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(struct resource *parent, resource_size_t start,
493 size_t size, unsigned long flags,
497 int type = 0; int other = 0;
501 res.end = start + size - 1;
503 for (p = parent->child; p ; p = p->sibling) {
504 bool is_type = (((p->flags & flags) == flags) &&
505 ((desc == IORES_DESC_NONE) ||
508 if (resource_overlaps(p, &res))
509 is_type ? type++ : other++;
513 return REGION_DISJOINT;
516 return REGION_INTERSECTS;
522 * region_intersects() - determine intersection of region with known resources
523 * @start: region start address
524 * @size: size of region
525 * @flags: flags of resource (in iomem_resource)
526 * @desc: descriptor of resource (in iomem_resource) or IORES_DESC_NONE
528 * Check if the specified region partially overlaps or fully eclipses a
529 * resource identified by @flags and @desc (optional with IORES_DESC_NONE).
530 * Return REGION_DISJOINT if the region does not overlap @flags/@desc,
531 * return REGION_MIXED if the region overlaps @flags/@desc and another
532 * resource, and return REGION_INTERSECTS if the region overlaps @flags/@desc
533 * and no other defined resource. Note that REGION_INTERSECTS is also
534 * returned in the case when the specified region overlaps RAM and undefined
537 * region_intersect() is used by memory remapping functions to ensure
538 * the user is not remapping RAM and is a vast speed up over walking
539 * through the resource table page by page.
541 int region_intersects(resource_size_t start, size_t size, unsigned long flags,
546 read_lock(&resource_lock);
547 ret = __region_intersects(&iomem_resource, start, size, flags, desc);
548 read_unlock(&resource_lock);
552 EXPORT_SYMBOL_GPL(region_intersects);
554 void __weak arch_remove_reservations(struct resource *avail)
558 static resource_size_t simple_align_resource(void *data,
559 const struct resource *avail,
560 resource_size_t size,
561 resource_size_t align)
566 static void resource_clip(struct resource *res, resource_size_t min,
569 if (res->start < min)
576 * Find empty slot in the resource tree with the given range and
577 * alignment constraints
579 static int __find_resource(struct resource *root, struct resource *old,
580 struct resource *new,
581 resource_size_t size,
582 struct resource_constraint *constraint)
584 struct resource *this = root->child;
585 struct resource tmp = *new, avail, alloc;
587 tmp.start = root->start;
589 * Skip past an allocated resource that starts at 0, since the assignment
590 * of this->start - 1 to tmp->end below would cause an underflow.
592 if (this && this->start == root->start) {
593 tmp.start = (this == old) ? old->start : this->end + 1;
594 this = this->sibling;
598 tmp.end = (this == old) ? this->end : this->start - 1;
602 if (tmp.end < tmp.start)
605 resource_clip(&tmp, constraint->min, constraint->max);
606 arch_remove_reservations(&tmp);
608 /* Check for overflow after ALIGN() */
609 avail.start = ALIGN(tmp.start, constraint->align);
611 avail.flags = new->flags & ~IORESOURCE_UNSET;
612 if (avail.start >= tmp.start) {
613 alloc.flags = avail.flags;
614 alloc.start = constraint->alignf(constraint->alignf_data, &avail,
615 size, constraint->align);
616 alloc.end = alloc.start + size - 1;
617 if (alloc.start <= alloc.end &&
618 resource_contains(&avail, &alloc)) {
619 new->start = alloc.start;
620 new->end = alloc.end;
625 next: if (!this || this->end == root->end)
629 tmp.start = this->end + 1;
630 this = this->sibling;
636 * Find empty slot in the resource tree given range and alignment.
638 static int find_resource(struct resource *root, struct resource *new,
639 resource_size_t size,
640 struct resource_constraint *constraint)
642 return __find_resource(root, NULL, new, size, constraint);
646 * reallocate_resource - allocate a slot in the resource tree given range & alignment.
647 * The resource will be relocated if the new size cannot be reallocated in the
650 * @root: root resource descriptor
651 * @old: resource descriptor desired by caller
652 * @newsize: new size of the resource descriptor
653 * @constraint: the size and alignment constraints to be met.
655 static int reallocate_resource(struct resource *root, struct resource *old,
656 resource_size_t newsize,
657 struct resource_constraint *constraint)
660 struct resource new = *old;
661 struct resource *conflict;
663 write_lock(&resource_lock);
665 if ((err = __find_resource(root, old, &new, newsize, constraint)))
668 if (resource_contains(&new, old)) {
669 old->start = new.start;
679 if (resource_contains(old, &new)) {
680 old->start = new.start;
683 __release_resource(old, true);
685 conflict = __request_resource(root, old);
689 write_unlock(&resource_lock);
695 * allocate_resource - allocate empty slot in the resource tree given range & alignment.
