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

TOMOYO Linux Cross Reference
Linux/mm/memblock.c

Version: ~ [ linux-5.16-rc3 ] ~ [ linux-5.15.5 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.82 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.162 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.218 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.256 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.291 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.293 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * Procedures for maintaining information about logical memory blocks.
  3  *
  4  * Peter Bergner, IBM Corp.     June 2001.
  5  * Copyright (C) 2001 Peter Bergner.
  6  *
  7  *      This program is free software; you can redistribute it and/or
  8  *      modify it under the terms of the GNU General Public License
  9  *      as published by the Free Software Foundation; either version
 10  *      2 of the License, or (at your option) any later version.
 11  */
 12 
 13 #include <linux/kernel.h>
 14 #include <linux/slab.h>
 15 #include <linux/init.h>
 16 #include <linux/bitops.h>
 17 #include <linux/poison.h>
 18 #include <linux/pfn.h>
 19 #include <linux/debugfs.h>
 20 #include <linux/seq_file.h>
 21 #include <linux/memblock.h>
 22 
 23 #include <asm/sections.h>
 24 #include <linux/io.h>
 25 
 26 #include "internal.h"
 27 
 28 static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
 29 static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
 30 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
 31 static struct memblock_region memblock_physmem_init_regions[INIT_PHYSMEM_REGIONS] __initdata_memblock;
 32 #endif
 33 
 34 struct memblock memblock __initdata_memblock = {
 35         .memory.regions         = memblock_memory_init_regions,
 36         .memory.cnt             = 1,    /* empty dummy entry */
 37         .memory.max             = INIT_MEMBLOCK_REGIONS,
 38         .memory.name            = "memory",
 39 
 40         .reserved.regions       = memblock_reserved_init_regions,
 41         .reserved.cnt           = 1,    /* empty dummy entry */
 42         .reserved.max           = INIT_MEMBLOCK_REGIONS,
 43         .reserved.name          = "reserved",
 44 
 45 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
 46         .physmem.regions        = memblock_physmem_init_regions,
 47         .physmem.cnt            = 1,    /* empty dummy entry */
 48         .physmem.max            = INIT_PHYSMEM_REGIONS,
 49         .physmem.name           = "physmem",
 50 #endif
 51 
 52         .bottom_up              = false,
 53         .current_limit          = MEMBLOCK_ALLOC_ANYWHERE,
 54 };
 55 
 56 int memblock_debug __initdata_memblock;
 57 #ifdef CONFIG_MOVABLE_NODE
 58 bool movable_node_enabled __initdata_memblock = false;
 59 #endif
 60 static bool system_has_some_mirror __initdata_memblock = false;
 61 static int memblock_can_resize __initdata_memblock;
 62 static int memblock_memory_in_slab __initdata_memblock = 0;
 63 static int memblock_reserved_in_slab __initdata_memblock = 0;
 64 
 65 ulong __init_memblock choose_memblock_flags(void)
 66 {
 67         return system_has_some_mirror ? MEMBLOCK_MIRROR : MEMBLOCK_NONE;
 68 }
 69 
 70 /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
 71 static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size)
 72 {
 73         return *size = min(*size, (phys_addr_t)ULLONG_MAX - base);
 74 }
 75 
 76 /*
 77  * Address comparison utilities
 78  */
 79 static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
 80                                        phys_addr_t base2, phys_addr_t size2)
 81 {
 82         return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
 83 }
 84 
 85 bool __init_memblock memblock_overlaps_region(struct memblock_type *type,
 86                                         phys_addr_t base, phys_addr_t size)
 87 {
 88         unsigned long i;
 89 
 90         for (i = 0; i < type->cnt; i++)
 91                 if (memblock_addrs_overlap(base, size, type->regions[i].base,
 92                                            type->regions[i].size))
 93                         break;
 94         return i < type->cnt;
 95 }
 96 
 97 /*
 98  * __memblock_find_range_bottom_up - find free area utility in bottom-up
 99  * @start: start of candidate range
100  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
101  * @size: size of free area to find
102  * @align: alignment of free area to find
103  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
104  * @flags: pick from blocks based on memory attributes
105  *
106  * Utility called from memblock_find_in_range_node(), find free area bottom-up.
107  *
108  * RETURNS:
109  * Found address on success, 0 on failure.
110  */
111 static phys_addr_t __init_memblock
112 __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end,
113                                 phys_addr_t size, phys_addr_t align, int nid,
114                                 ulong flags)
115 {
116         phys_addr_t this_start, this_end, cand;
117         u64 i;
118 
119         for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) {
120                 this_start = clamp(this_start, start, end);
121                 this_end = clamp(this_end, start, end);
122 
123                 cand = round_up(this_start, align);
124                 if (cand < this_end && this_end - cand >= size)
125                         return cand;
126         }
127 
128         return 0;
129 }
130 
131 /**
132  * __memblock_find_range_top_down - find free area utility, in top-down
133  * @start: start of candidate range
134  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
135  * @size: size of free area to find
136  * @align: alignment of free area to find
137  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
138  * @flags: pick from blocks based on memory attributes
139  *
140  * Utility called from memblock_find_in_range_node(), find free area top-down.
141  *
142  * RETURNS:
143  * Found address on success, 0 on failure.
144  */
145 static phys_addr_t __init_memblock
146 __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end,
147                                phys_addr_t size, phys_addr_t align, int nid,
148                                ulong flags)
149 {
150         phys_addr_t this_start, this_end, cand;
151         u64 i;
152 
153         for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end,
154                                         NULL) {
155                 this_start = clamp(this_start, start, end);
156                 this_end = clamp(this_end, start, end);
157 
158                 if (this_end < size)
159                         continue;
160 
161                 cand = round_down(this_end - size, align);
162                 if (cand >= this_start)
163                         return cand;
164         }
165 
166         return 0;
167 }
168 
169 /**
170  * memblock_find_in_range_node - find free area in given range and node
171  * @size: size of free area to find
172  * @align: alignment of free area to find
173  * @start: start of candidate range
174  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
175  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
176  * @flags: pick from blocks based on memory attributes
177  *
178  * Find @size free area aligned to @align in the specified range and node.
179  *
180  * When allocation direction is bottom-up, the @start should be greater
181  * than the end of the kernel image. Otherwise, it will be trimmed. The
182  * reason is that we want the bottom-up allocation just near the kernel
183  * image so it is highly likely that the allocated memory and the kernel
184  * will reside in the same node.
185  *
186  * If bottom-up allocation failed, will try to allocate memory top-down.
187  *
188  * RETURNS:
189  * Found address on success, 0 on failure.
190  */
191 phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size,
192                                         phys_addr_t align, phys_addr_t start,
193                                         phys_addr_t end, int nid, ulong flags)
194 {
195         phys_addr_t kernel_end, ret;
196 
197         /* pump up @end */
198         if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
199                 end = memblock.current_limit;
200 
201         /* avoid allocating the first page */
202         start = max_t(phys_addr_t, start, PAGE_SIZE);
203         end = max(start, end);
204         kernel_end = __pa_symbol(_end);
205 
206         /*
207          * try bottom-up allocation only when bottom-up mode
208          * is set and @end is above the kernel image.
209          */
210         if (memblock_bottom_up() && end > kernel_end) {
211                 phys_addr_t bottom_up_start;
212 
213                 /* make sure we will allocate above the kernel */
214                 bottom_up_start = max(start, kernel_end);
215 
216                 /* ok, try bottom-up allocation first */
217                 ret = __memblock_find_range_bottom_up(bottom_up_start, end,
218                                                       size, align, nid, flags);
219                 if (ret)
220                         return ret;
221 
222                 /*
223                  * we always limit bottom-up allocation above the kernel,
224                  * but top-down allocation doesn't have the limit, so
225                  * retrying top-down allocation may succeed when bottom-up
226                  * allocation failed.
