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 #include "hw.h"
  17 #include "pc.h"
  18 #include "pci.h"
  19 #include "msix.h"
  20 #include "kvm.h"
  21 #include "migration.h"
  22 #include "qerror.h"
  23
  24 #include <sys/mman.h>
  25 #include <sys/types.h>
  26
  27 #define IVSHMEM_IOEVENTFD   0
  28 #define IVSHMEM_MSI     1
  29
  30 #define IVSHMEM_PEER    0
  31 #define IVSHMEM_MASTER  1
  32
  33 #define IVSHMEM_REG_BAR_SIZE 0x100
  34
  35 //#define DEBUG_IVSHMEM
  36 #ifdef DEBUG_IVSHMEM
  37 #define IVSHMEM_DPRINTF(fmt, ...)        \
  38     do {printf("IVSHMEM: " fmt, ## __VA_ARGS__); } while (0)
  39 #else
  40 #define IVSHMEM_DPRINTF(fmt, ...)
  41 #endif
  42
  43 typedef struct Peer {
  44     int nb_eventfds;
  45     int *eventfds;
  46 } Peer;
  47
  48 typedef struct EventfdEntry {
  49     PCIDevice *pdev;
  50     int vector;
  51 } EventfdEntry;
  52
  53 typedef struct IVShmemState {
  54     PCIDevice dev;
  55     uint32_t intrmask;
  56     uint32_t intrstatus;
  57     uint32_t doorbell;
  58
  59     CharDriverState **eventfd_chr;
  60     CharDriverState *server_chr;
  61     MemoryRegion ivshmem_mmio;
  62
  63     pcibus_t mmio_addr;
  64     /* We might need to register the BAR before we actually have the memory.
  65      * So prepare a container MemoryRegion for the BAR immediately and
  66      * add a subregion when we have the memory.
  67      */
  68     MemoryRegion bar;
  69     MemoryRegion ivshmem;
  70     MemoryRegion msix_bar;
  71     uint64_t ivshmem_size; /* size of shared memory region */
  72     int shm_fd; /* shared memory file descriptor */
  73
  74     Peer *peers;
  75     int nb_peers; /* how many guests we have space for */
  76     int max_peer; /* maximum numbered peer */
  77
  78     int vm_id;
  79     uint32_t vectors;
  80     uint32_t features;
  81     EventfdEntry *eventfd_table;
  82
  83     Error *migration_blocker;
  84
  85     char * shmobj;
  86     char * sizearg;
  87     char * role;
  88     int role_val;   /* scalar to avoid multiple string comparisons */
  89 } IVShmemState;
  90
  91 /* registers for the Inter-VM shared memory device */
  92 enum ivshmem_registers {
  93     INTRMASK = 0,
  94     INTRSTATUS = 4,
  95     IVPOSITION = 8,
  96     DOORBELL = 12,
  97 };
  98
  99 static inline uint32_t ivshmem_has_feature(IVShmemState *ivs,
 100                                                     unsigned int feature) {
 101     return (ivs->features & (1 << feature));
 102 }
 103
 104 static inline bool is_power_of_two(uint64_t x) {
 105     return (x & (x - 1)) == 0;
 106 }
 107
 108 /* accessing registers - based on rtl8139 */
 109 static void ivshmem_update_irq(IVShmemState *s, int val)
 110 {
 111     int isr;
 112     isr = (s->intrstatus & s->intrmask) & 0xffffffff;
 113
 114     /* don't print ISR resets */
 115     if (isr) {
 116         IVSHMEM_DPRINTF("Set IRQ to %d (%04x %04x)\n",
 117            isr ? 1 : 0, s->intrstatus, s->intrmask);
 118     }
 119
 120     qemu_set_irq(s->dev.irq[0], (isr != 0));
 121 }
 122
 123 static void ivshmem_IntrMask_write(IVShmemState *s, uint32_t val)
 124 {
 125     IVSHMEM_DPRINTF("IntrMask write(w) val = 0x%04x\n", val);
 126
 127     s->intrmask = val;
 128
 129     ivshmem_update_irq(s, val);
 130 }
 131
 132 static uint32_t ivshmem_IntrMask_read(IVShmemState *s)
 133 {
 134     uint32_t ret = s->intrmask;
 135
 136     IVSHMEM_DPRINTF("intrmask read(w) val = 0x%04x\n", ret);
 137
 138     return ret;
 139 }
 140
 141 static void ivshmem_IntrStatus_write(IVShmemState *s, uint32_t val)
 142 {
 143     IVSHMEM_DPRINTF("IntrStatus write(w) val = 0x%04x\n", val);
 144
 145     s->intrstatus = val;
 146
