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/msi.h"
  23 #include "hw/pci/msix.h"
  24 #include "sysemu/kvm.h"
  25 #include "migration/migration.h"
  26 #include "qemu/error-report.h"
  27 #include "qemu/event_notifier.h"
  28 #include "qemu/fifo8.h"
  29 #include "sysemu/char.h"
  30 #include "sysemu/hostmem.h"
  31 #include "qapi/visitor.h"
  32
  33 #include "hw/misc/ivshmem.h"
  34
  35 #include <sys/mman.h>
  36 #include <sys/types.h>
  37 #include <limits.h>
  38
  39 #define PCI_VENDOR_ID_IVSHMEM   PCI_VENDOR_ID_REDHAT_QUMRANET
  40 #define PCI_DEVICE_ID_IVSHMEM   0x1110
  41
  42 #define IVSHMEM_MAX_PEERS G_MAXUINT16
  43 #define IVSHMEM_IOEVENTFD   0
  44 #define IVSHMEM_MSI     1
  45
  46 #define IVSHMEM_PEER    0
  47 #define IVSHMEM_MASTER  1
  48
  49 #define IVSHMEM_REG_BAR_SIZE 0x100
  50
  51 //#define DEBUG_IVSHMEM
  52 #ifdef DEBUG_IVSHMEM
  53 #define IVSHMEM_DPRINTF(fmt, ...)        \
  54     do {printf("IVSHMEM: " fmt, ## __VA_ARGS__); } while (0)
  55 #else
  56 #define IVSHMEM_DPRINTF(fmt, ...)
  57 #endif
  58
  59 #define TYPE_IVSHMEM "ivshmem"
  60 #define IVSHMEM(obj) \
  61     OBJECT_CHECK(IVShmemState, (obj), TYPE_IVSHMEM)
  62
  63 typedef struct Peer {
  64     int nb_eventfds;
  65     EventNotifier *eventfds;
  66 } Peer;
  67
  68 typedef struct MSIVector {
  69     PCIDevice *pdev;
  70     int virq;
  71 } MSIVector;
  72
  73 typedef struct IVShmemState {
  74     /*< private >*/
  75     PCIDevice parent_obj;
  76     /*< public >*/
  77
  78     HostMemoryBackend *hostmem;
  79     uint32_t intrmask;
  80     uint32_t intrstatus;
  81
  82     CharDriverState **eventfd_chr;
  83     CharDriverState *server_chr;
  84     Fifo8 incoming_fifo;
  85     MemoryRegion ivshmem_mmio;
  86
  87     /* We might need to register the BAR before we actually have the memory.
  88      * So prepare a container MemoryRegion for the BAR immediately and
  89      * add a subregion when we have the memory.
  90      */
  91     MemoryRegion bar;
  92     MemoryRegion ivshmem;
  93     uint64_t ivshmem_size; /* size of shared memory region */
  94     uint32_t ivshmem_64bit;
  95
  96     Peer *peers;
  97     int nb_peers; /* how many peers we have space for */
  98
  99     int vm_id;
 100     uint32_t vectors;
 101     uint32_t features;
 102     MSIVector *msi_vectors;
 103
 104     Error *migration_blocker;
 105
 106     char * shmobj;
 107     char * sizearg;
 108     char * role;
 109     int role_val;   /* scalar to avoid multiple string comparisons */
 110 } IVShmemState;
 111
 112 /* registers for the Inter-VM shared memory device */
 113 enum ivshmem_registers {
 114     INTRMASK = 0,
 115     INTRSTATUS = 4,
 116     IVPOSITION = 8,
 117     DOORBELL = 12,
 118 };
 119
 120 static inline uint32_t ivshmem_has_feature(IVShmemState *ivs,
 121                                                     unsigned int feature) {
 122     return (ivs->features & (1 << feature));
 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->nb_peers) {
 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             } else {
 213                 IVSHMEM_DPRINTF("Invalid destination vector %d on VM %d\n",
 214                                 vector, dest);
 215             }
 216             break;
 217         default:
 218             IVSHMEM_DPRINTF("Unhandled write " TARGET_FMT_plx "\n", addr);
 219     }
 220 }
 221
 222 static uint64_t ivshmem_io_read(void *opaque, hwaddr addr,
 223                                 unsigned size)
 224 {
 225
 226     IVShmemState *s = opaque;
 227     uint32_t ret;
 228
 229     switch (addr)
 230     {
 231         case INTRMASK:
 232             ret = ivshmem_IntrMask_read(s);
 233             break;
 234
 235         case INTRSTATUS:
 236             ret = ivshmem_IntrStatus_read(s);
 237             break;
 238
 239         case IVPOSITION:
 240             /* return my VM ID if the memory is mapped */
 241             if (memory_region_is_mapped(&s->ivshmem)) {
 242                 ret = s->vm_id;
 243             } else {
 244                 ret = -1;
 245             }
 246             break;
 247
 248         default:
 249             IVSHMEM_DPRINTF("why are we reading " TARGET_FMT_plx "\n", addr);
 250             ret = 0;
 251     }
 252
 253     return ret;
 254 }
 255
 256 static const MemoryRegionOps ivshmem_mmio_ops = {
 257     .read = ivshmem_io_read,
 258     .write = ivshmem_io_write,
 259     .endianness = DEVICE_NATIVE_ENDIAN,
 260     .impl = {
 261         .min_access_size = 4,
 262         .max_access_size = 4,
