1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * This kernel test validates architecture page table helpers and 4 * accessors and helps in verifying their continued compliance with 5 * expected generic MM semantics. 6 * 7 * Copyright (C) 2019 ARM Ltd. 8 * 9 * Author: Anshuman Khandual <anshuman.khandual@arm.com> 10 */ 11 #define pr_fmt(fmt) "debug_vm_pgtable: [%-25s]: " fmt, __func__ 12 13 #include <linux/gfp.h> 14 #include <linux/highmem.h> 15 #include <linux/hugetlb.h> 16 #include <linux/kernel.h> 17 #include <linux/kconfig.h> 18 #include <linux/mm.h> 19 #include <linux/mman.h> 20 #include <linux/mm_types.h> 21 #include <linux/module.h> 22 #include <linux/pfn_t.h> 23 #include <linux/printk.h> 24 #include <linux/pgtable.h> 25 #include <linux/random.h> 26 #include <linux/spinlock.h> 27 #include <linux/swap.h> 28 #include <linux/swapops.h> 29 #include <linux/start_kernel.h> 30 #include <linux/sched/mm.h> 31 #include <linux/io.h> 32 #include <asm/pgalloc.h> 33 #include <asm/tlbflush.h> 34 35 /* 36 * Please refer Documentation/vm/arch_pgtable_helpers.rst for the semantics 37 * expectations that are being validated here. All future changes in here 38 * or the documentation need to be in sync. 39 */ 40 41 #define VMFLAGS (VM_READ|VM_WRITE|VM_EXEC) 42 43 /* 44 * On s390 platform, the lower 4 bits are used to identify given page table 45 * entry type. But these bits might affect the ability to clear entries with 46 * pxx_clear() because of how dynamic page table folding works on s390. So 47 * while loading up the entries do not change the lower 4 bits. It does not 48 * have affect any other platform. Also avoid the 62nd bit on ppc64 that is 49 * used to mark a pte entry. 50 */ 51 #define S390_SKIP_MASK GENMASK(3, 0) 52 #if __BITS_PER_LONG == 64 53 #define PPC64_SKIP_MASK GENMASK(62, 62) 54 #else 55 #define PPC64_SKIP_MASK 0x0 56 #endif 57 #define ARCH_SKIP_MASK (S390_SKIP_MASK | PPC64_SKIP_MASK) 58 #define RANDOM_ORVALUE (GENMASK(BITS_PER_LONG - 1, 0) & ~ARCH_SKIP_MASK) 59 #define RANDOM_NZVALUE GENMASK(7, 0) 60 61 static void __init pte_basic_tests(unsigned long pfn, int idx) 62 { 63 pgprot_t prot = protection_map[idx]; 64 pte_t pte = pfn_pte(pfn, prot); 65 unsigned long val = idx, *ptr = &val; 66 67 pr_debug("Validating PTE basic (%pGv)\n", ptr); 68 69 /* 70 * This test needs to be executed after the given page table entry 71 * is created with pfn_pte() to make sure that protection_map[idx] 72 * does not have the dirty bit enabled from the beginning. This is 73 * important for platforms like arm64 where (!PTE_RDONLY) indicate 74 * dirty bit being set. 75 */ 76 WARN_ON(pte_dirty(pte_wrprotect(pte))); 77 78 WARN_ON(!pte_same(pte, pte)); 79 WARN_ON(!pte_young(pte_mkyoung(pte_mkold(pte)))); 80 WARN_ON(!pte_dirty(pte_mkdirty(pte_mkclean(pte)))); 81 WARN_ON(!pte_write(pte_mkwrite(pte_wrprotect(pte)))); 82 WARN_ON(pte_young(pte_mkold(pte_mkyoung(pte)))); 83 WARN_ON(pte_dirty(pte_mkclean(pte_mkdirty(pte)))); 84 WARN_ON(pte_write(pte_wrprotect(pte_mkwrite(pte)))); 85 WARN_ON(pte_dirty(pte_wrprotect(pte_mkclean(pte)))); 86 WARN_ON(!pte_dirty(pte_wrprotect(pte_mkdirty(pte)))); 87 } 88 89 static void __init pte_advanced_tests(struct mm_struct *mm, 90 struct vm_area_struct *vma, pte_t *ptep, 91 unsigned long pfn, unsigned long vaddr, 92 pgprot_t prot) 93 { 94 pte_t pte; 95 96 /* 97 * Architectures optimize set_pte_at by avoiding TLB flush. 98 * This requires set_pte_at to be not used to update an 99 * existing pte entry. Clear pte before we do set_pte_at 100 */ 101 102 pr_debug("Validating PTE advanced\n"); 103 pte = pfn_pte(pfn, prot); 104 set_pte_at(mm, vaddr, ptep, pte); 105 ptep_set_wrprotect(mm, vaddr, ptep); 106 pte = ptep_get(ptep); 107 WARN_ON(pte_write(pte)); 108 ptep_get_and_clear(mm, vaddr, ptep); 109 pte = ptep_get(ptep); 110 WARN_ON(!pte_none(pte)); 111 112 pte = pfn_pte(pfn, prot); 113 pte = pte_wrprotect(pte); 114 pte = pte_mkclean(pte); 115 set_pte_at(mm, vaddr, ptep, pte); 116 pte = pte_mkwrite(pte); 117 pte = pte_mkdirty(pte); 118 ptep_set_access_flags(vma, vaddr, ptep, pte, 1); 119 pte = ptep_get(ptep); 120 WARN_ON(!