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/memblock.h>
25 #include <linux/proc_fs.h>
26 #include <linux/pci.h>
27 #include <linux/pfn.h>
28 #include <linux/poison.h>
29 #include <linux/dma-mapping.h>
30 #include <linux/module.h>
31 #include <linux/memory.h>
32 #include <linux/memory_hotplug.h>
33 #include <linux/nmi.h>
34 #include <linux/gfp.h>
36 #include <asm/processor.h>
37 #include <asm/bios_ebda.h>
38 #include <asm/uaccess.h>
39 #include <asm/pgtable.h>
40 #include <asm/pgalloc.h>
42 #include <asm/fixmap.h>
46 #include <asm/mmu_context.h>
47 #include <asm/proto.h>
49 #include <asm/sections.h>
50 #include <asm/kdebug.h>
52 #include <asm/cacheflush.h>
54 #include <asm/uv/uv.h>
55 #include <asm/setup.h>
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 * When memory was added/removed make sure all the processes MM have
102 * suitable PGD entries in the local PGD level page.
104 void sync_global_pgds(unsigned long start, unsigned long end)
106 unsigned long address;
108 for (address = start; address <= end; address += PGDIR_SIZE) {
109 const pgd_t *pgd_ref = pgd_offset_k(address);
112 if (pgd_none(*pgd_ref))
115 spin_lock(&pgd_lock);
116 list_for_each_entry(page, &pgd_list, lru) {
118 spinlock_t *pgt_lock;
120 pgd = (pgd_t *)page_address(page) + pgd_index(address);
121 /* the pgt_lock only for Xen */
122 pgt_lock = &pgd_page_get_mm(page)->page_table_lock;
126 set_pgd(pgd, *pgd_ref);
128 BUG_ON(pgd_page_vaddr(*pgd)
129 != pgd_page_vaddr(*pgd_ref));
131 spin_unlock(pgt_lock);
133 spin_unlock(&pgd_lock);
138 * NOTE: This function is marked __ref because it calls __init function
139 * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0.
141 static __ref void *spp_getpage(void)
146 ptr = (void *) get_zeroed_page(GFP_ATOMIC | __GFP_NOTRACK);
148 ptr = alloc_bootmem_pages(PAGE_SIZE);
150 if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) {
151 panic("set_pte_phys: cannot allocate page data %s\n",
152 after_bootmem ? "after bootmem" : "");
155 pr_debug("spp_getpage %p\n", ptr);
160 static pud_t *fill_pud(pgd_t *pgd, unsigned long vaddr)
162 if (pgd_none(*pgd)) {
163 pud_t *pud = (pud_t *)spp_getpage();
164 pgd_populate(&init_mm, pgd, pud);
165 if (pud != pud_offset(pgd, 0))
166 printk(KERN_ERR "PAGETABLE BUG #00! %p <-> %p\n",
167 pud, pud_offset(pgd, 0));
169 return pud_offset(pgd, vaddr);
172 static pmd_t *fill_pmd(pud_t *pud, unsigned long vaddr)
174 if (pud_none(*pud)) {
175 pmd_t *pmd = (pmd_t *) spp_getpage();
176 pud_populate(&init_mm, pud, pmd);
177 if (pmd != pmd_offset(pud, 0))
178 printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n",
179 pmd, pmd_offset(pud, 0));
181 return pmd_offset(pud, vaddr);
184 static pte_t *fill_pte(pmd_t *pmd, unsigned long vaddr)
186 if (pmd_none(*pmd)) {
187 pte_t *pte = (pte_t *) spp_getpage();
188 pmd_populate_kernel(&init_mm, pmd, pte);
189 if (pte != pte_offset_kernel(pmd, 0))
190 printk(KERN_ERR "PAGETABLE BUG #02!\n");
192 return pte_offset_kernel(pmd, vaddr);
195 void set_pte_vaddr_pud(pud_t *pud_page, unsigned long vaddr, pte_t new_pte)
201 pud = pud_page + pud_index(vaddr);
202 pmd = fill_pmd(pud, vaddr);
203 pte = fill_pte(pmd, vaddr);
205 set_pte(pte, new_pte);
208 * It's enough to flush this one mapping.
