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 #define IVSHMEM_MEMDEV_PROP "memdev"
  64
  65 typedef struct Peer {
  66     int nb_eventfds;
  67     EventNotifier *eventfds;
  68 } Peer;
  69
  70 typedef struct MSIVector {
  71     PCIDevice *pdev;
  72     int virq;
  73 } MSIVector;
  74
  75 typedef struct IVShmemState {
  76     /*< private >*/
  77     PCIDevice parent_obj;
  78     /*< public >*/
  79
  80     HostMemoryBackend *hostmem;
  81     uint32_t intrmask;
  82     uint32_t intrstatus;
  83
  84     CharDriverState **eventfd_chr;
  85     CharDriverState *server_chr;
  86     Fifo8 incoming_fifo;
  87     MemoryRegion ivshmem_mmio;
  88
  89     /* We might need to register the BAR before we actually have the memory.
  90      * So prepare a container MemoryRegion for the BAR immediately and
  91      * add a subregion when we have the memory.
  92      */
  93     MemoryRegion bar;
  94     MemoryRegion ivshmem;
  95     uint64_t ivshmem_size; /* size of shared memory region */
  96     uint32_t ivshmem_64bit;
  97
  98     Peer *peers;
  99     int nb_peers; /* how many peers we have space for */
 100
 101     int vm_id;
 102     uint32_t vectors;
 103     uint32_t features;
 104     MSIVector *msi_vectors;
 105
 106     Error *migration_blocker;
 107
 108     char * shmobj;
 109     char * sizearg;
 110     char * role;
 111     int role_val;   /* scalar to avoid multiple string comparisons */
 112 } IVShmemState;
 113
 114 /* registers for the Inter-VM shared memory device */
 115 enum ivshmem_registers {
 116     INTRMASK = 0,
 117     INTRSTATUS = 4,
 118     IVPOSITION = 8,
 119     DOORBELL = 12,
 120 };
 121
 122 static inline uint32_t ivshmem_has_feature(IVShmemState *ivs,
 123                                                     unsigned int feature) {
 124     return (ivs->features & (1 << feature));
 125 }
 126
 127 /* accessing registers - based on rtl8139 */
 128 static void ivshmem_update_irq(IVShmemState *s)
 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);
 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);
 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);
 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->nb_peers) {
 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             } else {
 215                 IVSHMEM_DPRINTF("Invalid destination vector %d on VM %d\n",
 216                                 vector, dest);
 217             }
 218             break;
 219         default:
 220             IVSHMEM_DPRINTF("Unhandled write " TARGET_FMT_plx "\n", addr);
 221     }
 222 }
 223
 224 static uint64_t ivshmem_io_read(void *opaque, hwaddr addr,
 225                                 unsigned size)
 226 {
 227
 228     IVShmemState *s = opaque;
 229     uint32_t ret;
 230
 231     switch (addr)
 232     {
 233         case INTRMASK:
 234             ret = ivshmem_IntrMask_read(s);
 235             break;
 236
 237         case INTRSTATUS:
 238             ret = ivshmem_IntrStatus_read(s);
 239             break;
 240
 241         case IVPOSITION:
 242             /* return my VM ID if the memory is mapped */
 243             if (memory_region_is_mapped(&s->ivshmem)) {
 244                 ret = s->vm_id;
 245             } else {
 246                 ret = -1;
 247             }
 248             break;
 249
 250         default:
 251             IVSHMEM_DPRINTF("why are we reading " TARGET_FMT_plx "\n", addr);
 252             ret = 0;
 253     }
 254
 255     return ret;
 256 }
 257
 258 static const MemoryRegionOps ivshmem_mmio_ops = {
 259     .read = ivshmem_io_read,
 260     .write = ivshmem_io_write,
 261     .endianness = DEVICE_NATIVE_ENDIAN,
 262     .impl = {
 263         .min_access_size = 4,
 264         .max_access_size = 4,
 265     },
 266 };
 267
 268 static void ivshmem_receive(void *opaque, const uint8_t *buf, int size)
 269 {
 270     IVShmemState *s = opaque;
 271
 272     IVSHMEM_DPRINTF("ivshmem_receive 0x%02x size: %d\n", *buf, size);
 273
 274     ivshmem_IntrStatus_write(s, *buf);
 275 }
 276
 277 static int ivshmem_can_receive(void * opaque)
 278 {
 279     return sizeof(int64_t);
 280 }
 281
 282 static void ivshmem_event(void *opaque, int event)
 283 {
 284     IVSHMEM_DPRINTF("ivshmem_event %d\n", event);
 285 }
 286
 287 static void fake_irqfd(void *opaque, const uint8_t *buf, int size) {
 288
 289     MSIVector *entry = opaque;
 290     PCIDevice *pdev = entry->pdev;
