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_base; 41 unsigned long ioremap_bot; 42 EXPORT_SYMBOL(ioremap_bot); /* aka VMALLOC_END */ 43 44 #ifdef CONFIG_6xx 45 #define HAVE_BATS 1 46 #endif 47 48 #if defined(CONFIG_FSL_BOOKE) 49 #define HAVE_TLBCAM 1 50 #endif 51 52 extern char etext[], _stext[]; 53 54 #ifdef HAVE_BATS 55 extern phys_addr_t v_mapped_by_bats(unsigned long va); 56 extern unsigned long p_mapped_by_bats(phys_addr_t pa); 57 void setbat(int index, unsigned long virt, phys_addr_t phys, 58 unsigned int size, int flags); 59 60 #else /* !HAVE_BATS */ 61 #define v_mapped_by_bats(x) (0UL) 62 #define p_mapped_by_bats(x) (0UL) 63 #endif /* HAVE_BATS */ 64 65 #ifdef HAVE_TLBCAM 66 extern phys_addr_t v_mapped_by_tlbcam(unsigned long va); 67 extern unsigned long p_mapped_by_tlbcam(phys_addr_t pa); 68 #else /* !HAVE_TLBCAM */ 69 #define v_mapped_by_tlbcam(x) (0UL) 70 #define p_mapped_by_tlbcam(x) (0UL) 71 #endif /* HAVE_TLBCAM */ 72 73 #define PGDIR_ORDER (32 + PGD_T_LOG2 - PGDIR_SHIFT) 74 75 #ifndef CONFIG_PPC_4K_PAGES 76 static struct kmem_cache *pgtable_cache; 77 78 void pgtable_cache_init(void) 79 { 80 pgtable_cache = kmem_cache_create("PGDIR cache", 1 << PGDIR_ORDER, 81 1 << PGDIR_ORDER, 0, NULL); 82 if (pgtable_cache == NULL) 83 panic("Couldn't allocate pgtable caches"); 84 } 85 #endif 86 87 pgd_t *pgd_alloc(struct mm_struct *mm) 88 { 89 pgd_t *ret; 90 91 /* pgdir take page or two with 4K pages and a page fraction otherwise */ 92 #ifndef CONFIG_PPC_4K_PAGES 93 ret = kmem_cache_alloc(pgtable_cache, GFP_KERNEL | __GFP_ZERO); 94 #else 95 ret = (pgd_t *)__get_free_pages(GFP_KERNEL|__GFP_ZERO, 96 PGDIR_ORDER - PAGE_SHIFT); 97 #endif 98 return ret; 99 } 100 101 void pgd_free(struct mm_struct *mm, pgd_t *pgd) 102 { 103 #ifndef CONFIG_PPC_4K_PAGES 104 kmem_cache_free(pgtable_cache, (void *)pgd); 105 #else 106 free_pages((unsigned long)pgd, PGDIR_ORDER - PAGE_SHIFT); 107 #endif 108 } 109 110 __init_refok pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address) 111 { 112 pte_t *pte; 113 extern int mem_init_done; 114 115 if (mem_init_done) { 116 pte = (pte_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO); 117 } else { 118 pte = __va(memblock_alloc(PAGE_SIZE, PAGE_SIZE)); 119 if (pte) 120 clear_page(pte); 121 } 122 return pte; 123 } 124 125 pgtable_t pte_alloc_one(struct mm_struct *mm, unsigned long address) 126 { 127 struct page *ptepage; 128 129 gfp_t flags = GFP_KERNEL | __GFP_REPEAT | __GFP_ZERO; 130 131 ptepage = alloc_pages(flags, 0); 132 if (!ptepage) 133 return NULL; 134 if (!pgtable_page_ctor(ptepage)) { 135 __free_page(ptepage); 136 return NULL; 137 } 138 return ptepage; 139 } 140 141 void __iomem * 142 ioremap(phys_addr_t addr, unsigned long size) 143 { 144 return __ioremap_caller(addr, size, _PAGE_NO_CACHE | _PAGE_GUARDED, 145 __builtin_return_address(0)); 146 } 147 EXPORT_SYMBOL(ioremap); 148 149 void __iomem * 150 ioremap_wc(phys_addr_t addr, unsigned long size) 151 { 152 return __ioremap_caller(addr, size, _PAGE_NO_CACHE, 153 __builtin_return_address(0)); 154 } 155 EXPORT_SYMBOL(ioremap_wc); 156 157 void __iomem * 158 ioremap_prot(phys_addr_t addr, unsigned long size, unsigned long flags) 159 { 160 /* writeable implies dirty for kernel addresses */ 161 if ((flags & (_PAGE_RW | _PAGE_RO)) != _PAGE_RO) 162 flags |= _PAGE_DIRTY | _PAGE_HWWRITE; 163 164 /* we don't want to let _PAGE_USER and _PAGE_EXEC leak out */ 165 flags &= ~(_PAGE_USER | _PAGE_EXEC); 166 167 #ifdef _PAGE_BAP_SR 168 /* _PAGE_USER contains _PAGE_BAP_SR on BookE using the new PTE format 169 * which means that we just cleared supervisor access... oops ;-) This 170 * restores it 171 */ 172 flags |= _PAGE_BAP_SR; 173 #endif 174 175 return __ioremap_caller(addr, size, flags, __builtin_return_address(0)); 176 } 