arch/powerpc/mm/pgtable_64.c

   1 /*
   2  *  This file contains ioremap and related functions for 64-bit machines.
   3  *
   4  *  Derived from arch/ppc64/mm/init.c
   5  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
   6  *
   7  *  Modifications by Paul Mackerras (PowerMac) (paulus@samba.org)
   8  *  and Cort Dougan (PReP) (cort@cs.nmt.edu)
   9  *    Copyright (C) 1996 Paul Mackerras
  10  *
  11  *  Derived from "arch/i386/mm/init.c"
  12  *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
  13  *
  14  *  Dave Engebretsen <engebret@us.ibm.com>
  15  *      Rework for PPC64 port.
  16  *
  17  *  This program is free software; you can redistribute it and/or
  18  *  modify it under the terms of the GNU General Public License
  19  *  as published by the Free Software Foundation; either version
  20  *  2 of the License, or (at your option) any later version.
  21  *
  22  */
  23
  24 #include <linux/signal.h>
  25 #include <linux/sched.h>
  26 #include <linux/kernel.h>
  27 #include <linux/errno.h>
  28 #include <linux/string.h>
  29 #include <linux/export.h>
  30 #include <linux/types.h>
  31 #include <linux/mman.h>
  32 #include <linux/mm.h>
  33 #include <linux/swap.h>
  34 #include <linux/stddef.h>
  35 #include <linux/vmalloc.h>
  36 #include <linux/init.h>
  37 #include <linux/bootmem.h>
  38 #include <linux/memblock.h>
  39 #include <linux/slab.h>
  40
  41 #include <asm/pgalloc.h>
  42 #include <asm/page.h>
  43 #include <asm/prom.h>
  44 #include <asm/io.h>
  45 #include <asm/mmu_context.h>
  46 #include <asm/pgtable.h>
  47 #include <asm/mmu.h>
  48 #include <asm/smp.h>
  49 #include <asm/machdep.h>
  50 #include <asm/tlb.h>
  51 #include <asm/processor.h>
  52 #include <asm/cputable.h>
  53 #include <asm/sections.h>
  54 #include <asm/system.h>
  55 #include <asm/abs_addr.h>
  56 #include <asm/firmware.h>
  57
  58 #include "mmu_decl.h"
  59
  60 unsigned long ioremap_bot = IOREMAP_BASE;
  61
  62
  63 #ifdef CONFIG_PPC_MMU_NOHASH
  64 static void *early_alloc_pgtable(unsigned long size)
  65 {
  66         void *pt;
  67
  68         if (init_bootmem_done)
  69                 pt = __alloc_bootmem(size, size, __pa(MAX_DMA_ADDRESS));
  70         else
  71                 pt = __va(memblock_alloc_base(size, size,
  72                                          __pa(MAX_DMA_ADDRESS)));
  73         memset(pt, 0, size);
  74
  75         return pt;
  76 }
  77 #endif /* CONFIG_PPC_MMU_NOHASH */
  78
  79 /*
  80  * map_kernel_page currently only called by __ioremap
  81  * map_kernel_page adds an entry to the ioremap page table
  82  * and adds an entry to the HPT, possibly bolting it
  83  */
  84 int map_kernel_page(unsigned long ea, unsigned long pa, int flags)
  85 {
  86         pgd_t *pgdp;
  87         pud_t *pudp;
  88         pmd_t *pmdp;
  89         pte_t *ptep;
  90
  91         if (slab_is_available()) {
  92                 pgdp = pgd_offset_k(ea);
  93                 pudp = pud_alloc(&init_mm, pgdp, ea);
  94                 if (!pudp)
  95                         return -ENOMEM;
  96                 pmdp = pmd_alloc(&init_mm, pudp, ea);
  97                 if (!pmdp)
  98                         return -ENOMEM;
  99                 ptep = pte_alloc_kernel(pmdp, ea);
 100                 if (!ptep)
 101                         return -ENOMEM;
 102                 set_pte_at(&init_mm, ea, ptep, pfn_pte(pa >> PAGE_SHIFT,
 103                                                           __pgprot(flags)));
 104         } else {
 105 #ifdef CONFIG_PPC_MMU_NOHASH
 106                 /* Warning ! This will blow up if bootmem is not initialized
 107                  * which our ppc64 code is keen to do that, we'll need to
 108                  * fix it and/or be more careful
 109                  */
 110                 pgdp = pgd_offset_k(ea);
 111 #ifdef PUD_TABLE_SIZE
