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 long max_low_pfn_mapped
;
54 unsigned long max_pfn_mapped
;
56 DEFINE_PER_CPU(struct mmu_gather
, mmu_gathers
);
57 unsigned long highstart_pfn
, highend_pfn
;
59 static noinline
int do_test_wp_bit(void);
62 static unsigned long __initdata table_start
;
63 static unsigned long __meminitdata table_end
;
64 static unsigned long __meminitdata table_top
;
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
)) {
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
;
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_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(unsigned long start_pfn
,
231 unsigned long end_pfn
,
234 pgd_t
*pgd_base
= swapper_pg_dir
;
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.
362 static inline pte_t
*kmap_get_fixmap_pte(unsigned long vaddr
)
364 return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr
),
365 vaddr
), vaddr
), vaddr
);
368 static void __init
kmap_init(void)
370 unsigned long kmap_vstart
;
373 * Cache the first kmap pte:
375 kmap_vstart
= __fix_to_virt(FIX_KMAP_BEGIN
);
376 kmap_pte
= kmap_get_fixmap_pte(kmap_vstart
);
378 kmap_prot
= PAGE_KERNEL
;
381 #ifdef CONFIG_HIGHMEM
382 static void __init
permanent_kmaps_init(pgd_t
*pgd_base
)
391 page_table_range_init(vaddr
, vaddr
+ PAGE_SIZE
*LAST_PKMAP
, pgd_base
);
393 pgd
= swapper_pg_dir
+ pgd_index(vaddr
);
394 pud
= pud_offset(pgd
, vaddr
);
395 pmd
= pmd_offset(pud
, vaddr
);
396 pte
= pte_offset_kernel(pmd
, vaddr
);
397 pkmap_page_table
= pte
;
400 static void __init
add_one_highpage_init(struct page
*page
, int pfn
)
402 ClearPageReserved(page
);
403 init_page_count(page
);
408 struct add_highpages_data
{
409 unsigned long start_pfn
;
410 unsigned long end_pfn
;
413 static int __init
add_highpages_work_fn(unsigned long start_pfn
,
414 unsigned long end_pfn
, void *datax
)
418 unsigned long final_start_pfn
, final_end_pfn
;
419 struct add_highpages_data
*data
;
421 data
= (struct add_highpages_data
*)datax
;
423 final_start_pfn
= max(start_pfn
, data
->start_pfn
);
424 final_end_pfn
= min(end_pfn
, data
->end_pfn
);
425 if (final_start_pfn
>= final_end_pfn
)
428 for (node_pfn
= final_start_pfn
; node_pfn
< final_end_pfn
;
430 if (!pfn_valid(node_pfn
))
432 page
= pfn_to_page(node_pfn
);
433 add_one_highpage_init(page
, node_pfn
);
440 void __init
add_highpages_with_active_regions(int nid
, unsigned long start_pfn
,
441 unsigned long end_pfn
)
443 struct add_highpages_data data
;
445 data
.start_pfn
= start_pfn
;
446 data
.end_pfn
= end_pfn
;
448 work_with_active_regions(nid
, add_highpages_work_fn
, &data
);
452 static inline void permanent_kmaps_init(pgd_t
*pgd_base
)
455 #endif /* CONFIG_HIGHMEM */
457 void __init
native_pagetable_setup_start(pgd_t
*base
)
459 unsigned long pfn
, va
;
466 * Remove any mappings which extend past the end of physical
467 * memory from the boot time page table:
469 for (pfn
= max_low_pfn
+ 1; pfn
< 1<<(32-PAGE_SHIFT
); pfn
++) {
470 va
= PAGE_OFFSET
+ (pfn
<<PAGE_SHIFT
);
471 pgd
= base
+ pgd_index(va
);
472 if (!pgd_present(*pgd
))
475 pud
= pud_offset(pgd
, va
);
476 pmd
= pmd_offset(pud
, va
);
477 if (!pmd_present(*pmd
))
480 pte
= pte_offset_kernel(pmd
, va
);
481 if (!pte_present(*pte
))
484 pte_clear(NULL
, va
, pte
);
486 paravirt_alloc_pmd(&init_mm
, __pa(base
) >> PAGE_SHIFT
);
489 void __init
native_pagetable_setup_done(pgd_t
*base
)
494 * Build a proper pagetable for the kernel mappings. Up until this
495 * point, we've been running on some set of pagetables constructed by
498 * If we're booting on native hardware, this will be a pagetable
499 * constructed in arch/x86/kernel/head_32.S. The root of the
500 * pagetable will be swapper_pg_dir.
