2 * bootmem - A boot-time physical memory allocator and configurator
4 * Copyright (C) 1999 Ingo Molnar
5 * 1999 Kanoj Sarcar, SGI
8 * Access to this subsystem has to be serialized externally (which is true
9 * for the boot process anyway).
11 #include <linux/init.h>
12 #include <linux/pfn.h>
13 #include <linux/bootmem.h>
14 #include <linux/module.h>
15 #include <linux/kmemleak.h>
19 #include <asm/processor.h>
23 unsigned long max_low_pfn;
24 unsigned long min_low_pfn;
25 unsigned long max_pfn;
27 #ifdef CONFIG_CRASH_DUMP
29 * If we have booted due to a crash, max_pfn will be a very low value. We need
30 * to know the amount of memory that the previous kernel used.
32 unsigned long saved_max_pfn;
35 bootmem_data_t bootmem_node_data[MAX_NUMNODES] __initdata;
37 static struct list_head bdata_list __initdata = LIST_HEAD_INIT(bdata_list);
39 static int bootmem_debug;
41 static int __init bootmem_debug_setup(char *buf)
46 early_param("bootmem_debug", bootmem_debug_setup);
48 #define bdebug(fmt, args...) ({ \
49 if (unlikely(bootmem_debug)) \
55 static unsigned long __init bootmap_bytes(unsigned long pages)
57 unsigned long bytes = (pages + 7) / 8;
59 return ALIGN(bytes, sizeof(long));
63 * bootmem_bootmap_pages - calculate bitmap size in pages
64 * @pages: number of pages the bitmap has to represent
66 unsigned long __init bootmem_bootmap_pages(unsigned long pages)
68 unsigned long bytes = bootmap_bytes(pages);
70 return PAGE_ALIGN(bytes) >> PAGE_SHIFT;
76 static void __init link_bootmem(bootmem_data_t *bdata)
78 struct list_head *iter;
80 list_for_each(iter, &bdata_list) {
83 ent = list_entry(iter, bootmem_data_t, list);
84 if (bdata->node_min_pfn < ent->node_min_pfn)
87 list_add_tail(&bdata->list, iter);
91 * Called once to set up the allocator itself.
93 static unsigned long __init init_bootmem_core(bootmem_data_t *bdata,
94 unsigned long mapstart, unsigned long start, unsigned long end)
96 unsigned long mapsize;
98 mminit_validate_memmodel_limits(&start, &end);
99 bdata->node_bootmem_map = phys_to_virt(PFN_PHYS(mapstart));
100 bdata->node_min_pfn = start;
101 bdata->node_low_pfn = end;
105 * Initially all pages are reserved - setup_arch() has to
106 * register free RAM areas explicitly.
108 mapsize = bootmap_bytes(end - start);
109 memset(bdata->node_bootmem_map, 0xff, mapsize);
111 bdebug("nid=%td start=%lx map=%lx end=%lx mapsize=%lx\n",
112 bdata - bootmem_node_data, start, mapstart, end, mapsize);
118 * init_bootmem_node - register a node as boot memory
119 * @pgdat: node to register
120 * @freepfn: pfn where the bitmap for this node is to be placed
121 * @startpfn: first pfn on the node
122 * @endpfn: first pfn after the node
124 * Returns the number of bytes needed to hold the bitmap for this node.
126 unsigned long __init init_bootmem_node(pg_data_t *pgdat, unsigned long freepfn,
127 unsigned long startpfn, unsigned long endpfn)
129 return init_bootmem_core(pgdat->bdata, freepfn, startpfn, endpfn);
133 * init_bootmem - register boot memory
134 * @start: pfn where the bitmap is to be placed
135 * @pages: number of available physical pages
137 * Returns the number of bytes needed to hold the bitmap.
139 unsigned long __init init_bootmem(unsigned long start, unsigned long pages)
143 return init_bootmem_core(NODE_DATA(0)->bdata, start, 0, pages);
147 * free_bootmem_late - free bootmem pages directly to page allocator
148 * @addr: starting address of the range
149 * @size: size of the range in bytes
151 * This is only useful when the bootmem allocator has already been torn
152 * down, but we are still initializing the system. Pages are given directly
153 * to the page allocator, no bootmem metadata is updated because it is gone.
