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/config.h>
10 #include <linux/module.h>
11 #include <linux/signal.h>
12 #include <linux/sched.h>
13 #include <linux/kernel.h>
14 #include <linux/errno.h>
15 #include <linux/string.h>
16 #include <linux/types.h>
17 #include <linux/ptrace.h>
18 #include <linux/mman.h>
20 #include <linux/hugetlb.h>
21 #include <linux/swap.h>
22 #include <linux/smp.h>
23 #include <linux/init.h>
24 #include <linux/highmem.h>
25 #include <linux/pagemap.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/dma-mapping.h>
33 #include <linux/scatterlist.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>
44 #include <asm/tlbflush.h>
45 #include <asm/sections.h>
46 #include <asm/hypervisor.h>
47 #include <asm/swiotlb.h>
49 extern unsigned long *contiguous_bitmap;
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 int noinline 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 pmd_table = (pmd_t *) alloc_bootmem_low_pages(PAGE_SIZE);
70 make_lowmem_page_readonly(pmd_table, XENFEAT_writable_page_tables);
71 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
72 pud = pud_offset(pgd, 0);
73 if (pmd_table != pmd_offset(pud, 0))
76 pud = pud_offset(pgd, 0);
77 pmd_table = pmd_offset(pud, 0);
84 * Create a page table and place a pointer to it in a middle page
87 static pte_t * __init one_page_table_init(pmd_t *pmd)
90 pte_t *page_table = (pte_t *) alloc_bootmem_low_pages(PAGE_SIZE);
91 make_lowmem_page_readonly(page_table,
92 XENFEAT_writable_page_tables);
93 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
94 if (page_table != pte_offset_kernel(pmd, 0))
100 return pte_offset_kernel(pmd, 0);
104 * This function initializes a certain range of kernel virtual memory
105 * with new bootmem page tables, everywhere page tables are missing in
110 * NOTE: The pagetables are allocated contiguous on the physical space
111 * so we can cache the place of the first one and move around without
112 * checking the pgd every time.
114 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++) {
129 one_md_table_init(pgd);
130 pud = pud_offset(pgd, vaddr);
131 pmd = pmd_offset(pud, vaddr);
132 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end); pmd++, pmd_idx++) {
133 if (vaddr < HYPERVISOR_VIRT_START && pmd_none(*pmd))
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)
160 int pgd_idx, pmd_idx, pte_ofs;
162 unsigned long max_ram_pfn = xen_start_info->nr_pages;
163 if (max_ram_pfn > max_low_pfn)
164 max_ram_pfn = max_low_pfn;
166 pgd_idx = pgd_index(PAGE_OFFSET);
167 pgd = pgd_base + pgd_idx;
169 pmd_idx = pmd_index(PAGE_OFFSET);
170 pte_ofs = pte_index(PAGE_OFFSET);
172 for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
175 * Native linux hasn't PAE-paging enabled yet at this
176 * point. When running as xen domain we are in PAE
177 * mode already, thus we can't simply hook a empty
178 * pmd. That would kill the mappings we are currently
181 pmd = pmd_offset(pud_offset(pgd, PAGE_OFFSET), PAGE_OFFSET);
183 pmd = one_md_table_init(pgd);
185 if (pfn >= max_low_pfn)
188 for (; pmd_idx < PTRS_PER_PMD && pfn < max_low_pfn; pmd++, pmd_idx++) {
