2 * linux/arch/x86_64/mm/init.c
4 * Copyright (C) 1995 Linus Torvalds
5 * Copyright (C) 2000 Pavel Machek <pavel@ucw.cz>
6 * Copyright (C) 2002,2003 Andi Kleen <ak@suse.de>
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/swap.h>
19 #include <linux/smp.h>
20 #include <linux/init.h>
21 #include <linux/initrd.h>
22 #include <linux/pagemap.h>
23 #include <linux/bootmem.h>
24 #include <linux/proc_fs.h>
25 #include <linux/pci.h>
26 #include <linux/pfn.h>
27 #include <linux/poison.h>
28 #include <linux/dma-mapping.h>
29 #include <linux/module.h>
30 #include <linux/memory_hotplug.h>
31 #include <linux/nmi.h>
32 #include <linux/gfp.h>
34 #include <asm/processor.h>
35 #include <asm/bios_ebda.h>
36 #include <asm/system.h>
37 #include <asm/uaccess.h>
38 #include <asm/pgtable.h>
39 #include <asm/pgalloc.h>
41 #include <asm/fixmap.h>
45 #include <asm/mmu_context.h>
46 #include <asm/proto.h>
48 #include <asm/sections.h>
49 #include <asm/kdebug.h>
51 #include <asm/cacheflush.h>
53 #include <linux/bootmem.h>
55 static unsigned long dma_reserve __initdata;
57 static int __init parse_direct_gbpages_off(char *arg)
62 early_param("nogbpages", parse_direct_gbpages_off);
64 static int __init parse_direct_gbpages_on(char *arg)
69 early_param("gbpages", parse_direct_gbpages_on);
72 * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the
73 * physical space so we can cache the place of the first one and move
74 * around without checking the pgd every time.
77 pteval_t __supported_pte_mask __read_mostly = ~_PAGE_IOMAP;
78 EXPORT_SYMBOL_GPL(__supported_pte_mask);
80 int force_personality32;
84 * Control non executable heap for 32bit processes.
85 * To control the stack too use noexec=off
87 * on PROT_READ does not imply PROT_EXEC for 32-bit processes (default)
88 * off PROT_READ implies PROT_EXEC
90 static int __init nonx32_setup(char *str)
92 if (!strcmp(str, "on"))
93 force_personality32 &= ~READ_IMPLIES_EXEC;
94 else if (!strcmp(str, "off"))
95 force_personality32 |= READ_IMPLIES_EXEC;
98 __setup("noexec32=", nonx32_setup);
101 * NOTE: This function is marked __ref because it calls __init function
102 * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0.
104 static __ref void *spp_getpage(void)
109 ptr = (void *) get_zeroed_page(GFP_ATOMIC | __GFP_NOTRACK);
111 ptr = alloc_bootmem_pages(PAGE_SIZE);
113 if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) {
114 panic("set_pte_phys: cannot allocate page data %s\n",
115 after_bootmem ? "after bootmem" : "");
118 pr_debug("spp_getpage %p\n", ptr);
123 static pud_t *fill_pud(pgd_t *pgd, unsigned long vaddr)
125 if (pgd_none(*pgd)) {
126 pud_t *pud = (pud_t *)spp_getpage();
127 pgd_populate(&init_mm, pgd, pud);
128 if (pud != pud_offset(pgd, 0))
129 printk(KERN_ERR "PAGETABLE BUG #00! %p <-> %p\n",
130 pud, pud_offset(pgd, 0));
132 return pud_offset(pgd, vaddr);
135 static pmd_t *fill_pmd(pud_t *pud, unsigned long vaddr)
137 if (pud_none(*pud)) {
138 pmd_t *pmd = (pmd_t *) spp_getpage();
139 pud_populate(&init_mm, pud, pmd);
140 if (pmd != pmd_offset(pud, 0))
141 printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n",
142 pmd, pmd_offset(pud, 0));
144 return pmd_offset(pud, vaddr);
147 static pte_t *fill_pte(pmd_t *pmd, unsigned long vaddr)
149 if (pmd_none(*pmd)) {
150 pte_t *pte = (pte_t *) spp_getpage();
151 pmd_populate_kernel(&init_mm, pmd, pte);
152 if (pte != pte_offset_kernel(pmd, 0))
153 printk(KERN_ERR "PAGETABLE BUG #02!\n");
155 return pte_offset_kernel(pmd, vaddr);
158 void set_pte_vaddr_pud(pud_t *pud_page, unsigned long vaddr, pte_t new_pte)
164 pud = pud_page + pud_index(vaddr);
165 pmd = fill_pmd(pud, vaddr);
166 pte = fill_pte(pmd, vaddr);
168 set_pte(pte, new_pte);
171 * It's enough to flush this one mapping.
