Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 1 | #include <linux/mm.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 2 | #include <linux/gfp.h> |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 3 | #include <asm/pgalloc.h> |
Jeremy Fitzhardinge | ee5aa8d | 2008-03-17 16:37:03 -0700 | [diff] [blame] | 4 | #include <asm/pgtable.h> |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 5 | #include <asm/tlb.h> |
Ingo Molnar | a1d5a86 | 2008-06-20 15:34:46 +0200 | [diff] [blame] | 6 | #include <asm/fixmap.h> |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 7 | |
Vegard Nossum | 9e73023 | 2009-02-22 11:28:25 +0100 | [diff] [blame] | 8 | #define PGALLOC_GFP GFP_KERNEL | __GFP_NOTRACK | __GFP_REPEAT | __GFP_ZERO |
| 9 | |
Ian Campbell | 1431559 | 2010-02-17 10:38:10 +0000 | [diff] [blame] | 10 | #ifdef CONFIG_HIGHPTE |
| 11 | #define PGALLOC_USER_GFP __GFP_HIGHMEM |
| 12 | #else |
| 13 | #define PGALLOC_USER_GFP 0 |
| 14 | #endif |
| 15 | |
| 16 | gfp_t __userpte_alloc_gfp = PGALLOC_GFP | PGALLOC_USER_GFP; |
| 17 | |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 18 | pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address) |
| 19 | { |
Vegard Nossum | 9e73023 | 2009-02-22 11:28:25 +0100 | [diff] [blame] | 20 | return (pte_t *)__get_free_page(PGALLOC_GFP); |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 21 | } |
| 22 | |
| 23 | pgtable_t pte_alloc_one(struct mm_struct *mm, unsigned long address) |
| 24 | { |
| 25 | struct page *pte; |
| 26 | |
Ian Campbell | 1431559 | 2010-02-17 10:38:10 +0000 | [diff] [blame] | 27 | pte = alloc_pages(__userpte_alloc_gfp, 0); |
Kirill A. Shutemov | cecbd1b | 2013-11-14 14:31:47 -0800 | [diff] [blame] | 28 | if (!pte) |
| 29 | return NULL; |
| 30 | if (!pgtable_page_ctor(pte)) { |
| 31 | __free_page(pte); |
| 32 | return NULL; |
| 33 | } |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 34 | return pte; |
| 35 | } |
| 36 | |
Ian Campbell | 1431559 | 2010-02-17 10:38:10 +0000 | [diff] [blame] | 37 | static int __init setup_userpte(char *arg) |
| 38 | { |
| 39 | if (!arg) |
| 40 | return -EINVAL; |
| 41 | |
| 42 | /* |
| 43 | * "userpte=nohigh" disables allocation of user pagetables in |
| 44 | * high memory. |
| 45 | */ |
| 46 | if (strcmp(arg, "nohigh") == 0) |
| 47 | __userpte_alloc_gfp &= ~__GFP_HIGHMEM; |
| 48 | else |
| 49 | return -EINVAL; |
| 50 | return 0; |
| 51 | } |
| 52 | early_param("userpte", setup_userpte); |
| 53 | |
Benjamin Herrenschmidt | 9e1b32c | 2009-07-22 15:44:28 +1000 | [diff] [blame] | 54 | void ___pte_free_tlb(struct mmu_gather *tlb, struct page *pte) |
Jeremy Fitzhardinge | 397f687 | 2008-03-17 16:36:57 -0700 | [diff] [blame] | 55 | { |
| 56 | pgtable_page_dtor(pte); |
Jeremy Fitzhardinge | 6944a9c | 2008-03-17 16:37:01 -0700 | [diff] [blame] | 57 | paravirt_release_pte(page_to_pfn(pte)); |
Jeremy Fitzhardinge | 397f687 | 2008-03-17 16:36:57 -0700 | [diff] [blame] | 58 | tlb_remove_page(tlb, pte); |
| 59 | } |
| 60 | |
Jeremy Fitzhardinge | 170fdff | 2008-03-17 16:36:58 -0700 | [diff] [blame] | 61 | #if PAGETABLE_LEVELS > 2 |
Benjamin Herrenschmidt | 9e1b32c | 2009-07-22 15:44:28 +1000 | [diff] [blame] | 62 | void ___pmd_free_tlb(struct mmu_gather *tlb, pmd_t *pmd) |
Jeremy Fitzhardinge | 170fdff | 2008-03-17 16:36:58 -0700 | [diff] [blame] | 63 | { |
Kirill A. Shutemov | c283610 | 2013-11-21 14:32:09 -0800 | [diff] [blame] | 64 | struct page *page = virt_to_page(pmd); |
Jeremy Fitzhardinge | 6944a9c | 2008-03-17 16:37:01 -0700 | [diff] [blame] | 65 | paravirt_release_pmd(__pa(pmd) >> PAGE_SHIFT); |
