Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame^] | 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 | * Amiga/APUS changes by Jesper Skov (jskov@cygnus.co.uk). |
| 12 | * |
| 13 | * Derived from "arch/i386/mm/init.c" |
| 14 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds |
| 15 | * |
| 16 | * This program is free software; you can redistribute it and/or |
| 17 | * modify it under the terms of the GNU General Public License |
| 18 | * as published by the Free Software Foundation; either version |
| 19 | * 2 of the License, or (at your option) any later version. |
| 20 | * |
| 21 | */ |
| 22 | |
| 23 | #include <linux/config.h> |
| 24 | #include <linux/kernel.h> |
| 25 | #include <linux/module.h> |
| 26 | #include <linux/types.h> |
| 27 | #include <linux/mm.h> |
| 28 | #include <linux/vmalloc.h> |
| 29 | #include <linux/init.h> |
| 30 | #include <linux/highmem.h> |
| 31 | |
| 32 | #include <asm/pgtable.h> |
| 33 | #include <asm/pgalloc.h> |
| 34 | #include <asm/io.h> |
| 35 | |
| 36 | #include "mmu_decl.h" |
| 37 | |
| 38 | unsigned long ioremap_base; |
| 39 | unsigned long ioremap_bot; |
| 40 | int io_bat_index; |
| 41 | |
| 42 | #if defined(CONFIG_6xx) || defined(CONFIG_POWER3) |
| 43 | #define HAVE_BATS 1 |
| 44 | #endif |
| 45 | |
| 46 | #if defined(CONFIG_FSL_BOOKE) |
| 47 | #define HAVE_TLBCAM 1 |
| 48 | #endif |
| 49 | |
| 50 | extern char etext[], _stext[]; |
| 51 | |
| 52 | #ifdef CONFIG_SMP |
| 53 | extern void hash_page_sync(void); |
| 54 | #endif |
| 55 | |
| 56 | #ifdef HAVE_BATS |
| 57 | extern unsigned long v_mapped_by_bats(unsigned long va); |
| 58 | extern unsigned long p_mapped_by_bats(unsigned long pa); |
| 59 | void setbat(int index, unsigned long virt, unsigned long phys, |
| 60 | unsigned int size, int flags); |
| 61 | |
| 62 | #else /* !HAVE_BATS */ |
| 63 | #define v_mapped_by_bats(x) (0UL) |
| 64 | #define p_mapped_by_bats(x) (0UL) |
| 65 | #endif /* HAVE_BATS */ |
| 66 | |
| 67 | #ifdef HAVE_TLBCAM |
| 68 | extern unsigned int tlbcam_index; |
| 69 | extern unsigned int num_tlbcam_entries; |
| 70 | extern unsigned long v_mapped_by_tlbcam(unsigned long va); |
| 71 | extern unsigned long p_mapped_by_tlbcam(unsigned long pa); |
| 72 | #else /* !HAVE_TLBCAM */ |
| 73 | #define v_mapped_by_tlbcam(x) (0UL) |
| 74 | #define p_mapped_by_tlbcam(x) (0UL) |
| 75 | #endif /* HAVE_TLBCAM */ |
| 76 | |
| 77 | #ifdef CONFIG_44x |
| 78 | /* 44x uses an 8kB pgdir because it has 8-byte Linux PTEs. */ |
| 79 | #define PGDIR_ORDER 1 |
| 80 | #else |
| 81 | #define PGDIR_ORDER 0 |
| 82 | #endif |
| 83 | |
| 84 | pgd_t *pgd_alloc(struct mm_struct *mm) |
| 85 | { |
| 86 | pgd_t *ret; |
| 87 | |
| 88 | ret = (pgd_t *)__get_free_pages(GFP_KERNEL|__GFP_ZERO, PGDIR_ORDER); |
| 89 | return ret; |
| 90 | } |
| 91 | |
| 92 | void pgd_free(pgd_t *pgd) |
| 93 | { |
| 94 | free_pages((unsigned long)pgd, PGDIR_ORDER); |
