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/slab.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/cpu.h> 17 #include <linux/fs.h> 18 #include <linux/io.h> 19 #include <linux/mm.h> 20 #include <linux/nd.h> 21 #include "nd-core.h" 22 #include "nd.h" 23 #include "pfn.h" 24 25 int nvdimm_major; 26 static int nvdimm_bus_major; 27 static DEFINE_IDA(nd_ida); 28 29 static const struct class nd_class = { 30 .name = "nd", 31 }; 32 33 static int to_nd_device_type(const struct device *dev) 34 { 35 if (is_nvdimm(dev)) 36 return ND_DEVICE_DIMM; 37 else if (is_memory(dev)) 38 return ND_DEVICE_REGION_PMEM; 39 else if (is_nd_dax(dev)) 40 return ND_DEVICE_DAX_PMEM; 41 else if (is_nd_region(dev->parent)) 42 return nd_region_to_nstype(to_nd_region(dev->parent)); 43 44 return 0; 45 } 46 47 static int nvdimm_bus_uevent(const struct device *dev, struct kobj_uevent_env *env) 48 { 49 return add_uevent_var(env, "MODALIAS=" ND_DEVICE_MODALIAS_FMT, 50 to_nd_device_type(dev)); 51 } 52 53 static struct module *to_bus_provider(struct device *dev) 54 { 55 /* pin bus providers while regions are enabled */ 56 if (is_nd_region(dev)) { 57 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev); 58 59 return nvdimm_bus->nd_desc->module; 60 } 61 return NULL; 62 } 63 64 static void nvdimm_bus_probe_start(struct nvdimm_bus *nvdimm_bus) 65 { 66 guard(nvdimm_bus)(&nvdimm_bus->dev); 67 nvdimm_bus->probe_active++; 68 } 69 70 static void nvdimm_bus_probe_end(struct nvdimm_bus *nvdimm_bus) 71 { 72 guard(nvdimm_bus)(&nvdimm_bus->dev); 73 if (--nvdimm_bus->probe_active == 0) 74 wake_up(&nvdimm_bus->wait); 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, const 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_obj(*nvdimm_bus); 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, const struct device_driver *drv) 469 { 470 const 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 struct device *parent = dev->parent; 490 491 if (device_add(dev) != 0) { 492 dev_err(dev, "%s: failed\n", __func__); 493 put_device(dev); 494 } 495 put_device(dev); 496 if (parent) 497 put_device(parent); 498 } 499 500 static void nd_async_device_unregister(void *d, async_cookie_t cookie) 501 { 502 struct device *dev = d; 503 504 /* flush bus operations before delete */ 505 nvdimm_bus_lock(dev); 506 nvdimm_bus_unlock(dev); 507 508 device_unregister(dev); 509 put_device(dev); 510 } 511 512 static void __nd_device_register(struct device *dev, bool sync) 513 { 514 if (!dev) 515 return; 516 517 /* 518 * Ensure that region devices always have their NUMA node set as 519 * early as possible. This way we are able to make certain that 520 * any memory associated with the creation and the creation 521 * itself of the region is associated with the correct node. 522 */ 523 if (is_nd_region(dev)) 524 set_dev_node(dev, to_nd_region(dev)->numa_node); 525 526 dev->bus = &nvdimm_bus_type; 527 device_set_pm_not_required(dev); 528 if (dev->parent) { 529 get_device(dev->parent); 530 if (dev_to_node(dev) == NUMA_NO_NODE) 531 