(pte_write(pte) && pte_dirty(pte))); 121 ptep_get_and_clear_full(mm, vaddr, ptep, 1); 122 pte = ptep_get(ptep); 123 WARN_ON(!pte_none(pte)); 124 125 pte = pfn_pte(pfn, prot); 126 pte = pte_mkyoung(pte); 127 set_pte_at(mm, vaddr, ptep, pte); 128 ptep_test_and_clear_young(vma, vaddr, ptep); 129 pte = ptep_get(ptep); 130 WARN_ON(pte_young(pte)); 131 } 132 133 static void __init pte_savedwrite_tests(unsigned long pfn, pgprot_t prot) 134 { 135 pte_t pte = pfn_pte(pfn, prot); 136 137 if (!IS_ENABLED(CONFIG_NUMA_BALANCING)) 138 return; 139 140 pr_debug("Validating PTE saved write\n"); 141 WARN_ON(!pte_savedwrite(pte_mk_savedwrite(pte_clear_savedwrite(pte)))); 142 WARN_ON(pte_savedwrite(pte_clear_savedwrite(pte_mk_savedwrite(pte)))); 143 } 144 145 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 146 static void __init pmd_basic_tests(unsigned long pfn, int idx) 147 { 148 pgprot_t prot = protection_map[idx]; 149 unsigned long val = idx, *ptr = &val; 150 pmd_t pmd; 151 152 if (!has_transparent_hugepage()) 153 return; 154 155 pr_debug("Validating PMD basic (%pGv)\n", ptr); 156 pmd = pfn_pmd(pfn, prot); 157 158 /* 159 * This test needs to be executed after the given page table entry 160 * is created with pfn_pmd() to make sure that protection_map[idx] 161 * does not have the dirty bit enabled from the beginning. This is 162 * important for platforms like arm64 where (!PTE_RDONLY) indicate 163 * dirty bit being set. 164 */ 165 WARN_ON(pmd_dirty(pmd_wrprotect(pmd))); 166 167 168 WARN_ON(!pmd_same(pmd, pmd)); 169 WARN_ON(!pmd_young(pmd_mkyoung(pmd_mkold(pmd)))); 170 WARN_ON(!pmd_dirty(pmd_mkdirty(pmd_mkclean(pmd)))); 171 WARN_ON(!pmd_write(pmd_mkwrite(pmd_wrprotect(pmd)))); 172 WARN_ON(pmd_young(pmd_mkold(pmd_mkyoung(pmd)))); 173 WARN_ON(pmd_dirty(pmd_mkclean(pmd_mkdirty(pmd)))); 174 WARN_ON(pmd_write(pmd_wrprotect(pmd_mkwrite(pmd)))); 175 WARN_ON(pmd_dirty(pmd_wrprotect(pmd_mkclean(pmd)))); 176 WARN_ON(!pmd_dirty(pmd_wrprotect(pmd_mkdirty(pmd)))); 177 /* 178 * A huge page does not point to next level page table 179 * entry. Hence this must qualify as pmd_bad(). 180 */ 181 WARN_ON(!pmd_bad(pmd_mkhuge(pmd))); 182 } 183 184 static void __init pmd_advanced_tests(struct mm_struct *mm, 185 struct vm_area_struct *vma, pmd_t *pmdp, 186 unsigned long pfn, unsigned long vaddr, 187 pgprot_t prot, pgtable_t pgtable) 188 { 189 pmd_t pmd; 190 191 if (!has_transparent_hugepage()) 192 return; 193 194 pr_debug("Validating PMD advanced\n"); 195 /* Align the address wrt HPAGE_PMD_SIZE */ 196 vaddr &= HPAGE_PMD_MASK; 197 198 pgtable_trans_huge_deposit(mm, pmdp, pgtable); 199 200 pmd = pfn_pmd(pfn, prot); 201 set_pmd_at(mm, vaddr, pmdp, pmd); 202 pmdp_set_wrprotect(mm, vaddr, pmdp); 203 pmd = READ_ONCE(*pmdp); 204 WARN_ON(pmd_write(pmd)); 205 pmdp_huge_get_and_clear(mm, vaddr, pmdp); 206 pmd = READ_ONCE(*pmdp); 207 WARN_ON(!pmd_none(pmd)); 208 209 pmd = pfn_pmd(pfn, prot); 210 pmd = pmd_wrprotect(pmd); 211 pmd = pmd_mkclean(pmd); 212 set_pmd_at(mm, vaddr, pmdp, pmd); 213 pmd = pmd_mkwrite(pmd); 214 pmd = pmd_mkdirty(pmd); 215 pmdp_set_access_flags(vma, vaddr, pmdp, pmd, 1); 216 pmd = READ_ONCE(*pmdp); 217 WARN_ON(!(pmd_write(pmd) && pmd_dirty(pmd))); 218 pmdp_huge_get_and_clear_full(vma, vaddr, pmdp, 1); 219 pmd = READ_ONCE(*pmdp); 220 WARN_ON(!pmd_none(pmd)); 221 222 pmd = pmd_mkhuge(pfn_pmd(pfn, prot)); 223 pmd = pmd_mkyoung(pmd); 224 set_pmd_at(mm, vaddr, pmdp, pmd); 225 pmdp_test_and_clear_young(vma, vaddr, pmdp); 226 pmd = READ_ONCE(*pmdp); 227 WARN_ON(pmd_young(pmd)); 228 229 /* Clear the pte entries */ 230 pmdp_huge_get_and_clear(mm, vaddr, pmdp); 231 pgtable = pgtable_trans_huge_withdraw(mm, pmdp); 232 } 233 234 static void __init pmd_leaf_tests(unsigned long pfn, pgprot_t prot) 235 { 236 pmd_t pmd; 237 238 if (!has_transparent_hugepage()) 239 return; 240 241 pr_debug("Validating PMD leaf\n"); 242 pmd = pfn_pmd(pfn, prot); 243 244 /* 245 * PMD based THP is a leaf entry. 246 */ 247 pmd = pmd_mkhuge(pmd); 248 WARN_ON(!pmd_leaf(pmd)); 249 } 250 251 static void __init pmd_savedwrite_tests(unsigned long pfn, pgprot_t prot) 252 { 253 pmd_t pmd; 254 255 if (!IS_ENABLED(CONFIG_NUMA_BALANCING)) 256 return; 257 258 if (!has_transparent_hugepage()) 259 return; 260 261 pr_debug("Validating PMD saved write\n"); 262 pmd = pfn_pmd(pfn, prot); 263 WARN_ON(!pmd_savedwrite(pmd_mk_savedwrite(pmd_clear_savedwrite(pmd)))); 264 WARN_ON(pmd_savedwrite(pmd_clear_savedwrite(pmd_mk_savedwrite(pmd)))); 265 } 266 267 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD 268 static void __init pud_basic_tests(struct mm_struct *mm, unsigned long pfn, int idx) 269 { 270 pgprot_t prot = protection_map[idx]; 271 unsigned long val = idx, *ptr = &val; 272 pud_t pud; 273 274 if (!has_transparent_hugepage()) 275 return; 276 277 pr_debug("Validating PUD basic (%pGv)\n", ptr); 278 pud = pfn_pud(pfn, prot); 279 280 /* 