1 /* 2 * QEMU sPAPR PCI host originated from Uninorth PCI host 3 * 4 * Copyright (c) 2011 Alexey Kardashevskiy, IBM Corporation. 5 * Copyright (C) 2011 David Gibson, IBM Corporation. 6 * 7 * Permission is hereby granted, free of charge, to any person obtaining a copy 8 * of this software and associated documentation files (the "Software"), to deal 9 * in the Software without restriction, including without limitation the rights 10 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 11 * copies of the Software, and to permit persons to whom the Software is 12 * furnished to do so, subject to the following conditions: 13 * 14 * The above copyright notice and this permission notice shall be included in 15 * all copies or substantial portions of the Software. 16 * 17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 23 * THE SOFTWARE. 24 */ 25 26 #include "qemu/osdep.h" 27 #include "qapi/error.h" 28 #include "hw/irq.h" 29 #include "hw/sysbus.h" 30 #include "migration/vmstate.h" 31 #include "hw/pci/pci.h" 32 #include "hw/pci/msi.h" 33 #include "hw/pci/msix.h" 34 #include "hw/pci/pci_host.h" 35 #include "hw/ppc/spapr.h" 36 #include "hw/pci-host/spapr.h" 37 #include "exec/ram_addr.h" 38 #include <libfdt.h> 39 #include "trace.h" 40 #include "qemu/error-report.h" 41 #include "qemu/module.h" 42 #include "hw/ppc/fdt.h" 43 #include "hw/pci/pci_bridge.h" 44 #include "hw/pci/pci_bus.h" 45 #include "hw/pci/pci_ids.h" 46 #include "hw/ppc/spapr_drc.h" 47 #include "hw/qdev-properties.h" 48 #include "sysemu/device_tree.h" 49 #include "sysemu/kvm.h" 50 #include "sysemu/hostmem.h" 51 #include "sysemu/numa.h" 52 #include "hw/ppc/spapr_numa.h" 53 #include "qemu/log.h" 54 55 /* Copied from the kernel arch/powerpc/platforms/pseries/msi.c */ 56 #define RTAS_QUERY_FN 0 57 #define RTAS_CHANGE_FN 1 58 #define RTAS_RESET_FN 2 59 #define RTAS_CHANGE_MSI_FN 3 60 #define RTAS_CHANGE_MSIX_FN 4 61 62 /* Interrupt types to return on RTAS_CHANGE_* */ 63 #define RTAS_TYPE_MSI 1 64 #define RTAS_TYPE_MSIX 2 65 66 SpaprPhbState *spapr_pci_find_phb(SpaprMachineState *spapr, uint64_t buid) 67 { 68 SpaprPhbState *sphb; 69 70 QLIST_FOREACH(sphb, &spapr->phbs, list) { 71 if (sphb->buid != buid) { 72 continue; 73 } 74 return sphb; 75 } 76 77 return NULL; 78 } 79 80 PCIDevice *spapr_pci_find_dev(SpaprMachineState *spapr, uint64_t buid, 81 uint32_t config_addr) 82 { 83 SpaprPhbState *sphb = spapr_pci_find_phb(spapr, buid); 84 PCIHostState *phb = PCI_HOST_BRIDGE(sphb); 85 int bus_num = (config_addr >> 16) & 0xFF; 86 int devfn = (config_addr >> 8) & 0xFF; 87 88 if (!phb) { 89 return NULL; 90 } 91 92 return pci_find_device(phb->bus, bus_num, devfn); 93 } 94 95 static uint32_t rtas_pci_cfgaddr(uint32_t arg) 96 { 97 /* This handles the encoding of extended config space addresses */ 98 return ((arg >> 20) & 0xf00) | (arg & 0xff); 99 } 100 101 static void finish_read_pci_config(SpaprMachineState *spapr, uint64_t buid, 102 uint32_t addr, uint32_t size, 103 target_ulong rets) 104 { 105 PCIDevice *pci_dev; 106 uint32_t val; 107 108 if ((size != 1) && (size != 2) && (size != 4)) { 109 /* access must be 1, 2 or 4 bytes */ 110 rtas_st(rets, 0, RTAS_OUT_HW_ERROR); 111 return; 112 } 113 114 pci_dev = spapr_pci_find_dev(spapr, buid, addr); 115 addr = rtas_pci_cfgaddr(addr); 116 117 if (!pci_dev || (addr % size) || (addr >= pci_config_size(pci_dev))) { 118 /* Access must be to a valid device, within bounds and 119 * naturally aligned */ 120 rtas_st(rets, 0, RTAS_OUT_HW_ERROR); 121 return; 122 } 123 124 val = pci_host_config_read_common(pci_dev, addr, 125 pci_config_size(pci_dev), size); 126 127 rtas_st(rets, 0, RTAS_OUT_SUCCESS); 128 rtas_st(rets, 1, val); 129 } 130 131 static void rtas_ibm_read_pci_config(PowerPCCPU *cpu, SpaprMachineState *spapr, 132 uint32_t token, uint32_t nargs, 133 target_ulong args, 134 uint32_t nret, target_ulong rets) 135 { 136 uint64_t buid; 137 uint32_t size, addr; 138 139 if ((nargs != 4) || (nret != 2)) { 140 rtas_st(rets, 0, RTAS_OUT_HW_ERROR); 141 return; 142 } 143 144 buid = rtas_ldq(args, 1); 145 size = rtas_ld(args, 3); 146 addr = rtas_ld(args, 0); 147 148 finish_read_pci_config(spapr, buid, addr, size, rets); 149 } 150 151 static void rtas_read_pci_config(PowerPCCPU *cpu, SpaprMachineState *spapr, 152 uint32_t token, uint32_t nargs, 153 target_ulong args, 154 uint32_t nret, target_ulong rets) 155 { 156 uint32_t size, addr; 157 158 if ((nargs != 2) || (nret != 2)) { 159 rtas_st(rets, 0, RTAS_OUT_HW_ERROR); 160 return; 161 } 162 163 size = rtas_ld(args, 1); 164 addr = rtas_ld(args, 0); 165 166 finish_read_pci_config(spapr, 0, addr, size, rets); 167 } 168 169 static void finish_write_pci_config(SpaprMachineState *spapr, uint64_t buid, 170 uint32_t addr, uint32_t size, 171 uint32_t val, target_ulong rets) 172 { 173 PCIDevice *pci_dev; 174 175 if ((size != 1) && (size != 2) && (size != 4)) { 176 /* access must be 1, 2 or 4 bytes */ 177 rtas_st(rets, 0, RTAS_OUT_HW_ERROR); 178 return; 179 } 180 181 pci_dev = spapr_pci_find_dev(spapr, buid, addr); 182 addr = rtas_pci_cfgaddr(addr); 183 184 if (!pci_dev || (addr % size) || (addr >= pci_config_size(pci_dev))) { 185 /* Access must be to a valid device, within bounds and 186 * naturally aligned */ 187 rtas_st(rets, 0, RTAS_OUT_HW_ERROR); 188 return; 189 } 190 191 pci_host_config_write_common(pci_dev, addr, pci_config_size(pci_dev), 192 val, size); 193 194 rtas_st(rets, 0, RTAS_OUT_SUCCESS); 195 } 196 197 static void rtas_ibm_write_pci_config(PowerPCCPU *cpu, SpaprMachineState *spapr, 198 uint32_t token, uint32_t nargs, 199 target_ulong args, 200 uint32_t nret, target_ulong rets) 201 { 202 uint64_t buid; 203 uint32_t val, size, addr; 204 205 if ((nargs != 5) || (nret != 1)) { 206 rtas_st(rets, 0, RTAS_OUT_HW_ERROR); 207 return; 208 } 209 210 buid = rtas_ldq(args, 1); 211 val = rtas_ld(args, 4); 212 size = rtas_ld(args, 3); 213 addr = rtas_ld(args, 0); 214 215 finish_write_pci_config(spapr, buid, addr, size, val, rets); 216 } 217 218 static void rtas_write_pci_config(PowerPCCPU *cpu, SpaprMachineState *spapr, 219 uint32_t token, uint32_t nargs, 220 target_ulong args, 221 uint32_t nret, target_ulong rets) 222 { 223 uint32_t val, size, addr; 224 225 if ((nargs != 3) || (nret != 1)) { 226 rtas_st(rets, 0, RTAS_OUT_HW_ERROR); 227 return; 228 } 229 230 231 val = rtas_ld(args, 2); 232 size = rtas_ld(args, 1); 233 addr = rtas_ld(args, 0); 234 235 finish_write_pci_config(spapr, 0, addr, size, val, rets); 236 } 237 238 /* 239 * Set MSI/MSIX message data. 240 * This is required for msi_notify()/msix_notify() which 241 * will write at the addresses via spapr_msi_write(). 242 * 243 * If hwaddr == 0, all entries will have .data == first_irq i.e. 244 * table will be reset. 245 */ 246 static void spapr_msi_setmsg(PCIDevice *pdev, hwaddr addr, bool msix, 247 unsigned first_irq, unsigned req_num) 248 { 249 unsigned i; 250 MSIMessage msg = { .address = addr, .data = first_irq }; 251 252 if (!msix) { 253 msi_set_message(pdev, msg); 254 trace_spapr_pci_msi_setup(pdev->name, 0, msg.address); 255 return; 256 } 257 258 for (i = 0; i < req_num; ++i) { 259 msix_set_message(pdev, i, msg); 260 trace_spapr_pci_msi_setup(pdev->name, i, msg.address); 261 if (addr) { 262 ++msg.data; 263 } 264 } 265 } 266 267 static void rtas_ibm_change_msi(PowerPCCPU *cpu, SpaprMachineState *spapr, 268 uint32_t token, uint32_t nargs, 269 target_ulong args, uint32_t nret, 270 target_ulong rets) 271 { 272 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr); 273 uint32_t config_addr = rtas_ld(args, 0); 274 uint64_t buid = rtas_ldq(args, 1); 275 unsigned int func = rtas_ld(args, 3); 276 unsigned int req_num = rtas_ld(args, 4); /* 0 == remove all */ 277 unsigned int seq_num = rtas_ld(args, 5); 278 unsigned int ret_intr_type; 279 unsigned int irq, max_irqs = 0; 280 SpaprPhbState *phb = NULL; 281 PCIDevice *pdev = NULL; 282 SpaprPciMsi *msi; 283 int *config_addr_key; 284 Error *err = NULL; 285 int i; 286 287 /* Fins SpaprPhbState */ 288 phb = spapr_pci_find_phb(spapr, buid); 289 if (phb) { 290 pdev = spapr_pci_find_dev(spapr, buid, config_addr); 291 } 292 if (!phb || !pdev) { 293 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR); 294 return; 295 } 296 297 switch (func) { 298 case RTAS_CHANGE_FN: 299 if (msi_present(pdev)) { 300 ret_intr_type = RTAS_TYPE_MSI; 301 } else if (msix_present(pdev)) { 302 ret_intr_type = RTAS_TYPE_MSIX; 303 } else { 304 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR); 305 return; 306 } 307 break; 308 case RTAS_CHANGE_MSI_FN: 309 if (msi_present(pdev)) { 310 ret_intr_type = RTAS_TYPE_MSI; 311 } else { 312 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR); 313 return; 314 } 315 break; 316 case RTAS_CHANGE_MSIX_FN: 317 if (msix_present(pdev)) { 318 ret_intr_type = RTAS_TYPE_MSIX; 319 } else { 320 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR); 321 return; 322 } 323 break; 324 default: 325 error_report("rtas_ibm_change_msi(%u) is not implemented", func); 326 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR); 327 return; 328 } 329 330 msi = (SpaprPciMsi *) g_hash_table_lookup(phb->msi, &config_addr); 331 332 /* Releasing MSIs */ 333 if (!req_num) { 334 if (!msi) { 335 trace_spapr_pci_msi("Releasing wrong config", config_addr); 336 rtas_st(rets, 0, RTAS_OUT_HW_ERROR); 337 return; 338 } 339 340 if (msi_present(pdev)) { 341 spapr_msi_setmsg(pdev, 0, false, 0, 0); 342 } 343 if (msix_present(pdev)) { 344 spapr_msi_setmsg(pdev, 0, true, 0, 0); 345 } 346 g_hash_table_remove(phb->msi, &config_addr); 347 348 trace_spapr_pci_msi("Released MSIs", config_addr); 349 rtas_st(rets, 0, RTAS_OUT_SUCCESS); 350 rtas_st(rets, 1, 0); 351 return; 352 } 353 354 /* Enabling MSI */ 355 356 /* Check if the device supports