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>
43 #include <asm/tlbflush.h>
44 #include <asm/sections.h>
45 #include <asm/paravirt.h>
47 unsigned int __VMALLOC_RESERVE = 128 << 20;
49 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
50 unsigned long highstart_pfn, highend_pfn;
52 static int noinline do_test_wp_bit(void);
55 * Creates a middle page table and puts a pointer to it in the
56 * given global directory entry. This only returns the gd entry
57 * in non-PAE compilation mode, since the middle layer is folded.
59 static pmd_t * __init one_md_table_init(pgd_t *pgd)
65 if (!(pgd_val(*pgd) & _PAGE_PRESENT)) {
66 pmd_table = (pmd_t *) alloc_bootmem_low_pages(PAGE_SIZE);
68 paravirt_alloc_pd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT);
69 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
70 pud = pud_offset(pgd, 0);
71 if (pmd_table != pmd_offset(pud, 0))
75 pud = pud_offset(pgd, 0);
76 pmd_table = pmd_offset(pud, 0);
81 * Create a page table and place a pointer to it in a middle page
84 static pte_t * __init one_page_table_init(pmd_t *pmd)
86 if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
87 pte_t *page_table = NULL;
89 #ifdef CONFIG_DEBUG_PAGEALLOC
90 page_table = (pte_t *) alloc_bootmem_pages(PAGE_SIZE);
94 (pte_t *)alloc_bootmem_low_pages(PAGE_SIZE);
96 paravirt_alloc_pt(&init_mm, __pa(page_table) >> PAGE_SHIFT);
97 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
98 BUG_ON(page_table != pte_offset_kernel(pmd, 0));
101 return pte_offset_kernel(pmd, 0);
105 * This function initializes a certain range of kernel virtual memory
106 * with new bootmem page tables, everywhere page tables are missing in
111 * NOTE: The pagetables are allocated contiguous on the physical space
112 * so we can cache the place of the first one and move around without
113 * checking the pgd every time.
115 static void __init page_table_range_init (unsigned long start, unsigned long end, pgd_t *pgd_base)
119 int pgd_idx, pmd_idx;
123 pgd_idx = pgd_index(vaddr);
124 pmd_idx = pmd_index(vaddr);
125 pgd = pgd_base + pgd_idx;
127 for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
128 pmd = one_md_table_init(pgd);
129 pmd = pmd + pmd_index(vaddr);
130 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end); pmd++, pmd_idx++) {
131 one_page_table_init(pmd);
139 static inline int is_kernel_text(unsigned long addr)
141 if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end)
147 * This maps the physical memory to kernel virtual address space, a total
148 * of max_low_pfn pages, by creating page tables starting from address
151 static void __init kernel_physical_mapping_init(pgd_t *pgd_base)
157 int pgd_idx, pmd_idx, pte_ofs;
159 pgd_idx = pgd_index(PAGE_OFFSET);
160 pgd = pgd_base + pgd_idx;
163 for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
164 pmd = one_md_table_init(pgd);
165 if (pfn >= max_low_pfn)
168 pmd_idx < PTRS_PER_PMD && pfn < max_low_pfn;
170 unsigned int address = pfn * PAGE_SIZE + PAGE_OFFSET;
172 /* Map with big pages if possible, otherwise
173 create normal page tables. */
175 unsigned int address2;
176 pgprot_t prot = PAGE_KERNEL_LARGE;
178 address2 = (pfn + PTRS_PER_PTE - 1) * PAGE_SIZE +
179 PAGE_OFFSET + PAGE_SIZE-1;
181 if (is_kernel_text(address) ||
182 is_kernel_text(address2))
183 prot = PAGE_KERNEL_LARGE_EXEC;
185 set_pmd(pmd, pfn_pmd(pfn, prot));
189 pte = one_page_table_init(pmd);
192 pte_ofs < PTRS_PER_PTE && pfn < max_low_pfn;
193 pte++, pfn++, pte_ofs++, address += PAGE_SIZE) {
