hw/vfio/ccw.c

   1 /*
   2  * vfio based subchannel assignment support
   3  *
   4  * Copyright 2017 IBM Corp.
   5  * Copyright 2019 Red Hat, Inc.
   6  *
   7  * Author(s): Dong Jia Shi <bjsdjshi@linux.vnet.ibm.com>
   8  *            Xiao Feng Ren <renxiaof@linux.vnet.ibm.com>
   9  *            Pierre Morel <pmorel@linux.vnet.ibm.com>
  10  *            Cornelia Huck <cohuck@redhat.com>
  11  *
  12  * This work is licensed under the terms of the GNU GPL, version 2 or (at
  13  * your option) any later version. See the COPYING file in the top-level
  14  * directory.
  15  */
  16
  17 #include "qemu/osdep.h"
  18 #include <linux/vfio.h>
  19 #include <linux/vfio_ccw.h>
  20 #include <sys/ioctl.h>
  21
  22 #include "qapi/error.h"
  23 #include "hw/vfio/vfio.h"
  24 #include "hw/vfio/vfio-common.h"
  25 #include "hw/s390x/s390-ccw.h"
  26 #include "hw/s390x/vfio-ccw.h"
  27 #include "hw/qdev-properties.h"
  28 #include "hw/s390x/ccw-device.h"
  29 #include "exec/address-spaces.h"
  30 #include "qemu/error-report.h"
  31 #include "qemu/main-loop.h"
  32 #include "qemu/module.h"
  33
  34 struct VFIOCCWDevice {
  35     S390CCWDevice cdev;
  36     VFIODevice vdev;
  37     uint64_t io_region_size;
  38     uint64_t io_region_offset;
  39     struct ccw_io_region *io_region;
  40     uint64_t async_cmd_region_size;
  41     uint64_t async_cmd_region_offset;
  42     struct ccw_cmd_region *async_cmd_region;
  43     uint64_t schib_region_size;
  44     uint64_t schib_region_offset;
  45     struct ccw_schib_region *schib_region;
  46     uint64_t crw_region_size;
  47     uint64_t crw_region_offset;
  48     struct ccw_crw_region *crw_region;
  49     EventNotifier io_notifier;
  50     EventNotifier crw_notifier;
  51     EventNotifier req_notifier;
  52     bool force_orb_pfch;
  53     bool warned_orb_pfch;
  54 };
  55
  56 static inline void warn_once_pfch(VFIOCCWDevice *vcdev, SubchDev *sch,
  57                                   const char *msg)
  58 {
  59     warn_report_once_cond(&vcdev->warned_orb_pfch,
  60                           "vfio-ccw (devno %x.%x.%04x): %s",
  61                           sch->cssid, sch->ssid, sch->devno, msg);
  62 }
  63
  64 static void vfio_ccw_compute_needs_reset(VFIODevice *vdev)
  65 {
  66     vdev->needs_reset = false;
  67 }
  68
  69 /*
  70  * We don't need vfio_hot_reset_multi and vfio_eoi operations for
  71  * vfio_ccw device now.
