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>
53 static unsigned long dma_reserve __initdata
;
55 static int __init
parse_direct_gbpages_off(char *arg
)
60 early_param("nogbpages", parse_direct_gbpages_off
);
62 static int __init
parse_direct_gbpages_on(char *arg
)
67 early_param("gbpages", parse_direct_gbpages_on
);
70 * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the
71 * physical space so we can cache the place of the first one and move
72 * around without checking the pgd every time.
75 pteval_t __supported_pte_mask __read_mostly
= ~_PAGE_IOMAP
;
76 EXPORT_SYMBOL_GPL(__supported_pte_mask
);
78 int force_personality32
;
82 * Control non executable heap for 32bit processes.
83 * To control the stack too use noexec=off
85 * on PROT_READ does not imply PROT_EXEC for 32-bit processes (default)
86 * off PROT_READ implies PROT_EXEC
88 static int __init
nonx32_setup(char *str
)
90 if (!strcmp(str
, "on"))
91 force_personality32
&= ~READ_IMPLIES_EXEC
;
92 else if (!strcmp(str
, "off"))
93 force_personality32
|= READ_IMPLIES_EXEC
;
96 __setup("noexec32=", nonx32_setup
);
99 * NOTE: This function is marked __ref because it calls __init function
100 * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0.
102 static __ref
void *spp_getpage(void)
107 ptr
= (void *) get_zeroed_page(GFP_ATOMIC
| __GFP_NOTRACK
);
109 ptr
= alloc_bootmem_pages(PAGE_SIZE
);
111 if (!ptr
|| ((unsigned long)ptr
& ~PAGE_MASK
)) {
112 panic("set_pte_phys: cannot allocate page data %s\n",
113 after_bootmem
? "after bootmem" : "");
116 pr_debug("spp_getpage %p\n", ptr
);
121 static pud_t
*fill_pud(pgd_t
*pgd
, unsigned long vaddr
)
123 if (pgd_none(*pgd
)) {
124 pud_t
*pud
= (pud_t
*)spp_getpage();
125 pgd_populate(&init_mm
, pgd
, pud
);
126 if (pud
!= pud_offset(pgd
, 0))
127 printk(KERN_ERR
"PAGETABLE BUG #00! %p <-> %p\n",
128 pud
, pud_offset(pgd
, 0));
130 return pud_offset(pgd
, vaddr
);
133 static pmd_t
*fill_pmd(pud_t
*pud
, unsigned long vaddr
)
135 if (pud_none(*pud
)) {
136 pmd_t
*pmd
= (pmd_t
*) spp_getpage();
137 pud_populate(&init_mm
, pud
, pmd
);
138 if (pmd
!= pmd_offset(pud
, 0))
139 printk(KERN_ERR
"PAGETABLE BUG #01! %p <-> %p\n",
140 pmd
, pmd_offset(pud
, 0));
142 return pmd_offset(pud
, vaddr
);
145 static pte_t
*fill_pte(pmd_t
*pmd
, unsigned long vaddr
)
147 if (pmd_none(*pmd
)) {
148 pte_t
*pte
= (pte_t
*) spp_getpage();
149 pmd_populate_kernel(&init_mm
, pmd
, pte
);
150 if (pte
!= pte_offset_kernel(pmd
, 0))
151 printk(KERN_ERR
"PAGETABLE BUG #02!\n");
153 return pte_offset_kernel(pmd
, vaddr
);
156 void set_pte_vaddr_pud(pud_t
*pud_page
, unsigned long vaddr
, pte_t new_pte
)
162 pud
= pud_page
+ pud_index(vaddr
);
163 pmd
= fill_pmd(pud
, vaddr
);
164 pte
= fill_pte(pmd
, vaddr
);
166 set_pte(pte
, new_pte
);
169 * It's enough to flush this one mapping.