227                  *
228                  * bottom-up allocation is expected to be fail very rarely,
229                  * so we use WARN_ONCE() here to see the stack trace if
230                  * fail happens.
231                  */
232                 WARN_ONCE(1, "memblock: bottom-up allocation failed, memory hotunplug may be affected\n");
233         }
234 
235         return __memblock_find_range_top_down(start, end, size, align, nid,
236                                               flags);
237 }
238 
239 /**
240  * memblock_find_in_range - find free area in given range
241  * @start: start of candidate range
242  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
243  * @size: size of free area to find
244  * @align: alignment of free area to find
245  *
246  * Find @size free area aligned to @align in the specified range.
247  *
248  * RETURNS:
249  * Found address on success, 0 on failure.
250  */
251 phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
252                                         phys_addr_t end, phys_addr_t size,
253                                         phys_addr_t align)
254 {
255         phys_addr_t ret;
256         ulong flags = choose_memblock_flags();
257 
258 again:
259         ret = memblock_find_in_range_node(size, align, start, end,
260                                             NUMA_NO_NODE, flags);
261 
262         if (!ret && (flags & MEMBLOCK_MIRROR)) {
263                 pr_warn("Could not allocate %pap bytes of mirrored memory\n",
264                         &size);
265                 flags &= ~MEMBLOCK_MIRROR;
266                 goto again;
267         }
268 
269         return ret;
270 }
271 
272 static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
273 {
274         type->total_size -= type->regions[r].size;
275         memmove(&type->regions[r], &type->regions[r + 1],
276                 (type->cnt - (r + 1)) * sizeof(type->regions[r]));
277         type->cnt--;
278 
279         /* Special case for empty arrays */
280         if (type->cnt == 0) {
281                 WARN_ON(type->total_size != 0);
282                 type->cnt = 1;
283                 type->regions[0].base = 0;
284                 type->regions[0].size = 0;
285                 type->regions[0].flags = 0;
286                 memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
287         }
288 }
289 
290 #ifdef CONFIG_ARCH_DISCARD_MEMBLOCK
291 /**
292  * Discard memory and reserved arrays if they were allocated
293  */
294 void __init memblock_discard(void)
295 {
296         phys_addr_t addr, size;
297 
298         if (memblock.reserved.regions != memblock_reserved_init_regions) {
299                 addr = __pa(memblock.reserved.regions);
300                 size = PAGE_ALIGN(sizeof(struct memblock_region) *
301                                   memblock.reserved.max);
302                 __memblock_free_late(addr, size);
303         }
304 
305         if (memblock.memory.regions != memblock_memory_init_regions) {
306                 addr = __pa(memblock.memory.regions);
307                 size = PAGE_ALIGN(sizeof(struct memblock_region) *
308                                   memblock.memory.max);
309                 __memblock_free_late(addr, size);
310         }
311 }
312 #endif
313 
314 /**
315  * memblock_double_array - double the size of the memblock regions array
316  * @type: memblock type of the regions array being doubled
317  * @new_area_start: starting address of memory range to avoid overlap with
318  * @new_area_size: size of memory range to avoid overlap with
319  *
320  * Double the size of the @type regions array. If memblock is being used to
321  * allocate memory for a new reserved regions array and there is a previously
322  * allocated memory range [@new_area_start,@new_area_start+@new_area_size]
323  * waiting to be reserved, ensure the memory used by the new array does
324  * not overlap.
325  *
326  * RETURNS:
327  * 0 on success, -1 on failure.
328  */
329 static int __init_memblock memblock_double_array(struct memblock_type *type,
330                                                 phys_addr_t new_area_start,
331                                                 phys_addr_t new_area_size)
332 {
333         struct memblock_region *new_array, *old_array;
334         phys_addr_t old_alloc_size, new_alloc_size;
335         phys_addr_t old_size, new_size, addr;
336         int use_slab = slab_is_available();
337         int *in_slab;
338 
339         /* We don't allow resizing until we know about the reserved regions
340          * of memory that aren't suitable for allocation
341          */
342         if (!memblock_can_resize)
343                 return -1;
344 
345         /* Calculate new doubled size */
346         old_size = type->max * sizeof(struct memblock_region);
347         new_size = old_size << 1;
348         /*
349          * We need to allocated new one align to PAGE_SIZE,
350          *   so we can free them completely later.
351          */
352         old_alloc_size = PAGE_ALIGN(old_size);
353         new_alloc_size = PAGE_ALIGN(new_size);
354 
355         /* Retrieve the slab flag */
356         if (type == &memblock.memory)
357                 in_slab = &memblock_memory_in_slab;
358         else
359                 in_slab = &memblock_reserved_in_slab;
360 
361         /* Try to find some space for it.
362          *
363          * WARNING: We assume that either slab_is_available() and we use it or
364          * we use MEMBLOCK for allocations. That means that this is unsafe to
365          * use when bootmem is currently active (unless bootmem itself is
366          * implemented on top of MEMBLOCK which isn't the case yet)
367          *
368          * This should however not be an issue for now, as we currently only
369          * call into MEMBLOCK while it's still active, or much later when slab
370          * is active for memory hotplug operations
371          */
372         if (use_slab) {
373                 new_array = kmalloc(new_size, GFP_KERNEL);
374                 addr = new_array ? __pa(new_array) : 0;
375         } else {
376                 /* only exclude range when trying to double reserved.regions */
377                 if (type != &memblock.reserved)
378                         new_area_start = new_area_size = 0;
379 
380                 addr = memblock_find_in_range(new_area_start + new_area_size,
381                                                 memblock.current_limit,
382                                                 new_alloc_size, PAGE_SIZE);
383                 if (!addr && new_area_size)
384                         addr = memblock_find_in_range(0,
385                                 min(new_area_start, memblock.current_limit),
386                                 new_alloc_size, PAGE_SIZE);
387 
388                 new_array = addr ? __va(addr) : NULL;
389         }
390         if (!addr) {
391                 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
392                        type->name, type->max, type->max * 2);
393                 return -1;
394         }
395 
396         memblock_dbg("memblock: %s is doubled to %ld at [%#010llx-%#010llx]",
397                         type->name, type->max * 2, (u64)addr,
398                         (u64)addr + new_size - 1);
399 
400         /*
401          * Found space, we now need to move the array over before we add the
402          * reserved region since it may be our reserved array itself that is
403          * full.
404          */
405         memcpy(new_array, type->regions, old_size);
406         memset(new_array + type->max, 0, old_size);
407         old_array = type->regions;
408         type->regions = new_array;
409         type->max <<= 1;
410 
411         /* Free old array. We needn't free it if the array is the static one */
412         if (*in_slab)
413                 kfree(old_array);
414         else if (old_array != memblock_memory_init_regions &&
415                  old_array != memblock_reserved_init_regions)
416                 memblock_free(__pa(old_array), old_alloc_size);
417 
418         /*
419          * Reserve the new array if that comes from the memblock.  Otherwise, we
420          * needn't do it
421          */
422         if (!use_slab)
423                 BUG_ON(memblock_reserve(addr, new_alloc_size));
424 
425         /* Update slab flag */
426         *in_slab = use_slab;
427 
428         return 0;
429 }
430 
431 /**
432  * memblock_merge_regions - merge neighboring compatible regions
433  * @type: memblock type to scan
434  *
435  * Scan @type and merge neighboring compatible regions.