189 unsigned int address = pfn * PAGE_SIZE + PAGE_OFFSET;
190 if (address >= HYPERVISOR_VIRT_START)
193 /* Map with big pages if possible, otherwise create normal page tables. */
195 unsigned int address2 = (pfn + PTRS_PER_PTE - 1) * PAGE_SIZE + PAGE_OFFSET + PAGE_SIZE-1;
197 if (is_kernel_text(address) || is_kernel_text(address2))
198 set_pmd(pmd, pfn_pmd(pfn, PAGE_KERNEL_LARGE_EXEC));
200 set_pmd(pmd, pfn_pmd(pfn, PAGE_KERNEL_LARGE));
203 pte = one_page_table_init(pmd);
206 for (; pte_ofs < PTRS_PER_PTE && pfn < max_low_pfn; pte++, pfn++, pte_ofs++) {
207 /* XEN: Only map initial RAM allocation. */
208 if ((pfn >= max_ram_pfn) || pte_present(*pte))
210 if (is_kernel_text(address))
211 set_pte(pte, pfn_pte(pfn, PAGE_KERNEL_EXEC));
213 set_pte(pte, pfn_pte(pfn, PAGE_KERNEL));
224 static inline int page_kills_ppro(unsigned long pagenr)
226 if (pagenr >= 0x70000 && pagenr <= 0x7003F)
231 extern int is_available_memory(efi_memory_desc_t *);
233 int page_is_ram(unsigned long pagenr)
236 unsigned long addr, end;
239 efi_memory_desc_t *md;
242 for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
244 if (!is_available_memory(md))
246 addr = (md->phys_addr+PAGE_SIZE-1) >> PAGE_SHIFT;
247 end = (md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT)) >> PAGE_SHIFT;
249 if ((pagenr >= addr) && (pagenr < end))
255 for (i = 0; i < e820.nr_map; i++) {
257 if (e820.map[i].type != E820_RAM) /* not usable memory */
260 * !!!FIXME!!! Some BIOSen report areas as RAM that
261 * are not. Notably the 640->1Mb area. We need a sanity
264 addr = (e820.map[i].addr+PAGE_SIZE-1) >> PAGE_SHIFT;
265 end = (e820.map[i].addr+e820.map[i].size) >> PAGE_SHIFT;
266 if ((pagenr >= addr) && (pagenr < end))
272 #else /* CONFIG_XEN */
274 #define page_kills_ppro(p) 0
275 #define page_is_ram(p) 1
279 #ifdef CONFIG_HIGHMEM
283 #define kmap_get_fixmap_pte(vaddr) \
284 pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr), vaddr), (vaddr)), (vaddr))
286 static void __init kmap_init(void)
288 unsigned long kmap_vstart;
290 /* cache the first kmap pte */
291 kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
292 kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
294 kmap_prot = PAGE_KERNEL;
297 static void __init permanent_kmaps_init(pgd_t *pgd_base)
306 page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
308 pgd = swapper_pg_dir + pgd_index(vaddr);
309 pud = pud_offset(pgd, vaddr);
310 pmd = pmd_offset(pud, vaddr);
311 pte = pte_offset_kernel(pmd, vaddr);
312 pkmap_page_table = pte;
315 static void __meminit free_new_highpage(struct page *page, int pfn)
317 set_page_count(page, 1);
318 if (pfn < xen_start_info->nr_pages)
323 void __init add_one_highpage_init(struct page *page, int pfn, int bad_ppro)
325 if (page_is_ram(pfn) && !(bad_ppro && page_kills_ppro(pfn))) {
326 ClearPageReserved(page);
327 free_new_highpage(page, pfn);
329 SetPageReserved(page);
332 static int add_one_highpage_hotplug(struct page *page, unsigned long pfn)
334 free_new_highpage(page, pfn);
336 #ifdef CONFIG_FLATMEM
337 max_mapnr = max(pfn, max_mapnr);
344 * Not currently handling the NUMA case.
