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/pci.h>
25 #include <linux/pfn.h>
26 #include <linux/poison.h>
27 #include <linux/bootmem.h>
28 #include <linux/slab.h>
29 #include <linux/proc_fs.h>
30 #include <linux/memory_hotplug.h>
31 #include <linux/initrd.h>
32 #include <linux/cpumask.h>
35 #include <asm/bios_ebda.h>
36 #include <asm/processor.h>
37 #include <asm/system.h>
38 #include <asm/uaccess.h>
39 #include <asm/pgtable.h>
41 #include <asm/fixmap.h>
46 #include <asm/tlbflush.h>
47 #include <asm/pgalloc.h>
48 #include <asm/sections.h>
49 #include <asm/paravirt.h>
50 #include <asm/setup.h>
51 #include <asm/cacheflush.h>
53 unsigned int __VMALLOC_RESERVE
= 128 << 20;
55 unsigned long max_low_pfn_mapped
;
56 unsigned long max_pfn_mapped
;
58 DEFINE_PER_CPU(struct mmu_gather
, mmu_gathers
);
59 unsigned long highstart_pfn
, highend_pfn
;
61 static noinline
int do_test_wp_bit(void);
64 static unsigned long __initdata table_start
;
65 static unsigned long __meminitdata table_end
;
66 static unsigned long __meminitdata table_top
;
68 static int __initdata after_init_bootmem
;
70 static __init
void *alloc_low_page(void)
72 unsigned long pfn
= table_end
++;
76 panic("alloc_low_page: ran out of memory");
78 adr
= __va(pfn
* PAGE_SIZE
);
79 memset(adr
, 0, 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
)
94 if (!(pgd_val(*pgd
) & _PAGE_PRESENT
)) {
95 if (after_init_bootmem
)
96 pmd_table
= (pmd_t
*)alloc_bootmem_low_pages(PAGE_SIZE
);
98 pmd_table
= (pmd_t
*)alloc_low_page();
99 paravirt_alloc_pmd(&init_mm
, __pa(pmd_table
) >> PAGE_SHIFT
);
100 set_pgd(pgd
, __pgd(__pa(pmd_table
) | _PAGE_PRESENT
));
101 pud
= pud_offset(pgd
, 0);
102 BUG_ON(pmd_table
!= pmd_offset(pud
, 0));
107 pud
= pud_offset(pgd
, 0);
108 pmd_table
= pmd_offset(pud
, 0);
114 * Create a page table and place a pointer to it in a middle page
117 static pte_t
* __init
one_page_table_init(pmd_t
*pmd
)
119 if (!(pmd_val(*pmd
) & _PAGE_PRESENT
)) {
120 pte_t
*page_table
= NULL
;
122 if (after_init_bootmem
) {
123 #ifdef CONFIG_DEBUG_PAGEALLOC
124 page_table
= (pte_t
*) alloc_bootmem_pages(PAGE_SIZE
);
128 (pte_t
*)alloc_bootmem_low_pages(PAGE_SIZE
);
130 page_table
= (pte_t
*)alloc_low_page();
132 paravirt_alloc_pte(&init_mm
, __pa(page_table
) >> PAGE_SHIFT
);
133 set_pmd(pmd
, __pmd(__pa(page_table
) | _PAGE_TABLE
));
134 BUG_ON(page_table
!= pte_offset_kernel(pmd
, 0));
137 return pte_offset_kernel(pmd
, 0);
140 static pte_t
*__init
page_table_kmap_check(pte_t
*pte
, pmd_t
*pmd
,
141 unsigned long vaddr
, pte_t
*lastpte
)
143 #ifdef CONFIG_HIGHMEM
145 * Something (early fixmap) may already have put a pte
146 * page here, which causes the page table allocation
147 * to become nonlinear. Attempt to fix it, and if it
148 * is still nonlinear then we have to bug.
150 int pmd_idx_kmap_begin
= fix_to_virt(FIX_KMAP_END
) >> PMD_SHIFT
;
151 int pmd_idx_kmap_end
= fix_to_virt(FIX_KMAP_BEGIN
) >> PMD_SHIFT
;
153 if (pmd_idx_kmap_begin
!= pmd_idx_kmap_end
154 && (vaddr
>> PMD_SHIFT
) >= pmd_idx_kmap_begin
155 && (vaddr
>> PMD_SHIFT
) <= pmd_idx_kmap_end
156 && ((__pa(pte
) >> PAGE_SHIFT
) < table_start
157 || (__pa(pte
) >> PAGE_SHIFT
) >= table_end
)) {
161 BUG_ON(after_init_bootmem
);
162 newpte
= alloc_low_page();
163 for (i
= 0; i
< PTRS_PER_PTE
; i
++)
164 set_pte(newpte
+ i
, pte
[i
]);
166 paravirt_alloc_pte(&init_mm
, __pa(newpte
) >> PAGE_SHIFT
);
167 set_pmd(pmd
, __pmd(__pa(newpte
)|_PAGE_TABLE
));
168 BUG_ON(newpte
!= pte_offset_kernel(pmd
, 0));
171 paravirt_release_pte(__pa(pte
) >> PAGE_SHIFT
);
174 BUG_ON(vaddr
< fix_to_virt(FIX_KMAP_BEGIN
- 1)
175 && vaddr
> fix_to_virt(FIX_KMAP_END
)
176 && lastpte
&& lastpte
+ PTRS_PER_PTE
!= pte
);
182 * This function initializes a certain range of kernel virtual memory
183 * with new bootmem page tables, everywhere page tables are missing in
186 * NOTE: The pagetables are allocated contiguous on the physical space
187 * so we can cache the place of the first one and move around without
188 * checking the pgd every time.
