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, bool skip_children)
61 if (!skip_children && p->child)
63 while (!p->sibling && p->parent)
68 #define for_each_resource(_root, _p, _skip_children) \
69 for ((_p) = (_root)->child; (_p); (_p) = next_resource(_p, _skip_children))
73 enum { MAX_IORES_LEVEL = 5 };
75 static void *r_start(struct seq_file *m, loff_t *pos)
76 __acquires(resource_lock)
78 struct resource *root = pde_data(file_inode(m->file));
82 read_lock(&resource_lock);
83 for_each_resource(root, p, false) {
91 static void *r_next(struct seq_file *m, void *v, loff_t *pos)
93 struct resource *p = v;
97 return (void *)next_resource(p, false);
100 static void r_stop(struct seq_file *m, void *v)
101 __releases(resource_lock)
103 read_unlock(&resource_lock);
106 static int r_show(struct seq_file *m, void *v)
108 struct resource *root = pde_data(file_inode(m->file));
109 struct resource *r = v, *p;
110 unsigned long long start, end;
111 int width = root->end < 0x10000 ? 4 : 8;
114 for (depth = 0, p = r; depth < MAX_IORES_LEVEL; depth++, p = p->parent)
115 if (p->parent == root)
118 if (file_ns_capable(m->file, &init_user_ns, CAP_SYS_ADMIN)) {
125 seq_printf(m, "%*s%0*llx-%0*llx : %s\n",
129 r->name ? r->name : "<BAD>");
133 static const struct seq_operations resource_op = {
140 static int __init ioresources_init(void)
142 proc_create_seq_data("ioports", 0, NULL, &resource_op,
144 proc_create_seq_data("iomem", 0, NULL, &resource_op, &iomem_resource);
147 __initcall(ioresources_init);
149 #endif /* CONFIG_PROC_FS */
151 static void free_resource(struct resource *res)
154 * If the resource was allocated using memblock early during boot
155 * we'll leak it here: we can only return full pages back to the
156 * buddy and trying to be smart and reusing them eventually in
157 * alloc_resource() overcomplicates resource handling.
159 if (res && PageSlab(virt_to_head_page(res)))
163 static struct resource *alloc_resource(gfp_t flags)
165 return kzalloc(sizeof(struct resource), flags);
168 /* Return the conflict entry if you can't request it */
169 static struct resource * __request_resource(struct resource *root, struct resource *new)
171 resource_size_t start = new->start;
172 resource_size_t end = new->end;
173 struct resource *tmp, **p;
177 if (start < root->start)
184 if (!tmp || tmp->start > end) {
191 if (tmp->end < start)
197 static int __release_resource(struct resource *old, bool release_child)
199 struct resource *tmp, **p, *chd;
201 p = &old->parent->child;
207 if (release_child || !(tmp->child)) {
210 for (chd = tmp->child;; chd = chd->sibling) {
211 chd->parent = tmp->parent;
216 chd->sibling = tmp->sibling;
226 static void __release_child_resources(struct resource *r)
228 struct resource *tmp, *p;
229 resource_size_t size;
239 __release_child_resources(tmp);
241 printk(KERN_DEBUG "release child resource %pR\n", tmp);
242 /* need to restore size, and keep flags */
243 size = resource_size(tmp);
249 void release_child_resources(struct resource *r)
251 write_lock(&resource_lock);
252 __release_child_resources(r);
253 write_unlock(&resource_lock);
257 * request_resource_conflict - request and reserve an I/O or memory resource
258 * @root: root resource descriptor
259 * @new: resource descriptor desired by caller
261 * Returns 0 for success, conflict resource on error.
263 struct resource *request_resource_conflict(struct resource *root, struct resource *new)
265 struct resource *conflict;
267 write_lock(&resource_lock);
268 conflict = __request_resource(root, new);
269 write_unlock(&resource_lock);
274 * request_resource - request and reserve an I/O or memory resource
275 * @root: root resource descriptor
276 * @new: resource descriptor desired by caller
278 * Returns 0 for success, negative error code on error.
280 int request_resource(struct resource *root, struct resource *new)
282 struct resource *conflict;
284 conflict = request_resource_conflict(root, new);
285 return conflict ? -EBUSY : 0;
288 EXPORT_SYMBOL(request_resource);
291 * release_resource - release a previously reserved resource
292 * @old: resource pointer
294 int release_resource(struct resource *old)
298 write_lock(&resource_lock);
299 retval = __release_resource(old, true);
300 write_unlock(&resource_lock);
304 EXPORT_SYMBOL(release_resource);
307 * find_next_iomem_res - Finds the lowest iomem resource that covers part of
310 * If a resource is found, returns 0 and @*res is overwritten with the part
311 * of the resource that's within [@start..@end]; if none is found, returns
312 * -ENODEV. Returns -EINVAL for invalid parameters.
314 * @start: start address of the resource searched for
315 * @end: end address of same resource
316 * @flags: flags which the resource must have
317 * @desc: descriptor the resource must have
318 * @res: return ptr, if resource found
320 * The caller must specify @start, @end, @flags, and @desc
321 * (which may be IORES_DESC_NONE).
323 static int find_next_iomem_res(resource_size_t start, resource_size_t end,
324 unsigned long flags, unsigned long desc,
325 struct resource *res)
335 read_lock(&resource_lock);
337 for_each_resource(&iomem_resource, p, false) {
338 /* If we passed the resource we are looking for, stop */
339 if (p->start > end) {
344 /* Skip until we find a range that matches what we look for */
348 if ((p->flags & flags) != flags)
350 if ((desc != IORES_DESC_NONE) && (desc != p->desc))
353 /* Found a match, break */
359 *res = (struct resource) {
360 .start = max(start, p->start),
361 .end = min(end, p->end),
368 read_unlock(&resource_lock);
369 return p ? 0 : -ENODEV;
372 static int __walk_iomem_res_desc(resource_size_t start, resource_size_t end,
373 unsigned long flags, unsigned long desc,
375 int (*func)(struct resource *, void *))
380 while (start < end &&
381 !find_next_iomem_res(start, end, flags, desc, &res)) {
382 ret = (*func)(&res, arg);
393 * walk_iomem_res_desc - Walks through iomem resources and calls func()
394 * with matching resource ranges.
