Merge git://git.kernel.org/pub/scm/linux/kernel/git/lethal/sh-2.6
[linux-2.6] / arch / arm / mm / init.c
1 /*
2  *  linux/arch/arm/mm/init.c
3  *
4  *  Copyright (C) 1995-2005 Russell King
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 as
8  * published by the Free Software Foundation.
9  */
10 #include <linux/kernel.h>
11 #include <linux/errno.h>
12 #include <linux/swap.h>
13 #include <linux/init.h>
14 #include <linux/bootmem.h>
15 #include <linux/mman.h>
16 #include <linux/nodemask.h>
17 #include <linux/initrd.h>
18 #include <linux/highmem.h>
19
20 #include <asm/mach-types.h>
21 #include <asm/sections.h>
22 #include <asm/setup.h>
23 #include <asm/sizes.h>
24 #include <asm/tlb.h>
25
26 #include <asm/mach/arch.h>
27 #include <asm/mach/map.h>
28
29 #include "mm.h"
30
31 static unsigned long phys_initrd_start __initdata = 0;
32 static unsigned long phys_initrd_size __initdata = 0;
33
34 static void __init early_initrd(char **p)
35 {
36         unsigned long start, size;
37
38         start = memparse(*p, p);
39         if (**p == ',') {
40                 size = memparse((*p) + 1, p);
41
42                 phys_initrd_start = start;
43                 phys_initrd_size = size;
44         }
45 }
46 __early_param("initrd=", early_initrd);
47
48 static int __init parse_tag_initrd(const struct tag *tag)
49 {
50         printk(KERN_WARNING "ATAG_INITRD is deprecated; "
51                 "please update your bootloader.\n");
52         phys_initrd_start = __virt_to_phys(tag->u.initrd.start);
53         phys_initrd_size = tag->u.initrd.size;
54         return 0;
55 }
56
57 __tagtable(ATAG_INITRD, parse_tag_initrd);
58
59 static int __init parse_tag_initrd2(const struct tag *tag)
60 {
61         phys_initrd_start = tag->u.initrd.start;
62         phys_initrd_size = tag->u.initrd.size;
63         return 0;
64 }
65
66 __tagtable(ATAG_INITRD2, parse_tag_initrd2);
67
68 /*
69  * This keeps memory configuration data used by a couple memory
70  * initialization functions, as well as show_mem() for the skipping
71  * of holes in the memory map.  It is populated by arm_add_memory().
72  */
73 struct meminfo meminfo;
74
75 void show_mem(void)
76 {
77         int free = 0, total = 0, reserved = 0;
78         int shared = 0, cached = 0, slab = 0, node, i;
79         struct meminfo * mi = &meminfo;
80
81         printk("Mem-info:\n");
82         show_free_areas();
83         for_each_online_node(node) {
84                 pg_data_t *n = NODE_DATA(node);
85                 struct page *map = pgdat_page_nr(n, 0) - n->node_start_pfn;
86
87                 for_each_nodebank (i,mi,node) {
88                         struct membank *bank = &mi->bank[i];
89                         unsigned int pfn1, pfn2;
90                         struct page *page, *end;
91
92                         pfn1 = bank_pfn_start(bank);
93                         pfn2 = bank_pfn_end(bank);
94
95                         page = map + pfn1;
96                         end  = map + pfn2;
97
98                         do {
99                                 total++;
100                                 if (PageReserved(page))
101                                         reserved++;
102                                 else if (PageSwapCache(page))
103                                         cached++;
104                                 else if (PageSlab(page))
105                                         slab++;
106                                 else if (!page_count(page))
107                                         free++;
108                                 else
109                                         shared += page_count(page) - 1;
110                                 page++;
111                         } while (page < end);
112                 }
113         }
114
115         printk("%d pages of RAM\n", total);
116         printk("%d free pages\n", free);
117         printk("%d reserved pages\n", reserved);
118         printk("%d slab pages\n", slab);
119         printk("%d pages shared\n", shared);
120         printk("%d pages swap cached\n", cached);
121 }
122
123 /*
124  * FIXME: We really want to avoid allocating the bootmap bitmap
125  * over the top of the initrd.  Hopefully, this is located towards
126  * the start of a bank, so if we allocate the bootmap bitmap at
127  * the end, we won't clash.
