1 // SPDX-License-Identifier: GPL-2.0 2 /* Copyright(c) 2017-2018 Intel Corporation. All rights reserved. */ 3 #include <linux/memremap.h> 4 #include <linux/device.h> 5 #include <linux/mutex.h> 6 #include <linux/list.h> 7 #include <linux/slab.h> 8 #include <linux/dax.h> 9 #include <linux/io.h> 10 #include "dax-private.h" 11 #include "bus.h" 12 13 static struct class *dax_class; 14 15 static DEFINE_MUTEX(dax_bus_lock); 16 17 #define DAX_NAME_LEN 30 18 struct dax_id { 19 struct list_head list; 20 char dev_name[DAX_NAME_LEN]; 21 }; 22 23 static int dax_bus_uevent(struct device *dev, struct kobj_uevent_env *env) 24 { 25 /* 26 * We only ever expect to handle device-dax instances, i.e. the 27 * @type argument to MODULE_ALIAS_DAX_DEVICE() is always zero 28 */ 29 return add_uevent_var(env, "MODALIAS=" DAX_DEVICE_MODALIAS_FMT, 0); 30 } 31 32 static struct dax_device_driver *to_dax_drv(struct device_driver *drv) 33 { 34 return container_of(drv, struct dax_device_driver, drv); 35 } 36 37 static struct dax_id *__dax_match_id(struct dax_device_driver *dax_drv, 38 const char *dev_name) 39 { 40 struct dax_id *dax_id; 41 42 lockdep_assert_held(&dax_bus_lock); 43 44 list_for_each_entry(dax_id, &dax_drv->ids, list) 45 if (sysfs_streq(dax_id->dev_name, dev_name)) 46 return dax_id; 47 return NULL; 48 } 49 50 static int dax_match_id(struct dax_device_driver *dax_drv, struct device *dev) 51 { 52 int match; 53 54 mutex_lock(&dax_bus_lock); 55 match = !!__dax_match_id(dax_drv, dev_name(dev)); 56 mutex_unlock(&dax_bus_lock); 57 58 return match; 59 } 60 61 enum id_action { 62 ID_REMOVE, 63 ID_ADD, 64 }; 65 66 static ssize_t do_id_store(struct device_driver *drv, const char *buf, 67 size_t count, enum id_action action) 68 { 69 struct dax_device_driver *dax_drv = to_dax_drv(drv); 70 unsigned int region_id, id; 71 char devname[DAX_NAME_LEN]; 72 struct dax_id *dax_id; 73 ssize_t rc = count; 74 int fields; 75 76 fields = sscanf(buf, "dax%d.%d", ®ion_id, &id); 77 if (fields != 2) 78 return -EINVAL; 79 sprintf(devname, "dax%d.%d", region_id, id); 80 if (!sysfs_streq(buf, devname)) 81 return -EINVAL; 82 83 mutex_lock(&dax_bus_lock); 84 dax_id = __dax_match_id(dax_drv, buf); 85 if (!dax_id) { 86 if (action == ID_ADD) { 87 dax_id = kzalloc(sizeof(*dax_id), GFP_KERNEL); 88 if (dax_id) { 89 strncpy(dax_id->dev_name, buf, DAX_NAME_LEN); 90 list_add(&dax_id->list, &dax_drv->ids); 91 } else 92 rc = -ENOMEM; 93 } 94 } else if (action == ID_REMOVE) { 95 list_del(&dax_id->list); 96 kfree(dax_id); 97 } 98 mutex_unlock(&dax_bus_lock); 99 100 if (rc < 0) 101 return rc; 102 if (action == ID_ADD) 103 rc = driver_attach(drv); 104 if (rc) 105 return rc; 106 return count; 107 } 108 109 static ssize_t new_id_store(struct device_driver *drv, const char *buf, 110 size_t count) 111 { 112 return do_id_store(drv, buf, count, ID_ADD); 113 } 114 static DRIVER_ATTR_WO(new_id); 115 116 static ssize_t remove_id_store(struct device_driver *drv, const char *buf, 117 size_t count) 118 { 119 return do_id_store(drv, buf, count, ID_REMOVE); 120 } 121 static DRIVER_ATTR_WO(remove_id); 122 123 static struct attribute *dax_drv_attrs[] = { 124 &driver_attr_new_id.attr, 125 &driver_attr_remove_id.attr, 126 NULL, 127 }; 128 ATTRIBUTE_GROUPS(dax_drv); 129 130 static int dax_bus_match(struct device *dev, struct device_driver *drv); 131 132 static bool is_static(struct dax_region *dax_region) 133 { 134 return (dax_region->res.flags & IORESOURCE_DAX_STATIC) != 0; 135 } 136 137 static u64 dev_dax_size(struct dev_dax *dev_dax) 138 { 139 u64 size = 0; 140 int i; 141 142 device_lock_assert(&dev_dax->dev); 143 144 for (i = 0; i < dev_dax->nr_range; i++) 145 size += range_len(&dev_dax->ranges[i].range); 146 147 return size; 148 } 149 150 static int dax_bus_probe(struct device *dev) 151 { 152 struct dax_device_driver *dax_drv = to_dax_drv(dev->driver); 153 struct dev_dax *dev_dax = to_dev_dax(dev); 154 struct dax_region *dax_region = dev_dax->region; 155 int rc; 156 157 if (dev_dax_size(dev_dax) == 0 || dev_dax->id < 0) 158 return -ENXIO; 159 160 rc = dax_drv->probe(dev_dax); 161 162 if (rc || is_static(dax_region)) 163 return rc; 164 165 /* 166 * Track new seed creation only after successful probe of the 167 * previous seed. 168 */ 169 if (dax_region->seed == dev) 170 dax_region->seed = NULL; 171 172 return 0; 173 } 174 175 static void dax_bus_remove(struct device *dev) 176 { 177 struct dax_device_driver *dax_drv = to_dax_drv(dev->driver); 178 struct dev_dax *dev_dax = to_dev_dax(dev); 179 180 if (dax_drv->remove) 181 dax_drv->remove(dev_dax); 182 } 183 184 static struct bus_type dax_bus_type = { 185 .name = "dax", 186 .uevent = dax_bus_uevent, 187 .match = dax_bus_match, 188 .probe = dax_bus_probe, 189 .remove = dax_bus_remove, 190 .drv_groups = dax_drv_groups, 191 }; 192 193 static int dax_bus_match(struct device *dev, struct device_driver *drv) 194 { 195 struct dax_device_driver *dax_drv = to_dax_drv(drv); 196 197 /* 198 * All but the 'device-dax' driver, which has 'match_always' 199 * set, requires an exact id match. 