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/mm.h>
  19 #include <linux/mman.h>
  20 #include <linux/mm_types.h>
  21 #include <linux/module.h>
  22 #include <linux/pfn_t.h>
  23 #include <linux/printk.h>
  24 #include <linux/pgtable.h>
  25 #include <linux/random.h>
  26 #include <linux/spinlock.h>
  27 #include <linux/swap.h>
  28 #include <linux/swapops.h>
  29 #include <linux/start_kernel.h>
  30 #include <linux/sched/mm.h>
  31 #include <linux/io.h>
  32 #include <asm/pgalloc.h>
  33 #include <asm/tlbflush.h>
  34
  35 /*
  36  * Please refer Documentation/vm/arch_pgtable_helpers.rst for the semantics
  37  * expectations that are being validated here. All future changes in here
  38  * or the documentation need to be in sync.
  39  */
  40
  41 #define VMFLAGS (VM_READ|VM_WRITE|VM_EXEC)
  42
  43 /*
  44  * On s390 platform, the lower 4 bits are used to identify given page table
  45  * entry type. But these bits might affect the ability to clear entries with
  46  * pxx_clear() because of how dynamic page table folding works on s390. So
  47  * while loading up the entries do not change the lower 4 bits. It does not
  48  * have affect any other platform. Also avoid the 62nd bit on ppc64 that is
  49  * used to mark a pte entry.
  50  */
  51 #define S390_SKIP_MASK          GENMASK(3, 0)
  52 #if __BITS_PER_LONG == 64
  53 #define PPC64_SKIP_MASK         GENMASK(62, 62)
  54 #else
  55 #define PPC64_SKIP_MASK         0x0
  56 #endif
  57 #define ARCH_SKIP_MASK (S390_SKIP_MASK | PPC64_SKIP_MASK)
  58 #define RANDOM_ORVALUE (GENMASK(BITS_PER_LONG - 1, 0) & ~ARCH_SKIP_MASK)
  59 #define RANDOM_NZVALUE  GENMASK(7, 0)
  60
  61 struct pgtable_debug_args {
  62         struct mm_struct        *mm;
  63         struct vm_area_struct   *vma;
  64
  65         pgd_t                   *pgdp;
  66         p4d_t                   *p4dp;
  67         pud_t                   *pudp;
  68         pmd_t                   *pmdp;
  69         pte_t                   *ptep;
  70
  71         p4d_t                   *start_p4dp;
  72         pud_t                   *start_pudp;
  73         pmd_t                   *start_pmdp;
  74         pgtable_t               start_ptep;
  75
  76         unsigned long           vaddr;
  77         pgprot_t                page_prot;
  78         pgprot_t                page_prot_none;
  79
  80         bool                    is_contiguous_page;
  81         unsigned long           pud_pfn;
  82         unsigned long           pmd_pfn;
  83         unsigned long           pte_pfn;
  84
  85         unsigned long           fixed_pgd_pfn;
  86         unsigned long           fixed_p4d_pfn;
  87         unsigned long           fixed_pud_pfn;
  88         unsigned long           fixed_pmd_pfn;
  89         unsigned long           fixed_pte_pfn;
  90 };
  91
  92 static void __init pte_basic_tests(struct pgtable_debug_args *args, int idx)
  93 {
  94         pgprot_t prot = protection_map[idx];
  95         pte_t pte = pfn_pte(args->fixed_pte_pfn, prot);
  96         unsigned long val = idx, *ptr = &val;
  97
  98         pr_debug("Validating PTE basic (%pGv)\n", ptr);
  99
 100         /*
 101          * This test needs to be executed after the given page table entry
 102          * is created with pfn_pte() to make sure that protection_map[idx]
 103          * does not have the dirty bit enabled from the beginning. This is
 104          * important for platforms like arm64 where (!PTE_RDONLY) indicate
 105          * dirty bit being set.
 106          */
 107         WARN_ON(pte_dirty(pte_wrprotect(pte)));
 108
 109         WARN_ON(!pte_same(pte, pte));
 110         WARN_ON(!pte_young(pte_mkyoung(pte_mkold(pte))));
 111         WARN_ON(!pte_dirty(pte_mkdirty(pte_mkclean(pte))));
 112         WARN_ON(!pte_write(pte_mkwrite(pte_wrprotect(pte))));
 113         WARN_ON(pte_young(pte_mkold(pte_mkyoung(pte))));
 114         WARN_ON(pte_dirty(pte_mkclean(pte_mkdirty(pte))));
 115         WARN_ON(pte_write(pte_wrprotect(pte_mkwrite(pte))));
 116         WARN_ON(pte_dirty(pte_wrprotect(pte_mkclean(pte))));
 117         WARN_ON(!pte_dirty(pte_wrprotect(pte_mkdirty(pte))));
 118 }
 119
 120 static void __init pte_advanced_tests(struct mm_struct *mm,
 121                                       struct vm_area_struct *vma, pte_t *ptep,
 122                                       unsigned long pfn, unsigned long vaddr,
 123                                       pgprot_t prot)
 124 {
 125         pte_t pte;
 126
 127         /*
 128          * Architectures optimize set_pte_at by avoiding TLB flush.
 129          * This requires set_pte_at to be not used to update an
 130          * existing pte entry. Clear pte before we do set_pte_at
 131          */
 132
 133         pr_debug("Validating PTE advanced\n");
 134         pte = pfn_pte(pfn, prot);
 135         set_pte_at(mm, vaddr, ptep, pte);
 136         ptep_set_wrprotect(mm, vaddr, ptep);
 137         pte = ptep_get(ptep);
 138         WARN_ON(pte_write(pte));
 139         ptep_get_and_clear(mm, vaddr, ptep);
 140         pte = ptep_get(ptep);
 141         WARN_ON(!pte_none(pte));
 142
 143         pte = pfn_pte(pfn, prot);
 144         pte = pte_wrprotect(pte);
 145         pte = pte_mkclean(pte);
 146         set_pte_at(mm, vaddr, ptep, pte);
 147         pte = pte_mkwrite(pte);
 148         pte = pte_mkdirty(pte);
 149         ptep_set_access_flags(vma, vaddr, ptep, pte, 1);
 150         pte = ptep_get(ptep);
 151         WARN_ON(!(pte_write(pte) && pte_dirty(pte)));
 152         ptep_get_and_clear_full(mm, vaddr, ptep, 1);
 153         pte = ptep_get(ptep);
 154         WARN_ON(!pte_none(pte));
 155
 156         pte = pfn_pte(pfn, prot);
 157         pte = pte_mkyoung(pte);
 158         set_pte_at(mm, vaddr, ptep, pte);
 159         ptep_test_and_clear_young(vma, vaddr, ptep);
 160         pte = ptep_get(ptep);
 161         WARN_ON(pte_young(pte));
 162 }
 163
 164 static void __init pte_savedwrite_tests(struct pgtable_debug_args *args)
 165 {
 166         pte_t pte = pfn_pte(args->fixed_pte_pfn, args->page_prot_none);
 167
 168         if (!IS_ENABLED(CONFIG_NUMA_BALANCING))
 169                 return;
 170
 171         pr_debug("Validating PTE saved write\n");
 172         WARN_ON(!pte_savedwrite(pte_mk_savedwrite(pte_clear_savedwrite(pte))));
 173         WARN_ON(pte_savedwrite(pte_clear_savedwrite(pte_mk_savedwrite(pte))));
 174 }
 175
 176 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 177 static void __init pmd_basic_tests(struct pgtable_debug_args *args, int idx)
 178 {
 179         pgprot_t prot = protection_map[idx];
 180         unsigned long val = idx, *ptr = &val;
 181         pmd_t pmd;
 182
 183         if (!has_transparent_hugepage())
 184                 return;
 185
 186         pr_debug("Validating PMD basic (%pGv)\n", ptr);
 187         pmd = pfn_pmd(args->fixed_pmd_pfn, prot);
 188
 189         /*
 190          * This test needs to be executed after the given page table entry
 191          * is created with pfn_pmd() to make sure that protection_map[idx]
 192          * does not have the dirty bit enabled from the beginning. This is
 193          * important for platforms like arm64 where (!PTE_RDONLY) indicate
 194          * dirty bit being set.
