1 /* 2 * QEMU sPAPR IOMMU (TCE) code 3 * 4 * Copyright (c) 2010 David Gibson, IBM Corporation <dwg@au1.ibm.com> 5 * 6 * This library is free software; you can redistribute it and/or 7 * modify it under the terms of the GNU Lesser General Public 8 * License as published by the Free Software Foundation; either 9 * version 2 of the License, or (at your option) any later version. 10 * 11 * This library is distributed in the hope that it will be useful, 12 * but WITHOUT ANY WARRANTY; without even the implied warranty of 13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 14 * Lesser General Public License for more details. 15 * 16 * You should have received a copy of the GNU Lesser General Public 17 * License along with this library; if not, see <http://www.gnu.org/licenses/>. 18 */ 19 #include "qemu/osdep.h" 20 #include "qemu/error-report.h" 21 #include "hw/hw.h" 22 #include "qemu/log.h" 23 #include "sysemu/kvm.h" 24 #include "hw/qdev.h" 25 #include "kvm_ppc.h" 26 #include "sysemu/dma.h" 27 #include "exec/address-spaces.h" 28 #include "trace.h" 29 30 #include "hw/ppc/spapr.h" 31 #include "hw/ppc/spapr_vio.h" 32 33 #include <libfdt.h> 34 35 enum sPAPRTCEAccess { 36 SPAPR_TCE_FAULT = 0, 37 SPAPR_TCE_RO = 1, 38 SPAPR_TCE_WO = 2, 39 SPAPR_TCE_RW = 3, 40 }; 41 42 #define IOMMU_PAGE_SIZE(shift) (1ULL << (shift)) 43 #define IOMMU_PAGE_MASK(shift) (~(IOMMU_PAGE_SIZE(shift) - 1)) 44 45 static QLIST_HEAD(spapr_tce_tables, sPAPRTCETable) spapr_tce_tables; 46 47 sPAPRTCETable *spapr_tce_find_by_liobn(target_ulong liobn) 48 { 49 sPAPRTCETable *tcet; 50 51 if (liobn & 0xFFFFFFFF00000000ULL) { 52 hcall_dprintf("Request for out-of-bounds LIOBN 0x" TARGET_FMT_lx "\n", 53 liobn); 54 return NULL; 55 } 56 57 QLIST_FOREACH(tcet, &spapr_tce_tables, list) { 58 if (tcet->liobn == (uint32_t)liobn) { 59 return tcet; 60 } 61 } 62 63 return NULL; 64 } 65 66 static IOMMUAccessFlags spapr_tce_iommu_access_flags(uint64_t tce) 67 { 68 switch (tce & SPAPR_TCE_RW) { 69 case SPAPR_TCE_FAULT: 70 return IOMMU_NONE; 71 case SPAPR_TCE_RO: 72 return IOMMU_RO; 73 case SPAPR_TCE_WO: 74 return IOMMU_WO; 75 default: /* SPAPR_TCE_RW */ 76 return IOMMU_RW; 77 } 78 } 79 80 static uint64_t *spapr_tce_alloc_table(uint32_t liobn, 81 uint32_t page_shift, 82 uint32_t nb_table, 83 int *fd, 84 bool need_vfio) 85 { 86 uint64_t *table = NULL; 87 uint64_t window_size = (uint64_t)nb_table << page_shift; 88 89 if (kvm_enabled() && !(window_size >> 32)) { 90 table = kvmppc_create_spapr_tce(liobn, window_size, fd, need_vfio); 91 } 92 93 if (!table) { 94 *fd = -1; 95 table = g_malloc0(nb_table * sizeof(uint64_t)); 96 } 97 98 trace_spapr_iommu_new_table(liobn, table, *fd); 99 100 return table; 101 } 102 103 static void spapr_tce_free_table(uint64_t *table, int fd, uint32_t nb_table) 104 { 105 if (!kvm_enabled() || 106 (kvmppc_remove_spapr_tce(table, fd, nb_table) != 0)) { 107 g_free(table); 108 } 109 } 110 111 /* Called from RCU critical section */ 112 static IOMMUTLBEntry spapr_tce_translate_iommu(MemoryRegion *iommu, hwaddr addr, 113 bool is_write) 114 { 115 sPAPRTCETable *tcet = container_of(iommu, sPAPRTCETable, iommu); 116 uint64_t tce; 117 IOMMUTLBEntry ret = { 118 .target_as = &address_space_memory, 119 .iova = 0, 120 .translated_addr = 0, 121 .addr_mask = ~(hwaddr)0, 122 .perm = IOMMU_NONE, 123 }; 124 125 if ((addr >> tcet->page_shift) < tcet->nb_table) { 126 /* Check if we are in bound */ 127 hwaddr page_mask = IOMMU_PAGE_MASK(tcet->page_shift); 128 129 tce = tcet->table[addr >> tcet->page_shift]; 130 ret.iova = addr & page_mask; 131 ret.translated_addr = tce & page_mask; 132 ret.addr_mask = ~page_mask; 133 ret.perm = spapr_tce_iommu_access_flags(tce); 134 } 135 trace_spapr_iommu_xlate(tcet->liobn, addr, ret.iova, ret.perm, 136 ret.addr_mask); 137 138 return ret; 139 } 140 141 static void spapr_tce_table_pre_save(void *opaque) 142 { 143 sPAPRTCETable *tcet = SPAPR_TCE_TABLE(opaque); 144 145 tcet->mig_table = tcet->table; 146 tcet->mig_nb_table = tcet->nb_table; 147 148 trace_spapr_iommu_pre_save(tcet->liobn, tcet->mig_nb_table, 149 tcet->bus_offset, tcet->page_shift); 150 } 151 152 static uint64_t spapr_tce_get_min_page_size(MemoryRegion *iommu) 153 { 154 sPAPRTCETable *tcet = container_of(iommu, sPAPRTCETable, iommu); 155 156 return 1ULL << tcet->page_shift; 157 } 158 159 static void spapr_tce_notify_started(MemoryRegion *iommu) 160 { 161 spapr_tce_set_need_vfio(container_of(iommu, sPAPRTCETable, iommu), true); 162 } 163 164 static void spapr_tce_notify_stopped(MemoryRegion *iommu) 165 { 166 spapr_tce_set_need_vfio(container_of(iommu, sPAPRTCETable, iommu), false); 167 } 168 169 static int spapr_tce_table_post_load(void *opaque, int version_id) 170 { 171 sPAPRTCETable *tcet = SPAPR_TCE_TABLE(opaque); 172 uint32_t old_nb_table = tcet->nb_table; 173 uint64_t old_bus_offset = tcet->bus_offset; 174 uint32_t old_page_shift = tcet->page_shift; 175 176 if (tcet->vdev) { 177 spapr_vio_set_bypass(tcet->vdev, tcet->bypass); 178 } 179 180 if (tcet->mig_nb_table != tcet->nb_table) { 181 spapr_tce_table_disable(tcet); 182 } 183 184 if (tcet->mig_nb_table) { 185 if (!tcet->nb_table) { 186 spapr_tce_table_enable(tcet, old_page_shift, old_bus_offset, 187 tcet->mig_nb_table); 188 } 189 190 memcpy(tcet->table, tcet->mig_table, 191 tcet->nb_table * sizeof(tcet->table[0])); 192 193 free(tcet->mig_table); 194 tcet->mig_table = NULL; 195 } 196 197 trace_spapr_iommu_post_load(tcet->liobn, old_nb_table, tcet->nb_table, 198 tcet->bus_offset, tcet->page_shift); 199 200 return 0; 201 } 202 203 static bool spapr_tce_table_ex_needed(void *opaque) 204 { 205 sPAPRTCETable *tcet = opaque; 206 207 return tcet->bus_offset || tcet->page_shift != 0xC; 208 } 209 210 static const VMStateDescription vmstate_spapr_tce_table_ex = { 211 .name = "spapr_iommu_ex", 212 .version_id = 1, 213 .minimum_version_id = 1, 214 .needed = spapr_tce_table_ex_needed, 215 .fields = (VMStateField[]) { 216 VMSTATE_UINT64(bus_offset, sPAPRTCETable), 217 VMSTATE_UINT32(page_shift, sPAPRTCETable), 218 VMSTATE_END_OF_LIST() 219 }, 220 }; 221 222 static const VMStateDescription vmstate_spapr_tce_table = { 223 .name = "spapr_iommu", 224 .version_id = 2, 225 .minimum_version_id = 2, 226 .pre_save = spapr_tce_table_pre_save, 227 .post_load = spapr_tce_table_post_load, 228 .fields = (VMStateField []) { 229 /* Sanity check */ 230 VMSTATE_UINT32_EQUAL(liobn, sPAPRTCETable), 231 232 /* IOMMU state */ 233 VMSTATE_UINT32(mig_nb_table, sPAPRTCETable), 234 VMSTATE_BOOL(bypass, sPAPRTCETable), 235 VMSTATE_VARRAY_UINT32_ALLOC(mig_table, sPAPRTCETable, mig_nb_table, 0, 236 vmstate_info_uint64, uint64_t), 237 238 VMSTATE_END_OF_LIST() 239 }, 240 .subsections = (const VMStateDescription*[]) { 241 &vmstate_spapr_tce_table_ex, 242 NULL 243 } 244 }; 245 246 static MemoryRegionIOMMUOps spapr_iommu_ops = { 247 .translate = spapr_tce_translate_iommu, 248 .get_min_page_size = spapr_tce_get_min_page_size, 249 .notify_started = spapr_tce_notify_started, 250 .notify_stopped = spapr_tce_notify_stopped, 251 }; 252 253 static int spapr_tce_table_realize(DeviceState *dev) 254 { 255 sPAPRTCETable *tcet = SPAPR_TCE_TABLE(dev); 256 Object *tcetobj = OBJECT(tcet); 257 char tmp[32]; 258 259 tcet->fd = -1; 260 tcet->need_vfio = false; 261 snprintf(tmp, sizeof(tmp), "tce-root-%x", tcet->liobn); 262 memory_region_init(&tcet->root, tcetobj, tmp, UINT64_MAX); 263 264 snprintf(tmp, sizeof(tmp), "tce-iommu-%x", tcet->liobn); 265 memory_region_init_iommu(&tcet->iommu, tcetobj, &spapr_iommu_ops, tmp, 0); 266 267 QLIST_INSERT_HEAD(&spapr_tce_tables, tcet, list); 268 269 vmstate_register(DEVICE(tcet), tcet->liobn, &vmstate_spapr_tce_table, 270 tcet); 271 272 return 0; 273 } 274 275 void spapr_tce_set_need_vfio(sPAPRTCETable *tcet, bool need_vfio) 276 { 277 size_t table_size = tcet->nb_table * sizeof(uint64_t); 278 void *newtable; 279 280 if (need_vfio == tcet->need_vfio) { 281 /* Nothing to do */ 282 return; 283 } 284 285 if (!need_vfio) { 286 /* FIXME: We don't support transition back to KVM accelerated 287 * TCEs yet */ 288 return; 289 } 290 291 tcet->need_vfio = true; 292 293 if (tcet->fd < 0) { 294 /* Table is already in userspace, nothing to be do */ 295 return; 296 } 297 298 newtable = g_malloc(table_size); 299 memcpy(newtable, tcet->table, table_size); 300 301 kvmppc_remove_spapr_tce(tcet->table, tcet->fd, tcet->nb_table); 302 303 tcet->fd = -1; 304 tcet->table = newtable; 305 } 306 307 sPAPRTCETable *spapr_tce_new_table(DeviceState *owner, uint32_t liobn) 308 { 309 sPAPRTCETable *tcet; 310 char tmp[32]; 311 312 if (spapr_tce_find_by_liobn(liobn)) { 313 fprintf(stderr, "Attempted to create TCE table with duplicate" 314 " LIOBN 0x%x\n", liobn); 315 return NULL; 316 } 317 318 tcet = SPAPR_TCE_TABLE(object_new(TYPE_SPAPR_TCE_TABLE)); 319 tcet->liobn = liobn; 320 321 snprintf(tmp, sizeof(tmp), "tce-table-%x", liobn); 322 object_property_add_child(OBJECT(owner), tmp, OBJECT(tcet), NULL); 323 324 object_property_set_bool(OBJECT(tcet), true, "realized", NULL); 325 326 return tcet; 327 } 328 329 void spapr_tce_table_enable(sPAPRTCETable *tcet, 330 uint32_t page_shift, uint64_t bus_offset, 331 uint32_t nb_table) 332 { 333 if (tcet->nb_table) { 334 error_report("Warning: trying to enable already enabled TCE table"); 335 return; 336 } 337 338 tcet->bus_offset = bus_offset; 339 tcet->page_shift = page_shift; 340 tcet->nb_table = nb_table; 341 tcet->table = spapr_tce_alloc_table(tcet->liobn, 342 tcet->page_shift, 343 tcet->nb_table, 344 &tcet->fd, 345 tcet->need_vfio); 346 347 memory_region_set_size(&tcet->iommu, 348 (uint64_t)tcet->nb_table << tcet->page_shift); 349 memory_region_add_subregion(&tcet->root, tcet->bus_offset, &tcet->iommu); 350 } 351 352 void spapr_tce_table_disable(sPAPRTCETable *tcet) 353 { 354 if (!tcet->nb_table) { 355 return; 356 } 357 358 memory_region_del_subregion(&tcet->root, &tcet->iommu); 359 memory_region_set_size(&tcet->iommu, 0); 360 361 spapr_tce_free_table(tcet->table, tcet->fd, tcet->nb_table); 362 tcet->fd = -1; 363 tcet->table = NULL; 364 tcet->bus_offset = 0; 365 tcet->page_shift = 0; 366 tcet->nb_table = 0; 367 } 368 369 static void spapr_tce_table_unrealize(DeviceState *dev, Error **errp) 370 { 371 sPAPRTCETable *tcet = SPAPR_TCE_TABLE(dev); 372 373 QLIST_REMOVE(tcet, list); 374 375 spapr_tce_table_disable(tcet); 376 } 377 378 MemoryRegion *spapr_tce_get_iommu(sPAPRTCETable *tcet) 379 { 380 return &tcet->root; 381 } 382 383 static void spapr_tce_reset(DeviceState *dev) 384 { 385 sPAPRTCETable *tcet = SPAPR_TCE_TABLE(dev); 386 size_t table_size = tcet->nb_table * sizeof(uint64_t); 387 388 if (tcet->nb_table) { 389 memset(tcet->table, 0, table_size); 390 } 391 } 392 393 static target_ulong put_tce_emu(sPAPRTCETable *tcet, target_ulong ioba, 394 target_ulong tce) 395 { 396 IOMMUTLBEntry entry; 397 hwaddr page_mask = IOMMU_PAGE_MASK(tcet->page_shift); 398 unsigned long index = (ioba - tcet->bus_offset) >> tcet->page_shift; 399 400 if (index >= tcet->nb_table) { 401 hcall_dprintf("spapr_vio_put_tce on out-of-bounds IOBA 0x" 402 TARGET_FMT_lx "\n", ioba); 403 return H_PARAMETER; 404 } 405 406 tcet->table[index] = tce; 407 408 entry.target_as = &address_space_memory, 409 entry.iova = (ioba - tcet->bus_offset) & page_mask; 410 entry.translated_addr = tce & page_mask; 411 entry.addr_mask = ~page_mask; 412 entry.perm = spapr_tce_iommu_access_flags(tce); 413 memory_region_notify_iommu(&tcet->iommu, entry); 414 415 return H_SUCCESS; 416 } 417 418 static target_ulong h_put_tce_indirect(PowerPCCPU *cpu, 419 sPAPRMachineState *spapr, 420 target_ulong opcode, target_ulong *args) 421 { 422 int i; 423 target_ulong liobn = args[0]; 424 target_ulong