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>
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>
44 #include <asm/tlbflush.h>
45 #include <asm/pgalloc.h>
46 #include <asm/sections.h>
47 #include <asm/paravirt.h>
49 unsigned int __VMALLOC_RESERVE = 128 << 20;
51 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
52 unsigned long highstart_pfn, highend_pfn;
54 static noinline int do_test_wp_bit(void);
57 * Creates a middle page table and puts a pointer to it in the
58 * given global directory entry. This only returns the gd entry
59 * in non-PAE compilation mode, since the middle layer is folded.
61 static pmd_t * __init one_md_table_init(pgd_t *pgd)
67 if (!(pgd_val(*pgd) & _PAGE_PRESENT)) {
68 pmd_table = (pmd_t *) alloc_bootmem_low_pages(PAGE_SIZE);
70 paravirt_alloc_pd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT);
71 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
72 pud = pud_offset(pgd, 0);
73 BUG_ON(pmd_table != pmd_offset(pud, 0));
76 pud = pud_offset(pgd, 0);
77 pmd_table = pmd_offset(pud, 0);
83 * Create a page table and place a pointer to it in a middle page
86 static pte_t * __init one_page_table_init(pmd_t *pmd)
88 if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
89 pte_t *page_table = NULL;
91 #ifdef CONFIG_DEBUG_PAGEALLOC
92 page_table = (pte_t *) alloc_bootmem_pages(PAGE_SIZE);
96 (pte_t *)alloc_bootmem_low_pages(PAGE_SIZE);
99 paravirt_alloc_pt(&init_mm, __pa(page_table) >> PAGE_SHIFT);
100 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
101 BUG_ON(page_table != pte_offset_kernel(pmd, 0));
104 return pte_offset_kernel(pmd, 0);
108 * This function initializes a certain range of kernel virtual memory
109 * with new bootmem page tables, everywhere page tables are missing in
112 * NOTE: The pagetables are allocated contiguous on the physical space
113 * so we can cache the place of the first one and move around without
114 * checking the pgd every time.
117 page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base)
119 int pgd_idx, pmd_idx;
125 pgd_idx = pgd_index(vaddr);
126 pmd_idx = pmd_index(vaddr);
127 pgd = pgd_base + pgd_idx;
129 for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
130 pmd = one_md_table_init(pgd);
131 pmd = pmd + pmd_index(vaddr);
132 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
134 one_page_table_init(pmd);
142 static inline int is_kernel_text(unsigned long addr)
144 if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end)
150 * This maps the physical memory to kernel virtual address space, a total
151 * of max_low_pfn pages, by creating page tables starting from address
154 static void __init kernel_physical_mapping_init(pgd_t *pgd_base)
156 int pgd_idx, pmd_idx, pte_ofs;
162 pgd_idx = pgd_index(PAGE_OFFSET);
163 pgd = pgd_base + pgd_idx;
166 for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
167 pmd = one_md_table_init(pgd);
168 if (pfn >= max_low_pfn)
172 pmd_idx < PTRS_PER_PMD && pfn < max_low_pfn;
174 unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET;
177 * Map with big pages if possible, otherwise
178 * create normal page tables:
182 pgprot_t prot = PAGE_KERNEL_LARGE;
184 addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
185 PAGE_OFFSET + PAGE_SIZE-1;
187 if (is_kernel_text(addr) ||
188 is_kernel_text(addr2))
189 prot = PAGE_KERNEL_LARGE_EXEC;
191 set_pmd(pmd, pfn_pmd(pfn, prot));
196 pte = one_page_table_init(pmd);
199 pte_ofs < PTRS_PER_PTE && pfn < max_low_pfn;