281 * This test needs to be executed after the given page table entry 282 * is created with pfn_pud() to make sure that protection_map[idx] 283 * does not have the dirty bit enabled from the beginning. This is 284 * important for platforms like arm64 where (!PTE_RDONLY) indicate 285 * dirty bit being set. 286 */ 287 WARN_ON(pud_dirty(pud_wrprotect(pud))); 288 289 WARN_ON(!pud_same(pud, pud)); 290 WARN_ON(!pud_young(pud_mkyoung(pud_mkold(pud)))); 291 WARN_ON(!pud_dirty(pud_mkdirty(pud_mkclean(pud)))); 292 WARN_ON(pud_dirty(pud_mkclean(pud_mkdirty(pud)))); 293 WARN_ON(!pud_write(pud_mkwrite(pud_wrprotect(pud)))); 294 WARN_ON(pud_write(pud_wrprotect(pud_mkwrite(pud)))); 295 WARN_ON(pud_young(pud_mkold(pud_mkyoung(pud)))); 296 WARN_ON(pud_dirty(pud_wrprotect(pud_mkclean(pud)))); 297 WARN_ON(!pud_dirty(pud_wrprotect(pud_mkdirty(pud)))); 298 299 if (mm_pmd_folded(mm)) 300 return; 301 302 /* 303 * A huge page does not point to next level page table 304 * entry. Hence this must qualify as pud_bad(). 305 */ 306 WARN_ON(!pud_bad(pud_mkhuge(pud))); 307 } 308 309 static void __init pud_advanced_tests(struct mm_struct *mm, 310 struct vm_area_struct *vma, pud_t *pudp, 311 unsigned long pfn, unsigned long vaddr, 312 pgprot_t prot) 313 { 314 pud_t pud; 315 316 if (!has_transparent_hugepage()) 317 return; 318 319 pr_debug("Validating PUD advanced\n"); 320 /* Align the address wrt HPAGE_PUD_SIZE */ 321 vaddr &= HPAGE_PUD_MASK; 322 323 pud = pfn_pud(pfn, prot); 324 set_pud_at(mm, vaddr, pudp, pud); 325 pudp_set_wrprotect(mm, vaddr, pudp); 326 pud = READ_ONCE(*pudp); 327 WARN_ON(pud_write(pud)); 328 329 #ifndef __PAGETABLE_PMD_FOLDED 330 pudp_huge_get_and_clear(mm, vaddr, pudp); 331 pud = READ_ONCE(*pudp); 332 WARN_ON(!pud_none(pud)); 333 #endif /* __PAGETABLE_PMD_FOLDED */ 334 pud = pfn_pud(pfn, prot); 335 pud = pud_wrprotect(pud); 336 pud = pud_mkclean(pud); 337 set_pud_at(mm, vaddr, pudp, pud); 338 pud = pud_mkwrite(pud); 339 pud = pud_mkdirty(pud); 340 pudp_set_access_flags(vma, vaddr, pudp, pud, 1); 341 pud = READ_ONCE(*pudp); 342 WARN_ON(!(pud_write(pud) && pud_dirty(pud))); 343 344 #ifndef __PAGETABLE_PMD_FOLDED 345 pudp_huge_get_and_clear_full(mm, vaddr, pudp, 1); 346 pud = READ_ONCE(*pudp); 347 WARN_ON(!pud_none(pud)); 348 #endif /* __PAGETABLE_PMD_FOLDED */ 349 350 pud = pfn_pud(pfn, prot); 351 pud = pud_mkyoung(pud); 352 set_pud_at(mm, vaddr, pudp, pud); 353 pudp_test_and_clear_young(vma, vaddr, pudp); 354 pud = READ_ONCE(*pudp); 355 WARN_ON(pud_young(pud)); 356 357 pudp_huge_get_and_clear(mm, vaddr, pudp); 358 } 359 360 static void __init pud_leaf_tests(unsigned long pfn, pgprot_t prot) 361 { 362 pud_t pud; 363 364 if (!has_transparent_hugepage()) 365 return; 366 367 pr_debug("Validating PUD leaf\n"); 368 pud = pfn_pud(pfn, prot); 369 /* 370 * PUD based THP is a leaf entry. 371 */ 372 pud = pud_mkhuge(pud); 373 WARN_ON(!pud_leaf(pud)); 374 } 375 #else /* !CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */ 376 static void __init pud_basic_tests(struct mm_struct *mm, unsigned long pfn, int idx) { } 377 static void __init pud_advanced_tests(struct mm_struct *mm, 378 struct vm_area_struct *vma, pud_t *pudp, 379 unsigned long pfn, unsigned long vaddr, 380 pgprot_t prot) 381 { 382 } 383 static void __init pud_leaf_tests(unsigned long pfn, pgprot_t prot) { } 384 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */ 385 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */ 386 static void __init pmd_basic_tests(unsigned long pfn, int idx) { } 387 static void __init pud_basic_tests(struct mm_struct *mm, unsigned long pfn, int idx) { } 388 static void __init pmd_advanced_tests(struct mm_struct *mm, 389 struct vm_area_struct *vma, pmd_t *pmdp, 390 unsigned long pfn, unsigned long vaddr, 391 pgprot_t prot, pgtable_t pgtable) 392 { 393 } 394 static void __init pud_advanced_tests(struct mm_struct *mm, 395 struct vm_area_struct *vma, pud_t *pudp, 396 unsigned long pfn, unsigned long vaddr, 397 pgprot_t prot) 398 { 399 } 400 static void __init pmd_leaf_tests(unsigned long pfn, pgprot_t prot) { } 401 static void __init pud_leaf_tests(unsigned long pfn, pgprot_t prot) { } 402 static void __init pmd_savedwrite_tests(unsigned long pfn, pgprot_t prot) { } 403 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 404 405 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP 406 static void __init pmd_huge_tests(pmd_t *pmdp, unsigned long pfn, pgprot_t prot) 407 { 408 pmd_t pmd; 409 410 if (!arch_vmap_pmd_supported(prot)) 411 return; 412 413 pr_debug("Validating PMD huge\n"); 414 /* 415 * X86 defined pmd_set_huge() verifies that the given 416 * PMD is not a populated non-leaf entry. 