blob: 0e499bfca41ccda4752463f770dd495259fb255e [file] [log] [blame]
Dan Williamsab68f262016-05-18 09:15:08 -07001/*
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>
Dan Williams3bc52c42016-07-24 21:55:45 -070016#include <linux/mount.h>
Dan Williamsab68f262016-05-18 09:15:08 -070017#include <linux/pfn_t.h>
Dan Williams3bc52c42016-07-24 21:55:45 -070018#include <linux/hash.h>
Dan Williamsba09c012016-07-24 15:55:42 -070019#include <linux/cdev.h>
Dan Williamsab68f262016-05-18 09:15:08 -070020#include <linux/slab.h>
21#include <linux/dax.h>
22#include <linux/fs.h>
23#include <linux/mm.h>
Dan Williamsccdb07f2016-08-06 16:05:06 -070024#include "dax.h"
Dan Williamsab68f262016-05-18 09:15:08 -070025
Dan Williamsba09c012016-07-24 15:55:42 -070026static dev_t dax_devt;
Dan Williamsab68f262016-05-18 09:15:08 -070027static struct class *dax_class;
28static DEFINE_IDA(dax_minor_ida);
Dan Williamsba09c012016-07-24 15:55:42 -070029static int nr_dax = CONFIG_NR_DEV_DAX;
30module_param(nr_dax, int, S_IRUGO);
Dan Williams3bc52c42016-07-24 21:55:45 -070031static struct vfsmount *dax_mnt;
32static struct kmem_cache *dax_cache __read_mostly;
33static struct super_block *dax_superblock __read_mostly;
Dan Williamsba09c012016-07-24 15:55:42 -070034MODULE_PARM_DESC(nr_dax, "max number of device-dax instances");
Dan Williamsab68f262016-05-18 09:15:08 -070035
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 */
46struct 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
Dan Williamsba09c012016-07-24 15:55:42 -070061 * @cdev - core chardev data
Dan Williamsdee41072016-05-14 12:20:44 -070062 * @alive - !alive + rcu grace period == no new mappings can be established
Dan Williamsab68f262016-05-18 09:15:08 -070063 * @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 */
67struct dax_dev {
68 struct dax_region *region;
Dan Williams3bc52c42016-07-24 21:55:45 -070069 struct inode *inode;
Dan Williamsebd84d72016-08-11 00:41:51 -070070 struct device dev;
Dan Williamsba09c012016-07-24 15:55:42 -070071 struct cdev cdev;
Dan Williamsdee41072016-05-14 12:20:44 -070072 bool alive;
Dan Williamsab68f262016-05-18 09:15:08 -070073 int id;
74 int num_resources;
75 struct resource res[0];
76};
77
Dan Williams3bc52c42016-07-24 21:55:45 -070078static struct inode *dax_alloc_inode(struct super_block *sb)
79{
80 return kmem_cache_alloc(dax_cache, GFP_KERNEL);
81}
82
83static void dax_i_callback(struct rcu_head *head)
84{
85 struct inode *inode = container_of(head, struct inode, i_rcu);
86
87 kmem_cache_free(dax_cache, inode);
88}
89
90static void dax_destroy_inode(struct inode *inode)
91{
92 call_rcu(&inode->i_rcu, dax_i_callback);
93}
94
95static const struct super_operations dax_sops = {
96 .statfs = simple_statfs,
97 .alloc_inode = dax_alloc_inode,
98 .destroy_inode = dax_destroy_inode,
99 .drop_inode = generic_delete_inode,
100};
101
102static struct dentry *dax_mount(struct file_system_type *fs_type,
103 int flags, const char *dev_name, void *data)
