1 /* 2 * This file contains the routines setting up the linux page tables. 3 * -- paulus 4 * 5 * Derived from arch/ppc/mm/init.c: 6 * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org) 7 * 8 * Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au) 9 * and Cort Dougan (PReP) (cort@cs.nmt.edu) 10 * Copyright (C) 1996 Paul Mackerras 11 * 12 * Derived from "arch/i386/mm/init.c" 13 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds 14 * 15 * This program is free software; you can redistribute it and/or 16 * modify it under the terms of the GNU General Public License 17 * as published by the Free Software Foundation; either version 18 * 2 of the License, or (at your option) any later version. 19 * 20 */ 21 22 #include <linux/kernel.h> 23 #include <linux/module.h> 24 #include <linux/types.h> 25 #include <linux/mm.h> 26 #include <linux/vmalloc.h> 27 #include <linux/init.h> 28 #include <linux/highmem.h> 29 #include <linux/memblock.h> 30 #include <linux/slab.h> 31 32 #include <asm/pgtable.h> 33 #include <asm/pgalloc.h> 34 #include <asm/fixmap.h> 35 #include <asm/io.h> 36 #include <asm/setup.h> 37 38 #include "mmu_decl.h" 39 40 unsigned long ioremap_bot; 41 EXPORT_SYMBOL(ioremap_bot); /* aka VMALLOC_END */ 42 43 extern char etext[], _stext[], _sinittext[], _einittext[]; 44 45 __ref pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address) 46 { 47 pte_t *pte; 48 49 if (slab_is_available()) { 50 pte = (pte_t *)__get_free_page(GFP_KERNEL|__GFP_ZERO); 51 } else { 52 pte = __va(memblock_alloc(PAGE_SIZE, PAGE_SIZE)); 53 if (pte) 54 clear_page(pte); 55 } 56 return pte; 57 } 58 59 pgtable_t pte_alloc_one(struct mm_struct *mm, unsigned long address) 60 { 61 struct page *ptepage; 62 63 gfp_t flags = GFP_KERNEL | __GFP_ZERO | __GFP_ACCOUNT; 64 65 ptepage = alloc_pages(flags, 0); 66 if (!ptepage) 67 return NULL; 68 if (!pgtable_page_ctor(ptepage)) { 69 __free_page(ptepage); 70 return NULL; 71 } 72 return ptepage; 73 } 74 75 void __iomem * 76 ioremap(phys_addr_t addr, unsigned long size) 77 { 78 return __ioremap_caller(addr, size, _PAGE_NO_CACHE | _PAGE_GUARDED, 79 __builtin_return_address(0)); 80 } 81 EXPORT_SYMBOL(ioremap); 82 83 void __iomem * 84 ioremap_wc(phys_addr_t addr, unsigned long size) 85 { 86 return __ioremap_caller(addr, size, _PAGE_NO_CACHE, 87 __builtin_return_address(0)); 88 } 89 EXPORT_SYMBOL(ioremap_wc); 90 91 void __iomem * 92 ioremap_prot(phys_addr_t addr, unsigned long size, unsigned long flags) 93 { 94 /* writeable implies dirty for kernel addresses */ 95 if ((flags & (_PAGE_RW | _PAGE_RO)) != _PAGE_RO) 96 flags |= _PAGE_DIRTY | _PAGE_HWWRITE; 97 98 /* we don't want to let _PAGE_USER and _PAGE_EXEC leak out */ 99 flags &= ~(_PAGE_USER | _PAGE_EXEC); 100 101 #ifdef _PAGE_BAP_SR 102 /* _PAGE_USER contains _PAGE_BAP_SR on BookE using the new PTE format 103 * which means that we just cleared supervisor access... oops ;-) This 104 * restores it 105 */ 106 flags |= _PAGE_BAP_SR; 107 #endif 108 109 return __ioremap_caller(addr, size, flags, __builtin_return_address(0)); 110 } 111 EXPORT_SYMBOL(ioremap_prot); 112 113 void __iomem * 114 __ioremap(phys_addr_t addr, unsigned long size, unsigned long flags) 115 { 116 return __ioremap_caller(addr, size, flags, __builtin_return_address(0)); 117 } 118 119 void __iomem * 120 __ioremap_caller(phys_addr_t addr, unsigned long size, unsigned long flags, 121 void *caller) 122 { 123 unsigned long v, i; 124 phys_addr_t p; 125 int err; 126 127 /* Make sure we have the base flags */ 128 if ((flags & _PAGE_PRESENT) == 0) 129 flags |= pgprot_val(PAGE_KERNEL); 130 131 /* Non-cacheable page cannot be coherent */ 132 if (flags & _PAGE_NO_CACHE) 133 flags &= ~_PAGE_COHERENT; 134 135 /* 136 * Choose an address to map it to. 137 * Once the vmalloc system is running, we use it. 138 * Before then, we use space going down from IOREMAP_TOP 139 * (ioremap_bot records where we're up to). 