 291     IVShmemState *s = IVSHMEM(pdev);
 292     int vector = entry - s->msi_vectors;
 293
 294     IVSHMEM_DPRINTF("interrupt on vector %p %d\n", pdev, vector);
 295     msix_notify(pdev, vector);
 296 }
 297
 298 static int ivshmem_vector_unmask(PCIDevice *dev, unsigned vector,
 299                                  MSIMessage msg)
 300 {
 301     IVShmemState *s = IVSHMEM(dev);
 302     EventNotifier *n = &s->peers[s->vm_id].eventfds[vector];
 303     MSIVector *v = &s->msi_vectors[vector];
 304     int ret;
 305
 306     IVSHMEM_DPRINTF("vector unmask %p %d\n", dev, vector);
 307
 308     ret = kvm_irqchip_update_msi_route(kvm_state, v->virq, msg, dev);
 309     if (ret < 0) {
 310         return ret;
 311     }
 312
 313     return kvm_irqchip_add_irqfd_notifier_gsi(kvm_state, n, NULL, v->virq);
 314 }
 315
 316 static void ivshmem_vector_mask(PCIDevice *dev, unsigned vector)
 317 {
 318     IVShmemState *s = IVSHMEM(dev);
 319     EventNotifier *n = &s->peers[s->vm_id].eventfds[vector];
 320     int ret;
 321
 322     IVSHMEM_DPRINTF("vector mask %p %d\n", dev, vector);
 323
 324     ret = kvm_irqchip_remove_irqfd_notifier_gsi(kvm_state, n,
 325                                                 s->msi_vectors[vector].virq);
 326     if (ret != 0) {
 327         error_report("remove_irqfd_notifier_gsi failed");
 328     }
 329 }
 330
 331 static void ivshmem_vector_poll(PCIDevice *dev,
 332                                 unsigned int vector_start,
 333                                 unsigned int vector_end)
 334 {
 335     IVShmemState *s = IVSHMEM(dev);
 336     unsigned int vector;
 337
 338     IVSHMEM_DPRINTF("vector poll %p %d-%d\n", dev, vector_start, vector_end);
 339
 340     vector_end = MIN(vector_end, s->vectors);
 341
 342     for (vector = vector_start; vector < vector_end; vector++) {
 343         EventNotifier *notifier = &s->peers[s->vm_id].eventfds[vector];
 344
 345         if (!msix_is_masked(dev, vector)) {
 346             continue;
 347         }
 348
 349         if (event_notifier_test_and_clear(notifier)) {
 350             msix_set_pending(dev, vector);
 351         }
 352     }
 353 }
 354
 355 static CharDriverState* create_eventfd_chr_device(void * opaque, EventNotifier *n,
 356                                                   int vector)
 357 {
 358     /* create a event character device based on the passed eventfd */
 359     IVShmemState *s = opaque;
 360     PCIDevice *pdev = PCI_DEVICE(s);
 361     int eventfd = event_notifier_get_fd(n);
 362     CharDriverState *chr;
 363
 364     s->msi_vectors[vector].pdev = pdev;
 365
 366     chr = qemu_chr_open_eventfd(eventfd);
 367
 368     if (chr == NULL) {
 369         error_report("creating chardriver for eventfd %d failed", eventfd);
 370         return NULL;
 371     }
 372     qemu_chr_fe_claim_no_fail(chr);
 373
 374     /* if MSI is supported we need multiple interrupts */
 375     if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
 376         s->msi_vectors[vector].pdev = PCI_DEVICE(s);
 377
 378         qemu_chr_add_handlers(chr, ivshmem_can_receive, fake_irqfd,
 379                       ivshmem_event, &s->msi_vectors[vector]);
 380     } else {
 381         qemu_chr_add_handlers(chr, ivshmem_can_receive, ivshmem_receive,
 382                       ivshmem_event, s);
 383     }
 384
 385     return chr;
 386
 387 }
 388
 389 static int check_shm_size(IVShmemState *s, int fd, Error **errp)
 390 {
 391     /* check that the guest isn't going to try and map more memory than the
 392      * the object has allocated return -1 to indicate error */
 393
 394     struct stat buf;
 395
 396     if (fstat(fd, &buf) < 0) {
 397         error_setg(errp, "exiting: fstat on fd %d failed: %s",
 398                    fd, strerror(errno));
 399         return -1;
 400     }
 401
 402     if (s->ivshmem_size > buf.st_size) {
 403         error_setg(errp, "Requested memory size greater"
 404                    " than shared object size (%" PRIu64 " > %" PRIu64")",
 405                    s->ivshmem_size, (uint64_t)buf.st_size);
 406         return -1;
 407     } else {
 408         return 0;
 409     }
 410 }
 411
 412 /* create the shared memory BAR when we are not using the server, so we can
 413  * create the BAR and map the memory immediately */
 414 static int create_shared_memory_BAR(IVShmemState *s, int fd, uint8_t attr,
 415                                     Error **errp)
 416 {
 417     void * ptr;
 418
