aboutsummaryrefslogtreecommitdiffstats
path: root/mm/bootmem.c
diff options
context:
space:
mode:
Diffstat (limited to 'mm/bootmem.c')
-rw-r--r--mm/bootmem.c935
1 files changed, 514 insertions, 421 deletions
diff --git a/mm/bootmem.c b/mm/bootmem.c
index 8d9f60e06f62..4af15d0340ad 100644
--- a/mm/bootmem.c
+++ b/mm/bootmem.c
@@ -1,12 +1,12 @@
1/* 1/*
2 * linux/mm/bootmem.c 2 * bootmem - A boot-time physical memory allocator and configurator
3 * 3 *
4 * Copyright (C) 1999 Ingo Molnar 4 * Copyright (C) 1999 Ingo Molnar
5 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999 5 * 1999 Kanoj Sarcar, SGI
6 * 2008 Johannes Weiner
6 * 7 *
7 * simple boot-time physical memory area allocator and 8 * Access to this subsystem has to be serialized externally (which is true
8 * free memory collector. It's used to deal with reserved 9 * for the boot process anyway).
9 * system memory and memory holes as well.
10 */ 10 */
11#include <linux/init.h> 11#include <linux/init.h>
12#include <linux/pfn.h> 12#include <linux/pfn.h>
@@ -19,15 +19,10 @@
19 19
20#include "internal.h" 20#include "internal.h"
21 21
22/*
23 * Access to this subsystem has to be serialized externally. (this is
24 * true for the boot process anyway)
25 */
26unsigned long max_low_pfn; 22unsigned long max_low_pfn;
27unsigned long min_low_pfn; 23unsigned long min_low_pfn;
28unsigned long max_pfn; 24unsigned long max_pfn;
29 25
30static LIST_HEAD(bdata_list);
31#ifdef CONFIG_CRASH_DUMP 26#ifdef CONFIG_CRASH_DUMP
32/* 27/*
33 * If we have booted due to a crash, max_pfn will be a very low value. We need 28 * If we have booted due to a crash, max_pfn will be a very low value. We need
@@ -36,63 +31,72 @@ static LIST_HEAD(bdata_list);
36unsigned long saved_max_pfn; 31unsigned long saved_max_pfn;
37#endif 32#endif
38 33
39/* return the number of _pages_ that will be allocated for the boot bitmap */ 34bootmem_data_t bootmem_node_data[MAX_NUMNODES] __initdata;
40unsigned long __init bootmem_bootmap_pages(unsigned long pages) 35
36static struct list_head bdata_list __initdata = LIST_HEAD_INIT(bdata_list);
37
38static int bootmem_debug;
39
40static int __init bootmem_debug_setup(char *buf)
41{ 41{
42 unsigned long mapsize; 42 bootmem_debug = 1;
43 return 0;
44}
45early_param("bootmem_debug", bootmem_debug_setup);
43 46
44 mapsize = (pages+7)/8; 47#define bdebug(fmt, args...) ({ \
45 mapsize = (mapsize + ~PAGE_MASK) & PAGE_MASK; 48 if (unlikely(bootmem_debug)) \
46 mapsize >>= PAGE_SHIFT; 49 printk(KERN_INFO \
50 "bootmem::%s " fmt, \
51 __FUNCTION__, ## args); \
52})
47 53
48 return mapsize; 54static unsigned long __init bootmap_bytes(unsigned long pages)
55{
56 unsigned long bytes = (pages + 7) / 8;
57
58 return ALIGN(bytes, sizeof(long));
49} 59}
50 60
51/* 61/**
52 * link bdata in order 62 * bootmem_bootmap_pages - calculate bitmap size in pages
63 * @pages: number of pages the bitmap has to represent
53 */ 64 */
54static void __init link_bootmem(bootmem_data_t *bdata) 65unsigned long __init bootmem_bootmap_pages(unsigned long pages)
55{ 66{
56 bootmem_data_t *ent; 67 unsigned long bytes = bootmap_bytes(pages);
57 68
58 if (list_empty(&bdata_list)) { 69 return PAGE_ALIGN(bytes) >> PAGE_SHIFT;
59 list_add(&bdata->list, &bdata_list);
60 return;
61 }
62 /* insert in order */
63 list_for_each_entry(ent, &bdata_list, list) {
64 if (bdata->node_boot_start < ent->node_boot_start) {
65 list_add_tail(&bdata->list, &ent->list);
66 return;
67 }
68 }
69 list_add_tail(&bdata->list, &bdata_list);
70} 70}
71 71
72/* 72/*
73 * Given an initialised bdata, it returns the size of the boot bitmap 73 * link bdata in order
74 */ 74 */
75static unsigned long __init get_mapsize(bootmem_data_t *bdata) 75static void __init link_bootmem(bootmem_data_t *bdata)
76{ 76{
77 unsigned long mapsize; 77 struct list_head *iter;
78 unsigned long start = PFN_DOWN(bdata->node_boot_start);
79 unsigned long end = bdata->node_low_pfn;
80 78
81 mapsize = ((end - start) + 7) / 8; 79 list_for_each(iter, &bdata_list) {
82 return ALIGN(mapsize, sizeof(long)); 80 bootmem_data_t *ent;
81
82 ent = list_entry(iter, bootmem_data_t, list);
83 if (bdata->node_min_pfn < ent->node_min_pfn)
84 break;
85 }
86 list_add_tail(&bdata->list, iter);
83} 87}
84 88
85/* 89/*
86 * Called once to set up the allocator itself. 90 * Called once to set up the allocator itself.
