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/memblock.h>
29 #include <linux/proc_fs.h>
30 #include <linux/memory_hotplug.h>
31 #include <linux/initrd.h>
32 #include <linux/cpumask.h>
33 #include <linux/gfp.h>
36 #include <asm/bios_ebda.h>
37 #include <asm/processor.h>
38 #include <asm/uaccess.h>
39 #include <asm/pgtable.h>
41 #include <asm/fixmap.h>
46 #include <asm/tlbflush.h>
47 #include <asm/olpc_ofw.h>
48 #include <asm/pgalloc.h>
49 #include <asm/sections.h>
50 #include <asm/paravirt.h>
51 #include <asm/setup.h>
52 #include <asm/cacheflush.h>
53 #include <asm/page_types.h>
56 #include "mm_internal.h"
58 unsigned long highstart_pfn
, highend_pfn
;
60 static noinline
int do_test_wp_bit(void);
62 bool __read_mostly __vmalloc_start_set
= false;
65 * Creates a middle page table and puts a pointer to it in the
66 * given global directory entry. This only returns the gd entry
67 * in non-PAE compilation mode, since the middle layer is folded.
69 static pmd_t
* __init
one_md_table_init(pgd_t
*pgd
)
75 if (!(pgd_val(*pgd
) & _PAGE_PRESENT
)) {
77 pmd_table
= (pmd_t
*)alloc_bootmem_pages(PAGE_SIZE
);
79 pmd_table
= (pmd_t
*)alloc_low_page();
80 paravirt_alloc_pmd(&init_mm
, __pa(pmd_table
) >> PAGE_SHIFT
);
81 set_pgd(pgd
, __pgd(__pa(pmd_table
) | _PAGE_PRESENT
));
82 pud
= pud_offset(pgd
, 0);
83 BUG_ON(pmd_table
!= pmd_offset(pud
, 0));
88 pud
= pud_offset(pgd
, 0);
89 pmd_table
= pmd_offset(pud
, 0);
95 * Create a page table and place a pointer to it in a middle page
98 static pte_t
* __init
one_page_table_init(pmd_t
*pmd
)
100 if (!(pmd_val(*pmd
) & _PAGE_PRESENT
)) {
101 pte_t
*page_table
= NULL
;
104 #if defined(CONFIG_DEBUG_PAGEALLOC) || defined(CONFIG_KMEMCHECK)
105 page_table
= (pte_t
*) alloc_bootmem_pages(PAGE_SIZE
);
109 (pte_t
*)alloc_bootmem_pages(PAGE_SIZE
);
111 page_table
= (pte_t
*)alloc_low_page();
113 paravirt_alloc_pte(&init_mm
, __pa(page_table
) >> PAGE_SHIFT
);
114 set_pmd(pmd
, __pmd(__pa(page_table
) | _PAGE_TABLE
));
115 BUG_ON(page_table
!= pte_offset_kernel(pmd
, 0));
118 return pte_offset_kernel(pmd
, 0);
121 pmd_t
* __init
populate_extra_pmd(unsigned long vaddr
)
123 int pgd_idx
= pgd_index(vaddr
);
124 int pmd_idx
= pmd_index(vaddr
);
126 return one_md_table_init(swapper_pg_dir
+ pgd_idx
) + pmd_idx
;
129 pte_t
* __init
populate_extra_pte(unsigned long vaddr
)
131 int pte_idx
= pte_index(vaddr
);
134 pmd
= populate_extra_pmd(vaddr
);
135 return one_page_table_init(pmd
) + pte_idx
;
138 static unsigned long __init
139 page_table_range_init_count(unsigned long start
, unsigned long end
)
141 unsigned long count
= 0;
142 #ifdef CONFIG_HIGHMEM
143 int pmd_idx_kmap_begin
= fix_to_virt(FIX_KMAP_END
) >> PMD_SHIFT
;
144 int pmd_idx_kmap_end
= fix_to_virt(FIX_KMAP_BEGIN
) >> PMD_SHIFT
;
145 int pgd_idx
, pmd_idx
;
148 if (pmd_idx_kmap_begin
== pmd_idx_kmap_end
)
152 pgd_idx
= pgd_index(vaddr
);
154 for ( ; (pgd_idx
< PTRS_PER_PGD
) && (vaddr
!= end
); pgd_idx
++) {
155 for (; (pmd_idx
< PTRS_PER_PMD
) && (vaddr
!= end
);
157 if ((vaddr
>> PMD_SHIFT
) >= pmd_idx_kmap_begin
&&
158 (vaddr
>> PMD_SHIFT
) <= pmd_idx_kmap_end
)
168 static pte_t
*__init
page_table_kmap_check(pte_t
*pte
, pmd_t
*pmd
,
169 unsigned long vaddr
, pte_t
*lastpte
,
172 #ifdef CONFIG_HIGHMEM
174 * Something (early fixmap) may already have put a pte
175 * page here, which causes the page table allocation
176 * to become nonlinear. Attempt to fix it, and if it
177 * is still nonlinear then we have to bug.