 263     },
 264 };
 265
 266 static void ivshmem_receive(void *opaque, const uint8_t *buf, int size)
 267 {
 268     IVShmemState *s = opaque;
 269
 270     IVSHMEM_DPRINTF("ivshmem_receive 0x%02x size: %d\n", *buf, size);
 271
 272     ivshmem_IntrStatus_write(s, *buf);
 273 }
 274
 275 static int ivshmem_can_receive(void * opaque)
 276 {
 277     return sizeof(int64_t);
 278 }
 279
 280 static void ivshmem_event(void *opaque, int event)
 281 {
 282     IVSHMEM_DPRINTF("ivshmem_event %d\n", event);
 283 }
 284
 285 static void fake_irqfd(void *opaque, const uint8_t *buf, int size) {
 286
 287     MSIVector *entry = opaque;
 288     PCIDevice *pdev = entry->pdev;
 289     IVShmemState *s = IVSHMEM(pdev);
 290     int vector = entry - s->msi_vectors;
 291
 292     IVSHMEM_DPRINTF("interrupt on vector %p %d\n", pdev, vector);
 293     msix_notify(pdev, vector);
 294 }
 295
 296 static int ivshmem_vector_unmask(PCIDevice *dev, unsigned vector,
 297                                  MSIMessage msg)
 298 {
 299     IVShmemState *s = IVSHMEM(dev);
 300     EventNotifier *n = &s->peers[s->vm_id].eventfds[vector];
 301     MSIVector *v = &s->msi_vectors[vector];
 302     int ret;
 303
 304     IVSHMEM_DPRINTF("vector unmask %p %d\n", dev, vector);
 305
 306     ret = kvm_irqchip_update_msi_route(kvm_state, v->virq, msg, dev);
 307     if (ret < 0) {
 308         return ret;
 309     }
 310
 311     return kvm_irqchip_add_irqfd_notifier_gsi(kvm_state, n, NULL, v->virq);
 312 }
 313
 314 static void ivshmem_vector_mask(PCIDevice *dev, unsigned vector)
 315 {
 316     IVShmemState *s = IVSHMEM(dev);
 317     EventNotifier *n = &s->peers[s->vm_id].eventfds[vector];
 318     int ret;
 319
 320     IVSHMEM_DPRINTF("vector mask %p %d\n", dev, vector);
 321
 322     ret = kvm_irqchip_remove_irqfd_notifier_gsi(kvm_state, n,
 323                                                 s->msi_vectors[vector].virq);
 324     if (ret != 0) {
 325         error_report("remove_irqfd_notifier_gsi failed");
 326     }
 327 }
 328
 329 static void ivshmem_vector_poll(PCIDevice *dev,
 330                                 unsigned int vector_start,
 331                                 unsigned int vector_end)
 332 {
 333     IVShmemState *s = IVSHMEM(dev);
 334     unsigned int vector;
 335
 336     IVSHMEM_DPRINTF("vector poll %p %d-%d\n", dev, vector_start, vector_end);
 337
 338     vector_end = MIN(vector_end, s->vectors);
 339
 340     for (vector = vector_start; vector < vector_end; vector++) {
 341         EventNotifier *notifier = &s->peers[s->vm_id].eventfds[vector];
 342
 343         if (!msix_is_masked(dev, vector)) {
 344             continue;
 345         }
 346
 347         if (event_notifier_test_and_clear(notifier)) {
 348             msix_set_pending(dev, vector);
 349         }
 350     }
 351 }
 352
 353 static CharDriverState* create_eventfd_chr_device(void * opaque, EventNotifier *n,
 354                                                   int vector)
 355 {
 356     /* create a event character device based on the passed eventfd */
 357     IVShmemState *s = opaque;
 358     PCIDevice *pdev = PCI_DEVICE(s);
 359     int eventfd = event_notifier_get_fd(n);
 360     CharDriverState *chr;
 361
 362     s->msi_vectors[vector].pdev = pdev;
 363
 364     chr = qemu_chr_open_eventfd(eventfd);
 365
 366     if (chr == NULL) {
 367         error_report("creating chardriver for eventfd %d failed", eventfd);
 368         return NULL;
 369     }
 370     qemu_chr_fe_claim_no_fail(chr);
 371
 372     /* if MSI is supported we need multiple interrupts */
 373     if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
 374         s->msi_vectors[vector].pdev = PCI_DEVICE(s);
 375
 376         qemu_chr_add_handlers(chr, ivshmem_can_receive, fake_irqfd,
 377                       ivshmem_event, &s->msi_vectors[vector]);
 378     } else {
 379         qemu_chr_add_handlers(chr, ivshmem_can_receive, ivshmem_receive,
 380                       ivshmem_event, s);
 381     }
 382
 383     return chr;
 384
 385 }
 386
 387 static int check_shm_size(IVShmemState *s, int fd, Error **errp)
 388 {
 389     /* check that the guest isn't going to try and map more memory than the
 390      * the object has allocated return -1 to indicate error */
 391
 392     struct stat buf;
 393
 394     if (fstat(fd, &buf) < 0) {
 395         error_setg(errp, "exiting: fstat on fd %d failed: %s",
 396                    fd, strerror(errno));
 397         return -1;
 398     }
 399
 400     if (s->ivshmem_size > buf.st_size) {