502 * If we're booting paravirtualized under a hypervisor, then there are
503 * more options: we may already be running PAE, and the pagetable may
504 * or may not be based in swapper_pg_dir. In any case,
505 * paravirt_pagetable_setup_start() will set up swapper_pg_dir
506 * appropriately for the rest of the initialization to work.
508 * In general, pagetable_init() assumes that the pagetable may already
509 * be partially populated, and so it avoids stomping on any existing
512 static void __init
early_ioremap_page_table_range_init(void)
514 pgd_t
*pgd_base
= swapper_pg_dir
;
515 unsigned long vaddr
, end
;
518 * Fixed mappings, only the page table structure has to be
519 * created - mappings will be set by set_fixmap():
521 vaddr
= __fix_to_virt(__end_of_fixed_addresses
- 1) & PMD_MASK
;
522 end
= (FIXADDR_TOP
+ PMD_SIZE
- 1) & PMD_MASK
;
523 page_table_range_init(vaddr
, end
, pgd_base
);
524 early_ioremap_reset();
527 static void __init
pagetable_init(void)
529 pgd_t
*pgd_base
= swapper_pg_dir
;
531 permanent_kmaps_init(pgd_base
);
534 #ifdef CONFIG_ACPI_SLEEP
536 * ACPI suspend needs this for resume, because things like the intel-agp
537 * driver might have split up a kernel 4MB mapping.
539 char swsusp_pg_dir
[PAGE_SIZE
]
540 __attribute__ ((aligned(PAGE_SIZE
)));
542 static inline void save_pg_dir(void)
544 memcpy(swsusp_pg_dir
, swapper_pg_dir
, PAGE_SIZE
);
546 #else /* !CONFIG_ACPI_SLEEP */
547 static inline void save_pg_dir(void)
550 #endif /* !CONFIG_ACPI_SLEEP */
552 void zap_low_mappings(void)
557 * Zap initial low-memory mappings.
559 * Note that "pgd_clear()" doesn't do it for
560 * us, because pgd_clear() is a no-op on i386.
562 for (i
= 0; i
< KERNEL_PGD_BOUNDARY
; i
++) {
563 #ifdef CONFIG_X86_PAE
564 set_pgd(swapper_pg_dir
+i
, __pgd(1 + __pa(empty_zero_page
)));
566 set_pgd(swapper_pg_dir
+i
, __pgd(0));
574 pteval_t __supported_pte_mask __read_mostly
= ~(_PAGE_NX
| _PAGE_GLOBAL
| _PAGE_IOMAP
);
575 EXPORT_SYMBOL_GPL(__supported_pte_mask
);
577 #ifdef CONFIG_X86_PAE
579 static int disable_nx __initdata
;
584 * Control non executable mappings.
589 static int __init
noexec_setup(char *str
)
591 if (!str
|| !strcmp(str
, "on")) {
593 __supported_pte_mask
|= _PAGE_NX
;
597 if (!strcmp(str
, "off")) {
599 __supported_pte_mask
&= ~_PAGE_NX
;
607 early_param("noexec", noexec_setup
);
609 static void __init
set_nx(void)
611 unsigned int v
[4], l
, h
;
613 if (cpu_has_pae
&& (cpuid_eax(0x80000000) > 0x80000001)) {
614 cpuid(0x80000001, &v
[0], &v
[1], &v
[2], &v
[3]);
616 if ((v
[3] & (1 << 20)) && !disable_nx
) {
617 rdmsr(MSR_EFER
, l
, h
);
619 wrmsr(MSR_EFER
, l
, h
);
621 __supported_pte_mask
|= _PAGE_NX
;
627 /* user-defined highmem size */
628 static unsigned int highmem_pages
= -1;
631 * highmem=size forces highmem to be exactly 'size' bytes.
632 * This works even on boxes that have no highmem otherwise.
633 * This also works to reduce highmem size on bigger boxes.
