2 * linux/arch/arm/mm/init.c
4 * Copyright (C) 1995-2005 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/kernel.h>
11 #include <linux/errno.h>
12 #include <linux/ptrace.h>
13 #include <linux/swap.h>
14 #include <linux/init.h>
15 #include <linux/bootmem.h>
16 #include <linux/mman.h>
17 #include <linux/nodemask.h>
18 #include <linux/initrd.h>
20 #include <asm/mach-types.h>
21 #include <asm/setup.h>
22 #include <asm/sizes.h>
25 #include <asm/mach/arch.h>
26 #include <asm/mach/map.h>
28 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
30 extern pgd_t swapper_pg_dir[PTRS_PER_PGD];
31 extern void _stext, _text, _etext, __data_start, _end, __init_begin, __init_end;
32 extern unsigned long phys_initrd_start;
33 extern unsigned long phys_initrd_size;
36 * The sole use of this is to pass memory configuration
37 * data from paging_init to mem_init.
39 static struct meminfo meminfo __initdata = { 0, };
42 * empty_zero_page is a special page that is used for
43 * zero-initialized data and COW.
45 struct page *empty_zero_page;
49 int free = 0, total = 0, reserved = 0;
50 int shared = 0, cached = 0, slab = 0, node;
52 printk("Mem-info:\n");
54 printk("Free swap: %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10));
56 for_each_online_node(node) {
57 struct page *page, *end;
59 page = NODE_MEM_MAP(node);
60 end = page + NODE_DATA(node)->node_spanned_pages;
64 if (PageReserved(page))
66 else if (PageSwapCache(page))
68 else if (PageSlab(page))
70 else if (!page_count(page))
73 shared += page_count(page) - 1;
78 printk("%d pages of RAM\n", total);
79 printk("%d free pages\n", free);
80 printk("%d reserved pages\n", reserved);
81 printk("%d slab pages\n", slab);
82 printk("%d pages shared\n", shared);
83 printk("%d pages swap cached\n", cached);
86 static inline pmd_t *pmd_off(pgd_t *pgd, unsigned long virt)
88 return pmd_offset(pgd, virt);
91 static inline pmd_t *pmd_off_k(unsigned long virt)
93 return pmd_off(pgd_offset_k(virt), virt);
96 #define for_each_nodebank(iter,mi,no) \
97 for (iter = 0; iter < mi->nr_banks; iter++) \
98 if (mi->bank[iter].node == no)
101 * FIXME: We really want to avoid allocating the bootmap bitmap
102 * over the top of the initrd. Hopefully, this is located towards
103 * the start of a bank, so if we allocate the bootmap bitmap at
104 * the end, we won't clash.
106 static unsigned int __init
107 find_bootmap_pfn(int node, struct meminfo *mi, unsigned int bootmap_pages)
109 unsigned int start_pfn, bank, bootmap_pfn;
111 start_pfn = PAGE_ALIGN(__pa(&_end)) >> PAGE_SHIFT;
114 for_each_nodebank(bank, mi, node) {
115 unsigned int start, end;
117 start = mi->bank[bank].start >> PAGE_SHIFT;
118 end = (mi->bank[bank].size +
119 mi->bank[bank].start) >> PAGE_SHIFT;
124 if (start < start_pfn)
130 if (end - start >= bootmap_pages) {
136 if (bootmap_pfn == 0)
142 static int __init check_initrd(struct meminfo *mi)
144 int initrd_node = -2;
145 #ifdef CONFIG_BLK_DEV_INITRD
146 unsigned long end = phys_initrd_start + phys_initrd_size;
149 * Make sure that the initrd is within a valid area of
152 if (phys_initrd_size) {
157 for (i = 0; i < mi->nr_banks; i++) {
158 unsigned long bank_end;
160 bank_end = mi->bank[i].start + mi->bank[i].size;
162 if (mi->bank[i].start <= phys_initrd_start &&
164 initrd_node = mi->bank[i].node;
168 if (initrd_node == -1) {
169 printk(KERN_ERR "initrd (0x%08lx - 0x%08lx) extends beyond "
170 "physical memory - disabling initrd\n",
171 phys_initrd_start, end);
172 phys_initrd_start = phys_initrd_size = 0;
180 * Reserve the various regions of node 0
182 static __init void reserve_node_zero(pg_data_t *pgdat)
184 unsigned long res_size = 0;
187 * Register the kernel text and data with bootmem.
188 * Note that this can only be in node 0.
190 #ifdef CONFIG_XIP_KERNEL
191 reserve_bootmem_node(pgdat, __pa(&__data_start), &_end - &__data_start);
193 reserve_bootmem_node(pgdat, __pa(&_stext), &_end - &_stext);
197 * Reserve the page tables. These are already in use,
198 * and can only be in node 0.