140 */ 141 p = addr & PAGE_MASK; 142 size = PAGE_ALIGN(addr + size) - p; 143 144 /* 145 * If the address lies within the first 16 MB, assume it's in ISA 146 * memory space 147 */ 148 if (p < 16*1024*1024) 149 p += _ISA_MEM_BASE; 150 151 #ifndef CONFIG_CRASH_DUMP 152 /* 153 * Don't allow anybody to remap normal RAM that we're using. 154 * mem_init() sets high_memory so only do the check after that. 155 */ 156 if (slab_is_available() && (p < virt_to_phys(high_memory)) && 157 !(__allow_ioremap_reserved && memblock_is_region_reserved(p, size))) { 158 printk("__ioremap(): phys addr 0x%llx is RAM lr %ps\n", 159 (unsigned long long)p, __builtin_return_address(0)); 160 return NULL; 161 } 162 #endif 163 164 if (size == 0) 165 return NULL; 166 167 /* 168 * Is it already mapped? Perhaps overlapped by a previous 169 * mapping. 170 */ 171 v = p_block_mapped(p); 172 if (v) 173 goto out; 174 175 if (slab_is_available()) { 176 struct vm_struct *area; 177 area = get_vm_area_caller(size, VM_IOREMAP, caller); 178 if (area == 0) 179 return NULL; 180 area->phys_addr = p; 181 v = (unsigned long) area->addr; 182 } else { 183 v = (ioremap_bot -= size); 184 } 185 186 /* 187 * Should check if it is a candidate for a BAT mapping 188 */ 189 190 err = 0; 191 for (i = 0; i < size && err == 0; i += PAGE_SIZE) 192 err = map_kernel_page(v+i, p+i, flags); 193 if (err) { 194 if (slab_is_available()) 195 vunmap((void *)v); 196 return NULL; 197 } 198 199 out: 200 return (void __iomem *) (v + ((unsigned long)addr & ~PAGE_MASK)); 201 } 202 EXPORT_SYMBOL(__ioremap); 203 204 void iounmap(volatile void __iomem *addr) 205 { 206 /* 207 * If mapped by BATs then there is nothing to do. 208 * Calling vfree() generates a benign warning. 209 */ 210 if (v_block_mapped((unsigned long)addr)) 211 return; 212 213 if (addr > high_memory && (unsigned long) addr < ioremap_bot) 214 vunmap((void *) (PAGE_MASK & (unsigned long)addr)); 215 } 216 EXPORT_SYMBOL(iounmap); 217 218 int map_kernel_page(unsigned long va, phys_addr_t pa, int flags) 219 { 220 pmd_t *pd; 221 pte_t *pg; 222 int err = -ENOMEM; 223 224 /* Use upper 10 bits of VA to index the first level map */ 225 pd = pmd_offset(pud_offset(pgd_offset_k(va), va), va); 226 /* Use middle 10 bits of VA to index the second-level map */ 227 pg = pte_alloc_kernel(pd, va); 228 if (pg != 0) { 229 err = 0; 230 /* The PTE should never be already set nor present in the 231 * hash table 232 */ 233 BUG_ON((pte_val(*pg) & (_PAGE_PRESENT | _PAGE_HASHPTE)) && 234 flags); 235 set_pte_at(&init_mm, va, pg, pfn_pte(pa >> PAGE_SHIFT, 236 __pgprot(flags))); 237 } 238 smp_wmb(); 239 return err; 240 } 241 242 /* 243 * Map in a chunk of physical memory starting at start. 