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/pfn.h>
25 #include <linux/poison.h>
26 #include <linux/bootmem.h>
27 #include <linux/slab.h>
28 #include <linux/proc_fs.h>
29 #include <linux/memory_hotplug.h>
30 #include <linux/initrd.h>
31 #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>
48 #include <asm/setup.h>
49 #include <asm/cacheflush.h>
51 unsigned int __VMALLOC_RESERVE = 128 << 20;
53 unsigned long max_pfn_mapped;
55 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
56 unsigned long highstart_pfn, highend_pfn;
58 static noinline int do_test_wp_bit(void);
61 * Creates a middle page table and puts a pointer to it in the
62 * given global directory entry. This only returns the gd entry
63 * in non-PAE compilation mode, since the middle layer is folded.
65 static pmd_t * __init one_md_table_init(pgd_t *pgd)
71 if (!(pgd_val(*pgd) & _PAGE_PRESENT)) {
72 pmd_table = (pmd_t *) alloc_bootmem_low_pages(PAGE_SIZE);
74 paravirt_alloc_pmd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT);
75 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
76 pud = pud_offset(pgd, 0);
77 BUG_ON(pmd_table != pmd_offset(pud, 0));
80 pud = pud_offset(pgd, 0);
81 pmd_table = pmd_offset(pud, 0);
87 * Create a page table and place a pointer to it in a middle page
90 static pte_t * __init one_page_table_init(pmd_t *pmd)
92 if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
93 pte_t *page_table = NULL;
95 #ifdef CONFIG_DEBUG_PAGEALLOC
96 page_table = (pte_t *) alloc_bootmem_pages(PAGE_SIZE);
100 (pte_t *)alloc_bootmem_low_pages(PAGE_SIZE);
103 paravirt_alloc_pte(&init_mm, __pa(page_table) >> PAGE_SHIFT);
104 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
105 BUG_ON(page_table != pte_offset_kernel(pmd, 0));
108 return pte_offset_kernel(pmd, 0);
112 * This function initializes a certain range of kernel virtual memory
113 * with new bootmem page tables, everywhere page tables are missing in
116 * NOTE: The pagetables are allocated contiguous on the physical space
117 * so we can cache the place of the first one and move around without
118 * checking the pgd every time.
121 page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base)
123 int pgd_idx, pmd_idx;
129 pgd_idx = pgd_index(vaddr);
130 pmd_idx = pmd_index(vaddr);
131 pgd = pgd_base + pgd_idx;
133 for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
134 pmd = one_md_table_init(pgd);
135 pmd = pmd + pmd_index(vaddr);
136 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
138 one_page_table_init(pmd);
146 static inline int is_kernel_text(unsigned long addr)
148 if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end)
154 * This maps the physical memory to kernel virtual address space, a total
155 * of max_low_pfn pages, by creating page tables starting from address
158 static void __init kernel_physical_mapping_init(pgd_t *pgd_base)
160 int pgd_idx, pmd_idx, pte_ofs;
165 unsigned pages_2m = 0, pages_4k = 0;
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;
202 set_pmd(pmd, pfn_pmd(pfn, prot));
205 max_pfn_mapped = pfn;
208 pte = one_page_table_init(pmd);
211 pte_ofs < PTRS_PER_PTE && pfn < max_low_pfn;
212 pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
213 pgprot_t prot = PAGE_KERNEL;
215 if (is_kernel_text(addr))
216 prot = PAGE_KERNEL_EXEC;
219 set_pte(pte, pfn_pte(pfn, prot));
221 max_pfn_mapped = pfn;
224 update_page_count(PG_LEVEL_2M, pages_2m);
225 update_page_count(PG_LEVEL_4K, pages_4k);
228 static inline int page_kills_ppro(unsigned long pagenr)
230 if (pagenr >= 0x70000 && pagenr <= 0x7003F)
236 * devmem_is_allowed() checks to see if /dev/mem access to a certain address
237 * is valid. The argument is a physical page number.
240 * On x86, access has to be given to the first megabyte of ram because that area
241 * contains bios code and data regions used by X and dosemu and similar apps.
242 * Access has to be given to non-kernel-ram areas as well, these contain the PCI
243 * mmio resources as well as potential bios/acpi data regions.
