arch/ppc/mm/pgtable.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  *  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/kernel.h>
  24 #include <linux/module.h>
  25 #include <linux/types.h>
  26 #include <linux/mm.h>
  27 #include <linux/vmalloc.h>
  28 #include <linux/init.h>
  29 #include <linux/highmem.h>
  30
  31 #include <asm/pgtable.h>
  32 #include <asm/pgalloc.h>
  33 #include <asm/io.h>
  34
  35 #include "mmu_decl.h"
  36
  37 unsigned long ioremap_base;
  38 unsigned long ioremap_bot;
  39 int io_bat_index;
  40
  41 #if defined(CONFIG_6xx)
  42 #define HAVE_BATS       1
  43 #endif
  44
  45 extern char etext[], _stext[];
  46
  47 #ifdef CONFIG_SMP
  48 extern void hash_page_sync(void);
  49 #endif
  50
  51 #ifdef HAVE_BATS
  52 extern unsigned long v_mapped_by_bats(unsigned long va);
  53 extern unsigned long p_mapped_by_bats(unsigned long pa);
  54 void setbat(int index, unsigned long virt, unsigned long phys,
  55             unsigned int size, int flags);
  56
  57 #else /* !HAVE_BATS */
  58 #define v_mapped_by_bats(x)     (0UL)
  59 #define p_mapped_by_bats(x)     (0UL)
  60 #endif /* HAVE_BATS */
  61
  62 #ifdef CONFIG_PTE_64BIT
  63 /* 44x uses an 8kB pgdir because it has 8-byte Linux PTEs. */
  64 #define PGDIR_ORDER     1
  65 #else
  66 #define PGDIR_ORDER     0
  67 #endif
  68
  69 pgd_t *pgd_alloc(struct mm_struct *mm)
  70 {
  71         pgd_t *ret;
  72
  73         ret = (pgd_t *)__get_free_pages(GFP_KERNEL|__GFP_ZERO, PGDIR_ORDER);
  74         return ret;
  75 }
  76
  77 void pgd_free(struct mm_struct *mm, pgd_t *pgd)
  78 {
  79         free_pages((unsigned long)pgd, PGDIR_ORDER);
  80 }
  81
  82 __init_refok pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
  83 {
  84         pte_t *pte;
  85         extern int mem_init_done;
  86         extern void *early_get_page(void);
  87
  88         if (mem_init_done) {
  89                 pte = (pte_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO);
  90         } else {
  91                 pte = (pte_t *)early_get_page();
  92                 if (pte)
  93                         clear_page(pte);
  94         }
  95         return pte;
  96 }
  97
  98 pgtable_t pte_alloc_one(struct mm_struct *mm, unsigned long address)
  99 {
 100         struct page *ptepage;
 101
 102 #ifdef CONFIG_HIGHPTE
 103         gfp_t flags = GFP_KERNEL | __GFP_HIGHMEM | __GFP_REPEAT;
 104 #else
 105         gfp_t flags = GFP_KERNEL | __GFP_REPEAT;
 106 #endif
 107
 108         ptepage = alloc_pages(flags, 0);
 109         if (ptepage) {
 110                 clear_highpage(ptepage);
 111                 pgtable_page_ctor(ptepage);
 112         }
 113         return ptepage;
 114 }
 115
 116 void pte_free_kernel(struct mm_struct *mm, pte_t *pte)
 117 {
 118 #ifdef CONFIG_SMP
 119         hash_page_sync();
 120 #endif
 121         free_page((unsigned long)pte);
 122 }
 123
 124 void pte_free(struct mm_struct *mm, pgtable_t ptepage)
 125 {
 126 #ifdef CONFIG_SMP
 127         hash_page_sync();
 128 #endif
 129         pgtable_page_dtor(ptepage);
 130         __free_page(ptepage);
 131 }
 132
 133 #ifndef CONFIG_PHYS_64BIT
 134 void __iomem *
 135 ioremap(phys_addr_t addr, unsigned long size)
 136 {
 137         return __ioremap(addr, size, _PAGE_NO_CACHE);
 138 }
 139 #else /* CONFIG_PHYS_64BIT */
 140 void __iomem *
 141 ioremap64(unsigned long long addr, unsigned long size)
 142 {
 143         return __ioremap(addr, size, _PAGE_NO_CACHE);
 144 }
 145
 146 void __iomem *
 147 ioremap(phys_addr_t addr, unsigned long size)
 148 {
 149         phys_addr_t addr64 = fixup_bigphys_addr(addr, size);
 150
 151         return ioremap64(addr64, size);
 152 }
 153 #endif /* CONFIG_PHYS_64BIT */
 154
 155 void __iomem *
 156 __ioremap(phys_addr_t addr, unsigned long size, unsigned long flags)
 157 {
 158         unsigned long v, i;
 159         phys_addr_t p;
 160         int err;
 161
 162         /*
 163          * Choose an address to map it to.
