hw/s390x/s390-pci-bus.c

   1 /*
   2  * s390 PCI BUS
   3  *
   4  * Copyright 2014 IBM Corp.
   5  * Author(s): Frank Blaschka <frank.blaschka@de.ibm.com>
   6  *            Hong Bo Li <lihbbj@cn.ibm.com>
   7  *            Yi Min Zhao <zyimin@cn.ibm.com>
   8  *
   9  * This work is licensed under the terms of the GNU GPL, version 2 or (at
  10  * your option) any later version. See the COPYING file in the top-level
  11  * directory.
  12  */
  13
  14 #include "qemu/osdep.h"
  15 #include "qapi/error.h"
  16 #include "qapi/visitor.h"
  17 #include "hw/s390x/s390-pci-bus.h"
  18 #include "hw/s390x/s390-pci-inst.h"
  19 #include "hw/s390x/s390-pci-kvm.h"
  20 #include "hw/s390x/s390-pci-vfio.h"
  21 #include "hw/pci/pci_bus.h"
  22 #include "hw/qdev-properties.h"
  23 #include "hw/pci/pci_bridge.h"
  24 #include "hw/pci/msi.h"
  25 #include "qemu/error-report.h"
  26 #include "qemu/module.h"
  27 #include "sysemu/reset.h"
  28 #include "sysemu/runstate.h"
  29
  30 #include "trace.h"
  31
  32 S390pciState *s390_get_phb(void)
  33 {
  34     static S390pciState *phb;
  35
  36     if (!phb) {
  37         phb = S390_PCI_HOST_BRIDGE(
  38             object_resolve_path(TYPE_S390_PCI_HOST_BRIDGE, NULL));
  39         assert(phb != NULL);
  40     }
  41
  42     return phb;
  43 }
  44
  45 int pci_chsc_sei_nt2_get_event(void *res)
  46 {
  47     ChscSeiNt2Res *nt2_res = (ChscSeiNt2Res *)res;
  48     PciCcdfAvail *accdf;
  49     PciCcdfErr *eccdf;
  50     int rc = 1;
  51     SeiContainer *sei_cont;
  52     S390pciState *s = s390_get_phb();
  53
  54     sei_cont = QTAILQ_FIRST(&s->pending_sei);
  55     if (sei_cont) {
  56         QTAILQ_REMOVE(&s->pending_sei, sei_cont, link);
  57         nt2_res->nt = 2;
  58         nt2_res->cc = sei_cont->cc;
  59         nt2_res->length = cpu_to_be16(sizeof(ChscSeiNt2Res));
  60         switch (sei_cont->cc) {
  61         case 1: /* error event */
  62             eccdf = (PciCcdfErr *)nt2_res->ccdf;
  63             eccdf->fid = cpu_to_be32(sei_cont->fid);
  64             eccdf->fh = cpu_to_be32(sei_cont->fh);
  65             eccdf->e = cpu_to_be32(sei_cont->e);
  66             eccdf->faddr = cpu_to_be64(sei_cont->faddr);
  67             eccdf->pec = cpu_to_be16(sei_cont->pec);
  68             break;
  69         case 2: /* availability event */
  70             accdf = (PciCcdfAvail *)nt2_res->ccdf;
  71             accdf->fid = cpu_to_be32(sei_cont->fid);
  72             accdf->fh = cpu_to_be32(sei_cont->fh);
  73             accdf->pec = cpu_to_be16(sei_cont->pec);
  74             break;
  75         default:
  76             abort();
  77         }
  78         g_free(sei_cont);
  79         rc = 0;
  80     }
  81
  82     return rc;
  83 }
  84
  85 int pci_chsc_sei_nt2_have_event(void)
  86 {
  87     S390pciState *s = s390_get_phb();
  88
  89     return !QTAILQ_EMPTY(&s->pending_sei);
  90 }
  91
  92 S390PCIBusDevice *s390_pci_find_next_avail_dev(S390pciState *s,
  93                                                S390PCIBusDevice *pbdev)
  94 {
  95     S390PCIBusDevice *ret = pbdev ? QTAILQ_NEXT(pbdev, link) :
  96         QTAILQ_FIRST(&s->zpci_devs);
  97
  98     while (ret && ret->state == ZPCI_FS_RESERVED) {
  99         ret = QTAILQ_NEXT(ret, link);
 100     }
 101
 102     return ret;
 103 }
 104
 105 S390PCIBusDevice *s390_pci_find_dev_by_fid(S390pciState *s, uint32_t fid)
 106 {
 107     S390PCIBusDevice *pbdev;
 108
 109     QTAILQ_FOREACH(pbdev, &s->zpci_devs, link) {
 110         if (pbdev->fid == fid) {
 111             return pbdev;
 112         }
 113     }
 114
 115     return NULL;
 116 }
 117
 118 void s390_pci_sclp_configure(SCCB *sccb)
 119 {
 120     IoaCfgSccb *psccb = (IoaCfgSccb *)sccb;
 121     S390PCIBusDevice *pbdev = s390_pci_find_dev_by_fid(s390_get_phb(),
 122                                                        be32_to_cpu(psccb->aid));
 123     uint16_t rc;
 124
 125     if (!pbdev) {
 126         trace_s390_pci_sclp_nodev("configure", be32_to_cpu(psccb->aid));
 127         rc = SCLP_RC_ADAPTER_ID_NOT_RECOGNIZED;
 128         goto out;
 129     }
 130
 131     switch (pbdev->state) {
 132     case ZPCI_FS_RESERVED:
 133         rc = SCLP_RC_ADAPTER_IN_RESERVED_STATE;
 134         break;
 135     case ZPCI_FS_STANDBY:
 136         pbdev->state = ZPCI_FS_DISABLED;
 137         rc = SCLP_RC_NORMAL_COMPLETION;
 138         break;
 139     default:
 140         rc = SCLP_RC_NO_ACTION_REQUIRED;
 141     }
 142 out:
 143     psccb->header.response_code = cpu_to_be16(rc);
 144 }
 145
 146 static void s390_pci_shutdown_notifier(Notifier *n, void *opaque)
 147 {
 148     S390PCIBusDevice *pbdev = container_of(n, S390PCIBusDevice,
 149                                            shutdown_notifier);
 150
 151     pci_device_reset(pbdev->pdev);
 152 }
 153
 154 static void s390_pci_reset_cb(void *opaque)
 155 {
 156     S390PCIBusDevice *pbdev = opaque;
 157
 158     pci_device_reset(pbdev->pdev);
 159 }
 160
 161 static void s390_pci_perform_unplug(S390PCIBusDevice *pbdev)
 162 {
 163     HotplugHandler *hotplug_ctrl;
 164
 165     if (pbdev->pft == ZPCI_PFT_ISM) {
 166         notifier_remove(&pbdev->shutdown_notifier);
 167         qemu_unregister_reset(s390_pci_reset_cb, pbdev);
 168     }
 169
 170     /* Unplug the PCI device */
 171     if (pbdev->pdev) {
 172         DeviceState *pdev = DEVICE(pbdev->pdev);
 173
 174         hotplug_ctrl = qdev_get_hotplug_handler(pdev);
 175         hotplug_handler_unplug(hotplug_ctrl, pdev, &error_abort);
 176         object_unparent(OBJECT(pdev));
 177     }
 178
 179     /* Unplug the zPCI device */
 180     hotplug_ctrl = qdev_get_hotplug_handler(DEVICE(pbdev));
 181     hotplug_handler_unplug(hotplug_ctrl, DEVICE(pbdev), &error_abort);
 182     object_unparent(OBJECT(pbdev));
 183 }
 184
 185 void s390_pci_sclp_deconfigure(SCCB *sccb)
 186 {
 187     IoaCfgSccb *psccb = (IoaCfgSccb *)sccb;
 188     S390PCIBusDevice *pbdev = s390_pci_find_dev_by_fid(s390_get_phb(),
 189                                                        be32_to_cpu(psccb->aid));
 190     uint16_t rc;
 191
 192     if (!pbdev) {
 193         trace_s390_pci_sclp_nodev("deconfigure", be32_to_cpu(psccb->aid));
 194         rc = SCLP_RC_ADAPTER_ID_NOT_RECOGNIZED;
 195         goto out;
 196     }
 197
 198     switch (pbdev->state) {
 199     case ZPCI_FS_RESERVED:
 200         rc = SCLP_RC_ADAPTER_IN_RESERVED_STATE;
 201         break;
 202     case ZPCI_FS_STANDBY:
 203         rc = SCLP_RC_NO_ACTION_REQUIRED;
 204         break;
 205     default:
 206         if (pbdev->interp && (pbdev->fh & FH_MASK_ENABLE)) {
 207             /* Interpreted devices were using interrupt forwarding */
 208             s390_pci_kvm_aif_disable(pbdev);