 419     ptr = mmap(0, s->ivshmem_size, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
 420     if (ptr == MAP_FAILED) {
 421         error_setg_errno(errp, errno, "Failed to mmap shared memory");
 422         return -1;
 423     }
 424
 425     memory_region_init_ram_ptr(&s->ivshmem, OBJECT(s), "ivshmem.bar2",
 426                                s->ivshmem_size, ptr);
 427     vmstate_register_ram(&s->ivshmem, DEVICE(s));
 428     memory_region_add_subregion(&s->bar, 0, &s->ivshmem);
 429
 430     /* region for shared memory */
 431     pci_register_bar(PCI_DEVICE(s), 2, attr, &s->bar);
 432
 433     return 0;
 434 }
 435
 436 static void ivshmem_add_eventfd(IVShmemState *s, int posn, int i)
 437 {
 438     memory_region_add_eventfd(&s->ivshmem_mmio,
 439                               DOORBELL,
 440                               4,
 441                               true,
 442                               (posn << 16) | i,
 443                               &s->peers[posn].eventfds[i]);
 444 }
 445
 446 static void ivshmem_del_eventfd(IVShmemState *s, int posn, int i)
 447 {
 448     memory_region_del_eventfd(&s->ivshmem_mmio,
 449                               DOORBELL,
 450                               4,
 451                               true,
 452                               (posn << 16) | i,
 453                               &s->peers[posn].eventfds[i]);
 454 }
 455
 456 static void close_peer_eventfds(IVShmemState *s, int posn)
 457 {
 458     int i, n;
 459
 460     if (!ivshmem_has_feature(s, IVSHMEM_IOEVENTFD)) {
 461         return;
 462     }
 463     if (posn < 0 || posn >= s->nb_peers) {
 464         error_report("invalid peer %d", posn);
 465         return;
 466     }
 467
 468     n = s->peers[posn].nb_eventfds;
 469
 470     memory_region_transaction_begin();
 471     for (i = 0; i < n; i++) {
 472         ivshmem_del_eventfd(s, posn, i);
 473     }
 474     memory_region_transaction_commit();
 475     for (i = 0; i < n; i++) {
 476         event_notifier_cleanup(&s->peers[posn].eventfds[i]);
 477     }
 478
 479     g_free(s->peers[posn].eventfds);
 480     s->peers[posn].nb_eventfds = 0;
 481 }
 482
 483 /* this function increase the dynamic storage need to store data about other
 484  * peers */
 485 static int resize_peers(IVShmemState *s, int new_min_size)
 486 {
 487
 488     int j, old_size;
 489
 490     /* limit number of max peers */
 491     if (new_min_size <= 0 || new_min_size > IVSHMEM_MAX_PEERS) {
 492         return -1;
 493     }
 494     if (new_min_size <= s->nb_peers) {
 495         return 0;
 496     }
 497
 498     old_size = s->nb_peers;
 499     s->nb_peers = new_min_size;
 500
 501     IVSHMEM_DPRINTF("bumping storage to %d peers\n", s->nb_peers);
 502
 503     s->peers = g_realloc(s->peers, s->nb_peers * sizeof(Peer));
 504
 505     for (j = old_size; j < s->nb_peers; j++) {
 506         s->peers[j].eventfds = g_new0(EventNotifier, s->vectors);
 507         s->peers[j].nb_eventfds = 0;
 508     }
 509
 510     return 0;
 511 }
 512
 513 static bool fifo_update_and_get(IVShmemState *s, const uint8_t *buf, int size,
 514                                 void *data, size_t len)
 515 {
 516     const uint8_t *p;
 517     uint32_t num;
 518
 519     assert(len <= sizeof(int64_t)); /* limitation of the fifo */
 520     if (fifo8_is_empty(&s->incoming_fifo) && size == len) {
 521         memcpy(data, buf, size);
 522         return true;
 523     }
 524
 525     IVSHMEM_DPRINTF("short read of %d bytes\n", size);
 526
 527     num = MIN(size, sizeof(int64_t) - fifo8_num_used(&s->incoming_fifo));
 528     fifo8_push_all(&s->incoming_fifo, buf, num);
 529
 530     if (fifo8_num_used(&s->incoming_fifo) < len) {
 531         assert(num == 0);
 532         return false;
 533     }
 534
 535     size -= num;
 536     buf += num;
 537     p = fifo8_pop_buf(&s->incoming_fifo, len, &num);
 538     assert(num == len);
 539
 540     memcpy(data, p, len);
 541
 542     if (size > 0) {
 543         fifo8_push_all(&s->incoming_fifo, buf, size);
 544     }
 545
 546     return true;
 547 }
 548
 549 static bool fifo_update_and_get_i64(IVShmemState *s,
 550                                     const uint8_t *buf, int size, int64_t *i64)
 551 {
 552     if (fifo_update_and_get(s, buf, size, i64, sizeof(*i64))) {
 553         *i64 = GINT64_FROM_LE(*i64);
 554         return true;
 555     }
 556
 557     return false;
 558 }
 559
