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 | ||
79442ed1 | 23 | #include <asm-generic/sections.h> |
26f09e9b SS |
24 | #include <linux/io.h> |
25 | ||
26 | #include "internal.h" | |
79442ed1 | 27 | |
fe091c20 TH |
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 | ||
31 | struct memblock memblock __initdata_memblock = { | |
32 | .memory.regions = memblock_memory_init_regions, | |
33 | .memory.cnt = 1, /* empty dummy entry */ | |
34 | .memory.max = INIT_MEMBLOCK_REGIONS, | |
35 | ||
36 | .reserved.regions = memblock_reserved_init_regions, | |
37 | .reserved.cnt = 1, /* empty dummy entry */ | |
38 | .reserved.max = INIT_MEMBLOCK_REGIONS, | |
39 | ||
79442ed1 | 40 | .bottom_up = false, |
fe091c20 TH |
41 | .current_limit = MEMBLOCK_ALLOC_ANYWHERE, |
42 | }; | |
95f72d1e | 43 | |
10d06439 | 44 | int memblock_debug __initdata_memblock; |
55ac590c TC |
45 | #ifdef CONFIG_MOVABLE_NODE |
46 | bool movable_node_enabled __initdata_memblock = false; | |
47 | #endif | |
1aadc056 | 48 | static int memblock_can_resize __initdata_memblock; |
181eb394 GS |
49 | static int memblock_memory_in_slab __initdata_memblock = 0; |
50 | static int memblock_reserved_in_slab __initdata_memblock = 0; | |
95f72d1e | 51 | |
142b45a7 | 52 | /* inline so we don't get a warning when pr_debug is compiled out */ |
c2233116 RP |
53 | static __init_memblock const char * |
54 | memblock_type_name(struct memblock_type *type) | |
142b45a7 BH |
55 | { |
56 | if (type == &memblock.memory) | |
57 | return "memory"; | |
58 | else if (type == &memblock.reserved) | |
59 | return "reserved"; | |
60 | else | |
61 | return "unknown"; | |
62 | } | |
63 | ||
eb18f1b5 TH |
64 | /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */ |
65 | static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size) | |
66 | { | |
67 | return *size = min(*size, (phys_addr_t)ULLONG_MAX - base); | |
68 | } | |
69 | ||
6ed311b2 BH |
70 | /* |
71 | * Address comparison utilities | |
72 | */ | |
10d06439 | 73 | static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1, |
2898cc4c | 74 | phys_addr_t base2, phys_addr_t size2) |
95f72d1e YL |
75 | { |
76 | return ((base1 < (base2 + size2)) && (base2 < (base1 + size1))); | |
77 | } | |
78 | ||
2d7d3eb2 HS |
79 | static long __init_memblock memblock_overlaps_region(struct memblock_type *type, |
80 | phys_addr_t base, phys_addr_t size) | |
6ed311b2 BH |
81 | { |
82 | unsigned long i; | |
83 | ||
84 | for (i = 0; i < type->cnt; i++) { | |
85 | phys_addr_t rgnbase = type->regions[i].base; | |
86 | phys_addr_t rgnsize = type->regions[i].size; | |
87 | if (memblock_addrs_overlap(base, size, rgnbase, rgnsize)) | |
88 | break; | |
89 | } | |
90 | ||
91 | return (i < type->cnt) ? i : -1; | |
92 | } | |
93 | ||
79442ed1 TC |
94 | /* |
95 | * __memblock_find_range_bottom_up - find free area utility in bottom-up | |
96 | * @start: start of candidate range | |
97 | * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE} | |
98 | * @size: size of free area to find | |
99 | * @align: alignment of free area to find | |
b1154233 | 100 | * @nid: nid of the free area to find, %NUMA_NO_NODE for any node |
79442ed1 TC |
101 | * |
102 | * Utility called from memblock_find_in_range_node(), find free area bottom-up. | |
103 | * | |
104 | * RETURNS: | |
105 | * Found address on success, 0 on failure. | |
106 | */ | |
107 | static phys_addr_t __init_memblock | |
108 | __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end, | |
109 | phys_addr_t size, phys_addr_t align, int nid) | |
110 | { | |
111 | phys_addr_t this_start, this_end, cand; | |
112 | u64 i; | |
113 | ||
114 | for_each_free_mem_range(i, nid, &this_start, &this_end, NULL) { | |
115 | this_start = clamp(this_start, start, end); | |
116 | this_end = clamp(this_end, start, end); | |
117 | ||
118 | cand = round_up(this_start, align); | |
119 | if (cand < this_end && this_end - cand >= size) | |
120 | return cand; | |
121 | } | |
122 | ||
123 | return 0; | |
124 | } | |
125 | ||
7bd0b0f0 | 126 | /** |
1402899e | 127 | * __memblock_find_range_top_down - find free area utility, in top-down |
7bd0b0f0 TH |
128 | * @start: start of candidate range |
129 | * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE} | |
130 | * @size: size of free area to find | |
131 | * @align: alignment of free area to find | |
b1154233 | 132 | * @nid: nid of the free area to find, %NUMA_NO_NODE for any node |
7bd0b0f0 | 133 | * |
1402899e | 134 | * Utility called from memblock_find_in_range_node(), find free area top-down. |
7bd0b0f0 TH |
135 | * |
136 | * RETURNS: | |
79442ed1 | 137 | * Found address on success, 0 on failure. |
6ed311b2 | 138 | */ |
1402899e TC |
139 | static phys_addr_t __init_memblock |
140 | __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end, | |
141 | phys_addr_t size, phys_addr_t align, int nid) | |
f7210e6c TC |
142 | { |
143 | phys_addr_t this_start, this_end, cand; | |
144 | u64 i; | |
145 | ||
f7210e6c TC |
146 | for_each_free_mem_range_reverse(i, nid, &this_start, &this_end, NULL) { |
147 | this_start = clamp(this_start, start, end); | |
148 | this_end = clamp(this_end, start, end); | |
149 | ||
150 | if (this_end < size) | |
151 | continue; | |
152 | ||
153 | cand = round_down(this_end - size, align); | |
154 | if (cand >= this_start) | |
155 | return cand; | |
156 | } | |
1402899e | 157 | |
f7210e6c TC |
158 | return 0; |
159 | } | |
6ed311b2 | 160 | |
1402899e TC |
161 | /** |
162 | * memblock_find_in_range_node - find free area in given range and node | |
1402899e TC |
163 | * @size: size of free area to find |
164 | * @align: alignment of free area to find | |
87029ee9 GS |
165 | * @start: start of candidate range |
166 | * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE} | |
b1154233 | 167 | * @nid: nid of the free area to find, %NUMA_NO_NODE for any node |
1402899e TC |
168 | * |
169 | * Find @size free area aligned to @align in the specified range and node. | |
170 | * | |
79442ed1 TC |
171 | * When allocation direction is bottom-up, the @start should be greater |
172 | * than the end of the kernel image. Otherwise, it will be trimmed. The | |
173 | * reason is that we want the bottom-up allocation just near the kernel | |
174 | * image so it is highly likely that the allocated memory and the kernel | |
175 | * will reside in the same node. | |
176 | * | |
177 | * If bottom-up allocation failed, will try to allocate memory top-down. | |
178 | * | |
1402899e | 179 | * RETURNS: |
79442ed1 | 180 | * Found address on success, 0 on failure. |
1402899e | 181 | */ |
87029ee9 GS |
182 | phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size, |
183 | phys_addr_t align, phys_addr_t start, | |
184 | phys_addr_t end, int nid) | |
1402899e | 185 | { |
79442ed1 TC |
186 | int ret; |
187 | phys_addr_t kernel_end; | |
188 | ||
1402899e TC |
189 | /* pump up @end */ |
190 | if (end == MEMBLOCK_ALLOC_ACCESSIBLE) | |
191 | end = memblock.current_limit; | |
192 | ||
193 | /* avoid allocating the first page */ | |
194 | start = max_t(phys_addr_t, start, PAGE_SIZE); | |
195 | end = max(start, end); | |
79442ed1 TC |
196 | kernel_end = __pa_symbol(_end); |
197 | ||
198 | /* | |
199 | * try bottom-up allocation only when bottom-up mode | |
200 | * is set and @end is above the kernel image. | |
201 | */ | |
202 | if (memblock_bottom_up() && end > kernel_end) { | |
203 | phys_addr_t bottom_up_start; | |
204 | ||