200 */ 201 if (dax_drv->match_always) 202 return 1; 203 204 return dax_match_id(dax_drv, dev); 205 } 206 207 /* 208 * Rely on the fact that drvdata is set before the attributes are 209 * registered, and that the attributes are unregistered before drvdata 210 * is cleared to assume that drvdata is always valid. 211 */ 212 static ssize_t id_show(struct device *dev, 213 struct device_attribute *attr, char *buf) 214 { 215 struct dax_region *dax_region = dev_get_drvdata(dev); 216 217 return sprintf(buf, "%d\n", dax_region->id); 218 } 219 static DEVICE_ATTR_RO(id); 220 221 static ssize_t region_size_show(struct device *dev, 222 struct device_attribute *attr, char *buf) 223 { 224 struct dax_region *dax_region = dev_get_drvdata(dev); 225 226 return sprintf(buf, "%llu\n", (unsigned long long) 227 resource_size(&dax_region->res)); 228 } 229 static struct device_attribute dev_attr_region_size = __ATTR(size, 0444, 230 region_size_show, NULL); 231 232 static ssize_t region_align_show(struct device *dev, 233 struct device_attribute *attr, char *buf) 234 { 235 struct dax_region *dax_region = dev_get_drvdata(dev); 236 237 return sprintf(buf, "%u\n", dax_region->align); 238 } 239 static struct device_attribute dev_attr_region_align = 240 __ATTR(align, 0400, region_align_show, NULL); 241 242 #define for_each_dax_region_resource(dax_region, res) \ 243 for (res = (dax_region)->res.child; res; res = res->sibling) 244 245 static unsigned long long dax_region_avail_size(struct dax_region *dax_region) 246 { 247 resource_size_t size = resource_size(&dax_region->res); 248 struct resource *res; 249 250 device_lock_assert(dax_region->dev); 251 252 for_each_dax_region_resource(dax_region, res) 253 size -= resource_size(res); 254 return size; 255 } 256 257 static ssize_t available_size_show(struct device *dev, 258 struct device_attribute *attr, char *buf) 259 { 260 struct dax_region *dax_region = dev_get_drvdata(dev); 261 unsigned long long size; 262 263 device_lock(dev); 264 size = dax_region_avail_size(dax_region); 265 device_unlock(dev); 266 267 return sprintf(buf, "%llu\n", size); 268 } 269 static DEVICE_ATTR_RO(available_size); 270 271 static ssize_t seed_show(struct device *dev, 272 struct device_attribute *attr, char *buf) 273 { 274 struct dax_region *dax_region = dev_get_drvdata(dev); 275 struct device *seed; 276 ssize_t rc; 277 278 if (is_static(dax_region)) 279 return -EINVAL; 280 281 device_lock(dev); 282 seed = dax_region->seed; 283 rc = sprintf(buf, "%s\n", seed ? dev_name(seed) : ""); 284 device_unlock(dev); 285 286 return rc; 287 } 288 static DEVICE_ATTR_RO(seed); 289 290 static ssize_t create_show(struct device *dev, 291 struct device_attribute *attr, char *buf) 292 { 293 struct dax_region *dax_region = dev_get_drvdata(dev); 294 struct device *youngest; 295 ssize_t rc; 296 297 if (is_static(dax_region)) 298 return -EINVAL; 299 300 device_lock(dev); 301 youngest = dax_region->youngest; 302 rc = sprintf(buf, "%s\n", youngest ? dev_name(youngest) : ""); 303 device_unlock(dev); 304 305 return rc; 306 } 307 308 static ssize_t create_store(struct device *dev, struct device_attribute *attr, 309 const char *buf, size_t len) 310 { 311 struct dax_region *dax_region = dev_get_drvdata(dev); 312 unsigned long long avail; 313 ssize_t rc; 314 int val; 315 316 if (is_static(dax_region)) 317 return -EINVAL; 318 319 rc = kstrtoint(buf, 0, &val); 320 if (rc) 321 return rc; 322 if (val != 1) 323 return -EINVAL; 324 325 device_lock(dev); 326 avail = dax_region_avail_size(dax_region); 327 if (avail == 0) 328 rc = -ENOSPC; 329 else { 330 struct dev_dax_data data = { 331 .dax_region = dax_region, 332 .size = 0, 333 .id = -1, 334 }; 335 struct dev_dax *dev_dax = devm_create_dev_dax(&data); 336 337 if (IS_ERR(dev_dax)) 338 rc = PTR_ERR(dev_dax); 339 else { 340 /* 341 * In support of crafting multiple new devices 342 * simultaneously multiple seeds can be created, 343 * but only the first one that has not been 344 * successfully bound is tracked as the region 345 * seed. 346 */ 347 if (!dax_region->seed) 348 dax_region->seed = &dev_dax->dev; 349 dax_region->youngest = &dev_dax->dev; 350 rc = len; 351 } 352 } 353 device_unlock(dev); 354 355 return rc; 356 } 357 static DEVICE_ATTR_RW(create); 358 359 void kill_dev_dax(struct dev_dax *dev_dax) 360 { 361 struct dax_device *dax_dev = dev_dax->dax_dev; 362 struct inode *inode = dax_inode(dax_dev); 363 364 kill_dax(dax_dev); 365 unmap_mapping_range(inode->i_mapping, 0, 0, 1); 366 } 367 EXPORT_SYMBOL_GPL(kill_dev_dax); 368 369 static void trim_dev_dax_range(struct dev_dax *dev_dax) 370 { 371 int i = dev_dax->nr_range - 1; 372 struct range *range = &dev_dax->ranges[i].range; 373 struct dax_region *dax_region = dev_dax->region; 374 375 device_lock_assert(dax_region->dev); 376 dev_dbg(&dev_dax->dev, "delete range[%d]: %#llx:%#llx\n", i, 377 (unsigned long long)range->start, 378 (unsigned long long)range->end); 