3 * Copyright (C) 1995 Linus Torvalds
5 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
8 #include <linux/module.h>
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/hugetlb.h>
19 #include <linux/swap.h>
20 #include <linux/smp.h>
21 #include <linux/init.h>
22 #include <linux/highmem.h>
23 #include <linux/pagemap.h>
24 #include <linux/pfn.h>
25 #include <linux/poison.h>
26 #include <linux/bootmem.h>
27 #include <linux/slab.h>
28 #include <linux/proc_fs.h>
29 #include <linux/memory_hotplug.h>
30 #include <linux/initrd.h>
31 #include <linux/cpumask.h>
34 #include <asm/processor.h>
35 #include <asm/system.h>
36 #include <asm/uaccess.h>
37 #include <asm/pgtable.h>
39 #include <asm/fixmap.h>
44 #include <asm/tlbflush.h>
45 #include <asm/pgalloc.h>
46 #include <asm/sections.h>
47 #include <asm/paravirt.h>
48 #include <asm/setup.h>
49 #include <asm/cacheflush.h>
52 unsigned int __VMALLOC_RESERVE
= 128 << 20;
54 unsigned long max_low_pfn_mapped
;
55 unsigned long max_pfn_mapped
;
57 DEFINE_PER_CPU(struct mmu_gather
, mmu_gathers
);
58 unsigned long highstart_pfn
, highend_pfn
;
60 static noinline
int do_test_wp_bit(void);
63 static unsigned long __initdata table_start
;
64 static unsigned long __meminitdata table_end
;
65 static unsigned long __meminitdata table_top
;
67 static int __initdata after_init_bootmem
;
69 static __init
void *alloc_low_page(unsigned long *phys
)
71 unsigned long pfn
= table_end
++;
75 panic("alloc_low_page: ran out of memory");
77 adr
= __va(pfn
* PAGE_SIZE
);
78 memset(adr
, 0, PAGE_SIZE
);
79 *phys
= pfn
* PAGE_SIZE
;
84 * Creates a middle page table and puts a pointer to it in the
85 * given global directory entry. This only returns the gd entry
86 * in non-PAE compilation mode, since the middle layer is folded.
88 static pmd_t
* __init
one_md_table_init(pgd_t
*pgd
)
95 if (!(pgd_val(*pgd
) & _PAGE_PRESENT
)) {
96 if (after_init_bootmem
)
97 pmd_table
= (pmd_t
*)alloc_bootmem_low_pages(PAGE_SIZE
);
99 pmd_table
= (pmd_t
*)alloc_low_page(&phys
);
100 paravirt_alloc_pmd(&init_mm
, __pa(pmd_table
) >> PAGE_SHIFT
);
101 set_pgd(pgd
, __pgd(__pa(pmd_table
) | _PAGE_PRESENT
));
102 pud
= pud_offset(pgd
, 0);
103 BUG_ON(pmd_table
!= pmd_offset(pud
, 0));
106 pud
= pud_offset(pgd
, 0);
107 pmd_table
= pmd_offset(pud
, 0);
113 * Create a page table and place a pointer to it in a middle page
116 static pte_t
* __init
one_page_table_init(pmd_t
*pmd
)
118 if (!(pmd_val(*pmd
) & _PAGE_PRESENT
)) {
119 pte_t
*page_table
= NULL
;
121 if (after_init_bootmem
) {
122 #ifdef CONFIG_DEBUG_PAGEALLOC
123 page_table
= (pte_t
*) alloc_bootmem_pages(PAGE_SIZE
);
127 (pte_t
*)alloc_bootmem_low_pages(PAGE_SIZE
);
130 page_table
= (pte_t
*)alloc_low_page(&phys
);
133 paravirt_alloc_pte(&init_mm
, __pa(page_table
) >> PAGE_SHIFT
);
134 set_pmd(pmd
, __pmd(__pa(page_table
) | _PAGE_TABLE
));
135 BUG_ON(page_table
!= pte_offset_kernel(pmd
, 0));
138 return pte_offset_kernel(pmd
, 0);
142 * This function initializes a certain range of kernel virtual memory
143 * with new bootmem page tables, everywhere page tables are missing in
146 * NOTE: The pagetables are allocated contiguous on the physical space
147 * so we can cache the place of the first one and move around without
148 * checking the pgd every time.