179 int pmd_idx_kmap_begin
= fix_to_virt(FIX_KMAP_END
) >> PMD_SHIFT
;
180 int pmd_idx_kmap_end
= fix_to_virt(FIX_KMAP_BEGIN
) >> PMD_SHIFT
;
182 if (pmd_idx_kmap_begin
!= pmd_idx_kmap_end
183 && (vaddr
>> PMD_SHIFT
) >= pmd_idx_kmap_begin
184 && (vaddr
>> PMD_SHIFT
) <= pmd_idx_kmap_end
) {
188 BUG_ON(after_bootmem
);
190 for (i
= 0; i
< PTRS_PER_PTE
; i
++)
191 set_pte(newpte
+ i
, pte
[i
]);
192 *adr
= (void *)(((unsigned long)(*adr
)) + PAGE_SIZE
);
194 paravirt_alloc_pte(&init_mm
, __pa(newpte
) >> PAGE_SHIFT
);
195 set_pmd(pmd
, __pmd(__pa(newpte
)|_PAGE_TABLE
));
196 BUG_ON(newpte
!= pte_offset_kernel(pmd
, 0));
199 paravirt_release_pte(__pa(pte
) >> PAGE_SHIFT
);
202 BUG_ON(vaddr
< fix_to_virt(FIX_KMAP_BEGIN
- 1)
203 && vaddr
> fix_to_virt(FIX_KMAP_END
)
204 && lastpte
&& lastpte
+ PTRS_PER_PTE
!= pte
);
210 * This function initializes a certain range of kernel virtual memory
211 * with new bootmem page tables, everywhere page tables are missing in
214 * NOTE: The pagetables are allocated contiguous on the physical space
215 * so we can cache the place of the first one and move around without
216 * checking the pgd every time.
219 page_table_range_init(unsigned long start
, unsigned long end
, pgd_t
*pgd_base
)
221 int pgd_idx
, pmd_idx
;
226 unsigned long count
= page_table_range_init_count(start
, end
);
230 adr
= alloc_low_pages(count
);
233 pgd_idx
= pgd_index(vaddr
);
234 pmd_idx
= pmd_index(vaddr
);
235 pgd
= pgd_base
+ pgd_idx
;
237 for ( ; (pgd_idx
< PTRS_PER_PGD
) && (vaddr
!= end
); pgd
++, pgd_idx
++) {
238 pmd
= one_md_table_init(pgd
);
239 pmd
= pmd
+ pmd_index(vaddr
);
240 for (; (pmd_idx
< PTRS_PER_PMD
) && (vaddr
!= end
);
242 pte
= page_table_kmap_check(one_page_table_init(pmd
),
243 pmd
, vaddr
, pte
, &adr
);
251 static inline int is_kernel_text(unsigned long addr
)
253 if (addr
>= (unsigned long)_text
&& addr
<= (unsigned long)__init_end
)
259 * This maps the physical memory to kernel virtual address space, a total
260 * of max_low_pfn pages, by creating page tables starting from address
264 kernel_physical_mapping_init(unsigned long start
,
266 unsigned long page_size_mask
)
268 int use_pse
= page_size_mask
== (1<<PG_LEVEL_2M
);
269 unsigned long last_map_addr
= end
;
270 unsigned long start_pfn
, end_pfn
;
271 pgd_t
*pgd_base
= swapper_pg_dir
;
272 int pgd_idx
, pmd_idx
, pte_ofs
;
277 unsigned pages_2m
, pages_4k
;
280 start_pfn
= start
>> PAGE_SHIFT
;
281 end_pfn
= end
>> PAGE_SHIFT
;
284 * First iteration will setup identity mapping using large/small pages
285 * based on use_pse, with other attributes same as set by
286 * the early code in head_32.S
288 * Second iteration will setup the appropriate attributes (NX, GLOBAL..)