396 * @desc: I/O resource descriptor. Use IORES_DESC_NONE to skip @desc check.
397 * @flags: I/O resource flags
400 * @arg: function argument for the callback @func
401 * @func: callback function that is called for each qualifying resource area
403 * All the memory ranges which overlap start,end and also match flags and
404 * desc are valid candidates.
406 * NOTE: For a new descriptor search, define a new IORES_DESC in
407 * <linux/ioport.h> and set it in 'desc' of a target resource entry.
409 int walk_iomem_res_desc(unsigned long desc, unsigned long flags, u64 start,
410 u64 end, void *arg, int (*func)(struct resource *, void *))
412 return __walk_iomem_res_desc(start, end, flags, desc, arg, func);
414 EXPORT_SYMBOL_GPL(walk_iomem_res_desc);
417 * This function calls the @func callback against all memory ranges of type
418 * System RAM which are marked as IORESOURCE_SYSTEM_RAM and IORESOUCE_BUSY.
419 * Now, this function is only for System RAM, it deals with full ranges and
420 * not PFNs. If resources are not PFN-aligned, dealing with PFNs can truncate
423 int walk_system_ram_res(u64 start, u64 end, void *arg,
424 int (*func)(struct resource *, void *))
426 unsigned long flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
428 return __walk_iomem_res_desc(start, end, flags, IORES_DESC_NONE, arg,
433 * This function calls the @func callback against all memory ranges, which
434 * are ranges marked as IORESOURCE_MEM and IORESOUCE_BUSY.
436 int walk_mem_res(u64 start, u64 end, void *arg,
437 int (*func)(struct resource *, void *))
439 unsigned long flags = IORESOURCE_MEM | IORESOURCE_BUSY;
441 return __walk_iomem_res_desc(start, end, flags, IORES_DESC_NONE, arg,
446 * This function calls the @func callback against all memory ranges of type
447 * System RAM which are marked as IORESOURCE_SYSTEM_RAM and IORESOUCE_BUSY.
448 * It is to be used only for System RAM.
450 int walk_system_ram_range(unsigned long start_pfn, unsigned long nr_pages,
451 void *arg, int (*func)(unsigned long, unsigned long, void *))
453 resource_size_t start, end;
456 unsigned long pfn, end_pfn;
459 start = (u64) start_pfn << PAGE_SHIFT;
460 end = ((u64)(start_pfn + nr_pages) << PAGE_SHIFT) - 1;
461 flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
462 while (start < end &&
463 !find_next_iomem_res(start, end, flags, IORES_DESC_NONE, &res)) {
464 pfn = PFN_UP(res.start);
465 end_pfn = PFN_DOWN(res.end + 1);
467 ret = (*func)(pfn, end_pfn - pfn, arg);
475 static int __is_ram(unsigned long pfn, unsigned long nr_pages, void *arg)
481 * This generic page_is_ram() returns true if specified address is
482 * registered as System RAM in iomem_resource list.
484 int __weak page_is_ram(unsigned long pfn)
486 return walk_system_ram_range(pfn, 1, NULL, __is_ram) == 1;
488 EXPORT_SYMBOL_GPL(page_is_ram);
490 static int __region_intersects(struct resource *parent, resource_size_t start,
491 size_t size, unsigned long flags,
495 int type = 0; int other = 0;
499 res.end = start + size - 1;
501 for (p = parent->child; p ; p = p->sibling) {
502 bool is_type = (((p->flags & flags) == flags) &&
503 ((desc == IORES_DESC_NONE) ||
506 if (resource_overlaps(p, &res))
507 is_type ? type++ : other++;
511 return REGION_DISJOINT;
514 return REGION_INTERSECTS;
520 * region_intersects() - determine intersection of region with known resources
521 * @start: region start address
522 * @size: size of region
523 * @flags: flags of resource (in iomem_resource)
524 * @desc: descriptor of resource (in iomem_resource) or IORES_DESC_NONE
526 * Check if the specified region partially overlaps or fully eclipses a
527 * resource identified by @flags and @desc (optional with IORES_DESC_NONE).
528 * Return REGION_DISJOINT if the region does not overlap @flags/@desc,
529 * return REGION_MIXED if the region overlaps @flags/@desc and another
530 * resource, and return REGION_INTERSECTS if the region overlaps @flags/@desc
531 * and no other defined resource. Note that REGION_INTERSECTS is also
532 * returned in the case when the specified region overlaps RAM and undefined
535 * region_intersect() is used by memory remapping functions to ensure
536 * the user is not remapping RAM and is a vast speed up over walking
537 * through the resource table page by page.
539 int region_intersects(resource_size_t start, size_t size, unsigned long flags,
544 read_lock(&resource_lock);
545 ret = __region_intersects(&iomem_resource, start, size, flags, desc);
546 read_unlock(&resource_lock);
550 EXPORT_SYMBOL_GPL(region_intersects);
552 void __weak arch_remove_reservations(struct resource *avail)
556 static resource_size_t simple_align_resource(void *data,
557 const struct resource *avail,
558 resource_size_t size,
559 resource_size_t align)
564 static void resource_clip(struct resource *res, resource_size_t min,
567 if (res->start < min)
574 * Find empty slot in the resource tree with the given range and
575 * alignment constraints
577 static int __find_resource(struct resource *root, struct resource *old,
578 struct resource *new,
579 resource_size_t size,
580 struct resource_constraint *constraint)
582 struct resource *this = root->child;
583 struct resource tmp = *new, avail, alloc;
585 tmp.start = root->start;
587 * Skip past an allocated resource that starts at 0, since the assignment
588 * of this->start - 1 to tmp->end below would cause an underflow.
