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)
61 pmd_table = (pmd_t *) alloc_bootmem_low_pages(PAGE_SIZE);
62 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
63 pud = pud_offset(pgd, 0);
64 if (pmd_table != pmd_offset(pud, 0))
67 pud = pud_offset(pgd, 0);
68 pmd_table = pmd_offset(pud, 0);
75 * Create a page table and place a pointer to it in a middle page
78 static pte_t * __init one_page_table_init(pmd_t *pmd)
81 pte_t *page_table = (pte_t *) alloc_bootmem_low_pages(PAGE_SIZE);
82 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
83 if (page_table != pte_offset_kernel(pmd, 0))
89 return pte_offset_kernel(pmd, 0);
93 * This function initializes a certain range of kernel virtual memory
94 * with new bootmem page tables, everywhere page tables are missing in
99 * NOTE: The pagetables are allocated contiguous on the physical space
100 * so we can cache the place of the first one and move around without
101 * checking the pgd every time.
103 static void __init page_table_range_init (unsigned long start, unsigned long end, pgd_t *pgd_base)
108 int pgd_idx, pmd_idx;
112 pgd_idx = pgd_index(vaddr);
113 pmd_idx = pmd_index(vaddr);
114 pgd = pgd_base + pgd_idx;
116 for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
118 one_md_table_init(pgd);
119 pud = pud_offset(pgd, vaddr);
120 pmd = pmd_offset(pud, vaddr);
121 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end); pmd++, pmd_idx++) {
123 one_page_table_init(pmd);
131 static inline int is_kernel_text(unsigned long addr)
133 if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end)
139 * This maps the physical memory to kernel virtual address space, a total
140 * of max_low_pfn pages, by creating page tables starting from address
143 static void __init kernel_physical_mapping_init(pgd_t *pgd_base)
149 int pgd_idx, pmd_idx, pte_ofs;
151 pgd_idx = pgd_index(PAGE_OFFSET);
152 pgd = pgd_base + pgd_idx;
155 for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
156 pmd = one_md_table_init(pgd);
157 if (pfn >= max_low_pfn)
159 for (pmd_idx = 0; pmd_idx < PTRS_PER_PMD && pfn < max_low_pfn; pmd++, pmd_idx++) {
160 unsigned int address = pfn * PAGE_SIZE + PAGE_OFFSET;
162 /* Map with big pages if possible, otherwise create normal page tables. */
164 unsigned int address2 = (pfn + PTRS_PER_PTE - 1) * PAGE_SIZE + PAGE_OFFSET + PAGE_SIZE-1;
166 if (is_kernel_text(address) || is_kernel_text(address2))
167 set_pmd(pmd, pfn_pmd(pfn, PAGE_KERNEL_LARGE_EXEC));
169 set_pmd(pmd, pfn_pmd(pfn, PAGE_KERNEL_LARGE));
172 pte = one_page_table_init(pmd);
174 for (pte_ofs = 0; pte_ofs < PTRS_PER_PTE && pfn < max_low_pfn; pte++, pfn++, pte_ofs++) {
175 if (is_kernel_text(address))
176 set_pte(pte, pfn_pte(pfn, PAGE_KERNEL_EXEC));
178 set_pte(pte, pfn_pte(pfn, PAGE_KERNEL));
185 static inline int page_kills_ppro(unsigned long pagenr)
187 if (pagenr >= 0x70000 && pagenr <= 0x7003F)
192 extern int is_available_memory(efi_memory_desc_t *);
194 static inline int page_is_ram(unsigned long pagenr)
197 unsigned long addr, end;
200 efi_memory_desc_t *md;
202 for (i = 0; i < memmap.nr_map; i++) {
204 if (!is_available_memory(md))
206 addr = (md->phys_addr+PAGE_SIZE-1) >> PAGE_SHIFT;
207 end = (md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT)) >> PAGE_SHIFT;
209 if ((pagenr >= addr) && (pagenr < end))
215 for (i = 0; i < e820.nr_map; i++) {
217 if (e820.map[i].type != E820_RAM) /* not usable memory */
220 * !!!FIXME!!! Some BIOSen report areas as RAM that
221 * are not. Notably the 640->1Mb area. We need a sanity
224 addr = (e820.map[i].addr+PAGE_SIZE-1) >> PAGE_SHIFT;
225 end = (e820.map[i].addr+e820.map[i].size) >> PAGE_SHIFT;
226 if ((pagenr >= addr) && (pagenr < end))
232 unsigned long next_ram_page(unsigned long pagenr)
235 unsigned long addr, end;
236 unsigned long min_pageno = ULONG_MAX;
241 efi_memory_desc_t *md;
243 for (i = 0; i < memmap.nr_map; i++) {
245 if (!is_available_memory(md))
247 addr = (md->phys_addr+PAGE_SIZE-1) >> PAGE_SHIFT;
