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     uint64_t ivshmem_size; /* size of shared memory region */
  74     uint32_t ivshmem_attr;
  75     uint32_t ivshmem_64bit;
  76     int shm_fd; /* shared memory file descriptor */
  77
  78     Peer *peers;
  79     int nb_peers; /* how many guests we have space for */
  80     int max_peer; /* maximum numbered peer */
  81
  82     int vm_id;
  83     uint32_t vectors;
  84     uint32_t features;
  85     EventfdEntry *eventfd_table;
  86
  87     Error *migration_blocker;
  88
  89     char * shmobj;
  90     char * sizearg;
  91     char * role;
  92     int role_val;   /* scalar to avoid multiple string comparisons */
  93 } IVShmemState;
  94
  95 /* registers for the Inter-VM shared memory device */
  96 enum ivshmem_registers {
  97     INTRMASK = 0,
  98     INTRSTATUS = 4,
  99     IVPOSITION = 8,
 100     DOORBELL = 12,
 101 };
 102
 103 static inline uint32_t ivshmem_has_feature(IVShmemState *ivs,
 104                                                     unsigned int feature) {
 105     return (ivs->features & (1 << feature));
 106 }
 107
 108 static inline bool is_power_of_two(uint64_t x) {
 109     return (x & (x - 1)) == 0;
 110 }
 111
 112 /* accessing registers - based on rtl8139 */
 113 static void ivshmem_update_irq(IVShmemState *s, int val)
 114 {
 115     int isr;
 116     isr = (s->intrstatus & s->intrmask) & 0xffffffff;
 117
 118     /* don't print ISR resets */
 119     if (isr) {
 120         IVSHMEM_DPRINTF("Set IRQ to %d (%04x %04x)\n",
 121            isr ? 1 : 0, s->intrstatus, s->intrmask);
 122     }
 123
 124     qemu_set_irq(s->dev.irq[0], (isr != 0));
 125 }
 126
 127 static void ivshmem_IntrMask_write(IVShmemState *s, uint32_t val)
 128 {
 129     IVSHMEM_DPRINTF("IntrMask write(w) val = 0x%04x\n", val);
 130
 131     s->intrmask = val;
 132
 133     ivshmem_update_irq(s, val);
 134 }
 135
 136 static uint32_t ivshmem_IntrMask_read(IVShmemState *s)
 137 {
 138     uint32_t ret = s->intrmask;
 139
 140     IVSHMEM_DPRINTF("intrmask read(w) val = 0x%04x\n", ret);
 141
 142     return ret;
 143 }
 144
 145 static void ivshmem_IntrStatus_write(IVShmemState *s, uint32_t val)
 146 {
 147     IVSHMEM_DPRINTF("IntrStatus write(w) val = 0x%04x\n", val);
 148
 149     s->intrstatus = val;
 150
 151     ivshmem_update_irq(s, val);
 152     return;
 153 }
 154
 155 static uint32_t ivshmem_IntrStatus_read(IVShmemState *s)
 156 {
 157     uint32_t ret = s->intrstatus;
 158
 159     /* reading ISR clears all interrupts */
 160     s->intrstatus = 0;
 161
 162     ivshmem_update_irq(s, 0);
 163
 164     return ret;
 165 }
 166
 167 static void ivshmem_io_write(void *opaque, target_phys_addr_t addr,
 168                              uint64_t val, unsigned size)
 169 {
 170     IVShmemState *s = opaque;
 171
 172     uint16_t dest = val >> 16;
 173     uint16_t vector = val & 0xff;
 174
 175     addr &= 0xfc;
 176
 177     IVSHMEM_DPRINTF("writing to addr " TARGET_FMT_plx "\n", addr);
 178     switch (addr)
 179     {
 180         case INTRMASK:
 181             ivshmem_IntrMask_write(s, val);
 182             break;
 183
 184         case INTRSTATUS:
 185             ivshmem_IntrStatus_write(s, val);
 186             break;
 187
 188         case DOORBELL:
 189             /* check that dest VM ID is reasonable */
 190             if (dest > s->max_peer) {
 191                 IVSHMEM_DPRINTF("Invalid destination VM ID (%d)\n", dest);
 192                 break;
 193             }
 194
 195             /* check doorbell range */
 196             if (vector < s->peers[dest].nb_eventfds) {
 197                 IVSHMEM_DPRINTF("Notifying VM %d on vector %d\n", dest, vector);
 198                 event_notifier_set(&s->peers[dest].eventfds[vector]);
 199             }
 200             break;
 201         default:
 202             IVSHMEM_DPRINTF("Invalid VM Doorbell VM %d\n", dest);
 203     }
 204 }
 205
 206 static uint64_t ivshmem_io_read(void *opaque, target_phys_addr_t addr,
 207                                 unsigned size)
 208 {
 209
 210     IVShmemState *s = opaque;
 211     uint32_t ret;
 212
 213     switch (addr)
 214     {
 215         case INTRMASK:
