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