1ad757b6aSThomas Gleixner /* 2ad757b6aSThomas Gleixner * 3ad757b6aSThomas Gleixner * Copyright (C) 1995 Linus Torvalds 4ad757b6aSThomas Gleixner * 5ad757b6aSThomas Gleixner * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999 6ad757b6aSThomas Gleixner */ 7ad757b6aSThomas Gleixner 8ad757b6aSThomas Gleixner #include <linux/module.h> 9ad757b6aSThomas Gleixner #include <linux/signal.h> 10ad757b6aSThomas Gleixner #include <linux/sched.h> 11ad757b6aSThomas Gleixner #include <linux/kernel.h> 12ad757b6aSThomas Gleixner #include <linux/errno.h> 13ad757b6aSThomas Gleixner #include <linux/string.h> 14ad757b6aSThomas Gleixner #include <linux/types.h> 15ad757b6aSThomas Gleixner #include <linux/ptrace.h> 16ad757b6aSThomas Gleixner #include <linux/mman.h> 17ad757b6aSThomas Gleixner #include <linux/mm.h> 18ad757b6aSThomas Gleixner #include <linux/hugetlb.h> 19ad757b6aSThomas Gleixner #include <linux/swap.h> 20ad757b6aSThomas Gleixner #include <linux/smp.h> 21ad757b6aSThomas Gleixner #include <linux/init.h> 22ad757b6aSThomas Gleixner #include <linux/highmem.h> 23ad757b6aSThomas Gleixner #include <linux/pagemap.h> 24cfb80c9eSJeremy Fitzhardinge #include <linux/pci.h> 25ad757b6aSThomas Gleixner #include <linux/pfn.h> 26ad757b6aSThomas Gleixner #include <linux/poison.h> 27ad757b6aSThomas Gleixner #include <linux/bootmem.h> 28a9ce6bc1SYinghai Lu #include <linux/memblock.h> 29ad757b6aSThomas Gleixner #include <linux/proc_fs.h> 30ad757b6aSThomas Gleixner #include <linux/memory_hotplug.h> 31ad757b6aSThomas Gleixner #include <linux/initrd.h> 32ad757b6aSThomas Gleixner #include <linux/cpumask.h> 335a0e3ad6STejun Heo #include <linux/gfp.h> 34ad757b6aSThomas Gleixner 35f832ff18SH. Peter Anvin #include <asm/asm.h> 3646eaa670SIngo Molnar #include <asm/bios_ebda.h> 37ad757b6aSThomas Gleixner #include <asm/processor.h> 38ad757b6aSThomas Gleixner #include <asm/system.h> 39ad757b6aSThomas Gleixner #include <asm/uaccess.h> 40ad757b6aSThomas Gleixner #include <asm/pgtable.h> 41ad757b6aSThomas Gleixner #include <asm/dma.h> 42ad757b6aSThomas Gleixner #include <asm/fixmap.h> 43ad757b6aSThomas Gleixner #include <asm/e820.h> 44ad757b6aSThomas Gleixner #include <asm/apic.h> 458550eb99SIngo Molnar #include <asm/bugs.h> 46ad757b6aSThomas Gleixner #include <asm/tlb.h> 47ad757b6aSThomas Gleixner #include <asm/tlbflush.h> 48a5a19c63SJeremy Fitzhardinge #include <asm/pgalloc.h> 49ad757b6aSThomas Gleixner #include <asm/sections.h> 50ad757b6aSThomas Gleixner #include <asm/paravirt.h> 51551889a6SIan Campbell #include <asm/setup.h> 527bfeab9aSHarvey Harrison #include <asm/cacheflush.h> 532b72394eSPekka Enberg #include <asm/page_types.h> 544fcb2083SPekka Enberg #include <asm/init.h> 55ad757b6aSThomas Gleixner 56ad757b6aSThomas Gleixner unsigned long highstart_pfn, highend_pfn; 57ad757b6aSThomas Gleixner 588550eb99SIngo Molnar static noinline int do_test_wp_bit(void); 59ad757b6aSThomas Gleixner 60dc16ecf7SJeremy Fitzhardinge bool __read_mostly __vmalloc_start_set = false; 614e29684cSYinghai Lu 62d6be89adSJan Beulich static __init void *alloc_low_page(void) 634e29684cSYinghai Lu { 64298af9d8SPekka Enberg unsigned long pfn = e820_table_end++; 654e29684cSYinghai Lu void *adr; 664e29684cSYinghai Lu 67298af9d8SPekka Enberg if (pfn >= e820_table_top) 684e29684cSYinghai Lu panic("alloc_low_page: ran out of memory"); 694e29684cSYinghai Lu 704e29684cSYinghai Lu adr = __va(pfn * PAGE_SIZE); 714e29684cSYinghai Lu memset(adr, 0, PAGE_SIZE); 724e29684cSYinghai Lu return adr; 734e29684cSYinghai Lu } 744e29684cSYinghai Lu 75ad757b6aSThomas Gleixner /* 76ad757b6aSThomas Gleixner * Creates a middle page table and puts a pointer to it in the 77ad757b6aSThomas Gleixner * given global directory entry. This only returns the gd entry 78ad757b6aSThomas Gleixner * in non-PAE compilation mode, since the middle layer is folded. 79ad757b6aSThomas Gleixner */ 80ad757b6aSThomas Gleixner static pmd_t * __init one_md_table_init(pgd_t *pgd) 81ad757b6aSThomas Gleixner { 82ad757b6aSThomas Gleixner pud_t *pud; 83ad757b6aSThomas Gleixner pmd_t *pmd_table; 84ad757b6aSThomas Gleixner 85ad757b6aSThomas Gleixner #ifdef CONFIG_X86_PAE 86ad757b6aSThomas Gleixner if (!(pgd_val(*pgd) & _PAGE_PRESENT)) { 87c464573cSPekka Enberg if (after_bootmem) 883c1596efSJan Beulich pmd_table = (pmd_t *)alloc_bootmem_pages(PAGE_SIZE); 894e29684cSYinghai Lu else 90d6be89adSJan Beulich pmd_table = (pmd_t *)alloc_low_page(); 916944a9c8SJeremy Fitzhardinge paravirt_alloc_pmd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT); 92ad757b6aSThomas Gleixner set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT)); 93ad757b6aSThomas Gleixner pud = pud_offset(pgd, 0); 948550eb99SIngo Molnar BUG_ON(pmd_table != pmd_offset(pud, 0)); 95a376f30aSZhaolei 96a376f30aSZhaolei return pmd_table; 97ad757b6aSThomas Gleixner } 98ad757b6aSThomas Gleixner #endif 99ad757b6aSThomas Gleixner pud = pud_offset(pgd, 0); 100ad757b6aSThomas Gleixner pmd_table = pmd_offset(pud, 0); 1018550eb99SIngo Molnar 102ad757b6aSThomas Gleixner return pmd_table; 103ad757b6aSThomas Gleixner } 104ad757b6aSThomas Gleixner 105ad757b6aSThomas Gleixner /* 106ad757b6aSThomas Gleixner * Create a page table and place a pointer to it in a middle page 1078550eb99SIngo Molnar * directory entry: 108ad757b6aSThomas Gleixner */ 109ad757b6aSThomas Gleixner static pte_t * __init one_page_table_init(pmd_t *pmd) 110ad757b6aSThomas Gleixner { 111ad757b6aSThomas Gleixner if (!