4 * Copyright (C) 1999 Ingo Molnar
5 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
7 * simple boot-time physical memory area allocator and
8 * free memory collector. It's used to deal with reserved
9 * system memory and memory holes as well.
13 #include <linux/kernel_stat.h>
14 #include <linux/swap.h>
15 #include <linux/interrupt.h>
16 #include <linux/init.h>
17 #include <linux/bootmem.h>
18 #include <linux/mmzone.h>
19 #include <linux/module.h>
25 * Access to this subsystem has to be serialized externally. (this is
26 * true for the boot process anyway)
28 unsigned long max_low_pfn;
29 unsigned long min_low_pfn;
30 unsigned long max_pfn;
32 EXPORT_UNUSED_SYMBOL(max_pfn); /* June 2006 */
34 static LIST_HEAD(bdata_list);
35 #ifdef CONFIG_CRASH_DUMP
37 * If we have booted due to a crash, max_pfn will be a very low value. We need
38 * to know the amount of memory that the previous kernel used.
40 unsigned long saved_max_pfn;
43 /* return the number of _pages_ that will be allocated for the boot bitmap */
44 unsigned long __init bootmem_bootmap_pages (unsigned long pages)
46 unsigned long mapsize;
48 mapsize = (pages+7)/8;
49 mapsize = (mapsize + ~PAGE_MASK) & PAGE_MASK;
50 mapsize >>= PAGE_SHIFT;
57 static void link_bootmem(bootmem_data_t *bdata)
60 if (list_empty(&bdata_list)) {
61 list_add(&bdata->list, &bdata_list);
65 list_for_each_entry(ent, &bdata_list, list) {
66 if (bdata->node_boot_start < ent->node_boot_start) {
67 list_add_tail(&bdata->list, &ent->list);
71 list_add_tail(&bdata->list, &bdata_list);
77 * Called once to set up the allocator itself.
79 static unsigned long __init init_bootmem_core (pg_data_t *pgdat,
80 unsigned long mapstart, unsigned long start, unsigned long end)
82 bootmem_data_t *bdata = pgdat->bdata;
83 unsigned long mapsize = ((end - start)+7)/8;
85 mapsize = ALIGN(mapsize, sizeof(long));
86 bdata->node_bootmem_map = phys_to_virt(mapstart << PAGE_SHIFT);
87 bdata->node_boot_start = (start << PAGE_SHIFT);
88 bdata->node_low_pfn = end;
92 * Initially all pages are reserved - setup_arch() has to
93 * register free RAM areas explicitly.
95 memset(bdata->node_bootmem_map, 0xff, mapsize);
101 * Marks a particular physical memory range as unallocatable. Usable RAM
102 * might be used for boot-time allocations - or it might get added
103 * to the free page pool later on.
105 static void __init reserve_bootmem_core(bootmem_data_t *bdata, unsigned long addr, unsigned long size)
109 * round up, partially reserved pages are considered
112 unsigned long sidx = (addr - bdata->node_boot_start)/PAGE_SIZE;
113 unsigned long eidx = (addr + size - bdata->node_boot_start +
114 PAGE_SIZE-1)/PAGE_SIZE;
115 unsigned long end = (addr + size + PAGE_SIZE-1)/PAGE_SIZE;
118 BUG_ON(sidx >= eidx);
119 BUG_ON((addr >> PAGE_SHIFT) >= bdata->node_low_pfn);
120 BUG_ON(end > bdata->node_low_pfn);
122 for (i = sidx; i < eidx; i++)
123 if (test_and_set_bit(i, bdata->node_bootmem_map)) {
124 #ifdef CONFIG_DEBUG_BOOTMEM
125 printk("hm, page %08lx reserved twice.\n", i*PAGE_SIZE);
130 static void __init free_bootmem_core(bootmem_data_t *bdata, unsigned long addr, unsigned long size)
135 * round down end of usable mem, partially free pages are
136 * considered reserved.
139 unsigned long eidx = (addr + size - bdata->node_boot_start)/PAGE_SIZE;
140 unsigned long end = (addr + size)/PAGE_SIZE;
143 BUG_ON(end > bdata->node_low_pfn);
145 if (addr < bdata->last_success)
146 bdata->last_success = addr;
149 * Round up the beginning of the address.
