1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright(c) 2013-2015 Intel Corporation. All rights reserved. 4 */ 5 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 6 #include <linux/libnvdimm.h> 7 #include <linux/sched/mm.h> 8 #include <linux/vmalloc.h> 9 #include <linux/uaccess.h> 10 #include <linux/module.h> 11 #include <linux/blkdev.h> 12 #include <linux/fcntl.h> 13 #include <linux/async.h> 14 #include <linux/ndctl.h> 15 #include <linux/sched.h> 16 #include <linux/slab.h> 17 #include <linux/cpu.h> 18 #include <linux/fs.h> 19 #include <linux/io.h> 20 #include <linux/mm.h> 21 #include <linux/nd.h> 22 #include "nd-core.h" 23 #include "nd.h" 24 #include "pfn.h" 25 26 int nvdimm_major; 27 static int nvdimm_bus_major; 28 static struct class *nd_class; 29 static DEFINE_IDA(nd_ida); 30 31 static int to_nd_device_type(const struct device *dev) 32 { 33 if (is_nvdimm(dev)) 34 return ND_DEVICE_DIMM; 35 else if (is_memory(dev)) 36 return ND_DEVICE_REGION_PMEM; 37 else if (is_nd_dax(dev)) 38 return ND_DEVICE_DAX_PMEM; 39 else if (is_nd_region(dev->parent)) 40 return nd_region_to_nstype(to_nd_region(dev->parent)); 41 42 return 0; 43 } 44 45 static int nvdimm_bus_uevent(const struct device *dev, struct kobj_uevent_env *env) 46 { 47 return add_uevent_var(env, "MODALIAS=" ND_DEVICE_MODALIAS_FMT, 48 to_nd_device_type(dev)); 49 } 50 51 static struct module *to_bus_provider(struct device *dev) 52 { 53 /* pin bus providers while regions are enabled */ 54 if (is_nd_region(dev)) { 55 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev); 56 57 return nvdimm_bus->nd_desc->module; 58 } 59 return NULL; 60 } 61 62 static void nvdimm_bus_probe_start(struct nvdimm_bus *nvdimm_bus) 63 { 64 nvdimm_bus_lock(&nvdimm_bus->dev); 65 nvdimm_bus->probe_active++; 66 nvdimm_bus_unlock(&nvdimm_bus->dev); 67 } 68 69 static void nvdimm_bus_probe_end(struct nvdimm_bus *nvdimm_bus) 70 { 71 nvdimm_bus_lock(&nvdimm_bus->dev); 72 if (--nvdimm_bus->probe_active == 0) 73 wake_up(&nvdimm_bus->wait); 74 nvdimm_bus_unlock(&nvdimm_bus->dev); 75 } 76 77 static int nvdimm_bus_probe(struct device *dev) 78 { 79 struct nd_device_driver *nd_drv = to_nd_device_driver(dev->driver); 80 struct module *provider = to_bus_provider(dev); 81 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev); 82 int rc; 83 84 if (!try_module_get(provider)) 85 return -ENXIO; 86 87 dev_dbg(&nvdimm_bus->dev, "START: %s.probe(%s)\n", 88 dev->driver->name, dev_name(dev)); 89 90 nvdimm_bus_probe_start(nvdimm_bus); 91 rc = nd_drv->probe(dev); 92 if ((rc == 0 || rc == -EOPNOTSUPP) && 93 dev->parent && is_nd_region(dev->parent)) 94 nd_region_advance_seeds(to_nd_region(dev->parent), dev); 95 nvdimm_bus_probe_end(nvdimm_bus); 96 97 dev_dbg(&nvdimm_bus->dev, "END: %s.probe(%s) = %d\n", dev->driver->name, 98 dev_name(dev), rc); 99 100 if (rc != 0) 101 module_put(provider); 102 return rc; 103 } 104 105 static void nvdimm_bus_remove(struct device *dev) 106 { 107 struct nd_device_driver *nd_drv = to_nd_device_driver(dev->driver); 108 struct module *provider = to_bus_provider(dev); 109 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev); 110 111 if (nd_drv->remove) 112 nd_drv->remove(dev); 113 114 dev_dbg(&nvdimm_bus->dev, "%s.remove(%s)\n", dev->driver->name, 115 dev_name(dev)); 116 module_put(provider); 117 } 118 119 static void nvdimm_bus_shutdown(struct device *dev) 120 { 121 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev); 122 struct nd_device_driver *nd_drv = NULL; 123 124 if (dev->driver) 125 nd_drv = to_nd_device_driver(dev->driver); 126 127 if (nd_drv && nd_drv->shutdown) { 128 nd_drv->shutdown(dev); 129 dev_dbg(&nvdimm_bus->dev, "%s.shutdown(%s)\n", 130 dev->driver->name, dev_name(dev)); 131 } 132 } 133 134 void nd_device_notify(struct device *dev, enum nvdimm_event event) 135 { 136 device_lock(dev); 137 if (dev->driver) { 138 struct nd_device_driver *nd_drv; 139 140 nd_drv = to_nd_device_driver(dev->driver); 141 if (nd_drv->notify) 142 nd_drv->notify(dev, event); 143 } 144 device_unlock(dev); 145 } 146 EXPORT_SYMBOL(nd_device_notify); 147 148 void nvdimm_region_notify(struct nd_region *nd_region, enum nvdimm_event event) 149 { 150 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev); 151 152 if (!nvdimm_bus) 153 return; 154 155 /* caller is responsible for holding a reference on the device */ 156 nd_device_notify(&nd_region->dev, event); 157 } 158 EXPORT_SYMBOL_GPL(nvdimm_region_notify); 159 160 struct clear_badblocks_context { 161 resource_size_t phys, cleared; 162 }; 163 164 static int