172 * (PGE mappings get flushed as well)
174 __flush_tlb_one(vaddr);
177 void set_pte_vaddr(unsigned long vaddr, pte_t pteval)
182 pr_debug("set_pte_vaddr %lx to %lx\n", vaddr, native_pte_val(pteval));
184 pgd = pgd_offset_k(vaddr);
185 if (pgd_none(*pgd)) {
187 "PGD FIXMAP MISSING, it should be setup in head.S!\n");
190 pud_page = (pud_t*)pgd_page_vaddr(*pgd);
191 set_pte_vaddr_pud(pud_page, vaddr, pteval);
194 pmd_t * __init populate_extra_pmd(unsigned long vaddr)
199 pgd = pgd_offset_k(vaddr);
200 pud = fill_pud(pgd, vaddr);
201 return fill_pmd(pud, vaddr);
204 pte_t * __init populate_extra_pte(unsigned long vaddr)
208 pmd = populate_extra_pmd(vaddr);
209 return fill_pte(pmd, vaddr);
213 * Create large page table mappings for a range of physical addresses.
215 static void __init __init_extra_mapping(unsigned long phys, unsigned long size,
222 BUG_ON((phys & ~PMD_MASK) || (size & ~PMD_MASK));
223 for (; size; phys += PMD_SIZE, size -= PMD_SIZE) {
224 pgd = pgd_offset_k((unsigned long)__va(phys));
225 if (pgd_none(*pgd)) {
226 pud = (pud_t *) spp_getpage();
227 set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE |
230 pud = pud_offset(pgd, (unsigned long)__va(phys));
231 if (pud_none(*pud)) {
232 pmd = (pmd_t *) spp_getpage();
233 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE |
236 pmd = pmd_offset(pud, phys);
237 BUG_ON(!pmd_none(*pmd));
238 set_pmd(pmd, __pmd(phys | pgprot_val(prot)));
242 void __init init_extra_mapping_wb(unsigned long phys, unsigned long size)
244 __init_extra_mapping(phys, size, PAGE_KERNEL_LARGE);
247 void __init init_extra_mapping_uc(unsigned long phys, unsigned long size)
249 __init_extra_mapping(phys, size, PAGE_KERNEL_LARGE_NOCACHE);
253 * The head.S code sets up the kernel high mapping:
255 * from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
257 * phys_addr holds the negative offset to the kernel, which is added
258 * to the compile time generated pmds. This results in invalid pmds up
259 * to the point where we hit the physaddr 0 mapping.
261 * We limit the mappings to the region from _text to _end. _end is
262 * rounded up to the 2MB boundary. This catches the invalid pmds as
263 * well, as they are located before _text:
265 void __init cleanup_highmap(void)
267 unsigned long vaddr = __START_KERNEL_map;
268 unsigned long end = roundup((unsigned long)_end, PMD_SIZE) - 1;
269 pmd_t *pmd = level2_kernel_pgt;
270 pmd_t *last_pmd = pmd + PTRS_PER_PMD;
272 for (; pmd < last_pmd; pmd++, vaddr += PMD_SIZE) {
275 if (vaddr < (unsigned long) _text || vaddr > end)
276 set_pmd(pmd, __pmd(0));
280 static __ref void *alloc_low_page(unsigned long *phys)
282 unsigned long pfn = e820_table_end++;
286 adr = (void *)get_zeroed_page(GFP_ATOMIC | __GFP_NOTRACK);
292 if (pfn >= e820_table_top)
293 panic("alloc_low_page: ran out of memory");
295 adr = early_memremap(pfn * PAGE_SIZE, PAGE_SIZE);
296 memset(adr, 0, PAGE_SIZE);
297 *phys = pfn * PAGE_SIZE;
301 static __ref void unmap_low_page(void *adr)
306 early_iounmap(adr, PAGE_SIZE);
309 static unsigned long __meminit
310 phys_pte_init(pte_t *pte_page, unsigned long addr, unsigned long end,
314 unsigned long last_map_addr = end;
317 pte_t *pte = pte_page + pte_index(addr);
319 for(i = pte_index(addr); i < PTRS_PER_PTE; i++, addr += PAGE_SIZE, pte++) {
322 if (!after_bootmem) {
323 for(; i < PTRS_PER_PTE; i++, pte++)
324 set_pte(pte, __pte(0));
330 * We will re-use the existing mapping.
