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
  30 #include <asm/pgtable.h>
  31 #include <asm/pgalloc.h>
  32 #include <asm/fixmap.h>
  33 #include <asm/io.h>
  34
  35 #include "mmu_decl.h"
  36
  37 unsigned long ioremap_base;
  38 unsigned long ioremap_bot;
  39 EXPORT_SYMBOL(ioremap_bot);     /* aka VMALLOC_END */
  40
  41 #if defined(CONFIG_6xx) || defined(CONFIG_POWER3)
  42 #define HAVE_BATS       1
  43 #endif
  44
  45 #if defined(CONFIG_FSL_BOOKE)
  46 #define HAVE_TLBCAM     1
  47 #endif
  48
  49 extern char etext[], _stext[];
  50
  51 #ifdef CONFIG_SMP
  52 extern void hash_page_sync(void);
  53 #endif
  54
  55 #ifdef HAVE_BATS
  56 extern phys_addr_t v_mapped_by_bats(unsigned long va);
  57 extern unsigned long p_mapped_by_bats(phys_addr_t pa);
  58 void setbat(int index, unsigned long virt, phys_addr_t phys,
  59             unsigned int size, int flags);
  60
  61 #else /* !HAVE_BATS */
  62 #define v_mapped_by_bats(x)     (0UL)
  63 #define p_mapped_by_bats(x)     (0UL)
  64 #endif /* HAVE_BATS */
  65
  66 #ifdef HAVE_TLBCAM
  67 extern unsigned int tlbcam_index;
  68 extern unsigned long v_mapped_by_tlbcam(unsigned long va);
  69 extern unsigned long p_mapped_by_tlbcam(unsigned long pa);
  70 #else /* !HAVE_TLBCAM */
  71 #define v_mapped_by_tlbcam(x)   (0UL)
  72 #define p_mapped_by_tlbcam(x)   (0UL)
  73 #endif /* HAVE_TLBCAM */
  74
  75 #ifdef CONFIG_PTE_64BIT
  76 /* 44x uses an 8kB pgdir because it has 8-byte Linux PTEs. */
  77 #define PGDIR_ORDER     1
  78 #else
  79 #define PGDIR_ORDER     0
  80 #endif
  81
  82 pgd_t *pgd_alloc(struct mm_struct *mm)
  83 {
  84         pgd_t *ret;
  85
  86         ret = (pgd_t *)__get_free_pages(GFP_KERNEL|__GFP_ZERO, PGDIR_ORDER);
  87         return ret;
  88 }
  89
  90 void pgd_free(struct mm_struct *mm, pgd_t *pgd)
  91 {
  92         free_pages((unsigned long)pgd, PGDIR_ORDER);
  93 }
  94
  95 __init_refok pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
  96 {
  97         pte_t *pte;
  98         extern int mem_init_done;
  99         extern void *early_get_page(void);
 100
 101         if (mem_init_done) {
 102                 pte = (pte_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO);
 103         } else {
 104                 pte = (pte_t *)early_get_page();
 105                 if (pte)
 106                         clear_page(pte);
 107         }
 108         return pte;
 109 }
 110
 111 pgtable_t pte_alloc_one(struct mm_struct *mm, unsigned long address)
 112 {
 113         struct page *ptepage;
 114
 115 #ifdef CONFIG_HIGHPTE
 116         gfp_t flags = GFP_KERNEL | __GFP_HIGHMEM | __GFP_REPEAT | __GFP_ZERO;
 117 #else
 118         gfp_t flags = GFP_KERNEL | __GFP_REPEAT | __GFP_ZERO;
 119 #endif
 120
 121         ptepage = alloc_pages(flags, 0);
 122         if (!ptepage)
 123                 return NULL;
 124         pgtable_page_ctor(ptepage);
 125         return ptepage;
 126 }
 127
 128 void pte_free_kernel(struct mm_struct *mm, pte_t *pte)
 129 {
 130 #ifdef CONFIG_SMP
 131         hash_page_sync();
 132 #endif
 133         free_page((unsigned long)pte);
 134 }
 135
 136 void pte_free(struct mm_struct *mm, pgtable_t ptepage)
 137 {
 138 #ifdef CONFIG_SMP
 139         hash_page_sync();
 140 #endif
 141         pgtable_page_dtor(ptepage);
 142         __free_page(ptepage);
 143 }
 144
 145 void __iomem *
 146 ioremap(phys_addr_t addr, unsigned long size)
 147 {
 148         return __ioremap(addr, size, _PAGE_NO_CACHE | _PAGE_GUARDED);
 149 }
 150 EXPORT_SYMBOL(ioremap);
 151
 152 void __iomem *
 153 ioremap_flags(phys_addr_t addr, unsigned long size, unsigned long flags)
 154 {
 155         /* writeable implies dirty for kernel addresses */
 156         if (flags & _PAGE_RW)
 157                 flags |= _PAGE_DIRTY | _PAGE_HWWRITE;
 158
 159         /* we don't want to let _PAGE_USER and _PAGE_EXEC leak out */
 160         flags &= ~(_PAGE_USER | _PAGE_EXEC | _PAGE_HWEXEC);
 161
 162         return __ioremap(addr, size, flags);
 163 }
 164 EXPORT_SYMBOL(ioremap_flags);
 165
 166 void __iomem *
 167 __ioremap(phys_addr_t addr, unsigned long size, unsigned long flags)
 168 {
 169         unsigned long v, i;
 170         phys_addr_t p;
 171         int err;
 172
 173         /* Make sure we have the base flags */
 174         if ((flags & _PAGE_PRESENT) == 0)
 175                 flags |= _PAGE_KERNEL;
 176
 177         /* Non-cacheable page cannot be coherent */
 178         if (flags & _PAGE_NO_CACHE)
 179                 flags &= ~_PAGE_COHERENT;
 180
 181         /*
 182          * Choose an address to map it to.
