arch/powerpc/mm/pgtable_32.c

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