590 if (this && this->start == root->start) {
591 tmp.start = (this == old) ? old->start : this->end + 1;
592 this = this->sibling;
596 tmp.end = (this == old) ? this->end : this->start - 1;
600 if (tmp.end < tmp.start)
603 resource_clip(&tmp, constraint->min, constraint->max);
604 arch_remove_reservations(&tmp);
606 /* Check for overflow after ALIGN() */
607 avail.start = ALIGN(tmp.start, constraint->align);
609 avail.flags = new->flags & ~IORESOURCE_UNSET;
610 if (avail.start >= tmp.start) {
611 alloc.flags = avail.flags;
612 alloc.start = constraint->alignf(constraint->alignf_data, &avail,
613 size, constraint->align);
614 alloc.end = alloc.start + size - 1;
615 if (alloc.start <= alloc.end &&
616 resource_contains(&avail, &alloc)) {
617 new->start = alloc.start;
618 new->end = alloc.end;
623 next: if (!this || this->end == root->end)
627 tmp.start = this->end + 1;
628 this = this->sibling;
634 * Find empty slot in the resource tree given range and alignment.
636 static int find_resource(struct resource *root, struct resource *new,
637 resource_size_t size,
638 struct resource_constraint *constraint)
640 return __find_resource(root, NULL, new, size, constraint);
644 * reallocate_resource - allocate a slot in the resource tree given range & alignment.
645 * The resource will be relocated if the new size cannot be reallocated in the
648 * @root: root resource descriptor
649 * @old: resource descriptor desired by caller
650 * @newsize: new size of the resource descriptor
651 * @constraint: the size and alignment constraints to be met.
653 static int reallocate_resource(struct resource *root, struct resource *old,
654 resource_size_t newsize,
655 struct resource_constraint *constraint)
658 struct resource new = *old;
659 struct resource *conflict;
661 write_lock(&resource_lock);
663 if ((err = __find_resource(root, old, &new, newsize, constraint)))
666 if (resource_contains(&new, old)) {
667 old->start = new.start;
677 if (resource_contains(old, &new)) {
678 old->start = new.start;
681 __release_resource(old, true);
683 conflict = __request_resource(root, old);
687 write_unlock(&resource_lock);
693 * allocate_resource - allocate empty slot in the resource tree given range & alignment.
694 * The resource will be reallocated with a new size if it was already allocated
695 * @root: root resource descriptor
696 * @new: resource descriptor desired by caller
697 * @size: requested resource region size
698 * @min: minimum boundary to allocate
699 * @max: maximum boundary to allocate
700 * @align: alignment requested, in bytes
701 * @alignf: alignment function, optional, called if not NULL
702 * @alignf_data: arbitrary data to pass to the @alignf function
704 int allocate_resource(struct resource *root, struct resource *new,
705 resource_size_t size, resource_size_t min,
706 resource_size_t max, resource_size_t align,
707 resource_size_t (*alignf)(void *,
708 const struct resource *,
714 struct resource_constraint constraint;
717 alignf = simple_align_resource;
719 constraint.min = min;
720 constraint.max = max;
721 constraint.align = align;
722 constraint.alignf = alignf;
723 constraint.alignf_data = alignf_data;
726 /* resource is already allocated, try reallocating with
727 the new constraints */
728 return reallocate_resource(root, new, size, &constraint);
731 write_lock(&resource_lock);
732 err = find_resource(root, new, size, &constraint);
733 if (err >= 0 && __request_resource(root, new))
735 write_unlock(&resource_lock);
739 EXPORT_SYMBOL(allocate_resource);
742 * lookup_resource - find an existing resource by a resource start address
743 * @root: root resource descriptor
744 * @start: resource start address
746 * Returns a pointer to the resource if found, NULL otherwise
748 struct resource *lookup_resource(struct resource *root, resource_size_t start)
750 struct resource *res;
752 read_lock(&resource_lock);
753 for (res = root->child; res; res = res->sibling) {
754 if (res->start == start)
757 read_unlock(&resource_lock);
763 * Insert a resource into the resource tree. If successful, return NULL,
764 * otherwise return the conflicting resource (compare to __request_resource())
766 static struct resource * __insert_resource(struct resource *parent, struct resource *new)
768 struct resource *first, *next;
770 for (;; parent = first) {
771 first = __request_resource(parent, new);
777 if (WARN_ON(first == new)) /* duplicated insertion */
780 if ((first->start > new->start) || (first->end < new->end))
782 if ((first->start == new->start) && (first->end == new->end))
786 for (next = first; ; next = next->sibling) {
787 /* Partial overlap? Bad, and unfixable */
788 if (next->start < new->start || next->end > new->end)
792 if (next->sibling->start > new->end)
796 new->parent = parent;
797 new->sibling = next->sibling;
800 next->sibling = NULL;
801 for (next = first; next; next = next->sibling)
804 if (parent->child == first) {
807 next = parent->child;
808 while (next->sibling != first)
809 next = next->sibling;
816 * insert_resource_conflict - Inserts resource in the resource tree
817 * @parent: parent of the new resource
818 * @new: new resource to insert
820 * Returns 0 on success, conflict resource if the resource can't be inserted.
822 * This function is equivalent to request_resource_conflict when no conflict
823 * happens. If a conflict happens, and the conflicting resources
824 * entirely fit within the range of the new resource, then the new
825 * resource is inserted and the conflicting resources become children of
828 * This function is intended for producers of resources, such as FW modules
831 struct resource *insert_resource_conflict(struct resource *parent, struct resource *new)
833 struct resource *conflict;
835 write_lock(&resource_lock);
836 conflict = __insert_resource(parent, new);
837 write_unlock(&resource_lock);
842 * insert_resource - Inserts a resource in the resource tree
843 * @parent: parent of the new resource
844 * @new: new resource to insert
846 * Returns 0 on success, -EBUSY if the resource can't be inserted.
848 * This function is intended for producers of resources, such as FW modules
851 int insert_resource(struct resource *parent, struct resource *new)
853 struct resource *conflict;
855 conflict = insert_resource_conflict(parent, new);
856 return conflict ? -EBUSY : 0;
858 EXPORT_SYMBOL_GPL(insert_resource);
861 * insert_resource_expand_to_fit - Insert a resource into the resource tree
862 * @root: root resource descriptor
863 * @new: new resource to insert
865 * Insert a resource into the resource tree, possibly expanding it in order
866 * to make it encompass any conflicting resources.