| 95 | } |
| 96 | |
| 97 | pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address) |
| 98 | { |
| 99 | pte_t *pte; |
| 100 | extern int mem_init_done; |
| 101 | extern void *early_get_page(void); |
| 102 | |
| 103 | if (mem_init_done) { |
| 104 | pte = (pte_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO); |
| 105 | } else { |
| 106 | pte = (pte_t *)early_get_page(); |
| 107 | if (pte) |
| 108 | clear_page(pte); |
| 109 | } |
| 110 | return pte; |
| 111 | } |
| 112 | |
| 113 | struct page *pte_alloc_one(struct mm_struct *mm, unsigned long address) |
| 114 | { |
| 115 | struct page *ptepage; |
| 116 | |
| 117 | #ifdef CONFIG_HIGHPTE |
| 118 | int flags = GFP_KERNEL | __GFP_HIGHMEM | __GFP_REPEAT; |
| 119 | #else |
| 120 | int flags = GFP_KERNEL | __GFP_REPEAT; |
| 121 | #endif |
| 122 | |
| 123 | ptepage = alloc_pages(flags, 0); |
| 124 | if (ptepage) |
| 125 | clear_highpage(ptepage); |
| 126 | return ptepage; |
| 127 | } |
| 128 | |
| 129 | void pte_free_kernel(pte_t *pte) |
| 130 | { |
| 131 | #ifdef CONFIG_SMP |
| 132 | hash_page_sync(); |
| 133 | #endif |
| 134 | free_page((unsigned long)pte); |
| 135 | } |
| 136 | |
| 137 | void pte_free(struct page *ptepage) |
| 138 | { |
| 139 | #ifdef CONFIG_SMP |
| 140 | hash_page_sync(); |
| 141 | #endif |
| 142 | __free_page(ptepage); |
| 143 | } |
| 144 | |
| 145 | #ifndef CONFIG_44x |
| 146 | void __iomem * |
| 147 | ioremap(phys_addr_t addr, unsigned long size) |
| 148 | { |
| 149 | return __ioremap(addr, size, _PAGE_NO_CACHE); |
| 150 | } |
| 151 | #else /* CONFIG_44x */ |
| 152 | void __iomem * |
| 153 | ioremap64(unsigned long long addr, unsigned long size) |
| 154 | { |
| 155 | return __ioremap(addr, size, _PAGE_NO_CACHE); |
| 156 | } |
| 157 | |
| 158 | void __iomem * |
| 159 | ioremap(phys_addr_t addr, unsigned long size) |
| 160 | { |
| 161 | phys_addr_t addr64 = fixup_bigphys_addr(addr, size); |
| 162 | |
| 163 | return ioremap64(addr64, size); |
| 164 | } |
| 165 | #endif /* CONFIG_44x */ |
| 166 | |
| 167 | void __iomem * |
| 168 | __ioremap(phys_addr_t addr, unsigned long size, unsigned long flags) |
| 169 | { |
| 170 | unsigned long v, i; |
| 171 | phys_addr_t p; |
| 172 | int err; |
| 173 | |
| 174 | /* |
| 175 | * Choose an address to map it to. |
| 176 | * Once the vmalloc system is running, we use it. |
| 177 | * Before then, we use space going down from ioremap_base |
| 178 | * (ioremap_bot records where we're up to). |
| 179 | */ |
| 180 | p = addr & PAGE_MASK; |
| 181 | size = PAGE_ALIGN(addr + size) - p; |
| 182 | |
| 183 | /* |
| 184 | * If the address lies within the first 16 MB, assume it's in ISA |
| 185 | * memory space |
| 186 | */ |
| 187 | if (p < 16*1024*1024) |
| 188 | p += _ISA_MEM_BASE; |
| 189 | |
| 190 | /* |
| 191 | * Don't allow anybody to remap normal RAM that we're using. |
| 192 | * mem_init() sets high_memory so only do the check after that. |
