linux/kernel/resource.c
<<
>>
Prefs
   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 <asm/io.h>
  31
  32
  33struct resource ioport_resource = {
  34        .name   = "PCI IO",
  35        .start  = 0,
  36        .end    = IO_SPACE_LIMIT,
  37        .flags  = IORESOURCE_IO,
  38};
  39EXPORT_SYMBOL(ioport_resource);
  40
  41struct resource iomem_resource = {
  42        .name   = "PCI mem",
  43        .start  = 0,
  44        .end    = -1,
  45        .flags  = IORESOURCE_MEM,
  46};
  47EXPORT_SYMBOL(iomem_resource);
  48
  49/* constraints to be met while allocating resources */
  50struct resource_constraint {
  51        resource_size_t min, max, align;
  52        resource_size_t (*alignf)(void *, const struct resource *,
  53                        resource_size_t, resource_size_t);
  54        void *alignf_data;
  55};
  56
  57static DEFINE_RWLOCK(resource_lock);
  58
  59static struct resource *next_resource(struct resource *p)
  60{
  61        if (p->child)
  62                return p->child;
  63        while (!p->sibling && p->parent)
  64                p = p->parent;
  65        return p->sibling;
  66}
  67
  68static struct resource *next_resource_skip_children(struct resource *p)
  69{
  70        while (!p->sibling && p->parent)
  71                p = p->parent;
  72        return p->sibling;
  73}
  74
  75#define for_each_resource(_root, _p, _skip_children) \
  76        for ((_p) = (_root)->child; (_p); \
  77             (_p) = (_skip_children) ? next_resource_skip_children(_p) : \
  78                                       next_resource(_p))
  79
  80static void *r_next(struct seq_file *m, void *v, loff_t *pos)
  81{
  82        struct resource *p = v;
  83        (*pos)++;
  84        return (void *)next_resource(p);
  85}
  86
  87#ifdef CONFIG_PROC_FS
  88
  89enum { MAX_IORES_LEVEL = 5 };
  90
  91static void *r_start(struct seq_file *m, loff_t *pos)
  92        __acquires(resource_lock)
  93{
  94        struct resource *p = pde_data(file_inode(m->file));
  95        loff_t l = 0;
  96        read_lock(&resource_lock);
  97        for (p = p->child; p && l < *pos; p = r_next(m, p, &l))
  98                ;
  99        return p;
 100}
 101
 102static void r_stop(struct seq_file *m, void *v)
 103        __releases(resource_lock)
 104{
 105        read_unlock(&resource_lock);
 106}
 107
 108static int r_show(struct seq_file *m, void *v)
 109{
 110        struct resource *root = pde_data(file_inode(m->file));
 111        struct resource *r = v, *p;
 112        unsigned long long start, end;
 113        int width = root->end < 0x10000 ? 4 : 8;
 114        int depth;
 115
 116        for (depth = 0, p = r; depth < MAX_IORES_LEVEL; depth++, p = p->parent)
 117                if (p->parent == root)
 118                        break;
 119
 120        if (file_ns_capable(m->file, &init_user_ns, CAP_SYS_ADMIN)) {
 121                start = r->start;
 122                end = r->end;
 123        } else {
 124                start = end = 0;
 125        }
 126
 127        seq_printf(m, "%*s%0*llx-%0*llx : %s\n",
 128                        depth * 2, "",
 129                        width, start,
 130                        width, end,
 131                        r->name ? r->name : "<BAD>");
 132        return 0;
 133}
 134
 135static const struct seq_operations resource_op = {
 136        .start  = r_start,
 137        .next   = r_next,
 138        .stop   = r_stop,
 139        .show   = r_show,
 140};
 141
 142static int __init ioresources_init(void)
 143{
 144        proc_create_seq_data("ioports", 0, NULL, &resource_op,
 145                        &ioport_resource);
 146        proc_create_seq_data("iomem", 0, NULL, &resource_op, &iomem_resource);
 147        return 0;
 148}
 149__initcall(ioresources_init);
 150
 151#endif /* CONFIG_PROC_FS */
 152
 153static void free_resource(struct resource *res)
 154{
 155        /**
 156         * If the resource was allocated using memblock early during boot
 157         * we'll leak it here: we can only return full pages back to the
 158         * buddy and trying to be smart and reusing them eventually in
 159         * alloc_resource() overcomplicates resource handling.
 160         */
 161        if (res && PageSlab(virt_to_head_page(res)))
 162                kfree(res);
 163}
 164
 165static struct resource *alloc_resource(gfp_t flags)
 166{
 167        return kzalloc(sizeof(struct resource), flags);
 168}
 169
 170/* Return the conflict entry if you can't request it */
 171static struct resource * __request_resource(struct resource *root, struct resource *new)
 172{
 173        resource_size_t start = new->start;
 174        resource_size_t end = new->end;
 175        struct resource *tmp, **p;
 176
 177        if (end < start)
 178                return root;
 179        if (start < root->start)
 180                return root;
 181        if (end > root->end)
 182                return root;
 183        p = &root->child;
 184        for (;;) {
 185                tmp = *p;
 186                if (!tmp || tmp->start > end) {
 187                        new->sibling = tmp;
 188                        *p = new;
 189                        new->parent = root;
 190                        return NULL;
 191                }
 192                p = &tmp->sibling;
 193                if (tmp->end < start)
 194                        continue;
 195                return tmp;
 196        }
 197}
 198
 199static int __release_resource(struct resource *old, bool release_child)
 200{
 201        struct resource *tmp, **p, *chd;
 202
 203        p = &old->parent->child;
 204        for (;;) {
 205                tmp = *p;
 206                if (!tmp)
 207                        break;
 208                if (tmp == old) {
 209                        if (release_child || !(tmp->child)) {
 210                                *p = tmp->sibling;
 211                        } else {
 212                                for (chd = tmp->child;; chd = chd->sibling) {
 213                                        chd->parent = tmp->parent;
 214                                        if (!(chd->sibling))
 215                                                break;
 216                                }
 217                                *p = tmp->child;
 218                                chd->sibling = tmp->sibling;
 219                        }
 220                        old->parent = NULL;
 221                        return 0;
 222                }
 223                p = &tmp->sibling;
 224        }
 225        return -EINVAL;
 226}
 227
 228static void __release_child_resources(struct resource *r)
 229{
 230        struct resource *tmp, *p;
 231        resource_size_t size;
 232
 233        p = r->child;
 234        r->child = NULL;
 235        while (p) {
 236                tmp = p;
 237                p = p->sibling;
 238
 239                tmp->parent = NULL;
 240                tmp->sibling = NULL;
 241                __release_child_resources(tmp);
 242
 243                printk(KERN_DEBUG "release child resource %pR\n", tmp);
 244                /* need to restore size, and keep flags */
 245                size = resource_size(tmp);
 246                tmp->start = 0;
 247                tmp->end = size - 1;
 248        }
 249}
 250
 251void release_child_resources(struct resource *r)
 252{
 253        write_lock(&resource_lock);
 254        __release_child_resources(r);
 255        write_unlock(&resource_lock);
 256}
 257
 258/**
 259 * request_resource_conflict - request and reserve an I/O or memory resource
 260 * @root: root resource descriptor
 261 * @new: resource descriptor desired by caller
 262 *
 263 * Returns 0 for success, conflict resource on error.
 264 */
 265struct resource *request_resource_conflict(struct resource *root, struct resource *new)
 266{
 267        struct resource *conflict;
 268
 269        write_lock(&resource_lock);
 270        conflict = __request_resource(root, new);
 271        write_unlock(&resource_lock);
 272        return conflict;
 273}
 274
 275/**
 276 * request_resource - request and reserve an I/O or memory resource
 277 * @root: root resource descriptor
 278 * @new: resource descriptor desired by caller
 279 *
 280 * Returns 0 for success, negative error code on error.
 281 */
 282int request_resource(struct resource *root, struct resource *new)
 283{
 284        struct resource *conflict;
 285
 286        conflict = request_resource_conflict(root, new);
 287        return conflict ? -EBUSY : 0;
 288}
 289
 290EXPORT_SYMBOL(request_resource);
 291
 292/**
 293 * release_resource - release a previously reserved resource
 294 * @old: resource pointer
 295 */
 296int release_resource(struct resource *old)
 297{
 298        int retval;
 299
 300        write_lock(&resource_lock);
 301        retval = __release_resource(old, true);
 302        write_unlock(&resource_lock);
 303        return retval;
 304}
 305
 306EXPORT_SYMBOL(release_resource);
 307
 308/**
 309 * find_next_iomem_res - Finds the lowest iomem resource that covers part of
 310 *                       [@start..@end].
