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 DEFINE_MUTEX(dax_bus_lock); 14 15 /* 16 * All changes to the dax region configuration occur with this lock held 17 * for write. 18 */ 19 DECLARE_RWSEM(dax_region_rwsem); 20 21 /* 22 * All changes to the dax device configuration occur with this lock held 23 * for write. 24 */ 25 DECLARE_RWSEM(dax_dev_rwsem); 26 27 #define DAX_NAME_LEN 30 28 struct dax_id { 29 struct list_head list; 30 char dev_name[DAX_NAME_LEN]; 31 }; 32 33 static int dax_bus_uevent(const struct device *dev, struct kobj_uevent_env *env) 34 { 35 /* 36 * We only ever expect to handle device-dax instances, i.e. the 37 * @type argument to MODULE_ALIAS_DAX_DEVICE() is always zero 38 */ 39 return add_uevent_var(env, "MODALIAS=" DAX_DEVICE_MODALIAS_FMT, 0); 40 } 41 42 static struct dax_device_driver *to_dax_drv(struct device_driver *drv) 43 { 44 return container_of(drv, struct dax_device_driver, drv); 45 } 46 47 static struct dax_id *__dax_match_id(struct dax_device_driver *dax_drv, 48 const char *dev_name) 49 { 50 struct dax_id *dax_id; 51 52 lockdep_assert_held(&dax_bus_lock); 53 54 list_for_each_entry(dax_id, &dax_drv->ids, list) 55 if (sysfs_streq(dax_id->dev_name, dev_name)) 56 return dax_id; 57 return NULL; 58 } 59 60 static int dax_match_id(struct dax_device_driver *dax_drv, struct device *dev) 61 { 62 int match; 63 64 mutex_lock(&dax_bus_lock); 65 match = !!__dax_match_id(dax_drv, dev_name(dev)); 66 mutex_unlock(&dax_bus_lock); 67 68 return match; 69 } 70 71 static int dax_match_type(struct dax_device_driver *dax_drv, struct device *dev) 72 { 73 enum dax_driver_type type = DAXDRV_DEVICE_TYPE; 74 struct dev_dax *dev_dax = to_dev_dax(dev); 75 76 if (dev_dax->region->res.flags & IORESOURCE_DAX_KMEM) 77 type = DAXDRV_KMEM_TYPE; 78 79 if (dax_drv->type == type) 80 return 1; 81 82 /* default to device mode if dax_kmem is disabled */ 83 if (dax_drv->type == DAXDRV_DEVICE_TYPE && 84 !IS_ENABLED(CONFIG_DEV_DAX_KMEM)) 85 return 1; 86 87 return 0; 88 } 89 90 enum id_action { 91 ID_REMOVE, 92 ID_ADD, 93 }; 94 95 static ssize_t do_id_store(struct device_driver *drv, const char *buf, 96 size_t count, enum id_action action) 97 { 98 struct dax_device_driver *dax_drv = to_dax_drv(drv); 99 unsigned int region_id, id; 100 char devname[DAX_NAME_LEN]; 101 struct dax_id *dax_id; 102 ssize_t rc = count; 103 int fields; 104 105 fields = sscanf(buf, "dax%d.%d", ®ion_id, &id); 106 if (fields != 2) 107 return -EINVAL; 108 sprintf(devname, "dax%d.%d", region_id, id); 109 if (!sysfs_streq(buf, devname)) 110 return -EINVAL; 111 112 mutex_lock(&dax_bus_lock); 113 dax_id = __dax_match_id(dax_drv, buf); 114 if (!dax_id) { 115 if (action == ID_ADD) { 116 dax_id = kzalloc(sizeof(*dax_id), GFP_KERNEL); 117 if (dax_id) { 118 strscpy(dax_id->dev_name, buf, DAX_NAME_LEN); 119 list_add(&dax_id->list, &dax_drv->ids); 120 } else 121 rc = -ENOMEM; 122 } 123 } else if (action == ID_REMOVE) { 124 list_del(&dax_id->list); 125 kfree(dax_id); 126 } 127 mutex_unlock(&dax_bus_lock); 128 129 if (rc < 0) 130 return rc; 131 if (action == ID_ADD) 132 rc = driver_attach(drv); 133 if (rc) 134 return rc; 135 return count; 136 } 137 138 static ssize_t new_id_store(struct device_driver *drv, const char *buf, 139 size_t count) 140 { 141 return do_id_store(drv, buf, count, ID_ADD); 142 } 143 static DRIVER_ATTR_WO(new_id); 144 145 static ssize_t remove_id_store(struct device_driver *drv, const char *buf, 146 size_t count) 147 { 148 return do_id_store(drv, buf, count, ID_REMOVE); 149 } 150 static DRIVER_ATTR_WO(remove_id); 151 152 static struct attribute *dax_drv_attrs[] = { 153 &driver_attr_new_id.attr, 154 &driver_attr_remove_id.attr, 155 NULL, 156 }; 157 ATTRIBUTE_GROUPS(dax_drv); 158 159 static int dax_bus_match(struct device *dev, struct device_driver *drv); 160 161 /* 162 * Static dax regions are regions created by an external subsystem 163 * nvdimm where a single range is assigned. Its boundaries are by the external 164 * subsystem and are usually limited to one physical memory range. For example, 165 * for PMEM it is usually defined by NVDIMM Namespace boundaries (i.e. a 166 * single contiguous range) 167 * 168 * On dynamic dax regions, the assigned region can be partitioned by dax core 169 * into multiple subdivisions. A subdivision is represented into one 170 * /dev/daxN.M device composed by one or more potentially discontiguous ranges. 171 * 172 * When allocating a dax region, drivers must set whether it's static 173 * (IORESOURCE_DAX_STATIC). On static dax devices, the @pgmap is pre-assigned 174 * to dax core when calling devm_create_dev_dax(), whereas in dynamic dax 175 * devices it is NULL but afterwards allocated by dax core on device ->probe(). 176 * Care is needed to make sure that dynamic dax devices are torn down with a 177 * cleared @pgmap field (see kill_dev_dax()). 