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/signal.h> 9ad757b6aSThomas Gleixner #include <linux/sched.h> 10ad757b6aSThomas Gleixner #include <linux/kernel.h> 11ad757b6aSThomas Gleixner #include <linux/errno.h> 12ad757b6aSThomas Gleixner #include <linux/string.h> 13ad757b6aSThomas Gleixner #include <linux/types.h> 14ad757b6aSThomas Gleixner #include <linux/ptrace.h> 15ad757b6aSThomas Gleixner #include <linux/mman.h> 16ad757b6aSThomas Gleixner #include <linux/mm.h> 17ad757b6aSThomas Gleixner #include <linux/hugetlb.h> 18ad757b6aSThomas Gleixner #include <linux/swap.h> 19ad757b6aSThomas Gleixner #include <linux/smp.h> 20ad757b6aSThomas Gleixner #include <linux/init.h> 21ad757b6aSThomas Gleixner #include <linux/highmem.h> 22ad757b6aSThomas Gleixner #include <linux/pagemap.h> 23cfb80c9eSJeremy Fitzhardinge #include <linux/pci.h> 24ad757b6aSThomas Gleixner #include <linux/pfn.h> 25ad757b6aSThomas Gleixner #include <linux/poison.h> 26a9ce6bc1SYinghai Lu #include <linux/memblock.h> 27ad757b6aSThomas Gleixner #include <linux/proc_fs.h> 28ad757b6aSThomas Gleixner #include <linux/memory_hotplug.h> 29ad757b6aSThomas Gleixner #include <linux/initrd.h> 30ad757b6aSThomas Gleixner #include <linux/cpumask.h> 315a0e3ad6STejun Heo #include <linux/gfp.h> 32ad757b6aSThomas Gleixner 33f832ff18SH. Peter Anvin #include <asm/asm.h> 3446eaa670SIngo Molnar #include <asm/bios_ebda.h> 35ad757b6aSThomas Gleixner #include <asm/processor.h> 367c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 37ad757b6aSThomas Gleixner #include <asm/pgtable.h> 38ad757b6aSThomas Gleixner #include <asm/dma.h> 39ad757b6aSThomas Gleixner #include <asm/fixmap.h> 4066441bd3SIngo Molnar #include <asm/e820/api.h> 41ad757b6aSThomas Gleixner #include <asm/apic.h> 428550eb99SIngo Molnar #include <asm/bugs.h> 43ad757b6aSThomas Gleixner #include <asm/tlb.h> 44ad757b6aSThomas Gleixner #include <asm/tlbflush.h> 45c10d1e26SAndres Salomon #include <asm/olpc_ofw.h> 46a5a19c63SJeremy Fitzhardinge #include <asm/pgalloc.h> 47ad757b6aSThomas Gleixner #include <asm/sections.h> 48ad757b6aSThomas Gleixner #include <asm/paravirt.h> 49551889a6SIan Campbell #include <asm/setup.h> 50d1163651SLaura Abbott #include <asm/set_memory.h> 512b72394eSPekka Enberg #include <asm/page_types.h> 5292a0f81dSThomas Gleixner #include <asm/cpu_entry_area.h> 534fcb2083SPekka Enberg #include <asm/init.h> 54ad757b6aSThomas Gleixner 555c51bdbeSYinghai Lu #include "mm_internal.h" 565c51bdbeSYinghai Lu 57ad757b6aSThomas Gleixner unsigned long highstart_pfn, highend_pfn; 58ad757b6aSThomas Gleixner 59dc16ecf7SJeremy Fitzhardinge bool __read_mostly __vmalloc_start_set = false; 604e29684cSYinghai Lu 61ad757b6aSThomas Gleixner /* 62ad757b6aSThomas Gleixner * Creates a middle page table and puts a pointer to it in the 63ad757b6aSThomas Gleixner * given global directory entry. This only returns the gd entry 64ad757b6aSThomas Gleixner * in non-PAE compilation mode, since the middle layer is folded. 65ad757b6aSThomas Gleixner */ 66ad757b6aSThomas Gleixner static pmd_t * __init one_md_table_init(pgd_t *pgd) 67ad757b6aSThomas Gleixner { 68e0c4f675SKirill A. Shutemov p4d_t *p4d; 69ad757b6aSThomas Gleixner pud_t *pud; 70ad757b6aSThomas Gleixner pmd_t *pmd_table; 71ad757b6aSThomas Gleixner 72ad757b6aSThomas Gleixner #ifdef CONFIG_X86_PAE 73ad757b6aSThomas Gleixner if (!(pgd_val(*pgd) & _PAGE_PRESENT)) { 74d6be89adSJan Beulich pmd_table = (pmd_t *)alloc_low_page(); 756944a9c8SJeremy Fitzhardinge paravirt_alloc_pmd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT); 76ad757b6aSThomas Gleixner set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT)); 77e0c4f675SKirill A. Shutemov p4d = p4d_offset(pgd, 0); 78e0c4f675SKirill A. Shutemov pud = pud_offset(p4d, 0); 798550eb99SIngo Molnar BUG_ON(pmd_table != pmd_offset(pud, 0)); 80a376f30aSZhaolei 81a376f30aSZhaolei return pmd_table; 82ad757b6aSThomas Gleixner } 83ad757b6aSThomas Gleixner #endif 84e0c4f675SKirill A. Shutemov p4d = p4d_offset(pgd, 0); 85e0c4f675SKirill A. Shutemov pud = pud_offset(p4d, 0); 86ad757b6aSThomas Gleixner pmd_table = pmd_offset(pud, 0); 878550eb99SIngo Molnar 88ad757b6aSThomas Gleixner return pmd_table; 89ad757b6aSThomas Gleixner } 90ad757b6aSThomas Gleixner 91ad757b6aSThomas Gleixner /* 92ad757b6aSThomas Gleixner * Create a page table and place a pointer to it in a middle page 938550eb99SIngo Molnar * directory entry: 94ad757b6aSThomas Gleixner */ 95ad757b6aSThomas Gleixner static pte_t * __init one_page_table_init(pmd_t *pmd) 96ad757b6aSThomas Gleixner { 97ad757b6aSThomas Gleixner if (!(pmd_val(*pmd) & _PAGE_PRESENT)) { 984e37a890SYinghai Lu pte_t *page_table = (pte_t *)alloc_low_page(); 99ad757b6aSThomas Gleixner 1006944a9c8SJeremy Fitzhardinge paravirt_alloc_pte(&init_mm, __pa(page_table) >> PAGE_SHIFT); 101ad757b6aSThomas Gleixner set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE)); 102ad757b6aSThomas Gleixner BUG_ON(page_table != pte_offset_kernel(pmd, 0)); 103ad757b6aSThomas Gleixner } 104ad757b6aSThomas Gleixner 105ad757b6aSThomas Gleixner return pte_offset_kernel(pmd, 0); 106ad757b6aSThomas Gleixner } 107ad757b6aSThomas Gleixner 108458a3e64STejun Heo pmd_t * __init populate_extra_pmd(unsigned long vaddr) 10911124411STejun Heo { 11011124411STejun Heo int pgd_idx = pgd_index(vaddr); 11111124411STejun Heo int pmd_idx = pmd_index(vaddr); 112458a3e64STejun Heo 113458a3e64STejun Heo return one_md_table_init(swapper_pg_dir + pgd_idx) + pmd_idx; 114458a3e64STejun Heo } 115458a3e64STejun Heo 116458a3e64STejun Heo pte_t * __init populate_extra_pte(unsigned long vaddr) 117458a3e64STejun Heo { 118458a3e64STejun Heo int pte_idx = pte_index(vaddr); 11911124411STejun Heo pmd_t *pmd; 12011124411STejun Heo 121458a3e64STejun Heo pmd = populate_extra_pmd(vaddr); 122458a3e64STejun Heo return one_page_table_init(pmd) + pte_idx; 12311124411STejun Heo } 12411124411STejun Heo 125719272c4SYinghai Lu static unsigned long __init 126719272c4SYinghai Lu page_table_range_init_count(unsigned long start, unsigned long end) 127719272c4SYinghai Lu { 128719272c4SYinghai Lu unsigned long count = 0; 129719272c4SYinghai Lu #ifdef CONFIG_HIGHMEM 130719272c4SYinghai Lu int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT; 131719272c4SYinghai Lu int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT; 132719272c4SYinghai Lu int pgd_idx, pmd_idx; 133719272c4SYinghai Lu unsigned long vaddr; 134719272c4SYinghai Lu 135719272c4SYinghai Lu if (pmd_idx_kmap_begin == pmd_idx_kmap_end) 136719272c4SYinghai Lu return 0; 137719272c4SYinghai Lu 138719272c4SYinghai Lu vaddr = start; 139719272c4SYinghai Lu pgd_idx = pgd_index(vaddr); 1409962eea9SMinfei Huang pmd_idx = pmd_index(vaddr); 141719272c4SYinghai Lu 142719272c4SYinghai Lu for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd_idx++) { 143719272c4SYinghai Lu for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end); 144719272c4SYinghai Lu pmd_idx++) { 145719272c4SYinghai Lu if ((vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin && 146719272c4SYinghai Lu (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end) 147719272c4SYinghai Lu count++; 148719272c4SYinghai Lu vaddr += PMD_SIZE; 149719272c4SYinghai Lu } 150719272c4SYinghai Lu pmd_idx = 0; 151719272c4SYinghai Lu } 152719272c4SYinghai Lu #endif 153719272c4SYinghai Lu return count; 154719272c4SYinghai Lu } 155719272c4SYinghai Lu 156a3c6018eSJan Beulich static pte_t *__init page_table_kmap_check(pte_t *pte, pmd_t *pmd, 157719272c4SYinghai Lu unsigned long vaddr, pte_t *lastpte, 158719272c4SYinghai Lu void **adr) 159a3c6018eSJan Beulich { 160a3c6018eSJan Beulich #ifdef CONFIG_HIGHMEM 161a3c6018eSJan Beulich /* 162a3c6018eSJan Beulich * Something (early fixmap) may already have put a pte 163a3c6018eSJan Beulich * page here, which causes the page table allocation 164a3c6018eSJan Beulich * to become nonlinear. Attempt to fix it, and if it 165a3c6018eSJan Beulich * is still nonlinear then we have to bug. 166a3c6018eSJan Beulich */ 167a3c6018eSJan Beulich int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT; 168a3c6018eSJan Beulich int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT; 169a3c6018eSJan Beulich 170a3c6018eSJan Beulich if (pmd_idx_kmap_begin != pmd_idx_kmap_end 171a3c6018eSJan Beulich && (vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin 172719272c4SYinghai Lu && (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end) { 173a3c6018eSJan Beulich pte_t *newpte; 174a3c6018eSJan Beulich int i; 175a3c6018eSJan Beulich 176c464573cSPekka Enberg BUG_ON(after_bootmem); 177719272c4SYinghai Lu newpte = *adr; 178a3c6018eSJan Beulich for (i = 0; i < PTRS_PER_PTE; i++) 179a3c6018eSJan Beulich set_pte(newpte + i, pte[i]); 180719272c4SYinghai Lu *adr = (void *)(((unsigned long)(*adr)) + PAGE_SIZE); 181a3c6018eSJan Beulich 182a3c6018eSJan Beulich paravirt_alloc_pte(&init_mm, __pa(newpte) >> PAGE_SHIFT); 183a3c6018eSJan Beulich set_pmd(pmd, __pmd(__pa(newpte)|_PAGE_TABLE)); 184a3c6018eSJan Beulich BUG_ON(newpte != pte_offset_kernel(pmd, 0)); 185a3c6018eSJan Beulich __flush_tlb_all(); 186a3c6018eSJan Beulich 187a3c6018eSJan Beulich paravirt_release_pte(__pa(pte) >> PAGE_SHIFT); 188a3c6018eSJan Beulich pte = newpte; 189a3c6018eSJan Beulich } 190a3c6018eSJan Beulich BUG_ON(vaddr < fix_to_virt(FIX_KMAP_BEGIN - 1) 191a3c6018eSJan Beulich && vaddr > fix_to_virt(FIX_KMAP_END) 192a3c6018eSJan Beulich && lastpte && lastpte + PTRS_PER_PTE != pte); 193a3c6018eSJan Beulich #endif 194a3c6018eSJan Beulich return pte; 195a3c6018eSJan Beulich } 196a3c6018eSJan Beulich 197ad757b6aSThomas Gleixner /* 198ad757b6aSThomas Gleixner * This function initializes a certain range of kernel virtual memory 199ad757b6aSThomas Gleixner * with new bootmem page tables, everywhere page tables are missing in 200ad757b6aSThomas Gleixner * the given range. 2018550eb99SIngo Molnar * 202ad757b6aSThomas Gleixner * NOTE: The pagetables are allocated contiguous on the physical space 203ad757b6aSThomas Gleixner * so we can cache the place of the first one and move around without 204ad757b6aSThomas Gleixner * checking the pgd every time. 205ad757b6aSThomas Gleixner */ 2068550eb99SIngo Molnar static void __init 2078550eb99SIngo Molnar page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base) 208ad757b6aSThomas Gleixner { 209ad757b6aSThomas Gleixner int pgd_idx, pmd_idx; 210ad757b6aSThomas Gleixner unsigned long vaddr; 2118550eb99SIngo Molnar pgd_t *pgd; 2128550eb99SIngo Molnar pmd_t *pmd; 213a3c6018eSJan Beulich pte_t *pte = NULL; 214719272c4SYinghai Lu unsigned long count = page_table_range_init_count(start, end); 215719272c4SYinghai Lu void *adr = NULL; 216719272c4SYinghai Lu 217719272c4SYinghai Lu if (count) 218719272c4SYinghai Lu adr = alloc_low_pages(count); 219ad757b6aSThomas Gleixner 220ad757b6aSThomas Gleixner vaddr = start; 221ad757b6aSThomas Gleixner pgd_idx = pgd_index(vaddr); 222ad757b6aSThomas Gleixner pmd_idx = pmd_index(vaddr); 223ad757b6aSThomas Gleixner pgd = pgd_base + pgd_idx; 224ad757b6aSThomas Gleixner 225ad757b6aSThomas Gleixner for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) { 226ad757b6aSThomas Gleixner pmd = one_md_table_init(pgd); 227ad757b6aSThomas Gleixner pmd = pmd + pmd_index(vaddr); 2288550eb99SIngo Molnar for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end); 2298550eb99SIngo Molnar pmd++, pmd_idx++) { 230a3c6018eSJan Beulich pte = page_table_kmap_check(one_page_table_init(pmd), 231719272c4SYinghai Lu pmd, vaddr, pte, &adr); 232ad757b6aSThomas Gleixner 233ad757b6aSThomas Gleixner vaddr += PMD_SIZE; 234ad757b6aSThomas Gleixner } 235ad757b6aSThomas Gleixner pmd_idx = 0; 236ad757b6aSThomas Gleixner } 237ad757b6aSThomas Gleixner } 238ad757b6aSThomas Gleixner 239ad757b6aSThomas Gleixner static inline int is_kernel_text(unsigned long addr) 240ad757b6aSThomas Gleixner { 2415bd5a452SMatthieu Castet if (addr >= (unsigned long)_text && addr <= (unsigned long)__init_end) 242ad757b6aSThomas Gleixner return 1; 243ad757b6aSThomas Gleixner return 0; 244ad757b6aSThomas Gleixner } 245ad757b6aSThomas Gleixner 246ad757b6aSThomas Gleixner /* 247ad757b6aSThomas Gleixner * This maps the physical memory to kernel virtual address