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/pci.h>
25 #include <linux/pfn.h>
26 #include <linux/poison.h>
27 #include <linux/bootmem.h>
28 #include <linux/slab.h>
29 #include <linux/proc_fs.h>
30 #include <linux/memory_hotplug.h>
31 #include <linux/initrd.h>
32 #include <linux/cpumask.h>
35 #include <asm/bios_ebda.h>
36 #include <asm/processor.h>
37 #include <asm/system.h>
38 #include <asm/uaccess.h>
39 #include <asm/pgtable.h>
41 #include <asm/fixmap.h>
46 #include <asm/tlbflush.h>
47 #include <asm/pgalloc.h>
48 #include <asm/sections.h>
49 #include <asm/paravirt.h>
50 #include <asm/setup.h>
51 #include <asm/cacheflush.h>
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(void)
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);
82 * Creates a middle page table and puts a pointer to it in the
83 * given global directory entry. This only returns the gd entry
84 * in non-PAE compilation mode, since the middle layer is folded.
86 static pmd_t * __init one_md_table_init(pgd_t *pgd)
92 if (!(pgd_val(*pgd) & _PAGE_PRESENT)) {
93 if (after_init_bootmem)
94 pmd_table = (pmd_t *)alloc_bootmem_low_pages(PAGE_SIZE);
96 pmd_table = (pmd_t *)alloc_low_page();
97 paravirt_alloc_pmd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT);
98 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
99 pud = pud_offset(pgd, 0);
100 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);
128 page_table = (pte_t *)alloc_low_page();
130 paravirt_alloc_pte(&init_mm, __pa(page_table) >> PAGE_SHIFT);
131 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
132 BUG_ON(page_table != pte_offset_kernel(pmd, 0));
135 return pte_offset_kernel(pmd, 0);
138 pmd_t * __init populate_extra_pmd(unsigned long vaddr)
140 int pgd_idx = pgd_index(vaddr);
141 int pmd_idx = pmd_index(vaddr);
143 return one_md_table_init(swapper_pg_dir + pgd_idx) + pmd_idx;
146 pte_t * __init populate_extra_pte(unsigned long vaddr)
148 int pte_idx = pte_index(vaddr);
151 pmd = populate_extra_pmd(vaddr);
152 return one_page_table_init(pmd) + pte_idx;
155 static pte_t *__init page_table_kmap_check(pte_t *pte, pmd_t *pmd,
156 unsigned long vaddr, pte_t *lastpte)
158 #ifdef CONFIG_HIGHMEM
160 * Something (early fixmap) may already have put a pte
161 * page here, which causes the page table allocation
162 * to become nonlinear. Attempt to fix it, and if it
163 * is still nonlinear then we have to bug.
165 int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
166 int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
168 if (pmd_idx_kmap_begin != pmd_idx_kmap_end
169 && (vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin
170 && (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end
171 && ((__pa(pte) >> PAGE_SHIFT) < table_start
172 || (__pa(pte) >> PAGE_SHIFT) >= table_end)) {
176 BUG_ON(after_init_bootmem);
177 newpte = alloc_low_page();
178 for (i = 0; i < PTRS_PER_PTE; i++)
179 set_pte(newpte + i, pte[i]);
181 paravirt_alloc_pte(&init_mm, __pa(newpte) >> PAGE_SHIFT);
182 set_pmd(pmd, __pmd(__pa(newpte)|_PAGE_TABLE));
183 BUG_ON(newpte != pte_offset_kernel(pmd, 0));
186 paravirt_release_pte(__pa(pte) >> PAGE_SHIFT);
189 BUG_ON(vaddr < fix_to_virt(FIX_KMAP_BEGIN - 1)
190 && vaddr > fix_to_virt(FIX_KMAP_END)
191 && lastpte && lastpte + PTRS_PER_PTE != pte);
197 * This function initializes a certain range of kernel virtual memory
198 * with new bootmem page tables, everywhere page tables are missing in
201 * NOTE: The pagetables are allocated contiguous on the physical space
202 * so we can cache the place of the first one and move around without
203 * checking the pgd every time.
