2 * linux/arch/i386/mm/init.c
4 * Copyright (C) 1995 Linus Torvalds
6 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
9 #include <linux/config.h>
10 #include <linux/module.h>
11 #include <linux/signal.h>
12 #include <linux/sched.h>
13 #include <linux/kernel.h>
14 #include <linux/errno.h>
15 #include <linux/string.h>
16 #include <linux/types.h>
17 #include <linux/ptrace.h>
18 #include <linux/mman.h>
20 #include <linux/hugetlb.h>
21 #include <linux/swap.h>
22 #include <linux/smp.h>
23 #include <linux/init.h>
24 #include <linux/highmem.h>
25 #include <linux/pagemap.h>
26 #include <linux/bootmem.h>
27 #include <linux/slab.h>
28 #include <linux/proc_fs.h>
29 #include <linux/efi.h>
30 #include <linux/memory_hotplug.h>
31 #include <linux/initrd.h>
32 #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>
43 #include <asm/tlbflush.h>
44 #include <asm/sections.h>
46 unsigned int __VMALLOC_RESERVE
= 128 << 20;
48 DEFINE_PER_CPU(struct mmu_gather
, mmu_gathers
);
49 unsigned long highstart_pfn
, highend_pfn
;
51 static int noinline
do_test_wp_bit(void);
54 * Creates a middle page table and puts a pointer to it in the
55 * given global directory entry. This only returns the gd entry
56 * in non-PAE compilation mode, since the middle layer is folded.
58 static pmd_t
* __init
one_md_table_init(pgd_t
*pgd
)
64 pmd_table
= (pmd_t
*) alloc_bootmem_low_pages(PAGE_SIZE
);
65 set_pgd(pgd
, __pgd(__pa(pmd_table
) | _PAGE_PRESENT
));
66 pud
= pud_offset(pgd
, 0);
67 if (pmd_table
!= pmd_offset(pud
, 0))
70 pud
= pud_offset(pgd
, 0);
71 pmd_table
= pmd_offset(pud
, 0);
78 * Create a page table and place a pointer to it in a middle page
81 static pte_t
* __init
one_page_table_init(pmd_t
*pmd
)
84 pte_t
*page_table
= (pte_t
*) alloc_bootmem_low_pages(PAGE_SIZE
);
85 set_pmd(pmd
, __pmd(__pa(page_table
) | _PAGE_TABLE
));
86 if (page_table
!= pte_offset_kernel(pmd
, 0))
92 return pte_offset_kernel(pmd
, 0);
96 * This function initializes a certain range of kernel virtual memory
97 * with new bootmem page tables, everywhere page tables are missing in
102 * NOTE: The pagetables are allocated contiguous on the physical space
103 * so we can cache the place of the first one and move around without
104 * checking the pgd every time.
106 static void __init
page_table_range_init (unsigned long start
, unsigned long end
, pgd_t
*pgd_base
)
111 int pgd_idx
, pmd_idx
;
115 pgd_idx
= pgd_index(vaddr
);
116 pmd_idx
= pmd_index(vaddr
);
117 pgd
= pgd_base
+ pgd_idx
;
119 for ( ; (pgd_idx
< PTRS_PER_PGD
) && (vaddr
!= end
); pgd
++, pgd_idx
++) {
121 one_md_table_init(pgd
);
122 pud
= pud_offset(pgd
, vaddr
);
123 pmd
= pmd_offset(pud
, vaddr
);
124 for (; (pmd_idx
< PTRS_PER_PMD
) && (vaddr
!= end
); pmd
++, pmd_idx
++) {
126 one_page_table_init(pmd
);
134 static inline int is_kernel_text(unsigned long addr
)
136 if (addr
>= PAGE_OFFSET
&& addr
<= (unsigned long)__init_end
)
142 * This maps the physical memory to kernel virtual address space, a total
143 * of max_low_pfn pages, by creating page tables starting from address
146 static void __init
kernel_physical_mapping_init(pgd_t
*pgd_base
)
152 int pgd_idx
, pmd_idx
, pte_ofs
;
154 pgd_idx