 209         } else if (pbdev->summary_ind) {
 210             pci_dereg_irqs(pbdev);
 211         }
 212         if (pbdev->iommu->enabled) {
 213             pci_dereg_ioat(pbdev->iommu);
 214         }
 215         pbdev->state = ZPCI_FS_STANDBY;
 216         rc = SCLP_RC_NORMAL_COMPLETION;
 217
 218         if (pbdev->unplug_requested) {
 219             s390_pci_perform_unplug(pbdev);
 220         }
 221     }
 222 out:
 223     psccb->header.response_code = cpu_to_be16(rc);
 224 }
 225
 226 static S390PCIBusDevice *s390_pci_find_dev_by_uid(S390pciState *s, uint16_t uid)
 227 {
 228     S390PCIBusDevice *pbdev;
 229
 230     QTAILQ_FOREACH(pbdev, &s->zpci_devs, link) {
 231         if (pbdev->uid == uid) {
 232             return pbdev;
 233         }
 234     }
 235
 236     return NULL;
 237 }
 238
 239 S390PCIBusDevice *s390_pci_find_dev_by_target(S390pciState *s,
 240                                               const char *target)
 241 {
 242     S390PCIBusDevice *pbdev;
 243
 244     if (!target) {
 245         return NULL;
 246     }
 247
 248     QTAILQ_FOREACH(pbdev, &s->zpci_devs, link) {
 249         if (!strcmp(pbdev->target, target)) {
 250             return pbdev;
 251         }
 252     }
 253
 254     return NULL;
 255 }
 256
 257 static S390PCIBusDevice *s390_pci_find_dev_by_pci(S390pciState *s,
 258                                                   PCIDevice *pci_dev)
 259 {
 260     S390PCIBusDevice *pbdev;
 261
 262     if (!pci_dev) {
 263         return NULL;
 264     }
 265
 266     QTAILQ_FOREACH(pbdev, &s->zpci_devs, link) {
 267         if (pbdev->pdev == pci_dev) {
 268             return pbdev;
 269         }
 270     }
 271
 272     return NULL;
 273 }
 274
 275 S390PCIBusDevice *s390_pci_find_dev_by_idx(S390pciState *s, uint32_t idx)
 276 {
 277     return g_hash_table_lookup(s->zpci_table, &idx);
 278 }
 279
 280 S390PCIBusDevice *s390_pci_find_dev_by_fh(S390pciState *s, uint32_t fh)
 281 {
 282     uint32_t idx = FH_MASK_INDEX & fh;
 283     S390PCIBusDevice *pbdev = s390_pci_find_dev_by_idx(s, idx);
 284
 285     if (pbdev && pbdev->fh == fh) {
 286         return pbdev;
 287     }
 288
 289     return NULL;
 290 }
 291
 292 static void s390_pci_generate_event(uint8_t cc, uint16_t pec, uint32_t fh,
 293                                     uint32_t fid, uint64_t faddr, uint32_t e)
 294 {
 295     SeiContainer *sei_cont;
 296     S390pciState *s = s390_get_phb();
 297
 298     sei_cont = g_new0(SeiContainer, 1);
 299     sei_cont->fh = fh;
 300     sei_cont->fid = fid;
 301     sei_cont->cc = cc;
 302     sei_cont->pec = pec;
 303     sei_cont->faddr = faddr;
 304     sei_cont->e = e;
 305
 306     QTAILQ_INSERT_TAIL(&s->pending_sei, sei_cont, link);
 307     css_generate_css_crws(0);
 308 }
 309
 310 static void s390_pci_generate_plug_event(uint16_t pec, uint32_t fh,
 311                                          uint32_t fid)
 312 {
 313     s390_pci_generate_event(2, pec, fh, fid, 0, 0);
 314 }
 315
 316 void s390_pci_generate_error_event(uint16_t pec, uint32_t fh, uint32_t fid,
 317                                    uint64_t faddr, uint32_t e)
 318 {
 319     s390_pci_generate_event(1, pec, fh, fid, faddr, e);
 320 }
 321
 322 static void s390_pci_set_irq(void *opaque, int irq, int level)
 323 {
 324     /* nothing to do */
 325 }
 326
 327 static int s390_pci_map_irq(PCIDevice *pci_dev, int irq_num)
 328 {
 329     /* nothing to do */
 330     return 0;
 331 }
 332
 333 static uint64_t s390_pci_get_table_origin(uint64_t iota)
 334 {
 335     return iota & ~ZPCI_IOTA_RTTO_FLAG;
 336 }
 337
 338 static unsigned int calc_rtx(dma_addr_t ptr)
 339 {
 340     return ((unsigned long) ptr >> ZPCI_RT_SHIFT) & ZPCI_INDEX_MASK;
 341 }
 342
 343 static unsigned int calc_sx(dma_addr_t ptr)
 344 {
 345     return ((unsigned long) ptr >> ZPCI_ST_SHIFT) & ZPCI_INDEX_MASK;
 346 }
 347
 348 static unsigned int calc_px(dma_addr_t ptr)
 349 {
 350     return ((unsigned long) ptr >> TARGET_PAGE_BITS) & ZPCI_PT_MASK;
 351 }
 352
 353 static uint64_t get_rt_sto(uint64_t entry)
 354 {
 355     return ((entry & ZPCI_TABLE_TYPE_MASK) == ZPCI_TABLE_TYPE_RTX)
 356                 ? (entry & ZPCI_RTE_ADDR_MASK)
 357                 : 0;
 358 }
 359
 360 static uint64_t get_st_pto(uint64_t entry)
 361 {
 362     return ((entry & ZPCI_TABLE_TYPE_MASK) == ZPCI_TABLE_TYPE_SX)
 363             ? (entry & ZPCI_STE_ADDR_MASK)
 364             : 0;
 365 }
 366
 367 static bool rt_entry_isvalid(uint64_t entry)
 368 {
 369     return (entry & ZPCI_TABLE_VALID_MASK) == ZPCI_TABLE_VALID;
 370 }
 371
 372 static bool pt_entry_isvalid(uint64_t entry)
 373 {
 374     return (entry & ZPCI_PTE_VALID_MASK) == ZPCI_PTE_VALID;
 375 }
 376
 377 static bool entry_isprotected(uint64_t entry)
 378 {
 379     return (entry & ZPCI_TABLE_PROT_MASK) == ZPCI_TABLE_PROTECTED;
 380 }
 381
 382 /* ett is expected table type, -1 page table, 0 segment table, 1 region table */
 383 static uint64_t get_table_index(uint64_t iova, int8_t ett)
 384 {
 385     switch (ett) {
 386     case ZPCI_ETT_PT:
 387         return calc_px(iova);
 388     case ZPCI_ETT_ST:
 389         return calc_sx(iova);
 390     case ZPCI_ETT_RT:
 391         return calc_rtx(iova);
 392     }
 393
 394     return -1;
 395 }
 396
 397 static bool entry_isvalid(uint64_t entry, int8_t ett)
 398 {
 399     switch (ett) {
 400     case ZPCI_ETT_PT:
 401         return pt_entry_isvalid(entry);
 402     case ZPCI_ETT_ST:
 403     case ZPCI_ETT_RT:
 404         return rt_entry_isvalid(entry);
 405     }
 406
 407     return false;
 408 }
 409
 410 /* Return true if address translation is done */
 411 static bool translate_iscomplete(uint64_t entry, int8_t ett)
 412 {
 413     switch (ett) {
 414     case 0:
 415         return (entry & ZPCI_TABLE_FC) ? true : false;
 416     case 1:
 417         return false;
 418     }
 419
 420     return true;
 421 }
 422
 423 static uint64_t get_frame_size(int8_t ett)
 424 {
 425     switch (ett) {
 426     case ZPCI_ETT_PT:
 427         return 1ULL << 12;
 428     case ZPCI_ETT_ST:
 429         return 1ULL << 20;
 430     case ZPCI_ETT_RT:
 431         return 1ULL << 31;
 432     }
 433
 434     return 0;
 435 }
 436
 437 static uint64_t get_next_table_origin(uint64_t entry, int8_t ett)
 438 {
 439     switch (ett) {
 440     case ZPCI_ETT_PT:
 441         return entry & ZPCI_PTE_ADDR_MASK;
 442     case ZPCI_ETT_ST:
 443         return get_st_pto(entry);
 444     case ZPCI_ETT_RT:
 445         return get_rt_sto(entry);
 446     }
 447
 448     return 0;
 449 }
 450
 451 /**
 452  * table_translate: do translation within one table and return the following
 453  *                  table origin
 454  *
 455  * @entry: the entry being translated, the result is stored in this.
