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/module.h>
10 #include <linux/signal.h>
11 #include <linux/sched.h>
12 #include <linux/kernel.h>
13 #include <linux/errno.h>
14 #include <linux/string.h>
15 #include <linux/types.h>
16 #include <linux/ptrace.h>
17 #include <linux/mman.h>
19 #include <linux/hugetlb.h>
20 #include <linux/swap.h>
21 #include <linux/smp.h>
22 #include <linux/init.h>
23 #include <linux/highmem.h>
24 #include <linux/pagemap.h>
25 #include <linux/poison.h>
26 #include <linux/bootmem.h>
27 #include <linux/slab.h>
28 #include <linux/proc_fs.h>
29 #include <linux/efi.h>
30 #include <linux/memory_hotplug.h>
31 #include <linux/initrd.h>
32 #include <linux/cpumask.h>
34 #include <asm/processor.h>
35 #include <asm/system.h>
36 #include <asm/uaccess.h>
37 #include <asm/pgtable.h>
39 #include <asm/fixmap.h>
43 #include <asm/tlbflush.h>
44 #include <asm/sections.h>
46 unsigned int __VMALLOC_RESERVE = 128 << 20;
48 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
49 unsigned long highstart_pfn, highend_pfn;
51 static int noinline do_test_wp_bit(void);
54 * Creates a middle page table and puts a pointer to it in the
55 * given global directory entry. This only returns the gd entry
56 * in non-PAE compilation mode, since the middle layer is folded.
58 static pmd_t * __init one_md_table_init(pgd_t *pgd)
64 pmd_table = (pmd_t *) alloc_bootmem_low_pages(PAGE_SIZE);
65 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
66 pud = pud_offset(pgd, 0);
67 if (pmd_table != pmd_offset(pud, 0))
70 pud = pud_offset(pgd, 0);
71 pmd_table = pmd_offset(pud, 0);
78 * Create a page table and place a pointer to it in a middle page
81 static pte_t * __init one_page_table_init(pmd_t *pmd)
84 pte_t *page_table = (pte_t *) alloc_bootmem_low_pages(PAGE_SIZE);
85 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
86 if (page_table != pte_offset_kernel(pmd, 0))
92 return pte_offset_kernel(pmd, 0);
96 * This function initializes a certain range of kernel virtual memory
97 * with new bootmem page tables, everywhere page tables are missing in
102 * NOTE: The pagetables are allocated contiguous on the physical space
103 * so we can cache the place of the first one and move around without
104 * checking the pgd every time.
106 static void __init page_table_range_init (unsigned long start, unsigned long end, pgd_t *pgd_base)
111 int pgd_idx, pmd_idx;
115 pgd_idx = pgd_index(vaddr);
116 pmd_idx = pmd_index(vaddr);
117 pgd = pgd_base + pgd_idx;
119 for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
121 one_md_table_init(pgd);
122 pud = pud_offset(pgd, vaddr);
123 pmd = pmd_offset(pud, vaddr);
124 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end); pmd++, pmd_idx++) {
126 one_page_table_init(pmd);
134 static inline int is_kernel_text(unsigned long addr)
136 if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end)
142 * This maps the physical memory to kernel virtual address space, a total
143 * of max_low_pfn pages, by creating page tables starting from address
146 static void __init kernel_physical_mapping_init(pgd_t *pgd_base)
152 int pgd_idx, pmd_idx, pte_ofs;
154 pgd_idx = pgd_index(PAGE_OFFSET);
155 pgd = pgd_base + pgd_idx;
158 for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
159 pmd = one_md_table_init(pgd);
160 if (pfn >= max_low_pfn)
162 for (pmd_idx = 0; pmd_idx < PTRS_PER_PMD && pfn < max_low_pfn; pmd++, pmd_idx++) {