(pmd_val(*pmd) & _PAGE_PRESENT)) { 112509a80c4SIngo Molnar pte_t *page_table = NULL; 113509a80c4SIngo Molnar 114c464573cSPekka Enberg if (after_bootmem) { 115f8561296SVegard Nossum #if defined(CONFIG_DEBUG_PAGEALLOC) || defined(CONFIG_KMEMCHECK) 116509a80c4SIngo Molnar page_table = (pte_t *) alloc_bootmem_pages(PAGE_SIZE); 117509a80c4SIngo Molnar #endif 1184e29684cSYinghai Lu if (!page_table) 119509a80c4SIngo Molnar page_table = 1203c1596efSJan Beulich (pte_t *)alloc_bootmem_pages(PAGE_SIZE); 121d6be89adSJan Beulich } else 122d6be89adSJan Beulich page_table = (pte_t *)alloc_low_page(); 123ad757b6aSThomas Gleixner 1246944a9c8SJeremy Fitzhardinge paravirt_alloc_pte(&init_mm, __pa(page_table) >> PAGE_SHIFT); 125ad757b6aSThomas Gleixner set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE)); 126ad757b6aSThomas Gleixner BUG_ON(page_table != pte_offset_kernel(pmd, 0)); 127ad757b6aSThomas Gleixner } 128ad757b6aSThomas Gleixner 129ad757b6aSThomas Gleixner return pte_offset_kernel(pmd, 0); 130ad757b6aSThomas Gleixner } 131ad757b6aSThomas Gleixner 132458a3e64STejun Heo pmd_t * __init populate_extra_pmd(unsigned long vaddr) 13311124411STejun Heo { 13411124411STejun Heo int pgd_idx = pgd_index(vaddr); 13511124411STejun Heo int pmd_idx = pmd_index(vaddr); 136458a3e64STejun Heo 137458a3e64STejun Heo return one_md_table_init(swapper_pg_dir + pgd_idx) + pmd_idx; 138458a3e64STejun Heo } 139458a3e64STejun Heo 140458a3e64STejun Heo pte_t * __init populate_extra_pte(unsigned long vaddr) 141458a3e64STejun Heo { 142458a3e64STejun Heo int pte_idx = pte_index(vaddr); 14311124411STejun Heo pmd_t *pmd; 14411124411STejun Heo 145458a3e64STejun Heo pmd = populate_extra_pmd(vaddr); 146458a3e64STejun Heo return one_page_table_init(pmd) + pte_idx; 14711124411STejun Heo } 14811124411STejun Heo 149a3c6018eSJan Beulich static pte_t *__init page_table_kmap_check(pte_t *pte, pmd_t *pmd, 150a3c6018eSJan Beulich unsigned long vaddr, pte_t *lastpte) 151a3c6018eSJan Beulich { 152a3c6018eSJan Beulich #ifdef CONFIG_HIGHMEM 153a3c6018eSJan Beulich /* 154a3c6018eSJan Beulich * Something (early fixmap) may already have put a pte 155a3c6018eSJan Beulich * page here, which causes the page table allocation 156a3c6018eSJan Beulich * to become nonlinear. Attempt to fix it, and if it 157a3c6018eSJan Beulich * is still nonlinear then we have to bug. 158a3c6018eSJan Beulich */ 159a3c6018eSJan Beulich int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT; 160a3c6018eSJan Beulich int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT; 161a3c6018eSJan Beulich 162a3c6018eSJan Beulich if (pmd_idx_kmap_begin != pmd_idx_kmap_end 163a3c6018eSJan Beulich && (vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin 164a3c6018eSJan Beulich && (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end 165298af9d8SPekka Enberg && ((__pa(pte) >> PAGE_SHIFT) < e820_table_start 166298af9d8SPekka Enberg || (__pa(pte) >> PAGE_SHIFT) >= e820_table_end)) { 167a3c6018eSJan Beulich pte_t *newpte; 168a3c6018eSJan Beulich int i; 169a3c6018eSJan Beulich 170c464573cSPekka Enberg BUG_ON(after_bootmem); 171a3c6018eSJan Beulich newpte = alloc_low_page(); 172a3c6018eSJan Beulich for (i = 0; i < PTRS_PER_PTE; i++) 173a3c6018eSJan Beulich set_pte(newpte + i, pte[i]); 174a3c6018eSJan Beulich 175a3c6018eSJan Beulich paravirt_alloc_pte(&init_mm, __pa(newpte) >> PAGE_SHIFT); 176a3c6018eSJan Beulich set_pmd(pmd, __pmd(__pa(newpte)|_PAGE_TABLE)); 177a3c6018eSJan Beulich BUG_ON(newpte != pte_offset_kernel(pmd, 0)); 178a3c6018eSJan Beulich __flush_tlb_all(); 179a3c6018eSJan Beulich 180a3c6018eSJan Beulich paravirt_release_pte(__pa(pte) >> PAGE_SHIFT); 181a3c6018eSJan Beulich pte = newpte; 182a3c6018eSJan Beulich } 183a3c6018eSJan Beulich BUG_ON(vaddr < fix_to_virt(FIX_KMAP_BEGIN - 1) 184a3c6018eSJan Beulich && vaddr > fix_to_virt(FIX_KMAP_END) 185a3c6018eSJan Beulich && lastpte && lastpte + PTRS_PER_PTE != pte); 186a3c6018eSJan Beulich #endif 187a3c6018eSJan Beulich return pte; 188a3c6018eSJan Beulich } 189a3c6018eSJan Beulich 190ad757b6aSThomas Gleixner /* 191ad757b6aSThomas Gleixner * This function initializes a certain range of kernel virtual memory 192ad757b6aSThomas Gleixner * with new bootmem page tables, everywhere page tables are missing in 193ad757b6aSThomas Gleixner * the given range. 1948550eb99SIngo Molnar * 195ad757b6aSThomas Gleixner * NOTE: The pagetables are allocated contiguous on the physical space 196ad757b6aSThomas Gleixner * so we can cache the place of the first one and move around without 197ad757b6aSThomas Gleixner * checking the pgd every time. 198ad757b6aSThomas Gleixner */ 1998550eb99SIngo Molnar static void __init 2008550eb99SIngo Molnar page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base) 201ad757b6aSThomas Gleixner { 202ad757b6aSThomas Gleixner int pgd_idx, pmd_idx; 203ad757b6aSThomas Gleixner unsigned long vaddr; 2048550eb99SIngo Molnar pgd_t *pgd; 2058550eb99SIngo Molnar pmd_t *pmd; 206a3c6018eSJan Beulich pte_t *pte = NULL; 207ad757b6aSThomas Gleixner 208ad757b6aSThomas Gleixner vaddr = start; 209ad757b6aSThomas Gleixner pgd_idx = pgd_index(vaddr); 210ad757b6aSThomas Gleixner pmd_idx = pmd_index(vaddr); 211ad757b6aSThomas Gleixner pgd = pgd_base + pgd_idx; 212ad757b6aSThomas Gleixner 213ad757b6aSThomas Gleixner for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) { 214ad757b6aSThomas Gleixner pmd = one_md_table_init(pgd); 215ad757b6aSThomas Gleixner pmd = pmd + pmd_index(vaddr); 2168550eb99SIngo Molnar for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end); 2178550eb99SIngo Molnar pmd++, pmd_idx++) { 218a3c6018eSJan Beulich pte = page_table_kmap_check(one_page_table_init(pmd), 219a3c6018eSJan Beulich pmd, vaddr, pte); 220ad757b6aSThomas Gleixner 221ad757b6aSThomas Gleixner vaddr += PMD_SIZE; 222ad757b6aSThomas Gleixner } 223ad757b6aSThomas Gleixner pmd_idx = 0; 224ad757b6aSThomas Gleixner } 225ad757b6aSThomas Gleixner } 226ad757b6aSThomas Gleixner 227ad757b6aSThomas Gleixner static inline int is_kernel_text(unsigned long addr) 228ad757b6aSThomas Gleixner { 229ad757b6aSThomas Gleixner if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end) 230ad757b6aSThomas Gleixner return 1; 231ad757b6aSThomas Gleixner return 0; 232ad757b6aSThomas Gleixner } 233ad757b6aSThomas Gleixner 234ad757b6aSThomas Gleixner /* 235ad757b6aSThomas Gleixner * This maps the physical memory to kernel virtual address space, a total 236ad757b6aSThomas Gleixner * of max_low_pfn pages, by creating page tables starting from address 2378550eb99SIngo Molnar * PAGE_OFFSET: 238ad757b6aSThomas Gleixner */ 239e53fb04fSPekka Enberg unsigned long __init 240e53fb04fSPekka Enberg kernel_physical_mapping_init(unsigned long start, 241e53fb04fSPekka Enberg unsigned long end, 242e53fb04fSPekka Enberg unsigned long page_size_mask) 243ad757b6aSThomas Gleixner { 244e53fb04fSPekka Enberg int use_pse = page_size_mask == (1<<PG_LEVEL_2M); 245c1fd1b43SPekka Enberg unsigned long last_map_addr = end; 246e53fb04fSPekka Enberg unsigned long start_pfn, end_pfn; 247e7179853SPekka Enberg pgd_t *pgd_base = swapper_pg_dir; 2488550eb99SIngo Molnar int pgd_idx, pmd_idx, pte_ofs; 249ad757b6aSThomas Gleixner unsigned long pfn; 250ad757b6aSThomas Gleixner pgd_t *pgd; 251ad757b6aSThomas Gleixner pmd_t *pmd; 252ad757b6aSThomas Gleixner pte_t *pte; 253a2699e47SSuresh Siddha unsigned pages_2m, pages_4k; 254a2699e47SSuresh Siddha int mapping_iter; 255a2699e47SSuresh Siddha 256e53fb04fSPekka Enberg start_pfn = start >> PAGE_SHIFT; 257e53fb04fSPekka Enberg end_pfn = end >> PAGE_SHIFT; 258e53fb04fSPekka Enberg 259a2699e47SSuresh Siddha /* 260a2699e47SSuresh Siddha * First iteration will setup identity mapping using large/small pages 261a2699e47SSuresh Siddha * based on use_pse, with other attributes same as set by 262a2699e47SSuresh Siddha * the early code in head_32.S 263a2699e47SSuresh Siddha * 264a2699e47SSuresh Siddha * Second iteration will setup the appropriate attributes (NX, GLOBAL..) 