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>
54 * end_pfn only includes RAM, while max_pfn_mapped includes all e820 entries.
55 * The direct mapping extends to max_pfn_mapped, so that we can directly access
56 * apertures, ACPI and other tables without having to play with fixmaps.
58 unsigned long max_low_pfn_mapped;
59 unsigned long max_pfn_mapped;
61 static unsigned long dma_reserve __initdata;
63 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
65 static int __init parse_direct_gbpages_off(char *arg)
70 early_param("nogbpages", parse_direct_gbpages_off);
72 static int __init parse_direct_gbpages_on(char *arg)
77 early_param("gbpages", parse_direct_gbpages_on);
80 * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the
81 * physical space so we can cache the place of the first one and move
82 * around without checking the pgd every time.
85 pteval_t __supported_pte_mask __read_mostly = ~_PAGE_IOMAP;
86 EXPORT_SYMBOL_GPL(__supported_pte_mask);
88 static int do_not_nx __cpuinitdata;
92 * Control non-executable mappings for 64-bit processes.
97 static int __init nonx_setup(char *str)
101 if (!strncmp(str, "on", 2)) {
102 __supported_pte_mask |= _PAGE_NX;
104 } else if (!strncmp(str, "off", 3)) {
106 __supported_pte_mask &= ~_PAGE_NX;
110 early_param("noexec", nonx_setup);
112 void __cpuinit check_efer(void)
116 rdmsrl(MSR_EFER, efer);
117 if (!(efer & EFER_NX) || do_not_nx)
118 __supported_pte_mask &= ~_PAGE_NX;
121 int force_personality32;
125 * Control non executable heap for 32bit processes.
126 * To control the stack too use noexec=off
128 * on PROT_READ does not imply PROT_EXEC for 32-bit processes (default)
129 * off PROT_READ implies PROT_EXEC
131 static int __init nonx32_setup(char *str)
133 if (!strcmp(str, "on"))
134 force_personality32 &= ~READ_IMPLIES_EXEC;
135 else if (!strcmp(str, "off"))
136 force_personality32 |= READ_IMPLIES_EXEC;
139 __setup("noexec32=", nonx32_setup);
142 * NOTE: This function is marked __ref because it calls __init function
143 * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0.
145 static __ref void *spp_getpage(void)
150 ptr = (void *) get_zeroed_page(GFP_ATOMIC);
152 ptr = alloc_bootmem_pages(PAGE_SIZE);
154 if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) {
155 panic("set_pte_phys: cannot allocate page data %s\n",
156 after_bootmem ? "after bootmem" : "");
159 pr_debug("spp_getpage %p\n", ptr);
165 set_pte_vaddr_pud(pud_t *pud_page, unsigned long vaddr, pte_t new_pte)
171 pud = pud_page + pud_index(vaddr);
172 if (pud_none(*pud)) {
173 pmd = (pmd_t *) spp_getpage();
174 pud_populate(&init_mm, pud, pmd);
175 if (pmd != pmd_offset(pud, 0)) {
176 printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n",
177 pmd, pmd_offset(pud, 0));
181 pmd = pmd_offset(pud, vaddr);
182 if (pmd_none(*pmd)) {
183 pte = (pte_t *) spp_getpage();
184 pmd_populate_kernel(&init_mm, pmd, pte);
185 if (pte != pte_offset_kernel(pmd, 0)) {
186 printk(KERN_ERR "PAGETABLE BUG #02!\n");
191 pte = pte_offset_kernel(pmd, vaddr);
192 set_pte(pte, new_pte);
195 * It's enough to flush this one mapping.