635 static int __init
parse_highmem(char *arg
)
640 highmem_pages
= memparse(arg
, &arg
) >> PAGE_SHIFT
;
643 early_param("highmem", parse_highmem
);
645 #define MSG_HIGHMEM_TOO_BIG \
646 "highmem size (%luMB) is bigger than pages available (%luMB)!\n"
648 #define MSG_LOWMEM_TOO_SMALL \
649 "highmem size (%luMB) results in <64MB lowmem, ignoring it!\n"
651 * All of RAM fits into lowmem - but if user wants highmem
652 * artificially via the highmem=x boot parameter then create
655 void __init
lowmem_pfn_init(void)
657 /* max_low_pfn is 0, we already have early_res support */
658 max_low_pfn
= max_pfn
;
660 if (highmem_pages
== -1)
662 #ifdef CONFIG_HIGHMEM
663 if (highmem_pages
>= max_pfn
) {
664 printk(KERN_ERR MSG_HIGHMEM_TOO_BIG
,
665 pages_to_mb(highmem_pages
), pages_to_mb(max_pfn
));
669 if (max_low_pfn
- highmem_pages
< 64*1024*1024/PAGE_SIZE
) {
670 printk(KERN_ERR MSG_LOWMEM_TOO_SMALL
,
671 pages_to_mb(highmem_pages
));
674 max_low_pfn
-= highmem_pages
;
678 printk(KERN_ERR
"ignoring highmem size on non-highmem kernel!\n");
682 #define MSG_HIGHMEM_TOO_SMALL \
683 "only %luMB highmem pages available, ignoring highmem size of %luMB!\n"
685 #define MSG_HIGHMEM_TRIMMED \
686 "Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n"
688 * We have more RAM than fits into lowmem - we try to put it into
689 * highmem, also taking the highmem=x boot parameter into account:
691 void __init
highmem_pfn_init(void)
693 max_low_pfn
= MAXMEM_PFN
;
695 if (highmem_pages
== -1)
696 highmem_pages
= max_pfn
- MAXMEM_PFN
;
698 if (highmem_pages
+ MAXMEM_PFN
< max_pfn
)
699 max_pfn
= MAXMEM_PFN
+ highmem_pages
;
701 if (highmem_pages
+ MAXMEM_PFN
> max_pfn
) {
702 printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL
,
703 pages_to_mb(max_pfn
- MAXMEM_PFN
),
704 pages_to_mb(highmem_pages
));
707 #ifndef CONFIG_HIGHMEM
708 /* Maximum memory usable is what is directly addressable */
709 printk(KERN_WARNING
"Warning only %ldMB will be used.\n", MAXMEM
>>20);
710 if (max_pfn
> MAX_NONPAE_PFN
)
711 printk(KERN_WARNING
"Use a HIGHMEM64G enabled kernel.\n");
713 printk(KERN_WARNING
"Use a HIGHMEM enabled kernel.\n");
714 max_pfn
= MAXMEM_PFN
;
715 #else /* !CONFIG_HIGHMEM */
716 #ifndef CONFIG_HIGHMEM64G
717 if (max_pfn
> MAX_NONPAE_PFN
) {
718 max_pfn
= MAX_NONPAE_PFN
;
719 printk(KERN_WARNING MSG_HIGHMEM_TRIMMED
);
721 #endif /* !CONFIG_HIGHMEM64G */
722 #endif /* !CONFIG_HIGHMEM */
726 * Determine low and high memory ranges:
728 void __init
find_low_pfn_range(void)
730 /* it could update max_pfn */
732 if (max_pfn
<= MAXMEM_PFN
)
738 #ifndef CONFIG_NEED_MULTIPLE_NODES
739 void __init
initmem_init(unsigned long start_pfn
,
740 unsigned long end_pfn
)
742 #ifdef CONFIG_HIGHMEM
743 highstart_pfn
= highend_pfn
= max_pfn
;
744 if (max_pfn
> max_low_pfn
)
745 highstart_pfn
= max_low_pfn
;
746 memory_present(0, 0, highend_pfn
);
747 e820_register_active_regions(0, 0, highend_pfn
);
748 printk(KERN_NOTICE
"%ldMB HIGHMEM available.\n",
749 pages_to_mb(highend_pfn
- highstart_pfn
));
750 num_physpages
= highend_pfn
;
751 high_memory
= (void *) __va(highstart_pfn
* PAGE_SIZE
- 1) + 1;
753 memory_present(0, 0, max_low_pfn
);
754 e820_register_active_regions(0, 0, max_low_pfn
);
755 num_physpages
= max_low_pfn
;