345 * Assuming single node and all memory that
346 * has been added dynamically that would be
347 * onlined here is in HIGHMEM
349 void online_page(struct page *page)
351 ClearPageReserved(page);
352 add_one_highpage_hotplug(page, page_to_pfn(page));
357 extern void set_highmem_pages_init(int);
359 static void __init set_highmem_pages_init(int bad_ppro)
362 for (pfn = highstart_pfn; pfn < highend_pfn; pfn++)
363 add_one_highpage_init(pfn_to_page(pfn), pfn, bad_ppro);
364 totalram_pages += totalhigh_pages;
366 #endif /* CONFIG_FLATMEM */
369 #define kmap_init() do { } while (0)
370 #define permanent_kmaps_init(pgd_base) do { } while (0)
371 #define set_highmem_pages_init(bad_ppro) do { } while (0)
372 #endif /* CONFIG_HIGHMEM */
374 unsigned long long __PAGE_KERNEL = _PAGE_KERNEL;
375 EXPORT_SYMBOL(__PAGE_KERNEL);
376 unsigned long long __PAGE_KERNEL_EXEC = _PAGE_KERNEL_EXEC;
379 extern void __init remap_numa_kva(void);
381 #define remap_numa_kva() do {} while (0)
384 pgd_t *swapper_pg_dir;
386 static void __init pagetable_init (void)
389 pgd_t *pgd_base = (pgd_t *)xen_start_info->pt_base;
391 swapper_pg_dir = pgd_base;
392 init_mm.pgd = pgd_base;
394 /* Enable PSE if available */
396 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 vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
414 page_table_range_init(vaddr, 0, pgd_base);
416 permanent_kmaps_init(pgd_base);
419 #ifdef CONFIG_SOFTWARE_SUSPEND
421 * Swap suspend & friends need this for resume because things like the intel-agp
422 * driver might have split up a kernel 4MB mapping.
424 char __nosavedata swsusp_pg_dir[PAGE_SIZE]
425 __attribute__ ((aligned (PAGE_SIZE)));
427 static inline void save_pg_dir(void)
429 memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
432 static inline void save_pg_dir(void)
437 void zap_low_mappings (void)
444 * Zap initial low-memory mappings.
446 * Note that "pgd_clear()" doesn't do it for
447 * us, because pgd_clear() is a no-op on i386.
449 for (i = 0; i < USER_PTRS_PER_PGD; i++)
450 #if defined(CONFIG_X86_PAE) && !defined(CONFIG_XEN)
451 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
453 set_pgd(swapper_pg_dir+i, __pgd(0));
458 static int disable_nx __initdata = 0;
459 u64 __supported_pte_mask __read_mostly = ~_PAGE_NX;
460 EXPORT_SYMBOL(__supported_pte_mask);
465 * Control non executable mappings.
470 void __init noexec_setup(const char *str)
472 if (!strncmp(str, "on",2) && cpu_has_nx) {
473 __supported_pte_mask |= _PAGE_NX;
475 } else if (!strncmp(str,"off",3)) {
477 __supported_pte_mask &= ~_PAGE_NX;
482 #ifdef CONFIG_X86_PAE
484 static void __init set_nx(void)
486 unsigned int v[4], l, h;
488 if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) {
489 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]);
490 if ((v[3] & (1 << 20)) && !disable_nx) {
491 rdmsr(MSR_EFER, l, h);
493 wrmsr(MSR_EFER, l, h);
495 __supported_pte_mask |= _PAGE_NX;
501 * Enables/disables executability of a given kernel page and
502 * returns the previous setting.
504 int __init set_kernel_exec(unsigned long vaddr, int enable)
512 pte = lookup_address(vaddr);
515 if (!pte_exec_kernel(*pte))
519 pte->pte_high &= ~(1 << (_PAGE_BIT_NX - 32));
521 pte->pte_high |= 1 << (_PAGE_BIT_NX - 32);
530 * paging_init() sets up the page tables - note that the first 8MB are
531 * already mapped by head.S.
533 * This routines also unmaps the page at virtual kernel address 0, so
534 * that we can trap those pesky NULL-reference errors in the kernel.
536 void __init paging_init(void)
540 #ifdef CONFIG_X86_PAE
543 printk("NX (Execute Disable) protection: active\n");
548 #if defined(CONFIG_X86_PAE) && !defined(CONFIG_XEN)
550 * We will bail out later - printk doesn't work right now so
551 * the user would just see a hanging kernel.