128  */
129 static unsigned int __init
130 find_bootmap_pfn(int node, struct meminfo *mi, unsigned int bootmap_pages)
131 {
132         unsigned int start_pfn, i, bootmap_pfn;
133
134         start_pfn   = PAGE_ALIGN(__pa(_end)) >> PAGE_SHIFT;
135         bootmap_pfn = 0;
136
137         for_each_nodebank(i, mi, node) {
138                 struct membank *bank = &mi->bank[i];
139                 unsigned int start, end;
140
141                 start = bank_pfn_start(bank);
142                 end   = bank_pfn_end(bank);
143
144                 if (end < start_pfn)
145                         continue;
146
147                 if (start < start_pfn)
148                         start = start_pfn;
149
150                 if (end <= start)
151                         continue;
152
153                 if (end - start >= bootmap_pages) {
154                         bootmap_pfn = start;
155                         break;
156                 }
157         }
158
159         if (bootmap_pfn == 0)
160                 BUG();
161
162         return bootmap_pfn;
163 }
164
165 static int __init check_initrd(struct meminfo *mi)
166 {
167         int initrd_node = -2;
168 #ifdef CONFIG_BLK_DEV_INITRD
169         unsigned long end = phys_initrd_start + phys_initrd_size;
170
171         /*
172          * Make sure that the initrd is within a valid area of
173          * memory.
174          */
175         if (phys_initrd_size) {
176                 unsigned int i;
177
178                 initrd_node = -1;
179
180                 for (i = 0; i < mi->nr_banks; i++) {
181                         struct membank *bank = &mi->bank[i];
182                         if (bank_phys_start(bank) <= phys_initrd_start &&
183                             end <= bank_phys_end(bank))
184                                 initrd_node = bank->node;
185                 }
186         }
187
188         if (initrd_node == -1) {
189                 printk(KERN_ERR "INITRD: 0x%08lx+0x%08lx extends beyond "
190                        "physical memory - disabling initrd\n",
191                        phys_initrd_start, phys_initrd_size);
192                 phys_initrd_start = phys_initrd_size = 0;
193         }
194 #endif
195
196         return initrd_node;
197 }
198
199 static inline void map_memory_bank(struct membank *bank)
200 {
201 #ifdef CONFIG_MMU
202         struct map_desc map;
203
204         map.pfn = bank_pfn_start(bank);
205         map.virtual = __phys_to_virt(bank_phys_start(bank));
206         map.length = bank_phys_size(bank);
207         map.type = MT_MEMORY;
208
209         create_mapping(&map);
210 #endif
211 }
212
213 static unsigned long __init bootmem_init_node(int node, struct meminfo *mi)
214 {
215         unsigned long start_pfn, end_pfn, boot_pfn;
216         unsigned int boot_pages;
217         pg_data_t *pgdat;
218         int i;
219
220         start_pfn = -1UL;
221         end_pfn = 0;
222
223         /*
224          * Calculate the pfn range, and map the memory banks for this node.
225          */
226         for_each_nodebank(i, mi, node) {
227                 struct membank *bank = &mi->bank[i];
228                 unsigned long start, end;
229
230                 start = bank_pfn_start(bank);
231                 end = bank_pfn_end(bank);
232
233                 if (start_pfn > start)
234                         start_pfn = start;
235                 if (end_pfn < end)
236                         end_pfn = end;
237
238                 map_memory_bank(bank);
239         }
240
241         /*
242          * If there is no memory in this node, ignore it.
243          */
244         if (end_pfn == 0)
245                 return end_pfn;
246
247         /*
248          * Allocate the bootmem bitmap page.
