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 #include <asm/sections.h>
  38
  39 #include "mmu_decl.h"
  40
  41 unsigned long ioremap_bot;
  42 EXPORT_SYMBOL(ioremap_bot);     /* aka VMALLOC_END */
  43
  44 extern char etext[], _stext[], _sinittext[], _einittext[];
  45
  46 __ref pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
  47 {
  48         pte_t *pte;
  49
  50         if (slab_is_available()) {
  51                 pte = (pte_t *)__get_free_page(GFP_KERNEL|__GFP_ZERO);
  52         } else {
  53                 pte = __va(memblock_alloc(PAGE_SIZE, PAGE_SIZE));
  54                 if (pte)
  55                         clear_page(pte);
  56         }
  57         return pte;
  58 }
  59
  60 pgtable_t pte_alloc_one(struct mm_struct *mm, unsigned long address)
  61 {
  62         struct page *ptepage;
  63
  64         gfp_t flags = GFP_KERNEL | __GFP_ZERO | __GFP_ACCOUNT;
  65
  66         ptepage = alloc_pages(flags, 0);
  67         if (!ptepage)
  68                 return NULL;
  69         if (!pgtable_page_ctor(ptepage)) {
  70                 __free_page(ptepage);
  71                 return NULL;
  72         }
  73         return ptepage;
  74 }
  75
  76 void __iomem *
  77 ioremap(phys_addr_t addr, unsigned long size)
  78 {
  79         return __ioremap_caller(addr, size, _PAGE_NO_CACHE | _PAGE_GUARDED,
  80                                 __builtin_return_address(0));
  81 }
  82 EXPORT_SYMBOL(ioremap);
  83
  84 void __iomem *
  85 ioremap_wc(phys_addr_t addr, unsigned long size)
  86 {
  87         return __ioremap_caller(addr, size, _PAGE_NO_CACHE,
  88                                 __builtin_return_address(0));
  89 }
  90 EXPORT_SYMBOL(ioremap_wc);
  91
  92 void __iomem *
  93 ioremap_prot(phys_addr_t addr, unsigned long size, unsigned long flags)
  94 {
  95         /* writeable implies dirty for kernel addresses */
  96         if ((flags & (_PAGE_RW | _PAGE_RO)) != _PAGE_RO)
  97                 flags |= _PAGE_DIRTY | _PAGE_HWWRITE;
  98
  99         /* we don't want to let _PAGE_USER and _PAGE_EXEC leak out */
 100         flags &= ~(_PAGE_USER | _PAGE_EXEC);
 101         flags |= _PAGE_PRIVILEGED;
 102
 103         return __ioremap_caller(addr, size, flags, __builtin_return_address(0));
 104 }
 105 EXPORT_SYMBOL(ioremap_prot);
 106
 107 void __iomem *
 108 __ioremap(phys_addr_t addr, unsigned long size, unsigned long flags)
 109 {
 110         return __ioremap_caller(addr, size, flags, __builtin_return_address(0));
 111 }
 112
 113 void __iomem *
 114 __ioremap_caller(phys_addr_t addr, unsigned long size, unsigned long flags,
 115                  void *caller)
 116 {
 117         unsigned long v, i;
 118         phys_addr_t p;
 119         int err;
 120
 121         /* Make sure we have the base flags */
 122         if ((flags & _PAGE_PRESENT) == 0)
 123                 flags |= pgprot_val(PAGE_KERNEL);
 124
 125         /* Non-cacheable page cannot be coherent */
 126         if (flags & _PAGE_NO_CACHE)
 127                 flags &= ~_PAGE_COHERENT;
 128
 129         /*
 130          * Choose an address to map it to.
 131          * Once the vmalloc system is running, we use it.
 132          * Before then, we use space going down from IOREMAP_TOP
 133          * (ioremap_bot records where we're up to).
 134          */
 135         p = addr & PAGE_MASK;
 136         size = PAGE_ALIGN(addr + size) - p;
 137
 138         /*
 139          * If the address lies within the first 16 MB, assume it's in ISA
 140          * memory space
 141          */
 142         if (p < 16*1024*1024)
 143                 p += _ISA_MEM_BASE;
 144
 145 #ifndef CONFIG_CRASH_DUMP
 146         /*
 147          * Don't allow anybody to remap normal RAM that we're using.
 148          * mem_init() sets high_memory so only do the check after that.
 149          */
 150         if (slab_is_available() && (p < virt_to_phys(high_memory)) &&
 151             !(__allow_ioremap_reserved && memblock_is_region_reserved(p, size))) {
 152                 printk("__ioremap(): phys addr 0x%llx is RAM lr %ps\n",
 153                        (unsigned long long)p, __builtin_return_address(0));
 154                 return NULL;
 155         }
 156 #endif
 157
 158         if (size == 0)
 159                 return NULL;
 160
 161         /*
 162          * Is it already mapped?  Perhaps overlapped by a previous
 163          * mapping.
