drivers/dax/dax.c

   1 /*
   2  * Copyright(c) 2016 Intel Corporation. All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of version 2 of the GNU General Public License as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  */
  13 #include <linux/pagemap.h>
  14 #include <linux/module.h>
  15 #include <linux/device.h>
  16 #include <linux/mount.h>
  17 #include <linux/pfn_t.h>
  18 #include <linux/hash.h>
  19 #include <linux/cdev.h>
  20 #include <linux/slab.h>
  21 #include <linux/dax.h>
  22 #include <linux/fs.h>
  23 #include <linux/mm.h>
  24 #include "dax.h"
  25
  26 static dev_t dax_devt;
  27 static struct class *dax_class;
  28 static DEFINE_IDA(dax_minor_ida);
  29 static int nr_dax = CONFIG_NR_DEV_DAX;
  30 module_param(nr_dax, int, S_IRUGO);
  31 static struct vfsmount *dax_mnt;
  32 static struct kmem_cache *dax_cache __read_mostly;
  33 static struct super_block *dax_superblock __read_mostly;
  34 MODULE_PARM_DESC(nr_dax, "max number of device-dax instances");
  35
  36 /**
  37  * struct dax_region - mapping infrastructure for dax devices
  38  * @id: kernel-wide unique region for a memory range
  39  * @base: linear address corresponding to @res
  40  * @kref: to pin while other agents have a need to do lookups
  41  * @dev: parent device backing this region
  42  * @align: allocation and mapping alignment for child dax devices
  43  * @res: physical address range of the region
  44  * @pfn_flags: identify whether the pfns are paged back or not
  45  */
  46 struct dax_region {
  47         int id;
  48         struct ida ida;
  49         void *base;
  50         struct kref kref;
  51         struct device *dev;
  52         unsigned int align;
  53         struct resource res;
  54         unsigned long pfn_flags;
  55 };
  56
  57 /**
  58  * struct dax_dev - subdivision of a dax region
  59  * @region - parent region
  60  * @dev - device backing the character device
  61  * @cdev - core chardev data
  62  * @alive - !alive + rcu grace period == no new mappings can be established
  63  * @id - child id in the region
  64  * @num_resources - number of physical address extents in this device
  65  * @res - array of physical address ranges
  66  */
  67 struct dax_dev {
  68         struct dax_region *region;
  69         struct inode *inode;
  70         struct device dev;
  71         struct cdev cdev;
  72         bool alive;
  73         int id;
  74         int num_resources;
  75         struct resource res[0];
  76 };
  77
  78 static ssize_t id_show(struct device *dev,
  79                 struct device_attribute *attr, char *buf)
  80 {
  81         struct dax_region *dax_region;
  82         ssize_t rc = -ENXIO;
  83
  84         device_lock(dev);
  85         dax_region = dev_get_drvdata(dev);
  86         if (dax_region)
  87                 rc = sprintf(buf, "%d\n", dax_region->id);
  88         device_unlock(dev);
  89
  90         return rc;
  91 }
  92 static DEVICE_ATTR_RO(id);
  93
  94 static ssize_t region_size_show(struct device *dev,
  95                 struct device_attribute *attr, char *buf)
  96 {
  97         struct dax_region *dax_region;
  98         ssize_t rc = -ENXIO;
  99
 100         device_lock(dev);
 101         dax_region = dev_get_drvdata(dev);
 102         if (dax_region)
 103                 rc = sprintf(buf, "%llu\n", (unsigned long long)
 104                                 resource_size(&dax_region->res));
 105         device_unlock(dev);
 106
 107         return rc;
 108 }
 109 static struct device_attribute dev_attr_region_size = __ATTR(size, 0444,
 110                 region_size_show, NULL);
 111
 112 static ssize_t align_show(struct device *dev,
 113                 struct device_attribute *attr, char *buf)
 114 {
 115         struct dax_region *dax_region;
 116         ssize_t rc = -ENXIO;
 117
 118         device_lock(dev);
 119         dax_region = dev_get_drvdata(dev);
 120         if (dax_region)
 121                 rc = sprintf(buf, "%u\n", dax_region->align);
 122         device_unlock(dev);
 123
 124         return rc;
 125 }
 126 static DEVICE_ATTR_RO(align);
 127
 128 static struct attribute *dax_region_attributes[] = {
 129         &dev_attr_region_size.attr,
 130         &dev_attr_align.attr,
 131         &dev_attr_id.attr,
 132         NULL,
 133 };
 134
 135 static const struct attribute_group dax_region_attribute_group = {
 136         .name = "dax_region",
 137         .attrs = dax_region_attributes,
 138 };
 139
 140 static const struct attribute_group *dax_region_attribute_groups[] = {
