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_MAX_PEERS G_MAXUINT16
  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
  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_64bit;
  89     int shm_fd; /* shared memory file descriptor */
  90
  91     Peer *peers;
  92     int nb_peers; /* how many guests we have space for */
  93     int max_peer; /* maximum numbered peer */
  94
  95     int vm_id;
  96     uint32_t vectors;
  97     uint32_t features;
  98     EventfdEntry *eventfd_table;
  99
 100     Error *migration_blocker;
 101
 102     char * shmobj;
 103     char * sizearg;
 104     char * role;
 105     int role_val;   /* scalar to avoid multiple string comparisons */
 106 } IVShmemState;
 107
 108 /* registers for the Inter-VM shared memory device */
 109 enum ivshmem_registers {
 110     INTRMASK = 0,
 111     INTRSTATUS = 4,
 112     IVPOSITION = 8,
 113     DOORBELL = 12,
 114 };
 115
 116 static inline uint32_t ivshmem_has_feature(IVShmemState *ivs,
 117                                                     unsigned int feature) {
 118     return (ivs->features & (1 << feature));
 119 }
 120
 121 static inline bool is_power_of_two(uint64_t x) {
 122     return (x & (x - 1)) == 0;
 123 }
 124
 125 /* accessing registers - based on rtl8139 */
 126 static void ivshmem_update_irq(IVShmemState *s)
 127 {
 128     PCIDevice *d = PCI_DEVICE(s);
 129     int isr;
 130     isr = (s->intrstatus & s->intrmask) & 0xffffffff;
 131
 132     /* don't print ISR resets */
 133     if (isr) {
 134         IVSHMEM_DPRINTF("Set IRQ to %d (%04x %04x)\n",
 135                         isr ? 1 : 0, s->intrstatus, s->intrmask);
 136     }
 137
 138     pci_set_irq(d, (isr != 0));
 139 }
 140
 141 static void ivshmem_IntrMask_write(IVShmemState *s, uint32_t val)
 142 {
 143     IVSHMEM_DPRINTF("IntrMask write(w) val = 0x%04x\n", val);
 144
 145     s->intrmask = val;
 146
 147     ivshmem_update_irq(s);
 148 }
 149
 150 static uint32_t ivshmem_IntrMask_read(IVShmemState *s)
 151 {
 152     uint32_t ret = s->intrmask;
 153
 154     IVSHMEM_DPRINTF("intrmask read(w) val = 0x%04x\n", ret);
 155
 156     return ret;
 157 }
 158
 159 static void ivshmem_IntrStatus_write(IVShmemState *s, uint32_t val)
 160 {
 161     IVSHMEM_DPRINTF("IntrStatus write(w) val = 0x%04x\n", val);
 162
 163     s->intrstatus = val;
 164
 165     ivshmem_update_irq(s);
 166 }
 167
 168 static uint32_t ivshmem_IntrStatus_read(IVShmemState *s)
 169 {
 170     uint32_t ret = s->intrstatus;
 171
 172     /* reading ISR clears all interrupts */
 173     s->intrstatus = 0;
 174
 175     ivshmem_update_irq(s);
 176
 177     return ret;
 178 }
 179
 180 static void ivshmem_io_write(void *opaque, hwaddr addr,
 181                              uint64_t val, unsigned size)
 182 {
 183     IVShmemState *s = opaque;
 184
 185     uint16_t dest = val >> 16;
 186     uint16_t vector = val & 0xff;
 187
 188     addr &= 0xfc;
 189
 190     IVSHMEM_DPRINTF("writing to addr " TARGET_FMT_plx "\n", addr);
 191     switch (addr)
 192     {
 193         case INTRMASK:
 194             ivshmem_IntrMask_write(s, val);
 195             break;
 196
 197         case INTRSTATUS:
 198             ivshmem_IntrStatus_write(s, val);
 199             break;
 200
 201         case DOORBELL:
 202             /* check that dest VM ID is reasonable */
 203             if (dest > s->max_peer) {
 204                 IVSHMEM_DPRINTF("Invalid destination VM ID (%d)\n", dest);
 205                 break;
 206             }
 207
 208             /* check doorbell range */
 209             if (vector < s->peers[dest].nb_eventfds) {
 210                 IVSHMEM_DPRINTF("Notifying VM %d on vector %d\n", dest, vector);
 211                 event_notifier_set(&s->peers[dest].eventfds[vector]);
 212             }
 213             break;
 214         default:
 215             IVSHMEM_DPRINTF("Invalid VM Doorbell VM %d\n", dest);
 216     }
 217 }
 218
