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>
54 unsigned long max_low_pfn_mapped;
55 unsigned long max_pfn_mapped;
57 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
58 unsigned long highstart_pfn, highend_pfn;
60 static noinline int do_test_wp_bit(void);
62 bool __read_mostly __vmalloc_start_set = false;
64 static __init void *alloc_low_page(void)
66 unsigned long pfn = e820_table_end++;
69 if (pfn >= e820_table_top)
70 panic("alloc_low_page: ran out of memory");
72 adr = __va(pfn * PAGE_SIZE);
73 memset(adr, 0, PAGE_SIZE);
78 * Creates a middle page table and puts a pointer to it in the
79 * given global directory entry. This only returns the gd entry
80 * in non-PAE compilation mode, since the middle layer is folded.
82 static pmd_t * __init one_md_table_init(pgd_t *pgd)
88 if (!(pgd_val(*pgd) & _PAGE_PRESENT)) {
90 pmd_table = (pmd_t *)alloc_bootmem_low_pages(PAGE_SIZE);
92 pmd_table = (pmd_t *)alloc_low_page();
93 paravirt_alloc_pmd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT);
94 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
95 pud = pud_offset(pgd, 0);
96 BUG_ON(pmd_table != pmd_offset(pud, 0));
101 pud = pud_offset(pgd, 0);
102 pmd_table = pmd_offset(pud, 0);
108 * Create a page table and place a pointer to it in a middle page
111 static pte_t * __init one_page_table_init(pmd_t *pmd)
113 if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
114 pte_t *page_table = NULL;
117 #ifdef CONFIG_DEBUG_PAGEALLOC
118 page_table = (pte_t *) alloc_bootmem_pages(PAGE_SIZE);
122 (pte_t *)alloc_bootmem_low_pages(PAGE_SIZE);
124 page_table = (pte_t *)alloc_low_page();
126 paravirt_alloc_pte(&init_mm, __pa(page_table) >> PAGE_SHIFT);
127 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
128 BUG_ON(page_table != pte_offset_kernel(pmd, 0));
131 return pte_offset_kernel(pmd, 0);
134 pmd_t * __init populate_extra_pmd(unsigned long vaddr)
136 int pgd_idx = pgd_index(vaddr);
137 int pmd_idx = pmd_index(vaddr);
139 return one_md_table_init(swapper_pg_dir + pgd_idx) + pmd_idx;
142 pte_t * __init populate_extra_pte(unsigned long vaddr)
144 int pte_idx = pte_index(vaddr);
147 pmd = populate_extra_pmd(vaddr);
148 return one_page_table_init(pmd) + pte_idx;
151 static pte_t *__init page_table_kmap_check(pte_t *pte, pmd_t *pmd,
152 unsigned long vaddr, pte_t *lastpte)
154 #ifdef CONFIG_HIGHMEM
156 * Something (early fixmap) may already have put a pte
157 * page here, which causes the page table allocation
158 * to become nonlinear. Attempt to fix it, and if it
159 * is still nonlinear then we have to bug.
161 int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
162 int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
164 if (pmd_idx_kmap_begin != pmd_idx_kmap_end
165 && (vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin
166 && (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end
167 && ((__pa(pte) >> PAGE_SHIFT) < e820_table_start
168 || (__pa(pte) >> PAGE_SHIFT) >= e820_table_end)) {
172 BUG_ON(after_bootmem);
173 newpte = alloc_low_page();
174 for (i = 0; i < PTRS_PER_PTE; i++)
175 set_pte(newpte + i, pte[i]);
177 paravirt_alloc_pte(&init_mm, __pa(newpte) >> PAGE_SHIFT);
178 set_pmd(pmd, __pmd(__pa(newpte)|_PAGE_TABLE));
179 BUG_ON(newpte != pte_offset_kernel(pmd, 0));
182 paravirt_release_pte(__pa(pte) >> PAGE_SHIFT);
185 BUG_ON(vaddr < fix_to_virt(FIX_KMAP_BEGIN - 1)
186 && vaddr > fix_to_virt(FIX_KMAP_END)
187 && lastpte && lastpte + PTRS_PER_PTE != pte);
193 * This function initializes a certain range of kernel virtual memory
194 * with new bootmem page tables, everywhere page tables are missing in
197 * NOTE: The pagetables are allocated contiguous on the physical space
198 * so we can cache the place of the first one and move around without
199 * checking the pgd every time.