 311 *
 312 * If a resource is found, returns 0 and @*res is overwritten with the part
 313 * of the resource that's within [@start..@end]; if none is found, returns
 314 * -ENODEV.  Returns -EINVAL for invalid parameters.
 315 *
 316 * @start:      start address of the resource searched for
 317 * @end:        end address of same resource
 318 * @flags:      flags which the resource must have
 319 * @desc:       descriptor the resource must have
 320 * @res:        return ptr, if resource found
 321 *
 322 * The caller must specify @start, @end, @flags, and @desc
 323 * (which may be IORES_DESC_NONE).
 324 */
 325static int find_next_iomem_res(resource_size_t start, resource_size_t end,
 326                               unsigned long flags, unsigned long desc,
 327                               struct resource *res)
 328{
 329        struct resource *p;
 330
 331        if (!res)
 332                return -EINVAL;
 333
 334        if (start >= end)
 335                return -EINVAL;
 336
 337        read_lock(&resource_lock);
 338
 339        for (p = iomem_resource.child; p; p = next_resource(p)) {
 340                /* If we passed the resource we are looking for, stop */
 341                if (p->start > end) {
 342                        p = NULL;
 343                        break;
 344                }
 345
 346                /* Skip until we find a range that matches what we look for */
 347                if (p->end < start)
 348                        continue;
 349
 350                if ((p->flags & flags) != flags)
 351                        continue;
 352                if ((desc != IORES_DESC_NONE) && (desc != p->desc))
 353                        continue;
 354
 355                /* Found a match, break */
 356                break;
 357        }
 358
 359        if (p) {
 360                /* copy data */
 361                *res = (struct resource) {
 362                        .start = max(start, p->start),
 363                        .end = min(end, p->end),
 364                        .flags = p->flags,
 365                        .desc = p->desc,
 366                        .parent = p->parent,
 367                };
 368        }
 369
 370        read_unlock(&resource_lock);
 371        return p ? 0 : -ENODEV;
 372}
 373
 374static int __walk_iomem_res_desc(resource_size_t start, resource_size_t end,
 375                                 unsigned long flags, unsigned long desc,
 376                                 void *arg,
 377                                 int (*func)(struct resource *, void *))
 378{
 379        struct resource res;
 380        int ret = -EINVAL;
 381
 382        while (start < end &&
 383               !find_next_iomem_res(start, end, flags, desc, &res)) {
 384                ret = (*func)(&res, arg);
 385                if (ret)
 386                        break;
 387
 388                start = res.end + 1;
 389        }
 390
 391        return ret;
 392}
 393
 394/**
 395 * walk_iomem_res_desc - Walks through iomem resources and calls func()
 396 *                       with matching resource ranges.
 397 * *
 398 * @desc: I/O resource descriptor. Use IORES_DESC_NONE to skip @desc check.
 399 * @flags: I/O resource flags
 400 * @start: start addr
 401 * @end: end addr
 402 * @arg: function argument for the callback @func
 403 * @func: callback function that is called for each qualifying resource area
 404 *
 405 * All the memory ranges which overlap start,end and also match flags and
 406 * desc are valid candidates.
 407 *
 408 * NOTE: For a new descriptor search, define a new IORES_DESC in
 409 * <linux/ioport.h> and set it in 'desc' of a target resource entry.
 410 */
 411int walk_iomem_res_desc(unsigned long desc, unsigned long flags, u64 start,
 412                u64 end, void *arg, int (*func)(struct resource *, void *))
 413{
 414        return __walk_iomem_res_desc(start, end, flags, desc, arg, func);
 415}
 416EXPORT_SYMBOL_GPL(walk_iomem_res_desc);
 417
 418/*
 419 * This function calls the @func callback against all memory ranges of type
 420 * System RAM which are marked as IORESOURCE_SYSTEM_RAM and IORESOUCE_BUSY.
 421 * Now, this function is only for System RAM, it deals with full ranges and
 422 * not PFNs. If resources are not PFN-aligned, dealing with PFNs can truncate
 423 * ranges.
 424 */
 425int walk_system_ram_res(u64 start, u64 end, void *arg,
 426                        int (*func)(struct resource *, void *))
 427{
 428        unsigned long flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
 429
 430        return __walk_iomem_res_desc(start, end, flags, IORES_DESC_NONE, arg,
 431                                     func);
 432}
 433
 434/*
 435 * This function calls the @func callback against all memory ranges, which
 436 * are ranges marked as IORESOURCE_MEM and IORESOUCE_BUSY.
 437 */
 438int walk_mem_res(u64 start, u64 end, void *arg,
 439                 int (*func)(struct resource *, void *))
 440{
 441        unsigned long flags = IORESOURCE_MEM | IORESOURCE_BUSY;
 442
 443        return __walk_iomem_res_desc(start, end, flags, IORES_DESC_NONE, arg,
 444                                     func);
 445}
 446
 447/*
 448 * This function calls the @func callback against all memory ranges of type
 449 * System RAM which are marked as IORESOURCE_SYSTEM_RAM and IORESOUCE_BUSY.
 450 * It is to be used only for System RAM.
 451 */
 452int walk_system_ram_range(unsigned long start_pfn, unsigned long nr_pages,
 453                          void *arg, int (*func)(unsigned long, unsigned long, void *))
 454{
 455        resource_size_t start, end;
 456        unsigned long flags;
 457        struct resource res;
 458        unsigned long pfn, end_pfn;
 459        int ret = -EINVAL;
 460
 461        start = (u64) start_pfn << PAGE_SHIFT;
 462        end = ((u64)(start_pfn + nr_pages) << PAGE_SHIFT) - 1;
 463        flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
 464        while (start < end &&
 465               !find_next_iomem_res(start, end, flags, IORES_DESC_NONE, &res)) {
 466                pfn = PFN_UP(res.start);
 467                end_pfn = PFN_DOWN(res.end + 1);
 468                if (end_pfn > pfn)
 469                        ret = (*func)(pfn, end_pfn - pfn, arg);
 470                if (ret)
 471                        break;
 472                start = res.end + 1;
 473        }
 474        return ret;
 475}
 476
 477static int __is_ram(unsigned long pfn, unsigned long nr_pages, void *arg)
 478{
 479        return 1;
 480}
 481
 482/*
 483 * This generic page_is_ram() returns true if specified address is
 484 * registered as System RAM in iomem_resource list.
 485 */
 486int __weak page_is_ram(unsigned long pfn)
 487{
 488        return walk_system_ram_range(pfn, 1, NULL, __is_ram) == 1;
 489}
 490EXPORT_SYMBOL_GPL(page_is_ram);
 491
 492static int __region_intersects(struct resource *parent, resource_size_t start,
 493                               size_t size, unsigned long flags,
 494                               unsigned long desc)
 495{
 496        struct resource res;
 497        int type = 0; int other = 0;
 498        struct resource *p;
 499
 500        res.start = start;
 501        res.end = start + size - 1;
 502
 503        for (p = parent->child; p ; p = p->sibling) {
 504                bool is_type = (((p->flags & flags) == flags) &&
 505                                ((desc == IORES_DESC_NONE) ||
 506                                 (desc == p->desc)));
 507
 508                if (resource_overlaps(p, &res))
 509                        is_type ? type++ : other++;
 510        }
 511
 512        if (type == 0)
 513                return REGION_DISJOINT;
 514
 515        if (other == 0)
 516                return REGION_INTERSECTS;
 517
 518        return REGION_MIXED;
 519}
 520
 521/**
 522 * region_intersects() - determine intersection of region with known resources
 523 * @start: region start address
 524 * @size: size of region
 525 * @flags: flags of resource (in iomem_resource)
 526 * @desc: descriptor of resource (in iomem_resource) or IORES_DESC_NONE
 527 *
 528 * Check if the specified region partially overlaps or fully eclipses a
 529 * resource identified by @flags and @desc (optional with IORES_DESC_NONE).
 530 * Return REGION_DISJOINT if the region does not overlap @flags/@desc,
 531 * return REGION_MIXED if the region overlaps @flags/@desc and another
 532 * resource, and return REGION_INTERSECTS if the region overlaps @flags/@desc
 533 * and no other defined resource. Note that REGION_INTERSECTS is also
 534 * returned in the case when the specified region overlaps RAM and undefined
 535 * memory holes.
