1 // SPDX-License-Identifier: GPL-2.0-only
3 * linux/kernel/resource.c
5 * Copyright (C) 1999 Linus Torvalds
6 * Copyright (C) 1999 Martin Mares <mj@ucw.cz>
8 * Arbitrary resource management.
11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
13 #include <linux/export.h>
14 #include <linux/errno.h>
15 #include <linux/ioport.h>
16 #include <linux/init.h>
17 #include <linux/slab.h>
18 #include <linux/spinlock.h>
20 #include <linux/proc_fs.h>
21 #include <linux/pseudo_fs.h>
22 #include <linux/sched.h>
23 #include <linux/seq_file.h>
24 #include <linux/device.h>
25 #include <linux/pfn.h>
27 #include <linux/mount.h>
28 #include <linux/resource_ext.h>
29 #include <uapi/linux/magic.h>
33 struct resource ioport_resource = {
36 .end = IO_SPACE_LIMIT,
37 .flags = IORESOURCE_IO,
39 EXPORT_SYMBOL(ioport_resource);
41 struct resource iomem_resource = {
45 .flags = IORESOURCE_MEM,
47 EXPORT_SYMBOL(iomem_resource);
49 /* constraints to be met while allocating resources */
50 struct resource_constraint {
51 resource_size_t min, max, align;
52 resource_size_t (*alignf)(void *, const struct resource *,
53 resource_size_t, resource_size_t);
57 static DEFINE_RWLOCK(resource_lock);
59 static struct resource *next_resource(struct resource *p)
63 while (!p->sibling && p->parent)
68 static struct resource *next_resource_skip_children(struct resource *p)
70 while (!p->sibling && p->parent)
75 #define for_each_resource(_root, _p, _skip_children) \
76 for ((_p) = (_root)->child; (_p); \
77 (_p) = (_skip_children) ? next_resource_skip_children(_p) : \
80 static void *r_next(struct seq_file *m, void *v, loff_t *pos)
82 struct resource *p = v;
84 return (void *)next_resource(p);
89 enum { MAX_IORES_LEVEL = 5 };
91 static void *r_start(struct seq_file *m, loff_t *pos)
92 __acquires(resource_lock)
94 struct resource *p = pde_data(file_inode(m->file));
96 read_lock(&resource_lock);
97 for (p = p->child; p && l < *pos; p = r_next(m, p, &l))
102 static void r_stop(struct seq_file *m, void *v)
103 __releases(resource_lock)
105 read_unlock(&resource_lock);
108 static int r_show(struct seq_file *m, void *v)
110 struct resource *root = pde_data(file_inode(m->file));
111 struct resource *r = v, *p;
112 unsigned long long start, end;
113 int width = root->end < 0x10000 ? 4 : 8;
116 for (depth = 0, p = r; depth < MAX_IORES_LEVEL; depth++, p = p->parent)
117 if (p->parent == root)
120 if (file_ns_capable(m->file, &init_user_ns, CAP_SYS_ADMIN)) {
127 seq_printf(m, "%*s%0*llx-%0*llx : %s\n",
131 r->name ? r->name : "<BAD>");
135 static const struct seq_operations resource_op = {
142 static int __init ioresources_init(void)
144 proc_create_seq_data("ioports", 0, NULL, &resource_op,
146 proc_create_seq_data("iomem", 0, NULL, &resource_op, &iomem_resource);
149 __initcall(ioresources_init);
151 #endif /* CONFIG_PROC_FS */
153 static void free_resource(struct resource *res)
156 * If the resource was allocated using memblock early during boot
157 * we'll leak it here: we can only return full pages back to the
158 * buddy and trying to be smart and reusing them eventually in
159 * alloc_resource() overcomplicates resource handling.
161 if (res && PageSlab(virt_to_head_page(res)))
165 static struct resource *alloc_resource(gfp_t flags)
167 return kzalloc(sizeof(struct resource), flags);
170 /* Return the conflict entry if you can't request it */
171 static struct resource * __request_resource(struct resource *root, struct resource *new)
173 resource_size_t start = new->start;
174 resource_size_t end = new->end;
175 struct resource *tmp, **p;
179 if (start < root->start)
186 if (!tmp || tmp->start > end) {
193 if (tmp->end < start)
199 static int __release_resource(struct resource *old, bool release_child)
201 struct resource *tmp, **p, *chd;
203 p = &old->parent->child;
209 if (release_child || !(tmp->child)) {
212 for (chd = tmp->child;; chd = chd->sibling) {
213 chd->parent = tmp->parent;
218 chd->sibling = tmp->sibling;
228 static void __release_child_resources(struct resource *r)
230 struct resource *tmp, *p;
231 resource_size_t size;
241 __release_child_resources(tmp);
243 printk(KERN_DEBUG "release child resource %pR\n", tmp);
244 /* need to restore size, and keep flags */
245 size = resource_size(tmp);
251 void release_child_resources(struct resource *r)
253 write_lock(&resource_lock);
254 __release_child_resources(r);
255 write_unlock(&resource_lock);
259 * request_resource_conflict - request and reserve an I/O or memory resource
260 * @root: root resource descriptor
261 * @new: resource descriptor desired by caller
263 * Returns 0 for success, conflict resource on error.
265 struct resource *request_resource_conflict(struct resource *root, struct resource *new)
267 struct resource *conflict;
269 write_lock(&resource_lock);
270 conflict = __request_resource(root, new);
271 write_unlock(&resource_lock);
276 * request_resource - request and reserve an I/O or memory resource
277 * @root: root resource descriptor
278 * @new: resource descriptor desired by caller
280 * Returns 0 for success, negative error code on error.
