2 * linux/arch/arm/mm/init.c
4 * Copyright (C) 1995-2002 Russell King
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.
10 #include <linux/config.h>
11 #include <linux/kernel.h>
12 #include <linux/errno.h>
13 #include <linux/ptrace.h>
14 #include <linux/swap.h>
15 #include <linux/init.h>
16 #include <linux/bootmem.h>
17 #include <linux/mman.h>
18 #include <linux/nodemask.h>
19 #include <linux/initrd.h>
21 #include <asm/mach-types.h>
22 #include <asm/hardware.h>
23 #include <asm/setup.h>
26 #include <asm/mach/arch.h>
27 #include <asm/mach/map.h>
29 #define TABLE_SIZE (2 * PTRS_PER_PTE * sizeof(pte_t))
31 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
33 extern pgd_t swapper_pg_dir[PTRS_PER_PGD];
34 extern void _stext, _text, _etext, __data_start, _end, __init_begin, __init_end;
35 extern unsigned long phys_initrd_start;
36 extern unsigned long phys_initrd_size;
39 * The sole use of this is to pass memory configuration
40 * data from paging_init to mem_init.
42 static struct meminfo meminfo __initdata = { 0, };
45 * empty_zero_page is a special page that is used for
46 * zero-initialized data and COW.
48 struct page *empty_zero_page;
52 int free = 0, total = 0, reserved = 0;
53 int shared = 0, cached = 0, slab = 0, node;
55 printk("Mem-info:\n");
57 printk("Free swap: %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10));
59 for_each_online_node(node) {
60 struct page *page, *end;
62 page = NODE_MEM_MAP(node);
63 end = page + NODE_DATA(node)->node_spanned_pages;
67 if (PageReserved(page))
69 else if (PageSwapCache(page))
71 else if (PageSlab(page))
73 else if (!page_count(page))
76 shared += page_count(page) - 1;
81 printk("%d pages of RAM\n", total);
82 printk("%d free pages\n", free);
83 printk("%d reserved pages\n", reserved);
84 printk("%d slab pages\n", slab);
85 printk("%d pages shared\n", shared);
86 printk("%d pages swap cached\n", cached);
95 #define O_PFN_DOWN(x) ((x) >> PAGE_SHIFT)
96 #define O_PFN_UP(x) (PAGE_ALIGN(x) >> PAGE_SHIFT)
99 * FIXME: We really want to avoid allocating the bootmap bitmap
100 * over the top of the initrd. Hopefully, this is located towards
101 * the start of a bank, so if we allocate the bootmap bitmap at
102 * the end, we won't clash.
104 static unsigned int __init
105 find_bootmap_pfn(int node, struct meminfo *mi, unsigned int bootmap_pages)
107 unsigned int start_pfn, bank, bootmap_pfn;
109 start_pfn = O_PFN_UP(__pa(&_end));
112 for (bank = 0; bank < mi->nr_banks; bank ++) {
113 unsigned int start, end;
115 if (mi->bank[bank].node != node)
118 start = mi->bank[bank].start >> PAGE_SHIFT;
119 end = (mi->bank[bank].size +
120 mi->bank[bank].start) >> PAGE_SHIFT;
125 if (start < start_pfn)
131 if (end - start >= bootmap_pages) {
137 if (bootmap_pfn == 0)
144 * Scan the memory info structure and pull out:
145 * - the end of memory
146 * - the number of nodes
147 * - the pfn range of each node
148 * - the number of bootmem bitmap pages
150 static unsigned int __init
151 find_memend_and_nodes(struct meminfo *mi, struct node_info *np)
153 unsigned int i, bootmem_pages = 0, memend_pfn = 0;
155 for (i = 0; i < MAX_NUMNODES; i++) {
158 np[i].bootmap_pages = 0;
161 for (i = 0; i < mi->nr_banks; i++) {
162 unsigned long start, end;
165 if (mi->bank[i].size == 0) {
167 * Mark this bank with an invalid node number
169 mi->bank[i].node = -1;
173 node = mi->bank[i].node;
176 * Make sure we haven't exceeded the maximum number of nodes
177 * that we have in this configuration. If we have, we're in
178 * trouble. (maybe we ought to limit, instead of bugging?)