Dave Hansen | 1de14c3 | 2013-04-12 16:23:54 -0700 | [diff] [blame] | 66 | /* |
| 67 | * NOTE! For PAE, any changes to the top page-directory-pointer-table |
| 68 | * entries need a full cr3 reload to flush. |
| 69 | */ |
| 70 | #ifdef CONFIG_X86_PAE |
| 71 | tlb->need_flush_all = 1; |
| 72 | #endif |
Kirill A. Shutemov | c283610 | 2013-11-21 14:32:09 -0800 | [diff] [blame] | 73 | pgtable_pmd_page_dtor(page); |
| 74 | tlb_remove_page(tlb, page); |
Jeremy Fitzhardinge | 170fdff | 2008-03-17 16:36:58 -0700 | [diff] [blame] | 75 | } |
Jeremy Fitzhardinge | 5a5f8f4 | 2008-03-17 16:36:59 -0700 | [diff] [blame] | 76 | |
| 77 | #if PAGETABLE_LEVELS > 3 |
Benjamin Herrenschmidt | 9e1b32c | 2009-07-22 15:44:28 +1000 | [diff] [blame] | 78 | void ___pud_free_tlb(struct mmu_gather *tlb, pud_t *pud) |
Jeremy Fitzhardinge | 5a5f8f4 | 2008-03-17 16:36:59 -0700 | [diff] [blame] | 79 | { |
Jeremy Fitzhardinge | 2761fa0 | 2008-03-17 16:37:02 -0700 | [diff] [blame] | 80 | paravirt_release_pud(__pa(pud) >> PAGE_SHIFT); |
Jeremy Fitzhardinge | 5a5f8f4 | 2008-03-17 16:36:59 -0700 | [diff] [blame] | 81 | tlb_remove_page(tlb, virt_to_page(pud)); |
| 82 | } |
| 83 | #endif /* PAGETABLE_LEVELS > 3 */ |
Jeremy Fitzhardinge | 170fdff | 2008-03-17 16:36:58 -0700 | [diff] [blame] | 84 | #endif /* PAGETABLE_LEVELS > 2 */ |
| 85 | |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 86 | static inline void pgd_list_add(pgd_t *pgd) |
| 87 | { |
| 88 | struct page *page = virt_to_page(pgd); |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 89 | |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 90 | list_add(&page->lru, &pgd_list); |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 91 | } |
| 92 | |
| 93 | static inline void pgd_list_del(pgd_t *pgd) |
| 94 | { |
| 95 | struct page *page = virt_to_page(pgd); |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 96 | |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 97 | list_del(&page->lru); |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 98 | } |
| 99 | |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 100 | #define UNSHARED_PTRS_PER_PGD \ |
Jeremy Fitzhardinge | 68db065 | 2008-03-17 16:37:13 -0700 | [diff] [blame] | 101 | (SHARED_KERNEL_PMD ? KERNEL_PGD_BOUNDARY : PTRS_PER_PGD) |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 102 | |
Jeremy Fitzhardinge | 617d34d | 2010-09-21 12:01:51 -0700 | [diff] [blame] | 103 | |
| 104 | static void pgd_set_mm(pgd_t *pgd, struct mm_struct *mm) |
| 105 | { |
| 106 | BUILD_BUG_ON(sizeof(virt_to_page(pgd)->index) < sizeof(mm)); |
| 107 | virt_to_page(pgd)->index = (pgoff_t)mm; |
| 108 | } |
| 109 | |
| 110 | struct mm_struct *pgd_page_get_mm(struct page *page) |
| 111 | { |
| 112 | return (struct mm_struct *)page->index; |
| 113 | } |
| 114 | |
| 115 | static void pgd_ctor(struct mm_struct *mm, pgd_t *pgd) |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 116 | { |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 117 | /* If the pgd points to a shared pagetable level (either the |
| 118 | ptes in non-PAE, or shared PMD in PAE), then just copy the |
| 119 | references from swapper_pg_dir. */ |
| 120 | if (PAGETABLE_LEVELS == 2 || |
Jeremy Fitzhardinge | 85958b4 | 2008-03-17 16:37:14 -0700 | [diff] [blame] | 121 | (PAGETABLE_LEVELS == 3 && SHARED_KERNEL_PMD) || |
| 122 | PAGETABLE_LEVELS == 4) { |
Jeremy Fitzhardinge | 68db065 | 2008-03-17 16:37:13 -0700 | [diff] [blame] | 123 | clone_pgd_range(pgd + KERNEL_PGD_BOUNDARY, |
| 124 | swapper_pg_dir + KERNEL_PGD_BOUNDARY, |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 125 | KERNEL_PGD_PTRS); |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 126 | } |