155 void __init free_bootmem_late(unsigned long addr, unsigned long size)
157 unsigned long cursor, end;
159 kmemleak_free_part(__va(addr), size);
161 cursor = PFN_UP(addr);
162 end = PFN_DOWN(addr + size);
164 for (; cursor < end; cursor++) {
165 __free_pages_bootmem(pfn_to_page(cursor), 0);
170 static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata)
174 unsigned long start, end, pages, count = 0;
176 if (!bdata->node_bootmem_map)
179 start = bdata->node_min_pfn;
180 end = bdata->node_low_pfn;
183 * If the start is aligned to the machines wordsize, we might
184 * be able to free pages in bulks of that order.
186 aligned = !(start & (BITS_PER_LONG - 1));
188 bdebug("nid=%td start=%lx end=%lx aligned=%d\n",
189 bdata - bootmem_node_data, start, end, aligned);
191 while (start < end) {
192 unsigned long *map, idx, vec;
194 map = bdata->node_bootmem_map;
195 idx = start - bdata->node_min_pfn;
196 vec = ~map[idx / BITS_PER_LONG];
198 if (aligned && vec == ~0UL && start + BITS_PER_LONG < end) {
199 int order = ilog2(BITS_PER_LONG);
201 __free_pages_bootmem(pfn_to_page(start), order);
202 count += BITS_PER_LONG;
204 unsigned long off = 0;
206 while (vec && off < BITS_PER_LONG) {
208 page = pfn_to_page(start + off);
209 __free_pages_bootmem(page, 0);
216 start += BITS_PER_LONG;
219 page = virt_to_page(bdata->node_bootmem_map);
220 pages = bdata->node_low_pfn - bdata->node_min_pfn;
221 pages = bootmem_bootmap_pages(pages);
224 __free_pages_bootmem(page++, 0);
226 bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count);
232 * free_all_bootmem_node - release a node's free pages to the buddy allocator
233 * @pgdat: node to be released
235 * Returns the number of pages actually released.
237 unsigned long __init free_all_bootmem_node(pg_data_t *pgdat)
239 register_page_bootmem_info_node(pgdat);
240 return free_all_bootmem_core(pgdat->bdata);
244 * free_all_bootmem - release free pages to the buddy allocator
246 * Returns the number of pages actually released.
248 unsigned long __init free_all_bootmem(void)
250 return free_all_bootmem_core(NODE_DATA(0)->bdata);
253 static void __init __free(bootmem_data_t *bdata,
254 unsigned long sidx, unsigned long eidx)
258 bdebug("nid=%td start=%lx end=%lx\n", bdata - bootmem_node_data,
259 sidx + bdata->node_min_pfn,
260 eidx + bdata->node_min_pfn);
262 if (bdata->hint_idx > sidx)
263 bdata->hint_idx = sidx;
265 for (idx = sidx; idx < eidx; idx++)
266 if (!test_and_clear_bit(idx, bdata->node_bootmem_map))
270 static int __init __reserve(bootmem_data_t *bdata, unsigned long sidx,
271 unsigned long eidx, int flags)
274 int exclusive = flags & BOOTMEM_EXCLUSIVE;
276 bdebug("nid=%td start=%lx end=%lx flags=%x\n",
277 bdata - bootmem_node_data,
278 sidx + bdata->node_min_pfn,
279 eidx + bdata->node_min_pfn,
282 for (idx = sidx; idx < eidx; idx++)
283 if (test_and_set_bit(idx, bdata->node_bootmem_map)) {
285 __free(bdata, sidx, idx);
288 bdebug("silent double reserve of PFN %lx\n",
289 idx + bdata->node_min_pfn);
294 static int __init mark_bootmem_node(bootmem_data_t *bdata,
295 unsigned long start, unsigned long end,
296 int reserve, int flags)
298 unsigned long sidx, eidx;
300 bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n",
301 bdata - bootmem_node_data, start, end, reserve, flags);
303 BUG_ON(start < bdata->node_min_pfn);
304 BUG_ON(end > bdata->node_low_pfn);
306 sidx = start - bdata->node_min_pfn;
307 eidx = end - bdata->node_min_pfn;
310 return __reserve(bdata, sidx, eidx, flags);
312 __free(bdata, sidx, eidx);
316 static int __init mark_bootmem(unsigned long start, unsigned long end,
317 int reserve, int flags)
320 bootmem_data_t *bdata;
323 list_for_each_entry(bdata, &bdata_list, list) {
327 if (pos < bdata->node_min_pfn ||
328 pos >= bdata->node_low_pfn) {
329 BUG_ON(pos != start);
333 max = min(bdata->node_low_pfn, end);
335 err = mark_bootmem_node(bdata, pos, max, reserve, flags);
336 if (reserve && err) {
337 mark_bootmem(start, pos, 0, 0);
343 pos = bdata->node_low_pfn;
349 * free_bootmem_node - mark a page range as usable
350 * @pgdat: node the range resides on
351 * @physaddr: starting address of the range
352 * @size: size of the range in bytes
354 * Partial pages will be considered reserved and left as they are.