set_dev_node(dev, dev_to_node(dev->parent)); 532 } 533 get_device(dev); 534 535 if (sync) 536 nd_async_device_register(dev, 0); 537 else 538 async_schedule_dev_domain(nd_async_device_register, dev, 539 &nd_async_domain); 540 } 541 542 void nd_device_register(struct device *dev) 543 { 544 __nd_device_register(dev, false); 545 } 546 EXPORT_SYMBOL(nd_device_register); 547 548 void nd_device_register_sync(struct device *dev) 549 { 550 __nd_device_register(dev, true); 551 } 552 553 void nd_device_unregister(struct device *dev, enum nd_async_mode mode) 554 { 555 bool killed; 556 557 switch (mode) { 558 case ND_ASYNC: 559 /* 560 * In the async case this is being triggered with the 561 * device lock held and the unregistration work needs to 562 * be moved out of line iff this is thread has won the 563 * race to schedule the deletion. 564 */ 565 if (!kill_device(dev)) 566 return; 567 568 get_device(dev); 569 async_schedule_domain(nd_async_device_unregister, dev, 570 &nd_async_domain); 571 break; 572 case ND_SYNC: 573 /* 574 * In the sync case the device is being unregistered due 575 * to a state change of the parent. Claim the kill state 576 * to synchronize against other unregistration requests, 577 * or otherwise let the async path handle it if the 578 * unregistration was already queued. 579 */ 580 device_lock(dev); 581 killed = kill_device(dev); 582 device_unlock(dev); 583 584 if (!killed) 585 return; 586 587 nd_synchronize(); 588 device_unregister(dev); 589 break; 590 } 591 } 592 EXPORT_SYMBOL(nd_device_unregister); 593 594 /** 595 * __nd_driver_register() - register a region or a namespace driver 596 * @nd_drv: driver to register 597 * @owner: automatically set by nd_driver_register() macro 598 * @mod_name: automatically set by nd_driver_register() macro 599 */ 600 int __nd_driver_register(struct nd_device_driver *nd_drv, struct module *owner, 601 const char *mod_name) 602 { 603 struct device_driver *drv = &nd_drv->drv; 604 605 if (!nd_drv->type) { 606 pr_debug("driver type bitmask not set (%ps)\n", 607 __builtin_return_address(0)); 608 return -EINVAL; 609 } 610 611 if (!nd_drv->probe) { 612 pr_debug("%s ->probe() must be specified\n", mod_name); 613 return -EINVAL; 614 } 615 616 drv->bus = &nvdimm_bus_type; 617 drv->owner = owner; 618 drv->mod_name = mod_name; 619 620 return driver_register(drv); 621 } 622 EXPORT_SYMBOL(__nd_driver_register); 623 624 void nvdimm_check_and_set_ro(struct gendisk *disk) 625 { 626 struct device *dev = disk_to_dev(disk)->parent; 627 struct nd_region *nd_region = to_nd_region(dev->parent); 628 int disk_ro = get_disk_ro(disk); 629 630 /* catch the disk up with the region ro state */ 631 if (disk_ro == nd_region->ro) 632 return; 633 634 dev_info(dev, "%s read-%s, marking %s read-%s\n", 635 dev_name(&nd_region->dev), nd_region->ro ? "only" : "write", 636 disk->disk_name, nd_region->ro ? "only" : "write"); 637 set_disk_ro(disk, nd_region->ro); 638 } 639 EXPORT_SYMBOL(nvdimm_check_and_set_ro); 640 641 static ssize_t modalias_show(struct device *dev, struct device_attribute *attr, 642 char *buf) 643 { 644 return sprintf(buf, ND_DEVICE_MODALIAS_FMT "\n", 645 to_nd_device_type(dev)); 646 } 647 static DEVICE_ATTR_RO(modalias); 648 649 static ssize_t devtype_show(struct device *dev, struct device_attribute *attr, 650 char *buf) 651 { 652 return sprintf(buf, "%s\n", dev->type->name); 653 } 654 static DEVICE_ATTR_RO(devtype); 655 656 static struct attribute *nd_device_attributes[] = { 657 &dev_attr_modalias.attr, 658 &dev_attr_devtype.attr, 659 NULL, 660 }; 661 662 /* 663 * nd_device_attribute_group - generic attributes for all devices on an nd bus 664 */ 665 const struct attribute_group nd_device_attribute_group = { 666 .attrs = nd_device_attributes, 667 }; 668 669 static ssize_t numa_node_show(struct device *dev, 670 struct device_attribute *attr, char *buf) 671 { 672 return sprintf(buf, "%d\n", dev_to_node(dev)); 673 } 674 static DEVICE_ATTR_RO(numa_node); 675 676 static int nvdimm_dev_to_target_node(struct device *dev) 677 { 678 struct device *parent = dev->parent; 679 struct nd_region *nd_region = NULL; 680 681 if (is_nd_region(dev)) 682 nd_region = to_nd_region(dev); 683 else if (parent && is_nd_region(parent)) 684 nd_region = to_nd_region(parent); 685 686 if (!nd_region) 687 return NUMA_NO_NODE; 688 return nd_region->target_node; 689 } 690 691 static ssize_t target_node_show(struct device *dev, 692 struct device_attribute *attr, char *buf) 693 { 694 return sprintf(buf, "%d\n", nvdimm_dev_to_target_node(dev)); 695 } 696 static DEVICE_ATTR_RO(target_node); 697 698 static struct attribute *nd_numa_attributes[] = { 699 &dev_attr_numa_node.attr, 700 &dev_attr_target_node.attr, 701 NULL, 702 }; 703 704 static umode_t nd_numa_attr_visible(struct kobject *kobj, struct attribute *a, 705 int n) 706 { 707 struct device *dev = container_of(kobj, typeof(*dev), kobj); 708 709 if (!IS_ENABLED(CONFIG_NUMA)) 710 return 0; 711 712 if (a == &dev_attr_target_node.attr && 713 nvdimm_dev_to_target_node(dev) == NUMA_NO_NODE) 714 return 0; 715 716 return a->mode; 717 } 718 719 /* 720 * nd_numa_attribute_group - NUMA attributes for all devices on an nd bus 721 */ 722 const struct attribute_group nd_numa_attribute_group = { 723 .attrs = nd_numa_attributes, 724 .is_visible = nd_numa_attr_visible, 725 }; 726 727 static void ndctl_release(struct device *dev) 728 { 729 kfree(dev); 730 } 731 732 static struct lock_class_key nvdimm_ndctl_key; 733 734 int nvdimm_bus_create_ndctl(struct nvdimm_bus *nvdimm_bus) 735 { 736 dev_t devt = MKDEV(nvdimm_bus_major, nvdimm_bus->id); 737 struct device *dev; 738 int rc; 739 740 dev = kzalloc_obj(*dev); 741 if (!dev) 742 return -ENOMEM; 743 device_initialize(dev); 744 lockdep_set_class(&dev->mutex, &nvdimm_ndctl_key); 745 device_set_pm_not_required(dev); 746 dev->class = &nd_class; 747 dev->parent = &nvdimm_bus->dev; 748 dev->devt = devt; 749 dev->release = ndctl_release; 750 rc = dev_set_name(dev, "ndctl%d", nvdimm_bus->id); 751 if (rc) 752 goto err; 753 754 rc = device_add(dev); 755 if (rc) { 756 dev_dbg(&nvdimm_bus->dev, "failed to register