417 */ 418 WRITE_ONCE(*pmdp, __pmd(0)); 419 WARN_ON(!pmd_set_huge(pmdp, __pfn_to_phys(pfn), prot)); 420 WARN_ON(!pmd_clear_huge(pmdp)); 421 pmd = READ_ONCE(*pmdp); 422 WARN_ON(!pmd_none(pmd)); 423 } 424 425 static void __init pud_huge_tests(pud_t *pudp, unsigned long pfn, pgprot_t prot) 426 { 427 pud_t pud; 428 429 if (!arch_vmap_pud_supported(prot)) 430 return; 431 432 pr_debug("Validating PUD huge\n"); 433 /* 434 * X86 defined pud_set_huge() verifies that the given 435 * PUD is not a populated non-leaf entry. 436 */ 437 WRITE_ONCE(*pudp, __pud(0)); 438 WARN_ON(!pud_set_huge(pudp, __pfn_to_phys(pfn), prot)); 439 WARN_ON(!pud_clear_huge(pudp)); 440 pud = READ_ONCE(*pudp); 441 WARN_ON(!pud_none(pud)); 442 } 443 #else /* !CONFIG_HAVE_ARCH_HUGE_VMAP */ 444 static void __init pmd_huge_tests(pmd_t *pmdp, unsigned long pfn, pgprot_t prot) { } 445 static void __init pud_huge_tests(pud_t *pudp, unsigned long pfn, pgprot_t prot) { } 446 #endif /* CONFIG_HAVE_ARCH_HUGE_VMAP */ 447 448 static void __init p4d_basic_tests(unsigned long pfn, pgprot_t prot) 449 { 450 p4d_t p4d; 451 452 pr_debug("Validating P4D basic\n"); 453 memset(&p4d, RANDOM_NZVALUE, sizeof(p4d_t)); 454 WARN_ON(!p4d_same(p4d, p4d)); 455 } 456 457 static void __init pgd_basic_tests(unsigned long pfn, pgprot_t prot) 458 { 459 pgd_t pgd; 460 461 pr_debug("Validating PGD basic\n"); 462 memset(&pgd, RANDOM_NZVALUE, sizeof(pgd_t)); 463 WARN_ON(!pgd_same(pgd, pgd)); 464 } 465 466 #ifndef __PAGETABLE_PUD_FOLDED 467 static void __init pud_clear_tests(struct mm_struct *mm, pud_t *pudp) 468 { 469 pud_t pud = READ_ONCE(*pudp); 470 471 if (mm_pmd_folded(mm)) 472 return; 473 474 pr_debug("Validating PUD clear\n"); 475 pud = __pud(pud_val(pud) | RANDOM_ORVALUE); 476 WRITE_ONCE(*pudp, pud); 477 pud_clear(pudp); 478 pud = READ_ONCE(*pudp); 479 WARN_ON(!pud_none(pud)); 480 } 481 482 static void __init pud_populate_tests(struct mm_struct *mm, pud_t *pudp, 483 pmd_t *pmdp) 484 { 485 pud_t pud; 486 487 if (mm_pmd_folded(mm)) 488 return; 489 490 pr_debug("Validating PUD populate\n"); 491 /* 492 * This entry points to next level page table page. 493 * Hence this must not qualify as pud_bad(). 494 */ 495 pud_populate(mm, pudp, pmdp); 496 pud = READ_ONCE(*pudp); 497 WARN_ON(pud_bad(pud)); 498 } 499 #else /* !__PAGETABLE_PUD_FOLDED */ 500 static void __init pud_clear_tests(struct mm_struct *mm, pud_t *pudp) { } 501 static void __init pud_populate_tests(struct mm_struct *mm, pud_t *pudp, 502 pmd_t *pmdp) 503 { 504 } 505 #endif /* PAGETABLE_PUD_FOLDED */ 506 507 #ifndef __PAGETABLE_P4D_FOLDED 508 static void __init p4d_clear_tests(struct mm_struct *mm, p4d_t *p4dp) 509 { 510 p4d_t p4d = READ_ONCE(*p4dp); 511 512 if (mm_pud_folded(mm)) 513 return; 514 515 pr_debug("Validating P4D clear\n"); 516 p4d = __p4d(p4d_val(p4d) | RANDOM_ORVALUE); 517 WRITE_ONCE(*p4dp, p4d); 518 p4d_clear(p4dp); 519 p4d = READ_ONCE(*p4dp); 520 WARN_ON(!p4d_none(p4d)); 521 } 522 523 static void __init p4d_populate_tests(struct mm_struct *mm, p4d_t *p4dp, 524 pud_t *pudp) 525 { 526 p4d_t p4d; 527 528 if (mm_pud_folded(mm)) 529 return; 530 531 pr_debug("Validating P4D populate\n"); 532 /* 533 * This entry points to next level page table page. 534 * Hence this must not qualify as p4d_bad(). 535 */ 536 pud_clear(pudp); 537 p4d_clear(p4dp); 538 p4d_populate(mm, p4dp, pudp); 539 p4d = READ_ONCE(*p4dp); 540 WARN_ON(p4d_bad(p4d)); 541 } 542 543 static void __init pgd_clear_tests(struct mm_struct *mm, pgd_t *pgdp) 544 { 545 pgd_t pgd = READ_ONCE(*pgdp); 546 547 if (mm_p4d_folded(mm)) 548 return; 549 550 pr_debug("Validating PGD clear\n"); 551 pgd = __pgd(pgd_val(pgd) | RANDOM_ORVALUE); 552 WRITE_ONCE(*pgdp, pgd); 553 pgd_clear(pgdp); 554 pgd = READ_ONCE(*pgdp); 555 WARN_ON(!pgd_none(pgd)); 556 } 557 558 static void __init pgd_populate_tests(struct mm_struct *mm, pgd_t *pgdp, 559 p4d_t *p4dp) 560 { 561 pgd_t pgd; 562 563 if (mm_p4d_folded(mm)) 564 return; 565 566 pr_debug("Validating PGD populate\n"); 567 /* 568 * This entry points to next level page table page. 569 * Hence this must not qualify as pgd_bad(). 570 */ 571 p4d_clear(p4dp); 572 pgd_clear(pgdp); 573 pgd_populate(mm, pgdp, p4dp); 574 pgd = READ_ONCE(*pgdp); 575 WARN_ON(pgd_bad(pgd)); 576 } 577 #else /* !