2 * Copyright(c) 2016 Intel Corporation. All rights reserved.
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.
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.
13 #include <linux/pagemap.h>
14 #include <linux/module.h>
15 #include <linux/device.h>
16 #include <linux/pfn_t.h>
17 #include <linux/slab.h>
18 #include <linux/dax.h>
23 static struct class *dax_class
;
24 static DEFINE_IDA(dax_minor_ida
);
27 * struct dax_region - mapping infrastructure for dax devices
28 * @id: kernel-wide unique region for a memory range
29 * @base: linear address corresponding to @res
30 * @kref: to pin while other agents have a need to do lookups
31 * @dev: parent device backing this region
32 * @align: allocation and mapping alignment for child dax devices
33 * @res: physical address range of the region
34 * @pfn_flags: identify whether the pfns are paged back or not
44 unsigned long pfn_flags
;
48 * struct dax_dev - subdivision of a dax region
49 * @region - parent region
50 * @dev - device backing the character device
51 * @kref - enable this data to be tracked in filp->private_data
52 * @alive - !alive + rcu grace period == no new mappings can be established
53 * @id - child id in the region
54 * @num_resources - number of physical address extents in this device
55 * @res - array of physical address ranges
58 struct dax_region
*region
;
64 struct resource res
[0];
67 static void dax_region_free(struct kref
*kref
)
69 struct dax_region
*dax_region
;
71 dax_region
= container_of(kref
, struct dax_region
, kref
);
75 void dax_region_put(struct dax_region
*dax_region
)
77 kref_put(&dax_region
->kref
, dax_region_free
);
79 EXPORT_SYMBOL_GPL(dax_region_put
);
81 static void dax_dev_free(struct kref
*kref
)
83 struct dax_dev
*dax_dev
;
85 dax_dev
= container_of(kref
, struct dax_dev
, kref
);
86 dax_region_put(dax_dev
->region
);
90 static void dax_dev_put(struct dax_dev
*dax_dev
)
92 kref_put(&dax_dev
->kref
, dax_dev_free
);
95 struct dax_region
*alloc_dax_region(struct device
*parent
, int region_id
,
96 struct resource
*res
, unsigned int align
, void *addr
,
97 unsigned long pfn_flags
)
99 struct dax_region
*dax_region
;
101 dax_region
= kzalloc(sizeof(*dax_region
), GFP_KERNEL
);
106 memcpy(&dax_region
->res
, res
, sizeof(*res
));
107 dax_region
->pfn_flags
= pfn_flags
;
108 kref_init(&dax_region
->kref
);
109 dax_region
->id
= region_id
;
110 ida_init(&dax_region
->ida
);
111 dax_region
->align
= align
;
112 dax_region
->dev
= parent
;
113 dax_region
->base
= addr
;
117 EXPORT_SYMBOL_GPL(alloc_dax_region
);
119 static ssize_t
size_show(struct device
*dev
,
120 struct device_attribute
*attr
, char *buf
)
122 struct dax_dev
*dax_dev
= dev_get_drvdata(dev
);
123 unsigned long long size
= 0;
126 for (i
= 0; i
< dax_dev
->num_resources
; i
++)
127 size
+= resource_size(&dax_dev
->res
[i
]);
129 return sprintf(buf
, "%llu\n", size
);
131 static DEVICE_ATTR_RO(size
);
133 static struct attribute
*dax_device_attributes
[] = {
138 static const struct attribute_group dax_device_attribute_group
= {
139 .attrs
= dax_device_attributes
,
142 static const struct attribute_group
*dax_attribute_groups
[] = {
143 &dax_device_attribute_group
,
147 static void unregister_dax_dev(void *_dev
)
149 struct device
*dev
= _dev
;
150 struct dax_dev
*dax_dev
= dev_get_drvdata(dev
);
151 struct dax_region
*dax_region
= dax_dev
->region
;
153 dev_dbg(dev
, "%s\n", __func__
);
156 * Note, rcu is not protecting the liveness of dax_dev, rcu is
157 * ensuring that any fault handlers that might have seen
158 * dax_dev->alive == true, have completed. Any fault handlers
159 * that start after synchronize_rcu() has started will abort
160 * upon seeing dax_dev->alive == false.