200 pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
201 pgprot_t prot = PAGE_KERNEL;
203 if (is_kernel_text(addr))
204 prot = PAGE_KERNEL_EXEC;
206 set_pte(pte, pfn_pte(pfn, prot));
212 static inline int page_kills_ppro(unsigned long pagenr)
214 if (pagenr >= 0x70000 && pagenr <= 0x7003F)
219 #ifdef CONFIG_HIGHMEM
223 static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr)
225 return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
226 vaddr), vaddr), vaddr);
229 static void __init kmap_init(void)
231 unsigned long kmap_vstart;
234 * Cache the first kmap pte:
236 kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
237 kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
239 kmap_prot = PAGE_KERNEL;
242 static void __init permanent_kmaps_init(pgd_t *pgd_base)
251 page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
253 pgd = swapper_pg_dir + pgd_index(vaddr);
254 pud = pud_offset(pgd, vaddr);
255 pmd = pmd_offset(pud, vaddr);
256 pte = pte_offset_kernel(pmd, vaddr);
257 pkmap_page_table = pte;
260 static void __meminit free_new_highpage(struct page *page)
262 init_page_count(page);
267 void __init add_one_highpage_init(struct page *page, int pfn, int bad_ppro)
269 if (page_is_ram(pfn) && !(bad_ppro && page_kills_ppro(pfn))) {
270 ClearPageReserved(page);
271 free_new_highpage(page);
273 SetPageReserved(page);
277 add_one_highpage_hotplug(struct page *page, unsigned long pfn)
279 free_new_highpage(page);
281 #ifdef CONFIG_FLATMEM
282 max_mapnr = max(pfn, max_mapnr);
290 * Not currently handling the NUMA case.
291 * Assuming single node and all memory that
292 * has been added dynamically that would be
293 * onlined here is in HIGHMEM.
295 void __meminit online_page(struct page *page)
297 ClearPageReserved(page);
298 add_one_highpage_hotplug(page, page_to_pfn(page));
302 static void __init set_highmem_pages_init(int bad_ppro)
306 for (pfn = highstart_pfn; pfn < highend_pfn; pfn++) {
308 * Holes under sparsemem might not have no mem_map[]:
311 add_one_highpage_init(pfn_to_page(pfn), pfn, bad_ppro);
313 totalram_pages += totalhigh_pages;
315 #endif /* !CONFIG_NUMA */
318 # define kmap_init() do { } while (0)
319 # define permanent_kmaps_init(pgd_base) do { } while (0)
320 # define set_highmem_pages_init(bad_ppro) do { } while (0)
321 #endif /* CONFIG_HIGHMEM */
323 pteval_t __PAGE_KERNEL = _PAGE_KERNEL;
324 EXPORT_SYMBOL(__PAGE_KERNEL);
326 pteval_t __PAGE_KERNEL_EXEC = _PAGE_KERNEL_EXEC;
328 void __init native_pagetable_setup_start(pgd_t *base)
330 #ifdef CONFIG_X86_PAE
334 * Init entries of the first-level page table to the
335 * zero page, if they haven't already been set up.
337 * In a normal native boot, we'll be running on a
338 * pagetable rooted in swapper_pg_dir, but not in PAE
339 * mode, so this will end up clobbering the mappings
340 * for the lower 24Mbytes of the address space,
341 * without affecting the kernel address space.
343 for (i = 0; i < USER_PTRS_PER_PGD; i++)
345 __pgd(__pa(empty_zero_page) | _PAGE_PRESENT));
347 /* Make sure kernel address space is empty so that a pagetable
348 will be allocated for it. */
349 memset(&base[USER_PTRS_PER_PGD], 0,
350 KERNEL_PGD_PTRS * sizeof(pgd_t));
352 paravirt_alloc_pd(&init_mm, __pa(base) >> PAGE_SHIFT);
356 void __init native_pagetable_setup_done(pgd_t *base)
358 #ifdef CONFIG_X86_PAE
360 * Add low memory identity-mappings - SMP needs it when
361 * starting up on an AP from real-mode. In the non-PAE
362 * case we already have these mappings through head.S.