__PAGETABLE_P4D_FOLDED */ 578 static void __init p4d_clear_tests(struct mm_struct *mm, p4d_t *p4dp) { } 579 static void __init pgd_clear_tests(struct mm_struct *mm, pgd_t *pgdp) { } 580 static void __init p4d_populate_tests(struct mm_struct *mm, p4d_t *p4dp, 581 pud_t *pudp) 582 { 583 } 584 static void __init pgd_populate_tests(struct mm_struct *mm, pgd_t *pgdp, 585 p4d_t *p4dp) 586 { 587 } 588 #endif /* PAGETABLE_P4D_FOLDED */ 589 590 static void __init pte_clear_tests(struct mm_struct *mm, pte_t *ptep, 591 unsigned long pfn, unsigned long vaddr, 592 pgprot_t prot) 593 { 594 pte_t pte = pfn_pte(pfn, prot); 595 596 pr_debug("Validating PTE clear\n"); 597 #ifndef CONFIG_RISCV 598 pte = __pte(pte_val(pte) | RANDOM_ORVALUE); 599 #endif 600 set_pte_at(mm, vaddr, ptep, pte); 601 barrier(); 602 pte_clear(mm, vaddr, ptep); 603 pte = ptep_get(ptep); 604 WARN_ON(!pte_none(pte)); 605 } 606 607 static void __init pmd_clear_tests(struct mm_struct *mm, pmd_t *pmdp) 608 { 609 pmd_t pmd = READ_ONCE(*pmdp); 610 611 pr_debug("Validating PMD clear\n"); 612 pmd = __pmd(pmd_val(pmd) | RANDOM_ORVALUE); 613 WRITE_ONCE(*pmdp, pmd); 614 pmd_clear(pmdp); 615 pmd = READ_ONCE(*pmdp); 616 WARN_ON(!pmd_none(pmd)); 617 } 618 619 static void __init pmd_populate_tests(struct mm_struct *mm, pmd_t *pmdp, 620 pgtable_t pgtable) 621 { 622 pmd_t pmd; 623 624 pr_debug("Validating PMD populate\n"); 625 /* 626 * This entry points to next level page table page. 627 * Hence this must not qualify as pmd_bad(). 628 */ 629 pmd_populate(mm, pmdp, pgtable); 630 pmd = READ_ONCE(*pmdp); 631 WARN_ON(pmd_bad(pmd)); 632 } 633 634 static void __init pte_special_tests(unsigned long pfn, pgprot_t prot) 635 { 636 pte_t pte = pfn_pte(pfn, prot); 637 638 if (!IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL)) 639 return; 640 641 pr_debug("Validating PTE special\n"); 642 WARN_ON(!pte_special(pte_mkspecial(pte))); 643 } 644 645 static void __init pte_protnone_tests(unsigned long pfn, pgprot_t prot) 646 { 647 pte_t pte = pfn_pte(pfn, prot); 648 649 if (!IS_ENABLED(CONFIG_NUMA_BALANCING)) 650 return; 651 652 pr_debug("Validating PTE protnone\n"); 653 WARN_ON(!pte_protnone(pte)); 654 WARN_ON(!pte_present(pte)); 655 } 656 657 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 658 static void __init pmd_protnone_tests(unsigned long pfn, pgprot_t prot) 659 { 660 pmd_t pmd; 661 662 if (!IS_ENABLED(CONFIG_NUMA_BALANCING)) 663 return; 664 665 if (!has_transparent_hugepage()) 666 return; 667 668 pr_debug("Validating PMD protnone\n"); 669 pmd = pmd_mkhuge(pfn_pmd(pfn, prot)); 670 WARN_ON(!pmd_protnone(pmd)); 671 WARN_ON(!pmd_present(pmd)); 672 } 673 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */ 674 static void __init pmd_protnone_tests(unsigned long pfn, pgprot_t prot) { } 675 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 676 677 #ifdef CONFIG_ARCH_HAS_PTE_DEVMAP 678 static void __init pte_devmap_tests(unsigned long pfn, pgprot_t prot) 679 { 680 pte_t pte = pfn_pte(pfn, prot); 681 682 pr_debug("Validating PTE devmap\n"); 683 WARN_ON(!pte_devmap(pte_mkdevmap(pte))); 684 } 685 686 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 687 static void __init pmd_devmap_tests(unsigned long pfn, pgprot_t prot) 688 { 689 pmd_t pmd; 690 691 if (!has_transparent_hugepage()) 692 return; 693 694 pr_debug("Validating PMD devmap\n"); 695 pmd = pfn_pmd(pfn, prot); 696 WARN_ON(!pmd_devmap(pmd_mkdevmap(pmd))); 697 } 698 699 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD 700 static void __init pud_devmap_tests(unsigned long pfn, pgprot_t prot) 701 { 702 pud_t pud; 703 704 if (!has_transparent_hugepage()) 705 return; 706 707 pr_debug("Validating PUD devmap\n"); 708 pud = pfn_pud(pfn, prot); 709 WARN_ON(!pud_devmap(pud_mkdevmap(pud))); 710 } 711 #else /* !CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */ 712 static void __init pud_devmap_tests(unsigned long pfn, pgprot_t prot) { } 713 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */ 714 #else /* CONFIG_TRANSPARENT_HUGEPAGE */ 715 static void __init pmd_devmap_tests(unsigned long pfn, pgprot_t prot) { } 716 static void __init pud_devmap_tests(unsigned long pfn, pgprot_t prot) { } 717 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 718 #else 719 static void __init pte_devmap_tests(unsigned long pfn, pgprot_t prot) { } 720 static void __init pmd_devmap_tests(unsigned long pfn, pgprot_t prot) { } 721 static void __init pud_devmap_tests(unsigned long pfn, pgprot_t prot) { } 722 #endif /* CONFIG_ARCH_HAS_PTE_DEVMAP */ 723 724 static void __init