include/asm/pgtable_64.h

0001 /* SPDX-License-Identifier: GPL-2.0 */
0002 #ifndef _ASM_X86_PGTABLE_64_H
0003 #define _ASM_X86_PGTABLE_64_H
0004
0005 #include <linux/const.h>
0006 #include <asm/pgtable_64_types.h>
0007
0008 #ifndef __ASSEMBLY__
0009
0010 /*
0011  * This file contains the functions and defines necessary to modify and use
0012  * the x86-64 page table tree.
0013  */
0014 #include <asm/processor.h>
0015 #include <linux/bitops.h>
0016 #include <linux/threads.h>
0017 #include <asm/fixmap.h>
0018
0019 extern p4d_t level4_kernel_pgt[512];
0020 extern p4d_t level4_ident_pgt[512];
0021 extern pud_t level3_kernel_pgt[512];
0022 extern pud_t level3_ident_pgt[512];
0023 extern pmd_t level2_kernel_pgt[512];
0024 extern pmd_t level2_fixmap_pgt[512];
0025 extern pmd_t level2_ident_pgt[512];
0026 extern pte_t level1_fixmap_pgt[512 * FIXMAP_PMD_NUM];
0027 extern pgd_t init_top_pgt[];
0028
0029 #define swapper_pg_dir init_top_pgt
0030
0031 extern void paging_init(void);
0032 static inline void sync_initial_page_table(void) { }
0033
0034 #define pte_ERROR(e)                    \
0035     pr_err("%s:%d: bad pte %p(%016lx)\n",       \
0036            __FILE__, __LINE__, &(e), pte_val(e))
0037 #define pmd_ERROR(e)                    \
0038     pr_err("%s:%d: bad pmd %p(%016lx)\n",       \
0039            __FILE__, __LINE__, &(e), pmd_val(e))
0040 #define pud_ERROR(e)                    \
0041     pr_err("%s:%d: bad pud %p(%016lx)\n",       \
0042            __FILE__, __LINE__, &(e), pud_val(e))
0043
0044 #if CONFIG_PGTABLE_LEVELS >= 5
0045 #define p4d_ERROR(e)                    \
0046     pr_err("%s:%d: bad p4d %p(%016lx)\n",       \
0047            __FILE__, __LINE__, &(e), p4d_val(e))
0048 #endif
0049
0050 #define pgd_ERROR(e)                    \
0051     pr_err("%s:%d: bad pgd %p(%016lx)\n",       \
0052            __FILE__, __LINE__, &(e), pgd_val(e))
0053
0054 struct mm_struct;
0055
0056 #define mm_p4d_folded mm_p4d_folded
0057 static inline bool mm_p4d_folded(struct mm_struct *mm)
0058 {
0059     return !pgtable_l5_enabled();
0060 }
0061
0062 void set_pte_vaddr_p4d(p4d_t *p4d_page, unsigned long vaddr, pte_t new_pte);
0063 void set_pte_vaddr_pud(pud_t *pud_page, unsigned long vaddr, pte_t new_pte);
0064
0065 static inline void native_set_pte(pte_t *ptep, pte_t pte)
0066 {
0067     WRITE_ONCE(*ptep, pte);
0068 }
0069
0070 static inline void native_pte_clear(struct mm_struct *mm, unsigned long addr,
0071                     pte_t *ptep)
0072 {
0073     native_set_pte(ptep, native_make_pte(0));
0074 }
0075
0076 static inline void native_set_pte_atomic(pte_t *ptep, pte_t pte)
0077 {
0078     native_set_pte(ptep, pte);
0079 }
0080
0081 static inline void native_set_pmd(pmd_t *pmdp, pmd_t pmd)
0082 {
0083     WRITE_ONCE(*pmdp, pmd);
0084 }
0085
0086 static inline void native_pmd_clear(pmd_t *pmd)
0087 {
0088     native_set_pmd(pmd, native_make_pmd(0));
0089 }
0090
0091 static inline pte_t native_ptep_get_and_clear(pte_t *xp)
0092 {