436  */
437 static void __init_memblock memblock_merge_regions(struct memblock_type *type)
438 {
439         int i = 0;
440 
441         /* cnt never goes below 1 */
442         while (i < type->cnt - 1) {
443                 struct memblock_region *this = &type->regions[i];
444                 struct memblock_region *next = &type->regions[i + 1];
445 
446                 if (this->base + this->size != next->base ||
447                     memblock_get_region_node(this) !=
448                     memblock_get_region_node(next) ||
449                     this->flags != next->flags) {
450                         BUG_ON(this->base + this->size > next->base);
451                         i++;
452                         continue;
453                 }
454 
455                 this->size += next->size;
456                 /* move forward from next + 1, index of which is i + 2 */
457                 memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
458                 type->cnt--;
459         }
460 }
461 
462 /**
463  * memblock_insert_region - insert new memblock region
464  * @type:       memblock type to insert into
465  * @idx:        index for the insertion point
466  * @base:       base address of the new region
467  * @size:       size of the new region
468  * @nid:        node id of the new region
469  * @flags:      flags of the new region
470  *
471  * Insert new memblock region [@base,@base+@size) into @type at @idx.
472  * @type must already have extra room to accommodate the new region.
473  */
474 static void __init_memblock memblock_insert_region(struct memblock_type *type,
475                                                    int idx, phys_addr_t base,
476                                                    phys_addr_t size,
477                                                    int nid, unsigned long flags)
478 {
479         struct memblock_region *rgn = &type->regions[idx];
480 
481         BUG_ON(type->cnt >= type->max);
482         memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
483         rgn->base = base;
484         rgn->size = size;
485         rgn->flags = flags;
486         memblock_set_region_node(rgn, nid);
487         type->cnt++;
488         type->total_size += size;
489 }
490 
491 /**
492  * memblock_add_range - add new memblock region
493  * @type: memblock type to add new region into
494  * @base: base address of the new region
495  * @size: size of the new region
496  * @nid: nid of the new region
497  * @flags: flags of the new region
498  *
499  * Add new memblock region [@base,@base+@size) into @type.  The new region
500  * is allowed to overlap with existing ones - overlaps don't affect already
501  * existing regions.  @type is guaranteed to be minimal (all neighbouring
502  * compatible regions are merged) after the addition.
503  *
504  * RETURNS:
505  * 0 on success, -errno on failure.
506  */
507 int __init_memblock memblock_add_range(struct memblock_type *type,
508                                 phys_addr_t base, phys_addr_t size,
509                                 int nid, unsigned long flags)
510 {
511         bool insert = false;
512         phys_addr_t obase = base;
513         phys_addr_t end = base + memblock_cap_size(base, &size);
514         int idx, nr_new;
515         struct memblock_region *rgn;
516 
517         if (!size)
518                 return 0;
519 
520         /* special case for empty array */
521         if (type->regions[0].size == 0) {
522                 WARN_ON(type->cnt != 1 || type->total_size);
523                 type->regions[0].base = base;
524                 type->regions[0].size = size;
525                 type->regions[0].flags = flags;
526                 memblock_set_region_node(&type->regions[0], nid);
527                 type->total_size = size;
528                 return 0;
529         }
530 repeat:
531         /*
532          * The following is executed twice.  Once with %false @insert and
533          * then with %true.  The first counts the number of regions needed
534          * to accommodate the new area.  The second actually inserts them.
535          */
536         base = obase;
537         nr_new = 0;
538 
539         for_each_memblock_type(type, rgn) {
540                 phys_addr_t rbase = rgn->base;
541                 phys_addr_t rend = rbase + rgn->size;
542 
543                 if (rbase >= end)
544                         break;
545                 if (rend <= base)
546                         continue;
547                 /*
548                  * @rgn overlaps.  If it separates the lower part of new
549                  * area, insert that portion.
550                  */
551                 if (rbase > base) {
552 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
553                         WARN_ON(nid != memblock_get_region_node(rgn));
554 #endif
555                         WARN_ON(flags != rgn->flags);
556                         nr_new++;
557                         if (insert)
558                                 memblock_insert_region(type, idx++, base,
559                                                        rbase - base, nid,
560                                                        flags);
561                 }
562                 /* area below @rend is dealt with, forget about it */
563                 base = min(rend, end);
564         }
565 
566         /* insert the remaining portion */
567         if (base < end) {
568                 nr_new++;
569                 if (insert)
570                         memblock_insert_region(type, idx, base, end - base,
571                                                nid, flags);
572         }
573 
574         if (!nr_new)
575                 return 0;
576 
577         /*
578          * If this was the first round, resize array and repeat for actual
579          * insertions; otherwise, merge and return.
580          */
581         if (!insert) {
582                 while (type->cnt + nr_new > type->max)
583                         if (memblock_double_array(type, obase, size) < 0)
584                                 return -ENOMEM;
585                 insert = true;
586                 goto repeat;
587         } else {
588                 memblock_merge_regions(type);
589                 return 0;
590         }
591 }
592 
593 int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
594                                        int nid)
595 {
596         return memblock_add_range(&memblock.memory, base, size, nid, 0);
597 }
598 
599 int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
600 {
601         phys_addr_t end = base + size - 1;
602 
603         memblock_dbg("memblock_add: [%pa-%pa] %pF\n",
604                      &base, &end, (void *)_RET_IP_);
605 
606         return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0);
607 }
608 
609 /**
610  * memblock_isolate_range - isolate given range into disjoint memblocks
611  * @type: memblock type to isolate range for
612  * @base: base of range to isolate
613  * @size: size of range to isolate
614  * @start_rgn: out parameter for the start of isolated region
615  * @end_rgn: out parameter for the end of isolated region
616  *
617  * Walk @type and ensure that regions don't cross the boundaries defined by
618  * [@base,@base+@size).  Crossing regions are split at the boundaries,
619  * which may create at most two more regions.  The index of the first
620  * region inside the range is returned in *@start_rgn and end in *@end_rgn.
621  *
622  * RETURNS:
623  * 0 on success, -errno on failure.
624  */
625 static int __init_memblock memblock_isolate_range(struct memblock_type *type,
626                                         phys_addr_t base, phys_addr_t size,
627                                         int *start_rgn, int *end_rgn)
628 {
629         phys_addr_t end = base + memblock_cap_size(base, &size);
630         int idx;
631         struct memblock_region *rgn;
632 
633         *start_rgn = *end_rgn = 0;
634 
635         if (!size)
636                 return 0;
637 
638         /* we'll create at most two more regions */
639         while (type->cnt + 2 > type->max)
640                 if (memblock_double_array(type, base, size) < 0)
641                         return -ENOMEM;
642 
643         for_each_memblock_type(type, rgn) {
644                 phys_addr_t rbase = rgn->base;
645                 phys_addr_t rend = rbase + rgn->size;
646 
647                 if (rbase >= end)
648                         break;
649                 if (rend <= base)
650                         continue;
651 
652                 if (rbase < base) {
653                         /*
654                          * @rgn intersects from below.  Split and continue
655                          * to process the next region - the new top half.
656                          */
657                         rgn->base = base;
658                         rgn->size -= base - rbase;
659                         type->total_size -= base - rbase;
660                         memblock_insert_region(type, idx, rbase, base - rbase,
661                                                memblock_get_region_node(rgn),
662                                                rgn->flags);
663                 } else if (rend > end) {
664                         /*
665                          * @rgn intersects from above.  Split and redo the
666                          * current region - the new bottom half.
