Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/kernel/resource.c
4 *
5 * Copyright (C) 1999 Linus Torvalds
6 * Copyright (C) 1999 Martin Mares <mj@ucw.cz>
7 *
8 * Arbitrary resource management.
9 */
10
11#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12
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>
19#include <linux/fs.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>
26#include <linux/mm.h>
27#include <linux/mount.h>
28#include <linux/resource_ext.h>
29#include <uapi/linux/magic.h>
30#include <linux/string.h>
31#include <linux/vmalloc.h>
32#include <asm/io.h>
33
34
35struct resource ioport_resource = {
36 .name = "PCI IO",
37 .start = 0,
38 .end = IO_SPACE_LIMIT,
39 .flags = IORESOURCE_IO,
40};
41EXPORT_SYMBOL(ioport_resource);
42
43struct resource iomem_resource = {
44 .name = "PCI mem",
45 .start = 0,
46 .end = -1,
47 .flags = IORESOURCE_MEM,
48};
49EXPORT_SYMBOL(iomem_resource);
50
51struct resource soft_reserve_resource = {
52 .name = "Soft Reserved",
53 .start = 0,
54 .end = -1,
55 .desc = IORES_DESC_SOFT_RESERVED,
56 .flags = IORESOURCE_MEM,
57};
58
59static DEFINE_RWLOCK(resource_lock);
60
61/*
62 * Return the next node of @p in pre-order tree traversal. If
63 * @skip_children is true, skip the descendant nodes of @p in
64 * traversal. If @p is a descendant of @subtree_root, only traverse
65 * the subtree under @subtree_root.
66 */
67static struct resource *next_resource(struct resource *p, bool skip_children,
68 struct resource *subtree_root)
69{
70 if (!skip_children && p->child)
71 return p->child;
72 while (!p->sibling && p->parent) {
73 p = p->parent;
74 if (p == subtree_root)
75 return NULL;
76 }
77 return p->sibling;
78}
79
80/*
81 * Traverse the resource subtree under @_root in pre-order, excluding
82 * @_root itself.
83 *
84 * NOTE: '__p' is introduced to avoid shadowing '_p' outside of loop.
85 * And it is referenced to avoid unused variable warning.
86 */
87#define for_each_resource(_root, _p, _skip_children) \
88 for (typeof(_root) __root = (_root), __p = _p = __root->child; \
89 __p && _p; _p = next_resource(_p, _skip_children, __root))
90
91#ifdef CONFIG_PROC_FS
92
93enum { MAX_IORES_LEVEL = 8 };
94
95static void *r_start(struct seq_file *m, loff_t *pos)
96 __acquires(resource_lock)
97{
98 struct resource *root = pde_data(file_inode(m->file));
99 struct resource *p;
100 loff_t l = *pos;
101
102 read_lock(&resource_lock);
103 for_each_resource(root, p, false) {
104 if (l-- == 0)
105 break;
106 }
107
108 return p;
109}
110
111static void *r_next(struct seq_file *m, void *v, loff_t *pos)
112{
113 struct resource *p = v;
114
115 (*pos)++;
116
117 return (void *)next_resource(p, false, NULL);
118}
119
120static void r_stop(struct seq_file *m, void *v)
121 __releases(resource_lock)
122{
123 read_unlock(&resource_lock);
124}
125
126static int r_show(struct seq_file *m, void *v)
127{
128 struct resource *root = pde_data(file_inode(m->file));
129 struct resource *r = v, *p;
130 unsigned long long start, end;
131 int width = root->end < 0x10000 ? 4 : 8;
132 int depth;
133
134 for (depth = 0, p = r; depth < MAX_IORES_LEVEL; depth++, p = p->parent)
135 if (p->parent == root)
136 break;
137
138 if (file_ns_capable(m->file, &init_user_ns, CAP_SYS_ADMIN)) {
139 start = r->start;
140 end = r->end;
141 } else {
142 start = end = 0;
143 }
144
145 seq_printf(m, "%*s%0*llx-%0*llx : %s\n",
146 depth * 2, "",
147 width, start,
148 width, end,
149 r->name ? r->name : "<BAD>");
150 return 0;
151}
152
153static const struct seq_operations resource_op = {
154 .start = r_start,
155 .next = r_next,
156 .stop = r_stop,
157 .show = r_show,
158};
159
160static int __init ioresources_init(void)
161{
162 proc_create_seq_data("ioports", 0, NULL, &resource_op,
163 &ioport_resource);
164 proc_create_seq_data("iomem", 0, NULL, &resource_op, &iomem_resource);
165 return 0;
166}
167__initcall(ioresources_init);
168
169#endif /* CONFIG_PROC_FS */
170
171static void free_resource(struct resource *res)
172{
173 /**
174 * If the resource was allocated using memblock early during boot
175 * we'll leak it here: we can only return full pages back to the
176 * buddy and trying to be smart and reusing them eventually in
177 * alloc_resource() overcomplicates resource handling.
178 */
179 if (res && PageSlab(virt_to_head_page(res)))
180 kfree(res);
181}
182
183static struct resource *alloc_resource(gfp_t flags)
184{
185 return kzalloc_obj(struct resource, flags);
186}
187
188/* Return the conflict entry if you can't request it */
189static struct resource * __request_resource(struct resource *root, struct resource *new)
190{
191 resource_size_t start = new->start;
192 resource_size_t end = new->end;
193 struct resource *tmp, **p;
194
195 if (end < start)
196 return root;
197 if (start < root->start)
198 return root;
199 if (end > root->end)
200 return root;
201 p = &root->child;
202 for (;;) {
203 tmp = *p;
204 if (!tmp || tmp->start > end) {
205 new->sibling = tmp;
206 *p = new;
207 new->parent = root;
208 return NULL;
209 }
210 p = &tmp->sibling;
211 if (tmp->end < start)
212 continue;
213 return tmp;
214 }
215}
216
217static int __release_resource(struct resource *old, bool release_child)
218{
219 struct resource *tmp, **p, *chd;
220
221 p = &old->parent->child;
222 for (;;) {
223 tmp = *p;
224 if (!tmp)
225 break;
226 if (tmp == old) {
227 if (release_child || !(tmp->child)) {
228 *p = tmp->sibling;
229 } else {
230 for (chd = tmp->child;; chd = chd->sibling) {
231 chd->parent = tmp->parent;
232 if (!(chd->sibling))
233 break;
234 }
235 *p = tmp->child;
236 chd->sibling = tmp->sibling;
237 }
238 old->parent = NULL;
239 return 0;
240 }
241 p = &tmp->sibling;
242 }
243 return -EINVAL;
244}
245
246static void __release_child_resources(struct resource *r)
247{
248 struct resource *tmp, *p;
249 resource_size_t size;
250
251 p = r->child;
252 r->child = NULL;
253 while (p) {
254 tmp = p;
255 p = p->sibling;
256
257 tmp->parent = NULL;
258 tmp->sibling = NULL;
259 __release_child_resources(tmp);
260
261 printk(KERN_DEBUG "release child resource %pR\n", tmp);
262 /* need to restore size, and keep flags */
263 size = resource_size(tmp);
264 tmp->start = 0;
265 tmp->end = size - 1;
266 }
267}
268
269void release_child_resources(struct resource *r)
270{
271 write_lock(&resource_lock);
272 __release_child_resources(r);
273 write_unlock(&resource_lock);
274}
275
276/**
277 * request_resource_conflict - request and reserve an I/O or memory resource
278 * @root: root resource descriptor
279 * @new: resource descriptor desired by caller
280 *
281 * Returns 0 for success, conflict resource on error.
282 */
283struct resource *request_resource_conflict(struct resource *root, struct resource *new)
284{
285 struct resource *conflict;
286
287 write_lock(&resource_lock);
288 conflict = __request_resource(root, new);
289 write_unlock(&resource_lock);
290 return conflict;
291}
292
293/**
294 * request_resource - request and reserve an I/O or memory resource
295 * @root: root resource descriptor
296 * @new: resource descriptor desired by caller
297 *
298 * Returns 0 for success, negative error code on error.
299 */
300int request_resource(struct resource *root, struct resource *new)
301{
302 struct resource *conflict;
303
304 conflict = request_resource_conflict(root, new);
305 return conflict ? -EBUSY : 0;
306}
307
308EXPORT_SYMBOL(request_resource);
309
310/**
311 * release_resource - release a previously reserved resource
312 * @old: resource pointer
313 */
314int release_resource(struct resource *old)
315{
316 int retval;
317
318 write_lock(&resource_lock);
319 retval = __release_resource(old, true);
320 write_unlock(&resource_lock);
321 return retval;
322}
323
324EXPORT_SYMBOL(release_resource);
325
326static bool is_type_match(struct resource *p, unsigned long flags, unsigned long desc)
327{
328 return (p->flags & flags) == flags && (desc == IORES_DESC_NONE || desc == p->desc);
329}
330
331/**
332 * find_next_res - Finds the lowest resource that covers part of
333 * [@start..@end].
