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