mm/debug_vm_pgtable.c

   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 = READ_ONCE(*args->pmdp);
 235         WARN_ON(pmd_write(pmd));
 236         pmdp_huge_get_and_clear(args->mm, vaddr, args->pmdp);
 237         pmd = READ_ONCE(*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 = READ_ONCE(*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 = READ_ONCE(*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 = READ_ONCE(*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 = READ_ONCE(*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 = READ_ONCE(*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 = READ_ONCE(*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 = READ_ONCE(*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 = READ_ONCE(*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 = READ_ONCE(*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 = READ_ONCE(*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 = READ_ONCE(*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 = READ_ONCE(*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 = READ_ONCE(*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 = READ_ONCE(*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 = READ_ONCE(*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 = READ_ONCE(*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 = READ_ONCE(*(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 = READ_ONCE(*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 = READ_ONCE(*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 = READ_ONCE(*args->pmdp);
 631
 632         pr_debug("Validating PMD clear\n");
 633         WARN_ON(pmd_none(pmd));
 634         pmd_clear(args->pmdp);
 635         pmd = READ_ONCE(*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 = READ_ONCE(*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         struct page *page;
 914         pte_t pte;
 915
 916         pr_debug("Validating HugeTLB basic\n");
 917         /*
 918          * Accessing the page associated with the pfn is safe here,
 919          * as it was previously derived from a real kernel symbol.
 920          */
 921         page = pfn_to_page(args->fixed_pmd_pfn);
 922         pte = mk_huge_pte(page, args->page_prot);
 923
 924         WARN_ON(!huge_pte_dirty(huge_pte_mkdirty(pte)));
 925         WARN_ON(!huge_pte_write(huge_pte_mkwrite(huge_pte_wrprotect(pte))));
 926         WARN_ON(huge_pte_write(huge_pte_wrprotect(huge_pte_mkwrite(pte))));
 927
 928 #ifdef CONFIG_ARCH_WANT_GENERAL_HUGETLB
 929         pte = pfn_pte(args->fixed_pmd_pfn, args->page_prot);
 930
 931         WARN_ON(!pte_huge(arch_make_huge_pte(pte, PMD_SHIFT, VM_ACCESS_FLAGS)));
 932 #endif /* CONFIG_ARCH_WANT_GENERAL_HUGETLB */
 933 }
 934 #else  /* !CONFIG_HUGETLB_PAGE */
 935 static void __init hugetlb_basic_tests(struct pgtable_debug_args *args) { }
 936 #endif /* CONFIG_HUGETLB_PAGE */
 937
 938 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 939 static void __init pmd_thp_tests(struct pgtable_debug_args *args)
 940 {
 941         pmd_t pmd;
 942
 943         if (!has_transparent_hugepage())
 944                 return;
 945
 946         pr_debug("Validating PMD based THP\n");
 947         /*
 948          * pmd_trans_huge() and pmd_present() must return positive after
 949          * MMU invalidation with pmd_mkinvalid(). This behavior is an
 950          * optimization for transparent huge page. pmd_trans_huge() must
 951          * be true if pmd_page() returns a valid THP to avoid taking the
 952          * pmd_lock when others walk over non transhuge pmds (i.e. there
 953          * are no THP allocated). Especially when splitting a THP and
 954          * removing the present bit from the pmd, pmd_trans_huge() still
 955          * needs to return true. pmd_present() should be true whenever
 956          * pmd_trans_huge() returns true.
 957          */
 958         pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot);
 959         WARN_ON(!pmd_trans_huge(pmd_mkhuge(pmd)));
 960
 961 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
 962         WARN_ON(!pmd_trans_huge(pmd_mkinvalid(pmd_mkhuge(pmd))));
 963         WARN_ON(!pmd_present(pmd_mkinvalid(pmd_mkhuge(pmd))));
 964         WARN_ON(!pmd_leaf(pmd_mkinvalid(pmd_mkhuge(pmd))));
 965 #endif /* __HAVE_ARCH_PMDP_INVALIDATE */
 966 }
 967
 968 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
 969 static void __init pud_thp_tests(struct pgtable_debug_args *args)
 970 {
 971         pud_t pud;
 972
 973         if (!has_transparent_pud_hugepage())
 974                 return;
 975
 976         pr_debug("Validating PUD based THP\n");
 977         pud = pfn_pud(args->fixed_pud_pfn, args->page_prot);
 978         WARN_ON(!pud_trans_huge(pud_mkhuge(pud)));
 979
 980         /*
 981          * pud_mkinvalid() has been dropped for now. Enable back
 982          * these tests when it comes back with a modified pud_present().
