3 * Copyright (C) 1995 Linus Torvalds
5 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
8 #include <linux/module.h>
9 #include <linux/signal.h>
10 #include <linux/sched.h>
11 #include <linux/kernel.h>
12 #include <linux/errno.h>
13 #include <linux/string.h>
14 #include <linux/types.h>
15 #include <linux/ptrace.h>
16 #include <linux/mman.h>
18 #include <linux/hugetlb.h>
19 #include <linux/swap.h>
20 #include <linux/smp.h>
21 #include <linux/init.h>
22 #include <linux/highmem.h>
23 #include <linux/pagemap.h>
24 #include <linux/pfn.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/memory_hotplug.h>
30 #include <linux/initrd.h>
31 #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>
44 #include <asm/tlbflush.h>
45 #include <asm/pgalloc.h>
46 #include <asm/sections.h>
47 #include <asm/paravirt.h>
48 #include <asm/setup.h>
49 #include <asm/cacheflush.h>
51 unsigned int __VMALLOC_RESERVE = 128 << 20;
53 unsigned long max_low_pfn_mapped;
54 unsigned long max_pfn_mapped;
56 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
57 unsigned long highstart_pfn, highend_pfn;
59 static noinline int do_test_wp_bit(void);
62 static unsigned long __initdata table_start;
63 static unsigned long __meminitdata table_end;
64 static unsigned long __meminitdata table_top;
66 static int __initdata after_init_bootmem;
68 static __init void *alloc_low_page(unsigned long *phys)
70 unsigned long pfn = table_end++;
74 panic("alloc_low_page: ran out of memory");
76 adr = __va(pfn * PAGE_SIZE);
77 memset(adr, 0, PAGE_SIZE);
78 *phys = pfn * PAGE_SIZE;
83 * Creates a middle page table and puts a pointer to it in the
84 * given global directory entry. This only returns the gd entry
85 * in non-PAE compilation mode, since the middle layer is folded.
87 static pmd_t * __init one_md_table_init(pgd_t *pgd)
94 if (!(pgd_val(*pgd) & _PAGE_PRESENT)) {
95 if (after_init_bootmem)
96 pmd_table = (pmd_t *)alloc_bootmem_low_pages(PAGE_SIZE);
98 pmd_table = (pmd_t *)alloc_low_page(&phys);
99 paravirt_alloc_pmd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT);
100 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
101 pud = pud_offset(pgd, 0);
102 BUG_ON(pmd_table != pmd_offset(pud, 0));
105 pud = pud_offset(pgd, 0);
106 pmd_table = pmd_offset(pud, 0);
112 * Create a page table and place a pointer to it in a middle page
115 static pte_t * __init one_page_table_init(pmd_t *pmd)
117 if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
118 pte_t *page_table = NULL;
120 if (after_init_bootmem) {
121 #ifdef CONFIG_DEBUG_PAGEALLOC
122 page_table = (pte_t *) alloc_bootmem_pages(PAGE_SIZE);
126 (pte_t *)alloc_bootmem_low_pages(PAGE_SIZE);
129 page_table = (pte_t *)alloc_low_page(&phys);
132 paravirt_alloc_pte(&init_mm, __pa(page_table) >> PAGE_SHIFT);
133 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
134 BUG_ON(page_table != pte_offset_kernel(pmd, 0));
137 return pte_offset_kernel(pmd, 0);
141 * This function initializes a certain range of kernel virtual memory
142 * with new bootmem page tables, everywhere page tables are missing in
145 * NOTE: The pagetables are allocated contiguous on the physical space
146 * so we can cache the place of the first one and move around without
147 * checking the pgd every time.
150 page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base)
152 int pgd_idx, pmd_idx;
158 pgd_idx = pgd_index(vaddr);
159 pmd_idx = pmd_index(vaddr);
160 pgd = pgd_base + pgd_idx;
162 for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
163 pmd = one_md_table_init(pgd);
164 pmd = pmd + pmd_index(vaddr);
165 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
167 one_page_table_init(pmd);
175 static inline int is_kernel_text(unsigned long addr)
177 if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end)
183 * This maps the physical memory to kernel virtual address space, a total
184 * of max_low_pfn pages, by creating page tables starting from address
187 static void __init kernel_physical_mapping_init(pgd_t *pgd_base,
188 unsigned long start_pfn,
189 unsigned long end_pfn,
192 int pgd_idx, pmd_idx, pte_ofs;
197 unsigned pages_2m, pages_4k;
201 * First iteration will setup identity mapping using large/small pages
202 * based on use_pse, with other attributes same as set by
203 * the early code in head_32.S
205 * Second iteration will setup the appropriate attributes (NX, GLOBAL..)