163 unsigned int address = pfn * PAGE_SIZE + PAGE_OFFSET;
165 /* Map with big pages if possible, otherwise create normal page tables. */
167 unsigned int address2 = (pfn + PTRS_PER_PTE - 1) * PAGE_SIZE + PAGE_OFFSET + PAGE_SIZE-1;
169 if (is_kernel_text(address) || is_kernel_text(address2))
170 set_pmd(pmd, pfn_pmd(pfn, PAGE_KERNEL_LARGE_EXEC));
172 set_pmd(pmd, pfn_pmd(pfn, PAGE_KERNEL_LARGE));
175 pte = one_page_table_init(pmd);
177 for (pte_ofs = 0; pte_ofs < PTRS_PER_PTE && pfn < max_low_pfn; pte++, pfn++, pte_ofs++) {
178 if (is_kernel_text(address))
179 set_pte(pte, pfn_pte(pfn, PAGE_KERNEL_EXEC));
181 set_pte(pte, pfn_pte(pfn, PAGE_KERNEL));
188 static inline int page_kills_ppro(unsigned long pagenr)
190 if (pagenr >= 0x70000 && pagenr <= 0x7003F)
195 int page_is_ram(unsigned long pagenr)
198 unsigned long addr, end;
201 efi_memory_desc_t *md;
204 for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
206 if (!is_available_memory(md))
208 addr = (md->phys_addr+PAGE_SIZE-1) >> PAGE_SHIFT;
209 end = (md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT)) >> PAGE_SHIFT;
211 if ((pagenr >= addr) && (pagenr < end))
217 for (i = 0; i < e820.nr_map; i++) {
219 if (e820.map[i].type != E820_RAM) /* not usable memory */
222 * !!!FIXME!!! Some BIOSen report areas as RAM that
223 * are not. Notably the 640->1Mb area. We need a sanity
226 addr = (e820.map[i].addr+PAGE_SIZE-1) >> PAGE_SHIFT;
227 end = (e820.map[i].addr+e820.map[i].size) >> PAGE_SHIFT;
228 if ((pagenr >= addr) && (pagenr < end))
235 * devmem_is_allowed() checks to see if /dev/mem access to a certain address is
236 * valid. The argument is a physical page number.
239 * On x86, access has to be given to the first megabyte of ram because that area
240 * contains bios code and data regions used by X and dosemu and similar apps.
241 * Access has to be given to non-kernel-ram areas as well, these contain the PCI
242 * mmio resources as well as potential bios/acpi data regions.
244 int devmem_is_allowed(unsigned long pagenr)
248 if (!page_is_ram(pagenr))
253 EXPORT_SYMBOL_GPL(page_is_ram);
255 #ifdef CONFIG_HIGHMEM
259 #define kmap_get_fixmap_pte(vaddr) \
260 pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr), vaddr), (vaddr)), (vaddr))
262 static void __init kmap_init(void)
264 unsigned long kmap_vstart;
266 /* cache the first kmap pte */
267 kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
268 kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
270 kmap_prot = PAGE_KERNEL;
273 static void __init permanent_kmaps_init(pgd_t *pgd_base)
282 page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
284 pgd = swapper_pg_dir + pgd_index(vaddr);
285 pud = pud_offset(pgd, vaddr);
286 pmd = pmd_offset(pud, vaddr);
287 pte = pte_offset_kernel(pmd, vaddr);
288 pkmap_page_table = pte;
291 static void __meminit free_new_highpage(struct page *page)
293 init_page_count(page);
298 void __init add_one_highpage_init(struct page *page, int pfn, int bad_ppro)
300 if (page_is_ram(pfn) && !(bad_ppro && page_kills_ppro(pfn))) {
301 ClearPageReserved(page);
302 free_new_highpage(page);
304 SetPageReserved(page);
307 static int __meminit add_one_highpage_hotplug(struct page *page, unsigned long pfn)
309 free_new_highpage(page);
311 #ifdef CONFIG_FLATMEM
312 max_mapnr = max(pfn, max_mapnr);
319 * Not currently handling the NUMA case.