 456  * @to: the address of table origin.
 457  * @ett: expected table type, 1 region table, 0 segment table and -1 page table.
 458  * @error: error code
 459  */
 460 static uint64_t table_translate(S390IOTLBEntry *entry, uint64_t to, int8_t ett,
 461                                 uint16_t *error)
 462 {
 463     uint64_t tx, te, nto = 0;
 464     uint16_t err = 0;
 465
 466     tx = get_table_index(entry->iova, ett);
 467     te = address_space_ldq(&address_space_memory, to + tx * sizeof(uint64_t),
 468                            MEMTXATTRS_UNSPECIFIED, NULL);
 469
 470     if (!te) {
 471         err = ERR_EVENT_INVALTE;
 472         goto out;
 473     }
 474
 475     if (!entry_isvalid(te, ett)) {
 476         entry->perm &= IOMMU_NONE;
 477         goto out;
 478     }
 479
 480     if (ett == ZPCI_ETT_RT && ((te & ZPCI_TABLE_LEN_RTX) != ZPCI_TABLE_LEN_RTX
 481                                || te & ZPCI_TABLE_OFFSET_MASK)) {
 482         err = ERR_EVENT_INVALTL;
 483         goto out;
 484     }
 485
 486     nto = get_next_table_origin(te, ett);
 487     if (!nto) {
 488         err = ERR_EVENT_TT;
 489         goto out;
 490     }
 491
 492     if (entry_isprotected(te)) {
 493         entry->perm &= IOMMU_RO;
 494     } else {
 495         entry->perm &= IOMMU_RW;
 496     }
 497
 498     if (translate_iscomplete(te, ett)) {
 499         switch (ett) {
 500         case ZPCI_ETT_PT:
 501             entry->translated_addr = te & ZPCI_PTE_ADDR_MASK;
 502             break;
 503         case ZPCI_ETT_ST:
 504             entry->translated_addr = (te & ZPCI_SFAA_MASK) |
 505                 (entry->iova & ~ZPCI_SFAA_MASK);
 506             break;
 507         }
 508         nto = 0;
 509     }
 510 out:
 511     if (err) {
 512         entry->perm = IOMMU_NONE;
 513         *error = err;
 514     }
 515     entry->len = get_frame_size(ett);
 516     return nto;
 517 }
 518
 519 uint16_t s390_guest_io_table_walk(uint64_t g_iota, hwaddr addr,
 520                                   S390IOTLBEntry *entry)
 521 {
 522     uint64_t to = s390_pci_get_table_origin(g_iota);
 523     int8_t ett = 1;
 524     uint16_t error = 0;
 525
 526     entry->iova = addr & TARGET_PAGE_MASK;
 527     entry->translated_addr = 0;
 528     entry->perm = IOMMU_RW;
 529
 530     if (entry_isprotected(g_iota)) {
 531         entry->perm &= IOMMU_RO;
 532     }
 533
 534     while (to) {
 535         to = table_translate(entry, to, ett--, &error);
 536     }
 537
 538     return error;
 539 }
 540
 541 static IOMMUTLBEntry s390_translate_iommu(IOMMUMemoryRegion *mr, hwaddr addr,
 542                                           IOMMUAccessFlags flag, int iommu_idx)
 543 {
 544     S390PCIIOMMU *iommu = container_of(mr, S390PCIIOMMU, iommu_mr);
 545     S390IOTLBEntry *entry;
 546     uint64_t iova = addr & TARGET_PAGE_MASK;
 547     uint16_t error = 0;
 548     IOMMUTLBEntry ret = {
 549         .target_as = &address_space_memory,
 550         .iova = 0,
 551         .translated_addr = 0,
 552         .addr_mask = ~(hwaddr)0,
 553         .perm = IOMMU_NONE,
 554     };
 555
 556     switch (iommu->pbdev->state) {
 557     case ZPCI_FS_ENABLED:
 558     case ZPCI_FS_BLOCKED:
 559         if (!iommu->enabled) {
 560             return ret;
 561         }
 562         break;
 563     default:
 564         return ret;
 565     }
 566
 567     trace_s390_pci_iommu_xlate(addr);
 568
 569     if (addr < iommu->pba || addr > iommu->pal) {
 570         error = ERR_EVENT_OORANGE;
 571         goto err;
 572     }
 573
 574     entry = g_hash_table_lookup(iommu->iotlb, &iova);
 575     if (entry) {
 576         ret.iova = entry->iova;
 577         ret.translated_addr = entry->translated_addr;
 578         ret.addr_mask = entry->len - 1;
 579         ret.perm = entry->perm;
 580     } else {
 581         ret.iova = iova;
 582         ret.addr_mask = ~TARGET_PAGE_MASK;
 583         ret.perm = IOMMU_NONE;
 584     }
 585
 586     if (flag != IOMMU_NONE && !(flag & ret.perm)) {
 587         error = ERR_EVENT_TPROTE;
 588     }
 589 err:
 590     if (error) {
 591         iommu->pbdev->state = ZPCI_FS_ERROR;
 592         s390_pci_generate_error_event(error, iommu->pbdev->fh,
 593                                       iommu->pbdev->fid, addr, 0);
 594     }
 595     return ret;
 596 }
 597
 598 static void s390_pci_iommu_replay(IOMMUMemoryRegion *iommu,
 599                                   IOMMUNotifier *notifier)
 600 {
 601     /* It's impossible to plug a pci device on s390x that already has iommu
 602      * mappings which need to be replayed, that is due to the "one iommu per
 603      * zpci device" construct. But when we support migration of vfio-pci
 604      * devices in future, we need to revisit this.