= pgd_index(PAGE_OFFSET
);
155 pgd
= pgd_base
+ pgd_idx
;
158 for (; pgd_idx
< PTRS_PER_PGD
; pgd
++, pgd_idx
++) {
159 pmd
= one_md_table_init(pgd
);
160 if (pfn
>= max_low_pfn
)
162 for (pmd_idx
= 0; pmd_idx
< PTRS_PER_PMD
&& pfn
< max_low_pfn
; pmd
++, pmd_idx
++) {
163 unsigned int address
= pfn
* PAGE_SIZE
+ PAGE_OFFSET
;
165 /* Map with big pages if possible, otherwise create normal page tables. */
167 unsigned int address2
= (pfn
+ PTRS_PER_PTE
- 1) * PAGE_SIZE
+ PAGE_OFFSET
+ PAGE_SIZE
-1;
169 if (is_kernel_text(address
) || is_kernel_text(address2
))
170 set_pmd(pmd
, pfn_pmd(pfn
, PAGE_KERNEL_LARGE_EXEC
));
172 set_pmd(pmd
, pfn_pmd(pfn
, PAGE_KERNEL_LARGE
));
175 pte
= one_page_table_init(pmd
);
177 for (pte_ofs
= 0; pte_ofs
< PTRS_PER_PTE
&& pfn
< max_low_pfn
; pte
++, pfn
++, pte_ofs
++) {
178 if (is_kernel_text(address
))
179 set_pte(pte
, pfn_pte(pfn
, PAGE_KERNEL_EXEC
));
181 set_pte(pte
, pfn_pte(pfn
, PAGE_KERNEL
));
188 static inline int page_kills_ppro(unsigned long pagenr
)
190 if (pagenr
>= 0x70000 && pagenr
<= 0x7003F)
195 extern int is_available_memory(efi_memory_desc_t
*);
197 int page_is_ram(unsigned long pagenr
)
200 unsigned long addr
, end
;
203 efi_memory_desc_t
*md
;
206 for (p
= memmap
.map
; p
< memmap
.map_end
; p
+= memmap
.desc_size
) {
208 if (!is_available_memory(md
))
210 addr
= (md
->phys_addr
+PAGE_SIZE
-1) >> PAGE_SHIFT
;
211 end
= (md
->phys_addr
+ (md
->num_pages
<< EFI_PAGE_SHIFT
)) >> PAGE_SHIFT
;
213 if ((pagenr
>= addr
) && (pagenr
< end
))
219 for (i
= 0; i
< e820
.nr_map
; i
++) {
221 if (e820
.map
[i
].type
!= E820_RAM
) /* not usable memory */
224 * !!!FIXME!!! Some BIOSen report areas as RAM that
225 * are not. Notably the 640->1Mb area. We need a sanity
228 addr
= (e820
.map
[i
].addr
+PAGE_SIZE
-1) >> PAGE_SHIFT
;
229 end
= (e820
.map
[i
].addr
+e820
.map
[i
].size
) >> PAGE_SHIFT
;
230 if ((pagenr
>= addr
) && (pagenr
< end
))
236 #ifdef CONFIG_HIGHMEM
240 #define kmap_get_fixmap_pte(vaddr) \
241 pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr), vaddr), (vaddr)), (vaddr))
243 static void __init
kmap_init(void)
245 unsigned long kmap_vstart
;
247 /* cache the first kmap pte */
248 kmap_vstart
= __fix_to_virt(FIX_KMAP_BEGIN
);
249 kmap_pte
= kmap_get_fixmap_pte(kmap_vstart
);
251 kmap_prot
= PAGE_KERNEL
;
254 static void __init
permanent_kmaps_init(pgd_t
*pgd_base
)
263 page_table_range_init(vaddr
, vaddr
+ PAGE_SIZE
*LAST_PKMAP
, pgd_base
);
265 pgd
= swapper_pg_dir
+ pgd_index(vaddr
);
266 pud
= pud_offset(pgd
, vaddr
);
267 pmd
= pmd_offset(pud
, vaddr
);
268 pte
= pte_offset_kernel(pmd
, vaddr
);
269 pkmap_page_table
= pte
;
272 static void __meminit
free_new_highpage(struct page
*page
)
274 init_page_count(page
);
279 void __init
add_one_highpage_init(struct page
*page
, int pfn
, int bad_ppro
)
281 if (page_is_ram(pfn
) && !(bad_ppro
&& page_kills_ppro(pfn
))) {
282 ClearPageReserved(page
);
283 free_new_highpage(page
);
285 SetPageReserved(page
);
288 static int add_one_highpage_hotplug(struct page
*page
, unsigned long pfn
)
290 free_new_highpage(page
);
292 #ifdef CONFIG_FLATMEM
293 max_mapnr
= max(pfn
, max_mapnr
);
300 * Not currently handling the NUMA case.