289 * as desired for the kernel identity mapping.
291 * This two pass mechanism conforms to the TLB app note which says:
293 * "Software should not write to a paging-structure entry in a way
294 * that would change, for any linear address, both the page size
295 * and either the page frame or attributes."
303 pages_2m
= pages_4k
= 0;
305 pgd_idx
= pgd_index((pfn
<<PAGE_SHIFT
) + PAGE_OFFSET
);
306 pgd
= pgd_base
+ pgd_idx
;
307 for (; pgd_idx
< PTRS_PER_PGD
; pgd
++, pgd_idx
++) {
308 pmd
= one_md_table_init(pgd
);
312 #ifdef CONFIG_X86_PAE
313 pmd_idx
= pmd_index((pfn
<<PAGE_SHIFT
) + PAGE_OFFSET
);
318 for (; pmd_idx
< PTRS_PER_PMD
&& pfn
< end_pfn
;
320 unsigned int addr
= pfn
* PAGE_SIZE
+ PAGE_OFFSET
;
323 * Map with big pages if possible, otherwise
324 * create normal page tables:
328 pgprot_t prot
= PAGE_KERNEL_LARGE
;
330 * first pass will use the same initial
331 * identity mapping attribute + _PAGE_PSE.
334 __pgprot(PTE_IDENT_ATTR
|
337 pfn
&= PMD_MASK
>> PAGE_SHIFT
;
338 addr2
= (pfn
+ PTRS_PER_PTE
-1) * PAGE_SIZE
+
339 PAGE_OFFSET
+ PAGE_SIZE
-1;
341 if (is_kernel_text(addr
) ||
342 is_kernel_text(addr2
))
343 prot
= PAGE_KERNEL_LARGE_EXEC
;
346 if (mapping_iter
== 1)
347 set_pmd(pmd
, pfn_pmd(pfn
, init_prot
));
349 set_pmd(pmd
, pfn_pmd(pfn
, prot
));
354 pte
= one_page_table_init(pmd
);
356 pte_ofs
= pte_index((pfn
<<PAGE_SHIFT
) + PAGE_OFFSET
);
358 for (; pte_ofs
< PTRS_PER_PTE
&& pfn
< end_pfn
;
359 pte
++, pfn
++, pte_ofs
++, addr
+= PAGE_SIZE
) {
360 pgprot_t prot
= PAGE_KERNEL
;
362 * first pass will use the same initial
363 * identity mapping attribute.
365 pgprot_t init_prot
= __pgprot(PTE_IDENT_ATTR
);
367 if (is_kernel_text(addr
))
368 prot
= PAGE_KERNEL_EXEC
;
371 if (mapping_iter
== 1) {
372 set_pte(pte
, pfn_pte(pfn
, init_prot
));
373 last_map_addr
= (pfn
<< PAGE_SHIFT
) + PAGE_SIZE
;
375 set_pte(pte
, pfn_pte(pfn
, prot
));
379 if (mapping_iter
== 1) {
381 * update direct mapping page count only in the first
384 update_page_count(PG_LEVEL_2M
, pages_2m
);
385 update_page_count(PG_LEVEL_4K
, pages_4k
);
388 * local global flush tlb, which will flush the previous
389 * mappings present in both small and large page TLB's.
394 * Second iteration will set the actual desired PTE attributes.
