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 "dax-private.h"
25 static struct class *dax_class;
28 * Rely on the fact that drvdata is set before the attributes are
29 * registered, and that the attributes are unregistered before drvdata
30 * is cleared to assume that drvdata is always valid.
32 static ssize_t id_show(struct device *dev,
33 struct device_attribute *attr, char *buf)
35 struct dax_region *dax_region = dev_get_drvdata(dev);
37 return sprintf(buf, "%d\n", dax_region->id);
39 static DEVICE_ATTR_RO(id);
41 static ssize_t region_size_show(struct device *dev,
42 struct device_attribute *attr, char *buf)
44 struct dax_region *dax_region = dev_get_drvdata(dev);
46 return sprintf(buf, "%llu\n", (unsigned long long)
47 resource_size(&dax_region->res));
49 static struct device_attribute dev_attr_region_size = __ATTR(size, 0444,
50 region_size_show, NULL);
52 static ssize_t align_show(struct device *dev,
53 struct device_attribute *attr, char *buf)
55 struct dax_region *dax_region = dev_get_drvdata(dev);
57 return sprintf(buf, "%u\n", dax_region->align);
59 static DEVICE_ATTR_RO(align);
61 static struct attribute *dax_region_attributes[] = {
62 &dev_attr_region_size.attr,
68 static const struct attribute_group dax_region_attribute_group = {
70 .attrs = dax_region_attributes,
73 static const struct attribute_group *dax_region_attribute_groups[] = {
74 &dax_region_attribute_group,
78 static void dax_region_free(struct kref *kref)
80 struct dax_region *dax_region;
82 dax_region = container_of(kref, struct dax_region, kref);
86 void dax_region_put(struct dax_region *dax_region)
88 kref_put(&dax_region->kref, dax_region_free);
90 EXPORT_SYMBOL_GPL(dax_region_put);
92 static void dax_region_unregister(void *region)
94 struct dax_region *dax_region = region;
96 sysfs_remove_groups(&dax_region->dev->kobj,
97 dax_region_attribute_groups);
98 dax_region_put(dax_region);
101 struct dax_region *alloc_dax_region(struct device *parent, int region_id,
102 struct resource *res, unsigned int align, void *addr,
103 unsigned long pfn_flags)
105 struct dax_region *dax_region;
108 * The DAX core assumes that it can store its private data in
109 * parent->driver_data. This WARN is a reminder / safeguard for
110 * developers of device-dax drivers.
112 if (dev_get_drvdata(parent)) {
113 dev_WARN(parent, "dax core failed to setup private data\n");
117 if (!IS_ALIGNED(res->start, align)
118 || !IS_ALIGNED(resource_size(res), align))
121 dax_region = kzalloc(sizeof(*dax_region), GFP_KERNEL);
125 dev_set_drvdata(parent, dax_region);
126 memcpy(&dax_region->res, res, sizeof(*res));
127 dax_region->pfn_flags = pfn_flags;
128 kref_init(&dax_region->kref);
129 dax_region->id = region_id;
130 ida_init(&dax_region->ida);
131 dax_region->align = align;
132 dax_region->dev = parent;
133 dax_region->base = addr;
134 if (sysfs_create_groups(&parent->kobj, dax_region_attribute_groups)) {
139 kref_get(&dax_region->kref);
140 if (devm_add_action_or_reset(parent, dax_region_unregister, dax_region))
144 EXPORT_SYMBOL_GPL(alloc_dax_region);
146 static struct dev_dax *to_dev_dax(struct device *dev)
