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 WARN_ON_ONCE(!rwsem_is_locked(&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 WARN_ON_ONCE(!rwsem_is_locked(&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 WARN_ON_ONCE(!rwsem_is_locked(&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 kill_dev_dax(dev_dax); 475 device_del(dev); 476 free_dev_dax_ranges(dev_dax); 477 put_device(dev); 478 } 479 480 static void unregister_dev_dax(void *dev) 481 { 482 if (rwsem_is_locked(&dax_region_rwsem)) 483 return __unregister_dev_dax(dev); 484 485 if (WARN_ON_ONCE(down_write_killable(&dax_region_rwsem) != 0)) 486 return; 487 __unregister_dev_dax(dev); 488 up_write(&dax_region_rwsem); 489 } 490 491 static void dax_region_free(struct kref *kref) 492 { 493 struct dax_region *dax_region; 494 495 dax_region = container_of(kref, struct dax_region, kref); 496 kfree(dax_region); 497 } 498 499 static void dax_region_put(struct dax_region *dax_region) 500 { 501 kref_put(&dax_region->kref, dax_region_free); 502 } 503 504 /* a return value >= 0 indicates this invocation invalidated the id */ 505 static int __free_dev_dax_id(struct dev_dax *dev_dax) 506 { 507 struct dax_region *dax_region; 508 int rc = dev_dax->id; 509 510 WARN_ON_ONCE(!rwsem_is_locked(&dax_dev_rwsem)); 511 512 if (!dev_dax->dyn_id || dev_dax->id < 0) 513 return -1; 514 dax_region = dev_dax->region; 515 ida_free(&dax_region->ida, dev_dax->id); 516 dax_region_put(dax_region); 517 dev_dax->id = -1; 518 return rc; 519 } 520 521 static int free_dev_dax_id(struct dev_dax *dev_dax) 522 { 523 int rc; 524 525 rc = down_write_killable(&dax_dev_rwsem); 526 if (rc) 527 return rc; 528 rc = __free_dev_dax_id(dev_dax); 529 up_write(&dax_dev_rwsem); 530 return rc; 531 } 532 533 static int alloc_dev_dax_id(struct dev_dax *dev_dax) 534 { 535 struct dax_region *dax_region = dev_dax->region; 536 int id; 537 538 id = ida_alloc(&dax_region->ida, GFP_KERNEL); 539 if (id < 0) 540 return id; 541 kref_get(&dax_region->kref); 542 dev_dax->dyn_id = true; 543 dev_dax->id = id; 544 return id; 545 } 546 547 static ssize_t delete_store(struct device *dev, struct device_attribute *attr, 548 const char *buf, size_t len) 549 { 550 struct dax_region *dax_region = dev_get_drvdata(dev); 551 struct dev_dax *dev_dax; 552 struct device *victim; 553 bool do_del = false; 554 int rc; 555 556 if (is_static(dax_region)) 557 return -EINVAL; 558 559 victim = device_find_child_by_name(dax_region->dev, buf); 560 if (!victim) 561 return -ENXIO; 562 563 rc = down_write_killable(&dax_region_rwsem); 564 if (rc) 565 return rc; 566 rc = down_write_killable(&dax_dev_rwsem); 567 if (rc) { 568 up_write(&dax_region_rwsem); 569 return rc; 570 } 571 dev_dax = to_dev_dax(victim); 572 if (victim->driver || dev_dax_size(dev_dax)) 573 rc = -EBUSY; 574 else { 575 /* 576 * Invalidate the device so it does not become active 577 * again, but always preserve device-id-0 so that 578 * /sys/bus/dax/ is guaranteed to be populated while any 579 * dax_region is registered. 580 */ 581 if (dev_dax->id > 0) { 582 do_del = __free_dev_dax_id(dev_dax) >= 0; 583 rc = len; 584 if (dax_region->seed == victim) 585 dax_region->seed = NULL; 586 if (dax_region->youngest == victim) 587 dax_region->youngest = NULL; 588 } else 589 rc = -EBUSY; 590 } 591 up_write(&dax_dev_rwsem); 592 593 /* won the race to invalidate the device, clean it up */ 594 if (do_del) 595 devm_release_action(dev, unregister_dev_dax, victim); 596 up_write(&dax_region_rwsem); 597 put_device(victim); 598 599 return rc; 600 } 601 static DEVICE_ATTR_WO(delete); 602 603 static umode_t dax_region_visible(struct kobject *kobj, struct attribute *a, 604 int n) 605 { 606 struct device *dev = container_of(kobj, struct device, kobj); 607 struct dax_region *dax_region = dev_get_drvdata(dev); 608 609 if (is_static(dax_region)) 610 if (a == &dev_attr_available_size.attr 611 || a == &dev_attr_create.attr 612 || a == &dev_attr_seed.attr 613 || a == &dev_attr_delete.attr) 614 return 0; 615 return a->mode; 616 } 617 618 static struct attribute *dax_region_attributes[] = { 619 &dev_attr_available_size.attr, 620 &dev_attr_region_size.attr, 621 &dev_attr_region_align.attr, 622 &dev_attr_create.attr, 623 &dev_attr_seed.attr, 624 &dev_attr_delete.attr, 625 &dev_attr_id.attr, 626 NULL, 627 }; 628 629 static const struct attribute_group dax_region_attribute_group = { 630 .name = "dax_region", 631 .attrs = dax_region_attributes, 632 .is_visible = dax_region_visible, 633 }; 634 635 static const struct attribute_group *dax_region_attribute_groups[] = { 636 &dax_region_attribute_group, 637 NULL, 638 }; 639 640 static void dax_region_unregister(void *region) 641 { 642 struct dax_region *dax_region = region; 643 644 sysfs_remove_groups(&dax_region->dev->kobj, 645 dax_region_attribute_groups); 646 dax_region_put(dax_region); 647 } 648 649 struct dax_region *alloc_dax_region(struct device *parent, int region_id, 650 struct range *range, int target_node, unsigned int align, 651 unsigned long flags) 652 { 653 struct dax_region *dax_region; 654 655 /* 656 * The DAX core assumes that it can store its private data in 657 * parent->driver_data. This WARN is a reminder / safeguard for 658 * developers of device-dax drivers. 659 */ 660 if (dev_get_drvdata(parent)) { 661 dev_WARN(parent, "dax core failed to setup private data\n"); 662 return NULL; 663 } 664 665 if (!IS_ALIGNED(range->start, align) 666 || !IS_ALIGNED(range_len(range), align)) 667 return NULL; 668 669 dax_region = kzalloc(sizeof(*dax_region), GFP_KERNEL); 670 if (!dax_region) 671 return NULL; 672 673 dev_set_drvdata(parent, dax_region); 674 kref_init(&dax_region->kref); 675 dax_region->id = region_id; 676 dax_region->align = align; 677 dax_region->dev = parent; 678 dax_region->target_node = target_node; 679 ida_init(&dax_region->ida); 680 dax_region->res = (struct resource) { 681 .start = range->start, 682 .end = range->end, 683 .flags = IORESOURCE_MEM | flags, 684 }; 685 686 if (sysfs_create_groups(&parent->kobj, dax_region_attribute_groups)) { 687 kfree(dax_region); 688 return NULL; 689 } 690 691 if (devm_add_action_or_reset(parent, dax_region_unregister, dax_region)) 692 return NULL; 693 return dax_region; 694 } 695 EXPORT_SYMBOL_GPL(alloc_dax_region); 696 697 static void dax_mapping_release(struct device *dev) 698 { 699 struct dax_mapping *mapping = to_dax_mapping(dev); 700 struct device *parent = dev->parent; 701 struct dev_dax *dev_dax = to_dev_dax(parent); 702 703 ida_free(&dev_dax->ida, mapping->id); 704 kfree(mapping); 705 put_device(parent); 706 } 707 708 static void __unregister_dax_mapping(void *data) 709 { 710 struct device *dev = data; 711 struct dax_mapping *mapping = to_dax_mapping(dev); 712 struct dev_dax *dev_dax = to_dev_dax(dev->parent); 713 714 dev_dbg(dev, "%s\n", __func__); 715 716 WARN_ON_ONCE(!rwsem_is_locked(&dax_region_rwsem)); 717 718 dev_dax->ranges[mapping->range_id].mapping = NULL; 719 mapping->range_id = -1; 720 721 device_unregister(dev); 722 } 723 724 static void unregister_dax_mapping(void *data) 725 { 726 if (rwsem_is_locked(&dax_region_rwsem)) 727 return __unregister_dax_mapping(data); 728 729 if (WARN_ON_ONCE(down_write_killable(&dax_region_rwsem) != 0)) 730 return; 731 __unregister_dax_mapping(data); 732 up_write(&dax_region_rwsem); 733 } 734 735 static struct dev_dax_range *get_dax_range(struct device *dev) 736 { 737 struct dax_mapping *mapping = to_dax_mapping(dev); 738 struct dev_dax *dev_dax = to_dev_dax(dev->parent); 739 int rc; 740 741 rc = down_write_killable(&dax_region_rwsem); 742 if (rc) 743 return NULL; 744 if (mapping->range_id < 0) { 745 up_write(&dax_region_rwsem); 746 return NULL; 747 } 748 749 return &dev_dax->ranges[mapping->range_id]; 750 } 751 752 static void put_dax_range(void) 753 { 754 up_write(&dax_region_rwsem); 755 } 756 757 static ssize_t start_show(struct device *dev, 758 struct device_attribute *attr, char *buf) 759 { 760 