]>
Commit | Line | Data |
---|---|---|
95f72d1e YL |
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> | |
142b45a7 | 14 | #include <linux/slab.h> |
95f72d1e YL |
15 | #include <linux/init.h> |
16 | #include <linux/bitops.h> | |
449e8df3 | 17 | #include <linux/poison.h> |
c196f76f | 18 | #include <linux/pfn.h> |
6d03b885 BH |
19 | #include <linux/debugfs.h> |
20 | #include <linux/seq_file.h> | |
95f72d1e YL |
21 | #include <linux/memblock.h> |
22 | ||
10d06439 | 23 | struct memblock memblock __initdata_memblock; |
95f72d1e | 24 | |
10d06439 YL |
25 | int memblock_debug __initdata_memblock; |
26 | int memblock_can_resize __initdata_memblock; | |
27 | static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS + 1] __initdata_memblock; | |
28 | static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS + 1] __initdata_memblock; | |
95f72d1e | 29 | |
142b45a7 BH |
30 | /* inline so we don't get a warning when pr_debug is compiled out */ |
31 | static inline const char *memblock_type_name(struct memblock_type *type) | |
32 | { | |
33 | if (type == &memblock.memory) | |
34 | return "memory"; | |
35 | else if (type == &memblock.reserved) | |
36 | return "reserved"; | |
37 | else | |
38 | return "unknown"; | |
39 | } | |
40 | ||
6ed311b2 BH |
41 | /* |
42 | * Address comparison utilities | |
43 | */ | |
95f72d1e | 44 | |
10d06439 | 45 | static phys_addr_t __init_memblock memblock_align_down(phys_addr_t addr, phys_addr_t size) |
95f72d1e | 46 | { |
6ed311b2 | 47 | return addr & ~(size - 1); |
95f72d1e YL |
48 | } |
49 | ||
10d06439 | 50 | static phys_addr_t __init_memblock memblock_align_up(phys_addr_t addr, phys_addr_t size) |
95f72d1e | 51 | { |
6ed311b2 | 52 | return (addr + (size - 1)) & ~(size - 1); |
95f72d1e YL |
53 | } |
54 | ||
10d06439 | 55 | static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1, |
2898cc4c | 56 | phys_addr_t base2, phys_addr_t size2) |
95f72d1e YL |
57 | { |
58 | return ((base1 < (base2 + size2)) && (base2 < (base1 + size1))); | |
59 | } | |
60 | ||
10d06439 | 61 | static long __init_memblock memblock_addrs_adjacent(phys_addr_t base1, phys_addr_t size1, |
2898cc4c | 62 | phys_addr_t base2, phys_addr_t size2) |
95f72d1e YL |
63 | { |
64 | if (base2 == base1 + size1) | |
65 | return 1; | |
66 | else if (base1 == base2 + size2) | |
67 | return -1; | |
68 | ||
69 | return 0; | |
70 | } | |
71 | ||
10d06439 | 72 | static long __init_memblock memblock_regions_adjacent(struct memblock_type *type, |
2898cc4c | 73 | unsigned long r1, unsigned long r2) |
95f72d1e | 74 | { |
2898cc4c BH |
75 | phys_addr_t base1 = type->regions[r1].base; |
76 | phys_addr_t size1 = type->regions[r1].size; | |
77 | phys_addr_t base2 = type->regions[r2].base; | |
78 | phys_addr_t size2 = type->regions[r2].size; | |
95f72d1e YL |
79 | |
80 | return memblock_addrs_adjacent(base1, size1, base2, size2); | |
81 | } | |
82 | ||
10d06439 | 83 | long __init_memblock memblock_overlaps_region(struct memblock_type *type, phys_addr_t base, phys_addr_t size) |
6ed311b2 BH |
84 | { |
85 | unsigned long i; | |
86 | ||
87 | for (i = 0; i < type->cnt; i++) { | |
88 | phys_addr_t rgnbase = type->regions[i].base; | |
89 | phys_addr_t rgnsize = type->regions[i].size; | |
90 | if (memblock_addrs_overlap(base, size, rgnbase, rgnsize)) | |
91 | break; | |
92 | } | |
93 | ||
94 | return (i < type->cnt) ? i : -1; | |
95 | } | |
96 | ||
97 | /* | |
98 | * Find, allocate, deallocate or reserve unreserved regions. All allocations | |
99 | * are top-down. | |
100 | */ | |
101 | ||
102 | static phys_addr_t __init memblock_find_region(phys_addr_t start, phys_addr_t end, | |
103 | phys_addr_t size, phys_addr_t align) | |
104 | { | |
105 | phys_addr_t base, res_base; | |
106 | long j; | |
107 | ||
25818f0f BH |
108 | /* Prevent allocations returning 0 as it's also used to |
109 | * indicate an allocation failure | |
110 | */ | |
111 | if (start == 0) | |
112 | start = PAGE_SIZE; | |