 605      */
 606     return;
 607 }
 608
 609 static S390PCIIOMMU *s390_pci_get_iommu(S390pciState *s, PCIBus *bus,
 610                                         int devfn)
 611 {
 612     uint64_t key = (uintptr_t)bus;
 613     S390PCIIOMMUTable *table = g_hash_table_lookup(s->iommu_table, &key);
 614     S390PCIIOMMU *iommu;
 615
 616     if (!table) {
 617         table = g_new0(S390PCIIOMMUTable, 1);
 618         table->key = key;
 619         g_hash_table_insert(s->iommu_table, &table->key, table);
 620     }
 621
 622     iommu = table->iommu[PCI_SLOT(devfn)];
 623     if (!iommu) {
 624         iommu = S390_PCI_IOMMU(object_new(TYPE_S390_PCI_IOMMU));
 625
 626         char *mr_name = g_strdup_printf("iommu-root-%02x:%02x.%01x",
 627                                         pci_bus_num(bus),
 628                                         PCI_SLOT(devfn),
 629                                         PCI_FUNC(devfn));
 630         char *as_name = g_strdup_printf("iommu-pci-%02x:%02x.%01x",
 631                                         pci_bus_num(bus),
 632                                         PCI_SLOT(devfn),
 633                                         PCI_FUNC(devfn));
 634         memory_region_init(&iommu->mr, OBJECT(iommu), mr_name, UINT64_MAX);
 635         address_space_init(&iommu->as, &iommu->mr, as_name);
 636         iommu->iotlb = g_hash_table_new_full(g_int64_hash, g_int64_equal,
 637                                              NULL, g_free);
 638         table->iommu[PCI_SLOT(devfn)] = iommu;
 639
 640         g_free(mr_name);
 641         g_free(as_name);
 642     }
 643
 644     return iommu;
 645 }
 646
 647 static AddressSpace *s390_pci_dma_iommu(PCIBus *bus, void *opaque, int devfn)
 648 {
 649     S390pciState *s = opaque;
 650     S390PCIIOMMU *iommu = s390_pci_get_iommu(s, bus, devfn);
 651
 652     return &iommu->as;
 653 }
 654
 655 static const PCIIOMMUOps s390_iommu_ops = {
 656     .get_address_space = s390_pci_dma_iommu,
 657 };
 658
 659 static uint8_t set_ind_atomic(uint64_t ind_loc, uint8_t to_be_set)
 660 {
 661     uint8_t expected, actual;
 662     hwaddr len = 1;
 663     /* avoid  multiple fetches */
 664     uint8_t volatile *ind_addr;
 665
 666     ind_addr = cpu_physical_memory_map(ind_loc, &len, true);
 667     if (!ind_addr) {
 668         s390_pci_generate_error_event(ERR_EVENT_AIRERR, 0, 0, 0, 0);
 669         return -1;
 670     }
 671     actual = *ind_addr;
 672     do {
 673         expected = actual;
 674         actual = qatomic_cmpxchg(ind_addr, expected, expected | to_be_set);
 675     } while (actual != expected);
 676     cpu_physical_memory_unmap((void *)ind_addr, len, 1, len);
 677
 678     return actual;
 679 }
 680
 681 static void s390_msi_ctrl_write(void *opaque, hwaddr addr, uint64_t data,
 682                                 unsigned int size)
 683 {
 684     S390PCIBusDevice *pbdev = opaque;
 685     uint32_t vec = data & ZPCI_MSI_VEC_MASK;
 686     uint64_t ind_bit;
 687     uint32_t sum_bit;
 688
 689     assert(pbdev);
 690
 691     trace_s390_pci_msi_ctrl_write(data, pbdev->idx, vec);
 692
 693     if (pbdev->state != ZPCI_FS_ENABLED) {
 694         return;
 695     }
 696
 697     ind_bit = pbdev->routes.adapter.ind_offset;
 698     sum_bit = pbdev->routes.adapter.summary_offset;
 699
 700     set_ind_atomic(pbdev->routes.adapter.ind_addr + (ind_bit + vec) / 8,
 701                    0x80 >> ((ind_bit + vec) % 8));
 702     if (!set_ind_atomic(pbdev->routes.adapter.summary_addr + sum_bit / 8,
 703                                        0x80 >> (sum_bit % 8))) {
 704         css_adapter_interrupt(CSS_IO_ADAPTER_PCI, pbdev->isc);
 705     }
 706 }
 707
 708 static uint64_t s390_msi_ctrl_read(void *opaque, hwaddr addr, unsigned size)
 709 {
 710     return 0xffffffff;
 711 }
 712
 713 static const MemoryRegionOps s390_msi_ctrl_ops = {
 714     .write = s390_msi_ctrl_write,
 715     .read = s390_msi_ctrl_read,
 716     .endianness = DEVICE_LITTLE_ENDIAN,
 717 };
 718
 719 void s390_pci_iommu_enable(S390PCIIOMMU *iommu)
 720 {
 721     /*
 722      * The iommu region is initialized against a 0-mapped address space,
 723      * so the smallest IOMMU region we can define runs from 0 to the end
 724      * of the PCI address space.
 725      */
 726     char *name = g_strdup_printf("iommu-s390-%04x", iommu->pbdev->uid);
 727     memory_region_init_iommu(&iommu->iommu_mr, sizeof(iommu->iommu_mr),
 728                              TYPE_S390_IOMMU_MEMORY_REGION, OBJECT(&iommu->mr),
 729                              name, iommu->pal + 1);
 730     iommu->enabled = true;
 731     memory_region_add_subregion(&iommu->mr, 0, MEMORY_REGION(&iommu->iommu_mr));
 732     g_free(name);
 733 }
 734
 735 void s390_pci_iommu_disable(S390PCIIOMMU *iommu)
 736 {
 737     iommu->enabled = false;
 738     g_hash_table_remove_all(iommu->iotlb);
 739     memory_region_del_subregion(&iommu->mr, MEMORY_REGION(&iommu->iommu_mr));
 740     object_unparent(OBJECT(&iommu->iommu_mr));
 741 }
 742
 743 static void s390_pci_iommu_free(S390pciState *s, PCIBus *bus, int32_t devfn)
 744 {
 745     uint64_t key = (uintptr_t)bus;
 746     S390PCIIOMMUTable *table = g_hash_table_lookup(s->iommu_table, &key);
 747     S390PCIIOMMU *iommu = table ? table->iommu[PCI_SLOT(devfn)] : NULL;
 748
 749     if (!table || !iommu) {
 750         return;
 751     }
 752
 753     table->iommu[PCI_SLOT(devfn)] = NULL;
 754     g_hash_table_destroy(iommu->iotlb);
 755     /*
 756      * An attached PCI device may have memory listeners, eg. VFIO PCI.
 757      * The associated subregion will already have been unmapped in
 758      * s390_pci_iommu_disable in response to the guest deconfigure request.
 759      * Remove the listeners now before destroying the address space.
