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 static pte_t *__init page_table_kmap_check(pte_t *pte, pmd_t *pmd,
139 unsigned long vaddr, pte_t *lastpte)
141 #ifdef CONFIG_HIGHMEM
143 * Something (early fixmap) may already have put a pte
144 * page here, which causes the page table allocation
145 * to become nonlinear. Attempt to fix it, and if it
146 * is still nonlinear then we have to bug.
148 int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
149 int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
151 if (pmd_idx_kmap_begin != pmd_idx_kmap_end
152 && (vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin
153 && (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end
154 && ((__pa(pte) >> PAGE_SHIFT) < table_start
155 || (__pa(pte) >> PAGE_SHIFT) >= table_end)) {
159 BUG_ON(after_init_bootmem);
160 newpte = alloc_low_page();
161 for (i = 0; i < PTRS_PER_PTE; i++)
162 set_pte(newpte + i, pte[i]);
164 paravirt_alloc_pte(&init_mm, __pa(newpte) >> PAGE_SHIFT);
165 set_pmd(pmd, __pmd(__pa(newpte)|_PAGE_TABLE));
166 BUG_ON(newpte != pte_offset_kernel(pmd, 0));
169 paravirt_release_pte(__pa(pte) >> PAGE_SHIFT);
172 BUG_ON(vaddr < fix_to_virt(FIX_KMAP_BEGIN - 1)
173 && vaddr > fix_to_virt(FIX_KMAP_END)
174 && lastpte && lastpte + PTRS_PER_PTE != pte);
180 * This function initializes a certain range of kernel virtual memory
181 * with new bootmem page tables, everywhere page tables are missing in
184 * NOTE: The pagetables are allocated contiguous on the physical space
185 * so we can cache the place of the first one and move around without
186 * checking the pgd every time.
189 page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base)
191 int pgd_idx, pmd_idx;
198 pgd_idx = pgd_index(vaddr);
199 pmd_idx = pmd_index(vaddr);
200 pgd = pgd_base + pgd_idx;
202 for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
203 pmd = one_md_table_init(pgd);
204 pmd = pmd + pmd_index(vaddr);
205 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
207 pte = page_table_kmap_check(one_page_table_init(pmd),
216 static inline int is_kernel_text(unsigned long addr)
218 if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end)
224 * This maps the physical memory to kernel virtual address space, a total
225 * of max_low_pfn pages, by creating page tables starting from address
228 static void __init kernel_physical_mapping_init(pgd_t *pgd_base,
229 unsigned long start_pfn,
230 unsigned long end_pfn,
233 int pgd_idx, pmd_idx, pte_ofs;
238 unsigned pages_2m, pages_4k;
242 * First iteration will setup identity mapping using large/small pages
243 * based on use_pse, with other attributes same as set by
244 * the early code in head_32.S
246 * Second iteration will setup the appropriate attributes (NX, GLOBAL..)
247 * as desired for the kernel identity mapping.
249 * This two pass mechanism conforms to the TLB app note which says:
251 * "Software should not write to a paging-structure entry in a way
252 * that would change, for any linear address, both the page size
253 * and either the page frame or attributes."
261 pages_2m = pages_4k = 0;
263 pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
264 pgd = pgd_base + pgd_idx;
265 for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
266 pmd = one_md_table_init(pgd);
270 #ifdef CONFIG_X86_PAE
271 pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
276 for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn;
278 unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET;
281 * Map with big pages if possible, otherwise
282 * create normal page tables:
286 pgprot_t prot = PAGE_KERNEL_LARGE;
288 * first pass will use the same initial
289 * identity mapping attribute + _PAGE_PSE.
