2 * linux/arch/i386/mm/init.c
4 * Copyright (C) 1995 Linus Torvalds
6 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
9 #include <linux/config.h>
10 #include <linux/module.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>
20 #include <linux/hugetlb.h>
21 #include <linux/swap.h>
22 #include <linux/smp.h>
23 #include <linux/init.h>
24 #include <linux/highmem.h>
25 #include <linux/pagemap.h>
26 #include <linux/bootmem.h>
27 #include <linux/slab.h>
28 #include <linux/proc_fs.h>
29 #include <linux/efi.h>
31 #include <asm/processor.h>
32 #include <asm/system.h>
33 #include <asm/uaccess.h>
34 #include <asm/pgtable.h>
36 #include <asm/fixmap.h>
40 #include <asm/tlbflush.h>
41 #include <asm/sections.h>
43 unsigned int __VMALLOC_RESERVE = 128 << 20;
45 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
46 unsigned long highstart_pfn, highend_pfn;
48 static int noinline do_test_wp_bit(void);
51 * Creates a middle page table and puts a pointer to it in the
52 * given global directory entry. This only returns the gd entry
53 * in non-PAE compilation mode, since the middle layer is folded.
55 static pmd_t * __init one_md_table_init(pgd_t *pgd)
60 pmd_table = (pmd_t *) alloc_bootmem_low_pages(PAGE_SIZE);
61 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
62 if (pmd_table != pmd_offset(pgd, 0))
65 pmd_table = pmd_offset(pgd, 0);
72 * Create a page table and place a pointer to it in a middle page
75 static pte_t * __init one_page_table_init(pmd_t *pmd)
78 pte_t *page_table = (pte_t *) alloc_bootmem_low_pages(PAGE_SIZE);
79 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
80 if (page_table != pte_offset_kernel(pmd, 0))
86 return pte_offset_kernel(pmd, 0);
90 * This function initializes a certain range of kernel virtual memory
91 * with new bootmem page tables, everywhere page tables are missing in
96 * NOTE: The pagetables are allocated contiguous on the physical space
97 * so we can cache the place of the first one and move around without
98 * checking the pgd every time.
100 static void __init page_table_range_init (unsigned long start, unsigned long end, pgd_t *pgd_base)
104 int pgd_idx, pmd_idx;
108 pgd_idx = pgd_index(vaddr);
109 pmd_idx = pmd_index(vaddr);
110 pgd = pgd_base + pgd_idx;
112 for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
114 one_md_table_init(pgd);
116 pmd = pmd_offset(pgd, vaddr);
117 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end); pmd++, pmd_idx++) {
119 one_page_table_init(pmd);
127 static inline int is_kernel_text(unsigned long addr)
129 if (addr >= (unsigned long)_stext && addr <= (unsigned long)__init_end)
135 * This maps the physical memory to kernel virtual address space, a total
136 * of max_low_pfn pages, by creating page tables starting from address
139 static void __init kernel_physical_mapping_init(pgd_t *pgd_base)
145 int pgd_idx, pmd_idx, pte_ofs;
147 pgd_idx = pgd_index(PAGE_OFFSET);
148 pgd = pgd_base + pgd_idx;
151 for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
152 pmd = one_md_table_init(pgd);
153 if (pfn >= max_low_pfn)
155 for (pmd_idx = 0; pmd_idx < PTRS_PER_PMD && pfn < max_low_pfn; pmd++, pmd_idx++) {
156 unsigned int address = pfn * PAGE_SIZE + PAGE_OFFSET;
158 /* Map with big pages if possible, otherwise create normal page tables. */
160 unsigned int address2 = (pfn + PTRS_PER_PTE - 1) * PAGE_SIZE + PAGE_OFFSET + PAGE_SIZE-1;
162 if (is_kernel_text(address) || is_kernel_text(address2))
163 set_pmd(pmd, pfn_pmd(pfn, PAGE_KERNEL_LARGE_EXEC));
165 set_pmd(pmd, pfn_pmd(pfn, PAGE_KERNEL_LARGE));
168 pte = one_page_table_init(pmd);
170 for (pte_ofs = 0; pte_ofs < PTRS_PER_PTE && pfn < max_low_pfn; pte++, pfn++, pte_ofs++) {
171 if (is_kernel_text(address))
172 set_pte(pte, pfn_pte(pfn, PAGE_KERNEL_EXEC));
174 set_pte(pte, pfn_pte(pfn, PAGE_KERNEL));
181 static inline int page_kills_ppro(unsigned long pagenr)
183 if (pagenr >= 0x70000 && pagenr <= 0x7003F)
188 extern int is_available_memory(efi_memory_desc_t *);
190 static inline int page_is_ram(unsigned long pagenr)
193 unsigned long addr, end;
196 efi_memory_desc_t *md;
198 for (i = 0; i < memmap.nr_map; i++) {
200 if (!is_available_memory(md))
202 addr = (md->phys_addr+PAGE_SIZE-1) >> PAGE_SHIFT;
203 end = (md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT)) >> PAGE_SHIFT;
205 if ((pagenr >= addr) && (pagenr < end))
211 for (i = 0; i < e820.nr_map; i++) {
213 if (e820.map[i].type != E820_RAM) /* not usable memory */
216 * !!!FIXME!!! Some BIOSen report areas as RAM that
217 * are not. Notably the 640->1Mb area. We need a sanity
220 addr = (e820.map[i].addr+PAGE_SIZE-1) >> PAGE_SHIFT;
221 end = (e820.map[i].addr+e820.map[i].size) >> PAGE_SHIFT;
222 if ((pagenr >= addr) && (pagenr < end))
228 /* To enable modules to check if a page is in RAM */
229 int pfn_is_ram(unsigned long pfn)
231 return (page_is_ram(pfn));
236 * devmem_is_allowed() checks to see if /dev/mem access to a certain address is
237 * valid. The argument is a physical page number.