248 end = (md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT)) >> PAGE_SHIFT;
250 if ((pagenr >= addr) && (pagenr < end))
252 if ((pagenr < addr) && (addr < min_pageno))
258 for (i = 0; i < e820.nr_map; i++) {
260 if (e820.map[i].type != E820_RAM) /* not usable memory */
263 * !!!FIXME!!! Some BIOSen report areas as RAM that
264 * are not. Notably the 640->1Mb area. We need a sanity
267 addr = (e820.map[i].addr+PAGE_SIZE-1) >> PAGE_SHIFT;
268 end = (e820.map[i].addr+e820.map[i].size) >> PAGE_SHIFT;
269 if ((pagenr >= addr) && (pagenr < end))
271 if ((pagenr < addr) && (addr < min_pageno))
277 EXPORT_SYMBOL_GPL(next_ram_page);
280 * devmem_is_allowed() checks to see if /dev/mem access to a certain address is
281 * valid. The argument is a physical page number.
284 * On x86, access has to be given to the first megabyte of ram because that area
285 * contains bios code and data regions used by X and dosemu and similar apps.
286 * Access has to be given to non-kernel-ram areas as well, these contain the PCI
287 * mmio resources as well as potential bios/acpi data regions.
289 int devmem_is_allowed(unsigned long pagenr)
293 if (!page_is_ram(pagenr))
298 EXPORT_SYMBOL_GPL(page_is_ram);
300 #ifdef CONFIG_HIGHMEM
304 #define kmap_get_fixmap_pte(vaddr) \
305 pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr), vaddr), (vaddr)), (vaddr))
307 static void __init kmap_init(void)
309 unsigned long kmap_vstart;
311 /* cache the first kmap pte */
312 kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
313 kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
315 kmap_prot = PAGE_KERNEL;
318 static void __init permanent_kmaps_init(pgd_t *pgd_base)
327 page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
329 pgd = swapper_pg_dir + pgd_index(vaddr);
330 pud = pud_offset(pgd, vaddr);
331 pmd = pmd_offset(pud, vaddr);
332 pte = pte_offset_kernel(pmd, vaddr);
333 pkmap_page_table = pte;
336 void __init one_highpage_init(struct page *page, int pfn, int bad_ppro)
338 if (page_is_ram(pfn) && !(bad_ppro && page_kills_ppro(pfn))) {
339 ClearPageReserved(page);
340 set_bit(PG_highmem, &page->flags);
341 set_page_count(page, 1);
345 SetPageReserved(page);
348 #ifndef CONFIG_DISCONTIGMEM
349 static void __init set_highmem_pages_init(int bad_ppro)
352 for (pfn = highstart_pfn; pfn < highend_pfn; pfn++)
353 one_highpage_init(pfn_to_page(pfn), pfn, bad_ppro);
354 totalram_pages += totalhigh_pages;
357 extern void set_highmem_pages_init(int);
358 #endif /* !CONFIG_DISCONTIGMEM */
361 #define kmap_init() do { } while (0)
362 #define permanent_kmaps_init(pgd_base) do { } while (0)
363 #define set_highmem_pages_init(bad_ppro) do { } while (0)
364 #endif /* CONFIG_HIGHMEM */
366 unsigned long long __PAGE_KERNEL = _PAGE_KERNEL;
367 unsigned long long __PAGE_KERNEL_EXEC = _PAGE_KERNEL_EXEC;
369 #ifndef CONFIG_DISCONTIGMEM
370 #define remap_numa_kva() do {} while (0)
372 extern void __init remap_numa_kva(void);
375 static void __init pagetable_init (void)
378 pgd_t *pgd_base = swapper_pg_dir;
380 #ifdef CONFIG_X86_PAE
382 /* Init entries of the first-level page table to the zero page */
383 for (i = 0; i < PTRS_PER_PGD; i++)
384 set_pgd(pgd_base + i, __pgd(__pa(empty_zero_page) | _PAGE_PRESENT));
387 /* Enable PSE if available */
389 set_in_cr4(X86_CR4_PSE);
392 /* Enable PGE if available */
394 set_in_cr4(X86_CR4_PGE);
395 __PAGE_KERNEL |= _PAGE_GLOBAL;
396 __PAGE_KERNEL_EXEC |= _PAGE_GLOBAL;
399 kernel_physical_mapping_init(pgd_base);
403 * Fixed mappings, only the page table structure has to be
404 * created - mappings will be set by set_fixmap():
406 vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
407 page_table_range_init(vaddr, 0, pgd_base);
409 permanent_kmaps_init(pgd_base);
411 #ifdef CONFIG_X86_PAE
413 * Add low memory identity-mappings - SMP needs it when
414 * starting up on an AP from real-mode. In the non-PAE
415 * case we already have these mappings through head.S.
