1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright(c) 2015 Intel Corporation. All rights reserved. */
3 #include <linux/device.h>
5 #include <linux/kasan.h>
6 #include <linux/memory_hotplug.h>
8 #include <linux/pfn_t.h>
9 #include <linux/swap.h>
10 #include <linux/mmzone.h>
11 #include <linux/swapops.h>
12 #include <linux/types.h>
13 #include <linux/wait_bit.h>
14 #include <linux/xarray.h>
16 static DEFINE_XARRAY(pgmap_array
);
19 * The memremap() and memremap_pages() interfaces are alternately used
20 * to map persistent memory namespaces. These interfaces place different
21 * constraints on the alignment and size of the mapping (namespace).
22 * memremap() can map individual PAGE_SIZE pages. memremap_pages() can
23 * only map subsections (2MB), and at least one architecture (PowerPC)
24 * the minimum mapping granularity of memremap_pages() is 16MB.
26 * The role of memremap_compat_align() is to communicate the minimum
27 * arch supported alignment of a namespace such that it can freely
28 * switch modes without violating the arch constraint. Namely, do not
29 * allow a namespace to be PAGE_SIZE aligned since that namespace may be
30 * reconfigured into a mode that requires SUBSECTION_SIZE alignment.
32 #ifndef CONFIG_ARCH_HAS_MEMREMAP_COMPAT_ALIGN
33 unsigned long memremap_compat_align(void)
35 return SUBSECTION_SIZE
;
37 EXPORT_SYMBOL_GPL(memremap_compat_align
);
40 #ifdef CONFIG_DEV_PAGEMAP_OPS
41 DEFINE_STATIC_KEY_FALSE(devmap_managed_key
);
42 EXPORT_SYMBOL(devmap_managed_key
);
44 static void devmap_managed_enable_put(struct dev_pagemap
*pgmap
)
46 if (pgmap
->type
== MEMORY_DEVICE_PRIVATE
||
47 pgmap
->type
== MEMORY_DEVICE_FS_DAX
)
48 static_branch_dec(&devmap_managed_key
);
51 static void devmap_managed_enable_get(struct dev_pagemap
*pgmap
)
53 if (pgmap
->type
== MEMORY_DEVICE_PRIVATE
||
54 pgmap
->type
== MEMORY_DEVICE_FS_DAX
)
55 static_branch_inc(&devmap_managed_key
);
58 static void devmap_managed_enable_get(struct dev_pagemap
*pgmap
)
61 static void devmap_managed_enable_put(struct dev_pagemap
*pgmap
)
64 #endif /* CONFIG_DEV_PAGEMAP_OPS */
66 static void pgmap_array_delete(struct range
*range
)
68 xa_store_range(&pgmap_array
, PHYS_PFN(range
->start
), PHYS_PFN(range
->end
),
73 static unsigned long pfn_first(struct dev_pagemap
*pgmap
, int range_id
)
75 struct range
*range
= &pgmap
->ranges
[range_id
];
76 unsigned long pfn
= PHYS_PFN(range
->start
);
80 return pfn
+ vmem_altmap_offset(pgmap_altmap(pgmap
));
83 bool pgmap_pfn_valid(struct dev_pagemap
*pgmap
, unsigned long pfn
)
87 for (i
= 0; i
< pgmap
->nr_range
; i
++) {
88 struct range
*range
= &pgmap
->ranges
[i
];
90 if (pfn
>= PHYS_PFN(range
->start
) &&
91 pfn
<= PHYS_PFN(range
->end
))
92 return pfn
>= pfn_first(pgmap
, i
);
98 static unsigned long pfn_end(struct dev_pagemap
*pgmap
, int range_id
)
100 const struct range
*range
= &pgmap
->ranges
[range_id
];
102 return (range
->start
+ range_len(range
)) >> PAGE_SHIFT
;
105 static unsigned long pfn_next(unsigned long pfn
)
112 #define for_each_device_pfn(pfn, map, i) \
113 for (pfn = pfn_first(map, i); pfn < pfn_end(map, i); pfn = pfn_next(pfn))
115 static void dev_pagemap_kill(struct dev_pagemap
*pgmap
)
117 if (pgmap
->ops
&& pgmap
->ops
->kill
)
118 pgmap
->ops
->kill(pgmap
);
120 percpu_ref_kill(pgmap
->ref
);
123 static void dev_pagemap_cleanup(struct dev_pagemap
*pgmap
)
125 if (pgmap
->ops
&& pgmap
->ops
->cleanup
) {
126 pgmap
->ops
->cleanup(pgmap
);
128 wait_for_completion(&pgmap
->done
);
129 percpu_ref_exit(pgmap
->ref
);
132 * Undo the pgmap ref assignment for the internal case as the
133 * caller may re-enable the same pgmap.
