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/system.h>
35 #include <asm/uaccess.h>
36 #include <asm/pgtable.h>
37 #include <asm/pgalloc.h>
39 #include <asm/fixmap.h>
43 #include <asm/mmu_context.h>
44 #include <asm/proto.h>
46 #include <asm/sections.h>
47 #include <asm/kdebug.h>
49 #include <asm/cacheflush.h>
52 * end_pfn only includes RAM, while max_pfn_mapped includes all e820 entries.
53 * The direct mapping extends to max_pfn_mapped, so that we can directly access
54 * apertures, ACPI and other tables without having to play with fixmaps.
56 unsigned long max_low_pfn_mapped
;
57 unsigned long max_pfn_mapped
;
59 static unsigned long dma_reserve __initdata
;
61 DEFINE_PER_CPU(struct mmu_gather
, mmu_gathers
);
63 int direct_gbpages __meminitdata
64 #ifdef CONFIG_DIRECT_GBPAGES
69 static int __init
parse_direct_gbpages_off(char *arg
)
74 early_param("nogbpages", parse_direct_gbpages_off
);
76 static int __init
parse_direct_gbpages_on(char *arg
)
81 early_param("gbpages", parse_direct_gbpages_on
);
84 * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the
85 * physical space so we can cache the place of the first one and move
86 * around without checking the pgd every time.
91 static __init
void *spp_getpage(void)
96 ptr
= (void *) get_zeroed_page(GFP_ATOMIC
);
98 ptr
= alloc_bootmem_pages(PAGE_SIZE
);
100 if (!ptr
|| ((unsigned long)ptr
& ~PAGE_MASK
)) {
101 panic("set_pte_phys: cannot allocate page data %s\n",
102 after_bootmem
? "after bootmem" : "");
105 pr_debug("spp_getpage %p\n", ptr
);
111 set_pte_vaddr_pud(pud_t
*pud_page
, unsigned long vaddr
, pte_t new_pte
)
117 pud
= pud_page
+ pud_index(vaddr
);
118 if (pud_none(*pud
)) {
119 pmd
= (pmd_t
*) spp_getpage();
120 pud_populate(&init_mm
, pud
, pmd
);
121 if (pmd
!= pmd_offset(pud
, 0)) {
122 printk(KERN_ERR
"PAGETABLE BUG #01! %p <-> %p\n",
123 pmd
, pmd_offset(pud
, 0));
127 pmd
= pmd_offset(pud
, vaddr
);
128 if (pmd_none(*pmd
)) {
129 pte
= (pte_t
*) spp_getpage();
130 pmd_populate_kernel(&init_mm
, pmd
, pte
);
131 if (pte
!= pte_offset_kernel(pmd
, 0)) {
132 printk(KERN_ERR
"PAGETABLE BUG #02!\n");
137 pte
= pte_offset_kernel(pmd
, vaddr
);
138 if (!pte_none(*pte
) && pte_val(new_pte
) &&
139 pte_val(*pte
) != (pte_val(new_pte
) & __supported_pte_mask
))
141 set_pte(pte
, new_pte
);
144 * It's enough to flush this one mapping.
145 * (PGE mappings get flushed as well)
147 __flush_tlb_one(vaddr
);
151 set_pte_vaddr(unsigned long vaddr
, pte_t pteval
)
156 pr_debug("set_pte_vaddr %lx to %lx\n", vaddr
, native_pte_val(pteval
));
158 pgd
= pgd_offset_k(vaddr
);
159 if (pgd_none(*pgd
)) {
161 "PGD FIXMAP MISSING, it should be setup in head.S!\n");
164 pud_page
= (pud_t
*)pgd_page_vaddr(*pgd
);
165 set_pte_vaddr_pud(pud_page
, vaddr
, pteval
);
169 * Create large page table mappings for a range of physical addresses.
