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 * end_pfn only includes RAM, while max_pfn_mapped includes all e820 entries.
54 * The direct mapping extends to max_pfn_mapped, so that we can directly access
55 * apertures, ACPI and other tables without having to play with fixmaps.
57 unsigned long max_low_pfn_mapped
;
58 unsigned long max_pfn_mapped
;
60 static unsigned long dma_reserve __initdata
;
62 DEFINE_PER_CPU(struct mmu_gather
, mmu_gathers
);
65 #ifdef CONFIG_DIRECT_GBPAGES
70 static int __init
parse_direct_gbpages_off(char *arg
)
75 early_param("nogbpages", parse_direct_gbpages_off
);
77 static int __init
parse_direct_gbpages_on(char *arg
)
82 early_param("gbpages", parse_direct_gbpages_on
);
85 * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the
86 * physical space so we can cache the place of the first one and move
87 * around without checking the pgd every time.
92 pteval_t __supported_pte_mask __read_mostly
= ~_PAGE_IOMAP
;
93 EXPORT_SYMBOL_GPL(__supported_pte_mask
);
95 static int do_not_nx __cpuinitdata
;
99 * Control non-executable mappings for 64-bit processes.
101 * on Enable (default)
104 static int __init
nonx_setup(char *str
)
108 if (!strncmp(str
, "on", 2)) {
109 __supported_pte_mask
|= _PAGE_NX
;
111 } else if (!strncmp(str
, "off", 3)) {
113 __supported_pte_mask
&= ~_PAGE_NX
;
117 early_param("noexec", nonx_setup
);
119 void __cpuinit
check_efer(void)
123 rdmsrl(MSR_EFER
, efer
);
124 if (!(efer
& EFER_NX
) || do_not_nx
)
125 __supported_pte_mask
&= ~_PAGE_NX
;
128 int force_personality32
;
132 * Control non executable heap for 32bit processes.
133 * To control the stack too use noexec=off
135 * on PROT_READ does not imply PROT_EXEC for 32-bit processes (default)
136 * off PROT_READ implies PROT_EXEC
138 static int __init
nonx32_setup(char *str
)
140 if (!strcmp(str
, "on"))
141 force_personality32
&= ~READ_IMPLIES_EXEC
;
142 else if (!strcmp(str
, "off"))
143 force_personality32
|= READ_IMPLIES_EXEC
;
146 __setup("noexec32=", nonx32_setup
);
149 * NOTE: This function is marked __ref because it calls __init function
150 * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0.
152 static __ref
void *spp_getpage(void)
157 ptr
= (void *) get_zeroed_page(GFP_ATOMIC
);
159 ptr
= alloc_bootmem_pages(PAGE_SIZE
);
161 if (!ptr
|| ((unsigned long)ptr
& ~PAGE_MASK
)) {
162 panic("set_pte_phys: cannot allocate page data %s\n",
163 after_bootmem
? "after bootmem" : "");
166 pr_debug("spp_getpage %p\n", ptr
);
172 set_pte_vaddr_pud(pud_t
*pud_page
, unsigned long vaddr
, pte_t new_pte
)
178 pud
= pud_page
+ pud_index(vaddr
);
179 if (pud_none(*pud
)) {
180 pmd
= (pmd_t
*) spp_getpage();
181 pud_populate(&init_mm
, pud
, pmd
);
182 if (pmd
!= pmd_offset(pud
, 0)) {
183 printk(KERN_ERR
"PAGETABLE BUG #01! %p <-> %p\n",
184 pmd
, pmd_offset(pud
, 0));
188 pmd
= pmd_offset(pud
, vaddr
);
189 if (pmd_none(*pmd
)) {
190 pte
= (pte_t
*) spp_getpage();
191 pmd_populate_kernel(&init_mm
, pmd
, pte
);
192 if (pte
!= pte_offset_kernel(pmd
, 0)) {
193 printk(KERN_ERR
"PAGETABLE BUG #02!\n");
198 pte
= pte_offset_kernel(pmd
, vaddr
);
199 set_pte(pte
, new_pte
);
202 * It's enough to flush this one mapping.
203 * (PGE mappings get flushed as well)
205 __flush_tlb_one(vaddr
);
209 set_pte_vaddr(unsigned long vaddr
, pte_t pteval
)
214 pr_debug("set_pte_vaddr %lx to %lx\n", vaddr
, native_pte_val(pteval
));
216 pgd
= pgd_offset_k(vaddr
);
217 if (pgd_none(*pgd
)) {
219 "PGD FIXMAP MISSING, it should be setup in head.S!\n");
222 pud_page
= (pud_t
*)pgd_page_vaddr(*pgd
);
223 set_pte_vaddr_pud(pud_page
, vaddr
, pteval
);
227 * Create large page table mappings for a range of physical addresses.
