device.c

  1// SPDX-License-Identifier: GPL-2.0
  2/* Copyright(c) 2016-2018 Intel Corporation. All rights reserved. */
  3#include <linux/memremap.h>
  4#include <linux/pagemap.h>
  5#include <linux/module.h>
  6#include <linux/device.h>
  7#include <linux/cdev.h>
  8#include <linux/slab.h>
  9#include <linux/dax.h>
 10#include <linux/fs.h>
 11#include <linux/mm.h>
 12#include <linux/mman.h>
 13#include "dax-private.h"
 14#include "bus.h"
 15
 16static int __check_vma(struct dev_dax *dev_dax, vm_flags_t vm_flags,
 17		       unsigned long start, unsigned long end, struct file *file,
 18		       const char *func)
 19{
 20	struct device *dev = &dev_dax->dev;
 21	unsigned long mask;
 22
 23	if (!dax_alive(dev_dax->dax_dev))
 24		return -ENXIO;
 25
 26	/* prevent private mappings from being established */
 27	if ((vm_flags & VM_MAYSHARE) != VM_MAYSHARE) {
 28		dev_info_ratelimited(dev,
 29				"%s: %s: fail, attempted private mapping\n",
 30				current->comm, func);
 31		return -EINVAL;
 32	}
 33
 34	mask = dev_dax->align - 1;
 35	if (start & mask || end & mask) {
 36		dev_info_ratelimited(dev,
 37				"%s: %s: fail, unaligned vma (%#lx - %#lx, %#lx)\n",
 38				current->comm, func, start, end,
 39				mask);
 40		return -EINVAL;
 41	}
 42
 43	if (!file_is_dax(file)) {
 44		dev_info_ratelimited(dev,
 45				"%s: %s: fail, vma is not DAX capable\n",
 46				current->comm, func);
 47		return -EINVAL;
 48	}
 49
 50	return 0;
 51}
 52
 53static int check_vma(struct dev_dax *dev_dax, struct vm_area_struct *vma,
 54		     const char *func)
 55{
 56	return __check_vma(dev_dax, vma->vm_flags, vma->vm_start, vma->vm_end,
 57			   vma->vm_file, func);
 58}
 59
 60/* see "strong" declaration in tools/testing/nvdimm/dax-dev.c */
 61__weak phys_addr_t dax_pgoff_to_phys(struct dev_dax *dev_dax, pgoff_t pgoff,
 62		unsigned long size)
 63{
 64	int i;
 65
 66	for (i = 0; i < dev_dax->nr_range; i++) {
 67		struct dev_dax_range *dax_range = &dev_dax->ranges[i];
 68		struct range *range = &dax_range->range;
 69		unsigned long long pgoff_end;
 70		phys_addr_t phys;
 71
 72		pgoff_end = dax_range->pgoff + PHYS_PFN(range_len(range)) - 1;
 73		if (pgoff < dax_range->pgoff || pgoff > pgoff_end)
 74			continue;
 75		phys = PFN_PHYS(pgoff - dax_range->pgoff) + range->start;
 76		if (phys + size - 1 <= range->end)
 77			return phys;
 78		break;
 79	}
 80	return -1;
 81}
 82
 83static void dax_set_mapping(struct vm_fault *vmf, unsigned long pfn,
 84			      unsigned long fault_size)
 85{
 86	unsigned long i, nr_pages = fault_size / PAGE_SIZE;
 87	struct file *filp = vmf->vma->vm_file;
 88	struct dev_dax *dev_dax = filp->private_data;
 89	pgoff_t pgoff;
 90
 91	/* mapping is only set on the head */
 92	if (dev_dax->pgmap->vmemmap_shift)
 93		nr_pages = 1;
 94
 95	pgoff = linear_page_index(vmf->vma,
 96			ALIGN_DOWN(vmf->address, fault_size));
 97
 98	for (i = 0; i < nr_pages; i++) {
 99		struct folio *folio = pfn_folio(pfn + i);
100
101		if (folio->mapping)
102			continue;
103
104		folio->mapping = filp->f_mapping;
105		folio->index = pgoff + i;
106	}
107}
108
109static vm_fault_t __dev_dax_pte_fault(struct dev_dax *dev_dax,
110				struct vm_fault *vmf)
111{
112	struct device *dev = &dev_dax->dev;
113	phys_addr_t phys;
114	unsigned long pfn;
115	unsigned int fault_size = PAGE_SIZE;
116
117	if (check_vma(dev_dax, vmf->vma, __func__))
118		return VM_FAULT_SIGBUS;
119
120	if (dev_dax->align > PAGE_SIZE) {
121		dev_dbg(dev, "alignment (%#x) > fault size (%#x)\n",