205 | /* make sure we will allocate above the kernel */ | |
206 | bottom_up_start = max(start, kernel_end); | |
207 | ||
208 | /* ok, try bottom-up allocation first */ | |
209 | ret = __memblock_find_range_bottom_up(bottom_up_start, end, | |
210 | size, align, nid); | |
211 | if (ret) | |
212 | return ret; | |
213 | ||
214 | /* | |
215 | * we always limit bottom-up allocation above the kernel, | |
216 | * but top-down allocation doesn't have the limit, so | |
217 | * retrying top-down allocation may succeed when bottom-up | |
218 | * allocation failed. | |
219 | * | |
220 | * bottom-up allocation is expected to be fail very rarely, | |
221 | * so we use WARN_ONCE() here to see the stack trace if | |
222 | * fail happens. | |
223 | */ | |
224 | WARN_ONCE(1, "memblock: bottom-up allocation failed, " | |
225 | "memory hotunplug may be affected\n"); | |
226 | } | |
1402899e TC |
227 | |
228 | return __memblock_find_range_top_down(start, end, size, align, nid); | |
229 | } | |
230 | ||
7bd0b0f0 TH |
231 | /** |
232 | * memblock_find_in_range - find free area in given range | |
233 | * @start: start of candidate range | |
234 | * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE} | |
235 | * @size: size of free area to find | |
236 | * @align: alignment of free area to find | |
237 | * | |
238 | * Find @size free area aligned to @align in the specified range. | |
239 | * | |
240 | * RETURNS: | |
79442ed1 | 241 | * Found address on success, 0 on failure. |
fc769a8e | 242 | */ |
7bd0b0f0 TH |
243 | phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start, |
244 | phys_addr_t end, phys_addr_t size, | |
245 | phys_addr_t align) | |
6ed311b2 | 246 | { |
87029ee9 | 247 | return memblock_find_in_range_node(size, align, start, end, |
b1154233 | 248 | NUMA_NO_NODE); |
6ed311b2 BH |
249 | } |
250 | ||
10d06439 | 251 | static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r) |
95f72d1e | 252 | { |
1440c4e2 | 253 | type->total_size -= type->regions[r].size; |
7c0caeb8 TH |
254 | memmove(&type->regions[r], &type->regions[r + 1], |
255 | (type->cnt - (r + 1)) * sizeof(type->regions[r])); | |
e3239ff9 | 256 | type->cnt--; |
95f72d1e | 257 | |
8f7a6605 BH |
258 | /* Special case for empty arrays */ |
259 | if (type->cnt == 0) { | |
1440c4e2 | 260 | WARN_ON(type->total_size != 0); |
8f7a6605 BH |
261 | type->cnt = 1; |
262 | type->regions[0].base = 0; | |
263 | type->regions[0].size = 0; | |
66a20757 | 264 | type->regions[0].flags = 0; |
7c0caeb8 | 265 | memblock_set_region_node(&type->regions[0], MAX_NUMNODES); |
8f7a6605 | 266 | } |
95f72d1e YL |
267 | } |
268 | ||
354f17e1 PH |
269 | #ifdef CONFIG_ARCH_DISCARD_MEMBLOCK |
270 | ||
29f67386 YL |
271 | phys_addr_t __init_memblock get_allocated_memblock_reserved_regions_info( |
272 | phys_addr_t *addr) | |
273 | { | |
274 | if (memblock.reserved.regions == memblock_reserved_init_regions) | |
275 | return 0; | |
276 | ||
277 | *addr = __pa(memblock.reserved.regions); | |
278 | ||
279 | return PAGE_ALIGN(sizeof(struct memblock_region) * | |
280 | memblock.reserved.max); | |
281 | } | |
282 | ||
5e270e25 PH |
283 | phys_addr_t __init_memblock get_allocated_memblock_memory_regions_info( |
284 | phys_addr_t *addr) | |
285 | { | |
286 | if (memblock.memory.regions == memblock_memory_init_regions) | |
287 | return 0; | |
288 | ||
289 | *addr = __pa(memblock.memory.regions); | |
290 | ||
291 | return PAGE_ALIGN(sizeof(struct memblock_region) * | |
292 | memblock.memory.max); | |
293 | } | |
294 | ||
295 | #endif | |
296 | ||
48c3b583 GP |
297 | /** |
298 | * memblock_double_array - double the size of the memblock regions array | |
299 | * @type: memblock type of the regions array being doubled | |
300 | * @new_area_start: starting address of memory range to avoid overlap with | |
301 | * @new_area_size: size of memory range to avoid overlap with | |
302 | * | |
303 | * Double the size of the @type regions array. If memblock is being used to | |
304 | * allocate memory for a new reserved regions array and there is a previously | |
305 | * allocated memory range [@new_area_start,@new_area_start+@new_area_size] | |
306 | * waiting to be reserved, ensure the memory used by the new array does | |
307 | * not overlap. | |
308 | * | |
309 | * RETURNS: | |
310 | * 0 on success, -1 on failure. | |
311 | */ | |
312 | static int __init_memblock memblock_double_array(struct memblock_type *type, | |
313 | phys_addr_t new_area_start, | |
314 | phys_addr_t new_area_size) | |
142b45a7 BH |
315 | { |
316 | struct memblock_region *new_array, *old_array; | |
29f67386 | 317 | phys_addr_t old_alloc_size, new_alloc_size; |
142b45a7 BH |
318 | phys_addr_t old_size, new_size, addr; |
319 | int use_slab = slab_is_available(); | |
181eb394 | 320 | int *in_slab; |
142b45a7 BH |
321 | |
322 | /* We don't allow resizing until we know about the reserved regions | |
323 | * of memory that aren't suitable for allocation | |
324 | */ | |
325 | if (!memblock_can_resize) | |
326 | return -1; | |
327 | ||
142b45a7 BH |
328 | /* Calculate new doubled size */ |
329 | old_size = type->max * sizeof(struct memblock_region); | |
330 | new_size = old_size << 1; | |
29f67386 YL |
331 | /* |
332 | * We need to allocated new one align to PAGE_SIZE, | |
333 | * so we can free them completely later. | |
334 | */ | |
335 | old_alloc_size = PAGE_ALIGN(old_size); | |
336 | new_alloc_size = PAGE_ALIGN(new_size); | |
142b45a7 | 337 | |
181eb394 GS |
338 | /* Retrieve the slab flag */ |
339 | if (type == &memblock.memory) | |
340 | in_slab = &memblock_memory_in_slab; | |
341 | else | |
342 | in_slab = &memblock_reserved_in_slab; | |
343 | ||
142b45a7 BH |
344 | /* Try to find some space for it. |
345 | * | |
346 | * WARNING: We assume that either slab_is_available() and we use it or | |
fd07383b AM |
347 | * we use MEMBLOCK for allocations. That means that this is unsafe to |
348 | * use when bootmem is currently active (unless bootmem itself is | |
349 | * implemented on top of MEMBLOCK which isn't the case yet) | |
142b45a7 BH |
350 | * |
351 | * This should however not be an issue for now, as we currently only | |
fd07383b AM |
352 | * call into MEMBLOCK while it's still active, or much later when slab |
353 | * is active for memory hotplug operations | |
142b45a7 BH |
354 | */ |
355 | if (use_slab) { | |
356 | new_array = kmalloc(new_size, GFP_KERNEL); | |
1f5026a7 | 357 | addr = new_array ? __pa(new_array) : 0; |
4e2f0775 | 358 | } else { |
48c3b583 GP |
359 | /* only exclude range when trying to double reserved.regions */ |
360 | if (type != &memblock.reserved) | |
361 | new_area_start = new_area_size = 0; | |
362 | ||
363 | addr = memblock_find_in_range(new_area_start + new_area_size, | |
364 | memblock.current_limit, | |
29f67386 | 365 | new_alloc_size, PAGE_SIZE); |
48c3b583 GP |
366 | if (!addr && new_area_size) |
367 | addr = memblock_find_in_range(0, | |
fd07383b AM |
368 | min(new_area_start, memblock.current_limit), |
369 | new_alloc_size, PAGE_SIZE); | |
48c3b583 | 370 | |
15674868 | 371 | new_array = addr ? __va(addr) : NULL; |
4e2f0775 | 372 | } |
1f5026a7 | 373 | if (!addr) { |
142b45a7 BH |
374 | pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n", |
375 | memblock_type_name(type), type->max, type->max * 2); | |
376 | return -1; | |
377 | } | |
142b45a7 | 378 | |
fd07383b AM |
379 | memblock_dbg("memblock: %s is doubled to %ld at [%#010llx-%#010llx]", |
