~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/kernel/resource.c

Version: ~ [ linux-6.11.5 ] ~ [ linux-6.10.14 ] ~ [ linux-6.9.12 ] ~ [ linux-6.8.12 ] ~ [ linux-6.7.12 ] ~ [ linux-6.6.58 ] ~ [ linux-6.5.13 ] ~ [ linux-6.4.16 ] ~ [ linux-6.3.13 ] ~ [ linux-6.2.16 ] ~ [ linux-6.1.114 ] ~ [ linux-6.0.19 ] ~ [ linux-5.19.17 ] ~ [ linux-5.18.19 ] ~ [ linux-5.17.15 ] ~ [ linux-5.16.20 ] ~ [ linux-5.15.169 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.228 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.284 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.322 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.336 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.337 ] ~ [ linux-4.4.302 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.9 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

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

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

sflogo.php