151 start = (addr + PAGE_SIZE-1) / PAGE_SIZE;
152 sidx = start - (bdata->node_boot_start/PAGE_SIZE);
154 for (i = sidx; i < eidx; i++) {
155 if (unlikely(!test_and_clear_bit(i, bdata->node_bootmem_map)))
161 * We 'merge' subsequent allocations to save space. We might 'lose'
162 * some fraction of a page if allocations cannot be satisfied due to
163 * size constraints on boxes where there is physical RAM space
164 * fragmentation - in these cases (mostly large memory boxes) this
167 * On low memory boxes we get it right in 100% of the cases.
169 * alignment has to be a power of 2 value.
171 * NOTE: This function is _not_ reentrant.
174 __alloc_bootmem_core(struct bootmem_data *bdata, unsigned long size,
175 unsigned long align, unsigned long goal, unsigned long limit)
177 unsigned long offset, remaining_size, areasize, preferred;
178 unsigned long i, start = 0, incr, eidx, end_pfn = bdata->node_low_pfn;
182 printk("__alloc_bootmem_core(): zero-sized request\n");
185 BUG_ON(align & (align-1));
187 if (limit && bdata->node_boot_start >= limit)
191 if (limit && end_pfn > limit)
194 eidx = end_pfn - (bdata->node_boot_start >> PAGE_SHIFT);
197 (bdata->node_boot_start & (align - 1UL)) != 0)
198 offset = (align - (bdata->node_boot_start & (align - 1UL)));
199 offset >>= PAGE_SHIFT;
202 * We try to allocate bootmem pages above 'goal'
203 * first, then we try to allocate lower pages.
205 if (goal && (goal >= bdata->node_boot_start) &&
206 ((goal >> PAGE_SHIFT) < end_pfn)) {
207 preferred = goal - bdata->node_boot_start;
209 if (bdata->last_success >= preferred)
210 if (!limit || (limit && limit > bdata->last_success))
211 preferred = bdata->last_success;
215 preferred = ALIGN(preferred, align) >> PAGE_SHIFT;
217 areasize = (size+PAGE_SIZE-1)/PAGE_SIZE;
218 incr = align >> PAGE_SHIFT ? : 1;
221 for (i = preferred; i < eidx; i += incr) {
223 i = find_next_zero_bit(bdata->node_bootmem_map, eidx, i);
227 if (test_bit(i, bdata->node_bootmem_map))
229 for (j = i + 1; j < i + areasize; ++j) {
232 if (test_bit (j, bdata->node_bootmem_map))
241 if (preferred > offset) {
248 bdata->last_success = start << PAGE_SHIFT;
249 BUG_ON(start >= eidx);
252 * Is the next page of the previous allocation-end the start
253 * of this allocation's buffer? If yes then we can 'merge'
254 * the previous partial page with this allocation.
256 if (align < PAGE_SIZE &&
257 bdata->last_offset && bdata->last_pos+1 == start) {
258 offset = ALIGN(bdata->last_offset, align);
259 BUG_ON(offset > PAGE_SIZE);
260 remaining_size = PAGE_SIZE-offset;
261 if (size < remaining_size) {
263 /* last_pos unchanged */
264 bdata->last_offset = offset+size;
265 ret = phys_to_virt(bdata->last_pos*PAGE_SIZE + offset +
266 bdata->node_boot_start);
268 remaining_size = size - remaining_size;
269 areasize = (remaining_size+PAGE_SIZE-1)/PAGE_SIZE;
270 ret = phys_to_virt(bdata->last_pos*PAGE_SIZE + offset +
271 bdata->node_boot_start);
272 bdata->last_pos = start+areasize-1;
273 bdata->last_offset = remaining_size;
275 bdata->last_offset &= ~PAGE_MASK;
277 bdata->last_pos = start + areasize - 1;
278 bdata->last_offset = size & ~PAGE_MASK;
279 ret = phys_to_virt(start * PAGE_SIZE + bdata->node_boot_start);
283 * Reserve the area now:
285 for (i = start; i < start+areasize; i++)
286 if (unlikely(test_and_set_bit(i, bdata->node_bootmem_map)))
288 memset(ret, 0, size);
292 static unsigned long __init free_all_bootmem_core(pg_data_t *pgdat)
296 bootmem_data_t *bdata = pgdat->bdata;
297 unsigned long i, count, total = 0;
302 BUG_ON(!bdata->node_bootmem_map);
305 /* first extant page of the node */