87 */ 91 */
88static unsigned long __init init_bootmem_core(pg_data_t *pgdat, 92static unsigned long __init init_bootmem_core(bootmem_data_t *bdata,
89 unsigned long mapstart, unsigned long start, unsigned long end) 93 unsigned long mapstart, unsigned long start, unsigned long end)
90{ 94{
91 bootmem_data_t *bdata = pgdat->bdata;
92 unsigned long mapsize; 95 unsigned long mapsize;
93 96
97 mminit_validate_memmodel_limits(&start, &end);
94 bdata->node_bootmem_map = phys_to_virt(PFN_PHYS(mapstart)); 98 bdata->node_bootmem_map = phys_to_virt(PFN_PHYS(mapstart));
95 bdata->node_boot_start = PFN_PHYS(start); 99 bdata->node_min_pfn = start;
96 bdata->node_low_pfn = end; 100 bdata->node_low_pfn = end;
97 link_bootmem(bdata); 101 link_bootmem(bdata);
98 102
@@ -100,429 +104,461 @@ static unsigned long __init init_bootmem_core(pg_data_t *pgdat,
100 * Initially all pages are reserved - setup_arch() has to 104 * Initially all pages are reserved - setup_arch() has to
101 * register free RAM areas explicitly. 105 * register free RAM areas explicitly.
102 */ 106 */
103 mapsize = get_mapsize(bdata); 107 mapsize = bootmap_bytes(end - start);
104 memset(bdata->node_bootmem_map, 0xff, mapsize); 108 memset(bdata->node_bootmem_map, 0xff, mapsize);
105 109
110 bdebug("nid=%td start=%lx map=%lx end=%lx mapsize=%lx\n",
111 bdata - bootmem_node_data, start, mapstart, end, mapsize);
112
106 return mapsize; 113 return mapsize;
107} 114}
108 115
109/* 116/**
110 * Marks a particular physical memory range as unallocatable. Usable RAM 117 * init_bootmem_node - register a node as boot memory
111 * might be used for boot-time allocations - or it might get added 118 * @pgdat: node to register
112 * to the free page pool later on. 119 * @freepfn: pfn where the bitmap for this node is to be placed
120 * @startpfn: first pfn on the node
121 * @endpfn: first pfn after the node
122 *
123 * Returns the number of bytes needed to hold the bitmap for this node.
113 */ 124 */
114static int __init can_reserve_bootmem_core(bootmem_data_t *bdata, 125unsigned long __init init_bootmem_node(pg_data_t *pgdat, unsigned long freepfn,
115 unsigned long addr, unsigned long size, int flags) 126 unsigned long startpfn, unsigned long endpfn)
116{ 127{
117 unsigned long sidx, eidx; 128 return init_bootmem_core(pgdat->bdata, freepfn, startpfn, endpfn);
118 unsigned long i; 129}
119 130
120 BUG_ON(!size); 131/**
132 * init_bootmem - register boot memory
133 * @start: pfn where the bitmap is to be placed
134 * @pages: number of available physical pages
135 *
136 * Returns the number of bytes needed to hold the bitmap.
137 */
138unsigned long __init init_bootmem(unsigned long start, unsigned long pages)
139{
140 max_low_pfn = pages;
141 min_low_pfn = start;
142 return init_bootmem_core(NODE_DATA(0)->bdata, start, 0, pages);
143}
121 144
122 /* out of range, don't hold other */ 145static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata)
123 if (addr + size < bdata->node_boot_start || 146{
124 PFN_DOWN(addr) > bdata->node_low_pfn) 147 int aligned;
148 struct page *page;
149 unsigned long start, end, pages, count = 0;
150
151 if (!bdata->node_bootmem_map)
125 return 0; 152 return 0;
126 153
154 start = bdata->node_min_pfn;
155 end = bdata->node_low_pfn;
156
127 /* 157 /*
128 * Round up to index to the range. 158 * If the start is aligned to the machines wordsize, we might
159 * be able to free pages in bulks of that order.