756 high_memory
= (void *) __va(max_low_pfn
* PAGE_SIZE
- 1) + 1;
758 #ifdef CONFIG_FLATMEM
759 max_mapnr
= num_physpages
;
761 printk(KERN_NOTICE
"%ldMB LOWMEM available.\n",
762 pages_to_mb(max_low_pfn
));
764 setup_bootmem_allocator();
766 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
768 static void __init
zone_sizes_init(void)
770 unsigned long max_zone_pfns
[MAX_NR_ZONES
];
771 memset(max_zone_pfns
, 0, sizeof(max_zone_pfns
));
772 max_zone_pfns
[ZONE_DMA
] =
773 virt_to_phys((char *)MAX_DMA_ADDRESS
) >> PAGE_SHIFT
;
774 max_zone_pfns
[ZONE_NORMAL
] = max_low_pfn
;
775 #ifdef CONFIG_HIGHMEM
776 max_zone_pfns
[ZONE_HIGHMEM
] = highend_pfn
;
779 free_area_init_nodes(max_zone_pfns
);
782 static unsigned long __init
setup_node_bootmem(int nodeid
,
783 unsigned long start_pfn
,
784 unsigned long end_pfn
,
785 unsigned long bootmap
)
787 unsigned long bootmap_size
;
789 if (start_pfn
> max_low_pfn
)
791 if (end_pfn
> max_low_pfn
)
792 end_pfn
= max_low_pfn
;
794 /* don't touch min_low_pfn */
795 bootmap_size
= init_bootmem_node(NODE_DATA(nodeid
),
796 bootmap
>> PAGE_SHIFT
,
798 printk(KERN_INFO
" node %d low ram: %08lx - %08lx\n",
799 nodeid
, start_pfn
<<PAGE_SHIFT
, end_pfn
<<PAGE_SHIFT
);
800 printk(KERN_INFO
" node %d bootmap %08lx - %08lx\n",
801 nodeid
, bootmap
, bootmap
+ bootmap_size
);
802 free_bootmem_with_active_regions(nodeid
, end_pfn
);
803 early_res_to_bootmem(start_pfn
<<PAGE_SHIFT
, end_pfn
<<PAGE_SHIFT
);
805 return bootmap
+ bootmap_size
;
808 void __init
setup_bootmem_allocator(void)
811 unsigned long bootmap_size
, bootmap
;
813 * Initialize the boot-time allocator (with low memory only):
815 bootmap_size
= bootmem_bootmap_pages(max_low_pfn
)<<PAGE_SHIFT
;
816 bootmap
= find_e820_area(0, max_pfn_mapped
<<PAGE_SHIFT
, bootmap_size
,
819 panic("Cannot find bootmem map of size %ld\n", bootmap_size
);
820 reserve_early(bootmap
, bootmap
+ bootmap_size
, "BOOTMAP");
822 printk(KERN_INFO
" mapped low ram: 0 - %08lx\n",
823 max_pfn_mapped
<<PAGE_SHIFT
);
824 printk(KERN_INFO
" low ram: 0 - %08lx\n", max_low_pfn
<<PAGE_SHIFT
);
826 #ifdef CONFIG_NEED_MULTIPLE_NODES
827 for_each_online_node(nodeid
)
828 bootmap
= setup_node_bootmem(nodeid
, node_start_pfn
[nodeid
],
829 node_end_pfn
[nodeid
], bootmap
);
831 bootmap
= setup_node_bootmem(0, 0, max_low_pfn
, bootmap
);
837 static void __init
find_early_table_space(unsigned long end
, int use_pse
,
840 unsigned long puds
, pmds
, ptes
, tables
, start
;
842 puds
= (end
+ PUD_SIZE
- 1) >> PUD_SHIFT
;
843 tables
= roundup(puds
* sizeof(pud_t
), PAGE_SIZE
);
848 extra
= end
- ((end
>>PUD_SHIFT
) << PUD_SHIFT
);
849 pmds
= (extra
+ PMD_SIZE
- 1) >> PMD_SHIFT
;
851 pmds
= (end
+ PMD_SIZE
- 1) >> PMD_SHIFT
;
853 tables
+= roundup(pmds
* sizeof(pmd_t
), PAGE_SIZE
);
858 extra
= end
- ((end
>>PMD_SHIFT
) << PMD_SHIFT
);
862 ptes
= (extra
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
864 ptes
= (end
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
866 tables
+= roundup(ptes
* sizeof(pte_t
), PAGE_SIZE
);
870 tables
+= roundup(__end_of_fixed_addresses
* sizeof(pte_t
), PAGE_SIZE
);
874 * RED-PEN putting page tables only on node 0 could
875 * cause a hotspot and fill up ZONE_DMA. The page tables
876 * need roughly 0.5KB per GB.