104{
105 return mount_pseudo(fs_type, "dax:", &dax_sops, NULL, DAXFS_MAGIC);
106}
107
108static struct file_system_type dax_type = {
109 .name = "dax",
110 .mount = dax_mount,
111 .kill_sb = kill_anon_super,
112};
113
114static int dax_test(struct inode *inode, void *data)
115{
116 return inode->i_cdev == data;
117}
118
119static int dax_set(struct inode *inode, void *data)
120{
121 inode->i_cdev = data;
122 return 0;
123}
124
125static struct inode *dax_inode_get(struct cdev *cdev, dev_t devt)
126{
127 struct inode *inode;
128
129 inode = iget5_locked(dax_superblock, hash_32(devt + DAXFS_MAGIC, 31),
130 dax_test, dax_set, cdev);
131
132 if (!inode)
133 return NULL;
134
135 if (inode->i_state & I_NEW) {
136 inode->i_mode = S_IFCHR;
137 inode->i_flags = S_DAX;
138 inode->i_rdev = devt;
139 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
140 unlock_new_inode(inode);
141 }
142 return inode;
143}
144
145static void init_once(void *inode)
146{
147 inode_init_once(inode);
148}
149
150static int dax_inode_init(void)
151{
152 int rc;
153
154 dax_cache = kmem_cache_create("dax_cache", sizeof(struct inode), 0,
155 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
156 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
157 init_once);
158 if (!dax_cache)
159 return -ENOMEM;
160
161 rc = register_filesystem(&dax_type);
162 if (rc)
163 goto err_register_fs;
164
165 dax_mnt = kern_mount(&dax_type);
166 if (IS_ERR(dax_mnt)) {
167 rc = PTR_ERR(dax_mnt);
168 goto err_mount;
169 }
170 dax_superblock = dax_mnt->mnt_sb;
171
172 return 0;
173
174 err_mount:
175 unregister_filesystem(&dax_type);
176 err_register_fs:
177 kmem_cache_destroy(dax_cache);
178
179 return rc;
180}
181
182static void dax_inode_exit(void)
183{
184 kern_unmount(dax_mnt);
185 unregister_filesystem(&dax_type);
186 kmem_cache_destroy(dax_cache);
187}
188
Dan Williamsab68f262016-05-18 09:15:08 -0700189static void dax_region_free(struct kref *kref)
190{
191 struct dax_region *dax_region;
192
193 dax_region = container_of(kref, struct dax_region, kref);
194 kfree(dax_region);
195}
196
197void dax_region_put(struct dax_region *dax_region)
198{
199 kref_put(&dax_region->kref, dax_region_free);
200}
201EXPORT_SYMBOL_GPL(dax_region_put);
202
Dan Williamsab68f262016-05-18 09:15:08 -0700203struct dax_region *alloc_dax_region(struct device *parent, int region_id,
204 struct resource *res, unsigned int align, void *addr,
205 unsigned long pfn_flags)
206{
207 struct dax_region *dax_region;
208
Dan Williams9d2d01a2016-07-19 16:17:58 -0700209 if (!IS_ALIGNED(res->start, align)
210 || !IS_ALIGNED(resource_size(res), align))
211 return NULL;
Dan Williamsab68f262016-05-18 09:15:08 -0700212
Dan Williams9d2d01a2016-07-19 16:17:58 -0700213 dax_region = kzalloc(sizeof(*dax_region), GFP_KERNEL);
Dan Williamsab68f262016-05-18 09:15:08 -0700214 if (!dax_region)
215 return NULL;
216
217 memcpy(&dax_region->res, res, sizeof(*res));
218 dax_region->pfn_flags = pfn_flags;