113 | ||
6ed311b2 BH |
114 | base = memblock_align_down((end - size), align); |
115 | while (start <= base) { | |
116 | j = memblock_overlaps_region(&memblock.reserved, base, size); | |
117 | if (j < 0) | |
118 | return base; | |
119 | res_base = memblock.reserved.regions[j].base; | |
120 | if (res_base < size) | |
121 | break; | |
122 | base = memblock_align_down(res_base - size, align); | |
123 | } | |
124 | ||
125 | return MEMBLOCK_ERROR; | |
126 | } | |
127 | ||
3661ca66 YL |
128 | static phys_addr_t __init_memblock memblock_find_base(phys_addr_t size, |
129 | phys_addr_t align, phys_addr_t start, phys_addr_t end) | |
6ed311b2 BH |
130 | { |
131 | long i; | |
6ed311b2 BH |
132 | |
133 | BUG_ON(0 == size); | |
134 | ||
135 | size = memblock_align_up(size, align); | |
136 | ||
137 | /* Pump up max_addr */ | |
fef501d4 BH |
138 | if (end == MEMBLOCK_ALLOC_ACCESSIBLE) |
139 | end = memblock.current_limit; | |
6ed311b2 BH |
140 | |
141 | /* We do a top-down search, this tends to limit memory | |
142 | * fragmentation by keeping early boot allocs near the | |
143 | * top of memory | |
144 | */ | |
145 | for (i = memblock.memory.cnt - 1; i >= 0; i--) { | |
146 | phys_addr_t memblockbase = memblock.memory.regions[i].base; | |
147 | phys_addr_t memblocksize = memblock.memory.regions[i].size; | |
fef501d4 | 148 | phys_addr_t bottom, top, found; |
6ed311b2 BH |
149 | |
150 | if (memblocksize < size) | |
151 | continue; | |
fef501d4 BH |
152 | if ((memblockbase + memblocksize) <= start) |
153 | break; | |
154 | bottom = max(memblockbase, start); | |
155 | top = min(memblockbase + memblocksize, end); | |
156 | if (bottom >= top) | |
157 | continue; | |
158 | found = memblock_find_region(bottom, top, size, align); | |
159 | if (found != MEMBLOCK_ERROR) | |
160 | return found; | |
6ed311b2 BH |
161 | } |
162 | return MEMBLOCK_ERROR; | |
163 | } | |
164 | ||
5303b68f YL |
165 | /* |
166 | * Find a free area with specified alignment in a specific range. | |
167 | */ | |
168 | u64 __init_memblock memblock_find_in_range(u64 start, u64 end, u64 size, u64 align) | |
169 | { | |
170 | return memblock_find_base(size, align, start, end); | |
171 | } | |
172 | ||
7950c407 YL |
173 | /* |
174 | * Free memblock.reserved.regions | |
175 | */ | |
176 | int __init_memblock memblock_free_reserved_regions(void) | |
177 | { | |
178 | if (memblock.reserved.regions == memblock_reserved_init_regions) | |
179 | return 0; | |
180 | ||
181 | return memblock_free(__pa(memblock.reserved.regions), | |
182 | sizeof(struct memblock_region) * memblock.reserved.max); | |
183 | } | |
184 | ||
185 | /* | |
186 | * Reserve memblock.reserved.regions | |
187 | */ | |
188 | int __init_memblock memblock_reserve_reserved_regions(void) | |
189 | { | |
190 | if (memblock.reserved.regions == memblock_reserved_init_regions) | |
191 | return 0; | |
192 | ||
193 | return memblock_reserve(__pa(memblock.reserved.regions), | |
194 | sizeof(struct memblock_region) * memblock.reserved.max); | |
195 | } | |
196 | ||
10d06439 | 197 | static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r) |
95f72d1e YL |
198 | { |
199 | unsigned long i; | |
200 | ||
e3239ff9 BH |
201 | for (i = r; i < type->cnt - 1; i++) { |
202 | type->regions[i].base = type->regions[i + 1].base; | |
203 | type->regions[i].size = type->regions[i + 1].size; | |
95f72d1e | 204 | } |
e3239ff9 | 205 | type->cnt--; |
95f72d1e YL |
206 | } |
207 | ||
208 | /* Assumption: base addr of region 1 < base addr of region 2 */ | |
10d06439 | 209 | static void __init_memblock memblock_coalesce_regions(struct memblock_type *type, |
95f72d1e YL |
210 | unsigned long r1, unsigned long r2) |
211 | { | |
e3239ff9 BH |
212 | type->regions[r1].size += type->regions[r2].size; |
213 | memblock_remove_region(type, r2); | |
95f72d1e YL |
214 | } |
215 | ||
142b45a7 BH |
216 | /* Defined below but needed now */ |
217 | static long memblock_add_region(struct memblock_type *type, phys_addr_t base, phys_addr_t size); | |