320 * Assuming single node and all memory that
321 * has been added dynamically that would be
322 * onlined here is in HIGHMEM
324 void __meminit online_page(struct page *page)
326 ClearPageReserved(page);
327 add_one_highpage_hotplug(page, page_to_pfn(page));
332 extern void set_highmem_pages_init(int);
334 static void __init set_highmem_pages_init(int bad_ppro)
337 for (pfn = highstart_pfn; pfn < highend_pfn; pfn++)
338 add_one_highpage_init(pfn_to_page(pfn), pfn, bad_ppro);
339 totalram_pages += totalhigh_pages;
341 #endif /* CONFIG_FLATMEM */
344 #define kmap_init() do { } while (0)
345 #define permanent_kmaps_init(pgd_base) do { } while (0)
346 #define set_highmem_pages_init(bad_ppro) do { } while (0)
347 #endif /* CONFIG_HIGHMEM */
349 unsigned long long __PAGE_KERNEL = _PAGE_KERNEL;
350 EXPORT_SYMBOL(__PAGE_KERNEL);
351 unsigned long long __PAGE_KERNEL_EXEC = _PAGE_KERNEL_EXEC;
354 extern void __init remap_numa_kva(void);
356 #define remap_numa_kva() do {} while (0)
359 static void __init pagetable_init (void)
362 pgd_t *pgd_base = swapper_pg_dir;
364 #ifdef CONFIG_X86_PAE
366 /* Init entries of the first-level page table to the zero page */
367 for (i = 0; i < PTRS_PER_PGD; i++)
368 set_pgd(pgd_base + i, __pgd(__pa(empty_zero_page) | _PAGE_PRESENT));
371 /* Enable PSE if available */
373 set_in_cr4(X86_CR4_PSE);
376 /* Enable PGE if available */
378 set_in_cr4(X86_CR4_PGE);
379 __PAGE_KERNEL |= _PAGE_GLOBAL;
380 __PAGE_KERNEL_EXEC |= _PAGE_GLOBAL;
383 kernel_physical_mapping_init(pgd_base);
387 * Fixed mappings, only the page table structure has to be
388 * created - mappings will be set by set_fixmap():
390 vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
391 page_table_range_init(vaddr, 0, pgd_base);
393 permanent_kmaps_init(pgd_base);
395 #ifdef CONFIG_X86_PAE
397 * Add low memory identity-mappings - SMP needs it when
398 * starting up on an AP from real-mode. In the non-PAE
399 * case we already have these mappings through head.S.
400 * All user-space mappings are explicitly cleared after
403 set_pgd(&pgd_base[0], pgd_base[USER_PTRS_PER_PGD]);
407 #if defined(CONFIG_SOFTWARE_SUSPEND) || defined(CONFIG_ACPI_SLEEP)
409 * Swap suspend & friends need this for resume because things like the intel-agp
410 * driver might have split up a kernel 4MB mapping.
412 char __nosavedata swsusp_pg_dir[PAGE_SIZE]
413 __attribute__ ((aligned (PAGE_SIZE)));
415 static inline void save_pg_dir(void)
417 memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
420 static inline void save_pg_dir(void)
425 void zap_low_mappings (void)
432 * Zap initial low-memory mappings.
434 * Note that "pgd_clear()" doesn't do it for
435 * us, because pgd_clear() is a no-op on i386.
437 for (i = 0; i < USER_PTRS_PER_PGD; i++)
438 #ifdef CONFIG_X86_PAE
439 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
441 set_pgd(swapper_pg_dir+i, __pgd(0));
446 static int disable_nx __initdata = 0;
447 u64 __supported_pte_mask __read_mostly = ~_PAGE_NX;
452 * Control non executable mappings.
455 * off Disable (disables exec-shield too)
457 static int __init noexec_setup(char *str)
459 if (!str || !strcmp(str, "on")) {
461 __supported_pte_mask |= _PAGE_NX;
464 } else if (!strcmp(str,"off")) {
466 __supported_pte_mask &= ~_PAGE_NX;
473 early_param("noexec", noexec_setup);
476 #ifdef CONFIG_X86_PAE
478 static void __init set_nx(void)
480 unsigned int v[4], l, h;
482 if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) {
483 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]);
484 if ((v[3] & (1 << 20)) && !disable_nx) {
485 rdmsr(MSR_EFER, l, h);
487 wrmsr(MSR_EFER, l, h);
489 __supported_pte_mask |= _PAGE_NX;
495 * Enables/disables executability of a given kernel page and
496 * returns the previous setting.