245 int devmem_is_allowed(unsigned long pagenr)
249 if (!page_is_ram(pagenr))
254 #ifdef CONFIG_HIGHMEM
258 static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr)
260 return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
261 vaddr), vaddr), vaddr);
264 static void __init kmap_init(void)
266 unsigned long kmap_vstart;
269 * Cache the first kmap pte:
271 kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
272 kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
274 kmap_prot = PAGE_KERNEL;
277 static void __init permanent_kmaps_init(pgd_t *pgd_base)
286 page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
288 pgd = swapper_pg_dir + pgd_index(vaddr);
289 pud = pud_offset(pgd, vaddr);
290 pmd = pmd_offset(pud, vaddr);
291 pte = pte_offset_kernel(pmd, vaddr);
292 pkmap_page_table = pte;
295 void __init add_one_highpage_init(struct page *page, int pfn, int bad_ppro)
297 if (page_is_ram(pfn) && !(bad_ppro && page_kills_ppro(pfn))) {
298 ClearPageReserved(page);
299 init_page_count(page);
303 SetPageReserved(page);
307 static void __init set_highmem_pages_init(int bad_ppro)
311 for (pfn = highstart_pfn; pfn < highend_pfn; pfn++) {
313 * Holes under sparsemem might not have no mem_map[]:
316 add_one_highpage_init(pfn_to_page(pfn), pfn, bad_ppro);
318 totalram_pages += totalhigh_pages;
320 #endif /* !CONFIG_NUMA */
323 # define kmap_init() do { } while (0)
324 # define permanent_kmaps_init(pgd_base) do { } while (0)
325 # define set_highmem_pages_init(bad_ppro) do { } while (0)
326 #endif /* CONFIG_HIGHMEM */
328 pteval_t __PAGE_KERNEL = _PAGE_KERNEL;
329 EXPORT_SYMBOL(__PAGE_KERNEL);
331 pteval_t __PAGE_KERNEL_EXEC = _PAGE_KERNEL_EXEC;
333 void __init native_pagetable_setup_start(pgd_t *base)
335 unsigned long pfn, va;
342 * Remove any mappings which extend past the end of physical
343 * memory from the boot time page table:
345 for (pfn = max_low_pfn + 1; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
346 va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
347 pgd = base + pgd_index(va);
348 if (!pgd_present(*pgd))
351 pud = pud_offset(pgd, va);
352 pmd = pmd_offset(pud, va);
353 if (!pmd_present(*pmd))
356 pte = pte_offset_kernel(pmd, va);
357 if (!pte_present(*pte))
360 pte_clear(NULL, va, pte);
362 paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT);
365 void __init native_pagetable_setup_done(pgd_t *base)
370 * Build a proper pagetable for the kernel mappings. Up until this
371 * point, we've been running on some set of pagetables constructed by
374 * If we're booting on native hardware, this will be a pagetable
375 * constructed in arch/x86/kernel/head_32.S. The root of the
376 * pagetable will be swapper_pg_dir.
378 * If we're booting paravirtualized under a hypervisor, then there are
379 * more options: we may already be running PAE, and the pagetable may
380 * or may not be based in swapper_pg_dir. In any case,
381 * paravirt_pagetable_setup_start() will set up swapper_pg_dir
382 * appropriately for the rest of the initialization to work.
384 * In general, pagetable_init() assumes that the pagetable may already
385 * be partially populated, and so it avoids stomping on any existing
388 static void __init pagetable_init(void)
390 pgd_t *pgd_base = swapper_pg_dir;
391 unsigned long vaddr, end;
393 paravirt_pagetable_setup_start(pgd_base);
395 /* Enable PSE if available */
397 set_in_cr4(X86_CR4_PSE);
399 /* Enable PGE if available */
401 set_in_cr4(X86_CR4_PGE);
402 __PAGE_KERNEL |= _PAGE_GLOBAL;
403 __PAGE_KERNEL_EXEC |= _PAGE_GLOBAL;
406 kernel_physical_mapping_init(pgd_base);
410 * Fixed mappings, only the page table structure has to be
411 * created - mappings will be set by set_fixmap():
413 early_ioremap_clear();
414 vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
415 end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
416 page_table_range_init(vaddr, end, pgd_base);
417 early_ioremap_reset();
419 permanent_kmaps_init(pgd_base);
421 paravirt_pagetable_setup_done(pgd_base);
424 #ifdef CONFIG_ACPI_SLEEP
426 * ACPI suspend needs this for resume, because things like the intel-agp
427 * driver might have split up a kernel 4MB mapping.