218 | ||
10d06439 | 219 | static int __init_memblock memblock_double_array(struct memblock_type *type) |
142b45a7 BH |
220 | { |
221 | struct memblock_region *new_array, *old_array; | |
222 | phys_addr_t old_size, new_size, addr; | |
223 | int use_slab = slab_is_available(); | |
224 | ||
225 | /* We don't allow resizing until we know about the reserved regions | |
226 | * of memory that aren't suitable for allocation | |
227 | */ | |
228 | if (!memblock_can_resize) | |
229 | return -1; | |
230 | ||
142b45a7 BH |
231 | /* Calculate new doubled size */ |
232 | old_size = type->max * sizeof(struct memblock_region); | |
233 | new_size = old_size << 1; | |
234 | ||
235 | /* Try to find some space for it. | |
236 | * | |
237 | * WARNING: We assume that either slab_is_available() and we use it or | |
238 | * we use MEMBLOCK for allocations. That means that this is unsafe to use | |
239 | * when bootmem is currently active (unless bootmem itself is implemented | |
240 | * on top of MEMBLOCK which isn't the case yet) | |
241 | * | |
242 | * This should however not be an issue for now, as we currently only | |
243 | * call into MEMBLOCK while it's still active, or much later when slab is | |
244 | * active for memory hotplug operations | |
245 | */ | |
246 | if (use_slab) { | |
247 | new_array = kmalloc(new_size, GFP_KERNEL); | |
248 | addr = new_array == NULL ? MEMBLOCK_ERROR : __pa(new_array); | |
249 | } else | |
fef501d4 | 250 | addr = memblock_find_base(new_size, sizeof(phys_addr_t), 0, MEMBLOCK_ALLOC_ACCESSIBLE); |
142b45a7 BH |
251 | if (addr == MEMBLOCK_ERROR) { |
252 | pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n", | |
253 | memblock_type_name(type), type->max, type->max * 2); | |
254 | return -1; | |
255 | } | |
256 | new_array = __va(addr); | |
257 | ||
ea9e4376 YL |
258 | memblock_dbg("memblock: %s array is doubled to %ld at [%#010llx-%#010llx]", |
259 | memblock_type_name(type), type->max * 2, (u64)addr, (u64)addr + new_size - 1); | |
260 | ||
142b45a7 BH |
261 | /* Found space, we now need to move the array over before |
262 | * we add the reserved region since it may be our reserved | |
263 | * array itself that is full. | |
264 | */ | |
265 | memcpy(new_array, type->regions, old_size); | |
266 | memset(new_array + type->max, 0, old_size); | |
267 | old_array = type->regions; | |
268 | type->regions = new_array; | |
269 | type->max <<= 1; | |
270 | ||
271 | /* If we use SLAB that's it, we are done */ | |
272 | if (use_slab) | |
273 | return 0; | |
274 | ||
275 | /* Add the new reserved region now. Should not fail ! */ | |
276 | BUG_ON(memblock_add_region(&memblock.reserved, addr, new_size) < 0); | |
277 | ||
278 | /* If the array wasn't our static init one, then free it. We only do | |
279 | * that before SLAB is available as later on, we don't know whether | |
280 | * to use kfree or free_bootmem_pages(). Shouldn't be a big deal | |
281 | * anyways | |
282 | */ | |
283 | if (old_array != memblock_memory_init_regions && | |
284 | old_array != memblock_reserved_init_regions) | |
285 | memblock_free(__pa(old_array), old_size); | |
286 | ||
287 | return 0; | |
288 | } | |
289 | ||
10d06439 | 290 | extern int __init_memblock __weak memblock_memory_can_coalesce(phys_addr_t addr1, phys_addr_t size1, |
d2cd563b BH |
291 | phys_addr_t addr2, phys_addr_t size2) |
292 | { | |
293 | return 1; | |
294 | } | |
295 | ||
10d06439 | 296 | static long __init_memblock memblock_add_region(struct memblock_type *type, phys_addr_t base, phys_addr_t size) |
95f72d1e YL |
297 | { |
298 | unsigned long coalesced = 0; | |
299 | long adjacent, i; | |
300 | ||
e3239ff9 BH |
301 | if ((type->cnt == 1) && (type->regions[0].size == 0)) { |
302 | type->regions[0].base = base; | |
303 | type->regions[0].size = size; | |
95f72d1e YL |
304 | return 0; |
305 | } | |
306 | ||
307 | /* First try and coalesce this MEMBLOCK with another. */ | |