219 kref_init(&dax_region->kref);
220 dax_region->id = region_id;
221 ida_init(&dax_region->ida);
222 dax_region->align = align;
223 dax_region->dev = parent;
224 dax_region->base = addr;
225
226 return dax_region;
227}
228EXPORT_SYMBOL_GPL(alloc_dax_region);
229
Dan Williamsebd84d72016-08-11 00:41:51 -0700230static struct dax_dev *to_dax_dev(struct device *dev)
231{
232 return container_of(dev, struct dax_dev, dev);
233}
234
Dan Williamsab68f262016-05-18 09:15:08 -0700235static ssize_t size_show(struct device *dev,
236 struct device_attribute *attr, char *buf)
237{
Dan Williamsebd84d72016-08-11 00:41:51 -0700238 struct dax_dev *dax_dev = to_dax_dev(dev);
Dan Williamsab68f262016-05-18 09:15:08 -0700239 unsigned long long size = 0;
240 int i;
241
242 for (i = 0; i < dax_dev->num_resources; i++)
243 size += resource_size(&dax_dev->res[i]);
244
245 return sprintf(buf, "%llu\n", size);
246}
247static DEVICE_ATTR_RO(size);
248
249static struct attribute *dax_device_attributes[] = {
250 &dev_attr_size.attr,
251 NULL,
252};
253
254static const struct attribute_group dax_device_attribute_group = {
255 .attrs = dax_device_attributes,
256};
257
258static const struct attribute_group *dax_attribute_groups[] = {
259 &dax_device_attribute_group,
260 NULL,
261};
262
Dan Williamsdee41072016-05-14 12:20:44 -0700263static int check_vma(struct dax_dev *dax_dev, struct vm_area_struct *vma,
264 const char *func)
265{
266 struct dax_region *dax_region = dax_dev->region;
Dan Williamsebd84d72016-08-11 00:41:51 -0700267 struct device *dev = &dax_dev->dev;
Dan Williamsdee41072016-05-14 12:20:44 -0700268 unsigned long mask;
269
270 if (!dax_dev->alive)
271 return -ENXIO;
272
273 /* prevent private / writable mappings from being established */
274 if ((vma->vm_flags & (VM_NORESERVE|VM_SHARED|VM_WRITE)) == VM_WRITE) {
275 dev_info(dev, "%s: %s: fail, attempted private mapping\n",
276 current->comm, func);
277 return -EINVAL;
278 }
279
280 mask = dax_region->align - 1;
281 if (vma->vm_start & mask || vma->vm_end & mask) {
282 dev_info(dev, "%s: %s: fail, unaligned vma (%#lx - %#lx, %#lx)\n",
283 current->comm, func, vma->vm_start, vma->vm_end,
284 mask);
285 return -EINVAL;
286 }
287
288 if ((dax_region->pfn_flags & (PFN_DEV|PFN_MAP)) == PFN_DEV
289 && (vma->vm_flags & VM_DONTCOPY) == 0) {
290 dev_info(dev, "%s: %s: fail, dax range requires MADV_DONTFORK\n",
291 current->comm, func);
292 return -EINVAL;
293 }
294
295 if (!vma_is_dax(vma)) {
296 dev_info(dev, "%s: %s: fail, vma is not DAX capable\n",
297 current->comm, func);
298 return -EINVAL;
299 }
300
301 return 0;
302}
303
304static phys_addr_t pgoff_to_phys(struct dax_dev *dax_dev, pgoff_t pgoff,
305 unsigned long size)
306{
307 struct resource *res;
308 phys_addr_t phys;
309 int i;
310
311 for (i = 0; i < dax_dev->num_resources; i++) {
312 res = &dax_dev->res[i];
313 phys = pgoff * PAGE_SIZE + res->start;
314 if (phys >= res->start && phys <= res->end)
315 break;
316 pgoff -= PHYS_PFN(resource_size(res));