249          */
250         boot_pages = bootmem_bootmap_pages(end_pfn - start_pfn);
251         boot_pfn = find_bootmap_pfn(node, mi, boot_pages);
252
253         /*
254          * Initialise the bootmem allocator for this node, handing the
255          * memory banks over to bootmem.
256          */
257         node_set_online(node);
258         pgdat = NODE_DATA(node);
259         init_bootmem_node(pgdat, boot_pfn, start_pfn, end_pfn);
260
261         for_each_nodebank(i, mi, node) {
262                 struct membank *bank = &mi->bank[i];
263                 free_bootmem_node(pgdat, bank_phys_start(bank), bank_phys_size(bank));
264                 memory_present(node, bank_pfn_start(bank), bank_pfn_end(bank));
265         }
266
267         /*
268          * Reserve the bootmem bitmap for this node.
269          */
270         reserve_bootmem_node(pgdat, boot_pfn << PAGE_SHIFT,
271                              boot_pages << PAGE_SHIFT, BOOTMEM_DEFAULT);
272
273         return end_pfn;
274 }
275
276 static void __init bootmem_reserve_initrd(int node)
277 {
278 #ifdef CONFIG_BLK_DEV_INITRD
279         pg_data_t *pgdat = NODE_DATA(node);
280         int res;
281
282         res = reserve_bootmem_node(pgdat, phys_initrd_start,
283                              phys_initrd_size, BOOTMEM_EXCLUSIVE);
284
285         if (res == 0) {
286                 initrd_start = __phys_to_virt(phys_initrd_start);
287                 initrd_end = initrd_start + phys_initrd_size;
288         } else {
289                 printk(KERN_ERR
290                         "INITRD: 0x%08lx+0x%08lx overlaps in-use "
291                         "memory region - disabling initrd\n",
292                         phys_initrd_start, phys_initrd_size);
293         }
294 #endif
295 }
296
297 static void __init bootmem_free_node(int node, struct meminfo *mi)
298 {
299         unsigned long zone_size[MAX_NR_ZONES], zhole_size[MAX_NR_ZONES];
300         unsigned long start_pfn, end_pfn;
301         pg_data_t *pgdat = NODE_DATA(node);
302         int i;
303
304         start_pfn = pgdat->bdata->node_min_pfn;
305         end_pfn = pgdat->bdata->node_low_pfn;
306
307         /*
308          * initialise the zones within this node.
309          */
310         memset(zone_size, 0, sizeof(zone_size));
311         memset(zhole_size, 0, sizeof(zhole_size));
312
313         /*
314          * The size of this node has already been determined.  If we need
315          * to do anything fancy with the allocation of this memory to the
316          * zones, now is the time to do it.
317          */
318         zone_size[0] = end_pfn - start_pfn;
319
320         /*
321          * For each bank in this node, calculate the size of the holes.
322          *  holes = node_size - sum(bank_sizes_in_node)
323          */
324         zhole_size[0] = zone_size[0];
325         for_each_nodebank(i, mi, node)
326                 zhole_size[0] -= bank_pfn_size(&mi->bank[i]);
327
328         /*
329          * Adjust the sizes according to any special requirements for
330          * this machine type.
331          */
332         arch_adjust_zones(node, zone_size, zhole_size);
333
334         free_area_init_node(node, zone_size, start_pfn, zhole_size);
335 }
336
337 void __init bootmem_init(void)
338 {
339         struct meminfo *mi = &meminfo;
340         unsigned long memend_pfn = 0;
341         int node, initrd_node;
342
343         /*
344          * Locate which node contains the ramdisk image, if any.
345          */
346         initrd_node = check_initrd(mi);
347
348         /*
349          * Run through each node initialising the bootmem allocator.
350          */
351         for_each_node(node) {
352                 unsigned long end_pfn = bootmem_init_node(node, mi);
353
354                 /*
355                  * Reserve any special node zero regions.