e3239ff9 | 308 | for (i = 0; i < type->cnt; i++) { |
2898cc4c BH |
309 | phys_addr_t rgnbase = type->regions[i].base; |
310 | phys_addr_t rgnsize = type->regions[i].size; | |
95f72d1e YL |
311 | |
312 | if ((rgnbase == base) && (rgnsize == size)) | |
313 | /* Already have this region, so we're done */ | |
314 | return 0; | |
315 | ||
316 | adjacent = memblock_addrs_adjacent(base, size, rgnbase, rgnsize); | |
d2cd563b BH |
317 | /* Check if arch allows coalescing */ |
318 | if (adjacent != 0 && type == &memblock.memory && | |
319 | !memblock_memory_can_coalesce(base, size, rgnbase, rgnsize)) | |
320 | break; | |
95f72d1e | 321 | if (adjacent > 0) { |
e3239ff9 BH |
322 | type->regions[i].base -= size; |
323 | type->regions[i].size += size; | |
95f72d1e YL |
324 | coalesced++; |
325 | break; | |
326 | } else if (adjacent < 0) { | |
e3239ff9 | 327 | type->regions[i].size += size; |
95f72d1e YL |
328 | coalesced++; |
329 | break; | |
330 | } | |
331 | } | |
332 | ||
d2cd563b BH |
333 | /* If we plugged a hole, we may want to also coalesce with the |
334 | * next region | |
335 | */ | |
336 | if ((i < type->cnt - 1) && memblock_regions_adjacent(type, i, i+1) && | |
337 | ((type != &memblock.memory || memblock_memory_can_coalesce(type->regions[i].base, | |
338 | type->regions[i].size, | |
339 | type->regions[i+1].base, | |
340 | type->regions[i+1].size)))) { | |
e3239ff9 | 341 | memblock_coalesce_regions(type, i, i+1); |
95f72d1e YL |
342 | coalesced++; |
343 | } | |
344 | ||
345 | if (coalesced) | |
346 | return coalesced; | |
142b45a7 BH |
347 | |
348 | /* If we are out of space, we fail. It's too late to resize the array | |
349 | * but then this shouldn't have happened in the first place. | |
350 | */ | |
351 | if (WARN_ON(type->cnt >= type->max)) | |
95f72d1e YL |
352 | return -1; |
353 | ||
354 | /* Couldn't coalesce the MEMBLOCK, so add it to the sorted table. */ | |
e3239ff9 BH |
355 | for (i = type->cnt - 1; i >= 0; i--) { |
356 | if (base < type->regions[i].base) { | |
357 | type->regions[i+1].base = type->regions[i].base; | |
358 | type->regions[i+1].size = type->regions[i].size; | |
95f72d1e | 359 | } else { |
e3239ff9 BH |
360 | type->regions[i+1].base = base; |
361 | type->regions[i+1].size = size; | |
95f72d1e YL |
362 | break; |
363 | } | |
364 | } | |
365 | ||
e3239ff9 BH |
366 | if (base < type->regions[0].base) { |
367 | type->regions[0].base = base; | |
368 | type->regions[0].size = size; | |
95f72d1e | 369 | } |
e3239ff9 | 370 | type->cnt++; |
95f72d1e | 371 | |
142b45a7 BH |
372 | /* The array is full ? Try to resize it. If that fails, we undo |
373 | * our allocation and return an error | |
374 | */ | |
375 | if (type->cnt == type->max && memblock_double_array(type)) { | |
376 | type->cnt--; | |
377 | return -1; | |
378 | } | |
379 | ||
95f72d1e YL |
380 | return 0; |
381 | } | |
382 | ||
10d06439 | 383 | long __init_memblock memblock_add(phys_addr_t base, phys_addr_t size) |
95f72d1e | 384 | { |
e3239ff9 | 385 | return memblock_add_region(&memblock.memory, base, size); |
95f72d1e YL |
386 | |
387 | } | |
388 | ||
10d06439 | 389 | static long __init_memblock __memblock_remove(struct memblock_type *type, phys_addr_t base, phys_addr_t size) |
95f72d1e | 390 | { |
2898cc4c BH |
391 | phys_addr_t rgnbegin, rgnend; |
392 | phys_addr_t end = base + size; | |
95f72d1e YL |
393 | int i; |
394 | ||
395 | rgnbegin = rgnend = 0; /* supress gcc warnings */ | |
396 | ||
397 | /* Find the region where (base, size) belongs to */ | |
e3239ff9 BH |
398 | for (i=0; i < type->cnt; i++) { |
399 | rgnbegin = type->regions[i].base; | |
400 | rgnend = rgnbegin + type->regions[i].size; | |
95f72d1e YL |
401 | |
402 | if ((rgnbegin <= base) && (end <= rgnend)) | |
403 | break; | |
404 | } | |
405 | ||
406 | /* Didn't find the region */ | |
e3239ff9 | 407 | if (i == type->cnt) |
95f72d1e YL |
408 | return -1; |
409 | ||
410 | /* Check to see if we are removing entire region */ | |