162 dax_dev
->alive
= false;
166 device_unregister(dev
);
167 ida_simple_remove(&dax_region
->ida
, dax_dev
->id
);
168 ida_simple_remove(&dax_minor_ida
, MINOR(dev
->devt
));
170 dax_dev_put(dax_dev
);
173 int devm_create_dax_dev(struct dax_region
*dax_region
, struct resource
*res
,
176 struct device
*parent
= dax_region
->dev
;
177 struct dax_dev
*dax_dev
;
182 dax_dev
= kzalloc(sizeof(*dax_dev
) + sizeof(*res
) * count
, GFP_KERNEL
);
185 memcpy(dax_dev
->res
, res
, sizeof(*res
) * count
);
186 dax_dev
->num_resources
= count
;
187 kref_init(&dax_dev
->kref
);
188 dax_dev
->alive
= true;
189 dax_dev
->region
= dax_region
;
190 kref_get(&dax_region
->kref
);
192 dax_dev
->id
= ida_simple_get(&dax_region
->ida
, 0, 0, GFP_KERNEL
);
193 if (dax_dev
->id
< 0) {
198 minor
= ida_simple_get(&dax_minor_ida
, 0, 0, GFP_KERNEL
);
204 dev_t
= MKDEV(dax_major
, minor
);
205 dev
= device_create_with_groups(dax_class
, parent
, dev_t
, dax_dev
,
206 dax_attribute_groups
, "dax%d.%d", dax_region
->id
,
214 rc
= devm_add_action(dax_region
->dev
, unregister_dax_dev
, dev
);
216 unregister_dax_dev(dev
);
223 ida_simple_remove(&dax_minor_ida
, minor
);
225 ida_simple_remove(&dax_region
->ida
, dax_dev
->id
);
227 dax_dev_put(dax_dev
);
231 EXPORT_SYMBOL_GPL(devm_create_dax_dev
);
233 /* return an unmapped area aligned to the dax region specified alignment */
234 static unsigned long dax_dev_get_unmapped_area(struct file
*filp
,
235 unsigned long addr
, unsigned long len
, unsigned long pgoff
,
238 unsigned long off
, off_end
, off_align
, len_align
, addr_align
, align
;
239 struct dax_dev
*dax_dev
= filp
? filp
->private_data
: NULL
;
240 struct dax_region
*dax_region
;
242 if (!dax_dev
|| addr
)
245 dax_region
= dax_dev
->region
;
246 align
= dax_region
->align
;
247 off
= pgoff
<< PAGE_SHIFT
;
249 off_align
= round_up(off
, align
);
251 if ((off_end
<= off_align
) || ((off_end
- off_align
) < align
))
254 len_align
= len
+ align
;
255 if ((off
+ len_align
) < off
)
258 addr_align
= current
->mm
->get_unmapped_area(filp
, addr
, len_align
,
260 if (!IS_ERR_VALUE(addr_align
)) {
261 addr_align
+= (off
- addr_align
) & (align
- 1);
265 return current
->mm
->get_unmapped_area(filp
, addr
, len
, pgoff
, flags
);
268 static int __match_devt(struct device
*dev
, const void *data
)
270 const dev_t
*devt
= data
;
272 return dev
->devt
== *devt
;
275 static struct device
*dax_dev_find(dev_t dev_t
)
277 return class_find_device(dax_class
, NULL
, &dev_t
, __match_devt
);
280 static int dax_dev_open(struct inode
*inode
, struct file
*filp
)
282 struct dax_dev
*dax_dev
= NULL
;
285 dev
= dax_dev_find(inode
->i_rdev
);
290 dax_dev
= dev_get_drvdata(dev
);
292 dev_dbg(dev
, "%s\n", __func__
);
293 filp
->private_data
= dax_dev
;
294 kref_get(&dax_dev
->kref
);
295 inode
->i_flags
= S_DAX
;
306 static int dax_dev_release(struct inode
*inode
, struct file
*filp
)
308 struct dax_dev
*dax_dev
= filp
->private_data
;
309 struct device
*dev
= dax_dev
->dev
;
311 dev_dbg(dax_dev
->dev