317 }
318
319 if (i < dax_dev->num_resources) {
320 res = &dax_dev->res[i];
321 if (phys + size - 1 <= res->end)
322 return phys;
323 }
324
325 return -1;
326}
327
328static int __dax_dev_fault(struct dax_dev *dax_dev, struct vm_area_struct *vma,
329 struct vm_fault *vmf)
330{
331 unsigned long vaddr = (unsigned long) vmf->virtual_address;
Dan Williamsebd84d72016-08-11 00:41:51 -0700332 struct device *dev = &dax_dev->dev;
Dan Williamsdee41072016-05-14 12:20:44 -0700333 struct dax_region *dax_region;
334 int rc = VM_FAULT_SIGBUS;
335 phys_addr_t phys;
336 pfn_t pfn;
337
338 if (check_vma(dax_dev, vma, __func__))
339 return VM_FAULT_SIGBUS;
340
341 dax_region = dax_dev->region;
342 if (dax_region->align > PAGE_SIZE) {
343 dev_dbg(dev, "%s: alignment > fault size\n", __func__);
344 return VM_FAULT_SIGBUS;
345 }
346
347 phys = pgoff_to_phys(dax_dev, vmf->pgoff, PAGE_SIZE);
348 if (phys == -1) {
349 dev_dbg(dev, "%s: phys_to_pgoff(%#lx) failed\n", __func__,
350 vmf->pgoff);
351 return VM_FAULT_SIGBUS;
352 }
353
354 pfn = phys_to_pfn_t(phys, dax_region->pfn_flags);
355
356 rc = vm_insert_mixed(vma, vaddr, pfn);
357
358 if (rc == -ENOMEM)
359 return VM_FAULT_OOM;
360 if (rc < 0 && rc != -EBUSY)
361 return VM_FAULT_SIGBUS;
362
363 return VM_FAULT_NOPAGE;
364}
365
366static int dax_dev_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
367{
368 int rc;
369 struct file *filp = vma->vm_file;
370 struct dax_dev *dax_dev = filp->private_data;
371
Dan Williamsebd84d72016-08-11 00:41:51 -0700372 dev_dbg(&dax_dev->dev, "%s: %s: %s (%#lx - %#lx)\n", __func__,
Dan Williamsdee41072016-05-14 12:20:44 -0700373 current->comm, (vmf->flags & FAULT_FLAG_WRITE)
374 ? "write" : "read", vma->vm_start, vma->vm_end);
375 rcu_read_lock();
376 rc = __dax_dev_fault(dax_dev, vma, vmf);
377 rcu_read_unlock();
378
379 return rc;
380}
381
382static int __dax_dev_pmd_fault(struct dax_dev *dax_dev,
383 struct vm_area_struct *vma, unsigned long addr, pmd_t *pmd,
384 unsigned int flags)
385{
386 unsigned long pmd_addr = addr & PMD_MASK;
Dan Williamsebd84d72016-08-11 00:41:51 -0700387 struct device *dev = &dax_dev->dev;
Dan Williamsdee41072016-05-14 12:20:44 -0700388 struct dax_region *dax_region;
389 phys_addr_t phys;
390 pgoff_t pgoff;
391 pfn_t pfn;
392
393 if (check_vma(dax_dev, vma, __func__))
394 return VM_FAULT_SIGBUS;
395
396 dax_region = dax_dev->region;
397 if (dax_region->align > PMD_SIZE) {
398 dev_dbg(dev, "%s: alignment > fault size\n", __func__);
399 return VM_FAULT_SIGBUS;
400 }
401
402 /* dax pmd mappings require pfn_t_devmap() */
403 if ((dax_region->pfn_flags & (PFN_DEV|PFN_MAP)) != (PFN_DEV|PFN_MAP)) {
404 dev_dbg(dev, "%s: alignment > fault size\n", __func__);
405 return VM_FAULT_SIGBUS;
406 }
407
408 pgoff = linear_page_index(vma, pmd_addr);
Dan Williams4c3cb6e2016-09-03 10:36:00 -0700409 phys = pgoff_to_phys(dax_dev, pgoff, PMD_SIZE);
Dan Williamsdee41072016-05-14 12:20:44 -0700410 if (phys == -1) {
