memblock: Add "start" argument to memblock_find_base()
[linux-2.6/x86.git] / mm / memblock.c
blob468ff43a72b4a299f0d16d625a79a61d6627af52
1 /*
2 * Procedures for maintaining information about logical memory blocks.
4 * Peter Bergner, IBM Corp. June 2001.
5 * Copyright (C) 2001 Peter Bergner.
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.
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/memblock.h>
20 struct memblock memblock;
22 static int memblock_debug, memblock_can_resize;
23 static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS + 1];
24 static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS + 1];
26 #define MEMBLOCK_ERROR (~(phys_addr_t)0)
28 /* inline so we don't get a warning when pr_debug is compiled out */
29 static inline const char *memblock_type_name(struct memblock_type *type)
31 if (type == &memblock.memory)
32 return "memory";
33 else if (type == &memblock.reserved)
34 return "reserved";
35 else
36 return "unknown";
40 * Address comparison utilities
43 static phys_addr_t memblock_align_down(phys_addr_t addr, phys_addr_t size)
45 return addr & ~(size - 1);
48 static phys_addr_t memblock_align_up(phys_addr_t addr, phys_addr_t size)
50 return (addr + (size - 1)) & ~(size - 1);
53 static unsigned long memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
54 phys_addr_t base2, phys_addr_t size2)
56 return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
59 static long memblock_addrs_adjacent(phys_addr_t base1, phys_addr_t size1,
60 phys_addr_t base2, phys_addr_t size2)
62 if (base2 == base1 + size1)
63 return 1;
64 else if (base1 == base2 + size2)
65 return -1;
67 return 0;
70 static long memblock_regions_adjacent(struct memblock_type *type,
71 unsigned long r1, unsigned long r2)
73 phys_addr_t base1 = type->regions[r1].base;
74 phys_addr_t size1 = type->regions[r1].size;
75 phys_addr_t base2 = type->regions[r2].base;
76 phys_addr_t size2 = type->regions[r2].size;
78 return memblock_addrs_adjacent(base1, size1, base2, size2);
81 long memblock_overlaps_region(struct memblock_type *type, phys_addr_t base, phys_addr_t size)
83 unsigned long i;
85 for (i = 0; i < type->cnt; i++) {
86 phys_addr_t rgnbase = type->regions[i].base;
87 phys_addr_t rgnsize = type->regions[i].size;
88 if (memblock_addrs_overlap(base, size, rgnbase, rgnsize))
89 break;
92 return (i < type->cnt) ? i : -1;
96 * Find, allocate, deallocate or reserve unreserved regions. All allocations
97 * are top-down.
100 static phys_addr_t __init memblock_find_region(phys_addr_t start, phys_addr_t end,
101 phys_addr_t size, phys_addr_t align)
103 phys_addr_t base, res_base;
104 long j;
106 base = memblock_align_down((end - size), align);
107 while (start <= base) {
108 j = memblock_overlaps_region(&memblock.reserved, base, size);
109 if (j < 0)
110 return base;
111 res_base = memblock.reserved.regions[j].base;
112 if (res_base < size)
113 break;
114 base = memblock_align_down(res_base - size, align);
117 return MEMBLOCK_ERROR;
120 static phys_addr_t __init memblock_find_base(phys_addr_t size, phys_addr_t align,
121 phys_addr_t start, phys_addr_t end)
123 long i;
125 BUG_ON(0 == size);
127 size = memblock_align_up(size, align);
129 /* Pump up max_addr */
130 if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
131 end = memblock.current_limit;
133 /* We do a top-down search, this tends to limit memory
134 * fragmentation by keeping early boot allocs near the
135 * top of memory
137 for (i = memblock.memory.cnt - 1; i >= 0; i--) {
138 phys_addr_t memblockbase = memblock.memory.regions[i].base;
139 phys_addr_t memblocksize = memblock.memory.regions[i].size;
140 phys_addr_t bottom, top, found;