ndctl%d: %d\n", 757 nvdimm_bus->id, rc); 758 goto err; 759 } 760 return 0; 761 762 err: 763 put_device(dev); 764 return rc; 765 } 766 767 void nvdimm_bus_destroy_ndctl(struct nvdimm_bus *nvdimm_bus) 768 { 769 device_destroy(&nd_class, MKDEV(nvdimm_bus_major, nvdimm_bus->id)); 770 } 771 772 static const struct nd_cmd_desc __nd_cmd_dimm_descs[] = { 773 [ND_CMD_IMPLEMENTED] = { }, 774 [ND_CMD_SMART] = { 775 .out_num = 2, 776 .out_sizes = { 4, 128, }, 777 }, 778 [ND_CMD_SMART_THRESHOLD] = { 779 .out_num = 2, 780 .out_sizes = { 4, 8, }, 781 }, 782 [ND_CMD_DIMM_FLAGS] = { 783 .out_num = 2, 784 .out_sizes = { 4, 4 }, 785 }, 786 [ND_CMD_GET_CONFIG_SIZE] = { 787 .out_num = 3, 788 .out_sizes = { 4, 4, 4, }, 789 }, 790 [ND_CMD_GET_CONFIG_DATA] = { 791 .in_num = 2, 792 .in_sizes = { 4, 4, }, 793 .out_num = 2, 794 .out_sizes = { 4, UINT_MAX, }, 795 }, 796 [ND_CMD_SET_CONFIG_DATA] = { 797 .in_num = 3, 798 .in_sizes = { 4, 4, UINT_MAX, }, 799 .out_num = 1, 800 .out_sizes = { 4, }, 801 }, 802 [ND_CMD_VENDOR] = { 803 .in_num = 3, 804 .in_sizes = { 4, 4, UINT_MAX, }, 805 .out_num = 3, 806 .out_sizes = { 4, 4, UINT_MAX, }, 807 }, 808 [ND_CMD_CALL] = { 809 .in_num = 2, 810 .in_sizes = { sizeof(struct nd_cmd_pkg), UINT_MAX, }, 811 .out_num = 1, 812 .out_sizes = { UINT_MAX, }, 813 }, 814 }; 815 816 const struct nd_cmd_desc *nd_cmd_dimm_desc(int cmd) 817 { 818 if (cmd < ARRAY_SIZE(__nd_cmd_dimm_descs)) 819 return &__nd_cmd_dimm_descs[cmd]; 820 return NULL; 821 } 822 EXPORT_SYMBOL_GPL(nd_cmd_dimm_desc); 823 824 static const struct nd_cmd_desc __nd_cmd_bus_descs[] = { 825 [ND_CMD_IMPLEMENTED] = { }, 826 [ND_CMD_ARS_CAP] = { 827 .in_num = 2, 828 .in_sizes = { 8, 8, }, 829 .out_num = 4, 830 .out_sizes = { 4, 4, 4, 4, }, 831 }, 832 [ND_CMD_ARS_START] = { 833 .in_num = 5, 834 .in_sizes = { 8, 8, 2, 1, 5, }, 835 .out_num = 2, 836 .out_sizes = { 4, 4, }, 837 }, 838 [ND_CMD_ARS_STATUS] = { 839 .out_num = 3, 840 .out_sizes = { 4, 4, UINT_MAX, }, 841 }, 842 [ND_CMD_CLEAR_ERROR] = { 843 .in_num = 2, 844 .in_sizes = { 8, 8, }, 845 .out_num = 3, 846 .out_sizes = { 4, 4, 8, }, 847 }, 848 [ND_CMD_CALL] = { 849 .in_num = 2, 850 .in_sizes = { sizeof(struct nd_cmd_pkg), UINT_MAX, }, 851 .out_num = 1, 852 .out_sizes = { UINT_MAX, }, 853 }, 854 }; 855 856 const struct nd_cmd_desc *nd_cmd_bus_desc(int cmd) 857 { 858 if (cmd < ARRAY_SIZE(__nd_cmd_bus_descs)) 859 return &__nd_cmd_bus_descs[cmd]; 860 return NULL; 861 } 862 EXPORT_SYMBOL_GPL(nd_cmd_bus_desc); 863 864 u32 nd_cmd_in_size(struct nvdimm *nvdimm, int cmd, 865 const struct nd_cmd_desc *desc, int idx, void *buf) 866 { 867 if (idx >= desc->in_num) 868 return UINT_MAX; 869 870 if (desc->in_sizes[idx] < UINT_MAX) 871 return desc->in_sizes[idx]; 872 873 if (nvdimm && cmd == ND_CMD_SET_CONFIG_DATA && idx == 2) { 874 struct nd_cmd_set_config_hdr *hdr = buf; 875 876 return hdr->in_length; 877 } else if (nvdimm && cmd == ND_CMD_VENDOR && idx == 2) { 878 struct nd_cmd_vendor_hdr *hdr = buf; 879 880 return hdr->in_length; 881 } else if (cmd == ND_CMD_CALL) { 882 struct