206 page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base)
208 int pgd_idx, pmd_idx;
215 pgd_idx = pgd_index(vaddr);
216 pmd_idx = pmd_index(vaddr);
217 pgd = pgd_base + pgd_idx;
219 for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
220 pmd = one_md_table_init(pgd);
221 pmd = pmd + pmd_index(vaddr);
222 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
224 pte = page_table_kmap_check(one_page_table_init(pmd),
233 static inline int is_kernel_text(unsigned long addr)
235 if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end)
241 * This maps the physical memory to kernel virtual address space, a total
242 * of max_low_pfn pages, by creating page tables starting from address
245 static void __init kernel_physical_mapping_init(pgd_t *pgd_base,
246 unsigned long start_pfn,
247 unsigned long end_pfn,
250 int pgd_idx, pmd_idx, pte_ofs;
255 unsigned pages_2m, pages_4k;
259 * First iteration will setup identity mapping using large/small pages
260 * based on use_pse, with other attributes same as set by
261 * the early code in head_32.S
263 * Second iteration will setup the appropriate attributes (NX, GLOBAL..)
264 * as desired for the kernel identity mapping.
266 * This two pass mechanism conforms to the TLB app note which says:
268 * "Software should not write to a paging-structure entry in a way
269 * that would change, for any linear address, both the page size
270 * and either the page frame or attributes."
278 pages_2m = pages_4k = 0;
280 pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
281 pgd = pgd_base + pgd_idx;
282 for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
283 pmd = one_md_table_init(pgd);
287 #ifdef CONFIG_X86_PAE
288 pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
293 for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn;
295 unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET;
298 * Map with big pages if possible, otherwise
299 * create normal page tables:
303 pgprot_t prot = PAGE_KERNEL_LARGE;
305 * first pass will use the same initial
306 * identity mapping attribute + _PAGE_PSE.
309 __pgprot(PTE_IDENT_ATTR |
312 addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
313 PAGE_OFFSET + PAGE_SIZE-1;
315 if (is_kernel_text(addr) ||
316 is_kernel_text(addr2))
317 prot = PAGE_KERNEL_LARGE_EXEC;
320 if (mapping_iter == 1)
321 set_pmd(pmd, pfn_pmd(pfn, init_prot));
323 set_pmd(pmd, pfn_pmd(pfn, prot));
328 pte = one_page_table_init(pmd);
330 pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
332 for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn;
333 pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
334 pgprot_t prot = PAGE_KERNEL;
336 * first pass will use the same initial
337 * identity mapping attribute.
339 pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR);
341 if (is_kernel_text(addr))
342 prot = PAGE_KERNEL_EXEC;
345 if (mapping_iter == 1)
346 set_pte(pte, pfn_pte(pfn, init_prot));
348 set_pte(pte, pfn_pte(pfn, prot));
352 if (mapping_iter == 1) {
354 * update direct mapping page count only in the first
357 update_page_count(PG_LEVEL_2M, pages_2m);
358 update_page_count(PG_LEVEL_4K, pages_4k);
361 * local global flush tlb, which will flush the previous
362 * mappings present in both small and large page TLB's.
367 * Second iteration will set the actual desired PTE attributes.
375 * devmem_is_allowed() checks to see if /dev/mem access to a certain address
376 * is valid. The argument is a physical page number.
379 * On x86, access has to be given to the first megabyte of ram because that area
380 * contains bios code and data regions used by X and dosemu and similar apps.
381 * Access has to be given to non-kernel-ram areas as well, these contain the PCI
382 * mmio resources as well as potential bios/acpi data regions.