498 int __init set_kernel_exec(unsigned long vaddr, int enable)
506 pte = lookup_address(vaddr);
509 if (!pte_exec_kernel(*pte))
513 pte->pte_high &= ~(1 << (_PAGE_BIT_NX - 32));
515 pte->pte_high |= 1 << (_PAGE_BIT_NX - 32);
516 pte_update_defer(&init_mm, vaddr, pte);
525 * paging_init() sets up the page tables - note that the first 8MB are
526 * already mapped by head.S.
528 * This routines also unmaps the page at virtual kernel address 0, so
529 * that we can trap those pesky NULL-reference errors in the kernel.
531 void __init paging_init(void)
533 #ifdef CONFIG_X86_PAE
536 printk("NX (Execute Disable) protection: active\n");
540 printk("Using x86 segment limits to approximate NX protection\n");
544 load_cr3(swapper_pg_dir);
546 #ifdef CONFIG_X86_PAE
548 * We will bail out later - printk doesn't work right now so
549 * the user would just see a hanging kernel.
552 set_in_cr4(X86_CR4_PAE);
560 * Test if the WP bit works in supervisor mode. It isn't supported on 386's
561 * and also on some strange 486's (NexGen etc.). All 586+'s are OK. This
562 * used to involve black magic jumps to work around some nasty CPU bugs,
563 * but fortunately the switch to using exceptions got rid of all that.
566 static void __init test_wp_bit(void)
568 printk("Checking if this processor honours the WP bit even in supervisor mode... ");
570 /* Any page-aligned address will do, the test is non-destructive */
571 __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
572 boot_cpu_data.wp_works_ok = do_test_wp_bit();
573 clear_fixmap(FIX_WP_TEST);
575 if (!boot_cpu_data.wp_works_ok) {
577 #ifdef CONFIG_X86_WP_WORKS_OK
578 panic("This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
585 static struct kcore_list kcore_mem, kcore_vmalloc;
587 void __init mem_init(void)
589 extern int ppro_with_ram_bug(void);
590 int codesize, reservedpages, datasize, initsize;
594 #ifdef CONFIG_FLATMEM
598 bad_ppro = ppro_with_ram_bug();
600 #ifdef CONFIG_HIGHMEM
601 /* check that fixmap and pkmap do not overlap */
602 if (PKMAP_BASE+LAST_PKMAP*PAGE_SIZE >= FIXADDR_START) {
603 printk(KERN_ERR "fixmap and kmap areas overlap - this will crash\n");
604 printk(KERN_ERR "pkstart: %lxh pkend: %lxh fixstart %lxh\n",
605 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE, FIXADDR_START);
610 /* this will put all low memory onto the freelists */
611 totalram_pages += free_all_bootmem();
614 for (tmp = 0; tmp < max_low_pfn; tmp++)
616 * Only count reserved RAM pages
618 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
621 set_highmem_pages_init(bad_ppro);
623 codesize = (unsigned long) &_etext - (unsigned long) &_text;
624 datasize = (unsigned long) &_edata - (unsigned long) &_etext;
625 initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin;
627 kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
628 kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
629 VMALLOC_END-VMALLOC_START);
631 printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, %dk reserved, %dk data, %dk init, %ldk highmem)\n",
632 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
633 num_physpages << (PAGE_SHIFT-10),
635 reservedpages << (PAGE_SHIFT-10),
638 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10))
641 #if 1 /* double-sanity-check paranoia */
642 printk("virtual kernel memory layout:\n"
643 " fixmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
644 #ifdef CONFIG_HIGHMEM
645 " pkmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
647 " vmalloc : 0x%08lx - 0x%08lx (%4ld MB)\n"
648 " lowmem : 0x%08lx - 0x%08lx (%4ld MB)\n"
649 " .init : 0x%08lx - 0x%08lx (%4ld kB)\n"
650 " .data : 0x%08lx - 0x%08lx (%4ld kB)\n"
651 " .text : 0x%08lx - 0x%08lx (%4ld kB)\n",
652 FIXADDR_START, FIXADDR_TOP,
653 (FIXADDR_TOP - FIXADDR_START) >> 10,
655 #ifdef CONFIG_HIGHMEM
656 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
657 (LAST_PKMAP*PAGE_SIZE) >> 10,
660 VMALLOC_START, VMALLOC_END,
661 (VMALLOC_END - VMALLOC_START) >> 20,
663 (unsigned long)__va(0), (unsigned long)high_memory,
664 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
666 (unsigned long)&__init_begin, (unsigned long)&__init_end,
667 ((unsigned long)&__init_end - (unsigned long)&__init_begin) >> 10,
669 (unsigned long)&_etext, (unsigned long)&_edata,
670 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
672 (unsigned long)&_text, (unsigned long)&_etext,
673 ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
675 #ifdef CONFIG_HIGHMEM
676 BUG_ON(PKMAP_BASE+LAST_PKMAP*PAGE_SIZE > FIXADDR_START);
677 BUG_ON(VMALLOC_END > PKMAP_BASE);
679 BUG_ON(VMALLOC_START > VMALLOC_END);
680 BUG_ON((unsigned long)high_memory > VMALLOC_START);
681 #endif /* double-sanity-check paranoia */
683 #ifdef CONFIG_X86_PAE
685 panic("cannot execute a PAE-enabled kernel on a PAE-less CPU!");
687 if (boot_cpu_data.wp_works_ok < 0)
691 * Subtle. SMP is doing it's boot stuff late (because it has to
692 * fork idle threads) - but it also needs low mappings for the
693 * protected-mode entry to work. We zap these entries only after
694 * the WP-bit has been tested.