171 static void __init
__init_extra_mapping(unsigned long phys
, unsigned long size
,
178 BUG_ON((phys
& ~PMD_MASK
) || (size
& ~PMD_MASK
));
179 for (; size
; phys
+= PMD_SIZE
, size
-= PMD_SIZE
) {
180 pgd
= pgd_offset_k((unsigned long)__va(phys
));
181 if (pgd_none(*pgd
)) {
182 pud
= (pud_t
*) spp_getpage();
183 set_pgd(pgd
, __pgd(__pa(pud
) | _KERNPG_TABLE
|
186 pud
= pud_offset(pgd
, (unsigned long)__va(phys
));
187 if (pud_none(*pud
)) {
188 pmd
= (pmd_t
*) spp_getpage();
189 set_pud(pud
, __pud(__pa(pmd
) | _KERNPG_TABLE
|
192 pmd
= pmd_offset(pud
, phys
);
193 BUG_ON(!pmd_none(*pmd
));
194 set_pmd(pmd
, __pmd(phys
| pgprot_val(prot
)));
198 void __init
init_extra_mapping_wb(unsigned long phys
, unsigned long size
)
200 __init_extra_mapping(phys
, size
, PAGE_KERNEL_LARGE
);
203 void __init
init_extra_mapping_uc(unsigned long phys
, unsigned long size
)
205 __init_extra_mapping(phys
, size
, PAGE_KERNEL_LARGE_NOCACHE
);
209 * The head.S code sets up the kernel high mapping:
211 * from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
213 * phys_addr holds the negative offset to the kernel, which is added
214 * to the compile time generated pmds. This results in invalid pmds up
215 * to the point where we hit the physaddr 0 mapping.
217 * We limit the mappings to the region from _text to _end. _end is
218 * rounded up to the 2MB boundary. This catches the invalid pmds as
219 * well, as they are located before _text:
221 void __init
cleanup_highmap(void)
223 unsigned long vaddr
= __START_KERNEL_map
;
224 unsigned long end
= round_up((unsigned long)_end
, PMD_SIZE
) - 1;
225 pmd_t
*pmd
= level2_kernel_pgt
;
226 pmd_t
*last_pmd
= pmd
+ PTRS_PER_PMD
;
228 for (; pmd
< last_pmd
; pmd
++, vaddr
+= PMD_SIZE
) {
231 if (vaddr
< (unsigned long) _text
|| vaddr
> end
)
232 set_pmd(pmd
, __pmd(0));
236 static unsigned long __initdata table_start
;
237 static unsigned long __meminitdata table_end
;
238 static unsigned long __meminitdata table_top
;
240 static __meminit
void *alloc_low_page(unsigned long *phys
)
242 unsigned long pfn
= table_end
++;
246 adr
= (void *)get_zeroed_page(GFP_ATOMIC
);
252 if (pfn
>= table_top
)
253 panic("alloc_low_page: ran out of memory");
255 adr
= early_ioremap(pfn
* PAGE_SIZE
, PAGE_SIZE
);
256 memset(adr
, 0, PAGE_SIZE
);
257 *phys
= pfn
* PAGE_SIZE
;
261 static __meminit
void unmap_low_page(void *adr
)
266 early_iounmap(adr
, PAGE_SIZE
);
269 static unsigned long __meminit
270 phys_pte_init(pte_t
*pte_page
, unsigned long addr
, unsigned long end
)
273 unsigned long last_map_addr
= end
;
276 pte_t
*pte
= pte_page
+ pte_index(addr
);
278 for(i
= pte_index(addr
); i
< PTRS_PER_PTE
; i
++, addr
+= PAGE_SIZE
, pte
++) {
281 if (!after_bootmem
) {
282 for(; i
< PTRS_PER_PTE
; i
++, pte
++)
283 set_pte(pte
, __pte(0));
292 printk(" pte=%p addr=%lx pte=%016lx\n",
293 pte
, addr
, pfn_pte(addr
>> PAGE_SHIFT
, PAGE_KERNEL
).pte
);
294 set_pte(pte
, pfn_pte(addr
>> PAGE_SHIFT
, PAGE_KERNEL
));
295 last_map_addr
= (addr
& PAGE_MASK
) + PAGE_SIZE
;
298 update_page_count(PG_LEVEL_4K
, pages
);
300 return last_map_addr
;
303 static unsigned long __meminit
304 phys_pte_update(pmd_t
*pmd
, unsigned long address
, unsigned long end
)
306 pte_t
*pte
= (pte_t
*)pmd_page_vaddr(*pmd
);
308 return phys_pte_init(pte
, address
, end
);
311 static unsigned long __meminit
312 phys_pmd_init(pmd_t
*pmd_page
, unsigned long address
, unsigned long end
,
313 unsigned long page_size_mask
)
