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/sysbus.h"
  24 #include "hw/vfio/vfio.h"
  25 #include "hw/vfio/vfio-common.h"
  26 #include "hw/s390x/s390-ccw.h"
  27 #include "hw/s390x/vfio-ccw.h"
  28 #include "hw/qdev-properties.h"
  29 #include "hw/s390x/ccw-device.h"
  30 #include "exec/address-spaces.h"
  31 #include "qemu/error-report.h"
  32 #include "qemu/main-loop.h"
  33 #include "qemu/module.h"
  34
  35 struct VFIOCCWDevice {
  36     S390CCWDevice cdev;
  37     VFIODevice vdev;
  38     uint64_t io_region_size;
  39     uint64_t io_region_offset;
  40     struct ccw_io_region *io_region;
  41     uint64_t async_cmd_region_size;
  42     uint64_t async_cmd_region_offset;
  43     struct ccw_cmd_region *async_cmd_region;
  44     uint64_t schib_region_size;
  45     uint64_t schib_region_offset;
  46     struct ccw_schib_region *schib_region;
  47     uint64_t crw_region_size;
  48     uint64_t crw_region_offset;
  49     struct ccw_crw_region *crw_region;
  50     EventNotifier io_notifier;
  51     EventNotifier crw_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;
 107     } else {
 108         ret = region->ret_code;
 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;
 195     } else {
 196         ret = region->ret_code;
 197     }
 198     switch (ret) {
 199     case 0:
 200     case -ENODEV:
 201     case -EACCES:
 202         return 0;
 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;
 235     } else {
 236         ret = region->ret_code;
 237     }
 238     switch (ret) {
 239     case 0:
 240     case -EBUSY:
 241     case -ENODEV:
 242     case -EACCES:
 243         return 0;
 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_crw_notifier_handler(void *opaque)
 291 {
 292     VFIOCCWDevice *vcdev = opaque;
 293
 294     while (event_notifier_test_and_clear(&vcdev->crw_notifier)) {
 295         vfio_ccw_crw_read(vcdev);
 296     }
 297 }
 298
 299 static void vfio_ccw_io_notifier_handler(void *opaque)
 300 {
 301     VFIOCCWDevice *vcdev = opaque;
 302     struct ccw_io_region *region = vcdev->io_region;
 303     S390CCWDevice *cdev = S390_CCW_DEVICE(vcdev);
 304     CcwDevice *ccw_dev = CCW_DEVICE(cdev);
 305     SubchDev *sch = ccw_dev->sch;
 306     SCHIB *schib = &sch->curr_status;
 307     SCSW s;
 308     IRB irb;
 309     int size;
 310
 311     if (!event_notifier_test_and_clear(&vcdev->io_notifier)) {
 312         return;
 313     }
 314
 315     size = pread(vcdev->vdev.fd, region, vcdev->io_region_size,
 316                  vcdev->io_region_offset);
 317     if (size == -1) {
 318         switch (errno) {
 319         case ENODEV:
 320             /* Generate a deferred cc 3 condition. */
 321             schib->scsw.flags |= SCSW_FLAGS_MASK_CC;
 322             schib->scsw.ctrl &= ~SCSW_CTRL_MASK_STCTL;
 323             schib->scsw.ctrl |= (SCSW_STCTL_ALERT | SCSW_STCTL_STATUS_PEND);
 324             goto read_err;
 325         case EFAULT:
 326             /* Memory problem, generate channel data check. */
 327             schib->scsw.ctrl &= ~SCSW_ACTL_START_PEND;
 328             schib->scsw.cstat = SCSW_CSTAT_DATA_CHECK;
 329             schib->scsw.ctrl &= ~SCSW_CTRL_MASK_STCTL;
 330             schib->scsw.ctrl |= SCSW_STCTL_PRIMARY | SCSW_STCTL_SECONDARY |
 331                        SCSW_STCTL_ALERT | SCSW_STCTL_STATUS_PEND;
 332             goto read_err;
 333         default:
 334             /* Error, generate channel program check. */
 335             schib->scsw.ctrl &= ~SCSW_ACTL_START_PEND;
 336             schib->scsw.cstat = SCSW_CSTAT_PROG_CHECK;
 337             schib->scsw.ctrl &= ~SCSW_CTRL_MASK_STCTL;
 338             schib->scsw.ctrl |= SCSW_STCTL_PRIMARY | SCSW_STCTL_SECONDARY |
 339                        SCSW_STCTL_ALERT | SCSW_STCTL_STATUS_PEND;
 340             goto read_err;
 341         }
 342     } else if (size != vcdev->io_region_size) {
 343         /* Information transfer error, generate channel-control check. */
 344         schib->scsw.ctrl &= ~SCSW_ACTL_START_PEND;
 345         schib->scsw.cstat = SCSW_CSTAT_CHN_CTRL_CHK;
 346         schib->scsw.ctrl &= ~SCSW_CTRL_MASK_STCTL;
 347         schib->scsw.ctrl |= SCSW_STCTL_PRIMARY | SCSW_STCTL_SECONDARY |
