hw/ivshmem.c

   1 /*
   2  * Inter-VM Shared Memory PCI device.
   3  *
   4  * Author:
   5  *      Cam Macdonell <cam@cs.ualberta.ca>
   6  *
   7  * Based On: cirrus_vga.c
   8  *          Copyright (c) 2004 Fabrice Bellard
   9  *          Copyright (c) 2004 Makoto Suzuki (suzu)
  10  *
  11  *      and rtl8139.c
  12  *          Copyright (c) 2006 Igor Kovalenko
  13  *
  14  * This code is licensed under the GNU GPL v2.
  15  *
  16  * Contributions after 2012-01-13 are licensed under the terms of the
  17  * GNU GPL, version 2 or (at your option) any later version.
  18  */
  19 #include "hw.h"
  20 #include "pc.h"
  21 #include "pci.h"
  22 #include "msix.h"
  23 #include "kvm.h"
  24 #include "migration.h"
  25 #include "qerror.h"
  26 #include "event_notifier.h"
  27
  28 #include <sys/mman.h>
  29 #include <sys/types.h>
  30
  31 #define IVSHMEM_IOEVENTFD   0
  32 #define IVSHMEM_MSI     1
  33
  34 #define IVSHMEM_PEER    0
  35 #define IVSHMEM_MASTER  1
  36
  37 #define IVSHMEM_REG_BAR_SIZE 0x100
  38
  39 //#define DEBUG_IVSHMEM
  40 #ifdef DEBUG_IVSHMEM
  41 #define IVSHMEM_DPRINTF(fmt, ...)        \
  42     do {printf("IVSHMEM: " fmt, ## __VA_ARGS__); } while (0)
  43 #else
  44 #define IVSHMEM_DPRINTF(fmt, ...)
  45 #endif
  46
  47 typedef struct Peer {
  48     int nb_eventfds;
  49     EventNotifier *eventfds;
  50 } Peer;
  51
  52 typedef struct EventfdEntry {
  53     PCIDevice *pdev;
  54     int vector;
  55 } EventfdEntry;
  56
  57 typedef struct IVShmemState {
  58     PCIDevice dev;
  59     uint32_t intrmask;
  60     uint32_t intrstatus;
  61     uint32_t doorbell;
  62
  63     CharDriverState **eventfd_chr;
  64     CharDriverState *server_chr;
  65     MemoryRegion ivshmem_mmio;
  66
  67     /* We might need to register the BAR before we actually have the memory.
  68      * So prepare a container MemoryRegion for the BAR immediately and
  69      * add a subregion when we have the memory.
  70      */
  71     MemoryRegion bar;
  72     MemoryRegion ivshmem;
  73     MemoryRegion msix_bar;
  74     uint64_t ivshmem_size; /* size of shared memory region */
  75     int shm_fd; /* shared memory file descriptor */
  76
  77     Peer *peers;
  78     int nb_peers; /* how many guests we have space for */
  79     int max_peer; /* maximum numbered peer */
  80
  81     int vm_id;
  82     uint32_t vectors;
  83     uint32_t features;
  84     EventfdEntry *eventfd_table;
  85
  86     Error *migration_blocker;
  87
  88     char * shmobj;
  89     char * sizearg;
  90     char * role;
  91     int role_val;   /* scalar to avoid multiple string comparisons */
  92 } IVShmemState;
  93
  94 /* registers for the Inter-VM shared memory device */
  95 enum ivshmem_registers {
  96     INTRMASK = 0,
  97     INTRSTATUS = 4,
  98     IVPOSITION = 8,
  99     DOORBELL = 12,
 100 };
 101
 102 static inline uint32_t ivshmem_has_feature(IVShmemState *ivs,
 103                                                     unsigned int feature) {
 104     return (ivs->features & (1 << feature));
 105 }
 106
 107 static inline bool is_power_of_two(uint64_t x) {
 108     return (x & (x - 1)) == 0;
 109 }
 110
 111 /* accessing registers - based on rtl8139 */
 112 static void ivshmem_update_irq(IVShmemState *s, int val)
 113 {
 114     int isr;
 115     isr = (s->intrstatus & s->intrmask) & 0xffffffff;
 116