315 unsigned long pages
= 0;
316 unsigned long last_map_addr
= end
;
318 int i
= pmd_index(address
);
320 for (; i
< PTRS_PER_PMD
; i
++, address
+= PMD_SIZE
) {
321 unsigned long pte_phys
;
322 pmd_t
*pmd
= pmd_page
+ pmd_index(address
);
325 if (address
>= end
) {
326 if (!after_bootmem
) {
327 for (; i
< PTRS_PER_PMD
; i
++, pmd
++)
328 set_pmd(pmd
, __pmd(0));
334 if (!pmd_large(*pmd
))
335 last_map_addr
= phys_pte_update(pmd
, address
,
340 if (page_size_mask
& (1<<PG_LEVEL_2M
)) {
342 set_pte((pte_t
*)pmd
,
343 pfn_pte(address
>> PAGE_SHIFT
, PAGE_KERNEL_LARGE
));
344 last_map_addr
= (address
& PMD_MASK
) + PMD_SIZE
;
348 pte
= alloc_low_page(&pte_phys
);
349 last_map_addr
= phys_pte_init(pte
, address
, end
);
352 pmd_populate_kernel(&init_mm
, pmd
, __va(pte_phys
));
354 update_page_count(PG_LEVEL_2M
, pages
);
355 return last_map_addr
;
358 static unsigned long __meminit
359 phys_pmd_update(pud_t
*pud
, unsigned long address
, unsigned long end
,
360 unsigned long page_size_mask
)
362 pmd_t
*pmd
= pmd_offset(pud
, 0);
363 unsigned long last_map_addr
;
365 spin_lock(&init_mm
.page_table_lock
);
366 last_map_addr
= phys_pmd_init(pmd
, address
, end
, page_size_mask
);
367 spin_unlock(&init_mm
.page_table_lock
);
369 return last_map_addr
;
372 static unsigned long __meminit
373 phys_pud_init(pud_t
*pud_page
, unsigned long addr
, unsigned long end
,
374 unsigned long page_size_mask
)
376 unsigned long pages
= 0;
377 unsigned long last_map_addr
= end
;
378 int i
= pud_index(addr
);
380 for (; i
< PTRS_PER_PUD
; i
++, addr
= (addr
& PUD_MASK
) + PUD_SIZE
) {
381 unsigned long pmd_phys
;
382 pud_t
*pud
= pud_page
+ pud_index(addr
);
388 if (!after_bootmem
&&
389 !e820_any_mapped(addr
, addr
+PUD_SIZE
, 0)) {
390 set_pud(pud
, __pud(0));
395 if (!pud_large(*pud
))
396 last_map_addr
= phys_pmd_update(pud
, addr
, end
,
401 if (page_size_mask
& (1<<PG_LEVEL_1G
)) {
403 set_pte((pte_t
*)pud
,
404 pfn_pte(addr
>> PAGE_SHIFT
, PAGE_KERNEL_LARGE
));
405 last_map_addr
= (addr
& PUD_MASK
) + PUD_SIZE
;
409 pmd
= alloc_low_page(&pmd_phys
);
411 spin_lock(&init_mm
.page_table_lock
);
412 last_map_addr
= phys_pmd_init(pmd
, addr
, end
, page_size_mask
);
414 pud_populate(&init_mm
, pud
, __va(pmd_phys
));
415 spin_unlock(&init_mm
.page_table_lock
);
419 update_page_count(PG_LEVEL_1G
, pages
);
421 return last_map_addr
;
424 static unsigned long __meminit
425 phys_pud_update(pgd_t
*pgd
, unsigned long addr
, unsigned long end
,
426 unsigned long page_size_mask
)
430 pud
= (pud_t
*)pgd_page_vaddr(*pgd
);
432 return phys_pud_init(pud
, addr
, end
, page_size_mask
);
435 static void __init
find_early_table_space(unsigned long end
)
437 unsigned long puds
, pmds
, ptes
, tables
, start
;
439 puds
= (end
+ PUD_SIZE
- 1) >> PUD_SHIFT
;
440 tables
= round_up(puds
* sizeof(pud_t
), PAGE_SIZE
);
441 if (direct_gbpages
) {
443 extra
= end
- ((end
>>PUD_SHIFT
) << PUD_SHIFT
);
444 pmds
= (extra
+ PMD_SIZE
- 1) >> PMD_SHIFT
;
446 pmds
= (end
+ PMD_SIZE
- 1) >> PMD_SHIFT
;
447 tables
+= round_up(pmds
* sizeof(pmd_t
), PAGE_SIZE
);
451 extra
= end
- ((end
>>PMD_SHIFT
) << PMD_SHIFT
);
452 ptes
= (extra
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
454 ptes
= (end
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
455 tables
+= round_up(ptes
* sizeof(pte_t
), PAGE_SIZE
);
458 * RED-PEN putting page tables only on node 0 could
459 * cause a hotspot and fill up ZONE_DMA. The page tables
460 * need roughly 0.5KB per GB.