379 380 __release_region(&dax_region->res, range->start, range_len(range)); 381 if (--dev_dax->nr_range == 0) { 382 kfree(dev_dax->ranges); 383 dev_dax->ranges = NULL; 384 } 385 } 386 387 static void free_dev_dax_ranges(struct dev_dax *dev_dax) 388 { 389 while (dev_dax->nr_range) 390 trim_dev_dax_range(dev_dax); 391 } 392 393 static void unregister_dev_dax(void *dev) 394 { 395 struct dev_dax *dev_dax = to_dev_dax(dev); 396 397 dev_dbg(dev, "%s\n", __func__); 398 399 kill_dev_dax(dev_dax); 400 free_dev_dax_ranges(dev_dax); 401 device_del(dev); 402 put_device(dev); 403 } 404 405 /* a return value >= 0 indicates this invocation invalidated the id */ 406 static int __free_dev_dax_id(struct dev_dax *dev_dax) 407 { 408 struct dax_region *dax_region = dev_dax->region; 409 struct device *dev = &dev_dax->dev; 410 int rc = dev_dax->id; 411 412 device_lock_assert(dev); 413 414 if (is_static(dax_region) || dev_dax->id < 0) 415 return -1; 416 ida_free(&dax_region->ida, dev_dax->id); 417 dev_dax->id = -1; 418 return rc; 419 } 420 421 static int free_dev_dax_id(struct dev_dax *dev_dax) 422 { 423 struct device *dev = &dev_dax->dev; 424 int rc; 425 426 device_lock(dev); 427 rc = __free_dev_dax_id(dev_dax); 428 device_unlock(dev); 429 return rc; 430 } 431 432 static ssize_t delete_store(struct device *dev, struct device_attribute *attr, 433 const char *buf, size_t len) 434 { 435 struct dax_region *dax_region = dev_get_drvdata(dev); 436 struct dev_dax *dev_dax; 437 struct device *victim; 438 bool do_del = false; 439 int rc; 440 441 if (is_static(dax_region)) 442 return -EINVAL; 443 444 victim = device_find_child_by_name(dax_region->dev, buf); 445 if (!victim) 446 return -ENXIO; 447 448 device_lock(dev); 449 device_lock(victim); 450 dev_dax = to_dev_dax(victim); 451 if (victim->driver || dev_dax_size(dev_dax)) 452 rc = -EBUSY; 453 else { 454 /* 455 * Invalidate the device so it does not become active 456 * again, but always preserve device-id-0 so that 457 * /sys/bus/dax/ is guaranteed to be populated while any 458 * dax_region is registered. 459 */ 460 if (dev_dax->id > 0) { 461 do_del = __free_dev_dax_id(dev_dax) >= 0; 462 rc = len; 463 if (dax_region->seed == victim) 464 dax_region->seed = NULL; 465 if (dax_region->youngest == victim) 466 dax_region->youngest = NULL; 467 } else 468 rc = -EBUSY; 469 } 470 device_unlock(victim); 471 472 /* won the race to invalidate the device, clean it up */ 473 if (do_del) 474 devm_release_action(dev, unregister_dev_dax, victim); 475 device_unlock(dev); 476 put_device(victim); 477 478 return rc; 479 } 480 static DEVICE_ATTR_WO(delete); 481 482 static umode_t dax_region_visible(struct kobject *kobj, struct attribute *a, 483 int n) 484 { 485 struct device *dev = container_of(kobj, struct device, kobj); 486 struct dax_region *dax_region = dev_get_drvdata(dev); 487 488 if (is_static(dax_region)) 489 if (a == &dev_attr_available_size.attr 490 || a == &dev_attr_create.attr 491 || a == &dev_attr_seed.attr 492 || a == &dev_attr_delete.attr) 493 return 0; 494 return a->mode; 495 } 496 497 static struct attribute *dax_region_attributes[] = { 498 &dev_attr_available_size.attr, 499 &dev_attr_region_size.attr, 500 &dev_attr_region_align.attr, 501 &dev_attr_create.attr, 502 &dev_attr_seed.attr, 503 &dev_attr_delete.attr, 504 &dev_attr_id.attr, 505 NULL, 506 }; 507 508 static const struct attribute_group dax_region_attribute_group = { 509 .name = "dax_region", 510 .attrs = dax_region_attributes, 511 .is_visible = dax_region_visible, 512 }; 513 514 static const struct attribute_group *dax_region_attribute_groups[] = { 515 &dax_region_attribute_group, 516 NULL, 517 }; 518 519 static void dax_region_free(struct kref *kref) 520 { 521 struct dax_region *dax_region; 522 523 dax_region = container_of(kref, struct dax_region, kref); 524 kfree(dax_region); 525 } 526 527 void dax_region_put(struct dax_region *dax_region) 528 { 529 kref_put(&dax_region->kref, dax_region_free); 530 } 531 EXPORT_SYMBOL_GPL(dax_region_put); 532 533 static void dax_region_unregister(void *region) 534 { 535 struct dax_region *dax_region = region; 536 537 sysfs_remove_groups(&dax_region->dev->kobj, 538 dax_region_attribute_groups); 539 dax_region_put(dax_region); 540 } 541 542 struct dax_region *alloc_dax_region(struct device *parent, int region_id, 543 struct range *range, int target_node, unsigned int align, 544 unsigned long flags) 545 { 546 struct dax_region *dax_region; 547 548 /* 549 * The DAX core assumes that it can store its private data in 550 * parent->driver_data. This WARN is a reminder / safeguard for 551 * developers of device-dax drivers. 