Back to home page

OSCL-LXR
 
 

     

0001 // SPDX-License-Identifier: GPL-2.0-only
0002 /*
0003  * This kernel test validates architecture page table helpers and
0004  * accessors and helps in verifying their continued compliance with
0005  * expected generic MM semantics.
0006  *
0007  * Copyright (C) 2019 ARM Ltd.
0008  *
0009  * Author: Anshuman Khandual <anshuman.khandual@arm.com>
0010  */
0011 #define pr_fmt(fmt) "debug_vm_pgtable: [%-25s]: " fmt, __func__
0012 
0013 #include <linux/gfp.h>
0014 #include <linux/highmem.h>
0015 #include <linux/hugetlb.h>
0016 #include <linux/kernel.h>
0017 #include <linux/kconfig.h>
0018 #include <linux/mm.h>
0019 #include <linux/mman.h>
0020 #include <linux/mm_types.h>
0021 #include <linux/module.h>
0022 #include <linux/pfn_t.h>
0023 #include <linux/printk.h>
0024 #include <linux/pgtable.h>
0025 #include <linux/random.h>
0026 #include <linux/spinlock.h>
0027 #include <linux/swap.h>
0028 #include <linux/swapops.h>
0029 #include <linux/start_kernel.h>
0030 #include <linux/sched/mm.h>
0031 #include <linux/io.h>
0032 
0033 #include <asm/cacheflush.h>
0034 #include <asm/pgalloc.h>
0035 #include <asm/tlbflush.h>
0036 
0037 /*
0038  * Please refer Documentation/mm/arch_pgtable_helpers.rst for the semantics
0039  * expectations that are being validated here. All future changes in here
0040  * or the documentation need to be in sync.
0041  */
0042 
0043 #define VMFLAGS (VM_READ|VM_WRITE|VM_EXEC)
0044 
0045 /*
0046  * On s390 platform, the lower 4 bits are used to identify given page table
0047  * entry type. But these bits might affect the ability to clear entries with
0048  * pxx_clear() because of how dynamic page table folding works on s390. So
0049  * while loading up the entries do not change the lower 4 bits. It does not
0050  * have affect any other platform. Also avoid the 62nd bit on ppc64 that is
0051  * used to mark a pte entry.
0052  */
0053 #define S390_SKIP_MASK      GENMASK(3, 0)
0054 #if __BITS_PER_LONG == 64
0055 #define PPC64_SKIP_MASK     GENMASK(62, 62)
0056 #else
0057 #define PPC64_SKIP_MASK     0x0
0058 #endif
0059 #define ARCH_SKIP_MASK (S390_SKIP_MASK | PPC64_SKIP_MASK)
0060 #define RANDOM_ORVALUE (GENMASK(BITS_PER_LONG - 1, 0) & ~ARCH_SKIP_MASK)
0061 #define RANDOM_NZVALUE  GENMASK(7, 0)
0062 
0063 struct pgtable_debug_args {
0064     struct mm_struct    *mm;
0065     struct vm_area_struct   *vma;
0066 
0067     pgd_t           *pgdp;
0068     p4d_t           *p4dp;
0069     pud_t           *pudp;
0070     pmd_t           *pmdp;
0071     pte_t           *ptep;
0072 
0073     p4d_t           *start_p4dp;
0074     pud_t           *start_pudp;
0075     pmd_t           *start_pmdp;
0076     pgtable_t       start_ptep;
0077 
0078     unsigned long       vaddr;
0079     pgprot_t        page_prot;
0080     pgprot_t        page_prot_none;
0081 
0082     bool            is_contiguous_page;
0083     unsigned long       pud_pfn;
0084     unsigned long       pmd_pfn;
0085     unsigned long       pte_pfn;
0086 
0087     unsigned long       fixed_pgd_pfn;
0088     unsigned long       fixed_p4d_pfn;
0089     unsigned long       fixed_pud_pfn;
0090     unsigned long       fixed_pmd_pfn;
0091     unsigned long       fixed_pte_pfn;
0092 };
0093 
0094 static void __init pte_basic_tests(struct pgtable_debug_args *args, int idx)
0095 {
0096     pgprot_t prot = vm_get_page_prot(idx);
0097     pte_t pte = pfn_pte(args->fixed_pte_pfn, prot);
0098     unsigned long val = idx, *ptr = &val;
0099 
0100     pr_debug("Validating PTE basic (%pGv)\n", ptr);
0101 
0102     /*
0103      * This test needs to be executed after the given page table entry
0104      * is created with pfn_pte() to make sure that vm_get_page_prot(idx)
0105      * does not have the dirty bit enabled from the beginning. This is
0106      * important for platforms like arm64 where (!PTE_RDONLY) indicate
0107      * dirty bit being set.