463 table_start
= find_e820_area(start
, end
, tables
, PAGE_SIZE
);
464 if (table_start
== -1UL)
465 panic("Cannot find space for the kernel page tables");
467 table_start
>>= PAGE_SHIFT
;
468 table_end
= table_start
;
469 table_top
= table_start
+ (tables
>> PAGE_SHIFT
);
471 printk(KERN_DEBUG
"kernel direct mapping tables up to %lx @ %lx-%lx\n",
472 end
, table_start
<< PAGE_SHIFT
, table_top
<< PAGE_SHIFT
);
475 static void __init
init_gbpages(void)
477 if (direct_gbpages
&& cpu_has_gbpages
)
478 printk(KERN_INFO
"Using GB pages for direct mapping\n");
483 static unsigned long __init
kernel_physical_mapping_init(unsigned long start
,
485 unsigned long page_size_mask
)
488 unsigned long next
, last_map_addr
= end
;
490 start
= (unsigned long)__va(start
);
491 end
= (unsigned long)__va(end
);
493 for (; start
< end
; start
= next
) {
494 pgd_t
*pgd
= pgd_offset_k(start
);
495 unsigned long pud_phys
;
498 next
= (start
+ PGDIR_SIZE
) & PGDIR_MASK
;
503 last_map_addr
= phys_pud_update(pgd
, __pa(start
),
504 __pa(end
), page_size_mask
);
509 pud
= pud_offset(pgd
, start
& PGDIR_MASK
);
511 pud
= alloc_low_page(&pud_phys
);
513 last_map_addr
= phys_pud_init(pud
, __pa(start
), __pa(next
),
516 pgd_populate(&init_mm
, pgd_offset_k(start
),
520 return last_map_addr
;
526 unsigned page_size_mask
;
529 #define NR_RANGE_MR 5
531 static int save_mr(struct map_range
*mr
, int nr_range
,
532 unsigned long start_pfn
, unsigned long end_pfn
,
533 unsigned long page_size_mask
)
536 if (start_pfn
< end_pfn
) {
537 if (nr_range
>= NR_RANGE_MR
)
538 panic("run out of range for init_memory_mapping\n");
539 mr
[nr_range
].start
= start_pfn
<<PAGE_SHIFT
;
540 mr
[nr_range
].end
= end_pfn
<<PAGE_SHIFT
;
541 mr
[nr_range
].page_size_mask
= page_size_mask
;
549 * Setup the direct mapping of the physical memory at PAGE_OFFSET.
550 * This runs before bootmem is initialized and gets pages directly from
551 * the physical memory. To access them they are temporarily mapped.
553 unsigned long __init_refok
init_memory_mapping(unsigned long start
,
556 unsigned long last_map_addr
= 0;
557 unsigned long page_size_mask
= 0;
558 unsigned long start_pfn
, end_pfn
;
560 struct map_range mr
[NR_RANGE_MR
];
563 printk(KERN_INFO
"init_memory_mapping\n");
566 * Find space for the kernel direct mapping tables.
568 * Later we should allocate these tables in the local node of the
569 * memory mapped. Unfortunately this is done currently before the
570 * nodes are discovered.