 401         error_setg(errp, "Requested memory size greater"
 402                    " than shared object size (%" PRIu64 " > %" PRIu64")",
 403                    s->ivshmem_size, (uint64_t)buf.st_size);
 404         return -1;
 405     } else {
 406         return 0;
 407     }
 408 }
 409
 410 /* create the shared memory BAR when we are not using the server, so we can
 411  * create the BAR and map the memory immediately */
 412 static int create_shared_memory_BAR(IVShmemState *s, int fd, uint8_t attr,
 413                                     Error **errp)
 414 {
 415     void * ptr;
 416
 417     ptr = mmap(0, s->ivshmem_size, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
 418     if (ptr == MAP_FAILED) {
 419         error_setg_errno(errp, errno, "Failed to mmap shared memory");
 420         return -1;
 421     }
 422
 423     memory_region_init_ram_ptr(&s->ivshmem, OBJECT(s), "ivshmem.bar2",
 424                                s->ivshmem_size, ptr);
 425     vmstate_register_ram(&s->ivshmem, DEVICE(s));
 426     memory_region_add_subregion(&s->bar, 0, &s->ivshmem);
 427
 428     /* region for shared memory */
 429     pci_register_bar(PCI_DEVICE(s), 2, attr, &s->bar);
 430
 431     return 0;
 432 }
 433
 434 static void ivshmem_add_eventfd(IVShmemState *s, int posn, int i)
 435 {
 436     memory_region_add_eventfd(&s->ivshmem_mmio,
 437                               DOORBELL,
 438                               4,
 439                               true,
 440                               (posn << 16) | i,
 441                               &s->peers[posn].eventfds[i]);
 442 }
 443
 444 static void ivshmem_del_eventfd(IVShmemState *s, int posn, int i)
 445 {
 446     memory_region_del_eventfd(&s->ivshmem_mmio,
 447                               DOORBELL,
 448                               4,
 449                               true,
 450                               (posn << 16) | i,
 451                               &s->peers[posn].eventfds[i]);
 452 }
 453
 454 static void close_peer_eventfds(IVShmemState *s, int posn)
 455 {
 456     int i, n;
 457
 458     if (!ivshmem_has_feature(s, IVSHMEM_IOEVENTFD)) {
 459         return;
 460     }
 461     if (posn < 0 || posn >= s->nb_peers) {
 462         error_report("invalid peer %d", posn);
 463         return;
 464     }
 465
 466     n = s->peers[posn].nb_eventfds;
 467
 468     memory_region_transaction_begin();
 469     for (i = 0; i < n; i++) {
 470         ivshmem_del_eventfd(s, posn, i);
 471     }
 472     memory_region_transaction_commit();
 473     for (i = 0; i < n; i++) {
 474         event_notifier_cleanup(&s->peers[posn].eventfds[i]);
 475     }
 476
 477     g_free(s->peers[posn].eventfds);
 478     s->peers[posn].nb_eventfds = 0;
 479 }
 480
 481 /* this function increase the dynamic storage need to store data about other
 482  * peers */
 483 static int resize_peers(IVShmemState *s, int new_min_size)
 484 {
 485
 486     int j, old_size;
 487
 488     /* limit number of max peers */
 489     if (new_min_size <= 0 || new_min_size > IVSHMEM_MAX_PEERS) {
 490         return -1;
 491     }
 492     if (new_min_size <= s->nb_peers) {
 493         return 0;
 494     }
 495
 496     old_size = s->nb_peers;
 497     s->nb_peers = new_min_size;
 498
 499     IVSHMEM_DPRINTF("bumping storage to %d peers\n", s->nb_peers);
 500
 501     s->peers = g_realloc(s->peers, s->nb_peers * sizeof(Peer));
 502
 503     for (j = old_size; j < s->nb_peers; j++) {
 504         s->peers[j].eventfds = g_new0(EventNotifier, s->vectors);
 505         s->peers[j].nb_eventfds = 0;
 506     }
 507
 508     return 0;
 509 }
 510
 511 static bool fifo_update_and_get(IVShmemState *s, const uint8_t *buf, int size,
 512                                 void *data, size_t len)
 513 {
 514     const uint8_t *p;
 515     uint32_t num;
 516
 517     assert(len <= sizeof(int64_t)); /* limitation of the fifo */
 518     if (fifo8_is_empty(&s->incoming_fifo) && size == len) {
 519         memcpy(data, buf, size);
 520         return true;
 521     }
 522
 523     IVSHMEM_DPRINTF("short read of %d bytes\n", size);
 524
 525     num = MIN(size, sizeof(int64_t) - fifo8_num_used(&s->incoming_fifo));
 526     fifo8_push_all(&s->incoming_fifo, buf, num);
 527
 528     if (fifo8_num_used(&s->incoming_fifo) < len) {
 529         assert(num == 0);
 530         return false;
 531     }
 532
 533     size -= num;
 534     buf += num;
 535     p = fifo8_pop_buf(&s->incoming_fifo, len, &num);
 536     assert(num == len);
 537
 538     memcpy(data, p, len);
 539
 540     if (size > 0) {