399 return last_map_addr
;
405 static inline pte_t
*kmap_get_fixmap_pte(unsigned long vaddr
)
407 return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr
),
408 vaddr
), vaddr
), vaddr
);
411 static void __init
kmap_init(void)
413 unsigned long kmap_vstart
;
416 * Cache the first kmap pte:
418 kmap_vstart
= __fix_to_virt(FIX_KMAP_BEGIN
);
419 kmap_pte
= kmap_get_fixmap_pte(kmap_vstart
);
421 kmap_prot
= PAGE_KERNEL
;
424 #ifdef CONFIG_HIGHMEM
425 static void __init
permanent_kmaps_init(pgd_t
*pgd_base
)
434 page_table_range_init(vaddr
, vaddr
+ PAGE_SIZE
*LAST_PKMAP
, pgd_base
);
436 pgd
= swapper_pg_dir
+ pgd_index(vaddr
);
437 pud
= pud_offset(pgd
, vaddr
);
438 pmd
= pmd_offset(pud
, vaddr
);
439 pte
= pte_offset_kernel(pmd
, vaddr
);
440 pkmap_page_table
= pte
;
443 static void __init
add_one_highpage_init(struct page
*page
)
445 ClearPageReserved(page
);
446 init_page_count(page
);
451 void __init
add_highpages_with_active_regions(int nid
,
452 unsigned long start_pfn
, unsigned long end_pfn
)
454 phys_addr_t start
, end
;
457 for_each_free_mem_range(i
, nid
, &start
, &end
, NULL
) {
458 unsigned long pfn
= clamp_t(unsigned long, PFN_UP(start
),
460 unsigned long e_pfn
= clamp_t(unsigned long, PFN_DOWN(end
),
462 for ( ; pfn
< e_pfn
; pfn
++)
464 add_one_highpage_init(pfn_to_page(pfn
));
468 static inline void permanent_kmaps_init(pgd_t
*pgd_base
)
471 #endif /* CONFIG_HIGHMEM */
473 void __init
native_pagetable_init(void)
475 unsigned long pfn
, va
;
476 pgd_t
*pgd
, *base
= swapper_pg_dir
;
482 * Remove any mappings which extend past the end of physical
483 * memory from the boot time page table.
484 * In virtual address space, we should have at least two pages
485 * from VMALLOC_END to pkmap or fixmap according to VMALLOC_END
486 * definition. And max_low_pfn is set to VMALLOC_END physical
487 * address. If initial memory mapping is doing right job, we
488 * should have pte used near max_low_pfn or one pmd is not present.
490 for (pfn
= max_low_pfn
; pfn
< 1<<(32-PAGE_SHIFT
); pfn
++) {
491 va
= PAGE_OFFSET
+ (pfn
<<PAGE_SHIFT
);
492 pgd
= base
+ pgd_index(va
);
493 if (!pgd_present(*pgd
))
496 pud
= pud_offset(pgd
, va
);
497 pmd
= pmd_offset(pud
, va
);
498 if (!pmd_present(*pmd
))
501 /* should not be large page here */
502 if (pmd_large(*pmd
)) {
503 pr_warn("try to clear pte for ram above max_low_pfn: pfn: %lx pmd: %p pmd phys: %lx, but pmd is big page and is not using pte !\n",
504 pfn
, pmd
, __pa(pmd
));
508 pte
= pte_offset_kernel(pmd
, va
);
509 if (!pte_present(*pte
))
512 printk(KERN_DEBUG
"clearing pte for ram above max_low_pfn: pfn: %lx pmd: %p pmd phys: %lx pte: %p pte phys: %lx\n",
513 pfn
, pmd
, __pa(pmd
), pte
, __pa(pte
));
514 pte_clear(NULL
, va
, pte
);
516 paravirt_alloc_pmd(&init_mm
, __pa(base
) >> PAGE_SHIFT
);
521 * Build a proper pagetable for the kernel mappings. Up until this
522 * point, we've been running on some set of pagetables constructed by
525 * If we're booting on native hardware, this will be a pagetable
526 * constructed in arch/x86/kernel/head_32.S. The root of the
527 * pagetable will be swapper_pg_dir.
529 * If we're booting paravirtualized under a hypervisor, then there are
530 * more options: we may already be running PAE, and the pagetable may
531 * or may not be based in swapper_pg_dir. In any case,
532 * paravirt_pagetable_init() will set up swapper_pg_dir
533 * appropriately for the rest of the initialization to work.