334 *
335 * If a resource is found, returns 0 and @*res is overwritten with the part
336 * of the resource that's within [@start..@end]; if none is found, returns
337 * -ENODEV. Returns -EINVAL for invalid parameters.
338 *
339 * @parent: resource tree root to search
340 * @start: start address of the resource searched for
341 * @end: end address of same resource
342 * @flags: flags which the resource must have
343 * @desc: descriptor the resource must have
344 * @res: return ptr, if resource found
345 *
346 * The caller must specify @start, @end, @flags, and @desc
347 * (which may be IORES_DESC_NONE).
348 */
349static int find_next_res(struct resource *parent, resource_size_t start,
350 resource_size_t end, unsigned long flags,
351 unsigned long desc, struct resource *res)
352{
353 /* Skip children until we find a top level range that matches */
354 bool skip_children = true;
355 struct resource *p;
356
357 if (!res)
358 return -EINVAL;
359
360 if (start >= end)
361 return -EINVAL;
362
363 read_lock(&resource_lock);
364
365 for_each_resource(parent, p, skip_children) {
366 /* If we passed the resource we are looking for, stop */
367 if (p->start > end) {
368 p = NULL;
369 break;
370 }
371
372 /* Skip until we find a range that matches what we look for */
373 if (p->end < start)
374 continue;
375
376 /*
377 * We found a top level range that matches what we are looking
378 * for. Time to start checking children too.
379 */
380 skip_children = false;
381
382 /* Found a match, break */
383 if (is_type_match(p, flags, desc))
384 break;
385 }
386
387 if (p) {
388 /* copy data */
389 *res = (struct resource) {
390 .start = max(start, p->start),
391 .end = min(end, p->end),
392 .flags = p->flags,
393 .desc = p->desc,
394 .parent = p->parent,
395 };
396 }
397
398 read_unlock(&resource_lock);
399 return p ? 0 : -ENODEV;
400}
401
402static int find_next_iomem_res(resource_size_t start, resource_size_t end,
403 unsigned long flags, unsigned long desc,
404 struct resource *res)
405{
406 return find_next_res(&iomem_resource, start, end, flags, desc, res);
407}
408
409static int walk_res_desc(struct resource *parent, resource_size_t start,
410 resource_size_t end, unsigned long flags,
411 unsigned long desc, void *arg,
412 int (*func)(struct resource *, void *))
413{
414 struct resource res;
415 int ret = -EINVAL;
416
417 while (start < end &&
418 !find_next_res(parent, start, end, flags, desc, &res)) {
419 ret = (*func)(&res, arg);
420 if (ret)
421 break;
422
423 start = res.end + 1;
424 }
425
426 return ret;
427}
428
429static int __walk_iomem_res_desc(resource_size_t start, resource_size_t end,
430 unsigned long flags, unsigned long desc,
431 void *arg,
432 int (*func)(struct resource *, void *))
433{
434 return walk_res_desc(&iomem_resource, start, end, flags, desc, arg, func);
435}
436
437
438/**
439 * walk_iomem_res_desc - Walks through iomem resources and calls func()
440 * with matching resource ranges.
441 * *
442 * @desc: I/O resource descriptor. Use IORES_DESC_NONE to skip @desc check.
443 * @flags: I/O resource flags
444 * @start: start addr
445 * @end: end addr
446 * @arg: function argument for the callback @func
447 * @func: callback function that is called for each qualifying resource area
448 *
449 * All the memory ranges which overlap start,end and also match flags and
450 * desc are valid candidates.
451 *
452 * NOTE: For a new descriptor search, define a new IORES_DESC in
453 * <linux/ioport.h> and set it in 'desc' of a target resource entry.
454 */
455int walk_iomem_res_desc(unsigned long desc, unsigned long flags, u64 start,
456 u64 end, void *arg, int (*func)(struct resource *, void *))
457{
458 return __walk_iomem_res_desc(start, end, flags, desc, arg, func);
459}
460EXPORT_SYMBOL_GPL(walk_iomem_res_desc);
461
462/*
463 * In support of device drivers claiming Soft Reserved resources, walk the Soft
464 * Reserved resource deferral tree.
465 */
466int walk_soft_reserve_res(u64 start, u64 end, void *arg,
467 int (*func)(struct resource *, void *))
468{
469 return walk_res_desc(&soft_reserve_resource, start, end, IORESOURCE_MEM,
470 IORES_DESC_SOFT_RESERVED, arg, func);
471}
472EXPORT_SYMBOL_GPL(walk_soft_reserve_res);
473
474/*
475 * This function calls the @func callback against all memory ranges of type
476 * System RAM which are marked as IORESOURCE_SYSTEM_RAM and IORESOUCE_BUSY.
477 * Now, this function is only for System RAM, it deals with full ranges and
478 * not PFNs. If resources are not PFN-aligned, dealing with PFNs can truncate
479 * ranges.
480 */
481int walk_system_ram_res(u64 start, u64 end, void *arg,
482 int (*func)(struct resource *, void *))
483{
484 unsigned long flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
485
486 return __walk_iomem_res_desc(start, end, flags, IORES_DESC_NONE, arg,
487 func);
488}
489
490/*
491 * This function, being a variant of walk_system_ram_res(), calls the @func
492 * callback against all memory ranges of type System RAM which are marked as
493 * IORESOURCE_SYSTEM_RAM and IORESOUCE_BUSY in reversed order, i.e., from
494 * higher to lower.
495 */
496int walk_system_ram_res_rev(u64 start, u64 end, void *arg,
497 int (*func)(struct resource *, void *))
498{
499 struct resource res, *rams;
500 int rams_size = 16, i;
501 unsigned long flags;
502 int ret = -1;
503
504 /* create a list */
505 rams = kvzalloc_objs(struct resource, rams_size);
506 if (!rams)
507 return ret;
508
509 flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
510 i = 0;
511 while ((start < end) &&
512 (!find_next_iomem_res(start, end, flags, IORES_DESC_NONE, &res))) {
513 if (i >= rams_size) {
514 /* re-alloc */
515 struct resource *rams_new;
516
517 rams_new = kvrealloc(rams, (rams_size + 16) * sizeof(struct resource),
518 GFP_KERNEL);
519 if (!rams_new)
520 goto out;
521
522 rams = rams_new;
523 rams_size += 16;
524 }
525
526 rams[i++] = res;
527 start = res.end + 1;
528 }
529
530 /* go reverse */
531 for (i--; i >= 0; i--) {
532 ret = (*func)(&rams[i], arg);
533 if (ret)
534 break;
535 }
536
537out:
538 kvfree(rams);
539 return ret;
540}
541
542/*
543 * This function calls the @func callback against all memory ranges, which
544 * are ranges marked as IORESOURCE_MEM and IORESOUCE_BUSY.
545 */
546int walk_mem_res(u64 start, u64 end, void *arg,
547 int (*func)(struct resource *, void *))
548{
549 unsigned long flags = IORESOURCE_MEM | IORESOURCE_BUSY;
550
551 return __walk_iomem_res_desc(start, end, flags, IORES_DESC_NONE, arg,
552 func);
553}
554
555/*
556 * This function calls the @func callback against all memory ranges of type
557 * System RAM which are marked as IORESOURCE_SYSTEM_RAM and IORESOUCE_BUSY.
558 * It is to be used only for System RAM.
559 */
560int walk_system_ram_range(unsigned long start_pfn, unsigned long nr_pages,
561 void *arg, int (*func)(unsigned long, unsigned long, void *))
562{
563 resource_size_t start, end;
564 unsigned long flags;
565 struct resource res;
566 unsigned long pfn, end_pfn;
567 int ret = -EINVAL;
568
569 start = (u64) start_pfn << PAGE_SHIFT;
570 end = ((u64)(start_pfn + nr_pages) << PAGE_SHIFT) - 1;
571 flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
572 while (start < end &&
573 !find_next_iomem_res(start, end, flags, IORES_DESC_NONE, &res)) {
574 pfn = PFN_UP(res.start);
575 end_pfn = PFN_DOWN(res.end + 1);
576 if (end_pfn > pfn)
577 ret = (*func)(pfn, end_pfn - pfn, arg);
578 if (ret)
579 break;
580 start = res.end + 1;
581 }
582 return ret;
583}
584
585static int __is_ram(unsigned long pfn, unsigned long nr_pages, void *arg)
586{
587 return 1;
588}
589
590/*
591 * This generic page_is_ram() returns true if specified address is
592 * registered as System RAM in iomem_resource list.
