1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * IBM System z Huge TLB Page Support for Kernel. 4 * 5 * Copyright IBM Corp. 2007,2020 6 * Author(s): Gerald Schaefer <gerald.schaefer@de.ibm.com> 7 */ 8 9 #define KMSG_COMPONENT "hugetlb" 10 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt 11 12 #include <asm/pgalloc.h> 13 #include <linux/mm.h> 14 #include <linux/hugetlb.h> 15 #include <linux/mman.h> 16 #include <linux/sched/mm.h> 17 #include <linux/security.h> 18 19 /* 20 * If the bit selected by single-bit bitmask "a" is set within "x", move 21 * it to the position indicated by single-bit bitmask "b". 22 */ 23 #define move_set_bit(x, a, b) (((x) & (a)) >> ilog2(a) << ilog2(b)) 24 25 static inline unsigned long __pte_to_rste(pte_t pte) 26 { 27 unsigned long rste; 28 29 /* 30 * Convert encoding pte bits pmd / pud bits 31 * lIR.uswrdy.p dy..R...I...wr 32 * empty 010.000000.0 -> 00..0...1...00 33 * prot-none, clean, old 111.000000.1 -> 00..1...1...00 34 * prot-none, clean, young 111.000001.1 -> 01..1...1...00 35 * prot-none, dirty, old 111.000010.1 -> 10..1...1...00 36 * prot-none, dirty, young 111.000011.1 -> 11..1...1...00 37 * read-only, clean, old 111.000100.1 -> 00..1...1...01 38 * read-only, clean, young 101.000101.1 -> 01..1...0...01 39 * read-only, dirty, old 111.000110.1 -> 10..1...1...01 40 * read-only, dirty, young 101.000111.1 -> 11..1...0...01 41 * read-write, clean, old 111.001100.1 -> 00..1...1...11 42 * read-write, clean, young 101.001101.1 -> 01..1...0...11 43 * read-write, dirty, old 110.001110.1 -> 10..0...1...11 44 * read-write, dirty, young 100.001111.1 -> 11..0...0...11 45 * HW-bits: R read-only, I invalid 46 * SW-bits: p present, y young, d dirty, r read, w write, s special, 47 * u unused, l large 48 */ 49 if (pte_present(pte)) { 50 rste = pte_val(pte) & PAGE_MASK; 51 rste |= move_set_bit(pte_val(pte), _PAGE_READ, 52 _SEGMENT_ENTRY_READ); 53 rste |= move_set_bit(pte_val(pte), _PAGE_WRITE, 54 _SEGMENT_ENTRY_WRITE); 55 rste |= move_set_bit(pte_val(pte), _PAGE_INVALID, 56 _SEGMENT_ENTRY_INVALID); 57 rste |= move_set_bit(pte_val(pte), _PAGE_PROTECT, 58 _SEGMENT_ENTRY_PROTECT); 59 rste |= move_set_bit(pte_val(pte), _PAGE_DIRTY, 60 _SEGMENT_ENTRY_DIRTY); 61 rste |= move_set_bit(pte_val(pte), _PAGE_YOUNG, 62 _SEGMENT_ENTRY_YOUNG); 63 #ifdef CONFIG_MEM_SOFT_DIRTY 64 rste |= move_set_bit(pte_val(pte), _PAGE_SOFT_DIRTY, 65 _SEGMENT_ENTRY_SOFT_DIRTY); 66 #endif 67 rste |= move_set_bit(pte_val(pte), _PAGE_NOEXEC, 68 _SEGMENT_ENTRY_NOEXEC); 69 } else 70 rste = _SEGMENT_ENTRY_EMPTY; 71 return rste; 72 } 73 74 static inline pte_t __rste_to_pte(unsigned long rste) 75 { 76 int present; 77 pte_t pte; 78 79 if ((rste & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3) 80 present = pud_present(__pud(rste)); 81 else 82 present = pmd_present(__pmd(rste)); 83 84 /* 85 * Convert encoding pmd / pud bits pte bits 86 * dy..R...I...wr lIR.uswrdy.p 87 * empty 00..0...1...00 -> 010.000000.0 88 * prot-none, clean, old 00..1...1...00 -> 111.000000.1 89 * prot-none, clean, young 01..1...1...00 -> 111.000001.1 90 * prot-none, dirty, old 10..1...1...00 -> 111.000010.1 91 * prot-none, dirty, young 11..1...1...00 -> 111.000011.1 92 * read-only, clean, old 00..1...1...01 -> 111.000100.1 93 * read-only, clean, young 01..1...0...01 -> 101.000101.1 94 * read-only, dirty, old 10..1...1...01 -> 111.000110.1 95 * read-only, dirty, young 11..1...0...01 -> 101.000111.1 96 * read-write, clean, old 00..1...1...11 -> 111.001100.1 97 * read-write, clean, young 01..1...0...11 -> 101.001101.1 98 * read-write, dirty, old 