696 * The resource will be reallocated with a new size if it was already allocated
697 * @root: root resource descriptor
698 * @new: resource descriptor desired by caller
699 * @size: requested resource region size
700 * @min: minimum boundary to allocate
701 * @max: maximum boundary to allocate
702 * @align: alignment requested, in bytes
703 * @alignf: alignment function, optional, called if not NULL
704 * @alignf_data: arbitrary data to pass to the @alignf function
706 int allocate_resource(struct resource *root, struct resource *new,
707 resource_size_t size, resource_size_t min,
708 resource_size_t max, resource_size_t align,
709 resource_size_t (*alignf)(void *,
710 const struct resource *,
716 struct resource_constraint constraint;
719 alignf = simple_align_resource;
721 constraint.min = min;
722 constraint.max = max;
723 constraint.align = align;
724 constraint.alignf = alignf;
725 constraint.alignf_data = alignf_data;
728 /* resource is already allocated, try reallocating with
729 the new constraints */
730 return reallocate_resource(root, new, size, &constraint);
733 write_lock(&resource_lock);
734 err = find_resource(root, new, size, &constraint);
735 if (err >= 0 && __request_resource(root, new))
737 write_unlock(&resource_lock);
741 EXPORT_SYMBOL(allocate_resource);
744 * lookup_resource - find an existing resource by a resource start address
745 * @root: root resource descriptor
746 * @start: resource start address
748 * Returns a pointer to the resource if found, NULL otherwise
750 struct resource *lookup_resource(struct resource *root, resource_size_t start)
752 struct resource *res;
754 read_lock(&resource_lock);
755 for (res = root->child; res; res = res->sibling) {
756 if (res->start == start)
759 read_unlock(&resource_lock);
765 * Insert a resource into the resource tree. If successful, return NULL,
766 * otherwise return the conflicting resource (compare to __request_resource())
768 static struct resource * __insert_resource(struct resource *parent, struct resource *new)
770 struct resource *first, *next;
772 for (;; parent = first) {
773 first = __request_resource(parent, new);
779 if (WARN_ON(first == new)) /* duplicated insertion */
782 if ((first->start > new->start) || (first->end < new->end))
784 if ((first->start == new->start) && (first->end == new->end))
788 for (next = first; ; next = next->sibling) {
789 /* Partial overlap? Bad, and unfixable */
790 if (next->start < new->start || next->end > new->end)
794 if (next->sibling->start > new->end)
798 new->parent = parent;
799 new->sibling = next->sibling;
802 next->sibling = NULL;
803 for (next = first; next; next = next->sibling)
806 if (parent->child == first) {
809 next = parent->child;
810 while (next->sibling != first)
811 next = next->sibling;
818 * insert_resource_conflict - Inserts resource in the resource tree
819 * @parent: parent of the new resource
820 * @new: new resource to insert
822 * Returns 0 on success, conflict resource if the resource can't be inserted.
824 * This function is equivalent to request_resource_conflict when no conflict
825 * happens. If a conflict happens, and the conflicting resources
826 * entirely fit within the range of the new resource, then the new
827 * resource is inserted and the conflicting resources become children of
830 * This function is intended for producers of resources, such as FW modules
833 struct resource *insert_resource_conflict(struct resource *parent, struct resource *new)
835 struct resource *conflict;
837 write_lock(&resource_lock);
838 conflict = __insert_resource(parent, new);
839 write_unlock(&resource_lock);
844 * insert_resource - Inserts a resource in the resource tree
845 * @parent: parent of the new resource
846 * @new: new resource to insert
848 * Returns 0 on success, -EBUSY if the resource can't be inserted.
850 * This function is intended for producers of resources, such as FW modules
853 int insert_resource(struct resource *parent, struct resource *new)
855 struct resource *conflict;
857 conflict = insert_resource_conflict(parent, new);
858 return conflict ? -EBUSY : 0;
860 EXPORT_SYMBOL_GPL(insert_resource);
863 * insert_resource_expand_to_fit - Insert a resource into the resource tree
864 * @root: root resource descriptor
865 * @new: new resource to insert
867 * Insert a resource into the resource tree, possibly expanding it in order
868 * to make it encompass any conflicting resources.
870 void insert_resource_expand_to_fit(struct resource *root, struct resource *new)
875 write_lock(&resource_lock);
877 struct resource *conflict;
879 conflict = __insert_resource(root, new);
882 if (conflict == root)
885 /* Ok, expand resource to cover the conflict, then try again .. */
886 if (conflict->start < new->start)
887 new->start = conflict->start;
888 if (conflict->end > new->end)
889 new->end = conflict->end;
891 pr_info("Expanded resource %s due to conflict with %s\n", new->name, conflict->name);
893 write_unlock(&resource_lock);
896 * Not for general consumption, only early boot memory map parsing, PCI
897 * resource discovery, and late discovery of CXL resources are expected
898 * to use this interface. The former are built-in and only the latter,
901 EXPORT_SYMBOL_NS_GPL(insert_resource_expand_to_fit, CXL);
904 * remove_resource - Remove a resource in the resource tree
905 * @old: resource to remove
907 * Returns 0 on success, -EINVAL if the resource is not valid.