229 static void __init
__init_extra_mapping(unsigned long phys
, unsigned long size
,
236 BUG_ON((phys
& ~PMD_MASK
) || (size
& ~PMD_MASK
));
237 for (; size
; phys
+= PMD_SIZE
, size
-= PMD_SIZE
) {
238 pgd
= pgd_offset_k((unsigned long)__va(phys
));
239 if (pgd_none(*pgd
)) {
240 pud
= (pud_t
*) spp_getpage();
241 set_pgd(pgd
, __pgd(__pa(pud
) | _KERNPG_TABLE
|
244 pud
= pud_offset(pgd
, (unsigned long)__va(phys
));
245 if (pud_none(*pud
)) {
246 pmd
= (pmd_t
*) spp_getpage();
247 set_pud(pud
, __pud(__pa(pmd
) | _KERNPG_TABLE
|
250 pmd
= pmd_offset(pud
, phys
);
251 BUG_ON(!pmd_none(*pmd
));
252 set_pmd(pmd
, __pmd(phys
| pgprot_val(prot
)));
256 void __init
init_extra_mapping_wb(unsigned long phys
, unsigned long size
)
258 __init_extra_mapping(phys
, size
, PAGE_KERNEL_LARGE
);
261 void __init
init_extra_mapping_uc(unsigned long phys
, unsigned long size
)
263 __init_extra_mapping(phys
, size
, PAGE_KERNEL_LARGE_NOCACHE
);
267 * The head.S code sets up the kernel high mapping:
269 * from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
271 * phys_addr holds the negative offset to the kernel, which is added
272 * to the compile time generated pmds. This results in invalid pmds up
273 * to the point where we hit the physaddr 0 mapping.
275 * We limit the mappings to the region from _text to _end. _end is
276 * rounded up to the 2MB boundary. This catches the invalid pmds as
277 * well, as they are located before _text:
279 void __init
cleanup_highmap(void)
281 unsigned long vaddr
= __START_KERNEL_map
;
282 unsigned long end
= roundup((unsigned long)_end
, PMD_SIZE
) - 1;
283 pmd_t
*pmd
= level2_kernel_pgt
;
284 pmd_t
*last_pmd
= pmd
+ PTRS_PER_PMD
;
286 for (; pmd
< last_pmd
; pmd
++, vaddr
+= PMD_SIZE
) {
289 if (vaddr
< (unsigned long) _text
|| vaddr
> end
)
290 set_pmd(pmd
, __pmd(0));
294 static unsigned long __initdata table_start
;
295 static unsigned long __meminitdata table_end
;
296 static unsigned long __meminitdata table_top
;
298 static __ref
void *alloc_low_page(unsigned long *phys
)
300 unsigned long pfn
= table_end
++;
304 adr
= (void *)get_zeroed_page(GFP_ATOMIC
);
310 if (pfn
>= table_top
)
311 panic("alloc_low_page: ran out of memory");
313 adr
= early_memremap(pfn
* PAGE_SIZE
, PAGE_SIZE
);
314 memset(adr
, 0, PAGE_SIZE
);
315 *phys
= pfn
* PAGE_SIZE
;
319 static __ref
void unmap_low_page(void *adr
)
324 early_iounmap(adr
, PAGE_SIZE
);
327 static unsigned long __meminit
328 phys_pte_init(pte_t
*pte_page
, unsigned long addr
, unsigned long end
,
332 unsigned long last_map_addr
= end
;
335 pte_t
*pte
= pte_page
+ pte_index(addr
);
337 for(i
= pte_index(addr
); i
< PTRS_PER_PTE
; i
++, addr
+= PAGE_SIZE
, pte
++) {
340 if (!after_bootmem
) {
341 for(; i
< PTRS_PER_PTE
; i
++, pte
++)
342 set_pte(pte
, __pte(0));
348 * We will re-use the existing mapping.
349 * Xen for example has some special requirements, like mapping
350 * pagetable pages as RO. So assume someone who pre-setup
351 * these mappings are more intelligent.