0093 #ifdef CONFIG_SMP
0094     return native_make_pte(xchg(&xp->pte, 0));
0095 #else
0096     /* native_local_ptep_get_and_clear,
0097        but duplicated because of cyclic dependency */
0098     pte_t ret = *xp;
0099     native_pte_clear(NULL, 0, xp);
0100     return ret;
0101 #endif
0102 }
0103
0104 static inline pmd_t native_pmdp_get_and_clear(pmd_t *xp)
0105 {
0106 #ifdef CONFIG_SMP
0107     return native_make_pmd(xchg(&xp->pmd, 0));
0108 #else
0109     /* native_local_pmdp_get_and_clear,
0110        but duplicated because of cyclic dependency */
0111     pmd_t ret = *xp;
0112     native_pmd_clear(xp);
0113     return ret;
0114 #endif
0115 }
0116
0117 static inline void native_set_pud(pud_t *pudp, pud_t pud)
0118 {
0119     WRITE_ONCE(*pudp, pud);
0120 }
0121
0122 static inline void native_pud_clear(pud_t *pud)
0123 {
0124     native_set_pud(pud, native_make_pud(0));
0125 }
0126
0127 static inline pud_t native_pudp_get_and_clear(pud_t *xp)
0128 {
0129 #ifdef CONFIG_SMP
0130     return native_make_pud(xchg(&xp->pud, 0));
0131 #else
0132     /* native_local_pudp_get_and_clear,
0133      * but duplicated because of cyclic dependency
0134      */
0135     pud_t ret = *xp;
0136
0137     native_pud_clear(xp);
0138     return ret;
0139 #endif
0140 }
0141
0142 static inline void native_set_p4d(p4d_t *p4dp, p4d_t p4d)
0143 {
0144     pgd_t pgd;
0145
0146     if (pgtable_l5_enabled() || !IS_ENABLED(CONFIG_PAGE_TABLE_ISOLATION)) {
0147         WRITE_ONCE(*p4dp, p4d);
0148         return;
0149     }
0150
0151     pgd = native_make_pgd(native_p4d_val(p4d));
0152     pgd = pti_set_user_pgtbl((pgd_t *)p4dp, pgd);
0153     WRITE_ONCE(*p4dp, native_make_p4d(native_pgd_val(pgd)));
0154 }
0155
0156 static inline void native_p4d_clear(p4d_t *p4d)
0157 {
0158     native_set_p4d(p4d, native_make_p4d(0));
0159 }
0160
0161 static inline void native_set_pgd(pgd_t *pgdp, pgd_t pgd)
0162 {
0163     WRITE_ONCE(*pgdp, pti_set_user_pgtbl(pgdp, pgd));
0164 }
0165
0166 static inline void native_pgd_clear(pgd_t *pgd)
0167 {
0168     native_set_pgd(pgd, native_make_pgd(0));
0169 }
0170
0171 /*
0172  * Conversion functions: convert a page and protection to a page entry,
0173  * and a page entry and page directory to the page they refer to.
0174  */
0175
0176 /* PGD - Level 4 access */
0177
0178 /* PUD - Level 3 access */
0179
0180 /* PMD - Level 2 access */
0181
0182 /* PTE - Level 1 access */
0183
0184 /*
0185  * Encode and de-code a swap entry
0186  *
0187  * |     ...            | 11| 10|  9|8|7|6|5| 4| 3|2| 1|0| <- bit number
0188  * |     ...            |SW3|SW2|SW1|G|L|D|A|CD|WT|U| W|P| <- bit names
0189  * | TYPE (59-63) | ~OFFSET (9-58)  |0|0|X|X| X| E|F|SD|0| <- swp entry
0190  *
0191  * G (8) is aliased and used as a PROT_NONE indicator for
0192  * !present ptes.  We need to start storing swap entries above
0193  * there.  We also need to avoid using A and D because of an
0194  * erratum where they can be incorrectly set by hardware on
0195  * non-present PTEs.
