blob: 87108e77e476a326d69ec748f4e300d713fbd192 [file] [log] [blame]
Yinghai Lu95f72d12010-07-12 14:36:09 +10001/*
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>
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -070014#include <linux/slab.h>
Yinghai Lu95f72d12010-07-12 14:36:09 +100015#include <linux/init.h>
16#include <linux/bitops.h>
Benjamin Herrenschmidt449e8df2010-07-06 15:39:07 -070017#include <linux/poison.h>
Benjamin Herrenschmidtc196f762010-07-06 15:39:16 -070018#include <linux/pfn.h>
Benjamin Herrenschmidt6d03b882010-07-06 15:39:19 -070019#include <linux/debugfs.h>
20#include <linux/seq_file.h>
Yinghai Lu95f72d12010-07-12 14:36:09 +100021#include <linux/memblock.h>
22
Tang Chen79442ed2013-11-12 15:07:59 -080023#include <asm-generic/sections.h>
Santosh Shilimkar26f09e92014-01-21 15:50:19 -080024#include <linux/io.h>
25
26#include "internal.h"
Tang Chen79442ed2013-11-12 15:07:59 -080027
Tejun Heofe091c22011-12-08 10:22:07 -080028static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
29static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
Philipp Hachtmann70210ed2014-01-29 18:16:01 +010030#ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
31static struct memblock_region memblock_physmem_init_regions[INIT_PHYSMEM_REGIONS] __initdata_memblock;
32#endif
Tejun Heofe091c22011-12-08 10:22:07 -080033
34struct memblock memblock __initdata_memblock = {
35 .memory.regions = memblock_memory_init_regions,
36 .memory.cnt = 1, /* empty dummy entry */
37 .memory.max = INIT_MEMBLOCK_REGIONS,
38
39 .reserved.regions = memblock_reserved_init_regions,
40 .reserved.cnt = 1, /* empty dummy entry */
41 .reserved.max = INIT_MEMBLOCK_REGIONS,
42
Philipp Hachtmann70210ed2014-01-29 18:16:01 +010043#ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
44 .physmem.regions = memblock_physmem_init_regions,
45 .physmem.cnt = 1, /* empty dummy entry */
46 .physmem.max = INIT_PHYSMEM_REGIONS,
47#endif
48
Tang Chen79442ed2013-11-12 15:07:59 -080049 .bottom_up = false,
Tejun Heofe091c22011-12-08 10:22:07 -080050 .current_limit = MEMBLOCK_ALLOC_ANYWHERE,
51};
Yinghai Lu95f72d12010-07-12 14:36:09 +100052
Yinghai Lu10d06432010-07-28 15:43:02 +100053int memblock_debug __initdata_memblock;
Tang Chen55ac5902014-01-21 15:49:35 -080054#ifdef CONFIG_MOVABLE_NODE
55bool movable_node_enabled __initdata_memblock = false;
56#endif
Tony Lucka3f5baf2015-06-24 16:58:12 -070057static bool system_has_some_mirror __initdata_memblock = false;
Tejun Heo1aadc052011-12-08 10:22:08 -080058static int memblock_can_resize __initdata_memblock;
Gavin Shan181eb392012-05-29 15:06:50 -070059static int memblock_memory_in_slab __initdata_memblock = 0;
60static int memblock_reserved_in_slab __initdata_memblock = 0;
Yinghai Lu95f72d12010-07-12 14:36:09 +100061
Tony Lucka3f5baf2015-06-24 16:58:12 -070062ulong __init_memblock choose_memblock_flags(void)
63{
64 return system_has_some_mirror ? MEMBLOCK_MIRROR : MEMBLOCK_NONE;
65}
66
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -070067/* inline so we don't get a warning when pr_debug is compiled out */
Raghavendra D Prabhuc2233112012-10-08 16:33:55 -070068static __init_memblock const char *
69memblock_type_name(struct memblock_type *type)
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -070070{
71 if (type == &memblock.memory)
72 return "memory";
73 else if (type == &memblock.reserved)
74 return "reserved";
75 else
76 return "unknown";
77}
78
Tejun Heoeb18f1b2011-12-08 10:22:07 -080079/* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
80static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size)
81{
82 return *size = min(*size, (phys_addr_t)ULLONG_MAX - base);
83}
84
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +100085/*
86 * Address comparison utilities
87 */
Yinghai Lu10d06432010-07-28 15:43:02 +100088static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +100089 phys_addr_t base2, phys_addr_t size2)
Yinghai Lu95f72d12010-07-12 14:36:09 +100090{
91 return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
92}
93
H Hartley Sweeten2d7d3eb2011-10-31 17:09:15 -070094static long __init_memblock memblock_overlaps_region(struct memblock_type *type,
95 phys_addr_t base, phys_addr_t size)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +100096{
97 unsigned long i;
98
99 for (i = 0; i < type->cnt; i++) {
100 phys_addr_t rgnbase = type->regions[i].base;
101 phys_addr_t rgnsize = type->regions[i].size;
102 if (memblock_addrs_overlap(base, size, rgnbase, rgnsize))
103 break;
104 }
105
106 return (i < type->cnt) ? i : -1;
107}
108
Tang Chen79442ed2013-11-12 15:07:59 -0800109/*
110 * __memblock_find_range_bottom_up - find free area utility in bottom-up
111 * @start: start of candidate range
112 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
113 * @size: size of free area to find
114 * @align: alignment of free area to find
Grygorii Strashkob1154232014-01-21 15:50:16 -0800115 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
Tony Luckfc6daaf2015-06-24 16:58:09 -0700116 * @flags: pick from blocks based on memory attributes
Tang Chen79442ed2013-11-12 15:07:59 -0800117 *
118 * Utility called from memblock_find_in_range_node(), find free area bottom-up.
119 *
120 * RETURNS:
121 * Found address on success, 0 on failure.
122 */
123static phys_addr_t __init_memblock
124__memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end,
Tony Luckfc6daaf2015-06-24 16:58:09 -0700125 phys_addr_t size, phys_addr_t align, int nid,
126 ulong flags)
Tang Chen79442ed2013-11-12 15:07:59 -0800127{
128 phys_addr_t this_start, this_end, cand;
129 u64 i;
130
Tony Luckfc6daaf2015-06-24 16:58:09 -0700131 for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) {
Tang Chen79442ed2013-11-12 15:07:59 -0800132 this_start = clamp(this_start, start, end);
133 this_end = clamp(this_end, start, end);
134
135 cand = round_up(this_start, align);
136 if (cand < this_end && this_end - cand >= size)
137 return cand;
138 }
139
140 return 0;
141}
142
Tejun Heo7bd0b0f2011-12-08 10:22:09 -0800143/**
Tang Chen14028992013-11-12 15:07:57 -0800144 * __memblock_find_range_top_down - find free area utility, in top-down
145 * @start: start of candidate range
146 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
147 * @size: size of free area to find
148 * @align: alignment of free area to find
Grygorii Strashkob1154232014-01-21 15:50:16 -0800149 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
Tony Luckfc6daaf2015-06-24 16:58:09 -0700150 * @flags: pick from blocks based on memory attributes
Tang Chen14028992013-11-12 15:07:57 -0800151 *
152 * Utility called from memblock_find_in_range_node(), find free area top-down.
153 *
154 * RETURNS:
Tang Chen79442ed2013-11-12 15:07:59 -0800155 * Found address on success, 0 on failure.
Tang Chen14028992013-11-12 15:07:57 -0800156 */
157static phys_addr_t __init_memblock
158__memblock_find_range_top_down(phys_addr_t start, phys_addr_t end,
Tony Luckfc6daaf2015-06-24 16:58:09 -0700159 phys_addr_t size, phys_addr_t align, int nid,
160 ulong flags)
Tang Chen14028992013-11-12 15:07:57 -0800161{
162 phys_addr_t this_start, this_end, cand;
163 u64 i;
164
Tony Luckfc6daaf2015-06-24 16:58:09 -0700165 for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end,
166 NULL) {
Tang Chen14028992013-11-12 15:07:57 -0800167 this_start = clamp(this_start, start, end);
168 this_end = clamp(this_end, start, end);
169
170 if (this_end < size)
171 continue;
172
173 cand = round_down(this_end - size, align);
174 if (cand >= this_start)
175 return cand;
176 }
177
178 return 0;
179}
180
181/**
Tejun Heo7bd0b0f2011-12-08 10:22:09 -0800182 * memblock_find_in_range_node - find free area in given range and node
Tejun Heo7bd0b0f2011-12-08 10:22:09 -0800183 * @size: size of free area to find
184 * @align: alignment of free area to find
Grygorii Strashko87029ee2014-01-21 15:50:14 -0800185 * @start: start of candidate range
186 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
Grygorii Strashkob1154232014-01-21 15:50:16 -0800187 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
Tony Luckfc6daaf2015-06-24 16:58:09 -0700188 * @flags: pick from blocks based on memory attributes
Tejun Heo7bd0b0f2011-12-08 10:22:09 -0800189 *
190 * Find @size free area aligned to @align in the specified range and node.
191 *
Tang Chen79442ed2013-11-12 15:07:59 -0800192 * When allocation direction is bottom-up, the @start should be greater
193 * than the end of the kernel image. Otherwise, it will be trimmed. The
194 * reason is that we want the bottom-up allocation just near the kernel
195 * image so it is highly likely that the allocated memory and the kernel
196 * will reside in the same node.
197 *
198 * If bottom-up allocation failed, will try to allocate memory top-down.
199 *
Tejun Heo7bd0b0f2011-12-08 10:22:09 -0800200 * RETURNS:
Tang Chen79442ed2013-11-12 15:07:59 -0800201 * Found address on success, 0 on failure.