 195          */
 196         WARN_ON(pmd_dirty(pmd_wrprotect(pmd)));
 197
 198
 199         WARN_ON(!pmd_same(pmd, pmd));
 200         WARN_ON(!pmd_young(pmd_mkyoung(pmd_mkold(pmd))));
 201         WARN_ON(!pmd_dirty(pmd_mkdirty(pmd_mkclean(pmd))));
 202         WARN_ON(!pmd_write(pmd_mkwrite(pmd_wrprotect(pmd))));
 203         WARN_ON(pmd_young(pmd_mkold(pmd_mkyoung(pmd))));
 204         WARN_ON(pmd_dirty(pmd_mkclean(pmd_mkdirty(pmd))));
 205         WARN_ON(pmd_write(pmd_wrprotect(pmd_mkwrite(pmd))));
 206         WARN_ON(pmd_dirty(pmd_wrprotect(pmd_mkclean(pmd))));
 207         WARN_ON(!pmd_dirty(pmd_wrprotect(pmd_mkdirty(pmd))));
 208         /*
 209          * A huge page does not point to next level page table
 210          * entry. Hence this must qualify as pmd_bad().
 211          */
 212         WARN_ON(!pmd_bad(pmd_mkhuge(pmd)));
 213 }
 214
 215 static void __init pmd_advanced_tests(struct mm_struct *mm,
 216                                       struct vm_area_struct *vma, pmd_t *pmdp,
 217                                       unsigned long pfn, unsigned long vaddr,
 218                                       pgprot_t prot, pgtable_t pgtable)
 219 {
 220         pmd_t pmd;
 221
 222         if (!has_transparent_hugepage())
 223                 return;
 224
 225         pr_debug("Validating PMD advanced\n");
 226         /* Align the address wrt HPAGE_PMD_SIZE */
 227         vaddr &= HPAGE_PMD_MASK;
 228
 229         pgtable_trans_huge_deposit(mm, pmdp, pgtable);
 230
 231         pmd = pfn_pmd(pfn, prot);
 232         set_pmd_at(mm, vaddr, pmdp, pmd);
 233         pmdp_set_wrprotect(mm, vaddr, pmdp);
 234         pmd = READ_ONCE(*pmdp);
 235         WARN_ON(pmd_write(pmd));
 236         pmdp_huge_get_and_clear(mm, vaddr, pmdp);
 237         pmd = READ_ONCE(*pmdp);
 238         WARN_ON(!pmd_none(pmd));
 239
 240         pmd = pfn_pmd(pfn, prot);
 241         pmd = pmd_wrprotect(pmd);
 242         pmd = pmd_mkclean(pmd);
 243         set_pmd_at(mm, vaddr, pmdp, pmd);
 244         pmd = pmd_mkwrite(pmd);
 245         pmd = pmd_mkdirty(pmd);
 246         pmdp_set_access_flags(vma, vaddr, pmdp, pmd, 1);
 247         pmd = READ_ONCE(*pmdp);
 248         WARN_ON(!(pmd_write(pmd) && pmd_dirty(pmd)));
 249         pmdp_huge_get_and_clear_full(vma, vaddr, pmdp, 1);
 250         pmd = READ_ONCE(*pmdp);
 251         WARN_ON(!pmd_none(pmd));
 252
 253         pmd = pmd_mkhuge(pfn_pmd(pfn, prot));
 254         pmd = pmd_mkyoung(pmd);
 255         set_pmd_at(mm, vaddr, pmdp, pmd);
 256         pmdp_test_and_clear_young(vma, vaddr, pmdp);
 257         pmd = READ_ONCE(*pmdp);
 258         WARN_ON(pmd_young(pmd));
 259
 260         /*  Clear the pte entries  */
 261         pmdp_huge_get_and_clear(mm, vaddr, pmdp);
 262         pgtable = pgtable_trans_huge_withdraw(mm, pmdp);
 263 }
 264
 265 static void __init pmd_leaf_tests(struct pgtable_debug_args *args)
 266 {
 267         pmd_t pmd;
 268
 269         if (!has_transparent_hugepage())
 270                 return;
 271
 272         pr_debug("Validating PMD leaf\n");
 273         pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot);
 274
 275         /*
 276          * PMD based THP is a leaf entry.
 277          */
 278         pmd = pmd_mkhuge(pmd);
 279         WARN_ON(!pmd_leaf(pmd));
 280 }
 281
 282 static void __init pmd_savedwrite_tests(struct pgtable_debug_args *args)
 283 {
 284         pmd_t pmd;
 285
 286         if (!IS_ENABLED(CONFIG_NUMA_BALANCING))
 287                 return;
 288
 289         if (!has_transparent_hugepage())
 290                 return;
 291
 292         pr_debug("Validating PMD saved write\n");
 293         pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot_none);
 294         WARN_ON(!pmd_savedwrite(pmd_mk_savedwrite(pmd_clear_savedwrite(pmd))));
 295         WARN_ON(pmd_savedwrite(pmd_clear_savedwrite(pmd_mk_savedwrite(pmd))));
 296 }
 297
 298 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
 299 static void __init pud_basic_tests(struct pgtable_debug_args *args, int idx)
 300 {
 301         pgprot_t prot = protection_map[idx];
 302         unsigned long val = idx, *ptr = &val;
 303         pud_t pud;
 304
 305         if (!has_transparent_hugepage())
 306                 return;
 307
 308         pr_debug("Validating PUD basic (%pGv)\n", ptr);
 309         pud = pfn_pud(args->fixed_pud_pfn, prot);
 310
 311         /*
 312          * This test needs to be executed after the given page table entry
 313          * is created with pfn_pud() to make sure that protection_map[idx]
 314          * does not have the dirty bit enabled from the beginning. This is
 315          * important for platforms like arm64 where (!PTE_RDONLY) indicate
 316          * dirty bit being set.
 317          */
 318         WARN_ON(pud_dirty(pud_wrprotect(pud)));
 319
 320         WARN_ON(!pud_same(pud, pud));
 321         WARN_ON(!pud_young(pud_mkyoung(pud_mkold(pud))));
 322         WARN_ON(!pud_dirty(pud_mkdirty(pud_mkclean(pud))));
 323         WARN_ON(pud_dirty(pud_mkclean(pud_mkdirty(pud))));
 324         WARN_ON(!pud_write(pud_mkwrite(pud_wrprotect(pud))));
 325         WARN_ON(pud_write(pud_wrprotect(pud_mkwrite(pud))));
 326         WARN_ON(pud_young(pud_mkold(pud_mkyoung(pud))));
 327         WARN_ON(pud_dirty(pud_wrprotect(pud_mkclean(pud))));
 328         WARN_ON(!pud_dirty(pud_wrprotect(pud_mkdirty(pud))));
 329
 330         if (mm_pmd_folded(args->mm))
 331                 return;
 332
 333         /*
 334          * A huge page does not point to next level page table
 335          * entry. Hence this must qualify as pud_bad().