148 return container_of(dev, struct dev_dax, dev);
151 static ssize_t size_show(struct device *dev,
152 struct device_attribute *attr, char *buf)
154 struct dev_dax *dev_dax = to_dev_dax(dev);
155 unsigned long long size = 0;
158 for (i = 0; i < dev_dax->num_resources; i++)
159 size += resource_size(&dev_dax->res[i]);
161 return sprintf(buf, "%llu\n", size);
163 static DEVICE_ATTR_RO(size);
165 static struct attribute *dev_dax_attributes[] = {
170 static const struct attribute_group dev_dax_attribute_group = {
171 .attrs = dev_dax_attributes,
174 static const struct attribute_group *dax_attribute_groups[] = {
175 &dev_dax_attribute_group,
179 static int check_vma(struct dev_dax *dev_dax, struct vm_area_struct *vma,
182 struct dax_region *dax_region = dev_dax->region;
183 struct device *dev = &dev_dax->dev;
186 if (!dax_alive(dev_dax->dax_dev))
189 /* prevent private mappings from being established */
190 if ((vma->vm_flags & VM_MAYSHARE) != VM_MAYSHARE) {
191 dev_info(dev, "%s: %s: fail, attempted private mapping\n",
192 current->comm, func);
196 mask = dax_region->align - 1;
197 if (vma->vm_start & mask || vma->vm_end & mask) {
198 dev_info(dev, "%s: %s: fail, unaligned vma (%#lx - %#lx, %#lx)\n",
199 current->comm, func, vma->vm_start, vma->vm_end,
204 if ((dax_region->pfn_flags & (PFN_DEV|PFN_MAP)) == PFN_DEV
205 && (vma->vm_flags & VM_DONTCOPY) == 0) {
206 dev_info(dev, "%s: %s: fail, dax range requires MADV_DONTFORK\n",
207 current->comm, func);
211 if (!vma_is_dax(vma)) {
212 dev_info(dev, "%s: %s: fail, vma is not DAX capable\n",
213 current->comm, func);
220 /* see "strong" declaration in tools/testing/nvdimm/dax-dev.c */
221 __weak phys_addr_t dax_pgoff_to_phys(struct dev_dax *dev_dax, pgoff_t pgoff,
224 struct resource *res;
225 /* gcc-4.6.3-nolibc for i386 complains that this is uninitialized */
226 phys_addr_t uninitialized_var(phys);
229 for (i = 0; i < dev_dax->num_resources; i++) {
230 res = &dev_dax->res[i];
231 phys = pgoff * PAGE_SIZE + res->start;
232 if (phys >= res->start && phys <= res->end)
234 pgoff -= PHYS_PFN(resource_size(res));
237 if (i < dev_dax->num_resources) {
238 res = &dev_dax->res[i];
239 if (phys + size - 1 <= res->end)
246 static int __dev_dax_pte_fault(struct dev_dax *dev_dax, struct vm_fault *vmf)
248 struct device *dev = &dev_dax->dev;
249 struct dax_region *dax_region;
250 int rc = VM_FAULT_SIGBUS;
253 unsigned int fault_size = PAGE_SIZE;
255 if (check_vma(dev_dax, vmf->vma, __func__))
256 return VM_FAULT_SIGBUS;
258 dax_region = dev_dax->region;
259 if (dax_region->align > PAGE_SIZE) {
260 dev_dbg(dev, "alignment (%#x) > fault size (%#x)\n",
261 dax_region->align, fault_size);
262 return VM_FAULT_SIGBUS;
265 if (fault_size != dax_region->align)
266 return VM_FAULT_SIGBUS;
268 phys = dax_pgoff_to_phys(dev_dax, vmf->pgoff, PAGE_SIZE);
270 dev_dbg(dev, "pgoff_to_phys(%#lx) failed\n", vmf->pgoff);
271 return VM_FAULT_SIGBUS;
274 pfn = phys_to_pfn_t(phys, dax_region->pfn_flags);
276 rc = vm_insert_mixed(vmf->vma, vmf->address, pfn);
280 if (rc < 0 && rc != -EBUSY)
281 return VM_FAULT_SIGBUS;
283 return VM_FAULT_NOPAGE;
