File: /usr/src/linux/arch/arm/mm/init.c
1 /*
2 * linux/arch/arm/mm/init.c
3 *
4 * Copyright (C) 1995-2000 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/config.h>
11 #include <linux/signal.h>
12 #include <linux/sched.h>
13 #include <linux/kernel.h>
14 #include <linux/errno.h>
15 #include <linux/string.h>
16 #include <linux/types.h>
17 #include <linux/ptrace.h>
18 #include <linux/mman.h>
19 #include <linux/mm.h>
20 #include <linux/swap.h>
21 #include <linux/swapctl.h>
22 #include <linux/smp.h>
23 #include <linux/init.h>
24 #include <linux/bootmem.h>
25 #include <linux/blk.h>
26
27 #include <asm/segment.h>
28 #include <asm/mach-types.h>
29 #include <asm/pgalloc.h>
30 #include <asm/dma.h>
31 #include <asm/hardware.h>
32 #include <asm/setup.h>
33
34 #include <asm/mach/arch.h>
35 #include <asm/mach/map.h>
36
37 #ifndef CONFIG_DISCONTIGMEM
38 #define NR_NODES 1
39 #else
40 #define NR_NODES 4
41 #endif
42
43 #ifdef CONFIG_CPU_32
44 #define TABLE_OFFSET (PTRS_PER_PTE)
45 #else
46 #define TABLE_OFFSET 0
47 #endif
48
49 #define TABLE_SIZE ((TABLE_OFFSET + PTRS_PER_PTE) * sizeof(void *))
50
51 static unsigned long totalram_pages;
52 extern pgd_t swapper_pg_dir[PTRS_PER_PGD];
53 extern char _stext, _text, _etext, _end, __init_begin, __init_end;
54
55 /*
56 * The sole use of this is to pass memory configuration
57 * data from paging_init to mem_init.
58 */
59 static struct meminfo meminfo __initdata = { 0, };
60
61 /*
62 * empty_zero_page is a special page that is used for
63 * zero-initialized data and COW.
64 */
65 struct page *empty_zero_page;
66
67 #ifndef CONFIG_NO_PGT_CACHE
68 struct pgtable_cache_struct quicklists;
69
70 int do_check_pgt_cache(int low, int high)
71 {
72 int freed = 0;
73
74 if(pgtable_cache_size > high) {
75 do {
76 if(pgd_quicklist) {
77 free_pgd_slow(get_pgd_fast());
78 freed++;
79 }
80 if(pmd_quicklist) {
81 pmd_free_slow(pmd_alloc_one_fast(NULL, 0));
82 freed++;
83 }
84 if(pte_quicklist) {
85 pte_free_slow(pte_alloc_one_fast(NULL, 0));
86 freed++;
87 }
88 } while(pgtable_cache_size > low);
89 }
90 return freed;
91 }
92 #else
93 int do_check_pgt_cache(int low, int high)
94 {
95 return 0;
96 }
97 #endif
98
99 void show_mem(void)
100 {
101 int free = 0, total = 0, reserved = 0;
102 int shared = 0, cached = 0, node;
103
104 printk("Mem-info:\n");
105 show_free_areas();
106 printk("Free swap: %6dkB\n",nr_swap_pages<<(PAGE_SHIFT-10));
107
108 for (node = 0; node < numnodes; node++) {
109 struct page *page, *end;
110
111 page = NODE_MEM_MAP(node);
112 end = page + NODE_DATA(node)->node_size;
113
114 do {
115 /* This is currently broken
116 * PG_skip is used on sparc/sparc64 architectures to "skip" certain
117 * parts of the address space.