384 int devmem_is_allowed(unsigned long pagenr)
388 if (iomem_is_exclusive(pagenr << PAGE_SHIFT))
390 if (!page_is_ram(pagenr))
398 static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr)
400 return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
401 vaddr), vaddr), vaddr);
404 static void __init kmap_init(void)
406 unsigned long kmap_vstart;
409 * Cache the first kmap pte:
411 kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
412 kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
414 kmap_prot = PAGE_KERNEL;
417 #ifdef CONFIG_HIGHMEM
418 static void __init permanent_kmaps_init(pgd_t *pgd_base)
427 page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
429 pgd = swapper_pg_dir + pgd_index(vaddr);
430 pud = pud_offset(pgd, vaddr);
431 pmd = pmd_offset(pud, vaddr);
432 pte = pte_offset_kernel(pmd, vaddr);
433 pkmap_page_table = pte;
436 static void __init add_one_highpage_init(struct page *page, int pfn)
438 ClearPageReserved(page);
439 init_page_count(page);
444 struct add_highpages_data {
445 unsigned long start_pfn;
446 unsigned long end_pfn;
449 static int __init add_highpages_work_fn(unsigned long start_pfn,
450 unsigned long end_pfn, void *datax)
454 unsigned long final_start_pfn, final_end_pfn;
455 struct add_highpages_data *data;
457 data = (struct add_highpages_data *)datax;
459 final_start_pfn = max(start_pfn, data->start_pfn);
460 final_end_pfn = min(end_pfn, data->end_pfn);
461 if (final_start_pfn >= final_end_pfn)
464 for (node_pfn = final_start_pfn; node_pfn < final_end_pfn;
466 if (!pfn_valid(node_pfn))
468 page = pfn_to_page(node_pfn);
469 add_one_highpage_init(page, node_pfn);
476 void __init add_highpages_with_active_regions(int nid, unsigned long start_pfn,
477 unsigned long end_pfn)
479 struct add_highpages_data data;
481 data.start_pfn = start_pfn;
482 data.end_pfn = end_pfn;
484 work_with_active_regions(nid, add_highpages_work_fn, &data);
488 static inline void permanent_kmaps_init(pgd_t *pgd_base)
491 #endif /* CONFIG_HIGHMEM */
493 void __init native_pagetable_setup_start(pgd_t *base)
495 unsigned long pfn, va;
502 * Remove any mappings which extend past the end of physical
503 * memory from the boot time page table:
505 for (pfn = max_low_pfn + 1; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
506 va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
507 pgd = base + pgd_index(va);
508 if (!pgd_present(*pgd))
511 pud = pud_offset(pgd, va);
512 pmd = pmd_offset(pud, va);
513 if (!pmd_present(*pmd))
516 pte = pte_offset_kernel(pmd, va);
517 if (!pte_present(*pte))
520 pte_clear(NULL, va, pte);
522 paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT);
525 void __init native_pagetable_setup_done(pgd_t *base)
530 * Build a proper pagetable for the kernel mappings. Up until this
531 * point, we've been running on some set of pagetables constructed by
534 * If we're booting on native hardware, this will be a pagetable
535 * constructed in arch/x86/kernel/head_32.S. The root of the
536 * pagetable will be swapper_pg_dir.
538 * If we're booting paravirtualized under a hypervisor, then there are
539 * more options: we may already be running PAE, and the pagetable may
540 * or may not be based in swapper_pg_dir. In any case,
541 * paravirt_pagetable_setup_start() will set up swapper_pg_dir
542 * appropriately for the rest of the initialization to work.
544 * In general, pagetable_init() assumes that the pagetable may already
545 * be partially populated, and so it avoids stomping on any existing
548 static void __init early_ioremap_page_table_range_init(pgd_t *pgd_base)
550 unsigned long vaddr, end;
553 * Fixed mappings, only the page table structure has to be
554 * created - mappings will be set by set_fixmap():
556 vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
557 end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
558 page_table_range_init(vaddr, end, pgd_base);
559 early_ioremap_reset();
562 static void __init pagetable_init(void)
564 pgd_t *pgd_base = swapper_pg_dir;
566 permanent_kmaps_init(pgd_base);
569 #ifdef CONFIG_ACPI_SLEEP
571 * ACPI suspend needs this for resume, because things like the intel-agp
572 * driver might have split up a kernel 4MB mapping.
