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