2 * Copyright(c) 2016 - 2017 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/cdev.h>
18 #include <linux/slab.h>
19 #include <linux/dax.h>
22 #include <linux/mman.h>
23 #include "dax-private.h"
26 static struct class *dax_class
;
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.
33 static ssize_t
id_show(struct device
*dev
,
34 struct device_attribute
*attr
, char *buf
)
36 struct dax_region
*dax_region
= dev_get_drvdata(dev
);
38 return sprintf(buf
, "%d\n", dax_region
->id
);
40 static DEVICE_ATTR_RO(id
);
42 static ssize_t
region_size_show(struct device
*dev
,
43 struct device_attribute
*attr
, char *buf
)
45 struct dax_region
*dax_region
= dev_get_drvdata(dev
);
47 return sprintf(buf
, "%llu\n", (unsigned long long)
48 resource_size(&dax_region
->res
));
50 static struct device_attribute dev_attr_region_size
= __ATTR(size
, 0444,
51 region_size_show
, NULL
);
53 static ssize_t
align_show(struct device
*dev
,
54 struct device_attribute
*attr
, char *buf
)
56 struct dax_region
*dax_region
= dev_get_drvdata(dev
);
58 return sprintf(buf
, "%u\n", dax_region
->align
);
60 static DEVICE_ATTR_RO(align
);
62 static struct attribute
*dax_region_attributes
[] = {
63 &dev_attr_region_size
.attr
,
69 static const struct attribute_group dax_region_attribute_group
= {
71 .attrs
= dax_region_attributes
,
74 static const struct attribute_group
*dax_region_attribute_groups
[] = {
75 &dax_region_attribute_group
,
79 static void dax_region_free(struct kref
*kref
)
81 struct dax_region
*dax_region
;
83 dax_region
= container_of(kref
, struct dax_region
, kref
);
87 void dax_region_put(struct dax_region
*dax_region
)
89 kref_put(&dax_region
->kref
, dax_region_free
);
91 EXPORT_SYMBOL_GPL(dax_region_put
);
93 static void dax_region_unregister(void *region
)
95 struct dax_region
*dax_region
= region
;
97 sysfs_remove_groups(&dax_region
->dev
->kobj
,
98 dax_region_attribute_groups
);
99 dax_region_put(dax_region
);
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
)
106 struct dax_region
*dax_region
;
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.
113 if (dev_get_drvdata(parent
)) {
114 dev_WARN(parent
, "dax core failed to setup private data\n");
118 if (!IS_ALIGNED(res
->start
, align
)
119 || !IS_ALIGNED(resource_size(res
), align
))
122 dax_region
= kzalloc(sizeof(*dax_region
), GFP_KERNEL
);
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
)) {
140 kref_get(&dax_region
->kref
);
141 if (devm_add_action_or_reset(parent
, dax_region_unregister
, dax_region
))
145 EXPORT_SYMBOL_GPL(alloc_dax_region
);
147 static struct dev_dax
*to_dev_dax(struct device
*dev
)
149 return container_of(dev
, struct dev_dax
, dev
);
152 static ssize_t
size_show(struct device
*dev
,
153 struct device_attribute
*attr
, char *buf
)
155 struct dev_dax
*dev_dax
= to_dev_dax(dev
);
156 unsigned long long size
= 0;
159 for (i
= 0; i
< dev_dax
->num_resources
; i
++)
160 size
+= resource_size(&dev_dax
->res
[i
]);
162 return sprintf(buf
, "%llu\n", size
);
164 static DEVICE_ATTR_RO(size
);
166 static struct attribute
*dev_dax_attributes
[] = {