 760      */
 761     address_space_remove_listeners(&iommu->as);
 762     address_space_destroy(&iommu->as);
 763     object_unparent(OBJECT(&iommu->mr));
 764     object_unparent(OBJECT(iommu));
 765     object_unref(OBJECT(iommu));
 766 }
 767
 768 S390PCIGroup *s390_group_create(int id, int host_id)
 769 {
 770     S390PCIGroup *group;
 771     S390pciState *s = s390_get_phb();
 772
 773     group = g_new0(S390PCIGroup, 1);
 774     group->id = id;
 775     group->host_id = host_id;
 776     QTAILQ_INSERT_TAIL(&s->zpci_groups, group, link);
 777     return group;
 778 }
 779
 780 S390PCIGroup *s390_group_find(int id)
 781 {
 782     S390PCIGroup *group;
 783     S390pciState *s = s390_get_phb();
 784
 785     QTAILQ_FOREACH(group, &s->zpci_groups, link) {
 786         if (group->id == id) {
 787             return group;
 788         }
 789     }
 790     return NULL;
 791 }
 792
 793 S390PCIGroup *s390_group_find_host_sim(int host_id)
 794 {
 795     S390PCIGroup *group;
 796     S390pciState *s = s390_get_phb();
 797
 798     QTAILQ_FOREACH(group, &s->zpci_groups, link) {
 799         if (group->id >= ZPCI_SIM_GRP_START && group->host_id == host_id) {
 800             return group;
 801         }
 802     }
 803     return NULL;
 804 }
 805
 806 static void s390_pci_init_default_group(void)
 807 {
 808     S390PCIGroup *group;
 809     ClpRspQueryPciGrp *resgrp;
 810
 811     group = s390_group_create(ZPCI_DEFAULT_FN_GRP, ZPCI_DEFAULT_FN_GRP);
 812     resgrp = &group->zpci_group;
 813     resgrp->fr = 1;
 814     resgrp->dasm = 0;
 815     resgrp->msia = ZPCI_MSI_ADDR;
 816     resgrp->mui = DEFAULT_MUI;
 817     resgrp->i = 128;
 818     resgrp->maxstbl = 128;
 819     resgrp->version = 0;
 820     resgrp->dtsm = ZPCI_DTSM;
 821 }
 822
 823 static void set_pbdev_info(S390PCIBusDevice *pbdev)
 824 {
 825     pbdev->zpci_fn.sdma = ZPCI_SDMA_ADDR;
 826     pbdev->zpci_fn.edma = ZPCI_EDMA_ADDR;
 827     pbdev->zpci_fn.pchid = 0;
 828     pbdev->zpci_fn.pfgid = ZPCI_DEFAULT_FN_GRP;
 829     pbdev->zpci_fn.fid = pbdev->fid;
 830     pbdev->zpci_fn.uid = pbdev->uid;
 831     pbdev->pci_group = s390_group_find(ZPCI_DEFAULT_FN_GRP);
 832 }
 833
 834 static void s390_pcihost_realize(DeviceState *dev, Error **errp)
 835 {
 836     PCIBus *b;
 837     BusState *bus;
 838     PCIHostState *phb = PCI_HOST_BRIDGE(dev);
 839     S390pciState *s = S390_PCI_HOST_BRIDGE(dev);
 840
 841     trace_s390_pcihost("realize");
 842
 843     b = pci_register_root_bus(dev, NULL, s390_pci_set_irq, s390_pci_map_irq,
 844                               NULL, get_system_memory(), get_system_io(), 0,
 845                               64, TYPE_PCI_BUS);
 846     pci_setup_iommu(b, &s390_iommu_ops, s);
 847
 848     bus = BUS(b);
 849     qbus_set_hotplug_handler(bus, OBJECT(dev));
 850     phb->bus = b;
 851
 852     s->bus = S390_PCI_BUS(qbus_new(TYPE_S390_PCI_BUS, dev, NULL));
 853     qbus_set_hotplug_handler(BUS(s->bus), OBJECT(dev));
 854
 855     s->iommu_table = g_hash_table_new_full(g_int64_hash, g_int64_equal,
 856                                            NULL, g_free);
 857     s->zpci_table = g_hash_table_new_full(g_int_hash, g_int_equal, NULL, NULL);
 858     s->bus_no = 0;
 859     s->next_sim_grp = ZPCI_SIM_GRP_START;
 860     QTAILQ_INIT(&s->pending_sei);
 861     QTAILQ_INIT(&s->zpci_devs);
 862     QTAILQ_INIT(&s->zpci_dma_limit);
 863     QTAILQ_INIT(&s->zpci_groups);
 864
 865     s390_pci_init_default_group();
 866     css_register_io_adapters(CSS_IO_ADAPTER_PCI, true, false,
 867                              S390_ADAPTER_SUPPRESSIBLE, errp);
 868 }
 869
 870 static void s390_pcihost_unrealize(DeviceState *dev)
 871 {
 872     S390PCIGroup *group;
 873     S390pciState *s = S390_PCI_HOST_BRIDGE(dev);
 874
 875     while (!QTAILQ_EMPTY(&s->zpci_groups)) {
 876         group = QTAILQ_FIRST(&s->zpci_groups);
 877         QTAILQ_REMOVE(&s->zpci_groups, group, link);
 878     }
 879 }
 880
 881 static int s390_pci_msix_init(S390PCIBusDevice *pbdev)
 882 {
 883     char *name;
 884     uint8_t pos;
 885     uint16_t ctrl;
 886     uint32_t table, pba;
 887
 888     pos = pci_find_capability(pbdev->pdev, PCI_CAP_ID_MSIX);
 889     if (!pos) {
 890         return -1;
 891     }
 892
 893     ctrl = pci_host_config_read_common(pbdev->pdev, pos + PCI_MSIX_FLAGS,
 894              pci_config_size(pbdev->pdev), sizeof(ctrl));
 895     table = pci_host_config_read_common(pbdev->pdev, pos + PCI_MSIX_TABLE,
 896              pci_config_size(pbdev->pdev), sizeof(table));
 897     pba = pci_host_config_read_common(pbdev->pdev, pos + PCI_MSIX_PBA,
 898              pci_config_size(pbdev->pdev), sizeof(pba));
 899
 900     pbdev->msix.table_bar = table & PCI_MSIX_FLAGS_BIRMASK;
 901     pbdev->msix.table_offset = table & ~PCI_MSIX_FLAGS_BIRMASK;
 902     pbdev->msix.pba_bar = pba & PCI_MSIX_FLAGS_BIRMASK;
 903     pbdev->msix.pba_offset = pba & ~PCI_MSIX_FLAGS_BIRMASK;
 904     pbdev->msix.entries = (ctrl & PCI_MSIX_FLAGS_QSIZE) + 1;
 905
 906     name = g_strdup_printf("msix-s390-%04x", pbdev->uid);
 907     memory_region_init_io(&pbdev->msix_notify_mr, OBJECT(pbdev),
 908                           &s390_msi_ctrl_ops, pbdev, name, TARGET_PAGE_SIZE);
 909     memory_region_add_subregion(&pbdev->iommu->mr,
 910                                 pbdev->pci_group->zpci_group.msia,
 911                                 &pbdev->msix_notify_mr);
 912     g_free(name);
 913
 914     return 0;
 915 }
 916
 917 static void s390_pci_msix_free(S390PCIBusDevice *pbdev)
 918 {
 919     if (pbdev->msix.entries == 0) {
 920         return;
 921     }
 922
 923     memory_region_del_subregion(&pbdev->iommu->mr, &pbdev->msix_notify_mr);
 924     object_unparent(OBJECT(&pbdev->msix_notify_mr));
 925 }
 926
 927 static S390PCIBusDevice *s390_pci_device_new(S390pciState *s,
 928                                              const char *target, Error **errp)
 929 {
 930     Error *local_err = NULL;
 931     DeviceState *dev;
 932
 933     dev = qdev_try_new(TYPE_S390_PCI_DEVICE);
 934     if (!dev) {
 935         error_setg(errp, "zPCI device could not be created");
 936         return NULL;
 937     }
 938
 939     if (!object_property_set_str(OBJECT(dev), "target", target, &local_err)) {
 940         object_unparent(OBJECT(dev));
 941         error_propagate_prepend(errp, local_err,
 942                                 "zPCI device could not be created: ");
 943         return NULL;
 944     }
 945     if (!qdev_realize_and_unref(dev, BUS(s->bus), &local_err)) {
 946         object_unparent(OBJECT(dev));
 947         error_propagate_prepend(errp, local_err,
 948                                 "zPCI device could not be created: ");
 949         return NULL;
 950     }
 951
 952     return S390_PCI_DEVICE(dev);
 953 }
 954
 955 static bool s390_pci_alloc_idx(S390pciState *s, S390PCIBusDevice *pbdev)
 956 {
 957     uint32_t idx;
 958
 959     idx = s->next_idx;
 960     while (s390_pci_find_dev_by_idx(s, idx)) {
 961         idx = (idx + 1) & FH_MASK_INDEX;
 962         if (idx == s->next_idx) {
 963             return false;
 964         }
 965     }
 966
 967     pbdev->idx = idx;
 968     return true;
 969 }
 970
 971 static void s390_pcihost_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
 972                                    Error **errp)
 973 {
 974     S390pciState *s = S390_PCI_HOST_BRIDGE(hotplug_dev);
 975
 976     if (!s390_has_feat(S390_FEAT_ZPCI)) {
 977         warn_report("Plugging a PCI/zPCI device without the 'zpci' CPU "
 978                     "feature enabled; the guest will not be able to see/use "
 979                     "this device");
 980     }
 981
 982     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
 983         PCIDevice *pdev = PCI_DEVICE(dev);
 984
 985         if (pdev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
 986             error_setg(errp, "multifunction not supported in s390");
 987             return;
 988         }
 989     } else if (object_dynamic_cast(OBJECT(dev), TYPE_S390_PCI_DEVICE)) {
 990         S390PCIBusDevice *pbdev = S390_PCI_DEVICE(dev);
 991
 992         if (!s390_pci_alloc_idx(s, pbdev)) {
 993             error_setg(errp, "no slot for plugging zpci device");
 994             return;
 995         }
 996     }
 997 }
 998
 999 static void s390_pci_update_subordinate(PCIDevice *dev, uint32_t nr)
1000 {
1001     uint32_t old_nr;
1002
1003     pci_default_write_config(dev, PCI_SUBORDINATE_BUS, nr, 1);
1004     while (!pci_bus_is_root(pci_get_bus(dev))) {
1005         dev = pci_get_bus(dev)->parent_dev;
1006
1007         old_nr = pci_default_read_config(dev, PCI_SUBORDINATE_BUS, 1);
1008         if (old_nr < nr) {
1009             pci_default_write_config(dev, PCI_SUBORDINATE_BUS, nr, 1);
1010         }
1011     }
1012 }
1013
1014 static int s390_pci_interp_plug(S390pciState *s, S390PCIBusDevice *pbdev)
1015 {
1016     uint32_t idx, fh;
1017
1018     if (!s390_pci_get_host_fh(pbdev, &fh)) {
1019         return -EPERM;
1020     }
1021
1022     /*
1023      * The host device is already in an enabled state, but we always present
1024      * the initial device state to the guest as disabled (ZPCI_FS_DISABLED).