240 * On x86, access has to be given to the first megabyte of ram because that area
241 * contains bios code and data regions used by X and dosemu and similar apps.
242 * Access has to be given to non-kernel-ram areas as well, these contain the PCI
243 * mmio resources as well as potential bios/acpi data regions.
245 int devmem_is_allowed(unsigned long pagenr)
249 if (!page_is_ram(pagenr))
254 EXPORT_SYMBOL_GPL(page_is_ram);
256 #ifdef CONFIG_HIGHMEM
260 EXPORT_SYMBOL(kmap_prot);
261 EXPORT_SYMBOL(kmap_pte);
263 #define kmap_get_fixmap_pte(vaddr) \
264 pte_offset_kernel(pmd_offset(pgd_offset_k(vaddr), (vaddr)), (vaddr))
266 void __init kmap_init(void)
268 unsigned long kmap_vstart;
270 /* cache the first kmap pte */
271 kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
272 kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
274 kmap_prot = PAGE_KERNEL;
277 void __init permanent_kmaps_init(pgd_t *pgd_base)
285 page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
287 pgd = swapper_pg_dir + pgd_index(vaddr);
288 pmd = pmd_offset(pgd, vaddr);
289 pte = pte_offset_kernel(pmd, vaddr);
290 pkmap_page_table = pte;
293 void __init one_highpage_init(struct page *page, int pfn, int bad_ppro)
295 if (page_is_ram(pfn) && !(bad_ppro && page_kills_ppro(pfn))) {
296 ClearPageReserved(page);
297 set_bit(PG_highmem, &page->flags);
298 set_page_count(page, 1);
302 SetPageReserved(page);
305 #ifndef CONFIG_DISCONTIGMEM
306 void __init set_highmem_pages_init(int bad_ppro)
309 for (pfn = highstart_pfn; pfn < highend_pfn; pfn++)
310 one_highpage_init(pfn_to_page(pfn), pfn, bad_ppro);
311 totalram_pages += totalhigh_pages;
314 extern void set_highmem_pages_init(int);
315 #endif /* !CONFIG_DISCONTIGMEM */
318 #define kmap_init() do { } while (0)
319 #define permanent_kmaps_init(pgd_base) do { } while (0)
320 #define set_highmem_pages_init(bad_ppro) do { } while (0)
321 #endif /* CONFIG_HIGHMEM */
323 unsigned long long __PAGE_KERNEL = _PAGE_KERNEL;
324 unsigned long long __PAGE_KERNEL_EXEC = _PAGE_KERNEL_EXEC;
326 #ifndef CONFIG_DISCONTIGMEM
327 #define remap_numa_kva() do {} while (0)
329 extern void __init remap_numa_kva(void);
332 static void __init pagetable_init (void)
335 pgd_t *pgd_base = swapper_pg_dir;
337 #ifdef CONFIG_X86_PAE
339 /* Init entries of the first-level page table to the zero page */
340 for (i = 0; i < PTRS_PER_PGD; i++)
341 set_pgd(pgd_base + i, __pgd(__pa(empty_zero_page) | _PAGE_PRESENT));
344 /* Enable PSE if available */
346 set_in_cr4(X86_CR4_PSE);
349 /* Enable PGE if available */
351 set_in_cr4(X86_CR4_PGE);
352 __PAGE_KERNEL |= _PAGE_GLOBAL;
353 __PAGE_KERNEL_EXEC |= _PAGE_GLOBAL;
356 kernel_physical_mapping_init(pgd_base);
360 * Fixed mappings, only the page table structure has to be
361 * created - mappings will be set by set_fixmap():
363 vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
364 page_table_range_init(vaddr, 0, pgd_base);
366 permanent_kmaps_init(pgd_base);
368 #ifdef CONFIG_X86_PAE
370 * Add low memory identity-mappings - SMP needs it when
371 * starting up on an AP from real-mode. In the non-PAE
372 * case we already have these mappings through head.S.
