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/poison.h>
26 #include <linux/bootmem.h>
27 #include <linux/slab.h>
28 #include <linux/proc_fs.h>
29 #include <linux/efi.h>
30 #include <linux/memory_hotplug.h>
31 #include <linux/initrd.h>
32 #include <linux/cpumask.h>
33 #include <linux/dma-mapping.h>
34 #include <linux/scatterlist.h>
36 #include <asm/processor.h>
37 #include <asm/system.h>
38 #include <asm/uaccess.h>
39 #include <asm/pgtable.h>
41 #include <asm/fixmap.h>
45 #include <asm/tlbflush.h>
46 #include <asm/sections.h>
47 #include <asm/hypervisor.h>
48 #include <asm/swiotlb.h>
50 extern unsigned long *contiguous_bitmap;
52 unsigned int __VMALLOC_RESERVE = 128 << 20;
54 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
55 unsigned long highstart_pfn, highend_pfn;
57 static int noinline do_test_wp_bit(void);
60 * Creates a middle page table and puts a pointer to it in the
61 * given global directory entry. This only returns the gd entry
62 * in non-PAE compilation mode, since the middle layer is folded.
64 static pmd_t * __init one_md_table_init(pgd_t *pgd)
70 pmd_table = (pmd_t *) alloc_bootmem_low_pages(PAGE_SIZE);
71 make_lowmem_page_readonly(pmd_table, XENFEAT_writable_page_tables);
72 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
73 pud = pud_offset(pgd, 0);
74 if (pmd_table != pmd_offset(pud, 0))
77 pud = pud_offset(pgd, 0);
78 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)
91 pte_t *page_table = (pte_t *) alloc_bootmem_low_pages(PAGE_SIZE);
92 make_lowmem_page_readonly(page_table,
93 XENFEAT_writable_page_tables);
94 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
95 if (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)
120 int pgd_idx, pmd_idx;
124 pgd_idx = pgd_index(vaddr);
125 pmd_idx = pmd_index(vaddr);
126 pgd = pgd_base + pgd_idx;
128 for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
130 one_md_table_init(pgd);
131 pud = pud_offset(pgd, vaddr);
132 pmd = pmd_offset(pud, vaddr);
133 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end); pmd++, pmd_idx++) {
134 if (vaddr < HYPERVISOR_VIRT_START && pmd_none(*pmd))
135 one_page_table_init(pmd);
143 static inline int is_kernel_text(unsigned long addr)
145 if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end)
151 * This maps the physical memory to kernel virtual address space, a total
152 * of max_low_pfn pages, by creating page tables starting from address
155 static void __init kernel_physical_mapping_init(pgd_t *pgd_base)
161 int pgd_idx, pmd_idx, pte_ofs;
163 unsigned long max_ram_pfn = xen_start_info->nr_pages;
164 if (max_ram_pfn > max_low_pfn)
165 max_ram_pfn = max_low_pfn;
167 pgd_idx = pgd_index(PAGE_OFFSET);
168 pgd = pgd_base + pgd_idx;
170 pmd_idx = pmd_index(PAGE_OFFSET);
171 pte_ofs = pte_index(PAGE_OFFSET);
173 for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
176 * Native linux hasn't PAE-paging enabled yet at this
177 * point. When running as xen domain we are in PAE
178 * mode already, thus we can't simply hook a empty
179 * pmd. That would kill the mappings we are currently
182 pmd = pmd_offset(pud_offset(pgd, PAGE_OFFSET), PAGE_OFFSET);
184 pmd = one_md_table_init(pgd);
186 if (pfn >= max_low_pfn)
189 for (; pmd_idx < PTRS_PER_PMD && pfn < max_low_pfn; pmd++, pmd_idx++) {
190 unsigned int address = pfn * PAGE_SIZE + PAGE_OFFSET;
191 if (address >= HYPERVISOR_VIRT_START)
194 /* Map with big pages if possible, otherwise create normal page tables. */