535 * In general, pagetable_init() assumes that the pagetable may already
536 * be partially populated, and so it avoids stomping on any existing
539 void __init
early_ioremap_page_table_range_init(void)
541 pgd_t
*pgd_base
= swapper_pg_dir
;
542 unsigned long vaddr
, end
;
545 * Fixed mappings, only the page table structure has to be
546 * created - mappings will be set by set_fixmap():
548 vaddr
= __fix_to_virt(__end_of_fixed_addresses
- 1) & PMD_MASK
;
549 end
= (FIXADDR_TOP
+ PMD_SIZE
- 1) & PMD_MASK
;
550 page_table_range_init(vaddr
, end
, pgd_base
);
551 early_ioremap_reset();
554 static void __init
pagetable_init(void)
556 pgd_t
*pgd_base
= swapper_pg_dir
;
558 permanent_kmaps_init(pgd_base
);
561 pteval_t __supported_pte_mask __read_mostly
= ~(_PAGE_NX
| _PAGE_GLOBAL
| _PAGE_IOMAP
);
562 EXPORT_SYMBOL_GPL(__supported_pte_mask
);
564 /* user-defined highmem size */
565 static unsigned int highmem_pages
= -1;
568 * highmem=size forces highmem to be exactly 'size' bytes.
569 * This works even on boxes that have no highmem otherwise.
570 * This also works to reduce highmem size on bigger boxes.
572 static int __init
parse_highmem(char *arg
)
577 highmem_pages
= memparse(arg
, &arg
) >> PAGE_SHIFT
;
580 early_param("highmem", parse_highmem
);
582 #define MSG_HIGHMEM_TOO_BIG \
583 "highmem size (%luMB) is bigger than pages available (%luMB)!\n"
585 #define MSG_LOWMEM_TOO_SMALL \
586 "highmem size (%luMB) results in <64MB lowmem, ignoring it!\n"
588 * All of RAM fits into lowmem - but if user wants highmem
589 * artificially via the highmem=x boot parameter then create
592 static void __init
lowmem_pfn_init(void)
594 /* max_low_pfn is 0, we already have early_res support */
595 max_low_pfn
= max_pfn
;
597 if (highmem_pages
== -1)
599 #ifdef CONFIG_HIGHMEM
600 if (highmem_pages
>= max_pfn
) {
601 printk(KERN_ERR MSG_HIGHMEM_TOO_BIG
,
602 pages_to_mb(highmem_pages
), pages_to_mb(max_pfn
));
606 if (max_low_pfn
- highmem_pages
< 64*1024*1024/PAGE_SIZE
) {
607 printk(KERN_ERR MSG_LOWMEM_TOO_SMALL
,
608 pages_to_mb(highmem_pages
));
611 max_low_pfn
-= highmem_pages
;
615 printk(KERN_ERR
"ignoring highmem size on non-highmem kernel!\n");
619 #define MSG_HIGHMEM_TOO_SMALL \
620 "only %luMB highmem pages available, ignoring highmem size of %luMB!\n"
622 #define MSG_HIGHMEM_TRIMMED \
623 "Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n"
625 * We have more RAM than fits into lowmem - we try to put it into
626 * highmem, also taking the highmem=x boot parameter into account:
628 static void __init
highmem_pfn_init(void)
630 max_low_pfn
= MAXMEM_PFN
;
632 if (highmem_pages
== -1)
633 highmem_pages
= max_pfn
- MAXMEM_PFN
;
635 if (highmem_pages
+ MAXMEM_PFN
< max_pfn
)
636 max_pfn
= MAXMEM_PFN
+ highmem_pages
;