265a2699e47SSuresh Siddha * as desired for the kernel identity mapping. 266a2699e47SSuresh Siddha * 267a2699e47SSuresh Siddha * This two pass mechanism conforms to the TLB app note which says: 268a2699e47SSuresh Siddha * 269a2699e47SSuresh Siddha * "Software should not write to a paging-structure entry in a way 270a2699e47SSuresh Siddha * that would change, for any linear address, both the page size 271a2699e47SSuresh Siddha * and either the page frame or attributes." 272a2699e47SSuresh Siddha */ 273a2699e47SSuresh Siddha mapping_iter = 1; 274ad757b6aSThomas Gleixner 275a04ad82dSYinghai Lu if (!cpu_has_pse) 276a04ad82dSYinghai Lu use_pse = 0; 277a04ad82dSYinghai Lu 278a2699e47SSuresh Siddha repeat: 279a2699e47SSuresh Siddha pages_2m = pages_4k = 0; 280a04ad82dSYinghai Lu pfn = start_pfn; 281a04ad82dSYinghai Lu pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET); 282ad757b6aSThomas Gleixner pgd = pgd_base + pgd_idx; 283ad757b6aSThomas Gleixner for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) { 284ad757b6aSThomas Gleixner pmd = one_md_table_init(pgd); 2858550eb99SIngo Molnar 286a04ad82dSYinghai Lu if (pfn >= end_pfn) 287a04ad82dSYinghai Lu continue; 288a04ad82dSYinghai Lu #ifdef CONFIG_X86_PAE 289a04ad82dSYinghai Lu pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET); 290a04ad82dSYinghai Lu pmd += pmd_idx; 291a04ad82dSYinghai Lu #else 292a04ad82dSYinghai Lu pmd_idx = 0; 293a04ad82dSYinghai Lu #endif 294a04ad82dSYinghai Lu for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn; 295f3f20de8SJeremy Fitzhardinge pmd++, pmd_idx++) { 2968550eb99SIngo Molnar unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET; 297ad757b6aSThomas Gleixner 2988550eb99SIngo Molnar /* 2998550eb99SIngo Molnar * Map with big pages if possible, otherwise 3008550eb99SIngo Molnar * create normal page tables: 3018550eb99SIngo Molnar */ 302a04ad82dSYinghai Lu if (use_pse) { 3038550eb99SIngo Molnar unsigned int addr2; 304f3f20de8SJeremy Fitzhardinge pgprot_t prot = PAGE_KERNEL_LARGE; 305a2699e47SSuresh Siddha /* 306a2699e47SSuresh Siddha * first pass will use the same initial 307a2699e47SSuresh Siddha * identity mapping attribute + _PAGE_PSE. 308a2699e47SSuresh Siddha */ 309a2699e47SSuresh Siddha pgprot_t init_prot = 310a2699e47SSuresh Siddha __pgprot(PTE_IDENT_ATTR | 311a2699e47SSuresh Siddha _PAGE_PSE); 312f3f20de8SJeremy Fitzhardinge 3138550eb99SIngo Molnar addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE + 314f3f20de8SJeremy Fitzhardinge PAGE_OFFSET + PAGE_SIZE-1; 315f3f20de8SJeremy Fitzhardinge 3168550eb99SIngo Molnar if (is_kernel_text(addr) || 3178550eb99SIngo Molnar is_kernel_text(addr2)) 318f3f20de8SJeremy Fitzhardinge prot = PAGE_KERNEL_LARGE_EXEC; 319f3f20de8SJeremy Fitzhardinge 320ce0c0e50SAndi Kleen pages_2m++; 321a2699e47SSuresh Siddha if (mapping_iter == 1) 322a2699e47SSuresh Siddha set_pmd(pmd, pfn_pmd(pfn, init_prot)); 323a2699e47SSuresh Siddha else 324f3f20de8SJeremy Fitzhardinge set_pmd(pmd, pfn_pmd(pfn, prot)); 325ad757b6aSThomas Gleixner 326ad757b6aSThomas Gleixner pfn += PTRS_PER_PTE; 3278550eb99SIngo Molnar continue; 3288550eb99SIngo Molnar } 329ad757b6aSThomas Gleixner pte = one_page_table_init(pmd); 330ad757b6aSThomas Gleixner 331a04ad82dSYinghai Lu pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET); 332a04ad82dSYinghai Lu pte += pte_ofs; 333a04ad82dSYinghai Lu for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn; 3348550eb99SIngo Molnar pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) { 335f3f20de8SJeremy Fitzhardinge pgprot_t prot = PAGE_KERNEL; 336a2699e47SSuresh Siddha /* 337a2699e47SSuresh Siddha * first pass will use the same initial 338a2699e47SSuresh Siddha * identity mapping attribute. 339a2699e47SSuresh Siddha */ 340a2699e47SSuresh Siddha pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR); 341f3f20de8SJeremy Fitzhardinge 3428550eb99SIngo Molnar if (is_kernel_text(addr)) 343f3f20de8SJeremy Fitzhardinge prot = PAGE_KERNEL_EXEC; 344f3f20de8SJeremy Fitzhardinge 345ce0c0e50SAndi Kleen pages_4k++; 346c1fd1b43SPekka Enberg if (mapping_iter == 1) { 347a2699e47SSuresh Siddha set_pte(pte, pfn_pte(pfn, init_prot)); 348c1fd1b43SPekka Enberg last_map_addr = (pfn << PAGE_SHIFT) + PAGE_SIZE; 349c1fd1b43SPekka Enberg } else 350f3f20de8SJeremy Fitzhardinge set_pte(pte, pfn_pte(pfn, prot)); 351ad757b6aSThomas Gleixner } 352ad757b6aSThomas Gleixner } 353ad757b6aSThomas Gleixner } 354a2699e47SSuresh Siddha if (mapping_iter == 1) { 355a2699e47SSuresh Siddha /* 356a2699e47SSuresh Siddha * update direct mapping page count only in the first 357a2699e47SSuresh Siddha * iteration. 358a2699e47SSuresh Siddha */ 359ce0c0e50SAndi Kleen update_page_count(PG_LEVEL_2M, pages_2m); 360ce0c0e50SAndi Kleen update_page_count(PG_LEVEL_4K, pages_4k); 361a2699e47SSuresh Siddha 362a2699e47SSuresh Siddha /* 363a2699e47SSuresh Siddha * local global flush tlb, which will flush the previous 364a2699e47SSuresh Siddha * mappings present in both small and large page TLB's. 365a2699e47SSuresh Siddha */ 366a2699e47SSuresh Siddha __flush_tlb_all(); 367a2699e47SSuresh Siddha 368a2699e47SSuresh Siddha /* 369a2699e47SSuresh Siddha * Second iteration will set the actual desired PTE attributes. 