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000202 */
Grygorii Strashko87029ee2014-01-21 15:50:14 -0800203phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size,
204 phys_addr_t align, phys_addr_t start,
Tony Luckfc6daaf2015-06-24 16:58:09 -0700205 phys_addr_t end, int nid, ulong flags)
Tang Chenf7210e62013-02-22 16:33:51 -0800206{
Tang Chen0cfb8f02014-08-29 15:18:31 -0700207 phys_addr_t kernel_end, ret;
Tang Chen79442ed2013-11-12 15:07:59 -0800208
Tang Chenf7210e62013-02-22 16:33:51 -0800209 /* pump up @end */
210 if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
211 end = memblock.current_limit;
212
213 /* avoid allocating the first page */
214 start = max_t(phys_addr_t, start, PAGE_SIZE);
215 end = max(start, end);
Tang Chen79442ed2013-11-12 15:07:59 -0800216 kernel_end = __pa_symbol(_end);
217
218 /*
219 * try bottom-up allocation only when bottom-up mode
220 * is set and @end is above the kernel image.
221 */
222 if (memblock_bottom_up() && end > kernel_end) {
223 phys_addr_t bottom_up_start;
224
225 /* make sure we will allocate above the kernel */
226 bottom_up_start = max(start, kernel_end);
227
228 /* ok, try bottom-up allocation first */
229 ret = __memblock_find_range_bottom_up(bottom_up_start, end,
Tony Luckfc6daaf2015-06-24 16:58:09 -0700230 size, align, nid, flags);
Tang Chen79442ed2013-11-12 15:07:59 -0800231 if (ret)
232 return ret;
233
234 /*
235 * we always limit bottom-up allocation above the kernel,
236 * but top-down allocation doesn't have the limit, so
237 * retrying top-down allocation may succeed when bottom-up
238 * allocation failed.
239 *
240 * bottom-up allocation is expected to be fail very rarely,
241 * so we use WARN_ONCE() here to see the stack trace if
242 * fail happens.
243 */
244 WARN_ONCE(1, "memblock: bottom-up allocation failed, "
245 "memory hotunplug may be affected\n");
246 }
Tang Chenf7210e62013-02-22 16:33:51 -0800247
Tony Luckfc6daaf2015-06-24 16:58:09 -0700248 return __memblock_find_range_top_down(start, end, size, align, nid,
249 flags);
Tang Chenf7210e62013-02-22 16:33:51 -0800250}
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000251
Tejun Heo7bd0b0f2011-12-08 10:22:09 -0800252/**
253 * memblock_find_in_range - find free area in given range
254 * @start: start of candidate range
255 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
256 * @size: size of free area to find
257 * @align: alignment of free area to find
258 *
259 * Find @size free area aligned to @align in the specified range.
260 *
261 * RETURNS:
Tang Chen79442ed2013-11-12 15:07:59 -0800262 * Found address on success, 0 on failure.
Tejun Heo7bd0b0f2011-12-08 10:22:09 -0800263 */
264phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
265 phys_addr_t end, phys_addr_t size,
266 phys_addr_t align)
267{
Tony Lucka3f5baf2015-06-24 16:58:12 -0700268 phys_addr_t ret;
269 ulong flags = choose_memblock_flags();
270
271again:
272 ret = memblock_find_in_range_node(size, align, start, end,
273 NUMA_NO_NODE, flags);
274
275 if (!ret && (flags & MEMBLOCK_MIRROR)) {
276 pr_warn("Could not allocate %pap bytes of mirrored memory\n",
277 &size);
278 flags &= ~MEMBLOCK_MIRROR;
279 goto again;
280 }
281
282 return ret;
Tejun Heo7bd0b0f2011-12-08 10:22:09 -0800283}
284
Yinghai Lu10d06432010-07-28 15:43:02 +1000285static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000286{
Tejun Heo1440c4e2011-12-08 10:22:08 -0800287 type->total_size -= type->regions[r].size;
Tejun Heo7c0caeb2011-07-14 11:43:42 +0200288 memmove(&type->regions[r], &type->regions[r + 1],
289 (type->cnt - (r + 1)) * sizeof(type->regions[r]));
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000290 type->cnt--;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000291
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700292 /* Special case for empty arrays */
293 if (type->cnt == 0) {
Tejun Heo1440c4e2011-12-08 10:22:08 -0800294 WARN_ON(type->total_size != 0);
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700295 type->cnt = 1;
296 type->regions[0].base = 0;
297 type->regions[0].size = 0;
Tang Chen66a20752014-01-21 15:49:20 -0800298 type->regions[0].flags = 0;
Tejun Heo7c0caeb2011-07-14 11:43:42 +0200299 memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700300 }
Yinghai Lu95f72d12010-07-12 14:36:09 +1000301}
302
Philipp Hachtmann354f17e2014-01-23 15:53:24 -0800303#ifdef CONFIG_ARCH_DISCARD_MEMBLOCK
304
Yinghai Lu29f67382012-07-11 14:02:56 -0700305phys_addr_t __init_memblock get_allocated_memblock_reserved_regions_info(
306 phys_addr_t *addr)
307{
308 if (memblock.reserved.regions == memblock_reserved_init_regions)
309 return 0;
310
311 *addr = __pa(memblock.reserved.regions);
312
313 return PAGE_ALIGN(sizeof(struct memblock_region) *
314 memblock.reserved.max);
315}
316
Philipp Hachtmann5e270e22014-01-23 15:53:11 -0800317phys_addr_t __init_memblock get_allocated_memblock_memory_regions_info(
318 phys_addr_t *addr)
319{
320 if (memblock.memory.regions == memblock_memory_init_regions)
321 return 0;
322
323 *addr = __pa(memblock.memory.regions);
324
325 return PAGE_ALIGN(sizeof(struct memblock_region) *
326 memblock.memory.max);
327}
328
329#endif
330
Greg Pearson48c3b582012-06-20 12:53:05 -0700331/**
332 * memblock_double_array - double the size of the memblock regions array
333 * @type: memblock type of the regions array being doubled
334 * @new_area_start: starting address of memory range to avoid overlap with
335 * @new_area_size: size of memory range to avoid overlap with
336 *
337 * Double the size of the @type regions array. If memblock is being used to
338 * allocate memory for a new reserved regions array and there is a previously
339 * allocated memory range [@new_area_start,@new_area_start+@new_area_size]
340 * waiting to be reserved, ensure the memory used by the new array does
341 * not overlap.
342 *
343 * RETURNS:
344 * 0 on success, -1 on failure.
345 */
346static int __init_memblock memblock_double_array(struct memblock_type *type,
347 phys_addr_t new_area_start,
348 phys_addr_t new_area_size)
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700349{
350 struct memblock_region *new_array, *old_array;
Yinghai Lu29f67382012-07-11 14:02:56 -0700351 phys_addr_t old_alloc_size, new_alloc_size;
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700352 phys_addr_t old_size, new_size, addr;
353 int use_slab = slab_is_available();
Gavin Shan181eb392012-05-29 15:06:50 -0700354 int *in_slab;
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700355
356 /* We don't allow resizing until we know about the reserved regions
357 * of memory that aren't suitable for allocation
358 */
359 if (!memblock_can_resize)
360 return -1;
361
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700362 /* Calculate new doubled size */
363 old_size = type->max * sizeof(struct memblock_region);
364 new_size = old_size << 1;
Yinghai Lu29f67382012-07-11 14:02:56 -0700365 /*
366 * We need to allocated new one align to PAGE_SIZE,
367 * so we can free them completely later.
368 */
369 old_alloc_size = PAGE_ALIGN(old_size);
370 new_alloc_size = PAGE_ALIGN(new_size);
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700371
Gavin Shan181eb392012-05-29 15:06:50 -0700372 /* Retrieve the slab flag */
373 if (type == &memblock.memory)
374 in_slab = &memblock_memory_in_slab;
375 else
376 in_slab = &memblock_reserved_in_slab;
377
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700378 /* Try to find some space for it.
379 *
380 * WARNING: We assume that either slab_is_available() and we use it or
Andrew Mortonfd073832012-07-31 16:42:40 -0700381 * we use MEMBLOCK for allocations. That means that this is unsafe to
382 * use when bootmem is currently active (unless bootmem itself is
383 * implemented on top of MEMBLOCK which isn't the case yet)
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700384 *
385 * This should however not be an issue for now, as we currently only
Andrew Mortonfd073832012-07-31 16:42:40 -0700386 * call into MEMBLOCK while it's still active, or much later when slab
387 * is active for memory hotplug operations
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700388 */
389 if (use_slab) {
390 new_array = kmalloc(new_size, GFP_KERNEL);
Tejun Heo1f5026a2011-07-12 09:58:09 +0200391 addr = new_array ? __pa(new_array) : 0;
Gavin Shan4e2f0772012-05-29 15:06:50 -0700392 } else {
Greg Pearson48c3b582012-06-20 12:53:05 -0700393 /* only exclude range when trying to double reserved.regions */
394 if (type != &memblock.reserved)
395 new_area_start = new_area_size = 0;
396
397 addr = memblock_find_in_range(new_area_start + new_area_size,
398 memblock.current_limit,
Yinghai Lu29f67382012-07-11 14:02:56 -0700399 new_alloc_size, PAGE_SIZE);
Greg Pearson48c3b582012-06-20 12:53:05 -0700400 if (!addr && new_area_size)
401 addr = memblock_find_in_range(0,
Andrew Mortonfd073832012-07-31 16:42:40 -0700402 min(new_area_start, memblock.current_limit),
403 new_alloc_size, PAGE_SIZE);
Greg Pearson48c3b582012-06-20 12:53:05 -0700404
Sachin Kamat15674862012-09-04 13:55:05 +0530405 new_array = addr ? __va(addr) : NULL;
Gavin Shan4e2f0772012-05-29 15:06:50 -0700406 }
Tejun Heo1f5026a2011-07-12 09:58:09 +0200407 if (!addr) {
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700408 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
409 memblock_type_name(type), type->max, type->max * 2);
410 return -1;
411 }
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700412
Andrew Mortonfd073832012-07-31 16:42:40 -0700413 memblock_dbg("memblock: %s is doubled to %ld at [%#010llx-%#010llx]",
414 memblock_type_name(type), type->max * 2, (u64)addr,
415 (u64)addr + new_size - 1);
Yinghai Luea9e4372010-07-28 15:13:22 +1000416
Andrew Mortonfd073832012-07-31 16:42:40 -0700417 /*
418 * Found space, we now need to move the array over before we add the
419 * reserved region since it may be our reserved array itself that is
420 * full.
