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 #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 unsigned long v_mapped_by_bats(unsigned long va);
  57 extern unsigned long p_mapped_by_bats(unsigned long pa);
  58 void setbat(int index, unsigned long virt, unsigned long 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(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 struct page *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;
 117 #else
 118         gfp_t flags = GFP_KERNEL | __GFP_REPEAT;
 119 #endif
 120
 121         ptepage = alloc_pages(flags, 0);
 122         if (ptepage)
 123                 clear_highpage(ptepage);
 124         return ptepage;
 125 }
 126
 127 void pte_free_kernel(pte_t *pte)
 128 {
 129 #ifdef CONFIG_SMP
 130         hash_page_sync();
 131 #endif
 132         free_page((unsigned long)pte);
 133 }
 134
 135 void pte_free(struct page *ptepage)
 136 {
 137 #ifdef CONFIG_SMP
 138         hash_page_sync();
 139 #endif
 140         __free_page(ptepage);
 141 }
 142
 143 #ifndef CONFIG_PHYS_64BIT
 144 void __iomem *
 145 ioremap(phys_addr_t addr, unsigned long size)
 146 {
 147         return __ioremap(addr, size, _PAGE_NO_CACHE);
 148 }
 149 #else /* CONFIG_PHYS_64BIT */
 150 void __iomem *
 151 ioremap64(unsigned long long addr, unsigned long size)
 152 {
 153         return __ioremap(addr, size, _PAGE_NO_CACHE);
 154 }
 155
 156 void __iomem *
 157 ioremap(phys_addr_t addr, unsigned long size)
 158 {
 159         phys_addr_t addr64 = fixup_bigphys_addr(addr, size);
 160
 161         return ioremap64(addr64, size);
 162 }
 163 #endif /* CONFIG_PHYS_64BIT */
 164
 165 void __iomem *
 166 __ioremap(phys_addr_t addr, unsigned long size, unsigned long flags)
 167 {
 168         unsigned long v, i;
 169         phys_addr_t p;
 170         int err;
 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_base
 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         /*
 189          * Don't allow anybody to remap normal RAM that we're using.
 190          * mem_init() sets high_memory so only do the check after that.
 191          */
 192         if ( mem_init_done && (p < virt_to_phys(high_memory)) )
 193         {
 194                 printk("__ioremap(): phys addr "PHYS_FMT" is RAM lr %p\n", p,
 195                        __builtin_return_address(0));
 196                 return NULL;
 197         }
 198
 199         if (size == 0)
 200                 return NULL;
 201
 202         /*
 203          * Is it already mapped?  Perhaps overlapped by a previous
 204          * BAT mapping.  If the whole area is mapped then we're done,
 205          * otherwise remap it since we want to keep the virt addrs for
 206          * each request contiguous.
 207          *
 208          * We make the assumption here that if the bottom and top
 209          * of the range we want are mapped then it's mapped to the
 210          * same virt address (and this is contiguous).
 211          *  -- Cort
 212          */
 213         if ((v = p_mapped_by_bats(p)) /*&& p_mapped_by_bats(p+size-1)*/ )
 214                 goto out;
 215
 216         if ((v = p_mapped_by_tlbcam(p)))
 217                 goto out;
 218
 219         if (mem_init_done) {
 220                 struct vm_struct *area;
 221                 area = get_vm_area(size, VM_IOREMAP);
 222                 if (area == 0)
 223                         return NULL;
 224                 v = (unsigned long) area->addr;
 225         } else {
 226                 v = (ioremap_bot -= size);
 227         }
 228
 229         if ((flags & _PAGE_PRESENT) == 0)
 230                 flags |= _PAGE_KERNEL;
 231         if (flags & _PAGE_NO_CACHE)
 232                 flags |= _PAGE_GUARDED;
 233
 234         /*
 235          * Should check if it is a candidate for a BAT mapping
 236          */
 237
 238         err = 0;
 239         for (i = 0; i < size && err == 0; i += PAGE_SIZE)
 240                 err = map_page(v+i, p+i, flags);
 241         if (err) {
 242                 if (mem_init_done)
 243                         vunmap((void *)v);
 244                 return NULL;
 245         }
 246
 247 out:
 248         return (void __iomem *) (v + ((unsigned long)addr & ~PAGE_MASK));
 249 }
 250
 251 void iounmap(volatile void __iomem *addr)
 252 {
 253         /*
 254          * If mapped by BATs then there is nothing to do.
 255          * Calling vfree() generates a benign warning.
 256          */
 257         if (v_mapped_by_bats((unsigned long)addr)) return;
 258
 259         if (addr > high_memory && (unsigned long) addr < ioremap_bot)
 260                 vunmap((void *) (PAGE_MASK & (unsigned long)addr));
 261 }
 262
 263 void __iomem *ioport_map(unsigned long port, unsigned int len)
 264 {
 265         return (void __iomem *) (port + _IO_BASE);
 266 }
 267
 268 void ioport_unmap(void __iomem *addr)
 269 {
 270         /* Nothing to do */
 271 }
 272 EXPORT_SYMBOL(ioport_map);
 273 EXPORT_SYMBOL(ioport_unmap);
 274
 275 int
 276 map_page(unsigned long va, phys_addr_t pa, int flags)
 277 {
 278         pmd_t *pd;
 279         pte_t *pg;
 280         int err = -ENOMEM;
 281
 282         /* Use upper 10 bits of VA to index the first level map */
 283         pd = pmd_offset(pgd_offset_k(va), va);
 284         /* Use middle 10 bits of VA to index the second-level map */
 285         pg = pte_alloc_kernel(pd, va);
 286         if (pg != 0) {
 287                 err = 0;
 288                 set_pte_at(&init_mm, va, pg, pfn_pte(pa >> PAGE_SHIFT, __pgprot(flags)));
 289                 if (mem_init_done)
 290                         flush_HPTE(0, va, pmd_val(*pd));
 291         }
 292         return err;
 293 }
 294
 295 /*
 296  * Map in all of physical memory starting at KERNELBASE.