 141         &dax_region_attribute_group,
 142         NULL,
 143 };
 144
 145 static struct inode *dax_alloc_inode(struct super_block *sb)
 146 {
 147         return kmem_cache_alloc(dax_cache, GFP_KERNEL);
 148 }
 149
 150 static void dax_i_callback(struct rcu_head *head)
 151 {
 152         struct inode *inode = container_of(head, struct inode, i_rcu);
 153
 154         kmem_cache_free(dax_cache, inode);
 155 }
 156
 157 static void dax_destroy_inode(struct inode *inode)
 158 {
 159         call_rcu(&inode->i_rcu, dax_i_callback);
 160 }
 161
 162 static const struct super_operations dax_sops = {
 163         .statfs = simple_statfs,
 164         .alloc_inode = dax_alloc_inode,
 165         .destroy_inode = dax_destroy_inode,
 166         .drop_inode = generic_delete_inode,
 167 };
 168
 169 static struct dentry *dax_mount(struct file_system_type *fs_type,
 170                 int flags, const char *dev_name, void *data)
 171 {
 172         return mount_pseudo(fs_type, "dax:", &dax_sops, NULL, DAXFS_MAGIC);
 173 }
 174
 175 static struct file_system_type dax_type = {
 176         .name = "dax",
 177         .mount = dax_mount,
 178         .kill_sb = kill_anon_super,
 179 };
 180
 181 static int dax_test(struct inode *inode, void *data)
 182 {
 183         return inode->i_cdev == data;
 184 }
 185
 186 static int dax_set(struct inode *inode, void *data)
 187 {
 188         inode->i_cdev = data;
 189         return 0;
 190 }
 191
 192 static struct inode *dax_inode_get(struct cdev *cdev, dev_t devt)
 193 {
 194         struct inode *inode;
 195
 196         inode = iget5_locked(dax_superblock, hash_32(devt + DAXFS_MAGIC, 31),
 197                         dax_test, dax_set, cdev);
 198
 199         if (!inode)
 200                 return NULL;
 201
 202         if (inode->i_state & I_NEW) {
 203                 inode->i_mode = S_IFCHR;
 204                 inode->i_flags = S_DAX;
 205                 inode->i_rdev = devt;
 206                 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
 207                 unlock_new_inode(inode);
 208         }
 209         return inode;
 210 }
 211
 212 static void init_once(void *inode)
 213 {
 214         inode_init_once(inode);
 215 }
 216
 217 static int dax_inode_init(void)
 218 {
 219         int rc;
 220
 221         dax_cache = kmem_cache_create("dax_cache", sizeof(struct inode), 0,
 222                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
 223                          SLAB_MEM_SPREAD|SLAB_ACCOUNT),
 224                         init_once);
 225         if (!dax_cache)
 226                 return -ENOMEM;
 227
 228         rc = register_filesystem(&dax_type);
 229         if (rc)
 230                 goto err_register_fs;
 231
 232         dax_mnt = kern_mount(&dax_type);
 233         if (IS_ERR(dax_mnt)) {
 234                 rc = PTR_ERR(dax_mnt);
 235                 goto err_mount;
 236         }
 237         dax_superblock = dax_mnt->mnt_sb;
 238
 239         return 0;
 240
 241  err_mount:
 242         unregister_filesystem(&dax_type);
 243  err_register_fs:
 244         kmem_cache_destroy(dax_cache);
 245
 246         return rc;
 247 }
 248
 249 static void dax_inode_exit(void)
 250 {
 251         kern_unmount(dax_mnt);
 252         unregister_filesystem(&dax_type);
 253         kmem_cache_destroy(dax_cache);
 254 }
 255
 256 static void dax_region_free(struct kref *kref)
 257 {
 258         struct dax_region *dax_region;
 259
 260         dax_region = container_of(kref, struct dax_region, kref);
 261         kfree(dax_region);
 262 }
 263
 264 void dax_region_put(struct dax_region *dax_region)
 265 {
 266         kref_put(&dax_region->kref, dax_region_free);
 267 }
 268 EXPORT_SYMBOL_GPL(dax_region_put);
 269
 270 static void dax_region_unregister(void *region)
 271 {
 272         struct dax_region *dax_region = region;
 273
 274         sysfs_remove_groups(&dax_region->dev->kobj,
 275                         dax_region_attribute_groups);
 276         dax_region_put(dax_region);
 277 }
 278
 279 struct dax_region *alloc_dax_region(struct device *parent, int region_id,
 280                 struct resource *res, unsigned int align, void *addr,
 281                 unsigned long pfn_flags)
 282 {
 283         struct dax_region *dax_region;
 284
 285         /*
 286          * The DAX core assumes that it can store its private data in
 287          * parent->driver_data. This WARN is a reminder / safeguard for
 288          * developers of device-dax drivers.