356 * The range must reside completely on the specified node.
358 void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
361 unsigned long start, end;
363 kmemleak_free_part(__va(physaddr), size);
365 start = PFN_UP(physaddr);
366 end = PFN_DOWN(physaddr + size);
368 mark_bootmem_node(pgdat->bdata, start, end, 0, 0);
372 * free_bootmem - mark a page range as usable
373 * @addr: starting address of the range
374 * @size: size of the range in bytes
376 * Partial pages will be considered reserved and left as they are.
378 * The range must be contiguous but may span node boundaries.
380 void __init free_bootmem(unsigned long addr, unsigned long size)
382 unsigned long start, end;
384 kmemleak_free_part(__va(addr), size);
386 start = PFN_UP(addr);
387 end = PFN_DOWN(addr + size);
389 mark_bootmem(start, end, 0, 0);
393 * reserve_bootmem_node - mark a page range as reserved
394 * @pgdat: node the range resides on
395 * @physaddr: starting address of the range
396 * @size: size of the range in bytes
397 * @flags: reservation flags (see linux/bootmem.h)
399 * Partial pages will be reserved.
401 * The range must reside completely on the specified node.
403 int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
404 unsigned long size, int flags)
406 unsigned long start, end;
408 start = PFN_DOWN(physaddr);
409 end = PFN_UP(physaddr + size);
411 return mark_bootmem_node(pgdat->bdata, start, end, 1, flags);
415 * reserve_bootmem - mark a page range as usable
416 * @addr: starting address of the range
417 * @size: size of the range in bytes
418 * @flags: reservation flags (see linux/bootmem.h)
420 * Partial pages will be reserved.
422 * The range must be contiguous but may span node boundaries.
424 int __init reserve_bootmem(unsigned long addr, unsigned long size,
427 unsigned long start, end;
429 start = PFN_DOWN(addr);
430 end = PFN_UP(addr + size);
432 return mark_bootmem(start, end, 1, flags);
435 static unsigned long __init align_idx(struct bootmem_data *bdata,
436 unsigned long idx, unsigned long step)
438 unsigned long base = bdata->node_min_pfn;
441 * Align the index with respect to the node start so that the
442 * combination of both satisfies the requested alignment.
