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/memblock.h> 19 #include <linux/mm.h> 20 #include <linux/mman.h> 21 #include <linux/mm_types.h> 22 #include <linux/module.h> 23 #include <linux/pfn_t.h> 24 #include <linux/printk.h> 25 #include <linux/pgtable.h> 26 #include <linux/random.h> 27 #include <linux/spinlock.h> 28 #include <linux/swap.h> 29 #include <linux/swapops.h> 30 #include <linux/start_kernel.h> 31 #include <linux/sched/mm.h> 32 #include <linux/io.h> 33 #include <linux/vmalloc.h> 34 35 #include <asm/cacheflush.h> 36 #include <asm/pgalloc.h> 37 #include <asm/tlbflush.h> 38 39 /* 40 * Please refer Documentation/mm/arch_pgtable_helpers.rst for the semantics 41 * expectations that are being validated here. All future changes in here 42 * or the documentation need to be in sync. 43 */ 44 #define RANDOM_NZVALUE GENMASK(7, 0) 45 46 struct pgtable_debug_args { 47 struct mm_struct *mm; 48 struct vm_area_struct *vma; 49 50 pgd_t *pgdp; 51 p4d_t *p4dp; 52 pud_t *pudp; 53 pmd_t *pmdp; 54 pte_t *ptep; 55 56 p4d_t *start_p4dp; 57 pud_t *start_pudp; 58 pmd_t *start_pmdp; 59 pgtable_t start_ptep; 60 61 unsigned long vaddr; 62 pgprot_t page_prot; 63 pgprot_t page_prot_none; 64 65 bool is_contiguous_page; 66 unsigned long pud_pfn; 67 unsigned long pmd_pfn; 68 unsigned long pte_pfn; 69 70 unsigned long fixed_alignment; 71 unsigned long fixed_pgd_pfn; 72 unsigned long fixed_p4d_pfn; 73 unsigned long fixed_pud_pfn; 74 unsigned long fixed_pmd_pfn; 75 unsigned long fixed_pte_pfn; 76 }; 77 78 static void __init pte_basic_tests(struct pgtable_debug_args *args, int idx) 79 { 80 pgprot_t prot = vm_get_page_prot(idx); 81 pte_t pte = pfn_pte(args->fixed_pte_pfn, prot); 82 unsigned long val = idx, *ptr = &val; 83 84 pr_debug("Validating PTE basic (%pGv)\n", ptr); 85 86 /* 87 * This test needs to be executed after the given page table entry 88 * is created with pfn_pte() to make sure that vm_get_page_prot(idx) 89 * does not have the dirty bit enabled from the beginning. This is 90 * important for platforms like arm64 where (!PTE_RDONLY) indicate 91 * dirty bit being set. 92 */ 93 WARN_ON(pte_dirty(pte_wrprotect(pte))); 94 95 WARN_ON(!pte_same(pte, pte)); 96 WARN_ON(!pte_young(pte_mkyoung(pte_mkold(pte)))); 97 WARN_ON(!pte_dirty(pte_mkdirty(pte_mkclean(pte)))); 98 WARN_ON(!pte_write(pte_mkwrite(pte_wrprotect(pte), args->vma))); 99 WARN_ON(pte_young(pte_mkold(pte_mkyoung(pte)))); 100 WARN_ON(pte_dirty(pte_mkclean(pte_mkdirty(pte)))); 101 WARN_ON(pte_write(pte_wrprotect(pte_mkwrite(pte, args->vma)))); 102 WARN_ON(pte_dirty(pte_wrprotect(pte_mkclean(pte)))); 103 WARN_ON(!pte_dirty(pte_wrprotect(pte_mkdirty(pte)))); 104 } 105 106 static void __init pte_advanced_tests(struct pgtable_debug_args *args) 107 { 108 struct page *page; 109 pte_t pte; 110 111 /* 112 * Architectures optimize set_pte_at by avoiding TLB flush. 113 * This requires set_pte_at to be not used to update an 114 * existing pte entry. Clear pte before we do set_pte_at 115 * 116 * flush_dcache_page() is called after set_pte_at() to clear 117 * PG_arch_1 for the page on ARM64. The page flag isn't cleared 118 * when it's released and page allocation check will fail when 119 * the page is allocated again. For architectures other than ARM64, 120 * the unexpected overhead of cache flushing is acceptable. 121 */ 122 page = (args->pte_pfn != ULONG_MAX) ? pfn_to_page(args->pte_pfn) : NULL; 123 if (!page) 124 return; 125 126 pr_debug("Validating PTE advanced\n"); 127 if (WARN_ON(!args->ptep)) 128 return; 129 130 pte = pfn_pte(args->pte_pfn, args->page_prot); 131 set_pte_at(args->mm, args->vaddr, args->ptep, pte); 132 flush_dcache_page(page); 133 ptep_set_wrprotect(args->mm, args->vaddr, args->ptep); 134 pte = ptep_get(args->ptep); 135 WARN_ON(pte_write(pte)); 136 ptep_get_and_clear(args->mm, args->vaddr, args->ptep); 137 pte = ptep_get(args->ptep); 138 WARN_ON(!pte_none(pte)); 139 140 pte = pfn_pte(args->pte_pfn, args->page_prot); 141 pte = pte_wrprotect(pte); 142 pte = pte_mkclean(pte); 143 set_pte_at(args->mm, args->vaddr, args->ptep, pte); 144 flush_dcache_page(page); 145 pte = pte_mkwrite(pte, args->vma); 146 pte = pte_mkdirty(pte); 147 ptep_set_access_flags(args->vma, args->vaddr, args->ptep, pte, 1); 148 pte = ptep_get(args->ptep); 149 WARN_ON(!(pte_write(pte) && pte_dirty(pte))); 150 ptep_get_and_clear_full(args->mm, args->vaddr, args->ptep, 1); 151 pte = ptep_get(args->ptep); 152 WARN_ON(!pte_none(pte)); 153 154 pte = pfn_pte(args->pte_pfn, args->page_prot); 155 pte = pte_mkyoung(pte); 156 set_pte_at(args->mm, args->vaddr, args->ptep, pte); 157 flush_dcache_page(page); 158 ptep_test_and_clear_young(args->vma, args->vaddr, args->ptep); 159 pte = ptep_get(args->ptep); 160 WARN_ON(pte_young(pte)); 161 162 ptep_get_and_clear_full(args->mm, args->vaddr, args->ptep, 1); 163 } 164 165 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 166 static void __init pmd_basic_tests(struct pgtable_debug_args *args, int idx) 167 { 168 pgprot_t prot = vm_get_page_prot(idx); 169 unsigned long val = idx, *ptr = &val; 170 pmd_t pmd; 171 172 if (!has_transparent_hugepage()) 173 return; 174 175 pr_debug("Validating PMD basic (%pGv)\n", ptr); 176 pmd = pfn_pmd(args->fixed_pmd_pfn, prot); 177 178 /* 179 * This test needs to be executed after the given page table entry 180 * is created with pfn_pmd() to make sure that vm_get_page_prot(idx) 181 * does not have the dirty bit enabled from the beginning. This is 182 * important for platforms like arm64 where (!PTE_RDONLY) indicate 183 * dirty bit being set. 184 */ 185 WARN_ON(pmd_dirty(pmd_wrprotect(pmd))); 186 187 188 WARN_ON(!pmd_same(pmd, pmd)); 189 WARN_ON(!pmd_young(pmd_mkyoung(pmd_mkold(pmd)))); 190 WARN_ON(!pmd_dirty(pmd_mkdirty(pmd_mkclean(pmd)))); 191 WARN_ON(!pmd_write(pmd_mkwrite(pmd_wrprotect(pmd), args->vma))); 192 WARN_ON(pmd_young(pmd_mkold(pmd_mkyoung(pmd)))); 193 WARN_ON(pmd_dirty(pmd_mkclean(pmd_mkdirty(pmd)))); 194 WARN_ON(pmd_write(pmd_wrprotect(pmd_mkwrite(pmd, args->vma)))); 195 WARN_ON(pmd_dirty(pmd_wrprotect(pmd_mkclean(pmd)))); 196 WARN_ON(!pmd_dirty(pmd_wrprotect(pmd_mkdirty(pmd)))); 197 /* 198 * A huge page does not point to next level page table 199 * entry. Hence this must qualify as pmd_bad(). 