drivers/dax/device.c

   1 /*
   2  * Copyright(c) 2016 - 2017 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/pfn_t.h>
  17 #include <linux/cdev.h>
  18 #include <linux/slab.h>
  19 #include <linux/dax.h>
  20 #include <linux/fs.h>
  21 #include <linux/mm.h>
  22 #include <linux/mman.h>
  23 #include "dax-private.h"
  24 #include "dax.h"
  25
  26 static struct class *dax_class;
  27
  28 /*
  29  * Rely on the fact that drvdata is set before the attributes are
  30  * registered, and that the attributes are unregistered before drvdata
  31  * is cleared to assume that drvdata is always valid.
  32  */
  33 static ssize_t id_show(struct device *dev,
  34                 struct device_attribute *attr, char *buf)
  35 {
  36         struct dax_region *dax_region = dev_get_drvdata(dev);
  37
  38         return sprintf(buf, "%d\n", dax_region->id);
  39 }
  40 static DEVICE_ATTR_RO(id);
  41
  42 static ssize_t region_size_show(struct device *dev,
  43                 struct device_attribute *attr, char *buf)
  44 {
  45         struct dax_region *dax_region = dev_get_drvdata(dev);
  46
  47         return sprintf(buf, "%llu\n", (unsigned long long)
  48                         resource_size(&dax_region->res));
  49 }
  50 static struct device_attribute dev_attr_region_size = __ATTR(size, 0444,
  51                 region_size_show, NULL);
  52
  53 static ssize_t align_show(struct device *dev,
  54                 struct device_attribute *attr, char *buf)
  55 {
  56         struct dax_region *dax_region = dev_get_drvdata(dev);
  57
  58         return sprintf(buf, "%u\n", dax_region->align);
  59 }
  60 static DEVICE_ATTR_RO(align);
  61
  62 static struct attribute *dax_region_attributes[] = {
  63         &dev_attr_region_size.attr,
  64         &dev_attr_align.attr,
  65         &dev_attr_id.attr,
  66         NULL,
  67 };
  68
  69 static const struct attribute_group dax_region_attribute_group = {
  70         .name = "dax_region",
  71         .attrs = dax_region_attributes,
  72 };
  73
  74 static const struct attribute_group *dax_region_attribute_groups[] = {
  75         &dax_region_attribute_group,
  76         NULL,
  77 };
  78
  79 static void dax_region_free(struct kref *kref)
  80 {
  81         struct dax_region *dax_region;
  82
  83         dax_region = container_of(kref, struct dax_region, kref);
  84         kfree(dax_region);
  85 }
  86
  87 void dax_region_put(struct dax_region *dax_region)
  88 {
  89         kref_put(&dax_region->kref, dax_region_free);
  90 }
  91 EXPORT_SYMBOL_GPL(dax_region_put);
  92
  93 static void dax_region_unregister(void *region)
  94 {
  95         struct dax_region *dax_region = region;
  96
  97         sysfs_remove_groups(&dax_region->dev->kobj,
  98                         dax_region_attribute_groups);
  99         dax_region_put(dax_region);
 100 }
 101
 102 struct dax_region *alloc_dax_region(struct device *parent, int region_id,
 103                 struct resource *res, unsigned int align, void *addr,
 104                 unsigned long pfn_flags)
 105 {
 106         struct dax_region *dax_region;
 107
 108         /*
 109          * The DAX core assumes that it can store its private data in
 110          * parent->driver_data. This WARN is a reminder / safeguard for
 111          * developers of device-dax drivers.