667                          */
668                         rgn->base = end;
669                         rgn->size -= end - rbase;
670                         type->total_size -= end - rbase;
671                         memblock_insert_region(type, idx--, rbase, end - rbase,
672                                                memblock_get_region_node(rgn),
673                                                rgn->flags);
674                 } else {
675                         /* @rgn is fully contained, record it */
676                         if (!*end_rgn)
677                                 *start_rgn = idx;
678                         *end_rgn = idx + 1;
679                 }
680         }
681 
682         return 0;
683 }
684 
685 static int __init_memblock memblock_remove_range(struct memblock_type *type,
686                                           phys_addr_t base, phys_addr_t size)
687 {
688         int start_rgn, end_rgn;
689         int i, ret;
690 
691         ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
692         if (ret)
693                 return ret;
694 
695         for (i = end_rgn - 1; i >= start_rgn; i--)
696                 memblock_remove_region(type, i);
697         return 0;
698 }
699 
700 int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
701 {
702         return memblock_remove_range(&memblock.memory, base, size);
703 }
704 
705 
706 int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
707 {
708         phys_addr_t end = base + size - 1;
709 
710         memblock_dbg("   memblock_free: [%pa-%pa] %pF\n",
711                      &base, &end, (void *)_RET_IP_);
712 
713         kmemleak_free_part_phys(base, size);
714         return memblock_remove_range(&memblock.reserved, base, size);
715 }
716 
717 int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
718 {
719         phys_addr_t end = base + size - 1;
720 
721         memblock_dbg("memblock_reserve: [%pa-%pa] %pF\n",
722                      &base, &end, (void *)_RET_IP_);
723 
724         return memblock_add_range(&memblock.reserved, base, size, MAX_NUMNODES, 0);
725 }
726 
727 /**
728  *
729  * This function isolates region [@base, @base + @size), and sets/clears flag
730  *
731  * Return 0 on success, -errno on failure.
732  */
733 static int __init_memblock memblock_setclr_flag(phys_addr_t base,
734                                 phys_addr_t size, int set, int flag)
735 {
736         struct memblock_type *type = &memblock.memory;
737         int i, ret, start_rgn, end_rgn;
738 
739         ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
740         if (ret)
741                 return ret;
742 
743         for (i = start_rgn; i < end_rgn; i++)
744                 if (set)
745                         memblock_set_region_flags(&type->regions[i], flag);
746                 else
747                         memblock_clear_region_flags(&type->regions[i], flag);
748 
749         memblock_merge_regions(type);
750         return 0;
751 }
752 
753 /**
754  * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
755  * @base: the base phys addr of the region
756  * @size: the size of the region
757  *
758  * Return 0 on success, -errno on failure.
759  */
760 int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
761 {
762         return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG);
763 }
764 
765 /**
766  * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
767  * @base: the base phys addr of the region
768  * @size: the size of the region
769  *
770  * Return 0 on success, -errno on failure.
771  */
772 int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
773 {
774         return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG);
775 }
776 
777 /**
778  * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR.
779  * @base: the base phys addr of the region
780  * @size: the size of the region
781  *
782  * Return 0 on success, -errno on failure.
783  */
784 int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
785 {
786         system_has_some_mirror = true;
787 
788         return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR);
789 }
790 
791 /**
792  * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP.
793  * @base: the base phys addr of the region
794  * @size: the size of the region
795  *
796  * Return 0 on success, -errno on failure.
797  */
798 int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size)
799 {
800         return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP);
801 }
802 
803 /**
804  * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region.
805  * @base: the base phys addr of the region
806  * @size: the size of the region
807  *
808  * Return 0 on success, -errno on failure.
809  */
810 int __init_memblock memblock_clear_nomap(phys_addr_t base, phys_addr_t size)
811 {
812         return memblock_setclr_flag(base, size, 0, MEMBLOCK_NOMAP);
813 }
814 
815 /**
816  * __next_reserved_mem_region - next function for for_each_reserved_region()
817  * @idx: pointer to u64 loop variable
818  * @out_start: ptr to phys_addr_t for start address of the region, can be %NULL
819  * @out_end: ptr to phys_addr_t for end address of the region, can be %NULL
820  *
821  * Iterate over all reserved memory regions.
822  */
823 void __init_memblock __next_reserved_mem_region(u64 *idx,
824                                            phys_addr_t *out_start,
825                                            phys_addr_t *out_end)
826 {
827         struct memblock_type *type = &memblock.reserved;
828 
829         if (*idx < type->cnt) {
830                 struct memblock_region *r = &type->regions[*idx];
831                 phys_addr_t base = r->base;
832                 phys_addr_t size = r->size;
833 
834                 if (out_start)
835                         *out_start = base;
836                 if (out_end)
837                         *out_end = base + size - 1;
838 
839                 *idx += 1;
840                 return;
841         }
842 
843         /* signal end of iteration */
844         *idx = ULLONG_MAX;
845 }
846 
847 /**
848  * __next__mem_range - next function for for_each_free_mem_range() etc.
849  * @idx: pointer to u64 loop variable
850  * @nid: node selector, %NUMA_NO_NODE for all nodes
851  * @flags: pick from blocks based on memory attributes
852  * @type_a: pointer to memblock_type from where the range is taken
853  * @type_b: pointer to memblock_type which excludes memory from being taken
854  * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
855  * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
856  * @out_nid: ptr to int for nid of the range, can be %NULL
857  *
858  * Find the first area from *@idx which matches @nid, fill the out
859  * parameters, and update *@idx for the next iteration.  The lower 32bit of
860  * *@idx contains index into type_a and the upper 32bit indexes the
861  * areas before each region in type_b.  For example, if type_b regions
862  * look like the following,
863  *
864  *      0:[0-16), 1:[32-48), 2:[128-130)
865  *
866  * The upper 32bit indexes the following regions.
867  *
868  *      0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
869  *
870  * As both region arrays are sorted, the function advances the two indices
871  * in lockstep and returns each intersection.
872  */
873 void __init_memblock __next_mem_range(u64 *idx, int nid, ulong flags,
874                                       struct memblock_type *type_a,
875                                       struct memblock_type *type_b,
876                                       phys_addr_t *out_start,
877                                       phys_addr_t *out_end, int *out_nid)
878 {
879         int idx_a = *idx & 0xffffffff;
880         int idx_b = *idx >> 32;
881 
882         if (WARN_ONCE(nid == MAX_NUMNODES,
883         "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
884                 nid = NUMA_NO_NODE;
885 
886         for (; idx_a < type_a->cnt; idx_a++) {
887                 struct memblock_region *m = &type_a->regions[idx_a];
888 
889                 phys_addr_t m_start = m->base;
890                 phys_addr_t m_end = m->base + m->size;
891                 int         m_nid = memblock_get_region_node(m);
892 
893                 /* only memory regions are associated with nodes, check it */
894                 if (nid != NUMA_NO_NODE && nid != m_nid)
895                         continue;
896 
897                 /* skip hotpluggable memory regions if needed */
898                 if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
899                         continue;
900 
901                 /* if we want mirror memory skip non-mirror memory regions */
902                 if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
903                         continue;
904 
905                 /* skip nomap memory unless we were asked for it explicitly */
906                 if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
907                         continue;
908 
909                 if (!type_b) {
910                         if (out_start)
911                                 *out_start = m_start;
912                         if (out_end)
913                                 *out_end = m_end;
914                         if (out_nid)
915                                 *out_nid = m_nid;
916                         idx_a++;
917                         *idx = (u32)idx_a | (u64)idx_b << 32;
918                         return;
919                 }
920 
921                 /* scan areas before each reservation */
922                 for (; idx_b < type_b->cnt + 1; idx_b++) {
923                         struct memblock_region *r;
924                         phys_addr_t r_start;
925                         phys_addr_t r_end;
926 
927                         r = &type_b->regions[idx_b];
928                         r_start = idx_b ? r[-1].base + r[-1].size : 0;
929                         r_end = idx_b < type_b->cnt ?