170 * (PGE mappings get flushed as well)
172 __flush_tlb_one(vaddr
);
175 void set_pte_vaddr(unsigned long vaddr
, pte_t pteval
)
180 pr_debug("set_pte_vaddr %lx to %lx\n", vaddr
, native_pte_val(pteval
));
182 pgd
= pgd_offset_k(vaddr
);
183 if (pgd_none(*pgd
)) {
185 "PGD FIXMAP MISSING, it should be setup in head.S!\n");
188 pud_page
= (pud_t
*)pgd_page_vaddr(*pgd
);
189 set_pte_vaddr_pud(pud_page
, vaddr
, pteval
);
192 pmd_t
* __init
populate_extra_pmd(unsigned long vaddr
)
197 pgd
= pgd_offset_k(vaddr
);
198 pud
= fill_pud(pgd
, vaddr
);
199 return fill_pmd(pud
, vaddr
);
202 pte_t
* __init
populate_extra_pte(unsigned long vaddr
)
206 pmd
= populate_extra_pmd(vaddr
);
207 return fill_pte(pmd
, vaddr
);
211 * Create large page table mappings for a range of physical addresses.
213 static void __init
__init_extra_mapping(unsigned long phys
, unsigned long size
,
220 BUG_ON((phys
& ~PMD_MASK
) || (size
& ~PMD_MASK
));
221 for (; size
; phys
+= PMD_SIZE
, size
-= PMD_SIZE
) {
222 pgd
= pgd_offset_k((unsigned long)__va(phys
));
223 if (pgd_none(*pgd
)) {
224 pud
= (pud_t
*) spp_getpage();
225 set_pgd(pgd
, __pgd(__pa(pud
) | _KERNPG_TABLE
|
228 pud
= pud_offset(pgd
, (unsigned long)__va(phys
));
229 if (pud_none(*pud
)) {
230 pmd
= (pmd_t
*) spp_getpage();
231 set_pud(pud
, __pud(__pa(pmd
) | _KERNPG_TABLE
|
234 pmd
= pmd_offset(pud
, phys
);
235 BUG_ON(!pmd_none(*pmd
));
236 set_pmd(pmd
, __pmd(phys
| pgprot_val(prot
)));
240 void __init
init_extra_mapping_wb(unsigned long phys
, unsigned long size
)
242 __init_extra_mapping(phys
, size
, PAGE_KERNEL_LARGE
);
245 void __init
init_extra_mapping_uc(unsigned long phys
, unsigned long size
)
247 __init_extra_mapping(phys
, size
, PAGE_KERNEL_LARGE_NOCACHE
);
251 * The head.S code sets up the kernel high mapping:
253 * from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
255 * phys_addr holds the negative offset to the kernel, which is added
256 * to the compile time generated pmds. This results in invalid pmds up
257 * to the point where we hit the physaddr 0 mapping.
259 * We limit the mappings to the region from _text to _end. _end is
260 * rounded up to the 2MB boundary. This catches the invalid pmds as
261 * well, as they are located before _text:
263 void __init
cleanup_highmap(void)
265 unsigned long vaddr
= __START_KERNEL_map
;
266 unsigned long end
= roundup((unsigned long)_end
, PMD_SIZE
) - 1;
267 pmd_t
*pmd
= level2_kernel_pgt
;
268 pmd_t
*last_pmd
= pmd
+ PTRS_PER_PMD
;
270 for (; pmd
< last_pmd
; pmd
++, vaddr
+= PMD_SIZE
) {
273 if (vaddr
< (unsigned long) _text
|| vaddr
> end
)
274 set_pmd(pmd
, __pmd(0));
278 static __ref
void *alloc_low_page(unsigned long *phys
)
280 unsigned long pfn
= e820_table_end
++;
284 adr
= (void *)get_zeroed_page(GFP_ATOMIC
| __GFP_NOTRACK
);
290 if (pfn
>= e820_table_top
)
291 panic("alloc_low_page: ran out of memory");
293 adr
= early_memremap(pfn
* PAGE_SIZE
, PAGE_SIZE
);
294 memset(adr
, 0, PAGE_SIZE
);
295 *phys
= pfn
* PAGE_SIZE
;
299 static __ref
void unmap_low_page(void *adr
)
304 early_iounmap(adr
, PAGE_SIZE
);
307 static unsigned long __meminit
308 phys_pte_init(pte_t
*pte_page
, unsigned long addr
, unsigned long end
,
312 unsigned long last_map_addr
= end
;
315 pte_t
*pte
= pte_page
+ pte_index(addr
);
317 for(i
= pte_index(addr
); i
< PTRS_PER_PTE
; i
++, addr
+= PAGE_SIZE
, pte
++) {
320 if (!after_bootmem
) {
321 for(; i
< PTRS_PER_PTE
; i
++, pte
++)
322 set_pte(pte
, __pte(0));
328 * We will re-use the existing mapping.