as many IRQs as requested */ 357 if (ret_intr_type == RTAS_TYPE_MSI) { 358 max_irqs = msi_nr_vectors_allocated(pdev); 359 } else if (ret_intr_type == RTAS_TYPE_MSIX) { 360 max_irqs = pdev->msix_entries_nr; 361 } 362 if (!max_irqs) { 363 error_report("Requested interrupt type %d is not enabled for device %x", 364 ret_intr_type, config_addr); 365 rtas_st(rets, 0, -1); /* Hardware error */ 366 return; 367 } 368 /* Correct the number if the guest asked for too many */ 369 if (req_num > max_irqs) { 370 trace_spapr_pci_msi_retry(config_addr, req_num, max_irqs); 371 req_num = max_irqs; 372 irq = 0; /* to avoid misleading trace */ 373 goto out; 374 } 375 376 /* Allocate MSIs */ 377 if (smc->legacy_irq_allocation) { 378 irq = spapr_irq_find(spapr, req_num, ret_intr_type == RTAS_TYPE_MSI, 379 &err); 380 } else { 381 irq = spapr_irq_msi_alloc(spapr, req_num, 382 ret_intr_type == RTAS_TYPE_MSI, &err); 383 } 384 if (err) { 385 error_reportf_err(err, "Can't allocate MSIs for device %x: ", 386 config_addr); 387 rtas_st(rets, 0, RTAS_OUT_HW_ERROR); 388 return; 389 } 390 391 for (i = 0; i < req_num; i++) { 392 spapr_irq_claim(spapr, irq + i, false, &err); 393 if (err) { 394 if (i) { 395 spapr_irq_free(spapr, irq, i); 396 } 397 if (!smc->legacy_irq_allocation) { 398 spapr_irq_msi_free(spapr, irq, req_num); 399 } 400 error_reportf_err(err, "Can't allocate MSIs for device %x: ", 401 config_addr); 402 rtas_st(rets, 0, RTAS_OUT_HW_ERROR); 403 return; 404 } 405 } 406 407 /* Release previous MSIs */ 408 if (msi) { 409 g_hash_table_remove(phb->msi, &config_addr); 410 } 411 412 /* Setup MSI/MSIX vectors in the device (via cfgspace or MSIX BAR) */ 413 spapr_msi_setmsg(pdev, SPAPR_PCI_MSI_WINDOW, ret_intr_type == RTAS_TYPE_MSIX, 414 irq, req_num); 415 416 /* Add MSI device to cache */ 417 msi = g_new(SpaprPciMsi, 1); 418 msi->first_irq = irq; 419 msi->num = req_num; 420 config_addr_key = g_new(int, 1); 421 *config_addr_key = config_addr; 422 g_hash_table_insert(phb->msi, config_addr_key, msi); 423 424 out: 425 rtas_st(rets, 0, RTAS_OUT_SUCCESS); 426 rtas_st(rets, 1, req_num); 427 rtas_st(rets, 2, ++seq_num); 428 if (nret > 3) { 429 rtas_st(rets, 3, ret_intr_type); 430 } 431 432 trace_spapr_pci_rtas_ibm_change_msi(config_addr, func, req_num, irq); 433 } 434 435 static void rtas_ibm_query_interrupt_source_number(PowerPCCPU *cpu, 436 SpaprMachineState *spapr, 437 uint32_t token, 438 uint32_t nargs, 439 target_ulong args, 440 uint32_t nret, 441 target_ulong rets) 442 { 443 uint32_t config_addr = rtas_ld(args, 0); 444 uint64_t buid = rtas_ldq(args, 1); 445 unsigned int intr_src_num = -1, ioa_intr_num = rtas_ld(args, 3); 446 SpaprPhbState *phb = NULL; 447 PCIDevice *pdev = NULL; 448 SpaprPciMsi *msi; 449 450 /* Find SpaprPhbState */ 451 phb = spapr_pci_find_phb(spapr, buid); 452 if (phb) { 453 pdev = spapr_pci_find_dev(spapr, buid, config_addr); 454 } 455 if (!phb || !pdev) { 456 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR); 457 return; 458 } 459 460 /* Find device descriptor and start IRQ */ 461 msi = (SpaprPciMsi *) g_hash_table_lookup(phb->msi, &config_addr); 462 if (!msi || !msi->first_irq || !msi->num || (ioa_intr_num >= msi->num)) { 463 trace_spapr_pci_msi("Failed to return vector", config_addr); 464 rtas_st(rets, 0, RTAS_OUT_HW_ERROR); 465 return; 466 } 467 intr_src_num = msi->first_irq + ioa_intr_num; 468 trace_spapr_pci_rtas_ibm_query_interrupt_source_number(ioa_intr_num, 469 intr_src_num); 470 471 rtas_st(rets, 0, RTAS_OUT_SUCCESS); 472 rtas_st(rets, 1, intr_src_num); 473 rtas_st(rets, 2, 1);/* 0 == level; 1 == edge */ 474 } 475 476 static void rtas_ibm_set_eeh_option(PowerPCCPU *cpu, 477 SpaprMachineState *spapr, 478 uint32_t token, uint32_t nargs, 479 target_ulong args, uint32_t nret, 480 target_ulong rets) 481 { 482 SpaprPhbState *sphb; 483 uint32_t addr, option; 484 uint64_t buid; 485 int ret; 486 487 if ((nargs != 4) || (nret != 1)) { 488 goto param_error_exit; 489 } 490 491 buid = rtas_ldq(args, 1); 492 addr = rtas_ld(args, 0); 493 option = rtas_ld(args, 3); 494 495 sphb = spapr_pci_find_phb(spapr, buid); 496 if (!sphb) { 497 goto param_error_exit; 498 } 499 500 if (!spapr_phb_eeh_available(sphb)) { 501 goto param_error_exit; 502 } 503 504 ret = spapr_phb_vfio_eeh_set_option(sphb, addr, option); 505 rtas_st(rets, 0, ret); 506 return; 507 508 param_error_exit: 509 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR); 510 } 511 512 static void rtas_ibm_get_config_addr_info2(PowerPCCPU *cpu, 513 SpaprMachineState *spapr, 514 uint32_t token, uint32_t nargs, 515 target_ulong args, uint32_t nret, 516 target_ulong rets) 517 { 518 SpaprPhbState *sphb; 519 PCIDevice *pdev; 520 uint32_t addr, option; 521 uint64_t buid; 522 523 if ((nargs != 4) || (nret != 2)) { 524 goto param_error_exit; 525 } 526 527 buid = rtas_ldq(args, 1); 528 sphb = spapr_pci_find_phb(spapr, buid); 529 if (!sphb) { 530 goto param_error_exit; 531 } 532 533 if (!spapr_phb_eeh_available(sphb)) { 534 goto param_error_exit; 535 } 536 537 /* 538 * We always have PE address of form "00BB0001". "BB" 539 * represents the bus number of PE's primary bus. 540 */ 541 option = rtas_ld(args, 3); 542 switch (option) { 543 case RTAS_GET_PE_ADDR: 544 addr = rtas_ld(args, 0); 545 pdev = spapr_pci_find_dev(spapr, buid, addr); 546 if (!pdev) { 547 goto param_error_exit; 548 } 549 550 rtas_st(rets, 1, (pci_bus_num(pci_get_bus(pdev)) << 16) + 1); 551 break; 552 case RTAS_GET_PE_MODE: 553 rtas_st(rets, 1, RTAS_PE_MODE_SHARED); 554 break; 555 default: 556 goto param_error_exit; 557 } 558 559 rtas_st(rets, 0, RTAS_OUT_SUCCESS); 560 return; 561 562 param_error_exit: 563 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR); 564 } 565 566 static void rtas_ibm_read_slot_reset_state2(PowerPCCPU *cpu, 567 SpaprMachineState *spapr, 568 uint32_t token, uint32_t nargs, 569 target_ulong args, uint32_t nret, 570 target_ulong rets) 571 { 572 SpaprPhbState *sphb; 573 uint64_t buid; 574 int state, ret; 575 576 if ((nargs != 3) || (nret != 4 && nret != 5)) { 577 goto param_error_exit; 578 } 579 580 buid = rtas_ldq(args, 1); 581 sphb = spapr_pci_find_phb(spapr, buid); 582 if (!sphb) { 583 goto param_error_exit; 584 } 585 586 if (!spapr_phb_eeh_available(sphb)) { 587 goto param_error_exit; 588 } 589 590 ret = spapr_phb_vfio_eeh_get_state(sphb, &state); 591 rtas_st(rets, 0, ret); 592 if (ret != RTAS_OUT_SUCCESS) { 593 return; 594 } 595 596 rtas_st(rets, 1, state); 597 rtas_st(rets, 2, RTAS_EEH_SUPPORT); 598 rtas_st(rets, 3, RTAS_EEH_PE_UNAVAIL_INFO); 599 if (nret >= 5) { 600 rtas_st(rets, 4, RTAS_EEH_PE_RECOVER_INFO); 601 } 602 return; 603 604 param_error_exit: 605 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR); 606 } 607 608 static void rtas_ibm_set_slot_reset(PowerPCCPU *cpu, 609 SpaprMachineState *spapr, 610 uint32_t token, uint32_t nargs, 611 target_ulong args, uint32_t nret, 612 target_ulong rets) 613 { 614 SpaprPhbState *sphb; 615 uint32_t option; 616 uint64_t buid; 617 int ret; 618 619 if ((nargs != 4) || (nret != 1)) { 620 goto param_error_exit; 621 } 622 623 buid = rtas_ldq(args, 1); 624 option = rtas_ld(args, 3); 625 sphb = spapr_pci_find_phb(spapr, buid); 626 if (!sphb) { 627 goto param_error_exit; 628 } 629 630 if (!spapr_phb_eeh_available(sphb)) { 631 goto param_error_exit; 632 } 633 634 ret = spapr_phb_vfio_eeh_reset(sphb, option); 635 rtas_st(rets, 0, ret); 636 return; 637 638 param_error_exit: 639 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR); 640 } 641 642 static void rtas_ibm_configure_pe(PowerPCCPU *cpu, 643 SpaprMachineState *spapr, 644 uint32_t token, uint32_t nargs, 645 target_ulong args, uint32_t nret, 646 target_ulong rets) 647 { 648 SpaprPhbState *sphb; 649 uint64_t buid; 650 int ret; 651 652 if ((nargs != 3) || (nret != 1)) { 653 goto param_error_exit; 654 } 655 656 buid = rtas_ldq(args, 1); 657 sphb = spapr_pci_find_phb(spapr, buid); 658 if (!sphb) { 659 goto param_error_exit; 660 } 661 662 if (!spapr_phb_eeh_available(sphb)) { 663 goto param_error_exit; 664 } 665 666 ret = spapr_phb_vfio_eeh_configure(sphb); 667 rtas_st(rets, 0, ret); 668 return; 669 670 param_error_exit: 671 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR); 672 } 673 674 /* To support it later */ 675 static void rtas_ibm_slot_error_detail(PowerPCCPU *cpu, 676 SpaprMachineState *spapr, 677 uint32_t token, uint32_t nargs, 678 target_ulong args, uint32_t nret, 679 target_ulong rets) 680 { 681 SpaprPhbState *sphb; 682 int option; 683 uint64_t buid; 684 685 if ((nargs != 8) || (nret != 1)) { 686 goto param_error_exit; 687 } 688 689 buid = rtas_ldq(args, 1); 690 sphb = spapr_pci_find_phb(spapr, buid); 691 if (!sphb) { 692 goto param_error_exit; 693 } 694 695 if (!spapr_phb_eeh_available(sphb)) { 696 goto param_error_exit; 697 } 698 699 option = rtas_ld(args, 7); 700 switch (option) { 701 case RTAS_SLOT_TEMP_ERR_LOG: 702 case RTAS_SLOT_PERM_ERR_LOG: 703 break; 704 default: 705 goto param_error_exit; 706 } 707 708 /* We don't have error log yet */ 709 rtas_st(rets, 0, RTAS_OUT_NO_ERRORS_FOUND); 710 return; 711 712 param_error_exit: 713 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR); 714 } 715 716 static void pci_spapr_set_irq(void *opaque, int irq_num, int level) 717 { 718 /* 719 * Here we use the number returned by pci_swizzle_map_irq_fn to find a 720 * corresponding qemu_irq. 721 */ 722 SpaprPhbState *phb = opaque; 723 SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine()); 724 725 trace_spapr_pci_lsi_set(phb->dtbusname, irq_num, phb->lsi_table[irq_num].irq); 726 qemu_set_irq(spapr_qirq(spapr, phb->lsi_table[irq_num].irq), level); 727 } 728 729 static PCIINTxRoute spapr_route_intx_pin_to_irq(void *opaque, int pin) 730 { 731 SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(opaque); 732 PCIINTxRoute route; 733 734 route.mode = PCI_INTX_ENABLED; 735 route.irq = sphb->lsi_table[pin].irq; 736 737 return route; 738 } 739 740 static uint64_t spapr_msi_read(void *opaque, hwaddr addr, unsigned size) 741 { 742 qemu_log_mask(LOG_GUEST_ERROR, "%s: invalid access\n", __func__); 743 return 0; 744 } 745 746 /* 747 * MSI/MSIX memory region implementation. 