129 */ 160 */
130 if (addr > bdata->node_boot_start) 161 aligned = !(start & (BITS_PER_LONG - 1));
131 sidx= PFN_DOWN(addr - bdata->node_boot_start);
132 else
133 sidx = 0;
134 162
135 eidx = PFN_UP(addr + size - bdata->node_boot_start); 163 bdebug("nid=%td start=%lx end=%lx aligned=%d\n",
136 if (eidx > bdata->node_low_pfn - PFN_DOWN(bdata->node_boot_start)) 164 bdata - bootmem_node_data, start, end, aligned);
137 eidx = bdata->node_low_pfn - PFN_DOWN(bdata->node_boot_start);
138 165
139 for (i = sidx; i < eidx; i++) { 166 while (start < end) {
140 if (test_bit(i, bdata->node_bootmem_map)) { 167 unsigned long *map, idx, vec;
141 if (flags & BOOTMEM_EXCLUSIVE)
142 return -EBUSY;
143 }
144 }
145 168
146 return 0; 169 map = bdata->node_bootmem_map;
170 idx = start - bdata->node_min_pfn;
171 vec = ~map[idx / BITS_PER_LONG];
147 172
148} 173 if (aligned && vec == ~0UL && start + BITS_PER_LONG < end) {
174 int order = ilog2(BITS_PER_LONG);
149 175
150static void __init reserve_bootmem_core(bootmem_data_t *bdata, 176 __free_pages_bootmem(pfn_to_page(start), order);
151 unsigned long addr, unsigned long size, int flags) 177 count += BITS_PER_LONG;
152{ 178 } else {
153 unsigned long sidx, eidx; 179 unsigned long off = 0;
154 unsigned long i;
155
156 BUG_ON(!size);
157 180
158 /* out of range */ 181 while (vec && off < BITS_PER_LONG) {
159 if (addr + size < bdata->node_boot_start || 182 if (vec & 1) {
160 PFN_DOWN(addr) > bdata->node_low_pfn) 183 page = pfn_to_page(start + off);
161 return; 184 __free_pages_bootmem(page, 0);
185 count++;
186 }
187 vec >>= 1;
188 off++;
189 }
190 }
191 start += BITS_PER_LONG;
192 }
162 193
163 /* 194 page = virt_to_page(bdata->node_bootmem_map);
164 * Round up to index to the range. 195 pages = bdata->node_low_pfn - bdata->node_min_pfn;
165 */ 196 pages = bootmem_bootmap_pages(pages);
166 if (addr > bdata->node_boot_start) 197 count += pages;
167 sidx= PFN_DOWN(addr - bdata->node_boot_start); 198 while (pages--)
168 else 199 __free_pages_bootmem(page++, 0);
169 sidx = 0;
170 200
171 eidx = PFN_UP(addr + size - bdata->node_boot_start); 201 bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count);
172 if (eidx > bdata->node_low_pfn - PFN_DOWN(bdata->node_boot_start))
173 eidx = bdata->node_low_pfn - PFN_DOWN(bdata->node_boot_start);
174 202
175 for (i = sidx; i < eidx; i++) { 203 return count;
176 if (test_and_set_bit(i, bdata->node_bootmem_map)) {
177#ifdef CONFIG_DEBUG_BOOTMEM
178 printk("hm, page %08lx reserved twice.\n", i*PAGE_SIZE);
179#endif
180 }
181 }
182} 204}
183 205
184static void __init free_bootmem_core(bootmem_data_t *bdata, unsigned long addr, 206/**
185 unsigned long size) 207 * free_all_bootmem_node - release a node's free pages to the buddy allocator
208 * @pgdat: node to be released
209 *
210 * Returns the number of pages actually released.
211 */
212unsigned long __init free_all_bootmem_node(pg_data_t *pgdat)
186{ 213{
187 unsigned long sidx, eidx; 214 register_page_bootmem_info_node(pgdat);
188 unsigned long i; 215 return free_all_bootmem_core(pgdat->bdata);
189 216}
190 BUG_ON(!size);
191 217
192 /* out range */ 218/**
193 if (addr + size < bdata->node_boot_start || 219 * free_all_bootmem - release free pages to the buddy allocator
194 PFN_DOWN(addr) > bdata->node_low_pfn) 220 *
195 return; 221 * Returns the number of pages actually released.
196 /* 222 */
197 * round down end of usable mem, partially free pages are 223unsigned long __init free_all_bootmem(void)
198 * considered reserved. 224{
199 */ 225 return free_all_bootmem_core(NODE_DATA(0)->bdata);
226}
200 227
201 if (addr >= bdata->node_boot_start && addr < bdata->last_success) 228static void __init __free(bootmem_data_t *bdata,
202 bdata->last_success = addr; 229 unsigned long sidx, unsigned long eidx)
230{
231 unsigned long idx;
203 232
204 /* 233 bdebug("nid=%td start=%lx end=%lx\n", bdata - bootmem_node_data,
205 * Round up to index to the range. 234 sidx + bdata->node_min_pfn,
206 */ 235 eidx + bdata->node_min_pfn);
207 if (PFN_UP(addr) > PFN_DOWN(bdata->node_boot_start))
208 sidx = PFN_UP(addr) - PFN_DOWN(bdata->node_boot_start);
209 else
210 sidx = 0;
211 236
212 eidx = PFN_DOWN(addr + size - bdata->node_boot_start); 237 if (bdata->hint_idx > sidx)
213 if (eidx > bdata->node_low_pfn - PFN_DOWN(bdata->node_boot_start)) 238 bdata->hint_idx = sidx;
214 eidx = bdata->node_low_pfn - PFN_DOWN(bdata->node_boot_start);
215 239
216 for (i = sidx; i < eidx; i++) { 240 for (idx = sidx; idx < eidx; idx++)
217 if (unlikely(!test_and_clear_bit(i, bdata->node_bootmem_map))) 241 if (!test_and_clear_bit(idx, bdata->node_bootmem_map))
218 BUG(); 242 BUG();
219 }
220} 243}
221 244
222/* 245static int __init __reserve(bootmem_data_t *bdata, unsigned long sidx,
223 * We 'merge' subsequent allocations to save space. We might 'lose' 246 unsigned long eidx, int flags)
224 * some fraction of a page if allocations cannot be satisfied due to
225 * size constraints on boxes where there is physical RAM space
226 * fragmentation - in these cases (mostly large memory boxes) this
227 * is not a problem.
228 *
229 * On low memory boxes we get it right in 100% of the cases.