 536 *
 537 * region_intersect() is used by memory remapping functions to ensure
 538 * the user is not remapping RAM and is a vast speed up over walking
 539 * through the resource table page by page.
 540 */
 541int region_intersects(resource_size_t start, size_t size, unsigned long flags,
 542                      unsigned long desc)
 543{
 544        int ret;
 545
 546        read_lock(&resource_lock);
 547        ret = __region_intersects(&iomem_resource, start, size, flags, desc);
 548        read_unlock(&resource_lock);
 549
 550        return ret;
 551}
 552EXPORT_SYMBOL_GPL(region_intersects);
 553
 554void __weak arch_remove_reservations(struct resource *avail)
 555{
 556}
 557
 558static resource_size_t simple_align_resource(void *data,
 559                                             const struct resource *avail,
 560                                             resource_size_t size,
 561                                             resource_size_t align)
 562{
 563        return avail->start;
 564}
 565
 566static void resource_clip(struct resource *res, resource_size_t min,
 567                          resource_size_t max)
 568{
 569        if (res->start < min)
 570                res->start = min;
 571        if (res->end > max)
 572                res->end = max;
 573}
 574
 575/*
 576 * Find empty slot in the resource tree with the given range and
 577 * alignment constraints
 578 */
 579static int __find_resource(struct resource *root, struct resource *old,
 580                         struct resource *new,
 581                         resource_size_t  size,
 582                         struct resource_constraint *constraint)
 583{
 584        struct resource *this = root->child;
 585        struct resource tmp = *new, avail, alloc;
 586
 587        tmp.start = root->start;
 588        /*
 589         * Skip past an allocated resource that starts at 0, since the assignment
 590         * of this->start - 1 to tmp->end below would cause an underflow.
 591         */
 592        if (this && this->start == root->start) {
 593                tmp.start = (this == old) ? old->start : this->end + 1;
 594                this = this->sibling;
 595        }
 596        for(;;) {
 597                if (this)
 598                        tmp.end = (this == old) ?  this->end : this->start - 1;
 599                else
 600                        tmp.end = root->end;
 601
 602                if (tmp.end < tmp.start)
 603                        goto next;
 604
 605                resource_clip(&tmp, constraint->min, constraint->max);
 606                arch_remove_reservations(&tmp);
 607
 608                /* Check for overflow after ALIGN() */
 609                avail.start = ALIGN(tmp.start, constraint->align);
 610                avail.end = tmp.end;
 611                avail.flags = new->flags & ~IORESOURCE_UNSET;
 612                if (avail.start >= tmp.start) {
 613                        alloc.flags = avail.flags;
 614                        alloc.start = constraint->alignf(constraint->alignf_data, &avail,
 615                                        size, constraint->align);
 616                        alloc.end = alloc.start + size - 1;
 617                        if (alloc.start <= alloc.end &&
 618                            resource_contains(&avail, &alloc)) {
 619                                new->start = alloc.start;
 620                                new->end = alloc.end;
 621                                return 0;
 622                        }
 623                }
 624
 625next:           if (!this || this->end == root->end)
 626                        break;
 627
 628                if (this != old)
 629                        tmp.start = this->end + 1;
 630                this = this->sibling;
 631        }
 632        return -EBUSY;
 633}
 634
 635/*
 636 * Find empty slot in the resource tree given range and alignment.
 637 */
 638static int find_resource(struct resource *root, struct resource *new,
 639                        resource_size_t size,
 640                        struct resource_constraint  *constraint)
 641{
 642        return  __find_resource(root, NULL, new, size, constraint);
 643}
 644
 645/**
 646 * reallocate_resource - allocate a slot in the resource tree given range & alignment.
 647 *      The resource will be relocated if the new size cannot be reallocated in the
 648 *      current location.
 649 *
 650 * @root: root resource descriptor
 651 * @old:  resource descriptor desired by caller
 652 * @newsize: new size of the resource descriptor
 653 * @constraint: the size and alignment constraints to be met.
 654 */
 655static int reallocate_resource(struct resource *root, struct resource *old,
 656                               resource_size_t newsize,
 657                               struct resource_constraint *constraint)
 658{
 659        int err=0;
 660        struct resource new = *old;
 661        struct resource *conflict;
 662
 663        write_lock(&resource_lock);
 664
 665        if ((err = __find_resource(root, old, &new, newsize, constraint)))
 666                goto out;
 667
 668        if (resource_contains(&new, old)) {
 669                old->start = new.start;
 670                old->end = new.end;
 671                goto out;
 672        }
 673
 674        if (old->child) {
 675                err = -EBUSY;
 676                goto out;
 677        }
 678
 679        if (resource_contains(old, &new)) {
 680                old->start = new.start;
 681                old->end = new.end;
 682        } else {
 683                __release_resource(old, true);
 684                *old = new;
 685                conflict = __request_resource(root, old);
 686                BUG_ON(conflict);
 687        }
 688out:
 689        write_unlock(&resource_lock);
 690        return err;
 691}
 692
 693
 694/**
 695 * allocate_resource - allocate empty slot in the resource tree given range & alignment.
 696 *      The resource will be reallocated with a new size if it was already allocated
 697 * @root: root resource descriptor
 698 * @new: resource descriptor desired by caller
 699 * @size: requested resource region size
 700 * @min: minimum boundary to allocate
 701 * @max: maximum boundary to allocate
 702 * @align: alignment requested, in bytes
 703 * @alignf: alignment function, optional, called if not NULL
 704 * @alignf_data: arbitrary data to pass to the @alignf function
 705 */
 706int allocate_resource(struct resource *root, struct resource *new,
 707                      resource_size_t size, resource_size_t min,
 708                      resource_size_t max, resource_size_t align,
 709                      resource_size_t (*alignf)(void *,
 710                                                const struct resource *,
 711                                                resource_size_t,
 712                                                resource_size_t),
 713                      void *alignf_data)
 714{
 715        int err;
 716        struct resource_constraint constraint;
 717
 718        if (!alignf)
 719                alignf = simple_align_resource;
 720
 721        constraint.min = min;
 722        constraint.max = max;
 723        constraint.align = align;
 724        constraint.alignf = alignf;
 725        constraint.alignf_data = alignf_data;
 726
 727        if ( new->parent ) {
 728                /* resource is already allocated, try reallocating with
 729                   the new constraints */
 730                return reallocate_resource(root, new, size, &constraint);
 731        }
 732
 733        write_lock(&resource_lock);
 734        err = find_resource(root, new, size, &constraint);
 735        if (err >= 0 && __request_resource(root, new))
 736                err = -EBUSY;
 737        write_unlock(&resource_lock);
 738        return err;
 739}
 740
 741EXPORT_SYMBOL(allocate_resource);
 742
 743/**
 744 * lookup_resource - find an existing resource by a resource start address
 745 * @root: root resource descriptor
 746 * @start: resource start address
 747 *
 748 * Returns a pointer to the resource if found, NULL otherwise
 749 */
 750struct resource *lookup_resource(struct resource *root, resource_size_t start)
 751{
 752        struct resource *res;
 753
 754        read_lock(&resource_lock);
 755        for (res = root->child; res; res = res->sibling) {
 756                if (res->start == start)
 757                        break;
 758        }
 759        read_unlock(&resource_lock);
 760
 761        return res;
 762}
 763
 764/*
 765 * Insert a resource into the resource tree. If successful, return NULL,
 766 * otherwise return the conflicting resource (compare to __request_resource())
 767 */
 768static struct resource * __insert_resource(struct resource *parent, struct resource *new)
 769{
 770        struct resource *first, *next;
 771
 772        for (;; parent = first) {
 773                first = __request_resource(parent, new);
 774                if (!first)
 775                        return first;
 776
 777                if (first == parent)
 778                        return first;
 779                if (WARN_ON(first == new))      /* duplicated insertion */
 780                        return first;
 781
 782                if ((first->start > new->start) || (first->end < new->end))
 783                        break;
 784                if ((first->start == new->start) && (first->end == new->end))
 785                        break;
 786        }
 787
 788        for (next = first; ; next = next->sibling) {
 789                /* Partial overlap? Bad, and unfixable */
 790                if (next->start < new->start || next->end > new->end)
 791                        return next;
 792                if (!next->sibling)
 793                        break;
 794                if (next->sibling->start > new->end)
 795                        break;
 796        }
 797
 798        new->parent = parent;
 799        new->sibling = next->sibling;
 800        new->child = first;
 801
 802        next->sibling = NULL;
 803        for (next = first; next; next = next->sibling)
 804                next->parent = new;
 805
 806        if (parent->child == first) {
 807                parent->child = new;
 808        } else {
 809                next = parent->child;
 810                while (next->sibling != first)
 811                        next = next->sibling;
 812                next->sibling = new;
 813        }
 814        return NULL;
 815}
 816
 817/**
 818 * insert_resource_conflict - Inserts resource in the resource tree
 819 * @parent: parent of the new resource
 820 * @new: new resource to insert
 821 *
 822 * Returns 0 on success, conflict resource if the resource can't be inserted.