 117     /* don't print ISR resets */
 118     if (isr) {
 119         IVSHMEM_DPRINTF("Set IRQ to %d (%04x %04x)\n",
 120            isr ? 1 : 0, s->intrstatus, s->intrmask);
 121     }
 122
 123     qemu_set_irq(s->dev.irq[0], (isr != 0));
 124 }
 125
 126 static void ivshmem_IntrMask_write(IVShmemState *s, uint32_t val)
 127 {
 128     IVSHMEM_DPRINTF("IntrMask write(w) val = 0x%04x\n", val);
 129
 130     s->intrmask = val;
 131
 132     ivshmem_update_irq(s, val);
 133 }
 134
 135 static uint32_t ivshmem_IntrMask_read(IVShmemState *s)
 136 {
 137     uint32_t ret = s->intrmask;
 138
 139     IVSHMEM_DPRINTF("intrmask read(w) val = 0x%04x\n", ret);
 140
 141     return ret;
 142 }
 143
 144 static void ivshmem_IntrStatus_write(IVShmemState *s, uint32_t val)
 145 {
 146     IVSHMEM_DPRINTF("IntrStatus write(w) val = 0x%04x\n", val);
 147
 148     s->intrstatus = val;
 149
 150     ivshmem_update_irq(s, val);
 151     return;
 152 }
 153
 154 static uint32_t ivshmem_IntrStatus_read(IVShmemState *s)
 155 {
 156     uint32_t ret = s->intrstatus;
 157
 158     /* reading ISR clears all interrupts */
 159     s->intrstatus = 0;
 160
 161     ivshmem_update_irq(s, 0);
 162
 163     return ret;
 164 }
 165
 166 static void ivshmem_io_write(void *opaque, target_phys_addr_t addr,
 167                              uint64_t val, unsigned size)
 168 {
 169     IVShmemState *s = opaque;
 170
 171     uint16_t dest = val >> 16;
 172     uint16_t vector = val & 0xff;
 173
 174     addr &= 0xfc;
 175
 176     IVSHMEM_DPRINTF("writing to addr " TARGET_FMT_plx "\n", addr);
 177     switch (addr)
 178     {
 179         case INTRMASK:
 180             ivshmem_IntrMask_write(s, val);
 181             break;
 182
 183         case INTRSTATUS:
 184             ivshmem_IntrStatus_write(s, val);
 185             break;
 186
 187         case DOORBELL:
 188             /* check that dest VM ID is reasonable */
 189             if (dest > s->max_peer) {
 190                 IVSHMEM_DPRINTF("Invalid destination VM ID (%d)\n", dest);
 191                 break;
 192             }
 193
 194             /* check doorbell range */
 195             if (vector < s->peers[dest].nb_eventfds) {
 196                 IVSHMEM_DPRINTF("Notifying VM %d on vector %d\n", dest, vector);
 197                 event_notifier_set(&s->peers[dest].eventfds[vector]);
 198             }
 199             break;
 200         default:
 201             IVSHMEM_DPRINTF("Invalid VM Doorbell VM %d\n", dest);
 202     }
 203 }
 204
 205 static uint64_t ivshmem_io_read(void *opaque, target_phys_addr_t addr,
 206                                 unsigned size)
 207 {
 208
 209     IVShmemState *s = opaque;
 210     uint32_t ret;
 211
 212     switch (addr)
 213     {
 214         case INTRMASK:
 215             ret = ivshmem_IntrMask_read(s);
 216             break;
 217
 218         case INTRSTATUS:
 219             ret = ivshmem_IntrStatus_read(s);
 220             break;
 221
 222         case IVPOSITION:
 223             /* return my VM ID if the memory is mapped */
 224             if (s->shm_fd > 0) {
 225                 ret = s->vm_id;
 226             } else {
 227                 ret = -1;
 228             }
 229             break;
 230
 231         default:
 232             IVSHMEM_DPRINTF("why are we reading " TARGET_FMT_plx "\n", addr);
 233             ret = 0;