118 *
119 * #define PG_skip 10
120 * #define PageSkip(page) (machine_is_riscpc() && test_bit(PG_skip, &(page)->flags))
121 * if (PageSkip(page)) {
122 * page = page->next_hash;
123 * if (page == NULL)
124 * break;
125 * }
126 */
127 total++;
128 if (PageReserved(page))
129 reserved++;
130 else if (PageSwapCache(page))
131 cached++;
132 else if (!page_count(page))
133 free++;
134 else
135 shared += atomic_read(&page->count) - 1;
136 page++;
137 } while (page < end);
138 }
139
140 printk("%d pages of RAM\n", total);
141 printk("%d free pages\n", free);
142 printk("%d reserved pages\n", reserved);
143 printk("%d pages shared\n", shared);
144 printk("%d pages swap cached\n", cached);
145 #ifndef CONFIG_NO_PGT_CACHE
146 printk("%ld page tables cached\n", pgtable_cache_size);
147 #endif
148 show_buffers();
149 }
150
151 struct node_info {
152 unsigned int start;
153 unsigned int end;
154 int bootmap_pages;
155 };
156
157 #define O_PFN_DOWN(x) ((x) >> PAGE_SHIFT)
158 #define V_PFN_DOWN(x) O_PFN_DOWN(__pa(x))
159
160 #define O_PFN_UP(x) (PAGE_ALIGN(x) >> PAGE_SHIFT)
161 #define V_PFN_UP(x) O_PFN_UP(__pa(x))
162
163 #define PFN_SIZE(x) ((x) >> PAGE_SHIFT)
164 #define PFN_RANGE(s,e) PFN_SIZE(PAGE_ALIGN((unsigned long)(e)) - \
165 (((unsigned long)(s)) & PAGE_MASK))
166
167 /*
168 * FIXME: We really want to avoid allocating the bootmap bitmap
169 * over the top of the initrd. Hopefully, this is located towards
170 * the start of a bank, so if we allocate the bootmap bitmap at
171 * the end, we won't clash.
172 */
173 static unsigned int __init
174 find_bootmap_pfn(int node, struct meminfo *mi, unsigned int bootmap_pages)
175 {
176 unsigned int start_pfn, bank, bootmap_pfn;
177
178 start_pfn = V_PFN_UP(&_end);
179 bootmap_pfn = 0;
180
181 for (bank = 0; bank < mi->nr_banks; bank ++) {
182 unsigned int start, end;
183
184 if (mi->bank[bank].node != node)
185 continue;
186
187 start = O_PFN_UP(mi->bank[bank].start);
188 end = O_PFN_DOWN(mi->bank[bank].size +
189 mi->bank[bank].start);
190
191 if (end < start_pfn)
192 continue;
193
194 if (start < start_pfn)
195 start = start_pfn;
196
197 if (end <= start)
198 continue;
199
200 if (end - start >= bootmap_pages) {
201 bootmap_pfn = start;
202 break;
203 }
204 }
205
206 if (bootmap_pfn == 0)
207 BUG();
208
209 return bootmap_pfn;
210 }
211
212 /*
213 * Scan the memory info structure and pull out:
214 * - the end of memory
215 * - the number of nodes
216 * - the pfn range of each node
217 * - the number of bootmem bitmap pages
218 */
219 static unsigned int __init
220 find_memend_and_nodes(struct meminfo *mi, struct node_info *np)
221 {
222 unsigned int i, bootmem_pages = 0, memend_pfn = 0;
223
224 for (i = 0; i < NR_NODES; i++) {
225 np[i].start = -1U;
226 np[i].end = 0;
227 np[i].bootmap_pages = 0;
228 }
229
230 for (i = 0; i < mi->nr_banks; i++) {
231 unsigned long start, end;
232 int node;
233
234 if (mi->bank[i].size == 0) {
235 /*
236 * Mark this bank with an invalid node number
237 */
238 mi->bank[i].node = -1;
239 continue;
240 }
241
242 node = mi->bank[i].node;
243
244 if (node >= numnodes) {
245 numnodes = node + 1;
246
247 /*
248 * Make sure we haven't exceeded the maximum number
249 * of nodes that we have in this configuration. If
250 * we have, we're in trouble. (maybe we ought to
251 * limit, instead of bugging?)
252 */
253 if (numnodes > NR_NODES)
254 BUG();
255 }
256
257 /*
258 * Get the start and end pfns for this bank
259 */
260 start = O_PFN_UP(mi->bank[i].start);
261 end = O_PFN_DOWN(mi->bank[i].start + mi->bank[i].size);
262
263 if (np[node].start > start)
264 np[node].start = start;
265
266 if (np[node].end < end)
267 np[node].end = end;
268
269 if (memend_pfn < end)
270 memend_pfn = end;
271 }
272
273 /*
274 * Calculate the number of pages we require to
275 * store the bootmem bitmaps.
276 */
277 for (i = 0; i < numnodes; i++) {
278 if (np[i].end == 0)
279 continue;
280
281 np[i].bootmap_pages = bootmem_bootmap_pages(np[i].end -
282 np[i].start);
283 bootmem_pages += np[i].bootmap_pages;
284 }
285
286 /*
287 * This doesn't seem to be used by the Linux memory
288 * manager any more. If we can get rid of it, we
289 * also get rid of some of the stuff above as well.