  72  */
  73 struct VFIODeviceOps vfio_ccw_ops = {
  74     .vfio_compute_needs_reset = vfio_ccw_compute_needs_reset,
  75 };
  76
  77 static IOInstEnding vfio_ccw_handle_request(SubchDev *sch)
  78 {
  79     S390CCWDevice *cdev = sch->driver_data;
  80     VFIOCCWDevice *vcdev = DO_UPCAST(VFIOCCWDevice, cdev, cdev);
  81     struct ccw_io_region *region = vcdev->io_region;
  82     int ret;
  83
  84     if (!(sch->orb.ctrl0 & ORB_CTRL0_MASK_PFCH) && vcdev->force_orb_pfch) {
  85         sch->orb.ctrl0 |= ORB_CTRL0_MASK_PFCH;
  86         warn_once_pfch(vcdev, sch, "PFCH flag forced");
  87     }
  88
  89     QEMU_BUILD_BUG_ON(sizeof(region->orb_area) != sizeof(ORB));
  90     QEMU_BUILD_BUG_ON(sizeof(region->scsw_area) != sizeof(SCSW));
  91     QEMU_BUILD_BUG_ON(sizeof(region->irb_area) != sizeof(IRB));
  92
  93     memset(region, 0, sizeof(*region));
  94
  95     memcpy(region->orb_area, &sch->orb, sizeof(ORB));
  96     memcpy(region->scsw_area, &sch->curr_status.scsw, sizeof(SCSW));
  97
  98 again:
  99     ret = pwrite(vcdev->vdev.fd, region,
 100                  vcdev->io_region_size, vcdev->io_region_offset);
 101     if (ret != vcdev->io_region_size) {
 102         if (errno == EAGAIN) {
 103             goto again;
 104         }
 105         error_report("vfio-ccw: write I/O region failed with errno=%d", errno);
 106         ret = errno ? -errno : -EFAULT;
 107     } else {
 108         ret = 0;
 109     }
 110     switch (ret) {
 111     case 0:
 112         return IOINST_CC_EXPECTED;
 113     case -EBUSY:
 114         return IOINST_CC_BUSY;
 115     case -ENODEV:
 116     case -EACCES:
 117         return IOINST_CC_NOT_OPERATIONAL;
 118     case -EFAULT:
 119     default:
 120         sch_gen_unit_exception(sch);
 121         css_inject_io_interrupt(sch);
 122         return IOINST_CC_EXPECTED;
 123     }
 124 }
 125
 126 static IOInstEnding vfio_ccw_handle_store(SubchDev *sch)
 127 {
 128     S390CCWDevice *cdev = sch->driver_data;
 129     VFIOCCWDevice *vcdev = DO_UPCAST(VFIOCCWDevice, cdev, cdev);
 130     SCHIB *schib = &sch->curr_status;
 131     struct ccw_schib_region *region = vcdev->schib_region;
 132     SCHIB *s;
 133     int ret;
 134
 135     /* schib region not available so nothing else to do */
 136     if (!region) {
 137         return IOINST_CC_EXPECTED;
 138     }
 139
 140     memset(region, 0, sizeof(*region));
 141     ret = pread(vcdev->vdev.fd, region, vcdev->schib_region_size,
 142                 vcdev->schib_region_offset);
 143
 144     if (ret == -1) {
 145         /*
 146          * Device is probably damaged, but store subchannel does not
 147          * have a nonzero cc defined for this scenario.  Log an error,
 148          * and presume things are otherwise fine.
 149          */
 150         error_report("vfio-ccw: store region read failed with errno=%d", errno);
 151         return IOINST_CC_EXPECTED;
 152     }
 153
 154     /*
 155      * Selectively copy path-related bits of the SCHIB,
 156      * rather than copying the entire struct.
 157      */
 158     s = (SCHIB *)region->schib_area;
 159     schib->pmcw.pnom = s->pmcw.pnom;
 160     schib->pmcw.lpum = s->pmcw.lpum;
 161     schib->pmcw.pam = s->pmcw.pam;
 162     schib->pmcw.pom = s->pmcw.pom;
 163
 164     if (s->scsw.flags & SCSW_FLAGS_MASK_PNO) {
 165         schib->scsw.flags |= SCSW_FLAGS_MASK_PNO;
 166     }
 167
 168     return IOINST_CC_EXPECTED;
 169 }
 170
 171 static int vfio_ccw_handle_clear(SubchDev *sch)
 172 {
 173     S390CCWDevice *cdev = sch->driver_data;
 174     VFIOCCWDevice *vcdev = DO_UPCAST(VFIOCCWDevice, cdev, cdev);
 175     struct ccw_cmd_region *region = vcdev->async_cmd_region;