 112          */
 113         if (dev_get_drvdata(parent)) {
 114                 dev_WARN(parent, "dax core failed to setup private data\n");
 115                 return NULL;
 116         }
 117
 118         if (!IS_ALIGNED(res->start, align)
 119                         || !IS_ALIGNED(resource_size(res), align))
 120                 return NULL;
 121
 122         dax_region = kzalloc(sizeof(*dax_region), GFP_KERNEL);
 123         if (!dax_region)
 124                 return NULL;
 125
 126         dev_set_drvdata(parent, dax_region);
 127         memcpy(&dax_region->res, res, sizeof(*res));
 128         dax_region->pfn_flags = pfn_flags;
 129         kref_init(&dax_region->kref);
 130         dax_region->id = region_id;
 131         ida_init(&dax_region->ida);
 132         dax_region->align = align;
 133         dax_region->dev = parent;
 134         dax_region->base = addr;
 135         if (sysfs_create_groups(&parent->kobj, dax_region_attribute_groups)) {
 136                 kfree(dax_region);
 137                 return NULL;
 138         }
 139
 140         kref_get(&dax_region->kref);
 141         if (devm_add_action_or_reset(parent, dax_region_unregister, dax_region))
 142                 return NULL;
 143         return dax_region;
 144 }
 145 EXPORT_SYMBOL_GPL(alloc_dax_region);
 146
 147 static struct dev_dax *to_dev_dax(struct device *dev)
 148 {
 149         return container_of(dev, struct dev_dax, dev);
 150 }
 151
 152 static ssize_t size_show(struct device *dev,
 153                 struct device_attribute *attr, char *buf)
 154 {
 155         struct dev_dax *dev_dax = to_dev_dax(dev);
 156         unsigned long long size = 0;
 157         int i;
 158
 159         for (i = 0; i < dev_dax->num_resources; i++)
 160                 size += resource_size(&dev_dax->res[i]);
 161
 162         return sprintf(buf, "%llu\n", size);
 163 }
 164 static DEVICE_ATTR_RO(size);
 165
 166 static struct attribute *dev_dax_attributes[] = {
 167         &dev_attr_size.attr,
 168         NULL,
 169 };
 170
 171 static const struct attribute_group dev_dax_attribute_group = {
 172         .attrs = dev_dax_attributes,
 173 };
 174
 175 static const struct attribute_group *dax_attribute_groups[] = {
 176         &dev_dax_attribute_group,
 177         NULL,
 178 };
 179
 180 static int check_vma(struct dev_dax *dev_dax, struct vm_area_struct *vma,
 181                 const char *func)
 182 {
 183         struct dax_region *dax_region = dev_dax->region;
 184         struct device *dev = &dev_dax->dev;
 185         unsigned long mask;
 186
 187         if (!dax_alive(dev_dax->dax_dev))
 188                 return -ENXIO;
 189
 190         /* prevent private mappings from being established */
 191         if ((vma->vm_flags & VM_MAYSHARE) != VM_MAYSHARE) {
 192                 dev_info_ratelimited(dev,
 193                                 "%s: %s: fail, attempted private mapping\n",
 194                                 current->comm, func);
 195                 return -EINVAL;
 196         }
 197
 198         mask = dax_region->align - 1;
 199         if (vma->vm_start & mask || vma->vm_end & mask) {
 200                 dev_info_ratelimited(dev,
 201                                 "%s: %s: fail, unaligned vma (%#lx - %#lx, %#lx)\n",
 202                                 current->comm, func, vma->vm_start, vma->vm_end,
 203                                 mask);
 204                 return -EINVAL;
 205         }
 206
 207         if ((dax_region->pfn_flags & (PFN_DEV|PFN_MAP)) == PFN_DEV
 208                         && (vma->vm_flags & VM_DONTCOPY) == 0) {
 209                 dev_info_ratelimited(dev,
 210                                 "%s: %s: fail, dax range requires MADV_DONTFORK\n",
 211                                 current->comm, func);
 212                 return -EINVAL;
 213         }
 214
 215         if (!vma_is_dax(vma)) {
 216                 dev_info_ratelimited(dev,
 217                                 "%s: %s: fail, vma is not DAX capable\n",
 218                                 current->comm, func);
 219                 return -EINVAL;