191 page_table_range_init(unsigned long start
, unsigned long end
, pgd_t
*pgd_base
)
193 int pgd_idx
, pmd_idx
;
200 pgd_idx
= pgd_index(vaddr
);
201 pmd_idx
= pmd_index(vaddr
);
202 pgd
= pgd_base
+ pgd_idx
;
204 for ( ; (pgd_idx
< PTRS_PER_PGD
) && (vaddr
!= end
); pgd
++, pgd_idx
++) {
205 pmd
= one_md_table_init(pgd
);
206 pmd
= pmd
+ pmd_index(vaddr
);
207 for (; (pmd_idx
< PTRS_PER_PMD
) && (vaddr
!= end
);
209 pte
= page_table_kmap_check(one_page_table_init(pmd
),
218 static inline int is_kernel_text(unsigned long addr
)
220 if (addr
>= PAGE_OFFSET
&& addr
<= (unsigned long)__init_end
)
226 * This maps the physical memory to kernel virtual address space, a total
227 * of max_low_pfn pages, by creating page tables starting from address
230 static void __init
kernel_physical_mapping_init(pgd_t
*pgd_base
,
231 unsigned long start_pfn
,
232 unsigned long end_pfn
,
235 int pgd_idx
, pmd_idx
, pte_ofs
;
240 unsigned pages_2m
, pages_4k
;
244 * First iteration will setup identity mapping using large/small pages
245 * based on use_pse, with other attributes same as set by
246 * the early code in head_32.S
248 * Second iteration will setup the appropriate attributes (NX, GLOBAL..)
249 * as desired for the kernel identity mapping.
251 * This two pass mechanism conforms to the TLB app note which says:
253 * "Software should not write to a paging-structure entry in a way
254 * that would change, for any linear address, both the page size
255 * and either the page frame or attributes."
263 pages_2m
= pages_4k
= 0;
265 pgd_idx
= pgd_index((pfn
<<PAGE_SHIFT
) + PAGE_OFFSET
);
266 pgd
= pgd_base
+ pgd_idx
;
267 for (; pgd_idx
< PTRS_PER_PGD
; pgd
++, pgd_idx
++) {
268 pmd
= one_md_table_init(pgd
);
272 #ifdef CONFIG_X86_PAE
273 pmd_idx
= pmd_index((pfn
<<PAGE_SHIFT
) + PAGE_OFFSET
);
278 for (; pmd_idx
< PTRS_PER_PMD
&& pfn
< end_pfn
;
280 unsigned int addr
= pfn
* PAGE_SIZE
+ PAGE_OFFSET
;
283 * Map with big pages if possible, otherwise
284 * create normal page tables:
288 pgprot_t prot
= PAGE_KERNEL_LARGE
;
290 * first pass will use the same initial
291 * identity mapping attribute + _PAGE_PSE.
294 __pgprot(PTE_IDENT_ATTR
|
297 addr2
= (pfn
+ PTRS_PER_PTE
-1) * PAGE_SIZE
+
298 PAGE_OFFSET
+ PAGE_SIZE
-1;
300 if (is_kernel_text(addr
) ||
301 is_kernel_text(addr2
))
302 prot
= PAGE_KERNEL_LARGE_EXEC
;
305 if (mapping_iter
== 1)
306 set_pmd(pmd
, pfn_pmd(pfn
, init_prot
));
308 set_pmd(pmd
, pfn_pmd(pfn
, prot
));
313 pte
= one_page_table_init(pmd
);
315 pte_ofs
= pte_index((pfn
<<PAGE_SHIFT
) + PAGE_OFFSET
);
317 for (; pte_ofs
< PTRS_PER_PTE
&& pfn
< end_pfn
;
318 pte
++, pfn
++, pte_ofs
++, addr
+= PAGE_SIZE
) {
319 pgprot_t prot
= PAGE_KERNEL
;
321 * first pass will use the same initial
322 * identity mapping attribute.
324 pgprot_t init_prot
= __pgprot(PTE_IDENT_ATTR
);
326 if (is_kernel_text(addr
))
327 prot
= PAGE_KERNEL_EXEC
;
330 if (mapping_iter
== 1)
331 set_pte(pte
, pfn_pte(pfn
, init_prot
));
333 set_pte(pte
, pfn_pte(pfn
, prot
));
337 if (mapping_iter
== 1) {
339 * update direct mapping page count only in the first
342 update_page_count(PG_LEVEL_2M
, pages_2m
);
343 update_page_count(PG_LEVEL_4K
, pages_4k
);
346 * local global flush tlb, which will flush the previous
347 * mappings present in both small and large page TLB's.
352 * Second iteration will set the actual desired PTE attributes.