 541         fifo8_push_all(&s->incoming_fifo, buf, size);
 542     }
 543
 544     return true;
 545 }
 546
 547 static bool fifo_update_and_get_i64(IVShmemState *s,
 548                                     const uint8_t *buf, int size, int64_t *i64)
 549 {
 550     if (fifo_update_and_get(s, buf, size, i64, sizeof(*i64))) {
 551         *i64 = GINT64_FROM_LE(*i64);
 552         return true;
 553     }
 554
 555     return false;
 556 }
 557
 558 static int ivshmem_add_kvm_msi_virq(IVShmemState *s, int vector)
 559 {
 560     PCIDevice *pdev = PCI_DEVICE(s);
 561     MSIMessage msg = msix_get_message(pdev, vector);
 562     int ret;
 563
 564     IVSHMEM_DPRINTF("ivshmem_add_kvm_msi_virq vector:%d\n", vector);
 565
 566     if (s->msi_vectors[vector].pdev != NULL) {
 567         return 0;
 568     }
 569
 570     ret = kvm_irqchip_add_msi_route(kvm_state, msg, pdev);
 571     if (ret < 0) {
 572         error_report("ivshmem: kvm_irqchip_add_msi_route failed");
 573         return -1;
 574     }
 575
 576     s->msi_vectors[vector].virq = ret;
 577     s->msi_vectors[vector].pdev = pdev;
 578
 579     return 0;
 580 }
 581
 582 static void setup_interrupt(IVShmemState *s, int vector)
 583 {
 584     EventNotifier *n = &s->peers[s->vm_id].eventfds[vector];
 585     bool with_irqfd = kvm_msi_via_irqfd_enabled() &&
 586         ivshmem_has_feature(s, IVSHMEM_MSI);
 587     PCIDevice *pdev = PCI_DEVICE(s);
 588
 589     IVSHMEM_DPRINTF("setting up interrupt for vector: %d\n", vector);
 590
 591     if (!with_irqfd) {
 592         IVSHMEM_DPRINTF("with eventfd");
 593         s->eventfd_chr[vector] = create_eventfd_chr_device(s, n, vector);
 594     } else if (msix_enabled(pdev)) {
 595         IVSHMEM_DPRINTF("with irqfd");
 596         if (ivshmem_add_kvm_msi_virq(s, vector) < 0) {
 597             return;
 598         }
 599
 600         if (!msix_is_masked(pdev, vector)) {
 601             kvm_irqchip_add_irqfd_notifier_gsi(kvm_state, n, NULL,
 602                                                s->msi_vectors[vector].virq);
 603         }
 604     } else {
 605         /* it will be delayed until msix is enabled, in write_config */
 606         IVSHMEM_DPRINTF("with irqfd, delayed until msix enabled");
 607     }
 608 }
 609
 610 static void ivshmem_read(void *opaque, const uint8_t *buf, int size)
 611 {
 612     IVShmemState *s = opaque;
 613     int incoming_fd;
 614     int new_eventfd;
 615     int64_t incoming_posn;
 616     Error *err = NULL;
 617     Peer *peer;
 618
 619     if (!fifo_update_and_get_i64(s, buf, size, &incoming_posn)) {
 620         return;
 621     }
 622
 623     if (incoming_posn < -1) {
 624         IVSHMEM_DPRINTF("invalid incoming_posn %" PRId64 "\n", incoming_posn);
 625         return;
 626     }
 627
 628     /* pick off s->server_chr->msgfd and store it, posn should accompany msg */
 629     incoming_fd = qemu_chr_fe_get_msgfd(s->server_chr);
 630     IVSHMEM_DPRINTF("posn is %" PRId64 ", fd is %d\n",
 631                     incoming_posn, incoming_fd);
 632
 633     /* make sure we have enough space for this peer */
 634     if (incoming_posn >= s->nb_peers) {
 635         if (resize_peers(s, incoming_posn + 1) < 0) {
 636             error_report("failed to resize peers array");
 637             if (incoming_fd != -1) {
 638                 close(incoming_fd);
 639             }
 640             return;
 641         }
 642     }
 643
 644     peer = &s->peers[incoming_posn];
 645
 646     if (incoming_fd == -1) {
 647         /* if posn is positive and unseen before then this is our posn*/
 648         if (incoming_posn >= 0 && s->vm_id == -1) {
 649             /* receive our posn */
 650             s->vm_id = incoming_posn;
 651         } else {
 652             /* otherwise an fd == -1 means an existing peer has gone away */
 653             IVSHMEM_DPRINTF("posn %" PRId64 " has gone away\n", incoming_posn);
 654             close_peer_eventfds(s, incoming_posn);
 655         }
 656         return;
 657     }
 658
 659     /* if the position is -1, then it's shared memory region fd */
 660     if (incoming_posn == -1) {
 661         void * map_ptr;
 662
 663         if (memory_region_is_mapped(&s->ivshmem)) {
 664             error_report("shm already initialized");
 665             close(incoming_fd);
 666             return;
 667         }
 668
 669         if (check_shm_size(s, incoming_fd, &err) == -1) {
 670             error_report_err(err);
 671             close(incoming_fd);
 672             return;
 673         }
 674