380 | memblock_type_name(type), type->max * 2, (u64)addr, | |
381 | (u64)addr + new_size - 1); | |
ea9e4376 | 382 | |
fd07383b AM |
383 | /* |
384 | * Found space, we now need to move the array over before we add the | |
385 | * reserved region since it may be our reserved array itself that is | |
386 | * full. | |
142b45a7 BH |
387 | */ |
388 | memcpy(new_array, type->regions, old_size); | |
389 | memset(new_array + type->max, 0, old_size); | |
390 | old_array = type->regions; | |
391 | type->regions = new_array; | |
392 | type->max <<= 1; | |
393 | ||
fd07383b | 394 | /* Free old array. We needn't free it if the array is the static one */ |
181eb394 GS |
395 | if (*in_slab) |
396 | kfree(old_array); | |
397 | else if (old_array != memblock_memory_init_regions && | |
398 | old_array != memblock_reserved_init_regions) | |
29f67386 | 399 | memblock_free(__pa(old_array), old_alloc_size); |
142b45a7 | 400 | |
fd07383b AM |
401 | /* |
402 | * Reserve the new array if that comes from the memblock. Otherwise, we | |
403 | * needn't do it | |
181eb394 GS |
404 | */ |
405 | if (!use_slab) | |
29f67386 | 406 | BUG_ON(memblock_reserve(addr, new_alloc_size)); |
181eb394 GS |
407 | |
408 | /* Update slab flag */ | |
409 | *in_slab = use_slab; | |
410 | ||
142b45a7 BH |
411 | return 0; |
412 | } | |
413 | ||
784656f9 TH |
414 | /** |
415 | * memblock_merge_regions - merge neighboring compatible regions | |
416 | * @type: memblock type to scan | |
417 | * | |
418 | * Scan @type and merge neighboring compatible regions. | |
419 | */ | |
420 | static void __init_memblock memblock_merge_regions(struct memblock_type *type) | |
95f72d1e | 421 | { |
784656f9 | 422 | int i = 0; |
95f72d1e | 423 | |
784656f9 TH |
424 | /* cnt never goes below 1 */ |
425 | while (i < type->cnt - 1) { | |
426 | struct memblock_region *this = &type->regions[i]; | |
427 | struct memblock_region *next = &type->regions[i + 1]; | |
95f72d1e | 428 | |
7c0caeb8 TH |
429 | if (this->base + this->size != next->base || |
430 | memblock_get_region_node(this) != | |
66a20757 TC |
431 | memblock_get_region_node(next) || |
432 | this->flags != next->flags) { | |
784656f9 TH |
433 | BUG_ON(this->base + this->size > next->base); |
434 | i++; | |
435 | continue; | |
8f7a6605 BH |
436 | } |
437 | ||
784656f9 | 438 | this->size += next->size; |
c0232ae8 LF |
439 | /* move forward from next + 1, index of which is i + 2 */ |
440 | memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next)); | |
784656f9 | 441 | type->cnt--; |
95f72d1e | 442 | } |
784656f9 | 443 | } |
95f72d1e | 444 | |
784656f9 TH |
445 | /** |
446 | * memblock_insert_region - insert new memblock region | |
209ff86d TC |
447 | * @type: memblock type to insert into |
448 | * @idx: index for the insertion point | |
449 | * @base: base address of the new region | |
450 | * @size: size of the new region | |
451 | * @nid: node id of the new region | |
66a20757 | 452 | * @flags: flags of the new region |
784656f9 TH |
453 | * |
454 | * Insert new memblock region [@base,@base+@size) into @type at @idx. | |
455 | * @type must already have extra room to accomodate the new region. | |
456 | */ | |
457 | static void __init_memblock memblock_insert_region(struct memblock_type *type, | |
458 | int idx, phys_addr_t base, | |
66a20757 TC |
459 | phys_addr_t size, |
460 | int nid, unsigned long flags) | |
784656f9 TH |
461 | { |
462 | struct memblock_region *rgn = &type->regions[idx]; | |
463 | ||
464 | BUG_ON(type->cnt >= type->max); | |
465 | memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn)); | |
466 | rgn->base = base; | |
467 | rgn->size = size; | |
66a20757 | 468 | rgn->flags = flags; |
7c0caeb8 | 469 | memblock_set_region_node(rgn, nid); |
784656f9 | 470 | type->cnt++; |
1440c4e2 | 471 | type->total_size += size; |
784656f9 TH |
472 | } |
473 | ||
474 | /** | |
475 | * memblock_add_region - add new memblock region | |
476 | * @type: memblock type to add new region into | |
477 | * @base: base address of the new region | |
478 | * @size: size of the new region | |
7fb0bc3f | 479 | * @nid: nid of the new region |
66a20757 | 480 | * @flags: flags of the new region |
784656f9 TH |
481 | * |
482 | * Add new memblock region [@base,@base+@size) into @type. The new region | |
483 | * is allowed to overlap with existing ones - overlaps don't affect already | |
484 | * existing regions. @type is guaranteed to be minimal (all neighbouring | |
485 | * compatible regions are merged) after the addition. | |
486 | * | |
487 | * RETURNS: | |
488 | * 0 on success, -errno on failure. | |
489 | */ | |
581adcbe | 490 | static int __init_memblock memblock_add_region(struct memblock_type *type, |
66a20757 TC |
491 | phys_addr_t base, phys_addr_t size, |
492 | int nid, unsigned long flags) | |
784656f9 TH |
493 | { |
494 | bool insert = false; | |
eb18f1b5 TH |
495 | phys_addr_t obase = base; |
496 | phys_addr_t end = base + memblock_cap_size(base, &size); | |
784656f9 TH |
497 | int i, nr_new; |
498 | ||
b3dc627c TH |
499 | if (!size) |
500 | return 0; | |
501 | ||
784656f9 TH |
502 | /* special case for empty array */ |
503 | if (type->regions[0].size == 0) { | |
1440c4e2 | 504 | WARN_ON(type->cnt != 1 || type->total_size); |
8f7a6605 BH |
505 | type->regions[0].base = base; |
506 | type->regions[0].size = size; | |
66a20757 | 507 | type->regions[0].flags = flags; |
7fb0bc3f | 508 | memblock_set_region_node(&type->regions[0], nid); |
1440c4e2 | 509 | type->total_size = size; |
8f7a6605 | 510 | return 0; |
95f72d1e | 511 | } |
784656f9 TH |
512 | repeat: |
513 | /* | |
514 | * The following is executed twice. Once with %false @insert and | |
515 | * then with %true. The first counts the number of regions needed | |
516 | * to accomodate the new area. The second actually inserts them. | |
142b45a7 | 517 | */ |
784656f9 TH |
518 | base = obase; |
519 | nr_new = 0; | |
95f72d1e | 520 | |
784656f9 TH |
521 | for (i = 0; i < type->cnt; i++) { |
522 | struct memblock_region *rgn = &type->regions[i]; | |
523 | phys_addr_t rbase = rgn->base; | |
524 | phys_addr_t rend = rbase + rgn->size; | |
525 | ||
526 | if (rbase >= end) | |
95f72d1e | 527 | break; |
784656f9 TH |
528 | if (rend <= base) |
529 | continue; | |
530 | /* | |
531 | * @rgn overlaps. If it separates the lower part of new | |
532 | * area, insert that portion. | |
533 | */ | |
534 | if (rbase > base) { | |
535 | nr_new++; | |
536 | if (insert) | |
537 | memblock_insert_region(type, i++, base, | |
66a20757 TC |
538 | rbase - base, nid, |
539 | flags); | |
95f72d1e | 540 | } |
784656f9 TH |
541 | /* area below @rend is dealt with, forget about it */ |
542 | base = min(rend, end); | |
95f72d1e | 543 | } |
784656f9 TH |
544 | |
545 | /* insert the remaining portion */ | |
546 | if (base < end) { | |
547 | nr_new++; | |
548 | if (insert) | |
66a20757 TC |
549 | memblock_insert_region(type, i, base, end - base, |
550 | nid, flags); | |
95f72d1e | 551 | } |
95f72d1e | 552 | |
784656f9 TH |
553 | /* |
554 | * If this was the first round, resize array and repeat for actual | |
555 | * insertions; otherwise, merge and return. | |
142b45a7 | 556 | */ |
784656f9 TH |
557 | if (!insert) { |
558 | while (type->cnt + nr_new > type->max) | |
48c3b583 | 559 | if (memblock_double_array(type, obase, size) < 0) |
784656f9 TH |
560 | return -ENOMEM; |
561 | insert = true; | |
562 | goto repeat; | |
563 | } else { | |
564 | memblock_merge_regions(type); | |