 220         }
 221
 222         return 0;
 223 }
 224
 225 /* see "strong" declaration in tools/testing/nvdimm/dax-dev.c */
 226 __weak phys_addr_t dax_pgoff_to_phys(struct dev_dax *dev_dax, pgoff_t pgoff,
 227                 unsigned long size)
 228 {
 229         struct resource *res;
 230         /* gcc-4.6.3-nolibc for i386 complains that this is uninitialized */
 231         phys_addr_t uninitialized_var(phys);
 232         int i;
 233
 234         for (i = 0; i < dev_dax->num_resources; i++) {
 235                 res = &dev_dax->res[i];
 236                 phys = pgoff * PAGE_SIZE + res->start;
 237                 if (phys >= res->start && phys <= res->end)
 238                         break;
 239                 pgoff -= PHYS_PFN(resource_size(res));
 240         }
 241
 242         if (i < dev_dax->num_resources) {
 243                 res = &dev_dax->res[i];
 244                 if (phys + size - 1 <= res->end)
 245                         return phys;
 246         }
 247
 248         return -1;
 249 }
 250
 251 static vm_fault_t __dev_dax_pte_fault(struct dev_dax *dev_dax,
 252                                 struct vm_fault *vmf, pfn_t *pfn)
 253 {
 254         struct device *dev = &dev_dax->dev;
 255         struct dax_region *dax_region;
 256         phys_addr_t phys;
 257         unsigned int fault_size = PAGE_SIZE;
 258
 259         if (check_vma(dev_dax, vmf->vma, __func__))
 260                 return VM_FAULT_SIGBUS;
 261
 262         dax_region = dev_dax->region;
 263         if (dax_region->align > PAGE_SIZE) {
 264                 dev_dbg(dev, "alignment (%#x) > fault size (%#x)\n",
 265                         dax_region->align, fault_size);
 266                 return VM_FAULT_SIGBUS;
 267         }
 268
 269         if (fault_size != dax_region->align)
 270                 return VM_FAULT_SIGBUS;
 271
 272         phys = dax_pgoff_to_phys(dev_dax, vmf->pgoff, PAGE_SIZE);
 273         if (phys == -1) {
 274                 dev_dbg(dev, "pgoff_to_phys(%#lx) failed\n", vmf->pgoff);
 275                 return VM_FAULT_SIGBUS;
 276         }
 277
 278         *pfn = phys_to_pfn_t(phys, dax_region->pfn_flags);
 279
 280         return vmf_insert_mixed(vmf->vma, vmf->address, *pfn);
 281 }
 282
 283 static vm_fault_t __dev_dax_pmd_fault(struct dev_dax *dev_dax,
 284                                 struct vm_fault *vmf, pfn_t *pfn)
 285 {
 286         unsigned long pmd_addr = vmf->address & PMD_MASK;
 287         struct device *dev = &dev_dax->dev;
 288         struct dax_region *dax_region;
 289         phys_addr_t phys;
 290         pgoff_t pgoff;
 291         unsigned int fault_size = PMD_SIZE;
 292
 293         if (check_vma(dev_dax, vmf->vma, __func__))
 294                 return VM_FAULT_SIGBUS;
 295
 296         dax_region = dev_dax->region;
 297         if (dax_region->align > PMD_SIZE) {
 298                 dev_dbg(dev, "alignment (%#x) > fault size (%#x)\n",
 299                         dax_region->align, fault_size);
 300                 return VM_FAULT_SIGBUS;
 301         }
 302
 303         /* dax pmd mappings require pfn_t_devmap() */
 304         if ((dax_region->pfn_flags & (PFN_DEV|PFN_MAP)) != (PFN_DEV|PFN_MAP)) {
 305                 dev_dbg(dev, "region lacks devmap flags\n");
 306                 return VM_FAULT_SIGBUS;
 307         }
 308
 309         if (fault_size < dax_region->align)
 310                 return VM_FAULT_SIGBUS;
 311         else if (fault_size > dax_region->align)
 312                 return VM_FAULT_FALLBACK;
 313
 314         /* if we are outside of the VMA */
 315         if (pmd_addr < vmf->vma->vm_start ||
 316                         (pmd_addr + PMD_SIZE) > vmf->vma->vm_end)
 317                 return VM_FAULT_SIGBUS;
 318
 319         pgoff = linear_page_index(vmf->vma, pmd_addr);
 320         phys = dax_pgoff_to_phys(dev_dax, pgoff, PMD_SIZE);
 321         if (phys == -1) {
 322                 dev_dbg(dev, "pgoff_to_phys(%#lx) failed\n", pgoff);
 323                 return VM_FAULT_SIGBUS;
 324         }
 325
 326         *pfn = phys_to_pfn_t(phys, dax_region->pfn_flags);