 336          */
 337         WARN_ON(!pud_bad(pud_mkhuge(pud)));
 338 }
 339
 340 static void __init pud_advanced_tests(struct mm_struct *mm,
 341                                       struct vm_area_struct *vma, pud_t *pudp,
 342                                       unsigned long pfn, unsigned long vaddr,
 343                                       pgprot_t prot)
 344 {
 345         pud_t pud;
 346
 347         if (!has_transparent_hugepage())
 348                 return;
 349
 350         pr_debug("Validating PUD advanced\n");
 351         /* Align the address wrt HPAGE_PUD_SIZE */
 352         vaddr &= HPAGE_PUD_MASK;
 353
 354         pud = pfn_pud(pfn, prot);
 355         set_pud_at(mm, vaddr, pudp, pud);
 356         pudp_set_wrprotect(mm, vaddr, pudp);
 357         pud = READ_ONCE(*pudp);
 358         WARN_ON(pud_write(pud));
 359
 360 #ifndef __PAGETABLE_PMD_FOLDED
 361         pudp_huge_get_and_clear(mm, vaddr, pudp);
 362         pud = READ_ONCE(*pudp);
 363         WARN_ON(!pud_none(pud));
 364 #endif /* __PAGETABLE_PMD_FOLDED */
 365         pud = pfn_pud(pfn, prot);
 366         pud = pud_wrprotect(pud);
 367         pud = pud_mkclean(pud);
 368         set_pud_at(mm, vaddr, pudp, pud);
 369         pud = pud_mkwrite(pud);
 370         pud = pud_mkdirty(pud);
 371         pudp_set_access_flags(vma, vaddr, pudp, pud, 1);
 372         pud = READ_ONCE(*pudp);
 373         WARN_ON(!(pud_write(pud) && pud_dirty(pud)));
 374
 375 #ifndef __PAGETABLE_PMD_FOLDED
 376         pudp_huge_get_and_clear_full(mm, vaddr, pudp, 1);
 377         pud = READ_ONCE(*pudp);
 378         WARN_ON(!pud_none(pud));
 379 #endif /* __PAGETABLE_PMD_FOLDED */
 380
 381         pud = pfn_pud(pfn, prot);
 382         pud = pud_mkyoung(pud);
 383         set_pud_at(mm, vaddr, pudp, pud);
 384         pudp_test_and_clear_young(vma, vaddr, pudp);
 385         pud = READ_ONCE(*pudp);
 386         WARN_ON(pud_young(pud));
 387
 388         pudp_huge_get_and_clear(mm, vaddr, pudp);
 389 }
 390
 391 static void __init pud_leaf_tests(struct pgtable_debug_args *args)
 392 {
 393         pud_t pud;
 394
 395         if (!has_transparent_hugepage())
 396                 return;
 397
 398         pr_debug("Validating PUD leaf\n");
 399         pud = pfn_pud(args->fixed_pud_pfn, args->page_prot);
 400         /*
 401          * PUD based THP is a leaf entry.
 402          */
 403         pud = pud_mkhuge(pud);
 404         WARN_ON(!pud_leaf(pud));
 405 }
 406 #else  /* !CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
 407 static void __init pud_basic_tests(struct pgtable_debug_args *args, int idx) { }
 408 static void __init pud_advanced_tests(struct mm_struct *mm,
 409                                       struct vm_area_struct *vma, pud_t *pudp,
 410                                       unsigned long pfn, unsigned long vaddr,
 411                                       pgprot_t prot)
 412 {
 413 }
 414 static void __init pud_leaf_tests(struct pgtable_debug_args *args) { }
 415 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
 416 #else  /* !CONFIG_TRANSPARENT_HUGEPAGE */
 417 static void __init pmd_basic_tests(struct pgtable_debug_args *args, int idx) { }
 418 static void __init pud_basic_tests(struct pgtable_debug_args *args, int idx) { }
 419 static void __init pmd_advanced_tests(struct mm_struct *mm,
 420                                       struct vm_area_struct *vma, pmd_t *pmdp,
 421                                       unsigned long pfn, unsigned long vaddr,
 422                                       pgprot_t prot, pgtable_t pgtable)
 423 {
 424 }
 425 static void __init pud_advanced_tests(struct mm_struct *mm,
 426                                       struct vm_area_struct *vma, pud_t *pudp,
 427                                       unsigned long pfn, unsigned long vaddr,
 428                                       pgprot_t prot)
 429 {
 430 }
 431 static void __init pmd_leaf_tests(struct pgtable_debug_args *args) { }
 432 static void __init pud_leaf_tests(struct pgtable_debug_args *args) { }
 433 static void __init pmd_savedwrite_tests(struct pgtable_debug_args *args) { }
 434 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 435
 436 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
 437 static void __init pmd_huge_tests(pmd_t *pmdp, unsigned long pfn, pgprot_t prot)
 438 {
 439         pmd_t pmd;
 440
 441         if (!arch_vmap_pmd_supported(prot))
 442                 return;
 443
 444         pr_debug("Validating PMD huge\n");
 445         /*
 446          * X86 defined pmd_set_huge() verifies that the given
 447          * PMD is not a populated non-leaf entry.
 448          */
 449         WRITE_ONCE(*pmdp, __pmd(0));
 450         WARN_ON(!pmd_set_huge(pmdp, __pfn_to_phys(pfn), prot));
 451         WARN_ON(!pmd_clear_huge(pmdp));
 452         pmd = READ_ONCE(*pmdp);
 453         WARN_ON(!pmd_none(pmd));
 454 }
 455
 456 static void __init pud_huge_tests(pud_t *pudp, unsigned long pfn, pgprot_t prot)
 457 {
 458         pud_t pud;
 459
 460         if (!arch_vmap_pud_supported(prot))
 461                 return;
 462
 463         pr_debug("Validating PUD huge\n");
 464         /*
 465          * X86 defined pud_set_huge() verifies that the given
 466          * PUD is not a populated non-leaf entry.
 467          */
 468         WRITE_ONCE(*pudp, __pud(0));
 469         WARN_ON(!pud_set_huge(pudp, __pfn_to_phys(pfn), prot));
 470         WARN_ON(!pud_clear_huge(pudp));
 471         pud = READ_ONCE(*pudp);
 472         WARN_ON(!pud_none(pud));
 473 }
 474 #else /* !CONFIG_HAVE_ARCH_HUGE_VMAP */
 475 static void __init pmd_huge_tests(pmd_t *pmdp, unsigned long pfn, pgprot_t prot) { }
 476 static void __init pud_huge_tests(pud_t *pudp, unsigned long pfn, pgprot_t prot) { }
 477 #endif /* CONFIG_HAVE_ARCH_HUGE_VMAP */
 478
 479 static void __init p4d_basic_tests(struct pgtable_debug_args *args)
 480 {
 481         p4d_t p4d;
 482
 483         pr_debug("Validating P4D basic\n");
 484         memset(&p4d, RANDOM_NZVALUE, sizeof(p4d_t));
 485         WARN_ON(!p4d_same(p4d, p4d));
 486 }
 487
 488 static void __init pgd_basic_tests(struct pgtable_debug_args *args)
 489 {
 490         pgd_t pgd;
 491
 492         pr_debug("Validating PGD basic\n");
 493         memset(&pgd, RANDOM_NZVALUE, sizeof(pgd_t));
 494         WARN_ON(!pgd_same(pgd, pgd));
 495 }
 496
 497 #ifndef __PAGETABLE_PUD_FOLDED
 498 static void __init pud_clear_tests(struct mm_struct *mm, pud_t *pudp)
 499 {
 500         pud_t pud = READ_ONCE(*pudp);
 501
 502         if (mm_pmd_folded(mm))
 503                 return;
 504
 505         pr_debug("Validating PUD clear\n");
 506         pud = __pud(pud_val(pud) | RANDOM_ORVALUE);
 507         WRITE_ONCE(*pudp, pud);
 508         pud_clear(pudp);
 509         pud = READ_ONCE(*pudp);
 510         WARN_ON(!pud_none(pud));
 511 }
 512
 513 static void __init pud_populate_tests(struct mm_struct *mm, pud_t *pudp,
 514                                       pmd_t *pmdp)
 515 {
 516         pud_t pud;
 517
 518         if (mm_pmd_folded(mm))
 519                 return;
 520
 521         pr_debug("Validating PUD populate\n");
 522         /*
 523          * This entry points to next level page table page.
