]>
Commit | Line | Data |
---|---|---|
ab68f262 DW |
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> | |
3bc52c45 | 16 | #include <linux/mount.h> |
ab68f262 | 17 | #include <linux/pfn_t.h> |
3bc52c45 | 18 | #include <linux/hash.h> |
ba09c01d | 19 | #include <linux/cdev.h> |
ab68f262 DW |
20 | #include <linux/slab.h> |
21 | #include <linux/dax.h> | |
22 | #include <linux/fs.h> | |
23 | #include <linux/mm.h> | |
ccdb07f6 | 24 | #include "dax.h" |
ab68f262 | 25 | |
ba09c01d | 26 | static dev_t dax_devt; |
ab68f262 DW |
27 | static struct class *dax_class; |
28 | static DEFINE_IDA(dax_minor_ida); | |
ba09c01d DW |
29 | static int nr_dax = CONFIG_NR_DEV_DAX; |
30 | module_param(nr_dax, int, S_IRUGO); | |
3bc52c45 DW |
31 | static struct vfsmount *dax_mnt; |
32 | static struct kmem_cache *dax_cache __read_mostly; | |
33 | static struct super_block *dax_superblock __read_mostly; | |
ba09c01d | 34 | MODULE_PARM_DESC(nr_dax, "max number of device-dax instances"); |
ab68f262 DW |
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 | |
ba09c01d | 61 | * @cdev - core chardev data |
dee41079 | 62 | * @alive - !alive + rcu grace period == no new mappings can be established |
ab68f262 DW |
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; | |
3bc52c45 | 69 | struct inode *inode; |
ebd84d72 | 70 | struct device dev; |
ba09c01d | 71 | struct cdev cdev; |
dee41079 | 72 | bool alive; |
ab68f262 DW |
73 | int id; |
74 | int num_resources; | |
75 | struct resource res[0]; | |
76 | }; | |
77 | ||
d7fe1a67 DW |
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 | ||
3bc52c45 | 145 | static struct inode *dax_alloc_inode(struct super_block *sb) |
ab68f262 | 146 | { |
3bc52c45 DW |
147 | return kmem_cache_alloc(dax_cache, GFP_KERNEL); |
148 | } | |
ab68f262 | 149 | |
3bc52c45 DW |
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); | |
ab68f262 DW |
155 | } |
156 | ||
3bc52c45 | 157 | static void dax_destroy_inode(struct inode *inode) |
ab68f262 | 158 | { |
3bc52c45 | 159 | call_rcu(&inode->i_rcu, dax_i_callback); |
ab68f262 | 160 | } |
ab68f262 | 161 | |
3bc52c45 DW |
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) | |
ab68f262 | 171 | { |
3bc52c45 DW |
172 | return mount_pseudo(fs_type, "dax:", &dax_sops, NULL, DAXFS_MAGIC); |
173 | } | |
ab68f262 | 174 | |
3bc52c45 DW |
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; | |
ab68f262 DW |
247 | } |
248 | ||
3bc52c45 DW |
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 | ||
ab68f262 DW |
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) | |
ab68f262 | 265 | { |
ab68f262 | 266 | kref_put(&dax_region->kref, dax_region_free); |
ab68f262 | 267 | } |
ab68f262 | 268 | EXPORT_SYMBOL_GPL(dax_region_put); |
ab68f262 | 269 | |
d7fe1a67 DW |
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 | ||
ab68f262 DW |
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 | ||
d7fe1a67 DW |
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 | ||
9d2d01a0 DW |
295 | if (!IS_ALIGNED(res->start, align) |