0108      */
0109     WARN_ON(pte_dirty(pte_wrprotect(pte)));
0110 
0111     WARN_ON(!pte_same(pte, pte));
0112     WARN_ON(!pte_young(pte_mkyoung(pte_mkold(pte))));
0113     WARN_ON(!pte_dirty(pte_mkdirty(pte_mkclean(pte))));
0114     WARN_ON(!pte_write(pte_mkwrite(pte_wrprotect(pte))));
0115     WARN_ON(pte_young(pte_mkold(pte_mkyoung(pte))));
0116     WARN_ON(pte_dirty(pte_mkclean(pte_mkdirty(pte))));
0117     WARN_ON(pte_write(pte_wrprotect(pte_mkwrite(pte))));
0118     WARN_ON(pte_dirty(pte_wrprotect(pte_mkclean(pte))));
0119     WARN_ON(!pte_dirty(pte_wrprotect(pte_mkdirty(pte))));
0120 }
0121 
0122 static void __init pte_advanced_tests(struct pgtable_debug_args *args)
0123 {
0124     struct page *page;
0125     pte_t pte;
0126 
0127     /*
0128      * Architectures optimize set_pte_at by avoiding TLB flush.
0129      * This requires set_pte_at to be not used to update an
0130      * existing pte entry. Clear pte before we do set_pte_at
0131      *
0132      * flush_dcache_page() is called after set_pte_at() to clear
0133      * PG_arch_1 for the page on ARM64. The page flag isn't cleared
0134      * when it's released and page allocation check will fail when
0135      * the page is allocated again. For architectures other than ARM64,
0136      * the unexpected overhead of cache flushing is acceptable.
0137      */
0138     page = (args->pte_pfn != ULONG_MAX) ? pfn_to_page(args->pte_pfn) : NULL;
0139     if (!page)
0140         return;
0141 
0142     pr_debug("Validating PTE advanced\n");
0143     pte = pfn_pte(args->pte_pfn, args->page_prot);
0144     set_pte_at(args->mm, args->vaddr, args->ptep, pte);
0145     flush_dcache_page(page);
0146     ptep_set_wrprotect(args->mm, args->vaddr, args->ptep);
0147     pte = ptep_get(args->ptep);
0148     WARN_ON(pte_write(pte));
0149     ptep_get_and_clear(args->mm, args->vaddr, args->ptep);
0150     pte = ptep_get(args->ptep);
0151     WARN_ON(!pte_none(pte));
0152 
0153     pte = pfn_pte(args->pte_pfn, args->page_prot);
0154     pte = pte_wrprotect(pte);
0155     pte = pte_mkclean(pte);
0156     set_pte_at(args->mm, args->vaddr, args->ptep, pte);
0157     flush_dcache_page(page);
0158     pte = pte_mkwrite(pte);
0159     pte = pte_mkdirty(pte);
0160     ptep_set_access_flags(args->vma, args->vaddr, args->ptep, pte, 1);
0161     pte = ptep_get(args->ptep);
0162     WARN_ON(!(pte_write(pte) && pte_dirty(pte)));
0163     ptep_get_and_clear_full(args->mm, args->vaddr, args->ptep, 1);
0164     pte = ptep_get(args->ptep);
0165     WARN_ON(!pte_none(pte));
0166 
0167     pte = pfn_pte(args->pte_pfn, args->page_prot);
0168     pte = pte_mkyoung(pte);
0169     set_pte_at(args->mm, args->vaddr, args->ptep, pte);
0170     flush_dcache_page(page);
0171     ptep_test_and_clear_young(args->vma, args->vaddr, args->ptep);
0172     pte = ptep_get(args->ptep);
0173     WARN_ON(pte_young(pte));
0174 