122			dev_dax->align, fault_size);
123		return VM_FAULT_SIGBUS;
124	}
125
126	if (fault_size != dev_dax->align)
127		return VM_FAULT_SIGBUS;
128
129	phys = dax_pgoff_to_phys(dev_dax, vmf->pgoff, PAGE_SIZE);
130	if (phys == -1) {
131		dev_dbg(dev, "pgoff_to_phys(%#lx) failed\n", vmf->pgoff);
132		return VM_FAULT_SIGBUS;
133	}
134
135	pfn = PHYS_PFN(phys);
136
137	dax_set_mapping(vmf, pfn, fault_size);
138
139	return vmf_insert_page_mkwrite(vmf, pfn_to_page(pfn),
140					vmf->flags & FAULT_FLAG_WRITE);
141}
142
143static vm_fault_t __dev_dax_pmd_fault(struct dev_dax *dev_dax,
144				struct vm_fault *vmf)
145{
146	unsigned long pmd_addr = vmf->address & PMD_MASK;
147	struct device *dev = &dev_dax->dev;
148	phys_addr_t phys;
149	pgoff_t pgoff;
150	unsigned long pfn;
151	unsigned int fault_size = PMD_SIZE;
152
153	if (check_vma(dev_dax, vmf->vma, __func__))
154		return VM_FAULT_SIGBUS;
155
156	if (dev_dax->align > PMD_SIZE) {
157		dev_dbg(dev, "alignment (%#x) > fault size (%#x)\n",
158			dev_dax->align, fault_size);
159		return VM_FAULT_SIGBUS;
160	}
161
162	if (fault_size < dev_dax->align)
163		return VM_FAULT_SIGBUS;
164	else if (fault_size > dev_dax->align)
165		return VM_FAULT_FALLBACK;
166
167	/* if we are outside of the VMA */
168	if (pmd_addr < vmf->vma->vm_start ||
169			(pmd_addr + PMD_SIZE) > vmf->vma->vm_end)
170		return VM_FAULT_SIGBUS;
171
172	pgoff = linear_page_index(vmf->vma, pmd_addr);
173	phys = dax_pgoff_to_phys(dev_dax, pgoff, PMD_SIZE);
174	if (phys == -1) {
175		dev_dbg(dev, "pgoff_to_phys(%#lx) failed\n", pgoff);
176		return VM_FAULT_SIGBUS;
177	}
178
179	pfn = PHYS_PFN(phys);
180
181	dax_set_mapping(vmf, pfn, fault_size);
182
183	return vmf_insert_folio_pmd(vmf, page_folio(pfn_to_page(pfn)),
184				vmf->flags & FAULT_FLAG_WRITE);
185}
186
187#ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
188static vm_fault_t __dev_dax_pud_fault(struct dev_dax *dev_dax,
189				struct vm_fault *vmf)
190{
191	unsigned long pud_addr = vmf->address & PUD_MASK;
192	struct device *dev = &dev_dax->dev;
193	phys_addr_t phys;
194	pgoff_t pgoff;
195	unsigned long pfn;
196	unsigned int fault_size = PUD_SIZE;
197
198
199	if (check_vma(dev_dax, vmf->vma, __func__))
200		return VM_FAULT_SIGBUS;
201
202	if (dev_dax->align > PUD_SIZE) {
203		dev_dbg(dev, "alignment (%#x) > fault size (%#x)\n",
204			dev_dax->align, fault_size);
205		return VM_FAULT_SIGBUS;
206	}
207
208	if (fault_size < dev_dax->align)
209		return VM_FAULT_SIGBUS;
210	else if (fault_size > dev_dax->align)
211		return VM_FAULT_FALLBACK;
212
213	/* if we are outside of the VMA */
214	if (pud_addr < vmf->vma->vm_start ||
215			(pud_addr + PUD_SIZE) > vmf->vma->vm_end)
216		return VM_FAULT_SIGBUS;
217
218	pgoff = linear_page_index(vmf->vma, pud_addr);
219	phys = dax_pgoff_to_phys(dev_dax, pgoff, PUD_SIZE);
220	if (phys == -1) {
221		dev_dbg(dev, "pgoff_to_phys(%#lx) failed\n", pgoff);
222		return VM_FAULT_SIGBUS;
223	}
224
225	pfn = PHYS_PFN(phys);
226
227	dax_set_mapping(vmf, pfn, fault_size);
228
229	return vmf_insert_folio_pud(vmf, page_folio(pfn_to_page(pfn)),
230				vmf->flags & FAULT_FLAG_WRITE);
231}
232#else
233static vm_fault_t __dev_dax_pud_fault(struct dev_dax *dev_dax,
234				struct vm_fault *vmf)
235{
236	return VM_FAULT_FALLBACK;
237}
238#endif /* !CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
239
240static vm_fault_t dev_dax_huge_fault(struct vm_fault *vmf, unsigned int order)
241{
242	struct file *filp = vmf->vma->vm_file;
243	vm_fault_t rc = VM_FAULT_SIGBUS;
