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 37 #include "mmu_decl.h" 38 39 unsigned long ioremap_base; 40 unsigned long ioremap_bot; 41 EXPORT_SYMBOL(ioremap_bot); /* aka VMALLOC_END */ 42 43 #if defined(CONFIG_6xx) || defined(CONFIG_POWER3) 44 #define HAVE_BATS 1 45 #endif 46 47 #if defined(CONFIG_FSL_BOOKE) 48 #define HAVE_TLBCAM 1 49 #endif 50 51 extern char etext[], _stext[]; 52 53 #ifdef HAVE_BATS 54 extern phys_addr_t v_mapped_by_bats(unsigned long va); 55 extern unsigned long p_mapped_by_bats(phys_addr_t pa); 56 void setbat(int index, unsigned long virt, phys_addr_t phys, 57 unsigned int size, int flags); 58 59 #else /* !HAVE_BATS */ 60 #define v_mapped_by_bats(x) (0UL) 61 #define p_mapped_by_bats(x) (0UL) 62 #endif /* HAVE_BATS */ 63 64 #ifdef HAVE_TLBCAM 65 extern unsigned int tlbcam_index; 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 pgd_t *pgd_alloc(struct mm_struct *mm) 76 { 77 pgd_t *ret; 78 79 /* pgdir take page or two with 4K pages and a page fraction otherwise */ 80 #ifndef CONFIG_PPC_4K_PAGES 81 ret = kzalloc(1 << PGDIR_ORDER, GFP_KERNEL); 82 #else 83 ret = (pgd_t *)__get_free_pages(GFP_KERNEL|__GFP_ZERO, 84 PGDIR_ORDER - PAGE_SHIFT); 85 #endif 86 return ret; 87 } 88 89 void pgd_free(struct mm_struct *mm, pgd_t *pgd) 90 { 91 #ifndef CONFIG_PPC_4K_PAGES 92 kfree((void *)pgd); 93 #else 94 free_pages((unsigned long)pgd, PGDIR_ORDER - PAGE_SHIFT); 95 #endif 96 } 97 98 __init_refok pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address) 99 { 100 pte_t *pte; 101 extern int mem_init_done; 102 extern void *early_get_page(void); 103 104 if (mem_init_done) { 105 pte = (pte_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO); 106 } else { 107 pte = (pte_t *)early_get_page(); 108 if (pte) 109 clear_page(pte); 110 } 111 return pte; 112 } 113 114 pgtable_t pte_alloc_one(struct mm_struct *mm, unsigned long address) 115 { 116 struct page *ptepage; 117 118 gfp_t flags = GFP_KERNEL | __GFP_REPEAT | __GFP_ZERO; 119 120 ptepage = alloc_pages(flags, 0); 121 if (!ptepage) 122 return NULL; 123 pgtable_page_ctor(ptepage); 124 return ptepage; 125 } 126 127 void __iomem * 128 ioremap(phys_addr_t addr, unsigned long size) 129 { 130 return __ioremap_caller(addr, size, _PAGE_NO_CACHE | _PAGE_GUARDED, 131 __builtin_return_address(0)); 132 } 133 EXPORT_SYMBOL(ioremap); 134 135 void __iomem * 136 ioremap_wc(phys_addr_t addr, unsigned long size) 137 { 138 return __ioremap_caller(addr, size, _PAGE_NO_CACHE, 139 __builtin_return_address(0)); 140 } 141 EXPORT_SYMBOL(ioremap_wc); 142 143 void __iomem * 144 ioremap_prot(phys_addr_t addr, unsigned long size, unsigned long flags) 145 { 146 /* writeable implies dirty for kernel addresses */ 147 if (flags & _PAGE_RW) 148 flags |= _PAGE_DIRTY | _PAGE_HWWRITE; 149 150 /* we don't want to let _PAGE_USER and _PAGE_EXEC leak out */ 151 flags &= ~(_PAGE_USER | _PAGE_EXEC); 152 153 #ifdef _PAGE_BAP_SR 154 /* _PAGE_USER contains _PAGE_BAP_SR on BookE using the new PTE format 155 * which means that we just cleared supervisor access... oops ;-) This 156 * restores it 157 */ 158 flags |= _PAGE_BAP_SR; 159 #endif 160 161 return __ioremap_caller(addr, size, flags, __builtin_return_address(0)); 162 } 163 EXPORT_SYMBOL(ioremap_prot); 164 165 void __iomem * 166 __ioremap(phys_addr_t addr, unsigned long size, unsigned long flags) 167 { 168 return __ioremap_caller(addr, size, flags, __builtin_return_address(0)); 169 } 170 171 void __iomem * 172 __ioremap_caller(phys_addr_t addr, unsigned long size, unsigned long flags, 173 void *caller) 174 { 175 unsigned long v, i; 176 phys_addr_t p; 177 int err; 178 179 /* Make sure we have the base flags */ 180 if ((flags & _PAGE_PRESENT) == 0) 181 flags |= PAGE_KERNEL; 182 183 /* Non-cacheable page cannot be coherent */ 184 if (flags & _PAGE_NO_CACHE) 185 flags &= ~_PAGE_COHERENT; 186 187 /* 188 * Choose an address to map it to. 189 * Once the vmalloc system is running, we use it. 190 * Before then, we use space going down from ioremap_base 191 * (ioremap_bot records where we're up to). 