 219 static uint64_t ivshmem_io_read(void *opaque, hwaddr addr,
 220                                 unsigned size)
 221 {
 222
 223     IVShmemState *s = opaque;
 224     uint32_t ret;
 225
 226     switch (addr)
 227     {
 228         case INTRMASK:
 229             ret = ivshmem_IntrMask_read(s);
 230             break;
 231
 232         case INTRSTATUS:
 233             ret = ivshmem_IntrStatus_read(s);
 234             break;
 235
 236         case IVPOSITION:
 237             /* return my VM ID if the memory is mapped */
 238             if (s->shm_fd > 0) {
 239                 ret = s->vm_id;
 240             } else {
 241                 ret = -1;
 242             }
 243             break;
 244
 245         default:
 246             IVSHMEM_DPRINTF("why are we reading " TARGET_FMT_plx "\n", addr);
 247             ret = 0;
 248     }
 249
 250     return ret;
 251 }
 252
 253 static const MemoryRegionOps ivshmem_mmio_ops = {
 254     .read = ivshmem_io_read,
 255     .write = ivshmem_io_write,
 256     .endianness = DEVICE_NATIVE_ENDIAN,
 257     .impl = {
 258         .min_access_size = 4,
 259         .max_access_size = 4,
 260     },
 261 };
 262
 263 static void ivshmem_receive(void *opaque, const uint8_t *buf, int size)
 264 {
 265     IVShmemState *s = opaque;
 266
 267     ivshmem_IntrStatus_write(s, *buf);
 268
 269     IVSHMEM_DPRINTF("ivshmem_receive 0x%02x\n", *buf);
 270 }
 271
 272 static int ivshmem_can_receive(void * opaque)
 273 {
 274     return sizeof(long);
 275 }
 276
 277 static void ivshmem_event(void *opaque, int event)
 278 {
 279     IVSHMEM_DPRINTF("ivshmem_event %d\n", event);
 280 }
 281
 282 static void fake_irqfd(void *opaque, const uint8_t *buf, int size) {
 283
 284     EventfdEntry *entry = opaque;
 285     PCIDevice *pdev = entry->pdev;
 286
 287     IVSHMEM_DPRINTF("interrupt on vector %p %d\n", pdev, entry->vector);
 288     msix_notify(pdev, entry->vector);
 289 }
 290
 291 static CharDriverState* create_eventfd_chr_device(void * opaque, EventNotifier *n,
 292                                                   int vector)
 293 {
 294     /* create a event character device based on the passed eventfd */
 295     IVShmemState *s = opaque;
 296     CharDriverState * chr;
 297     int eventfd = event_notifier_get_fd(n);
 298
 299     chr = qemu_chr_open_eventfd(eventfd);
 300
 301     if (chr == NULL) {
 302         error_report("creating eventfd for eventfd %d failed", eventfd);
 303         return NULL;
 304     }
 305     qemu_chr_fe_claim_no_fail(chr);
 306
 307     /* if MSI is supported we need multiple interrupts */
 308     if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
 309         s->eventfd_table[vector].pdev = PCI_DEVICE(s);
 310         s->eventfd_table[vector].vector = vector;
 311
 312         qemu_chr_add_handlers(chr, ivshmem_can_receive, fake_irqfd,
 313                       ivshmem_event, &s->eventfd_table[vector]);
 314     } else {
 315         qemu_chr_add_handlers(chr, ivshmem_can_receive, ivshmem_receive,
 316                       ivshmem_event, s);
 317     }
 318
 319     return chr;
 320
 321 }
 322
 323 static int check_shm_size(IVShmemState *s, int fd, Error **errp)
 324 {
 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_setg(errp, "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_setg(errp, "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 int create_shared_memory_BAR(IVShmemState *s, int fd, uint8_t attr,
 349                                     Error **errp)
 350 {
 351     void * ptr;
 352
 353     ptr = mmap(0, s->ivshmem_size, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
 354     if (ptr == MAP_FAILED) {
 355         error_setg_errno(errp, errno, "Failed to mmap shared memory");
 356         return -1;
 357     }
 358
 359     s->shm_fd = fd;
 360
 361     memory_region_init_ram_ptr(&s->ivshmem, OBJECT(s), "ivshmem.bar2",
 362                                s->ivshmem_size, ptr);
 363     vmstate_register_ram(&s->ivshmem, DEVICE(s));