, "%s\n", __func__
);
312 dax_dev_put(dax_dev
);
318 static int check_vma(struct dax_dev
*dax_dev
, struct vm_area_struct
*vma
,
321 struct dax_region
*dax_region
= dax_dev
->region
;
322 struct device
*dev
= dax_dev
->dev
;
328 /* prevent private / writable mappings from being established */
329 if ((vma
->vm_flags
& (VM_NORESERVE
|VM_SHARED
|VM_WRITE
)) == VM_WRITE
) {
330 dev_info(dev
, "%s: %s: fail, attempted private mapping\n",
331 current
->comm
, func
);
335 mask
= dax_region
->align
- 1;
336 if (vma
->vm_start
& mask
|| vma
->vm_end
& mask
) {
337 dev_info(dev
, "%s: %s: fail, unaligned vma (%#lx - %#lx, %#lx)\n",
338 current
->comm
, func
, vma
->vm_start
, vma
->vm_end
,
343 if ((dax_region
->pfn_flags
& (PFN_DEV
|PFN_MAP
)) == PFN_DEV
344 && (vma
->vm_flags
& VM_DONTCOPY
) == 0) {
345 dev_info(dev
, "%s: %s: fail, dax range requires MADV_DONTFORK\n",
346 current
->comm
, func
);
350 if (!vma_is_dax(vma
)) {
351 dev_info(dev
, "%s: %s: fail, vma is not DAX capable\n",
352 current
->comm
, func
);
359 static phys_addr_t
pgoff_to_phys(struct dax_dev
*dax_dev
, pgoff_t pgoff
,
362 struct resource
*res
;
366 for (i
= 0; i
< dax_dev
->num_resources
; i
++) {
367 res
= &dax_dev
->res
[i
];
368 phys
= pgoff
* PAGE_SIZE
+ res
->start
;
369 if (phys
>= res
->start
&& phys
<= res
->end
)
371 pgoff
-= PHYS_PFN(resource_size(res
));
374 if (i
< dax_dev
->num_resources
) {
375 res
= &dax_dev
->res
[i
];
376 if (phys
+ size
- 1 <= res
->end
)
383 static int __dax_dev_fault(struct dax_dev
*dax_dev
, struct vm_area_struct
*vma
,
384 struct vm_fault
*vmf
)
386 unsigned long vaddr
= (unsigned long) vmf
->virtual_address
;
387 struct device
*dev
= dax_dev
->dev
;
388 struct dax_region
*dax_region
;
389 int rc
= VM_FAULT_SIGBUS
;
393 if (check_vma(dax_dev
, vma
, __func__
))
394 return VM_FAULT_SIGBUS
;
396 dax_region
= dax_dev
->region
;
397 if (dax_region
->align
> PAGE_SIZE
) {
398 dev_dbg(dev
, "%s: alignment > fault size\n", __func__
);
399 return VM_FAULT_SIGBUS
;
402 phys
= pgoff_to_phys(dax_dev
, vmf
->pgoff
, PAGE_SIZE
);
404 dev_dbg(dev
, "%s: phys_to_pgoff(%#lx) failed\n", __func__
,
406 return VM_FAULT_SIGBUS
;
409 pfn
= phys_to_pfn_t(phys
, dax_region
->pfn_flags
);
411 rc
= vm_insert_mixed(vma
, vaddr
, pfn
);
415 if (rc
< 0 && rc
!= -EBUSY
)
416 return VM_FAULT_SIGBUS
;
418 return VM_FAULT_NOPAGE
;
421 static int dax_dev_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
424 struct file
*filp
= vma
->vm_file
;
425 struct dax_dev
*dax_dev
= filp
->private_data
;
427 dev_dbg(dax_dev
->dev
, "%s: %s: %s (%#lx - %#lx)\n", __func__
,
428 current
->comm
, (vmf
->flags
& FAULT_FLAG_WRITE
)
429 ? "write" : "read", vma
->vm_start
, vma
->vm_end
);
431 rc
= __dax_dev_fault(dax_dev
, vma
, vmf
);
437 static int __dax_dev_pmd_fault(struct dax_dev
*dax_dev
,
438 struct vm_area_struct
*vma
, unsigned long addr
, pmd_t
*pmd
,
441 unsigned long pmd_addr
= addr
& PMD_MASK
;
442 struct device
*dev
= dax_dev
->dev
;
443 struct dax_region
*dax_region
;