 164          * Once the vmalloc system is running, we use it.
 165          * Before then, we use space going down from ioremap_base
 166          * (ioremap_bot records where we're up to).
 167          */
 168         p = addr & PAGE_MASK;
 169         size = PAGE_ALIGN(addr + size) - p;
 170
 171         /*
 172          * If the address lies within the first 16 MB, assume it's in ISA
 173          * memory space
 174          */
 175         if (p < 16*1024*1024)
 176                 p += _ISA_MEM_BASE;
 177
 178         /*
 179          * Don't allow anybody to remap normal RAM that we're using.
 180          * mem_init() sets high_memory so only do the check after that.
 181          */
 182         if ( mem_init_done && (p < virt_to_phys(high_memory)) )
 183         {
 184                 printk("__ioremap(): phys addr "PHYS_FMT" is RAM lr %p\n", p,
 185                        __builtin_return_address(0));
 186                 return NULL;
 187         }
 188
 189         if (size == 0)
 190                 return NULL;
 191
 192         /*
 193          * Is it already mapped?  Perhaps overlapped by a previous
 194          * BAT mapping.  If the whole area is mapped then we're done,
 195          * otherwise remap it since we want to keep the virt addrs for
 196          * each request contiguous.
 197          *
 198          * We make the assumption here that if the bottom and top
 199          * of the range we want are mapped then it's mapped to the
 200          * same virt address (and this is contiguous).
 201          *  -- Cort
 202          */
 203         if ((v = p_mapped_by_bats(p)) /*&& p_mapped_by_bats(p+size-1)*/ )
 204                 goto out;
 205
 206         if (mem_init_done) {
 207                 struct vm_struct *area;
 208                 area = get_vm_area(size, VM_IOREMAP);
 209                 if (area == 0)
 210                         return NULL;
 211                 v = (unsigned long) area->addr;
 212         } else {
 213                 v = (ioremap_bot -= size);
 214         }
 215
 216         if ((flags & _PAGE_PRESENT) == 0)
 217                 flags |= _PAGE_KERNEL;
 218         if (flags & _PAGE_NO_CACHE)
 219                 flags |= _PAGE_GUARDED;
 220
 221         /*
 222          * Should check if it is a candidate for a BAT mapping
 223          */
 224
 225         err = 0;
 226         for (i = 0; i < size && err == 0; i += PAGE_SIZE)
 227                 err = map_page(v+i, p+i, flags);
 228         if (err) {
 229                 if (mem_init_done)
 230                         vunmap((void *)v);
 231                 return NULL;
 232         }
 233
 234 out:
 235         return (void __iomem *) (v + ((unsigned long)addr & ~PAGE_MASK));
 236 }
 237
 238 void iounmap(volatile void __iomem *addr)
 239 {
 240         /*
 241          * If mapped by BATs then there is nothing to do.
 242          * Calling vfree() generates a benign warning.
 243          */
 244         if (v_mapped_by_bats((unsigned long)addr)) return;
 245
 246         if (addr > high_memory && (unsigned long) addr < ioremap_bot)
 247                 vunmap((void *) (PAGE_MASK & (unsigned long)addr));
 248 }
 249
 250 void __iomem *ioport_map(unsigned long port, unsigned int len)
 251 {
 252         return (void __iomem *) (port + _IO_BASE);
 253 }
 254
 255 void ioport_unmap(void __iomem *addr)
 256 {
 257         /* Nothing to do */
 258 }
 259 EXPORT_SYMBOL(ioport_map);
 260 EXPORT_SYMBOL(ioport_unmap);
 261
 262 int
 263 map_page(unsigned long va, phys_addr_t pa, int flags)
 264 {
 265         pmd_t *pd;
 266         pte_t *pg;
 267         int err = -ENOMEM;
 268
 269         /* Use upper 10 bits of VA to index the first level map */
 270         pd = pmd_offset(pgd_offset_k(va), va);
 271         /* Use middle 10 bits of VA to index the second-level map */
 272         pg = pte_alloc_kernel(pd, va);
 273         if (pg != 0) {
 274                 err = 0;
 275                 set_pte_at(&init_mm, va, pg, pfn_pte(pa >> PAGE_SHIFT, __pgprot(flags)));
 276                 if (mem_init_done)
 277                         flush_HPTE(0, va, pmd_val(*pd));
 278         }
 279         return err;
 280 }
 281
 282 /*
 283  * Map in all of physical memory starting at KERNELBASE.