909 * This function removes a resource previously inserted by insert_resource()
910 * or insert_resource_conflict(), and moves the children (if any) up to
911 * where they were before. insert_resource() and insert_resource_conflict()
912 * insert a new resource, and move any conflicting resources down to the
913 * children of the new resource.
915 * insert_resource(), insert_resource_conflict() and remove_resource() are
916 * intended for producers of resources, such as FW modules and bus drivers.
918 int remove_resource(struct resource *old)
922 write_lock(&resource_lock);
923 retval = __release_resource(old, false);
924 write_unlock(&resource_lock);
927 EXPORT_SYMBOL_GPL(remove_resource);
929 static int __adjust_resource(struct resource *res, resource_size_t start,
930 resource_size_t size)
932 struct resource *tmp, *parent = res->parent;
933 resource_size_t end = start + size - 1;
939 if ((start < parent->start) || (end > parent->end))
942 if (res->sibling && (res->sibling->start <= end))
947 while (tmp->sibling != res)
949 if (start <= tmp->end)
954 for (tmp = res->child; tmp; tmp = tmp->sibling)
955 if ((tmp->start < start) || (tmp->end > end))
967 * adjust_resource - modify a resource's start and size
968 * @res: resource to modify
969 * @start: new start value
972 * Given an existing resource, change its start and size to match the
973 * arguments. Returns 0 on success, -EBUSY if it can't fit.
974 * Existing children of the resource are assumed to be immutable.
976 int adjust_resource(struct resource *res, resource_size_t start,
977 resource_size_t size)
981 write_lock(&resource_lock);
982 result = __adjust_resource(res, start, size);
983 write_unlock(&resource_lock);
986 EXPORT_SYMBOL(adjust_resource);
989 __reserve_region_with_split(struct resource *root, resource_size_t start,
990 resource_size_t end, const char *name)
992 struct resource *parent = root;
993 struct resource *conflict;
994 struct resource *res = alloc_resource(GFP_ATOMIC);
995 struct resource *next_res = NULL;
996 int type = resource_type(root);
1004 res->flags = type | IORESOURCE_BUSY;
1005 res->desc = IORES_DESC_NONE;
1009 conflict = __request_resource(parent, res);
1018 /* conflict covered whole area */
1019 if (conflict->start <= res->start &&
1020 conflict->end >= res->end) {
1026 /* failed, split and try again */
1027 if (conflict->start > res->start) {
1029 res->end = conflict->start - 1;
1030 if (conflict->end < end) {
1031 next_res = alloc_resource(GFP_ATOMIC);
1036 next_res->name = name;
1037 next_res->start = conflict->end + 1;
1038 next_res->end = end;
1039 next_res->flags = type | IORESOURCE_BUSY;
1040 next_res->desc = IORES_DESC_NONE;
1043 res->start = conflict->end + 1;
1050 reserve_region_with_split(struct resource *root, resource_size_t start,
1051 resource_size_t end, const char *name)
1055 write_lock(&resource_lock);
1056 if (root->start > start || root->end < end) {
1057 pr_err("requested range [0x%llx-0x%llx] not in root %pr\n",
1058 (unsigned long long)start, (unsigned long long)end,
1060 if (start > root->end || end < root->start)
1063 if (end > root->end)
1065 if (start < root->start)
1066 start = root->start;
1067 pr_err("fixing request to [0x%llx-0x%llx]\n",
1068 (unsigned long long)start,
1069 (unsigned long long)end);
1074 __reserve_region_with_split(root, start, end, name);
1075 write_unlock(&resource_lock);
1079 * resource_alignment - calculate resource's alignment
1080 * @res: resource pointer
1082 * Returns alignment on success, 0 (invalid alignment) on failure.
1084 resource_size_t resource_alignment(struct resource *res)
1086 switch (res->flags & (IORESOURCE_SIZEALIGN | IORESOURCE_STARTALIGN)) {
1087 case IORESOURCE_SIZEALIGN:
1088 return resource_size(res);
1089 case IORESOURCE_STARTALIGN:
1097 * This is compatibility stuff for IO resources.
1099 * Note how this, unlike the above, knows about
1100 * the IO flag meanings (busy etc).
1102 * request_region creates a new busy region.
1104 * release_region releases a matching busy region.
1107 static DECLARE_WAIT_QUEUE_HEAD(muxed_resource_wait);
1109 static struct inode *iomem_inode;
1111 #ifdef CONFIG_IO_STRICT_DEVMEM
1112 static void revoke_iomem(struct resource *res)
1114 /* pairs with smp_store_release() in iomem_init_inode() */
1115 struct inode *inode = smp_load_acquire(&iomem_inode);
1118 * Check that the initialization has completed. Losing the race
1119 * is ok because it means drivers are claiming resources before
1120 * the fs_initcall level of init and prevent iomem_get_mapping users
1121 * from establishing mappings.