301 * Assuming single node and all memory that
302 * has been added dynamically that would be
303 * onlined here is in HIGHMEM
305 void online_page(struct page
*page
)
307 ClearPageReserved(page
);
308 add_one_highpage_hotplug(page
, page_to_pfn(page
));
313 extern void set_highmem_pages_init(int);
315 static void __init
set_highmem_pages_init(int bad_ppro
)
318 for (pfn
= highstart_pfn
; pfn
< highend_pfn
; pfn
++)
319 add_one_highpage_init(pfn_to_page(pfn
), pfn
, bad_ppro
);
320 totalram_pages
+= totalhigh_pages
;
322 #endif /* CONFIG_FLATMEM */
325 #define kmap_init() do { } while (0)
326 #define permanent_kmaps_init(pgd_base) do { } while (0)
327 #define set_highmem_pages_init(bad_ppro) do { } while (0)
328 #endif /* CONFIG_HIGHMEM */
330 unsigned long long __PAGE_KERNEL
= _PAGE_KERNEL
;
331 EXPORT_SYMBOL(__PAGE_KERNEL
);
332 unsigned long long __PAGE_KERNEL_EXEC
= _PAGE_KERNEL_EXEC
;
335 extern void __init
remap_numa_kva(void);
337 #define remap_numa_kva() do {} while (0)
340 static void __init
pagetable_init (void)
343 pgd_t
*pgd_base
= swapper_pg_dir
;
345 #ifdef CONFIG_X86_PAE
347 /* Init entries of the first-level page table to the zero page */
348 for (i
= 0; i
< PTRS_PER_PGD
; i
++)
349 set_pgd(pgd_base
+ i
, __pgd(__pa(empty_zero_page
) | _PAGE_PRESENT
));
352 /* Enable PSE if available */
354 set_in_cr4(X86_CR4_PSE
);
357 /* Enable PGE if available */
359 set_in_cr4(X86_CR4_PGE
);
360 __PAGE_KERNEL
|= _PAGE_GLOBAL
;
361 __PAGE_KERNEL_EXEC
|= _PAGE_GLOBAL
;
364 kernel_physical_mapping_init(pgd_base
);
368 * Fixed mappings, only the page table structure has to be
369 * created - mappings will be set by set_fixmap():
371 vaddr
= __fix_to_virt(__end_of_fixed_addresses
- 1) & PMD_MASK
;
372 page_table_range_init(vaddr
, 0, pgd_base
);
374 permanent_kmaps_init(pgd_base
);
376 #ifdef CONFIG_X86_PAE
378 * Add low memory identity-mappings - SMP needs it when
379 * starting up on an AP from real-mode. In the non-PAE
380 * case we already have these mappings through head.S.
381 * All user-space mappings are explicitly cleared after
384 set_pgd(&pgd_base
[0], pgd_base
[USER_PTRS_PER_PGD
]);
388 #if defined(CONFIG_SOFTWARE_SUSPEND) || defined(CONFIG_ACPI_SLEEP)
390 * Swap suspend & friends need this for resume because things like the intel-agp
391 * driver might have split up a kernel 4MB mapping.
393 char __nosavedata swsusp_pg_dir
[PAGE_SIZE
]
394 __attribute__ ((aligned (PAGE_SIZE
)));
396 static inline void save_pg_dir(void)
398 memcpy(swsusp_pg_dir
, swapper_pg_dir
, PAGE_SIZE
);
401 static inline void save_pg_dir(void)
406 void zap_low_mappings (void)
413 * Zap initial low-memory mappings.
415 * Note that "pgd_clear()" doesn't do it for
416 * us, because pgd_clear() is a no-op on i386.
418 for (i
= 0; i
< USER_PTRS_PER_PGD
; i
++)
419 #ifdef CONFIG_X86_PAE
420 set_pgd(swapper_pg_dir
+i
, __pgd(1 + __pa(empty_zero_page
)));
422 set_pgd(swapper_pg_dir
+i
, __pgd(0));
427 static int disable_nx __initdata
= 0;
428 u64 __supported_pte_mask __read_mostly
= ~_PAGE_NX
;
433 * Control non executable mappings.
438 void __init
noexec_setup(const char *str
)
440 if (!strncmp(str
, "on",2) && cpu_has_nx
) {
441 __supported_pte_mask
|= _PAGE_NX
;
443 } else if (!strncmp(str
,"off",3)) {
445 __supported_pte_mask
&= ~_PAGE_NX
;
450 #ifdef CONFIG_X86_PAE
452 static void __init
set_nx(void)
454 unsigned int v
[4], l
, h
;
456 if (cpu_has_pae
&& (cpuid_eax(0x80000000) > 0x80000001)) {
457 cpuid(0x80000001, &v
[0], &v
[1], &v
[2], &v
[3]);
458 if ((v
[3] & (1 << 20)) && !disable_nx
) {
459 rdmsr(MSR_EFER
, l
, h
);
461 wrmsr(MSR_EFER
, l
, h
);
463 __supported_pte_mask
|= _PAGE_NX
;
469 * Enables/disables executability of a given kernel page and
470 * returns the previous setting.