 823 *
 824 * This function is equivalent to request_resource_conflict when no conflict
 825 * happens. If a conflict happens, and the conflicting resources
 826 * entirely fit within the range of the new resource, then the new
 827 * resource is inserted and the conflicting resources become children of
 828 * the new resource.
 829 *
 830 * This function is intended for producers of resources, such as FW modules
 831 * and bus drivers.
 832 */
 833struct resource *insert_resource_conflict(struct resource *parent, struct resource *new)
 834{
 835        struct resource *conflict;
 836
 837        write_lock(&resource_lock);
 838        conflict = __insert_resource(parent, new);
 839        write_unlock(&resource_lock);
 840        return conflict;
 841}
 842
 843/**
 844 * insert_resource - Inserts a resource in the resource tree
 845 * @parent: parent of the new resource
 846 * @new: new resource to insert
 847 *
 848 * Returns 0 on success, -EBUSY if the resource can't be inserted.
 849 *
 850 * This function is intended for producers of resources, such as FW modules
 851 * and bus drivers.
 852 */
 853int insert_resource(struct resource *parent, struct resource *new)
 854{
 855        struct resource *conflict;
 856
 857        conflict = insert_resource_conflict(parent, new);
 858        return conflict ? -EBUSY : 0;
 859}
 860EXPORT_SYMBOL_GPL(insert_resource);
 861
 862/**
 863 * insert_resource_expand_to_fit - Insert a resource into the resource tree
 864 * @root: root resource descriptor
 865 * @new: new resource to insert
 866 *
 867 * Insert a resource into the resource tree, possibly expanding it in order
 868 * to make it encompass any conflicting resources.
 869 */
 870void insert_resource_expand_to_fit(struct resource *root, struct resource *new)
 871{
 872        if (new->parent)
 873                return;
 874
 875        write_lock(&resource_lock);
 876        for (;;) {
 877                struct resource *conflict;
 878
 879                conflict = __insert_resource(root, new);
 880                if (!conflict)
 881                        break;
 882                if (conflict == root)
 883                        break;
 884
 885                /* Ok, expand resource to cover the conflict, then try again .. */
 886                if (conflict->start < new->start)
 887                        new->start = conflict->start;
 888                if (conflict->end > new->end)
 889                        new->end = conflict->end;
 890
 891                printk("Expanded resource %s due to conflict with %s\n", new->name, conflict->name);
 892        }
 893        write_unlock(&resource_lock);
 894}
 895/*
 896 * Not for general consumption, only early boot memory map parsing, PCI
 897 * resource discovery, and late discovery of CXL resources are expected
 898 * to use this interface. The former are built-in and only the latter,
 899 * CXL, is a module.
 900 */
 901EXPORT_SYMBOL_NS_GPL(insert_resource_expand_to_fit, CXL);
 902
 903/**
 904 * remove_resource - Remove a resource in the resource tree
 905 * @old: resource to remove
 906 *
 907 * Returns 0 on success, -EINVAL if the resource is not valid.
 908 *
 909 * This function removes a resource previously inserted by insert_resource()
 910 * or insert_resource_conflict(), and moves the children (if any) up to
 911 * where they were before.  insert_resource() and insert_resource_conflict()
 912 * insert a new resource, and move any conflicting resources down to the
 913 * children of the new resource.
 914 *
 915 * insert_resource(), insert_resource_conflict() and remove_resource() are
 916 * intended for producers of resources, such as FW modules and bus drivers.
 917 */
 918int remove_resource(struct resource *old)
 919{
 920        int retval;
 921
 922        write_lock(&resource_lock);
 923        retval = __release_resource(old, false);
 924        write_unlock(&resource_lock);
 925        return retval;
 926}
 927EXPORT_SYMBOL_GPL(remove_resource);
 928
 929static int __adjust_resource(struct resource *res, resource_size_t start,
 930                                resource_size_t size)
 931{
 932        struct resource *tmp, *parent = res->parent;
 933        resource_size_t end = start + size - 1;
 934        int result = -EBUSY;
 935
 936        if (!parent)
 937                goto skip;
 938
 939        if ((start < parent->start) || (end > parent->end))
 940                goto out;
 941
 942        if (res->sibling && (res->sibling->start <= end))
 943                goto out;
 944
 945        tmp = parent->child;
 946        if (tmp != res) {
 947                while (tmp->sibling != res)
 948                        tmp = tmp->sibling;
 949                if (start <= tmp->end)
 950                        goto out;
 951        }
 952
 953skip:
 954        for (tmp = res->child; tmp; tmp = tmp->sibling)
 955                if ((tmp->start < start) || (tmp->end > end))
 956                        goto out;
 957
 958        res->start = start;
 959        res->end = end;
 960        result = 0;
 961
 962 out:
 963        return result;
 964}
 965
 966/**
 967 * adjust_resource - modify a resource's start and size
 968 * @res: resource to modify
 969 * @start: new start value
 970 * @size: new size
 971 *
 972 * Given an existing resource, change its start and size to match the
 973 * arguments.  Returns 0 on success, -EBUSY if it can't fit.
 974 * Existing children of the resource are assumed to be immutable.
 975 */
 976int adjust_resource(struct resource *res, resource_size_t start,
 977                    resource_size_t size)
 978{
 979        int result;
 980
 981        write_lock(&resource_lock);
 982        result = __adjust_resource(res, start, size);
 983        write_unlock(&resource_lock);
 984        return result;
 985}
 986EXPORT_SYMBOL(adjust_resource);
 987
 988static void __init
 989__reserve_region_with_split(struct resource *root, resource_size_t start,
 990                            resource_size_t end, const char *name)
 991{
 992        struct resource *parent = root;
 993        struct resource *conflict;
 994        struct resource *res = alloc_resource(GFP_ATOMIC);
 995        struct resource *next_res = NULL;
 996        int type = resource_type(root);
 997
 998        if (!res)
 999                return;
1000
1001        res->name = name;
1002        res->start = start;
1003        res->end = end;
1004        res->flags = type | IORESOURCE_BUSY;
1005        res->desc = IORES_DESC_NONE;
1006
1007        while (1) {
1008
1009                conflict = __request_resource(parent, res);
1010                if (!conflict) {
1011                        if (!next_res)
1012                                break;
1013                        res = next_res;
1014                        next_res = NULL;
1015                        continue;
1016                }
1017
1018                /* conflict covered whole area */
1019                if (conflict->start <= res->start &&
1020                                conflict->end >= res->end) {
1021                        free_resource(res);
1022                        WARN_ON(next_res);
1023                        break;
1024                }
1025
1026                /* failed, split and try again */
1027                if (conflict->start > res->start) {
1028                        end = res->end;
1029                        res->end = conflict->start - 1;
1030                        if (conflict->end < end) {
1031                                next_res = alloc_resource(GFP_ATOMIC);
1032                                if (!next_res) {
1033                                        free_resource(res);
1034                                        break;
1035                                }
1036                                next_res->name = name;
1037                                next_res->start = conflict->end + 1;
1038                                next_res->end = end;
1039                                next_res->flags = type | IORESOURCE_BUSY;
1040                                next_res->desc = IORES_DESC_NONE;
1041                        }
1042                } else {
1043                        res->start = conflict->end + 1;
1044                }
1045        }
1046
1047}
1048
1049void __init
1050reserve_region_with_split(struct resource *root, resource_size_t start,
1051                          resource_size_t end, const char *name)
1052{
1053        int abort = 0;
1054
1055        write_lock(&resource_lock);
1056        if (root->start > start || root->end < end) {
1057                pr_err("requested range [0x%llx-0x%llx] not in root %pr\n",
1058                       (unsigned long long)start, (unsigned long long)end,
1059                       root);
1060                if (start > root->end || end < root->start)
1061                        abort = 1;
1062                else {
1063                        if (end > root->end)
1064                                end = root->end;
1065                        if (start < root->start)
1066                                start = root->start;
1067                        pr_err("fixing request to [0x%llx-0x%llx]\n",
1068                               (unsigned long long)start,
1069                               (unsigned long long)end);
1070                }
1071                dump_stack();
1072        }
1073        if (!abort)
1074                __reserve_region_with_split(root, start, end, name);
1075        write_unlock(&resource_lock);
1076}
1077
1078/**
1079 * resource_alignment - calculate resource's alignment
1080 * @res: resource pointer
1081 *
1082 * Returns alignment on success, 0 (invalid alignment) on failure.