202 page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base)
204 int pgd_idx, pmd_idx;
211 pgd_idx = pgd_index(vaddr);
212 pmd_idx = pmd_index(vaddr);
213 pgd = pgd_base + pgd_idx;
215 for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
216 pmd = one_md_table_init(pgd);
217 pmd = pmd + pmd_index(vaddr);
218 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
220 pte = page_table_kmap_check(one_page_table_init(pmd),
229 static inline int is_kernel_text(unsigned long addr)
231 if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end)
237 * This maps the physical memory to kernel virtual address space, a total
238 * of max_low_pfn pages, by creating page tables starting from address
242 kernel_physical_mapping_init(unsigned long start,
244 unsigned long page_size_mask)
246 int use_pse = page_size_mask == (1<<PG_LEVEL_2M);
247 unsigned long start_pfn, end_pfn;
248 pgd_t *pgd_base = swapper_pg_dir;
249 int pgd_idx, pmd_idx, pte_ofs;
254 unsigned pages_2m, pages_4k;
257 start_pfn = start >> PAGE_SHIFT;
258 end_pfn = end >> PAGE_SHIFT;
261 * First iteration will setup identity mapping using large/small pages
262 * based on use_pse, with other attributes same as set by
263 * the early code in head_32.S
265 * Second iteration will setup the appropriate attributes (NX, GLOBAL..)
266 * as desired for the kernel identity mapping.
268 * This two pass mechanism conforms to the TLB app note which says:
270 * "Software should not write to a paging-structure entry in a way
271 * that would change, for any linear address, both the page size
272 * and either the page frame or attributes."
280 pages_2m = pages_4k = 0;
282 pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
283 pgd = pgd_base + pgd_idx;
284 for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
285 pmd = one_md_table_init(pgd);
289 #ifdef CONFIG_X86_PAE
290 pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
295 for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn;
297 unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET;
300 * Map with big pages if possible, otherwise
301 * create normal page tables:
305 pgprot_t prot = PAGE_KERNEL_LARGE;
307 * first pass will use the same initial
308 * identity mapping attribute + _PAGE_PSE.
311 __pgprot(PTE_IDENT_ATTR |
314 addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
315 PAGE_OFFSET + PAGE_SIZE-1;
317 if (is_kernel_text(addr) ||
318 is_kernel_text(addr2))
319 prot = PAGE_KERNEL_LARGE_EXEC;
322 if (mapping_iter == 1)
323 set_pmd(pmd, pfn_pmd(pfn, init_prot));
325 set_pmd(pmd, pfn_pmd(pfn, prot));
330 pte = one_page_table_init(pmd);
332 pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
334 for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn;
335 pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
336 pgprot_t prot = PAGE_KERNEL;
338 * first pass will use the same initial
339 * identity mapping attribute.
341 pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR);
343 if (is_kernel_text(addr))
344 prot = PAGE_KERNEL_EXEC;
347 if (mapping_iter == 1)
348 set_pte(pte, pfn_pte(pfn, init_prot));
350 set_pte(pte, pfn_pte(pfn, prot));
354 if (mapping_iter == 1) {
356 * update direct mapping page count only in the first
359 update_page_count(PG_LEVEL_2M, pages_2m);
360 update_page_count(PG_LEVEL_4K, pages_4k);
363 * local global flush tlb, which will flush the previous
364 * mappings present in both small and large page TLB's.
369 * Second iteration will set the actual desired PTE attributes.