360 * devmem_is_allowed() checks to see if /dev/mem access to a certain address
361 * is valid. The argument is a physical page number.
364 * On x86, access has to be given to the first megabyte of ram because that area
365 * contains bios code and data regions used by X and dosemu and similar apps.
366 * Access has to be given to non-kernel-ram areas as well, these contain the PCI
367 * mmio resources as well as potential bios/acpi data regions.
369 int devmem_is_allowed(unsigned long pagenr
)
373 if (iomem_is_exclusive(pagenr
<< PAGE_SHIFT
))
375 if (!page_is_ram(pagenr
))
383 static inline pte_t
*kmap_get_fixmap_pte(unsigned long vaddr
)
385 return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr
),
386 vaddr
), vaddr
), vaddr
);
389 static void __init
kmap_init(void)
391 unsigned long kmap_vstart
;
394 * Cache the first kmap pte:
396 kmap_vstart
= __fix_to_virt(FIX_KMAP_BEGIN
);
397 kmap_pte
= kmap_get_fixmap_pte(kmap_vstart
);
399 kmap_prot
= PAGE_KERNEL
;
402 #ifdef CONFIG_HIGHMEM
403 static void __init
permanent_kmaps_init(pgd_t
*pgd_base
)
412 page_table_range_init(vaddr
, vaddr
+ PAGE_SIZE
*LAST_PKMAP
, pgd_base
);
414 pgd
= swapper_pg_dir
+ pgd_index(vaddr
);
415 pud
= pud_offset(pgd
, vaddr
);
416 pmd
= pmd_offset(pud
, vaddr
);
417 pte
= pte_offset_kernel(pmd
, vaddr
);
418 pkmap_page_table
= pte
;
421 static void __init
add_one_highpage_init(struct page
*page
, int pfn
)
423 ClearPageReserved(page
);
424 init_page_count(page
);
429 struct add_highpages_data
{
430 unsigned long start_pfn
;
431 unsigned long end_pfn
;
434 static int __init
add_highpages_work_fn(unsigned long start_pfn
,
435 unsigned long end_pfn
, void *datax
)
439 unsigned long final_start_pfn
, final_end_pfn
;
440 struct add_highpages_data
*data
;
442 data
= (struct add_highpages_data
*)datax
;
444 final_start_pfn
= max(start_pfn
, data
->start_pfn
);
445 final_end_pfn
= min(end_pfn
, data
->end_pfn
);
446 if (final_start_pfn
>= final_end_pfn
)
449 for (node_pfn
= final_start_pfn
; node_pfn
< final_end_pfn
;
451 if (!pfn_valid(node_pfn
))
453 page
= pfn_to_page(node_pfn
);
454 add_one_highpage_init(page
, node_pfn
);
461 void __init
add_highpages_with_active_regions(int nid
, unsigned long start_pfn
,
462 unsigned long end_pfn
)
464 struct add_highpages_data data
;
466 data
.start_pfn
= start_pfn
;
467 data
.end_pfn
= end_pfn
;
469 work_with_active_regions(nid
, add_highpages_work_fn
, &data
);
473 static void __init
set_highmem_pages_init(void)
475 add_highpages_with_active_regions(0, highstart_pfn
, highend_pfn
);
477 totalram_pages
+= totalhigh_pages
;
479 #endif /* !CONFIG_NUMA */
482 static inline void permanent_kmaps_init(pgd_t
*pgd_base
)
485 static inline void set_highmem_pages_init(void)
488 #endif /* CONFIG_HIGHMEM */
490 void __init
native_pagetable_setup_start(pgd_t
*base
)
492 unsigned long pfn
, va
;
499 * Remove any mappings which extend past the end of physical
500 * memory from the boot time page table:
502 for (pfn
= max_low_pfn
+ 1; pfn
< 1<<(32-PAGE_SHIFT
); pfn
++) {
503 va
= PAGE_OFFSET
+ (pfn
<<PAGE_SHIFT
);
504 pgd
= base
+ pgd_index(va
);
505 if (!pgd_present(*pgd
))
508 pud
= pud_offset(pgd
, va
);
509 pmd
= pmd_offset(pud
, va
);
510 if (!pmd_present(*pmd
))
513 pte
= pte_offset_kernel(pmd
, va
);
514 if (!pte_present(*pte
))
517 pte_clear(NULL
, va
, pte
);
519 paravirt_alloc_pmd(&init_mm
, __pa(base
) >> PAGE_SHIFT
);
522 void __init
native_pagetable_setup_done(pgd_t
*base
)
527 * Build a proper pagetable for the kernel mappings. Up until this
528 * point, we've been running on some set of pagetables constructed by
531 * If we're booting on native hardware, this will be a pagetable
532 * constructed in arch/x86/kernel/head_32.S. The root of the
533 * pagetable will be swapper_pg_dir.
535 * If we're booting paravirtualized under a hypervisor, then there are
536 * more options: we may already be running PAE, and the pagetable may
537 * or may not be based in swapper_pg_dir. In any case,
538 * paravirt_pagetable_setup_start() will set up swapper_pg_dir
539 * appropriately for the rest of the initialization to work.
