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