nd_cmd_pkg *pkg = buf; 883 884 return pkg->nd_size_in; 885 } 886 887 return UINT_MAX; 888 } 889 EXPORT_SYMBOL_GPL(nd_cmd_in_size); 890 891 u32 nd_cmd_out_size(struct nvdimm *nvdimm, int cmd, 892 const struct nd_cmd_desc *desc, int idx, const u32 *in_field, 893 const u32 *out_field, unsigned long remainder) 894 { 895 if (idx >= desc->out_num) 896 return UINT_MAX; 897 898 if (desc->out_sizes[idx] < UINT_MAX) 899 return desc->out_sizes[idx]; 900 901 if (nvdimm && cmd == ND_CMD_GET_CONFIG_DATA && idx == 1) 902 return in_field[1]; 903 else if (nvdimm && cmd == ND_CMD_VENDOR && idx == 2) 904 return out_field[1]; 905 else if (!nvdimm && cmd == ND_CMD_ARS_STATUS && idx == 2) { 906 /* 907 * Per table 9-276 ARS Data in ACPI 6.1, out_field[1] is 908 * "Size of Output Buffer in bytes, including this 909 * field." 910 */ 911 if (out_field[1] < 4) 912 return 0; 913 /* 914 * ACPI 6.1 is ambiguous if 'status' is included in the 915 * output size. If we encounter an output size that 916 * overshoots the remainder by 4 bytes, assume it was 917 * including 'status'. 918 */ 919 if (out_field[1] - 4 == remainder) 920 return remainder; 921 return out_field[1] - 8; 922 } else if (cmd == ND_CMD_CALL) { 923 struct nd_cmd_pkg *pkg = (struct nd_cmd_pkg *) in_field; 924 925 return pkg->nd_size_out; 926 } 927 928 929 return UINT_MAX; 930 } 931 EXPORT_SYMBOL_GPL(nd_cmd_out_size); 932 933 void wait_nvdimm_bus_probe_idle(struct device *dev) 934 { 935 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev); 936 937 do { 938 if (nvdimm_bus->probe_active == 0) 939 break; 940 nvdimm_bus_unlock(dev); 941 device_unlock(dev); 942 wait_event(nvdimm_bus->wait, 943 nvdimm_bus->probe_active == 0); 944 device_lock(dev); 945 nvdimm_bus_lock(dev); 946 } while (true); 947 } 948 949 static int nd_pmem_forget_poison_check(struct device *dev, void *data) 950 { 951 struct nd_cmd_clear_error *clear_err = 952 (struct nd_cmd_clear_error *)data; 953 struct nd_btt *nd_btt = is_nd_btt(dev) ? to_nd_btt(dev) : NULL; 954 struct nd_pfn *nd_pfn = is_nd_pfn(dev) ? to_nd_pfn(dev) : NULL; 955 struct nd_dax *nd_dax = is_nd_dax(dev) ? to_nd_dax(dev) : NULL; 956 struct nd_namespace_common *ndns = NULL; 957 struct nd_namespace_io *nsio; 958 resource_size_t offset = 0, end_trunc = 0, start, end, pstart, pend; 959 960 if (nd_dax || !dev->driver) 961 return 0; 962 963 start = clear_err->address; 964 end = clear_err->address + clear_err->cleared - 1; 965 966 if (nd_btt || nd_pfn || nd_dax) { 967 if (nd_btt) 968 ndns = nd_btt->ndns; 969 else if (nd_pfn) 970 ndns = nd_pfn->ndns; 971 else if (nd_dax) 972 ndns = nd_dax->nd_pfn.ndns; 973 974 if (!ndns) 975 return 0; 976 } else 977 ndns = to_ndns(dev); 978 979 nsio = to_nd_namespace_io(&ndns->dev); 980 pstart = nsio->res.start + offset; 981 pend = nsio->res.end - end_trunc; 982 983 if ((pstart >= start) && (pend <= end)) 984 return -EBUSY; 985 986 return 0; 987 988 } 989 990 static int nd_ns_forget_poison_check(struct device *dev, void *data) 991 { 992 return device_for_each_child(dev, data, nd_pmem_forget_poison_check); 