0175     ptep_get_and_clear_full(args->mm, args->vaddr, args->ptep, 1);
0176 }
0177 
0178 static void __init pte_savedwrite_tests(struct pgtable_debug_args *args)
0179 {
0180     pte_t pte = pfn_pte(args->fixed_pte_pfn, args->page_prot_none);
0181 
0182     if (!IS_ENABLED(CONFIG_NUMA_BALANCING))
0183         return;
0184 
0185     pr_debug("Validating PTE saved write\n");
0186     WARN_ON(!pte_savedwrite(pte_mk_savedwrite(pte_clear_savedwrite(pte))));
0187     WARN_ON(pte_savedwrite(pte_clear_savedwrite(pte_mk_savedwrite(pte))));
0188 }
0189 
0190 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
0191 static void __init pmd_basic_tests(struct pgtable_debug_args *args, int idx)
0192 {
0193     pgprot_t prot = vm_get_page_prot(idx);
0194     unsigned long val = idx, *ptr = &val;
0195     pmd_t pmd;
0196 
0197     if (!has_transparent_hugepage())
0198         return;
0199 
0200     pr_debug("Validating PMD basic (%pGv)\n", ptr);
0201     pmd = pfn_pmd(args->fixed_pmd_pfn, prot);
0202 
0203     /*
0204      * This test needs to be executed after the given page table entry
0205      * is created with pfn_pmd() to make sure that vm_get_page_prot(idx)
0206      * does not have the dirty bit enabled from the beginning. This is
0207      * important for platforms like arm64 where (!PTE_RDONLY) indicate
0208      * dirty bit being set.
0209      */
0210     WARN_ON(pmd_dirty(pmd_wrprotect(pmd)));
0211 
0212 
0213     WARN_ON(!pmd_same(pmd, pmd));
0214     WARN_ON(!pmd_young(pmd_mkyoung(pmd_mkold(pmd))));
0215     WARN_ON(!pmd_dirty(pmd_mkdirty(pmd_mkclean(pmd))));
0216     WARN_ON(!pmd_write(pmd_mkwrite(pmd_wrprotect(pmd))));
0217     WARN_ON(pmd_young(pmd_mkold(pmd_mkyoung(pmd))));
0218     WARN_ON(pmd_dirty(pmd_mkclean(pmd_mkdirty(pmd))));
0219     WARN_ON(pmd_write(pmd_wrprotect(pmd_mkwrite(pmd))));
0220     WARN_ON(pmd_dirty(pmd_wrprotect(pmd_mkclean(pmd))));
0221     WARN_ON(!pmd_dirty(pmd_wrprotect(pmd_mkdirty(pmd))));
0222     /*
0223      * A huge page does not point to next level page table
0224      * entry. Hence this must qualify as pmd_bad().
0225      */
0226     WARN_ON(!pmd_bad(pmd_mkhuge(pmd)));
0227 }
0228 
0229 static void __init pmd_advanced_tests(struct pgtable_debug_args *args)
0230 {
0231     struct page *page;
0232     pmd_t pmd;
0233     unsigned long vaddr = args->vaddr;
0234 
0235     if (!has_transparent_hugepage())
0236         return;
0237 
0238     page = (args->pmd_pfn != ULONG_MAX) ? pfn_to_page(args->pmd_pfn) : NULL;
0239     if (!page)
0240         return;
0241 
0242     /*
0243      * flush_dcache_page() is called after set_pmd_at() to clear
0244      * PG_arch_1 for the page on ARM64. The page flag isn't cleared
0245      * when it's released and page allocation check will fail when
0246      * the page is allocated again. For architectures other than ARM64,
0247      * the unexpected overhead of cache flushing is acceptable.