329 * Xen for example has some special requirements, like mapping
330 * pagetable pages as RO. So assume someone who pre-setup
331 * these mappings are more intelligent.
339 printk(" pte=%p addr=%lx pte=%016lx\n",
340 pte
, addr
, pfn_pte(addr
>> PAGE_SHIFT
, PAGE_KERNEL
).pte
);
342 set_pte(pte
, pfn_pte(addr
>> PAGE_SHIFT
, prot
));
343 last_map_addr
= (addr
& PAGE_MASK
) + PAGE_SIZE
;
346 update_page_count(PG_LEVEL_4K
, pages
);
348 return last_map_addr
;
351 static unsigned long __meminit
352 phys_pte_update(pmd_t
*pmd
, unsigned long address
, unsigned long end
,
355 pte_t
*pte
= (pte_t
*)pmd_page_vaddr(*pmd
);
357 return phys_pte_init(pte
, address
, end
, prot
);
360 static unsigned long __meminit
361 phys_pmd_init(pmd_t
*pmd_page
, unsigned long address
, unsigned long end
,
362 unsigned long page_size_mask
, pgprot_t prot
)
364 unsigned long pages
= 0;
365 unsigned long last_map_addr
= end
;
367 int i
= pmd_index(address
);
369 for (; i
< PTRS_PER_PMD
; i
++, address
+= PMD_SIZE
) {
370 unsigned long pte_phys
;
371 pmd_t
*pmd
= pmd_page
+ pmd_index(address
);
373 pgprot_t new_prot
= prot
;
375 if (address
>= end
) {
376 if (!after_bootmem
) {
377 for (; i
< PTRS_PER_PMD
; i
++, pmd
++)
378 set_pmd(pmd
, __pmd(0));
384 if (!pmd_large(*pmd
)) {
385 spin_lock(&init_mm
.page_table_lock
);
386 last_map_addr
= phys_pte_update(pmd
, address
,
388 spin_unlock(&init_mm
.page_table_lock
);
392 * If we are ok with PG_LEVEL_2M mapping, then we will
393 * use the existing mapping,
395 * Otherwise, we will split the large page mapping but
396 * use the same existing protection bits except for
397 * large page, so that we don't violate Intel's TLB
398 * Application note (317080) which says, while changing
399 * the page sizes, new and old translations should
400 * not differ with respect to page frame and
403 if (page_size_mask
& (1 << PG_LEVEL_2M
)) {
407 new_prot
= pte_pgprot(pte_clrhuge(*(pte_t
*)pmd
));
410 if (page_size_mask
& (1<<PG_LEVEL_2M
)) {
412 spin_lock(&init_mm
.page_table_lock
);
413 set_pte((pte_t
*)pmd
,
414 pfn_pte(address
>> PAGE_SHIFT
,
415 __pgprot(pgprot_val(prot
) | _PAGE_PSE
)));
416 spin_unlock(&init_mm
.page_table_lock
);
417 last_map_addr
= (address
& PMD_MASK
) + PMD_SIZE
;
421 pte
= alloc_low_page(&pte_phys
);
422 last_map_addr
= phys_pte_init(pte
, address
, end
, new_prot
);
425 spin_lock(&init_mm
.page_table_lock
);
426 pmd_populate_kernel(&init_mm
, pmd
, __va(pte_phys
));
427 spin_unlock(&init_mm
.page_table_lock
);
429 update_page_count(PG_LEVEL_2M
, pages
);
430 return last_map_addr
;
433 static unsigned long __meminit
434 phys_pmd_update(pud_t
*pud
, unsigned long address
, unsigned long end
,
435 unsigned long page_size_mask
, pgprot_t prot
)
437 pmd_t
*pmd
= pmd_offset(pud
, 0);
438 unsigned long last_map_addr
;
440 last_map_addr
= phys_pmd_init(pmd
, address
, end
, page_size_mask
, prot
);