380 static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr)
382 return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
383 vaddr), vaddr), vaddr);
386 static void __init kmap_init(void)
388 unsigned long kmap_vstart;
391 * Cache the first kmap pte:
393 kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
394 kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
396 kmap_prot = PAGE_KERNEL;
399 #ifdef CONFIG_HIGHMEM
400 static void __init permanent_kmaps_init(pgd_t *pgd_base)
409 page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
411 pgd = swapper_pg_dir + pgd_index(vaddr);
412 pud = pud_offset(pgd, vaddr);
413 pmd = pmd_offset(pud, vaddr);
414 pte = pte_offset_kernel(pmd, vaddr);
415 pkmap_page_table = pte;
418 static void __init add_one_highpage_init(struct page *page, int pfn)
420 ClearPageReserved(page);
421 init_page_count(page);
426 struct add_highpages_data {
427 unsigned long start_pfn;
428 unsigned long end_pfn;
431 static int __init add_highpages_work_fn(unsigned long start_pfn,
432 unsigned long end_pfn, void *datax)
436 unsigned long final_start_pfn, final_end_pfn;
437 struct add_highpages_data *data;
439 data = (struct add_highpages_data *)datax;
441 final_start_pfn = max(start_pfn, data->start_pfn);
442 final_end_pfn = min(end_pfn, data->end_pfn);
443 if (final_start_pfn >= final_end_pfn)
446 for (node_pfn = final_start_pfn; node_pfn < final_end_pfn;
448 if (!pfn_valid(node_pfn))
450 page = pfn_to_page(node_pfn);
451 add_one_highpage_init(page, node_pfn);
458 void __init add_highpages_with_active_regions(int nid, unsigned long start_pfn,
459 unsigned long end_pfn)
461 struct add_highpages_data data;
463 data.start_pfn = start_pfn;
464 data.end_pfn = end_pfn;
466 work_with_active_regions(nid, add_highpages_work_fn, &data);
470 static inline void permanent_kmaps_init(pgd_t *pgd_base)
473 #endif /* CONFIG_HIGHMEM */
475 void __init native_pagetable_setup_start(pgd_t *base)
477 unsigned long pfn, va;
484 * Remove any mappings which extend past the end of physical
485 * memory from the boot time page table:
487 for (pfn = max_low_pfn + 1; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
488 va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
489 pgd = base + pgd_index(va);
490 if (!pgd_present(*pgd))
493 pud = pud_offset(pgd, va);
494 pmd = pmd_offset(pud, va);
495 if (!pmd_present(*pmd))
498 pte = pte_offset_kernel(pmd, va);
499 if (!pte_present(*pte))
502 pte_clear(NULL, va, pte);
504 paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT);
507 void __init native_pagetable_setup_done(pgd_t *base)
512 * Build a proper pagetable for the kernel mappings. Up until this
513 * point, we've been running on some set of pagetables constructed by
516 * If we're booting on native hardware, this will be a pagetable
517 * constructed in arch/x86/kernel/head_32.S. The root of the
518 * pagetable will be swapper_pg_dir.
520 * If we're booting paravirtualized under a hypervisor, then there are
521 * more options: we may already be running PAE, and the pagetable may
522 * or may not be based in swapper_pg_dir. In any case,
523 * paravirt_pagetable_setup_start() will set up swapper_pg_dir
524 * appropriately for the rest of the initialization to work.
526 * In general, pagetable_init() assumes that the pagetable may already
527 * be partially populated, and so it avoids stomping on any existing
530 void __init early_ioremap_page_table_range_init(void)
532 pgd_t *pgd_base = swapper_pg_dir;
533 unsigned long vaddr, end;
536 * Fixed mappings, only the page table structure has to be
537 * created - mappings will be set by set_fixmap():
539 vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
540 end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
541 page_table_range_init(vaddr, end, pgd_base);
542 early_ioremap_reset();
545 static void __init pagetable_init(void)
547 pgd_t *pgd_base = swapper_pg_dir;
549 permanent_kmaps_init(pgd_base);
552 #ifdef CONFIG_ACPI_SLEEP
554 * ACPI suspend needs this for resume, because things like the intel-agp
555 * driver might have split up a kernel 4MB mapping.
557 char swsusp_pg_dir[PAGE_SIZE]
558 __attribute__ ((aligned(PAGE_SIZE)));
560 static inline void save_pg_dir(void)
562 memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
564 #else /* !CONFIG_ACPI_SLEEP */
565 static inline void save_pg_dir(void)
568 #endif /* !CONFIG_ACPI_SLEEP */
570 void zap_low_mappings(void)
575 * Zap initial low-memory mappings.
577 * Note that "pgd_clear()" doesn't do it for
578 * us, because pgd_clear() is a no-op on i386.
580 for (i = 0; i < KERNEL_PGD_BOUNDARY; i++) {
581 #ifdef CONFIG_X86_PAE
582 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
584 set_pgd(swapper_pg_dir+i, __pgd(0));
590 pteval_t __supported_pte_mask __read_mostly = ~(_PAGE_NX | _PAGE_GLOBAL | _PAGE_IOMAP);
591 EXPORT_SYMBOL_GPL(__supported_pte_mask);
593 /* user-defined highmem size */
594 static unsigned int highmem_pages = -1;
597 * highmem=size forces highmem to be exactly 'size' bytes.
598 * This works even on boxes that have no highmem otherwise.
599 * This also works to reduce highmem size on bigger boxes.