0248      */
0249     pr_debug("Validating PMD advanced\n");
0250     /* Align the address wrt HPAGE_PMD_SIZE */
0251     vaddr &= HPAGE_PMD_MASK;
0252 
0253     pgtable_trans_huge_deposit(args->mm, args->pmdp, args->start_ptep);
0254 
0255     pmd = pfn_pmd(args->pmd_pfn, args->page_prot);
0256     set_pmd_at(args->mm, vaddr, args->pmdp, pmd);
0257     flush_dcache_page(page);
0258     pmdp_set_wrprotect(args->mm, vaddr, args->pmdp);
0259     pmd = READ_ONCE(*args->pmdp);
0260     WARN_ON(pmd_write(pmd));
0261     pmdp_huge_get_and_clear(args->mm, vaddr, args->pmdp);
0262     pmd = READ_ONCE(*args->pmdp);
0263     WARN_ON(!pmd_none(pmd));
0264 
0265     pmd = pfn_pmd(args->pmd_pfn, args->page_prot);
0266     pmd = pmd_wrprotect(pmd);
0267     pmd = pmd_mkclean(pmd);
0268     set_pmd_at(args->mm, vaddr, args->pmdp, pmd);
0269     flush_dcache_page(page);
0270     pmd = pmd_mkwrite(pmd);
0271     pmd = pmd_mkdirty(pmd);
0272     pmdp_set_access_flags(args->vma, vaddr, args->pmdp, pmd, 1);
0273     pmd = READ_ONCE(*args->pmdp);
0274     WARN_ON(!(pmd_write(pmd) && pmd_dirty(pmd)));
0275     pmdp_huge_get_and_clear_full(args->vma, vaddr, args->pmdp, 1);
0276     pmd = READ_ONCE(*args->pmdp);
0277     WARN_ON(!pmd_none(pmd));
0278 
0279     pmd = pmd_mkhuge(pfn_pmd(args->pmd_pfn, args->page_prot));
0280     pmd = pmd_mkyoung(pmd);
0281     set_pmd_at(args->mm, vaddr, args->pmdp, pmd);
0282     flush_dcache_page(page);
0283     pmdp_test_and_clear_young(args->vma, vaddr, args->pmdp);
0284     pmd = READ_ONCE(*args->pmdp);
0285     WARN_ON(pmd_young(pmd));
0286 
0287     /*  Clear the pte entries  */
0288     pmdp_huge_get_and_clear(args->mm, vaddr, args->pmdp);
0289     pgtable_trans_huge_withdraw(args->mm, args->pmdp);
0290 }
0291 
0292 static void __init pmd_leaf_tests(struct pgtable_debug_args *args)
0293 {
0294     pmd_t pmd;
0295 
0296     if (!has_transparent_hugepage())
0297         return;
0298 
0299     pr_debug("Validating PMD leaf\n");
0300     pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot);
0301 
0302     /*
0303      * PMD based THP is a leaf entry.
0304      */
0305     pmd = pmd_mkhuge(pmd);
0306     WARN_ON(!pmd_leaf(pmd));
0307 }
0308 
0309 static void __init pmd_savedwrite_tests(struct pgtable_debug_args *args)
0310 {
0311     pmd_t pmd;
0312 
0313     if (!IS_ENABLED(CONFIG_NUMA_BALANCING))
0314         return;
0315 
0316     if (!has_transparent_hugepage())
0317         return;
0318 
0319     pr_debug("Validating PMD saved write\n");
0320     pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot_none);
0321     WARN_ON(!pmd_savedwrite(pmd_mk_savedwrite(pmd_clear_savedwrite(pmd))));
0322     WARN_ON(pmd_savedwrite(pmd_clear_savedwrite(pmd_mk_savedwrite(pmd))));
0323 }
0324 
0325 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
0326 static void __init pud_basic_tests(struct pgtable_debug_args *args, int idx)
0327 {
0328     pgprot_t prot = vm_get_page_prot(idx);
0329     unsigned long val = idx, *ptr = &val;
0330     pud_t pud;
0331 
0332     if (!has_transparent_hugepage())
0333         return;
0334 
0335     pr_debug("Validating PUD basic (%pGv)\n", ptr);
0336     pud = pfn_pud(args->fixed_pud_pfn, prot);
0337 
0338     /*
0339      * This test needs to be executed after the given page table entry
0340      * is created with pfn_pud() to make sure that vm_get_page_prot(idx)
0341      * does not have the dirty bit enabled from the beginning. This is
0342      * important for platforms like arm64 where (!PTE_RDONLY) indicate
0343      * dirty bit being set.
