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