574 char swsusp_pg_dir[PAGE_SIZE]
575 __attribute__ ((aligned(PAGE_SIZE)));
577 static inline void save_pg_dir(void)
579 memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
581 #else /* !CONFIG_ACPI_SLEEP */
582 static inline void save_pg_dir(void)
585 #endif /* !CONFIG_ACPI_SLEEP */
587 void zap_low_mappings(void)
592 * Zap initial low-memory mappings.
594 * Note that "pgd_clear()" doesn't do it for
595 * us, because pgd_clear() is a no-op on i386.
597 for (i = 0; i < KERNEL_PGD_BOUNDARY; i++) {
598 #ifdef CONFIG_X86_PAE
599 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
601 set_pgd(swapper_pg_dir+i, __pgd(0));
609 pteval_t __supported_pte_mask __read_mostly = ~(_PAGE_NX | _PAGE_GLOBAL | _PAGE_IOMAP);
610 EXPORT_SYMBOL_GPL(__supported_pte_mask);
612 #ifdef CONFIG_X86_PAE
614 static int disable_nx __initdata;
619 * Control non executable mappings.
624 static int __init noexec_setup(char *str)
626 if (!str || !strcmp(str, "on")) {
628 __supported_pte_mask |= _PAGE_NX;
632 if (!strcmp(str, "off")) {
634 __supported_pte_mask &= ~_PAGE_NX;
642 early_param("noexec", noexec_setup);
644 static void __init set_nx(void)
646 unsigned int v[4], l, h;
648 if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) {
649 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]);
651 if ((v[3] & (1 << 20)) && !disable_nx) {
652 rdmsr(MSR_EFER, l, h);
654 wrmsr(MSR_EFER, l, h);
656 __supported_pte_mask |= _PAGE_NX;
662 /* user-defined highmem size */
663 static unsigned int highmem_pages = -1;
666 * highmem=size forces highmem to be exactly 'size' bytes.
667 * This works even on boxes that have no highmem otherwise.
668 * This also works to reduce highmem size on bigger boxes.
670 static int __init parse_highmem(char *arg)
675 highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
678 early_param("highmem", parse_highmem);
680 #define MSG_HIGHMEM_TOO_BIG \
681 "highmem size (%luMB) is bigger than pages available (%luMB)!\n"
683 #define MSG_LOWMEM_TOO_SMALL \
684 "highmem size (%luMB) results in <64MB lowmem, ignoring it!\n"
686 * All of RAM fits into lowmem - but if user wants highmem
687 * artificially via the highmem=x boot parameter then create
690 void __init lowmem_pfn_init(void)
692 /* max_low_pfn is 0, we already have early_res support */
693 max_low_pfn = max_pfn;
695 if (highmem_pages == -1)
697 #ifdef CONFIG_HIGHMEM
698 if (highmem_pages >= max_pfn) {
699 printk(KERN_ERR MSG_HIGHMEM_TOO_BIG,
700 pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
704 if (max_low_pfn - highmem_pages < 64*1024*1024/PAGE_SIZE) {
705 printk(KERN_ERR MSG_LOWMEM_TOO_SMALL,
706 pages_to_mb(highmem_pages));
709 max_low_pfn -= highmem_pages;
713 printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
717 #define MSG_HIGHMEM_TOO_SMALL \
718 "only %luMB highmem pages available, ignoring highmem size of %luMB!\n"
720 #define MSG_HIGHMEM_TRIMMED \
721 "Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n"
723 * We have more RAM than fits into lowmem - we try to put it into
724 * highmem, also taking the highmem=x boot parameter into account:
726 void __init highmem_pfn_init(void)
728 max_low_pfn = MAXMEM_PFN;
730 if (highmem_pages == -1)
731 highmem_pages = max_pfn - MAXMEM_PFN;
733 if (highmem_pages + MAXMEM_PFN < max_pfn)
734 max_pfn = MAXMEM_PFN + highmem_pages;
736 if (highmem_pages + MAXMEM_PFN > max_pfn) {
737 printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL,