194 pgprot_t prot = PAGE_KERNEL;
196 if (is_kernel_text(address))
197 prot = PAGE_KERNEL_EXEC;
199 set_pte(pte, pfn_pte(pfn, prot));
206 static inline int page_kills_ppro(unsigned long pagenr)
208 if (pagenr >= 0x70000 && pagenr <= 0x7003F)
213 int page_is_ram(unsigned long pagenr)
216 unsigned long addr, end;
218 for (i = 0; i < e820.nr_map; i++) {
220 if (e820.map[i].type != E820_RAM) /* not usable memory */
223 * !!!FIXME!!! Some BIOSen report areas as RAM that
224 * are not. Notably the 640->1Mb area. We need a sanity
227 addr = (e820.map[i].addr+PAGE_SIZE-1) >> PAGE_SHIFT;
228 end = (e820.map[i].addr+e820.map[i].size) >> PAGE_SHIFT;
229 if ((pagenr >= addr) && (pagenr < end))
235 #ifdef CONFIG_HIGHMEM
239 #define kmap_get_fixmap_pte(vaddr) \
240 pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr), vaddr), (vaddr)), (vaddr))
242 static void __init kmap_init(void)
244 unsigned long kmap_vstart;
246 /* cache the first kmap pte */
247 kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
248 kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
250 kmap_prot = PAGE_KERNEL;
253 static void __init permanent_kmaps_init(pgd_t *pgd_base)
262 page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
264 pgd = swapper_pg_dir + pgd_index(vaddr);
265 pud = pud_offset(pgd, vaddr);
266 pmd = pmd_offset(pud, vaddr);
267 pte = pte_offset_kernel(pmd, vaddr);
268 pkmap_page_table = pte;
271 static void __meminit free_new_highpage(struct page *page)
273 init_page_count(page);
278 void __init add_one_highpage_init(struct page *page, int pfn, int bad_ppro)
280 if (page_is_ram(pfn) && !(bad_ppro && page_kills_ppro(pfn))) {
281 ClearPageReserved(page);
282 free_new_highpage(page);
284 SetPageReserved(page);
287 static int __meminit add_one_highpage_hotplug(struct page *page, unsigned long pfn)
289 free_new_highpage(page);
291 #ifdef CONFIG_FLATMEM
292 max_mapnr = max(pfn, max_mapnr);
299 * Not currently handling the NUMA case.
300 * Assuming single node and all memory that
301 * has been added dynamically that would be
302 * onlined here is in HIGHMEM
304 void __meminit online_page(struct page *page)
306 ClearPageReserved(page);
307 add_one_highpage_hotplug(page, page_to_pfn(page));
312 extern void set_highmem_pages_init(int);
314 static void __init set_highmem_pages_init(int bad_ppro)
317 for (pfn = highstart_pfn; pfn < highend_pfn; pfn++) {
319 * Holes under sparsemem might not have no mem_map[]:
322 add_one_highpage_init(pfn_to_page(pfn), pfn, bad_ppro);
324 totalram_pages += totalhigh_pages;
326 #endif /* CONFIG_FLATMEM */
329 #define kmap_init() do { } while (0)
330 #define permanent_kmaps_init(pgd_base) do { } while (0)
331 #define set_highmem_pages_init(bad_ppro) do { } while (0)
332 #endif /* CONFIG_HIGHMEM */
334 unsigned long long __PAGE_KERNEL = _PAGE_KERNEL;
335 EXPORT_SYMBOL(__PAGE_KERNEL);
336 unsigned long long __PAGE_KERNEL_EXEC = _PAGE_KERNEL_EXEC;
339 extern void __init remap_numa_kva(void);
341 #define remap_numa_kva() do {} while (0)
344 void __init native_pagetable_setup_start(pgd_t *base)
346 #ifdef CONFIG_X86_PAE
350 * Init entries of the first-level page table to the
351 * zero page, if they haven't already been set up.
353 * In a normal native boot, we'll be running on a
354 * pagetable rooted in swapper_pg_dir, but not in PAE
355 * mode, so this will end up clobbering the mappings
356 * for the lower 24Mbytes of the address space,
357 * without affecting the kernel address space.