411 | if ((rgnbegin == base) && (rgnend == end)) { | |
e3239ff9 | 412 | memblock_remove_region(type, i); |
95f72d1e YL |
413 | return 0; |
414 | } | |
415 | ||
416 | /* Check to see if region is matching at the front */ | |
417 | if (rgnbegin == base) { | |
e3239ff9 BH |
418 | type->regions[i].base = end; |
419 | type->regions[i].size -= size; | |
95f72d1e YL |
420 | return 0; |
421 | } | |
422 | ||
423 | /* Check to see if the region is matching at the end */ | |
424 | if (rgnend == end) { | |
e3239ff9 | 425 | type->regions[i].size -= size; |
95f72d1e YL |
426 | return 0; |
427 | } | |
428 | ||
429 | /* | |
430 | * We need to split the entry - adjust the current one to the | |
431 | * beginging of the hole and add the region after hole. | |
432 | */ | |
e3239ff9 BH |
433 | type->regions[i].size = base - type->regions[i].base; |
434 | return memblock_add_region(type, end, rgnend - end); | |
95f72d1e YL |
435 | } |
436 | ||
10d06439 | 437 | long __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size) |
95f72d1e YL |
438 | { |
439 | return __memblock_remove(&memblock.memory, base, size); | |
440 | } | |
441 | ||
3661ca66 | 442 | long __init_memblock memblock_free(phys_addr_t base, phys_addr_t size) |
95f72d1e YL |
443 | { |
444 | return __memblock_remove(&memblock.reserved, base, size); | |
445 | } | |
446 | ||
3661ca66 | 447 | long __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size) |
95f72d1e | 448 | { |
e3239ff9 | 449 | struct memblock_type *_rgn = &memblock.reserved; |
95f72d1e YL |
450 | |
451 | BUG_ON(0 == size); | |
452 | ||
453 | return memblock_add_region(_rgn, base, size); | |
454 | } | |
455 | ||
6ed311b2 | 456 | phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr) |
95f72d1e | 457 | { |
6ed311b2 | 458 | phys_addr_t found; |
95f72d1e | 459 | |
6ed311b2 BH |
460 | /* We align the size to limit fragmentation. Without this, a lot of |
461 | * small allocs quickly eat up the whole reserve array on sparc | |
462 | */ | |
463 | size = memblock_align_up(size, align); | |
95f72d1e | 464 | |
fef501d4 | 465 | found = memblock_find_base(size, align, 0, max_addr); |
6ed311b2 BH |
466 | if (found != MEMBLOCK_ERROR && |
467 | memblock_add_region(&memblock.reserved, found, size) >= 0) | |
468 | return found; | |
95f72d1e | 469 | |
6ed311b2 | 470 | return 0; |
95f72d1e YL |
471 | } |
472 | ||
6ed311b2 | 473 | phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr) |
95f72d1e | 474 | { |
6ed311b2 BH |
475 | phys_addr_t alloc; |
476 | ||
477 | alloc = __memblock_alloc_base(size, align, max_addr); | |
478 | ||
479 | if (alloc == 0) | |
480 | panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n", | |
481 | (unsigned long long) size, (unsigned long long) max_addr); | |
482 | ||
483 | return alloc; | |
95f72d1e YL |
484 | } |
485 | ||
6ed311b2 | 486 | phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align) |
95f72d1e | 487 | { |
6ed311b2 BH |
488 | return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE); |
489 | } | |
95f72d1e | 490 | |
95f72d1e | 491 | |
6ed311b2 BH |
492 | /* |
493 | * Additional node-local allocators. Search for node memory is bottom up | |
494 | * and walks memblock regions within that node bottom-up as well, but allocation | |
c196f76f BH |
495 | * within an memblock region is top-down. XXX I plan to fix that at some stage |
496 | * | |
497 | * WARNING: Only available after early_node_map[] has been populated, | |
498 | * on some architectures, that is after all the calls to add_active_range() | |
499 | * have been done to populate it. | |
6ed311b2 | 500 | */ |
95f72d1e | 501 | |
2898cc4c | 502 | phys_addr_t __weak __init memblock_nid_range(phys_addr_t start, phys_addr_t end, int *nid) |
c3f72b57 | 503 | { |
c196f76f BH |
504 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
505 | /* | |
506 | * This code originates from sparc which really wants use to walk by addresses | |