552 * when running as xen domain we are already in PAE mode at
556 set_in_cr4(X86_CR4_PAE);
562 if (!xen_feature(XENFEAT_auto_translated_physmap) ||
563 xen_start_info->shared_info >= xen_start_info->nr_pages) {
564 /* Switch to the real shared_info page, and clear the
566 set_fixmap(FIX_SHARED_INFO, xen_start_info->shared_info);
567 HYPERVISOR_shared_info =
568 (shared_info_t *)fix_to_virt(FIX_SHARED_INFO);
569 memset(empty_zero_page, 0, sizeof(empty_zero_page));
572 /* Setup mapping of lower 1st MB */
573 for (i = 0; i < NR_FIX_ISAMAPS; i++)
574 if (xen_start_info->flags & SIF_PRIVILEGED)
575 set_fixmap(FIX_ISAMAP_BEGIN - i, i * PAGE_SIZE);
577 __set_fixmap(FIX_ISAMAP_BEGIN - i,
578 virt_to_machine(empty_zero_page),
583 * Test if the WP bit works in supervisor mode. It isn't supported on 386's
584 * and also on some strange 486's (NexGen etc.). All 586+'s are OK. This
585 * used to involve black magic jumps to work around some nasty CPU bugs,
586 * but fortunately the switch to using exceptions got rid of all that.
589 static void __init test_wp_bit(void)
591 printk("Checking if this processor honours the WP bit even in supervisor mode... ");
593 /* Any page-aligned address will do, the test is non-destructive */
594 __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
595 boot_cpu_data.wp_works_ok = do_test_wp_bit();
596 clear_fixmap(FIX_WP_TEST);
598 if (!boot_cpu_data.wp_works_ok) {
600 #ifdef CONFIG_X86_WP_WORKS_OK
601 panic("This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
608 static void __init set_max_mapnr_init(void)
610 #ifdef CONFIG_HIGHMEM
611 num_physpages = highend_pfn;
613 num_physpages = max_low_pfn;
615 #ifdef CONFIG_FLATMEM
616 max_mapnr = num_physpages;
620 static struct kcore_list kcore_mem, kcore_vmalloc;
622 void __init mem_init(void)
624 extern int ppro_with_ram_bug(void);
625 int codesize, reservedpages, datasize, initsize;
630 contiguous_bitmap = alloc_bootmem_low_pages(
631 (max_low_pfn + 2*BITS_PER_LONG) >> 3);
632 BUG_ON(!contiguous_bitmap);
633 memset(contiguous_bitmap, 0, (max_low_pfn + 2*BITS_PER_LONG) >> 3);
635 #if defined(CONFIG_SWIOTLB)
639 #ifdef CONFIG_FLATMEM
644 bad_ppro = ppro_with_ram_bug();
646 #ifdef CONFIG_HIGHMEM
647 /* check that fixmap and pkmap do not overlap */
648 if (PKMAP_BASE+LAST_PKMAP*PAGE_SIZE >= FIXADDR_START) {
649 printk(KERN_ERR "fixmap and kmap areas overlap - this will crash\n");
650 printk(KERN_ERR "pkstart: %lxh pkend: %lxh fixstart %lxh\n",
651 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE, FIXADDR_START);
656 set_max_mapnr_init();
658 #ifdef CONFIG_HIGHMEM
659 high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
661 high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
663 printk("vmalloc area: %lx-%lx, maxmem %lx\n",
664 VMALLOC_START,VMALLOC_END,MAXMEM);
665 BUG_ON(VMALLOC_START > VMALLOC_END);
667 /* this will put all low memory onto the freelists */
668 totalram_pages += free_all_bootmem();
669 /* XEN: init and count low-mem pages outside initial allocation. */
670 for (pfn = xen_start_info->nr_pages; pfn < max_low_pfn; pfn++) {
671 ClearPageReserved(&mem_map[pfn]);
672 set_page_count(&mem_map[pfn], 1);
677 for (tmp = 0; tmp < max_low_pfn; tmp++)
679 * Only count reserved RAM pages
681 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
684 set_highmem_pages_init(bad_ppro);
686 codesize = (unsigned long) &_etext - (unsigned long) &_text;
687 datasize = (unsigned long) &_edata - (unsigned long) &_etext;
688 initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin;
690 kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
691 kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
692 VMALLOC_END-VMALLOC_START);
694 printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, %dk reserved, %dk data, %dk init, %ldk highmem)\n",
695 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
696 num_physpages << (PAGE_SHIFT-10),
698 reservedpages << (PAGE_SHIFT-10),
701 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10))
704 #ifdef CONFIG_X86_PAE
706 panic("cannot execute a PAE-enabled kernel on a PAE-less CPU!");
708 if (boot_cpu_data.wp_works_ok < 0)
712 * Subtle. SMP is doing it's boot stuff late (because it has to
713 * fork idle threads) - but it also needs low mappings for the
714 * protected-mode entry to work. We zap these entries only after
715 * the WP-bit has been tested.