292 __pgprot(PTE_IDENT_ATTR |
295 addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
296 PAGE_OFFSET + PAGE_SIZE-1;
298 if (is_kernel_text(addr) ||
299 is_kernel_text(addr2))
300 prot = PAGE_KERNEL_LARGE_EXEC;
303 if (mapping_iter == 1)
304 set_pmd(pmd, pfn_pmd(pfn, init_prot));
306 set_pmd(pmd, pfn_pmd(pfn, prot));
311 pte = one_page_table_init(pmd);
313 pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
315 for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn;
316 pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
317 pgprot_t prot = PAGE_KERNEL;
319 * first pass will use the same initial
320 * identity mapping attribute.
322 pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR);
324 if (is_kernel_text(addr))
325 prot = PAGE_KERNEL_EXEC;
328 if (mapping_iter == 1)
329 set_pte(pte, pfn_pte(pfn, init_prot));
331 set_pte(pte, pfn_pte(pfn, prot));
335 if (mapping_iter == 1) {
337 * update direct mapping page count only in the first
340 update_page_count(PG_LEVEL_2M, pages_2m);
341 update_page_count(PG_LEVEL_4K, pages_4k);
344 * local global flush tlb, which will flush the previous
345 * mappings present in both small and large page TLB's.
350 * Second iteration will set the actual desired PTE attributes.
360 static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr)
362 return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
363 vaddr), vaddr), vaddr);
366 static void __init kmap_init(void)
368 unsigned long kmap_vstart;
371 * Cache the first kmap pte:
373 kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
374 kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
376 kmap_prot = PAGE_KERNEL;
379 #ifdef CONFIG_HIGHMEM
380 static void __init permanent_kmaps_init(pgd_t *pgd_base)
389 page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
391 pgd = swapper_pg_dir + pgd_index(vaddr);
392 pud = pud_offset(pgd, vaddr);
393 pmd = pmd_offset(pud, vaddr);
394 pte = pte_offset_kernel(pmd, vaddr);
395 pkmap_page_table = pte;
398 static void __init add_one_highpage_init(struct page *page, int pfn)
400 ClearPageReserved(page);
401 init_page_count(page);
406 struct add_highpages_data {
407 unsigned long start_pfn;
408 unsigned long end_pfn;
411 static int __init add_highpages_work_fn(unsigned long start_pfn,
412 unsigned long end_pfn, void *datax)
416 unsigned long final_start_pfn, final_end_pfn;
417 struct add_highpages_data *data;
419 data = (struct add_highpages_data *)datax;
421 final_start_pfn = max(start_pfn, data->start_pfn);
422 final_end_pfn = min(end_pfn, data->end_pfn);
423 if (final_start_pfn >= final_end_pfn)
426 for (node_pfn = final_start_pfn; node_pfn < final_end_pfn;
428 if (!pfn_valid(node_pfn))
430 page = pfn_to_page(node_pfn);
431 add_one_highpage_init(page, node_pfn);
438 void __init add_highpages_with_active_regions(int nid, unsigned long start_pfn,
439 unsigned long end_pfn)
441 struct add_highpages_data data;
443 data.start_pfn = start_pfn;
444 data.end_pfn = end_pfn;
446 work_with_active_regions(nid, add_highpages_work_fn, &data);
450 static inline void permanent_kmaps_init(pgd_t *pgd_base)
453 #endif /* CONFIG_HIGHMEM */
455 void __init native_pagetable_setup_start(pgd_t *base)
457 unsigned long pfn, va;
464 * Remove any mappings which extend past the end of physical
465 * memory from the boot time page table:
467 for (pfn = max_low_pfn + 1; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
468 va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
469 pgd = base + pgd_index(va);
470 if (!pgd_present(*pgd))
473 pud = pud_offset(pgd, va);
474 pmd = pmd_offset(pud, va);
475 if (!pmd_present(*pmd))
478 pte = pte_offset_kernel(pmd, va);
479 if (!pte_present(*pte))
482 pte_clear(NULL, va, pte);
484 paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT);
487 void __init native_pagetable_setup_done(pgd_t *base)
492 * Build a proper pagetable for the kernel mappings. Up until this
493 * point, we've been running on some set of pagetables constructed by
496 * If we're booting on native hardware, this will be a pagetable
497 * constructed in arch/x86/kernel/head_32.S. The root of the
498 * pagetable will be swapper_pg_dir.