601 static int __init parse_highmem(char *arg)
606 highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
609 early_param("highmem", parse_highmem);
611 #define MSG_HIGHMEM_TOO_BIG \
612 "highmem size (%luMB) is bigger than pages available (%luMB)!\n"
614 #define MSG_LOWMEM_TOO_SMALL \
615 "highmem size (%luMB) results in <64MB lowmem, ignoring it!\n"
617 * All of RAM fits into lowmem - but if user wants highmem
618 * artificially via the highmem=x boot parameter then create
621 void __init lowmem_pfn_init(void)
623 /* max_low_pfn is 0, we already have early_res support */
624 max_low_pfn = max_pfn;
626 if (highmem_pages == -1)
628 #ifdef CONFIG_HIGHMEM
629 if (highmem_pages >= max_pfn) {
630 printk(KERN_ERR MSG_HIGHMEM_TOO_BIG,
631 pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
635 if (max_low_pfn - highmem_pages < 64*1024*1024/PAGE_SIZE) {
636 printk(KERN_ERR MSG_LOWMEM_TOO_SMALL,
637 pages_to_mb(highmem_pages));
640 max_low_pfn -= highmem_pages;
644 printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
648 #define MSG_HIGHMEM_TOO_SMALL \
649 "only %luMB highmem pages available, ignoring highmem size of %luMB!\n"
651 #define MSG_HIGHMEM_TRIMMED \
652 "Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n"
654 * We have more RAM than fits into lowmem - we try to put it into
655 * highmem, also taking the highmem=x boot parameter into account:
657 void __init highmem_pfn_init(void)
659 max_low_pfn = MAXMEM_PFN;
661 if (highmem_pages == -1)
662 highmem_pages = max_pfn - MAXMEM_PFN;
664 if (highmem_pages + MAXMEM_PFN < max_pfn)
665 max_pfn = MAXMEM_PFN + highmem_pages;
667 if (highmem_pages + MAXMEM_PFN > max_pfn) {
668 printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL,
669 pages_to_mb(max_pfn - MAXMEM_PFN),
670 pages_to_mb(highmem_pages));
673 #ifndef CONFIG_HIGHMEM
674 /* Maximum memory usable is what is directly addressable */
675 printk(KERN_WARNING "Warning only %ldMB will be used.\n", MAXMEM>>20);
676 if (max_pfn > MAX_NONPAE_PFN)
677 printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
679 printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
680 max_pfn = MAXMEM_PFN;
681 #else /* !CONFIG_HIGHMEM */
682 #ifndef CONFIG_HIGHMEM64G
683 if (max_pfn > MAX_NONPAE_PFN) {
684 max_pfn = MAX_NONPAE_PFN;
685 printk(KERN_WARNING MSG_HIGHMEM_TRIMMED);
687 #endif /* !CONFIG_HIGHMEM64G */
688 #endif /* !CONFIG_HIGHMEM */
692 * Determine low and high memory ranges:
694 void __init find_low_pfn_range(void)
696 /* it could update max_pfn */
698 if (max_pfn <= MAXMEM_PFN)
704 #ifndef CONFIG_NEED_MULTIPLE_NODES
705 void __init initmem_init(unsigned long start_pfn,
706 unsigned long end_pfn)
708 #ifdef CONFIG_HIGHMEM
709 highstart_pfn = highend_pfn = max_pfn;
710 if (max_pfn > max_low_pfn)
711 highstart_pfn = max_low_pfn;
712 memory_present(0, 0, highend_pfn);
713 e820_register_active_regions(0, 0, highend_pfn);
714 printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
715 pages_to_mb(highend_pfn - highstart_pfn));
716 num_physpages = highend_pfn;
717 high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
719 memory_present(0, 0, max_low_pfn);
720 e820_register_active_regions(0, 0, max_low_pfn);
721 num_physpages = max_low_pfn;
722 high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
724 #ifdef CONFIG_FLATMEM
725 max_mapnr = num_physpages;
727 __vmalloc_start_set = true;
729 printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
730 pages_to_mb(max_low_pfn));
732 setup_bootmem_allocator();
734 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
736 static void __init zone_sizes_init(void)
738 unsigned long max_zone_pfns[MAX_NR_ZONES];
739 memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
740 max_zone_pfns[ZONE_DMA] =
741 virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
742 max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
743 #ifdef CONFIG_HIGHMEM
744 max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
747 free_area_init_nodes(max_zone_pfns);
750 static unsigned long __init setup_node_bootmem(int nodeid,
751 unsigned long start_pfn,
752 unsigned long end_pfn,
753 unsigned long bootmap)
755 unsigned long bootmap_size;
757 /* don't touch min_low_pfn */
758 bootmap_size = init_bootmem_node(NODE_DATA(nodeid),
759 bootmap >> PAGE_SHIFT,
761 printk(KERN_INFO " node %d low ram: %08lx - %08lx\n",
762 nodeid, start_pfn<<PAGE_SHIFT, end_pfn<<PAGE_SHIFT);
763 printk(KERN_INFO " node %d bootmap %08lx - %08lx\n",
764 nodeid, bootmap, bootmap + bootmap_size);
765 free_bootmem_with_active_regions(nodeid, end_pfn);