 216             ret = ivshmem_IntrMask_read(s);
 217             break;
 218
 219         case INTRSTATUS:
 220             ret = ivshmem_IntrStatus_read(s);
 221             break;
 222
 223         case IVPOSITION:
 224             /* return my VM ID if the memory is mapped */
 225             if (s->shm_fd > 0) {
 226                 ret = s->vm_id;
 227             } else {
 228                 ret = -1;
 229             }
 230             break;
 231
 232         default:
 233             IVSHMEM_DPRINTF("why are we reading " TARGET_FMT_plx "\n", addr);
 234             ret = 0;
 235     }
 236
 237     return ret;
 238 }
 239
 240 static const MemoryRegionOps ivshmem_mmio_ops = {
 241     .read = ivshmem_io_read,
 242     .write = ivshmem_io_write,
 243     .endianness = DEVICE_NATIVE_ENDIAN,
 244     .impl = {
 245         .min_access_size = 4,
 246         .max_access_size = 4,
 247     },
 248 };
 249
 250 static void ivshmem_receive(void *opaque, const uint8_t *buf, int size)
 251 {
 252     IVShmemState *s = opaque;
 253
 254     ivshmem_IntrStatus_write(s, *buf);
 255
 256     IVSHMEM_DPRINTF("ivshmem_receive 0x%02x\n", *buf);
 257 }
 258
 259 static int ivshmem_can_receive(void * opaque)
 260 {
 261     return 8;
 262 }
 263
 264 static void ivshmem_event(void *opaque, int event)
 265 {
 266     IVSHMEM_DPRINTF("ivshmem_event %d\n", event);
 267 }
 268
 269 static void fake_irqfd(void *opaque, const uint8_t *buf, int size) {
 270
 271     EventfdEntry *entry = opaque;
 272     PCIDevice *pdev = entry->pdev;
 273
 274     IVSHMEM_DPRINTF("interrupt on vector %p %d\n", pdev, entry->vector);
 275     msix_notify(pdev, entry->vector);
 276 }
 277
 278 static CharDriverState* create_eventfd_chr_device(void * opaque, EventNotifier *n,
 279                                                   int vector)
 280 {
 281     /* create a event character device based on the passed eventfd */
 282     IVShmemState *s = opaque;
 283     CharDriverState * chr;
 284     int eventfd = event_notifier_get_fd(n);
 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, s->ivshmem_attr, &s->bar);
 345 }
 346
 347 static void ivshmem_add_eventfd(IVShmemState *s, int posn, int i)
 348 {
 349     memory_region_add_eventfd(&s->ivshmem_mmio,
 350                               DOORBELL,
 351                               4,
 352                               true,
 353                               (posn << 16) | i,
 354                               &s->peers[posn].eventfds[i]);
 355 }
 356
 357 static void ivshmem_del_eventfd(IVShmemState *s, int posn, int i)
 358 {
 359     memory_region_del_eventfd(&s->ivshmem_mmio,
 360                               DOORBELL,
 361                               4,
 362                               true,
 363                               (posn << 16) | i,
 364                               &s->peers[posn].eventfds[i]);
 365 }
 366
 367 static void close_guest_eventfds(IVShmemState *s, int posn)
 368 {
 369     int i, guest_curr_max;
 370
 371     if (!ivshmem_has_feature(s, IVSHMEM_IOEVENTFD)) {
 372         return;
 373     }
 374
 375     guest_curr_max = s->peers[posn].nb_eventfds;
 376
 377     memory_region_transaction_begin();
 378     for (i = 0; i < guest_curr_max; i++) {
 379         ivshmem_del_eventfd(s, posn, i);
 380     }
 381     memory_region_transaction_commit();
 382     for (i = 0; i < guest_curr_max; i++) {
 383         event_notifier_cleanup(&s->peers[posn].eventfds[i]);
 384     }
 385
 386     g_free(s->peers[posn].eventfds);
 387     s->peers[posn].nb_eventfds = 0;
 388 }
 389
 390 /* this function increase the dynamic storage need to store data about other
 391  * guests */
 392 static void increase_dynamic_storage(IVShmemState *s, int new_min_size) {
 393
 394     int j, old_nb_alloc;
 395
 396     old_nb_alloc = s->nb_peers;
 397
 398     while (new_min_size >= s->nb_peers)
 399         s->nb_peers = s->nb_peers * 2;
 400
 401     IVSHMEM_DPRINTF("bumping storage to %d guests\n", s->nb_peers);
 402     s->peers = g_realloc(s->peers, s->nb_peers * sizeof(Peer));
 403
 404     /* zero out new pointers */
 405     for (j = old_nb_alloc; j < s->nb_peers; j++) {
 406         s->peers[j].eventfds = NULL;