500 * If we're booting paravirtualized under a hypervisor, then there are
501 * more options: we may already be running PAE, and the pagetable may
502 * or may not be based in swapper_pg_dir. In any case,
503 * paravirt_pagetable_setup_start() will set up swapper_pg_dir
504 * appropriately for the rest of the initialization to work.
506 * In general, pagetable_init() assumes that the pagetable may already
507 * be partially populated, and so it avoids stomping on any existing
510 static void __init early_ioremap_page_table_range_init(pgd_t *pgd_base)
512 unsigned long vaddr, end;
515 * Fixed mappings, only the page table structure has to be
516 * created - mappings will be set by set_fixmap():
518 vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
519 end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
520 page_table_range_init(vaddr, end, pgd_base);
521 early_ioremap_reset();
524 static void __init pagetable_init(void)
526 pgd_t *pgd_base = swapper_pg_dir;
528 permanent_kmaps_init(pgd_base);
531 #ifdef CONFIG_ACPI_SLEEP
533 * ACPI suspend needs this for resume, because things like the intel-agp
534 * driver might have split up a kernel 4MB mapping.
536 char swsusp_pg_dir[PAGE_SIZE]
537 __attribute__ ((aligned(PAGE_SIZE)));
539 static inline void save_pg_dir(void)
541 memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
543 #else /* !CONFIG_ACPI_SLEEP */
544 static inline void save_pg_dir(void)
547 #endif /* !CONFIG_ACPI_SLEEP */
549 void zap_low_mappings(void)
554 * Zap initial low-memory mappings.
556 * Note that "pgd_clear()" doesn't do it for
557 * us, because pgd_clear() is a no-op on i386.
559 for (i = 0; i < KERNEL_PGD_BOUNDARY; i++) {
560 #ifdef CONFIG_X86_PAE
561 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
563 set_pgd(swapper_pg_dir+i, __pgd(0));
571 pteval_t __supported_pte_mask __read_mostly = ~(_PAGE_NX | _PAGE_GLOBAL | _PAGE_IOMAP);
572 EXPORT_SYMBOL_GPL(__supported_pte_mask);
574 #ifdef CONFIG_X86_PAE
576 static int disable_nx __initdata;
581 * Control non executable mappings.
586 static int __init noexec_setup(char *str)
588 if (!str || !strcmp(str, "on")) {
590 __supported_pte_mask |= _PAGE_NX;
594 if (!strcmp(str, "off")) {
596 __supported_pte_mask &= ~_PAGE_NX;
604 early_param("noexec", noexec_setup);
606 static void __init set_nx(void)
608 unsigned int v[4], l, h;
610 if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) {
611 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]);
613 if ((v[3] & (1 << 20)) && !disable_nx) {
614 rdmsr(MSR_EFER, l, h);
616 wrmsr(MSR_EFER, l, h);
618 __supported_pte_mask |= _PAGE_NX;
624 /* user-defined highmem size */
625 static unsigned int highmem_pages = -1;
628 * highmem=size forces highmem to be exactly 'size' bytes.
629 * This works even on boxes that have no highmem otherwise.
630 * This also works to reduce highmem size on bigger boxes.