 524          * Hence this must not qualify as pud_bad().
 525          */
 526         pud_populate(mm, pudp, pmdp);
 527         pud = READ_ONCE(*pudp);
 528         WARN_ON(pud_bad(pud));
 529 }
 530 #else  /* !__PAGETABLE_PUD_FOLDED */
 531 static void __init pud_clear_tests(struct mm_struct *mm, pud_t *pudp) { }
 532 static void __init pud_populate_tests(struct mm_struct *mm, pud_t *pudp,
 533                                       pmd_t *pmdp)
 534 {
 535 }
 536 #endif /* PAGETABLE_PUD_FOLDED */
 537
 538 #ifndef __PAGETABLE_P4D_FOLDED
 539 static void __init p4d_clear_tests(struct mm_struct *mm, p4d_t *p4dp)
 540 {
 541         p4d_t p4d = READ_ONCE(*p4dp);
 542
 543         if (mm_pud_folded(mm))
 544                 return;
 545
 546         pr_debug("Validating P4D clear\n");
 547         p4d = __p4d(p4d_val(p4d) | RANDOM_ORVALUE);
 548         WRITE_ONCE(*p4dp, p4d);
 549         p4d_clear(p4dp);
 550         p4d = READ_ONCE(*p4dp);
 551         WARN_ON(!p4d_none(p4d));
 552 }
 553
 554 static void __init p4d_populate_tests(struct mm_struct *mm, p4d_t *p4dp,
 555                                       pud_t *pudp)
 556 {
 557         p4d_t p4d;
 558
 559         if (mm_pud_folded(mm))
 560                 return;
 561
 562         pr_debug("Validating P4D populate\n");
 563         /*
 564          * This entry points to next level page table page.
 565          * Hence this must not qualify as p4d_bad().
 566          */
 567         pud_clear(pudp);
 568         p4d_clear(p4dp);
 569         p4d_populate(mm, p4dp, pudp);
 570         p4d = READ_ONCE(*p4dp);
 571         WARN_ON(p4d_bad(p4d));
 572 }
 573
 574 static void __init pgd_clear_tests(struct mm_struct *mm, pgd_t *pgdp)
 575 {
 576         pgd_t pgd = READ_ONCE(*pgdp);
 577
 578         if (mm_p4d_folded(mm))
 579                 return;
 580
 581         pr_debug("Validating PGD clear\n");
 582         pgd = __pgd(pgd_val(pgd) | RANDOM_ORVALUE);
 583         WRITE_ONCE(*pgdp, pgd);
 584         pgd_clear(pgdp);
 585         pgd = READ_ONCE(*pgdp);
 586         WARN_ON(!pgd_none(pgd));
 587 }
 588
 589 static void __init pgd_populate_tests(struct mm_struct *mm, pgd_t *pgdp,
 590                                       p4d_t *p4dp)
 591 {
 592         pgd_t pgd;
 593
 594         if (mm_p4d_folded(mm))
 595                 return;
 596
 597         pr_debug("Validating PGD populate\n");
 598         /*
 599          * This entry points to next level page table page.
 600          * Hence this must not qualify as pgd_bad().
 601          */
 602         p4d_clear(p4dp);
 603         pgd_clear(pgdp);
 604         pgd_populate(mm, pgdp, p4dp);
 605         pgd = READ_ONCE(*pgdp);
 606         WARN_ON(pgd_bad(pgd));
 607 }
 608 #else  /* !__PAGETABLE_P4D_FOLDED */
 609 static void __init p4d_clear_tests(struct mm_struct *mm, p4d_t *p4dp) { }
 610 static void __init pgd_clear_tests(struct mm_struct *mm, pgd_t *pgdp) { }
 611 static void __init p4d_populate_tests(struct mm_struct *mm, p4d_t *p4dp,
 612                                       pud_t *pudp)
 613 {
 614 }
 615 static void __init pgd_populate_tests(struct mm_struct *mm, pgd_t *pgdp,
 616                                       p4d_t *p4dp)
 617 {
 618 }
 619 #endif /* PAGETABLE_P4D_FOLDED */
 620
 621 static void __init pte_clear_tests(struct mm_struct *mm, pte_t *ptep,
 622                                    unsigned long pfn, unsigned long vaddr,
 623                                    pgprot_t prot)
 624 {
 625         pte_t pte = pfn_pte(pfn, prot);
 626
 627         pr_debug("Validating PTE clear\n");
 628 #ifndef CONFIG_RISCV
 629         pte = __pte(pte_val(pte) | RANDOM_ORVALUE);
 630 #endif
 631         set_pte_at(mm, vaddr, ptep, pte);
 632         barrier();
 633         pte_clear(mm, vaddr, ptep);
 634         pte = ptep_get(ptep);
 635         WARN_ON(!pte_none(pte));
 636 }
 637
 638 static void __init pmd_clear_tests(struct mm_struct *mm, pmd_t *pmdp)
 639 {
 640         pmd_t pmd = READ_ONCE(*pmdp);
 641
 642         pr_debug("Validating PMD clear\n");
 643         pmd = __pmd(pmd_val(pmd) | RANDOM_ORVALUE);
 644         WRITE_ONCE(*pmdp, pmd);
 645         pmd_clear(pmdp);
 646         pmd = READ_ONCE(*pmdp);
 647         WARN_ON(!pmd_none(pmd));
 648 }
 649
 650 static void __init pmd_populate_tests(struct mm_struct *mm, pmd_t *pmdp,
 651                                       pgtable_t pgtable)
 652 {
 653         pmd_t pmd;
 654
 655         pr_debug("Validating PMD populate\n");
 656         /*
 657          * This entry points to next level page table page.
 658          * Hence this must not qualify as pmd_bad().