282 int request_resource(struct resource *root, struct resource *new)
284 struct resource *conflict;
286 conflict = request_resource_conflict(root, new);
287 return conflict ? -EBUSY : 0;
290 EXPORT_SYMBOL(request_resource);
293 * release_resource - release a previously reserved resource
294 * @old: resource pointer
296 int release_resource(struct resource *old)
300 write_lock(&resource_lock);
301 retval = __release_resource(old, true);
302 write_unlock(&resource_lock);
306 EXPORT_SYMBOL(release_resource);
309 * find_next_iomem_res - Finds the lowest iomem resource that covers part of
312 * If a resource is found, returns 0 and @*res is overwritten with the part
313 * of the resource that's within [@start..@end]; if none is found, returns
314 * -ENODEV. Returns -EINVAL for invalid parameters.
316 * @start: start address of the resource searched for
317 * @end: end address of same resource
318 * @flags: flags which the resource must have
319 * @desc: descriptor the resource must have
320 * @res: return ptr, if resource found
322 * The caller must specify @start, @end, @flags, and @desc
323 * (which may be IORES_DESC_NONE).
325 static int find_next_iomem_res(resource_size_t start, resource_size_t end,
326 unsigned long flags, unsigned long desc,
327 struct resource *res)
337 read_lock(&resource_lock);
339 for (p = iomem_resource.child; p; p = next_resource(p)) {
340 /* If we passed the resource we are looking for, stop */
341 if (p->start > end) {
346 /* Skip until we find a range that matches what we look for */
350 if ((p->flags & flags) != flags)
352 if ((desc != IORES_DESC_NONE) && (desc != p->desc))
355 /* Found a match, break */
361 *res = (struct resource) {
362 .start = max(start, p->start),
363 .end = min(end, p->end),
370 read_unlock(&resource_lock);
371 return p ? 0 : -ENODEV;
374 static int __walk_iomem_res_desc(resource_size_t start, resource_size_t end,
375 unsigned long flags, unsigned long desc,
377 int (*func)(struct resource *, void *))
382 while (start < end &&
383 !find_next_iomem_res(start, end, flags, desc, &res)) {
384 ret = (*func)(&res, arg);
395 * walk_iomem_res_desc - Walks through iomem resources and calls func()
396 * with matching resource ranges.
398 * @desc: I/O resource descriptor. Use IORES_DESC_NONE to skip @desc check.
399 * @flags: I/O resource flags
402 * @arg: function argument for the callback @func
403 * @func: callback function that is called for each qualifying resource area
405 * All the memory ranges which overlap start,end and also match flags and
406 * desc are valid candidates.
408 * NOTE: For a new descriptor search, define a new IORES_DESC in
409 * <linux/ioport.h> and set it in 'desc' of a target resource entry.
411 int walk_iomem_res_desc(unsigned long desc, unsigned long flags, u64 start,
412 u64 end, void *arg, int (*func)(struct resource *, void *))
414 return __walk_iomem_res_desc(start, end, flags, desc, arg, func);
416 EXPORT_SYMBOL_GPL(walk_iomem_res_desc);
419 * This function calls the @func callback against all memory ranges of type
420 * System RAM which are marked as IORESOURCE_SYSTEM_RAM and IORESOUCE_BUSY.
421 * Now, this function is only for System RAM, it deals with full ranges and
422 * not PFNs. If resources are not PFN-aligned, dealing with PFNs can truncate
425 int walk_system_ram_res(u64 start, u64 end, void *arg,
426 int (*func)(struct resource *, void *))
428 unsigned long flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
430 return __walk_iomem_res_desc(start, end, flags, IORES_DESC_NONE, arg,
435 * This function calls the @func callback against all memory ranges, which
436 * are ranges marked as IORESOURCE_MEM and IORESOUCE_BUSY.
438 int walk_mem_res(u64 start, u64 end, void *arg,
439 int (*func)(struct resource *, void *))
441 unsigned long flags = IORESOURCE_MEM | IORESOURCE_BUSY;
443 return __walk_iomem_res_desc(start, end, flags, IORES_DESC_NONE, arg,
448 * This function calls the @func callback against all memory ranges of type
449 * System RAM which are marked as IORESOURCE_SYSTEM_RAM and IORESOUCE_BUSY.
450 * It is to be used only for System RAM.
452 int walk_system_ram_range(unsigned long start_pfn, unsigned long nr_pages,
453 void *arg, int (*func)(unsigned long, unsigned long, void *))
455 resource_size_t start, end;
458 unsigned long pfn, end_pfn;
461 start = (u64) start_pfn << PAGE_SHIFT;
462 end = ((u64)(start_pfn + nr_pages) << PAGE_SHIFT) - 1;
463 flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
464 while (start < end &&
465 !find_next_iomem_res(start, end, flags, IORES_DESC_NONE, &res)) {
466 pfn = PFN_UP(res.start);
467 end_pfn = PFN_DOWN(res.end + 1);
469 ret = (*func)(pfn, end_pfn - pfn, arg);
477 static int __is_ram(unsigned long pfn, unsigned long nr_pages, void *arg)
483 * This generic page_is_ram() returns true if specified address is
484 * registered as System RAM in iomem_resource list.
486 int __weak page_is_ram(unsigned long pfn)
488 return walk_system_ram_range(pfn, 1, NULL, __is_ram) == 1;
490 EXPORT_SYMBOL_GPL(page_is_ram);
492 static int __region_intersects(resource_size_t start, size_t size,
493 unsigned long flags, unsigned long desc)
496 int type = 0; int other = 0;
500 res.end = start + size - 1;
502 for (p = iomem_resource.child; p ; p = p->sibling) {
503 bool is_type = (((p->flags & flags) == flags) &&
504 ((desc == IORES_DESC_NONE) ||
507 if (resource_overlaps(p, &res))
508 is_type ? type++ : other++;
512 return REGION_DISJOINT;
515 return REGION_INTERSECTS;
521 * region_intersects() - determine intersection of region with known resources
522 * @start: region start address
523 * @size: size of region
524 * @flags: flags of resource (in iomem_resource)
525 * @desc: descriptor of resource (in iomem_resource) or IORES_DESC_NONE
527 * Check if the specified region partially overlaps or fully eclipses a
528 * resource identified by @flags and @desc (optional with IORES_DESC_NONE).
