arch/powerpc/mm/pgtable_32.c

   1 /*
   2  * This file contains the routines setting up the linux page tables.
   3  *  -- paulus
   4  *
   5  *  Derived from arch/ppc/mm/init.c:
   6  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
   7  *
   8  *  Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
   9  *  and Cort Dougan (PReP) (cort@cs.nmt.edu)
  10  *    Copyright (C) 1996 Paul Mackerras
  11  *
  12  *  Derived from "arch/i386/mm/init.c"
  13  *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
  14  *
  15  *  This program is free software; you can redistribute it and/or
  16  *  modify it under the terms of the GNU General Public License
  17  *  as published by the Free Software Foundation; either version
  18  *  2 of the License, or (at your option) any later version.
  19  *
  20  */
  21
  22 #include <linux/kernel.h>
  23 #include <linux/module.h>
  24 #include <linux/types.h>
  25 #include <linux/mm.h>
  26 #include <linux/vmalloc.h>
  27 #include <linux/init.h>
  28 #include <linux/highmem.h>
  29 #include <linux/memblock.h>
  30 #include <linux/slab.h>
  31
  32 #include <asm/pgtable.h>
  33 #include <asm/pgalloc.h>
  34 #include <asm/fixmap.h>
  35 #include <asm/io.h>
  36 #include <asm/setup.h>
  37 #include <asm/sections.h>
  38
  39 #include "mmu_decl.h"
  40
  41 unsigned long ioremap_bot;
  42 EXPORT_SYMBOL(ioremap_bot);     /* aka VMALLOC_END */
  43
  44 extern char etext[], _stext[], _sinittext[], _einittext[];
  45
  46 __ref pte_t *pte_alloc_one_kernel(struct mm_struct *mm)
  47 {
  48         if (!slab_is_available())
  49                 return memblock_alloc(PTE_FRAG_SIZE, PTE_FRAG_SIZE);
  50
  51         return (pte_t *)pte_fragment_alloc(mm, 1);
  52 }
  53
  54 pgtable_t pte_alloc_one(struct mm_struct *mm)
  55 {
  56         return (pgtable_t)pte_fragment_alloc(mm, 0);
  57 }
  58
  59 void __iomem *
  60 ioremap(phys_addr_t addr, unsigned long size)
  61 {
  62         pgprot_t prot = pgprot_noncached(PAGE_KERNEL);
  63
  64         return __ioremap_caller(addr, size, prot, __builtin_return_address(0));
  65 }
  66 EXPORT_SYMBOL(ioremap);
  67
  68 void __iomem *
  69 ioremap_wc(phys_addr_t addr, unsigned long size)
  70 {
  71         pgprot_t prot = pgprot_noncached_wc(PAGE_KERNEL);
  72
  73         return __ioremap_caller(addr, size, prot, __builtin_return_address(0));
  74 }
  75 EXPORT_SYMBOL(ioremap_wc);
  76
  77 void __iomem *
  78 ioremap_wt(phys_addr_t addr, unsigned long size)
  79 {
  80         pgprot_t prot = pgprot_cached_wthru(PAGE_KERNEL);
  81
  82         return __ioremap_caller(addr, size, prot, __builtin_return_address(0));
  83 }
  84 EXPORT_SYMBOL(ioremap_wt);
  85
  86 void __iomem *
  87 ioremap_coherent(phys_addr_t addr, unsigned long size)
  88 {
  89         pgprot_t prot = pgprot_cached(PAGE_KERNEL);
  90
  91         return __ioremap_caller(addr, size, prot, __builtin_return_address(0));
  92 }
  93 EXPORT_SYMBOL(ioremap_coherent);
  94
  95 void __iomem *
  96 ioremap_prot(phys_addr_t addr, unsigned long size, unsigned long flags)
  97 {
  98         pte_t pte = __pte(flags);
  99
 100         /* writeable implies dirty for kernel addresses */
 101         if (pte_write(pte))
 102                 pte = pte_mkdirty(pte);
 103
 104         /* we don't want to let _PAGE_USER and _PAGE_EXEC leak out */
 105         pte = pte_exprotect(pte);
 106         pte = pte_mkprivileged(pte);
 107
 108         return __ioremap_caller(addr, size, pte_pgprot(pte), __builtin_return_address(0));
 109 }
 110 EXPORT_SYMBOL(ioremap_prot);
 111
 112 void __iomem *
 113 __ioremap(phys_addr_t addr, unsigned long size, unsigned long flags)
 114 {
 115         return __ioremap_caller(addr, size, __pgprot(flags), __builtin_return_address(0));
 116 }
 117
 118 void __iomem *
 119 __ioremap_caller(phys_addr_t addr, unsigned long size, pgprot_t prot, void *caller)
 120 {
 121         unsigned long v, i;
 122         phys_addr_t p;
 123         int err;
 124
 125         /*
 126          * Choose an address to map it to.
 127          * Once the vmalloc system is running, we use it.
 128          * Before then, we use space going down from IOREMAP_TOP
 129          * (ioremap_bot records where we're up to).