 983          *
 984          * WARN_ON(!pud_trans_huge(pud_mkinvalid(pud_mkhuge(pud))));
 985          * WARN_ON(!pud_present(pud_mkinvalid(pud_mkhuge(pud))));
 986          */
 987 }
 988 #else  /* !CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
 989 static void __init pud_thp_tests(struct pgtable_debug_args *args) { }
 990 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
 991 #else  /* !CONFIG_TRANSPARENT_HUGEPAGE */
 992 static void __init pmd_thp_tests(struct pgtable_debug_args *args) { }
 993 static void __init pud_thp_tests(struct pgtable_debug_args *args) { }
 994 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 995
 996 static unsigned long __init get_random_vaddr(void)
 997 {
 998         unsigned long random_vaddr, random_pages, total_user_pages;
 999
1000         total_user_pages = (TASK_SIZE - FIRST_USER_ADDRESS) / PAGE_SIZE;
1001
1002         random_pages = get_random_long() % total_user_pages;
1003         random_vaddr = FIRST_USER_ADDRESS + random_pages * PAGE_SIZE;
1004
1005         return random_vaddr;
1006 }
1007
1008 static void __init destroy_args(struct pgtable_debug_args *args)
1009 {
1010         struct page *page = NULL;
1011
1012         /* Free (huge) page */
1013         if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
1014             has_transparent_pud_hugepage() &&
1015             args->pud_pfn != ULONG_MAX) {
1016                 if (args->is_contiguous_page) {
1017                         free_contig_range(args->pud_pfn,
1018                                           (1 << (HPAGE_PUD_SHIFT - PAGE_SHIFT)));
1019                 } else {
1020                         page = pfn_to_page(args->pud_pfn);
1021                         __free_pages(page, HPAGE_PUD_SHIFT - PAGE_SHIFT);
1022                 }
1023
1024                 args->pud_pfn = ULONG_MAX;
1025                 args->pmd_pfn = ULONG_MAX;
1026                 args->pte_pfn = ULONG_MAX;
1027         }
1028
1029         if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
1030             has_transparent_hugepage() &&
1031             args->pmd_pfn != ULONG_MAX) {
1032                 if (args->is_contiguous_page) {
1033                         free_contig_range(args->pmd_pfn, (1 << HPAGE_PMD_ORDER));
1034                 } else {
1035                         page = pfn_to_page(args->pmd_pfn);
1036                         __free_pages(page, HPAGE_PMD_ORDER);
1037                 }
1038
1039                 args->pmd_pfn = ULONG_MAX;
1040                 args->pte_pfn = ULONG_MAX;
1041         }
1042
1043         if (args->pte_pfn != ULONG_MAX) {
1044                 page = pfn_to_page(args->pte_pfn);
1045                 __free_page(page);
1046
1047                 args->pte_pfn = ULONG_MAX;
1048         }
1049
1050         /* Free page table entries */
1051         if (args->start_ptep) {
1052                 pte_free(args->mm, args->start_ptep);
1053                 mm_dec_nr_ptes(args->mm);
1054         }
1055
1056         if (args->start_pmdp) {
1057                 pmd_free(args->mm, args->start_pmdp);
1058                 mm_dec_nr_pmds(args->mm);
1059         }
1060
1061         if (args->start_pudp) {
1062                 pud_free(args->mm, args->start_pudp);
1063                 mm_dec_nr_puds(args->mm);
1064         }
1065
1066         if (args->start_p4dp)
1067                 p4d_free(args->mm, args->start_p4dp);
1068
1069         /* Free vma and mm struct */
1070         if (args->vma)
1071                 vm_area_free(args->vma);
1072
1073         if (args->mm)
1074                 mmdrop(args->mm);
1075 }
1076
1077 static struct page * __init
1078 debug_vm_pgtable_alloc_huge_page(struct pgtable_debug_args *args, int order)
1079 {
1080         struct page *page = NULL;
1081
1082 #ifdef CONFIG_CONTIG_ALLOC
1083         if (order > MAX_PAGE_ORDER) {
1084                 page = alloc_contig_pages((1 << order), GFP_KERNEL,
1085                                           first_online_node, NULL);
1086                 if (page) {
1087                         args->is_contiguous_page = true;
1088                         return page;
1089                 }
1090         }
1091 #endif
1092
1093         if (order <= MAX_PAGE_ORDER)
1094                 page = alloc_pages(GFP_KERNEL, order);
1095
1096         return page;
1097 }
1098
1099 /*
1100  * Check if a physical memory range described by <pstart, pend> contains
1101  * an area that is of size psize, and aligned to psize.