593 */
594int __weak page_is_ram(unsigned long pfn)
595{
596 return walk_system_ram_range(pfn, 1, NULL, __is_ram) == 1;
597}
598EXPORT_SYMBOL_GPL(page_is_ram);
599
600static int __region_intersects(struct resource *parent, resource_size_t start,
601 size_t size, unsigned long flags,
602 unsigned long desc)
603{
604 int type = 0; int other = 0;
605 struct resource *p, *dp;
606 struct resource res, o;
607 bool covered;
608
609 res = DEFINE_RES(start, size, 0);
610
611 for (p = parent->child; p ; p = p->sibling) {
612 if (!resource_intersection(p, &res, &o))
613 continue;
614 if (is_type_match(p, flags, desc)) {
615 type++;
616 continue;
617 }
618 /*
619 * Continue to search in descendant resources as if the
620 * matched descendant resources cover some ranges of 'p'.
621 *
622 * |------------- "CXL Window 0" ------------|
623 * |-- "System RAM" --|
624 *
625 * will behave similar as the following fake resource
626 * tree when searching "System RAM".
627 *
628 * |-- "System RAM" --||-- "CXL Window 0a" --|
629 */
630 covered = false;
631 for_each_resource(p, dp, false) {
632 if (!resource_overlaps(dp, &res))
633 continue;
634 if (is_type_match(dp, flags, desc)) {
635 type++;
636 /*
637 * Range from 'o.start' to 'dp->start'
638 * isn't covered by matched resource.
639 */
640 if (dp->start > o.start)
641 break;
642 if (dp->end >= o.end) {
643 covered = true;
644 break;
645 }
646 /* Remove covered range */
647 o.start = max(o.start, dp->end + 1);
648 }
649 }
650 if (!covered)
651 other++;
652 }
653
654 if (type == 0)
655 return REGION_DISJOINT;
656
657 if (other == 0)
658 return REGION_INTERSECTS;
659
660 return REGION_MIXED;
661}
662
663/**
664 * region_intersects() - determine intersection of region with known resources
665 * @start: region start address
666 * @size: size of region
667 * @flags: flags of resource (in iomem_resource)
668 * @desc: descriptor of resource (in iomem_resource) or IORES_DESC_NONE
669 *
670 * Check if the specified region partially overlaps or fully eclipses a
671 * resource identified by @flags and @desc (optional with IORES_DESC_NONE).
672 * Return REGION_DISJOINT if the region does not overlap @flags/@desc,
673 * return REGION_MIXED if the region overlaps @flags/@desc and another
674 * resource, and return REGION_INTERSECTS if the region overlaps @flags/@desc
675 * and no other defined resource. Note that REGION_INTERSECTS is also
676 * returned in the case when the specified region overlaps RAM and undefined
677 * memory holes.
678 *
679 * region_intersect() is used by memory remapping functions to ensure
680 * the user is not remapping RAM and is a vast speed up over walking
681 * through the resource table page by page.
682 */
683int region_intersects(resource_size_t start, size_t size, unsigned long flags,
684 unsigned long desc)
685{
686 int ret;
687
688 read_lock(&resource_lock);
689 ret = __region_intersects(&iomem_resource, start, size, flags, desc);
690 read_unlock(&resource_lock);
691
692 return ret;
693}
694EXPORT_SYMBOL_GPL(region_intersects);
695
696/*
697 * Check if the provided range is registered in the Soft Reserved resource
698 * deferral tree for driver consideration.
699 */
700int region_intersects_soft_reserve(resource_size_t start, size_t size)
701{
702 guard(read_lock)(&resource_lock);
703 return __region_intersects(&soft_reserve_resource, start, size,
704 IORESOURCE_MEM, IORES_DESC_SOFT_RESERVED);
705}
706EXPORT_SYMBOL_GPL(region_intersects_soft_reserve);
707
708void __weak arch_remove_reservations(struct resource *avail)
709{
710}
711
712static void resource_clip(struct resource *res, resource_size_t min,
713 resource_size_t max)
714{
715 if (res->start < min)
716 res->start = min;
717 if (res->end > max)
718 res->end = max;
719}
720
721/*
722 * Find empty space in the resource tree with the given range and
723 * alignment constraints
724 */
725static int __find_resource_space(struct resource *root, struct resource *old,
726 struct resource *new, resource_size_t size,
727 struct resource_constraint *constraint)
728{
729 struct resource *this = root->child;
730 struct resource full_avail = *new, avail, alloc;
731 resource_alignf alignf = constraint->alignf;
732
733 full_avail.start = root->start;
734 /*
735 * Skip past an allocated resource that starts at 0, since the assignment
736 * of this->start - 1 to full_avail->end below would cause an underflow.
737 */
738 if (this && this->start == root->start) {
739 full_avail.start = (this == old) ? old->start : this->end + 1;
740 this = this->sibling;
741 }
742 for(;;) {
743 if (this)
744 full_avail.end = (this == old) ? this->end : this->start - 1;
745 else
746 full_avail.end = root->end;
747
748 if (full_avail.end < full_avail.start)
749 goto next;
750
751 resource_clip(&full_avail, constraint->min, constraint->max);
752 arch_remove_reservations(&full_avail);
753
754 /* Check for overflow after ALIGN() */
755 avail.start = ALIGN(full_avail.start, constraint->align);
756 avail.end = full_avail.end;
757 avail.flags = new->flags;
758 if (avail.start >= full_avail.start) {
759 alloc.flags = avail.flags;
760 if (alignf) {
761 alloc.start = alignf(constraint->alignf_data,
762 &avail, &full_avail,
763 size, constraint->align);
764 } else {
765 alloc.start = avail.start;
766 }
767 alloc.end = alloc.start + size - 1;
768 if (alloc.start <= alloc.end &&
769 __resource_contains_unbound(&full_avail, &alloc)) {
770 new->start = alloc.start;
771 new->end = alloc.end;
772 return 0;
773 }
774 }
775
776next: if (!this || this->end == root->end)
777 break;
778
779 if (this != old)
780 full_avail.start = this->end + 1;
781 this = this->sibling;
782 }
783 return -EBUSY;
784}
785
786/**
787 * find_resource_space - Find empty space in the resource tree
788 * @root: Root resource descriptor
789 * @new: Resource descriptor awaiting an empty resource space
790 * @size: The minimum size of the empty space
791 * @constraint: The range and alignment constraints to be met
792 *
793 * Finds an empty space under @root in the resource tree satisfying range and
794 * alignment @constraints.
795 *
796 * Return:
797 * * %0 - if successful, @new members start, end, and flags are altered.
798 * * %-EBUSY - if no empty space was found.
799 */
800int find_resource_space(struct resource *root, struct resource *new,
801 resource_size_t size,
802 struct resource_constraint *constraint)
803{
804 return __find_resource_space(root, NULL, new, size, constraint);
805}
806EXPORT_SYMBOL_GPL(find_resource_space);
807
808/**
809 * reallocate_resource - allocate a slot in the resource tree given range & alignment.
810 * The resource will be relocated if the new size cannot be reallocated in the
811 * current location.
812 *
813 * @root: root resource descriptor
814 * @old: resource descriptor desired by caller
815 * @newsize: new size of the resource descriptor
816 * @constraint: the memory range and alignment constraints to be met.
817 */
818static int reallocate_resource(struct resource *root, struct resource *old,
819 resource_size_t newsize,
820 struct resource_constraint *constraint)
821{
822 int err=0;
823 struct resource new = *old;
824 struct resource *conflict;
825
826 write_lock(&resource_lock);
827
828 if ((err = __find_resource_space(root, old, &new, newsize, constraint)))
829 goto out;
830
831 if (resource_contains(&new, old)) {
832 old->start = new.start;
833 old->end = new.end;
834 goto out;
835 }
836
837 if (old->child) {
838 err = -EBUSY;
839 goto out;
840 }
841
842 if (resource_contains(old, &new)) {
843 old->start = new.start;
844 old->end = new.end;
845 } else {
846 __release_resource(old, true);
847 *old = new;
848 conflict = __request_resource(root, old);
849 BUG_ON(conflict);
850 }
851out:
852 write_unlock(&resource_lock);
853 return err;
854}
855
856
857/**
858 * allocate_resource - allocate empty slot in the resource tree given range & alignment.
