hw/misc/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/hw.h"
  20 #include "hw/i386/pc.h"
  21 #include "hw/pci/pci.h"
  22 #include "hw/pci/msix.h"
  23 #include "sysemu/kvm.h"
  24 #include "migration/migration.h"
  25 #include "qemu/error-report.h"
  26 #include "qemu/event_notifier.h"
  27 #include "qemu/fifo8.h"
  28 #include "sysemu/char.h"
  29
  30 #include <sys/mman.h>
  31 #include <sys/types.h>
  32 #include <limits.h>
  33
  34 #define PCI_VENDOR_ID_IVSHMEM   PCI_VENDOR_ID_REDHAT_QUMRANET
  35 #define PCI_DEVICE_ID_IVSHMEM   0x1110
  36
  37 #define IVSHMEM_IOEVENTFD   0
  38 #define IVSHMEM_MSI     1
  39
  40 #define IVSHMEM_PEER    0
  41 #define IVSHMEM_MASTER  1
  42
  43 #define IVSHMEM_REG_BAR_SIZE 0x100
  44
  45 //#define DEBUG_IVSHMEM
  46 #ifdef DEBUG_IVSHMEM
  47 #define IVSHMEM_DPRINTF(fmt, ...)        \
  48     do {printf("IVSHMEM: " fmt, ## __VA_ARGS__); } while (0)
  49 #else
  50 #define IVSHMEM_DPRINTF(fmt, ...)
  51 #endif
  52
  53 #define TYPE_IVSHMEM "ivshmem"
  54 #define IVSHMEM(obj) \
  55     OBJECT_CHECK(IVShmemState, (obj), TYPE_IVSHMEM)
  56
  57 typedef struct Peer {
  58     int nb_eventfds;
  59     EventNotifier *eventfds;
  60 } Peer;
  61
  62 typedef struct EventfdEntry {
  63     PCIDevice *pdev;
  64     int vector;
  65 } EventfdEntry;
  66
  67 typedef struct IVShmemState {
  68     /*< private >*/
  69     PCIDevice parent_obj;
  70     /*< public >*/
  71
  72     uint32_t intrmask;
  73     uint32_t intrstatus;
  74     uint32_t doorbell;
  75
  76     CharDriverState **eventfd_chr;
  77     CharDriverState *server_chr;
  78     Fifo8 incoming_fifo;
  79     MemoryRegion ivshmem_mmio;
  80
  81     /* We might need to register the BAR before we actually have the memory.
  82      * So prepare a container MemoryRegion for the BAR immediately and
  83      * add a subregion when we have the memory.
  84      */
  85     MemoryRegion bar;
  86     MemoryRegion ivshmem;
  87     uint64_t ivshmem_size; /* size of shared memory region */
  88     uint32_t ivshmem_attr;
  89     uint32_t ivshmem_64bit;
  90     int shm_fd; /* shared memory file descriptor */
  91
  92     Peer *peers;
  93     int nb_peers; /* how many guests we have space for */
  94     int max_peer; /* maximum numbered peer */
  95
  96     int vm_id;
  97     uint32_t vectors;
  98     uint32_t features;
  99     EventfdEntry *eventfd_table;
 100
 101     Error *migration_blocker;
 102
 103     char * shmobj;
 104     char * sizearg;
 105     char * role;
 106     int role_val;   /* scalar to avoid multiple string comparisons */
 107 } IVShmemState;
 108
 109 /* registers for the Inter-VM shared memory device */
 110 enum ivshmem_registers {
 111     INTRMASK = 0,
 112     INTRSTATUS = 4,
 113     IVPOSITION = 8,
 114     DOORBELL = 12,
 115 };
 116
 117 static inline uint32_t ivshmem_has_feature(IVShmemState *ivs,
 118                                                     unsigned int feature) {
 119     return (ivs->features & (1 << feature));
 120 }
 121
 122 static inline bool is_power_of_two(uint64_t x) {
 123     return (x & (x - 1)) == 0;
 124 }
 125
 126 /* accessing registers - based on rtl8139 */
 127 static void ivshmem_update_irq(IVShmemState *s, int val)
 128 {
 129     PCIDevice *d = PCI_DEVICE(s);
 130     int isr;
 131     isr = (s->intrstatus & s->intrmask) & 0xffffffff;
 132
 133     /* don't print ISR resets */
 134     if (isr) {
 135         IVSHMEM_DPRINTF("Set IRQ to %d (%04x %04x)\n",
 136                         isr ? 1 : 0, s->intrstatus, s->intrmask);
 137     }
 138
 139     pci_set_irq(d, (isr != 0));
 140 }
 141
 142 static void ivshmem_IntrMask_write(IVShmemState *s, uint32_t val)
