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_perform_unplug(S390PCIBusDevice *pbdev)
 155 {
 156     HotplugHandler *hotplug_ctrl;
 157
 158     if (pbdev->pft == ZPCI_PFT_ISM) {
 159         notifier_remove(&pbdev->shutdown_notifier);
 160     }
 161
 162     /* Unplug the PCI device */
 163     if (pbdev->pdev) {
 164         DeviceState *pdev = DEVICE(pbdev->pdev);
 165
 166         hotplug_ctrl = qdev_get_hotplug_handler(pdev);
 167         hotplug_handler_unplug(hotplug_ctrl, pdev, &error_abort);
 168         object_unparent(OBJECT(pdev));
 169     }
 170
 171     /* Unplug the zPCI device */
 172     hotplug_ctrl = qdev_get_hotplug_handler(DEVICE(pbdev));
 173     hotplug_handler_unplug(hotplug_ctrl, DEVICE(pbdev), &error_abort);
 174     object_unparent(OBJECT(pbdev));
 175 }
 176
 177 void s390_pci_sclp_deconfigure(SCCB *sccb)
 178 {
 179     IoaCfgSccb *psccb = (IoaCfgSccb *)sccb;
 180     S390PCIBusDevice *pbdev = s390_pci_find_dev_by_fid(s390_get_phb(),
 181                                                        be32_to_cpu(psccb->aid));
 182     uint16_t rc;
 183
 184     if (!pbdev) {
 185         trace_s390_pci_sclp_nodev("deconfigure", be32_to_cpu(psccb->aid));
 186         rc = SCLP_RC_ADAPTER_ID_NOT_RECOGNIZED;
 187         goto out;
 188     }
 189
 190     switch (pbdev->state) {
 191     case ZPCI_FS_RESERVED:
 192         rc = SCLP_RC_ADAPTER_IN_RESERVED_STATE;
 193         break;
 194     case ZPCI_FS_STANDBY:
 195         rc = SCLP_RC_NO_ACTION_REQUIRED;
 196         break;
 197     default:
 198         if (pbdev->interp && (pbdev->fh & FH_MASK_ENABLE)) {
 199             /* Interpreted devices were using interrupt forwarding */
 200             s390_pci_kvm_aif_disable(pbdev);
 201         } else if (pbdev->summary_ind) {
 202             pci_dereg_irqs(pbdev);
 203         }
 204         if (pbdev->iommu->enabled) {
 205             pci_dereg_ioat(pbdev->iommu);
 206         }
 207         pbdev->state = ZPCI_FS_STANDBY;
 208         rc = SCLP_RC_NORMAL_COMPLETION;
 209
 210         if (pbdev->unplug_requested) {
 211             s390_pci_perform_unplug(pbdev);
 212         }
 213     }
 214 out:
 215     psccb->header.response_code = cpu_to_be16(rc);
 216 }
 217
 218 static S390PCIBusDevice *s390_pci_find_dev_by_uid(S390pciState *s, uint16_t uid)
 219 {
 220     S390PCIBusDevice *pbdev;
 221
 222     QTAILQ_FOREACH(pbdev, &s->zpci_devs, link) {
 223         if (pbdev->uid == uid) {
 224             return pbdev;
 225         }
 226     }
 227
 228     return NULL;
 229 }
 230
 231 S390PCIBusDevice *s390_pci_find_dev_by_target(S390pciState *s,
 232                                               const char *target)
 233 {
 234     S390PCIBusDevice *pbdev;
 235
 236     if (!target) {
 237         return NULL;
 238     }
 239
 240     QTAILQ_FOREACH(pbdev, &s->zpci_devs, link) {
 241         if (!strcmp(pbdev->target, target)) {
 242             return pbdev;
 243         }
 244     }
 245
 246     return NULL;
 247 }
 248
 249 static S390PCIBusDevice *s390_pci_find_dev_by_pci(S390pciState *s,
 250                                                   PCIDevice *pci_dev)
 251 {
 252     S390PCIBusDevice *pbdev;
 253
 254     if (!pci_dev) {
 255         return NULL;
 256     }
 257
 258     QTAILQ_FOREACH(pbdev, &s->zpci_devs, link) {
 259         if (pbdev->pdev == pci_dev) {
 260             return pbdev;
 261         }
 262     }
 263
 264     return NULL;
 265 }
 266
 267 S390PCIBusDevice *s390_pci_find_dev_by_idx(S390pciState *s, uint32_t idx)
 268 {
 269     return g_hash_table_lookup(s->zpci_table, &idx);
 270 }
 271
 272 S390PCIBusDevice *s390_pci_find_dev_by_fh(S390pciState *s, uint32_t fh)
 273 {
 274     uint32_t idx = FH_MASK_INDEX & fh;
 275     S390PCIBusDevice *pbdev = s390_pci_find_dev_by_idx(s, idx);
 276
 277     if (pbdev && pbdev->fh == fh) {
 278         return pbdev;
 279     }
 280
 281     return NULL;
 282 }
 283
 284 static void s390_pci_generate_event(uint8_t cc, uint16_t pec, uint32_t fh,
 285                                     uint32_t fid, uint64_t faddr, uint32_t e)
 286 {
 287     SeiContainer *sei_cont;
 288     S390pciState *s = s390_get_phb();
 289
 290     sei_cont = g_new0(SeiContainer, 1);
 291     sei_cont->fh = fh;
 292     sei_cont->fid = fid;
 293     sei_cont->cc = cc;
 294     sei_cont->pec = pec;
 295     sei_cont->faddr = faddr;
 296     sei_cont->e = e;
 297
 298     QTAILQ_INSERT_TAIL(&s->pending_sei, sei_cont, link);
 299     css_generate_css_crws(0);
 300 }
 301
 302 static void s390_pci_generate_plug_event(uint16_t pec, uint32_t fh,
 303                                          uint32_t fid)
 304 {
 305     s390_pci_generate_event(2, pec, fh, fid, 0, 0);
 306 }
 307
 308 void s390_pci_generate_error_event(uint16_t pec, uint32_t fh, uint32_t fid,
 309                                    uint64_t faddr, uint32_t e)
 310 {
 311     s390_pci_generate_event(1, pec, fh, fid, faddr, e);
 312 }
 313
 314 static void s390_pci_set_irq(void *opaque, int irq, int level)
 315 {
 316     /* nothing to do */
 317 }
 318
 319 static int s390_pci_map_irq(PCIDevice *pci_dev, int irq_num)
 320 {
 321     /* nothing to do */
 322     return 0;
 323 }
 324
 325 static uint64_t s390_pci_get_table_origin(uint64_t iota)
 326 {
 327     return iota & ~ZPCI_IOTA_RTTO_FLAG;
 328 }
 329
 330 static unsigned int calc_rtx(dma_addr_t ptr)
 331 {
 332     return ((unsigned long) ptr >> ZPCI_RT_SHIFT) & ZPCI_INDEX_MASK;
 333 }
 334
 335 static unsigned int calc_sx(dma_addr_t ptr)
 336 {
 337     return ((unsigned long) ptr >> ZPCI_ST_SHIFT) & ZPCI_INDEX_MASK;
 338 }
 339
 340 static unsigned int calc_px(dma_addr_t ptr)
 341 {
 342     return ((unsigned long) ptr >> TARGET_PAGE_BITS) & ZPCI_PT_MASK;
 343 }
 344
 345 static uint64_t get_rt_sto(uint64_t entry)
 346 {
 347     return ((entry & ZPCI_TABLE_TYPE_MASK) == ZPCI_TABLE_TYPE_RTX)
 348                 ? (entry & ZPCI_RTE_ADDR_MASK)
 349                 : 0;
 350 }
 351
 352 static uint64_t get_st_pto(uint64_t entry)
 353 {
 354     return ((entry & ZPCI_TABLE_TYPE_MASK) == ZPCI_TABLE_TYPE_SX)
 355             ? (entry & ZPCI_STE_ADDR_MASK)
 356             : 0;
 357 }
 358
 359 static bool rt_entry_isvalid(uint64_t entry)
 360 {
 361     return (entry & ZPCI_TABLE_VALID_MASK) == ZPCI_TABLE_VALID;
 362 }
 363
 364 static bool pt_entry_isvalid(uint64_t entry)
 365 {
 366     return (entry & ZPCI_PTE_VALID_MASK) == ZPCI_PTE_VALID;
 367 }
 368
 369 static bool entry_isprotected(uint64_t entry)
 370 {
 371     return (entry & ZPCI_TABLE_PROT_MASK) == ZPCI_TABLE_PROTECTED;
 372 }
 373