196 * (PGE mappings get flushed as well)
198 __flush_tlb_one(vaddr);
202 set_pte_vaddr(unsigned long vaddr, pte_t pteval)
207 pr_debug("set_pte_vaddr %lx to %lx\n", vaddr, native_pte_val(pteval));
209 pgd = pgd_offset_k(vaddr);
210 if (pgd_none(*pgd)) {
212 "PGD FIXMAP MISSING, it should be setup in head.S!\n");
215 pud_page = (pud_t*)pgd_page_vaddr(*pgd);
216 set_pte_vaddr_pud(pud_page, vaddr, pteval);
220 * Create large page table mappings for a range of physical addresses.
222 static void __init __init_extra_mapping(unsigned long phys, unsigned long size,
229 BUG_ON((phys & ~PMD_MASK) || (size & ~PMD_MASK));
230 for (; size; phys += PMD_SIZE, size -= PMD_SIZE) {
231 pgd = pgd_offset_k((unsigned long)__va(phys));
232 if (pgd_none(*pgd)) {
233 pud = (pud_t *) spp_getpage();
234 set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE |
237 pud = pud_offset(pgd, (unsigned long)__va(phys));
238 if (pud_none(*pud)) {
239 pmd = (pmd_t *) spp_getpage();
240 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE |
243 pmd = pmd_offset(pud, phys);
244 BUG_ON(!pmd_none(*pmd));
245 set_pmd(pmd, __pmd(phys | pgprot_val(prot)));
249 void __init init_extra_mapping_wb(unsigned long phys, unsigned long size)
251 __init_extra_mapping(phys, size, PAGE_KERNEL_LARGE);
254 void __init init_extra_mapping_uc(unsigned long phys, unsigned long size)
256 __init_extra_mapping(phys, size, PAGE_KERNEL_LARGE_NOCACHE);
260 * The head.S code sets up the kernel high mapping:
262 * from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
264 * phys_addr holds the negative offset to the kernel, which is added
265 * to the compile time generated pmds. This results in invalid pmds up
266 * to the point where we hit the physaddr 0 mapping.
268 * We limit the mappings to the region from _text to _end. _end is
269 * rounded up to the 2MB boundary. This catches the invalid pmds as
270 * well, as they are located before _text:
272 void __init cleanup_highmap(void)
274 unsigned long vaddr = __START_KERNEL_map;
275 unsigned long end = roundup((unsigned long)_end, PMD_SIZE) - 1;
276 pmd_t *pmd = level2_kernel_pgt;
277 pmd_t *last_pmd = pmd + PTRS_PER_PMD;
279 for (; pmd < last_pmd; pmd++, vaddr += PMD_SIZE) {
282 if (vaddr < (unsigned long) _text || vaddr > end)
283 set_pmd(pmd, __pmd(0));
287 static __ref void *alloc_low_page(unsigned long *phys)
289 unsigned long pfn = e820_table_end++;
293 adr = (void *)get_zeroed_page(GFP_ATOMIC);
299 if (pfn >= e820_table_top)
300 panic("alloc_low_page: ran out of memory");
302 adr = early_memremap(pfn * PAGE_SIZE, PAGE_SIZE);
303 memset(adr, 0, PAGE_SIZE);
304 *phys = pfn * PAGE_SIZE;
308 static __ref void unmap_low_page(void *adr)
313 early_iounmap(adr, PAGE_SIZE);
316 static unsigned long __meminit
317 phys_pte_init(pte_t *pte_page, unsigned long addr, unsigned long end,
321 unsigned long last_map_addr = end;
324 pte_t *pte = pte_page + pte_index(addr);
326 for(i = pte_index(addr); i < PTRS_PER_PTE; i++, addr += PAGE_SIZE, pte++) {
329 if (!after_bootmem) {
330 for(; i < PTRS_PER_PTE; i++, pte++)
331 set_pte(pte, __pte(0));
337 * We will re-use the existing mapping.
338 * Xen for example has some special requirements, like mapping
339 * pagetable pages as RO. So assume someone who pre-setup
340 * these mappings are more intelligent.