638 if (highmem_pages
+ MAXMEM_PFN
> max_pfn
) {
639 printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL
,
640 pages_to_mb(max_pfn
- MAXMEM_PFN
),
641 pages_to_mb(highmem_pages
));
644 #ifndef CONFIG_HIGHMEM
645 /* Maximum memory usable is what is directly addressable */
646 printk(KERN_WARNING
"Warning only %ldMB will be used.\n", MAXMEM
>>20);
647 if (max_pfn
> MAX_NONPAE_PFN
)
648 printk(KERN_WARNING
"Use a HIGHMEM64G enabled kernel.\n");
650 printk(KERN_WARNING
"Use a HIGHMEM enabled kernel.\n");
651 max_pfn
= MAXMEM_PFN
;
652 #else /* !CONFIG_HIGHMEM */
653 #ifndef CONFIG_HIGHMEM64G
654 if (max_pfn
> MAX_NONPAE_PFN
) {
655 max_pfn
= MAX_NONPAE_PFN
;
656 printk(KERN_WARNING MSG_HIGHMEM_TRIMMED
);
658 #endif /* !CONFIG_HIGHMEM64G */
659 #endif /* !CONFIG_HIGHMEM */
663 * Determine low and high memory ranges:
665 void __init
find_low_pfn_range(void)
667 /* it could update max_pfn */
669 if (max_pfn
<= MAXMEM_PFN
)
675 #ifndef CONFIG_NEED_MULTIPLE_NODES
676 void __init
initmem_init(void)
678 #ifdef CONFIG_HIGHMEM
679 highstart_pfn
= highend_pfn
= max_pfn
;
680 if (max_pfn
> max_low_pfn
)
681 highstart_pfn
= max_low_pfn
;
682 printk(KERN_NOTICE
"%ldMB HIGHMEM available.\n",
683 pages_to_mb(highend_pfn
- highstart_pfn
));
684 num_physpages
= highend_pfn
;
685 high_memory
= (void *) __va(highstart_pfn
* PAGE_SIZE
- 1) + 1;
687 num_physpages
= max_low_pfn
;
688 high_memory
= (void *) __va(max_low_pfn
* PAGE_SIZE
- 1) + 1;
691 memblock_set_node(0, (phys_addr_t
)ULLONG_MAX
, 0);
692 sparse_memory_present_with_active_regions(0);
694 #ifdef CONFIG_FLATMEM
695 max_mapnr
= num_physpages
;
697 __vmalloc_start_set
= true;
699 printk(KERN_NOTICE
"%ldMB LOWMEM available.\n",
700 pages_to_mb(max_low_pfn
));
702 setup_bootmem_allocator();
704 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
706 void __init
setup_bootmem_allocator(void)
708 printk(KERN_INFO
" mapped low ram: 0 - %08lx\n",
709 max_pfn_mapped
<<PAGE_SHIFT
);
710 printk(KERN_INFO
" low ram: 0 - %08lx\n", max_low_pfn
<<PAGE_SHIFT
);
716 * paging_init() sets up the page tables - note that the first 8MB are
717 * already mapped by head.S.
719 * This routines also unmaps the page at virtual kernel address 0, so
720 * that we can trap those pesky NULL-reference errors in the kernel.
722 void __init
paging_init(void)
731 * NOTE: at this point the bootmem allocator is fully available.
733 olpc_dt_build_devicetree();
734 sparse_memory_present_with_active_regions(MAX_NUMNODES
);
740 * Test if the WP bit works in supervisor mode. It isn't supported on 386's
741 * and also on some strange 486's. All 586+'s are OK. This used to involve
742 * black magic jumps to work around some nasty CPU bugs, but fortunately the
743 * switch to using exceptions got rid of all that.