171 static const struct attribute_group dev_dax_attribute_group
= {
172 .attrs
= dev_dax_attributes
,
175 static const struct attribute_group
*dax_attribute_groups
[] = {
176 &dev_dax_attribute_group
,
180 static int check_vma(struct dev_dax
*dev_dax
, struct vm_area_struct
*vma
,
183 struct dax_region
*dax_region
= dev_dax
->region
;
184 struct device
*dev
= &dev_dax
->dev
;
187 if (!dax_alive(dev_dax
->dax_dev
))
190 /* prevent private mappings from being established */
191 if ((vma
->vm_flags
& VM_MAYSHARE
) != VM_MAYSHARE
) {
192 dev_info(dev
, "%s: %s: fail, attempted private mapping\n",
193 current
->comm
, func
);
197 mask
= dax_region
->align
- 1;
198 if (vma
->vm_start
& mask
|| vma
->vm_end
& mask
) {
199 dev_info(dev
, "%s: %s: fail, unaligned vma (%#lx - %#lx, %#lx)\n",
200 current
->comm
, func
, vma
->vm_start
, vma
->vm_end
,
205 if ((dax_region
->pfn_flags
& (PFN_DEV
|PFN_MAP
)) == PFN_DEV
206 && (vma
->vm_flags
& VM_DONTCOPY
) == 0) {
207 dev_info(dev
, "%s: %s: fail, dax range requires MADV_DONTFORK\n",
208 current
->comm
, func
);
212 if (!vma_is_dax(vma
)) {
213 dev_info(dev
, "%s: %s: fail, vma is not DAX capable\n",
214 current
->comm
, func
);
221 /* see "strong" declaration in tools/testing/nvdimm/dax-dev.c */
222 __weak phys_addr_t
dax_pgoff_to_phys(struct dev_dax
*dev_dax
, pgoff_t pgoff
,
225 struct resource
*res
;
226 /* gcc-4.6.3-nolibc for i386 complains that this is uninitialized */
227 phys_addr_t
uninitialized_var(phys
);
230 for (i
= 0; i
< dev_dax
->num_resources
; i
++) {
231 res
= &dev_dax
->res
[i
];
232 phys
= pgoff
* PAGE_SIZE
+ res
->start
;
233 if (phys
>= res
->start
&& phys
<= res
->end
)
235 pgoff
-= PHYS_PFN(resource_size(res
));
238 if (i
< dev_dax
->num_resources
) {
239 res
= &dev_dax
->res
[i
];
240 if (phys
+ size
- 1 <= res
->end
)
247 static int __dev_dax_pte_fault(struct dev_dax
*dev_dax
, struct vm_fault
*vmf
)
249 struct device
*dev
= &dev_dax
->dev
;
250 struct dax_region
*dax_region
;
251 int rc
= VM_FAULT_SIGBUS
;
254 unsigned int fault_size
= PAGE_SIZE
;
256 if (check_vma(dev_dax
, vmf
->vma
, __func__
))
257 return VM_FAULT_SIGBUS
;
259 dax_region
= dev_dax
->region
;
260 if (dax_region
->align
> PAGE_SIZE
) {
261 dev_dbg(dev
, "%s: alignment (%#x) > fault size (%#x)\n",
262 __func__
, dax_region
->align
, fault_size
);
263 return VM_FAULT_SIGBUS
;
266 if (fault_size
!= dax_region
->align
)
267 return VM_FAULT_SIGBUS
;
269 phys
= dax_pgoff_to_phys(dev_dax
, vmf
->pgoff
, PAGE_SIZE
);
271 dev_dbg(dev
, "%s: pgoff_to_phys(%#lx) failed\n", __func__
,
273 return VM_FAULT_SIGBUS
;
276 pfn
= phys_to_pfn_t(phys
, dax_region
->pfn_flags
);
278 rc
= vm_insert_mixed(vmf
->vma
, vmf
->address
, pfn
);
282 if (rc
< 0 && rc
!= -EBUSY
)
283 return VM_FAULT_SIGBUS
;
285 return VM_FAULT_NOPAGE
;
288 static int __dev_dax_pmd_fault(struct dev_dax
*dev_dax
, struct vm_fault
*vmf
)
290 unsigned long pmd_addr
= vmf
->address
& PMD_MASK
;
291 struct device
*dev
= &dev_dax
->dev
;
292 struct dax_region
*dax_region
;
296 unsigned int fault_size
= PMD_SIZE
;
298 if (check_vma(dev_dax
, vmf
->vma
, __func__
))