space, a total 248ad757b6aSThomas Gleixner * of max_low_pfn pages, by creating page tables starting from address 2498550eb99SIngo Molnar * PAGE_OFFSET: 250ad757b6aSThomas Gleixner */ 251e53fb04fSPekka Enberg unsigned long __init 252e53fb04fSPekka Enberg kernel_physical_mapping_init(unsigned long start, 253e53fb04fSPekka Enberg unsigned long end, 254e53fb04fSPekka Enberg unsigned long page_size_mask) 255ad757b6aSThomas Gleixner { 256e53fb04fSPekka Enberg int use_pse = page_size_mask == (1<<PG_LEVEL_2M); 257c1fd1b43SPekka Enberg unsigned long last_map_addr = end; 258e53fb04fSPekka Enberg unsigned long start_pfn, end_pfn; 259e7179853SPekka Enberg pgd_t *pgd_base = swapper_pg_dir; 2608550eb99SIngo Molnar int pgd_idx, pmd_idx, pte_ofs; 261ad757b6aSThomas Gleixner unsigned long pfn; 262ad757b6aSThomas Gleixner pgd_t *pgd; 263ad757b6aSThomas Gleixner pmd_t *pmd; 264ad757b6aSThomas Gleixner pte_t *pte; 265a2699e47SSuresh Siddha unsigned pages_2m, pages_4k; 266a2699e47SSuresh Siddha int mapping_iter; 267a2699e47SSuresh Siddha 268e53fb04fSPekka Enberg start_pfn = start >> PAGE_SHIFT; 269e53fb04fSPekka Enberg end_pfn = end >> PAGE_SHIFT; 270e53fb04fSPekka Enberg 271a2699e47SSuresh Siddha /* 272a2699e47SSuresh Siddha * First iteration will setup identity mapping using large/small pages 273a2699e47SSuresh Siddha * based on use_pse, with other attributes same as set by 274a2699e47SSuresh Siddha * the early code in head_32.S 275a2699e47SSuresh Siddha * 276a2699e47SSuresh Siddha * Second iteration will setup the appropriate attributes (NX, GLOBAL..) 277a2699e47SSuresh Siddha * as desired for the kernel identity mapping. 278a2699e47SSuresh Siddha * 279a2699e47SSuresh Siddha * This two pass mechanism conforms to the TLB app note which says: 280a2699e47SSuresh Siddha * 281a2699e47SSuresh Siddha * "Software should not write to a paging-structure entry in a way 282a2699e47SSuresh Siddha * that would change, for any linear address, both the page size 283a2699e47SSuresh Siddha * and either the page frame or attributes." 284a2699e47SSuresh Siddha */ 285a2699e47SSuresh Siddha mapping_iter = 1; 286ad757b6aSThomas Gleixner 28716bf9226SBorislav Petkov if (!boot_cpu_has(X86_FEATURE_PSE)) 288a04ad82dSYinghai Lu use_pse = 0; 289a04ad82dSYinghai Lu 290a2699e47SSuresh Siddha repeat: 291a2699e47SSuresh Siddha pages_2m = pages_4k = 0; 292a04ad82dSYinghai Lu pfn = start_pfn; 293a04ad82dSYinghai Lu pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET); 294ad757b6aSThomas Gleixner pgd = pgd_base + pgd_idx; 295ad757b6aSThomas Gleixner for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) { 296ad757b6aSThomas Gleixner pmd = one_md_table_init(pgd); 2978550eb99SIngo Molnar 298a04ad82dSYinghai Lu if (pfn >= end_pfn) 299a04ad82dSYinghai Lu continue; 300a04ad82dSYinghai Lu #ifdef CONFIG_X86_PAE 301a04ad82dSYinghai Lu pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET); 302a04ad82dSYinghai Lu pmd += pmd_idx; 303a04ad82dSYinghai Lu #else 304a04ad82dSYinghai Lu pmd_idx = 0; 305a04ad82dSYinghai Lu #endif 306a04ad82dSYinghai Lu for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn; 307f3f20de8SJeremy Fitzhardinge pmd++, pmd_idx++) { 3088550eb99SIngo Molnar unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET; 309ad757b6aSThomas Gleixner 3108550eb99SIngo Molnar /* 3118550eb99SIngo Molnar * Map with big pages if possible, otherwise 3128550eb99SIngo Molnar * create normal page tables: 3138550eb99SIngo Molnar */ 314a04ad82dSYinghai Lu if (use_pse) { 3158550eb99SIngo Molnar unsigned int addr2; 316f3f20de8SJeremy Fitzhardinge pgprot_t prot = PAGE_KERNEL_LARGE; 317a2699e47SSuresh Siddha /* 318a2699e47SSuresh Siddha * first pass will use the same initial 319a2699e47SSuresh Siddha * identity mapping attribute + _PAGE_PSE. 320a2699e47SSuresh Siddha */ 321a2699e47SSuresh Siddha pgprot_t init_prot = 322a2699e47SSuresh Siddha __pgprot(PTE_IDENT_ATTR | 323a2699e47SSuresh Siddha _PAGE_PSE); 324f3f20de8SJeremy Fitzhardinge 325960ddb4fSYinghai Lu pfn &= PMD_MASK >> PAGE_SHIFT; 3268550eb99SIngo Molnar addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE + 327f3f20de8SJeremy Fitzhardinge PAGE_OFFSET + PAGE_SIZE-1; 328f3f20de8SJeremy Fitzhardinge 3298550eb99SIngo Molnar if (is_kernel_text(addr) || 3308550eb99SIngo Molnar is_kernel_text(addr2)) 331f3f20de8SJeremy Fitzhardinge prot = PAGE_KERNEL_LARGE_EXEC; 332f3f20de8SJeremy Fitzhardinge 333ce0c0e50SAndi Kleen pages_2m++; 334a2699e47SSuresh Siddha if (mapping_iter == 1) 335a2699e47SSuresh Siddha set_pmd(pmd, pfn_pmd(pfn, init_prot)); 336a2699e47SSuresh Siddha else 337f3f20de8SJeremy Fitzhardinge set_pmd(pmd, pfn_pmd(pfn, prot)); 338ad757b6aSThomas Gleixner 339ad757b6aSThomas Gleixner pfn += PTRS_PER_PTE; 3408550eb99SIngo Molnar continue; 3418550eb99SIngo Molnar } 342ad757b6aSThomas Gleixner pte = one_page_table_init(pmd); 343ad757b6aSThomas Gleixner 344a04ad82dSYinghai Lu pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET); 345a04ad82dSYinghai Lu pte += pte_ofs; 346a04ad82dSYinghai Lu for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn; 3478550eb99SIngo Molnar pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) { 348f3f20de8SJeremy Fitzhardinge pgprot_t prot = PAGE_KERNEL; 349a2699e47SSuresh Siddha /* 350a2699e47SSuresh Siddha * first pass will use the same initial 351a2699e47SSuresh Siddha * identity mapping attribute. 352a2699e47SSuresh Siddha */ 353a2699e47SSuresh Siddha pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR); 354f3f20de8SJeremy Fitzhardinge 3558550eb99SIngo Molnar if (is_kernel_text(addr)) 356f3f20de8SJeremy Fitzhardinge prot = PAGE_KERNEL_EXEC; 357f3f20de8SJeremy Fitzhardinge 358ce0c0e50SAndi Kleen pages_4k++; 359c1fd1b43SPekka Enberg if (mapping_iter == 1) { 360a2699e47SSuresh Siddha set_pte(pte, pfn_pte(pfn, init_prot)); 361c1fd1b43SPekka Enberg last_map_addr = (pfn << PAGE_SHIFT) + PAGE_SIZE; 362c1fd1b43SPekka Enberg } else 363f3f20de8SJeremy Fitzhardinge set_pte(pte, pfn_pte(pfn, prot)); 364ad757b6aSThomas Gleixner } 365ad757b6aSThomas Gleixner } 366ad757b6aSThomas Gleixner } 367a2699e47SSuresh Siddha if (mapping_iter == 1) { 368a2699e47SSuresh Siddha /* 369a2699e47SSuresh Siddha * update direct mapping page count only in the first 370a2699e47SSuresh Siddha * iteration. 