ioba = args[1]; 425 target_ulong ioba1 = ioba; 426 target_ulong tce_list = args[2]; 427 target_ulong npages = args[3]; 428 target_ulong ret = H_PARAMETER, tce = 0; 429 sPAPRTCETable *tcet = spapr_tce_find_by_liobn(liobn); 430 CPUState *cs = CPU(cpu); 431 hwaddr page_mask, page_size; 432 433 if (!tcet) { 434 return H_PARAMETER; 435 } 436 437 if ((npages > 512) || (tce_list & SPAPR_TCE_PAGE_MASK)) { 438 return H_PARAMETER; 439 } 440 441 page_mask = IOMMU_PAGE_MASK(tcet->page_shift); 442 page_size = IOMMU_PAGE_SIZE(tcet->page_shift); 443 ioba &= page_mask; 444 445 for (i = 0; i < npages; ++i, ioba += page_size) { 446 tce = ldq_be_phys(cs->as, tce_list + i * sizeof(target_ulong)); 447 448 ret = put_tce_emu(tcet, ioba, tce); 449 if (ret) { 450 break; 451 } 452 } 453 454 /* Trace last successful or the first problematic entry */ 455 i = i ? (i - 1) : 0; 456 if (SPAPR_IS_PCI_LIOBN(liobn)) { 457 trace_spapr_iommu_pci_indirect(liobn, ioba1, tce_list, i, tce, ret); 458 } else { 459 trace_spapr_iommu_indirect(liobn, ioba1, tce_list, i, tce, ret); 460 } 461 return ret; 462 } 463 464 static target_ulong h_stuff_tce(PowerPCCPU *cpu, sPAPRMachineState *spapr, 465 target_ulong opcode, target_ulong *args) 466 { 467 int i; 468 target_ulong liobn = args[0]; 469 target_ulong ioba = args[1]; 470 target_ulong tce_value = args[2]; 471 target_ulong npages = args[3]; 472 target_ulong ret = H_PARAMETER; 473 sPAPRTCETable *tcet = spapr_tce_find_by_liobn(liobn); 474 hwaddr page_mask, page_size; 475 476 if (!tcet) { 477 return H_PARAMETER; 478 } 479 480 if (npages > tcet->nb_table) { 481 return H_PARAMETER; 482 } 483 484 page_mask = IOMMU_PAGE_MASK(tcet->page_shift); 485 page_size = IOMMU_PAGE_SIZE(tcet->page_shift); 486 ioba &= page_mask; 487 488 for (i = 0; i < npages; ++i, ioba += page_size) { 489 ret = put_tce_emu(tcet, ioba, tce_value); 490 if (ret) { 491 break; 492 } 493 } 494 if (SPAPR_IS_PCI_LIOBN(liobn)) { 495 trace_spapr_iommu_pci_stuff(liobn, ioba, tce_value, npages, ret); 496 } else { 497 trace_spapr_iommu_stuff(liobn, ioba, tce_value, npages, ret); 498 } 499 500 return ret; 501 } 502 503 static target_ulong h_put_tce(PowerPCCPU *cpu, sPAPRMachineState *spapr, 504 target_ulong opcode, target_ulong *args) 505 { 506 target_ulong liobn = args[0]; 507 target_ulong ioba = args[1]; 508 target_ulong tce = args[2]; 509 target_ulong ret = H_PARAMETER; 510 sPAPRTCETable *tcet = spapr_tce_find_by_liobn(liobn); 511 512 if (tcet) { 513 hwaddr page_mask = IOMMU_PAGE_MASK(tcet->page_shift); 514 515 ioba &= page_mask; 516 517 ret = put_tce_emu(tcet, ioba, tce); 518 } 519 if (SPAPR_IS_PCI_LIOBN(liobn)) { 520 trace_spapr_iommu_pci_put(liobn, ioba, tce, ret); 521 } else { 522 trace_spapr_iommu_put(liobn, ioba, tce, ret); 523 } 524 525 return ret; 526 } 527 528 static target_ulong get_tce_emu(sPAPRTCETable *tcet, target_ulong ioba, 529 target_ulong *tce) 530 { 531 unsigned long index = (ioba - tcet->bus_offset) >> tcet->page_shift; 532 533 if (index >= tcet->nb_table) { 534 hcall_dprintf("spapr_iommu_get_tce on out-of-bounds IOBA 0x" 535 TARGET_FMT_lx "\n", ioba); 536 return H_PARAMETER; 537 } 538 539 *tce = tcet->table[index]; 540 541 return H_SUCCESS; 542 } 543 544 static target_ulong h_get_tce(PowerPCCPU *cpu, sPAPRMachineState *spapr, 545 target_ulong opcode, target_ulong *args) 546 { 547 target_ulong liobn = args[0]; 548 target_ulong ioba = args[1]; 549 target_ulong tce = 0; 550 target_ulong ret = H_PARAMETER; 551 sPAPRTCETable *tcet = spapr_tce_find_by_liobn(liobn); 552 553 if (tcet) { 554 hwaddr page_mask = IOMMU_PAGE_MASK(tcet->page_shift); 555 556 ioba &= page_mask; 557 558 ret = get_tce_emu(tcet, ioba, &tce); 559 if (!ret) { 560 args[0] = tce; 561 } 562 } 563 if (SPAPR_IS_PCI_LIOBN(liobn)) { 564 trace_spapr_iommu_pci_get(liobn, ioba, ret, tce); 565 } else { 566 trace_spapr_iommu_get(liobn, ioba, ret, tce); 567 } 568 569 return ret; 570 } 571 572 int spapr_dma_dt(void *fdt, int node_off, const char *propname, 573 uint32_t liobn, uint64_t window, uint32_t size) 574 { 575 uint32_t dma_prop[5]; 576 int ret; 577 578 dma_prop[0] = cpu_to_be32(liobn); 579 dma_prop[1] = cpu_to_be32(window >> 32); 580 dma_prop[2] = cpu_to_be32(window & 0xFFFFFFFF); 581 dma_prop[3] = 0; /* window size is 32 bits */ 582 dma_prop[4] = cpu_to_be32(size); 583 584 ret = fdt_setprop_cell(fdt, node_off, "ibm,#dma-address-cells", 2); 585 if (ret < 0) { 586 return ret; 587 } 588 589 ret = fdt_setprop_cell(fdt, node_off, "ibm,#dma-size-cells", 2); 590 if (ret < 0) { 591 return ret; 592 } 593 594 ret = fdt_setprop(fdt, node_off, propname, dma_prop, sizeof(dma_prop)); 595 if (ret < 0) { 596 return ret; 597 } 598 599 return 0; 600 } 601 602 int spapr_tcet_dma_dt(void *fdt, int node_off, const char *propname, 603 sPAPRTCETable *tcet) 604 { 605 if (!tcet) { 606 return 0; 607 } 608 609 return spapr_dma_dt(fdt, node_off, propname, 610 tcet->liobn, 0, tcet->nb_table << tcet->page_shift); 611 } 612 613 static void spapr_tce_table_class_init(ObjectClass *klass, void *data) 614 { 615 DeviceClass *dc = DEVICE_CLASS(klass); 616 dc->init = spapr_tce_table_realize; 617 dc->reset = spapr_tce_reset; 618 dc->unrealize = spapr_tce_table_unrealize; 619 620 QLIST_INIT(&spapr_tce_tables); 621 622 /* hcall-tce */ 623 spapr_register_hypercall(H_PUT_TCE, h_put_tce); 624 spapr_register_hypercall(H_GET_TCE, h_get_tce); 625 spapr_register_hypercall(H_PUT_TCE_INDIRECT, h_put_tce_indirect); 626 spapr_register_hypercall(H_STUFF_TCE, h_stuff_tce); 627 } 628 629 static TypeInfo spapr_tce_table_info = { 630 .name = TYPE_SPAPR_TCE_TABLE, 631 .parent = TYPE_DEVICE, 632 .instance_size = sizeof(sPAPRTCETable), 633 .class_init = spapr_tce_table_class_init, 634 }; 635 636 static void register_types(void) 637 { 638 type_register_static(&spapr_tce_table_info); 639 } 640 641 type_init(register_types); 642