 364     memory_region_add_subregion(&s->bar, 0, &s->ivshmem);
 365
 366     /* region for shared memory */
 367     pci_register_bar(PCI_DEVICE(s), 2, attr, &s->bar);
 368
 369     return 0;
 370 }
 371
 372 static void ivshmem_add_eventfd(IVShmemState *s, int posn, int i)
 373 {
 374     memory_region_add_eventfd(&s->ivshmem_mmio,
 375                               DOORBELL,
 376                               4,
 377                               true,
 378                               (posn << 16) | i,
 379                               &s->peers[posn].eventfds[i]);
 380 }
 381
 382 static void ivshmem_del_eventfd(IVShmemState *s, int posn, int i)
 383 {
 384     memory_region_del_eventfd(&s->ivshmem_mmio,
 385                               DOORBELL,
 386                               4,
 387                               true,
 388                               (posn << 16) | i,
 389                               &s->peers[posn].eventfds[i]);
 390 }
 391
 392 static void close_guest_eventfds(IVShmemState *s, int posn)
 393 {
 394     int i, guest_curr_max;
 395
 396     if (!ivshmem_has_feature(s, IVSHMEM_IOEVENTFD)) {
 397         return;
 398     }
 399     if (posn < 0 || posn >= s->nb_peers) {
 400         return;
 401     }
 402
 403     guest_curr_max = s->peers[posn].nb_eventfds;
 404
 405     memory_region_transaction_begin();
 406     for (i = 0; i < guest_curr_max; i++) {
 407         ivshmem_del_eventfd(s, posn, i);
 408     }
 409     memory_region_transaction_commit();
 410     for (i = 0; i < guest_curr_max; i++) {
 411         event_notifier_cleanup(&s->peers[posn].eventfds[i]);
 412     }
 413
 414     g_free(s->peers[posn].eventfds);
 415     s->peers[posn].nb_eventfds = 0;
 416 }
 417
 418 /* this function increase the dynamic storage need to store data about other
 419  * guests */
 420 static int resize_peers(IVShmemState *s, int new_min_size)
 421 {
 422
 423     int j, old_size;
 424
 425     /* limit number of max peers */
 426     if (new_min_size <= 0 || new_min_size > IVSHMEM_MAX_PEERS) {
 427         return -1;
 428     }
 429     if (new_min_size <= s->nb_peers) {
 430         return 0;
 431     }
 432
 433     old_size = s->nb_peers;
 434     s->nb_peers = new_min_size;
 435
 436     IVSHMEM_DPRINTF("bumping storage to %d guests\n", s->nb_peers);
 437
 438     s->peers = g_realloc(s->peers, s->nb_peers * sizeof(Peer));
 439
 440     for (j = old_size; j < s->nb_peers; j++) {
 441         s->peers[j].eventfds = g_new0(EventNotifier, s->vectors);
 442         s->peers[j].nb_eventfds = 0;
 443     }
 444
 445     return 0;
 446 }
 447
 448 static bool fifo_update_and_get(IVShmemState *s, const uint8_t *buf, int size,
 449                                 void *data, size_t len)
 450 {
 451     const uint8_t *p;
 452     uint32_t num;
 453
 454     assert(len <= sizeof(long)); /* limitation of the fifo */
 455     if (fifo8_is_empty(&s->incoming_fifo) && size == len) {
 456         memcpy(data, buf, size);
 457         return true;
 458     }
 459
 460     IVSHMEM_DPRINTF("short read of %d bytes\n", size);
 461
 462     num = MIN(size, sizeof(long) - fifo8_num_used(&s->incoming_fifo));
 463     fifo8_push_all(&s->incoming_fifo, buf, num);
 464
 465     if (fifo8_num_used(&s->incoming_fifo) < len) {
 466         assert(num == 0);
 467         return false;
 468     }
 469
 470     size -= num;
 471     buf += num;
 472     p = fifo8_pop_buf(&s->incoming_fifo, len, &num);
 473     assert(num == len);
 474
 475     memcpy(data, p, len);
 476
 477     if (size > 0) {
 478         fifo8_push_all(&s->incoming_fifo, buf, size);
 479     }
 480
 481     return true;
 482 }
 483
 484 static void ivshmem_read(void *opaque, const uint8_t *buf, int size)
 485 {
 486     IVShmemState *s = opaque;
 487     int incoming_fd;
 488     int guest_max_eventfd;
 489     long incoming_posn;
 490     Error *err = NULL;
 491
 492     if (!fifo_update_and_get(s, buf, size,