244	int id;
245	struct dev_dax *dev_dax = filp->private_data;
246
247	dev_dbg(&dev_dax->dev, "%s: op=%s addr=%#lx order=%d\n", current->comm,
248		(vmf->flags & FAULT_FLAG_WRITE) ? "write" : "read",
249		vmf->address & ~((1UL << (order + PAGE_SHIFT)) - 1), order);
250
251	id = dax_read_lock();
252	if (order == 0)
253		rc = __dev_dax_pte_fault(dev_dax, vmf);
254	else if (order == PMD_ORDER)
255		rc = __dev_dax_pmd_fault(dev_dax, vmf);
256	else if (order == PUD_ORDER)
257		rc = __dev_dax_pud_fault(dev_dax, vmf);
258	else
259		rc = VM_FAULT_SIGBUS;
260
261	dax_read_unlock(id);
262
263	return rc;
264}
265
266static vm_fault_t dev_dax_fault(struct vm_fault *vmf)
267{
268	return dev_dax_huge_fault(vmf, 0);
269}
270
271static int dev_dax_may_split(struct vm_area_struct *vma, unsigned long addr)
272{
273	struct file *filp = vma->vm_file;
274	struct dev_dax *dev_dax = filp->private_data;
275
276	if (!IS_ALIGNED(addr, dev_dax->align))
277		return -EINVAL;
278	return 0;
279}
280
281static unsigned long dev_dax_pagesize(struct vm_area_struct *vma)
282{
283	struct file *filp = vma->vm_file;
284	struct dev_dax *dev_dax = filp->private_data;
285
286	return dev_dax->align;
287}
288
289static const struct vm_operations_struct dax_vm_ops = {
290	.fault = dev_dax_fault,
291	.huge_fault = dev_dax_huge_fault,
292	.may_split = dev_dax_may_split,
293	.pagesize = dev_dax_pagesize,
294};
295
296static int dax_mmap_prepare(struct vm_area_desc *desc)
297{
298	struct file *filp = desc->file;
299	struct dev_dax *dev_dax = filp->private_data;
300	int rc, id;
301
302	dev_dbg(&dev_dax->dev, "trace\n");
303
304	/*
305	 * We lock to check dax_dev liveness and will re-check at
306	 * fault time.
307	 */
308	id = dax_read_lock();
309	rc = __check_vma(dev_dax, desc->vm_flags, desc->start, desc->end, filp,
310			 __func__);
311	dax_read_unlock(id);
312	if (rc)
313		return rc;
314
315	desc->vm_ops = &dax_vm_ops;
316	desc->vm_flags |= VM_HUGEPAGE;
317	return 0;
318}
319
320/* return an unmapped area aligned to the dax region specified alignment */
321static unsigned long dax_get_unmapped_area(struct file *filp,
322		unsigned long addr, unsigned long len, unsigned long pgoff,
323		unsigned long flags)
324{
325	unsigned long off, off_end, off_align, len_align, addr_align, align;
326	struct dev_dax *dev_dax = filp ? filp->private_data : NULL;
327
328	if (!dev_dax || addr)
329		goto out;
330
331	align = dev_dax->align;
332	off = pgoff << PAGE_SHIFT;
333	off_end = off + len;
334	off_align = round_up(off, align);
335
336	if ((off_end <= off_align) || ((off_end - off_align) < align))
337		goto out;
338
339	len_align = len + align;
340	if ((off + len_align) < off)
341		goto out;
342
343	addr_align = mm_get_unmapped_area(filp, addr, len_align, pgoff, flags);
344	if (!IS_ERR_VALUE(addr_align)) {
345		addr_align += (off - addr_align) & (align - 1);
346		return addr_align;
347	}
348 out:
349	return mm_get_unmapped_area(filp, addr, len, pgoff, flags);
350}
351
352static const struct address_space_operations dev_dax_aops = {
353	.dirty_folio	= noop_dirty_folio,
354};
355
356static int dax_open(struct inode *inode, struct file *filp)
357{
358	struct dax_device *dax_dev = inode_dax(inode);
359	struct inode *__dax_inode = dax_inode(dax_dev);
360	struct dev_dax *dev_dax = dax_get_private(dax_dev);
361
362	dev_dbg(&dev_dax->dev, "trace\n");
363	inode->i_mapping = __dax_inode->i_mapping;
364	inode->i_mapping->host = __dax_inode;
365	inode->i_mapping->a_ops = &dev_dax_aops;