748 * The handler handles both MSI and MSIX. 749 * The vector number is encoded in least bits in data. 750 */ 751 static void spapr_msi_write(void *opaque, hwaddr addr, 752 uint64_t data, unsigned size) 753 { 754 SpaprMachineState *spapr = opaque; 755 uint32_t irq = data; 756 757 trace_spapr_pci_msi_write(addr, data, irq); 758 759 qemu_irq_pulse(spapr_qirq(spapr, irq)); 760 } 761 762 static const MemoryRegionOps spapr_msi_ops = { 763 /* 764 * .read result is undefined by PCI spec. 765 * define .read method to avoid assert failure in memory_region_init_io 766 */ 767 .read = spapr_msi_read, 768 .write = spapr_msi_write, 769 .endianness = DEVICE_LITTLE_ENDIAN 770 }; 771 772 /* 773 * PHB PCI device 774 */ 775 static AddressSpace *spapr_pci_dma_iommu(PCIBus *bus, void *opaque, int devfn) 776 { 777 SpaprPhbState *phb = opaque; 778 779 return &phb->iommu_as; 780 } 781 782 static const PCIIOMMUOps spapr_iommu_ops = { 783 .get_address_space = spapr_pci_dma_iommu, 784 }; 785 786 static char *spapr_phb_vfio_get_loc_code(SpaprPhbState *sphb, PCIDevice *pdev) 787 { 788 g_autofree char *path = NULL; 789 g_autofree char *host = NULL; 790 g_autofree char *devspec = NULL; 791 char *buf = NULL; 792 793 /* Get the PCI VFIO host id */ 794 host = object_property_get_str(OBJECT(pdev), "host", NULL); 795 if (!host) { 796 return NULL; 797 } 798 799 /* Construct the path of the file that will give us the DT location */ 800 path = g_strdup_printf("/sys/bus/pci/devices/%s/devspec", host); 801 if (!g_file_get_contents(path, &devspec, NULL, NULL)) { 802 return NULL; 803 } 804 805 /* Construct and read from host device tree the loc-code */ 806 g_free(path); 807 path = g_strdup_printf("/proc/device-tree%s/ibm,loc-code", devspec); 808 if (!g_file_get_contents(path, &buf, NULL, NULL)) { 809 return NULL; 810 } 811 return buf; 812 } 813 814 static char *spapr_phb_get_loc_code(SpaprPhbState *sphb, PCIDevice *pdev) 815 { 816 char *buf; 817 const char *devtype = "qemu"; 818 uint32_t busnr = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(pdev)))); 819 820 if (object_dynamic_cast(OBJECT(pdev), "vfio-pci")) { 821 buf = spapr_phb_vfio_get_loc_code(sphb, pdev); 822 if (buf) { 823 return buf; 824 } 825 devtype = "vfio"; 826 } 827 /* 828 * For emulated devices and VFIO-failure case, make up 829 * the loc-code. 830 */ 831 buf = g_strdup_printf("%s_%s:%04x:%02x:%02x.%x", 832 devtype, pdev->name, sphb->index, busnr, 833 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn)); 834 return buf; 835 } 836 837 /* Macros to operate with address in OF binding to PCI */ 838 #define b_x(x, p, l) (((x) & ((1<<(l))-1)) << (p)) 839 #define b_n(x) b_x((x), 31, 1) /* 0 if relocatable */ 840 #define b_p(x) b_x((x), 30, 1) /* 1 if prefetchable */ 841 #define b_t(x) b_x((x), 29, 1) /* 1 if the address is aliased */ 842 #define b_ss(x) b_x((x), 24, 2) /* the space code */ 843 #define b_bbbbbbbb(x) b_x((x), 16, 8) /* bus number */ 844 #define b_ddddd(x) b_x((x), 11, 5) /* device number */ 845 #define b_fff(x) b_x((x), 8, 3) /* function number */ 846 #define b_rrrrrrrr(x) b_x((x), 0, 8) /* register number */ 847 848 /* for 'reg' OF properties */ 849 #define RESOURCE_CELLS_SIZE 2 850 #define RESOURCE_CELLS_ADDRESS 3 851 852 typedef struct ResourceFields { 853 uint32_t phys_hi; 854 uint32_t phys_mid; 855 uint32_t phys_lo; 856 uint32_t size_hi; 857 uint32_t size_lo; 858 } QEMU_PACKED ResourceFields; 859 860 typedef struct ResourceProps { 861 ResourceFields reg[8]; 862 uint32_t reg_len; 863 } ResourceProps; 864 865 /* fill in the 'reg' OF properties for 866 * a PCI device. 'reg' describes resource requirements for a 867 * device's IO/MEM regions. 868 * 869 * the property is an array of ('phys-addr', 'size') pairs describing 870 * the addressable regions of the PCI device, where 'phys-addr' is a 871 * RESOURCE_CELLS_ADDRESS-tuple of 32-bit integers corresponding to 872 * (phys.hi, phys.mid, phys.lo), and 'size' is a 873 * RESOURCE_CELLS_SIZE-tuple corresponding to (size.hi, size.lo). 874 * 875 * phys.hi = 0xYYXXXXZZ, where: 876 * 0xYY = npt000ss 877 * ||| | 878 * ||| +-- space code 879 * ||| | 880 * ||| + 00 if configuration space 881 * ||| + 01 if IO region, 882 * ||| + 10 if 32-bit MEM region 883 * ||| + 11 if 64-bit MEM region 884 * ||| 885 * ||+------ for non-relocatable IO: 1 if aliased 886 * || for relocatable IO: 1 if below 64KB 887 * || for MEM: 1 if below 1MB 888 * |+------- 1 if region is prefetchable 889 * +-------- 1 if region is non-relocatable 890 * 0xXXXX = bbbbbbbb dddddfff, encoding bus, slot, and function 891 * bits respectively 892 * 0xZZ = rrrrrrrr, the register number of the BAR corresponding 893 * to the region 894 * 895 * phys.mid and phys.lo correspond respectively to the hi/lo portions 896 * of the actual address of the region. 897 * 898 * note also that addresses defined in this property are, at least 899 * for PAPR guests, relative to the PHBs IO/MEM windows, and 900 * correspond directly to the addresses in the BARs. 901 * 902 * in accordance with PCI Bus Binding to Open Firmware, 903 * IEEE Std 1275-1994, section 4.1.1, as implemented by PAPR+ v2.7, 904 * Appendix C. 905 */ 906 static void populate_resource_props(PCIDevice *d, ResourceProps *rp) 907 { 908 int bus_num = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(d)))); 909 uint32_t dev_id = (b_bbbbbbbb(bus_num) | 910 b_ddddd(PCI_SLOT(d->devfn)) | 911 b_fff(PCI_FUNC(d->devfn))); 912 ResourceFields *reg; 913 int i, reg_idx = 0; 914 915 /* config space region */ 916 reg = &rp->reg[reg_idx++]; 917 reg->phys_hi = cpu_to_be32(dev_id); 918 reg->phys_mid = 0; 919 reg->phys_lo = 0; 920 reg->size_hi = 0; 921 reg->size_lo = 0; 922 923 for (i = 0; i < PCI_NUM_REGIONS; i++) { 924 if (!d->io_regions[i].size) { 925 continue; 926 } 927 928 reg = &rp->reg[reg_idx++]; 929 930 reg->phys_hi = cpu_to_be32(dev_id | b_rrrrrrrr(pci_bar(d, i))); 931 if (d->io_regions[i].type & PCI_BASE_ADDRESS_SPACE_IO) { 932 reg->phys_hi |= cpu_to_be32(b_ss(1)); 933 } else if (d->io_regions[i].type & PCI_BASE_ADDRESS_MEM_TYPE_64) { 934 reg->phys_hi |= cpu_to_be32(b_ss(3)); 935 } else { 936 reg->phys_hi |= cpu_to_be32(b_ss(2)); 937 } 938 reg->phys_mid = 0; 939 reg->phys_lo = 0; 940 reg->size_hi = cpu_to_be32(d->io_regions[i].size >> 32); 941 reg->size_lo = cpu_to_be32(d->io_regions[i].size); 942 } 943 944 rp->reg_len = reg_idx * sizeof(ResourceFields); 945 } 946 947 typedef struct PCIClass PCIClass; 948 typedef struct PCISubClass PCISubClass; 949 typedef struct PCIIFace PCIIFace; 950 951 struct PCIIFace { 952 int iface; 953 const char *name; 954 }; 955 956 struct PCISubClass { 957 int subclass; 958 const char *name; 959 const PCIIFace *iface; 960 }; 961 962 struct PCIClass { 963 const char *name; 964 const PCISubClass *subc; 965 }; 966 967 static const PCISubClass undef_subclass[] = { 968 { PCI_CLASS_NOT_DEFINED_VGA, "display", NULL }, 969 { 0xFF, NULL, NULL }, 970 }; 971 972 static const PCISubClass mass_subclass[] = { 973 { PCI_CLASS_STORAGE_SCSI, "scsi", NULL }, 974 { PCI_CLASS_STORAGE_IDE, "ide", NULL }, 975 { PCI_CLASS_STORAGE_FLOPPY, "fdc", NULL }, 976 { PCI_CLASS_STORAGE_IPI, "ipi", NULL }, 977 { PCI_CLASS_STORAGE_RAID, "raid", NULL }, 978 { PCI_CLASS_STORAGE_ATA, "ata", NULL }, 979 { PCI_CLASS_STORAGE_SATA, "sata", NULL }, 980 { PCI_CLASS_STORAGE_SAS, "sas", NULL }, 981 { 0xFF, NULL, NULL }, 982 }; 983 984 static const PCISubClass net_subclass[] = { 985 { PCI_CLASS_NETWORK_ETHERNET, "ethernet", NULL }, 986 { PCI_CLASS_NETWORK_TOKEN_RING, "token-ring", NULL }, 987 { PCI_CLASS_NETWORK_FDDI, "fddi", NULL }, 988 { PCI_CLASS_NETWORK_ATM, "atm", NULL }, 989 { PCI_CLASS_NETWORK_ISDN, "isdn", NULL }, 990 { PCI_CLASS_NETWORK_WORLDFIP, "worldfip", NULL }, 991 { PCI_CLASS_NETWORK_PICMG214, "picmg", NULL }, 992 { 0xFF, NULL, NULL }, 993 }; 994 995 static const PCISubClass displ_subclass[] = { 996 { PCI_CLASS_DISPLAY_VGA, "vga", NULL }, 997 { PCI_CLASS_DISPLAY_XGA, "xga", NULL }, 998 { PCI_CLASS_DISPLAY_3D, "3d-controller", NULL }, 999 { 0xFF, NULL, NULL }, 1000 }; 1001 1002 static const PCISubClass media_subclass[] = { 1003 { PCI_CLASS_MULTIMEDIA_VIDEO, "video", NULL }, 1004 { PCI_CLASS_MULTIMEDIA_AUDIO, "sound", NULL }, 1005 { PCI_CLASS_MULTIMEDIA_PHONE, "telephony", NULL }, 1006 { 0xFF, NULL, NULL }, 1007 }; 1008 1009 static const PCISubClass mem_subclass[] = { 1010 { PCI_CLASS_MEMORY_RAM, "memory", NULL }, 1011 { PCI_CLASS_MEMORY_FLASH, "flash", NULL }, 1012 { 0xFF, NULL, NULL }, 1013 }; 1014 1015 static const PCISubClass bridg_subclass[] = { 1016 { PCI_CLASS_BRIDGE_HOST, "host", NULL }, 1017 { PCI_CLASS_BRIDGE_ISA, "isa", NULL }, 1018 { PCI_CLASS_BRIDGE_EISA, "eisa", NULL }, 1019 { PCI_CLASS_BRIDGE_MC, "mca", NULL }, 1020 { PCI_CLASS_BRIDGE_PCI, "pci", NULL }, 1021 { PCI_CLASS_BRIDGE_PCMCIA, "pcmcia", NULL }, 1022 { PCI_CLASS_BRIDGE_NUBUS, "nubus", NULL }, 1023 { PCI_CLASS_BRIDGE_CARDBUS, "cardbus", NULL }, 1024 { PCI_CLASS_BRIDGE_RACEWAY, "raceway", NULL }, 1025 { PCI_CLASS_BRIDGE_PCI_SEMITP, "semi-transparent-pci", NULL }, 1026 { PCI_CLASS_BRIDGE_IB_PCI, "infiniband", NULL }, 1027 { 0xFF, NULL, NULL }, 1028 }; 1029 1030 static const PCISubClass comm_subclass[] = { 1031 { PCI_CLASS_COMMUNICATION_SERIAL, "serial", NULL }, 1032 { PCI_CLASS_COMMUNICATION_PARALLEL, "parallel", NULL }, 1033 { PCI_CLASS_COMMUNICATION_MULTISERIAL, "multiport-serial", NULL }, 1034 { PCI_CLASS_COMMUNICATION_MODEM, "modem", NULL }, 1035 { PCI_CLASS_COMMUNICATION_GPIB, "gpib", NULL }, 1036 { PCI_CLASS_COMMUNICATION_SC, "smart-card", NULL }, 1037 { 0xFF, NULL, NULL, }, 1038 }; 1039 1040 static const PCIIFace pic_iface[] = { 1041 { PCI_CLASS_SYSTEM_PIC_IOAPIC, "io-apic" }, 1042 { PCI_CLASS_SYSTEM_PIC_IOXAPIC, "io-xapic" }, 1043 { 0xFF, NULL }, 1044 }; 1045 1046 static const PCISubClass sys_subclass[] = { 1047 { PCI_CLASS_SYSTEM_PIC, "interrupt-controller", pic_iface }, 1048 { PCI_CLASS_SYSTEM_DMA, "dma-controller", NULL }, 1049 { PCI_CLASS_SYSTEM_TIMER, "timer", NULL }, 1050 { PCI_CLASS_SYSTEM_RTC, "rtc", NULL }, 1051 { PCI_CLASS_SYSTEM_PCI_HOTPLUG, "hot-plug-controller", NULL }, 1052 { PCI_CLASS_SYSTEM_SDHCI, "sd-host-controller", NULL }, 1053 { 0xFF, NULL, NULL }, 1054 }; 1055 1056 static const PCISubClass inp_subclass[] = { 1057 { PCI_CLASS_INPUT_KEYBOARD, "keyboard", NULL }, 1058 { PCI_CLASS_INPUT_PEN, "pen", NULL }, 1059 { PCI_CLASS_INPUT_MOUSE, "mouse", NULL }, 1060 { PCI_CLASS_INPUT_SCANNER, "scanner", NULL }, 1061 { PCI_CLASS_INPUT_GAMEPORT, "gameport", NULL }, 1062 { 0xFF, NULL, NULL }, 1063 }; 1064 1065 static const PCISubClass dock_subclass[] = { 1066 { PCI_CLASS_DOCKING_GENERIC, "dock", NULL }, 1067 { 0xFF, NULL, NULL }, 1068 }; 1069 1070 static const PCISubClass cpu_subclass[] = { 1071 { PCI_CLASS_PROCESSOR_PENTIUM, "pentium", NULL }, 1072 { PCI_CLASS_PROCESSOR_POWERPC, "powerpc", NULL }, 1073 { PCI_CLASS_PROCESSOR_MIPS, "mips", NULL }, 1074 { PCI_CLASS_PROCESSOR_CO, "co-processor", NULL }, 1075 { 0xFF, NULL, NULL }, 1076 }; 1077 1078 static const PCIIFace usb_iface[] = { 1079 { PCI_CLASS_SERIAL_USB_UHCI, "usb-uhci" }, 1080 { PCI_CLASS_SERIAL_USB_OHCI, "usb-ohci", }, 1081 { PCI_CLASS_SERIAL_USB_EHCI, "usb-ehci" }, 1082 { PCI_CLASS_SERIAL_USB_XHCI, "usb-xhci" }, 1083 { PCI_CLASS_SERIAL_USB_UNKNOWN, "usb-unknown" }, 1084 { PCI_CLASS_SERIAL_USB_DEVICE, "usb-device" }, 1085 { 0xFF, NULL }, 1086 }; 1087 1088 static const PCISubClass ser_subclass[] = { 1089 { PCI_CLASS_SERIAL_FIREWIRE, "firewire", NULL }, 1090 { PCI_CLASS_SERIAL_ACCESS, "access-bus", NULL }, 1091 { PCI_CLASS_SERIAL_SSA, "ssa", NULL }, 1092 { PCI_CLASS_SERIAL_USB, "usb", usb_iface }, 1093 { PCI_CLASS_SERIAL_FIBER, "fibre-channel", NULL }, 1094 { PCI_CLASS_SERIAL_SMBUS, "smb", NULL }, 1095 { PCI_CLASS_SERIAL_IB, "infiniband", NULL }, 1096 { PCI_CLASS_SERIAL_IPMI, "ipmi", NULL }, 1097 { PCI_CLASS_SERIAL_SERCOS, "sercos", NULL }, 1098 { PCI_CLASS_SERIAL_CANBUS, "canbus", NULL }, 1099 { 0xFF, NULL, NULL }, 1100 }; 1101 1102 static const PCISubClass wrl_subclass[] = { 1103 { PCI_CLASS_WIRELESS_IRDA, "irda", NULL }, 1104 { PCI_CLASS_WIRELESS_CIR, "consumer-ir", NULL }, 1105 { PCI_CLASS_WIRELESS_RF_CONTROLLER, "rf-controller", NULL }, 1106 { PCI_CLASS_WIRELESS_BLUETOOTH, "bluetooth", NULL }, 1107 { PCI_CLASS_WIRELESS_BROADBAND, "broadband", NULL }, 1108 { 0xFF, NULL, NULL }, 1109 }; 1110 1111 static const PCISubClass sat_subclass[] = { 1112 { PCI_CLASS_SATELLITE_TV, "satellite-tv", NULL }, 1113 { PCI_CLASS_SATELLITE_AUDIO, "satellite-audio", NULL }, 1114 { PCI_CLASS_SATELLITE_VOICE, "satellite-voice", NULL }, 1115 { PCI_CLASS_SATELLITE_DATA, "satellite-data", NULL }, 1116 { 0xFF, NULL, NULL }, 1117 }; 1118 1119 static const PCISubClass crypt_subclass[] = { 1120 { PCI_CLASS_CRYPT_NETWORK, "network-encryption", NULL }, 1121 { PCI_CLASS_CRYPT_ENTERTAINMENT, 1122 "entertainment-encryption", NULL }, 1123 { 0xFF, NULL, NULL }, 1124 }; 1125 1126 static const PCISubClass spc_subclass[] = { 1127 { PCI_CLASS_SP_DPIO, "dpio", NULL }, 1128 { PCI_CLASS_SP_PERF, "counter", NULL }, 1129 { PCI_CLASS_SP_SYNCH, "measurement", NULL }, 1130 { PCI_CLASS_SP_MANAGEMENT, "management-card", NULL }, 1131 { 0xFF, NULL, NULL }, 1132 }; 1133 1134 static const PCIClass pci_classes[] = { 1135 { "legacy-device", undef_subclass }, 1136 { "mass-storage", mass_subclass }, 1137 { "network", net_subclass }, 1138 { "display", displ_subclass, }, 1139 { "multimedia-device", media_subclass }, 1140 { "memory-controller", mem_subclass }, 1141 { "unknown-bridge", bridg_subclass }, 1142 { "communication-controller", comm_subclass}, 1143 { "system-peripheral", sys_subclass }, 1144 { "input-controller", inp_subclass }, 1145 { "docking-station", dock_subclass }, 1146 { "cpu", cpu_subclass }, 1147 { "serial-bus", ser_subclass }, 1148 { "wireless-controller", wrl_subclass }, 1149 { "intelligent-io", NULL }, 1150 { "satellite-device", sat_subclass }, 1151 { "encryption", crypt_subclass }, 1152 { "data-processing-controller", spc_subclass }, 1153 }; 1154 1155 static const char *dt_name_from_class(uint8_t class, uint8_t subclass, 1156 uint8_t iface) 1157 { 1158 const PCIClass *pclass; 1159 const PCISubClass *psubclass; 1160 const PCIIFace *piface; 1161 const char *name; 1162 1163 if (class >= ARRAY_SIZE(pci_classes)) { 1164 return "pci"; 1165 } 1166 1167 pclass = pci_classes + class; 1168 name = pclass->name; 1169 1170 if (pclass->subc == NULL) { 1171 return name; 1172 } 1173 1174 psubclass = pclass->subc; 1175 while ((psubclass->subclass & 0xff) != 0xff) { 1176 if ((psubclass->subclass & 0xff) == subclass) { 1177 name = psubclass->name; 1178 break; 1179 } 1180 psubclass++; 1181 } 1182 1183 piface = psubclass->iface; 1184 if (piface == NULL) { 1185 return name; 1186 } 1187 while ((piface->iface & 0xff) != 0xff) { 1188 if ((piface->iface & 0xff) == iface) { 1189 name = piface->name; 1190 break; 1191 } 1192 piface++; 1193 } 1194 1195 return name; 1196 } 1197 1198 /* 1199 * DRC helper functions 1200 */ 1201 1202 static uint32_t drc_id_from_devfn(SpaprPhbState *phb, 1203 uint8_t chassis, int32_t devfn) 1204 { 1205 return (phb->index << 16) | (chassis << 8) | devfn; 1206 } 1207 1208 static SpaprDrc *drc_from_devfn(SpaprPhbState *phb, 1209 uint8_t chassis, int32_t devfn) 1210 { 1211 return spapr_drc_by_id(TYPE_SPAPR_DRC_PCI, 1212 drc_id_from_devfn(phb, chassis, devfn)); 1213 } 1214 1215 static uint8_t chassis_from_bus(PCIBus *bus) 1216 { 1217 if (pci_bus_is_root(bus)) { 1218 return 0; 1219 } else { 1220 PCIDevice *bridge = pci_bridge_get_device(bus); 1221 1222 return object_property_get_uint(OBJECT(bridge), "chassis_nr", 1223 &error_abort); 1224 } 1225 } 1226 1227 static SpaprDrc *drc_from_dev(SpaprPhbState *phb, PCIDevice *dev) 1228 { 1229 uint8_t chassis = chassis_from_bus(pci_get_bus(dev)); 1230 1231 return drc_from_devfn(phb, chassis, dev->devfn); 1232 } 1233 1234 static void add_drcs(SpaprPhbState *phb, PCIBus *bus) 1235 { 1236 Object *owner; 1237 int i; 1238 uint8_t chassis; 1239 1240 if (!phb->dr_enabled) { 1241 return; 1242 } 1243 1244 chassis = chassis_from_bus(bus); 1245 1246 if (pci_bus_is_root(bus)) { 1247 owner = OBJECT(phb); 1248 } else { 1249 owner = OBJECT(pci_bridge_get_device(bus)); 1250 } 1251 1252 for (i = 0; i < PCI_SLOT_MAX * PCI_FUNC_MAX; i++) { 1253 spapr_dr_connector_new(owner, TYPE_SPAPR_DRC_PCI, 1254 drc_id_from_devfn(phb, chassis, i)); 1255 } 1256 } 1257 1258 static void remove_drcs(SpaprPhbState *phb, PCIBus *bus) 1259 { 1260 int i; 1261 uint8_t chassis; 1262 1263 if (!phb->dr_enabled) { 1264 return; 1265 } 1266 1267 chassis = chassis_from_bus(bus); 1268 1269 for (i = PCI_SLOT_MAX * PCI_FUNC_MAX - 1; i >= 0; i--) { 1270 SpaprDrc *drc = drc_from_devfn(phb, chassis, i); 1271 1272 if (drc) { 1273 object_unparent(OBJECT(drc)); 1274 } 1275 } 1276 } 1277 1278 typedef struct PciWalkFdt { 1279 void *fdt; 1280 int offset; 1281 SpaprPhbState *sphb; 1282 int err; 1283 } PciWalkFdt; 1284 1285 static int spapr_dt_pci_device(SpaprPhbState *sphb, PCIDevice *dev, 1286 void *fdt, int parent_offset); 1287 1288 static void spapr_dt_pci_device_cb(PCIBus *bus, PCIDevice *pdev, 1289 void *opaque) 1290 { 1291 PciWalkFdt *p = opaque; 1292 int err; 1293 1294 if (p->err) { 1295 /* Something's already broken, don't keep going */ 1296 return; 1297 } 1298 1299 if (!pdev->enabled) { 1300 return; 1301 } 1302 1303 err = spapr_dt_pci_device(p->sphb, pdev, p->fdt, p->offset); 1304 if (err < 0) { 1305 p->err = err; 1306 } 1307 } 1308 1309 /* Augment PCI device node with bridge specific information */ 1310 static int spapr_dt_pci_bus(SpaprPhbState *sphb, PCIBus *bus, 1311 void *fdt, int offset) 1312 { 1313 Object *owner; 1314 PciWalkFdt cbinfo = { 1315 .fdt = fdt, 1316 .offset = offset, 1317 .sphb = sphb, 1318 .err = 0, 1319 }; 1320 int ret; 1321 1322 _FDT(fdt_setprop_cell(fdt, offset, "#address-cells", 1323 RESOURCE_CELLS_ADDRESS)); 1324 _FDT(fdt_setprop_cell(fdt, offset, "#size-cells", 1325 RESOURCE_CELLS_SIZE)); 1326 1327 assert(bus); 1328 pci_for_each_device_under_bus_reverse(bus, spapr_dt_pci_device_cb, &cbinfo); 1329 if (cbinfo.err) { 1330 return cbinfo.err; 1331 } 1332 1333 if (pci_bus_is_root(bus)) { 1334 owner = OBJECT(sphb); 1335 } else { 1336 owner = OBJECT(pci_bridge_get_device(bus)); 1337 } 1338 1339 ret = spapr_dt_drc(fdt, offset, owner, 1340 SPAPR_DR_CONNECTOR_TYPE_PCI); 1341 if (ret) { 1342 return ret; 1343 } 1344 1345 return offset; 1346 } 1347 1348 char *spapr_pci_fw_dev_name(PCIDevice *dev) 1349 { 1350 const gchar *basename; 1351 int slot = PCI_SLOT(dev->devfn); 1352 int func = PCI_FUNC(dev->devfn); 1353 uint32_t ccode = pci_default_read_config(dev, PCI_CLASS_PROG, 3); 1354 1355 basename = dt_name_from_class((ccode >> 16) & 0xff, (ccode >> 8) & 0xff, 1356 ccode & 0xff); 1357 1358 if (func != 0) { 1359 return g_strdup_printf("%s@%x,%x", basename, slot, func); 1360 } else { 1361 return g_strdup_printf("%s@%x", basename, slot); 1362 } 1363 } 1364 1365 /* create OF node for pci device and required OF DT properties */ 1366 static int spapr_dt_pci_device(SpaprPhbState *sphb, PCIDevice *dev, 1367 void *fdt, int parent_offset) 1368 { 1369 int offset; 1370 g_autofree gchar *nodename = spapr_pci_fw_dev_name(dev); 1371 ResourceProps rp; 1372 SpaprDrc *drc = drc_from_dev(sphb, dev); 1373 uint32_t vendor_id = pci_default_read_config(dev, PCI_VENDOR_ID, 2); 1374 uint32_t device_id = pci_default_read_config(dev, PCI_DEVICE_ID, 2); 1375 uint32_t revision_id = pci_default_read_config(dev, PCI_REVISION_ID, 1); 1376 uint32_t ccode = pci_default_read_config(dev, PCI_CLASS_PROG, 3); 1377 uint32_t irq_pin = pci_default_read_config(dev, PCI_INTERRUPT_PIN, 1); 1378 uint32_t subsystem_id = pci_default_read_config(dev, PCI_SUBSYSTEM_ID, 2); 1379 uint32_t subsystem_vendor_id = 1380 pci_default_read_config(dev, PCI_SUBSYSTEM_VENDOR_ID, 2); 1381 uint32_t cache_line_size = 1382 pci_default_read_config(dev, PCI_CACHE_LINE_SIZE, 1); 1383 uint32_t pci_status = pci_default_read_config(dev, PCI_STATUS, 2); 1384 gchar *loc_code; 1385 1386 _FDT(offset = fdt_add_subnode(fdt, parent_offset, nodename)); 1387 1388 /* in accordance with PAPR+ v2.7 13.6.3, Table 181 */ 1389 _FDT(fdt_setprop_cell(fdt, offset, "vendor-id", vendor_id)); 1390 _FDT(fdt_setprop_cell(fdt, offset, "device-id", device_id)); 1391 _FDT(fdt_setprop_cell(fdt, offset, "revision-id", revision_id)); 1392 1393 _FDT(fdt_setprop_cell(fdt, offset, "class-code", ccode)); 1394 if (irq_pin) { 1395 _FDT(fdt_setprop_cell(fdt, offset, "interrupts", irq_pin)); 1396 } 1397 1398 if (subsystem_id) { 1399 _FDT(fdt_setprop_cell(fdt, offset, "subsystem-id", subsystem_id)); 1400 } 1401 1402 if (subsystem_vendor_id) { 1403 _FDT(fdt_setprop_cell(fdt, offset, "subsystem-vendor-id", 1404 subsystem_vendor_id)); 1405 } 1406 1407 _FDT(fdt_setprop_cell(fdt, offset, "cache-line-size", cache_line_size)); 1408 1409 1410 /* the following fdt cells are masked off the pci status register */ 1411 _FDT(fdt_setprop_cell(fdt, offset, "devsel-speed", 1412 PCI_STATUS_DEVSEL_MASK & pci_status)); 1413 1414 if (pci_status & PCI_STATUS_FAST_BACK) { 1415 _FDT(fdt_setprop(fdt, offset, "fast-back-to-back", NULL, 0)); 1416 } 1417 if (pci_status & PCI_STATUS_66MHZ) { 1418 _FDT(fdt_setprop(fdt, offset, "66mhz-capable", NULL, 0)); 1419 } 1420 if (pci_status & PCI_STATUS_UDF) { 1421 _FDT(fdt_setprop(fdt, offset, "udf-supported", NULL, 0)); 1422 } 1423 1424 loc_code = spapr_phb_get_loc_code(sphb, dev); 1425 _FDT(fdt_setprop_string(fdt, offset, "ibm,loc-code", loc_code)); 1426 g_free(loc_code); 1427 1428 if (drc) { 1429 _FDT(fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", 1430 spapr_drc_index(drc))); 1431 } 1432 1433 if (msi_present(dev)) { 1434 uint32_t max_msi = msi_nr_vectors_allocated(dev); 1435 if (max_msi) { 1436 _FDT(fdt_setprop_cell(fdt, offset, "ibm,req#msi", max_msi)); 1437 } 1438 } 1439 if (msix_present(dev)) { 1440 uint32_t max_msix = dev->msix_entries_nr; 1441 if (max_msix) { 1442 _FDT(fdt_setprop_cell(fdt, offset, "ibm,req#msi-x", max_msix)); 1443 } 1444 } 1445 1446 populate_resource_props(dev, &rp); 1447 _FDT(fdt_setprop(fdt, offset, "reg", (uint8_t *)rp.reg, rp.reg_len)); 1448 1449 if (sphb->pcie_ecs && pci_is_express(dev)) { 1450 _FDT(fdt_setprop_cell(fdt, offset, "ibm,pci-config-space-type", 0x1)); 1451 } 1452 1453 if (!IS_PCI_BRIDGE(dev)) { 1454 /* Properties only for non-bridges */ 1455 uint32_t min_grant = pci_default_read_config(dev, PCI_MIN_GNT, 1); 1456 uint32_t max_latency = pci_default_read_config(dev, PCI_MAX_LAT, 1); 1457 _FDT(fdt_setprop_cell(fdt, offset, "min-grant", min_grant)); 1458 _FDT(fdt_setprop_cell(fdt, offset, "max-latency", max_latency)); 1459 return offset; 1460 } else { 1461 PCIBus *sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(dev)); 1462 1463 return spapr_dt_pci_bus(sphb, sec_bus, fdt, offset); 1464 } 1465 } 1466 1467 /* Callback to be called during DRC release. */ 1468 void spapr_phb_remove_pci_device_cb(DeviceState *dev) 1469 { 1470 HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev); 1471 1472 hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort); 1473 object_unparent(OBJECT(dev)); 1474 } 1475 1476 int spapr_pci_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr, 1477 void *fdt, int *fdt_start_offset, Error **errp) 1478 { 1479 HotplugHandler *plug_handler = qdev_get_hotplug_handler(drc->dev); 1480 SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(plug_handler); 1481 PCIDevice *pdev = PCI_DEVICE(drc->dev); 1482 1483 *fdt_start_offset = spapr_dt_pci_device(sphb, pdev, fdt, 0); 1484 return 0; 1485 } 1486 1487 static void spapr_pci_bridge_plug(SpaprPhbState *phb, 1488 PCIBridge *bridge) 1489 { 1490 PCIBus *bus = pci_bridge_get_sec_bus(bridge); 1491 1492 add_drcs(phb, bus); 1493 } 1494 1495 /* Returns non-zero if the value of "chassis_nr" is already in use */ 1496 static int check_chassis_nr(Object *obj, void *opaque) 1497 { 1498 int new_chassis_nr = 1499 object_property_get_uint(opaque, "chassis_nr", &error_abort); 1500 int chassis_nr = 1501 object_property_get_uint(obj, "chassis_nr", NULL); 1502 1503 if (!object_dynamic_cast(obj, TYPE_PCI_BRIDGE)) { 1504 return 0; 1505 } 1506 1507 /* Skip unsupported bridge types */ 1508 if (!chassis_nr) { 1509 return 0; 1510 } 1511 1512 /* Skip self */ 1513 if (obj == opaque) { 1514 return 0; 1515 } 1516 1517 return chassis_nr == new_chassis_nr; 1518 } 1519 1520 static bool bridge_has_valid_chassis_nr(Object *bridge, Error **errp) 1521 { 1522 int chassis_nr = 1523 object_property_get_uint(bridge, "chassis_nr", NULL); 1524 1525 /* 1526 * slotid_cap_init() already ensures that "chassis_nr" isn't null for 1527 * standard PCI bridges, so this really tells if "chassis_nr" is present 1528 * or not. 1529 */ 1530 if (!chassis_nr) { 1531 error_setg(errp, "PCI Bridge lacks a \"chassis_nr\" property"); 1532 error_append_hint(errp, "Try -device pci-bridge instead.\n"); 1533 return false; 1534 } 1535 1536 /* We want unique values for "chassis_nr" */ 1537 if (object_child_foreach_recursive(object_get_root(), check_chassis_nr, 1538 bridge)) { 1539 error_setg(errp, "Bridge chassis %d already in use", chassis_nr); 1540 return false; 1541 } 1542 1543 return true; 1544 } 1545 1546 static void spapr_pci_pre_plug(HotplugHandler *plug_handler, 1547 DeviceState *plugged_dev, Error **errp) 1548 { 1549 SpaprPhbState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler)); 1550 PCIDevice *pdev = PCI_DEVICE(plugged_dev); 1551 SpaprDrc *drc = drc_from_dev(phb, pdev); 1552 PCIBus *bus = PCI_BUS(qdev_get_parent_bus(DEVICE(pdev))); 1553 uint32_t slotnr = PCI_SLOT(pdev->devfn); 1554 1555 if (!phb->dr_enabled) { 1556 /* if this is a hotplug operation initiated by the user 1557 * we need to let them know it's not enabled 1558 */ 1559 if (plugged_dev->hotplugged) { 1560 error_setg(errp, "Bus '%s' does not support hotplugging", 1561 phb->parent_obj.bus->qbus.name); 1562 return; 1563 } 1564 } 1565 1566 if (IS_PCI_BRIDGE(plugged_dev)) { 1567 if (!bridge_has_valid_chassis_nr(OBJECT(plugged_dev), errp)) { 1568 return; 1569 } 1570 } 1571 1572 /* Following the QEMU convention used for PCIe multifunction 1573 * hotplug, we do not allow functions to be hotplugged to a 1574 * slot that already has function 0 present 1575 */ 1576 if (plugged_dev->hotplugged && 1577 !pci_is_vf(pdev) && 1578 bus->devices[PCI_DEVFN(slotnr, 0)] && 1579 PCI_FUNC(pdev->devfn) != 0) { 1580 error_setg(errp, "PCI: slot %d function 0 already occupied by %s," 1581 " additional functions can no longer be exposed to guest.", 1582 slotnr, bus->devices[PCI_DEVFN(slotnr, 0)]->name); 1583 } 1584 1585 if (drc && drc->dev) { 1586 error_setg(errp, "PCI: slot %d already occupied by %s", slotnr, 1587 pci_get_function_0(PCI_DEVICE(drc->dev))->name); 1588 return; 1589 } 1590 } 1591 1592 static void spapr_pci_plug(HotplugHandler *plug_handler, 1593 DeviceState *plugged_dev, Error **errp) 1594 { 1595 SpaprPhbState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler)); 1596 PCIDevice *pdev = PCI_DEVICE(plugged_dev); 1597 SpaprDrc *drc = drc_from_dev(phb, pdev); 1598 uint32_t slotnr = PCI_SLOT(pdev->devfn); 1599 1600 /* 1601 * If DR is disabled we don't need to do anything in the case of 1602 * hotplug or coldplug callbacks. 1603 */ 1604 if (!phb->dr_enabled) { 1605 return; 1606 } 1607 1608 g_assert(drc); 1609 1610 if (IS_PCI_BRIDGE(plugged_dev)) { 1611 spapr_pci_bridge_plug(phb, PCI_BRIDGE(plugged_dev)); 1612 } 1613 1614 /* spapr_pci_pre_plug() already checked the DRC is attachable */ 1615 spapr_drc_attach(drc, DEVICE(pdev)); 1616 1617 /* If this is function 0, signal hotplug for all the device functions. 1618 * Otherwise defer sending the hotplug event. 1619 */ 1620 if (!spapr_drc_hotplugged(plugged_dev)) { 1621 spapr_drc_reset(drc); 1622 } else if (PCI_FUNC(pdev->devfn) == 0) { 1623 int i; 1624 uint8_t chassis = chassis_from_bus(pci_get_bus(pdev)); 1625 1626 for (i = 0; i < 8; i++) { 1627 SpaprDrc *func_drc; 1628 SpaprDrcClass *func_drck; 1629 SpaprDREntitySense state; 1630 1631 func_drc = drc_from_devfn(phb, chassis, PCI_DEVFN(slotnr, i)); 1632 func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc); 1633 state = func_drck->dr_entity_sense(func_drc); 1634 1635 if (state == SPAPR_DR_ENTITY_SENSE_PRESENT) { 1636 spapr_hotplug_req_add_by_index(func_drc); 1637 } 1638 } 1639 } 1640 } 1641 1642 static void spapr_pci_bridge_unplug(SpaprPhbState *phb, 1643 PCIBridge *bridge) 1644 { 1645 PCIBus *bus = pci_bridge_get_sec_bus(bridge); 1646 1647 remove_drcs(phb, bus); 1648 } 1649 1650 static void spapr_pci_unplug(HotplugHandler *plug_handler, 1651 DeviceState *plugged_dev, Error **errp) 1652 { 1653 SpaprPhbState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler)); 1654 1655 /* some version guests do not wait for completion of a device 1656 * cleanup (generally done asynchronously by the kernel) before 1657 * signaling to QEMU that the device is safe, but instead sleep 1658 * for some 'safe' period of time. unfortunately on a busy host 1659 * this sleep isn't guaranteed to be long enough, resulting in 1660 * bad things like IRQ lines being left asserted during final 1661 * device removal. to deal with this we call reset just prior 1662 * to finalizing the device, which will put the device back into 1663 * an 'idle' state, as the device cleanup code expects. 