541 * In general, pagetable_init() assumes that the pagetable may already
542 * be partially populated, and so it avoids stomping on any existing
545 static void __init
early_ioremap_page_table_range_init(pgd_t
*pgd_base
)
547 unsigned long vaddr
, end
;
550 * Fixed mappings, only the page table structure has to be
551 * created - mappings will be set by set_fixmap():
553 vaddr
= __fix_to_virt(__end_of_fixed_addresses
- 1) & PMD_MASK
;
554 end
= (FIXADDR_TOP
+ PMD_SIZE
- 1) & PMD_MASK
;
555 page_table_range_init(vaddr
, end
, pgd_base
);
556 early_ioremap_reset();
559 static void __init
pagetable_init(void)
561 pgd_t
*pgd_base
= swapper_pg_dir
;
563 permanent_kmaps_init(pgd_base
);
566 #ifdef CONFIG_ACPI_SLEEP
568 * ACPI suspend needs this for resume, because things like the intel-agp
569 * driver might have split up a kernel 4MB mapping.
571 char swsusp_pg_dir
[PAGE_SIZE
]
572 __attribute__ ((aligned(PAGE_SIZE
)));
574 static inline void save_pg_dir(void)
576 memcpy(swsusp_pg_dir
, swapper_pg_dir
, PAGE_SIZE
);
578 #else /* !CONFIG_ACPI_SLEEP */
579 static inline void save_pg_dir(void)
582 #endif /* !CONFIG_ACPI_SLEEP */
584 void zap_low_mappings(void)
589 * Zap initial low-memory mappings.
591 * Note that "pgd_clear()" doesn't do it for
592 * us, because pgd_clear() is a no-op on i386.
594 for (i
= 0; i
< KERNEL_PGD_BOUNDARY
; i
++) {
595 #ifdef CONFIG_X86_PAE
596 set_pgd(swapper_pg_dir
+i
, __pgd(1 + __pa(empty_zero_page
)));
598 set_pgd(swapper_pg_dir
+i
, __pgd(0));
606 pteval_t __supported_pte_mask __read_mostly
= ~(_PAGE_NX
| _PAGE_GLOBAL
| _PAGE_IOMAP
);
607 EXPORT_SYMBOL_GPL(__supported_pte_mask
);
609 #ifdef CONFIG_X86_PAE
611 static int disable_nx __initdata
;
616 * Control non executable mappings.
621 static int __init
noexec_setup(char *str
)
623 if (!str
|| !strcmp(str
, "on")) {
625 __supported_pte_mask
|= _PAGE_NX
;
629 if (!strcmp(str
, "off")) {
631 __supported_pte_mask
&= ~_PAGE_NX
;
639 early_param("noexec", noexec_setup
);
641 static void __init
set_nx(void)
643 unsigned int v
[4], l
, h
;
645 if (cpu_has_pae
&& (cpuid_eax(0x80000000) > 0x80000001)) {
646 cpuid(0x80000001, &v
[0], &v
[1], &v
[2], &v
[3]);
648 if ((v
[3] & (1 << 20)) && !disable_nx
) {
649 rdmsr(MSR_EFER
, l
, h
);
651 wrmsr(MSR_EFER
, l
, h
);
653 __supported_pte_mask
|= _PAGE_NX
;
659 /* user-defined highmem size */
660 static unsigned int highmem_pages
= -1;
663 * highmem=size forces highmem to be exactly 'size' bytes.
664 * This works even on boxes that have no highmem otherwise.
665 * This also works to reduce highmem size on bigger boxes.
667 static int __init
parse_highmem(char *arg
)
672 highmem_pages
= memparse(arg
, &arg
) >> PAGE_SHIFT
;
675 early_param("highmem", parse_highmem
);
677 #define MSG_HIGHMEM_TOO_BIG \
678 "highmem size (%luMB) is bigger than pages available (%luMB)!\n"
680 #define MSG_LOWMEM_TOO_SMALL \
681 "highmem size (%luMB) results in <64MB lowmem, ignoring it!\n"
683 * All of RAM fits into lowmem - but if user wants highmem
684 * artificially via the highmem=x boot parameter then create
687 void __init
lowmem_pfn_init(void)
689 /* max_low_pfn is 0, we already have early_res support */
690 max_low_pfn
= max_pfn
;
692 if (highmem_pages
== -1)
694 #ifdef CONFIG_HIGHMEM
695 if (highmem_pages
>= max_pfn
) {
696 printk(KERN_ERR MSG_HIGHMEM_TOO_BIG
,
697 pages_to_mb(highmem_pages
), pages_to_mb(max_pfn
));
701 if (max_low_pfn
- highmem_pages
< 64*1024*1024/PAGE_SIZE
) {
702 printk(KERN_ERR MSG_LOWMEM_TOO_SMALL
,
703 pages_to_mb(highmem_pages
));
706 max_low_pfn
-= highmem_pages
;
710 printk(KERN_ERR
"ignoring highmem size on non-highmem kernel!\n");
714 #define MSG_HIGHMEM_TOO_SMALL \
715 "only %luMB highmem pages available, ignoring highmem size of %luMB!\n"
717 #define MSG_HIGHMEM_TRIMMED \
718 "Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n"
720 * We have more RAM than fits into lowmem - we try to put it into
721 * highmem, also taking the highmem=x boot parameter into account:
723 void __init
highmem_pfn_init(void)
725 max_low_pfn
= MAXMEM_PFN
;
727 if (highmem_pages
== -1)
728 highmem_pages
= max_pfn
- MAXMEM_PFN
;
730 if (highmem_pages
+ MAXMEM_PFN
< max_pfn
)
731 max_pfn
= MAXMEM_PFN