442 return last_map_addr
;
445 static unsigned long __meminit
446 phys_pud_init(pud_t
*pud_page
, unsigned long addr
, unsigned long end
,
447 unsigned long page_size_mask
)
449 unsigned long pages
= 0;
450 unsigned long last_map_addr
= end
;
451 int i
= pud_index(addr
);
453 for (; i
< PTRS_PER_PUD
; i
++, addr
= (addr
& PUD_MASK
) + PUD_SIZE
) {
454 unsigned long pmd_phys
;
455 pud_t
*pud
= pud_page
+ pud_index(addr
);
457 pgprot_t prot
= PAGE_KERNEL
;
462 if (!after_bootmem
&&
463 !e820_any_mapped(addr
, addr
+PUD_SIZE
, 0)) {
464 set_pud(pud
, __pud(0));
469 if (!pud_large(*pud
)) {
470 last_map_addr
= phys_pmd_update(pud
, addr
, end
,
471 page_size_mask
, prot
);
475 * If we are ok with PG_LEVEL_1G mapping, then we will
476 * use the existing mapping.
478 * Otherwise, we will split the gbpage mapping but use
479 * the same existing protection bits except for large
480 * page, so that we don't violate Intel's TLB
481 * Application note (317080) which says, while changing
482 * the page sizes, new and old translations should
483 * not differ with respect to page frame and
486 if (page_size_mask
& (1 << PG_LEVEL_1G
)) {
490 prot
= pte_pgprot(pte_clrhuge(*(pte_t
*)pud
));
493 if (page_size_mask
& (1<<PG_LEVEL_1G
)) {
495 spin_lock(&init_mm
.page_table_lock
);
496 set_pte((pte_t
*)pud
,
497 pfn_pte(addr
>> PAGE_SHIFT
, PAGE_KERNEL_LARGE
));
498 spin_unlock(&init_mm
.page_table_lock
);
499 last_map_addr
= (addr
& PUD_MASK
) + PUD_SIZE
;
503 pmd
= alloc_low_page(&pmd_phys
);
504 last_map_addr
= phys_pmd_init(pmd
, addr
, end
, page_size_mask
,
508 spin_lock(&init_mm
.page_table_lock
);
509 pud_populate(&init_mm
, pud
, __va(pmd_phys
));
510 spin_unlock(&init_mm
.page_table_lock
);
514 update_page_count(PG_LEVEL_1G
, pages
);
516 return last_map_addr
;
519 static unsigned long __meminit
520 phys_pud_update(pgd_t
*pgd
, unsigned long addr
, unsigned long end
,
521 unsigned long page_size_mask
)
525 pud
= (pud_t
*)pgd_page_vaddr(*pgd
);
527 return phys_pud_init(pud
, addr
, end
, page_size_mask
);
530 unsigned long __meminit
531 kernel_physical_mapping_init(unsigned long start
,
533 unsigned long page_size_mask
)
536 unsigned long next
, last_map_addr
= end
;
538 start
= (unsigned long)__va(start
);
539 end
= (unsigned long)__va(end
);
541 for (; start
< end
; start
= next
) {
542 pgd_t
*pgd
= pgd_offset_k(start
);
543 unsigned long pud_phys
;
546 next
= (start
+ PGDIR_SIZE
) & PGDIR_MASK
;
551 last_map_addr
= phys_pud_update(pgd
, __pa(start
),
552 __pa(end
), page_size_mask
);
556 pud
= alloc_low_page(&pud_phys
);
557 last_map_addr
= phys_pud_init(pud
, __pa(start
), __pa(next
),
561 spin_lock(&init_mm
.page_table_lock
);
562 pgd_populate(&init_mm
, pgd
, __va(pud_phys
));
563 spin_unlock(&init_mm
.page_table_lock
);
567 return last_map_addr
;
571 void __init
initmem_init(unsigned long start_pfn
, unsigned long end_pfn
)
573 unsigned long bootmap_size