 147     ivshmem_update_irq(s, val);
 148     return;
 149 }
 150
 151 static uint32_t ivshmem_IntrStatus_read(IVShmemState *s)
 152 {
 153     uint32_t ret = s->intrstatus;
 154
 155     /* reading ISR clears all interrupts */
 156     s->intrstatus = 0;
 157
 158     ivshmem_update_irq(s, 0);
 159
 160     return ret;
 161 }
 162
 163 static void ivshmem_io_write(void *opaque, target_phys_addr_t addr,
 164                              uint64_t val, unsigned size)
 165 {
 166     IVShmemState *s = opaque;
 167
 168     uint64_t write_one = 1;
 169     uint16_t dest = val >> 16;
 170     uint16_t vector = val & 0xff;
 171
 172     addr &= 0xfc;
 173
 174     IVSHMEM_DPRINTF("writing to addr " TARGET_FMT_plx "\n", addr);
 175     switch (addr)
 176     {
 177         case INTRMASK:
 178             ivshmem_IntrMask_write(s, val);
 179             break;
 180
 181         case INTRSTATUS:
 182             ivshmem_IntrStatus_write(s, val);
 183             break;
 184
 185         case DOORBELL:
 186             /* check that dest VM ID is reasonable */
 187             if (dest > s->max_peer) {
 188                 IVSHMEM_DPRINTF("Invalid destination VM ID (%d)\n", dest);
 189                 break;
 190             }
 191
 192             /* check doorbell range */
 193             if (vector < s->peers[dest].nb_eventfds) {
 194                 IVSHMEM_DPRINTF("Writing %" PRId64 " to VM %d on vector %d\n",
 195                                                     write_one, dest, vector);
 196                 if (write(s->peers[dest].eventfds[vector],
 197                                                     &(write_one), 8) != 8) {
 198                     IVSHMEM_DPRINTF("error writing to eventfd\n");
 199                 }
 200             }
 201             break;
 202         default:
 203             IVSHMEM_DPRINTF("Invalid VM Doorbell VM %d\n", dest);
 204     }
 205 }
 206
 207 static uint64_t ivshmem_io_read(void *opaque, target_phys_addr_t addr,
 208                                 unsigned size)
 209 {
 210
 211     IVShmemState *s = opaque;
 212     uint32_t ret;
 213
 214     switch (addr)
 215     {
 216         case INTRMASK:
 217             ret = ivshmem_IntrMask_read(s);
 218             break;
 219
 220         case INTRSTATUS:
 221             ret = ivshmem_IntrStatus_read(s);
 222             break;
 223
 224         case IVPOSITION:
 225             /* return my VM ID if the memory is mapped */
 226             if (s->shm_fd > 0) {
 227                 ret = s->vm_id;
 228             } else {
 229                 ret = -1;
 230             }
 231             break;
 232
 233         default:
 234             IVSHMEM_DPRINTF("why are we reading " TARGET_FMT_plx "\n", addr);
 235             ret = 0;
 236     }
 237
 238     return ret;
 239 }
 240
 241 static const MemoryRegionOps ivshmem_mmio_ops = {
 242     .read = ivshmem_io_read,
 243     .write = ivshmem_io_write,
 244     .endianness = DEVICE_NATIVE_ENDIAN,
 245     .impl = {
 246         .min_access_size = 4,
 247         .max_access_size = 4,
 248     },
 249 };
 250
 251 static void ivshmem_receive(void *opaque, const uint8_t *buf, int size)
 252 {
 253     IVShmemState *s = opaque;
 254
 255     ivshmem_IntrStatus_write(s, *buf);
 256
 257     IVSHMEM_DPRINTF("ivshmem_receive 0x%02x\n", *buf);
 258 }
 259
 260 static int ivshmem_can_receive(void * opaque)
 261 {
 262     return 8;
 263 }
 264
 265 static void ivshmem_event(void *opaque, int event)
 266 {
 267     IVSHMEM_DPRINTF("ivshmem_event %d\n", event);
 268 }
 269
 270 static void fake_irqfd(void *opaque, const uint8_t *buf, int size) {
 271
 272     EventfdEntry *entry = opaque;
 273     PCIDevice *pdev = entry->pdev;
