2 * linux/arch/x86_64/mm/init.c
4 * Copyright (C) 1995 Linus Torvalds
5 * Copyright (C) 2000 Pavel Machek <pavel@suse.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>
33 #include <asm/processor.h>
34 #include <asm/bios_ebda.h>
35 #include <asm/system.h>
36 #include <asm/uaccess.h>
37 #include <asm/pgtable.h>
38 #include <asm/pgalloc.h>
40 #include <asm/fixmap.h>
44 #include <asm/mmu_context.h>
45 #include <asm/proto.h>
47 #include <asm/sections.h>
48 #include <asm/kdebug.h>
50 #include <asm/cacheflush.h>
52 #include <linux/bootmem.h>
54 static unsigned long dma_reserve __initdata;
56 static int __init parse_direct_gbpages_off(char *arg)
61 early_param("nogbpages", parse_direct_gbpages_off);
63 static int __init parse_direct_gbpages_on(char *arg)
68 early_param("gbpages", parse_direct_gbpages_on);
71 * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the
72 * physical space so we can cache the place of the first one and move
73 * around without checking the pgd every time.
76 pteval_t __supported_pte_mask __read_mostly = ~_PAGE_IOMAP;
77 EXPORT_SYMBOL_GPL(__supported_pte_mask);
79 int force_personality32;
83 * Control non executable heap for 32bit processes.
84 * To control the stack too use noexec=off
86 * on PROT_READ does not imply PROT_EXEC for 32-bit processes (default)
87 * off PROT_READ implies PROT_EXEC
89 static int __init nonx32_setup(char *str)
91 if (!strcmp(str, "on"))
92 force_personality32 &= ~READ_IMPLIES_EXEC;
93 else if (!strcmp(str, "off"))
94 force_personality32 |= READ_IMPLIES_EXEC;
97 __setup("noexec32=", nonx32_setup);
100 * NOTE: This function is marked __ref because it calls __init function
101 * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0.
103 static __ref void *spp_getpage(void)
108 ptr = (void *) get_zeroed_page(GFP_ATOMIC | __GFP_NOTRACK);
110 ptr = alloc_bootmem_pages(PAGE_SIZE);
112 if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) {
113 panic("set_pte_phys: cannot allocate page data %s\n",
114 after_bootmem ? "after bootmem" : "");
117 pr_debug("spp_getpage %p\n", ptr);
122 static pud_t *fill_pud(pgd_t *pgd, unsigned long vaddr)
124 if (pgd_none(*pgd)) {
125 pud_t *pud = (pud_t *)spp_getpage();
126 pgd_populate(&init_mm, pgd, pud);
127 if (pud != pud_offset(pgd, 0))
128 printk(KERN_ERR "PAGETABLE BUG #00! %p <-> %p\n",
129 pud, pud_offset(pgd, 0));
131 return pud_offset(pgd, vaddr);
134 static pmd_t *fill_pmd(pud_t *pud, unsigned long vaddr)
136 if (pud_none(*pud)) {
137 pmd_t *pmd = (pmd_t *) spp_getpage();
138 pud_populate(&init_mm, pud, pmd);
139 if (pmd != pmd_offset(pud, 0))
140 printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n",
141 pmd, pmd_offset(pud, 0));
143 return pmd_offset(pud, vaddr);
146 static pte_t *fill_pte(pmd_t *pmd, unsigned long vaddr)
148 if (pmd_none(*pmd)) {
149 pte_t *pte = (pte_t *) spp_getpage();
150 pmd_populate_kernel(&init_mm, pmd, pte);
151 if (pte != pte_offset_kernel(pmd, 0))
152 printk(KERN_ERR "PAGETABLE BUG #02!\n");
154 return pte_offset_kernel(pmd, vaddr);
157 void set_pte_vaddr_pud(pud_t *pud_page, unsigned long vaddr, pte_t new_pte)
163 pud = pud_page + pud_index(vaddr);
164 pmd = fill_pmd(pud, vaddr);
165 pte = fill_pte(pmd, vaddr);
167 set_pte(pte, new_pte);
170 * It's enough to flush this one mapping.