142 if (memblocksize < size)
143 continue;
144 if ((memblockbase + memblocksize) <= start)
145 break;
146 bottom = max(memblockbase, start);
147 top = min(memblockbase + memblocksize, end);
148 if (bottom >= top)
149 continue;
150 found = memblock_find_region(bottom, top, size, align);
151 if (found != MEMBLOCK_ERROR)
152 return found;
154 return MEMBLOCK_ERROR;
157 static void memblock_remove_region(struct memblock_type *type, unsigned long r)
159 unsigned long i;
161 for (i = r; i < type->cnt - 1; i++) {
162 type->regions[i].base = type->regions[i + 1].base;
163 type->regions[i].size = type->regions[i + 1].size;
165 type->cnt--;
168 /* Assumption: base addr of region 1 < base addr of region 2 */
169 static void memblock_coalesce_regions(struct memblock_type *type,
170 unsigned long r1, unsigned long r2)
172 type->regions[r1].size += type->regions[r2].size;
173 memblock_remove_region(type, r2);
176 /* Defined below but needed now */
177 static long memblock_add_region(struct memblock_type *type, phys_addr_t base, phys_addr_t size);
179 static int memblock_double_array(struct memblock_type *type)
181 struct memblock_region *new_array, *old_array;
182 phys_addr_t old_size, new_size, addr;
183 int use_slab = slab_is_available();
185 /* We don't allow resizing until we know about the reserved regions
186 * of memory that aren't suitable for allocation
188 if (!memblock_can_resize)
189 return -1;
191 pr_debug("memblock: %s array full, doubling...", memblock_type_name(type));
193 /* Calculate new doubled size */
194 old_size = type->max * sizeof(struct memblock_region);
195 new_size = old_size << 1;
197 /* Try to find some space for it.
199 * WARNING: We assume that either slab_is_available() and we use it or
200 * we use MEMBLOCK for allocations. That means that this is unsafe to use
201 * when bootmem is currently active (unless bootmem itself is implemented
202 * on top of MEMBLOCK which isn't the case yet)
204 * This should however not be an issue for now, as we currently only
205 * call into MEMBLOCK while it's still active, or much later when slab is
206 * active for memory hotplug operations
208 if (use_slab) {
209 new_array = kmalloc(new_size, GFP_KERNEL);
210 addr = new_array == NULL ? MEMBLOCK_ERROR : __pa(new_array);
211 } else
212 addr = memblock_find_base(new_size, sizeof(phys_addr_t), 0, MEMBLOCK_ALLOC_ACCESSIBLE);
213 if (addr == MEMBLOCK_ERROR) {
214 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
215 memblock_type_name(type), type->max, type->max * 2);
216 return -1;
218 new_array = __va(addr);
220 /* Found space, we now need to move the array over before
221 * we add the reserved region since it may be our reserved
222 * array itself that is full.
224 memcpy(new_array, type->regions, old_size);
225 memset(new_array + type->max, 0, old_size);
226 old_array = type->regions;
227 type->regions = new_array;
228 type->max <<= 1;
230 /* If we use SLAB that's it, we are done */
231 if (use_slab)
232 return 0;
234 /* Add the new reserved region now. Should not fail ! */
235 BUG_ON(memblock_add_region(&memblock.reserved, addr, new_size) < 0);
237 /* If the array wasn't our static init one, then free it. We only do
238 * that before SLAB is available as later on, we don't know whether
239 * to use kfree or free_bootmem_pages(). Shouldn't be a big deal
240 * anyways
242 if (old_array != memblock_memory_init_regions &&
243 old_array != memblock_reserved_init_regions)
244 memblock_free(__pa(old_array), old_size);
246 return 0;
249 extern int __weak memblock_memory_can_coalesce(phys_addr_t addr1, phys_addr_t size1,
250 phys_addr_t addr2, phys_addr_t size2)
252 return 1;