552 */ 553 if (dev_get_drvdata(parent)) { 554 dev_WARN(parent, "dax core failed to setup private data\n"); 555 return NULL; 556 } 557 558 if (!IS_ALIGNED(range->start, align) 559 || !IS_ALIGNED(range_len(range), align)) 560 return NULL; 561 562 dax_region = kzalloc(sizeof(*dax_region), GFP_KERNEL); 563 if (!dax_region) 564 return NULL; 565 566 dev_set_drvdata(parent, dax_region); 567 kref_init(&dax_region->kref); 568 dax_region->id = region_id; 569 dax_region->align = align; 570 dax_region->dev = parent; 571 dax_region->target_node = target_node; 572 ida_init(&dax_region->ida); 573 dax_region->res = (struct resource) { 574 .start = range->start, 575 .end = range->end, 576 .flags = IORESOURCE_MEM | flags, 577 }; 578 579 if (sysfs_create_groups(&parent->kobj, dax_region_attribute_groups)) { 580 kfree(dax_region); 581 return NULL; 582 } 583 584 kref_get(&dax_region->kref); 585 if (devm_add_action_or_reset(parent, dax_region_unregister, dax_region)) 586 return NULL; 587 return dax_region; 588 } 589 EXPORT_SYMBOL_GPL(alloc_dax_region); 590 591 static void dax_mapping_release(struct device *dev) 592 { 593 struct dax_mapping *mapping = to_dax_mapping(dev); 594 struct dev_dax *dev_dax = to_dev_dax(dev->parent); 595 596 ida_free(&dev_dax->ida, mapping->id); 597 kfree(mapping); 598 } 599 600 static void unregister_dax_mapping(void *data) 601 { 602 struct device *dev = data; 603 struct dax_mapping *mapping = to_dax_mapping(dev); 604 struct dev_dax *dev_dax = to_dev_dax(dev->parent); 605 struct dax_region *dax_region = dev_dax->region; 606 607 dev_dbg(dev, "%s\n", __func__); 608 609 device_lock_assert(dax_region->dev); 610 611 dev_dax->ranges[mapping->range_id].mapping = NULL; 612 mapping->range_id = -1; 613 614 device_del(dev); 615 put_device(dev); 616 } 617 618 static struct dev_dax_range *get_dax_range(struct device *dev) 619 { 620 struct dax_mapping *mapping = to_dax_mapping(dev); 621 struct dev_dax *dev_dax = to_dev_dax(dev->parent); 622 struct dax_region *dax_region = dev_dax->region; 623 624 device_lock(dax_region->dev); 625 if (mapping->range_id < 0) { 626 device_unlock(dax_region->dev); 627 return NULL; 628 } 629 630 return &dev_dax->ranges[mapping->range_id]; 631 } 632 633 static void put_dax_range(struct dev_dax_range *dax_range) 634 { 635 struct dax_mapping *mapping = dax_range->mapping; 636 struct dev_dax *dev_dax = to_dev_dax(mapping->dev.parent); 637 struct dax_region *dax_region = dev_dax->region; 638 639 device_unlock(dax_region->dev); 640 } 641 642 static ssize_t start_show(struct device *dev, 643 struct device_attribute *attr, char *buf) 644 { 645 struct dev_dax_range *dax_range; 646 ssize_t rc; 647 648 dax_range = get_dax_range(dev); 649 if (!dax_range) 650 return -ENXIO; 651 rc = sprintf(buf, "%#llx\n", dax_range->range.start); 652 put_dax_range(dax_range); 653 654 return rc; 655 } 656 static DEVICE_ATTR(start, 0400, start_show, NULL); 657 658 static ssize_t end_show(struct device *dev, 659 struct device_attribute *attr, char *buf) 660 { 661 struct dev_dax_range *dax_range; 662 ssize_t rc; 663 664 dax_range = get_dax_range(dev); 665 if (!dax_range) 666 return -ENXIO; 667 rc = sprintf(buf, "%#llx\n", dax_range->range.end); 668 put_dax_range(dax_range); 669 670 return rc; 671 } 672 static DEVICE_ATTR(end, 0400, end_show, NULL); 673 674 static ssize_t pgoff_show(struct device *dev, 675 struct device_attribute *attr, char *buf) 676 { 677 struct dev_dax_range *dax_range; 678 ssize_t rc; 679 680 dax_range = get_dax_range(dev); 681 if (!dax_range) 682 return -ENXIO; 683 rc = sprintf(buf, "%#lx\n", dax_range->pgoff); 684 put_dax_range(dax_range); 685 686 return rc; 687 } 688 static DEVICE_ATTR(page_offset, 0400, pgoff_show, NULL); 689 690 static struct attribute *dax_mapping_attributes[] = { 691 &dev_attr_start.attr, 692 &dev_attr_end.attr, 693 &dev_attr_page_offset.attr, 694 NULL, 695 }; 696 697 static const struct attribute_group dax_mapping_attribute_group = { 698 .attrs = dax_mapping_attributes, 699 }; 700 701 static const struct attribute_group *dax_mapping_attribute_groups[] = { 702 &dax_mapping_attribute_group, 703 NULL, 704 }; 705 706 static struct device_type dax_mapping_type = { 707 .release = dax_mapping_release, 708 .groups = dax_mapping_attribute_groups, 709 }; 710 711 static int devm_register_dax_mapping(struct dev_dax *dev_dax, int range_id) 712 { 713 struct dax_region *dax_region = dev_dax->region; 714 struct dax_mapping *mapping; 715 struct device *dev; 716 int rc; 717 718 device_lock_assert(dax_region->dev); 719 720 if (dev_WARN_ONCE(&dev_dax->dev, !dax_region->dev->driver, 721 "region disabled\n")) 722 return -ENXIO; 723 724 mapping = kzalloc(sizeof(*mapping), GFP_KERNEL); 725 if (!mapping) 726 return -ENOMEM; 727 mapping->range_id = range_id; 728 mapping->id = ida_alloc(&dev_dax->ida, GFP_KERNEL); 729 if (mapping->id < 0) { 730 kfree(mapping); 731 return -ENOMEM; 732 } 733 dev_dax->ranges[range_id].mapping = mapping; 734 dev = &mapping->dev; 735 device_initialize(dev); 736 dev->parent = &dev_dax->dev; 737 dev->type = &dax_mapping_type; 738 dev_set_name(dev, "mapping%d", mapping->id); 739 rc = device_add(dev); 740 if (rc) { 741 put_device(dev); 742 return rc; 743 } 744 745 rc = devm_add_action_or_reset(dax_region->dev, unregister_dax_mapping, 746 dev); 747 if (rc) 748 return rc; 749 return 0; 750 } 