 176     int ret;
 177
 178     if (!vcdev->async_cmd_region) {
 179         /* Async command region not available, fall back to emulation */
 180         return -ENOSYS;
 181     }
 182
 183     memset(region, 0, sizeof(*region));
 184     region->command = VFIO_CCW_ASYNC_CMD_CSCH;
 185
 186 again:
 187     ret = pwrite(vcdev->vdev.fd, region,
 188                  vcdev->async_cmd_region_size, vcdev->async_cmd_region_offset);
 189     if (ret != vcdev->async_cmd_region_size) {
 190         if (errno == EAGAIN) {
 191             goto again;
 192         }
 193         error_report("vfio-ccw: write cmd region failed with errno=%d", errno);
 194         ret = errno ? -errno : -EFAULT;
 195     } else {
 196         ret = 0;
 197     }
 198     switch (ret) {
 199     case 0:
 200     case -ENODEV:
 201     case -EACCES:
 202         return ret;
 203     case -EFAULT:
 204     default:
 205         sch_gen_unit_exception(sch);
 206         css_inject_io_interrupt(sch);
 207         return 0;
 208     }
 209 }
 210
 211 static int vfio_ccw_handle_halt(SubchDev *sch)
 212 {
 213     S390CCWDevice *cdev = sch->driver_data;
 214     VFIOCCWDevice *vcdev = DO_UPCAST(VFIOCCWDevice, cdev, cdev);
 215     struct ccw_cmd_region *region = vcdev->async_cmd_region;
 216     int ret;
 217
 218     if (!vcdev->async_cmd_region) {
 219         /* Async command region not available, fall back to emulation */
 220         return -ENOSYS;
 221     }
 222
 223     memset(region, 0, sizeof(*region));
 224     region->command = VFIO_CCW_ASYNC_CMD_HSCH;
 225
 226 again:
 227     ret = pwrite(vcdev->vdev.fd, region,
 228                  vcdev->async_cmd_region_size, vcdev->async_cmd_region_offset);
 229     if (ret != vcdev->async_cmd_region_size) {
 230         if (errno == EAGAIN) {
 231             goto again;
 232         }
 233         error_report("vfio-ccw: write cmd region failed with errno=%d", errno);
 234         ret = errno ? -errno : -EFAULT;
 235     } else {
 236         ret = 0;
 237     }
 238     switch (ret) {
 239     case 0:
 240     case -EBUSY:
 241     case -ENODEV:
 242     case -EACCES:
 243         return ret;
 244     case -EFAULT:
 245     default:
 246         sch_gen_unit_exception(sch);
 247         css_inject_io_interrupt(sch);
 248         return 0;
 249     }
 250 }
 251
 252 static void vfio_ccw_reset(DeviceState *dev)
 253 {
 254     CcwDevice *ccw_dev = DO_UPCAST(CcwDevice, parent_obj, dev);
 255     S390CCWDevice *cdev = DO_UPCAST(S390CCWDevice, parent_obj, ccw_dev);
 256     VFIOCCWDevice *vcdev = DO_UPCAST(VFIOCCWDevice, cdev, cdev);
 257
 258     ioctl(vcdev->vdev.fd, VFIO_DEVICE_RESET);
 259 }
 260
 261 static void vfio_ccw_crw_read(VFIOCCWDevice *vcdev)
 262 {
 263     struct ccw_crw_region *region = vcdev->crw_region;
 264     CRW crw;
 265     int size;
 266
 267     /* Keep reading CRWs as long as data is returned */
 268     do {
 269         memset(region, 0, sizeof(*region));
 270         size = pread(vcdev->vdev.fd, region, vcdev->crw_region_size,
 271                      vcdev->crw_region_offset);
 272
 273         if (size == -1) {
 274             error_report("vfio-ccw: Read crw region failed with errno=%d",
 275                          errno);
 276             break;
 277         }
 278
 279         if (region->crw == 0) {
 280             /* No more CRWs to queue */
 281             break;
 282         }
 283
 284         memcpy(&crw, &region->crw, sizeof(CRW));
 285
 286         css_crw_add_to_queue(crw);
 287     } while (1);
 288 }
 289
 290 static void vfio_ccw_req_notifier_handler(void *opaque)
 291 {
 292     VFIOCCWDevice *vcdev = opaque;
 293     Error *err = NULL;
 294