 374 /* ett is expected table type, -1 page table, 0 segment table, 1 region table */
 375 static uint64_t get_table_index(uint64_t iova, int8_t ett)
 376 {
 377     switch (ett) {
 378     case ZPCI_ETT_PT:
 379         return calc_px(iova);
 380     case ZPCI_ETT_ST:
 381         return calc_sx(iova);
 382     case ZPCI_ETT_RT:
 383         return calc_rtx(iova);
 384     }
 385
 386     return -1;
 387 }
 388
 389 static bool entry_isvalid(uint64_t entry, int8_t ett)
 390 {
 391     switch (ett) {
 392     case ZPCI_ETT_PT:
 393         return pt_entry_isvalid(entry);
 394     case ZPCI_ETT_ST:
 395     case ZPCI_ETT_RT:
 396         return rt_entry_isvalid(entry);
 397     }
 398
 399     return false;
 400 }
 401
 402 /* Return true if address translation is done */
 403 static bool translate_iscomplete(uint64_t entry, int8_t ett)
 404 {
 405     switch (ett) {
 406     case 0:
 407         return (entry & ZPCI_TABLE_FC) ? true : false;
 408     case 1:
 409         return false;
 410     }
 411
 412     return true;
 413 }
 414
 415 static uint64_t get_frame_size(int8_t ett)
 416 {
 417     switch (ett) {
 418     case ZPCI_ETT_PT:
 419         return 1ULL << 12;
 420     case ZPCI_ETT_ST:
 421         return 1ULL << 20;
 422     case ZPCI_ETT_RT:
 423         return 1ULL << 31;
 424     }
 425
 426     return 0;
 427 }
 428
 429 static uint64_t get_next_table_origin(uint64_t entry, int8_t ett)
 430 {
 431     switch (ett) {
 432     case ZPCI_ETT_PT:
 433         return entry & ZPCI_PTE_ADDR_MASK;
 434     case ZPCI_ETT_ST:
 435         return get_st_pto(entry);
 436     case ZPCI_ETT_RT:
 437         return get_rt_sto(entry);
 438     }
 439
 440     return 0;
 441 }
 442
 443 /**
 444  * table_translate: do translation within one table and return the following
 445  *                  table origin
 446  *
 447  * @entry: the entry being translated, the result is stored in this.
 448  * @to: the address of table origin.
 449  * @ett: expected table type, 1 region table, 0 segment table and -1 page table.
 450  * @error: error code
 451  */
 452 static uint64_t table_translate(S390IOTLBEntry *entry, uint64_t to, int8_t ett,
 453                                 uint16_t *error)
 454 {
 455     uint64_t tx, te, nto = 0;
 456     uint16_t err = 0;
 457
 458     tx = get_table_index(entry->iova, ett);
 459     te = address_space_ldq(&address_space_memory, to + tx * sizeof(uint64_t),
 460                            MEMTXATTRS_UNSPECIFIED, NULL);
 461
 462     if (!te) {
 463         err = ERR_EVENT_INVALTE;
 464         goto out;
 465     }
 466
 467     if (!entry_isvalid(te, ett)) {
 468         entry->perm &= IOMMU_NONE;
 469         goto out;
 470     }
 471
 472     if (ett == ZPCI_ETT_RT && ((te & ZPCI_TABLE_LEN_RTX) != ZPCI_TABLE_LEN_RTX
 473                                || te & ZPCI_TABLE_OFFSET_MASK)) {
 474         err = ERR_EVENT_INVALTL;
 475         goto out;
 476     }
 477
 478     nto = get_next_table_origin(te, ett);
 479     if (!nto) {
 480         err = ERR_EVENT_TT;
 481         goto out;
 482     }
 483
 484     if (entry_isprotected(te)) {
 485         entry->perm &= IOMMU_RO;
 486     } else {
 487         entry->perm &= IOMMU_RW;
 488     }
 489
 490     if (translate_iscomplete(te, ett)) {
 491         switch (ett) {
 492         case ZPCI_ETT_PT:
 493             entry->translated_addr = te & ZPCI_PTE_ADDR_MASK;
 494             break;
 495         case ZPCI_ETT_ST:
 496             entry->translated_addr = (te & ZPCI_SFAA_MASK) |
 497                 (entry->iova & ~ZPCI_SFAA_MASK);
 498             break;
 499         }
 500         nto = 0;
 501     }
 502 out:
 503     if (err) {
 504         entry->perm = IOMMU_NONE;
 505         *error = err;
 506     }
 507     entry->len = get_frame_size(ett);
 508     return nto;
 509 }
 510
 511 uint16_t s390_guest_io_table_walk(uint64_t g_iota, hwaddr addr,
 512                                   S390IOTLBEntry *entry)
 513 {
 514     uint64_t to = s390_pci_get_table_origin(g_iota);
 515     int8_t ett = 1;
 516     uint16_t error = 0;
 517
 518     entry->iova = addr & TARGET_PAGE_MASK;
 519     entry->translated_addr = 0;
 520     entry->perm = IOMMU_RW;
 521
 522     if (entry_isprotected(g_iota)) {
 523         entry->perm &= IOMMU_RO;
 524     }
 525
 526     while (to) {
 527         to = table_translate(entry, to, ett--, &error);
 528     }
 529
 530     return error;
 531 }
 532
 533 static IOMMUTLBEntry s390_translate_iommu(IOMMUMemoryRegion *mr, hwaddr addr,
 534                                           IOMMUAccessFlags flag, int iommu_idx)
 535 {
 536     S390PCIIOMMU *iommu = container_of(mr, S390PCIIOMMU, iommu_mr);
 537     S390IOTLBEntry *entry;
 538     uint64_t iova = addr & TARGET_PAGE_MASK;
 539     uint16_t error = 0;
 540     IOMMUTLBEntry ret = {
 541         .target_as = &address_space_memory,
 542         .iova = 0,
 543         .translated_addr = 0,
 544         .addr_mask = ~(hwaddr)0,
 545         .perm = IOMMU_NONE,
 546     };
 547
 548     switch (iommu->pbdev->state) {
 549     case ZPCI_FS_ENABLED:
 550     case ZPCI_FS_BLOCKED:
 551         if (!iommu->enabled) {
 552             return ret;
 553         }
 554         break;
 555     default:
 556         return ret;
 557     }
 558
 559     trace_s390_pci_iommu_xlate(addr);
 560
 561     if (addr < iommu->pba || addr > iommu->pal) {
 562         error = ERR_EVENT_OORANGE;
 563         goto err;
 564     }
 565
 566     entry = g_hash_table_lookup(iommu->iotlb, &iova);
 567     if (entry) {
 568         ret.iova = entry->iova;
 569         ret.translated_addr = entry->translated_addr;
 570         ret.addr_mask = entry->len - 1;
 571         ret.perm = entry->perm;
 572     } else {
 573         ret.iova = iova;
 574         ret.addr_mask = ~TARGET_PAGE_MASK;
 575         ret.perm = IOMMU_NONE;
 576     }
 577
 578     if (flag != IOMMU_NONE && !(flag & ret.perm)) {
 579         error = ERR_EVENT_TPROTE;
 580     }
 581 err:
 582     if (error) {
 583         iommu->pbdev->state = ZPCI_FS_ERROR;
 584         s390_pci_generate_error_event(error, iommu->pbdev->fh,
 585                                       iommu->pbdev->fid, addr, 0);
 586     }
 587     return ret;
 588 }
 589
 590 static void s390_pci_iommu_replay(IOMMUMemoryRegion *iommu,
 591                                   IOMMUNotifier *notifier)
 592 {
 593     /* It's impossible to plug a pci device on s390x that already has iommu
 594      * mappings which need to be replayed, that is due to the "one iommu per
 595      * zpci device" construct. But when we support migration of vfio-pci
 596      * devices in future, we need to revisit this.