| 127 | |
| 128 | /* list required to sync kernel mapping updates */ |
Jeremy Fitzhardinge | 617d34d | 2010-09-21 12:01:51 -0700 | [diff] [blame] | 129 | if (!SHARED_KERNEL_PMD) { |
| 130 | pgd_set_mm(pgd, mm); |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 131 | pgd_list_add(pgd); |
Jeremy Fitzhardinge | 617d34d | 2010-09-21 12:01:51 -0700 | [diff] [blame] | 132 | } |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 133 | } |
| 134 | |
Jan Beulich | 17b7462 | 2008-08-29 12:51:32 +0100 | [diff] [blame] | 135 | static void pgd_dtor(pgd_t *pgd) |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 136 | { |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 137 | if (SHARED_KERNEL_PMD) |
| 138 | return; |
| 139 | |
Andrea Arcangeli | a79e53d | 2011-02-16 15:45:22 -0800 | [diff] [blame] | 140 | spin_lock(&pgd_lock); |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 141 | pgd_list_del(pgd); |
Andrea Arcangeli | a79e53d | 2011-02-16 15:45:22 -0800 | [diff] [blame] | 142 | spin_unlock(&pgd_lock); |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 143 | } |
| 144 | |
Jeremy Fitzhardinge | 85958b4 | 2008-03-17 16:37:14 -0700 | [diff] [blame] | 145 | /* |
| 146 | * List of all pgd's needed for non-PAE so it can invalidate entries |
| 147 | * in both cached and uncached pgd's; not needed for PAE since the |
| 148 | * kernel pmd is shared. If PAE were not to share the pmd a similar |
| 149 | * tactic would be needed. This is essentially codepath-based locking |
| 150 | * against pageattr.c; it is the unique case in which a valid change |
| 151 | * of kernel pagetables can't be lazily synchronized by vmalloc faults. |
| 152 | * vmalloc faults work because attached pagetables are never freed. |
Nadia Yvette Chambers | 6d49e35 | 2012-12-06 10:39:54 +0100 | [diff] [blame] | 153 | * -- nyc |
Jeremy Fitzhardinge | 85958b4 | 2008-03-17 16:37:14 -0700 | [diff] [blame] | 154 | */ |
| 155 | |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 156 | #ifdef CONFIG_X86_PAE |
| 157 | /* |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 158 | * In PAE mode, we need to do a cr3 reload (=tlb flush) when |
| 159 | * updating the top-level pagetable entries to guarantee the |
| 160 | * processor notices the update. Since this is expensive, and |
| 161 | * all 4 top-level entries are used almost immediately in a |
| 162 | * new process's life, we just pre-populate them here. |
| 163 | * |
| 164 | * Also, if we're in a paravirt environment where the kernel pmd is |
| 165 | * not shared between pagetables (!SHARED_KERNEL_PMDS), we allocate |
| 166 | * and initialize the kernel pmds here. |
| 167 | */ |
Jeremy Fitzhardinge | d8d5900 | 2008-06-25 00:19:13 -0400 | [diff] [blame] | 168 | #define PREALLOCATED_PMDS UNSHARED_PTRS_PER_PGD |
Ingo Molnar | 1ec1fe7 | 2008-03-19 20:30:40 +0100 | [diff] [blame] | 169 | |
| 170 | void pud_populate(struct mm_struct *mm, pud_t *pudp, pmd_t *pmd) |
| 171 | { |
Jeremy Fitzhardinge | 6944a9c | 2008-03-17 16:37:01 -0700 | [diff] [blame] | 172 | paravirt_alloc_pmd(mm, __pa(pmd) >> PAGE_SHIFT); |
Ingo Molnar | 1ec1fe7 | 2008-03-19 20:30:40 +0100 | [diff] [blame] | 173 | |
| 174 | /* Note: almost everything apart from _PAGE_PRESENT is |
| 175 | reserved at the pmd (PDPT) level. */ |
| 176 | set_pud(pudp, __pud(__pa(pmd) | _PAGE_PRESENT)); |
| 177 | |
| 178 | /* |
| 179 | * According to Intel App note "TLBs, Paging-Structure Caches, |
| 180 | * and Their Invalidation", April 2007, document 317080-001, |
| 181 | * section 8.1: in PAE mode we explicitly have to flush the |