359 printk(" pte=%p addr=%lx pte=%016lx\n",
360 pte
, addr
, pfn_pte(addr
>> PAGE_SHIFT
, PAGE_KERNEL
).pte
);
362 set_pte(pte
, pfn_pte(addr
>> PAGE_SHIFT
, prot
));
363 last_map_addr
= (addr
& PAGE_MASK
) + PAGE_SIZE
;
366 update_page_count(PG_LEVEL_4K
, pages
);
368 return last_map_addr
;
371 static unsigned long __meminit
372 phys_pte_update(pmd_t
*pmd
, unsigned long address
, unsigned long end
,
375 pte_t
*pte
= (pte_t
*)pmd_page_vaddr(*pmd
);
377 return phys_pte_init(pte
, address
, end
, prot
);
380 static unsigned long __meminit
381 phys_pmd_init(pmd_t
*pmd_page
, unsigned long address
, unsigned long end
,
382 unsigned long page_size_mask
, pgprot_t prot
)
384 unsigned long pages
= 0;
385 unsigned long last_map_addr
= end
;
387 int i
= pmd_index(address
);
389 for (; i
< PTRS_PER_PMD
; i
++, address
+= PMD_SIZE
) {
390 unsigned long pte_phys
;
391 pmd_t
*pmd
= pmd_page
+ pmd_index(address
);
393 pgprot_t new_prot
= prot
;
395 if (address
>= end
) {
396 if (!after_bootmem
) {
397 for (; i
< PTRS_PER_PMD
; i
++, pmd
++)
398 set_pmd(pmd
, __pmd(0));
404 if (!pmd_large(*pmd
)) {
405 spin_lock(&init_mm
.page_table_lock
);
406 last_map_addr
= phys_pte_update(pmd
, address
,
408 spin_unlock(&init_mm
.page_table_lock
);
412 * If we are ok with PG_LEVEL_2M mapping, then we will
413 * use the existing mapping,
415 * Otherwise, we will split the large page mapping but
416 * use the same existing protection bits except for
417 * large page, so that we don't violate Intel's TLB
418 * Application note (317080) which says, while changing
419 * the page sizes, new and old translations should
420 * not differ with respect to page frame and
423 if (page_size_mask
& (1 << PG_LEVEL_2M
)) {
427 new_prot
= pte_pgprot(pte_clrhuge(*(pte_t
*)pmd
));
430 if (page_size_mask
& (1<<PG_LEVEL_2M
)) {
432 spin_lock(&init_mm
.page_table_lock
);
433 set_pte((pte_t
*)pmd
,
434 pfn_pte(address
>> PAGE_SHIFT
,
435 __pgprot(pgprot_val(prot
) | _PAGE_PSE
)));
436 spin_unlock(&init_mm
.page_table_lock
);
437 last_map_addr
= (address
& PMD_MASK
) + PMD_SIZE
;
441 pte
= alloc_low_page(&pte_phys
);
442 last_map_addr
= phys_pte_init(pte
, address
, end
, new_prot
);
445 spin_lock(&init_mm
.page_table_lock
);
446 pmd_populate_kernel(&init_mm
, pmd
, __va(pte_phys
));
447 spin_unlock(&init_mm
.page_table_lock
);
449 update_page_count(PG_LEVEL_2M
, pages
);
450 return last_map_addr
;
453 static unsigned long __meminit
454 phys_pmd_update(pud_t
*pud
, unsigned long address
, unsigned long end
,
455 unsigned long page_size_mask
, pgprot_t prot
)
457 pmd_t
*pmd
= pmd_offset(pud
, 0);
458 unsigned long last_map_addr
;
460 last_map_addr
= phys_pmd_init(pmd
, address
, end
, page_size_mask
, prot
);
462 return last_map_addr
;
465 static unsigned long __meminit
466 phys_pud_init(pud_t
*pud_page
, unsigned long addr
, unsigned long end
,
467 unsigned long page_size_mask
)
469 unsigned long pages
= 0;
470 unsigned long last_map_addr
= end
;
471 int i
= pud_index(addr
);
473 for (; i
< PTRS_PER_PUD
; i
++, addr
= (addr
& PUD_MASK
) + PUD_SIZE
) {
474 unsigned long pmd_phys
;
475 pud_t
*pud
= pud_page
+ pud_index(addr
);
477 pgprot_t prot
= PAGE_KERNEL
;
482 if (!after_bootmem
&&
483 !e820_any_mapped(addr
, addr
+PUD_SIZE
, 0)) {
484 set_pud(pud
, __pud(0));
489 if (!pud_large(*pud
)) {
490 last_map_addr
= phys_pmd_update(pud
, addr
, end
,
491 page_size_mask
, prot
);
495 * If we are ok with PG_LEVEL_1G mapping, then we will
496 * use the existing mapping.