 289          */
 290         if (dev_get_drvdata(parent)) {
 291                 dev_WARN(parent, "dax core failed to setup private data\n");
 292                 return NULL;
 293         }
 294
 295         if (!IS_ALIGNED(res->start, align)
 296                         || !IS_ALIGNED(resource_size(res), align))
 297                 return NULL;
 298
 299         dax_region = kzalloc(sizeof(*dax_region), GFP_KERNEL);
 300         if (!dax_region)
 301                 return NULL;
 302
 303         dev_set_drvdata(parent, dax_region);
 304         memcpy(&dax_region->res, res, sizeof(*res));
 305         dax_region->pfn_flags = pfn_flags;
 306         kref_init(&dax_region->kref);
 307         dax_region->id = region_id;
 308         ida_init(&dax_region->ida);
 309         dax_region->align = align;
 310         dax_region->dev = parent;
 311         dax_region->base = addr;
 312         if (sysfs_create_groups(&parent->kobj, dax_region_attribute_groups)) {
 313                 kfree(dax_region);
 314                 return NULL;;
 315         }
 316
 317         kref_get(&dax_region->kref);
 318         if (devm_add_action_or_reset(parent, dax_region_unregister, dax_region))
 319                 return NULL;
 320         return dax_region;
 321 }
 322 EXPORT_SYMBOL_GPL(alloc_dax_region);
 323
 324 static struct dax_dev *to_dax_dev(struct device *dev)
 325 {
 326         return container_of(dev, struct dax_dev, dev);
 327 }
 328
 329 static ssize_t size_show(struct device *dev,
 330                 struct device_attribute *attr, char *buf)
 331 {
 332         struct dax_dev *dax_dev = to_dax_dev(dev);
 333         unsigned long long size = 0;
 334         int i;
 335
 336         for (i = 0; i < dax_dev->num_resources; i++)
 337                 size += resource_size(&dax_dev->res[i]);
 338
 339         return sprintf(buf, "%llu\n", size);
 340 }
 341 static DEVICE_ATTR_RO(size);
 342
 343 static struct attribute *dax_device_attributes[] = {
 344         &dev_attr_size.attr,
 345         NULL,
 346 };
 347
 348 static const struct attribute_group dax_device_attribute_group = {
 349         .attrs = dax_device_attributes,
 350 };
 351
 352 static const struct attribute_group *dax_attribute_groups[] = {
 353         &dax_device_attribute_group,
 354         NULL,
 355 };
 356
 357 static int check_vma(struct dax_dev *dax_dev, struct vm_area_struct *vma,
 358                 const char *func)
 359 {
 360         struct dax_region *dax_region = dax_dev->region;
 361         struct device *dev = &dax_dev->dev;
 362         unsigned long mask;
 363
 364         if (!dax_dev->alive)
 365                 return -ENXIO;
 366
 367         /* prevent private mappings from being established */
 368         if ((vma->vm_flags & VM_MAYSHARE) != VM_MAYSHARE) {
 369                 dev_info(dev, "%s: %s: fail, attempted private mapping\n",
 370                                 current->comm, func);
 371                 return -EINVAL;
 372         }
 373
 374         mask = dax_region->align - 1;
 375         if (vma->vm_start & mask || vma->vm_end & mask) {
 376                 dev_info(dev, "%s: %s: fail, unaligned vma (%#lx - %#lx, %#lx)\n",
 377                                 current->comm, func, vma->vm_start, vma->vm_end,