 327
 328         return vmf_insert_pfn_pmd(vmf, *pfn, vmf->flags & FAULT_FLAG_WRITE);
 329 }
 330
 331 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
 332 static vm_fault_t __dev_dax_pud_fault(struct dev_dax *dev_dax,
 333                                 struct vm_fault *vmf, pfn_t *pfn)
 334 {
 335         unsigned long pud_addr = vmf->address & PUD_MASK;
 336         struct device *dev = &dev_dax->dev;
 337         struct dax_region *dax_region;
 338         phys_addr_t phys;
 339         pgoff_t pgoff;
 340         unsigned int fault_size = PUD_SIZE;
 341
 342
 343         if (check_vma(dev_dax, vmf->vma, __func__))
 344                 return VM_FAULT_SIGBUS;
 345
 346         dax_region = dev_dax->region;
 347         if (dax_region->align > PUD_SIZE) {
 348                 dev_dbg(dev, "alignment (%#x) > fault size (%#x)\n",
 349                         dax_region->align, fault_size);
 350                 return VM_FAULT_SIGBUS;
 351         }
 352
 353         /* dax pud mappings require pfn_t_devmap() */
 354         if ((dax_region->pfn_flags & (PFN_DEV|PFN_MAP)) != (PFN_DEV|PFN_MAP)) {
 355                 dev_dbg(dev, "region lacks devmap flags\n");
 356                 return VM_FAULT_SIGBUS;
 357         }
 358
 359         if (fault_size < dax_region->align)
 360                 return VM_FAULT_SIGBUS;
 361         else if (fault_size > dax_region->align)
 362                 return VM_FAULT_FALLBACK;
 363
 364         /* if we are outside of the VMA */
 365         if (pud_addr < vmf->vma->vm_start ||
 366                         (pud_addr + PUD_SIZE) > vmf->vma->vm_end)
 367                 return VM_FAULT_SIGBUS;
 368
 369         pgoff = linear_page_index(vmf->vma, pud_addr);
 370         phys = dax_pgoff_to_phys(dev_dax, pgoff, PUD_SIZE);
 371         if (phys == -1) {
 372                 dev_dbg(dev, "pgoff_to_phys(%#lx) failed\n", pgoff);
 373                 return VM_FAULT_SIGBUS;
 374         }
 375
 376         *pfn = phys_to_pfn_t(phys, dax_region->pfn_flags);
 377
 378         return vmf_insert_pfn_pud(vmf, *pfn, vmf->flags & FAULT_FLAG_WRITE);
 379 }
 380 #else
 381 static vm_fault_t __dev_dax_pud_fault(struct dev_dax *dev_dax,
 382                                 struct vm_fault *vmf, pfn_t *pfn)
 383 {
 384         return VM_FAULT_FALLBACK;
 385 }
 386 #endif /* !CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
 387
 388 static vm_fault_t dev_dax_huge_fault(struct vm_fault *vmf,
 389                 enum page_entry_size pe_size)
 390 {
 391         struct file *filp = vmf->vma->vm_file;
 392         unsigned long fault_size;
 393         vm_fault_t rc = VM_FAULT_SIGBUS;
 394         int id;
 395         pfn_t pfn;
 396         struct dev_dax *dev_dax = filp->private_data;
 397
 398         dev_dbg(&dev_dax->dev, "%s: %s (%#lx - %#lx) size = %d\n", current->comm,
 399                         (vmf->flags & FAULT_FLAG_WRITE) ? "write" : "read",
 400                         vmf->vma->vm_start, vmf->vma->vm_end, pe_size);
 401
 402         id = dax_read_lock();
 403         switch (pe_size) {
 404         case PE_SIZE_PTE:
 405                 fault_size = PAGE_SIZE;
 406                 rc = __dev_dax_pte_fault(dev_dax, vmf, &pfn);
 407                 break;
 408         case PE_SIZE_PMD:
 409                 fault_size = PMD_SIZE;
 410                 rc = __dev_dax_pmd_fault(dev_dax, vmf, &pfn);
 411                 break;
 412         case PE_SIZE_PUD:
 413                 fault_size = PUD_SIZE;
 414                 rc = __dev_dax_pud_fault(dev_dax, vmf, &pfn);
 415                 break;
 416         default:
 417                 rc = VM_FAULT_SIGBUS;
 418         }
 419
 420         if (rc == VM_FAULT_NOPAGE) {
 421                 unsigned long i;
 422                 pgoff_t pgoff;
 423
 424                 /*
 425                  * In the device-dax case the only possibility for a
 426                  * VM_FAULT_NOPAGE result is when device-dax capacity is
 427                  * mapped. No need to consider the zero page, or racing
 428                  * conflicting mappings.