299 return VM_FAULT_SIGBUS
;
301 dax_region
= dev_dax
->region
;
302 if (dax_region
->align
> PMD_SIZE
) {
303 dev_dbg(dev
, "%s: alignment (%#x) > fault size (%#x)\n",
304 __func__
, dax_region
->align
, fault_size
);
305 return VM_FAULT_SIGBUS
;
308 /* dax pmd mappings require pfn_t_devmap() */
309 if ((dax_region
->pfn_flags
& (PFN_DEV
|PFN_MAP
)) != (PFN_DEV
|PFN_MAP
)) {
310 dev_dbg(dev
, "%s: region lacks devmap flags\n", __func__
);
311 return VM_FAULT_SIGBUS
;
314 if (fault_size
< dax_region
->align
)
315 return VM_FAULT_SIGBUS
;
316 else if (fault_size
> dax_region
->align
)
317 return VM_FAULT_FALLBACK
;
319 /* if we are outside of the VMA */
320 if (pmd_addr
< vmf
->vma
->vm_start
||
321 (pmd_addr
+ PMD_SIZE
) > vmf
->vma
->vm_end
)
322 return VM_FAULT_SIGBUS
;
324 pgoff
= linear_page_index(vmf
->vma
, pmd_addr
);
325 phys
= dax_pgoff_to_phys(dev_dax
, pgoff
, PMD_SIZE
);
327 dev_dbg(dev
, "%s: pgoff_to_phys(%#lx) failed\n", __func__
,
329 return VM_FAULT_SIGBUS
;
332 pfn
= phys_to_pfn_t(phys
, dax_region
->pfn_flags
);
334 return vmf_insert_pfn_pmd(vmf
->vma
, vmf
->address
, vmf
->pmd
, pfn
,
335 vmf
->flags
& FAULT_FLAG_WRITE
);
338 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
339 static int __dev_dax_pud_fault(struct dev_dax
*dev_dax
, struct vm_fault
*vmf
)
341 unsigned long pud_addr
= vmf
->address
& PUD_MASK
;
342 struct device
*dev
= &dev_dax
->dev
;
343 struct dax_region
*dax_region
;
347 unsigned int fault_size
= PUD_SIZE
;
350 if (check_vma(dev_dax
, vmf
->vma
, __func__
))
351 return VM_FAULT_SIGBUS
;
353 dax_region
= dev_dax
->region
;
354 if (dax_region
->align
> PUD_SIZE
) {
355 dev_dbg(dev
, "%s: alignment (%#x) > fault size (%#x)\n",
356 __func__
, dax_region
->align
, fault_size
);
357 return VM_FAULT_SIGBUS
;
360 /* dax pud mappings require pfn_t_devmap() */
361 if ((dax_region
->pfn_flags
& (PFN_DEV
|PFN_MAP
)) != (PFN_DEV
|PFN_MAP
)) {
362 dev_dbg(dev
, "%s: region lacks devmap flags\n", __func__
);
363 return VM_FAULT_SIGBUS
;
366 if (fault_size
< dax_region
->align
)
367 return VM_FAULT_SIGBUS
;
368 else if (fault_size
> dax_region
->align
)
369 return VM_FAULT_FALLBACK
;
371 /* if we are outside of the VMA */
372 if (pud_addr
< vmf
->vma
->vm_start
||
373 (pud_addr
+ PUD_SIZE
) > vmf
->vma
->vm_end
)
374 return VM_FAULT_SIGBUS
;
376 pgoff
= linear_page_index(vmf
->vma
, pud_addr
);
377 phys
= dax_pgoff_to_phys(dev_dax
, pgoff
, PUD_SIZE
);
379 dev_dbg(dev
, "%s: pgoff_to_phys(%#lx) failed\n", __func__
,
381 return VM_FAULT_SIGBUS
;
384 pfn
= phys_to_pfn_t(phys
, dax_region
->pfn_flags
);
386 return vmf_insert_pfn_pud(vmf
->vma
, vmf
->address
, vmf
->pud
, pfn
,
387 vmf
->flags
& FAULT_FLAG_WRITE
);
390 static int __dev_dax_pud_fault(struct dev_dax
*dev_dax
, struct vm_fault
*vmf
)
392 return VM_FAULT_FALLBACK
;
394 #endif /* !CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
396 static int dev_dax_huge_fault(struct vm_fault
*vmf
,
397 enum page_entry_size pe_size
)
400 struct file
*filp
= vmf
->vma
->vm_file
;
401 struct dev_dax
*dev_dax
= filp
->private_data
;
403 dev_dbg(&dev_dax