151 page_table_range_init(unsigned long start
, unsigned long end
, pgd_t
*pgd_base
)
153 int pgd_idx
, pmd_idx
;
159 pgd_idx
= pgd_index(vaddr
);
160 pmd_idx
= pmd_index(vaddr
);
161 pgd
= pgd_base
+ pgd_idx
;
163 for ( ; (pgd_idx
< PTRS_PER_PGD
) && (vaddr
!= end
); pgd
++, pgd_idx
++) {
164 pmd
= one_md_table_init(pgd
);
165 pmd
= pmd
+ pmd_index(vaddr
);
166 for (; (pmd_idx
< PTRS_PER_PMD
) && (vaddr
!= end
);
168 one_page_table_init(pmd
);
176 static inline int is_kernel_text(unsigned long addr
)
178 if (addr
>= PAGE_OFFSET
&& addr
<= (unsigned long)__init_end
)
184 * This maps the physical memory to kernel virtual address space, a total
185 * of max_low_pfn pages, by creating page tables starting from address
188 static void __init
kernel_physical_mapping_init(pgd_t
*pgd_base
,
189 unsigned long start_pfn
,
190 unsigned long end_pfn
,
193 int pgd_idx
, pmd_idx
, pte_ofs
;
198 unsigned pages_2m
, pages_4k
;
202 * First iteration will setup identity mapping using large/small pages
203 * based on use_pse, with other attributes same as set by
204 * the early code in head_32.S
206 * Second iteration will setup the appropriate attributes (NX, GLOBAL..)
207 * as desired for the kernel identity mapping.
209 * This two pass mechanism conforms to the TLB app note which says:
211 * "Software should not write to a paging-structure entry in a way
212 * that would change, for any linear address, both the page size
213 * and either the page frame or attributes."
221 pages_2m
= pages_4k
= 0;
223 pgd_idx
= pgd_index((pfn
<<PAGE_SHIFT
) + PAGE_OFFSET
);
224 pgd
= pgd_base
+ pgd_idx
;
225 for (; pgd_idx
< PTRS_PER_PGD
; pgd
++, pgd_idx
++) {
226 pmd
= one_md_table_init(pgd
);
230 #ifdef CONFIG_X86_PAE
231 pmd_idx
= pmd_index((pfn
<<PAGE_SHIFT
) + PAGE_OFFSET
);
236 for (; pmd_idx
< PTRS_PER_PMD
&& pfn
< end_pfn
;
238 unsigned int addr
= pfn
* PAGE_SIZE
+ PAGE_OFFSET
;
241 * Map with big pages if possible, otherwise
242 * create normal page tables:
246 pgprot_t prot
= PAGE_KERNEL_LARGE
;
248 * first pass will use the same initial
249 * identity mapping attribute + _PAGE_PSE.
252 __pgprot(PTE_IDENT_ATTR
|
255 addr2
= (pfn
+ PTRS_PER_PTE
-1) * PAGE_SIZE
+
256 PAGE_OFFSET
+ PAGE_SIZE
-1;
258 if (is_kernel_text(addr
) ||
259 is_kernel_text(addr2
))
260 prot
= PAGE_KERNEL_LARGE_EXEC
;
263 if (mapping_iter
== 1)
264 set_pmd(pmd
, pfn_pmd(pfn
, init_prot
));
266 set_pmd(pmd
, pfn_pmd(pfn
, prot
));
271 pte
= one_page_table_init(pmd
);
273 pte_ofs
= pte_index((pfn
<<PAGE_SHIFT
) + PAGE_OFFSET
);
275 for (; pte_ofs
< PTRS_PER_PTE
&& pfn
< end_pfn
;
276 pte
++, pfn
++, pte_ofs
++, addr
+= PAGE_SIZE
) {
277 pgprot_t prot
= PAGE_KERNEL
;
279 * first pass will use the same initial
280 * identity mapping attribute.
282 pgprot_t init_prot
= __pgprot(PTE_IDENT_ATTR
);
284 if (is_kernel_text(addr
))
285 prot
= PAGE_KERNEL_EXEC
;
288 if (mapping_iter
== 1)
289 set_pte(pte
, pfn_pte(pfn
, init_prot
));
291 set_pte(pte
, pfn_pte(pfn
, prot
));
295 if (mapping_iter
== 1) {
297 * update direct mapping page count only in the first
300 update_page_count(PG_LEVEL_2M
, pages_2m
);
301 update_page_count(PG_LEVEL_4K
, pages_4k
);
304 * local global flush tlb, which will flush the previous
305 * mappings present in both small and large page TLB's.
310 * Second iteration will set the actual desired PTE attributes.
318 * devmem_is_allowed() checks to see if /dev/mem access to a certain address
319 * is valid. The argument is a physical page number.
322 * On x86, access has to be given to the first megabyte of ram because that area
323 * contains bios code and data regions used by X and dosemu and similar apps.
324 * Access has to be given to non-kernel-ram areas as well, these contain the PCI
325 * mmio resources as well as potential bios/acpi data regions.