1083 */
1084resource_size_t resource_alignment(struct resource *res)
1085{
1086        switch (res->flags & (IORESOURCE_SIZEALIGN | IORESOURCE_STARTALIGN)) {
1087        case IORESOURCE_SIZEALIGN:
1088                return resource_size(res);
1089        case IORESOURCE_STARTALIGN:
1090                return res->start;
1091        default:
1092                return 0;
1093        }
1094}
1095
1096/*
1097 * This is compatibility stuff for IO resources.
1098 *
1099 * Note how this, unlike the above, knows about
1100 * the IO flag meanings (busy etc).
1101 *
1102 * request_region creates a new busy region.
1103 *
1104 * release_region releases a matching busy region.
1105 */
1106
1107static DECLARE_WAIT_QUEUE_HEAD(muxed_resource_wait);
1108
1109static struct inode *iomem_inode;
1110
1111#ifdef CONFIG_IO_STRICT_DEVMEM
1112static void revoke_iomem(struct resource *res)
1113{
1114        /* pairs with smp_store_release() in iomem_init_inode() */
1115        struct inode *inode = smp_load_acquire(&iomem_inode);
1116
1117        /*
1118         * Check that the initialization has completed. Losing the race
1119         * is ok because it means drivers are claiming resources before
1120         * the fs_initcall level of init and prevent iomem_get_mapping users
1121         * from establishing mappings.
1122         */
1123        if (!inode)
1124                return;
1125
1126        /*
1127         * The expectation is that the driver has successfully marked
1128         * the resource busy by this point, so devmem_is_allowed()
1129         * should start returning false, however for performance this
1130         * does not iterate the entire resource range.
1131         */
1132        if (devmem_is_allowed(PHYS_PFN(res->start)) &&
1133            devmem_is_allowed(PHYS_PFN(res->end))) {
1134                /*
1135                 * *cringe* iomem=relaxed says "go ahead, what's the
1136                 * worst that can happen?"
1137                 */
1138                return;
1139        }
1140
1141        unmap_mapping_range(inode->i_mapping, res->start, resource_size(res), 1);
1142}
1143#else
1144static void revoke_iomem(struct resource *res) {}
1145#endif
1146
1147struct address_space *iomem_get_mapping(void)
1148{
1149        /*
1150         * This function is only called from file open paths, hence guaranteed
1151         * that fs_initcalls have completed and no need to check for NULL. But
1152         * since revoke_iomem can be called before the initcall we still need
1153         * the barrier to appease checkers.
1154         */
1155        return smp_load_acquire(&iomem_inode)->i_mapping;
1156}
1157
1158static int __request_region_locked(struct resource *res, struct resource *parent,
1159                                   resource_size_t start, resource_size_t n,
1160                                   const char *name, int flags)
1161{
1162        DECLARE_WAITQUEUE(wait, current);
1163
1164        res->name = name;
1165        res->start = start;
1166        res->end = start + n - 1;
1167
1168        for (;;) {
1169                struct resource *conflict;
1170
1171                res->flags = resource_type(parent) | resource_ext_type(parent);
1172                res->flags |= IORESOURCE_BUSY | flags;
1173                res->desc = parent->desc;
1174
1175                conflict = __request_resource(parent, res);
1176                if (!conflict)
1177                        break;
1178                /*
1179                 * mm/hmm.c reserves physical addresses which then
1180                 * become unavailable to other users.  Conflicts are
1181                 * not expected.  Warn to aid debugging if encountered.
1182                 */
1183                if (conflict->desc == IORES_DESC_DEVICE_PRIVATE_MEMORY) {
1184                        pr_warn("Unaddressable device %s %pR conflicts with %pR",
1185                                conflict->name, conflict, res);
1186                }
1187                if (conflict != parent) {
1188                        if (!(conflict->flags & IORESOURCE_BUSY)) {
1189                                parent = conflict;
1190                                continue;
1191                        }
1192                }
1193                if (conflict->flags & flags & IORESOURCE_MUXED) {
1194                        add_wait_queue(&muxed_resource_wait, &wait);
1195                        write_unlock(&resource_lock);
1196                        set_current_state(TASK_UNINTERRUPTIBLE);
1197                        schedule();
1198                        remove_wait_queue(&muxed_resource_wait, &wait);
1199                        write_lock(&resource_lock);
1200                        continue;
1201                }
1202                /* Uhhuh, that didn't work out.. */
1203                return -EBUSY;
1204        }
1205
1206        return 0;
1207}
1208
1209/**
1210 * __request_region - create a new busy resource region
1211 * @parent: parent resource descriptor
1212 * @start: resource start address
1213 * @n: resource region size
1214 * @name: reserving caller's ID string
1215 * @flags: IO resource flags
1216 */
1217struct resource *__request_region(struct resource *parent,
1218                                  resource_size_t start, resource_size_t n,
1219                                  const char *name, int flags)
1220{
1221        struct resource *res = alloc_resource(GFP_KERNEL);
1222        int ret;
1223
1224        if (!res)
1225                return NULL;
1226
1227        write_lock(&resource_lock);
1228        ret = __request_region_locked(res, parent, start, n, name, flags);
1229        write_unlock(&resource_lock);
1230
1231        if (ret) {
1232                free_resource(res);
1233                return NULL;
1234        }
1235
1236        if (parent == &iomem_resource)
1237                revoke_iomem(res);
1238
1239        return res;
1240}
1241EXPORT_SYMBOL(__request_region);
1242
1243/**
1244 * __release_region - release a previously reserved resource region
1245 * @parent: parent resource descriptor
1246 * @start: resource start address
1247 * @n: resource region size
1248 *
1249 * The described resource region must match a currently busy region.
1250 */
1251void __release_region(struct resource *parent, resource_size_t start,
1252                      resource_size_t n)
1253{
1254        struct resource **p;
1255        resource_size_t end;
1256
1257        p = &parent->child;
1258        end = start + n - 1;
1259
1260        write_lock(&resource_lock);
1261
1262        for (;;) {
1263                struct resource *res = *p;
1264
1265                if (!res)
1266                        break;
1267                if (res->start <= start && res->end >= end) {
1268                        if (!(res->flags & IORESOURCE_BUSY)) {
1269                                p = &res->child;
1270                                continue;
1271                        }
1272                        if (res->start != start || res->end != end)
1273                                break;
1274                        *p = res->sibling;
1275                        write_unlock(&resource_lock);
1276                        if (res->flags & IORESOURCE_MUXED)
1277                                wake_up(&muxed_resource_wait);
1278                        free_resource(res);
1279                        return;
1280                }
1281                p = &res->sibling;
1282        }
1283
1284        write_unlock(&resource_lock);
1285
1286        printk(KERN_WARNING "Trying to free nonexistent resource "
1287                "<%016llx-%016llx>\n", (unsigned long long)start,
1288                (unsigned long long)end);
1289}
1290EXPORT_SYMBOL(__release_region);
1291
1292#ifdef CONFIG_MEMORY_HOTREMOVE
1293/**
1294 * release_mem_region_adjustable - release a previously reserved memory region
1295 * @start: resource start address
1296 * @size: resource region size
1297 *
1298 * This interface is intended for memory hot-delete.  The requested region
1299 * is released from a currently busy memory resource.  The requested region
1300 * must either match exactly or fit into a single busy resource entry.  In
1301 * the latter case, the remaining resource is adjusted accordingly.
1302 * Existing children of the busy memory resource must be immutable in the
1303 * request.
1304 *
1305 * Note:
1306 * - Additional release conditions, such as overlapping region, can be
1307 *   supported after they are confirmed as valid cases.
1308 * - When a busy memory resource gets split into two entries, the code
1309 *   assumes that all children remain in the lower address entry for
1310 *   simplicity.  Enhance this logic when necessary.
