1/*
2 *  linux/mm/bootmem.c
3 *
4 *  Copyright (C) 1999 Ingo Molnar
5 *  Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
6 *
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.
10 */
11#include <linux/init.h>
12#include <linux/pfn.h>
13#include <linux/bootmem.h>
14#include <linux/module.h>
15
16#include <asm/bug.h>
17#include <asm/io.h>
18#include <asm/processor.h>
19
20#include "internal.h"
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;
27unsigned long min_low_pfn;
28unsigned long max_pfn;
29
30static LIST_HEAD(bdata_list);
31#ifdef CONFIG_CRASH_DUMP
32/*
33 * If we have booted due to a crash, max_pfn will be a very low value. We need
34 * to know the amount of memory that the previous kernel used.
35 */
36unsigned long saved_max_pfn;
37#endif
38
39/* return the number of _pages_ that will be allocated for the boot bitmap */
40unsigned long __init bootmem_bootmap_pages(unsigned long pages)
41{
42	unsigned long mapsize;
43
44	mapsize = (pages+7)/8;
45	mapsize = (mapsize + ~PAGE_MASK) & PAGE_MASK;
46	mapsize >>= PAGE_SHIFT;
47
48	return mapsize;
49}
50
51/*
52 * link bdata in order
53 */
54static void __init link_bootmem(bootmem_data_t *bdata)
55{
56	bootmem_data_t *ent;
57
58	if (list_empty(&bdata_list)) {
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}
71
72/*
73 * Given an initialised bdata, it returns the size of the boot bitmap
74 */
75static unsigned long __init get_mapsize(bootmem_data_t *bdata)
76{
77	unsigned long mapsize;
78	unsigned long start = PFN_DOWN(bdata->node_boot_start);
79	unsigned long end = bdata->node_low_pfn;
80
81	mapsize = ((end - start) + 7) / 8;
82	return ALIGN(mapsize, sizeof(long));
83}
84
85/*
86 * Called once to set up the allocator itself.
87 */
88static unsigned long __init init_bootmem_core(pg_data_t *pgdat,
89	unsigned long mapstart, unsigned long start, unsigned long end)
90{
91	bootmem_data_t *bdata = pgdat->bdata;
92	unsigned long mapsize;
93
94	bdata->node_bootmem_map = phys_to_virt(PFN_PHYS(mapstart));
95	bdata->node_boot_start = PFN_PHYS(start);
96	bdata->node_low_pfn = end;
97	link_bootmem(bdata);
98
99	/*
100	 * Initially all pages are reserved - setup_arch() has to
101	 * register free RAM areas explicitly.
102	 */
103	mapsize = get_mapsize(bdata);
104#if     !defined(CONFIG_HWSIM) || defined(CONFIG_HWSIM_ZMEM)
105	memset(bdata->node_bootmem_map, 0xff, mapsize);
106#endif
107
108	return mapsize;
109}
110
111/*
112 * Marks a particular physical memory range as unallocatable. Usable RAM
113 * might be used for boot-time allocations - or it might get added
114 * to the free page pool later on.
115 */
116static void __init reserve_bootmem_core(bootmem_data_t *bdata, unsigned long addr,
117					unsigned long size)
118{
119	unsigned long sidx, eidx;
120	unsigned long i;
121
122	/*
123	 * round up, partially reserved pages are considered
124	 * fully reserved.
125	 */
126	BUG_ON(!size);
127	BUG_ON(PFN_DOWN(addr) >= bdata->node_low_pfn);
128	BUG_ON(PFN_UP(addr + size) > bdata->node_low_pfn);
129
130	sidx = PFN_DOWN(addr - bdata->node_boot_start);
131	eidx = PFN_UP(addr + size - bdata->node_boot_start);
132
133	for (i = sidx; i < eidx; i++)
134		if (test_and_set_bit(i, bdata->node_bootmem_map)) {
135#ifdef CONFIG_DEBUG_BOOTMEM
136			printk("hm, page %08lx reserved twice.\n", i*PAGE_SIZE);
137#endif
138		}
139}
140
141static void __init free_bootmem_core(bootmem_data_t *bdata, unsigned long addr,
142				     unsigned long size)
143{
144	unsigned long sidx, eidx;
145	unsigned long i;
146
147	/*
148	 * round down end of usable mem, partially free pages are
149	 * considered reserved.