230 *
231 * alignment has to be a power of 2 value.
232 *
233 * NOTE: This function is _not_ reentrant.
234 */
235void * __init
236__alloc_bootmem_core(struct bootmem_data *bdata, unsigned long size,
237 unsigned long align, unsigned long goal, unsigned long limit)
238{ 247{
239 unsigned long areasize, preferred; 248 unsigned long idx;
240 unsigned long i, start = 0, incr, eidx, end_pfn; 249 int exclusive = flags & BOOTMEM_EXCLUSIVE;
241 void *ret; 250
242 unsigned long node_boot_start; 251 bdebug("nid=%td start=%lx end=%lx flags=%x\n",
243 void *node_bootmem_map; 252 bdata - bootmem_node_data,
244 253 sidx + bdata->node_min_pfn,
245 if (!size) { 254 eidx + bdata->node_min_pfn,
246 printk("__alloc_bootmem_core(): zero-sized request\n"); 255 flags);
247 BUG(); 256
248 } 257 for (idx = sidx; idx < eidx; idx++)
249 BUG_ON(align & (align-1)); 258 if (test_and_set_bit(idx, bdata->node_bootmem_map)) {
250 259 if (exclusive) {
251 /* on nodes without memory - bootmem_map is NULL */ 260 __free(bdata, sidx, idx);
252 if (!bdata->node_bootmem_map) 261 return -EBUSY;
253 return NULL; 262 }
263 bdebug("silent double reserve of PFN %lx\n",
264 idx + bdata->node_min_pfn);
265 }
266 return 0;
267}
254 268
255 /* bdata->node_boot_start is supposed to be (12+6)bits alignment on x86_64 ? */ 269static int __init mark_bootmem_node(bootmem_data_t *bdata,
256 node_boot_start = bdata->node_boot_start; 270 unsigned long start, unsigned long end,
257 node_bootmem_map = bdata->node_bootmem_map; 271 int reserve, int flags)
258 if (align) { 272{
259 node_boot_start = ALIGN(bdata->node_boot_start, align); 273 unsigned long sidx, eidx;
260 if (node_boot_start > bdata->node_boot_start)
261 node_bootmem_map = (unsigned long *)bdata->node_bootmem_map +
262 PFN_DOWN(node_boot_start - bdata->node_boot_start)/BITS_PER_LONG;
263 }
264 274
265 if (limit && node_boot_start >= limit) 275 bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n",
266 return NULL; 276 bdata - bootmem_node_data, start, end, reserve, flags);
267 277
268 end_pfn = bdata->node_low_pfn; 278 BUG_ON(start < bdata->node_min_pfn);
269 limit = PFN_DOWN(limit); 279 BUG_ON(end > bdata->node_low_pfn);
270 if (limit && end_pfn > limit)
271 end_pfn = limit;
272 280
273 eidx = end_pfn - PFN_DOWN(node_boot_start); 281 sidx = start - bdata->node_min_pfn;
282 eidx = end - bdata->node_min_pfn;
274 283
275 /* 284 if (reserve)
276 * We try to allocate bootmem pages above 'goal' 285 return __reserve(bdata, sidx, eidx, flags);
277 * first, then we try to allocate lower pages. 286 else
278 */ 287 __free(bdata, sidx, eidx);
279 preferred = 0; 288 return 0;
280 if (goal && PFN_DOWN(goal) < end_pfn) { 289}
281 if (goal > node_boot_start)
282 preferred = goal - node_boot_start;
283
284 if (bdata->last_success > node_boot_start &&
285 bdata->last_success - node_boot_start >= preferred)
286 if (!limit || (limit && limit > bdata->last_success))
287 preferred = bdata->last_success - node_boot_start;
288 }
289 290
290 preferred = PFN_DOWN(ALIGN(preferred, align)); 291static int __init mark_bootmem(unsigned long start, unsigned long end,
291 areasize = (size + PAGE_SIZE-1) / PAGE_SIZE; 292 int reserve, int flags)
292 incr = align >> PAGE_SHIFT ? : 1; 293{
294 unsigned long pos;
295 bootmem_data_t *bdata;
293 296
294restart_scan: 297 pos = start;
295 for (i = preferred; i < eidx;) { 298 list_for_each_entry(bdata, &bdata_list, list) {
296 unsigned long j; 299 int err;
300 unsigned long max;
297 301
298 i = find_next_zero_bit(node_bootmem_map, eidx, i); 302 if (pos < bdata->node_min_pfn ||
299 i = ALIGN(i, incr); 303 pos >= bdata->node_low_pfn) {
300 if (i >= eidx) 304 BUG_ON(pos != start);
301 break;
302 if (test_bit(i, node_bootmem_map)) {
303 i += incr;
304 continue; 305 continue;
305 } 306 }
306 for (j = i + 1; j < i + areasize; ++j) {
307 if (j >= eidx)
308 goto fail_block;
309 if (test_bit(j, node_bootmem_map))
310 goto fail_block;
311 }
312 start = i;
313 goto found;
314 fail_block:
315 i = ALIGN(j, incr);
316 if (i == j)
317 i += incr;
318 }
319 307
320 if (preferred > 0) { 308 max = min(bdata->node_low_pfn, end);
321 preferred = 0;
322 goto restart_scan;
323 }
324 return NULL;
325 309
326found: 310 err = mark_bootmem_node(bdata, pos, max, reserve, flags);
327 bdata->last_success = PFN_PHYS(start) + node_boot_start; 311 if (reserve && err) {
328 BUG_ON(start >= eidx); 312 mark_bootmem(start, pos, 0, 0);
329 313 return err;
330 /*
331 * Is the next page of the previous allocation-end the start
332 * of this allocation's buffer? If yes then we can 'merge'
333 * the previous partial page with this allocation.