859 * The resource will be reallocated with a new size if it was already allocated
860 * @root: root resource descriptor
861 * @new: resource descriptor desired by caller
862 * @size: requested resource region size
863 * @min: minimum boundary to allocate
864 * @max: maximum boundary to allocate
865 * @align: alignment requested, in bytes
866 * @alignf: alignment function, optional, called if not NULL
867 * @alignf_data: arbitrary data to pass to the @alignf function
868 */
869int allocate_resource(struct resource *root, struct resource *new,
870 resource_size_t size, resource_size_t min,
871 resource_size_t max, resource_size_t align,
872 resource_alignf alignf,
873 void *alignf_data)
874{
875 int err;
876 struct resource_constraint constraint;
877
878 constraint.min = min;
879 constraint.max = max;
880 constraint.align = align;
881 constraint.alignf = alignf;
882 constraint.alignf_data = alignf_data;
883
884 if ( new->parent ) {
885 /* resource is already allocated, try reallocating with
886 the new constraints */
887 return reallocate_resource(root, new, size, &constraint);
888 }
889
890 write_lock(&resource_lock);
891 err = find_resource_space(root, new, size, &constraint);
892 if (err >= 0 && __request_resource(root, new))
893 err = -EBUSY;
894 write_unlock(&resource_lock);
895 return err;
896}
897
898EXPORT_SYMBOL(allocate_resource);
899
900/**
901 * lookup_resource - find an existing resource by a resource start address
902 * @root: root resource descriptor
903 * @start: resource start address
904 *
905 * Returns a pointer to the resource if found, NULL otherwise
906 */
907struct resource *lookup_resource(struct resource *root, resource_size_t start)
908{
909 struct resource *res;
910
911 read_lock(&resource_lock);
912 for (res = root->child; res; res = res->sibling) {
913 if (res->start == start)
914 break;
915 }
916 read_unlock(&resource_lock);
917
918 return res;
919}
920
921/*
922 * Insert a resource into the resource tree. If successful, return NULL,
923 * otherwise return the conflicting resource (compare to __request_resource())
924 */
925static struct resource * __insert_resource(struct resource *parent, struct resource *new)
926{
927 struct resource *first, *next;
928
929 for (;; parent = first) {
930 first = __request_resource(parent, new);
931 if (!first)
932 return first;
933
934 if (first == parent)
935 return first;
936 if (WARN_ON(first == new)) /* duplicated insertion */
937 return first;
938
939 if ((first->start > new->start) || (first->end < new->end))
940 break;
941 if ((first->start == new->start) && (first->end == new->end))
942 break;
943 }
944
945 for (next = first; ; next = next->sibling) {
946 /* Partial overlap? Bad, and unfixable */
947 if (next->start < new->start || next->end > new->end)
948 return next;
949 if (!next->sibling)
950 break;
951 if (next->sibling->start > new->end)
952 break;
953 }
954
955 new->parent = parent;
956 new->sibling = next->sibling;
957 new->child = first;
958
959 next->sibling = NULL;
960 for (next = first; next; next = next->sibling)
961 next->parent = new;
962
963 if (parent->child == first) {
964 parent->child = new;
965 } else {
966 next = parent->child;
967 while (next->sibling != first)
968 next = next->sibling;
969 next->sibling = new;
970 }
971 return NULL;
972}
973
974/**
975 * insert_resource_conflict - Inserts resource in the resource tree
976 * @parent: parent of the new resource
977 * @new: new resource to insert
978 *
979 * Returns 0 on success, conflict resource if the resource can't be inserted.
980 *
981 * This function is equivalent to request_resource_conflict when no conflict
982 * happens. If a conflict happens, and the conflicting resources
983 * entirely fit within the range of the new resource, then the new
984 * resource is inserted and the conflicting resources become children of
985 * the new resource.
986 *
987 * This function is intended for producers of resources, such as FW modules
988 * and bus drivers.
989 */
990struct resource *insert_resource_conflict(struct resource *parent, struct resource *new)
991{
992 struct resource *conflict;
993
994 write_lock(&resource_lock);
995 conflict = __insert_resource(parent, new);
996 write_unlock(&resource_lock);
997 return conflict;
998}
999
1000/**
1001 * insert_resource - Inserts a resource in the resource tree
1002 * @parent: parent of the new resource
1003 * @new: new resource to insert
1004 *
1005 * Returns 0 on success, -EBUSY if the resource can't be inserted.
1006 *
1007 * This function is intended for producers of resources, such as FW modules
1008 * and bus drivers.
1009 */
1010int insert_resource(struct resource *parent, struct resource *new)
1011{
1012 struct resource *conflict;
1013
1014 conflict = insert_resource_conflict(parent, new);
1015 return conflict ? -EBUSY : 0;
1016}
1017EXPORT_SYMBOL_GPL(insert_resource);
1018
1019/**
1020 * insert_resource_expand_to_fit - Insert a resource into the resource tree
1021 * @root: root resource descriptor
1022 * @new: new resource to insert
1023 *
1024 * Insert a resource into the resource tree, possibly expanding it in order
1025 * to make it encompass any conflicting resources.
1026 */
1027void insert_resource_expand_to_fit(struct resource *root, struct resource *new)
1028{
1029 if (new->parent)
1030 return;
1031
1032 write_lock(&resource_lock);
1033 for (;;) {
1034 struct resource *conflict;
1035
1036 conflict = __insert_resource(root, new);
1037 if (!conflict)
1038 break;
1039 if (conflict == root)
1040 break;
1041
1042 /* Ok, expand resource to cover the conflict, then try again .. */
1043 if (conflict->start < new->start)
1044 new->start = conflict->start;
1045 if (conflict->end > new->end)
1046 new->end = conflict->end;
1047
1048 pr_info("Expanded resource %s due to conflict with %s\n", new->name, conflict->name);
1049 }
1050 write_unlock(&resource_lock);
1051}
1052/*
1053 * Not for general consumption, only early boot memory map parsing, PCI
1054 * resource discovery, and late discovery of CXL resources are expected
1055 * to use this interface. The former are built-in and only the latter,
1056 * CXL, is a module.
1057 */
1058EXPORT_SYMBOL_NS_GPL(insert_resource_expand_to_fit, "CXL");
1059
1060/**
1061 * remove_resource - Remove a resource in the resource tree
1062 * @old: resource to remove
1063 *
1064 * Returns 0 on success, -EINVAL if the resource is not valid.
1065 *
1066 * This function removes a resource previously inserted by insert_resource()
1067 * or insert_resource_conflict(), and moves the children (if any) up to
1068 * where they were before. insert_resource() and insert_resource_conflict()
1069 * insert a new resource, and move any conflicting resources down to the
1070 * children of the new resource.
1071 *
1072 * insert_resource(), insert_resource_conflict() and remove_resource() are
1073 * intended for producers of resources, such as FW modules and bus drivers.
1074 */
1075int remove_resource(struct resource *old)
1076{
1077 int retval;
1078
1079 write_lock(&resource_lock);
1080 retval = __release_resource(old, false);
1081 write_unlock(&resource_lock);
1082 return retval;
1083}
1084EXPORT_SYMBOL_GPL(remove_resource);
1085
1086static int __adjust_resource(struct resource *res, resource_size_t start,
1087 resource_size_t size)
1088{
1089 struct resource *tmp, *parent = res->parent;
1090 resource_size_t end = start + size - 1;
1091 int result = -EBUSY;
1092
1093 if (!parent)
1094 goto skip;
1095
1096 if ((start < parent->start) || (end > parent->end))
1097 goto out;
1098
1099 if (res->sibling && (res->sibling->start <= end))
1100 goto out;
1101
1102 tmp = parent->child;
1103 if (tmp != res) {
1104 while (tmp->sibling != res)
1105 tmp = tmp->sibling;
1106 if (start <= tmp->end)
1107 goto out;
1108 }
1109
1110skip:
1111 for (tmp = res->child; tmp; tmp = tmp->sibling)
1112 if ((tmp->start < start) || (tmp->end > end))
1113 goto out;
1114
1115 res->start = start;
1116 res->end = end;
1117 result = 0;
1118
1119 out:
1120 return result;
1121}
1122
1123/**
1124 * adjust_resource - modify a resource's start and size
1125 * @res: resource to modify
1126 * @start: new start value
1127 * @size: new size
1128 *
1129 * Given an existing resource, change its start and size to match the
1130 * arguments. Returns 0 on success, -EBUSY if it can't fit.
1131 * Existing children of the resource are assumed to be immutable.