359 for (i = 0; i < USER_PTRS_PER_PGD; i++)
361 __pgd(__pa(empty_zero_page) | _PAGE_PRESENT));
363 /* Make sure kernel address space is empty so that a pagetable
364 will be allocated for it. */
365 memset(&base[USER_PTRS_PER_PGD], 0,
366 KERNEL_PGD_PTRS * sizeof(pgd_t));
368 paravirt_alloc_pd(&init_mm, __pa(base) >> PAGE_SHIFT);
372 void __init native_pagetable_setup_done(pgd_t *base)
374 #ifdef CONFIG_X86_PAE
376 * Add low memory identity-mappings - SMP needs it when
377 * starting up on an AP from real-mode. In the non-PAE
378 * case we already have these mappings through head.S.
379 * All user-space mappings are explicitly cleared after
382 set_pgd(&base[0], base[USER_PTRS_PER_PGD]);
387 * Build a proper pagetable for the kernel mappings. Up until this
388 * point, we've been running on some set of pagetables constructed by
391 * If we're booting on native hardware, this will be a pagetable
392 * constructed in arch/i386/kernel/head.S, and not running in PAE mode
393 * (even if we'll end up running in PAE). The root of the pagetable
394 * will be swapper_pg_dir.
396 * If we're booting paravirtualized under a hypervisor, then there are
397 * more options: we may already be running PAE, and the pagetable may
398 * or may not be based in swapper_pg_dir. In any case,
399 * paravirt_pagetable_setup_start() will set up swapper_pg_dir
400 * appropriately for the rest of the initialization to work.
402 * In general, pagetable_init() assumes that the pagetable may already
403 * be partially populated, and so it avoids stomping on any existing
406 static void __init pagetable_init (void)
408 unsigned long vaddr, end;
409 pgd_t *pgd_base = swapper_pg_dir;
411 paravirt_pagetable_setup_start(pgd_base);
413 /* Enable PSE if available */
415 set_in_cr4(X86_CR4_PSE);
417 /* Enable PGE if available */
419 set_in_cr4(X86_CR4_PGE);
420 __PAGE_KERNEL |= _PAGE_GLOBAL;
421 __PAGE_KERNEL_EXEC |= _PAGE_GLOBAL;
424 kernel_physical_mapping_init(pgd_base);
428 * Fixed mappings, only the page table structure has to be
429 * created - mappings will be set by set_fixmap():
431 vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
432 end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
433 page_table_range_init(vaddr, end, pgd_base);
435 permanent_kmaps_init(pgd_base);
437 paravirt_pagetable_setup_done(pgd_base);
440 #if defined(CONFIG_HIBERNATION) || defined(CONFIG_ACPI)
442 * Swap suspend & friends need this for resume because things like the intel-agp
443 * driver might have split up a kernel 4MB mapping.
445 char __nosavedata swsusp_pg_dir[PAGE_SIZE]
446 __attribute__ ((aligned (PAGE_SIZE)));
448 static inline void save_pg_dir(void)
450 memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
453 static inline void save_pg_dir(void)
458 void zap_low_mappings (void)
465 * Zap initial low-memory mappings.
467 * Note that "pgd_clear()" doesn't do it for
468 * us, because pgd_clear() is a no-op on i386.
470 for (i = 0; i < USER_PTRS_PER_PGD; i++)
471 #ifdef CONFIG_X86_PAE
472 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
474 set_pgd(swapper_pg_dir+i, __pgd(0));
481 pteval_t __supported_pte_mask __read_mostly = ~_PAGE_NX;
482 EXPORT_SYMBOL_GPL(__supported_pte_mask);
484 #ifdef CONFIG_X86_PAE
486 static int disable_nx __initdata = 0;
491 * Control non executable mappings.
496 static int __init noexec_setup(char *str)
498 if (!str || !strcmp(str, "on")) {
500 __supported_pte_mask |= _PAGE_NX;
503 } else if (!strcmp(str,"off")) {
505 __supported_pte_mask &= ~_PAGE_NX;
511 early_param("noexec", noexec_setup);
513 static void __init set_nx(void)
515 unsigned int v[4], l, h;
517 if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) {
518 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]);
519 if ((v[3] & (1 << 20)) && !disable_nx) {
520 rdmsr(MSR_EFER, l, h);
522 wrmsr(MSR_EFER, l, h);
524 __supported_pte_mask |= _PAGE_NX;
530 * Enables/disables executability of a given kernel page and
531 * returns the previous setting.