nvdimm_clear_badblocks_region(struct device *dev, void *data) 165 { 166 struct clear_badblocks_context *ctx = data; 167 struct nd_region *nd_region; 168 resource_size_t ndr_end; 169 sector_t sector; 170 171 /* make sure device is a region */ 172 if (!is_memory(dev)) 173 return 0; 174 175 nd_region = to_nd_region(dev); 176 ndr_end = nd_region->ndr_start + nd_region->ndr_size - 1; 177 178 /* make sure we are in the region */ 179 if (ctx->phys < nd_region->ndr_start || 180 (ctx->phys + ctx->cleared - 1) > ndr_end) 181 return 0; 182 183 sector = (ctx->phys - nd_region->ndr_start) / 512; 184 badblocks_clear(&nd_region->bb, sector, ctx->cleared / 512); 185 186 if (nd_region->bb_state) 187 sysfs_notify_dirent(nd_region->bb_state); 188 189 return 0; 190 } 191 192 static void nvdimm_clear_badblocks_regions(struct nvdimm_bus *nvdimm_bus, 193 phys_addr_t phys, u64 cleared) 194 { 195 struct clear_badblocks_context ctx = { 196 .phys = phys, 197 .cleared = cleared, 198 }; 199 200 device_for_each_child(&nvdimm_bus->dev, &ctx, 201 nvdimm_clear_badblocks_region); 202 } 203 204 static void nvdimm_account_cleared_poison(struct nvdimm_bus *nvdimm_bus, 205 phys_addr_t phys, u64 cleared) 206 { 207 if (cleared > 0) 208 badrange_forget(&nvdimm_bus->badrange, phys, cleared); 209 210 if (cleared > 0 && cleared / 512) 211 nvdimm_clear_badblocks_regions(nvdimm_bus, phys, cleared); 212 } 213 214 long nvdimm_clear_poison(struct device *dev, phys_addr_t phys, 215 unsigned int len) 216 { 217 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev); 218 struct nvdimm_bus_descriptor *nd_desc; 219 struct nd_cmd_clear_error clear_err; 220 struct nd_cmd_ars_cap ars_cap; 221 u32 clear_err_unit, mask; 222 unsigned int noio_flag; 223 int cmd_rc, rc; 224 225 if (!nvdimm_bus) 226 return -ENXIO; 227 228 nd_desc = nvdimm_bus->nd_desc; 229 /* 230 * if ndctl does not exist, it's PMEM_LEGACY and 231 * we want to just pretend everything is handled. 232 */ 233 if (!nd_desc->ndctl) 234 return len; 235 236 memset(&ars_cap, 0, sizeof(ars_cap)); 237 ars_cap.address = phys; 238 ars_cap.length = len; 239 noio_flag = memalloc_noio_save(); 240 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, &ars_cap, 241 sizeof(ars_cap), &cmd_rc); 242 memalloc_noio_restore(noio_flag); 243 if (rc < 0) 244 return rc; 245 if (cmd_rc < 0) 246 return cmd_rc; 247 clear_err_unit = ars_cap.clear_err_unit; 248 if (!clear_err_unit || !is_power_of_2(clear_err_unit)) 249 return -ENXIO; 250 251 mask = clear_err_unit - 1; 252 if ((phys | len) & mask) 253 return -ENXIO; 254 memset(&clear_err, 0, sizeof(clear_err)); 255 clear_err.address = phys; 256 clear_err.length = len; 257 noio_flag = memalloc_noio_save(); 258 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_CLEAR_ERROR, &clear_err, 259 sizeof(clear_err), &cmd_rc); 260 memalloc_noio_restore(noio_flag); 261 if (rc < 0) 262 return rc; 263 if (cmd_rc < 0) 264 return cmd_rc; 265 266 nvdimm_account_cleared_poison(nvdimm_bus, phys, clear_err.cleared); 267 268 return clear_err.cleared; 269 } 270 EXPORT_SYMBOL_GPL(nvdimm_clear_poison); 271 272 static int nvdimm_bus_match(struct device *dev, struct device_driver *drv); 273 274 static const struct bus_type nvdimm_bus_type = { 275 .name = "nd", 276 .uevent = nvdimm_bus_uevent, 277 .match = nvdimm_bus_match, 278 .probe = nvdimm_bus_probe, 279 .remove = nvdimm_bus_remove, 280 .shutdown = nvdimm_bus_shutdown, 281 }; 282 283 static void nvdimm_bus_release(struct device *dev) 284 { 285 struct nvdimm_bus *nvdimm_bus; 286 287 nvdimm_bus = container_of(dev, struct nvdimm_bus, dev); 288 ida_free(&nd_ida, nvdimm_bus->id); 289 kfree(nvdimm_bus); 290 } 291 292 static const struct device_type nvdimm_bus_dev_type = { 293 .release = nvdimm_bus_release, 294 .groups = nvdimm_bus_attribute_groups, 295 }; 296 297 bool is_nvdimm_bus(struct device *dev) 298 { 299 return dev->type == &nvdimm_bus_dev_type; 300 } 301 302 struct nvdimm_bus *walk_to_nvdimm_bus(struct device *nd_dev) 303 { 304 struct device *dev; 305 306 for (dev = nd_dev; dev; dev = dev->parent) 307 if (is_nvdimm_bus(dev)) 308 break; 309 dev_WARN_ONCE(nd_dev, !dev, "invalid dev, not on nd bus\n"); 310 if (dev) 311 return to_nvdimm_bus(dev); 312 return NULL; 313 } 314 315 struct nvdimm_bus *to_nvdimm_bus(struct device *dev) 316 { 317 struct nvdimm_bus *nvdimm_bus; 318 319 nvdimm_bus = container_of(dev, struct nvdimm_bus, dev); 320 WARN_ON(!is_nvdimm_bus(dev)); 321 return nvdimm_bus; 322 } 323 EXPORT_SYMBOL_GPL(to_nvdimm_bus); 324 325 struct nvdimm_bus *nvdimm_to_bus(struct nvdimm *nvdimm) 326 { 327 return