 295     if (!event_notifier_test_and_clear(&vcdev->req_notifier)) {
 296         return;
 297     }
 298
 299     qdev_unplug(DEVICE(vcdev), &err);
 300     if (err) {
 301         warn_reportf_err(err, VFIO_MSG_PREFIX, vcdev->vdev.name);
 302     }
 303 }
 304
 305 static void vfio_ccw_crw_notifier_handler(void *opaque)
 306 {
 307     VFIOCCWDevice *vcdev = opaque;
 308
 309     while (event_notifier_test_and_clear(&vcdev->crw_notifier)) {
 310         vfio_ccw_crw_read(vcdev);
 311     }
 312 }
 313
 314 static void vfio_ccw_io_notifier_handler(void *opaque)
 315 {
 316     VFIOCCWDevice *vcdev = opaque;
 317     struct ccw_io_region *region = vcdev->io_region;
 318     S390CCWDevice *cdev = S390_CCW_DEVICE(vcdev);
 319     CcwDevice *ccw_dev = CCW_DEVICE(cdev);
 320     SubchDev *sch = ccw_dev->sch;
 321     SCHIB *schib = &sch->curr_status;
 322     SCSW s;
 323     IRB irb;
 324     ESW esw;
 325     int size;
 326
 327     if (!event_notifier_test_and_clear(&vcdev->io_notifier)) {
 328         return;
 329     }
 330
 331     size = pread(vcdev->vdev.fd, region, vcdev->io_region_size,
 332                  vcdev->io_region_offset);
 333     if (size == -1) {
 334         switch (errno) {
 335         case ENODEV:
 336             /* Generate a deferred cc 3 condition. */
 337             schib->scsw.flags |= SCSW_FLAGS_MASK_CC;
 338             schib->scsw.ctrl &= ~SCSW_CTRL_MASK_STCTL;
 339             schib->scsw.ctrl |= (SCSW_STCTL_ALERT | SCSW_STCTL_STATUS_PEND);
 340             goto read_err;
 341         case EFAULT:
 342             /* Memory problem, generate channel data check. */
 343             schib->scsw.ctrl &= ~SCSW_ACTL_START_PEND;
 344             schib->scsw.cstat = SCSW_CSTAT_DATA_CHECK;
 345             schib->scsw.ctrl &= ~SCSW_CTRL_MASK_STCTL;
 346             schib->scsw.ctrl |= SCSW_STCTL_PRIMARY | SCSW_STCTL_SECONDARY |
 347                        SCSW_STCTL_ALERT | SCSW_STCTL_STATUS_PEND;
 348             goto read_err;
 349         default:
 350             /* Error, generate channel program check. */
 351             schib->scsw.ctrl &= ~SCSW_ACTL_START_PEND;
 352             schib->scsw.cstat = SCSW_CSTAT_PROG_CHECK;
 353             schib->scsw.ctrl &= ~SCSW_CTRL_MASK_STCTL;
 354             schib->scsw.ctrl |= SCSW_STCTL_PRIMARY | SCSW_STCTL_SECONDARY |
 355                        SCSW_STCTL_ALERT | SCSW_STCTL_STATUS_PEND;
 356             goto read_err;
 357         }
 358     } else if (size != vcdev->io_region_size) {
 359         /* Information transfer error, generate channel-control check. */
 360         schib->scsw.ctrl &= ~SCSW_ACTL_START_PEND;
 361         schib->scsw.cstat = SCSW_CSTAT_CHN_CTRL_CHK;
 362         schib->scsw.ctrl &= ~SCSW_CTRL_MASK_STCTL;
 363         schib->scsw.ctrl |= SCSW_STCTL_PRIMARY | SCSW_STCTL_SECONDARY |
 364                    SCSW_STCTL_ALERT | SCSW_STCTL_STATUS_PEND;
 365         goto read_err;
 366     }
 367
 368     memcpy(&irb, region->irb_area, sizeof(IRB));
 369
 370     /* Update control block via irb. */
 371     s = schib->scsw;
 372     copy_scsw_to_guest(&s, &irb.scsw);
 373     schib->scsw = s;
 374
 375     copy_esw_to_guest(&esw, &irb.esw);
 376     sch->esw = esw;
 377
 378     /* If a uint check is pending, copy sense data. */
 379     if ((schib->scsw.dstat & SCSW_DSTAT_UNIT_CHECK) &&
 380         (schib->pmcw.chars & PMCW_CHARS_MASK_CSENSE)) {
 381         memcpy(sch->sense_data, irb.ecw, sizeof(irb.ecw));
 382     }
 383
 384 read_err:
 385     css_inject_io_interrupt(sch);
 386 }
 387
 388 static void vfio_ccw_register_irq_notifier(VFIOCCWDevice *vcdev,
 389                                            unsigned int irq,
 390                                            Error **errp)