 597      */
 598     return;
 599 }
 600
 601 static S390PCIIOMMU *s390_pci_get_iommu(S390pciState *s, PCIBus *bus,
 602                                         int devfn)
 603 {
 604     uint64_t key = (uintptr_t)bus;
 605     S390PCIIOMMUTable *table = g_hash_table_lookup(s->iommu_table, &key);
 606     S390PCIIOMMU *iommu;
 607
 608     if (!table) {
 609         table = g_new0(S390PCIIOMMUTable, 1);
 610         table->key = key;
 611         g_hash_table_insert(s->iommu_table, &table->key, table);
 612     }
 613
 614     iommu = table->iommu[PCI_SLOT(devfn)];
 615     if (!iommu) {
 616         iommu = S390_PCI_IOMMU(object_new(TYPE_S390_PCI_IOMMU));
 617
 618         char *mr_name = g_strdup_printf("iommu-root-%02x:%02x.%01x",
 619                                         pci_bus_num(bus),
 620                                         PCI_SLOT(devfn),
 621                                         PCI_FUNC(devfn));
 622         char *as_name = g_strdup_printf("iommu-pci-%02x:%02x.%01x",
 623                                         pci_bus_num(bus),
 624                                         PCI_SLOT(devfn),
 625                                         PCI_FUNC(devfn));
 626         memory_region_init(&iommu->mr, OBJECT(iommu), mr_name, UINT64_MAX);
 627         address_space_init(&iommu->as, &iommu->mr, as_name);
 628         iommu->iotlb = g_hash_table_new_full(g_int64_hash, g_int64_equal,
 629                                              NULL, g_free);
 630         table->iommu[PCI_SLOT(devfn)] = iommu;
 631
 632         g_free(mr_name);
 633         g_free(as_name);
 634     }
 635
 636     return iommu;
 637 }
 638
 639 static AddressSpace *s390_pci_dma_iommu(PCIBus *bus, void *opaque, int devfn)
 640 {
 641     S390pciState *s = opaque;
 642     S390PCIIOMMU *iommu = s390_pci_get_iommu(s, bus, devfn);
 643
 644     return &iommu->as;
 645 }
 646
 647 static const PCIIOMMUOps s390_iommu_ops = {
 648     .get_address_space = s390_pci_dma_iommu,
 649 };
 650
 651 static uint8_t set_ind_atomic(uint64_t ind_loc, uint8_t to_be_set)
 652 {
 653     uint8_t expected, actual;
 654     hwaddr len = 1;
 655     /* avoid  multiple fetches */
 656     uint8_t volatile *ind_addr;
 657
 658     ind_addr = cpu_physical_memory_map(ind_loc, &len, true);
 659     if (!ind_addr) {
 660         s390_pci_generate_error_event(ERR_EVENT_AIRERR, 0, 0, 0, 0);
 661         return -1;
 662     }
 663     actual = *ind_addr;
 664     do {
 665         expected = actual;
 666         actual = qatomic_cmpxchg(ind_addr, expected, expected | to_be_set);
 667     } while (actual != expected);
 668     cpu_physical_memory_unmap((void *)ind_addr, len, 1, len);
 669
 670     return actual;
 671 }
 672
 673 static void s390_msi_ctrl_write(void *opaque, hwaddr addr, uint64_t data,
 674                                 unsigned int size)
 675 {
 676     S390PCIBusDevice *pbdev = opaque;
 677     uint32_t vec = data & ZPCI_MSI_VEC_MASK;
 678     uint64_t ind_bit;
 679     uint32_t sum_bit;
 680
 681     assert(pbdev);
 682
 683     trace_s390_pci_msi_ctrl_write(data, pbdev->idx, vec);
 684
 685     if (pbdev->state != ZPCI_FS_ENABLED) {
 686         return;
 687     }
 688
 689     ind_bit = pbdev->routes.adapter.ind_offset;
 690     sum_bit = pbdev->routes.adapter.summary_offset;
 691
 692     set_ind_atomic(pbdev->routes.adapter.ind_addr + (ind_bit + vec) / 8,
 693                    0x80 >> ((ind_bit + vec) % 8));
 694     if (!set_ind_atomic(pbdev->routes.adapter.summary_addr + sum_bit / 8,
 695                                        0x80 >> (sum_bit % 8))) {
 696         css_adapter_interrupt(CSS_IO_ADAPTER_PCI, pbdev->isc);
 697     }
 698 }
 699
 700 static uint64_t s390_msi_ctrl_read(void *opaque, hwaddr addr, unsigned size)
 701 {
 702     return 0xffffffff;
 703 }
 704
 705 static const MemoryRegionOps s390_msi_ctrl_ops = {
 706     .write = s390_msi_ctrl_write,
 707     .read = s390_msi_ctrl_read,
 708     .endianness = DEVICE_LITTLE_ENDIAN,
 709 };
 710
 711 void s390_pci_iommu_enable(S390PCIIOMMU *iommu)
 712 {
 713     /*
 714      * The iommu region is initialized against a 0-mapped address space,
 715      * so the smallest IOMMU region we can define runs from 0 to the end
 716      * of the PCI address space.