 378                                 mask);
 379                 return -EINVAL;
 380         }
 381
 382         if ((dax_region->pfn_flags & (PFN_DEV|PFN_MAP)) == PFN_DEV
 383                         && (vma->vm_flags & VM_DONTCOPY) == 0) {
 384                 dev_info(dev, "%s: %s: fail, dax range requires MADV_DONTFORK\n",
 385                                 current->comm, func);
 386                 return -EINVAL;
 387         }
 388
 389         if (!vma_is_dax(vma)) {
 390                 dev_info(dev, "%s: %s: fail, vma is not DAX capable\n",
 391                                 current->comm, func);
 392                 return -EINVAL;
 393         }
 394
 395         return 0;
 396 }
 397
 398 static phys_addr_t pgoff_to_phys(struct dax_dev *dax_dev, pgoff_t pgoff,
 399                 unsigned long size)
 400 {
 401         struct resource *res;
 402         phys_addr_t phys;
 403         int i;
 404
 405         for (i = 0; i < dax_dev->num_resources; i++) {
 406                 res = &dax_dev->res[i];
 407                 phys = pgoff * PAGE_SIZE + res->start;
 408                 if (phys >= res->start && phys <= res->end)
 409                         break;
 410                 pgoff -= PHYS_PFN(resource_size(res));
 411         }
 412
 413         if (i < dax_dev->num_resources) {
 414                 res = &dax_dev->res[i];
 415                 if (phys + size - 1 <= res->end)
 416                         return phys;
 417         }
 418
 419         return -1;
 420 }
 421
 422 static int __dax_dev_fault(struct dax_dev *dax_dev, struct vm_area_struct *vma,
 423                 struct vm_fault *vmf)
 424 {
 425         struct device *dev = &dax_dev->dev;
 426         struct dax_region *dax_region;
 427         int rc = VM_FAULT_SIGBUS;
 428         phys_addr_t phys;
 429         pfn_t pfn;
 430
 431         if (check_vma(dax_dev, vma, __func__))
 432                 return VM_FAULT_SIGBUS;
 433
 434         dax_region = dax_dev->region;
 435         if (dax_region->align > PAGE_SIZE) {
 436                 dev_dbg(dev, "%s: alignment > fault size\n", __func__);
 437                 return VM_FAULT_SIGBUS;
 438         }
 439
 440         phys = pgoff_to_phys(dax_dev, vmf->pgoff, PAGE_SIZE);
 441         if (phys == -1) {
 442                 dev_dbg(dev, "%s: phys_to_pgoff(%#lx) failed\n", __func__,
 443                                 vmf->pgoff);
 444                 return VM_FAULT_SIGBUS;
 445         }
 446
 447         pfn = phys_to_pfn_t(phys, dax_region->pfn_flags);
 448
 449         rc = vm_insert_mixed(vma, vmf->address, pfn);
 450
 451         if (rc == -ENOMEM)
 452                 return VM_FAULT_OOM;
 453         if (rc < 0 && rc != -EBUSY)
 454                 return VM_FAULT_SIGBUS;
 455
 456         return VM_FAULT_NOPAGE;
 457 }
 458
 459 static int dax_dev_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 460 {
 461         int rc;
 462         struct file *filp = vma->vm_file;
 463         struct dax_dev *dax_dev = filp->private_data;
 464
 465         dev_dbg(&dax_dev->dev, "%s: %s: %s (%#lx - %#lx)\n", __func__,
 466                         current->comm, (vmf->flags & FAULT_FLAG_WRITE)
 467                         ? "write" : "read", vma->vm_start, vma->vm_end);