 560 static int ivshmem_add_kvm_msi_virq(IVShmemState *s, int vector)
 561 {
 562     PCIDevice *pdev = PCI_DEVICE(s);
 563     MSIMessage msg = msix_get_message(pdev, vector);
 564     int ret;
 565
 566     IVSHMEM_DPRINTF("ivshmem_add_kvm_msi_virq vector:%d\n", vector);
 567
 568     if (s->msi_vectors[vector].pdev != NULL) {
 569         return 0;
 570     }
 571
 572     ret = kvm_irqchip_add_msi_route(kvm_state, msg, pdev);
 573     if (ret < 0) {
 574         error_report("ivshmem: kvm_irqchip_add_msi_route failed");
 575         return -1;
 576     }
 577
 578     s->msi_vectors[vector].virq = ret;
 579     s->msi_vectors[vector].pdev = pdev;
 580
 581     return 0;
 582 }
 583
 584 static void setup_interrupt(IVShmemState *s, int vector)
 585 {
 586     EventNotifier *n = &s->peers[s->vm_id].eventfds[vector];
 587     bool with_irqfd = kvm_msi_via_irqfd_enabled() &&
 588         ivshmem_has_feature(s, IVSHMEM_MSI);
 589     PCIDevice *pdev = PCI_DEVICE(s);
 590
 591     IVSHMEM_DPRINTF("setting up interrupt for vector: %d\n", vector);
 592
 593     if (!with_irqfd) {
 594         IVSHMEM_DPRINTF("with eventfd");
 595         s->eventfd_chr[vector] = create_eventfd_chr_device(s, n, vector);
 596     } else if (msix_enabled(pdev)) {
 597         IVSHMEM_DPRINTF("with irqfd");
 598         if (ivshmem_add_kvm_msi_virq(s, vector) < 0) {
 599             return;
 600         }
 601
 602         if (!msix_is_masked(pdev, vector)) {
 603             kvm_irqchip_add_irqfd_notifier_gsi(kvm_state, n, NULL,
 604                                                s->msi_vectors[vector].virq);
 605         }
 606     } else {
 607         /* it will be delayed until msix is enabled, in write_config */
 608         IVSHMEM_DPRINTF("with irqfd, delayed until msix enabled");
 609     }
 610 }
 611
 612 static void ivshmem_read(void *opaque, const uint8_t *buf, int size)
 613 {
 614     IVShmemState *s = opaque;
 615     int incoming_fd;
 616     int new_eventfd;
 617     int64_t incoming_posn;
 618     Error *err = NULL;
 619     Peer *peer;
 620
 621     if (!fifo_update_and_get_i64(s, buf, size, &incoming_posn)) {
 622         return;
 623     }
 624
 625     if (incoming_posn < -1) {
 626         IVSHMEM_DPRINTF("invalid incoming_posn %" PRId64 "\n", incoming_posn);
 627         return;
 628     }
 629
 630     /* pick off s->server_chr->msgfd and store it, posn should accompany msg */
 631     incoming_fd = qemu_chr_fe_get_msgfd(s->server_chr);
 632     IVSHMEM_DPRINTF("posn is %" PRId64 ", fd is %d\n",
 633                     incoming_posn, incoming_fd);
 634
 635     /* make sure we have enough space for this peer */
 636     if (incoming_posn >= s->nb_peers) {
 637         if (resize_peers(s, incoming_posn + 1) < 0) {
 638             error_report("failed to resize peers array");
 639             if (incoming_fd != -1) {
 640                 close(incoming_fd);
 641             }
 642             return;
 643         }
 644     }
 645
 646     peer = &s->peers[incoming_posn];
 647
 648     if (incoming_fd == -1) {
 649         /* if posn is positive and unseen before then this is our posn*/
 650         if (incoming_posn >= 0 && s->vm_id == -1) {
 651             /* receive our posn */
 652             s->vm_id = incoming_posn;
 653         } else {
 654             /* otherwise an fd == -1 means an existing peer has gone away */
 655             IVSHMEM_DPRINTF("posn %" PRId64 " has gone away\n", incoming_posn);
 656             close_peer_eventfds(s, incoming_posn);
 657         }
 658         return;
 659     }
 660
 661     /* if the position is -1, then it's shared memory region fd */
 662     if (incoming_posn == -1) {
 663         void * map_ptr;
 664
 665         if (memory_region_is_mapped(&s->ivshmem)) {
 666             error_report("shm already initialized");
 667             close(incoming_fd);
 668             return;
 669         }
 670
 671         if (check_shm_size(s, incoming_fd, &err) == -1) {
 672             error_report_err(err);
 673             close(incoming_fd);
 674             return;
 675         }
 676
 677         /* mmap the region and map into the BAR2 */
 678         map_ptr = mmap(0, s->ivshmem_size, PROT_READ|PROT_WRITE, MAP_SHARED,
 679                                                             incoming_fd, 0);
 680         if (map_ptr == MAP_FAILED) {
 681             error_report("Failed to mmap shared memory %s", strerror(errno));
 682             close(incoming_fd);