327 int devmem_is_allowed(unsigned long pagenr
)
331 if (!page_is_ram(pagenr
))
336 #ifdef CONFIG_HIGHMEM
340 static inline pte_t
*kmap_get_fixmap_pte(unsigned long vaddr
)
342 return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr
),
343 vaddr
), vaddr
), vaddr
);
346 static void __init
kmap_init(void)
348 unsigned long kmap_vstart
;
351 * Cache the first kmap pte:
353 kmap_vstart
= __fix_to_virt(FIX_KMAP_BEGIN
);
354 kmap_pte
= kmap_get_fixmap_pte(kmap_vstart
);
356 kmap_prot
= PAGE_KERNEL
;
359 static void __init
permanent_kmaps_init(pgd_t
*pgd_base
)
368 page_table_range_init(vaddr
, vaddr
+ PAGE_SIZE
*LAST_PKMAP
, pgd_base
);
370 pgd
= swapper_pg_dir
+ pgd_index(vaddr
);
371 pud
= pud_offset(pgd
, vaddr
);
372 pmd
= pmd_offset(pud
, vaddr
);
373 pte
= pte_offset_kernel(pmd
, vaddr
);
374 pkmap_page_table
= pte
;
377 static void __init
add_one_highpage_init(struct page
*page
, int pfn
)
379 ClearPageReserved(page
);
380 init_page_count(page
);
385 struct add_highpages_data
{
386 unsigned long start_pfn
;
387 unsigned long end_pfn
;
390 static int __init
add_highpages_work_fn(unsigned long start_pfn
,
391 unsigned long end_pfn
, void *datax
)
395 unsigned long final_start_pfn
, final_end_pfn
;
396 struct add_highpages_data
*data
;
398 data
= (struct add_highpages_data
*)datax
;
400 final_start_pfn
= max(start_pfn
, data
->start_pfn
);
401 final_end_pfn
= min(end_pfn
, data
->end_pfn
);
402 if (final_start_pfn
>= final_end_pfn
)
405 for (node_pfn
= final_start_pfn
; node_pfn
< final_end_pfn
;
407 if (!pfn_valid(node_pfn
))
409 page
= pfn_to_page(node_pfn
);
410 add_one_highpage_init(page
, node_pfn
);
417 void __init
add_highpages_with_active_regions(int nid
, unsigned long start_pfn
,
418 unsigned long end_pfn
)
420 struct add_highpages_data data
;
422 data
.start_pfn
= start_pfn
;
423 data
.end_pfn
= end_pfn
;
425 work_with_active_regions(nid
, add_highpages_work_fn
, &data
);
429 static void __init
set_highmem_pages_init(void)
431 add_highpages_with_active_regions(0, highstart_pfn
, highend_pfn
);
433 totalram_pages
+= totalhigh_pages
;
435 #endif /* !CONFIG_NUMA */
438 # define kmap_init() do { } while (0)
439 # define permanent_kmaps_init(pgd_base) do { } while (0)
440 # define set_highmem_pages_init() do { } while (0)
441 #endif /* CONFIG_HIGHMEM */
443 void __init
native_pagetable_setup_start(pgd_t
*base
)
445 unsigned long pfn
, va
;
452 * Remove any mappings which extend past the end of physical
453 * memory from the boot time page table:
455 for (pfn
= max_low_pfn
+ 1; pfn
< 1<<(32-PAGE_SHIFT
); pfn
++) {
456 va
= PAGE_OFFSET
+ (pfn
<<PAGE_SHIFT
);
457 pgd
= base
+ pgd_index(va
);
458 if (!pgd_present(*pgd
))
461 pud
= pud_offset(pgd
, va
);
462 pmd
= pmd_offset(pud
, va
);
463 if (!pmd_present(*pmd
))
466 pte
= pte_offset_kernel(pmd
, va
);
467 if (!pte_present(*pte
))
470 pte_clear(NULL
, va
, pte
);
472 paravirt_alloc_pmd(&init_mm
, __pa(base
) >> PAGE_SHIFT
);
475 void __init
native_pagetable_setup_done(pgd_t
*base
)
480 * Build a proper pagetable for the kernel mappings. Up until this
481 * point, we've been running on some set of pagetables constructed by
484 * If we're booting on native hardware, this will be a pagetable
485 * constructed in arch/x86/kernel/head_32.S. The root of the
486 * pagetable will be swapper_pg_dir.
488 * If we're booting paravirtualized under a hypervisor, then there are
489 * more options: we may already be running PAE, and the pagetable may
490 * or may not be based in swapper_pg_dir. In any case,
491 * paravirt_pagetable_setup_start() will set up swapper_pg_dir
492 * appropriately for the rest of the initialization to work.