192 */ 193 p = addr & PAGE_MASK; 194 size = PAGE_ALIGN(addr + size) - p; 195 196 /* 197 * If the address lies within the first 16 MB, assume it's in ISA 198 * memory space 199 */ 200 if (p < 16*1024*1024) 201 p += _ISA_MEM_BASE; 202 203 #ifndef CONFIG_CRASH_DUMP 204 /* 205 * Don't allow anybody to remap normal RAM that we're using. 206 * mem_init() sets high_memory so only do the check after that. 207 */ 208 if (mem_init_done && (p < virt_to_phys(high_memory)) && 209 !(__allow_ioremap_reserved && memblock_is_region_reserved(p, size))) { 210 printk("__ioremap(): phys addr 0x%llx is RAM lr %p\n", 211 (unsigned long long)p, __builtin_return_address(0)); 212 return NULL; 213 } 214 #endif 215 216 if (size == 0) 217 return NULL; 218 219 /* 220 * Is it already mapped? Perhaps overlapped by a previous 221 * BAT mapping. If the whole area is mapped then we're done, 222 * otherwise remap it since we want to keep the virt addrs for 223 * each request contiguous. 224 * 225 * We make the assumption here that if the bottom and top 226 * of the range we want are mapped then it's mapped to the 227 * same virt address (and this is contiguous). 228 * -- Cort 229 */ 230 if ((v = p_mapped_by_bats(p)) /*&& p_mapped_by_bats(p+size-1)*/ ) 231 goto out; 232 233 if ((v = p_mapped_by_tlbcam(p))) 234 goto out; 235 236 if (mem_init_done) { 237 struct vm_struct *area; 238 area = get_vm_area_caller(size, VM_IOREMAP, caller); 239 if (area == 0) 240 return NULL; 241 area->phys_addr = p; 242 v = (unsigned long) area->addr; 243 } else { 244 v = (ioremap_bot -= size); 245 } 246 247 /* 248 * Should check if it is a candidate for a BAT mapping 249 */ 250 251 err = 0; 252 for (i = 0; i < size && err == 0; i += PAGE_SIZE) 253 err = map_page(v+i, p+i, flags); 254 if (err) { 255 if (mem_init_done) 256 vunmap((void *)v); 257 return NULL; 258 } 259 260 out: 261 return (void __iomem *) (v + ((unsigned long)addr & ~PAGE_MASK)); 262 } 263 EXPORT_SYMBOL(__ioremap); 264 265 void iounmap(volatile void __iomem *addr) 266 { 267 /* 268 * If mapped by BATs then there is nothing to do. 269 * Calling vfree() generates a benign warning. 270 */ 271 if (v_mapped_by_bats((unsigned long)addr)) return; 272 273 if (addr > high_memory && (unsigned long) addr < ioremap_bot) 274 vunmap((void *) (PAGE_MASK & (unsigned long)addr)); 275 } 276 EXPORT_SYMBOL(iounmap); 277 278 int map_page(unsigned long va, phys_addr_t pa, int flags) 279 { 280 pmd_t *pd; 281 pte_t *pg; 282 int err = -ENOMEM; 283 284 /* Use upper 10 bits of VA to index the first level map */ 285 pd = pmd_offset(pud_offset(pgd_offset_k(va), va), va); 286 /* Use middle 10 bits of VA to index the second-level map */ 287 pg = pte_alloc_kernel(pd, va); 288 if (pg != 0) { 289 err = 0; 290 /* The PTE should never be already set nor present in the 291 * hash table 292 */ 293 BUG_ON((pte_val(*pg) & (_PAGE_PRESENT | _PAGE_HASHPTE)) && 294 flags); 295 set_pte_at(&init_mm, va, pg, pfn_pte(pa >> PAGE_SHIFT, 296 __pgprot(flags))); 297 } 298 return err; 299 } 300 301 /* 302 * Map in a chunk of physical memory starting at start. 