200 reserve_bootmem_node(pgdat, __pa(swapper_pg_dir),
201 PTRS_PER_PGD * sizeof(pgd_t));
204 * Hmm... This should go elsewhere, but we really really need to
205 * stop things allocating the low memory; ideally we need a better
206 * implementation of GFP_DMA which does not assume that DMA-able
207 * memory starts at zero.
209 if (machine_is_integrator() || machine_is_cintegrator())
210 res_size = __pa(swapper_pg_dir) - PHYS_OFFSET;
213 * These should likewise go elsewhere. They pre-reserve the
214 * screen memory region at the start of main system memory.
216 if (machine_is_edb7211())
217 res_size = 0x00020000;
218 if (machine_is_p720t())
219 res_size = 0x00014000;
223 * Because of the SA1111 DMA bug, we want to preserve our
224 * precious DMA-able memory...
226 res_size = __pa(swapper_pg_dir) - PHYS_OFFSET;
229 reserve_bootmem_node(pgdat, PHYS_OFFSET, res_size);
232 void __init build_mem_type_table(void);
233 void __init create_mapping(struct map_desc *md);
235 static unsigned long __init
236 bootmem_init_node(int node, int initrd_node, struct meminfo *mi)
238 unsigned long zone_size[MAX_NR_ZONES], zhole_size[MAX_NR_ZONES];
239 unsigned long start_pfn, end_pfn, boot_pfn;
240 unsigned int boot_pages;
248 * Calculate the pfn range, and map the memory banks for this node.
250 for_each_nodebank(i, mi, node) {
251 unsigned long start, end;
254 start = mi->bank[i].start >> PAGE_SHIFT;
255 end = (mi->bank[i].start + mi->bank[i].size) >> PAGE_SHIFT;
257 if (start_pfn > start)
262 map.pfn = __phys_to_pfn(mi->bank[i].start);
263 map.virtual = __phys_to_virt(mi->bank[i].start);
264 map.length = mi->bank[i].size;
265 map.type = MT_MEMORY;
267 create_mapping(&map);
271 * If there is no memory in this node, ignore it.
277 * Allocate the bootmem bitmap page.
279 boot_pages = bootmem_bootmap_pages(end_pfn - start_pfn);
280 boot_pfn = find_bootmap_pfn(node, mi, boot_pages);
283 * Initialise the bootmem allocator for this node, handing the
284 * memory banks over to bootmem.
286 node_set_online(node);
287 pgdat = NODE_DATA(node);
288 init_bootmem_node(pgdat, boot_pfn, start_pfn, end_pfn);
290 for_each_nodebank(i, mi, node)
291 free_bootmem_node(pgdat, mi->bank[i].start, mi->bank[i].size);
294 * Reserve the bootmem bitmap for this node.
296 reserve_bootmem_node(pgdat, boot_pfn << PAGE_SHIFT,
297 boot_pages << PAGE_SHIFT);
299 #ifdef CONFIG_BLK_DEV_INITRD
301 * If the initrd is in this node, reserve its memory.
303 if (node == initrd_node) {
304 reserve_bootmem_node(pgdat, phys_initrd_start,
306 initrd_start = __phys_to_virt(phys_initrd_start);
307 initrd_end = initrd_start + phys_initrd_size;
312 * Finally, reserve any node zero regions.
315 reserve_node_zero(pgdat);
318 * initialise the zones within this node.
320 memset(zone_size, 0, sizeof(zone_size));
321 memset(zhole_size, 0, sizeof(zhole_size));
324 * The size of this node has already been determined. If we need
325 * to do anything fancy with the allocation of this memory to the
326 * zones, now is the time to do it.
328 zone_size[0] = end_pfn - start_pfn;
331 * For each bank in this node, calculate the size of the holes.
332 * holes = node_size - sum(bank_sizes_in_node)
334 zhole_size[0] = zone_size[0];
335 for_each_nodebank(i, mi, node)
336 zhole_size[0] -= mi->bank[i].size >> PAGE_SHIFT;
339 * Adjust the sizes according to any special requirements for
342 arch_adjust_zones(node, zone_size, zhole_size);
344 free_area_init_node(node, pgdat, zone_size, start_pfn, zhole_size);
349 static void __init bootmem_init(struct meminfo *mi)
351 unsigned long addr, memend_pfn = 0;
352 int node, initrd_node, i;
355 * Invalidate the node number for empty or invalid memory banks
357 for (i = 0; i < mi->nr_banks; i++)
358 if (mi->bank[i].size == 0 || mi->bank[i].node >= MAX_NUMNODES)
359 mi->bank[i].node = -1;
361 memcpy(&meminfo, mi, sizeof(meminfo));
364 * Clear out all the mappings below the kernel image.