1127 * The expectation is that the driver has successfully marked
1128 * the resource busy by this point, so devmem_is_allowed()
1129 * should start returning false, however for performance this
1130 * does not iterate the entire resource range.
1132 if (devmem_is_allowed(PHYS_PFN(res->start)) &&
1133 devmem_is_allowed(PHYS_PFN(res->end))) {
1135 * *cringe* iomem=relaxed says "go ahead, what's the
1136 * worst that can happen?"
1141 unmap_mapping_range(inode->i_mapping, res->start, resource_size(res), 1);
1144 static void revoke_iomem(struct resource *res) {}
1147 struct address_space *iomem_get_mapping(void)
1150 * This function is only called from file open paths, hence guaranteed
1151 * that fs_initcalls have completed and no need to check for NULL. But
1152 * since revoke_iomem can be called before the initcall we still need
1153 * the barrier to appease checkers.
1155 return smp_load_acquire(&iomem_inode)->i_mapping;
1158 static int __request_region_locked(struct resource *res, struct resource *parent,
1159 resource_size_t start, resource_size_t n,
1160 const char *name, int flags)
1162 DECLARE_WAITQUEUE(wait, current);
1166 res->end = start + n - 1;
1169 struct resource *conflict;
1171 res->flags = resource_type(parent) | resource_ext_type(parent);
1172 res->flags |= IORESOURCE_BUSY | flags;
1173 res->desc = parent->desc;
1175 conflict = __request_resource(parent, res);
1179 * mm/hmm.c reserves physical addresses which then
1180 * become unavailable to other users. Conflicts are
1181 * not expected. Warn to aid debugging if encountered.
1183 if (conflict->desc == IORES_DESC_DEVICE_PRIVATE_MEMORY) {
1184 pr_warn("Unaddressable device %s %pR conflicts with %pR",
1185 conflict->name, conflict, res);
1187 if (conflict != parent) {
1188 if (!(conflict->flags & IORESOURCE_BUSY)) {
1193 if (conflict->flags & flags & IORESOURCE_MUXED) {
1194 add_wait_queue(&muxed_resource_wait, &wait);
1195 write_unlock(&resource_lock);
1196 set_current_state(TASK_UNINTERRUPTIBLE);
1198 remove_wait_queue(&muxed_resource_wait, &wait);
1199 write_lock(&resource_lock);
1202 /* Uhhuh, that didn't work out.. */
1210 * __request_region - create a new busy resource region
1211 * @parent: parent resource descriptor
1212 * @start: resource start address
1213 * @n: resource region size
1214 * @name: reserving caller's ID string
1215 * @flags: IO resource flags
1217 struct resource *__request_region(struct resource *parent,
1218 resource_size_t start, resource_size_t n,
1219 const char *name, int flags)
1221 struct resource *res = alloc_resource(GFP_KERNEL);
1227 write_lock(&resource_lock);
1228 ret = __request_region_locked(res, parent, start, n, name, flags);
1229 write_unlock(&resource_lock);
1236 if (parent == &iomem_resource)
1241 EXPORT_SYMBOL(__request_region);
1244 * __release_region - release a previously reserved resource region
1245 * @parent: parent resource descriptor
1246 * @start: resource start address
1247 * @n: resource region size
1249 * The described resource region must match a currently busy region.
1251 void __release_region(struct resource *parent, resource_size_t start,
1254 struct resource **p;
1255 resource_size_t end;
1258 end = start + n - 1;
1260 write_lock(&resource_lock);
1263 struct resource *res = *p;
1267 if (res->start <= start && res->end >= end) {
1268 if (!(res->flags & IORESOURCE_BUSY)) {
1272 if (res->start != start || res->end != end)
1275 write_unlock(&resource_lock);
1276 if (res->flags & IORESOURCE_MUXED)
1277 wake_up(&muxed_resource_wait);
1284 write_unlock(&resource_lock);
1286 pr_warn("Trying to free nonexistent resource <%pa-%pa>\n", &start, &end);
1288 EXPORT_SYMBOL(__release_region);
1290 #ifdef CONFIG_MEMORY_HOTREMOVE
1292 * release_mem_region_adjustable - release a previously reserved memory region
1293 * @start: resource start address
1294 * @size: resource region size
1296 * This interface is intended for memory hot-delete. The requested region
1297 * is released from a currently busy memory resource. The requested region
1298 * must either match exactly or fit into a single busy resource entry. In
1299 * the latter case, the remaining resource is adjusted accordingly.
1300 * Existing children of the busy memory resource must be immutable in the
1304 * - Additional release conditions, such as overlapping region, can be
1305 * supported after they are confirmed as valid cases.
1306 * - When a busy memory resource gets split into two entries, the code
1307 * assumes that all children remain in the lower address entry for
1308 * simplicity. Enhance this logic when necessary.