296 | || !IS_ALIGNED(resource_size(res), align)) | |
297 | return NULL; | |
ab68f262 | 298 | |
9d2d01a0 | 299 | dax_region = kzalloc(sizeof(*dax_region), GFP_KERNEL); |
ab68f262 DW |
300 | if (!dax_region) |
301 | return NULL; | |
302 | ||
d7fe1a67 | 303 | dev_set_drvdata(parent, dax_region); |
ab68f262 DW |
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; | |
d7fe1a67 DW |
312 | if (sysfs_create_groups(&parent->kobj, dax_region_attribute_groups)) { |
313 | kfree(dax_region); | |
314 | return NULL;; | |
315 | } | |
ab68f262 | 316 | |
d7fe1a67 DW |
317 | kref_get(&dax_region->kref); |
318 | if (devm_add_action_or_reset(parent, dax_region_unregister, dax_region)) | |
319 | return NULL; | |
ab68f262 DW |
320 | return dax_region; |
321 | } | |
322 | EXPORT_SYMBOL_GPL(alloc_dax_region); | |
323 | ||
ebd84d72 DW |
324 | static struct dax_dev *to_dax_dev(struct device *dev) |
325 | { | |
326 | return container_of(dev, struct dax_dev, dev); | |
327 | } | |
328 | ||
ab68f262 DW |
329 | static ssize_t size_show(struct device *dev, |
330 | struct device_attribute *attr, char *buf) | |
331 | { | |
ebd84d72 | 332 | struct dax_dev *dax_dev = to_dax_dev(dev); |
ab68f262 DW |
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 | ||
dee41079 DW |
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; | |
ebd84d72 | 361 | struct device *dev = &dax_dev->dev; |
dee41079 DW |
362 | unsigned long mask; |
363 | ||
364 | if (!dax_dev->alive) | |
365 | return -ENXIO; | |
366 | ||
4cb19355 | 367 | /* prevent private mappings from being established */ |
325896ff | 368 | if ((vma->vm_flags & VM_MAYSHARE) != VM_MAYSHARE) { |
dee41079 DW |
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 | ||
11bac800 | 422 | static int __dax_dev_fault(struct dax_dev *dax_dev, struct vm_fault *vmf) |
dee41079 | 423 | { |
ebd84d72 | 424 | struct device *dev = &dax_dev->dev; |
dee41079 DW |
425 | struct dax_region *dax_region; |
426 | int rc = VM_FAULT_SIGBUS; | |
427 | phys_addr_t phys; | |
428 | pfn_t pfn; | |
429 | ||
11bac800 | 430 | if (check_vma(dax_dev, vmf->vma, __func__)) |
dee41079 DW |
431 | return VM_FAULT_SIGBUS; |
432 | ||
433 | dax_region = dax_dev->region; | |
434 | if (dax_region->align > PAGE_SIZE) { | |
435 | dev_dbg(dev, "%s: alignment > fault size\n", __func__); | |
436 | return VM_FAULT_SIGBUS; | |
437 | } | |
438 | ||
439 | phys = pgoff_to_phys(dax_dev, vmf->pgoff, PAGE_SIZE); | |
440 | if (phys == -1) { | |
441 | dev_dbg(dev, "%s: phys_to_pgoff(%#lx) failed\n", __func__, | |
442 | vmf->pgoff); | |
443 | return VM_FAULT_SIGBUS; | |
444 | } | |
445 | ||
446 | pfn = phys_to_pfn_t(phys, dax_region->pfn_flags); | |
447 | ||
11bac800 | 448 | rc = vm_insert_mixed(vmf->vma, vmf->address, pfn); |
dee41079 DW |
449 | |
450 | if (rc == -ENOMEM) | |
451 | return VM_FAULT_OOM; | |
452 | if (rc < 0 && rc != -EBUSY) | |
453 | return VM_FAULT_SIGBUS; | |
454 | ||
455 | return VM_FAULT_NOPAGE; | |
456 | } | |
457 | ||
11bac800 | 458 | static int dax_dev_fault(struct vm_fault *vmf) |
dee41079 | 459 | { |