286 static int __dev_dax_pmd_fault(struct dev_dax *dev_dax, struct vm_fault *vmf)
288 unsigned long pmd_addr = vmf->address & PMD_MASK;
289 struct device *dev = &dev_dax->dev;
290 struct dax_region *dax_region;
294 unsigned int fault_size = PMD_SIZE;
296 if (check_vma(dev_dax, vmf->vma, __func__))
297 return VM_FAULT_SIGBUS;
299 dax_region = dev_dax->region;
300 if (dax_region->align > PMD_SIZE) {
301 dev_dbg(dev, "alignment (%#x) > fault size (%#x)\n",
302 dax_region->align, fault_size);
303 return VM_FAULT_SIGBUS;
306 /* dax pmd mappings require pfn_t_devmap() */
307 if ((dax_region->pfn_flags & (PFN_DEV|PFN_MAP)) != (PFN_DEV|PFN_MAP)) {
308 dev_dbg(dev, "region lacks devmap flags\n");
309 return VM_FAULT_SIGBUS;
312 if (fault_size < dax_region->align)
313 return VM_FAULT_SIGBUS;
314 else if (fault_size > dax_region->align)
315 return VM_FAULT_FALLBACK;
317 /* if we are outside of the VMA */
318 if (pmd_addr < vmf->vma->vm_start ||
319 (pmd_addr + PMD_SIZE) > vmf->vma->vm_end)
320 return VM_FAULT_SIGBUS;
322 pgoff = linear_page_index(vmf->vma, pmd_addr);
323 phys = dax_pgoff_to_phys(dev_dax, pgoff, PMD_SIZE);
325 dev_dbg(dev, "pgoff_to_phys(%#lx) failed\n", pgoff);
326 return VM_FAULT_SIGBUS;
329 pfn = phys_to_pfn_t(phys, dax_region->pfn_flags);
331 return vmf_insert_pfn_pmd(vmf->vma, vmf->address, vmf->pmd, pfn,
332 vmf->flags & FAULT_FLAG_WRITE);
335 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
336 static int __dev_dax_pud_fault(struct dev_dax *dev_dax, struct vm_fault *vmf)
338 unsigned long pud_addr = vmf->address & PUD_MASK;
339 struct device *dev = &dev_dax->dev;
340 struct dax_region *dax_region;
344 unsigned int fault_size = PUD_SIZE;
347 if (check_vma(dev_dax, vmf->vma, __func__))
348 return VM_FAULT_SIGBUS;
350 dax_region = dev_dax->region;
351 if (dax_region->align > PUD_SIZE) {
352 dev_dbg(dev, "alignment (%#x) > fault size (%#x)\n",
353 dax_region->align, fault_size);
354 return VM_FAULT_SIGBUS;
357 /* dax pud mappings require pfn_t_devmap() */
358 if ((dax_region->pfn_flags & (PFN_DEV|PFN_MAP)) != (PFN_DEV|PFN_MAP)) {
359 dev_dbg(dev, "region lacks devmap flags\n");
360 return VM_FAULT_SIGBUS;
363 if (fault_size < dax_region->align)
364 return VM_FAULT_SIGBUS;
365 else if (fault_size > dax_region->align)
366 return VM_FAULT_FALLBACK;
368 /* if we are outside of the VMA */
369 if (pud_addr < vmf->vma->vm_start ||
370 (pud_addr + PUD_SIZE) > vmf->vma->vm_end)
371 return VM_FAULT_SIGBUS;
373 pgoff = linear_page_index(vmf->vma, pud_addr);
374 phys = dax_pgoff_to_phys(dev_dax, pgoff, PUD_SIZE);
376 dev_dbg(dev, "pgoff_to_phys(%#lx) failed\n", pgoff);
377 return VM_FAULT_SIGBUS;
380 pfn = phys_to_pfn_t(phys, dax_region->pfn_flags);
382 return vmf_insert_pfn_pud(vmf->vma, vmf->address, vmf->pud, pfn,
383 vmf->flags & FAULT_FLAG_WRITE);
386 static int __dev_dax_pud_fault(struct dev_dax *dev_dax, struct vm_fault *vmf)
388 return VM_FAULT_FALLBACK;
390 #endif /* !CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
392 static int dev_dax_huge_fault(struct vm_fault *vmf,
393 enum page_entry_size pe_size)
396 struct file *filp = vmf->vma->vm_file;