206 * as desired for the kernel identity mapping.
208 * This two pass mechanism conforms to the TLB app note which says:
210 * "Software should not write to a paging-structure entry in a way
211 * that would change, for any linear address, both the page size
212 * and either the page frame or attributes."
220 pages_2m = pages_4k = 0;
222 pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
223 pgd = pgd_base + pgd_idx;
224 for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
225 pmd = one_md_table_init(pgd);
229 #ifdef CONFIG_X86_PAE
230 pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
235 for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn;
237 unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET;
240 * Map with big pages if possible, otherwise
241 * create normal page tables:
245 pgprot_t prot = PAGE_KERNEL_LARGE;
247 * first pass will use the same initial
248 * identity mapping attribute + _PAGE_PSE.
251 __pgprot(PTE_IDENT_ATTR |
254 addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
255 PAGE_OFFSET + PAGE_SIZE-1;
257 if (is_kernel_text(addr) ||
258 is_kernel_text(addr2))
259 prot = PAGE_KERNEL_LARGE_EXEC;
262 if (mapping_iter == 1)
263 set_pmd(pmd, pfn_pmd(pfn, init_prot));
265 set_pmd(pmd, pfn_pmd(pfn, prot));
270 pte = one_page_table_init(pmd);
272 pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
274 for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn;
275 pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
276 pgprot_t prot = PAGE_KERNEL;
278 * first pass will use the same initial
279 * identity mapping attribute.
281 pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR);
283 if (is_kernel_text(addr))
284 prot = PAGE_KERNEL_EXEC;
287 if (mapping_iter == 1)
288 set_pte(pte, pfn_pte(pfn, init_prot));
290 set_pte(pte, pfn_pte(pfn, prot));
294 if (mapping_iter == 1) {
296 * update direct mapping page count only in the first
299 update_page_count(PG_LEVEL_2M, pages_2m);
300 update_page_count(PG_LEVEL_4K, pages_4k);
303 * local global flush tlb, which will flush the previous
304 * mappings present in both small and large page TLB's.
309 * Second iteration will set the actual desired PTE attributes.
317 * devmem_is_allowed() checks to see if /dev/mem access to a certain address
318 * is valid. The argument is a physical page number.
321 * On x86, access has to be given to the first megabyte of ram because that area
322 * contains bios code and data regions used by X and dosemu and similar apps.
323 * Access has to be given to non-kernel-ram areas as well, these contain the PCI
324 * mmio resources as well as potential bios/acpi data regions.
326 int devmem_is_allowed(unsigned long pagenr)
330 if (!page_is_ram(pagenr))
335 #ifdef CONFIG_HIGHMEM
339 static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr)
341 return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
342 vaddr), vaddr), vaddr);
345 static void __init kmap_init(void)
347 unsigned long kmap_vstart;
350 * Cache the first kmap pte:
352 kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
353 kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
355 kmap_prot = PAGE_KERNEL;
358 static void __init permanent_kmaps_init(pgd_t *pgd_base)
367 page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
369 pgd = swapper_pg_dir + pgd_index(vaddr);
370 pud = pud_offset(pgd, vaddr);
371 pmd = pmd_offset(pud, vaddr);
372 pte = pte_offset_kernel(pmd, vaddr);
373 pkmap_page_table = pte;
376 static void __init add_one_highpage_init(struct page *page, int pfn)
378 ClearPageReserved(page);
379 init_page_count(page);
384 struct add_highpages_data {
385 unsigned long start_pfn;
386 unsigned long end_pfn;
389 static int __init add_highpages_work_fn(unsigned long start_pfn,
390 unsigned long end_pfn, void *datax)
394 unsigned long final_start_pfn, final_end_pfn;
395 struct add_highpages_data *data;
397 data = (struct add_highpages_data *)datax;
399 final_start_pfn = max(start_pfn, data->start_pfn);
400 final_end_pfn = min(end_pfn, data->end_pfn);