429 char swsusp_pg_dir[PAGE_SIZE]
430 __attribute__ ((aligned(PAGE_SIZE)));
432 static inline void save_pg_dir(void)
434 memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
436 #else /* !CONFIG_ACPI_SLEEP */
437 static inline void save_pg_dir(void)
440 #endif /* !CONFIG_ACPI_SLEEP */
442 void zap_low_mappings(void)
447 * Zap initial low-memory mappings.
449 * Note that "pgd_clear()" doesn't do it for
450 * us, because pgd_clear() is a no-op on i386.
452 for (i = 0; i < KERNEL_PGD_BOUNDARY; i++) {
453 #ifdef CONFIG_X86_PAE
454 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
456 set_pgd(swapper_pg_dir+i, __pgd(0));
464 pteval_t __supported_pte_mask __read_mostly = ~_PAGE_NX;
465 EXPORT_SYMBOL_GPL(__supported_pte_mask);
467 #ifdef CONFIG_X86_PAE
469 static int disable_nx __initdata;
474 * Control non executable mappings.
479 static int __init noexec_setup(char *str)
481 if (!str || !strcmp(str, "on")) {
483 __supported_pte_mask |= _PAGE_NX;
487 if (!strcmp(str, "off")) {
489 __supported_pte_mask &= ~_PAGE_NX;
497 early_param("noexec", noexec_setup);
499 static void __init set_nx(void)
501 unsigned int v[4], l, h;
503 if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) {
504 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]);
506 if ((v[3] & (1 << 20)) && !disable_nx) {
507 rdmsr(MSR_EFER, l, h);
509 wrmsr(MSR_EFER, l, h);
511 __supported_pte_mask |= _PAGE_NX;
518 * paging_init() sets up the page tables - note that the first 8MB are
519 * already mapped by head.S.
521 * This routines also unmaps the page at virtual kernel address 0, so
522 * that we can trap those pesky NULL-reference errors in the kernel.
524 void __init paging_init(void)
526 #ifdef CONFIG_X86_PAE
529 printk(KERN_INFO "NX (Execute Disable) protection: active\n");
533 load_cr3(swapper_pg_dir);
541 * Test if the WP bit works in supervisor mode. It isn't supported on 386's
542 * and also on some strange 486's. All 586+'s are OK. This used to involve
543 * black magic jumps to work around some nasty CPU bugs, but fortunately the
544 * switch to using exceptions got rid of all that.
546 static void __init test_wp_bit(void)
549 "Checking if this processor honours the WP bit even in supervisor mode...");
551 /* Any page-aligned address will do, the test is non-destructive */
552 __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
553 boot_cpu_data.wp_works_ok = do_test_wp_bit();
554 clear_fixmap(FIX_WP_TEST);
556 if (!boot_cpu_data.wp_works_ok) {
557 printk(KERN_CONT "No.\n");
558 #ifdef CONFIG_X86_WP_WORKS_OK
560 "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
563 printk(KERN_CONT "Ok.\n");
567 static struct kcore_list kcore_mem, kcore_vmalloc;
569 void __init mem_init(void)
571 int codesize, reservedpages, datasize, initsize;
574 #ifdef CONFIG_FLATMEM
577 bad_ppro = ppro_with_ram_bug();
579 /* this will put all low memory onto the freelists */
580 totalram_pages += free_all_bootmem();
583 for (tmp = 0; tmp < max_low_pfn; tmp++)
585 * Only count reserved RAM pages:
587 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
590 set_highmem_pages_init(bad_ppro);
592 codesize = (unsigned long) &_etext - (unsigned long) &_text;
593 datasize = (unsigned long) &_edata - (unsigned long) &_etext;
594 initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin;
596 kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
597 kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
598 VMALLOC_END-VMALLOC_START);
600 printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, "
601 "%dk reserved, %dk data, %dk init, %ldk highmem)\n",
602 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
603 num_physpages << (PAGE_SHIFT-10),
605 reservedpages << (PAGE_SHIFT-10),
608 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10))
611 printk(KERN_INFO "virtual kernel memory layout:\n"
612 " fixmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
613 #ifdef CONFIG_HIGHMEM
614 " pkmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
616 " vmalloc : 0x%08lx - 0x%08lx (%4ld MB)\n"
