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