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>
51 unsigned int __VMALLOC_RESERVE = 128 << 20;
53 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
54 unsigned long highstart_pfn, highend_pfn;
56 static noinline int do_test_wp_bit(void);
59 * Creates a middle page table and puts a pointer to it in the
60 * given global directory entry. This only returns the gd entry
61 * in non-PAE compilation mode, since the middle layer is folded.
63 static pmd_t * __init one_md_table_init(pgd_t *pgd)
69 if (!(pgd_val(*pgd) & _PAGE_PRESENT)) {
70 pmd_table = (pmd_t *) alloc_bootmem_low_pages(PAGE_SIZE);
72 paravirt_alloc_pd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT);
73 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
74 pud = pud_offset(pgd, 0);
75 BUG_ON(pmd_table != pmd_offset(pud, 0));
78 pud = pud_offset(pgd, 0);
79 pmd_table = pmd_offset(pud, 0);
85 * Create a page table and place a pointer to it in a middle page
88 static pte_t * __init one_page_table_init(pmd_t *pmd)
90 if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
91 pte_t *page_table = NULL;
93 #ifdef CONFIG_DEBUG_PAGEALLOC
94 page_table = (pte_t *) alloc_bootmem_pages(PAGE_SIZE);
98 (pte_t *)alloc_bootmem_low_pages(PAGE_SIZE);
101 paravirt_alloc_pt(&init_mm, __pa(page_table) >> PAGE_SHIFT);
102 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
103 BUG_ON(page_table != pte_offset_kernel(pmd, 0));
106 return pte_offset_kernel(pmd, 0);
110 * This function initializes a certain range of kernel virtual memory
111 * with new bootmem page tables, everywhere page tables are missing in
114 * NOTE: The pagetables are allocated contiguous on the physical space
115 * so we can cache the place of the first one and move around without
116 * checking the pgd every time.
119 page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base)
121 int pgd_idx, pmd_idx;
127 pgd_idx = pgd_index(vaddr);
128 pmd_idx = pmd_index(vaddr);
129 pgd = pgd_base + pgd_idx;
131 for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
132 pmd = one_md_table_init(pgd);
133 pmd = pmd + pmd_index(vaddr);
134 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
136 one_page_table_init(pmd);
144 static inline int is_kernel_text(unsigned long addr)
146 if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end)
152 * This maps the physical memory to kernel virtual address space, a total
153 * of max_low_pfn pages, by creating page tables starting from address
156 static void __init kernel_physical_mapping_init(pgd_t *pgd_base)
158 int pgd_idx, pmd_idx, pte_ofs;
164 pgd_idx = pgd_index(PAGE_OFFSET);
165 pgd = pgd_base + pgd_idx;
168 for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
169 pmd = one_md_table_init(pgd);
170 if (pfn >= max_low_pfn)
174 pmd_idx < PTRS_PER_PMD && pfn < max_low_pfn;
176 unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET;
179 * Map with big pages if possible, otherwise
180 * create normal page tables:
184 pgprot_t prot = PAGE_KERNEL_LARGE;
186 addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
187 PAGE_OFFSET + PAGE_SIZE-1;
189 if (is_kernel_text(addr) ||
190 is_kernel_text(addr2))
191 prot = PAGE_KERNEL_LARGE_EXEC;
193 set_pmd(pmd, pfn_pmd(pfn, prot));
198 pte = one_page_table_init(pmd);
201 pte_ofs < PTRS_PER_PTE && pfn < max_low_pfn;
202 pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
203 pgprot_t prot = PAGE_KERNEL;
205 if (is_kernel_text(addr))
206 prot = PAGE_KERNEL_EXEC;
208 set_pte(pte, pfn_pte(pfn, prot));
214 static inline int page_kills_ppro(unsigned long pagenr)
216 if (pagenr >= 0x70000 && pagenr <= 0x7003F)
221 #ifdef CONFIG_HIGHMEM
225 static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr)
227 return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
228 vaddr), vaddr), vaddr);
231 static void __init kmap_init(void)
233 unsigned long kmap_vstart;
236 * Cache the first kmap pte:
238 kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
239 kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
241 kmap_prot = PAGE_KERNEL;
244 static void __init permanent_kmaps_init(pgd_t *pgd_base)
253 page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
255 pgd = swapper_pg_dir + pgd_index(vaddr);
256 pud = pud_offset(pgd, vaddr);
257 pmd = pmd_offset(pud, vaddr);
258 pte = pte_offset_kernel(pmd, vaddr);
259 pkmap_page_table = pte;
262 static void __meminit free_new_highpage(struct page *page)
264 init_page_count(page);
269 void __init add_one_highpage_init(struct page *page, int pfn, int bad_ppro)
271 if (page_is_ram(pfn) && !(bad_ppro && page_kills_ppro(pfn))) {
272 ClearPageReserved(page);
273 free_new_highpage(page);
275 SetPageReserved(page);
279 add_one_highpage_hotplug(struct page *page, unsigned long pfn)
281 free_new_highpage(page);
283 #ifdef CONFIG_FLATMEM
284 max_mapnr = max(pfn, max_mapnr);
292 * Not currently handling the NUMA case.