993 } 994 995 /* set_config requires an idle interleave set */ 996 static int nd_cmd_clear_to_send(struct nvdimm_bus *nvdimm_bus, 997 struct nvdimm *nvdimm, unsigned int cmd, void *data) 998 { 999 struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc; 1000 1001 /* ask the bus provider if it would like to block this request */ 1002 if (nd_desc->clear_to_send) { 1003 int rc = nd_desc->clear_to_send(nd_desc, nvdimm, cmd, data); 1004 1005 if (rc) 1006 return rc; 1007 } 1008 1009 /* require clear error to go through the pmem driver */ 1010 if (!nvdimm && cmd == ND_CMD_CLEAR_ERROR) 1011 return device_for_each_child(&nvdimm_bus->dev, data, 1012 nd_ns_forget_poison_check); 1013 1014 if (!nvdimm || cmd != ND_CMD_SET_CONFIG_DATA) 1015 return 0; 1016 1017 /* prevent label manipulation while the kernel owns label updates */ 1018 wait_nvdimm_bus_probe_idle(&nvdimm_bus->dev); 1019 if (atomic_read(&nvdimm->busy)) 1020 return -EBUSY; 1021 return 0; 1022 } 1023 1024 static int __nd_ioctl(struct nvdimm_bus *nvdimm_bus, struct nvdimm *nvdimm, 1025 int read_only, unsigned int ioctl_cmd, unsigned long arg) 1026 { 1027 struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc; 1028 const struct nd_cmd_desc *desc = NULL; 1029 unsigned int cmd = _IOC_NR(ioctl_cmd); 1030 struct device *dev = &nvdimm_bus->dev; 1031 void __user *p = (void __user *) arg; 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 u64 buf_len = 0; 1039 1040 if (nvdimm) { 1041 desc = nd_cmd_dimm_desc(cmd); 1042 cmd_name = nvdimm_cmd_name(cmd); 1043 cmd_mask = nvdimm->cmd_mask; 1044 dimm_name = dev_name(&nvdimm->dev); 1045 } else { 1046 desc = nd_cmd_bus_desc(cmd); 1047 cmd_name = nvdimm_bus_cmd_name(cmd); 1048 cmd_mask = nd_desc->cmd_mask; 1049 dimm_name = "bus"; 1050 } 1051 1052 /* Validate command family support against bus declared support */ 1053 if (cmd == ND_CMD_CALL) { 1054 unsigned long *mask; 1055 1056 if (copy_from_user(&pkg, p, sizeof(pkg))) 1057 return -EFAULT; 1058 1059 if (nvdimm) { 1060 if (pkg.nd_family > NVDIMM_FAMILY_MAX) 1061 return -EINVAL; 1062 mask = &nd_desc->dimm_family_mask; 1063 } else { 1064 if (pkg.nd_family > NVDIMM_BUS_FAMILY_MAX) 1065 return -EINVAL; 1066 mask = &nd_desc->bus_family_mask; 1067 } 1068 1069 if (!test_bit(pkg.nd_family, mask)) 1070 return -EINVAL; 1071 } 1072 1073 if (!desc || 1074 (desc->out_num + desc->in_num == 0) || 1075 cmd > ND_CMD_CALL || 1076 !test_bit(cmd, &cmd_mask)) 1077 return -ENOTTY; 1078 1079 /* fail write commands (when read-only) */ 1080 if (read_only) 1081 switch (cmd) { 1082 case ND_CMD_VENDOR: 1083 case ND_CMD_SET_CONFIG_DATA: 1084 case ND_CMD_ARS_START: 1085 case ND_CMD_CLEAR_ERROR: 1086 case ND_CMD_CALL: 1087 dev_dbg(dev, "'%s' command while read-only.