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