171 * (PGE mappings get flushed as well)
173 __flush_tlb_one(vaddr);
176 void set_pte_vaddr(unsigned long vaddr, pte_t pteval)
181 pr_debug("set_pte_vaddr %lx to %lx\n", vaddr, native_pte_val(pteval));
183 pgd = pgd_offset_k(vaddr);
184 if (pgd_none(*pgd)) {
186 "PGD FIXMAP MISSING, it should be setup in head.S!\n");
189 pud_page = (pud_t*)pgd_page_vaddr(*pgd);
190 set_pte_vaddr_pud(pud_page, vaddr, pteval);
193 pmd_t * __init populate_extra_pmd(unsigned long vaddr)
198 pgd = pgd_offset_k(vaddr);
199 pud = fill_pud(pgd, vaddr);
200 return fill_pmd(pud, vaddr);
203 pte_t * __init populate_extra_pte(unsigned long vaddr)
207 pmd = populate_extra_pmd(vaddr);
208 return fill_pte(pmd, vaddr);
212 * Create large page table mappings for a range of physical addresses.
214 static void __init __init_extra_mapping(unsigned long phys, unsigned long size,
221 BUG_ON((phys & ~PMD_MASK) || (size & ~PMD_MASK));
222 for (; size; phys += PMD_SIZE, size -= PMD_SIZE) {
223 pgd = pgd_offset_k((unsigned long)__va(phys));
224 if (pgd_none(*pgd)) {
225 pud = (pud_t *) spp_getpage();
226 set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE |
229 pud = pud_offset(pgd, (unsigned long)__va(phys));
230 if (pud_none(*pud)) {
231 pmd = (pmd_t *) spp_getpage();
232 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE |
235 pmd = pmd_offset(pud, phys);
236 BUG_ON(!pmd_none(*pmd));
237 set_pmd(pmd, __pmd(phys | pgprot_val(prot)));
241 void __init init_extra_mapping_wb(unsigned long phys, unsigned long size)
243 __init_extra_mapping(phys, size, PAGE_KERNEL_LARGE);
246 void __init init_extra_mapping_uc(unsigned long phys, unsigned long size)
248 __init_extra_mapping(phys, size, PAGE_KERNEL_LARGE_NOCACHE);
252 * The head.S code sets up the kernel high mapping:
254 * from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
256 * phys_addr holds the negative offset to the kernel, which is added
257 * to the compile time generated pmds. This results in invalid pmds up
258 * to the point where we hit the physaddr 0 mapping.
260 * We limit the mappings to the region from _text to _end. _end is
261 * rounded up to the 2MB boundary. This catches the invalid pmds as
262 * well, as they are located before _text:
264 void __init cleanup_highmap(void)
266 unsigned long vaddr = __START_KERNEL_map;
267 unsigned long end = roundup((unsigned long)_end, PMD_SIZE) - 1;
268 pmd_t *pmd = level2_kernel_pgt;
269 pmd_t *last_pmd = pmd + PTRS_PER_PMD;
271 for (; pmd < last_pmd; pmd++, vaddr += PMD_SIZE) {
274 if (vaddr < (unsigned long) _text || vaddr > end)
275 set_pmd(pmd, __pmd(0));
279 static __ref void *alloc_low_page(unsigned long *phys)
281 unsigned long pfn = e820_table_end++;
285 adr = (void *)get_zeroed_page(GFP_ATOMIC | __GFP_NOTRACK);
291 if (pfn >= e820_table_top)
292 panic("alloc_low_page: ran out of memory");
294 adr = early_memremap(pfn * PAGE_SIZE, PAGE_SIZE);
295 memset(adr, 0, PAGE_SIZE);
296 *phys = pfn * PAGE_SIZE;
300 static __ref void unmap_low_page(void *adr)
305 early_iounmap(adr, PAGE_SIZE);
308 static unsigned long __meminit
309 phys_pte_init(pte_t *pte_page, unsigned long addr, unsigned long end,
313 unsigned long last_map_addr = end;
316 pte_t *pte = pte_page + pte_index(addr);
318 for(i = pte_index(addr); i < PTRS_PER_PTE; i++, addr += PAGE_SIZE, pte++) {
321 if (!after_bootmem) {
322 for(; i < PTRS_PER_PTE; i++, pte++)
323 set_pte(pte, __pte(0));
329 * We will re-use the existing mapping.
330 * Xen for example has some special requirements, like mapping
331 * pagetable pages as RO. So assume someone who pre-setup
332 * these mappings are more intelligent.