to_nvdimm_bus(nvdimm->dev.parent); 328 } 329 EXPORT_SYMBOL_GPL(nvdimm_to_bus); 330 331 static struct lock_class_key nvdimm_bus_key; 332 333 struct nvdimm_bus *nvdimm_bus_register(struct device *parent, 334 struct nvdimm_bus_descriptor *nd_desc) 335 { 336 struct nvdimm_bus *nvdimm_bus; 337 int rc; 338 339 nvdimm_bus = kzalloc(sizeof(*nvdimm_bus), GFP_KERNEL); 340 if (!nvdimm_bus) 341 return NULL; 342 INIT_LIST_HEAD(&nvdimm_bus->list); 343 INIT_LIST_HEAD(&nvdimm_bus->mapping_list); 344 init_waitqueue_head(&nvdimm_bus->wait); 345 nvdimm_bus->id = ida_alloc(&nd_ida, GFP_KERNEL); 346 if (nvdimm_bus->id < 0) { 347 kfree(nvdimm_bus); 348 return NULL; 349 } 350 mutex_init(&nvdimm_bus->reconfig_mutex); 351 badrange_init(&nvdimm_bus->badrange); 352 nvdimm_bus->nd_desc = nd_desc; 353 nvdimm_bus->dev.parent = parent; 354 nvdimm_bus->dev.type = &nvdimm_bus_dev_type; 355 nvdimm_bus->dev.groups = nd_desc->attr_groups; 356 nvdimm_bus->dev.bus = &nvdimm_bus_type; 357 nvdimm_bus->dev.of_node = nd_desc->of_node; 358 device_initialize(&nvdimm_bus->dev); 359 lockdep_set_class(&nvdimm_bus->dev.mutex, &nvdimm_bus_key); 360 device_set_pm_not_required(&nvdimm_bus->dev); 361 rc = dev_set_name(&nvdimm_bus->dev, "ndbus%d", nvdimm_bus->id); 362 if (rc) 363 goto err; 364 365 rc = device_add(&nvdimm_bus->dev); 366 if (rc) { 367 dev_dbg(&nvdimm_bus->dev, "registration failed: %d\n", rc); 368 goto err; 369 } 370 371 return nvdimm_bus; 372 err: 373 put_device(&nvdimm_bus->dev); 374 return NULL; 375 } 376 EXPORT_SYMBOL_GPL(nvdimm_bus_register); 377 378 void nvdimm_bus_unregister(struct nvdimm_bus *nvdimm_bus) 379 { 380 if (!nvdimm_bus) 381 return; 382 device_unregister(&nvdimm_bus->dev); 383 } 384 EXPORT_SYMBOL_GPL(nvdimm_bus_unregister); 385 386 static int child_unregister(struct device *dev, void *data) 387 { 388 /* 389 * the singular ndctl class device per bus needs to be 390 * "device_destroy"ed, so skip it here 391 * 392 * i.e. remove classless children 393 */ 394 if (dev->class) 395 return 0; 396 397 if (is_nvdimm(dev)) 398 nvdimm_delete(to_nvdimm(dev)); 399 else 400 nd_device_unregister(dev, ND_SYNC); 401 402 return 0; 403 } 404 405 static void free_badrange_list(struct list_head *badrange_list) 406 { 407 struct badrange_entry *bre, *next; 408 409 list_for_each_entry_safe(bre, next, badrange_list, list) { 410 list_del(&bre->list); 411 kfree(bre); 412 } 413 list_del_init(badrange_list); 414 } 415 416 static void nd_bus_remove(struct device *dev) 417 { 418 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev); 419 420 mutex_lock(&nvdimm_bus_list_mutex); 421 list_del_init(&nvdimm_bus->list); 422 mutex_unlock(&nvdimm_bus_list_mutex); 423 424 wait_event(nvdimm_bus->wait, 425 atomic_read(&nvdimm_bus->ioctl_active) == 0); 426 427 nd_synchronize(); 428 device_for_each_child(&nvdimm_bus->dev, NULL, child_unregister); 429 430 spin_lock(&nvdimm_bus->badrange.lock); 431 free_badrange_list(&nvdimm_bus->badrange.list); 432 spin_unlock(&nvdimm_bus->badrange.lock); 433 434 nvdimm_bus_destroy_ndctl(nvdimm_bus); 435 } 436 437 static int nd_bus_probe(struct device *dev) 438 { 439 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev); 440 int rc; 441 442 rc = nvdimm_bus_create_ndctl(nvdimm_bus); 443 if (rc) 444 return rc; 445 446 mutex_lock(&nvdimm_bus_list_mutex); 447 list_add_tail(&nvdimm_bus->list, &nvdimm_bus_list); 448 mutex_unlock(&nvdimm_bus_list_mutex); 449 450 /* enable bus provider attributes to look up their local context */ 451 dev_set_drvdata(dev, nvdimm_bus->nd_desc); 452 453 return 0; 454 } 455 456 static struct nd_device_driver nd_bus_driver = { 457 .probe = nd_bus_probe, 458 .remove = nd_bus_remove, 459 .drv = { 460 .name = "nd_bus", 461 .suppress_bind_attrs = true, 462 .bus = &nvdimm_bus_type, 463 .owner = THIS_MODULE, 464 .mod_name = KBUILD_MODNAME, 465 }, 466 }; 467 468 static int nvdimm_bus_match(struct device *dev, struct device_driver *drv) 469 { 470 struct nd_device_driver *nd_drv = to_nd_device_driver(drv); 471 472 if (is_nvdimm_bus(dev) && nd_drv == &nd_bus_driver) 473 return true; 474 475 return !!test_bit(to_nd_device_type(dev), &nd_drv->type); 476 } 477 478 static ASYNC_DOMAIN_EXCLUSIVE(nd_async_domain); 479 480 void nd_synchronize(void) 481 { 482 async_synchronize_full_domain(&nd_async_domain); 483 } 484 EXPORT_SYMBOL_GPL(nd_synchronize); 485 486 static void nd_async_device_register(void *d, async_cookie_t cookie) 487 { 488 struct device *dev = d; 489 490 if (device_add(dev) != 0) { 491 dev_err(dev, "%s: failed\n", __func__); 492 put_device(dev); 493 } 494 put_device(dev); 