1310 void release_mem_region_adjustable(resource_size_t start, resource_size_t size)
1312 struct resource *parent = &iomem_resource;
1313 struct resource *new_res = NULL;
1314 bool alloc_nofail = false;
1315 struct resource **p;
1316 struct resource *res;
1317 resource_size_t end;
1319 end = start + size - 1;
1320 if (WARN_ON_ONCE((start < parent->start) || (end > parent->end)))
1324 * We free up quite a lot of memory on memory hotunplug (esp., memap),
1325 * just before releasing the region. This is highly unlikely to
1326 * fail - let's play save and make it never fail as the caller cannot
1327 * perform any error handling (e.g., trying to re-add memory will fail
1331 new_res = alloc_resource(GFP_KERNEL | (alloc_nofail ? __GFP_NOFAIL : 0));
1334 write_lock(&resource_lock);
1336 while ((res = *p)) {
1337 if (res->start >= end)
1340 /* look for the next resource if it does not fit into */
1341 if (res->start > start || res->end < end) {
1346 if (!(res->flags & IORESOURCE_MEM))
1349 if (!(res->flags & IORESOURCE_BUSY)) {
1354 /* found the target resource; let's adjust accordingly */
1355 if (res->start == start && res->end == end) {
1356 /* free the whole entry */
1359 } else if (res->start == start && res->end != end) {
1360 /* adjust the start */
1361 WARN_ON_ONCE(__adjust_resource(res, end + 1,
1363 } else if (res->start != start && res->end == end) {
1364 /* adjust the end */
1365 WARN_ON_ONCE(__adjust_resource(res, res->start,
1366 start - res->start));
1368 /* split into two entries - we need a new resource */
1370 new_res = alloc_resource(GFP_ATOMIC);
1372 alloc_nofail = true;
1373 write_unlock(&resource_lock);
1377 new_res->name = res->name;
1378 new_res->start = end + 1;
1379 new_res->end = res->end;
1380 new_res->flags = res->flags;
1381 new_res->desc = res->desc;
1382 new_res->parent = res->parent;
1383 new_res->sibling = res->sibling;
1384 new_res->child = NULL;
1386 if (WARN_ON_ONCE(__adjust_resource(res, res->start,
1387 start - res->start)))
1389 res->sibling = new_res;
1396 write_unlock(&resource_lock);
1397 free_resource(new_res);
1399 #endif /* CONFIG_MEMORY_HOTREMOVE */
1401 #ifdef CONFIG_MEMORY_HOTPLUG
1402 static bool system_ram_resources_mergeable(struct resource *r1,
1403 struct resource *r2)
1405 /* We assume either r1 or r2 is IORESOURCE_SYSRAM_MERGEABLE. */
1406 return r1->flags == r2->flags && r1->end + 1 == r2->start &&
1407 r1->name == r2->name && r1->desc == r2->desc &&
1408 !r1->child && !r2->child;
1412 * merge_system_ram_resource - mark the System RAM resource mergeable and try to
1413 * merge it with adjacent, mergeable resources
1414 * @res: resource descriptor
1416 * This interface is intended for memory hotplug, whereby lots of contiguous
1417 * system ram resources are added (e.g., via add_memory*()) by a driver, and
1418 * the actual resource boundaries are not of interest (e.g., it might be
1419 * relevant for DIMMs). Only resources that are marked mergeable, that have the
1420 * same parent, and that don't have any children are considered. All mergeable
1421 * resources must be immutable during the request.
1424 * - The caller has to make sure that no pointers to resources that are
1425 * marked mergeable are used anymore after this call - the resource might
1426 * be freed and the pointer might be stale!
1427 * - release_mem_region_adjustable() will split on demand on memory hotunplug
1429 void merge_system_ram_resource(struct resource *res)
1431 const unsigned long flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
1432 struct resource *cur;
1434 if (WARN_ON_ONCE((res->flags & flags) != flags))
1437 write_lock(&resource_lock);
1438 res->flags |= IORESOURCE_SYSRAM_MERGEABLE;
1440 /* Try to merge with next item in the list. */
1442 if (cur && system_ram_resources_mergeable(res, cur)) {
1443 res->end = cur->end;
1444 res->sibling = cur->sibling;
1448 /* Try to merge with previous item in the list. */
1449 cur = res->parent->child;
1450 while (cur && cur->sibling != res)
1452 if (cur && system_ram_resources_mergeable(cur, res)) {
1453 cur->end = res->end;
1454 cur->sibling = res->sibling;
1457 write_unlock(&resource_lock);
1459 #endif /* CONFIG_MEMORY_HOTPLUG */
1462 * Managed region resource
1464 static void devm_resource_release(struct device *dev, void *ptr)
1466 struct resource **r = ptr;
1468 release_resource(*r);
1472 * devm_request_resource() - request and reserve an I/O or memory resource
1473 * @dev: device for which to request the resource
1474 * @root: root of the resource tree from which to request the resource
1475 * @new: descriptor of the resource to request
1477 * This is a device-managed version of request_resource(). There is usually
1478 * no need to release resources requested by this function explicitly since
1479 * that will be taken care of when the device is unbound from its driver.