 112                 if (pgd_none(*pgdp)) {
 113                         pudp = early_alloc_pgtable(PUD_TABLE_SIZE);
 114                         BUG_ON(pudp == NULL);
 115                         pgd_populate(&init_mm, pgdp, pudp);
 116                 }
 117 #endif /* PUD_TABLE_SIZE */
 118                 pudp = pud_offset(pgdp, ea);
 119                 if (pud_none(*pudp)) {
 120                         pmdp = early_alloc_pgtable(PMD_TABLE_SIZE);
 121                         BUG_ON(pmdp == NULL);
 122                         pud_populate(&init_mm, pudp, pmdp);
 123                 }
 124                 pmdp = pmd_offset(pudp, ea);
 125                 if (!pmd_present(*pmdp)) {
 126                         ptep = early_alloc_pgtable(PAGE_SIZE);
 127                         BUG_ON(ptep == NULL);
 128                         pmd_populate_kernel(&init_mm, pmdp, ptep);
 129                 }
 130                 ptep = pte_offset_kernel(pmdp, ea);
 131                 set_pte_at(&init_mm, ea, ptep, pfn_pte(pa >> PAGE_SHIFT,
 132                                                           __pgprot(flags)));
 133 #else /* CONFIG_PPC_MMU_NOHASH */
 134                 /*
 135                  * If the mm subsystem is not fully up, we cannot create a
 136                  * linux page table entry for this mapping.  Simply bolt an
 137                  * entry in the hardware page table.
 138                  *
 139                  */
 140                 if (htab_bolt_mapping(ea, ea + PAGE_SIZE, pa, flags,
 141                                       mmu_io_psize, mmu_kernel_ssize)) {
 142                         printk(KERN_ERR "Failed to do bolted mapping IO "
 143                                "memory at %016lx !\n", pa);
 144                         return -ENOMEM;
 145                 }
 146 #endif /* !CONFIG_PPC_MMU_NOHASH */
 147         }
 148         return 0;
 149 }
 150
 151
 152 /**
 153  * __ioremap_at - Low level function to establish the page tables
 154  *                for an IO mapping
 155  */
 156 void __iomem * __ioremap_at(phys_addr_t pa, void *ea, unsigned long size,
 157                             unsigned long flags)
 158 {
 159         unsigned long i;
 160
 161         /* Make sure we have the base flags */
 162         if ((flags & _PAGE_PRESENT) == 0)
 163                 flags |= pgprot_val(PAGE_KERNEL);
 164
 165         /* Non-cacheable page cannot be coherent */
 166         if (flags & _PAGE_NO_CACHE)
 167                 flags &= ~_PAGE_COHERENT;
 168
 169         /* We don't support the 4K PFN hack with ioremap */
 170         if (flags & _PAGE_4K_PFN)
 171                 return NULL;
 172
 173         WARN_ON(pa & ~PAGE_MASK);
 174         WARN_ON(((unsigned long)ea) & ~PAGE_MASK);
 175         WARN_ON(size & ~PAGE_MASK);
 176
 177         for (i = 0; i < size; i += PAGE_SIZE)
 178                 if (map_kernel_page((unsigned long)ea+i, pa+i, flags))
 179                         return NULL;
 180
 181         return (void __iomem *)ea;
 182 }
 183
 184 /**
 185  * __iounmap_from - Low level function to tear down the page tables
 186  *                  for an IO mapping. This is used for mappings that
 187  *                  are manipulated manually, like partial unmapping of
 188  *                  PCI IOs or ISA space.
 189  */
 190 void __iounmap_at(void *ea, unsigned long size)
 191 {
 192         WARN_ON(((unsigned long)ea) & ~PAGE_MASK);
 193         WARN_ON(size & ~PAGE_MASK);
 194
 195         unmap_kernel_range((unsigned long)ea, size);
 196 }
 197
 198 void __iomem * __ioremap_caller(phys_addr_t addr, unsigned long size,
 199                                 unsigned long flags, void *caller)
 200 {
 201         phys_addr_t paligned;
 202         void __iomem *ret;
 203
 204         /*
 205          * Choose an address to map it to.
 206          * Once the imalloc system is running, we use it.