196 unsigned int address2 = (pfn + PTRS_PER_PTE - 1) * PAGE_SIZE + PAGE_OFFSET + PAGE_SIZE-1;
198 if (is_kernel_text(address) || is_kernel_text(address2))
199 set_pmd(pmd, pfn_pmd(pfn, PAGE_KERNEL_LARGE_EXEC));
201 set_pmd(pmd, pfn_pmd(pfn, PAGE_KERNEL_LARGE));
204 pte = one_page_table_init(pmd);
207 for (; pte_ofs < PTRS_PER_PTE && pfn < max_low_pfn; pte++, pfn++, pte_ofs++) {
208 /* XEN: Only map initial RAM allocation. */
209 if ((pfn >= max_ram_pfn) || pte_present(*pte))
211 if (is_kernel_text(address))
212 set_pte(pte, pfn_pte(pfn, PAGE_KERNEL_EXEC));
214 set_pte(pte, pfn_pte(pfn, PAGE_KERNEL));
225 static inline int page_kills_ppro(unsigned long pagenr)
227 if (pagenr >= 0x70000 && pagenr <= 0x7003F)
234 #define page_kills_ppro(p) 0
238 extern int is_available_memory(efi_memory_desc_t *);
240 int page_is_ram(unsigned long pagenr)
243 unsigned long addr, end;
246 efi_memory_desc_t *md;
249 for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
251 if (!is_available_memory(md))
253 addr = (md->phys_addr+PAGE_SIZE-1) >> PAGE_SHIFT;
254 end = (md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT)) >> PAGE_SHIFT;
256 if ((pagenr >= addr) && (pagenr < end))
262 for (i = 0; i < e820.nr_map; i++) {
264 if (e820.map[i].type != E820_RAM) /* not usable memory */
267 * !!!FIXME!!! Some BIOSen report areas as RAM that
268 * are not. Notably the 640->1Mb area. We need a sanity
271 addr = (e820.map[i].addr+PAGE_SIZE-1) >> PAGE_SHIFT;
272 end = (e820.map[i].addr+e820.map[i].size) >> PAGE_SHIFT;
273 if ((pagenr >= addr) && (pagenr < end))
278 EXPORT_SYMBOL_GPL(page_is_ram);
281 * devmem_is_allowed() checks to see if /dev/mem access to a certain address is
282 * valid. The argument is a physical page number.
285 * On x86, access has to be given to the first megabyte of ram because that area
286 * contains bios code and data regions used by X and dosemu and similar apps.
287 * Access has to be given to non-kernel-ram areas as well, these contain the PCI
288 * mmio resources as well as potential bios/acpi data regions.
290 int devmem_is_allowed(unsigned long pagenr)
294 if (!page_is_ram(pagenr))
299 #ifdef CONFIG_HIGHMEM
303 #define kmap_get_fixmap_pte(vaddr) \
304 pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr), vaddr), (vaddr)), (vaddr))
306 static void __init kmap_init(void)
308 unsigned long kmap_vstart;
310 /* cache the first kmap pte */
311 kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
312 kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
314 kmap_prot = PAGE_KERNEL;
317 static void __init permanent_kmaps_init(pgd_t *pgd_base)
326 page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
328 pgd = swapper_pg_dir + pgd_index(vaddr);
329 pud = pud_offset(pgd, vaddr);
330 pmd = pmd_offset(pud, vaddr);
331 pte = pte_offset_kernel(pmd, vaddr);
332 pkmap_page_table = pte;
335 static void __meminit free_new_highpage(struct page *page, int pfn)
337 init_page_count(page);
338 if (pfn < xen_start_info->nr_pages)
343 void __init add_one_highpage_init(struct page *page, int pfn, int bad_ppro)
345 if (page_is_ram(pfn) && !(bad_ppro && page_kills_ppro(pfn))) {
346 ClearPageReserved(page);
347 free_new_highpage(page, pfn);
349 SetPageReserved(page);
352 static int add_one_highpage_hotplug(struct page *page, unsigned long pfn)
354 free_new_highpage(page, pfn);
356 #ifdef CONFIG_FLATMEM
357 max_mapnr = max(pfn, max_mapnr);
364 * Not currently handling the NUMA case.