 391 {
 392     VFIODevice *vdev = &vcdev->vdev;
 393     struct vfio_irq_info *irq_info;
 394     size_t argsz;
 395     int fd;
 396     EventNotifier *notifier;
 397     IOHandler *fd_read;
 398
 399     switch (irq) {
 400     case VFIO_CCW_IO_IRQ_INDEX:
 401         notifier = &vcdev->io_notifier;
 402         fd_read = vfio_ccw_io_notifier_handler;
 403         break;
 404     case VFIO_CCW_CRW_IRQ_INDEX:
 405         notifier = &vcdev->crw_notifier;
 406         fd_read = vfio_ccw_crw_notifier_handler;
 407         break;
 408     case VFIO_CCW_REQ_IRQ_INDEX:
 409         notifier = &vcdev->req_notifier;
 410         fd_read = vfio_ccw_req_notifier_handler;
 411         break;
 412     default:
 413         error_setg(errp, "vfio: Unsupported device irq(%d)", irq);
 414         return;
 415     }
 416
 417     if (vdev->num_irqs < irq + 1) {
 418         error_setg(errp, "vfio: IRQ %u not available (number of irqs %u)",
 419                    irq, vdev->num_irqs);
 420         return;
 421     }
 422
 423     argsz = sizeof(*irq_info);
 424     irq_info = g_malloc0(argsz);
 425     irq_info->index = irq;
 426     irq_info->argsz = argsz;
 427     if (ioctl(vdev->fd, VFIO_DEVICE_GET_IRQ_INFO,
 428               irq_info) < 0 || irq_info->count < 1) {
 429         error_setg_errno(errp, errno, "vfio: Error getting irq info");
 430         goto out_free_info;
 431     }
 432
 433     if (event_notifier_init(notifier, 0)) {
 434         error_setg_errno(errp, errno,
 435                          "vfio: Unable to init event notifier for irq (%d)",
 436                          irq);
 437         goto out_free_info;
 438     }
 439
 440     fd = event_notifier_get_fd(notifier);
 441     qemu_set_fd_handler(fd, fd_read, NULL, vcdev);
 442
 443     if (vfio_set_irq_signaling(vdev, irq, 0,
 444                                VFIO_IRQ_SET_ACTION_TRIGGER, fd, errp)) {
 445         qemu_set_fd_handler(fd, NULL, NULL, vcdev);
 446         event_notifier_cleanup(notifier);
 447     }
 448
 449 out_free_info:
 450     g_free(irq_info);
 451 }
 452
 453 static void vfio_ccw_unregister_irq_notifier(VFIOCCWDevice *vcdev,
 454                                              unsigned int irq)
 455 {
 456     Error *err = NULL;
 457     EventNotifier *notifier;
 458
 459     switch (irq) {
 460     case VFIO_CCW_IO_IRQ_INDEX:
 461         notifier = &vcdev->io_notifier;
 462         break;
 463     case VFIO_CCW_CRW_IRQ_INDEX:
 464         notifier = &vcdev->crw_notifier;
 465         break;
 466     case VFIO_CCW_REQ_IRQ_INDEX:
 467         notifier = &vcdev->req_notifier;
 468         break;
 469     default:
 470         error_report("vfio: Unsupported device irq(%d)", irq);
 471         return;
 472     }
 473
 474     if (vfio_set_irq_signaling(&vcdev->vdev, irq, 0,
 475                                VFIO_IRQ_SET_ACTION_TRIGGER, -1, &err)) {
 476         warn_reportf_err(err, VFIO_MSG_PREFIX, vcdev->vdev.name);
 477     }
 478
 479     qemu_set_fd_handler(event_notifier_get_fd(notifier),
 480                         NULL, NULL, vcdev);
 481     event_notifier_cleanup(notifier);
 482 }
 483
 484 static void vfio_ccw_get_region(VFIOCCWDevice *vcdev, Error **errp)
 485 {
 486     VFIODevice *vdev = &vcdev->vdev;
 487     struct vfio_region_info *info;
 488     int ret;
 489
 490     /* Sanity check device */
 491     if (!(vdev->flags & VFIO_DEVICE_FLAGS_CCW)) {
 492         error_setg(errp, "vfio: Um, this isn't a vfio-ccw device");
 493         return;
 494     }
 495
 496     /*
 497      * We always expect at least the I/O region to be present. We also
 498      * may have a variable number of regions governed by capabilities.