738 pages_to_mb(max_pfn - MAXMEM_PFN),
739 pages_to_mb(highmem_pages));
742 #ifndef CONFIG_HIGHMEM
743 /* Maximum memory usable is what is directly addressable */
744 printk(KERN_WARNING "Warning only %ldMB will be used.\n", MAXMEM>>20);
745 if (max_pfn > MAX_NONPAE_PFN)
746 printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
748 printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
749 max_pfn = MAXMEM_PFN;
750 #else /* !CONFIG_HIGHMEM */
751 #ifndef CONFIG_HIGHMEM64G
752 if (max_pfn > MAX_NONPAE_PFN) {
753 max_pfn = MAX_NONPAE_PFN;
754 printk(KERN_WARNING MSG_HIGHMEM_TRIMMED);
756 #endif /* !CONFIG_HIGHMEM64G */
757 #endif /* !CONFIG_HIGHMEM */
761 * Determine low and high memory ranges:
763 void __init find_low_pfn_range(void)
765 /* it could update max_pfn */
767 if (max_pfn <= MAXMEM_PFN)
773 #ifndef CONFIG_NEED_MULTIPLE_NODES
774 void __init initmem_init(unsigned long start_pfn,
775 unsigned long end_pfn)
777 #ifdef CONFIG_HIGHMEM
778 highstart_pfn = highend_pfn = max_pfn;
779 if (max_pfn > max_low_pfn)
780 highstart_pfn = max_low_pfn;
781 memory_present(0, 0, highend_pfn);
782 e820_register_active_regions(0, 0, highend_pfn);
783 printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
784 pages_to_mb(highend_pfn - highstart_pfn));
785 num_physpages = highend_pfn;
786 high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
788 memory_present(0, 0, max_low_pfn);
789 e820_register_active_regions(0, 0, max_low_pfn);
790 num_physpages = max_low_pfn;
791 high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
793 #ifdef CONFIG_FLATMEM
794 max_mapnr = num_physpages;
796 printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
797 pages_to_mb(max_low_pfn));
799 setup_bootmem_allocator();
801 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
803 static void __init zone_sizes_init(void)
805 unsigned long max_zone_pfns[MAX_NR_ZONES];
806 memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
807 max_zone_pfns[ZONE_DMA] =
808 virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
809 max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
810 #ifdef CONFIG_HIGHMEM
811 max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
814 free_area_init_nodes(max_zone_pfns);
817 void __init setup_bootmem_allocator(void)
820 unsigned long bootmap_size, bootmap;
822 * Initialize the boot-time allocator (with low memory only):
824 bootmap_size = bootmem_bootmap_pages(max_low_pfn)<<PAGE_SHIFT;
825 bootmap = find_e820_area(min_low_pfn<<PAGE_SHIFT,
826 max_pfn_mapped<<PAGE_SHIFT, bootmap_size,
829 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
830 reserve_early(bootmap, bootmap + bootmap_size, "BOOTMAP");
832 /* don't touch min_low_pfn */
833 bootmap_size = init_bootmem_node(NODE_DATA(0), bootmap >> PAGE_SHIFT,
834 min_low_pfn, max_low_pfn);
835 printk(KERN_INFO " mapped low ram: 0 - %08lx\n",
836 max_pfn_mapped<<PAGE_SHIFT);
837 printk(KERN_INFO " low ram: %08lx - %08lx\n",
838 min_low_pfn<<PAGE_SHIFT, max_low_pfn<<PAGE_SHIFT);
839 printk(KERN_INFO " bootmap %08lx - %08lx\n",
840 bootmap, bootmap + bootmap_size);
841 for_each_online_node(i)
842 free_bootmem_with_active_regions(i, max_low_pfn);
843 early_res_to_bootmem(0, max_low_pfn<<PAGE_SHIFT);
845 after_init_bootmem = 1;
848 static void __init find_early_table_space(unsigned long end, int use_pse)
850 unsigned long puds, pmds, ptes, tables, start;
852 puds = (end + PUD_SIZE - 1) >> PUD_SHIFT;
853 tables = roundup(puds * sizeof(pud_t), PAGE_SIZE);