348 printk(" pte=%p addr=%lx pte=%016lx\n",
349 pte, addr, pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL).pte);
351 set_pte(pte, pfn_pte(addr >> PAGE_SHIFT, prot));
352 last_map_addr = (addr & PAGE_MASK) + PAGE_SIZE;
355 update_page_count(PG_LEVEL_4K, pages);
357 return last_map_addr;
360 static unsigned long __meminit
361 phys_pte_update(pmd_t *pmd, unsigned long address, unsigned long end,
364 pte_t *pte = (pte_t *)pmd_page_vaddr(*pmd);
366 return phys_pte_init(pte, address, end, prot);
369 static unsigned long __meminit
370 phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end,
371 unsigned long page_size_mask, pgprot_t prot)
373 unsigned long pages = 0;
374 unsigned long last_map_addr = end;
376 int i = pmd_index(address);
378 for (; i < PTRS_PER_PMD; i++, address += PMD_SIZE) {
379 unsigned long pte_phys;
380 pmd_t *pmd = pmd_page + pmd_index(address);
382 pgprot_t new_prot = prot;
384 if (address >= end) {
385 if (!after_bootmem) {
386 for (; i < PTRS_PER_PMD; i++, pmd++)
387 set_pmd(pmd, __pmd(0));
393 if (!pmd_large(*pmd)) {
394 spin_lock(&init_mm.page_table_lock);
395 last_map_addr = phys_pte_update(pmd, address,
397 spin_unlock(&init_mm.page_table_lock);
401 * If we are ok with PG_LEVEL_2M mapping, then we will
402 * use the existing mapping,
404 * Otherwise, we will split the large page mapping but
405 * use the same existing protection bits except for
406 * large page, so that we don't violate Intel's TLB
407 * Application note (317080) which says, while changing
408 * the page sizes, new and old translations should
409 * not differ with respect to page frame and
412 if (page_size_mask & (1 << PG_LEVEL_2M)) {
416 new_prot = pte_pgprot(pte_clrhuge(*(pte_t *)pmd));
419 if (page_size_mask & (1<<PG_LEVEL_2M)) {
421 spin_lock(&init_mm.page_table_lock);
422 set_pte((pte_t *)pmd,
423 pfn_pte(address >> PAGE_SHIFT,
424 __pgprot(pgprot_val(prot) | _PAGE_PSE)));
425 spin_unlock(&init_mm.page_table_lock);
426 last_map_addr = (address & PMD_MASK) + PMD_SIZE;
430 pte = alloc_low_page(&pte_phys);
431 last_map_addr = phys_pte_init(pte, address, end, new_prot);
434 spin_lock(&init_mm.page_table_lock);
435 pmd_populate_kernel(&init_mm, pmd, __va(pte_phys));
436 spin_unlock(&init_mm.page_table_lock);
438 update_page_count(PG_LEVEL_2M, pages);
439 return last_map_addr;
442 static unsigned long __meminit
443 phys_pmd_update(pud_t *pud, unsigned long address, unsigned long end,
444 unsigned long page_size_mask, pgprot_t prot)
446 pmd_t *pmd = pmd_offset(pud, 0);
447 unsigned long last_map_addr;
449 last_map_addr = phys_pmd_init(pmd, address, end, page_size_mask, prot);
451 return last_map_addr;
454 static unsigned long __meminit
455 phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end,
456 unsigned long page_size_mask)
458 unsigned long pages = 0;
459 unsigned long last_map_addr = end;
460 int i = pud_index(addr);
462 for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE) {
463 unsigned long pmd_phys;
464 pud_t *pud = pud_page + pud_index(addr);
466 pgprot_t prot = PAGE_KERNEL;
471 if (!after_bootmem &&
472 !e820_any_mapped(addr, addr+PUD_SIZE, 0)) {
473 set_pud(pud, __pud(0));
478 if (!pud_large(*pud)) {
479 last_map_addr = phys_pmd_update(pud, addr, end,
480 page_size_mask, prot);
484 * If we are ok with PG_LEVEL_1G mapping, then we will
485 * use the existing mapping.