366	filp->f_mapping = inode->i_mapping;
367	filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
368	filp->f_sb_err = file_sample_sb_err(filp);
369	filp->private_data = dev_dax;
370	inode->i_flags = S_DAX;
371
372	return 0;
373}
374
375static int dax_release(struct inode *inode, struct file *filp)
376{
377	struct dev_dax *dev_dax = filp->private_data;
378
379	dev_dbg(&dev_dax->dev, "trace\n");
380	return 0;
381}
382
383static const struct file_operations dax_fops = {
384	.llseek = noop_llseek,
385	.owner = THIS_MODULE,
386	.open = dax_open,
387	.release = dax_release,
388	.get_unmapped_area = dax_get_unmapped_area,
389	.mmap_prepare = dax_mmap_prepare,
390	.fop_flags = FOP_MMAP_SYNC,
391};
392
393static void dev_dax_cdev_del(void *cdev)
394{
395	cdev_del(cdev);
396}
397
398static void dev_dax_kill(void *dev_dax)
399{
400	kill_dev_dax(dev_dax);
401}
402
403static int dev_dax_probe(struct dev_dax *dev_dax)
404{
405	struct dax_device *dax_dev = dev_dax->dax_dev;
406	struct device *dev = &dev_dax->dev;
407	struct dev_pagemap *pgmap;
408	struct inode *inode;
409	struct cdev *cdev;
410	void *addr;
411	int rc, i;
412
413	if (static_dev_dax(dev_dax))  {
414		if (dev_dax->nr_range > 1) {
415			dev_warn(dev,
416				"static pgmap / multi-range device conflict\n");
417			return -EINVAL;
418		}
419
420		pgmap = dev_dax->pgmap;
421	} else {
422		if (dev_dax->pgmap) {
423			dev_warn(dev,
424				 "dynamic-dax with pre-populated page map\n");
425			return -EINVAL;
426		}
427
428		pgmap = devm_kzalloc(dev,
429                       struct_size(pgmap, ranges, dev_dax->nr_range - 1),
430                       GFP_KERNEL);
431		if (!pgmap)
432			return -ENOMEM;
433
434		pgmap->nr_range = dev_dax->nr_range;
435		dev_dax->pgmap = pgmap;
436
437		for (i = 0; i < dev_dax->nr_range; i++) {
438			struct range *range = &dev_dax->ranges[i].range;
439			pgmap->ranges[i] = *range;
440		}
441	}
442
443	for (i = 0; i < dev_dax->nr_range; i++) {
444		struct range *range = &dev_dax->ranges[i].range;
445
446		if (!devm_request_mem_region(dev, range->start,
447					range_len(range), dev_name(dev))) {
448			dev_warn(dev, "mapping%d: %#llx-%#llx could not reserve range\n",
449					i, range->start, range->end);
450			return -EBUSY;
451		}
452	}
453
454	pgmap->type = MEMORY_DEVICE_GENERIC;
455	if (dev_dax->align > PAGE_SIZE)
456		pgmap->vmemmap_shift =
457			order_base_2(dev_dax->align >> PAGE_SHIFT);
458	addr = devm_memremap_pages(dev, pgmap);
459	if (IS_ERR(addr))
460		return PTR_ERR(addr);
461
462	inode = dax_inode(dax_dev);
463	cdev = inode->i_cdev;
464	cdev_init(cdev, &dax_fops);
465	cdev->owner = dev->driver->owner;
466	cdev_set_parent(cdev, &dev->kobj);
467	rc = cdev_add(cdev, dev->devt, 1);
468	if (rc)
469		return rc;
470
471	rc = devm_add_action_or_reset(dev, dev_dax_cdev_del, cdev);
472	if (rc)
473		return rc;
474
475	run_dax(dax_dev);
476	return devm_add_action_or_reset(dev, dev_dax_kill, dev_dax);
477}
478
479static struct dax_device_driver device_dax_driver = {
480	.probe = dev_dax_probe,
481	.type = DAXDRV_DEVICE_TYPE,
482};
483
484static int __init dax_init(void)
485{
486	return dax_driver_register(&device_dax_driver);
487}
488
489static void __exit dax_exit(void)
490{
491	dax_driver_unregister(&device_dax_driver);
492}
493
494MODULE_AUTHOR("Intel Corporation");
495MODULE_DESCRIPTION("Device DAX: direct access device driver");
496MODULE_LICENSE("GPL v2");
497module_init(dax_init);
498module_exit(dax_exit);
499MODULE_ALIAS_DAX_DEVICE(0);