150	 */
151	BUG_ON(!size);
152	BUG_ON(PFN_DOWN(addr + size) > bdata->node_low_pfn);
153
154	if (addr < bdata->last_success)
155		bdata->last_success = addr;
156
157	/*
158	 * Round up the beginning of the address.
159	 */
160	sidx = PFN_UP(addr) - PFN_DOWN(bdata->node_boot_start);
161	eidx = PFN_DOWN(addr + size - bdata->node_boot_start);
162
163	for (i = sidx; i < eidx; i++) {
164		if (unlikely(!test_and_clear_bit(i, bdata->node_bootmem_map)))
165			BUG();
166	}
167}
168
169/*
170 * We 'merge' subsequent allocations to save space. We might 'lose'
171 * some fraction of a page if allocations cannot be satisfied due to
172 * size constraints on boxes where there is physical RAM space
173 * fragmentation - in these cases (mostly large memory boxes) this
174 * is not a problem.
175 *
176 * On low memory boxes we get it right in 100% of the cases.
177 *
178 * alignment has to be a power of 2 value.
179 *
180 * NOTE:  This function is _not_ reentrant.
181 */
182void * __init
183__alloc_bootmem_core(struct bootmem_data *bdata, unsigned long size,
184	      unsigned long align, unsigned long goal, unsigned long limit)
185{
186	unsigned long offset, remaining_size, areasize, preferred;
187	unsigned long i, start = 0, incr, eidx, end_pfn;
188	void *ret;
189
190	if (!size) {
191		printk("__alloc_bootmem_core(): zero-sized request\n");
192		BUG();
193	}
194	BUG_ON(align & (align-1));
195
196	if (limit && bdata->node_boot_start >= limit)
197		return NULL;
198
199	/* on nodes without memory - bootmem_map is NULL */
200	if (!bdata->node_bootmem_map)
201		return NULL;
202
203	end_pfn = bdata->node_low_pfn;
204	limit = PFN_DOWN(limit);
205	if (limit && end_pfn > limit)
206		end_pfn = limit;
207
208	eidx = end_pfn - PFN_DOWN(bdata->node_boot_start);
209	offset = 0;
210	if (align && (bdata->node_boot_start & (align - 1UL)) != 0)
211		offset = align - (bdata->node_boot_start & (align - 1UL));
212	offset = PFN_DOWN(offset);
213
214	/*
215	 * We try to allocate bootmem pages above 'goal'
216	 * first, then we try to allocate lower pages.
217	 */
218	if (goal && goal >= bdata->node_boot_start && PFN_DOWN(goal) < end_pfn) {
219		preferred = goal - bdata->node_boot_start;
220
221		if (bdata->last_success >= preferred)
222			if (!limit || (limit && limit > bdata->last_success))
223				preferred = bdata->last_success;
224	} else
225		preferred = 0;
226
227	preferred = PFN_DOWN(ALIGN(preferred, align)) + offset;
228	areasize = (size + PAGE_SIZE-1) / PAGE_SIZE;
229	incr = align >> PAGE_SHIFT ? : 1;
230
231restart_scan:
232	for (i = preferred; i < eidx; i += incr) {
233		unsigned long j;
234		i = find_next_zero_bit(bdata->node_bootmem_map, eidx, i);
235		i = ALIGN(i, incr);
236		if (i >= eidx)
237			break;
238		if (test_bit(i, bdata->node_bootmem_map))
239			continue;
240		for (j = i + 1; j < i + areasize; ++j) {
241			if (j >= eidx)
242				goto fail_block;
243			if (test_bit(j, bdata->node_bootmem_map))
244				goto fail_block;
245		}
246		start = i;
247		goto found;
248	fail_block:
249		i = ALIGN(j, incr);
250	}
251
252	if (preferred > offset) {
253		preferred = offset;
254		goto restart_scan;
255	}
256	return NULL;
257
258found:
259	bdata->last_success = PFN_PHYS(start);
260	BUG_ON(start >= eidx);
261
262	/*
263	 * Is the next page of the previous allocation-end the start
264	 * of this allocation's buffer? If yes then we can 'merge'
265	 * the previous partial page with this allocation.
