1 /* 2 * vfio based subchannel assignment support 3 * 4 * Copyright 2017 IBM Corp. 5 * Author(s): Dong Jia Shi <bjsdjshi@linux.vnet.ibm.com> 6 * Xiao Feng Ren <renxiaof@linux.vnet.ibm.com> 7 * Pierre Morel <pmorel@linux.vnet.ibm.com> 8 * 9 * This work is licensed under the terms of the GNU GPL, version 2 or (at 10 * your option) any later version. See the COPYING file in the top-level 11 * directory. 12 */ 13 14 #include "qemu/osdep.h" 15 #include <linux/vfio.h> 16 #include <linux/vfio_ccw.h> 17 #include <sys/ioctl.h> 18 19 #include "qapi/error.h" 20 #include "hw/sysbus.h" 21 #include "hw/vfio/vfio.h" 22 #include "hw/vfio/vfio-common.h" 23 #include "hw/s390x/s390-ccw.h" 24 #include "hw/s390x/vfio-ccw.h" 25 #include "hw/s390x/ccw-device.h" 26 #include "exec/address-spaces.h" 27 #include "qemu/error-report.h" 28 #include "qemu/module.h" 29 30 struct VFIOCCWDevice { 31 S390CCWDevice cdev; 32 VFIODevice vdev; 33 uint64_t io_region_size; 34 uint64_t io_region_offset; 35 struct ccw_io_region *io_region; 36 EventNotifier io_notifier; 37 bool force_orb_pfch; 38 bool warned_orb_pfch; 39 }; 40 41 static inline void warn_once_pfch(VFIOCCWDevice *vcdev, SubchDev *sch, 42 const char *msg) 43 { 44 warn_report_once_cond(&vcdev->warned_orb_pfch, 45 "vfio-ccw (devno %x.%x.%04x): %s", 46 sch->cssid, sch->ssid, sch->devno, msg); 47 } 48 49 static void vfio_ccw_compute_needs_reset(VFIODevice *vdev) 50 { 51 vdev->needs_reset = false; 52 } 53 54 /* 55 * We don't need vfio_hot_reset_multi and vfio_eoi operations for 56 * vfio_ccw device now. 57 */ 58 struct VFIODeviceOps vfio_ccw_ops = { 59 .vfio_compute_needs_reset = vfio_ccw_compute_needs_reset, 60 }; 61 62 static IOInstEnding vfio_ccw_handle_request(SubchDev *sch) 63 { 64 S390CCWDevice *cdev = sch->driver_data; 65 VFIOCCWDevice *vcdev = DO_UPCAST(VFIOCCWDevice, cdev, cdev); 66 struct ccw_io_region *region = vcdev->io_region; 67 int ret; 68 69 if (!(sch->orb.ctrl0 & ORB_CTRL0_MASK_PFCH)) { 70 if (!(vcdev->force_orb_pfch)) { 71 warn_once_pfch(vcdev, sch, "requires PFCH flag set"); 72 sch_gen_unit_exception(sch); 73 css_inject_io_interrupt(sch); 74 return IOINST_CC_EXPECTED; 75 } else { 76 sch->orb.ctrl0 |= ORB_CTRL0_MASK_PFCH; 77 warn_once_pfch(vcdev, sch, "PFCH flag forced"); 78 } 79 } 80 81 QEMU_BUILD_BUG_ON(sizeof(region->orb_area) != sizeof(ORB)); 82 QEMU_BUILD_BUG_ON(sizeof(region->scsw_area) != sizeof(SCSW)); 83 QEMU_BUILD_BUG_ON(sizeof(region->irb_area) != sizeof(IRB)); 84 85 memset(region, 0, sizeof(*region)); 86 87 memcpy(region->orb_area, &sch->orb, sizeof(ORB)); 88 memcpy(region->scsw_area, &sch->curr_status.scsw, sizeof(SCSW)); 89 90 again: 91 ret = pwrite(vcdev->vdev.fd, region, 92 vcdev->io_region_size, vcdev->io_region_offset); 93 if (ret != vcdev->io_region_size) { 94 if (errno == EAGAIN) { 95 goto again; 96 } 97 error_report("vfio-ccw: wirte I/O region failed with errno=%d", errno); 98 ret = -errno; 99 } else { 100 ret = region->ret_code; 101 } 102 switch (ret) { 