498 * Otherwise, we will split the gbpage mapping but use
499 * the same existing protection bits except for large
500 * page, so that we don't violate Intel's TLB
501 * Application note (317080) which says, while changing
502 * the page sizes, new and old translations should
503 * not differ with respect to page frame and
506 if (page_size_mask
& (1 << PG_LEVEL_1G
)) {
510 prot
= pte_pgprot(pte_clrhuge(*(pte_t
*)pud
));
513 if (page_size_mask
& (1<<PG_LEVEL_1G
)) {
515 spin_lock(&init_mm
.page_table_lock
);
516 set_pte((pte_t
*)pud
,
517 pfn_pte(addr
>> PAGE_SHIFT
, PAGE_KERNEL_LARGE
));
518 spin_unlock(&init_mm
.page_table_lock
);
519 last_map_addr
= (addr
& PUD_MASK
) + PUD_SIZE
;
523 pmd
= alloc_low_page(&pmd_phys
);
524 last_map_addr
= phys_pmd_init(pmd
, addr
, end
, page_size_mask
,
528 spin_lock(&init_mm
.page_table_lock
);
529 pud_populate(&init_mm
, pud
, __va(pmd_phys
));
530 spin_unlock(&init_mm
.page_table_lock
);
534 update_page_count(PG_LEVEL_1G
, pages
);
536 return last_map_addr
;
539 static unsigned long __meminit
540 phys_pud_update(pgd_t
*pgd
, unsigned long addr
, unsigned long end
,
541 unsigned long page_size_mask
)
545 pud
= (pud_t
*)pgd_page_vaddr(*pgd
);
547 return phys_pud_init(pud
, addr
, end
, page_size_mask
);
550 static void __init
find_early_table_space(unsigned long end
, int use_pse
,
553 unsigned long puds
, pmds
, ptes
, tables
, start
;
555 puds
= (end
+ PUD_SIZE
- 1) >> PUD_SHIFT
;
556 tables
= roundup(puds
* sizeof(pud_t
), PAGE_SIZE
);
559 extra
= end
- ((end
>>PUD_SHIFT
) << PUD_SHIFT
);
560 pmds
= (extra
+ PMD_SIZE
- 1) >> PMD_SHIFT
;
562 pmds
= (end
+ PMD_SIZE
- 1) >> PMD_SHIFT
;
563 tables
+= roundup(pmds
* sizeof(pmd_t
), PAGE_SIZE
);
567 extra
= end
- ((end
>>PMD_SHIFT
) << PMD_SHIFT
);
568 ptes
= (extra
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
570 ptes
= (end
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
571 tables
+= roundup(ptes
* sizeof(pte_t
), PAGE_SIZE
);
574 * RED-PEN putting page tables only on node 0 could
575 * cause a hotspot and fill up ZONE_DMA. The page tables
576 * need roughly 0.5KB per GB.
579 table_start
= find_e820_area(start
, end
, tables
, PAGE_SIZE
);
580 if (table_start
== -1UL)
581 panic("Cannot find space for the kernel page tables");
583 table_start
>>= PAGE_SHIFT
;
584 table_end
= table_start
;
585 table_top
= table_start
+ (tables
>> PAGE_SHIFT
);
587 printk(KERN_DEBUG
"kernel direct mapping tables up to %lx @ %lx-%lx\n",
588 end
, table_start
<< PAGE_SHIFT
, table_top
<< PAGE_SHIFT
);
591 static void __init
init_gbpages(void)
593 if (direct_gbpages
&& cpu_has_gbpages
)
594 printk(KERN_INFO
"Using GB pages for direct mapping\n");
599 static unsigned long __init
kernel_physical_mapping_init(unsigned long start
,
601 unsigned long page_size_mask
)
604 unsigned long next
, last_map_addr
= end
;
606 start
= (unsigned long)__va(start
);
607 end
= (unsigned long)__va(end
);
609 for (; start
< end
; start
= next
) {
610 pgd_t
*pgd
= pgd_offset_k(start
);
611 unsigned long pud_phys
;
614 next
= (start
+ PGDIR_SIZE
) & PGDIR_MASK
;
619 last_map_addr
= phys_pud_update(pgd
, __pa(start
),
620 __pa(end
), page_size_mask
);
624 pud
= alloc_low_page(&pud_phys
);
625 last_map_addr
= phys_pud_init(pud
, __pa(start
), __pa(next
),
629 spin_lock(&init_mm
.page_table_lock
);
630 pgd_populate(&init_mm
, pgd
, __va(pud_phys
));
631 spin_unlock(&init_mm
.page_table_lock
);
635 return last_map_addr
;
641 unsigned page_size_mask
;
644 #define NR_RANGE_MR 5
646 static int save_mr(struct map_range
*mr
, int nr_range
,
647 unsigned long start_pfn
, unsigned long end_pfn
,
648 unsigned long page_size_mask
)
651 if (start_pfn
< end_pfn
) {
652 if (nr_range
>= NR_RANGE_MR
)
653 panic("run out of range for init_memory_mapping\n");
654 mr
[nr_range
].start
= start_pfn
<<PAGE_SHIFT
;
655 mr
[nr_range
].end
= end_pfn
<<PAGE_SHIFT
;
656 mr
[nr_range
].page_size_mask
= page_size_mask
;
664 * Setup the direct mapping of the physical memory at PAGE_OFFSET.