+ highmem_pages
;
733 if (highmem_pages
+ MAXMEM_PFN
> max_pfn
) {
734 printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL
,
735 pages_to_mb(max_pfn
- MAXMEM_PFN
),
736 pages_to_mb(highmem_pages
));
739 #ifndef CONFIG_HIGHMEM
740 /* Maximum memory usable is what is directly addressable */
741 printk(KERN_WARNING
"Warning only %ldMB will be used.\n", MAXMEM
>>20);
742 if (max_pfn
> MAX_NONPAE_PFN
)
743 printk(KERN_WARNING
"Use a HIGHMEM64G enabled kernel.\n");
745 printk(KERN_WARNING
"Use a HIGHMEM enabled kernel.\n");
746 max_pfn
= MAXMEM_PFN
;
747 #else /* !CONFIG_HIGHMEM */
748 #ifndef CONFIG_HIGHMEM64G
749 if (max_pfn
> MAX_NONPAE_PFN
) {
750 max_pfn
= MAX_NONPAE_PFN
;
751 printk(KERN_WARNING MSG_HIGHMEM_TRIMMED
);
753 #endif /* !CONFIG_HIGHMEM64G */
754 #endif /* !CONFIG_HIGHMEM */
758 * Determine low and high memory ranges:
760 void __init
find_low_pfn_range(void)
762 /* it could update max_pfn */
764 if (max_pfn
<= MAXMEM_PFN
)
770 #ifndef CONFIG_NEED_MULTIPLE_NODES
771 void __init
initmem_init(unsigned long start_pfn
,
772 unsigned long end_pfn
)
774 #ifdef CONFIG_HIGHMEM
775 highstart_pfn
= highend_pfn
= max_pfn
;
776 if (max_pfn
> max_low_pfn
)
777 highstart_pfn
= max_low_pfn
;
778 memory_present(0, 0, highend_pfn
);
779 e820_register_active_regions(0, 0, highend_pfn
);
780 printk(KERN_NOTICE
"%ldMB HIGHMEM available.\n",
781 pages_to_mb(highend_pfn
- highstart_pfn
));
782 num_physpages
= highend_pfn
;
783 high_memory
= (void *) __va(highstart_pfn
* PAGE_SIZE
- 1) + 1;
785 memory_present(0, 0, max_low_pfn
);
786 e820_register_active_regions(0, 0, max_low_pfn
);
787 num_physpages
= max_low_pfn
;
788 high_memory
= (void *) __va(max_low_pfn
* PAGE_SIZE
- 1) + 1;
790 #ifdef CONFIG_FLATMEM
791 max_mapnr
= num_physpages
;
793 printk(KERN_NOTICE
"%ldMB LOWMEM available.\n",
794 pages_to_mb(max_low_pfn
));
796 setup_bootmem_allocator();
798 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
800 static void __init
zone_sizes_init(void)
802 unsigned long max_zone_pfns
[MAX_NR_ZONES
];
803 memset(max_zone_pfns
, 0, sizeof(max_zone_pfns
));
804 max_zone_pfns
[ZONE_DMA
] =
805 virt_to_phys((char *)MAX_DMA_ADDRESS
) >> PAGE_SHIFT
;
806 max_zone_pfns
[ZONE_NORMAL
] = max_low_pfn
;
807 #ifdef CONFIG_HIGHMEM
808 max_zone_pfns
[ZONE_HIGHMEM
] = highend_pfn
;
811 free_area_init_nodes(max_zone_pfns
);
814 void __init
setup_bootmem_allocator(void)
817 unsigned long bootmap_size
, bootmap
;
819 * Initialize the boot-time allocator (with low memory only):
821 bootmap_size
= bootmem_bootmap_pages(max_low_pfn
)<<PAGE_SHIFT
;
822 bootmap
= find_e820_area(min_low_pfn
<<PAGE_SHIFT
,
823 max_pfn_mapped
<<PAGE_SHIFT
, bootmap_size
,
826 panic("Cannot find bootmem map of size %ld\n", bootmap_size
);
827 reserve_early(bootmap
, bootmap
+ bootmap_size
, "BOOTMAP");
829 /* don't touch min_low_pfn */
830 bootmap_size
= init_bootmem_node(NODE_DATA(0), bootmap
>> PAGE_SHIFT
,
831 min_low_pfn
, max_low_pfn
);
832 printk(KERN_INFO
" mapped low ram: 0 - %08lx\n",
833 max_pfn_mapped
<<PAGE_SHIFT
);
834 printk(KERN_INFO
" low ram: %08lx - %08lx\n",
835 min_low_pfn
<<PAGE_SHIFT
, max_low_pfn
<<PAGE_SHIFT
);
836 printk(KERN_INFO
" bootmap %08lx - %08lx\n",
837 bootmap
, bootmap
+ bootmap_size
);
838 for_each_online_node(i
)
839 free_bootmem_with_active_regions(i
, max_low_pfn
);
840 early_res_to_bootmem(0, max_low_pfn
<<PAGE_SHIFT
);
842 after_init_bootmem
= 1;
845 static void __init
find_early_table_space(unsigned long end
, int use_pse
)
847 unsigned long puds
, pmds
, ptes
, tables
, start
;
849 puds
= (end
+ PUD_SIZE
- 1) >> PUD_SHIFT
;
850 tables
= PAGE_ALIGN(puds
* sizeof(pud_t
));
852 pmds
= (end
+ PMD_SIZE
- 1) >> PMD_SHIFT
;
853 tables
+= PAGE_ALIGN(pmds
* sizeof(pmd_t
));
858 extra
= end
- ((end
>>PMD_SHIFT
) << PMD_SHIFT
);
860 ptes
= (extra
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
862 ptes
= (end
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
864 tables
+= PAGE_ALIGN(ptes
* sizeof(pte_t
));
867 tables
+= PAGE_ALIGN(__end_of_fixed_addresses
* sizeof(pte_t
));
870 * RED-PEN putting page tables only on node 0 could
871 * cause a hotspot and fill up ZONE_DMA. The page tables
872 * need roughly 0.5KB per GB.