868 void insert_resource_expand_to_fit(struct resource *root, struct resource *new)
873 write_lock(&resource_lock);
875 struct resource *conflict;
877 conflict = __insert_resource(root, new);
880 if (conflict == root)
883 /* Ok, expand resource to cover the conflict, then try again .. */
884 if (conflict->start < new->start)
885 new->start = conflict->start;
886 if (conflict->end > new->end)
887 new->end = conflict->end;
889 pr_info("Expanded resource %s due to conflict with %s\n", new->name, conflict->name);
891 write_unlock(&resource_lock);
894 * Not for general consumption, only early boot memory map parsing, PCI
895 * resource discovery, and late discovery of CXL resources are expected
896 * to use this interface. The former are built-in and only the latter,
899 EXPORT_SYMBOL_NS_GPL(insert_resource_expand_to_fit, CXL);
902 * remove_resource - Remove a resource in the resource tree
903 * @old: resource to remove
905 * Returns 0 on success, -EINVAL if the resource is not valid.
907 * This function removes a resource previously inserted by insert_resource()
908 * or insert_resource_conflict(), and moves the children (if any) up to
909 * where they were before. insert_resource() and insert_resource_conflict()
910 * insert a new resource, and move any conflicting resources down to the
911 * children of the new resource.
913 * insert_resource(), insert_resource_conflict() and remove_resource() are
914 * intended for producers of resources, such as FW modules and bus drivers.
916 int remove_resource(struct resource *old)
920 write_lock(&resource_lock);
921 retval = __release_resource(old, false);
922 write_unlock(&resource_lock);
925 EXPORT_SYMBOL_GPL(remove_resource);
927 static int __adjust_resource(struct resource *res, resource_size_t start,
928 resource_size_t size)
930 struct resource *tmp, *parent = res->parent;
931 resource_size_t end = start + size - 1;
937 if ((start < parent->start) || (end > parent->end))
940 if (res->sibling && (res->sibling->start <= end))
945 while (tmp->sibling != res)
947 if (start <= tmp->end)
952 for (tmp = res->child; tmp; tmp = tmp->sibling)
953 if ((tmp->start < start) || (tmp->end > end))
965 * adjust_resource - modify a resource's start and size
966 * @res: resource to modify
967 * @start: new start value
970 * Given an existing resource, change its start and size to match the
971 * arguments. Returns 0 on success, -EBUSY if it can't fit.
972 * Existing children of the resource are assumed to be immutable.
974 int adjust_resource(struct resource *res, resource_size_t start,
975 resource_size_t size)
979 write_lock(&resource_lock);
980 result = __adjust_resource(res, start, size);
981 write_unlock(&resource_lock);
984 EXPORT_SYMBOL(adjust_resource);
987 __reserve_region_with_split(struct resource *root, resource_size_t start,
988 resource_size_t end, const char *name)
990 struct resource *parent = root;
991 struct resource *conflict;
992 struct resource *res = alloc_resource(GFP_ATOMIC);
993 struct resource *next_res = NULL;
994 int type = resource_type(root);
1002 res->flags = type | IORESOURCE_BUSY;
1003 res->desc = IORES_DESC_NONE;
1007 conflict = __request_resource(parent, res);
1016 /* conflict covered whole area */
1017 if (conflict->start <= res->start &&
1018 conflict->end >= res->end) {
1024 /* failed, split and try again */
1025 if (conflict->start > res->start) {
1027 res->end = conflict->start - 1;
1028 if (conflict->end < end) {
1029 next_res = alloc_resource(GFP_ATOMIC);
1034 next_res->name = name;
1035 next_res->start = conflict->end + 1;
1036 next_res->end = end;
1037 next_res->flags = type | IORESOURCE_BUSY;
1038 next_res->desc = IORES_DESC_NONE;
1041 res->start = conflict->end + 1;
1048 reserve_region_with_split(struct resource *root, resource_size_t start,
1049 resource_size_t end, const char *name)
1053 write_lock(&resource_lock);
1054 if (root->start > start || root->end < end) {
1055 pr_err("requested range [0x%llx-0x%llx] not in root %pr\n",
1056 (unsigned long long)start, (unsigned long long)end,
1058 if (start > root->end || end < root->start)
1061 if (end > root->end)
1063 if (start < root->start)
1064 start = root->start;
1065 pr_err("fixing request to [0x%llx-0x%llx]\n",
1066 (unsigned long long)start,
1067 (unsigned long long)end);
1072 __reserve_region_with_split(root, start, end, name);
1073 write_unlock(&resource_lock);
1077 * resource_alignment - calculate resource's alignment
1078 * @res: resource pointer
1080 * Returns alignment on success, 0 (invalid alignment) on failure.
1082 resource_size_t resource_alignment(struct resource *res)
1084 switch (res->flags & (IORESOURCE_SIZEALIGN | IORESOURCE_STARTALIGN)) {
1085 case IORESOURCE_SIZEALIGN:
1086 return resource_size(res);
1087 case IORESOURCE_STARTALIGN:
1095 * This is compatibility stuff for IO resources.
1097 * Note how this, unlike the above, knows about
1098 * the IO flag meanings (busy etc).
1100 * request_region creates a new busy region.
1102 * release_region releases a matching busy region.
1105 static DECLARE_WAIT_QUEUE_HEAD(muxed_resource_wait);
1107 static struct inode *iomem_inode;
1109 #ifdef CONFIG_IO_STRICT_DEVMEM
1110 static void revoke_iomem(struct resource *res)
1112 /* pairs with smp_store_release() in iomem_init_inode() */
1113 struct inode *inode = smp_load_acquire(&iomem_inode);
1116 * Check that the initialization has completed. Losing the race
1117 * is ok because it means drivers are claiming resources before
1118 * the fs_initcall level of init and prevent iomem_get_mapping users
1119 * from establishing mappings.
1125 * The expectation is that the driver has successfully marked
1126 * the resource busy by this point, so devmem_is_allowed()
1127 * should start returning false, however for performance this
1128 * does not iterate the entire resource range.