290 */
291 max_low_pfn = memend_pfn - O_PFN_DOWN(PHYS_OFFSET);
292 mi->end = memend_pfn << PAGE_SHIFT;
293
294 return bootmem_pages;
295 }
296
297 static int __init check_initrd(struct meminfo *mi)
298 {
299 int initrd_node = -2;
300
301 #ifdef CONFIG_BLK_DEV_INITRD
302 /*
303 * Make sure that the initrd is within a valid area of
304 * memory.
305 */
306 if (initrd_start) {
307 unsigned long phys_initrd_start, phys_initrd_end;
308 unsigned int i;
309
310 phys_initrd_start = __pa(initrd_start);
311 phys_initrd_end = __pa(initrd_end);
312
313 for (i = 0; i < mi->nr_banks; i++) {
314 unsigned long bank_end;
315
316 bank_end = mi->bank[i].start + mi->bank[i].size;
317
318 if (mi->bank[i].start <= phys_initrd_start &&
319 phys_initrd_end <= bank_end)
320 initrd_node = mi->bank[i].node;
321 }
322 }
323
324 if (initrd_node == -1) {
325 printk(KERN_ERR "initrd (0x%08lx - 0x%08lx) extends beyond "
326 "physical memory - disabling initrd\n",
327 initrd_start, initrd_end);
328 initrd_start = initrd_end = 0;
329 }
330 #endif
331
332 return initrd_node;
333 }
334
335 /*
336 * Reserve the various regions of node 0
337 */
338 static __init void reserve_node_zero(unsigned int bootmap_pfn, unsigned int bootmap_pages)
339 {
340 pg_data_t *pgdat = NODE_DATA(0);
341
342 /*
343 * Register the kernel text and data with bootmem.
344 * Note that this can only be in node 0.
345 */
346 reserve_bootmem_node(pgdat, __pa(&_stext), &_end - &_stext);
347
348 #ifdef CONFIG_CPU_32
349 /*
350 * Reserve the page tables. These are already in use,
351 * and can only be in node 0.
352 */
353 reserve_bootmem_node(pgdat, __pa(swapper_pg_dir),
354 PTRS_PER_PGD * sizeof(void *));
355 #endif
356 /*
357 * And don't forget to reserve the allocator bitmap,
358 * which will be freed later.
359 */
360 reserve_bootmem_node(pgdat, bootmap_pfn << PAGE_SHIFT,
361 bootmap_pages << PAGE_SHIFT);
362
363 /*
364 * Hmm... This should go elsewhere, but we really really
365 * need to stop things allocating the low memory; we need
366 * a better implementation of GFP_DMA which does not assume
367 * that DMA-able memory starts at zero.
368 */
369 if (machine_is_integrator())
370 reserve_bootmem_node(pgdat, 0, __pa(swapper_pg_dir));
371 /*
372 * These should likewise go elsewhere. They pre-reserve
373 * the screen memory region at the start of main system
374 * memory.
375 */
376 if (machine_is_archimedes() || machine_is_a5k())
377 reserve_bootmem_node(pgdat, 0x02000000, 0x00080000);
378 if (machine_is_p720t())
379 reserve_bootmem_node(pgdat, PHYS_OFFSET, 0x00014000);
380 #ifdef CONFIG_SA1111
381 /*
382 * Because of the SA1111 DMA bug, we want to preserve
383 * our precious DMA-able memory...
384 */
385 reserve_bootmem_node(pgdat, PHYS_OFFSET, __pa(swapper_pg_dir)-PHYS_OFFSET);
386 #endif
387 }
388
389 /*
390 * Register all available RAM in this node with the bootmem allocator.
391 */
392 static inline void free_bootmem_node_bank(int node, struct meminfo *mi)
393 {
394 pg_data_t *pgdat = NODE_DATA(node);
395 int bank;
396
397 for (bank = 0; bank < mi->nr_banks; bank++)
398 if (mi->bank[bank].node == node)
399 free_bootmem_node(pgdat, mi->bank[bank].start,
400 mi->bank[bank].size);
401 }
402
403 /*
404 * Initialise the bootmem allocator for all nodes. This is called
405 * early during the architecture specific initialisation.