200 */ 201 WARN_ON(!pmd_bad(pmd_mkhuge(pmd))); 202 } 203 204 static void __init pmd_advanced_tests(struct pgtable_debug_args *args) 205 { 206 struct page *page; 207 pmd_t pmd; 208 unsigned long vaddr = args->vaddr; 209 210 if (!has_transparent_hugepage()) 211 return; 212 213 page = (args->pmd_pfn != ULONG_MAX) ? pfn_to_page(args->pmd_pfn) : NULL; 214 if (!page) 215 return; 216 217 /* 218 * flush_dcache_page() is called after set_pmd_at() to clear 219 * PG_arch_1 for the page on ARM64. The page flag isn't cleared 220 * when it's released and page allocation check will fail when 221 * the page is allocated again. For architectures other than ARM64, 222 * the unexpected overhead of cache flushing is acceptable. 223 */ 224 pr_debug("Validating PMD advanced\n"); 225 /* Align the address wrt HPAGE_PMD_SIZE */ 226 vaddr &= HPAGE_PMD_MASK; 227 228 pgtable_trans_huge_deposit(args->mm, args->pmdp, args->start_ptep); 229 230 pmd = pfn_pmd(args->pmd_pfn, args->page_prot); 231 set_pmd_at(args->mm, vaddr, args->pmdp, pmd); 232 flush_dcache_page(page); 233 pmdp_set_wrprotect(args->mm, vaddr, args->pmdp); 234 pmd = pmdp_get(args->pmdp); 235 WARN_ON(pmd_write(pmd)); 236 pmdp_huge_get_and_clear(args->mm, vaddr, args->pmdp); 237 pmd = pmdp_get(args->pmdp); 238 WARN_ON(!pmd_none(pmd)); 239 240 pmd = pfn_pmd(args->pmd_pfn, args->page_prot); 241 pmd = pmd_wrprotect(pmd); 242 pmd = pmd_mkclean(pmd); 243 set_pmd_at(args->mm, vaddr, args->pmdp, pmd); 244 flush_dcache_page(page); 245 pmd = pmd_mkwrite(pmd, args->vma); 246 pmd = pmd_mkdirty(pmd); 247 pmdp_set_access_flags(args->vma, vaddr, args->pmdp, pmd, 1); 248 pmd = pmdp_get(args->pmdp); 249 WARN_ON(!(pmd_write(pmd) && pmd_dirty(pmd))); 250 pmdp_huge_get_and_clear_full(args->vma, vaddr, args->pmdp, 1); 251 pmd = pmdp_get(args->pmdp); 252 WARN_ON(!pmd_none(pmd)); 253 254 pmd = pmd_mkhuge(pfn_pmd(args->pmd_pfn, args->page_prot)); 255 pmd = pmd_mkyoung(pmd); 256 set_pmd_at(args->mm, vaddr, args->pmdp, pmd); 257 flush_dcache_page(page); 258 pmdp_test_and_clear_young(args->vma, vaddr, args->pmdp); 259 pmd = pmdp_get(args->pmdp); 260 WARN_ON(pmd_young(pmd)); 261 262 /* Clear the pte entries */ 263 pmdp_huge_get_and_clear(args->mm, vaddr, args->pmdp); 264 pgtable_trans_huge_withdraw(args->mm, args->pmdp); 265 } 266 267 static void __init pmd_leaf_tests(struct pgtable_debug_args *args) 268 { 269 pmd_t pmd; 270 271 if (!has_transparent_hugepage()) 272 return; 273 274 pr_debug("Validating PMD leaf\n"); 275 pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot); 276 277 /* 278 * PMD based THP is a leaf entry. 279 */ 280 pmd = pmd_mkhuge(pmd); 281 WARN_ON(!pmd_leaf(pmd)); 282 } 283 284 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD 285 static void __init pud_basic_tests(struct pgtable_debug_args *args, int idx) 286 { 287 pgprot_t prot = vm_get_page_prot(idx); 288 unsigned long val = idx, *ptr = &val; 289 pud_t pud; 290 291 if (!has_transparent_pud_hugepage()) 292 return; 293 294 pr_debug("Validating PUD basic (%pGv)\n", ptr); 295 pud = pfn_pud(args->fixed_pud_pfn, prot); 296 297 /* 298 * This test needs to be executed after the given page table entry 299 * is created with pfn_pud() to make sure that vm_get_page_prot(idx) 300 * does not have the dirty bit enabled from the beginning. This is 301 * important for platforms like arm64 where (!PTE_RDONLY) indicate 302 * dirty bit being set. 303 */ 304 WARN_ON(pud_dirty(pud_wrprotect(pud))); 305 306 WARN_ON(!pud_same(pud, pud)); 307 WARN_ON(!pud_young(pud_mkyoung(pud_mkold(pud)))); 308 WARN_ON(!pud_dirty(pud_mkdirty(pud_mkclean(pud)))); 309 WARN_ON(pud_dirty(pud_mkclean(pud_mkdirty(pud)))); 310 WARN_ON(!pud_write(pud_mkwrite(pud_wrprotect(pud)))); 311 WARN_ON(pud_write(pud_wrprotect(pud_mkwrite(pud)))); 312 WARN_ON(pud_young(pud_mkold(pud_mkyoung(pud)))); 313 WARN_ON(pud_dirty(pud_wrprotect(pud_mkclean(pud)))); 314 WARN_ON(!pud_dirty(pud_wrprotect(pud_mkdirty(pud)))); 315 316 if (mm_pmd_folded(args->mm)) 317 return; 318 319 /* 320 * A huge page does not point to next level page table 321 * entry. Hence this must qualify as pud_bad(). 322 */ 323 WARN_ON(!pud_bad(pud_mkhuge(pud))); 324 } 325 326 static void __init pud_advanced_tests(struct pgtable_debug_args *args) 327 { 328 struct page *page; 329 unsigned long vaddr = args->vaddr; 330 pud_t pud; 331 332 if (!has_transparent_pud_hugepage()) 333 return; 334 335 page = (args->pud_pfn != ULONG_MAX) ? pfn_to_page(args->pud_pfn) : NULL; 336 if (!page) 337 return; 338 339 /* 340 * flush_dcache_page() is called after set_pud_at() to clear 341 * PG_arch_1 for the page on ARM64. The page flag isn't cleared 342 * when it's released and page allocation check will fail when 343 * the page is allocated again. For architectures other than ARM64, 344 * the unexpected overhead of cache flushing is acceptable. 