 468         rcu_read_lock();
 469         rc = __dax_dev_fault(dax_dev, vma, vmf);
 470         rcu_read_unlock();
 471
 472         return rc;
 473 }
 474
 475 static int __dax_dev_pmd_fault(struct dax_dev *dax_dev,
 476                 struct vm_area_struct *vma, unsigned long addr, pmd_t *pmd,
 477                 unsigned int flags)
 478 {
 479         unsigned long pmd_addr = addr & PMD_MASK;
 480         struct device *dev = &dax_dev->dev;
 481         struct dax_region *dax_region;
 482         phys_addr_t phys;
 483         pgoff_t pgoff;
 484         pfn_t pfn;
 485
 486         if (check_vma(dax_dev, vma, __func__))
 487                 return VM_FAULT_SIGBUS;
 488
 489         dax_region = dax_dev->region;
 490         if (dax_region->align > PMD_SIZE) {
 491                 dev_dbg(dev, "%s: alignment > fault size\n", __func__);
 492                 return VM_FAULT_SIGBUS;
 493         }
 494
 495         /* dax pmd mappings require pfn_t_devmap() */
 496         if ((dax_region->pfn_flags & (PFN_DEV|PFN_MAP)) != (PFN_DEV|PFN_MAP)) {
 497                 dev_dbg(dev, "%s: alignment > fault size\n", __func__);
 498                 return VM_FAULT_SIGBUS;
 499         }
 500
 501         pgoff = linear_page_index(vma, pmd_addr);
 502         phys = pgoff_to_phys(dax_dev, pgoff, PMD_SIZE);
 503         if (phys == -1) {
 504                 dev_dbg(dev, "%s: phys_to_pgoff(%#lx) failed\n", __func__,
 505                                 pgoff);
 506                 return VM_FAULT_SIGBUS;
 507         }
 508
 509         pfn = phys_to_pfn_t(phys, dax_region->pfn_flags);
 510
 511         return vmf_insert_pfn_pmd(vma, addr, pmd, pfn,
 512                         flags & FAULT_FLAG_WRITE);
 513 }
 514
 515 static int dax_dev_pmd_fault(struct vm_area_struct *vma, unsigned long addr,
 516                 pmd_t *pmd, unsigned int flags)
 517 {
 518         int rc;
 519         struct file *filp = vma->vm_file;
 520         struct dax_dev *dax_dev = filp->private_data;
 521
 522         dev_dbg(&dax_dev->dev, "%s: %s: %s (%#lx - %#lx)\n", __func__,
 523                         current->comm, (flags & FAULT_FLAG_WRITE)
 524                         ? "write" : "read", vma->vm_start, vma->vm_end);
 525
 526         rcu_read_lock();
 527         rc = __dax_dev_pmd_fault(dax_dev, vma, addr, pmd, flags);
 528         rcu_read_unlock();
 529
 530         return rc;
 531 }
 532
 533 static const struct vm_operations_struct dax_dev_vm_ops = {
 534         .fault = dax_dev_fault,
 535         .pmd_fault = dax_dev_pmd_fault,
 536 };
 537
 538 static int dax_mmap(struct file *filp, struct vm_area_struct *vma)
 539 {
 540         struct dax_dev *dax_dev = filp->private_data;
 541         int rc;
 542
 543         dev_dbg(&dax_dev->dev, "%s\n", __func__);
 544
 545         rc = check_vma(dax_dev, vma, __func__);
 546         if (rc)
 547                 return rc;
 548
 549         vma->vm_ops = &dax_dev_vm_ops;
 550         vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
 551         return 0;
 552 }
 553
 554 /* return an unmapped area aligned to the dax region specified alignment */
 555 static unsigned long dax_get_unmapped_area(struct file *filp,