 659          */
 660         pmd_populate(mm, pmdp, pgtable);
 661         pmd = READ_ONCE(*pmdp);
 662         WARN_ON(pmd_bad(pmd));
 663 }
 664
 665 static void __init pte_special_tests(unsigned long pfn, pgprot_t prot)
 666 {
 667         pte_t pte = pfn_pte(pfn, prot);
 668
 669         if (!IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL))
 670                 return;
 671
 672         pr_debug("Validating PTE special\n");
 673         WARN_ON(!pte_special(pte_mkspecial(pte)));
 674 }
 675
 676 static void __init pte_protnone_tests(unsigned long pfn, pgprot_t prot)
 677 {
 678         pte_t pte = pfn_pte(pfn, prot);
 679
 680         if (!IS_ENABLED(CONFIG_NUMA_BALANCING))
 681                 return;
 682
 683         pr_debug("Validating PTE protnone\n");
 684         WARN_ON(!pte_protnone(pte));
 685         WARN_ON(!pte_present(pte));
 686 }
 687
 688 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 689 static void __init pmd_protnone_tests(unsigned long pfn, pgprot_t prot)
 690 {
 691         pmd_t pmd;
 692
 693         if (!IS_ENABLED(CONFIG_NUMA_BALANCING))
 694                 return;
 695
 696         if (!has_transparent_hugepage())
 697                 return;
 698
 699         pr_debug("Validating PMD protnone\n");
 700         pmd = pmd_mkhuge(pfn_pmd(pfn, prot));
 701         WARN_ON(!pmd_protnone(pmd));
 702         WARN_ON(!pmd_present(pmd));
 703 }
 704 #else  /* !CONFIG_TRANSPARENT_HUGEPAGE */
 705 static void __init pmd_protnone_tests(unsigned long pfn, pgprot_t prot) { }
 706 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 707
 708 #ifdef CONFIG_ARCH_HAS_PTE_DEVMAP
 709 static void __init pte_devmap_tests(unsigned long pfn, pgprot_t prot)
 710 {
 711         pte_t pte = pfn_pte(pfn, prot);
 712
 713         pr_debug("Validating PTE devmap\n");
 714         WARN_ON(!pte_devmap(pte_mkdevmap(pte)));
 715 }
 716
 717 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 718 static void __init pmd_devmap_tests(unsigned long pfn, pgprot_t prot)
 719 {
 720         pmd_t pmd;
 721
 722         if (!has_transparent_hugepage())
 723                 return;
 724
 725         pr_debug("Validating PMD devmap\n");
 726         pmd = pfn_pmd(pfn, prot);
 727         WARN_ON(!pmd_devmap(pmd_mkdevmap(pmd)));
 728 }
 729
 730 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
 731 static void __init pud_devmap_tests(unsigned long pfn, pgprot_t prot)
 732 {
 733         pud_t pud;
 734
 735         if (!has_transparent_hugepage())
 736                 return;
 737
 738         pr_debug("Validating PUD devmap\n");
 739         pud = pfn_pud(pfn, prot);
 740         WARN_ON(!pud_devmap(pud_mkdevmap(pud)));
 741 }
 742 #else  /* !CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
 743 static void __init pud_devmap_tests(unsigned long pfn, pgprot_t prot) { }
 744 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
 745 #else  /* CONFIG_TRANSPARENT_HUGEPAGE */
 746 static void __init pmd_devmap_tests(unsigned long pfn, pgprot_t prot) { }
 747 static void __init pud_devmap_tests(unsigned long pfn, pgprot_t prot) { }
 748 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 749 #else
 750 static void __init pte_devmap_tests(unsigned long pfn, pgprot_t prot) { }
 751 static void __init pmd_devmap_tests(unsigned long pfn, pgprot_t prot) { }
 752 static void __init pud_devmap_tests(unsigned long pfn, pgprot_t prot) { }
 753 #endif /* CONFIG_ARCH_HAS_PTE_DEVMAP */
 754
 755 static void __init pte_soft_dirty_tests(unsigned long pfn, pgprot_t prot)
 756 {
 757         pte_t pte = pfn_pte(pfn, prot);
 758
 759         if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY))
 760                 return;
 761
 762         pr_debug("Validating PTE soft dirty\n");
 763         WARN_ON(!pte_soft_dirty(pte_mksoft_dirty(pte)));
 764         WARN_ON(pte_soft_dirty(pte_clear_soft_dirty(pte)));
 765 }
 766
 767 static void __init pte_swap_soft_dirty_tests(unsigned long pfn, pgprot_t prot)
 768 {
 769         pte_t pte = pfn_pte(pfn, prot);
 770
 771         if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY))
 772                 return;
 773
 774         pr_debug("Validating PTE swap soft dirty\n");
 775         WARN_ON(!pte_swp_soft_dirty(pte_swp_mksoft_dirty(pte)));
 776         WARN_ON(pte_swp_soft_dirty(pte_swp_clear_soft_dirty(pte)));
 777 }
 778
 779 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 780 static void __init pmd_soft_dirty_tests(unsigned long pfn, pgprot_t prot)
 781 {
 782         pmd_t pmd;
 783
 784         if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY))
 785                 return;
 786
 787         if (!has_transparent_hugepage())
 788                 return;
 789
 790         pr_debug("Validating PMD soft dirty\n");
 791         pmd = pfn_pmd(pfn, prot);
 792         WARN_ON(!pmd_soft_dirty(pmd_mksoft_dirty(pmd)));
 793         WARN_ON(pmd_soft_dirty(pmd_clear_soft_dirty(pmd)));
 794 }
 795
 796 static void __init pmd_swap_soft_dirty_tests(unsigned long pfn, pgprot_t prot)
 797 {
 798         pmd_t pmd;
 799
 800         if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) ||
 801                 !IS_ENABLED(CONFIG_ARCH_ENABLE_THP_MIGRATION))
 802                 return;
 803
 804         if (!has_transparent_hugepage())
 805                 return;
 806
 807         pr_debug("Validating PMD swap soft dirty\n");
 808         pmd = pfn_pmd(pfn, prot);
 809         WARN_ON(!pmd_swp_soft_dirty(pmd_swp_mksoft_dirty(pmd)));
 810         WARN_ON(pmd_swp_soft_dirty(pmd_swp_clear_soft_dirty(pmd)));
 811 }
 812 #else  /* !CONFIG_TRANSPARENT_HUGEPAGE */
 813 static void __init pmd_soft_dirty_tests(unsigned long pfn, pgprot_t prot) { }
 814 static void __init pmd_swap_soft_dirty_tests(unsigned long pfn, pgprot_t prot)
 815 {
 816 }
 817 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 818
 819 static void __init pte_swap_tests(unsigned long pfn, pgprot_t prot)
 820 {
 821         swp_entry_t swp;
 822         pte_t pte;
 823
 824         pr_debug("Validating PTE swap\n");
 825         pte = pfn_pte(pfn, prot);
 826         swp = __pte_to_swp_entry(pte);
 827         pte = __swp_entry_to_pte(swp);
 828         WARN_ON(pfn != pte_pfn(pte));
 829 }
 830
 831 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
 832 static void __init pmd_swap_tests(unsigned long pfn, pgprot_t prot)
 833 {
 834         swp_entry_t swp;
 835         pmd_t pmd;
 836
 837         if (!has_transparent_hugepage())
 838                 return;
 839
 840         pr_debug("Validating PMD swap\n");
 841         pmd = pfn_pmd(pfn, prot);
 842         swp = __pmd_to_swp_entry(pmd);
 843         pmd = __swp_entry_to_pmd(swp);
 844         WARN_ON(pfn != pmd_pfn(pmd));
 845 }
 846 #else  /* !CONFIG_ARCH_ENABLE_THP_MIGRATION */
 847 static void __init pmd_swap_tests(unsigned long pfn, pgprot_t prot) { }
 848 #endif /* CONFIG_ARCH_ENABLE_THP_MIGRATION */
 849
 850 static void __init swap_migration_tests(void)
 851 {
 852         struct page *page;
 853         swp_entry_t swp;
 854
 855         if (!IS_ENABLED(CONFIG_MIGRATION))
 856                 return;
 857
 858         pr_debug("Validating swap migration\n");
 859         /*
 860          * swap_migration_tests() requires a dedicated page as it needs to
 861          * be locked before creating a migration entry from it. Locking the
 862          * page that actually maps kernel text ('start_kernel') can be real
 863          * problematic. Lets allocate a dedicated page explicitly for this
 864          * purpose that will be freed subsequently.