178 */ 179 static bool is_static(struct dax_region *dax_region) 180 { 181 return (dax_region->res.flags & IORESOURCE_DAX_STATIC) != 0; 182 } 183 184 bool static_dev_dax(struct dev_dax *dev_dax) 185 { 186 return is_static(dev_dax->region); 187 } 188 EXPORT_SYMBOL_GPL(static_dev_dax); 189 190 static u64 dev_dax_size(struct dev_dax *dev_dax) 191 { 192 u64 size = 0; 193 int i; 194 195 lockdep_assert_held(&dax_dev_rwsem); 196 197 for (i = 0; i < dev_dax->nr_range; i++) 198 size += range_len(&dev_dax->ranges[i].range); 199 200 return size; 201 } 202 203 static int dax_bus_probe(struct device *dev) 204 { 205 struct dax_device_driver *dax_drv = to_dax_drv(dev->driver); 206 struct dev_dax *dev_dax = to_dev_dax(dev); 207 struct dax_region *dax_region = dev_dax->region; 208 int rc; 209 u64 size; 210 211 rc = down_read_interruptible(&dax_dev_rwsem); 212 if (rc) 213 return rc; 214 size = dev_dax_size(dev_dax); 215 up_read(&dax_dev_rwsem); 216 217 if (size == 0 || dev_dax->id < 0) 218 return -ENXIO; 219 220 rc = dax_drv->probe(dev_dax); 221 222 if (rc || is_static(dax_region)) 223 return rc; 224 225 /* 226 * Track new seed creation only after successful probe of the 227 * previous seed. 228 */ 229 if (dax_region->seed == dev) 230 dax_region->seed = NULL; 231 232 return 0; 233 } 234 235 static void dax_bus_remove(struct device *dev) 236 { 237 struct dax_device_driver *dax_drv = to_dax_drv(dev->driver); 238 struct dev_dax *dev_dax = to_dev_dax(dev); 239 240 if (dax_drv->remove) 241 dax_drv->remove(dev_dax); 242 } 243 244 static const struct bus_type dax_bus_type = { 245 .name = "dax", 246 .uevent = dax_bus_uevent, 247 .match = dax_bus_match, 248 .probe = dax_bus_probe, 249 .remove = dax_bus_remove, 250 .drv_groups = dax_drv_groups, 251 }; 252 253 static int dax_bus_match(struct device *dev, struct device_driver *drv) 254 { 255 struct dax_device_driver *dax_drv = to_dax_drv(drv); 256 257 if (dax_match_id(dax_drv, dev)) 258 return 1; 259 return dax_match_type(dax_drv, dev); 260 } 261 262 /* 263 * Rely on the fact that drvdata is set before the attributes are 264 * registered, and that the attributes are unregistered before drvdata 265 * is cleared to assume that drvdata is always valid. 266 */ 267 static ssize_t id_show(struct device *dev, 268 struct device_attribute *attr, char *buf) 269 { 270 struct dax_region *dax_region = dev_get_drvdata(dev); 271 272 return sysfs_emit(buf, "%d\n", dax_region->id); 273 } 274 static DEVICE_ATTR_RO(id); 275 276 static ssize_t region_size_show(struct device *dev, 277 struct device_attribute *attr, char *buf) 278 { 279 struct dax_region *dax_region = dev_get_drvdata(dev); 280 281 return sysfs_emit(buf, "%llu\n", 282 (unsigned long long)resource_size(&dax_region->res)); 283 } 284 static struct device_attribute dev_attr_region_size = __ATTR(size, 0444, 285 region_size_show, NULL); 286 287 static ssize_t region_align_show(struct device *dev, 288 struct device_attribute *attr, char *buf) 289 { 290 struct dax_region *dax_region = dev_get_drvdata(dev); 291 292 return sysfs_emit(buf, "%u\n", dax_region->align); 293 } 294 static struct device_attribute dev_attr_region_align = 295 __ATTR(align, 0400, region_align_show, NULL); 296 297 #define for_each_dax_region_resource(dax_region, res) \ 298 for (res = (dax_region)->res.child; res; res = res->sibling) 299 300 static unsigned long long dax_region_avail_size(struct dax_region *dax_region) 301 { 302 resource_size_t size = resource_size(&dax_region->res); 303 struct resource *res; 304 305 lockdep_assert_held(&dax_region_rwsem); 306 307 for_each_dax_region_resource(dax_region, res) 308 size -= resource_size(res); 309 return size; 310 } 311 312 static ssize_t available_size_show(struct device *dev, 313 struct device_attribute *attr, char *buf) 314 { 315 struct dax_region *dax_region = dev_get_drvdata(dev); 316 unsigned long long size; 317 int rc; 318 319 rc = down_read_interruptible(&dax_region_rwsem); 320 if (rc) 321 return rc; 322 size = dax_region_avail_size(dax_region); 323 up_read(&dax_region_rwsem); 324 325 return sysfs_emit(buf, "%llu\n", size); 326 } 327 static DEVICE_ATTR_RO(available_size); 328 329 static ssize_t seed_show(struct device *dev, 330 struct device_attribute *attr, char *buf) 331 { 332 struct dax_region *dax_region = dev_get_drvdata(dev); 333 struct device *seed; 334 ssize_t rc; 335 336 if (is_static(dax_region)) 337 return -EINVAL; 338 339 rc = down_read_interruptible(&dax_region_rwsem); 340 if (rc) 341 return rc; 342 seed = dax_region->seed; 343 rc = sysfs_emit(buf, "%s\n", seed ? dev_name(seed) : ""); 344 up_read(&dax_region_rwsem); 345 346 return rc; 347 } 348 static DEVICE_ATTR_RO(seed); 349 350 static ssize_t create_show(struct device *dev, 351 struct device_attribute *attr, char *buf) 352 { 353 struct dax_region *dax_region = dev_get_drvdata(dev); 354 struct device *youngest; 355 ssize_t rc; 356 357 if (is_static(dax_region)) 358 return -EINVAL; 359 360 rc = down_read_interruptible(&dax_region_rwsem); 361 if (rc) 362 return rc; 363 youngest = dax_region->youngest; 364 rc = sysfs_emit(buf, "%s\n", youngest ? dev_name(youngest) : ""); 365 up_read(&dax_region_rwsem); 366 367 return rc; 368 } 369 370 static struct dev_dax *__devm_create_dev_dax(struct dev_dax_data *data); 371 372 static ssize_t create_store(struct device *dev, struct device_attribute *attr, 373 const char *buf, size_t len) 374 { 375 struct dax_region *dax_region = dev_get_drvdata(dev); 376 unsigned long long avail; 377 ssize_t rc; 378 int val; 379 380 if (is_static(dax_region)) 381 return -EINVAL; 382 383 rc = kstrtoint(buf, 0, &val); 384 if (rc) 385 return rc; 386 if (val != 1) 387 return -EINVAL; 388 389 rc = down_write_killable(&dax_region_rwsem); 390 if (rc) 391 return rc; 392 avail = dax_region_avail_size(dax_region); 393 if (avail == 0) 394 rc = -ENOSPC; 395 else { 396 struct dev_dax_data data = { 397 .dax_region = dax_region, 398 .size = 0, 399 .id = -1, 400 .memmap_on_memory = false, 401 }; 402 struct dev_dax *dev_dax = __devm_create_dev_dax(&data); 403 404 if (IS_ERR(dev_dax)) 405 rc = PTR_ERR(dev_dax); 406 else { 407 /* 408 * In support of crafting multiple new devices 409 * simultaneously multiple seeds can be created, 410 * but only the first one that has not been 411 * successfully bound is tracked as the region 412 * seed. 413 */ 414 if (!dax_region->seed) 415 dax_region->seed = &dev_dax->dev; 416 dax_region->youngest = &dev_dax->dev; 417 rc = len; 418 } 419 } 420 up_write(&dax_region_rwsem); 421 422 return rc; 423 } 424 static DEVICE_ATTR_RW(create); 425 426 void kill_dev_dax(struct dev_dax *dev_dax) 427 { 428 struct dax_device *dax_dev = dev_dax->dax_dev; 429 struct inode *inode = dax_inode(dax_dev); 430 431 kill_dax(dax_dev); 432 unmap_mapping_range(inode->i_mapping, 0, 0, 1); 433 434 /* 435 * Dynamic dax region have the pgmap allocated via dev_kzalloc() 436 * and thus freed by devm. Clear the pgmap to not have stale pgmap 437 * ranges on probe() from previous reconfigurations of region devices. 438 */ 439 if (!static_dev_dax(dev_dax)) 440 dev_dax->pgmap = NULL; 441 } 442 EXPORT_SYMBOL_GPL(kill_dev_dax); 443 444 static void trim_dev_dax_range(struct dev_dax *dev_dax) 445 { 446 int i = dev_dax->nr_range - 1; 447 struct range *range = &dev_dax->ranges[i].range; 448 struct dax_region *dax_region = dev_dax->region; 449 450 lockdep_assert_held_write(&dax_region_rwsem); 451 dev_dbg(&dev_dax->dev, "delete range[%d]: %#llx:%#llx\n", i, 452 (unsigned long long)range->start, 453 (unsigned long long)range->end); 454 455 __release_region(&dax_region->res, range->start, range_len(range)); 456 if (--dev_dax->nr_range == 0) { 457 kfree(dev_dax->ranges); 458 dev_dax->ranges = NULL; 459 } 460 } 461 462 static void free_dev_dax_ranges(struct dev_dax *dev_dax) 463 { 464 while (dev_dax->nr_range) 465 trim_dev_dax_range(dev_dax); 466 } 467 468 static void unregister_dev_dax(void *dev) 469 { 470 struct dev_dax *dev_dax = to_dev_dax(dev); 471 472 dev_dbg(dev, "%s\n", __func__); 473 474 down_write(&dax_region_rwsem); 475 kill_dev_dax(dev_dax); 476 device_del(dev); 477 free_dev_dax_ranges(dev_dax); 478 put_device(dev); 479 up_write(&dax_region_rwsem); 480 } 481 482 static void dax_region_free(struct kref *kref) 483 { 484 struct dax_region *dax_region; 485 486 dax_region = container_of(kref, struct dax_region, kref); 487 kfree(dax_region); 488 } 489 490 static void dax_region_put(struct dax_region *dax_region) 491 { 492 kref_put(&dax_region->kref, dax_region_free); 493 } 494 495 /* a return value >= 0 indicates this invocation invalidated the id */ 496 static int __free_dev_dax_id(struct dev_dax *dev_dax) 497 { 498 struct dax_region *dax_region; 499 int rc = dev_dax->id; 500 501 lockdep_assert_held_write(&dax_dev_rwsem); 502 503 if (!dev_dax->dyn_id || dev_dax->id < 0) 504 return -1; 505 dax_region = dev_dax->region; 506 ida_free(&dax_region->ida, dev_dax->id); 507 dax_region_put(dax_region); 508 dev_dax->id = -1; 509 return rc; 510 } 511 512 static int free_dev_dax_id(struct dev_dax *dev_dax) 513 { 514 int rc; 515 516 rc = down_write_killable(&dax_dev_rwsem); 517 if (rc) 518 return rc; 519 rc = __free_dev_dax_id(dev_dax); 520 up_write(&dax_dev_rwsem); 521 return rc; 522 } 523 524 static int alloc_dev_dax_id(struct dev_dax *dev_dax) 525 { 526 struct dax_region *dax_region = dev_dax->region; 527 int id; 528 529 id = ida_alloc(&dax_region->ida, GFP_KERNEL); 530 if (id < 0) 531 return id; 532 kref_get(&dax_region->kref); 533 dev_dax->dyn_id = true; 534 dev_dax->id = id; 535 return id; 536 } 537 538 static ssize_t delete_store(struct device *dev, struct device_attribute *attr, 539 const char *buf, size_t len) 540 { 541 struct dax_region *dax_region = dev_get_drvdata(dev); 542 struct dev_dax *dev_dax; 543 struct device *victim; 544 bool do_del = false; 545 int rc; 546 547 if (is_static(dax_region)) 548 return -EINVAL; 549 550 victim = device_find_child_by_name(dax_region->dev, buf); 551 if (!victim) 552 return -ENXIO; 553 554 device_lock(dev); 555 device_lock(victim); 556 dev_dax = to_dev_dax(victim); 557 down_write(&dax_dev_rwsem); 558 if (victim->driver || dev_dax_size(dev_dax)) 559 rc = -EBUSY; 560 else { 561 /* 562 * Invalidate the device so it does not become active 563 * again, but always preserve device-id-0 so that 564 * /sys/bus/dax/ is guaranteed to be populated while any 565 * dax_region is registered. 566 */ 567 if (dev_dax->id > 0) { 568 do_del = __free_dev_dax_id(dev_dax) >= 0; 569 rc = len; 570 if (dax_region->seed == victim) 571 dax_region->seed = NULL; 572 if (dax_region->youngest == victim) 573 dax_region->youngest = NULL; 574 } else 575 rc = -EBUSY; 576 } 577 up_write(&dax_dev_rwsem); 578 device_unlock(victim); 579 580 /* won the race to invalidate the device, clean it up */ 581 if (do_del) 582 devm_release_action(dev, unregister_dev_dax, victim); 583 device_unlock(dev); 584 put_device(victim); 585 586 return rc; 587 } 588 static DEVICE_ATTR_WO(delete); 589 590 static umode_t dax_region_visible(struct kobject *kobj, struct attribute *a, 591 int n) 592 { 593 struct device *dev = container_of(kobj, struct device, kobj); 594 struct dax_region *dax_region = dev_get_drvdata(dev); 595 596 if (is_static(dax_region)) 597 if (a == &dev_attr_available_size.attr 598 || a == &dev_attr_create.attr 599 || a == &dev_attr_seed.attr 600 || a == &dev_attr_delete.attr) 601 return 0; 602 return a->mode; 603 } 604 605 static struct attribute *dax_region_attributes[] = { 