373 * All user-space mappings are explicitly cleared after
376 pgd_base[0] = pgd_base[USER_PTRS_PER_PGD];
380 #if defined(CONFIG_PM_DISK) || defined(CONFIG_SOFTWARE_SUSPEND)
382 * Swap suspend & friends need this for resume because things like the intel-agp
383 * driver might have split up a kernel 4MB mapping.
385 char __nosavedata swsusp_pg_dir[PAGE_SIZE]
386 __attribute__ ((aligned (PAGE_SIZE)));
388 static inline void save_pg_dir(void)
390 memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
393 static inline void save_pg_dir(void)
398 void zap_low_mappings (void)
405 * Zap initial low-memory mappings.
407 * Note that "pgd_clear()" doesn't do it for
408 * us, because pgd_clear() is a no-op on i386.
410 for (i = 0; i < USER_PTRS_PER_PGD; i++)
411 #ifdef CONFIG_X86_PAE
412 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
414 set_pgd(swapper_pg_dir+i, __pgd(0));
419 #ifndef CONFIG_DISCONTIGMEM
420 void __init zone_sizes_init(void)
422 unsigned long zones_size[MAX_NR_ZONES] = {0, 0, 0};
423 unsigned int max_dma, high, low;
425 max_dma = virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
430 zones_size[ZONE_DMA] = low;
432 zones_size[ZONE_DMA] = max_dma;
433 zones_size[ZONE_NORMAL] = low - max_dma;
434 #ifdef CONFIG_HIGHMEM
435 zones_size[ZONE_HIGHMEM] = high - low;
438 free_area_init(zones_size);
441 extern void zone_sizes_init(void);
442 #endif /* !CONFIG_DISCONTIGMEM */
444 static int disable_nx __initdata = 0;
445 u64 __supported_pte_mask = ~_PAGE_NX;
450 * Control non executable mappings.
453 * off Disable (disables exec-shield too)
455 static int __init noexec_setup(char *str)
457 if (!strncmp(str, "on",2) && cpu_has_nx) {
458 __supported_pte_mask |= _PAGE_NX;
460 } else if (!strncmp(str,"off",3)) {
462 __supported_pte_mask &= ~_PAGE_NX;
468 __setup("noexec=", noexec_setup);
470 #ifdef CONFIG_X86_PAE
473 static void __init set_nx(void)
475 unsigned int v[4], l, h;
477 if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) {
478 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]);
479 if ((v[3] & (1 << 20)) && !disable_nx) {
480 rdmsr(MSR_EFER, l, h);
482 wrmsr(MSR_EFER, l, h);
484 __supported_pte_mask |= _PAGE_NX;
490 * Enables/disables executability of a given kernel page and
491 * returns the previous setting.
493 int __init set_kernel_exec(unsigned long vaddr, int enable)
501 pte = lookup_address(vaddr);
504 if (!pte_exec_kernel(*pte))
508 pte->pte_high &= ~(1 << (_PAGE_BIT_NX - 32));
510 pte->pte_high |= 1 << (_PAGE_BIT_NX - 32);
519 * paging_init() sets up the page tables - note that the first 8MB are
520 * already mapped by head.S.
522 * This routines also unmaps the page at virtual kernel address 0, so
523 * that we can trap those pesky NULL-reference errors in the kernel.
525 void __init paging_init(void)
527 #ifdef CONFIG_X86_PAE
530 printk("NX (Execute Disable) protection: active\n");
535 load_cr3(swapper_pg_dir);
537 #ifdef CONFIG_X86_PAE
539 * We will bail out later - printk doesn't work right now so
540 * the user would just see a hanging kernel.
543 set_in_cr4(X86_CR4_PAE);
552 * Test if the WP bit works in supervisor mode. It isn't supported on 386's
553 * and also on some strange 486's (NexGen etc.). All 586+'s are OK. This
554 * used to involve black magic jumps to work around some nasty CPU bugs,
555 * but fortunately the switch to using exceptions got rid of all that.
