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>
18 #include <asm/processor.h>
22 unsigned long max_low_pfn;
23 unsigned long min_low_pfn;
24 unsigned long max_pfn;
26 #ifdef CONFIG_CRASH_DUMP
28 * If we have booted due to a crash, max_pfn will be a very low value. We need
29 * to know the amount of memory that the previous kernel used.
31 unsigned long saved_max_pfn;
34 bootmem_data_t bootmem_node_data[MAX_NUMNODES] __initdata;
36 static struct list_head bdata_list __initdata = LIST_HEAD_INIT(bdata_list);
38 static int bootmem_debug;
40 static int __init bootmem_debug_setup(char *buf)
45 early_param("bootmem_debug", bootmem_debug_setup);
47 #define bdebug(fmt, args...) ({ \
48 if (unlikely(bootmem_debug)) \
51 __FUNCTION__, ## args); \
54 static unsigned long __init bootmap_bytes(unsigned long pages)
56 unsigned long bytes = (pages + 7) / 8;
58 return ALIGN(bytes, sizeof(long));
62 * bootmem_bootmap_pages - calculate bitmap size in pages
63 * @pages: number of pages the bitmap has to represent
65 unsigned long __init bootmem_bootmap_pages(unsigned long pages)
67 unsigned long bytes = bootmap_bytes(pages);
69 return PAGE_ALIGN(bytes) >> PAGE_SHIFT;
75 static void __init link_bootmem(bootmem_data_t *bdata)
77 struct list_head *iter;
79 list_for_each(iter, &bdata_list) {
82 ent = list_entry(iter, bootmem_data_t, list);
83 if (bdata->node_boot_start < ent->node_boot_start)
86 list_add_tail(&bdata->list, iter);
90 * Called once to set up the allocator itself.
92 static unsigned long __init init_bootmem_core(bootmem_data_t *bdata,
93 unsigned long mapstart, unsigned long start, unsigned long end)
95 unsigned long mapsize;
97 mminit_validate_memmodel_limits(&start, &end);
98 bdata->node_bootmem_map = phys_to_virt(PFN_PHYS(mapstart));
99 bdata->node_boot_start = PFN_PHYS(start);
100 bdata->node_low_pfn = end;
104 * Initially all pages are reserved - setup_arch() has to
105 * register free RAM areas explicitly.
107 mapsize = bootmap_bytes(end - start);
108 memset(bdata->node_bootmem_map, 0xff, mapsize);
110 bdebug("nid=%td start=%lx map=%lx end=%lx mapsize=%lx\n",
111 bdata - bootmem_node_data, start, mapstart, end, mapsize);
117 * init_bootmem_node - register a node as boot memory
118 * @pgdat: node to register
119 * @freepfn: pfn where the bitmap for this node is to be placed
120 * @startpfn: first pfn on the node
121 * @endpfn: first pfn after the node
123 * Returns the number of bytes needed to hold the bitmap for this node.
125 unsigned long __init init_bootmem_node(pg_data_t *pgdat, unsigned long freepfn,
126 unsigned long startpfn, unsigned long endpfn)
128 return init_bootmem_core(pgdat->bdata, freepfn, startpfn, endpfn);
132 * init_bootmem - register boot memory
133 * @start: pfn where the bitmap is to be placed
134 * @pages: number of available physical pages
136 * Returns the number of bytes needed to hold the bitmap.
138 unsigned long __init init_bootmem(unsigned long start, unsigned long pages)
142 return init_bootmem_core(NODE_DATA(0)->bdata, start, 0, pages);
145 static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata)
149 unsigned long start, end, pages, count = 0;
151 if (!bdata->node_bootmem_map)
154 start = PFN_DOWN(bdata->node_boot_start);
155 end = bdata->node_low_pfn;
158 * If the start is aligned to the machines wordsize, we might
159 * be able to free pages in bulks of that order.