606 &dev_attr_available_size.attr, 607 &dev_attr_region_size.attr, 608 &dev_attr_region_align.attr, 609 &dev_attr_create.attr, 610 &dev_attr_seed.attr, 611 &dev_attr_delete.attr, 612 &dev_attr_id.attr, 613 NULL, 614 }; 615 616 static const struct attribute_group dax_region_attribute_group = { 617 .name = "dax_region", 618 .attrs = dax_region_attributes, 619 .is_visible = dax_region_visible, 620 }; 621 622 static const struct attribute_group *dax_region_attribute_groups[] = { 623 &dax_region_attribute_group, 624 NULL, 625 }; 626 627 static void dax_region_unregister(void *region) 628 { 629 struct dax_region *dax_region = region; 630 631 sysfs_remove_groups(&dax_region->dev->kobj, 632 dax_region_attribute_groups); 633 dax_region_put(dax_region); 634 } 635 636 struct dax_region *alloc_dax_region(struct device *parent, int region_id, 637 struct range *range, int target_node, unsigned int align, 638 unsigned long flags) 639 { 640 struct dax_region *dax_region; 641 642 /* 643 * The DAX core assumes that it can store its private data in 644 * parent->driver_data. This WARN is a reminder / safeguard for 645 * developers of device-dax drivers. 646 */ 647 if (dev_get_drvdata(parent)) { 648 dev_WARN(parent, "dax core failed to setup private data\n"); 649 return NULL; 650 } 651 652 if (!IS_ALIGNED(range->start, align) 653 || !IS_ALIGNED(range_len(range), align)) 654 return NULL; 655 656 dax_region = kzalloc(sizeof(*dax_region), GFP_KERNEL); 657 if (!dax_region) 658 return NULL; 659 660 dev_set_drvdata(parent, dax_region); 661 kref_init(&dax_region->kref); 662 dax_region->id = region_id; 663 dax_region->align = align; 664 dax_region->dev = parent; 665 dax_region->target_node = target_node; 666 ida_init(&dax_region->ida); 667 dax_region->res = (struct resource) { 668 .start = range->start, 669 .end = range->end, 670 .flags = IORESOURCE_MEM | flags, 671 }; 672 673 if (sysfs_create_groups(&parent->kobj, dax_region_attribute_groups)) { 674 kfree(dax_region); 675 return NULL; 676 } 677 678 if (devm_add_action_or_reset(parent, dax_region_unregister, dax_region)) 679 return NULL; 680 return dax_region; 681 } 682 EXPORT_SYMBOL_GPL(alloc_dax_region); 683 684 static void dax_mapping_release(struct device *dev) 685 { 686 struct dax_mapping *mapping = to_dax_mapping(dev); 687 struct device *parent = dev->parent; 688 struct dev_dax *dev_dax = to_dev_dax(parent); 689 690 ida_free(&dev_dax->ida, mapping->id); 691 kfree(mapping); 692 put_device(parent); 693 } 694 695 static void unregister_dax_mapping(void *data) 696 { 697 struct device *dev = data; 698 struct dax_mapping *mapping = to_dax_mapping(dev); 699 struct dev_dax *dev_dax = to_dev_dax(dev->parent); 700 701 dev_dbg(dev, "%s\n", __func__); 702 703 dev_dax->ranges[mapping->range_id].mapping = NULL; 704 mapping->range_id = -1; 705 706 device_unregister(dev); 707 } 708 709 static struct dev_dax_range *get_dax_range(struct device *dev) 710 { 711 struct dax_mapping *mapping = to_dax_mapping(dev); 712 struct dev_dax *dev_dax = to_dev_dax(dev->parent); 713 int rc; 714 715 rc = down_write_killable(&dax_region_rwsem); 716 if (rc) 717 return NULL; 718 if (mapping->range_id < 0) { 719 up_write(&dax_region_rwsem); 720 return NULL; 721 } 722 723 return &dev_dax->ranges[mapping->range_id]; 724 } 725 726 static void put_dax_range(void) 727 { 728 up_write(&dax_region_rwsem); 729 } 730 731 static ssize_t start_show(struct device *dev, 732 struct device_attribute *attr, char *buf) 733 { 734 struct dev_dax_range *dax_range; 735 ssize_t rc; 736 737 dax_range = get_dax_range(dev); 738 if (!dax_range) 739 return -ENXIO; 740 rc = sysfs_emit(buf, "%#llx\n", dax_range->range.start); 741 put_dax_range(); 742 743 return rc; 744 } 745 static DEVICE_ATTR(start, 0400, start_show, NULL); 746 747 static ssize_t end_show(struct device *dev, 748 struct device_attribute *attr, char *buf) 749 { 750 struct dev_dax_range *dax_range; 751 ssize_t rc; 752 753 dax_range = get_dax_range(dev); 754 if (!dax_range) 755 return -ENXIO; 756 rc = sysfs_emit(buf, "%#llx\n", dax_range->range.end); 757 put_dax_range(); 758 759 return rc; 760 } 761 static DEVICE_ATTR(end, 0400, end_show, NULL); 762 763 static ssize_t pgoff_show(struct device *dev, 764 struct device_attribute *attr, char *buf) 765 { 766 struct dev_dax_range *dax_range; 767 ssize_t rc; 768 769 dax_range = get_dax_range(dev); 770 if (!dax_range) 771 return -ENXIO; 772 rc = sysfs_emit(buf, "%#lx\n", dax_range->pgoff); 773 put_dax_range(); 774 775 return rc; 776 } 777 static DEVICE_ATTR(page_offset, 0400, pgoff_show, NULL); 778 779 static struct attribute *dax_mapping_attributes[] = { 780 &dev_attr_start.attr, 781 &dev_attr_end.attr, 782 &dev_attr_page_offset.attr, 783 NULL, 784 }; 785 786 static const struct attribute_group dax_mapping_attribute_group = { 787 .attrs = dax_mapping_attributes, 788 }; 789 790 static const struct attribute_group *dax_mapping_attribute_groups[] = { 791 &dax_mapping_attribute_group, 792 NULL, 793 }; 794 795 static const struct device_type dax_mapping_type = { 796 .release = dax_mapping_release, 797 .groups = dax_mapping_attribute_groups, 798 }; 799 800 static int devm_register_dax_mapping(struct dev_dax *dev_dax, int range_id) 801 { 802 struct dax_region *dax_region = dev_dax->region; 803 struct dax_mapping *mapping; 804 struct device *dev; 805 int rc; 806 807 lockdep_assert_held_write(&dax_region_rwsem); 808 809 if (dev_WARN_ONCE(&dev_dax->dev, !dax_region->dev->driver, 810 "region disabled\n")) 811 return -ENXIO; 812 813 mapping = kzalloc(sizeof(*mapping), GFP_KERNEL); 814 if (!mapping) 815 return -ENOMEM; 816 mapping->range_id = range_id; 817 mapping->id = ida_alloc(&dev_dax->ida, GFP_KERNEL); 818 if (mapping->id < 0) { 819 kfree(mapping); 820 return -ENOMEM; 821 } 822 dev_dax->ranges[range_id].mapping = mapping; 823 dev = &mapping->dev; 824 device_initialize(dev); 825 dev->parent = &dev_dax->dev; 826 get_device(dev->parent); 827 dev->type = &dax_mapping_type; 828 dev_set_name(dev, "mapping%d", mapping->id); 829 rc = device_add(dev); 830 if (rc) { 831 put_device(dev); 832 return rc; 833 } 834 835 rc = devm_add_action_or_reset(dax_region->dev, unregister_dax_mapping, 836 dev); 837 if (rc) 838 return rc; 839 return 0; 840 } 841 842 static int alloc_dev_dax_range(struct dev_dax *dev_dax, u64 start, 843 resource_size_t size) 844 { 845 struct dax_region *dax_region = dev_dax->region; 846 struct resource *res = &dax_region->res; 847 struct device *dev = &dev_dax->dev; 848 struct dev_dax_range *ranges; 849 unsigned long pgoff = 0; 850 struct resource *alloc; 851 int i, rc; 852 853 lockdep_assert_held_write(&dax_region_rwsem); 854 855 /* handle the seed alloc special case */ 856 if (!size) { 857 if (dev_WARN_ONCE(dev, dev_dax->nr_range, 858 "0-size allocation must be first\n")) 859 return -EBUSY; 860 /* nr_range == 0 is elsewhere special cased as 0-size device */ 861 return 0; 862 } 863 864 alloc = __request_region(res, start, size, dev_name(dev), 0); 865 if (!alloc) 866 return -ENOMEM; 867 868 ranges = krealloc(dev_dax->ranges, sizeof(*ranges) 869 * (dev_dax->nr_range + 1), GFP_KERNEL); 870 if (!ranges) { 871 __release_region(res, alloc->start, resource_size(alloc)); 872 return -ENOMEM; 873 } 874 875 for (i = 0; i < dev_dax->nr_range; i++) 876 pgoff += PHYS_PFN(range_len(&ranges[i].range)); 877 dev_dax->ranges = ranges; 878 ranges[dev_dax->nr_range++] = (struct dev_dax_range) { 879 .pgoff = pgoff, 880 .range = { 881 .start = alloc->start, 882 .end = alloc->end, 883 }, 884 }; 885 886 dev_dbg(dev, "alloc range[%d]: %pa:%pa\n", dev_dax->nr_range - 1, 887 &alloc->start, &alloc->end); 888 /* 889 * A dev_dax instance must be registered before mapping device 890 * children can be added. Defer to devm_create_dev_dax() to add 891 * the initial mapping device. 892 */ 893 if (!device_is_registered(&dev_dax->dev)) 894 return 0; 895 896 rc = devm_register_dax_mapping(dev_dax, dev_dax->nr_range - 1); 897 if (rc) 898 trim_dev_dax_range(dev_dax); 899 900 return rc; 901 } 902 903 static int adjust_dev_dax_range(struct dev_dax *dev_dax, struct resource *res, resource_size_t size) 904 { 905 int last_range = dev_dax->nr_range - 1; 906 struct dev_dax_range *dax_range = &dev_dax->ranges[last_range]; 907 bool is_shrink = resource_size(res) > size; 908 struct range *range = &dax_range->range; 909 struct device *dev = &dev_dax->dev; 910 int rc; 911 912 lockdep_assert_held_write(&dax_region_rwsem); 913 914 if (dev_WARN_ONCE(dev, !size, "deletion is handled by dev_dax_shrink\n")) 915 return -EINVAL; 916 917 rc = adjust_resource(res, range->start, size); 918 if (rc) 919 return rc; 920 921 *range = (struct range) { 922 .start = range->start, 923 .end = range->start + size - 1, 924 }; 925 926 dev_dbg(dev, "%s range[%d]: %#llx:%#llx\n", is_shrink ? "shrink" : "extend", 927 last_range, (unsigned long long) range->start, 928 (unsigned long long) range->end); 929 930 return 0; 931 } 932 933 static ssize_t size_show(struct device *dev, 934 struct device_attribute *attr, char *buf) 935 { 936 struct dev_dax *dev_dax = to_dev_dax(dev); 937 unsigned long long size; 938 int rc; 939 940 rc = down_read_interruptible(&dax_dev_rwsem); 941 if (rc) 942 return rc; 943 size = dev_dax_size(dev_dax); 944 up_read(&dax_dev_rwsem); 945 946 return sysfs_emit(buf, "%llu\n", size); 947 } 948 949 static bool alloc_is_aligned(struct dev_dax *dev_dax, resource_size_t size) 950 { 951 /* 952 * The minimum mapping granularity for a device instance is a 953 * single subsection, unless the arch says otherwise. 954 */ 955 return IS_ALIGNED(size, max_t(unsigned long, dev_dax->align, memremap_compat_align())); 956 } 957 958 static int dev_dax_shrink(struct dev_dax *dev_dax, resource_size_t size) 959 { 960 resource_size_t to_shrink = dev_dax_size(dev_dax) - size; 961 struct dax_region *dax_region = dev_dax->region; 962 struct device *dev = &dev_dax->dev; 963 int i; 964 965 for (i = dev_dax->nr_range - 1; i >= 0; i--) { 966 struct range *range = &dev_dax->ranges[i].range; 967 struct dax_mapping *mapping = dev_dax->ranges[i].mapping; 968 struct resource *adjust = NULL, *res; 969 resource_size_t shrink; 970 971 shrink = min_t(u64, to_shrink, range_len(range)); 972 if (shrink >= range_len(range)) { 973 devm_release_action(dax_region->dev, 974 unregister_dax_mapping, &mapping->dev); 975 trim_dev_dax_range(dev_dax); 976 to_shrink -= shrink; 977 if (!to_shrink) 978 break; 979 continue; 980 } 981 982 for_each_dax_region_resource(dax_region, res) 983 if (strcmp(res->name, dev_name(dev)) == 0 984 && res->start == range->start) { 985 adjust = res; 986 break; 987 } 988 989 if (dev_WARN_ONCE(dev, !adjust || i != dev_dax->nr_range - 1, 990 "failed to find matching resource\n")) 991 return -ENXIO; 992 return adjust_dev_dax_range(dev_dax, adjust, range_len(range) 993 - shrink); 994 } 995 return 0; 996 } 997 998 /* 999 * Only allow adjustments that preserve the relative pgoff of existing 1000 * allocations. I.e. the dev_dax->ranges array is ordered by increasing pgoff. 