1132 */
1133int adjust_resource(struct resource *res, resource_size_t start,
1134 resource_size_t size)
1135{
1136 int result;
1137
1138 write_lock(&resource_lock);
1139 result = __adjust_resource(res, start, size);
1140 write_unlock(&resource_lock);
1141 return result;
1142}
1143EXPORT_SYMBOL(adjust_resource);
1144
1145static void __init
1146__reserve_region_with_split(struct resource *root, resource_size_t start,
1147 resource_size_t end, const char *name)
1148{
1149 struct resource *parent = root;
1150 struct resource *conflict;
1151 struct resource *res = alloc_resource(GFP_ATOMIC);
1152 struct resource *next_res = NULL;
1153 int type = resource_type(root);
1154
1155 if (!res)
1156 return;
1157
1158 res->name = name;
1159 res->start = start;
1160 res->end = end;
1161 res->flags = type | IORESOURCE_BUSY;
1162 res->desc = IORES_DESC_NONE;
1163
1164 while (1) {
1165
1166 conflict = __request_resource(parent, res);
1167 if (!conflict) {
1168 if (!next_res)
1169 break;
1170 res = next_res;
1171 next_res = NULL;
1172 continue;
1173 }
1174
1175 /* conflict covered whole area */
1176 if (conflict->start <= res->start &&
1177 conflict->end >= res->end) {
1178 free_resource(res);
1179 WARN_ON(next_res);
1180 break;
1181 }
1182
1183 /* failed, split and try again */
1184 if (conflict->start > res->start) {
1185 end = res->end;
1186 res->end = conflict->start - 1;
1187 if (conflict->end < end) {
1188 next_res = alloc_resource(GFP_ATOMIC);
1189 if (!next_res) {
1190 free_resource(res);
1191 break;
1192 }
1193 next_res->name = name;
1194 next_res->start = conflict->end + 1;
1195 next_res->end = end;
1196 next_res->flags = type | IORESOURCE_BUSY;
1197 next_res->desc = IORES_DESC_NONE;
1198 }
1199 } else {
1200 res->start = conflict->end + 1;
1201 }
1202 }
1203
1204}
1205
1206void __init
1207reserve_region_with_split(struct resource *root, resource_size_t start,
1208 resource_size_t end, const char *name)
1209{
1210 int abort = 0;
1211
1212 write_lock(&resource_lock);
1213 if (root->start > start || root->end < end) {
1214 pr_err("requested range [0x%llx-0x%llx] not in root %pr\n",
1215 (unsigned long long)start, (unsigned long long)end,
1216 root);
1217 if (start > root->end || end < root->start)
1218 abort = 1;
1219 else {
1220 if (end > root->end)
1221 end = root->end;
1222 if (start < root->start)
1223 start = root->start;
1224 pr_err("fixing request to [0x%llx-0x%llx]\n",
1225 (unsigned long long)start,
1226 (unsigned long long)end);
1227 }
1228 dump_stack();
1229 }
1230 if (!abort)
1231 __reserve_region_with_split(root, start, end, name);
1232 write_unlock(&resource_lock);
1233}
1234
1235/**
1236 * resource_alignment - calculate resource's alignment
1237 * @res: resource pointer
1238 *
1239 * Returns alignment on success, 0 (invalid alignment) on failure.
1240 */
1241resource_size_t resource_alignment(struct resource *res)
1242{
1243 switch (res->flags & (IORESOURCE_SIZEALIGN | IORESOURCE_STARTALIGN)) {
1244 case IORESOURCE_SIZEALIGN:
1245 return resource_size(res);
1246 case IORESOURCE_STARTALIGN:
1247 return res->start;
1248 default:
1249 return 0;
1250 }
1251}
1252
1253/*
1254 * This is compatibility stuff for IO resources.
1255 *
1256 * Note how this, unlike the above, knows about
1257 * the IO flag meanings (busy etc).
1258 *
1259 * request_region creates a new busy region.
1260 *
1261 * release_region releases a matching busy region.
1262 */
1263
1264static DECLARE_WAIT_QUEUE_HEAD(muxed_resource_wait);
1265
1266static struct inode *iomem_inode;
1267
1268#ifdef CONFIG_IO_STRICT_DEVMEM
1269static void revoke_iomem(struct resource *res)
1270{
1271 /* pairs with smp_store_release() in iomem_init_inode() */
1272 struct inode *inode = smp_load_acquire(&iomem_inode);
1273
1274 /*
1275 * Check that the initialization has completed. Losing the race
1276 * is ok because it means drivers are claiming resources before
1277 * the fs_initcall level of init and prevent iomem_get_mapping users
1278 * from establishing mappings.
1279 */
1280 if (!inode)
1281 return;
1282
1283 /*
1284 * The expectation is that the driver has successfully marked
1285 * the resource busy by this point, so devmem_is_allowed()
1286 * should start returning false, however for performance this
1287 * does not iterate the entire resource range.
1288 */
1289 if (devmem_is_allowed(PHYS_PFN(res->start)) &&
1290 devmem_is_allowed(PHYS_PFN(res->end))) {
1291 /*
1292 * *cringe* iomem=relaxed says "go ahead, what's the
1293 * worst that can happen?"
1294 */
1295 return;
1296 }
1297
1298 unmap_mapping_range(inode->i_mapping, res->start, resource_size(res), 1);
1299}
1300#else
1301static void revoke_iomem(struct resource *res) {}
1302#endif
1303
1304struct address_space *iomem_get_mapping(void)
1305{
1306 /*
1307 * This function is only called from file open paths, hence guaranteed
1308 * that fs_initcalls have completed and no need to check for NULL. But
1309 * since revoke_iomem can be called before the initcall we still need
1310 * the barrier to appease checkers.
1311 */
1312 return smp_load_acquire(&iomem_inode)->i_mapping;
1313}
1314
1315static int __request_region_locked(struct resource *res, struct resource *parent,
1316 resource_size_t start, resource_size_t n,
1317 const char *name, int flags)
1318{
1319 DECLARE_WAITQUEUE(wait, current);
1320
1321 res->name = name;
1322 res->start = start;
1323 res->end = start + n - 1;
1324
1325 for (;;) {
1326 struct resource *conflict;
1327
1328 res->flags = resource_type(parent) | resource_ext_type(parent);
1329 res->flags |= IORESOURCE_BUSY | flags;
1330 res->desc = parent->desc;
1331
1332 conflict = __request_resource(parent, res);
1333 if (!conflict)
1334 break;
1335 /*
1336 * mm/hmm.c reserves physical addresses which then
1337 * become unavailable to other users. Conflicts are
1338 * not expected. Warn to aid debugging if encountered.
1339 */
1340 if (parent == &iomem_resource &&
1341 conflict->desc == IORES_DESC_DEVICE_PRIVATE_MEMORY) {
1342 pr_warn("Unaddressable device %s %pR conflicts with %pR\n",
1343 conflict->name, conflict, res);
1344 }
1345 if (conflict != parent) {
1346 if (!(conflict->flags & IORESOURCE_BUSY)) {
1347 parent = conflict;
1348 continue;
1349 }
1350 }
1351 if (conflict->flags & flags & IORESOURCE_MUXED) {
1352 add_wait_queue(&muxed_resource_wait, &wait);
1353 write_unlock(&resource_lock);
1354 set_current_state(TASK_UNINTERRUPTIBLE);
1355 schedule();
1356 remove_wait_queue(&muxed_resource_wait, &wait);
1357 write_lock(&resource_lock);
1358 continue;
1359 }
1360 /* Uhhuh, that didn't work out.. */
1361 return -EBUSY;
1362 }
1363
1364 return 0;
1365}
1366
1367/**
1368 * __request_region - create a new busy resource region
1369 * @parent: parent resource descriptor
1370 * @start: resource start address
1371 * @n: resource region size
1372 * @name: reserving caller's ID string
1373 * @flags: IO resource flags
1374 */
1375struct resource *__request_region(struct resource *parent,
1376 resource_size_t start, resource_size_t n,
1377 const char *name, int flags)
1378{
1379 struct resource *res = alloc_resource(GFP_KERNEL);
1380 int ret;
1381
1382 if (!res)
1383 return NULL;
1384
1385 write_lock(&resource_lock);
1386 ret = __request_region_locked(res, parent, start, n, name, flags);
1387 write_unlock(&resource_lock);
1388
1389 if (ret) {
1390 free_resource(res);
1391 return NULL;
1392 }
1393
1394 if (parent == &iomem_resource)
1395 revoke_iomem(res);
1396
1397 return res;
1398}
1399EXPORT_SYMBOL(__request_region);
1400
1401/**
1402 * __release_region - release a previously reserved resource region
1403 * @parent: parent resource descriptor
1404 * @start: resource start address
1405 * @n: resource region size
1406 *
1407 * The described resource region must match a currently busy region.
1408 */
1409void __release_region(struct resource *parent, resource_size_t start,
1410 resource_size_t n)
1411{
1412 struct resource **p;
1413 resource_size_t end;
1414
1415 p = &parent->child;
1416 end = start + n - 1;
1417
1418 write_lock(&resource_lock);
1419
1420 for (;;) {
1421 struct resource *res = *p;
1422
1423 if (!res)
1424 break;
1425 if (res->start <= start && res->end >= end) {
1426 if (!(res->flags & IORESOURCE_BUSY)) {
1427 p = &res->child;
1428 continue;
1429 }
1430 if (res->start != start || res->end != end)
1431 break;
1432 *p = res->sibling;
1433 write_unlock(&resource_lock);
1434 if (res->flags & IORESOURCE_MUXED)
1435 wake_up(&muxed_resource_wait);
1436 free_resource(res);
1437 return;
1438 }
1439 p = &res->sibling;
1440 }
1441
1442 write_unlock(&resource_lock);
1443
1444 pr_warn("Trying to free nonexistent resource <%pa-%pa>\n", &start, &end);
1445}
1446EXPORT_SYMBOL(__release_region);
1447
1448#ifdef CONFIG_MEMORY_HOTREMOVE
1449static void append_child_to_parent(struct resource *new_parent, struct resource *new_child)
1450{
1451 struct resource *child;
1452
1453 child = new_parent->child;
1454 if (child) {
1455 while (child->sibling)
1456 child = child->sibling;
1457 child->sibling = new_child;
1458 } else {
1459 new_parent->child = new_child;
1460 }
1461 new_child->parent = new_parent;
1462 new_child->sibling = NULL;
1463}
1464
1465/*
1466 * Reparent all child resources that no longer belong to "low" after a split to
1467 * "high". Note that "high" does not have any children, because "low" is the
1468 * original resource and "high" is a new resource. Treat "low" as the original
1469 * resource being split and defer its range adjustment to __adjust_resource().