0344      */
0345     WARN_ON(pud_dirty(pud_wrprotect(pud)));
0346 
0347     WARN_ON(!pud_same(pud, pud));
0348     WARN_ON(!pud_young(pud_mkyoung(pud_mkold(pud))));
0349     WARN_ON(!pud_dirty(pud_mkdirty(pud_mkclean(pud))));
0350     WARN_ON(pud_dirty(pud_mkclean(pud_mkdirty(pud))));
0351     WARN_ON(!pud_write(pud_mkwrite(pud_wrprotect(pud))));
0352     WARN_ON(pud_write(pud_wrprotect(pud_mkwrite(pud))));
0353     WARN_ON(pud_young(pud_mkold(pud_mkyoung(pud))));
0354     WARN_ON(pud_dirty(pud_wrprotect(pud_mkclean(pud))));
0355     WARN_ON(!pud_dirty(pud_wrprotect(pud_mkdirty(pud))));
0356 
0357     if (mm_pmd_folded(args->mm))
0358         return;
0359 
0360     /*
0361      * A huge page does not point to next level page table
0362      * entry. Hence this must qualify as pud_bad().
0363      */
0364     WARN_ON(!pud_bad(pud_mkhuge(pud)));
0365 }
0366 
0367 static void __init pud_advanced_tests(struct pgtable_debug_args *args)
0368 {
0369     struct page *page;
0370     unsigned long vaddr = args->vaddr;
0371     pud_t pud;
0372 
0373     if (!has_transparent_hugepage())
0374         return;
0375 
0376     page = (args->pud_pfn != ULONG_MAX) ? pfn_to_page(args->pud_pfn) : NULL;
0377     if (!page)
0378         return;
0379 
0380     /*
0381      * flush_dcache_page() is called after set_pud_at() to clear
0382      * PG_arch_1 for the page on ARM64. The page flag isn't cleared
0383      * when it's released and page allocation check will fail when
0384      * the page is allocated again. For architectures other than ARM64,
0385      * the unexpected overhead of cache flushing is acceptable.
0386      */
0387     pr_debug("Validating PUD advanced\n");
0388     /* Align the address wrt HPAGE_PUD_SIZE */
0389     vaddr &= HPAGE_PUD_MASK;
0390 
0391     pud = pfn_pud(args->pud_pfn, args->page_prot);
0392     set_pud_at(args->mm, vaddr, args->pudp, pud);
0393     flush_dcache_page(page);
0394     pudp_set_wrprotect(args->mm, vaddr, args->pudp);
0395     pud = READ_ONCE(*args->pudp);
0396     WARN_ON(pud_write(pud));
0397 
0398 #ifndef __PAGETABLE_PMD_FOLDED
0399     pudp_huge_get_and_clear(args->mm, vaddr, args->pudp);
0400     pud = READ_ONCE(*args->pudp);
0401     WARN_ON(!pud_none(pud));
0402 #endif /* __PAGETABLE_PMD_FOLDED */
0403     pud = pfn_pud(args->pud_pfn, args->page_prot);
0404     pud = pud_wrprotect(pud);
0405     pud = pud_mkclean(pud);
0406     set_pud_at(args->mm, vaddr, args->pudp, pud);
0407     flush_dcache_page(page);
0408     pud = pud_mkwrite(pud);
0409     pud = pud_mkdirty(pud);
0410     pudp_set_access_flags(args->vma, vaddr, args->pudp, pud, 1);
0411     pud = READ_ONCE(*args->pudp);
0412     WARN_ON(!(pud_write(pud) && pud_dirty(pud)));
0413 
0414 #ifndef __PAGETABLE_PMD_FOLDED
0415     pudp_huge_get_and_clear_full(args->mm, vaddr, args->pudp, 1);
0416     pud = READ_ONCE(*args->pudp);
0417     WARN_ON(!pud_none(pud));
0418 #endif /* __PAGETABLE_PMD_FOLDED */
0419 
0420     pud = pfn_pud(args->pud_pfn, args->page_prot);
0421     pud = pud_mkyoung(pud);
0422     set_pud_at(args->mm, vaddr, args->pudp, pud);
0423     flush_dcache_page(page);
0424     pudp_test_and_clear_young(args->vma, vaddr, args->pudp);
0425     pud = READ_ONCE(*args->pudp);
0426     WARN_ON(pud_young(pud));
0427 
0428     pudp_huge_get_and_clear(args->mm, vaddr, args->pudp);
0429 }
0430 
0431 static void __init pud_leaf_tests(struct pgtable_debug_args *args)
0432 {
0433     pud_t pud;
0434 
0435     if (!has_transparent_hugepage())
0436         return;
0437 
0438     pr_debug("Validating PUD leaf\n");
0439     pud = pfn_pud(args->fixed_pud_pfn, args->page_prot);
0440     /*
0441      * PUD based THP is a leaf entry.