632 static int __init parse_highmem(char *arg)
637 highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
640 early_param("highmem", parse_highmem);
642 #define MSG_HIGHMEM_TOO_BIG \
643 "highmem size (%luMB) is bigger than pages available (%luMB)!\n"
645 #define MSG_LOWMEM_TOO_SMALL \
646 "highmem size (%luMB) results in <64MB lowmem, ignoring it!\n"
648 * All of RAM fits into lowmem - but if user wants highmem
649 * artificially via the highmem=x boot parameter then create
652 void __init lowmem_pfn_init(void)
654 /* max_low_pfn is 0, we already have early_res support */
655 max_low_pfn = max_pfn;
657 if (highmem_pages == -1)
659 #ifdef CONFIG_HIGHMEM
660 if (highmem_pages >= max_pfn) {
661 printk(KERN_ERR MSG_HIGHMEM_TOO_BIG,
662 pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
666 if (max_low_pfn - highmem_pages < 64*1024*1024/PAGE_SIZE) {
667 printk(KERN_ERR MSG_LOWMEM_TOO_SMALL,
668 pages_to_mb(highmem_pages));
671 max_low_pfn -= highmem_pages;
675 printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
679 #define MSG_HIGHMEM_TOO_SMALL \
680 "only %luMB highmem pages available, ignoring highmem size of %luMB!\n"
682 #define MSG_HIGHMEM_TRIMMED \
683 "Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n"
685 * We have more RAM than fits into lowmem - we try to put it into
686 * highmem, also taking the highmem=x boot parameter into account:
688 void __init highmem_pfn_init(void)
690 max_low_pfn = MAXMEM_PFN;
692 if (highmem_pages == -1)
693 highmem_pages = max_pfn - MAXMEM_PFN;
695 if (highmem_pages + MAXMEM_PFN < max_pfn)
696 max_pfn = MAXMEM_PFN + highmem_pages;
698 if (highmem_pages + MAXMEM_PFN > max_pfn) {
699 printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL,
700 pages_to_mb(max_pfn - MAXMEM_PFN),
701 pages_to_mb(highmem_pages));
704 #ifndef CONFIG_HIGHMEM
705 /* Maximum memory usable is what is directly addressable */
706 printk(KERN_WARNING "Warning only %ldMB will be used.\n", MAXMEM>>20);
707 if (max_pfn > MAX_NONPAE_PFN)
708 printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
710 printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
711 max_pfn = MAXMEM_PFN;
712 #else /* !CONFIG_HIGHMEM */
713 #ifndef CONFIG_HIGHMEM64G
714 if (max_pfn > MAX_NONPAE_PFN) {
715 max_pfn = MAX_NONPAE_PFN;
716 printk(KERN_WARNING MSG_HIGHMEM_TRIMMED);
718 #endif /* !CONFIG_HIGHMEM64G */
719 #endif /* !CONFIG_HIGHMEM */
723 * Determine low and high memory ranges:
725 void __init find_low_pfn_range(void)
727 /* it could update max_pfn */
729 if (max_pfn <= MAXMEM_PFN)
735 #ifndef CONFIG_NEED_MULTIPLE_NODES
736 void __init initmem_init(unsigned long start_pfn,
737 unsigned long end_pfn)
739 #ifdef CONFIG_HIGHMEM
740 highstart_pfn = highend_pfn = max_pfn;
741 if (max_pfn > max_low_pfn)
742 highstart_pfn = max_low_pfn;
743 memory_present(0, 0, highend_pfn);
744 e820_register_active_regions(0, 0, highend_pfn);
745 printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
746 pages_to_mb(highend_pfn - highstart_pfn));
747 num_physpages = highend_pfn;
748 high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
750 memory_present(0, 0, max_low_pfn);
751 e820_register_active_regions(0, 0, max_low_pfn);
752 num_physpages = max_low_pfn;
753 high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