180 if (node >= MAX_NUMNODES)
182 node_set_online(node);
185 * Get the start and end pfns for this bank
187 start = mi->bank[i].start >> PAGE_SHIFT;
188 end = (mi->bank[i].start + mi->bank[i].size) >> PAGE_SHIFT;
190 if (np[node].start > start)
191 np[node].start = start;
193 if (np[node].end < end)
196 if (memend_pfn < end)
201 * Calculate the number of pages we require to
202 * store the bootmem bitmaps.
204 for_each_online_node(i) {
208 np[i].bootmap_pages = bootmem_bootmap_pages(np[i].end -
210 bootmem_pages += np[i].bootmap_pages;
213 high_memory = __va(memend_pfn << PAGE_SHIFT);
216 * This doesn't seem to be used by the Linux memory
217 * manager any more. If we can get rid of it, we
218 * also get rid of some of the stuff above as well.
220 * Note: max_low_pfn and max_pfn reflect the number
221 * of _pages_ in the system, not the maximum PFN.
223 max_low_pfn = memend_pfn - O_PFN_DOWN(PHYS_OFFSET);
224 max_pfn = memend_pfn - O_PFN_DOWN(PHYS_OFFSET);
226 return bootmem_pages;
229 static int __init check_initrd(struct meminfo *mi)
231 int initrd_node = -2;
232 #ifdef CONFIG_BLK_DEV_INITRD
233 unsigned long end = phys_initrd_start + phys_initrd_size;
236 * Make sure that the initrd is within a valid area of
239 if (phys_initrd_size) {
244 for (i = 0; i < mi->nr_banks; i++) {
245 unsigned long bank_end;
247 bank_end = mi->bank[i].start + mi->bank[i].size;
249 if (mi->bank[i].start <= phys_initrd_start &&
251 initrd_node = mi->bank[i].node;
255 if (initrd_node == -1) {
256 printk(KERN_ERR "initrd (0x%08lx - 0x%08lx) extends beyond "
257 "physical memory - disabling initrd\n",
258 phys_initrd_start, end);
259 phys_initrd_start = phys_initrd_size = 0;
267 * Reserve the various regions of node 0
269 static __init void reserve_node_zero(unsigned int bootmap_pfn, unsigned int bootmap_pages)
271 pg_data_t *pgdat = NODE_DATA(0);
272 unsigned long res_size = 0;
275 * Register the kernel text and data with bootmem.
276 * Note that this can only be in node 0.
278 #ifdef CONFIG_XIP_KERNEL
279 reserve_bootmem_node(pgdat, __pa(&__data_start), &_end - &__data_start);
281 reserve_bootmem_node(pgdat, __pa(&_stext), &_end - &_stext);
285 * Reserve the page tables. These are already in use,
286 * and can only be in node 0.
288 reserve_bootmem_node(pgdat, __pa(swapper_pg_dir),
289 PTRS_PER_PGD * sizeof(pgd_t));
292 * And don't forget to reserve the allocator bitmap,
293 * which will be freed later.
295 reserve_bootmem_node(pgdat, bootmap_pfn << PAGE_SHIFT,
296 bootmap_pages << PAGE_SHIFT);
299 * Hmm... This should go elsewhere, but we really really need to
300 * stop things allocating the low memory; ideally we need a better
301 * implementation of GFP_DMA which does not assume that DMA-able
302 * memory starts at zero.
304 if (machine_is_integrator() || machine_is_cintegrator())
305 res_size = __pa(swapper_pg_dir) - PHYS_OFFSET;
308 * These should likewise go elsewhere. They pre-reserve the
309 * screen memory region at the start of main system memory.
311 if (machine_is_edb7211())
312 res_size = 0x00020000;
313 if (machine_is_p720t())
314 res_size = 0x00014000;
318 * Because of the SA1111 DMA bug, we want to preserve our
319 * precious DMA-able memory...
321 res_size = __pa(swapper_pg_dir) - PHYS_OFFSET;
324 reserve_bootmem_node(pgdat, PHYS_OFFSET, res_size);
328 * Register all available RAM in this node with the bootmem allocator.
330 static inline void free_bootmem_node_bank(int node, struct meminfo *mi)
332 pg_data_t *pgdat = NODE_DATA(node);
335 for (bank = 0; bank < mi->nr_banks; bank++)
336 if (mi->bank[bank].node == node)
337 free_bootmem_node(pgdat, mi->bank[bank].start,
338 mi->bank[bank].size);
342 * Initialise the bootmem allocator for all nodes. This is called
343 * early during the architecture specific initialisation.