930                                 r->base : ULLONG_MAX;
931 
932                         /*
933                          * if idx_b advanced past idx_a,
934                          * break out to advance idx_a
935                          */
936                         if (r_start >= m_end)
937                                 break;
938                         /* if the two regions intersect, we're done */
939                         if (m_start < r_end) {
940                                 if (out_start)
941                                         *out_start =
942                                                 max(m_start, r_start);
943                                 if (out_end)
944                                         *out_end = min(m_end, r_end);
945                                 if (out_nid)
946                                         *out_nid = m_nid;
947                                 /*
948                                  * The region which ends first is
949                                  * advanced for the next iteration.
950                                  */
951                                 if (m_end <= r_end)
952                                         idx_a++;
953                                 else
954                                         idx_b++;
955                                 *idx = (u32)idx_a | (u64)idx_b << 32;
956                                 return;
957                         }
958                 }
959         }
960 
961         /* signal end of iteration */
962         *idx = ULLONG_MAX;
963 }
964 
965 /**
966  * __next_mem_range_rev - generic next function for for_each_*_range_rev()
967  *
968  * Finds the next range from type_a which is not marked as unsuitable
969  * in type_b.
970  *
971  * @idx: pointer to u64 loop variable
972  * @nid: node selector, %NUMA_NO_NODE for all nodes
973  * @flags: pick from blocks based on memory attributes
974  * @type_a: pointer to memblock_type from where the range is taken
975  * @type_b: pointer to memblock_type which excludes memory from being taken
976  * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
977  * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
978  * @out_nid: ptr to int for nid of the range, can be %NULL
979  *
980  * Reverse of __next_mem_range().
981  */
982 void __init_memblock __next_mem_range_rev(u64 *idx, int nid, ulong flags,
983                                           struct memblock_type *type_a,
984                                           struct memblock_type *type_b,
985                                           phys_addr_t *out_start,
986                                           phys_addr_t *out_end, int *out_nid)
987 {
988         int idx_a = *idx & 0xffffffff;
989         int idx_b = *idx >> 32;
990 
991         if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
992                 nid = NUMA_NO_NODE;
993 
994         if (*idx == (u64)ULLONG_MAX) {
995                 idx_a = type_a->cnt - 1;
996                 if (type_b != NULL)
997                         idx_b = type_b->cnt;
998                 else
999                         idx_b = 0;
1000         }
1001 
1002         for (; idx_a >= 0; idx_a--) {
1003                 struct memblock_region *m = &type_a->regions[idx_a];
1004 
1005                 phys_addr_t m_start = m->base;
1006                 phys_addr_t m_end = m->base + m->size;
1007                 int m_nid = memblock_get_region_node(m);
1008 
1009                 /* only memory regions are associated with nodes, check it */
1010                 if (nid != NUMA_NO_NODE && nid != m_nid)
1011                         continue;
1012 
1013                 /* skip hotpluggable memory regions if needed */
1014                 if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
1015                         continue;
1016 
1017                 /* if we want mirror memory skip non-mirror memory regions */
1018                 if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
1019                         continue;
1020 
1021                 /* skip nomap memory unless we were asked for it explicitly */
1022                 if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
1023                         continue;
1024 
1025                 if (!type_b) {
1026                         if (out_start)
1027                                 *out_start = m_start;
1028                         if (out_end)
1029                                 *out_end = m_end;
1030                         if (out_nid)
1031                                 *out_nid = m_nid;
1032                         idx_a--;
1033                         *idx = (u32)idx_a | (u64)idx_b << 32;
1034                         return;
1035                 }
1036 
1037                 /* scan areas before each reservation */
1038                 for (; idx_b >= 0; idx_b--) {
1039                         struct memblock_region *r;
1040                         phys_addr_t r_start;
1041                         phys_addr_t r_end;
1042 
1043                         r = &type_b->regions[idx_b];
1044                         r_start = idx_b ? r[-1].base + r[-1].size : 0;
1045                         r_end = idx_b < type_b->cnt ?
1046                                 r->base : ULLONG_MAX;
1047                         /*
1048                          * if idx_b advanced past idx_a,
1049                          * break out to advance idx_a
1050                          */
1051 
1052                         if (r_end <= m_start)
1053                                 break;
1054                         /* if the two regions intersect, we're done */
1055                         if (m_end > r_start) {
1056                                 if (out_start)
1057                                         *out_start = max(m_start, r_start);
1058                                 if (out_end)
1059                                         *out_end = min(m_end, r_end);
1060                                 if (out_nid)
1061                                         *out_nid = m_nid;
1062                                 if (m_start >= r_start)
1063                                         idx_a--;
1064                                 else
1065                                         idx_b--;
1066                                 *idx = (u32)idx_a | (u64)idx_b << 32;
1067                                 return;
1068                         }
1069                 }
1070         }
1071         /* signal end of iteration */
1072         *idx = ULLONG_MAX;
1073 }
1074 
1075 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1076 /*
1077  * Common iterator interface used to define for_each_mem_range().
1078  */
1079 void __init_memblock __next_mem_pfn_range(int *idx, int nid,
1080                                 unsigned long *out_start_pfn,
1081                                 unsigned long *out_end_pfn, int *out_nid)
1082 {
1083         struct memblock_type *type = &memblock.memory;
1084         struct memblock_region *r;
1085 
1086         while (++*idx < type->cnt) {
1087                 r = &type->regions[*idx];
1088 
1089                 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
1090                         continue;
1091                 if (nid == MAX_NUMNODES || nid == r->nid)
1092                         break;
1093         }
1094         if (*idx >= type->cnt) {
1095                 *idx = -1;
1096                 return;
1097         }
1098 
1099         if (out_start_pfn)
1100                 *out_start_pfn = PFN_UP(r->base);
1101         if (out_end_pfn)
1102                 *out_end_pfn = PFN_DOWN(r->base + r->size);
1103         if (out_nid)
1104                 *out_nid = r->nid;
1105 }
1106 
1107 unsigned long __init_memblock memblock_next_valid_pfn(unsigned long pfn,
1108                                                       unsigned long max_pfn)
1109 {
1110         struct memblock_type *type = &memblock.memory;
1111         unsigned int right = type->cnt;
1112         unsigned int mid, left = 0;
1113         phys_addr_t addr = PFN_PHYS(pfn + 1);
1114 
1115         do {
1116                 mid = (right + left) / 2;
1117 
1118                 if (addr < type->regions[mid].base)
1119                         right = mid;
1120                 else if (addr >= (type->regions[mid].base +
1121                                   type->regions[mid].size))
1122                         left = mid + 1;
1123                 else {
1124                         /* addr is within the region, so pfn + 1 is valid */
1125                         return min(pfn + 1, max_pfn);
1126                 }
1127         } while (left < right);
1128 
1129         if (right == type->cnt)
1130                 return max_pfn;
1131         else
1132                 return min(PHYS_PFN(type->regions[right].base), max_pfn);
1133 }
1134 
1135 /**
1136  * memblock_set_node - set node ID on memblock regions
1137  * @base: base of area to set node ID for
1138  * @size: size of area to set node ID for
1139  * @type: memblock type to set node ID for
1140  * @nid: node ID to set
1141  *
1142  * Set the nid of memblock @type regions in [@base,@base+@size) to @nid.
1143  * Regions which cross the area boundaries are split as necessary.
1144  *
1145  * RETURNS:
1146  * 0 on success, -errno on failure.