 348                    SCSW_STCTL_ALERT | SCSW_STCTL_STATUS_PEND;
 349         goto read_err;
 350     }
 351
 352     memcpy(&irb, region->irb_area, sizeof(IRB));
 353
 354     /* Update control block via irb. */
 355     s = schib->scsw;
 356     copy_scsw_to_guest(&s, &irb.scsw);
 357     schib->scsw = s;
 358
 359     /* If a uint check is pending, copy sense data. */
 360     if ((schib->scsw.dstat & SCSW_DSTAT_UNIT_CHECK) &&
 361         (schib->pmcw.chars & PMCW_CHARS_MASK_CSENSE)) {
 362         memcpy(sch->sense_data, irb.ecw, sizeof(irb.ecw));
 363     }
 364
 365 read_err:
 366     css_inject_io_interrupt(sch);
 367 }
 368
 369 static void vfio_ccw_register_irq_notifier(VFIOCCWDevice *vcdev,
 370                                            unsigned int irq,
 371                                            Error **errp)
 372 {
 373     VFIODevice *vdev = &vcdev->vdev;
 374     struct vfio_irq_info *irq_info;
 375     size_t argsz;
 376     int fd;
 377     EventNotifier *notifier;
 378     IOHandler *fd_read;
 379
 380     switch (irq) {
 381     case VFIO_CCW_IO_IRQ_INDEX:
 382         notifier = &vcdev->io_notifier;
 383         fd_read = vfio_ccw_io_notifier_handler;
 384         break;
 385     case VFIO_CCW_CRW_IRQ_INDEX:
 386         notifier = &vcdev->crw_notifier;
 387         fd_read = vfio_ccw_crw_notifier_handler;
 388         break;
 389     default:
 390         error_setg(errp, "vfio: Unsupported device irq(%d)", irq);
 391         return;
 392     }
 393
 394     if (vdev->num_irqs < irq + 1) {
 395         error_setg(errp, "vfio: unexpected number of irqs %u",
 396                    vdev->num_irqs);
 397         return;
 398     }
 399
 400     argsz = sizeof(*irq_info);
 401     irq_info = g_malloc0(argsz);
 402     irq_info->index = irq;
 403     irq_info->argsz = argsz;
 404     if (ioctl(vdev->fd, VFIO_DEVICE_GET_IRQ_INFO,
 405               irq_info) < 0 || irq_info->count < 1) {
 406         error_setg_errno(errp, errno, "vfio: Error getting irq info");
 407         goto out_free_info;
 408     }
 409
 410     if (event_notifier_init(notifier, 0)) {
 411         error_setg_errno(errp, errno,
 412                          "vfio: Unable to init event notifier for irq (%d)",
 413                          irq);
 414         goto out_free_info;
 415     }
 416
 417     fd = event_notifier_get_fd(notifier);
 418     qemu_set_fd_handler(fd, fd_read, NULL, vcdev);
 419
 420     if (vfio_set_irq_signaling(vdev, irq, 0,
 421                                VFIO_IRQ_SET_ACTION_TRIGGER, fd, errp)) {
 422         qemu_set_fd_handler(fd, NULL, NULL, vcdev);
 423         event_notifier_cleanup(notifier);
 424     }
 425
 426 out_free_info:
 427     g_free(irq_info);
 428 }
 429
 430 static void vfio_ccw_unregister_irq_notifier(VFIOCCWDevice *vcdev,
 431                                              unsigned int irq)
 432 {
 433     Error *err = NULL;
 434     EventNotifier *notifier;
 435
 436     switch (irq) {
 437     case VFIO_CCW_IO_IRQ_INDEX:
 438         notifier = &vcdev->io_notifier;
 439         break;
 440     case VFIO_CCW_CRW_IRQ_INDEX:
 441         notifier = &vcdev->crw_notifier;
 442         break;
 443     default:
 444         error_report("vfio: Unsupported device irq(%d)", irq);
 445         return;
 446     }
 447
 448     if (vfio_set_irq_signaling(&vcdev->vdev, irq, 0,
 449                                VFIO_IRQ_SET_ACTION_TRIGGER, -1, &err)) {
 450         error_reportf_err(err, VFIO_MSG_PREFIX, vcdev->vdev.name);
 451     }
 452
 453     qemu_set_fd_handler(event_notifier_get_fd(notifier),
 454                         NULL, NULL, vcdev);
 455     event_notifier_cleanup(notifier);
 456 }
 457
 458 static void vfio_ccw_get_region(VFIOCCWDevice *vcdev, Error **errp)
 459 {
 460     VFIODevice *vdev = &vcdev->vdev;
 461     struct vfio_region_info *info;
 462     int ret;
 463
 464     /* Sanity check device */
 465     if (!(vdev->flags & VFIO_DEVICE_FLAGS_CCW)) {
 466         error_setg(errp, "vfio: Um, this isn't a vfio-ccw device");
 467         return;
 468     }
 469
 470     /*
 471      * We always expect at least the I/O region to be present. We also
 472      * may have a variable number of regions governed by capabilities.