 499      */
 500     if (vdev->num_regions < VFIO_CCW_CONFIG_REGION_INDEX + 1) {
 501         error_setg(errp, "vfio: too few regions (%u), expected at least %u",
 502                    vdev->num_regions, VFIO_CCW_CONFIG_REGION_INDEX + 1);
 503         return;
 504     }
 505
 506     ret = vfio_get_region_info(vdev, VFIO_CCW_CONFIG_REGION_INDEX, &info);
 507     if (ret) {
 508         error_setg_errno(errp, -ret, "vfio: Error getting config info");
 509         return;
 510     }
 511
 512     vcdev->io_region_size = info->size;
 513     if (sizeof(*vcdev->io_region) != vcdev->io_region_size) {
 514         error_setg(errp, "vfio: Unexpected size of the I/O region");
 515         goto out_err;
 516     }
 517
 518     vcdev->io_region_offset = info->offset;
 519     vcdev->io_region = g_malloc0(info->size);
 520     g_free(info);
 521
 522     /* check for the optional async command region */
 523     ret = vfio_get_dev_region_info(vdev, VFIO_REGION_TYPE_CCW,
 524                                    VFIO_REGION_SUBTYPE_CCW_ASYNC_CMD, &info);
 525     if (!ret) {
 526         vcdev->async_cmd_region_size = info->size;
 527         if (sizeof(*vcdev->async_cmd_region) != vcdev->async_cmd_region_size) {
 528             error_setg(errp, "vfio: Unexpected size of the async cmd region");
 529             goto out_err;
 530         }
 531         vcdev->async_cmd_region_offset = info->offset;
 532         vcdev->async_cmd_region = g_malloc0(info->size);
 533         g_free(info);
 534     }
 535
 536     ret = vfio_get_dev_region_info(vdev, VFIO_REGION_TYPE_CCW,
 537                                    VFIO_REGION_SUBTYPE_CCW_SCHIB, &info);
 538     if (!ret) {
 539         vcdev->schib_region_size = info->size;
 540         if (sizeof(*vcdev->schib_region) != vcdev->schib_region_size) {
 541             error_setg(errp, "vfio: Unexpected size of the schib region");
 542             goto out_err;
 543         }
 544         vcdev->schib_region_offset = info->offset;
 545         vcdev->schib_region = g_malloc(info->size);
 546         g_free(info);
 547     }
 548
 549     ret = vfio_get_dev_region_info(vdev, VFIO_REGION_TYPE_CCW,
 550                                    VFIO_REGION_SUBTYPE_CCW_CRW, &info);
 551
 552     if (!ret) {
 553         vcdev->crw_region_size = info->size;
 554         if (sizeof(*vcdev->crw_region) != vcdev->crw_region_size) {
 555             error_setg(errp, "vfio: Unexpected size of the CRW region");
 556             goto out_err;
 557         }
 558         vcdev->crw_region_offset = info->offset;
 559         vcdev->crw_region = g_malloc(info->size);
 560         g_free(info);
 561     }
 562
 563     return;
 564
 565 out_err:
 566     g_free(vcdev->crw_region);
 567     g_free(vcdev->schib_region);
 568     g_free(vcdev->async_cmd_region);
 569     g_free(vcdev->io_region);
 570     g_free(info);
 571     return;
 572 }
 573
 574 static void vfio_ccw_put_region(VFIOCCWDevice *vcdev)
 575 {
 576     g_free(vcdev->crw_region);
 577     g_free(vcdev->schib_region);
 578     g_free(vcdev->async_cmd_region);
 579     g_free(vcdev->io_region);
 580 }
 581
 582 static void vfio_ccw_put_device(VFIOCCWDevice *vcdev)
 583 {
 584     g_free(vcdev->vdev.name);
 585     vfio_put_base_device(&vcdev->vdev);
 586 }
 587
 588 static void vfio_ccw_get_device(VFIOGroup *group, VFIOCCWDevice *vcdev,
 589                                 Error **errp)
 590 {
 591     char *name = g_strdup_printf("%x.%x.%04x", vcdev->cdev.hostid.cssid,
 592                                  vcdev->cdev.hostid.ssid,
 593                                  vcdev->cdev.hostid.devid);
 594     VFIODevice *vbasedev;
 595
 596     QLIST_FOREACH(vbasedev, &group->device_list, next) {
 597         if (strcmp(vbasedev->name, name) == 0) {
 598             error_setg(errp, "vfio: subchannel %s has already been attached",
 599                        name);
 600             goto out_err;
 601         }
 602     }
 603
 604     /*
 605      * All vfio-ccw devices are believed to operate in a way compatible with
 606      * discarding of memory in RAM blocks, ie. pages pinned in the host are
 607      * in the current working set of the guest driver and therefore never
 608      * overlap e.g., with pages available to the guest balloon driver.  This
 609      * needs to be set before vfio_get_device() for vfio common to handle
 610      * ram_block_discard_disable().