303 */ 304 void __init __mapin_ram_chunk(unsigned long offset, unsigned long top) 305 { 306 unsigned long v, s, f; 307 phys_addr_t p; 308 int ktext; 309 310 s = offset; 311 v = PAGE_OFFSET + s; 312 p = memstart_addr + s; 313 for (; s < top; s += PAGE_SIZE) { 314 ktext = ((char *) v >= _stext && (char *) v < etext); 315 f = ktext ? PAGE_KERNEL_TEXT : PAGE_KERNEL; 316 map_page(v, p, f); 317 #ifdef CONFIG_PPC_STD_MMU_32 318 if (ktext) 319 hash_preload(&init_mm, v, 0, 0x300); 320 #endif 321 v += PAGE_SIZE; 322 p += PAGE_SIZE; 323 } 324 } 325 326 void __init mapin_ram(void) 327 { 328 unsigned long s, top; 329 330 #ifndef CONFIG_WII 331 top = total_lowmem; 332 s = mmu_mapin_ram(top); 333 __mapin_ram_chunk(s, top); 334 #else 335 if (!wii_hole_size) { 336 s = mmu_mapin_ram(total_lowmem); 337 __mapin_ram_chunk(s, total_lowmem); 338 } else { 339 top = wii_hole_start; 340 s = mmu_mapin_ram(top); 341 __mapin_ram_chunk(s, top); 342 343 top = memblock_end_of_DRAM(); 344 s = wii_mmu_mapin_mem2(top); 345 __mapin_ram_chunk(s, top); 346 } 347 #endif 348 } 349 350 /* Scan the real Linux page tables and return a PTE pointer for 351 * a virtual address in a context. 352 * Returns true (1) if PTE was found, zero otherwise. The pointer to 353 * the PTE pointer is unmodified if PTE is not found. 354 */ 355 int 356 get_pteptr(struct mm_struct *mm, unsigned long addr, pte_t **ptep, pmd_t **pmdp) 357 { 358 pgd_t *pgd; 359 pud_t *pud; 360 pmd_t *pmd; 361 pte_t *pte; 362 int retval = 0; 363 364 pgd = pgd_offset(mm, addr & PAGE_MASK); 365 if (pgd) { 366 pud = pud_offset(pgd, addr & PAGE_MASK); 367 if (pud && pud_present(*pud)) { 368 pmd = pmd_offset(pud, addr & PAGE_MASK); 369 if (pmd_present(*pmd)) { 370 pte = pte_offset_map(pmd, addr & PAGE_MASK); 371 if (pte) { 372 retval = 1; 373 *ptep = pte; 374 if (pmdp) 375 *pmdp = pmd; 376 /* XXX caller needs to do pte_unmap, yuck */ 377 } 378 } 379 } 380 } 381 return(retval); 382 } 383 384 #ifdef CONFIG_DEBUG_PAGEALLOC 385 386 static int __change_page_attr(struct page *page, pgprot_t prot) 387 { 388 pte_t *kpte; 389 pmd_t *kpmd; 390 unsigned long address; 391 392 BUG_ON(PageHighMem(page)); 393 address = (unsigned long)page_address(page); 394 395 if (v_mapped_by_bats(address) || v_mapped_by_tlbcam(address)) 396 return 0; 397 if (!get_pteptr(&init_mm, address, &kpte, &kpmd)) 398 return -EINVAL; 399 __set_pte_at(&init_mm, address, kpte, mk_pte(page, prot), 0); 400 wmb(); 401 flush_tlb_page(NULL, address); 402 pte_unmap(kpte); 403 404 return 0; 405 } 406 407 /* 408 * Change the page attributes of an page in the linear mapping. 409 * 410 * THIS CONFLICTS WITH BAT MAPPINGS, DEBUG USE ONLY 411 */ 412 static int change_page_attr(struct page *page, int numpages, pgprot_t prot) 413 { 414 int i, err = 0; 415 unsigned long flags; 416 417 local_irq_save(flags); 418 for (i = 0; i < numpages; i++, page++) { 419 err = __change_page_attr(page, prot); 420 if (err) 421 break; 422 } 423 local_irq_restore(flags); 424 return err; 425 } 426 427 428 void kernel_map_pages(struct page *page, int numpages, int enable) 429 { 430 if (PageHighMem(page)) 431 return; 432 433 change_page_attr(page, numpages, enable ? PAGE_KERNEL : __pgprot(0)); 434 } 435 #endif /* CONFIG_DEBUG_PAGEALLOC */ 436 437 static int fixmaps; 438 439 void __set_fixmap (enum fixed_addresses idx, phys_addr_t phys, pgprot_t flags) 440 { 441 unsigned long address = __fix_to_virt(idx); 442 443 if (idx >= __end_of_fixed_addresses) { 444 BUG(); 445 return; 446 } 447 448 map_page(address, phys, pgprot_val(flags)); 449 fixmaps++; 450 } 451 452 void __this_fixmap_does_not_exist(void) 453 { 454 WARN_ON(1); 455 } 456