751 752 static int alloc_dev_dax_range(struct dev_dax *dev_dax, u64 start, 753 resource_size_t size) 754 { 755 struct dax_region *dax_region = dev_dax->region; 756 struct resource *res = &dax_region->res; 757 struct device *dev = &dev_dax->dev; 758 struct dev_dax_range *ranges; 759 unsigned long pgoff = 0; 760 struct resource *alloc; 761 int i, rc; 762 763 device_lock_assert(dax_region->dev); 764 765 /* handle the seed alloc special case */ 766 if (!size) { 767 if (dev_WARN_ONCE(dev, dev_dax->nr_range, 768 "0-size allocation must be first\n")) 769 return -EBUSY; 770 /* nr_range == 0 is elsewhere special cased as 0-size device */ 771 return 0; 772 } 773 774 alloc = __request_region(res, start, size, dev_name(dev), 0); 775 if (!alloc) 776 return -ENOMEM; 777 778 ranges = krealloc(dev_dax->ranges, sizeof(*ranges) 779 * (dev_dax->nr_range + 1), GFP_KERNEL); 780 if (!ranges) { 781 __release_region(res, alloc->start, resource_size(alloc)); 782 return -ENOMEM; 783 } 784 785 for (i = 0; i < dev_dax->nr_range; i++) 786 pgoff += PHYS_PFN(range_len(&ranges[i].range)); 787 dev_dax->ranges = ranges; 788 ranges[dev_dax->nr_range++] = (struct dev_dax_range) { 789 .pgoff = pgoff, 790 .range = { 791 .start = alloc->start, 792 .end = alloc->end, 793 }, 794 }; 795 796 dev_dbg(dev, "alloc range[%d]: %pa:%pa\n", dev_dax->nr_range - 1, 797 &alloc->start, &alloc->end); 798 /* 799 * A dev_dax instance must be registered before mapping device 800 * children can be added. Defer to devm_create_dev_dax() to add 801 * the initial mapping device. 802 */ 803 if (!device_is_registered(&dev_dax->dev)) 804 return 0; 805 806 rc = devm_register_dax_mapping(dev_dax, dev_dax->nr_range - 1); 807 if (rc) 808 trim_dev_dax_range(dev_dax); 809 810 return rc; 811 } 812 813 static int adjust_dev_dax_range(struct dev_dax *dev_dax, struct resource *res, resource_size_t size) 814 { 815 int last_range = dev_dax->nr_range - 1; 816 struct dev_dax_range *dax_range = &dev_dax->ranges[last_range]; 817 struct dax_region *dax_region = dev_dax->region; 818 bool is_shrink = resource_size(res) > size; 819 struct range *range = &dax_range->range; 820 struct device *dev = &dev_dax->dev; 821 int rc; 822 823 device_lock_assert(dax_region->dev); 824 825 if (dev_WARN_ONCE(dev, !size, "deletion is handled by dev_dax_shrink\n")) 826 return -EINVAL; 827 828 rc = adjust_resource(res, range->start, size); 829 if (rc) 830 return rc; 831 832 *range = (struct range) { 833 .start = range->start, 834 .end = range->start + size - 1, 835 }; 836 837 dev_dbg(dev, "%s range[%d]: %#llx:%#llx\n", is_shrink ? "shrink" : "extend", 838 last_range, (unsigned long long) range->start, 839 (unsigned long long) range->end); 840 841 return 0; 842 } 843 844 static ssize_t size_show(struct device *dev, 845 struct device_attribute *attr, char *buf) 846 { 847 struct dev_dax *dev_dax = to_dev_dax(dev); 848 unsigned long long size; 849 850 device_lock(dev); 851 size = dev_dax_size(dev_dax); 852 device_unlock(dev); 853 854 return sprintf(buf, "%llu\n", size); 855 } 856 857 static bool alloc_is_aligned(struct dev_dax *dev_dax, resource_size_t size) 858 { 859 /* 860 * The minimum mapping granularity for a device instance is a 861 * single subsection, unless the arch says otherwise. 862 */ 863 return IS_ALIGNED(size, max_t(unsigned long, dev_dax->align, memremap_compat_align())); 864 } 865 866 static int dev_dax_shrink(struct dev_dax *dev_dax, resource_size_t size) 867 { 868 resource_size_t to_shrink = dev_dax_size(dev_dax) - size; 869 struct dax_region *dax_region = dev_dax->region; 870 struct device *dev = &dev_dax->dev; 871 int i; 872 873 for (i = dev_dax->nr_range - 1; i >= 0; i--) { 874 struct range *range = &dev_dax->ranges[i].range; 875 struct dax_mapping *mapping = dev_dax->ranges[i].mapping; 876 struct resource *adjust = NULL, *res; 877 resource_size_t shrink; 878 879 shrink = min_t(u64, to_shrink, range_len(range)); 880 if (shrink >= range_len(range)) { 881 devm_release_action(dax_region->dev, 882 unregister_dax_mapping, &mapping->dev); 883 trim_dev_dax_range(dev_dax); 884 to_shrink -= shrink; 885 if (!to_shrink) 886 break; 887 continue; 888 } 889 890 for_each_dax_region_resource(dax_region, res) 891 if (strcmp(res->name, dev_name(dev)) == 0 892 && res->start == range->start) { 893 adjust = res; 894 break; 895 } 896 897 if (dev_WARN_ONCE(dev, !adjust || i != dev_dax->nr_range - 1, 898 "failed to find matching resource\n")) 899 return -ENXIO; 900 return adjust_dev_dax_range(dev_dax, adjust, range_len(range) 901 - shrink); 902 } 903 return 0; 904 } 905 906 /* 907 * Only allow adjustments that preserve the relative pgoff of existing 908 * allocations. I.e. the dev_dax->ranges array is ordered by increasing pgoff. 