 493                              &incoming_posn, sizeof(incoming_posn))) {
 494         return;
 495     }
 496
 497     if (incoming_posn < -1) {
 498         IVSHMEM_DPRINTF("invalid incoming_posn %ld\n", incoming_posn);
 499         return;
 500     }
 501
 502     /* pick off s->server_chr->msgfd and store it, posn should accompany msg */
 503     incoming_fd = qemu_chr_fe_get_msgfd(s->server_chr);
 504     IVSHMEM_DPRINTF("posn is %ld, fd is %d\n", incoming_posn, incoming_fd);
 505
 506     /* make sure we have enough space for this guest */
 507     if (incoming_posn >= s->nb_peers) {
 508         if (resize_peers(s, incoming_posn + 1) < 0) {
 509             error_report("failed to resize peers array");
 510             if (incoming_fd != -1) {
 511                 close(incoming_fd);
 512             }
 513             return;
 514         }
 515     }
 516
 517     if (incoming_fd == -1) {
 518         /* if posn is positive and unseen before then this is our posn*/
 519         if (incoming_posn >= 0 && s->vm_id == -1) {
 520             /* receive our posn */
 521             s->vm_id = incoming_posn;
 522             return;
 523         } else {
 524             /* otherwise an fd == -1 means an existing guest has gone away */
 525             IVSHMEM_DPRINTF("posn %ld has gone away\n", incoming_posn);
 526             close_guest_eventfds(s, incoming_posn);
 527             return;
 528         }
 529     }
 530
 531     /* if the position is -1, then it's shared memory region fd */
 532     if (incoming_posn == -1) {
 533         void * map_ptr;
 534
 535         if (check_shm_size(s, incoming_fd, &err) == -1) {
 536             error_report_err(err);
 537             close(incoming_fd);
 538             return;
 539         }
 540
 541         /* mmap the region and map into the BAR2 */
 542         map_ptr = mmap(0, s->ivshmem_size, PROT_READ|PROT_WRITE, MAP_SHARED,
 543                                                             incoming_fd, 0);
 544         if (map_ptr == MAP_FAILED) {
 545             error_report("Failed to mmap shared memory %s", strerror(errno));
 546             close(incoming_fd);
 547             return;
 548         }
 549         memory_region_init_ram_ptr(&s->ivshmem, OBJECT(s),
 550                                    "ivshmem.bar2", s->ivshmem_size, map_ptr);
 551         vmstate_register_ram(&s->ivshmem, DEVICE(s));
 552
 553         IVSHMEM_DPRINTF("guest h/w addr = %p, size = %" PRIu64 "\n",
 554                         map_ptr, s->ivshmem_size);
 555
 556         memory_region_add_subregion(&s->bar, 0, &s->ivshmem);
 557
 558         /* only store the fd if it is successfully mapped */
 559         s->shm_fd = incoming_fd;
 560
 561         return;
 562     }
 563
 564     /* each guest has an array of eventfds, and we keep track of how many
 565      * guests for each VM */
 566     guest_max_eventfd = s->peers[incoming_posn].nb_eventfds;
 567
 568     /* this is an eventfd for a particular guest VM */
 569     IVSHMEM_DPRINTF("eventfds[%ld][%d] = %d\n", incoming_posn,
 570                     guest_max_eventfd, incoming_fd);
 571     event_notifier_init_fd(&s->peers[incoming_posn].eventfds[guest_max_eventfd],
 572                            incoming_fd);
 573
 574     /* increment count for particular guest */
 575     s->peers[incoming_posn].nb_eventfds++;
 576
 577     /* keep track of the maximum VM ID */
 578     if (incoming_posn > s->max_peer) {
 579         s->max_peer = incoming_posn;
 580     }
 581
 582     if (incoming_posn == s->vm_id) {
 583         s->eventfd_chr[guest_max_eventfd] = create_eventfd_chr_device(s,
 584                    &s->peers[s->vm_id].eventfds[guest_max_eventfd],
 585                    guest_max_eventfd);
 586     }
 587
 588     if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD)) {
 589         ivshmem_add_eventfd(s, incoming_posn, guest_max_eventfd);
 590     }
 591 }
 592
 593 /* Select the MSI-X vectors used by device.