 717      */
 718     char *name = g_strdup_printf("iommu-s390-%04x", iommu->pbdev->uid);
 719     memory_region_init_iommu(&iommu->iommu_mr, sizeof(iommu->iommu_mr),
 720                              TYPE_S390_IOMMU_MEMORY_REGION, OBJECT(&iommu->mr),
 721                              name, iommu->pal + 1);
 722     iommu->enabled = true;
 723     memory_region_add_subregion(&iommu->mr, 0, MEMORY_REGION(&iommu->iommu_mr));
 724     g_free(name);
 725 }
 726
 727 void s390_pci_iommu_disable(S390PCIIOMMU *iommu)
 728 {
 729     iommu->enabled = false;
 730     g_hash_table_remove_all(iommu->iotlb);
 731     memory_region_del_subregion(&iommu->mr, MEMORY_REGION(&iommu->iommu_mr));
 732     object_unparent(OBJECT(&iommu->iommu_mr));
 733 }
 734
 735 static void s390_pci_iommu_free(S390pciState *s, PCIBus *bus, int32_t devfn)
 736 {
 737     uint64_t key = (uintptr_t)bus;
 738     S390PCIIOMMUTable *table = g_hash_table_lookup(s->iommu_table, &key);
 739     S390PCIIOMMU *iommu = table ? table->iommu[PCI_SLOT(devfn)] : NULL;
 740
 741     if (!table || !iommu) {
 742         return;
 743     }
 744
 745     table->iommu[PCI_SLOT(devfn)] = NULL;
 746     g_hash_table_destroy(iommu->iotlb);
 747     /*
 748      * An attached PCI device may have memory listeners, eg. VFIO PCI.
 749      * The associated subregion will already have been unmapped in
 750      * s390_pci_iommu_disable in response to the guest deconfigure request.
 751      * Remove the listeners now before destroying the address space.
 752      */
 753     address_space_remove_listeners(&iommu->as);
 754     address_space_destroy(&iommu->as);
 755     object_unparent(OBJECT(&iommu->mr));
 756     object_unparent(OBJECT(iommu));
 757     object_unref(OBJECT(iommu));
 758 }
 759
 760 S390PCIGroup *s390_group_create(int id, int host_id)
 761 {
 762     S390PCIGroup *group;
 763     S390pciState *s = s390_get_phb();
 764
 765     group = g_new0(S390PCIGroup, 1);
 766     group->id = id;
 767     group->host_id = host_id;
 768     QTAILQ_INSERT_TAIL(&s->zpci_groups, group, link);
 769     return group;
 770 }
 771
 772 S390PCIGroup *s390_group_find(int id)
 773 {
 774     S390PCIGroup *group;
 775     S390pciState *s = s390_get_phb();
 776
 777     QTAILQ_FOREACH(group, &s->zpci_groups, link) {
 778         if (group->id == id) {
 779             return group;
 780         }
 781     }
 782     return NULL;
 783 }
 784
 785 S390PCIGroup *s390_group_find_host_sim(int host_id)
 786 {
 787     S390PCIGroup *group;
 788     S390pciState *s = s390_get_phb();
 789
 790     QTAILQ_FOREACH(group, &s->zpci_groups, link) {
 791         if (group->id >= ZPCI_SIM_GRP_START && group->host_id == host_id) {
 792             return group;
 793         }
 794     }
 795     return NULL;
 796 }
 797
 798 static void s390_pci_init_default_group(void)
 799 {
 800     S390PCIGroup *group;
 801     ClpRspQueryPciGrp *resgrp;
 802
 803     group = s390_group_create(ZPCI_DEFAULT_FN_GRP, ZPCI_DEFAULT_FN_GRP);
 804     resgrp = &group->zpci_group;
 805     resgrp->fr = 1;
 806     resgrp->dasm = 0;
 807     resgrp->msia = ZPCI_MSI_ADDR;
 808     resgrp->mui = DEFAULT_MUI;
 809     resgrp->i = 128;
 810     resgrp->maxstbl = 128;
 811     resgrp->version = 0;
 812     resgrp->dtsm = ZPCI_DTSM;
 813 }
 814
 815 static void set_pbdev_info(S390PCIBusDevice *pbdev)
 816 {
 817     pbdev->zpci_fn.sdma = ZPCI_SDMA_ADDR;
 818     pbdev->zpci_fn.edma = ZPCI_EDMA_ADDR;
 819     pbdev->zpci_fn.pchid = 0;
 820     pbdev->zpci_fn.pfgid = ZPCI_DEFAULT_FN_GRP;
 821     pbdev->zpci_fn.fid = pbdev->fid;
 822     pbdev->zpci_fn.uid = pbdev->uid;
 823     pbdev->pci_group = s390_group_find(ZPCI_DEFAULT_FN_GRP);
 824 }
 825
 826 static void s390_pcihost_realize(DeviceState *dev, Error **errp)
 827 {
 828     PCIBus *b;
 829     BusState *bus;
 830     PCIHostState *phb = PCI_HOST_BRIDGE(dev);
 831     S390pciState *s = S390_PCI_HOST_BRIDGE(dev);
 832
 833     trace_s390_pcihost("realize");
 834
 835     b = pci_register_root_bus(dev, NULL, s390_pci_set_irq, s390_pci_map_irq,
 836                               NULL, get_system_memory(), get_system_io(), 0,
 837                               64, TYPE_PCI_BUS);
 838     pci_setup_iommu(b, &s390_iommu_ops, s);
 839
 840     bus = BUS(b);
 841     qbus_set_hotplug_handler(bus, OBJECT(dev));
 842     phb->bus = b;
 843
 844     s->bus = S390_PCI_BUS(qbus_new(TYPE_S390_PCI_BUS, dev, NULL));
 845     qbus_set_hotplug_handler(BUS(s->bus), OBJECT(dev));
 846
 847     s->iommu_table = g_hash_table_new_full(g_int64_hash, g_int64_equal,
 848                                            NULL, g_free);
 849     s->zpci_table = g_hash_table_new_full(g_int_hash, g_int_equal, NULL, NULL);
 850     s->bus_no = 0;
 851     s->next_sim_grp = ZPCI_SIM_GRP_START;
 852     QTAILQ_INIT(&s->pending_sei);
 853     QTAILQ_INIT(&s->zpci_devs);
 854     QTAILQ_INIT(&s->zpci_dma_limit);
 855     QTAILQ_INIT(&s->zpci_groups);
 856
 857     s390_pci_init_default_group();
 858     css_register_io_adapters(CSS_IO_ADAPTER_PCI, true, false,
 859                              S390_ADAPTER_SUPPRESSIBLE, errp);
 860 }
 861
 862 static void s390_pcihost_unrealize(DeviceState *dev)
 863 {
 864     S390PCIGroup *group;
 865     S390pciState *s = S390_PCI_HOST_BRIDGE(dev);
 866
 867     while (!QTAILQ_EMPTY(&s->zpci_groups)) {
 868         group = QTAILQ_FIRST(&s->zpci_groups);
 869         QTAILQ_REMOVE(&s->zpci_groups, group, link);