340 printk(" pte=%p addr=%lx pte=%016lx\n",
341 pte, addr, pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL).pte);
343 set_pte(pte, pfn_pte(addr >> PAGE_SHIFT, prot));
344 last_map_addr = (addr & PAGE_MASK) + PAGE_SIZE;
347 update_page_count(PG_LEVEL_4K, pages);
349 return last_map_addr;
352 static unsigned long __meminit
353 phys_pte_update(pmd_t *pmd, unsigned long address, unsigned long end,
356 pte_t *pte = (pte_t *)pmd_page_vaddr(*pmd);
358 return phys_pte_init(pte, address, end, prot);
361 static unsigned long __meminit
362 phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end,
363 unsigned long page_size_mask, pgprot_t prot)
365 unsigned long pages = 0;
366 unsigned long last_map_addr = end;
368 int i = pmd_index(address);
370 for (; i < PTRS_PER_PMD; i++, address += PMD_SIZE) {
371 unsigned long pte_phys;
372 pmd_t *pmd = pmd_page + pmd_index(address);
374 pgprot_t new_prot = prot;
376 if (address >= end) {
377 if (!after_bootmem) {
378 for (; i < PTRS_PER_PMD; i++, pmd++)
379 set_pmd(pmd, __pmd(0));
385 if (!pmd_large(*pmd)) {
386 spin_lock(&init_mm.page_table_lock);
387 last_map_addr = phys_pte_update(pmd, address,
389 spin_unlock(&init_mm.page_table_lock);
393 * If we are ok with PG_LEVEL_2M mapping, then we will
394 * use the existing mapping,
396 * Otherwise, we will split the large page mapping but
397 * use the same existing protection bits except for
398 * large page, so that we don't violate Intel's TLB
399 * Application note (317080) which says, while changing
400 * the page sizes, new and old translations should
401 * not differ with respect to page frame and
404 if (page_size_mask & (1 << PG_LEVEL_2M)) {
408 new_prot = pte_pgprot(pte_clrhuge(*(pte_t *)pmd));
411 if (page_size_mask & (1<<PG_LEVEL_2M)) {
413 spin_lock(&init_mm.page_table_lock);
414 set_pte((pte_t *)pmd,
415 pfn_pte(address >> PAGE_SHIFT,
416 __pgprot(pgprot_val(prot) | _PAGE_PSE)));
417 spin_unlock(&init_mm.page_table_lock);
418 last_map_addr = (address & PMD_MASK) + PMD_SIZE;
422 pte = alloc_low_page(&pte_phys);
423 last_map_addr = phys_pte_init(pte, address, end, new_prot);
426 spin_lock(&init_mm.page_table_lock);
427 pmd_populate_kernel(&init_mm, pmd, __va(pte_phys));
428 spin_unlock(&init_mm.page_table_lock);
430 update_page_count(PG_LEVEL_2M, pages);
431 return last_map_addr;
434 static unsigned long __meminit
435 phys_pmd_update(pud_t *pud, unsigned long address, unsigned long end,
436 unsigned long page_size_mask, pgprot_t prot)
438 pmd_t *pmd = pmd_offset(pud, 0);
439 unsigned long last_map_addr;
441 last_map_addr = phys_pmd_init(pmd, address, end, page_size_mask, prot);
443 return last_map_addr;
446 static unsigned long __meminit
447 phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end,
448 unsigned long page_size_mask)
450 unsigned long pages = 0;
451 unsigned long last_map_addr = end;
452 int i = pud_index(addr);
454 for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE) {
455 unsigned long pmd_phys;
456 pud_t *pud = pud_page + pud_index(addr);
458 pgprot_t prot = PAGE_KERNEL;
463 if (!after_bootmem &&
464 !e820_any_mapped(addr, addr+PUD_SIZE, 0)) {
465 set_pud(pud, __pud(0));
470 if (!pud_large(*pud)) {
471 last_map_addr = phys_pmd_update(pud, addr, end,
472 page_size_mask, prot);
476 * If we are ok with PG_LEVEL_1G mapping, then we will
477 * use the existing mapping.