| 182 | * TLB via cr3 if the top-level pgd is changed... |
| 183 | */ |
Shaohua Li | 4981d01 | 2011-03-16 11:37:29 +0800 | [diff] [blame] | 184 | flush_tlb_mm(mm); |
Ingo Molnar | 1ec1fe7 | 2008-03-19 20:30:40 +0100 | [diff] [blame] | 185 | } |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 186 | #else /* !CONFIG_X86_PAE */ |
Jeremy Fitzhardinge | d8d5900 | 2008-06-25 00:19:13 -0400 | [diff] [blame] | 187 | |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 188 | /* No need to prepopulate any pagetable entries in non-PAE modes. */ |
Jeremy Fitzhardinge | d8d5900 | 2008-06-25 00:19:13 -0400 | [diff] [blame] | 189 | #define PREALLOCATED_PMDS 0 |
| 190 | |
| 191 | #endif /* CONFIG_X86_PAE */ |
| 192 | |
| 193 | static void free_pmds(pmd_t *pmds[]) |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 194 | { |
Jeremy Fitzhardinge | d8d5900 | 2008-06-25 00:19:13 -0400 | [diff] [blame] | 195 | int i; |
| 196 | |
| 197 | for(i = 0; i < PREALLOCATED_PMDS; i++) |
Kirill A. Shutemov | 09ef493 | 2013-11-14 14:31:13 -0800 | [diff] [blame] | 198 | if (pmds[i]) { |
| 199 | pgtable_pmd_page_dtor(virt_to_page(pmds[i])); |
Jeremy Fitzhardinge | d8d5900 | 2008-06-25 00:19:13 -0400 | [diff] [blame] | 200 | free_page((unsigned long)pmds[i]); |
Kirill A. Shutemov | 09ef493 | 2013-11-14 14:31:13 -0800 | [diff] [blame] | 201 | } |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 202 | } |
| 203 | |
Jeremy Fitzhardinge | d8d5900 | 2008-06-25 00:19:13 -0400 | [diff] [blame] | 204 | static int preallocate_pmds(pmd_t *pmds[]) |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 205 | { |
Jeremy Fitzhardinge | d8d5900 | 2008-06-25 00:19:13 -0400 | [diff] [blame] | 206 | int i; |
| 207 | bool failed = false; |
| 208 | |
| 209 | for(i = 0; i < PREALLOCATED_PMDS; i++) { |
Vegard Nossum | 9e73023 | 2009-02-22 11:28:25 +0100 | [diff] [blame] | 210 | pmd_t *pmd = (pmd_t *)__get_free_page(PGALLOC_GFP); |
Kirill A. Shutemov | 09ef493 | 2013-11-14 14:31:13 -0800 | [diff] [blame] | 211 | if (!pmd) |
Jeremy Fitzhardinge | d8d5900 | 2008-06-25 00:19:13 -0400 | [diff] [blame] | 212 | failed = true; |
Kirill A. Shutemov | 09ef493 | 2013-11-14 14:31:13 -0800 | [diff] [blame] | 213 | if (pmd && !pgtable_pmd_page_ctor(virt_to_page(pmd))) { |
Al Viro | 2a46eed | 2013-11-20 22:16:36 +0000 | [diff] [blame] | 214 | free_page((unsigned long)pmd); |
Kirill A. Shutemov | 09ef493 | 2013-11-14 14:31:13 -0800 | [diff] [blame] | 215 | pmd = NULL; |
| 216 | failed = true; |
| 217 | } |
Jeremy Fitzhardinge | d8d5900 | 2008-06-25 00:19:13 -0400 | [diff] [blame] | 218 | pmds[i] = pmd; |
| 219 | } |
| 220 | |
| 221 | if (failed) { |
| 222 | free_pmds(pmds); |
| 223 | return -ENOMEM; |
| 224 | } |
| 225 | |
| 226 | return 0; |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 227 | } |
Jeremy Fitzhardinge | d8d5900 | 2008-06-25 00:19:13 -0400 | [diff] [blame] | 228 | |
| 229 | /* |
| 230 | * Mop up any pmd pages which may still be attached to the pgd. |
| 231 | * Normally they will be freed by munmap/exit_mmap, but any pmd we |
| 232 | * preallocate which never got a corresponding vma will need to be |
| 233 | * freed manually. |
| 234 | */ |
| 235 | static void pgd_mop_up_pmds(struct mm_struct *mm, pgd_t *pgdp) |
| 236 | { |
| 237 | int i; |
| 238 | |
| 239 | for(i = 0; i < PREALLOCATED_PMDS; i++) { |
| 240 | pgd_t pgd = pgdp[i]; |
| 241 | |
| 242 | if (pgd_val(pgd) != 0) { |
| 243 | pmd_t *pmd = (pmd_t *)pgd_page_vaddr(pgd); |
| 244 | |
| 245 | pgdp[i] = native_make_pgd(0); |
| 246 | |
| 247 | paravirt_release_pmd(pgd_val(pgd) >> PAGE_SHIFT); |
| 248 | pmd_free(mm, pmd); |
| 249 | } |