334 */
335 if (align < PAGE_SIZE &&
336 bdata->last_offset && bdata->last_pos+1 == start) {
337 unsigned long offset, remaining_size;
338 offset = ALIGN(bdata->last_offset, align);
339 BUG_ON(offset > PAGE_SIZE);
340 remaining_size = PAGE_SIZE - offset;
341 if (size < remaining_size) {
342 areasize = 0;
343 /* last_pos unchanged */
344 bdata->last_offset = offset + size;
345 ret = phys_to_virt(bdata->last_pos * PAGE_SIZE +
346 offset + node_boot_start);
347 } else {
348 remaining_size = size - remaining_size;
349 areasize = (remaining_size + PAGE_SIZE-1) / PAGE_SIZE;
350 ret = phys_to_virt(bdata->last_pos * PAGE_SIZE +
351 offset + node_boot_start);
352 bdata->last_pos = start + areasize - 1;
353 bdata->last_offset = remaining_size;
354 } 314 }
355 bdata->last_offset &= ~PAGE_MASK;
356 } else {
357 bdata->last_pos = start + areasize - 1;
358 bdata->last_offset = size & ~PAGE_MASK;
359 ret = phys_to_virt(start * PAGE_SIZE + node_boot_start);
360 }
361 315
362 /* 316 if (max == end)
363 * Reserve the area now: 317 return 0;
364 */ 318 pos = bdata->node_low_pfn;
365 for (i = start; i < start + areasize; i++) 319 }
366 if (unlikely(test_and_set_bit(i, node_bootmem_map))) 320 BUG();
367 BUG();
368 memset(ret, 0, size);
369 return ret;
370} 321}
371 322
372static unsigned long __init free_all_bootmem_core(pg_data_t *pgdat) 323/**
324 * free_bootmem_node - mark a page range as usable
325 * @pgdat: node the range resides on
326 * @physaddr: starting address of the range
327 * @size: size of the range in bytes
328 *
329 * Partial pages will be considered reserved and left as they are.
330 *
331 * The range must reside completely on the specified node.
332 */
333void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
334 unsigned long size)
373{ 335{
374 struct page *page; 336 unsigned long start, end;
375 unsigned long pfn;
376 bootmem_data_t *bdata = pgdat->bdata;
377 unsigned long i, count, total = 0;
378 unsigned long idx;
379 unsigned long *map;
380 int gofast = 0;
381
382 BUG_ON(!bdata->node_bootmem_map);
383
384 count = 0;
385 /* first extant page of the node */
386 pfn = PFN_DOWN(bdata->node_boot_start);
387 idx = bdata->node_low_pfn - pfn;
388 map = bdata->node_bootmem_map;
389 /* Check physaddr is O(LOG2(BITS_PER_LONG)) page aligned */
390 if (bdata->node_boot_start == 0 ||
391 ffs(bdata->node_boot_start) - PAGE_SHIFT > ffs(BITS_PER_LONG))
392 gofast = 1;
393 for (i = 0; i < idx; ) {
394 unsigned long v = ~map[i / BITS_PER_LONG];
395
396 if (gofast && v == ~0UL) {
397 int order;
398
399 page = pfn_to_page(pfn);
400 count += BITS_PER_LONG;
401 order = ffs(BITS_PER_LONG) - 1;
402 __free_pages_bootmem(page, order);
403 i += BITS_PER_LONG;
404 page += BITS_PER_LONG;
405 } else if (v) {
406 unsigned long m;
407
408 page = pfn_to_page(pfn);
409 for (m = 1; m && i < idx; m<<=1, page++, i++) {
410 if (v & m) {
411 count++;
412 __free_pages_bootmem(page, 0);
413 }
414 }
415 } else {
416 i += BITS_PER_LONG;
417 }
418 pfn += BITS_PER_LONG;
419 }
420 total += count;
421 337
422 /* 338 start = PFN_UP(physaddr);
423 * Now free the allocator bitmap itself, it's not 339 end = PFN_DOWN(physaddr + size);
424 * needed anymore:
425 */
426 page = virt_to_page(bdata->node_bootmem_map);
427 count = 0;
428 idx = (get_mapsize(bdata) + PAGE_SIZE-1) >> PAGE_SHIFT;
429 for (i = 0; i < idx; i++, page++) {
430 __free_pages_bootmem(page, 0);
431 count++;
432 }
433 total += count;
434 bdata->node_bootmem_map = NULL;
435 340
436 return total; 341 mark_bootmem_node(pgdat->bdata, start, end, 0, 0);
437} 342}
438 343
439unsigned long __init init_bootmem_node(pg_data_t *pgdat, unsigned long freepfn, 344/**
440 unsigned long startpfn, unsigned long endpfn) 345 * free_bootmem - mark a page range as usable
441{ 346 * @addr: starting address of the range
442 return init_bootmem_core(pgdat, freepfn, startpfn, endpfn); 347 * @size: size of the range in bytes
443} 348 *
444 349 * Partial pages will be considered reserved and left as they are.