10..0...1...11 -> 110.001110.1 99 * read-write, dirty, young 11..0...0...11 -> 100.001111.1 100 * HW-bits: R read-only, I invalid 101 * SW-bits: p present, y young, d dirty, r read, w write, s special, 102 * u unused, l large 103 */ 104 if (present) { 105 pte_val(pte) = rste & _SEGMENT_ENTRY_ORIGIN_LARGE; 106 pte_val(pte) |= _PAGE_LARGE | _PAGE_PRESENT; 107 pte_val(pte) |= move_set_bit(rste, _SEGMENT_ENTRY_READ, 108 _PAGE_READ); 109 pte_val(pte) |= move_set_bit(rste, _SEGMENT_ENTRY_WRITE, 110 _PAGE_WRITE); 111 pte_val(pte) |= move_set_bit(rste, _SEGMENT_ENTRY_INVALID, 112 _PAGE_INVALID); 113 pte_val(pte) |= move_set_bit(rste, _SEGMENT_ENTRY_PROTECT, 114 _PAGE_PROTECT); 115 pte_val(pte) |= move_set_bit(rste, _SEGMENT_ENTRY_DIRTY, 116 _PAGE_DIRTY); 117 pte_val(pte) |= move_set_bit(rste, _SEGMENT_ENTRY_YOUNG, 118 _PAGE_YOUNG); 119 #ifdef CONFIG_MEM_SOFT_DIRTY 120 pte_val(pte) |= move_set_bit(rste, _SEGMENT_ENTRY_SOFT_DIRTY, 121 _PAGE_SOFT_DIRTY); 122 #endif 123 pte_val(pte) |= move_set_bit(rste, _SEGMENT_ENTRY_NOEXEC, 124 _PAGE_NOEXEC); 125 } else 126 pte_val(pte) = _PAGE_INVALID; 127 return pte; 128 } 129 130 static void clear_huge_pte_skeys(struct mm_struct *mm, unsigned long rste) 131 { 132 struct page *page; 133 unsigned long size, paddr; 134 135 if (!mm_uses_skeys(mm) || 136 rste & _SEGMENT_ENTRY_INVALID) 137 return; 138 139 if ((rste & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3) { 140 page = pud_page(__pud(rste)); 141 size = PUD_SIZE; 142 paddr = rste & PUD_MASK; 143 } else { 144 page = pmd_page(__pmd(rste)); 145 size = PMD_SIZE; 146 paddr = rste & PMD_MASK; 147 } 148 149 if (!test_and_set_bit(PG_arch_1, &page->flags)) 150 __storage_key_init_range(paddr, paddr + size - 1); 151 } 152 153 void set_huge_pte_at(struct mm_struct *mm, unsigned long addr, 154 pte_t *ptep, pte_t pte) 155 { 156 unsigned long rste; 157 158 rste = __pte_to_rste(pte); 159 if (!MACHINE_HAS_NX) 160 rste &= ~_SEGMENT_ENTRY_NOEXEC; 161 162 /* Set correct table type for 2G hugepages */ 163 if ((pte_val(*ptep) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3) { 164 if (likely(pte_present(pte))) 165 rste |= _REGION3_ENTRY_LARGE; 166 rste |= _REGION_ENTRY_TYPE_R3; 167 } else if (likely(pte_present(pte))) 168 rste |= _SEGMENT_ENTRY_LARGE; 169 170 clear_huge_pte_skeys(mm, rste); 171 pte_val(*ptep) = rste; 172 } 173 174 pte_t huge_ptep_get(pte_t *ptep) 175 { 176 return __rste_to_pte(pte_val(*ptep)); 177 } 178 179 pte_t huge_ptep_get_and_clear(struct mm_struct *mm, 180 unsigned long addr, pte_t *ptep) 181 { 182 pte_t pte = huge_ptep_get(ptep); 183 pmd_t *pmdp = (pmd_t *) ptep; 184 pud_t *pudp = (pud_t *) ptep; 185 186 if ((pte_val(*ptep) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3) 187 pudp_xchg_direct(mm, addr, pudp, __pud(_REGION3_ENTRY_EMPTY)); 188 else 189 pmdp_xchg_direct(mm, addr, pmdp, __pmd(_SEGMENT_ENTRY_EMPTY)); 190 return pte; 191 } 192 193 pte_t *huge_pte_alloc(struct mm_struct *mm, struct vm_area_struct *vma, 194 unsigned long addr, unsigned long sz) 195 { 196 pgd_t *pgdp; 197 p4d_t *p4dp; 198 pud_t *pudp; 199 pmd_t *pmdp = NULL; 200 201 pgdp = pgd_offset(mm, addr); 202 p4dp = p4d_alloc(mm, pgdp, addr); 203 if (p4dp) { 204 pudp = pud_alloc(mm, p4dp, addr); 205 if (pudp) { 206 if (sz == PUD_SIZE) 207 return (pte_t *) pudp; 208 else if (sz == PMD_SIZE) 209 pmdp = pmd_alloc(mm, pudp, addr); 210 } 211 } 212 return (pte_t *) pmdp; 213 } 214 215 pte_t *huge_pte_offset(struct mm_struct *mm, 216 unsigned long addr, unsigned long sz) 217 { 218 pgd_t *pgdp; 219 p4d_t *p4dp; 220 pud_t *pudp; 221 pmd_t *pmdp = NULL; 222 