include/asm-x86/pgtable.h

   1 #ifndef _ASM_X86_PGTABLE_H
   2 #define _ASM_X86_PGTABLE_H
   3
   4 #define USER_PTRS_PER_PGD       ((TASK_SIZE-1)/PGDIR_SIZE+1)
   5 #define FIRST_USER_ADDRESS      0
   6
   7 #define _PAGE_BIT_PRESENT       0
   8 #define _PAGE_BIT_RW            1
   9 #define _PAGE_BIT_USER          2
  10 #define _PAGE_BIT_PWT           3
  11 #define _PAGE_BIT_PCD           4
  12 #define _PAGE_BIT_ACCESSED      5
  13 #define _PAGE_BIT_DIRTY         6
  14 #define _PAGE_BIT_FILE          6
  15 #define _PAGE_BIT_PSE           7       /* 4 MB (or 2MB) page */
  16 #define _PAGE_BIT_GLOBAL        8       /* Global TLB entry PPro+ */
  17 #define _PAGE_BIT_UNUSED1       9       /* available for programmer */
  18 #define _PAGE_BIT_UNUSED2       10
  19 #define _PAGE_BIT_UNUSED3       11
  20 #define _PAGE_BIT_NX           63       /* No execute: only valid after cpuid check */
  21
  22 /*
  23  * Note: we use _AC(1, L) instead of _AC(1, UL) so that we get a
  24  * sign-extended value on 32-bit with all 1's in the upper word,
  25  * which preserves the upper pte values on 64-bit ptes:
  26  */
  27 #define _PAGE_PRESENT   (_AC(1, L)<<_PAGE_BIT_PRESENT)
  28 #define _PAGE_RW        (_AC(1, L)<<_PAGE_BIT_RW)
  29 #define _PAGE_USER      (_AC(1, L)<<_PAGE_BIT_USER)
  30 #define _PAGE_PWT       (_AC(1, L)<<_PAGE_BIT_PWT)
  31 #define _PAGE_PCD       (_AC(1, L)<<_PAGE_BIT_PCD)
  32 #define _PAGE_ACCESSED  (_AC(1, L)<<_PAGE_BIT_ACCESSED)
  33 #define _PAGE_DIRTY     (_AC(1, L)<<_PAGE_BIT_DIRTY)
  34 #define _PAGE_PSE       (_AC(1, L)<<_PAGE_BIT_PSE)      /* 2MB page */
  35 #define _PAGE_GLOBAL    (_AC(1, L)<<_PAGE_BIT_GLOBAL)   /* Global TLB entry */
  36 #define _PAGE_UNUSED1   (_AC(1, L)<<_PAGE_BIT_UNUSED1)
  37 #define _PAGE_UNUSED2   (_AC(1, L)<<_PAGE_BIT_UNUSED2)
  38 #define _PAGE_UNUSED3   (_AC(1, L)<<_PAGE_BIT_UNUSED3)
  39
  40 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
  41 #define _PAGE_NX        (_AC(1, ULL) << _PAGE_BIT_NX)
  42 #else
  43 #define _PAGE_NX        0
  44 #endif
  45
  46 /* If _PAGE_PRESENT is clear, we use these: */
  47 #define _PAGE_FILE      _PAGE_DIRTY     /* nonlinear file mapping, saved PTE; unset:swap */
  48 #define _PAGE_PROTNONE  _PAGE_PSE       /* if the user mapped it with PROT_NONE;
  49                                            pte_present gives true */
  50
  51 #define _PAGE_TABLE     (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER | _PAGE_ACCESSED | _PAGE_DIRTY)
  52 #define _KERNPG_TABLE   (_PAGE_PRESENT | _PAGE_RW | _PAGE_ACCESSED | _PAGE_DIRTY)
  53
  54 #define _PAGE_CHG_MASK  (PTE_MASK | _PAGE_ACCESSED | _PAGE_DIRTY)
  55
  56 #define PAGE_NONE       __pgprot(_PAGE_PROTNONE | _PAGE_ACCESSED)
  57 #define PAGE_SHARED     __pgprot(_PAGE_PRESENT | _PAGE_RW | _PAGE_USER | _PAGE_ACCESSED | _PAGE_NX)