445int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr, 350 *
446 unsigned long size, int flags) 351 * The range must be contiguous but may span node boundaries.
352 */
353void __init free_bootmem(unsigned long addr, unsigned long size)
447{ 354{
448 int ret; 355 unsigned long start, end;
449 356
450 ret = can_reserve_bootmem_core(pgdat->bdata, physaddr, size, flags); 357 start = PFN_UP(addr);
451 if (ret < 0) 358 end = PFN_DOWN(addr + size);
452 return -ENOMEM;
453 reserve_bootmem_core(pgdat->bdata, physaddr, size, flags);
454 359
455 return 0; 360 mark_bootmem(start, end, 0, 0);
456} 361}
457 362
458void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr, 363/**
459 unsigned long size) 364 * reserve_bootmem_node - mark a page range as reserved
365 * @pgdat: node the range resides on
366 * @physaddr: starting address of the range
367 * @size: size of the range in bytes
368 * @flags: reservation flags (see linux/bootmem.h)
369 *
370 * Partial pages will be reserved.
371 *
372 * The range must reside completely on the specified node.
373 */
374int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
375 unsigned long size, int flags)
460{ 376{
461 free_bootmem_core(pgdat->bdata, physaddr, size); 377 unsigned long start, end;
462}
463 378
464unsigned long __init free_all_bootmem_node(pg_data_t *pgdat) 379 start = PFN_DOWN(physaddr);
465{ 380 end = PFN_UP(physaddr + size);
466 register_page_bootmem_info_node(pgdat);
467 return free_all_bootmem_core(pgdat);
468}
469 381
470unsigned long __init init_bootmem(unsigned long start, unsigned long pages) 382 return mark_bootmem_node(pgdat->bdata, start, end, 1, flags);
471{
472 max_low_pfn = pages;
473 min_low_pfn = start;
474 return init_bootmem_core(NODE_DATA(0), start, 0, pages);
475} 383}
476 384
477#ifndef CONFIG_HAVE_ARCH_BOOTMEM_NODE 385#ifndef CONFIG_HAVE_ARCH_BOOTMEM_NODE
386/**
387 * reserve_bootmem - mark a page range as usable
388 * @addr: starting address of the range
389 * @size: size of the range in bytes
390 * @flags: reservation flags (see linux/bootmem.h)
391 *
392 * Partial pages will be reserved.
393 *
394 * The range must be contiguous but may span node boundaries.
395 */
478int __init reserve_bootmem(unsigned long addr, unsigned long size, 396int __init reserve_bootmem(unsigned long addr, unsigned long size,
479 int flags) 397 int flags)
480{ 398{
481 bootmem_data_t *bdata; 399 unsigned long start, end;
482 int ret;
483 400
484 list_for_each_entry(bdata, &bdata_list, list) { 401 start = PFN_DOWN(addr);
485 ret = can_reserve_bootmem_core(bdata, addr, size, flags); 402 end = PFN_UP(addr + size);
486 if (ret < 0)
487 return ret;
488 }
489 list_for_each_entry(bdata, &bdata_list, list)
490 reserve_bootmem_core(bdata, addr, size, flags);
491 403
492 return 0; 404 return mark_bootmem(start, end, 1, flags);
493} 405}
494#endif /* !CONFIG_HAVE_ARCH_BOOTMEM_NODE */ 406#endif /* !CONFIG_HAVE_ARCH_BOOTMEM_NODE */
495 407
496void __init free_bootmem(unsigned long addr, unsigned long size) 408static void * __init alloc_bootmem_core(struct bootmem_data *bdata,
409 unsigned long size, unsigned long align,
410 unsigned long goal, unsigned long limit)
497{ 411{
498 bootmem_data_t *bdata; 412 unsigned long fallback = 0;
499 list_for_each_entry(bdata, &bdata_list, list) 413 unsigned long min, max, start, sidx, midx, step;
500 free_bootmem_core(bdata, addr, size);
501}
502 414
503unsigned long __init free_all_bootmem(void) 415 BUG_ON(!size);
504{ 416 BUG_ON(align & (align - 1));
505 return free_all_bootmem_core(NODE_DATA(0)); 417 BUG_ON(limit && goal + size > limit);
418
419 if (!bdata->node_bootmem_map)
420 return NULL;
421
422 bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n",
423 bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT,
424 align, goal, limit);
425
426 min = bdata->node_min_pfn;
427 max = bdata->node_low_pfn;
428
429 goal >>= PAGE_SHIFT;
430 limit >>= PAGE_SHIFT;
431
432 if (limit && max > limit)
433 max = limit;
434 if (max <= min)
435 return NULL;
436
437 step = max(align >> PAGE_SHIFT, 1UL);
438
439 if (goal && min < goal && goal < max)
440 start = ALIGN(goal, step);
441 else
442 start = ALIGN(min, step);
443
444 sidx = start - bdata->node_min_pfn;;
445 midx = max - bdata->node_min_pfn;
446
447 if (bdata->hint_idx > sidx) {
448 /*
449 * Handle the valid case of sidx being zero and still
450 * catch the fallback below.
