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/slab.h>
  37 #include <linux/hugetlb.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/firmware.h>
  53 #include <asm/dma.h>
  54
  55 #include "mmu_decl.h"
  56
  57
  58 #ifdef CONFIG_PPC_BOOK3S_64
  59 /*
  60  * partition table and process table for ISA 3.0
  61  */
  62 struct prtb_entry *process_tb;
  63 struct patb_entry *partition_tb;
  64 /*
  65  * page table size
  66  */
  67 unsigned long __pte_index_size;
  68 EXPORT_SYMBOL(__pte_index_size);
  69 unsigned long __pmd_index_size;
  70 EXPORT_SYMBOL(__pmd_index_size);
  71 unsigned long __pud_index_size;
  72 EXPORT_SYMBOL(__pud_index_size);
  73 unsigned long __pgd_index_size;
  74 EXPORT_SYMBOL(__pgd_index_size);
  75 unsigned long __pud_cache_index;
  76 EXPORT_SYMBOL(__pud_cache_index);
  77 unsigned long __pte_table_size;
  78 EXPORT_SYMBOL(__pte_table_size);
  79 unsigned long __pmd_table_size;
  80 EXPORT_SYMBOL(__pmd_table_size);
  81 unsigned long __pud_table_size;
  82 EXPORT_SYMBOL(__pud_table_size);
  83 unsigned long __pgd_table_size;
  84 EXPORT_SYMBOL(__pgd_table_size);
  85 unsigned long __pmd_val_bits;
  86 EXPORT_SYMBOL(__pmd_val_bits);
  87 unsigned long __pud_val_bits;
  88 EXPORT_SYMBOL(__pud_val_bits);
  89 unsigned long __pgd_val_bits;
  90 EXPORT_SYMBOL(__pgd_val_bits);
  91 unsigned long __kernel_virt_start;
  92 EXPORT_SYMBOL(__kernel_virt_start);
  93 unsigned long __kernel_virt_size;
  94 EXPORT_SYMBOL(__kernel_virt_size);
  95 unsigned long __vmalloc_start;
  96 EXPORT_SYMBOL(__vmalloc_start);
  97 unsigned long __vmalloc_end;
  98 EXPORT_SYMBOL(__vmalloc_end);
  99 unsigned long __kernel_io_start;
 100 EXPORT_SYMBOL(__kernel_io_start);
 101 struct page *vmemmap;
 102 EXPORT_SYMBOL(vmemmap);
 103 unsigned long __pte_frag_nr;
 104 EXPORT_SYMBOL(__pte_frag_nr);
 105 unsigned long __pte_frag_size_shift;
 106 EXPORT_SYMBOL(__pte_frag_size_shift);
 107 unsigned long ioremap_bot;
 108 #else /* !CONFIG_PPC_BOOK3S_64 */
 109 unsigned long ioremap_bot = IOREMAP_BASE;
 110 #endif
 111
 112 /**
 113  * __ioremap_at - Low level function to establish the page tables
 114  *                for an IO mapping
 115  */
 116 void __iomem *__ioremap_at(phys_addr_t pa, void *ea, unsigned long size, pgprot_t prot)
 117 {
 118         unsigned long i;
 119
 120         /* We don't support the 4K PFN hack with ioremap */
 121         if (pgprot_val(prot) & H_PAGE_4K_PFN)
 122                 return NULL;
 123
 124         WARN_ON(pa & ~PAGE_MASK);
 125         WARN_ON(((unsigned long)ea) & ~PAGE_MASK);
 126         WARN_ON(size & ~PAGE_MASK);
 127
 128         for (i = 0; i < size; i += PAGE_SIZE)
 129                 if (map_kernel_page((unsigned long)ea + i, pa + i, prot))
 130                         return NULL;
 131
 132         return (void __iomem *)ea;
 133 }
 134
 135 /**
 136  * __iounmap_from - Low level function to tear down the page tables
 137  *                  for an IO mapping. This is used for mappings that
 138  *                  are manipulated manually, like partial unmapping of
 139  *                  PCI IOs or ISA space.