1025      * Therefore, mask off the enable bit from the passthrough handle until
1026      * the guest issues a CLP SET PCI FN later to enable the device.
1027      */
1028     pbdev->fh = fh & ~FH_MASK_ENABLE;
1029
1030     /* Next, see if the idx is already in-use */
1031     idx = pbdev->fh & FH_MASK_INDEX;
1032     if (pbdev->idx != idx) {
1033         if (s390_pci_find_dev_by_idx(s, idx)) {
1034             return -EINVAL;
1035         }
1036         /*
1037          * Update the idx entry with the passed through idx
1038          * If the relinquished idx is lower than next_idx, use it
1039          * to replace next_idx
1040          */
1041         g_hash_table_remove(s->zpci_table, &pbdev->idx);
1042         if (idx < s->next_idx) {
1043             s->next_idx = idx;
1044         }
1045         pbdev->idx = idx;
1046         g_hash_table_insert(s->zpci_table, &pbdev->idx, pbdev);
1047     }
1048
1049     return 0;
1050 }
1051
1052 static void s390_pcihost_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
1053                               Error **errp)
1054 {
1055     S390pciState *s = S390_PCI_HOST_BRIDGE(hotplug_dev);
1056     PCIDevice *pdev = NULL;
1057     S390PCIBusDevice *pbdev = NULL;
1058     int rc;
1059
1060     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_BRIDGE)) {
1061         PCIBridge *pb = PCI_BRIDGE(dev);
1062
1063         pdev = PCI_DEVICE(dev);
1064         pci_bridge_map_irq(pb, dev->id, s390_pci_map_irq);
1065         pci_setup_iommu(&pb->sec_bus, &s390_iommu_ops, s);
1066
1067         qbus_set_hotplug_handler(BUS(&pb->sec_bus), OBJECT(s));
1068
1069         if (dev->hotplugged) {
1070             pci_default_write_config(pdev, PCI_PRIMARY_BUS,
1071                                      pci_dev_bus_num(pdev), 1);
1072             s->bus_no += 1;
1073             pci_default_write_config(pdev, PCI_SECONDARY_BUS, s->bus_no, 1);
1074
1075             s390_pci_update_subordinate(pdev, s->bus_no);
1076         }
1077     } else if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
1078         pdev = PCI_DEVICE(dev);
1079
1080         if (!dev->id) {
1081             /* In the case the PCI device does not define an id */
1082             /* we generate one based on the PCI address         */
1083             dev->id = g_strdup_printf("auto_%02x:%02x.%01x",
1084                                       pci_dev_bus_num(pdev),
1085                                       PCI_SLOT(pdev->devfn),
1086                                       PCI_FUNC(pdev->devfn));
1087         }
1088
1089         pbdev = s390_pci_find_dev_by_target(s, dev->id);
1090         if (!pbdev) {
1091             pbdev = s390_pci_device_new(s, dev->id, errp);
1092             if (!pbdev) {
1093                 return;
1094             }
1095         }
1096
1097         pbdev->pdev = pdev;
1098         pbdev->iommu = s390_pci_get_iommu(s, pci_get_bus(pdev), pdev->devfn);
1099         pbdev->iommu->pbdev = pbdev;
1100         pbdev->state = ZPCI_FS_DISABLED;
1101         set_pbdev_info(pbdev);
1102
1103         if (object_dynamic_cast(OBJECT(dev), "vfio-pci")) {
1104             /*
1105              * By default, interpretation is always requested; if the available
1106              * facilities indicate it is not available, fallback to the
1107              * interception model.
1108              */
1109             if (pbdev->interp) {
1110                 if (s390_pci_kvm_interp_allowed()) {
1111                     rc = s390_pci_interp_plug(s, pbdev);
1112                     if (rc) {
1113                         error_setg(errp, "Plug failed for zPCI device in "
1114                                    "interpretation mode: %d", rc);
1115                         return;
1116                     }
1117                 } else {
1118                     trace_s390_pcihost("zPCI interpretation missing");
1119                     pbdev->interp = false;
1120                     pbdev->forwarding_assist = false;
1121                 }
1122             }
1123             pbdev->iommu->dma_limit = s390_pci_start_dma_count(s, pbdev);
1124             /* Fill in CLP information passed via the vfio region */
1125             s390_pci_get_clp_info(pbdev);
1126             if (!pbdev->interp) {
1127                 /* Do vfio passthrough but intercept for I/O */
1128                 pbdev->fh |= FH_SHM_VFIO;
1129                 pbdev->forwarding_assist = false;
1130             }
1131             /* Register shutdown notifier and reset callback for ISM devices */
1132             if (pbdev->pft == ZPCI_PFT_ISM) {
1133                 pbdev->shutdown_notifier.notify = s390_pci_shutdown_notifier;
1134                 qemu_register_shutdown_notifier(&pbdev->shutdown_notifier);
1135                 qemu_register_reset(s390_pci_reset_cb, pbdev);
1136             }
1137         } else {
1138             pbdev->fh |= FH_SHM_EMUL;
1139             /* Always intercept emulated devices */
1140             pbdev->interp = false;
1141             pbdev->forwarding_assist = false;
1142         }
1143
1144         if (s390_pci_msix_init(pbdev) && !pbdev->interp) {
1145             error_setg(errp, "MSI-X support is mandatory "
1146                        "in the S390 architecture");
1147             return;
1148         }
1149
1150         if (dev->hotplugged) {
1151             s390_pci_generate_plug_event(HP_EVENT_TO_CONFIGURED ,
1152                                          pbdev->fh, pbdev->fid);
1153         }
1154     } else if (object_dynamic_cast(OBJECT(dev), TYPE_S390_PCI_DEVICE)) {
1155         pbdev = S390_PCI_DEVICE(dev);
1156
1157         /* the allocated idx is actually getting used */
1158         s->next_idx = (pbdev->idx + 1) & FH_MASK_INDEX;
1159         pbdev->fh = pbdev->idx;
1160         QTAILQ_INSERT_TAIL(&s->zpci_devs, pbdev, link);
1161         g_hash_table_insert(s->zpci_table, &pbdev->idx, pbdev);
1162     } else {