161 aligned = !(start & (BITS_PER_LONG - 1));
163 bdebug("nid=%td start=%lx end=%lx aligned=%d\n",
164 bdata - bootmem_node_data, start, end, aligned);
166 while (start < end) {
167 unsigned long *map, idx, vec;
169 map = bdata->node_bootmem_map;
170 idx = start - PFN_DOWN(bdata->node_boot_start);
171 vec = ~map[idx / BITS_PER_LONG];
173 if (aligned && vec == ~0UL && start + BITS_PER_LONG < end) {
174 int order = ilog2(BITS_PER_LONG);
176 __free_pages_bootmem(pfn_to_page(start), order);
177 count += BITS_PER_LONG;
179 unsigned long off = 0;
181 while (vec && off < BITS_PER_LONG) {
183 page = pfn_to_page(start + off);
184 __free_pages_bootmem(page, 0);
191 start += BITS_PER_LONG;
194 page = virt_to_page(bdata->node_bootmem_map);
195 pages = bdata->node_low_pfn - PFN_DOWN(bdata->node_boot_start);
196 pages = bootmem_bootmap_pages(pages);
199 __free_pages_bootmem(page++, 0);
201 bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count);
207 * free_all_bootmem_node - release a node's free pages to the buddy allocator
208 * @pgdat: node to be released
210 * Returns the number of pages actually released.
212 unsigned long __init free_all_bootmem_node(pg_data_t *pgdat)
214 register_page_bootmem_info_node(pgdat);
215 return free_all_bootmem_core(pgdat->bdata);
219 * free_all_bootmem - release free pages to the buddy allocator
221 * Returns the number of pages actually released.
223 unsigned long __init free_all_bootmem(void)
225 return free_all_bootmem_core(NODE_DATA(0)->bdata);
228 static void __init free_bootmem_core(bootmem_data_t *bdata, unsigned long addr,
231 unsigned long sidx, eidx;
237 if (addr + size < bdata->node_boot_start ||
238 PFN_DOWN(addr) > bdata->node_low_pfn)
241 * round down end of usable mem, partially free pages are
242 * considered reserved.
245 if (addr >= bdata->node_boot_start && addr < bdata->last_success)
246 bdata->last_success = addr;
249 * Round up to index to the range.
251 if (PFN_UP(addr) > PFN_DOWN(bdata->node_boot_start))
252 sidx = PFN_UP(addr) - PFN_DOWN(bdata->node_boot_start);
256 eidx = PFN_DOWN(addr + size - bdata->node_boot_start);
257 if (eidx > bdata->node_low_pfn - PFN_DOWN(bdata->node_boot_start))
258 eidx = bdata->node_low_pfn - PFN_DOWN(bdata->node_boot_start);
260 bdebug("nid=%td start=%lx end=%lx\n", bdata - bootmem_node_data,
261 sidx + PFN_DOWN(bdata->node_boot_start),
262 eidx + PFN_DOWN(bdata->node_boot_start));
264 for (i = sidx; i < eidx; i++) {
265 if (unlikely(!test_and_clear_bit(i, bdata->node_bootmem_map)))
271 * free_bootmem_node - mark a page range as usable
272 * @pgdat: node the range resides on
273 * @physaddr: starting address of the range
274 * @size: size of the range in bytes
276 * Partial pages will be considered reserved and left as they are.
278 * Only physical pages that actually reside on @pgdat are marked.
280 void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
283 free_bootmem_core(pgdat->bdata, physaddr, size);
287 * free_bootmem - mark a page range as usable
288 * @addr: starting address of the range
289 * @size: size of the range in bytes
291 * Partial pages will be considered reserved and left as they are.
293 * All physical pages within the range are marked, no matter what
294 * node they reside on.
296 void __init free_bootmem(unsigned long addr, unsigned long size)
298 bootmem_data_t *bdata;
299 list_for_each_entry(bdata, &bdata_list, list)
300 free_bootmem_core(bdata, addr, size);
304 * Marks a particular physical memory range as unallocatable. Usable RAM
305 * might be used for boot-time allocations - or it might get added
306 * to the free page pool later on.
308 static int __init can_reserve_bootmem_core(bootmem_data_t *bdata,
309 unsigned long addr, unsigned long size, int flags)
311 unsigned long sidx, eidx;
316 /* out of range, don't hold other */
317 if (addr + size < bdata->node_boot_start ||
318 PFN_DOWN(addr) > bdata->node_low_pfn)
322 * Round up to index to the range.