665 * This runs before bootmem is initialized and gets pages directly from
666 * the physical memory. To access them they are temporarily mapped.
668 unsigned long __init_refok
init_memory_mapping(unsigned long start
,
671 unsigned long last_map_addr
= 0;
672 unsigned long page_size_mask
= 0;
673 unsigned long start_pfn
, end_pfn
;
675 struct map_range mr
[NR_RANGE_MR
];
677 int use_pse
, use_gbpages
;
679 printk(KERN_INFO
"init_memory_mapping\n");
682 * Find space for the kernel direct mapping tables.
684 * Later we should allocate these tables in the local node of the
685 * memory mapped. Unfortunately this is done currently before the
686 * nodes are discovered.
691 #ifdef CONFIG_DEBUG_PAGEALLOC
693 * For CONFIG_DEBUG_PAGEALLOC, identity mapping will use small pages.
694 * This will simplify cpa(), which otherwise needs to support splitting
695 * large pages into small in interrupt context, etc.
697 use_pse
= use_gbpages
= 0;
699 use_pse
= cpu_has_pse
;
700 use_gbpages
= direct_gbpages
;
704 page_size_mask
|= 1 << PG_LEVEL_1G
;
706 page_size_mask
|= 1 << PG_LEVEL_2M
;
708 memset(mr
, 0, sizeof(mr
));
711 /* head if not big page alignment ?*/
712 start_pfn
= start
>> PAGE_SHIFT
;
713 end_pfn
= ((start
+ (PMD_SIZE
- 1)) >> PMD_SHIFT
)
714 << (PMD_SHIFT
- PAGE_SHIFT
);
715 nr_range
= save_mr(mr
, nr_range
, start_pfn
, end_pfn
, 0);
717 /* big page (2M) range*/
718 start_pfn
= ((start
+ (PMD_SIZE
- 1))>>PMD_SHIFT
)
719 << (PMD_SHIFT
- PAGE_SHIFT
);
720 end_pfn
= ((start
+ (PUD_SIZE
- 1))>>PUD_SHIFT
)
721 << (PUD_SHIFT
- PAGE_SHIFT
);
722 if (end_pfn
> ((end
>>PUD_SHIFT
)<<(PUD_SHIFT
- PAGE_SHIFT
)))
723 end_pfn
= ((end
>>PUD_SHIFT
)<<(PUD_SHIFT
- PAGE_SHIFT
));
724 nr_range
= save_mr(mr
, nr_range
, start_pfn
, end_pfn
,
725 page_size_mask
& (1<<PG_LEVEL_2M
));
727 /* big page (1G) range */
729 end_pfn
= (end
>>PUD_SHIFT
) << (PUD_SHIFT
- PAGE_SHIFT
);
730 nr_range
= save_mr(mr
, nr_range
, start_pfn
, end_pfn
,
732 ((1<<PG_LEVEL_2M
)|(1<<PG_LEVEL_1G
)));
734 /* tail is not big page (1G) alignment */
736 end_pfn
= (end
>>PMD_SHIFT
) << (PMD_SHIFT
- PAGE_SHIFT
);
737 nr_range
= save_mr(mr
, nr_range
, start_pfn
, end_pfn
,
738 page_size_mask
& (1<<PG_LEVEL_2M
));
740 /* tail is not big page (2M) alignment */
742 end_pfn
= end
>>PAGE_SHIFT
;
743 nr_range
= save_mr(mr
, nr_range
, start_pfn
, end_pfn
, 0);
745 /* try to merge same page size and continuous */
746 for (i
= 0; nr_range
> 1 && i
< nr_range
- 1; i
++) {
747 unsigned long old_start
;
748 if (mr
[i
].end
!= mr
[i
+1].start
||
749 mr
[i
].page_size_mask
!= mr
[i
+1].page_size_mask
)
752 old_start
= mr
[i
].start
;
753 memmove(&mr
[i
], &mr
[i
+1],
754 (nr_range
- 1 - i
) * sizeof (struct map_range
));
755 mr
[i
--].start
= old_start
;
759 for (i
= 0; i
< nr_range
; i
++)
760 printk(KERN_DEBUG
" %010lx - %010lx page %s\n",
761 mr
[i
].start
, mr
[i
].end
,
762 (mr
[i
].page_size_mask
& (1<<PG_LEVEL_1G
))?"1G":(
763 (mr
[i
].page_size_mask