 870     }
 871 }
 872
 873 static int s390_pci_msix_init(S390PCIBusDevice *pbdev)
 874 {
 875     char *name;
 876     uint8_t pos;
 877     uint16_t ctrl;
 878     uint32_t table, pba;
 879
 880     pos = pci_find_capability(pbdev->pdev, PCI_CAP_ID_MSIX);
 881     if (!pos) {
 882         return -1;
 883     }
 884
 885     ctrl = pci_host_config_read_common(pbdev->pdev, pos + PCI_MSIX_FLAGS,
 886              pci_config_size(pbdev->pdev), sizeof(ctrl));
 887     table = pci_host_config_read_common(pbdev->pdev, pos + PCI_MSIX_TABLE,
 888              pci_config_size(pbdev->pdev), sizeof(table));
 889     pba = pci_host_config_read_common(pbdev->pdev, pos + PCI_MSIX_PBA,
 890              pci_config_size(pbdev->pdev), sizeof(pba));
 891
 892     pbdev->msix.table_bar = table & PCI_MSIX_FLAGS_BIRMASK;
 893     pbdev->msix.table_offset = table & ~PCI_MSIX_FLAGS_BIRMASK;
 894     pbdev->msix.pba_bar = pba & PCI_MSIX_FLAGS_BIRMASK;
 895     pbdev->msix.pba_offset = pba & ~PCI_MSIX_FLAGS_BIRMASK;
 896     pbdev->msix.entries = (ctrl & PCI_MSIX_FLAGS_QSIZE) + 1;
 897
 898     name = g_strdup_printf("msix-s390-%04x", pbdev->uid);
 899     memory_region_init_io(&pbdev->msix_notify_mr, OBJECT(pbdev),
 900                           &s390_msi_ctrl_ops, pbdev, name, TARGET_PAGE_SIZE);
 901     memory_region_add_subregion(&pbdev->iommu->mr,
 902                                 pbdev->pci_group->zpci_group.msia,
 903                                 &pbdev->msix_notify_mr);
 904     g_free(name);
 905
 906     return 0;
 907 }
 908
 909 static void s390_pci_msix_free(S390PCIBusDevice *pbdev)
 910 {
 911     if (pbdev->msix.entries == 0) {
 912         return;
 913     }
 914
 915     memory_region_del_subregion(&pbdev->iommu->mr, &pbdev->msix_notify_mr);
 916     object_unparent(OBJECT(&pbdev->msix_notify_mr));
 917 }
 918
 919 static S390PCIBusDevice *s390_pci_device_new(S390pciState *s,
 920                                              const char *target, Error **errp)
 921 {
 922     Error *local_err = NULL;
 923     DeviceState *dev;
 924
 925     dev = qdev_try_new(TYPE_S390_PCI_DEVICE);
 926     if (!dev) {
 927         error_setg(errp, "zPCI device could not be created");
 928         return NULL;
 929     }
 930
 931     if (!object_property_set_str(OBJECT(dev), "target", target, &local_err)) {
 932         object_unparent(OBJECT(dev));
 933         error_propagate_prepend(errp, local_err,
 934                                 "zPCI device could not be created: ");
 935         return NULL;
 936     }
 937     if (!qdev_realize_and_unref(dev, BUS(s->bus), &local_err)) {
 938         object_unparent(OBJECT(dev));
 939         error_propagate_prepend(errp, local_err,
 940                                 "zPCI device could not be created: ");
 941         return NULL;
 942     }
 943
 944     return S390_PCI_DEVICE(dev);
 945 }
 946
 947 static bool s390_pci_alloc_idx(S390pciState *s, S390PCIBusDevice *pbdev)
 948 {
 949     uint32_t idx;
 950
 951     idx = s->next_idx;
 952     while (s390_pci_find_dev_by_idx(s, idx)) {
 953         idx = (idx + 1) & FH_MASK_INDEX;
 954         if (idx == s->next_idx) {
 955             return false;
 956         }
 957     }
 958
 959     pbdev->idx = idx;
 960     return true;
 961 }
 962
 963 static void s390_pcihost_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
 964                                    Error **errp)
 965 {
 966     S390pciState *s = S390_PCI_HOST_BRIDGE(hotplug_dev);
 967
 968     if (!s390_has_feat(S390_FEAT_ZPCI)) {
 969         warn_report("Plugging a PCI/zPCI device without the 'zpci' CPU "
 970                     "feature enabled; the guest will not be able to see/use "
 971                     "this device");
 972     }
 973
 974     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
 975         PCIDevice *pdev = PCI_DEVICE(dev);
 976
 977         if (pdev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
 978             error_setg(errp, "multifunction not supported in s390");
 979             return;
 980         }
 981     } else if (object_dynamic_cast(OBJECT(dev), TYPE_S390_PCI_DEVICE)) {
 982         S390PCIBusDevice *pbdev = S390_PCI_DEVICE(dev);
 983
 984         if (!s390_pci_alloc_idx(s, pbdev)) {
 985             error_setg(errp, "no slot for plugging zpci device");
 986             return;
 987         }
 988     }
 989 }
 990
 991 static void s390_pci_update_subordinate(PCIDevice *dev, uint32_t nr)
 992 {
 993     uint32_t old_nr;
 994
 995     pci_default_write_config(dev, PCI_SUBORDINATE_BUS, nr, 1);
 996     while (!pci_bus_is_root(pci_get_bus(dev))) {
 997         dev = pci_get_bus(dev)->parent_dev;
 998
 999         old_nr = pci_default_read_config(dev, PCI_SUBORDINATE_BUS, 1);
1000         if (old_nr < nr) {
1001             pci_default_write_config(dev, PCI_SUBORDINATE_BUS, nr, 1);
1002         }
1003     }
1004 }
1005
1006 static int s390_pci_interp_plug(S390pciState *s, S390PCIBusDevice *pbdev)
1007 {
1008     uint32_t idx, fh;
1009
1010     if (!s390_pci_get_host_fh(pbdev, &fh)) {
1011         return -EPERM;
1012     }
1013
1014     /*
1015      * The host device is already in an enabled state, but we always present
1016      * the initial device state to the guest as disabled (ZPCI_FS_DISABLED).
1017      * Therefore, mask off the enable bit from the passthrough handle until
1018      * the guest issues a CLP SET PCI FN later to enable the device.