565 | return 0; | |
142b45a7 | 566 | } |
95f72d1e YL |
567 | } |
568 | ||
7fb0bc3f TH |
569 | int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size, |
570 | int nid) | |
571 | { | |
66a20757 | 572 | return memblock_add_region(&memblock.memory, base, size, nid, 0); |
7fb0bc3f TH |
573 | } |
574 | ||
581adcbe | 575 | int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size) |
95f72d1e | 576 | { |
66a20757 TC |
577 | return memblock_add_region(&memblock.memory, base, size, |
578 | MAX_NUMNODES, 0); | |
95f72d1e YL |
579 | } |
580 | ||
6a9ceb31 TH |
581 | /** |
582 | * memblock_isolate_range - isolate given range into disjoint memblocks | |
583 | * @type: memblock type to isolate range for | |
584 | * @base: base of range to isolate | |
585 | * @size: size of range to isolate | |
586 | * @start_rgn: out parameter for the start of isolated region | |
587 | * @end_rgn: out parameter for the end of isolated region | |
588 | * | |
589 | * Walk @type and ensure that regions don't cross the boundaries defined by | |
590 | * [@base,@base+@size). Crossing regions are split at the boundaries, | |
591 | * which may create at most two more regions. The index of the first | |
592 | * region inside the range is returned in *@start_rgn and end in *@end_rgn. | |
593 | * | |
594 | * RETURNS: | |
595 | * 0 on success, -errno on failure. | |
596 | */ | |
597 | static int __init_memblock memblock_isolate_range(struct memblock_type *type, | |
598 | phys_addr_t base, phys_addr_t size, | |
599 | int *start_rgn, int *end_rgn) | |
600 | { | |
eb18f1b5 | 601 | phys_addr_t end = base + memblock_cap_size(base, &size); |
6a9ceb31 TH |
602 | int i; |
603 | ||
604 | *start_rgn = *end_rgn = 0; | |
605 | ||
b3dc627c TH |
606 | if (!size) |
607 | return 0; | |
608 | ||
6a9ceb31 TH |
609 | /* we'll create at most two more regions */ |
610 | while (type->cnt + 2 > type->max) | |
48c3b583 | 611 | if (memblock_double_array(type, base, size) < 0) |
6a9ceb31 TH |
612 | return -ENOMEM; |
613 | ||
614 | for (i = 0; i < type->cnt; i++) { | |
615 | struct memblock_region *rgn = &type->regions[i]; | |
616 | phys_addr_t rbase = rgn->base; | |
617 | phys_addr_t rend = rbase + rgn->size; | |
618 | ||
619 | if (rbase >= end) | |
620 | break; | |
621 | if (rend <= base) | |
622 | continue; | |
623 | ||
624 | if (rbase < base) { | |
625 | /* | |
626 | * @rgn intersects from below. Split and continue | |
627 | * to process the next region - the new top half. | |
628 | */ | |
629 | rgn->base = base; | |
1440c4e2 TH |
630 | rgn->size -= base - rbase; |
631 | type->total_size -= base - rbase; | |
6a9ceb31 | 632 | memblock_insert_region(type, i, rbase, base - rbase, |
66a20757 TC |
633 | memblock_get_region_node(rgn), |
634 | rgn->flags); | |
6a9ceb31 TH |
635 | } else if (rend > end) { |
636 | /* | |
637 | * @rgn intersects from above. Split and redo the | |
638 | * current region - the new bottom half. | |
639 | */ | |
640 | rgn->base = end; | |
1440c4e2 TH |
641 | rgn->size -= end - rbase; |
642 | type->total_size -= end - rbase; | |
6a9ceb31 | 643 | memblock_insert_region(type, i--, rbase, end - rbase, |
66a20757 TC |
644 | memblock_get_region_node(rgn), |
645 | rgn->flags); | |
6a9ceb31 TH |
646 | } else { |
647 | /* @rgn is fully contained, record it */ | |
648 | if (!*end_rgn) | |
649 | *start_rgn = i; | |
650 | *end_rgn = i + 1; | |
651 | } | |
652 | } | |
653 | ||
654 | return 0; | |
655 | } | |
6a9ceb31 | 656 | |
581adcbe TH |
657 | static int __init_memblock __memblock_remove(struct memblock_type *type, |
658 | phys_addr_t base, phys_addr_t size) | |
95f72d1e | 659 | { |
71936180 TH |
660 | int start_rgn, end_rgn; |
661 | int i, ret; | |
95f72d1e | 662 | |
71936180 TH |
663 | ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); |
664 | if (ret) | |
665 | return ret; | |
95f72d1e | 666 | |
71936180 TH |
667 | for (i = end_rgn - 1; i >= start_rgn; i--) |
668 | memblock_remove_region(type, i); | |
8f7a6605 | 669 | return 0; |
95f72d1e YL |
670 | } |
671 | ||
581adcbe | 672 | int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size) |
95f72d1e YL |
673 | { |
674 | return __memblock_remove(&memblock.memory, base, size); | |
675 | } | |
676 | ||
581adcbe | 677 | int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size) |
95f72d1e | 678 | { |
24aa0788 | 679 | memblock_dbg(" memblock_free: [%#016llx-%#016llx] %pF\n", |
a150439c | 680 | (unsigned long long)base, |
931d13f5 | 681 | (unsigned long long)base + size - 1, |
a150439c | 682 | (void *)_RET_IP_); |
24aa0788 | 683 | |
95f72d1e YL |
684 | return __memblock_remove(&memblock.reserved, base, size); |
685 | } | |
686 | ||
66a20757 TC |
687 | static int __init_memblock memblock_reserve_region(phys_addr_t base, |
688 | phys_addr_t size, | |
689 | int nid, | |
690 | unsigned long flags) | |
95f72d1e | 691 | { |
e3239ff9 | 692 | struct memblock_type *_rgn = &memblock.reserved; |
95f72d1e | 693 | |
66a20757 | 694 | memblock_dbg("memblock_reserve: [%#016llx-%#016llx] flags %#02lx %pF\n", |
a150439c | 695 | (unsigned long long)base, |
931d13f5 | 696 | (unsigned long long)base + size - 1, |
66a20757 TC |
697 | flags, (void *)_RET_IP_); |
698 | ||
699 | return memblock_add_region(_rgn, base, size, nid, flags); | |
700 | } | |
95f72d1e | 701 | |
66a20757 TC |
702 | int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size) |
703 | { | |
704 | return memblock_reserve_region(base, size, MAX_NUMNODES, 0); | |
95f72d1e YL |
705 | } |
706 | ||
66b16edf TC |
707 | /** |
708 | * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG. | |
709 | * @base: the base phys addr of the region | |
710 | * @size: the size of the region | |
711 | * | |
712 | * This function isolates region [@base, @base + @size), and mark it with flag | |
713 | * MEMBLOCK_HOTPLUG. | |
714 | * | |
715 | * Return 0 on succees, -errno on failure. | |
716 | */ | |
717 | int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size) | |
718 | { | |
719 | struct memblock_type *type = &memblock.memory; | |
720 | int i, ret, start_rgn, end_rgn; | |
721 | ||
722 | ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); | |
723 | if (ret) | |
724 | return ret; | |
725 | ||
726 | for (i = start_rgn; i < end_rgn; i++) | |
727 | memblock_set_region_flags(&type->regions[i], MEMBLOCK_HOTPLUG); | |
728 | ||
729 | memblock_merge_regions(type); | |
730 | return 0; | |
731 | } | |
732 | ||
733 | /** | |
734 | * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region. | |
735 | * @base: the base phys addr of the region | |
736 | * @size: the size of the region | |
737 | * | |
738 | * This function isolates region [@base, @base + @size), and clear flag | |
739 | * MEMBLOCK_HOTPLUG for the isolated regions. | |
740 | * | |
741 | * Return 0 on succees, -errno on failure. | |
742 | */ | |
743 | int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size) | |
744 | { | |
745 | struct memblock_type *type = &memblock.memory; | |
746 | int i, ret, start_rgn, end_rgn; | |
747 | ||
748 | ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); | |
749 | if (ret) | |
750 | return ret; | |
751 | ||
752 | for (i = start_rgn; i < end_rgn; i++) | |
753 | memblock_clear_region_flags(&type->regions[i], | |
754 | MEMBLOCK_HOTPLUG); | |
755 | ||
756 | memblock_merge_regions(type); | |
757 | return 0; | |
758 | } | |
759 | ||
35fd0808 TH |
760 | /** |
761 | * __next_free_mem_range - next function for for_each_free_mem_range() | |