366 for (addr = 0; addr < MODULE_START; addr += PGDIR_SIZE)
367 pmd_clear(pmd_off_k(addr));
368 #ifdef CONFIG_XIP_KERNEL
369 /* The XIP kernel is mapped in the module area -- skip over it */
370 addr = ((unsigned long)&_etext + PGDIR_SIZE - 1) & PGDIR_MASK;
372 for ( ; addr < PAGE_OFFSET; addr += PGDIR_SIZE)
373 pmd_clear(pmd_off_k(addr));
376 * Clear out all the kernel space mappings, except for the first
377 * memory bank, up to the end of the vmalloc region.
379 for (addr = __phys_to_virt(mi->bank[0].start + mi->bank[0].size);
380 addr < VMALLOC_END; addr += PGDIR_SIZE)
381 pmd_clear(pmd_off_k(addr));
384 * Locate which node contains the ramdisk image, if any.
386 initrd_node = check_initrd(mi);
389 * Run through each node initialising the bootmem allocator.
391 for_each_node(node) {
392 unsigned long end_pfn;
394 end_pfn = bootmem_init_node(node, initrd_node, mi);
397 * Remember the highest memory PFN.
399 if (end_pfn > memend_pfn)
400 memend_pfn = end_pfn;
403 high_memory = __va(memend_pfn << PAGE_SHIFT);
406 * This doesn't seem to be used by the Linux memory manager any
407 * more, but is used by ll_rw_block. If we can get rid of it, we
408 * also get rid of some of the stuff above as well.
410 * Note: max_low_pfn and max_pfn reflect the number of _pages_ in
411 * the system, not the maximum PFN.
413 max_pfn = max_low_pfn = memend_pfn - PHYS_PFN_OFFSET;
417 * Set up device the mappings. Since we clear out the page tables for all
418 * mappings above VMALLOC_END, we will remove any debug device mappings.
419 * This means you have to be careful how you debug this function, or any
420 * called function. This means you can't use any function or debugging
421 * method which may touch any device, otherwise the kernel _will_ crash.
423 static void __init devicemaps_init(struct machine_desc *mdesc)
430 * Allocate the vector page early.
432 vectors = alloc_bootmem_low_pages(PAGE_SIZE);
435 for (addr = VMALLOC_END; addr; addr += PGDIR_SIZE)
436 pmd_clear(pmd_off_k(addr));
439 * Map the kernel if it is XIP.
440 * It is always first in the modulearea.
442 #ifdef CONFIG_XIP_KERNEL
443 map.pfn = __phys_to_pfn(CONFIG_XIP_PHYS_ADDR & PGDIR_MASK);
444 map.virtual = MODULE_START;
445 map.length = ((unsigned long)&_etext - map.virtual + ~PGDIR_MASK) & PGDIR_MASK;
447 create_mapping(&map);
451 * Map the cache flushing regions.
454 map.pfn = __phys_to_pfn(FLUSH_BASE_PHYS);
455 map.virtual = FLUSH_BASE;
457 map.type = MT_CACHECLEAN;
458 create_mapping(&map);
460 #ifdef FLUSH_BASE_MINICACHE
461 map.pfn = __phys_to_pfn(FLUSH_BASE_PHYS + SZ_1M);
462 map.virtual = FLUSH_BASE_MINICACHE;
464 map.type = MT_MINICLEAN;
465 create_mapping(&map);
469 * Create a mapping for the machine vectors at the high-vectors
470 * location (0xffff0000). If we aren't using high-vectors, also
471 * create a mapping at the low-vectors virtual address.
473 map.pfn = __phys_to_pfn(virt_to_phys(vectors));
474 map.virtual = 0xffff0000;
475 map.length = PAGE_SIZE;
476 map.type = MT_HIGH_VECTORS;
477 create_mapping(&map);
479 if (!vectors_high()) {
481 map.type = MT_LOW_VECTORS;
482 create_mapping(&map);
486 * Ask the machine support to map in the statically mapped devices.
492 * Finally flush the caches and tlb to ensure that we're in a
493 * consistent state wrt the writebuffer. This also ensures that
494 * any write-allocated cache lines in the vector page are written
495 * back. After this point, we can start to touch devices again.
497 local_flush_tlb_all();
502 * paging_init() sets up the page tables, initialises the zone memory
503 * maps, and sets up the zero page, bad page and bad page tables.
505 void __init paging_init(struct meminfo *mi, struct machine_desc *mdesc)
509 build_mem_type_table();
511 devicemaps_init(mdesc);
513 top_pmd = pmd_off_k(0xffff0000);
516 * allocate the zero page. Note that we count on this going ok.