 143 {
 144     IVSHMEM_DPRINTF("IntrMask write(w) val = 0x%04x\n", val);
 145
 146     s->intrmask = val;
 147
 148     ivshmem_update_irq(s, val);
 149 }
 150
 151 static uint32_t ivshmem_IntrMask_read(IVShmemState *s)
 152 {
 153     uint32_t ret = s->intrmask;
 154
 155     IVSHMEM_DPRINTF("intrmask read(w) val = 0x%04x\n", ret);
 156
 157     return ret;
 158 }
 159
 160 static void ivshmem_IntrStatus_write(IVShmemState *s, uint32_t val)
 161 {
 162     IVSHMEM_DPRINTF("IntrStatus write(w) val = 0x%04x\n", val);
 163
 164     s->intrstatus = val;
 165
 166     ivshmem_update_irq(s, val);
 167 }
 168
 169 static uint32_t ivshmem_IntrStatus_read(IVShmemState *s)
 170 {
 171     uint32_t ret = s->intrstatus;
 172
 173     /* reading ISR clears all interrupts */
 174     s->intrstatus = 0;
 175
 176     ivshmem_update_irq(s, 0);
 177
 178     return ret;
 179 }
 180
 181 static void ivshmem_io_write(void *opaque, hwaddr addr,
 182                              uint64_t val, unsigned size)
 183 {
 184     IVShmemState *s = opaque;
 185
 186     uint16_t dest = val >> 16;
 187     uint16_t vector = val & 0xff;
 188
 189     addr &= 0xfc;
 190
 191     IVSHMEM_DPRINTF("writing to addr " TARGET_FMT_plx "\n", addr);
 192     switch (addr)
 193     {
 194         case INTRMASK:
 195             ivshmem_IntrMask_write(s, val);
 196             break;
 197
 198         case INTRSTATUS:
 199             ivshmem_IntrStatus_write(s, val);
 200             break;
 201
 202         case DOORBELL:
 203             /* check that dest VM ID is reasonable */
 204             if (dest > s->max_peer) {
 205                 IVSHMEM_DPRINTF("Invalid destination VM ID (%d)\n", dest);
 206                 break;
 207             }
 208
 209             /* check doorbell range */
 210             if (vector < s->peers[dest].nb_eventfds) {
 211                 IVSHMEM_DPRINTF("Notifying VM %d on vector %d\n", dest, vector);
 212                 event_notifier_set(&s->peers[dest].eventfds[vector]);
 213             }
 214             break;
 215         default:
 216             IVSHMEM_DPRINTF("Invalid VM Doorbell VM %d\n", dest);
 217     }
 218 }
 219
 220 static uint64_t ivshmem_io_read(void *opaque, hwaddr addr,
 221                                 unsigned size)
 222 {
 223
 224     IVShmemState *s = opaque;
 225     uint32_t ret;
 226
 227     switch (addr)
 228     {
 229         case INTRMASK:
 230             ret = ivshmem_IntrMask_read(s);
 231             break;
 232
 233         case INTRSTATUS:
 234             ret = ivshmem_IntrStatus_read(s);
 235             break;
 236
 237         case IVPOSITION:
 238             /* return my VM ID if the memory is mapped */
 239             if (s->shm_fd > 0) {
 240                 ret = s->vm_id;
 241             } else {
 242                 ret = -1;
 243             }
 244             break;
 245
 246         default:
 247             IVSHMEM_DPRINTF("why are we reading " TARGET_FMT_plx "\n", addr);
 248             ret = 0;
 249     }
 250
 251     return ret;
 252 }
 253
 254 static const MemoryRegionOps ivshmem_mmio_ops = {
 255     .read = ivshmem_io_read,
 256     .write = ivshmem_io_write,
 257     .endianness = DEVICE_NATIVE_ENDIAN,
 258     .impl = {
 259         .min_access_size = 4,
 260         .max_access_size = 4,
 261     },
 262 };
 263
 264 static void ivshmem_receive(void *opaque, const uint8_t *buf, int size)
 265 {
 266     IVShmemState *s = opaque;
 267
 268     ivshmem_IntrStatus_write(s, *buf);
 269
 270     IVSHMEM_DPRINTF("ivshmem_receive 0x%02x\n", *buf);
 271 }
 272
 273 static int ivshmem_can_receive(void * opaque)