755 #ifdef CONFIG_FLATMEM
756 max_mapnr = num_physpages;
758 printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
759 pages_to_mb(max_low_pfn));
761 setup_bootmem_allocator();
763 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
765 static void __init zone_sizes_init(void)
767 unsigned long max_zone_pfns[MAX_NR_ZONES];
768 memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
769 max_zone_pfns[ZONE_DMA] =
770 virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
771 max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
772 #ifdef CONFIG_HIGHMEM
773 max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
776 free_area_init_nodes(max_zone_pfns);
779 #ifdef CONFIG_NEED_MULTIPLE_NODES
780 static unsigned long __init setup_node_bootmem(int nodeid,
781 unsigned long start_pfn,
782 unsigned long end_pfn,
783 unsigned long bootmap)
785 unsigned long bootmap_size;
787 if (start_pfn > max_low_pfn)
789 if (end_pfn > max_low_pfn)
790 end_pfn = max_low_pfn;
792 /* don't touch min_low_pfn */
793 bootmap_size = init_bootmem_node(NODE_DATA(nodeid),
794 bootmap >> PAGE_SHIFT,
796 printk(KERN_INFO " node %d low ram: %08lx - %08lx\n",
797 nodeid, start_pfn<<PAGE_SHIFT, end_pfn<<PAGE_SHIFT);
798 printk(KERN_INFO " node %d bootmap %08lx - %08lx\n",
799 nodeid, bootmap, bootmap + bootmap_size);
800 free_bootmem_with_active_regions(nodeid, end_pfn);
801 early_res_to_bootmem(start_pfn<<PAGE_SHIFT, end_pfn<<PAGE_SHIFT);
803 return bootmap + bootmap_size;
807 void __init setup_bootmem_allocator(void)
810 unsigned long bootmap_size, bootmap;
812 * Initialize the boot-time allocator (with low memory only):
814 bootmap_size = bootmem_bootmap_pages(max_low_pfn)<<PAGE_SHIFT;
815 bootmap = find_e820_area(min_low_pfn<<PAGE_SHIFT,
816 max_pfn_mapped<<PAGE_SHIFT, bootmap_size,
819 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
820 reserve_early(bootmap, bootmap + bootmap_size, "BOOTMAP");
822 printk(KERN_INFO " mapped low ram: 0 - %08lx\n",
823 max_pfn_mapped<<PAGE_SHIFT);
824 printk(KERN_INFO " low ram: %08lx - %08lx\n",
825 min_low_pfn<<PAGE_SHIFT, max_low_pfn<<PAGE_SHIFT);
827 #ifdef CONFIG_NEED_MULTIPLE_NODES
828 for_each_online_node(nodeid)
829 bootmap = setup_node_bootmem(nodeid, node_start_pfn[nodeid],
830 node_end_pfn[nodeid], 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 " bootmap %08lx - %08lx\n",
836 bootmap, bootmap + bootmap_size);
837 free_bootmem_with_active_regions(0, max_low_pfn);
838 early_res_to_bootmem(0, max_low_pfn<<PAGE_SHIFT);
841 after_init_bootmem = 1;
844 static void __init find_early_table_space(unsigned long end, int use_pse)
846 unsigned long puds, pmds, ptes, tables, start;
848 puds = (end + PUD_SIZE - 1) >> PUD_SHIFT;
849 tables = roundup(puds * sizeof(pud_t), PAGE_SIZE);
851 pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT;
852 tables += roundup(pmds * sizeof(pmd_t), PAGE_SIZE);
857 extra = end - ((end>>PMD_SHIFT) << PMD_SHIFT);
859 ptes = (extra + PAGE_SIZE - 1) >> PAGE_SHIFT;
861 ptes = (end + PAGE_SIZE - 1) >> PAGE_SHIFT;
863 tables += roundup(ptes * sizeof(pte_t), PAGE_SIZE);
866 tables += roundup(__end_of_fixed_addresses * sizeof(pte_t), PAGE_SIZE);
869 * RED-PEN putting page tables only on node 0 could
870 * cause a hotspot and fill up ZONE_DMA. The page tables
871 * need roughly 0.5KB per GB.