401 if (final_start_pfn >= final_end_pfn)
404 for (node_pfn = final_start_pfn; node_pfn < final_end_pfn;
406 if (!pfn_valid(node_pfn))
408 page = pfn_to_page(node_pfn);
409 add_one_highpage_init(page, node_pfn);
416 void __init add_highpages_with_active_regions(int nid, unsigned long start_pfn,
417 unsigned long end_pfn)
419 struct add_highpages_data data;
421 data.start_pfn = start_pfn;
422 data.end_pfn = end_pfn;
424 work_with_active_regions(nid, add_highpages_work_fn, &data);
428 static void __init set_highmem_pages_init(void)
430 add_highpages_with_active_regions(0, highstart_pfn, highend_pfn);
432 totalram_pages += totalhigh_pages;
434 #endif /* !CONFIG_NUMA */
437 # define kmap_init() do { } while (0)
438 # define permanent_kmaps_init(pgd_base) do { } while (0)
439 # define set_highmem_pages_init() do { } while (0)
440 #endif /* CONFIG_HIGHMEM */
442 void __init native_pagetable_setup_start(pgd_t *base)
444 unsigned long pfn, va;
451 * Remove any mappings which extend past the end of physical
452 * memory from the boot time page table:
454 for (pfn = max_low_pfn + 1; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
455 va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
456 pgd = base + pgd_index(va);
457 if (!pgd_present(*pgd))
460 pud = pud_offset(pgd, va);
461 pmd = pmd_offset(pud, va);
462 if (!pmd_present(*pmd))
465 pte = pte_offset_kernel(pmd, va);
466 if (!pte_present(*pte))
469 pte_clear(NULL, va, pte);
471 paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT);
474 void __init native_pagetable_setup_done(pgd_t *base)
479 * Build a proper pagetable for the kernel mappings. Up until this
480 * point, we've been running on some set of pagetables constructed by
483 * If we're booting on native hardware, this will be a pagetable
484 * constructed in arch/x86/kernel/head_32.S. The root of the
485 * pagetable will be swapper_pg_dir.
487 * If we're booting paravirtualized under a hypervisor, then there are
488 * more options: we may already be running PAE, and the pagetable may
489 * or may not be based in swapper_pg_dir. In any case,
490 * paravirt_pagetable_setup_start() will set up swapper_pg_dir
491 * appropriately for the rest of the initialization to work.
493 * In general, pagetable_init() assumes that the pagetable may already
494 * be partially populated, and so it avoids stomping on any existing
497 static void __init early_ioremap_page_table_range_init(pgd_t *pgd_base)
499 unsigned long vaddr, end;
502 * Fixed mappings, only the page table structure has to be
503 * created - mappings will be set by set_fixmap():
505 early_ioremap_clear();
506 vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
507 end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
508 page_table_range_init(vaddr, end, pgd_base);
509 early_ioremap_reset();
512 static void __init pagetable_init(void)
514 pgd_t *pgd_base = swapper_pg_dir;
516 paravirt_pagetable_setup_start(pgd_base);
518 permanent_kmaps_init(pgd_base);
520 paravirt_pagetable_setup_done(pgd_base);
523 #ifdef CONFIG_ACPI_SLEEP
525 * ACPI suspend needs this for resume, because things like the intel-agp
526 * driver might have split up a kernel 4MB mapping.
528 char swsusp_pg_dir[PAGE_SIZE]
529 __attribute__ ((aligned(PAGE_SIZE)));
531 static inline void save_pg_dir(void)
533 memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
535 #else /* !CONFIG_ACPI_SLEEP */
536 static inline void save_pg_dir(void)
539 #endif /* !CONFIG_ACPI_SLEEP */
541 void zap_low_mappings(void)
546 * Zap initial low-memory mappings.
548 * Note that "pgd_clear()" doesn't do it for
549 * us, because pgd_clear() is a no-op on i386.
551 for (i = 0; i < KERNEL_PGD_BOUNDARY; i++) {
552 #ifdef CONFIG_X86_PAE
553 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
555 set_pgd(swapper_pg_dir+i, __pgd(0));
563 pteval_t __supported_pte_mask __read_mostly = ~(_PAGE_NX | _PAGE_GLOBAL);
564 EXPORT_SYMBOL_GPL(__supported_pte_mask);
566 #ifdef CONFIG_X86_PAE
568 static int disable_nx __initdata;
573 * Control non executable mappings.