 234     }
 235
 236     return ret;
 237 }
 238
 239 static const MemoryRegionOps ivshmem_mmio_ops = {
 240     .read = ivshmem_io_read,
 241     .write = ivshmem_io_write,
 242     .endianness = DEVICE_NATIVE_ENDIAN,
 243     .impl = {
 244         .min_access_size = 4,
 245         .max_access_size = 4,
 246     },
 247 };
 248
 249 static void ivshmem_receive(void *opaque, const uint8_t *buf, int size)
 250 {
 251     IVShmemState *s = opaque;
 252
 253     ivshmem_IntrStatus_write(s, *buf);
 254
 255     IVSHMEM_DPRINTF("ivshmem_receive 0x%02x\n", *buf);
 256 }
 257
 258 static int ivshmem_can_receive(void * opaque)
 259 {
 260     return 8;
 261 }
 262
 263 static void ivshmem_event(void *opaque, int event)
 264 {
 265     IVSHMEM_DPRINTF("ivshmem_event %d\n", event);
 266 }
 267
 268 static void fake_irqfd(void *opaque, const uint8_t *buf, int size) {
 269
 270     EventfdEntry *entry = opaque;
 271     PCIDevice *pdev = entry->pdev;
 272
 273     IVSHMEM_DPRINTF("interrupt on vector %p %d\n", pdev, entry->vector);
 274     msix_notify(pdev, entry->vector);
 275 }
 276
 277 static CharDriverState* create_eventfd_chr_device(void * opaque, EventNotifier *n,
 278                                                   int vector)
 279 {
 280     /* create a event character device based on the passed eventfd */
 281     IVShmemState *s = opaque;
 282     CharDriverState * chr;
 283     int eventfd = event_notifier_get_fd(n);
 284
 285     chr = qemu_chr_open_eventfd(eventfd);
 286
 287     if (chr == NULL) {
 288         fprintf(stderr, "creating eventfd for eventfd %d failed\n", eventfd);
 289         exit(-1);
 290     }
 291
 292     /* if MSI is supported we need multiple interrupts */
 293     if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
 294         s->eventfd_table[vector].pdev = &s->dev;
 295         s->eventfd_table[vector].vector = vector;
 296
 297         qemu_chr_add_handlers(chr, ivshmem_can_receive, fake_irqfd,
 298                       ivshmem_event, &s->eventfd_table[vector]);
 299     } else {
 300         qemu_chr_add_handlers(chr, ivshmem_can_receive, ivshmem_receive,
 301                       ivshmem_event, s);
 302     }
 303
 304     return chr;
 305
 306 }
 307
 308 static int check_shm_size(IVShmemState *s, int fd) {
 309     /* check that the guest isn't going to try and map more memory than the
 310      * the object has allocated return -1 to indicate error */
 311
 312     struct stat buf;
 313
 314     fstat(fd, &buf);
 315
 316     if (s->ivshmem_size > buf.st_size) {
 317         fprintf(stderr,
 318                 "IVSHMEM ERROR: Requested memory size greater"
 319                 " than shared object size (%" PRIu64 " > %" PRIu64")\n",
 320                 s->ivshmem_size, (uint64_t)buf.st_size);
 321         return -1;
 322     } else {
 323         return 0;
 324     }
 325 }
 326
 327 /* create the shared memory BAR when we are not using the server, so we can
 328  * create the BAR and map the memory immediately */
 329 static void create_shared_memory_BAR(IVShmemState *s, int fd) {
 330
 331     void * ptr;
 332
 333     s->shm_fd = fd;
 334
 335     ptr = mmap(0, s->ivshmem_size, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
 336
 337     memory_region_init_ram_ptr(&s->ivshmem, "ivshmem.bar2",
 338                                s->ivshmem_size, ptr);