 130          */
 131         p = addr & PAGE_MASK;
 132         size = PAGE_ALIGN(addr + size) - p;
 133
 134         /*
 135          * If the address lies within the first 16 MB, assume it's in ISA
 136          * memory space
 137          */
 138         if (p < 16*1024*1024)
 139                 p += _ISA_MEM_BASE;
 140
 141 #ifndef CONFIG_CRASH_DUMP
 142         /*
 143          * Don't allow anybody to remap normal RAM that we're using.
 144          * mem_init() sets high_memory so only do the check after that.
 145          */
 146         if (slab_is_available() && p <= virt_to_phys(high_memory - 1) &&
 147             page_is_ram(__phys_to_pfn(p))) {
 148                 printk("__ioremap(): phys addr 0x%llx is RAM lr %ps\n",
 149                        (unsigned long long)p, __builtin_return_address(0));
 150                 return NULL;
 151         }
 152 #endif
 153
 154         if (size == 0)
 155                 return NULL;
 156
 157         /*
 158          * Is it already mapped?  Perhaps overlapped by a previous
 159          * mapping.
 160          */
 161         v = p_block_mapped(p);
 162         if (v)
 163                 goto out;
 164
 165         if (slab_is_available()) {
 166                 struct vm_struct *area;
 167                 area = get_vm_area_caller(size, VM_IOREMAP, caller);
 168                 if (area == 0)
 169                         return NULL;
 170                 area->phys_addr = p;
 171                 v = (unsigned long) area->addr;
 172         } else {
 173                 v = (ioremap_bot -= size);
 174         }
 175
 176         /*
 177          * Should check if it is a candidate for a BAT mapping
 178          */
 179
 180         err = 0;
 181         for (i = 0; i < size && err == 0; i += PAGE_SIZE)
 182                 err = map_kernel_page(v + i, p + i, prot);
 183         if (err) {
 184                 if (slab_is_available())
 185                         vunmap((void *)v);
 186                 return NULL;
 187         }
 188
 189 out:
 190         return (void __iomem *) (v + ((unsigned long)addr & ~PAGE_MASK));
 191 }
 192 EXPORT_SYMBOL(__ioremap);
 193
 194 void iounmap(volatile void __iomem *addr)
 195 {
 196         /*
 197          * If mapped by BATs then there is nothing to do.
 198          * Calling vfree() generates a benign warning.
 199          */
 200         if (v_block_mapped((unsigned long)addr))
 201                 return;
 202
 203         if (addr > high_memory && (unsigned long) addr < ioremap_bot)
 204                 vunmap((void *) (PAGE_MASK & (unsigned long)addr));
 205 }
 206 EXPORT_SYMBOL(iounmap);
 207
 208 int map_kernel_page(unsigned long va, phys_addr_t pa, pgprot_t prot)
 209 {
 210         pmd_t *pd;
 211         pte_t *pg;
 212         int err = -ENOMEM;
 213
 214         /* Use upper 10 bits of VA to index the first level map */
 215         pd = pmd_offset(pud_offset(pgd_offset_k(va), va), va);
 216         /* Use middle 10 bits of VA to index the second-level map */
 217         pg = pte_alloc_kernel(pd, va);
 218         if (pg != 0) {
 219                 err = 0;
 220                 /* The PTE should never be already set nor present in the
 221                  * hash table
 222                  */
 223                 BUG_ON((pte_present(*pg) | pte_hashpte(*pg)) && pgprot_val(prot));
 224                 set_pte_at(&init_mm, va, pg, pfn_pte(pa >> PAGE_SHIFT, prot));
 225         }
 226         smp_wmb();
 227         return err;
 228 }
 229
 230 /*
 231  * Map in a chunk of physical memory starting at start.
 232  */
 233 static void __init __mapin_ram_chunk(unsigned long offset, unsigned long top)
 234 {
 235         unsigned long v, s;
 236         phys_addr_t p;
 237         int ktext;
 238
 239         s = offset;
 240         v = PAGE_OFFSET + s;
 241         p = memstart_addr + s;
 242         for (; s < top; s += PAGE_SIZE) {
 243                 ktext = ((char *)v >= _stext && (char *)v < etext) ||
 244                         ((char *)v >= _sinittext && (char *)v < _einittext);
 245                 map_kernel_page(v, p, ktext ? PAGE_KERNEL_TEXT : PAGE_KERNEL);
 246 #ifdef CONFIG_PPC_BOOK3S_32
 247                 if (ktext)
 248                         hash_preload(&init_mm, v, false, 0x300);
 249 #endif
 250                 v += PAGE_SIZE;
 251                 p += PAGE_SIZE;
 252         }
 253 }
 254
 255 void __init mapin_ram(void)
 256 {
 257         struct memblock_region *reg;
 258
 259         for_each_memblock(memory, reg) {
 260                 phys_addr_t base = reg->base;
 261                 phys_addr_t top = min(base + reg->size, total_lowmem);
 262
 263                 if (base >= top)
 264                         continue;
 265                 base = mmu_mapin_ram(base, top);
 266                 if (IS_ENABLED(CONFIG_BDI_SWITCH))
 267                         __mapin_ram_chunk(reg->base, top);
 268                 else
 269                         __mapin_ram_chunk(base, top);
 270         }
 271 }
 272
 273 /* Scan the real Linux page tables and return a PTE pointer for
 274  * a virtual address in a context.