406 */
407 void __init bootmem_init(struct meminfo *mi)
408 {
409 struct node_info node_info[NR_NODES], *np = node_info;
410 unsigned int bootmap_pages, bootmap_pfn, map_pg;
411 int node, initrd_node;
412
413 bootmap_pages = find_memend_and_nodes(mi, np);
414 bootmap_pfn = find_bootmap_pfn(0, mi, bootmap_pages);
415 initrd_node = check_initrd(mi);
416
417 map_pg = bootmap_pfn;
418
419 /*
420 * Initialise the bootmem nodes.
421 *
422 * What we really want to do is:
423 *
424 * unmap_all_regions_except_kernel();
425 * for_each_node_in_reverse_order(node) {
426 * map_node(node);
427 * allocate_bootmem_map(node);
428 * init_bootmem_node(node);
429 * free_bootmem_node(node);
430 * }
431 *
432 * but this is a 2.5-type change. For now, we just set
433 * the nodes up in reverse order.
434 *
435 * (we could also do with rolling bootmem_init and paging_init
436 * into one generic "memory_init" type function).
437 */
438 np += numnodes - 1;
439 for (node = numnodes - 1; node >= 0; node--, np--) {
440 /*
441 * If there are no pages in this node, ignore it.
442 * Note that node 0 must always have some pages.
443 */
444 if (np->end == 0) {
445 if (node == 0)
446 BUG();
447 continue;
448 }
449
450 /*
451 * Initialise the bootmem allocator.
452 */
453 init_bootmem_node(NODE_DATA(node), map_pg, np->start, np->end);
454 free_bootmem_node_bank(node, mi);
455 map_pg += np->bootmap_pages;
456
457 /*
458 * If this is node 0, we need to reserve some areas ASAP -
459 * we may use bootmem on node 0 to setup the other nodes.
460 */
461 if (node == 0)
462 reserve_node_zero(bootmap_pfn, bootmap_pages);
463 }
464
465
466 #ifdef CONFIG_BLK_DEV_INITRD
467 if (initrd_node >= 0)
468 reserve_bootmem_node(NODE_DATA(initrd_node), __pa(initrd_start),
469 initrd_end - initrd_start);
470 #endif
471
472 if (map_pg != bootmap_pfn + bootmap_pages)
473 BUG();
474 }
475
476 /*
477 * paging_init() sets up the page tables, initialises the zone memory
478 * maps, and sets up the zero page, bad page and bad page tables.
479 */
480 void __init paging_init(struct meminfo *mi, struct machine_desc *mdesc)
481 {
482 void *zero_page;
483 int node;
484
485 memcpy(&meminfo, mi, sizeof(meminfo));
486
487 /*
488 * allocate the zero page. Note that we count on this going ok.
489 */
490 zero_page = alloc_bootmem_low_pages(PAGE_SIZE);
491
492 /*
493 * initialise the page tables.
494 */
495 memtable_init(mi);
496 if (mdesc->map_io)
497 mdesc->map_io();
498 flush_tlb_all();
499
500 /*
501 * initialise the zones within each node
502 */
503 for (node = 0; node < numnodes; node++) {
504 unsigned long zone_size[MAX_NR_ZONES];
505 unsigned long zhole_size[MAX_NR_ZONES];
506 struct bootmem_data *bdata;
507 pg_data_t *pgdat;
508 int i;
509
510 /*
511 * Initialise the zone size information.
512 */
513 for (i = 0; i < MAX_NR_ZONES; i++) {
514 zone_size[i] = 0;
515 zhole_size[i] = 0;
516 }
517
518 pgdat = NODE_DATA(node);
519 bdata = pgdat->bdata;
520
521 /*
522 * The size of this node has already been determined.
523 * If we need to do anything fancy with the allocation
524 * of this memory to the zones, now is the time to do
525 * it.
526 */
527 zone_size[0] = bdata->node_low_pfn -
528 (bdata->node_boot_start >> PAGE_SHIFT);
529
530 /*
531 * For each bank in this node, calculate the size of the
532 * holes. holes = node_size - sum(bank_sizes_in_node)
533 */
534 zhole_size[0] = zone_size[0];
535 for (i = 0; i < mi->nr_banks; i++) {
536 if (mi->bank[i].node != node)
537 continue;
538
539 zhole_size[0] -= mi->bank[i].size >> PAGE_SHIFT;
540 }
541
542 /*
543 * Adjust the sizes according to any special
544 * requirements for this machine type.