209 * (PGE mappings get flushed as well)
211 __flush_tlb_one(vaddr);
214 void set_pte_vaddr(unsigned long vaddr, pte_t pteval)
219 pr_debug("set_pte_vaddr %lx to %lx\n", vaddr, native_pte_val(pteval));
221 pgd = pgd_offset_k(vaddr);
222 if (pgd_none(*pgd)) {
224 "PGD FIXMAP MISSING, it should be setup in head.S!\n");
227 pud_page = (pud_t*)pgd_page_vaddr(*pgd);
228 set_pte_vaddr_pud(pud_page, vaddr, pteval);
231 pmd_t * __init populate_extra_pmd(unsigned long vaddr)
236 pgd = pgd_offset_k(vaddr);
237 pud = fill_pud(pgd, vaddr);
238 return fill_pmd(pud, vaddr);
241 pte_t * __init populate_extra_pte(unsigned long vaddr)
245 pmd = populate_extra_pmd(vaddr);
246 return fill_pte(pmd, vaddr);
250 * Create large page table mappings for a range of physical addresses.
252 static void __init __init_extra_mapping(unsigned long phys, unsigned long size,
259 BUG_ON((phys & ~PMD_MASK) || (size & ~PMD_MASK));
260 for (; size; phys += PMD_SIZE, size -= PMD_SIZE) {
261 pgd = pgd_offset_k((unsigned long)__va(phys));
262 if (pgd_none(*pgd)) {
263 pud = (pud_t *) spp_getpage();
264 set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE |
267 pud = pud_offset(pgd, (unsigned long)__va(phys));
268 if (pud_none(*pud)) {
269 pmd = (pmd_t *) spp_getpage();
270 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE |
273 pmd = pmd_offset(pud, phys);
274 BUG_ON(!pmd_none(*pmd));
275 set_pmd(pmd, __pmd(phys | pgprot_val(prot)));
279 void __init init_extra_mapping_wb(unsigned long phys, unsigned long size)
281 __init_extra_mapping(phys, size, PAGE_KERNEL_LARGE);
284 void __init init_extra_mapping_uc(unsigned long phys, unsigned long size)
286 __init_extra_mapping(phys, size, PAGE_KERNEL_LARGE_NOCACHE);
290 * The head.S code sets up the kernel high mapping:
292 * from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
294 * phys_addr holds the negative offset to the kernel, which is added
295 * to the compile time generated pmds. This results in invalid pmds up
296 * to the point where we hit the physaddr 0 mapping.
298 * We limit the mappings to the region from _text to _brk_end. _brk_end
299 * is rounded up to the 2MB boundary. This catches the invalid pmds as
300 * well, as they are located before _text:
302 void __init cleanup_highmap(void)
304 unsigned long vaddr = __START_KERNEL_map;
305 unsigned long vaddr_end = __START_KERNEL_map + (max_pfn_mapped << PAGE_SHIFT);
306 unsigned long end = roundup((unsigned long)_brk_end, PMD_SIZE) - 1;
307 pmd_t *pmd = level2_kernel_pgt;
309 for (; vaddr + PMD_SIZE - 1 < vaddr_end; pmd++, vaddr += PMD_SIZE) {
312 if (vaddr < (unsigned long) _text || vaddr > end)
313 set_pmd(pmd, __pmd(0));
317 static __ref void *alloc_low_page(void)
323 adr = (void *)get_zeroed_page(GFP_ATOMIC | __GFP_NOTRACK);
328 if ((pgt_buf_end + 1) >= pgt_buf_top) {
330 if (min_pfn_mapped >= max_pfn_mapped)
331 panic("alloc_low_page: ran out of memory");
332 ret = memblock_find_in_range(min_pfn_mapped << PAGE_SHIFT,
333 max_pfn_mapped << PAGE_SHIFT,
334 PAGE_SIZE, PAGE_SIZE);
336 panic("alloc_low_page: can not alloc memory");
337 memblock_reserve(ret, PAGE_SIZE);
338 pfn = ret >> PAGE_SHIFT;
342 adr = __va(pfn * PAGE_SIZE);
347 static unsigned long __meminit
348 phys_pte_init(pte_t *pte_page, unsigned long addr, unsigned long end,
351 unsigned long pages = 0, next;
352 unsigned long last_map_addr = end;
355 pte_t *pte = pte_page + pte_index(addr);
357 for (i = pte_index(addr); i < PTRS_PER_PTE; i++, addr = next, pte++) {
358 next = (addr & PAGE_MASK) + PAGE_SIZE;
360 if (!after_bootmem &&
361 !e820_any_mapped(addr & PAGE_MASK, next, E820_RAM) &&
362 !e820_any_mapped(addr & PAGE_MASK, next, E820_RESERVED_KERN))
363 set_pte(pte, __pte(0));
368 * We will re-use the existing mapping.