874 table_start = find_e820_area(start, max_pfn_mapped<<PAGE_SHIFT,
876 if (table_start == -1UL)
877 panic("Cannot find space for the kernel page tables");
879 table_start >>= PAGE_SHIFT;
880 table_end = table_start;
881 table_top = table_start + (tables>>PAGE_SHIFT);
883 printk(KERN_DEBUG "kernel direct mapping tables up to %lx @ %lx-%lx\n",
884 end, table_start << PAGE_SHIFT,
885 (table_start << PAGE_SHIFT) + tables);
891 unsigned page_size_mask;
894 #define NR_RANGE_MR 3
896 static int save_mr(struct map_range *mr, int nr_range,
897 unsigned long start_pfn, unsigned long end_pfn,
898 unsigned long page_size_mask)
900 if (start_pfn < end_pfn) {
901 if (nr_range >= NR_RANGE_MR)
902 panic("run out of range for init_memory_mapping\n");
903 mr[nr_range].start = start_pfn<<PAGE_SHIFT;
904 mr[nr_range].end = end_pfn<<PAGE_SHIFT;
905 mr[nr_range].page_size_mask = page_size_mask;
912 unsigned long __init_refok init_memory_mapping(unsigned long start,
915 pgd_t *pgd_base = swapper_pg_dir;
916 unsigned long page_size_mask = 0;
917 unsigned long start_pfn, end_pfn;
920 struct map_range mr[NR_RANGE_MR];
924 printk(KERN_INFO "init_memory_mapping: %08lx-%08lx\n", start, end);
926 #ifdef CONFIG_DEBUG_PAGEALLOC
928 * For CONFIG_DEBUG_PAGEALLOC, identity mapping will use small pages.
929 * This will simplify cpa(), which otherwise needs to support splitting
930 * large pages into small in interrupt context, etc.
934 use_pse = cpu_has_pse;
937 #ifdef CONFIG_X86_PAE
940 printk(KERN_INFO "NX (Execute Disable) protection: active\n");
943 /* Enable PSE if available */
945 set_in_cr4(X86_CR4_PSE);
947 /* Enable PGE if available */
949 set_in_cr4(X86_CR4_PGE);
950 __supported_pte_mask |= _PAGE_GLOBAL;
953 memset(mr, 0, sizeof(mr));
957 page_size_mask |= 1 << PG_LEVEL_2M;
960 * Don't use a large page for the first 2/4MB of memory
961 * because there are often fixed size MTRRs in there
962 * and overlapping MTRRs into large pages can cause
965 /* head could not be big page alignment ? */
966 start_pfn = start >> PAGE_SHIFT;
967 pos = start_pfn << PAGE_SHIFT;
969 end_pfn = 1<<(PMD_SHIFT - PAGE_SHIFT);
971 end_pfn = ((pos + (PMD_SIZE - 1))>>PMD_SHIFT)
972 << (PMD_SHIFT - PAGE_SHIFT);
973 if (end_pfn > (end>>PAGE_SHIFT))
974 end_pfn = end>>PAGE_SHIFT;
975 if (start_pfn < end_pfn) {
976 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0);
977 pos = end_pfn << PAGE_SHIFT;
981 start_pfn = ((pos + (PMD_SIZE - 1))>>PMD_SHIFT)
982 << (PMD_SHIFT - PAGE_SHIFT);
983 end_pfn = (end>>PMD_SHIFT) << (PMD_SHIFT - PAGE_SHIFT);
984 if (start_pfn < end_pfn) {
985 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
986 page_size_mask & (1<<PG_LEVEL_2M));
987 pos = end_pfn << PAGE_SHIFT;
990 /* tail is not big page alignment ? */
991 start_pfn = pos>>PAGE_SHIFT;
992 end_pfn = end>>PAGE_SHIFT;
993 if (start_pfn < end_pfn)
994 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0);
996 /* try to merge same page size and continuous */
997 for (i = 0; nr_range > 1 && i < nr_range - 1; i++) {
998 unsigned long old_start;
999 if (mr[i].end != mr[i+1].start ||
1000 mr[i].page_size_mask != mr[i+1].page_size_mask)
1003 old_start = mr[i].start;
1004 memmove(&mr[i], &mr[i+1],
1005 (nr_range - 1 - i) * sizeof(struct map_range));
1006 mr[i--].start = old_start;
1010 for (i = 0; i < nr_range; i++)
1011 printk(KERN_DEBUG " %08lx - %08lx page %s\n",
1012 mr[i].start, mr[i].end,
1013 (mr[i].page_size_mask & (1<<PG_LEVEL_2M)) ?