 183          * Once the vmalloc system is running, we use it.
 184          * Before then, we use space going down from ioremap_base
 185          * (ioremap_bot records where we're up to).
 186          */
 187         p = addr & PAGE_MASK;
 188         size = PAGE_ALIGN(addr + size) - p;
 189
 190         /*
 191          * If the address lies within the first 16 MB, assume it's in ISA
 192          * memory space
 193          */
 194         if (p < 16*1024*1024)
 195                 p += _ISA_MEM_BASE;
 196
 197         /*
 198          * Don't allow anybody to remap normal RAM that we're using.
 199          * mem_init() sets high_memory so only do the check after that.
 200          */
 201         if (mem_init_done && (p < virt_to_phys(high_memory))) {
 202                 printk("__ioremap(): phys addr 0x%llx is RAM lr %p\n",
 203                        (unsigned long long)p, __builtin_return_address(0));
 204                 return NULL;
 205         }
 206
 207         if (size == 0)
 208                 return NULL;
 209
 210         /*
 211          * Is it already mapped?  Perhaps overlapped by a previous
 212          * BAT mapping.  If the whole area is mapped then we're done,
 213          * otherwise remap it since we want to keep the virt addrs for
 214          * each request contiguous.
 215          *
 216          * We make the assumption here that if the bottom and top
 217          * of the range we want are mapped then it's mapped to the
 218          * same virt address (and this is contiguous).
 219          *  -- Cort
 220          */
 221         if ((v = p_mapped_by_bats(p)) /*&& p_mapped_by_bats(p+size-1)*/ )
 222                 goto out;
 223
 224         if ((v = p_mapped_by_tlbcam(p)))
 225                 goto out;
 226
 227         if (mem_init_done) {
 228                 struct vm_struct *area;
 229                 area = get_vm_area(size, VM_IOREMAP);
 230                 if (area == 0)
 231                         return NULL;
 232                 v = (unsigned long) area->addr;
 233         } else {
 234                 v = (ioremap_bot -= size);
 235         }
 236
 237         /*
 238          * Should check if it is a candidate for a BAT mapping
 239          */
 240
 241         err = 0;
 242         for (i = 0; i < size && err == 0; i += PAGE_SIZE)
 243                 err = map_page(v+i, p+i, flags);
 244         if (err) {
 245                 if (mem_init_done)
 246                         vunmap((void *)v);
 247                 return NULL;
 248         }
 249
 250 out:
 251         return (void __iomem *) (v + ((unsigned long)addr & ~PAGE_MASK));
 252 }
 253 EXPORT_SYMBOL(__ioremap);
 254
 255 void iounmap(volatile void __iomem *addr)
 256 {
 257         /*
 258          * If mapped by BATs then there is nothing to do.
 259          * Calling vfree() generates a benign warning.
 260          */
 261         if (v_mapped_by_bats((unsigned long)addr)) return;
 262
 263         if (addr > high_memory && (unsigned long) addr < ioremap_bot)
 264                 vunmap((void *) (PAGE_MASK & (unsigned long)addr));
 265 }
 266 EXPORT_SYMBOL(iounmap);
 267
 268 int map_page(unsigned long va, phys_addr_t pa, int flags)
 269 {
 270         pmd_t *pd;
 271         pte_t *pg;
 272         int err = -ENOMEM;
 273
 274         /* Use upper 10 bits of VA to index the first level map */
 275         pd = pmd_offset(pud_offset(pgd_offset_k(va), va), va);
 276         /* Use middle 10 bits of VA to index the second-level map */
 277         pg = pte_alloc_kernel(pd, va);
 278         if (pg != 0) {
 279                 err = 0;
 280                 /* The PTE should never be already set nor present in the
 281                  * hash table
 282                  */
 283                 BUG_ON(pte_val(*pg) & (_PAGE_PRESENT | _PAGE_HASHPTE));
 284                 set_pte_at(&init_mm, va, pg, pfn_pte(pa >> PAGE_SHIFT,
 285                                                      __pgprot(flags)));
 286         }
 287         return err;
 288 }
 289
 290 /*
 291  * Map in all of physical memory starting at KERNELBASE.