445 return ALIGN(base + idx, step) - base;
448 static unsigned long __init align_off(struct bootmem_data *bdata,
449 unsigned long off, unsigned long align)
451 unsigned long base = PFN_PHYS(bdata->node_min_pfn);
453 /* Same as align_idx for byte offsets */
455 return ALIGN(base + off, align) - base;
458 static void * __init alloc_bootmem_core(struct bootmem_data *bdata,
459 unsigned long size, unsigned long align,
460 unsigned long goal, unsigned long limit)
462 unsigned long fallback = 0;
463 unsigned long min, max, start, sidx, midx, step;
465 bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n",
466 bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT,
470 BUG_ON(align & (align - 1));
471 BUG_ON(limit && goal + size > limit);
473 if (!bdata->node_bootmem_map)
476 min = bdata->node_min_pfn;
477 max = bdata->node_low_pfn;
480 limit >>= PAGE_SHIFT;
482 if (limit && max > limit)
487 step = max(align >> PAGE_SHIFT, 1UL);
489 if (goal && min < goal && goal < max)
490 start = ALIGN(goal, step);
492 start = ALIGN(min, step);
494 sidx = start - bdata->node_min_pfn;
495 midx = max - bdata->node_min_pfn;
497 if (bdata->hint_idx > sidx) {
499 * Handle the valid case of sidx being zero and still
500 * catch the fallback below.
503 sidx = align_idx(bdata, bdata->hint_idx, step);
509 unsigned long eidx, i, start_off, end_off;
511 sidx = find_next_zero_bit(bdata->node_bootmem_map, midx, sidx);
512 sidx = align_idx(bdata, sidx, step);
513 eidx = sidx + PFN_UP(size);
515 if (sidx >= midx || eidx > midx)
518 for (i = sidx; i < eidx; i++)
519 if (test_bit(i, bdata->node_bootmem_map)) {
520 sidx = align_idx(bdata, i, step);
526 if (bdata->last_end_off & (PAGE_SIZE - 1) &&
527 PFN_DOWN(bdata->last_end_off) + 1 == sidx)
528 start_off = align_off(bdata, bdata->last_end_off, align);
530 start_off = PFN_PHYS(sidx);
532 merge = PFN_DOWN(start_off) < sidx;
533 end_off = start_off + size;
535 bdata->last_end_off = end_off;
536 bdata->hint_idx = PFN_UP(end_off);
539 * Reserve the area now:
541 if (__reserve(bdata, PFN_DOWN(start_off) + merge,
542 PFN_UP(end_off), BOOTMEM_EXCLUSIVE))
545 region = phys_to_virt(PFN_PHYS(bdata->node_min_pfn) +
547 memset(region, 0, size);
549 * The min_count is set to 0 so that bootmem allocated blocks
550 * are never reported as leaks.
552 kmemleak_alloc(region, size, 0, 0);
557 sidx = align_idx(bdata, fallback - 1, step);
565 static void * __init alloc_arch_preferred_bootmem(bootmem_data_t *bdata,
566 unsigned long size, unsigned long align,
567 unsigned long goal, unsigned long limit)
569 if (WARN_ON_ONCE(slab_is_available()))
570 return kzalloc(size, GFP_NOWAIT);
572 #ifdef CONFIG_HAVE_ARCH_BOOTMEM
574 bootmem_data_t *p_bdata;
576 p_bdata = bootmem_arch_preferred_node(bdata, size, align,
579 return alloc_bootmem_core(p_bdata, size, align,
586 static void * __init ___alloc_bootmem_nopanic(unsigned long size,
591 bootmem_data_t *bdata;
595 region = alloc_arch_preferred_bootmem(NULL, size, align, goal, limit);
599 list_for_each_entry(bdata, &bdata_list, list) {
600 if (goal && bdata->node_low_pfn <= PFN_DOWN(goal))
602 if (limit && bdata->node_min_pfn >= PFN_DOWN(limit))
605 region = alloc_bootmem_core(bdata, size, align, goal, limit);
619 * __alloc_bootmem_nopanic - allocate boot memory without panicking
620 * @size: size of the request in bytes
621 * @align: alignment of the region
622 * @goal: preferred starting address of the region
624 * The goal is dropped if it can not be satisfied and the allocation will
625 * fall back to memory below @goal.
627 * Allocation may happen on any node in the system.
629 * Returns NULL on failure.
631 void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
634 return ___alloc_bootmem_nopanic(size, align, goal, 0);
637 static void * __init ___alloc_bootmem(unsigned long size, unsigned long align,
638 unsigned long goal, unsigned long limit)
640 void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit);
645 * Whoops, we cannot satisfy the allocation request.