507 | * and returns the nid. This is not very convenient for early_pfn_map[] users | |
508 | * as the map isn't sorted yet, and it really wants to be walked by nid. | |
509 | * | |
510 | * For now, I implement the inefficient method below which walks the early | |
511 | * map multiple times. Eventually we may want to use an ARCH config option | |
512 | * to implement a completely different method for both case. | |
513 | */ | |
514 | unsigned long start_pfn, end_pfn; | |
515 | int i; | |
516 | ||
517 | for (i = 0; i < MAX_NUMNODES; i++) { | |
518 | get_pfn_range_for_nid(i, &start_pfn, &end_pfn); | |
519 | if (start < PFN_PHYS(start_pfn) || start >= PFN_PHYS(end_pfn)) | |
520 | continue; | |
521 | *nid = i; | |
522 | return min(end, PFN_PHYS(end_pfn)); | |
523 | } | |
524 | #endif | |
c3f72b57 BH |
525 | *nid = 0; |
526 | ||
527 | return end; | |
528 | } | |
529 | ||
2898cc4c BH |
530 | static phys_addr_t __init memblock_alloc_nid_region(struct memblock_region *mp, |
531 | phys_addr_t size, | |
532 | phys_addr_t align, int nid) | |
95f72d1e | 533 | { |
2898cc4c | 534 | phys_addr_t start, end; |
95f72d1e YL |
535 | |
536 | start = mp->base; | |
537 | end = start + mp->size; | |
538 | ||
539 | start = memblock_align_up(start, align); | |
540 | while (start < end) { | |
2898cc4c | 541 | phys_addr_t this_end; |
95f72d1e YL |
542 | int this_nid; |
543 | ||
35a1f0bd | 544 | this_end = memblock_nid_range(start, end, &this_nid); |
95f72d1e | 545 | if (this_nid == nid) { |
3a9c2c81 | 546 | phys_addr_t ret = memblock_find_region(start, this_end, size, align); |
4d629f9a | 547 | if (ret != MEMBLOCK_ERROR && |
3a9c2c81 | 548 | memblock_add_region(&memblock.reserved, ret, size) >= 0) |
95f72d1e YL |
549 | return ret; |
550 | } | |
551 | start = this_end; | |
552 | } | |
553 | ||
4d629f9a | 554 | return MEMBLOCK_ERROR; |
95f72d1e YL |
555 | } |
556 | ||
2898cc4c | 557 | phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid) |
95f72d1e | 558 | { |
e3239ff9 | 559 | struct memblock_type *mem = &memblock.memory; |
95f72d1e YL |
560 | int i; |
561 | ||
562 | BUG_ON(0 == size); | |
563 | ||
7f219c73 BH |
564 | /* We align the size to limit fragmentation. Without this, a lot of |
565 | * small allocs quickly eat up the whole reserve array on sparc | |
566 | */ | |
567 | size = memblock_align_up(size, align); | |
568 | ||
c3f72b57 BH |
569 | /* We do a bottom-up search for a region with the right |
570 | * nid since that's easier considering how memblock_nid_range() | |
571 | * works | |
572 | */ | |
95f72d1e | 573 | for (i = 0; i < mem->cnt; i++) { |
2898cc4c | 574 | phys_addr_t ret = memblock_alloc_nid_region(&mem->regions[i], |
95f72d1e | 575 | size, align, nid); |
4d629f9a | 576 | if (ret != MEMBLOCK_ERROR) |
95f72d1e YL |
577 | return ret; |
578 | } | |
579 | ||
9d1e2492 BH |
580 | return 0; |
581 | } | |
582 | ||
583 | phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid) | |
584 | { | |
585 | phys_addr_t res = memblock_alloc_nid(size, align, nid); | |
586 | ||
587 | if (res) | |
588 | return res; | |
918fe8d6 | 589 | return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ANYWHERE); |
95f72d1e YL |
590 | } |
591 | ||
9d1e2492 BH |
592 | |
593 | /* | |
594 | * Remaining API functions | |
595 | */ | |
596 | ||
95f72d1e | 597 | /* You must call memblock_analyze() before this. */ |
2898cc4c | 598 | phys_addr_t __init memblock_phys_mem_size(void) |
95f72d1e | 599 | { |
4734b594 | 600 | return memblock.memory_size; |
95f72d1e YL |
601 | } |
602 | ||
10d06439 | 603 | phys_addr_t __init_memblock memblock_end_of_DRAM(void) |
95f72d1e YL |
604 | { |
605 | int idx = memblock.memory.cnt - 1; | |
606 | ||
e3239ff9 | 607 | return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size); |
95f72d1e YL |
608 | } |
609 | ||
610 | /* You must call memblock_analyze() after this. */ | |
2898cc4c | 611 | void __init memblock_enforce_memory_limit(phys_addr_t memory_limit) |
95f72d1e YL |