| 250 | } |
| 251 | } |
| 252 | |
| 253 | static void pgd_prepopulate_pmd(struct mm_struct *mm, pgd_t *pgd, pmd_t *pmds[]) |
| 254 | { |
| 255 | pud_t *pud; |
Jeremy Fitzhardinge | d8d5900 | 2008-06-25 00:19:13 -0400 | [diff] [blame] | 256 | int i; |
| 257 | |
Jeremy Fitzhardinge | cf3e505 | 2008-08-08 13:46:07 -0700 | [diff] [blame] | 258 | if (PREALLOCATED_PMDS == 0) /* Work around gcc-3.4.x bug */ |
| 259 | return; |
| 260 | |
Jeremy Fitzhardinge | d8d5900 | 2008-06-25 00:19:13 -0400 | [diff] [blame] | 261 | pud = pud_offset(pgd, 0); |
| 262 | |
Wanpeng Li | 73b44ff | 2013-07-08 16:00:17 -0700 | [diff] [blame] | 263 | for (i = 0; i < PREALLOCATED_PMDS; i++, pud++) { |
Jeremy Fitzhardinge | d8d5900 | 2008-06-25 00:19:13 -0400 | [diff] [blame] | 264 | pmd_t *pmd = pmds[i]; |
| 265 | |
| 266 | if (i >= KERNEL_PGD_BOUNDARY) |
| 267 | memcpy(pmd, (pmd_t *)pgd_page_vaddr(swapper_pg_dir[i]), |
| 268 | sizeof(pmd_t) * PTRS_PER_PMD); |
| 269 | |
| 270 | pud_populate(mm, pud, pmd); |
| 271 | } |
| 272 | } |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 273 | |
Fenghua Yu | 1db491f | 2015-01-15 20:30:01 -0800 | [diff] [blame^] | 274 | /* |
| 275 | * Xen paravirt assumes pgd table should be in one page. 64 bit kernel also |
| 276 | * assumes that pgd should be in one page. |
| 277 | * |
| 278 | * But kernel with PAE paging that is not running as a Xen domain |
| 279 | * only needs to allocate 32 bytes for pgd instead of one page. |
| 280 | */ |
| 281 | #ifdef CONFIG_X86_PAE |
| 282 | |
| 283 | #include <linux/slab.h> |
| 284 | |
| 285 | #define PGD_SIZE (PTRS_PER_PGD * sizeof(pgd_t)) |
| 286 | #define PGD_ALIGN 32 |
| 287 | |
| 288 | static struct kmem_cache *pgd_cache; |
| 289 | |
| 290 | static int __init pgd_cache_init(void) |
| 291 | { |
| 292 | /* |
| 293 | * When PAE kernel is running as a Xen domain, it does not use |
| 294 | * shared kernel pmd. And this requires a whole page for pgd. |
| 295 | */ |
| 296 | if (!SHARED_KERNEL_PMD) |
| 297 | return 0; |
| 298 | |
| 299 | /* |
| 300 | * when PAE kernel is not running as a Xen domain, it uses |
| 301 | * shared kernel pmd. Shared kernel pmd does not require a whole |
| 302 | * page for pgd. We are able to just allocate a 32-byte for pgd. |
| 303 | * During boot time, we create a 32-byte slab for pgd table allocation. |
| 304 | */ |
| 305 | pgd_cache = kmem_cache_create("pgd_cache", PGD_SIZE, PGD_ALIGN, |
| 306 | SLAB_PANIC, NULL); |
| 307 | if (!pgd_cache) |
| 308 | return -ENOMEM; |
| 309 | |
| 310 | return 0; |
| 311 | } |
| 312 | core_initcall(pgd_cache_init); |
| 313 | |
| 314 | static inline pgd_t *_pgd_alloc(void) |
| 315 | { |
| 316 | /* |
| 317 | * If no SHARED_KERNEL_PMD, PAE kernel is running as a Xen domain. |
| 318 | * We allocate one page for pgd. |
| 319 | */ |
| 320 | if (!SHARED_KERNEL_PMD) |
| 321 | return (pgd_t *)__get_free_page(PGALLOC_GFP); |
| 322 | |
| 323 | /* |
| 324 | * Now PAE kernel is not running as a Xen domain. We can allocate |
| 325 | * a 32-byte slab for pgd to save memory space. |
| 326 | */ |
| 327 | return kmem_cache_alloc(pgd_cache, PGALLOC_GFP); |
| 328 | } |
| 329 | |
| 330 | static inline void _pgd_free(pgd_t *pgd) |
| 331 | { |
| 332 | if (!SHARED_KERNEL_PMD) |
| 333 | free_page((unsigned long)pgd); |
| 334 | else |
| 335 | kmem_cache_free(pgd_cache, pgd); |
| 336 | } |
| 337 | #else |
| 338 | static inline pgd_t *_pgd_alloc(void) |
| 339 | { |
| 340 | return (pgd_t *)__get_free_page(PGALLOC_GFP); |
| 341 | } |
| 342 | |
| 343 | static inline void _pgd_free(pgd_t *pgd) |
| 344 | { |
| 345 | free_page((unsigned long)pgd); |
| 346 | } |