135 if (pgmap
->ref
== &pgmap
->internal_ref
)
139 static void pageunmap_range(struct dev_pagemap
*pgmap
, int range_id
)
141 struct range
*range
= &pgmap
->ranges
[range_id
];
142 struct page
*first_page
;
144 /* make sure to access a memmap that was actually initialized */
145 first_page
= pfn_to_page(pfn_first(pgmap
, range_id
));
147 /* pages are dead and unused, undo the arch mapping */
149 remove_pfn_range_from_zone(page_zone(first_page
), PHYS_PFN(range
->start
),
150 PHYS_PFN(range_len(range
)));
151 if (pgmap
->type
== MEMORY_DEVICE_PRIVATE
) {
152 __remove_pages(PHYS_PFN(range
->start
),
153 PHYS_PFN(range_len(range
)), NULL
);
155 arch_remove_memory(range
->start
, range_len(range
),
156 pgmap_altmap(pgmap
));
157 kasan_remove_zero_shadow(__va(range
->start
), range_len(range
));
161 untrack_pfn(NULL
, PHYS_PFN(range
->start
), range_len(range
));
162 pgmap_array_delete(range
);
165 void memunmap_pages(struct dev_pagemap
*pgmap
)
170 dev_pagemap_kill(pgmap
);
171 for (i
= 0; i
< pgmap
->nr_range
; i
++)
172 for_each_device_pfn(pfn
, pgmap
, i
)
173 put_page(pfn_to_page(pfn
));
174 dev_pagemap_cleanup(pgmap
);
176 for (i
= 0; i
< pgmap
->nr_range
; i
++)
177 pageunmap_range(pgmap
, i
);
179 WARN_ONCE(pgmap
->altmap
.alloc
, "failed to free all reserved pages\n");
180 devmap_managed_enable_put(pgmap
);
182 EXPORT_SYMBOL_GPL(memunmap_pages
);
184 static void devm_memremap_pages_release(void *data
)
186 memunmap_pages(data
);
189 static void dev_pagemap_percpu_release(struct percpu_ref
*ref
)
191 struct dev_pagemap
*pgmap
=
192 container_of(ref
, struct dev_pagemap
, internal_ref
);
194 complete(&pgmap
->done
);
197 static int pagemap_range(struct dev_pagemap
*pgmap
, struct mhp_params
*params
,
198 int range_id
, int nid
)
200 const bool is_private
= pgmap
->type
== MEMORY_DEVICE_PRIVATE
;
201 struct range
*range
= &pgmap
->ranges
[range_id
];
202 struct dev_pagemap
*conflict_pgmap
;
205 if (WARN_ONCE(pgmap_altmap(pgmap
) && range_id
> 0,
206 "altmap not supported for multiple ranges\n"))
209 conflict_pgmap
= get_dev_pagemap(PHYS_PFN(range
->start
), NULL
);
210 if (conflict_pgmap
) {
211 WARN(1, "Conflicting mapping in same section\n");
212 put_dev_pagemap(conflict_pgmap
);
216 conflict_pgmap
= get_dev_pagemap(PHYS_PFN(range
->end
), NULL
);
217 if (conflict_pgmap
) {
218 WARN(1, "Conflicting mapping in same section\n");
219 put_dev_pagemap(conflict_pgmap
);
223 is_ram
= region_intersects(range
->start
, range_len(range
),
224 IORESOURCE_SYSTEM_RAM
, IORES_DESC_NONE
);
226 if (is_ram
!= REGION_DISJOINT
) {
227 WARN_ONCE(1, "attempted on %s region %#llx-%#llx\n",
228 is_ram
== REGION_MIXED
? "mixed" : "ram",
229 range
->start
, range
->end
);
233 error
= xa_err(xa_store_range(&pgmap_array
, PHYS_PFN(range
->start
),
234 PHYS_PFN(range
->end
), pgmap
, GFP_KERNEL
));
241 error
= track_pfn_remap(NULL
, ¶ms
->pgprot
, PHYS_PFN(range
->start
), 0,
246 if (!mhp_range_allowed(range
->start
, range_len(range
), !is_private
)) {
254 * For device private memory we call add_pages() as we only need to
255 * allocate and initialize struct page for the device memory. More-
256 * over the device memory is un-accessible thus we do not want to
257 * create a linear mapping for the memory like arch_add_memory()
260 * For all other device memory types, which are accessible by
261 * the CPU, we do want the linear mapping and thus use
265 error
= add_pages(nid
, PHYS_PFN(range
->start
),
266 PHYS_PFN(range_len(range
)), params
);
268 error
= kasan_add_zero_shadow(__va(range
->start
), range_len(range
));
274 error
= arch_add_memory(nid
, range
->start
, range_len(range
),
281 zone
= &NODE_DATA(nid
)->node_zones
[ZONE_DEVICE
];
282 move_pfn_range_to_zone(zone
, PHYS_PFN(range
->start
),
283 PHYS_PFN(range_len(range
)), params
->altmap
,
292 * Initialization of the pages has been deferred until now in order
293 * to allow us to do the work while not holding the hotplug lock.