855 pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT;
856 tables += roundup(pmds * sizeof(pmd_t), PAGE_SIZE);
861 extra = end - ((end>>PMD_SHIFT) << PMD_SHIFT);
863 ptes = (extra + PAGE_SIZE - 1) >> PAGE_SHIFT;
865 ptes = (end + PAGE_SIZE - 1) >> PAGE_SHIFT;
867 tables += roundup(ptes * sizeof(pte_t), PAGE_SIZE);
870 tables += roundup(__end_of_fixed_addresses * sizeof(pte_t), PAGE_SIZE);
873 * RED-PEN putting page tables only on node 0 could
874 * cause a hotspot and fill up ZONE_DMA. The page tables
875 * need roughly 0.5KB per GB.
878 table_start = find_e820_area(start, max_pfn_mapped<<PAGE_SHIFT,
880 if (table_start == -1UL)
881 panic("Cannot find space for the kernel page tables");
883 table_start >>= PAGE_SHIFT;
884 table_end = table_start;
885 table_top = table_start + (tables>>PAGE_SHIFT);
887 printk(KERN_DEBUG "kernel direct mapping tables up to %lx @ %lx-%lx\n",
888 end, table_start << PAGE_SHIFT,
889 (table_start << PAGE_SHIFT) + tables);
892 unsigned long __init_refok init_memory_mapping(unsigned long start,
895 pgd_t *pgd_base = swapper_pg_dir;
896 unsigned long start_pfn, end_pfn;
897 unsigned long big_page_start;
898 #ifdef CONFIG_DEBUG_PAGEALLOC
900 * For CONFIG_DEBUG_PAGEALLOC, identity mapping will use small pages.
901 * This will simplify cpa(), which otherwise needs to support splitting
902 * large pages into small in interrupt context, etc.
906 int use_pse = cpu_has_pse;
910 * Find space for the kernel direct mapping tables.
912 if (!after_init_bootmem)
913 find_early_table_space(end, use_pse);
915 #ifdef CONFIG_X86_PAE
918 printk(KERN_INFO "NX (Execute Disable) protection: active\n");
921 /* Enable PSE if available */
923 set_in_cr4(X86_CR4_PSE);
925 /* Enable PGE if available */
927 set_in_cr4(X86_CR4_PGE);
928 __supported_pte_mask |= _PAGE_GLOBAL;
932 * Don't use a large page for the first 2/4MB of memory
933 * because there are often fixed size MTRRs in there
934 * and overlapping MTRRs into large pages can cause
937 big_page_start = PMD_SIZE;
939 if (start < big_page_start) {
940 start_pfn = start >> PAGE_SHIFT;
941 end_pfn = min(big_page_start>>PAGE_SHIFT, end>>PAGE_SHIFT);
943 /* head is not big page alignment ? */
944 start_pfn = start >> PAGE_SHIFT;
945 end_pfn = ((start + (PMD_SIZE - 1))>>PMD_SHIFT)
946 << (PMD_SHIFT - PAGE_SHIFT);
948 if (start_pfn < end_pfn)
949 kernel_physical_mapping_init(pgd_base, start_pfn, end_pfn, 0);
952 start_pfn = ((start + (PMD_SIZE - 1))>>PMD_SHIFT)
953 << (PMD_SHIFT - PAGE_SHIFT);
954 if (start_pfn < (big_page_start >> PAGE_SHIFT))
955 start_pfn = big_page_start >> PAGE_SHIFT;
956 end_pfn = (end>>PMD_SHIFT) << (PMD_SHIFT - PAGE_SHIFT);
957 if (start_pfn < end_pfn)
958 kernel_physical_mapping_init(pgd_base, start_pfn, end_pfn,
961 /* tail is not big page alignment ? */
963 if (start_pfn > (big_page_start>>PAGE_SHIFT)) {
964 end_pfn = end >> PAGE_SHIFT;
965 if (start_pfn < end_pfn)
966 kernel_physical_mapping_init(pgd_base, start_pfn,
970 early_ioremap_page_table_range_init(pgd_base);
972 load_cr3(swapper_pg_dir);
976 if (!after_init_bootmem)
977 reserve_early(table_start << PAGE_SHIFT,
978 table_end << PAGE_SHIFT, "PGTABLE");
980 if (!after_init_bootmem)
981 early_memtest(start, end);
983 return end >> PAGE_SHIFT;
988 * paging_init() sets up the page tables - note that the first 8MB are
989 * already mapped by head.S.
991 * This routines also unmaps the page at virtual kernel address 0, so
