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