 683             return;
 684         }
 685         memory_region_init_ram_ptr(&s->ivshmem, OBJECT(s),
 686                                    "ivshmem.bar2", s->ivshmem_size, map_ptr);
 687         vmstate_register_ram(&s->ivshmem, DEVICE(s));
 688
 689         IVSHMEM_DPRINTF("guest h/w addr = %p, size = %" PRIu64 "\n",
 690                         map_ptr, s->ivshmem_size);
 691
 692         memory_region_add_subregion(&s->bar, 0, &s->ivshmem);
 693
 694         close(incoming_fd);
 695         return;
 696     }
 697
 698     /* each peer has an associated array of eventfds, and we keep
 699      * track of how many eventfds received so far */
 700     /* get a new eventfd: */
 701     if (peer->nb_eventfds >= s->vectors) {
 702         error_report("Too many eventfd received, device has %d vectors",
 703                      s->vectors);
 704         close(incoming_fd);
 705         return;
 706     }
 707
 708     new_eventfd = peer->nb_eventfds++;
 709
 710     /* this is an eventfd for a particular peer VM */
 711     IVSHMEM_DPRINTF("eventfds[%" PRId64 "][%d] = %d\n", incoming_posn,
 712                     new_eventfd, incoming_fd);
 713     event_notifier_init_fd(&peer->eventfds[new_eventfd], incoming_fd);
 714     fcntl_setfl(incoming_fd, O_NONBLOCK); /* msix/irqfd poll non block */
 715
 716     if (incoming_posn == s->vm_id) {
 717         setup_interrupt(s, new_eventfd);
 718     }
 719
 720     if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD)) {
 721         ivshmem_add_eventfd(s, incoming_posn, new_eventfd);
 722     }
 723 }
 724
 725 static void ivshmem_check_version(void *opaque, const uint8_t * buf, int size)
 726 {
 727     IVShmemState *s = opaque;
 728     int tmp;
 729     int64_t version;
 730
 731     if (!fifo_update_and_get_i64(s, buf, size, &version)) {
 732         return;
 733     }
 734
 735     tmp = qemu_chr_fe_get_msgfd(s->server_chr);
 736     if (tmp != -1 || version != IVSHMEM_PROTOCOL_VERSION) {
 737         fprintf(stderr, "incompatible version, you are connecting to a ivshmem-"
 738                 "server using a different protocol please check your setup\n");
 739         qemu_chr_delete(s->server_chr);
 740         s->server_chr = NULL;
 741         return;
 742     }
 743
 744     IVSHMEM_DPRINTF("version check ok, switch to real chardev handler\n");
 745     qemu_chr_add_handlers(s->server_chr, ivshmem_can_receive, ivshmem_read,
 746                           ivshmem_event, s);
 747 }
 748
 749 /* Select the MSI-X vectors used by device.
 750  * ivshmem maps events to vectors statically, so
 751  * we just enable all vectors on init and after reset. */
 752 static void ivshmem_use_msix(IVShmemState * s)
 753 {
 754     PCIDevice *d = PCI_DEVICE(s);
 755     int i;
 756
 757     IVSHMEM_DPRINTF("%s, msix present: %d\n", __func__, msix_present(d));
 758     if (!msix_present(d)) {
 759         return;
 760     }
 761
 762     for (i = 0; i < s->vectors; i++) {
 763         msix_vector_use(d, i);
 764     }
 765 }
 766
 767 static void ivshmem_reset(DeviceState *d)
 768 {
 769     IVShmemState *s = IVSHMEM(d);
 770
 771     s->intrstatus = 0;
 772     s->intrmask = 0;
 773     ivshmem_use_msix(s);
 774 }
 775
 776 static int ivshmem_setup_msi(IVShmemState * s)
 777 {
 778     if (msix_init_exclusive_bar(PCI_DEVICE(s), s->vectors, 1)) {
 779         return -1;
 780     }
 781
 782     IVSHMEM_DPRINTF("msix initialized (%d vectors)\n", s->vectors);
 783
 784     /* allocate QEMU char devices for receiving interrupts */
 785     s->msi_vectors = g_malloc0(s->vectors * sizeof(MSIVector));
 786
 787     ivshmem_use_msix(s);
 788     return 0;
 789 }
 790
 791 static void ivshmem_enable_irqfd(IVShmemState *s)
 792 {
 793     PCIDevice *pdev = PCI_DEVICE(s);
 794     int i;
 795
 796     for (i = 0; i < s->peers[s->vm_id].nb_eventfds; i++) {
 797         ivshmem_add_kvm_msi_virq(s, i);
 798     }
 799
 800     if (msix_set_vector_notifiers(pdev,
 801                                   ivshmem_vector_unmask,
 802                                   ivshmem_vector_mask,
 803                                   ivshmem_vector_poll)) {
 804         error_report("ivshmem: msix_set_vector_notifiers failed");
 805     }
 806 }
 807
 808 static void ivshmem_remove_kvm_msi_virq(IVShmemState *s, int vector)
 809 {
 810     IVSHMEM_DPRINTF("ivshmem_remove_kvm_msi_virq vector:%d\n", vector);
 811
 812     if (s->msi_vectors[vector].pdev == NULL) {