992 * that we can trap those pesky NULL-reference errors in the kernel.
994 void __init paging_init(void)
1003 * NOTE: at this point the bootmem allocator is fully available.
1010 * Test if the WP bit works in supervisor mode. It isn't supported on 386's
1011 * and also on some strange 486's. All 586+'s are OK. This used to involve
1012 * black magic jumps to work around some nasty CPU bugs, but fortunately the
1013 * switch to using exceptions got rid of all that.
1015 static void __init test_wp_bit(void)
1018 "Checking if this processor honours the WP bit even in supervisor mode...");
1020 /* Any page-aligned address will do, the test is non-destructive */
1021 __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
1022 boot_cpu_data.wp_works_ok = do_test_wp_bit();
1023 clear_fixmap(FIX_WP_TEST);
1025 if (!boot_cpu_data.wp_works_ok) {
1026 printk(KERN_CONT "No.\n");
1027 #ifdef CONFIG_X86_WP_WORKS_OK
1029 "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
1032 printk(KERN_CONT "Ok.\n");
1036 static struct kcore_list kcore_mem, kcore_vmalloc;
1038 void __init mem_init(void)
1040 int codesize, reservedpages, datasize, initsize;
1045 #ifdef CONFIG_FLATMEM
1048 /* this will put all low memory onto the freelists */
1049 totalram_pages += free_all_bootmem();
1052 for (tmp = 0; tmp < max_low_pfn; tmp++)
1054 * Only count reserved RAM pages:
1056 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
1059 set_highmem_pages_init();
1061 codesize = (unsigned long) &_etext - (unsigned long) &_text;
1062 datasize = (unsigned long) &_edata - (unsigned long) &_etext;
1063 initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin;
1065 kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
1066 kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
1067 VMALLOC_END-VMALLOC_START);
1069 printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, "
1070 "%dk reserved, %dk data, %dk init, %ldk highmem)\n",
1071 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
1072 num_physpages << (PAGE_SHIFT-10),
1074 reservedpages << (PAGE_SHIFT-10),
1077 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10))
1080 printk(KERN_INFO "virtual kernel memory layout:\n"
1081 " fixmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
1082 #ifdef CONFIG_HIGHMEM
1083 " pkmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
1085 " vmalloc : 0x%08lx - 0x%08lx (%4ld MB)\n"
1086 " lowmem : 0x%08lx - 0x%08lx (%4ld MB)\n"
1087 " .init : 0x%08lx - 0x%08lx (%4ld kB)\n"
1088 " .data : 0x%08lx - 0x%08lx (%4ld kB)\n"
1089 " .text : 0x%08lx - 0x%08lx (%4ld kB)\n",
1090 FIXADDR_START, FIXADDR_TOP,
1091 (FIXADDR_TOP - FIXADDR_START) >> 10,
1093 #ifdef CONFIG_HIGHMEM
1094 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
1095 (LAST_PKMAP*PAGE_SIZE) >> 10,
1098 VMALLOC_START, VMALLOC_END,
1099 (VMALLOC_END - VMALLOC_START) >> 20,
1101 (unsigned long)__va(0), (unsigned long)high_memory,
1102 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
1104 (unsigned long)&__init_begin, (unsigned long)&__init_end,
1105 ((unsigned long)&__init_end -
1106 (unsigned long)&__init_begin) >> 10,
1108 (unsigned long)&_etext, (unsigned long)&_edata,
1109 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
1111 (unsigned long)&_text, (unsigned long)&_etext,
1112 ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
1115 * Check boundaries twice: Some fundamental inconsistencies can
1116 * be detected at build time already.