371a2699e47SSuresh Siddha */ 372ce0c0e50SAndi Kleen update_page_count(PG_LEVEL_2M, pages_2m); 373ce0c0e50SAndi Kleen update_page_count(PG_LEVEL_4K, pages_4k); 374a2699e47SSuresh Siddha 375a2699e47SSuresh Siddha /* 376a2699e47SSuresh Siddha * local global flush tlb, which will flush the previous 377a2699e47SSuresh Siddha * mappings present in both small and large page TLB's. 378a2699e47SSuresh Siddha */ 379a2699e47SSuresh Siddha __flush_tlb_all(); 380a2699e47SSuresh Siddha 381a2699e47SSuresh Siddha /* 382a2699e47SSuresh Siddha * Second iteration will set the actual desired PTE attributes. 383a2699e47SSuresh Siddha */ 384a2699e47SSuresh Siddha mapping_iter = 2; 385a2699e47SSuresh Siddha goto repeat; 386a2699e47SSuresh Siddha } 387c1fd1b43SPekka Enberg return last_map_addr; 388ae531c26SArjan van de Ven } 389ae531c26SArjan van de Ven 390ad757b6aSThomas Gleixner pte_t *kmap_pte; 391ad757b6aSThomas Gleixner 3928550eb99SIngo Molnar static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr) 3938550eb99SIngo Molnar { 394e0c4f675SKirill A. Shutemov pgd_t *pgd = pgd_offset_k(vaddr); 395e0c4f675SKirill A. Shutemov p4d_t *p4d = p4d_offset(pgd, vaddr); 396e0c4f675SKirill A. Shutemov pud_t *pud = pud_offset(p4d, vaddr); 397e0c4f675SKirill A. Shutemov pmd_t *pmd = pmd_offset(pud, vaddr); 398e0c4f675SKirill A. Shutemov return pte_offset_kernel(pmd, vaddr); 3998550eb99SIngo Molnar } 400ad757b6aSThomas Gleixner 401ad757b6aSThomas Gleixner static void __init kmap_init(void) 402ad757b6aSThomas Gleixner { 403ad757b6aSThomas Gleixner unsigned long kmap_vstart; 404ad757b6aSThomas Gleixner 4058550eb99SIngo Molnar /* 4068550eb99SIngo Molnar * Cache the first kmap pte: 4078550eb99SIngo Molnar */ 408ad757b6aSThomas Gleixner kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN); 409ad757b6aSThomas Gleixner kmap_pte = kmap_get_fixmap_pte(kmap_vstart); 410ad757b6aSThomas Gleixner } 411ad757b6aSThomas Gleixner 412fd940934SKeith Packard #ifdef CONFIG_HIGHMEM 413ad757b6aSThomas Gleixner static void __init permanent_kmaps_init(pgd_t *pgd_base) 414ad757b6aSThomas Gleixner { 4158550eb99SIngo Molnar unsigned long vaddr; 416ad757b6aSThomas Gleixner pgd_t *pgd; 417e0c4f675SKirill A. Shutemov p4d_t *p4d; 418ad757b6aSThomas Gleixner pud_t *pud; 419ad757b6aSThomas Gleixner pmd_t *pmd; 420ad757b6aSThomas Gleixner pte_t *pte; 421ad757b6aSThomas Gleixner 422ad757b6aSThomas Gleixner vaddr = PKMAP_BASE; 423ad757b6aSThomas Gleixner page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base); 424ad757b6aSThomas Gleixner 425ad757b6aSThomas Gleixner pgd = swapper_pg_dir + pgd_index(vaddr); 426e0c4f675SKirill A. Shutemov p4d = p4d_offset(pgd, vaddr); 427e0c4f675SKirill A. Shutemov pud = pud_offset(p4d, vaddr); 428ad757b6aSThomas Gleixner pmd = pmd_offset(pud, vaddr); 429ad757b6aSThomas Gleixner pte = pte_offset_kernel(pmd, vaddr); 430ad757b6aSThomas Gleixner pkmap_page_table = pte; 431ad757b6aSThomas Gleixner } 432ad757b6aSThomas Gleixner 4331d931264SYinghai Lu void __init add_highpages_with_active_regions(int nid, 4341d931264SYinghai Lu unsigned long start_pfn, unsigned long end_pfn) 435b5bc6c0eSYinghai Lu { 4368a9ca34cSTejun Heo phys_addr_t start, end; 4378a9ca34cSTejun Heo u64 i; 4381d931264SYinghai Lu 439fc6daaf9STony Luck for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &start, &end, NULL) { 4408a9ca34cSTejun Heo unsigned long pfn = clamp_t(unsigned long, PFN_UP(start), 4418a9ca34cSTejun Heo start_pfn, end_pfn); 4428a9ca34cSTejun Heo unsigned long e_pfn = clamp_t(unsigned long, PFN_DOWN(end), 4438a9ca34cSTejun Heo start_pfn, end_pfn); 4448a9ca34cSTejun Heo for ( ; pfn < e_pfn; pfn++) 4458a9ca34cSTejun Heo if (pfn_valid(pfn)) 4465e7ccf86SJiang Liu free_highmem_page(pfn_to_page(pfn)); 447b5bc6c0eSYinghai Lu } 448ad757b6aSThomas Gleixner } 449ad757b6aSThomas Gleixner #else 450e8e32326SIngo Brueckl static inline void permanent_kmaps_init(pgd_t *pgd_base) 451e8e32326SIngo Brueckl { 452e8e32326SIngo Brueckl } 453ad757b6aSThomas Gleixner #endif /* CONFIG_HIGHMEM */ 454ad757b6aSThomas Gleixner 455945fd17aSThomas Gleixner void __init sync_initial_page_table(void) 456945fd17aSThomas Gleixner { 457945fd17aSThomas Gleixner clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY, 458945fd17aSThomas Gleixner swapper_pg_dir + KERNEL_PGD_BOUNDARY, 459945fd17aSThomas Gleixner KERNEL_PGD_PTRS); 460945fd17aSThomas Gleixner 461945fd17aSThomas Gleixner /* 462945fd17aSThomas Gleixner * sync back low identity map too. It is used for example 463945fd17aSThomas Gleixner * in the 32-bit EFI stub. 464945fd17aSThomas Gleixner */ 465945fd17aSThomas Gleixner clone_pgd_range(initial_page_table, 466945fd17aSThomas Gleixner swapper_pg_dir + KERNEL_PGD_BOUNDARY, 467945fd17aSThomas Gleixner min(KERNEL_PGD_PTRS, KERNEL_PGD_BOUNDARY)); 468945fd17aSThomas Gleixner } 469945fd17aSThomas Gleixner 4707737b215SAttilio Rao void __init native_pagetable_init(void) 471ad757b6aSThomas Gleixner { 472551889a6SIan Campbell unsigned long pfn, va; 47373090f89SAttilio Rao pgd_t *pgd, *base = swapper_pg_dir; 474e0c4f675SKirill A. Shutemov p4d_t *p4d; 475551889a6SIan Campbell pud_t *pud; 476551889a6SIan Campbell pmd_t *pmd; 477551889a6SIan Campbell pte_t *pte; 478ad757b6aSThomas Gleixner 479ad757b6aSThomas Gleixner /* 480551889a6SIan Campbell * Remove any mappings which extend past the end of physical 48111ed9e92SYinghai Lu * memory from the boot time page table. 48211ed9e92SYinghai Lu * In virtual address space, we should have at least two pages 48311ed9e92SYinghai Lu * from VMALLOC_END to pkmap or fixmap according to VMALLOC_END 48411ed9e92SYinghai Lu * definition. And max_low_pfn is set to VMALLOC_END physical 48511ed9e92SYinghai Lu * address. If initial memory mapping is doing right job, we 48611ed9e92SYinghai Lu * should have pte used near max_low_pfn or one pmd is not present. 