1147  */
1148 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
1149                                       struct memblock_type *type, int nid)
1150 {
1151         int start_rgn, end_rgn;
1152         int i, ret;
1153 
1154         ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
1155         if (ret)
1156                 return ret;
1157 
1158         for (i = start_rgn; i < end_rgn; i++)
1159                 memblock_set_region_node(&type->regions[i], nid);
1160 
1161         memblock_merge_regions(type);
1162         return 0;
1163 }
1164 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
1165 
1166 static phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
1167                                         phys_addr_t align, phys_addr_t start,
1168                                         phys_addr_t end, int nid, ulong flags)
1169 {
1170         phys_addr_t found;
1171 
1172         if (!align)
1173                 align = SMP_CACHE_BYTES;
1174 
1175         found = memblock_find_in_range_node(size, align, start, end, nid,
1176                                             flags);
1177         if (found && !memblock_reserve(found, size)) {
1178                 /*
1179                  * The min_count is set to 0 so that memblock allocations are
1180                  * never reported as leaks.
1181                  */
1182                 kmemleak_alloc_phys(found, size, 0, 0);
1183                 return found;
1184         }
1185         return 0;
1186 }
1187 
1188 phys_addr_t __init memblock_alloc_range(phys_addr_t size, phys_addr_t align,
1189                                         phys_addr_t start, phys_addr_t end,
1190                                         ulong flags)
1191 {
1192         return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE,
1193                                         flags);
1194 }
1195 
1196 static phys_addr_t __init memblock_alloc_base_nid(phys_addr_t size,
1197                                         phys_addr_t align, phys_addr_t max_addr,
1198                                         int nid, ulong flags)
1199 {
1200         return memblock_alloc_range_nid(size, align, 0, max_addr, nid, flags);
1201 }
1202 
1203 phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
1204 {
1205         ulong flags = choose_memblock_flags();
1206         phys_addr_t ret;
1207 
1208 again:
1209         ret = memblock_alloc_base_nid(size, align, MEMBLOCK_ALLOC_ACCESSIBLE,
1210                                       nid, flags);
1211 
1212         if (!ret && (flags & MEMBLOCK_MIRROR)) {
1213                 flags &= ~MEMBLOCK_MIRROR;
1214                 goto again;
1215         }
1216         return ret;
1217 }
1218 
1219 phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1220 {
1221         return memblock_alloc_base_nid(size, align, max_addr, NUMA_NO_NODE,
1222                                        MEMBLOCK_NONE);
1223 }
1224 
1225 phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1226 {
1227         phys_addr_t alloc;
1228 
1229         alloc = __memblock_alloc_base(size, align, max_addr);
1230 
1231         if (alloc == 0)
1232                 panic("ERROR: Failed to allocate %pa bytes below %pa.\n",
1233                       &size, &max_addr);
1234 
1235         return alloc;
1236 }
1237 
1238 phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
1239 {
1240         return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1241 }
1242 
1243 phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1244 {
1245         phys_addr_t res = memblock_alloc_nid(size, align, nid);
1246 
1247         if (res)
1248                 return res;
1249         return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1250 }
1251 
1252 /**
1253  * memblock_virt_alloc_internal - allocate boot memory block
1254  * @size: size of memory block to be allocated in bytes
1255  * @align: alignment of the region and block's size
1256  * @min_addr: the lower bound of the memory region to allocate (phys address)
1257  * @max_addr: the upper bound of the memory region to allocate (phys address)
1258  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1259  *
1260  * The @min_addr limit is dropped if it can not be satisfied and the allocation
1261  * will fall back to memory below @min_addr. Also, allocation may fall back
1262  * to any node in the system if the specified node can not
1263  * hold the requested memory.
1264  *
1265  * The allocation is performed from memory region limited by
1266  * memblock.current_limit if @max_addr == %BOOTMEM_ALLOC_ACCESSIBLE.
1267  *
1268  * The memory block is aligned on SMP_CACHE_BYTES if @align == 0.
1269  *
1270  * The phys address of allocated boot memory block is converted to virtual and
1271  * allocated memory is reset to 0.
1272  *
1273  * In addition, function sets the min_count to 0 using kmemleak_alloc for
1274  * allocated boot memory block, so that it is never reported as leaks.
1275  *
1276  * RETURNS:
1277  * Virtual address of allocated memory block on success, NULL on failure.
1278  */
1279 static void * __init memblock_virt_alloc_internal(
1280                                 phys_addr_t size, phys_addr_t align,
1281                                 phys_addr_t min_addr, phys_addr_t max_addr,
1282                                 int nid)
1283 {
1284         phys_addr_t alloc;
1285         void *ptr;
1286         ulong flags = choose_memblock_flags();
1287 
1288         if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1289                 nid = NUMA_NO_NODE;
1290 
1291         /*
1292          * Detect any accidental use of these APIs after slab is ready, as at
1293          * this moment memblock may be deinitialized already and its
1294          * internal data may be destroyed (after execution of free_all_bootmem)
1295          */
1296         if (WARN_ON_ONCE(slab_is_available()))
1297                 return kzalloc_node(size, GFP_NOWAIT, nid);
1298 
1299         if (!align)
1300                 align = SMP_CACHE_BYTES;
1301 
1302         if (max_addr > memblock.current_limit)
1303                 max_addr = memblock.current_limit;
1304 again:
1305         alloc = memblock_find_in_range_node(size, align, min_addr, max_addr,
1306                                             nid, flags);
1307         if (alloc && !memblock_reserve(alloc, size))
1308                 goto done;
1309 
1310         if (nid != NUMA_NO_NODE) {
1311                 alloc = memblock_find_in_range_node(size, align, min_addr,
1312                                                     max_addr, NUMA_NO_NODE,
1313                                                     flags);
1314                 if (alloc && !memblock_reserve(alloc, size))
1315                         goto done;
1316         }
1317 
1318         if (min_addr) {
1319                 min_addr = 0;
1320                 goto again;
1321         }
1322 
1323         if (flags & MEMBLOCK_MIRROR) {
1324                 flags &= ~MEMBLOCK_MIRROR;
1325                 pr_warn("Could not allocate %pap bytes of mirrored memory\n",
1326                         &size);
1327                 goto again;
1328         }
1329 
1330         return NULL;
1331 done:
1332         ptr = phys_to_virt(alloc);
1333         memset(ptr, 0, size);
1334 
1335         /*
1336          * The min_count is set to 0 so that bootmem allocated blocks
1337          * are never reported as leaks. This is because many of these blocks
1338          * are only referred via the physical address which is not
1339          * looked up by kmemleak.
1340          */
1341         kmemleak_alloc(ptr, size, 0, 0);
1342 
1343         return ptr;
1344 }
1345 
1346 /**
1347  * memblock_virt_alloc_try_nid_nopanic - allocate boot memory block
1348  * @size: size of memory block to be allocated in bytes
1349  * @align: alignment of the region and block's size
1350  * @min_addr: the lower bound of the memory region from where the allocation
1351  *        is preferred (phys address)
1352  * @max_addr: the upper bound of the memory region from where the allocation
1353  *            is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1354  *            allocate only from memory limited by memblock.current_limit value
1355  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1356  *
1357  * Public version of _memblock_virt_alloc_try_nid_nopanic() which provides
1358  * additional debug information (including caller info), if enabled.
1359  *
1360  * RETURNS:
1361  * Virtual address of allocated memory block on success, NULL on failure.
1362  */
1363 void * __init memblock_virt_alloc_try_nid_nopanic(
1364                                 phys_addr_t size, phys_addr_t align,
1365                                 phys_addr_t min_addr, phys_addr_t max_addr,
1366                                 int nid)
1367 {
1368         memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1369                      __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1370                      (u64)max_addr, (void *)_RET_IP_);
1371         return memblock_virt_alloc_internal(size, align, min_addr,
1372                                              max_addr, nid);
1373 }
1374 
1375 /**
1376  * memblock_virt_alloc_try_nid - allocate boot memory block with panicking
1377  * @size: size of memory block to be allocated in bytes
1378  * @align: alignment of the region and block's size
1379  * @min_addr: the lower bound of the memory region from where the allocation
1380  *        is preferred (phys address)
1381  * @max_addr: the upper bound of the memory region from where the allocation
1382  *            is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1383  *            allocate only from memory limited by memblock.current_limit value
1384  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1385  *
1386  * Public panicking version of _memblock_virt_alloc_try_nid_nopanic()
1387  * which provides debug information (including caller info), if enabled,
1388  * and panics if the request can not be satisfied.
