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>
30 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
32 extern pgd_t swapper_pg_dir[PTRS_PER_PGD];
33 extern void _stext, _text, _etext, __data_start, _end, __init_begin, __init_end;
34 extern unsigned long phys_initrd_start;
35 extern unsigned long phys_initrd_size;
38 * The sole use of this is to pass memory configuration
39 * data from paging_init to mem_init.
41 static struct meminfo meminfo __initdata = { 0, };
44 * empty_zero_page is a special page that is used for
45 * zero-initialized data and COW.
47 struct page *empty_zero_page;
50 * The pmd table for the upper-most set of pages.
56 int free = 0, total = 0, reserved = 0;
57 int shared = 0, cached = 0, slab = 0, node;
59 printk("Mem-info:\n");
61 printk("Free swap: %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10));
63 for_each_online_node(node) {
64 struct page *page, *end;
66 page = NODE_MEM_MAP(node);
67 end = page + NODE_DATA(node)->node_spanned_pages;
71 if (PageReserved(page))
73 else if (PageSwapCache(page))
75 else if (PageSlab(page))
77 else if (!page_count(page))
80 shared += page_count(page) - 1;
85 printk("%d pages of RAM\n", total);
86 printk("%d free pages\n", free);
87 printk("%d reserved pages\n", reserved);
88 printk("%d slab pages\n", slab);
89 printk("%d pages shared\n", shared);
90 printk("%d pages swap cached\n", cached);
93 #define for_each_nodebank(iter,mi,no) \
94 for (iter = 0; iter < mi->nr_banks; iter++) \
95 if (mi->bank[iter].node == no)
98 * FIXME: We really want to avoid allocating the bootmap bitmap
99 * over the top of the initrd. Hopefully, this is located towards
100 * the start of a bank, so if we allocate the bootmap bitmap at
101 * the end, we won't clash.
103 static unsigned int __init
104 find_bootmap_pfn(int node, struct meminfo *mi, unsigned int bootmap_pages)
106 unsigned int start_pfn, bank, bootmap_pfn;
108 start_pfn = PAGE_ALIGN(__pa(&_end)) >> PAGE_SHIFT;
111 for_each_nodebank(bank, mi, node) {
112 unsigned int start, end;
114 start = mi->bank[bank].start >> PAGE_SHIFT;
115 end = (mi->bank[bank].size +
116 mi->bank[bank].start) >> PAGE_SHIFT;
121 if (start < start_pfn)
127 if (end - start >= bootmap_pages) {
133 if (bootmap_pfn == 0)
139 static int __init check_initrd(struct meminfo *mi)
141 int initrd_node = -2;
142 #ifdef CONFIG_BLK_DEV_INITRD
143 unsigned long end = phys_initrd_start + phys_initrd_size;
146 * Make sure that the initrd is within a valid area of
149 if (phys_initrd_size) {
154 for (i = 0; i < mi->nr_banks; i++) {
155 unsigned long bank_end;
157 bank_end = mi->bank[i].start + mi->bank[i].size;
159 if (mi->bank[i].start <= phys_initrd_start &&
161 initrd_node = mi->bank[i].node;
165 if (initrd_node == -1) {
166 printk(KERN_ERR "initrd (0x%08lx - 0x%08lx) extends beyond "
167 "physical memory - disabling initrd\n",
168 phys_initrd_start, end);
169 phys_initrd_start = phys_initrd_size = 0;
177 * Reserve the various regions of node 0
179 static __init void reserve_node_zero(pg_data_t *pgdat)
181 unsigned long res_size = 0;
184 * Register the kernel text and data with bootmem.
185 * Note that this can only be in node 0.
187 #ifdef CONFIG_XIP_KERNEL
188 reserve_bootmem_node(pgdat, __pa(&__data_start), &_end - &__data_start);
190 reserve_bootmem_node(pgdat, __pa(&_stext), &_end - &_stext);
194 * Reserve the page tables. These are already in use,
195 * and can only be in node 0.
197 reserve_bootmem_node(pgdat, __pa(swapper_pg_dir),
198 PTRS_PER_PGD * sizeof(pgd_t));
201 * Hmm... This should go elsewhere, but we really really need to
202 * stop things allocating the low memory; ideally we need a better
203 * implementation of GFP_DMA which does not assume that DMA-able
204 * memory starts at zero.
206 if (machine_is_integrator() || machine_is_cintegrator())
207 res_size = __pa(swapper_pg_dir) - PHYS_OFFSET;
210 * These should likewise go elsewhere. They pre-reserve the
211 * screen memory region at the start of main system memory.