, bootmap
;
575 bootmap_size
= bootmem_bootmap_pages(end_pfn
)<<PAGE_SHIFT
;
576 bootmap
= find_e820_area(0, end_pfn
<<PAGE_SHIFT
, bootmap_size
,
579 panic("Cannot find bootmem map of size %ld\n", bootmap_size
);
580 /* don't touch min_low_pfn */
581 bootmap_size
= init_bootmem_node(NODE_DATA(0), bootmap
>> PAGE_SHIFT
,
583 e820_register_active_regions(0, start_pfn
, end_pfn
);
584 free_bootmem_with_active_regions(0, end_pfn
);
585 early_res_to_bootmem(0, end_pfn
<<PAGE_SHIFT
);
586 reserve_bootmem(bootmap
, bootmap_size
, BOOTMEM_DEFAULT
);
590 void __init
paging_init(void)
592 unsigned long max_zone_pfns
[MAX_NR_ZONES
];
594 memset(max_zone_pfns
, 0, sizeof(max_zone_pfns
));
595 max_zone_pfns
[ZONE_DMA
] = MAX_DMA_PFN
;
596 max_zone_pfns
[ZONE_DMA32
] = MAX_DMA32_PFN
;
597 max_zone_pfns
[ZONE_NORMAL
] = max_pfn
;
599 sparse_memory_present_with_active_regions(MAX_NUMNODES
);
601 free_area_init_nodes(max_zone_pfns
);
605 * Memory hotplug specific functions
607 #ifdef CONFIG_MEMORY_HOTPLUG
609 * Memory is added always to NORMAL zone. This means you will never get
610 * additional DMA/DMA32 memory.
612 int arch_add_memory(int nid
, u64 start
, u64 size
)
614 struct pglist_data
*pgdat
= NODE_DATA(nid
);
615 struct zone
*zone
= pgdat
->node_zones
+ ZONE_NORMAL
;
616 unsigned long last_mapped_pfn
, start_pfn
= start
>> PAGE_SHIFT
;
617 unsigned long nr_pages
= size
>> PAGE_SHIFT
;
620 last_mapped_pfn
= init_memory_mapping(start
, start
+ size
);
621 if (last_mapped_pfn
> max_pfn_mapped
)
622 max_pfn_mapped
= last_mapped_pfn
;
624 ret
= __add_pages(nid
, zone
, start_pfn
, nr_pages
);
629 EXPORT_SYMBOL_GPL(arch_add_memory
);
631 #if !defined(CONFIG_ACPI_NUMA) && defined(CONFIG_NUMA)
632 int memory_add_physaddr_to_nid(u64 start
)
636 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid
);
639 #endif /* CONFIG_MEMORY_HOTPLUG */
641 static struct kcore_list kcore_mem
, kcore_vmalloc
, kcore_kernel
,
642 kcore_modules
, kcore_vsyscall
;
644 void __init
mem_init(void)
646 long codesize
, reservedpages
, datasize
, initsize
;
647 unsigned long absent_pages
;
651 /* clear_bss() already clear the empty_zero_page */
655 /* this will put all low memory onto the freelists */
657 totalram_pages
= numa_free_all_bootmem();
659 totalram_pages
= free_all_bootmem();
662 absent_pages
= absent_pages_in_range(0, max_pfn
);
663 reservedpages
= max_pfn
- totalram_pages
- absent_pages
;
666 codesize
= (unsigned long) &_etext
- (unsigned long) &_text
;
667 datasize
= (unsigned long) &_edata
- (unsigned long) &_etext
;
668 initsize
= (unsigned long) &__init_end
- (unsigned long) &__init_begin
;
670 /* Register memory areas for /proc/kcore */
671 kclist_add(&kcore_mem
, __va(0), max_low_pfn
<< PAGE_SHIFT
);
672 kclist_add(&kcore_vmalloc
, (void *)VMALLOC_START
,
673 VMALLOC_END
-VMALLOC_START
);
674 kclist_add(&kcore_kernel
, &_stext
, _end
- _stext
);