529 * Return REGION_DISJOINT if the region does not overlap @flags/@desc,
530 * return REGION_MIXED if the region overlaps @flags/@desc and another
531 * resource, and return REGION_INTERSECTS if the region overlaps @flags/@desc
532 * and no other defined resource. Note that REGION_INTERSECTS is also
533 * returned in the case when the specified region overlaps RAM and undefined
536 * region_intersect() is used by memory remapping functions to ensure
537 * the user is not remapping RAM and is a vast speed up over walking
538 * through the resource table page by page.
540 int region_intersects(resource_size_t start, size_t size, unsigned long flags,
545 read_lock(&resource_lock);
546 ret = __region_intersects(start, size, flags, desc);
547 read_unlock(&resource_lock);
551 EXPORT_SYMBOL_GPL(region_intersects);
553 void __weak arch_remove_reservations(struct resource *avail)
557 static resource_size_t simple_align_resource(void *data,
558 const struct resource *avail,
559 resource_size_t size,
560 resource_size_t align)
565 static void resource_clip(struct resource *res, resource_size_t min,
568 if (res->start < min)
575 * Find empty slot in the resource tree with the given range and
576 * alignment constraints
578 static int __find_resource(struct resource *root, struct resource *old,
579 struct resource *new,
580 resource_size_t size,
581 struct resource_constraint *constraint)
583 struct resource *this = root->child;
584 struct resource tmp = *new, avail, alloc;
586 tmp.start = root->start;
588 * Skip past an allocated resource that starts at 0, since the assignment
589 * of this->start - 1 to tmp->end below would cause an underflow.
591 if (this && this->start == root->start) {
592 tmp.start = (this == old) ? old->start : this->end + 1;
593 this = this->sibling;
597 tmp.end = (this == old) ? this->end : this->start - 1;
601 if (tmp.end < tmp.start)
604 resource_clip(&tmp, constraint->min, constraint->max);
605 arch_remove_reservations(&tmp);
607 /* Check for overflow after ALIGN() */
608 avail.start = ALIGN(tmp.start, constraint->align);
610 avail.flags = new->flags & ~IORESOURCE_UNSET;
611 if (avail.start >= tmp.start) {
612 alloc.flags = avail.flags;
613 alloc.start = constraint->alignf(constraint->alignf_data, &avail,
614 size, constraint->align);
615 alloc.end = alloc.start + size - 1;
616 if (alloc.start <= alloc.end &&
617 resource_contains(&avail, &alloc)) {
618 new->start = alloc.start;
619 new->end = alloc.end;
624 next: if (!this || this->end == root->end)
628 tmp.start = this->end + 1;
629 this = this->sibling;
635 * Find empty slot in the resource tree given range and alignment.
637 static int find_resource(struct resource *root, struct resource *new,
638 resource_size_t size,
639 struct resource_constraint *constraint)
641 return __find_resource(root, NULL, new, size, constraint);
645 * reallocate_resource - allocate a slot in the resource tree given range & alignment.
646 * The resource will be relocated if the new size cannot be reallocated in the
649 * @root: root resource descriptor
650 * @old: resource descriptor desired by caller
651 * @newsize: new size of the resource descriptor
652 * @constraint: the size and alignment constraints to be met.
654 static int reallocate_resource(struct resource *root, struct resource *old,
655 resource_size_t newsize,
656 struct resource_constraint *constraint)
659 struct resource new = *old;
660 struct resource *conflict;
662 write_lock(&resource_lock);
664 if ((err = __find_resource(root, old, &new, newsize, constraint)))
667 if (resource_contains(&new, old)) {
668 old->start = new.start;
678 if (resource_contains(old, &new)) {
679 old->start = new.start;
682 __release_resource(old, true);
684 conflict = __request_resource(root, old);
688 write_unlock(&resource_lock);
694 * allocate_resource - allocate empty slot in the resource tree given range & alignment.
695 * The resource will be reallocated with a new size if it was already allocated
696 * @root: root resource descriptor
697 * @new: resource descriptor desired by caller
698 * @size: requested resource region size
699 * @min: minimum boundary to allocate
700 * @max: maximum boundary to allocate
701 * @align: alignment requested, in bytes
702 * @alignf: alignment function, optional, called if not NULL
703 * @alignf_data: arbitrary data to pass to the @alignf function
705 int allocate_resource(struct resource *root, struct resource *new,
706 resource_size_t size, resource_size_t min,
707 resource_size_t max, resource_size_t align,
708 resource_size_t (*alignf)(void *,
709 const struct resource *,
715 struct resource_constraint constraint;
718 alignf = simple_align_resource;
720 constraint.min = min;
721 constraint.max = max;
722 constraint.align = align;
723 constraint.alignf = alignf;
724 constraint.alignf_data = alignf_data;
727 /* resource is already allocated, try reallocating with
728 the new constraints */
729 return reallocate_resource(root, new, size, &constraint);
732 write_lock(&resource_lock);
733 err = find_resource(root, new, size, &constraint);
734 if (err >= 0 && __request_resource(root, new))
736 write_unlock(&resource_lock);
740 EXPORT_SYMBOL(allocate_resource);
743 * lookup_resource - find an existing resource by a resource start address
744 * @root: root resource descriptor
745 * @start: resource start address
747 * Returns a pointer to the resource if found, NULL otherwise
749 struct resource *lookup_resource(struct resource *root, resource_size_t start)
751 struct resource *res;
753 read_lock(&resource_lock);
754 for (res = root->child; res; res = res->sibling) {
755 if (res->start == start)
758 read_unlock(&resource_lock);
764 * Insert a resource into the resource tree. If successful, return NULL,
765 * otherwise return the conflicting resource (compare to __request_resource())
767 static struct resource * __insert_resource(struct resource *parent, struct resource *new)
769 struct resource *first, *next;
771 for (;; parent = first) {
772 first = __request_resource(parent, new);
778 if (WARN_ON(first == new)) /* duplicated insertion */
781 if ((first->start > new->start) || (first->end < new->end))
783 if ((first->start == new->start) && (first->end == new->end))
787 for (next = first; ; next = next->sibling) {
788 /* Partial overlap? Bad, and unfixable */
789 if (next->start < new->start || next->end > new->end)
793 if (next->sibling->start > new->end)
797 new->parent = parent;
798 new->sibling = next->sibling;
801 next->sibling = NULL;
802 for (next = first; next; next = next->sibling)
805 if (parent->child == first) {
808 next = parent->child;
809 while (next->sibling != first)
810 next = next->sibling;
817 * insert_resource_conflict - Inserts resource in the resource tree
818 * @parent: parent of the new resource
819 * @new: new resource to insert
821 * Returns 0 on success, conflict resource if the resource can't be inserted.