324 if (addr > bdata->node_boot_start)
325 sidx= PFN_DOWN(addr - bdata->node_boot_start);
329 eidx = PFN_UP(addr + size - bdata->node_boot_start);
330 if (eidx > bdata->node_low_pfn - PFN_DOWN(bdata->node_boot_start))
331 eidx = bdata->node_low_pfn - PFN_DOWN(bdata->node_boot_start);
333 for (i = sidx; i < eidx; i++) {
334 if (test_bit(i, bdata->node_bootmem_map)) {
335 if (flags & BOOTMEM_EXCLUSIVE)
344 static void __init reserve_bootmem_core(bootmem_data_t *bdata,
345 unsigned long addr, unsigned long size, int flags)
347 unsigned long sidx, eidx;
353 if (addr + size < bdata->node_boot_start ||
354 PFN_DOWN(addr) > bdata->node_low_pfn)
358 * Round up to index to the range.
360 if (addr > bdata->node_boot_start)
361 sidx= PFN_DOWN(addr - bdata->node_boot_start);
365 eidx = PFN_UP(addr + size - bdata->node_boot_start);
366 if (eidx > bdata->node_low_pfn - PFN_DOWN(bdata->node_boot_start))
367 eidx = bdata->node_low_pfn - PFN_DOWN(bdata->node_boot_start);
369 bdebug("nid=%td start=%lx end=%lx flags=%x\n",
370 bdata - bootmem_node_data,
371 sidx + PFN_DOWN(bdata->node_boot_start),
372 eidx + PFN_DOWN(bdata->node_boot_start),
375 for (i = sidx; i < eidx; i++)
376 if (test_and_set_bit(i, bdata->node_bootmem_map))
377 bdebug("hm, page %lx reserved twice.\n",
378 PFN_DOWN(bdata->node_boot_start) + i);
382 * reserve_bootmem_node - mark a page range as reserved
383 * @pgdat: node the range resides on
384 * @physaddr: starting address of the range
385 * @size: size of the range in bytes
386 * @flags: reservation flags (see linux/bootmem.h)
388 * Partial pages will be reserved.
390 * Only physical pages that actually reside on @pgdat are marked.
392 int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
393 unsigned long size, int flags)
397 ret = can_reserve_bootmem_core(pgdat->bdata, physaddr, size, flags);
400 reserve_bootmem_core(pgdat->bdata, physaddr, size, flags);
404 #ifndef CONFIG_HAVE_ARCH_BOOTMEM_NODE
406 * reserve_bootmem - mark a page range as usable
407 * @addr: starting address of the range
408 * @size: size of the range in bytes
409 * @flags: reservation flags (see linux/bootmem.h)
411 * Partial pages will be reserved.
413 * All physical pages within the range are marked, no matter what
414 * node they reside on.
416 int __init reserve_bootmem(unsigned long addr, unsigned long size,
419 bootmem_data_t *bdata;
422 list_for_each_entry(bdata, &bdata_list, list) {
423 ret = can_reserve_bootmem_core(bdata, addr, size, flags);
427 list_for_each_entry(bdata, &bdata_list, list)
428 reserve_bootmem_core(bdata, addr, size, flags);
432 #endif /* !CONFIG_HAVE_ARCH_BOOTMEM_NODE */
435 * We 'merge' subsequent allocations to save space. We might 'lose'
436 * some fraction of a page if allocations cannot be satisfied due to
437 * size constraints on boxes where there is physical RAM space
438 * fragmentation - in these cases (mostly large memory boxes) this
441 * On low memory boxes we get it right in 100% of the cases.
443 * alignment has to be a power of 2 value.
445 * NOTE: This function is _not_ reentrant.