 339     vmstate_register_ram(&s->ivshmem, &s->dev.qdev);
 340     memory_region_add_subregion(&s->bar, 0, &s->ivshmem);
 341
 342     /* region for shared memory */
 343     pci_register_bar(&s->dev, 2, PCI_BASE_ADDRESS_SPACE_MEMORY, &s->bar);
 344 }
 345
 346 static void ivshmem_add_eventfd(IVShmemState *s, int posn, int i)
 347 {
 348     memory_region_add_eventfd(&s->ivshmem_mmio,
 349                               DOORBELL,
 350                               4,
 351                               true,
 352                               (posn << 16) | i,
 353                               &s->peers[posn].eventfds[i]);
 354 }
 355
 356 static void ivshmem_del_eventfd(IVShmemState *s, int posn, int i)
 357 {
 358     memory_region_del_eventfd(&s->ivshmem_mmio,
 359                               DOORBELL,
 360                               4,
 361                               true,
 362                               (posn << 16) | i,
 363                               &s->peers[posn].eventfds[i]);
 364 }
 365
 366 static void close_guest_eventfds(IVShmemState *s, int posn)
 367 {
 368     int i, guest_curr_max;
 369
 370     guest_curr_max = s->peers[posn].nb_eventfds;
 371
 372     memory_region_transaction_begin();
 373     for (i = 0; i < guest_curr_max; i++) {
 374         ivshmem_del_eventfd(s, posn, i);
 375     }
 376     memory_region_transaction_commit();
 377     for (i = 0; i < guest_curr_max; i++) {
 378         event_notifier_cleanup(&s->peers[posn].eventfds[i]);
 379     }
 380
 381     g_free(s->peers[posn].eventfds);
 382     s->peers[posn].nb_eventfds = 0;
 383 }
 384
 385 static void setup_ioeventfds(IVShmemState *s) {
 386
 387     int i, j;
 388
 389     for (i = 0; i <= s->max_peer; i++) {
 390         for (j = 0; j < s->peers[i].nb_eventfds; j++) {
 391             ivshmem_add_eventfd(s, i, j);
 392         }
 393     }
 394 }
 395
 396 /* this function increase the dynamic storage need to store data about other
 397  * guests */
 398 static void increase_dynamic_storage(IVShmemState *s, int new_min_size) {
 399
 400     int j, old_nb_alloc;
 401
 402     old_nb_alloc = s->nb_peers;
 403
 404     while (new_min_size >= s->nb_peers)
 405         s->nb_peers = s->nb_peers * 2;
 406
 407     IVSHMEM_DPRINTF("bumping storage to %d guests\n", s->nb_peers);
 408     s->peers = g_realloc(s->peers, s->nb_peers * sizeof(Peer));
 409
 410     /* zero out new pointers */
 411     for (j = old_nb_alloc; j < s->nb_peers; j++) {
 412         s->peers[j].eventfds = NULL;
 413         s->peers[j].nb_eventfds = 0;
 414     }
 415 }
 416
 417 static void ivshmem_read(void *opaque, const uint8_t * buf, int flags)
 418 {
 419     IVShmemState *s = opaque;
 420     int incoming_fd, tmp_fd;
 421     int guest_max_eventfd;
 422     long incoming_posn;
 423
 424     memcpy(&incoming_posn, buf, sizeof(long));
 425     /* pick off s->server_chr->msgfd and store it, posn should accompany msg */
 426     tmp_fd = qemu_chr_fe_get_msgfd(s->server_chr);
 427     IVSHMEM_DPRINTF("posn is %ld, fd is %d\n", incoming_posn, tmp_fd);
 428
 429     /* make sure we have enough space for this guest */
 430     if (incoming_posn >= s->nb_peers) {
 431         increase_dynamic_storage(s, incoming_posn);
 432     }
 433
 434     if (tmp_fd == -1) {
 435         /* if posn is positive and unseen before then this is our posn*/
 436         if ((incoming_posn >= 0) &&