| 347 | #endif /* CONFIG_X86_PAE */ |
| 348 | |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 349 | pgd_t *pgd_alloc(struct mm_struct *mm) |
| 350 | { |
Jeremy Fitzhardinge | d8d5900 | 2008-06-25 00:19:13 -0400 | [diff] [blame] | 351 | pgd_t *pgd; |
| 352 | pmd_t *pmds[PREALLOCATED_PMDS]; |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 353 | |
Fenghua Yu | 1db491f | 2015-01-15 20:30:01 -0800 | [diff] [blame^] | 354 | pgd = _pgd_alloc(); |
Jeremy Fitzhardinge | d8d5900 | 2008-06-25 00:19:13 -0400 | [diff] [blame] | 355 | |
| 356 | if (pgd == NULL) |
| 357 | goto out; |
| 358 | |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 359 | mm->pgd = pgd; |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 360 | |
Jeremy Fitzhardinge | d8d5900 | 2008-06-25 00:19:13 -0400 | [diff] [blame] | 361 | if (preallocate_pmds(pmds) != 0) |
| 362 | goto out_free_pgd; |
| 363 | |
| 364 | if (paravirt_pgd_alloc(mm) != 0) |
| 365 | goto out_free_pmds; |
| 366 | |
| 367 | /* |
| 368 | * Make sure that pre-populating the pmds is atomic with |
| 369 | * respect to anything walking the pgd_list, so that they |
| 370 | * never see a partially populated pgd. |
| 371 | */ |
Andrea Arcangeli | a79e53d | 2011-02-16 15:45:22 -0800 | [diff] [blame] | 372 | spin_lock(&pgd_lock); |
Jeremy Fitzhardinge | d8d5900 | 2008-06-25 00:19:13 -0400 | [diff] [blame] | 373 | |
Jeremy Fitzhardinge | 617d34d | 2010-09-21 12:01:51 -0700 | [diff] [blame] | 374 | pgd_ctor(mm, pgd); |
Jeremy Fitzhardinge | d8d5900 | 2008-06-25 00:19:13 -0400 | [diff] [blame] | 375 | pgd_prepopulate_pmd(mm, pgd, pmds); |
| 376 | |
Andrea Arcangeli | a79e53d | 2011-02-16 15:45:22 -0800 | [diff] [blame] | 377 | spin_unlock(&pgd_lock); |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 378 | |
| 379 | return pgd; |
Jeremy Fitzhardinge | d8d5900 | 2008-06-25 00:19:13 -0400 | [diff] [blame] | 380 | |
| 381 | out_free_pmds: |
| 382 | free_pmds(pmds); |
| 383 | out_free_pgd: |
Fenghua Yu | 1db491f | 2015-01-15 20:30:01 -0800 | [diff] [blame^] | 384 | _pgd_free(pgd); |
Jeremy Fitzhardinge | d8d5900 | 2008-06-25 00:19:13 -0400 | [diff] [blame] | 385 | out: |
| 386 | return NULL; |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 387 | } |
| 388 | |
| 389 | void pgd_free(struct mm_struct *mm, pgd_t *pgd) |
| 390 | { |
| 391 | pgd_mop_up_pmds(mm, pgd); |
| 392 | pgd_dtor(pgd); |
Jeremy Fitzhardinge | eba0045 | 2008-06-25 00:19:12 -0400 | [diff] [blame] | 393 | paravirt_pgd_free(mm, pgd); |
Fenghua Yu | 1db491f | 2015-01-15 20:30:01 -0800 | [diff] [blame^] | 394 | _pgd_free(pgd); |
Jeremy Fitzhardinge | 4f76cd3 | 2008-03-17 16:36:55 -0700 | [diff] [blame] | 395 | } |
Jeremy Fitzhardinge | ee5aa8d | 2008-03-17 16:37:03 -0700 | [diff] [blame] | 396 | |
Rik van Riel | 0f9a921 | 2012-11-06 09:54:47 +0000 | [diff] [blame] | 397 | /* |
| 398 | * Used to set accessed or dirty bits in the page table entries |
| 399 | * on other architectures. On x86, the accessed and dirty bits |
| 400 | * are tracked by hardware. However, do_wp_page calls this function |
| 401 | * to also make the pte writeable at the same time the dirty bit is |
| 402 | * set. In that case we do actually need to write the PTE. |
| 403 | */ |
Jeremy Fitzhardinge | ee5aa8d | 2008-03-17 16:37:03 -0700 | [diff] [blame] | 404 | int ptep_set_access_flags(struct vm_area_struct *vma, |
| 405 | unsigned long address, pte_t *ptep, |
| 406 | pte_t entry, int dirty) |
| 407 | { |
| 408 | int changed = !pte_same(*ptep, entry); |
| 409 | |
| 410 | if (changed && dirty) { |
| 411 | *ptep = entry; |
| 412 | pte_update_defer(vma->vm_mm, address, ptep); |