823 * This function is equivalent to request_resource_conflict when no conflict
824 * happens. If a conflict happens, and the conflicting resources
825 * entirely fit within the range of the new resource, then the new
826 * resource is inserted and the conflicting resources become children of
829 * This function is intended for producers of resources, such as FW modules
832 struct resource *insert_resource_conflict(struct resource *parent, struct resource *new)
834 struct resource *conflict;
836 write_lock(&resource_lock);
837 conflict = __insert_resource(parent, new);
838 write_unlock(&resource_lock);
843 * insert_resource - Inserts a resource in the resource tree
844 * @parent: parent of the new resource
845 * @new: new resource to insert
847 * Returns 0 on success, -EBUSY if the resource can't be inserted.
849 * This function is intended for producers of resources, such as FW modules
852 int insert_resource(struct resource *parent, struct resource *new)
854 struct resource *conflict;
856 conflict = insert_resource_conflict(parent, new);
857 return conflict ? -EBUSY : 0;
859 EXPORT_SYMBOL_GPL(insert_resource);
862 * insert_resource_expand_to_fit - Insert a resource into the resource tree
863 * @root: root resource descriptor
864 * @new: new resource to insert
866 * Insert a resource into the resource tree, possibly expanding it in order
867 * to make it encompass any conflicting resources.
869 void insert_resource_expand_to_fit(struct resource *root, struct resource *new)
874 write_lock(&resource_lock);
876 struct resource *conflict;
878 conflict = __insert_resource(root, new);
881 if (conflict == root)
884 /* Ok, expand resource to cover the conflict, then try again .. */
885 if (conflict->start < new->start)
886 new->start = conflict->start;
887 if (conflict->end > new->end)
888 new->end = conflict->end;
890 printk("Expanded resource %s due to conflict with %s\n", new->name, conflict->name);
892 write_unlock(&resource_lock);
895 * Not for general consumption, only early boot memory map parsing, PCI
896 * resource discovery, and late discovery of CXL resources are expected
897 * to use this interface. The former are built-in and only the latter,
900 EXPORT_SYMBOL_NS_GPL(insert_resource_expand_to_fit, CXL);
903 * remove_resource - Remove a resource in the resource tree
904 * @old: resource to remove
906 * Returns 0 on success, -EINVAL if the resource is not valid.
908 * This function removes a resource previously inserted by insert_resource()
909 * or insert_resource_conflict(), and moves the children (if any) up to
910 * where they were before. insert_resource() and insert_resource_conflict()
911 * insert a new resource, and move any conflicting resources down to the
912 * children of the new resource.
914 * insert_resource(), insert_resource_conflict() and remove_resource() are
915 * intended for producers of resources, such as FW modules and bus drivers.
917 int remove_resource(struct resource *old)
921 write_lock(&resource_lock);
922 retval = __release_resource(old, false);
923 write_unlock(&resource_lock);
926 EXPORT_SYMBOL_GPL(remove_resource);
928 static int __adjust_resource(struct resource *res, resource_size_t start,
929 resource_size_t size)
931 struct resource *tmp, *parent = res->parent;
932 resource_size_t end = start + size - 1;
938 if ((start < parent->start) || (end > parent->end))
941 if (res->sibling && (res->sibling->start <= end))
946 while (tmp->sibling != res)
948 if (start <= tmp->end)
953 for (tmp = res->child; tmp; tmp = tmp->sibling)
954 if ((tmp->start < start) || (tmp->end > end))
966 * adjust_resource - modify a resource's start and size
967 * @res: resource to modify
968 * @start: new start value
971 * Given an existing resource, change its start and size to match the
972 * arguments. Returns 0 on success, -EBUSY if it can't fit.
973 * Existing children of the resource are assumed to be immutable.
975 int adjust_resource(struct resource *res, resource_size_t start,
976 resource_size_t size)
980 write_lock(&resource_lock);
981 result = __adjust_resource(res, start, size);
982 write_unlock(&resource_lock);
985 EXPORT_SYMBOL(adjust_resource);
988 __reserve_region_with_split(struct resource *root, resource_size_t start,
989 resource_size_t end, const char *name)
991 struct resource *parent = root;
992 struct resource *conflict;
993 struct resource *res = alloc_resource(GFP_ATOMIC);
994 struct resource *next_res = NULL;
995 int type = resource_type(root);
1003 res->flags = type | IORESOURCE_BUSY;
1004 res->desc = IORES_DESC_NONE;
1008 conflict = __request_resource(parent, res);
1017 /* conflict covered whole area */
1018 if (conflict->start <= res->start &&
1019 conflict->end >= res->end) {
1025 /* failed, split and try again */
1026 if (conflict->start > res->start) {
1028 res->end = conflict->start - 1;
1029 if (conflict->end < end) {
1030 next_res = alloc_resource(GFP_ATOMIC);
1035 next_res->name = name;
1036 next_res->start = conflict->end + 1;
1037 next_res->end = end;
1038 next_res->flags = type | IORESOURCE_BUSY;
1039 next_res->desc = IORES_DESC_NONE;
1042 res->start = conflict->end + 1;
1049 reserve_region_with_split(struct resource *root, resource_size_t start,
1050 resource_size_t end, const char *name)
1054 write_lock(&resource_lock);
1055 if (root->start > start || root->end < end) {
1056 pr_err("requested range [0x%llx-0x%llx] not in root %pr\n",
1057 (unsigned long long)start, (unsigned long long)end,
1059 if (start > root->end || end < root->start)
1062 if (end > root->end)
1064 if (start < root->start)
1065 start = root->start;
1066 pr_err("fixing request to [0x%llx-0x%llx]\n",
1067 (unsigned long long)start,
1068 (unsigned long long)end);
1073 __reserve_region_with_split(root, start, end, name);
1074 write_unlock(&resource_lock);
1078 * resource_alignment - calculate resource's alignment
1079 * @res: resource pointer
1081 * Returns alignment on success, 0 (invalid alignment) on failure.