293 * Assuming single node and all memory that
294 * has been added dynamically that would be
295 * onlined here is in HIGHMEM.
297 void __meminit online_page(struct page *page)
299 ClearPageReserved(page);
300 add_one_highpage_hotplug(page, page_to_pfn(page));
304 static void __init set_highmem_pages_init(int bad_ppro)
308 for (pfn = highstart_pfn; pfn < highend_pfn; pfn++) {
310 * Holes under sparsemem might not have no mem_map[]:
313 add_one_highpage_init(pfn_to_page(pfn), pfn, bad_ppro);
315 totalram_pages += totalhigh_pages;
317 #endif /* !CONFIG_NUMA */
320 # define kmap_init() do { } while (0)
321 # define permanent_kmaps_init(pgd_base) do { } while (0)
322 # define set_highmem_pages_init(bad_ppro) do { } while (0)
323 #endif /* CONFIG_HIGHMEM */
325 pteval_t __PAGE_KERNEL = _PAGE_KERNEL;
326 EXPORT_SYMBOL(__PAGE_KERNEL);
328 pteval_t __PAGE_KERNEL_EXEC = _PAGE_KERNEL_EXEC;
330 void __init native_pagetable_setup_start(pgd_t *base)
332 unsigned long pfn, va;
339 * Remove any mappings which extend past the end of physical
340 * memory from the boot time page table:
342 for (pfn = max_low_pfn + 1; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
343 va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
344 pgd = base + pgd_index(va);
345 if (!pgd_present(*pgd))
348 pud = pud_offset(pgd, va);
349 pmd = pmd_offset(pud, va);
350 if (!pmd_present(*pmd))
353 pte = pte_offset_kernel(pmd, va);
354 if (!pte_present(*pte))
357 pte_clear(NULL, va, pte);
359 paravirt_alloc_pd(&init_mm, __pa(base) >> PAGE_SHIFT);
362 void __init native_pagetable_setup_done(pgd_t *base)
367 * Build a proper pagetable for the kernel mappings. Up until this
368 * point, we've been running on some set of pagetables constructed by
371 * If we're booting on native hardware, this will be a pagetable
372 * constructed in arch/x86/kernel/head_32.S. The root of the
373 * pagetable will be swapper_pg_dir.
375 * If we're booting paravirtualized under a hypervisor, then there are
376 * more options: we may already be running PAE, and the pagetable may
377 * or may not be based in swapper_pg_dir. In any case,
378 * paravirt_pagetable_setup_start() will set up swapper_pg_dir
379 * appropriately for the rest of the initialization to work.
381 * In general, pagetable_init() assumes that the pagetable may already
382 * be partially populated, and so it avoids stomping on any existing
385 static void __init pagetable_init(void)
387 pgd_t *pgd_base = swapper_pg_dir;
388 unsigned long vaddr, end;
390 paravirt_pagetable_setup_start(pgd_base);
392 /* Enable PSE if available */
394 set_in_cr4(X86_CR4_PSE);
396 /* Enable PGE if available */
398 set_in_cr4(X86_CR4_PGE);
399 __PAGE_KERNEL |= _PAGE_GLOBAL;
400 __PAGE_KERNEL_EXEC |= _PAGE_GLOBAL;
403 kernel_physical_mapping_init(pgd_base);
407 * Fixed mappings, only the page table structure has to be
408 * created - mappings will be set by set_fixmap():
410 early_ioremap_clear();
411 vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
412 end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
413 page_table_range_init(vaddr, end, pgd_base);
414 early_ioremap_reset();
416 permanent_kmaps_init(pgd_base);
418 paravirt_pagetable_setup_done(pgd_base);
421 #ifdef CONFIG_ACPI_SLEEP
423 * ACPI suspend needs this for resume, because things like the intel-agp
424 * driver might have split up a kernel 4MB mapping.