1664 */ 1665 pci_device_reset(PCI_DEVICE(plugged_dev)); 1666 1667 if (IS_PCI_BRIDGE(plugged_dev)) { 1668 spapr_pci_bridge_unplug(phb, PCI_BRIDGE(plugged_dev)); 1669 return; 1670 } 1671 1672 qdev_unrealize(plugged_dev); 1673 } 1674 1675 static void spapr_pci_unplug_request(HotplugHandler *plug_handler, 1676 DeviceState *plugged_dev, Error **errp) 1677 { 1678 SpaprPhbState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler)); 1679 PCIDevice *pdev = PCI_DEVICE(plugged_dev); 1680 SpaprDrc *drc = drc_from_dev(phb, pdev); 1681 1682 if (!phb->dr_enabled) { 1683 error_setg(errp, "Bus '%s' does not support hotplugging", 1684 phb->parent_obj.bus->qbus.name); 1685 return; 1686 } 1687 1688 g_assert(drc); 1689 g_assert(drc->dev == plugged_dev); 1690 1691 if (!spapr_drc_unplug_requested(drc)) { 1692 uint32_t slotnr = PCI_SLOT(pdev->devfn); 1693 SpaprDrc *func_drc; 1694 SpaprDrcClass *func_drck; 1695 SpaprDREntitySense state; 1696 int i; 1697 uint8_t chassis = chassis_from_bus(pci_get_bus(pdev)); 1698 1699 if (IS_PCI_BRIDGE(plugged_dev)) { 1700 error_setg(errp, "PCI: Hot unplug of PCI bridges not supported"); 1701 return; 1702 } 1703 if (object_property_get_uint(OBJECT(pdev), "nvlink2-tgt", NULL)) { 1704 error_setg(errp, "PCI: Cannot unplug NVLink2 devices"); 1705 return; 1706 } 1707 1708 /* ensure any other present functions are pending unplug */ 1709 if (PCI_FUNC(pdev->devfn) == 0) { 1710 for (i = 1; i < 8; i++) { 1711 func_drc = drc_from_devfn(phb, chassis, PCI_DEVFN(slotnr, i)); 1712 func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc); 1713 state = func_drck->dr_entity_sense(func_drc); 1714 if (state == SPAPR_DR_ENTITY_SENSE_PRESENT 1715 && !spapr_drc_unplug_requested(func_drc)) { 1716 /* 1717 * Attempting to remove function 0 of a multifunction 1718 * device will will cascade into removing all child 1719 * functions, even if their unplug weren't requested 1720 * beforehand. 1721 */ 1722 spapr_drc_unplug_request(func_drc); 1723 } 1724 } 1725 } 1726 1727 spapr_drc_unplug_request(drc); 1728 1729 /* if this isn't func 0, defer unplug event. otherwise signal removal 1730 * for all present functions 1731 */ 1732 if (PCI_FUNC(pdev->devfn) == 0) { 1733 for (i = 7; i >= 0; i--) { 1734 func_drc = drc_from_devfn(phb, chassis, PCI_DEVFN(slotnr, i)); 1735 func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc); 1736 state = func_drck->dr_entity_sense(func_drc); 1737 if (state == SPAPR_DR_ENTITY_SENSE_PRESENT) { 1738 spapr_hotplug_req_remove_by_index(func_drc); 1739 } 1740 } 1741 } 1742 } else { 1743 error_setg(errp, 1744 "PCI device unplug already in progress for device %s", 1745 drc->dev->id); 1746 } 1747 } 1748 1749 static void spapr_phb_finalizefn(Object *obj) 1750 { 1751 SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(obj); 1752 1753 g_free(sphb->dtbusname); 1754 sphb->dtbusname = NULL; 1755 } 1756 1757 static void spapr_phb_unrealize(DeviceState *dev) 1758 { 1759 SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine()); 1760 SysBusDevice *s = SYS_BUS_DEVICE(dev); 1761 PCIHostState *phb = PCI_HOST_BRIDGE(s); 1762 SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(phb); 1763 SpaprTceTable *tcet; 1764 int i; 1765 const unsigned windows_supported = spapr_phb_windows_supported(sphb); 1766 1767 if (sphb->msi) { 1768 g_hash_table_unref(sphb->msi); 1769 sphb->msi = NULL; 1770 } 1771 1772 /* 1773 * Remove IO/MMIO subregions and aliases, rest should get cleaned 1774 * via PHB's unrealize->object_finalize 1775 */ 1776 for (i = windows_supported - 1; i >= 0; i--) { 1777 tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[i]); 1778 if (tcet) { 1779 memory_region_del_subregion(&sphb->iommu_root, 1780 spapr_tce_get_iommu(tcet)); 1781 } 1782 } 1783 1784 remove_drcs(sphb, phb->bus); 1785 1786 for (i = PCI_NUM_PINS - 1; i >= 0; i--) { 1787 if (sphb->lsi_table[i].irq) { 1788 spapr_irq_free(spapr, sphb->lsi_table[i].irq, 1); 1789 sphb->lsi_table[i].irq = 0; 1790 } 1791 } 1792 1793 QLIST_REMOVE(sphb, list); 1794 1795 memory_region_del_subregion(&sphb->iommu_root, &sphb->msiwindow); 1796 1797 /* 1798 * An attached PCI device may have memory listeners, eg. VFIO PCI. We have 1799 * unmapped all sections. Remove the listeners now, before destroying the 1800 * address space. 1801 */ 1802 address_space_remove_listeners(&sphb->iommu_as); 1803 address_space_destroy(&sphb->iommu_as); 1804 1805 qbus_set_hotplug_handler(BUS(phb->bus), NULL); 1806 pci_unregister_root_bus(phb->bus); 1807 1808 memory_region_del_subregion(get_system_memory(), &sphb->iowindow); 1809 if (sphb->mem64_win_pciaddr != (hwaddr)-1) { 1810 memory_region_del_subregion(get_system_memory(), &sphb->mem64window); 1811 } 1812 memory_region_del_subregion(get_system_memory(), &sphb->mem32window); 1813 } 1814 1815 static void spapr_phb_destroy_msi(gpointer opaque) 1816 { 1817 SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine()); 1818 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr); 1819 SpaprPciMsi *msi = opaque; 1820 1821 if (!smc->legacy_irq_allocation) { 1822 spapr_irq_msi_free(spapr, msi->first_irq, msi->num); 1823 } 1824 spapr_irq_free(spapr, msi->first_irq, msi->num); 1825 g_free(msi); 1826 } 1827 1828 static void spapr_phb_realize(DeviceState *dev, Error **errp) 1829 { 1830 ERRP_GUARD(); 1831 /* We don't use SPAPR_MACHINE() in order to exit gracefully if the user 1832 * tries to add a sPAPR PHB to a non-pseries machine. 1833 */ 1834 SpaprMachineState *spapr = 1835 (SpaprMachineState *) object_dynamic_cast(qdev_get_machine(), 1836 TYPE_SPAPR_MACHINE); 1837 SpaprMachineClass *smc = spapr ? SPAPR_MACHINE_GET_CLASS(spapr) : NULL; 1838 SysBusDevice *sbd = SYS_BUS_DEVICE(dev); 1839 SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(sbd); 1840 PCIHostState *phb = PCI_HOST_BRIDGE(sbd); 1841 MachineState *ms = MACHINE(spapr); 1842 char *namebuf; 1843 int i; 1844 PCIBus *bus; 1845 uint64_t msi_window_size = 4096; 1846 SpaprTceTable *tcet; 1847 const unsigned windows_supported = spapr_phb_windows_supported(sphb); 1848 1849 if (!spapr) { 1850 error_setg(errp, TYPE_SPAPR_PCI_HOST_BRIDGE " needs a pseries machine"); 1851 return; 1852 } 1853 1854 assert(sphb->index != (uint32_t)-1); /* checked in spapr_phb_pre_plug() */ 1855 1856 if (sphb->mem64_win_size != 0) { 1857 if (sphb->mem_win_size > SPAPR_PCI_MEM32_WIN_SIZE) { 1858 error_setg(errp, "32-bit memory window of size 0x%"HWADDR_PRIx 1859 " (max 2 GiB)", sphb->mem_win_size); 1860 return; 1861 } 1862 1863 /* 64-bit window defaults to identity mapping */ 1864 sphb->mem64_win_pciaddr = sphb->mem64_win_addr; 1865 } else if (sphb->mem_win_size > SPAPR_PCI_MEM32_WIN_SIZE) { 1866 /* 1867 * For compatibility with old configuration, if no 64-bit MMIO 1868 * window is specified, but the ordinary (32-bit) memory 1869 * window is specified as > 2GiB, we treat it as a 2GiB 32-bit 1870 * window, with a 64-bit MMIO window following on immediately 1871 * afterwards 1872 */ 1873 sphb->mem64_win_size = sphb->mem_win_size - SPAPR_PCI_MEM32_WIN_SIZE; 1874 sphb->mem64_win_addr = sphb->mem_win_addr + SPAPR_PCI_MEM32_WIN_SIZE; 1875 sphb->mem64_win_pciaddr = 1876 SPAPR_PCI_MEM_WIN_BUS_OFFSET + SPAPR_PCI_MEM32_WIN_SIZE; 1877 sphb->mem_win_size = SPAPR_PCI_MEM32_WIN_SIZE; 1878 } 1879 1880 if (spapr_pci_find_phb(spapr, sphb->buid)) { 1881 SpaprPhbState *s; 1882 1883 error_setg(errp, "PCI host bridges must have unique indexes"); 1884 error_append_hint(errp, "The following indexes are already in use:"); 1885 QLIST_FOREACH(s, &spapr->phbs, list) { 1886 error_append_hint(errp, " %d", s->index); 1887 } 1888 error_append_hint(errp, "\nTry another value for the index property\n"); 1889 return; 1890 } 1891 1892 if (sphb->numa_node != -1 && 1893 (sphb->numa_node >= MAX_NODES || 1894 !ms->numa_state->nodes[sphb->numa_node].present)) { 1895 error_setg(errp, "Invalid NUMA node ID for PCI host bridge"); 1896 return; 1897 } 1898 1899 sphb->dtbusname = g_strdup_printf("pci@%" PRIx64, sphb->buid); 1900 1901 /* Initialize memory regions */ 1902 namebuf = g_strdup_printf("%s.mmio", sphb->dtbusname); 1903 memory_region_init(&sphb->memspace, OBJECT(sphb), namebuf, UINT64_MAX); 1904 g_free(namebuf); 1905 1906 namebuf = g_strdup_printf("%s.mmio32-alias", sphb->dtbusname); 1907 memory_region_init_alias(&sphb->mem32window, OBJECT(sphb), 1908 namebuf, &sphb->memspace, 1909 SPAPR_PCI_MEM_WIN_BUS_OFFSET, sphb->mem_win_size); 1910 g_free(namebuf); 1911 memory_region_add_subregion(get_system_memory(), sphb->mem_win_addr, 1912 &sphb->mem32window); 1913 1914 if (sphb->mem64_win_size != 0) { 1915 namebuf = g_strdup_printf("%s.mmio64-alias", sphb->dtbusname); 1916 memory_region_init_alias(&sphb->mem64window, OBJECT(sphb), 1917 namebuf, &sphb->memspace, 1918 sphb->mem64_win_pciaddr, sphb->mem64_win_size); 1919 g_free(namebuf); 1920 1921 memory_region_add_subregion(get_system_memory(), 1922 sphb->mem64_win_addr, 1923 &sphb->mem64window); 1924 } 1925 1926 /* Initialize IO regions */ 1927 namebuf = g_strdup_printf("%s.io", sphb->dtbusname); 1928 memory_region_init(&sphb->iospace, OBJECT(sphb), 1929 namebuf, SPAPR_PCI_IO_WIN_SIZE); 1930 g_free(namebuf); 1931 1932 namebuf = g_strdup_printf("%s.io-alias", sphb->dtbusname); 1933 memory_region_init_alias(&sphb->iowindow, OBJECT(sphb), namebuf, 1934 &sphb->iospace, 0, SPAPR_PCI_IO_WIN_SIZE); 1935 g_free(namebuf); 1936 memory_region_add_subregion(get_system_memory(), sphb->io_win_addr, 1937 &sphb->iowindow); 1938 1939 bus = pci_register_root_bus(dev, NULL, 1940 pci_spapr_set_irq, pci_swizzle_map_irq_fn, sphb, 1941 &sphb->memspace, &sphb->iospace, 1942 PCI_DEVFN(0, 0), PCI_NUM_PINS, 1943 TYPE_PCI_BUS); 1944 1945 /* 1946 * Despite resembling a vanilla PCI bus in most ways, the PAPR 1947 * para-virtualized PCI bus *does* permit PCI-E extended config 1948 * space access 1949 */ 1950 if (sphb->pcie_ecs) { 1951 bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE; 1952 } 1953 phb->bus = bus; 1954 qbus_set_hotplug_handler(BUS(phb->bus), OBJECT(sphb)); 1955 1956 /* 1957 * Initialize PHB address space. 