& (1<<PG_LEVEL_2M
))?"2M":"4k"));
766 find_early_table_space(end
, use_pse
, use_gbpages
);
768 for (i
= 0; i
< nr_range
; i
++)
769 last_map_addr
= kernel_physical_mapping_init(
770 mr
[i
].start
, mr
[i
].end
,
771 mr
[i
].page_size_mask
);
774 mmu_cr4_features
= read_cr4();
777 if (!after_bootmem
&& table_end
> table_start
)
778 reserve_early(table_start
<< PAGE_SHIFT
,
779 table_end
<< PAGE_SHIFT
, "PGTABLE");
781 printk(KERN_INFO
"last_map_addr: %lx end: %lx\n",
785 early_memtest(start
, end
);
787 return last_map_addr
>> PAGE_SHIFT
;
791 void __init
initmem_init(unsigned long start_pfn
, unsigned long end_pfn
)
793 unsigned long bootmap_size
, bootmap
;
795 bootmap_size
= bootmem_bootmap_pages(end_pfn
)<<PAGE_SHIFT
;
796 bootmap
= find_e820_area(0, end_pfn
<<PAGE_SHIFT
, bootmap_size
,
799 panic("Cannot find bootmem map of size %ld\n", bootmap_size
);
800 /* don't touch min_low_pfn */
801 bootmap_size
= init_bootmem_node(NODE_DATA(0), bootmap
>> PAGE_SHIFT
,
803 e820_register_active_regions(0, start_pfn
, end_pfn
);
804 free_bootmem_with_active_regions(0, end_pfn
);
805 early_res_to_bootmem(0, end_pfn
<<PAGE_SHIFT
);
806 reserve_bootmem(bootmap
, bootmap_size
, BOOTMEM_DEFAULT
);
809 void __init
paging_init(void)
811 unsigned long max_zone_pfns
[MAX_NR_ZONES
];
813 memset(max_zone_pfns
, 0, sizeof(max_zone_pfns
));
814 max_zone_pfns
[ZONE_DMA
] = MAX_DMA_PFN
;
815 max_zone_pfns
[ZONE_DMA32
] = MAX_DMA32_PFN
;
816 max_zone_pfns
[ZONE_NORMAL
] = max_pfn
;
818 memory_present(0, 0, max_pfn
);
820 free_area_init_nodes(max_zone_pfns
);
825 * Memory hotplug specific functions
827 #ifdef CONFIG_MEMORY_HOTPLUG
829 * Memory is added always to NORMAL zone. This means you will never get
830 * additional DMA/DMA32 memory.
832 int arch_add_memory(int nid
, u64 start
, u64 size
)
834 struct pglist_data
*pgdat
= NODE_DATA(nid
);
835 struct zone
*zone
= pgdat
->node_zones
+ ZONE_NORMAL
;
836 unsigned long last_mapped_pfn
, start_pfn
= start
>> PAGE_SHIFT
;
837 unsigned long nr_pages
= size
>> PAGE_SHIFT
;
840 last_mapped_pfn
= init_memory_mapping(start
, start
+ size
);
841 if (last_mapped_pfn
> max_pfn_mapped
)
842 max_pfn_mapped
= last_mapped_pfn
;
844 ret
= __add_pages(zone
, start_pfn
, nr_pages
);
849 EXPORT_SYMBOL_GPL(arch_add_memory
);
851 #if !defined(CONFIG_ACPI_NUMA) && defined(CONFIG_NUMA)
852 int memory_add_physaddr_to_nid(u64 start
)
856 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid
);
859 #endif /* CONFIG_MEMORY_HOTPLUG */
862 * devmem_is_allowed() checks to see if /dev/mem access to a certain address
863 * is valid. The argument is a physical page number.
866 * On x86, access has to be given to the first megabyte of ram because that area
867 * contains bios code and data regions used by X and dosemu and similar apps.
868 * Access has to be given to non-kernel-ram areas as well, these contain the PCI
869 * mmio resources as well as potential bios/acpi data regions.