1311 */
1312void release_mem_region_adjustable(resource_size_t start, resource_size_t size)
1313{
1314        struct resource *parent = &iomem_resource;
1315        struct resource *new_res = NULL;
1316        bool alloc_nofail = false;
1317        struct resource **p;
1318        struct resource *res;
1319        resource_size_t end;
1320
1321        end = start + size - 1;
1322        if (WARN_ON_ONCE((start < parent->start) || (end > parent->end)))
1323                return;
1324
1325        /*
1326         * We free up quite a lot of memory on memory hotunplug (esp., memap),
1327         * just before releasing the region. This is highly unlikely to
1328         * fail - let's play save and make it never fail as the caller cannot
1329         * perform any error handling (e.g., trying to re-add memory will fail
1330         * similarly).
1331         */
1332retry:
1333        new_res = alloc_resource(GFP_KERNEL | (alloc_nofail ? __GFP_NOFAIL : 0));
1334
1335        p = &parent->child;
1336        write_lock(&resource_lock);
1337
1338        while ((res = *p)) {
1339                if (res->start >= end)
1340                        break;
1341
1342                /* look for the next resource if it does not fit into */
1343                if (res->start > start || res->end < end) {
1344                        p = &res->sibling;
1345                        continue;
1346                }
1347
1348                /*
1349                 * All memory regions added from memory-hotplug path have the
1350                 * flag IORESOURCE_SYSTEM_RAM. If the resource does not have
1351                 * this flag, we know that we are dealing with a resource coming
1352                 * from HMM/devm. HMM/devm use another mechanism to add/release
1353                 * a resource. This goes via devm_request_mem_region and
1354                 * devm_release_mem_region.
1355                 * HMM/devm take care to release their resources when they want,
1356                 * so if we are dealing with them, let us just back off here.
1357                 */
1358                if (!(res->flags & IORESOURCE_SYSRAM)) {
1359                        break;
1360                }
1361
1362                if (!(res->flags & IORESOURCE_MEM))
1363                        break;
1364
1365                if (!(res->flags & IORESOURCE_BUSY)) {
1366                        p = &res->child;
1367                        continue;
1368                }
1369
1370                /* found the target resource; let's adjust accordingly */
1371                if (res->start == start && res->end == end) {
1372                        /* free the whole entry */
1373                        *p = res->sibling;
1374                        free_resource(res);
1375                } else if (res->start == start && res->end != end) {
1376                        /* adjust the start */
1377                        WARN_ON_ONCE(__adjust_resource(res, end + 1,
1378                                                       res->end - end));
1379                } else if (res->start != start && res->end == end) {
1380                        /* adjust the end */
1381                        WARN_ON_ONCE(__adjust_resource(res, res->start,
1382                                                       start - res->start));
1383                } else {
1384                        /* split into two entries - we need a new resource */
1385                        if (!new_res) {
1386                                new_res = alloc_resource(GFP_ATOMIC);
1387                                if (!new_res) {
1388                                        alloc_nofail = true;
1389                                        write_unlock(&resource_lock);
1390                                        goto retry;
1391                                }
1392                        }
1393                        new_res->name = res->name;
1394                        new_res->start = end + 1;
1395                        new_res->end = res->end;
1396                        new_res->flags = res->flags;
1397                        new_res->desc = res->desc;
1398                        new_res->parent = res->parent;
1399                        new_res->sibling = res->sibling;
1400                        new_res->child = NULL;
1401
1402                        if (WARN_ON_ONCE(__adjust_resource(res, res->start,
1403                                                           start - res->start)))
1404                                break;
1405                        res->sibling = new_res;
1406                        new_res = NULL;
1407                }
1408
1409                break;
1410        }
1411
1412        write_unlock(&resource_lock);
1413        free_resource(new_res);
1414}
1415#endif  /* CONFIG_MEMORY_HOTREMOVE */
1416
1417#ifdef CONFIG_MEMORY_HOTPLUG
1418static bool system_ram_resources_mergeable(struct resource *r1,
1419                                           struct resource *r2)
1420{
1421        /* We assume either r1 or r2 is IORESOURCE_SYSRAM_MERGEABLE. */
1422        return r1->flags == r2->flags && r1->end + 1 == r2->start &&
1423               r1->name == r2->name && r1->desc == r2->desc &&
1424               !r1->child && !r2->child;
1425}
1426
1427/**
1428 * merge_system_ram_resource - mark the System RAM resource mergeable and try to
1429 *      merge it with adjacent, mergeable resources
1430 * @res: resource descriptor
1431 *
1432 * This interface is intended for memory hotplug, whereby lots of contiguous
1433 * system ram resources are added (e.g., via add_memory*()) by a driver, and
1434 * the actual resource boundaries are not of interest (e.g., it might be
1435 * relevant for DIMMs). Only resources that are marked mergeable, that have the
1436 * same parent, and that don't have any children are considered. All mergeable
1437 * resources must be immutable during the request.
1438 *
1439 * Note:
1440 * - The caller has to make sure that no pointers to resources that are
1441 *   marked mergeable are used anymore after this call - the resource might
1442 *   be freed and the pointer might be stale!
1443 * - release_mem_region_adjustable() will split on demand on memory hotunplug
1444 */
1445void merge_system_ram_resource(struct resource *res)
1446{
1447        const unsigned long flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
1448        struct resource *cur;
1449
1450        if (WARN_ON_ONCE((res->flags & flags) != flags))
1451                return;
1452
1453        write_lock(&resource_lock);
1454        res->flags |= IORESOURCE_SYSRAM_MERGEABLE;
1455
1456        /* Try to merge with next item in the list. */
1457        cur = res->sibling;
1458        if (cur && system_ram_resources_mergeable(res, cur)) {
1459                res->end = cur->end;
1460                res->sibling = cur->sibling;
1461                free_resource(cur);
1462        }
1463
1464        /* Try to merge with previous item in the list. */
1465        cur = res->parent->child;
1466        while (cur && cur->sibling != res)
1467                cur = cur->sibling;
1468        if (cur && system_ram_resources_mergeable(cur, res)) {
1469                cur->end = res->end;
1470                cur->sibling = res->sibling;
1471                free_resource(res);
1472        }
1473        write_unlock(&resource_lock);
1474}
1475#endif  /* CONFIG_MEMORY_HOTPLUG */
1476
1477/*
1478 * Managed region resource
1479 */
1480static void devm_resource_release(struct device *dev, void *ptr)
1481{
1482        struct resource **r = ptr;
1483
1484        release_resource(*r);
1485}
1486
1487/**
1488 * devm_request_resource() - request and reserve an I/O or memory resource
1489 * @dev: device for which to request the resource
1490 * @root: root of the resource tree from which to request the resource
1491 * @new: descriptor of the resource to request
1492 *
1493 * This is a device-managed version of request_resource(). There is usually
1494 * no need to release resources requested by this function explicitly since
1495 * that will be taken care of when the device is unbound from its driver.
1496 * If for some reason the resource needs to be released explicitly, because
1497 * of ordering issues for example, drivers must call devm_release_resource()
1498 * rather than the regular release_resource().
1499 *
1500 * When a conflict is detected between any existing resources and the newly
1501 * requested resource, an error message will be printed.
1502 *
1503 * Returns 0 on success or a negative error code on failure.
1504 */
1505int devm_request_resource(struct device *dev, struct resource *root,
1506                          struct resource *new)
1507{
1508        struct resource *conflict, **ptr;
1509
1510        ptr = devres_alloc(devm_resource_release, sizeof(*ptr), GFP_KERNEL);
1511        if (!ptr)
1512                return -ENOMEM;
1513
1514        *ptr = new;
1515
1516        conflict = request_resource_conflict(root, new);
1517        if (conflict) {
1518                dev_err(dev, "resource collision: %pR conflicts with %s %pR\n",
1519                        new, conflict->name, conflict);
1520                devres_free(ptr);
1521                return -EBUSY;
1522        }
1523
1524        devres_add(dev, ptr);
1525        return 0;
1526}
1527EXPORT_SYMBOL(devm_request_resource);
1528
1529static int devm_resource_match(struct device *dev, void *res, void *data)
1530{
1531        struct resource **ptr = res;
1532
1533        return *ptr == data;
1534}
1535
1536/**
1537 * devm_release_resource() - release a previously requested resource
1538 * @dev: device for which to release the resource
1539 * @new: descriptor of the resource to release
1540 *
1541 * Releases a resource previously requested using devm_request_resource().
