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/mount.h>
17 #include <linux/pfn_t.h>
18 #include <linux/hash.h>
19 #include <linux/cdev.h>
20 #include <linux/slab.h>
21 #include <linux/dax.h>
26 static dev_t dax_devt;
27 static struct class *dax_class;
28 static DEFINE_IDA(dax_minor_ida);
29 static int nr_dax = CONFIG_NR_DEV_DAX;
30 module_param(nr_dax, int, S_IRUGO);
31 static struct vfsmount *dax_mnt;
32 static struct kmem_cache *dax_cache __read_mostly;
33 static struct super_block *dax_superblock __read_mostly;
34 MODULE_PARM_DESC(nr_dax, "max number of device-dax instances");
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
54 unsigned long pfn_flags;
58 * struct dax_dev - subdivision of a dax region
59 * @region - parent region
60 * @dev - device backing the character device
61 * @cdev - core chardev data
62 * @alive - !alive + rcu grace period == no new mappings can be established
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
68 struct dax_region *region;
75 struct resource res[0];
78 static struct inode *dax_alloc_inode(struct super_block *sb)
80 return kmem_cache_alloc(dax_cache, GFP_KERNEL);
83 static void dax_i_callback(struct rcu_head *head)
85 struct inode *inode = container_of(head, struct inode, i_rcu);
87 kmem_cache_free(dax_cache, inode);
90 static void dax_destroy_inode(struct inode *inode)
92 call_rcu(&inode->i_rcu, dax_i_callback);
95 static 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,
102 static struct dentry *dax_mount(struct file_system_type *fs_type,
103 int flags, const char *dev_name, void *data)
105 return mount_pseudo(fs_type, "dax:", &dax_sops, NULL, DAXFS_MAGIC);
108 static struct file_system_type dax_type = {
111 .kill_sb = kill_anon_super,
114 static int dax_test(struct inode *inode, void *data)
116 return inode->i_cdev == data;
119 static int dax_set(struct inode *inode, void *data)
121 inode->i_cdev = data;
125 static struct inode *dax_inode_get(struct cdev *cdev, dev_t devt)
129 inode = iget5_locked(dax_superblock, hash_32(devt + DAXFS_MAGIC, 31),
130 dax_test, dax_set, cdev);
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);
145 static void init_once(void *inode)
147 inode_init_once(inode);
150 static int dax_inode_init(void)
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),
161 rc = register_filesystem(&dax_type);
163 goto err_register_fs;
165 dax_mnt = kern_mount(&dax_type);
166 if (IS_ERR(dax_mnt)) {
167 rc = PTR_ERR(dax_mnt);
170 dax_superblock = dax_mnt->mnt_sb;
175 unregister_filesystem(&dax_type);
177 kmem_cache_destroy(dax_cache);
182 static void dax_inode_exit(void)
184 kern_unmount(dax_mnt);
185 unregister_filesystem(&dax_type);
186 kmem_cache_destroy(dax_cache);
189 static void dax_region_free(struct kref *kref)
191 struct dax_region *dax_region;
193 dax_region = container_of(kref, struct dax_region, kref);
197 void dax_region_put(struct dax_region *dax_region)
199 kref_put(&dax_region->kref, dax_region_free);
201 EXPORT_SYMBOL_GPL(dax_region_put);
203 struct 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)
207 struct dax_region *dax_region;
209 if (!IS_ALIGNED(res->start, align)
210 || !IS_ALIGNED(resource_size(res), align))
213 dax_region = kzalloc(sizeof(*dax_region), GFP_KERNEL);
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;
228 EXPORT_SYMBOL_GPL(alloc_dax_region);
230 static struct dax_dev *to_dax_dev(struct device *dev)
232 return container_of(dev, struct dax_dev, dev);
235 static ssize_t size_show(struct device *dev,