495 if (dev->parent) 496 put_device(dev->parent); 497 } 498 499 static void nd_async_device_unregister(void *d, async_cookie_t cookie) 500 { 501 struct device *dev = d; 502 503 /* flush bus operations before delete */ 504 nvdimm_bus_lock(dev); 505 nvdimm_bus_unlock(dev); 506 507 device_unregister(dev); 508 put_device(dev); 509 } 510 511 static void __nd_device_register(struct device *dev, bool sync) 512 { 513 if (!dev) 514 return; 515 516 /* 517 * Ensure that region devices always have their NUMA node set as 518 * early as possible. This way we are able to make certain that 519 * any memory associated with the creation and the creation 520 * itself of the region is associated with the correct node. 521 */ 522 if (is_nd_region(dev)) 523 set_dev_node(dev, to_nd_region(dev)->numa_node); 524 525 dev->bus = &nvdimm_bus_type; 526 device_set_pm_not_required(dev); 527 if (dev->parent) { 528 get_device(dev->parent); 529 if (dev_to_node(dev) == NUMA_NO_NODE) 530 set_dev_node(dev, dev_to_node(dev->parent)); 531 } 532 get_device(dev); 533 534 if (sync) 535 nd_async_device_register(dev, 0); 536 else 537 async_schedule_dev_domain(nd_async_device_register, dev, 538 &nd_async_domain); 539 } 540 541 void nd_device_register(struct device *dev) 542 { 543 __nd_device_register(dev, false); 544 } 545 EXPORT_SYMBOL(nd_device_register); 546 547 void nd_device_register_sync(struct device *dev) 548 { 549 __nd_device_register(dev, true); 550 } 551 552 void nd_device_unregister(struct device *dev, enum nd_async_mode mode) 553 { 554 bool killed; 555 556 switch (mode) { 557 case ND_ASYNC: 558 /* 559 * In the async case this is being triggered with the 560 * device lock held and the unregistration work needs to 561 * be moved out of line iff this is thread has won the 562 * race to schedule the deletion. 563 */ 564 if (!kill_device(dev)) 565 return; 566 567 get_device(dev); 568 async_schedule_domain(nd_async_device_unregister, dev, 569 &nd_async_domain); 570 break; 571 case ND_SYNC: 572 /* 573 * In the sync case the device is being unregistered due 574 * to a state change of the parent. Claim the kill state 575 * to synchronize against other unregistration requests, 576 * or otherwise let the async path handle it if the 577 * unregistration was already queued. 578 */ 579 device_lock(dev); 580 killed = kill_device(dev); 581 device_unlock(dev); 582 583 if (!killed) 584 return; 585 586 nd_synchronize(); 587 device_unregister(dev); 588 break; 589 } 590 } 591 EXPORT_SYMBOL(nd_device_unregister); 592 593 /** 594 * __nd_driver_register() - register a region or a namespace driver 595 * @nd_drv: driver to register 596 * @owner: automatically set by nd_driver_register() macro 597 * @mod_name: automatically set by nd_driver_register() macro 598 */ 599 int __nd_driver_register(struct nd_device_driver *nd_drv, struct module *owner, 600 const char *mod_name) 601 { 602 struct device_driver *drv = &nd_drv->drv; 603 604 if (!nd_drv->type) { 605 pr_debug("driver type bitmask not set (%ps)\n", 606 __builtin_return_address(0)); 607 return -EINVAL; 608 } 609 610 if (!nd_drv->probe) { 611 pr_debug("%s ->probe() must be specified\n", mod_name); 612 return -EINVAL; 613 } 614 615 drv->bus = &nvdimm_bus_type; 616 drv->owner = owner; 617 drv->mod_name = mod_name; 618 619 return driver_register(drv); 620 } 621 EXPORT_SYMBOL(__nd_driver_register); 622 623 void nvdimm_check_and_set_ro(struct gendisk *disk) 624 { 625 struct device *dev = disk_to_dev(disk)->parent; 626 struct nd_region *nd_region = to_nd_region(dev->parent); 627 int disk_ro = get_disk_ro(disk); 628 629 /* catch the disk up with the region ro state */ 630 if (disk_ro == nd_region->ro) 631 return; 632 633 dev_info(dev, "%s read-%s, marking %s read-%s\n", 634 dev_name(&nd_region->dev), nd_region->ro ? "only" : "write", 635 disk->disk_name, nd_region->ro ? "only" : "write"); 636 set_disk_ro(disk, nd_region->ro); 637 } 638 EXPORT_SYMBOL(nvdimm_check_and_set_ro); 639 640 static ssize_t modalias_show(struct device *dev, struct device_attribute *attr, 641 char *buf) 642 { 643 return sprintf(buf, ND_DEVICE_MODALIAS_FMT "\n", 644 to_nd_device_type(dev)); 645 } 646 static DEVICE_ATTR_RO(modalias); 647 648 static ssize_t devtype_show(struct device *dev, struct device_attribute *attr, 649 char *buf) 650 { 651 return sprintf(buf, "%s\n", dev->type->name); 652 } 653 static DEVICE_ATTR_RO(devtype); 654 655 static struct attribute *nd_device_attributes[] = { 656 &dev_attr_modalias.attr, 657 &dev_attr_devtype.attr, 658 NULL, 659 }; 660 661 /* 662 * nd_device_attribute_group - generic attributes for all devices on an nd bus 663 */ 664 const struct attribute_group nd_device_attribute_group = { 665 .attrs = nd_device_attributes, 666 }; 667 668 static ssize_t numa_node_show(struct device *dev, 669 struct device_attribute *attr, char *buf) 670 { 671 return sprintf(buf, "%d\n", dev_to_node(dev)); 672 } 673 static DEVICE_ATTR_RO(numa_node); 674 675 static int nvdimm_dev_to_target_node(struct device *dev) 676 { 677 struct device *parent = dev->parent; 678 struct nd_region *nd_region = NULL; 679 680 if (is_nd_region(dev)) 681 nd_region = to_nd_region(dev); 682 else if (parent && is_nd_region(parent)) 683 nd_region = to_nd_region(parent); 684 685 if (!nd_region) 686 return NUMA_NO_NODE; 687 return nd_region->target_node; 688 } 689 690 static ssize_t target_node_show(struct device *dev, 691 struct device_attribute *attr, char *buf) 692 { 693 return sprintf(buf, "%d\n", nvdimm_dev_to_target_node(dev)); 694 } 695 static DEVICE_ATTR_RO(target_node); 696 697 static struct attribute *nd_numa_attributes[] = { 698 &dev_attr_numa_node.attr, 699 &dev_attr_target_node.attr, 700 NULL, 701 }; 702 703 static umode_t nd_numa_attr_visible(struct kobject *kobj, struct attribute *a, 704 int n) 705 { 706 struct device *dev = container_of(kobj, typeof(*dev), kobj); 707 708 if (!IS_ENABLED(CONFIG_NUMA)) 709 return 0; 710 711 if (a == &dev_attr_target_node.attr && 712 nvdimm_dev_to_target_node(dev) == NUMA_NO_NODE) 713 return 0; 714 715 return a->mode; 716 } 717 718 /* 719 * nd_numa_attribute_group - NUMA attributes for all devices on an nd bus 720 */ 721 const struct attribute_group nd_numa_attribute_group = { 722 .attrs = nd_numa_attributes, 723 .is_visible = nd_numa_attr_visible, 724 }; 725 726 static void ndctl_release(struct device *dev) 727 { 728 kfree(dev); 729 } 730 731 static struct lock_class_key nvdimm_ndctl_key; 732 733 int nvdimm_bus_create_ndctl(struct nvdimm_bus *nvdimm_bus) 734 { 735 dev_t devt = MKDEV(nvdimm_bus_major, nvdimm_bus->id); 736 struct device *dev; 737 int rc; 738 739 dev = kzalloc(sizeof(*dev), GFP_KERNEL); 740 if (!dev) 741 return -ENOMEM; 742 device_initialize(dev); 743 lockdep_set_class(&dev->mutex, &nvdimm_ndctl_key); 744 device_set_pm_not_required(dev); 745 dev->class = nd_class; 746 dev->parent = &nvdimm_bus->dev; 747 dev->devt = devt; 748 dev->release = ndctl_release; 749 rc = dev_set_name(dev, "ndctl%d", nvdimm_bus->id); 750 if (rc) 751 goto err; 752 753 rc = device_add(dev); 754 if (rc) { 755 dev_dbg(&nvdimm_bus->dev, "failed to register ndctl%d: %d\n", 756 nvdimm_bus->id, rc); 757 goto err; 758 } 759 return 0; 760 761 err: 762 put_device(dev); 763 return rc; 764 } 765 766 void nvdimm_bus_destroy_ndctl(struct nvdimm_bus *nvdimm_bus) 767 { 768 device_destroy(nd_class, MKDEV(nvdimm_bus_major, nvdimm_bus->id)); 769 } 770 771 static const struct nd_cmd_desc __nd_cmd_dimm_descs[] = { 772 [ND_CMD_IMPLEMENTED] = { }, 773 [ND_CMD_SMART] = { 774 .out_num = 2, 775 .out_sizes = { 4, 128, }, 776 }, 777 [ND_CMD_SMART_THRESHOLD] = { 778 .out_num = 2, 779 .out_sizes = { 4, 8, }, 780 }, 781 [ND_CMD_DIMM_FLAGS] = { 782 .out_num = 2, 783 .out_sizes = { 4, 4 }, 784 }, 785 [ND_CMD_GET_CONFIG_SIZE] = { 786 .out_num = 3, 787 .out_sizes = { 4, 4, 4, }, 788 }, 789 [ND_CMD_GET_CONFIG_DATA] = { 790 .in_num = 2, 791 .in_sizes = { 4, 4, }, 792 .out_num = 2, 793 .out_sizes = { 4, UINT_MAX, }, 794 }, 795 [ND_CMD_SET_CONFIG_DATA] = { 796 .in_num = 3, 797 .in_sizes = { 4, 4, UINT_MAX, }, 798 .out_num = 1, 799 .out_sizes = { 4, }, 800 }, 801 [ND_CMD_VENDOR] = { 802 .in_num = 3, 803 .in_sizes = { 4, 4, UINT_MAX, }, 804 .out_num = 3, 805 .out_sizes = { 4, 4, UINT_MAX, }, 806 }, 807 [ND_CMD_CALL] = { 808 .in_num = 2, 809 .in_sizes = { sizeof(struct nd_cmd_pkg), UINT_MAX, }, 810 .out_num = 1, 811 .out_sizes = { UINT_MAX, }, 812 }, 813 }; 814 815 const struct nd_cmd_desc *nd_cmd_dimm_desc(int cmd) 816 { 817 if (cmd < ARRAY_SIZE(__nd_cmd_dimm_descs)) 818 return &__nd_cmd_dimm_descs[cmd]; 819 return NULL; 820 } 821 EXPORT_SYMBOL_GPL(nd_cmd_dimm_desc); 822 823 static const struct nd_cmd_desc __nd_cmd_bus_descs[] = { 824 [ND_CMD_IMPLEMENTED] = { }, 825 [ND_CMD_ARS_CAP] = { 826 .in_num = 2, 827 .in_sizes = { 8, 8, }, 828 .out_num = 4, 829 .out_sizes = { 4, 4, 4, 4, }, 830 }, 831 [ND_CMD_ARS_START] = { 832 .in_num = 5, 833 .in_sizes = { 8, 8, 2, 1, 5, }, 834 .out_num = 2, 835 .out_sizes = { 4, 4, }, 836 }, 837 [ND_CMD_ARS_STATUS] = { 838 .out_num = 3, 839 .out_sizes = { 4, 4, UINT_MAX, }, 840 }, 841 [ND_CMD_CLEAR_ERROR] = { 842 .in_num = 