745 static void __init
test_wp_bit(void)
748 "Checking if this processor honours the WP bit even in supervisor mode...");
750 /* Any page-aligned address will do, the test is non-destructive */
751 __set_fixmap(FIX_WP_TEST
, __pa(&swapper_pg_dir
), PAGE_KERNEL_RO
);
752 boot_cpu_data
.wp_works_ok
= do_test_wp_bit();
753 clear_fixmap(FIX_WP_TEST
);
755 if (!boot_cpu_data
.wp_works_ok
) {
756 printk(KERN_CONT
"No.\n");
757 #ifdef CONFIG_X86_WP_WORKS_OK
759 "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
762 printk(KERN_CONT
"Ok.\n");
766 void __init
mem_init(void)
768 int codesize
, reservedpages
, datasize
, initsize
;
773 #ifdef CONFIG_FLATMEM
777 * With CONFIG_DEBUG_PAGEALLOC initialization of highmem pages has to
778 * be done before free_all_bootmem(). Memblock use free low memory for
779 * temporary data (see find_range_array()) and for this purpose can use
780 * pages that was already passed to the buddy allocator, hence marked as
781 * not accessible in the page tables when compiled with
782 * CONFIG_DEBUG_PAGEALLOC. Otherwise order of initialization is not
785 set_highmem_pages_init();
787 /* this will put all low memory onto the freelists */
788 totalram_pages
+= free_all_bootmem();
791 for (tmp
= 0; tmp
< max_low_pfn
; tmp
++)
793 * Only count reserved RAM pages:
795 if (page_is_ram(tmp
) && PageReserved(pfn_to_page(tmp
)))
798 codesize
= (unsigned long) &_etext
- (unsigned long) &_text
;
799 datasize
= (unsigned long) &_edata
- (unsigned long) &_etext
;
800 initsize
= (unsigned long) &__init_end
- (unsigned long) &__init_begin
;
802 printk(KERN_INFO
"Memory: %luk/%luk available (%dk kernel code, "
803 "%dk reserved, %dk data, %dk init, %ldk highmem)\n",
804 nr_free_pages() << (PAGE_SHIFT
-10),
805 num_physpages
<< (PAGE_SHIFT
-10),
807 reservedpages
<< (PAGE_SHIFT
-10),
810 totalhigh_pages
<< (PAGE_SHIFT
-10));
812 printk(KERN_INFO
"virtual kernel memory layout:\n"
813 " fixmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
814 #ifdef CONFIG_HIGHMEM
815 " pkmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
817 " vmalloc : 0x%08lx - 0x%08lx (%4ld MB)\n"
818 " lowmem : 0x%08lx - 0x%08lx (%4ld MB)\n"
819 " .init : 0x%08lx - 0x%08lx (%4ld kB)\n"
820 " .data : 0x%08lx - 0x%08lx (%4ld kB)\n"
821 " .text : 0x%08lx - 0x%08lx (%4ld kB)\n",
822 FIXADDR_START
, FIXADDR_TOP
,
823 (FIXADDR_TOP
- FIXADDR_START
) >> 10,
825 #ifdef CONFIG_HIGHMEM
826 PKMAP_BASE
, PKMAP_BASE
+LAST_PKMAP
*PAGE_SIZE
,
827 (LAST_PKMAP
*PAGE_SIZE
) >> 10,
830 VMALLOC_START
, VMALLOC_END
,
831 (VMALLOC_END
- VMALLOC_START
) >> 20,
833 (unsigned long)__va(0), (unsigned long)high_memory
,
834 ((unsigned long)high_memory
- (unsigned long)__va(0)) >> 20,
836 (unsigned long)&__init_begin
, (unsigned long)&__init_end
,
837 ((unsigned long)&__init_end
-
838 (unsigned long)&__init_begin
) >> 10,
840 (unsigned long)&_etext
, (unsigned long)&_edata
,
841 ((unsigned long)&_edata
- (unsigned long)&_etext
) >> 10,
843 (unsigned long)&_text
, (unsigned long)&_etext
,
844 ((unsigned long)&_etext
- (unsigned long)&_text
) >> 10);
847 * Check boundaries twice: Some fundamental inconsistencies can
848 * be detected at build time already.