1019      */
1020     pbdev->fh = fh & ~FH_MASK_ENABLE;
1021
1022     /* Next, see if the idx is already in-use */
1023     idx = pbdev->fh & FH_MASK_INDEX;
1024     if (pbdev->idx != idx) {
1025         if (s390_pci_find_dev_by_idx(s, idx)) {
1026             return -EINVAL;
1027         }
1028         /*
1029          * Update the idx entry with the passed through idx
1030          * If the relinquished idx is lower than next_idx, use it
1031          * to replace next_idx
1032          */
1033         g_hash_table_remove(s->zpci_table, &pbdev->idx);
1034         if (idx < s->next_idx) {
1035             s->next_idx = idx;
1036         }
1037         pbdev->idx = idx;
1038         g_hash_table_insert(s->zpci_table, &pbdev->idx, pbdev);
1039     }
1040
1041     return 0;
1042 }
1043
1044 static void s390_pcihost_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
1045                               Error **errp)
1046 {
1047     S390pciState *s = S390_PCI_HOST_BRIDGE(hotplug_dev);
1048     PCIDevice *pdev = NULL;
1049     S390PCIBusDevice *pbdev = NULL;
1050     int rc;
1051
1052     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_BRIDGE)) {
1053         PCIBridge *pb = PCI_BRIDGE(dev);
1054
1055         pdev = PCI_DEVICE(dev);
1056         pci_bridge_map_irq(pb, dev->id, s390_pci_map_irq);
1057         pci_setup_iommu(&pb->sec_bus, &s390_iommu_ops, s);
1058
1059         qbus_set_hotplug_handler(BUS(&pb->sec_bus), OBJECT(s));
1060
1061         if (dev->hotplugged) {
1062             pci_default_write_config(pdev, PCI_PRIMARY_BUS,
1063                                      pci_dev_bus_num(pdev), 1);
1064             s->bus_no += 1;
1065             pci_default_write_config(pdev, PCI_SECONDARY_BUS, s->bus_no, 1);
1066
1067             s390_pci_update_subordinate(pdev, s->bus_no);
1068         }
1069     } else if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
1070         pdev = PCI_DEVICE(dev);
1071
1072         if (!dev->id) {
1073             /* In the case the PCI device does not define an id */
1074             /* we generate one based on the PCI address         */
1075             dev->id = g_strdup_printf("auto_%02x:%02x.%01x",
1076                                       pci_dev_bus_num(pdev),
1077                                       PCI_SLOT(pdev->devfn),
1078                                       PCI_FUNC(pdev->devfn));
1079         }
1080
1081         pbdev = s390_pci_find_dev_by_target(s, dev->id);
1082         if (!pbdev) {
1083             pbdev = s390_pci_device_new(s, dev->id, errp);
1084             if (!pbdev) {
1085                 return;
1086             }
1087         }
1088
1089         pbdev->pdev = pdev;
1090         pbdev->iommu = s390_pci_get_iommu(s, pci_get_bus(pdev), pdev->devfn);
1091         pbdev->iommu->pbdev = pbdev;
1092         pbdev->state = ZPCI_FS_DISABLED;
1093         set_pbdev_info(pbdev);
1094
1095         if (object_dynamic_cast(OBJECT(dev), "vfio-pci")) {
1096             /*
1097              * By default, interpretation is always requested; if the available
1098              * facilities indicate it is not available, fallback to the
1099              * interception model.
1100              */
1101             if (pbdev->interp) {
1102                 if (s390_pci_kvm_interp_allowed()) {
1103                     rc = s390_pci_interp_plug(s, pbdev);
1104                     if (rc) {
1105                         error_setg(errp, "Plug failed for zPCI device in "
1106                                    "interpretation mode: %d", rc);
1107                         return;
1108                     }
1109                 } else {
1110                     trace_s390_pcihost("zPCI interpretation missing");
1111                     pbdev->interp = false;
1112                     pbdev->forwarding_assist = false;
1113                 }
1114             }
1115             pbdev->iommu->dma_limit = s390_pci_start_dma_count(s, pbdev);
1116             /* Fill in CLP information passed via the vfio region */
1117             s390_pci_get_clp_info(pbdev);
1118             if (!pbdev->interp) {
1119                 /* Do vfio passthrough but intercept for I/O */
1120                 pbdev->fh |= FH_SHM_VFIO;
1121                 pbdev->forwarding_assist = false;
1122             }
1123             /* Register shutdown notifier and reset callback for ISM devices */
1124             if (pbdev->pft == ZPCI_PFT_ISM) {
1125                 pbdev->shutdown_notifier.notify = s390_pci_shutdown_notifier;
1126                 qemu_register_shutdown_notifier(&pbdev->shutdown_notifier);
1127             }
1128         } else {
1129             pbdev->fh |= FH_SHM_EMUL;
1130             /* Always intercept emulated devices */
1131             pbdev->interp = false;
1132             pbdev->forwarding_assist = false;
1133         }
1134
1135         if (s390_pci_msix_init(pbdev) && !pbdev->interp) {
1136             error_setg(errp, "MSI-X support is mandatory "
1137                        "in the S390 architecture");
1138             return;
1139         }
1140
1141         if (dev->hotplugged) {
1142             s390_pci_generate_plug_event(HP_EVENT_TO_CONFIGURED ,
1143                                          pbdev->fh, pbdev->fid);
1144         }
1145     } else if (object_dynamic_cast(OBJECT(dev), TYPE_S390_PCI_DEVICE)) {
1146         pbdev = S390_PCI_DEVICE(dev);
1147
1148         /* the allocated idx is actually getting used */
1149         s->next_idx = (pbdev->idx + 1) & FH_MASK_INDEX;
1150         pbdev->fh = pbdev->idx;
1151         QTAILQ_INSERT_TAIL(&s->zpci_devs, pbdev, link);
1152         g_hash_table_insert(s->zpci_table, &pbdev->idx, pbdev);
1153     } else {
1154         g_assert_not_reached();
1155     }
1156 }
1157
1158 static void s390_pcihost_unplug(HotplugHandler *hotplug_dev, DeviceState *dev,
1159                                 Error **errp)
1160 {
1161     S390pciState *s = S390_PCI_HOST_BRIDGE(hotplug_dev);
1162     S390PCIBusDevice *pbdev = NULL;
1163
1164     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
1165         PCIDevice *pci_dev = PCI_DEVICE(dev);
1166         PCIBus *bus;
1167         int32_t devfn;
1168
1169         pbdev = s390_pci_find_dev_by_pci(s, PCI_DEVICE(dev));
1170         g_assert(pbdev);
1171
1172         s390_pci_generate_plug_event(HP_EVENT_STANDBY_TO_RESERVED,
1173                                      pbdev->fh, pbdev->fid);
1174         bus = pci_get_bus(pci_dev);
1175         devfn = pci_dev->devfn;
1176         qdev_unrealize(dev);
1177
1178         s390_pci_msix_free(pbdev);
1179         s390_pci_iommu_free(s, bus, devfn);
1180         pbdev->pdev = NULL;
1181         pbdev->state = ZPCI_FS_RESERVED;
1182     } else if (object_dynamic_cast(OBJECT(dev), TYPE_S390_PCI_DEVICE)) {
1183         pbdev = S390_PCI_DEVICE(dev);
1184         pbdev->fid = 0;
1185         QTAILQ_REMOVE(&s->zpci_devs, pbdev, link);
1186         g_hash_table_remove(s->zpci_table, &pbdev->idx);
1187         if (pbdev->iommu->dma_limit) {
1188             s390_pci_end_dma_count(s, pbdev->iommu->dma_limit);
1189         }
1190         qdev_unrealize(dev);
1191     }
1192 }
1193
1194 static void s390_pcihost_unplug_request(HotplugHandler *hotplug_dev,
1195                                         DeviceState *dev,
1196                                         Error **errp)
1197 {
1198     S390pciState *s = S390_PCI_HOST_BRIDGE(hotplug_dev);
1199     S390PCIBusDevice *pbdev;
1200
1201     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_BRIDGE)) {
1202         error_setg(errp, "PCI bridge hot unplug currently not supported");
1203     } else if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
1204         /*
1205          * Redirect the unplug request to the zPCI device and remember that
1206          * we've checked the PCI device already (to prevent endless recursion).