103 case 0: 104 return IOINST_CC_EXPECTED; 105 case -EBUSY: 106 return IOINST_CC_BUSY; 107 case -ENODEV: 108 case -EACCES: 109 return IOINST_CC_NOT_OPERATIONAL; 110 case -EFAULT: 111 default: 112 sch_gen_unit_exception(sch); 113 css_inject_io_interrupt(sch); 114 return IOINST_CC_EXPECTED; 115 } 116 } 117 118 static void vfio_ccw_reset(DeviceState *dev) 119 { 120 CcwDevice *ccw_dev = DO_UPCAST(CcwDevice, parent_obj, dev); 121 S390CCWDevice *cdev = DO_UPCAST(S390CCWDevice, parent_obj, ccw_dev); 122 VFIOCCWDevice *vcdev = DO_UPCAST(VFIOCCWDevice, cdev, cdev); 123 124 ioctl(vcdev->vdev.fd, VFIO_DEVICE_RESET); 125 } 126 127 static void vfio_ccw_io_notifier_handler(void *opaque) 128 { 129 VFIOCCWDevice *vcdev = opaque; 130 struct ccw_io_region *region = vcdev->io_region; 131 S390CCWDevice *cdev = S390_CCW_DEVICE(vcdev); 132 CcwDevice *ccw_dev = CCW_DEVICE(cdev); 133 SubchDev *sch = ccw_dev->sch; 134 SCHIB *schib = &sch->curr_status; 135 SCSW s; 136 IRB irb; 137 int size; 138 139 if (!event_notifier_test_and_clear(&vcdev->io_notifier)) { 140 return; 141 } 142 143 size = pread(vcdev->vdev.fd, region, vcdev->io_region_size, 144 vcdev->io_region_offset); 145 if (size == -1) { 146 switch (errno) { 147 case ENODEV: 148 /* Generate a deferred cc 3 condition. */ 149 schib->scsw.flags |= SCSW_FLAGS_MASK_CC; 150 schib->scsw.ctrl &= ~SCSW_CTRL_MASK_STCTL; 151 schib->scsw.ctrl |= (SCSW_STCTL_ALERT | SCSW_STCTL_STATUS_PEND); 152 goto read_err; 153 case EFAULT: 154 /* Memory problem, generate channel data check. */ 155 schib->scsw.ctrl &= ~SCSW_ACTL_START_PEND; 156 schib->scsw.cstat = SCSW_CSTAT_DATA_CHECK; 157 schib->scsw.ctrl &= ~SCSW_CTRL_MASK_STCTL; 158 schib->scsw.ctrl |= SCSW_STCTL_PRIMARY | SCSW_STCTL_SECONDARY | 159 SCSW_STCTL_ALERT | SCSW_STCTL_STATUS_PEND; 160 goto read_err; 161 default: 162 /* Error, generate channel program check. */ 163 schib->scsw.ctrl &= ~SCSW_ACTL_START_PEND; 164 schib->scsw.cstat = SCSW_CSTAT_PROG_CHECK; 165 schib->scsw.ctrl &= ~SCSW_CTRL_MASK_STCTL; 166 schib->scsw.ctrl |= SCSW_STCTL_PRIMARY | SCSW_STCTL_SECONDARY | 167 SCSW_STCTL_ALERT | SCSW_STCTL_STATUS_PEND; 168 goto read_err; 169 } 170 } else if (size != vcdev->io_region_size) { 171 /* Information transfer error, generate channel-control check. */ 172 schib->scsw.ctrl &= ~SCSW_ACTL_START_PEND; 173 schib->scsw.cstat = SCSW_CSTAT_CHN_CTRL_CHK; 174 schib->scsw.ctrl &= ~SCSW_CTRL_MASK_STCTL; 175 schib->scsw.ctrl |= SCSW_STCTL_PRIMARY | SCSW_STCTL_SECONDARY | 176 SCSW_STCTL_ALERT | SCSW_STCTL_STATUS_PEND; 177 goto read_err; 178 } 179 180 memcpy(&irb, region->irb_area, sizeof(IRB)); 181 182 /* Update control block via irb. */ 183 s = schib->scsw; 184 copy_scsw_to_guest(&s, &irb.scsw); 185 schib->scsw = s; 186 187 /* If a uint check is pending, copy sense data. */ 188 if ((schib->scsw.dstat & SCSW_DSTAT_UNIT_CHECK) && 189 (schib->pmcw.chars & PMCW_CHARS_MASK_CSENSE)) { 190 memcpy(sch->sense_data, irb.ecw, sizeof(irb.ecw)); 191 } 192 193 