365 * Assuming single node and all memory that
366 * has been added dynamically that would be
367 * onlined here is in HIGHMEM
369 void online_page(struct page *page)
371 ClearPageReserved(page);
372 add_one_highpage_hotplug(page, page_to_pfn(page));
377 extern void set_highmem_pages_init(int);
379 static void __init set_highmem_pages_init(int bad_ppro)
382 for (pfn = highstart_pfn; pfn < highend_pfn; pfn++)
383 add_one_highpage_init(pfn_to_page(pfn), pfn, bad_ppro);
384 totalram_pages += totalhigh_pages;
386 #endif /* CONFIG_FLATMEM */
389 #define kmap_init() do { } while (0)
390 #define permanent_kmaps_init(pgd_base) do { } while (0)
391 #define set_highmem_pages_init(bad_ppro) do { } while (0)
392 #endif /* CONFIG_HIGHMEM */
394 unsigned long long __PAGE_KERNEL = _PAGE_KERNEL;
395 EXPORT_SYMBOL(__PAGE_KERNEL);
396 unsigned long long __PAGE_KERNEL_EXEC = _PAGE_KERNEL_EXEC;
399 extern void __init remap_numa_kva(void);
401 #define remap_numa_kva() do {} while (0)
404 pgd_t *swapper_pg_dir;
406 static void __init pagetable_init (void)
409 pgd_t *pgd_base = (pgd_t *)xen_start_info->pt_base;
411 swapper_pg_dir = pgd_base;
412 init_mm.pgd = pgd_base;
414 /* Enable PSE if available */
416 set_in_cr4(X86_CR4_PSE);
419 /* Enable PGE if available */
421 set_in_cr4(X86_CR4_PGE);
422 __PAGE_KERNEL |= _PAGE_GLOBAL;
423 __PAGE_KERNEL_EXEC |= _PAGE_GLOBAL;
426 kernel_physical_mapping_init(pgd_base);
430 * Fixed mappings, only the page table structure has to be
431 * created - mappings will be set by set_fixmap():
433 vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
434 page_table_range_init(vaddr, 0, pgd_base);
436 permanent_kmaps_init(pgd_base);
439 #if defined(CONFIG_SOFTWARE_SUSPEND) || defined(CONFIG_ACPI_SLEEP)
441 * Swap suspend & friends need this for resume because things like the intel-agp
442 * driver might have split up a kernel 4MB mapping.
444 char __nosavedata swsusp_pg_dir[PAGE_SIZE]
445 __attribute__ ((aligned (PAGE_SIZE)));
447 static inline void save_pg_dir(void)
449 memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
452 static inline void save_pg_dir(void)
457 void zap_low_mappings (void)
464 * Zap initial low-memory mappings.
466 * Note that "pgd_clear()" doesn't do it for
467 * us, because pgd_clear() is a no-op on i386.
469 for (i = 0; i < USER_PTRS_PER_PGD; i++)
470 #if defined(CONFIG_X86_PAE) && !defined(CONFIG_XEN)
471 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
473 set_pgd(swapper_pg_dir+i, __pgd(0));
478 static int disable_nx __initdata = 0;
479 u64 __supported_pte_mask __read_mostly = ~_PAGE_NX;
480 EXPORT_SYMBOL(__supported_pte_mask);
485 * Control non executable mappings.
488 * off Disable (disables exec-shield too)
490 void __init noexec_setup(const char *str)
492 if (!strncmp(str, "on",2) && cpu_has_nx) {
493 __supported_pte_mask |= _PAGE_NX;
495 } else if (!strncmp(str,"off",3)) {
497 __supported_pte_mask &= ~_PAGE_NX;
503 #ifdef CONFIG_X86_PAE
505 static void __init set_nx(void)
507 unsigned int v[4], l, h;
509 if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) {
510 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]);
511 if ((v[3] & (1 << 20)) && !disable_nx) {
512 rdmsr(MSR_EFER, l, h);
514 wrmsr(MSR_EFER, l, h);
516 __supported_pte_mask |= _PAGE_NX;
522 * Enables/disables executability of a given kernel page and
523 * returns the previous setting.
525 int __init set_kernel_exec(unsigned long vaddr, int enable)
533 pte = lookup_address(vaddr);
536 if (!pte_exec_kernel(*pte))
540 pte->pte_high &= ~(1 << (_PAGE_BIT_NX - 32));
542 pte->pte_high |= 1 << (_PAGE_BIT_NX - 32);
551 * paging_init() sets up the page tables - note that the first 8MB are
552 * already mapped by head.S.
554 * This routines also unmaps the page at virtual kernel address 0, so
555 * that we can trap those pesky NULL-reference errors in the kernel.