1083 resource_size_t resource_alignment(struct resource *res)
1085 switch (res->flags & (IORESOURCE_SIZEALIGN | IORESOURCE_STARTALIGN)) {
1086 case IORESOURCE_SIZEALIGN:
1087 return resource_size(res);
1088 case IORESOURCE_STARTALIGN:
1096 * This is compatibility stuff for IO resources.
1098 * Note how this, unlike the above, knows about
1099 * the IO flag meanings (busy etc).
1101 * request_region creates a new busy region.
1103 * release_region releases a matching busy region.
1106 static DECLARE_WAIT_QUEUE_HEAD(muxed_resource_wait);
1108 static struct inode *iomem_inode;
1110 #ifdef CONFIG_IO_STRICT_DEVMEM
1111 static void revoke_iomem(struct resource *res)
1113 /* pairs with smp_store_release() in iomem_init_inode() */
1114 struct inode *inode = smp_load_acquire(&iomem_inode);
1117 * Check that the initialization has completed. Losing the race
1118 * is ok because it means drivers are claiming resources before
1119 * the fs_initcall level of init and prevent iomem_get_mapping users
1120 * from establishing mappings.
1126 * The expectation is that the driver has successfully marked
1127 * the resource busy by this point, so devmem_is_allowed()
1128 * should start returning false, however for performance this
1129 * does not iterate the entire resource range.
1131 if (devmem_is_allowed(PHYS_PFN(res->start)) &&
1132 devmem_is_allowed(PHYS_PFN(res->end))) {
1134 * *cringe* iomem=relaxed says "go ahead, what's the
1135 * worst that can happen?"
1140 unmap_mapping_range(inode->i_mapping, res->start, resource_size(res), 1);
1143 static void revoke_iomem(struct resource *res) {}
1146 struct address_space *iomem_get_mapping(void)
1149 * This function is only called from file open paths, hence guaranteed
1150 * that fs_initcalls have completed and no need to check for NULL. But
1151 * since revoke_iomem can be called before the initcall we still need
1152 * the barrier to appease checkers.
1154 return smp_load_acquire(&iomem_inode)->i_mapping;
1157 static int __request_region_locked(struct resource *res, struct resource *parent,
1158 resource_size_t start, resource_size_t n,
1159 const char *name, int flags)
1161 DECLARE_WAITQUEUE(wait, current);
1165 res->end = start + n - 1;
1168 struct resource *conflict;
1170 res->flags = resource_type(parent) | resource_ext_type(parent);
1171 res->flags |= IORESOURCE_BUSY | flags;
1172 res->desc = parent->desc;
1174 conflict = __request_resource(parent, res);
1178 * mm/hmm.c reserves physical addresses which then
1179 * become unavailable to other users. Conflicts are
1180 * not expected. Warn to aid debugging if encountered.
1182 if (conflict->desc == IORES_DESC_DEVICE_PRIVATE_MEMORY) {
1183 pr_warn("Unaddressable device %s %pR conflicts with %pR",
1184 conflict->name, conflict, res);
1186 if (conflict != parent) {
1187 if (!(conflict->flags & IORESOURCE_BUSY)) {
1192 if (conflict->flags & flags & IORESOURCE_MUXED) {
1193 add_wait_queue(&muxed_resource_wait, &wait);
1194 write_unlock(&resource_lock);
1195 set_current_state(TASK_UNINTERRUPTIBLE);
1197 remove_wait_queue(&muxed_resource_wait, &wait);
1198 write_lock(&resource_lock);
1201 /* Uhhuh, that didn't work out.. */
1209 * __request_region - create a new busy resource region
1210 * @parent: parent resource descriptor
1211 * @start: resource start address
1212 * @n: resource region size
1213 * @name: reserving caller's ID string
1214 * @flags: IO resource flags
1216 struct resource *__request_region(struct resource *parent,
1217 resource_size_t start, resource_size_t n,
1218 const char *name, int flags)
1220 struct resource *res = alloc_resource(GFP_KERNEL);
1226 write_lock(&resource_lock);
1227 ret = __request_region_locked(res, parent, start, n, name, flags);
1228 write_unlock(&resource_lock);
1235 if (parent == &iomem_resource)
1240 EXPORT_SYMBOL(__request_region);
1243 * __release_region - release a previously reserved resource region
1244 * @parent: parent resource descriptor
1245 * @start: resource start address
1246 * @n: resource region size
1248 * The described resource region must match a currently busy region.
1250 void __release_region(struct resource *parent, resource_size_t start,
1253 struct resource **p;
1254 resource_size_t end;
1257 end = start + n - 1;
1259 write_lock(&resource_lock);
1262 struct resource *res = *p;
1266 if (res->start <= start && res->end >= end) {
1267 if (!(res->flags & IORESOURCE_BUSY)) {
1271 if (res->start != start || res->end != end)
1274 write_unlock(&resource_lock);
1275 if (res->flags & IORESOURCE_MUXED)
1276 wake_up(&muxed_resource_wait);
1283 write_unlock(&resource_lock);
1285 printk(KERN_WARNING "Trying to free nonexistent resource "
1286 "<%016llx-%016llx>\n", (unsigned long long)start,
1287 (unsigned long long)end);
1289 EXPORT_SYMBOL(__release_region);
1291 #ifdef CONFIG_MEMORY_HOTREMOVE
1293 * release_mem_region_adjustable - release a previously reserved memory region
1294 * @start: resource start address
1295 * @size: resource region size
1297 * This interface is intended for memory hot-delete. The requested region
1298 * is released from a currently busy memory resource. The requested region
1299 * must either match exactly or fit into a single busy resource entry. In
1300 * the latter case, the remaining resource is adjusted accordingly.