1542 */
1543void devm_release_resource(struct device *dev, struct resource *new)
1544{
1545        WARN_ON(devres_release(dev, devm_resource_release, devm_resource_match,
1546                               new));
1547}
1548EXPORT_SYMBOL(devm_release_resource);
1549
1550struct region_devres {
1551        struct resource *parent;
1552        resource_size_t start;
1553        resource_size_t n;
1554};
1555
1556static void devm_region_release(struct device *dev, void *res)
1557{
1558        struct region_devres *this = res;
1559
1560        __release_region(this->parent, this->start, this->n);
1561}
1562
1563static int devm_region_match(struct device *dev, void *res, void *match_data)
1564{
1565        struct region_devres *this = res, *match = match_data;
1566
1567        return this->parent == match->parent &&
1568                this->start == match->start && this->n == match->n;
1569}
1570
1571struct resource *
1572__devm_request_region(struct device *dev, struct resource *parent,
1573                      resource_size_t start, resource_size_t n, const char *name)
1574{
1575        struct region_devres *dr = NULL;
1576        struct resource *res;
1577
1578        dr = devres_alloc(devm_region_release, sizeof(struct region_devres),
1579                          GFP_KERNEL);
1580        if (!dr)
1581                return NULL;
1582
1583        dr->parent = parent;
1584        dr->start = start;
1585        dr->n = n;
1586
1587        res = __request_region(parent, start, n, name, 0);
1588        if (res)
1589                devres_add(dev, dr);
1590        else
1591                devres_free(dr);
1592
1593        return res;
1594}
1595EXPORT_SYMBOL(__devm_request_region);
1596
1597void __devm_release_region(struct device *dev, struct resource *parent,
1598                           resource_size_t start, resource_size_t n)
1599{
1600        struct region_devres match_data = { parent, start, n };
1601
1602        __release_region(parent, start, n);
1603        WARN_ON(devres_destroy(dev, devm_region_release, devm_region_match,
1604                               &match_data));
1605}
1606EXPORT_SYMBOL(__devm_release_region);
1607
1608/*
1609 * Reserve I/O ports or memory based on "reserve=" kernel parameter.
1610 */
1611#define MAXRESERVE 4
1612static int __init reserve_setup(char *str)
1613{
1614        static int reserved;
1615        static struct resource reserve[MAXRESERVE];
1616
1617        for (;;) {
1618                unsigned int io_start, io_num;
1619                int x = reserved;
1620                struct resource *parent;
1621
1622                if (get_option(&str, &io_start) != 2)
1623                        break;
1624                if (get_option(&str, &io_num) == 0)
1625                        break;
1626                if (x < MAXRESERVE) {
1627                        struct resource *res = reserve + x;
1628
1629                        /*
1630                         * If the region starts below 0x10000, we assume it's
1631                         * I/O port space; otherwise assume it's memory.
1632                         */
1633                        if (io_start < 0x10000) {
1634                                res->flags = IORESOURCE_IO;
1635                                parent = &ioport_resource;
1636                        } else {
1637                                res->flags = IORESOURCE_MEM;
1638                                parent = &iomem_resource;
1639                        }
1640                        res->name = "reserved";
1641                        res->start = io_start;
1642                        res->end = io_start + io_num - 1;
1643                        res->flags |= IORESOURCE_BUSY;
1644                        res->desc = IORES_DESC_NONE;
1645                        res->child = NULL;
1646                        if (request_resource(parent, res) == 0)
1647                                reserved = x+1;
1648                }
1649        }
1650        return 1;
1651}
1652__setup("reserve=", reserve_setup);
1653
1654/*
1655 * Check if the requested addr and size spans more than any slot in the
1656 * iomem resource tree.
1657 */
1658int iomem_map_sanity_check(resource_size_t addr, unsigned long size)
1659{
1660        struct resource *p = &iomem_resource;
1661        int err = 0;
1662        loff_t l;
1663
1664        read_lock(&resource_lock);
1665        for (p = p->child; p ; p = r_next(NULL, p, &l)) {
1666                /*
1667                 * We can probably skip the resources without
1668                 * IORESOURCE_IO attribute?
1669                 */
1670                if (p->start >= addr + size)
1671                        continue;
1672                if (p->end < addr)
1673                        continue;
1674                if (PFN_DOWN(p->start) <= PFN_DOWN(addr) &&
1675                    PFN_DOWN(p->end) >= PFN_DOWN(addr + size - 1))
1676                        continue;
1677                /*
1678                 * if a resource is "BUSY", it's not a hardware resource
1679                 * but a driver mapping of such a resource; we don't want
1680                 * to warn for those; some drivers legitimately map only
1681                 * partial hardware resources. (example: vesafb)
1682                 */
1683                if (p->flags & IORESOURCE_BUSY)
1684                        continue;
1685
1686                printk(KERN_WARNING "resource sanity check: requesting [mem %#010llx-%#010llx], which spans more than %s %pR\n",
1687                       (unsigned long long)addr,
1688                       (unsigned long long)(addr + size - 1),
1689                       p->name, p);
1690                err = -1;
1691                break;
1692        }
1693        read_unlock(&resource_lock);
1694
1695        return err;
1696}
1697
1698#ifdef CONFIG_STRICT_DEVMEM
1699static int strict_iomem_checks = 1;
1700#else
1701static int strict_iomem_checks;
1702#endif
1703
1704/*
1705 * Check if an address is exclusive to the kernel and must not be mapped to
1706 * user space, for example, via /dev/mem.
1707 *
1708 * Returns true if exclusive to the kernel, otherwise returns false.
1709 */
1710bool iomem_is_exclusive(u64 addr)
1711{
1712        const unsigned int exclusive_system_ram = IORESOURCE_SYSTEM_RAM |
1713                                                  IORESOURCE_EXCLUSIVE;
1714        bool skip_children = false, err = false;
1715        int size = PAGE_SIZE;
1716        struct resource *p;
1717
1718        addr = addr & PAGE_MASK;
1719
1720        read_lock(&resource_lock);
1721        for_each_resource(&iomem_resource, p, skip_children) {
1722                if (p->start >= addr + size)
1723                        break;
1724                if (p->end < addr) {
1725                        skip_children = true;
1726                        continue;
1727                }
1728                skip_children = false;
1729
1730                /*
1731                 * IORESOURCE_SYSTEM_RAM resources are exclusive if
1732                 * IORESOURCE_EXCLUSIVE is set, even if they
1733                 * are not busy and even if "iomem=relaxed" is set. The
1734                 * responsible driver dynamically adds/removes system RAM within
1735                 * such an area and uncontrolled access is dangerous.
1736                 */
1737                if ((p->flags & exclusive_system_ram) == exclusive_system_ram) {
1738                        err = true;
1739                        break;
1740                }
1741
1742                /*
1743                 * A resource is exclusive if IORESOURCE_EXCLUSIVE is set
1744                 * or CONFIG_IO_STRICT_DEVMEM is enabled and the
1745                 * resource is busy.
1746                 */
1747                if (!strict_iomem_checks || !(p->flags & IORESOURCE_BUSY))
1748                        continue;
1749                if (IS_ENABLED(CONFIG_IO_STRICT_DEVMEM)
1750                                || p->flags & IORESOURCE_EXCLUSIVE) {
1751                        err = true;
1752                        break;
1753                }
1754        }
1755        read_unlock(&resource_lock);
1756
1757        return err;
1758}
1759
1760struct resource_entry *resource_list_create_entry(struct resource *res,
1761                                                  size_t extra_size)
1762{
1763        struct resource_entry *entry;
1764
1765        entry = kzalloc(sizeof(*entry) + extra_size, GFP_KERNEL);
1766        if (entry) {
1767                INIT_LIST_HEAD(&entry->node);
1768                entry->res = res ? res : &entry->__res;
1769        }
1770
1771        return entry;
1772}
1773EXPORT_SYMBOL(resource_list_create_entry);
1774
1775void resource_list_free(struct list_head *head)
1776{
1777        struct resource_entry *entry, *tmp;
1778
1779        list_for_each_entry_safe(entry, tmp, head, node)
1780                resource_list_destroy_entry(entry);
1781}
1782EXPORT_SYMBOL(resource_list_free);
1783
1784#ifdef CONFIG_GET_FREE_REGION
1785#define GFR_DESCENDING          (1UL << 0)
1786#define GFR_REQUEST_REGION      (1UL << 1)
1787#define GFR_DEFAULT_ALIGN (1UL << PA_SECTION_SHIFT)
1788
1789static resource_size_t gfr_start(struct resource *base, resource_size_t size,
1790                                 resource_size_t align, unsigned long flags)
1791{
1792        if (flags & GFR_DESCENDING) {
1793                resource_size_t end;
1794
1795                end = min_t(resource_size_t, base->end,
1796                            (1ULL << MAX_PHYSMEM_BITS) - 1);
1797                return end - size + 1;
1798        }
1799
1800        return ALIGN(base->start, align);
1801}
1802
1803static bool gfr_continue(struct resource *base, resource_size_t addr,
1804                         resource_size_t size, unsigned long flags)
1805{
1806        if (flags & GFR_DESCENDING)
1807                return addr > size && addr >= base->start;
1808        /*
1809         * In the ascend case be careful that the last increment by
1810         * @size did not wrap 0.