 274 {
 275     return 8;
 276 }
 277
 278 static void ivshmem_event(void *opaque, int event)
 279 {
 280     IVSHMEM_DPRINTF("ivshmem_event %d\n", event);
 281 }
 282
 283 static void fake_irqfd(void *opaque, const uint8_t *buf, int size) {
 284
 285     EventfdEntry *entry = opaque;
 286     PCIDevice *pdev = entry->pdev;
 287
 288     IVSHMEM_DPRINTF("interrupt on vector %p %d\n", pdev, entry->vector);
 289     msix_notify(pdev, entry->vector);
 290 }
 291
 292 static CharDriverState* create_eventfd_chr_device(void * opaque, EventNotifier *n,
 293                                                   int vector)
 294 {
 295     /* create a event character device based on the passed eventfd */
 296     IVShmemState *s = opaque;
 297     CharDriverState * chr;
 298     int eventfd = event_notifier_get_fd(n);
 299
 300     chr = qemu_chr_open_eventfd(eventfd);
 301
 302     if (chr == NULL) {
 303         error_report("creating eventfd for eventfd %d failed", eventfd);
 304         exit(1);
 305     }
 306     qemu_chr_fe_claim_no_fail(chr);
 307
 308     /* if MSI is supported we need multiple interrupts */
 309     if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
 310         s->eventfd_table[vector].pdev = PCI_DEVICE(s);
 311         s->eventfd_table[vector].vector = vector;
 312
 313         qemu_chr_add_handlers(chr, ivshmem_can_receive, fake_irqfd,
 314                       ivshmem_event, &s->eventfd_table[vector]);
 315     } else {
 316         qemu_chr_add_handlers(chr, ivshmem_can_receive, ivshmem_receive,
 317                       ivshmem_event, s);
 318     }
 319
 320     return chr;
 321
 322 }
 323
 324 static int check_shm_size(IVShmemState *s, int fd) {
 325     /* check that the guest isn't going to try and map more memory than the
 326      * the object has allocated return -1 to indicate error */
 327
 328     struct stat buf;
 329
 330     if (fstat(fd, &buf) < 0) {
 331         error_report("exiting: fstat on fd %d failed: %s",
 332                      fd, strerror(errno));
 333         return -1;
 334     }
 335
 336     if (s->ivshmem_size > buf.st_size) {
 337         error_report("Requested memory size greater"
 338                      " than shared object size (%" PRIu64 " > %" PRIu64")",
 339                      s->ivshmem_size, (uint64_t)buf.st_size);
 340         return -1;
 341     } else {
 342         return 0;
 343     }
 344 }
 345
 346 /* create the shared memory BAR when we are not using the server, so we can
 347  * create the BAR and map the memory immediately */
 348 static void create_shared_memory_BAR(IVShmemState *s, int fd) {
 349
 350     void * ptr;
 351
 352     s->shm_fd = fd;
 353
 354     ptr = mmap(0, s->ivshmem_size, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
 355
 356     memory_region_init_ram_ptr(&s->ivshmem, OBJECT(s), "ivshmem.bar2",
 357                                s->ivshmem_size, ptr);
 358     vmstate_register_ram(&s->ivshmem, DEVICE(s));
 359     memory_region_add_subregion(&s->bar, 0, &s->ivshmem);
 360
 361     /* region for shared memory */
 362     pci_register_bar(PCI_DEVICE(s), 2, s->ivshmem_attr, &s->bar);
 363 }
 364
 365 static void ivshmem_add_eventfd(IVShmemState *s, int posn, int i)
 366 {
 367     memory_region_add_eventfd(&s->ivshmem_mmio,
 368                               DOORBELL,
 369                               4,
 370                               true,
 371                               (posn << 16) | i,
 372                               &s->peers[posn].eventfds[i]);
 373 }
 374
 375 static void ivshmem_del_eventfd(IVShmemState *s, int posn, int i)
 376 {
 377     memory_region_del_eventfd(&s->ivshmem_mmio,