487 * Otherwise, we will split the gbpage mapping but use
488 * the same existing protection bits except for large
489 * page, so that we don't violate Intel's TLB
490 * Application note (317080) which says, while changing
491 * the page sizes, new and old translations should
492 * not differ with respect to page frame and
495 if (page_size_mask & (1 << PG_LEVEL_1G)) {
499 prot = pte_pgprot(pte_clrhuge(*(pte_t *)pud));
502 if (page_size_mask & (1<<PG_LEVEL_1G)) {
504 spin_lock(&init_mm.page_table_lock);
505 set_pte((pte_t *)pud,
506 pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL_LARGE));
507 spin_unlock(&init_mm.page_table_lock);
508 last_map_addr = (addr & PUD_MASK) + PUD_SIZE;
512 pmd = alloc_low_page(&pmd_phys);
513 last_map_addr = phys_pmd_init(pmd, addr, end, page_size_mask,
517 spin_lock(&init_mm.page_table_lock);
518 pud_populate(&init_mm, pud, __va(pmd_phys));
519 spin_unlock(&init_mm.page_table_lock);
523 update_page_count(PG_LEVEL_1G, pages);
525 return last_map_addr;
528 static unsigned long __meminit
529 phys_pud_update(pgd_t *pgd, unsigned long addr, unsigned long end,
530 unsigned long page_size_mask)
534 pud = (pud_t *)pgd_page_vaddr(*pgd);
536 return phys_pud_init(pud, addr, end, page_size_mask);
540 kernel_physical_mapping_init(unsigned long start,
542 unsigned long page_size_mask)
545 unsigned long next, last_map_addr = end;
547 start = (unsigned long)__va(start);
548 end = (unsigned long)__va(end);
550 for (; start < end; start = next) {
551 pgd_t *pgd = pgd_offset_k(start);
552 unsigned long pud_phys;
555 next = (start + PGDIR_SIZE) & PGDIR_MASK;
560 last_map_addr = phys_pud_update(pgd, __pa(start),
561 __pa(end), page_size_mask);
565 pud = alloc_low_page(&pud_phys);
566 last_map_addr = phys_pud_init(pud, __pa(start), __pa(next),
570 spin_lock(&init_mm.page_table_lock);
571 pgd_populate(&init_mm, pgd, __va(pud_phys));
572 spin_unlock(&init_mm.page_table_lock);
576 return last_map_addr;
580 void __init initmem_init(unsigned long start_pfn, unsigned long end_pfn)
582 unsigned long bootmap_size, bootmap;
584 bootmap_size = bootmem_bootmap_pages(end_pfn)<<PAGE_SHIFT;
585 bootmap = find_e820_area(0, end_pfn<<PAGE_SHIFT, bootmap_size,
588 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
589 /* don't touch min_low_pfn */
590 bootmap_size = init_bootmem_node(NODE_DATA(0), bootmap >> PAGE_SHIFT,
592 e820_register_active_regions(0, start_pfn, end_pfn);
593 free_bootmem_with_active_regions(0, end_pfn);
594 early_res_to_bootmem(0, end_pfn<<PAGE_SHIFT);
595 reserve_bootmem(bootmap, bootmap_size, BOOTMEM_DEFAULT);
598 void __init paging_init(void)
600 unsigned long max_zone_pfns[MAX_NR_ZONES];
602 memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
603 max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN;
604 max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN;
605 max_zone_pfns[ZONE_NORMAL] = max_pfn;
607 memory_present(0, 0, max_pfn);
609 free_area_init_nodes(max_zone_pfns);
614 * Memory hotplug specific functions
616 #ifdef CONFIG_MEMORY_HOTPLUG
618 * Memory is added always to NORMAL zone. This means you will never get
619 * additional DMA/DMA32 memory.