331 * Xen for example has some special requirements, like mapping
332 * pagetable pages as RO. So assume someone who pre-setup
333 * these mappings are more intelligent.
341 printk(" pte=%p addr=%lx pte=%016lx\n",
342 pte, addr, pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL).pte);
344 set_pte(pte, pfn_pte(addr >> PAGE_SHIFT, prot));
345 last_map_addr = (addr & PAGE_MASK) + PAGE_SIZE;
348 update_page_count(PG_LEVEL_4K, pages);
350 return last_map_addr;
353 static unsigned long __meminit
354 phys_pte_update(pmd_t *pmd, unsigned long address, unsigned long end,
357 pte_t *pte = (pte_t *)pmd_page_vaddr(*pmd);
359 return phys_pte_init(pte, address, end, prot);
362 static unsigned long __meminit
363 phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end,
364 unsigned long page_size_mask, pgprot_t prot)
366 unsigned long pages = 0;
367 unsigned long last_map_addr = end;
369 int i = pmd_index(address);
371 for (; i < PTRS_PER_PMD; i++, address += PMD_SIZE) {
372 unsigned long pte_phys;
373 pmd_t *pmd = pmd_page + pmd_index(address);
375 pgprot_t new_prot = prot;
377 if (address >= end) {
378 if (!after_bootmem) {
379 for (; i < PTRS_PER_PMD; i++, pmd++)
380 set_pmd(pmd, __pmd(0));
386 if (!pmd_large(*pmd)) {
387 spin_lock(&init_mm.page_table_lock);
388 last_map_addr = phys_pte_update(pmd, address,
390 spin_unlock(&init_mm.page_table_lock);
394 * If we are ok with PG_LEVEL_2M mapping, then we will
395 * use the existing mapping,
397 * Otherwise, we will split the large page mapping but
398 * use the same existing protection bits except for
399 * large page, so that we don't violate Intel's TLB
400 * Application note (317080) which says, while changing
401 * the page sizes, new and old translations should
402 * not differ with respect to page frame and
405 if (page_size_mask & (1 << PG_LEVEL_2M)) {
409 new_prot = pte_pgprot(pte_clrhuge(*(pte_t *)pmd));
412 if (page_size_mask & (1<<PG_LEVEL_2M)) {
414 spin_lock(&init_mm.page_table_lock);
415 set_pte((pte_t *)pmd,
416 pfn_pte(address >> PAGE_SHIFT,
417 __pgprot(pgprot_val(prot) | _PAGE_PSE)));
418 spin_unlock(&init_mm.page_table_lock);
419 last_map_addr = (address & PMD_MASK) + PMD_SIZE;
423 pte = alloc_low_page(&pte_phys);
424 last_map_addr = phys_pte_init(pte, address, end, new_prot);
427 spin_lock(&init_mm.page_table_lock);
428 pmd_populate_kernel(&init_mm, pmd, __va(pte_phys));
429 spin_unlock(&init_mm.page_table_lock);
431 update_page_count(PG_LEVEL_2M, pages);
432 return last_map_addr;
435 static unsigned long __meminit
436 phys_pmd_update(pud_t *pud, unsigned long address, unsigned long end,
437 unsigned long page_size_mask, pgprot_t prot)
439 pmd_t *pmd = pmd_offset(pud, 0);
440 unsigned long last_map_addr;
442 last_map_addr = phys_pmd_init(pmd, address, end, page_size_mask, prot);
444 return last_map_addr;
447 static unsigned long __meminit
448 phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end,
449 unsigned long page_size_mask)
451 unsigned long pages = 0;
452 unsigned long last_map_addr = end;
453 int i = pud_index(addr);
455 for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE) {
456 unsigned long pmd_phys;
457 pud_t *pud = pud_page + pud_index(addr);
459 pgprot_t prot = PAGE_KERNEL;
464 if (!after_bootmem &&
465 !e820_any_mapped(addr, addr+PUD_SIZE, 0)) {
466 set_pud(pud, __pud(0));
471 if (!pud_large(*pud)) {
472 last_map_addr = phys_pmd_update(pud, addr, end,
473 page_size_mask, prot);
477 * If we are ok with PG_LEVEL_1G mapping, then we will
478 * use the existing mapping.