850 #define __FIXADDR_TOP (-PAGE_SIZE)
851 #ifdef CONFIG_HIGHMEM
852 BUILD_BUG_ON(PKMAP_BASE
+ LAST_PKMAP
*PAGE_SIZE
> FIXADDR_START
);
853 BUILD_BUG_ON(VMALLOC_END
> PKMAP_BASE
);
855 #define high_memory (-128UL << 20)
856 BUILD_BUG_ON(VMALLOC_START
>= VMALLOC_END
);
860 #ifdef CONFIG_HIGHMEM
861 BUG_ON(PKMAP_BASE
+ LAST_PKMAP
*PAGE_SIZE
> FIXADDR_START
);
862 BUG_ON(VMALLOC_END
> PKMAP_BASE
);
864 BUG_ON(VMALLOC_START
>= VMALLOC_END
);
865 BUG_ON((unsigned long)high_memory
> VMALLOC_START
);
867 if (boot_cpu_data
.wp_works_ok
< 0)
871 #ifdef CONFIG_MEMORY_HOTPLUG
872 int arch_add_memory(int nid
, u64 start
, u64 size
)
874 struct pglist_data
*pgdata
= NODE_DATA(nid
);
875 struct zone
*zone
= pgdata
->node_zones
+ ZONE_HIGHMEM
;
876 unsigned long start_pfn
= start
>> PAGE_SHIFT
;
877 unsigned long nr_pages
= size
>> PAGE_SHIFT
;
879 return __add_pages(nid
, zone
, start_pfn
, nr_pages
);
884 * This function cannot be __init, since exceptions don't work in that
885 * section. Put this after the callers, so that it cannot be inlined.
887 static noinline
int do_test_wp_bit(void)
892 __asm__
__volatile__(
898 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST
)),
907 #ifdef CONFIG_DEBUG_RODATA
908 const int rodata_test_data
= 0xC3;
909 EXPORT_SYMBOL_GPL(rodata_test_data
);
911 int kernel_set_to_readonly __read_mostly
;
913 void set_kernel_text_rw(void)
915 unsigned long start
= PFN_ALIGN(_text
);
916 unsigned long size
= PFN_ALIGN(_etext
) - start
;
918 if (!kernel_set_to_readonly
)
921 pr_debug("Set kernel text: %lx - %lx for read write\n",
924 set_pages_rw(virt_to_page(start
), size
>> PAGE_SHIFT
);
927 void set_kernel_text_ro(void)
929 unsigned long start
= PFN_ALIGN(_text
);
930 unsigned long size
= PFN_ALIGN(_etext
) - start
;
932 if (!kernel_set_to_readonly
)
935 pr_debug("Set kernel text: %lx - %lx for read only\n",
938 set_pages_ro(virt_to_page(start
), size
>> PAGE_SHIFT
);
941 static void mark_nxdata_nx(void)
944 * When this called, init has already been executed and released,
945 * so everything past _etext should be NX.
947 unsigned long start
= PFN_ALIGN(_etext
);
949 * This comes from is_kernel_text upper limit. Also HPAGE where used:
951 unsigned long size
= (((unsigned long)__init_end
+ HPAGE_SIZE
) & HPAGE_MASK
) - start
;
953 if (__supported_pte_mask
& _PAGE_NX
)
954 printk(KERN_INFO
"NX-protecting the kernel data: %luk\n", size
>> 10);
955 set_pages_nx(virt_to_page(start
), size
>> PAGE_SHIFT
);
958 void mark_rodata_ro(void)
960 unsigned long start
= PFN_ALIGN(_text
);
961 unsigned long size
= PFN_ALIGN(_etext
) - start
;
963 set_pages_ro(virt_to_page(start
), size
>> PAGE_SHIFT
);
964 printk(KERN_INFO
"Write protecting the kernel text: %luk\n",
967 kernel_set_to_readonly
= 1;
969 #ifdef CONFIG_CPA_DEBUG
970 printk(KERN_INFO
"Testing CPA: Reverting %lx-%lx\n",
972 set_pages_rw(virt_to_page(start
), size
>>PAGE_SHIFT
);
974 printk(KERN_INFO
"Testing CPA: write protecting again\n");
975 set_pages_ro(virt_to_page(start
), size
>>PAGE_SHIFT
);
979 size
= (unsigned long)__end_rodata
- start
;
980 set_pages_ro(virt_to_page(start
), size
>> PAGE_SHIFT
);
981 printk(KERN_INFO
"Write protecting the kernel read-only data: %luk\n",
985 #ifdef CONFIG_CPA_DEBUG
986 printk(KERN_INFO
"Testing CPA: undo %lx-%lx\n", start
, start
+ size
);
987 set_pages_rw(virt_to_page(start
), size
>> PAGE_SHIFT
);
989 printk(KERN_INFO
"Testing CPA: write protecting again\n");
990 set_pages_ro(virt_to_page(start
), size
>> PAGE_SHIFT
);