0442      */
0443     pud = pud_mkhuge(pud);
0444     WARN_ON(!pud_leaf(pud));
0445 }
0446 #else  /* !CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
0447 static void __init pud_basic_tests(struct pgtable_debug_args *args, int idx) { }
0448 static void __init pud_advanced_tests(struct pgtable_debug_args *args) { }
0449 static void __init pud_leaf_tests(struct pgtable_debug_args *args) { }
0450 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
0451 #else  /* !CONFIG_TRANSPARENT_HUGEPAGE */
0452 static void __init pmd_basic_tests(struct pgtable_debug_args *args, int idx) { }
0453 static void __init pud_basic_tests(struct pgtable_debug_args *args, int idx) { }
0454 static void __init pmd_advanced_tests(struct pgtable_debug_args *args) { }
0455 static void __init pud_advanced_tests(struct pgtable_debug_args *args) { }
0456 static void __init pmd_leaf_tests(struct pgtable_debug_args *args) { }
0457 static void __init pud_leaf_tests(struct pgtable_debug_args *args) { }
0458 static void __init pmd_savedwrite_tests(struct pgtable_debug_args *args) { }
0459 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
0460 
0461 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
0462 static void __init pmd_huge_tests(struct pgtable_debug_args *args)
0463 {
0464     pmd_t pmd;
0465 
0466     if (!arch_vmap_pmd_supported(args->page_prot))
0467         return;
0468 
0469     pr_debug("Validating PMD huge\n");
0470     /*
0471      * X86 defined pmd_set_huge() verifies that the given
0472      * PMD is not a populated non-leaf entry.
0473      */
0474     WRITE_ONCE(*args->pmdp, __pmd(0));
0475     WARN_ON(!pmd_set_huge(args->pmdp, __pfn_to_phys(args->fixed_pmd_pfn), args->page_prot));
0476     WARN_ON(!pmd_clear_huge(args->pmdp));
0477     pmd = READ_ONCE(*args->pmdp);
0478     WARN_ON(!pmd_none(pmd));
0479 }
0480 
0481 static void __init pud_huge_tests(struct pgtable_debug_args *args)
0482 {
0483     pud_t pud;
0484 
0485     if (!arch_vmap_pud_supported(args->page_prot))
0486         return;
0487 
0488     pr_debug("Validating PUD huge\n");
0489     /*
0490      * X86 defined pud_set_huge() verifies that the given
0491      * PUD is not a populated non-leaf entry.
0492      */
0493     WRITE_ONCE(*args->pudp, __pud(0));
0494     WARN_ON(!pud_set_huge(args->pudp, __pfn_to_phys(args->fixed_pud_pfn), args->page_prot));
0495     WARN_ON(!pud_clear_huge(args->pudp));
0496     pud = READ_ONCE(*args->pudp);
0497     WARN_ON(!pud_none(pud));
0498 }
0499 #else /* !CONFIG_HAVE_ARCH_HUGE_VMAP */
0500 static void __init pmd_huge_tests(struct pgtable_debug_args *args) { }
0501 static void __init pud_huge_tests(struct pgtable_debug_args *args) { }
0502 #endif /* CONFIG_HAVE_ARCH_HUGE_VMAP */
0503 
0504 static void __init p4d_basic_tests(struct pgtable_debug_args *args)
0505 {
0506     p4d_t p4d;
0507 
0508     pr_debug("Validating P4D basic\n");
0509     memset(&p4d, RANDOM_NZVALUE, sizeof(p4d_t));
0510     WARN_ON(!p4d_same(p4d, p4d));
0511 }
0512 
0513 static void __init pgd_basic_tests(struct pgtable_debug_args *args)
0514 {
0515     pgd_t pgd;
0516 
0517     pr_debug("Validating PGD basic\n");
0518     memset(&pgd, RANDOM_NZVALUE, sizeof(pgd_t));
0519     WARN_ON(!pgd_same(pgd, pgd));
0520 }
0521 
0522 #ifndef __PAGETABLE_PUD_FOLDED
0523 static void __init pud_clear_tests(struct pgtable_debug_args *args)
0524 {
0525     pud_t pud = READ_ONCE(*args->pudp);
0526 
0527     if (mm_pmd_folded(args->mm))
0528         return;
0529 
0530     pr_debug("Validating PUD clear\n");
0531     pud = __pud(pud_val(pud) | RANDOM_ORVALUE);
0532     WRITE_ONCE(*args->pudp, pud);
0533     pud_clear(args->pudp);
0534     pud = READ_ONCE(*args->pudp);
0535     WARN_ON(!pud_none(pud));
0536 }
0537 
0538 static void __init pud_populate_tests(struct pgtable_debug_args *args)
0539 {
0540     pud_t pud;
0541 
0542     if (mm_pmd_folded(args->mm))
0543         return;
0544 
0545     pr_debug("Validating PUD populate\n");
0546     /*
0547      * This entry points to next level page table page.