pte_soft_dirty_tests(unsigned long pfn, pgprot_t prot) 725 { 726 pte_t pte = pfn_pte(pfn, prot); 727 728 if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY)) 729 return; 730 731 pr_debug("Validating PTE soft dirty\n"); 732 WARN_ON(!pte_soft_dirty(pte_mksoft_dirty(pte))); 733 WARN_ON(pte_soft_dirty(pte_clear_soft_dirty(pte))); 734 } 735 736 static void __init pte_swap_soft_dirty_tests(unsigned long pfn, pgprot_t prot) 737 { 738 pte_t pte = pfn_pte(pfn, prot); 739 740 if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY)) 741 return; 742 743 pr_debug("Validating PTE swap soft dirty\n"); 744 WARN_ON(!pte_swp_soft_dirty(pte_swp_mksoft_dirty(pte))); 745 WARN_ON(pte_swp_soft_dirty(pte_swp_clear_soft_dirty(pte))); 746 } 747 748 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 749 static void __init pmd_soft_dirty_tests(unsigned long pfn, pgprot_t prot) 750 { 751 pmd_t pmd; 752 753 if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY)) 754 return; 755 756 if (!has_transparent_hugepage()) 757 return; 758 759 pr_debug("Validating PMD soft dirty\n"); 760 pmd = pfn_pmd(pfn, prot); 761 WARN_ON(!pmd_soft_dirty(pmd_mksoft_dirty(pmd))); 762 WARN_ON(pmd_soft_dirty(pmd_clear_soft_dirty(pmd))); 763 } 764 765 static void __init pmd_swap_soft_dirty_tests(unsigned long pfn, pgprot_t prot) 766 { 767 pmd_t pmd; 768 769 if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) || 770 !IS_ENABLED(CONFIG_ARCH_ENABLE_THP_MIGRATION)) 771 return; 772 773 if (!has_transparent_hugepage()) 774 return; 775 776 pr_debug("Validating PMD swap soft dirty\n"); 777 pmd = pfn_pmd(pfn, prot); 778 WARN_ON(!pmd_swp_soft_dirty(pmd_swp_mksoft_dirty(pmd))); 779 WARN_ON(pmd_swp_soft_dirty(pmd_swp_clear_soft_dirty(pmd))); 780 } 781 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */ 782 static void __init pmd_soft_dirty_tests(unsigned long pfn, pgprot_t prot) { } 783 static void __init pmd_swap_soft_dirty_tests(unsigned long pfn, pgprot_t prot) 784 { 785 } 786 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 787 788 static void __init pte_swap_tests(unsigned long pfn, pgprot_t prot) 789 { 790 swp_entry_t swp; 791 pte_t pte; 792 793 pr_debug("Validating PTE swap\n"); 794 pte = pfn_pte(pfn, prot); 795 swp = __pte_to_swp_entry(pte); 796 pte = __swp_entry_to_pte(swp); 797 WARN_ON(pfn != pte_pfn(pte)); 798 } 799 800 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION 801 static void __init pmd_swap_tests(unsigned long pfn, pgprot_t prot) 802 { 803 swp_entry_t swp; 804 pmd_t pmd; 805 806 if (!has_transparent_hugepage()) 807 return; 808 809 pr_debug("Validating PMD swap\n"); 810 pmd = pfn_pmd(pfn, prot); 811 swp = __pmd_to_swp_entry(pmd); 812 pmd = __swp_entry_to_pmd(swp); 813 WARN_ON(pfn != pmd_pfn(pmd)); 814 } 815 #else /* !CONFIG_ARCH_ENABLE_THP_MIGRATION */ 816 static void __init pmd_swap_tests(unsigned long pfn, pgprot_t prot) { } 817 #endif /* CONFIG_ARCH_ENABLE_THP_MIGRATION */ 818 819 static void __init swap_migration_tests(void) 820 { 821 struct page *page; 822 swp_entry_t swp; 823 824 if (!IS_ENABLED(CONFIG_MIGRATION)) 825 return; 826 827 pr_debug("Validating swap migration\n"); 828 /* 829 * swap_migration_tests() requires a dedicated page as it needs to 830 * be locked before creating a migration entry from it. Locking the 831 * page that actually maps kernel text ('start_kernel') can be real 832 * problematic. Lets allocate a dedicated page explicitly for this 833 * purpose that will be freed subsequently. 834 */ 835 page = alloc_page(GFP_KERNEL); 836 if (!page) { 837 pr_err("page allocation failed\n"); 838 return; 839 } 840 841 /* 842 * make_migration_entry() expects given page to be 843 * locked, otherwise it stumbles upon a BUG_ON(). 844 */ 845 __SetPageLocked(page); 846 swp = make_writable_migration_entry(page_to_pfn(page)); 847 WARN_ON(!is_migration_entry(swp)); 848 WARN_ON(!is_writable_migration_entry(swp)); 849 850 swp = make_readable_migration_entry(swp_offset(swp)); 851 WARN_ON(!is_migration_entry(swp)); 852 WARN_ON(is_writable_migration_entry(swp)); 853 854 swp = make_readable_migration_entry(page_to_pfn(page)); 855 WARN_ON(!is_migration_entry(swp)); 856 WARN_ON(is_writable_migration_entry(swp)); 857 __ClearPageLocked(page); 858 __free_page(page); 859 } 860 861 #ifdef CONFIG_HUGETLB_PAGE 862 static void __init hugetlb_basic_tests(unsigned long pfn, pgprot_t prot) 863 { 864 struct page *page; 865 pte_t pte; 866 867 pr_debug("Validating HugeTLB basic\n"); 868 /* 869 * Accessing the page associated with the pfn is safe here, 870 * as it was previously derived from a real kernel symbol. 