177 EXPORT_SYMBOL(ioremap_prot); 178 179 void __iomem * 180 __ioremap(phys_addr_t addr, unsigned long size, unsigned long flags) 181 { 182 return __ioremap_caller(addr, size, flags, __builtin_return_address(0)); 183 } 184 185 void __iomem * 186 __ioremap_caller(phys_addr_t addr, unsigned long size, unsigned long flags, 187 void *caller) 188 { 189 unsigned long v, i; 190 phys_addr_t p; 191 int err; 192 193 /* Make sure we have the base flags */ 194 if ((flags & _PAGE_PRESENT) == 0) 195 flags |= PAGE_KERNEL; 196 197 /* Non-cacheable page cannot be coherent */ 198 if (flags & _PAGE_NO_CACHE) 199 flags &= ~_PAGE_COHERENT; 200 201 /* 202 * Choose an address to map it to. 203 * Once the vmalloc system is running, we use it. 204 * Before then, we use space going down from ioremap_base 205 * (ioremap_bot records where we're up to). 206 */ 207 p = addr & PAGE_MASK; 208 size = PAGE_ALIGN(addr + size) - p; 209 210 /* 211 * If the address lies within the first 16 MB, assume it's in ISA 212 * memory space 213 */ 214 if (p < 16*1024*1024) 215 p += _ISA_MEM_BASE; 216 217 #ifndef CONFIG_CRASH_DUMP 218 /* 219 * Don't allow anybody to remap normal RAM that we're using. 220 * mem_init() sets high_memory so only do the check after that. 221 */ 222 if (mem_init_done && (p < virt_to_phys(high_memory)) && 223 !(__allow_ioremap_reserved && memblock_is_region_reserved(p, size))) { 224 printk("__ioremap(): phys addr 0x%llx is RAM lr %pf\n", 225 (unsigned long long)p, __builtin_return_address(0)); 226 return NULL; 227 } 228 #endif 229 230 if (size == 0) 231 return NULL; 232 233 /* 234 * Is it already mapped? Perhaps overlapped by a previous 235 * BAT mapping. If the whole area is mapped then we're done, 236 * otherwise remap it since we want to keep the virt addrs for 237 * each request contiguous. 238 * 239 * We make the assumption here that if the bottom and top 240 * of the range we want are mapped then it's mapped to the 241 * same virt address (and this is contiguous). 242 * -- Cort 243 */ 244 if ((v = p_mapped_by_bats(p)) /*&& p_mapped_by_bats(p+size-1)*/ ) 245 goto out; 246 247 if ((v = p_mapped_by_tlbcam(p))) 248 goto out; 249 250 if (mem_init_done) { 251 struct vm_struct *area; 252 area = get_vm_area_caller(size, VM_IOREMAP, caller); 253 if (area == 0) 254 return NULL; 255 area->phys_addr = p; 256 v = (unsigned long) area->addr; 257 } else { 258 v = (ioremap_bot -= size); 259 } 260 261 /* 262 * Should check if it is a candidate for a BAT mapping 263 */ 264 265 err = 0; 266 for (i = 0; i < size && err == 0; i += PAGE_SIZE) 267 err = map_page(v+i, p+i, flags); 268 if (err) { 269 if (mem_init_done) 270 vunmap((void *)v); 271 return NULL; 272 } 273 274 out: 275 return (void __iomem *) (v + ((unsigned long)addr & ~PAGE_MASK)); 276 } 277 EXPORT_SYMBOL(__ioremap); 278 279 void iounmap(volatile void __iomem *addr) 280 { 281 /* 282 * If mapped by BATs then there is nothing to do. 283 * Calling vfree() generates a benign warning. 284 */ 285 if (v_mapped_by_bats((unsigned long)addr)) return; 286 287 if (addr > high_memory && (unsigned long) addr < ioremap_bot) 288 vunmap((void *) (PAGE_MASK & (unsigned long)addr)); 289 } 290 EXPORT_SYMBOL(iounmap); 291 292 int map_page(unsigned long va, phys_addr_t pa, int flags) 293 { 294 pmd_t *pd; 295 pte_t *pg; 296 int err = -ENOMEM; 297 298 /* Use upper 10 bits of VA to index the first level map */ 299 pd = pmd_offset(pud_offset(pgd_offset_k(va), va), va); 300 /* Use middle 10 bits of VA to index the second-level map */ 301 pg = pte_alloc_kernel(pd, va); 302 if (pg != 0) { 303 err = 0; 304 /* The PTE should never be already set nor present in the 305 * hash table 306 */ 307 BUG_ON((pte_val(*pg) & (_PAGE_PRESENT | _PAGE_HASHPTE)) && 308 flags); 309 set_pte_at(&init_mm, va, pg, pfn_pte(pa >> PAGE_SHIFT, 310 __pgprot(flags))); 311 } 312 smp_wmb(); 313 return err; 314 } 315 316 /* 317 * Map in a chunk of physical memory starting at start. 