 429                  */
 430                 pgoff = linear_page_index(vmf->vma, vmf->address
 431                                 & ~(fault_size - 1));
 432                 for (i = 0; i < fault_size / PAGE_SIZE; i++) {
 433                         struct page *page;
 434
 435                         page = pfn_to_page(pfn_t_to_pfn(pfn) + i);
 436                         if (page->mapping)
 437                                 continue;
 438                         page->mapping = filp->f_mapping;
 439                         page->index = pgoff + i;
 440                 }
 441         }
 442         dax_read_unlock(id);
 443
 444         return rc;
 445 }
 446
 447 static vm_fault_t dev_dax_fault(struct vm_fault *vmf)
 448 {
 449         return dev_dax_huge_fault(vmf, PE_SIZE_PTE);
 450 }
 451
 452 static int dev_dax_split(struct vm_area_struct *vma, unsigned long addr)
 453 {
 454         struct file *filp = vma->vm_file;
 455         struct dev_dax *dev_dax = filp->private_data;
 456         struct dax_region *dax_region = dev_dax->region;
 457
 458         if (!IS_ALIGNED(addr, dax_region->align))
 459                 return -EINVAL;
 460         return 0;
 461 }
 462
 463 static unsigned long dev_dax_pagesize(struct vm_area_struct *vma)
 464 {
 465         struct file *filp = vma->vm_file;
 466         struct dev_dax *dev_dax = filp->private_data;
 467         struct dax_region *dax_region = dev_dax->region;
 468
 469         return dax_region->align;
 470 }
 471
 472 static const struct vm_operations_struct dax_vm_ops = {
 473         .fault = dev_dax_fault,
 474         .huge_fault = dev_dax_huge_fault,
 475         .split = dev_dax_split,
 476         .pagesize = dev_dax_pagesize,
 477 };
 478
 479 static int dax_mmap(struct file *filp, struct vm_area_struct *vma)
 480 {
 481         struct dev_dax *dev_dax = filp->private_data;
 482         int rc, id;
 483
 484         dev_dbg(&dev_dax->dev, "trace\n");
 485
 486         /*
 487          * We lock to check dax_dev liveness and will re-check at
 488          * fault time.