213 if (machine_is_edb7211())
214 res_size = 0x00020000;
215 if (machine_is_p720t())
216 res_size = 0x00014000;
220 * Because of the SA1111 DMA bug, we want to preserve our
221 * precious DMA-able memory...
223 res_size = __pa(swapper_pg_dir) - PHYS_OFFSET;
226 reserve_bootmem_node(pgdat, PHYS_OFFSET, res_size);
229 static inline void prepare_page_table(struct meminfo *mi)
234 * Clear out all the mappings below the kernel image.
236 for (addr = 0; addr < MODULE_START; addr += PGDIR_SIZE)
237 pmd_clear(pmd_off_k(addr));
239 #ifdef CONFIG_XIP_KERNEL
240 /* The XIP kernel is mapped in the module area -- skip over it */
241 addr = ((unsigned long)&_etext + PGDIR_SIZE - 1) & PGDIR_MASK;
243 for ( ; addr < PAGE_OFFSET; addr += PGDIR_SIZE)
244 pmd_clear(pmd_off_k(addr));
247 * Clear out all the kernel space mappings, except for the first
248 * memory bank, up to the end of the vmalloc region.
250 for (addr = __phys_to_virt(mi->bank[0].start + mi->bank[0].size);
251 addr < VMALLOC_END; addr += PGDIR_SIZE)
252 pmd_clear(pmd_off_k(addr));
255 static inline void map_memory_bank(struct membank *bank)
259 map.pfn = __phys_to_pfn(bank->start);
260 map.virtual = __phys_to_virt(bank->start);
261 map.length = bank->size;
262 map.type = MT_MEMORY;
264 create_mapping(&map);
267 static unsigned long __init
268 bootmem_init_node(int node, int initrd_node, struct meminfo *mi)
270 unsigned long zone_size[MAX_NR_ZONES], zhole_size[MAX_NR_ZONES];
271 unsigned long start_pfn, end_pfn, boot_pfn;
272 unsigned int boot_pages;
280 * Calculate the pfn range, and map the memory banks for this node.
282 for_each_nodebank(i, mi, node) {
283 struct membank *bank = &mi->bank[i];
284 unsigned long start, end;
286 start = bank->start >> PAGE_SHIFT;
287 end = (bank->start + bank->size) >> PAGE_SHIFT;
289 if (start_pfn > start)
294 map_memory_bank(bank);
298 * If there is no memory in this node, ignore it.
304 * Allocate the bootmem bitmap page.
306 boot_pages = bootmem_bootmap_pages(end_pfn - start_pfn);
307 boot_pfn = find_bootmap_pfn(node, mi, boot_pages);
310 * Initialise the bootmem allocator for this node, handing the
311 * memory banks over to bootmem.
313 node_set_online(node);
314 pgdat = NODE_DATA(node);
315 init_bootmem_node(pgdat, boot_pfn, start_pfn, end_pfn);
317 for_each_nodebank(i, mi, node)
318 free_bootmem_node(pgdat, mi->bank[i].start, mi->bank[i].size);
321 * Reserve the bootmem bitmap for this node.
323 reserve_bootmem_node(pgdat, boot_pfn << PAGE_SHIFT,
324 boot_pages << PAGE_SHIFT);
326 #ifdef CONFIG_BLK_DEV_INITRD
328 * If the initrd is in this node, reserve its memory.
330 if (node == initrd_node) {
331 reserve_bootmem_node(pgdat, phys_initrd_start,
333 initrd_start = __phys_to_virt(phys_initrd_start);
334 initrd_end = initrd_start + phys_initrd_size;
339 * Finally, reserve any node zero regions.
342 reserve_node_zero(pgdat);
345 * initialise the zones within this node.
347 memset(zone_size, 0, sizeof(zone_size));
348 memset(zhole_size, 0, sizeof(zhole_size));
351 * The size of this node has already been determined. If we need
352 * to do anything fancy with the allocation of this memory to the
353 * zones, now is the time to do it.
355 zone_size[0] = end_pfn - start_pfn;
358 * For each bank in this node, calculate the size of the holes.