 437                             (s->peers[incoming_posn].eventfds == NULL)) {
 438             /* receive our posn */
 439             s->vm_id = incoming_posn;
 440             return;
 441         } else {
 442             /* otherwise an fd == -1 means an existing guest has gone away */
 443             IVSHMEM_DPRINTF("posn %ld has gone away\n", incoming_posn);
 444             close_guest_eventfds(s, incoming_posn);
 445             return;
 446         }
 447     }
 448
 449     /* because of the implementation of get_msgfd, we need a dup */
 450     incoming_fd = dup(tmp_fd);
 451
 452     if (incoming_fd == -1) {
 453         fprintf(stderr, "could not allocate file descriptor %s\n",
 454                                                             strerror(errno));
 455         return;
 456     }
 457
 458     /* if the position is -1, then it's shared memory region fd */
 459     if (incoming_posn == -1) {
 460
 461         void * map_ptr;
 462
 463         s->max_peer = 0;
 464
 465         if (check_shm_size(s, incoming_fd) == -1) {
 466             exit(-1);
 467         }
 468
 469         /* mmap the region and map into the BAR2 */
 470         map_ptr = mmap(0, s->ivshmem_size, PROT_READ|PROT_WRITE, MAP_SHARED,
 471                                                             incoming_fd, 0);
 472         memory_region_init_ram_ptr(&s->ivshmem,
 473                                    "ivshmem.bar2", s->ivshmem_size, map_ptr);
 474         vmstate_register_ram(&s->ivshmem, &s->dev.qdev);
 475
 476         IVSHMEM_DPRINTF("guest h/w addr = %" PRIu64 ", size = %" PRIu64 "\n",
 477                          s->ivshmem_offset, s->ivshmem_size);
 478
 479         memory_region_add_subregion(&s->bar, 0, &s->ivshmem);
 480
 481         /* only store the fd if it is successfully mapped */
 482         s->shm_fd = incoming_fd;
 483
 484         return;
 485     }
 486
 487     /* each guest has an array of eventfds, and we keep track of how many
 488      * guests for each VM */
 489     guest_max_eventfd = s->peers[incoming_posn].nb_eventfds;
 490
 491     if (guest_max_eventfd == 0) {
 492         /* one eventfd per MSI vector */
 493         s->peers[incoming_posn].eventfds = g_new(EventNotifier, s->vectors);
 494     }
 495
 496     /* this is an eventfd for a particular guest VM */
 497     IVSHMEM_DPRINTF("eventfds[%ld][%d] = %d\n", incoming_posn,
 498                                             guest_max_eventfd, incoming_fd);
 499     event_notifier_init_fd(&s->peers[incoming_posn].eventfds[guest_max_eventfd],
 500                            incoming_fd);
 501
 502     /* increment count for particular guest */
 503     s->peers[incoming_posn].nb_eventfds++;
 504
 505     /* keep track of the maximum VM ID */
 506     if (incoming_posn > s->max_peer) {
 507         s->max_peer = incoming_posn;
 508     }
 509
 510     if (incoming_posn == s->vm_id) {
 511         s->eventfd_chr[guest_max_eventfd] = create_eventfd_chr_device(s,
 512                    &s->peers[s->vm_id].eventfds[guest_max_eventfd],
 513                    guest_max_eventfd);
 514     }
 515
 516     if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD)) {
 517         ivshmem_add_eventfd(s, incoming_posn, guest_max_eventfd);
 518     }
 519
 520     return;
 521 }
 522
 523 /* Select the MSI-X vectors used by device.