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700421 */
422 memcpy(new_array, type->regions, old_size);
423 memset(new_array + type->max, 0, old_size);
424 old_array = type->regions;
425 type->regions = new_array;
426 type->max <<= 1;
427
Andrew Mortonfd073832012-07-31 16:42:40 -0700428 /* Free old array. We needn't free it if the array is the static one */
Gavin Shan181eb392012-05-29 15:06:50 -0700429 if (*in_slab)
430 kfree(old_array);
431 else if (old_array != memblock_memory_init_regions &&
432 old_array != memblock_reserved_init_regions)
Yinghai Lu29f67382012-07-11 14:02:56 -0700433 memblock_free(__pa(old_array), old_alloc_size);
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700434
Andrew Mortonfd073832012-07-31 16:42:40 -0700435 /*
436 * Reserve the new array if that comes from the memblock. Otherwise, we
437 * needn't do it
Gavin Shan181eb392012-05-29 15:06:50 -0700438 */
439 if (!use_slab)
Yinghai Lu29f67382012-07-11 14:02:56 -0700440 BUG_ON(memblock_reserve(addr, new_alloc_size));
Gavin Shan181eb392012-05-29 15:06:50 -0700441
442 /* Update slab flag */
443 *in_slab = use_slab;
444
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700445 return 0;
446}
447
Tejun Heo784656f92011-07-12 11:15:55 +0200448/**
449 * memblock_merge_regions - merge neighboring compatible regions
450 * @type: memblock type to scan
451 *
452 * Scan @type and merge neighboring compatible regions.
453 */
454static void __init_memblock memblock_merge_regions(struct memblock_type *type)
455{
456 int i = 0;
457
458 /* cnt never goes below 1 */
459 while (i < type->cnt - 1) {
460 struct memblock_region *this = &type->regions[i];
461 struct memblock_region *next = &type->regions[i + 1];
462
Tejun Heo7c0caeb2011-07-14 11:43:42 +0200463 if (this->base + this->size != next->base ||
464 memblock_get_region_node(this) !=
Tang Chen66a20752014-01-21 15:49:20 -0800465 memblock_get_region_node(next) ||
466 this->flags != next->flags) {
Tejun Heo784656f92011-07-12 11:15:55 +0200467 BUG_ON(this->base + this->size > next->base);
468 i++;
469 continue;
470 }
471
472 this->size += next->size;
Lin Fengc0232ae2013-01-11 14:31:44 -0800473 /* move forward from next + 1, index of which is i + 2 */
474 memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
Tejun Heo784656f92011-07-12 11:15:55 +0200475 type->cnt--;
476 }
477}
478
479/**
480 * memblock_insert_region - insert new memblock region
Tang Chen209ff862013-04-29 15:08:41 -0700481 * @type: memblock type to insert into
482 * @idx: index for the insertion point
483 * @base: base address of the new region
484 * @size: size of the new region
485 * @nid: node id of the new region
Tang Chen66a20752014-01-21 15:49:20 -0800486 * @flags: flags of the new region
Tejun Heo784656f92011-07-12 11:15:55 +0200487 *
488 * Insert new memblock region [@base,@base+@size) into @type at @idx.
489 * @type must already have extra room to accomodate the new region.
490 */
491static void __init_memblock memblock_insert_region(struct memblock_type *type,
492 int idx, phys_addr_t base,
Tang Chen66a20752014-01-21 15:49:20 -0800493 phys_addr_t size,
494 int nid, unsigned long flags)
Tejun Heo784656f92011-07-12 11:15:55 +0200495{
496 struct memblock_region *rgn = &type->regions[idx];
497
498 BUG_ON(type->cnt >= type->max);
499 memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
500 rgn->base = base;
501 rgn->size = size;
Tang Chen66a20752014-01-21 15:49:20 -0800502 rgn->flags = flags;
Tejun Heo7c0caeb2011-07-14 11:43:42 +0200503 memblock_set_region_node(rgn, nid);
Tejun Heo784656f92011-07-12 11:15:55 +0200504 type->cnt++;
Tejun Heo1440c4e2011-12-08 10:22:08 -0800505 type->total_size += size;
Tejun Heo784656f92011-07-12 11:15:55 +0200506}
507
508/**
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100509 * memblock_add_range - add new memblock region
Tejun Heo784656f92011-07-12 11:15:55 +0200510 * @type: memblock type to add new region into
511 * @base: base address of the new region
512 * @size: size of the new region
Tejun Heo7fb0bc32011-12-08 10:22:08 -0800513 * @nid: nid of the new region
Tang Chen66a20752014-01-21 15:49:20 -0800514 * @flags: flags of the new region
Tejun Heo784656f92011-07-12 11:15:55 +0200515 *
516 * Add new memblock region [@base,@base+@size) into @type. The new region
517 * is allowed to overlap with existing ones - overlaps don't affect already
518 * existing regions. @type is guaranteed to be minimal (all neighbouring
519 * compatible regions are merged) after the addition.
520 *
521 * RETURNS:
522 * 0 on success, -errno on failure.
523 */
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100524int __init_memblock memblock_add_range(struct memblock_type *type,
Tang Chen66a20752014-01-21 15:49:20 -0800525 phys_addr_t base, phys_addr_t size,
526 int nid, unsigned long flags)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000527{
Tejun Heo784656f92011-07-12 11:15:55 +0200528 bool insert = false;
Tejun Heoeb18f1b2011-12-08 10:22:07 -0800529 phys_addr_t obase = base;
530 phys_addr_t end = base + memblock_cap_size(base, &size);
Tejun Heo784656f92011-07-12 11:15:55 +0200531 int i, nr_new;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000532
Tejun Heob3dc6272012-04-20 08:31:34 -0700533 if (!size)
534 return 0;
535
Tejun Heo784656f92011-07-12 11:15:55 +0200536 /* special case for empty array */
537 if (type->regions[0].size == 0) {
Tejun Heo1440c4e2011-12-08 10:22:08 -0800538 WARN_ON(type->cnt != 1 || type->total_size);
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000539 type->regions[0].base = base;
540 type->regions[0].size = size;
Tang Chen66a20752014-01-21 15:49:20 -0800541 type->regions[0].flags = flags;
Tejun Heo7fb0bc32011-12-08 10:22:08 -0800542 memblock_set_region_node(&type->regions[0], nid);
Tejun Heo1440c4e2011-12-08 10:22:08 -0800543 type->total_size = size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000544 return 0;
545 }
Tejun Heo784656f92011-07-12 11:15:55 +0200546repeat:
547 /*
548 * The following is executed twice. Once with %false @insert and
549 * then with %true. The first counts the number of regions needed
550 * to accomodate the new area. The second actually inserts them.
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700551 */
Tejun Heo784656f92011-07-12 11:15:55 +0200552 base = obase;
553 nr_new = 0;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000554
Tejun Heo784656f92011-07-12 11:15:55 +0200555 for (i = 0; i < type->cnt; i++) {
556 struct memblock_region *rgn = &type->regions[i];
557 phys_addr_t rbase = rgn->base;
558 phys_addr_t rend = rbase + rgn->size;
559
560 if (rbase >= end)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000561 break;
Tejun Heo784656f92011-07-12 11:15:55 +0200562 if (rend <= base)
563 continue;
564 /*
565 * @rgn overlaps. If it separates the lower part of new
566 * area, insert that portion.
567 */
568 if (rbase > base) {
569 nr_new++;
570 if (insert)
571 memblock_insert_region(type, i++, base,
Tang Chen66a20752014-01-21 15:49:20 -0800572 rbase - base, nid,
573 flags);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000574 }
Tejun Heo784656f92011-07-12 11:15:55 +0200575 /* area below @rend is dealt with, forget about it */
576 base = min(rend, end);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000577 }
Yinghai Lu95f72d12010-07-12 14:36:09 +1000578
Tejun Heo784656f92011-07-12 11:15:55 +0200579 /* insert the remaining portion */
580 if (base < end) {
581 nr_new++;
582 if (insert)
Tang Chen66a20752014-01-21 15:49:20 -0800583 memblock_insert_region(type, i, base, end - base,
584 nid, flags);
Tejun Heo784656f92011-07-12 11:15:55 +0200585 }
586
587 /*
588 * If this was the first round, resize array and repeat for actual
589 * insertions; otherwise, merge and return.