494 * In general, pagetable_init() assumes that the pagetable may already
495 * be partially populated, and so it avoids stomping on any existing
498 static void __init
early_ioremap_page_table_range_init(pgd_t
*pgd_base
)
500 unsigned long vaddr
, end
;
503 * Fixed mappings, only the page table structure has to be
504 * created - mappings will be set by set_fixmap():
506 early_ioremap_clear();
507 vaddr
= __fix_to_virt(__end_of_fixed_addresses
- 1) & PMD_MASK
;
508 end
= (FIXADDR_TOP
+ PMD_SIZE
- 1) & PMD_MASK
;
509 page_table_range_init(vaddr
, end
, pgd_base
);
510 early_ioremap_reset();
513 static void __init
pagetable_init(void)
515 pgd_t
*pgd_base
= swapper_pg_dir
;
517 permanent_kmaps_init(pgd_base
);
520 #ifdef CONFIG_ACPI_SLEEP
522 * ACPI suspend needs this for resume, because things like the intel-agp
523 * driver might have split up a kernel 4MB mapping.
525 char swsusp_pg_dir
[PAGE_SIZE
]
526 __attribute__ ((aligned(PAGE_SIZE
)));
528 static inline void save_pg_dir(void)
530 memcpy(swsusp_pg_dir
, swapper_pg_dir
, PAGE_SIZE
);
532 #else /* !CONFIG_ACPI_SLEEP */
533 static inline void save_pg_dir(void)
536 #endif /* !CONFIG_ACPI_SLEEP */
538 void zap_low_mappings(void)
543 * Zap initial low-memory mappings.
545 * Note that "pgd_clear()" doesn't do it for
546 * us, because pgd_clear() is a no-op on i386.
548 for (i
= 0; i
< KERNEL_PGD_BOUNDARY
; i
++) {
549 #ifdef CONFIG_X86_PAE
550 set_pgd(swapper_pg_dir
+i
, __pgd(1 + __pa(empty_zero_page
)));
552 set_pgd(swapper_pg_dir
+i
, __pgd(0));
560 pteval_t __supported_pte_mask __read_mostly
= ~(_PAGE_NX
| _PAGE_GLOBAL
);
561 EXPORT_SYMBOL_GPL(__supported_pte_mask
);
563 #ifdef CONFIG_X86_PAE
565 static int disable_nx __initdata
;
570 * Control non executable mappings.
575 static int __init
noexec_setup(char *str
)
577 if (!str
|| !strcmp(str
, "on")) {
579 __supported_pte_mask
|= _PAGE_NX
;
583 if (!strcmp(str
, "off")) {
585 __supported_pte_mask
&= ~_PAGE_NX
;
593 early_param("noexec", noexec_setup
);
595 static void __init
set_nx(void)
597 unsigned int v
[4], l
, h
;
599 if (cpu_has_pae
&& (cpuid_eax(0x80000000) > 0x80000001)) {
600 cpuid(0x80000001, &v
[0], &v
[1], &v
[2], &v
[3]);
602 if ((v
[3] & (1 << 20)) && !disable_nx
) {
603 rdmsr(MSR_EFER
, l
, h
);
605 wrmsr(MSR_EFER
, l
, h
);
607 __supported_pte_mask
|= _PAGE_NX
;
613 /* user-defined highmem size */
614 static unsigned int highmem_pages
= -1;
617 * highmem=size forces highmem to be exactly 'size' bytes.
618 * This works even on boxes that have no highmem otherwise.
619 * This also works to reduce highmem size on bigger boxes.
621 static int __init
parse_highmem(char *arg
)
626 highmem_pages
= memparse(arg
, &arg
) >> PAGE_SHIFT
;
629 early_param("highmem", parse_highmem
);
632 * Determine low and high memory ranges:
634 void __init
find_low_pfn_range(void)
636 /* it could update max_pfn */
638 /* max_low_pfn is 0, we already have early_res support */
640 max_low_pfn
= max_pfn
;
641 if (max_low_pfn
> MAXMEM_PFN
) {
642 if (highmem_pages
== -1)
643 highmem_pages
= max_pfn
- MAXMEM_PFN
;
644 if (highmem_pages
+ MAXMEM_PFN
< max_pfn
)
645 max_pfn
= MAXMEM_PFN
+ highmem_pages
;
646 if (highmem_pages
+ MAXMEM_PFN
> max_pfn
) {
647 printk(KERN_WARNING
"only %luMB highmem pages "
648 "available, ignoring highmem size of %uMB.\n",
649 pages_to_mb(max_pfn
- MAXMEM_PFN
),
650 pages_to_mb(highmem_pages
));
653 max_low_pfn
= MAXMEM_PFN
;
654 #ifndef CONFIG_HIGHMEM
655 /* Maximum memory usable is what is directly addressable */
656 printk(KERN_WARNING
"Warning only %ldMB will be used.\n",
658 if (max_pfn
> MAX_NONPAE_PFN
)
660 "Use a HIGHMEM64G enabled kernel.\n");
662 printk(KERN_WARNING
"Use a HIGHMEM enabled kernel.\n");
663 max_pfn
= MAXMEM_PFN
;
664 #else /* !CONFIG_HIGHMEM */
665 #ifndef CONFIG_HIGHMEM64G
666 if (max_pfn
> MAX_NONPAE_PFN
) {
667 max_pfn
= MAX_NONPAE_PFN
;
668 printk(KERN_WARNING
"Warning only 4GB will be used."