 489          */
 490         id = dax_read_lock();
 491         rc = check_vma(dev_dax, vma, __func__);
 492         dax_read_unlock(id);
 493         if (rc)
 494                 return rc;
 495
 496         vma->vm_ops = &dax_vm_ops;
 497         vma->vm_flags |= VM_HUGEPAGE;
 498         return 0;
 499 }
 500
 501 /* return an unmapped area aligned to the dax region specified alignment */
 502 static unsigned long dax_get_unmapped_area(struct file *filp,
 503                 unsigned long addr, unsigned long len, unsigned long pgoff,
 504                 unsigned long flags)
 505 {
 506         unsigned long off, off_end, off_align, len_align, addr_align, align;
 507         struct dev_dax *dev_dax = filp ? filp->private_data : NULL;
 508         struct dax_region *dax_region;
 509
 510         if (!dev_dax || addr)
 511                 goto out;
 512
 513         dax_region = dev_dax->region;
 514         align = dax_region->align;
 515         off = pgoff << PAGE_SHIFT;
 516         off_end = off + len;
 517         off_align = round_up(off, align);
 518
 519         if ((off_end <= off_align) || ((off_end - off_align) < align))
 520                 goto out;
 521
 522         len_align = len + align;
 523         if ((off + len_align) < off)
 524                 goto out;
 525
 526         addr_align = current->mm->get_unmapped_area(filp, addr, len_align,
 527                         pgoff, flags);
 528         if (!IS_ERR_VALUE(addr_align)) {
 529                 addr_align += (off - addr_align) & (align - 1);
 530                 return addr_align;
 531         }
 532  out:
 533         return current->mm->get_unmapped_area(filp, addr, len, pgoff, flags);
 534 }
 535
 536 static const struct address_space_operations dev_dax_aops = {
 537         .set_page_dirty         = noop_set_page_dirty,
 538         .invalidatepage         = noop_invalidatepage,
 539 };
 540
 541 static int dax_open(struct inode *inode, struct file *filp)
 542 {
 543         struct dax_device *dax_dev = inode_dax(inode);
 544         struct inode *__dax_inode = dax_inode(dax_dev);
 545         struct dev_dax *dev_dax = dax_get_private(dax_dev);
 546
 547         dev_dbg(&dev_dax->dev, "trace\n");
 548         inode->i_mapping = __dax_inode->i_mapping;
 549         inode->i_mapping->host = __dax_inode;
 550         inode->i_mapping->a_ops = &dev_dax_aops;
 551         filp->f_mapping = inode->i_mapping;
 552         filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
 553         filp->private_data = dev_dax;
 554         inode->i_flags = S_DAX;
 555
 556         return 0;
 557 }
 558
 559 static int dax_release(struct inode *inode, struct file *filp)
 560 {
 561         struct dev_dax *dev_dax = filp->private_data;
 562
 563         dev_dbg(&dev_dax->dev, "trace\n");
 564         return 0;
 565 }
 566
 567 static const struct file_operations dax_fops = {
 568         .llseek = noop_llseek,
 569         .owner = THIS_MODULE,
 570         .open = dax_open,
 571         .release = dax_release,
 572         .get_unmapped_area = dax_get_unmapped_area,
 573         .mmap = dax_mmap,
 574         .mmap_supported_flags = MAP_SYNC,
 575 };
 576
 577 static void dev_dax_release(struct device *dev)
 578 {
 579         struct dev_dax *dev_dax = to_dev_dax(dev);
 580         struct dax_region *dax_region = dev_dax->region;
 581         struct dax_device *dax_dev = dev_dax->dax_dev;
 582
 583         if (dev_dax->id >= 0)
 584                 ida_simple_remove(&dax_region->ida, dev_dax->id);
 585         dax_region_put(dax_region);
 586         put_dax(dax_dev);
 587         kfree(dev_dax);
 588 }
 589
 590 static void kill_dev_dax(struct dev_dax *dev_dax)
 591 {
 592         struct dax_device *dax_dev = dev_dax->dax_dev;
 593         struct inode *inode = dax_inode(dax_dev);
 594
 595         kill_dax(dax_dev);
 596         unmap_mapping_range(inode->i_mapping, 0, 0, 1);
 597 }
 598
 599 static void unregister_dev_dax(void *dev)
 600 {
 601         struct dev_dax *dev_dax = to_dev_dax(dev);
 602         struct dax_device *dax_dev = dev_dax->dax_dev;