11bac800 | 460 | struct vm_area_struct *vma = vmf->vma; |
dee41079 DW |
461 | int rc; |
462 | struct file *filp = vma->vm_file; | |
463 | struct dax_dev *dax_dev = filp->private_data; | |
464 | ||
ebd84d72 | 465 | dev_dbg(&dax_dev->dev, "%s: %s: %s (%#lx - %#lx)\n", __func__, |
dee41079 DW |
466 | current->comm, (vmf->flags & FAULT_FLAG_WRITE) |
467 | ? "write" : "read", vma->vm_start, vma->vm_end); | |
468 | rcu_read_lock(); | |
11bac800 | 469 | rc = __dax_dev_fault(dax_dev, vmf); |
dee41079 DW |
470 | rcu_read_unlock(); |
471 | ||
472 | return rc; | |
473 | } | |
474 | ||
f4200391 | 475 | static int __dax_dev_pmd_fault(struct dax_dev *dax_dev, struct vm_fault *vmf) |
dee41079 | 476 | { |
d8a849e1 | 477 | unsigned long pmd_addr = vmf->address & PMD_MASK; |
ebd84d72 | 478 | struct device *dev = &dax_dev->dev; |
dee41079 DW |
479 | struct dax_region *dax_region; |
480 | phys_addr_t phys; | |
481 | pgoff_t pgoff; | |
482 | pfn_t pfn; | |
483 | ||
f4200391 | 484 | if (check_vma(dax_dev, vmf->vma, __func__)) |
dee41079 DW |
485 | return VM_FAULT_SIGBUS; |
486 | ||
487 | dax_region = dax_dev->region; | |
488 | if (dax_region->align > PMD_SIZE) { | |
489 | dev_dbg(dev, "%s: alignment > fault size\n", __func__); | |
490 | return VM_FAULT_SIGBUS; | |
491 | } | |
492 | ||
493 | /* dax pmd mappings require pfn_t_devmap() */ | |
494 | if ((dax_region->pfn_flags & (PFN_DEV|PFN_MAP)) != (PFN_DEV|PFN_MAP)) { | |
495 | dev_dbg(dev, "%s: alignment > fault size\n", __func__); | |
496 | return VM_FAULT_SIGBUS; | |
497 | } | |
498 | ||
f4200391 | 499 | pgoff = linear_page_index(vmf->vma, pmd_addr); |
4c3cb6e9 | 500 | phys = pgoff_to_phys(dax_dev, pgoff, PMD_SIZE); |
dee41079 DW |
501 | if (phys == -1) { |
502 | dev_dbg(dev, "%s: phys_to_pgoff(%#lx) failed\n", __func__, | |
503 | pgoff); | |
504 | return VM_FAULT_SIGBUS; | |
505 | } | |
506 | ||
507 | pfn = phys_to_pfn_t(phys, dax_region->pfn_flags); | |
508 | ||
f4200391 | 509 | return vmf_insert_pfn_pmd(vmf->vma, vmf->address, vmf->pmd, pfn, |
d8a849e1 | 510 | vmf->flags & FAULT_FLAG_WRITE); |
dee41079 DW |
511 | } |
512 | ||
f4200391 | 513 | static int dax_dev_pmd_fault(struct vm_fault *vmf) |
dee41079 DW |
514 | { |
515 | int rc; | |
f4200391 | 516 | struct file *filp = vmf->vma->vm_file; |
dee41079 DW |
517 | struct dax_dev *dax_dev = filp->private_data; |
518 | ||
ebd84d72 | 519 | dev_dbg(&dax_dev->dev, "%s: %s: %s (%#lx - %#lx)\n", __func__, |
d8a849e1 | 520 | current->comm, (vmf->flags & FAULT_FLAG_WRITE) |
f4200391 DJ |
521 | ? "write" : "read", |
522 | vmf->vma->vm_start, vmf->vma->vm_end); | |
dee41079 DW |
523 | |
524 | rcu_read_lock(); | |
f4200391 | 525 | rc = __dax_dev_pmd_fault(dax_dev, vmf); |
dee41079 DW |
526 | rcu_read_unlock(); |
527 | ||
528 | return rc; | |
529 | } | |
530 | ||
dee41079 DW |
531 | static const struct vm_operations_struct dax_dev_vm_ops = { |
532 | .fault = dax_dev_fault, | |
533 | .pmd_fault = dax_dev_pmd_fault, | |
dee41079 DW |
534 | }; |
535 | ||