345 */ 346 pr_debug("Validating PUD advanced\n"); 347 /* Align the address wrt HPAGE_PUD_SIZE */ 348 vaddr &= HPAGE_PUD_MASK; 349 350 pud = pfn_pud(args->pud_pfn, args->page_prot); 351 /* 352 * Some architectures have debug checks to make sure 353 * huge pud mapping are only found with devmap entries 354 * For now test with only devmap entries. 355 */ 356 pud = pud_mkdevmap(pud); 357 set_pud_at(args->mm, vaddr, args->pudp, pud); 358 flush_dcache_page(page); 359 pudp_set_wrprotect(args->mm, vaddr, args->pudp); 360 pud = pudp_get(args->pudp); 361 WARN_ON(pud_write(pud)); 362 363 #ifndef __PAGETABLE_PMD_FOLDED 364 pudp_huge_get_and_clear(args->mm, vaddr, args->pudp); 365 pud = pudp_get(args->pudp); 366 WARN_ON(!pud_none(pud)); 367 #endif /* __PAGETABLE_PMD_FOLDED */ 368 pud = pfn_pud(args->pud_pfn, args->page_prot); 369 pud = pud_mkdevmap(pud); 370 pud = pud_wrprotect(pud); 371 pud = pud_mkclean(pud); 372 set_pud_at(args->mm, vaddr, args->pudp, pud); 373 flush_dcache_page(page); 374 pud = pud_mkwrite(pud); 375 pud = pud_mkdirty(pud); 376 pudp_set_access_flags(args->vma, vaddr, args->pudp, pud, 1); 377 pud = pudp_get(args->pudp); 378 WARN_ON(!(pud_write(pud) && pud_dirty(pud))); 379 380 #ifndef __PAGETABLE_PMD_FOLDED 381 pudp_huge_get_and_clear_full(args->vma, vaddr, args->pudp, 1); 382 pud = pudp_get(args->pudp); 383 WARN_ON(!pud_none(pud)); 384 #endif /* __PAGETABLE_PMD_FOLDED */ 385 386 pud = pfn_pud(args->pud_pfn, args->page_prot); 387 pud = pud_mkdevmap(pud); 388 pud = pud_mkyoung(pud); 389 set_pud_at(args->mm, vaddr, args->pudp, pud); 390 flush_dcache_page(page); 391 pudp_test_and_clear_young(args->vma, vaddr, args->pudp); 392 pud = pudp_get(args->pudp); 393 WARN_ON(pud_young(pud)); 394 395 pudp_huge_get_and_clear(args->mm, vaddr, args->pudp); 396 } 397 398 static void __init pud_leaf_tests(struct pgtable_debug_args *args) 399 { 400 pud_t pud; 401 402 if (!has_transparent_pud_hugepage()) 403 return; 404 405 pr_debug("Validating PUD leaf\n"); 406 pud = pfn_pud(args->fixed_pud_pfn, args->page_prot); 407 /* 408 * PUD based THP is a leaf entry. 409 */ 410 pud = pud_mkhuge(pud); 411 WARN_ON(!pud_leaf(pud)); 412 } 413 #else /* !CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */ 414 static void __init pud_basic_tests(struct pgtable_debug_args *args, int idx) { } 415 static void __init pud_advanced_tests(struct pgtable_debug_args *args) { } 416 static void __init pud_leaf_tests(struct pgtable_debug_args *args) { } 417 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */ 418 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */ 419 static void __init pmd_basic_tests(struct pgtable_debug_args *args, int idx) { } 420 static void __init pud_basic_tests(struct pgtable_debug_args *args, int idx) { } 421 static void __init pmd_advanced_tests(struct pgtable_debug_args *args) { } 422 static void __init pud_advanced_tests(struct pgtable_debug_args *args) { } 423 static void __init pmd_leaf_tests(struct pgtable_debug_args *args) { } 424 static void __init pud_leaf_tests(struct pgtable_debug_args *args) { } 425 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 426 427 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP 428 static void __init pmd_huge_tests(struct pgtable_debug_args *args) 429 { 430 pmd_t pmd; 431 432 if (!arch_vmap_pmd_supported(args->page_prot) || 433 args->fixed_alignment < PMD_SIZE) 434 return; 435 436 pr_debug("Validating PMD huge\n"); 437 /* 438 * X86 defined pmd_set_huge() verifies that the given 439 * PMD is not a populated non-leaf entry. 440 */ 441 WRITE_ONCE(*args->pmdp, __pmd(0)); 442 WARN_ON(!pmd_set_huge(args->pmdp, __pfn_to_phys(args->fixed_pmd_pfn), args->page_prot)); 443 WARN_ON(!pmd_clear_huge(args->pmdp)); 444 pmd = pmdp_get(args->pmdp); 445 WARN_ON(!pmd_none(pmd)); 446 } 447 448 static void __init pud_huge_tests(struct pgtable_debug_args *args) 449 { 450 pud_t pud; 451 452 if (!arch_vmap_pud_supported(args->page_prot) || 453 args->fixed_alignment < PUD_SIZE) 454 return; 455 456 pr_debug("Validating PUD huge\n"); 457 /* 458 * X86 defined pud_set_huge() verifies that the given 459 * PUD is not a populated non-leaf entry. 460 */ 461 WRITE_ONCE(*args->pudp, __pud(0)); 462 WARN_ON(!pud_set_huge(args->pudp, __pfn_to_phys(args->fixed_pud_pfn), args->page_prot)); 463 WARN_ON(!pud_clear_huge(args->pudp)); 464 pud = pudp_get(args->pudp); 465 WARN_ON(!pud_none(pud)); 466 } 467 #else /* !CONFIG_HAVE_ARCH_HUGE_VMAP */ 468 static void __init pmd_huge_tests(struct pgtable_debug_args *args) { } 469 static void __init pud_huge_tests(struct pgtable_debug_args *args) { } 470 #endif /* CONFIG_HAVE_ARCH_HUGE_VMAP */ 471 472 static void __init p4d_basic_tests(struct pgtable_debug_args *args) 473 { 474 p4d_t p4d; 475 476 pr_debug("Validating P4D basic\n"); 477 memset(&p4d, RANDOM_NZVALUE, sizeof(p4d_t)); 478 WARN_ON(!p4d_same(p4d, p4d)); 479 } 480 481 static void __init pgd_basic_tests(struct pgtable_debug_args *args) 482 { 483 pgd_t pgd; 484 485 pr_debug("Validating PGD basic\n"); 486 memset(&pgd, RANDOM_NZVALUE, sizeof(pgd_t)); 487 WARN_ON(!pgd_same(pgd, pgd)); 488 } 489 490 #ifndef __PAGETABLE_PUD_FOLDED 491 static void __init pud_clear_tests(struct pgtable_debug_args *args) 492 { 493 pud_t pud = pudp_get(args->pudp); 494 495 if (mm_pmd_folded(args->mm)) 496 return; 497 498 pr_debug("Validating PUD clear\n"); 499 WARN_ON(pud_none(pud)); 500 pud_clear(args->pudp); 501 pud = pudp_get(args->pudp); 502 WARN_ON(!pud_none(pud)); 503 } 504 505 static void __init pud_populate_tests(struct pgtable_debug_args *args) 506 { 507 pud_t pud; 508 509 if (mm_pmd_folded(args->mm)) 510 return; 511 512 pr_debug("Validating PUD populate\n"); 513 /* 514 * This entry points to next level page table page. 515 * Hence this must not qualify as pud_bad(). 