 274
 275     IVSHMEM_DPRINTF("interrupt on vector %p %d\n", pdev, entry->vector);
 276     msix_notify(pdev, entry->vector);
 277 }
 278
 279 static CharDriverState* create_eventfd_chr_device(void * opaque, int eventfd,
 280                                                                     int vector)
 281 {
 282     /* create a event character device based on the passed eventfd */
 283     IVShmemState *s = opaque;
 284     CharDriverState * chr;
 285
 286     chr = qemu_chr_open_eventfd(eventfd);
 287
 288     if (chr == NULL) {
 289         fprintf(stderr, "creating eventfd for eventfd %d failed\n", eventfd);
 290         exit(-1);
 291     }
 292
 293     /* if MSI is supported we need multiple interrupts */
 294     if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
 295         s->eventfd_table[vector].pdev = &s->dev;
 296         s->eventfd_table[vector].vector = vector;
 297
 298         qemu_chr_add_handlers(chr, ivshmem_can_receive, fake_irqfd,
 299                       ivshmem_event, &s->eventfd_table[vector]);
 300     } else {
 301         qemu_chr_add_handlers(chr, ivshmem_can_receive, ivshmem_receive,
 302                       ivshmem_event, s);
 303     }
 304
 305     return chr;
 306
 307 }
 308
 309 static int check_shm_size(IVShmemState *s, int fd) {
 310     /* check that the guest isn't going to try and map more memory than the
 311      * the object has allocated return -1 to indicate error */
 312
 313     struct stat buf;
 314
 315     fstat(fd, &buf);
 316
 317     if (s->ivshmem_size > buf.st_size) {
 318         fprintf(stderr,
 319                 "IVSHMEM ERROR: Requested memory size greater"
 320                 " than shared object size (%" PRIu64 " > %" PRIu64")\n",
 321                 s->ivshmem_size, (uint64_t)buf.st_size);
 322         return -1;
 323     } else {
 324         return 0;
 325     }
 326 }
 327
 328 /* create the shared memory BAR when we are not using the server, so we can
 329  * create the BAR and map the memory immediately */
 330 static void create_shared_memory_BAR(IVShmemState *s, int fd) {
 331
 332     void * ptr;
 333
 334     s->shm_fd = fd;
 335
 336     ptr = mmap(0, s->ivshmem_size, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
 337
 338     memory_region_init_ram_ptr(&s->ivshmem, "ivshmem.bar2",
 339                                s->ivshmem_size, ptr);
 340     vmstate_register_ram(&s->ivshmem, &s->dev.qdev);
 341     memory_region_add_subregion(&s->bar, 0, &s->ivshmem);
 342
 343     /* region for shared memory */
 344     pci_register_bar(&s->dev, 2, PCI_BASE_ADDRESS_SPACE_MEMORY, &s->bar);
 345 }
 346
 347 static void close_guest_eventfds(IVShmemState *s, int posn)
 348 {
 349     int i, guest_curr_max;
 350
 351     guest_curr_max = s->peers[posn].nb_eventfds;
 352
 353     for (i = 0; i < guest_curr_max; i++) {
 354         kvm_set_ioeventfd_mmio_long(s->peers[posn].eventfds[i],
 355                     s->mmio_addr + DOORBELL, (posn << 16) | i, 0);
 356         close(s->peers[posn].eventfds[i]);
 357     }
 358
 359     g_free(s->peers[posn].eventfds);
 360     s->peers[posn].nb_eventfds = 0;
 361 }
 362
 363 static void setup_ioeventfds(IVShmemState *s) {
 364
 365     int i, j;
 366
 367     for (i = 0; i <= s->max_peer; i++) {
 368         for (j = 0; j < s->peers[i].nb_eventfds; j++) {
 369             memory_region_add_eventfd(&s->ivshmem_mmio,
 370                                       DOORBELL,
 371                                       4,
 372                                       true,
 373                                       (i << 16) | j,
 374                                       s->peers[i].eventfds[j]);
 375         }