1130 if (devmem_is_allowed(PHYS_PFN(res->start)) &&
1131 devmem_is_allowed(PHYS_PFN(res->end))) {
1133 * *cringe* iomem=relaxed says "go ahead, what's the
1134 * worst that can happen?"
1139 unmap_mapping_range(inode->i_mapping, res->start, resource_size(res), 1);
1142 static void revoke_iomem(struct resource *res) {}
1145 struct address_space *iomem_get_mapping(void)
1148 * This function is only called from file open paths, hence guaranteed
1149 * that fs_initcalls have completed and no need to check for NULL. But
1150 * since revoke_iomem can be called before the initcall we still need
1151 * the barrier to appease checkers.
1153 return smp_load_acquire(&iomem_inode)->i_mapping;
1156 static int __request_region_locked(struct resource *res, struct resource *parent,
1157 resource_size_t start, resource_size_t n,
1158 const char *name, int flags)
1160 DECLARE_WAITQUEUE(wait, current);
1164 res->end = start + n - 1;
1167 struct resource *conflict;
1169 res->flags = resource_type(parent) | resource_ext_type(parent);
1170 res->flags |= IORESOURCE_BUSY | flags;
1171 res->desc = parent->desc;
1173 conflict = __request_resource(parent, res);
1177 * mm/hmm.c reserves physical addresses which then
1178 * become unavailable to other users. Conflicts are
1179 * not expected. Warn to aid debugging if encountered.
1181 if (conflict->desc == IORES_DESC_DEVICE_PRIVATE_MEMORY) {
1182 pr_warn("Unaddressable device %s %pR conflicts with %pR",
1183 conflict->name, conflict, res);
1185 if (conflict != parent) {
1186 if (!(conflict->flags & IORESOURCE_BUSY)) {
1191 if (conflict->flags & flags & IORESOURCE_MUXED) {
1192 add_wait_queue(&muxed_resource_wait, &wait);
1193 write_unlock(&resource_lock);
1194 set_current_state(TASK_UNINTERRUPTIBLE);
1196 remove_wait_queue(&muxed_resource_wait, &wait);
1197 write_lock(&resource_lock);
1200 /* Uhhuh, that didn't work out.. */
1208 * __request_region - create a new busy resource region
1209 * @parent: parent resource descriptor
1210 * @start: resource start address
1211 * @n: resource region size
1212 * @name: reserving caller's ID string
1213 * @flags: IO resource flags
1215 struct resource *__request_region(struct resource *parent,
1216 resource_size_t start, resource_size_t n,
1217 const char *name, int flags)
1219 struct resource *res = alloc_resource(GFP_KERNEL);
1225 write_lock(&resource_lock);
1226 ret = __request_region_locked(res, parent, start, n, name, flags);
1227 write_unlock(&resource_lock);
1234 if (parent == &iomem_resource)
1239 EXPORT_SYMBOL(__request_region);
1242 * __release_region - release a previously reserved resource region
1243 * @parent: parent resource descriptor
1244 * @start: resource start address
1245 * @n: resource region size
1247 * The described resource region must match a currently busy region.
1249 void __release_region(struct resource *parent, resource_size_t start,
1252 struct resource **p;
1253 resource_size_t end;
1256 end = start + n - 1;
1258 write_lock(&resource_lock);
1261 struct resource *res = *p;
1265 if (res->start <= start && res->end >= end) {
1266 if (!(res->flags & IORESOURCE_BUSY)) {
1270 if (res->start != start || res->end != end)
1273 write_unlock(&resource_lock);
1274 if (res->flags & IORESOURCE_MUXED)
1275 wake_up(&muxed_resource_wait);
1282 write_unlock(&resource_lock);
1284 pr_warn("Trying to free nonexistent resource <%pa-%pa>\n", &start, &end);
1286 EXPORT_SYMBOL(__release_region);
1288 #ifdef CONFIG_MEMORY_HOTREMOVE
1290 * release_mem_region_adjustable - release a previously reserved memory region
1291 * @start: resource start address
1292 * @size: resource region size
1294 * This interface is intended for memory hot-delete. The requested region
1295 * is released from a currently busy memory resource. The requested region
1296 * must either match exactly or fit into a single busy resource entry. In
1297 * the latter case, the remaining resource is adjusted accordingly.
1298 * Existing children of the busy memory resource must be immutable in the
1302 * - Additional release conditions, such as overlapping region, can be
1303 * supported after they are confirmed as valid cases.
1304 * - When a busy memory resource gets split into two entries, the code
1305 * assumes that all children remain in the lower address entry for
1306 * simplicity. Enhance this logic when necessary.