363 * All user-space mappings are explicitly cleared after
366 set_pgd(&base[0], base[USER_PTRS_PER_PGD]);
371 * Build a proper pagetable for the kernel mappings. Up until this
372 * point, we've been running on some set of pagetables constructed by
375 * If we're booting on native hardware, this will be a pagetable
376 * constructed in arch/i386/kernel/head.S, and not running in PAE mode
377 * (even if we'll end up running in PAE). The root of the pagetable
378 * will be swapper_pg_dir.
380 * If we're booting paravirtualized under a hypervisor, then there are
381 * more options: we may already be running PAE, and the pagetable may
382 * or may not be based in swapper_pg_dir. In any case,
383 * paravirt_pagetable_setup_start() will set up swapper_pg_dir
384 * appropriately for the rest of the initialization to work.
386 * In general, pagetable_init() assumes that the pagetable may already
387 * be partially populated, and so it avoids stomping on any existing
390 static void __init pagetable_init(void)
392 pgd_t *pgd_base = swapper_pg_dir;
393 unsigned long vaddr, end;
395 paravirt_pagetable_setup_start(pgd_base);
397 /* Enable PSE if available */
399 set_in_cr4(X86_CR4_PSE);
401 /* Enable PGE if available */
403 set_in_cr4(X86_CR4_PGE);
404 __PAGE_KERNEL |= _PAGE_GLOBAL;
405 __PAGE_KERNEL_EXEC |= _PAGE_GLOBAL;
408 kernel_physical_mapping_init(pgd_base);
412 * Fixed mappings, only the page table structure has to be
413 * created - mappings will be set by set_fixmap():
415 early_ioremap_clear();
416 vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
417 end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
418 page_table_range_init(vaddr, end, pgd_base);
419 early_ioremap_reset();
421 permanent_kmaps_init(pgd_base);
423 paravirt_pagetable_setup_done(pgd_base);
426 #if defined(CONFIG_HIBERNATION) || defined(CONFIG_ACPI)
428 * Swap suspend & friends need this for resume because things like the intel-agp
429 * driver might have split up a kernel 4MB mapping.
431 char __nosavedata swsusp_pg_dir[PAGE_SIZE]
432 __attribute__ ((aligned(PAGE_SIZE)));
434 static inline void save_pg_dir(void)
436 memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
439 static inline void save_pg_dir(void)
444 void zap_low_mappings(void)
451 * Zap initial low-memory mappings.
453 * Note that "pgd_clear()" doesn't do it for
454 * us, because pgd_clear() is a no-op on i386.
456 for (i = 0; i < USER_PTRS_PER_PGD; i++) {
457 #ifdef CONFIG_X86_PAE
458 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
460 set_pgd(swapper_pg_dir+i, __pgd(0));
468 pteval_t __supported_pte_mask __read_mostly = ~_PAGE_NX;
469 EXPORT_SYMBOL_GPL(__supported_pte_mask);
471 #ifdef CONFIG_X86_PAE
473 static int disable_nx __initdata;
478 * Control non executable mappings.
483 static int __init noexec_setup(char *str)
485 if (!str || !strcmp(str, "on")) {
487 __supported_pte_mask |= _PAGE_NX;
491 if (!strcmp(str, "off")) {
493 __supported_pte_mask &= ~_PAGE_NX;
501 early_param("noexec", noexec_setup);
503 static void __init set_nx(void)
505 unsigned int v[4], l, h;
507 if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) {
508 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]);
510 if ((v[3] & (1 << 20)) && !disable_nx) {
511 rdmsr(MSR_EFER, l, h);
513 wrmsr(MSR_EFER, l, h);
515 __supported_pte_mask |= _PAGE_NX;
522 * paging_init() sets up the page tables - note that the first 8MB are
523 * already mapped by head.S.
525 * This routines also unmaps the page at virtual kernel address 0, so
526 * that we can trap those pesky NULL-reference errors in the kernel.