  58
  59 #define PAGE_SHARED_EXEC        __pgprot(_PAGE_PRESENT | _PAGE_RW | _PAGE_USER | _PAGE_ACCESSED)
  60 #define PAGE_COPY_NOEXEC        __pgprot(_PAGE_PRESENT | _PAGE_USER | _PAGE_ACCESSED | _PAGE_NX)
  61 #define PAGE_COPY_EXEC          __pgprot(_PAGE_PRESENT | _PAGE_USER | _PAGE_ACCESSED)
  62 #define PAGE_COPY               PAGE_COPY_NOEXEC
  63 #define PAGE_READONLY           __pgprot(_PAGE_PRESENT | _PAGE_USER | _PAGE_ACCESSED | _PAGE_NX)
  64 #define PAGE_READONLY_EXEC      __pgprot(_PAGE_PRESENT | _PAGE_USER | _PAGE_ACCESSED)
  65
  66 #ifdef CONFIG_X86_32
  67 #define _PAGE_KERNEL_EXEC \
  68         (_PAGE_PRESENT | _PAGE_RW | _PAGE_DIRTY | _PAGE_ACCESSED)
  69 #define _PAGE_KERNEL (_PAGE_KERNEL_EXEC | _PAGE_NX)
  70
  71 #ifndef __ASSEMBLY__
  72 extern pteval_t __PAGE_KERNEL, __PAGE_KERNEL_EXEC;
  73 #endif  /* __ASSEMBLY__ */
  74 #else
  75 #define __PAGE_KERNEL_EXEC                                              \
  76         (_PAGE_PRESENT | _PAGE_RW | _PAGE_DIRTY | _PAGE_ACCESSED)
  77 #define __PAGE_KERNEL           (__PAGE_KERNEL_EXEC | _PAGE_NX)
  78 #endif
  79
  80 #define __PAGE_KERNEL_RO                (__PAGE_KERNEL & ~_PAGE_RW)
  81 #define __PAGE_KERNEL_RX                (__PAGE_KERNEL_EXEC & ~_PAGE_RW)
  82 #define __PAGE_KERNEL_NOCACHE           (__PAGE_KERNEL | _PAGE_PCD | _PAGE_PWT)
  83 #define __PAGE_KERNEL_VSYSCALL          (__PAGE_KERNEL_RX | _PAGE_USER)
  84 #define __PAGE_KERNEL_VSYSCALL_NOCACHE  (__PAGE_KERNEL_VSYSCALL | _PAGE_PCD | _PAGE_PWT)
  85 #define __PAGE_KERNEL_LARGE             (__PAGE_KERNEL | _PAGE_PSE)
  86 #define __PAGE_KERNEL_LARGE_EXEC        (__PAGE_KERNEL_EXEC | _PAGE_PSE)
  87
  88 #ifdef CONFIG_X86_32
  89 # define MAKE_GLOBAL(x)                 __pgprot((x))
  90 #else
  91 # define MAKE_GLOBAL(x)                 __pgprot((x) | _PAGE_GLOBAL)
  92 #endif
  93
  94 #define PAGE_KERNEL                     MAKE_GLOBAL(__PAGE_KERNEL)
  95 #define PAGE_KERNEL_RO                  MAKE_GLOBAL(__PAGE_KERNEL_RO)
  96 #define PAGE_KERNEL_EXEC                MAKE_GLOBAL(__PAGE_KERNEL_EXEC)
  97 #define PAGE_KERNEL_RX                  MAKE_GLOBAL(__PAGE_KERNEL_RX)
  98 #define PAGE_KERNEL_NOCACHE             MAKE_GLOBAL(__PAGE_KERNEL_NOCACHE)
  99 #define PAGE_KERNEL_LARGE               MAKE_GLOBAL(__PAGE_KERNEL_LARGE)
 100 #define PAGE_KERNEL_LARGE_EXEC          MAKE_GLOBAL(__PAGE_KERNEL_LARGE_EXEC)
 101 #define PAGE_KERNEL_VSYSCALL            MAKE_GLOBAL(__PAGE_KERNEL_VSYSCALL)
 102 #define PAGE_KERNEL_VSYSCALL_NOCACHE    MAKE_GLOBAL(__PAGE_KERNEL_VSYSCALL_NOCACHE)
 103
 104 /*         xwr */
 105 #define __P000  PAGE_NONE
 106 #define __P001  PAGE_READONLY
 107 #define __P010  PAGE_COPY
 108 #define __P011  PAGE_COPY
 109 #define __P100  PAGE_READONLY_EXEC
 110 #define __P101  PAGE_READONLY_EXEC