 378                               DOORBELL,
 379                               4,
 380                               true,
 381                               (posn << 16) | i,
 382                               &s->peers[posn].eventfds[i]);
 383 }
 384
 385 static void close_guest_eventfds(IVShmemState *s, int posn)
 386 {
 387     int i, guest_curr_max;
 388
 389     if (!ivshmem_has_feature(s, IVSHMEM_IOEVENTFD)) {
 390         return;
 391     }
 392     if (posn < 0 || posn >= s->nb_peers) {
 393         return;
 394     }
 395
 396     guest_curr_max = s->peers[posn].nb_eventfds;
 397
 398     memory_region_transaction_begin();
 399     for (i = 0; i < guest_curr_max; i++) {
 400         ivshmem_del_eventfd(s, posn, i);
 401     }
 402     memory_region_transaction_commit();
 403     for (i = 0; i < guest_curr_max; i++) {
 404         event_notifier_cleanup(&s->peers[posn].eventfds[i]);
 405     }
 406
 407     g_free(s->peers[posn].eventfds);
 408     s->peers[posn].nb_eventfds = 0;
 409 }
 410
 411 /* this function increase the dynamic storage need to store data about other
 412  * guests */
 413 static int increase_dynamic_storage(IVShmemState *s, int new_min_size)
 414 {
 415
 416     int j, old_nb_alloc;
 417
 418     /* check for integer overflow */
 419     if (new_min_size >= INT_MAX / sizeof(Peer) - 1 || new_min_size <= 0) {
 420         return -1;
 421     }
 422
 423     old_nb_alloc = s->nb_peers;
 424
 425     if (new_min_size >= s->nb_peers) {
 426         /* +1 because #new_min_size is used as last array index */
 427         s->nb_peers = new_min_size + 1;
 428     } else {
 429         return 0;
 430     }
 431
 432     IVSHMEM_DPRINTF("bumping storage to %d guests\n", s->nb_peers);
 433     s->peers = g_realloc(s->peers, s->nb_peers * sizeof(Peer));
 434
 435     /* zero out new pointers */
 436     for (j = old_nb_alloc; j < s->nb_peers; j++) {
 437         s->peers[j].eventfds = NULL;
 438         s->peers[j].nb_eventfds = 0;
 439     }
 440
 441     return 0;
 442 }
 443
 444 static void ivshmem_read(void *opaque, const uint8_t *buf, int size)
 445 {
 446     IVShmemState *s = opaque;
 447     int incoming_fd, tmp_fd;
 448     int guest_max_eventfd;
 449     long incoming_posn;
 450
 451     if (fifo8_is_empty(&s->incoming_fifo) && size == sizeof(incoming_posn)) {
 452         memcpy(&incoming_posn, buf, size);
 453     } else {
 454         const uint8_t *p;
 455         uint32_t num;
 456
 457         IVSHMEM_DPRINTF("short read of %d bytes\n", size);
 458         num = MAX(size, sizeof(long) - fifo8_num_used(&s->incoming_fifo));
 459         fifo8_push_all(&s->incoming_fifo, buf, num);
 460         if (fifo8_num_used(&s->incoming_fifo) < sizeof(incoming_posn)) {
 461             return;
 462         }
 463         size -= num;
 464         buf += num;
 465         p = fifo8_pop_buf(&s->incoming_fifo, sizeof(incoming_posn), &num);
 466         g_assert(num == sizeof(incoming_posn));
 467         memcpy(&incoming_posn, p, sizeof(incoming_posn));
 468         if (size > 0) {
 469             fifo8_push_all(&s->incoming_fifo, buf, size);
 470         }
 471     }
 472
 473     if (incoming_posn < -1) {
 474         IVSHMEM_DPRINTF("invalid incoming_posn %ld\n", incoming_posn);
 475         return;
 476     }
 477
 478     /* pick off s->server_chr->msgfd and store it, posn should accompany msg */
 479     tmp_fd = qemu_chr_fe_get_msgfd(s->server_chr);
 480     IVSHMEM_DPRINTF("posn is %ld, fd is %d\n", incoming_posn, tmp_fd);
 481
 482     /* make sure we have enough space for this guest */
 483     if (incoming_posn >= s->nb_peers) {
