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_SYMBOL(max_pfn); /* This is exported so
33 * dma_get_required_mask(), which uses
34 * it, can be an inline function */
36 #ifdef CONFIG_CRASH_DUMP
38 * If we have booted due to a crash, max_pfn will be a very low value. We need
39 * to know the amount of memory that the previous kernel used.
41 unsigned long saved_max_pfn;
44 /* return the number of _pages_ that will be allocated for the boot bitmap */
45 unsigned long __init bootmem_bootmap_pages (unsigned long pages)
47 unsigned long mapsize;
49 mapsize = (pages+7)/8;
50 mapsize = (mapsize + ~PAGE_MASK) & PAGE_MASK;
51 mapsize >>= PAGE_SHIFT;
57 * Called once to set up the allocator itself.
59 static unsigned long __init init_bootmem_core (pg_data_t *pgdat,
60 unsigned long mapstart, unsigned long start, unsigned long end)
62 bootmem_data_t *bdata = pgdat->bdata;
63 unsigned long mapsize = ((end - start)+7)/8;
65 pgdat->pgdat_next = pgdat_list;
68 mapsize = ALIGN(mapsize, sizeof(long));
69 bdata->node_bootmem_map = phys_to_virt(mapstart << PAGE_SHIFT);
70 bdata->node_boot_start = (start << PAGE_SHIFT);
71 bdata->node_low_pfn = end;
74 * Initially all pages are reserved - setup_arch() has to
75 * register free RAM areas explicitly.
77 memset(bdata->node_bootmem_map, 0xff, mapsize);
83 * Marks a particular physical memory range as unallocatable. Usable RAM
84 * might be used for boot-time allocations - or it might get added
85 * to the free page pool later on.
87 static void __init reserve_bootmem_core(bootmem_data_t *bdata, unsigned long addr, unsigned long size)
91 * round up, partially reserved pages are considered
94 unsigned long sidx = (addr - bdata->node_boot_start)/PAGE_SIZE;
95 unsigned long eidx = (addr + size - bdata->node_boot_start +
96 PAGE_SIZE-1)/PAGE_SIZE;
97 unsigned long end = (addr + size + PAGE_SIZE-1)/PAGE_SIZE;
100 BUG_ON(sidx >= eidx);
101 BUG_ON((addr >> PAGE_SHIFT) >= bdata->node_low_pfn);
102 BUG_ON(end > bdata->node_low_pfn);
104 for (i = sidx; i < eidx; i++)
105 if (test_and_set_bit(i, bdata->node_bootmem_map)) {
106 #ifdef CONFIG_DEBUG_BOOTMEM
107 printk("hm, page %08lx reserved twice.\n", i*PAGE_SIZE);
112 static void __init free_bootmem_core(bootmem_data_t *bdata, unsigned long addr, unsigned long size)
117 * round down end of usable mem, partially free pages are
118 * considered reserved.
121 unsigned long eidx = (addr + size - bdata->node_boot_start)/PAGE_SIZE;
122 unsigned long end = (addr + size)/PAGE_SIZE;
125 BUG_ON(end > bdata->node_low_pfn);
127 if (addr < bdata->last_success)
128 bdata->last_success = addr;
131 * Round up the beginning of the address.
133 start = (addr + PAGE_SIZE-1) / PAGE_SIZE;
134 sidx = start - (bdata->node_boot_start/PAGE_SIZE);
136 for (i = sidx; i < eidx; i++) {
137 if (unlikely(!test_and_clear_bit(i, bdata->node_bootmem_map)))
143 * We 'merge' subsequent allocations to save space. We might 'lose'
144 * some fraction of a page if allocations cannot be satisfied due to
145 * size constraints on boxes where there is physical RAM space
146 * fragmentation - in these cases (mostly large memory boxes) this
149 * On low memory boxes we get it right in 100% of the cases.
151 * alignment has to be a power of 2 value.
153 * NOTE: This function is _not_ reentrant.