0548      * Hence this must not qualify as pud_bad().
0549      */
0550     pud_populate(args->mm, args->pudp, args->start_pmdp);
0551     pud = READ_ONCE(*args->pudp);
0552     WARN_ON(pud_bad(pud));
0553 }
0554 #else  /* !__PAGETABLE_PUD_FOLDED */
0555 static void __init pud_clear_tests(struct pgtable_debug_args *args) { }
0556 static void __init pud_populate_tests(struct pgtable_debug_args *args) { }
0557 #endif /* PAGETABLE_PUD_FOLDED */
0558 
0559 #ifndef __PAGETABLE_P4D_FOLDED
0560 static void __init p4d_clear_tests(struct pgtable_debug_args *args)
0561 {
0562     p4d_t p4d = READ_ONCE(*args->p4dp);
0563 
0564     if (mm_pud_folded(args->mm))
0565         return;
0566 
0567     pr_debug("Validating P4D clear\n");
0568     p4d = __p4d(p4d_val(p4d) | RANDOM_ORVALUE);
0569     WRITE_ONCE(*args->p4dp, p4d);
0570     p4d_clear(args->p4dp);
0571     p4d = READ_ONCE(*args->p4dp);
0572     WARN_ON(!p4d_none(p4d));
0573 }
0574 
0575 static void __init p4d_populate_tests(struct pgtable_debug_args *args)
0576 {
0577     p4d_t p4d;
0578 
0579     if (mm_pud_folded(args->mm))
0580         return;
0581 
0582     pr_debug("Validating P4D populate\n");
0583     /*
0584      * This entry points to next level page table page.
0585      * Hence this must not qualify as p4d_bad().
0586      */
0587     pud_clear(args->pudp);
0588     p4d_clear(args->p4dp);
0589     p4d_populate(args->mm, args->p4dp, args->start_pudp);
0590     p4d = READ_ONCE(*args->p4dp);
0591     WARN_ON(p4d_bad(p4d));
0592 }
0593 
0594 static void __init pgd_clear_tests(struct pgtable_debug_args *args)
0595 {
0596     pgd_t pgd = READ_ONCE(*(args->pgdp));
0597 
0598     if (mm_p4d_folded(args->mm))
0599         return;
0600 
0601     pr_debug("Validating PGD clear\n");
0602     pgd = __pgd(pgd_val(pgd) | RANDOM_ORVALUE);
0603     WRITE_ONCE(*args->pgdp, pgd);
0604     pgd_clear(args->pgdp);
0605     pgd = READ_ONCE(*args->pgdp);
0606     WARN_ON(!pgd_none(pgd));
0607 }
0608 
0609 static void __init pgd_populate_tests(struct pgtable_debug_args *args)
0610 {
0611     pgd_t pgd;
0612 
0613     if (mm_p4d_folded(args->mm))
0614         return;
0615 
0616     pr_debug("Validating PGD populate\n");
0617     /*
0618      * This entry points to next level page table page.
0619      * Hence this must not qualify as pgd_bad().