1301 * Existing children of the busy memory resource must be immutable in the
1305 * - Additional release conditions, such as overlapping region, can be
1306 * supported after they are confirmed as valid cases.
1307 * - When a busy memory resource gets split into two entries, the code
1308 * assumes that all children remain in the lower address entry for
1309 * simplicity. Enhance this logic when necessary.
1311 void release_mem_region_adjustable(resource_size_t start, resource_size_t size)
1313 struct resource *parent = &iomem_resource;
1314 struct resource *new_res = NULL;
1315 bool alloc_nofail = false;
1316 struct resource **p;
1317 struct resource *res;
1318 resource_size_t end;
1320 end = start + size - 1;
1321 if (WARN_ON_ONCE((start < parent->start) || (end > parent->end)))
1325 * We free up quite a lot of memory on memory hotunplug (esp., memap),
1326 * just before releasing the region. This is highly unlikely to
1327 * fail - let's play save and make it never fail as the caller cannot
1328 * perform any error handling (e.g., trying to re-add memory will fail
1332 new_res = alloc_resource(GFP_KERNEL | (alloc_nofail ? __GFP_NOFAIL : 0));
1335 write_lock(&resource_lock);
1337 while ((res = *p)) {
1338 if (res->start >= end)
1341 /* look for the next resource if it does not fit into */
1342 if (res->start > start || res->end < end) {
1348 * All memory regions added from memory-hotplug path have the
1349 * flag IORESOURCE_SYSTEM_RAM. If the resource does not have
1350 * this flag, we know that we are dealing with a resource coming
1351 * from HMM/devm. HMM/devm use another mechanism to add/release
1352 * a resource. This goes via devm_request_mem_region and
1353 * devm_release_mem_region.
1354 * HMM/devm take care to release their resources when they want,
1355 * so if we are dealing with them, let us just back off here.
1357 if (!(res->flags & IORESOURCE_SYSRAM)) {
1361 if (!(res->flags & IORESOURCE_MEM))
1364 if (!(res->flags & IORESOURCE_BUSY)) {
1369 /* found the target resource; let's adjust accordingly */
1370 if (res->start == start && res->end == end) {
1371 /* free the whole entry */
1374 } else if (res->start == start && res->end != end) {
1375 /* adjust the start */
1376 WARN_ON_ONCE(__adjust_resource(res, end + 1,
1378 } else if (res->start != start && res->end == end) {
1379 /* adjust the end */
1380 WARN_ON_ONCE(__adjust_resource(res, res->start,
1381 start - res->start));
1383 /* split into two entries - we need a new resource */
1385 new_res = alloc_resource(GFP_ATOMIC);
1387 alloc_nofail = true;
1388 write_unlock(&resource_lock);
1392 new_res->name = res->name;
1393 new_res->start = end + 1;
1394 new_res->end = res->end;
1395 new_res->flags = res->flags;
1396 new_res->desc = res->desc;
1397 new_res->parent = res->parent;
1398 new_res->sibling = res->sibling;
1399 new_res->child = NULL;
1401 if (WARN_ON_ONCE(__adjust_resource(res, res->start,
1402 start - res->start)))
1404 res->sibling = new_res;
1411 write_unlock(&resource_lock);
1412 free_resource(new_res);
1414 #endif /* CONFIG_MEMORY_HOTREMOVE */
1416 #ifdef CONFIG_MEMORY_HOTPLUG
1417 static bool system_ram_resources_mergeable(struct resource *r1,
1418 struct resource *r2)
1420 /* We assume either r1 or r2 is IORESOURCE_SYSRAM_MERGEABLE. */
1421 return r1->flags == r2->flags && r1->end + 1 == r2->start &&
1422 r1->name == r2->name && r1->desc == r2->desc &&
1423 !r1->child && !r2->child;
1427 * merge_system_ram_resource - mark the System RAM resource mergeable and try to
1428 * merge it with adjacent, mergeable resources
1429 * @res: resource descriptor
1431 * This interface is intended for memory hotplug, whereby lots of contiguous
1432 * system ram resources are added (e.g., via add_memory*()) by a driver, and
1433 * the actual resource boundaries are not of interest (e.g., it might be
1434 * relevant for DIMMs). Only resources that are marked mergeable, that have the
1435 * same parent, and that don't have any children are considered. All mergeable
1436 * resources must be immutable during the request.
1439 * - The caller has to make sure that no pointers to resources that are
1440 * marked mergeable are used anymore after this call - the resource might
1441 * be freed and the pointer might be stale!
1442 * - release_mem_region_adjustable() will split on demand on memory hotunplug
1444 void merge_system_ram_resource(struct resource *res)
1446 const unsigned long flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
1447 struct resource *cur;
1449 if (WARN_ON_ONCE((res->flags & flags) != flags))
1452 write_lock(&resource_lock);
1453 res->flags |= IORESOURCE_SYSRAM_MERGEABLE;
1455 /* Try to merge with next item in the list. */
1457 if (cur && system_ram_resources_mergeable(res, cur)) {
1458 res->end = cur->end;
1459 res->sibling = cur->sibling;
1463 /* Try to merge with previous item in the list. */
1464 cur = res->parent->child;
1465 while (cur && cur->sibling != res)
1467 if (cur && system_ram_resources_mergeable(cur, res)) {
1468 cur->end = res->end;
1469 cur->sibling = res->sibling;
1472 write_unlock(&resource_lock);
1474 #endif /* CONFIG_MEMORY_HOTPLUG */
1477 * Managed region resource
1479 static void devm_resource_release(struct device *dev, void *ptr)
1481 struct resource **r = ptr;
1483 release_resource(*r);
1487 * devm_request_resource() - request and reserve an I/O or memory resource
1488 * @dev: device for which to request the resource
1489 * @root: root of the resource tree from which to request the resource
1490 * @new: descriptor of the resource to request
1492 * This is a device-managed version of request_resource(). There is usually
1493 * no need to release resources requested by this function explicitly since
1494 * that will be taken care of when the device is unbound from its driver.