359 * holes = node_size - sum(bank_sizes_in_node)
361 zhole_size[0] = zone_size[0];
362 for_each_nodebank(i, mi, node)
363 zhole_size[0] -= mi->bank[i].size >> PAGE_SHIFT;
366 * Adjust the sizes according to any special requirements for
369 arch_adjust_zones(node, zone_size, zhole_size);
371 free_area_init_node(node, pgdat, zone_size, start_pfn, zhole_size);
376 static void __init bootmem_init(struct meminfo *mi)
378 unsigned long memend_pfn = 0;
379 int node, initrd_node, i;
382 * Invalidate the node number for empty or invalid memory banks
384 for (i = 0; i < mi->nr_banks; i++)
385 if (mi->bank[i].size == 0 || mi->bank[i].node >= MAX_NUMNODES)
386 mi->bank[i].node = -1;
388 memcpy(&meminfo, mi, sizeof(meminfo));
390 prepare_page_table(mi);
393 * Locate which node contains the ramdisk image, if any.
395 initrd_node = check_initrd(mi);
398 * Run through each node initialising the bootmem allocator.
400 for_each_node(node) {
401 unsigned long end_pfn;
403 end_pfn = bootmem_init_node(node, initrd_node, mi);
406 * Remember the highest memory PFN.
408 if (end_pfn > memend_pfn)
409 memend_pfn = end_pfn;
412 high_memory = __va(memend_pfn << PAGE_SHIFT);
415 * This doesn't seem to be used by the Linux memory manager any
416 * more, but is used by ll_rw_block. If we can get rid of it, we
417 * also get rid of some of the stuff above as well.
419 * Note: max_low_pfn and max_pfn reflect the number of _pages_ in
420 * the system, not the maximum PFN.
422 max_pfn = max_low_pfn = memend_pfn - PHYS_PFN_OFFSET;
426 * Set up device the mappings. Since we clear out the page tables for all
427 * mappings above VMALLOC_END, we will remove any debug device mappings.
428 * This means you have to be careful how you debug this function, or any
429 * called function. This means you can't use any function or debugging
430 * method which may touch any device, otherwise the kernel _will_ crash.
432 static void __init devicemaps_init(struct machine_desc *mdesc)
439 * Allocate the vector page early.
441 vectors = alloc_bootmem_low_pages(PAGE_SIZE);
444 for (addr = VMALLOC_END; addr; addr += PGDIR_SIZE)
445 pmd_clear(pmd_off_k(addr));
448 * Map the kernel if it is XIP.
449 * It is always first in the modulearea.
451 #ifdef CONFIG_XIP_KERNEL
452 map.pfn = __phys_to_pfn(CONFIG_XIP_PHYS_ADDR & SECTION_MASK);
453 map.virtual = MODULE_START;
454 map.length = ((unsigned long)&_etext - map.virtual + ~SECTION_MASK) & SECTION_MASK;
456 create_mapping(&map);
460 * Map the cache flushing regions.
463 map.pfn = __phys_to_pfn(FLUSH_BASE_PHYS);
464 map.virtual = FLUSH_BASE;
466 map.type = MT_CACHECLEAN;
467 create_mapping(&map);
469 #ifdef FLUSH_BASE_MINICACHE
470 map.pfn = __phys_to_pfn(FLUSH_BASE_PHYS + SZ_1M);
471 map.virtual = FLUSH_BASE_MINICACHE;
473 map.type = MT_MINICLEAN;
474 create_mapping(&map);
478 * Create a mapping for the machine vectors at the high-vectors
479 * location (0xffff0000). If we aren't using high-vectors, also
480 * create a mapping at the low-vectors virtual address.
482 map.pfn = __phys_to_pfn(virt_to_phys(vectors));
483 map.virtual = 0xffff0000;
484 map.length = PAGE_SIZE;
485 map.type = MT_HIGH_VECTORS;
486 create_mapping(&map);
488 if (!vectors_high()) {
490 map.type = MT_LOW_VECTORS;
491 create_mapping(&map);
495 * Ask the machine support to map in the statically mapped devices.
501 * Finally flush the caches and tlb to ensure that we're in a
502 * consistent state wrt the writebuffer. This also ensures that
503 * any write-allocated cache lines in the vector page are written
504 * back. After this point, we can start to touch devices again.
506 local_flush_tlb_all();
511 * paging_init() sets up the page tables, initialises the zone memory
512 * maps, and sets up the zero page, bad page and bad page tables.
514 void __init paging_init(struct meminfo *mi, struct machine_desc *mdesc)
518 build_mem_type_table();
520 devicemaps_init(mdesc);
522 top_pmd = pmd_off_k(0xffff0000);
525 * allocate the zero page. Note that we count on this going ok.