read_err: 194 css_inject_io_interrupt(sch); 195 } 196 197 static void vfio_ccw_register_io_notifier(VFIOCCWDevice *vcdev, Error **errp) 198 { 199 VFIODevice *vdev = &vcdev->vdev; 200 struct vfio_irq_info *irq_info; 201 struct vfio_irq_set *irq_set; 202 size_t argsz; 203 int32_t *pfd; 204 205 if (vdev->num_irqs < VFIO_CCW_IO_IRQ_INDEX + 1) { 206 error_setg(errp, "vfio: unexpected number of io irqs %u", 207 vdev->num_irqs); 208 return; 209 } 210 211 argsz = sizeof(*irq_info); 212 irq_info = g_malloc0(argsz); 213 irq_info->index = VFIO_CCW_IO_IRQ_INDEX; 214 irq_info->argsz = argsz; 215 if (ioctl(vdev->fd, VFIO_DEVICE_GET_IRQ_INFO, 216 irq_info) < 0 || irq_info->count < 1) { 217 error_setg_errno(errp, errno, "vfio: Error getting irq info"); 218 goto out_free_info; 219 } 220 221 if (event_notifier_init(&vcdev->io_notifier, 0)) { 222 error_setg_errno(errp, errno, 223 "vfio: Unable to init event notifier for IO"); 224 goto out_free_info; 225 } 226 227 argsz = sizeof(*irq_set) + sizeof(*pfd); 228 irq_set = g_malloc0(argsz); 229 irq_set->argsz = argsz; 230 irq_set->flags = VFIO_IRQ_SET_DATA_EVENTFD | 231 VFIO_IRQ_SET_ACTION_TRIGGER; 232 irq_set->index = VFIO_CCW_IO_IRQ_INDEX; 233 irq_set->start = 0; 234 irq_set->count = 1; 235 pfd = (int32_t *) &irq_set->data; 236 237 *pfd = event_notifier_get_fd(&vcdev->io_notifier); 238 qemu_set_fd_handler(*pfd, vfio_ccw_io_notifier_handler, NULL, vcdev); 239 if (ioctl(vdev->fd, VFIO_DEVICE_SET_IRQS, irq_set)) { 240 error_setg(errp, "vfio: Failed to set up io notification"); 241 qemu_set_fd_handler(*pfd, NULL, NULL, vcdev); 242 event_notifier_cleanup(&vcdev->io_notifier); 243 } 244 245 g_free(irq_set); 246 247 out_free_info: 248 g_free(irq_info); 249 } 250 251 static void vfio_ccw_unregister_io_notifier(VFIOCCWDevice *vcdev) 252 { 253 struct vfio_irq_set *irq_set; 254 size_t argsz; 255 int32_t *pfd; 256 257 argsz = sizeof(*irq_set) + sizeof(*pfd); 258 irq_set = g_malloc0(argsz); 259 irq_set->argsz = argsz; 260 irq_set->flags = VFIO_IRQ_SET_DATA_EVENTFD | 261 VFIO_IRQ_SET_ACTION_TRIGGER; 262 irq_set->index = VFIO_CCW_IO_IRQ_INDEX; 263 irq_set->start = 0; 264 irq_set->count = 1; 265 pfd = (int32_t *) &irq_set->data; 266 *pfd = -1; 267 268 if (ioctl(vcdev->vdev.fd, VFIO_DEVICE_SET_IRQS, irq_set)) { 269 error_report("vfio: Failed to de-assign device io fd: %m"); 270 } 271 272 qemu_set_fd_handler(event_notifier_get_fd(&vcdev->io_notifier), 273 NULL, NULL, vcdev); 274 event_notifier_cleanup(&vcdev->io_notifier); 275 276 g_free(irq_set); 277 } 278 279 static void vfio_ccw_get_region(VFIOCCWDevice *vcdev, Error **errp) 280 { 281 VFIODevice *vdev = &vcdev->vdev; 282 struct vfio_region_info *info; 283 int ret; 284 285 /* Sanity check device */ 286 if (!