669 "Use a HIGHMEM64G enabled kernel.\n");
671 #endif /* !CONFIG_HIGHMEM64G */
672 #endif /* !CONFIG_HIGHMEM */
674 if (highmem_pages
== -1)
676 #ifdef CONFIG_HIGHMEM
677 if (highmem_pages
>= max_pfn
) {
678 printk(KERN_ERR
"highmem size specified (%uMB) is "
679 "bigger than pages available (%luMB)!.\n",
680 pages_to_mb(highmem_pages
),
681 pages_to_mb(max_pfn
));
685 if (max_low_pfn
- highmem_pages
<
686 64*1024*1024/PAGE_SIZE
){
687 printk(KERN_ERR
"highmem size %uMB results in "
688 "smaller than 64MB lowmem, ignoring it.\n"
689 , pages_to_mb(highmem_pages
));
692 max_low_pfn
-= highmem_pages
;
696 printk(KERN_ERR
"ignoring highmem size on non-highmem"
702 #ifndef CONFIG_NEED_MULTIPLE_NODES
703 void __init
initmem_init(unsigned long start_pfn
,
704 unsigned long end_pfn
)
706 #ifdef CONFIG_HIGHMEM
707 highstart_pfn
= highend_pfn
= max_pfn
;
708 if (max_pfn
> max_low_pfn
)
709 highstart_pfn
= max_low_pfn
;
710 memory_present(0, 0, highend_pfn
);
711 e820_register_active_regions(0, 0, highend_pfn
);
712 printk(KERN_NOTICE
"%ldMB HIGHMEM available.\n",
713 pages_to_mb(highend_pfn
- highstart_pfn
));
714 num_physpages
= highend_pfn
;
715 high_memory
= (void *) __va(highstart_pfn
* PAGE_SIZE
- 1) + 1;
717 memory_present(0, 0, max_low_pfn
);
718 e820_register_active_regions(0, 0, max_low_pfn
);
719 num_physpages
= max_low_pfn
;
720 high_memory
= (void *) __va(max_low_pfn
* PAGE_SIZE
- 1) + 1;
722 #ifdef CONFIG_FLATMEM
723 max_mapnr
= num_physpages
;
725 printk(KERN_NOTICE
"%ldMB LOWMEM available.\n",
726 pages_to_mb(max_low_pfn
));
728 setup_bootmem_allocator();
730 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
732 static void __init
zone_sizes_init(void)
734 unsigned long max_zone_pfns
[MAX_NR_ZONES
];
735 memset(max_zone_pfns
, 0, sizeof(max_zone_pfns
));
736 max_zone_pfns
[ZONE_DMA
] =
737 virt_to_phys((char *)MAX_DMA_ADDRESS
) >> PAGE_SHIFT
;
738 max_zone_pfns
[ZONE_NORMAL
] = max_low_pfn
;
739 #ifdef CONFIG_HIGHMEM
740 max_zone_pfns
[ZONE_HIGHMEM
] = highend_pfn
;
743 free_area_init_nodes(max_zone_pfns
);
746 void __init
setup_bootmem_allocator(void)
749 unsigned long bootmap_size
, bootmap
;
751 * Initialize the boot-time allocator (with low memory only):
753 bootmap_size
= bootmem_bootmap_pages(max_low_pfn
)<<PAGE_SHIFT
;
754 bootmap
= find_e820_area(min_low_pfn
<<PAGE_SHIFT
,
755 max_pfn_mapped
<<PAGE_SHIFT
, bootmap_size
,
758 panic("Cannot find bootmem map of size %ld\n", bootmap_size
);
759 reserve_early(bootmap
, bootmap
+ bootmap_size
, "BOOTMAP");
761 /* don't touch min_low_pfn */
762 bootmap_size
= init_bootmem_node(NODE_DATA(0), bootmap
>> PAGE_SHIFT
,
763 min_low_pfn
, max_low_pfn
);
764 printk(KERN_INFO
" mapped low ram: 0 - %08lx\n",
765 max_pfn_mapped
<<PAGE_SHIFT
);
766 printk(KERN_INFO
" low ram: %08lx - %08lx\n",
767 min_low_pfn
<<PAGE_SHIFT
, max_low_pfn
<<PAGE_SHIFT
);
768 printk(KERN_INFO
" bootmap %08lx - %08lx\n",
769 bootmap
, bootmap
+ bootmap_size
);
770 for_each_online_node(i
)
771 free_bootmem_with_active_regions(i
, max_low_pfn
);
772 early_res_to_bootmem(0, max_low_pfn
<<PAGE_SHIFT
);
774 after_init_bootmem
= 1;
777 static void __init
find_early_table_space(unsigned long end
, int use_pse
)
779 unsigned long puds
, pmds
, ptes
, tables
, start
;
781 puds
= (end
+ PUD_SIZE
- 1) >> PUD_SHIFT
;
782 tables
= PAGE_ALIGN(puds
* sizeof(pud_t
));
784 pmds
= (end
+ PMD_SIZE
- 1) >> PMD_SHIFT
;
785 tables
+= PAGE_ALIGN(pmds
* sizeof(pmd_t
));
790 extra
= end
- ((end
>>PMD_SHIFT
) << PMD_SHIFT
);
792 ptes
= (extra
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
794 ptes
= (end
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
796 tables
+= PAGE_ALIGN(ptes
* sizeof(pte_t
));
799 tables
+= PAGE_SIZE
* 2;
802 * RED-PEN putting page tables only on node 0 could
803 * cause a hotspot and fill up ZONE_DMA. The page tables
804 * need roughly 0.5KB per GB.