 473      */
 474     if (vdev->num_regions < VFIO_CCW_CONFIG_REGION_INDEX + 1) {
 475         error_setg(errp, "vfio: too few regions (%u), expected at least %u",
 476                    vdev->num_regions, VFIO_CCW_CONFIG_REGION_INDEX + 1);
 477         return;
 478     }
 479
 480     ret = vfio_get_region_info(vdev, VFIO_CCW_CONFIG_REGION_INDEX, &info);
 481     if (ret) {
 482         error_setg_errno(errp, -ret, "vfio: Error getting config info");
 483         return;
 484     }
 485
 486     vcdev->io_region_size = info->size;
 487     if (sizeof(*vcdev->io_region) != vcdev->io_region_size) {
 488         error_setg(errp, "vfio: Unexpected size of the I/O region");
 489         goto out_err;
 490     }
 491
 492     vcdev->io_region_offset = info->offset;
 493     vcdev->io_region = g_malloc0(info->size);
 494
 495     /* check for the optional async command region */
 496     ret = vfio_get_dev_region_info(vdev, VFIO_REGION_TYPE_CCW,
 497                                    VFIO_REGION_SUBTYPE_CCW_ASYNC_CMD, &info);
 498     if (!ret) {
 499         vcdev->async_cmd_region_size = info->size;
 500         if (sizeof(*vcdev->async_cmd_region) != vcdev->async_cmd_region_size) {
 501             error_setg(errp, "vfio: Unexpected size of the async cmd region");
 502             goto out_err;
 503         }
 504         vcdev->async_cmd_region_offset = info->offset;
 505         vcdev->async_cmd_region = g_malloc0(info->size);
 506     }
 507
 508     ret = vfio_get_dev_region_info(vdev, VFIO_REGION_TYPE_CCW,
 509                                    VFIO_REGION_SUBTYPE_CCW_SCHIB, &info);
 510     if (!ret) {
 511         vcdev->schib_region_size = info->size;
 512         if (sizeof(*vcdev->schib_region) != vcdev->schib_region_size) {
 513             error_setg(errp, "vfio: Unexpected size of the schib region");
 514             goto out_err;
 515         }
 516         vcdev->schib_region_offset = info->offset;
 517         vcdev->schib_region = g_malloc(info->size);
 518     }
 519
 520     ret = vfio_get_dev_region_info(vdev, VFIO_REGION_TYPE_CCW,
 521                                    VFIO_REGION_SUBTYPE_CCW_CRW, &info);
 522
 523     if (!ret) {
 524         vcdev->crw_region_size = info->size;
 525         if (sizeof(*vcdev->crw_region) != vcdev->crw_region_size) {
 526             error_setg(errp, "vfio: Unexpected size of the CRW region");
 527             goto out_err;
 528         }
 529         vcdev->crw_region_offset = info->offset;
 530         vcdev->crw_region = g_malloc(info->size);
 531     }
 532
 533     g_free(info);
 534     return;
 535
 536 out_err:
 537     g_free(vcdev->crw_region);
 538     g_free(vcdev->schib_region);
 539     g_free(vcdev->async_cmd_region);
 540     g_free(vcdev->io_region);
 541     g_free(info);
 542     return;
 543 }
 544
 545 static void vfio_ccw_put_region(VFIOCCWDevice *vcdev)
 546 {
 547     g_free(vcdev->crw_region);
 548     g_free(vcdev->schib_region);
 549     g_free(vcdev->async_cmd_region);
 550     g_free(vcdev->io_region);
 551 }
 552
 553 static void vfio_ccw_put_device(VFIOCCWDevice *vcdev)
 554 {
 555     g_free(vcdev->vdev.name);
 556     vfio_put_base_device(&vcdev->vdev);
 557 }
 558
 559 static void vfio_ccw_get_device(VFIOGroup *group, VFIOCCWDevice *vcdev,