766 early_res_to_bootmem(start_pfn<<PAGE_SHIFT, end_pfn<<PAGE_SHIFT);
768 return bootmap + bootmap_size;
771 void __init setup_bootmem_allocator(void)
774 unsigned long bootmap_size, bootmap;
776 * Initialize the boot-time allocator (with low memory only):
778 bootmap_size = bootmem_bootmap_pages(max_low_pfn)<<PAGE_SHIFT;
779 bootmap = find_e820_area(0, max_pfn_mapped<<PAGE_SHIFT, bootmap_size,
782 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
783 reserve_early(bootmap, bootmap + bootmap_size, "BOOTMAP");
785 printk(KERN_INFO " mapped low ram: 0 - %08lx\n",
786 max_pfn_mapped<<PAGE_SHIFT);
787 printk(KERN_INFO " low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT);
789 for_each_online_node(nodeid) {
790 unsigned long start_pfn, end_pfn;
792 #ifdef CONFIG_NEED_MULTIPLE_NODES
793 start_pfn = node_start_pfn[nodeid];
794 end_pfn = node_end_pfn[nodeid];
795 if (start_pfn > max_low_pfn)
797 if (end_pfn > max_low_pfn)
798 end_pfn = max_low_pfn;
801 end_pfn = max_low_pfn;
803 bootmap = setup_node_bootmem(nodeid, start_pfn, end_pfn,
811 * paging_init() sets up the page tables - note that the first 8MB are
812 * already mapped by head.S.
814 * This routines also unmaps the page at virtual kernel address 0, so
815 * that we can trap those pesky NULL-reference errors in the kernel.
817 void __init paging_init(void)
826 * NOTE: at this point the bootmem allocator is fully available.
833 * Test if the WP bit works in supervisor mode. It isn't supported on 386's
834 * and also on some strange 486's. All 586+'s are OK. This used to involve
835 * black magic jumps to work around some nasty CPU bugs, but fortunately the
836 * switch to using exceptions got rid of all that.
838 static void __init test_wp_bit(void)
841 "Checking if this processor honours the WP bit even in supervisor mode...");
843 /* Any page-aligned address will do, the test is non-destructive */
844 __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
845 boot_cpu_data.wp_works_ok = do_test_wp_bit();
846 clear_fixmap(FIX_WP_TEST);
848 if (!boot_cpu_data.wp_works_ok) {
849 printk(KERN_CONT "No.\n");
850 #ifdef CONFIG_X86_WP_WORKS_OK
852 "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
855 printk(KERN_CONT "Ok.\n");
859 static struct kcore_list kcore_mem, kcore_vmalloc;
861 void __init mem_init(void)
863 int codesize, reservedpages, datasize, initsize;
868 #ifdef CONFIG_FLATMEM
871 /* this will put all low memory onto the freelists */
872 totalram_pages += free_all_bootmem();
875 for (tmp = 0; tmp < max_low_pfn; tmp++)
877 * Only count reserved RAM pages:
879 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
882 set_highmem_pages_init();
884 codesize = (unsigned long) &_etext - (unsigned long) &_text;
885 datasize = (unsigned long) &_edata - (unsigned long) &_etext;
886 initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin;
888 kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
889 kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
890 VMALLOC_END-VMALLOC_START);
892 printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, "
893 "%dk reserved, %dk data, %dk init, %ldk highmem)\n",
894 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
895 num_physpages << (PAGE_SHIFT-10),
897 reservedpages << (PAGE_SHIFT-10),
900 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10))
903 printk(KERN_INFO "virtual kernel memory layout:\n"
904 " fixmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
905 #ifdef CONFIG_HIGHMEM
906 " pkmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
908 " vmalloc : 0x%08lx - 0x%08lx (%4ld MB)\n"
909 " lowmem : 0x%08lx - 0x%08lx (%4ld MB)\n"
910 " .init : 0x%08lx - 0x%08lx (%4ld kB)\n"
911 " .data : 0x%08lx - 0x%08lx (%4ld kB)\n"
912 " .text : 0x%08lx - 0x%08lx (%4ld kB)\n",
913 FIXADDR_START, FIXADDR_TOP,
914 (FIXADDR_TOP - FIXADDR_START) >> 10,
916 #ifdef CONFIG_HIGHMEM
917 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
918 (LAST_PKMAP*PAGE_SIZE) >> 10,
921 VMALLOC_START, VMALLOC_END,
922 (VMALLOC_END - VMALLOC_START) >> 20,
924 (unsigned long)__va(0), (unsigned long)high_memory,
925 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
927 (unsigned long)&__init_begin, (unsigned long)&__init_end,
928 ((unsigned long)&__init_end -
929 (unsigned long)&__init_begin) >> 10,
931 (unsigned long)&_etext, (unsigned long)&_edata,
932 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
934 (unsigned long)&_text, (unsigned long)&_etext,
935 ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
938 * Check boundaries twice: Some fundamental inconsistencies can
939 * be detected at build time already.