527 zero_page = alloc_bootmem_low_pages(PAGE_SIZE);
528 memzero(zero_page, PAGE_SIZE);
529 empty_zero_page = virt_to_page(zero_page);
530 flush_dcache_page(empty_zero_page);
533 static inline void free_area(unsigned long addr, unsigned long end, char *s)
535 unsigned int size = (end - addr) >> 10;
537 for (; addr < end; addr += PAGE_SIZE) {
538 struct page *page = virt_to_page(addr);
539 ClearPageReserved(page);
540 init_page_count(page);
546 printk(KERN_INFO "Freeing %s memory: %dK\n", s, size);
550 free_memmap(int node, unsigned long start_pfn, unsigned long end_pfn)
552 struct page *start_pg, *end_pg;
553 unsigned long pg, pgend;
556 * Convert start_pfn/end_pfn to a struct page pointer.
558 start_pg = pfn_to_page(start_pfn);
559 end_pg = pfn_to_page(end_pfn);
562 * Convert to physical addresses, and
563 * round start upwards and end downwards.
565 pg = PAGE_ALIGN(__pa(start_pg));
566 pgend = __pa(end_pg) & PAGE_MASK;
569 * If there are free pages between these,
570 * free the section of the memmap array.
573 free_bootmem_node(NODE_DATA(node), pg, pgend - pg);
577 * The mem_map array can get very big. Free the unused area of the memory map.
579 static void __init free_unused_memmap_node(int node, struct meminfo *mi)
581 unsigned long bank_start, prev_bank_end = 0;
585 * [FIXME] This relies on each bank being in address order. This
586 * may not be the case, especially if the user has provided the
587 * information on the command line.
589 for_each_nodebank(i, mi, node) {
590 bank_start = mi->bank[i].start >> PAGE_SHIFT;
591 if (bank_start < prev_bank_end) {
592 printk(KERN_ERR "MEM: unordered memory banks. "
593 "Not freeing memmap.\n");
598 * If we had a previous bank, and there is a space
599 * between the current bank and the previous, free it.
601 if (prev_bank_end && prev_bank_end != bank_start)
602 free_memmap(node, prev_bank_end, bank_start);
604 prev_bank_end = (mi->bank[i].start +
605 mi->bank[i].size) >> PAGE_SHIFT;
610 * mem_init() marks the free areas in the mem_map and tells us how much
611 * memory is free. This is done after various parts of the system have
612 * claimed their memory after the kernel image.
614 void __init mem_init(void)
616 unsigned int codepages, datapages, initpages;
619 codepages = &_etext - &_text;
620 datapages = &_end - &__data_start;
621 initpages = &__init_end - &__init_begin;
623 #ifndef CONFIG_DISCONTIGMEM
624 max_mapnr = virt_to_page(high_memory) - mem_map;
627 /* this will put all unused low memory onto the freelists */
628 for_each_online_node(node) {
629 pg_data_t *pgdat = NODE_DATA(node);
631 free_unused_memmap_node(node, &meminfo);
633 if (pgdat->node_spanned_pages != 0)
634 totalram_pages += free_all_bootmem_node(pgdat);
638 /* now that our DMA memory is actually so designated, we can free it */
639 free_area(PAGE_OFFSET, (unsigned long)swapper_pg_dir, NULL);
643 * Since our memory may not be contiguous, calculate the
644 * real number of pages we have in this system
646 printk(KERN_INFO "Memory:");
649 for (i = 0; i < meminfo.nr_banks; i++) {
650 num_physpages += meminfo.bank[i].size >> PAGE_SHIFT;
651 printk(" %ldMB", meminfo.bank[i].size >> 20);
654 printk(" = %luMB total\n", num_physpages >> (20 - PAGE_SHIFT));
655 printk(KERN_NOTICE "Memory: %luKB available (%dK code, "
656 "%dK data, %dK init)\n",
657 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
658 codepages >> 10, datapages >> 10, initpages >> 10);
660 if (PAGE_SIZE >= 16384 && num_physpages <= 128) {
661 extern int sysctl_overcommit_memory;
663 * On a machine this small we won't get
664 * anywhere without overcommit, so turn
667 sysctl_overcommit_memory = OVERCOMMIT_ALWAYS;
671 void free_initmem(void)
673 if (!machine_is_integrator() && !machine_is_cintegrator()) {
674 free_area((unsigned long)(&__init_begin),
675 (unsigned long)(&__init_end),
680 #ifdef CONFIG_BLK_DEV_INITRD
682 static int keep_initrd;
684 void free_initrd_mem(unsigned long start, unsigned long end)
687 free_area(start, end, "initrd");
690 static int __init keepinitrd_setup(char *__unused)
696 __setup("keepinitrd", keepinitrd_setup);