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