 484         if (increase_dynamic_storage(s, incoming_posn) < 0) {
 485             error_report("increase_dynamic_storage() failed");
 486             if (tmp_fd != -1) {
 487                 close(tmp_fd);
 488             }
 489             return;
 490         }
 491     }
 492
 493     if (tmp_fd == -1) {
 494         /* if posn is positive and unseen before then this is our posn*/
 495         if ((incoming_posn >= 0) &&
 496                             (s->peers[incoming_posn].eventfds == NULL)) {
 497             /* receive our posn */
 498             s->vm_id = incoming_posn;
 499             return;
 500         } else {
 501             /* otherwise an fd == -1 means an existing guest has gone away */
 502             IVSHMEM_DPRINTF("posn %ld has gone away\n", incoming_posn);
 503             close_guest_eventfds(s, incoming_posn);
 504             return;
 505         }
 506     }
 507
 508     /* because of the implementation of get_msgfd, we need a dup */
 509     incoming_fd = dup(tmp_fd);
 510
 511     if (incoming_fd == -1) {
 512         error_report("could not allocate file descriptor %s", strerror(errno));
 513         close(tmp_fd);
 514         return;
 515     }
 516
 517     /* if the position is -1, then it's shared memory region fd */
 518     if (incoming_posn == -1) {
 519
 520         void * map_ptr;
 521
 522         s->max_peer = 0;
 523
 524         if (check_shm_size(s, incoming_fd) == -1) {
 525             exit(1);
 526         }
 527
 528         /* mmap the region and map into the BAR2 */
 529         map_ptr = mmap(0, s->ivshmem_size, PROT_READ|PROT_WRITE, MAP_SHARED,
 530                                                             incoming_fd, 0);
 531         memory_region_init_ram_ptr(&s->ivshmem, OBJECT(s),
 532                                    "ivshmem.bar2", s->ivshmem_size, map_ptr);
 533         vmstate_register_ram(&s->ivshmem, DEVICE(s));
 534
 535         IVSHMEM_DPRINTF("guest h/w addr = %p, size = %" PRIu64 "\n",
 536                         map_ptr, s->ivshmem_size);
 537
 538         memory_region_add_subregion(&s->bar, 0, &s->ivshmem);
 539
 540         /* only store the fd if it is successfully mapped */
 541         s->shm_fd = incoming_fd;
 542
 543         return;
 544     }
 545
 546     /* each guest has an array of eventfds, and we keep track of how many
 547      * guests for each VM */
 548     guest_max_eventfd = s->peers[incoming_posn].nb_eventfds;
 549
 550     if (guest_max_eventfd == 0) {
 551         /* one eventfd per MSI vector */
 552         s->peers[incoming_posn].eventfds = g_new(EventNotifier, s->vectors);
 553     }
 554
 555     /* this is an eventfd for a particular guest VM */
 556     IVSHMEM_DPRINTF("eventfds[%ld][%d] = %d\n", incoming_posn,
 557                     guest_max_eventfd, incoming_fd);
 558     event_notifier_init_fd(&s->peers[incoming_posn].eventfds[guest_max_eventfd],
 559                            incoming_fd);
 560
 561     /* increment count for particular guest */
 562     s->peers[incoming_posn].nb_eventfds++;
 563
 564     /* keep track of the maximum VM ID */
 565     if (incoming_posn > s->max_peer) {
 566         s->max_peer = incoming_posn;
 567     }
 568
 569     if (incoming_posn == s->vm_id) {
 570         s->eventfd_chr[guest_max_eventfd] = create_eventfd_chr_device(s,
 571                    &s->peers[s->vm_id].eventfds[guest_max_eventfd],
 572                    guest_max_eventfd);
 573     }
 574
 575     if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD)) {
 576         ivshmem_add_eventfd(s, incoming_posn, guest_max_eventfd);
 577     }
 578 }
 579
 580 /* Select the MSI-X vectors used by device.