487ad757b6aSThomas Gleixner */ 48811ed9e92SYinghai Lu for (pfn = max_low_pfn; pfn < 1<<(32-PAGE_SHIFT); pfn++) { 489551889a6SIan Campbell va = PAGE_OFFSET + (pfn<<PAGE_SHIFT); 490551889a6SIan Campbell pgd = base + pgd_index(va); 491551889a6SIan Campbell if (!pgd_present(*pgd)) 492551889a6SIan Campbell break; 493ad757b6aSThomas Gleixner 494e0c4f675SKirill A. Shutemov p4d = p4d_offset(pgd, va); 495e0c4f675SKirill A. Shutemov pud = pud_offset(p4d, va); 496551889a6SIan Campbell pmd = pmd_offset(pud, va); 497551889a6SIan Campbell if (!pmd_present(*pmd)) 498551889a6SIan Campbell break; 499551889a6SIan Campbell 50011ed9e92SYinghai Lu /* should not be large page here */ 50111ed9e92SYinghai Lu if (pmd_large(*pmd)) { 50211ed9e92SYinghai Lu pr_warn("try to clear pte for ram above max_low_pfn: pfn: %lx pmd: %p pmd phys: %lx, but pmd is big page and is not using pte !\n", 50311ed9e92SYinghai Lu pfn, pmd, __pa(pmd)); 50411ed9e92SYinghai Lu BUG_ON(1); 50511ed9e92SYinghai Lu } 50611ed9e92SYinghai Lu 507551889a6SIan Campbell pte = pte_offset_kernel(pmd, va); 508551889a6SIan Campbell if (!pte_present(*pte)) 509551889a6SIan Campbell break; 510551889a6SIan Campbell 51111ed9e92SYinghai Lu printk(KERN_DEBUG "clearing pte for ram above max_low_pfn: pfn: %lx pmd: %p pmd phys: %lx pte: %p pte phys: %lx\n", 51211ed9e92SYinghai Lu pfn, pmd, __pa(pmd), pte, __pa(pte)); 513551889a6SIan Campbell pte_clear(NULL, va, pte); 514551889a6SIan Campbell } 5156944a9c8SJeremy Fitzhardinge paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT); 516843b8ed2SAttilio Rao paging_init(); 517ad757b6aSThomas Gleixner } 518ad757b6aSThomas Gleixner 519ad757b6aSThomas Gleixner /* 520ad757b6aSThomas Gleixner * Build a proper pagetable for the kernel mappings. Up until this 521ad757b6aSThomas Gleixner * point, we've been running on some set of pagetables constructed by 522ad757b6aSThomas Gleixner * the boot process. 523ad757b6aSThomas Gleixner * 524ad757b6aSThomas Gleixner * If we're booting on native hardware, this will be a pagetable 525551889a6SIan Campbell * constructed in arch/x86/kernel/head_32.S. The root of the 526551889a6SIan Campbell * pagetable will be swapper_pg_dir. 527ad757b6aSThomas Gleixner * 528ad757b6aSThomas Gleixner * If we're booting paravirtualized under a hypervisor, then there are 529ad757b6aSThomas Gleixner * more options: we may already be running PAE, and the pagetable may 530ad757b6aSThomas Gleixner * or may not be based in swapper_pg_dir. In any case, 5317737b215SAttilio Rao * paravirt_pagetable_init() will set up swapper_pg_dir 532ad757b6aSThomas Gleixner * appropriately for the rest of the initialization to work. 533ad757b6aSThomas Gleixner * 534ad757b6aSThomas Gleixner * In general, pagetable_init() assumes that the pagetable may already 535ad757b6aSThomas Gleixner * be partially populated, and so it avoids stomping on any existing 536ad757b6aSThomas Gleixner * mappings. 537ad757b6aSThomas Gleixner */ 538f765090aSPekka Enberg void __init early_ioremap_page_table_range_init(void) 539ad757b6aSThomas Gleixner { 540e7179853SPekka Enberg pgd_t *pgd_base = swapper_pg_dir; 5418550eb99SIngo Molnar unsigned long vaddr, end; 542ad757b6aSThomas Gleixner 543ad757b6aSThomas Gleixner /* 544ad757b6aSThomas Gleixner * Fixed mappings, only the page table structure has to be 545ad757b6aSThomas Gleixner * created - mappings will be set by set_fixmap(): 546ad757b6aSThomas Gleixner */ 547ad757b6aSThomas Gleixner vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK; 548ad757b6aSThomas Gleixner end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK; 549ad757b6aSThomas Gleixner page_table_range_init(vaddr, end, pgd_base); 550beacfaacSHuang, Ying early_ioremap_reset(); 551e7b37895SYinghai Lu } 552e7b37895SYinghai Lu 553e7b37895SYinghai Lu static void __init pagetable_init(void) 554e7b37895SYinghai Lu { 555e7b37895SYinghai Lu pgd_t *pgd_base = swapper_pg_dir; 556e7b37895SYinghai Lu 557ad757b6aSThomas Gleixner permanent_kmaps_init(pgd_base); 558ad757b6aSThomas Gleixner } 559ad757b6aSThomas Gleixner 5608a57f484SDave Hansen #define DEFAULT_PTE_MASK ~(_PAGE_NX | _PAGE_GLOBAL) 5618a57f484SDave Hansen /* Bits supported by the hardware: */ 5628a57f484SDave Hansen pteval_t __supported_pte_mask __read_mostly = DEFAULT_PTE_MASK; 5638a57f484SDave Hansen /* Bits allowed in normal kernel mappings: */ 5648a57f484SDave Hansen pteval_t __default_kernel_pte_mask __read_mostly = DEFAULT_PTE_MASK; 5656fdc05d4SJeremy Fitzhardinge EXPORT_SYMBOL_GPL(__supported_pte_mask); 5668a57f484SDave Hansen /* Used in PAGE_KERNEL_* macros which are reasonably used out-of-tree: */ 5678a57f484SDave Hansen EXPORT_SYMBOL(__default_kernel_pte_mask); 5686fdc05d4SJeremy Fitzhardinge 56990d967e0SYinghai Lu /* user-defined highmem size */ 57090d967e0SYinghai Lu static unsigned int highmem_pages = -1; 57190d967e0SYinghai Lu 57290d967e0SYinghai Lu /* 57390d967e0SYinghai Lu * highmem=size forces highmem to be exactly 'size' bytes. 57490d967e0SYinghai Lu * This works even on boxes that have no highmem otherwise. 57590d967e0SYinghai Lu * This also works to reduce highmem size on bigger boxes. 57690d967e0SYinghai Lu */ 57790d967e0SYinghai Lu static int __init parse_highmem(char *arg) 57890d967e0SYinghai Lu { 57990d967e0SYinghai Lu if (!arg) 58090d967e0SYinghai Lu return -EINVAL; 58190d967e0SYinghai Lu 58290d967e0SYinghai Lu highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT; 58390d967e0SYinghai Lu return 0; 58490d967e0SYinghai Lu } 58590d967e0SYinghai Lu early_param("highmem", parse_highmem); 58690d967e0SYinghai Lu 5874769843bSIngo Molnar #define MSG_HIGHMEM_TOO_BIG \ 5884769843bSIngo Molnar "highmem size (%luMB) is bigger than pages available (%luMB)!\n" 5894769843bSIngo Molnar 5904769843bSIngo Molnar #define MSG_LOWMEM_TOO_SMALL \ 5914769843bSIngo Molnar "highmem size (%luMB) results in <64MB lowmem, ignoring it!