871 int devmem_is_allowed(unsigned long pagenr
)
875 if (!page_is_ram(pagenr
))
881 static struct kcore_list kcore_mem
, kcore_vmalloc
, kcore_kernel
,
882 kcore_modules
, kcore_vsyscall
;
884 void __init
mem_init(void)
886 long codesize
, reservedpages
, datasize
, initsize
;
887 unsigned long absent_pages
;
889 start_periodic_check_for_corruption();
893 /* clear_bss() already clear the empty_zero_page */
897 /* this will put all low memory onto the freelists */
899 totalram_pages
= numa_free_all_bootmem();
901 totalram_pages
= free_all_bootmem();
904 absent_pages
= absent_pages_in_range(0, max_pfn
);
905 reservedpages
= max_pfn
- totalram_pages
- absent_pages
;
908 codesize
= (unsigned long) &_etext
- (unsigned long) &_text
;
909 datasize
= (unsigned long) &_edata
- (unsigned long) &_etext
;
910 initsize
= (unsigned long) &__init_end
- (unsigned long) &__init_begin
;
912 /* Register memory areas for /proc/kcore */
913 kclist_add(&kcore_mem
, __va(0), max_low_pfn
<< PAGE_SHIFT
);
914 kclist_add(&kcore_vmalloc
, (void *)VMALLOC_START
,
915 VMALLOC_END
-VMALLOC_START
);
916 kclist_add(&kcore_kernel
, &_stext
, _end
- _stext
);
917 kclist_add(&kcore_modules
, (void *)MODULES_VADDR
, MODULES_LEN
);
918 kclist_add(&kcore_vsyscall
, (void *)VSYSCALL_START
,
919 VSYSCALL_END
- VSYSCALL_START
);
921 printk(KERN_INFO
"Memory: %luk/%luk available (%ldk kernel code, "
922 "%ldk absent, %ldk reserved, %ldk data, %ldk init)\n",
923 (unsigned long) nr_free_pages() << (PAGE_SHIFT
-10),
924 max_pfn
<< (PAGE_SHIFT
-10),
926 absent_pages
<< (PAGE_SHIFT
-10),
927 reservedpages
<< (PAGE_SHIFT
-10),
932 void free_init_pages(char *what
, unsigned long begin
, unsigned long end
)
934 unsigned long addr
= begin
;
940 * If debugging page accesses then do not free this memory but
941 * mark them not present - any buggy init-section access will
942 * create a kernel page fault:
944 #ifdef CONFIG_DEBUG_PAGEALLOC
945 printk(KERN_INFO
"debug: unmapping init memory %08lx..%08lx\n",
946 begin
, PAGE_ALIGN(end
));
947 set_memory_np(begin
, (end
- begin
) >> PAGE_SHIFT
);
949 printk(KERN_INFO
"Freeing %s: %luk freed\n", what
, (end
- begin
) >> 10);
951 for (; addr
< end
; addr
+= PAGE_SIZE
) {
952 ClearPageReserved(virt_to_page(addr
));
953 init_page_count(virt_to_page(addr
));
954 memset((void *)(addr
& ~(PAGE_SIZE
-1)),
955 POISON_FREE_INITMEM
, PAGE_SIZE
);
962 void free_initmem(void)
964 free_init_pages("unused kernel memory",
965 (unsigned long)(&__init_begin
),
966 (unsigned long)(&__init_end
));
969 #ifdef CONFIG_DEBUG_RODATA
970 const int rodata_test_data
= 0xC3;
971 EXPORT_SYMBOL_GPL(rodata_test_data
);
973 void mark_rodata_ro(void)
975 unsigned long start
= PFN_ALIGN(_stext
), end
= PFN_ALIGN(__end_rodata
);
976 unsigned long rodata_start
=
977 ((unsigned long)__start_rodata
+ PAGE_SIZE
- 1) & PAGE_MASK
;
979 #ifdef CONFIG_DYNAMIC_FTRACE
980 /* Dynamic tracing modifies the kernel text section */
981 start
= rodata_start
;
984 printk(KERN_INFO
"Write protecting the kernel read-only data: %luk\n",
985 (end
- start
) >> 10);
986 set_memory_ro(start
, (end
- start
) >> PAGE_SHIFT
);
989 * The rodata section (but not the kernel text!) should also be
992 set_memory_nx(rodata_start
, (end
- rodata_start
) >> PAGE_SHIFT
);
996 #ifdef CONFIG_CPA_DEBUG
997 printk(KERN_INFO
"Testing CPA: undo %lx-%lx\n", start
, end
);
998 set_memory_rw(start
, (end
-start
) >> PAGE_SHIFT
);
1000 printk(KERN_INFO
"Testing CPA: again\n");
1001 set_memory_ro(start
, (end
-start
) >> PAGE_SHIFT
);
1007 #ifdef CONFIG_BLK_DEV_INITRD
1008 void free_initrd_mem(unsigned long start
, unsigned long end
)
1010 free_init_pages("initrd memory", start
, end
);
1014 int __init
reserve_bootmem_generic(unsigned long phys
, unsigned long len
,
1021 unsigned long pfn
= phys
>> PAGE_SHIFT
;
1023 if (pfn
>= max_pfn
) {