576 page_size_mask
|= 1 << PG_LEVEL_1G
;
578 page_size_mask
|= 1 << PG_LEVEL_2M
;
580 memset(mr
, 0, sizeof(mr
));
583 /* head if not big page alignment ?*/
584 start_pfn
= start
>> PAGE_SHIFT
;
585 end_pfn
= ((start
+ (PMD_SIZE
- 1)) >> PMD_SHIFT
)
586 << (PMD_SHIFT
- PAGE_SHIFT
);
587 nr_range
= save_mr(mr
, nr_range
, start_pfn
, end_pfn
, 0);
589 /* big page (2M) range*/
590 start_pfn
= ((start
+ (PMD_SIZE
- 1))>>PMD_SHIFT
)
591 << (PMD_SHIFT
- PAGE_SHIFT
);
592 end_pfn
= ((start
+ (PUD_SIZE
- 1))>>PUD_SHIFT
)
593 << (PUD_SHIFT
- PAGE_SHIFT
);
594 if (end_pfn
> ((end
>>PUD_SHIFT
)<<(PUD_SHIFT
- PAGE_SHIFT
)))
595 end_pfn
= ((end
>>PUD_SHIFT
)<<(PUD_SHIFT
- PAGE_SHIFT
));
596 nr_range
= save_mr(mr
, nr_range
, start_pfn
, end_pfn
,
597 page_size_mask
& (1<<PG_LEVEL_2M
));
599 /* big page (1G) range */
601 end_pfn
= (end
>>PUD_SHIFT
) << (PUD_SHIFT
- PAGE_SHIFT
);
602 nr_range
= save_mr(mr
, nr_range
, start_pfn
, end_pfn
,
604 ((1<<PG_LEVEL_2M
)|(1<<PG_LEVEL_1G
)));
606 /* tail is not big page (1G) alignment */
608 end_pfn
= (end
>>PMD_SHIFT
) << (PMD_SHIFT
- PAGE_SHIFT
);
609 nr_range
= save_mr(mr
, nr_range
, start_pfn
, end_pfn
,
610 page_size_mask
& (1<<PG_LEVEL_2M
));
612 /* tail is not big page (2M) alignment */
614 end_pfn
= end
>>PAGE_SHIFT
;
615 nr_range
= save_mr(mr
, nr_range
, start_pfn
, end_pfn
, 0);
617 /* try to merge same page size and continuous */
618 for (i
= 0; nr_range
> 1 && i
< nr_range
- 1; i
++) {
619 unsigned long old_start
;
620 if (mr
[i
].end
!= mr
[i
+1].start
||
621 mr
[i
].page_size_mask
!= mr
[i
+1].page_size_mask
)
624 old_start
= mr
[i
].start
;
625 memmove(&mr
[i
], &mr
[i
+1],
626 (nr_range
- 1 - i
) * sizeof (struct map_range
));
627 mr
[i
].start
= old_start
;
631 for (i
= 0; i
< nr_range
; i
++)
632 printk(KERN_DEBUG
" %010lx - %010lx page %s\n",
633 mr
[i
].start
, mr
[i
].end
,
634 (mr
[i
].page_size_mask
& (1<<PG_LEVEL_1G
))?"1G":(
635 (mr
[i
].page_size_mask
& (1<<PG_LEVEL_2M
))?"2M":"4k"));
638 find_early_table_space(end
);
640 for (i
= 0; i
< nr_range
; i
++)
641 last_map_addr
= kernel_physical_mapping_init(
642 mr
[i
].start
, mr
[i
].end
,
643 mr
[i
].page_size_mask
);
646 mmu_cr4_features
= read_cr4();
649 if (!after_bootmem
&& table_end
> table_start
)
650 reserve_early(table_start
<< PAGE_SHIFT
,
651 table_end
<< PAGE_SHIFT
, "PGTABLE");
653 printk(KERN_INFO
"last_map_addr: %lx end: %lx\n",
657 early_memtest(start
, end
);
659 return last_map_addr
>> PAGE_SHIFT
;
663 void __init
initmem_init(unsigned long start_pfn
, unsigned long end_pfn
)
665 unsigned long bootmap_size
, bootmap
;
667 bootmap_size
= bootmem_bootmap_pages(end_pfn
)<<PAGE_SHIFT
;
668 bootmap
= find_e820_area(0, end_pfn
<<PAGE_SHIFT
, bootmap_size
,
671 panic("Cannot find bootmem map of size %ld\n", bootmap_size
);
672 /* don't touch min_low_pfn */
673 bootmap_size
= init_bootmem_node(NODE_DATA(0), bootmap
>> PAGE_SHIFT
,
675 e820_register_active_regions(0, start_pfn
, end_pfn
);
676 free_bootmem_with_active_regions(0, end_pfn
);
677 early_res_to_bootmem(0, end_pfn
<<PAGE_SHIFT
);