236 struct device_attribute *attr, char *buf)
238 struct dax_dev *dax_dev = to_dax_dev(dev);
239 unsigned long long size = 0;
242 for (i = 0; i < dax_dev->num_resources; i++)
243 size += resource_size(&dax_dev->res[i]);
245 return sprintf(buf, "%llu\n", size);
247 static DEVICE_ATTR_RO(size);
249 static struct attribute *dax_device_attributes[] = {
254 static const struct attribute_group dax_device_attribute_group = {
255 .attrs = dax_device_attributes,
258 static const struct attribute_group *dax_attribute_groups[] = {
259 &dax_device_attribute_group,
263 static int check_vma(struct dax_dev *dax_dev, struct vm_area_struct *vma,
266 struct dax_region *dax_region = dax_dev->region;
267 struct device *dev = &dax_dev->dev;
273 /* prevent private mappings from being established */
274 if ((vma->vm_flags & VM_MAYSHARE) != VM_MAYSHARE) {
275 dev_info(dev, "%s: %s: fail, attempted private mapping\n",
276 current->comm, func);
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,
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);
295 if (!vma_is_dax(vma)) {
296 dev_info(dev, "%s: %s: fail, vma is not DAX capable\n",
297 current->comm, func);
304 static phys_addr_t pgoff_to_phys(struct dax_dev *dax_dev, pgoff_t pgoff,
307 struct resource *res;
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)
316 pgoff -= PHYS_PFN(resource_size(res));
319 if (i < dax_dev->num_resources) {
320 res = &dax_dev->res[i];
321 if (phys + size - 1 <= res->end)
328 static int __dax_dev_fault(struct dax_dev *dax_dev, struct vm_area_struct *vma,
329 struct vm_fault *vmf)
331 struct device *dev = &dax_dev->dev;
332 struct dax_region *dax_region;
333 int rc = VM_FAULT_SIGBUS;
337 if (check_vma(dax_dev, vma, __func__))
338 return VM_FAULT_SIGBUS;
340 dax_region = dax_dev->region;
341 if (dax_region->align > PAGE_SIZE) {
342 dev_dbg(dev, "%s: alignment > fault size\n", __func__);
343 return VM_FAULT_SIGBUS;
346 phys = pgoff_to_phys(dax_dev, vmf->pgoff, PAGE_SIZE);
348 dev_dbg(dev, "%s: phys_to_pgoff(%#lx) failed\n", __func__,
350 return VM_FAULT_SIGBUS;
353 pfn = phys_to_pfn_t(phys, dax_region->pfn_flags);
355 rc = vm_insert_mixed(vma, vmf->address, pfn);
359 if (rc < 0 && rc != -EBUSY)
360 return VM_FAULT_SIGBUS;
362 return VM_FAULT_NOPAGE;
365 static int dax_dev_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
368 struct file *filp = vma->vm_file;
369 struct dax_dev *dax_dev = filp->private_data;
371 dev_dbg(&dax_dev->dev, "%s: %s: %s (%#lx - %#lx)\n", __func__,
372 current->comm, (vmf->flags & FAULT_FLAG_WRITE)
373 ? "write" : "read", vma->vm_start, vma->vm_end);
375 rc = __dax_dev_fault(dax_dev, vma, vmf);
381 static int __dax_dev_pmd_fault(struct dax_dev *dax_dev,
382 struct vm_area_struct *vma, unsigned long addr, pmd_t *pmd,
385 unsigned long pmd_addr = addr & PMD_MASK;
386 struct device *dev = &dax_dev->dev;
387 struct dax_region *dax_region;
392 if (check_vma(dax_dev, vma, __func__))
393 return VM_FAULT_SIGBUS;
395 dax_region = dax_dev->region;
396 if (dax_region->align > PMD_SIZE) {
397 dev_dbg(dev, "%s: alignment > fault size\n", __func__);
398 return VM_FAULT_SIGBUS;
401 /* dax pmd mappings require pfn_t_devmap() */
402 if ((dax_region->pfn_flags & (PFN_DEV|PFN_MAP)) != (PFN_DEV|PFN_MAP)) {
403 dev_dbg(dev, "%s: alignment > fault size\n", __func__);
404 return VM_FAULT_SIGBUS;
407 pgoff = linear_page_index(vma, pmd_addr);
408 phys = pgoff_to_phys(dax_dev, pgoff, PMD_SIZE);
410 dev_dbg(dev, "%s: phys_to_pgoff(%#lx) failed\n", __func__,