1389  *
1390  * RETURNS:
1391  * Virtual address of allocated memory block on success, NULL on failure.
1392  */
1393 void * __init memblock_virt_alloc_try_nid(
1394                         phys_addr_t size, phys_addr_t align,
1395                         phys_addr_t min_addr, phys_addr_t max_addr,
1396                         int nid)
1397 {
1398         void *ptr;
1399 
1400         memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1401                      __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1402                      (u64)max_addr, (void *)_RET_IP_);
1403         ptr = memblock_virt_alloc_internal(size, align,
1404                                            min_addr, max_addr, nid);
1405         if (ptr)
1406                 return ptr;
1407 
1408         panic("%s: Failed to allocate %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx\n",
1409               __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1410               (u64)max_addr);
1411         return NULL;
1412 }
1413 
1414 /**
1415  * __memblock_free_early - free boot memory block
1416  * @base: phys starting address of the  boot memory block
1417  * @size: size of the boot memory block in bytes
1418  *
1419  * Free boot memory block previously allocated by memblock_virt_alloc_xx() API.
1420  * The freeing memory will not be released to the buddy allocator.
1421  */
1422 void __init __memblock_free_early(phys_addr_t base, phys_addr_t size)
1423 {
1424         memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1425                      __func__, (u64)base, (u64)base + size - 1,
1426                      (void *)_RET_IP_);
1427         kmemleak_free_part_phys(base, size);
1428         memblock_remove_range(&memblock.reserved, base, size);
1429 }
1430 
1431 /*
1432  * __memblock_free_late - free bootmem block pages directly to buddy allocator
1433  * @addr: phys starting address of the  boot memory block
1434  * @size: size of the boot memory block in bytes
1435  *
1436  * This is only useful when the bootmem allocator has already been torn
1437  * down, but we are still initializing the system.  Pages are released directly
1438  * to the buddy allocator, no bootmem metadata is updated because it is gone.
1439  */
1440 void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
1441 {
1442         u64 cursor, end;
1443 
1444         memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1445                      __func__, (u64)base, (u64)base + size - 1,
1446                      (void *)_RET_IP_);
1447         kmemleak_free_part_phys(base, size);
1448         cursor = PFN_UP(base);
1449         end = PFN_DOWN(base + size);
1450 
1451         for (; cursor < end; cursor++) {
1452                 __free_pages_bootmem(pfn_to_page(cursor), cursor, 0);
1453                 totalram_pages++;
1454         }
1455 }
1456 
1457 /*
1458  * Remaining API functions
1459  */
1460 
1461 phys_addr_t __init_memblock memblock_phys_mem_size(void)
1462 {
1463         return memblock.memory.total_size;
1464 }
1465 
1466 phys_addr_t __init_memblock memblock_reserved_size(void)
1467 {
1468         return memblock.reserved.total_size;
1469 }
1470 
1471 phys_addr_t __init memblock_mem_size(unsigned long limit_pfn)
1472 {
1473         unsigned long pages = 0;
1474         struct memblock_region *r;
1475         unsigned long start_pfn, end_pfn;
1476 
1477         for_each_memblock(memory, r) {
1478                 start_pfn = memblock_region_memory_base_pfn(r);
1479                 end_pfn = memblock_region_memory_end_pfn(r);
1480                 start_pfn = min_t(unsigned long, start_pfn, limit_pfn);
1481                 end_pfn = min_t(unsigned long, end_pfn, limit_pfn);
1482                 pages += end_pfn - start_pfn;
1483         }
1484 
1485         return PFN_PHYS(pages);
1486 }
1487 
1488 /* lowest address */
1489 phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1490 {
1491         return memblock.memory.regions[0].base;
1492 }
1493 
1494 phys_addr_t __init_memblock memblock_end_of_DRAM(void)
1495 {
1496         int idx = memblock.memory.cnt - 1;
1497 
1498         return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
1499 }
1500 
1501 static phys_addr_t __init_memblock __find_max_addr(phys_addr_t limit)
1502 {
1503         phys_addr_t max_addr = (phys_addr_t)ULLONG_MAX;
1504         struct memblock_region *r;
1505 
1506         /*
1507          * translate the memory @limit size into the max address within one of
1508          * the memory memblock regions, if the @limit exceeds the total size
1509          * of those regions, max_addr will keep original value ULLONG_MAX
1510          */
1511         for_each_memblock(memory, r) {
1512                 if (limit <= r->size) {
1513                         max_addr = r->base + limit;
1514                         break;
1515                 }
1516                 limit -= r->size;
1517         }
1518 
1519         return max_addr;
1520 }
1521 
1522 void __init memblock_enforce_memory_limit(phys_addr_t limit)
1523 {
1524         phys_addr_t max_addr = (phys_addr_t)ULLONG_MAX;
1525 
1526         if (!limit)
1527                 return;
1528 
1529         max_addr = __find_max_addr(limit);
1530 
1531         /* @limit exceeds the total size of the memory, do nothing */
1532         if (max_addr == (phys_addr_t)ULLONG_MAX)
1533                 return;
1534 
1535         /* truncate both memory and reserved regions */
1536         memblock_remove_range(&memblock.memory, max_addr,
1537                               (phys_addr_t)ULLONG_MAX);
1538         memblock_remove_range(&memblock.reserved, max_addr,
1539                               (phys_addr_t)ULLONG_MAX);
1540 }
1541 
1542 void __init memblock_cap_memory_range(phys_addr_t base, phys_addr_t size)
1543 {
1544         int start_rgn, end_rgn;
1545         int i, ret;
1546 
1547         if (!size)
1548                 return;
1549 
1550         ret = memblock_isolate_range(&memblock.memory, base, size,
1551                                                 &start_rgn, &end_rgn);
1552         if (ret)
1553                 return;
1554 
1555         /* remove all the MAP regions */
1556         for (i = memblock.memory.cnt - 1; i >= end_rgn; i--)
1557                 if (!memblock_is_nomap(&memblock.memory.regions[i]))
1558                         memblock_remove_region(&memblock.memory, i);
1559 
1560         for (i = start_rgn - 1; i >= 0; i--)
1561                 if (!memblock_is_nomap(&memblock.memory.regions[i]))
1562                         memblock_remove_region(&memblock.memory, i);
1563 
1564         /* truncate the reserved regions */
1565         memblock_remove_range(&memblock.reserved, 0, base);
1566         memblock_remove_range(&memblock.reserved,
1567                         base + size, (phys_addr_t)ULLONG_MAX);
1568 }
1569 
1570 void __init memblock_mem_limit_remove_map(phys_addr_t limit)
1571 {
1572         phys_addr_t max_addr;
1573 
1574         if (!limit)
1575                 return;
1576 
1577         max_addr = __find_max_addr(limit);
1578 
1579         /* @limit exceeds the total size of the memory, do nothing */
1580         if (max_addr == (phys_addr_t)ULLONG_MAX)
1581                 return;
1582 
1583         memblock_cap_memory_range(0, max_addr);
1584 }
1585 
1586 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
1587 {
1588         unsigned int left = 0, right = type->cnt;
1589 
1590         do {
1591                 unsigned int mid = (right + left) / 2;
1592 
1593                 if (addr < type->regions[mid].base)
1594                         right = mid;
1595                 else if (addr >= (type->regions[mid].base +
1596                                   type->regions[mid].size))
1597                         left = mid + 1;
1598                 else
1599                         return mid;
1600         } while (left < right);
1601         return -1;
1602 }
1603 
1604 bool __init memblock_is_reserved(phys_addr_t addr)
1605 {
1606         return memblock_search(&memblock.reserved, addr) != -1;
1607 }
1608 
1609 bool __init_memblock memblock_is_memory(phys_addr_t addr)
1610 {
1611         return memblock_search(&memblock.memory, addr) != -1;
1612 }
1613 
1614 int __init_memblock memblock_is_map_memory(phys_addr_t addr)
1615 {
1616         int i = memblock_search(&memblock.memory, addr);
1617 
1618         if (i == -1)
1619                 return false;
1620         return !memblock_is_nomap(&memblock.memory.regions[i]);
1621 }
1622 
1623 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1624 int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1625                          unsigned long *start_pfn, unsigned long *end_pfn)
1626 {
1627         struct memblock_type *type = &memblock.memory;
1628         int mid = memblock_search(type, PFN_PHYS(pfn));
1629 
1630         if (mid == -1)
1631                 return -1;
1632 
1633         *start_pfn = PFN_DOWN(type->regions[mid].base);
1634         *end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
1635 
1636         return type->regions[mid].nid;
1637 }
1638 #endif
1639 
1640 /**
1641  * memblock_is_region_memory - check if a region is a subset of memory
1642  * @base: base of region to check
1643  * @size: size of region to check
1644  *
1645  * Check if the region [@base, @base+@size) is a subset of a memory block.