 284  */
 285 void __init mapin_ram(void)
 286 {
 287         unsigned long v, p, s, f;
 288
 289         s = mmu_mapin_ram();
 290         v = KERNELBASE + s;
 291         p = PPC_MEMSTART + s;
 292         for (; s < total_lowmem; s += PAGE_SIZE) {
 293                 if ((char *) v >= _stext && (char *) v < etext)
 294                         f = _PAGE_RAM_TEXT;
 295                 else
 296                         f = _PAGE_RAM;
 297                 map_page(v, p, f);
 298                 v += PAGE_SIZE;
 299                 p += PAGE_SIZE;
 300         }
 301 }
 302
 303 /* is x a power of 4? */
 304 #define is_power_of_4(x)        is_power_of_2(x) && (ffs(x) & 1)
 305
 306 /*
 307  * Set up a mapping for a block of I/O.
 308  * virt, phys, size must all be page-aligned.
 309  * This should only be called before ioremap is called.
 310  */
 311 void __init io_block_mapping(unsigned long virt, phys_addr_t phys,
 312                              unsigned int size, int flags)
 313 {
 314         int i;
 315
 316         if (virt > KERNELBASE && virt < ioremap_bot)
 317                 ioremap_bot = ioremap_base = virt;
 318
 319 #ifdef HAVE_BATS
 320         /*
 321          * Use a BAT for this if possible...
 322          */
 323         if (io_bat_index < 2 && is_power_of_2(size)
 324             && (virt & (size - 1)) == 0 && (phys & (size - 1)) == 0) {
 325                 setbat(io_bat_index, virt, phys, size, flags);
 326                 ++io_bat_index;
 327                 return;
 328         }
 329 #endif /* HAVE_BATS */
 330
 331         /* No BATs available, put it in the page tables. */
 332         for (i = 0; i < size; i += PAGE_SIZE)
 333                 map_page(virt + i, phys + i, flags);
 334 }
 335
 336 /* Scan the real Linux page tables and return a PTE pointer for
 337  * a virtual address in a context.
 338  * Returns true (1) if PTE was found, zero otherwise.  The pointer to
 339  * the PTE pointer is unmodified if PTE is not found.
 340  */
 341 int
 342 get_pteptr(struct mm_struct *mm, unsigned long addr, pte_t **ptep, pmd_t **pmdp)
 343 {
 344         pgd_t   *pgd;
 345         pmd_t   *pmd;
 346         pte_t   *pte;
 347         int     retval = 0;
 348
 349         pgd = pgd_offset(mm, addr & PAGE_MASK);
 350         if (pgd) {
 351                 pmd = pmd_offset(pgd, addr & PAGE_MASK);
 352                 if (pmd_present(*pmd)) {
 353                         pte = pte_offset_map(pmd, addr & PAGE_MASK);
 354                         if (pte) {
 355                                 retval = 1;
 356                                 *ptep = pte;
 357                                 if (pmdp)
 358                                         *pmdp = pmd;
 359                                 /* XXX caller needs to do pte_unmap, yuck */
 360                         }
 361                 }
 362         }
 363         return(retval);
 364 }
 365
 366 /* Find physical address for this virtual address.  Normally used by
 367  * I/O functions, but anyone can call it.
 368  */
 369 unsigned long iopa(unsigned long addr)
 370 {
 371         unsigned long pa;
 372
 373         /* I don't know why this won't work on PMacs or CHRP.  It
 374          * appears there is some bug, or there is some implicit
 375          * mapping done not properly represented by BATs or in page
 376          * tables.......I am actively working on resolving this, but
 377          * can't hold up other stuff.  -- Dan
 378          */
 379         pte_t *pte;
 380         struct mm_struct *mm;
 381
 382         /* Check the BATs */
 383         pa = v_mapped_by_bats(addr);
 384         if (pa)
 385                 return pa;
 386
 387         /* Allow mapping of user addresses (within the thread)
 388          * for DMA if necessary.
 389          */
 390         if (addr < TASK_SIZE)
 391                 mm = current->mm;
 392         else
 393                 mm = &init_mm;
 394
 395         pa = 0;
 396         if (get_pteptr(mm, addr, &pte, NULL)) {
 397                 pa = (pte_val(*pte) & PAGE_MASK) | (addr & ~PAGE_MASK);
 398                 pte_unmap(pte);
 399         }
 400
 401         return(pa);
 402 }
 403