578 static int __init noexec_setup(char *str)
580 if (!str || !strcmp(str, "on")) {
582 __supported_pte_mask |= _PAGE_NX;
586 if (!strcmp(str, "off")) {
588 __supported_pte_mask &= ~_PAGE_NX;
596 early_param("noexec", noexec_setup);
598 static void __init set_nx(void)
600 unsigned int v[4], l, h;
602 if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) {
603 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]);
605 if ((v[3] & (1 << 20)) && !disable_nx) {
606 rdmsr(MSR_EFER, l, h);
608 wrmsr(MSR_EFER, l, h);
610 __supported_pte_mask |= _PAGE_NX;
616 /* user-defined highmem size */
617 static unsigned int highmem_pages = -1;
620 * highmem=size forces highmem to be exactly 'size' bytes.
621 * This works even on boxes that have no highmem otherwise.
622 * This also works to reduce highmem size on bigger boxes.
624 static int __init parse_highmem(char *arg)
629 highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
632 early_param("highmem", parse_highmem);
635 * Determine low and high memory ranges:
637 void __init find_low_pfn_range(void)
639 /* it could update max_pfn */
641 /* max_low_pfn is 0, we already have early_res support */
643 max_low_pfn = max_pfn;
644 if (max_low_pfn > MAXMEM_PFN) {
645 if (highmem_pages == -1)
646 highmem_pages = max_pfn - MAXMEM_PFN;
647 if (highmem_pages + MAXMEM_PFN < max_pfn)
648 max_pfn = MAXMEM_PFN + highmem_pages;
649 if (highmem_pages + MAXMEM_PFN > max_pfn) {
650 printk(KERN_WARNING "only %luMB highmem pages "
651 "available, ignoring highmem size of %uMB.\n",
652 pages_to_mb(max_pfn - MAXMEM_PFN),
653 pages_to_mb(highmem_pages));
656 max_low_pfn = MAXMEM_PFN;
657 #ifndef CONFIG_HIGHMEM
658 /* Maximum memory usable is what is directly addressable */
659 printk(KERN_WARNING "Warning only %ldMB will be used.\n",
661 if (max_pfn > MAX_NONPAE_PFN)
663 "Use a HIGHMEM64G enabled kernel.\n");
665 printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
666 max_pfn = MAXMEM_PFN;
667 #else /* !CONFIG_HIGHMEM */
668 #ifndef CONFIG_HIGHMEM64G
669 if (max_pfn > MAX_NONPAE_PFN) {
670 max_pfn = MAX_NONPAE_PFN;
671 printk(KERN_WARNING "Warning only 4GB will be used."
672 "Use a HIGHMEM64G enabled kernel.\n");
674 #endif /* !CONFIG_HIGHMEM64G */
675 #endif /* !CONFIG_HIGHMEM */
677 if (highmem_pages == -1)
679 #ifdef CONFIG_HIGHMEM
680 if (highmem_pages >= max_pfn) {
681 printk(KERN_ERR "highmem size specified (%uMB) is "
682 "bigger than pages available (%luMB)!.\n",
683 pages_to_mb(highmem_pages),
684 pages_to_mb(max_pfn));
688 if (max_low_pfn - highmem_pages <
689 64*1024*1024/PAGE_SIZE){
690 printk(KERN_ERR "highmem size %uMB results in "
691 "smaller than 64MB lowmem, ignoring it.\n"
692 , pages_to_mb(highmem_pages));
695 max_low_pfn -= highmem_pages;
699 printk(KERN_ERR "ignoring highmem size on non-highmem"
705 #ifndef CONFIG_NEED_MULTIPLE_NODES
706 void __init initmem_init(unsigned long start_pfn,
707 unsigned long end_pfn)
709 #ifdef CONFIG_HIGHMEM
710 highstart_pfn = highend_pfn = max_pfn;
711 if (max_pfn > max_low_pfn)
712 highstart_pfn = max_low_pfn;
713 memory_present(0, 0, highend_pfn);
714 e820_register_active_regions(0, 0, highend_pfn);
715 printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
716 pages_to_mb(highend_pfn - highstart_pfn));
717 num_physpages = highend_pfn;
718 high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
720 memory_present(0, 0, max_low_pfn);
721 e820_register_active_regions(0, 0, max_low_pfn);