448 if (check_vma(dax_dev
, vma
, __func__
))
449 return VM_FAULT_SIGBUS
;
451 dax_region
= dax_dev
->region
;
452 if (dax_region
->align
> PMD_SIZE
) {
453 dev_dbg(dev
, "%s: alignment > fault size\n", __func__
);
454 return VM_FAULT_SIGBUS
;
457 /* dax pmd mappings require pfn_t_devmap() */
458 if ((dax_region
->pfn_flags
& (PFN_DEV
|PFN_MAP
)) != (PFN_DEV
|PFN_MAP
)) {
459 dev_dbg(dev
, "%s: alignment > fault size\n", __func__
);
460 return VM_FAULT_SIGBUS
;
463 pgoff
= linear_page_index(vma
, pmd_addr
);
464 phys
= pgoff_to_phys(dax_dev
, pgoff
, PAGE_SIZE
);
466 dev_dbg(dev
, "%s: phys_to_pgoff(%#lx) failed\n", __func__
,
468 return VM_FAULT_SIGBUS
;
471 pfn
= phys_to_pfn_t(phys
, dax_region
->pfn_flags
);
473 return vmf_insert_pfn_pmd(vma
, addr
, pmd
, pfn
,
474 flags
& FAULT_FLAG_WRITE
);
477 static int dax_dev_pmd_fault(struct vm_area_struct
*vma
, unsigned long addr
,
478 pmd_t
*pmd
, unsigned int flags
)
481 struct file
*filp
= vma
->vm_file
;
482 struct dax_dev
*dax_dev
= filp
->private_data
;
484 dev_dbg(dax_dev
->dev
, "%s: %s: %s (%#lx - %#lx)\n", __func__
,
485 current
->comm
, (flags
& FAULT_FLAG_WRITE
)
486 ? "write" : "read", vma
->vm_start
, vma
->vm_end
);
489 rc
= __dax_dev_pmd_fault(dax_dev
, vma
, addr
, pmd
, flags
);
495 static void dax_dev_vm_open(struct vm_area_struct
*vma
)
497 struct file
*filp
= vma
->vm_file
;
498 struct dax_dev
*dax_dev
= filp
->private_data
;
500 dev_dbg(dax_dev
->dev
, "%s\n", __func__
);
501 kref_get(&dax_dev
->kref
);
504 static void dax_dev_vm_close(struct vm_area_struct
*vma
)
506 struct file
*filp
= vma
->vm_file
;
507 struct dax_dev
*dax_dev
= filp
->private_data
;
509 dev_dbg(dax_dev
->dev
, "%s\n", __func__
);
510 dax_dev_put(dax_dev
);
513 static const struct vm_operations_struct dax_dev_vm_ops
= {
514 .fault
= dax_dev_fault
,
515 .pmd_fault
= dax_dev_pmd_fault
,
516 .open
= dax_dev_vm_open
,
517 .close
= dax_dev_vm_close
,
520 static int dax_dev_mmap(struct file
*filp
, struct vm_area_struct
*vma
)
522 struct dax_dev
*dax_dev
= filp
->private_data
;
525 dev_dbg(dax_dev
->dev
, "%s\n", __func__
);
527 rc
= check_vma(dax_dev
, vma
, __func__
);
531 kref_get(&dax_dev
->kref
);
532 vma
->vm_ops
= &dax_dev_vm_ops
;
533 vma
->vm_flags
|= VM_MIXEDMAP
| VM_HUGEPAGE
;
538 static const struct file_operations dax_fops
= {
539 .llseek
= noop_llseek
,
540 .owner
= THIS_MODULE
,
541 .open
= dax_dev_open
,
542 .release
= dax_dev_release
,
543 .get_unmapped_area
= dax_dev_get_unmapped_area
,
544 .mmap
= dax_dev_mmap
,
547 static int __init
dax_init(void)
551 rc
= register_chrdev(0, "dax", &dax_fops
);
556 dax_class
= class_create(THIS_MODULE
, "dax");
557 if (IS_ERR(dax_class
)) {
558 unregister_chrdev(dax_major
, "dax");
559 return PTR_ERR(dax_class
);
565 static void __exit
dax_exit(void)
567 class_destroy(dax_class
);
568 unregister_chrdev(dax_major
, "dax");
569 ida_destroy(&dax_minor_ida
);
572 MODULE_AUTHOR("Intel Corporation");
573 MODULE_LICENSE("GPL v2");
574 subsys_initcall(dax_init
);
575 module_exit(dax_exit
);