2, 843 .in_sizes = { 8, 8, }, 844 .out_num = 3, 845 .out_sizes = { 4, 4, 8, }, 846 }, 847 [ND_CMD_CALL] = { 848 .in_num = 2, 849 .in_sizes = { sizeof(struct nd_cmd_pkg), UINT_MAX, }, 850 .out_num = 1, 851 .out_sizes = { UINT_MAX, }, 852 }, 853 }; 854 855 const struct nd_cmd_desc *nd_cmd_bus_desc(int cmd) 856 { 857 if (cmd < ARRAY_SIZE(__nd_cmd_bus_descs)) 858 return &__nd_cmd_bus_descs[cmd]; 859 return NULL; 860 } 861 EXPORT_SYMBOL_GPL(nd_cmd_bus_desc); 862 863 u32 nd_cmd_in_size(struct nvdimm *nvdimm, int cmd, 864 const struct nd_cmd_desc *desc, int idx, void *buf) 865 { 866 if (idx >= desc->in_num) 867 return UINT_MAX; 868 869 if (desc->in_sizes[idx] < UINT_MAX) 870 return desc->in_sizes[idx]; 871 872 if (nvdimm && cmd == ND_CMD_SET_CONFIG_DATA && idx == 2) { 873 struct nd_cmd_set_config_hdr *hdr = buf; 874 875 return hdr->in_length; 876 } else if (nvdimm && cmd == ND_CMD_VENDOR && idx == 2) { 877 struct nd_cmd_vendor_hdr *hdr = buf; 878 879 return hdr->in_length; 880 } else if (cmd == ND_CMD_CALL) { 881 struct nd_cmd_pkg *pkg = buf; 882 883 return pkg->nd_size_in; 884 } 885 886 return UINT_MAX; 887 } 888 EXPORT_SYMBOL_GPL(nd_cmd_in_size); 889 890 u32 nd_cmd_out_size(struct nvdimm *nvdimm, int cmd, 891 const struct nd_cmd_desc *desc, int idx, const u32 *in_field, 892 const u32 *out_field, unsigned long remainder) 893 { 894 if (idx >= desc->out_num) 895 return UINT_MAX; 896 897 if (desc->out_sizes[idx] < UINT_MAX) 898 return desc->out_sizes[idx]; 899 900 if (nvdimm && cmd == ND_CMD_GET_CONFIG_DATA && idx == 1) 901 return in_field[1]; 902 else if (nvdimm && cmd == ND_CMD_VENDOR && idx == 2) 903 return out_field[1]; 904 else if (!nvdimm && cmd == ND_CMD_ARS_STATUS && idx == 2) { 905 /* 906 * Per table 9-276 ARS Data in ACPI 6.1, out_field[1] is 907 * "Size of Output Buffer in bytes, including this 908 * field." 909 */ 910 if (out_field[1] < 4) 911 return 0; 912 /* 913 * ACPI 6.1 is ambiguous if 'status' is included in the 914 * output size. If we encounter an output size that 915 * overshoots the remainder by 4 bytes, assume it was 916 * including 'status'. 917 */ 918 if (out_field[1] - 4 == remainder) 919 return remainder; 920 return out_field[1] - 8; 921 } else if (cmd == ND_CMD_CALL) { 922 struct nd_cmd_pkg *pkg = (struct nd_cmd_pkg *) in_field; 923 924 return pkg->nd_size_out; 925 } 926 927 928 return UINT_MAX; 929 } 930 EXPORT_SYMBOL_GPL(nd_cmd_out_size); 931 932 void wait_nvdimm_bus_probe_idle(struct device *dev) 933 { 934 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev); 935 936 do { 937 if (nvdimm_bus->probe_active == 0) 938 break; 939 nvdimm_bus_unlock(dev); 940 device_unlock(dev); 941 wait_event(nvdimm_bus->wait, 942 nvdimm_bus->probe_active == 0); 943 device_lock(dev); 944 nvdimm_bus_lock(dev); 945 } while (true); 946 } 947 948 static int nd_pmem_forget_poison_check(struct device *dev, void *data) 949 { 950 struct nd_cmd_clear_error *clear_err = 951 (struct nd_cmd_clear_error *)data; 952 struct nd_btt *nd_btt = is_nd_btt(dev) ? to_nd_btt(dev) : NULL; 953 struct nd_pfn *nd_pfn = is_nd_pfn(dev) ? to_nd_pfn(dev) : NULL; 954 struct nd_dax *nd_dax = is_nd_dax(dev) ? to_nd_dax(dev) : NULL; 955 struct nd_namespace_common *ndns = NULL; 956 struct nd_namespace_io *nsio; 957 resource_size_t offset = 0, end_trunc = 0, start, end, pstart, pend; 958 959 if (nd_dax || !dev->driver) 960 return 0; 961 962 start = clear_err->address; 963 end = clear_err->address + clear_err->cleared - 1; 964 965 if (nd_btt || nd_pfn || nd_dax) { 966 if (nd_btt) 967 ndns = nd_btt->ndns; 968 else if (nd_pfn) 969 ndns = nd_pfn->ndns; 970 else if (nd_dax) 971 ndns = nd_dax->nd_pfn.ndns; 972 973 if (!ndns) 974 return 0; 975 } else 976 ndns = to_ndns(dev); 977 978 nsio = to_nd_namespace_io(&ndns->dev); 979 pstart = nsio->res.start + offset; 980 pend = nsio->res.end - end_trunc; 981 982 if ((pstart >= start) && (pend <= end)) 983 return -EBUSY; 984 985 return 0; 986 987 } 988 989 static int nd_ns_forget_poison_check(struct device *dev, void *data) 990 { 991 return device_for_each_child(dev, data, nd_pmem_forget_poison_check); 992 } 993 994 /* set_config requires an idle interleave set */ 995 static int nd_cmd_clear_to_send(struct nvdimm_bus *nvdimm_bus, 996 struct nvdimm *nvdimm, unsigned int cmd, void *data) 997 { 998 struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc; 999 1000 /* ask the bus provider if it would like to block this request */ 1001 if (nd_desc->clear_to_send) { 1002 int rc = nd_desc->clear_to_send(nd_desc, nvdimm, cmd, data); 1003 1004 if (rc) 1005 