244 */ 245 void __init __mapin_ram_chunk(unsigned long offset, unsigned long top) 246 { 247 unsigned long v, s, f; 248 phys_addr_t p; 249 int ktext; 250 251 s = offset; 252 v = PAGE_OFFSET + s; 253 p = memstart_addr + s; 254 for (; s < top; s += PAGE_SIZE) { 255 ktext = ((char *)v >= _stext && (char *)v < etext) || 256 ((char *)v >= _sinittext && (char *)v < _einittext); 257 f = ktext ? pgprot_val(PAGE_KERNEL_TEXT) : pgprot_val(PAGE_KERNEL); 258 map_kernel_page(v, p, f); 259 #ifdef CONFIG_PPC_STD_MMU_32 260 if (ktext) 261 hash_preload(&init_mm, v, 0, 0x300); 262 #endif 263 v += PAGE_SIZE; 264 p += PAGE_SIZE; 265 } 266 } 267 268 void __init mapin_ram(void) 269 { 270 unsigned long s, top; 271 272 #ifndef CONFIG_WII 273 top = total_lowmem; 274 s = mmu_mapin_ram(top); 275 __mapin_ram_chunk(s, top); 276 #else 277 if (!wii_hole_size) { 278 s = mmu_mapin_ram(total_lowmem); 279 __mapin_ram_chunk(s, total_lowmem); 280 } else { 281 top = wii_hole_start; 282 s = mmu_mapin_ram(top); 283 __mapin_ram_chunk(s, top); 284 285 top = memblock_end_of_DRAM(); 286 s = wii_mmu_mapin_mem2(top); 287 __mapin_ram_chunk(s, top); 288 } 289 #endif 290 } 291 292 /* Scan the real Linux page tables and return a PTE pointer for 293 * a virtual address in a context. 294 * Returns true (1) if PTE was found, zero otherwise. The pointer to 295 * the PTE pointer is unmodified if PTE is not found. 296 */ 297 int 298 get_pteptr(struct mm_struct *mm, unsigned long addr, pte_t **ptep, pmd_t **pmdp) 299 { 300 pgd_t *pgd; 301 pud_t *pud; 302 pmd_t *pmd; 303 pte_t *pte; 304 int retval = 0; 305 306 pgd = pgd_offset(mm, addr & PAGE_MASK); 307 if (pgd) { 308 pud = pud_offset(pgd, addr & PAGE_MASK); 309 if (pud && pud_present(*pud)) { 310 pmd = pmd_offset(pud, addr & PAGE_MASK); 311 if (pmd_present(*pmd)) { 312 pte = pte_offset_map(pmd, addr & PAGE_MASK); 313 if (pte) { 314 retval = 1; 315 *ptep = pte; 316 if (pmdp) 317 *pmdp = pmd; 318 /* XXX caller needs to do pte_unmap, yuck */ 319 } 320 } 321 } 322 } 323 return(retval); 324 } 325 326 #ifdef CONFIG_DEBUG_PAGEALLOC 327 328 static int __change_page_attr(struct page *page, pgprot_t prot) 329 { 330 pte_t *kpte; 331 pmd_t *kpmd; 332 unsigned long address; 333 334 BUG_ON(PageHighMem(page)); 335 address = (unsigned long)page_address(page); 336 337 if (v_block_mapped(address)) 338 return 0; 339 if (!get_pteptr(&init_mm, address, &kpte, &kpmd)) 340 return -EINVAL; 341 __set_pte_at(&init_mm, address, kpte, mk_pte(page, prot), 0); 342 wmb(); 343 flush_tlb_page(NULL, address); 344 pte_unmap(kpte); 345 346 return 0; 347 } 348 349 /* 350 * Change the page attributes of an page in the linear mapping. 351 * 352 * THIS CONFLICTS WITH BAT MAPPINGS, DEBUG USE ONLY 353 */ 354 static int change_page_attr(struct page *page, int numpages, pgprot_t prot) 355 { 356 int i, err = 0; 357 unsigned long flags; 358 359 local_irq_save(flags); 360 for (i = 0; i < numpages; i++, page++) { 361 err = __change_page_attr(page, prot); 362 if (err) 363 break; 364 } 365 local_irq_restore(flags); 366 return err; 367 } 368 369 370 void __kernel_map_pages(struct page *page, int numpages, int enable) 371 { 372 if (PageHighMem(page)) 373 return; 374 375 change_page_attr(page, numpages, enable ? PAGE_KERNEL : __pgprot(0)); 376 } 377 #endif /* CONFIG_DEBUG_PAGEALLOC */ 378 379 static int fixmaps; 380 381 void __set_fixmap (enum fixed_addresses idx, phys_addr_t phys, pgprot_t flags) 382 { 383 unsigned long address = __fix_to_virt(idx); 384 385 if (idx >= __end_of_fixed_addresses) { 386 BUG(); 387 return; 388 } 389 390 map_kernel_page(address, phys, pgprot_val(flags)); 391 fixmaps++; 392 } 393