1207          */
1208         pbdev = s390_pci_find_dev_by_pci(s, PCI_DEVICE(dev));
1209         g_assert(pbdev);
1210         pbdev->pci_unplug_request_processed = true;
1211         qdev_unplug(DEVICE(pbdev), errp);
1212     } else if (object_dynamic_cast(OBJECT(dev), TYPE_S390_PCI_DEVICE)) {
1213         pbdev = S390_PCI_DEVICE(dev);
1214
1215         /*
1216          * If unplug was initially requested for the zPCI device, we
1217          * first have to redirect to the PCI device, which will in return
1218          * redirect back to us after performing its checks (if the request
1219          * is not blocked, e.g. because it's a PCI bridge).
1220          */
1221         if (pbdev->pdev && !pbdev->pci_unplug_request_processed) {
1222             qdev_unplug(DEVICE(pbdev->pdev), errp);
1223             return;
1224         }
1225         pbdev->pci_unplug_request_processed = false;
1226
1227         switch (pbdev->state) {
1228         case ZPCI_FS_STANDBY:
1229         case ZPCI_FS_RESERVED:
1230             s390_pci_perform_unplug(pbdev);
1231             break;
1232         default:
1233             /*
1234              * Allow to send multiple requests, e.g. if the guest crashed
1235              * before releasing the device, we would not be able to send
1236              * another request to the same VM (e.g. fresh OS).
1237              */
1238             pbdev->unplug_requested = true;
1239             s390_pci_generate_plug_event(HP_EVENT_DECONFIGURE_REQUEST,
1240                                          pbdev->fh, pbdev->fid);
1241         }
1242     } else {
1243         g_assert_not_reached();
1244     }
1245 }
1246
1247 static void s390_pci_enumerate_bridge(PCIBus *bus, PCIDevice *pdev,
1248                                       void *opaque)
1249 {
1250     S390pciState *s = opaque;
1251     PCIBus *sec_bus = NULL;
1252
1253     if ((pci_default_read_config(pdev, PCI_HEADER_TYPE, 1) !=
1254          PCI_HEADER_TYPE_BRIDGE)) {
1255         return;
1256     }
1257
1258     (s->bus_no)++;
1259     pci_default_write_config(pdev, PCI_PRIMARY_BUS, pci_dev_bus_num(pdev), 1);
1260     pci_default_write_config(pdev, PCI_SECONDARY_BUS, s->bus_no, 1);
1261     pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, s->bus_no, 1);
1262
1263     sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(pdev));
1264     if (!sec_bus) {
1265         return;
1266     }
1267
1268     /* Assign numbers to all child bridges. The last is the highest number. */
1269     pci_for_each_device_under_bus(sec_bus, s390_pci_enumerate_bridge, s);
1270     pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, s->bus_no, 1);
1271 }
1272
1273 void s390_pci_ism_reset(void)
1274 {
1275     S390pciState *s = s390_get_phb();
1276
1277     S390PCIBusDevice *pbdev, *next;
1278
1279     /* Trigger reset event for each passthrough ISM device currently in-use */
1280     QTAILQ_FOREACH_SAFE(pbdev, &s->zpci_devs, link, next) {
1281         if (pbdev->interp && pbdev->pft == ZPCI_PFT_ISM &&
1282             pbdev->fh & FH_MASK_ENABLE) {
1283             s390_pci_kvm_aif_disable(pbdev);
1284
1285             pci_device_reset(pbdev->pdev);
1286         }
1287     }
1288 }
1289
1290 static void s390_pcihost_reset(DeviceState *dev)
1291 {
1292     S390pciState *s = S390_PCI_HOST_BRIDGE(dev);
1293     PCIBus *bus = s->parent_obj.bus;
1294     S390PCIBusDevice *pbdev, *next;
1295
1296     /* Process all pending unplug requests */
1297     QTAILQ_FOREACH_SAFE(pbdev, &s->zpci_devs, link, next) {
1298         if (pbdev->unplug_requested) {
1299             if (pbdev->interp && (pbdev->fh & FH_MASK_ENABLE)) {
1300                 /* Interpreted devices were using interrupt forwarding */
1301                 s390_pci_kvm_aif_disable(pbdev);
1302             } else if (pbdev->summary_ind) {
1303                 pci_dereg_irqs(pbdev);
1304             }
1305             if (pbdev->iommu->enabled) {
1306                 pci_dereg_ioat(pbdev->iommu);
1307             }
1308             pbdev->state = ZPCI_FS_STANDBY;
1309             s390_pci_perform_unplug(pbdev);
1310         }
1311     }
1312
1313     /*
1314      * When resetting a PCI bridge, the assigned numbers are set to 0. So
1315      * on every system reset, we also have to reassign numbers.
1316      */
1317     s->bus_no = 0;
1318     pci_for_each_device_under_bus(bus, s390_pci_enumerate_bridge, s);
1319 }
1320
1321 static void s390_pcihost_class_init(ObjectClass *klass, void *data)
1322 {
1323     DeviceClass *dc = DEVICE_CLASS(klass);
1324     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(klass);
1325
1326     dc->reset = s390_pcihost_reset;
1327     dc->realize = s390_pcihost_realize;
1328     dc->unrealize = s390_pcihost_unrealize;
1329     hc->pre_plug = s390_pcihost_pre_plug;
1330     hc->plug = s390_pcihost_plug;
1331     hc->unplug_request = s390_pcihost_unplug_request;
1332     hc->unplug = s390_pcihost_unplug;
1333     msi_nonbroken = true;
1334 }
1335
1336 static const TypeInfo s390_pcihost_info = {
1337     .name          = TYPE_S390_PCI_HOST_BRIDGE,
1338     .parent        = TYPE_PCI_HOST_BRIDGE,
1339     .instance_size = sizeof(S390pciState),
1340     .class_init    = s390_pcihost_class_init,
1341     .interfaces = (InterfaceInfo[]) {
1342         { TYPE_HOTPLUG_HANDLER },
1343         { }
1344     }
1345 };
1346
1347 static const TypeInfo s390_pcibus_info = {
1348     .name = TYPE_S390_PCI_BUS,
1349     .parent = TYPE_BUS,
1350     .instance_size = sizeof(S390PCIBus),
1351 };
1352
1353 static uint16_t s390_pci_generate_uid(S390pciState *s)
1354 {
1355     uint16_t uid = 0;
1356
1357     do {
1358         uid++;
1359         if (!s390_pci_find_dev_by_uid(s, uid)) {
1360             return uid;
1361         }
1362     } while (uid < ZPCI_MAX_UID);
1363
1364     return UID_UNDEFINED;
1365 }
1366
1367 static uint32_t s390_pci_generate_fid(S390pciState *s, Error **errp)
1368 {
1369     uint32_t fid = 0;
1370
1371     do {
1372         if (!s390_pci_find_dev_by_fid(s, fid)) {