411 dev_dbg(dev, "%s: phys_to_pgoff(%#lx) failed\n", __func__,
412 pgoff);
413 return VM_FAULT_SIGBUS;
414 }
415
416 pfn = phys_to_pfn_t(phys, dax_region->pfn_flags);
417
418 return vmf_insert_pfn_pmd(vma, addr, pmd, pfn,
419 flags & FAULT_FLAG_WRITE);
420}
421
422static int dax_dev_pmd_fault(struct vm_area_struct *vma, unsigned long addr,
423 pmd_t *pmd, unsigned int flags)
424{
425 int rc;
426 struct file *filp = vma->vm_file;
427 struct dax_dev *dax_dev = filp->private_data;
428
Dan Williamsebd84d72016-08-11 00:41:51 -0700429 dev_dbg(&dax_dev->dev, "%s: %s: %s (%#lx - %#lx)\n", __func__,
Dan Williamsdee41072016-05-14 12:20:44 -0700430 current->comm, (flags & FAULT_FLAG_WRITE)
431 ? "write" : "read", vma->vm_start, vma->vm_end);
432
433 rcu_read_lock();
434 rc = __dax_dev_pmd_fault(dax_dev, vma, addr, pmd, flags);
435 rcu_read_unlock();
436
437 return rc;
438}
439
Dan Williamsdee41072016-05-14 12:20:44 -0700440static const struct vm_operations_struct dax_dev_vm_ops = {
441 .fault = dax_dev_fault,
442 .pmd_fault = dax_dev_pmd_fault,
Dan Williamsdee41072016-05-14 12:20:44 -0700443};
444
Dan Williamsaf69f512016-08-11 00:38:03 -0700445static int dax_mmap(struct file *filp, struct vm_area_struct *vma)
Dan Williamsdee41072016-05-14 12:20:44 -0700446{
447 struct dax_dev *dax_dev = filp->private_data;
448 int rc;
449
Dan Williamsebd84d72016-08-11 00:41:51 -0700450 dev_dbg(&dax_dev->dev, "%s\n", __func__);
Dan Williamsdee41072016-05-14 12:20:44 -0700451
452 rc = check_vma(dax_dev, vma, __func__);
453 if (rc)
454 return rc;
455
Dan Williamsdee41072016-05-14 12:20:44 -0700456 vma->vm_ops = &dax_dev_vm_ops;
457 vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
458 return 0;
Dan Williams043a9252016-08-07 08:23:56 -0700459}
Dan Williamsdee41072016-05-14 12:20:44 -0700460
Dan Williams043a9252016-08-07 08:23:56 -0700461/* return an unmapped area aligned to the dax region specified alignment */
Dan Williamsaf69f512016-08-11 00:38:03 -0700462static unsigned long dax_get_unmapped_area(struct file *filp,
Dan Williams043a9252016-08-07 08:23:56 -0700463 unsigned long addr, unsigned long len, unsigned long pgoff,
464 unsigned long flags)
465{
466 unsigned long off, off_end, off_align, len_align, addr_align, align;
467 struct dax_dev *dax_dev = filp ? filp->private_data : NULL;
468 struct dax_region *dax_region;
469
470 if (!dax_dev || addr)
471 goto out;
472
473 dax_region = dax_dev->region;
474 align = dax_region->align;
475 off = pgoff << PAGE_SHIFT;
476 off_end = off + len;
477 off_align = round_up(off, align);
478
479 if ((off_end <= off_align) || ((off_end - off_align) < align))
480 goto out;
481
482 len_align = len + align;
483 if ((off + len_align) < off)
484 goto out;
485
486 addr_align = current->mm->get_unmapped_area(filp, addr, len_align,
487 pgoff, flags);
488 if (!IS_ERR_VALUE(addr_align)) {
489 addr_align += (off - addr_align) & (align - 1);
490 return addr_align;
491 }
492 out:
493 return current->mm->get_unmapped_area(filp, addr, len, pgoff, flags);