266	 */
267	if (align < PAGE_SIZE &&
268	    bdata->last_offset && bdata->last_pos+1 == start) {
269		offset = ALIGN(bdata->last_offset, align);
270		BUG_ON(offset > PAGE_SIZE);
271		remaining_size = PAGE_SIZE - offset;
272		if (size < remaining_size) {
273			areasize = 0;
274			/* last_pos unchanged */
275			bdata->last_offset = offset + size;
276			ret = phys_to_virt(bdata->last_pos * PAGE_SIZE +
277					   offset +
278					   bdata->node_boot_start);
279		} else {
280			remaining_size = size - remaining_size;
281			areasize = (remaining_size + PAGE_SIZE-1) / PAGE_SIZE;
282			ret = phys_to_virt(bdata->last_pos * PAGE_SIZE +
283					   offset +
284					   bdata->node_boot_start);
285			bdata->last_pos = start + areasize - 1;
286			bdata->last_offset = remaining_size;
287		}
288		bdata->last_offset &= ~PAGE_MASK;
289	} else {
290		bdata->last_pos = start + areasize - 1;
291		bdata->last_offset = size & ~PAGE_MASK;
292		ret = phys_to_virt(start * PAGE_SIZE + bdata->node_boot_start);
293	}
294
295	/*
296	 * Reserve the area now:
297	 */
298	for (i = start; i < start + areasize; i++)
299		if (unlikely(test_and_set_bit(i, bdata->node_bootmem_map)))
300			BUG();
301#ifndef CONFIG_HWSIM
302	memset(ret, 0, size);
303#endif
304	return ret;
305}
306
307static unsigned long __init free_all_bootmem_core(pg_data_t *pgdat)
308{
309	struct page *page;
310	unsigned long pfn;
311	bootmem_data_t *bdata = pgdat->bdata;
312	unsigned long i, count, total = 0;
313	unsigned long idx;
314	unsigned long *map;
315	int gofast = 0;
316
317	BUG_ON(!bdata->node_bootmem_map);
318
319	count = 0;
320	/* first extant page of the node */
321	pfn = PFN_DOWN(bdata->node_boot_start);
322	idx = bdata->node_low_pfn - pfn;
323	map = bdata->node_bootmem_map;
324	/* Check physaddr is O(LOG2(BITS_PER_LONG)) page aligned */
325	if (bdata->node_boot_start == 0 ||
326	    ffs(bdata->node_boot_start) - PAGE_SHIFT > ffs(BITS_PER_LONG))
327		gofast = 1;
328	for (i = 0; i < idx; ) {
329		unsigned long v = ~map[i / BITS_PER_LONG];
330
331		if (gofast && v == ~0UL) {
332			int order;
333
334			page = pfn_to_page(pfn);
335			count += BITS_PER_LONG;
336			order = ffs(BITS_PER_LONG) - 1;
337			__free_pages_bootmem(page, order);
338			i += BITS_PER_LONG;
339			page += BITS_PER_LONG;
340		} else if (v) {
341			unsigned long m;
342
343			page = pfn_to_page(pfn);
344			for (m = 1; m && i < idx; m<<=1, page++, i++) {
345				if (v & m) {
346					count++;
347					__free_pages_bootmem(page, 0);
348				}
349			}
350		} else {
351			i += BITS_PER_LONG;
352		}
353		pfn += BITS_PER_LONG;
354	}
355	total += count;
356
357	/*
358	 * Now free the allocator bitmap itself, it's not
359	 * needed anymore:
360	 */
361	page = virt_to_page(bdata->node_bootmem_map);
362	count = 0;
363	idx = (get_mapsize(bdata) + PAGE_SIZE-1) >> PAGE_SHIFT;
364	for (i = 0; i < idx; i++, page++) {
365		__free_pages_bootmem(page, 0);
366		count++;
367	}
368	total += count;
369	bdata->node_bootmem_map = NULL;
370
371	return total;
372}
373
374unsigned long __init init_bootmem_node(pg_data_t *pgdat, unsigned long freepfn,