1646  *
1647  * RETURNS:
1648  * 0 if false, non-zero if true
1649  */
1650 int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
1651 {
1652         int idx = memblock_search(&memblock.memory, base);
1653         phys_addr_t end = base + memblock_cap_size(base, &size);
1654 
1655         if (idx == -1)
1656                 return 0;
1657         return (memblock.memory.regions[idx].base +
1658                  memblock.memory.regions[idx].size) >= end;
1659 }
1660 
1661 /**
1662  * memblock_is_region_reserved - check if a region intersects reserved memory
1663  * @base: base of region to check
1664  * @size: size of region to check
1665  *
1666  * Check if the region [@base, @base+@size) intersects a reserved memory block.
1667  *
1668  * RETURNS:
1669  * True if they intersect, false if not.
1670  */
1671 bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
1672 {
1673         memblock_cap_size(base, &size);
1674         return memblock_overlaps_region(&memblock.reserved, base, size);
1675 }
1676 
1677 void __init_memblock memblock_trim_memory(phys_addr_t align)
1678 {
1679         phys_addr_t start, end, orig_start, orig_end;
1680         struct memblock_region *r;
1681 
1682         for_each_memblock(memory, r) {
1683                 orig_start = r->base;
1684                 orig_end = r->base + r->size;
1685                 start = round_up(orig_start, align);
1686                 end = round_down(orig_end, align);
1687 
1688                 if (start == orig_start && end == orig_end)
1689                         continue;
1690 
1691                 if (start < end) {
1692                         r->base = start;
1693                         r->size = end - start;
1694                 } else {
1695                         memblock_remove_region(&memblock.memory,
1696                                                r - memblock.memory.regions);
1697                         r--;
1698                 }
1699         }
1700 }
1701 
1702 void __init_memblock memblock_set_current_limit(phys_addr_t limit)
1703 {
1704         memblock.current_limit = limit;
1705 }
1706 
1707 phys_addr_t __init_memblock memblock_get_current_limit(void)
1708 {
1709         return memblock.current_limit;
1710 }
1711 
1712 static void __init_memblock memblock_dump(struct memblock_type *type)
1713 {
1714         phys_addr_t base, end, size;
1715         unsigned long flags;
1716         int idx;
1717         struct memblock_region *rgn;
1718 
1719         pr_info(" %s.cnt  = 0x%lx\n", type->name, type->cnt);
1720 
1721         for_each_memblock_type(type, rgn) {
1722                 char nid_buf[32] = "";
1723 
1724                 base = rgn->base;
1725                 size = rgn->size;
1726                 end = base + size - 1;
1727                 flags = rgn->flags;
1728 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1729                 if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1730                         snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1731                                  memblock_get_region_node(rgn));
1732 #endif
1733                 pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#lx\n",
1734                         type->name, idx, &base, &end, &size, nid_buf, flags);
1735         }
1736 }
1737 
1738 extern unsigned long __init_memblock
1739 memblock_reserved_memory_within(phys_addr_t start_addr, phys_addr_t end_addr)
1740 {
1741         struct memblock_region *rgn;
1742         unsigned long size = 0;
1743         int idx;
1744 
1745         for_each_memblock_type((&memblock.reserved), rgn) {
1746                 phys_addr_t start, end;
1747 
1748                 if (rgn->base + rgn->size < start_addr)
1749                         continue;
1750                 if (rgn->base > end_addr)
1751                         continue;
1752 
1753                 start = rgn->base;
1754                 end = start + rgn->size;
1755                 size += end - start;
1756         }
1757 
1758         return size;
1759 }
1760 
1761 void __init_memblock __memblock_dump_all(void)
1762 {
1763         pr_info("MEMBLOCK configuration:\n");
1764         pr_info(" memory size = %pa reserved size = %pa\n",
1765                 &memblock.memory.total_size,
1766                 &memblock.reserved.total_size);
1767 
1768         memblock_dump(&memblock.memory);
1769         memblock_dump(&memblock.reserved);
1770 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1771         memblock_dump(&memblock.physmem);
1772 #endif
1773 }
1774 
1775 void __init memblock_allow_resize(void)
1776 {
1777         memblock_can_resize = 1;
1778 }
1779 
1780 static int __init early_memblock(char *p)
1781 {
1782         if (p && strstr(p, "debug"))
1783                 memblock_debug = 1;
1784         return 0;
1785 }
1786 early_param("memblock", early_memblock);
1787 
1788 #if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK)
1789 
1790 static int memblock_debug_show(struct seq_file *m, void *private)
1791 {
1792         struct memblock_type *type = m->private;
1793         struct memblock_region *reg;
1794         int i;
1795         phys_addr_t end;
1796 
1797         for (i = 0; i < type->cnt; i++) {
1798                 reg = &type->regions[i];
1799                 end = reg->base + reg->size - 1;
1800 
1801                 seq_printf(m, "%4d: ", i);
1802                 seq_printf(m, "%pa..%pa\n", &reg->base, &end);
1803         }
1804         return 0;
1805 }
1806 
1807 static int memblock_debug_open(struct inode *inode, struct file *file)
1808 {
1809         return single_open(file, memblock_debug_show, inode->i_private);
1810 }
1811 
1812 static const struct file_operations memblock_debug_fops = {
1813         .open = memblock_debug_open,
1814         .read = seq_read,
1815         .llseek = seq_lseek,
1816         .release = single_release,
1817 };
1818 
1819 static int __init memblock_init_debugfs(void)
1820 {
1821         struct dentry *root = debugfs_create_dir("memblock", NULL);
1822         if (!root)
1823                 return -ENXIO;
1824         debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops);
1825         debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops);
1826 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1827         debugfs_create_file("physmem", S_IRUGO, root, &memblock.physmem, &memblock_debug_fops);
1828 #endif
1829 
1830         return 0;
1831 }
1832 __initcall(memblock_init_debugfs);
1833 
1834 #endif /* CONFIG_DEBUG_FS */
1835 

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

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | 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.

osdn.jp