1025 * This can happen with kdump kernels when accessing
1028 if (pfn
< max_pfn_mapped
)
1031 printk(KERN_ERR
"reserve_bootmem: illegal reserve %lx %lu\n",
1036 /* Should check here against the e820 map to avoid double free */
1038 nid
= phys_to_nid(phys
);
1039 next_nid
= phys_to_nid(phys
+ len
- 1);
1040 if (nid
== next_nid
)
1041 ret
= reserve_bootmem_node(NODE_DATA(nid
), phys
, len
, flags
);
1043 ret
= reserve_bootmem(phys
, len
, flags
);
1049 reserve_bootmem(phys
, len
, BOOTMEM_DEFAULT
);
1052 if (phys
+len
<= MAX_DMA_PFN
*PAGE_SIZE
) {
1053 dma_reserve
+= len
/ PAGE_SIZE
;
1054 set_dma_reserve(dma_reserve
);
1060 int kern_addr_valid(unsigned long addr
)
1062 unsigned long above
= ((long)addr
) >> __VIRTUAL_MASK_SHIFT
;
1068 if (above
!= 0 && above
!= -1UL)
1071 pgd
= pgd_offset_k(addr
);
1075 pud
= pud_offset(pgd
, addr
);
1079 pmd
= pmd_offset(pud
, addr
);
1083 if (pmd_large(*pmd
))
1084 return pfn_valid(pmd_pfn(*pmd
));
1086 pte
= pte_offset_kernel(pmd
, addr
);
1090 return pfn_valid(pte_pfn(*pte
));
1094 * A pseudo VMA to allow ptrace access for the vsyscall page. This only
1095 * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
1096 * not need special handling anymore:
1098 static struct vm_area_struct gate_vma
= {
1099 .vm_start
= VSYSCALL_START
,
1100 .vm_end
= VSYSCALL_START
+ (VSYSCALL_MAPPED_PAGES
* PAGE_SIZE
),
1101 .vm_page_prot
= PAGE_READONLY_EXEC
,
1102 .vm_flags
= VM_READ
| VM_EXEC
1105 struct vm_area_struct
*get_gate_vma(struct task_struct
*tsk
)
1107 #ifdef CONFIG_IA32_EMULATION
1108 if (test_tsk_thread_flag(tsk
, TIF_IA32
))
1114 int in_gate_area(struct task_struct
*task
, unsigned long addr
)
1116 struct vm_area_struct
*vma
= get_gate_vma(task
);
1121 return (addr
>= vma
->vm_start
) && (addr
< vma
->vm_end
);
1125 * Use this when you have no reliable task/vma, typically from interrupt
1126 * context. It is less reliable than using the task's vma and may give
1129 int in_gate_area_no_task(unsigned long addr
)
1131 return (addr
>= VSYSCALL_START
) && (addr
< VSYSCALL_END
);
1134 const char *arch_vma_name(struct vm_area_struct
*vma
)
1136 if (vma
->vm_mm
&& vma
->vm_start
== (long)vma
->vm_mm
->context
.vdso
)
1138 if (vma
== &gate_vma
)
1139 return "[vsyscall]";
1143 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1145 * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
1147 static long __meminitdata addr_start
, addr_end
;
1148 static void __meminitdata
*p_start
, *p_end
;
1149 static int __meminitdata node_start
;
1152 vmemmap_populate(struct page
*start_page
, unsigned long size
, int node
)
1154 unsigned long addr
= (unsigned long)start_page
;
1155 unsigned long end
= (unsigned long)(start_page
+ size
);
1161 for (; addr
< end
; addr
= next
) {
1164 pgd
= vmemmap_pgd_populate(addr
, node
);
1168 pud
= vmemmap_pud_populate(pgd
, addr
, node
);
1173 next
= (addr
+ PAGE_SIZE
) & PAGE_MASK
;
1174 pmd
= vmemmap_pmd_populate(pud
, addr
, node
);
1179 p
= vmemmap_pte_populate(pmd
, addr
, node
);
1184 addr_end
= addr
+ PAGE_SIZE
;
1185 p_end
= p
+ PAGE_SIZE
;
1187 next
= pmd_addr_end(addr
, end
);
1189 pmd
= pmd_offset(pud
, addr
);
1190 if (pmd_none(*pmd
)) {
1193 p
= vmemmap_alloc_block(PMD_SIZE
, node
);
1197 entry
= pfn_pte(__pa(p
) >> PAGE_SHIFT
,
1199 set_pmd(pmd
, __pmd(pte_val(entry
)));
1201 /* check to see if we have contiguous blocks */
1202 if (p_end
!= p
|| node_start
!= node
) {
1204 printk(KERN_DEBUG
" [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1205 addr_start
, addr_end
-1, p_start
, p_end
-1, node_start
);
1211 addr_end
= addr
+ PMD_SIZE
;
1212 p_end
= p
+ PMD_SIZE
;
1214 vmemmap_verify((pte_t
*)pmd
, node
, addr
, next
);
1221 void __meminit
vmemmap_populate_print_last(void)
1224 printk(KERN_DEBUG
" [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1225 addr_start
, addr_end
-1, p_start
, p_end
-1, node_start
);