345 static void __init bootmem_init(struct meminfo *mi)
347 struct node_info node_info[MAX_NUMNODES], *np = node_info;
348 unsigned int bootmap_pages, bootmap_pfn, map_pg;
349 int node, initrd_node;
351 bootmap_pages = find_memend_and_nodes(mi, np);
352 bootmap_pfn = find_bootmap_pfn(0, mi, bootmap_pages);
353 initrd_node = check_initrd(mi);
355 map_pg = bootmap_pfn;
358 * Initialise the bootmem nodes.
360 * What we really want to do is:
362 * unmap_all_regions_except_kernel();
363 * for_each_node_in_reverse_order(node) {
365 * allocate_bootmem_map(node);
366 * init_bootmem_node(node);
367 * free_bootmem_node(node);
370 * but this is a 2.5-type change. For now, we just set
371 * the nodes up in reverse order.
373 * (we could also do with rolling bootmem_init and paging_init
374 * into one generic "memory_init" type function).
376 np += num_online_nodes() - 1;
377 for (node = num_online_nodes() - 1; node >= 0; node--, np--) {
379 * If there are no pages in this node, ignore it.
380 * Note that node 0 must always have some pages.
382 if (np->end == 0 || !node_online(node)) {
389 * Initialise the bootmem allocator.
391 init_bootmem_node(NODE_DATA(node), map_pg, np->start, np->end);
392 free_bootmem_node_bank(node, mi);
393 map_pg += np->bootmap_pages;
396 * If this is node 0, we need to reserve some areas ASAP -
397 * we may use bootmem on node 0 to setup the other nodes.
400 reserve_node_zero(bootmap_pfn, bootmap_pages);
404 #ifdef CONFIG_BLK_DEV_INITRD
405 if (phys_initrd_size && initrd_node >= 0) {
406 reserve_bootmem_node(NODE_DATA(initrd_node), phys_initrd_start,
408 initrd_start = __phys_to_virt(phys_initrd_start);
409 initrd_end = initrd_start + phys_initrd_size;
413 BUG_ON(map_pg != bootmap_pfn + bootmap_pages);
417 * paging_init() sets up the page tables, initialises the zone memory
418 * maps, and sets up the zero page, bad page and bad page tables.
420 void __init paging_init(struct meminfo *mi, struct machine_desc *mdesc)
427 memcpy(&meminfo, mi, sizeof(meminfo));
430 * allocate the zero page. Note that we count on this going ok.
432 zero_page = alloc_bootmem_low_pages(PAGE_SIZE);
435 * initialise the page tables.
440 local_flush_tlb_all();
443 * initialise the zones within each node
445 for_each_online_node(node) {
446 unsigned long zone_size[MAX_NR_ZONES];
447 unsigned long zhole_size[MAX_NR_ZONES];
448 struct bootmem_data *bdata;
453 * Initialise the zone size information.
455 for (i = 0; i < MAX_NR_ZONES; i++) {
460 pgdat = NODE_DATA(node);
461 bdata = pgdat->bdata;
464 * The size of this node has already been determined.
465 * If we need to do anything fancy with the allocation
466 * of this memory to the zones, now is the time to do
469 zone_size[0] = bdata->node_low_pfn -
470 (bdata->node_boot_start >> PAGE_SHIFT);
473 * If this zone has zero size, skip it.
479 * For each bank in this node, calculate the size of the
480 * holes. holes = node_size - sum(bank_sizes_in_node)
482 zhole_size[0] = zone_size[0];
483 for (i = 0; i < mi->nr_banks; i++) {
484 if (mi->bank[i].node != node)
487 zhole_size[0] -= mi->bank[i].size >> PAGE_SHIFT;
491 * Adjust the sizes according to any special
492 * requirements for this machine type.
494 arch_adjust_zones(node, zone_size, zhole_size);
496 free_area_init_node(node, pgdat, zone_size,
497 bdata->node_boot_start >> PAGE_SHIFT, zhole_size);
501 * finish off the bad pages once
502 * the mem_map is initialised
504 memzero(zero_page, PAGE_SIZE);
505 empty_zero_page = virt_to_page(zero_page);
506 flush_dcache_page(empty_zero_page);
509 static inline void free_area(unsigned long addr, unsigned long end, char *s)
511 unsigned int size = (end - addr) >> 10;
513 for (; addr < end; addr += PAGE_SIZE) {
514 struct page *page = virt_to_page(addr);
515 ClearPageReserved(page);
516 set_page_count(page, 1);
522 printk(KERN_INFO "Freeing %s memory: %dK\n", s, size);
526 free_memmap(int node, unsigned long start_pfn, unsigned long end_pfn)
528 struct page *start_pg, *end_pg;
529 unsigned long pg, pgend;
532 * Convert start_pfn/end_pfn to a struct page pointer.