472 int __init
set_kernel_exec(unsigned long vaddr
, int enable
)
480 pte
= lookup_address(vaddr
);
483 if (!pte_exec_kernel(*pte
))
487 pte
->pte_high
&= ~(1 << (_PAGE_BIT_NX
- 32));
489 pte
->pte_high
|= 1 << (_PAGE_BIT_NX
- 32);
498 * paging_init() sets up the page tables - note that the first 8MB are
499 * already mapped by head.S.
501 * This routines also unmaps the page at virtual kernel address 0, so
502 * that we can trap those pesky NULL-reference errors in the kernel.
504 void __init
paging_init(void)
506 #ifdef CONFIG_X86_PAE
509 printk("NX (Execute Disable) protection: active\n");
514 load_cr3(swapper_pg_dir
);
516 #ifdef CONFIG_X86_PAE
518 * We will bail out later - printk doesn't work right now so
519 * the user would just see a hanging kernel.
522 set_in_cr4(X86_CR4_PAE
);
530 * Test if the WP bit works in supervisor mode. It isn't supported on 386's
531 * and also on some strange 486's (NexGen etc.). All 586+'s are OK. This
532 * used to involve black magic jumps to work around some nasty CPU bugs,
533 * but fortunately the switch to using exceptions got rid of all that.
536 static void __init
test_wp_bit(void)
538 printk("Checking if this processor honours the WP bit even in supervisor mode... ");
540 /* Any page-aligned address will do, the test is non-destructive */
541 __set_fixmap(FIX_WP_TEST
, __pa(&swapper_pg_dir
), PAGE_READONLY
);
542 boot_cpu_data
.wp_works_ok
= do_test_wp_bit();
543 clear_fixmap(FIX_WP_TEST
);
545 if (!boot_cpu_data
.wp_works_ok
) {
547 #ifdef CONFIG_X86_WP_WORKS_OK
548 panic("This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
555 static void __init
set_max_mapnr_init(void)
557 #ifdef CONFIG_HIGHMEM
558 num_physpages
= highend_pfn
;
560 num_physpages
= max_low_pfn
;
562 #ifdef CONFIG_FLATMEM
563 max_mapnr
= num_physpages
;
567 static struct kcore_list kcore_mem
, kcore_vmalloc
;
569 void __init
mem_init(void)
571 extern int ppro_with_ram_bug(void);
572 int codesize
, reservedpages
, datasize
, initsize
;
576 #ifdef CONFIG_FLATMEM
581 bad_ppro
= ppro_with_ram_bug();
583 #ifdef CONFIG_HIGHMEM
584 /* check that fixmap and pkmap do not overlap */
585 if (PKMAP_BASE
+LAST_PKMAP
*PAGE_SIZE
>= FIXADDR_START
) {
586 printk(KERN_ERR
"fixmap and kmap areas overlap - this will crash\n");
587 printk(KERN_ERR
"pkstart: %lxh pkend: %lxh fixstart %lxh\n",
588 PKMAP_BASE
, PKMAP_BASE
+LAST_PKMAP
*PAGE_SIZE
, FIXADDR_START
);
593 set_max_mapnr_init();
595 #ifdef CONFIG_HIGHMEM
596 high_memory
= (void *) __va(highstart_pfn
* PAGE_SIZE
- 1) + 1;
598 high_memory
= (void *) __va(max_low_pfn
* PAGE_SIZE
- 1) + 1;
601 /* this will put all low memory onto the freelists */
602 totalram_pages
+= free_all_bootmem();
605 for (tmp
= 0; tmp
< max_low_pfn
; tmp
++)
607 * Only count reserved RAM pages
609 if (page_is_ram(tmp
) && PageReserved(pfn_to_page(tmp
)))
612 set_highmem_pages_init(bad_ppro
);
614 codesize
= (unsigned long) &_etext
- (unsigned long) &_text
;
615 datasize
= (unsigned long) &_edata
- (unsigned long) &_etext
;
616 initsize
= (unsigned long) &__init_end
- (unsigned long) &__init_begin
;
618 kclist_add(&kcore_mem
, __va(0), max_low_pfn
<< PAGE_SHIFT
);
619 kclist_add(&kcore_vmalloc
, (void *)VMALLOC_START
,
620 VMALLOC_END
-VMALLOC_START
);
622 printk(KERN_INFO
"Memory: %luk/%luk available (%dk kernel code, %dk reserved, %dk data, %dk init, %ldk highmem)\n",
623 (unsigned long) nr_free_pages() << (PAGE_SHIFT
-10),
624 num_physpages
<< (PAGE_SHIFT
-10),
626 reservedpages
<< (PAGE_SHIFT
-10),
629 (unsigned long) (totalhigh_pages
<< (PAGE_SHIFT
-10))
632 #ifdef CONFIG_X86_PAE
634 panic("cannot execute a PAE-enabled kernel on a PAE-less CPU!");
636 if (boot_cpu_data
.wp_works_ok
< 0)
640 * Subtle. SMP is doing it's boot stuff late (because it has to
641 * fork idle threads) - but it also needs low mappings for the
642 * protected-mode entry to work. We zap these entries only after
643 * the WP-bit has been tested.