621 int arch_add_memory(int nid, u64 start, u64 size)
623 struct pglist_data *pgdat = NODE_DATA(nid);
624 struct zone *zone = pgdat->node_zones + ZONE_NORMAL;
625 unsigned long last_mapped_pfn, start_pfn = start >> PAGE_SHIFT;
626 unsigned long nr_pages = size >> PAGE_SHIFT;
629 last_mapped_pfn = init_memory_mapping(start, start + size);
630 if (last_mapped_pfn > max_pfn_mapped)
631 max_pfn_mapped = last_mapped_pfn;
633 ret = __add_pages(nid, zone, start_pfn, nr_pages);
638 EXPORT_SYMBOL_GPL(arch_add_memory);
640 #if !defined(CONFIG_ACPI_NUMA) && defined(CONFIG_NUMA)
641 int memory_add_physaddr_to_nid(u64 start)
645 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
648 #endif /* CONFIG_MEMORY_HOTPLUG */
650 static struct kcore_list kcore_mem, kcore_vmalloc, kcore_kernel,
651 kcore_modules, kcore_vsyscall;
653 void __init mem_init(void)
655 long codesize, reservedpages, datasize, initsize;
656 unsigned long absent_pages;
660 /* clear_bss() already clear the empty_zero_page */
664 /* this will put all low memory onto the freelists */
666 totalram_pages = numa_free_all_bootmem();
668 totalram_pages = free_all_bootmem();
671 absent_pages = absent_pages_in_range(0, max_pfn);
672 reservedpages = max_pfn - totalram_pages - absent_pages;
675 codesize = (unsigned long) &_etext - (unsigned long) &_text;
676 datasize = (unsigned long) &_edata - (unsigned long) &_etext;
677 initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin;
679 /* Register memory areas for /proc/kcore */
680 kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
681 kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
682 VMALLOC_END-VMALLOC_START);
683 kclist_add(&kcore_kernel, &_stext, _end - _stext);
684 kclist_add(&kcore_modules, (void *)MODULES_VADDR, MODULES_LEN);
685 kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START,
686 VSYSCALL_END - VSYSCALL_START);
688 printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, "
689 "%ldk absent, %ldk reserved, %ldk data, %ldk init)\n",
690 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
691 max_pfn << (PAGE_SHIFT-10),
693 absent_pages << (PAGE_SHIFT-10),
694 reservedpages << (PAGE_SHIFT-10),
699 #ifdef CONFIG_DEBUG_RODATA
700 const int rodata_test_data = 0xC3;
701 EXPORT_SYMBOL_GPL(rodata_test_data);
703 void mark_rodata_ro(void)
705 unsigned long start = PFN_ALIGN(_stext), end = PFN_ALIGN(__end_rodata);
706 unsigned long rodata_start =
707 ((unsigned long)__start_rodata + PAGE_SIZE - 1) & PAGE_MASK;
709 #ifdef CONFIG_DYNAMIC_FTRACE
710 /* Dynamic tracing modifies the kernel text section */
711 start = rodata_start;
714 printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
715 (end - start) >> 10);
716 set_memory_ro(start, (end - start) >> PAGE_SHIFT);
719 * The rodata section (but not the kernel text!) should also be
722 set_memory_nx(rodata_start, (end - rodata_start) >> PAGE_SHIFT);
726 #ifdef CONFIG_CPA_DEBUG
727 printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end);
728 set_memory_rw(start, (end-start) >> PAGE_SHIFT);
730 printk(KERN_INFO "Testing CPA: again\n");
731 set_memory_ro(start, (end-start) >> PAGE_SHIFT);
737 int __init reserve_bootmem_generic(unsigned long phys, unsigned long len,
744 unsigned long pfn = phys >> PAGE_SHIFT;
746 if (pfn >= max_pfn) {
748 * This can happen with kdump kernels when accessing
751 if (pfn < max_pfn_mapped)
754 printk(KERN_ERR "reserve_bootmem: illegal reserve %lx %lu\n",