 560                                 Error **errp)
 561 {
 562     char *name = g_strdup_printf("%x.%x.%04x", vcdev->cdev.hostid.cssid,
 563                                  vcdev->cdev.hostid.ssid,
 564                                  vcdev->cdev.hostid.devid);
 565     VFIODevice *vbasedev;
 566
 567     QLIST_FOREACH(vbasedev, &group->device_list, next) {
 568         if (strcmp(vbasedev->name, name) == 0) {
 569             error_setg(errp, "vfio: subchannel %s has already been attached",
 570                        name);
 571             goto out_err;
 572         }
 573     }
 574
 575     /*
 576      * All vfio-ccw devices are believed to operate in a way compatible with
 577      * memory ballooning, ie. pages pinned in the host are in the current
 578      * working set of the guest driver and therefore never overlap with pages
 579      * available to the guest balloon driver.  This needs to be set before
 580      * vfio_get_device() for vfio common to handle the balloon inhibitor.
 581      */
 582     vcdev->vdev.balloon_allowed = true;
 583
 584     if (vfio_get_device(group, vcdev->cdev.mdevid, &vcdev->vdev, errp)) {
 585         goto out_err;
 586     }
 587
 588     vcdev->vdev.ops = &vfio_ccw_ops;
 589     vcdev->vdev.type = VFIO_DEVICE_TYPE_CCW;
 590     vcdev->vdev.name = name;
 591     vcdev->vdev.dev = &vcdev->cdev.parent_obj.parent_obj;
 592
 593     return;
 594
 595 out_err:
 596     g_free(name);
 597 }
 598
 599 static VFIOGroup *vfio_ccw_get_group(S390CCWDevice *cdev, Error **errp)
 600 {
 601     char *tmp, group_path[PATH_MAX];
 602     ssize_t len;
 603     int groupid;
 604
 605     tmp = g_strdup_printf("/sys/bus/css/devices/%x.%x.%04x/%s/iommu_group",
 606                           cdev->hostid.cssid, cdev->hostid.ssid,
 607                           cdev->hostid.devid, cdev->mdevid);
 608     len = readlink(tmp, group_path, sizeof(group_path));
 609     g_free(tmp);
 610
 611     if (len <= 0 || len >= sizeof(group_path)) {
 612         error_setg(errp, "vfio: no iommu_group found");
 613         return NULL;
 614     }
 615
 616     group_path[len] = 0;
 617
 618     if (sscanf(basename(group_path), "%d", &groupid) != 1) {
 619         error_setg(errp, "vfio: failed to read %s", group_path);
 620         return NULL;
 621     }
 622
 623     return vfio_get_group(groupid, &address_space_memory, errp);
 624 }
 625
 626 static void vfio_ccw_realize(DeviceState *dev, Error **errp)
 627 {
 628     VFIOGroup *group;
 629     CcwDevice *ccw_dev = DO_UPCAST(CcwDevice, parent_obj, dev);
 630     S390CCWDevice *cdev = DO_UPCAST(S390CCWDevice, parent_obj, ccw_dev);
 631     VFIOCCWDevice *vcdev = DO_UPCAST(VFIOCCWDevice, cdev, cdev);
 632     S390CCWDeviceClass *cdc = S390_CCW_DEVICE_GET_CLASS(cdev);
 633     Error *err = NULL;
 634
 635     /* Call the class init function for subchannel. */
 636     if (cdc->realize) {
 637         cdc->realize(cdev, vcdev->vdev.sysfsdev, &err);
 638         if (err) {
 639             goto out_err_propagate;
 640         }
 641     }
 642
 643     group = vfio_ccw_get_group(cdev, &err);
 644     if (!group) {
 645         goto out_group_err;
 646     }
 647
 648     vfio_ccw_get_device(group, vcdev, &err);