 524  * ivshmem maps events to vectors statically, so
 525  * we just enable all vectors on init and after reset. */
 526 static void ivshmem_use_msix(IVShmemState * s)
 527 {
 528     int i;
 529
 530     if (!msix_present(&s->dev)) {
 531         return;
 532     }
 533
 534     for (i = 0; i < s->vectors; i++) {
 535         msix_vector_use(&s->dev, i);
 536     }
 537 }
 538
 539 static void ivshmem_reset(DeviceState *d)
 540 {
 541     IVShmemState *s = DO_UPCAST(IVShmemState, dev.qdev, d);
 542
 543     s->intrstatus = 0;
 544     ivshmem_use_msix(s);
 545     return;
 546 }
 547
 548 static uint64_t ivshmem_get_size(IVShmemState * s) {
 549
 550     uint64_t value;
 551     char *ptr;
 552
 553     value = strtoull(s->sizearg, &ptr, 10);
 554     switch (*ptr) {
 555         case 0: case 'M': case 'm':
 556             value <<= 20;
 557             break;
 558         case 'G': case 'g':
 559             value <<= 30;
 560             break;
 561         default:
 562             fprintf(stderr, "qemu: invalid ram size: %s\n", s->sizearg);
 563             exit(1);
 564     }
 565
 566     /* BARs must be a power of 2 */
 567     if (!is_power_of_two(value)) {
 568         fprintf(stderr, "ivshmem: size must be power of 2\n");
 569         exit(1);
 570     }
 571
 572     return value;
 573 }
 574
 575 static void ivshmem_setup_msi(IVShmemState * s)
 576 {
 577     memory_region_init(&s->msix_bar, "ivshmem-msix", 4096);
 578     if (!msix_init(&s->dev, s->vectors, &s->msix_bar, 1, 0)) {
 579         pci_register_bar(&s->dev, 1, PCI_BASE_ADDRESS_SPACE_MEMORY,
 580                          &s->msix_bar);
 581         IVSHMEM_DPRINTF("msix initialized (%d vectors)\n", s->vectors);
 582     } else {
 583         IVSHMEM_DPRINTF("msix initialization failed\n");
 584         exit(1);
 585     }
 586
 587     /* allocate QEMU char devices for receiving interrupts */
 588     s->eventfd_table = g_malloc0(s->vectors * sizeof(EventfdEntry));
 589
 590     ivshmem_use_msix(s);
 591 }
 592
 593 static void ivshmem_save(QEMUFile* f, void *opaque)
 594 {
 595     IVShmemState *proxy = opaque;
 596
 597     IVSHMEM_DPRINTF("ivshmem_save\n");
 598     pci_device_save(&proxy->dev, f);
 599
 600     if (ivshmem_has_feature(proxy, IVSHMEM_MSI)) {
 601         msix_save(&proxy->dev, f);
 602     } else {
 603         qemu_put_be32(f, proxy->intrstatus);
 604         qemu_put_be32(f, proxy->intrmask);
 605     }
 606
 607 }
 608
 609 static int ivshmem_load(QEMUFile* f, void *opaque, int version_id)
 610 {
 611     IVSHMEM_DPRINTF("ivshmem_load\n");
 612
 613     IVShmemState *proxy = opaque;
 614     int ret;
 615
 616     if (version_id > 0) {
 617         return -EINVAL;
 618     }
 619
 620     if (proxy->role_val == IVSHMEM_PEER) {
 621         fprintf(stderr, "ivshmem: 'peer' devices are not migratable\n");
 622         return -EINVAL;
 623     }
 624
 625     ret = pci_device_load(&proxy->dev, f);
 626     if (ret) {
 627         return ret;
 628     }
 629
 630     if (ivshmem_has_feature(proxy, IVSHMEM_MSI)) {
 631         msix_load(&proxy->dev, f);
 632         ivshmem_use_msix(proxy);
 633     } else {
 634         proxy->intrstatus = qemu_get_be32(f);
 635         proxy->intrmask = qemu_get_be32(f);
 636     }
 637
 638     return 0;
 639 }
 640
 641 static void ivshmem_write_config(PCIDevice *pci_dev, uint32_t address,
 642                                  uint32_t val, int len)
 643 {
 644     pci_default_write_config(pci_dev, address, val, len);