909 */ 910 static bool adjust_ok(struct dev_dax *dev_dax, struct resource *res) 911 { 912 struct dev_dax_range *last; 913 int i; 914 915 if (dev_dax->nr_range == 0) 916 return false; 917 if (strcmp(res->name, dev_name(&dev_dax->dev)) != 0) 918 return false; 919 last = &dev_dax->ranges[dev_dax->nr_range - 1]; 920 if (last->range.start != res->start || last->range.end != res->end) 921 return false; 922 for (i = 0; i < dev_dax->nr_range - 1; i++) { 923 struct dev_dax_range *dax_range = &dev_dax->ranges[i]; 924 925 if (dax_range->pgoff > last->pgoff) 926 return false; 927 } 928 929 return true; 930 } 931 932 static ssize_t dev_dax_resize(struct dax_region *dax_region, 933 struct dev_dax *dev_dax, resource_size_t size) 934 { 935 resource_size_t avail = dax_region_avail_size(dax_region), to_alloc; 936 resource_size_t dev_size = dev_dax_size(dev_dax); 937 struct resource *region_res = &dax_region->res; 938 struct device *dev = &dev_dax->dev; 939 struct resource *res, *first; 940 resource_size_t alloc = 0; 941 int rc; 942 943 if (dev->driver) 944 return -EBUSY; 945 if (size == dev_size) 946 return 0; 947 if (size > dev_size && size - dev_size > avail) 948 return -ENOSPC; 949 if (size < dev_size) 950 return dev_dax_shrink(dev_dax, size); 951 952 to_alloc = size - dev_size; 953 if (dev_WARN_ONCE(dev, !alloc_is_aligned(dev_dax, to_alloc), 954 "resize of %pa misaligned\n", &to_alloc)) 955 return -ENXIO; 956 957 /* 958 * Expand the device into the unused portion of the region. This 959 * may involve adjusting the end of an existing resource, or 960 * allocating a new resource. 961 */ 962 retry: 963 first = region_res->child; 964 if (!first) 965 return alloc_dev_dax_range(dev_dax, dax_region->res.start, to_alloc); 966 967 rc = -ENOSPC; 968 for (res = first; res; res = res->sibling) { 969 struct resource *next = res->sibling; 970 971 /* space at the beginning of the region */ 972 if (res == first && res->start > dax_region->res.start) { 973 alloc = min(res->start - dax_region->res.start, to_alloc); 974 rc = alloc_dev_dax_range(dev_dax, dax_region->res.start, alloc); 975 break; 976 } 977 978 alloc = 0; 979 /* space between allocations */ 980 if (next && next->start > res->end + 1) 981 alloc = min(next->start - (res->end + 1), to_alloc); 982 983 /* space at the end of the region */ 984 if (!alloc && !next && res->end < region_res->end) 985 alloc = min(region_res->end - res->end, to_alloc); 986 987 if (!alloc) 988 continue; 989 990 if (adjust_ok(dev_dax, res)) { 991 rc = adjust_dev_dax_range(dev_dax, res, resource_size(res) + alloc); 992 break; 993 } 994 rc = alloc_dev_dax_range(dev_dax, res->end + 1, alloc); 995 break; 996 } 997 if (rc) 998 return rc; 999 to_alloc -= alloc; 1000 if (to_alloc) 1001 goto retry; 1002 return 0; 1003 } 1004 1005 static ssize_t size_store(struct device *dev, struct device_attribute *attr, 1006 const char *buf, size_t len) 1007 { 1008 ssize_t rc; 1009 unsigned long long val; 1010 struct dev_dax *dev_dax = to_dev_dax(dev); 1011 struct dax_region *dax_region = dev_dax->region; 1012 1013 rc = kstrtoull(buf, 0, &val); 1014 if (rc) 1015 return rc; 1016 1017 if (!alloc_is_aligned(dev_dax, val)) { 1018 dev_dbg(dev, "%s: size: %lld misaligned\n", __func__, val); 1019 return -EINVAL; 1020 } 1021 1022 device_lock(dax_region->dev); 1023 if (!dax_region->dev->driver) { 1024 device_unlock(dax_region->dev); 1025 return -ENXIO; 1026 } 1027 device_lock(dev); 1028 rc = dev_dax_resize(dax_region, dev_dax, val); 1029 device_unlock(dev); 1030 device_unlock(dax_region->dev); 1031 1032 return rc == 0 ? len : rc; 1033 } 1034 static DEVICE_ATTR_RW(size); 1035 1036 static ssize_t range_parse(const char *opt, size_t len, struct range *range) 1037 { 1038 unsigned long long addr = 0; 1039 char *start, *end, *str; 1040 ssize_t rc = -EINVAL; 1041 1042 str = kstrdup(opt, GFP_KERNEL); 1043 if (!str) 1044 return rc; 1045 1046 end = str; 1047 start = strsep(&end, "-"); 1048 if (!start || !end) 1049 goto err; 1050 1051 rc = kstrtoull(start, 16, &addr); 1052 if (rc) 1053 goto err; 1054 range->start = addr; 1055 1056 rc = kstrtoull(end, 16, &addr); 1057 if (rc) 1058 goto err; 1059 range->end = addr; 1060 1061 err: 1062 kfree(str); 1063 return rc; 1064 } 1065 1066 static ssize_t mapping_store(struct device *dev, struct device_attribute *attr, 1067 const char *buf, size_t len) 1068 { 1069 struct dev_dax *dev_dax = to_dev_dax(dev); 1070 struct dax_region *dax_region = dev_dax->region; 1071 size_t to_alloc; 1072 struct range r; 1073 ssize_t rc; 1074 1075 rc = range_parse(buf, len, &r); 1076 if (rc) 1077 return rc; 1078 1079 rc = -ENXIO; 1080 device_lock(dax_region->dev); 1081 if (!dax_region->dev->driver) { 1082 device_unlock(dax_region->dev); 1083 return rc; 1084 } 1085 device_lock(dev); 1086 1087 to_alloc = range_len(&r); 1088 if (alloc_is_aligned(dev_dax, to_alloc)) 1089 rc = alloc_dev_dax_range(dev_dax, r.start, to_alloc); 1090 device_unlock(dev); 1091 device_unlock(dax_region->dev); 1092 1093 return rc == 0 ? len : rc; 1094 } 1095 static DEVICE_ATTR_WO(mapping); 1096 1097 static ssize_t align_show(struct device *dev, 1098 struct device_attribute *attr, char *buf) 1099 { 1100 struct dev_dax *dev_dax = to_dev_dax(dev); 1101 1102 return