678 reserve_bootmem(bootmap
, bootmap_size
, BOOTMEM_DEFAULT
);
681 void __init
paging_init(void)
683 unsigned long max_zone_pfns
[MAX_NR_ZONES
];
685 memset(max_zone_pfns
, 0, sizeof(max_zone_pfns
));
686 max_zone_pfns
[ZONE_DMA
] = MAX_DMA_PFN
;
687 max_zone_pfns
[ZONE_DMA32
] = MAX_DMA32_PFN
;
688 max_zone_pfns
[ZONE_NORMAL
] = max_pfn
;
690 memory_present(0, 0, max_pfn
);
692 free_area_init_nodes(max_zone_pfns
);
697 * Memory hotplug specific functions
699 #ifdef CONFIG_MEMORY_HOTPLUG
701 * Memory is added always to NORMAL zone. This means you will never get
702 * additional DMA/DMA32 memory.
704 int arch_add_memory(int nid
, u64 start
, u64 size
)
706 struct pglist_data
*pgdat
= NODE_DATA(nid
);
707 struct zone
*zone
= pgdat
->node_zones
+ ZONE_NORMAL
;
708 unsigned long last_mapped_pfn
, start_pfn
= start
>> PAGE_SHIFT
;
709 unsigned long nr_pages
= size
>> PAGE_SHIFT
;
712 last_mapped_pfn
= init_memory_mapping(start
, start
+ size
-1);
713 if (last_mapped_pfn
> max_pfn_mapped
)
714 max_pfn_mapped
= last_mapped_pfn
;
716 ret
= __add_pages(zone
, start_pfn
, nr_pages
);
721 EXPORT_SYMBOL_GPL(arch_add_memory
);
723 #if !defined(CONFIG_ACPI_NUMA) && defined(CONFIG_NUMA)
724 int memory_add_physaddr_to_nid(u64 start
)
728 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid
);
731 #endif /* CONFIG_MEMORY_HOTPLUG */
734 * devmem_is_allowed() checks to see if /dev/mem access to a certain address
735 * is valid. The argument is a physical page number.
738 * On x86, access has to be given to the first megabyte of ram because that area
739 * contains bios code and data regions used by X and dosemu and similar apps.
740 * Access has to be given to non-kernel-ram areas as well, these contain the PCI
741 * mmio resources as well as potential bios/acpi data regions.
743 int devmem_is_allowed(unsigned long pagenr
)
747 if (!page_is_ram(pagenr
))
753 static struct kcore_list kcore_mem
, kcore_vmalloc
, kcore_kernel
,
754 kcore_modules
, kcore_vsyscall
;
756 void __init
mem_init(void)
758 long codesize
, reservedpages
, datasize
, initsize
;
762 /* clear_bss() already clear the empty_zero_page */
766 /* this will put all low memory onto the freelists */
768 totalram_pages
= numa_free_all_bootmem();
770 totalram_pages
= free_all_bootmem();
772 reservedpages
= max_pfn
- totalram_pages
-
773 absent_pages_in_range(0, max_pfn
);
776 codesize
= (unsigned long) &_etext
- (unsigned long) &_text
;
777 datasize
= (unsigned long) &_edata
- (unsigned long) &_etext
;
778 initsize
= (unsigned long) &__init_end
- (unsigned long) &__init_begin
;
780 /* Register memory areas for /proc/kcore */
781 kclist_add(&kcore_mem
, __va(0), max_low_pfn
<< PAGE_SHIFT
);
782 kclist_add(&kcore_vmalloc
, (void *)VMALLOC_START
,
783 VMALLOC_END
-VMALLOC_START
);
784 kclist_add(&kcore_kernel
, &_stext
, _end
- _stext
);
785 kclist_add(&kcore_modules
, (void *)MODULES_VADDR
, MODULES_LEN
);
786 kclist_add(&kcore_vsyscall
, (void *)VSYSCALL_START
,
787 VSYSCALL_END
- VSYSCALL_START
);
789 printk(KERN_INFO
"Memory: %luk/%luk available (%ldk kernel code, "