 675         /* mmap the region and map into the BAR2 */
 676         map_ptr = mmap(0, s->ivshmem_size, PROT_READ|PROT_WRITE, MAP_SHARED,
 677                                                             incoming_fd, 0);
 678         if (map_ptr == MAP_FAILED) {
 679             error_report("Failed to mmap shared memory %s", strerror(errno));
 680             close(incoming_fd);
 681             return;
 682         }
 683         memory_region_init_ram_ptr(&s->ivshmem, OBJECT(s),
 684                                    "ivshmem.bar2", s->ivshmem_size, map_ptr);
 685         vmstate_register_ram(&s->ivshmem, DEVICE(s));
 686
 687         IVSHMEM_DPRINTF("guest h/w addr = %p, size = %" PRIu64 "\n",
 688                         map_ptr, s->ivshmem_size);
 689
 690         memory_region_add_subregion(&s->bar, 0, &s->ivshmem);
 691
 692         close(incoming_fd);
 693         return;
 694     }
 695
 696     /* each peer has an associated array of eventfds, and we keep
 697      * track of how many eventfds received so far */
 698     /* get a new eventfd: */
 699     if (peer->nb_eventfds >= s->vectors) {
 700         error_report("Too many eventfd received, device has %d vectors",
 701                      s->vectors);
 702         close(incoming_fd);
 703         return;
 704     }
 705
 706     new_eventfd = peer->nb_eventfds++;
 707
 708     /* this is an eventfd for a particular peer VM */
 709     IVSHMEM_DPRINTF("eventfds[%" PRId64 "][%d] = %d\n", incoming_posn,
 710                     new_eventfd, incoming_fd);
 711     event_notifier_init_fd(&peer->eventfds[new_eventfd], incoming_fd);
 712     fcntl_setfl(incoming_fd, O_NONBLOCK); /* msix/irqfd poll non block */
 713
 714     if (incoming_posn == s->vm_id) {
 715         setup_interrupt(s, new_eventfd);
 716     }
 717
 718     if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD)) {
 719         ivshmem_add_eventfd(s, incoming_posn, new_eventfd);
 720     }
 721 }
 722
 723 static void ivshmem_check_version(void *opaque, const uint8_t * buf, int size)
 724 {
 725     IVShmemState *s = opaque;
 726     int tmp;
 727     int64_t version;
 728
 729     if (!fifo_update_and_get_i64(s, buf, size, &version)) {
 730         return;
 731     }
 732
 733     tmp = qemu_chr_fe_get_msgfd(s->server_chr);
 734     if (tmp != -1 || version != IVSHMEM_PROTOCOL_VERSION) {
 735         fprintf(stderr, "incompatible version, you are connecting to a ivshmem-"
 736                 "server using a different protocol please check your setup\n");
 737         qemu_chr_delete(s->server_chr);
 738         s->server_chr = NULL;
 739         return;
 740     }
 741
 742     IVSHMEM_DPRINTF("version check ok, switch to real chardev handler\n");
 743     qemu_chr_add_handlers(s->server_chr, ivshmem_can_receive, ivshmem_read,
 744                           ivshmem_event, s);
 745 }
 746
 747 /* Select the MSI-X vectors used by device.
 748  * ivshmem maps events to vectors statically, so
 749  * we just enable all vectors on init and after reset. */
 750 static void ivshmem_use_msix(IVShmemState * s)
 751 {
 752     PCIDevice *d = PCI_DEVICE(s);
 753     int i;
 754
 755     IVSHMEM_DPRINTF("%s, msix present: %d\n", __func__, msix_present(d));
 756     if (!msix_present(d)) {
 757         return;
 758     }
 759
 760     for (i = 0; i < s->vectors; i++) {
 761         msix_vector_use(d, i);
 762     }
 763 }
 764
 765 static void ivshmem_reset(DeviceState *d)
 766 {
 767     IVShmemState *s = IVSHMEM(d);
 768
 769     s->intrstatus = 0;
 770     s->intrmask = 0;
 771     ivshmem_use_msix(s);
 772 }
 773
 774 static int ivshmem_setup_msi(IVShmemState * s)
 775 {
 776     if (msix_init_exclusive_bar(PCI_DEVICE(s), s->vectors, 1)) {
 777         return -1;
 778     }
 779
 780     IVSHMEM_DPRINTF("msix initialized (%d vectors)\n", s->vectors);
 781
 782     /* allocate QEMU char devices for receiving interrupts */
 783     s->msi_vectors = g_malloc0(s->vectors * sizeof(MSIVector));
 784
 785     ivshmem_use_msix(s);
 786     return 0;
 787 }
 788
 789 static void ivshmem_enable_irqfd(IVShmemState *s)
 790 {
 791     PCIDevice *pdev = PCI_DEVICE(s);
 792     int i;
 793
 794     for (i = 0; i < s->peers[s->vm_id].nb_eventfds; i++) {
 795         ivshmem_add_kvm_msi_virq(s, i);
 796     }
 797
 798     if (msix_set_vector_notifiers(pdev,
 799                                   ivshmem_vector_unmask,
 800                                   ivshmem_vector_mask,
 801                                   ivshmem_vector_poll)) {
 802         error_report("ivshmem: msix_set_vector_notifiers failed");