1102  *
1103  * Don't use address 0, an all-zeroes physical address might mask bugs, and
1104  * it's not used on x86.
1105  */
1106 static void  __init phys_align_check(phys_addr_t pstart,
1107                                      phys_addr_t pend, unsigned long psize,
1108                                      phys_addr_t *physp, unsigned long *alignp)
1109 {
1110         phys_addr_t aligned_start, aligned_end;
1111
1112         if (pstart == 0)
1113                 pstart = PAGE_SIZE;
1114
1115         aligned_start = ALIGN(pstart, psize);
1116         aligned_end = aligned_start + psize;
1117
1118         if (aligned_end > aligned_start && aligned_end <= pend) {
1119                 *alignp = psize;
1120                 *physp = aligned_start;
1121         }
1122 }
1123
1124 static void __init init_fixed_pfns(struct pgtable_debug_args *args)
1125 {
1126         u64 idx;
1127         phys_addr_t phys, pstart, pend;
1128
1129         /*
1130          * Initialize the fixed pfns. To do this, try to find a
1131          * valid physical range, preferably aligned to PUD_SIZE,
1132          * but settling for aligned to PMD_SIZE as a fallback. If
1133          * neither of those is found, use the physical address of
1134          * the start_kernel symbol.
1135          *
1136          * The memory doesn't need to be allocated, it just needs to exist
1137          * as usable memory. It won't be touched.
1138          *
1139          * The alignment is recorded, and can be checked to see if we
1140          * can run the tests that require an actual valid physical
1141          * address range on some architectures ({pmd,pud}_huge_test
1142          * on x86).
1143          */
1144
1145         phys = __pa_symbol(&start_kernel);
1146         args->fixed_alignment = PAGE_SIZE;
1147
1148         for_each_mem_range(idx, &pstart, &pend) {
1149                 /* First check for a PUD-aligned area */
1150                 phys_align_check(pstart, pend, PUD_SIZE, &phys,
1151                                  &args->fixed_alignment);
1152
1153                 /* If a PUD-aligned area is found, we're done */
1154                 if (args->fixed_alignment == PUD_SIZE)
1155                         break;
1156
1157                 /*
1158                  * If no PMD-aligned area found yet, check for one,
1159                  * but continue the loop to look for a PUD-aligned area.
1160                  */
1161                 if (args->fixed_alignment < PMD_SIZE)
1162                         phys_align_check(pstart, pend, PMD_SIZE, &phys,
1163                                          &args->fixed_alignment);
1164         }
1165
1166         args->fixed_pgd_pfn = __phys_to_pfn(phys & PGDIR_MASK);
1167         args->fixed_p4d_pfn = __phys_to_pfn(phys & P4D_MASK);
1168         args->fixed_pud_pfn = __phys_to_pfn(phys & PUD_MASK);
1169         args->fixed_pmd_pfn = __phys_to_pfn(phys & PMD_MASK);
1170         args->fixed_pte_pfn = __phys_to_pfn(phys & PAGE_MASK);
1171         WARN_ON(!pfn_valid(args->fixed_pte_pfn));
1172 }
1173
1174
1175 static int __init init_args(struct pgtable_debug_args *args)
1176 {
1177         struct page *page = NULL;
1178         int ret = 0;
1179
1180         /*
1181          * Initialize the debugging data.
1182          *
1183          * vm_get_page_prot(VM_NONE) or vm_get_page_prot(VM_SHARED|VM_NONE)
1184          * will help create page table entries with PROT_NONE permission as
1185          * required for pxx_protnone_tests().