875 table_start
= find_e820_area(start
, max_pfn_mapped
<<PAGE_SHIFT
,
877 if (table_start
== -1UL)
878 panic("Cannot find space for the kernel page tables");
880 table_start
>>= PAGE_SHIFT
;
881 table_end
= table_start
;
882 table_top
= table_start
+ (tables
>>PAGE_SHIFT
);
884 printk(KERN_DEBUG
"kernel direct mapping tables up to %lx @ %lx-%lx\n",
885 end
, table_start
<< PAGE_SHIFT
,
886 (table_start
<< PAGE_SHIFT
) + tables
);
889 unsigned long __init_refok
init_memory_mapping(unsigned long start
,
892 pgd_t
*pgd_base
= swapper_pg_dir
;
893 unsigned long start_pfn
, end_pfn
;
894 unsigned long big_page_start
;
895 #ifdef CONFIG_DEBUG_PAGEALLOC
897 * For CONFIG_DEBUG_PAGEALLOC, identity mapping will use small pages.
898 * This will simplify cpa(), which otherwise needs to support splitting
899 * large pages into small in interrupt context, etc.
903 int use_pse
= cpu_has_pse
;
907 * Find space for the kernel direct mapping tables.
909 if (!after_init_bootmem
)
910 find_early_table_space(end
, use_pse
);
912 #ifdef CONFIG_X86_PAE
915 printk(KERN_INFO
"NX (Execute Disable) protection: active\n");
918 /* Enable PSE if available */
920 set_in_cr4(X86_CR4_PSE
);
922 /* Enable PGE if available */
924 set_in_cr4(X86_CR4_PGE
);
925 __supported_pte_mask
|= _PAGE_GLOBAL
;
929 * Don't use a large page for the first 2/4MB of memory
930 * because there are often fixed size MTRRs in there
931 * and overlapping MTRRs into large pages can cause
934 big_page_start
= PMD_SIZE
;
936 if (start
< big_page_start
) {
937 start_pfn
= start
>> PAGE_SHIFT
;
938 end_pfn
= min(big_page_start
>>PAGE_SHIFT
, end
>>PAGE_SHIFT
);
940 /* head is not big page alignment ? */
941 start_pfn
= start
>> PAGE_SHIFT
;
942 end_pfn
= ((start
+ (PMD_SIZE
- 1))>>PMD_SHIFT
)
943 << (PMD_SHIFT
- PAGE_SHIFT
);
945 if (start_pfn
< end_pfn
)
946 kernel_physical_mapping_init(pgd_base
, start_pfn
, end_pfn
, 0);
949 start_pfn
= ((start
+ (PMD_SIZE
- 1))>>PMD_SHIFT
)
950 << (PMD_SHIFT
- PAGE_SHIFT
);
951 if (start_pfn
< (big_page_start
>> PAGE_SHIFT
))
952 start_pfn
= big_page_start
>> PAGE_SHIFT
;
953 end_pfn
= (end
>>PMD_SHIFT
) << (PMD_SHIFT
- PAGE_SHIFT
);
954 if (start_pfn
< end_pfn
)
955 kernel_physical_mapping_init(pgd_base
, start_pfn
, end_pfn
,
958 /* tail is not big page alignment ? */
960 if (start_pfn
> (big_page_start
>>PAGE_SHIFT
)) {
961 end_pfn
= end
>> PAGE_SHIFT
;
962 if (start_pfn
< end_pfn
)
963 kernel_physical_mapping_init(pgd_base
, start_pfn
,
967 early_ioremap_page_table_range_init(pgd_base
);
969 load_cr3(swapper_pg_dir
);
973 if (!after_init_bootmem
)
974 reserve_early(table_start
<< PAGE_SHIFT
,
975 table_end
<< PAGE_SHIFT
, "PGTABLE");
977 if (!after_init_bootmem
)
978 early_memtest(start
, end
);
980 return end
>> PAGE_SHIFT
;
985 * paging_init() sets up the page tables - note that the first 8MB are
986 * already mapped by head.S.