479 * Otherwise, we will split the gbpage mapping but use
480 * the same existing protection bits except for large
481 * page, so that we don't violate Intel's TLB
482 * Application note (317080) which says, while changing
483 * the page sizes, new and old translations should
484 * not differ with respect to page frame and
487 if (page_size_mask & (1 << PG_LEVEL_1G)) {
491 prot = pte_pgprot(pte_clrhuge(*(pte_t *)pud));
494 if (page_size_mask & (1<<PG_LEVEL_1G)) {
496 spin_lock(&init_mm.page_table_lock);
497 set_pte((pte_t *)pud,
498 pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL_LARGE));
499 spin_unlock(&init_mm.page_table_lock);
500 last_map_addr = (addr & PUD_MASK) + PUD_SIZE;
504 pmd = alloc_low_page(&pmd_phys);
505 last_map_addr = phys_pmd_init(pmd, addr, end, page_size_mask,
509 spin_lock(&init_mm.page_table_lock);
510 pud_populate(&init_mm, pud, __va(pmd_phys));
511 spin_unlock(&init_mm.page_table_lock);
515 update_page_count(PG_LEVEL_1G, pages);
517 return last_map_addr;
520 static unsigned long __meminit
521 phys_pud_update(pgd_t *pgd, unsigned long addr, unsigned long end,
522 unsigned long page_size_mask)
526 pud = (pud_t *)pgd_page_vaddr(*pgd);
528 return phys_pud_init(pud, addr, end, page_size_mask);
531 unsigned long __meminit
532 kernel_physical_mapping_init(unsigned long start,
534 unsigned long page_size_mask)
537 unsigned long next, last_map_addr = end;
539 start = (unsigned long)__va(start);
540 end = (unsigned long)__va(end);
542 for (; start < end; start = next) {
543 pgd_t *pgd = pgd_offset_k(start);
544 unsigned long pud_phys;
547 next = (start + PGDIR_SIZE) & PGDIR_MASK;
552 last_map_addr = phys_pud_update(pgd, __pa(start),
553 __pa(end), page_size_mask);
557 pud = alloc_low_page(&pud_phys);
558 last_map_addr = phys_pud_init(pud, __pa(start), __pa(next),
562 spin_lock(&init_mm.page_table_lock);
563 pgd_populate(&init_mm, pgd, __va(pud_phys));
564 spin_unlock(&init_mm.page_table_lock);
568 return last_map_addr;
572 void __init initmem_init(unsigned long start_pfn, unsigned long end_pfn,
575 unsigned long bootmap_size, bootmap;
577 bootmap_size = bootmem_bootmap_pages(end_pfn)<<PAGE_SHIFT;
578 bootmap = find_e820_area(0, end_pfn<<PAGE_SHIFT, bootmap_size,
581 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
582 reserve_early(bootmap, bootmap + bootmap_size, "BOOTMAP");
583 /* don't touch min_low_pfn */
584 bootmap_size = init_bootmem_node(NODE_DATA(0), bootmap >> PAGE_SHIFT,
586 e820_register_active_regions(0, start_pfn, end_pfn);
587 free_bootmem_with_active_regions(0, end_pfn);
591 void __init paging_init(void)
593 unsigned long max_zone_pfns[MAX_NR_ZONES];
595 memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
596 max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN;
597 max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN;
598 max_zone_pfns[ZONE_NORMAL] = max_pfn;
600 sparse_memory_present_with_active_regions(MAX_NUMNODES);
604 * clear the default setting with node 0
605 * note: don't use nodes_clear here, that is really clearing when
606 * numa support is not compiled in, and later node_set_state
607 * will not set it back.
609 node_clear_state(0, N_NORMAL_MEMORY);
611 free_area_init_nodes(max_zone_pfns);
615 * Memory hotplug specific functions
617 #ifdef CONFIG_MEMORY_HOTPLUG
619 * After memory hotplug the variables max_pfn, max_low_pfn and high_memory need
622 static void update_end_of_memory_vars(u64 start, u64 size)
624 unsigned long end_pfn = PFN_UP(start + size);
626 if (end_pfn > max_pfn) {
628 max_low_pfn = end_pfn;
629 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
634 * Memory is added always to NORMAL zone. This means you will never get
635 * additional DMA/DMA32 memory.