516 */ 517 pud_populate(args->mm, args->pudp, args->start_pmdp); 518 pud = pudp_get(args->pudp); 519 WARN_ON(pud_bad(pud)); 520 } 521 #else /* !__PAGETABLE_PUD_FOLDED */ 522 static void __init pud_clear_tests(struct pgtable_debug_args *args) { } 523 static void __init pud_populate_tests(struct pgtable_debug_args *args) { } 524 #endif /* PAGETABLE_PUD_FOLDED */ 525 526 #ifndef __PAGETABLE_P4D_FOLDED 527 static void __init p4d_clear_tests(struct pgtable_debug_args *args) 528 { 529 p4d_t p4d = p4dp_get(args->p4dp); 530 531 if (mm_pud_folded(args->mm)) 532 return; 533 534 pr_debug("Validating P4D clear\n"); 535 WARN_ON(p4d_none(p4d)); 536 p4d_clear(args->p4dp); 537 p4d = p4dp_get(args->p4dp); 538 WARN_ON(!p4d_none(p4d)); 539 } 540 541 static void __init p4d_populate_tests(struct pgtable_debug_args *args) 542 { 543 p4d_t p4d; 544 545 if (mm_pud_folded(args->mm)) 546 return; 547 548 pr_debug("Validating P4D populate\n"); 549 /* 550 * This entry points to next level page table page. 551 * Hence this must not qualify as p4d_bad(). 552 */ 553 pud_clear(args->pudp); 554 p4d_clear(args->p4dp); 555 p4d_populate(args->mm, args->p4dp, args->start_pudp); 556 p4d = p4dp_get(args->p4dp); 557 WARN_ON(p4d_bad(p4d)); 558 } 559 560 static void __init pgd_clear_tests(struct pgtable_debug_args *args) 561 { 562 pgd_t pgd = pgdp_get(args->pgdp); 563 564 if (mm_p4d_folded(args->mm)) 565 return; 566 567 pr_debug("Validating PGD clear\n"); 568 WARN_ON(pgd_none(pgd)); 569 pgd_clear(args->pgdp); 570 pgd = pgdp_get(args->pgdp); 571 WARN_ON(!pgd_none(pgd)); 572 } 573 574 static void __init pgd_populate_tests(struct pgtable_debug_args *args) 575 { 576 pgd_t pgd; 577 578 if (mm_p4d_folded(args->mm)) 579 return; 580 581 pr_debug("Validating PGD populate\n"); 582 /* 583 * This entry points to next level page table page. 584 * Hence this must not qualify as pgd_bad(). 585 */ 586 p4d_clear(args->p4dp); 587 pgd_clear(args->pgdp); 588 pgd_populate(args->mm, args->pgdp, args->start_p4dp); 589 pgd = pgdp_get(args->pgdp); 590 WARN_ON(pgd_bad(pgd)); 591 } 592 #else /* !__PAGETABLE_P4D_FOLDED */ 593 static void __init p4d_clear_tests(struct pgtable_debug_args *args) { } 594 static void __init pgd_clear_tests(struct pgtable_debug_args *args) { } 595 static void __init p4d_populate_tests(struct pgtable_debug_args *args) { } 596 static void __init pgd_populate_tests(struct pgtable_debug_args *args) { } 597 #endif /* PAGETABLE_P4D_FOLDED */ 598 599 static void __init pte_clear_tests(struct pgtable_debug_args *args) 600 { 601 struct page *page; 602 pte_t pte = pfn_pte(args->pte_pfn, args->page_prot); 603 604 page = (args->pte_pfn != ULONG_MAX) ? pfn_to_page(args->pte_pfn) : NULL; 605 if (!page) 606 return; 607 608 /* 609 * flush_dcache_page() is called after set_pte_at() to clear 610 * PG_arch_1 for the page on ARM64. The page flag isn't cleared 611 * when it's released and page allocation check will fail when 612 * the page is allocated again. For architectures other than ARM64, 613 * the unexpected overhead of cache flushing is acceptable. 614 */ 615 pr_debug("Validating PTE clear\n"); 616 if (WARN_ON(!args->ptep)) 617 return; 618 619 set_pte_at(args->mm, args->vaddr, args->ptep, pte); 620 WARN_ON(pte_none(pte)); 621 flush_dcache_page(page); 622 barrier(); 623 ptep_clear(args->mm, args->vaddr, args->ptep); 624 pte = ptep_get(args->ptep); 625 WARN_ON(!pte_none(pte)); 626 } 627 628 static void __init pmd_clear_tests(struct pgtable_debug_args *args) 629 { 630 pmd_t pmd = pmdp_get(args->pmdp); 631 632 pr_debug("Validating PMD clear\n"); 633 WARN_ON(pmd_none(pmd)); 634 pmd_clear(args->pmdp); 635 pmd = pmdp_get(args->pmdp); 636 WARN_ON(!pmd_none(pmd)); 637 } 638 639 static void __init pmd_populate_tests(struct pgtable_debug_args *args) 640 { 641 pmd_t pmd; 642 643 pr_debug("Validating PMD populate\n"); 644 /* 645 * This entry points to next level page table page. 646 * Hence this must not qualify as pmd_bad(). 647 */ 648 pmd_populate(args->mm, args->pmdp, args->start_ptep); 649 pmd = pmdp_get(args->pmdp); 650 WARN_ON(pmd_bad(pmd)); 651 } 652 653 static void __init pte_special_tests(struct pgtable_debug_args *args) 654 { 655 pte_t pte = pfn_pte(args->fixed_pte_pfn, args->page_prot); 656 657 if (!IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL)) 658 return; 659 660 pr_debug("Validating PTE special\n"); 661 WARN_ON(!pte_special(pte_mkspecial(pte))); 662 } 663 664 static void __init pte_protnone_tests(struct pgtable_debug_args *args) 665 { 666 pte_t pte = pfn_pte(args->fixed_pte_pfn, args->page_prot_none); 667 668 if (!IS_ENABLED(CONFIG_NUMA_BALANCING)) 669 return; 670 671 pr_debug("Validating PTE protnone\n"); 672 WARN_ON(!pte_protnone(pte)); 673 WARN_ON(!pte_present(pte)); 674 } 675 676 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 677 static void __init pmd_protnone_tests(struct pgtable_debug_args *args) 678 { 679 pmd_t pmd; 680 681 if (!IS_ENABLED(CONFIG_NUMA_BALANCING)) 682 return; 683 684 if (!has_transparent_hugepage()) 685 return; 686 687 pr_debug("Validating PMD protnone\n"); 688 pmd = pmd_mkhuge(pfn_pmd(args->fixed_pmd_pfn, args->page_prot_none)); 689 WARN_ON(!pmd_protnone(pmd)); 690 WARN_ON(!pmd_present(pmd)); 691 } 692 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */ 693 static void __init pmd_protnone_tests(struct pgtable_debug_args *args) { } 694 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 695 696 #ifdef CONFIG_ARCH_HAS_PTE_DEVMAP 