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/memblock.h>
  37 #include <linux/slab.h>
  38 #include <linux/hugetlb.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/firmware.h>
  54 #include <asm/dma.h>
  55
  56 #include "mmu_decl.h"
  57
  58 #ifdef CONFIG_PPC_STD_MMU_64
  59 #if TASK_SIZE_USER64 > (1UL << (ESID_BITS + SID_SHIFT))
  60 #error TASK_SIZE_USER64 exceeds user VSID range
  61 #endif
  62 #endif
  63
  64 #ifdef CONFIG_PPC_BOOK3S_64
  65 /*
  66  * partition table and process table for ISA 3.0
  67  */
  68 struct prtb_entry *process_tb;
  69 struct patb_entry *partition_tb;
  70 /*
  71  * page table size
  72  */
  73 unsigned long __pte_index_size;
  74 EXPORT_SYMBOL(__pte_index_size);
  75 unsigned long __pmd_index_size;
  76 EXPORT_SYMBOL(__pmd_index_size);
  77 unsigned long __pud_index_size;
  78 EXPORT_SYMBOL(__pud_index_size);
  79 unsigned long __pgd_index_size;
  80 EXPORT_SYMBOL(__pgd_index_size);
  81 unsigned long __pmd_cache_index;
  82 EXPORT_SYMBOL(__pmd_cache_index);
  83 unsigned long __pte_table_size;
  84 EXPORT_SYMBOL(__pte_table_size);
  85 unsigned long __pmd_table_size;
  86 EXPORT_SYMBOL(__pmd_table_size);
  87 unsigned long __pud_table_size;
  88 EXPORT_SYMBOL(__pud_table_size);
  89 unsigned long __pgd_table_size;
  90 EXPORT_SYMBOL(__pgd_table_size);
  91 unsigned long __pmd_val_bits;
  92 EXPORT_SYMBOL(__pmd_val_bits);
  93 unsigned long __pud_val_bits;
  94 EXPORT_SYMBOL(__pud_val_bits);
  95 unsigned long __pgd_val_bits;
  96 EXPORT_SYMBOL(__pgd_val_bits);
  97 unsigned long __kernel_virt_start;
  98 EXPORT_SYMBOL(__kernel_virt_start);
  99 unsigned long __kernel_virt_size;
 100 EXPORT_SYMBOL(__kernel_virt_size);
 101 unsigned long __vmalloc_start;
 102 EXPORT_SYMBOL(__vmalloc_start);
 103 unsigned long __vmalloc_end;
 104 EXPORT_SYMBOL(__vmalloc_end);
 105 struct page *vmemmap;
 106 EXPORT_SYMBOL(vmemmap);
 107 unsigned long __pte_frag_nr;
 108 EXPORT_SYMBOL(__pte_frag_nr);
 109 unsigned long __pte_frag_size_shift;
 110 EXPORT_SYMBOL(__pte_frag_size_shift);
 111 unsigned long ioremap_bot;
 112 #else /* !CONFIG_PPC_BOOK3S_64 */
 113 unsigned long ioremap_bot = IOREMAP_BASE;
 114 #endif
 115
 116 /**
 117  * __ioremap_at - Low level function to establish the page tables
 118  *                for an IO mapping
 119  */
 120 void __iomem * __ioremap_at(phys_addr_t pa, void *ea, unsigned long size,
 121                             unsigned long flags)
 122 {
 123         unsigned long i;
 124
 125         /* Make sure we have the base flags */
 126         if ((flags & _PAGE_PRESENT) == 0)
 127                 flags |= pgprot_val(PAGE_KERNEL);
 128
 129         /* We don't support the 4K PFN hack with ioremap */
 130         if (flags & H_PAGE_4K_PFN)
 131                 return NULL;
 132
 133         WARN_ON(pa & ~PAGE_MASK);
 134         WARN_ON(((unsigned long)ea) & ~PAGE_MASK);
 135         WARN_ON(size & ~PAGE_MASK);
 136
 137         for (i = 0; i < size; i += PAGE_SIZE)
 138                 if (map_kernel_page((unsigned long)ea+i, pa+i, flags))
 139                         return NULL;
 140
 141         return (void __iomem *)ea;
 142 }
 143
 144 /**
 145  * __iounmap_from - Low level function to tear down the page tables
 146  *                  for an IO mapping. This is used for mappings that
 147  *                  are manipulated manually, like partial unmapping of
 148  *                  PCI IOs or ISA space.
 149  */
 150 void __iounmap_at(void *ea, unsigned long size)
 151 {
 152         WARN_ON(((unsigned long)ea) & ~PAGE_MASK);
 153         WARN_ON(size & ~PAGE_MASK);
 154
 155         unmap_kernel_range((unsigned long)ea, size);
 156 }
 157
 158 void __iomem * __ioremap_caller(phys_addr_t addr, unsigned long size,
 159                                 unsigned long flags, void *caller)
 160 {
 161         phys_addr_t paligned;
 162         void __iomem *ret;
 163
 164         /*
 165          * Choose an address to map it to.
 166          * Once the imalloc system is running, we use it.