1470 */
1471static void reparent_children_after_split(struct resource *low,
1472 struct resource *high,
1473 resource_size_t split_addr)
1474{
1475 struct resource *child, *next, **p;
1476
1477 p = &low->child;
1478 while ((child = *p)) {
1479 next = child->sibling;
1480 if (child->start > split_addr) {
1481 /* unlink child */
1482 *p = next;
1483 append_child_to_parent(high, child);
1484 } else {
1485 p = &child->sibling;
1486 }
1487 }
1488}
1489
1490/**
1491 * release_mem_region_adjustable - release a previously reserved memory region
1492 * @start: resource start address
1493 * @size: resource region size
1494 *
1495 * This interface is intended for memory hot-delete. The requested region
1496 * is released from a currently busy memory resource. The requested region
1497 * must either match exactly or fit into a single busy resource entry. In
1498 * the latter case, the remaining resource is adjusted accordingly.
1499 *
1500 * Note:
1501 * - Additional release conditions, such as overlapping region, can be
1502 * supported after they are confirmed as valid cases.
1503 * - When a busy memory resource gets split into two entries, its children are
1504 * reassigned to the correct parent based on their range. If a child memory
1505 * resource overlaps with more than one parent, enhance the logic as needed.
1506 */
1507void release_mem_region_adjustable(resource_size_t start, resource_size_t size)
1508{
1509 struct resource *parent = &iomem_resource;
1510 struct resource *new_res = NULL;
1511 bool alloc_nofail = false;
1512 struct resource **p;
1513 struct resource *res;
1514 resource_size_t end;
1515
1516 end = start + size - 1;
1517 if (WARN_ON_ONCE((start < parent->start) || (end > parent->end)))
1518 return;
1519
1520 /*
1521 * We free up quite a lot of memory on memory hotunplug (esp., memap),
1522 * just before releasing the region. This is highly unlikely to
1523 * fail - let's play save and make it never fail as the caller cannot
1524 * perform any error handling (e.g., trying to re-add memory will fail
1525 * similarly).
1526 */
1527retry:
1528 new_res = alloc_resource(GFP_KERNEL | (alloc_nofail ? __GFP_NOFAIL : 0));
1529
1530 p = &parent->child;
1531 write_lock(&resource_lock);
1532
1533 while ((res = *p)) {
1534 if (res->start >= end)
1535 break;
1536
1537 /* look for the next resource if it does not fit into */
1538 if (res->start > start || res->end < end) {
1539 p = &res->sibling;
1540 continue;
1541 }
1542
1543 if (!(res->flags & IORESOURCE_MEM))
1544 break;
1545
1546 if (!(res->flags & IORESOURCE_BUSY)) {
1547 p = &res->child;
1548 continue;
1549 }
1550
1551 /* found the target resource; let's adjust accordingly */
1552 if (res->start == start && res->end == end) {
1553 /* free the whole entry */
1554 *p = res->sibling;
1555 free_resource(res);
1556 } else if (res->start == start && res->end != end) {
1557 /* adjust the start */
1558 WARN_ON_ONCE(__adjust_resource(res, end + 1,
1559 res->end - end));
1560 } else if (res->start != start && res->end == end) {
1561 /* adjust the end */
1562 WARN_ON_ONCE(__adjust_resource(res, res->start,
1563 start - res->start));
1564 } else {
1565 /* split into two entries - we need a new resource */
1566 if (!new_res) {
1567 new_res = alloc_resource(GFP_ATOMIC);
1568 if (!new_res) {
1569 alloc_nofail = true;
1570 write_unlock(&resource_lock);
1571 goto retry;
1572 }
1573 }
1574 new_res->name = res->name;
1575 new_res->start = end + 1;
1576 new_res->end = res->end;
1577 new_res->flags = res->flags;
1578 new_res->desc = res->desc;
1579 new_res->parent = res->parent;
1580 new_res->sibling = res->sibling;
1581 new_res->child = NULL;
1582 reparent_children_after_split(res, new_res, end);
1583
1584 if (WARN_ON_ONCE(__adjust_resource(res, res->start,
1585 start - res->start)))
1586 break;
1587 res->sibling = new_res;
1588 new_res = NULL;
1589 }
1590
1591 break;
1592 }
1593
1594 write_unlock(&resource_lock);
1595 free_resource(new_res);
1596}
1597#endif /* CONFIG_MEMORY_HOTREMOVE */
1598
1599#ifdef CONFIG_MEMORY_HOTPLUG
1600static bool system_ram_resources_mergeable(struct resource *r1,
1601 struct resource *r2)
1602{
1603 /* We assume either r1 or r2 is IORESOURCE_SYSRAM_MERGEABLE. */
1604 return r1->flags == r2->flags && r1->end + 1 == r2->start &&
1605 r1->name == r2->name && r1->desc == r2->desc &&
1606 !r1->child && !r2->child;
1607}
1608
1609/**
1610 * merge_system_ram_resource - mark the System RAM resource mergeable and try to
1611 * merge it with adjacent, mergeable resources
1612 * @res: resource descriptor
1613 *
1614 * This interface is intended for memory hotplug, whereby lots of contiguous
1615 * system ram resources are added (e.g., via add_memory*()) by a driver, and
1616 * the actual resource boundaries are not of interest (e.g., it might be
1617 * relevant for DIMMs). Only resources that are marked mergeable, that have the
1618 * same parent, and that don't have any children are considered. All mergeable
1619 * resources must be immutable during the request.
1620 *
1621 * Note:
1622 * - The caller has to make sure that no pointers to resources that are
1623 * marked mergeable are used anymore after this call - the resource might
1624 * be freed and the pointer might be stale!
1625 * - release_mem_region_adjustable() will split on demand on memory hotunplug
1626 */
1627void merge_system_ram_resource(struct resource *res)
1628{
1629 const unsigned long flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
1630 struct resource *cur;
1631
1632 if (WARN_ON_ONCE((res->flags & flags) != flags))
1633 return;
1634
1635 write_lock(&resource_lock);
1636 res->flags |= IORESOURCE_SYSRAM_MERGEABLE;
1637
1638 /* Try to merge with next item in the list. */
1639 cur = res->sibling;
1640 if (cur && system_ram_resources_mergeable(res, cur)) {
1641 res->end = cur->end;
1642 res->sibling = cur->sibling;
1643 free_resource(cur);
1644 }
1645
1646 /* Try to merge with previous item in the list. */
1647 cur = res->parent->child;
1648 while (cur && cur->sibling != res)
1649 cur = cur->sibling;
1650 if (cur && system_ram_resources_mergeable(cur, res)) {
1651 cur->end = res->end;
1652 cur->sibling = res->sibling;
1653 free_resource(res);
1654 }
1655 write_unlock(&resource_lock);
1656}
1657#endif /* CONFIG_MEMORY_HOTPLUG */
1658
1659/*
1660 * Managed region resource
1661 */
1662static void devm_resource_release(struct device *dev, void *ptr)
1663{
1664 struct resource **r = ptr;
1665
1666 release_resource(*r);
1667}
1668
1669/**
1670 * devm_request_resource() - request and reserve an I/O or memory resource
1671 * @dev: device for which to request the resource
1672 * @root: root of the resource tree from which to request the resource
1673 * @new: descriptor of the resource to request
1674 *
1675 * This is a device-managed version of request_resource(). There is usually
1676 * no need to release resources requested by this function explicitly since
1677 * that will be taken care of when the device is unbound from its driver.
1678 * If for some reason the resource needs to be released explicitly, because
1679 * of ordering issues for example, drivers must call devm_release_resource()
1680 * rather than the regular release_resource().
1681 *
1682 * When a conflict is detected between any existing resources and the newly
1683 * requested resource, an error message will be printed.
1684 *
1685 * Returns 0 on success or a negative error code on failure.