 594  * ivshmem maps events to vectors statically, so
 595  * we just enable all vectors on init and after reset. */
 596 static void ivshmem_use_msix(IVShmemState * s)
 597 {
 598     PCIDevice *d = PCI_DEVICE(s);
 599     int i;
 600
 601     if (!msix_present(d)) {
 602         return;
 603     }
 604
 605     for (i = 0; i < s->vectors; i++) {
 606         msix_vector_use(d, i);
 607     }
 608 }
 609
 610 static void ivshmem_reset(DeviceState *d)
 611 {
 612     IVShmemState *s = IVSHMEM(d);
 613
 614     s->intrstatus = 0;
 615     ivshmem_use_msix(s);
 616 }
 617
 618 static uint64_t ivshmem_get_size(IVShmemState * s, Error **errp) {
 619
 620     uint64_t value;
 621     char *ptr;
 622
 623     value = strtoull(s->sizearg, &ptr, 10);
 624     switch (*ptr) {
 625         case 0: case 'M': case 'm':
 626             value <<= 20;
 627             break;
 628         case 'G': case 'g':
 629             value <<= 30;
 630             break;
 631         default:
 632             error_setg(errp, "invalid ram size: %s", s->sizearg);
 633             return 0;
 634     }
 635
 636     /* BARs must be a power of 2 */
 637     if (!is_power_of_two(value)) {
 638         error_setg(errp, "size must be power of 2");
 639         return 0;
 640     }
 641
 642     return value;
 643 }
 644
 645 static int ivshmem_setup_msi(IVShmemState * s)
 646 {
 647     if (msix_init_exclusive_bar(PCI_DEVICE(s), s->vectors, 1)) {
 648         return -1;
 649     }
 650
 651     IVSHMEM_DPRINTF("msix initialized (%d vectors)\n", s->vectors);
 652
 653     /* allocate QEMU char devices for receiving interrupts */
 654     s->eventfd_table = g_malloc0(s->vectors * sizeof(EventfdEntry));
 655
 656     ivshmem_use_msix(s);
 657     return 0;
 658 }
 659
 660 static void ivshmem_save(QEMUFile* f, void *opaque)
 661 {
 662     IVShmemState *proxy = opaque;
 663     PCIDevice *pci_dev = PCI_DEVICE(proxy);
 664
 665     IVSHMEM_DPRINTF("ivshmem_save\n");
 666     pci_device_save(pci_dev, f);
 667
 668     if (ivshmem_has_feature(proxy, IVSHMEM_MSI)) {
 669         msix_save(pci_dev, f);
 670     } else {
 671         qemu_put_be32(f, proxy->intrstatus);
 672         qemu_put_be32(f, proxy->intrmask);
 673     }
 674
 675 }
 676
 677 static int ivshmem_load(QEMUFile* f, void *opaque, int version_id)
 678 {
 679     IVSHMEM_DPRINTF("ivshmem_load\n");
 680
 681     IVShmemState *proxy = opaque;
 682     PCIDevice *pci_dev = PCI_DEVICE(proxy);
 683     int ret;
 684
 685     if (version_id > 0) {
 686         return -EINVAL;
 687     }
 688
 689     if (proxy->role_val == IVSHMEM_PEER) {
 690         error_report("'peer' devices are not migratable");
 691         return -EINVAL;
 692     }
 693
 694     ret = pci_device_load(pci_dev, f);
 695     if (ret) {
 696         return ret;
 697     }
 698
 699     if (ivshmem_has_feature(proxy, IVSHMEM_MSI)) {
 700         msix_load(pci_dev, f);
 701         ivshmem_use_msix(proxy);
 702     } else {
 703         proxy->intrstatus = qemu_get_be32(f);
 704         proxy->intrmask = qemu_get_be32(f);
 705     }
 706
 707     return 0;
 708 }
 709
 710 static void ivshmem_write_config(PCIDevice *pci_dev, uint32_t address,
 711                                  uint32_t val, int len)
 712 {
 713     pci_default_write_config(pci_dev, address, val, len);
 714 }
 715
 716 static void pci_ivshmem_realize(PCIDevice *dev, Error **errp)
 717 {
 718     IVShmemState *s = IVSHMEM(dev);
 719     uint8_t *pci_conf;