af69f51e | 536 | static int dax_mmap(struct file *filp, struct vm_area_struct *vma) |
dee41079 DW |
537 | { |
538 | struct dax_dev *dax_dev = filp->private_data; | |
539 | int rc; | |
540 | ||
ebd84d72 | 541 | dev_dbg(&dax_dev->dev, "%s\n", __func__); |
dee41079 DW |
542 | |
543 | rc = check_vma(dax_dev, vma, __func__); | |
544 | if (rc) | |
545 | return rc; | |
546 | ||
dee41079 DW |
547 | vma->vm_ops = &dax_dev_vm_ops; |
548 | vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE; | |
549 | return 0; | |
043a9255 DW |
550 | } |
551 | ||
552 | /* return an unmapped area aligned to the dax region specified alignment */ | |
af69f51e | 553 | static unsigned long dax_get_unmapped_area(struct file *filp, |
043a9255 DW |
554 | unsigned long addr, unsigned long len, unsigned long pgoff, |
555 | unsigned long flags) | |
556 | { | |
557 | unsigned long off, off_end, off_align, len_align, addr_align, align; | |
558 | struct dax_dev *dax_dev = filp ? filp->private_data : NULL; | |
559 | struct dax_region *dax_region; | |
560 | ||
561 | if (!dax_dev || addr) | |
562 | goto out; | |
563 | ||
564 | dax_region = dax_dev->region; | |
565 | align = dax_region->align; | |
566 | off = pgoff << PAGE_SHIFT; | |
567 | off_end = off + len; | |
568 | off_align = round_up(off, align); | |
569 | ||
570 | if ((off_end <= off_align) || ((off_end - off_align) < align)) | |
571 | goto out; | |
572 | ||
573 | len_align = len + align; | |
574 | if ((off + len_align) < off) | |
575 | goto out; | |
dee41079 | 576 | |
043a9255 DW |
577 | addr_align = current->mm->get_unmapped_area(filp, addr, len_align, |
578 | pgoff, flags); | |
579 | if (!IS_ERR_VALUE(addr_align)) { | |
580 | addr_align += (off - addr_align) & (align - 1); | |
581 | return addr_align; | |
582 | } | |
583 | out: | |
584 | return current->mm->get_unmapped_area(filp, addr, len, pgoff, flags); | |
585 | } | |
586 | ||
af69f51e | 587 | static int dax_open(struct inode *inode, struct file *filp) |
043a9255 | 588 | { |
ba09c01d | 589 | struct dax_dev *dax_dev; |
043a9255 | 590 | |
ba09c01d DW |
591 | dax_dev = container_of(inode->i_cdev, struct dax_dev, cdev); |
592 | dev_dbg(&dax_dev->dev, "%s\n", __func__); | |
3bc52c45 DW |
593 | inode->i_mapping = dax_dev->inode->i_mapping; |
594 | inode->i_mapping->host = dax_dev->inode; | |
595 | filp->f_mapping = inode->i_mapping; | |
ebd84d72 DW |
596 | filp->private_data = dax_dev; |
597 | inode->i_flags = S_DAX; | |
043a9255 | 598 | |
043a9255 DW |
599 | return 0; |
600 | } | |
dee41079 | 601 | |
af69f51e | 602 | static int dax_release(struct inode *inode, struct file *filp) |
043a9255 DW |
603 | { |
604 | struct dax_dev *dax_dev = filp->private_data; | |
043a9255 | 605 | |
ba09c01d | 606 | dev_dbg(&dax_dev->dev, "%s\n", __func__); |
043a9255 | 607 | return 0; |
dee41079 DW |
608 | } |
609 | ||
ab68f262 DW |
610 | static const struct file_operations dax_fops = { |
611 | .llseek = noop_llseek, | |
612 | .owner = THIS_MODULE, | |
af69f51e DW |
613 | .open = dax_open, |
614 | .release = dax_release, | |