 803     }
 804 }
 805
 806 static void ivshmem_remove_kvm_msi_virq(IVShmemState *s, int vector)
 807 {
 808     IVSHMEM_DPRINTF("ivshmem_remove_kvm_msi_virq vector:%d\n", vector);
 809
 810     if (s->msi_vectors[vector].pdev == NULL) {
 811         return;
 812     }
 813
 814     /* it was cleaned when masked in the frontend. */
 815     kvm_irqchip_release_virq(kvm_state, s->msi_vectors[vector].virq);
 816
 817     s->msi_vectors[vector].pdev = NULL;
 818 }
 819
 820 static void ivshmem_disable_irqfd(IVShmemState *s)
 821 {
 822     PCIDevice *pdev = PCI_DEVICE(s);
 823     int i;
 824
 825     for (i = 0; i < s->peers[s->vm_id].nb_eventfds; i++) {
 826         ivshmem_remove_kvm_msi_virq(s, i);
 827     }
 828
 829     msix_unset_vector_notifiers(pdev);
 830 }
 831
 832 static void ivshmem_write_config(PCIDevice *pdev, uint32_t address,
 833                                  uint32_t val, int len)
 834 {
 835     IVShmemState *s = IVSHMEM(pdev);
 836     int is_enabled, was_enabled = msix_enabled(pdev);
 837
 838     pci_default_write_config(pdev, address, val, len);
 839     is_enabled = msix_enabled(pdev);
 840
 841     if (kvm_msi_via_irqfd_enabled() && s->vm_id != -1) {
 842         if (!was_enabled && is_enabled) {
 843             ivshmem_enable_irqfd(s);
 844         } else if (was_enabled && !is_enabled) {
 845             ivshmem_disable_irqfd(s);
 846         }
 847     }
 848 }
 849
 850 static void pci_ivshmem_realize(PCIDevice *dev, Error **errp)
 851 {
 852     IVShmemState *s = IVSHMEM(dev);
 853     uint8_t *pci_conf;
 854     uint8_t attr = PCI_BASE_ADDRESS_SPACE_MEMORY |
 855         PCI_BASE_ADDRESS_MEM_PREFETCH;
 856
 857     if (!!s->server_chr + !!s->shmobj + !!s->hostmem != 1) {
 858         error_setg(errp,
 859                    "You must specify either 'shm', 'chardev' or 'x-memdev'");
 860         return;
 861     }
 862
 863     if (s->hostmem) {
 864         MemoryRegion *mr;
 865
 866         if (s->sizearg) {
 867             g_warning("size argument ignored with hostmem");
 868         }
 869
 870         mr = host_memory_backend_get_memory(s->hostmem, errp);
 871         s->ivshmem_size = memory_region_size(mr);
 872     } else if (s->sizearg == NULL) {
 873         s->ivshmem_size = 4 << 20; /* 4 MB default */
 874     } else {
 875         char *end;
 876         int64_t size = qemu_strtosz(s->sizearg, &end);
 877         if (size < 0 || *end != '\0' || !is_power_of_2(size)) {
 878             error_setg(errp, "Invalid size %s", s->sizearg);
 879             return;
 880         }
 881         s->ivshmem_size = size;
 882     }
 883
 884     fifo8_create(&s->incoming_fifo, sizeof(int64_t));
 885
 886     /* IRQFD requires MSI */
 887     if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD) &&
 888         !ivshmem_has_feature(s, IVSHMEM_MSI)) {
 889         error_setg(errp, "ioeventfd/irqfd requires MSI");
 890         return;
 891     }
 892
 893     /* check that role is reasonable */
 894     if (s->role) {
 895         if (strncmp(s->role, "peer", 5) == 0) {
 896             s->role_val = IVSHMEM_PEER;
 897         } else if (strncmp(s->role, "master", 7) == 0) {
 898             s->role_val = IVSHMEM_MASTER;
 899         } else {
 900             error_setg(errp, "'role' must be 'peer' or 'master'");
 901             return;
 902         }
 903     } else {
 904         s->role_val = IVSHMEM_MASTER; /* default */
 905     }
 906
 907     if (s->role_val == IVSHMEM_PEER) {
 908         error_setg(&s->migration_blocker,
 909                    "Migration is disabled when using feature 'peer mode' in device 'ivshmem'");
 910         migrate_add_blocker(s->migration_blocker);
 911     }
 912
 913     pci_conf = dev->config;
 914     pci_conf[PCI_COMMAND] = PCI_COMMAND_IO | PCI_COMMAND_MEMORY;
 915
 916     pci_config_set_interrupt_pin(pci_conf, 1);
 917
 918     memory_region_init_io(&s->ivshmem_mmio, OBJECT(s), &ivshmem_mmio_ops, s,
 919                           "ivshmem-mmio", IVSHMEM_REG_BAR_SIZE);
 920
 921     /* region for registers*/
 922     pci_register_bar(dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY,
 923                      &s->ivshmem_mmio);
 924
 925     memory_region_init(&s->bar, OBJECT(s), "ivshmem-bar2-container", s->ivshmem_size);
 926     if (s->ivshmem_64bit) {
 927         attr |= PCI_BASE_ADDRESS_MEM_TYPE_64;
 928     }
 929
 930     if (s->hostmem != NULL) {
 931         MemoryRegion *mr;
 932
 933         IVSHMEM_DPRINTF("using hostmem\n");
 934