370a2699e47SSuresh Siddha */ 371a2699e47SSuresh Siddha mapping_iter = 2; 372a2699e47SSuresh Siddha goto repeat; 373a2699e47SSuresh Siddha } 374c1fd1b43SPekka Enberg return last_map_addr; 375ae531c26SArjan van de Ven } 376ae531c26SArjan van de Ven 377ad757b6aSThomas Gleixner pte_t *kmap_pte; 378ad757b6aSThomas Gleixner pgprot_t kmap_prot; 379ad757b6aSThomas Gleixner 3808550eb99SIngo Molnar static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr) 3818550eb99SIngo Molnar { 3828550eb99SIngo Molnar return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr), 3838550eb99SIngo Molnar vaddr), vaddr), vaddr); 3848550eb99SIngo Molnar } 385ad757b6aSThomas Gleixner 386ad757b6aSThomas Gleixner static void __init kmap_init(void) 387ad757b6aSThomas Gleixner { 388ad757b6aSThomas Gleixner unsigned long kmap_vstart; 389ad757b6aSThomas Gleixner 3908550eb99SIngo Molnar /* 3918550eb99SIngo Molnar * Cache the first kmap pte: 3928550eb99SIngo Molnar */ 393ad757b6aSThomas Gleixner kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN); 394ad757b6aSThomas Gleixner kmap_pte = kmap_get_fixmap_pte(kmap_vstart); 395ad757b6aSThomas Gleixner 396ad757b6aSThomas Gleixner kmap_prot = PAGE_KERNEL; 397ad757b6aSThomas Gleixner } 398ad757b6aSThomas Gleixner 399fd940934SKeith Packard #ifdef CONFIG_HIGHMEM 400ad757b6aSThomas Gleixner static void __init permanent_kmaps_init(pgd_t *pgd_base) 401ad757b6aSThomas Gleixner { 4028550eb99SIngo Molnar unsigned long vaddr; 403ad757b6aSThomas Gleixner pgd_t *pgd; 404ad757b6aSThomas Gleixner pud_t *pud; 405ad757b6aSThomas Gleixner pmd_t *pmd; 406ad757b6aSThomas Gleixner pte_t *pte; 407ad757b6aSThomas Gleixner 408ad757b6aSThomas Gleixner vaddr = PKMAP_BASE; 409ad757b6aSThomas Gleixner page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base); 410ad757b6aSThomas Gleixner 411ad757b6aSThomas Gleixner pgd = swapper_pg_dir + pgd_index(vaddr); 412ad757b6aSThomas Gleixner pud = pud_offset(pgd, vaddr); 413ad757b6aSThomas Gleixner pmd = pmd_offset(pud, vaddr); 414ad757b6aSThomas Gleixner pte = pte_offset_kernel(pmd, vaddr); 415ad757b6aSThomas Gleixner pkmap_page_table = pte; 416ad757b6aSThomas Gleixner } 417ad757b6aSThomas Gleixner 418b1258ac2SMinchan Kim static void __init add_one_highpage_init(struct page *page) 419ad757b6aSThomas Gleixner { 420ad757b6aSThomas Gleixner ClearPageReserved(page); 421180c06efSJeremy Fitzhardinge init_page_count(page); 422180c06efSJeremy Fitzhardinge __free_page(page); 423180c06efSJeremy Fitzhardinge totalhigh_pages++; 424ad757b6aSThomas Gleixner } 425ad757b6aSThomas Gleixner 426*1d931264SYinghai Lu void __init add_highpages_with_active_regions(int nid, 427*1d931264SYinghai Lu unsigned long start_pfn, unsigned long end_pfn) 428b5bc6c0eSYinghai Lu { 429*1d931264SYinghai Lu struct range *range; 430*1d931264SYinghai Lu int nr_range; 431*1d931264SYinghai Lu int i; 432*1d931264SYinghai Lu 433*1d931264SYinghai Lu nr_range = __get_free_all_memory_range(&range, nid, start_pfn, end_pfn); 434*1d931264SYinghai Lu 435*1d931264SYinghai Lu for (i = 0; i < nr_range; i++) { 436b5bc6c0eSYinghai Lu struct page *page; 437*1d931264SYinghai Lu int node_pfn; 438b5bc6c0eSYinghai Lu 439*1d931264SYinghai Lu for (node_pfn = range[i].start; node_pfn < range[i].end; 440b5bc6c0eSYinghai Lu node_pfn++) { 441b5bc6c0eSYinghai Lu if (!pfn_valid(node_pfn)) 442b5bc6c0eSYinghai Lu continue; 443b5bc6c0eSYinghai Lu page = pfn_to_page(node_pfn); 444b1258ac2SMinchan Kim add_one_highpage_init(page); 445b5bc6c0eSYinghai Lu } 446b5bc6c0eSYinghai Lu } 447ad757b6aSThomas Gleixner } 448ad757b6aSThomas Gleixner #else 449e8e32326SIngo Brueckl static inline void permanent_kmaps_init(pgd_t *pgd_base) 450e8e32326SIngo Brueckl { 451e8e32326SIngo Brueckl } 452ad757b6aSThomas Gleixner #endif /* CONFIG_HIGHMEM */ 453ad757b6aSThomas Gleixner 454ad757b6aSThomas Gleixner void __init native_pagetable_setup_start(pgd_t *base) 455ad757b6aSThomas Gleixner { 456551889a6SIan Campbell unsigned long pfn, va; 457551889a6SIan Campbell pgd_t *pgd; 458551889a6SIan Campbell pud_t *pud; 459551889a6SIan Campbell pmd_t *pmd; 460551889a6SIan Campbell pte_t *pte; 461ad757b6aSThomas Gleixner 462ad757b6aSThomas Gleixner /* 463551889a6SIan Campbell * Remove any mappings which extend past the end of physical 464551889a6SIan Campbell * memory from the boot time page table: 465ad757b6aSThomas Gleixner */ 466551889a6SIan Campbell for (pfn = max_low_pfn + 1; pfn < 1<<(32-PAGE_SHIFT); pfn++) { 467551889a6SIan Campbell va = PAGE_OFFSET + (pfn<<PAGE_SHIFT); 468551889a6SIan Campbell pgd = base + pgd_index(va); 469551889a6SIan Campbell if (!pgd_present(*pgd)) 470551889a6SIan Campbell break; 471ad757b6aSThomas Gleixner 472551889a6SIan Campbell pud = pud_offset(pgd, va); 473551889a6SIan Campbell pmd = pmd_offset(pud, va); 474551889a6SIan Campbell if (!pmd_present(*pmd)) 475551889a6SIan Campbell break; 476551889a6SIan Campbell 477551889a6SIan Campbell pte = pte_offset_kernel(pmd, va); 478551889a6SIan Campbell if (!pte_present(*pte)) 479551889a6SIan Campbell break; 480551889a6SIan Campbell 481551889a6SIan Campbell pte_clear(NULL, va, pte); 482551889a6SIan Campbell } 4836944a9c8SJeremy Fitzhardinge paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT); 484ad757b6aSThomas Gleixner } 485ad757b6aSThomas Gleixner 486ad757b6aSThomas Gleixner void __init native_pagetable_setup_done(pgd_t *base) 487ad757b6aSThomas Gleixner { 488ad757b6aSThomas Gleixner } 489ad757b6aSThomas Gleixner 490ad757b6aSThomas Gleixner /* 491ad757b6aSThomas Gleixner * Build a proper pagetable for the kernel mappings. Up until this 492ad757b6aSThomas Gleixner * point, we've been running on some set of pagetables constructed by 493ad757b6aSThomas Gleixner * the boot process. 494ad757b6aSThomas Gleixner * 495ad757b6aSThomas Gleixner * If we're booting on native hardware, this will be a pagetable 496551889a6SIan Campbell * constructed in arch/x86/kernel/head_32.S. The root of the 497551889a6SIan Campbell * pagetable will be swapper_pg_dir. 498ad757b6aSThomas Gleixner * 499ad757b6aSThomas Gleixner * If we're booting paravirtualized under a hypervisor, then there are 500ad757b6aSThomas Gleixner * more options: we may already be running PAE, and the pagetable may 501ad757b6aSThomas Gleixner * or may not be based in swapper_pg_dir. In any case, 502ad757b6aSThomas Gleixner * paravirt_pagetable_setup_start() will set up swapper_pg_dir 503ad757b6aSThomas Gleixner * appropriately for the rest of the initialization to work. 504ad757b6aSThomas Gleixner * 505ad757b6aSThomas Gleixner * In general, pagetable_init() assumes that the pagetable may already 506ad757b6aSThomas Gleixner * be partially populated, and so it avoids stomping on any existing 507ad757b6aSThomas Gleixner * mappings. 