557 void __init paging_init(void)
561 #ifdef CONFIG_X86_PAE
564 printk("NX (Execute Disable) protection: active\n");
568 printk("Using x86 segment limits to approximate NX protection\n");
572 #if defined(CONFIG_X86_PAE) && !defined(CONFIG_XEN)
574 * We will bail out later - printk doesn't work right now so
575 * the user would just see a hanging kernel.
576 * when running as xen domain we are already in PAE mode at
580 set_in_cr4(X86_CR4_PAE);
586 /* Switch to the real shared_info page, and clear the
588 set_fixmap(FIX_SHARED_INFO, xen_start_info->shared_info);
589 HYPERVISOR_shared_info = (shared_info_t *)fix_to_virt(FIX_SHARED_INFO);
590 memset(empty_zero_page, 0, sizeof(empty_zero_page));
592 /* Setup mapping of lower 1st MB */
593 for (i = 0; i < NR_FIX_ISAMAPS; i++)
594 if (is_initial_xendomain())
595 set_fixmap(FIX_ISAMAP_BEGIN - i, i * PAGE_SIZE);
597 __set_fixmap(FIX_ISAMAP_BEGIN - i,
598 virt_to_machine(empty_zero_page),
603 * Test if the WP bit works in supervisor mode. It isn't supported on 386's
604 * and also on some strange 486's (NexGen etc.). All 586+'s are OK. This
605 * used to involve black magic jumps to work around some nasty CPU bugs,
606 * but fortunately the switch to using exceptions got rid of all that.
609 static void __init test_wp_bit(void)
611 printk("Checking if this processor honours the WP bit even in supervisor mode... ");
613 /* Any page-aligned address will do, the test is non-destructive */
614 __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
615 boot_cpu_data.wp_works_ok = do_test_wp_bit();
616 clear_fixmap(FIX_WP_TEST);
618 if (!boot_cpu_data.wp_works_ok) {
620 #ifdef CONFIG_X86_WP_WORKS_OK
621 panic("This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
628 static void __init set_max_mapnr_init(void)
630 #ifdef CONFIG_HIGHMEM
631 num_physpages = highend_pfn;
633 num_physpages = max_low_pfn;
635 #ifdef CONFIG_FLATMEM
636 max_mapnr = num_physpages;
640 static struct kcore_list kcore_mem, kcore_vmalloc;
642 void __init mem_init(void)
644 extern int ppro_with_ram_bug(void);
645 int codesize, reservedpages, datasize, initsize;
650 contiguous_bitmap = alloc_bootmem_low_pages(
651 (max_low_pfn + 2*BITS_PER_LONG) >> 3);
652 BUG_ON(!contiguous_bitmap);
653 memset(contiguous_bitmap, 0, (max_low_pfn + 2*BITS_PER_LONG) >> 3);
655 #if defined(CONFIG_SWIOTLB)
659 #ifdef CONFIG_FLATMEM
664 bad_ppro = ppro_with_ram_bug();
666 #ifdef CONFIG_HIGHMEM
667 /* check that fixmap and pkmap do not overlap */
668 if (PKMAP_BASE+LAST_PKMAP*PAGE_SIZE >= FIXADDR_START) {
669 printk(KERN_ERR "fixmap and kmap areas overlap - this will crash\n");
670 printk(KERN_ERR "pkstart: %lxh pkend: %lxh fixstart %lxh\n",
671 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE, FIXADDR_START);
676 set_max_mapnr_init();
678 #ifdef CONFIG_HIGHMEM
679 high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
681 high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
683 printk("vmalloc area: %lx-%lx, maxmem %lx\n",
684 VMALLOC_START,VMALLOC_END,MAXMEM);
685 BUG_ON(VMALLOC_START > VMALLOC_END);
687 /* this will put all low memory onto the freelists */
688 totalram_pages += free_all_bootmem();
689 /* XEN: init and count low-mem pages outside initial allocation. */
690 for (pfn = xen_start_info->nr_pages; pfn < max_low_pfn; pfn++) {
691 ClearPageReserved(&mem_map[pfn]);
692 init_page_count(&mem_map[pfn]);
697 for (tmp = 0; tmp < max_low_pfn; tmp++)
699 * Only count reserved RAM pages
701 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
704 set_highmem_pages_init(bad_ppro);
706 codesize = (unsigned long) &_etext - (unsigned long) &_text;
707 datasize = (unsigned long) &_edata - (unsigned long) &_etext;
708 initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin;
710 kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
711 kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
712 VMALLOC_END-VMALLOC_START);
714 printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, %dk reserved, %dk data, %dk init, %ldk highmem)\n",
715 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
716 num_physpages << (PAGE_SHIFT-10),
718 reservedpages << (PAGE_SHIFT-10),
721 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10))
724 #ifdef CONFIG_X86_PAE
726 panic("cannot execute a PAE-enabled kernel on a PAE-less CPU!");
728 if (boot_cpu_data.wp_works_ok < 0)
732 * Subtle. SMP is doing it's boot stuff late (because it has to
733 * fork idle threads) - but it also needs low mappings for the
734 * protected-mode entry to work. We zap these entries only after
735 * the WP-bit has been tested.