 935         mr = host_memory_backend_get_memory(MEMORY_BACKEND(s->hostmem), errp);
 936         vmstate_register_ram(mr, DEVICE(s));
 937         memory_region_add_subregion(&s->bar, 0, mr);
 938         pci_register_bar(PCI_DEVICE(s), 2, attr, &s->bar);
 939     } else if (s->server_chr != NULL) {
 940         /* FIXME do not rely on what chr drivers put into filename */
 941         if (strncmp(s->server_chr->filename, "unix:", 5)) {
 942             error_setg(errp, "chardev is not a unix client socket");
 943             return;
 944         }
 945
 946         /* if we get a UNIX socket as the parameter we will talk
 947          * to the ivshmem server to receive the memory region */
 948
 949         IVSHMEM_DPRINTF("using shared memory server (socket = %s)\n",
 950                         s->server_chr->filename);
 951
 952         if (ivshmem_has_feature(s, IVSHMEM_MSI) &&
 953             ivshmem_setup_msi(s)) {
 954             error_setg(errp, "msix initialization failed");
 955             return;
 956         }
 957
 958         /* we allocate enough space for 16 peers and grow as needed */
 959         resize_peers(s, 16);
 960         s->vm_id = -1;
 961
 962         pci_register_bar(dev, 2, attr, &s->bar);
 963
 964         s->eventfd_chr = g_malloc0(s->vectors * sizeof(CharDriverState *));
 965
 966         qemu_chr_add_handlers(s->server_chr, ivshmem_can_receive,
 967                               ivshmem_check_version, ivshmem_event, s);
 968     } else {
 969         /* just map the file immediately, we're not using a server */
 970         int fd;
 971
 972         IVSHMEM_DPRINTF("using shm_open (shm object = %s)\n", s->shmobj);
 973
 974         /* try opening with O_EXCL and if it succeeds zero the memory
 975          * by truncating to 0 */
 976         if ((fd = shm_open(s->shmobj, O_CREAT|O_RDWR|O_EXCL,
 977                         S_IRWXU|S_IRWXG|S_IRWXO)) > 0) {
 978            /* truncate file to length PCI device's memory */
 979             if (ftruncate(fd, s->ivshmem_size) != 0) {
 980                 error_report("could not truncate shared file");
 981             }
 982
 983         } else if ((fd = shm_open(s->shmobj, O_CREAT|O_RDWR,
 984                         S_IRWXU|S_IRWXG|S_IRWXO)) < 0) {
 985             error_setg(errp, "could not open shared file");
 986             return;
 987         }
 988
 989         if (check_shm_size(s, fd, errp) == -1) {
 990             return;
 991         }
 992
 993         create_shared_memory_BAR(s, fd, attr, errp);
 994         close(fd);
 995     }
 996 }
 997
 998 static void pci_ivshmem_exit(PCIDevice *dev)
 999 {
1000     IVShmemState *s = IVSHMEM(dev);
1001     int i;
1002
1003     fifo8_destroy(&s->incoming_fifo);
1004
1005     if (s->migration_blocker) {
1006         migrate_del_blocker(s->migration_blocker);
1007         error_free(s->migration_blocker);
1008     }
1009
1010     if (memory_region_is_mapped(&s->ivshmem)) {
1011         if (!s->hostmem) {
1012             void *addr = memory_region_get_ram_ptr(&s->ivshmem);
1013
1014             if (munmap(addr, s->ivshmem_size) == -1) {
1015                 error_report("Failed to munmap shared memory %s",
1016                              strerror(errno));
1017             }
1018         }
1019
1020         vmstate_unregister_ram(&s->ivshmem, DEVICE(dev));
1021         memory_region_del_subregion(&s->bar, &s->ivshmem);
1022     }
1023
1024     if (s->eventfd_chr) {
1025         for (i = 0; i < s->vectors; i++) {
1026             if (s->eventfd_chr[i]) {
1027                 qemu_chr_free(s->eventfd_chr[i]);
1028             }
1029         }
1030         g_free(s->eventfd_chr);
1031     }
1032
1033     if (s->peers) {
1034         for (i = 0; i < s->nb_peers; i++) {
1035             close_peer_eventfds(s, i);
1036         }
1037         g_free(s->peers);
1038     }
1039
1040     if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
1041         msix_uninit_exclusive_bar(dev);
1042     }
1043
1044     g_free(s->msi_vectors);
1045 }
1046
1047 static bool test_msix(void *opaque, int version_id)
1048 {
1049     IVShmemState *s = opaque;
1050
1051     return ivshmem_has_feature(s, IVSHMEM_MSI);
1052 }
1053
1054 static bool test_no_msix(void *opaque, int version_id)
1055 {
1056     return !test_msix(opaque, version_id);
1057 }
1058
1059 static int ivshmem_pre_load(void *opaque)
1060 {
1061     IVShmemState *s = opaque;
1062
1063     if (s->role_val == IVSHMEM_PEER) {
1064         error_report("'peer' devices are not migratable");
1065         return -EINVAL;
1066     }
1067
1068     return 0;