->dev
, "%s: %s: %s (%#lx - %#lx) size = %d\n", __func__
,
404 current
->comm
, (vmf
->flags
& FAULT_FLAG_WRITE
)
406 vmf
->vma
->vm_start
, vmf
->vma
->vm_end
, pe_size
);
408 id
= dax_read_lock();
411 rc
= __dev_dax_pte_fault(dev_dax
, vmf
);
414 rc
= __dev_dax_pmd_fault(dev_dax
, vmf
);
417 rc
= __dev_dax_pud_fault(dev_dax
, vmf
);
420 rc
= VM_FAULT_SIGBUS
;
427 static int dev_dax_fault(struct vm_fault
*vmf
)
429 return dev_dax_huge_fault(vmf
, PE_SIZE_PTE
);
432 static int dev_dax_split(struct vm_area_struct
*vma
, unsigned long addr
)
434 struct file
*filp
= vma
->vm_file
;
435 struct dev_dax
*dev_dax
= filp
->private_data
;
436 struct dax_region
*dax_region
= dev_dax
->region
;
438 if (!IS_ALIGNED(addr
, dax_region
->align
))
443 static const struct vm_operations_struct dax_vm_ops
= {
444 .fault
= dev_dax_fault
,
445 .huge_fault
= dev_dax_huge_fault
,
446 .split
= dev_dax_split
,
449 static int dax_mmap(struct file
*filp
, struct vm_area_struct
*vma
)
451 struct dev_dax
*dev_dax
= filp
->private_data
;
454 dev_dbg(&dev_dax
->dev
, "%s\n", __func__
);
457 * We lock to check dax_dev liveness and will re-check at
460 id
= dax_read_lock();
461 rc
= check_vma(dev_dax
, vma
, __func__
);
466 vma
->vm_ops
= &dax_vm_ops
;
467 vma
->vm_flags
|= VM_MIXEDMAP
| VM_HUGEPAGE
;
471 /* return an unmapped area aligned to the dax region specified alignment */
472 static unsigned long dax_get_unmapped_area(struct file
*filp
,
473 unsigned long addr
, unsigned long len
, unsigned long pgoff
,
476 unsigned long off
, off_end
, off_align
, len_align
, addr_align
, align
;
477 struct dev_dax
*dev_dax
= filp
? filp
->private_data
: NULL
;
478 struct dax_region
*dax_region
;
480 if (!dev_dax
|| addr
)
483 dax_region
= dev_dax
->region
;
484 align
= dax_region
->align
;
485 off
= pgoff
<< PAGE_SHIFT
;
487 off_align
= round_up(off
, align
);
489 if ((off_end
<= off_align
) || ((off_end
- off_align
) < align
))
492 len_align
= len
+ align
;
493 if ((off
+ len_align
) < off
)
496 addr_align
= current
->mm
->get_unmapped_area(filp
, addr
, len_align
,
498 if (!IS_ERR_VALUE(addr_align
)) {
499 addr_align
+= (off
- addr_align
) & (align
- 1);
503 return current
->mm
->get_unmapped_area(filp
, addr
, len
, pgoff
, flags
);
506 static int dax_open(struct inode
*inode
, struct file
*filp
)
508 struct dax_device
*dax_dev
= inode_dax(inode
);
509 struct inode
*__dax_inode
= dax_inode(dax_dev
);
510 struct dev_dax
*dev_dax
= dax_get_private(dax_dev
);
512 dev_dbg(&dev_dax
->dev
, "%s\n", __func__
);
513 inode
->i_mapping
= __dax_inode
->i_mapping
;
514 inode
->i_mapping
->host
= __dax_inode
;
515 filp
->f_mapping
= inode
->i_mapping
;
516 filp
->f_wb_err
= filemap_sample_wb_err(filp
->f_mapping
);
517 filp
->private_data
= dev_dax
;
518 inode
->i_flags
= S_DAX
;
523 static int dax_release(struct inode
*inode
, struct file
*filp
)
525 struct dev_dax
*dev_dax
= filp
->private_data
;
527 dev_dbg(&dev_dax
->dev
, "%s\n", __func__
);
531 static const struct file_operations dax_fops
= {
532 .llseek
= noop_llseek
,
533 .owner
= THIS_MODULE
,
535 .release
= dax_release
,