 297  */
 298 void __init mapin_ram(void)
 299 {
 300         unsigned long v, p, s, f;
 301
 302         s = mmu_mapin_ram();
 303         v = KERNELBASE + s;
 304         p = PPC_MEMSTART + s;
 305         for (; s < total_lowmem; s += PAGE_SIZE) {
 306                 if ((char *) v >= _stext && (char *) v < etext)
 307                         f = _PAGE_RAM_TEXT;
 308                 else
 309                         f = _PAGE_RAM;
 310                 map_page(v, p, f);
 311                 v += PAGE_SIZE;
 312                 p += PAGE_SIZE;
 313         }
 314 }
 315
 316 /* is x a power of 4? */
 317 #define is_power_of_4(x)        is_power_of_2(x) && (ffs(x) & 1)
 318
 319 /*
 320  * Set up a mapping for a block of I/O.
 321  * virt, phys, size must all be page-aligned.
 322  * This should only be called before ioremap is called.
 323  */
 324 void __init io_block_mapping(unsigned long virt, phys_addr_t phys,
 325                              unsigned int size, int flags)
 326 {
 327         int i;
 328
 329         if (virt > KERNELBASE && virt < ioremap_bot)
 330                 ioremap_bot = ioremap_base = virt;
 331
 332 #ifdef HAVE_BATS
 333         /*
 334          * Use a BAT for this if possible...
 335          */
 336         if (io_bat_index < 2 && is_power_of_2(size)
 337             && (virt & (size - 1)) == 0 && (phys & (size - 1)) == 0) {
 338                 setbat(io_bat_index, virt, phys, size, flags);
 339                 ++io_bat_index;
 340                 return;
 341         }
 342 #endif /* HAVE_BATS */
 343
 344 #ifdef HAVE_TLBCAM
 345         /*
 346          * Use a CAM for this if possible...
 347          */
 348         if (tlbcam_index < num_tlbcam_entries && is_power_of_4(size)
 349             && (virt & (size - 1)) == 0 && (phys & (size - 1)) == 0) {
 350                 settlbcam(tlbcam_index, virt, phys, size, flags, 0);
 351                 ++tlbcam_index;
 352                 return;
 353         }
 354 #endif /* HAVE_TLBCAM */
 355
 356         /* No BATs available, put it in the page tables. */
 357         for (i = 0; i < size; i += PAGE_SIZE)
 358                 map_page(virt + i, phys + i, flags);
 359 }
 360
 361 /* Scan the real Linux page tables and return a PTE pointer for
 362  * a virtual address in a context.
 363  * Returns true (1) if PTE was found, zero otherwise.  The pointer to
 364  * the PTE pointer is unmodified if PTE is not found.
 365  */
 366 int
 367 get_pteptr(struct mm_struct *mm, unsigned long addr, pte_t **ptep, pmd_t **pmdp)
 368 {
 369         pgd_t   *pgd;
 370         pmd_t   *pmd;
 371         pte_t   *pte;
 372         int     retval = 0;
 373
 374         pgd = pgd_offset(mm, addr & PAGE_MASK);
 375         if (pgd) {
 376                 pmd = pmd_offset(pgd, addr & PAGE_MASK);
 377                 if (pmd_present(*pmd)) {
 378                         pte = pte_offset_map(pmd, addr & PAGE_MASK);
 379                         if (pte) {
 380                                 retval = 1;
 381                                 *ptep = pte;
 382                                 if (pmdp)
 383                                         *pmdp = pmd;
 384                                 /* XXX caller needs to do pte_unmap, yuck */
 385                         }
 386                 }
 387         }
 388         return(retval);
 389 }
 390
 391 /* Find physical address for this virtual address.  Normally used by
 392  * I/O functions, but anyone can call it.
 393  */
 394 unsigned long iopa(unsigned long addr)
 395 {
 396         unsigned long pa;
 397
 398         /* I don't know why this won't work on PMacs or CHRP.  It
 399          * appears there is some bug, or there is some implicit
 400          * mapping done not properly represented by BATs or in page
 401          * tables.......I am actively working on resolving this, but
 402          * can't hold up other stuff.  -- Dan
 403          */
 404         pte_t *pte;
 405         struct mm_struct *mm;
 406
 407         /* Check the BATs */
 408         pa = v_mapped_by_bats(addr);
 409         if (pa)
 410                 return pa;
 411
 412         /* Allow mapping of user addresses (within the thread)
 413          * for DMA if necessary.
 414          */
 415         if (addr < TASK_SIZE)
 416                 mm = current->mm;
 417         else
 418                 mm = &init_mm;
 419
 420         pa = 0;
 421         if (get_pteptr(mm, addr, &pte, NULL)) {
 422                 pa = (pte_val(*pte) & PAGE_MASK) | (addr & ~PAGE_MASK);
 423                 pte_unmap(pte);
 424         }
 425
 426         return(pa);
 427 }
 428