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