1017 * Find space for the kernel direct mapping tables.
1019 if (!after_init_bootmem)
1020 find_early_table_space(end, use_pse);
1022 for (i = 0; i < nr_range; i++)
1023 kernel_physical_mapping_init(pgd_base,
1024 mr[i].start >> PAGE_SHIFT,
1025 mr[i].end >> PAGE_SHIFT,
1026 mr[i].page_size_mask == (1<<PG_LEVEL_2M));
1028 early_ioremap_page_table_range_init(pgd_base);
1030 load_cr3(swapper_pg_dir);
1034 if (!after_init_bootmem)
1035 reserve_early(table_start << PAGE_SHIFT,
1036 table_end << PAGE_SHIFT, "PGTABLE");
1038 if (!after_init_bootmem)
1039 early_memtest(start, end);
1041 return end >> PAGE_SHIFT;
1046 * paging_init() sets up the page tables - note that the first 8MB are
1047 * already mapped by head.S.
1049 * This routines also unmaps the page at virtual kernel address 0, so
1050 * that we can trap those pesky NULL-reference errors in the kernel.
1052 void __init paging_init(void)
1061 * NOTE: at this point the bootmem allocator is fully available.
1068 * Test if the WP bit works in supervisor mode. It isn't supported on 386's
1069 * and also on some strange 486's. All 586+'s are OK. This used to involve
1070 * black magic jumps to work around some nasty CPU bugs, but fortunately the
1071 * switch to using exceptions got rid of all that.
1073 static void __init test_wp_bit(void)
1076 "Checking if this processor honours the WP bit even in supervisor mode...");
1078 /* Any page-aligned address will do, the test is non-destructive */
1079 __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
1080 boot_cpu_data.wp_works_ok = do_test_wp_bit();
1081 clear_fixmap(FIX_WP_TEST);
1083 if (!boot_cpu_data.wp_works_ok) {
1084 printk(KERN_CONT "No.\n");
1085 #ifdef CONFIG_X86_WP_WORKS_OK
1087 "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
1090 printk(KERN_CONT "Ok.\n");
1094 static struct kcore_list kcore_mem, kcore_vmalloc;
1096 void __init mem_init(void)
1098 int codesize, reservedpages, datasize, initsize;
1103 #ifdef CONFIG_FLATMEM
1106 /* this will put all low memory onto the freelists */
1107 totalram_pages += free_all_bootmem();
1110 for (tmp = 0; tmp < max_low_pfn; tmp++)
1112 * Only count reserved RAM pages:
1114 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
1117 set_highmem_pages_init();
1119 codesize = (unsigned long) &_etext - (unsigned long) &_text;
1120 datasize = (unsigned long) &_edata - (unsigned long) &_etext;
1121 initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin;
1123 kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
1124 kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
1125 VMALLOC_END-VMALLOC_START);
1127 printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, "
1128 "%dk reserved, %dk data, %dk init, %ldk highmem)\n",
1129 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
1130 num_physpages << (PAGE_SHIFT-10),
1132 reservedpages << (PAGE_SHIFT-10),
1135 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10))
1138 printk(KERN_INFO "virtual kernel memory layout:\n"
1139 " fixmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
1140 #ifdef CONFIG_HIGHMEM
1141 " pkmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
1143 " vmalloc : 0x%08lx - 0x%08lx (%4ld MB)\n"
1144 " lowmem : 0x%08lx - 0x%08lx (%4ld MB)\n"
1145 " .init : 0x%08lx - 0x%08lx (%4ld kB)\n"
1146 " .data : 0x%08lx - 0x%08lx (%4ld kB)\n"
1147 " .text : 0x%08lx - 0x%08lx (%4ld kB)\n",
1148 FIXADDR_START, FIXADDR_TOP,
1149 (FIXADDR_TOP - FIXADDR_START) >> 10,
1151 #ifdef CONFIG_HIGHMEM
1152 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
1153 (LAST_PKMAP*PAGE_SIZE) >> 10,
1156 VMALLOC_START, VMALLOC_END,
1157 (VMALLOC_END - VMALLOC_START) >> 20,
1159 (unsigned long)__va(0), (unsigned long)high_memory,
1160 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
1162 (unsigned long)&__init_begin, (unsigned long)&__init_end,
1163 ((unsigned long)&__init_end -
1164 (unsigned long)&__init_begin) >> 10,
1166 (unsigned long)&_etext, (unsigned long)&_edata,
1167 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
1169 (unsigned long)&_text, (unsigned long)&_etext,
1170 ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
1173 * Check boundaries twice: Some fundamental inconsistencies can
1174 * be detected at build time already.