 376     }
 377 }
 378
 379 /* this function increase the dynamic storage need to store data about other
 380  * guests */
 381 static void increase_dynamic_storage(IVShmemState *s, int new_min_size) {
 382
 383     int j, old_nb_alloc;
 384
 385     old_nb_alloc = s->nb_peers;
 386
 387     while (new_min_size >= s->nb_peers)
 388         s->nb_peers = s->nb_peers * 2;
 389
 390     IVSHMEM_DPRINTF("bumping storage to %d guests\n", s->nb_peers);
 391     s->peers = g_realloc(s->peers, s->nb_peers * sizeof(Peer));
 392
 393     /* zero out new pointers */
 394     for (j = old_nb_alloc; j < s->nb_peers; j++) {
 395         s->peers[j].eventfds = NULL;
 396         s->peers[j].nb_eventfds = 0;
 397     }
 398 }
 399
 400 static void ivshmem_read(void *opaque, const uint8_t * buf, int flags)
 401 {
 402     IVShmemState *s = opaque;
 403     int incoming_fd, tmp_fd;
 404     int guest_max_eventfd;
 405     long incoming_posn;
 406
 407     memcpy(&incoming_posn, buf, sizeof(long));
 408     /* pick off s->server_chr->msgfd and store it, posn should accompany msg */
 409     tmp_fd = qemu_chr_fe_get_msgfd(s->server_chr);
 410     IVSHMEM_DPRINTF("posn is %ld, fd is %d\n", incoming_posn, tmp_fd);
 411
 412     /* make sure we have enough space for this guest */
 413     if (incoming_posn >= s->nb_peers) {
 414         increase_dynamic_storage(s, incoming_posn);
 415     }
 416
 417     if (tmp_fd == -1) {
 418         /* if posn is positive and unseen before then this is our posn*/
 419         if ((incoming_posn >= 0) &&
 420                             (s->peers[incoming_posn].eventfds == NULL)) {
 421             /* receive our posn */
 422             s->vm_id = incoming_posn;
 423             return;
 424         } else {
 425             /* otherwise an fd == -1 means an existing guest has gone away */
 426             IVSHMEM_DPRINTF("posn %ld has gone away\n", incoming_posn);
 427             close_guest_eventfds(s, incoming_posn);
 428             return;
 429         }
 430     }
 431
 432     /* because of the implementation of get_msgfd, we need a dup */
 433     incoming_fd = dup(tmp_fd);
 434
 435     if (incoming_fd == -1) {
 436         fprintf(stderr, "could not allocate file descriptor %s\n",
 437                                                             strerror(errno));
 438         return;
 439     }
 440
 441     /* if the position is -1, then it's shared memory region fd */
 442     if (incoming_posn == -1) {
 443
 444         void * map_ptr;
 445
 446         s->max_peer = 0;
 447
 448         if (check_shm_size(s, incoming_fd) == -1) {
 449             exit(-1);
 450         }
 451
 452         /* mmap the region and map into the BAR2 */
 453         map_ptr = mmap(0, s->ivshmem_size, PROT_READ|PROT_WRITE, MAP_SHARED,
 454                                                             incoming_fd, 0);
 455         memory_region_init_ram_ptr(&s->ivshmem,
 456                                    "ivshmem.bar2", s->ivshmem_size, map_ptr);
 457         vmstate_register_ram(&s->ivshmem, &s->dev.qdev);
 458
 459         IVSHMEM_DPRINTF("guest h/w addr = %" PRIu64 ", size = %" PRIu64 "\n",
 460                          s->ivshmem_offset, s->ivshmem_size);
 461
 462         memory_region_add_subregion(&s->bar, 0, &s->ivshmem);
 463
 464         /* only store the fd if it is successfully mapped */
 465         s->shm_fd = incoming_fd;
 466
 467         return;
 468     }
 469
 470     /* each guest has an array of eventfds, and we keep track of how many
 471      * guests for each VM */
 472     guest_max_eventfd = s->peers[incoming_posn].nb_eventfds;
 473
 474     if (guest_max_eventfd == 0) {
 475         /* one eventfd per MSI vector */