721 set_bit(PG_pinned, &virt_to_page(init_mm.pgd)->flags);
725 * this is for the non-NUMA, single node SMP system case.
726 * Specifically, in the case of x86, we will always add
727 * memory to the highmem for now.
729 #ifndef CONFIG_NEED_MULTIPLE_NODES
730 int add_memory(u64 start, u64 size)
732 struct pglist_data *pgdata = &contig_page_data;
733 struct zone *zone = pgdata->node_zones + MAX_NR_ZONES-1;
734 unsigned long start_pfn = start >> PAGE_SHIFT;
735 unsigned long nr_pages = size >> PAGE_SHIFT;
737 return __add_pages(zone, start_pfn, nr_pages);
740 int remove_memory(u64 start, u64 size)
746 kmem_cache_t *pgd_cache;
747 kmem_cache_t *pmd_cache;
749 void __init pgtable_cache_init(void)
751 if (PTRS_PER_PMD > 1) {
752 pmd_cache = kmem_cache_create("pmd",
753 PTRS_PER_PMD*sizeof(pmd_t),
754 PTRS_PER_PMD*sizeof(pmd_t),
759 panic("pgtable_cache_init(): cannot create pmd cache");
761 pgd_cache = kmem_cache_create("pgd",
763 PTRS_PER_PGD*sizeof(pgd_t),
764 PTRS_PER_PGD*sizeof(pgd_t),
773 panic("pgtable_cache_init(): Cannot create pgd cache");
777 * This function cannot be __init, since exceptions don't work in that
778 * section. Put this after the callers, so that it cannot be inlined.
780 static int noinline do_test_wp_bit(void)
785 __asm__ __volatile__(
790 ".section __ex_table,\"a\"\n"
794 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
803 void free_initmem(void)
807 addr = (unsigned long)(&__init_begin);
808 for (; addr < (unsigned long)(&__init_end); addr += PAGE_SIZE) {
809 ClearPageReserved(virt_to_page(addr));
810 set_page_count(virt_to_page(addr), 1);
811 memset((void *)addr, 0xcc, PAGE_SIZE);
815 printk (KERN_INFO "Freeing unused kernel memory: %dk freed\n", (__init_end - __init_begin) >> 10);
818 #ifdef CONFIG_DEBUG_RODATA
820 extern char __start_rodata, __end_rodata;
821 void mark_rodata_ro(void)
823 unsigned long addr = (unsigned long)&__start_rodata;
825 for (; addr < (unsigned long)&__end_rodata; addr += PAGE_SIZE)
826 change_page_attr(virt_to_page(addr), 1, PAGE_KERNEL_RO);
828 printk ("Write protecting the kernel read-only data: %luk\n",
829 (unsigned long)(&__end_rodata - &__start_rodata) >> 10);
832 * change_page_attr() requires a global_flush_tlb() call after it.
833 * We do this after the printk so that if something went wrong in the
834 * change, the printk gets out at least to give a better debug hint
835 * of who is the culprit.
842 #ifdef CONFIG_BLK_DEV_INITRD
843 void free_initrd_mem(unsigned long start, unsigned long end)
846 printk (KERN_INFO "Freeing initrd memory: %ldk freed\n", (end - start) >> 10);
847 for (; start < end; start += PAGE_SIZE) {
848 ClearPageReserved(virt_to_page(start));
849 set_page_count(virt_to_page(start), 1);