615 | .get_unmapped_area = dax_get_unmapped_area, | |
616 | .mmap = dax_mmap, | |
ab68f262 DW |
617 | }; |
618 | ||
ebd84d72 | 619 | static void dax_dev_release(struct device *dev) |
043a9255 | 620 | { |
ebd84d72 | 621 | struct dax_dev *dax_dev = to_dax_dev(dev); |
043a9255 DW |
622 | struct dax_region *dax_region = dax_dev->region; |
623 | ||
ebd84d72 DW |
624 | ida_simple_remove(&dax_region->ida, dax_dev->id); |
625 | ida_simple_remove(&dax_minor_ida, MINOR(dev->devt)); | |
626 | dax_region_put(dax_region); | |
3bc52c45 | 627 | iput(dax_dev->inode); |
ebd84d72 DW |
628 | kfree(dax_dev); |
629 | } | |
630 | ||
631 | static void unregister_dax_dev(void *dev) | |
632 | { | |
633 | struct dax_dev *dax_dev = to_dax_dev(dev); | |
ba09c01d | 634 | struct cdev *cdev = &dax_dev->cdev; |
ebd84d72 | 635 | |
043a9255 DW |
636 | dev_dbg(dev, "%s\n", __func__); |
637 | ||
638 | /* | |
639 | * Note, rcu is not protecting the liveness of dax_dev, rcu is | |
640 | * ensuring that any fault handlers that might have seen | |
641 | * dax_dev->alive == true, have completed. Any fault handlers | |
642 | * that start after synchronize_rcu() has started will abort | |
643 | * upon seeing dax_dev->alive == false. | |
644 | */ | |
645 | dax_dev->alive = false; | |
646 | synchronize_rcu(); | |
9dc1e492 | 647 | unmap_mapping_range(dax_dev->inode->i_mapping, 0, 0, 1); |
ba09c01d | 648 | cdev_del(cdev); |
043a9255 | 649 | device_unregister(dev); |
043a9255 DW |
650 | } |
651 | ||
d76911ee DW |
652 | struct dax_dev *devm_create_dax_dev(struct dax_region *dax_region, |
653 | struct resource *res, int count) | |
043a9255 DW |
654 | { |
655 | struct device *parent = dax_region->dev; | |
656 | struct dax_dev *dax_dev; | |
9d2d01a0 | 657 | int rc = 0, minor, i; |
043a9255 | 658 | struct device *dev; |
ba09c01d | 659 | struct cdev *cdev; |
043a9255 DW |
660 | dev_t dev_t; |
661 | ||
662 | dax_dev = kzalloc(sizeof(*dax_dev) + sizeof(*res) * count, GFP_KERNEL); | |
663 | if (!dax_dev) | |
d76911ee | 664 | return ERR_PTR(-ENOMEM); |
043a9255 | 665 | |
9d2d01a0 DW |
666 | for (i = 0; i < count; i++) { |
667 | if (!IS_ALIGNED(res[i].start, dax_region->align) | |
668 | || !IS_ALIGNED(resource_size(&res[i]), | |
669 | dax_region->align)) { | |
670 | rc = -EINVAL; | |
671 | break; | |
672 | } | |
673 | dax_dev->res[i].start = res[i].start; | |
674 | dax_dev->res[i].end = res[i].end; | |
675 | } | |
676 | ||
677 | if (i < count) | |
678 | goto err_id; | |
679 | ||
043a9255 DW |
680 | dax_dev->id = ida_simple_get(&dax_region->ida, 0, 0, GFP_KERNEL); |
681 | if (dax_dev->id < 0) { | |
682 | rc = dax_dev->id; | |
683 | goto err_id; | |
684 | } | |
685 | ||
686 | minor = ida_simple_get(&dax_minor_ida, 0, 0, GFP_KERNEL); | |
687 | if (minor < 0) { | |
688 | rc = minor; | |
689 | goto err_minor; | |
690 | } | |
691 | ||
bc0a0fe9 AB |
692 | dev_t = MKDEV(MAJOR(dax_devt), minor); |
693 | dev = &dax_dev->dev; | |
3bc52c45 DW |
694 | dax_dev->inode = dax_inode_get(&dax_dev->cdev, dev_t); |
695 | if (!dax_dev->inode) { | |
696 | rc = -ENOMEM; | |