397 struct dev_dax *dev_dax = filp->private_data;
399 dev_dbg(&dev_dax->dev, "%s: %s (%#lx - %#lx) size = %d\n", current->comm,
400 (vmf->flags & FAULT_FLAG_WRITE) ? "write" : "read",
401 vmf->vma->vm_start, vmf->vma->vm_end, pe_size);
403 id = dax_read_lock();
406 rc = __dev_dax_pte_fault(dev_dax, vmf);
409 rc = __dev_dax_pmd_fault(dev_dax, vmf);
412 rc = __dev_dax_pud_fault(dev_dax, vmf);
415 rc = VM_FAULT_SIGBUS;
422 static int dev_dax_fault(struct vm_fault *vmf)
424 return dev_dax_huge_fault(vmf, PE_SIZE_PTE);
427 static int dev_dax_split(struct vm_area_struct *vma, unsigned long addr)
429 struct file *filp = vma->vm_file;
430 struct dev_dax *dev_dax = filp->private_data;
431 struct dax_region *dax_region = dev_dax->region;
433 if (!IS_ALIGNED(addr, dax_region->align))
438 static unsigned long dev_dax_pagesize(struct vm_area_struct *vma)
440 struct file *filp = vma->vm_file;
441 struct dev_dax *dev_dax = filp->private_data;
442 struct dax_region *dax_region = dev_dax->region;
444 return dax_region->align;
447 static const struct vm_operations_struct dax_vm_ops = {
448 .fault = dev_dax_fault,
449 .huge_fault = dev_dax_huge_fault,
450 .split = dev_dax_split,
451 .pagesize = dev_dax_pagesize,
454 static int dax_mmap(struct file *filp, struct vm_area_struct *vma)
456 struct dev_dax *dev_dax = filp->private_data;
459 dev_dbg(&dev_dax->dev, "trace\n");
462 * We lock to check dax_dev liveness and will re-check at
465 id = dax_read_lock();
466 rc = check_vma(dev_dax, vma, __func__);
471 vma->vm_ops = &dax_vm_ops;
472 vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
476 /* return an unmapped area aligned to the dax region specified alignment */
477 static unsigned long dax_get_unmapped_area(struct file *filp,
478 unsigned long addr, unsigned long len, unsigned long pgoff,
481 unsigned long off, off_end, off_align, len_align, addr_align, align;
482 struct dev_dax *dev_dax = filp ? filp->private_data : NULL;
483 struct dax_region *dax_region;
485 if (!dev_dax || addr)
488 dax_region = dev_dax->region;
489 align = dax_region->align;
490 off = pgoff << PAGE_SHIFT;
492 off_align = round_up(off, align);
494 if ((off_end <= off_align) || ((off_end - off_align) < align))
497 len_align = len + align;
498 if ((off + len_align) < off)
501 addr_align = current->mm->get_unmapped_area(filp, addr, len_align,
503 if (!IS_ERR_VALUE(addr_align)) {
504 addr_align += (off - addr_align) & (align - 1);
508 return current->mm->get_unmapped_area(filp, addr, len, pgoff, flags);
511 static int dax_open(struct inode *inode, struct file *filp)
513 struct dax_device *dax_dev = inode_dax(inode);
514 struct inode *__dax_inode = dax_inode(dax_dev);
515 struct dev_dax *dev_dax = dax_get_private(dax_dev);
517 dev_dbg(&dev_dax->dev, "trace\n");
518 inode->i_mapping = __dax_inode->i_mapping;
519 inode->i_mapping->host = __dax_inode;
520 filp->f_mapping = inode->i_mapping;
521 filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
522 filp->private_data = dev_dax;
523 inode->i_flags = S_DAX;
528 static int dax_release(struct inode *inode, struct file *filp)
530 struct dev_dax *dev_dax = filp->private_data;