sprintf(buf, "%d\n", dev_dax->align); 1103 } 1104 1105 static ssize_t dev_dax_validate_align(struct dev_dax *dev_dax) 1106 { 1107 struct device *dev = &dev_dax->dev; 1108 int i; 1109 1110 for (i = 0; i < dev_dax->nr_range; i++) { 1111 size_t len = range_len(&dev_dax->ranges[i].range); 1112 1113 if (!alloc_is_aligned(dev_dax, len)) { 1114 dev_dbg(dev, "%s: align %u invalid for range %d\n", 1115 __func__, dev_dax->align, i); 1116 return -EINVAL; 1117 } 1118 } 1119 1120 return 0; 1121 } 1122 1123 static ssize_t align_store(struct device *dev, struct device_attribute *attr, 1124 const char *buf, size_t len) 1125 { 1126 struct dev_dax *dev_dax = to_dev_dax(dev); 1127 struct dax_region *dax_region = dev_dax->region; 1128 unsigned long val, align_save; 1129 ssize_t rc; 1130 1131 rc = kstrtoul(buf, 0, &val); 1132 if (rc) 1133 return -ENXIO; 1134 1135 if (!dax_align_valid(val)) 1136 return -EINVAL; 1137 1138 device_lock(dax_region->dev); 1139 if (!dax_region->dev->driver) { 1140 device_unlock(dax_region->dev); 1141 return -ENXIO; 1142 } 1143 1144 device_lock(dev); 1145 if (dev->driver) { 1146 rc = -EBUSY; 1147 goto out_unlock; 1148 } 1149 1150 align_save = dev_dax->align; 1151 dev_dax->align = val; 1152 rc = dev_dax_validate_align(dev_dax); 1153 if (rc) 1154 dev_dax->align = align_save; 1155 out_unlock: 1156 device_unlock(dev); 1157 device_unlock(dax_region->dev); 1158 return rc == 0 ? len : rc; 1159 } 1160 static DEVICE_ATTR_RW(align); 1161 1162 static int dev_dax_target_node(struct dev_dax *dev_dax) 1163 { 1164 struct dax_region *dax_region = dev_dax->region; 1165 1166 return dax_region->target_node; 1167 } 1168 1169 static ssize_t target_node_show(struct device *dev, 1170 struct device_attribute *attr, char *buf) 1171 { 1172 struct dev_dax *dev_dax = to_dev_dax(dev); 1173 1174 return sprintf(buf, "%d\n", dev_dax_target_node(dev_dax)); 1175 } 1176 static DEVICE_ATTR_RO(target_node); 1177 1178 static ssize_t resource_show(struct device *dev, 1179 struct device_attribute *attr, char *buf) 1180 { 1181 struct dev_dax *dev_dax = to_dev_dax(dev); 1182 struct dax_region *dax_region = dev_dax->region; 1183 unsigned long long start; 1184 1185 if (dev_dax->nr_range < 1) 1186 start = dax_region->res.start; 1187 else 1188 start = dev_dax->ranges[0].range.start; 1189 1190 return sprintf(buf, "%#llx\n", start); 1191 } 1192 static DEVICE_ATTR(resource, 0400, resource_show, NULL); 1193 1194 static ssize_t modalias_show(struct device *dev, struct device_attribute *attr, 1195 char *buf) 1196 { 1197 /* 1198 * We only ever expect to handle device-dax instances, i.e. the 1199 * @type argument to MODULE_ALIAS_DAX_DEVICE() is always zero 1200 */ 1201 return sprintf(buf, DAX_DEVICE_MODALIAS_FMT "\n", 0); 1202 } 1203 static DEVICE_ATTR_RO(modalias); 1204 1205 static ssize_t numa_node_show(struct device *dev, 1206 struct device_attribute *attr, char *buf) 1207 { 1208 return sprintf(buf, "%d\n", dev_to_node(dev)); 1209 } 1210 static DEVICE_ATTR_RO(numa_node); 1211 1212 static umode_t dev_dax_visible(struct kobject *kobj, struct attribute *a, int n) 1213 { 1214 struct device *dev = container_of(kobj, struct device, kobj); 1215 struct dev_dax *dev_dax = to_dev_dax(dev); 1216 struct dax_region *dax_region = dev_dax->region; 1217 1218 if (a == &dev_attr_target_node.attr && dev_dax_target_node(dev_dax) < 0) 1219 return 0; 1220 if (a == &dev_attr_numa_node.attr && !IS_ENABLED(CONFIG_NUMA)) 1221 return 0; 1222 if (a == &dev_attr_mapping.attr && is_static(dax_region)) 1223 return 0; 1224 if ((a == &dev_attr_align.attr || 1225 a == &dev_attr_size.attr) && is_static(dax_region)) 1226 return 0444; 1227 return a->mode; 1228 } 1229 1230 static struct attribute *dev_dax_attributes[] = { 1231 &dev_attr_modalias.attr, 1232 &dev_attr_size.attr, 1233 &dev_attr_mapping.attr, 1234 &dev_attr_target_node.attr, 1235 &dev_attr_align.attr, 1236 &dev_attr_resource.attr, 1237 &dev_attr_numa_node.attr, 1238 NULL, 1239 }; 1240 1241 static const struct attribute_group dev_dax_attribute_group = { 1242 .attrs = dev_dax_attributes, 1243 .is_visible = dev_dax_visible, 1244 }; 1245 1246 static const struct attribute_group *dax_attribute_groups[] = { 1247 &dev_dax_attribute_group, 1248 NULL, 1249 }; 1250 1251 static void dev_dax_release(struct device *dev) 1252 { 1253 struct dev_dax *dev_dax = to_dev_dax(dev); 1254 struct dax_region *dax_region = dev_dax->region; 1255 struct dax_device *dax_dev = dev_dax->dax_dev; 1256 1257 put_dax(dax_dev); 1258 free_dev_dax_id(dev_dax); 1259 dax_region_put(dax_region); 1260 kfree(dev_dax->pgmap); 1261 kfree(dev_dax); 1262 } 1263 1264 static const struct device_type dev_dax_type = { 1265 .release = dev_dax_release, 1266 .groups = dax_attribute_groups, 1267 }; 1268 1269 struct dev_dax *devm_create_dev_dax(struct dev_dax_data *data) 1270 { 1271 struct dax_region *dax_region = data->dax_region; 1272 struct device *parent = dax_region->dev; 1273 struct dax_device *dax_dev; 1274 struct dev_dax *dev_dax; 1275 struct inode *inode; 