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700590 */
Tejun Heo784656f92011-07-12 11:15:55 +0200591 if (!insert) {
592 while (type->cnt + nr_new > type->max)
Greg Pearson48c3b582012-06-20 12:53:05 -0700593 if (memblock_double_array(type, obase, size) < 0)
Tejun Heo784656f92011-07-12 11:15:55 +0200594 return -ENOMEM;
595 insert = true;
596 goto repeat;
597 } else {
598 memblock_merge_regions(type);
599 return 0;
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700600 }
Yinghai Lu95f72d12010-07-12 14:36:09 +1000601}
602
Tejun Heo7fb0bc32011-12-08 10:22:08 -0800603int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
604 int nid)
605{
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100606 return memblock_add_range(&memblock.memory, base, size, nid, 0);
Tejun Heo7fb0bc32011-12-08 10:22:08 -0800607}
608
Alexander Kuleshov6a4055b2015-04-15 16:14:44 -0700609static int __init_memblock memblock_add_region(phys_addr_t base,
610 phys_addr_t size,
611 int nid,
612 unsigned long flags)
613{
614 struct memblock_type *_rgn = &memblock.memory;
615
616 memblock_dbg("memblock_add: [%#016llx-%#016llx] flags %#02lx %pF\n",
617 (unsigned long long)base,
618 (unsigned long long)base + size - 1,
619 flags, (void *)_RET_IP_);
620
621 return memblock_add_range(_rgn, base, size, nid, flags);
622}
623
Tejun Heo581adcb2011-12-08 10:22:06 -0800624int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000625{
Alexander Kuleshov6a4055b2015-04-15 16:14:44 -0700626 return memblock_add_region(base, size, MAX_NUMNODES, 0);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000627}
628
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800629/**
630 * memblock_isolate_range - isolate given range into disjoint memblocks
631 * @type: memblock type to isolate range for
632 * @base: base of range to isolate
633 * @size: size of range to isolate
634 * @start_rgn: out parameter for the start of isolated region
635 * @end_rgn: out parameter for the end of isolated region
636 *
637 * Walk @type and ensure that regions don't cross the boundaries defined by
638 * [@base,@base+@size). Crossing regions are split at the boundaries,
639 * which may create at most two more regions. The index of the first
640 * region inside the range is returned in *@start_rgn and end in *@end_rgn.
641 *
642 * RETURNS:
643 * 0 on success, -errno on failure.
644 */
645static int __init_memblock memblock_isolate_range(struct memblock_type *type,
646 phys_addr_t base, phys_addr_t size,
647 int *start_rgn, int *end_rgn)
648{
Tejun Heoeb18f1b2011-12-08 10:22:07 -0800649 phys_addr_t end = base + memblock_cap_size(base, &size);
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800650 int i;
651
652 *start_rgn = *end_rgn = 0;
653
Tejun Heob3dc6272012-04-20 08:31:34 -0700654 if (!size)
655 return 0;
656
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800657 /* we'll create at most two more regions */
658 while (type->cnt + 2 > type->max)
Greg Pearson48c3b582012-06-20 12:53:05 -0700659 if (memblock_double_array(type, base, size) < 0)
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800660 return -ENOMEM;
661
662 for (i = 0; i < type->cnt; i++) {
663 struct memblock_region *rgn = &type->regions[i];
664 phys_addr_t rbase = rgn->base;
665 phys_addr_t rend = rbase + rgn->size;
666
667 if (rbase >= end)
668 break;
669 if (rend <= base)
670 continue;
671
672 if (rbase < base) {
673 /*
674 * @rgn intersects from below. Split and continue
675 * to process the next region - the new top half.
676 */
677 rgn->base = base;
Tejun Heo1440c4e2011-12-08 10:22:08 -0800678 rgn->size -= base - rbase;
679 type->total_size -= base - rbase;
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800680 memblock_insert_region(type, i, rbase, base - rbase,
Tang Chen66a20752014-01-21 15:49:20 -0800681 memblock_get_region_node(rgn),
682 rgn->flags);
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800683 } else if (rend > end) {
684 /*
685 * @rgn intersects from above. Split and redo the
686 * current region - the new bottom half.
687 */
688 rgn->base = end;
Tejun Heo1440c4e2011-12-08 10:22:08 -0800689 rgn->size -= end - rbase;
690 type->total_size -= end - rbase;
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800691 memblock_insert_region(type, i--, rbase, end - rbase,
Tang Chen66a20752014-01-21 15:49:20 -0800692 memblock_get_region_node(rgn),
693 rgn->flags);
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800694 } else {
695 /* @rgn is fully contained, record it */
696 if (!*end_rgn)
697 *start_rgn = i;
698 *end_rgn = i + 1;
699 }
700 }
701
702 return 0;
703}
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800704
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100705int __init_memblock memblock_remove_range(struct memblock_type *type,
706 phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000707{
Tejun Heo71936182011-12-08 10:22:07 -0800708 int start_rgn, end_rgn;
709 int i, ret;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000710
Tejun Heo71936182011-12-08 10:22:07 -0800711 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
712 if (ret)
713 return ret;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000714
Tejun Heo71936182011-12-08 10:22:07 -0800715 for (i = end_rgn - 1; i >= start_rgn; i--)
716 memblock_remove_region(type, i);
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700717 return 0;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000718}
719
Tejun Heo581adcb2011-12-08 10:22:06 -0800720int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000721{
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100722 return memblock_remove_range(&memblock.memory, base, size);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000723}
724
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100725
Tejun Heo581adcb2011-12-08 10:22:06 -0800726int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000727{
Tejun Heo24aa0782011-07-12 11:16:06 +0200728 memblock_dbg(" memblock_free: [%#016llx-%#016llx] %pF\n",
H. Peter Anvina1504392011-07-14 11:57:10 -0700729 (unsigned long long)base,
Grygorii Strashko931d13f2014-01-21 15:49:17 -0800730 (unsigned long long)base + size - 1,
H. Peter Anvina1504392011-07-14 11:57:10 -0700731 (void *)_RET_IP_);
Tejun Heo24aa0782011-07-12 11:16:06 +0200732
Catalin Marinasaedf95e2014-06-06 14:38:20 -0700733 kmemleak_free_part(__va(base), size);
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100734 return memblock_remove_range(&memblock.reserved, base, size);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000735}
736
Tang Chen66a20752014-01-21 15:49:20 -0800737static int __init_memblock memblock_reserve_region(phys_addr_t base,
738 phys_addr_t size,
739 int nid,
740 unsigned long flags)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000741{
Baoquan He7fc825b2015-04-14 15:44:48 -0700742 struct memblock_type *type = &memblock.reserved;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000743
Tang Chen66a20752014-01-21 15:49:20 -0800744 memblock_dbg("memblock_reserve: [%#016llx-%#016llx] flags %#02lx %pF\n",
H. Peter Anvina1504392011-07-14 11:57:10 -0700745 (unsigned long long)base,
Grygorii Strashko931d13f2014-01-21 15:49:17 -0800746 (unsigned long long)base + size - 1,
Tang Chen66a20752014-01-21 15:49:20 -0800747 flags, (void *)_RET_IP_);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000748
Baoquan He7fc825b2015-04-14 15:44:48 -0700749 return memblock_add_range(type, base, size, nid, flags);
Tang Chen66a20752014-01-21 15:49:20 -0800750}
751
752int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
753{
754 return memblock_reserve_region(base, size, MAX_NUMNODES, 0);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000755}
756
Tejun Heo35fd0802011-07-12 11:15:59 +0200757/**
Tang Chen66b16ed2014-01-21 15:49:23 -0800758 *
Tony Luck4308ce12014-12-12 16:54:59 -0800759 * This function isolates region [@base, @base + @size), and sets/clears flag
Tang Chen66b16ed2014-01-21 15:49:23 -0800760 *
761 * Return 0 on succees, -errno on failure.
762 */
Tony Luck4308ce12014-12-12 16:54:59 -0800763static int __init_memblock memblock_setclr_flag(phys_addr_t base,
764 phys_addr_t size, int set, int flag)
Tang Chen66b16ed2014-01-21 15:49:23 -0800765{
766 struct memblock_type *type = &memblock.memory;
767 int i, ret, start_rgn, end_rgn;
768
769 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
770 if (ret)
771 return ret;
772
773 for (i = start_rgn; i < end_rgn; i++)
Tony Luck4308ce12014-12-12 16:54:59 -0800774 if (set)
775 memblock_set_region_flags(&type->regions[i], flag);
776 else
777 memblock_clear_region_flags(&type->regions[i], flag);
Tang Chen66b16ed2014-01-21 15:49:23 -0800778
779 memblock_merge_regions(type);
780 return 0;
781}
782
783/**
Tony Luck4308ce12014-12-12 16:54:59 -0800784 * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
785 * @base: the base phys addr of the region
786 * @size: the size of the region
787 *
788 * Return 0 on succees, -errno on failure.
789 */
790int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
791{
792 return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG);
793}
794
795/**
Tang Chen66b16ed2014-01-21 15:49:23 -0800796 * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
797 * @base: the base phys addr of the region
798 * @size: the size of the region
799 *
Tang Chen66b16ed2014-01-21 15:49:23 -0800800 * Return 0 on succees, -errno on failure.
801 */
802int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
803{
Tony Luck4308ce12014-12-12 16:54:59 -0800804 return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG);
Tang Chen66b16ed2014-01-21 15:49:23 -0800805}
806
807/**
Tony Lucka3f5baf2015-06-24 16:58:12 -0700808 * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR.
809 * @base: the base phys addr of the region
810 * @size: the size of the region
811 *
812 * Return 0 on succees, -errno on failure.
813 */
814int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
815{
816 system_has_some_mirror = true;
817
818 return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR);
819}
820
821
822/**
Robin Holt8e7a7f82015-06-30 14:56:41 -0700823 * __next_reserved_mem_region - next function for for_each_reserved_region()
824 * @idx: pointer to u64 loop variable
825 * @out_start: ptr to phys_addr_t for start address of the region, can be %NULL
826 * @out_end: ptr to phys_addr_t for end address of the region, can be %NULL
827 *
828 * Iterate over all reserved memory regions.