1480 * If for some reason the resource needs to be released explicitly, because
1481 * of ordering issues for example, drivers must call devm_release_resource()
1482 * rather than the regular release_resource().
1484 * When a conflict is detected between any existing resources and the newly
1485 * requested resource, an error message will be printed.
1487 * Returns 0 on success or a negative error code on failure.
1489 int devm_request_resource(struct device *dev, struct resource *root,
1490 struct resource *new)
1492 struct resource *conflict, **ptr;
1494 ptr = devres_alloc(devm_resource_release, sizeof(*ptr), GFP_KERNEL);
1500 conflict = request_resource_conflict(root, new);
1502 dev_err(dev, "resource collision: %pR conflicts with %s %pR\n",
1503 new, conflict->name, conflict);
1508 devres_add(dev, ptr);
1511 EXPORT_SYMBOL(devm_request_resource);
1513 static int devm_resource_match(struct device *dev, void *res, void *data)
1515 struct resource **ptr = res;
1517 return *ptr == data;
1521 * devm_release_resource() - release a previously requested resource
1522 * @dev: device for which to release the resource
1523 * @new: descriptor of the resource to release
1525 * Releases a resource previously requested using devm_request_resource().
1527 void devm_release_resource(struct device *dev, struct resource *new)
1529 WARN_ON(devres_release(dev, devm_resource_release, devm_resource_match,
1532 EXPORT_SYMBOL(devm_release_resource);
1534 struct region_devres {
1535 struct resource *parent;
1536 resource_size_t start;
1540 static void devm_region_release(struct device *dev, void *res)
1542 struct region_devres *this = res;
1544 __release_region(this->parent, this->start, this->n);
1547 static int devm_region_match(struct device *dev, void *res, void *match_data)
1549 struct region_devres *this = res, *match = match_data;
1551 return this->parent == match->parent &&
1552 this->start == match->start && this->n == match->n;
1556 __devm_request_region(struct device *dev, struct resource *parent,
1557 resource_size_t start, resource_size_t n, const char *name)
1559 struct region_devres *dr = NULL;
1560 struct resource *res;
1562 dr = devres_alloc(devm_region_release, sizeof(struct region_devres),
1567 dr->parent = parent;
1571 res = __request_region(parent, start, n, name, 0);
1573 devres_add(dev, dr);
1579 EXPORT_SYMBOL(__devm_request_region);
1581 void __devm_release_region(struct device *dev, struct resource *parent,
1582 resource_size_t start, resource_size_t n)
1584 struct region_devres match_data = { parent, start, n };
1586 __release_region(parent, start, n);
1587 WARN_ON(devres_destroy(dev, devm_region_release, devm_region_match,
1590 EXPORT_SYMBOL(__devm_release_region);
1593 * Reserve I/O ports or memory based on "reserve=" kernel parameter.
1595 #define MAXRESERVE 4
1596 static int __init reserve_setup(char *str)
1598 static int reserved;
1599 static struct resource reserve[MAXRESERVE];
1602 unsigned int io_start, io_num;
1604 struct resource *parent;
1606 if (get_option(&str, &io_start) != 2)
1608 if (get_option(&str, &io_num) == 0)
1610 if (x < MAXRESERVE) {
1611 struct resource *res = reserve + x;
1614 * If the region starts below 0x10000, we assume it's
1615 * I/O port space; otherwise assume it's memory.
1617 if (io_start < 0x10000) {
1618 res->flags = IORESOURCE_IO;
1619 parent = &ioport_resource;
1621 res->flags = IORESOURCE_MEM;
1622 parent = &iomem_resource;
1624 res->name = "reserved";
1625 res->start = io_start;
1626 res->end = io_start + io_num - 1;
1627 res->flags |= IORESOURCE_BUSY;
1628 res->desc = IORES_DESC_NONE;
1630 if (request_resource(parent, res) == 0)
1636 __setup("reserve=", reserve_setup);
1639 * Check if the requested addr and size spans more than any slot in the
1640 * iomem resource tree.
1642 int iomem_map_sanity_check(resource_size_t addr, unsigned long size)
1644 struct resource *p = &iomem_resource;
1645 resource_size_t end = addr + size - 1;
1649 read_lock(&resource_lock);
1650 for (p = p->child; p ; p = r_next(NULL, p, &l)) {
1652 * We can probably skip the resources without
1653 * IORESOURCE_IO attribute?
1659 if (PFN_DOWN(p->start) <= PFN_DOWN(addr) &&
1660 PFN_DOWN(p->end) >= PFN_DOWN(end))
1663 * if a resource is "BUSY", it's not a hardware resource
1664 * but a driver mapping of such a resource; we don't want
1665 * to warn for those; some drivers legitimately map only
1666 * partial hardware resources. (example: vesafb)
1668 if (p->flags & IORESOURCE_BUSY)
1671 pr_warn("resource sanity check: requesting [mem %pa-%pa], which spans more than %s %pR\n",
1672 &addr, &end, p->name, p);
1676 read_unlock(&resource_lock);
1681 #ifdef CONFIG_STRICT_DEVMEM
1682 static int strict_iomem_checks = 1;
1684 static int strict_iomem_checks;
1688 * Check if an address is exclusive to the kernel and must not be mapped to
1689 * user space, for example, via /dev/mem.