 720     uint8_t attr = PCI_BASE_ADDRESS_SPACE_MEMORY |
 721         PCI_BASE_ADDRESS_MEM_PREFETCH;
 722     Error *local_err = NULL;
 723
 724     s->max_peer = -1;
 725
 726     if (s->sizearg == NULL) {
 727         s->ivshmem_size = 4 << 20; /* 4 MB default */
 728     } else {
 729         s->ivshmem_size = ivshmem_get_size(s, &local_err);
 730         if (local_err) {
 731             error_propagate(errp, local_err);
 732             return;
 733         }
 734     }
 735
 736     fifo8_create(&s->incoming_fifo, sizeof(long));
 737     register_savevm(DEVICE(dev), "ivshmem", 0, 0, ivshmem_save, ivshmem_load,
 738                                                                         dev);
 739     /* IRQFD requires MSI */
 740     if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD) &&
 741         !ivshmem_has_feature(s, IVSHMEM_MSI)) {
 742         error_setg(errp, "ioeventfd/irqfd requires MSI");
 743         return;
 744     }
 745
 746     /* check that role is reasonable */
 747     if (s->role) {
 748         if (strncmp(s->role, "peer", 5) == 0) {
 749             s->role_val = IVSHMEM_PEER;
 750         } else if (strncmp(s->role, "master", 7) == 0) {
 751             s->role_val = IVSHMEM_MASTER;
 752         } else {
 753             error_setg(errp, "'role' must be 'peer' or 'master'");
 754             return;
 755         }
 756     } else {
 757         s->role_val = IVSHMEM_MASTER; /* default */
 758     }
 759
 760     if (s->role_val == IVSHMEM_PEER) {
 761         error_setg(&s->migration_blocker,
 762                    "Migration is disabled when using feature 'peer mode' in device 'ivshmem'");
 763         migrate_add_blocker(s->migration_blocker);
 764     }
 765
 766     pci_conf = dev->config;
 767     pci_conf[PCI_COMMAND] = PCI_COMMAND_IO | PCI_COMMAND_MEMORY;
 768
 769     pci_config_set_interrupt_pin(pci_conf, 1);
 770
 771     s->shm_fd = 0;
 772
 773     memory_region_init_io(&s->ivshmem_mmio, OBJECT(s), &ivshmem_mmio_ops, s,
 774                           "ivshmem-mmio", IVSHMEM_REG_BAR_SIZE);
 775
 776     /* region for registers*/
 777     pci_register_bar(dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY,
 778                      &s->ivshmem_mmio);
 779
 780     memory_region_init(&s->bar, OBJECT(s), "ivshmem-bar2-container", s->ivshmem_size);
 781     if (s->ivshmem_64bit) {
 782         attr |= PCI_BASE_ADDRESS_MEM_TYPE_64;
 783     }
 784
 785     if ((s->server_chr != NULL) &&
 786                         (strncmp(s->server_chr->filename, "unix:", 5) == 0)) {
 787         /* if we get a UNIX socket as the parameter we will talk
 788          * to the ivshmem server to receive the memory region */
 789
 790         if (s->shmobj != NULL) {
 791             error_setg(errp, "do not specify both 'chardev' "
 792                        "and 'shm' with ivshmem");
 793             return;
 794         }
 795
 796         IVSHMEM_DPRINTF("using shared memory server (socket = %s)\n",
 797                         s->server_chr->filename);
 798
 799         if (ivshmem_has_feature(s, IVSHMEM_MSI) &&
 800             ivshmem_setup_msi(s)) {
 801             error_setg(errp, "msix initialization failed");
 802             return;
 803         }
 804
 805         /* we allocate enough space for 16 guests and grow as needed */
 806         resize_peers(s, 16);
 807         s->vm_id = -1;
 808
 809         pci_register_bar(dev, 2, attr, &s->bar);
 810
 811         s->eventfd_chr = g_malloc0(s->vectors * sizeof(CharDriverState *));