\n", 1088 nvdimm ? nvdimm_cmd_name(cmd) 1089 : nvdimm_bus_cmd_name(cmd)); 1090 return -EPERM; 1091 default: 1092 break; 1093 } 1094 1095 /* process an input envelope */ 1096 char *in_env __free(kfree) = kzalloc(ND_CMD_MAX_ENVELOPE, GFP_KERNEL); 1097 if (!in_env) 1098 return -ENOMEM; 1099 for (i = 0; i < desc->in_num; i++) { 1100 u32 in_size, copy; 1101 1102 in_size = nd_cmd_in_size(nvdimm, cmd, desc, i, in_env); 1103 if (in_size == UINT_MAX) { 1104 dev_err(dev, "%s:%s unknown input size cmd: %s field: %d\n", 1105 __func__, dimm_name, cmd_name, i); 1106 return -ENXIO; 1107 } 1108 if (in_len < ND_CMD_MAX_ENVELOPE) 1109 copy = min_t(u32, ND_CMD_MAX_ENVELOPE - in_len, in_size); 1110 else 1111 copy = 0; 1112 if (copy && copy_from_user(&in_env[in_len], p + in_len, copy)) 1113 return -EFAULT; 1114 in_len += in_size; 1115 } 1116 1117 if (cmd == ND_CMD_CALL) { 1118 func = pkg.nd_command; 1119 dev_dbg(dev, "%s, idx: %llu, in: %u, out: %u, len %llu\n", 1120 dimm_name, pkg.nd_command, 1121 in_len, out_len, buf_len); 1122 } 1123 1124 /* process an output envelope */ 1125 char *out_env __free(kfree) = kzalloc(ND_CMD_MAX_ENVELOPE, GFP_KERNEL); 1126 if (!out_env) 1127 return -ENOMEM; 1128 1129 for (i = 0; i < desc->out_num; i++) { 1130 u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i, 1131 (u32 *) in_env, (u32 *) out_env, 0); 1132 u32 copy; 1133 1134 if (out_size == UINT_MAX) { 1135 dev_dbg(dev, "%s unknown output size cmd: %s field: %d\n", 1136 dimm_name, cmd_name, i); 1137 return -EFAULT; 1138 } 1139 if (out_len < ND_CMD_MAX_ENVELOPE) 1140 copy = min_t(u32, ND_CMD_MAX_ENVELOPE - out_len, out_size); 1141 else 1142 copy = 0; 1143 if (copy && copy_from_user(&out_env[out_len], 1144 p + in_len + out_len, copy)) { 1145 return -EFAULT; 1146 } 1147 out_len += out_size; 1148 } 1149 1150 buf_len = (u64) out_len + (u64) in_len; 1151 if (buf_len > ND_IOCTL_MAX_BUFLEN) { 1152 dev_dbg(dev, "%s cmd: %s buf_len: %llu > %d\n", dimm_name, 1153 cmd_name, buf_len, ND_IOCTL_MAX_BUFLEN); 1154 return -EINVAL; 1155 } 1156 1157 void *buf __free(kvfree) = kvzalloc(buf_len, GFP_KERNEL); 1158 if (!buf) 1159 return -ENOMEM; 1160 1161 if (copy_from_user(buf, p, buf_len)) 1162 return -EFAULT; 1163 1164 guard(device)(dev); 1165 guard(nvdimm_bus)(dev); 1166 rc = nd_cmd_clear_to_send(nvdimm_bus, nvdimm, func, buf); 1167 if (rc) 1168 return rc; 1169 1170 rc = nd_desc->ndctl(nd_desc, nvdimm, cmd, buf, buf_len, &cmd_rc); 1171 if (rc < 0) 1172 return rc; 1173 1174 if (!nvdimm && cmd == ND_CMD_CLEAR_ERROR && cmd_rc >= 0) { 1175 struct nd_cmd_clear_error *clear_err = buf; 1176 1177 nvdimm_account_cleared_poison(nvdimm_bus, clear_err->address, 1178 clear_err->cleared); 1179 } 1180 1181 if (copy_to_user(p, buf, buf_len)) 1182 return -EFAULT; 1183 1184 return 0; 1185 } 1186 1187 enum nd_ioctl_mode { 1188 BUS_IOCTL, 1189 DIMM_IOCTL, 1190 }; 1191 1192 static int match_dimm(struct device *dev, const void *data) 1193 { 1194 long id = (long) data; 1195 1196 if (is_nvdimm(dev)) { 1197 struct nvdimm *nvdimm = to_nvdimm(dev); 1198 1199 return nvdimm->id == id; 1200 } 1201 1202 return 0; 1203 } 1204 1205 static long nd_ioctl(struct file *file, unsigned int cmd, unsigned long arg, 1206 enum nd_ioctl_mode mode) 1207 1208 { 1209 struct nvdimm_bus *nvdimm_bus, *found = NULL; 1210 long id = (long) file->private_data; 1211 struct