0196  *
0197  * SD Bits 1-4 are not used in non-present format and available for
0198  * special use described below:
0199  *
0200  * SD (1) in swp entry is used to store soft dirty bit, which helps us
0201  * remember soft dirty over page migration
0202  *
0203  * F (2) in swp entry is used to record when a pagetable is
0204  * writeprotected by userfaultfd WP support.
0205  *
0206  * E (3) in swp entry is used to rememeber PG_anon_exclusive.
0207  *
0208  * Bit 7 in swp entry should be 0 because pmd_present checks not only P,
0209  * but also L and G.
0210  *
0211  * The offset is inverted by a binary not operation to make the high
0212  * physical bits set.
0213  */
0214 #define SWP_TYPE_BITS       5
0215
0216 #define SWP_OFFSET_FIRST_BIT    (_PAGE_BIT_PROTNONE + 1)
0217
0218 /* We always extract/encode the offset by shifting it all the way up, and then down again */
0219 #define SWP_OFFSET_SHIFT    (SWP_OFFSET_FIRST_BIT+SWP_TYPE_BITS)
0220
0221 #define MAX_SWAPFILES_CHECK() BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > SWP_TYPE_BITS)
0222
0223 /* Extract the high bits for type */
0224 #define __swp_type(x) ((x).val >> (64 - SWP_TYPE_BITS))
0225
0226 /* Shift up (to get rid of type), then down to get value */
0227 #define __swp_offset(x) (~(x).val << SWP_TYPE_BITS >> SWP_OFFSET_SHIFT)
0228
0229 /*
0230  * Shift the offset up "too far" by TYPE bits, then down again
0231  * The offset is inverted by a binary not operation to make the high
0232  * physical bits set.
0233  */
0234 #define __swp_entry(type, offset) ((swp_entry_t) { \
0235     (~(unsigned long)(offset) << SWP_OFFSET_SHIFT >> SWP_TYPE_BITS) \
0236     | ((unsigned long)(type) << (64-SWP_TYPE_BITS)) })
0237
0238 #define __pte_to_swp_entry(pte)     ((swp_entry_t) { pte_val((pte)) })
0239 #define __pmd_to_swp_entry(pmd)     ((swp_entry_t) { pmd_val((pmd)) })
0240 #define __swp_entry_to_pte(x)       ((pte_t) { .pte = (x).val })
0241 #define __swp_entry_to_pmd(x)       ((pmd_t) { .pmd = (x).val })
0242
0243 extern int kern_addr_valid(unsigned long addr);
0244 extern void cleanup_highmap(void);
0245
0246 #define HAVE_ARCH_UNMAPPED_AREA
0247 #define HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
0248
0249 #define PAGE_AGP    PAGE_KERNEL_NOCACHE
0250 #define HAVE_PAGE_AGP 1
0251
0252 /* fs/proc/kcore.c */
0253 #define kc_vaddr_to_offset(v) ((v) & __VIRTUAL_MASK)
0254 #define kc_offset_to_vaddr(o) ((o) | ~__VIRTUAL_MASK)
0255
0256 #define __HAVE_ARCH_PTE_SAME
0257
0258 #define vmemmap ((struct page *)VMEMMAP_START)
0259
0260 extern void init_extra_mapping_uc(unsigned long phys, unsigned long size);
0261 extern void init_extra_mapping_wb(unsigned long phys, unsigned long size);
0262
0263 #define gup_fast_permitted gup_fast_permitted
0264 static inline bool gup_fast_permitted(unsigned long start, unsigned long end)
0265 {
0266     if (end >> __VIRTUAL_MASK_SHIFT)
0267         return false;
0268     return true;
0269 }
0270
0271 #include <asm/pgtable-invert.h>
0272
0273 #endif /* !__ASSEMBLY__ */
0274 #endif /* _ASM_X86_PGTABLE_64_H */