 581  * ivshmem maps events to vectors statically, so
 582  * we just enable all vectors on init and after reset. */
 583 static void ivshmem_use_msix(IVShmemState * s)
 584 {
 585     PCIDevice *d = PCI_DEVICE(s);
 586     int i;
 587
 588     if (!msix_present(d)) {
 589         return;
 590     }
 591
 592     for (i = 0; i < s->vectors; i++) {
 593         msix_vector_use(d, i);
 594     }
 595 }
 596
 597 static void ivshmem_reset(DeviceState *d)
 598 {
 599     IVShmemState *s = IVSHMEM(d);
 600
 601     s->intrstatus = 0;
 602     ivshmem_use_msix(s);
 603 }
 604
 605 static uint64_t ivshmem_get_size(IVShmemState * s) {
 606
 607     uint64_t value;
 608     char *ptr;
 609
 610     value = strtoull(s->sizearg, &ptr, 10);
 611     switch (*ptr) {
 612         case 0: case 'M': case 'm':
 613             value <<= 20;
 614             break;
 615         case 'G': case 'g':
 616             value <<= 30;
 617             break;
 618         default:
 619             error_report("invalid ram size: %s", s->sizearg);
 620             exit(1);
 621     }
 622
 623     /* BARs must be a power of 2 */
 624     if (!is_power_of_two(value)) {
 625         error_report("size must be power of 2");
 626         exit(1);
 627     }
 628
 629     return value;
 630 }
 631
 632 static void ivshmem_setup_msi(IVShmemState * s)
 633 {
 634     if (msix_init_exclusive_bar(PCI_DEVICE(s), s->vectors, 1)) {
 635         IVSHMEM_DPRINTF("msix initialization failed\n");
 636         exit(1);
 637     }
 638
 639     IVSHMEM_DPRINTF("msix initialized (%d vectors)\n", s->vectors);
 640
 641     /* allocate QEMU char devices for receiving interrupts */
 642     s->eventfd_table = g_malloc0(s->vectors * sizeof(EventfdEntry));
 643
 644     ivshmem_use_msix(s);
 645 }
 646
 647 static void ivshmem_save(QEMUFile* f, void *opaque)
 648 {
 649     IVShmemState *proxy = opaque;
 650     PCIDevice *pci_dev = PCI_DEVICE(proxy);
 651
 652     IVSHMEM_DPRINTF("ivshmem_save\n");
 653     pci_device_save(pci_dev, f);
 654
 655     if (ivshmem_has_feature(proxy, IVSHMEM_MSI)) {
 656         msix_save(pci_dev, f);
 657     } else {
 658         qemu_put_be32(f, proxy->intrstatus);
 659         qemu_put_be32(f, proxy->intrmask);
 660     }
 661
 662 }
 663
 664 static int ivshmem_load(QEMUFile* f, void *opaque, int version_id)
 665 {
 666     IVSHMEM_DPRINTF("ivshmem_load\n");
 667
 668     IVShmemState *proxy = opaque;
 669     PCIDevice *pci_dev = PCI_DEVICE(proxy);
 670     int ret;
 671
 672     if (version_id > 0) {
 673         return -EINVAL;
 674     }
 675
 676     if (proxy->role_val == IVSHMEM_PEER) {
 677         error_report("'peer' devices are not migratable");
 678         return -EINVAL;
 679     }
 680
 681     ret = pci_device_load(pci_dev, f);
 682     if (ret) {
 683         return ret;
 684     }
 685
 686     if (ivshmem_has_feature(proxy, IVSHMEM_MSI)) {
 687         msix_load(pci_dev, f);
 688         ivshmem_use_msix(proxy);
 689     } else {
 690         proxy->intrstatus = qemu_get_be32(f);
 691         proxy->intrmask = qemu_get_be32(f);
 692     }
 693
 694     return 0;
 695 }
 696
 697 static void ivshmem_write_config(PCIDevice *pci_dev, uint32_t address,
 698                                  uint32_t val, int len)
 699 {
 700     pci_default_write_config(pci_dev, address, val, len);
 701 }
 702
 703 static int pci_ivshmem_init(PCIDevice *dev)
 704 {
 705     IVShmemState *s = IVSHMEM(dev);