 407         s->peers[j].nb_eventfds = 0;
 408     }
 409 }
 410
 411 static void ivshmem_read(void *opaque, const uint8_t * buf, int flags)
 412 {
 413     IVShmemState *s = opaque;
 414     int incoming_fd, tmp_fd;
 415     int guest_max_eventfd;
 416     long incoming_posn;
 417
 418     memcpy(&incoming_posn, buf, sizeof(long));
 419     /* pick off s->server_chr->msgfd and store it, posn should accompany msg */
 420     tmp_fd = qemu_chr_fe_get_msgfd(s->server_chr);
 421     IVSHMEM_DPRINTF("posn is %ld, fd is %d\n", incoming_posn, tmp_fd);
 422
 423     /* make sure we have enough space for this guest */
 424     if (incoming_posn >= s->nb_peers) {
 425         increase_dynamic_storage(s, incoming_posn);
 426     }
 427
 428     if (tmp_fd == -1) {
 429         /* if posn is positive and unseen before then this is our posn*/
 430         if ((incoming_posn >= 0) &&
 431                             (s->peers[incoming_posn].eventfds == NULL)) {
 432             /* receive our posn */
 433             s->vm_id = incoming_posn;
 434             return;
 435         } else {
 436             /* otherwise an fd == -1 means an existing guest has gone away */
 437             IVSHMEM_DPRINTF("posn %ld has gone away\n", incoming_posn);
 438             close_guest_eventfds(s, incoming_posn);
 439             return;
 440         }
 441     }
 442
 443     /* because of the implementation of get_msgfd, we need a dup */
 444     incoming_fd = dup(tmp_fd);
 445
 446     if (incoming_fd == -1) {
 447         fprintf(stderr, "could not allocate file descriptor %s\n",
 448                                                             strerror(errno));
 449         return;
 450     }
 451
 452     /* if the position is -1, then it's shared memory region fd */
 453     if (incoming_posn == -1) {
 454
 455         void * map_ptr;
 456
 457         s->max_peer = 0;
 458
 459         if (check_shm_size(s, incoming_fd) == -1) {
 460             exit(-1);
 461         }
 462
 463         /* mmap the region and map into the BAR2 */
 464         map_ptr = mmap(0, s->ivshmem_size, PROT_READ|PROT_WRITE, MAP_SHARED,
 465                                                             incoming_fd, 0);
 466         memory_region_init_ram_ptr(&s->ivshmem,
 467                                    "ivshmem.bar2", s->ivshmem_size, map_ptr);
 468         vmstate_register_ram(&s->ivshmem, &s->dev.qdev);
 469
 470         IVSHMEM_DPRINTF("guest h/w addr = %" PRIu64 ", size = %" PRIu64 "\n",
 471                          s->ivshmem_offset, s->ivshmem_size);
 472
 473         memory_region_add_subregion(&s->bar, 0, &s->ivshmem);
 474
 475         /* only store the fd if it is successfully mapped */
 476         s->shm_fd = incoming_fd;
 477
 478         return;
 479     }
 480
 481     /* each guest has an array of eventfds, and we keep track of how many
 482      * guests for each VM */
 483     guest_max_eventfd = s->peers[incoming_posn].nb_eventfds;
 484
 485     if (guest_max_eventfd == 0) {
 486         /* one eventfd per MSI vector */
 487         s->peers[incoming_posn].eventfds = g_new(EventNotifier, s->vectors);
 488     }
 489
 490     /* this is an eventfd for a particular guest VM */
 491     IVSHMEM_DPRINTF("eventfds[%ld][%d] = %d\n", incoming_posn,
 492                                             guest_max_eventfd, incoming_fd);
 493     event_notifier_init_fd(&s->peers[incoming_posn].eventfds[guest_max_eventfd],
 494                            incoming_fd);
 495
 496     /* increment count for particular guest */
 497     s->peers[incoming_posn].nb_eventfds++;
 498
 499     /* keep track of the maximum VM ID */
 500     if (incoming_posn > s->max_peer) {
 501         s->max_peer = incoming_posn;
 502     }
 503
 504     if (incoming_posn == s->vm_id) {
 505         s->eventfd_chr[guest_max_eventfd] = create_eventfd_chr_device(s,
 506                    &s->peers[s->vm_id].eventfds[guest_max_eventfd],
 507                    guest_max_eventfd);
 508     }
 509
 510     if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD)) {
 511         ivshmem_add_eventfd(s, incoming_posn, guest_max_eventfd);
 512     }
 513
 514     return;
 515 }
 516
 517 /* Select the MSI-X vectors used by device.