1373             return fid;
1374         }
1375     } while (fid++ != ZPCI_MAX_FID);
1376
1377     error_setg(errp, "no free fid could be found");
1378     return 0;
1379 }
1380
1381 static void s390_pci_device_realize(DeviceState *dev, Error **errp)
1382 {
1383     S390PCIBusDevice *zpci = S390_PCI_DEVICE(dev);
1384     S390pciState *s = s390_get_phb();
1385
1386     if (!zpci->target) {
1387         error_setg(errp, "target must be defined");
1388         return;
1389     }
1390
1391     if (s390_pci_find_dev_by_target(s, zpci->target)) {
1392         error_setg(errp, "target %s already has an associated zpci device",
1393                    zpci->target);
1394         return;
1395     }
1396
1397     if (zpci->uid == UID_UNDEFINED) {
1398         zpci->uid = s390_pci_generate_uid(s);
1399         if (!zpci->uid) {
1400             error_setg(errp, "no free uid could be found");
1401             return;
1402         }
1403     } else if (s390_pci_find_dev_by_uid(s, zpci->uid)) {
1404         error_setg(errp, "uid %u already in use", zpci->uid);
1405         return;
1406     }
1407
1408     if (!zpci->fid_defined) {
1409         Error *local_error = NULL;
1410
1411         zpci->fid = s390_pci_generate_fid(s, &local_error);
1412         if (local_error) {
1413             error_propagate(errp, local_error);
1414             return;
1415         }
1416     } else if (s390_pci_find_dev_by_fid(s, zpci->fid)) {
1417         error_setg(errp, "fid %u already in use", zpci->fid);
1418         return;
1419     }
1420
1421     zpci->state = ZPCI_FS_RESERVED;
1422     zpci->fmb.format = ZPCI_FMB_FORMAT;
1423 }
1424
1425 static void s390_pci_device_reset(DeviceState *dev)
1426 {
1427     S390PCIBusDevice *pbdev = S390_PCI_DEVICE(dev);
1428
1429     switch (pbdev->state) {
1430     case ZPCI_FS_RESERVED:
1431         return;
1432     case ZPCI_FS_STANDBY:
1433         break;
1434     default:
1435         pbdev->fh &= ~FH_MASK_ENABLE;
1436         pbdev->state = ZPCI_FS_DISABLED;
1437         break;
1438     }
1439
1440     if (pbdev->interp && (pbdev->fh & FH_MASK_ENABLE)) {
1441         /* Interpreted devices were using interrupt forwarding */
1442         s390_pci_kvm_aif_disable(pbdev);
1443     } else if (pbdev->summary_ind) {
1444         pci_dereg_irqs(pbdev);
1445     }
1446     if (pbdev->iommu->enabled) {
1447         pci_dereg_ioat(pbdev->iommu);
1448     }
1449
1450     fmb_timer_free(pbdev);
1451 }
1452
1453 static void s390_pci_get_fid(Object *obj, Visitor *v, const char *name,
1454                          void *opaque, Error **errp)
1455 {
1456     Property *prop = opaque;
1457     uint32_t *ptr = object_field_prop_ptr(obj, prop);
1458
1459     visit_type_uint32(v, name, ptr, errp);
1460 }
1461
1462 static void s390_pci_set_fid(Object *obj, Visitor *v, const char *name,
1463                          void *opaque, Error **errp)
1464 {
1465     S390PCIBusDevice *zpci = S390_PCI_DEVICE(obj);
1466     Property *prop = opaque;
1467     uint32_t *ptr = object_field_prop_ptr(obj, prop);
1468
1469     if (!visit_type_uint32(v, name, ptr, errp)) {
1470         return;
1471     }
1472     zpci->fid_defined = true;
1473 }
1474
1475 static const PropertyInfo s390_pci_fid_propinfo = {
1476     .name = "zpci_fid",
1477     .get = s390_pci_get_fid,
1478     .set = s390_pci_set_fid,
1479 };
1480
1481 #define DEFINE_PROP_S390_PCI_FID(_n, _s, _f) \
1482     DEFINE_PROP(_n, _s, _f, s390_pci_fid_propinfo, uint32_t)
1483
1484 static Property s390_pci_device_properties[] = {
1485     DEFINE_PROP_UINT16("uid", S390PCIBusDevice, uid, UID_UNDEFINED),
1486     DEFINE_PROP_S390_PCI_FID("fid", S390PCIBusDevice, fid),
1487     DEFINE_PROP_STRING("target", S390PCIBusDevice, target),
1488     DEFINE_PROP_BOOL("interpret", S390PCIBusDevice, interp, true),
1489     DEFINE_PROP_BOOL("forwarding-assist", S390PCIBusDevice, forwarding_assist,
1490                      true),
1491     DEFINE_PROP_END_OF_LIST(),
1492 };
1493
1494 static const VMStateDescription s390_pci_device_vmstate = {
1495     .name = TYPE_S390_PCI_DEVICE,
1496     /*
1497      * TODO: add state handling here, so migration works at least with
1498      * emulated pci devices on s390x
1499      */
1500     .unmigratable = 1,
1501 };
1502
1503 static void s390_pci_device_class_init(ObjectClass *klass, void *data)
1504 {
1505     DeviceClass *dc = DEVICE_CLASS(klass);
1506
1507     dc->desc = "zpci device";
1508     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
1509     dc->reset = s390_pci_device_reset;
1510     dc->bus_type = TYPE_S390_PCI_BUS;
1511     dc->realize = s390_pci_device_realize;
1512     device_class_set_props(dc, s390_pci_device_properties);
1513     dc->vmsd = &s390_pci_device_vmstate;
1514 }
1515
1516 static const TypeInfo s390_pci_device_info = {
1517     .name = TYPE_S390_PCI_DEVICE,
1518     .parent = TYPE_DEVICE,
1519     .instance_size = sizeof(S390PCIBusDevice),
1520     .class_init = s390_pci_device_class_init,
1521 };
1522
1523 static const TypeInfo s390_pci_iommu_info = {
1524     .name = TYPE_S390_PCI_IOMMU,
1525     .parent = TYPE_OBJECT,
1526     .instance_size = sizeof(S390PCIIOMMU),
1527 };
1528
1529 static void s390_iommu_memory_region_class_init(ObjectClass *klass, void *data)
1530 {
1531     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_CLASS(klass);
1532
1533     imrc->translate = s390_translate_iommu;
1534     imrc->replay = s390_pci_iommu_replay;
1535 }
1536
1537 static const TypeInfo s390_iommu_memory_region_info = {
1538     .parent = TYPE_IOMMU_MEMORY_REGION,
1539     .name = TYPE_S390_IOMMU_MEMORY_REGION,
1540     .class_init = s390_iommu_memory_region_class_init,
1541 };
1542
1543 static void s390_pci_register_types(void)
1544 {
1545     type_register_static(&s390_pcihost_info);
1546     type_register_static(&s390_pcibus_info);
1547     type_register_static(&s390_pci_device_info);
1548     type_register_static(&s390_pci_iommu_info);
1549     type_register_static(&s390_iommu_memory_region_info);
1550 }
1551
1552 type_init(s390_pci_register_types)