 865          */
 866         page = alloc_page(GFP_KERNEL);
 867         if (!page) {
 868                 pr_err("page allocation failed\n");
 869                 return;
 870         }
 871
 872         /*
 873          * make_migration_entry() expects given page to be
 874          * locked, otherwise it stumbles upon a BUG_ON().
 875          */
 876         __SetPageLocked(page);
 877         swp = make_writable_migration_entry(page_to_pfn(page));
 878         WARN_ON(!is_migration_entry(swp));
 879         WARN_ON(!is_writable_migration_entry(swp));
 880
 881         swp = make_readable_migration_entry(swp_offset(swp));
 882         WARN_ON(!is_migration_entry(swp));
 883         WARN_ON(is_writable_migration_entry(swp));
 884
 885         swp = make_readable_migration_entry(page_to_pfn(page));
 886         WARN_ON(!is_migration_entry(swp));
 887         WARN_ON(is_writable_migration_entry(swp));
 888         __ClearPageLocked(page);
 889         __free_page(page);
 890 }
 891
 892 #ifdef CONFIG_HUGETLB_PAGE
 893 static void __init hugetlb_basic_tests(struct pgtable_debug_args *args)
 894 {
 895         struct page *page;
 896         pte_t pte;
 897
 898         pr_debug("Validating HugeTLB basic\n");
 899         /*
 900          * Accessing the page associated with the pfn is safe here,
 901          * as it was previously derived from a real kernel symbol.
 902          */
 903         page = pfn_to_page(args->fixed_pmd_pfn);
 904         pte = mk_huge_pte(page, args->page_prot);
 905
 906         WARN_ON(!huge_pte_dirty(huge_pte_mkdirty(pte)));
 907         WARN_ON(!huge_pte_write(huge_pte_mkwrite(huge_pte_wrprotect(pte))));
 908         WARN_ON(huge_pte_write(huge_pte_wrprotect(huge_pte_mkwrite(pte))));
 909
 910 #ifdef CONFIG_ARCH_WANT_GENERAL_HUGETLB
 911         pte = pfn_pte(args->fixed_pmd_pfn, args->page_prot);
 912
 913         WARN_ON(!pte_huge(pte_mkhuge(pte)));
 914 #endif /* CONFIG_ARCH_WANT_GENERAL_HUGETLB */
 915 }
 916 #else  /* !CONFIG_HUGETLB_PAGE */
 917 static void __init hugetlb_basic_tests(struct pgtable_debug_args *args) { }
 918 #endif /* CONFIG_HUGETLB_PAGE */
 919
 920 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 921 static void __init pmd_thp_tests(unsigned long pfn, pgprot_t prot)
 922 {
 923         pmd_t pmd;
 924
 925         if (!has_transparent_hugepage())
 926                 return;
 927
 928         pr_debug("Validating PMD based THP\n");
 929         /*
 930          * pmd_trans_huge() and pmd_present() must return positive after
 931          * MMU invalidation with pmd_mkinvalid(). This behavior is an
 932          * optimization for transparent huge page. pmd_trans_huge() must
 933          * be true if pmd_page() returns a valid THP to avoid taking the
 934          * pmd_lock when others walk over non transhuge pmds (i.e. there
 935          * are no THP allocated). Especially when splitting a THP and
 936          * removing the present bit from the pmd, pmd_trans_huge() still
 937          * needs to return true. pmd_present() should be true whenever
 938          * pmd_trans_huge() returns true.
 939          */
 940         pmd = pfn_pmd(pfn, prot);
 941         WARN_ON(!pmd_trans_huge(pmd_mkhuge(pmd)));
 942
 943 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
 944         WARN_ON(!pmd_trans_huge(pmd_mkinvalid(pmd_mkhuge(pmd))));
 945         WARN_ON(!pmd_present(pmd_mkinvalid(pmd_mkhuge(pmd))));
 946 #endif /* __HAVE_ARCH_PMDP_INVALIDATE */
 947 }
 948
 949 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
 950 static void __init pud_thp_tests(unsigned long pfn, pgprot_t prot)
 951 {
 952         pud_t pud;
 953
 954         if (!has_transparent_hugepage())
 955                 return;
 956
 957         pr_debug("Validating PUD based THP\n");
 958         pud = pfn_pud(pfn, prot);
 959         WARN_ON(!pud_trans_huge(pud_mkhuge(pud)));
 960
 961         /*
 962          * pud_mkinvalid() has been dropped for now. Enable back
 963          * these tests when it comes back with a modified pud_present().
 964          *
 965          * WARN_ON(!pud_trans_huge(pud_mkinvalid(pud_mkhuge(pud))));
 966          * WARN_ON(!pud_present(pud_mkinvalid(pud_mkhuge(pud))));
 967          */
 968 }
 969 #else  /* !CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
 970 static void __init pud_thp_tests(unsigned long pfn, pgprot_t prot) { }
 971 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
 972 #else  /* !CONFIG_TRANSPARENT_HUGEPAGE */
 973 static void __init pmd_thp_tests(unsigned long pfn, pgprot_t prot) { }
 974 static void __init pud_thp_tests(unsigned long pfn, pgprot_t prot) { }
 975 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 976
 977 static unsigned long __init get_random_vaddr(void)
 978 {
 979         unsigned long random_vaddr, random_pages, total_user_pages;
 980
 981         total_user_pages = (TASK_SIZE - FIRST_USER_ADDRESS) / PAGE_SIZE;
 982
 983         random_pages = get_random_long() % total_user_pages;
 984         random_vaddr = FIRST_USER_ADDRESS + random_pages * PAGE_SIZE;
 985
 986         return random_vaddr;
 987 }
 988
 989 static void __init destroy_args(struct pgtable_debug_args *args)
 990 {
 991         struct page *page = NULL;
 992
 993         /* Free (huge) page */
 994         if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
 995             IS_ENABLED(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD) &&
 996             has_transparent_hugepage() &&
 997             args->pud_pfn != ULONG_MAX) {
 998                 if (args->is_contiguous_page) {
 999                         free_contig_range(args->pud_pfn,
1000                                           (1 << (HPAGE_PUD_SHIFT - PAGE_SHIFT)));
1001                 } else {
1002                         page = pfn_to_page(args->pud_pfn);
1003                         __free_pages(page, HPAGE_PUD_SHIFT - PAGE_SHIFT);
1004                 }
1005
1006                 args->pud_pfn = ULONG_MAX;
1007                 args->pmd_pfn = ULONG_MAX;
1008                 args->pte_pfn = ULONG_MAX;
1009         }
1010
1011         if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
1012             has_transparent_hugepage() &&
1013             args->pmd_pfn != ULONG_MAX) {
1014                 if (args->is_contiguous_page) {
1015                         free_contig_range(args->pmd_pfn, (1 << HPAGE_PMD_ORDER));
1016                 } else {
1017                         page = pfn_to_page(args->pmd_pfn);
1018                         __free_pages(page, HPAGE_PMD_ORDER);
1019                 }
1020
1021                 args->pmd_pfn = ULONG_MAX;
1022                 args->pte_pfn = ULONG_MAX;
1023         }
1024
1025         if (args->pte_pfn != ULONG_MAX) {
1026                 page = pfn_to_page(args->pte_pfn);
1027                 __free_pages(page, 0);
1028
1029                 args->pte_pfn = ULONG_MAX;
1030         }
1031
1032         /* Free page table entries */
1033         if (args->start_ptep) {
1034                 pte_free(args->mm, args->start_ptep);
1035                 mm_dec_nr_ptes(args->mm);
1036         }
1037
1038         if (args->start_pmdp) {
1039                 pmd_free(args->mm, args->start_pmdp);
1040                 mm_dec_nr_pmds(args->mm);
1041         }
1042
1043         if (args->start_pudp) {
1044                 pud_free(args->mm, args->start_pudp);
1045                 mm_dec_nr_puds(args->mm);
1046         }
1047
1048         if (args->start_p4dp)
1049                 p4d_free(args->mm, args->start_p4dp);
1050
1051         /* Free vma and mm struct */
1052         if (args->vma)
1053                 vm_area_free(args->vma);
1054
1055         if (args->mm)
1056                 mmdrop(args->mm);
1057 }
1058
1059 static struct page * __init
1060 debug_vm_pgtable_alloc_huge_page(struct pgtable_debug_args *args, int order)
1061 {
1062         struct page *page = NULL;
1063
1064 #ifdef CONFIG_CONTIG_ALLOC
1065         if (order >= MAX_ORDER) {
1066                 page = alloc_contig_pages((1 << order), GFP_KERNEL,
1067                                           first_online_node, NULL);
1068                 if (page) {
1069                         args->is_contiguous_page = true;
1070                         return page;
1071                 }
1072         }
1073 #endif
1074
1075         if (order < MAX_ORDER)
1076                 page = alloc_pages(GFP_KERNEL, order);
1077
1078         return page;
1079 }
1080
1081 static int __init init_args(struct pgtable_debug_args *args)
1082 {
1083         struct page *page = NULL;
1084         phys_addr_t phys;
1085         int ret = 0;
1086
1087         /*
1088          * Initialize the debugging data.