829 */
830void __init_memblock __next_reserved_mem_region(u64 *idx,
831 phys_addr_t *out_start,
832 phys_addr_t *out_end)
833{
834 struct memblock_type *rsv = &memblock.reserved;
835
836 if (*idx >= 0 && *idx < rsv->cnt) {
837 struct memblock_region *r = &rsv->regions[*idx];
838 phys_addr_t base = r->base;
839 phys_addr_t size = r->size;
840
841 if (out_start)
842 *out_start = base;
843 if (out_end)
844 *out_end = base + size - 1;
845
846 *idx += 1;
847 return;
848 }
849
850 /* signal end of iteration */
851 *idx = ULLONG_MAX;
852}
853
854/**
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100855 * __next__mem_range - next function for for_each_free_mem_range() etc.
Tejun Heo35fd0802011-07-12 11:15:59 +0200856 * @idx: pointer to u64 loop variable
Grygorii Strashkob1154232014-01-21 15:50:16 -0800857 * @nid: node selector, %NUMA_NO_NODE for all nodes
Tony Luckfc6daaf2015-06-24 16:58:09 -0700858 * @flags: pick from blocks based on memory attributes
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100859 * @type_a: pointer to memblock_type from where the range is taken
860 * @type_b: pointer to memblock_type which excludes memory from being taken
Wanpeng Lidad75572012-06-20 12:53:01 -0700861 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
862 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
863 * @out_nid: ptr to int for nid of the range, can be %NULL
Tejun Heo35fd0802011-07-12 11:15:59 +0200864 *
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100865 * Find the first area from *@idx which matches @nid, fill the out
Tejun Heo35fd0802011-07-12 11:15:59 +0200866 * parameters, and update *@idx for the next iteration. The lower 32bit of
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100867 * *@idx contains index into type_a and the upper 32bit indexes the
868 * areas before each region in type_b. For example, if type_b regions
Tejun Heo35fd0802011-07-12 11:15:59 +0200869 * look like the following,
870 *
871 * 0:[0-16), 1:[32-48), 2:[128-130)
872 *
873 * The upper 32bit indexes the following regions.
874 *
875 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
876 *
877 * As both region arrays are sorted, the function advances the two indices
878 * in lockstep and returns each intersection.
879 */
Tony Luckfc6daaf2015-06-24 16:58:09 -0700880void __init_memblock __next_mem_range(u64 *idx, int nid, ulong flags,
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100881 struct memblock_type *type_a,
882 struct memblock_type *type_b,
883 phys_addr_t *out_start,
884 phys_addr_t *out_end, int *out_nid)
Tejun Heo35fd0802011-07-12 11:15:59 +0200885{
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100886 int idx_a = *idx & 0xffffffff;
887 int idx_b = *idx >> 32;
Grygorii Strashkob1154232014-01-21 15:50:16 -0800888
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100889 if (WARN_ONCE(nid == MAX_NUMNODES,
890 "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
Grygorii Strashko560dca272014-01-21 15:50:55 -0800891 nid = NUMA_NO_NODE;
Tejun Heo35fd0802011-07-12 11:15:59 +0200892
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100893 for (; idx_a < type_a->cnt; idx_a++) {
894 struct memblock_region *m = &type_a->regions[idx_a];
895
Tejun Heo35fd0802011-07-12 11:15:59 +0200896 phys_addr_t m_start = m->base;
897 phys_addr_t m_end = m->base + m->size;
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100898 int m_nid = memblock_get_region_node(m);
Tejun Heo35fd0802011-07-12 11:15:59 +0200899
900 /* only memory regions are associated with nodes, check it */
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100901 if (nid != NUMA_NO_NODE && nid != m_nid)
Tejun Heo35fd0802011-07-12 11:15:59 +0200902 continue;
903
Xishi Qiu0a313a92014-09-09 14:50:46 -0700904 /* skip hotpluggable memory regions if needed */
905 if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
906 continue;
907
Tony Lucka3f5baf2015-06-24 16:58:12 -0700908 /* if we want mirror memory skip non-mirror memory regions */
909 if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
910 continue;
911
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100912 if (!type_b) {
913 if (out_start)
914 *out_start = m_start;
915 if (out_end)
916 *out_end = m_end;
917 if (out_nid)
918 *out_nid = m_nid;
919 idx_a++;
920 *idx = (u32)idx_a | (u64)idx_b << 32;
921 return;
922 }
Tejun Heo35fd0802011-07-12 11:15:59 +0200923
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100924 /* scan areas before each reservation */
925 for (; idx_b < type_b->cnt + 1; idx_b++) {
926 struct memblock_region *r;
927 phys_addr_t r_start;
928 phys_addr_t r_end;
929
930 r = &type_b->regions[idx_b];
931 r_start = idx_b ? r[-1].base + r[-1].size : 0;
932 r_end = idx_b < type_b->cnt ?
933 r->base : ULLONG_MAX;
934
935 /*
936 * if idx_b advanced past idx_a,
937 * break out to advance idx_a
938 */
Tejun Heo35fd0802011-07-12 11:15:59 +0200939 if (r_start >= m_end)
940 break;
941 /* if the two regions intersect, we're done */
942 if (m_start < r_end) {
943 if (out_start)
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100944 *out_start =
945 max(m_start, r_start);
Tejun Heo35fd0802011-07-12 11:15:59 +0200946 if (out_end)
947 *out_end = min(m_end, r_end);
948 if (out_nid)
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100949 *out_nid = m_nid;
Tejun Heo35fd0802011-07-12 11:15:59 +0200950 /*
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100951 * The region which ends first is
952 * advanced for the next iteration.
Tejun Heo35fd0802011-07-12 11:15:59 +0200953 */
954 if (m_end <= r_end)
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100955 idx_a++;
Tejun Heo35fd0802011-07-12 11:15:59 +0200956 else
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100957 idx_b++;
958 *idx = (u32)idx_a | (u64)idx_b << 32;
Tejun Heo35fd0802011-07-12 11:15:59 +0200959 return;
960 }
961 }
962 }
963
964 /* signal end of iteration */
965 *idx = ULLONG_MAX;
966}
967
Tejun Heo7bd0b0f2011-12-08 10:22:09 -0800968/**
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100969 * __next_mem_range_rev - generic next function for for_each_*_range_rev()
970 *
971 * Finds the next range from type_a which is not marked as unsuitable
972 * in type_b.
973 *
Tejun Heo7bd0b0f2011-12-08 10:22:09 -0800974 * @idx: pointer to u64 loop variable
Grygorii Strashkob1154232014-01-21 15:50:16 -0800975 * @nid: nid: node selector, %NUMA_NO_NODE for all nodes
Tony Luckfc6daaf2015-06-24 16:58:09 -0700976 * @flags: pick from blocks based on memory attributes
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100977 * @type_a: pointer to memblock_type from where the range is taken
978 * @type_b: pointer to memblock_type which excludes memory from being taken
Wanpeng Lidad75572012-06-20 12:53:01 -0700979 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
980 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
981 * @out_nid: ptr to int for nid of the range, can be %NULL
Tejun Heo7bd0b0f2011-12-08 10:22:09 -0800982 *
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100983 * Reverse of __next_mem_range().
Tejun Heo7bd0b0f2011-12-08 10:22:09 -0800984 */
Tony Luckfc6daaf2015-06-24 16:58:09 -0700985void __init_memblock __next_mem_range_rev(u64 *idx, int nid, ulong flags,
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100986 struct memblock_type *type_a,
987 struct memblock_type *type_b,
988 phys_addr_t *out_start,
989 phys_addr_t *out_end, int *out_nid)
Tejun Heo7bd0b0f2011-12-08 10:22:09 -0800990{
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100991 int idx_a = *idx & 0xffffffff;
992 int idx_b = *idx >> 32;
Grygorii Strashkob1154232014-01-21 15:50:16 -0800993
Grygorii Strashko560dca272014-01-21 15:50:55 -0800994 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
995 nid = NUMA_NO_NODE;
Tejun Heo7bd0b0f2011-12-08 10:22:09 -0800996
997 if (*idx == (u64)ULLONG_MAX) {
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100998 idx_a = type_a->cnt - 1;
999 idx_b = type_b->cnt;
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001000 }
1001
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001002 for (; idx_a >= 0; idx_a--) {
1003 struct memblock_region *m = &type_a->regions[idx_a];
1004
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001005 phys_addr_t m_start = m->base;
1006 phys_addr_t m_end = m->base + m->size;
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001007 int m_nid = memblock_get_region_node(m);
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001008
1009 /* only memory regions are associated with nodes, check it */
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001010 if (nid != NUMA_NO_NODE && nid != m_nid)
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001011 continue;
1012
Tang Chen55ac5902014-01-21 15:49:35 -08001013 /* skip hotpluggable memory regions if needed */
1014 if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
1015 continue;
1016
Tony Lucka3f5baf2015-06-24 16:58:12 -07001017 /* if we want mirror memory skip non-mirror memory regions */
1018 if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
1019 continue;
1020
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001021 if (!type_b) {
1022 if (out_start)
1023 *out_start = m_start;
1024 if (out_end)
1025 *out_end = m_end;
1026 if (out_nid)
1027 *out_nid = m_nid;
1028 idx_a++;
1029 *idx = (u32)idx_a | (u64)idx_b << 32;
1030 return;
1031 }
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001032
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001033 /* scan areas before each reservation */
1034 for (; idx_b >= 0; idx_b--) {
1035 struct memblock_region *r;
1036 phys_addr_t r_start;
1037 phys_addr_t r_end;
1038
1039 r = &type_b->regions[idx_b];
1040 r_start = idx_b ? r[-1].base + r[-1].size : 0;
1041 r_end = idx_b < type_b->cnt ?