675 kclist_add(&kcore_modules
, (void *)MODULES_VADDR
, MODULES_LEN
);
676 kclist_add(&kcore_vsyscall
, (void *)VSYSCALL_START
,
677 VSYSCALL_END
- VSYSCALL_START
);
679 printk(KERN_INFO
"Memory: %luk/%luk available (%ldk kernel code, "
680 "%ldk absent, %ldk reserved, %ldk data, %ldk init)\n",
681 (unsigned long) nr_free_pages() << (PAGE_SHIFT
-10),
682 max_pfn
<< (PAGE_SHIFT
-10),
684 absent_pages
<< (PAGE_SHIFT
-10),
685 reservedpages
<< (PAGE_SHIFT
-10),
690 #ifdef CONFIG_DEBUG_RODATA
691 const int rodata_test_data
= 0xC3;
692 EXPORT_SYMBOL_GPL(rodata_test_data
);
694 static int kernel_set_to_readonly
;
696 void set_kernel_text_rw(void)
698 unsigned long start
= PFN_ALIGN(_stext
);
699 unsigned long end
= PFN_ALIGN(__start_rodata
);
701 if (!kernel_set_to_readonly
)
704 pr_debug("Set kernel text: %lx - %lx for read write\n",
707 set_memory_rw(start
, (end
- start
) >> PAGE_SHIFT
);
710 void set_kernel_text_ro(void)
712 unsigned long start
= PFN_ALIGN(_stext
);
713 unsigned long end
= PFN_ALIGN(__start_rodata
);
715 if (!kernel_set_to_readonly
)
718 pr_debug("Set kernel text: %lx - %lx for read only\n",
721 set_memory_ro(start
, (end
- start
) >> PAGE_SHIFT
);
724 void mark_rodata_ro(void)
726 unsigned long start
= PFN_ALIGN(_stext
), end
= PFN_ALIGN(__end_rodata
);
727 unsigned long rodata_start
=
728 ((unsigned long)__start_rodata
+ PAGE_SIZE
- 1) & PAGE_MASK
;
730 printk(KERN_INFO
"Write protecting the kernel read-only data: %luk\n",
731 (end
- start
) >> 10);
732 set_memory_ro(start
, (end
- start
) >> PAGE_SHIFT
);
734 kernel_set_to_readonly
= 1;
737 * The rodata section (but not the kernel text!) should also be
740 set_memory_nx(rodata_start
, (end
- rodata_start
) >> PAGE_SHIFT
);
744 #ifdef CONFIG_CPA_DEBUG
745 printk(KERN_INFO
"Testing CPA: undo %lx-%lx\n", start
, end
);
746 set_memory_rw(start
, (end
-start
) >> PAGE_SHIFT
);
748 printk(KERN_INFO
"Testing CPA: again\n");
749 set_memory_ro(start
, (end
-start
) >> PAGE_SHIFT
);
755 int __init
reserve_bootmem_generic(unsigned long phys
, unsigned long len
,
762 unsigned long pfn
= phys
>> PAGE_SHIFT
;
764 if (pfn
>= max_pfn
) {
766 * This can happen with kdump kernels when accessing
769 if (pfn
< max_pfn_mapped
)
772 printk(KERN_ERR
"reserve_bootmem: illegal reserve %lx %lu\n",
777 /* Should check here against the e820 map to avoid double free */
779 nid
= phys_to_nid(phys
);
780 next_nid
= phys_to_nid(phys
+ len
- 1);
782 ret
= reserve_bootmem_node(NODE_DATA(nid
), phys
, len
, flags
);
784 ret
= reserve_bootmem(phys
, len
, flags
);
790 reserve_bootmem(phys
, len
, BOOTMEM_DEFAULT
);
793 if (phys
+len
<= MAX_DMA_PFN
*PAGE_SIZE
) {
794 dma_reserve
+= len
/ PAGE_SIZE
;
795 set_dma_reserve(dma_reserve
);
801 int kern_addr_valid(unsigned long addr
)
803 unsigned long above
= ((long)addr
) >> __VIRTUAL_MASK_SHIFT
;
809 if (above
!= 0 && above
!= -1UL)
812 pgd
= pgd_offset_k(addr