518 zero_page = alloc_bootmem_low_pages(PAGE_SIZE);
519 memzero(zero_page, PAGE_SIZE);
520 empty_zero_page = virt_to_page(zero_page);
521 flush_dcache_page(empty_zero_page);
524 static inline void free_area(unsigned long addr, unsigned long end, char *s)
526 unsigned int size = (end - addr) >> 10;
528 for (; addr < end; addr += PAGE_SIZE) {
529 struct page *page = virt_to_page(addr);
530 ClearPageReserved(page);
531 init_page_count(page);
537 printk(KERN_INFO "Freeing %s memory: %dK\n", s, size);
541 free_memmap(int node, unsigned long start_pfn, unsigned long end_pfn)
543 struct page *start_pg, *end_pg;
544 unsigned long pg, pgend;
547 * Convert start_pfn/end_pfn to a struct page pointer.
549 start_pg = pfn_to_page(start_pfn);
550 end_pg = pfn_to_page(end_pfn);
553 * Convert to physical addresses, and
554 * round start upwards and end downwards.
556 pg = PAGE_ALIGN(__pa(start_pg));
557 pgend = __pa(end_pg) & PAGE_MASK;
560 * If there are free pages between these,
561 * free the section of the memmap array.
564 free_bootmem_node(NODE_DATA(node), pg, pgend - pg);
568 * The mem_map array can get very big. Free the unused area of the memory map.
570 static void __init free_unused_memmap_node(int node, struct meminfo *mi)
572 unsigned long bank_start, prev_bank_end = 0;
576 * [FIXME] This relies on each bank being in address order. This
577 * may not be the case, especially if the user has provided the
578 * information on the command line.
580 for_each_nodebank(i, mi, node) {
581 bank_start = mi->bank[i].start >> PAGE_SHIFT;
582 if (bank_start < prev_bank_end) {
583 printk(KERN_ERR "MEM: unordered memory banks. "
584 "Not freeing memmap.\n");
589 * If we had a previous bank, and there is a space
590 * between the current bank and the previous, free it.
592 if (prev_bank_end && prev_bank_end != bank_start)
593 free_memmap(node, prev_bank_end, bank_start);
595 prev_bank_end = (mi->bank[i].start +
596 mi->bank[i].size) >> PAGE_SHIFT;
601 * mem_init() marks the free areas in the mem_map and tells us how much
602 * memory is free. This is done after various parts of the system have
603 * claimed their memory after the kernel image.
605 void __init mem_init(void)
607 unsigned int codepages, datapages, initpages;
610 codepages = &_etext - &_text;
611 datapages = &_end - &__data_start;
612 initpages = &__init_end - &__init_begin;
614 #ifndef CONFIG_DISCONTIGMEM
615 max_mapnr = virt_to_page(high_memory) - mem_map;
618 /* this will put all unused low memory onto the freelists */
619 for_each_online_node(node) {
620 pg_data_t *pgdat = NODE_DATA(node);
622 free_unused_memmap_node(node, &meminfo);
624 if (pgdat->node_spanned_pages != 0)
625 totalram_pages += free_all_bootmem_node(pgdat);
629 /* now that our DMA memory is actually so designated, we can free it */
630 free_area(PAGE_OFFSET, (unsigned long)swapper_pg_dir, NULL);
634 * Since our memory may not be contiguous, calculate the
635 * real number of pages we have in this system
637 printk(KERN_INFO "Memory:");
640 for (i = 0; i < meminfo.nr_banks; i++) {
641 num_physpages += meminfo.bank[i].size >> PAGE_SHIFT;
642 printk(" %ldMB", meminfo.bank[i].size >> 20);
645 printk(" = %luMB total\n", num_physpages >> (20 - PAGE_SHIFT));
646 printk(KERN_NOTICE "Memory: %luKB available (%dK code, "
647 "%dK data, %dK init)\n",
648 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
649 codepages >> 10, datapages >> 10, initpages >> 10);
651 if (PAGE_SIZE >= 16384 && num_physpages <= 128) {
652 extern int sysctl_overcommit_memory;
654 * On a machine this small we won't get
655 * anywhere without overcommit, so turn
658 sysctl_overcommit_memory = OVERCOMMIT_ALWAYS;
662 void free_initmem(void)
664 if (!machine_is_integrator() && !machine_is_cintegrator()) {
665 free_area((unsigned long)(&__init_begin),
666 (unsigned long)(&__init_end),
671 #ifdef CONFIG_BLK_DEV_INITRD
673 static int keep_initrd;
675 void free_initrd_mem(unsigned long start, unsigned long end)
678 free_area(start, end, "initrd");
681 static int __init keepinitrd_setup(char *__unused)
687 __setup("keepinitrd", keepinitrd_setup);