451 */
452 fallback = sidx + 1;
453 sidx = ALIGN(bdata->hint_idx, step);
454 }
455
456 while (1) {
457 int merge;
458 void *region;
459 unsigned long eidx, i, start_off, end_off;
460find_block:
461 sidx = find_next_zero_bit(bdata->node_bootmem_map, midx, sidx);
462 sidx = ALIGN(sidx, step);
463 eidx = sidx + PFN_UP(size);
464
465 if (sidx >= midx || eidx > midx)
466 break;
467
468 for (i = sidx; i < eidx; i++)
469 if (test_bit(i, bdata->node_bootmem_map)) {
470 sidx = ALIGN(i, step);
471 if (sidx == i)
472 sidx += step;
473 goto find_block;
474 }
475
476 if (bdata->last_end_off &&
477 PFN_DOWN(bdata->last_end_off) + 1 == sidx)
478 start_off = ALIGN(bdata->last_end_off, align);
479 else
480 start_off = PFN_PHYS(sidx);
481
482 merge = PFN_DOWN(start_off) < sidx;
483 end_off = start_off + size;
484
485 bdata->last_end_off = end_off;
486 bdata->hint_idx = PFN_UP(end_off);
487
488 /*
489 * Reserve the area now:
490 */
491 if (__reserve(bdata, PFN_DOWN(start_off) + merge,
492 PFN_UP(end_off), BOOTMEM_EXCLUSIVE))
493 BUG();
494
495 region = phys_to_virt(PFN_PHYS(bdata->node_min_pfn) +
496 start_off);
497 memset(region, 0, size);
498 return region;
499 }
500
501 if (fallback) {
502 sidx = ALIGN(fallback - 1, step);
503 fallback = 0;
504 goto find_block;
505 }
506
507 return NULL;
506} 508}
507 509
508void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align, 510static void * __init ___alloc_bootmem_nopanic(unsigned long size,
509 unsigned long goal) 511 unsigned long align,
512 unsigned long goal,
513 unsigned long limit)
510{ 514{
511 bootmem_data_t *bdata; 515 bootmem_data_t *bdata;
512 void *ptr;
513 516
517restart:
514 list_for_each_entry(bdata, &bdata_list, list) { 518 list_for_each_entry(bdata, &bdata_list, list) {
515 ptr = __alloc_bootmem_core(bdata, size, align, goal, 0); 519 void *region;
516 if (ptr) 520
517 return ptr; 521 if (goal && bdata->node_low_pfn <= PFN_DOWN(goal))
522 continue;
523 if (limit && bdata->node_min_pfn >= PFN_DOWN(limit))
524 break;
525
526 region = alloc_bootmem_core(bdata, size, align, goal, limit);
527 if (region)
528 return region;
529 }
530
531 if (goal) {
532 goal = 0;
533 goto restart;
518 } 534 }
535
519 return NULL; 536 return NULL;
520} 537}
521 538
522void * __init __alloc_bootmem(unsigned long size, unsigned long align, 539/**
523 unsigned long goal) 540 * __alloc_bootmem_nopanic - allocate boot memory without panicking
541 * @size: size of the request in bytes
542 * @align: alignment of the region
543 * @goal: preferred starting address of the region
544 *
545 * The goal is dropped if it can not be satisfied and the allocation will
546 * fall back to memory below @goal.
547 *
548 * Allocation may happen on any node in the system.
549 *
550 * Returns NULL on failure.
551 */
552void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
553 unsigned long goal)
524{ 554{
525 void *mem = __alloc_bootmem_nopanic(size,align,goal); 555 return ___alloc_bootmem_nopanic(size, align, goal, 0);
556}
557
558static void * __init ___alloc_bootmem(unsigned long size, unsigned long align,
559 unsigned long goal, unsigned long limit)
560{
561 void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit);
526 562
527 if (mem) 563 if (mem)
528 return mem; 564 return mem;
@@ -534,78 +570,135 @@ void * __init __alloc_bootmem(unsigned long size, unsigned long align,
534 return NULL; 570 return NULL;
535} 571}
536 572
573/**
574 * __alloc_bootmem - allocate boot memory
575 * @size: size of the request in bytes
576 * @align: alignment of the region
577 * @goal: preferred starting address of the region
578 *
579 * The goal is dropped if it can not be satisfied and the allocation will
580 * fall back to memory below @goal.
581 *
582 * Allocation may happen on any node in the system.
583 *
584 * The function panics if the request can not be satisfied.
585 */
586void * __init __alloc_bootmem(unsigned long size, unsigned long align,
587 unsigned long goal)
588{
589 return ___alloc_bootmem(size, align, goal, 0);
590}
537 591
538void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size, 592static void * __init ___alloc_bootmem_node(bootmem_data_t *bdata,
539 unsigned long align, unsigned long goal) 593 unsigned long size, unsigned long align,
594 unsigned long goal, unsigned long limit)
540{ 595{
541 void *ptr; 596 void *ptr;
542 597
543 ptr = __alloc_bootmem_core(pgdat->bdata, size, align, goal, 0); 598 ptr = alloc_bootmem_core(bdata, size, align, goal, limit);
544 if (ptr) 599 if (ptr)
545 return ptr; 600 return ptr;
546 601
547 return __alloc_bootmem(size, align, goal); 602 return ___alloc_bootmem(size, align, goal, limit);
603}
604
605/**
606 * __alloc_bootmem_node - allocate boot memory from a specific node
607 * @pgdat: node to allocate from
608 * @size: size of the request in bytes
609 * @align: alignment of the region
610 * @goal: preferred starting address of the region
611 *
612 * The goal is dropped if it can not be satisfied and the allocation will
613 * fall back to memory below @goal.