 140  */
 141 void __iounmap_at(void *ea, unsigned long size)
 142 {
 143         WARN_ON(((unsigned long)ea) & ~PAGE_MASK);
 144         WARN_ON(size & ~PAGE_MASK);
 145
 146         unmap_kernel_range((unsigned long)ea, size);
 147 }
 148
 149 void __iomem * __ioremap_caller(phys_addr_t addr, unsigned long size,
 150                                 pgprot_t prot, void *caller)
 151 {
 152         phys_addr_t paligned;
 153         void __iomem *ret;
 154
 155         /*
 156          * Choose an address to map it to.
 157          * Once the imalloc system is running, we use it.
 158          * Before that, we map using addresses going
 159          * up from ioremap_bot.  imalloc will use
 160          * the addresses from ioremap_bot through
 161          * IMALLOC_END
 162          *
 163          */
 164         paligned = addr & PAGE_MASK;
 165         size = PAGE_ALIGN(addr + size) - paligned;
 166
 167         if ((size == 0) || (paligned == 0))
 168                 return NULL;
 169
 170         if (slab_is_available()) {
 171                 struct vm_struct *area;
 172
 173                 area = __get_vm_area_caller(size, VM_IOREMAP,
 174                                             ioremap_bot, IOREMAP_END,
 175                                             caller);
 176                 if (area == NULL)
 177                         return NULL;
 178
 179                 area->phys_addr = paligned;
 180                 ret = __ioremap_at(paligned, area->addr, size, prot);
 181                 if (!ret)
 182                         vunmap(area->addr);
 183         } else {
 184                 ret = __ioremap_at(paligned, (void *)ioremap_bot, size, prot);
 185                 if (ret)
 186                         ioremap_bot += size;
 187         }
 188
 189         if (ret)
 190                 ret += addr & ~PAGE_MASK;
 191         return ret;
 192 }
 193
 194 void __iomem * __ioremap(phys_addr_t addr, unsigned long size,
 195                          unsigned long flags)
 196 {
 197         return __ioremap_caller(addr, size, __pgprot(flags), __builtin_return_address(0));
 198 }
 199
 200 void __iomem * ioremap(phys_addr_t addr, unsigned long size)
 201 {
 202         pgprot_t prot = pgprot_noncached(PAGE_KERNEL);
 203         void *caller = __builtin_return_address(0);
 204
 205         if (ppc_md.ioremap)
 206                 return ppc_md.ioremap(addr, size, prot, caller);
 207         return __ioremap_caller(addr, size, prot, caller);
 208 }
 209
 210 void __iomem * ioremap_wc(phys_addr_t addr, unsigned long size)
 211 {
 212         pgprot_t prot = pgprot_noncached_wc(PAGE_KERNEL);
 213         void *caller = __builtin_return_address(0);
 214
 215         if (ppc_md.ioremap)
 216                 return ppc_md.ioremap(addr, size, prot, caller);
 217         return __ioremap_caller(addr, size, prot, caller);
 218 }
 219
 220 void __iomem *ioremap_coherent(phys_addr_t addr, unsigned long size)
 221 {
 222         pgprot_t prot = pgprot_cached(PAGE_KERNEL);
 223         void *caller = __builtin_return_address(0);
 224
 225         if (ppc_md.ioremap)
 226                 return ppc_md.ioremap(addr, size, prot, caller);
 227         return __ioremap_caller(addr, size, prot, caller);
 228 }
 229
 230 void __iomem * ioremap_prot(phys_addr_t addr, unsigned long size,
 231                              unsigned long flags)
 232 {
 233         pte_t pte = __pte(flags);
 234         void *caller = __builtin_return_address(0);
 235
 236         /* writeable implies dirty for kernel addresses */
 237         if (pte_write(pte))
 238                 pte = pte_mkdirty(pte);
 239
 240         /* we don't want to let _PAGE_EXEC leak out */
 241         pte = pte_exprotect(pte);
 242         /*
 243          * Force kernel mapping.