 207          * Before that, we map using addresses going
 208          * up from ioremap_bot.  imalloc will use
 209          * the addresses from ioremap_bot through
 210          * IMALLOC_END
 211          *
 212          */
 213         paligned = addr & PAGE_MASK;
 214         size = PAGE_ALIGN(addr + size) - paligned;
 215
 216         if ((size == 0) || (paligned == 0))
 217                 return NULL;
 218
 219         if (mem_init_done) {
 220                 struct vm_struct *area;
 221
 222                 area = __get_vm_area_caller(size, VM_IOREMAP,
 223                                             ioremap_bot, IOREMAP_END,
 224                                             caller);
 225                 if (area == NULL)
 226                         return NULL;
 227
 228                 area->phys_addr = paligned;
 229                 ret = __ioremap_at(paligned, area->addr, size, flags);
 230                 if (!ret)
 231                         vunmap(area->addr);
 232         } else {
 233                 ret = __ioremap_at(paligned, (void *)ioremap_bot, size, flags);
 234                 if (ret)
 235                         ioremap_bot += size;
 236         }
 237
 238         if (ret)
 239                 ret += addr & ~PAGE_MASK;
 240         return ret;
 241 }
 242
 243 void __iomem * __ioremap(phys_addr_t addr, unsigned long size,
 244                          unsigned long flags)
 245 {
 246         return __ioremap_caller(addr, size, flags, __builtin_return_address(0));
 247 }
 248
 249 void __iomem * ioremap(phys_addr_t addr, unsigned long size)
 250 {
 251         unsigned long flags = _PAGE_NO_CACHE | _PAGE_GUARDED;
 252         void *caller = __builtin_return_address(0);
 253
 254         if (ppc_md.ioremap)
 255                 return ppc_md.ioremap(addr, size, flags, caller);
 256         return __ioremap_caller(addr, size, flags, caller);
 257 }
 258
 259 void __iomem * ioremap_wc(phys_addr_t addr, unsigned long size)
 260 {
 261         unsigned long flags = _PAGE_NO_CACHE;
 262         void *caller = __builtin_return_address(0);
 263
 264         if (ppc_md.ioremap)
 265                 return ppc_md.ioremap(addr, size, flags, caller);
 266         return __ioremap_caller(addr, size, flags, caller);
 267 }
 268
 269 void __iomem * ioremap_prot(phys_addr_t addr, unsigned long size,
 270                              unsigned long flags)
 271 {
 272         void *caller = __builtin_return_address(0);
 273
 274         /* writeable implies dirty for kernel addresses */
 275         if (flags & _PAGE_RW)
 276                 flags |= _PAGE_DIRTY;
 277
 278         /* we don't want to let _PAGE_USER and _PAGE_EXEC leak out */
 279         flags &= ~(_PAGE_USER | _PAGE_EXEC);
 280
 281 #ifdef _PAGE_BAP_SR
 282         /* _PAGE_USER contains _PAGE_BAP_SR on BookE using the new PTE format
 283          * which means that we just cleared supervisor access... oops ;-) This
 284          * restores it
 285          */
 286         flags |= _PAGE_BAP_SR;
 287 #endif
 288
 289         if (ppc_md.ioremap)
 290                 return ppc_md.ioremap(addr, size, flags, caller);
 291         return __ioremap_caller(addr, size, flags, caller);
 292 }
 293
 294
 295 /*
 296  * Unmap an IO region and remove it from imalloc'd list.
 297  * Access to IO memory should be serialized by driver.
 298  */
 299 void __iounmap(volatile void __iomem *token)
 300 {
 301         void *addr;
 302
 303         if (!mem_init_done)
 304                 return;
 305
 306         addr = (void *) ((unsigned long __force)
 307                          PCI_FIX_ADDR(token) & PAGE_MASK);
 308         if ((unsigned long)addr < ioremap_bot) {
 309                 printk(KERN_WARNING "Attempt to iounmap early bolted mapping"
 310                        " at 0x%p\n", addr);
 311                 return;
 312         }
 313         vunmap(addr);
 314 }
 315
 316 void iounmap(volatile void __iomem *token)
 317 {
 318         if (ppc_md.iounmap)
 319                 ppc_md.iounmap(token);
 320         else
 321                 __iounmap(token);
 322 }
 323
 324 EXPORT_SYMBOL(ioremap);
 325 EXPORT_SYMBOL(ioremap_wc);
 326 EXPORT_SYMBOL(ioremap_prot);
 327 EXPORT_SYMBOL(__ioremap);
 328 EXPORT_SYMBOL(__ioremap_at);
 329 EXPORT_SYMBOL(iounmap);
 330 EXPORT_SYMBOL(__iounmap);
 331 EXPORT_SYMBOL(__iounmap_at);