\n" 59290d967e0SYinghai Lu /* 5934769843bSIngo Molnar * All of RAM fits into lowmem - but if user wants highmem 5944769843bSIngo Molnar * artificially via the highmem=x boot parameter then create 5954769843bSIngo Molnar * it: 59690d967e0SYinghai Lu */ 597f836e35aSYinghai Lu static void __init lowmem_pfn_init(void) 59890d967e0SYinghai Lu { 599346cafecSYinghai Lu /* max_low_pfn is 0, we already have early_res support */ 60090d967e0SYinghai Lu max_low_pfn = max_pfn; 601d88316c2SIngo Molnar 6024769843bSIngo Molnar if (highmem_pages == -1) 6034769843bSIngo Molnar highmem_pages = 0; 6044769843bSIngo Molnar #ifdef CONFIG_HIGHMEM 6054769843bSIngo Molnar if (highmem_pages >= max_pfn) { 6064769843bSIngo Molnar printk(KERN_ERR MSG_HIGHMEM_TOO_BIG, 6074769843bSIngo Molnar pages_to_mb(highmem_pages), pages_to_mb(max_pfn)); 6084769843bSIngo Molnar highmem_pages = 0; 6094769843bSIngo Molnar } 6104769843bSIngo Molnar if (highmem_pages) { 6114769843bSIngo Molnar if (max_low_pfn - highmem_pages < 64*1024*1024/PAGE_SIZE) { 6124769843bSIngo Molnar printk(KERN_ERR MSG_LOWMEM_TOO_SMALL, 6134769843bSIngo Molnar pages_to_mb(highmem_pages)); 6144769843bSIngo Molnar highmem_pages = 0; 6154769843bSIngo Molnar } 6164769843bSIngo Molnar max_low_pfn -= highmem_pages; 6174769843bSIngo Molnar } 6184769843bSIngo Molnar #else 6194769843bSIngo Molnar if (highmem_pages) 6204769843bSIngo Molnar printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n"); 6214769843bSIngo Molnar #endif 6224769843bSIngo Molnar } 6234769843bSIngo Molnar 6244769843bSIngo Molnar #define MSG_HIGHMEM_TOO_SMALL \ 6254769843bSIngo Molnar "only %luMB highmem pages available, ignoring highmem size of %luMB!\n" 6264769843bSIngo Molnar 6274769843bSIngo Molnar #define MSG_HIGHMEM_TRIMMED \ 6284769843bSIngo Molnar "Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n" 6294769843bSIngo Molnar /* 6304769843bSIngo Molnar * We have more RAM than fits into lowmem - we try to put it into 6314769843bSIngo Molnar * highmem, also taking the highmem=x boot parameter into account: 6324769843bSIngo Molnar */ 633f836e35aSYinghai Lu static void __init highmem_pfn_init(void) 6344769843bSIngo Molnar { 635d88316c2SIngo Molnar max_low_pfn = MAXMEM_PFN; 636d88316c2SIngo Molnar 63790d967e0SYinghai Lu if (highmem_pages == -1) 63890d967e0SYinghai Lu highmem_pages = max_pfn - MAXMEM_PFN; 6394769843bSIngo Molnar 64090d967e0SYinghai Lu if (highmem_pages + MAXMEM_PFN < max_pfn) 64190d967e0SYinghai Lu max_pfn = MAXMEM_PFN + highmem_pages; 6424769843bSIngo Molnar 64390d967e0SYinghai Lu if (highmem_pages + MAXMEM_PFN > max_pfn) { 6444769843bSIngo Molnar printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL, 64590d967e0SYinghai Lu pages_to_mb(max_pfn - MAXMEM_PFN), 64690d967e0SYinghai Lu pages_to_mb(highmem_pages)); 64790d967e0SYinghai Lu highmem_pages = 0; 64890d967e0SYinghai Lu } 64990d967e0SYinghai Lu #ifndef CONFIG_HIGHMEM 65090d967e0SYinghai Lu /* Maximum memory usable is what is directly addressable */ 6514769843bSIngo Molnar printk(KERN_WARNING "Warning only %ldMB will be used.\n", MAXMEM>>20); 65290d967e0SYinghai Lu if (max_pfn > MAX_NONPAE_PFN) 6534769843bSIngo Molnar printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n"); 65490d967e0SYinghai Lu else 65590d967e0SYinghai Lu printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n"); 65690d967e0SYinghai Lu max_pfn = MAXMEM_PFN; 65790d967e0SYinghai Lu #else /* !CONFIG_HIGHMEM */ 65890d967e0SYinghai Lu #ifndef CONFIG_HIGHMEM64G 65990d967e0SYinghai Lu if (max_pfn > MAX_NONPAE_PFN) { 66090d967e0SYinghai Lu max_pfn = MAX_NONPAE_PFN; 6614769843bSIngo Molnar printk(KERN_WARNING MSG_HIGHMEM_TRIMMED); 66290d967e0SYinghai Lu } 66390d967e0SYinghai Lu #endif /* !CONFIG_HIGHMEM64G */ 66490d967e0SYinghai Lu #endif /* !CONFIG_HIGHMEM */ 66590d967e0SYinghai Lu } 6664769843bSIngo Molnar 66790d967e0SYinghai Lu /* 66890d967e0SYinghai Lu * Determine low and high memory ranges: 66990d967e0SYinghai Lu */ 67090d967e0SYinghai Lu void __init find_low_pfn_range(void) 67190d967e0SYinghai Lu { 67290d967e0SYinghai Lu /* it could update max_pfn */ 67390d967e0SYinghai Lu 674d88316c2SIngo Molnar if (max_pfn <= MAXMEM_PFN) 6754769843bSIngo Molnar lowmem_pfn_init(); 676d88316c2SIngo Molnar else 677d88316c2SIngo Molnar highmem_pfn_init(); 67890d967e0SYinghai Lu } 67990d967e0SYinghai Lu 680b2ac82a0SYinghai Lu #ifndef CONFIG_NEED_MULTIPLE_NODES 681d8fc3afcSTejun Heo void __init initmem_init(void) 682b2ac82a0SYinghai Lu { 683b2ac82a0SYinghai Lu #ifdef CONFIG_HIGHMEM 684b2ac82a0SYinghai Lu highstart_pfn = highend_pfn = max_pfn; 685b2ac82a0SYinghai Lu if (max_pfn > max_low_pfn) 686b2ac82a0SYinghai Lu highstart_pfn = max_low_pfn; 687b2ac82a0SYinghai Lu printk(KERN_NOTICE "%ldMB HIGHMEM available.\n", 688b2ac82a0SYinghai Lu pages_to_mb(highend_pfn - highstart_pfn)); 689b2ac82a0SYinghai Lu high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1; 690b2ac82a0SYinghai Lu #else 691b2ac82a0SYinghai Lu high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1; 692b2ac82a0SYinghai Lu #endif 6930608f70cSTejun Heo 694d7dc899aSStefan Agner memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0); 6950608f70cSTejun Heo sparse_memory_present_with_active_regions(0); 6960608f70cSTejun Heo 697b2ac82a0SYinghai Lu #ifdef CONFIG_FLATMEM 69846a84132SJiang Liu max_mapnr = IS_ENABLED(CONFIG_HIGHMEM) ? highend_pfn : max_low_pfn; 699b2ac82a0SYinghai Lu #endif 700dc16ecf7SJeremy Fitzhardinge __vmalloc_start_set = true; 701dc16ecf7SJeremy Fitzhardinge 702b2ac82a0SYinghai Lu printk(KERN_NOTICE "%ldMB LOWMEM available.\n", 703b2ac82a0SYinghai Lu pages_to_mb(max_low_pfn)); 704b2ac82a0SYinghai Lu 705b2ac82a0SYinghai Lu setup_bootmem_allocator(); 706b2ac82a0SYinghai Lu } 707cb95a13aSYinghai Lu #endif /* !CONFIG_NEED_MULTIPLE_NODES */ 708b2ac82a0SYinghai Lu 709b2ac82a0SYinghai Lu void __init setup_bootmem_allocator(void) 710b2ac82a0SYinghai Lu { 711b2ac82a0SYinghai Lu printk(KERN_INFO " mapped low ram: 0 - %08lx\n", 712b2ac82a0SYinghai Lu max_pfn_mapped<<PAGE_SHIFT); 713fc5efe39SYinghai Lu printk(KERN_INFO " low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT); 7144e29684cSYinghai Lu } 7154e29684cSYinghai Lu 716a04ad82dSYinghai Lu /* 717ad757b6aSThomas Gleixner * paging_init() sets up the page tables - note that the first 8MB are 718ad757b6aSThomas Gleixner * already mapped by head.S. 719ad757b6aSThomas Gleixner * 720ad757b6aSThomas Gleixner * This routines also unmaps the page at virtual kernel address 0, so 721ad757b6aSThomas Gleixner * that we can trap those pesky NULL-reference errors in the kernel. 