612 | { |
613 | unsigned long i; | |
2898cc4c | 614 | phys_addr_t limit; |
e3239ff9 | 615 | struct memblock_region *p; |
95f72d1e YL |
616 | |
617 | if (!memory_limit) | |
618 | return; | |
619 | ||
620 | /* Truncate the memblock regions to satisfy the memory limit. */ | |
621 | limit = memory_limit; | |
622 | for (i = 0; i < memblock.memory.cnt; i++) { | |
e3239ff9 BH |
623 | if (limit > memblock.memory.regions[i].size) { |
624 | limit -= memblock.memory.regions[i].size; | |
95f72d1e YL |
625 | continue; |
626 | } | |
627 | ||
e3239ff9 | 628 | memblock.memory.regions[i].size = limit; |
95f72d1e YL |
629 | memblock.memory.cnt = i + 1; |
630 | break; | |
631 | } | |
632 | ||
95f72d1e YL |
633 | memory_limit = memblock_end_of_DRAM(); |
634 | ||
635 | /* And truncate any reserves above the limit also. */ | |
636 | for (i = 0; i < memblock.reserved.cnt; i++) { | |
e3239ff9 | 637 | p = &memblock.reserved.regions[i]; |
95f72d1e YL |
638 | |
639 | if (p->base > memory_limit) | |
640 | p->size = 0; | |
641 | else if ((p->base + p->size) > memory_limit) | |
642 | p->size = memory_limit - p->base; | |
643 | ||
644 | if (p->size == 0) { | |
645 | memblock_remove_region(&memblock.reserved, i); | |
646 | i--; | |
647 | } | |
648 | } | |
649 | } | |
650 | ||
2898cc4c | 651 | static int memblock_search(struct memblock_type *type, phys_addr_t addr) |
72d4b0b4 BH |
652 | { |
653 | unsigned int left = 0, right = type->cnt; | |
654 | ||
655 | do { | |
656 | unsigned int mid = (right + left) / 2; | |
657 | ||
658 | if (addr < type->regions[mid].base) | |
659 | right = mid; | |
660 | else if (addr >= (type->regions[mid].base + | |
661 | type->regions[mid].size)) | |
662 | left = mid + 1; | |
663 | else | |
664 | return mid; | |
665 | } while (left < right); | |
666 | return -1; | |
667 | } | |
668 | ||
2898cc4c | 669 | int __init memblock_is_reserved(phys_addr_t addr) |
95f72d1e | 670 | { |
72d4b0b4 BH |
671 | return memblock_search(&memblock.reserved, addr) != -1; |
672 | } | |
95f72d1e | 673 | |
3661ca66 | 674 | int __init_memblock memblock_is_memory(phys_addr_t addr) |
72d4b0b4 BH |
675 | { |
676 | return memblock_search(&memblock.memory, addr) != -1; | |
677 | } | |
678 | ||
3661ca66 | 679 | int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size) |
72d4b0b4 BH |
680 | { |
681 | int idx = memblock_search(&memblock.reserved, base); | |
682 | ||
683 | if (idx == -1) | |
684 | return 0; | |
685 | return memblock.reserved.regions[idx].base <= base && | |
686 | (memblock.reserved.regions[idx].base + | |
687 | memblock.reserved.regions[idx].size) >= (base + size); | |
95f72d1e YL |
688 | } |
689 | ||
10d06439 | 690 | int __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size) |
95f72d1e | 691 | { |
f1c2c19c | 692 | return memblock_overlaps_region(&memblock.reserved, base, size) >= 0; |
95f72d1e YL |
693 | } |
694 | ||
e63075a3 | 695 | |
3661ca66 | 696 | void __init_memblock memblock_set_current_limit(phys_addr_t limit) |
e63075a3 BH |
697 | { |
698 | memblock.current_limit = limit; | |
699 | } | |
700 | ||
10d06439 | 701 | static void __init_memblock memblock_dump(struct memblock_type *region, char *name) |
6ed311b2 BH |
702 | { |
703 | unsigned long long base, size; | |
704 | int i; | |
705 | ||
706 | pr_info(" %s.cnt = 0x%lx\n", name, region->cnt); | |
707 | ||
708 | for (i = 0; i < region->cnt; i++) { | |
709 | base = region->regions[i].base; | |
710 | size = region->regions[i].size; | |
711 | ||
ea9e4376 | 712 | pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes\n", |
6ed311b2 BH |
713 | name, i, base, base + size - 1, size); |
714 | } | |
715 | } | |
716 | ||
10d06439 | 717 | void __init_memblock memblock_dump_all(void) |
6ed311b2 BH |
718 | { |
719 | if (!memblock_debug) | |
720 | return; | |
721 | ||
722 | pr_info("MEMBLOCK configuration:\n"); | |
723 | pr_info(" memory size = 0x%llx\n", (unsigned long long)memblock.memory_size); | |
724 | ||