807 table_start
= find_e820_area(start
, max_pfn_mapped
<<PAGE_SHIFT
,
809 if (table_start
== -1UL)
810 panic("Cannot find space for the kernel page tables");
812 table_start
>>= PAGE_SHIFT
;
813 table_end
= table_start
;
814 table_top
= table_start
+ (tables
>>PAGE_SHIFT
);
816 printk(KERN_DEBUG
"kernel direct mapping tables up to %lx @ %lx-%lx\n",
817 end
, table_start
<< PAGE_SHIFT
,
818 (table_start
<< PAGE_SHIFT
) + tables
);
821 unsigned long __init_refok
init_memory_mapping(unsigned long start
,
824 pgd_t
*pgd_base
= swapper_pg_dir
;
825 unsigned long start_pfn
, end_pfn
;
826 unsigned long big_page_start
;
827 #ifdef CONFIG_DEBUG_PAGEALLOC
829 * For CONFIG_DEBUG_PAGEALLOC, identity mapping will use small pages.
830 * This will simplify cpa(), which otherwise needs to support splitting
831 * large pages into small in interrupt context, etc.
835 int use_pse
= cpu_has_pse
;
839 * Find space for the kernel direct mapping tables.
841 if (!after_init_bootmem
)
842 find_early_table_space(end
, use_pse
);
844 #ifdef CONFIG_X86_PAE
847 printk(KERN_INFO
"NX (Execute Disable) protection: active\n");
850 /* Enable PSE if available */
852 set_in_cr4(X86_CR4_PSE
);
854 /* Enable PGE if available */
856 set_in_cr4(X86_CR4_PGE
);
857 __supported_pte_mask
|= _PAGE_GLOBAL
;
861 * Don't use a large page for the first 2/4MB of memory
862 * because there are often fixed size MTRRs in there
863 * and overlapping MTRRs into large pages can cause
866 big_page_start
= PMD_SIZE
;
868 if (start
< big_page_start
) {
869 start_pfn
= start
>> PAGE_SHIFT
;
870 end_pfn
= min(big_page_start
>>PAGE_SHIFT
, end
>>PAGE_SHIFT
);
872 /* head is not big page alignment ? */
873 start_pfn
= start
>> PAGE_SHIFT
;
874 end_pfn
= ((start
+ (PMD_SIZE
- 1))>>PMD_SHIFT
)
875 << (PMD_SHIFT
- PAGE_SHIFT
);
877 if (start_pfn
< end_pfn
)
878 kernel_physical_mapping_init(pgd_base
, start_pfn
, end_pfn
, 0);
881 start_pfn
= ((start
+ (PMD_SIZE
- 1))>>PMD_SHIFT
)
882 << (PMD_SHIFT
- PAGE_SHIFT
);
883 if (start_pfn
< (big_page_start
>> PAGE_SHIFT
))
884 start_pfn
= big_page_start
>> PAGE_SHIFT
;
885 end_pfn
= (end
>>PMD_SHIFT
) << (PMD_SHIFT
- PAGE_SHIFT
);
886 if (start_pfn
< end_pfn
)
887 kernel_physical_mapping_init(pgd_base
, start_pfn
, end_pfn
,
890 /* tail is not big page alignment ? */
892 if (start_pfn
> (big_page_start
>>PAGE_SHIFT
)) {
893 end_pfn
= end
>> PAGE_SHIFT
;
894 if (start_pfn
< end_pfn
)
895 kernel_physical_mapping_init(pgd_base
, start_pfn
,
899 early_ioremap_page_table_range_init(pgd_base
);
901 load_cr3(swapper_pg_dir
);
905 if (!after_init_bootmem
)
906 reserve_early(table_start
<< PAGE_SHIFT
,
907 table_end
<< PAGE_SHIFT
, "PGTABLE");
909 if (!after_init_bootmem
)
910 early_memtest(start
, end
);
912 return end
>> PAGE_SHIFT
;
917 * paging_init() sets up the page tables - note that the first 8MB are
918 * already mapped by head.S.