1958 * By default there will be at least one subregion for default 1959 * 32bit DMA window. 1960 * Later the guest might want to create another DMA window 1961 * which will become another memory subregion. 1962 */ 1963 namebuf = g_strdup_printf("%s.iommu-root", sphb->dtbusname); 1964 memory_region_init(&sphb->iommu_root, OBJECT(sphb), 1965 namebuf, UINT64_MAX); 1966 g_free(namebuf); 1967 address_space_init(&sphb->iommu_as, &sphb->iommu_root, 1968 sphb->dtbusname); 1969 1970 /* 1971 * As MSI/MSIX interrupts trigger by writing at MSI/MSIX vectors, 1972 * we need to allocate some memory to catch those writes coming 1973 * from msi_notify()/msix_notify(). 1974 * As MSIMessage:addr is going to be the same and MSIMessage:data 1975 * is going to be a VIRQ number, 4 bytes of the MSI MR will only 1976 * be used. 1977 * 1978 * For KVM we want to ensure that this memory is a full page so that 1979 * our memory slot is of page size granularity. 1980 */ 1981 if (kvm_enabled()) { 1982 msi_window_size = qemu_real_host_page_size(); 1983 } 1984 1985 memory_region_init_io(&sphb->msiwindow, OBJECT(sphb), &spapr_msi_ops, spapr, 1986 "msi", msi_window_size); 1987 memory_region_add_subregion(&sphb->iommu_root, SPAPR_PCI_MSI_WINDOW, 1988 &sphb->msiwindow); 1989 1990 pci_setup_iommu(bus, &spapr_iommu_ops, sphb); 1991 1992 pci_bus_set_route_irq_fn(bus, spapr_route_intx_pin_to_irq); 1993 1994 QLIST_INSERT_HEAD(&spapr->phbs, sphb, list); 1995 1996 /* Initialize the LSI table */ 1997 for (i = 0; i < PCI_NUM_PINS; i++) { 1998 int irq = SPAPR_IRQ_PCI_LSI + sphb->index * PCI_NUM_PINS + i; 1999 2000 if (smc->legacy_irq_allocation) { 2001 irq = spapr_irq_findone(spapr, errp); 2002 if (irq < 0) { 2003 error_prepend(errp, "can't allocate LSIs: "); 2004 /* 2005 * Older machines will never support PHB hotplug, ie, this is an 2006 * init only path and QEMU will terminate. No need to rollback. 2007 */ 2008 return; 2009 } 2010 } 2011 2012 if (spapr_irq_claim(spapr, irq, true, errp) < 0) { 2013 error_prepend(errp, "can't allocate LSIs: "); 2014 goto unrealize; 2015 } 2016 2017 sphb->lsi_table[i].irq = irq; 2018 } 2019 2020 /* allocate connectors for child PCI devices */ 2021 add_drcs(sphb, phb->bus); 2022 2023 /* DMA setup */ 2024 for (i = 0; i < windows_supported; ++i) { 2025 tcet = spapr_tce_new_table(DEVICE(sphb), sphb->dma_liobn[i]); 2026 if (!tcet) { 2027 error_setg(errp, "Creating window#%d failed for %s", 2028 i, sphb->dtbusname); 2029 goto unrealize; 2030 } 2031 memory_region_add_subregion(&sphb->iommu_root, 0, 2032 spapr_tce_get_iommu(tcet)); 2033 } 2034 2035 sphb->msi = g_hash_table_new_full(g_int_hash, g_int_equal, g_free, 2036 spapr_phb_destroy_msi); 2037 return; 2038 2039 unrealize: 2040 spapr_phb_unrealize(dev); 2041 } 2042 2043 static int spapr_phb_children_reset(Object *child, void *opaque) 2044 { 2045 DeviceState *dev = (DeviceState *) object_dynamic_cast(child, TYPE_DEVICE); 2046 2047 if (dev) { 2048 device_cold_reset(dev); 2049 } 2050 2051 return 0; 2052 } 2053 2054 void spapr_phb_dma_reset(SpaprPhbState *sphb) 2055 { 2056 int i; 2057 SpaprTceTable *tcet; 2058 2059 for (i = 0; i < SPAPR_PCI_DMA_MAX_WINDOWS; ++i) { 2060 tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[i]); 2061 2062 if (tcet && tcet->nb_table) { 2063 spapr_tce_table_disable(tcet); 2064 } 2065 } 2066 2067 /* Register default 32bit DMA window */ 2068 tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[0]); 2069 spapr_tce_table_enable(tcet, SPAPR_TCE_PAGE_SHIFT, sphb->dma_win_addr, 2070 sphb->dma_win_size >> SPAPR_TCE_PAGE_SHIFT); 2071 tcet->def_win = true; 2072 } 2073 2074 static void spapr_phb_reset(DeviceState *qdev) 2075 { 2076 SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(qdev); 2077 2078 spapr_phb_dma_reset(sphb); 2079 2080 /* Reset the IOMMU state */ 2081 object_child_foreach(OBJECT(qdev), spapr_phb_children_reset, NULL); 2082 2083 if (spapr_phb_eeh_available(SPAPR_PCI_HOST_BRIDGE(qdev))) { 2084 spapr_phb_vfio_reset(qdev); 2085 } 2086 2087 g_hash_table_remove_all(sphb->msi); 2088 } 2089 2090 static Property spapr_phb_properties[] = { 2091 DEFINE_PROP_UINT32("index", SpaprPhbState, index, -1), 2092 DEFINE_PROP_UINT64("mem_win_size", SpaprPhbState, mem_win_size, 2093 SPAPR_PCI_MEM32_WIN_SIZE), 2094 DEFINE_PROP_UINT64("mem64_win_size", SpaprPhbState, mem64_win_size, 2095 SPAPR_PCI_MEM64_WIN_SIZE), 2096 DEFINE_PROP_UINT64("io_win_size", SpaprPhbState, io_win_size, 2097 SPAPR_PCI_IO_WIN_SIZE), 2098 DEFINE_PROP_BOOL("dynamic-reconfiguration", SpaprPhbState, dr_enabled, 2099 true), 2100 /* Default DMA window is 0..1GB */ 2101 DEFINE_PROP_UINT64("dma_win_addr", SpaprPhbState, dma_win_addr, 0), 2102 DEFINE_PROP_UINT64("dma_win_size", SpaprPhbState, dma_win_size, 0x40000000), 2103 DEFINE_PROP_UINT64("dma64_win_addr", SpaprPhbState, dma64_win_addr, 2104 0x800000000000000ULL), 2105 DEFINE_PROP_BOOL("ddw", SpaprPhbState, ddw_enabled, true), 2106 DEFINE_PROP_UINT64("pgsz", SpaprPhbState, page_size_mask, 2107 (1ULL << 12) | (1ULL << 16) 2108 | (1ULL << 21) | (1ULL << 24)), 2109 DEFINE_PROP_UINT32("numa_node", SpaprPhbState, numa_node, -1), 2110 DEFINE_PROP_BOOL("pre-2.8-migration", SpaprPhbState, 2111 pre_2_8_migration, false), 2112 DEFINE_PROP_BOOL("pcie-extended-configuration-space", SpaprPhbState, 2113 pcie_ecs, true), 2114 DEFINE_PROP_BOOL("pre-5.1-associativity", SpaprPhbState, 2115 pre_5_1_assoc, false), 2116 DEFINE_PROP_END_OF_LIST(), 2117 }; 2118 2119 static const VMStateDescription vmstate_spapr_pci_lsi = { 2120 .name = "spapr_pci/lsi", 2121 .version_id = 1, 2122 .minimum_version_id = 1, 2123 .fields = (const VMStateField[]) { 2124 VMSTATE_UINT32_EQUAL(irq, SpaprPciLsi, NULL), 2125 2126 VMSTATE_END_OF_LIST() 2127 }, 2128 }; 2129 2130 static const VMStateDescription vmstate_spapr_pci_msi = { 2131 .name = "spapr_pci/msi", 2132 .version_id = 1, 2133 .minimum_version_id = 1, 2134 .fields = (const VMStateField []) { 2135 VMSTATE_UINT32(key, SpaprPciMsiMig), 2136 VMSTATE_UINT32(value.first_irq, SpaprPciMsiMig), 2137 VMSTATE_UINT32(value.num, SpaprPciMsiMig), 2138 VMSTATE_END_OF_LIST() 2139 }, 2140 }; 2141 2142 static int spapr_pci_pre_save(void *opaque) 2143 { 2144 SpaprPhbState *sphb = opaque; 2145 GHashTableIter iter; 2146 gpointer key, value; 2147 int i; 2148 2149 if (sphb->pre_2_8_migration) { 2150 sphb->mig_liobn = sphb->dma_liobn[0]; 2151 sphb->mig_mem_win_addr = sphb->mem_win_addr; 2152 sphb->mig_mem_win_size = sphb->mem_win_size; 2153 sphb->mig_io_win_addr = sphb->io_win_addr; 2154 sphb->mig_io_win_size = sphb->io_win_size; 2155 2156 if ((sphb->mem64_win_size != 0) 2157 && (sphb->mem64_win_addr 2158 == (sphb->mem_win_addr + sphb->mem_win_size))) { 2159 sphb->mig_mem_win_size += sphb->mem64_win_size; 2160 } 2161 } 2162 2163 g_free(sphb->msi_devs); 2164 sphb->msi_devs = NULL; 2165 sphb->msi_devs_num = g_hash_table_size(sphb->msi); 2166 if (!sphb->msi_devs_num) { 2167 return 0; 2168 } 2169 sphb->msi_devs = g_new(SpaprPciMsiMig, sphb->msi_devs_num); 2170 2171 g_hash_table_iter_init(&iter, sphb->msi); 2172 for (i = 0; g_hash_table_iter_next(&iter, &key, &value); ++i) { 2173 sphb->msi_devs[i].key = *(uint32_t *) key; 2174 sphb->msi_devs[i].value = *(SpaprPciMsi *) value; 2175 } 2176 2177 return 0; 2178 } 2179 2180 static int spapr_pci_post_save(void *opaque) 2181 { 2182 SpaprPhbState *sphb = opaque; 2183 2184 g_free(sphb->msi_devs); 2185 sphb->msi_devs = NULL; 2186 sphb->msi_devs_num = 0; 2187 return 0; 2188 } 2189 2190 static int spapr_pci_post_load(void *opaque, int version_id) 2191 { 2192 SpaprPhbState *sphb = opaque; 2193 gpointer key, value; 2194 int i; 2195 2196 for (i = 0; i < sphb->msi_devs_num; ++i) { 2197 key = g_memdup2(&sphb->msi_devs[i].key, sizeof(sphb->msi_devs[i].key)); 2198 value = g_memdup2(&sphb->msi_devs[i].value, 2199 sizeof(sphb->msi_devs[i].value)); 2200 g_hash_table_insert(sphb->msi, key, value); 2201 } 2202 g_free(sphb->msi_devs); 2203 sphb->msi_devs = NULL; 2204 sphb->msi_devs_num = 0; 2205 2206 return 0; 2207 } 2208 2209 static bool pre_2_8_migration(void *opaque, int version_id) 2210 { 2211 SpaprPhbState *sphb = opaque; 2212 2213 return sphb->pre_2_8_migration; 2214 } 2215 2216 static const VMStateDescription vmstate_spapr_pci = { 2217 .name = "spapr_pci", 2218 .version_id = 2, 2219 .minimum_version_id = 2, 2220 .pre_save = spapr_pci_pre_save, 2221 .post_save = spapr_pci_post_save, 2222 .post_load = spapr_pci_post_load, 2223 .fields = (const VMStateField[]) { 2224 VMSTATE_UINT64_EQUAL(buid, SpaprPhbState, NULL), 2225 VMSTATE_UINT32_TEST(mig_liobn, SpaprPhbState, pre_2_8_migration), 2226 VMSTATE_UINT64_TEST(mig_mem_win_addr, SpaprPhbState, pre_2_8_migration), 2227 VMSTATE_UINT64_TEST(mig_mem_win_size, SpaprPhbState, pre_2_8_migration), 2228 VMSTATE_UINT64_TEST(mig_io_win_addr, SpaprPhbState, pre_2_8_migration), 2229 VMSTATE_UINT64_TEST(mig_io_win_size, SpaprPhbState, pre_2_8_migration), 2230 VMSTATE_STRUCT_ARRAY(lsi_table, SpaprPhbState, PCI_NUM_PINS, 0, 2231 vmstate_spapr_pci_lsi, SpaprPciLsi), 2232 VMSTATE_INT32(msi_devs_num, SpaprPhbState), 2233 VMSTATE_STRUCT_VARRAY_ALLOC(msi_devs, SpaprPhbState, msi_devs_num, 