struct dev_dax_range *dax_range; 761 ssize_t rc; 762 763 dax_range = get_dax_range(dev); 764 if (!dax_range) 765 return -ENXIO; 766 rc = sysfs_emit(buf, "%#llx\n", dax_range->range.start); 767 put_dax_range(); 768 769 return rc; 770 } 771 static DEVICE_ATTR(start, 0400, start_show, NULL); 772 773 static ssize_t end_show(struct device *dev, 774 struct device_attribute *attr, char *buf) 775 { 776 struct dev_dax_range *dax_range; 777 ssize_t rc; 778 779 dax_range = get_dax_range(dev); 780 if (!dax_range) 781 return -ENXIO; 782 rc = sysfs_emit(buf, "%#llx\n", dax_range->range.end); 783 put_dax_range(); 784 785 return rc; 786 } 787 static DEVICE_ATTR(end, 0400, end_show, NULL); 788 789 static ssize_t pgoff_show(struct device *dev, 790 struct device_attribute *attr, char *buf) 791 { 792 struct dev_dax_range *dax_range; 793 ssize_t rc; 794 795 dax_range = get_dax_range(dev); 796 if (!dax_range) 797 return -ENXIO; 798 rc = sysfs_emit(buf, "%#lx\n", dax_range->pgoff); 799 put_dax_range(); 800 801 return rc; 802 } 803 static DEVICE_ATTR(page_offset, 0400, pgoff_show, NULL); 804 805 static struct attribute *dax_mapping_attributes[] = { 806 &dev_attr_start.attr, 807 &dev_attr_end.attr, 808 &dev_attr_page_offset.attr, 809 NULL, 810 }; 811 812 static const struct attribute_group dax_mapping_attribute_group = { 813 .attrs = dax_mapping_attributes, 814 }; 815 816 static const struct attribute_group *dax_mapping_attribute_groups[] = { 817 &dax_mapping_attribute_group, 818 NULL, 819 }; 820 821 static struct device_type dax_mapping_type = { 822 .release = dax_mapping_release, 823 .groups = dax_mapping_attribute_groups, 824 }; 825 826 static int devm_register_dax_mapping(struct dev_dax *dev_dax, int range_id) 827 { 828 struct dax_region *dax_region = dev_dax->region; 829 struct dax_mapping *mapping; 830 struct device *dev; 831 int rc; 832 833 WARN_ON_ONCE(!rwsem_is_locked(&dax_region_rwsem)); 834 835 if (dev_WARN_ONCE(&dev_dax->dev, !dax_region->dev->driver, 836 "region disabled\n")) 837 return -ENXIO; 838 839 mapping = kzalloc(sizeof(*mapping), GFP_KERNEL); 840 if (!mapping) 841 return -ENOMEM; 842 mapping->range_id = range_id; 843 mapping->id = ida_alloc(&dev_dax->ida, GFP_KERNEL); 844 if (mapping->id < 0) { 845 kfree(mapping); 846 return -ENOMEM; 847 } 848 dev_dax->ranges[range_id].mapping = mapping; 849 dev = &mapping->dev; 850 device_initialize(dev); 851 dev->parent = &dev_dax->dev; 852 get_device(dev->parent); 853 dev->type = &dax_mapping_type; 854 dev_set_name(dev, "mapping%d", mapping->id); 855 rc = device_add(dev); 856 if (rc) { 857 put_device(dev); 858 return rc; 859 } 860 861 rc = devm_add_action_or_reset(dax_region->dev, unregister_dax_mapping, 862 dev); 863 if (rc) 864 return rc; 865 return 0; 866 } 867 868 static int alloc_dev_dax_range(struct dev_dax *dev_dax, u64 start, 869 resource_size_t size) 870 { 871 struct dax_region *dax_region = dev_dax->region; 872 struct resource *res = &dax_region->res; 873 struct device *dev = &dev_dax->dev; 874 struct dev_dax_range *ranges; 875 unsigned long pgoff = 0; 876 struct resource *alloc; 877 int i, rc; 878 879 WARN_ON_ONCE(!rwsem_is_locked(&dax_region_rwsem)); 880 881 /* handle the seed alloc special case */ 882 if (!size) { 883 if (dev_WARN_ONCE(dev, dev_dax->nr_range, 884 "0-size allocation must be first\n")) 885 return -EBUSY; 886 /* nr_range == 0 is elsewhere special cased as 0-size device */ 887 return 0; 888 } 889 890 alloc = __request_region(res, start, size, dev_name(dev), 0); 891 if (!alloc) 892 return -ENOMEM; 893 894 ranges = krealloc(dev_dax->ranges, sizeof(*ranges) 895 * (dev_dax->nr_range + 1), GFP_KERNEL); 896 if (!ranges) { 897 __release_region(res, alloc->start, resource_size(alloc)); 898 return -ENOMEM; 899 } 900 901 for (i = 0; i < dev_dax->nr_range; i++) 902 pgoff += PHYS_PFN(range_len(&ranges[i].range)); 903 dev_dax->ranges = ranges; 904 ranges[dev_dax->nr_range++] = (struct dev_dax_range) { 905 .pgoff = pgoff, 906 .range = { 907 .start = alloc->start, 908 .end = alloc->end, 909 }, 910 }; 911 912 dev_dbg(dev, "alloc range[%d]: %pa:%pa\n", dev_dax->nr_range - 1, 913 &alloc->start, &alloc->end); 914 /* 915 * A dev_dax instance must be registered before mapping device 916 * children can be added. Defer to devm_create_dev_dax() to add 917 * the initial mapping device. 