416 * All user-space mappings are explicitly cleared after
419 pgd_base[0] = pgd_base[USER_PTRS_PER_PGD];
423 #if defined(CONFIG_PM_DISK) || defined(CONFIG_SOFTWARE_SUSPEND)
425 * Swap suspend & friends need this for resume because things like the intel-agp
426 * driver might have split up a kernel 4MB mapping.
428 char __nosavedata swsusp_pg_dir[PAGE_SIZE]
429 __attribute__ ((aligned (PAGE_SIZE)));
431 static inline void save_pg_dir(void)
433 memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
436 static inline void save_pg_dir(void)
441 void zap_low_mappings (void)
448 * Zap initial low-memory mappings.
450 * Note that "pgd_clear()" doesn't do it for
451 * us, because pgd_clear() is a no-op on i386.
453 for (i = 0; i < USER_PTRS_PER_PGD; i++)
454 #ifdef CONFIG_X86_PAE
455 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
457 set_pgd(swapper_pg_dir+i, __pgd(0));
462 static int disable_nx __initdata = 0;
463 u64 __supported_pte_mask = ~_PAGE_NX;
468 * Control non executable mappings.
471 * off Disable (disables exec-shield too)
473 void __init noexec_setup(const char *str)
475 if (!strncmp(str, "on",2) && cpu_has_nx) {
476 __supported_pte_mask |= _PAGE_NX;
478 } else if (!strncmp(str,"off",3)) {
480 __supported_pte_mask &= ~_PAGE_NX;
486 #ifdef CONFIG_X86_PAE
488 static void __init set_nx(void)
490 unsigned int v[4], l, h;
492 if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) {
493 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]);
494 if ((v[3] & (1 << 20)) && !disable_nx) {
495 rdmsr(MSR_EFER, l, h);
497 wrmsr(MSR_EFER, l, h);
499 __supported_pte_mask |= _PAGE_NX;
505 * Enables/disables executability of a given kernel page and
506 * returns the previous setting.
508 int __init set_kernel_exec(unsigned long vaddr, int enable)
516 pte = lookup_address(vaddr);
519 if (!pte_exec_kernel(*pte))
523 pte->pte_high &= ~(1 << (_PAGE_BIT_NX - 32));
525 pte->pte_high |= 1 << (_PAGE_BIT_NX - 32);
534 * paging_init() sets up the page tables - note that the first 8MB are
535 * already mapped by head.S.
537 * This routines also unmaps the page at virtual kernel address 0, so
538 * that we can trap those pesky NULL-reference errors in the kernel.
540 void __init paging_init(void)
542 #ifdef CONFIG_X86_PAE
545 printk("NX (Execute Disable) protection: active\n");
549 printk("Using x86 segment limits to approximate NX protection\n");
553 load_cr3(swapper_pg_dir);
555 #ifdef CONFIG_X86_PAE
557 * We will bail out later - printk doesn't work right now so
558 * the user would just see a hanging kernel.
561 set_in_cr4(X86_CR4_PAE);
569 * Test if the WP bit works in supervisor mode. It isn't supported on 386's
570 * and also on some strange 486's (NexGen etc.). All 586+'s are OK. This
571 * used to involve black magic jumps to work around some nasty CPU bugs,
572 * but fortunately the switch to using exceptions got rid of all that.