 556                 unsigned long addr, unsigned long len, unsigned long pgoff,
 557                 unsigned long flags)
 558 {
 559         unsigned long off, off_end, off_align, len_align, addr_align, align;
 560         struct dax_dev *dax_dev = filp ? filp->private_data : NULL;
 561         struct dax_region *dax_region;
 562
 563         if (!dax_dev || addr)
 564                 goto out;
 565
 566         dax_region = dax_dev->region;
 567         align = dax_region->align;
 568         off = pgoff << PAGE_SHIFT;
 569         off_end = off + len;
 570         off_align = round_up(off, align);
 571
 572         if ((off_end <= off_align) || ((off_end - off_align) < align))
 573                 goto out;
 574
 575         len_align = len + align;
 576         if ((off + len_align) < off)
 577                 goto out;
 578
 579         addr_align = current->mm->get_unmapped_area(filp, addr, len_align,
 580                         pgoff, flags);
 581         if (!IS_ERR_VALUE(addr_align)) {
 582                 addr_align += (off - addr_align) & (align - 1);
 583                 return addr_align;
 584         }
 585  out:
 586         return current->mm->get_unmapped_area(filp, addr, len, pgoff, flags);
 587 }
 588
 589 static int dax_open(struct inode *inode, struct file *filp)
 590 {
 591         struct dax_dev *dax_dev;
 592
 593         dax_dev = container_of(inode->i_cdev, struct dax_dev, cdev);
 594         dev_dbg(&dax_dev->dev, "%s\n", __func__);
 595         inode->i_mapping = dax_dev->inode->i_mapping;
 596         inode->i_mapping->host = dax_dev->inode;
 597         filp->f_mapping = inode->i_mapping;
 598         filp->private_data = dax_dev;
 599         inode->i_flags = S_DAX;
 600
 601         return 0;
 602 }
 603
 604 static int dax_release(struct inode *inode, struct file *filp)
 605 {
 606         struct dax_dev *dax_dev = filp->private_data;
 607
 608         dev_dbg(&dax_dev->dev, "%s\n", __func__);
 609         return 0;
 610 }
 611
 612 static const struct file_operations dax_fops = {
 613         .llseek = noop_llseek,
 614         .owner = THIS_MODULE,
 615         .open = dax_open,
 616         .release = dax_release,
 617         .get_unmapped_area = dax_get_unmapped_area,
 618         .mmap = dax_mmap,
 619 };
 620
 621 static void dax_dev_release(struct device *dev)
 622 {
 623         struct dax_dev *dax_dev = to_dax_dev(dev);
 624         struct dax_region *dax_region = dax_dev->region;
 625
 626         ida_simple_remove(&dax_region->ida, dax_dev->id);
 627         ida_simple_remove(&dax_minor_ida, MINOR(dev->devt));
 628         dax_region_put(dax_region);
 629         iput(dax_dev->inode);
 630         kfree(dax_dev);
 631 }
 632
 633 static void unregister_dax_dev(void *dev)
 634 {
 635         struct dax_dev *dax_dev = to_dax_dev(dev);
 636         struct cdev *cdev = &dax_dev->cdev;
 637
 638         dev_dbg(dev, "%s\n", __func__);
 639
 640         /*
 641          * Note, rcu is not protecting the liveness of dax_dev, rcu is
 642          * ensuring that any fault handlers that might have seen
 643          * dax_dev->alive == true, have completed.  Any fault handlers
 644          * that start after synchronize_rcu() has started will abort
 645          * upon seeing dax_dev->alive == false.