369 * Xen for example has some special requirements, like mapping
370 * pagetable pages as RO. So assume someone who pre-setup
371 * these mappings are more intelligent.
380 printk(" pte=%p addr=%lx pte=%016lx\n",
381 pte, addr, pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL).pte);
383 set_pte(pte, pfn_pte(addr >> PAGE_SHIFT, prot));
384 last_map_addr = (addr & PAGE_MASK) + PAGE_SIZE;
387 update_page_count(PG_LEVEL_4K, pages);
389 return last_map_addr;
392 static unsigned long __meminit
393 phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end,
394 unsigned long page_size_mask, pgprot_t prot)
396 unsigned long pages = 0, next;
397 unsigned long last_map_addr = end;
399 int i = pmd_index(address);
401 for (; i < PTRS_PER_PMD; i++, address = next) {
402 pmd_t *pmd = pmd_page + pmd_index(address);
404 pgprot_t new_prot = prot;
406 next = (address & PMD_MASK) + PMD_SIZE;
407 if (address >= end) {
408 if (!after_bootmem &&
409 !e820_any_mapped(address & PMD_MASK, next, E820_RAM) &&
410 !e820_any_mapped(address & PMD_MASK, next, E820_RESERVED_KERN))
411 set_pmd(pmd, __pmd(0));
416 if (!pmd_large(*pmd)) {
417 spin_lock(&init_mm.page_table_lock);
418 pte = (pte_t *)pmd_page_vaddr(*pmd);
419 last_map_addr = phys_pte_init(pte, address,
421 spin_unlock(&init_mm.page_table_lock);
425 * If we are ok with PG_LEVEL_2M mapping, then we will
426 * use the existing mapping,
428 * Otherwise, we will split the large page mapping but
429 * use the same existing protection bits except for
430 * large page, so that we don't violate Intel's TLB
431 * Application note (317080) which says, while changing
432 * the page sizes, new and old translations should
433 * not differ with respect to page frame and
436 if (page_size_mask & (1 << PG_LEVEL_2M)) {
439 last_map_addr = next;
442 new_prot = pte_pgprot(pte_clrhuge(*(pte_t *)pmd));
445 if (page_size_mask & (1<<PG_LEVEL_2M)) {
447 spin_lock(&init_mm.page_table_lock);
448 set_pte((pte_t *)pmd,
449 pfn_pte((address & PMD_MASK) >> PAGE_SHIFT,
450 __pgprot(pgprot_val(prot) | _PAGE_PSE)));
451 spin_unlock(&init_mm.page_table_lock);
452 last_map_addr = next;
456 pte = alloc_low_page();
457 last_map_addr = phys_pte_init(pte, address, end, new_prot);
459 spin_lock(&init_mm.page_table_lock);
460 pmd_populate_kernel(&init_mm, pmd, pte);
461 spin_unlock(&init_mm.page_table_lock);
463 update_page_count(PG_LEVEL_2M, pages);
464 return last_map_addr;
467 static unsigned long __meminit
468 phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end,
469 unsigned long page_size_mask)
471 unsigned long pages = 0, next;
472 unsigned long last_map_addr = end;
473 int i = pud_index(addr);
475 for (; i < PTRS_PER_PUD; i++, addr = next) {
476 pud_t *pud = pud_page + pud_index(addr);
478 pgprot_t prot = PAGE_KERNEL;
480 next = (addr & PUD_MASK) + PUD_SIZE;
482 if (!after_bootmem &&
483 !e820_any_mapped(addr & PUD_MASK, next, E820_RAM) &&
484 !e820_any_mapped(addr & PUD_MASK, next, E820_RESERVED_KERN))
485 set_pud(pud, __pud(0));
490 if (!pud_large(*pud)) {
491 pmd = pmd_offset(pud, 0);
492 last_map_addr = phys_pmd_init(pmd, addr, end,
493 page_size_mask, prot);
498 * If we are ok with PG_LEVEL_1G mapping, then we will
499 * use the existing mapping.