651 * this is for the non-NUMA, single node SMP system case.
652 * Specifically, in the case of x86, we will always add
653 * memory to the highmem for now.
655 #ifdef CONFIG_MEMORY_HOTPLUG
656 #ifndef CONFIG_NEED_MULTIPLE_NODES
657 int add_memory(u64 start
, u64 size
)
659 struct pglist_data
*pgdata
= &contig_page_data
;
660 struct zone
*zone
= pgdata
->node_zones
+ MAX_NR_ZONES
-1;
661 unsigned long start_pfn
= start
>> PAGE_SHIFT
;
662 unsigned long nr_pages
= size
>> PAGE_SHIFT
;
664 return __add_pages(zone
, start_pfn
, nr_pages
);
667 int remove_memory(u64 start
, u64 size
)
674 kmem_cache_t
*pgd_cache
;
675 kmem_cache_t
*pmd_cache
;
677 void __init
pgtable_cache_init(void)
679 if (PTRS_PER_PMD
> 1) {
680 pmd_cache
= kmem_cache_create("pmd",
681 PTRS_PER_PMD
*sizeof(pmd_t
),
682 PTRS_PER_PMD
*sizeof(pmd_t
),
687 panic("pgtable_cache_init(): cannot create pmd cache");
689 pgd_cache
= kmem_cache_create("pgd",
690 PTRS_PER_PGD
*sizeof(pgd_t
),
691 PTRS_PER_PGD
*sizeof(pgd_t
),
694 PTRS_PER_PMD
== 1 ? pgd_dtor
: NULL
);
696 panic("pgtable_cache_init(): Cannot create pgd cache");
700 * This function cannot be __init, since exceptions don't work in that
701 * section. Put this after the callers, so that it cannot be inlined.
703 static int noinline
do_test_wp_bit(void)
708 __asm__
__volatile__(
713 ".section __ex_table,\"a\"\n"
717 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST
)),
726 #ifdef CONFIG_DEBUG_RODATA
728 extern char __start_rodata
, __end_rodata
;
729 void mark_rodata_ro(void)
731 unsigned long addr
= (unsigned long)&__start_rodata
;
733 for (; addr
< (unsigned long)&__end_rodata
; addr
+= PAGE_SIZE
)
734 change_page_attr(virt_to_page(addr
), 1, PAGE_KERNEL_RO
);
736 printk ("Write protecting the kernel read-only data: %luk\n",
737 (unsigned long)(&__end_rodata
- &__start_rodata
) >> 10);
740 * change_page_attr() requires a global_flush_tlb() call after it.
741 * We do this after the printk so that if something went wrong in the
742 * change, the printk gets out at least to give a better debug hint
743 * of who is the culprit.
749 void free_init_pages(char *what
, unsigned long begin
, unsigned long end
)
753 for (addr
= begin
; addr
< end
; addr
+= PAGE_SIZE
) {
754 ClearPageReserved(virt_to_page(addr
));
755 init_page_count(virt_to_page(addr
));
756 memset((void *)addr
, 0xcc, PAGE_SIZE
);
760 printk(KERN_INFO
"Freeing %s: %ldk freed\n", what
, (end
- begin
) >> 10);
763 void free_initmem(void)
765 free_init_pages("unused kernel memory",
766 (unsigned long)(&__init_begin
),
767 (unsigned long)(&__init_end
));
770 #ifdef CONFIG_BLK_DEV_INITRD
771 void free_initrd_mem(unsigned long start
, unsigned long end
)
773 free_init_pages("initrd memory", start
, end
);