resource.c

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