);
816 pud
= pud_offset(pgd
, addr
);
820 pmd
= pmd_offset(pud
, addr
);
825 return pfn_valid(pmd_pfn(*pmd
));
827 pte
= pte_offset_kernel(pmd
, addr
);
831 return pfn_valid(pte_pfn(*pte
));
835 * A pseudo VMA to allow ptrace access for the vsyscall page. This only
836 * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
837 * not need special handling anymore:
839 static struct vm_area_struct gate_vma
= {
840 .vm_start
= VSYSCALL_START
,
841 .vm_end
= VSYSCALL_START
+ (VSYSCALL_MAPPED_PAGES
* PAGE_SIZE
),
842 .vm_page_prot
= PAGE_READONLY_EXEC
,
843 .vm_flags
= VM_READ
| VM_EXEC
846 struct vm_area_struct
*get_gate_vma(struct task_struct
*tsk
)
848 #ifdef CONFIG_IA32_EMULATION
849 if (test_tsk_thread_flag(tsk
, TIF_IA32
))
855 int in_gate_area(struct task_struct
*task
, unsigned long addr
)
857 struct vm_area_struct
*vma
= get_gate_vma(task
);
862 return (addr
>= vma
->vm_start
) && (addr
< vma
->vm_end
);
866 * Use this when you have no reliable task/vma, typically from interrupt
867 * context. It is less reliable than using the task's vma and may give
870 int in_gate_area_no_task(unsigned long addr
)
872 return (addr
>= VSYSCALL_START
) && (addr
< VSYSCALL_END
);
875 const char *arch_vma_name(struct vm_area_struct
*vma
)
877 if (vma
->vm_mm
&& vma
->vm_start
== (long)vma
->vm_mm
->context
.vdso
)
879 if (vma
== &gate_vma
)
884 #ifdef CONFIG_SPARSEMEM_VMEMMAP
886 * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
888 static long __meminitdata addr_start
, addr_end
;
889 static void __meminitdata
*p_start
, *p_end
;
890 static int __meminitdata node_start
;
893 vmemmap_populate(struct page
*start_page
, unsigned long size
, int node
)
895 unsigned long addr
= (unsigned long)start_page
;
896 unsigned long end
= (unsigned long)(start_page
+ size
);
902 for (; addr
< end
; addr
= next
) {
905 pgd
= vmemmap_pgd_populate(addr
, node
);
909 pud
= vmemmap_pud_populate(pgd
, addr
, node
);
914 next
= (addr
+ PAGE_SIZE
) & PAGE_MASK
;
915 pmd
= vmemmap_pmd_populate(pud
, addr
, node
);
920 p
= vmemmap_pte_populate(pmd
, addr
, node
);
925 addr_end
= addr
+ PAGE_SIZE
;
926 p_end
= p
+ PAGE_SIZE
;
928 next
= pmd_addr_end(addr
, end
);
930 pmd
= pmd_offset(pud
, addr
);
931 if (pmd_none(*pmd
)) {
934 p
= vmemmap_alloc_block(PMD_SIZE
, node
);
938 entry
= pfn_pte(__pa(p
) >> PAGE_SHIFT
,
940 set_pmd(pmd
, __pmd(pte_val(entry
)));
942 /* check to see if we have contiguous blocks */
943 if (p_end
!= p
|| node_start
!= node
) {
945 printk(KERN_DEBUG
" [%lx-%lx] PMD -> [%p-%p] on node %d\n",
946 addr_start
, addr_end
-1, p_start
, p_end
-1, node_start
);
952 addr_end
= addr
+ PMD_SIZE
;
953 p_end
= p
+ PMD_SIZE
;
955 vmemmap_verify((pte_t
*)pmd
, node
, addr
, next
);
962 void __meminit
vmemmap_populate_print_last(void)
965 printk(KERN_DEBUG
" [%lx-%lx] PMD -> [%p-%p] on node %d\n",
966 addr_start
, addr_end
-1, p_start
, p_end
-1, node_start
);