508ad757b6aSThomas Gleixner */ 509f765090aSPekka Enberg void __init early_ioremap_page_table_range_init(void) 510ad757b6aSThomas Gleixner { 511e7179853SPekka Enberg pgd_t *pgd_base = swapper_pg_dir; 5128550eb99SIngo Molnar unsigned long vaddr, end; 513ad757b6aSThomas Gleixner 514ad757b6aSThomas Gleixner /* 515ad757b6aSThomas Gleixner * Fixed mappings, only the page table structure has to be 516ad757b6aSThomas Gleixner * created - mappings will be set by set_fixmap(): 517ad757b6aSThomas Gleixner */ 518ad757b6aSThomas Gleixner vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK; 519ad757b6aSThomas Gleixner end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK; 520ad757b6aSThomas Gleixner page_table_range_init(vaddr, end, pgd_base); 521beacfaacSHuang, Ying early_ioremap_reset(); 522e7b37895SYinghai Lu } 523e7b37895SYinghai Lu 524e7b37895SYinghai Lu static void __init pagetable_init(void) 525e7b37895SYinghai Lu { 526e7b37895SYinghai Lu pgd_t *pgd_base = swapper_pg_dir; 527e7b37895SYinghai Lu 528ad757b6aSThomas Gleixner permanent_kmaps_init(pgd_base); 529ad757b6aSThomas Gleixner } 530ad757b6aSThomas Gleixner 531a6eb84bcSRafael J. Wysocki #ifdef CONFIG_ACPI_SLEEP 532ad757b6aSThomas Gleixner /* 533a6eb84bcSRafael J. Wysocki * ACPI suspend needs this for resume, because things like the intel-agp 534ad757b6aSThomas Gleixner * driver might have split up a kernel 4MB mapping. 535ad757b6aSThomas Gleixner */ 536a6eb84bcSRafael J. Wysocki char swsusp_pg_dir[PAGE_SIZE] 537ad757b6aSThomas Gleixner __attribute__ ((aligned(PAGE_SIZE))); 538ad757b6aSThomas Gleixner 539ad757b6aSThomas Gleixner static inline void save_pg_dir(void) 540ad757b6aSThomas Gleixner { 541ad757b6aSThomas Gleixner memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE); 542ad757b6aSThomas Gleixner } 543a6eb84bcSRafael J. Wysocki #else /* !CONFIG_ACPI_SLEEP */ 544ad757b6aSThomas Gleixner static inline void save_pg_dir(void) 545ad757b6aSThomas Gleixner { 546ad757b6aSThomas Gleixner } 547a6eb84bcSRafael J. Wysocki #endif /* !CONFIG_ACPI_SLEEP */ 548ad757b6aSThomas Gleixner 54955cd6367SYinghai Lu void zap_low_mappings(bool early) 550ad757b6aSThomas Gleixner { 551ad757b6aSThomas Gleixner int i; 552ad757b6aSThomas Gleixner 553ad757b6aSThomas Gleixner /* 554ad757b6aSThomas Gleixner * Zap initial low-memory mappings. 555ad757b6aSThomas Gleixner * 556ad757b6aSThomas Gleixner * Note that "pgd_clear()" doesn't do it for 557ad757b6aSThomas Gleixner * us, because pgd_clear() is a no-op on i386. 558ad757b6aSThomas Gleixner */ 55968db065cSJeremy Fitzhardinge for (i = 0; i < KERNEL_PGD_BOUNDARY; i++) { 560ad757b6aSThomas Gleixner #ifdef CONFIG_X86_PAE 561ad757b6aSThomas Gleixner set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page))); 562ad757b6aSThomas Gleixner #else 563ad757b6aSThomas Gleixner set_pgd(swapper_pg_dir+i, __pgd(0)); 564ad757b6aSThomas Gleixner #endif 5658550eb99SIngo Molnar } 56655cd6367SYinghai Lu 56755cd6367SYinghai Lu if (early) 56855cd6367SYinghai Lu __flush_tlb(); 56955cd6367SYinghai Lu else 570ad757b6aSThomas Gleixner flush_tlb_all(); 571ad757b6aSThomas Gleixner } 572ad757b6aSThomas Gleixner 573be43d728SJeremy Fitzhardinge pteval_t __supported_pte_mask __read_mostly = ~(_PAGE_NX | _PAGE_GLOBAL | _PAGE_IOMAP); 5746fdc05d4SJeremy Fitzhardinge EXPORT_SYMBOL_GPL(__supported_pte_mask); 5756fdc05d4SJeremy Fitzhardinge 57690d967e0SYinghai Lu /* user-defined highmem size */ 57790d967e0SYinghai Lu static unsigned int highmem_pages = -1; 57890d967e0SYinghai Lu 57990d967e0SYinghai Lu /* 58090d967e0SYinghai Lu * highmem=size forces highmem to be exactly 'size' bytes. 58190d967e0SYinghai Lu * This works even on boxes that have no highmem otherwise. 58290d967e0SYinghai Lu * This also works to reduce highmem size on bigger boxes. 58390d967e0SYinghai Lu */ 58490d967e0SYinghai Lu static int __init parse_highmem(char *arg) 58590d967e0SYinghai Lu { 58690d967e0SYinghai Lu if (!arg) 58790d967e0SYinghai Lu return -EINVAL; 58890d967e0SYinghai Lu 58990d967e0SYinghai Lu highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT; 59090d967e0SYinghai Lu return 0; 59190d967e0SYinghai Lu } 59290d967e0SYinghai Lu early_param("highmem", parse_highmem); 59390d967e0SYinghai Lu 5944769843bSIngo Molnar #define MSG_HIGHMEM_TOO_BIG \ 5954769843bSIngo Molnar "highmem size (%luMB) is bigger than pages available (%luMB)!\n" 5964769843bSIngo Molnar 5974769843bSIngo Molnar #define MSG_LOWMEM_TOO_SMALL \ 5984769843bSIngo Molnar "highmem size (%luMB) results in <64MB lowmem, ignoring it!\n" 59990d967e0SYinghai Lu /* 6004769843bSIngo Molnar * All of RAM fits into lowmem - but if user wants highmem 6014769843bSIngo Molnar * artificially via the highmem=x boot parameter then create 6024769843bSIngo Molnar * it: 60390d967e0SYinghai Lu */ 6044769843bSIngo Molnar void __init lowmem_pfn_init(void) 60590d967e0SYinghai Lu { 606346cafecSYinghai Lu /* max_low_pfn is 0, we already have early_res support */ 60790d967e0SYinghai Lu max_low_pfn = max_pfn; 608d88316c2SIngo Molnar 6094769843bSIngo Molnar if (highmem_pages == -1) 6104769843bSIngo Molnar highmem_pages = 0; 6114769843bSIngo Molnar #ifdef CONFIG_HIGHMEM 6124769843bSIngo Molnar if (highmem_pages >= max_pfn) { 6134769843bSIngo Molnar printk(KERN_ERR MSG_HIGHMEM_TOO_BIG, 6144769843bSIngo Molnar pages_to_mb(highmem_pages), pages_to_mb(max_pfn)); 6154769843bSIngo Molnar highmem_pages = 0; 6164769843bSIngo Molnar } 6174769843bSIngo Molnar if (highmem_pages) { 6184769843bSIngo Molnar if (max_low_pfn - highmem_pages < 64*1024*1024/PAGE_SIZE) { 6194769843bSIngo Molnar printk(KERN_ERR MSG_LOWMEM_TOO_SMALL, 6204769843bSIngo Molnar pages_to_mb(highmem_pages)); 6214769843bSIngo Molnar highmem_pages = 0; 6224769843bSIngo Molnar } 6234769843bSIngo Molnar max_low_pfn -= highmem_pages; 6244769843bSIngo Molnar } 6254769843bSIngo Molnar #else 6264769843bSIngo Molnar if (highmem_pages) 6274769843bSIngo Molnar printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n"); 6284769843bSIngo Molnar #endif 6294769843bSIngo Molnar } 6304769843bSIngo Molnar 6314769843bSIngo Molnar #define MSG_HIGHMEM_TOO_SMALL \ 6324769843bSIngo Molnar "only %luMB highmem pages available, ignoring highmem size of %luMB!