880 table_start
= find_e820_area(start
, max_pfn_mapped
<<PAGE_SHIFT
,
882 #else /* CONFIG_X86_64 */
884 table_start
= find_e820_area(start
, end
, tables
, PAGE_SIZE
);
886 if (table_start
== -1UL)
887 panic("Cannot find space for the kernel page tables");
889 table_start
>>= PAGE_SHIFT
;
890 table_end
= table_start
;
891 table_top
= table_start
+ (tables
>> PAGE_SHIFT
);
893 printk(KERN_DEBUG
"kernel direct mapping tables up to %lx @ %lx-%lx\n",
894 end
, table_start
<< PAGE_SHIFT
, table_top
<< PAGE_SHIFT
);
900 unsigned page_size_mask
;
903 #define NR_RANGE_MR 3
905 static int save_mr(struct map_range
*mr
, int nr_range
,
906 unsigned long start_pfn
, unsigned long end_pfn
,
907 unsigned long page_size_mask
)
909 if (start_pfn
< end_pfn
) {
910 if (nr_range
>= NR_RANGE_MR
)
911 panic("run out of range for init_memory_mapping\n");
912 mr
[nr_range
].start
= start_pfn
<<PAGE_SHIFT
;
913 mr
[nr_range
].end
= end_pfn
<<PAGE_SHIFT
;
914 mr
[nr_range
].page_size_mask
= page_size_mask
;
922 * Setup the direct mapping of the physical memory at PAGE_OFFSET.
923 * This runs before bootmem is initialized and gets pages directly from
924 * the physical memory. To access them they are temporarily mapped.
926 unsigned long __init_refok
init_memory_mapping(unsigned long start
,
929 unsigned long page_size_mask
= 0;
930 unsigned long start_pfn
, end_pfn
;
934 struct map_range mr
[NR_RANGE_MR
];
936 int use_pse
, use_gbpages
;
938 printk(KERN_INFO
"init_memory_mapping: %016lx-%016lx\n", start
, end
);
940 #ifdef CONFIG_DEBUG_PAGEALLOC
942 * For CONFIG_DEBUG_PAGEALLOC, identity mapping will use small pages.
943 * This will simplify cpa(), which otherwise needs to support splitting
944 * large pages into small in interrupt context, etc.