 244          */
 245         pte = pte_mkprivileged(pte);
 246
 247         if (ppc_md.ioremap)
 248                 return ppc_md.ioremap(addr, size, pte_pgprot(pte), caller);
 249         return __ioremap_caller(addr, size, pte_pgprot(pte), caller);
 250 }
 251
 252
 253 /*
 254  * Unmap an IO region and remove it from imalloc'd list.
 255  * Access to IO memory should be serialized by driver.
 256  */
 257 void __iounmap(volatile void __iomem *token)
 258 {
 259         void *addr;
 260
 261         if (!slab_is_available())
 262                 return;
 263
 264         addr = (void *) ((unsigned long __force)
 265                          PCI_FIX_ADDR(token) & PAGE_MASK);
 266         if ((unsigned long)addr < ioremap_bot) {
 267                 printk(KERN_WARNING "Attempt to iounmap early bolted mapping"
 268                        " at 0x%p\n", addr);
 269                 return;
 270         }
 271         vunmap(addr);
 272 }
 273
 274 void iounmap(volatile void __iomem *token)
 275 {
 276         if (ppc_md.iounmap)
 277                 ppc_md.iounmap(token);
 278         else
 279                 __iounmap(token);
 280 }
 281
 282 EXPORT_SYMBOL(ioremap);
 283 EXPORT_SYMBOL(ioremap_wc);
 284 EXPORT_SYMBOL(ioremap_prot);
 285 EXPORT_SYMBOL(__ioremap);
 286 EXPORT_SYMBOL(__ioremap_at);
 287 EXPORT_SYMBOL(iounmap);
 288 EXPORT_SYMBOL(__iounmap);
 289 EXPORT_SYMBOL(__iounmap_at);
 290
 291 #ifndef __PAGETABLE_PUD_FOLDED
 292 /* 4 level page table */
 293 struct page *pgd_page(pgd_t pgd)
 294 {
 295         if (pgd_huge(pgd))
 296                 return pte_page(pgd_pte(pgd));
 297         return virt_to_page(pgd_page_vaddr(pgd));
 298 }
 299 #endif
 300
 301 struct page *pud_page(pud_t pud)
 302 {
 303         if (pud_huge(pud))
 304                 return pte_page(pud_pte(pud));
 305         return virt_to_page(pud_page_vaddr(pud));
 306 }
 307
 308 /*
 309  * For hugepage we have pfn in the pmd, we use PTE_RPN_SHIFT bits for flags
 310  * For PTE page, we have a PTE_FRAG_SIZE (4K) aligned virtual address.
 311  */
 312 struct page *pmd_page(pmd_t pmd)
 313 {
 314         if (pmd_large(pmd) || pmd_huge(pmd) || pmd_devmap(pmd))
 315                 return pte_page(pmd_pte(pmd));
 316         return virt_to_page(pmd_page_vaddr(pmd));
 317 }
 318
 319 #ifdef CONFIG_STRICT_KERNEL_RWX
 320 void mark_rodata_ro(void)
 321 {
 322         if (!mmu_has_feature(MMU_FTR_KERNEL_RO)) {
 323                 pr_warn("Warning: Unable to mark rodata read only on this CPU.\n");
 324                 return;
 325         }
 326
 327         if (radix_enabled())
 328                 radix__mark_rodata_ro();
 329         else
 330                 hash__mark_rodata_ro();
 331 }
 332
 333 void mark_initmem_nx(void)
 334 {
 335         if (radix_enabled())
 336                 radix__mark_initmem_nx();
 337         else
 338                 hash__mark_initmem_nx();
 339 }
 340 #endif