1686 */
1687int devm_request_resource(struct device *dev, struct resource *root,
1688 struct resource *new)
1689{
1690 struct resource *conflict, **ptr;
1691
1692 ptr = devres_alloc(devm_resource_release, sizeof(*ptr), GFP_KERNEL);
1693 if (!ptr)
1694 return -ENOMEM;
1695
1696 *ptr = new;
1697
1698 conflict = request_resource_conflict(root, new);
1699 if (conflict) {
1700 dev_err(dev, "resource collision: %pR conflicts with %s %pR\n",
1701 new, conflict->name, conflict);
1702 devres_free(ptr);
1703 return -EBUSY;
1704 }
1705
1706 devres_add(dev, ptr);
1707 return 0;
1708}
1709EXPORT_SYMBOL(devm_request_resource);
1710
1711static int devm_resource_match(struct device *dev, void *res, void *data)
1712{
1713 struct resource **ptr = res;
1714
1715 return *ptr == data;
1716}
1717
1718/**
1719 * devm_release_resource() - release a previously requested resource
1720 * @dev: device for which to release the resource
1721 * @new: descriptor of the resource to release
1722 *
1723 * Releases a resource previously requested using devm_request_resource().
1724 */
1725void devm_release_resource(struct device *dev, struct resource *new)
1726{
1727 WARN_ON(devres_release(dev, devm_resource_release, devm_resource_match,
1728 new));
1729}
1730EXPORT_SYMBOL(devm_release_resource);
1731
1732struct region_devres {
1733 struct resource *parent;
1734 resource_size_t start;
1735 resource_size_t n;
1736};
1737
1738static void devm_region_release(struct device *dev, void *res)
1739{
1740 struct region_devres *this = res;
1741
1742 __release_region(this->parent, this->start, this->n);
1743}
1744
1745static int devm_region_match(struct device *dev, void *res, void *match_data)
1746{
1747 struct region_devres *this = res, *match = match_data;
1748
1749 return this->parent == match->parent &&
1750 this->start == match->start && this->n == match->n;
1751}
1752
1753struct resource *
1754__devm_request_region(struct device *dev, struct resource *parent,
1755 resource_size_t start, resource_size_t n, const char *name)
1756{
1757 struct region_devres *dr = NULL;
1758 struct resource *res;
1759
1760 dr = devres_alloc(devm_region_release, sizeof(struct region_devres),
1761 GFP_KERNEL);
1762 if (!dr)
1763 return NULL;
1764
1765 dr->parent = parent;
1766 dr->start = start;
1767 dr->n = n;
1768
1769 res = __request_region(parent, start, n, name, 0);
1770 if (res)
1771 devres_add(dev, dr);
1772 else
1773 devres_free(dr);
1774
1775 return res;
1776}
1777EXPORT_SYMBOL(__devm_request_region);
1778
1779void __devm_release_region(struct device *dev, struct resource *parent,
1780 resource_size_t start, resource_size_t n)
1781{
1782 struct region_devres match_data = { parent, start, n };
1783
1784 WARN_ON(devres_release(dev, devm_region_release, devm_region_match,
1785 &match_data));
1786}
1787EXPORT_SYMBOL(__devm_release_region);
1788
1789/*
1790 * Reserve I/O ports or memory based on "reserve=" kernel parameter.
1791 */
1792#define MAXRESERVE 4
1793static int __init reserve_setup(char *str)
1794{
1795 static int reserved;
1796 static struct resource reserve[MAXRESERVE];
1797
1798 for (;;) {
1799 unsigned int io_start, io_num;
1800 int x = reserved;
1801 struct resource *parent;
1802
1803 if (get_option(&str, &io_start) != 2)
1804 break;
1805 if (get_option(&str, &io_num) == 0)
1806 break;
1807 if (x < MAXRESERVE) {
1808 struct resource *res = reserve + x;
1809
1810 /*
1811 * If the region starts below 0x10000, we assume it's
1812 * I/O port space; otherwise assume it's memory.
1813 */
1814 if (io_start < 0x10000) {
1815 *res = DEFINE_RES_IO_NAMED(io_start, io_num, "reserved");
1816 parent = &ioport_resource;
1817 } else {
1818 *res = DEFINE_RES_MEM_NAMED(io_start, io_num, "reserved");
1819 parent = &iomem_resource;
1820 }
1821 res->flags |= IORESOURCE_BUSY;
1822 if (request_resource(parent, res) == 0)
1823 reserved = x+1;
1824 }
1825 }
1826 return 1;
1827}
1828__setup("reserve=", reserve_setup);
1829
1830/*
1831 * Check if the requested addr and size spans more than any slot in the
1832 * iomem resource tree.
1833 */
1834int iomem_map_sanity_check(resource_size_t addr, unsigned long size)
1835{
1836 resource_size_t end = addr + size - 1;
1837 struct resource *p;
1838 int err = 0;
1839
1840 read_lock(&resource_lock);
1841 for_each_resource(&iomem_resource, p, false) {
1842 /*
1843 * We can probably skip the resources without
1844 * IORESOURCE_IO attribute?
1845 */
1846 if (p->start > end)
1847 continue;
1848 if (p->end < addr)
1849 continue;
1850 if (PFN_DOWN(p->start) <= PFN_DOWN(addr) &&
1851 PFN_DOWN(p->end) >= PFN_DOWN(end))
1852 continue;
1853 /*
1854 * if a resource is "BUSY", it's not a hardware resource
1855 * but a driver mapping of such a resource; we don't want
1856 * to warn for those; some drivers legitimately map only
1857 * partial hardware resources. (example: vesafb)
1858 */
1859 if (p->flags & IORESOURCE_BUSY)
1860 continue;
1861
1862 pr_warn("resource sanity check: requesting [mem %pa-%pa], which spans more than %s %pR\n",
1863 &addr, &end, p->name, p);
1864 err = -1;
1865 break;
1866 }
1867 read_unlock(&resource_lock);
1868
1869 return err;
1870}
1871
1872#ifdef CONFIG_STRICT_DEVMEM
1873static int strict_iomem_checks = 1;
1874#else
1875static int strict_iomem_checks;
1876#endif
1877
1878/*
1879 * Check if an address is exclusive to the kernel and must not be mapped to
1880 * user space, for example, via /dev/mem.
1881 *
1882 * Returns true if exclusive to the kernel, otherwise returns false.
1883 */
1884bool resource_is_exclusive(struct resource *root, u64 addr, resource_size_t size)
1885{
1886 const unsigned int exclusive_system_ram = IORESOURCE_SYSTEM_RAM |
1887 IORESOURCE_EXCLUSIVE;
1888 bool skip_children = false, err = false;
1889 struct resource *p;
1890
1891 read_lock(&resource_lock);
1892 for_each_resource(root, p, skip_children) {
1893 if (p->start >= addr + size)
1894 break;
1895 if (p->end < addr) {
1896 skip_children = true;
1897 continue;
1898 }
1899 skip_children = false;
1900
1901 /*
1902 * IORESOURCE_SYSTEM_RAM resources are exclusive if
1903 * IORESOURCE_EXCLUSIVE is set, even if they
1904 * are not busy and even if "iomem=relaxed" is set. The
1905 * responsible driver dynamically adds/removes system RAM within
1906 * such an area and uncontrolled access is dangerous.
1907 */
1908 if ((p->flags & exclusive_system_ram) == exclusive_system_ram) {
1909 err = true;
1910 break;
1911 }
1912
1913 /*
1914 * A resource is exclusive if IORESOURCE_EXCLUSIVE is set
1915 * or CONFIG_IO_STRICT_DEVMEM is enabled and the
1916 * resource is busy.
1917 */
1918 if (!strict_iomem_checks || !(p->flags & IORESOURCE_BUSY))
1919 continue;
1920 if (IS_ENABLED(CONFIG_IO_STRICT_DEVMEM)
1921 || p->flags & IORESOURCE_EXCLUSIVE) {
1922 err = true;
1923 break;
1924 }
1925 }
1926 read_unlock(&resource_lock);
1927
1928 return err;
1929}
1930
1931bool iomem_is_exclusive(u64 addr)
1932{
1933 return resource_is_exclusive(&iomem_resource, addr & PAGE_MASK,
1934 PAGE_SIZE);
1935}
1936
1937struct resource_entry *resource_list_create_entry(struct resource *res,
1938 size_t extra_size)
1939{
1940 struct resource_entry *entry;
1941
1942 entry = kzalloc(sizeof(*entry) + extra_size, GFP_KERNEL);
1943 if (entry) {
1944 INIT_LIST_HEAD(&entry->node);
1945 entry->res = res ? res : &entry->__res;
1946 }
1947
1948 return entry;
1949}
1950EXPORT_SYMBOL(resource_list_create_entry);
1951
1952void resource_list_free(struct list_head *head)
1953{
1954 struct resource_entry *entry, *tmp;
1955
1956 list_for_each_entry_safe(entry, tmp, head, node)
1957 resource_list_destroy_entry(entry);
1958}
1959EXPORT_SYMBOL(resource_list_free);
1960
1961#ifdef CONFIG_GET_FREE_REGION
1962#define GFR_DESCENDING (1UL << 0)
1963#define GFR_REQUEST_REGION (1UL << 1)
1964#ifdef PA_SECTION_SHIFT
1965#define GFR_DEFAULT_ALIGN (1UL << PA_SECTION_SHIFT)
1966#else
1967#define GFR_DEFAULT_ALIGN PAGE_SIZE
1968#endif
1969
1970static resource_size_t gfr_start(struct resource *base, resource_size_t size,
1971 resource_size_t align, unsigned long flags)
1972{
1973 if (flags & GFR_DESCENDING) {
1974 resource_size_t end;
1975
1976 end = min_t(resource_size_t, base->end, DIRECT_MAP_PHYSMEM_END);
1977 return end - size + 1;
1978 }
1979
1980 return ALIGN(max(base->start, align), align);
1981}
1982
1983static bool gfr_continue(struct resource *base, resource_size_t addr,
1984 resource_size_t size, unsigned long flags)
1985{
1986 if (flags & GFR_DESCENDING)
1987 return addr > size && addr >= base->start;
1988 /*
1989 * In the ascend case be careful that the last increment by
1990 * @size did not wrap 0.