 813         return;
 814     }
 815
 816     /* it was cleaned when masked in the frontend. */
 817     kvm_irqchip_release_virq(kvm_state, s->msi_vectors[vector].virq);
 818
 819     s->msi_vectors[vector].pdev = NULL;
 820 }
 821
 822 static void ivshmem_disable_irqfd(IVShmemState *s)
 823 {
 824     PCIDevice *pdev = PCI_DEVICE(s);
 825     int i;
 826
 827     for (i = 0; i < s->peers[s->vm_id].nb_eventfds; i++) {
 828         ivshmem_remove_kvm_msi_virq(s, i);
 829     }
 830
 831     msix_unset_vector_notifiers(pdev);
 832 }
 833
 834 static void ivshmem_write_config(PCIDevice *pdev, uint32_t address,
 835                                  uint32_t val, int len)
 836 {
 837     IVShmemState *s = IVSHMEM(pdev);
 838     int is_enabled, was_enabled = msix_enabled(pdev);
 839
 840     pci_default_write_config(pdev, address, val, len);
 841     is_enabled = msix_enabled(pdev);
 842
 843     if (kvm_msi_via_irqfd_enabled() && s->vm_id != -1) {
 844         if (!was_enabled && is_enabled) {
 845             ivshmem_enable_irqfd(s);
 846         } else if (was_enabled && !is_enabled) {
 847             ivshmem_disable_irqfd(s);
 848         }
 849     }
 850 }
 851
 852 static void pci_ivshmem_realize(PCIDevice *dev, Error **errp)
 853 {
 854     IVShmemState *s = IVSHMEM(dev);
 855     uint8_t *pci_conf;
 856     uint8_t attr = PCI_BASE_ADDRESS_SPACE_MEMORY |
 857         PCI_BASE_ADDRESS_MEM_PREFETCH;
 858
 859     if (!!s->server_chr + !!s->shmobj + !!s->hostmem != 1) {
 860         error_setg(errp, "You must specify either a shmobj, a chardev"
 861                    " or a hostmem");
 862         return;
 863     }
 864
 865     if (s->hostmem) {
 866         MemoryRegion *mr;
 867
 868         if (s->sizearg) {
 869             g_warning("size argument ignored with hostmem");
 870         }
 871
 872         mr = host_memory_backend_get_memory(s->hostmem, errp);
 873         s->ivshmem_size = memory_region_size(mr);
 874     } else if (s->sizearg == NULL) {
 875         s->ivshmem_size = 4 << 20; /* 4 MB default */
 876     } else {
 877         char *end;
 878         int64_t size = qemu_strtosz(s->sizearg, &end);
 879         if (size < 0 || *end != '\0' || !is_power_of_2(size)) {
 880             error_setg(errp, "Invalid size %s", s->sizearg);
 881             return;
 882         }
 883         s->ivshmem_size = size;
 884     }
 885
 886     fifo8_create(&s->incoming_fifo, sizeof(int64_t));
 887
 888     /* IRQFD requires MSI */
 889     if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD) &&
 890         !ivshmem_has_feature(s, IVSHMEM_MSI)) {
 891         error_setg(errp, "ioeventfd/irqfd requires MSI");
 892         return;
 893     }
 894
 895     /* check that role is reasonable */
 896     if (s->role) {
 897         if (strncmp(s->role, "peer", 5) == 0) {
 898             s->role_val = IVSHMEM_PEER;
 899         } else if (strncmp(s->role, "master", 7) == 0) {
 900             s->role_val = IVSHMEM_MASTER;
 901         } else {
 902             error_setg(errp, "'role' must be 'peer' or 'master'");
 903             return;
 904         }
 905     } else {
 906         s->role_val = IVSHMEM_MASTER; /* default */
 907     }
 908
 909     if (s->role_val == IVSHMEM_PEER) {
 910         error_setg(&s->migration_blocker,
 911                    "Migration is disabled when using feature 'peer mode' in device 'ivshmem'");
 912         migrate_add_blocker(s->migration_blocker);
 913     }
 914
 915     pci_conf = dev->config;
 916     pci_conf[PCI_COMMAND] = PCI_COMMAND_IO | PCI_COMMAND_MEMORY;
 917
 918     pci_config_set_interrupt_pin(pci_conf, 1);
 919
 920     memory_region_init_io(&s->ivshmem_mmio, OBJECT(s), &ivshmem_mmio_ops, s,
 921                           "ivshmem-mmio", IVSHMEM_REG_BAR_SIZE);
 922
 923     /* region for registers*/
 924     pci_register_bar(dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY,
 925                      &s->ivshmem_mmio);
 926
 927     memory_region_init(&s->bar, OBJECT(s), "ivshmem-bar2-container", s->ivshmem_size);
 928     if (s->ivshmem_64bit) {
 929         attr |= PCI_BASE_ADDRESS_MEM_TYPE_64;
 930     }
 931
 932     if (s->hostmem != NULL) {
 933         MemoryRegion *mr;
 934
 935         IVSHMEM_DPRINTF("using hostmem\n");
 936
 937         mr = host_memory_backend_get_memory(MEMORY_BACKEND(s->hostmem), errp);
 938         vmstate_register_ram(mr, DEVICE(s));
 939         memory_region_add_subregion(&s->bar, 0, mr);