762 | * @idx: pointer to u64 loop variable | |
b1154233 | 763 | * @nid: node selector, %NUMA_NO_NODE for all nodes |
dad7557e WL |
764 | * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL |
765 | * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL | |
766 | * @out_nid: ptr to int for nid of the range, can be %NULL | |
35fd0808 TH |
767 | * |
768 | * Find the first free area from *@idx which matches @nid, fill the out | |
769 | * parameters, and update *@idx for the next iteration. The lower 32bit of | |
770 | * *@idx contains index into memory region and the upper 32bit indexes the | |
771 | * areas before each reserved region. For example, if reserved regions | |
772 | * look like the following, | |
773 | * | |
774 | * 0:[0-16), 1:[32-48), 2:[128-130) | |
775 | * | |
776 | * The upper 32bit indexes the following regions. | |
777 | * | |
778 | * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX) | |
779 | * | |
780 | * As both region arrays are sorted, the function advances the two indices | |
781 | * in lockstep and returns each intersection. | |
782 | */ | |
783 | void __init_memblock __next_free_mem_range(u64 *idx, int nid, | |
784 | phys_addr_t *out_start, | |
785 | phys_addr_t *out_end, int *out_nid) | |
786 | { | |
787 | struct memblock_type *mem = &memblock.memory; | |
788 | struct memblock_type *rsv = &memblock.reserved; | |
789 | int mi = *idx & 0xffffffff; | |
790 | int ri = *idx >> 32; | |
b1154233 | 791 | |
560dca27 GS |
792 | if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n")) |
793 | nid = NUMA_NO_NODE; | |
35fd0808 TH |
794 | |
795 | for ( ; mi < mem->cnt; mi++) { | |
796 | struct memblock_region *m = &mem->regions[mi]; | |
797 | phys_addr_t m_start = m->base; | |
798 | phys_addr_t m_end = m->base + m->size; | |
799 | ||
800 | /* only memory regions are associated with nodes, check it */ | |
560dca27 | 801 | if (nid != NUMA_NO_NODE && nid != memblock_get_region_node(m)) |
35fd0808 TH |
802 | continue; |
803 | ||
804 | /* scan areas before each reservation for intersection */ | |
805 | for ( ; ri < rsv->cnt + 1; ri++) { | |
806 | struct memblock_region *r = &rsv->regions[ri]; | |
807 | phys_addr_t r_start = ri ? r[-1].base + r[-1].size : 0; | |
808 | phys_addr_t r_end = ri < rsv->cnt ? r->base : ULLONG_MAX; | |
809 | ||
810 | /* if ri advanced past mi, break out to advance mi */ | |
811 | if (r_start >= m_end) | |
812 | break; | |
813 | /* if the two regions intersect, we're done */ | |
814 | if (m_start < r_end) { | |
815 | if (out_start) | |
816 | *out_start = max(m_start, r_start); | |
817 | if (out_end) | |
818 | *out_end = min(m_end, r_end); | |
819 | if (out_nid) | |
820 | *out_nid = memblock_get_region_node(m); | |
821 | /* | |
822 | * The region which ends first is advanced | |
823 | * for the next iteration. | |
824 | */ | |
825 | if (m_end <= r_end) | |
826 | mi++; | |
827 | else | |
828 | ri++; | |
829 | *idx = (u32)mi | (u64)ri << 32; | |
830 | return; | |
831 | } | |
832 | } | |
833 | } | |
834 | ||
835 | /* signal end of iteration */ | |
836 | *idx = ULLONG_MAX; | |
837 | } | |
838 | ||
7bd0b0f0 TH |
839 | /** |
840 | * __next_free_mem_range_rev - next function for for_each_free_mem_range_reverse() | |
841 | * @idx: pointer to u64 loop variable | |
b1154233 | 842 | * @nid: nid: node selector, %NUMA_NO_NODE for all nodes |
dad7557e WL |
843 | * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL |
844 | * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL | |
845 | * @out_nid: ptr to int for nid of the range, can be %NULL | |
7bd0b0f0 TH |
846 | * |
847 | * Reverse of __next_free_mem_range(). | |
55ac590c TC |
848 | * |
849 | * Linux kernel cannot migrate pages used by itself. Memory hotplug users won't | |
850 | * be able to hot-remove hotpluggable memory used by the kernel. So this | |
851 | * function skip hotpluggable regions if needed when allocating memory for the | |
852 | * kernel. | |
7bd0b0f0 TH |
853 | */ |
854 | void __init_memblock __next_free_mem_range_rev(u64 *idx, int nid, | |
855 | phys_addr_t *out_start, | |
856 | phys_addr_t *out_end, int *out_nid) | |
857 | { | |
858 | struct memblock_type *mem = &memblock.memory; | |
859 | struct memblock_type *rsv = &memblock.reserved; | |
860 | int mi = *idx & 0xffffffff; | |
861 | int ri = *idx >> 32; | |
b1154233 | 862 | |
560dca27 GS |
863 | if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n")) |
864 | nid = NUMA_NO_NODE; | |
7bd0b0f0 TH |
865 | |
866 | if (*idx == (u64)ULLONG_MAX) { | |
867 | mi = mem->cnt - 1; | |
868 | ri = rsv->cnt; | |
869 | } | |
870 | ||
871 | for ( ; mi >= 0; mi--) { | |
872 | struct memblock_region *m = &mem->regions[mi]; | |
873 | phys_addr_t m_start = m->base; | |
874 | phys_addr_t m_end = m->base + m->size; | |
875 | ||
876 | /* only memory regions are associated with nodes, check it */ | |
560dca27 | 877 | if (nid != NUMA_NO_NODE && nid != memblock_get_region_node(m)) |
7bd0b0f0 TH |
878 | continue; |
879 | ||
55ac590c TC |
880 | /* skip hotpluggable memory regions if needed */ |
881 | if (movable_node_is_enabled() && memblock_is_hotpluggable(m)) | |
882 | continue; | |
883 | ||
7bd0b0f0 TH |
884 | /* scan areas before each reservation for intersection */ |
885 | for ( ; ri >= 0; ri--) { | |
886 | struct memblock_region *r = &rsv->regions[ri]; | |
887 | phys_addr_t r_start = ri ? r[-1].base + r[-1].size : 0; | |
888 | phys_addr_t r_end = ri < rsv->cnt ? r->base : ULLONG_MAX; | |
889 | ||
890 | /* if ri advanced past mi, break out to advance mi */ | |
891 | if (r_end <= m_start) | |
892 | break; | |
893 | /* if the two regions intersect, we're done */ | |
894 | if (m_end > r_start) { | |
895 | if (out_start) | |
896 | *out_start = max(m_start, r_start); | |
897 | if (out_end) | |
898 | *out_end = min(m_end, r_end); | |
899 | if (out_nid) | |
900 | *out_nid = memblock_get_region_node(m); | |
901 | ||
902 | if (m_start >= r_start) | |
903 | mi--; | |
904 | else | |
905 | ri--; | |
906 | *idx = (u32)mi | (u64)ri << 32; | |
907 | return; | |
908 | } | |
909 | } | |
910 | } | |
911 | ||
912 | *idx = ULLONG_MAX; | |
913 | } | |
914 | ||
7c0caeb8 TH |
915 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
916 | /* | |
917 | * Common iterator interface used to define for_each_mem_range(). | |
918 | */ | |
919 | void __init_memblock __next_mem_pfn_range(int *idx, int nid, | |
920 | unsigned long *out_start_pfn, | |
921 | unsigned long *out_end_pfn, int *out_nid) | |
922 | { | |
923 | struct memblock_type *type = &memblock.memory; | |
924 | struct memblock_region *r; | |
925 | ||
926 | while (++*idx < type->cnt) { | |
927 | r = &type->regions[*idx]; | |
928 | ||
929 | if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size)) | |
930 | continue; | |
931 | if (nid == MAX_NUMNODES || nid == r->nid) | |
932 | break; | |
933 | } | |
934 | if (*idx >= type->cnt) { | |
935 | *idx = -1; | |
936 | return; | |
937 | } | |
938 | ||
939 | if (out_start_pfn) | |
940 | *out_start_pfn = PFN_UP(r->base); | |
941 | if (out_end_pfn) | |
942 | *out_end_pfn = PFN_DOWN(r->base + r->size); | |
943 | if (out_nid) | |
944 | *out_nid = r->nid; | |
945 | } | |
946 | ||
947 | /** | |
948 | * memblock_set_node - set node ID on memblock regions | |
949 | * @base: base of area to set node ID for | |
950 | * @size: size of area to set node ID for | |
e7e8de59 | 951 | * @type: memblock type to set node ID for |
7c0caeb8 TH |