558 void __init test_wp_bit(void)
560 printk("Checking if this processor honours the WP bit even in supervisor mode... ");
562 /* Any page-aligned address will do, the test is non-destructive */
563 __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
564 boot_cpu_data.wp_works_ok = do_test_wp_bit();
565 clear_fixmap(FIX_WP_TEST);
567 if (!boot_cpu_data.wp_works_ok) {
569 #ifdef CONFIG_X86_WP_WORKS_OK
570 panic("This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
577 #ifndef CONFIG_DISCONTIGMEM
578 static void __init set_max_mapnr_init(void)
580 #ifdef CONFIG_HIGHMEM
581 highmem_start_page = pfn_to_page(highstart_pfn);
582 max_mapnr = num_physpages = highend_pfn;
584 max_mapnr = num_physpages = max_low_pfn;
587 #define __free_all_bootmem() free_all_bootmem()
589 #define __free_all_bootmem() free_all_bootmem_node(NODE_DATA(0))
590 extern void set_max_mapnr_init(void);
591 #endif /* !CONFIG_DISCONTIGMEM */
593 static struct kcore_list kcore_mem, kcore_vmalloc;
595 void __init mem_init(void)
597 extern int ppro_with_ram_bug(void);
598 int codesize, reservedpages, datasize, initsize;
602 #ifndef CONFIG_DISCONTIGMEM
607 bad_ppro = ppro_with_ram_bug();
609 #ifdef CONFIG_HIGHMEM
610 /* check that fixmap and pkmap do not overlap */
611 if (PKMAP_BASE+LAST_PKMAP*PAGE_SIZE >= FIXADDR_START) {
612 printk(KERN_ERR "fixmap and kmap areas overlap - this will crash\n");
613 printk(KERN_ERR "pkstart: %lxh pkend: %lxh fixstart %lxh\n",
614 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE, FIXADDR_START);
619 set_max_mapnr_init();
621 #ifdef CONFIG_HIGHMEM
622 high_memory = (void *) __va(highstart_pfn * PAGE_SIZE);
624 high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
627 /* this will put all low memory onto the freelists */
628 totalram_pages += __free_all_bootmem();
631 for (tmp = 0; tmp < max_low_pfn; tmp++)
633 * Only count reserved RAM pages
635 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
638 set_highmem_pages_init(bad_ppro);
640 codesize = (unsigned long) &_etext - (unsigned long) &_text;
641 datasize = (unsigned long) &_edata - (unsigned long) &_etext;
642 initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin;
644 kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
645 kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
646 VMALLOC_END-VMALLOC_START);
648 printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, %dk reserved, %dk data, %dk init, %ldk highmem)\n",
649 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
650 num_physpages << (PAGE_SHIFT-10),
652 reservedpages << (PAGE_SHIFT-10),
655 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10))
658 #ifdef CONFIG_X86_PAE
660 panic("cannot execute a PAE-enabled kernel on a PAE-less CPU!");
662 if (boot_cpu_data.wp_works_ok < 0)
666 * Subtle. SMP is doing it's boot stuff late (because it has to
667 * fork idle threads) - but it also needs low mappings for the
668 * protected-mode entry to work. We zap these entries only after
669 * the WP-bit has been tested.
676 kmem_cache_t *pgd_cache;
677 kmem_cache_t *pmd_cache;
679 void __init pgtable_cache_init(void)
681 if (PTRS_PER_PMD > 1) {
682 pmd_cache = kmem_cache_create("pmd",
683 PTRS_PER_PMD*sizeof(pmd_t),
684 PTRS_PER_PMD*sizeof(pmd_t),
689 panic("pgtable_cache_init(): cannot create pmd cache");
691 pgd_cache = kmem_cache_create("pgd",
692 PTRS_PER_PGD*sizeof(pgd_t),
693 PTRS_PER_PGD*sizeof(pgd_t),
696 PTRS_PER_PMD == 1 ? pgd_dtor : NULL);
698 panic("pgtable_cache_init(): Cannot create pgd cache");
702 * This function cannot be __init, since exceptions don't work in that
703 * section. Put this after the callers, so that it cannot be inlined.
705 static int noinline do_test_wp_bit(void)
710 __asm__ __volatile__(
715 ".section __ex_table,\"a\"\n"
719 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
728 void free_initmem(void)
732 addr = (unsigned long)(&__init_begin);
733 for (; addr < (unsigned long)(&__init_end); addr += PAGE_SIZE) {
734 ClearPageReserved(virt_to_page(addr));
735 set_page_count(virt_to_page(addr), 1);
739 printk (KERN_INFO "Freeing unused kernel memory: %dk freed\n", (__init_end - __init_begin) >> 10);
742 #ifdef CONFIG_BLK_DEV_INITRD
743 void free_initrd_mem(unsigned long start, unsigned long end)
746 printk (KERN_INFO "Freeing initrd memory: %ldk freed\n", (end - start) >> 10);
747 for (; start < end; start += PAGE_SIZE) {
748 ClearPageReserved(virt_to_page(start));
749 set_page_count(virt_to_page(start), 1);