722 num_physpages = max_low_pfn;
723 high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
725 #ifdef CONFIG_FLATMEM
726 max_mapnr = num_physpages;
728 printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
729 pages_to_mb(max_low_pfn));
731 setup_bootmem_allocator();
733 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
735 static void __init zone_sizes_init(void)
737 unsigned long max_zone_pfns[MAX_NR_ZONES];
738 memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
739 max_zone_pfns[ZONE_DMA] =
740 virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
741 max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
742 #ifdef CONFIG_HIGHMEM
743 max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
746 free_area_init_nodes(max_zone_pfns);
749 void __init setup_bootmem_allocator(void)
752 unsigned long bootmap_size, bootmap;
754 * Initialize the boot-time allocator (with low memory only):
756 bootmap_size = bootmem_bootmap_pages(max_low_pfn)<<PAGE_SHIFT;
757 bootmap = find_e820_area(min_low_pfn<<PAGE_SHIFT,
758 max_pfn_mapped<<PAGE_SHIFT, bootmap_size,
761 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
762 reserve_early(bootmap, bootmap + bootmap_size, "BOOTMAP");
764 /* don't touch min_low_pfn */
765 bootmap_size = init_bootmem_node(NODE_DATA(0), bootmap >> PAGE_SHIFT,
766 min_low_pfn, max_low_pfn);
767 printk(KERN_INFO " mapped low ram: 0 - %08lx\n",
768 max_pfn_mapped<<PAGE_SHIFT);
769 printk(KERN_INFO " low ram: %08lx - %08lx\n",
770 min_low_pfn<<PAGE_SHIFT, max_low_pfn<<PAGE_SHIFT);
771 printk(KERN_INFO " bootmap %08lx - %08lx\n",
772 bootmap, bootmap + bootmap_size);
773 for_each_online_node(i)
774 free_bootmem_with_active_regions(i, max_low_pfn);
775 early_res_to_bootmem(0, max_low_pfn<<PAGE_SHIFT);
777 after_init_bootmem = 1;
780 static void __init find_early_table_space(unsigned long end, int use_pse)
782 unsigned long puds, pmds, ptes, tables, start;
784 puds = (end + PUD_SIZE - 1) >> PUD_SHIFT;
785 tables = PAGE_ALIGN(puds * sizeof(pud_t));
787 pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT;
788 tables += PAGE_ALIGN(pmds * sizeof(pmd_t));
793 extra = end - ((end>>PMD_SHIFT) << PMD_SHIFT);
795 ptes = (extra + PAGE_SIZE - 1) >> PAGE_SHIFT;
797 ptes = (end + PAGE_SIZE - 1) >> PAGE_SHIFT;
799 tables += PAGE_ALIGN(ptes * sizeof(pte_t));
802 tables += PAGE_SIZE * 2;
805 * RED-PEN putting page tables only on node 0 could
806 * cause a hotspot and fill up ZONE_DMA. The page tables
807 * need roughly 0.5KB per GB.
810 table_start = find_e820_area(start, max_pfn_mapped<<PAGE_SHIFT,
812 if (table_start == -1UL)
813 panic("Cannot find space for the kernel page tables");
815 table_start >>= PAGE_SHIFT;
816 table_end = table_start;
817 table_top = table_start + (tables>>PAGE_SHIFT);
819 printk(KERN_DEBUG "kernel direct mapping tables up to %lx @ %lx-%lx\n",
820 end, table_start << PAGE_SHIFT,
821 (table_start << PAGE_SHIFT) + tables);
824 unsigned long __init_refok init_memory_mapping(unsigned long start,
827 pgd_t *pgd_base = swapper_pg_dir;
828 unsigned long start_pfn, end_pfn;
829 unsigned long big_page_start;
830 #ifdef CONFIG_DEBUG_PAGEALLOC
832 * For CONFIG_DEBUG_PAGEALLOC, identity mapping will use small pages.
833 * This will simplify cpa(), which otherwise needs to support splitting
834 * large pages into small in interrupt context, etc.
838 int use_pse = cpu_has_pse;
842 * Find space for the kernel direct mapping tables.