295 memmap_init_zone_device(&NODE_DATA(nid
)->node_zones
[ZONE_DEVICE
],
296 PHYS_PFN(range
->start
),
297 PHYS_PFN(range_len(range
)), pgmap
);
298 percpu_ref_get_many(pgmap
->ref
, pfn_end(pgmap
, range_id
)
299 - pfn_first(pgmap
, range_id
));
303 kasan_remove_zero_shadow(__va(range
->start
), range_len(range
));
305 untrack_pfn(NULL
, PHYS_PFN(range
->start
), range_len(range
));
307 pgmap_array_delete(range
);
313 * Not device managed version of dev_memremap_pages, undone by
314 * memunmap_pages(). Please use dev_memremap_pages if you have a struct
317 void *memremap_pages(struct dev_pagemap
*pgmap
, int nid
)
319 struct mhp_params params
= {
320 .altmap
= pgmap_altmap(pgmap
),
321 .pgprot
= PAGE_KERNEL
,
323 const int nr_range
= pgmap
->nr_range
;
326 if (WARN_ONCE(!nr_range
, "nr_range must be specified\n"))
327 return ERR_PTR(-EINVAL
);
329 switch (pgmap
->type
) {
330 case MEMORY_DEVICE_PRIVATE
:
331 if (!IS_ENABLED(CONFIG_DEVICE_PRIVATE
)) {
332 WARN(1, "Device private memory not supported\n");
333 return ERR_PTR(-EINVAL
);
335 if (!pgmap
->ops
|| !pgmap
->ops
->migrate_to_ram
) {
336 WARN(1, "Missing migrate_to_ram method\n");
337 return ERR_PTR(-EINVAL
);
339 if (!pgmap
->ops
->page_free
) {
340 WARN(1, "Missing page_free method\n");
341 return ERR_PTR(-EINVAL
);
344 WARN(1, "Missing owner\n");
345 return ERR_PTR(-EINVAL
);
348 case MEMORY_DEVICE_FS_DAX
:
349 if (!IS_ENABLED(CONFIG_ZONE_DEVICE
) ||
350 IS_ENABLED(CONFIG_FS_DAX_LIMITED
)) {
351 WARN(1, "File system DAX not supported\n");
352 return ERR_PTR(-EINVAL
);
355 case MEMORY_DEVICE_GENERIC
:
357 case MEMORY_DEVICE_PCI_P2PDMA
:
358 params
.pgprot
= pgprot_noncached(params
.pgprot
);
361 WARN(1, "Invalid pgmap type %d\n", pgmap
->type
);
366 if (pgmap
->ops
&& (pgmap
->ops
->kill
|| pgmap
->ops
->cleanup
))
367 return ERR_PTR(-EINVAL
);
369 init_completion(&pgmap
->done
);
370 error
= percpu_ref_init(&pgmap
->internal_ref
,
371 dev_pagemap_percpu_release
, 0, GFP_KERNEL
);
373 return ERR_PTR(error
);
374 pgmap
->ref
= &pgmap
->internal_ref
;
376 if (!pgmap
->ops
|| !pgmap
->ops
->kill
|| !pgmap
->ops
->cleanup
) {
377 WARN(1, "Missing reference count teardown definition\n");
378 return ERR_PTR(-EINVAL
);
382 devmap_managed_enable_get(pgmap
);
385 * Clear the pgmap nr_range as it will be incremented for each
386 * successfully processed range. This communicates how many
387 * regions to unwind in the abort case.