1691 * Returns true if exclusive to the kernel, otherwise returns false.
1693 bool resource_is_exclusive(struct resource *root, u64 addr, resource_size_t size)
1695 const unsigned int exclusive_system_ram = IORESOURCE_SYSTEM_RAM |
1696 IORESOURCE_EXCLUSIVE;
1697 bool skip_children = false, err = false;
1700 read_lock(&resource_lock);
1701 for_each_resource(root, p, skip_children) {
1702 if (p->start >= addr + size)
1704 if (p->end < addr) {
1705 skip_children = true;
1708 skip_children = false;
1711 * IORESOURCE_SYSTEM_RAM resources are exclusive if
1712 * IORESOURCE_EXCLUSIVE is set, even if they
1713 * are not busy and even if "iomem=relaxed" is set. The
1714 * responsible driver dynamically adds/removes system RAM within
1715 * such an area and uncontrolled access is dangerous.
1717 if ((p->flags & exclusive_system_ram) == exclusive_system_ram) {
1723 * A resource is exclusive if IORESOURCE_EXCLUSIVE is set
1724 * or CONFIG_IO_STRICT_DEVMEM is enabled and the
1727 if (!strict_iomem_checks || !(p->flags & IORESOURCE_BUSY))
1729 if (IS_ENABLED(CONFIG_IO_STRICT_DEVMEM)
1730 || p->flags & IORESOURCE_EXCLUSIVE) {
1735 read_unlock(&resource_lock);
1740 bool iomem_is_exclusive(u64 addr)
1742 return resource_is_exclusive(&iomem_resource, addr & PAGE_MASK,
1746 struct resource_entry *resource_list_create_entry(struct resource *res,
1749 struct resource_entry *entry;
1751 entry = kzalloc(sizeof(*entry) + extra_size, GFP_KERNEL);
1753 INIT_LIST_HEAD(&entry->node);
1754 entry->res = res ? res : &entry->__res;
1759 EXPORT_SYMBOL(resource_list_create_entry);
1761 void resource_list_free(struct list_head *head)
1763 struct resource_entry *entry, *tmp;
1765 list_for_each_entry_safe(entry, tmp, head, node)
1766 resource_list_destroy_entry(entry);
1768 EXPORT_SYMBOL(resource_list_free);
1770 #ifdef CONFIG_GET_FREE_REGION
1771 #define GFR_DESCENDING (1UL << 0)
1772 #define GFR_REQUEST_REGION (1UL << 1)
1773 #define GFR_DEFAULT_ALIGN (1UL << PA_SECTION_SHIFT)
1775 static resource_size_t gfr_start(struct resource *base, resource_size_t size,
1776 resource_size_t align, unsigned long flags)
1778 if (flags & GFR_DESCENDING) {
1779 resource_size_t end;
1781 end = min_t(resource_size_t, base->end,
1782 (1ULL << MAX_PHYSMEM_BITS) - 1);
1783 return end - size + 1;
1786 return ALIGN(base->start, align);
1789 static bool gfr_continue(struct resource *base, resource_size_t addr,
1790 resource_size_t size, unsigned long flags)
1792 if (flags & GFR_DESCENDING)
1793 return addr > size && addr >= base->start;
1795 * In the ascend case be careful that the last increment by
1796 * @size did not wrap 0.