\n" 6334769843bSIngo Molnar 6344769843bSIngo Molnar #define MSG_HIGHMEM_TRIMMED \ 6354769843bSIngo Molnar "Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n" 6364769843bSIngo Molnar /* 6374769843bSIngo Molnar * We have more RAM than fits into lowmem - we try to put it into 6384769843bSIngo Molnar * highmem, also taking the highmem=x boot parameter into account: 6394769843bSIngo Molnar */ 6404769843bSIngo Molnar void __init highmem_pfn_init(void) 6414769843bSIngo Molnar { 642d88316c2SIngo Molnar max_low_pfn = MAXMEM_PFN; 643d88316c2SIngo Molnar 64490d967e0SYinghai Lu if (highmem_pages == -1) 64590d967e0SYinghai Lu highmem_pages = max_pfn - MAXMEM_PFN; 6464769843bSIngo Molnar 64790d967e0SYinghai Lu if (highmem_pages + MAXMEM_PFN < max_pfn) 64890d967e0SYinghai Lu max_pfn = MAXMEM_PFN + highmem_pages; 6494769843bSIngo Molnar 65090d967e0SYinghai Lu if (highmem_pages + MAXMEM_PFN > max_pfn) { 6514769843bSIngo Molnar printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL, 65290d967e0SYinghai Lu pages_to_mb(max_pfn - MAXMEM_PFN), 65390d967e0SYinghai Lu pages_to_mb(highmem_pages)); 65490d967e0SYinghai Lu highmem_pages = 0; 65590d967e0SYinghai Lu } 65690d967e0SYinghai Lu #ifndef CONFIG_HIGHMEM 65790d967e0SYinghai Lu /* Maximum memory usable is what is directly addressable */ 6584769843bSIngo Molnar printk(KERN_WARNING "Warning only %ldMB will be used.\n", MAXMEM>>20); 65990d967e0SYinghai Lu if (max_pfn > MAX_NONPAE_PFN) 6604769843bSIngo Molnar printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n"); 66190d967e0SYinghai Lu else 66290d967e0SYinghai Lu printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n"); 66390d967e0SYinghai Lu max_pfn = MAXMEM_PFN; 66490d967e0SYinghai Lu #else /* !CONFIG_HIGHMEM */ 66590d967e0SYinghai Lu #ifndef CONFIG_HIGHMEM64G 66690d967e0SYinghai Lu if (max_pfn > MAX_NONPAE_PFN) { 66790d967e0SYinghai Lu max_pfn = MAX_NONPAE_PFN; 6684769843bSIngo Molnar printk(KERN_WARNING MSG_HIGHMEM_TRIMMED); 66990d967e0SYinghai Lu } 67090d967e0SYinghai Lu #endif /* !CONFIG_HIGHMEM64G */ 67190d967e0SYinghai Lu #endif /* !CONFIG_HIGHMEM */ 67290d967e0SYinghai Lu } 6734769843bSIngo Molnar 67490d967e0SYinghai Lu /* 67590d967e0SYinghai Lu * Determine low and high memory ranges: 67690d967e0SYinghai Lu */ 67790d967e0SYinghai Lu void __init find_low_pfn_range(void) 67890d967e0SYinghai Lu { 67990d967e0SYinghai Lu /* it could update max_pfn */ 68090d967e0SYinghai Lu 681d88316c2SIngo Molnar if (max_pfn <= MAXMEM_PFN) 6824769843bSIngo Molnar lowmem_pfn_init(); 683d88316c2SIngo Molnar else 684d88316c2SIngo Molnar highmem_pfn_init(); 68590d967e0SYinghai Lu } 68690d967e0SYinghai Lu 687b2ac82a0SYinghai Lu #ifndef CONFIG_NEED_MULTIPLE_NODES 6888ee2debcSDavid Rientjes void __init initmem_init(unsigned long start_pfn, unsigned long end_pfn, 6898ee2debcSDavid Rientjes int acpi, int k8) 690b2ac82a0SYinghai Lu { 691b2ac82a0SYinghai Lu #ifdef CONFIG_HIGHMEM 692b2ac82a0SYinghai Lu highstart_pfn = highend_pfn = max_pfn; 693b2ac82a0SYinghai Lu if (max_pfn > max_low_pfn) 694b2ac82a0SYinghai Lu highstart_pfn = max_low_pfn; 695a9ce6bc1SYinghai Lu memblock_x86_register_active_regions(0, 0, highend_pfn); 696ed077b58SShaohua Li sparse_memory_present_with_active_regions(0); 697b2ac82a0SYinghai Lu printk(KERN_NOTICE "%ldMB HIGHMEM available.\n", 698b2ac82a0SYinghai Lu pages_to_mb(highend_pfn - highstart_pfn)); 699b2ac82a0SYinghai Lu num_physpages = highend_pfn; 700b2ac82a0SYinghai Lu high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1; 701b2ac82a0SYinghai Lu #else 702a9ce6bc1SYinghai Lu memblock_x86_register_active_regions(0, 0, max_low_pfn); 703ed077b58SShaohua Li sparse_memory_present_with_active_regions(0); 704b2ac82a0SYinghai Lu num_physpages = max_low_pfn; 705b2ac82a0SYinghai Lu high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1; 706b2ac82a0SYinghai Lu #endif 707b2ac82a0SYinghai Lu #ifdef CONFIG_FLATMEM 708b2ac82a0SYinghai Lu max_mapnr = num_physpages; 709b2ac82a0SYinghai Lu #endif 710dc16ecf7SJeremy Fitzhardinge __vmalloc_start_set = true; 711dc16ecf7SJeremy Fitzhardinge 712b2ac82a0SYinghai Lu printk(KERN_NOTICE "%ldMB LOWMEM available.\n", 713b2ac82a0SYinghai Lu pages_to_mb(max_low_pfn)); 714b2ac82a0SYinghai Lu 715b2ac82a0SYinghai Lu setup_bootmem_allocator(); 716b2ac82a0SYinghai Lu } 717cb95a13aSYinghai Lu #endif /* !CONFIG_NEED_MULTIPLE_NODES */ 718b2ac82a0SYinghai Lu 719cb95a13aSYinghai Lu static void __init zone_sizes_init(void) 720b2ac82a0SYinghai Lu { 721b2ac82a0SYinghai Lu unsigned long max_zone_pfns[MAX_NR_ZONES]; 722b2ac82a0SYinghai Lu memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); 723b2ac82a0SYinghai Lu max_zone_pfns[ZONE_DMA] = 724b2ac82a0SYinghai Lu virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT; 725b2ac82a0SYinghai Lu max_zone_pfns[ZONE_NORMAL] = max_low_pfn; 726b2ac82a0SYinghai Lu #ifdef CONFIG_HIGHMEM 727b2ac82a0SYinghai Lu max_zone_pfns[ZONE_HIGHMEM] = highend_pfn; 728b2ac82a0SYinghai Lu #endif 729b2ac82a0SYinghai Lu 730b2ac82a0SYinghai Lu free_area_init_nodes(max_zone_pfns); 731b2ac82a0SYinghai Lu } 732b2ac82a0SYinghai Lu 733b2ac82a0SYinghai Lu void __init setup_bootmem_allocator(void) 734b2ac82a0SYinghai Lu { 735b2ac82a0SYinghai Lu printk(KERN_INFO " mapped low ram: 0 - %08lx\n", 736b2ac82a0SYinghai Lu max_pfn_mapped<<PAGE_SHIFT); 737fc5efe39SYinghai Lu printk(KERN_INFO " low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT); 738b2ac82a0SYinghai Lu 739c464573cSPekka Enberg after_bootmem = 1; 7404e29684cSYinghai Lu } 7414e29684cSYinghai Lu 742a04ad82dSYinghai Lu /* 743ad757b6aSThomas Gleixner * paging_init() sets up the page tables - note that the first 8MB are 744ad757b6aSThomas Gleixner * already mapped by head.S. 745ad757b6aSThomas Gleixner * 746ad757b6aSThomas Gleixner * This routines also unmaps the page at virtual kernel address 0, so 747ad757b6aSThomas Gleixner * that we can trap those pesky NULL-reference errors in the kernel. 748ad757b6aSThomas Gleixner */ 749ad757b6aSThomas Gleixner void __init paging_init(void) 750ad757b6aSThomas Gleixner { 751ad757b6aSThomas Gleixner pagetable_init(); 752ad757b6aSThomas Gleixner 753ad757b6aSThomas Gleixner __flush_tlb_all(); 754ad757b6aSThomas Gleixner 755ad757b6aSThomas Gleixner kmap_init(); 75611cd0bc1SYinghai Lu 75711cd0bc1SYinghai Lu /* 75811cd0bc1SYinghai Lu * NOTE: at this point the bootmem allocator is fully available. 