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/module.h>
10 #include <linux/signal.h>
11 #include <linux/sched.h>
12 #include <linux/kernel.h>
13 #include <linux/errno.h>
14 #include <linux/string.h>
15 #include <linux/types.h>
16 #include <linux/ptrace.h>
17 #include <linux/mman.h>
19 #include <linux/hugetlb.h>
20 #include <linux/swap.h>
21 #include <linux/smp.h>
22 #include <linux/init.h>
23 #include <linux/highmem.h>
24 #include <linux/pagemap.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/processor.h>
36 #include <asm/system.h>
37 #include <asm/uaccess.h>
38 #include <asm/pgtable.h>
40 #include <asm/fixmap.h>
45 #include <asm/tlbflush.h>
46 #include <asm/pgalloc.h>
47 #include <asm/sections.h>
48 #include <asm/paravirt.h>
49 #include <asm/setup.h>
50 #include <asm/cacheflush.h>
52 unsigned int __VMALLOC_RESERVE = 128 << 20;
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 * Creates a middle page table and puts a pointer to it in the
63 * given global directory entry. This only returns the gd entry
64 * in non-PAE compilation mode, since the middle layer is folded.
66 static pmd_t * __init one_md_table_init(pgd_t *pgd)
72 if (!(pgd_val(*pgd) & _PAGE_PRESENT)) {
73 pmd_table = (pmd_t *) alloc_bootmem_low_pages(PAGE_SIZE);
75 paravirt_alloc_pd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT);
76 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
77 pud = pud_offset(pgd, 0);
78 BUG_ON(pmd_table != pmd_offset(pud, 0));
81 pud = pud_offset(pgd, 0);
82 pmd_table = pmd_offset(pud, 0);
88 * Create a page table and place a pointer to it in a middle page
91 static pte_t * __init one_page_table_init(pmd_t *pmd)
93 if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
94 pte_t *page_table = NULL;
96 #ifdef CONFIG_DEBUG_PAGEALLOC
97 page_table = (pte_t *) alloc_bootmem_pages(PAGE_SIZE);
101 (pte_t *)alloc_bootmem_low_pages(PAGE_SIZE);
104 paravirt_alloc_pt(&init_mm, __pa(page_table) >> PAGE_SHIFT);
105 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
106 BUG_ON(page_table != pte_offset_kernel(pmd, 0));
109 return pte_offset_kernel(pmd, 0);
113 * This function initializes a certain range of kernel virtual memory
114 * with new bootmem page tables, everywhere page tables are missing in
117 * NOTE: The pagetables are allocated contiguous on the physical space
118 * so we can cache the place of the first one and move around without
119 * checking the pgd every time.
122 page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base)
124 int pgd_idx, pmd_idx;
130 pgd_idx = pgd_index(vaddr);
131 pmd_idx = pmd_index(vaddr);
132 pgd = pgd_base + pgd_idx;
134 for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
135 pmd = one_md_table_init(pgd);
136 pmd = pmd + pmd_index(vaddr);
137 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
139 one_page_table_init(pmd);
147 static inline int is_kernel_text(unsigned long addr)
149 if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end)
155 * This maps the physical memory to kernel virtual address space, a total
156 * of max_low_pfn pages, by creating page tables starting from address
159 static void __init kernel_physical_mapping_init(pgd_t *pgd_base)
161 int pgd_idx, pmd_idx, pte_ofs;
167 pgd_idx = pgd_index(PAGE_OFFSET);
168 pgd = pgd_base + pgd_idx;
171 for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
172 pmd = one_md_table_init(pgd);
173 if (pfn >= max_low_pfn)
177 pmd_idx < PTRS_PER_PMD && pfn < max_low_pfn;
179 unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET;
182 * Map with big pages if possible, otherwise
183 * create normal page tables:
185 * Don't use a large page for the first 2/4MB of memory
186 * because there are often fixed size MTRRs in there
187 * and overlapping MTRRs into large pages can cause
190 if (cpu_has_pse && !(pgd_idx == 0 && pmd_idx == 0)) {
192 pgprot_t prot = PAGE_KERNEL_LARGE;
194 addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
195 PAGE_OFFSET + PAGE_SIZE-1;
197 if (is_kernel_text(addr) ||
198 is_kernel_text(addr2))
199 prot = PAGE_KERNEL_LARGE_EXEC;
201 set_pmd(pmd, pfn_pmd(pfn, prot));
204 max_pfn_mapped = pfn;
207 pte = one_page_table_init(pmd);
210 pte_ofs < PTRS_PER_PTE && pfn < max_low_pfn;
211 pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
212 pgprot_t prot = PAGE_KERNEL;
214 if (is_kernel_text(addr))
215 prot = PAGE_KERNEL_EXEC;
217 set_pte(pte, pfn_pte(pfn, prot));
219 max_pfn_mapped = pfn;
224 static inline int page_kills_ppro(unsigned long pagenr)
226 if (pagenr >= 0x70000 && pagenr <= 0x7003F)
231 #ifdef CONFIG_HIGHMEM
235 static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr)
237 return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
238 vaddr), vaddr), vaddr);
241 static void __init kmap_init(void)
243 unsigned long kmap_vstart;
246 * 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);
289 add_one_highpage_hotplug(struct page *page, unsigned long pfn)
291 free_new_highpage(page);
293 #ifdef CONFIG_FLATMEM
294 max_mapnr = max(pfn, max_mapnr);
302 * Not currently handling the NUMA case.