 940         pci_register_bar(PCI_DEVICE(s), 2, attr, &s->bar);
 941     } else if (s->server_chr != NULL) {
 942         if (strncmp(s->server_chr->filename, "unix:", 5)) {
 943             error_setg(errp, "chardev is not a unix client socket");
 944             return;
 945         }
 946
 947         /* if we get a UNIX socket as the parameter we will talk
 948          * to the ivshmem server to receive the memory region */
 949
 950         IVSHMEM_DPRINTF("using shared memory server (socket = %s)\n",
 951                         s->server_chr->filename);
 952
 953         if (ivshmem_has_feature(s, IVSHMEM_MSI) &&
 954             ivshmem_setup_msi(s)) {
 955             error_setg(errp, "msix initialization failed");
 956             return;
 957         }
 958
 959         /* we allocate enough space for 16 peers and grow as needed */
 960         resize_peers(s, 16);
 961         s->vm_id = -1;
 962
 963         pci_register_bar(dev, 2, attr, &s->bar);
 964
 965         s->eventfd_chr = g_malloc0(s->vectors * sizeof(CharDriverState *));
 966
 967         qemu_chr_add_handlers(s->server_chr, ivshmem_can_receive,
 968                               ivshmem_check_version, ivshmem_event, s);
 969     } else {
 970         /* just map the file immediately, we're not using a server */
 971         int fd;
 972
 973         IVSHMEM_DPRINTF("using shm_open (shm object = %s)\n", s->shmobj);
 974
 975         /* try opening with O_EXCL and if it succeeds zero the memory
 976          * by truncating to 0 */
 977         if ((fd = shm_open(s->shmobj, O_CREAT|O_RDWR|O_EXCL,
 978                         S_IRWXU|S_IRWXG|S_IRWXO)) > 0) {
 979            /* truncate file to length PCI device's memory */
 980             if (ftruncate(fd, s->ivshmem_size) != 0) {
 981                 error_report("could not truncate shared file");
 982             }
 983
 984         } else if ((fd = shm_open(s->shmobj, O_CREAT|O_RDWR,
 985                         S_IRWXU|S_IRWXG|S_IRWXO)) < 0) {
 986             error_setg(errp, "could not open shared file");
 987             return;
 988         }
 989
 990         if (check_shm_size(s, fd, errp) == -1) {
 991             return;
 992         }
 993
 994         create_shared_memory_BAR(s, fd, attr, errp);
 995         close(fd);
 996     }
 997 }
 998
 999 static void pci_ivshmem_exit(PCIDevice *dev)
1000 {
1001     IVShmemState *s = IVSHMEM(dev);
1002     int i;
1003
1004     fifo8_destroy(&s->incoming_fifo);
1005
1006     if (s->migration_blocker) {
1007         migrate_del_blocker(s->migration_blocker);
1008         error_free(s->migration_blocker);
1009     }
1010
1011     if (memory_region_is_mapped(&s->ivshmem)) {
1012         if (!s->hostmem) {
1013             void *addr = memory_region_get_ram_ptr(&s->ivshmem);
1014
1015             if (munmap(addr, s->ivshmem_size) == -1) {
1016                 error_report("Failed to munmap shared memory %s",
1017                              strerror(errno));
1018             }
1019         }
1020
1021         vmstate_unregister_ram(&s->ivshmem, DEVICE(dev));
1022         memory_region_del_subregion(&s->bar, &s->ivshmem);
1023     }
1024
1025     if (s->eventfd_chr) {
1026         for (i = 0; i < s->vectors; i++) {
1027             if (s->eventfd_chr[i]) {
1028                 qemu_chr_free(s->eventfd_chr[i]);
1029             }
1030         }
1031         g_free(s->eventfd_chr);
1032     }
1033
1034     if (s->peers) {
1035         for (i = 0; i < s->nb_peers; i++) {
1036             close_peer_eventfds(s, i);
1037         }
1038         g_free(s->peers);
1039     }
1040
1041     if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
1042         msix_uninit_exclusive_bar(dev);
1043     }
1044
1045     g_free(s->msi_vectors);
1046 }
1047
1048 static bool test_msix(void *opaque, int version_id)
1049 {
1050     IVShmemState *s = opaque;
1051
1052     return ivshmem_has_feature(s, IVSHMEM_MSI);
1053 }
1054
1055 static bool test_no_msix(void *opaque, int version_id)
1056 {
1057     return !test_msix(opaque, version_id);
1058 }
1059
1060 static int ivshmem_pre_load(void *opaque)
1061 {
1062     IVShmemState *s = opaque;
1063
1064     if (s->role_val == IVSHMEM_PEER) {
1065         error_report("'peer' devices are not migratable");
1066         return -EINVAL;
1067     }
1068
1069     return 0;
1070 }
1071
1072 static int ivshmem_post_load(void *opaque, int version_id)
1073 {
1074     IVShmemState *s = opaque;