Jeremy Fitzhardinge | ee5aa8d | 2008-03-17 16:37:03 -0700 | [diff] [blame] | 413 | } |
| 414 | |
| 415 | return changed; |
| 416 | } |
Jeremy Fitzhardinge | f9fbf1a | 2008-03-17 16:37:04 -0700 | [diff] [blame] | 417 | |
Andrea Arcangeli | db3eb96f | 2011-01-13 15:46:41 -0800 | [diff] [blame] | 418 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
| 419 | int pmdp_set_access_flags(struct vm_area_struct *vma, |
| 420 | unsigned long address, pmd_t *pmdp, |
| 421 | pmd_t entry, int dirty) |
| 422 | { |
| 423 | int changed = !pmd_same(*pmdp, entry); |
| 424 | |
| 425 | VM_BUG_ON(address & ~HPAGE_PMD_MASK); |
| 426 | |
| 427 | if (changed && dirty) { |
| 428 | *pmdp = entry; |
| 429 | pmd_update_defer(vma->vm_mm, address, pmdp); |
Ingo Molnar | 5e4bf1a | 2012-11-20 13:02:51 +0100 | [diff] [blame] | 430 | /* |
| 431 | * We had a write-protection fault here and changed the pmd |
| 432 | * to to more permissive. No need to flush the TLB for that, |
| 433 | * #PF is architecturally guaranteed to do that and in the |
| 434 | * worst-case we'll generate a spurious fault. |
| 435 | */ |
Andrea Arcangeli | db3eb96f | 2011-01-13 15:46:41 -0800 | [diff] [blame] | 436 | } |
| 437 | |
| 438 | return changed; |
| 439 | } |
| 440 | #endif |
| 441 | |
Jeremy Fitzhardinge | f9fbf1a | 2008-03-17 16:37:04 -0700 | [diff] [blame] | 442 | int ptep_test_and_clear_young(struct vm_area_struct *vma, |
| 443 | unsigned long addr, pte_t *ptep) |
| 444 | { |
| 445 | int ret = 0; |
| 446 | |
| 447 | if (pte_young(*ptep)) |
| 448 | ret = test_and_clear_bit(_PAGE_BIT_ACCESSED, |
Thomas Gleixner | 48e2395 | 2008-05-24 17:24:34 +0200 | [diff] [blame] | 449 | (unsigned long *) &ptep->pte); |
Jeremy Fitzhardinge | f9fbf1a | 2008-03-17 16:37:04 -0700 | [diff] [blame] | 450 | |
| 451 | if (ret) |
| 452 | pte_update(vma->vm_mm, addr, ptep); |
| 453 | |
| 454 | return ret; |
| 455 | } |
Jeremy Fitzhardinge | c20311e | 2008-03-17 16:37:05 -0700 | [diff] [blame] | 456 | |
Andrea Arcangeli | db3eb96f | 2011-01-13 15:46:41 -0800 | [diff] [blame] | 457 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
| 458 | int pmdp_test_and_clear_young(struct vm_area_struct *vma, |
| 459 | unsigned long addr, pmd_t *pmdp) |
| 460 | { |
| 461 | int ret = 0; |
| 462 | |
| 463 | if (pmd_young(*pmdp)) |
| 464 | ret = test_and_clear_bit(_PAGE_BIT_ACCESSED, |
Johannes Weiner | f2d6bfe | 2011-01-13 15:47:01 -0800 | [diff] [blame] | 465 | (unsigned long *)pmdp); |
Andrea Arcangeli | db3eb96f | 2011-01-13 15:46:41 -0800 | [diff] [blame] | 466 | |
| 467 | if (ret) |
| 468 | pmd_update(vma->vm_mm, addr, pmdp); |
| 469 | |
| 470 | return ret; |
| 471 | } |
| 472 | #endif |
| 473 | |
Jeremy Fitzhardinge | c20311e | 2008-03-17 16:37:05 -0700 | [diff] [blame] | 474 | int ptep_clear_flush_young(struct vm_area_struct *vma, |
| 475 | unsigned long address, pte_t *ptep) |
| 476 | { |
Shaohua Li | b13b1d2 | 2014-04-08 15:58:09 +0800 | [diff] [blame] | 477 | /* |
| 478 | * On x86 CPUs, clearing the accessed bit without a TLB flush |
| 479 | * doesn't cause data corruption. [ It could cause incorrect |
| 480 | * page aging and the (mistaken) reclaim of hot pages, but the |
| 481 | * chance of that should be relatively low. ] |
| 482 | * |
| 483 | * So as a performance optimization don't flush the TLB when |
| 484 | * clearing the accessed bit, it will eventually be flushed by |
| 485 | * a context switch or a VM operation anyway. [ In the rare |
| 486 | * event of it not getting flushed for a long time the delay |
| 487 | * shouldn't really matter because there's no real memory |
| 488 | * pressure for swapout to react to. ] |