494}
495
Dan Williamsaf69f512016-08-11 00:38:03 -0700496static int dax_open(struct inode *inode, struct file *filp)
Dan Williams043a9252016-08-07 08:23:56 -0700497{
Dan Williamsba09c012016-07-24 15:55:42 -0700498 struct dax_dev *dax_dev;
Dan Williams043a9252016-08-07 08:23:56 -0700499
Dan Williamsba09c012016-07-24 15:55:42 -0700500 dax_dev = container_of(inode->i_cdev, struct dax_dev, cdev);
501 dev_dbg(&dax_dev->dev, "%s\n", __func__);
Dan Williams3bc52c42016-07-24 21:55:45 -0700502 inode->i_mapping = dax_dev->inode->i_mapping;
503 inode->i_mapping->host = dax_dev->inode;
504 filp->f_mapping = inode->i_mapping;
Dan Williamsebd84d72016-08-11 00:41:51 -0700505 filp->private_data = dax_dev;
506 inode->i_flags = S_DAX;
Dan Williams043a9252016-08-07 08:23:56 -0700507
Dan Williams043a9252016-08-07 08:23:56 -0700508 return 0;
509}
510
Dan Williamsaf69f512016-08-11 00:38:03 -0700511static int dax_release(struct inode *inode, struct file *filp)
Dan Williams043a9252016-08-07 08:23:56 -0700512{
513 struct dax_dev *dax_dev = filp->private_data;
Dan Williams043a9252016-08-07 08:23:56 -0700514
Dan Williamsba09c012016-07-24 15:55:42 -0700515 dev_dbg(&dax_dev->dev, "%s\n", __func__);
Dan Williams043a9252016-08-07 08:23:56 -0700516 return 0;
Dan Williamsdee41072016-05-14 12:20:44 -0700517}
518
Dan Williamsab68f262016-05-18 09:15:08 -0700519static const struct file_operations dax_fops = {
520 .llseek = noop_llseek,
521 .owner = THIS_MODULE,
Dan Williamsaf69f512016-08-11 00:38:03 -0700522 .open = dax_open,
523 .release = dax_release,
524 .get_unmapped_area = dax_get_unmapped_area,
525 .mmap = dax_mmap,
Dan Williamsab68f262016-05-18 09:15:08 -0700526};
527
Dan Williamsebd84d72016-08-11 00:41:51 -0700528static void dax_dev_release(struct device *dev)
Dan Williams043a9252016-08-07 08:23:56 -0700529{
Dan Williamsebd84d72016-08-11 00:41:51 -0700530 struct dax_dev *dax_dev = to_dax_dev(dev);
Dan Williams043a9252016-08-07 08:23:56 -0700531 struct dax_region *dax_region = dax_dev->region;
532
Dan Williamsebd84d72016-08-11 00:41:51 -0700533 ida_simple_remove(&dax_region->ida, dax_dev->id);
534 ida_simple_remove(&dax_minor_ida, MINOR(dev->devt));
535 dax_region_put(dax_region);
Dan Williams3bc52c42016-07-24 21:55:45 -0700536 iput(dax_dev->inode);
Dan Williamsebd84d72016-08-11 00:41:51 -0700537 kfree(dax_dev);
538}
539
540static void unregister_dax_dev(void *dev)
541{
542 struct dax_dev *dax_dev = to_dax_dev(dev);
Dan Williamsba09c012016-07-24 15:55:42 -0700543 struct cdev *cdev = &dax_dev->cdev;
Dan Williamsebd84d72016-08-11 00:41:51 -0700544
Dan Williams043a9252016-08-07 08:23:56 -0700545 dev_dbg(dev, "%s\n", __func__);
546
547 /*
548 * Note, rcu is not protecting the liveness of dax_dev, rcu is
549 * ensuring that any fault handlers that might have seen
550 * dax_dev->alive == true, have completed. Any fault handlers
551 * that start after synchronize_rcu() has started will abort
552 * upon seeing dax_dev->alive == false.