918 */ 919 if (!device_is_registered(&dev_dax->dev)) 920 return 0; 921 922 rc = devm_register_dax_mapping(dev_dax, dev_dax->nr_range - 1); 923 if (rc) 924 trim_dev_dax_range(dev_dax); 925 926 return rc; 927 } 928 929 static int adjust_dev_dax_range(struct dev_dax *dev_dax, struct resource *res, resource_size_t size) 930 { 931 int last_range = dev_dax->nr_range - 1; 932 struct dev_dax_range *dax_range = &dev_dax->ranges[last_range]; 933 bool is_shrink = resource_size(res) > size; 934 struct range *range = &dax_range->range; 935 struct device *dev = &dev_dax->dev; 936 int rc; 937 938 WARN_ON_ONCE(!rwsem_is_locked(&dax_region_rwsem)); 939 940 if (dev_WARN_ONCE(dev, !size, "deletion is handled by dev_dax_shrink\n")) 941 return -EINVAL; 942 943 rc = adjust_resource(res, range->start, size); 944 if (rc) 945 return rc; 946 947 *range = (struct range) { 948 .start = range->start, 949 .end = range->start + size - 1, 950 }; 951 952 dev_dbg(dev, "%s range[%d]: %#llx:%#llx\n", is_shrink ? "shrink" : "extend", 953 last_range, (unsigned long long) range->start, 954 (unsigned long long) range->end); 955 956 return 0; 957 } 958 959 static ssize_t size_show(struct device *dev, 960 struct device_attribute *attr, char *buf) 961 { 962 struct dev_dax *dev_dax = to_dev_dax(dev); 963 unsigned long long size; 964 int rc; 965 966 rc = down_write_killable(&dax_dev_rwsem); 967 if (rc) 968 return rc; 969 size = dev_dax_size(dev_dax); 970 up_write(&dax_dev_rwsem); 971 972 return sysfs_emit(buf, "%llu\n", size); 973 } 974 975 static bool alloc_is_aligned(struct dev_dax *dev_dax, resource_size_t size) 976 { 977 /* 978 * The minimum mapping granularity for a device instance is a 979 * single subsection, unless the arch says otherwise. 980 */ 981 return IS_ALIGNED(size, max_t(unsigned long, dev_dax->align, memremap_compat_align())); 982 } 983 984 static int dev_dax_shrink(struct dev_dax *dev_dax, resource_size_t size) 985 { 986 resource_size_t to_shrink = dev_dax_size(dev_dax) - size; 987 struct dax_region *dax_region = dev_dax->region; 988 struct device *dev = &dev_dax->dev; 989 int i; 990 991 for (i = dev_dax->nr_range - 1; i >= 0; i--) { 992 struct range *range = &dev_dax->ranges[i].range; 993 struct dax_mapping *mapping = dev_dax->ranges[i].mapping; 994 struct resource *adjust = NULL, *res; 995 resource_size_t shrink; 996 997 shrink = min_t(u64, to_shrink, range_len(range)); 998 if (shrink >= range_len(range)) { 999 devm_release_action(dax_region->dev, 1000 unregister_dax_mapping, &mapping->dev); 1001 trim_dev_dax_range(dev_dax); 1002 to_shrink -= shrink; 1003 if (!to_shrink) 1004 break; 1005 continue; 1006 } 1007 1008 for_each_dax_region_resource(dax_region, res) 1009 if (strcmp(res->name, dev_name(dev)) == 0 1010 && res->start == range->start) { 1011 adjust = res; 1012 break; 1013 } 1014 1015 if (dev_WARN_ONCE(dev, !adjust || i != dev_dax->nr_range - 1, 1016 "failed to find matching resource\n")) 1017 return -ENXIO; 1018 return adjust_dev_dax_range(dev_dax, adjust, range_len(range) 1019 - shrink); 1020 } 1021 return 0; 1022 } 1023 1024 /* 1025 * Only allow adjustments that preserve the relative pgoff of existing 1026 * allocations. I.e. the dev_dax->ranges array is ordered by increasing pgoff. 