nvdimm *nvdimm = NULL; 1212 int rc, ro; 1213 1214 ro = ((file->f_flags & O_ACCMODE) == O_RDONLY); 1215 mutex_lock(&nvdimm_bus_list_mutex); 1216 list_for_each_entry(nvdimm_bus, &nvdimm_bus_list, list) { 1217 if (mode == DIMM_IOCTL) { 1218 struct device *dev; 1219 1220 dev = device_find_child(&nvdimm_bus->dev, 1221 file->private_data, match_dimm); 1222 if (!dev) 1223 continue; 1224 nvdimm = to_nvdimm(dev); 1225 found = nvdimm_bus; 1226 } else if (nvdimm_bus->id == id) { 1227 found = nvdimm_bus; 1228 } 1229 1230 if (found) { 1231 atomic_inc(&nvdimm_bus->ioctl_active); 1232 break; 1233 } 1234 } 1235 mutex_unlock(&nvdimm_bus_list_mutex); 1236 1237 if (!found) 1238 return -ENXIO; 1239 1240 nvdimm_bus = found; 1241 rc = __nd_ioctl(nvdimm_bus, nvdimm, ro, cmd, arg); 1242 1243 if (nvdimm) 1244 put_device(&nvdimm->dev); 1245 if (atomic_dec_and_test(&nvdimm_bus->ioctl_active)) 1246 wake_up(&nvdimm_bus->wait); 1247 1248 return rc; 1249 } 1250 1251 static long bus_ioctl(struct file *file, unsigned int cmd, unsigned long arg) 1252 { 1253 return nd_ioctl(file, cmd, arg, BUS_IOCTL); 1254 } 1255 1256 static long dimm_ioctl(struct file *file, unsigned int cmd, unsigned long arg) 1257 { 1258 return nd_ioctl(file, cmd, arg, DIMM_IOCTL); 1259 } 1260 1261 static int nd_open(struct inode *inode, struct file *file) 1262 { 1263 long minor = iminor(inode); 1264 1265 file->private_data = (void *) minor; 1266 return 0; 1267 } 1268 1269 static const struct file_operations nvdimm_bus_fops = { 1270 .owner = THIS_MODULE, 1271 .open = nd_open, 1272 .unlocked_ioctl = bus_ioctl, 1273 .compat_ioctl = compat_ptr_ioctl, 1274 .llseek = noop_llseek, 1275 }; 1276 1277 static const struct file_operations nvdimm_fops = { 1278 .owner = THIS_MODULE, 1279 .open = nd_open, 1280 .unlocked_ioctl = dimm_ioctl, 1281 .compat_ioctl = compat_ptr_ioctl, 1282 .llseek = noop_llseek, 1283 }; 1284 1285 int __init nvdimm_bus_init(void) 1286 { 1287 int rc; 1288 1289 rc = bus_register(&nvdimm_bus_type); 1290 if (rc) 1291 return rc; 1292 1293 rc = register_chrdev(0, "ndctl", &nvdimm_bus_fops); 1294 if (rc < 0) 1295 goto err_bus_chrdev; 1296 nvdimm_bus_major = rc; 1297 1298 rc = register_chrdev(0, "dimmctl", &nvdimm_fops); 1299 if (rc < 0) 1300 goto err_dimm_chrdev; 1301 nvdimm_major = rc; 1302 1303 rc = class_register(&nd_class); 1304 if (rc) 1305 goto err_class; 1306 1307 rc = driver_register(&nd_bus_driver.drv); 1308 if (rc) 1309 goto err_nd_bus; 1310 1311 return 0; 1312 1313 err_nd_bus: 1314 class_unregister(&nd_class); 1315 err_class: 1316 unregister_chrdev(nvdimm_major, "dimmctl"); 1317 err_dimm_chrdev: 1318 unregister_chrdev(nvdimm_bus_major, "ndctl"); 1319 err_bus_chrdev: 1320 bus_unregister(&nvdimm_bus_type); 1321 1322 return rc; 1323 } 1324 1325 void nvdimm_bus_exit(void) 1326 { 1327 driver_unregister(&nd_bus_driver.drv); 1328 class_unregister(&nd_class); 1329 unregister_chrdev(nvdimm_bus_major, "ndctl"); 1330 unregister_chrdev(nvdimm_major, "dimmctl"); 1331 bus_unregister(&nvdimm_bus_type); 1332 ida_destroy(&nd_ida); 1333 } 1334