(vdev->flags & VFIO_DEVICE_FLAGS_CCW)) { 287 error_setg(errp, "vfio: Um, this isn't a vfio-ccw device"); 288 return; 289 } 290 291 if (vdev->num_regions < VFIO_CCW_CONFIG_REGION_INDEX + 1) { 292 error_setg(errp, "vfio: Unexpected number of the I/O region %u", 293 vdev->num_regions); 294 return; 295 } 296 297 ret = vfio_get_region_info(vdev, VFIO_CCW_CONFIG_REGION_INDEX, &info); 298 if (ret) { 299 error_setg_errno(errp, -ret, "vfio: Error getting config info"); 300 return; 301 } 302 303 vcdev->io_region_size = info->size; 304 if (sizeof(*vcdev->io_region) != vcdev->io_region_size) { 305 error_setg(errp, "vfio: Unexpected size of the I/O region"); 306 g_free(info); 307 return; 308 } 309 310 vcdev->io_region_offset = info->offset; 311 vcdev->io_region = g_malloc0(info->size); 312 313 g_free(info); 314 } 315 316 static void vfio_ccw_put_region(VFIOCCWDevice *vcdev) 317 { 318 g_free(vcdev->io_region); 319 } 320 321 static void vfio_ccw_put_device(VFIOCCWDevice *vcdev) 322 { 323 g_free(vcdev->vdev.name); 324 vfio_put_base_device(&vcdev->vdev); 325 } 326 327 static void vfio_ccw_get_device(VFIOGroup *group, VFIOCCWDevice *vcdev, 328 Error **errp) 329 { 330 char *name = g_strdup_printf("%x.%x.%04x", vcdev->cdev.hostid.cssid, 331 vcdev->cdev.hostid.ssid, 332 vcdev->cdev.hostid.devid); 333 VFIODevice *vbasedev; 334 335 QLIST_FOREACH(vbasedev, &group->device_list, next) { 336 if (strcmp(vbasedev->name, name) == 0) { 337 error_setg(errp, "vfio: subchannel %s has already been attached", 338 name); 339 goto out_err; 340 } 341 } 342 343 /* 344 * All vfio-ccw devices are believed to operate in a way compatible with 345 * memory ballooning, ie. pages pinned in the host are in the current 346 * working set of the guest driver and therefore never overlap with pages 347 * available to the guest balloon driver. This needs to be set before 348 * vfio_get_device() for vfio common to handle the balloon inhibitor. 349 */ 350 vcdev->vdev.balloon_allowed = true; 351 352 if (vfio_get_device(group, vcdev->cdev.mdevid, &vcdev->vdev, errp)) { 353 goto out_err; 354 } 355 356 vcdev->vdev.ops = &vfio_ccw_ops; 357 vcdev->vdev.type = VFIO_DEVICE_TYPE_CCW; 358 vcdev->vdev.name = name; 359 vcdev->vdev.dev = &vcdev->cdev.parent_obj.parent_obj; 360 361 return; 362 363 out_err: 364 g_free(name); 365 } 366 367 static VFIOGroup *vfio_ccw_get_group(S390CCWDevice *cdev, Error **errp) 368 { 369 char *tmp, group_path[PATH_MAX]; 370 ssize_t len; 371 int groupid; 372 373 tmp = g_strdup_printf("/sys/bus/css/devices/%x.%x.%04x/%s/iommu_group", 374 cdev->hostid.cssid, cdev->hostid.ssid, 375 cdev->hostid.devid, cdev->mdevid); 376 len = readlink(tmp, group_path, sizeof(group_path)); 377 g_free(tmp); 378 379 if (len <= 0 || len >= sizeof(group_path)) { 380 error_setg(errp, "vfio: no iommu_group found"); 381 return NULL; 382 } 383 384 group_path[len] = 0; 385 386 if (sscanf(basename(group_path), "%d", &groupid) != 1) { 387 error_setg(errp, "vfio: failed to read %s", group_path); 388 return NULL; 389 } 390 391 return vfio_get_group(groupid, &address_space_memory, errp); 392 } 393 394 static void vfio_ccw_realize(DeviceState *dev, Error **errp) 395 { 396 VFIOGroup *group; 397 CcwDevice *ccw_dev = DO_UPCAST(CcwDevice, parent_obj, dev); 398 S390CCWDevice *cdev = DO_UPCAST(S390CCWDevice, parent_obj, ccw_dev); 399 VFIOCCWDevice *vcdev = DO_UPCAST(VFIOCCWDevice, cdev, cdev); 400 S390CCWDeviceClass *cdc = S390_CCW_DEVICE_GET_CLASS(cdev); 401 Error *err = NULL; 402 403 /* Call the class init function for subchannel. */ 404 if (cdc->realize) { 405 cdc->realize(cdev, vcdev->vdev.sysfsdev, &err); 406 if (err) { 407 goto out_err_propagate; 408 } 409 } 410 411 group = vfio_ccw_get_group(cdev, &err); 412 if (!group) { 413 goto out_group_err; 414 } 415 416 vfio_ccw_get_device(group, vcdev, &err); 417 if (err) { 418 goto out_device_err; 419 } 420 421 vfio_ccw_get_region(vcdev, &err); 422 if (err) { 423 goto out_region_err; 424 } 425 426 vfio_ccw_register_io_notifier(vcdev, &err); 427 if (err) { 428 goto out_notifier_err; 429 } 430 431 return; 432 433 out_notifier_err: 434 vfio_ccw_put_region(vcdev); 435 out_region_err: 436 vfio_ccw_put_device(vcdev); 437 out_device_err: 438 vfio_put_group(group); 439 out_group_err: 440 if (cdc->unrealize) { 441 cdc->unrealize(cdev, NULL); 442 } 443 out_err_propagate: 444 error_propagate(errp, err); 445 } 446 447 static void vfio_ccw_unrealize(DeviceState *dev, Error **errp) 448 { 449 CcwDevice *ccw_dev = DO_UPCAST(CcwDevice, parent_obj, dev); 450 S390CCWDevice *cdev = DO_UPCAST(S390CCWDevice, parent_obj, ccw_dev); 451 VFIOCCWDevice *vcdev = DO_UPCAST(VFIOCCWDevice, cdev, cdev); 452 S390CCWDeviceClass *cdc = S390_CCW_DEVICE_GET_CLASS(cdev); 453 VFIOGroup *group = vcdev->vdev.group; 454 455 vfio_ccw_unregister_io_notifier(vcdev); 456 vfio_ccw_put_region(vcdev); 457 vfio_ccw_put_device(vcdev); 458 vfio_put_group(group); 459 460 if (cdc->unrealize) { 461 cdc->unrealize(cdev, errp); 462 } 463 } 464 465 static Property vfio_ccw_properties[] = { 466 DEFINE_PROP_STRING("sysfsdev", VFIOCCWDevice, vdev.sysfsdev), 467 DEFINE_PROP_BOOL("force-orb-pfch", VFIOCCWDevice, force_orb_pfch, false), 468 DEFINE_PROP_END_OF_LIST(), 469 }; 470 471 static const VMStateDescription vfio_ccw_vmstate = { 472 .name = "vfio-ccw", 473 .unmigratable = 1, 474 }; 475 476 static void vfio_ccw_class_init(ObjectClass *klass, void *data) 477 { 478 DeviceClass *dc = DEVICE_CLASS(klass); 479 S390CCWDeviceClass *cdc = S390_CCW_DEVICE_CLASS(klass); 480 481 dc->props = vfio_ccw_properties; 482 dc->vmsd = &vfio_ccw_vmstate; 483 dc->desc = "VFIO-based subchannel assignment"; 484 set_bit(DEVICE_CATEGORY_MISC, dc->categories); 485 dc->realize = vfio_ccw_realize; 486 dc->unrealize = vfio_ccw_unrealize; 487 dc->reset = vfio_ccw_reset; 488 489 cdc->handle_request = vfio_ccw_handle_request; 490 } 491 492 static const TypeInfo vfio_ccw_info = { 493 .name = TYPE_VFIO_CCW, 494 .parent = TYPE_S390_CCW, 495 .instance_size = sizeof(VFIOCCWDevice), 496 .class_init = vfio_ccw_class_init, 497 }; 498 499 static void register_vfio_ccw_type(void) 500 { 501 type_register_static(&vfio_ccw_info); 502 } 503 504 type_init(register_vfio_ccw_type) 505