 111 #define __P110  PAGE_COPY_EXEC
 112 #define __P111  PAGE_COPY_EXEC
 113
 114 #define __S000  PAGE_NONE
 115 #define __S001  PAGE_READONLY
 116 #define __S010  PAGE_SHARED
 117 #define __S011  PAGE_SHARED
 118 #define __S100  PAGE_READONLY_EXEC
 119 #define __S101  PAGE_READONLY_EXEC
 120 #define __S110  PAGE_SHARED_EXEC
 121 #define __S111  PAGE_SHARED_EXEC
 122
 123 #ifndef __ASSEMBLY__
 124
 125 /*
 126  * ZERO_PAGE is a global shared page that is always zero: used
 127  * for zero-mapped memory areas etc..
 128  */
 129 extern unsigned long empty_zero_page[PAGE_SIZE/sizeof(unsigned long)];
 130 #define ZERO_PAGE(vaddr) (virt_to_page(empty_zero_page))
 131
 132
 133 /*
 134  * The following only work if pte_present() is true.
 135  * Undefined behaviour if not..
 136  */
 137 static inline int pte_dirty(pte_t pte)          { return pte_val(pte) & _PAGE_DIRTY; }
 138 static inline int pte_young(pte_t pte)          { return pte_val(pte) & _PAGE_ACCESSED; }
 139 static inline int pte_write(pte_t pte)          { return pte_val(pte) & _PAGE_RW; }
 140 static inline int pte_file(pte_t pte)           { return pte_val(pte) & _PAGE_FILE; }
 141 static inline int pte_huge(pte_t pte)           { return pte_val(pte) & _PAGE_PSE; }
 142 static inline int pte_global(pte_t pte)         { return pte_val(pte) & _PAGE_GLOBAL; }
 143 static inline int pte_exec(pte_t pte)           { return !(pte_val(pte) & _PAGE_NX); }
 144
 145 static inline int pmd_large(pmd_t pte) {
 146         return (pmd_val(pte) & (_PAGE_PSE|_PAGE_PRESENT)) ==
 147                 (_PAGE_PSE|_PAGE_PRESENT);
 148 }
 149
 150 static inline pte_t pte_mkclean(pte_t pte)      { return __pte(pte_val(pte) & ~(pteval_t)_PAGE_DIRTY); }
 151 static inline pte_t pte_mkold(pte_t pte)        { return __pte(pte_val(pte) & ~(pteval_t)_PAGE_ACCESSED); }
 152 static inline pte_t pte_wrprotect(pte_t pte)    { return __pte(pte_val(pte) & ~(pteval_t)_PAGE_RW); }
 153 static inline pte_t pte_mkexec(pte_t pte)       { return __pte(pte_val(pte) & ~(pteval_t)_PAGE_NX); }
 154 static inline pte_t pte_mkdirty(pte_t pte)      { return __pte(pte_val(pte) | _PAGE_DIRTY); }
 155 static inline pte_t pte_mkyoung(pte_t pte)      { return __pte(pte_val(pte) | _PAGE_ACCESSED); }
 156 static inline pte_t pte_mkwrite(pte_t pte)      { return __pte(pte_val(pte) | _PAGE_RW); }
 157 static inline pte_t pte_mkhuge(pte_t pte)       { return __pte(pte_val(pte) | _PAGE_PSE); }
 158 static inline pte_t pte_clrhuge(pte_t pte)      { return __pte(pte_val(pte) & ~(pteval_t)_PAGE_PSE); }
 159 static inline pte_t pte_mkglobal(pte_t pte)     { return __pte(pte_val(pte) | _PAGE_GLOBAL); }
 160 static inline pte_t pte_clrglobal(pte_t pte)    { return __pte(pte_val(pte) & ~(pteval_t)_PAGE_GLOBAL); }
 161
 162 extern pteval_t __supported_pte_mask;
 163