412 return VM_FAULT_SIGBUS;
415 pfn = phys_to_pfn_t(phys, dax_region->pfn_flags);
417 return vmf_insert_pfn_pmd(vma, addr, pmd, pfn,
418 flags & FAULT_FLAG_WRITE);
421 static int dax_dev_pmd_fault(struct vm_area_struct *vma, unsigned long addr,
422 pmd_t *pmd, unsigned int flags)
425 struct file *filp = vma->vm_file;
426 struct dax_dev *dax_dev = filp->private_data;
428 dev_dbg(&dax_dev->dev, "%s: %s: %s (%#lx - %#lx)\n", __func__,
429 current->comm, (flags & FAULT_FLAG_WRITE)
430 ? "write" : "read", vma->vm_start, vma->vm_end);
433 rc = __dax_dev_pmd_fault(dax_dev, vma, addr, pmd, flags);
439 static const struct vm_operations_struct dax_dev_vm_ops = {
440 .fault = dax_dev_fault,
441 .pmd_fault = dax_dev_pmd_fault,
444 static int dax_mmap(struct file *filp, struct vm_area_struct *vma)
446 struct dax_dev *dax_dev = filp->private_data;
449 dev_dbg(&dax_dev->dev, "%s\n", __func__);
451 rc = check_vma(dax_dev, vma, __func__);
455 vma->vm_ops = &dax_dev_vm_ops;
456 vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
460 /* return an unmapped area aligned to the dax region specified alignment */
461 static unsigned long dax_get_unmapped_area(struct file *filp,
462 unsigned long addr, unsigned long len, unsigned long pgoff,
465 unsigned long off, off_end, off_align, len_align, addr_align, align;
466 struct dax_dev *dax_dev = filp ? filp->private_data : NULL;
467 struct dax_region *dax_region;
469 if (!dax_dev || addr)
472 dax_region = dax_dev->region;
473 align = dax_region->align;
474 off = pgoff << PAGE_SHIFT;
476 off_align = round_up(off, align);
478 if ((off_end <= off_align) || ((off_end - off_align) < align))
481 len_align = len + align;
482 if ((off + len_align) < off)
485 addr_align = current->mm->get_unmapped_area(filp, addr, len_align,
487 if (!IS_ERR_VALUE(addr_align)) {
488 addr_align += (off - addr_align) & (align - 1);
492 return current->mm->get_unmapped_area(filp, addr, len, pgoff, flags);
495 static int dax_open(struct inode *inode, struct file *filp)
497 struct dax_dev *dax_dev;
499 dax_dev = container_of(inode->i_cdev, struct dax_dev, cdev);
500 dev_dbg(&dax_dev->dev, "%s\n", __func__);
501 inode->i_mapping = dax_dev->inode->i_mapping;
502 inode->i_mapping->host = dax_dev->inode;
503 filp->f_mapping = inode->i_mapping;
504 filp->private_data = dax_dev;
505 inode->i_flags = S_DAX;
510 static int dax_release(struct inode *inode, struct file *filp)
512 struct dax_dev *dax_dev = filp->private_data;
514 dev_dbg(&dax_dev->dev, "%s\n", __func__);
518 static const struct file_operations dax_fops = {
519 .llseek = noop_llseek,
520 .owner = THIS_MODULE,
522 .release = dax_release,
523 .get_unmapped_area = dax_get_unmapped_area,
527 static void dax_dev_release(struct device *dev)
529 struct dax_dev *dax_dev = to_dax_dev(dev);
530 struct dax_region *dax_region = dax_dev->region;
532 ida_simple_remove(&dax_region->ida, dax_dev->id);
533 ida_simple_remove(&dax_minor_ida, MINOR(dev->devt));
534 dax_region_put(dax_region);
535 iput(dax_dev->inode);
539 static void unregister_dax_dev(void *dev)
541 struct dax_dev *dax_dev = to_dax_dev(dev);
542 struct cdev *cdev = &dax_dev->cdev;
544 dev_dbg(dev, "%s\n", __func__);
547 * Note, rcu is not protecting the liveness of dax_dev, rcu is
548 * ensuring that any fault handlers that might have seen
549 * dax_dev->alive == true, have completed. Any fault handlers
550 * that start after synchronize_rcu() has started will abort
551 * upon seeing dax_dev->alive == false.