356                  */
357                 if (node == 0)
358                         reserve_node_zero(NODE_DATA(node));
359
360                 /*
361                  * If the initrd is in this node, reserve its memory.
362                  */
363                 if (node == initrd_node)
364                         bootmem_reserve_initrd(node);
365
366                 /*
367                  * Remember the highest memory PFN.
368                  */
369                 if (end_pfn > memend_pfn)
370                         memend_pfn = end_pfn;
371         }
372
373         /*
374          * sparse_init() needs the bootmem allocator up and running.
375          */
376         sparse_init();
377
378         /*
379          * Now free memory in each node - free_area_init_node needs
380          * the sparse mem_map arrays initialized by sparse_init()
381          * for memmap_init_zone(), otherwise all PFNs are invalid.
382          */
383         for_each_node(node)
384                 bootmem_free_node(node, mi);
385
386         high_memory = __va((memend_pfn << PAGE_SHIFT) - 1) + 1;
387
388         /*
389          * This doesn't seem to be used by the Linux memory manager any
390          * more, but is used by ll_rw_block.  If we can get rid of it, we
391          * also get rid of some of the stuff above as well.
392          *
393          * Note: max_low_pfn and max_pfn reflect the number of _pages_ in
394          * the system, not the maximum PFN.
395          */
396         max_pfn = max_low_pfn = memend_pfn - PHYS_PFN_OFFSET;
397 }
398
399 static inline int free_area(unsigned long pfn, unsigned long end, char *s)
400 {
401         unsigned int pages = 0, size = (end - pfn) << (PAGE_SHIFT - 10);
402
403         for (; pfn < end; pfn++) {
404                 struct page *page = pfn_to_page(pfn);
405                 ClearPageReserved(page);
406                 init_page_count(page);
407                 __free_page(page);
408                 pages++;
409         }
410
411         if (size && s)
412                 printk(KERN_INFO "Freeing %s memory: %dK\n", s, size);
413
414         return pages;
415 }
416
417 static inline void
418 free_memmap(int node, unsigned long start_pfn, unsigned long end_pfn)
419 {
420         struct page *start_pg, *end_pg;
421         unsigned long pg, pgend;
422
423         /*
424          * Convert start_pfn/end_pfn to a struct page pointer.
425          */
426         start_pg = pfn_to_page(start_pfn);
427         end_pg = pfn_to_page(end_pfn);
428
429         /*
430          * Convert to physical addresses, and
431          * round start upwards and end downwards.
432          */
433         pg = PAGE_ALIGN(__pa(start_pg));
434         pgend = __pa(end_pg) & PAGE_MASK;
435
436         /*
437          * If there are free pages between these,
438          * free the section of the memmap array.
439          */
440         if (pg < pgend)
441                 free_bootmem_node(NODE_DATA(node), pg, pgend - pg);
442 }
443
444 /*
445  * The mem_map array can get very big.  Free the unused area of the memory map.
446  */
447 static void __init free_unused_memmap_node(int node, struct meminfo *mi)
448 {
449         unsigned long bank_start, prev_bank_end = 0;
450         unsigned int i;
451
452         /*
453          * [FIXME] This relies on each bank being in address order.  This
454          * may not be the case, especially if the user has provided the
455          * information on the command line.
456          */
457         for_each_nodebank(i, mi, node) {
458                 struct membank *bank = &mi->bank[i];
459
460                 bank_start = bank_pfn_start(bank);
461                 if (bank_start < prev_bank_end) {
462                         printk(KERN_ERR "MEM: unordered memory banks.  "
463                                 "Not freeing memmap.\n");
464                         break;
465                 }
466
467                 /*
468                  * If we had a previous bank, and there is a space
469                  * between the current bank and the previous, free it.
470                  */
471                 if (prev_bank_end && prev_bank_end != bank_start)
472                         free_memmap(node, prev_bank_end, bank_start);
473
474                 prev_bank_end = bank_pfn_end(bank);
475         }
476 }
477
478 /*
479  * mem_init() marks the free areas in the mem_map and tells us how much
480  * memory is free.  This is done after various parts of the system have
481  * claimed their memory after the kernel image.