946 use_pse
= use_gbpages
= 0;
948 use_pse
= cpu_has_pse
;
949 use_gbpages
= direct_gbpages
;
953 #ifdef CONFIG_X86_PAE
956 printk(KERN_INFO
"NX (Execute Disable) protection: active\n");
959 /* Enable PSE if available */
961 set_in_cr4(X86_CR4_PSE
);
963 /* Enable PGE if available */
965 set_in_cr4(X86_CR4_PGE
);
966 __supported_pte_mask
|= _PAGE_GLOBAL
;
971 page_size_mask
|= 1 << PG_LEVEL_1G
;
973 page_size_mask
|= 1 << PG_LEVEL_2M
;
975 memset(mr
, 0, sizeof(mr
));
979 * Don't use a large page for the first 2/4MB of memory
980 * because there are often fixed size MTRRs in there
981 * and overlapping MTRRs into large pages can cause
984 /* head if not big page alignment ? */
985 start_pfn
= start
>> PAGE_SHIFT
;
986 pos
= start_pfn
<< PAGE_SHIFT
;
988 end_pfn
= 1<<(PMD_SHIFT
- PAGE_SHIFT
);
990 end_pfn
= ((pos
+ (PMD_SIZE
- 1))>>PMD_SHIFT
)
991 << (PMD_SHIFT
- PAGE_SHIFT
);
992 if (end_pfn
> (end
>> PAGE_SHIFT
))
993 end_pfn
= end
>> PAGE_SHIFT
;
994 if (start_pfn
< end_pfn
) {
995 nr_range
= save_mr(mr
, nr_range
, start_pfn
, end_pfn
, 0);
996 pos
= end_pfn
<< PAGE_SHIFT
;
999 /* big page (2M) range */
1000 start_pfn
= ((pos
+ (PMD_SIZE
- 1))>>PMD_SHIFT
)
1001 << (PMD_SHIFT
- PAGE_SHIFT
);
1002 end_pfn
= (end
>>PMD_SHIFT
) << (PMD_SHIFT
- PAGE_SHIFT
);
1003 if (start_pfn
< end_pfn
) {
1004 nr_range
= save_mr(mr
, nr_range
, start_pfn
, end_pfn
,
1005 page_size_mask
& (1<<PG_LEVEL_2M
));
1006 pos
= end_pfn
<< PAGE_SHIFT
;
1009 /* tail is not big page (2M) alignment */
1010 start_pfn
= pos
>>PAGE_SHIFT
;
1011 end_pfn
= end
>>PAGE_SHIFT
;
1012 nr_range
= save_mr(mr
, nr_range
, start_pfn
, end_pfn
, 0);
1014 /* try to merge same page size and continuous */
1015 for (i
= 0; nr_range
> 1 && i
< nr_range
- 1; i
++) {
1016 unsigned long old_start
;
1017 if (mr
[i
].end
!= mr
[i
+1].start
||
1018 mr
[i
].page_size_mask
!= mr
[i
+1].page_size_mask
)
1021 old_start
= mr
[i
].start
;
1022 memmove(&mr
[i
], &mr
[i
+1],
1023 (nr_range
- 1 - i
) * sizeof(struct map_range
));
1024 mr
[i
--].start
= old_start
;
1028 for (i
= 0; i
< nr_range
; i
++)
1029 printk(KERN_DEBUG
" %010lx - %010lx page %s\n",
1030 mr
[i
].start
, mr
[i
].end
,
1031 (mr
[i
].page_size_mask
& (1<<PG_LEVEL_1G
))?"1G":(
1032 (mr
[i
].page_size_mask
& (1<<PG_LEVEL_2M
))?"2M":"4k"));
1035 * Find space for the kernel direct mapping tables.
1037 * Later we should allocate these tables in the local node of the
1038 * memory mapped. Unfortunately this is done currently before the
1039 * nodes are discovered.
1042 find_early_table_space(end
, use_pse
, use_gbpages
);
1044 #ifdef CONFIG_X86_32
1045 for (i
= 0; i
< nr_range
; i
++)
1046 kernel_physical_mapping_init(
1047 mr
[i
].start
>> PAGE_SHIFT
,
1048 mr
[i
].end
>> PAGE_SHIFT
,
1049 mr
[i
].page_size_mask
== (1<<PG_LEVEL_2M
));
1051 #else /* CONFIG_X86_64 */
1052 for (i
= 0; i
< nr_range
; i
++)
1053 ret
= kernel_physical_mapping_init(mr
[i
].start
, mr
[i
].end
,
1054 mr
[i
].page_size_mask
);
1057 #ifdef CONFIG_X86_32
1058 early_ioremap_page_table_range_init();
1060 load_cr3(swapper_pg_dir
);
1063 #ifdef CONFIG_X86_64
1065 mmu_cr4_features
= read_cr4();
1069 if (!after_bootmem
&& table_end
> table_start
)
1070 reserve_early(table_start
<< PAGE_SHIFT
,
1071 table_end
<< PAGE_SHIFT
, "PGTABLE");
1074 early_memtest(start
, end
);
1076 return ret
>> PAGE_SHIFT
;
1081 * paging_init() sets up the page tables - note that the first 8MB are
1082 * already mapped by head.S.
1084 * This routines also unmaps the page at virtual kernel address 0, so
1085 * that we can trap those pesky NULL-reference errors in the kernel.
1087 void __init
paging_init(void)
1096 * NOTE: at this point the bootmem allocator is fully available.