 812
 813         qemu_chr_add_handlers(s->server_chr, ivshmem_can_receive, ivshmem_read,
 814                      ivshmem_event, s);
 815     } else {
 816         /* just map the file immediately, we're not using a server */
 817         int fd;
 818
 819         if (s->shmobj == NULL) {
 820             error_setg(errp, "Must specify 'chardev' or 'shm' to ivshmem");
 821             return;
 822         }
 823
 824         IVSHMEM_DPRINTF("using shm_open (shm object = %s)\n", s->shmobj);
 825
 826         /* try opening with O_EXCL and if it succeeds zero the memory
 827          * by truncating to 0 */
 828         if ((fd = shm_open(s->shmobj, O_CREAT|O_RDWR|O_EXCL,
 829                         S_IRWXU|S_IRWXG|S_IRWXO)) > 0) {
 830            /* truncate file to length PCI device's memory */
 831             if (ftruncate(fd, s->ivshmem_size) != 0) {
 832                 error_report("could not truncate shared file");
 833             }
 834
 835         } else if ((fd = shm_open(s->shmobj, O_CREAT|O_RDWR,
 836                         S_IRWXU|S_IRWXG|S_IRWXO)) < 0) {
 837             error_setg(errp, "could not open shared file");
 838             return;
 839         }
 840
 841         if (check_shm_size(s, fd, errp) == -1) {
 842             return;
 843         }
 844
 845         create_shared_memory_BAR(s, fd, attr, errp);
 846     }
 847 }
 848
 849 static void pci_ivshmem_exit(PCIDevice *dev)
 850 {
 851     IVShmemState *s = IVSHMEM(dev);
 852
 853     if (s->migration_blocker) {
 854         migrate_del_blocker(s->migration_blocker);
 855         error_free(s->migration_blocker);
 856     }
 857
 858     memory_region_del_subregion(&s->bar, &s->ivshmem);
 859     vmstate_unregister_ram(&s->ivshmem, DEVICE(dev));
 860     unregister_savevm(DEVICE(dev), "ivshmem", s);
 861     fifo8_destroy(&s->incoming_fifo);
 862 }
 863
 864 static Property ivshmem_properties[] = {
 865     DEFINE_PROP_CHR("chardev", IVShmemState, server_chr),
 866     DEFINE_PROP_STRING("size", IVShmemState, sizearg),
 867     DEFINE_PROP_UINT32("vectors", IVShmemState, vectors, 1),
 868     DEFINE_PROP_BIT("ioeventfd", IVShmemState, features, IVSHMEM_IOEVENTFD, false),
 869     DEFINE_PROP_BIT("msi", IVShmemState, features, IVSHMEM_MSI, true),
 870     DEFINE_PROP_STRING("shm", IVShmemState, shmobj),
 871     DEFINE_PROP_STRING("role", IVShmemState, role),
 872     DEFINE_PROP_UINT32("use64", IVShmemState, ivshmem_64bit, 1),
 873     DEFINE_PROP_END_OF_LIST(),
 874 };
 875
 876 static void ivshmem_class_init(ObjectClass *klass, void *data)
 877 {
 878     DeviceClass *dc = DEVICE_CLASS(klass);
 879     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
 880
 881     k->realize = pci_ivshmem_realize;
 882     k->exit = pci_ivshmem_exit;
 883     k->config_write = ivshmem_write_config;
 884     k->vendor_id = PCI_VENDOR_ID_IVSHMEM;
 885     k->device_id = PCI_DEVICE_ID_IVSHMEM;
 886     k->class_id = PCI_CLASS_MEMORY_RAM;
 887     dc->reset = ivshmem_reset;
 888     dc->props = ivshmem_properties;
 889     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
 890 }
 891
 892 static const TypeInfo ivshmem_info = {
 893     .name          = TYPE_IVSHMEM,
 894     .parent        = TYPE_PCI_DEVICE,
 895     .instance_size = sizeof(IVShmemState),
 896     .class_init    = ivshmem_class_init,
 897 };
 898
 899 static void ivshmem_register_types(void)
 900 {
 901     type_register_static(&ivshmem_info);
 902 }
 903
 904 type_init(ivshmem_register_types)