 611      */
 612     vcdev->vdev.ram_block_discard_allowed = true;
 613
 614     if (vfio_get_device(group, vcdev->cdev.mdevid, &vcdev->vdev, errp)) {
 615         goto out_err;
 616     }
 617
 618     vcdev->vdev.ops = &vfio_ccw_ops;
 619     vcdev->vdev.type = VFIO_DEVICE_TYPE_CCW;
 620     vcdev->vdev.name = name;
 621     vcdev->vdev.dev = &vcdev->cdev.parent_obj.parent_obj;
 622
 623     return;
 624
 625 out_err:
 626     g_free(name);
 627 }
 628
 629 static VFIOGroup *vfio_ccw_get_group(S390CCWDevice *cdev, Error **errp)
 630 {
 631     char *tmp, group_path[PATH_MAX];
 632     ssize_t len;
 633     int groupid;
 634
 635     tmp = g_strdup_printf("/sys/bus/css/devices/%x.%x.%04x/%s/iommu_group",
 636                           cdev->hostid.cssid, cdev->hostid.ssid,
 637                           cdev->hostid.devid, cdev->mdevid);
 638     len = readlink(tmp, group_path, sizeof(group_path));
 639     g_free(tmp);
 640
 641     if (len <= 0 || len >= sizeof(group_path)) {
 642         error_setg(errp, "vfio: no iommu_group found");
 643         return NULL;
 644     }
 645
 646     group_path[len] = 0;
 647
 648     if (sscanf(basename(group_path), "%d", &groupid) != 1) {
 649         error_setg(errp, "vfio: failed to read %s", group_path);
 650         return NULL;
 651     }
 652
 653     return vfio_get_group(groupid, &address_space_memory, errp);
 654 }
 655
 656 static void vfio_ccw_realize(DeviceState *dev, Error **errp)
 657 {
 658     VFIOGroup *group;
 659     CcwDevice *ccw_dev = DO_UPCAST(CcwDevice, parent_obj, dev);
 660     S390CCWDevice *cdev = DO_UPCAST(S390CCWDevice, parent_obj, ccw_dev);
 661     VFIOCCWDevice *vcdev = DO_UPCAST(VFIOCCWDevice, cdev, cdev);
 662     S390CCWDeviceClass *cdc = S390_CCW_DEVICE_GET_CLASS(cdev);
 663     Error *err = NULL;
 664
 665     /* Call the class init function for subchannel. */
 666     if (cdc->realize) {
 667         cdc->realize(cdev, vcdev->vdev.sysfsdev, &err);
 668         if (err) {
 669             goto out_err_propagate;
 670         }
 671     }
 672
 673     group = vfio_ccw_get_group(cdev, &err);
 674     if (!group) {
 675         goto out_group_err;
 676     }
 677
 678     vfio_ccw_get_device(group, vcdev, &err);
 679     if (err) {
 680         goto out_device_err;
 681     }
 682
 683     vfio_ccw_get_region(vcdev, &err);
 684     if (err) {
 685         goto out_region_err;
 686     }
 687
 688     vfio_ccw_register_irq_notifier(vcdev, VFIO_CCW_IO_IRQ_INDEX, &err);
 689     if (err) {
 690         goto out_io_notifier_err;
 691     }
 692
 693     if (vcdev->crw_region) {
 694         vfio_ccw_register_irq_notifier(vcdev, VFIO_CCW_CRW_IRQ_INDEX, &err);
 695         if (err) {
 696             goto out_irq_notifier_err;
 697         }
 698     }
 699
 700     vfio_ccw_register_irq_notifier(vcdev, VFIO_CCW_REQ_IRQ_INDEX, &err);
 701     if (err) {
 702         /*
 703          * Report this error, but do not make it a failing condition.