1103 * Test if the WP bit works in supervisor mode. It isn't supported on 386's
1104 * and also on some strange 486's. All 586+'s are OK. This used to involve
1105 * black magic jumps to work around some nasty CPU bugs, but fortunately the
1106 * switch to using exceptions got rid of all that.
1108 static void __init
test_wp_bit(void)
1111 "Checking if this processor honours the WP bit even in supervisor mode...");
1113 /* Any page-aligned address will do, the test is non-destructive */
1114 __set_fixmap(FIX_WP_TEST
, __pa(&swapper_pg_dir
), PAGE_READONLY
);
1115 boot_cpu_data
.wp_works_ok
= do_test_wp_bit();
1116 clear_fixmap(FIX_WP_TEST
);
1118 if (!boot_cpu_data
.wp_works_ok
) {
1119 printk(KERN_CONT
"No.\n");
1120 #ifdef CONFIG_X86_WP_WORKS_OK
1122 "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
1125 printk(KERN_CONT
"Ok.\n");
1129 static struct kcore_list kcore_mem
, kcore_vmalloc
;
1131 void __init
mem_init(void)
1133 int codesize
, reservedpages
, datasize
, initsize
;
1138 #ifdef CONFIG_FLATMEM
1141 /* this will put all low memory onto the freelists */
1142 totalram_pages
+= free_all_bootmem();
1145 for (tmp
= 0; tmp
< max_low_pfn
; tmp
++)
1147 * Only count reserved RAM pages:
1149 if (page_is_ram(tmp
) && PageReserved(pfn_to_page(tmp
)))
1152 set_highmem_pages_init();
1154 codesize
= (unsigned long) &_etext
- (unsigned long) &_text
;
1155 datasize
= (unsigned long) &_edata
- (unsigned long) &_etext
;
1156 initsize
= (unsigned long) &__init_end
- (unsigned long) &__init_begin
;
1158 kclist_add(&kcore_mem
, __va(0), max_low_pfn
<< PAGE_SHIFT
);
1159 kclist_add(&kcore_vmalloc
, (void *)VMALLOC_START
,
1160 VMALLOC_END
-VMALLOC_START
);
1162 printk(KERN_INFO
"Memory: %luk/%luk available (%dk kernel code, "
1163 "%dk reserved, %dk data, %dk init, %ldk highmem)\n",
1164 (unsigned long) nr_free_pages() << (PAGE_SHIFT
-10),
1165 num_physpages
<< (PAGE_SHIFT
-10),
1167 reservedpages
<< (PAGE_SHIFT
-10),
1170 (unsigned long) (totalhigh_pages
<< (PAGE_SHIFT
-10))
1173 printk(KERN_INFO
"virtual kernel memory layout:\n"
1174 " fixmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
1175 #ifdef CONFIG_HIGHMEM
1176 " pkmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
1178 " vmalloc : 0x%08lx - 0x%08lx (%4ld MB)\n"
1179 " lowmem : 0x%08lx - 0x%08lx (%4ld MB)\n"
1180 " .init : 0x%08lx - 0x%08lx (%4ld kB)\n"
1181 " .data : 0x%08lx - 0x%08lx (%4ld kB)\n"
1182 " .text : 0x%08lx - 0x%08lx (%4ld kB)\n",
1183 FIXADDR_START
, FIXADDR_TOP
,
1184 (FIXADDR_TOP
- FIXADDR_START
) >> 10,
1186 #ifdef CONFIG_HIGHMEM
1187 PKMAP_BASE
, PKMAP_BASE
+LAST_PKMAP
*PAGE_SIZE
,
1188 (LAST_PKMAP
*PAGE_SIZE
) >> 10,
1191 VMALLOC_START
, VMALLOC_END
,
1192 (VMALLOC_END
- VMALLOC_START
) >> 20,
1194 (unsigned long)__va(0), (unsigned long)high_memory
,
1195 ((unsigned long)high_memory
- (unsigned long)__va(0)) >> 20,
1197 (unsigned long)&__init_begin
, (unsigned long)&__init_end
,
1198 ((unsigned long)&__init_end
-
1199 (unsigned long)&__init_begin
) >> 10,
1201 (unsigned long)&_etext
, (unsigned long)&_edata
,
1202 ((unsigned long)&_edata
- (unsigned long)&_etext
) >> 10,
1204 (unsigned long)&_text
, (unsigned long)&_etext
,
1205 ((unsigned long)&_etext
- (unsigned long)&_text
) >> 10);
1208 * Check boundaries twice: Some fundamental inconsistencies can
1209 * be detected at build time already.