480 * Otherwise, we will split the gbpage mapping but use
481 * the same existing protection bits except for large
482 * page, so that we don't violate Intel's TLB
483 * Application note (317080) which says, while changing
484 * the page sizes, new and old translations should
485 * not differ with respect to page frame and
488 if (page_size_mask & (1 << PG_LEVEL_1G)) {
492 prot = pte_pgprot(pte_clrhuge(*(pte_t *)pud));
495 if (page_size_mask & (1<<PG_LEVEL_1G)) {
497 spin_lock(&init_mm.page_table_lock);
498 set_pte((pte_t *)pud,
499 pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL_LARGE));
500 spin_unlock(&init_mm.page_table_lock);
501 last_map_addr = (addr & PUD_MASK) + PUD_SIZE;
505 pmd = alloc_low_page(&pmd_phys);
506 last_map_addr = phys_pmd_init(pmd, addr, end, page_size_mask,
510 spin_lock(&init_mm.page_table_lock);
511 pud_populate(&init_mm, pud, __va(pmd_phys));
512 spin_unlock(&init_mm.page_table_lock);
516 update_page_count(PG_LEVEL_1G, pages);
518 return last_map_addr;
521 static unsigned long __meminit
522 phys_pud_update(pgd_t *pgd, unsigned long addr, unsigned long end,
523 unsigned long page_size_mask)
527 pud = (pud_t *)pgd_page_vaddr(*pgd);
529 return phys_pud_init(pud, addr, end, page_size_mask);
532 unsigned long __meminit
533 kernel_physical_mapping_init(unsigned long start,
535 unsigned long page_size_mask)
538 unsigned long next, last_map_addr = end;
540 start = (unsigned long)__va(start);
541 end = (unsigned long)__va(end);
543 for (; start < end; start = next) {
544 pgd_t *pgd = pgd_offset_k(start);
545 unsigned long pud_phys;
548 next = (start + PGDIR_SIZE) & PGDIR_MASK;
553 last_map_addr = phys_pud_update(pgd, __pa(start),
554 __pa(end), page_size_mask);
558 pud = alloc_low_page(&pud_phys);
559 last_map_addr = phys_pud_init(pud, __pa(start), __pa(next),
563 spin_lock(&init_mm.page_table_lock);
564 pgd_populate(&init_mm, pgd, __va(pud_phys));
565 spin_unlock(&init_mm.page_table_lock);
569 return last_map_addr;
573 void __init initmem_init(unsigned long start_pfn, unsigned long end_pfn,
576 #ifndef CONFIG_NO_BOOTMEM
577 unsigned long bootmap_size, bootmap;
579 bootmap_size = bootmem_bootmap_pages(end_pfn)<<PAGE_SHIFT;
580 bootmap = find_e820_area(0, end_pfn<<PAGE_SHIFT, bootmap_size,
583 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
584 reserve_early(bootmap, bootmap + bootmap_size, "BOOTMAP");
585 /* don't touch min_low_pfn */
586 bootmap_size = init_bootmem_node(NODE_DATA(0), bootmap >> PAGE_SHIFT,
588 e820_register_active_regions(0, start_pfn, end_pfn);
589 free_bootmem_with_active_regions(0, end_pfn);
591 e820_register_active_regions(0, start_pfn, end_pfn);
596 void __init paging_init(void)
598 unsigned long max_zone_pfns[MAX_NR_ZONES];
600 memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
601 max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN;
602 max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN;
603 max_zone_pfns[ZONE_NORMAL] = max_pfn;
605 sparse_memory_present_with_active_regions(MAX_NUMNODES);
609 * clear the default setting with node 0
610 * note: don't use nodes_clear here, that is really clearing when
611 * numa support is not compiled in, and later node_set_state
612 * will not set it back.
614 node_clear_state(0, N_NORMAL_MEMORY);
616 free_area_init_nodes(max_zone_pfns);
620 * Memory hotplug specific functions
622 #ifdef CONFIG_MEMORY_HOTPLUG
624 * After memory hotplug the variables max_pfn, max_low_pfn and high_memory need
627 static void update_end_of_memory_vars(u64 start, u64 size)
629 unsigned long end_pfn = PFN_UP(start + size);
631 if (end_pfn > max_pfn) {
633 max_low_pfn = end_pfn;
634 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
639 * Memory is added always to NORMAL zone. This means you will never get
640 * additional DMA/DMA32 memory.