536 .get_unmapped_area
= dax_get_unmapped_area
,
538 .mmap_supported_flags
= MAP_SYNC
,
541 static void dev_dax_release(struct device
*dev
)
543 struct dev_dax
*dev_dax
= to_dev_dax(dev
);
544 struct dax_region
*dax_region
= dev_dax
->region
;
545 struct dax_device
*dax_dev
= dev_dax
->dax_dev
;
547 if (dev_dax
->id
>= 0)
548 ida_simple_remove(&dax_region
->ida
, dev_dax
->id
);
549 dax_region_put(dax_region
);
554 static void kill_dev_dax(struct dev_dax
*dev_dax
)
556 struct dax_device
*dax_dev
= dev_dax
->dax_dev
;
557 struct inode
*inode
= dax_inode(dax_dev
);
560 unmap_mapping_range(inode
->i_mapping
, 0, 0, 1);
563 static void unregister_dev_dax(void *dev
)
565 struct dev_dax
*dev_dax
= to_dev_dax(dev
);
566 struct dax_device
*dax_dev
= dev_dax
->dax_dev
;
567 struct inode
*inode
= dax_inode(dax_dev
);
568 struct cdev
*cdev
= inode
->i_cdev
;
570 dev_dbg(dev
, "%s\n", __func__
);
572 kill_dev_dax(dev_dax
);
573 cdev_device_del(cdev
, dev
);
577 struct dev_dax
*devm_create_dev_dax(struct dax_region
*dax_region
,
578 int id
, struct resource
*res
, int count
)
580 struct device
*parent
= dax_region
->dev
;
581 struct dax_device
*dax_dev
;
582 struct dev_dax
*dev_dax
;
589 return ERR_PTR(-EINVAL
);
591 dev_dax
= kzalloc(sizeof(*dev_dax
) + sizeof(*res
) * count
, GFP_KERNEL
);
593 return ERR_PTR(-ENOMEM
);
595 for (i
= 0; i
< count
; i
++) {
596 if (!IS_ALIGNED(res
[i
].start
, dax_region
->align
)
597 || !IS_ALIGNED(resource_size(&res
[i
]),
598 dax_region
->align
)) {
602 dev_dax
->res
[i
].start
= res
[i
].start
;
603 dev_dax
->res
[i
].end
= res
[i
].end
;
610 id
= ida_simple_get(&dax_region
->ida
, 0, 0, GFP_KERNEL
);
617 /* region provider owns @id lifetime */
622 * No 'host' or dax_operations since there is no access to this
623 * device outside of mmap of the resulting character device.
625 dax_dev
= alloc_dax(dev_dax
, NULL
, NULL
);
631 /* from here on we're committed to teardown via dax_dev_release() */
633 device_initialize(dev
);
635 inode
= dax_inode(dax_dev
);
636 cdev
= inode
->i_cdev
;
637 cdev_init(cdev
, &dax_fops
);
638 cdev
->owner
= parent
->driver
->owner
;
640 dev_dax
->num_resources
= count
;
641 dev_dax
->dax_dev
= dax_dev
;
642 dev_dax
->region
= dax_region
;
643 kref_get(&dax_region
->kref
);
645 dev
->devt
= inode
->i_rdev
;
646 dev
->class = dax_class
;
647 dev
->parent
= parent
;
648 dev
->groups
= dax_attribute_groups
;
649 dev
->release
= dev_dax_release
;
650 dev_set_name(dev
, "dax%d.%d", dax_region
->id
, id
);
652 rc
= cdev_device_add(cdev
, dev
);
654 kill_dev_dax(dev_dax
);
659 rc
= devm_add_action_or_reset(dax_region
->dev
, unregister_dev_dax
, dev
);
666 if (dev_dax
->id
>= 0)
667 ida_simple_remove(&dax_region
->ida
, dev_dax
->id
);
673 EXPORT_SYMBOL_GPL(devm_create_dev_dax
);
675 static int __init
dax_init(void)
677 dax_class
= class_create(THIS_MODULE
, "dax");
678 return PTR_ERR_OR_ZERO(dax_class
);
681 static void __exit
dax_exit(void)
683 class_destroy(dax_class
);
686 MODULE_AUTHOR("Intel Corporation");
687 MODULE_LICENSE("GPL v2");
688 subsys_initcall(dax_init
);
689 module_exit(dax_exit
);