844 if (!after_init_bootmem)
845 find_early_table_space(end, use_pse);
847 #ifdef CONFIG_X86_PAE
850 printk(KERN_INFO "NX (Execute Disable) protection: active\n");
853 /* Enable PSE if available */
855 set_in_cr4(X86_CR4_PSE);
857 /* Enable PGE if available */
859 set_in_cr4(X86_CR4_PGE);
860 __supported_pte_mask |= _PAGE_GLOBAL;
864 * Don't use a large page for the first 2/4MB of memory
865 * because there are often fixed size MTRRs in there
866 * and overlapping MTRRs into large pages can cause
869 big_page_start = PMD_SIZE;
871 if (start < big_page_start) {
872 start_pfn = start >> PAGE_SHIFT;
873 end_pfn = min(big_page_start>>PAGE_SHIFT, end>>PAGE_SHIFT);
875 /* head is not big page alignment ? */
876 start_pfn = start >> PAGE_SHIFT;
877 end_pfn = ((start + (PMD_SIZE - 1))>>PMD_SHIFT)
878 << (PMD_SHIFT - PAGE_SHIFT);
880 if (start_pfn < end_pfn)
881 kernel_physical_mapping_init(pgd_base, start_pfn, end_pfn, 0);
884 start_pfn = ((start + (PMD_SIZE - 1))>>PMD_SHIFT)
885 << (PMD_SHIFT - PAGE_SHIFT);
886 if (start_pfn < (big_page_start >> PAGE_SHIFT))
887 start_pfn = big_page_start >> PAGE_SHIFT;
888 end_pfn = (end>>PMD_SHIFT) << (PMD_SHIFT - PAGE_SHIFT);
889 if (start_pfn < end_pfn)
890 kernel_physical_mapping_init(pgd_base, start_pfn, end_pfn,
893 /* tail is not big page alignment ? */
895 if (start_pfn > (big_page_start>>PAGE_SHIFT)) {
896 end_pfn = end >> PAGE_SHIFT;
897 if (start_pfn < end_pfn)
898 kernel_physical_mapping_init(pgd_base, start_pfn,
902 early_ioremap_page_table_range_init(pgd_base);
904 load_cr3(swapper_pg_dir);
908 if (!after_init_bootmem)
909 reserve_early(table_start << PAGE_SHIFT,
910 table_end << PAGE_SHIFT, "PGTABLE");
912 if (!after_init_bootmem)
913 early_memtest(start, end);
915 return end >> PAGE_SHIFT;
920 * paging_init() sets up the page tables - note that the first 8MB are
921 * already mapped by head.S.
923 * This routines also unmaps the page at virtual kernel address 0, so
924 * that we can trap those pesky NULL-reference errors in the kernel.
926 void __init paging_init(void)
935 * NOTE: at this point the bootmem allocator is fully available.
942 * Test if the WP bit works in supervisor mode. It isn't supported on 386's
943 * and also on some strange 486's. All 586+'s are OK. This used to involve
944 * black magic jumps to work around some nasty CPU bugs, but fortunately the
945 * switch to using exceptions got rid of all that.
947 static void __init test_wp_bit(void)
950 "Checking if this processor honours the WP bit even in supervisor mode...");
952 /* Any page-aligned address will do, the test is non-destructive */
953 __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
954 boot_cpu_data.wp_works_ok = do_test_wp_bit();
955 clear_fixmap(FIX_WP_TEST);
957 if (!boot_cpu_data.wp_works_ok) {
958 printk(KERN_CONT "No.\n");
959 #ifdef CONFIG_X86_WP_WORKS_OK
961 "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
964 printk(KERN_CONT "Ok.\n");
968 static struct kcore_list kcore_mem, kcore_vmalloc;
970 void __init mem_init(void)
972 int codesize, reservedpages, datasize, initsize;
975 #ifdef CONFIG_FLATMEM
978 /* this will put all low memory onto the freelists */
979 totalram_pages += free_all_bootmem();
982 for (tmp = 0; tmp < max_low_pfn; tmp++)
984 * Only count reserved RAM pages:
986 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
989 set_highmem_pages_init();
991 codesize = (unsigned long) &_etext - (unsigned long) &_text;
992 datasize = (unsigned long) &_edata - (unsigned long) &_etext;
993 initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin;
995 kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
996 kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
997 VMALLOC_END-VMALLOC_START);
999 printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, "
1000 "%dk reserved, %dk data, %dk init, %ldk highmem)\n",
1001 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
1002 num_physpages << (PAGE_SHIFT-10),
1004 reservedpages << (PAGE_SHIFT-10),
1007 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10))
1010 printk(KERN_INFO "virtual kernel memory layout:\n"
1011 " fixmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
1012 #ifdef CONFIG_HIGHMEM
1013 " pkmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
1015 " vmalloc : 0x%08lx - 0x%08lx (%4ld MB)\n"
1016 " lowmem : 0x%08lx - 0x%08lx (%4ld MB)\n"