75911cd0bc1SYinghai Lu */ 76011cd0bc1SYinghai Lu sparse_init(); 76111cd0bc1SYinghai Lu zone_sizes_init(); 762ad757b6aSThomas Gleixner } 763ad757b6aSThomas Gleixner 764ad757b6aSThomas Gleixner /* 765ad757b6aSThomas Gleixner * Test if the WP bit works in supervisor mode. It isn't supported on 386's 766f7f17a67SDmitri Vorobiev * and also on some strange 486's. All 586+'s are OK. This used to involve 767f7f17a67SDmitri Vorobiev * black magic jumps to work around some nasty CPU bugs, but fortunately the 768f7f17a67SDmitri Vorobiev * switch to using exceptions got rid of all that. 769ad757b6aSThomas Gleixner */ 770ad757b6aSThomas Gleixner static void __init test_wp_bit(void) 771ad757b6aSThomas Gleixner { 772d7d119d7SIngo Molnar printk(KERN_INFO 773d7d119d7SIngo Molnar "Checking if this processor honours the WP bit even in supervisor mode..."); 774ad757b6aSThomas Gleixner 775ad757b6aSThomas Gleixner /* Any page-aligned address will do, the test is non-destructive */ 776ad757b6aSThomas Gleixner __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY); 777ad757b6aSThomas Gleixner boot_cpu_data.wp_works_ok = do_test_wp_bit(); 778ad757b6aSThomas Gleixner clear_fixmap(FIX_WP_TEST); 779ad757b6aSThomas Gleixner 780ad757b6aSThomas Gleixner if (!boot_cpu_data.wp_works_ok) { 781d7d119d7SIngo Molnar printk(KERN_CONT "No.\n"); 782ad757b6aSThomas Gleixner #ifdef CONFIG_X86_WP_WORKS_OK 783d7d119d7SIngo Molnar panic( 784d7d119d7SIngo Molnar "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!"); 785ad757b6aSThomas Gleixner #endif 786ad757b6aSThomas Gleixner } else { 787d7d119d7SIngo Molnar printk(KERN_CONT "Ok.\n"); 788ad757b6aSThomas Gleixner } 789ad757b6aSThomas Gleixner } 790ad757b6aSThomas Gleixner 791ad757b6aSThomas Gleixner void __init mem_init(void) 792ad757b6aSThomas Gleixner { 793ad757b6aSThomas Gleixner int codesize, reservedpages, datasize, initsize; 794cc9f7a0cSYinghai Lu int tmp; 795ad757b6aSThomas Gleixner 796cfb80c9eSJeremy Fitzhardinge pci_iommu_alloc(); 797cfb80c9eSJeremy Fitzhardinge 798ad757b6aSThomas Gleixner #ifdef CONFIG_FLATMEM 799ad757b6aSThomas Gleixner BUG_ON(!mem_map); 800ad757b6aSThomas Gleixner #endif 801ad757b6aSThomas Gleixner /* this will put all low memory onto the freelists */ 802ad757b6aSThomas Gleixner totalram_pages += free_all_bootmem(); 803ad757b6aSThomas Gleixner 804ad757b6aSThomas Gleixner reservedpages = 0; 805ad757b6aSThomas Gleixner for (tmp = 0; tmp < max_low_pfn; tmp++) 806ad757b6aSThomas Gleixner /* 8078550eb99SIngo Molnar * Only count reserved RAM pages: 808ad757b6aSThomas Gleixner */ 809ad757b6aSThomas Gleixner if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp))) 810ad757b6aSThomas Gleixner reservedpages++; 811ad757b6aSThomas Gleixner 812cc9f7a0cSYinghai Lu set_highmem_pages_init(); 813ad757b6aSThomas Gleixner 814ad757b6aSThomas Gleixner codesize = (unsigned long) &_etext - (unsigned long) &_text; 815ad757b6aSThomas Gleixner datasize = (unsigned long) &_edata - (unsigned long) &_etext; 816ad757b6aSThomas Gleixner initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin; 817ad757b6aSThomas Gleixner 8188550eb99SIngo Molnar printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, " 8198550eb99SIngo Molnar "%dk reserved, %dk data, %dk init, %ldk highmem)\n", 820cc013a88SGeert Uytterhoeven nr_free_pages() << (PAGE_SHIFT-10), 821ad757b6aSThomas Gleixner num_physpages << (PAGE_SHIFT-10), 822ad757b6aSThomas Gleixner codesize >> 10, 823ad757b6aSThomas Gleixner reservedpages << (PAGE_SHIFT-10), 824ad757b6aSThomas Gleixner datasize >> 10, 825ad757b6aSThomas Gleixner initsize >> 10, 8264b529401SAndreas Fenkart totalhigh_pages << (PAGE_SHIFT-10)); 827ad757b6aSThomas Gleixner 828d7d119d7SIngo Molnar printk(KERN_INFO "virtual kernel memory layout:\n" 829ad757b6aSThomas Gleixner " fixmap : 0x%08lx - 0x%08lx (%4ld kB)\n" 830ad757b6aSThomas Gleixner #ifdef CONFIG_HIGHMEM 831ad757b6aSThomas Gleixner " pkmap : 0x%08lx - 0x%08lx (%4ld kB)\n" 832ad757b6aSThomas Gleixner #endif 833ad757b6aSThomas Gleixner " vmalloc : 0x%08lx - 0x%08lx (%4ld MB)\n" 834ad757b6aSThomas Gleixner " lowmem : 0x%08lx - 0x%08lx (%4ld MB)\n" 835ad757b6aSThomas Gleixner " .init : 0x%08lx - 0x%08lx (%4ld kB)\n" 836ad757b6aSThomas Gleixner " .data : 0x%08lx - 0x%08lx (%4ld kB)\n" 837ad757b6aSThomas Gleixner " .text : 0x%08lx - 0x%08lx (%4ld kB)\n", 838ad757b6aSThomas Gleixner FIXADDR_START, FIXADDR_TOP, 839ad757b6aSThomas Gleixner (FIXADDR_TOP - FIXADDR_START) >> 10, 840ad757b6aSThomas Gleixner 841ad757b6aSThomas Gleixner #ifdef CONFIG_HIGHMEM 842ad757b6aSThomas Gleixner PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE, 843ad757b6aSThomas Gleixner (LAST_PKMAP*PAGE_SIZE) >> 10, 844ad757b6aSThomas Gleixner #endif 845ad757b6aSThomas Gleixner 846ad757b6aSThomas Gleixner VMALLOC_START, VMALLOC_END, 847ad757b6aSThomas Gleixner (VMALLOC_END - VMALLOC_START) >> 20, 848ad757b6aSThomas Gleixner 849ad757b6aSThomas Gleixner (unsigned long)__va(0), (unsigned long)high_memory, 850ad757b6aSThomas Gleixner ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20, 851ad757b6aSThomas Gleixner 852ad757b6aSThomas Gleixner (unsigned long)&__init_begin, (unsigned long)&__init_end, 8538550eb99SIngo Molnar ((unsigned long)&__init_end - 8548550eb99SIngo Molnar (unsigned long)&__init_begin) >> 10, 855ad757b6aSThomas Gleixner 856ad757b6aSThomas Gleixner (unsigned long)&_etext, (unsigned long)&_edata, 857ad757b6aSThomas Gleixner ((unsigned long)&_edata - (unsigned long)&_etext) >> 10, 858ad757b6aSThomas Gleixner 859ad757b6aSThomas Gleixner (unsigned long)&_text, (unsigned long)&_etext, 860ad757b6aSThomas Gleixner ((unsigned long)&_etext - (unsigned long)&_text) >> 10); 861ad757b6aSThomas Gleixner 862beeb4195SJan Beulich /* 863beeb4195SJan Beulich * Check boundaries twice: Some fundamental inconsistencies can 864beeb4195SJan Beulich * be detected at build time already. 