1027 */ 1028 static bool adjust_ok(struct dev_dax *dev_dax, struct resource *res) 1029 { 1030 struct dev_dax_range *last; 1031 int i; 1032 1033 if (dev_dax->nr_range == 0) 1034 return false; 1035 if (strcmp(res->name, dev_name(&dev_dax->dev)) != 0) 1036 return false; 1037 last = &dev_dax->ranges[dev_dax->nr_range - 1]; 1038 if (last->range.start != res->start || last->range.end != res->end) 1039 return false; 1040 for (i = 0; i < dev_dax->nr_range - 1; i++) { 1041 struct dev_dax_range *dax_range = &dev_dax->ranges[i]; 1042 1043 if (dax_range->pgoff > last->pgoff) 1044 return false; 1045 } 1046 1047 return true; 1048 } 1049 1050 static ssize_t dev_dax_resize(struct dax_region *dax_region, 1051 struct dev_dax *dev_dax, resource_size_t size) 1052 { 1053 resource_size_t avail = dax_region_avail_size(dax_region), to_alloc; 1054 resource_size_t dev_size = dev_dax_size(dev_dax); 1055 struct resource *region_res = &dax_region->res; 1056 struct device *dev = &dev_dax->dev; 1057 struct resource *res, *first; 1058 resource_size_t alloc = 0; 1059 int rc; 1060 1061 if (dev->driver) 1062 return -EBUSY; 1063 if (size == dev_size) 1064 return 0; 1065 if (size > dev_size && size - dev_size > avail) 1066 return -ENOSPC; 1067 if (size < dev_size) 1068 return dev_dax_shrink(dev_dax, size); 1069 1070 to_alloc = size - dev_size; 1071 if (dev_WARN_ONCE(dev, !alloc_is_aligned(dev_dax, to_alloc), 1072 "resize of %pa misaligned\n", &to_alloc)) 1073 return -ENXIO; 1074 1075 /* 1076 * Expand the device into the unused portion of the region. This 1077 * may involve adjusting the end of an existing resource, or 1078 * allocating a new resource. 1079 */ 1080 retry: 1081 first = region_res->child; 1082 if (!first) 1083 return alloc_dev_dax_range(dev_dax, dax_region->res.start, to_alloc); 1084 1085 rc = -ENOSPC; 1086 for (res = first; res; res = res->sibling) { 1087 struct resource *next = res->sibling; 1088 1089 /* space at the beginning of the region */ 1090 if (res == first && res->start > dax_region->res.start) { 1091 alloc = min(res->start - dax_region->res.start, to_alloc); 1092 rc = alloc_dev_dax_range(dev_dax, dax_region->res.start, alloc); 1093 break; 1094 } 1095 1096 alloc = 0; 1097 /* space between allocations */ 1098 if (next && next->start > res->end + 1) 1099 alloc = min(next->start - (res->end + 1), to_alloc); 1100 1101 /* space at the end of the region */ 1102 if (!alloc && !next && res->end < region_res->end) 1103 alloc = min(region_res->end - res->end, to_alloc); 1104 1105 if (!alloc) 1106 continue; 1107 1108 if (adjust_ok(dev_dax, res)) { 1109 rc = adjust_dev_dax_range(dev_dax, res, resource_size(res) + alloc); 1110 break; 1111 } 1112 rc = alloc_dev_dax_range(dev_dax, res->end + 1, alloc); 1113 break; 1114 } 1115 if (rc) 1116 return rc; 1117 to_alloc -= alloc; 1118 if (to_alloc) 1119 goto retry; 1120 return 0; 1121 } 1122 1123 static ssize_t size_store(struct device *dev, struct device_attribute *attr, 1124 const char *buf, size_t len) 1125 { 1126 ssize_t rc; 1127 unsigned long long val; 1128 struct dev_dax *dev_dax = to_dev_dax(dev); 1129 struct dax_region *dax_region = dev_dax->region; 1130 1131 rc = kstrtoull(buf, 0, &val); 1132 if (rc) 1133 return rc; 1134 1135 if (!alloc_is_aligned(dev_dax, val)) { 1136 dev_dbg(dev, "%s: size: %lld misaligned\n", __func__, val); 1137 return -EINVAL; 1138 } 1139 1140 rc = down_write_killable(&dax_region_rwsem); 1141 if (rc) 1142 return rc; 1143 if (!dax_region->dev->driver) { 1144 rc = -ENXIO; 1145 goto err_region; 1146 } 1147 rc = down_write_killable(&dax_dev_rwsem); 1148 if (rc) 1149 goto err_dev; 1150 1151 rc = dev_dax_resize(dax_region, dev_dax, val); 1152 1153 err_dev: 1154 up_write(&dax_dev_rwsem); 1155 err_region: 1156 up_write(&dax_region_rwsem); 1157 1158 if (rc == 0) 1159 return len; 1160 return rc; 1161 } 1162 static DEVICE_ATTR_RW(size); 1163 1164 static ssize_t range_parse(const char *opt, size_t len, struct range *range) 1165 { 1166 unsigned long long addr = 0; 1167 char *start, *end, *str; 1168 ssize_t rc = -EINVAL; 1169 1170 str = kstrdup(opt, GFP_KERNEL); 1171 if (!str) 1172 return rc; 1173 1174 end = str; 1175 start = strsep(&end, "-"); 1176 if (!start || !end) 1177 goto err; 1178 1179 rc = kstrtoull(start, 16, &addr); 1180 if (rc) 1181 goto err; 1182 range->start = addr; 1183 1184 rc = kstrtoull(end, 