1308 void release_mem_region_adjustable(resource_size_t start, resource_size_t size)
1310 struct resource *parent = &iomem_resource;
1311 struct resource *new_res = NULL;
1312 bool alloc_nofail = false;
1313 struct resource **p;
1314 struct resource *res;
1315 resource_size_t end;
1317 end = start + size - 1;
1318 if (WARN_ON_ONCE((start < parent->start) || (end > parent->end)))
1322 * We free up quite a lot of memory on memory hotunplug (esp., memap),
1323 * just before releasing the region. This is highly unlikely to
1324 * fail - let's play save and make it never fail as the caller cannot
1325 * perform any error handling (e.g., trying to re-add memory will fail
1329 new_res = alloc_resource(GFP_KERNEL | (alloc_nofail ? __GFP_NOFAIL : 0));
1332 write_lock(&resource_lock);
1334 while ((res = *p)) {
1335 if (res->start >= end)
1338 /* look for the next resource if it does not fit into */
1339 if (res->start > start || res->end < end) {
1344 if (!(res->flags & IORESOURCE_MEM))
1347 if (!(res->flags & IORESOURCE_BUSY)) {
1352 /* found the target resource; let's adjust accordingly */
1353 if (res->start == start && res->end == end) {
1354 /* free the whole entry */
1357 } else if (res->start == start && res->end != end) {
1358 /* adjust the start */
1359 WARN_ON_ONCE(__adjust_resource(res, end + 1,
1361 } else if (res->start != start && res->end == end) {
1362 /* adjust the end */
1363 WARN_ON_ONCE(__adjust_resource(res, res->start,
1364 start - res->start));
1366 /* split into two entries - we need a new resource */
1368 new_res = alloc_resource(GFP_ATOMIC);
1370 alloc_nofail = true;
1371 write_unlock(&resource_lock);
1375 new_res->name = res->name;
1376 new_res->start = end + 1;
1377 new_res->end = res->end;
1378 new_res->flags = res->flags;
1379 new_res->desc = res->desc;
1380 new_res->parent = res->parent;
1381 new_res->sibling = res->sibling;
1382 new_res->child = NULL;
1384 if (WARN_ON_ONCE(__adjust_resource(res, res->start,
1385 start - res->start)))
1387 res->sibling = new_res;
1394 write_unlock(&resource_lock);
1395 free_resource(new_res);
1397 #endif /* CONFIG_MEMORY_HOTREMOVE */
1399 #ifdef CONFIG_MEMORY_HOTPLUG
1400 static bool system_ram_resources_mergeable(struct resource *r1,
1401 struct resource *r2)
1403 /* We assume either r1 or r2 is IORESOURCE_SYSRAM_MERGEABLE. */
1404 return r1->flags == r2->flags && r1->end + 1 == r2->start &&
1405 r1->name == r2->name && r1->desc == r2->desc &&
1406 !r1->child && !r2->child;
1410 * merge_system_ram_resource - mark the System RAM resource mergeable and try to
1411 * merge it with adjacent, mergeable resources
1412 * @res: resource descriptor
1414 * This interface is intended for memory hotplug, whereby lots of contiguous
1415 * system ram resources are added (e.g., via add_memory*()) by a driver, and
1416 * the actual resource boundaries are not of interest (e.g., it might be
1417 * relevant for DIMMs). Only resources that are marked mergeable, that have the
1418 * same parent, and that don't have any children are considered. All mergeable
1419 * resources must be immutable during the request.
1422 * - The caller has to make sure that no pointers to resources that are
1423 * marked mergeable are used anymore after this call - the resource might
1424 * be freed and the pointer might be stale!
1425 * - release_mem_region_adjustable() will split on demand on memory hotunplug
1427 void merge_system_ram_resource(struct resource *res)
1429 const unsigned long flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
1430 struct resource *cur;
1432 if (WARN_ON_ONCE((res->flags & flags) != flags))
1435 write_lock(&resource_lock);
1436 res->flags |= IORESOURCE_SYSRAM_MERGEABLE;
1438 /* Try to merge with next item in the list. */
1440 if (cur && system_ram_resources_mergeable(res, cur)) {
1441 res->end = cur->end;
1442 res->sibling = cur->sibling;
1446 /* Try to merge with previous item in the list. */
1447 cur = res->parent->child;
1448 while (cur && cur->sibling != res)
1450 if (cur && system_ram_resources_mergeable(cur, res)) {
1451 cur->end = res->end;
1452 cur->sibling = res->sibling;
1455 write_unlock(&resource_lock);
1457 #endif /* CONFIG_MEMORY_HOTPLUG */
1460 * Managed region resource
1462 static void devm_resource_release(struct device *dev, void *ptr)
1464 struct resource **r = ptr;
1466 release_resource(*r);
1470 * devm_request_resource() - request and reserve an I/O or memory resource
1471 * @dev: device for which to request the resource
1472 * @root: root of the resource tree from which to request the resource
1473 * @new: descriptor of the resource to request
1475 * This is a device-managed version of request_resource(). There is usually
1476 * no need to release resources requested by this function explicitly since
1477 * that will be taken care of when the device is unbound from its driver.
1478 * If for some reason the resource needs to be released explicitly, because
1479 * of ordering issues for example, drivers must call devm_release_resource()
1480 * rather than the regular release_resource().
1482 * When a conflict is detected between any existing resources and the newly
1483 * requested resource, an error message will be printed.
1485 * Returns 0 on success or a negative error code on failure.
1487 int devm_request_resource(struct device *dev, struct resource *root,
1488 struct resource *new)
1490 struct resource *conflict, **ptr;
1492 ptr = devres_alloc(devm_resource_release, sizeof(*ptr), GFP_KERNEL);
1498 conflict = request_resource_conflict(root, new);
1500 dev_err(dev, "resource collision: %pR conflicts with %s %pR\n",
1501 new, conflict->name, conflict);
1506 devres_add(dev, ptr);
1509 EXPORT_SYMBOL(devm_request_resource);
1511 static int devm_resource_match(struct device *dev, void *res, void *data)
1513 struct resource **ptr = res;
1515 return *ptr == data;
1519 * devm_release_resource() - release a previously requested resource
1520 * @dev: device for which to release the resource
1521 * @new: descriptor of the resource to release
1523 * Releases a resource previously requested using devm_request_resource().
1525 void devm_release_resource(struct device *dev, struct resource *new)
1527 WARN_ON(devres_release(dev, devm_resource_release, devm_resource_match,
1530 EXPORT_SYMBOL(devm_release_resource);
1532 struct region_devres {
1533 struct resource *parent;
1534 resource_size_t start;
1538 static void devm_region_release(struct device *dev, void *res)
1540 struct region_devres *this = res;
1542 __release_region(this->parent, this->start, this->n);
1545 static int devm_region_match(struct device *dev, void *res, void *match_data)
1547 struct region_devres *this = res, *match = match_data;
1549 return this->parent == match->parent &&
1550 this->start == match->start && this->n == match->n;
1554 __devm_request_region(struct device *dev, struct resource *parent,
1555 resource_size_t start, resource_size_t n, const char *name)
1557 struct region_devres *dr = NULL;
1558 struct resource *res;
1560 dr = devres_alloc(devm_region_release, sizeof(struct region_devres),
1565 dr->parent = parent;
1569 res = __request_region(parent, start, n, name, 0);
1571 devres_add(dev, dr);
1577 EXPORT_SYMBOL(__devm_request_region);
1579 void __devm_release_region(struct device *dev, struct resource *parent,
1580 resource_size_t start, resource_size_t n)
1582 struct region_devres match_data = { parent, start, n };
1584 __release_region(parent, start, n);
1585 WARN_ON(devres_destroy(dev, devm_region_release, devm_region_match,
1588 EXPORT_SYMBOL(__devm_release_region);
1591 * Reserve I/O ports or memory based on "reserve=" kernel parameter.