 645     msix_write_config(pci_dev, address, val, len);
 646 }
 647
 648 static int pci_ivshmem_init(PCIDevice *dev)
 649 {
 650     IVShmemState *s = DO_UPCAST(IVShmemState, dev, dev);
 651     uint8_t *pci_conf;
 652
 653     if (s->sizearg == NULL)
 654         s->ivshmem_size = 4 << 20; /* 4 MB default */
 655     else {
 656         s->ivshmem_size = ivshmem_get_size(s);
 657     }
 658
 659     register_savevm(&s->dev.qdev, "ivshmem", 0, 0, ivshmem_save, ivshmem_load,
 660                                                                         dev);
 661
 662     /* IRQFD requires MSI */
 663     if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD) &&
 664         !ivshmem_has_feature(s, IVSHMEM_MSI)) {
 665         fprintf(stderr, "ivshmem: ioeventfd/irqfd requires MSI\n");
 666         exit(1);
 667     }
 668
 669     /* check that role is reasonable */
 670     if (s->role) {
 671         if (strncmp(s->role, "peer", 5) == 0) {
 672             s->role_val = IVSHMEM_PEER;
 673         } else if (strncmp(s->role, "master", 7) == 0) {
 674             s->role_val = IVSHMEM_MASTER;
 675         } else {
 676             fprintf(stderr, "ivshmem: 'role' must be 'peer' or 'master'\n");
 677             exit(1);
 678         }
 679     } else {
 680         s->role_val = IVSHMEM_MASTER; /* default */
 681     }
 682
 683     if (s->role_val == IVSHMEM_PEER) {
 684         error_set(&s->migration_blocker, QERR_DEVICE_FEATURE_BLOCKS_MIGRATION, "ivshmem", "peer mode");
 685         migrate_add_blocker(s->migration_blocker);
 686     }
 687
 688     pci_conf = s->dev.config;
 689     pci_conf[PCI_COMMAND] = PCI_COMMAND_IO | PCI_COMMAND_MEMORY;
 690
 691     pci_config_set_interrupt_pin(pci_conf, 1);
 692
 693     s->shm_fd = 0;
 694
 695     memory_region_init_io(&s->ivshmem_mmio, &ivshmem_mmio_ops, s,
 696                           "ivshmem-mmio", IVSHMEM_REG_BAR_SIZE);
 697
 698     if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD)) {
 699         setup_ioeventfds(s);
 700     }
 701
 702     /* region for registers*/
 703     pci_register_bar(&s->dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY,
 704                      &s->ivshmem_mmio);
 705
 706     memory_region_init(&s->bar, "ivshmem-bar2-container", s->ivshmem_size);
 707
 708     if ((s->server_chr != NULL) &&
 709                         (strncmp(s->server_chr->filename, "unix:", 5) == 0)) {
 710         /* if we get a UNIX socket as the parameter we will talk
 711          * to the ivshmem server to receive the memory region */
 712
 713         if (s->shmobj != NULL) {
 714             fprintf(stderr, "WARNING: do not specify both 'chardev' "
 715                                                 "and 'shm' with ivshmem\n");
 716         }
 717
 718         IVSHMEM_DPRINTF("using shared memory server (socket = %s)\n",
 719                                                     s->server_chr->filename);
 720
 721         if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
 722             ivshmem_setup_msi(s);
 723         }
 724
 725         /* we allocate enough space for 16 guests and grow as needed */
 726         s->nb_peers = 16;
 727         s->vm_id = -1;
 728
 729         /* allocate/initialize space for interrupt handling */
 730         s->peers = g_malloc0(s->nb_peers * sizeof(Peer));
 731
 732         pci_register_bar(&s->dev, 2,
 733                          PCI_BASE_ADDRESS_SPACE_MEMORY, &s->bar);
 734
 735         s->eventfd_chr = g_malloc0(s->vectors * sizeof(CharDriverState *));