16, &addr); 1185 if (rc) 1186 goto err; 1187 range->end = addr; 1188 1189 err: 1190 kfree(str); 1191 return rc; 1192 } 1193 1194 static ssize_t mapping_store(struct device *dev, struct device_attribute *attr, 1195 const char *buf, size_t len) 1196 { 1197 struct dev_dax *dev_dax = to_dev_dax(dev); 1198 struct dax_region *dax_region = dev_dax->region; 1199 size_t to_alloc; 1200 struct range r; 1201 ssize_t rc; 1202 1203 rc = range_parse(buf, len, &r); 1204 if (rc) 1205 return rc; 1206 1207 rc = -ENXIO; 1208 rc = down_write_killable(&dax_region_rwsem); 1209 if (rc) 1210 return rc; 1211 if (!dax_region->dev->driver) { 1212 up_write(&dax_region_rwsem); 1213 return rc; 1214 } 1215 rc = down_write_killable(&dax_dev_rwsem); 1216 if (rc) { 1217 up_write(&dax_region_rwsem); 1218 return rc; 1219 } 1220 1221 to_alloc = range_len(&r); 1222 if (alloc_is_aligned(dev_dax, to_alloc)) 1223 rc = alloc_dev_dax_range(dev_dax, r.start, to_alloc); 1224 up_write(&dax_dev_rwsem); 1225 up_write(&dax_region_rwsem); 1226 1227 return rc == 0 ? len : rc; 1228 } 1229 static DEVICE_ATTR_WO(mapping); 1230 1231 static ssize_t align_show(struct device *dev, 1232 struct device_attribute *attr, char *buf) 1233 { 1234 struct dev_dax *dev_dax = to_dev_dax(dev); 1235 1236 return sysfs_emit(buf, "%d\n", dev_dax->align); 1237 } 1238 1239 static ssize_t dev_dax_validate_align(struct dev_dax *dev_dax) 1240 { 1241 struct device *dev = &dev_dax->dev; 1242 int i; 1243 1244 for (i = 0; i < dev_dax->nr_range; i++) { 1245 size_t len = range_len(&dev_dax->ranges[i].range); 1246 1247 if (!alloc_is_aligned(dev_dax, len)) { 1248 dev_dbg(dev, "%s: align %u invalid for range %d\n", 1249 __func__, dev_dax->align, i); 1250 return -EINVAL; 1251 } 1252 } 1253 1254 return 0; 1255 } 1256 1257 static ssize_t align_store(struct device *dev, struct device_attribute *attr, 1258 const char *buf, size_t len) 1259 { 1260 struct dev_dax *dev_dax = to_dev_dax(dev); 1261 struct dax_region *dax_region = dev_dax->region; 1262 unsigned long val, align_save; 1263 ssize_t rc; 1264 1265 rc = kstrtoul(buf, 0, &val); 1266 if (rc) 1267 return -ENXIO; 1268 1269 if (!dax_align_valid(val)) 1270 return -EINVAL; 1271 1272 rc = down_write_killable(&dax_region_rwsem); 1273 if (rc) 1274 return rc; 1275 if (!dax_region->dev->driver) { 1276 up_write(&dax_region_rwsem); 1277 return -ENXIO; 1278 } 1279 1280 rc = down_write_killable(&dax_dev_rwsem); 1281 if (rc) { 1282 up_write(&dax_region_rwsem); 1283 return rc; 1284 } 1285 if (dev->driver) { 1286 rc = -EBUSY; 1287 goto out_unlock; 1288 } 1289 1290 align_save = dev_dax->align; 1291 dev_dax->align = val; 1292 rc = dev_dax_validate_align(dev_dax); 1293 if (rc) 1294 dev_dax->align = align_save; 1295 out_unlock: 1296 up_write(&dax_dev_rwsem); 1297 up_write(&dax_region_rwsem); 1298 return rc == 0 ? len : rc; 1299 } 1300 static DEVICE_ATTR_RW(align); 1301 1302 static int dev_dax_target_node(struct dev_dax *dev_dax) 1303 { 1304 struct dax_region *dax_region = dev_dax->region; 1305 1306 return dax_region->target_node; 1307 } 1308 1309 static ssize_t target_node_show(struct device *dev, 1310 struct device_attribute *attr, char *buf) 1311 { 1312 struct dev_dax *dev_dax = to_dev_dax(dev); 1313 1314 return sysfs_emit(buf, "%d\n", dev_dax_target_node(dev_dax)); 1315 } 1316 static DEVICE_ATTR_RO(target_node); 1317 1318 static ssize_t resource_show(struct device *dev, 1319 struct device_attribute *attr, char *buf) 1320 { 1321 struct dev_dax *dev_dax = to_dev_dax(dev); 1322 struct dax_region *dax_region = dev_dax->region; 1323 unsigned long long start; 1324 1325 if (dev_dax->nr_range < 1) 1326 start = dax_region->res.start; 1327 else 1328 start = dev_dax->ranges[0].range.start; 1329 1330 return sysfs_emit(buf, "%#llx\n", start); 1331 } 1332 static DEVICE_ATTR(resource, 0400, resource_show, NULL); 1333 1334 static ssize_t modalias_show(struct device *dev, struct device_attribute *attr, 1335 char *buf) 1336 { 1337 /* 1338 * We only ever expect