1593 #define MAXRESERVE 4
1594 static int __init reserve_setup(char *str)
1596 static int reserved;
1597 static struct resource reserve[MAXRESERVE];
1600 unsigned int io_start, io_num;
1602 struct resource *parent;
1604 if (get_option(&str, &io_start) != 2)
1606 if (get_option(&str, &io_num) == 0)
1608 if (x < MAXRESERVE) {
1609 struct resource *res = reserve + x;
1612 * If the region starts below 0x10000, we assume it's
1613 * I/O port space; otherwise assume it's memory.
1615 if (io_start < 0x10000) {
1616 res->flags = IORESOURCE_IO;
1617 parent = &ioport_resource;
1619 res->flags = IORESOURCE_MEM;
1620 parent = &iomem_resource;
1622 res->name = "reserved";
1623 res->start = io_start;
1624 res->end = io_start + io_num - 1;
1625 res->flags |= IORESOURCE_BUSY;
1626 res->desc = IORES_DESC_NONE;
1628 if (request_resource(parent, res) == 0)
1634 __setup("reserve=", reserve_setup);
1637 * Check if the requested addr and size spans more than any slot in the
1638 * iomem resource tree.
1640 int iomem_map_sanity_check(resource_size_t addr, unsigned long size)
1642 resource_size_t end = addr + size - 1;
1646 read_lock(&resource_lock);
1647 for_each_resource(&iomem_resource, p, false) {
1649 * We can probably skip the resources without
1650 * IORESOURCE_IO attribute?
1656 if (PFN_DOWN(p->start) <= PFN_DOWN(addr) &&
1657 PFN_DOWN(p->end) >= PFN_DOWN(end))
1660 * if a resource is "BUSY", it's not a hardware resource
1661 * but a driver mapping of such a resource; we don't want
1662 * to warn for those; some drivers legitimately map only
1663 * partial hardware resources. (example: vesafb)
1665 if (p->flags & IORESOURCE_BUSY)
1668 pr_warn("resource sanity check: requesting [mem %pa-%pa], which spans more than %s %pR\n",
1669 &addr, &end, p->name, p);
1673 read_unlock(&resource_lock);
1678 #ifdef CONFIG_STRICT_DEVMEM
1679 static int strict_iomem_checks = 1;
1681 static int strict_iomem_checks;
1685 * Check if an address is exclusive to the kernel and must not be mapped to
1686 * user space, for example, via /dev/mem.
1688 * Returns true if exclusive to the kernel, otherwise returns false.
1690 bool resource_is_exclusive(struct resource *root, u64 addr, resource_size_t size)
1692 const unsigned int exclusive_system_ram = IORESOURCE_SYSTEM_RAM |
1693 IORESOURCE_EXCLUSIVE;
1694 bool skip_children = false, err = false;
1697 read_lock(&resource_lock);
1698 for_each_resource(root, p, skip_children) {
1699 if (p->start >= addr + size)
1701 if (p->end < addr) {
1702 skip_children = true;
1705 skip_children = false;
1708 * IORESOURCE_SYSTEM_RAM resources are exclusive if
1709 * IORESOURCE_EXCLUSIVE is set, even if they
1710 * are not busy and even if "iomem=relaxed" is set. The
1711 * responsible driver dynamically adds/removes system RAM within
1712 * such an area and uncontrolled access is dangerous.
1714 if ((p->flags & exclusive_system_ram) == exclusive_system_ram) {
1720 * A resource is exclusive if IORESOURCE_EXCLUSIVE is set
1721 * or CONFIG_IO_STRICT_DEVMEM is enabled and the
1724 if (!strict_iomem_checks || !(p->flags & IORESOURCE_BUSY))
1726 if (IS_ENABLED(CONFIG_IO_STRICT_DEVMEM)
1727 || p->flags & IORESOURCE_EXCLUSIVE) {
1732 read_unlock(&resource_lock);
1737 bool iomem_is_exclusive(u64 addr)
1739 return resource_is_exclusive(&iomem_resource, addr & PAGE_MASK,
1743 struct resource_entry *resource_list_create_entry(struct resource *res,
1746 struct resource_entry *entry;
1748 entry = kzalloc(sizeof(*entry) + extra_size, GFP_KERNEL);
1750 INIT_LIST_HEAD(&entry->node);
1751 entry->res = res ? res : &entry->__res;
1756 EXPORT_SYMBOL(resource_list_create_entry);
1758 void resource_list_free(struct list_head *head)
1760 struct resource_entry *entry, *tmp;
1762 list_for_each_entry_safe(entry, tmp, head, node)
1763 resource_list_destroy_entry(entry);
1765 EXPORT_SYMBOL(resource_list_free);
1767 #ifdef CONFIG_GET_FREE_REGION
1768 #define GFR_DESCENDING (1UL << 0)
1769 #define GFR_REQUEST_REGION (1UL << 1)
1770 #define GFR_DEFAULT_ALIGN (1UL << PA_SECTION_SHIFT)
1772 static resource_size_t gfr_start(struct resource *base, resource_size_t size,
1773 resource_size_t align, unsigned long flags)
1775 if (flags & GFR_DESCENDING) {
1776 resource_size_t end;
1778 end = min_t(resource_size_t, base->end,
1779 (1ULL << MAX_PHYSMEM_BITS) - 1);
1780 return end - size + 1;
1783 return ALIGN(base->start, align);
1786 static bool gfr_continue(struct resource *base, resource_size_t addr,
1787 resource_size_t size, unsigned long flags)
1789 if (flags & GFR_DESCENDING)
1790 return addr > size && addr >= base->start;
1792 * In the ascend case be careful that the last increment by
1793 * @size did not wrap 0.