1089          *
1090          * __P000 (or even __S000) will help create page table entries with
1091          * PROT_NONE permission as required for pxx_protnone_tests().
1092          */
1093         memset(args, 0, sizeof(*args));
1094         args->vaddr              = get_random_vaddr();
1095         args->page_prot          = vm_get_page_prot(VMFLAGS);
1096         args->page_prot_none     = __P000;
1097         args->is_contiguous_page = false;
1098         args->pud_pfn            = ULONG_MAX;
1099         args->pmd_pfn            = ULONG_MAX;
1100         args->pte_pfn            = ULONG_MAX;
1101         args->fixed_pgd_pfn      = ULONG_MAX;
1102         args->fixed_p4d_pfn      = ULONG_MAX;
1103         args->fixed_pud_pfn      = ULONG_MAX;
1104         args->fixed_pmd_pfn      = ULONG_MAX;
1105         args->fixed_pte_pfn      = ULONG_MAX;
1106
1107         /* Allocate mm and vma */
1108         args->mm = mm_alloc();
1109         if (!args->mm) {
1110                 pr_err("Failed to allocate mm struct\n");
1111                 ret = -ENOMEM;
1112                 goto error;
1113         }
1114
1115         args->vma = vm_area_alloc(args->mm);
1116         if (!args->vma) {
1117                 pr_err("Failed to allocate vma\n");
1118                 ret = -ENOMEM;
1119                 goto error;
1120         }
1121
1122         /*
1123          * Allocate page table entries. They will be modified in the tests.
1124          * Lets save the page table entries so that they can be released
1125          * when the tests are completed.
1126          */
1127         args->pgdp = pgd_offset(args->mm, args->vaddr);
1128         args->p4dp = p4d_alloc(args->mm, args->pgdp, args->vaddr);
1129         if (!args->p4dp) {
1130                 pr_err("Failed to allocate p4d entries\n");
1131                 ret = -ENOMEM;
1132                 goto error;
1133         }
1134         args->start_p4dp = p4d_offset(args->pgdp, 0UL);
1135         WARN_ON(!args->start_p4dp);
1136
1137         args->pudp = pud_alloc(args->mm, args->p4dp, args->vaddr);
1138         if (!args->pudp) {
1139                 pr_err("Failed to allocate pud entries\n");
1140                 ret = -ENOMEM;
1141                 goto error;
1142         }
1143         args->start_pudp = pud_offset(args->p4dp, 0UL);
1144         WARN_ON(!args->start_pudp);
1145
1146         args->pmdp = pmd_alloc(args->mm, args->pudp, args->vaddr);
1147         if (!args->pmdp) {
1148                 pr_err("Failed to allocate pmd entries\n");
1149                 ret = -ENOMEM;
1150                 goto error;
1151         }
1152         args->start_pmdp = pmd_offset(args->pudp, 0UL);
1153         WARN_ON(!args->start_pmdp);
1154
1155         if (pte_alloc(args->mm, args->pmdp)) {
1156                 pr_err("Failed to allocate pte entries\n");
1157                 ret = -ENOMEM;
1158                 goto error;
1159         }
1160         args->start_ptep = pmd_pgtable(READ_ONCE(*args->pmdp));
1161         WARN_ON(!args->start_ptep);
1162
1163         /*
1164          * PFN for mapping at PTE level is determined from a standard kernel
1165          * text symbol. But pfns for higher page table levels are derived by
1166          * masking lower bits of this real pfn. These derived pfns might not
1167          * exist on the platform but that does not really matter as pfn_pxx()
1168          * helpers will still create appropriate entries for the test. This
1169          * helps avoid large memory block allocations to be used for mapping
1170          * at higher page table levels in some of the tests.
1171          */
1172         phys = __pa_symbol(&start_kernel);
1173         args->fixed_pgd_pfn = __phys_to_pfn(phys & PGDIR_MASK);
1174         args->fixed_p4d_pfn = __phys_to_pfn(phys & P4D_MASK);
1175         args->fixed_pud_pfn = __phys_to_pfn(phys & PUD_MASK);
1176         args->fixed_pmd_pfn = __phys_to_pfn(phys & PMD_MASK);
1177         args->fixed_pte_pfn = __phys_to_pfn(phys & PAGE_MASK);
1178         WARN_ON(!pfn_valid(args->fixed_pte_pfn));
1179
1180         /*
1181          * Allocate (huge) pages because some of the tests need to access
1182          * the data in the pages. The corresponding tests will be skipped
1183          * if we fail to allocate (huge) pages.
1184          */
1185         if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
1186             IS_ENABLED(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD) &&
1187             has_transparent_hugepage()) {
1188                 page = debug_vm_pgtable_alloc_huge_page(args,
1189                                 HPAGE_PUD_SHIFT - PAGE_SHIFT);
1190                 if (page) {
1191                         args->pud_pfn = page_to_pfn(page);
1192                         args->pmd_pfn = args->pud_pfn;
1193                         args->pte_pfn = args->pud_pfn;
1194                         return 0;
1195                 }
1196         }
1197
1198         if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
1199             has_transparent_hugepage()) {
1200                 page = debug_vm_pgtable_alloc_huge_page(args, HPAGE_PMD_ORDER);
1201                 if (page) {
1202                         args->pmd_pfn = page_to_pfn(page);
1203                         args->pte_pfn = args->pmd_pfn;
1204                         return 0;
1205                 }
1206         }
1207
1208         page = alloc_pages(GFP_KERNEL, 0);
1209         if (page)
1210                 args->pte_pfn = page_to_pfn(page);
1211
1212         return 0;
1213
1214 error:
1215         destroy_args(args);
1216         return ret;
1217 }
1218
1219 static int __init debug_vm_pgtable(void)
1220 {
1221         struct pgtable_debug_args args;
1222         struct vm_area_struct *vma;
1223         struct mm_struct *mm;
1224         pgd_t *pgdp;
1225         p4d_t *p4dp, *saved_p4dp;
1226         pud_t *pudp, *saved_pudp;
1227         pmd_t *pmdp, *saved_pmdp, pmd;
1228         pte_t *ptep;
1229         pgtable_t saved_ptep;
1230         pgprot_t prot, protnone;
1231         phys_addr_t paddr;
1232         unsigned long vaddr, pte_aligned, pmd_aligned;
1233         unsigned long pud_aligned;
1234         spinlock_t *ptl = NULL;
1235         int idx, ret;
1236
1237         pr_info("Validating architecture page table helpers\n");
1238         ret = init_args(&args);
1239         if (ret)
1240                 return ret;
1241
1242         prot = vm_get_page_prot(VMFLAGS);
1243         vaddr = get_random_vaddr();
1244         mm = mm_alloc();
1245         if (!mm) {
1246                 pr_err("mm_struct allocation failed\n");
1247                 return 1;
1248         }
1249
1250         /*
1251          * __P000 (or even __S000) will help create page table entries with
1252          * PROT_NONE permission as required for pxx_protnone_tests().