952 | * @nid: node ID to set |
953 | * | |
e7e8de59 | 954 | * Set the nid of memblock @type regions in [@base,@base+@size) to @nid. |
7c0caeb8 TH |
955 | * Regions which cross the area boundaries are split as necessary. |
956 | * | |
957 | * RETURNS: | |
958 | * 0 on success, -errno on failure. | |
959 | */ | |
960 | int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size, | |
e7e8de59 | 961 | struct memblock_type *type, int nid) |
7c0caeb8 | 962 | { |
6a9ceb31 TH |
963 | int start_rgn, end_rgn; |
964 | int i, ret; | |
7c0caeb8 | 965 | |
6a9ceb31 TH |
966 | ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); |
967 | if (ret) | |
968 | return ret; | |
7c0caeb8 | 969 | |
6a9ceb31 | 970 | for (i = start_rgn; i < end_rgn; i++) |
e9d24ad3 | 971 | memblock_set_region_node(&type->regions[i], nid); |
7c0caeb8 TH |
972 | |
973 | memblock_merge_regions(type); | |
974 | return 0; | |
975 | } | |
976 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ | |
977 | ||
7bd0b0f0 TH |
978 | static phys_addr_t __init memblock_alloc_base_nid(phys_addr_t size, |
979 | phys_addr_t align, phys_addr_t max_addr, | |
980 | int nid) | |
95f72d1e | 981 | { |
6ed311b2 | 982 | phys_addr_t found; |
95f72d1e | 983 | |
79f40fab GS |
984 | if (!align) |
985 | align = SMP_CACHE_BYTES; | |
94f3d3af | 986 | |
847854f5 TH |
987 | /* align @size to avoid excessive fragmentation on reserved array */ |
988 | size = round_up(size, align); | |
989 | ||
87029ee9 | 990 | found = memblock_find_in_range_node(size, align, 0, max_addr, nid); |
9c8c27e2 | 991 | if (found && !memblock_reserve(found, size)) |
6ed311b2 | 992 | return found; |
95f72d1e | 993 | |
6ed311b2 | 994 | return 0; |
95f72d1e YL |
995 | } |
996 | ||
7bd0b0f0 TH |
997 | phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid) |
998 | { | |
999 | return memblock_alloc_base_nid(size, align, MEMBLOCK_ALLOC_ACCESSIBLE, nid); | |
1000 | } | |
1001 | ||
1002 | phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr) | |
1003 | { | |
b1154233 | 1004 | return memblock_alloc_base_nid(size, align, max_addr, NUMA_NO_NODE); |
7bd0b0f0 TH |
1005 | } |
1006 | ||
6ed311b2 | 1007 | phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr) |
95f72d1e | 1008 | { |
6ed311b2 BH |
1009 | phys_addr_t alloc; |
1010 | ||
1011 | alloc = __memblock_alloc_base(size, align, max_addr); | |
1012 | ||
1013 | if (alloc == 0) | |
1014 | panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n", | |
1015 | (unsigned long long) size, (unsigned long long) max_addr); | |
1016 | ||
1017 | return alloc; | |
95f72d1e YL |
1018 | } |
1019 | ||
6ed311b2 | 1020 | phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align) |
95f72d1e | 1021 | { |
6ed311b2 BH |
1022 | return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE); |
1023 | } | |
95f72d1e | 1024 | |
9d1e2492 BH |
1025 | phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid) |
1026 | { | |
1027 | phys_addr_t res = memblock_alloc_nid(size, align, nid); | |
1028 | ||
1029 | if (res) | |
1030 | return res; | |
15fb0972 | 1031 | return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE); |
95f72d1e YL |
1032 | } |
1033 | ||
26f09e9b SS |
1034 | /** |
1035 | * memblock_virt_alloc_internal - allocate boot memory block | |
1036 | * @size: size of memory block to be allocated in bytes | |
1037 | * @align: alignment of the region and block's size | |
1038 | * @min_addr: the lower bound of the memory region to allocate (phys address) | |
1039 | * @max_addr: the upper bound of the memory region to allocate (phys address) | |
1040 | * @nid: nid of the free area to find, %NUMA_NO_NODE for any node | |
1041 | * | |
1042 | * The @min_addr limit is dropped if it can not be satisfied and the allocation | |
1043 | * will fall back to memory below @min_addr. Also, allocation may fall back | |
1044 | * to any node in the system if the specified node can not | |
1045 | * hold the requested memory. | |
1046 | * | |
1047 | * The allocation is performed from memory region limited by | |
1048 | * memblock.current_limit if @max_addr == %BOOTMEM_ALLOC_ACCESSIBLE. | |
1049 | * | |
1050 | * The memory block is aligned on SMP_CACHE_BYTES if @align == 0. | |
1051 | * | |
1052 | * The phys address of allocated boot memory block is converted to virtual and | |
1053 | * allocated memory is reset to 0. | |
1054 | * | |
1055 | * In addition, function sets the min_count to 0 using kmemleak_alloc for | |
1056 | * allocated boot memory block, so that it is never reported as leaks. | |
1057 | * | |
1058 | * RETURNS: | |
1059 | * Virtual address of allocated memory block on success, NULL on failure. | |
1060 | */ | |
1061 | static void * __init memblock_virt_alloc_internal( | |
1062 | phys_addr_t size, phys_addr_t align, | |
1063 | phys_addr_t min_addr, phys_addr_t max_addr, | |
1064 | int nid) | |
1065 | { | |
1066 | phys_addr_t alloc; | |
1067 | void *ptr; | |
1068 | ||
560dca27 GS |
1069 | if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n")) |
1070 | nid = NUMA_NO_NODE; | |
26f09e9b SS |
1071 | |
1072 | /* | |
1073 | * Detect any accidental use of these APIs after slab is ready, as at | |
1074 | * this moment memblock may be deinitialized already and its | |
1075 | * internal data may be destroyed (after execution of free_all_bootmem) | |
1076 | */ | |
1077 | if (WARN_ON_ONCE(slab_is_available())) | |
1078 | return kzalloc_node(size, GFP_NOWAIT, nid); | |
1079 | ||
1080 | if (!align) | |
1081 | align = SMP_CACHE_BYTES; | |
1082 | ||
1083 | /* align @size to avoid excessive fragmentation on reserved array */ | |
1084 | size = round_up(size, align); | |
1085 | ||
1086 | again: | |
1087 | alloc = memblock_find_in_range_node(size, align, min_addr, max_addr, | |
1088 | nid); | |
1089 | if (alloc) | |
1090 | goto done; | |
1091 | ||
1092 | if (nid != NUMA_NO_NODE) { | |
1093 | alloc = memblock_find_in_range_node(size, align, min_addr, | |
1094 | max_addr, NUMA_NO_NODE); | |
1095 | if (alloc) | |
1096 | goto done; | |
1097 | } | |
1098 | ||
1099 | if (min_addr) { | |
1100 | min_addr = 0; | |
1101 | goto again; | |
1102 | } else { | |
1103 | goto error; | |
1104 | } | |
1105 | ||
1106 | done: | |
1107 | memblock_reserve(alloc, size); | |
1108 | ptr = phys_to_virt(alloc); | |
1109 | memset(ptr, 0, size); | |
1110 | ||
1111 | /* | |
1112 | * The min_count is set to 0 so that bootmem allocated blocks | |
1113 | * are never reported as leaks. This is because many of these blocks | |
1114 | * are only referred via the physical address which is not | |
1115 | * looked up by kmemleak. | |
1116 | */ | |
1117 | kmemleak_alloc(ptr, size, 0, 0); | |
1118 | ||
1119 | return ptr; | |
1120 | ||
1121 | error: | |
1122 | return NULL; | |
1123 | } | |
1124 | ||
1125 | /** | |
1126 | * memblock_virt_alloc_try_nid_nopanic - allocate boot memory block | |
1127 | * @size: size of memory block to be allocated in bytes | |
1128 | * @align: alignment of the region and block's size | |
1129 | * @min_addr: the lower bound of the memory region from where the allocation | |
1130 | * is preferred (phys address) | |
1131 | * @max_addr: the upper bound of the memory region from where the allocation | |
1132 | * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to | |
1133 | * allocate only from memory limited by memblock.current_limit value | |
1134 | * @nid: nid of the free area to find, %NUMA_NO_NODE for any node | |
1135 | * | |