697 static void __init pte_devmap_tests(struct pgtable_debug_args *args) 698 { 699 pte_t pte = pfn_pte(args->fixed_pte_pfn, args->page_prot); 700 701 pr_debug("Validating PTE devmap\n"); 702 WARN_ON(!pte_devmap(pte_mkdevmap(pte))); 703 } 704 705 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 706 static void __init pmd_devmap_tests(struct pgtable_debug_args *args) 707 { 708 pmd_t pmd; 709 710 if (!has_transparent_hugepage()) 711 return; 712 713 pr_debug("Validating PMD devmap\n"); 714 pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot); 715 WARN_ON(!pmd_devmap(pmd_mkdevmap(pmd))); 716 } 717 718 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD 719 static void __init pud_devmap_tests(struct pgtable_debug_args *args) 720 { 721 pud_t pud; 722 723 if (!has_transparent_pud_hugepage()) 724 return; 725 726 pr_debug("Validating PUD devmap\n"); 727 pud = pfn_pud(args->fixed_pud_pfn, args->page_prot); 728 WARN_ON(!pud_devmap(pud_mkdevmap(pud))); 729 } 730 #else /* !CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */ 731 static void __init pud_devmap_tests(struct pgtable_debug_args *args) { } 732 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */ 733 #else /* CONFIG_TRANSPARENT_HUGEPAGE */ 734 static void __init pmd_devmap_tests(struct pgtable_debug_args *args) { } 735 static void __init pud_devmap_tests(struct pgtable_debug_args *args) { } 736 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 737 #else 738 static void __init pte_devmap_tests(struct pgtable_debug_args *args) { } 739 static void __init pmd_devmap_tests(struct pgtable_debug_args *args) { } 740 static void __init pud_devmap_tests(struct pgtable_debug_args *args) { } 741 #endif /* CONFIG_ARCH_HAS_PTE_DEVMAP */ 742 743 static void __init pte_soft_dirty_tests(struct pgtable_debug_args *args) 744 { 745 pte_t pte = pfn_pte(args->fixed_pte_pfn, args->page_prot); 746 747 if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY)) 748 return; 749 750 pr_debug("Validating PTE soft dirty\n"); 751 WARN_ON(!pte_soft_dirty(pte_mksoft_dirty(pte))); 752 WARN_ON(pte_soft_dirty(pte_clear_soft_dirty(pte))); 753 } 754 755 static void __init pte_swap_soft_dirty_tests(struct pgtable_debug_args *args) 756 { 757 pte_t pte = pfn_pte(args->fixed_pte_pfn, args->page_prot); 758 759 if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY)) 760 return; 761 762 pr_debug("Validating PTE swap soft dirty\n"); 763 WARN_ON(!pte_swp_soft_dirty(pte_swp_mksoft_dirty(pte))); 764 WARN_ON(pte_swp_soft_dirty(pte_swp_clear_soft_dirty(pte))); 765 } 766 767 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 768 static void __init pmd_soft_dirty_tests(struct pgtable_debug_args *args) 769 { 770 pmd_t pmd; 771 772 if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY)) 773 return; 774 775 if (!has_transparent_hugepage()) 776 return; 777 778 pr_debug("Validating PMD soft dirty\n"); 779 pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot); 780 WARN_ON(!pmd_soft_dirty(pmd_mksoft_dirty(pmd))); 781 WARN_ON(pmd_soft_dirty(pmd_clear_soft_dirty(pmd))); 782 } 783 784 static void __init pmd_swap_soft_dirty_tests(struct pgtable_debug_args *args) 785 { 786 pmd_t pmd; 787 788 if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) || 789 !IS_ENABLED(CONFIG_ARCH_ENABLE_THP_MIGRATION)) 790 return; 791 792 if (!has_transparent_hugepage()) 793 return; 794 795 pr_debug("Validating PMD swap soft dirty\n"); 796 pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot); 797 WARN_ON(!pmd_swp_soft_dirty(pmd_swp_mksoft_dirty(pmd))); 798 WARN_ON(pmd_swp_soft_dirty(pmd_swp_clear_soft_dirty(pmd))); 799 } 800 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */ 801 static void __init pmd_soft_dirty_tests(struct pgtable_debug_args *args) { } 802 static void __init pmd_swap_soft_dirty_tests(struct pgtable_debug_args *args) { } 803 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 804 805 static void __init pte_swap_exclusive_tests(struct pgtable_debug_args *args) 806 { 807 unsigned long max_swap_offset; 808 swp_entry_t entry, entry2; 809 pte_t pte; 810 811 pr_debug("Validating PTE swap exclusive\n"); 812 813 /* See generic_max_swapfile_size(): probe the maximum offset */ 814 max_swap_offset = swp_offset(pte_to_swp_entry(swp_entry_to_pte(swp_entry(0, ~0UL)))); 815 816 /* Create a swp entry with all possible bits set */ 817 entry = swp_entry((1 << MAX_SWAPFILES_SHIFT) - 1, max_swap_offset); 818 819 pte = swp_entry_to_pte(entry); 820 WARN_ON(pte_swp_exclusive(pte)); 821 WARN_ON(!is_swap_pte(pte)); 822 entry2 = pte_to_swp_entry(pte); 823 WARN_ON(memcmp(&entry, &entry2, sizeof(entry))); 824 825 pte = pte_swp_mkexclusive(pte); 826 WARN_ON(!pte_swp_exclusive(pte)); 827 WARN_ON(!is_swap_pte(pte)); 828 WARN_ON(pte_swp_soft_dirty(pte)); 829 entry2 = pte_to_swp_entry(pte); 830 WARN_ON(memcmp(&entry, &entry2, sizeof(entry))); 831 832 pte = pte_swp_clear_exclusive(pte); 833 WARN_ON(pte_swp_exclusive(pte)); 834 WARN_ON(!is_swap_pte(pte)); 835 entry2 = pte_to_swp_entry(pte); 836 WARN_ON(memcmp(&entry, &entry2, sizeof(entry))); 837 } 838 839 static void __init pte_swap_tests(struct pgtable_debug_args *args) 840 { 841 swp_entry_t swp; 842 pte_t pte; 843 844 pr_debug("Validating PTE swap\n"); 845 pte = pfn_pte(args->fixed_pte_pfn, args->page_prot); 846 swp = __pte_to_swp_entry(pte); 847 pte = __swp_entry_to_pte(swp); 848 WARN_ON(args->fixed_pte_pfn != pte_pfn(pte)); 849 } 850 851 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION 852 static void __init pmd_swap_tests(struct pgtable_debug_args *args) 853 { 854 swp_entry_t swp; 