 518  * ivshmem maps events to vectors statically, so
 519  * we just enable all vectors on init and after reset. */
 520 static void ivshmem_use_msix(IVShmemState * s)
 521 {
 522     int i;
 523
 524     if (!msix_present(&s->dev)) {
 525         return;
 526     }
 527
 528     for (i = 0; i < s->vectors; i++) {
 529         msix_vector_use(&s->dev, i);
 530     }
 531 }
 532
 533 static void ivshmem_reset(DeviceState *d)
 534 {
 535     IVShmemState *s = DO_UPCAST(IVShmemState, dev.qdev, d);
 536
 537     s->intrstatus = 0;
 538     ivshmem_use_msix(s);
 539     return;
 540 }
 541
 542 static uint64_t ivshmem_get_size(IVShmemState * s) {
 543
 544     uint64_t value;
 545     char *ptr;
 546
 547     value = strtoull(s->sizearg, &ptr, 10);
 548     switch (*ptr) {
 549         case 0: case 'M': case 'm':
 550             value <<= 20;
 551             break;
 552         case 'G': case 'g':
 553             value <<= 30;
 554             break;
 555         default:
 556             fprintf(stderr, "qemu: invalid ram size: %s\n", s->sizearg);
 557             exit(1);
 558     }
 559
 560     /* BARs must be a power of 2 */
 561     if (!is_power_of_two(value)) {
 562         fprintf(stderr, "ivshmem: size must be power of 2\n");
 563         exit(1);
 564     }
 565
 566     return value;
 567 }
 568
 569 static void ivshmem_setup_msi(IVShmemState * s)
 570 {
 571     if (msix_init_exclusive_bar(&s->dev, s->vectors, 1)) {
 572         IVSHMEM_DPRINTF("msix initialization failed\n");
 573         exit(1);
 574     }
 575
 576     IVSHMEM_DPRINTF("msix initialized (%d vectors)\n", s->vectors);
 577
 578     /* allocate QEMU char devices for receiving interrupts */
 579     s->eventfd_table = g_malloc0(s->vectors * sizeof(EventfdEntry));
 580
 581     ivshmem_use_msix(s);
 582 }
 583
 584 static void ivshmem_save(QEMUFile* f, void *opaque)
 585 {
 586     IVShmemState *proxy = opaque;
 587
 588     IVSHMEM_DPRINTF("ivshmem_save\n");
 589     pci_device_save(&proxy->dev, f);
 590
 591     if (ivshmem_has_feature(proxy, IVSHMEM_MSI)) {
 592         msix_save(&proxy->dev, f);
 593     } else {
 594         qemu_put_be32(f, proxy->intrstatus);
 595         qemu_put_be32(f, proxy->intrmask);
 596     }
 597
 598 }
 599
 600 static int ivshmem_load(QEMUFile* f, void *opaque, int version_id)
 601 {
 602     IVSHMEM_DPRINTF("ivshmem_load\n");
 603
 604     IVShmemState *proxy = opaque;
 605     int ret;
 606
 607     if (version_id > 0) {
 608         return -EINVAL;
 609     }
 610
 611     if (proxy->role_val == IVSHMEM_PEER) {
 612         fprintf(stderr, "ivshmem: 'peer' devices are not migratable\n");
 613         return -EINVAL;
 614     }
 615
 616     ret = pci_device_load(&proxy->dev, f);
 617     if (ret) {
 618         return ret;
 619     }
 620
 621     if (ivshmem_has_feature(proxy, IVSHMEM_MSI)) {
 622         msix_load(&proxy->dev, f);
 623         ivshmem_use_msix(proxy);
 624     } else {
 625         proxy->intrstatus = qemu_get_be32(f);
 626         proxy->intrmask = qemu_get_be32(f);
 627     }
 628
 629     return 0;
 630 }
 631
 632 static void ivshmem_write_config(PCIDevice *pci_dev, uint32_t address,
 633                                  uint32_t val, int len)
 634 {
 635     pci_default_write_config(pci_dev, address, val, len);
 636     msix_write_config(pci_dev, address, val, len);
 637 }
 638
 639 static int pci_ivshmem_init(PCIDevice *dev)
 640 {
 641     IVShmemState *s = DO_UPCAST(IVShmemState, dev, dev);