1163         g_assert_not_reached();
1164     }
1165 }
1166
1167 static void s390_pcihost_unplug(HotplugHandler *hotplug_dev, DeviceState *dev,
1168                                 Error **errp)
1169 {
1170     S390pciState *s = S390_PCI_HOST_BRIDGE(hotplug_dev);
1171     S390PCIBusDevice *pbdev = NULL;
1172
1173     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
1174         PCIDevice *pci_dev = PCI_DEVICE(dev);
1175         PCIBus *bus;
1176         int32_t devfn;
1177
1178         pbdev = s390_pci_find_dev_by_pci(s, PCI_DEVICE(dev));
1179         g_assert(pbdev);
1180
1181         s390_pci_generate_plug_event(HP_EVENT_STANDBY_TO_RESERVED,
1182                                      pbdev->fh, pbdev->fid);
1183         bus = pci_get_bus(pci_dev);
1184         devfn = pci_dev->devfn;
1185         qdev_unrealize(dev);
1186
1187         s390_pci_msix_free(pbdev);
1188         s390_pci_iommu_free(s, bus, devfn);
1189         pbdev->pdev = NULL;
1190         pbdev->state = ZPCI_FS_RESERVED;
1191     } else if (object_dynamic_cast(OBJECT(dev), TYPE_S390_PCI_DEVICE)) {
1192         pbdev = S390_PCI_DEVICE(dev);
1193         pbdev->fid = 0;
1194         QTAILQ_REMOVE(&s->zpci_devs, pbdev, link);
1195         g_hash_table_remove(s->zpci_table, &pbdev->idx);
1196         if (pbdev->iommu->dma_limit) {
1197             s390_pci_end_dma_count(s, pbdev->iommu->dma_limit);
1198         }
1199         qdev_unrealize(dev);
1200     }
1201 }
1202
1203 static void s390_pcihost_unplug_request(HotplugHandler *hotplug_dev,
1204                                         DeviceState *dev,
1205                                         Error **errp)
1206 {
1207     S390pciState *s = S390_PCI_HOST_BRIDGE(hotplug_dev);
1208     S390PCIBusDevice *pbdev;
1209
1210     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_BRIDGE)) {
1211         error_setg(errp, "PCI bridge hot unplug currently not supported");
1212     } else if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
1213         /*
1214          * Redirect the unplug request to the zPCI device and remember that
1215          * we've checked the PCI device already (to prevent endless recursion).
1216          */
1217         pbdev = s390_pci_find_dev_by_pci(s, PCI_DEVICE(dev));
1218         g_assert(pbdev);
1219         pbdev->pci_unplug_request_processed = true;
1220         qdev_unplug(DEVICE(pbdev), errp);
1221     } else if (object_dynamic_cast(OBJECT(dev), TYPE_S390_PCI_DEVICE)) {
1222         pbdev = S390_PCI_DEVICE(dev);
1223
1224         /*
1225          * If unplug was initially requested for the zPCI device, we
1226          * first have to redirect to the PCI device, which will in return
1227          * redirect back to us after performing its checks (if the request
1228          * is not blocked, e.g. because it's a PCI bridge).
1229          */
1230         if (pbdev->pdev && !pbdev->pci_unplug_request_processed) {
1231             qdev_unplug(DEVICE(pbdev->pdev), errp);
1232             return;
1233         }
1234         pbdev->pci_unplug_request_processed = false;
1235
1236         switch (pbdev->state) {
1237         case ZPCI_FS_STANDBY:
1238         case ZPCI_FS_RESERVED:
1239             s390_pci_perform_unplug(pbdev);
1240             break;
1241         default:
1242             /*
1243              * Allow to send multiple requests, e.g. if the guest crashed
1244              * before releasing the device, we would not be able to send
1245              * another request to the same VM (e.g. fresh OS).
1246              */
1247             pbdev->unplug_requested = true;
1248             s390_pci_generate_plug_event(HP_EVENT_DECONFIGURE_REQUEST,
1249                                          pbdev->fh, pbdev->fid);
1250         }
1251     } else {
1252         g_assert_not_reached();
1253     }
1254 }
1255
1256 static void s390_pci_enumerate_bridge(PCIBus *bus, PCIDevice *pdev,
1257                                       void *opaque)
1258 {
1259     S390pciState *s = opaque;
1260     PCIBus *sec_bus = NULL;
1261
1262     if ((pci_default_read_config(pdev, PCI_HEADER_TYPE, 1) !=
1263          PCI_HEADER_TYPE_BRIDGE)) {
1264         return;
1265     }
1266
1267     (s->bus_no)++;
1268     pci_default_write_config(pdev, PCI_PRIMARY_BUS, pci_dev_bus_num(pdev), 1);
1269     pci_default_write_config(pdev, PCI_SECONDARY_BUS, s->bus_no, 1);
1270     pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, s->bus_no, 1);
1271
1272     sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(pdev));
1273     if (!sec_bus) {
1274         return;
1275     }
1276
1277     /* Assign numbers to all child bridges. The last is the highest number. */
1278     pci_for_each_device_under_bus(sec_bus, s390_pci_enumerate_bridge, s);
1279     pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, s->bus_no, 1);
1280 }
1281
1282 static void s390_pcihost_reset(DeviceState *dev)
1283 {
1284     S390pciState *s = S390_PCI_HOST_BRIDGE(dev);
1285     PCIBus *bus = s->parent_obj.bus;
1286     S390PCIBusDevice *pbdev, *next;
1287
1288     /* Process all pending unplug requests */
1289     QTAILQ_FOREACH_SAFE(pbdev, &s->zpci_devs, link, next) {
1290         if (pbdev->unplug_requested) {
1291             if (pbdev->interp && (pbdev->fh & FH_MASK_ENABLE)) {
1292                 /* Interpreted devices were using interrupt forwarding */
1293                 s390_pci_kvm_aif_disable(pbdev);
1294             } else if (pbdev->summary_ind) {
1295                 pci_dereg_irqs(pbdev);
1296             }
1297             if (pbdev->iommu->enabled) {
1298                 pci_dereg_ioat(pbdev->iommu);
1299             }
1300             pbdev->state = ZPCI_FS_STANDBY;
1301             s390_pci_perform_unplug(pbdev);
1302         }
1303     }
1304
1305     /*
1306      * When resetting a PCI bridge, the assigned numbers are set to 0. So
1307      * on every system reset, we also have to reassign numbers.