637 int arch_add_memory(int nid, u64 start, u64 size)
639 struct pglist_data *pgdat = NODE_DATA(nid);
640 struct zone *zone = pgdat->node_zones + ZONE_NORMAL;
641 unsigned long last_mapped_pfn, start_pfn = start >> PAGE_SHIFT;
642 unsigned long nr_pages = size >> PAGE_SHIFT;
645 last_mapped_pfn = init_memory_mapping(start, start + size);
646 if (last_mapped_pfn > max_pfn_mapped)
647 max_pfn_mapped = last_mapped_pfn;
649 ret = __add_pages(nid, zone, start_pfn, nr_pages);
652 /* update max_pfn, max_low_pfn and high_memory */
653 update_end_of_memory_vars(start, size);
657 EXPORT_SYMBOL_GPL(arch_add_memory);
659 #if !defined(CONFIG_ACPI_NUMA) && defined(CONFIG_NUMA)
660 int memory_add_physaddr_to_nid(u64 start)
664 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
667 #endif /* CONFIG_MEMORY_HOTPLUG */
669 static struct kcore_list kcore_vsyscall;
671 void __init mem_init(void)
673 long codesize, reservedpages, datasize, initsize;
674 unsigned long absent_pages;
678 /* clear_bss() already clear the empty_zero_page */
682 /* this will put all low memory onto the freelists */
684 totalram_pages = numa_free_all_bootmem();
686 totalram_pages = free_all_bootmem();
689 absent_pages = absent_pages_in_range(0, max_pfn);
690 reservedpages = max_pfn - totalram_pages - absent_pages;
693 codesize = (unsigned long) &_etext - (unsigned long) &_text;
694 datasize = (unsigned long) &_edata - (unsigned long) &_etext;
695 initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin;
697 /* Register memory areas for /proc/kcore */
698 kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START,
699 VSYSCALL_END - VSYSCALL_START, KCORE_OTHER);
701 printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, "
702 "%ldk absent, %ldk reserved, %ldk data, %ldk init)\n",
703 nr_free_pages() << (PAGE_SHIFT-10),
704 max_pfn << (PAGE_SHIFT-10),
706 absent_pages << (PAGE_SHIFT-10),
707 reservedpages << (PAGE_SHIFT-10),
712 #ifdef CONFIG_DEBUG_RODATA
713 const int rodata_test_data = 0xC3;
714 EXPORT_SYMBOL_GPL(rodata_test_data);
716 int kernel_set_to_readonly;
718 void set_kernel_text_rw(void)
720 unsigned long start = PFN_ALIGN(_text);
721 unsigned long end = PFN_ALIGN(__stop___ex_table);
723 if (!kernel_set_to_readonly)
726 pr_debug("Set kernel text: %lx - %lx for read write\n",
730 * Make the kernel identity mapping for text RW. Kernel text
731 * mapping will always be RO. Refer to the comment in
732 * static_protections() in pageattr.c
734 set_memory_rw(start, (end - start) >> PAGE_SHIFT);
737 void set_kernel_text_ro(void)
739 unsigned long start = PFN_ALIGN(_text);
740 unsigned long end = PFN_ALIGN(__stop___ex_table);
742 if (!kernel_set_to_readonly)
745 pr_debug("Set kernel text: %lx - %lx for read only\n",
749 * Set the kernel identity mapping for text RO.
751 set_memory_ro(start, (end - start) >> PAGE_SHIFT);
754 void mark_rodata_ro(void)
756 unsigned long start = PFN_ALIGN(_text);
757 unsigned long rodata_start =
758 ((unsigned long)__start_rodata + PAGE_SIZE - 1) & PAGE_MASK;
759 unsigned long end = (unsigned long) &__end_rodata_hpage_align;
760 unsigned long text_end = PAGE_ALIGN((unsigned long) &__stop___ex_table);
761 unsigned long rodata_end = PAGE_ALIGN((unsigned long) &__end_rodata);
762 unsigned long data_start = (unsigned long) &_sdata;
764 printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
765 (end - start) >> 10);
766 set_memory_ro(start, (end - start) >> PAGE_SHIFT);
768 kernel_set_to_readonly = 1;
771 * The rodata section (but not the kernel text!) should also be
774 set_memory_nx(rodata_start, (end - rodata_start) >> PAGE_SHIFT);
778 #ifdef CONFIG_CPA_DEBUG
779 printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end);
780 set_memory_rw(start, (end-start) >> PAGE_SHIFT);
782 printk(KERN_INFO "Testing CPA: again\n");
783 set_memory_ro(start, (end-start) >> PAGE_SHIFT);
786 free_init_pages("unused kernel memory",
787 (unsigned long) page_address(virt_to_page(text_end)),