759 /* Should check here against the e820 map to avoid double free */
761 nid = phys_to_nid(phys);
762 next_nid = phys_to_nid(phys + len - 1);
764 ret = reserve_bootmem_node(NODE_DATA(nid), phys, len, flags);
766 ret = reserve_bootmem(phys, len, flags);
772 reserve_bootmem(phys, len, BOOTMEM_DEFAULT);
775 if (phys+len <= MAX_DMA_PFN*PAGE_SIZE) {
776 dma_reserve += len / PAGE_SIZE;
777 set_dma_reserve(dma_reserve);
783 int kern_addr_valid(unsigned long addr)
785 unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
791 if (above != 0 && above != -1UL)
794 pgd = pgd_offset_k(addr);
798 pud = pud_offset(pgd, addr);
802 pmd = pmd_offset(pud, addr);
807 return pfn_valid(pmd_pfn(*pmd));
809 pte = pte_offset_kernel(pmd, addr);
813 return pfn_valid(pte_pfn(*pte));
817 * A pseudo VMA to allow ptrace access for the vsyscall page. This only
818 * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
819 * not need special handling anymore:
821 static struct vm_area_struct gate_vma = {
822 .vm_start = VSYSCALL_START,
823 .vm_end = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE),
824 .vm_page_prot = PAGE_READONLY_EXEC,
825 .vm_flags = VM_READ | VM_EXEC
828 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
830 #ifdef CONFIG_IA32_EMULATION
831 if (test_tsk_thread_flag(tsk, TIF_IA32))
837 int in_gate_area(struct task_struct *task, unsigned long addr)
839 struct vm_area_struct *vma = get_gate_vma(task);
844 return (addr >= vma->vm_start) && (addr < vma->vm_end);
848 * Use this when you have no reliable task/vma, typically from interrupt
849 * context. It is less reliable than using the task's vma and may give
852 int in_gate_area_no_task(unsigned long addr)
854 return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END);
857 const char *arch_vma_name(struct vm_area_struct *vma)
859 if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
861 if (vma == &gate_vma)
866 #ifdef CONFIG_SPARSEMEM_VMEMMAP
868 * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
870 static long __meminitdata addr_start, addr_end;
871 static void __meminitdata *p_start, *p_end;
872 static int __meminitdata node_start;
875 vmemmap_populate(struct page *start_page, unsigned long size, int node)
877 unsigned long addr = (unsigned long)start_page;
878 unsigned long end = (unsigned long)(start_page + size);
884 for (; addr < end; addr = next) {
887 pgd = vmemmap_pgd_populate(addr, node);
891 pud = vmemmap_pud_populate(pgd, addr, node);
896 next = (addr + PAGE_SIZE) & PAGE_MASK;
897 pmd = vmemmap_pmd_populate(pud, addr, node);
902 p = vmemmap_pte_populate(pmd, addr, node);
907 addr_end = addr + PAGE_SIZE;
908 p_end = p + PAGE_SIZE;
910 next = pmd_addr_end(addr, end);
912 pmd = pmd_offset(pud, addr);
913 if (pmd_none(*pmd)) {
916 p = vmemmap_alloc_block(PMD_SIZE, node);
920 entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
922 set_pmd(pmd, __pmd(pte_val(entry)));
924 /* check to see if we have contiguous blocks */
925 if (p_end != p || node_start != node) {
927 printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n",
928 addr_start, addr_end-1, p_start, p_end-1, node_start);
934 addr_end = addr + PMD_SIZE;
935 p_end = p + PMD_SIZE;
937 vmemmap_verify((pte_t *)pmd, node, addr, next);
944 void __meminit vmemmap_populate_print_last(void)
947 printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n",
948 addr_start, addr_end-1, p_start, p_end-1, node_start);