 649     if (err) {
 650         goto out_device_err;
 651     }
 652
 653     vfio_ccw_get_region(vcdev, &err);
 654     if (err) {
 655         goto out_region_err;
 656     }
 657
 658     vfio_ccw_register_irq_notifier(vcdev, VFIO_CCW_IO_IRQ_INDEX, &err);
 659     if (err) {
 660         goto out_notifier_err;
 661     }
 662
 663     if (vcdev->crw_region) {
 664         vfio_ccw_register_irq_notifier(vcdev, VFIO_CCW_CRW_IRQ_INDEX, &err);
 665         if (err) {
 666             vfio_ccw_unregister_irq_notifier(vcdev, VFIO_CCW_IO_IRQ_INDEX);
 667             goto out_notifier_err;
 668         }
 669     }
 670
 671     return;
 672
 673 out_notifier_err:
 674     vfio_ccw_put_region(vcdev);
 675 out_region_err:
 676     vfio_ccw_put_device(vcdev);
 677 out_device_err:
 678     vfio_put_group(group);
 679 out_group_err:
 680     if (cdc->unrealize) {
 681         cdc->unrealize(cdev);
 682     }
 683 out_err_propagate:
 684     error_propagate(errp, err);
 685 }
 686
 687 static void vfio_ccw_unrealize(DeviceState *dev)
 688 {
 689     CcwDevice *ccw_dev = DO_UPCAST(CcwDevice, parent_obj, dev);
 690     S390CCWDevice *cdev = DO_UPCAST(S390CCWDevice, parent_obj, ccw_dev);
 691     VFIOCCWDevice *vcdev = DO_UPCAST(VFIOCCWDevice, cdev, cdev);
 692     S390CCWDeviceClass *cdc = S390_CCW_DEVICE_GET_CLASS(cdev);
 693     VFIOGroup *group = vcdev->vdev.group;
 694
 695     vfio_ccw_unregister_irq_notifier(vcdev, VFIO_CCW_CRW_IRQ_INDEX);
 696     vfio_ccw_unregister_irq_notifier(vcdev, VFIO_CCW_IO_IRQ_INDEX);
 697     vfio_ccw_put_region(vcdev);
 698     vfio_ccw_put_device(vcdev);
 699     vfio_put_group(group);
 700
 701     if (cdc->unrealize) {
 702         cdc->unrealize(cdev);
 703     }
 704 }
 705
 706 static Property vfio_ccw_properties[] = {
 707     DEFINE_PROP_STRING("sysfsdev", VFIOCCWDevice, vdev.sysfsdev),
 708     DEFINE_PROP_BOOL("force-orb-pfch", VFIOCCWDevice, force_orb_pfch, false),
 709     DEFINE_PROP_END_OF_LIST(),
 710 };
 711
 712 static const VMStateDescription vfio_ccw_vmstate = {
 713     .name = "vfio-ccw",
 714     .unmigratable = 1,
 715 };
 716
 717 static void vfio_ccw_class_init(ObjectClass *klass, void *data)
 718 {
 719     DeviceClass *dc = DEVICE_CLASS(klass);
 720     S390CCWDeviceClass *cdc = S390_CCW_DEVICE_CLASS(klass);
 721
 722     device_class_set_props(dc, vfio_ccw_properties);
 723     dc->vmsd = &vfio_ccw_vmstate;
 724     dc->desc = "VFIO-based subchannel assignment";
 725     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
 726     dc->realize = vfio_ccw_realize;
 727     dc->unrealize = vfio_ccw_unrealize;
 728     dc->reset = vfio_ccw_reset;
 729
 730     cdc->handle_request = vfio_ccw_handle_request;
 731     cdc->handle_halt = vfio_ccw_handle_halt;
 732     cdc->handle_clear = vfio_ccw_handle_clear;
 733     cdc->handle_store = vfio_ccw_handle_store;
 734 }
 735
 736 static const TypeInfo vfio_ccw_info = {
 737     .name = TYPE_VFIO_CCW,
 738     .parent = TYPE_S390_CCW,
 739     .instance_size = sizeof(VFIOCCWDevice),
 740     .class_init = vfio_ccw_class_init,
 741 };
 742
 743 static void register_vfio_ccw_type(void)
 744 {
 745     type_register_static(&vfio_ccw_info);
 746 }
 747
 748 type_init(register_vfio_ccw_type)