482  */
483 void __init mem_init(void)
484 {
485         unsigned int codesize, datasize, initsize;
486         int i, node;
487
488 #ifndef CONFIG_DISCONTIGMEM
489         max_mapnr   = pfn_to_page(max_pfn + PHYS_PFN_OFFSET) - mem_map;
490 #endif
491
492         /* this will put all unused low memory onto the freelists */
493         for_each_online_node(node) {
494                 pg_data_t *pgdat = NODE_DATA(node);
495
496                 free_unused_memmap_node(node, &meminfo);
497
498                 if (pgdat->node_spanned_pages != 0)
499                         totalram_pages += free_all_bootmem_node(pgdat);
500         }
501
502 #ifdef CONFIG_SA1111
503         /* now that our DMA memory is actually so designated, we can free it */
504         totalram_pages += free_area(PHYS_PFN_OFFSET,
505                                     __phys_to_pfn(__pa(swapper_pg_dir)), NULL);
506 #endif
507
508 #ifdef CONFIG_HIGHMEM
509         /* set highmem page free */
510         for_each_online_node(node) {
511                 for_each_nodebank (i, &meminfo, node) {
512                         unsigned long start = bank_pfn_start(&meminfo.bank[i]);
513                         unsigned long end = bank_pfn_end(&meminfo.bank[i]);
514                         if (start >= max_low_pfn + PHYS_PFN_OFFSET)
515                                 totalhigh_pages += free_area(start, end, NULL);
516                 }
517         }
518         totalram_pages += totalhigh_pages;
519 #endif
520
521         /*
522          * Since our memory may not be contiguous, calculate the
523          * real number of pages we have in this system
524          */
525         printk(KERN_INFO "Memory:");
526         num_physpages = 0;
527         for (i = 0; i < meminfo.nr_banks; i++) {
528                 num_physpages += bank_pfn_size(&meminfo.bank[i]);
529                 printk(" %ldMB", bank_phys_size(&meminfo.bank[i]) >> 20);
530         }
531         printk(" = %luMB total\n", num_physpages >> (20 - PAGE_SHIFT));
532
533         codesize = _etext - _text;
534         datasize = _end - _data;
535         initsize = __init_end - __init_begin;
536
537         printk(KERN_NOTICE "Memory: %luKB available (%dK code, "
538                 "%dK data, %dK init, %luK highmem)\n",
539                 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
540                 codesize >> 10, datasize >> 10, initsize >> 10,
541                 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10)));
542
543         if (PAGE_SIZE >= 16384 && num_physpages <= 128) {
544                 extern int sysctl_overcommit_memory;
545                 /*
546                  * On a machine this small we won't get
547                  * anywhere without overcommit, so turn
548                  * it on by default.
549                  */
550                 sysctl_overcommit_memory = OVERCOMMIT_ALWAYS;
551         }
552 }
553
554 void free_initmem(void)
555 {
556         if (!machine_is_integrator() && !machine_is_cintegrator())
557                 totalram_pages += free_area(__phys_to_pfn(__pa(__init_begin)),
558                                             __phys_to_pfn(__pa(__init_end)),
559                                             "init");
560 }
561
562 #ifdef CONFIG_BLK_DEV_INITRD
563
564 static int keep_initrd;
565
566 void free_initrd_mem(unsigned long start, unsigned long end)
567 {
568         if (!keep_initrd)
569                 totalram_pages += free_area(__phys_to_pfn(__pa(start)),
570                                             __phys_to_pfn(__pa(end)),
571                                             "initrd");
572 }
573
574 static int __init keepinitrd_setup(char *__unused)
575 {
576         keep_initrd = 1;
577         return 1;
578 }
579
580 __setup("keepinitrd", keepinitrd_setup);
581 #endif