 164 static inline pte_t pfn_pte(unsigned long page_nr, pgprot_t pgprot)
 165 {
 166         return __pte((((phys_addr_t)page_nr << PAGE_SHIFT) |
 167                       pgprot_val(pgprot)) & __supported_pte_mask);
 168 }
 169
 170 static inline pmd_t pfn_pmd(unsigned long page_nr, pgprot_t pgprot)
 171 {
 172         return __pmd((((phys_addr_t)page_nr << PAGE_SHIFT) |
 173                       pgprot_val(pgprot)) & __supported_pte_mask);
 174 }
 175
 176 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
 177 {
 178         pteval_t val = pte_val(pte);
 179
 180         /*
 181          * Chop off the NX bit (if present), and add the NX portion of
 182          * the newprot (if present):
 183          */
 184         val &= _PAGE_CHG_MASK & ~_PAGE_NX;
 185         val |= pgprot_val(newprot) & __supported_pte_mask;
 186
 187         return __pte(val);
 188 }
 189
 190 #define pte_pgprot(x) __pgprot(pte_val(x) & (0xfff | _PAGE_NX))
 191
 192 #define canon_pgprot(p) __pgprot(pgprot_val(p) & __supported_pte_mask)
 193
 194 #ifdef CONFIG_PARAVIRT
 195 #include <asm/paravirt.h>
 196 #else  /* !CONFIG_PARAVIRT */
 197 #define set_pte(ptep, pte)              native_set_pte(ptep, pte)
 198 #define set_pte_at(mm, addr, ptep, pte) native_set_pte_at(mm, addr, ptep, pte)
 199
 200 #define set_pte_present(mm, addr, ptep, pte)                            \
 201         native_set_pte_present(mm, addr, ptep, pte)
 202 #define set_pte_atomic(ptep, pte)                                       \
 203         native_set_pte_atomic(ptep, pte)
 204
 205 #define set_pmd(pmdp, pmd)              native_set_pmd(pmdp, pmd)
 206
 207 #ifndef __PAGETABLE_PUD_FOLDED
 208 #define set_pgd(pgdp, pgd)              native_set_pgd(pgdp, pgd)
 209 #define pgd_clear(pgd)                  native_pgd_clear(pgd)
 210 #endif
 211
 212 #ifndef set_pud
 213 # define set_pud(pudp, pud)             native_set_pud(pudp, pud)
 214 #endif
 215
 216 #ifndef __PAGETABLE_PMD_FOLDED
 217 #define pud_clear(pud)                  native_pud_clear(pud)
 218 #endif
 219
 220 #define pte_clear(mm, addr, ptep)       native_pte_clear(mm, addr, ptep)
 221 #define pmd_clear(pmd)                  native_pmd_clear(pmd)
 222
 223 #define pte_update(mm, addr, ptep)              do { } while (0)
 224 #define pte_update_defer(mm, addr, ptep)        do { } while (0)
 225 #endif  /* CONFIG_PARAVIRT */
 226
 227 #endif  /* __ASSEMBLY__ */
 228
 229 #ifdef CONFIG_X86_32
 230 # include "pgtable_32.h"
 231 #else
 232 # include "pgtable_64.h"
 233 #endif
 234
 235 #ifndef __ASSEMBLY__
 236
 237 /*
 238  * Helper function that returns the kernel pagetable entry controlling
 239  * the virtual address 'address'. NULL means no pagetable entry present.
 240  * NOTE: the return type is pte_t but if the pmd is PSE then we return it
 241  * as a pte too.