1176 #define __FIXADDR_TOP (-PAGE_SIZE)
1177 #ifdef CONFIG_HIGHMEM
1178 BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START);
1179 BUILD_BUG_ON(VMALLOC_END > PKMAP_BASE);
1181 #define high_memory (-128UL << 20)
1182 BUILD_BUG_ON(VMALLOC_START >= VMALLOC_END);
1184 #undef __FIXADDR_TOP
1186 #ifdef CONFIG_HIGHMEM
1187 BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START);
1188 BUG_ON(VMALLOC_END > PKMAP_BASE);
1190 BUG_ON(VMALLOC_START >= VMALLOC_END);
1191 BUG_ON((unsigned long)high_memory > VMALLOC_START);
1193 if (boot_cpu_data.wp_works_ok < 0)
1200 #ifdef CONFIG_MEMORY_HOTPLUG
1201 int arch_add_memory(int nid, u64 start, u64 size)
1203 struct pglist_data *pgdata = NODE_DATA(nid);
1204 struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
1205 unsigned long start_pfn = start >> PAGE_SHIFT;
1206 unsigned long nr_pages = size >> PAGE_SHIFT;
1208 return __add_pages(nid, zone, start_pfn, nr_pages);
1213 * This function cannot be __init, since exceptions don't work in that
1214 * section. Put this after the callers, so that it cannot be inlined.
1216 static noinline int do_test_wp_bit(void)
1221 __asm__ __volatile__(
1227 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
1236 #ifdef CONFIG_DEBUG_RODATA
1237 const int rodata_test_data = 0xC3;
1238 EXPORT_SYMBOL_GPL(rodata_test_data);
1240 void mark_rodata_ro(void)
1242 unsigned long start = PFN_ALIGN(_text);
1243 unsigned long size = PFN_ALIGN(_etext) - start;
1245 #ifndef CONFIG_DYNAMIC_FTRACE
1246 /* Dynamic tracing modifies the kernel text section */
1247 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1248 printk(KERN_INFO "Write protecting the kernel text: %luk\n",
1251 #ifdef CONFIG_CPA_DEBUG
1252 printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n",
1254 set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT);
1256 printk(KERN_INFO "Testing CPA: write protecting again\n");
1257 set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT);
1259 #endif /* CONFIG_DYNAMIC_FTRACE */
1262 size = (unsigned long)__end_rodata - start;
1263 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1264 printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
1268 #ifdef CONFIG_CPA_DEBUG
1269 printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size);
1270 set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
1272 printk(KERN_INFO "Testing CPA: write protecting again\n");
1273 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1278 #ifdef CONFIG_BLK_DEV_INITRD
1279 void free_initrd_mem(unsigned long start, unsigned long end)
1281 free_init_pages("initrd memory", start, end);
1285 int __init reserve_bootmem_generic(unsigned long phys, unsigned long len,
1288 return reserve_bootmem(phys, len, flags);