617 " lowmem : 0x%08lx - 0x%08lx (%4ld MB)\n"
618 " .init : 0x%08lx - 0x%08lx (%4ld kB)\n"
619 " .data : 0x%08lx - 0x%08lx (%4ld kB)\n"
620 " .text : 0x%08lx - 0x%08lx (%4ld kB)\n",
621 FIXADDR_START, FIXADDR_TOP,
622 (FIXADDR_TOP - FIXADDR_START) >> 10,
624 #ifdef CONFIG_HIGHMEM
625 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
626 (LAST_PKMAP*PAGE_SIZE) >> 10,
629 VMALLOC_START, VMALLOC_END,
630 (VMALLOC_END - VMALLOC_START) >> 20,
632 (unsigned long)__va(0), (unsigned long)high_memory,
633 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
635 (unsigned long)&__init_begin, (unsigned long)&__init_end,
636 ((unsigned long)&__init_end -
637 (unsigned long)&__init_begin) >> 10,
639 (unsigned long)&_etext, (unsigned long)&_edata,
640 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
642 (unsigned long)&_text, (unsigned long)&_etext,
643 ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
645 #ifdef CONFIG_HIGHMEM
646 BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START);
647 BUG_ON(VMALLOC_END > PKMAP_BASE);
649 BUG_ON(VMALLOC_START > VMALLOC_END);
650 BUG_ON((unsigned long)high_memory > VMALLOC_START);
652 if (boot_cpu_data.wp_works_ok < 0)
660 #ifdef CONFIG_MEMORY_HOTPLUG
661 int arch_add_memory(int nid, u64 start, u64 size)
663 struct pglist_data *pgdata = NODE_DATA(nid);
664 struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
665 unsigned long start_pfn = start >> PAGE_SHIFT;
666 unsigned long nr_pages = size >> PAGE_SHIFT;
668 return __add_pages(zone, start_pfn, nr_pages);
673 * This function cannot be __init, since exceptions don't work in that
674 * section. Put this after the callers, so that it cannot be inlined.
676 static noinline int do_test_wp_bit(void)
681 __asm__ __volatile__(
687 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
696 #ifdef CONFIG_DEBUG_RODATA
697 const int rodata_test_data = 0xC3;
698 EXPORT_SYMBOL_GPL(rodata_test_data);
700 void mark_rodata_ro(void)
702 unsigned long start = PFN_ALIGN(_text);
703 unsigned long size = PFN_ALIGN(_etext) - start;
705 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
706 printk(KERN_INFO "Write protecting the kernel text: %luk\n",
709 #ifdef CONFIG_CPA_DEBUG
710 printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n",
712 set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT);
714 printk(KERN_INFO "Testing CPA: write protecting again\n");
715 set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT);
718 size = (unsigned long)__end_rodata - start;
719 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
720 printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
724 #ifdef CONFIG_CPA_DEBUG
725 printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size);
726 set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
728 printk(KERN_INFO "Testing CPA: write protecting again\n");
729 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
734 void free_init_pages(char *what, unsigned long begin, unsigned long end)
736 #ifdef CONFIG_DEBUG_PAGEALLOC
738 * If debugging page accesses then do not free this memory but
739 * mark them not present - any buggy init-section access will
740 * create a kernel page fault:
742 printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n",
743 begin, PAGE_ALIGN(end));
744 set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
749 * We just marked the kernel text read only above, now that
750 * we are going to free part of that, we need to make that
753 set_memory_rw(begin, (end - begin) >> PAGE_SHIFT);
755 for (addr = begin; addr < end; addr += PAGE_SIZE) {
756 ClearPageReserved(virt_to_page(addr));
757 init_page_count(virt_to_page(addr));
758 memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE);
762 printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
766 void free_initmem(void)
768 free_init_pages("unused kernel memory",
769 (unsigned long)(&__init_begin),
770 (unsigned long)(&__init_end));
773 #ifdef CONFIG_BLK_DEV_INITRD
774 void free_initrd_mem(unsigned long start, unsigned long end)
776 free_init_pages("initrd memory", start, end);