156 __alloc_bootmem_core(struct bootmem_data *bdata, unsigned long size,
157 unsigned long align, unsigned long goal, unsigned long limit)
159 unsigned long offset, remaining_size, areasize, preferred;
160 unsigned long i, start = 0, incr, eidx, end_pfn = bdata->node_low_pfn;
164 printk("__alloc_bootmem_core(): zero-sized request\n");
167 BUG_ON(align & (align-1));
169 if (limit && bdata->node_boot_start >= limit)
173 if (limit && end_pfn > limit)
176 eidx = end_pfn - (bdata->node_boot_start >> PAGE_SHIFT);
179 (bdata->node_boot_start & (align - 1UL)) != 0)
180 offset = (align - (bdata->node_boot_start & (align - 1UL)));
181 offset >>= PAGE_SHIFT;
184 * We try to allocate bootmem pages above 'goal'
185 * first, then we try to allocate lower pages.
187 if (goal && (goal >= bdata->node_boot_start) &&
188 ((goal >> PAGE_SHIFT) < end_pfn)) {
189 preferred = goal - bdata->node_boot_start;
191 if (bdata->last_success >= preferred)
192 if (!limit || (limit && limit > bdata->last_success))
193 preferred = bdata->last_success;
197 preferred = ALIGN(preferred, align) >> PAGE_SHIFT;
199 areasize = (size+PAGE_SIZE-1)/PAGE_SIZE;
200 incr = align >> PAGE_SHIFT ? : 1;
203 for (i = preferred; i < eidx; i += incr) {
205 i = find_next_zero_bit(bdata->node_bootmem_map, eidx, i);
207 if (test_bit(i, bdata->node_bootmem_map))
209 for (j = i + 1; j < i + areasize; ++j) {
212 if (test_bit (j, bdata->node_bootmem_map))
221 if (preferred > offset) {
228 bdata->last_success = start << PAGE_SHIFT;
229 BUG_ON(start >= eidx);
232 * Is the next page of the previous allocation-end the start
233 * of this allocation's buffer? If yes then we can 'merge'
234 * the previous partial page with this allocation.
236 if (align < PAGE_SIZE &&
237 bdata->last_offset && bdata->last_pos+1 == start) {
238 offset = ALIGN(bdata->last_offset, align);
239 BUG_ON(offset > PAGE_SIZE);
240 remaining_size = PAGE_SIZE-offset;
241 if (size < remaining_size) {
243 /* last_pos unchanged */
244 bdata->last_offset = offset+size;
245 ret = phys_to_virt(bdata->last_pos*PAGE_SIZE + offset +
246 bdata->node_boot_start);
248 remaining_size = size - remaining_size;
249 areasize = (remaining_size+PAGE_SIZE-1)/PAGE_SIZE;
250 ret = phys_to_virt(bdata->last_pos*PAGE_SIZE + offset +
251 bdata->node_boot_start);
252 bdata->last_pos = start+areasize-1;
253 bdata->last_offset = remaining_size;
255 bdata->last_offset &= ~PAGE_MASK;
257 bdata->last_pos = start + areasize - 1;
258 bdata->last_offset = size & ~PAGE_MASK;
259 ret = phys_to_virt(start * PAGE_SIZE + bdata->node_boot_start);
263 * Reserve the area now:
265 for (i = start; i < start+areasize; i++)
266 if (unlikely(test_and_set_bit(i, bdata->node_bootmem_map)))
268 memset(ret, 0, size);
272 static unsigned long __init free_all_bootmem_core(pg_data_t *pgdat)
276 bootmem_data_t *bdata = pgdat->bdata;
277 unsigned long i, count, total = 0;
282 BUG_ON(!bdata->node_bootmem_map);
285 /* first extant page of the node */
286 pfn = bdata->node_boot_start >> PAGE_SHIFT;
287 idx = bdata->node_low_pfn - (bdata->node_boot_start >> PAGE_SHIFT);