647 printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
648 panic("Out of memory");
653 * __alloc_bootmem - allocate boot memory
654 * @size: size of the request in bytes
655 * @align: alignment of the region
656 * @goal: preferred starting address of the region
658 * The goal is dropped if it can not be satisfied and the allocation will
659 * fall back to memory below @goal.
661 * Allocation may happen on any node in the system.
663 * The function panics if the request can not be satisfied.
665 void * __init __alloc_bootmem(unsigned long size, unsigned long align,
668 return ___alloc_bootmem(size, align, goal, 0);
671 static void * __init ___alloc_bootmem_node(bootmem_data_t *bdata,
672 unsigned long size, unsigned long align,
673 unsigned long goal, unsigned long limit)
677 ptr = alloc_arch_preferred_bootmem(bdata, size, align, goal, limit);
681 ptr = alloc_bootmem_core(bdata, size, align, goal, limit);
685 return ___alloc_bootmem(size, align, goal, limit);
689 * __alloc_bootmem_node - allocate boot memory from a specific node
690 * @pgdat: node to allocate from
691 * @size: size of the request in bytes
692 * @align: alignment of the region
693 * @goal: preferred starting address of the region
695 * The goal is dropped if it can not be satisfied and the allocation will
696 * fall back to memory below @goal.
698 * Allocation may fall back to any node in the system if the specified node
699 * can not hold the requested memory.
701 * The function panics if the request can not be satisfied.
703 void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
704 unsigned long align, unsigned long goal)
706 if (WARN_ON_ONCE(slab_is_available()))
707 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
709 return ___alloc_bootmem_node(pgdat->bdata, size, align, goal, 0);
712 #ifdef CONFIG_SPARSEMEM
714 * alloc_bootmem_section - allocate boot memory from a specific section
715 * @size: size of the request in bytes
716 * @section_nr: sparse map section to allocate from
718 * Return NULL on failure.
720 void * __init alloc_bootmem_section(unsigned long size,
721 unsigned long section_nr)
723 bootmem_data_t *bdata;
724 unsigned long pfn, goal, limit;
726 pfn = section_nr_to_pfn(section_nr);
727 goal = pfn << PAGE_SHIFT;
728 limit = section_nr_to_pfn(section_nr + 1) << PAGE_SHIFT;
729 bdata = &bootmem_node_data[early_pfn_to_nid(pfn)];
731 return alloc_bootmem_core(bdata, size, SMP_CACHE_BYTES, goal, limit);
735 void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
736 unsigned long align, unsigned long goal)
740 if (WARN_ON_ONCE(slab_is_available()))
741 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
743 ptr = alloc_arch_preferred_bootmem(pgdat->bdata, size, align, goal, 0);
747 ptr = alloc_bootmem_core(pgdat->bdata, size, align, goal, 0);
751 return __alloc_bootmem_nopanic(size, align, goal);
754 #ifndef ARCH_LOW_ADDRESS_LIMIT
755 #define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL
759 * __alloc_bootmem_low - allocate low boot memory
760 * @size: size of the request in bytes
761 * @align: alignment of the region
762 * @goal: preferred starting address of the region
764 * The goal is dropped if it can not be satisfied and the allocation will
765 * fall back to memory below @goal.
767 * Allocation may happen on any node in the system.
769 * The function panics if the request can not be satisfied.
771 void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
774 return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT);
778 * __alloc_bootmem_low_node - allocate low boot memory from a specific node
779 * @pgdat: node to allocate from
780 * @size: size of the request in bytes
781 * @align: alignment of the region
782 * @goal: preferred starting address of the region
784 * The goal is dropped if it can not be satisfied and the allocation will
785 * fall back to memory below @goal.
787 * Allocation may fall back to any node in the system if the specified node
788 * can not hold the requested memory.
790 * The function panics if the request can not be satisfied.
792 void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
793 unsigned long align, unsigned long goal)
795 if (WARN_ON_ONCE(slab_is_available()))
796 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
798 return ___alloc_bootmem_node(pgdat->bdata, size, align,
799 goal, ARCH_LOW_ADDRESS_LIMIT);