528 void __init paging_init(void)
530 #ifdef CONFIG_X86_PAE
533 printk(KERN_INFO "NX (Execute Disable) protection: active\n");
537 load_cr3(swapper_pg_dir);
539 #ifdef CONFIG_X86_PAE
541 * We will bail out later - printk doesn't work right now so
542 * the user would just see a hanging kernel.
545 set_in_cr4(X86_CR4_PAE);
553 * Test if the WP bit works in supervisor mode. It isn't supported on 386's
554 * and also on some strange 486's (NexGen etc.). All 586+'s are OK. This
555 * used to involve black magic jumps to work around some nasty CPU bugs,
556 * but fortunately the switch to using exceptions got rid of all that.
558 static void __init test_wp_bit(void)
561 "Checking if this processor honours the WP bit even in supervisor mode...");
563 /* Any page-aligned address will do, the test is non-destructive */
564 __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
565 boot_cpu_data.wp_works_ok = do_test_wp_bit();
566 clear_fixmap(FIX_WP_TEST);
568 if (!boot_cpu_data.wp_works_ok) {
569 printk(KERN_CONT "No.\n");
570 #ifdef CONFIG_X86_WP_WORKS_OK
572 "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
575 printk(KERN_CONT "Ok.\n");
579 static struct kcore_list kcore_mem, kcore_vmalloc;
581 void __init mem_init(void)
583 int codesize, reservedpages, datasize, initsize;
586 #ifdef CONFIG_FLATMEM
589 bad_ppro = ppro_with_ram_bug();
591 #ifdef CONFIG_HIGHMEM
592 /* check that fixmap and pkmap do not overlap */
593 if (PKMAP_BASE + LAST_PKMAP*PAGE_SIZE >= FIXADDR_START) {
595 "fixmap and kmap areas overlap - this will crash\n");
596 printk(KERN_ERR "pkstart: %lxh pkend: %lxh fixstart %lxh\n",
597 PKMAP_BASE, PKMAP_BASE + LAST_PKMAP*PAGE_SIZE,
602 /* this will put all low memory onto the freelists */
603 totalram_pages += free_all_bootmem();
606 for (tmp = 0; tmp < max_low_pfn; tmp++)
608 * Only count reserved RAM pages:
610 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
613 set_highmem_pages_init(bad_ppro);
615 codesize = (unsigned long) &_etext - (unsigned long) &_text;
616 datasize = (unsigned long) &_edata - (unsigned long) &_etext;
617 initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin;
619 kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
620 kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
621 VMALLOC_END-VMALLOC_START);
623 printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, "
624 "%dk reserved, %dk data, %dk init, %ldk highmem)\n",
625 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
626 num_physpages << (PAGE_SHIFT-10),
628 reservedpages << (PAGE_SHIFT-10),
631 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10))
634 #if 1 /* double-sanity-check paranoia */
635 printk(KERN_INFO "virtual kernel memory layout:\n"
636 " fixmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
637 #ifdef CONFIG_HIGHMEM
638 " pkmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
640 " vmalloc : 0x%08lx - 0x%08lx (%4ld MB)\n"
641 " lowmem : 0x%08lx - 0x%08lx (%4ld MB)\n"
642 " .init : 0x%08lx - 0x%08lx (%4ld kB)\n"
643 " .data : 0x%08lx - 0x%08lx (%4ld kB)\n"
644 " .text : 0x%08lx - 0x%08lx (%4ld kB)\n",
645 FIXADDR_START, FIXADDR_TOP,
646 (FIXADDR_TOP - FIXADDR_START) >> 10,
648 #ifdef CONFIG_HIGHMEM
649 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
650 (LAST_PKMAP*PAGE_SIZE) >> 10,
653 VMALLOC_START, VMALLOC_END,
654 (VMALLOC_END - VMALLOC_START) >> 20,
656 (unsigned long)__va(0), (unsigned long)high_memory,
657 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
659 (unsigned long)&__init_begin, (unsigned long)&__init_end,
660 ((unsigned long)&__init_end -
661 (unsigned long)&__init_begin) >> 10,
663 (unsigned long)&_etext, (unsigned long)&_edata,
664 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
666 (unsigned long)&_text, (unsigned long)&_etext,
667 ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
669 #ifdef CONFIG_HIGHMEM
670 BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START);
671 BUG_ON(VMALLOC_END > PKMAP_BASE);
673 BUG_ON(VMALLOC_START > VMALLOC_END);
674 BUG_ON((unsigned long)high_memory > VMALLOC_START);
675 #endif /* double-sanity-check paranoia */
677 #ifdef CONFIG_X86_PAE
679 panic("cannot execute a PAE-enabled kernel on a PAE-less CPU!");
681 if (boot_cpu_data.wp_works_ok < 0)
685 * Subtle. SMP is doing it's boot stuff late (because it has to
686 * fork idle threads) - but it also needs low mappings for the
687 * protected-mode entry to work. We zap these entries only after
688 * the WP-bit has been tested.