255 static long memblock_add_region(struct memblock_type *type, phys_addr_t base, phys_addr_t size)
257 unsigned long coalesced = 0;
258 long adjacent, i;
260 if ((type->cnt == 1) && (type->regions[0].size == 0)) {
261 type->regions[0].base = base;
262 type->regions[0].size = size;
263 return 0;
266 /* First try and coalesce this MEMBLOCK with another. */
267 for (i = 0; i < type->cnt; i++) {
268 phys_addr_t rgnbase = type->regions[i].base;
269 phys_addr_t rgnsize = type->regions[i].size;
271 if ((rgnbase == base) && (rgnsize == size))
272 /* Already have this region, so we're done */
273 return 0;
275 adjacent = memblock_addrs_adjacent(base, size, rgnbase, rgnsize);
276 /* Check if arch allows coalescing */
277 if (adjacent != 0 && type == &memblock.memory &&
278 !memblock_memory_can_coalesce(base, size, rgnbase, rgnsize))
279 break;
280 if (adjacent > 0) {
281 type->regions[i].base -= size;
282 type->regions[i].size += size;
283 coalesced++;
284 break;
285 } else if (adjacent < 0) {
286 type->regions[i].size += size;
287 coalesced++;
288 break;
292 /* If we plugged a hole, we may want to also coalesce with the
293 * next region
295 if ((i < type->cnt - 1) && memblock_regions_adjacent(type, i, i+1) &&
296 ((type != &memblock.memory || memblock_memory_can_coalesce(type->regions[i].base,
297 type->regions[i].size,
298 type->regions[i+1].base,
299 type->regions[i+1].size)))) {
300 memblock_coalesce_regions(type, i, i+1);
301 coalesced++;
304 if (coalesced)
305 return coalesced;
307 /* If we are out of space, we fail. It's too late to resize the array
308 * but then this shouldn't have happened in the first place.
310 if (WARN_ON(type->cnt >= type->max))
311 return -1;
313 /* Couldn't coalesce the MEMBLOCK, so add it to the sorted table. */
314 for (i = type->cnt - 1; i >= 0; i--) {
315 if (base < type->regions[i].base) {
316 type->regions[i+1].base = type->regions[i].base;
317 type->regions[i+1].size = type->regions[i].size;
318 } else {
319 type->regions[i+1].base = base;
320 type->regions[i+1].size = size;
321 break;
325 if (base < type->regions[0].base) {
326 type->regions[0].base = base;
327 type->regions[0].size = size;
329 type->cnt++;
331 /* The array is full ? Try to resize it. If that fails, we undo
332 * our allocation and return an error
334 if (type->cnt == type->max && memblock_double_array(type)) {
335 type->cnt--;
336 return -1;
339 return 0;
342 long memblock_add(phys_addr_t base, phys_addr_t size)
344 return memblock_add_region(&memblock.memory, base, size);
348 static long __memblock_remove(struct memblock_type *type, phys_addr_t base, phys_addr_t size)
350 phys_addr_t rgnbegin, rgnend;
351 phys_addr_t end = base + size;
352 int i;
354 rgnbegin = rgnend = 0; /* supress gcc warnings */
356 /* Find the region where (base, size) belongs to */
357 for (i=0; i < type->cnt; i++) {
358 rgnbegin = type->regions[i].base;
359 rgnend = rgnbegin + type->regions[i].size;
361 if ((rgnbegin <= base) && (end <= rgnend))
362 break;
365 /* Didn't find the region */
366 if (i == type->cnt)
367 return -1;
369 /* Check to see if we are removing entire region */
370 if ((rgnbegin == base) && (rgnend == end)) {
371 memblock_remove_region(type, i);
372 return 0;
375 /* Check to see if region is matching at the front */
376 if (rgnbegin == base) {
377 type->regions[i].base = end;
378 type->regions[i].size -= size;
379 return 0;
382 /* Check to see if the region is matching at the end */
383 if (rgnend == end) {
384 type->regions[i].size -= size;
385 return 0;
389 * We need to split the entry - adjust the current one to the
390 * beginging of the hole and add the region after hole.