1042 r->base : ULLONG_MAX;
1043 /*
1044 * if idx_b advanced past idx_a,
1045 * break out to advance idx_a
1046 */
1047
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001048 if (r_end <= m_start)
1049 break;
1050 /* if the two regions intersect, we're done */
1051 if (m_end > r_start) {
1052 if (out_start)
1053 *out_start = max(m_start, r_start);
1054 if (out_end)
1055 *out_end = min(m_end, r_end);
1056 if (out_nid)
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001057 *out_nid = m_nid;
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001058 if (m_start >= r_start)
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001059 idx_a--;
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001060 else
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001061 idx_b--;
1062 *idx = (u32)idx_a | (u64)idx_b << 32;
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001063 return;
1064 }
1065 }
1066 }
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001067 /* signal end of iteration */
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001068 *idx = ULLONG_MAX;
1069}
1070
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001071#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1072/*
1073 * Common iterator interface used to define for_each_mem_range().
1074 */
1075void __init_memblock __next_mem_pfn_range(int *idx, int nid,
1076 unsigned long *out_start_pfn,
1077 unsigned long *out_end_pfn, int *out_nid)
1078{
1079 struct memblock_type *type = &memblock.memory;
1080 struct memblock_region *r;
1081
1082 while (++*idx < type->cnt) {
1083 r = &type->regions[*idx];
1084
1085 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
1086 continue;
1087 if (nid == MAX_NUMNODES || nid == r->nid)
1088 break;
1089 }
1090 if (*idx >= type->cnt) {
1091 *idx = -1;
1092 return;
1093 }
1094
1095 if (out_start_pfn)
1096 *out_start_pfn = PFN_UP(r->base);
1097 if (out_end_pfn)
1098 *out_end_pfn = PFN_DOWN(r->base + r->size);
1099 if (out_nid)
1100 *out_nid = r->nid;
1101}
1102
1103/**
1104 * memblock_set_node - set node ID on memblock regions
1105 * @base: base of area to set node ID for
1106 * @size: size of area to set node ID for
Tang Chene7e8de52014-01-21 15:49:26 -08001107 * @type: memblock type to set node ID for
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001108 * @nid: node ID to set
1109 *
Tang Chene7e8de52014-01-21 15:49:26 -08001110 * Set the nid of memblock @type regions in [@base,@base+@size) to @nid.
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001111 * Regions which cross the area boundaries are split as necessary.
1112 *
1113 * RETURNS:
1114 * 0 on success, -errno on failure.
1115 */
1116int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
Tang Chene7e8de52014-01-21 15:49:26 -08001117 struct memblock_type *type, int nid)
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001118{
Tejun Heo6a9ceb32011-12-08 10:22:07 -08001119 int start_rgn, end_rgn;
1120 int i, ret;
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001121
Tejun Heo6a9ceb32011-12-08 10:22:07 -08001122 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
1123 if (ret)
1124 return ret;
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001125
Tejun Heo6a9ceb32011-12-08 10:22:07 -08001126 for (i = start_rgn; i < end_rgn; i++)
Wanpeng Lie9d24ad2012-10-08 16:32:21 -07001127 memblock_set_region_node(&type->regions[i], nid);
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001128
1129 memblock_merge_regions(type);
1130 return 0;
1131}
1132#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
1133
Akinobu Mita2bfc2862014-06-04 16:06:53 -07001134static phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
1135 phys_addr_t align, phys_addr_t start,
Tony Luckfc6daaf2015-06-24 16:58:09 -07001136 phys_addr_t end, int nid, ulong flags)
Yinghai Lu95f72d12010-07-12 14:36:09 +10001137{
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001138 phys_addr_t found;
Yinghai Lu95f72d12010-07-12 14:36:09 +10001139
Grygorii Strashko79f40fa2014-01-21 15:50:12 -08001140 if (!align)
1141 align = SMP_CACHE_BYTES;
Vineet Gupta94f3d3a2013-04-29 15:06:15 -07001142
Tony Luckfc6daaf2015-06-24 16:58:09 -07001143 found = memblock_find_in_range_node(size, align, start, end, nid,
1144 flags);
Catalin Marinasaedf95e2014-06-06 14:38:20 -07001145 if (found && !memblock_reserve(found, size)) {
1146 /*
1147 * The min_count is set to 0 so that memblock allocations are
1148 * never reported as leaks.
1149 */
1150 kmemleak_alloc(__va(found), size, 0, 0);
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001151 return found;
Catalin Marinasaedf95e2014-06-06 14:38:20 -07001152 }
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001153 return 0;
Yinghai Lu95f72d12010-07-12 14:36:09 +10001154}
1155
Akinobu Mita2bfc2862014-06-04 16:06:53 -07001156phys_addr_t __init memblock_alloc_range(phys_addr_t size, phys_addr_t align,
Tony Luckfc6daaf2015-06-24 16:58:09 -07001157 phys_addr_t start, phys_addr_t end,
1158 ulong flags)
Akinobu Mita2bfc2862014-06-04 16:06:53 -07001159{
Tony Luckfc6daaf2015-06-24 16:58:09 -07001160 return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE,
1161 flags);
Akinobu Mita2bfc2862014-06-04 16:06:53 -07001162}
1163
1164static phys_addr_t __init memblock_alloc_base_nid(phys_addr_t size,
1165 phys_addr_t align, phys_addr_t max_addr,
Tony Luckfc6daaf2015-06-24 16:58:09 -07001166 int nid, ulong flags)
Akinobu Mita2bfc2862014-06-04 16:06:53 -07001167{
Tony Luckfc6daaf2015-06-24 16:58:09 -07001168 return memblock_alloc_range_nid(size, align, 0, max_addr, nid, flags);
Akinobu Mita2bfc2862014-06-04 16:06:53 -07001169}
1170
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001171phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
1172{
Tony Lucka3f5baf2015-06-24 16:58:12 -07001173 ulong flags = choose_memblock_flags();
1174 phys_addr_t ret;
1175
1176again:
1177 ret = memblock_alloc_base_nid(size, align, MEMBLOCK_ALLOC_ACCESSIBLE,
1178 nid, flags);
1179
1180 if (!ret && (flags & MEMBLOCK_MIRROR)) {
1181 flags &= ~MEMBLOCK_MIRROR;
1182 goto again;
1183 }
1184 return ret;
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001185}
1186
1187phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1188{
Tony Luckfc6daaf2015-06-24 16:58:09 -07001189 return memblock_alloc_base_nid(size, align, max_addr, NUMA_NO_NODE,
1190 MEMBLOCK_NONE);
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001191}
1192
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001193phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
Yinghai Lu95f72d12010-07-12 14:36:09 +10001194{
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001195 phys_addr_t alloc;
1196
1197 alloc = __memblock_alloc_base(size, align, max_addr);
1198
1199 if (alloc == 0)
1200 panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n",
1201 (unsigned long long) size, (unsigned long long) max_addr);
1202
1203 return alloc;
Yinghai Lu95f72d12010-07-12 14:36:09 +10001204}
1205
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001206phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
Yinghai Lu95f72d12010-07-12 14:36:09 +10001207{
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001208 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
Yinghai Lu95f72d12010-07-12 14:36:09 +10001209}
1210
Benjamin Herrenschmidt9d1e2492010-07-06 15:39:17 -07001211phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1212{
1213 phys_addr_t res = memblock_alloc_nid(size, align, nid);
1214
1215 if (res)
1216 return res;
Tejun Heo15fb0972011-07-12 09:58:07 +02001217 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
Yinghai Lu95f72d12010-07-12 14:36:09 +10001218}
1219
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001220/**
1221 * memblock_virt_alloc_internal - allocate boot memory block
1222 * @size: size of memory block to be allocated in bytes
1223 * @align: alignment of the region and block's size
1224 * @min_addr: the lower bound of the memory region to allocate (phys address)
1225 * @max_addr: the upper bound of the memory region to allocate (phys address)
1226 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1227 *
1228 * The @min_addr limit is dropped if it can not be satisfied and the allocation
1229 * will fall back to memory below @min_addr. Also, allocation may fall back
1230 * to any node in the system if the specified node can not
1231 * hold the requested memory.
1232 *
1233 * The allocation is performed from memory region limited by
1234 * memblock.current_limit if @max_addr == %BOOTMEM_ALLOC_ACCESSIBLE.
1235 *
1236 * The memory block is aligned on SMP_CACHE_BYTES if @align == 0.
1237 *
1238 * The phys address of allocated boot memory block is converted to virtual and
1239 * allocated memory is reset to 0.
1240 *
1241 * In addition, function sets the min_count to 0 using kmemleak_alloc for
1242 * allocated boot memory block, so that it is never reported as leaks.
1243 *
1244 * RETURNS:
1245 * Virtual address of allocated memory block on success, NULL on failure.