303 * Assuming single node and all memory that
304 * has been added dynamically that would be
305 * onlined here is in HIGHMEM.
307 void __meminit online_page(struct page *page)
309 ClearPageReserved(page);
310 add_one_highpage_hotplug(page, page_to_pfn(page));
314 static void __init set_highmem_pages_init(int bad_ppro)
318 for (pfn = highstart_pfn; pfn < highend_pfn; pfn++) {
320 * Holes under sparsemem might not have no mem_map[]:
323 add_one_highpage_init(pfn_to_page(pfn), pfn, bad_ppro);
325 totalram_pages += totalhigh_pages;
327 #endif /* !CONFIG_NUMA */
330 # define kmap_init() do { } while (0)
331 # define permanent_kmaps_init(pgd_base) do { } while (0)
332 # define set_highmem_pages_init(bad_ppro) do { } while (0)
333 #endif /* CONFIG_HIGHMEM */
335 pteval_t __PAGE_KERNEL = _PAGE_KERNEL;
336 EXPORT_SYMBOL(__PAGE_KERNEL);
338 pteval_t __PAGE_KERNEL_EXEC = _PAGE_KERNEL_EXEC;
340 void __init native_pagetable_setup_start(pgd_t *base)
342 unsigned long pfn, va;
349 * Remove any mappings which extend past the end of physical
350 * memory from the boot time page table:
352 for (pfn = max_low_pfn + 1; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
353 va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
354 pgd = base + pgd_index(va);
355 if (!pgd_present(*pgd))
358 pud = pud_offset(pgd, va);
359 pmd = pmd_offset(pud, va);
360 if (!pmd_present(*pmd))
363 pte = pte_offset_kernel(pmd, va);
364 if (!pte_present(*pte))
367 pte_clear(NULL, va, pte);
369 paravirt_alloc_pd(&init_mm, __pa(base) >> PAGE_SHIFT);
372 void __init native_pagetable_setup_done(pgd_t *base)
377 * Build a proper pagetable for the kernel mappings. Up until this
378 * point, we've been running on some set of pagetables constructed by
381 * If we're booting on native hardware, this will be a pagetable
382 * constructed in arch/x86/kernel/head_32.S. The root of the
383 * pagetable will be swapper_pg_dir.
385 * If we're booting paravirtualized under a hypervisor, then there are
386 * more options: we may already be running PAE, and the pagetable may
387 * or may not be based in swapper_pg_dir. In any case,
388 * paravirt_pagetable_setup_start() will set up swapper_pg_dir
389 * appropriately for the rest of the initialization to work.
391 * In general, pagetable_init() assumes that the pagetable may already
392 * be partially populated, and so it avoids stomping on any existing
395 static void __init pagetable_init(void)
397 pgd_t *pgd_base = swapper_pg_dir;
398 unsigned long vaddr, end;
400 paravirt_pagetable_setup_start(pgd_base);
402 /* Enable PSE if available */
404 set_in_cr4(X86_CR4_PSE);
406 /* Enable PGE if available */
408 set_in_cr4(X86_CR4_PGE);
409 __PAGE_KERNEL |= _PAGE_GLOBAL;
410 __PAGE_KERNEL_EXEC |= _PAGE_GLOBAL;
413 kernel_physical_mapping_init(pgd_base);
417 * Fixed mappings, only the page table structure has to be
418 * created - mappings will be set by set_fixmap():
420 early_ioremap_clear();
421 vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
422 end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
423 page_table_range_init(vaddr, end, pgd_base);
424 early_ioremap_reset();
426 permanent_kmaps_init(pgd_base);
428 paravirt_pagetable_setup_done(pgd_base);
431 #ifdef CONFIG_ACPI_SLEEP
433 * ACPI suspend needs this for resume, because things like the intel-agp
434 * driver might have split up a kernel 4MB mapping.