1075
1076     if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
1077         ivshmem_use_msix(s);
1078     }
1079
1080     return 0;
1081 }
1082
1083 static int ivshmem_load_old(QEMUFile *f, void *opaque, int version_id)
1084 {
1085     IVShmemState *s = opaque;
1086     PCIDevice *pdev = PCI_DEVICE(s);
1087     int ret;
1088
1089     IVSHMEM_DPRINTF("ivshmem_load_old\n");
1090
1091     if (version_id != 0) {
1092         return -EINVAL;
1093     }
1094
1095     if (s->role_val == IVSHMEM_PEER) {
1096         error_report("'peer' devices are not migratable");
1097         return -EINVAL;
1098     }
1099
1100     ret = pci_device_load(pdev, f);
1101     if (ret) {
1102         return ret;
1103     }
1104
1105     if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
1106         msix_load(pdev, f);
1107         ivshmem_use_msix(s);
1108     } else {
1109         s->intrstatus = qemu_get_be32(f);
1110         s->intrmask = qemu_get_be32(f);
1111     }
1112
1113     return 0;
1114 }
1115
1116 static const VMStateDescription ivshmem_vmsd = {
1117     .name = "ivshmem",
1118     .version_id = 1,
1119     .minimum_version_id = 1,
1120     .pre_load = ivshmem_pre_load,
1121     .post_load = ivshmem_post_load,
1122     .fields = (VMStateField[]) {
1123         VMSTATE_PCI_DEVICE(parent_obj, IVShmemState),
1124
1125         VMSTATE_MSIX_TEST(parent_obj, IVShmemState, test_msix),
1126         VMSTATE_UINT32_TEST(intrstatus, IVShmemState, test_no_msix),
1127         VMSTATE_UINT32_TEST(intrmask, IVShmemState, test_no_msix),
1128
1129         VMSTATE_END_OF_LIST()
1130     },
1131     .load_state_old = ivshmem_load_old,
1132     .minimum_version_id_old = 0
1133 };
1134
1135 static Property ivshmem_properties[] = {
1136     DEFINE_PROP_CHR("chardev", IVShmemState, server_chr),
1137     DEFINE_PROP_STRING("size", IVShmemState, sizearg),
1138     DEFINE_PROP_UINT32("vectors", IVShmemState, vectors, 1),
1139     DEFINE_PROP_BIT("ioeventfd", IVShmemState, features, IVSHMEM_IOEVENTFD, false),
1140     DEFINE_PROP_BIT("msi", IVShmemState, features, IVSHMEM_MSI, true),
1141     DEFINE_PROP_STRING("shm", IVShmemState, shmobj),
1142     DEFINE_PROP_STRING("role", IVShmemState, role),
1143     DEFINE_PROP_UINT32("use64", IVShmemState, ivshmem_64bit, 1),
1144     DEFINE_PROP_END_OF_LIST(),
1145 };
1146
1147 static void ivshmem_class_init(ObjectClass *klass, void *data)
1148 {
1149     DeviceClass *dc = DEVICE_CLASS(klass);
1150     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
1151
1152     k->realize = pci_ivshmem_realize;
1153     k->exit = pci_ivshmem_exit;
1154     k->config_write = ivshmem_write_config;
1155     k->vendor_id = PCI_VENDOR_ID_IVSHMEM;
1156     k->device_id = PCI_DEVICE_ID_IVSHMEM;
1157     k->class_id = PCI_CLASS_MEMORY_RAM;
1158     dc->reset = ivshmem_reset;
1159     dc->props = ivshmem_properties;
1160     dc->vmsd = &ivshmem_vmsd;
1161     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
1162     dc->desc = "Inter-VM shared memory";
1163 }
1164
1165 static void ivshmem_check_memdev_is_busy(Object *obj, const char *name,
1166                                          Object *val, Error **errp)
1167 {
1168     MemoryRegion *mr;
1169
1170     mr = host_memory_backend_get_memory(MEMORY_BACKEND(val), errp);
1171     if (memory_region_is_mapped(mr)) {
1172         char *path = object_get_canonical_path_component(val);
1173         error_setg(errp, "can't use already busy memdev: %s", path);
1174         g_free(path);
1175     } else {
1176         qdev_prop_allow_set_link_before_realize(obj, name, val, errp);
1177     }
1178 }
1179
1180 static void ivshmem_init(Object *obj)
1181 {
1182     IVShmemState *s = IVSHMEM(obj);
1183
1184     object_property_add_link(obj, IVSHMEM_MEMDEV_PROP, TYPE_MEMORY_BACKEND,
1185                              (Object **)&s->hostmem,
1186                              ivshmem_check_memdev_is_busy,
1187                              OBJ_PROP_LINK_UNREF_ON_RELEASE,
1188                              &error_abort);
1189 }
1190
1191 static const TypeInfo ivshmem_info = {
1192     .name          = TYPE_IVSHMEM,
1193     .parent        = TYPE_PCI_DEVICE,
1194     .instance_size = sizeof(IVShmemState),
1195     .instance_init = ivshmem_init,
1196     .class_init    = ivshmem_class_init,
1197 };
1198
1199 static void ivshmem_register_types(void)
1200 {
1201     type_register_static(&ivshmem_info);
1202 }
1203
1204 type_init(ivshmem_register_types)