722ad757b6aSThomas Gleixner */ 723ad757b6aSThomas Gleixner void __init paging_init(void) 724ad757b6aSThomas Gleixner { 725ad757b6aSThomas Gleixner pagetable_init(); 726ad757b6aSThomas Gleixner 727ad757b6aSThomas Gleixner __flush_tlb_all(); 728ad757b6aSThomas Gleixner 729ad757b6aSThomas Gleixner kmap_init(); 73011cd0bc1SYinghai Lu 73111cd0bc1SYinghai Lu /* 73211cd0bc1SYinghai Lu * NOTE: at this point the bootmem allocator is fully available. 73311cd0bc1SYinghai Lu */ 734c10d1e26SAndres Salomon olpc_dt_build_devicetree(); 735797390d8STejun Heo sparse_memory_present_with_active_regions(MAX_NUMNODES); 73611cd0bc1SYinghai Lu sparse_init(); 73711cd0bc1SYinghai Lu zone_sizes_init(); 738ad757b6aSThomas Gleixner } 739ad757b6aSThomas Gleixner 740ad757b6aSThomas Gleixner /* 741ad757b6aSThomas Gleixner * Test if the WP bit works in supervisor mode. It isn't supported on 386's 742f7f17a67SDmitri Vorobiev * and also on some strange 486's. All 586+'s are OK. This used to involve 743f7f17a67SDmitri Vorobiev * black magic jumps to work around some nasty CPU bugs, but fortunately the 744f7f17a67SDmitri Vorobiev * switch to using exceptions got rid of all that. 745ad757b6aSThomas Gleixner */ 746ad757b6aSThomas Gleixner static void __init test_wp_bit(void) 747ad757b6aSThomas Gleixner { 7484af17110SAndy Lutomirski char z = 0; 7496415813bSMathias Krause 750952a6c2cSBorislav Petkov printk(KERN_INFO "Checking if this processor honours the WP bit even in supervisor mode..."); 751ad757b6aSThomas Gleixner 7524af17110SAndy Lutomirski __set_fixmap(FIX_WP_TEST, __pa_symbol(empty_zero_page), PAGE_KERNEL_RO); 753ad757b6aSThomas Gleixner 754952a6c2cSBorislav Petkov if (probe_kernel_write((char *)fix_to_virt(FIX_WP_TEST), &z, 1)) { 755952a6c2cSBorislav Petkov clear_fixmap(FIX_WP_TEST); 756952a6c2cSBorislav Petkov printk(KERN_CONT "Ok.\n"); 757952a6c2cSBorislav Petkov return; 758ad757b6aSThomas Gleixner } 7594af17110SAndy Lutomirski 760952a6c2cSBorislav Petkov printk(KERN_CONT "No.\n"); 761952a6c2cSBorislav Petkov panic("Linux doesn't support CPUs with broken WP."); 762ad757b6aSThomas Gleixner } 763ad757b6aSThomas Gleixner 764ad757b6aSThomas Gleixner void __init mem_init(void) 765ad757b6aSThomas Gleixner { 766cfb80c9eSJeremy Fitzhardinge pci_iommu_alloc(); 767cfb80c9eSJeremy Fitzhardinge 768ad757b6aSThomas Gleixner #ifdef CONFIG_FLATMEM 769ad757b6aSThomas Gleixner BUG_ON(!mem_map); 770ad757b6aSThomas Gleixner #endif 771855c743aSStanislaw Gruszka /* 772855c743aSStanislaw Gruszka * With CONFIG_DEBUG_PAGEALLOC initialization of highmem pages has to 773c6ffc5caSMike Rapoport * be done before memblock_free_all(). Memblock use free low memory for 774855c743aSStanislaw Gruszka * temporary data (see find_range_array()) and for this purpose can use 775855c743aSStanislaw Gruszka * pages that was already passed to the buddy allocator, hence marked as 776855c743aSStanislaw Gruszka * not accessible in the page tables when compiled with 777855c743aSStanislaw Gruszka * CONFIG_DEBUG_PAGEALLOC. Otherwise order of initialization is not 778855c743aSStanislaw Gruszka * important here. 779855c743aSStanislaw Gruszka */ 780855c743aSStanislaw Gruszka set_highmem_pages_init(); 781855c743aSStanislaw Gruszka 782ad757b6aSThomas Gleixner /* this will put all low memory onto the freelists */ 783c6ffc5caSMike Rapoport memblock_free_all(); 784ad757b6aSThomas Gleixner 7854e37a890SYinghai Lu after_bootmem = 1; 7866f84f8d1SPavel Tatashin x86_init.hyper.init_after_bootmem(); 7874e37a890SYinghai Lu 78846a84132SJiang Liu mem_init_print_info(NULL); 789d7d119d7SIngo Molnar printk(KERN_INFO "virtual kernel memory layout:\n" 790ad757b6aSThomas Gleixner " fixmap : 0x%08lx - 0x%08lx (%4ld kB)\n" 79192a0f81dSThomas Gleixner " cpu_entry : 0x%08lx - 0x%08lx (%4ld kB)\n" 792ad757b6aSThomas Gleixner #ifdef CONFIG_HIGHMEM 793ad757b6aSThomas Gleixner " pkmap : 0x%08lx - 0x%08lx (%4ld kB)\n" 794ad757b6aSThomas Gleixner #endif 795ad757b6aSThomas Gleixner " vmalloc : 0x%08lx - 0x%08lx (%4ld MB)\n" 796ad757b6aSThomas Gleixner " lowmem : 0x%08lx - 0x%08lx (%4ld MB)\n" 797ad757b6aSThomas Gleixner " .init : 0x%08lx - 0x%08lx (%4ld kB)\n" 798ad757b6aSThomas Gleixner " .data : 0x%08lx - 0x%08lx (%4ld kB)\n" 799ad757b6aSThomas Gleixner " .text : 0x%08lx - 0x%08lx (%4ld kB)\n", 800ad757b6aSThomas Gleixner FIXADDR_START, FIXADDR_TOP, 801ad757b6aSThomas Gleixner (FIXADDR_TOP - FIXADDR_START) >> 10, 802ad757b6aSThomas Gleixner 80392a0f81dSThomas Gleixner CPU_ENTRY_AREA_BASE, 80492a0f81dSThomas Gleixner CPU_ENTRY_AREA_BASE + CPU_ENTRY_AREA_MAP_SIZE, 80592a0f81dSThomas Gleixner CPU_ENTRY_AREA_MAP_SIZE >> 10, 80692a0f81dSThomas Gleixner 807ad757b6aSThomas Gleixner #ifdef CONFIG_HIGHMEM 808ad757b6aSThomas Gleixner PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE, 809ad757b6aSThomas Gleixner (LAST_PKMAP*PAGE_SIZE) >> 10, 810ad757b6aSThomas Gleixner #endif 811ad757b6aSThomas Gleixner 812ad757b6aSThomas Gleixner VMALLOC_START, VMALLOC_END, 813ad757b6aSThomas Gleixner (VMALLOC_END - VMALLOC_START) >> 20, 814ad757b6aSThomas Gleixner 815ad757b6aSThomas Gleixner (unsigned long)__va(0), (unsigned long)high_memory, 816ad757b6aSThomas Gleixner ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20, 817ad757b6aSThomas Gleixner 818ad757b6aSThomas Gleixner (unsigned long)&__init_begin, (unsigned long)&__init_end, 8198550eb99SIngo Molnar ((unsigned long)&__init_end - 8208550eb99SIngo Molnar (unsigned long)&__init_begin) >> 10, 821ad757b6aSThomas Gleixner 822ad757b6aSThomas Gleixner (unsigned long)&_etext, (unsigned long)&_edata, 823ad757b6aSThomas Gleixner ((unsigned long)&_edata - (unsigned long)&_etext) >> 10, 824ad757b6aSThomas Gleixner 825ad757b6aSThomas Gleixner (unsigned long)&_text, (unsigned long)&_etext, 826ad757b6aSThomas Gleixner ((unsigned long)&_etext - (unsigned long)&_text) >> 10); 827ad757b6aSThomas Gleixner 828beeb4195SJan Beulich /* 829beeb4195SJan Beulich * Check boundaries twice: Some fundamental inconsistencies can 830beeb4195SJan Beulich * be detected at build time already. 