 476         s->peers[incoming_posn].eventfds = (int *) g_malloc(s->vectors *
 477                                                                 sizeof(int));
 478     }
 479
 480     /* this is an eventfd for a particular guest VM */
 481     IVSHMEM_DPRINTF("eventfds[%ld][%d] = %d\n", incoming_posn,
 482                                             guest_max_eventfd, incoming_fd);
 483     s->peers[incoming_posn].eventfds[guest_max_eventfd] = incoming_fd;
 484
 485     /* increment count for particular guest */
 486     s->peers[incoming_posn].nb_eventfds++;
 487
 488     /* keep track of the maximum VM ID */
 489     if (incoming_posn > s->max_peer) {
 490         s->max_peer = incoming_posn;
 491     }
 492
 493     if (incoming_posn == s->vm_id) {
 494         s->eventfd_chr[guest_max_eventfd] = create_eventfd_chr_device(s,
 495                    s->peers[s->vm_id].eventfds[guest_max_eventfd],
 496                    guest_max_eventfd);
 497     }
 498
 499     if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD)) {
 500         if (kvm_set_ioeventfd_mmio_long(incoming_fd, s->mmio_addr + DOORBELL,
 501                         (incoming_posn << 16) | guest_max_eventfd, 1) < 0) {
 502             fprintf(stderr, "ivshmem: ioeventfd not available\n");
 503         }
 504     }
 505
 506     return;
 507 }
 508
 509 static void ivshmem_reset(DeviceState *d)
 510 {
 511     IVShmemState *s = DO_UPCAST(IVShmemState, dev.qdev, d);
 512
 513     s->intrstatus = 0;
 514     return;
 515 }
 516
 517 static uint64_t ivshmem_get_size(IVShmemState * s) {
 518
 519     uint64_t value;
 520     char *ptr;
 521
 522     value = strtoull(s->sizearg, &ptr, 10);
 523     switch (*ptr) {
 524         case 0: case 'M': case 'm':
 525             value <<= 20;
 526             break;
 527         case 'G': case 'g':
 528             value <<= 30;
 529             break;
 530         default:
 531             fprintf(stderr, "qemu: invalid ram size: %s\n", s->sizearg);
 532             exit(1);
 533     }
 534
 535     /* BARs must be a power of 2 */
 536     if (!is_power_of_two(value)) {
 537         fprintf(stderr, "ivshmem: size must be power of 2\n");
 538         exit(1);
 539     }
 540
 541     return value;
 542 }
 543
 544 static void ivshmem_setup_msi(IVShmemState * s) {
 545
 546     int i;
 547
 548     /* allocate the MSI-X vectors */
 549
 550     memory_region_init(&s->msix_bar, "ivshmem-msix", 4096);
 551     if (!msix_init(&s->dev, s->vectors, &s->msix_bar, 1, 0)) {
 552         pci_register_bar(&s->dev, 1, PCI_BASE_ADDRESS_SPACE_MEMORY,
 553                          &s->msix_bar);
 554         IVSHMEM_DPRINTF("msix initialized (%d vectors)\n", s->vectors);
 555     } else {
 556         IVSHMEM_DPRINTF("msix initialization failed\n");
 557         exit(1);
 558     }
 559
 560     /* 'activate' the vectors */
 561     for (i = 0; i < s->vectors; i++) {
 562         msix_vector_use(&s->dev, i);
 563     }
 564
 565     /* allocate Qemu char devices for receiving interrupts */
 566     s->eventfd_table = g_malloc0(s->vectors * sizeof(EventfdEntry));
 567 }
 568
 569 static void ivshmem_save(QEMUFile* f, void *opaque)
 570 {
 571     IVShmemState *proxy = opaque;
 572
 573     IVSHMEM_DPRINTF("ivshmem_save\n");
 574     pci_device_save(&proxy->dev, f);
 575
 576     if (ivshmem_has_feature(proxy, IVSHMEM_MSI)) {
 577         msix_save(&proxy->dev, f);
 578     } else {
 579         qemu_put_be32(f, proxy->intrstatus);
 580         qemu_put_be32(f, proxy->intrmask);
 581     }
 582
 583 }
 584
 585 static int ivshmem_load(QEMUFile* f, void *opaque, int version_id)
 586 {
 587     IVSHMEM_DPRINTF("ivshmem_load\n");
 588