1211 #define __FIXADDR_TOP (-PAGE_SIZE)
1212 #ifdef CONFIG_HIGHMEM
1213 BUILD_BUG_ON(PKMAP_BASE
+ LAST_PKMAP
*PAGE_SIZE
> FIXADDR_START
);
1214 BUILD_BUG_ON(VMALLOC_END
> PKMAP_BASE
);
1216 #define high_memory (-128UL << 20)
1217 BUILD_BUG_ON(VMALLOC_START
>= VMALLOC_END
);
1219 #undef __FIXADDR_TOP
1221 #ifdef CONFIG_HIGHMEM
1222 BUG_ON(PKMAP_BASE
+ LAST_PKMAP
*PAGE_SIZE
> FIXADDR_START
);
1223 BUG_ON(VMALLOC_END
> PKMAP_BASE
);
1225 BUG_ON(VMALLOC_START
>= VMALLOC_END
);
1226 BUG_ON((unsigned long)high_memory
> VMALLOC_START
);
1228 if (boot_cpu_data
.wp_works_ok
< 0)
1235 #ifdef CONFIG_MEMORY_HOTPLUG
1236 int arch_add_memory(int nid
, u64 start
, u64 size
)
1238 struct pglist_data
*pgdata
= NODE_DATA(nid
);
1239 struct zone
*zone
= pgdata
->node_zones
+ ZONE_HIGHMEM
;
1240 unsigned long start_pfn
= start
>> PAGE_SHIFT
;
1241 unsigned long nr_pages
= size
>> PAGE_SHIFT
;
1243 return __add_pages(nid
, zone
, start_pfn
, nr_pages
);
1248 * This function cannot be __init, since exceptions don't work in that
1249 * section. Put this after the callers, so that it cannot be inlined.
1251 static noinline
int do_test_wp_bit(void)
1256 __asm__
__volatile__(
1262 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST
)),
1271 #ifdef CONFIG_DEBUG_RODATA
1272 const int rodata_test_data
= 0xC3;
1273 EXPORT_SYMBOL_GPL(rodata_test_data
);
1275 void mark_rodata_ro(void)
1277 unsigned long start
= PFN_ALIGN(_text
);
1278 unsigned long size
= PFN_ALIGN(_etext
) - start
;
1280 #ifndef CONFIG_DYNAMIC_FTRACE
1281 /* Dynamic tracing modifies the kernel text section */
1282 set_pages_ro(virt_to_page(start
), size
>> PAGE_SHIFT
);
1283 printk(KERN_INFO
"Write protecting the kernel text: %luk\n",
1286 #ifdef CONFIG_CPA_DEBUG
1287 printk(KERN_INFO
"Testing CPA: Reverting %lx-%lx\n",
1289 set_pages_rw(virt_to_page(start
), size
>>PAGE_SHIFT
);
1291 printk(KERN_INFO
"Testing CPA: write protecting again\n");
1292 set_pages_ro(virt_to_page(start
), size
>>PAGE_SHIFT
);
1294 #endif /* CONFIG_DYNAMIC_FTRACE */
1297 size
= (unsigned long)__end_rodata
- start
;
1298 set_pages_ro(virt_to_page(start
), size
>> PAGE_SHIFT
);
1299 printk(KERN_INFO
"Write protecting the kernel read-only data: %luk\n",
1303 #ifdef CONFIG_CPA_DEBUG
1304 printk(KERN_INFO
"Testing CPA: undo %lx-%lx\n", start
, start
+ size
);
1305 set_pages_rw(virt_to_page(start
), size
>> PAGE_SHIFT
);
1307 printk(KERN_INFO
"Testing CPA: write protecting again\n");
1308 set_pages_ro(virt_to_page(start
), size
>> PAGE_SHIFT
);
1313 int __init
reserve_bootmem_generic(unsigned long phys
, unsigned long len
,
1316 return reserve_bootmem(phys
, len
, flags
);