532 dev_dbg(&dev_dax->dev, "trace\n");
536 static const struct file_operations dax_fops = {
537 .llseek = noop_llseek,
538 .owner = THIS_MODULE,
540 .release = dax_release,
541 .get_unmapped_area = dax_get_unmapped_area,
545 static void dev_dax_release(struct device *dev)
547 struct dev_dax *dev_dax = to_dev_dax(dev);
548 struct dax_region *dax_region = dev_dax->region;
549 struct dax_device *dax_dev = dev_dax->dax_dev;
551 if (dev_dax->id >= 0)
552 ida_simple_remove(&dax_region->ida, dev_dax->id);
553 dax_region_put(dax_region);
558 static void kill_dev_dax(struct dev_dax *dev_dax)
560 struct dax_device *dax_dev = dev_dax->dax_dev;
561 struct inode *inode = dax_inode(dax_dev);
564 unmap_mapping_range(inode->i_mapping, 0, 0, 1);
567 static void unregister_dev_dax(void *dev)
569 struct dev_dax *dev_dax = to_dev_dax(dev);
570 struct dax_device *dax_dev = dev_dax->dax_dev;
571 struct inode *inode = dax_inode(dax_dev);
572 struct cdev *cdev = inode->i_cdev;
574 dev_dbg(dev, "trace\n");
576 kill_dev_dax(dev_dax);
577 cdev_device_del(cdev, dev);
581 struct dev_dax *devm_create_dev_dax(struct dax_region *dax_region,
582 int id, struct resource *res, int count)
584 struct device *parent = dax_region->dev;
585 struct dax_device *dax_dev;
586 struct dev_dax *dev_dax;
593 return ERR_PTR(-EINVAL);
595 dev_dax = kzalloc(sizeof(*dev_dax) + sizeof(*res) * count, GFP_KERNEL);
597 return ERR_PTR(-ENOMEM);
599 for (i = 0; i < count; i++) {
600 if (!IS_ALIGNED(res[i].start, dax_region->align)
601 || !IS_ALIGNED(resource_size(&res[i]),
602 dax_region->align)) {
606 dev_dax->res[i].start = res[i].start;
607 dev_dax->res[i].end = res[i].end;
614 id = ida_simple_get(&dax_region->ida, 0, 0, GFP_KERNEL);
621 /* region provider owns @id lifetime */
626 * No 'host' or dax_operations since there is no access to this
627 * device outside of mmap of the resulting character device.
629 dax_dev = alloc_dax(dev_dax, NULL, NULL);
635 /* from here on we're committed to teardown via dax_dev_release() */
637 device_initialize(dev);
639 inode = dax_inode(dax_dev);
640 cdev = inode->i_cdev;
641 cdev_init(cdev, &dax_fops);
642 cdev->owner = parent->driver->owner;
644 dev_dax->num_resources = count;
645 dev_dax->dax_dev = dax_dev;
646 dev_dax->region = dax_region;
647 kref_get(&dax_region->kref);
649 dev->devt = inode->i_rdev;
650 dev->class = dax_class;
651 dev->parent = parent;
652 dev->groups = dax_attribute_groups;
653 dev->release = dev_dax_release;
654 dev_set_name(dev, "dax%d.%d", dax_region->id, id);
656 rc = cdev_device_add(cdev, dev);
658 kill_dev_dax(dev_dax);
663 rc = devm_add_action_or_reset(dax_region->dev, unregister_dev_dax, dev);
670 if (dev_dax->id >= 0)
671 ida_simple_remove(&dax_region->ida, dev_dax->id);
677 EXPORT_SYMBOL_GPL(devm_create_dev_dax);
679 static int __init dax_init(void)
681 dax_class = class_create(THIS_MODULE, "dax");
682 return PTR_ERR_OR_ZERO(dax_class);
685 static void __exit dax_exit(void)
687 class_destroy(dax_class);
690 MODULE_AUTHOR("Intel Corporation");
691 MODULE_LICENSE("GPL v2");
692 subsys_initcall(dax_init);
693 module_exit(dax_exit);