to handle device-dax instances, i.e. the 1339 * @type argument to MODULE_ALIAS_DAX_DEVICE() is always zero 1340 */ 1341 return sysfs_emit(buf, DAX_DEVICE_MODALIAS_FMT "\n", 0); 1342 } 1343 static DEVICE_ATTR_RO(modalias); 1344 1345 static ssize_t numa_node_show(struct device *dev, 1346 struct device_attribute *attr, char *buf) 1347 { 1348 return sysfs_emit(buf, "%d\n", dev_to_node(dev)); 1349 } 1350 static DEVICE_ATTR_RO(numa_node); 1351 1352 static ssize_t memmap_on_memory_show(struct device *dev, 1353 struct device_attribute *attr, char *buf) 1354 { 1355 struct dev_dax *dev_dax = to_dev_dax(dev); 1356 1357 return sysfs_emit(buf, "%d\n", dev_dax->memmap_on_memory); 1358 } 1359 1360 static ssize_t memmap_on_memory_store(struct device *dev, 1361 struct device_attribute *attr, 1362 const char *buf, size_t len) 1363 { 1364 struct dev_dax *dev_dax = to_dev_dax(dev); 1365 bool val; 1366 int rc; 1367 1368 rc = kstrtobool(buf, &val); 1369 if (rc) 1370 return rc; 1371 1372 if (val == true && !mhp_supports_memmap_on_memory()) { 1373 dev_dbg(dev, "memmap_on_memory is not available\n"); 1374 return -EOPNOTSUPP; 1375 } 1376 1377 rc = down_write_killable(&dax_dev_rwsem); 1378 if (rc) 1379 return rc; 1380 1381 if (dev_dax->memmap_on_memory != val && dev->driver && 1382 to_dax_drv(dev->driver)->type == DAXDRV_KMEM_TYPE) { 1383 up_write(&dax_dev_rwsem); 1384 return -EBUSY; 1385 } 1386 1387 dev_dax->memmap_on_memory = val; 1388 up_write(&dax_dev_rwsem); 1389 1390 return len; 1391 } 1392 static DEVICE_ATTR_RW(memmap_on_memory); 1393 1394 static umode_t dev_dax_visible(struct kobject *kobj, struct attribute *a, int n) 1395 { 1396 struct device *dev = container_of(kobj, struct device, kobj); 1397 struct dev_dax *dev_dax = to_dev_dax(dev); 1398 struct dax_region *dax_region = dev_dax->region; 1399 1400 if (a == &dev_attr_target_node.attr && dev_dax_target_node(dev_dax) < 0) 1401 return 0; 1402 if (a == &dev_attr_numa_node.attr && !IS_ENABLED(CONFIG_NUMA)) 1403 return 0; 1404 if (a == &dev_attr_mapping.attr && is_static(dax_region)) 1405 return 0; 1406 if ((a == &dev_attr_align.attr || 1407 a == &dev_attr_size.attr) && is_static(dax_region)) 1408 return 0444; 1409 return a->mode; 1410 } 1411 1412 static struct attribute *dev_dax_attributes[] = { 1413 &dev_attr_modalias.attr, 1414 &dev_attr_size.attr, 1415 &dev_attr_mapping.attr, 1416 &dev_attr_target_node.attr, 1417 &dev_attr_align.attr, 1418 &dev_attr_resource.attr, 1419 &dev_attr_numa_node.attr, 1420 &dev_attr_memmap_on_memory.attr, 1421 NULL, 1422 }; 1423 1424 static const struct attribute_group dev_dax_attribute_group = { 1425 .attrs = dev_dax_attributes, 1426 .is_visible = dev_dax_visible, 1427 }; 1428 1429 static const struct attribute_group *dax_attribute_groups[] = { 1430 &dev_dax_attribute_group, 1431 NULL, 1432 }; 1433 1434 static void dev_dax_release(struct device *dev) 1435 { 1436 struct dev_dax *dev_dax = to_dev_dax(dev); 1437 struct dax_device *dax_dev = dev_dax->dax_dev; 1438 1439 put_dax(dax_dev); 1440 free_dev_dax_id(dev_dax); 1441 kfree(dev_dax->pgmap); 1442 kfree(dev_dax); 1443 } 1444 1445 static const struct device_type dev_dax_type = { 1446 .release = dev_dax_release, 1447 .groups = dax_attribute_groups, 1448 }; 1449 1450 static struct dev_dax *__devm_create_dev_dax(struct dev_dax_data *data) 1451 { 1452 struct dax_region *dax_region = data->dax_region; 1453 struct device *parent = dax_region->dev; 1454 struct dax_device *dax_dev; 1455 struct dev_dax *dev_dax; 1456 struct inode *inode; 1457 struct device *dev; 1458 int rc; 1459 1460 dev_dax = kzalloc(sizeof(*dev_dax), GFP_KERNEL); 1461 if (!dev_dax) 1462 return ERR_PTR(-ENOMEM); 1463 1464 dev_dax->region = dax_region; 1465 if (is_static(dax_region)) { 1466 if (dev_WARN_ONCE(parent, data->id < 0, 1467 "dynamic id specified to static region\n")) { 1468 rc = -EINVAL; 1469 goto err_id; 1470 } 1471 1472 dev_dax->id = data->id; 1473 } else { 1474 if (dev_WARN_ONCE(parent, data->id >= 0, 1475 "static id specified to dynamic region\n")) { 1476 rc = -EINVAL; 1477 goto err_id; 1478 } 1479 1480 rc = alloc_dev_dax_id(dev_dax); 1481 if (rc < 0) 1482 goto err_id; 1483 } 1484 1485 dev = &dev_dax->dev; 1486 device_initialize(dev); 1487 dev_set_name(dev, "dax%d.