865beeb4195SJan Beulich */ 866beeb4195SJan Beulich #define __FIXADDR_TOP (-PAGE_SIZE) 867beeb4195SJan Beulich #ifdef CONFIG_HIGHMEM 868beeb4195SJan Beulich BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START); 869beeb4195SJan Beulich BUILD_BUG_ON(VMALLOC_END > PKMAP_BASE); 870beeb4195SJan Beulich #endif 871beeb4195SJan Beulich #define high_memory (-128UL << 20) 872beeb4195SJan Beulich BUILD_BUG_ON(VMALLOC_START >= VMALLOC_END); 873beeb4195SJan Beulich #undef high_memory 874beeb4195SJan Beulich #undef __FIXADDR_TOP 875beeb4195SJan Beulich 876ad757b6aSThomas Gleixner #ifdef CONFIG_HIGHMEM 877ad757b6aSThomas Gleixner BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START); 878ad757b6aSThomas Gleixner BUG_ON(VMALLOC_END > PKMAP_BASE); 879ad757b6aSThomas Gleixner #endif 880beeb4195SJan Beulich BUG_ON(VMALLOC_START >= VMALLOC_END); 881ad757b6aSThomas Gleixner BUG_ON((unsigned long)high_memory > VMALLOC_START); 882ad757b6aSThomas Gleixner 883ad757b6aSThomas Gleixner if (boot_cpu_data.wp_works_ok < 0) 884ad757b6aSThomas Gleixner test_wp_bit(); 885ad757b6aSThomas Gleixner 88661165d7aSHugh Dickins save_pg_dir(); 88755cd6367SYinghai Lu zap_low_mappings(true); 888ad757b6aSThomas Gleixner } 889ad757b6aSThomas Gleixner 890ad757b6aSThomas Gleixner #ifdef CONFIG_MEMORY_HOTPLUG 891ad757b6aSThomas Gleixner int arch_add_memory(int nid, u64 start, u64 size) 892ad757b6aSThomas Gleixner { 893ad757b6aSThomas Gleixner struct pglist_data *pgdata = NODE_DATA(nid); 894ad757b6aSThomas Gleixner struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM; 895ad757b6aSThomas Gleixner unsigned long start_pfn = start >> PAGE_SHIFT; 896ad757b6aSThomas Gleixner unsigned long nr_pages = size >> PAGE_SHIFT; 897ad757b6aSThomas Gleixner 898c04fc586SGary Hade return __add_pages(nid, zone, start_pfn, nr_pages); 899ad757b6aSThomas Gleixner } 900ad757b6aSThomas Gleixner #endif 901ad757b6aSThomas Gleixner 902ad757b6aSThomas Gleixner /* 903ad757b6aSThomas Gleixner * This function cannot be __init, since exceptions don't work in that 904ad757b6aSThomas Gleixner * section. Put this after the callers, so that it cannot be inlined. 905ad757b6aSThomas Gleixner */ 9068550eb99SIngo Molnar static noinline int do_test_wp_bit(void) 907ad757b6aSThomas Gleixner { 908ad757b6aSThomas Gleixner char tmp_reg; 909ad757b6aSThomas Gleixner int flag; 910ad757b6aSThomas Gleixner 911ad757b6aSThomas Gleixner __asm__ __volatile__( 912ad757b6aSThomas Gleixner " movb %0, %1 \n" 913ad757b6aSThomas Gleixner "1: movb %1, %0 \n" 914ad757b6aSThomas Gleixner " xorl %2, %2 \n" 915ad757b6aSThomas Gleixner "2: \n" 916f832ff18SH. Peter Anvin _ASM_EXTABLE(1b,2b) 917ad757b6aSThomas Gleixner :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)), 918ad757b6aSThomas Gleixner "=q" (tmp_reg), 919ad757b6aSThomas Gleixner "=r" (flag) 920ad757b6aSThomas Gleixner :"2" (1) 921ad757b6aSThomas Gleixner :"memory"); 922ad757b6aSThomas Gleixner 923ad757b6aSThomas Gleixner return flag; 924ad757b6aSThomas Gleixner } 925ad757b6aSThomas Gleixner 926ad757b6aSThomas Gleixner #ifdef CONFIG_DEBUG_RODATA 927edeed305SArjan van de Ven const int rodata_test_data = 0xC3; 928edeed305SArjan van de Ven EXPORT_SYMBOL_GPL(rodata_test_data); 929ad757b6aSThomas Gleixner 930502f6604SSuresh Siddha int kernel_set_to_readonly __read_mostly; 93116239630SSteven Rostedt 93216239630SSteven Rostedt void set_kernel_text_rw(void) 93316239630SSteven Rostedt { 93416239630SSteven Rostedt unsigned long start = PFN_ALIGN(_text); 93516239630SSteven Rostedt unsigned long size = PFN_ALIGN(_etext) - start; 93616239630SSteven Rostedt 93716239630SSteven Rostedt if (!kernel_set_to_readonly) 93816239630SSteven Rostedt return; 93916239630SSteven Rostedt 94016239630SSteven Rostedt pr_debug("Set kernel text: %lx - %lx for read write\n", 94116239630SSteven Rostedt start, start+size); 94216239630SSteven Rostedt 94316239630SSteven Rostedt set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT); 94416239630SSteven Rostedt } 94516239630SSteven Rostedt 94616239630SSteven Rostedt void set_kernel_text_ro(void) 94716239630SSteven Rostedt { 94816239630SSteven Rostedt unsigned long start = PFN_ALIGN(_text); 94916239630SSteven Rostedt unsigned long size = PFN_ALIGN(_etext) - start; 95016239630SSteven Rostedt 95116239630SSteven Rostedt if (!kernel_set_to_readonly) 95216239630SSteven Rostedt return; 95316239630SSteven Rostedt 95416239630SSteven Rostedt pr_debug("Set kernel text: %lx - %lx for read only\n", 95516239630SSteven Rostedt start, start+size); 95616239630SSteven Rostedt 95716239630SSteven Rostedt set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT); 95816239630SSteven Rostedt } 95916239630SSteven Rostedt 960ad757b6aSThomas Gleixner void mark_rodata_ro(void) 961ad757b6aSThomas Gleixner { 962ad757b6aSThomas Gleixner unsigned long start = PFN_ALIGN(_text); 963ad757b6aSThomas Gleixner unsigned long size = PFN_ALIGN(_etext) - start; 964ad757b6aSThomas Gleixner 9656d238cc4SArjan van de Ven set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT); 966d7d119d7SIngo Molnar printk(KERN_INFO "Write protecting the kernel text: %luk\n", 967d7d119d7SIngo Molnar size >> 10); 9680c42f392SAndi Kleen 96916239630SSteven Rostedt kernel_set_to_readonly = 1; 97016239630SSteven Rostedt 9710c42f392SAndi Kleen #ifdef CONFIG_CPA_DEBUG 972d7d119d7SIngo Molnar printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n", 973d7d119d7SIngo Molnar start, start+size); 9746d238cc4SArjan van de Ven set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT); 9750c42f392SAndi Kleen 976d7d119d7SIngo Molnar printk(KERN_INFO "Testing CPA: write protecting again\n"); 9776d238cc4SArjan van de Ven set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT); 9780c42f392SAndi Kleen #endif 9798f0f996eSSteven Rostedt 980ad757b6aSThomas Gleixner start += size; 981ad757b6aSThomas Gleixner size = (unsigned long)__end_rodata - start; 9826d238cc4SArjan van de Ven set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT); 983d7d119d7SIngo Molnar printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n", 984ad757b6aSThomas Gleixner size >> 10); 985edeed305SArjan van de Ven rodata_test(); 986ad757b6aSThomas Gleixner 9870c42f392SAndi Kleen #ifdef CONFIG_CPA_DEBUG 988d7d119d7SIngo Molnar printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size); 9896d238cc4SArjan van de Ven set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT); 9900c42f392SAndi Kleen 991d7d119d7SIngo Molnar printk(KERN_INFO "Testing CPA: write protecting again\n"); 9926d238cc4SArjan van de Ven set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT); 9930c42f392SAndi Kleen #endif 994ad757b6aSThomas Gleixner } 995ad757b6aSThomas Gleixner #endif 996a9ce6bc1SYinghai Lu 997