 292  */
 293 void __init mapin_ram(void)
 294 {
 295         unsigned long v, s, f;
 296         phys_addr_t p;
 297         int ktext;
 298
 299         s = mmu_mapin_ram();
 300         v = KERNELBASE + s;
 301         p = memstart_addr + s;
 302         for (; s < total_lowmem; s += PAGE_SIZE) {
 303                 ktext = ((char *) v >= _stext && (char *) v < etext);
 304                 f = ktext ?_PAGE_RAM_TEXT : _PAGE_RAM;
 305                 map_page(v, p, f);
 306 #ifdef CONFIG_PPC_STD_MMU_32
 307                 if (ktext)
 308                         hash_preload(&init_mm, v, 0, 0x300);
 309 #endif
 310                 v += PAGE_SIZE;
 311                 p += PAGE_SIZE;
 312         }
 313 }
 314
 315 /* Scan the real Linux page tables and return a PTE pointer for
 316  * a virtual address in a context.
 317  * Returns true (1) if PTE was found, zero otherwise.  The pointer to
 318  * the PTE pointer is unmodified if PTE is not found.
 319  */
 320 int
 321 get_pteptr(struct mm_struct *mm, unsigned long addr, pte_t **ptep, pmd_t **pmdp)
 322 {
 323         pgd_t   *pgd;
 324         pud_t   *pud;
 325         pmd_t   *pmd;
 326         pte_t   *pte;
 327         int     retval = 0;
 328
 329         pgd = pgd_offset(mm, addr & PAGE_MASK);
 330         if (pgd) {
 331                 pud = pud_offset(pgd, addr & PAGE_MASK);
 332                 if (pud && pud_present(*pud)) {
 333                         pmd = pmd_offset(pud, addr & PAGE_MASK);
 334                         if (pmd_present(*pmd)) {
 335                                 pte = pte_offset_map(pmd, addr & PAGE_MASK);
 336                                 if (pte) {
 337                                         retval = 1;
 338                                         *ptep = pte;
 339                                         if (pmdp)
 340                                                 *pmdp = pmd;
 341                                         /* XXX caller needs to do pte_unmap, yuck */
 342                                 }
 343                         }
 344                 }
 345         }
 346         return(retval);
 347 }
 348
 349 #ifdef CONFIG_DEBUG_PAGEALLOC
 350
 351 static int __change_page_attr(struct page *page, pgprot_t prot)
 352 {
 353         pte_t *kpte;
 354         pmd_t *kpmd;
 355         unsigned long address;
 356
 357         BUG_ON(PageHighMem(page));
 358         address = (unsigned long)page_address(page);
 359
 360         if (v_mapped_by_bats(address) || v_mapped_by_tlbcam(address))
 361                 return 0;
 362         if (!get_pteptr(&init_mm, address, &kpte, &kpmd))
 363                 return -EINVAL;
 364         set_pte_at(&init_mm, address, kpte, mk_pte(page, prot));
 365         wmb();
 366         flush_HPTE(0, address, pmd_val(*kpmd));
 367         pte_unmap(kpte);
 368
 369         return 0;
 370 }
 371
 372 /*
 373  * Change the page attributes of an page in the linear mapping.
 374  *
 375  * THIS CONFLICTS WITH BAT MAPPINGS, DEBUG USE ONLY
 376  */
 377 static int change_page_attr(struct page *page, int numpages, pgprot_t prot)
 378 {
 379         int i, err = 0;
 380         unsigned long flags;
 381
 382         local_irq_save(flags);
 383         for (i = 0; i < numpages; i++, page++) {
 384                 err = __change_page_attr(page, prot);
 385                 if (err)
 386                         break;
 387         }
 388         local_irq_restore(flags);
 389         return err;
 390 }
 391
 392
 393 void kernel_map_pages(struct page *page, int numpages, int enable)
 394 {
 395         if (PageHighMem(page))
 396                 return;
 397
 398         change_page_attr(page, numpages, enable ? PAGE_KERNEL : __pgprot(0));
 399 }
 400 #endif /* CONFIG_DEBUG_PAGEALLOC */
 401
 402 static int fixmaps;
 403 unsigned long FIXADDR_TOP = 0xfffff000;
 404 EXPORT_SYMBOL(FIXADDR_TOP);
 405
 406 void __set_fixmap (enum fixed_addresses idx, phys_addr_t phys, pgprot_t flags)
 407 {
 408         unsigned long address = __fix_to_virt(idx);
 409
 410         if (idx >= __end_of_fixed_addresses) {
 411                 BUG();
 412                 return;
 413         }
 414
 415         map_page(address, phys, pgprot_val(flags));
 416         fixmaps++;
 417 }
 418
 419 void __this_fixmap_does_not_exist(void)
 420 {
 421         WARN_ON(1);
 422 }