575 static void __init test_wp_bit(void)
577 printk("Checking if this processor honours the WP bit even in supervisor mode... ");
579 /* Any page-aligned address will do, the test is non-destructive */
580 __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
581 boot_cpu_data.wp_works_ok = do_test_wp_bit();
582 clear_fixmap(FIX_WP_TEST);
584 if (!boot_cpu_data.wp_works_ok) {
586 #ifdef CONFIG_X86_WP_WORKS_OK
587 panic("This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
594 static void __init set_max_mapnr_init(void)
596 #ifdef CONFIG_HIGHMEM
597 num_physpages = highend_pfn;
599 num_physpages = max_low_pfn;
601 #ifndef CONFIG_DISCONTIGMEM
602 max_mapnr = num_physpages;
606 static struct kcore_list kcore_mem, kcore_vmalloc;
608 void __init mem_init(void)
610 extern int ppro_with_ram_bug(void);
611 int codesize, reservedpages, datasize, initsize;
615 #ifndef CONFIG_DISCONTIGMEM
620 bad_ppro = ppro_with_ram_bug();
622 #ifdef CONFIG_HIGHMEM
623 /* check that fixmap and pkmap do not overlap */
624 if (PKMAP_BASE+LAST_PKMAP*PAGE_SIZE >= FIXADDR_START) {
625 printk(KERN_ERR "fixmap and kmap areas overlap - this will crash\n");
626 printk(KERN_ERR "pkstart: %lxh pkend: %lxh fixstart %lxh\n",
627 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE, FIXADDR_START);
632 set_max_mapnr_init();
634 #ifdef CONFIG_HIGHMEM
635 high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
637 high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
640 /* this will put all low memory onto the freelists */
641 totalram_pages += free_all_bootmem();
644 for (tmp = 0; tmp < max_low_pfn; tmp++)
646 * Only count reserved RAM pages
648 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
651 set_highmem_pages_init(bad_ppro);
653 codesize = (unsigned long) &_etext - (unsigned long) &_text;
654 datasize = (unsigned long) &_edata - (unsigned long) &_etext;
655 initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin;
657 kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
658 kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
659 VMALLOC_END-VMALLOC_START);
661 printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, %dk reserved, %dk data, %dk init, %ldk highmem)\n",
662 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
663 num_physpages << (PAGE_SHIFT-10),
665 reservedpages << (PAGE_SHIFT-10),
668 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10))
671 #ifdef CONFIG_X86_PAE
673 panic("cannot execute a PAE-enabled kernel on a PAE-less CPU!");
675 if (boot_cpu_data.wp_works_ok < 0)
679 * Subtle. SMP is doing it's boot stuff late (because it has to
680 * fork idle threads) - but it also needs low mappings for the
681 * protected-mode entry to work. We zap these entries only after
682 * the WP-bit has been tested.
689 kmem_cache_t *pgd_cache;
690 kmem_cache_t *pmd_cache;
692 void __init pgtable_cache_init(void)
694 if (PTRS_PER_PMD > 1) {
695 pmd_cache = kmem_cache_create("pmd",
696 PTRS_PER_PMD*sizeof(pmd_t),
697 PTRS_PER_PMD*sizeof(pmd_t),
702 panic("pgtable_cache_init(): cannot create pmd cache");
704 pgd_cache = kmem_cache_create("pgd",
705 PTRS_PER_PGD*sizeof(pgd_t),
706 PTRS_PER_PGD*sizeof(pgd_t),
709 PTRS_PER_PMD == 1 ? pgd_dtor : NULL);
711 panic("pgtable_cache_init(): Cannot create pgd cache");
715 * This function cannot be __init, since exceptions don't work in that
716 * section. Put this after the callers, so that it cannot be inlined.
718 static int noinline do_test_wp_bit(void)
723 __asm__ __volatile__(
728 ".section __ex_table,\"a\"\n"
732 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
741 void free_initmem(void)
745 addr = (unsigned long)(&__init_begin);
746 for (; addr < (unsigned long)(&__init_end); addr += PAGE_SIZE) {
747 ClearPageReserved(virt_to_page(addr));
748 set_page_count(virt_to_page(addr), 1);
749 memset((void *)addr, 0xcc, PAGE_SIZE);
753 printk (KERN_INFO "Freeing unused kernel memory: %dk freed\n", (__init_end - __init_begin) >> 10);
756 #ifdef CONFIG_BLK_DEV_INITRD
757 void free_initrd_mem(unsigned long start, unsigned long end)
760 printk (KERN_INFO "Freeing initrd memory: %ldk freed\n", (end - start) >> 10);
761 for (; start < end; start += PAGE_SIZE) {
762 ClearPageReserved(virt_to_page(start));
763 set_page_count(virt_to_page(start), 1);