PC_MACHINE(qdev_get_machine()); 2608 PCIBus *bus = pcms->bus; 2609 IntelIOMMUState *s = INTEL_IOMMU_DEVICE(dev); 2610 X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(dev); 2611 2612 VTD_DPRINTF(GENERAL, ""); 2613 x86_iommu->type = TYPE_INTEL; 2614 2615 if (!vtd_decide_config(s, errp)) { 2616 return; 2617 } 2618 2619 memset(s->vtd_as_by_bus_num, 0, sizeof(s->vtd_as_by_bus_num)); 2620 memory_region_init_io(&s->csrmem, OBJECT(s), &vtd_mem_ops, s, 2621 "intel_iommu", DMAR_REG_SIZE); 2622 sysbus_init_mmio(SYS_BUS_DEVICE(s), &s->csrmem); 2623 /* No corresponding destroy */ 2624 s->iotlb = g_hash_table_new_full(vtd_uint64_hash, vtd_uint64_equal, 2625 g_free, g_free); 2626 s->vtd_as_by_busptr = g_hash_table_new_full(vtd_uint64_hash, vtd_uint64_equal, 2627 g_free, g_free); 2628 vtd_init(s); 2629 sysbus_mmio_map(SYS_BUS_DEVICE(s), 0, Q35_HOST_BRIDGE_IOMMU_ADDR); 2630 pci_setup_iommu(bus, vtd_host_dma_iommu, dev); 2631 /* Pseudo address space under root PCI bus. */ 2632 pcms->ioapic_as = vtd_host_dma_iommu(bus, s, Q35_PSEUDO_DEVFN_IOAPIC); 2633 } 2634 2635 static void vtd_class_init(ObjectClass *klass, void *data) 2636 { 2637 DeviceClass *dc = DEVICE_CLASS(klass); 2638 X86IOMMUClass *x86_class = X86_IOMMU_CLASS(klass); 2639 2640 dc->reset = vtd_reset; 2641 dc->vmsd = &vtd_vmstate; 2642 dc->props = vtd_properties; 2643 dc->hotpluggable = false; 2644 x86_class->realize = vtd_realize; 2645 x86_class->int_remap = vtd_int_remap; 2646 } 2647 2648 static const TypeInfo vtd_info = { 2649 .name = TYPE_INTEL_IOMMU_DEVICE, 2650 .parent = TYPE_X86_IOMMU_DEVICE, 2651 .instance_size = sizeof(IntelIOMMUState), 2652 .class_init = vtd_class_init, 2653 }; 2654 2655 static void vtd_register_types(void) 2656 { 2657 VTD_DPRINTF(GENERAL, ""); 2658 type_register_static(&vtd_info); 2659 } 2660 2661 type_init(vtd_register_types) 2662