 275  * Returns true (1) if PTE was found, zero otherwise.  The pointer to
 276  * the PTE pointer is unmodified if PTE is not found.
 277  */
 278 static int
 279 get_pteptr(struct mm_struct *mm, unsigned long addr, pte_t **ptep, pmd_t **pmdp)
 280 {
 281         pgd_t   *pgd;
 282         pud_t   *pud;
 283         pmd_t   *pmd;
 284         pte_t   *pte;
 285         int     retval = 0;
 286
 287         pgd = pgd_offset(mm, addr & PAGE_MASK);
 288         if (pgd) {
 289                 pud = pud_offset(pgd, addr & PAGE_MASK);
 290                 if (pud && pud_present(*pud)) {
 291                         pmd = pmd_offset(pud, addr & PAGE_MASK);
 292                         if (pmd_present(*pmd)) {
 293                                 pte = pte_offset_map(pmd, addr & PAGE_MASK);
 294                                 if (pte) {
 295                                         retval = 1;
 296                                         *ptep = pte;
 297                                         if (pmdp)
 298                                                 *pmdp = pmd;
 299                                         /* XXX caller needs to do pte_unmap, yuck */
 300                                 }
 301                         }
 302                 }
 303         }
 304         return(retval);
 305 }
 306
 307 static int __change_page_attr_noflush(struct page *page, pgprot_t prot)
 308 {
 309         pte_t *kpte;
 310         pmd_t *kpmd;
 311         unsigned long address;
 312
 313         BUG_ON(PageHighMem(page));
 314         address = (unsigned long)page_address(page);
 315
 316         if (v_block_mapped(address))
 317                 return 0;
 318         if (!get_pteptr(&init_mm, address, &kpte, &kpmd))
 319                 return -EINVAL;
 320         __set_pte_at(&init_mm, address, kpte, mk_pte(page, prot), 0);
 321         pte_unmap(kpte);
 322
 323         return 0;
 324 }
 325
 326 /*
 327  * Change the page attributes of an page in the linear mapping.
 328  *
 329  * THIS DOES NOTHING WITH BAT MAPPINGS, DEBUG USE ONLY
 330  */
 331 static int change_page_attr(struct page *page, int numpages, pgprot_t prot)
 332 {
 333         int i, err = 0;
 334         unsigned long flags;
 335         struct page *start = page;
 336
 337         local_irq_save(flags);
 338         for (i = 0; i < numpages; i++, page++) {
 339                 err = __change_page_attr_noflush(page, prot);
 340                 if (err)
 341                         break;
 342         }
 343         wmb();
 344         local_irq_restore(flags);
 345         flush_tlb_kernel_range((unsigned long)page_address(start),
 346                                (unsigned long)page_address(page));
 347         return err;
 348 }
 349
 350 void mark_initmem_nx(void)
 351 {
 352         struct page *page = virt_to_page(_sinittext);
 353         unsigned long numpages = PFN_UP((unsigned long)_einittext) -
 354                                  PFN_DOWN((unsigned long)_sinittext);
 355
 356         if (v_block_mapped((unsigned long)_stext) + 1)
 357                 mmu_mark_initmem_nx();
 358         else
 359                 change_page_attr(page, numpages, PAGE_KERNEL);
 360 }
 361
 362 #ifdef CONFIG_STRICT_KERNEL_RWX
 363 void mark_rodata_ro(void)
 364 {
 365         struct page *page;
 366         unsigned long numpages;
 367
 368         if (v_block_mapped((unsigned long)_sinittext)) {
 369                 mmu_mark_rodata_ro();
 370                 return;
 371         }
 372
 373         page = virt_to_page(_stext);
 374         numpages = PFN_UP((unsigned long)_etext) -
 375                    PFN_DOWN((unsigned long)_stext);
 376
 377         change_page_attr(page, numpages, PAGE_KERNEL_ROX);
 378         /*
 379          * mark .rodata as read only. Use __init_begin rather than __end_rodata
 380          * to cover NOTES and EXCEPTION_TABLE.
 381          */
 382         page = virt_to_page(__start_rodata);
 383         numpages = PFN_UP((unsigned long)__init_begin) -
 384                    PFN_DOWN((unsigned long)__start_rodata);
 385
 386         change_page_attr(page, numpages, PAGE_KERNEL_RO);
 387 }
 388 #endif
 389
 390 #ifdef CONFIG_DEBUG_PAGEALLOC
 391 void __kernel_map_pages(struct page *page, int numpages, int enable)
 392 {
 393         if (PageHighMem(page))
 394                 return;
 395
 396         change_page_attr(page, numpages, enable ? PAGE_KERNEL : __pgprot(0));
 397 }
 398 #endif /* CONFIG_DEBUG_PAGEALLOC */