1991 */
1992 return addr > addr - size &&
1993 addr <= min_t(resource_size_t, base->end, DIRECT_MAP_PHYSMEM_END);
1994}
1995
1996static resource_size_t gfr_next(resource_size_t addr, resource_size_t size,
1997 unsigned long flags)
1998{
1999 if (flags & GFR_DESCENDING)
2000 return addr - size;
2001 return addr + size;
2002}
2003
2004static void remove_free_mem_region(void *_res)
2005{
2006 struct resource *res = _res;
2007
2008 if (res->parent)
2009 remove_resource(res);
2010 free_resource(res);
2011}
2012
2013static struct resource *
2014get_free_mem_region(struct device *dev, struct resource *base,
2015 resource_size_t size, const unsigned long align,
2016 const char *name, const unsigned long desc,
2017 const unsigned long flags)
2018{
2019 resource_size_t addr;
2020 struct resource *res;
2021 struct region_devres *dr = NULL;
2022
2023 size = ALIGN(size, align);
2024
2025 res = alloc_resource(GFP_KERNEL);
2026 if (!res)
2027 return ERR_PTR(-ENOMEM);
2028
2029 if (dev && (flags & GFR_REQUEST_REGION)) {
2030 dr = devres_alloc(devm_region_release,
2031 sizeof(struct region_devres), GFP_KERNEL);
2032 if (!dr) {
2033 free_resource(res);
2034 return ERR_PTR(-ENOMEM);
2035 }
2036 } else if (dev) {
2037 if (devm_add_action_or_reset(dev, remove_free_mem_region, res))
2038 return ERR_PTR(-ENOMEM);
2039 }
2040
2041 write_lock(&resource_lock);
2042 for (addr = gfr_start(base, size, align, flags);
2043 gfr_continue(base, addr, align, flags);
2044 addr = gfr_next(addr, align, flags)) {
2045 if (__region_intersects(base, addr, size, 0, IORES_DESC_NONE) !=
2046 REGION_DISJOINT)
2047 continue;
2048
2049 if (flags & GFR_REQUEST_REGION) {
2050 if (__request_region_locked(res, &iomem_resource, addr,
2051 size, name, 0))
2052 break;
2053
2054 if (dev) {
2055 dr->parent = &iomem_resource;
2056 dr->start = addr;
2057 dr->n = size;
2058 devres_add(dev, dr);
2059 }
2060
2061 res->desc = desc;
2062 write_unlock(&resource_lock);
2063
2064
2065 /*
2066 * A driver is claiming this region so revoke any
2067 * mappings.
2068 */
2069 revoke_iomem(res);
2070 } else {
2071 *res = DEFINE_RES_NAMED_DESC(addr, size, name, IORESOURCE_MEM, desc);
2072
2073 /*
2074 * Only succeed if the resource hosts an exclusive
2075 * range after the insert
2076 */
2077 if (__insert_resource(base, res) || res->child)
2078 break;
2079
2080 write_unlock(&resource_lock);
2081 }
2082
2083 return res;
2084 }
2085 write_unlock(&resource_lock);
2086
2087 if (flags & GFR_REQUEST_REGION) {
2088 free_resource(res);
2089 devres_free(dr);
2090 } else if (dev)
2091 devm_release_action(dev, remove_free_mem_region, res);
2092
2093 return ERR_PTR(-ERANGE);
2094}
2095
2096/**
2097 * devm_request_free_mem_region - find free region for device private memory
2098 *
2099 * @dev: device struct to bind the resource to
2100 * @size: size in bytes of the device memory to add
2101 * @base: resource tree to look in
2102 *
2103 * This function tries to find an empty range of physical address big enough to
2104 * contain the new resource, so that it can later be hotplugged as ZONE_DEVICE
2105 * memory, which in turn allocates struct pages.
2106 */
2107struct resource *devm_request_free_mem_region(struct device *dev,
2108 struct resource *base, unsigned long size)
2109{
2110 unsigned long flags = GFR_DESCENDING | GFR_REQUEST_REGION;
2111
2112 return get_free_mem_region(dev, base, size, GFR_DEFAULT_ALIGN,
2113 dev_name(dev),
2114 IORES_DESC_DEVICE_PRIVATE_MEMORY, flags);
2115}
2116EXPORT_SYMBOL_GPL(devm_request_free_mem_region);
2117
2118struct resource *request_free_mem_region(struct resource *base,
2119 unsigned long size, const char *name)
2120{
2121 unsigned long flags = GFR_DESCENDING | GFR_REQUEST_REGION;
2122
2123 return get_free_mem_region(NULL, base, size, GFR_DEFAULT_ALIGN, name,
2124 IORES_DESC_DEVICE_PRIVATE_MEMORY, flags);
2125}
2126EXPORT_SYMBOL_GPL(request_free_mem_region);
2127
2128/**
2129 * alloc_free_mem_region - find a free region relative to @base
2130 * @base: resource that will parent the new resource
2131 * @size: size in bytes of memory to allocate from @base
2132 * @align: alignment requirements for the allocation
2133 * @name: resource name
2134 *
2135 * Buses like CXL, that can dynamically instantiate new memory regions,
2136 * need a method to allocate physical address space for those regions.
2137 * Allocate and insert a new resource to cover a free, unclaimed by a
2138 * descendant of @base, range in the span of @base.
2139 */
2140struct resource *alloc_free_mem_region(struct resource *base,
2141 unsigned long size, unsigned long align,
2142 const char *name)
2143{
2144 /* Default of ascending direction and insert resource */
2145 unsigned long flags = 0;
2146
2147 return get_free_mem_region(NULL, base, size, align, name,
2148 IORES_DESC_NONE, flags);
2149}
2150EXPORT_SYMBOL_GPL(alloc_free_mem_region);
2151#endif /* CONFIG_GET_FREE_REGION */
2152
2153static int __init strict_iomem(char *str)
2154{
2155 if (strstr(str, "relaxed"))
2156 strict_iomem_checks = 0;
2157 if (strstr(str, "strict"))
2158 strict_iomem_checks = 1;
2159 return 1;
2160}
2161
2162static int iomem_fs_init_fs_context(struct fs_context *fc)
2163{
2164 return init_pseudo(fc, DEVMEM_MAGIC) ? 0 : -ENOMEM;
2165}
2166
2167static struct file_system_type iomem_fs_type = {
2168 .name = "iomem",
2169 .owner = THIS_MODULE,
2170 .init_fs_context = iomem_fs_init_fs_context,
2171 .kill_sb = kill_anon_super,
2172};
2173
2174static int __init iomem_init_inode(void)
2175{
2176 static struct vfsmount *iomem_vfs_mount;
2177 static int iomem_fs_cnt;
2178 struct inode *inode;
2179 int rc;
2180
2181 rc = simple_pin_fs(&iomem_fs_type, &iomem_vfs_mount, &iomem_fs_cnt);
2182 if (rc < 0) {
2183 pr_err("Cannot mount iomem pseudo filesystem: %d\n", rc);
2184 return rc;
2185 }
2186
2187 inode = alloc_anon_inode(iomem_vfs_mount->mnt_sb);
2188 if (IS_ERR(inode)) {
2189 rc = PTR_ERR(inode);
2190 pr_err("Cannot allocate inode for iomem: %d\n", rc);
2191 simple_release_fs(&iomem_vfs_mount, &iomem_fs_cnt);
2192 return rc;
2193 }
2194
2195 /*
2196 * Publish iomem revocation inode initialized.
2197 * Pairs with smp_load_acquire() in revoke_iomem().
2198 */
2199 smp_store_release(&iomem_inode, inode);
2200
2201 return 0;
2202}
2203
2204fs_initcall(iomem_init_inode);
2205
2206__setup("iomem=", strict_iomem);