| 489 | */ |
| 490 | return ptep_test_and_clear_young(vma, address, ptep); |
Jeremy Fitzhardinge | c20311e | 2008-03-17 16:37:05 -0700 | [diff] [blame] | 491 | } |
Jeremy Fitzhardinge | 7c7e6e0 | 2008-06-17 11:41:54 -0700 | [diff] [blame] | 492 | |
Andrea Arcangeli | db3eb96f | 2011-01-13 15:46:41 -0800 | [diff] [blame] | 493 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
| 494 | int pmdp_clear_flush_young(struct vm_area_struct *vma, |
| 495 | unsigned long address, pmd_t *pmdp) |
| 496 | { |
| 497 | int young; |
| 498 | |
| 499 | VM_BUG_ON(address & ~HPAGE_PMD_MASK); |
| 500 | |
| 501 | young = pmdp_test_and_clear_young(vma, address, pmdp); |
| 502 | if (young) |
| 503 | flush_tlb_range(vma, address, address + HPAGE_PMD_SIZE); |
| 504 | |
| 505 | return young; |
| 506 | } |
| 507 | |
| 508 | void pmdp_splitting_flush(struct vm_area_struct *vma, |
| 509 | unsigned long address, pmd_t *pmdp) |
| 510 | { |
| 511 | int set; |
| 512 | VM_BUG_ON(address & ~HPAGE_PMD_MASK); |
| 513 | set = !test_and_set_bit(_PAGE_BIT_SPLITTING, |
Johannes Weiner | f2d6bfe | 2011-01-13 15:47:01 -0800 | [diff] [blame] | 514 | (unsigned long *)pmdp); |
Andrea Arcangeli | db3eb96f | 2011-01-13 15:46:41 -0800 | [diff] [blame] | 515 | if (set) { |
| 516 | pmd_update(vma->vm_mm, address, pmdp); |
| 517 | /* need tlb flush only to serialize against gup-fast */ |
| 518 | flush_tlb_range(vma, address, address + HPAGE_PMD_SIZE); |
| 519 | } |
| 520 | } |
| 521 | #endif |
| 522 | |
Gustavo F. Padovan | fd862dd | 2009-02-15 21:48:54 -0300 | [diff] [blame] | 523 | /** |
| 524 | * reserve_top_address - reserves a hole in the top of kernel address space |
| 525 | * @reserve - size of hole to reserve |
| 526 | * |
| 527 | * Can be used to relocate the fixmap area and poke a hole in the top |
| 528 | * of kernel address space to make room for a hypervisor. |
| 529 | */ |
| 530 | void __init reserve_top_address(unsigned long reserve) |
| 531 | { |
| 532 | #ifdef CONFIG_X86_32 |
| 533 | BUG_ON(fixmaps_set > 0); |
Andy Lutomirski | 73159fd | 2014-05-05 12:19:31 -0700 | [diff] [blame] | 534 | __FIXADDR_TOP = round_down(-reserve, 1 << PMD_SHIFT) - PAGE_SIZE; |
| 535 | printk(KERN_INFO "Reserving virtual address space above 0x%08lx (rounded to 0x%08lx)\n", |
| 536 | -reserve, __FIXADDR_TOP + PAGE_SIZE); |
Gustavo F. Padovan | fd862dd | 2009-02-15 21:48:54 -0300 | [diff] [blame] | 537 | #endif |
| 538 | } |
| 539 | |
Jeremy Fitzhardinge | 7c7e6e0 | 2008-06-17 11:41:54 -0700 | [diff] [blame] | 540 | int fixmaps_set; |
| 541 | |
Jeremy Fitzhardinge | aeaaa59 | 2008-06-17 11:42:01 -0700 | [diff] [blame] | 542 | void __native_set_fixmap(enum fixed_addresses idx, pte_t pte) |
Jeremy Fitzhardinge | 7c7e6e0 | 2008-06-17 11:41:54 -0700 | [diff] [blame] | 543 | { |
| 544 | unsigned long address = __fix_to_virt(idx); |
| 545 | |
| 546 | if (idx >= __end_of_fixed_addresses) { |
| 547 | BUG(); |
| 548 | return; |
| 549 | } |
Jeremy Fitzhardinge | aeaaa59 | 2008-06-17 11:42:01 -0700 | [diff] [blame] | 550 | set_pte_vaddr(address, pte); |
Jeremy Fitzhardinge | 7c7e6e0 | 2008-06-17 11:41:54 -0700 | [diff] [blame] | 551 | fixmaps_set++; |
| 552 | } |
Jeremy Fitzhardinge | aeaaa59 | 2008-06-17 11:42:01 -0700 | [diff] [blame] | 553 | |
Masami Hiramatsu | 3b3809a | 2009-04-09 10:55:33 -0700 | [diff] [blame] | 554 | void native_set_fixmap(enum fixed_addresses idx, phys_addr_t phys, |
| 555 | pgprot_t flags) |
Jeremy Fitzhardinge | aeaaa59 | 2008-06-17 11:42:01 -0700 | [diff] [blame] | 556 | { |
| 557 | __native_set_fixmap(idx, pfn_pte(phys >> PAGE_SHIFT, flags)); |
| 558 | } |