 706     uint8_t *pci_conf;
 707
 708     if (s->sizearg == NULL)
 709         s->ivshmem_size = 4 << 20; /* 4 MB default */
 710     else {
 711         s->ivshmem_size = ivshmem_get_size(s);
 712     }
 713
 714     fifo8_create(&s->incoming_fifo, sizeof(long));
 715
 716     register_savevm(DEVICE(dev), "ivshmem", 0, 0, ivshmem_save, ivshmem_load,
 717                                                                         dev);
 718
 719     /* IRQFD requires MSI */
 720     if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD) &&
 721         !ivshmem_has_feature(s, IVSHMEM_MSI)) {
 722         error_report("ioeventfd/irqfd requires MSI");
 723         exit(1);
 724     }
 725
 726     /* check that role is reasonable */
 727     if (s->role) {
 728         if (strncmp(s->role, "peer", 5) == 0) {
 729             s->role_val = IVSHMEM_PEER;
 730         } else if (strncmp(s->role, "master", 7) == 0) {
 731             s->role_val = IVSHMEM_MASTER;
 732         } else {
 733             error_report("'role' must be 'peer' or 'master'");
 734             exit(1);
 735         }
 736     } else {
 737         s->role_val = IVSHMEM_MASTER; /* default */
 738     }
 739
 740     if (s->role_val == IVSHMEM_PEER) {
 741         error_setg(&s->migration_blocker,
 742                    "Migration is disabled when using feature 'peer mode' in device 'ivshmem'");
 743         migrate_add_blocker(s->migration_blocker);
 744     }
 745
 746     pci_conf = dev->config;
 747     pci_conf[PCI_COMMAND] = PCI_COMMAND_IO | PCI_COMMAND_MEMORY;
 748
 749     pci_config_set_interrupt_pin(pci_conf, 1);
 750
 751     s->shm_fd = 0;
 752
 753     memory_region_init_io(&s->ivshmem_mmio, OBJECT(s), &ivshmem_mmio_ops, s,
 754                           "ivshmem-mmio", IVSHMEM_REG_BAR_SIZE);
 755
 756     /* region for registers*/
 757     pci_register_bar(dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY,
 758                      &s->ivshmem_mmio);
 759
 760     memory_region_init(&s->bar, OBJECT(s), "ivshmem-bar2-container", s->ivshmem_size);
 761     s->ivshmem_attr = PCI_BASE_ADDRESS_SPACE_MEMORY |
 762         PCI_BASE_ADDRESS_MEM_PREFETCH;
 763     if (s->ivshmem_64bit) {
 764         s->ivshmem_attr |= PCI_BASE_ADDRESS_MEM_TYPE_64;
 765     }
 766
 767     if ((s->server_chr != NULL) &&
 768                         (strncmp(s->server_chr->filename, "unix:", 5) == 0)) {
 769         /* if we get a UNIX socket as the parameter we will talk
 770          * to the ivshmem server to receive the memory region */
 771
 772         if (s->shmobj != NULL) {
 773             error_report("WARNING: do not specify both 'chardev' "
 774                          "and 'shm' with ivshmem");
 775         }
 776
 777         IVSHMEM_DPRINTF("using shared memory server (socket = %s)\n",
 778                         s->server_chr->filename);
 779
 780         if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
 781             ivshmem_setup_msi(s);
 782         }
 783
 784         /* we allocate enough space for 16 guests and grow as needed */
 785         s->nb_peers = 16;
 786         s->vm_id = -1;
 787
 788         /* allocate/initialize space for interrupt handling */
 789         s->peers = g_malloc0(s->nb_peers * sizeof(Peer));
 790
 791         pci_register_bar(dev, 2, s->ivshmem_attr, &s->bar);
 792
 793         s->eventfd_chr = g_malloc0(s->vectors * sizeof(CharDriverState *));
 794
 795         qemu_chr_add_handlers(s->server_chr, ivshmem_can_receive, ivshmem_read,