741 set_bit(PG_pinned, &virt_to_page(init_mm.pgd)->flags);
745 * this is for the non-NUMA, single node SMP system case.
746 * Specifically, in the case of x86, we will always add
747 * memory to the highmem for now.
749 #ifdef CONFIG_MEMORY_HOTPLUG
750 #ifndef CONFIG_NEED_MULTIPLE_NODES
751 int arch_add_memory(u64 start, u64 size)
753 struct pglist_data *pgdata = &contig_page_data;
754 struct zone *zone = pgdata->node_zones + MAX_NR_ZONES-1;
755 unsigned long start_pfn = start >> PAGE_SHIFT;
756 unsigned long nr_pages = size >> PAGE_SHIFT;
758 return __add_pages(zone, start_pfn, nr_pages);
761 int remove_memory(u64 start, u64 size)
768 kmem_cache_t *pgd_cache;
769 kmem_cache_t *pmd_cache;
771 void __init pgtable_cache_init(void)
773 if (PTRS_PER_PMD > 1) {
774 pmd_cache = kmem_cache_create("pmd",
775 PTRS_PER_PMD*sizeof(pmd_t),
776 PTRS_PER_PMD*sizeof(pmd_t),
781 panic("pgtable_cache_init(): cannot create pmd cache");
783 pgd_cache = kmem_cache_create("pgd",
785 PTRS_PER_PGD*sizeof(pgd_t),
786 PTRS_PER_PGD*sizeof(pgd_t),
793 PTRS_PER_PMD == 1 ? pgd_dtor : NULL);
795 panic("pgtable_cache_init(): Cannot create pgd cache");
799 * This function cannot be __init, since exceptions don't work in that
800 * section. Put this after the callers, so that it cannot be inlined.
802 static int noinline do_test_wp_bit(void)
807 __asm__ __volatile__(
812 ".section __ex_table,\"a\"\n"
816 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
825 #ifdef CONFIG_DEBUG_RODATA
827 void mark_rodata_ro(void)
829 unsigned long addr = (unsigned long)__start_rodata;
831 for (; addr < (unsigned long)__end_rodata; addr += PAGE_SIZE)
832 change_page_attr(virt_to_page(addr), 1, PAGE_KERNEL_RO);
834 printk("Write protecting the kernel read-only data: %uk\n",
835 (__end_rodata - __start_rodata) >> 10);
839 * change_page_attr() requires a global_flush_tlb() call after it.
840 * We do this after the printk so that if something went wrong in the
841 * change, the printk gets out at least to give a better debug hint
842 * of who is the culprit.
848 void free_init_pages(char *what, unsigned long begin, unsigned long end)
852 for (addr = begin; addr < end; addr += PAGE_SIZE) {
853 ClearPageReserved(virt_to_page(addr));
854 init_page_count(virt_to_page(addr));
855 memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE);
859 printk(KERN_INFO "Freeing %s: %ldk freed\n", what, (end - begin) >> 10);
862 void free_initmem(void)
864 free_init_pages("unused kernel memory",
865 (unsigned long)(&__init_begin),
866 (unsigned long)(&__init_end));
869 #ifdef CONFIG_BLK_DEV_INITRD
870 void free_initrd_mem(unsigned long start, unsigned long end)
872 free_init_pages("initrd memory", start, end);