 589     IVShmemState *proxy = opaque;
 590     int ret, i;
 591
 592     if (version_id > 0) {
 593         return -EINVAL;
 594     }
 595
 596     if (proxy->role_val == IVSHMEM_PEER) {
 597         fprintf(stderr, "ivshmem: 'peer' devices are not migratable\n");
 598         return -EINVAL;
 599     }
 600
 601     ret = pci_device_load(&proxy->dev, f);
 602     if (ret) {
 603         return ret;
 604     }
 605
 606     if (ivshmem_has_feature(proxy, IVSHMEM_MSI)) {
 607         msix_load(&proxy->dev, f);
 608         for (i = 0; i < proxy->vectors; i++) {
 609             msix_vector_use(&proxy->dev, i);
 610         }
 611     } else {
 612         proxy->intrstatus = qemu_get_be32(f);
 613         proxy->intrmask = qemu_get_be32(f);
 614     }
 615
 616     return 0;
 617 }
 618
 619 static int pci_ivshmem_init(PCIDevice *dev)
 620 {
 621     IVShmemState *s = DO_UPCAST(IVShmemState, dev, dev);
 622     uint8_t *pci_conf;
 623
 624     if (s->sizearg == NULL)
 625         s->ivshmem_size = 4 << 20; /* 4 MB default */
 626     else {
 627         s->ivshmem_size = ivshmem_get_size(s);
 628     }
 629
 630     register_savevm(&s->dev.qdev, "ivshmem", 0, 0, ivshmem_save, ivshmem_load,
 631                                                                         dev);
 632
 633     /* IRQFD requires MSI */
 634     if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD) &&
 635         !ivshmem_has_feature(s, IVSHMEM_MSI)) {
 636         fprintf(stderr, "ivshmem: ioeventfd/irqfd requires MSI\n");
 637         exit(1);
 638     }
 639
 640     /* check that role is reasonable */
 641     if (s->role) {
 642         if (strncmp(s->role, "peer", 5) == 0) {
 643             s->role_val = IVSHMEM_PEER;
 644         } else if (strncmp(s->role, "master", 7) == 0) {
 645             s->role_val = IVSHMEM_MASTER;
 646         } else {
 647             fprintf(stderr, "ivshmem: 'role' must be 'peer' or 'master'\n");
 648             exit(1);
 649         }
 650     } else {
 651         s->role_val = IVSHMEM_MASTER; /* default */
 652     }
 653
 654     if (s->role_val == IVSHMEM_PEER) {
 655         error_set(&s->migration_blocker, QERR_DEVICE_FEATURE_BLOCKS_MIGRATION, "ivshmem", "peer mode");
 656         migrate_add_blocker(s->migration_blocker);
 657     }
 658
 659     pci_conf = s->dev.config;
 660     pci_conf[PCI_COMMAND] = PCI_COMMAND_IO | PCI_COMMAND_MEMORY;
 661
 662     pci_config_set_interrupt_pin(pci_conf, 1);
 663
 664     s->shm_fd = 0;
 665
 666     memory_region_init_io(&s->ivshmem_mmio, &ivshmem_mmio_ops, s,
 667                           "ivshmem-mmio", IVSHMEM_REG_BAR_SIZE);
 668
 669     if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD)) {
 670         setup_ioeventfds(s);
 671     }
 672
 673     /* region for registers*/
 674     pci_register_bar(&s->dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY,
 675                      &s->ivshmem_mmio);
 676
 677     memory_region_init(&s->bar, "ivshmem-bar2-container", s->ivshmem_size);
 678
 679     if ((s->server_chr != NULL) &&
 680                         (strncmp(s->server_chr->filename, "unix:", 5) == 0)) {
 681         /* if we get a UNIX socket as the parameter we will talk
 682          * to the ivshmem server to receive the memory region */
 683
 684         if (s->shmobj != NULL) {
 685             fprintf(stderr, "WARNING: do not specify both 'chardev' "
 686                                                 "and 'shm' with ivshmem\n");
 687         }
 688
 689         IVSHMEM_DPRINTF("using shared memory server (socket = %s)\n",
 690                                                     s->server_chr->filename);
 691
 692         if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