1795 return addr > addr - size &&
1796 addr <= min_t(resource_size_t, base->end,
1797 (1ULL << MAX_PHYSMEM_BITS) - 1);
1800 static resource_size_t gfr_next(resource_size_t addr, resource_size_t size,
1801 unsigned long flags)
1803 if (flags & GFR_DESCENDING)
1808 static void remove_free_mem_region(void *_res)
1810 struct resource *res = _res;
1813 remove_resource(res);
1817 static struct resource *
1818 get_free_mem_region(struct device *dev, struct resource *base,
1819 resource_size_t size, const unsigned long align,
1820 const char *name, const unsigned long desc,
1821 const unsigned long flags)
1823 resource_size_t addr;
1824 struct resource *res;
1825 struct region_devres *dr = NULL;
1827 size = ALIGN(size, align);
1829 res = alloc_resource(GFP_KERNEL);
1831 return ERR_PTR(-ENOMEM);
1833 if (dev && (flags & GFR_REQUEST_REGION)) {
1834 dr = devres_alloc(devm_region_release,
1835 sizeof(struct region_devres), GFP_KERNEL);
1838 return ERR_PTR(-ENOMEM);
1841 if (devm_add_action_or_reset(dev, remove_free_mem_region, res))
1842 return ERR_PTR(-ENOMEM);
1845 write_lock(&resource_lock);
1846 for (addr = gfr_start(base, size, align, flags);
1847 gfr_continue(base, addr, size, flags);
1848 addr = gfr_next(addr, size, flags)) {
1849 if (__region_intersects(base, addr, size, 0, IORES_DESC_NONE) !=
1853 if (flags & GFR_REQUEST_REGION) {
1854 if (__request_region_locked(res, &iomem_resource, addr,
1859 dr->parent = &iomem_resource;
1862 devres_add(dev, dr);
1866 write_unlock(&resource_lock);
1870 * A driver is claiming this region so revoke any
1876 res->end = addr + size - 1;
1879 res->flags = IORESOURCE_MEM;
1882 * Only succeed if the resource hosts an exclusive
1883 * range after the insert
1885 if (__insert_resource(base, res) || res->child)
1888 write_unlock(&resource_lock);
1893 write_unlock(&resource_lock);
1895 if (flags & GFR_REQUEST_REGION) {
1899 devm_release_action(dev, remove_free_mem_region, res);
1901 return ERR_PTR(-ERANGE);
1905 * devm_request_free_mem_region - find free region for device private memory
1907 * @dev: device struct to bind the resource to
1908 * @size: size in bytes of the device memory to add
1909 * @base: resource tree to look in
1911 * This function tries to find an empty range of physical address big enough to
1912 * contain the new resource, so that it can later be hotplugged as ZONE_DEVICE
1913 * memory, which in turn allocates struct pages.
1915 struct resource *devm_request_free_mem_region(struct device *dev,
1916 struct resource *base, unsigned long size)
1918 unsigned long flags = GFR_DESCENDING | GFR_REQUEST_REGION;
1920 return get_free_mem_region(dev, base, size, GFR_DEFAULT_ALIGN,
1922 IORES_DESC_DEVICE_PRIVATE_MEMORY, flags);
1924 EXPORT_SYMBOL_GPL(devm_request_free_mem_region);
1926 struct resource *request_free_mem_region(struct resource *base,
1927 unsigned long size, const char *name)
1929 unsigned long flags = GFR_DESCENDING | GFR_REQUEST_REGION;
1931 return get_free_mem_region(NULL, base, size, GFR_DEFAULT_ALIGN, name,
1932 IORES_DESC_DEVICE_PRIVATE_MEMORY, flags);
1934 EXPORT_SYMBOL_GPL(request_free_mem_region);
1937 * alloc_free_mem_region - find a free region relative to @base
1938 * @base: resource that will parent the new resource
1939 * @size: size in bytes of memory to allocate from @base
1940 * @align: alignment requirements for the allocation
1941 * @name: resource name
1943 * Buses like CXL, that can dynamically instantiate new memory regions,
1944 * need a method to allocate physical address space for those regions.
1945 * Allocate and insert a new resource to cover a free, unclaimed by a
1946 * descendant of @base, range in the span of @base.
1948 struct resource *alloc_free_mem_region(struct resource *base,
1949 unsigned long size, unsigned long align,
1952 /* Default of ascending direction and insert resource */
1953 unsigned long flags = 0;
1955 return get_free_mem_region(NULL, base, size, align, name,
1956 IORES_DESC_NONE, flags);
1958 EXPORT_SYMBOL_NS_GPL(alloc_free_mem_region, CXL);
1959 #endif /* CONFIG_GET_FREE_REGION */
1961 static int __init strict_iomem(char *str)
1963 if (strstr(str, "relaxed"))
1964 strict_iomem_checks = 0;
1965 if (strstr(str, "strict"))
1966 strict_iomem_checks = 1;
1970 static int iomem_fs_init_fs_context(struct fs_context *fc)
1972 return init_pseudo(fc, DEVMEM_MAGIC) ? 0 : -ENOMEM;
1975 static struct file_system_type iomem_fs_type = {
1977 .owner = THIS_MODULE,
1978 .init_fs_context = iomem_fs_init_fs_context,
1979 .kill_sb = kill_anon_super,
1982 static int __init iomem_init_inode(void)
1984 static struct vfsmount *iomem_vfs_mount;
1985 static int iomem_fs_cnt;
1986 struct inode *inode;
1989 rc = simple_pin_fs(&iomem_fs_type, &iomem_vfs_mount, &iomem_fs_cnt);
1991 pr_err("Cannot mount iomem pseudo filesystem: %d\n", rc);
1995 inode = alloc_anon_inode(iomem_vfs_mount->mnt_sb);
1996 if (IS_ERR(inode)) {
1997 rc = PTR_ERR(inode);
1998 pr_err("Cannot allocate inode for iomem: %d\n", rc);
1999 simple_release_fs(&iomem_vfs_mount, &iomem_fs_cnt);
2004 * Publish iomem revocation inode initialized.
2005 * Pairs with smp_load_acquire() in revoke_iomem().
2007 smp_store_release(&iomem_inode, inode);
2012 fs_initcall(iomem_init_inode);
2014 __setup("iomem=", strict_iomem);