| 193 | */ |
| 194 | if ( mem_init_done && (p < virt_to_phys(high_memory)) ) |
| 195 | { |
| 196 | printk("__ioremap(): phys addr "PTE_FMT" is RAM lr %p\n", p, |
| 197 | __builtin_return_address(0)); |
| 198 | return NULL; |
| 199 | } |
| 200 | |
| 201 | if (size == 0) |
| 202 | return NULL; |
| 203 | |
| 204 | /* |
| 205 | * Is it already mapped? Perhaps overlapped by a previous |
| 206 | * BAT mapping. If the whole area is mapped then we're done, |
| 207 | * otherwise remap it since we want to keep the virt addrs for |
| 208 | * each request contiguous. |
| 209 | * |
| 210 | * We make the assumption here that if the bottom and top |
| 211 | * of the range we want are mapped then it's mapped to the |
| 212 | * same virt address (and this is contiguous). |
| 213 | * -- Cort |
| 214 | */ |
| 215 | if ((v = p_mapped_by_bats(p)) /*&& p_mapped_by_bats(p+size-1)*/ ) |
| 216 | goto out; |
| 217 | |
| 218 | if ((v = p_mapped_by_tlbcam(p))) |
| 219 | goto out; |
| 220 | |
| 221 | if (mem_init_done) { |
| 222 | struct vm_struct *area; |
| 223 | area = get_vm_area(size, VM_IOREMAP); |
| 224 | if (area == 0) |
| 225 | return NULL; |
| 226 | v = (unsigned long) area->addr; |
| 227 | } else { |
| 228 | v = (ioremap_bot -= size); |
| 229 | } |
| 230 | |
| 231 | if ((flags & _PAGE_PRESENT) == 0) |
| 232 | flags |= _PAGE_KERNEL; |
| 233 | if (flags & _PAGE_NO_CACHE) |
| 234 | flags |= _PAGE_GUARDED; |
| 235 | |
| 236 | /* |
| 237 | * Should check if it is a candidate for a BAT mapping |
| 238 | */ |
| 239 | |
| 240 | err = 0; |
| 241 | for (i = 0; i < size && err == 0; i += PAGE_SIZE) |
| 242 | err = map_page(v+i, p+i, flags); |
| 243 | if (err) { |
| 244 | if (mem_init_done) |
| 245 | vunmap((void *)v); |
| 246 | return NULL; |
| 247 | } |
| 248 | |
| 249 | out: |
| 250 | return (void __iomem *) (v + ((unsigned long)addr & ~PAGE_MASK)); |
| 251 | } |
| 252 | |
| 253 | void iounmap(volatile void __iomem *addr) |
| 254 | { |
| 255 | /* |
| 256 | * If mapped by BATs then there is nothing to do. |
| 257 | * Calling vfree() generates a benign warning. |
| 258 | */ |
| 259 | if (v_mapped_by_bats((unsigned long)addr)) return; |
| 260 | |
| 261 | if (addr > high_memory && (unsigned long) addr < ioremap_bot) |
| 262 | vunmap((void *) (PAGE_MASK & (unsigned long)addr)); |
| 263 | } |
| 264 | |
| 265 | void __iomem *ioport_map(unsigned long port, unsigned int len) |
| 266 | { |
| 267 | return (void __iomem *) (port + _IO_BASE); |
| 268 | } |
| 269 | |
| 270 | void ioport_unmap(void __iomem *addr) |
| 271 | { |
| 272 | /* Nothing to do */ |
| 273 | } |
| 274 | EXPORT_SYMBOL(ioport_map); |
| 275 | EXPORT_SYMBOL(ioport_unmap); |
| 276 | |
| 277 | int |
| 278 | 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 | spin_lock(&init_mm.page_table_lock); |
| 285 | /* Use upper 10 bits of VA to index the first level map */ |
| 286 | pd = pmd_offset(pgd_offset_k(va), va); |