533 int __init set_kernel_exec(unsigned long vaddr, int enable)
541 pte = lookup_address(vaddr);
544 if (!pte_exec_kernel(*pte))
548 pte->pte_high &= ~(1 << (_PAGE_BIT_NX - 32));
550 pte->pte_high |= 1 << (_PAGE_BIT_NX - 32);
551 pte_update_defer(&init_mm, vaddr, pte);
560 * paging_init() sets up the page tables - note that the first 8MB are
561 * already mapped by head.S.
563 * This routines also unmaps the page at virtual kernel address 0, so
564 * that we can trap those pesky NULL-reference errors in the kernel.
566 void __init paging_init(void)
568 #ifdef CONFIG_X86_PAE
571 printk("NX (Execute Disable) protection: active\n");
576 load_cr3(swapper_pg_dir);
578 #ifdef CONFIG_X86_PAE
580 * We will bail out later - printk doesn't work right now so
581 * the user would just see a hanging kernel.
584 set_in_cr4(X86_CR4_PAE);
592 * Test if the WP bit works in supervisor mode. It isn't supported on 386's
593 * and also on some strange 486's (NexGen etc.). All 586+'s are OK. This
594 * used to involve black magic jumps to work around some nasty CPU bugs,
595 * but fortunately the switch to using exceptions got rid of all that.
598 static void __init test_wp_bit(void)
600 printk("Checking if this processor honours the WP bit even in supervisor mode... ");
602 /* Any page-aligned address will do, the test is non-destructive */
603 __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
604 boot_cpu_data.wp_works_ok = do_test_wp_bit();
605 clear_fixmap(FIX_WP_TEST);
607 if (!boot_cpu_data.wp_works_ok) {
609 #ifdef CONFIG_X86_WP_WORKS_OK
610 panic("This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
617 static struct kcore_list kcore_mem, kcore_vmalloc;
619 void __init mem_init(void)
621 extern int ppro_with_ram_bug(void);
622 int codesize, reservedpages, datasize, initsize;
626 #ifdef CONFIG_FLATMEM
630 bad_ppro = ppro_with_ram_bug();
632 #ifdef CONFIG_HIGHMEM
633 /* check that fixmap and pkmap do not overlap */
634 if (PKMAP_BASE+LAST_PKMAP*PAGE_SIZE >= FIXADDR_START) {
635 printk(KERN_ERR "fixmap and kmap areas overlap - this will crash\n");
636 printk(KERN_ERR "pkstart: %lxh pkend: %lxh fixstart %lxh\n",
637 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE, FIXADDR_START);
642 /* this will put all low memory onto the freelists */
643 totalram_pages += free_all_bootmem();
646 for (tmp = 0; tmp < max_low_pfn; tmp++)
648 * Only count reserved RAM pages
650 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
653 set_highmem_pages_init(bad_ppro);
655 codesize = (unsigned long) &_etext - (unsigned long) &_text;
656 datasize = (unsigned long) &_edata - (unsigned long) &_etext;
657 initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin;
659 kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
660 kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
661 VMALLOC_END-VMALLOC_START);
663 printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, %dk reserved, %dk data, %dk init, %ldk highmem)\n",
664 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
665 num_physpages << (PAGE_SHIFT-10),
667 reservedpages << (PAGE_SHIFT-10),
670 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10))
673 #if 1 /* double-sanity-check paranoia */
674 printk("virtual kernel memory layout:\n"
675 " fixmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
676 #ifdef CONFIG_HIGHMEM
677 " pkmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
679 " vmalloc : 0x%08lx - 0x%08lx (%4ld MB)\n"
680 " lowmem : 0x%08lx - 0x%08lx (%4ld MB)\n"
681 " .init : 0x%08lx - 0x%08lx (%4ld kB)\n"
682 " .data : 0x%08lx - 0x%08lx (%4ld kB)\n"
683 " .text : 0x%08lx - 0x%08lx (%4ld kB)\n",
684 FIXADDR_START, FIXADDR_TOP,
685 (FIXADDR_TOP - FIXADDR_START) >> 10,