501 * Otherwise, we will split the gbpage mapping but use
502 * the same existing protection bits except for large
503 * page, so that we don't violate Intel's TLB
504 * Application note (317080) which says, while changing
505 * the page sizes, new and old translations should
506 * not differ with respect to page frame and
509 if (page_size_mask & (1 << PG_LEVEL_1G)) {
512 last_map_addr = next;
515 prot = pte_pgprot(pte_clrhuge(*(pte_t *)pud));
518 if (page_size_mask & (1<<PG_LEVEL_1G)) {
520 spin_lock(&init_mm.page_table_lock);
521 set_pte((pte_t *)pud,
522 pfn_pte((addr & PUD_MASK) >> PAGE_SHIFT,
524 spin_unlock(&init_mm.page_table_lock);
525 last_map_addr = next;
529 pmd = alloc_low_page();
530 last_map_addr = phys_pmd_init(pmd, addr, end, page_size_mask,
533 spin_lock(&init_mm.page_table_lock);
534 pud_populate(&init_mm, pud, pmd);
535 spin_unlock(&init_mm.page_table_lock);
539 update_page_count(PG_LEVEL_1G, pages);
541 return last_map_addr;
544 unsigned long __meminit
545 kernel_physical_mapping_init(unsigned long start,
547 unsigned long page_size_mask)
549 bool pgd_changed = false;
550 unsigned long next, last_map_addr = end;
553 start = (unsigned long)__va(start);
554 end = (unsigned long)__va(end);
557 for (; start < end; start = next) {
558 pgd_t *pgd = pgd_offset_k(start);
561 next = (start + PGDIR_SIZE) & PGDIR_MASK;
566 pud = (pud_t *)pgd_page_vaddr(*pgd);
567 last_map_addr = phys_pud_init(pud, __pa(start),
568 __pa(end), page_size_mask);
572 pud = alloc_low_page();
573 last_map_addr = phys_pud_init(pud, __pa(start), __pa(next),
576 spin_lock(&init_mm.page_table_lock);
577 pgd_populate(&init_mm, pgd, pud);
578 spin_unlock(&init_mm.page_table_lock);
583 sync_global_pgds(addr, end);
587 return last_map_addr;
591 void __init initmem_init(void)
593 memblock_set_node(0, (phys_addr_t)ULLONG_MAX, 0);
597 void __init paging_init(void)
599 sparse_memory_present_with_active_regions(MAX_NUMNODES);
603 * clear the default setting with node 0
604 * note: don't use nodes_clear here, that is really clearing when
605 * numa support is not compiled in, and later node_set_state
606 * will not set it back.
608 node_clear_state(0, N_NORMAL_MEMORY);
614 * Memory hotplug specific functions
616 #ifdef CONFIG_MEMORY_HOTPLUG
618 * After memory hotplug the variables max_pfn, max_low_pfn and high_memory need
621 static void update_end_of_memory_vars(u64 start, u64 size)
623 unsigned long end_pfn = PFN_UP(start + size);
625 if (end_pfn > max_pfn) {
627 max_low_pfn = end_pfn;
628 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
633 * Memory is added always to NORMAL zone. This means you will never get
634 * additional DMA/DMA32 memory.