697 | goto err_inode; | |
698 | } | |
699 | ||
ba09c01d | 700 | /* device_initialize() so cdev can reference kobj parent */ |
ebd84d72 | 701 | device_initialize(dev); |
ba09c01d DW |
702 | |
703 | cdev = &dax_dev->cdev; | |
704 | cdev_init(cdev, &dax_fops); | |
705 | cdev->owner = parent->driver->owner; | |
706 | cdev->kobj.parent = &dev->kobj; | |
707 | rc = cdev_add(&dax_dev->cdev, dev_t, 1); | |
708 | if (rc) | |
709 | goto err_cdev; | |
710 | ||
711 | /* from here on we're committed to teardown via dax_dev_release() */ | |
ba09c01d DW |
712 | dax_dev->num_resources = count; |
713 | dax_dev->alive = true; | |
714 | dax_dev->region = dax_region; | |
715 | kref_get(&dax_region->kref); | |
716 | ||
ebd84d72 DW |
717 | dev->devt = dev_t; |
718 | dev->class = dax_class; | |
719 | dev->parent = parent; | |
720 | dev->groups = dax_attribute_groups; | |
721 | dev->release = dax_dev_release; | |
722 | dev_set_name(dev, "dax%d.%d", dax_region->id, dax_dev->id); | |
723 | rc = device_add(dev); | |
724 | if (rc) { | |
725 | put_device(dev); | |
d76911ee | 726 | return ERR_PTR(rc); |
ebd84d72 | 727 | } |
043a9255 | 728 | |
d76911ee DW |
729 | rc = devm_add_action_or_reset(dax_region->dev, unregister_dax_dev, dev); |
730 | if (rc) | |
731 | return ERR_PTR(rc); | |
732 | ||
733 | return dax_dev; | |
043a9255 | 734 | |
ba09c01d | 735 | err_cdev: |
3bc52c45 DW |
736 | iput(dax_dev->inode); |
737 | err_inode: | |
ba09c01d | 738 | ida_simple_remove(&dax_minor_ida, minor); |
043a9255 DW |
739 | err_minor: |
740 | ida_simple_remove(&dax_region->ida, dax_dev->id); | |
741 | err_id: | |
ebd84d72 | 742 | kfree(dax_dev); |
043a9255 | 743 | |
d76911ee | 744 | return ERR_PTR(rc); |
043a9255 DW |
745 | } |
746 | EXPORT_SYMBOL_GPL(devm_create_dax_dev); | |
747 | ||
ab68f262 DW |
748 | static int __init dax_init(void) |
749 | { | |
750 | int rc; | |
751 | ||
3bc52c45 DW |
752 | rc = dax_inode_init(); |
753 | if (rc) | |
ab68f262 | 754 | return rc; |
3bc52c45 | 755 | |
ba09c01d DW |
756 | nr_dax = max(nr_dax, 256); |
757 | rc = alloc_chrdev_region(&dax_devt, 0, nr_dax, "dax"); | |
758 | if (rc) | |
3bc52c45 | 759 | goto err_chrdev; |
ab68f262 DW |
760 | |
761 | dax_class = class_create(THIS_MODULE, "dax"); | |
762 | if (IS_ERR(dax_class)) { | |
3bc52c45 DW |
763 | rc = PTR_ERR(dax_class); |
764 | goto err_class; | |
ab68f262 DW |
765 | } |
766 | ||
767 | return 0; | |
3bc52c45 DW |
768 | |
769 | err_class: | |
770 | unregister_chrdev_region(dax_devt, nr_dax); | |
771 | err_chrdev: | |
772 | dax_inode_exit(); | |
773 | return rc; | |
ab68f262 DW |
774 | } |
775 | ||
776 | static void __exit dax_exit(void) | |
777 | { | |
778 | class_destroy(dax_class); | |
ba09c01d | 779 | unregister_chrdev_region(dax_devt, nr_dax); |
ab68f262 | 780 | ida_destroy(&dax_minor_ida); |
3bc52c45 | 781 | dax_inode_exit(); |
ab68f262 DW |
782 | } |
783 | ||
784 | MODULE_AUTHOR("Intel Corporation"); | |
785 | MODULE_LICENSE("GPL v2"); | |
786 | subsys_initcall(dax_init); | |
787 | module_exit(dax_exit); |