 642     uint8_t *pci_conf;
 643
 644     if (s->sizearg == NULL)
 645         s->ivshmem_size = 4 << 20; /* 4 MB default */
 646     else {
 647         s->ivshmem_size = ivshmem_get_size(s);
 648     }
 649
 650     register_savevm(&s->dev.qdev, "ivshmem", 0, 0, ivshmem_save, ivshmem_load,
 651                                                                         dev);
 652
 653     /* IRQFD requires MSI */
 654     if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD) &&
 655         !ivshmem_has_feature(s, IVSHMEM_MSI)) {
 656         fprintf(stderr, "ivshmem: ioeventfd/irqfd requires MSI\n");
 657         exit(1);
 658     }
 659
 660     /* check that role is reasonable */
 661     if (s->role) {
 662         if (strncmp(s->role, "peer", 5) == 0) {
 663             s->role_val = IVSHMEM_PEER;
 664         } else if (strncmp(s->role, "master", 7) == 0) {
 665             s->role_val = IVSHMEM_MASTER;
 666         } else {
 667             fprintf(stderr, "ivshmem: 'role' must be 'peer' or 'master'\n");
 668             exit(1);
 669         }
 670     } else {
 671         s->role_val = IVSHMEM_MASTER; /* default */
 672     }
 673
 674     if (s->role_val == IVSHMEM_PEER) {
 675         error_set(&s->migration_blocker, QERR_DEVICE_FEATURE_BLOCKS_MIGRATION,
 676                   "peer mode", "ivshmem");
 677         migrate_add_blocker(s->migration_blocker);
 678     }
 679
 680     pci_conf = s->dev.config;
 681     pci_conf[PCI_COMMAND] = PCI_COMMAND_IO | PCI_COMMAND_MEMORY;
 682
 683     pci_config_set_interrupt_pin(pci_conf, 1);
 684
 685     s->shm_fd = 0;
 686
 687     memory_region_init_io(&s->ivshmem_mmio, &ivshmem_mmio_ops, s,
 688                           "ivshmem-mmio", IVSHMEM_REG_BAR_SIZE);
 689
 690     /* region for registers*/
 691     pci_register_bar(&s->dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY,
 692                      &s->ivshmem_mmio);
 693
 694     memory_region_init(&s->bar, "ivshmem-bar2-container", s->ivshmem_size);
 695     s->ivshmem_attr = PCI_BASE_ADDRESS_SPACE_MEMORY |
 696         PCI_BASE_ADDRESS_MEM_PREFETCH;
 697     if (s->ivshmem_64bit) {
 698         s->ivshmem_attr |= PCI_BASE_ADDRESS_MEM_TYPE_64;
 699     }
 700
 701     if ((s->server_chr != NULL) &&
 702                         (strncmp(s->server_chr->filename, "unix:", 5) == 0)) {
 703         /* if we get a UNIX socket as the parameter we will talk
 704          * to the ivshmem server to receive the memory region */
 705
 706         if (s->shmobj != NULL) {
 707             fprintf(stderr, "WARNING: do not specify both 'chardev' "
 708                                                 "and 'shm' with ivshmem\n");
 709         }
 710
 711         IVSHMEM_DPRINTF("using shared memory server (socket = %s)\n",
 712                                                     s->server_chr->filename);
 713
 714         if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
 715             ivshmem_setup_msi(s);
 716         }
 717
 718         /* we allocate enough space for 16 guests and grow as needed */
 719         s->nb_peers = 16;
 720         s->vm_id = -1;
 721
 722         /* allocate/initialize space for interrupt handling */
 723         s->peers = g_malloc0(s->nb_peers * sizeof(Peer));
 724
 725         pci_register_bar(&s->dev, 2, s->ivshmem_attr, &s->bar);
 726
 727         s->eventfd_chr = g_malloc0(s->vectors * sizeof(CharDriverState *));
 728
 729         qemu_chr_add_handlers(s->server_chr, ivshmem_can_receive, ivshmem_read,