436 char swsusp_pg_dir[PAGE_SIZE]
437 __attribute__ ((aligned(PAGE_SIZE)));
439 static inline void save_pg_dir(void)
441 memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
443 #else /* !CONFIG_ACPI_SLEEP */
444 static inline void save_pg_dir(void)
447 #endif /* !CONFIG_ACPI_SLEEP */
449 void zap_low_mappings(void)
456 * Zap initial low-memory mappings.
458 * Note that "pgd_clear()" doesn't do it for
459 * us, because pgd_clear() is a no-op on i386.
461 for (i = 0; i < USER_PTRS_PER_PGD; i++) {
462 #ifdef CONFIG_X86_PAE
463 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
465 set_pgd(swapper_pg_dir+i, __pgd(0));
473 pteval_t __supported_pte_mask __read_mostly = ~_PAGE_NX;
474 EXPORT_SYMBOL_GPL(__supported_pte_mask);
476 #ifdef CONFIG_X86_PAE
478 static int disable_nx __initdata;
483 * Control non executable mappings.
488 static int __init noexec_setup(char *str)
490 if (!str || !strcmp(str, "on")) {
492 __supported_pte_mask |= _PAGE_NX;
496 if (!strcmp(str, "off")) {
498 __supported_pte_mask &= ~_PAGE_NX;
506 early_param("noexec", noexec_setup);
508 static void __init set_nx(void)
510 unsigned int v[4], l, h;
512 if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) {
513 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]);
515 if ((v[3] & (1 << 20)) && !disable_nx) {
516 rdmsr(MSR_EFER, l, h);
518 wrmsr(MSR_EFER, l, h);
520 __supported_pte_mask |= _PAGE_NX;
527 * paging_init() sets up the page tables - note that the first 8MB are
528 * already mapped by head.S.
530 * This routines also unmaps the page at virtual kernel address 0, so
531 * that we can trap those pesky NULL-reference errors in the kernel.
533 void __init paging_init(void)
535 #ifdef CONFIG_X86_PAE
538 printk(KERN_INFO "NX (Execute Disable) protection: active\n");
542 load_cr3(swapper_pg_dir);
550 * Test if the WP bit works in supervisor mode. It isn't supported on 386's
551 * and also on some strange 486's (NexGen etc.). All 586+'s are OK. This
552 * used to involve black magic jumps to work around some nasty CPU bugs,
553 * but fortunately the switch to using exceptions got rid of all that.
555 static void __init test_wp_bit(void)
558 "Checking if this processor honours the WP bit even in supervisor mode...");
560 /* Any page-aligned address will do, the test is non-destructive */
561 __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
562 boot_cpu_data.wp_works_ok = do_test_wp_bit();
563 clear_fixmap(FIX_WP_TEST);
565 if (!boot_cpu_data.wp_works_ok) {
566 printk(KERN_CONT "No.\n");
567 #ifdef CONFIG_X86_WP_WORKS_OK
569 "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
572 printk(KERN_CONT "Ok.\n");
576 static struct kcore_list kcore_mem, kcore_vmalloc;
578 void __init mem_init(void)
580 int codesize, reservedpages, datasize, initsize;
583 #ifdef CONFIG_FLATMEM
586 bad_ppro = ppro_with_ram_bug();
588 #ifdef CONFIG_HIGHMEM
589 /* check that fixmap and pkmap do not overlap */
590 if (PKMAP_BASE + LAST_PKMAP*PAGE_SIZE >= FIXADDR_START) {
592 "fixmap and kmap areas overlap - this will crash\n");
593 printk(KERN_ERR "pkstart: %lxh pkend: %lxh fixstart %lxh\n",
594 PKMAP_BASE, PKMAP_BASE + LAST_PKMAP*PAGE_SIZE,
599 /* this will put all low memory onto the freelists */
600 totalram_pages += free_all_bootmem();
603 for (tmp = 0; tmp < max_low_pfn; tmp++)
605 * Only count reserved RAM pages:
607 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
610 set_highmem_pages_init(bad_ppro);
612 codesize = (unsigned long) &_etext - (unsigned long) &_text;
613 datasize = (unsigned long) &_edata - (unsigned long) &_etext;
614 initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin;
616 kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
617 kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
618 VMALLOC_END-VMALLOC_START);
620 printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, "
621 "%dk reserved, %dk data, %dk init, %ldk highmem)\n",
622 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
623 num_physpages << (PAGE_SHIFT-10),
625 reservedpages << (PAGE_SHIFT-10),
628 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10))
631 #if 1 /* double-sanity-check paranoia */
632 printk(KERN_INFO "virtual kernel memory layout:\n"
633 " fixmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
634 #ifdef CONFIG_HIGHMEM
635 " pkmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