920 * This routines also unmaps the page at virtual kernel address 0, so
921 * that we can trap those pesky NULL-reference errors in the kernel.
923 void __init
paging_init(void)
932 * NOTE: at this point the bootmem allocator is fully available.
939 * Test if the WP bit works in supervisor mode. It isn't supported on 386's
940 * and also on some strange 486's. All 586+'s are OK. This used to involve
941 * black magic jumps to work around some nasty CPU bugs, but fortunately the
942 * switch to using exceptions got rid of all that.
944 static void __init
test_wp_bit(void)
947 "Checking if this processor honours the WP bit even in supervisor mode...");
949 /* Any page-aligned address will do, the test is non-destructive */
950 __set_fixmap(FIX_WP_TEST
, __pa(&swapper_pg_dir
), PAGE_READONLY
);
951 boot_cpu_data
.wp_works_ok
= do_test_wp_bit();
952 clear_fixmap(FIX_WP_TEST
);
954 if (!boot_cpu_data
.wp_works_ok
) {
955 printk(KERN_CONT
"No.\n");
956 #ifdef CONFIG_X86_WP_WORKS_OK
958 "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
961 printk(KERN_CONT
"Ok.\n");
965 static struct kcore_list kcore_mem
, kcore_vmalloc
;
967 void __init
mem_init(void)
969 int codesize
, reservedpages
, datasize
, initsize
;
972 #ifdef CONFIG_FLATMEM
975 /* this will put all low memory onto the freelists */
976 totalram_pages
+= free_all_bootmem();
979 for (tmp
= 0; tmp
< max_low_pfn
; tmp
++)
981 * Only count reserved RAM pages:
983 if (page_is_ram(tmp
) && PageReserved(pfn_to_page(tmp
)))
986 set_highmem_pages_init();
988 codesize
= (unsigned long) &_etext
- (unsigned long) &_text
;
989 datasize
= (unsigned long) &_edata
- (unsigned long) &_etext
;
990 initsize
= (unsigned long) &__init_end
- (unsigned long) &__init_begin
;
992 kclist_add(&kcore_mem
, __va(0), max_low_pfn
<< PAGE_SHIFT
);
993 kclist_add(&kcore_vmalloc
, (void *)VMALLOC_START
,
994 VMALLOC_END
-VMALLOC_START
);
996 printk(KERN_INFO
"Memory: %luk/%luk available (%dk kernel code, "
997 "%dk reserved, %dk data, %dk init, %ldk highmem)\n",
998 (unsigned long) nr_free_pages() << (PAGE_SHIFT
-10),
999 num_physpages
<< (PAGE_SHIFT
-10),
1001 reservedpages
<< (PAGE_SHIFT
-10),
1004 (unsigned long) (totalhigh_pages
<< (PAGE_SHIFT
-10))
1007 printk(KERN_INFO
"virtual kernel memory layout:\n"
1008 " fixmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
1009 #ifdef CONFIG_HIGHMEM
1010 " pkmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
1012 " vmalloc : 0x%08lx - 0x%08lx (%4ld MB)\n"
1013 " lowmem : 0x%08lx - 0x%08lx (%4ld MB)\n"
1014 " .init : 0x%08lx - 0x%08lx (%4ld kB)\n"
1015 " .data : 0x%08lx - 0x%08lx (%4ld kB)\n"
1016 " .text : 0x%08lx - 0x%08lx (%4ld kB)\n",
1017 FIXADDR_START
, FIXADDR_TOP
,
1018 (FIXADDR_TOP
- FIXADDR_START
) >> 10,
1020 #ifdef CONFIG_HIGHMEM
1021 PKMAP_BASE
, PKMAP_BASE
+LAST_PKMAP
*PAGE_SIZE
,
1022 (LAST_PKMAP
*PAGE_SIZE
) >> 10,
1025 VMALLOC_START
, VMALLOC_END
,
1026 (VMALLOC_END
- VMALLOC_START
) >> 20,
1028 (unsigned long)__va(0), (unsigned long)high_memory
,
1029 ((unsigned long)high_memory
- (unsigned long)__va(0)) >> 20,
1031 (unsigned long)&__init_begin
, (unsigned long)&__init_end
,
1032 ((unsigned long)&__init_end
-
1033 (unsigned long)&__init_begin
) >> 10,
1035 (unsigned long)&_etext
, (unsigned long)&_edata
,
1036 ((unsigned long)&_edata
- (unsigned long)&_etext
) >> 10,
1038 (unsigned long)&_text
, (unsigned long)&_etext
,
1039 ((unsigned long)&_etext
- (unsigned long)&_text
) >> 10);
1041 #ifdef CONFIG_HIGHMEM
1042 BUG_ON(PKMAP_BASE
+ LAST_PKMAP
*PAGE_SIZE
> FIXADDR_START
);
1043 BUG_ON(VMALLOC_END
> PKMAP_BASE
);
1045 BUG_ON(VMALLOC_START
> VMALLOC_END
);
1046 BUG_ON((unsigned long)high_memory
> VMALLOC_START
);
1048 if (boot_cpu_data
.wp_works_ok
< 0)
1055 #ifdef CONFIG_MEMORY_HOTPLUG
1056 int arch_add_memory(int nid
, u64 start
, u64 size
)
1058 struct pglist_data
*pgdata
= NODE_DATA(nid
);
1059 struct zone
*zone
= pgdata
->node_zones
+ ZONE_HIGHMEM
;
1060 unsigned long start_pfn
= start
>> PAGE_SHIFT
;
1061 unsigned long nr_pages
= size
>> PAGE_SHIFT
;
1063 return __add_pages(zone
, start_pfn
, nr_pages
);
1068 * This function cannot be __init, since exceptions don't work in that
1069 * section. Put this after the callers, so that it cannot be inlined.