642 int arch_add_memory(int nid, u64 start, u64 size)
644 struct pglist_data *pgdat = NODE_DATA(nid);
645 struct zone *zone = pgdat->node_zones + ZONE_NORMAL;
646 unsigned long last_mapped_pfn, start_pfn = start >> PAGE_SHIFT;
647 unsigned long nr_pages = size >> PAGE_SHIFT;
650 last_mapped_pfn = init_memory_mapping(start, start + size);
651 if (last_mapped_pfn > max_pfn_mapped)
652 max_pfn_mapped = last_mapped_pfn;
654 ret = __add_pages(nid, zone, start_pfn, nr_pages);
657 /* update max_pfn, max_low_pfn and high_memory */
658 update_end_of_memory_vars(start, size);
662 EXPORT_SYMBOL_GPL(arch_add_memory);
664 #if !defined(CONFIG_ACPI_NUMA) && defined(CONFIG_NUMA)
665 int memory_add_physaddr_to_nid(u64 start)
669 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
672 #endif /* CONFIG_MEMORY_HOTPLUG */
674 static struct kcore_list kcore_vsyscall;
676 void __init mem_init(void)
678 long codesize, reservedpages, datasize, initsize;
679 unsigned long absent_pages;
683 /* clear_bss() already clear the empty_zero_page */
687 /* this will put all low memory onto the freelists */
689 totalram_pages = numa_free_all_bootmem();
691 totalram_pages = free_all_bootmem();
694 absent_pages = absent_pages_in_range(0, max_pfn);
695 reservedpages = max_pfn - totalram_pages - absent_pages;
698 codesize = (unsigned long) &_etext - (unsigned long) &_text;
699 datasize = (unsigned long) &_edata - (unsigned long) &_etext;
700 initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin;
702 /* Register memory areas for /proc/kcore */
703 kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START,
704 VSYSCALL_END - VSYSCALL_START, KCORE_OTHER);
706 printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, "
707 "%ldk absent, %ldk reserved, %ldk data, %ldk init)\n",
708 nr_free_pages() << (PAGE_SHIFT-10),
709 max_pfn << (PAGE_SHIFT-10),
711 absent_pages << (PAGE_SHIFT-10),
712 reservedpages << (PAGE_SHIFT-10),
717 #ifdef CONFIG_DEBUG_RODATA
718 const int rodata_test_data = 0xC3;
719 EXPORT_SYMBOL_GPL(rodata_test_data);
721 int kernel_set_to_readonly;
723 void set_kernel_text_rw(void)
725 unsigned long start = PFN_ALIGN(_text);
726 unsigned long end = PFN_ALIGN(__stop___ex_table);
728 if (!kernel_set_to_readonly)
731 pr_debug("Set kernel text: %lx - %lx for read write\n",
735 * Make the kernel identity mapping for text RW. Kernel text
736 * mapping will always be RO. Refer to the comment in
737 * static_protections() in pageattr.c
739 set_memory_rw(start, (end - start) >> PAGE_SHIFT);
742 void set_kernel_text_ro(void)
744 unsigned long start = PFN_ALIGN(_text);
745 unsigned long end = PFN_ALIGN(__stop___ex_table);
747 if (!kernel_set_to_readonly)
750 pr_debug("Set kernel text: %lx - %lx for read only\n",
754 * Set the kernel identity mapping for text RO.