1136 | * Public version of _memblock_virt_alloc_try_nid_nopanic() which provides | |
1137 | * additional debug information (including caller info), if enabled. | |
1138 | * | |
1139 | * RETURNS: | |
1140 | * Virtual address of allocated memory block on success, NULL on failure. | |
1141 | */ | |
1142 | void * __init memblock_virt_alloc_try_nid_nopanic( | |
1143 | phys_addr_t size, phys_addr_t align, | |
1144 | phys_addr_t min_addr, phys_addr_t max_addr, | |
1145 | int nid) | |
1146 | { | |
1147 | memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n", | |
1148 | __func__, (u64)size, (u64)align, nid, (u64)min_addr, | |
1149 | (u64)max_addr, (void *)_RET_IP_); | |
1150 | return memblock_virt_alloc_internal(size, align, min_addr, | |
1151 | max_addr, nid); | |
1152 | } | |
1153 | ||
1154 | /** | |
1155 | * memblock_virt_alloc_try_nid - allocate boot memory block with panicking | |
1156 | * @size: size of memory block to be allocated in bytes | |
1157 | * @align: alignment of the region and block's size | |
1158 | * @min_addr: the lower bound of the memory region from where the allocation | |
1159 | * is preferred (phys address) | |
1160 | * @max_addr: the upper bound of the memory region from where the allocation | |
1161 | * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to | |
1162 | * allocate only from memory limited by memblock.current_limit value | |
1163 | * @nid: nid of the free area to find, %NUMA_NO_NODE for any node | |
1164 | * | |
1165 | * Public panicking version of _memblock_virt_alloc_try_nid_nopanic() | |
1166 | * which provides debug information (including caller info), if enabled, | |
1167 | * and panics if the request can not be satisfied. | |
1168 | * | |
1169 | * RETURNS: | |
1170 | * Virtual address of allocated memory block on success, NULL on failure. | |
1171 | */ | |
1172 | void * __init memblock_virt_alloc_try_nid( | |
1173 | phys_addr_t size, phys_addr_t align, | |
1174 | phys_addr_t min_addr, phys_addr_t max_addr, | |
1175 | int nid) | |
1176 | { | |
1177 | void *ptr; | |
1178 | ||
1179 | memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n", | |
1180 | __func__, (u64)size, (u64)align, nid, (u64)min_addr, | |
1181 | (u64)max_addr, (void *)_RET_IP_); | |
1182 | ptr = memblock_virt_alloc_internal(size, align, | |
1183 | min_addr, max_addr, nid); | |
1184 | if (ptr) | |
1185 | return ptr; | |
1186 | ||
1187 | panic("%s: Failed to allocate %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx\n", | |
1188 | __func__, (u64)size, (u64)align, nid, (u64)min_addr, | |
1189 | (u64)max_addr); | |
1190 | return NULL; | |
1191 | } | |
1192 | ||
1193 | /** | |
1194 | * __memblock_free_early - free boot memory block | |
1195 | * @base: phys starting address of the boot memory block | |
1196 | * @size: size of the boot memory block in bytes | |
1197 | * | |
1198 | * Free boot memory block previously allocated by memblock_virt_alloc_xx() API. | |
1199 | * The freeing memory will not be released to the buddy allocator. | |
1200 | */ | |
1201 | void __init __memblock_free_early(phys_addr_t base, phys_addr_t size) | |
1202 | { | |
1203 | memblock_dbg("%s: [%#016llx-%#016llx] %pF\n", | |
1204 | __func__, (u64)base, (u64)base + size - 1, | |
1205 | (void *)_RET_IP_); | |
1206 | kmemleak_free_part(__va(base), size); | |
1207 | __memblock_remove(&memblock.reserved, base, size); | |
1208 | } | |
1209 | ||
1210 | /* | |
1211 | * __memblock_free_late - free bootmem block pages directly to buddy allocator | |
1212 | * @addr: phys starting address of the boot memory block | |
1213 | * @size: size of the boot memory block in bytes | |
1214 | * | |
1215 | * This is only useful when the bootmem allocator has already been torn | |
1216 | * down, but we are still initializing the system. Pages are released directly | |
1217 | * to the buddy allocator, no bootmem metadata is updated because it is gone. | |
1218 | */ | |
1219 | void __init __memblock_free_late(phys_addr_t base, phys_addr_t size) | |
1220 | { | |
1221 | u64 cursor, end; | |
1222 | ||
1223 | memblock_dbg("%s: [%#016llx-%#016llx] %pF\n", | |
1224 | __func__, (u64)base, (u64)base + size - 1, | |
1225 | (void *)_RET_IP_); | |
1226 | kmemleak_free_part(__va(base), size); | |
1227 | cursor = PFN_UP(base); | |
1228 | end = PFN_DOWN(base + size); | |
1229 | ||
1230 | for (; cursor < end; cursor++) { | |
1231 | __free_pages_bootmem(pfn_to_page(cursor), 0); | |
1232 | totalram_pages++; | |
1233 | } | |
1234 | } | |
9d1e2492 BH |
1235 | |
1236 | /* | |
1237 | * Remaining API functions | |
1238 | */ | |
1239 | ||
2898cc4c | 1240 | phys_addr_t __init memblock_phys_mem_size(void) |
95f72d1e | 1241 | { |
1440c4e2 | 1242 | return memblock.memory.total_size; |
95f72d1e YL |
1243 | } |
1244 | ||
595ad9af YL |
1245 | phys_addr_t __init memblock_mem_size(unsigned long limit_pfn) |
1246 | { | |
1247 | unsigned long pages = 0; | |
1248 | struct memblock_region *r; | |
1249 | unsigned long start_pfn, end_pfn; | |
1250 | ||
1251 | for_each_memblock(memory, r) { | |
1252 | start_pfn = memblock_region_memory_base_pfn(r); | |
1253 | end_pfn = memblock_region_memory_end_pfn(r); | |
1254 | start_pfn = min_t(unsigned long, start_pfn, limit_pfn); | |
1255 | end_pfn = min_t(unsigned long, end_pfn, limit_pfn); | |
1256 | pages += end_pfn - start_pfn; | |
1257 | } | |
1258 | ||
1259 | return (phys_addr_t)pages << PAGE_SHIFT; | |
1260 | } | |
1261 | ||
0a93ebef SR |
1262 | /* lowest address */ |
1263 | phys_addr_t __init_memblock memblock_start_of_DRAM(void) | |
1264 | { | |
1265 | return memblock.memory.regions[0].base; | |
1266 | } | |
1267 | ||
10d06439 | 1268 | phys_addr_t __init_memblock memblock_end_of_DRAM(void) |
95f72d1e YL |
1269 | { |
1270 | int idx = memblock.memory.cnt - 1; | |
1271 | ||
e3239ff9 | 1272 | return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size); |
95f72d1e YL |
1273 | } |
1274 | ||
c0ce8fef | 1275 | void __init memblock_enforce_memory_limit(phys_addr_t limit) |
95f72d1e YL |
1276 | { |
1277 | unsigned long i; | |
c0ce8fef | 1278 | phys_addr_t max_addr = (phys_addr_t)ULLONG_MAX; |
95f72d1e | 1279 | |
c0ce8fef | 1280 | if (!limit) |
95f72d1e YL |
1281 | return; |
1282 | ||
c0ce8fef | 1283 | /* find out max address */ |
95f72d1e | 1284 | for (i = 0; i < memblock.memory.cnt; i++) { |
c0ce8fef | 1285 | struct memblock_region *r = &memblock.memory.regions[i]; |
95f72d1e | 1286 | |
c0ce8fef TH |
1287 | if (limit <= r->size) { |
1288 | max_addr = r->base + limit; | |
1289 | break; | |
95f72d1e | 1290 | } |
c0ce8fef | 1291 | limit -= r->size; |
95f72d1e | 1292 | } |
c0ce8fef TH |
1293 | |
1294 | /* truncate both memory and reserved regions */ | |
1295 | __memblock_remove(&memblock.memory, max_addr, (phys_addr_t)ULLONG_MAX); | |
1296 | __memblock_remove(&memblock.reserved, max_addr, (phys_addr_t)ULLONG_MAX); | |
95f72d1e YL |
1297 | } |
1298 | ||
cd79481d | 1299 | static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr) |
72d4b0b4 BH |
1300 | { |
1301 | unsigned int left = 0, right = type->cnt; | |
1302 | ||
1303 | do { | |
1304 | unsigned int mid = (right + left) / 2; | |
1305 | ||
1306 | if (addr < type->regions[mid].base) | |
1307 | right = mid; | |
1308 | else if (addr >= (type->regions[mid].base + | |
1309 | type->regions[mid].size)) | |
1310 | left = mid + 1; | |
1311 | else | |
1312 | return mid; | |
1313 | } while (left < right); | |
1314 | return -1; | |
1315 | } | |
1316 | ||