375				unsigned long startpfn, unsigned long endpfn)
376{
377	return init_bootmem_core(pgdat, freepfn, startpfn, endpfn);
378}
379
380void __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
381				 unsigned long size)
382{
383	reserve_bootmem_core(pgdat->bdata, physaddr, size);
384}
385
386void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
387			      unsigned long size)
388{
389	free_bootmem_core(pgdat->bdata, physaddr, size);
390}
391
392unsigned long __init free_all_bootmem_node(pg_data_t *pgdat)
393{
394	return free_all_bootmem_core(pgdat);
395}
396
397unsigned long __init init_bootmem(unsigned long start, unsigned long pages)
398{
399	max_low_pfn = pages;
400	min_low_pfn = start;
401	return init_bootmem_core(NODE_DATA(0), start, 0, pages);
402}
403
404#ifndef CONFIG_HAVE_ARCH_BOOTMEM_NODE
405void __init reserve_bootmem(unsigned long addr, unsigned long size)
406{
407	reserve_bootmem_core(NODE_DATA(0)->bdata, addr, size);
408}
409#endif /* !CONFIG_HAVE_ARCH_BOOTMEM_NODE */
410
411void __init free_bootmem(unsigned long addr, unsigned long size)
412{
413	free_bootmem_core(NODE_DATA(0)->bdata, addr, size);
414}
415
416unsigned long __init free_all_bootmem(void)
417{
418	return free_all_bootmem_core(NODE_DATA(0));
419}
420
421void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
422				      unsigned long goal)
423{
424	bootmem_data_t *bdata;
425	void *ptr;
426
427	list_for_each_entry(bdata, &bdata_list, list) {
428		ptr = __alloc_bootmem_core(bdata, size, align, goal, 0);
429		if (ptr)
430			return ptr;
431	}
432	return NULL;
433}
434
435void * __init __alloc_bootmem(unsigned long size, unsigned long align,
436			      unsigned long goal)
437{
438	void *mem = __alloc_bootmem_nopanic(size,align,goal);
439
440	if (mem)
441		return mem;
442	/*
443	 * Whoops, we cannot satisfy the allocation request.
444	 */
445	printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
446	panic("Out of memory");
447	return NULL;
448}
449
450
451void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
452				   unsigned long align, unsigned long goal)
453{
454	void *ptr;
455
456	ptr = __alloc_bootmem_core(pgdat->bdata, size, align, goal, 0);
457	if (ptr)
458		return ptr;
459
460	return __alloc_bootmem(size, align, goal);
461}
462
463#ifndef ARCH_LOW_ADDRESS_LIMIT
464#define ARCH_LOW_ADDRESS_LIMIT	0xffffffffUL
465#endif
466
467void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
468				  unsigned long goal)
469{
470	bootmem_data_t *bdata;
471	void *ptr;
472
473	list_for_each_entry(bdata, &bdata_list, list) {
474		ptr = __alloc_bootmem_core(bdata, size, align, goal,
475						ARCH_LOW_ADDRESS_LIMIT);
476		if (ptr)
477			return ptr;
478	}
479
480	/*
481	 * Whoops, we cannot satisfy the allocation request.
482	 */
483	printk(KERN_ALERT "low bootmem alloc of %lu bytes failed!\n", size);
484	panic("Out of low memory");
485	return NULL;
486}
487
488void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
489				       unsigned long align, unsigned long goal)
490{
491	return __alloc_bootmem_core(pgdat->bdata, size, align, goal,
492				    ARCH_LOW_ADDRESS_LIMIT);
493}
494