756 set_memory_ro(start, (end - start) >> PAGE_SHIFT);
759 void mark_rodata_ro(void)
761 unsigned long start = PFN_ALIGN(_text);
762 unsigned long rodata_start =
763 ((unsigned long)__start_rodata + PAGE_SIZE - 1) & PAGE_MASK;
764 unsigned long end = (unsigned long) &__end_rodata_hpage_align;
765 unsigned long text_end = PAGE_ALIGN((unsigned long) &__stop___ex_table);
766 unsigned long rodata_end = PAGE_ALIGN((unsigned long) &__end_rodata);
767 unsigned long data_start = (unsigned long) &_sdata;
769 printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
770 (end - start) >> 10);
771 set_memory_ro(start, (end - start) >> PAGE_SHIFT);
773 kernel_set_to_readonly = 1;
776 * The rodata section (but not the kernel text!) should also be
779 set_memory_nx(rodata_start, (end - rodata_start) >> PAGE_SHIFT);
783 #ifdef CONFIG_CPA_DEBUG
784 printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end);
785 set_memory_rw(start, (end-start) >> PAGE_SHIFT);
787 printk(KERN_INFO "Testing CPA: again\n");
788 set_memory_ro(start, (end-start) >> PAGE_SHIFT);
791 free_init_pages("unused kernel memory",
792 (unsigned long) page_address(virt_to_page(text_end)),
794 page_address(virt_to_page(rodata_start)));
795 free_init_pages("unused kernel memory",
796 (unsigned long) page_address(virt_to_page(rodata_end)),
797 (unsigned long) page_address(virt_to_page(data_start)));
802 int __init reserve_bootmem_generic(unsigned long phys, unsigned long len,
809 unsigned long pfn = phys >> PAGE_SHIFT;
811 if (pfn >= max_pfn) {
813 * This can happen with kdump kernels when accessing
816 if (pfn < max_pfn_mapped)
819 printk(KERN_ERR "reserve_bootmem: illegal reserve %lx %lu\n",
824 /* Should check here against the e820 map to avoid double free */
826 nid = phys_to_nid(phys);
827 next_nid = phys_to_nid(phys + len - 1);
829 ret = reserve_bootmem_node(NODE_DATA(nid), phys, len, flags);
831 ret = reserve_bootmem(phys, len, flags);
837 reserve_bootmem(phys, len, flags);
840 if (phys+len <= MAX_DMA_PFN*PAGE_SIZE) {
841 dma_reserve += len / PAGE_SIZE;
842 set_dma_reserve(dma_reserve);
848 int kern_addr_valid(unsigned long addr)
850 unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
856 if (above != 0 && above != -1UL)
859 pgd = pgd_offset_k(addr);
863 pud = pud_offset(pgd, addr);
867 pmd = pmd_offset(pud, addr);
872 return pfn_valid(pmd_pfn(*pmd));
874 pte = pte_offset_kernel(pmd, addr);
878 return pfn_valid(pte_pfn(*pte));
882 * A pseudo VMA to allow ptrace access for the vsyscall page. This only
883 * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
884 * not need special handling anymore:
886 static struct vm_area_struct gate_vma = {
887 .vm_start = VSYSCALL_START,
888 .vm_end = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE),
889 .vm_page_prot = PAGE_READONLY_EXEC,
890 .vm_flags = VM_READ | VM_EXEC
893 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
895 #ifdef CONFIG_IA32_EMULATION
896 if (test_tsk_thread_flag(tsk, TIF_IA32))
902 int in_gate_area(struct task_struct *task, unsigned long addr)
904 struct vm_area_struct *vma = get_gate_vma(task);
909 return (addr >= vma->vm_start) && (addr < vma->vm_end);
913 * Use this when you have no reliable task/vma, typically from interrupt
914 * context. It is less reliable than using the task's vma and may give
917 int in_gate_area_no_task(unsigned long addr)
919 return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END);
922 const char *arch_vma_name(struct vm_area_struct *vma)
924 if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
926 if (vma == &gate_vma)
931 #ifdef CONFIG_SPARSEMEM_VMEMMAP
933 * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
935 static long __meminitdata addr_start, addr_end;
936 static void __meminitdata *p_start, *p_end;
937 static int __meminitdata node_start;
940 vmemmap_populate(struct page *start_page, unsigned long size, int node)
942 unsigned long addr = (unsigned long)start_page;
943 unsigned long end = (unsigned long)(start_page + size);
949 for (; addr < end; addr = next) {
952 pgd = vmemmap_pgd_populate(addr, node);
956 pud = vmemmap_pud_populate(pgd, addr, node);
961 next = (addr + PAGE_SIZE) & PAGE_MASK;
962 pmd = vmemmap_pmd_populate(pud, addr, node);
967 p = vmemmap_pte_populate(pmd, addr, node);
972 addr_end = addr + PAGE_SIZE;
973 p_end = p + PAGE_SIZE;
975 next = pmd_addr_end(addr, end);
977 pmd = pmd_offset(pud, addr);
978 if (pmd_none(*pmd)) {
981 p = vmemmap_alloc_block_buf(PMD_SIZE, node);
985 entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
987 set_pmd(pmd, __pmd(pte_val(entry)));
989 /* check to see if we have contiguous blocks */
990 if (p_end != p || node_start != node) {
992 printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n",
993 addr_start, addr_end-1, p_start, p_end-1, node_start);
999 addr_end = addr + PMD_SIZE;
1000 p_end = p + PMD_SIZE;
1002 vmemmap_verify((pte_t *)pmd, node, addr, next);
1009 void __meminit vmemmap_populate_print_last(void)
1012 printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1013 addr_start, addr_end-1, p_start, p_end-1, node_start);