871 */ 872 page = pfn_to_page(pfn); 873 pte = mk_huge_pte(page, prot); 874 875 WARN_ON(!huge_pte_dirty(huge_pte_mkdirty(pte))); 876 WARN_ON(!huge_pte_write(huge_pte_mkwrite(huge_pte_wrprotect(pte)))); 877 WARN_ON(huge_pte_write(huge_pte_wrprotect(huge_pte_mkwrite(pte)))); 878 879 #ifdef CONFIG_ARCH_WANT_GENERAL_HUGETLB 880 pte = pfn_pte(pfn, prot); 881 882 WARN_ON(!pte_huge(pte_mkhuge(pte))); 883 #endif /* CONFIG_ARCH_WANT_GENERAL_HUGETLB */ 884 } 885 #else /* !CONFIG_HUGETLB_PAGE */ 886 static void __init hugetlb_basic_tests(unsigned long pfn, pgprot_t prot) { } 887 #endif /* CONFIG_HUGETLB_PAGE */ 888 889 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 890 static void __init pmd_thp_tests(unsigned long pfn, pgprot_t prot) 891 { 892 pmd_t pmd; 893 894 if (!has_transparent_hugepage()) 895 return; 896 897 pr_debug("Validating PMD based THP\n"); 898 /* 899 * pmd_trans_huge() and pmd_present() must return positive after 900 * MMU invalidation with pmd_mkinvalid(). This behavior is an 901 * optimization for transparent huge page. pmd_trans_huge() must 902 * be true if pmd_page() returns a valid THP to avoid taking the 903 * pmd_lock when others walk over non transhuge pmds (i.e. there 904 * are no THP allocated). Especially when splitting a THP and 905 * removing the present bit from the pmd, pmd_trans_huge() still 906 * needs to return true. pmd_present() should be true whenever 907 * pmd_trans_huge() returns true. 908 */ 909 pmd = pfn_pmd(pfn, prot); 910 WARN_ON(!pmd_trans_huge(pmd_mkhuge(pmd))); 911 912 #ifndef __HAVE_ARCH_PMDP_INVALIDATE 913 WARN_ON(!pmd_trans_huge(pmd_mkinvalid(pmd_mkhuge(pmd)))); 914 WARN_ON(!pmd_present(pmd_mkinvalid(pmd_mkhuge(pmd)))); 915 #endif /* __HAVE_ARCH_PMDP_INVALIDATE */ 916 } 917 918 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD 919 static void __init pud_thp_tests(unsigned long pfn, pgprot_t prot) 920 { 921 pud_t pud; 922 923 if (!has_transparent_hugepage()) 924 return; 925 926 pr_debug("Validating PUD based THP\n"); 927 pud = pfn_pud(pfn, prot); 928 WARN_ON(!pud_trans_huge(pud_mkhuge(pud))); 929 930 /* 931 * pud_mkinvalid() has been dropped for now. Enable back 932 * these tests when it comes back with a modified pud_present(). 933 * 934 * WARN_ON(!pud_trans_huge(pud_mkinvalid(pud_mkhuge(pud)))); 935 * WARN_ON(!pud_present(pud_mkinvalid(pud_mkhuge(pud)))); 936 */ 937 } 938 #else /* !CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */ 939 static void __init pud_thp_tests(unsigned long pfn, pgprot_t prot) { } 940 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */ 941 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */ 942 static void __init pmd_thp_tests(unsigned long pfn, pgprot_t prot) { } 943 static void __init pud_thp_tests(unsigned long pfn, pgprot_t prot) { } 944 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 945 946 static unsigned long __init get_random_vaddr(void) 947 { 948 unsigned long random_vaddr, random_pages, total_user_pages; 949 950 total_user_pages = (TASK_SIZE - FIRST_USER_ADDRESS) / PAGE_SIZE; 951 952 random_pages = get_random_long() % total_user_pages; 953 random_vaddr = FIRST_USER_ADDRESS + random_pages * PAGE_SIZE; 954 955 return random_vaddr; 956 } 957 958 static int __init debug_vm_pgtable(void) 959 { 960 struct vm_area_struct *vma; 961 struct mm_struct *mm; 962 pgd_t *pgdp; 963 p4d_t *p4dp, *saved_p4dp; 964 pud_t *pudp, *saved_pudp; 965 pmd_t *pmdp, *saved_pmdp, pmd; 966 pte_t *ptep; 967 pgtable_t saved_ptep; 968 pgprot_t prot, protnone; 969 phys_addr_t paddr; 970 unsigned long vaddr, pte_aligned, pmd_aligned; 971 unsigned long pud_aligned, p4d_aligned, pgd_aligned; 972 spinlock_t *ptl = NULL; 973 int idx; 974 975 pr_info("Validating architecture page table helpers\n"); 976 prot = vm_get_page_prot(VMFLAGS); 977 vaddr = get_random_vaddr(); 978 mm = mm_alloc(); 979 if (!mm) { 980 pr_err("mm_struct allocation failed\n"); 981 return 1; 982 } 983 984 /* 985 * __P000 (or even __S000) will help create page table entries with 986 * PROT_NONE permission as required for pxx_protnone_tests(). 987 */ 988 protnone = __P000; 989 990 vma = vm_area_alloc(mm); 991 if (!vma) { 992 pr_err("vma allocation failed\n"); 993 return 1; 994 } 995 996 /* 997 * PFN for mapping at PTE level is determined from a standard kernel 998 * text symbol. But pfns for higher page table levels are derived by 999 * masking lower bits of this real pfn. These derived pfns might not 1000 * exist on the platform but that does not really matter as pfn_pxx() 1001 * helpers will still create appropriate entries for the test. This 1002 * helps avoid large memory block allocations to be used for mapping 1003 * at higher page table levels. 