%d", dax_region->id, dev_dax->id); 1488 1489 rc = alloc_dev_dax_range(dev_dax, dax_region->res.start, data->size); 1490 if (rc) 1491 goto err_range; 1492 1493 if (data->pgmap) { 1494 dev_WARN_ONCE(parent, !is_static(dax_region), 1495 "custom dev_pagemap requires a static dax_region\n"); 1496 1497 dev_dax->pgmap = kmemdup(data->pgmap, 1498 sizeof(struct dev_pagemap), GFP_KERNEL); 1499 if (!dev_dax->pgmap) { 1500 rc = -ENOMEM; 1501 goto err_pgmap; 1502 } 1503 } 1504 1505 /* 1506 * No dax_operations since there is no access to this device outside of 1507 * mmap of the resulting character device. 1508 */ 1509 dax_dev = alloc_dax(dev_dax, NULL); 1510 if (IS_ERR(dax_dev)) { 1511 rc = PTR_ERR(dax_dev); 1512 goto err_alloc_dax; 1513 } 1514 set_dax_synchronous(dax_dev); 1515 set_dax_nocache(dax_dev); 1516 set_dax_nomc(dax_dev); 1517 1518 /* a device_dax instance is dead while the driver is not attached */ 1519 kill_dax(dax_dev); 1520 1521 dev_dax->dax_dev = dax_dev; 1522 dev_dax->target_node = dax_region->target_node; 1523 dev_dax->align = dax_region->align; 1524 ida_init(&dev_dax->ida); 1525 1526 dev_dax->memmap_on_memory = data->memmap_on_memory; 1527 1528 inode = dax_inode(dax_dev); 1529 dev->devt = inode->i_rdev; 1530 dev->bus = &dax_bus_type; 1531 dev->parent = parent; 1532 dev->type = &dev_dax_type; 1533 1534 rc = device_add(dev); 1535 if (rc) { 1536 kill_dev_dax(dev_dax); 1537 put_device(dev); 1538 return ERR_PTR(rc); 1539 } 1540 1541 rc = devm_add_action_or_reset(dax_region->dev, unregister_dev_dax, dev); 1542 if (rc) 1543 return ERR_PTR(rc); 1544 1545 /* register mapping device for the initial allocation range */ 1546 if (dev_dax->nr_range && range_len(&dev_dax->ranges[0].range)) { 1547 rc = devm_register_dax_mapping(dev_dax, 0); 1548 if (rc) 1549 return ERR_PTR(rc); 1550 } 1551 1552 return dev_dax; 1553 1554 err_alloc_dax: 1555 kfree(dev_dax->pgmap); 1556 err_pgmap: 1557 free_dev_dax_ranges(dev_dax); 1558 err_range: 1559 free_dev_dax_id(dev_dax); 1560 err_id: 1561 kfree(dev_dax); 1562 1563 return ERR_PTR(rc); 1564 } 1565 1566 struct dev_dax *devm_create_dev_dax(struct dev_dax_data *data) 1567 { 1568 struct dev_dax *dev_dax; 1569 int rc; 1570 1571 rc = down_write_killable(&dax_region_rwsem); 1572 if (rc) 1573 return ERR_PTR(rc); 1574 1575 dev_dax = __devm_create_dev_dax(data); 1576 up_write(&dax_region_rwsem); 1577 1578 return dev_dax; 1579 } 1580 EXPORT_SYMBOL_GPL(devm_create_dev_dax); 1581 1582 int __dax_driver_register(struct dax_device_driver *dax_drv, 1583 struct module *module, const char *mod_name) 1584 { 1585 struct device_driver *drv = &dax_drv->drv; 1586 1587 /* 1588 * dax_bus_probe() calls dax_drv->probe() unconditionally. 1589 * So better be safe than sorry and ensure it is provided. 1590 */ 1591 if (!dax_drv->probe) 1592 return -EINVAL; 1593 1594 INIT_LIST_HEAD(&dax_drv->ids); 1595 drv->owner = module; 1596 drv->name = mod_name; 1597 drv->mod_name = mod_name; 1598 drv->bus = &dax_bus_type; 1599 1600 return driver_register(drv); 1601 } 1602 EXPORT_SYMBOL_GPL(__dax_driver_register); 1603 1604 void dax_driver_unregister(struct dax_device_driver *dax_drv) 1605 { 1606 struct device_driver *drv = &dax_drv->drv; 1607 struct dax_id *dax_id, *_id; 1608 1609 mutex_lock(&dax_bus_lock); 1610 list_for_each_entry_safe(dax_id, _id, &dax_drv->ids, list) { 1611 list_del(&dax_id->list); 1612 kfree(dax_id); 1613 } 1614 mutex_unlock(&dax_bus_lock); 1615 driver_unregister(drv); 1616 } 1617 EXPORT_SYMBOL_GPL(dax_driver_unregister); 1618 1619 int __init dax_bus_init(void) 1620 { 1621 return bus_register(&dax_bus_type); 1622 } 1623 1624 void __exit dax_bus_exit(void) 1625 { 1626 bus_unregister(&dax_bus_type); 1627 } 1628