 704          * Lack of this IRQ in the host does not prevent normal operation.
 705          */
 706         error_report_err(err);
 707     }
 708
 709     return;
 710
 711 out_irq_notifier_err:
 712     vfio_ccw_unregister_irq_notifier(vcdev, VFIO_CCW_REQ_IRQ_INDEX);
 713     vfio_ccw_unregister_irq_notifier(vcdev, VFIO_CCW_CRW_IRQ_INDEX);
 714     vfio_ccw_unregister_irq_notifier(vcdev, VFIO_CCW_IO_IRQ_INDEX);
 715 out_io_notifier_err:
 716     vfio_ccw_put_region(vcdev);
 717 out_region_err:
 718     vfio_ccw_put_device(vcdev);
 719 out_device_err:
 720     vfio_put_group(group);
 721 out_group_err:
 722     if (cdc->unrealize) {
 723         cdc->unrealize(cdev);
 724     }
 725 out_err_propagate:
 726     error_propagate(errp, err);
 727 }
 728
 729 static void vfio_ccw_unrealize(DeviceState *dev)
 730 {
 731     CcwDevice *ccw_dev = DO_UPCAST(CcwDevice, parent_obj, dev);
 732     S390CCWDevice *cdev = DO_UPCAST(S390CCWDevice, parent_obj, ccw_dev);
 733     VFIOCCWDevice *vcdev = DO_UPCAST(VFIOCCWDevice, cdev, cdev);
 734     S390CCWDeviceClass *cdc = S390_CCW_DEVICE_GET_CLASS(cdev);
 735     VFIOGroup *group = vcdev->vdev.group;
 736
 737     vfio_ccw_unregister_irq_notifier(vcdev, VFIO_CCW_REQ_IRQ_INDEX);
 738     vfio_ccw_unregister_irq_notifier(vcdev, VFIO_CCW_CRW_IRQ_INDEX);
 739     vfio_ccw_unregister_irq_notifier(vcdev, VFIO_CCW_IO_IRQ_INDEX);
 740     vfio_ccw_put_region(vcdev);
 741     vfio_ccw_put_device(vcdev);
 742     vfio_put_group(group);
 743
 744     if (cdc->unrealize) {
 745         cdc->unrealize(cdev);
 746     }
 747 }
 748
 749 static Property vfio_ccw_properties[] = {
 750     DEFINE_PROP_STRING("sysfsdev", VFIOCCWDevice, vdev.sysfsdev),
 751     DEFINE_PROP_BOOL("force-orb-pfch", VFIOCCWDevice, force_orb_pfch, false),
 752     DEFINE_PROP_END_OF_LIST(),
 753 };
 754
 755 static const VMStateDescription vfio_ccw_vmstate = {
 756     .name = "vfio-ccw",
 757     .unmigratable = 1,
 758 };
 759
 760 static void vfio_ccw_class_init(ObjectClass *klass, void *data)
 761 {
 762     DeviceClass *dc = DEVICE_CLASS(klass);
 763     S390CCWDeviceClass *cdc = S390_CCW_DEVICE_CLASS(klass);
 764
 765     device_class_set_props(dc, vfio_ccw_properties);
 766     dc->vmsd = &vfio_ccw_vmstate;
 767     dc->desc = "VFIO-based subchannel assignment";
 768     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
 769     dc->realize = vfio_ccw_realize;
 770     dc->unrealize = vfio_ccw_unrealize;
 771     dc->reset = vfio_ccw_reset;
 772
 773     cdc->handle_request = vfio_ccw_handle_request;
 774     cdc->handle_halt = vfio_ccw_handle_halt;
 775     cdc->handle_clear = vfio_ccw_handle_clear;
 776     cdc->handle_store = vfio_ccw_handle_store;
 777 }
 778
 779 static const TypeInfo vfio_ccw_info = {
 780     .name = TYPE_VFIO_CCW,
 781     .parent = TYPE_S390_CCW,
 782     .instance_size = sizeof(VFIOCCWDevice),
 783     .class_init = vfio_ccw_class_init,
 784 };
 785
 786 static void register_vfio_ccw_type(void)
 787 {
 788     type_register_static(&vfio_ccw_info);
 789 }
 790
 791 type_init(register_vfio_ccw_type)