855 pmd_t pmd; 856 857 if (!has_transparent_hugepage()) 858 return; 859 860 pr_debug("Validating PMD swap\n"); 861 pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot); 862 swp = __pmd_to_swp_entry(pmd); 863 pmd = __swp_entry_to_pmd(swp); 864 WARN_ON(args->fixed_pmd_pfn != pmd_pfn(pmd)); 865 } 866 #else /* !CONFIG_ARCH_ENABLE_THP_MIGRATION */ 867 static void __init pmd_swap_tests(struct pgtable_debug_args *args) { } 868 #endif /* CONFIG_ARCH_ENABLE_THP_MIGRATION */ 869 870 static void __init swap_migration_tests(struct pgtable_debug_args *args) 871 { 872 struct page *page; 873 swp_entry_t swp; 874 875 if (!IS_ENABLED(CONFIG_MIGRATION)) 876 return; 877 878 /* 879 * swap_migration_tests() requires a dedicated page as it needs to 880 * be locked before creating a migration entry from it. Locking the 881 * page that actually maps kernel text ('start_kernel') can be real 882 * problematic. Lets use the allocated page explicitly for this 883 * purpose. 884 */ 885 page = (args->pte_pfn != ULONG_MAX) ? pfn_to_page(args->pte_pfn) : NULL; 886 if (!page) 887 return; 888 889 pr_debug("Validating swap migration\n"); 890 891 /* 892 * make_[readable|writable]_migration_entry() expects given page to 893 * be locked, otherwise it stumbles upon a BUG_ON(). 894 */ 895 __SetPageLocked(page); 896 swp = make_writable_migration_entry(page_to_pfn(page)); 897 WARN_ON(!is_migration_entry(swp)); 898 WARN_ON(!is_writable_migration_entry(swp)); 899 900 swp = make_readable_migration_entry(swp_offset(swp)); 901 WARN_ON(!is_migration_entry(swp)); 902 WARN_ON(is_writable_migration_entry(swp)); 903 904 swp = make_readable_migration_entry(page_to_pfn(page)); 905 WARN_ON(!is_migration_entry(swp)); 906 WARN_ON(is_writable_migration_entry(swp)); 907 __ClearPageLocked(page); 908 } 909 910 #ifdef CONFIG_HUGETLB_PAGE 911 static void __init hugetlb_basic_tests(struct pgtable_debug_args *args) 912 { 913 pte_t pte; 914 915 pr_debug("Validating HugeTLB basic\n"); 916 pte = pfn_pte(args->fixed_pmd_pfn, args->page_prot); 917 pte = arch_make_huge_pte(pte, PMD_SHIFT, VM_ACCESS_FLAGS); 918 919 #ifdef CONFIG_ARCH_WANT_GENERAL_HUGETLB 920 WARN_ON(!pte_huge(pte)); 921 #endif 922 WARN_ON(!huge_pte_dirty(huge_pte_mkdirty(pte))); 923 WARN_ON(!huge_pte_write(huge_pte_mkwrite(huge_pte_wrprotect(pte)))); 924 WARN_ON(huge_pte_write(huge_pte_wrprotect(huge_pte_mkwrite(pte)))); 925 } 926 #else /* !CONFIG_HUGETLB_PAGE */ 927 static void __init hugetlb_basic_tests(struct pgtable_debug_args *args) { } 928 #endif /* CONFIG_HUGETLB_PAGE */ 929 930 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 931 static void __init pmd_thp_tests(struct pgtable_debug_args *args) 932 { 933 pmd_t pmd; 934 935 if (!has_transparent_hugepage()) 936 return; 937 938 pr_debug("Validating PMD based THP\n"); 939 /* 940 * pmd_trans_huge() and pmd_present() must return positive after 941 * MMU invalidation with pmd_mkinvalid(). This behavior is an 942 * optimization for transparent huge page. pmd_trans_huge() must 943 * be true if pmd_page() returns a valid THP to avoid taking the 944 * pmd_lock when others walk over non transhuge pmds (i.e. there 945 * are no THP allocated). Especially when splitting a THP and 946 * removing the present bit from the pmd, pmd_trans_huge() still 947 * needs to return true. pmd_present() should be true whenever 948 * pmd_trans_huge() returns true. 949 */ 950 pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot); 951 WARN_ON(!pmd_trans_huge(pmd_mkhuge(pmd))); 952 953 #ifndef __HAVE_ARCH_PMDP_INVALIDATE 954 WARN_ON(!pmd_trans_huge(pmd_mkinvalid(pmd_mkhuge(pmd)))); 955 WARN_ON(!pmd_present(pmd_mkinvalid(pmd_mkhuge(pmd)))); 956 WARN_ON(!pmd_leaf(pmd_mkinvalid(pmd_mkhuge(pmd)))); 957 #endif /* __HAVE_ARCH_PMDP_INVALIDATE */ 958 } 959 960 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD 961 static void __init pud_thp_tests(struct pgtable_debug_args *args) 962 { 963 pud_t pud; 964 965 if (!has_transparent_pud_hugepage()) 966 return; 967 968 pr_debug("Validating PUD based THP\n"); 969 pud = pfn_pud(args->fixed_pud_pfn, args->page_prot); 970 WARN_ON(!pud_trans_huge(pud_mkhuge(pud))); 971 972 /* 973 * pud_mkinvalid() has been dropped for now. Enable back 974 * these tests when it comes back with a modified pud_present(). 975 * 976 * WARN_ON(!pud_trans_huge(pud_mkinvalid(pud_mkhuge(pud)))); 977 * WARN_ON(!pud_present(pud_mkinvalid(pud_mkhuge(pud)))); 978 */ 979 } 980 #else /* !CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */ 981 static void __init pud_thp_tests(struct pgtable_debug_args *args) { } 982 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */ 983 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */ 984 static void __init pmd_thp_tests(struct pgtable_debug_args *args) { } 985 static void __init pud_thp_tests(struct pgtable_debug_args *args) { } 986 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 987 988 static unsigned long __init get_random_vaddr(void) 989 { 990 unsigned long random_vaddr, random_pages, total_user_pages; 991 992 total_user_pages = (TASK_SIZE - FIRST_USER_ADDRESS) / PAGE_SIZE; 993 994 random_pages = get_random_long() % total_user_pages; 995 random_vaddr = FIRST_USER_ADDRESS + random_pages * PAGE_SIZE; 996 997 return random_vaddr; 998 } 999 1000 static void __init destroy_args(struct pgtable_debug_args *args) 1001 { 1002 struct page *page = NULL; 1003 1004 /* Free (huge) page */ 1005 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && 1006 has_transparent_pud_hugepage() && 1007 args->pud_pfn != ULONG_MAX) { 1008 if (args->is_contiguous_page) { 