 164          */
 165         v = p_block_mapped(p);
 166         if (v)
 167                 goto out;
 168
 169         if (slab_is_available()) {
 170                 struct vm_struct *area;
 171                 area = get_vm_area_caller(size, VM_IOREMAP, caller);
 172                 if (area == 0)
 173                         return NULL;
 174                 area->phys_addr = p;
 175                 v = (unsigned long) area->addr;
 176         } else {
 177                 v = (ioremap_bot -= size);
 178         }
 179
 180         /*
 181          * Should check if it is a candidate for a BAT mapping
 182          */
 183
 184         err = 0;
 185         for (i = 0; i < size && err == 0; i += PAGE_SIZE)
 186                 err = map_kernel_page(v+i, p+i, flags);
 187         if (err) {
 188                 if (slab_is_available())
 189                         vunmap((void *)v);
 190                 return NULL;
 191         }
 192
 193 out:
 194         return (void __iomem *) (v + ((unsigned long)addr & ~PAGE_MASK));
 195 }
 196 EXPORT_SYMBOL(__ioremap);
 197
 198 void iounmap(volatile void __iomem *addr)
 199 {
 200         /*
 201          * If mapped by BATs then there is nothing to do.
 202          * Calling vfree() generates a benign warning.
 203          */
 204         if (v_block_mapped((unsigned long)addr))
 205                 return;
 206
 207         if (addr > high_memory && (unsigned long) addr < ioremap_bot)
 208                 vunmap((void *) (PAGE_MASK & (unsigned long)addr));
 209 }
 210 EXPORT_SYMBOL(iounmap);
 211
 212 int map_kernel_page(unsigned long va, phys_addr_t pa, int flags)
 213 {
 214         pmd_t *pd;
 215         pte_t *pg;
 216         int err = -ENOMEM;
 217
 218         /* Use upper 10 bits of VA to index the first level map */
 219         pd = pmd_offset(pud_offset(pgd_offset_k(va), va), va);
 220         /* Use middle 10 bits of VA to index the second-level map */
 221         pg = pte_alloc_kernel(pd, va);
 222         if (pg != 0) {
 223                 err = 0;
 224                 /* The PTE should never be already set nor present in the
 225                  * hash table
 226                  */
 227                 BUG_ON((pte_val(*pg) & (_PAGE_PRESENT | _PAGE_HASHPTE)) &&
 228                        flags);
 229                 set_pte_at(&init_mm, va, pg, pfn_pte(pa >> PAGE_SHIFT,
 230                                                      __pgprot(flags)));
 231         }
 232         smp_wmb();
 233         return err;
 234 }
 235
 236 /*
 237  * Map in a chunk of physical memory starting at start.
 238  */
 239 static void __init __mapin_ram_chunk(unsigned long offset, unsigned long top)
 240 {
 241         unsigned long v, s, f;
 242         phys_addr_t p;
 243         int ktext;
 244
 245         s = offset;
 246         v = PAGE_OFFSET + s;
 247         p = memstart_addr + s;
 248         for (; s < top; s += PAGE_SIZE) {
 249                 ktext = ((char *)v >= _stext && (char *)v < etext) ||
 250                         ((char *)v >= _sinittext && (char *)v < _einittext);
 251                 f = ktext ? pgprot_val(PAGE_KERNEL_TEXT) : pgprot_val(PAGE_KERNEL);
 252                 map_kernel_page(v, p, f);
 253 #ifdef CONFIG_PPC_STD_MMU_32
 254                 if (ktext)
 255                         hash_preload(&init_mm, v, 0, 0x300);
 256 #endif
 257                 v += PAGE_SIZE;
 258                 p += PAGE_SIZE;
 259         }
 260 }
 261
 262 void __init mapin_ram(void)
 263 {
 264         unsigned long s, top;
 265
 266 #ifndef CONFIG_WII
 267         top = total_lowmem;
 268         s = mmu_mapin_ram(top);
 269         __mapin_ram_chunk(s, top);
 270 #else
 271         if (!wii_hole_size) {
 272                 s = mmu_mapin_ram(total_lowmem);
 273                 __mapin_ram_chunk(s, total_lowmem);
 274         } else {
 275                 top = wii_hole_start;
 276                 s = mmu_mapin_ram(top);
 277                 __mapin_ram_chunk(s, top);
 278
 279                 top = memblock_end_of_DRAM();
 280                 s = wii_mmu_mapin_mem2(top);
 281                 __mapin_ram_chunk(s, top);
 282         }
 283 #endif
 284 }
 285
 286 /* Scan the real Linux page tables and return a PTE pointer for
 287  * a virtual address in a context.