| 287 | /* Use middle 10 bits of VA to index the second-level map */ |
| 288 | pg = pte_alloc_kernel(&init_mm, pd, va); |
| 289 | if (pg != 0) { |
| 290 | err = 0; |
| 291 | set_pte_at(&init_mm, va, pg, pfn_pte(pa >> PAGE_SHIFT, __pgprot(flags))); |
| 292 | if (mem_init_done) |
| 293 | flush_HPTE(0, va, pmd_val(*pd)); |
| 294 | } |
| 295 | spin_unlock(&init_mm.page_table_lock); |
| 296 | return err; |
| 297 | } |
| 298 | |
| 299 | /* |
| 300 | * Map in all of physical memory starting at KERNELBASE. |
| 301 | */ |
| 302 | void __init mapin_ram(void) |
| 303 | { |
| 304 | unsigned long v, p, s, f; |
| 305 | |
| 306 | s = mmu_mapin_ram(); |
| 307 | v = KERNELBASE + s; |
| 308 | p = PPC_MEMSTART + s; |
| 309 | for (; s < total_lowmem; s += PAGE_SIZE) { |
| 310 | if ((char *) v >= _stext && (char *) v < etext) |
| 311 | f = _PAGE_RAM_TEXT; |
| 312 | else |
| 313 | f = _PAGE_RAM; |
| 314 | map_page(v, p, f); |
| 315 | v += PAGE_SIZE; |
| 316 | p += PAGE_SIZE; |
| 317 | } |
| 318 | } |
| 319 | |
| 320 | /* is x a power of 2? */ |
| 321 | #define is_power_of_2(x) ((x) != 0 && (((x) & ((x) - 1)) == 0)) |
| 322 | |
| 323 | /* is x a power of 4? */ |
| 324 | #define is_power_of_4(x) ((x) != 0 && (((x) & (x-1)) == 0) && (ffs(x) & 1)) |
| 325 | |
| 326 | /* |
| 327 | * Set up a mapping for a block of I/O. |
| 328 | * virt, phys, size must all be page-aligned. |
| 329 | * This should only be called before ioremap is called. |
| 330 | */ |
| 331 | void __init io_block_mapping(unsigned long virt, phys_addr_t phys, |
| 332 | unsigned int size, int flags) |
| 333 | { |
| 334 | int i; |
| 335 | |
| 336 | if (virt > KERNELBASE && virt < ioremap_bot) |
| 337 | ioremap_bot = ioremap_base = virt; |
| 338 | |
| 339 | #ifdef HAVE_BATS |
| 340 | /* |
| 341 | * Use a BAT for this if possible... |
| 342 | */ |
| 343 | if (io_bat_index < 2 && is_power_of_2(size) |
| 344 | && (virt & (size - 1)) == 0 && (phys & (size - 1)) == 0) { |
| 345 | setbat(io_bat_index, virt, phys, size, flags); |
| 346 | ++io_bat_index; |
| 347 | return; |
| 348 | } |
| 349 | #endif /* HAVE_BATS */ |
| 350 | |
| 351 | #ifdef HAVE_TLBCAM |
| 352 | /* |
| 353 | * Use a CAM for this if possible... |
| 354 | */ |
| 355 | if (tlbcam_index < num_tlbcam_entries && is_power_of_4(size) |
| 356 | && (virt & (size - 1)) == 0 && (phys & (size - 1)) == 0) { |
| 357 | settlbcam(tlbcam_index, virt, phys, size, flags, 0); |
| 358 | ++tlbcam_index; |
| 359 | return; |
| 360 | } |
| 361 | #endif /* HAVE_TLBCAM */ |
| 362 | |
| 363 | /* No BATs available, put it in the page tables. */ |
| 364 | for (i = 0; i < size; i += PAGE_SIZE) |
| 365 | map_page(virt + i, phys + i, flags); |
| 366 | } |
| 367 | |
| 368 | /* Scan the real Linux page tables and return a PTE pointer for |
| 369 | * a virtual address in a context. |
| 370 | * Returns true (1) if PTE was found, zero otherwise. The pointer to |