1798 return addr > addr - size &&
1799 addr <= min_t(resource_size_t, base->end,
1800 (1ULL << MAX_PHYSMEM_BITS) - 1);
1803 static resource_size_t gfr_next(resource_size_t addr, resource_size_t size,
1804 unsigned long flags)
1806 if (flags & GFR_DESCENDING)
1811 static void remove_free_mem_region(void *_res)
1813 struct resource *res = _res;
1816 remove_resource(res);
1820 static struct resource *
1821 get_free_mem_region(struct device *dev, struct resource *base,
1822 resource_size_t size, const unsigned long align,
1823 const char *name, const unsigned long desc,
1824 const unsigned long flags)
1826 resource_size_t addr;
1827 struct resource *res;
1828 struct region_devres *dr = NULL;
1830 size = ALIGN(size, align);
1832 res = alloc_resource(GFP_KERNEL);
1834 return ERR_PTR(-ENOMEM);
1836 if (dev && (flags & GFR_REQUEST_REGION)) {
1837 dr = devres_alloc(devm_region_release,
1838 sizeof(struct region_devres), GFP_KERNEL);
1841 return ERR_PTR(-ENOMEM);
1844 if (devm_add_action_or_reset(dev, remove_free_mem_region, res))
1845 return ERR_PTR(-ENOMEM);
1848 write_lock(&resource_lock);
1849 for (addr = gfr_start(base, size, align, flags);
1850 gfr_continue(base, addr, size, flags);
1851 addr = gfr_next(addr, size, flags)) {
1852 if (__region_intersects(base, addr, size, 0, IORES_DESC_NONE) !=
1856 if (flags & GFR_REQUEST_REGION) {
1857 if (__request_region_locked(res, &iomem_resource, addr,
1862 dr->parent = &iomem_resource;
1865 devres_add(dev, dr);
1869 write_unlock(&resource_lock);
1873 * A driver is claiming this region so revoke any
1879 res->end = addr + size - 1;
1882 res->flags = IORESOURCE_MEM;
1885 * Only succeed if the resource hosts an exclusive
1886 * range after the insert
1888 if (__insert_resource(base, res) || res->child)
1891 write_unlock(&resource_lock);
1896 write_unlock(&resource_lock);
1898 if (flags & GFR_REQUEST_REGION) {
1902 devm_release_action(dev, remove_free_mem_region, res);
1904 return ERR_PTR(-ERANGE);
1908 * devm_request_free_mem_region - find free region for device private memory
1910 * @dev: device struct to bind the resource to
1911 * @size: size in bytes of the device memory to add
1912 * @base: resource tree to look in
1914 * This function tries to find an empty range of physical address big enough to
1915 * contain the new resource, so that it can later be hotplugged as ZONE_DEVICE
1916 * memory, which in turn allocates struct pages.
1918 struct resource *devm_request_free_mem_region(struct device *dev,
1919 struct resource *base, unsigned long size)
1921 unsigned long flags = GFR_DESCENDING | GFR_REQUEST_REGION;
1923 return get_free_mem_region(dev, base, size, GFR_DEFAULT_ALIGN,
1925 IORES_DESC_DEVICE_PRIVATE_MEMORY, flags);
1927 EXPORT_SYMBOL_GPL(devm_request_free_mem_region);
1929 struct resource *request_free_mem_region(struct resource *base,
1930 unsigned long size, const char *name)
1932 unsigned long flags = GFR_DESCENDING | GFR_REQUEST_REGION;
1934 return get_free_mem_region(NULL, base, size, GFR_DEFAULT_ALIGN, name,
1935 IORES_DESC_DEVICE_PRIVATE_MEMORY, flags);
1937 EXPORT_SYMBOL_GPL(request_free_mem_region);
1940 * alloc_free_mem_region - find a free region relative to @base
1941 * @base: resource that will parent the new resource
1942 * @size: size in bytes of memory to allocate from @base
1943 * @align: alignment requirements for the allocation
1944 * @name: resource name
1946 * Buses like CXL, that can dynamically instantiate new memory regions,
1947 * need a method to allocate physical address space for those regions.
1948 * Allocate and insert a new resource to cover a free, unclaimed by a
1949 * descendant of @base, range in the span of @base.
1951 struct resource *alloc_free_mem_region(struct resource *base,
1952 unsigned long size, unsigned long align,
1955 /* Default of ascending direction and insert resource */
1956 unsigned long flags = 0;
1958 return get_free_mem_region(NULL, base, size, align, name,
1959 IORES_DESC_NONE, flags);
1961 EXPORT_SYMBOL_NS_GPL(alloc_free_mem_region, CXL);
1962 #endif /* CONFIG_GET_FREE_REGION */
1964 static int __init strict_iomem(char *str)
1966 if (strstr(str, "relaxed"))
1967 strict_iomem_checks = 0;
1968 if (strstr(str, "strict"))
1969 strict_iomem_checks = 1;
1973 static int iomem_fs_init_fs_context(struct fs_context *fc)
1975 return init_pseudo(fc, DEVMEM_MAGIC) ? 0 : -ENOMEM;
1978 static struct file_system_type iomem_fs_type = {
1980 .owner = THIS_MODULE,
1981 .init_fs_context = iomem_fs_init_fs_context,
1982 .kill_sb = kill_anon_super,
1985 static int __init iomem_init_inode(void)
1987 static struct vfsmount *iomem_vfs_mount;
1988 static int iomem_fs_cnt;
1989 struct inode *inode;
1992 rc = simple_pin_fs(&iomem_fs_type, &iomem_vfs_mount, &iomem_fs_cnt);
1994 pr_err("Cannot mount iomem pseudo filesystem: %d\n", rc);
1998 inode = alloc_anon_inode(iomem_vfs_mount->mnt_sb);
1999 if (IS_ERR(inode)) {
2000 rc = PTR_ERR(inode);
2001 pr_err("Cannot allocate inode for iomem: %d\n", rc);
2002 simple_release_fs(&iomem_vfs_mount, &iomem_fs_cnt);
2007 * Publish iomem revocation inode initialized.
2008 * Pairs with smp_load_acquire() in revoke_iomem().
2010 smp_store_release(&iomem_inode, inode);
2015 fs_initcall(iomem_init_inode);
2017 __setup("iomem=", strict_iomem);