941 #define __FIXADDR_TOP (-PAGE_SIZE)
942 #ifdef CONFIG_HIGHMEM
943 BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START);
944 BUILD_BUG_ON(VMALLOC_END > PKMAP_BASE);
946 #define high_memory (-128UL << 20)
947 BUILD_BUG_ON(VMALLOC_START >= VMALLOC_END);
951 #ifdef CONFIG_HIGHMEM
952 BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START);
953 BUG_ON(VMALLOC_END > PKMAP_BASE);
955 BUG_ON(VMALLOC_START >= VMALLOC_END);
956 BUG_ON((unsigned long)high_memory > VMALLOC_START);
958 if (boot_cpu_data.wp_works_ok < 0)
965 #ifdef CONFIG_MEMORY_HOTPLUG
966 int arch_add_memory(int nid, u64 start, u64 size)
968 struct pglist_data *pgdata = NODE_DATA(nid);
969 struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
970 unsigned long start_pfn = start >> PAGE_SHIFT;
971 unsigned long nr_pages = size >> PAGE_SHIFT;
973 return __add_pages(nid, zone, start_pfn, nr_pages);
978 * This function cannot be __init, since exceptions don't work in that
979 * section. Put this after the callers, so that it cannot be inlined.
981 static noinline int do_test_wp_bit(void)
986 __asm__ __volatile__(
992 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
1001 #ifdef CONFIG_DEBUG_RODATA
1002 const int rodata_test_data = 0xC3;
1003 EXPORT_SYMBOL_GPL(rodata_test_data);
1005 static int kernel_set_to_readonly;
1007 void set_kernel_text_rw(void)
1009 unsigned long start = PFN_ALIGN(_text);
1010 unsigned long size = PFN_ALIGN(_etext) - start;
1012 if (!kernel_set_to_readonly)
1015 pr_debug("Set kernel text: %lx - %lx for read write\n",
1018 set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
1021 void set_kernel_text_ro(void)
1023 unsigned long start = PFN_ALIGN(_text);
1024 unsigned long size = PFN_ALIGN(_etext) - start;
1026 if (!kernel_set_to_readonly)
1029 pr_debug("Set kernel text: %lx - %lx for read only\n",
1032 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1035 void mark_rodata_ro(void)
1037 unsigned long start = PFN_ALIGN(_text);
1038 unsigned long size = PFN_ALIGN(_etext) - start;
1040 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1041 printk(KERN_INFO "Write protecting the kernel text: %luk\n",
1044 kernel_set_to_readonly = 1;
1046 #ifdef CONFIG_CPA_DEBUG
1047 printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n",
1049 set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT);
1051 printk(KERN_INFO "Testing CPA: write protecting again\n");
1052 set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT);
1056 size = (unsigned long)__end_rodata - start;
1057 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1058 printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
1062 #ifdef CONFIG_CPA_DEBUG
1063 printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size);
1064 set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
1066 printk(KERN_INFO "Testing CPA: write protecting again\n");
1067 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1072 int __init reserve_bootmem_generic(unsigned long phys, unsigned long len,
1075 return reserve_bootmem(phys, len, flags);