1246 */
1247static void * __init memblock_virt_alloc_internal(
1248 phys_addr_t size, phys_addr_t align,
1249 phys_addr_t min_addr, phys_addr_t max_addr,
1250 int nid)
1251{
1252 phys_addr_t alloc;
1253 void *ptr;
Tony Lucka3f5baf2015-06-24 16:58:12 -07001254 ulong flags = choose_memblock_flags();
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001255
Grygorii Strashko560dca272014-01-21 15:50:55 -08001256 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1257 nid = NUMA_NO_NODE;
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001258
1259 /*
1260 * Detect any accidental use of these APIs after slab is ready, as at
1261 * this moment memblock may be deinitialized already and its
1262 * internal data may be destroyed (after execution of free_all_bootmem)
1263 */
1264 if (WARN_ON_ONCE(slab_is_available()))
1265 return kzalloc_node(size, GFP_NOWAIT, nid);
1266
1267 if (!align)
1268 align = SMP_CACHE_BYTES;
1269
Yinghai Luf544e142014-01-29 14:05:52 -08001270 if (max_addr > memblock.current_limit)
1271 max_addr = memblock.current_limit;
1272
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001273again:
1274 alloc = memblock_find_in_range_node(size, align, min_addr, max_addr,
Tony Lucka3f5baf2015-06-24 16:58:12 -07001275 nid, flags);
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001276 if (alloc)
1277 goto done;
1278
1279 if (nid != NUMA_NO_NODE) {
1280 alloc = memblock_find_in_range_node(size, align, min_addr,
Tony Luckfc6daaf2015-06-24 16:58:09 -07001281 max_addr, NUMA_NO_NODE,
Tony Lucka3f5baf2015-06-24 16:58:12 -07001282 flags);
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001283 if (alloc)
1284 goto done;
1285 }
1286
1287 if (min_addr) {
1288 min_addr = 0;
1289 goto again;
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001290 }
1291
Tony Lucka3f5baf2015-06-24 16:58:12 -07001292 if (flags & MEMBLOCK_MIRROR) {
1293 flags &= ~MEMBLOCK_MIRROR;
1294 pr_warn("Could not allocate %pap bytes of mirrored memory\n",
1295 &size);
1296 goto again;
1297 }
1298
1299 return NULL;
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001300done:
1301 memblock_reserve(alloc, size);
1302 ptr = phys_to_virt(alloc);
1303 memset(ptr, 0, size);
1304
1305 /*
1306 * The min_count is set to 0 so that bootmem allocated blocks
1307 * are never reported as leaks. This is because many of these blocks
1308 * are only referred via the physical address which is not
1309 * looked up by kmemleak.
1310 */
1311 kmemleak_alloc(ptr, size, 0, 0);
1312
1313 return ptr;
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001314}
1315
1316/**
1317 * memblock_virt_alloc_try_nid_nopanic - allocate boot memory block
1318 * @size: size of memory block to be allocated in bytes
1319 * @align: alignment of the region and block's size
1320 * @min_addr: the lower bound of the memory region from where the allocation
1321 * is preferred (phys address)
1322 * @max_addr: the upper bound of the memory region from where the allocation
1323 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1324 * allocate only from memory limited by memblock.current_limit value
1325 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1326 *
1327 * Public version of _memblock_virt_alloc_try_nid_nopanic() which provides
1328 * additional debug information (including caller info), if enabled.
1329 *
1330 * RETURNS:
1331 * Virtual address of allocated memory block on success, NULL on failure.
1332 */
1333void * __init memblock_virt_alloc_try_nid_nopanic(
1334 phys_addr_t size, phys_addr_t align,
1335 phys_addr_t min_addr, phys_addr_t max_addr,
1336 int nid)
1337{
1338 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1339 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1340 (u64)max_addr, (void *)_RET_IP_);
1341 return memblock_virt_alloc_internal(size, align, min_addr,
1342 max_addr, nid);
1343}
1344
1345/**
1346 * memblock_virt_alloc_try_nid - allocate boot memory block with panicking
1347 * @size: size of memory block to be allocated in bytes
1348 * @align: alignment of the region and block's size
1349 * @min_addr: the lower bound of the memory region from where the allocation
1350 * is preferred (phys address)
1351 * @max_addr: the upper bound of the memory region from where the allocation
1352 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1353 * allocate only from memory limited by memblock.current_limit value
1354 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1355 *
1356 * Public panicking version of _memblock_virt_alloc_try_nid_nopanic()
1357 * which provides debug information (including caller info), if enabled,
1358 * and panics if the request can not be satisfied.
1359 *
1360 * RETURNS:
1361 * Virtual address of allocated memory block on success, NULL on failure.
1362 */
1363void * __init memblock_virt_alloc_try_nid(
1364 phys_addr_t size, phys_addr_t align,
1365 phys_addr_t min_addr, phys_addr_t max_addr,
1366 int nid)
1367{
1368 void *ptr;
1369
1370 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1371 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1372 (u64)max_addr, (void *)_RET_IP_);
1373 ptr = memblock_virt_alloc_internal(size, align,
1374 min_addr, max_addr, nid);
1375 if (ptr)
1376 return ptr;
1377
1378 panic("%s: Failed to allocate %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx\n",
1379 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1380 (u64)max_addr);
1381 return NULL;
1382}
1383
1384/**
1385 * __memblock_free_early - free boot memory block
1386 * @base: phys starting address of the boot memory block
1387 * @size: size of the boot memory block in bytes
1388 *
1389 * Free boot memory block previously allocated by memblock_virt_alloc_xx() API.
1390 * The freeing memory will not be released to the buddy allocator.
1391 */
1392void __init __memblock_free_early(phys_addr_t base, phys_addr_t size)
1393{
1394 memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1395 __func__, (u64)base, (u64)base + size - 1,
1396 (void *)_RET_IP_);
1397 kmemleak_free_part(__va(base), size);
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001398 memblock_remove_range(&memblock.reserved, base, size);
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001399}
1400
1401/*
1402 * __memblock_free_late - free bootmem block pages directly to buddy allocator
1403 * @addr: phys starting address of the boot memory block
1404 * @size: size of the boot memory block in bytes
1405 *
1406 * This is only useful when the bootmem allocator has already been torn
1407 * down, but we are still initializing the system. Pages are released directly
1408 * to the buddy allocator, no bootmem metadata is updated because it is gone.
1409 */
1410void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
1411{
1412 u64 cursor, end;
1413
1414 memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1415 __func__, (u64)base, (u64)base + size - 1,
1416 (void *)_RET_IP_);
1417 kmemleak_free_part(__va(base), size);
1418 cursor = PFN_UP(base);
1419 end = PFN_DOWN(base + size);
1420
1421 for (; cursor < end; cursor++) {
Mel Gormand70ddd72015-06-30 14:56:52 -07001422 __free_pages_bootmem(pfn_to_page(cursor), cursor, 0);
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001423 totalram_pages++;
1424 }
1425}
Benjamin Herrenschmidt9d1e2492010-07-06 15:39:17 -07001426
1427/*
1428 * Remaining API functions
1429 */
1430
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +10001431phys_addr_t __init memblock_phys_mem_size(void)
Yinghai Lu95f72d12010-07-12 14:36:09 +10001432{
Tejun Heo1440c4e2011-12-08 10:22:08 -08001433 return memblock.memory.total_size;
Yinghai Lu95f72d12010-07-12 14:36:09 +10001434}
1435
Yinghai Lu595ad9a2013-01-24 12:20:09 -08001436phys_addr_t __init memblock_mem_size(unsigned long limit_pfn)
1437{
1438 unsigned long pages = 0;
1439 struct memblock_region *r;
1440 unsigned long start_pfn, end_pfn;
1441
1442 for_each_memblock(memory, r) {
1443 start_pfn = memblock_region_memory_base_pfn(r);
1444 end_pfn = memblock_region_memory_end_pfn(r);
1445 start_pfn = min_t(unsigned long, start_pfn, limit_pfn);
1446 end_pfn = min_t(unsigned long, end_pfn, limit_pfn);
1447 pages += end_pfn - start_pfn;
1448 }
1449
Fabian Frederick16763232014-04-07 15:37:53 -07001450 return PFN_PHYS(pages);
Yinghai Lu595ad9a2013-01-24 12:20:09 -08001451}
1452
Sam Ravnborg0a93ebe2011-10-31 17:08:16 -07001453/* lowest address */
1454phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1455{
1456 return memblock.memory.regions[0].base;
1457}
1458
Yinghai Lu10d06432010-07-28 15:43:02 +10001459phys_addr_t __init_memblock memblock_end_of_DRAM(void)
Yinghai Lu95f72d12010-07-12 14:36:09 +10001460{
1461 int idx = memblock.memory.cnt - 1;
1462
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +10001463 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
Yinghai Lu95f72d12010-07-12 14:36:09 +10001464}
1465
Tejun Heoc0ce8fe2011-12-08 10:22:07 -08001466void __init memblock_enforce_memory_limit(phys_addr_t limit)
Yinghai Lu95f72d12010-07-12 14:36:09 +10001467{
Tejun Heoc0ce8fe2011-12-08 10:22:07 -08001468 phys_addr_t max_addr = (phys_addr_t)ULLONG_MAX;
Emil Medve136199f2014-04-07 15:37:52 -07001469 struct memblock_region *r;
Yinghai Lu95f72d12010-07-12 14:36:09 +10001470
Tejun Heoc0ce8fe2011-12-08 10:22:07 -08001471 if (!limit)
Yinghai Lu95f72d12010-07-12 14:36:09 +10001472 return;
1473
Tejun Heoc0ce8fe2011-12-08 10:22:07 -08001474 /* find out max address */
Emil Medve136199f2014-04-07 15:37:52 -07001475 for_each_memblock(memory, r) {
Tejun Heoc0ce8fe2011-12-08 10:22:07 -08001476 if (limit <= r->size) {
1477 max_addr = r->base + limit;
1478 break;
1479 }
1480 limit -= r->size;
Yinghai Lu95f72d12010-07-12 14:36:09 +10001481 }
1482
Tejun Heoc0ce8fe2011-12-08 10:22:07 -08001483 /* truncate both memory and reserved regions */
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001484 memblock_remove_range(&memblock.memory, max_addr,
1485 (phys_addr_t)ULLONG_MAX);
1486 memblock_remove_range(&memblock.reserved, max_addr,
1487 (phys_addr_t)ULLONG_MAX);
Yinghai Lu95f72d12010-07-12 14:36:09 +10001488}
1489
Yinghai Lucd794812010-10-11 12:34:09 -07001490static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +10001491{
1492 unsigned int left = 0, right = type->cnt;
1493
1494 do {
1495 unsigned int mid = (right + left) / 2;
1496
1497 if (addr < type->regions[mid].base)
1498 right = mid;
1499 else if (addr >= (type->regions[mid].base +
1500 type->regions[mid].size))
1501 left = mid + 1;
1502 else
1503 return mid;
1504 } while (left < right);
1505 return -1;
1506}
1507
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +10001508int __init memblock_is_reserved(phys_addr_t addr)
Yinghai Lu95f72d12010-07-12 14:36:09 +10001509{
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +10001510 return memblock_search(&memblock.reserved, addr) != -1;
1511}
Yinghai Lu95f72d12010-07-12 14:36:09 +10001512
Yinghai Lu3661ca62010-09-15 13:05:29 -07001513int __init_memblock memblock_is_memory(phys_addr_t addr)
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +10001514{
1515 return memblock_search(&memblock.memory, addr) != -1;
1516}
1517
Yinghai Lue76b63f2013-09-11 14:22:17 -07001518#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1519int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1520 unsigned long *start_pfn, unsigned long *end_pfn)
1521{
1522 struct memblock_type *type = &memblock.memory;
Fabian Frederick16763232014-04-07 15:37:53 -07001523 int mid = memblock_search(type, PFN_PHYS(pfn));
Yinghai Lue76b63f2013-09-11 14:22:17 -07001524
1525 if (mid == -1)
1526 return -1;
1527
Fabian Frederickf7e2f7e2014-06-04 16:07:51 -07001528 *start_pfn = PFN_DOWN(type->regions[mid].base);
1529 *end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
Yinghai Lue76b63f2013-09-11 14:22:17 -07001530
1531 return type->regions[mid].nid;
1532}
1533#endif
1534
Stephen Boydeab30942012-05-24 00:45:21 -07001535/**
1536 * memblock_is_region_memory - check if a region is a subset of memory
1537 * @base: base of region to check
1538 * @size: size of region to check
1539 *
1540 * Check if the region [@base, @base+@size) is a subset of a memory block.