448 alloc_bootmem_core(struct bootmem_data *bdata, unsigned long size,
449 unsigned long align, unsigned long goal, unsigned long limit)
451 unsigned long areasize, preferred;
452 unsigned long i, start = 0, incr, eidx, end_pfn;
454 unsigned long node_boot_start;
455 void *node_bootmem_map;
458 printk("alloc_bootmem_core(): zero-sized request\n");
461 BUG_ON(align & (align-1));
463 /* on nodes without memory - bootmem_map is NULL */
464 if (!bdata->node_bootmem_map)
467 bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n",
468 bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT,
471 /* bdata->node_boot_start is supposed to be (12+6)bits alignment on x86_64 ? */
472 node_boot_start = bdata->node_boot_start;
473 node_bootmem_map = bdata->node_bootmem_map;
475 node_boot_start = ALIGN(bdata->node_boot_start, align);
476 if (node_boot_start > bdata->node_boot_start)
477 node_bootmem_map = (unsigned long *)bdata->node_bootmem_map +
478 PFN_DOWN(node_boot_start - bdata->node_boot_start)/BITS_PER_LONG;
481 if (limit && node_boot_start >= limit)
484 end_pfn = bdata->node_low_pfn;
485 limit = PFN_DOWN(limit);
486 if (limit && end_pfn > limit)
489 eidx = end_pfn - PFN_DOWN(node_boot_start);
492 * We try to allocate bootmem pages above 'goal'
493 * first, then we try to allocate lower pages.
496 if (goal && PFN_DOWN(goal) < end_pfn) {
497 if (goal > node_boot_start)
498 preferred = goal - node_boot_start;
500 if (bdata->last_success > node_boot_start &&
501 bdata->last_success - node_boot_start >= preferred)
502 if (!limit || (limit && limit > bdata->last_success))
503 preferred = bdata->last_success - node_boot_start;
506 preferred = PFN_DOWN(ALIGN(preferred, align));
507 areasize = (size + PAGE_SIZE-1) / PAGE_SIZE;
508 incr = align >> PAGE_SHIFT ? : 1;
511 for (i = preferred; i < eidx;) {
514 i = find_next_zero_bit(node_bootmem_map, eidx, i);
518 if (test_bit(i, node_bootmem_map)) {
522 for (j = i + 1; j < i + areasize; ++j) {
525 if (test_bit(j, node_bootmem_map))
543 bdata->last_success = PFN_PHYS(start) + node_boot_start;
544 BUG_ON(start >= eidx);
547 * Is the next page of the previous allocation-end the start
548 * of this allocation's buffer? If yes then we can 'merge'
549 * the previous partial page with this allocation.
551 if (align < PAGE_SIZE &&
552 bdata->last_offset && bdata->last_pos+1 == start) {
553 unsigned long offset, remaining_size;
554 offset = ALIGN(bdata->last_offset, align);
555 BUG_ON(offset > PAGE_SIZE);
556 remaining_size = PAGE_SIZE - offset;
557 if (size < remaining_size) {
559 /* last_pos unchanged */
560 bdata->last_offset = offset + size;
561 ret = phys_to_virt(bdata->last_pos * PAGE_SIZE +
562 offset + node_boot_start);
564 remaining_size = size - remaining_size;
565 areasize = (remaining_size + PAGE_SIZE-1) / PAGE_SIZE;
566 ret = phys_to_virt(bdata->last_pos * PAGE_SIZE +
567 offset + node_boot_start);
568 bdata->last_pos = start + areasize - 1;
569 bdata->last_offset = remaining_size;
571 bdata->last_offset &= ~PAGE_MASK;
573 bdata->last_pos = start + areasize - 1;
574 bdata->last_offset = size & ~PAGE_MASK;
575 ret = phys_to_virt(start * PAGE_SIZE + node_boot_start);
578 bdebug("nid=%td start=%lx end=%lx\n",
579 bdata - bootmem_node_data,
580 start + PFN_DOWN(bdata->node_boot_start),
581 start + areasize + PFN_DOWN(bdata->node_boot_start));
584 * Reserve the area now:
586 for (i = start; i < start + areasize; i++)
587 if (unlikely(test_and_set_bit(i, node_bootmem_map)))
589 memset(ret, 0, size);
594 * __alloc_bootmem_nopanic - allocate boot memory without panicking
595 * @size: size of the request in bytes
596 * @align: alignment of the region
597 * @goal: preferred starting address of the region
599 * The goal is dropped if it can not be satisfied and the allocation will
600 * fall back to memory below @goal.
602 * Allocation may happen on any node in the system.
604 * Returns NULL on failure.
606 void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
609 bootmem_data_t *bdata;
612 list_for_each_entry(bdata, &bdata_list, list) {
613 ptr = alloc_bootmem_core(bdata, size, align, goal, 0);
621 * __alloc_bootmem - allocate boot memory
622 * @size: size of the request in bytes
623 * @align: alignment of the region
624 * @goal: preferred starting address of the region
626 * The goal is dropped if it can not be satisfied and the allocation will
627 * fall back to memory below @goal.