725 | memblock_dump(&memblock.memory, "memory"); | |
726 | memblock_dump(&memblock.reserved, "reserved"); | |
727 | } | |
728 | ||
729 | void __init memblock_analyze(void) | |
730 | { | |
731 | int i; | |
732 | ||
733 | /* Check marker in the unused last array entry */ | |
734 | WARN_ON(memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS].base | |
735 | != (phys_addr_t)RED_INACTIVE); | |
736 | WARN_ON(memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS].base | |
737 | != (phys_addr_t)RED_INACTIVE); | |
738 | ||
739 | memblock.memory_size = 0; | |
740 | ||
741 | for (i = 0; i < memblock.memory.cnt; i++) | |
742 | memblock.memory_size += memblock.memory.regions[i].size; | |
142b45a7 BH |
743 | |
744 | /* We allow resizing from there */ | |
745 | memblock_can_resize = 1; | |
6ed311b2 BH |
746 | } |
747 | ||
7590abe8 BH |
748 | void __init memblock_init(void) |
749 | { | |
750 | /* Hookup the initial arrays */ | |
751 | memblock.memory.regions = memblock_memory_init_regions; | |
752 | memblock.memory.max = INIT_MEMBLOCK_REGIONS; | |
753 | memblock.reserved.regions = memblock_reserved_init_regions; | |
754 | memblock.reserved.max = INIT_MEMBLOCK_REGIONS; | |
755 | ||
756 | /* Write a marker in the unused last array entry */ | |
757 | memblock.memory.regions[INIT_MEMBLOCK_REGIONS].base = (phys_addr_t)RED_INACTIVE; | |
758 | memblock.reserved.regions[INIT_MEMBLOCK_REGIONS].base = (phys_addr_t)RED_INACTIVE; | |
759 | ||
760 | /* Create a dummy zero size MEMBLOCK which will get coalesced away later. | |
761 | * This simplifies the memblock_add() code below... | |
762 | */ | |
763 | memblock.memory.regions[0].base = 0; | |
764 | memblock.memory.regions[0].size = 0; | |
765 | memblock.memory.cnt = 1; | |
766 | ||
767 | /* Ditto. */ | |
768 | memblock.reserved.regions[0].base = 0; | |
769 | memblock.reserved.regions[0].size = 0; | |
770 | memblock.reserved.cnt = 1; | |
771 | ||
772 | memblock.current_limit = MEMBLOCK_ALLOC_ANYWHERE; | |
773 | } | |
774 | ||
6ed311b2 BH |
775 | static int __init early_memblock(char *p) |
776 | { | |
777 | if (p && strstr(p, "debug")) | |
778 | memblock_debug = 1; | |
779 | return 0; | |
780 | } | |
781 | early_param("memblock", early_memblock); | |
782 | ||
10d06439 | 783 | #if defined(CONFIG_DEBUG_FS) && !defined(ARCH_DISCARD_MEMBLOCK) |
6d03b885 BH |
784 | |
785 | static int memblock_debug_show(struct seq_file *m, void *private) | |
786 | { | |
787 | struct memblock_type *type = m->private; | |
788 | struct memblock_region *reg; | |
789 | int i; | |
790 | ||
791 | for (i = 0; i < type->cnt; i++) { | |
792 | reg = &type->regions[i]; | |
793 | seq_printf(m, "%4d: ", i); | |
794 | if (sizeof(phys_addr_t) == 4) | |
795 | seq_printf(m, "0x%08lx..0x%08lx\n", | |
796 | (unsigned long)reg->base, | |
797 | (unsigned long)(reg->base + reg->size - 1)); | |
798 | else | |
799 | seq_printf(m, "0x%016llx..0x%016llx\n", | |
800 | (unsigned long long)reg->base, | |
801 | (unsigned long long)(reg->base + reg->size - 1)); | |
802 | ||
803 | } | |
804 | return 0; | |
805 | } | |
806 | ||
807 | static int memblock_debug_open(struct inode *inode, struct file *file) | |
808 | { | |
809 | return single_open(file, memblock_debug_show, inode->i_private); | |
810 | } | |
811 | ||
812 | static const struct file_operations memblock_debug_fops = { | |
813 | .open = memblock_debug_open, | |
814 | .read = seq_read, | |
815 | .llseek = seq_lseek, | |
816 | .release = single_release, | |
817 | }; | |
818 | ||
819 | static int __init memblock_init_debugfs(void) | |
820 | { | |
821 | struct dentry *root = debugfs_create_dir("memblock", NULL); | |
822 | if (!root) | |
823 | return -ENXIO; | |
824 | debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops); | |
825 | debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops); | |
826 | ||
827 | return 0; | |
828 | } | |
829 | __initcall(memblock_init_debugfs); | |
830 | ||
831 | #endif /* CONFIG_DEBUG_FS */ |