 646          */
 647         dax_dev->alive = false;
 648         synchronize_rcu();
 649         unmap_mapping_range(dax_dev->inode->i_mapping, 0, 0, 1);
 650         cdev_del(cdev);
 651         device_unregister(dev);
 652 }
 653
 654 struct dax_dev *devm_create_dax_dev(struct dax_region *dax_region,
 655                 struct resource *res, int count)
 656 {
 657         struct device *parent = dax_region->dev;
 658         struct dax_dev *dax_dev;
 659         int rc = 0, minor, i;
 660         struct device *dev;
 661         struct cdev *cdev;
 662         dev_t dev_t;
 663
 664         dax_dev = kzalloc(sizeof(*dax_dev) + sizeof(*res) * count, GFP_KERNEL);
 665         if (!dax_dev)
 666                 return ERR_PTR(-ENOMEM);
 667
 668         for (i = 0; i < count; i++) {
 669                 if (!IS_ALIGNED(res[i].start, dax_region->align)
 670                                 || !IS_ALIGNED(resource_size(&res[i]),
 671                                         dax_region->align)) {
 672                         rc = -EINVAL;
 673                         break;
 674                 }
 675                 dax_dev->res[i].start = res[i].start;
 676                 dax_dev->res[i].end = res[i].end;
 677         }
 678
 679         if (i < count)
 680                 goto err_id;
 681
 682         dax_dev->id = ida_simple_get(&dax_region->ida, 0, 0, GFP_KERNEL);
 683         if (dax_dev->id < 0) {
 684                 rc = dax_dev->id;
 685                 goto err_id;
 686         }
 687
 688         minor = ida_simple_get(&dax_minor_ida, 0, 0, GFP_KERNEL);
 689         if (minor < 0) {
 690                 rc = minor;
 691                 goto err_minor;
 692         }
 693
 694         dev_t = MKDEV(MAJOR(dax_devt), minor);
 695         dev = &dax_dev->dev;
 696         dax_dev->inode = dax_inode_get(&dax_dev->cdev, dev_t);
 697         if (!dax_dev->inode) {
 698                 rc = -ENOMEM;
 699                 goto err_inode;
 700         }
 701
 702         /* device_initialize() so cdev can reference kobj parent */
 703         device_initialize(dev);
 704
 705         cdev = &dax_dev->cdev;
 706         cdev_init(cdev, &dax_fops);
 707         cdev->owner = parent->driver->owner;
 708         cdev->kobj.parent = &dev->kobj;
 709         rc = cdev_add(&dax_dev->cdev, dev_t, 1);
 710         if (rc)
 711                 goto err_cdev;
 712
 713         /* from here on we're committed to teardown via dax_dev_release() */
 714         dax_dev->num_resources = count;
 715         dax_dev->alive = true;
 716         dax_dev->region = dax_region;
 717         kref_get(&dax_region->kref);
 718
 719         dev->devt = dev_t;
 720         dev->class = dax_class;
 721         dev->parent = parent;
 722         dev->groups = dax_attribute_groups;
 723         dev->release = dax_dev_release;
 724         dev_set_name(dev, "dax%d.%d", dax_region->id, dax_dev->id);
 725         rc = device_add(dev);
 726         if (rc) {
 727                 put_device(dev);
 728                 return ERR_PTR(rc);
 729         }
 730
 731         rc = devm_add_action_or_reset(dax_region->dev, unregister_dax_dev, dev);
 732         if (rc)
 733                 return ERR_PTR(rc);
 734
 735         return dax_dev;
 736
 737  err_cdev:
 738         iput(dax_dev->inode);
 739  err_inode:
 740         ida_simple_remove(&dax_minor_ida, minor);
 741  err_minor:
 742         ida_simple_remove(&dax_region->ida, dax_dev->id);
 743  err_id:
 744         kfree(dax_dev);
 745
 746         return ERR_PTR(rc);
 747 }
 748 EXPORT_SYMBOL_GPL(devm_create_dax_dev);
 749
 750 static int __init dax_init(void)
 751 {
 752         int rc;
 753
 754         rc = dax_inode_init();
 755         if (rc)
 756                 return rc;
 757
 758         nr_dax = max(nr_dax, 256);
 759         rc = alloc_chrdev_region(&dax_devt, 0, nr_dax, "dax");
 760         if (rc)
 761                 goto err_chrdev;
 762
 763         dax_class = class_create(THIS_MODULE, "dax");
 764         if (IS_ERR(dax_class)) {
 765                 rc = PTR_ERR(dax_class);
 766                 goto err_class;
 767         }
 768
 769         return 0;
 770
 771  err_class:
 772         unregister_chrdev_region(dax_devt, nr_dax);
 773  err_chrdev:
 774         dax_inode_exit();
 775         return rc;
 776 }
 777
 778 static void __exit dax_exit(void)
 779 {
 780         class_destroy(dax_class);
 781         unregister_chrdev_region(dax_devt, nr_dax);
 782         ida_destroy(&dax_minor_ida);
 783         dax_inode_exit();
 784 }
 785
 786 MODULE_AUTHOR("Intel Corporation");
 787 MODULE_LICENSE("GPL v2");
 788 subsys_initcall(dax_init);
 789 module_exit(dax_exit);