1495 * If for some reason the resource needs to be released explicitly, because
1496 * of ordering issues for example, drivers must call devm_release_resource()
1497 * rather than the regular release_resource().
1499 * When a conflict is detected between any existing resources and the newly
1500 * requested resource, an error message will be printed.
1502 * Returns 0 on success or a negative error code on failure.
1504 int devm_request_resource(struct device *dev, struct resource *root,
1505 struct resource *new)
1507 struct resource *conflict, **ptr;
1509 ptr = devres_alloc(devm_resource_release, sizeof(*ptr), GFP_KERNEL);
1515 conflict = request_resource_conflict(root, new);
1517 dev_err(dev, "resource collision: %pR conflicts with %s %pR\n",
1518 new, conflict->name, conflict);
1523 devres_add(dev, ptr);
1526 EXPORT_SYMBOL(devm_request_resource);
1528 static int devm_resource_match(struct device *dev, void *res, void *data)
1530 struct resource **ptr = res;
1532 return *ptr == data;
1536 * devm_release_resource() - release a previously requested resource
1537 * @dev: device for which to release the resource
1538 * @new: descriptor of the resource to release
1540 * Releases a resource previously requested using devm_request_resource().
1542 void devm_release_resource(struct device *dev, struct resource *new)
1544 WARN_ON(devres_release(dev, devm_resource_release, devm_resource_match,
1547 EXPORT_SYMBOL(devm_release_resource);
1549 struct region_devres {
1550 struct resource *parent;
1551 resource_size_t start;
1555 static void devm_region_release(struct device *dev, void *res)
1557 struct region_devres *this = res;
1559 __release_region(this->parent, this->start, this->n);
1562 static int devm_region_match(struct device *dev, void *res, void *match_data)
1564 struct region_devres *this = res, *match = match_data;
1566 return this->parent == match->parent &&
1567 this->start == match->start && this->n == match->n;
1571 __devm_request_region(struct device *dev, struct resource *parent,
1572 resource_size_t start, resource_size_t n, const char *name)
1574 struct region_devres *dr = NULL;
1575 struct resource *res;
1577 dr = devres_alloc(devm_region_release, sizeof(struct region_devres),
1582 dr->parent = parent;
1586 res = __request_region(parent, start, n, name, 0);
1588 devres_add(dev, dr);
1594 EXPORT_SYMBOL(__devm_request_region);
1596 void __devm_release_region(struct device *dev, struct resource *parent,
1597 resource_size_t start, resource_size_t n)
1599 struct region_devres match_data = { parent, start, n };
1601 __release_region(parent, start, n);
1602 WARN_ON(devres_destroy(dev, devm_region_release, devm_region_match,
1605 EXPORT_SYMBOL(__devm_release_region);
1608 * Reserve I/O ports or memory based on "reserve=" kernel parameter.
1610 #define MAXRESERVE 4
1611 static int __init reserve_setup(char *str)
1613 static int reserved;
1614 static struct resource reserve[MAXRESERVE];
1617 unsigned int io_start, io_num;
1619 struct resource *parent;
1621 if (get_option(&str, &io_start) != 2)
1623 if (get_option(&str, &io_num) == 0)
1625 if (x < MAXRESERVE) {
1626 struct resource *res = reserve + x;
1629 * If the region starts below 0x10000, we assume it's
1630 * I/O port space; otherwise assume it's memory.
1632 if (io_start < 0x10000) {
1633 res->flags = IORESOURCE_IO;
1634 parent = &ioport_resource;
1636 res->flags = IORESOURCE_MEM;
1637 parent = &iomem_resource;
1639 res->name = "reserved";
1640 res->start = io_start;
1641 res->end = io_start + io_num - 1;
1642 res->flags |= IORESOURCE_BUSY;
1643 res->desc = IORES_DESC_NONE;
1645 if (request_resource(parent, res) == 0)
1651 __setup("reserve=", reserve_setup);
1654 * Check if the requested addr and size spans more than any slot in the
1655 * iomem resource tree.
1657 int iomem_map_sanity_check(resource_size_t addr, unsigned long size)
1659 struct resource *p = &iomem_resource;
1663 read_lock(&resource_lock);
1664 for (p = p->child; p ; p = r_next(NULL, p, &l)) {
1666 * We can probably skip the resources without
1667 * IORESOURCE_IO attribute?
1669 if (p->start >= addr + size)
1673 if (PFN_DOWN(p->start) <= PFN_DOWN(addr) &&
1674 PFN_DOWN(p->end) >= PFN_DOWN(addr + size - 1))
1677 * if a resource is "BUSY", it's not a hardware resource
1678 * but a driver mapping of such a resource; we don't want
1679 * to warn for those; some drivers legitimately map only
1680 * partial hardware resources. (example: vesafb)
1682 if (p->flags & IORESOURCE_BUSY)
1685 printk(KERN_WARNING "resource sanity check: requesting [mem %#010llx-%#010llx], which spans more than %s %pR\n",
1686 (unsigned long long)addr,
1687 (unsigned long long)(addr + size - 1),
1692 read_unlock(&resource_lock);
1697 #ifdef CONFIG_STRICT_DEVMEM
1698 static int strict_iomem_checks = 1;
1700 static int strict_iomem_checks;
1704 * Check if an address is exclusive to the kernel and must not be mapped to
1705 * user space, for example, via /dev/mem.
1707 * Returns true if exclusive to the kernel, otherwise returns false.