 242  */
 243 extern pte_t *lookup_address(unsigned long address, int *level);
 244
 245 /* local pte updates need not use xchg for locking */
 246 static inline pte_t native_local_ptep_get_and_clear(pte_t *ptep)
 247 {
 248         pte_t res = *ptep;
 249
 250         /* Pure native function needs no input for mm, addr */
 251         native_pte_clear(NULL, 0, ptep);
 252         return res;
 253 }
 254
 255 static inline void native_set_pte_at(struct mm_struct *mm, unsigned long addr,
 256                                      pte_t *ptep , pte_t pte)
 257 {
 258         native_set_pte(ptep, pte);
 259 }
 260
 261 #ifndef CONFIG_PARAVIRT
 262 /*
 263  * Rules for using pte_update - it must be called after any PTE update which
 264  * has not been done using the set_pte / clear_pte interfaces.  It is used by
 265  * shadow mode hypervisors to resynchronize the shadow page tables.  Kernel PTE
 266  * updates should either be sets, clears, or set_pte_atomic for P->P
 267  * transitions, which means this hook should only be called for user PTEs.
 268  * This hook implies a P->P protection or access change has taken place, which
 269  * requires a subsequent TLB flush.  The notification can optionally be delayed
 270  * until the TLB flush event by using the pte_update_defer form of the
 271  * interface, but care must be taken to assure that the flush happens while
 272  * still holding the same page table lock so that the shadow and primary pages
 273  * do not become out of sync on SMP.
 274  */
 275 #define pte_update(mm, addr, ptep)              do { } while (0)
 276 #define pte_update_defer(mm, addr, ptep)        do { } while (0)
 277 #endif
 278
 279 /*
 280  * We only update the dirty/accessed state if we set
 281  * the dirty bit by hand in the kernel, since the hardware
 282  * will do the accessed bit for us, and we don't want to
 283  * race with other CPU's that might be updating the dirty
 284  * bit at the same time.
 285  */
 286 #define  __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
 287 #define ptep_set_access_flags(vma, address, ptep, entry, dirty)         \
 288 ({                                                                      \
 289         int __changed = !pte_same(*(ptep), entry);                      \
 290         if (__changed && dirty) {                                       \
 291                 *ptep = entry;                                          \
 292                 pte_update_defer((vma)->vm_mm, (address), (ptep));      \
 293                 flush_tlb_page(vma, address);                           \
 294         }                                                               \
 295         __changed;                                                      \
 296 })
 297
 298 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
 299 #define ptep_test_and_clear_young(vma, addr, ptep) ({                   \
 300         int __ret = 0;                                                  \
 301         if (pte_young(*(ptep)))                                         \
 302                 __ret = test_and_clear_bit(_PAGE_BIT_ACCESSED,          \
 303                                            &(ptep)->pte);               \
 304         if (__ret)                                                      \
 305                 pte_update((vma)->vm_mm, addr, ptep);                   \
 306         __ret;                                                          \
 307 })
 308
 309 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
 310 #define ptep_clear_flush_young(vma, address, ptep)                      \
 311 ({                                                                      \
 312         int __young;                                                    \
 313         __young = ptep_test_and_clear_young((vma), (address), (ptep));  \
 314         if (__young)                                                    \
 315                 flush_tlb_page(vma, address);                           \
 316         __young;                                                        \
 317 })
 318
 319 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
 320 static inline pte_t ptep_get_and_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
 321 {
 322         pte_t pte = native_ptep_get_and_clear(ptep);
 323         pte_update(mm, addr, ptep);
 324         return pte;
 325 }
 326
 327 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
 328 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm, unsigned long addr, pte_t *ptep, int full)
 329 {
 330         pte_t pte;
 331         if (full) {
 332                 /*
 333                  * Full address destruction in progress; paravirt does not
 334                  * care about updates and native needs no locking
 335                  */
 336                 pte = native_local_ptep_get_and_clear(ptep);
 337         } else {
 338                 pte = ptep_get_and_clear(mm, addr, ptep);
 339         }
 340         return pte;
 341 }
 342
 343 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
 344 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
 345 {
 346         clear_bit(_PAGE_BIT_RW, (unsigned long *)&ptep->pte);
 347         pte_update(mm, addr, ptep);
 348 }
 349
 350 #include <asm-generic/pgtable.h>
 351 #endif  /* __ASSEMBLY__ */
 352
 353 #endif  /* _ASM_X86_PGTABLE_H */