391 for (i
= 0; i
< nr_range
; i
++) {
392 error
= pagemap_range(pgmap
, ¶ms
, i
, nid
);
399 memunmap_pages(pgmap
);
400 pgmap
->nr_range
= nr_range
;
401 return ERR_PTR(error
);
404 return __va(pgmap
->ranges
[0].start
);
406 EXPORT_SYMBOL_GPL(memremap_pages
);
409 * devm_memremap_pages - remap and provide memmap backing for the given resource
410 * @dev: hosting device for @res
411 * @pgmap: pointer to a struct dev_pagemap
414 * 1/ At a minimum the res and type members of @pgmap must be initialized
415 * by the caller before passing it to this function
417 * 2/ The altmap field may optionally be initialized, in which case
418 * PGMAP_ALTMAP_VALID must be set in pgmap->flags.
420 * 3/ The ref field may optionally be provided, in which pgmap->ref must be
421 * 'live' on entry and will be killed and reaped at
422 * devm_memremap_pages_release() time, or if this routine fails.
424 * 4/ range is expected to be a host memory range that could feasibly be
425 * treated as a "System RAM" range, i.e. not a device mmio range, but
426 * this is not enforced.
428 void *devm_memremap_pages(struct device
*dev
, struct dev_pagemap
*pgmap
)
433 ret
= memremap_pages(pgmap
, dev_to_node(dev
));
437 error
= devm_add_action_or_reset(dev
, devm_memremap_pages_release
,
440 return ERR_PTR(error
);
443 EXPORT_SYMBOL_GPL(devm_memremap_pages
);
445 void devm_memunmap_pages(struct device
*dev
, struct dev_pagemap
*pgmap
)
447 devm_release_action(dev
, devm_memremap_pages_release
, pgmap
);
449 EXPORT_SYMBOL_GPL(devm_memunmap_pages
);
451 unsigned long vmem_altmap_offset(struct vmem_altmap
*altmap
)
453 /* number of pfns from base where pfn_to_page() is valid */
455 return altmap
->reserve
+ altmap
->free
;
459 void vmem_altmap_free(struct vmem_altmap
*altmap
, unsigned long nr_pfns
)
461 altmap
->alloc
-= nr_pfns
;
465 * get_dev_pagemap() - take a new live reference on the dev_pagemap for @pfn
466 * @pfn: page frame number to lookup page_map
467 * @pgmap: optional known pgmap that already has a reference
469 * If @pgmap is non-NULL and covers @pfn it will be returned as-is. If @pgmap
470 * is non-NULL but does not cover @pfn the reference to it will be released.
472 struct dev_pagemap
*get_dev_pagemap(unsigned long pfn
,
473 struct dev_pagemap
*pgmap
)
475 resource_size_t phys
= PFN_PHYS(pfn
);
478 * In the cached case we're already holding a live reference.
481 if (phys
>= pgmap
->range
.start
&& phys
<= pgmap
->range
.end
)
483 put_dev_pagemap(pgmap
);
486 /* fall back to slow path lookup */
488 pgmap
= xa_load(&pgmap_array
, PHYS_PFN(phys
));
489 if (pgmap
&& !percpu_ref_tryget_live(pgmap
->ref
))
495 EXPORT_SYMBOL_GPL(get_dev_pagemap
);
497 #ifdef CONFIG_DEV_PAGEMAP_OPS
498 void free_devmap_managed_page(struct page
*page
)
500 /* notify page idle for dax */
501 if (!is_device_private_page(page
)) {
502 wake_up_var(&page
->_refcount
);
506 __ClearPageWaiters(page
);
508 mem_cgroup_uncharge(page
);
511 * When a device_private page is freed, the page->mapping field
512 * may still contain a (stale) mapping value. For example, the
513 * lower bits of page->mapping may still identify the page as an
514 * anonymous page. Ultimately, this entire field is just stale
515 * and wrong, and it will cause errors if not cleared. One
518 * migrate_vma_pages()
519 * migrate_vma_insert_page()
520 * page_add_new_anon_rmap()
521 * __page_set_anon_rmap()
522 * ...checks page->mapping, via PageAnon(page) call,
523 * and incorrectly concludes that the page is an
524 * anonymous page. Therefore, it incorrectly,
525 * silently fails to set up the new anon rmap.
527 * For other types of ZONE_DEVICE pages, migration is either
528 * handled differently or not done at all, so there is no need
529 * to clear page->mapping.
531 page
->mapping
= NULL
;
532 page
->pgmap
->ops
->page_free(page
);
534 #endif /* CONFIG_DEV_PAGEMAP_OPS */