1709 bool iomem_is_exclusive(u64 addr)
1711 const unsigned int exclusive_system_ram = IORESOURCE_SYSTEM_RAM |
1712 IORESOURCE_EXCLUSIVE;
1713 bool skip_children = false, err = false;
1714 int size = PAGE_SIZE;
1717 addr = addr & PAGE_MASK;
1719 read_lock(&resource_lock);
1720 for_each_resource(&iomem_resource, p, skip_children) {
1721 if (p->start >= addr + size)
1723 if (p->end < addr) {
1724 skip_children = true;
1727 skip_children = false;
1730 * IORESOURCE_SYSTEM_RAM resources are exclusive if
1731 * IORESOURCE_EXCLUSIVE is set, even if they
1732 * are not busy and even if "iomem=relaxed" is set. The
1733 * responsible driver dynamically adds/removes system RAM within
1734 * such an area and uncontrolled access is dangerous.
1736 if ((p->flags & exclusive_system_ram) == exclusive_system_ram) {
1742 * A resource is exclusive if IORESOURCE_EXCLUSIVE is set
1743 * or CONFIG_IO_STRICT_DEVMEM is enabled and the
1746 if (!strict_iomem_checks || !(p->flags & IORESOURCE_BUSY))
1748 if (IS_ENABLED(CONFIG_IO_STRICT_DEVMEM)
1749 || p->flags & IORESOURCE_EXCLUSIVE) {
1754 read_unlock(&resource_lock);
1759 struct resource_entry *resource_list_create_entry(struct resource *res,
1762 struct resource_entry *entry;
1764 entry = kzalloc(sizeof(*entry) + extra_size, GFP_KERNEL);
1766 INIT_LIST_HEAD(&entry->node);
1767 entry->res = res ? res : &entry->__res;
1772 EXPORT_SYMBOL(resource_list_create_entry);
1774 void resource_list_free(struct list_head *head)
1776 struct resource_entry *entry, *tmp;
1778 list_for_each_entry_safe(entry, tmp, head, node)
1779 resource_list_destroy_entry(entry);
1781 EXPORT_SYMBOL(resource_list_free);
1783 #ifdef CONFIG_DEVICE_PRIVATE
1784 static struct resource *__request_free_mem_region(struct device *dev,
1785 struct resource *base, unsigned long size, const char *name)
1787 resource_size_t end, addr;
1788 struct resource *res;
1789 struct region_devres *dr = NULL;
1791 size = ALIGN(size, 1UL << PA_SECTION_SHIFT);
1792 end = min_t(unsigned long, base->end, (1UL << MAX_PHYSMEM_BITS) - 1);
1793 addr = end - size + 1UL;
1795 res = alloc_resource(GFP_KERNEL);
1797 return ERR_PTR(-ENOMEM);
1800 dr = devres_alloc(devm_region_release,
1801 sizeof(struct region_devres), GFP_KERNEL);
1804 return ERR_PTR(-ENOMEM);
1808 write_lock(&resource_lock);
1809 for (; addr > size && addr >= base->start; addr -= size) {
1810 if (__region_intersects(addr, size, 0, IORES_DESC_NONE) !=
1814 if (__request_region_locked(res, &iomem_resource, addr, size,
1819 dr->parent = &iomem_resource;
1822 devres_add(dev, dr);
1825 res->desc = IORES_DESC_DEVICE_PRIVATE_MEMORY;
1826 write_unlock(&resource_lock);
1829 * A driver is claiming this region so revoke any mappings.
1834 write_unlock(&resource_lock);
1840 return ERR_PTR(-ERANGE);
1844 * devm_request_free_mem_region - find free region for device private memory
1846 * @dev: device struct to bind the resource to
1847 * @size: size in bytes of the device memory to add
1848 * @base: resource tree to look in
1850 * This function tries to find an empty range of physical address big enough to
1851 * contain the new resource, so that it can later be hotplugged as ZONE_DEVICE
1852 * memory, which in turn allocates struct pages.
1854 struct resource *devm_request_free_mem_region(struct device *dev,
1855 struct resource *base, unsigned long size)
1857 return __request_free_mem_region(dev, base, size, dev_name(dev));
1859 EXPORT_SYMBOL_GPL(devm_request_free_mem_region);
1861 struct resource *request_free_mem_region(struct resource *base,
1862 unsigned long size, const char *name)
1864 return __request_free_mem_region(NULL, base, size, name);
1866 EXPORT_SYMBOL_GPL(request_free_mem_region);
1868 #endif /* CONFIG_DEVICE_PRIVATE */
1870 static int __init strict_iomem(char *str)
1872 if (strstr(str, "relaxed"))
1873 strict_iomem_checks = 0;
1874 if (strstr(str, "strict"))
1875 strict_iomem_checks = 1;
1879 static int iomem_fs_init_fs_context(struct fs_context *fc)
1881 return init_pseudo(fc, DEVMEM_MAGIC) ? 0 : -ENOMEM;
1884 static struct file_system_type iomem_fs_type = {
1886 .owner = THIS_MODULE,
1887 .init_fs_context = iomem_fs_init_fs_context,
1888 .kill_sb = kill_anon_super,
1891 static int __init iomem_init_inode(void)
1893 static struct vfsmount *iomem_vfs_mount;
1894 static int iomem_fs_cnt;
1895 struct inode *inode;
1898 rc = simple_pin_fs(&iomem_fs_type, &iomem_vfs_mount, &iomem_fs_cnt);
1900 pr_err("Cannot mount iomem pseudo filesystem: %d\n", rc);
1904 inode = alloc_anon_inode(iomem_vfs_mount->mnt_sb);
1905 if (IS_ERR(inode)) {
1906 rc = PTR_ERR(inode);
1907 pr_err("Cannot allocate inode for iomem: %d\n", rc);
1908 simple_release_fs(&iomem_vfs_mount, &iomem_fs_cnt);
1913 * Publish iomem revocation inode initialized.
1914 * Pairs with smp_load_acquire() in revoke_iomem().
1916 smp_store_release(&iomem_inode, inode);
1921 fs_initcall(iomem_init_inode);
1923 __setup("iomem=", strict_iomem);