 736
 737         qemu_chr_add_handlers(s->server_chr, ivshmem_can_receive, ivshmem_read,
 738                      ivshmem_event, s);
 739     } else {
 740         /* just map the file immediately, we're not using a server */
 741         int fd;
 742
 743         if (s->shmobj == NULL) {
 744             fprintf(stderr, "Must specify 'chardev' or 'shm' to ivshmem\n");
 745         }
 746
 747         IVSHMEM_DPRINTF("using shm_open (shm object = %s)\n", s->shmobj);
 748
 749         /* try opening with O_EXCL and if it succeeds zero the memory
 750          * by truncating to 0 */
 751         if ((fd = shm_open(s->shmobj, O_CREAT|O_RDWR|O_EXCL,
 752                         S_IRWXU|S_IRWXG|S_IRWXO)) > 0) {
 753            /* truncate file to length PCI device's memory */
 754             if (ftruncate(fd, s->ivshmem_size) != 0) {
 755                 fprintf(stderr, "ivshmem: could not truncate shared file\n");
 756             }
 757
 758         } else if ((fd = shm_open(s->shmobj, O_CREAT|O_RDWR,
 759                         S_IRWXU|S_IRWXG|S_IRWXO)) < 0) {
 760             fprintf(stderr, "ivshmem: could not open shared file\n");
 761             exit(-1);
 762
 763         }
 764
 765         if (check_shm_size(s, fd) == -1) {
 766             exit(-1);
 767         }
 768
 769         create_shared_memory_BAR(s, fd);
 770
 771     }
 772
 773     s->dev.config_write = ivshmem_write_config;
 774
 775     return 0;
 776 }
 777
 778 static int pci_ivshmem_uninit(PCIDevice *dev)
 779 {
 780     IVShmemState *s = DO_UPCAST(IVShmemState, dev, dev);
 781
 782     if (s->migration_blocker) {
 783         migrate_del_blocker(s->migration_blocker);
 784         error_free(s->migration_blocker);
 785     }
 786
 787     memory_region_destroy(&s->ivshmem_mmio);
 788     memory_region_del_subregion(&s->bar, &s->ivshmem);
 789     vmstate_unregister_ram(&s->ivshmem, &s->dev.qdev);
 790     memory_region_destroy(&s->ivshmem);
 791     memory_region_destroy(&s->bar);
 792     unregister_savevm(&dev->qdev, "ivshmem", s);
 793
 794     return 0;
 795 }
 796
 797 static Property ivshmem_properties[] = {
 798     DEFINE_PROP_CHR("chardev", IVShmemState, server_chr),
 799     DEFINE_PROP_STRING("size", IVShmemState, sizearg),
 800     DEFINE_PROP_UINT32("vectors", IVShmemState, vectors, 1),
 801     DEFINE_PROP_BIT("ioeventfd", IVShmemState, features, IVSHMEM_IOEVENTFD, false),
 802     DEFINE_PROP_BIT("msi", IVShmemState, features, IVSHMEM_MSI, true),
 803     DEFINE_PROP_STRING("shm", IVShmemState, shmobj),
 804     DEFINE_PROP_STRING("role", IVShmemState, role),
 805     DEFINE_PROP_END_OF_LIST(),
 806 };
 807
 808 static void ivshmem_class_init(ObjectClass *klass, void *data)
 809 {
 810     DeviceClass *dc = DEVICE_CLASS(klass);
 811     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
 812
 813     k->init = pci_ivshmem_init;
 814     k->exit = pci_ivshmem_uninit;
 815     k->vendor_id = PCI_VENDOR_ID_REDHAT_QUMRANET;
 816     k->device_id = 0x1110;
 817     k->class_id = PCI_CLASS_MEMORY_RAM;
 818     dc->reset = ivshmem_reset;
 819     dc->props = ivshmem_properties;
 820 }
 821
 822 static TypeInfo ivshmem_info = {
 823     .name          = "ivshmem",
 824     .parent        = TYPE_PCI_DEVICE,
 825     .instance_size = sizeof(IVShmemState),
 826     .class_init    = ivshmem_class_init,
 827 };
 828
 829 static void ivshmem_register_types(void)
 830 {
 831     type_register_static(&ivshmem_info);
 832 }
 833
 834 type_init(ivshmem_register_types)