 288  * Returns true (1) if PTE was found, zero otherwise.  The pointer to
 289  * the PTE pointer is unmodified if PTE is not found.
 290  */
 291 static int
 292 get_pteptr(struct mm_struct *mm, unsigned long addr, pte_t **ptep, pmd_t **pmdp)
 293 {
 294         pgd_t   *pgd;
 295         pud_t   *pud;
 296         pmd_t   *pmd;
 297         pte_t   *pte;
 298         int     retval = 0;
 299
 300         pgd = pgd_offset(mm, addr & PAGE_MASK);
 301         if (pgd) {
 302                 pud = pud_offset(pgd, addr & PAGE_MASK);
 303                 if (pud && pud_present(*pud)) {
 304                         pmd = pmd_offset(pud, addr & PAGE_MASK);
 305                         if (pmd_present(*pmd)) {
 306                                 pte = pte_offset_map(pmd, addr & PAGE_MASK);
 307                                 if (pte) {
 308                                         retval = 1;
 309                                         *ptep = pte;
 310                                         if (pmdp)
 311                                                 *pmdp = pmd;
 312                                         /* XXX caller needs to do pte_unmap, yuck */
 313                                 }
 314                         }
 315                 }
 316         }
 317         return(retval);
 318 }
 319
 320 static int __change_page_attr_noflush(struct page *page, pgprot_t prot)
 321 {
 322         pte_t *kpte;
 323         pmd_t *kpmd;
 324         unsigned long address;
 325
 326         BUG_ON(PageHighMem(page));
 327         address = (unsigned long)page_address(page);
 328
 329         if (v_block_mapped(address))
 330                 return 0;
 331         if (!get_pteptr(&init_mm, address, &kpte, &kpmd))
 332                 return -EINVAL;
 333         __set_pte_at(&init_mm, address, kpte, mk_pte(page, prot), 0);
 334         pte_unmap(kpte);
 335
 336         return 0;
 337 }
 338
 339 /*
 340  * Change the page attributes of an page in the linear mapping.
 341  *
 342  * THIS DOES NOTHING WITH BAT MAPPINGS, DEBUG USE ONLY
 343  */
 344 static int change_page_attr(struct page *page, int numpages, pgprot_t prot)
 345 {
 346         int i, err = 0;
 347         unsigned long flags;
 348         struct page *start = page;
 349
 350         local_irq_save(flags);
 351         for (i = 0; i < numpages; i++, page++) {
 352                 err = __change_page_attr_noflush(page, prot);
 353                 if (err)
 354                         break;
 355         }
 356         wmb();
 357         local_irq_restore(flags);
 358         flush_tlb_kernel_range((unsigned long)page_address(start),
 359                                (unsigned long)page_address(page));
 360         return err;
 361 }
 362
 363 void mark_initmem_nx(void)
 364 {
 365         struct page *page = virt_to_page(_sinittext);
 366         unsigned long numpages = PFN_UP((unsigned long)_einittext) -
 367                                  PFN_DOWN((unsigned long)_sinittext);
 368
 369         change_page_attr(page, numpages, PAGE_KERNEL);
 370 }
 371
 372 #ifdef CONFIG_STRICT_KERNEL_RWX
 373 void mark_rodata_ro(void)
 374 {
 375         struct page *page;
 376         unsigned long numpages;
 377
 378         page = virt_to_page(_stext);
 379         numpages = PFN_UP((unsigned long)_etext) -
 380                    PFN_DOWN((unsigned long)_stext);
 381
 382         change_page_attr(page, numpages, PAGE_KERNEL_ROX);
 383         /*
 384          * mark .rodata as read only. Use __init_begin rather than __end_rodata
 385          * to cover NOTES and EXCEPTION_TABLE.
 386          */
 387         page = virt_to_page(__start_rodata);
 388         numpages = PFN_UP((unsigned long)__init_begin) -
 389                    PFN_DOWN((unsigned long)__start_rodata);
 390
 391         change_page_attr(page, numpages, PAGE_KERNEL_RO);
 392 }
 393 #endif
 394
 395 #ifdef CONFIG_DEBUG_PAGEALLOC
 396 void __kernel_map_pages(struct page *page, int numpages, int enable)
 397 {
 398         if (PageHighMem(page))
 399                 return;
 400
 401         change_page_attr(page, numpages, enable ? PAGE_KERNEL : __pgprot(0));
 402 }
 403 #endif /* CONFIG_DEBUG_PAGEALLOC */