1253          */
1254         protnone = __P000;
1255
1256         vma = vm_area_alloc(mm);
1257         if (!vma) {
1258                 pr_err("vma allocation failed\n");
1259                 return 1;
1260         }
1261
1262         /*
1263          * PFN for mapping at PTE level is determined from a standard kernel
1264          * text symbol. But pfns for higher page table levels are derived by
1265          * masking lower bits of this real pfn. These derived pfns might not
1266          * exist on the platform but that does not really matter as pfn_pxx()
1267          * helpers will still create appropriate entries for the test. This
1268          * helps avoid large memory block allocations to be used for mapping
1269          * at higher page table levels.
1270          */
1271         paddr = __pa_symbol(&start_kernel);
1272
1273         pte_aligned = (paddr & PAGE_MASK) >> PAGE_SHIFT;
1274         pmd_aligned = (paddr & PMD_MASK) >> PAGE_SHIFT;
1275         pud_aligned = (paddr & PUD_MASK) >> PAGE_SHIFT;
1276         WARN_ON(!pfn_valid(pte_aligned));
1277
1278         pgdp = pgd_offset(mm, vaddr);
1279         p4dp = p4d_alloc(mm, pgdp, vaddr);
1280         pudp = pud_alloc(mm, p4dp, vaddr);
1281         pmdp = pmd_alloc(mm, pudp, vaddr);
1282         /*
1283          * Allocate pgtable_t
1284          */
1285         if (pte_alloc(mm, pmdp)) {
1286                 pr_err("pgtable allocation failed\n");
1287                 return 1;
1288         }
1289
1290         /*
1291          * Save all the page table page addresses as the page table
1292          * entries will be used for testing with random or garbage
1293          * values. These saved addresses will be used for freeing
1294          * page table pages.
1295          */
1296         pmd = READ_ONCE(*pmdp);
1297         saved_p4dp = p4d_offset(pgdp, 0UL);
1298         saved_pudp = pud_offset(p4dp, 0UL);
1299         saved_pmdp = pmd_offset(pudp, 0UL);
1300         saved_ptep = pmd_pgtable(pmd);
1301
1302         /*
1303          * Iterate over the protection_map[] to make sure that all
1304          * the basic page table transformation validations just hold
1305          * true irrespective of the starting protection value for a
1306          * given page table entry.
1307          */
1308         for (idx = 0; idx < ARRAY_SIZE(protection_map); idx++) {
1309                 pte_basic_tests(&args, idx);
1310                 pmd_basic_tests(&args, idx);
1311                 pud_basic_tests(&args, idx);
1312         }
1313
1314         /*
1315          * Both P4D and PGD level tests are very basic which do not
1316          * involve creating page table entries from the protection
1317          * value and the given pfn. Hence just keep them out from
1318          * the above iteration for now to save some test execution
1319          * time.
1320          */
1321         p4d_basic_tests(&args);
1322         pgd_basic_tests(&args);
1323
1324         pmd_leaf_tests(&args);
1325         pud_leaf_tests(&args);
1326
1327         pte_savedwrite_tests(&args);
1328         pmd_savedwrite_tests(&args);
1329
1330         pte_special_tests(pte_aligned, prot);
1331         pte_protnone_tests(pte_aligned, protnone);
1332         pmd_protnone_tests(pmd_aligned, protnone);
1333
1334         pte_devmap_tests(pte_aligned, prot);
1335         pmd_devmap_tests(pmd_aligned, prot);
1336         pud_devmap_tests(pud_aligned, prot);
1337
1338         pte_soft_dirty_tests(pte_aligned, prot);
1339         pmd_soft_dirty_tests(pmd_aligned, prot);
1340         pte_swap_soft_dirty_tests(pte_aligned, prot);
1341         pmd_swap_soft_dirty_tests(pmd_aligned, prot);
1342
1343         pte_swap_tests(pte_aligned, prot);
1344         pmd_swap_tests(pmd_aligned, prot);
1345
1346         swap_migration_tests();
1347
1348         pmd_thp_tests(pmd_aligned, prot);
1349         pud_thp_tests(pud_aligned, prot);
1350
1351         hugetlb_basic_tests(&args);
1352
1353         /*
1354          * Page table modifying tests. They need to hold
1355          * proper page table lock.
1356          */
1357
1358         ptep = pte_offset_map_lock(mm, pmdp, vaddr, &ptl);
1359         pte_clear_tests(mm, ptep, pte_aligned, vaddr, prot);
1360         pte_advanced_tests(mm, vma, ptep, pte_aligned, vaddr, prot);
1361         pte_unmap_unlock(ptep, ptl);
1362
1363         ptl = pmd_lock(mm, pmdp);
1364         pmd_clear_tests(mm, pmdp);
1365         pmd_advanced_tests(mm, vma, pmdp, pmd_aligned, vaddr, prot, saved_ptep);
1366         pmd_huge_tests(pmdp, pmd_aligned, prot);
1367         pmd_populate_tests(mm, pmdp, saved_ptep);
1368         spin_unlock(ptl);
1369
1370         ptl = pud_lock(mm, pudp);
1371         pud_clear_tests(mm, pudp);
1372         pud_advanced_tests(mm, vma, pudp, pud_aligned, vaddr, prot);
1373         pud_huge_tests(pudp, pud_aligned, prot);
1374         pud_populate_tests(mm, pudp, saved_pmdp);
1375         spin_unlock(ptl);
1376
1377         spin_lock(&mm->page_table_lock);
1378         p4d_clear_tests(mm, p4dp);
1379         pgd_clear_tests(mm, pgdp);
1380         p4d_populate_tests(mm, p4dp, saved_pudp);
1381         pgd_populate_tests(mm, pgdp, saved_p4dp);
1382         spin_unlock(&mm->page_table_lock);
1383
1384         p4d_free(mm, saved_p4dp);
1385         pud_free(mm, saved_pudp);
1386         pmd_free(mm, saved_pmdp);
1387         pte_free(mm, saved_ptep);
1388
1389         vm_area_free(vma);
1390         mm_dec_nr_puds(mm);
1391         mm_dec_nr_pmds(mm);
1392         mm_dec_nr_ptes(mm);
1393         mmdrop(mm);
1394
1395         destroy_args(&args);
1396         return 0;
1397 }
1398 late_initcall(debug_vm_pgtable);