426 char swsusp_pg_dir[PAGE_SIZE]
427 __attribute__ ((aligned(PAGE_SIZE)));
429 static inline void save_pg_dir(void)
431 memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
433 #else /* !CONFIG_ACPI_SLEEP */
434 static inline void save_pg_dir(void)
437 #endif /* !CONFIG_ACPI_SLEEP */
439 void zap_low_mappings(void)
446 * Zap initial low-memory mappings.
448 * Note that "pgd_clear()" doesn't do it for
449 * us, because pgd_clear() is a no-op on i386.
451 for (i = 0; i < USER_PTRS_PER_PGD; i++) {
452 #ifdef CONFIG_X86_PAE
453 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
455 set_pgd(swapper_pg_dir+i, __pgd(0));
463 pteval_t __supported_pte_mask __read_mostly = ~_PAGE_NX;
464 EXPORT_SYMBOL_GPL(__supported_pte_mask);
466 #ifdef CONFIG_X86_PAE
468 static int disable_nx __initdata;
473 * Control non executable mappings.
478 static int __init noexec_setup(char *str)
480 if (!str || !strcmp(str, "on")) {
482 __supported_pte_mask |= _PAGE_NX;
486 if (!strcmp(str, "off")) {
488 __supported_pte_mask &= ~_PAGE_NX;
496 early_param("noexec", noexec_setup);
498 static void __init set_nx(void)
500 unsigned int v[4], l, h;
502 if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) {
503 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]);
505 if ((v[3] & (1 << 20)) && !disable_nx) {
506 rdmsr(MSR_EFER, l, h);
508 wrmsr(MSR_EFER, l, h);
510 __supported_pte_mask |= _PAGE_NX;
517 * paging_init() sets up the page tables - note that the first 8MB are
518 * already mapped by head.S.
520 * This routines also unmaps the page at virtual kernel address 0, so
521 * that we can trap those pesky NULL-reference errors in the kernel.
523 void __init paging_init(void)
525 #ifdef CONFIG_X86_PAE
528 printk(KERN_INFO "NX (Execute Disable) protection: active\n");
532 load_cr3(swapper_pg_dir);
540 * Test if the WP bit works in supervisor mode. It isn't supported on 386's
541 * and also on some strange 486's (NexGen etc.). All 586+'s are OK. This
542 * used to involve black magic jumps to work around some nasty CPU bugs,
543 * but fortunately the switch to using exceptions got rid of all that.
545 static void __init test_wp_bit(void)
548 "Checking if this processor honours the WP bit even in supervisor mode...");
550 /* Any page-aligned address will do, the test is non-destructive */
551 __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
552 boot_cpu_data.wp_works_ok = do_test_wp_bit();
553 clear_fixmap(FIX_WP_TEST);
555 if (!boot_cpu_data.wp_works_ok) {
556 printk(KERN_CONT "No.\n");
557 #ifdef CONFIG_X86_WP_WORKS_OK
559 "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
562 printk(KERN_CONT "Ok.\n");
566 static struct kcore_list kcore_mem, kcore_vmalloc;
568 void __init mem_init(void)
570 int codesize, reservedpages, datasize, initsize;
573 #ifdef CONFIG_FLATMEM
576 bad_ppro = ppro_with_ram_bug();
578 #ifdef CONFIG_HIGHMEM
579 /* check that fixmap and pkmap do not overlap */
580 if (PKMAP_BASE + LAST_PKMAP*PAGE_SIZE >= FIXADDR_START) {
582 "fixmap and kmap areas overlap - this will crash\n");
583 printk(KERN_ERR "pkstart: %lxh pkend: %lxh fixstart %lxh\n",
584 PKMAP_BASE, PKMAP_BASE + LAST_PKMAP*PAGE_SIZE,
589 /* this will put all low memory onto the freelists */
590 totalram_pages += free_all_bootmem();
593 for (tmp = 0; tmp < max_low_pfn; tmp++)
595 * Only count reserved RAM pages:
597 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
600 set_highmem_pages_init(bad_ppro);
602 codesize = (unsigned long) &_etext - (unsigned long) &_text;
603 datasize = (unsigned long) &_edata - (unsigned long) &_etext;
604 initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin;
606 kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
607 kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
608 VMALLOC_END-VMALLOC_START);
610 printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, "
611 "%dk reserved, %dk data, %dk init, %ldk highmem)\n",
612 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
613 num_physpages << (PAGE_SHIFT-10),
615 reservedpages << (PAGE_SHIFT-10),
618 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10))
621 #if 1 /* double-sanity-check paranoia */
622 printk(KERN_INFO "virtual kernel memory layout:\n"
623 " fixmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
624 #ifdef CONFIG_HIGHMEM
625 " pkmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
627 " vmalloc : 0x%08lx - 0x%08lx (%4ld MB)\n"
628 " lowmem : 0x%08lx - 0x%08lx (%4ld MB)\n"