 796                      ivshmem_event, s);
 797     } else {
 798         /* just map the file immediately, we're not using a server */
 799         int fd;
 800
 801         if (s->shmobj == NULL) {
 802             error_report("Must specify 'chardev' or 'shm' to ivshmem");
 803             exit(1);
 804         }
 805
 806         IVSHMEM_DPRINTF("using shm_open (shm object = %s)\n", s->shmobj);
 807
 808         /* try opening with O_EXCL and if it succeeds zero the memory
 809          * by truncating to 0 */
 810         if ((fd = shm_open(s->shmobj, O_CREAT|O_RDWR|O_EXCL,
 811                         S_IRWXU|S_IRWXG|S_IRWXO)) > 0) {
 812            /* truncate file to length PCI device's memory */
 813             if (ftruncate(fd, s->ivshmem_size) != 0) {
 814                 error_report("could not truncate shared file");
 815             }
 816
 817         } else if ((fd = shm_open(s->shmobj, O_CREAT|O_RDWR,
 818                         S_IRWXU|S_IRWXG|S_IRWXO)) < 0) {
 819             error_report("could not open shared file");
 820             exit(1);
 821
 822         }
 823
 824         if (check_shm_size(s, fd) == -1) {
 825             exit(1);
 826         }
 827
 828         create_shared_memory_BAR(s, fd);
 829
 830     }
 831
 832     dev->config_write = ivshmem_write_config;
 833
 834     return 0;
 835 }
 836
 837 static void pci_ivshmem_uninit(PCIDevice *dev)
 838 {
 839     IVShmemState *s = IVSHMEM(dev);
 840
 841     if (s->migration_blocker) {
 842         migrate_del_blocker(s->migration_blocker);
 843         error_free(s->migration_blocker);
 844     }
 845
 846     memory_region_del_subregion(&s->bar, &s->ivshmem);
 847     vmstate_unregister_ram(&s->ivshmem, DEVICE(dev));
 848     unregister_savevm(DEVICE(dev), "ivshmem", s);
 849     fifo8_destroy(&s->incoming_fifo);
 850 }
 851
 852 static Property ivshmem_properties[] = {
 853     DEFINE_PROP_CHR("chardev", IVShmemState, server_chr),
 854     DEFINE_PROP_STRING("size", IVShmemState, sizearg),
 855     DEFINE_PROP_UINT32("vectors", IVShmemState, vectors, 1),
 856     DEFINE_PROP_BIT("ioeventfd", IVShmemState, features, IVSHMEM_IOEVENTFD, false),
 857     DEFINE_PROP_BIT("msi", IVShmemState, features, IVSHMEM_MSI, true),
 858     DEFINE_PROP_STRING("shm", IVShmemState, shmobj),
 859     DEFINE_PROP_STRING("role", IVShmemState, role),
 860     DEFINE_PROP_UINT32("use64", IVShmemState, ivshmem_64bit, 1),
 861     DEFINE_PROP_END_OF_LIST(),
 862 };
 863
 864 static void ivshmem_class_init(ObjectClass *klass, void *data)
 865 {
 866     DeviceClass *dc = DEVICE_CLASS(klass);
 867     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
 868
 869     k->init = pci_ivshmem_init;
 870     k->exit = pci_ivshmem_uninit;
 871     k->vendor_id = PCI_VENDOR_ID_IVSHMEM;
 872     k->device_id = PCI_DEVICE_ID_IVSHMEM;
 873     k->class_id = PCI_CLASS_MEMORY_RAM;
 874     dc->reset = ivshmem_reset;
 875     dc->props = ivshmem_properties;
 876     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
 877 }
 878
 879 static const TypeInfo ivshmem_info = {
 880     .name          = TYPE_IVSHMEM,
 881     .parent        = TYPE_PCI_DEVICE,
 882     .instance_size = sizeof(IVShmemState),
 883     .class_init    = ivshmem_class_init,
 884 };
 885
 886 static void ivshmem_register_types(void)
 887 {
 888     type_register_static(&ivshmem_info);
 889 }
 890
 891 type_init(ivshmem_register_types)