 167          * Before that, we map using addresses going
 168          * up from ioremap_bot.  imalloc will use
 169          * the addresses from ioremap_bot through
 170          * IMALLOC_END
 171          *
 172          */
 173         paligned = addr & PAGE_MASK;
 174         size = PAGE_ALIGN(addr + size) - paligned;
 175
 176         if ((size == 0) || (paligned == 0))
 177                 return NULL;
 178
 179         if (slab_is_available()) {
 180                 struct vm_struct *area;
 181
 182                 area = __get_vm_area_caller(size, VM_IOREMAP,
 183                                             ioremap_bot, IOREMAP_END,
 184                                             caller);
 185                 if (area == NULL)
 186                         return NULL;
 187
 188                 area->phys_addr = paligned;
 189                 ret = __ioremap_at(paligned, area->addr, size, flags);
 190                 if (!ret)
 191                         vunmap(area->addr);
 192         } else {
 193                 ret = __ioremap_at(paligned, (void *)ioremap_bot, size, flags);
 194                 if (ret)
 195                         ioremap_bot += size;
 196         }
 197
 198         if (ret)
 199                 ret += addr & ~PAGE_MASK;
 200         return ret;
 201 }
 202
 203 void __iomem * __ioremap(phys_addr_t addr, unsigned long size,
 204                          unsigned long flags)
 205 {
 206         return __ioremap_caller(addr, size, flags, __builtin_return_address(0));
 207 }
 208
 209 void __iomem * ioremap(phys_addr_t addr, unsigned long size)
 210 {
 211         unsigned long flags = pgprot_val(pgprot_noncached(__pgprot(0)));
 212         void *caller = __builtin_return_address(0);
 213
 214         if (ppc_md.ioremap)
 215                 return ppc_md.ioremap(addr, size, flags, caller);
 216         return __ioremap_caller(addr, size, flags, caller);
 217 }
 218
 219 void __iomem * ioremap_wc(phys_addr_t addr, unsigned long size)
 220 {
 221         unsigned long flags = pgprot_val(pgprot_noncached_wc(__pgprot(0)));
 222         void *caller = __builtin_return_address(0);
 223
 224         if (ppc_md.ioremap)
 225                 return ppc_md.ioremap(addr, size, flags, caller);
 226         return __ioremap_caller(addr, size, flags, caller);
 227 }
 228
 229 void __iomem * ioremap_prot(phys_addr_t addr, unsigned long size,
 230                              unsigned long flags)
 231 {
 232         void *caller = __builtin_return_address(0);
 233
 234         /* writeable implies dirty for kernel addresses */
 235         if (flags & _PAGE_WRITE)
 236                 flags |= _PAGE_DIRTY;
 237
 238         /* we don't want to let _PAGE_EXEC leak out */
 239         flags &= ~_PAGE_EXEC;
 240         /*
 241          * Force kernel mapping.
 242          */
 243 #if defined(CONFIG_PPC_BOOK3S_64)
 244         flags |= _PAGE_PRIVILEGED;
 245 #else
 246         flags &= ~_PAGE_USER;
 247 #endif
 248
 249
 250 #ifdef _PAGE_BAP_SR
 251         /* _PAGE_USER contains _PAGE_BAP_SR on BookE using the new PTE format
 252          * which means that we just cleared supervisor access... oops ;-) This
 253          * restores it
 254          */
 255         flags |= _PAGE_BAP_SR;
 256 #endif
 257
 258         if (ppc_md.ioremap)
 259                 return ppc_md.ioremap(addr, size, flags, caller);
 260         return __ioremap_caller(addr, size, flags, caller);
 261 }
 262
 263
 264 /*
 265  * Unmap an IO region and remove it from imalloc'd list.
 266  * Access to IO memory should be serialized by driver.
 267  */
 268 void __iounmap(volatile void __iomem *token)
 269 {
 270         void *addr;
 271
 272         if (!slab_is_available())
 273                 return;
 274
 275         addr = (void *) ((unsigned long __force)
 276                          PCI_FIX_ADDR(token) & PAGE_MASK);
 277         if ((unsigned long)addr < ioremap_bot) {
 278                 printk(KERN_WARNING "Attempt to iounmap early bolted mapping"
 279                        " at 0x%p\n", addr);
 280                 return;
 281         }
 282         vunmap(addr);
 283 }
 284
 285 void iounmap(volatile void __iomem *token)
 286 {
 287         if (ppc_md.iounmap)
 288                 ppc_md.iounmap(token);
 289         else
 290                 __iounmap(token);
 291 }
 292
 293 EXPORT_SYMBOL(ioremap);
 294 EXPORT_SYMBOL(ioremap_wc);
 295 EXPORT_SYMBOL(ioremap_prot);
 296 EXPORT_SYMBOL(__ioremap);
 297 EXPORT_SYMBOL(__ioremap_at);
 298 EXPORT_SYMBOL(iounmap);
 299 EXPORT_SYMBOL(__iounmap);
 300 EXPORT_SYMBOL(__iounmap_at);
 301
 302 #ifndef __PAGETABLE_PUD_FOLDED
 303 /* 4 level page table */
 304 struct page *pgd_page(pgd_t pgd)
 305 {
 306         if (pgd_huge(pgd))
 307                 return pte_page(pgd_pte(pgd));
 308         return virt_to_page(pgd_page_vaddr(pgd));
 309 }
 310 #endif
 311
 312 struct page *pud_page(pud_t pud)
 313 {
 314         if (pud_huge(pud))
 315                 return pte_page(pud_pte(pud));
 316         return virt_to_page(pud_page_vaddr(pud));
 317 }
 318
 319 /*
 320  * For hugepage we have pfn in the pmd, we use PTE_RPN_SHIFT bits for flags
 321  * For PTE page, we have a PTE_FRAG_SIZE (4K) aligned virtual address.