 730                      ivshmem_event, s);
 731     } else {
 732         /* just map the file immediately, we're not using a server */
 733         int fd;
 734
 735         if (s->shmobj == NULL) {
 736             fprintf(stderr, "Must specify 'chardev' or 'shm' to ivshmem\n");
 737         }
 738
 739         IVSHMEM_DPRINTF("using shm_open (shm object = %s)\n", s->shmobj);
 740
 741         /* try opening with O_EXCL and if it succeeds zero the memory
 742          * by truncating to 0 */
 743         if ((fd = shm_open(s->shmobj, O_CREAT|O_RDWR|O_EXCL,
 744                         S_IRWXU|S_IRWXG|S_IRWXO)) > 0) {
 745            /* truncate file to length PCI device's memory */
 746             if (ftruncate(fd, s->ivshmem_size) != 0) {
 747                 fprintf(stderr, "ivshmem: could not truncate shared file\n");
 748             }
 749
 750         } else if ((fd = shm_open(s->shmobj, O_CREAT|O_RDWR,
 751                         S_IRWXU|S_IRWXG|S_IRWXO)) < 0) {
 752             fprintf(stderr, "ivshmem: could not open shared file\n");
 753             exit(-1);
 754
 755         }
 756
 757         if (check_shm_size(s, fd) == -1) {
 758             exit(-1);
 759         }
 760
 761         create_shared_memory_BAR(s, fd);
 762
 763     }
 764
 765     s->dev.config_write = ivshmem_write_config;
 766
 767     return 0;
 768 }
 769
 770 static void pci_ivshmem_uninit(PCIDevice *dev)
 771 {
 772     IVShmemState *s = DO_UPCAST(IVShmemState, dev, dev);
 773
 774     if (s->migration_blocker) {
 775         migrate_del_blocker(s->migration_blocker);
 776         error_free(s->migration_blocker);
 777     }
 778
 779     memory_region_destroy(&s->ivshmem_mmio);
 780     memory_region_del_subregion(&s->bar, &s->ivshmem);
 781     vmstate_unregister_ram(&s->ivshmem, &s->dev.qdev);
 782     memory_region_destroy(&s->ivshmem);
 783     memory_region_destroy(&s->bar);
 784     unregister_savevm(&dev->qdev, "ivshmem", s);
 785 }
 786
 787 static Property ivshmem_properties[] = {
 788     DEFINE_PROP_CHR("chardev", IVShmemState, server_chr),
 789     DEFINE_PROP_STRING("size", IVShmemState, sizearg),
 790     DEFINE_PROP_UINT32("vectors", IVShmemState, vectors, 1),
 791     DEFINE_PROP_BIT("ioeventfd", IVShmemState, features, IVSHMEM_IOEVENTFD, false),
 792     DEFINE_PROP_BIT("msi", IVShmemState, features, IVSHMEM_MSI, true),
 793     DEFINE_PROP_STRING("shm", IVShmemState, shmobj),
 794     DEFINE_PROP_STRING("role", IVShmemState, role),
 795     DEFINE_PROP_UINT32("use64", IVShmemState, ivshmem_64bit, 1),
 796     DEFINE_PROP_END_OF_LIST(),
 797 };
 798
 799 static void ivshmem_class_init(ObjectClass *klass, void *data)
 800 {
 801     DeviceClass *dc = DEVICE_CLASS(klass);
 802     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
 803
 804     k->init = pci_ivshmem_init;
 805     k->exit = pci_ivshmem_uninit;
 806     k->vendor_id = PCI_VENDOR_ID_REDHAT_QUMRANET;
 807     k->device_id = 0x1110;
 808     k->class_id = PCI_CLASS_MEMORY_RAM;
 809     dc->reset = ivshmem_reset;
 810     dc->props = ivshmem_properties;
 811 }
 812
 813 static TypeInfo ivshmem_info = {
 814     .name          = "ivshmem",
 815     .parent        = TYPE_PCI_DEVICE,
 816     .instance_size = sizeof(IVShmemState),
 817     .class_init    = ivshmem_class_init,
 818 };
 819
 820 static void ivshmem_register_types(void)
 821 {
 822     type_register_static(&ivshmem_info);
 823 }
 824
 825 type_init(ivshmem_register_types)