701 #ifdef CONFIG_MEMORY_HOTPLUG
702 int arch_add_memory(int nid, u64 start, u64 size)
704 struct pglist_data *pgdata = NODE_DATA(nid);
705 struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
706 unsigned long start_pfn = start >> PAGE_SHIFT;
707 unsigned long nr_pages = size >> PAGE_SHIFT;
709 return __add_pages(zone, start_pfn, nr_pages);
712 int remove_memory(u64 start, u64 size)
716 EXPORT_SYMBOL_GPL(remove_memory);
719 struct kmem_cache *pgd_cache;
720 struct kmem_cache *pmd_cache;
722 void __init pgtable_cache_init(void)
724 if (PTRS_PER_PMD > 1) {
725 pmd_cache = kmem_cache_create("pmd",
726 PTRS_PER_PMD*sizeof(pmd_t),
727 PTRS_PER_PMD*sizeof(pmd_t),
732 panic("pgtable_cache_init(): cannot create pmd cache");
734 pgd_cache = kmem_cache_create("pgd",
735 PTRS_PER_PGD*sizeof(pgd_t),
736 PTRS_PER_PGD*sizeof(pgd_t),
739 PTRS_PER_PMD == 1 ? pgd_dtor : NULL);
741 panic("pgtable_cache_init(): Cannot create pgd cache");
745 * This function cannot be __init, since exceptions don't work in that
746 * section. Put this after the callers, so that it cannot be inlined.
748 static int noinline do_test_wp_bit(void)
753 __asm__ __volatile__(
758 ".section __ex_table,\"a\"\n"
762 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
771 #ifdef CONFIG_DEBUG_RODATA
773 void mark_rodata_ro(void)
775 unsigned long addr = (unsigned long)__start_rodata;
777 for (; addr < (unsigned long)__end_rodata; addr += PAGE_SIZE)
778 change_page_attr(virt_to_page(addr), 1, PAGE_KERNEL_RO);
780 printk("Write protecting the kernel read-only data: %uk\n",
781 (__end_rodata - __start_rodata) >> 10);
784 * change_page_attr() requires a global_flush_tlb() call after it.
785 * We do this after the printk so that if something went wrong in the
786 * change, the printk gets out at least to give a better debug hint
787 * of who is the culprit.
793 void free_init_pages(char *what, unsigned long begin, unsigned long end)
797 for (addr = begin; addr < end; addr += PAGE_SIZE) {
798 ClearPageReserved(virt_to_page(addr));
799 init_page_count(virt_to_page(addr));
800 memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE);
804 printk(KERN_INFO "Freeing %s: %ldk freed\n", what, (end - begin) >> 10);
807 void free_initmem(void)
809 free_init_pages("unused kernel memory",
810 (unsigned long)(&__init_begin),
811 (unsigned long)(&__init_end));
814 #ifdef CONFIG_BLK_DEV_INITRD
815 void free_initrd_mem(unsigned long start, unsigned long end)
817 free_init_pages("initrd memory", start, end);