1017 " .init : 0x%08lx - 0x%08lx (%4ld kB)\n"
1018 " .data : 0x%08lx - 0x%08lx (%4ld kB)\n"
1019 " .text : 0x%08lx - 0x%08lx (%4ld kB)\n",
1020 FIXADDR_START, FIXADDR_TOP,
1021 (FIXADDR_TOP - FIXADDR_START) >> 10,
1023 #ifdef CONFIG_HIGHMEM
1024 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
1025 (LAST_PKMAP*PAGE_SIZE) >> 10,
1028 VMALLOC_START, VMALLOC_END,
1029 (VMALLOC_END - VMALLOC_START) >> 20,
1031 (unsigned long)__va(0), (unsigned long)high_memory,
1032 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
1034 (unsigned long)&__init_begin, (unsigned long)&__init_end,
1035 ((unsigned long)&__init_end -
1036 (unsigned long)&__init_begin) >> 10,
1038 (unsigned long)&_etext, (unsigned long)&_edata,
1039 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
1041 (unsigned long)&_text, (unsigned long)&_etext,
1042 ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
1044 #ifdef CONFIG_HIGHMEM
1045 BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START);
1046 BUG_ON(VMALLOC_END > PKMAP_BASE);
1048 BUG_ON(VMALLOC_START > VMALLOC_END);
1049 BUG_ON((unsigned long)high_memory > VMALLOC_START);
1051 if (boot_cpu_data.wp_works_ok < 0)
1058 #ifdef CONFIG_MEMORY_HOTPLUG
1059 int arch_add_memory(int nid, u64 start, u64 size)
1061 struct pglist_data *pgdata = NODE_DATA(nid);
1062 struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
1063 unsigned long start_pfn = start >> PAGE_SHIFT;
1064 unsigned long nr_pages = size >> PAGE_SHIFT;
1066 return __add_pages(zone, start_pfn, nr_pages);
1071 * This function cannot be __init, since exceptions don't work in that
1072 * section. Put this after the callers, so that it cannot be inlined.
1074 static noinline int do_test_wp_bit(void)
1079 __asm__ __volatile__(
1085 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
1094 #ifdef CONFIG_DEBUG_RODATA
1095 const int rodata_test_data = 0xC3;
1096 EXPORT_SYMBOL_GPL(rodata_test_data);
1098 void mark_rodata_ro(void)
1100 unsigned long start = PFN_ALIGN(_text);
1101 unsigned long size = PFN_ALIGN(_etext) - start;
1103 #ifndef CONFIG_DYNAMIC_FTRACE
1104 /* Dynamic tracing modifies the kernel text section */
1105 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1106 printk(KERN_INFO "Write protecting the kernel text: %luk\n",
1109 #ifdef CONFIG_CPA_DEBUG
1110 printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n",
1112 set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT);
1114 printk(KERN_INFO "Testing CPA: write protecting again\n");
1115 set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT);
1117 #endif /* CONFIG_DYNAMIC_FTRACE */
1120 size = (unsigned long)__end_rodata - start;
1121 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1122 printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
1126 #ifdef CONFIG_CPA_DEBUG
1127 printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size);
1128 set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
1130 printk(KERN_INFO "Testing CPA: write protecting again\n");
1131 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1136 void free_init_pages(char *what, unsigned long begin, unsigned long end)
1138 #ifdef CONFIG_DEBUG_PAGEALLOC
1140 * If debugging page accesses then do not free this memory but
1141 * mark them not present - any buggy init-section access will
1142 * create a kernel page fault:
1144 printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n",
1145 begin, PAGE_ALIGN(end));
1146 set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
1151 * We just marked the kernel text read only above, now that
1152 * we are going to free part of that, we need to make that
1155 set_memory_rw(begin, (end - begin) >> PAGE_SHIFT);
1157 for (addr = begin; addr < end; addr += PAGE_SIZE) {
1158 ClearPageReserved(virt_to_page(addr));
1159 init_page_count(virt_to_page(addr));
1160 memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE);
1164 printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
1168 void free_initmem(void)
1170 free_init_pages("unused kernel memory",
1171 (unsigned long)(&__init_begin),
1172 (unsigned long)(&__init_end));
1175 #ifdef CONFIG_BLK_DEV_INITRD
1176 void free_initrd_mem(unsigned long start, unsigned long end)
1178 free_init_pages("initrd memory", start, end);
1182 int __init reserve_bootmem_generic(unsigned long phys, unsigned long len,
1185 return reserve_bootmem(phys, len, flags);