1186          */
1187         memset(args, 0, sizeof(*args));
1188         args->vaddr              = get_random_vaddr();
1189         args->page_prot          = vm_get_page_prot(VM_ACCESS_FLAGS);
1190         args->page_prot_none     = vm_get_page_prot(VM_NONE);
1191         args->is_contiguous_page = false;
1192         args->pud_pfn            = ULONG_MAX;
1193         args->pmd_pfn            = ULONG_MAX;
1194         args->pte_pfn            = ULONG_MAX;
1195         args->fixed_pgd_pfn      = ULONG_MAX;
1196         args->fixed_p4d_pfn      = ULONG_MAX;
1197         args->fixed_pud_pfn      = ULONG_MAX;
1198         args->fixed_pmd_pfn      = ULONG_MAX;
1199         args->fixed_pte_pfn      = ULONG_MAX;
1200
1201         /* Allocate mm and vma */
1202         args->mm = mm_alloc();
1203         if (!args->mm) {
1204                 pr_err("Failed to allocate mm struct\n");
1205                 ret = -ENOMEM;
1206                 goto error;
1207         }
1208
1209         args->vma = vm_area_alloc(args->mm);
1210         if (!args->vma) {
1211                 pr_err("Failed to allocate vma\n");
1212                 ret = -ENOMEM;
1213                 goto error;
1214         }
1215
1216         /*
1217          * Allocate page table entries. They will be modified in the tests.
1218          * Lets save the page table entries so that they can be released
1219          * when the tests are completed.
1220          */
1221         args->pgdp = pgd_offset(args->mm, args->vaddr);
1222         args->p4dp = p4d_alloc(args->mm, args->pgdp, args->vaddr);
1223         if (!args->p4dp) {
1224                 pr_err("Failed to allocate p4d entries\n");
1225                 ret = -ENOMEM;
1226                 goto error;
1227         }
1228         args->start_p4dp = p4d_offset(args->pgdp, 0UL);
1229         WARN_ON(!args->start_p4dp);
1230
1231         args->pudp = pud_alloc(args->mm, args->p4dp, args->vaddr);
1232         if (!args->pudp) {
1233                 pr_err("Failed to allocate pud entries\n");
1234                 ret = -ENOMEM;
1235                 goto error;
1236         }
1237         args->start_pudp = pud_offset(args->p4dp, 0UL);
1238         WARN_ON(!args->start_pudp);
1239
1240         args->pmdp = pmd_alloc(args->mm, args->pudp, args->vaddr);
1241         if (!args->pmdp) {
1242                 pr_err("Failed to allocate pmd entries\n");
1243                 ret = -ENOMEM;
1244                 goto error;
1245         }
1246         args->start_pmdp = pmd_offset(args->pudp, 0UL);
1247         WARN_ON(!args->start_pmdp);
1248
1249         if (pte_alloc(args->mm, args->pmdp)) {
1250                 pr_err("Failed to allocate pte entries\n");
1251                 ret = -ENOMEM;
1252                 goto error;
1253         }
1254         args->start_ptep = pmd_pgtable(READ_ONCE(*args->pmdp));
1255         WARN_ON(!args->start_ptep);
1256
1257         init_fixed_pfns(args);
1258
1259         /*
1260          * Allocate (huge) pages because some of the tests need to access
1261          * the data in the pages. The corresponding tests will be skipped
1262          * if we fail to allocate (huge) pages.