988 * This routines also unmaps the page at virtual kernel address 0, so
989 * that we can trap those pesky NULL-reference errors in the kernel.
991 void __init
paging_init(void)
1000 * NOTE: at this point the bootmem allocator is fully available.
1007 * Test if the WP bit works in supervisor mode. It isn't supported on 386's
1008 * and also on some strange 486's. All 586+'s are OK. This used to involve
1009 * black magic jumps to work around some nasty CPU bugs, but fortunately the
1010 * switch to using exceptions got rid of all that.
1012 static void __init
test_wp_bit(void)
1015 "Checking if this processor honours the WP bit even in supervisor mode...");
1017 /* Any page-aligned address will do, the test is non-destructive */
1018 __set_fixmap(FIX_WP_TEST
, __pa(&swapper_pg_dir
), PAGE_READONLY
);
1019 boot_cpu_data
.wp_works_ok
= do_test_wp_bit();
1020 clear_fixmap(FIX_WP_TEST
);
1022 if (!boot_cpu_data
.wp_works_ok
) {
1023 printk(KERN_CONT
"No.\n");
1024 #ifdef CONFIG_X86_WP_WORKS_OK
1026 "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
1029 printk(KERN_CONT
"Ok.\n");
1033 static struct kcore_list kcore_mem
, kcore_vmalloc
;
1035 void __init
mem_init(void)
1037 int codesize
, reservedpages
, datasize
, initsize
;
1042 #ifdef CONFIG_FLATMEM
1045 /* this will put all low memory onto the freelists */
1046 totalram_pages
+= free_all_bootmem();
1049 for (tmp
= 0; tmp
< max_low_pfn
; tmp
++)
1051 * Only count reserved RAM pages:
1053 if (page_is_ram(tmp
) && PageReserved(pfn_to_page(tmp
)))
1056 set_highmem_pages_init();
1058 codesize
= (unsigned long) &_etext
- (unsigned long) &_text
;
1059 datasize
= (unsigned long) &_edata
- (unsigned long) &_etext
;
1060 initsize
= (unsigned long) &__init_end
- (unsigned long) &__init_begin
;
1062 kclist_add(&kcore_mem
, __va(0), max_low_pfn
<< PAGE_SHIFT
);
1063 kclist_add(&kcore_vmalloc
, (void *)VMALLOC_START
,
1064 VMALLOC_END
-VMALLOC_START
);
1066 printk(KERN_INFO
"Memory: %luk/%luk available (%dk kernel code, "
1067 "%dk reserved, %dk data, %dk init, %ldk highmem)\n",
1068 (unsigned long) nr_free_pages() << (PAGE_SHIFT
-10),
1069 num_physpages
<< (PAGE_SHIFT
-10),
1071 reservedpages
<< (PAGE_SHIFT
-10),
1074 (unsigned long) (totalhigh_pages
<< (PAGE_SHIFT
-10))
1077 printk(KERN_INFO
"virtual kernel memory layout:\n"
1078 " fixmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
1079 #ifdef CONFIG_HIGHMEM
1080 " pkmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
1082 " vmalloc : 0x%08lx - 0x%08lx (%4ld MB)\n"
1083 " lowmem : 0x%08lx - 0x%08lx (%4ld MB)\n"
1084 " .init : 0x%08lx - 0x%08lx (%4ld kB)\n"
1085 " .data : 0x%08lx - 0x%08lx (%4ld kB)\n"
1086 " .text : 0x%08lx - 0x%08lx (%4ld kB)\n",
1087 FIXADDR_START
, FIXADDR_TOP
,
1088 (FIXADDR_TOP
- FIXADDR_START
) >> 10,
1090 #ifdef CONFIG_HIGHMEM
1091 PKMAP_BASE
, PKMAP_BASE
+LAST_PKMAP
*PAGE_SIZE
,
1092 (LAST_PKMAP
*PAGE_SIZE
) >> 10,
1095 VMALLOC_START
, VMALLOC_END
,
1096 (VMALLOC_END
- VMALLOC_START
) >> 20,
1098 (unsigned long)__va(0), (unsigned long)high_memory
,
1099 ((unsigned long)high_memory
- (unsigned long)__va(0)) >> 20,
1101 (unsigned long)&__init_begin
, (unsigned long)&__init_end
,
1102 ((unsigned long)&__init_end
-
1103 (unsigned long)&__init_begin
) >> 10,
1105 (unsigned long)&_etext
, (unsigned long)&_edata
,
1106 ((unsigned long)&_edata
- (unsigned long)&_etext
) >> 10,
1108 (unsigned long)&_text
, (unsigned long)&_etext
,
1109 ((unsigned long)&_etext
- (unsigned long)&_text
) >> 10);
1112 * Check boundaries twice: Some fundamental inconsistencies can
1113 * be detected at build time already.