 693             ivshmem_setup_msi(s);
 694         }
 695
 696         /* we allocate enough space for 16 guests and grow as needed */
 697         s->nb_peers = 16;
 698         s->vm_id = -1;
 699
 700         /* allocate/initialize space for interrupt handling */
 701         s->peers = g_malloc0(s->nb_peers * sizeof(Peer));
 702
 703         pci_register_bar(&s->dev, 2,
 704                          PCI_BASE_ADDRESS_SPACE_MEMORY, &s->bar);
 705
 706         s->eventfd_chr = g_malloc0(s->vectors * sizeof(CharDriverState *));
 707
 708         qemu_chr_add_handlers(s->server_chr, ivshmem_can_receive, ivshmem_read,
 709                      ivshmem_event, s);
 710     } else {
 711         /* just map the file immediately, we're not using a server */
 712         int fd;
 713
 714         if (s->shmobj == NULL) {
 715             fprintf(stderr, "Must specify 'chardev' or 'shm' to ivshmem\n");
 716         }
 717
 718         IVSHMEM_DPRINTF("using shm_open (shm object = %s)\n", s->shmobj);
 719
 720         /* try opening with O_EXCL and if it succeeds zero the memory
 721          * by truncating to 0 */
 722         if ((fd = shm_open(s->shmobj, O_CREAT|O_RDWR|O_EXCL,
 723                         S_IRWXU|S_IRWXG|S_IRWXO)) > 0) {
 724            /* truncate file to length PCI device's memory */
 725             if (ftruncate(fd, s->ivshmem_size) != 0) {
 726                 fprintf(stderr, "ivshmem: could not truncate shared file\n");
 727             }
 728
 729         } else if ((fd = shm_open(s->shmobj, O_CREAT|O_RDWR,
 730                         S_IRWXU|S_IRWXG|S_IRWXO)) < 0) {
 731             fprintf(stderr, "ivshmem: could not open shared file\n");
 732             exit(-1);
 733
 734         }
 735
 736         if (check_shm_size(s, fd) == -1) {
 737             exit(-1);
 738         }
 739
 740         create_shared_memory_BAR(s, fd);
 741
 742     }
 743
 744     return 0;
 745 }
 746
 747 static int pci_ivshmem_uninit(PCIDevice *dev)
 748 {
 749     IVShmemState *s = DO_UPCAST(IVShmemState, dev, dev);
 750
 751     if (s->migration_blocker) {
 752         migrate_del_blocker(s->migration_blocker);
 753         error_free(s->migration_blocker);
 754     }
 755
 756     memory_region_destroy(&s->ivshmem_mmio);
 757     memory_region_del_subregion(&s->bar, &s->ivshmem);
 758     vmstate_unregister_ram(&s->ivshmem, &s->dev.qdev);
 759     memory_region_destroy(&s->ivshmem);
 760     memory_region_destroy(&s->bar);
 761     unregister_savevm(&dev->qdev, "ivshmem", s);
 762
 763     return 0;
 764 }
 765
 766 static PCIDeviceInfo ivshmem_info = {
 767     .qdev.name  = "ivshmem",
 768     .qdev.size  = sizeof(IVShmemState),
 769     .qdev.reset = ivshmem_reset,
 770     .init       = pci_ivshmem_init,
 771     .exit       = pci_ivshmem_uninit,
 772     .vendor_id  = PCI_VENDOR_ID_REDHAT_QUMRANET,
 773     .device_id  = 0x1110,
 774     .class_id   = PCI_CLASS_MEMORY_RAM,
 775     .qdev.props = (Property[]) {
 776         DEFINE_PROP_CHR("chardev", IVShmemState, server_chr),
 777         DEFINE_PROP_STRING("size", IVShmemState, sizearg),
 778         DEFINE_PROP_UINT32("vectors", IVShmemState, vectors, 1),
 779         DEFINE_PROP_BIT("ioeventfd", IVShmemState, features, IVSHMEM_IOEVENTFD, false),
 780         DEFINE_PROP_BIT("msi", IVShmemState, features, IVSHMEM_MSI, true),
 781         DEFINE_PROP_STRING("shm", IVShmemState, shmobj),
 782         DEFINE_PROP_STRING("role", IVShmemState, role),
 783         DEFINE_PROP_END_OF_LIST(),
 784     }
 785 };
 786
 787 static void ivshmem_register_devices(void)
 788 {
 789     pci_qdev_register(&ivshmem_info);
 790 }
 791
 792 device_init(ivshmem_register_devices)