1001 */ 1002 static bool adjust_ok(struct dev_dax *dev_dax, struct resource *res) 1003 { 1004 struct dev_dax_range *last; 1005 int i; 1006 1007 if (dev_dax->nr_range == 0) 1008 return false; 1009 if (strcmp(res->name, dev_name(&dev_dax->dev)) != 0) 1010 return false; 1011 last = &dev_dax->ranges[dev_dax->nr_range - 1]; 1012 if (last->range.start != res->start || last->range.end != res->end) 1013 return false; 1014 for (i = 0; i < dev_dax->nr_range - 1; i++) { 1015 struct dev_dax_range *dax_range = &dev_dax->ranges[i]; 1016 1017 if (dax_range->pgoff > last->pgoff) 1018 return false; 1019 } 1020 1021 return true; 1022 } 1023 1024 static ssize_t dev_dax_resize(struct dax_region *dax_region, 1025 struct dev_dax *dev_dax, resource_size_t size) 1026 { 1027 resource_size_t avail = dax_region_avail_size(dax_region), to_alloc; 1028 resource_size_t dev_size = dev_dax_size(dev_dax); 1029 struct resource *region_res = &dax_region->res; 1030 struct device *dev = &dev_dax->dev; 1031 struct resource *res, *first; 1032 resource_size_t alloc = 0; 1033 int rc; 1034 1035 if (dev->driver) 1036 return -EBUSY; 1037 if (size == dev_size) 1038 return 0; 1039 if (size > dev_size && size - dev_size > avail) 1040 return -ENOSPC; 1041 if (size < dev_size) 1042 return dev_dax_shrink(dev_dax, size); 1043 1044 to_alloc = size - dev_size; 1045 if (dev_WARN_ONCE(dev, !alloc_is_aligned(dev_dax, to_alloc), 1046 "resize of %pa misaligned\n", &to_alloc)) 1047 return -ENXIO; 1048 1049 /* 1050 * Expand the device into the unused portion of the region. This 1051 * may involve adjusting the end of an existing resource, or 1052 * allocating a new resource. 1053 */ 1054 retry: 1055 first = region_res->child; 1056 if (!first) 1057 return alloc_dev_dax_range(dev_dax, dax_region->res.start, to_alloc); 1058 1059 rc = -ENOSPC; 1060 for (res = first; res; res = res->sibling) { 1061 struct resource *next = res->sibling; 1062 1063 /* space at the beginning of the region */ 1064 if (res == first && res->start > dax_region->res.start) { 1065 alloc = min(res->start - dax_region->res.start, to_alloc); 1066 rc = alloc_dev_dax_range(dev_dax, dax_region->res.start, alloc); 1067 break; 1068 } 1069 1070 alloc = 0; 1071 /* space between allocations */ 1072 if (next && next->start > res->end + 1) 1073 alloc = min(next->start - (res->end + 1), to_alloc); 1074 1075 /* space at the end of the region */ 1076 if (!alloc && !next && res->end < region_res->end) 1077 alloc = min(region_res->end - res->end, to_alloc); 1078 1079 if (!alloc) 1080 continue; 1081 1082 if (adjust_ok(dev_dax, res)) { 1083 rc = adjust_dev_dax_range(dev_dax, res, resource_size(res) + alloc); 1084 break; 1085 } 1086 rc = alloc_dev_dax_range(dev_dax, res->end + 1, alloc); 1087 break; 1088 } 1089 if (rc) 1090 return rc; 1091 to_alloc -= alloc; 1092 if (to_alloc) 1093 goto retry; 1094 return 0; 1095 } 1096 1097 static ssize_t size_store(struct device *dev, struct device_attribute *attr, 1098 const char *buf, size_t len) 1099 { 1100 ssize_t rc; 1101 unsigned long long val; 1102 struct dev_dax *dev_dax = to_dev_dax(dev); 1103 struct dax_region *dax_region = dev_dax->region; 1104 1105 rc = kstrtoull(buf, 0, &val); 1106 if (rc) 1107 return rc; 1108 1109 if (!alloc_is_aligned(dev_dax, val)) { 1110 dev_dbg(dev, "%s: size: %lld misaligned\n", __func__, val); 1111 return -EINVAL; 1112 } 1113 1114 rc = down_write_killable(&dax_region_rwsem); 1115 if (rc) 1116 return rc; 1117 if (!dax_region->dev->driver) { 1118 rc = -ENXIO; 1119 goto err_region; 1120 } 1121 rc = down_write_killable(&dax_dev_rwsem); 1122 if (rc) 1123 goto err_dev; 1124 1125 rc = dev_dax_resize(dax_region, dev_dax, val); 1126 1127 err_dev: 1128 up_write(&dax_dev_rwsem); 1129 err_region: 1130 up_write(&dax_region_rwsem); 1131 1132 if (rc == 0) 1133 return len; 1134 return rc; 1135 } 1136 static DEVICE_ATTR_RW(size); 1137 1138 static ssize_t range_parse(const char *opt, size_t len, struct range *range) 1139 { 1140 unsigned long long addr = 0; 1141 char *start, *end, *str; 1142 ssize_t rc = -EINVAL; 1143 1144 str = kstrdup(opt, GFP_KERNEL); 1145 if (!str) 1146 return rc; 1147 1148 end = str; 1149 start = strsep(&end, "-"); 1150 if (!start || !end) 1151 goto err; 1152 1153 rc = kstrtoull(start, 16, &addr); 1154 if (rc) 1155 goto err; 1156 range->start = addr; 1157 1158 rc = kstrtoull(end, 16, &addr); 1159 if (rc) 1160 goto err; 1161 range->end = addr; 1162 1163 err: 1164 kfree(str); 1165 return rc; 1166 } 1167 1168 static ssize_t mapping_store(struct device *dev, struct device_attribute *attr, 1169 const char *buf, size_t len) 1170 { 1171 struct dev_dax *dev_dax = to_dev_dax(dev); 1172 struct dax_region *dax_region = dev_dax->region; 1173 size_t to_alloc; 1174 struct range r; 1175 ssize_t rc; 1176 1177 rc = range_parse(buf, len, &r); 1178 if (rc) 1179 return rc; 1180 1181 rc = down_write_killable(&dax_region_rwsem); 1182 if (rc) 1183 return rc; 1184 if (!dax_region->dev->driver) { 1185 up_write(&dax_region_rwsem); 1186 return rc; 1187 } 1188 rc = down_write_killable(&dax_dev_rwsem); 1189 if (rc) { 1190 up_write(&dax_region_rwsem); 1191 return rc; 1192 } 1193 1194 to_alloc = range_len(&r); 1195 if (alloc_is_aligned(dev_dax, to_alloc)) 1196 rc = alloc_dev_dax_range(dev_dax, r.start, to_alloc); 1197 up_write(&dax_dev_rwsem); 