695 #ifdef CONFIG_MEMORY_HOTPLUG
696 int arch_add_memory(int nid, u64 start, u64 size)
698 struct pglist_data *pgdata = NODE_DATA(nid);
699 struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
700 unsigned long start_pfn = start >> PAGE_SHIFT;
701 unsigned long nr_pages = size >> PAGE_SHIFT;
703 return __add_pages(zone, start_pfn, nr_pages);
708 * This function cannot be __init, since exceptions don't work in that
709 * section. Put this after the callers, so that it cannot be inlined.
711 static noinline int do_test_wp_bit(void)
716 __asm__ __volatile__(
721 ".section __ex_table, \"a\"\n"
725 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
734 #ifdef CONFIG_DEBUG_RODATA
735 const int rodata_test_data = 0xC3;
736 EXPORT_SYMBOL_GPL(rodata_test_data);
738 void mark_rodata_ro(void)
740 unsigned long start = PFN_ALIGN(_text);
741 unsigned long size = PFN_ALIGN(_etext) - start;
743 #ifndef CONFIG_KPROBES
744 #ifdef CONFIG_HOTPLUG_CPU
745 /* It must still be possible to apply SMP alternatives. */
746 if (num_possible_cpus() <= 1)
749 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
750 printk(KERN_INFO "Write protecting the kernel text: %luk\n",
753 #ifdef CONFIG_CPA_DEBUG
754 printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n",
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 size = (unsigned long)__end_rodata - start;
765 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
766 printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
770 #ifdef CONFIG_CPA_DEBUG
771 printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size);
772 set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
774 printk(KERN_INFO "Testing CPA: write protecting again\n");
775 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
780 void free_init_pages(char *what, unsigned long begin, unsigned long end)
782 #ifdef CONFIG_DEBUG_PAGEALLOC
784 * If debugging page accesses then do not free this memory but
785 * mark them not present - any buggy init-section access will
786 * create a kernel page fault:
788 printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n",
789 begin, PAGE_ALIGN(end));
790 set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
795 * We just marked the kernel text read only above, now that
796 * we are going to free part of that, we need to make that
799 set_memory_rw(begin, (end - begin) >> PAGE_SHIFT);
801 for (addr = begin; addr < end; addr += PAGE_SIZE) {
802 ClearPageReserved(virt_to_page(addr));
803 init_page_count(virt_to_page(addr));
804 memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE);
808 printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
812 void free_initmem(void)
814 free_init_pages("unused kernel memory",
815 (unsigned long)(&__init_begin),
816 (unsigned long)(&__init_end));
819 #ifdef CONFIG_BLK_DEV_INITRD
820 void free_initrd_mem(unsigned long start, unsigned long end)
822 free_init_pages("initrd memory", start, end);