629 " .init : 0x%08lx - 0x%08lx (%4ld kB)\n"
630 " .data : 0x%08lx - 0x%08lx (%4ld kB)\n"
631 " .text : 0x%08lx - 0x%08lx (%4ld kB)\n",
632 FIXADDR_START, FIXADDR_TOP,
633 (FIXADDR_TOP - FIXADDR_START) >> 10,
635 #ifdef CONFIG_HIGHMEM
636 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
637 (LAST_PKMAP*PAGE_SIZE) >> 10,
640 VMALLOC_START, VMALLOC_END,
641 (VMALLOC_END - VMALLOC_START) >> 20,
643 (unsigned long)__va(0), (unsigned long)high_memory,
644 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
646 (unsigned long)&__init_begin, (unsigned long)&__init_end,
647 ((unsigned long)&__init_end -
648 (unsigned long)&__init_begin) >> 10,
650 (unsigned long)&_etext, (unsigned long)&_edata,
651 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
653 (unsigned long)&_text, (unsigned long)&_etext,
654 ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
656 #ifdef CONFIG_HIGHMEM
657 BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START);
658 BUG_ON(VMALLOC_END > PKMAP_BASE);
660 BUG_ON(VMALLOC_START > VMALLOC_END);
661 BUG_ON((unsigned long)high_memory > VMALLOC_START);
662 #endif /* double-sanity-check paranoia */
664 if (boot_cpu_data.wp_works_ok < 0)
670 * Subtle. SMP is doing it's boot stuff late (because it has to
671 * fork idle threads) - but it also needs low mappings for the
672 * protected-mode entry to work. We zap these entries only after
673 * the WP-bit has been tested.
680 #ifdef CONFIG_MEMORY_HOTPLUG
681 int arch_add_memory(int nid, u64 start, u64 size)
683 struct pglist_data *pgdata = NODE_DATA(nid);
684 struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
685 unsigned long start_pfn = start >> PAGE_SHIFT;
686 unsigned long nr_pages = size >> PAGE_SHIFT;
688 return __add_pages(zone, start_pfn, nr_pages);
693 * This function cannot be __init, since exceptions don't work in that
694 * section. Put this after the callers, so that it cannot be inlined.
696 static noinline int do_test_wp_bit(void)
701 __asm__ __volatile__(
707 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
716 #ifdef CONFIG_DEBUG_RODATA
717 const int rodata_test_data = 0xC3;
718 EXPORT_SYMBOL_GPL(rodata_test_data);
720 void mark_rodata_ro(void)
722 unsigned long start = PFN_ALIGN(_text);
723 unsigned long size = PFN_ALIGN(_etext) - start;
725 #ifndef CONFIG_KPROBES
726 #ifdef CONFIG_HOTPLUG_CPU
727 /* It must still be possible to apply SMP alternatives. */
728 if (num_possible_cpus() <= 1)
731 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
732 printk(KERN_INFO "Write protecting the kernel text: %luk\n",
735 #ifdef CONFIG_CPA_DEBUG
736 printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n",
738 set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT);
740 printk(KERN_INFO "Testing CPA: write protecting again\n");
741 set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT);
746 size = (unsigned long)__end_rodata - start;
747 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
748 printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
752 #ifdef CONFIG_CPA_DEBUG
753 printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size);
754 set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
756 printk(KERN_INFO "Testing CPA: write protecting again\n");
757 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
762 void free_init_pages(char *what, unsigned long begin, unsigned long end)
764 #ifdef CONFIG_DEBUG_PAGEALLOC
766 * If debugging page accesses then do not free this memory but
767 * mark them not present - any buggy init-section access will
768 * create a kernel page fault:
770 printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n",
771 begin, PAGE_ALIGN(end));
772 set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
777 * We just marked the kernel text read only above, now that
778 * we are going to free part of that, we need to make that
781 set_memory_rw(begin, (end - begin) >> PAGE_SHIFT);
783 for (addr = begin; addr < end; addr += PAGE_SIZE) {
784 ClearPageReserved(virt_to_page(addr));
785 init_page_count(virt_to_page(addr));
786 memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE);
790 printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
794 void free_initmem(void)
796 free_init_pages("unused kernel memory",
797 (unsigned long)(&__init_begin),
798 (unsigned long)(&__init_end));
801 #ifdef CONFIG_BLK_DEV_INITRD
802 void free_initrd_mem(unsigned long start, unsigned long end)
804 free_init_pages("initrd memory", start, end);