629 * Allocation may happen on any node in the system.
631 * The function panics if the request can not be satisfied.
633 void * __init __alloc_bootmem(unsigned long size, unsigned long align,
636 void *mem = __alloc_bootmem_nopanic(size,align,goal);
641 * Whoops, we cannot satisfy the allocation request.
643 printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
644 panic("Out of memory");
649 * __alloc_bootmem_node - allocate boot memory from a specific node
650 * @pgdat: node to allocate from
651 * @size: size of the request in bytes
652 * @align: alignment of the region
653 * @goal: preferred starting address of the region
655 * The goal is dropped if it can not be satisfied and the allocation will
656 * fall back to memory below @goal.
658 * Allocation may fall back to any node in the system if the specified node
659 * can not hold the requested memory.
661 * The function panics if the request can not be satisfied.
663 void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
664 unsigned long align, unsigned long goal)
668 ptr = alloc_bootmem_core(pgdat->bdata, size, align, goal, 0);
672 return __alloc_bootmem(size, align, goal);
675 #ifdef CONFIG_SPARSEMEM
677 * alloc_bootmem_section - allocate boot memory from a specific section
678 * @size: size of the request in bytes
679 * @section_nr: sparse map section to allocate from
681 * Return NULL on failure.
683 void * __init alloc_bootmem_section(unsigned long size,
684 unsigned long section_nr)
687 unsigned long limit, goal, start_nr, end_nr, pfn;
688 struct pglist_data *pgdat;
690 pfn = section_nr_to_pfn(section_nr);
691 goal = PFN_PHYS(pfn);
692 limit = PFN_PHYS(section_nr_to_pfn(section_nr + 1)) - 1;
693 pgdat = NODE_DATA(early_pfn_to_nid(pfn));
694 ptr = alloc_bootmem_core(pgdat->bdata, size, SMP_CACHE_BYTES, goal,
700 start_nr = pfn_to_section_nr(PFN_DOWN(__pa(ptr)));
701 end_nr = pfn_to_section_nr(PFN_DOWN(__pa(ptr) + size));
702 if (start_nr != section_nr || end_nr != section_nr) {
703 printk(KERN_WARNING "alloc_bootmem failed on section %ld.\n",
705 free_bootmem_core(pgdat->bdata, __pa(ptr), size);
713 void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
714 unsigned long align, unsigned long goal)
718 ptr = alloc_bootmem_core(pgdat->bdata, size, align, goal, 0);
722 return __alloc_bootmem_nopanic(size, align, goal);
725 #ifndef ARCH_LOW_ADDRESS_LIMIT
726 #define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL
730 * __alloc_bootmem_low - allocate low boot memory
731 * @size: size of the request in bytes
732 * @align: alignment of the region
733 * @goal: preferred starting address of the region
735 * The goal is dropped if it can not be satisfied and the allocation will
736 * fall back to memory below @goal.
738 * Allocation may happen on any node in the system.
740 * The function panics if the request can not be satisfied.
742 void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
745 bootmem_data_t *bdata;
748 list_for_each_entry(bdata, &bdata_list, list) {
749 ptr = alloc_bootmem_core(bdata, size, align, goal,
750 ARCH_LOW_ADDRESS_LIMIT);
756 * Whoops, we cannot satisfy the allocation request.
758 printk(KERN_ALERT "low bootmem alloc of %lu bytes failed!\n", size);
759 panic("Out of low memory");
764 * __alloc_bootmem_low_node - allocate low boot memory from a specific node
765 * @pgdat: node to allocate from
766 * @size: size of the request in bytes
767 * @align: alignment of the region
768 * @goal: preferred starting address of the region
770 * The goal is dropped if it can not be satisfied and the allocation will
771 * fall back to memory below @goal.
773 * Allocation may fall back to any node in the system if the specified node
774 * can not hold the requested memory.
776 * The function panics if the request can not be satisfied.
778 void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
779 unsigned long align, unsigned long goal)
781 return alloc_bootmem_core(pgdat->bdata, size, align, goal,
782 ARCH_LOW_ADDRESS_LIMIT);