return rc; 1006 } 1007 1008 /* require clear error to go through the pmem driver */ 1009 if (!nvdimm && cmd == ND_CMD_CLEAR_ERROR) 1010 return device_for_each_child(&nvdimm_bus->dev, data, 1011 nd_ns_forget_poison_check); 1012 1013 if (!nvdimm || cmd != ND_CMD_SET_CONFIG_DATA) 1014 return 0; 1015 1016 /* prevent label manipulation while the kernel owns label updates */ 1017 wait_nvdimm_bus_probe_idle(&nvdimm_bus->dev); 1018 if (atomic_read(&nvdimm->busy)) 1019 return -EBUSY; 1020 return 0; 1021 } 1022 1023 static int __nd_ioctl(struct nvdimm_bus *nvdimm_bus, struct nvdimm *nvdimm, 1024 int read_only, unsigned int ioctl_cmd, unsigned long arg) 1025 { 1026 struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc; 1027 const struct nd_cmd_desc *desc = NULL; 1028 unsigned int cmd = _IOC_NR(ioctl_cmd); 1029 struct device *dev = &nvdimm_bus->dev; 1030 void __user *p = (void __user *) arg; 1031 char *out_env = NULL, *in_env = NULL; 1032 const char *cmd_name, *dimm_name; 1033 u32 in_len = 0, out_len = 0; 1034 unsigned int func = cmd; 1035 unsigned long cmd_mask; 1036 struct nd_cmd_pkg pkg; 1037 int rc, i, cmd_rc; 1038 void *buf = NULL; 1039 u64 buf_len = 0; 1040 1041 if (nvdimm) { 1042 desc = nd_cmd_dimm_desc(cmd); 1043 cmd_name = nvdimm_cmd_name(cmd); 1044 cmd_mask = nvdimm->cmd_mask; 1045 dimm_name = dev_name(&nvdimm->dev); 1046 } else { 1047 desc = nd_cmd_bus_desc(cmd); 1048 cmd_name = nvdimm_bus_cmd_name(cmd); 1049 cmd_mask = nd_desc->cmd_mask; 1050 dimm_name = "bus"; 1051 } 1052 1053 /* Validate command family support against bus declared support */ 1054 if (cmd == ND_CMD_CALL) { 1055 unsigned long *mask; 1056 1057 if (copy_from_user(&pkg, p, sizeof(pkg))) 1058 return -EFAULT; 1059 1060 if (nvdimm) { 1061 if (pkg.nd_family > NVDIMM_FAMILY_MAX) 1062 return -EINVAL; 1063 mask = &nd_desc->dimm_family_mask; 1064 } else { 1065 if (pkg.nd_family > NVDIMM_BUS_FAMILY_MAX) 1066 return -EINVAL; 1067 mask = &nd_desc->bus_family_mask; 1068 } 1069 1070 if (!test_bit(pkg.nd_family, mask)) 1071 return -EINVAL; 1072 } 1073 1074 if (!desc || 1075 (desc->out_num + desc->in_num == 0) || 1076 cmd > ND_CMD_CALL || 1077 !test_bit(cmd, &cmd_mask)) 1078 return -ENOTTY; 1079 1080 /* fail write commands (when read-only) */ 1081 if (read_only) 1082 switch (cmd) { 1083 case ND_CMD_VENDOR: 1084 case ND_CMD_SET_CONFIG_DATA: 1085 case ND_CMD_ARS_START: 1086 case ND_CMD_CLEAR_ERROR: 1087 case ND_CMD_CALL: 1088 dev_dbg(dev, "'%s' command while read-only.\n", 1089 nvdimm ? nvdimm_cmd_name(cmd) 1090 : nvdimm_bus_cmd_name(cmd)); 1091 return -EPERM; 1092 default: 1093 break; 1094 } 1095 1096 /* process an input envelope */ 1097 in_env = kzalloc(ND_CMD_MAX_ENVELOPE, GFP_KERNEL); 1098 if (!in_env) 1099 return -ENOMEM; 1100 for (i = 0; i < desc->in_num; i++) { 1101 u32 in_size, copy; 1102 1103 in_size = nd_cmd_in_size(nvdimm, cmd, desc, i, in_env); 1104 if (in_size == UINT_MAX) { 1105 dev_err(dev, "%s:%s unknown input size cmd: %s field: %d\n", 1106 __func__, dimm_name, cmd_name, i); 1107 rc = -ENXIO; 1108 goto out; 1109 } 1110 if (in_len < ND_CMD_MAX_ENVELOPE) 1111 copy = min_t(u32, ND_CMD_MAX_ENVELOPE - in_len, in_size); 1112 else 1113 copy = 0; 1114 if (copy && copy_from_user(&in_env[in_len], p + in_len, copy)) { 1115 rc = -EFAULT; 1116 goto out; 1117 } 1118 in_len += in_size; 1119 } 1120 1121 if (cmd == ND_CMD_CALL) { 1122 func = pkg.nd_command; 1123 dev_dbg(dev, "%s, idx: %llu, in: %u, out: %u, len %llu\n", 1124 dimm_name, pkg.nd_command, 1125 in_len, out_len, buf_len); 1126 } 1127 1128 /* process an output envelope */ 1129 out_env = kzalloc(ND_CMD_MAX_ENVELOPE, GFP_KERNEL); 1130 if (!out_env) { 1131 rc = -ENOMEM; 1132 goto out; 1133 } 1134 1135 for (i = 0; i < desc->out_num; i++) { 1136 u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i, 1137 (u32 *) in_env, (u32 *) out_env, 0); 1138 u32 copy; 1139 1140 if (out_size == UINT_MAX) { 1141 dev_dbg(dev, "%s unknown output size cmd: %s field: %d\n", 1142 dimm_name, cmd_name, i); 1143 rc = -EFAULT; 1144 goto out; 1145 } 1146 if (out_len < ND_CMD_MAX_ENVELOPE) 1147 copy = min_t(u32, ND_CMD_MAX_ENVELOPE - out_len, out_size); 1148 else 1149 copy = 0; 1150 if (copy && copy_from_user(&out_env[out_len], 1151 p + in_len + out_len, copy)) { 1152 rc = -EFAULT; 1153 goto out; 1154 } 1155 out_len += out_size; 1156 } 1157 1158 buf_len = (u64) out_len + (u64) in_len; 1159 if (buf_len > ND_IOCTL_MAX_BUFLEN) { 1160 dev_dbg(dev, "%s cmd: %s buf_len: %llu > %d\n", dimm_name, 1161 cmd_name, buf_len, ND_IOCTL_MAX_BUFLEN); 1162 rc = -EINVAL; 1163 goto out; 