| 371 | * the PTE pointer is unmodified if PTE is not found. |
| 372 | */ |
| 373 | int |
| 374 | get_pteptr(struct mm_struct *mm, unsigned long addr, pte_t **ptep) |
| 375 | { |
| 376 | pgd_t *pgd; |
| 377 | pmd_t *pmd; |
| 378 | pte_t *pte; |
| 379 | int retval = 0; |
| 380 | |
| 381 | pgd = pgd_offset(mm, addr & PAGE_MASK); |
| 382 | if (pgd) { |
| 383 | pmd = pmd_offset(pgd, 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 | /* XXX caller needs to do pte_unmap, yuck */ |
| 390 | } |
| 391 | } |
| 392 | } |
| 393 | return(retval); |
| 394 | } |
| 395 | |
| 396 | /* Find physical address for this virtual address. Normally used by |
| 397 | * I/O functions, but anyone can call it. |
| 398 | */ |
| 399 | unsigned long iopa(unsigned long addr) |
| 400 | { |
| 401 | unsigned long pa; |
| 402 | |
| 403 | /* I don't know why this won't work on PMacs or CHRP. It |
| 404 | * appears there is some bug, or there is some implicit |
| 405 | * mapping done not properly represented by BATs or in page |
| 406 | * tables.......I am actively working on resolving this, but |
| 407 | * can't hold up other stuff. -- Dan |
| 408 | */ |
| 409 | pte_t *pte; |
| 410 | struct mm_struct *mm; |
| 411 | |
| 412 | /* Check the BATs */ |
| 413 | pa = v_mapped_by_bats(addr); |
| 414 | if (pa) |
| 415 | return pa; |
| 416 | |
| 417 | /* Allow mapping of user addresses (within the thread) |
| 418 | * for DMA if necessary. |
| 419 | */ |
| 420 | if (addr < TASK_SIZE) |
| 421 | mm = current->mm; |
| 422 | else |
| 423 | mm = &init_mm; |
| 424 | |
| 425 | pa = 0; |
| 426 | if (get_pteptr(mm, addr, &pte)) { |
| 427 | pa = (pte_val(*pte) & PAGE_MASK) | (addr & ~PAGE_MASK); |
| 428 | pte_unmap(pte); |
| 429 | } |
| 430 | |
| 431 | return(pa); |
| 432 | } |
| 433 | |
| 434 | /* This is will find the virtual address for a physical one.... |
| 435 | * Swiped from APUS, could be dangerous :-). |
| 436 | * This is only a placeholder until I really find a way to make this |
| 437 | * work. -- Dan |
| 438 | */ |
| 439 | unsigned long |
| 440 | mm_ptov (unsigned long paddr) |
| 441 | { |
| 442 | unsigned long ret; |
| 443 | #if 0 |
| 444 | if (paddr < 16*1024*1024) |
| 445 | ret = ZTWO_VADDR(paddr); |
| 446 | else { |
| 447 | int i; |
| 448 | |
| 449 | for (i = 0; i < kmap_chunk_count;){ |
| 450 | unsigned long phys = kmap_chunks[i++]; |
| 451 | unsigned long size = kmap_chunks[i++]; |
| 452 | unsigned long virt = kmap_chunks[i++]; |
| 453 | if (paddr >= phys |
| 454 | && paddr < (phys + size)){ |
| 455 | ret = virt + paddr - phys; |
| 456 | goto exit; |
| 457 | } |
| 458 | } |
| 459 | |
| 460 | ret = (unsigned long) __va(paddr); |
| 461 | } |
| 462 | exit: |
| 463 | #ifdef DEBUGPV |
| 464 | printk ("PTOV(%lx)=%lx\n", paddr, ret); |
| 465 | #endif |
| 466 | #else |
| 467 | ret = (unsigned long)paddr + KERNELBASE; |
| 468 | #endif |
| 469 | return ret; |
| 470 | } |
| 471 | |