831beeb4195SJan Beulich */ 832beeb4195SJan Beulich #define __FIXADDR_TOP (-PAGE_SIZE) 833beeb4195SJan Beulich #ifdef CONFIG_HIGHMEM 834beeb4195SJan Beulich BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START); 835beeb4195SJan Beulich BUILD_BUG_ON(VMALLOC_END > PKMAP_BASE); 836beeb4195SJan Beulich #endif 837beeb4195SJan Beulich #define high_memory (-128UL << 20) 838beeb4195SJan Beulich BUILD_BUG_ON(VMALLOC_START >= VMALLOC_END); 839beeb4195SJan Beulich #undef high_memory 840beeb4195SJan Beulich #undef __FIXADDR_TOP 841beeb4195SJan Beulich 842ad757b6aSThomas Gleixner #ifdef CONFIG_HIGHMEM 843ad757b6aSThomas Gleixner BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START); 844ad757b6aSThomas Gleixner BUG_ON(VMALLOC_END > PKMAP_BASE); 845ad757b6aSThomas Gleixner #endif 846beeb4195SJan Beulich BUG_ON(VMALLOC_START >= VMALLOC_END); 847ad757b6aSThomas Gleixner BUG_ON((unsigned long)high_memory > VMALLOC_START); 848ad757b6aSThomas Gleixner 849ad757b6aSThomas Gleixner test_wp_bit(); 850ad757b6aSThomas Gleixner } 851ad757b6aSThomas Gleixner 852ad757b6aSThomas Gleixner #ifdef CONFIG_MEMORY_HOTPLUG 853*940519f0SMichal Hocko int arch_add_memory(int nid, u64 start, u64 size, 854*940519f0SMichal Hocko struct mhp_restrictions *restrictions) 855ad757b6aSThomas Gleixner { 856ad757b6aSThomas Gleixner unsigned long start_pfn = start >> PAGE_SHIFT; 857ad757b6aSThomas Gleixner unsigned long nr_pages = size >> PAGE_SHIFT; 858ad757b6aSThomas Gleixner 859*940519f0SMichal Hocko return __add_pages(nid, start_pfn, nr_pages, restrictions); 860ad757b6aSThomas Gleixner } 86124d335caSWen Congyang 86224d335caSWen Congyang #ifdef CONFIG_MEMORY_HOTREMOVE 8632c2a5af6SOscar Salvador int arch_remove_memory(int nid, u64 start, u64 size, struct vmem_altmap *altmap) 86424d335caSWen Congyang { 86524d335caSWen Congyang unsigned long start_pfn = start >> PAGE_SHIFT; 86624d335caSWen Congyang unsigned long nr_pages = size >> PAGE_SHIFT; 86724d335caSWen Congyang struct zone *zone; 86824d335caSWen Congyang 86924d335caSWen Congyang zone = page_zone(pfn_to_page(start_pfn)); 870da024512SChristoph Hellwig return __remove_pages(zone, start_pfn, nr_pages, altmap); 87124d335caSWen Congyang } 87224d335caSWen Congyang #endif 873ad757b6aSThomas Gleixner #endif 874ad757b6aSThomas Gleixner 875502f6604SSuresh Siddha int kernel_set_to_readonly __read_mostly; 87616239630SSteven Rostedt 87716239630SSteven Rostedt void set_kernel_text_rw(void) 87816239630SSteven Rostedt { 87916239630SSteven Rostedt unsigned long start = PFN_ALIGN(_text); 88016239630SSteven Rostedt unsigned long size = PFN_ALIGN(_etext) - start; 88116239630SSteven Rostedt 88216239630SSteven Rostedt if (!kernel_set_to_readonly) 88316239630SSteven Rostedt return; 88416239630SSteven Rostedt 88516239630SSteven Rostedt pr_debug("Set kernel text: %lx - %lx for read write\n", 88616239630SSteven Rostedt start, start+size); 88716239630SSteven Rostedt 88816239630SSteven Rostedt set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT); 88916239630SSteven Rostedt } 89016239630SSteven Rostedt 89116239630SSteven Rostedt void set_kernel_text_ro(void) 89216239630SSteven Rostedt { 89316239630SSteven Rostedt unsigned long start = PFN_ALIGN(_text); 89416239630SSteven Rostedt unsigned long size = PFN_ALIGN(_etext) - start; 89516239630SSteven Rostedt 89616239630SSteven Rostedt if (!kernel_set_to_readonly) 89716239630SSteven Rostedt return; 89816239630SSteven Rostedt 89916239630SSteven Rostedt pr_debug("Set kernel text: %lx - %lx for read only\n", 90016239630SSteven Rostedt start, start+size); 90116239630SSteven Rostedt 90216239630SSteven Rostedt set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT); 90316239630SSteven Rostedt } 90416239630SSteven Rostedt 9055bd5a452SMatthieu Castet static void mark_nxdata_nx(void) 9065bd5a452SMatthieu Castet { 9075bd5a452SMatthieu Castet /* 9085bd5a452SMatthieu Castet * When this called, init has already been executed and released, 9090d2eb44fSLucas De Marchi * so everything past _etext should be NX. 9105bd5a452SMatthieu Castet */ 9115bd5a452SMatthieu Castet unsigned long start = PFN_ALIGN(_etext); 9125bd5a452SMatthieu Castet /* 9135bd5a452SMatthieu Castet * This comes from is_kernel_text upper limit. Also HPAGE where used: 9145bd5a452SMatthieu Castet */ 9155bd5a452SMatthieu Castet unsigned long size = (((unsigned long)__init_end + HPAGE_SIZE) & HPAGE_MASK) - start; 9165bd5a452SMatthieu Castet 9175bd5a452SMatthieu Castet if (__supported_pte_mask & _PAGE_NX) 9185bd5a452SMatthieu Castet printk(KERN_INFO "NX-protecting the kernel data: %luk\n", size >> 10); 9195bd5a452SMatthieu Castet set_pages_nx(virt_to_page(start), size >> PAGE_SHIFT); 9205bd5a452SMatthieu Castet } 9215bd5a452SMatthieu Castet 922ad757b6aSThomas Gleixner void mark_rodata_ro(void) 923ad757b6aSThomas Gleixner { 924ad757b6aSThomas Gleixner unsigned long start = PFN_ALIGN(_text); 9252a25dc7cSThomas Gleixner unsigned long size = (unsigned long)__end_rodata - start; 926ad757b6aSThomas Gleixner 9276d238cc4SArjan van de Ven set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT); 9282a25dc7cSThomas Gleixner pr_info("Write protecting kernel text and read-only data: %luk\n", 929d7d119d7SIngo Molnar size >> 10); 9300c42f392SAndi Kleen 93116239630SSteven Rostedt kernel_set_to_readonly = 1; 93216239630SSteven Rostedt 9330c42f392SAndi Kleen #ifdef CONFIG_CPA_DEBUG 9342a25dc7cSThomas Gleixner pr_info("Testing CPA: Reverting %lx-%lx\n", start, start + size); 9356d238cc4SArjan van de Ven set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT); 9360c42f392SAndi Kleen 9372a25dc7cSThomas Gleixner pr_info("Testing CPA: write protecting again\n"); 9386d238cc4SArjan van de Ven set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT); 9390c42f392SAndi Kleen #endif 9405bd5a452SMatthieu Castet mark_nxdata_nx(); 941e1a58320SStephen Smalley if (__supported_pte_mask & _PAGE_NX) 942e1a58320SStephen Smalley debug_checkwx(); 943ad757b6aSThomas Gleixner } 944