0, 2234 vmstate_spapr_pci_msi, SpaprPciMsiMig), 2235 VMSTATE_END_OF_LIST() 2236 }, 2237 }; 2238 2239 static const char *spapr_phb_root_bus_path(PCIHostState *host_bridge, 2240 PCIBus *rootbus) 2241 { 2242 SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(host_bridge); 2243 2244 return sphb->dtbusname; 2245 } 2246 2247 static void spapr_phb_class_init(ObjectClass *klass, void *data) 2248 { 2249 PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_CLASS(klass); 2250 DeviceClass *dc = DEVICE_CLASS(klass); 2251 HotplugHandlerClass *hp = HOTPLUG_HANDLER_CLASS(klass); 2252 2253 hc->root_bus_path = spapr_phb_root_bus_path; 2254 dc->realize = spapr_phb_realize; 2255 dc->unrealize = spapr_phb_unrealize; 2256 device_class_set_props(dc, spapr_phb_properties); 2257 dc->reset = spapr_phb_reset; 2258 dc->vmsd = &vmstate_spapr_pci; 2259 /* Supported by TYPE_SPAPR_MACHINE */ 2260 dc->user_creatable = true; 2261 set_bit(DEVICE_CATEGORY_BRIDGE, dc->categories); 2262 hp->pre_plug = spapr_pci_pre_plug; 2263 hp->plug = spapr_pci_plug; 2264 hp->unplug = spapr_pci_unplug; 2265 hp->unplug_request = spapr_pci_unplug_request; 2266 } 2267 2268 static const TypeInfo spapr_phb_info = { 2269 .name = TYPE_SPAPR_PCI_HOST_BRIDGE, 2270 .parent = TYPE_PCI_HOST_BRIDGE, 2271 .instance_size = sizeof(SpaprPhbState), 2272 .instance_finalize = spapr_phb_finalizefn, 2273 .class_init = spapr_phb_class_init, 2274 .interfaces = (InterfaceInfo[]) { 2275 { TYPE_HOTPLUG_HANDLER }, 2276 { } 2277 } 2278 }; 2279 2280 static void spapr_phb_pci_enumerate_bridge(PCIBus *bus, PCIDevice *pdev, 2281 void *opaque) 2282 { 2283 unsigned int *bus_no = opaque; 2284 PCIBus *sec_bus = NULL; 2285 2286 if ((pci_default_read_config(pdev, PCI_HEADER_TYPE, 1) != 2287 PCI_HEADER_TYPE_BRIDGE)) { 2288 return; 2289 } 2290 2291 (*bus_no)++; 2292 pci_default_write_config(pdev, PCI_PRIMARY_BUS, pci_dev_bus_num(pdev), 1); 2293 pci_default_write_config(pdev, PCI_SECONDARY_BUS, *bus_no, 1); 2294 pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, *bus_no, 1); 2295 2296 sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(pdev)); 2297 if (!sec_bus) { 2298 return; 2299 } 2300 2301 pci_for_each_device_under_bus(sec_bus, spapr_phb_pci_enumerate_bridge, 2302 bus_no); 2303 pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, *bus_no, 1); 2304 } 2305 2306 static void spapr_phb_pci_enumerate(SpaprPhbState *phb) 2307 { 2308 PCIBus *bus = PCI_HOST_BRIDGE(phb)->bus; 2309 unsigned int bus_no = 0; 2310 2311 pci_for_each_device_under_bus(bus, spapr_phb_pci_enumerate_bridge, 2312 &bus_no); 2313 2314 } 2315 2316 int spapr_dt_phb(SpaprMachineState *spapr, SpaprPhbState *phb, 2317 uint32_t intc_phandle, void *fdt, int *node_offset) 2318 { 2319 int bus_off, i, j, ret; 2320 uint32_t bus_range[] = { cpu_to_be32(0), cpu_to_be32(0xff) }; 2321 struct { 2322 uint32_t hi; 2323 uint64_t child; 2324 uint64_t parent; 2325 uint64_t size; 2326 } QEMU_PACKED ranges[] = { 2327 { 2328 cpu_to_be32(b_ss(1)), cpu_to_be64(0), 2329 cpu_to_be64(phb->io_win_addr), 2330 cpu_to_be64(memory_region_size(&phb->iospace)), 2331 }, 2332 { 2333 cpu_to_be32(b_ss(2)), cpu_to_be64(SPAPR_PCI_MEM_WIN_BUS_OFFSET), 2334 cpu_to_be64(phb->mem_win_addr), 2335 cpu_to_be64(phb->mem_win_size), 2336 }, 2337 { 2338 cpu_to_be32(b_ss(3)), cpu_to_be64(phb->mem64_win_pciaddr), 2339 cpu_to_be64(phb->mem64_win_addr), 2340 cpu_to_be64(phb->mem64_win_size), 2341 }, 2342 }; 2343 const unsigned sizeof_ranges = 2344 (phb->mem64_win_size ? 3 : 2) * sizeof(ranges[0]); 2345 uint64_t bus_reg[] = { cpu_to_be64(phb->buid), 0 }; 2346 uint32_t interrupt_map_mask[] = { 2347 cpu_to_be32(b_ddddd(-1)|b_fff(0)), 0x0, 0x0, cpu_to_be32(-1)}; 2348 uint32_t interrupt_map[PCI_SLOT_MAX * PCI_NUM_PINS][7]; 2349 uint32_t ddw_applicable[] = { 2350 cpu_to_be32(RTAS_IBM_QUERY_PE_DMA_WINDOW), 2351 cpu_to_be32(RTAS_IBM_CREATE_PE_DMA_WINDOW), 2352 cpu_to_be32(RTAS_IBM_REMOVE_PE_DMA_WINDOW) 2353 }; 2354 uint32_t ddw_extensions[] = { 2355 cpu_to_be32(2), 2356 cpu_to_be32(RTAS_IBM_RESET_PE_DMA_WINDOW), 2357 cpu_to_be32(1), /* 1: ibm,query-pe-dma-window 6 outputs, PAPR 2.8 */ 2358 }; 2359 SpaprTceTable *tcet; 2360 SpaprDrc *drc; 2361 2362 /* Start populating the FDT */ 2363 _FDT(bus_off = fdt_add_subnode(fdt, 0, phb->dtbusname)); 2364 if (node_offset) { 2365 *node_offset = bus_off; 2366 } 2367 2368 /* Write PHB properties */ 2369 _FDT(fdt_setprop_string(fdt, bus_off, "device_type", "pci")); 2370 _FDT(fdt_setprop_string(fdt, bus_off, "compatible", "IBM,Logical_PHB")); 2371 _FDT(fdt_setprop_cell(fdt, bus_off, "#interrupt-cells", 0x1)); 2372 _FDT(fdt_setprop(fdt, bus_off, "used-by-rtas", NULL, 0)); 2373 _FDT(fdt_setprop(fdt, bus_off, "bus-range", &bus_range, sizeof(bus_range))); 2374 _FDT(fdt_setprop(fdt, bus_off, "ranges", &ranges, sizeof_ranges)); 2375 _FDT(fdt_setprop(fdt, bus_off, "reg", &bus_reg, sizeof(bus_reg))); 2376 _FDT(fdt_setprop_cell(fdt, bus_off, "ibm,pci-config-space-type", 0x1)); 2377 _FDT(fdt_setprop_cell(fdt, bus_off, "ibm,pe-total-#msi", 2378 spapr_irq_nr_msis(spapr))); 2379 2380 /* Dynamic DMA window */ 2381 if (phb->ddw_enabled) { 2382 _FDT(fdt_setprop(fdt, bus_off, "ibm,ddw-applicable", &ddw_applicable, 2383 sizeof(ddw_applicable))); 2384 _FDT(fdt_setprop(fdt, bus_off, "ibm,ddw-extensions", 2385 &ddw_extensions, sizeof(ddw_extensions))); 2386 } 2387 2388 /* Advertise NUMA via ibm,associativity */ 2389 if (phb->numa_node != -1) { 2390 spapr_numa_write_associativity_dt(spapr, fdt, bus_off, phb->numa_node); 2391 } 2392 2393 /* Build the interrupt-map, this must matches what is done 2394 * in pci_swizzle_map_irq_fn 2395 */ 2396 _FDT(fdt_setprop(fdt, bus_off, "interrupt-map-mask", 2397 &interrupt_map_mask, sizeof(interrupt_map_mask))); 2398 for (i = 0; i < PCI_SLOT_MAX; i++) { 2399 for (j = 0; j < PCI_NUM_PINS; j++) { 2400 uint32_t *irqmap = interrupt_map[i*PCI_NUM_PINS + j]; 2401 int lsi_num = pci_swizzle(i, j); 2402 2403 irqmap[0] = cpu_to_be32(b_ddddd(i)|b_fff(0)); 2404 irqmap[1] = 0; 2405 irqmap[2] = 0; 2406 irqmap[3] = cpu_to_be32(j+1); 2407 irqmap[4] = cpu_to_be32(intc_phandle); 2408 spapr_dt_irq(&irqmap[5], phb->lsi_table[lsi_num].irq, true); 2409 } 2410 } 2411 /* Write interrupt map */ 2412 _FDT(fdt_setprop(fdt, bus_off, "interrupt-map", &interrupt_map, 2413 sizeof(interrupt_map))); 2414 2415 tcet = spapr_tce_find_by_liobn(phb->dma_liobn[0]); 2416 if (!tcet) { 2417 return -1; 2418 } 2419 spapr_dma_dt(fdt, bus_off, "ibm,dma-window", 2420 tcet->liobn, tcet->bus_offset, 2421 tcet->nb_table << tcet->page_shift); 2422 2423 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, phb->index); 2424 if (drc) { 2425 uint32_t drc_index = cpu_to_be32(spapr_drc_index(drc)); 2426 2427 _FDT(fdt_setprop(fdt, bus_off, "ibm,my-drc-index", &drc_index, 2428 sizeof(drc_index))); 2429 } 2430 2431 /* Walk the bridges and program the bus numbers*/ 2432 spapr_phb_pci_enumerate(phb); 2433 _FDT(fdt_setprop_cell(fdt, bus_off, "qemu,phb-enumerated", 0x1)); 2434 2435 /* Walk the bridge and subordinate buses */ 2436 ret = spapr_dt_pci_bus(phb, PCI_HOST_BRIDGE(phb)->bus, fdt, bus_off); 2437 if (ret < 0) { 2438 return ret; 2439 } 2440 2441 return 0; 2442 } 2443 2444 void spapr_pci_rtas_init(void) 2445 { 2446 spapr_rtas_register(RTAS_READ_PCI_CONFIG, "read-pci-config", 2447 rtas_read_pci_config); 2448 spapr_rtas_register(RTAS_WRITE_PCI_CONFIG, "write-pci-config", 2449 rtas_write_pci_config); 2450 spapr_rtas_register(RTAS_IBM_READ_PCI_CONFIG, "ibm,read-pci-config", 2451 rtas_ibm_read_pci_config); 2452 spapr_rtas_register(RTAS_IBM_WRITE_PCI_CONFIG, "ibm,write-pci-config", 2453 rtas_ibm_write_pci_config); 2454 if (msi_nonbroken) { 2455 spapr_rtas_register(RTAS_IBM_QUERY_INTERRUPT_SOURCE_NUMBER, 2456 "ibm,query-interrupt-source-number", 2457 rtas_ibm_query_interrupt_source_number); 2458 spapr_rtas_register(RTAS_IBM_CHANGE_MSI, "ibm,change-msi", 2459 rtas_ibm_change_msi); 2460 } 2461 2462 spapr_rtas_register(RTAS_IBM_SET_EEH_OPTION, 2463 "ibm,set-eeh-option", 2464 rtas_ibm_set_eeh_option); 2465 spapr_rtas_register(RTAS_IBM_GET_CONFIG_ADDR_INFO2, 2466 "ibm,get-config-addr-info2", 2467 rtas_ibm_get_config_addr_info2); 2468 spapr_rtas_register(RTAS_IBM_READ_SLOT_RESET_STATE2, 2469 "ibm,read-slot-reset-state2", 2470 rtas_ibm_read_slot_reset_state2); 2471 spapr_rtas_register(RTAS_IBM_SET_SLOT_RESET, 2472 "ibm,set-slot-reset", 2473 rtas_ibm_set_slot_reset); 2474 spapr_rtas_register(RTAS_IBM_CONFIGURE_PE, 2475 "ibm,configure-pe", 2476 rtas_ibm_configure_pe); 2477 spapr_rtas_register(RTAS_IBM_SLOT_ERROR_DETAIL, 2478 "ibm,slot-error-detail", 2479 rtas_ibm_slot_error_detail); 2480 } 2481 2482 static void spapr_pci_register_types(void) 2483 { 2484 type_register_static(&spapr_phb_info); 2485 } 2486 2487 type_init(spapr_pci_register_types) 2488 2489 static int spapr_switch_one_vga(DeviceState *dev, void *opaque) 2490 { 2491 bool be = *(bool *)opaque; 2492 2493 if (object_dynamic_cast(OBJECT(dev), "VGA") 2494 || object_dynamic_cast(OBJECT(dev), "secondary-vga") 2495 || object_dynamic_cast(OBJECT(dev), "bochs-display") 2496 || object_dynamic_cast(OBJECT(dev), "virtio-vga")) { 2497 object_property_set_bool(OBJECT(dev), "big-endian-framebuffer", be, 2498 &error_abort); 2499 } 2500 return 0; 2501 } 2502 2503 void spapr_pci_switch_vga(SpaprMachineState *spapr, bool big_endian) 2504 { 2505 SpaprPhbState *sphb; 2506 2507 /* 2508 * For backward compatibility with existing guests, we switch 2509 * the endianness of the VGA controller when changing the guest 2510 * interrupt mode 2511 */ 2512 QLIST_FOREACH(sphb, &spapr->phbs, list) { 2513 BusState *bus = &PCI_HOST_BRIDGE(sphb)->bus->qbus; 2514 qbus_walk_children(bus, spapr_switch_one_vga, NULL, NULL, NULL, 2515 &big_endian); 2516 } 2517 } 2518