636 int arch_add_memory(int nid, u64 start, u64 size)
638 struct pglist_data *pgdat = NODE_DATA(nid);
639 struct zone *zone = pgdat->node_zones + ZONE_NORMAL;
640 unsigned long start_pfn = start >> PAGE_SHIFT;
641 unsigned long nr_pages = size >> PAGE_SHIFT;
644 init_memory_mapping(start, start + size);
646 ret = __add_pages(nid, zone, start_pfn, nr_pages);
649 /* update max_pfn, max_low_pfn and high_memory */
650 update_end_of_memory_vars(start, size);
654 EXPORT_SYMBOL_GPL(arch_add_memory);
656 #endif /* CONFIG_MEMORY_HOTPLUG */
658 static struct kcore_list kcore_vsyscall;
660 void __init mem_init(void)
662 long codesize, reservedpages, datasize, initsize;
663 unsigned long absent_pages;
667 /* clear_bss() already clear the empty_zero_page */
671 /* this will put all low memory onto the freelists */
673 totalram_pages = numa_free_all_bootmem();
675 totalram_pages = free_all_bootmem();
678 absent_pages = absent_pages_in_range(0, max_pfn);
679 reservedpages = max_pfn - totalram_pages - absent_pages;
682 codesize = (unsigned long) &_etext - (unsigned long) &_text;
683 datasize = (unsigned long) &_edata - (unsigned long) &_etext;
684 initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin;
686 /* Register memory areas for /proc/kcore */
687 kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START,
688 VSYSCALL_END - VSYSCALL_START, KCORE_OTHER);
690 printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, "
691 "%ldk absent, %ldk reserved, %ldk data, %ldk init)\n",
692 nr_free_pages() << (PAGE_SHIFT-10),
693 max_pfn << (PAGE_SHIFT-10),
695 absent_pages << (PAGE_SHIFT-10),
696 reservedpages << (PAGE_SHIFT-10),
701 #ifdef CONFIG_DEBUG_RODATA
702 const int rodata_test_data = 0xC3;
703 EXPORT_SYMBOL_GPL(rodata_test_data);
705 int kernel_set_to_readonly;
707 void set_kernel_text_rw(void)
709 unsigned long start = PFN_ALIGN(_text);
710 unsigned long end = PFN_ALIGN(__stop___ex_table);
712 if (!kernel_set_to_readonly)
715 pr_debug("Set kernel text: %lx - %lx for read write\n",
719 * Make the kernel identity mapping for text RW. Kernel text
720 * mapping will always be RO. Refer to the comment in
721 * static_protections() in pageattr.c
723 set_memory_rw(start, (end - start) >> PAGE_SHIFT);
726 void set_kernel_text_ro(void)
728 unsigned long start = PFN_ALIGN(_text);
729 unsigned long end = PFN_ALIGN(__stop___ex_table);
731 if (!kernel_set_to_readonly)
734 pr_debug("Set kernel text: %lx - %lx for read only\n",
738 * Set the kernel identity mapping for text RO.