1811         */
1812        return addr > addr - size &&
1813               addr <= min_t(resource_size_t, base->end,
1814                             (1ULL << MAX_PHYSMEM_BITS) - 1);
1815}
1816
1817static resource_size_t gfr_next(resource_size_t addr, resource_size_t size,
1818                                unsigned long flags)
1819{
1820        if (flags & GFR_DESCENDING)
1821                return addr - size;
1822        return addr + size;
1823}
1824
1825static void remove_free_mem_region(void *_res)
1826{
1827        struct resource *res = _res;
1828
1829        if (res->parent)
1830                remove_resource(res);
1831        free_resource(res);
1832}
1833
1834static struct resource *
1835get_free_mem_region(struct device *dev, struct resource *base,
1836                    resource_size_t size, const unsigned long align,
1837                    const char *name, const unsigned long desc,
1838                    const unsigned long flags)
1839{
1840        resource_size_t addr;
1841        struct resource *res;
1842        struct region_devres *dr = NULL;
1843
1844        size = ALIGN(size, align);
1845
1846        res = alloc_resource(GFP_KERNEL);
1847        if (!res)
1848                return ERR_PTR(-ENOMEM);
1849
1850        if (dev && (flags & GFR_REQUEST_REGION)) {
1851                dr = devres_alloc(devm_region_release,
1852                                sizeof(struct region_devres), GFP_KERNEL);
1853                if (!dr) {
1854                        free_resource(res);
1855                        return ERR_PTR(-ENOMEM);
1856                }
1857        } else if (dev) {
1858                if (devm_add_action_or_reset(dev, remove_free_mem_region, res))
1859                        return ERR_PTR(-ENOMEM);
1860        }
1861
1862        write_lock(&resource_lock);
1863        for (addr = gfr_start(base, size, align, flags);
1864             gfr_continue(base, addr, size, flags);
1865             addr = gfr_next(addr, size, flags)) {
1866                if (__region_intersects(base, addr, size, 0, IORES_DESC_NONE) !=
1867                    REGION_DISJOINT)
1868                        continue;
1869
1870                if (flags & GFR_REQUEST_REGION) {
1871                        if (__request_region_locked(res, &iomem_resource, addr,
1872                                                    size, name, 0))
1873                                break;
1874
1875                        if (dev) {
1876                                dr->parent = &iomem_resource;
1877                                dr->start = addr;
1878                                dr->n = size;
1879                                devres_add(dev, dr);
1880                        }
1881
1882                        res->desc = desc;
1883                        write_unlock(&resource_lock);
1884
1885
1886                        /*
1887                         * A driver is claiming this region so revoke any
1888                         * mappings.
1889                         */
1890                        revoke_iomem(res);
1891                } else {
1892                        res->start = addr;
1893                        res->end = addr + size - 1;
1894                        res->name = name;
1895                        res->desc = desc;
1896                        res->flags = IORESOURCE_MEM;
1897
1898                        /*
1899                         * Only succeed if the resource hosts an exclusive
1900                         * range after the insert
1901                         */
1902                        if (__insert_resource(base, res) || res->child)
1903                                break;
1904
1905                        write_unlock(&resource_lock);
1906                }
1907
1908                return res;
1909        }
1910        write_unlock(&resource_lock);
1911
1912        if (flags & GFR_REQUEST_REGION) {
1913                free_resource(res);
1914                devres_free(dr);
1915        } else if (dev)
1916                devm_release_action(dev, remove_free_mem_region, res);
1917
1918        return ERR_PTR(-ERANGE);
1919}
1920
1921/**
1922 * devm_request_free_mem_region - find free region for device private memory
1923 *
1924 * @dev: device struct to bind the resource to
1925 * @size: size in bytes of the device memory to add
1926 * @base: resource tree to look in
1927 *
1928 * This function tries to find an empty range of physical address big enough to
1929 * contain the new resource, so that it can later be hotplugged as ZONE_DEVICE
1930 * memory, which in turn allocates struct pages.
1931 */
1932struct resource *devm_request_free_mem_region(struct device *dev,
1933                struct resource *base, unsigned long size)
1934{
1935        unsigned long flags = GFR_DESCENDING | GFR_REQUEST_REGION;
1936
1937        return get_free_mem_region(dev, base, size, GFR_DEFAULT_ALIGN,
1938                                   dev_name(dev),
1939                                   IORES_DESC_DEVICE_PRIVATE_MEMORY, flags);
1940}
1941EXPORT_SYMBOL_GPL(devm_request_free_mem_region);
1942
1943struct resource *request_free_mem_region(struct resource *base,
1944                unsigned long size, const char *name)
1945{
1946        unsigned long flags = GFR_DESCENDING | GFR_REQUEST_REGION;
1947
1948        return get_free_mem_region(NULL, base, size, GFR_DEFAULT_ALIGN, name,
1949                                   IORES_DESC_DEVICE_PRIVATE_MEMORY, flags);
1950}
1951EXPORT_SYMBOL_GPL(request_free_mem_region);
1952
1953/**
1954 * alloc_free_mem_region - find a free region relative to @base
1955 * @base: resource that will parent the new resource
1956 * @size: size in bytes of memory to allocate from @base
1957 * @align: alignment requirements for the allocation
1958 * @name: resource name
1959 *
1960 * Buses like CXL, that can dynamically instantiate new memory regions,
1961 * need a method to allocate physical address space for those regions.
1962 * Allocate and insert a new resource to cover a free, unclaimed by a
1963 * descendant of @base, range in the span of @base.
1964 */
1965struct resource *alloc_free_mem_region(struct resource *base,
1966                                       unsigned long size, unsigned long align,
1967                                       const char *name)
1968{
1969        /* Default of ascending direction and insert resource */
1970        unsigned long flags = 0;
1971
1972        return get_free_mem_region(NULL, base, size, align, name,
1973                                   IORES_DESC_NONE, flags);
1974}
1975EXPORT_SYMBOL_NS_GPL(alloc_free_mem_region, CXL);
1976#endif /* CONFIG_GET_FREE_REGION */
1977
1978static int __init strict_iomem(char *str)
1979{
1980        if (strstr(str, "relaxed"))
1981                strict_iomem_checks = 0;
1982        if (strstr(str, "strict"))
1983                strict_iomem_checks = 1;
1984        return 1;
1985}
1986
1987static int iomem_fs_init_fs_context(struct fs_context *fc)
1988{
1989        return init_pseudo(fc, DEVMEM_MAGIC) ? 0 : -ENOMEM;
1990}
1991
1992static struct file_system_type iomem_fs_type = {
1993        .name           = "iomem",
1994        .owner          = THIS_MODULE,
1995        .init_fs_context = iomem_fs_init_fs_context,
1996        .kill_sb        = kill_anon_super,
1997};
1998
1999static int __init iomem_init_inode(void)
2000{
2001        static struct vfsmount *iomem_vfs_mount;
2002        static int iomem_fs_cnt;
2003        struct inode *inode;
2004        int rc;
2005
2006        rc = simple_pin_fs(&iomem_fs_type, &iomem_vfs_mount, &iomem_fs_cnt);
2007        if (rc < 0) {
2008                pr_err("Cannot mount iomem pseudo filesystem: %d\n", rc);
2009                return rc;
2010        }
2011
2012        inode = alloc_anon_inode(iomem_vfs_mount->mnt_sb);
2013        if (IS_ERR(inode)) {
2014                rc = PTR_ERR(inode);
2015                pr_err("Cannot allocate inode for iomem: %d\n", rc);
2016                simple_release_fs(&iomem_vfs_mount, &iomem_fs_cnt);
2017                return rc;
2018        }
2019
2020        /*
2021         * Publish iomem revocation inode initialized.
2022         * Pairs with smp_load_acquire() in revoke_iomem().
2023         */
2024        smp_store_release(&iomem_inode, inode);
2025
2026        return 0;
2027}
2028
2029fs_initcall(iomem_init_inode);
2030
2031__setup("iomem=", strict_iomem);
2032