392 type->regions[i].size = base - type->regions[i].base;
393 return memblock_add_region(type, end, rgnend - end);
396 long memblock_remove(phys_addr_t base, phys_addr_t size)
398 return __memblock_remove(&memblock.memory, base, size);
401 long __init memblock_free(phys_addr_t base, phys_addr_t size)
403 return __memblock_remove(&memblock.reserved, base, size);
406 long __init memblock_reserve(phys_addr_t base, phys_addr_t size)
408 struct memblock_type *_rgn = &memblock.reserved;
410 BUG_ON(0 == size);
412 return memblock_add_region(_rgn, base, size);
415 phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
417 phys_addr_t found;
419 /* We align the size to limit fragmentation. Without this, a lot of
420 * small allocs quickly eat up the whole reserve array on sparc
422 size = memblock_align_up(size, align);
424 found = memblock_find_base(size, align, 0, max_addr);
425 if (found != MEMBLOCK_ERROR &&
426 memblock_add_region(&memblock.reserved, found, size) >= 0)
427 return found;
429 return 0;
432 phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
434 phys_addr_t alloc;
436 alloc = __memblock_alloc_base(size, align, max_addr);
438 if (alloc == 0)
439 panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n",
440 (unsigned long long) size, (unsigned long long) max_addr);
442 return alloc;
445 phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
447 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
452 * Additional node-local allocators. Search for node memory is bottom up
453 * and walks memblock regions within that node bottom-up as well, but allocation
454 * within an memblock region is top-down.
457 phys_addr_t __weak __init memblock_nid_range(phys_addr_t start, phys_addr_t end, int *nid)
459 *nid = 0;
461 return end;
464 static phys_addr_t __init memblock_alloc_nid_region(struct memblock_region *mp,
465 phys_addr_t size,
466 phys_addr_t align, int nid)
468 phys_addr_t start, end;
470 start = mp->base;
471 end = start + mp->size;
473 start = memblock_align_up(start, align);
474 while (start < end) {
475 phys_addr_t this_end;
476 int this_nid;
478 this_end = memblock_nid_range(start, end, &this_nid);
479 if (this_nid == nid) {
480 phys_addr_t ret = memblock_find_region(start, this_end, size, align);
481 if (ret != MEMBLOCK_ERROR &&
482 memblock_add_region(&memblock.reserved, ret, size) >= 0)
483 return ret;
485 start = this_end;
488 return MEMBLOCK_ERROR;
491 phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
493 struct memblock_type *mem = &memblock.memory;
494 int i;
496 BUG_ON(0 == size);
498 /* We align the size to limit fragmentation. Without this, a lot of
499 * small allocs quickly eat up the whole reserve array on sparc
501 size = memblock_align_up(size, align);
503 /* We do a bottom-up search for a region with the right
504 * nid since that's easier considering how memblock_nid_range()
505 * works
507 for (i = 0; i < mem->cnt; i++) {
508 phys_addr_t ret = memblock_alloc_nid_region(&mem->regions[i],
509 size, align, nid);
510 if (ret != MEMBLOCK_ERROR)
511 return ret;
514 return memblock_alloc(size, align);
517 /* You must call memblock_analyze() before this. */
518 phys_addr_t __init memblock_phys_mem_size(void)
520 return memblock.memory_size;
523 phys_addr_t memblock_end_of_DRAM(void)
525 int idx = memblock.memory.cnt - 1;
527 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
530 /* You must call memblock_analyze() after this. */
531 void __init memblock_enforce_memory_limit(phys_addr_t memory_limit)
533 unsigned long i;
534 phys_addr_t limit;
535 struct memblock_region *p;
537 if (!memory_limit)
538 return;
540 /* Truncate the memblock regions to satisfy the memory limit. */
541 limit = memory_limit;
542 for (i = 0; i < memblock.memory.cnt; i++) {
543 if (limit > memblock.memory.regions[i].size) {
544 limit -= memblock.memory.regions[i].size;
545 continue;
548 memblock.memory.regions[i].size = limit;
549 memblock.memory.cnt = i + 1;