1198 up_write(&dax_region_rwsem); 1199 1200 return rc == 0 ? len : rc; 1201 } 1202 static DEVICE_ATTR_WO(mapping); 1203 1204 static ssize_t align_show(struct device *dev, 1205 struct device_attribute *attr, char *buf) 1206 { 1207 struct dev_dax *dev_dax = to_dev_dax(dev); 1208 1209 return sysfs_emit(buf, "%d\n", dev_dax->align); 1210 } 1211 1212 static ssize_t dev_dax_validate_align(struct dev_dax *dev_dax) 1213 { 1214 struct device *dev = &dev_dax->dev; 1215 int i; 1216 1217 for (i = 0; i < dev_dax->nr_range; i++) { 1218 size_t len = range_len(&dev_dax->ranges[i].range); 1219 1220 if (!alloc_is_aligned(dev_dax, len)) { 1221 dev_dbg(dev, "%s: align %u invalid for range %d\n", 1222 __func__, dev_dax->align, i); 1223 return -EINVAL; 1224 } 1225 } 1226 1227 return 0; 1228 } 1229 1230 static ssize_t align_store(struct device *dev, struct device_attribute *attr, 1231 const char *buf, size_t len) 1232 { 1233 struct dev_dax *dev_dax = to_dev_dax(dev); 1234 struct dax_region *dax_region = dev_dax->region; 1235 unsigned long val, align_save; 1236 ssize_t rc; 1237 1238 rc = kstrtoul(buf, 0, &val); 1239 if (rc) 1240 return -ENXIO; 1241 1242 if (!dax_align_valid(val)) 1243 return -EINVAL; 1244 1245 rc = down_write_killable(&dax_region_rwsem); 1246 if (rc) 1247 return rc; 1248 if (!dax_region->dev->driver) { 1249 up_write(&dax_region_rwsem); 1250 return -ENXIO; 1251 } 1252 1253 rc = down_write_killable(&dax_dev_rwsem); 1254 if (rc) { 1255 up_write(&dax_region_rwsem); 1256 return rc; 1257 } 1258 if (dev->driver) { 1259 rc = -EBUSY; 1260 goto out_unlock; 1261 } 1262 1263 align_save = dev_dax->align; 1264 dev_dax->align = val; 1265 rc = dev_dax_validate_align(dev_dax); 1266 if (rc) 1267 dev_dax->align = align_save; 1268 out_unlock: 1269 up_write(&dax_dev_rwsem); 1270 up_write(&dax_region_rwsem); 1271 return rc == 0 ? len : rc; 1272 } 1273 static DEVICE_ATTR_RW(align); 1274 1275 static int dev_dax_target_node(struct dev_dax *dev_dax) 1276 { 1277 struct dax_region *dax_region = dev_dax->region; 1278 1279 return dax_region->target_node; 1280 } 1281 1282 static ssize_t target_node_show(struct device *dev, 1283 struct device_attribute *attr, char *buf) 1284 { 1285 struct dev_dax *dev_dax = to_dev_dax(dev); 1286 1287 return sysfs_emit(buf, "%d\n", dev_dax_target_node(dev_dax)); 1288 } 1289 static DEVICE_ATTR_RO(target_node); 1290 1291 static ssize_t resource_show(struct device *dev, 1292 struct device_attribute *attr, char *buf) 1293 { 1294 struct dev_dax *dev_dax = to_dev_dax(dev); 1295 struct dax_region *dax_region = dev_dax->region; 1296 unsigned long long start; 1297 1298 if (dev_dax->nr_range < 1) 1299 start = dax_region->res.start; 1300 else 1301 start = dev_dax->ranges[0].range.start; 1302 1303 return sysfs_emit(buf, "%#llx\n", start); 1304 } 1305 static DEVICE_ATTR(resource, 0400, resource_show, NULL); 1306 1307 static ssize_t modalias_show(struct device *dev, struct device_attribute *attr, 1308 char *buf) 1309 { 1310 /* 1311 * We only ever expect to handle device-dax instances, i.e. the 1312 * @type argument to MODULE_ALIAS_DAX_DEVICE() is always zero 1313 */ 1314 return sysfs_emit(buf, DAX_DEVICE_MODALIAS_FMT "\n", 0); 1315 } 1316 static DEVICE_ATTR_RO(modalias); 1317 1318 static ssize_t numa_node_show(struct device *dev, 1319 struct device_attribute *attr, char *buf) 1320 { 1321 return sysfs_emit(buf, "%d\n", dev_to_node(dev)); 1322 } 1323 static DEVICE_ATTR_RO(numa_node); 1324 1325 static ssize_t memmap_on_memory_show(struct device *dev, 1326 struct device_attribute *attr, char *buf) 1327 { 1328 struct dev_dax *dev_dax = to_dev_dax(dev); 1329 1330 return sysfs_emit(buf, "%d\n", dev_dax->memmap_on_memory); 1331 } 1332 1333 static ssize_t memmap_on_memory_store(struct device *dev, 1334 struct device_attribute *attr, 1335 const char *buf, size_t len) 1336 { 1337 struct dev_dax *dev_dax = to_dev_dax(dev); 1338 bool val; 1339 int rc; 1340 1341 rc = kstrtobool(buf, &val); 1342 if (rc) 1343 return rc; 1344 1345 if (val == true && !mhp_supports_memmap_on_memory()) { 1346 dev_dbg(dev, "memmap_on_memory is not available\n"); 1347 return -EOPNOTSUPP; 1348 } 1349 1350 rc = down_write_killable(&dax_dev_rwsem); 1351 if (rc) 1352 return rc; 1353 1354 if (dev_dax->memmap_on_memory != val && dev->driver && 1355 to_dax_drv(dev->driver)->type == DAXDRV_KMEM_TYPE) { 1356 up_write(&dax_dev_rwsem); 1357 return -EBUSY; 1358 } 1359 1360 dev_dax->memmap_on_memory = val; 1361 up_write(&dax_dev_rwsem); 1362 1363 return len; 1364 } 1365 static DEVICE_ATTR_RW(memmap_on_memory); 1366 1367 static umode_t dev_dax_visible(struct kobject *kobj, struct attribute *a, int n) 1368 { 1369 struct device *dev = container_of(kobj, struct device, kobj); 1370 struct dev_dax *dev_dax = to_dev_dax(dev); 1371 struct dax_region *dax_region = dev_dax->region; 1372 1373 if (a == &dev_attr_target_node.attr && dev_dax_target_node(dev_dax) < 0) 1374 return 0; 1375 if (a == &dev_attr_numa_node.attr && !IS_ENABLED(CONFIG_NUMA)) 1376 return 0; 1377 if (a == &dev_attr_mapping.attr && is_static(dax_region)) 1378 return 0; 1379 if ((a == &dev_attr_align.attr || 1380 a == &dev_attr_size.attr) && is_static(dax_region)) 1381 return 0444; 1382 return a->mode; 1383 } 1384 1385 static struct attribute *dev_dax_attributes[] = { 1386 &dev_attr_modalias.attr, 1387 &dev_attr_size.attr, 1388 &dev_attr_mapping.attr, 1389 &dev_attr_target_node.attr, 1390 &dev_attr_align.attr, 1391 &dev_attr_resource.attr, 1392 &dev_attr_numa_node.attr, 1393 &dev_attr_memmap_on_memory.attr, 1394 NULL, 1395 }; 1396 1397 static const struct attribute_group dev_dax_attribute_group = { 1398 .attrs = dev_dax_attributes, 1399 .is_visible = dev_dax_visible, 1400 }; 1401 1402 static const struct attribute_group *dax_attribute_groups[] = { 1403 &dev_dax_attribute_group, 1404 NULL, 1405 }; 1406 1407 static void dev_dax_release(struct device *dev) 1408 { 1409 struct dev_dax *dev_dax = to_dev_dax(dev); 1410 struct dax_device *dax_dev = dev_dax->dax_dev; 1411 1412 put_dax(dax_dev); 1413 free_dev_dax_id(dev_dax); 1414 kfree(dev_dax->pgmap); 1415 kfree(dev_dax); 1416 } 1417 1418 static const struct device_type dev_dax_type = { 1419 .release = dev_dax_release, 1420 .groups = dax_attribute_groups, 1421 }; 1422 1423 static struct dev_dax *__devm_create_dev_dax(struct dev_dax_data *data) 1424 { 1425 struct dax_region *dax_region = data->dax_region; 1426 struct device *parent = dax_region->dev; 1427 struct dax_device *dax_dev; 1428 struct dev_dax *dev_dax; 1429 struct inode *inode; 1430 struct device *dev; 1431 int rc; 1432 1433 dev_dax = kzalloc(sizeof(*dev_dax), GFP_KERNEL); 1434 if (!dev_dax) 1435 return ERR_PTR(-ENOMEM); 1436 1437 dev_dax->region = dax_region; 1438 if (is_static(dax_region)) { 1439 if (dev_WARN_ONCE(parent, data->id < 0, 1440 "dynamic id specified to static region\n")) { 1441 rc = -EINVAL; 1442 goto err_id; 1443 } 1444 1445 dev_dax->id = data->id; 1446 } else { 1447 if (dev_WARN_ONCE(parent, data->id >= 0, 1448 "static id specified to dynamic region\n")) { 1449 rc = -EINVAL; 1450 goto err_id; 1451 } 1452 1453 rc = alloc_dev_dax_id(dev_dax); 1454 if (rc < 0) 1455 goto err_id; 1456 } 1457 1458 dev = &dev_dax->dev; 1459 device_initialize(dev); 1460 dev_set_name(dev, "dax%d.%d", dax_region->id, dev_dax->id); 1461 1462 rc = alloc_dev_dax_range(dev_dax, dax_region->res.start, data->size); 1463 if (rc) 1464 goto err_range; 1465 1466 if (data->pgmap) { 1467 dev_WARN_ONCE(parent, !is_static(dax_region), 1468 "custom dev_pagemap requires a static dax_region\n"); 1469 1470 dev_dax->pgmap = kmemdup(data->pgmap, 1471 sizeof(struct dev_pagemap), GFP_KERNEL); 1472 if (!dev_dax->pgmap) { 1473 rc = -ENOMEM; 1474 goto err_pgmap; 1475 } 1476 } 1477 1478 /* 1479 * No dax_operations since there is no access to this device outside of 1480 * mmap of the resulting character device. 1481 */ 1482 dax_dev = alloc_dax(dev_dax, NULL); 1483 if (IS_ERR(dax_dev)) { 1484 rc = PTR_ERR(dax_dev); 1485 goto err_alloc_dax; 1486 } 1487 set_dax_synchronous(dax_dev); 1488 set_dax_nocache(dax_dev); 1489 set_dax_nomc(dax_dev); 1490 1491 /* a device_dax instance is dead while the driver is not attached */ 1492 kill_dax(dax_dev); 1493 1494 dev_dax->dax_dev = dax_dev; 1495 dev_dax->target_node = dax_region->target_node; 1496 dev_dax->align = dax_region->align; 1497 ida_init(&dev_dax->ida); 1498 1499 dev_dax->memmap_on_memory = data->memmap_on_memory; 1500 1501 inode = dax_inode(dax_dev); 1502 dev->devt = inode->i_rdev; 1503 dev->bus = &dax_bus_type; 1504 dev->parent = parent; 1505 dev->type = &dev_dax_type; 1506 1507 rc = device_add(dev); 1508 if (rc) { 1509 kill_dev_dax(dev_dax); 1510 put_device(dev); 1511 return ERR_PTR(rc); 1512 } 1513 1514 rc = devm_add_action_or_reset(dax_region->dev, unregister_dev_dax, dev); 1515 if (rc) 1516 return ERR_PTR(rc); 1517 1518 /* register mapping device for the initial allocation range */ 1519 if (dev_dax->nr_range && range_len(&dev_dax->ranges[0].range)) { 1520 rc = devm_register_dax_mapping(dev_dax, 0); 1521 if (rc) 1522 return ERR_PTR(rc); 1523 } 1524 1525 return dev_dax; 1526 1527 err_alloc_dax: 1528 kfree(dev_dax->pgmap); 1529 err_pgmap: 1530 free_dev_dax_ranges(dev_dax); 1531 err_range: 1532 free_dev_dax_id(dev_dax); 1533 err_id: 1534 kfree(dev_dax); 1535 1536 return ERR_PTR(rc); 1537 } 1538 1539 struct dev_dax *devm_create_dev_dax(struct dev_dax_data *data) 1540 { 1541 struct dev_dax *dev_dax; 1542 1543 down_write(&dax_region_rwsem); 1544 dev_dax = __devm_create_dev_dax(data); 1545 up_write(&dax_region_rwsem); 1546 1547 return dev_dax; 1548 } 1549 EXPORT_SYMBOL_GPL(devm_create_dev_dax); 1550 1551 int __dax_driver_register(struct dax_device_driver *dax_drv, 1552 struct module *module, const char *mod_name) 1553 { 1554 struct device_driver *drv = &dax_drv->drv; 1555 1556 /* 1557 * dax_bus_probe() calls dax_drv->probe() unconditionally. 1558 * So better be safe than sorry and ensure it is provided. 1559 */ 1560 if (!dax_drv->probe) 1561 return -EINVAL; 1562 1563 INIT_LIST_HEAD(&dax_drv->ids); 1564 drv->owner = module; 1565 drv->name = mod_name; 1566 drv->mod_name = mod_name; 1567 drv->bus = &dax_bus_type; 1568 1569 return driver_register(drv); 1570 } 1571 EXPORT_SYMBOL_GPL(__dax_driver_register); 1572 1573 void dax_driver_unregister(struct dax_device_driver *dax_drv) 1574 { 1575 struct device_driver *drv = &dax_drv->drv; 1576 struct dax_id *dax_id, *_id; 1577 1578 mutex_lock(&dax_bus_lock); 1579 list_for_each_entry_safe(dax_id, _id, &dax_drv->ids, list) { 1580 list_del(&dax_id->list); 1581 kfree(dax_id); 1582 } 1583 mutex_unlock(&dax_bus_lock); 1584 driver_unregister(drv); 1585 } 1586 EXPORT_SYMBOL_GPL(dax_driver_unregister); 1587 1588 int __init dax_bus_init(void) 1589 { 1590 return bus_register(&dax_bus_type); 1591 } 1592 1593 void __exit dax_bus_exit(void) 1594 { 1595 bus_unregister(&dax_bus_type); 1596 } 1597