553 */
554 dax_dev->alive = false;
555 synchronize_rcu();
Dan Williams9dc1e492016-08-04 16:53:50 -0700556 unmap_mapping_range(dax_dev->inode->i_mapping, 0, 0, 1);
Dan Williamsba09c012016-07-24 15:55:42 -0700557 cdev_del(cdev);
Dan Williams043a9252016-08-07 08:23:56 -0700558 device_unregister(dev);
Dan Williams043a9252016-08-07 08:23:56 -0700559}
560
Dan Williamsd76911e2016-07-19 17:51:40 -0700561struct dax_dev *devm_create_dax_dev(struct dax_region *dax_region,
562 struct resource *res, int count)
Dan Williams043a9252016-08-07 08:23:56 -0700563{
564 struct device *parent = dax_region->dev;
565 struct dax_dev *dax_dev;
Dan Williams9d2d01a2016-07-19 16:17:58 -0700566 int rc = 0, minor, i;
Dan Williams043a9252016-08-07 08:23:56 -0700567 struct device *dev;
Dan Williamsba09c012016-07-24 15:55:42 -0700568 struct cdev *cdev;
Dan Williams043a9252016-08-07 08:23:56 -0700569 dev_t dev_t;
570
571 dax_dev = kzalloc(sizeof(*dax_dev) + sizeof(*res) * count, GFP_KERNEL);
572 if (!dax_dev)
Dan Williamsd76911e2016-07-19 17:51:40 -0700573 return ERR_PTR(-ENOMEM);
Dan Williams043a9252016-08-07 08:23:56 -0700574
Dan Williams9d2d01a2016-07-19 16:17:58 -0700575 for (i = 0; i < count; i++) {
576 if (!IS_ALIGNED(res[i].start, dax_region->align)
577 || !IS_ALIGNED(resource_size(&res[i]),
578 dax_region->align)) {
579 rc = -EINVAL;
580 break;
581 }
582 dax_dev->res[i].start = res[i].start;
583 dax_dev->res[i].end = res[i].end;
584 }
585
586 if (i < count)
587 goto err_id;
588
Dan Williams043a9252016-08-07 08:23:56 -0700589 dax_dev->id = ida_simple_get(&dax_region->ida, 0, 0, GFP_KERNEL);
590 if (dax_dev->id < 0) {
591 rc = dax_dev->id;
592 goto err_id;
593 }
594
595 minor = ida_simple_get(&dax_minor_ida, 0, 0, GFP_KERNEL);
596 if (minor < 0) {
597 rc = minor;
598 goto err_minor;
599 }
600
Arnd Bergmannbc0a0fe2016-09-08 15:53:28 +0200601 dev_t = MKDEV(MAJOR(dax_devt), minor);
602 dev = &dax_dev->dev;
Dan Williams3bc52c42016-07-24 21:55:45 -0700603 dax_dev->inode = dax_inode_get(&dax_dev->cdev, dev_t);
604 if (!dax_dev->inode) {
605 rc = -ENOMEM;
606 goto err_inode;
607 }
608
Dan Williamsba09c012016-07-24 15:55:42 -0700609 /* device_initialize() so cdev can reference kobj parent */
Dan Williamsebd84d72016-08-11 00:41:51 -0700610 device_initialize(dev);
Dan Williamsba09c012016-07-24 15:55:42 -0700611
612 cdev = &dax_dev->cdev;
613 cdev_init(cdev, &dax_fops);
614 cdev->owner = parent->driver->owner;
615 cdev->kobj.parent = &dev->kobj;
616 rc = cdev_add(&dax_dev->cdev, dev_t, 1);
617 if (rc)
618 goto err_cdev;
619
620 /* from here on we're committed to teardown via dax_dev_release() */
Dan Williamsba09c012016-07-24 15:55:42 -0700621 dax_dev->num_resources = count;
622 dax_dev->alive = true;
623 dax_dev->region = dax_region;
624 kref_get(&dax_region->kref);
625