550 break;
553 memory_limit = memblock_end_of_DRAM();
555 /* And truncate any reserves above the limit also. */
556 for (i = 0; i < memblock.reserved.cnt; i++) {
557 p = &memblock.reserved.regions[i];
559 if (p->base > memory_limit)
560 p->size = 0;
561 else if ((p->base + p->size) > memory_limit)
562 p->size = memory_limit - p->base;
564 if (p->size == 0) {
565 memblock_remove_region(&memblock.reserved, i);
566 i--;
571 static int memblock_search(struct memblock_type *type, phys_addr_t addr)
573 unsigned int left = 0, right = type->cnt;
575 do {
576 unsigned int mid = (right + left) / 2;
578 if (addr < type->regions[mid].base)
579 right = mid;
580 else if (addr >= (type->regions[mid].base +
581 type->regions[mid].size))
582 left = mid + 1;
583 else
584 return mid;
585 } while (left < right);
586 return -1;
589 int __init memblock_is_reserved(phys_addr_t addr)
591 return memblock_search(&memblock.reserved, addr) != -1;
594 int memblock_is_memory(phys_addr_t addr)
596 return memblock_search(&memblock.memory, addr) != -1;
599 int memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
601 int idx = memblock_search(&memblock.reserved, base);
603 if (idx == -1)
604 return 0;
605 return memblock.reserved.regions[idx].base <= base &&
606 (memblock.reserved.regions[idx].base +
607 memblock.reserved.regions[idx].size) >= (base + size);
610 int memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
612 return memblock_overlaps_region(&memblock.reserved, base, size) >= 0;
616 void __init memblock_set_current_limit(phys_addr_t limit)
618 memblock.current_limit = limit;
621 static void memblock_dump(struct memblock_type *region, char *name)
623 unsigned long long base, size;
624 int i;
626 pr_info(" %s.cnt = 0x%lx\n", name, region->cnt);
628 for (i = 0; i < region->cnt; i++) {
629 base = region->regions[i].base;
630 size = region->regions[i].size;
632 pr_info(" %s[0x%x]\t0x%016llx - 0x%016llx, 0x%llx bytes\n",
633 name, i, base, base + size - 1, size);
637 void memblock_dump_all(void)
639 if (!memblock_debug)
640 return;
642 pr_info("MEMBLOCK configuration:\n");
643 pr_info(" memory size = 0x%llx\n", (unsigned long long)memblock.memory_size);
645 memblock_dump(&memblock.memory, "memory");
646 memblock_dump(&memblock.reserved, "reserved");
649 void __init memblock_analyze(void)
651 int i;
653 /* Check marker in the unused last array entry */
654 WARN_ON(memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS].base
655 != (phys_addr_t)RED_INACTIVE);
656 WARN_ON(memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS].base
657 != (phys_addr_t)RED_INACTIVE);
659 memblock.memory_size = 0;
661 for (i = 0; i < memblock.memory.cnt; i++)
662 memblock.memory_size += memblock.memory.regions[i].size;
664 /* We allow resizing from there */
665 memblock_can_resize = 1;
668 void __init memblock_init(void)
670 /* Hookup the initial arrays */
671 memblock.memory.regions = memblock_memory_init_regions;
672 memblock.memory.max = INIT_MEMBLOCK_REGIONS;
673 memblock.reserved.regions = memblock_reserved_init_regions;
674 memblock.reserved.max = INIT_MEMBLOCK_REGIONS;
676 /* Write a marker in the unused last array entry */
677 memblock.memory.regions[INIT_MEMBLOCK_REGIONS].base = (phys_addr_t)RED_INACTIVE;
678 memblock.reserved.regions[INIT_MEMBLOCK_REGIONS].base = (phys_addr_t)RED_INACTIVE;
680 /* Create a dummy zero size MEMBLOCK which will get coalesced away later.
681 * This simplifies the memblock_add() code below...
683 memblock.memory.regions[0].base = 0;
684 memblock.memory.regions[0].size = 0;
685 memblock.memory.cnt = 1;
687 /* Ditto. */
688 memblock.reserved.regions[0].base = 0;
689 memblock.reserved.regions[0].size = 0;
690 memblock.reserved.cnt = 1;
692 memblock.current_limit = MEMBLOCK_ALLOC_ANYWHERE;
695 static int __init early_memblock(char *p)
697 if (p && strstr(p, "debug"))
698 memblock_debug = 1;
699 return 0;
701 early_param("memblock", early_memblock);