2898cc4c | 1317 | int __init memblock_is_reserved(phys_addr_t addr) |
95f72d1e | 1318 | { |
72d4b0b4 BH |
1319 | return memblock_search(&memblock.reserved, addr) != -1; |
1320 | } | |
95f72d1e | 1321 | |
3661ca66 | 1322 | int __init_memblock memblock_is_memory(phys_addr_t addr) |
72d4b0b4 BH |
1323 | { |
1324 | return memblock_search(&memblock.memory, addr) != -1; | |
1325 | } | |
1326 | ||
e76b63f8 YL |
1327 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
1328 | int __init_memblock memblock_search_pfn_nid(unsigned long pfn, | |
1329 | unsigned long *start_pfn, unsigned long *end_pfn) | |
1330 | { | |
1331 | struct memblock_type *type = &memblock.memory; | |
1332 | int mid = memblock_search(type, (phys_addr_t)pfn << PAGE_SHIFT); | |
1333 | ||
1334 | if (mid == -1) | |
1335 | return -1; | |
1336 | ||
1337 | *start_pfn = type->regions[mid].base >> PAGE_SHIFT; | |
1338 | *end_pfn = (type->regions[mid].base + type->regions[mid].size) | |
1339 | >> PAGE_SHIFT; | |
1340 | ||
1341 | return type->regions[mid].nid; | |
1342 | } | |
1343 | #endif | |
1344 | ||
eab30949 SB |
1345 | /** |
1346 | * memblock_is_region_memory - check if a region is a subset of memory | |
1347 | * @base: base of region to check | |
1348 | * @size: size of region to check | |
1349 | * | |
1350 | * Check if the region [@base, @base+@size) is a subset of a memory block. | |
1351 | * | |
1352 | * RETURNS: | |
1353 | * 0 if false, non-zero if true | |
1354 | */ | |
3661ca66 | 1355 | int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size) |
72d4b0b4 | 1356 | { |
abb65272 | 1357 | int idx = memblock_search(&memblock.memory, base); |
eb18f1b5 | 1358 | phys_addr_t end = base + memblock_cap_size(base, &size); |
72d4b0b4 BH |
1359 | |
1360 | if (idx == -1) | |
1361 | return 0; | |
abb65272 TV |
1362 | return memblock.memory.regions[idx].base <= base && |
1363 | (memblock.memory.regions[idx].base + | |
eb18f1b5 | 1364 | memblock.memory.regions[idx].size) >= end; |
95f72d1e YL |
1365 | } |
1366 | ||
eab30949 SB |
1367 | /** |
1368 | * memblock_is_region_reserved - check if a region intersects reserved memory | |
1369 | * @base: base of region to check | |
1370 | * @size: size of region to check | |
1371 | * | |
1372 | * Check if the region [@base, @base+@size) intersects a reserved memory block. | |
1373 | * | |
1374 | * RETURNS: | |
1375 | * 0 if false, non-zero if true | |
1376 | */ | |
10d06439 | 1377 | int __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size) |
95f72d1e | 1378 | { |
eb18f1b5 | 1379 | memblock_cap_size(base, &size); |
f1c2c19c | 1380 | return memblock_overlaps_region(&memblock.reserved, base, size) >= 0; |
95f72d1e YL |
1381 | } |
1382 | ||
6ede1fd3 YL |
1383 | void __init_memblock memblock_trim_memory(phys_addr_t align) |
1384 | { | |
1385 | int i; | |
1386 | phys_addr_t start, end, orig_start, orig_end; | |
1387 | struct memblock_type *mem = &memblock.memory; | |
1388 | ||
1389 | for (i = 0; i < mem->cnt; i++) { | |
1390 | orig_start = mem->regions[i].base; | |
1391 | orig_end = mem->regions[i].base + mem->regions[i].size; | |
1392 | start = round_up(orig_start, align); | |
1393 | end = round_down(orig_end, align); | |
1394 | ||
1395 | if (start == orig_start && end == orig_end) | |
1396 | continue; | |
1397 | ||
1398 | if (start < end) { | |
1399 | mem->regions[i].base = start; | |
1400 | mem->regions[i].size = end - start; | |
1401 | } else { | |
1402 | memblock_remove_region(mem, i); | |
1403 | i--; | |
1404 | } | |
1405 | } | |
1406 | } | |
e63075a3 | 1407 | |
3661ca66 | 1408 | void __init_memblock memblock_set_current_limit(phys_addr_t limit) |
e63075a3 BH |
1409 | { |
1410 | memblock.current_limit = limit; | |
1411 | } | |
1412 | ||
7c0caeb8 | 1413 | static void __init_memblock memblock_dump(struct memblock_type *type, char *name) |
6ed311b2 BH |
1414 | { |
1415 | unsigned long long base, size; | |
66a20757 | 1416 | unsigned long flags; |
6ed311b2 BH |
1417 | int i; |
1418 | ||
7c0caeb8 | 1419 | pr_info(" %s.cnt = 0x%lx\n", name, type->cnt); |
6ed311b2 | 1420 | |
7c0caeb8 TH |
1421 | for (i = 0; i < type->cnt; i++) { |
1422 | struct memblock_region *rgn = &type->regions[i]; | |
1423 | char nid_buf[32] = ""; | |
1424 | ||
1425 | base = rgn->base; | |
1426 | size = rgn->size; | |
66a20757 | 1427 | flags = rgn->flags; |
7c0caeb8 TH |
1428 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
1429 | if (memblock_get_region_node(rgn) != MAX_NUMNODES) | |
1430 | snprintf(nid_buf, sizeof(nid_buf), " on node %d", | |
1431 | memblock_get_region_node(rgn)); | |
1432 | #endif | |
66a20757 TC |
1433 | pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes%s flags: %#lx\n", |
1434 | name, i, base, base + size - 1, size, nid_buf, flags); | |
6ed311b2 BH |
1435 | } |
1436 | } | |
1437 | ||
4ff7b82f | 1438 | void __init_memblock __memblock_dump_all(void) |
6ed311b2 | 1439 | { |
6ed311b2 | 1440 | pr_info("MEMBLOCK configuration:\n"); |
1440c4e2 TH |
1441 | pr_info(" memory size = %#llx reserved size = %#llx\n", |
1442 | (unsigned long long)memblock.memory.total_size, | |
1443 | (unsigned long long)memblock.reserved.total_size); | |
6ed311b2 BH |
1444 | |
1445 | memblock_dump(&memblock.memory, "memory"); | |
1446 | memblock_dump(&memblock.reserved, "reserved"); | |
1447 | } | |
1448 | ||
1aadc056 | 1449 | void __init memblock_allow_resize(void) |
6ed311b2 | 1450 | { |
142b45a7 | 1451 | memblock_can_resize = 1; |
6ed311b2 BH |
1452 | } |
1453 | ||
6ed311b2 BH |
1454 | static int __init early_memblock(char *p) |
1455 | { | |
1456 | if (p && strstr(p, "debug")) | |
1457 | memblock_debug = 1; | |
1458 | return 0; | |
1459 | } | |
1460 | early_param("memblock", early_memblock); | |
1461 | ||
c378ddd5 | 1462 | #if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK) |
6d03b885 BH |
1463 | |
1464 | static int memblock_debug_show(struct seq_file *m, void *private) | |
1465 | { | |
1466 | struct memblock_type *type = m->private; | |
1467 | struct memblock_region *reg; | |
1468 | int i; | |
1469 | ||
1470 | for (i = 0; i < type->cnt; i++) { | |
1471 | reg = &type->regions[i]; | |
1472 | seq_printf(m, "%4d: ", i); | |
1473 | if (sizeof(phys_addr_t) == 4) | |
1474 | seq_printf(m, "0x%08lx..0x%08lx\n", | |
1475 | (unsigned long)reg->base, | |
1476 | (unsigned long)(reg->base + reg->size - 1)); | |
1477 | else | |
1478 | seq_printf(m, "0x%016llx..0x%016llx\n", | |
1479 | (unsigned long long)reg->base, | |
1480 | (unsigned long long)(reg->base + reg->size - 1)); | |
1481 | ||
1482 | } | |
1483 | return 0; | |
1484 | } | |
1485 | ||
1486 | static int memblock_debug_open(struct inode *inode, struct file *file) | |
1487 | { | |
1488 | return single_open(file, memblock_debug_show, inode->i_private); | |
1489 | } | |
1490 | ||
1491 | static const struct file_operations memblock_debug_fops = { | |
1492 | .open = memblock_debug_open, | |
1493 | .read = seq_read, | |
1494 | .llseek = seq_lseek, | |
1495 | .release = single_release, | |
1496 | }; | |
1497 | ||
1498 | static int __init memblock_init_debugfs(void) | |
1499 | { | |
1500 | struct dentry *root = debugfs_create_dir("memblock", NULL); | |
1501 | if (!root) | |
1502 | return -ENXIO; | |
1503 | debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops); | |
1504 | debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops); | |
1505 | ||
1506 | return 0; | |
1507 | } | |
1508 | __initcall(memblock_init_debugfs); | |
1509 | ||
1510 | #endif /* CONFIG_DEBUG_FS */ |