545 */
546 arch_adjust_zones(node, zone_size, zhole_size);
547
548 free_area_init_node(node, pgdat, 0, zone_size,
549 bdata->node_boot_start, zhole_size);
550 }
551
552 /*
553 * finish off the bad pages once
554 * the mem_map is initialised
555 */
556 memzero(zero_page, PAGE_SIZE);
557 empty_zero_page = virt_to_page(zero_page);
558 flush_dcache_page(empty_zero_page);
559 }
560
561 static inline void free_area(unsigned long addr, unsigned long end, char *s)
562 {
563 unsigned int size = (end - addr) >> 10;
564
565 for (; addr < end; addr += PAGE_SIZE) {
566 struct page *page = virt_to_page(addr);
567 ClearPageReserved(page);
568 set_page_count(page, 1);
569 free_page(addr);
570 totalram_pages++;
571 }
572
573 if (size && s)
574 printk("Freeing %s memory: %dK\n", s, size);
575 }
576
577 /*
578 * mem_init() marks the free areas in the mem_map and tells us how much
579 * memory is free. This is done after various parts of the system have
580 * claimed their memory after the kernel image.
581 */
582 void __init mem_init(void)
583 {
584 unsigned int codepages, datapages, initpages;
585 int i, node;
586
587 codepages = &_etext - &_text;
588 datapages = &_end - &_etext;
589 initpages = &__init_end - &__init_begin;
590
591 high_memory = (void *)__va(meminfo.end);
592 max_mapnr = virt_to_page(high_memory) - mem_map;
593
594 /*
595 * We may have non-contiguous memory.
596 */
597 if (meminfo.nr_banks != 1)
598 create_memmap_holes(&meminfo);
599
600 /* this will put all unused low memory onto the freelists */
601 for (node = 0; node < numnodes; node++)
602 totalram_pages += free_all_bootmem_node(NODE_DATA(node));
603
604 #ifdef CONFIG_SA1111
605 /* now that our DMA memory is actually so designated, we can free it */
606 free_area(PAGE_OFFSET, (unsigned long)swapper_pg_dir, NULL);
607 #endif
608
609 /*
610 * Since our memory may not be contiguous, calculate the
611 * real number of pages we have in this system
612 */
613 printk(KERN_INFO "Memory:");
614
615 num_physpages = 0;
616 for (i = 0; i < meminfo.nr_banks; i++) {
617 num_physpages += meminfo.bank[i].size >> PAGE_SHIFT;
618 printk(" %ldMB", meminfo.bank[i].size >> 20);
619 }
620
621 printk(" = %luMB total\n", num_physpages >> (20 - PAGE_SHIFT));
622 printk(KERN_NOTICE "Memory: %luKB available (%dK code, "
623 "%dK data, %dK init)\n",
624 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
625 codepages >> 10, datapages >> 10, initpages >> 10);
626
627 if (PAGE_SIZE >= 16384 && num_physpages <= 128) {
628 extern int sysctl_overcommit_memory;
629 /*
630 * On a machine this small we won't get
631 * anywhere without overcommit, so turn
632 * it on by default.
633 */
634 sysctl_overcommit_memory = 1;
635 }
636 }
637
638 void free_initmem(void)
639 {
640 if (!machine_is_integrator()) {
641 free_area((unsigned long)(&__init_begin),
642 (unsigned long)(&__init_end),
643 "init");
644 }
645 }
646
647 #ifdef CONFIG_BLK_DEV_INITRD
648
649 static int keep_initrd;
650
651 void free_initrd_mem(unsigned long start, unsigned long end)
652 {
653 if (!keep_initrd)
654 free_area(start, end, "initrd");
655 }
656
657 static int __init keepinitrd_setup(char *__unused)
658 {
659 keep_initrd = 1;
660 return 1;
661 }
662
663 __setup("keepinitrd", keepinitrd_setup);
664 #endif
665
666 void si_meminfo(struct sysinfo *val)
667 {
668 val->totalram = totalram_pages;
669 val->sharedram = 0;
670 val->freeram = nr_free_pages();
671 val->bufferram = atomic_read(&buffermem_pages);
672 val->totalhigh = 0;
673 val->freehigh = 0;
674 val->mem_unit = PAGE_SIZE;
675 }
676