1118 #define __FIXADDR_TOP (-PAGE_SIZE)
1119 #ifdef CONFIG_HIGHMEM
1120 BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START);
1121 BUILD_BUG_ON(VMALLOC_END > PKMAP_BASE);
1123 #define high_memory (-128UL << 20)
1124 BUILD_BUG_ON(VMALLOC_START >= VMALLOC_END);
1126 #undef __FIXADDR_TOP
1128 #ifdef CONFIG_HIGHMEM
1129 BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START);
1130 BUG_ON(VMALLOC_END > PKMAP_BASE);
1132 BUG_ON(VMALLOC_START >= VMALLOC_END);
1133 BUG_ON((unsigned long)high_memory > VMALLOC_START);
1135 if (boot_cpu_data.wp_works_ok < 0)
1142 #ifdef CONFIG_MEMORY_HOTPLUG
1143 int arch_add_memory(int nid, u64 start, u64 size)
1145 struct pglist_data *pgdata = NODE_DATA(nid);
1146 struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
1147 unsigned long start_pfn = start >> PAGE_SHIFT;
1148 unsigned long nr_pages = size >> PAGE_SHIFT;
1150 return __add_pages(nid, zone, start_pfn, nr_pages);
1155 * This function cannot be __init, since exceptions don't work in that
1156 * section. Put this after the callers, so that it cannot be inlined.
1158 static noinline int do_test_wp_bit(void)
1163 __asm__ __volatile__(
1169 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
1178 #ifdef CONFIG_DEBUG_RODATA
1179 const int rodata_test_data = 0xC3;
1180 EXPORT_SYMBOL_GPL(rodata_test_data);
1182 void mark_rodata_ro(void)
1184 unsigned long start = PFN_ALIGN(_text);
1185 unsigned long size = PFN_ALIGN(_etext) - start;
1187 #ifndef CONFIG_DYNAMIC_FTRACE
1188 /* Dynamic tracing modifies the kernel text section */
1189 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1190 printk(KERN_INFO "Write protecting the kernel text: %luk\n",
1193 #ifdef CONFIG_CPA_DEBUG
1194 printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n",
1196 set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT);
1198 printk(KERN_INFO "Testing CPA: write protecting again\n");
1199 set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT);
1201 #endif /* CONFIG_DYNAMIC_FTRACE */
1204 size = (unsigned long)__end_rodata - start;
1205 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1206 printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
1210 #ifdef CONFIG_CPA_DEBUG
1211 printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size);
1212 set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
1214 printk(KERN_INFO "Testing CPA: write protecting again\n");
1215 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1220 #ifdef CONFIG_BLK_DEV_INITRD
1221 void free_initrd_mem(unsigned long start, unsigned long end)
1223 free_init_pages("initrd memory", start, end);
1227 int __init reserve_bootmem_generic(unsigned long phys, unsigned long len,
1230 return reserve_bootmem(phys, len, flags);