 603         struct inode *inode = dax_inode(dax_dev);
 604         struct cdev *cdev = inode->i_cdev;
 605
 606         dev_dbg(dev, "trace\n");
 607
 608         kill_dev_dax(dev_dax);
 609         cdev_device_del(cdev, dev);
 610         put_device(dev);
 611 }
 612
 613 struct dev_dax *devm_create_dev_dax(struct dax_region *dax_region,
 614                 int id, struct resource *res, int count)
 615 {
 616         struct device *parent = dax_region->dev;
 617         struct dax_device *dax_dev;
 618         struct dev_dax *dev_dax;
 619         struct inode *inode;
 620         struct device *dev;
 621         struct cdev *cdev;
 622         int rc, i;
 623
 624         if (!count)
 625                 return ERR_PTR(-EINVAL);
 626
 627         dev_dax = kzalloc(struct_size(dev_dax, res, count), GFP_KERNEL);
 628         if (!dev_dax)
 629                 return ERR_PTR(-ENOMEM);
 630
 631         for (i = 0; i < count; i++) {
 632                 if (!IS_ALIGNED(res[i].start, dax_region->align)
 633                                 || !IS_ALIGNED(resource_size(&res[i]),
 634                                         dax_region->align)) {
 635                         rc = -EINVAL;
 636                         break;
 637                 }
 638                 dev_dax->res[i].start = res[i].start;
 639                 dev_dax->res[i].end = res[i].end;
 640         }
 641
 642         if (i < count)
 643                 goto err_id;
 644
 645         if (id < 0) {
 646                 id = ida_simple_get(&dax_region->ida, 0, 0, GFP_KERNEL);
 647                 dev_dax->id = id;
 648                 if (id < 0) {
 649                         rc = id;
 650                         goto err_id;
 651                 }
 652         } else {
 653                 /* region provider owns @id lifetime */
 654                 dev_dax->id = -1;
 655         }
 656
 657         /*
 658          * No 'host' or dax_operations since there is no access to this
 659          * device outside of mmap of the resulting character device.
 660          */
 661         dax_dev = alloc_dax(dev_dax, NULL, NULL);
 662         if (!dax_dev) {
 663                 rc = -ENOMEM;
 664                 goto err_dax;
 665         }
 666
 667         /* from here on we're committed to teardown via dax_dev_release() */
 668         dev = &dev_dax->dev;
 669         device_initialize(dev);
 670
 671         inode = dax_inode(dax_dev);
 672         cdev = inode->i_cdev;
 673         cdev_init(cdev, &dax_fops);
 674         cdev->owner = parent->driver->owner;
 675
 676         dev_dax->num_resources = count;
 677         dev_dax->dax_dev = dax_dev;
 678         dev_dax->region = dax_region;
 679         kref_get(&dax_region->kref);
 680
 681         dev->devt = inode->i_rdev;
 682         dev->class = dax_class;
 683         dev->parent = parent;
 684         dev->groups = dax_attribute_groups;
 685         dev->release = dev_dax_release;
 686         dev_set_name(dev, "dax%d.%d", dax_region->id, id);
 687
 688         rc = cdev_device_add(cdev, dev);
 689         if (rc) {
 690                 kill_dev_dax(dev_dax);
 691                 put_device(dev);
 692                 return ERR_PTR(rc);
 693         }
 694
 695         rc = devm_add_action_or_reset(dax_region->dev, unregister_dev_dax, dev);
 696         if (rc)
 697                 return ERR_PTR(rc);
 698
 699         return dev_dax;
 700
 701  err_dax:
 702         if (dev_dax->id >= 0)
 703                 ida_simple_remove(&dax_region->ida, dev_dax->id);
 704  err_id:
 705         kfree(dev_dax);
 706
 707         return ERR_PTR(rc);
 708 }
 709 EXPORT_SYMBOL_GPL(devm_create_dev_dax);
 710
 711 static int __init dax_init(void)
 712 {
 713         dax_class = class_create(THIS_MODULE, "dax");
 714         return PTR_ERR_OR_ZERO(dax_class);
 715 }
 716
 717 static void __exit dax_exit(void)
 718 {
 719         class_destroy(dax_class);
 720 }
 721
 722 MODULE_AUTHOR("Intel Corporation");
 723 MODULE_LICENSE("GPL v2");
 724 subsys_initcall(dax_init);
 725 module_exit(dax_exit);