740 set_memory_ro(start, (end - start) >> PAGE_SHIFT);
743 void mark_rodata_ro(void)
745 unsigned long start = PFN_ALIGN(_text);
746 unsigned long rodata_start =
747 ((unsigned long)__start_rodata + PAGE_SIZE - 1) & PAGE_MASK;
748 unsigned long end = (unsigned long) &__end_rodata_hpage_align;
749 unsigned long text_end = PAGE_ALIGN((unsigned long) &__stop___ex_table);
750 unsigned long rodata_end = PAGE_ALIGN((unsigned long) &__end_rodata);
751 unsigned long data_start = (unsigned long) &_sdata;
753 printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
754 (end - start) >> 10);
755 set_memory_ro(start, (end - start) >> PAGE_SHIFT);
757 kernel_set_to_readonly = 1;
760 * The rodata section (but not the kernel text!) should also be
763 set_memory_nx(rodata_start, (end - rodata_start) >> PAGE_SHIFT);
767 #ifdef CONFIG_CPA_DEBUG
768 printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end);
769 set_memory_rw(start, (end-start) >> PAGE_SHIFT);
771 printk(KERN_INFO "Testing CPA: again\n");
772 set_memory_ro(start, (end-start) >> PAGE_SHIFT);
775 free_init_pages("unused kernel memory",
776 (unsigned long) page_address(virt_to_page(text_end)),
778 page_address(virt_to_page(rodata_start)));
779 free_init_pages("unused kernel memory",
780 (unsigned long) page_address(virt_to_page(rodata_end)),
781 (unsigned long) page_address(virt_to_page(data_start)));
786 int kern_addr_valid(unsigned long addr)
788 unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
794 if (above != 0 && above != -1UL)
797 pgd = pgd_offset_k(addr);
801 pud = pud_offset(pgd, addr);
805 pmd = pmd_offset(pud, addr);
810 return pfn_valid(pmd_pfn(*pmd));
812 pte = pte_offset_kernel(pmd, addr);
816 return pfn_valid(pte_pfn(*pte));
820 * A pseudo VMA to allow ptrace access for the vsyscall page. This only
821 * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
822 * not need special handling anymore:
824 static struct vm_area_struct gate_vma = {
825 .vm_start = VSYSCALL_START,
826 .vm_end = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE),
827 .vm_page_prot = PAGE_READONLY_EXEC,
828 .vm_flags = VM_READ | VM_EXEC
831 struct vm_area_struct *get_gate_vma(struct mm_struct *mm)
833 #ifdef CONFIG_IA32_EMULATION
834 if (!mm || mm->context.ia32_compat)
840 int in_gate_area(struct mm_struct *mm, unsigned long addr)
842 struct vm_area_struct *vma = get_gate_vma(mm);
847 return (addr >= vma->vm_start) && (addr < vma->vm_end);
851 * Use this when you have no reliable mm, typically from interrupt
852 * context. It is less reliable than using a task's mm and may give
855 int in_gate_area_no_mm(unsigned long addr)
857 return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END);
860 const char *arch_vma_name(struct vm_area_struct *vma)
862 if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
864 if (vma == &gate_vma)
870 unsigned long memory_block_size_bytes(void)
872 if (is_uv_system()) {
873 printk(KERN_INFO "UV: memory block size 2GB\n");
874 return 2UL * 1024 * 1024 * 1024;
876 return MIN_MEMORY_BLOCK_SIZE;
880 #ifdef CONFIG_SPARSEMEM_VMEMMAP
882 * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
884 static long __meminitdata addr_start, addr_end;
885 static void __meminitdata *p_start, *p_end;
886 static int __meminitdata node_start;
889 vmemmap_populate(struct page *start_page, unsigned long size, int node)
891 unsigned long addr = (unsigned long)start_page;
892 unsigned long end = (unsigned long)(start_page + size);
898 for (; addr < end; addr = next) {
901 pgd = vmemmap_pgd_populate(addr, node);
905 pud = vmemmap_pud_populate(pgd, addr, node);
910 next = (addr + PAGE_SIZE) & PAGE_MASK;
911 pmd = vmemmap_pmd_populate(pud, addr, node);
916 p = vmemmap_pte_populate(pmd, addr, node);
921 addr_end = addr + PAGE_SIZE;
922 p_end = p + PAGE_SIZE;
924 next = pmd_addr_end(addr, end);
926 pmd = pmd_offset(pud, addr);
927 if (pmd_none(*pmd)) {
930 p = vmemmap_alloc_block_buf(PMD_SIZE, node);
934 entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
936 set_pmd(pmd, __pmd(pte_val(entry)));
938 /* check to see if we have contiguous blocks */
939 if (p_end != p || node_start != node) {
941 printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n",
942 addr_start, addr_end-1, p_start, p_end-1, node_start);
948 addr_end = addr + PMD_SIZE;
949 p_end = p + PMD_SIZE;
951 vmemmap_verify((pte_t *)pmd, node, addr, next);
955 sync_global_pgds((unsigned long)start_page, end);
959 void __meminit vmemmap_populate_print_last(void)
962 printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n",
963 addr_start, addr_end-1, p_start, p_end-1, node_start);