1009 free_contig_range(args->pud_pfn, 1010 (1 << (HPAGE_PUD_SHIFT - PAGE_SHIFT))); 1011 } else { 1012 page = pfn_to_page(args->pud_pfn); 1013 __free_pages(page, HPAGE_PUD_SHIFT - PAGE_SHIFT); 1014 } 1015 1016 args->pud_pfn = ULONG_MAX; 1017 args->pmd_pfn = ULONG_MAX; 1018 args->pte_pfn = ULONG_MAX; 1019 } 1020 1021 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && 1022 has_transparent_hugepage() && 1023 args->pmd_pfn != ULONG_MAX) { 1024 if (args->is_contiguous_page) { 1025 free_contig_range(args->pmd_pfn, (1 << HPAGE_PMD_ORDER)); 1026 } else { 1027 page = pfn_to_page(args->pmd_pfn); 1028 __free_pages(page, HPAGE_PMD_ORDER); 1029 } 1030 1031 args->pmd_pfn = ULONG_MAX; 1032 args->pte_pfn = ULONG_MAX; 1033 } 1034 1035 if (args->pte_pfn != ULONG_MAX) { 1036 page = pfn_to_page(args->pte_pfn); 1037 __free_page(page); 1038 1039 args->pte_pfn = ULONG_MAX; 1040 } 1041 1042 /* Free page table entries */ 1043 if (args->start_ptep) { 1044 pte_free(args->mm, args->start_ptep); 1045 mm_dec_nr_ptes(args->mm); 1046 } 1047 1048 if (args->start_pmdp) { 1049 pmd_free(args->mm, args->start_pmdp); 1050 mm_dec_nr_pmds(args->mm); 1051 } 1052 1053 if (args->start_pudp) { 1054 pud_free(args->mm, args->start_pudp); 1055 mm_dec_nr_puds(args->mm); 1056 } 1057 1058 if (args->start_p4dp) 1059 p4d_free(args->mm, args->start_p4dp); 1060 1061 /* Free vma and mm struct */ 1062 if (args->vma) 1063 vm_area_free(args->vma); 1064 1065 if (args->mm) 1066 mmdrop(args->mm); 1067 } 1068 1069 static struct page * __init 1070 debug_vm_pgtable_alloc_huge_page(struct pgtable_debug_args *args, int order) 1071 { 1072 struct page *page = NULL; 1073 1074 #ifdef CONFIG_CONTIG_ALLOC 1075 if (order > MAX_PAGE_ORDER) { 1076 page = alloc_contig_pages((1 << order), GFP_KERNEL, 1077 first_online_node, NULL); 1078 if (page) { 1079 args->is_contiguous_page = true; 1080 return page; 1081 } 1082 } 1083 #endif 1084 1085 if (order <= MAX_PAGE_ORDER) 1086 page = alloc_pages(GFP_KERNEL, order); 1087 1088 return page; 1089 } 1090 1091 /* 1092 * Check if a physical memory range described by <pstart, pend> contains 1093 * an area that is of size psize, and aligned to psize. 1094 * 1095 * Don't use address 0, an all-zeroes physical address might mask bugs, and 1096 * it's not used on x86. 1097 */ 1098 static void __init phys_align_check(phys_addr_t pstart, 1099 phys_addr_t pend, unsigned long psize, 1100 phys_addr_t *physp, unsigned long *alignp) 1101 { 1102 phys_addr_t aligned_start, aligned_end; 1103 1104 if (pstart == 0) 1105 pstart = PAGE_SIZE; 1106 1107 aligned_start = ALIGN(pstart, psize); 1108 aligned_end = aligned_start + psize; 1109 1110 if (aligned_end > aligned_start && aligned_end <= pend) { 1111 *alignp = psize; 1112 *physp = aligned_start; 1113 } 1114 } 1115 1116 static void __init init_fixed_pfns(struct pgtable_debug_args *args) 1117 { 1118 u64 idx; 1119 phys_addr_t phys, pstart, pend; 1120 1121 /* 1122 * Initialize the fixed pfns. To do this, try to find a 1123 * valid physical range, preferably aligned to PUD_SIZE, 1124 * but settling for aligned to PMD_SIZE as a fallback. If 1125 * neither of those is found, use the physical address of 1126 * the start_kernel symbol. 1127 * 1128 * The memory doesn't need to be allocated, it just needs to exist 1129 * as usable memory. It won't be touched. 1130 * 1131 * The alignment is recorded, and can be checked to see if we 1132 * can run the tests that require an actual valid physical 1133 * address range on some architectures ({pmd,pud}_huge_test 1134 * on x86). 1135 */ 1136 1137 phys = __pa_symbol(&start_kernel); 1138 args->fixed_alignment = PAGE_SIZE; 1139 1140 for_each_mem_range(idx, &pstart, &pend) { 1141 /* First check for a PUD-aligned area */ 1142 phys_align_check(pstart, pend, PUD_SIZE, &phys, 1143 &args->fixed_alignment); 1144 1145 /* If a PUD-aligned area is found, we're done */ 1146 if (args->fixed_alignment == PUD_SIZE) 1147 break; 1148 1149 /* 1150 * If no PMD-aligned area found yet, check for one, 1151 * but continue the loop to look for a PUD-aligned area. 1152 */ 1153 if (args->fixed_alignment < PMD_SIZE) 1154 phys_align_check(pstart, pend, PMD_SIZE, &phys, 1155 &args->fixed_alignment); 1156 } 1157 1158 args->fixed_pgd_pfn = __phys_to_pfn(phys & PGDIR_MASK); 1159 args->fixed_p4d_pfn = __phys_to_pfn(phys & P4D_MASK); 1160 args->fixed_pud_pfn = __phys_to_pfn(phys & PUD_MASK); 1161 args->fixed_pmd_pfn = __phys_to_pfn(phys & PMD_MASK); 1162 args->fixed_pte_pfn = __phys_to_pfn(phys & PAGE_MASK); 1163 WARN_ON(!pfn_valid(args->fixed_pte_pfn)); 1164 } 1165 1166 1167 static int __init init_args(struct pgtable_debug_args *args) 1168 { 1169 struct page *page = NULL; 1170 int ret = 0; 1171 1172 /* 1173 * Initialize the debugging data. 1174 * 1175 * vm_get_page_prot(VM_NONE) or vm_get_page_prot(VM_SHARED|VM_NONE) 1176 * will help create page table entries with PROT_NONE permission as 1177 * required for pxx_protnone_tests(). 1178 */ 1179 memset(args, 0, sizeof(*args)); 1180 args->vaddr = get_random_vaddr(); 1181 args->page_prot = vm_get_page_prot(VM_ACCESS_FLAGS); 1182 args->page_prot_none = vm_get_page_prot(VM_NONE); 1183 args->is_contiguous_page = false; 1184 args->pud_pfn = ULONG_MAX; 1185 args->pmd_pfn = ULONG_MAX; 1186 args->pte_pfn = ULONG_MAX; 1187 args->fixed_pgd_pfn = ULONG_MAX; 1188 args->fixed_p4d_pfn = ULONG_MAX; 1189 args->fixed_pud_pfn = ULONG_MAX; 1190 args->fixed_pmd_pfn = ULONG_MAX; 1191 args->fixed_pte_pfn = ULONG_MAX; 1192 1193 /* Allocate mm and vma */ 1194 args->mm = mm_alloc(); 1195 if (!args->mm) { 1196 pr_err("Failed to allocate mm struct\n"); 1197 ret = -ENOMEM; 1198 goto error; 1199 } 1200 1201 args->vma = vm_area_alloc(args->mm); 1202 if (!args->vma) { 1203 pr_err("Failed to allocate vma\n"); 1204 ret = -ENOMEM; 1205 goto error; 1206 } 1207 1208 /* 1209 * Allocate page table entries. They will be modified in the tests. 