1541 *
1542 * RETURNS:
1543 * 0 if false, non-zero if true
1544 */
Yinghai Lu3661ca62010-09-15 13:05:29 -07001545int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +10001546{
Tomi Valkeinenabb65272011-01-20 14:44:20 -08001547 int idx = memblock_search(&memblock.memory, base);
Tejun Heoeb18f1b2011-12-08 10:22:07 -08001548 phys_addr_t end = base + memblock_cap_size(base, &size);
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +10001549
1550 if (idx == -1)
1551 return 0;
Tomi Valkeinenabb65272011-01-20 14:44:20 -08001552 return memblock.memory.regions[idx].base <= base &&
1553 (memblock.memory.regions[idx].base +
Tejun Heoeb18f1b2011-12-08 10:22:07 -08001554 memblock.memory.regions[idx].size) >= end;
Yinghai Lu95f72d12010-07-12 14:36:09 +10001555}
1556
Stephen Boydeab30942012-05-24 00:45:21 -07001557/**
1558 * memblock_is_region_reserved - check if a region intersects reserved memory
1559 * @base: base of region to check
1560 * @size: size of region to check
1561 *
1562 * Check if the region [@base, @base+@size) intersects a reserved memory block.
1563 *
1564 * RETURNS:
1565 * 0 if false, non-zero if true
1566 */
Yinghai Lu10d06432010-07-28 15:43:02 +10001567int __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +10001568{
Tejun Heoeb18f1b2011-12-08 10:22:07 -08001569 memblock_cap_size(base, &size);
Benjamin Herrenschmidtf1c2c192010-08-04 14:17:17 +10001570 return memblock_overlaps_region(&memblock.reserved, base, size) >= 0;
Yinghai Lu95f72d12010-07-12 14:36:09 +10001571}
1572
Yinghai Lu6ede1fd2012-10-22 16:35:18 -07001573void __init_memblock memblock_trim_memory(phys_addr_t align)
1574{
Yinghai Lu6ede1fd2012-10-22 16:35:18 -07001575 phys_addr_t start, end, orig_start, orig_end;
Emil Medve136199f2014-04-07 15:37:52 -07001576 struct memblock_region *r;
Yinghai Lu6ede1fd2012-10-22 16:35:18 -07001577
Emil Medve136199f2014-04-07 15:37:52 -07001578 for_each_memblock(memory, r) {
1579 orig_start = r->base;
1580 orig_end = r->base + r->size;
Yinghai Lu6ede1fd2012-10-22 16:35:18 -07001581 start = round_up(orig_start, align);
1582 end = round_down(orig_end, align);
1583
1584 if (start == orig_start && end == orig_end)
1585 continue;
1586
1587 if (start < end) {
Emil Medve136199f2014-04-07 15:37:52 -07001588 r->base = start;
1589 r->size = end - start;
Yinghai Lu6ede1fd2012-10-22 16:35:18 -07001590 } else {
Emil Medve136199f2014-04-07 15:37:52 -07001591 memblock_remove_region(&memblock.memory,
1592 r - memblock.memory.regions);
1593 r--;
Yinghai Lu6ede1fd2012-10-22 16:35:18 -07001594 }
1595 }
1596}
Benjamin Herrenschmidte63075a2010-07-06 15:39:01 -07001597
Yinghai Lu3661ca62010-09-15 13:05:29 -07001598void __init_memblock memblock_set_current_limit(phys_addr_t limit)
Benjamin Herrenschmidte63075a2010-07-06 15:39:01 -07001599{
1600 memblock.current_limit = limit;
1601}
1602
Laura Abbottfec51012014-02-27 01:23:43 +01001603phys_addr_t __init_memblock memblock_get_current_limit(void)
1604{
1605 return memblock.current_limit;
1606}
1607
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001608static void __init_memblock memblock_dump(struct memblock_type *type, char *name)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001609{
1610 unsigned long long base, size;
Tang Chen66a20752014-01-21 15:49:20 -08001611 unsigned long flags;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001612 int i;
1613
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001614 pr_info(" %s.cnt = 0x%lx\n", name, type->cnt);
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001615
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001616 for (i = 0; i < type->cnt; i++) {
1617 struct memblock_region *rgn = &type->regions[i];
1618 char nid_buf[32] = "";
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001619
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001620 base = rgn->base;
1621 size = rgn->size;
Tang Chen66a20752014-01-21 15:49:20 -08001622 flags = rgn->flags;
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001623#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1624 if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1625 snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1626 memblock_get_region_node(rgn));
1627#endif
Tang Chen66a20752014-01-21 15:49:20 -08001628 pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes%s flags: %#lx\n",
1629 name, i, base, base + size - 1, size, nid_buf, flags);
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001630 }
1631}
1632
Tejun Heo4ff7b822011-12-08 10:22:06 -08001633void __init_memblock __memblock_dump_all(void)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001634{
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001635 pr_info("MEMBLOCK configuration:\n");
Tejun Heo1440c4e2011-12-08 10:22:08 -08001636 pr_info(" memory size = %#llx reserved size = %#llx\n",
1637 (unsigned long long)memblock.memory.total_size,
1638 (unsigned long long)memblock.reserved.total_size);
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001639
1640 memblock_dump(&memblock.memory, "memory");
1641 memblock_dump(&memblock.reserved, "reserved");
1642}
1643
Tejun Heo1aadc052011-12-08 10:22:08 -08001644void __init memblock_allow_resize(void)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001645{
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -07001646 memblock_can_resize = 1;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001647}
1648
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001649static int __init early_memblock(char *p)
1650{
1651 if (p && strstr(p, "debug"))
1652 memblock_debug = 1;
1653 return 0;
1654}
1655early_param("memblock", early_memblock);
1656
Tejun Heoc378ddd2011-07-14 11:46:03 +02001657#if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK)
Benjamin Herrenschmidt6d03b882010-07-06 15:39:19 -07001658
1659static int memblock_debug_show(struct seq_file *m, void *private)
1660{
1661 struct memblock_type *type = m->private;
1662 struct memblock_region *reg;
1663 int i;
1664
1665 for (i = 0; i < type->cnt; i++) {
1666 reg = &type->regions[i];
1667 seq_printf(m, "%4d: ", i);
1668 if (sizeof(phys_addr_t) == 4)
1669 seq_printf(m, "0x%08lx..0x%08lx\n",
1670 (unsigned long)reg->base,
1671 (unsigned long)(reg->base + reg->size - 1));
1672 else
1673 seq_printf(m, "0x%016llx..0x%016llx\n",
1674 (unsigned long long)reg->base,
1675 (unsigned long long)(reg->base + reg->size - 1));
1676
1677 }
1678 return 0;
1679}
1680
1681static int memblock_debug_open(struct inode *inode, struct file *file)
1682{
1683 return single_open(file, memblock_debug_show, inode->i_private);
1684}
1685
1686static const struct file_operations memblock_debug_fops = {
1687 .open = memblock_debug_open,
1688 .read = seq_read,
1689 .llseek = seq_lseek,
1690 .release = single_release,
1691};
1692
1693static int __init memblock_init_debugfs(void)
1694{
1695 struct dentry *root = debugfs_create_dir("memblock", NULL);
1696 if (!root)
1697 return -ENXIO;
1698 debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops);
1699 debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops);
Philipp Hachtmann70210ed2014-01-29 18:16:01 +01001700#ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1701 debugfs_create_file("physmem", S_IRUGO, root, &memblock.physmem, &memblock_debug_fops);
1702#endif
Benjamin Herrenschmidt6d03b882010-07-06 15:39:19 -07001703
1704 return 0;
1705}
1706__initcall(memblock_init_debugfs);
1707
1708#endif /* CONFIG_DEBUG_FS */