1071 static noinline
int do_test_wp_bit(void)
1076 __asm__
__volatile__(
1082 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST
)),
1091 #ifdef CONFIG_DEBUG_RODATA
1092 const int rodata_test_data
= 0xC3;
1093 EXPORT_SYMBOL_GPL(rodata_test_data
);
1095 void mark_rodata_ro(void)
1097 unsigned long start
= PFN_ALIGN(_text
);
1098 unsigned long size
= PFN_ALIGN(_etext
) - start
;
1100 #ifndef CONFIG_DYNAMIC_FTRACE
1101 /* Dynamic tracing modifies the kernel text section */
1102 set_pages_ro(virt_to_page(start
), size
>> PAGE_SHIFT
);
1103 printk(KERN_INFO
"Write protecting the kernel text: %luk\n",
1106 #ifdef CONFIG_CPA_DEBUG
1107 printk(KERN_INFO
"Testing CPA: Reverting %lx-%lx\n",
1109 set_pages_rw(virt_to_page(start
), size
>>PAGE_SHIFT
);
1111 printk(KERN_INFO
"Testing CPA: write protecting again\n");
1112 set_pages_ro(virt_to_page(start
), size
>>PAGE_SHIFT
);
1114 #endif /* CONFIG_DYNAMIC_FTRACE */
1117 size
= (unsigned long)__end_rodata
- start
;
1118 set_pages_ro(virt_to_page(start
), size
>> PAGE_SHIFT
);
1119 printk(KERN_INFO
"Write protecting the kernel read-only data: %luk\n",
1123 #ifdef CONFIG_CPA_DEBUG
1124 printk(KERN_INFO
"Testing CPA: undo %lx-%lx\n", start
, start
+ size
);
1125 set_pages_rw(virt_to_page(start
), size
>> PAGE_SHIFT
);
1127 printk(KERN_INFO
"Testing CPA: write protecting again\n");
1128 set_pages_ro(virt_to_page(start
), size
>> PAGE_SHIFT
);
1133 void free_init_pages(char *what
, unsigned long begin
, unsigned long end
)
1135 #ifdef CONFIG_DEBUG_PAGEALLOC
1137 * If debugging page accesses then do not free this memory but
1138 * mark them not present - any buggy init-section access will
1139 * create a kernel page fault:
1141 printk(KERN_INFO
"debug: unmapping init memory %08lx..%08lx\n",
1142 begin
, PAGE_ALIGN(end
));
1143 set_memory_np(begin
, (end
- begin
) >> PAGE_SHIFT
);
1148 * We just marked the kernel text read only above, now that
1149 * we are going to free part of that, we need to make that
1152 set_memory_rw(begin
, (end
- begin
) >> PAGE_SHIFT
);
1154 for (addr
= begin
; addr
< end
; addr
+= PAGE_SIZE
) {
1155 ClearPageReserved(virt_to_page(addr
));
1156 init_page_count(virt_to_page(addr
));
1157 memset((void *)addr
, POISON_FREE_INITMEM
, PAGE_SIZE
);
1161 printk(KERN_INFO
"Freeing %s: %luk freed\n", what
, (end
- begin
) >> 10);
1165 void free_initmem(void)
1167 free_init_pages("unused kernel memory",
1168 (unsigned long)(&__init_begin
),
1169 (unsigned long)(&__init_end
));
1172 #ifdef CONFIG_BLK_DEV_INITRD
1173 void free_initrd_mem(unsigned long start
, unsigned long end
)
1175 free_init_pages("initrd memory", start
, end
);
1179 int __init
reserve_bootmem_generic(unsigned long phys
, unsigned long len
,
1182 return reserve_bootmem(phys
, len
, flags
);