318 */ 319 void __init __mapin_ram_chunk(unsigned long offset, unsigned long top) 320 { 321 unsigned long v, s, f; 322 phys_addr_t p; 323 int ktext; 324 325 s = offset; 326 v = PAGE_OFFSET + s; 327 p = memstart_addr + s; 328 for (; s < top; s += PAGE_SIZE) { 329 ktext = ((char *) v >= _stext && (char *) v < etext); 330 f = ktext ? PAGE_KERNEL_TEXT : PAGE_KERNEL; 331 map_page(v, p, f); 332 #ifdef CONFIG_PPC_STD_MMU_32 333 if (ktext) 334 hash_preload(&init_mm, v, 0, 0x300); 335 #endif 336 v += PAGE_SIZE; 337 p += PAGE_SIZE; 338 } 339 } 340 341 void __init mapin_ram(void) 342 { 343 unsigned long s, top; 344 345 #ifndef CONFIG_WII 346 top = total_lowmem; 347 s = mmu_mapin_ram(top); 348 __mapin_ram_chunk(s, top); 349 #else 350 if (!wii_hole_size) { 351 s = mmu_mapin_ram(total_lowmem); 352 __mapin_ram_chunk(s, total_lowmem); 353 } else { 354 top = wii_hole_start; 355 s = mmu_mapin_ram(top); 356 __mapin_ram_chunk(s, top); 357 358 top = memblock_end_of_DRAM(); 359 s = wii_mmu_mapin_mem2(top); 360 __mapin_ram_chunk(s, top); 361 } 362 #endif 363 } 364 365 /* Scan the real Linux page tables and return a PTE pointer for 366 * a virtual address in a context. 367 * Returns true (1) if PTE was found, zero otherwise. The pointer to 368 * the PTE pointer is unmodified if PTE is not found. 369 */ 370 int 371 get_pteptr(struct mm_struct *mm, unsigned long addr, pte_t **ptep, pmd_t **pmdp) 372 { 373 pgd_t *pgd; 374 pud_t *pud; 375 pmd_t *pmd; 376 pte_t *pte; 377 int retval = 0; 378 379 pgd = pgd_offset(mm, addr & PAGE_MASK); 380 if (pgd) { 381 pud = pud_offset(pgd, addr & PAGE_MASK); 382 if (pud && pud_present(*pud)) { 383 pmd = pmd_offset(pud, addr & PAGE_MASK); 384 if (pmd_present(*pmd)) { 385 pte = pte_offset_map(pmd, addr & PAGE_MASK); 386 if (pte) { 387 retval = 1; 388 *ptep = pte; 389 if (pmdp) 390 *pmdp = pmd; 391 /* XXX caller needs to do pte_unmap, yuck */ 392 } 393 } 394 } 395 } 396 return(retval); 397 } 398 399 #ifdef CONFIG_DEBUG_PAGEALLOC 400 401 static int __change_page_attr(struct page *page, pgprot_t prot) 402 { 403 pte_t *kpte; 404 pmd_t *kpmd; 405 unsigned long address; 406 407 BUG_ON(PageHighMem(page)); 408 address = (unsigned long)page_address(page); 409 410 if (v_mapped_by_bats(address) || v_mapped_by_tlbcam(address)) 411 return 0; 412 if (!get_pteptr(&init_mm, address, &kpte, &kpmd)) 413 return -EINVAL; 414 __set_pte_at(&init_mm, address, kpte, mk_pte(page, prot), 0); 415 wmb(); 416 flush_tlb_page(NULL, address); 417 pte_unmap(kpte); 418 419 return 0; 420 } 421 422 /* 423 * Change the page attributes of an page in the linear mapping. 424 * 425 * THIS CONFLICTS WITH BAT MAPPINGS, DEBUG USE ONLY 426 */ 427 static int change_page_attr(struct page *page, int numpages, pgprot_t prot) 428 { 429 int i, err = 0; 430 unsigned long flags; 431 432 local_irq_save(flags); 433 for (i = 0; i < numpages; i++, page++) { 434 err = __change_page_attr(page, prot); 435 if (err) 436 break; 437 } 438 local_irq_restore(flags); 439 return err; 440 } 441 442 443 void __kernel_map_pages(struct page *page, int numpages, int enable) 444 { 445 if (PageHighMem(page)) 446 return; 447 448 change_page_attr(page, numpages, enable ? PAGE_KERNEL : __pgprot(0)); 449 } 450 #endif /* CONFIG_DEBUG_PAGEALLOC */ 451 452 static int fixmaps; 453 454 void __set_fixmap (enum fixed_addresses idx, phys_addr_t phys, pgprot_t flags) 455 { 456 unsigned long address = __fix_to_virt(idx); 457 458 if (idx >= __end_of_fixed_addresses) { 459 BUG(); 460 return; 461 } 462 463 map_page(address, phys, pgprot_val(flags)); 464 fixmaps++; 465 } 466 467 void __this_fixmap_does_not_exist(void) 468 { 469 WARN_ON(1); 470 } 471