1276 struct device *dev; 1277 int rc; 1278 1279 dev_dax = kzalloc(sizeof(*dev_dax), GFP_KERNEL); 1280 if (!dev_dax) 1281 return ERR_PTR(-ENOMEM); 1282 1283 if (is_static(dax_region)) { 1284 if (dev_WARN_ONCE(parent, data->id < 0, 1285 "dynamic id specified to static region\n")) { 1286 rc = -EINVAL; 1287 goto err_id; 1288 } 1289 1290 dev_dax->id = data->id; 1291 } else { 1292 if (dev_WARN_ONCE(parent, data->id >= 0, 1293 "static id specified to dynamic region\n")) { 1294 rc = -EINVAL; 1295 goto err_id; 1296 } 1297 1298 rc = ida_alloc(&dax_region->ida, GFP_KERNEL); 1299 if (rc < 0) 1300 goto err_id; 1301 dev_dax->id = rc; 1302 } 1303 1304 dev_dax->region = dax_region; 1305 dev = &dev_dax->dev; 1306 device_initialize(dev); 1307 dev_set_name(dev, "dax%d.%d", dax_region->id, dev_dax->id); 1308 1309 rc = alloc_dev_dax_range(dev_dax, dax_region->res.start, data->size); 1310 if (rc) 1311 goto err_range; 1312 1313 if (data->pgmap) { 1314 dev_WARN_ONCE(parent, !is_static(dax_region), 1315 "custom dev_pagemap requires a static dax_region\n"); 1316 1317 dev_dax->pgmap = kmemdup(data->pgmap, 1318 sizeof(struct dev_pagemap), GFP_KERNEL); 1319 if (!dev_dax->pgmap) { 1320 rc = -ENOMEM; 1321 goto err_pgmap; 1322 } 1323 } 1324 1325 /* 1326 * No 'host' or dax_operations since there is no access to this 1327 * device outside of mmap of the resulting character device. 1328 */ 1329 dax_dev = alloc_dax(dev_dax, NULL, NULL, DAXDEV_F_SYNC); 1330 if (IS_ERR(dax_dev)) { 1331 rc = PTR_ERR(dax_dev); 1332 goto err_alloc_dax; 1333 } 1334 1335 /* a device_dax instance is dead while the driver is not attached */ 1336 kill_dax(dax_dev); 1337 1338 dev_dax->dax_dev = dax_dev; 1339 dev_dax->target_node = dax_region->target_node; 1340 dev_dax->align = dax_region->align; 1341 ida_init(&dev_dax->ida); 1342 kref_get(&dax_region->kref); 1343 1344 inode = dax_inode(dax_dev); 1345 dev->devt = inode->i_rdev; 1346 if (data->subsys == DEV_DAX_BUS) 1347 dev->bus = &dax_bus_type; 1348 else 1349 dev->class = dax_class; 1350 dev->parent = parent; 1351 dev->type = &dev_dax_type; 1352 1353 rc = device_add(dev); 1354 if (rc) { 1355 kill_dev_dax(dev_dax); 1356 put_device(dev); 1357 return ERR_PTR(rc); 1358 } 1359 1360 rc = devm_add_action_or_reset(dax_region->dev, unregister_dev_dax, dev); 1361 if (rc) 1362 return ERR_PTR(rc); 1363 1364 /* register mapping device for the initial allocation range */ 1365 if (dev_dax->nr_range && range_len(&dev_dax->ranges[0].range)) { 1366 rc = devm_register_dax_mapping(dev_dax, 0); 1367 if (rc) 1368 return ERR_PTR(rc); 1369 } 1370 1371 return dev_dax; 1372 1373 err_alloc_dax: 1374 kfree(dev_dax->pgmap); 1375 err_pgmap: 1376 free_dev_dax_ranges(dev_dax); 1377 err_range: 1378 free_dev_dax_id(dev_dax); 1379 err_id: 1380 kfree(dev_dax); 1381 1382 return ERR_PTR(rc); 1383 } 1384 EXPORT_SYMBOL_GPL(devm_create_dev_dax); 1385 1386 static int match_always_count; 1387 1388 int __dax_driver_register(struct dax_device_driver *dax_drv, 1389 struct module *module, const char *mod_name) 1390 { 1391 struct device_driver *drv = &dax_drv->drv; 1392 int rc = 0; 1393 1394 /* 1395 * dax_bus_probe() calls dax_drv->probe() unconditionally. 1396 * So better be safe than sorry and ensure it is provided. 1397 */ 1398 if (!dax_drv->probe) 1399 return -EINVAL; 1400 1401 INIT_LIST_HEAD(&dax_drv->ids); 1402 drv->owner = module; 1403 drv->name = mod_name; 1404 drv->mod_name = mod_name; 1405 drv->bus = &dax_bus_type; 1406 1407 /* there can only be one default driver */ 1408 mutex_lock(&dax_bus_lock); 1409 match_always_count += dax_drv->match_always; 1410 if (match_always_count > 1) { 1411 match_always_count--; 1412 WARN_ON(1); 1413 rc = -EINVAL; 1414 } 1415 mutex_unlock(&dax_bus_lock); 1416 if (rc) 1417 return rc; 1418 1419 rc = driver_register(drv); 1420 if (rc && dax_drv->match_always) { 1421 mutex_lock(&dax_bus_lock); 1422 match_always_count -= dax_drv->match_always; 1423 mutex_unlock(&dax_bus_lock); 1424 } 1425 1426 return rc; 1427 } 1428 EXPORT_SYMBOL_GPL(__dax_driver_register); 1429 1430 void dax_driver_unregister(struct dax_device_driver *dax_drv) 1431 { 1432 struct device_driver *drv = &dax_drv->drv; 1433 struct dax_id *dax_id, *_id; 1434 1435 mutex_lock(&dax_bus_lock); 1436 match_always_count -= dax_drv->match_always; 1437 list_for_each_entry_safe(dax_id, _id, &dax_drv->ids, list) { 1438 list_del(&dax_id->list); 1439 kfree(dax_id); 1440 } 1441 mutex_unlock(&dax_bus_lock); 1442 driver_unregister(drv); 1443 } 1444 EXPORT_SYMBOL_GPL(dax_driver_unregister); 1445 1446 int __init dax_bus_init(void) 1447 { 1448 int rc; 1449 1450 if (IS_ENABLED(CONFIG_DEV_DAX_PMEM_COMPAT)) { 1451 dax_class = class_create(THIS_MODULE, "dax"); 1452 if (IS_ERR(dax_class)) 1453 return PTR_ERR(dax_class); 1454 } 1455 1456 rc = bus_register(&dax_bus_type); 1457 if (rc) 1458 class_destroy(dax_class); 1459 return rc; 1460 } 1461 1462 void __exit dax_bus_exit(void) 1463 { 1464 bus_unregister(&dax_bus_type); 1465 class_destroy(dax_class); 1466 } 1467