790 "%ldk reserved, %ldk data, %ldk init)\n",
791 (unsigned long) nr_free_pages() << (PAGE_SHIFT
-10),
792 max_pfn
<< (PAGE_SHIFT
-10),
794 reservedpages
<< (PAGE_SHIFT
-10),
801 void free_init_pages(char *what
, unsigned long begin
, unsigned long end
)
803 unsigned long addr
= begin
;
809 * If debugging page accesses then do not free this memory but
810 * mark them not present - any buggy init-section access will
811 * create a kernel page fault:
813 #ifdef CONFIG_DEBUG_PAGEALLOC
814 printk(KERN_INFO
"debug: unmapping init memory %08lx..%08lx\n",
815 begin
, PAGE_ALIGN(end
));
816 set_memory_np(begin
, (end
- begin
) >> PAGE_SHIFT
);
818 printk(KERN_INFO
"Freeing %s: %luk freed\n", what
, (end
- begin
) >> 10);
820 for (; addr
< end
; addr
+= PAGE_SIZE
) {
821 ClearPageReserved(virt_to_page(addr
));
822 init_page_count(virt_to_page(addr
));
823 memset((void *)(addr
& ~(PAGE_SIZE
-1)),
824 POISON_FREE_INITMEM
, PAGE_SIZE
);
831 void free_initmem(void)
833 free_init_pages("unused kernel memory",
834 (unsigned long)(&__init_begin
),
835 (unsigned long)(&__init_end
));
838 #ifdef CONFIG_DEBUG_RODATA
839 const int rodata_test_data
= 0xC3;
840 EXPORT_SYMBOL_GPL(rodata_test_data
);
842 void mark_rodata_ro(void)
844 unsigned long start
= PFN_ALIGN(_stext
), end
= PFN_ALIGN(__end_rodata
);
845 unsigned long rodata_start
=
846 ((unsigned long)__start_rodata
+ PAGE_SIZE
- 1) & PAGE_MASK
;
848 #ifdef CONFIG_DYNAMIC_FTRACE
849 /* Dynamic tracing modifies the kernel text section */
850 start
= rodata_start
;
853 printk(KERN_INFO
"Write protecting the kernel read-only data: %luk\n",
854 (end
- start
) >> 10);
855 set_memory_ro(start
, (end
- start
) >> PAGE_SHIFT
);
858 * The rodata section (but not the kernel text!) should also be
861 set_memory_nx(rodata_start
, (end
- rodata_start
) >> PAGE_SHIFT
);
865 #ifdef CONFIG_CPA_DEBUG
866 printk(KERN_INFO
"Testing CPA: undo %lx-%lx\n", start
, end
);
867 set_memory_rw(start
, (end
-start
) >> PAGE_SHIFT
);
869 printk(KERN_INFO
"Testing CPA: again\n");
870 set_memory_ro(start
, (end
-start
) >> PAGE_SHIFT
);
876 #ifdef CONFIG_BLK_DEV_INITRD
877 void free_initrd_mem(unsigned long start
, unsigned long end
)
879 free_init_pages("initrd memory", start
, end
);
883 int __init
reserve_bootmem_generic(unsigned long phys
, unsigned long len
,
890 unsigned long pfn
= phys
>> PAGE_SHIFT
;
892 if (pfn
>= max_pfn
) {
894 * This can happen with kdump kernels when accessing
897 if (pfn
< max_pfn_mapped
)
900 printk(KERN_ERR
"reserve_bootmem: illegal reserve %lx %lu\n",
905 /* Should check here against the e820 map to avoid double free */
907 nid
= phys_to_nid(phys
);
908 next_nid
= phys_to_nid(phys
+ len
- 1);
910 ret
= reserve_bootmem_node(NODE_DATA(nid
), phys
, len
, flags
);
912 ret
= reserve_bootmem(phys
, len
, flags
);
918 reserve_bootmem(phys
, len
, BOOTMEM_DEFAULT
);
921 if (phys
+len
<= MAX_DMA_PFN
*PAGE_SIZE
) {
922 dma_reserve
+= len
/ PAGE_SIZE
;
923 set_dma_reserve(dma_reserve
);
929 int kern_addr_valid(unsigned long addr
)
931 unsigned long above
= ((long)addr
) >> __VIRTUAL_MASK_SHIFT