789 page_address(virt_to_page(rodata_start)));
790 free_init_pages("unused kernel memory",
791 (unsigned long) page_address(virt_to_page(rodata_end)),
792 (unsigned long) page_address(virt_to_page(data_start)));
797 int __init reserve_bootmem_generic(unsigned long phys, unsigned long len,
804 unsigned long pfn = phys >> PAGE_SHIFT;
806 if (pfn >= max_pfn) {
808 * This can happen with kdump kernels when accessing
811 if (pfn < max_pfn_mapped)
814 printk(KERN_ERR "reserve_bootmem: illegal reserve %lx %lu\n",
819 /* Should check here against the e820 map to avoid double free */
821 nid = phys_to_nid(phys);
822 next_nid = phys_to_nid(phys + len - 1);
824 ret = reserve_bootmem_node(NODE_DATA(nid), phys, len, flags);
826 ret = reserve_bootmem(phys, len, flags);
832 reserve_bootmem(phys, len, flags);
835 if (phys+len <= MAX_DMA_PFN*PAGE_SIZE) {
836 dma_reserve += len / PAGE_SIZE;
837 set_dma_reserve(dma_reserve);
843 int kern_addr_valid(unsigned long addr)
845 unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
851 if (above != 0 && above != -1UL)
854 pgd = pgd_offset_k(addr);
858 pud = pud_offset(pgd, addr);
862 pmd = pmd_offset(pud, addr);
867 return pfn_valid(pmd_pfn(*pmd));
869 pte = pte_offset_kernel(pmd, addr);
873 return pfn_valid(pte_pfn(*pte));
877 * A pseudo VMA to allow ptrace access for the vsyscall page. This only
878 * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
879 * not need special handling anymore:
881 static struct vm_area_struct gate_vma = {
882 .vm_start = VSYSCALL_START,
883 .vm_end = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE),
884 .vm_page_prot = PAGE_READONLY_EXEC,
885 .vm_flags = VM_READ | VM_EXEC
888 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
890 #ifdef CONFIG_IA32_EMULATION
891 if (test_tsk_thread_flag(tsk, TIF_IA32))
897 int in_gate_area(struct task_struct *task, unsigned long addr)
899 struct vm_area_struct *vma = get_gate_vma(task);
904 return (addr >= vma->vm_start) && (addr < vma->vm_end);
908 * Use this when you have no reliable task/vma, typically from interrupt
909 * context. It is less reliable than using the task's vma and may give
912 int in_gate_area_no_task(unsigned long addr)
914 return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END);
917 const char *arch_vma_name(struct vm_area_struct *vma)
919 if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
921 if (vma == &gate_vma)
926 #ifdef CONFIG_SPARSEMEM_VMEMMAP
928 * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
930 static long __meminitdata addr_start, addr_end;
931 static void __meminitdata *p_start, *p_end;
932 static int __meminitdata node_start;
935 vmemmap_populate(struct page *start_page, unsigned long size, int node)
937 unsigned long addr = (unsigned long)start_page;
938 unsigned long end = (unsigned long)(start_page + size);
944 for (; addr < end; addr = next) {
947 pgd = vmemmap_pgd_populate(addr, node);
951 pud = vmemmap_pud_populate(pgd, addr, node);
956 next = (addr + PAGE_SIZE) & PAGE_MASK;
957 pmd = vmemmap_pmd_populate(pud, addr, node);
962 p = vmemmap_pte_populate(pmd, addr, node);
967 addr_end = addr + PAGE_SIZE;
968 p_end = p + PAGE_SIZE;
970 next = pmd_addr_end(addr, end);
972 pmd = pmd_offset(pud, addr);
973 if (pmd_none(*pmd)) {
976 p = vmemmap_alloc_block(PMD_SIZE, node);
980 entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
982 set_pmd(pmd, __pmd(pte_val(entry)));
984 /* check to see if we have contiguous blocks */
985 if (p_end != p || node_start != node) {
987 printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n",
988 addr_start, addr_end-1, p_start, p_end-1, node_start);
994 addr_end = addr + PMD_SIZE;
995 p_end = p + PMD_SIZE;
997 vmemmap_verify((pte_t *)pmd, node, addr, next);
1004 void __meminit vmemmap_populate_print_last(void)
1007 printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1008 addr_start, addr_end-1, p_start, p_end-1, node_start);