306 pfn = bdata->node_boot_start >> PAGE_SHIFT;
307 idx = bdata->node_low_pfn - (bdata->node_boot_start >> PAGE_SHIFT);
308 map = bdata->node_bootmem_map;
309 /* Check physaddr is O(LOG2(BITS_PER_LONG)) page aligned */
310 if (bdata->node_boot_start == 0 ||
311 ffs(bdata->node_boot_start) - PAGE_SHIFT > ffs(BITS_PER_LONG))
313 for (i = 0; i < idx; ) {
314 unsigned long v = ~map[i / BITS_PER_LONG];
316 if (gofast && v == ~0UL) {
319 page = pfn_to_page(pfn);
320 count += BITS_PER_LONG;
321 order = ffs(BITS_PER_LONG) - 1;
322 __free_pages_bootmem(page, order);
324 page += BITS_PER_LONG;
328 page = pfn_to_page(pfn);
329 for (m = 1; m && i < idx; m<<=1, page++, i++) {
332 __free_pages_bootmem(page, 0);
338 pfn += BITS_PER_LONG;
343 * Now free the allocator bitmap itself, it's not
346 page = virt_to_page(bdata->node_bootmem_map);
348 for (i = 0; i < ((bdata->node_low_pfn-(bdata->node_boot_start >> PAGE_SHIFT))/8 + PAGE_SIZE-1)/PAGE_SIZE; i++,page++) {
350 __free_pages_bootmem(page, 0);
353 bdata->node_bootmem_map = NULL;
358 unsigned long __init init_bootmem_node (pg_data_t *pgdat, unsigned long freepfn, unsigned long startpfn, unsigned long endpfn)
360 return(init_bootmem_core(pgdat, freepfn, startpfn, endpfn));
363 void __init reserve_bootmem_node (pg_data_t *pgdat, unsigned long physaddr, unsigned long size)
365 reserve_bootmem_core(pgdat->bdata, physaddr, size);
368 void __init free_bootmem_node (pg_data_t *pgdat, unsigned long physaddr, unsigned long size)
370 free_bootmem_core(pgdat->bdata, physaddr, size);
373 unsigned long __init free_all_bootmem_node (pg_data_t *pgdat)
375 return(free_all_bootmem_core(pgdat));
378 unsigned long __init init_bootmem (unsigned long start, unsigned long pages)
382 return(init_bootmem_core(NODE_DATA(0), start, 0, pages));
385 #ifndef CONFIG_HAVE_ARCH_BOOTMEM_NODE
386 void __init reserve_bootmem (unsigned long addr, unsigned long size)
388 reserve_bootmem_core(NODE_DATA(0)->bdata, addr, size);
390 #endif /* !CONFIG_HAVE_ARCH_BOOTMEM_NODE */
392 void __init free_bootmem (unsigned long addr, unsigned long size)
394 free_bootmem_core(NODE_DATA(0)->bdata, addr, size);
397 unsigned long __init free_all_bootmem (void)
399 return(free_all_bootmem_core(NODE_DATA(0)));
402 void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align, unsigned long goal)
404 bootmem_data_t *bdata;
407 list_for_each_entry(bdata, &bdata_list, list)
408 if ((ptr = __alloc_bootmem_core(bdata, size, align, goal, 0)))
413 void * __init __alloc_bootmem(unsigned long size, unsigned long align, unsigned long goal)
415 void *mem = __alloc_bootmem_nopanic(size,align,goal);
419 * Whoops, we cannot satisfy the allocation request.
421 printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
422 panic("Out of memory");
427 void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size, unsigned long align,
432 ptr = __alloc_bootmem_core(pgdat->bdata, size, align, goal, 0);
436 return __alloc_bootmem(size, align, goal);
439 #define LOW32LIMIT 0xffffffff
441 void * __init __alloc_bootmem_low(unsigned long size, unsigned long align, unsigned long goal)
443 bootmem_data_t *bdata;
446 list_for_each_entry(bdata, &bdata_list, list)
447 if ((ptr = __alloc_bootmem_core(bdata, size,
448 align, goal, LOW32LIMIT)))
452 * Whoops, we cannot satisfy the allocation request.
454 printk(KERN_ALERT "low bootmem alloc of %lu bytes failed!\n", size);
455 panic("Out of low memory");
459 void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
460 unsigned long align, unsigned long goal)
462 return __alloc_bootmem_core(pgdat->bdata, size, align, goal, LOW32LIMIT);