1210 * Lets save the page table entries so that they can be released 1211 * when the tests are completed. 1212 */ 1213 args->pgdp = pgd_offset(args->mm, args->vaddr); 1214 args->p4dp = p4d_alloc(args->mm, args->pgdp, args->vaddr); 1215 if (!args->p4dp) { 1216 pr_err("Failed to allocate p4d entries\n"); 1217 ret = -ENOMEM; 1218 goto error; 1219 } 1220 args->start_p4dp = p4d_offset(args->pgdp, 0UL); 1221 WARN_ON(!args->start_p4dp); 1222 1223 args->pudp = pud_alloc(args->mm, args->p4dp, args->vaddr); 1224 if (!args->pudp) { 1225 pr_err("Failed to allocate pud entries\n"); 1226 ret = -ENOMEM; 1227 goto error; 1228 } 1229 args->start_pudp = pud_offset(args->p4dp, 0UL); 1230 WARN_ON(!args->start_pudp); 1231 1232 args->pmdp = pmd_alloc(args->mm, args->pudp, args->vaddr); 1233 if (!args->pmdp) { 1234 pr_err("Failed to allocate pmd entries\n"); 1235 ret = -ENOMEM; 1236 goto error; 1237 } 1238 args->start_pmdp = pmd_offset(args->pudp, 0UL); 1239 WARN_ON(!args->start_pmdp); 1240 1241 if (pte_alloc(args->mm, args->pmdp)) { 1242 pr_err("Failed to allocate pte entries\n"); 1243 ret = -ENOMEM; 1244 goto error; 1245 } 1246 args->start_ptep = pmd_pgtable(pmdp_get(args->pmdp)); 1247 WARN_ON(!args->start_ptep); 1248 1249 init_fixed_pfns(args); 1250 1251 /* 1252 * Allocate (huge) pages because some of the tests need to access 1253 * the data in the pages. The corresponding tests will be skipped 1254 * if we fail to allocate (huge) pages. 1255 */ 1256 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && 1257 has_transparent_pud_hugepage()) { 1258 page = debug_vm_pgtable_alloc_huge_page(args, 1259 HPAGE_PUD_SHIFT - PAGE_SHIFT); 1260 if (page) { 1261 args->pud_pfn = page_to_pfn(page); 1262 args->pmd_pfn = args->pud_pfn; 1263 args->pte_pfn = args->pud_pfn; 1264 return 0; 1265 } 1266 } 1267 1268 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && 1269 has_transparent_hugepage()) { 1270 page = debug_vm_pgtable_alloc_huge_page(args, HPAGE_PMD_ORDER); 1271 if (page) { 1272 args->pmd_pfn = page_to_pfn(page); 1273 args->pte_pfn = args->pmd_pfn; 1274 return 0; 1275 } 1276 } 1277 1278 page = alloc_page(GFP_KERNEL); 1279 if (page) 1280 args->pte_pfn = page_to_pfn(page); 1281 1282 return 0; 1283 1284 error: 1285 destroy_args(args); 1286 return ret; 1287 } 1288 1289 static int __init debug_vm_pgtable(void) 1290 { 1291 struct pgtable_debug_args args; 1292 spinlock_t *ptl = NULL; 1293 int idx, ret; 1294 1295 pr_info("Validating architecture page table helpers\n"); 1296 ret = init_args(&args); 1297 if (ret) 1298 return ret; 1299 1300 /* 1301 * Iterate over each possible vm_flags to make sure that all 1302 * the basic page table transformation validations just hold 1303 * true irrespective of the starting protection value for a 1304 * given page table entry. 1305 * 1306 * Protection based vm_flags combinations are always linear 1307 * and increasing i.e starting from VM_NONE and going up to 1308 * (VM_SHARED | READ | WRITE | EXEC). 1309 */ 1310 #define VM_FLAGS_START (VM_NONE) 1311 #define VM_FLAGS_END (VM_SHARED | VM_EXEC | VM_WRITE | VM_READ) 1312 1313 for (idx = VM_FLAGS_START; idx <= VM_FLAGS_END; idx++) { 1314 pte_basic_tests(&args, idx); 1315 pmd_basic_tests(&args, idx); 1316 pud_basic_tests(&args, idx); 1317 } 1318 1319 /* 1320 * Both P4D and PGD level tests are very basic which do not 1321 * involve creating page table entries from the protection 1322 * value and the given pfn. Hence just keep them out from 1323 * the above iteration for now to save some test execution 1324 * time. 1325 */ 1326 p4d_basic_tests(&args); 1327 pgd_basic_tests(&args); 1328 1329 pmd_leaf_tests(&args); 1330 pud_leaf_tests(&args); 1331 1332 pte_special_tests(&args); 1333 pte_protnone_tests(&args); 1334 pmd_protnone_tests(&args); 1335 1336 pte_devmap_tests(&args); 1337 pmd_devmap_tests(&args); 1338 pud_devmap_tests(&args); 1339 1340 pte_soft_dirty_tests(&args); 1341 pmd_soft_dirty_tests(&args); 1342 pte_swap_soft_dirty_tests(&args); 1343 pmd_swap_soft_dirty_tests(&args); 1344 1345 pte_swap_exclusive_tests(&args); 1346 1347 pte_swap_tests(&args); 1348 pmd_swap_tests(&args); 1349 1350 swap_migration_tests(&args); 1351 1352 pmd_thp_tests(&args); 1353 pud_thp_tests(&args); 1354 1355 hugetlb_basic_tests(&args); 1356 1357 /* 1358 * Page table modifying tests. They need to hold 1359 * proper page table lock. 1360 */ 1361 1362 args.ptep = pte_offset_map_lock(args.mm, args.pmdp, args.vaddr, &ptl); 1363 pte_clear_tests(&args); 1364 pte_advanced_tests(&args); 1365 if (args.ptep) 1366 pte_unmap_unlock(args.ptep, ptl); 1367 1368 ptl = pmd_lock(args.mm, args.pmdp); 1369 pmd_clear_tests(&args); 1370 pmd_advanced_tests(&args); 1371 pmd_huge_tests(&args); 1372 pmd_populate_tests(&args); 1373 spin_unlock(ptl); 1374 1375 ptl = pud_lock(args.mm, args.pudp); 1376 pud_clear_tests(&args); 1377 pud_advanced_tests(&args); 1378 pud_huge_tests(&args); 1379 pud_populate_tests(&args); 1380 spin_unlock(ptl); 1381 1382 spin_lock(&(args.mm->page_table_lock)); 1383 p4d_clear_tests(&args); 1384 pgd_clear_tests(&args); 1385 p4d_populate_tests(&args); 1386 pgd_populate_tests(&args); 1387 spin_unlock(&(args.mm->page_table_lock)); 1388 1389 destroy_args(&args); 1390 return 0; 1391 } 1392 late_initcall(debug_vm_pgtable); 1393