1069 }
1070
1071 static int ivshmem_post_load(void *opaque, int version_id)
1072 {
1073     IVShmemState *s = opaque;
1074
1075     if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
1076         ivshmem_use_msix(s);
1077     }
1078
1079     return 0;
1080 }
1081
1082 static int ivshmem_load_old(QEMUFile *f, void *opaque, int version_id)
1083 {
1084     IVShmemState *s = opaque;
1085     PCIDevice *pdev = PCI_DEVICE(s);
1086     int ret;
1087
1088     IVSHMEM_DPRINTF("ivshmem_load_old\n");
1089
1090     if (version_id != 0) {
1091         return -EINVAL;
1092     }
1093
1094     if (s->role_val == IVSHMEM_PEER) {
1095         error_report("'peer' devices are not migratable");
1096         return -EINVAL;
1097     }
1098
1099     ret = pci_device_load(pdev, f);
1100     if (ret) {
1101         return ret;
1102     }
1103
1104     if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
1105         msix_load(pdev, f);
1106         ivshmem_use_msix(s);
1107     } else {
1108         s->intrstatus = qemu_get_be32(f);
1109         s->intrmask = qemu_get_be32(f);
1110     }
1111
1112     return 0;
1113 }
1114
1115 static const VMStateDescription ivshmem_vmsd = {
1116     .name = "ivshmem",
1117     .version_id = 1,
1118     .minimum_version_id = 1,
1119     .pre_load = ivshmem_pre_load,
1120     .post_load = ivshmem_post_load,
1121     .fields = (VMStateField[]) {
1122         VMSTATE_PCI_DEVICE(parent_obj, IVShmemState),
1123
1124         VMSTATE_MSIX_TEST(parent_obj, IVShmemState, test_msix),
1125         VMSTATE_UINT32_TEST(intrstatus, IVShmemState, test_no_msix),
1126         VMSTATE_UINT32_TEST(intrmask, IVShmemState, test_no_msix),
1127
1128         VMSTATE_END_OF_LIST()
1129     },
1130     .load_state_old = ivshmem_load_old,
1131     .minimum_version_id_old = 0
1132 };
1133
1134 static Property ivshmem_properties[] = {
1135     DEFINE_PROP_CHR("chardev", IVShmemState, server_chr),
1136     DEFINE_PROP_STRING("size", IVShmemState, sizearg),
1137     DEFINE_PROP_UINT32("vectors", IVShmemState, vectors, 1),
1138     DEFINE_PROP_BIT("ioeventfd", IVShmemState, features, IVSHMEM_IOEVENTFD, false),
1139     DEFINE_PROP_BIT("msi", IVShmemState, features, IVSHMEM_MSI, true),
1140     DEFINE_PROP_STRING("shm", IVShmemState, shmobj),
1141     DEFINE_PROP_STRING("role", IVShmemState, role),
1142     DEFINE_PROP_UINT32("use64", IVShmemState, ivshmem_64bit, 1),
1143     DEFINE_PROP_END_OF_LIST(),
1144 };
1145
1146 static void ivshmem_class_init(ObjectClass *klass, void *data)
1147 {
1148     DeviceClass *dc = DEVICE_CLASS(klass);
1149     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
1150
1151     k->realize = pci_ivshmem_realize;
1152     k->exit = pci_ivshmem_exit;
1153     k->config_write = ivshmem_write_config;
1154     k->vendor_id = PCI_VENDOR_ID_IVSHMEM;
1155     k->device_id = PCI_DEVICE_ID_IVSHMEM;
1156     k->class_id = PCI_CLASS_MEMORY_RAM;
1157     dc->reset = ivshmem_reset;
1158     dc->props = ivshmem_properties;
1159     dc->vmsd = &ivshmem_vmsd;
1160     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
1161     dc->desc = "Inter-VM shared memory";
1162 }
1163
1164 static void ivshmem_check_memdev_is_busy(Object *obj, const char *name,
1165                                          Object *val, Error **errp)
1166 {
1167     MemoryRegion *mr;
1168
1169     mr = host_memory_backend_get_memory(MEMORY_BACKEND(val), errp);
1170     if (memory_region_is_mapped(mr)) {
1171         char *path = object_get_canonical_path_component(val);
1172         error_setg(errp, "can't use already busy memdev: %s", path);
1173         g_free(path);
1174     } else {
1175         qdev_prop_allow_set_link_before_realize(obj, name, val, errp);
1176     }
1177 }
1178
1179 static void ivshmem_init(Object *obj)
1180 {
1181     IVShmemState *s = IVSHMEM(obj);
1182
1183     object_property_add_link(obj, "x-memdev", TYPE_MEMORY_BACKEND,
1184                              (Object **)&s->hostmem,
1185                              ivshmem_check_memdev_is_busy,
1186                              OBJ_PROP_LINK_UNREF_ON_RELEASE,
1187                              &error_abort);
1188 }
1189
1190 static const TypeInfo ivshmem_info = {
1191     .name          = TYPE_IVSHMEM,
1192     .parent        = TYPE_PCI_DEVICE,
1193     .instance_size = sizeof(IVShmemState),
1194     .instance_init = ivshmem_init,
1195     .class_init    = ivshmem_class_init,
1196 };
1197
1198 static void ivshmem_register_types(void)
1199 {
1200     type_register_static(&ivshmem_info);
1201 }
1202
1203 type_init(ivshmem_register_types)