223 pgdp = pgd_offset(mm, addr); 224 if (pgd_present(*pgdp)) { 225 p4dp = p4d_offset(pgdp, addr); 226 if (p4d_present(*p4dp)) { 227 pudp = pud_offset(p4dp, addr); 228 if (pud_present(*pudp)) { 229 if (pud_large(*pudp)) 230 return (pte_t *) pudp; 231 pmdp = pmd_offset(pudp, addr); 232 } 233 } 234 } 235 return (pte_t *) pmdp; 236 } 237 238 int pmd_huge(pmd_t pmd) 239 { 240 return pmd_large(pmd); 241 } 242 243 int pud_huge(pud_t pud) 244 { 245 return pud_large(pud); 246 } 247 248 struct page * 249 follow_huge_pud(struct mm_struct *mm, unsigned long address, 250 pud_t *pud, int flags) 251 { 252 if (flags & FOLL_GET) 253 return NULL; 254 255 return pud_page(*pud) + ((address & ~PUD_MASK) >> PAGE_SHIFT); 256 } 257 258 bool __init arch_hugetlb_valid_size(unsigned long size) 259 { 260 if (MACHINE_HAS_EDAT1 && size == PMD_SIZE) 261 return true; 262 else if (MACHINE_HAS_EDAT2 && size == PUD_SIZE) 263 return true; 264 else 265 return false; 266 } 267 268 static unsigned long hugetlb_get_unmapped_area_bottomup(struct file *file, 269 unsigned long addr, unsigned long len, 270 unsigned long pgoff, unsigned long flags) 271 { 272 struct hstate *h = hstate_file(file); 273 struct vm_unmapped_area_info info; 274 275 info.flags = 0; 276 info.length = len; 277 info.low_limit = current->mm->mmap_base; 278 info.high_limit = TASK_SIZE; 279 info.align_mask = PAGE_MASK & ~huge_page_mask(h); 280 info.align_offset = 0; 281 return vm_unmapped_area(&info); 282 } 283 284 static unsigned long hugetlb_get_unmapped_area_topdown(struct file *file, 285 unsigned long addr0, unsigned long len, 286 unsigned long pgoff, unsigned long flags) 287 { 288 struct hstate *h = hstate_file(file); 289 struct vm_unmapped_area_info info; 290 unsigned long addr; 291 292 info.flags = VM_UNMAPPED_AREA_TOPDOWN; 293 info.length = len; 294 info.low_limit = max(PAGE_SIZE, mmap_min_addr); 295 info.high_limit = current->mm->mmap_base; 296 info.align_mask = PAGE_MASK & ~huge_page_mask(h); 297 info.align_offset = 0; 298 addr = vm_unmapped_area(&info); 299 300 /* 301 * A failed mmap() very likely causes application failure, 302 * so fall back to the bottom-up function here. This scenario 303 * can happen with large stack limits and large mmap() 304 * allocations. 305 */ 306 if (addr & ~PAGE_MASK) { 307 VM_BUG_ON(addr != -ENOMEM); 308 info.flags = 0; 309 info.low_limit = TASK_UNMAPPED_BASE; 310 info.high_limit = TASK_SIZE; 311 addr = vm_unmapped_area(&info); 312 } 313 314 return addr; 315 } 316 317 unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr, 318 unsigned long len, unsigned long pgoff, unsigned long flags) 319 { 320 struct hstate *h = hstate_file(file); 321 struct mm_struct *mm = current->mm; 322 struct vm_area_struct *vma; 323 324 if (len & ~huge_page_mask(h)) 325 return -EINVAL; 326 if (len > TASK_SIZE - mmap_min_addr) 327 return -ENOMEM; 328 329 if (flags & MAP_FIXED) { 330 if (prepare_hugepage_range(file, addr, len)) 331 return -EINVAL; 332 goto check_asce_limit; 333 } 334 335 if (addr) { 336 addr = ALIGN(addr, huge_page_size(h)); 337 vma = find_vma(mm, addr); 338 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr && 339 (!vma || addr + len <= vm_start_gap(vma))) 340 goto check_asce_limit; 341 } 342 343 if (mm->get_unmapped_area == arch_get_unmapped_area) 344 addr = hugetlb_get_unmapped_area_bottomup(file, addr, len, 345 pgoff, flags); 346 else 347 addr = hugetlb_get_unmapped_area_topdown(file, addr, len, 348 pgoff, flags); 349 if (offset_in_page(addr)) 350 return addr; 351 352 check_asce_limit: 353 return check_asce_limit(mm, addr, len); 354 } 355