534 start_pg = pfn_to_page(start_pfn);
535 end_pg = pfn_to_page(end_pfn);
538 * Convert to physical addresses, and
539 * round start upwards and end downwards.
541 pg = PAGE_ALIGN(__pa(start_pg));
542 pgend = __pa(end_pg) & PAGE_MASK;
545 * If there are free pages between these,
546 * free the section of the memmap array.
549 free_bootmem_node(NODE_DATA(node), pg, pgend - pg);
553 * The mem_map array can get very big. Free the unused area of the memory map.
555 static void __init free_unused_memmap_node(int node, struct meminfo *mi)
557 unsigned long bank_start, prev_bank_end = 0;
561 * [FIXME] This relies on each bank being in address order. This
562 * may not be the case, especially if the user has provided the
563 * information on the command line.
565 for (i = 0; i < mi->nr_banks; i++) {
566 if (mi->bank[i].size == 0 || mi->bank[i].node != node)
569 bank_start = mi->bank[i].start >> PAGE_SHIFT;
570 if (bank_start < prev_bank_end) {
571 printk(KERN_ERR "MEM: unordered memory banks. "
572 "Not freeing memmap.\n");
577 * If we had a previous bank, and there is a space
578 * between the current bank and the previous, free it.
580 if (prev_bank_end && prev_bank_end != bank_start)
581 free_memmap(node, prev_bank_end, bank_start);
583 prev_bank_end = (mi->bank[i].start +
584 mi->bank[i].size) >> PAGE_SHIFT;
589 * mem_init() marks the free areas in the mem_map and tells us how much
590 * memory is free. This is done after various parts of the system have
591 * claimed their memory after the kernel image.
593 void __init mem_init(void)
595 unsigned int codepages, datapages, initpages;
598 codepages = &_etext - &_text;
599 datapages = &_end - &__data_start;
600 initpages = &__init_end - &__init_begin;
602 #ifndef CONFIG_DISCONTIGMEM
603 max_mapnr = virt_to_page(high_memory) - mem_map;
606 /* this will put all unused low memory onto the freelists */
607 for_each_online_node(node) {
608 pg_data_t *pgdat = NODE_DATA(node);
610 free_unused_memmap_node(node, &meminfo);
612 if (pgdat->node_spanned_pages != 0)
613 totalram_pages += free_all_bootmem_node(pgdat);
617 /* now that our DMA memory is actually so designated, we can free it */
618 free_area(PAGE_OFFSET, (unsigned long)swapper_pg_dir, NULL);
622 * Since our memory may not be contiguous, calculate the
623 * real number of pages we have in this system
625 printk(KERN_INFO "Memory:");
628 for (i = 0; i < meminfo.nr_banks; i++) {
629 num_physpages += meminfo.bank[i].size >> PAGE_SHIFT;
630 printk(" %ldMB", meminfo.bank[i].size >> 20);
633 printk(" = %luMB total\n", num_physpages >> (20 - PAGE_SHIFT));
634 printk(KERN_NOTICE "Memory: %luKB available (%dK code, "
635 "%dK data, %dK init)\n",
636 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
637 codepages >> 10, datapages >> 10, initpages >> 10);
639 if (PAGE_SIZE >= 16384 && num_physpages <= 128) {
640 extern int sysctl_overcommit_memory;
642 * On a machine this small we won't get
643 * anywhere without overcommit, so turn
646 sysctl_overcommit_memory = OVERCOMMIT_ALWAYS;
650 void free_initmem(void)
652 if (!machine_is_integrator() && !machine_is_cintegrator()) {
653 free_area((unsigned long)(&__init_begin),
654 (unsigned long)(&__init_end),
659 #ifdef CONFIG_BLK_DEV_INITRD
661 static int keep_initrd;
663 void free_initrd_mem(unsigned long start, unsigned long end)
666 free_area(start, end, "initrd");
669 static int __init keepinitrd_setup(char *__unused)
675 __setup("keepinitrd", keepinitrd_setup);