1263          */
1264         if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
1265             has_transparent_pud_hugepage()) {
1266                 page = debug_vm_pgtable_alloc_huge_page(args,
1267                                 HPAGE_PUD_SHIFT - PAGE_SHIFT);
1268                 if (page) {
1269                         args->pud_pfn = page_to_pfn(page);
1270                         args->pmd_pfn = args->pud_pfn;
1271                         args->pte_pfn = args->pud_pfn;
1272                         return 0;
1273                 }
1274         }
1275
1276         if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
1277             has_transparent_hugepage()) {
1278                 page = debug_vm_pgtable_alloc_huge_page(args, HPAGE_PMD_ORDER);
1279                 if (page) {
1280                         args->pmd_pfn = page_to_pfn(page);
1281                         args->pte_pfn = args->pmd_pfn;
1282                         return 0;
1283                 }
1284         }
1285
1286         page = alloc_page(GFP_KERNEL);
1287         if (page)
1288                 args->pte_pfn = page_to_pfn(page);
1289
1290         return 0;
1291
1292 error:
1293         destroy_args(args);
1294         return ret;
1295 }
1296
1297 static int __init debug_vm_pgtable(void)
1298 {
1299         struct pgtable_debug_args args;
1300         spinlock_t *ptl = NULL;
1301         int idx, ret;
1302
1303         pr_info("Validating architecture page table helpers\n");
1304         ret = init_args(&args);
1305         if (ret)
1306                 return ret;
1307
1308         /*
1309          * Iterate over each possible vm_flags to make sure that all
1310          * the basic page table transformation validations just hold
1311          * true irrespective of the starting protection value for a
1312          * given page table entry.
1313          *
1314          * Protection based vm_flags combinations are always linear
1315          * and increasing i.e starting from VM_NONE and going up to
1316          * (VM_SHARED | READ | WRITE | EXEC).
1317          */
1318 #define VM_FLAGS_START  (VM_NONE)
1319 #define VM_FLAGS_END    (VM_SHARED | VM_EXEC | VM_WRITE | VM_READ)
1320
1321         for (idx = VM_FLAGS_START; idx <= VM_FLAGS_END; idx++) {
1322                 pte_basic_tests(&args, idx);
1323                 pmd_basic_tests(&args, idx);
1324                 pud_basic_tests(&args, idx);
1325         }
1326
1327         /*
1328          * Both P4D and PGD level tests are very basic which do not
1329          * involve creating page table entries from the protection
1330          * value and the given pfn. Hence just keep them out from
1331          * the above iteration for now to save some test execution
1332          * time.
1333          */
1334         p4d_basic_tests(&args);
1335         pgd_basic_tests(&args);
1336
1337         pmd_leaf_tests(&args);
1338         pud_leaf_tests(&args);
1339
1340         pte_special_tests(&args);
1341         pte_protnone_tests(&args);
1342         pmd_protnone_tests(&args);
1343
1344         pte_devmap_tests(&args);
1345         pmd_devmap_tests(&args);
1346         pud_devmap_tests(&args);
1347
1348         pte_soft_dirty_tests(&args);
1349         pmd_soft_dirty_tests(&args);
1350         pte_swap_soft_dirty_tests(&args);
1351         pmd_swap_soft_dirty_tests(&args);
1352
1353         pte_swap_exclusive_tests(&args);
1354
1355         pte_swap_tests(&args);
1356         pmd_swap_tests(&args);
1357
1358         swap_migration_tests(&args);
1359
1360         pmd_thp_tests(&args);
1361         pud_thp_tests(&args);
1362
1363         hugetlb_basic_tests(&args);
1364
1365         /*
1366          * Page table modifying tests. They need to hold
1367          * proper page table lock.
1368          */
1369
1370         args.ptep = pte_offset_map_lock(args.mm, args.pmdp, args.vaddr, &ptl);
1371         pte_clear_tests(&args);
1372         pte_advanced_tests(&args);
1373         if (args.ptep)
1374                 pte_unmap_unlock(args.ptep, ptl);
1375
1376         ptl = pmd_lock(args.mm, args.pmdp);
1377         pmd_clear_tests(&args);
1378         pmd_advanced_tests(&args);
1379         pmd_huge_tests(&args);
1380         pmd_populate_tests(&args);
1381         spin_unlock(ptl);
1382
1383         ptl = pud_lock(args.mm, args.pudp);
1384         pud_clear_tests(&args);
1385         pud_advanced_tests(&args);
1386         pud_huge_tests(&args);
1387         pud_populate_tests(&args);
1388         spin_unlock(ptl);
1389
1390         spin_lock(&(args.mm->page_table_lock));
1391         p4d_clear_tests(&args);
1392         pgd_clear_tests(&args);
1393         p4d_populate_tests(&args);
1394         pgd_populate_tests(&args);
1395         spin_unlock(&(args.mm->page_table_lock));
1396
1397         destroy_args(&args);
1398         return 0;
1399 }
1400 late_initcall(debug_vm_pgtable);