1004 */ 1005 paddr = __pa_symbol(&start_kernel); 1006 1007 pte_aligned = (paddr & PAGE_MASK) >> PAGE_SHIFT; 1008 pmd_aligned = (paddr & PMD_MASK) >> PAGE_SHIFT; 1009 pud_aligned = (paddr & PUD_MASK) >> PAGE_SHIFT; 1010 p4d_aligned = (paddr & P4D_MASK) >> PAGE_SHIFT; 1011 pgd_aligned = (paddr & PGDIR_MASK) >> PAGE_SHIFT; 1012 WARN_ON(!pfn_valid(pte_aligned)); 1013 1014 pgdp = pgd_offset(mm, vaddr); 1015 p4dp = p4d_alloc(mm, pgdp, vaddr); 1016 pudp = pud_alloc(mm, p4dp, vaddr); 1017 pmdp = pmd_alloc(mm, pudp, vaddr); 1018 /* 1019 * Allocate pgtable_t 1020 */ 1021 if (pte_alloc(mm, pmdp)) { 1022 pr_err("pgtable allocation failed\n"); 1023 return 1; 1024 } 1025 1026 /* 1027 * Save all the page table page addresses as the page table 1028 * entries will be used for testing with random or garbage 1029 * values. These saved addresses will be used for freeing 1030 * page table pages. 1031 */ 1032 pmd = READ_ONCE(*pmdp); 1033 saved_p4dp = p4d_offset(pgdp, 0UL); 1034 saved_pudp = pud_offset(p4dp, 0UL); 1035 saved_pmdp = pmd_offset(pudp, 0UL); 1036 saved_ptep = pmd_pgtable(pmd); 1037 1038 /* 1039 * Iterate over the protection_map[] to make sure that all 1040 * the basic page table transformation validations just hold 1041 * true irrespective of the starting protection value for a 1042 * given page table entry. 1043 */ 1044 for (idx = 0; idx < ARRAY_SIZE(protection_map); idx++) { 1045 pte_basic_tests(pte_aligned, idx); 1046 pmd_basic_tests(pmd_aligned, idx); 1047 pud_basic_tests(mm, pud_aligned, idx); 1048 } 1049 1050 /* 1051 * Both P4D and PGD level tests are very basic which do not 1052 * involve creating page table entries from the protection 1053 * value and the given pfn. Hence just keep them out from 1054 * the above iteration for now to save some test execution 1055 * time. 1056 */ 1057 p4d_basic_tests(p4d_aligned, prot); 1058 pgd_basic_tests(pgd_aligned, prot); 1059 1060 pmd_leaf_tests(pmd_aligned, prot); 1061 pud_leaf_tests(pud_aligned, prot); 1062 1063 pte_savedwrite_tests(pte_aligned, protnone); 1064 pmd_savedwrite_tests(pmd_aligned, protnone); 1065 1066 pte_special_tests(pte_aligned, prot); 1067 pte_protnone_tests(pte_aligned, protnone); 1068 pmd_protnone_tests(pmd_aligned, protnone); 1069 1070 pte_devmap_tests(pte_aligned, prot); 1071 pmd_devmap_tests(pmd_aligned, prot); 1072 pud_devmap_tests(pud_aligned, prot); 1073 1074 pte_soft_dirty_tests(pte_aligned, prot); 1075 pmd_soft_dirty_tests(pmd_aligned, prot); 1076 pte_swap_soft_dirty_tests(pte_aligned, prot); 1077 pmd_swap_soft_dirty_tests(pmd_aligned, prot); 1078 1079 pte_swap_tests(pte_aligned, prot); 1080 pmd_swap_tests(pmd_aligned, prot); 1081 1082 swap_migration_tests(); 1083 1084 pmd_thp_tests(pmd_aligned, prot); 1085 pud_thp_tests(pud_aligned, prot); 1086 1087 hugetlb_basic_tests(pte_aligned, prot); 1088 1089 /* 1090 * Page table modifying tests. They need to hold 1091 * proper page table lock. 1092 */ 1093 1094 ptep = pte_offset_map_lock(mm, pmdp, vaddr, &ptl); 1095 pte_clear_tests(mm, ptep, pte_aligned, vaddr, prot); 1096 pte_advanced_tests(mm, vma, ptep, pte_aligned, vaddr, prot); 1097 pte_unmap_unlock(ptep, ptl); 1098 1099 ptl = pmd_lock(mm, pmdp); 1100 pmd_clear_tests(mm, pmdp); 1101 pmd_advanced_tests(mm, vma, pmdp, pmd_aligned, vaddr, prot, saved_ptep); 1102 pmd_huge_tests(pmdp, pmd_aligned, prot); 1103 pmd_populate_tests(mm, pmdp, saved_ptep); 1104 spin_unlock(ptl); 1105 1106 ptl = pud_lock(mm, pudp); 1107 pud_clear_tests(mm, pudp); 1108 pud_advanced_tests(mm, vma, pudp, pud_aligned, vaddr, prot); 1109 pud_huge_tests(pudp, pud_aligned, prot); 1110 pud_populate_tests(mm, pudp, saved_pmdp); 1111 spin_unlock(ptl); 1112 1113 spin_lock(&mm->page_table_lock); 1114 p4d_clear_tests(mm, p4dp); 1115 pgd_clear_tests(mm, pgdp); 1116 p4d_populate_tests(mm, p4dp, saved_pudp); 1117 pgd_populate_tests(mm, pgdp, saved_p4dp); 1118 spin_unlock(&mm->page_table_lock); 1119 1120 p4d_free(mm, saved_p4dp); 1121 pud_free(mm, saved_pudp); 1122 pmd_free(mm, saved_pmdp); 1123 pte_free(mm, saved_ptep); 1124 1125 vm_area_free(vma); 1126 mm_dec_nr_puds(mm); 1127 mm_dec_nr_pmds(mm); 1128 mm_dec_nr_ptes(mm); 1129 mmdrop(mm); 1130 return 0; 1131 } 1132 late_initcall(debug_vm_pgtable); 1133