Dan Williamsebd84d72016-08-11 00:41:51 -0700626 dev->devt = dev_t;
627 dev->class = dax_class;
628 dev->parent = parent;
629 dev->groups = dax_attribute_groups;
630 dev->release = dax_dev_release;
631 dev_set_name(dev, "dax%d.%d", dax_region->id, dax_dev->id);
632 rc = device_add(dev);
633 if (rc) {
634 put_device(dev);
Dan Williamsd76911e2016-07-19 17:51:40 -0700635 return ERR_PTR(rc);
Dan Williamsebd84d72016-08-11 00:41:51 -0700636 }
Dan Williams043a9252016-08-07 08:23:56 -0700637
Dan Williamsd76911e2016-07-19 17:51:40 -0700638 rc = devm_add_action_or_reset(dax_region->dev, unregister_dax_dev, dev);
639 if (rc)
640 return ERR_PTR(rc);
641
642 return dax_dev;
Dan Williams043a9252016-08-07 08:23:56 -0700643
Dan Williamsba09c012016-07-24 15:55:42 -0700644 err_cdev:
Dan Williams3bc52c42016-07-24 21:55:45 -0700645 iput(dax_dev->inode);
646 err_inode:
Dan Williamsba09c012016-07-24 15:55:42 -0700647 ida_simple_remove(&dax_minor_ida, minor);
Dan Williams043a9252016-08-07 08:23:56 -0700648 err_minor:
649 ida_simple_remove(&dax_region->ida, dax_dev->id);
650 err_id:
Dan Williamsebd84d72016-08-11 00:41:51 -0700651 kfree(dax_dev);
Dan Williams043a9252016-08-07 08:23:56 -0700652
Dan Williamsd76911e2016-07-19 17:51:40 -0700653 return ERR_PTR(rc);
Dan Williams043a9252016-08-07 08:23:56 -0700654}
655EXPORT_SYMBOL_GPL(devm_create_dax_dev);
656
Dan Williamsab68f262016-05-18 09:15:08 -0700657static int __init dax_init(void)
658{
659 int rc;
660
Dan Williams3bc52c42016-07-24 21:55:45 -0700661 rc = dax_inode_init();
Dan Williamsba09c012016-07-24 15:55:42 -0700662 if (rc)
Dan Williamsab68f262016-05-18 09:15:08 -0700663 return rc;
Dan Williamsab68f262016-05-18 09:15:08 -0700664
Dan Williams3bc52c42016-07-24 21:55:45 -0700665 nr_dax = max(nr_dax, 256);
666 rc = alloc_chrdev_region(&dax_devt, 0, nr_dax, "dax");
667 if (rc)
668 goto err_chrdev;
Dan Williamsab68f262016-05-18 09:15:08 -0700669
670 dax_class = class_create(THIS_MODULE, "dax");
671 if (IS_ERR(dax_class)) {
Dan Williams3bc52c42016-07-24 21:55:45 -0700672 rc = PTR_ERR(dax_class);
673 goto err_class;
Dan Williamsab68f262016-05-18 09:15:08 -0700674 }
675
676 return 0;
Dan Williams3bc52c42016-07-24 21:55:45 -0700677
678 err_class:
679 unregister_chrdev_region(dax_devt, nr_dax);
680 err_chrdev:
681 dax_inode_exit();
682 return rc;
Dan Williamsab68f262016-05-18 09:15:08 -0700683}
684
685static void __exit dax_exit(void)
686{
687 class_destroy(dax_class);
Dan Williamsba09c012016-07-24 15:55:42 -0700688 unregister_chrdev_region(dax_devt, nr_dax);
Dan Williamsab68f262016-05-18 09:15:08 -0700689 ida_destroy(&dax_minor_ida);
Dan Williams3bc52c42016-07-24 21:55:45 -0700690 dax_inode_exit();
Dan Williamsab68f262016-05-18 09:15:08 -0700691}
692
693MODULE_AUTHOR("Intel Corporation");
694MODULE_LICENSE("GPL v2");
695subsys_initcall(dax_init);
696module_exit(dax_exit);