1115 #define __FIXADDR_TOP (-PAGE_SIZE)
1116 #ifdef CONFIG_HIGHMEM
1117 BUILD_BUG_ON(PKMAP_BASE
+ LAST_PKMAP
*PAGE_SIZE
> FIXADDR_START
);
1118 BUILD_BUG_ON(VMALLOC_END
> PKMAP_BASE
);
1120 #define high_memory (-128UL << 20)
1121 BUILD_BUG_ON(VMALLOC_START
>= VMALLOC_END
);
1123 #undef __FIXADDR_TOP
1125 #ifdef CONFIG_HIGHMEM
1126 BUG_ON(PKMAP_BASE
+ LAST_PKMAP
*PAGE_SIZE
> FIXADDR_START
);
1127 BUG_ON(VMALLOC_END
> PKMAP_BASE
);
1129 BUG_ON(VMALLOC_START
>= VMALLOC_END
);
1130 BUG_ON((unsigned long)high_memory
> VMALLOC_START
);
1132 if (boot_cpu_data
.wp_works_ok
< 0)
1139 #ifdef CONFIG_MEMORY_HOTPLUG
1140 int arch_add_memory(int nid
, u64 start
, u64 size
)
1142 struct pglist_data
*pgdata
= NODE_DATA(nid
);
1143 struct zone
*zone
= pgdata
->node_zones
+ ZONE_HIGHMEM
;
1144 unsigned long start_pfn
= start
>> PAGE_SHIFT
;
1145 unsigned long nr_pages
= size
>> PAGE_SHIFT
;
1147 return __add_pages(nid
, zone
, start_pfn
, nr_pages
);
1152 * This function cannot be __init, since exceptions don't work in that
1153 * section. Put this after the callers, so that it cannot be inlined.
1155 static noinline
int do_test_wp_bit(void)
1160 __asm__
__volatile__(
1166 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST
)),
1175 #ifdef CONFIG_DEBUG_RODATA
1176 const int rodata_test_data
= 0xC3;
1177 EXPORT_SYMBOL_GPL(rodata_test_data
);
1179 void mark_rodata_ro(void)
1181 unsigned long start
= PFN_ALIGN(_text
);
1182 unsigned long size
= PFN_ALIGN(_etext
) - start
;
1184 #ifndef CONFIG_DYNAMIC_FTRACE
1185 /* Dynamic tracing modifies the kernel text section */
1186 set_pages_ro(virt_to_page(start
), size
>> PAGE_SHIFT
);
1187 printk(KERN_INFO
"Write protecting the kernel text: %luk\n",
1190 #ifdef CONFIG_CPA_DEBUG
1191 printk(KERN_INFO
"Testing CPA: Reverting %lx-%lx\n",
1193 set_pages_rw(virt_to_page(start
), size
>>PAGE_SHIFT
);
1195 printk(KERN_INFO
"Testing CPA: write protecting again\n");
1196 set_pages_ro(virt_to_page(start
), size
>>PAGE_SHIFT
);
1198 #endif /* CONFIG_DYNAMIC_FTRACE */
1201 size
= (unsigned long)__end_rodata
- start
;
1202 set_pages_ro(virt_to_page(start
), size
>> PAGE_SHIFT
);
1203 printk(KERN_INFO
"Write protecting the kernel read-only data: %luk\n",
1207 #ifdef CONFIG_CPA_DEBUG
1208 printk(KERN_INFO
"Testing CPA: undo %lx-%lx\n", start
, start
+ size
);
1209 set_pages_rw(virt_to_page(start
), size
>> PAGE_SHIFT
);
1211 printk(KERN_INFO
"Testing CPA: write protecting again\n");
1212 set_pages_ro(virt_to_page(start
), size
>> PAGE_SHIFT
);
1217 void free_init_pages(char *what
, unsigned long begin
, unsigned long end
)
1219 #ifdef CONFIG_DEBUG_PAGEALLOC
1221 * If debugging page accesses then do not free this memory but
1222 * mark them not present - any buggy init-section access will
1223 * create a kernel page fault:
1225 printk(KERN_INFO
"debug: unmapping init memory %08lx..%08lx\n",
1226 begin
, PAGE_ALIGN(end
));
1227 set_memory_np(begin
, (end
- begin
) >> PAGE_SHIFT
);
1232 * We just marked the kernel text read only above, now that
1233 * we are going to free part of that, we need to make that
1236 set_memory_rw(begin
, (end
- begin
) >> PAGE_SHIFT
);
1238 for (addr
= begin
; addr
< end
; addr
+= PAGE_SIZE
) {
1239 ClearPageReserved(virt_to_page(addr
));
1240 init_page_count(virt_to_page(addr
));
1241 memset((void *)addr
, POISON_FREE_INITMEM
, PAGE_SIZE
);
1245 printk(KERN_INFO
"Freeing %s: %luk freed\n", what
, (end
- begin
) >> 10);
1249 void free_initmem(void)
1251 free_init_pages("unused kernel memory",
1252 (unsigned long)(&__init_begin
),
1253 (unsigned long)(&__init_end
));
1256 #ifdef CONFIG_BLK_DEV_INITRD
1257 void free_initrd_mem(unsigned long start
, unsigned long end
)
1259 free_init_pages("initrd memory", start
, end
);
1263 int __init
reserve_bootmem_generic(unsigned long phys
, unsigned long len
,
1266 return reserve_bootmem(phys
, len
, flags
);