 322  */
 323 struct page *pmd_page(pmd_t pmd)
 324 {
 325         if (pmd_trans_huge(pmd) || pmd_huge(pmd))
 326                 return pte_page(pmd_pte(pmd));
 327         return virt_to_page(pmd_page_vaddr(pmd));
 328 }
 329
 330 #ifdef CONFIG_PPC_64K_PAGES
 331 static pte_t *get_from_cache(struct mm_struct *mm)
 332 {
 333         void *pte_frag, *ret;
 334
 335         spin_lock(&mm->page_table_lock);
 336         ret = mm->context.pte_frag;
 337         if (ret) {
 338                 pte_frag = ret + PTE_FRAG_SIZE;
 339                 /*
 340                  * If we have taken up all the fragments mark PTE page NULL
 341                  */
 342                 if (((unsigned long)pte_frag & ~PAGE_MASK) == 0)
 343                         pte_frag = NULL;
 344                 mm->context.pte_frag = pte_frag;
 345         }
 346         spin_unlock(&mm->page_table_lock);
 347         return (pte_t *)ret;
 348 }
 349
 350 static pte_t *__alloc_for_cache(struct mm_struct *mm, int kernel)
 351 {
 352         void *ret = NULL;
 353         struct page *page = alloc_page(GFP_KERNEL | __GFP_NOTRACK |
 354                                        __GFP_REPEAT | __GFP_ZERO);
 355         if (!page)
 356                 return NULL;
 357         if (!kernel && !pgtable_page_ctor(page)) {
 358                 __free_page(page);
 359                 return NULL;
 360         }
 361
 362         ret = page_address(page);
 363         spin_lock(&mm->page_table_lock);
 364         /*
 365          * If we find pgtable_page set, we return
 366          * the allocated page with single fragement
 367          * count.
 368          */
 369         if (likely(!mm->context.pte_frag)) {
 370                 set_page_count(page, PTE_FRAG_NR);
 371                 mm->context.pte_frag = ret + PTE_FRAG_SIZE;
 372         }
 373         spin_unlock(&mm->page_table_lock);
 374
 375         return (pte_t *)ret;
 376 }
 377
 378 pte_t *pte_fragment_alloc(struct mm_struct *mm, unsigned long vmaddr, int kernel)
 379 {
 380         pte_t *pte;
 381
 382         pte = get_from_cache(mm);
 383         if (pte)
 384                 return pte;
 385
 386         return __alloc_for_cache(mm, kernel);
 387 }
 388 #endif /* CONFIG_PPC_64K_PAGES */
 389
 390 void pte_fragment_free(unsigned long *table, int kernel)
 391 {
 392         struct page *page = virt_to_page(table);
 393         if (put_page_testzero(page)) {
 394                 if (!kernel)
 395                         pgtable_page_dtor(page);
 396                 free_hot_cold_page(page, 0);
 397         }
 398 }
 399
 400 #ifdef CONFIG_SMP
 401 void pgtable_free_tlb(struct mmu_gather *tlb, void *table, int shift)
 402 {
 403         unsigned long pgf = (unsigned long)table;
 404
 405         BUG_ON(shift > MAX_PGTABLE_INDEX_SIZE);
 406         pgf |= shift;
 407         tlb_remove_table(tlb, (void *)pgf);
 408 }
 409
 410 void __tlb_remove_table(void *_table)
 411 {
 412         void *table = (void *)((unsigned long)_table & ~MAX_PGTABLE_INDEX_SIZE);
 413         unsigned shift = (unsigned long)_table & MAX_PGTABLE_INDEX_SIZE;
 414
 415         if (!shift)
 416                 /* PTE page needs special handling */
 417                 pte_fragment_free(table, 0);
 418         else {
 419                 BUG_ON(shift > MAX_PGTABLE_INDEX_SIZE);
 420                 kmem_cache_free(PGT_CACHE(shift), table);
 421         }
 422 }
 423 #else
 424 void pgtable_free_tlb(struct mmu_gather *tlb, void *table, int shift)
 425 {
 426         if (!shift) {
 427                 /* PTE page needs special handling */
 428                 pte_fragment_free(table, 0);
 429         } else {
 430                 BUG_ON(shift > MAX_PGTABLE_INDEX_SIZE);
 431                 kmem_cache_free(PGT_CACHE(shift), table);
 432         }
 433 }
 434 #endif