553 dax_dev->alive = false;
555 unmap_mapping_range(dax_dev->inode->i_mapping, 0, 0, 1);
557 device_unregister(dev);
560 struct dax_dev *devm_create_dax_dev(struct dax_region *dax_region,
561 struct resource *res, int count)
563 struct device *parent = dax_region->dev;
564 struct dax_dev *dax_dev;
565 int rc = 0, minor, i;
570 dax_dev = kzalloc(sizeof(*dax_dev) + sizeof(*res) * count, GFP_KERNEL);
572 return ERR_PTR(-ENOMEM);
574 for (i = 0; i < count; i++) {
575 if (!IS_ALIGNED(res[i].start, dax_region->align)
576 || !IS_ALIGNED(resource_size(&res[i]),
577 dax_region->align)) {
581 dax_dev->res[i].start = res[i].start;
582 dax_dev->res[i].end = res[i].end;
588 dax_dev->id = ida_simple_get(&dax_region->ida, 0, 0, GFP_KERNEL);
589 if (dax_dev->id < 0) {
594 minor = ida_simple_get(&dax_minor_ida, 0, 0, GFP_KERNEL);
600 dev_t = MKDEV(MAJOR(dax_devt), minor);
602 dax_dev->inode = dax_inode_get(&dax_dev->cdev, dev_t);
603 if (!dax_dev->inode) {
608 /* device_initialize() so cdev can reference kobj parent */
609 device_initialize(dev);
611 cdev = &dax_dev->cdev;
612 cdev_init(cdev, &dax_fops);
613 cdev->owner = parent->driver->owner;
614 cdev->kobj.parent = &dev->kobj;
615 rc = cdev_add(&dax_dev->cdev, dev_t, 1);
619 /* from here on we're committed to teardown via dax_dev_release() */
620 dax_dev->num_resources = count;
621 dax_dev->alive = true;
622 dax_dev->region = dax_region;
623 kref_get(&dax_region->kref);
626 dev->class = dax_class;
627 dev->parent = parent;
628 dev->groups = dax_attribute_groups;
629 dev->release = dax_dev_release;
630 dev_set_name(dev, "dax%d.%d", dax_region->id, dax_dev->id);
631 rc = device_add(dev);
637 rc = devm_add_action_or_reset(dax_region->dev, unregister_dax_dev, dev);
644 iput(dax_dev->inode);
646 ida_simple_remove(&dax_minor_ida, minor);
648 ida_simple_remove(&dax_region->ida, dax_dev->id);
654 EXPORT_SYMBOL_GPL(devm_create_dax_dev);
656 static int __init dax_init(void)
660 rc = dax_inode_init();
664 nr_dax = max(nr_dax, 256);
665 rc = alloc_chrdev_region(&dax_devt, 0, nr_dax, "dax");
669 dax_class = class_create(THIS_MODULE, "dax");
670 if (IS_ERR(dax_class)) {
671 rc = PTR_ERR(dax_class);
678 unregister_chrdev_region(dax_devt, nr_dax);
684 static void __exit dax_exit(void)
686 class_destroy(dax_class);
687 unregister_chrdev_region(dax_devt, nr_dax);
688 ida_destroy(&dax_minor_ida);
692 MODULE_AUTHOR("Intel Corporation");
693 MODULE_LICENSE("GPL v2");
694 subsys_initcall(dax_init);
695 module_exit(dax_exit);