288 map = bdata->node_bootmem_map;
289 /* Check physaddr is O(LOG2(BITS_PER_LONG)) page aligned */
290 if (bdata->node_boot_start == 0 ||
291 ffs(bdata->node_boot_start) - PAGE_SHIFT > ffs(BITS_PER_LONG))
293 for (i = 0; i < idx; ) {
294 unsigned long v = ~map[i / BITS_PER_LONG];
296 if (gofast && v == ~0UL) {
299 page = pfn_to_page(pfn);
300 count += BITS_PER_LONG;
301 __ClearPageReserved(page);
302 order = ffs(BITS_PER_LONG) - 1;
303 set_page_refs(page, order);
304 for (j = 1; j < BITS_PER_LONG; j++) {
305 if (j + 16 < BITS_PER_LONG)
306 prefetchw(page + j + 16);
307 __ClearPageReserved(page + j);
308 set_page_count(page + j, 0);
310 __free_pages(page, order);
312 page += BITS_PER_LONG;
316 page = pfn_to_page(pfn);
317 for (m = 1; m && i < idx; m<<=1, page++, i++) {
320 __ClearPageReserved(page);
321 set_page_refs(page, 0);
328 pfn += BITS_PER_LONG;
333 * Now free the allocator bitmap itself, it's not
336 page = virt_to_page(bdata->node_bootmem_map);
338 for (i = 0; i < ((bdata->node_low_pfn-(bdata->node_boot_start >> PAGE_SHIFT))/8 + PAGE_SIZE-1)/PAGE_SIZE; i++,page++) {
340 __ClearPageReserved(page);
341 set_page_count(page, 1);
345 bdata->node_bootmem_map = NULL;
350 unsigned long __init init_bootmem_node (pg_data_t *pgdat, unsigned long freepfn, unsigned long startpfn, unsigned long endpfn)
352 return(init_bootmem_core(pgdat, freepfn, startpfn, endpfn));
355 void __init reserve_bootmem_node (pg_data_t *pgdat, unsigned long physaddr, unsigned long size)
357 reserve_bootmem_core(pgdat->bdata, physaddr, size);
360 void __init free_bootmem_node (pg_data_t *pgdat, unsigned long physaddr, unsigned long size)
362 free_bootmem_core(pgdat->bdata, physaddr, size);
365 unsigned long __init free_all_bootmem_node (pg_data_t *pgdat)
367 return(free_all_bootmem_core(pgdat));
370 unsigned long __init init_bootmem (unsigned long start, unsigned long pages)
374 return(init_bootmem_core(NODE_DATA(0), start, 0, pages));
377 #ifndef CONFIG_HAVE_ARCH_BOOTMEM_NODE
378 void __init reserve_bootmem (unsigned long addr, unsigned long size)
380 reserve_bootmem_core(NODE_DATA(0)->bdata, addr, size);
382 #endif /* !CONFIG_HAVE_ARCH_BOOTMEM_NODE */
384 void __init free_bootmem (unsigned long addr, unsigned long size)
386 free_bootmem_core(NODE_DATA(0)->bdata, addr, size);
389 unsigned long __init free_all_bootmem (void)
391 return(free_all_bootmem_core(NODE_DATA(0)));
394 void * __init __alloc_bootmem_limit (unsigned long size, unsigned long align, unsigned long goal,
397 pg_data_t *pgdat = pgdat_list;
400 for_each_pgdat(pgdat)
401 if ((ptr = __alloc_bootmem_core(pgdat->bdata, size,
402 align, goal, limit)))
406 * Whoops, we cannot satisfy the allocation request.
408 printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
409 panic("Out of memory");
414 void * __init __alloc_bootmem_node_limit (pg_data_t *pgdat, unsigned long size, unsigned long align,
415 unsigned long goal, unsigned long limit)
419 ptr = __alloc_bootmem_core(pgdat->bdata, size, align, goal, limit);
423 return __alloc_bootmem_limit(size, align, goal, limit);