1164 } 1165 1166 buf = vmalloc(buf_len); 1167 if (!buf) { 1168 rc = -ENOMEM; 1169 goto out; 1170 } 1171 1172 if (copy_from_user(buf, p, buf_len)) { 1173 rc = -EFAULT; 1174 goto out; 1175 } 1176 1177 device_lock(dev); 1178 nvdimm_bus_lock(dev); 1179 rc = nd_cmd_clear_to_send(nvdimm_bus, nvdimm, func, buf); 1180 if (rc) 1181 goto out_unlock; 1182 1183 rc = nd_desc->ndctl(nd_desc, nvdimm, cmd, buf, buf_len, &cmd_rc); 1184 if (rc < 0) 1185 goto out_unlock; 1186 1187 if (!nvdimm && cmd == ND_CMD_CLEAR_ERROR && cmd_rc >= 0) { 1188 struct nd_cmd_clear_error *clear_err = buf; 1189 1190 nvdimm_account_cleared_poison(nvdimm_bus, clear_err->address, 1191 clear_err->cleared); 1192 } 1193 1194 if (copy_to_user(p, buf, buf_len)) 1195 rc = -EFAULT; 1196 1197 out_unlock: 1198 nvdimm_bus_unlock(dev); 1199 device_unlock(dev); 1200 out: 1201 kfree(in_env); 1202 kfree(out_env); 1203 vfree(buf); 1204 return rc; 1205 } 1206 1207 enum nd_ioctl_mode { 1208 BUS_IOCTL, 1209 DIMM_IOCTL, 1210 }; 1211 1212 static int match_dimm(struct device *dev, void *data) 1213 { 1214 long id = (long) data; 1215 1216 if (is_nvdimm(dev)) { 1217 struct nvdimm *nvdimm = to_nvdimm(dev); 1218 1219 return nvdimm->id == id; 1220 } 1221 1222 return 0; 1223 } 1224 1225 static long nd_ioctl(struct file *file, unsigned int cmd, unsigned long arg, 1226 enum nd_ioctl_mode mode) 1227 1228 { 1229 struct nvdimm_bus *nvdimm_bus, *found = NULL; 1230 long id = (long) file->private_data; 1231 struct nvdimm *nvdimm = NULL; 1232 int rc, ro; 1233 1234 ro = ((file->f_flags & O_ACCMODE) == O_RDONLY); 1235 mutex_lock(&nvdimm_bus_list_mutex); 1236 list_for_each_entry(nvdimm_bus, &nvdimm_bus_list, list) { 1237 if (mode == DIMM_IOCTL) { 1238 struct device *dev; 1239 1240 dev = device_find_child(&nvdimm_bus->dev, 1241 file->private_data, match_dimm); 1242 if (!dev) 1243 continue; 1244 nvdimm = to_nvdimm(dev); 1245 found = nvdimm_bus; 1246 } else if (nvdimm_bus->id == id) { 1247 found = nvdimm_bus; 1248 } 1249 1250 if (found) { 1251 atomic_inc(&nvdimm_bus->ioctl_active); 1252 break; 1253 } 1254 } 1255 mutex_unlock(&nvdimm_bus_list_mutex); 1256 1257 if (!found) 1258 return -ENXIO; 1259 1260 nvdimm_bus = found; 1261 rc = __nd_ioctl(nvdimm_bus, nvdimm, ro, cmd, arg); 1262 1263 if (nvdimm) 1264 put_device(&nvdimm->dev); 1265 if (atomic_dec_and_test(&nvdimm_bus->ioctl_active)) 1266 wake_up(&nvdimm_bus->wait); 1267 1268 return rc; 1269 } 1270 1271 static long bus_ioctl(struct file *file, unsigned int cmd, unsigned long arg) 1272 { 1273 return nd_ioctl(file, cmd, arg, BUS_IOCTL); 1274 } 1275 1276 static long dimm_ioctl(struct file *file, unsigned int cmd, unsigned long arg) 1277 { 1278 return nd_ioctl(file, cmd, arg, DIMM_IOCTL); 1279 } 1280 1281 static int nd_open(struct inode *inode, struct file *file) 1282 { 1283 long minor = iminor(inode); 1284 1285 file->private_data = (void *) minor; 1286 return 0; 1287 } 1288 1289 static const struct file_operations nvdimm_bus_fops = { 1290 .owner = THIS_MODULE, 1291 .open = nd_open, 1292 .unlocked_ioctl = bus_ioctl, 1293 .compat_ioctl = compat_ptr_ioctl, 1294 .llseek = noop_llseek, 1295 }; 1296 1297 static const struct file_operations nvdimm_fops = { 1298 .owner = THIS_MODULE, 1299 .open = nd_open, 1300 .unlocked_ioctl = dimm_ioctl, 1301 .compat_ioctl = compat_ptr_ioctl, 1302 .llseek = noop_llseek, 1303 }; 1304 1305 int __init nvdimm_bus_init(void) 1306 { 1307 int rc; 1308 1309 rc = bus_register(&nvdimm_bus_type); 1310 if (rc) 1311 return rc; 1312 1313 rc = register_chrdev(0, "ndctl", &nvdimm_bus_fops); 1314 if (rc < 0) 1315 goto err_bus_chrdev; 1316 nvdimm_bus_major = rc; 1317 1318 rc = register_chrdev(0, "dimmctl", &nvdimm_fops); 1319 if (rc < 0) 1320 goto err_dimm_chrdev; 1321 nvdimm_major = rc; 1322 1323 nd_class = class_create("nd"); 1324 if (IS_ERR(nd_class)) { 1325 rc = PTR_ERR(nd_class); 1326 goto err_class; 1327 } 1328 1329 rc = driver_register(&nd_bus_driver.drv); 1330 if (rc) 1331 goto err_nd_bus; 1332 1333 return 0; 1334 1335 err_nd_bus: 1336 class_destroy(nd_class); 1337 err_class: 1338 unregister_chrdev(nvdimm_major, "dimmctl"); 1339 err_dimm_chrdev: 1340 unregister_chrdev(nvdimm_bus_major, "ndctl"); 1341 err_bus_chrdev: 1342 bus_unregister(&nvdimm_bus_type); 1343 1344 return rc; 1345 } 1346 1347 void nvdimm_bus_exit(void) 1348 { 1349 driver_unregister(&nd_bus_driver.drv); 1350 class_destroy(nd_class); 1351 unregister_chrdev(nvdimm_bus_major, "ndctl"); 1352 unregister_chrdev(nvdimm_major, "dimmctl"); 1353 bus_unregister(&nvdimm_bus_type); 1354 ida_destroy(&nd_ida); 1355 } 1356