637 " vmalloc : 0x%08lx - 0x%08lx (%4ld MB)\n"
638 " lowmem : 0x%08lx - 0x%08lx (%4ld MB)\n"
639 " .init : 0x%08lx - 0x%08lx (%4ld kB)\n"
640 " .data : 0x%08lx - 0x%08lx (%4ld kB)\n"
641 " .text : 0x%08lx - 0x%08lx (%4ld kB)\n",
642 FIXADDR_START, FIXADDR_TOP,
643 (FIXADDR_TOP - FIXADDR_START) >> 10,
645 #ifdef CONFIG_HIGHMEM
646 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
647 (LAST_PKMAP*PAGE_SIZE) >> 10,
650 VMALLOC_START, VMALLOC_END,
651 (VMALLOC_END - VMALLOC_START) >> 20,
653 (unsigned long)__va(0), (unsigned long)high_memory,
654 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
656 (unsigned long)&__init_begin, (unsigned long)&__init_end,
657 ((unsigned long)&__init_end -
658 (unsigned long)&__init_begin) >> 10,
660 (unsigned long)&_etext, (unsigned long)&_edata,
661 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
663 (unsigned long)&_text, (unsigned long)&_etext,
664 ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
666 #ifdef CONFIG_HIGHMEM
667 BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START);
668 BUG_ON(VMALLOC_END > PKMAP_BASE);
670 BUG_ON(VMALLOC_START > VMALLOC_END);
671 BUG_ON((unsigned long)high_memory > VMALLOC_START);
672 #endif /* double-sanity-check paranoia */
674 if (boot_cpu_data.wp_works_ok < 0)
680 * Subtle. SMP is doing it's boot stuff late (because it has to
681 * fork idle threads) - but it also needs low mappings for the
682 * protected-mode entry to work. We zap these entries only after
683 * the WP-bit has been tested.
690 #ifdef CONFIG_MEMORY_HOTPLUG
691 int arch_add_memory(int nid, u64 start, u64 size)
693 struct pglist_data *pgdata = NODE_DATA(nid);
694 struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
695 unsigned long start_pfn = start >> PAGE_SHIFT;
696 unsigned long nr_pages = size >> PAGE_SHIFT;
698 return __add_pages(zone, start_pfn, nr_pages);
703 * This function cannot be __init, since exceptions don't work in that
704 * section. Put this after the callers, so that it cannot be inlined.
706 static noinline int do_test_wp_bit(void)
711 __asm__ __volatile__(
717 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
726 #ifdef CONFIG_DEBUG_RODATA
727 const int rodata_test_data = 0xC3;
728 EXPORT_SYMBOL_GPL(rodata_test_data);
730 void mark_rodata_ro(void)
732 unsigned long start = PFN_ALIGN(_text);
733 unsigned long size = PFN_ALIGN(_etext) - start;
735 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
736 printk(KERN_INFO "Write protecting the kernel text: %luk\n",
739 #ifdef CONFIG_CPA_DEBUG
740 printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n",
742 set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT);
744 printk(KERN_INFO "Testing CPA: write protecting again\n");
745 set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT);
748 size = (unsigned long)__end_rodata - start;
749 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
750 printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
754 #ifdef CONFIG_CPA_DEBUG
755 printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size);
756 set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
758 printk(KERN_INFO "Testing CPA: write protecting again\n");
759 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
764 void free_init_pages(char *what, unsigned long begin, unsigned long end)
766 #ifdef CONFIG_DEBUG_PAGEALLOC
768 * If debugging page accesses then do not free this memory but
769 * mark them not present - any buggy init-section access will
770 * create a kernel page fault:
772 printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n",
773 begin, PAGE_ALIGN(end));
774 set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
779 * We just marked the kernel text read only above, now that
780 * we are going to free part of that, we need to make that
783 set_memory_rw(begin, (end - begin) >> PAGE_SHIFT);
785 for (addr = begin; addr < end; addr += PAGE_SIZE) {
786 ClearPageReserved(virt_to_page(addr));
787 init_page_count(virt_to_page(addr));
788 memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE);
792 printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
796 void free_initmem(void)
798 free_init_pages("unused kernel memory",
799 (unsigned long)(&__init_begin),
800 (unsigned long)(&__init_end));
803 #ifdef CONFIG_BLK_DEV_INITRD
804 void free_initrd_mem(unsigned long start, unsigned long end)
806 free_init_pages("initrd memory", start, end);