1308      */
1309     s->bus_no = 0;
1310     pci_for_each_device_under_bus(bus, s390_pci_enumerate_bridge, s);
1311 }
1312
1313 static void s390_pcihost_class_init(ObjectClass *klass, void *data)
1314 {
1315     DeviceClass *dc = DEVICE_CLASS(klass);
1316     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(klass);
1317
1318     dc->reset = s390_pcihost_reset;
1319     dc->realize = s390_pcihost_realize;
1320     dc->unrealize = s390_pcihost_unrealize;
1321     hc->pre_plug = s390_pcihost_pre_plug;
1322     hc->plug = s390_pcihost_plug;
1323     hc->unplug_request = s390_pcihost_unplug_request;
1324     hc->unplug = s390_pcihost_unplug;
1325     msi_nonbroken = true;
1326 }
1327
1328 static const TypeInfo s390_pcihost_info = {
1329     .name          = TYPE_S390_PCI_HOST_BRIDGE,
1330     .parent        = TYPE_PCI_HOST_BRIDGE,
1331     .instance_size = sizeof(S390pciState),
1332     .class_init    = s390_pcihost_class_init,
1333     .interfaces = (InterfaceInfo[]) {
1334         { TYPE_HOTPLUG_HANDLER },
1335         { }
1336     }
1337 };
1338
1339 static const TypeInfo s390_pcibus_info = {
1340     .name = TYPE_S390_PCI_BUS,
1341     .parent = TYPE_BUS,
1342     .instance_size = sizeof(S390PCIBus),
1343 };
1344
1345 static uint16_t s390_pci_generate_uid(S390pciState *s)
1346 {
1347     uint16_t uid = 0;
1348
1349     do {
1350         uid++;
1351         if (!s390_pci_find_dev_by_uid(s, uid)) {
1352             return uid;
1353         }
1354     } while (uid < ZPCI_MAX_UID);
1355
1356     return UID_UNDEFINED;
1357 }
1358
1359 static uint32_t s390_pci_generate_fid(S390pciState *s, Error **errp)
1360 {
1361     uint32_t fid = 0;
1362
1363     do {
1364         if (!s390_pci_find_dev_by_fid(s, fid)) {
1365             return fid;
1366         }
1367     } while (fid++ != ZPCI_MAX_FID);
1368
1369     error_setg(errp, "no free fid could be found");
1370     return 0;
1371 }
1372
1373 static void s390_pci_device_realize(DeviceState *dev, Error **errp)
1374 {
1375     S390PCIBusDevice *zpci = S390_PCI_DEVICE(dev);
1376     S390pciState *s = s390_get_phb();
1377
1378     if (!zpci->target) {
1379         error_setg(errp, "target must be defined");
1380         return;
1381     }
1382
1383     if (s390_pci_find_dev_by_target(s, zpci->target)) {
1384         error_setg(errp, "target %s already has an associated zpci device",
1385                    zpci->target);
1386         return;
1387     }
1388
1389     if (zpci->uid == UID_UNDEFINED) {
1390         zpci->uid = s390_pci_generate_uid(s);
1391         if (!zpci->uid) {
1392             error_setg(errp, "no free uid could be found");
1393             return;
1394         }
1395     } else if (s390_pci_find_dev_by_uid(s, zpci->uid)) {
1396         error_setg(errp, "uid %u already in use", zpci->uid);
1397         return;
1398     }
1399
1400     if (!zpci->fid_defined) {
1401         Error *local_error = NULL;
1402
1403         zpci->fid = s390_pci_generate_fid(s, &local_error);
1404         if (local_error) {
1405             error_propagate(errp, local_error);
1406             return;
1407         }
1408     } else if (s390_pci_find_dev_by_fid(s, zpci->fid)) {
1409         error_setg(errp, "fid %u already in use", zpci->fid);
1410         return;
1411     }
1412
1413     zpci->state = ZPCI_FS_RESERVED;
1414     zpci->fmb.format = ZPCI_FMB_FORMAT;
1415 }
1416
1417 static void s390_pci_device_reset(DeviceState *dev)
1418 {
1419     S390PCIBusDevice *pbdev = S390_PCI_DEVICE(dev);
1420
1421     switch (pbdev->state) {
1422     case ZPCI_FS_RESERVED:
1423         return;
1424     case ZPCI_FS_STANDBY:
1425         break;
1426     default:
1427         pbdev->fh &= ~FH_MASK_ENABLE;
1428         pbdev->state = ZPCI_FS_DISABLED;
1429         break;
1430     }
1431
1432     if (pbdev->interp && (pbdev->fh & FH_MASK_ENABLE)) {
1433         /* Interpreted devices were using interrupt forwarding */
1434         s390_pci_kvm_aif_disable(pbdev);
1435     } else if (pbdev->summary_ind) {
1436         pci_dereg_irqs(pbdev);
1437     }
1438     if (pbdev->iommu->enabled) {
1439         pci_dereg_ioat(pbdev->iommu);
1440     }
1441
1442     fmb_timer_free(pbdev);
1443 }
1444
1445 static void s390_pci_get_fid(Object *obj, Visitor *v, const char *name,
1446                          void *opaque, Error **errp)
1447 {
1448     Property *prop = opaque;
1449     uint32_t *ptr = object_field_prop_ptr(obj, prop);
1450
1451     visit_type_uint32(v, name, ptr, errp);
1452 }
1453
1454 static void s390_pci_set_fid(Object *obj, Visitor *v, const char *name,
1455                          void *opaque, Error **errp)
1456 {
1457     S390PCIBusDevice *zpci = S390_PCI_DEVICE(obj);
1458     Property *prop = opaque;
1459     uint32_t *ptr = object_field_prop_ptr(obj, prop);
1460
1461     if (!visit_type_uint32(v, name, ptr, errp)) {
1462         return;
1463     }
1464     zpci->fid_defined = true;
1465 }
1466
1467 static const PropertyInfo s390_pci_fid_propinfo = {
1468     .name = "zpci_fid",
1469     .get = s390_pci_get_fid,
1470     .set = s390_pci_set_fid,
1471 };
1472
1473 #define DEFINE_PROP_S390_PCI_FID(_n, _s, _f) \
1474     DEFINE_PROP(_n, _s, _f, s390_pci_fid_propinfo, uint32_t)
1475
1476 static Property s390_pci_device_properties[] = {
1477     DEFINE_PROP_UINT16("uid", S390PCIBusDevice, uid, UID_UNDEFINED),
1478     DEFINE_PROP_S390_PCI_FID("fid", S390PCIBusDevice, fid),
1479     DEFINE_PROP_STRING("target", S390PCIBusDevice, target),
1480     DEFINE_PROP_BOOL("interpret", S390PCIBusDevice, interp, true),
1481     DEFINE_PROP_BOOL("forwarding-assist", S390PCIBusDevice, forwarding_assist,
1482                      true),
1483     DEFINE_PROP_END_OF_LIST(),
1484 };
1485
1486 static const VMStateDescription s390_pci_device_vmstate = {
1487     .name = TYPE_S390_PCI_DEVICE,
1488     /*
1489      * TODO: add state handling here, so migration works at least with
1490      * emulated pci devices on s390x
1491      */
1492     .unmigratable = 1,
1493 };
1494
1495 static void s390_pci_device_class_init(ObjectClass *klass, void *data)
1496 {
1497     DeviceClass *dc = DEVICE_CLASS(klass);
1498
1499     dc->desc = "zpci device";
1500     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
1501     dc->reset = s390_pci_device_reset;
1502     dc->bus_type = TYPE_S390_PCI_BUS;
1503     dc->realize = s390_pci_device_realize;
1504     device_class_set_props(dc, s390_pci_device_properties);
1505     dc->vmsd = &s390_pci_device_vmstate;
1506 }
1507
1508 static const TypeInfo s390_pci_device_info = {
1509     .name = TYPE_S390_PCI_DEVICE,
1510     .parent = TYPE_DEVICE,
1511     .instance_size = sizeof(S390PCIBusDevice),
1512     .class_init = s390_pci_device_class_init,
1513 };
1514
1515 static const TypeInfo s390_pci_iommu_info = {
1516     .name = TYPE_S390_PCI_IOMMU,
1517     .parent = TYPE_OBJECT,
1518     .instance_size = sizeof(S390PCIIOMMU),
1519 };
1520
1521 static void s390_iommu_memory_region_class_init(ObjectClass *klass, void *data)
1522 {
1523     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_CLASS(klass);
1524
1525     imrc->translate = s390_translate_iommu;
1526     imrc->replay = s390_pci_iommu_replay;
1527 }
1528
1529 static const TypeInfo s390_iommu_memory_region_info = {
1530     .parent = TYPE_IOMMU_MEMORY_REGION,
1531     .name = TYPE_S390_IOMMU_MEMORY_REGION,
1532     .class_init = s390_iommu_memory_region_class_init,
1533 };
1534
1535 static void s390_pci_register_types(void)
1536 {
1537     type_register_static(&s390_pcihost_info);
1538     type_register_static(&s390_pcibus_info);
1539     type_register_static(&s390_pci_device_info);
1540     type_register_static(&s390_pci_iommu_info);
1541     type_register_static(&s390_iommu_memory_region_info);
1542 }
1543
1544 type_init(s390_pci_register_types)