687 #ifdef CONFIG_HIGHMEM
688 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
689 (LAST_PKMAP*PAGE_SIZE) >> 10,
692 VMALLOC_START, VMALLOC_END,
693 (VMALLOC_END - VMALLOC_START) >> 20,
695 (unsigned long)__va(0), (unsigned long)high_memory,
696 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
698 (unsigned long)&__init_begin, (unsigned long)&__init_end,
699 ((unsigned long)&__init_end - (unsigned long)&__init_begin) >> 10,
701 (unsigned long)&_etext, (unsigned long)&_edata,
702 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
704 (unsigned long)&_text, (unsigned long)&_etext,
705 ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
707 #ifdef CONFIG_HIGHMEM
708 BUG_ON(PKMAP_BASE+LAST_PKMAP*PAGE_SIZE > FIXADDR_START);
709 BUG_ON(VMALLOC_END > PKMAP_BASE);
711 BUG_ON(VMALLOC_START > VMALLOC_END);
712 BUG_ON((unsigned long)high_memory > VMALLOC_START);
713 #endif /* double-sanity-check paranoia */
715 #ifdef CONFIG_X86_PAE
717 panic("cannot execute a PAE-enabled kernel on a PAE-less CPU!");
719 if (boot_cpu_data.wp_works_ok < 0)
723 * Subtle. SMP is doing it's boot stuff late (because it has to
724 * fork idle threads) - but it also needs low mappings for the
725 * protected-mode entry to work. We zap these entries only after
726 * the WP-bit has been tested.
733 #ifdef CONFIG_MEMORY_HOTPLUG
734 int arch_add_memory(int nid, u64 start, u64 size)
736 struct pglist_data *pgdata = NODE_DATA(nid);
737 struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
738 unsigned long start_pfn = start >> PAGE_SHIFT;
739 unsigned long nr_pages = size >> PAGE_SHIFT;
741 return __add_pages(zone, start_pfn, nr_pages);
746 struct kmem_cache *pmd_cache;
748 void __init pgtable_cache_init(void)
750 if (PTRS_PER_PMD > 1)
751 pmd_cache = kmem_cache_create("pmd",
752 PTRS_PER_PMD*sizeof(pmd_t),
753 PTRS_PER_PMD*sizeof(pmd_t),
759 * This function cannot be __init, since exceptions don't work in that
760 * section. Put this after the callers, so that it cannot be inlined.
762 static int noinline do_test_wp_bit(void)
767 __asm__ __volatile__(
772 ".section __ex_table,\"a\"\n"
776 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
785 #ifdef CONFIG_DEBUG_RODATA
787 void mark_rodata_ro(void)
789 unsigned long start = PFN_ALIGN(_text);
790 unsigned long size = PFN_ALIGN(_etext) - start;
792 #ifndef CONFIG_KPROBES
793 #ifdef CONFIG_HOTPLUG_CPU
794 /* It must still be possible to apply SMP alternatives. */
795 if (num_possible_cpus() <= 1)
798 change_page_attr(virt_to_page(start),
799 size >> PAGE_SHIFT, PAGE_KERNEL_RX);
800 printk("Write protecting the kernel text: %luk\n", size >> 10);
804 size = (unsigned long)__end_rodata - start;
805 change_page_attr(virt_to_page(start),
806 size >> PAGE_SHIFT, PAGE_KERNEL_RO);
807 printk("Write protecting the kernel read-only data: %luk\n",
811 * change_page_attr() requires a global_flush_tlb() call after it.
812 * We do this after the printk so that if something went wrong in the
813 * change, the printk gets out at least to give a better debug hint
814 * of who is the culprit.
820 void free_init_pages(char *what, unsigned long begin, unsigned long end)
824 for (addr = begin; addr < end; addr += PAGE_SIZE) {
825 ClearPageReserved(virt_to_page(addr));
826 init_page_count(virt_to_page(addr));
827 memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE);
831 printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
834 void free_initmem(void)
836 free_init_pages("unused kernel memory",
837 (unsigned long)(&__init_begin),
838 (unsigned long)(&__init_end));
841 #ifdef CONFIG_BLK_DEV_INITRD
842 void free_initrd_mem(unsigned long start, unsigned long end)
844 free_init_pages("initrd memory", start, end);