614 *
615 * Allocation may fall back to any node in the system if the specified node
616 * can not hold the requested memory.
617 *
618 * The function panics if the request can not be satisfied.
619 */
620void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
621 unsigned long align, unsigned long goal)
622{
623 return ___alloc_bootmem_node(pgdat->bdata, size, align, goal, 0);
548} 624}
549 625
550#ifdef CONFIG_SPARSEMEM 626#ifdef CONFIG_SPARSEMEM
627/**
628 * alloc_bootmem_section - allocate boot memory from a specific section
629 * @size: size of the request in bytes
630 * @section_nr: sparse map section to allocate from
631 *
632 * Return NULL on failure.
633 */
551void * __init alloc_bootmem_section(unsigned long size, 634void * __init alloc_bootmem_section(unsigned long size,
552 unsigned long section_nr) 635 unsigned long section_nr)
553{ 636{
554 void *ptr; 637 bootmem_data_t *bdata;
555 unsigned long limit, goal, start_nr, end_nr, pfn; 638 unsigned long pfn, goal, limit;
556 struct pglist_data *pgdat;
557 639
558 pfn = section_nr_to_pfn(section_nr); 640 pfn = section_nr_to_pfn(section_nr);
559 goal = PFN_PHYS(pfn); 641 goal = pfn << PAGE_SHIFT;
560 limit = PFN_PHYS(section_nr_to_pfn(section_nr + 1)) - 1; 642 limit = section_nr_to_pfn(section_nr + 1) << PAGE_SHIFT;
561 pgdat = NODE_DATA(early_pfn_to_nid(pfn)); 643 bdata = &bootmem_node_data[early_pfn_to_nid(pfn)];
562 ptr = __alloc_bootmem_core(pgdat->bdata, size, SMP_CACHE_BYTES, goal,
563 limit);
564 644
565 if (!ptr) 645 return alloc_bootmem_core(bdata, size, SMP_CACHE_BYTES, goal, limit);
566 return NULL; 646}
647#endif
567 648
568 start_nr = pfn_to_section_nr(PFN_DOWN(__pa(ptr))); 649void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
569 end_nr = pfn_to_section_nr(PFN_DOWN(__pa(ptr) + size)); 650 unsigned long align, unsigned long goal)
570 if (start_nr != section_nr || end_nr != section_nr) { 651{
571 printk(KERN_WARNING "alloc_bootmem failed on section %ld.\n", 652 void *ptr;
572 section_nr);
573 free_bootmem_core(pgdat->bdata, __pa(ptr), size);
574 ptr = NULL;
575 }
576 653
577 return ptr; 654 ptr = alloc_bootmem_core(pgdat->bdata, size, align, goal, 0);
655 if (ptr)
656 return ptr;
657
658 return __alloc_bootmem_nopanic(size, align, goal);
578} 659}
579#endif
580 660
581#ifndef ARCH_LOW_ADDRESS_LIMIT 661#ifndef ARCH_LOW_ADDRESS_LIMIT
582#define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL 662#define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL
583#endif 663#endif
584 664
665/**
666 * __alloc_bootmem_low - allocate low boot memory
667 * @size: size of the request in bytes
668 * @align: alignment of the region
669 * @goal: preferred starting address of the region
670 *
671 * The goal is dropped if it can not be satisfied and the allocation will
672 * fall back to memory below @goal.
673 *
674 * Allocation may happen on any node in the system.
675 *
676 * The function panics if the request can not be satisfied.
677 */
585void * __init __alloc_bootmem_low(unsigned long size, unsigned long align, 678void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
586 unsigned long goal) 679 unsigned long goal)
587{ 680{
588 bootmem_data_t *bdata; 681 return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT);
589 void *ptr;
590
591 list_for_each_entry(bdata, &bdata_list, list) {
592 ptr = __alloc_bootmem_core(bdata, size, align, goal,
593 ARCH_LOW_ADDRESS_LIMIT);
594 if (ptr)
595 return ptr;
596 }
597
598 /*
599 * Whoops, we cannot satisfy the allocation request.
600 */
601 printk(KERN_ALERT "low bootmem alloc of %lu bytes failed!\n", size);
602 panic("Out of low memory");
603 return NULL;
604} 682}
605 683
684/**
685 * __alloc_bootmem_low_node - allocate low boot memory from a specific node
686 * @pgdat: node to allocate from
687 * @size: size of the request in bytes
688 * @align: alignment of the region
689 * @goal: preferred starting address of the region
690 *
691 * The goal is dropped if it can not be satisfied and the allocation will
692 * fall back to memory below @goal.
693 *
694 * Allocation may fall back to any node in the system if the specified node
695 * can not hold the requested memory.
696 *
697 * The function panics if the request can not be satisfied.
698 */
606void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size, 699void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
607 unsigned long align, unsigned long goal) 700 unsigned long align, unsigned long goal)
608{ 701{
609 return __alloc_bootmem_core(pgdat->bdata, size, align, goal, 702 return ___alloc_bootmem_node(pgdat->bdata, size, align,
610 ARCH_LOW_ADDRESS_LIMIT); 703 goal, ARCH_LOW_ADDRESS_LIMIT);
611} 704}
generated by cgit v1.2.2 (git 2.25.0) at 2025-11-13 08:06:54 -0500