;
937 if (above
!= 0 && above
!= -1UL)
940 pgd
= pgd_offset_k(addr
);
944 pud
= pud_offset(pgd
, addr
);
948 pmd
= pmd_offset(pud
, addr
);
953 return pfn_valid(pmd_pfn(*pmd
));
955 pte
= pte_offset_kernel(pmd
, addr
);
959 return pfn_valid(pte_pfn(*pte
));
963 * A pseudo VMA to allow ptrace access for the vsyscall page. This only
964 * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
965 * not need special handling anymore:
967 static struct vm_area_struct gate_vma
= {
968 .vm_start
= VSYSCALL_START
,
969 .vm_end
= VSYSCALL_START
+ (VSYSCALL_MAPPED_PAGES
* PAGE_SIZE
),
970 .vm_page_prot
= PAGE_READONLY_EXEC
,
971 .vm_flags
= VM_READ
| VM_EXEC
974 struct vm_area_struct
*get_gate_vma(struct task_struct
*tsk
)
976 #ifdef CONFIG_IA32_EMULATION
977 if (test_tsk_thread_flag(tsk
, TIF_IA32
))
983 int in_gate_area(struct task_struct
*task
, unsigned long addr
)
985 struct vm_area_struct
*vma
= get_gate_vma(task
);
990 return (addr
>= vma
->vm_start
) && (addr
< vma
->vm_end
);
994 * Use this when you have no reliable task/vma, typically from interrupt
995 * context. It is less reliable than using the task's vma and may give
998 int in_gate_area_no_task(unsigned long addr
)
1000 return (addr
>= VSYSCALL_START
) && (addr
< VSYSCALL_END
);
1003 const char *arch_vma_name(struct vm_area_struct
*vma
)
1005 if (vma
->vm_mm
&& vma
->vm_start
== (long)vma
->vm_mm
->context
.vdso
)
1007 if (vma
== &gate_vma
)
1008 return "[vsyscall]";
1012 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1014 * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
1016 static long __meminitdata addr_start
, addr_end
;
1017 static void __meminitdata
*p_start
, *p_end
;
1018 static int __meminitdata node_start
;
1021 vmemmap_populate(struct page
*start_page
, unsigned long size
, int node
)
1023 unsigned long addr
= (unsigned long)start_page
;
1024 unsigned long end
= (unsigned long)(start_page
+ size
);
1030 for (; addr
< end
; addr
= next
) {
1033 pgd
= vmemmap_pgd_populate(addr
, node
);
1037 pud
= vmemmap_pud_populate(pgd
, addr
, node
);
1042 next
= (addr
+ PAGE_SIZE
) & PAGE_MASK
;
1043 pmd
= vmemmap_pmd_populate(pud
, addr
, node
);
1048 p
= vmemmap_pte_populate(pmd
, addr
, node
);
1053 addr_end
= addr
+ PAGE_SIZE
;
1054 p_end
= p
+ PAGE_SIZE
;
1056 next
= pmd_addr_end(addr
, end
);
1058 pmd
= pmd_offset(pud
, addr
);
1059 if (pmd_none(*pmd
)) {
1062 p
= vmemmap_alloc_block(PMD_SIZE
, node
);
1066 entry
= pfn_pte(__pa(p
) >> PAGE_SHIFT
,
1068 set_pmd(pmd
, __pmd(pte_val(entry
)));
1070 /* check to see if we have contiguous blocks */
1071 if (p_end
!= p
|| node_start
!= node
) {
1073 printk(KERN_DEBUG
" [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1074 addr_start
, addr_end
-1, p_start
, p_end
-1, node_start
);
1080 addr_end
= addr
+ PMD_SIZE
;
1081 p_end
= p
+ PMD_SIZE
;
1083 vmemmap_verify((pte_t
*)pmd
, node
, addr
, next
);
1090 void __meminit
vmemmap_populate_print_last(void)
1093 printk(KERN_DEBUG
" [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1094 addr_start
, addr_end
-1, p_start
, p_end
-1, node_start
);