0620      */
0621     p4d_clear(args->p4dp);
0622     pgd_clear(args->pgdp);
0623     pgd_populate(args->mm, args->pgdp, args->start_p4dp);
0624     pgd = READ_ONCE(*args->pgdp);
0625     WARN_ON(pgd_bad(pgd));
0626 }
0627 #else  /* !__PAGETABLE_P4D_FOLDED */
0628 static void __init p4d_clear_tests(struct pgtable_debug_args *args) { }
0629 static void __init pgd_clear_tests(struct pgtable_debug_args *args) { }
0630 static void __init p4d_populate_tests(struct pgtable_debug_args *args) { }
0631 static void __init pgd_populate_tests(struct pgtable_debug_args *args) { }
0632 #endif /* PAGETABLE_P4D_FOLDED */
0633 
0634 static void __init pte_clear_tests(struct pgtable_debug_args *args)
0635 {
0636     struct page *page;
0637     pte_t pte = pfn_pte(args->pte_pfn, args->page_prot);
0638 
0639     page = (args->pte_pfn != ULONG_MAX) ? pfn_to_page(args->pte_pfn) : NULL;
0640     if (!page)
0641         return;
0642 
0643     /*
0644      * flush_dcache_page() is called after set_pte_at() to clear
0645      * PG_arch_1 for the page on ARM64. The page flag isn't cleared
0646      * when it's released and page allocation check will fail when
0647      * the page is allocated again. For architectures other than ARM64,
0648      * the unexpected overhead of cache flushing is acceptable.
0649      */
0650     pr_debug("Validating PTE clear\n");
0651 #ifndef CONFIG_RISCV
0652     pte = __pte(pte_val(pte) | RANDOM_ORVALUE);
0653 #endif
0654     set_pte_at(args->mm, args->vaddr, args->ptep, pte);
0655     flush_dcache_page(page);
0656     barrier();
0657     ptep_clear(args->mm, args->vaddr, args->ptep);
0658     pte = ptep_get(args->ptep);
0659     WARN_ON(!pte_none(pte));
0660 }
0661 
0662 static void __init pmd_clear_tests(struct pgtable_debug_args *args)
0663 {
0664     pmd_t pmd = READ_ONCE(*args->pmdp);
0665 
0666     pr_debug("Validating PMD clear\n");
0667     pmd = __pmd(pmd_val(pmd) | RANDOM_ORVALUE);
0668     WRITE_ONCE(*args->pmdp, pmd);
0669     pmd_clear(args->pmdp);
0670     pmd = READ_ONCE(*args->pmdp);
0671     WARN_ON(!pmd_none(pmd));
0672 }
0673 
0674 static void __init pmd_populate_tests(struct pgtable_debug_args *args)
0675 {
0676     pmd_t pmd;
0677 
0678     pr_debug("Validating PMD populate\n");
0679     /*
0680      * This entry points to next level page table page.
0681      * Hence this must not qualify as pmd_bad().
0682      */
0683     pmd_populate(args->mm, args->pmdp, args->start_ptep);
0684     pmd = READ_ONCE(*args->pmdp);
0685     WARN_ON(pmd_bad(pmd));
0686 }
0687 
0688 static void __init pte_special_tests(struct pgtable_debug_args *args)
0689 {
0690     pte_t pte = pfn_pte(args->fixed_pte_pfn, args->page_prot);
0691 
0692     if (!IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL))
0693         return;
0694 
0695     pr_debug("Validating PTE special\n");
0696     WARN_ON(!pte_special(pte_mkspecial(pte)));
0697 }
0698 
0699 static void __init pte_protnone_tests(struct pgtable_debug_args *args)
0700 {
0701     pte_t pte = pfn_pte(args->fixed_pte_pfn, args->page_prot_none);
0702 
0703     if (!IS_ENABLED(CONFIG_NUMA_BALANCING))
0704         return;
0705 
0706     pr_debug("Validating PTE protnone\n");
0707     WARN_ON(!pte_protnone(pte));
0708     WARN_ON(!pte_present(pte));
0709 }
0710 
0711 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
0712 static void __init pmd_protnone_tests(struct pgtable_debug_args *args)
0713 {
0714     pmd_t pmd;
0715 
0716     if (!IS_ENABLED(CONFIG_NUMA_BALANCING))
0717         return;
0718 
0719     if (!has_transparent_hugepage())
0720         return;
0721 
0722     pr_debug("Validating PMD protnone\n");
0723     pmd = pmd_mkhuge(pfn_pmd(args->fixed_pmd_pfn, args->page_prot_none));
0724     WARN_ON(!pmd_protnone(pmd));
0725     WARN_ON(!pmd_present(pmd));
0726 }
0727 #else  /* !CONFIG_TRANSPARENT_HUGEPAGE */
0728 static void __init pmd_protnone_tests(struct pgtable_debug_args *args) { }
0729 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
0730 
0731 #ifdef CONFIG_ARCH_HAS_PTE_DEVMAP
0732 static void __init pte_devmap_tests(struct pgtable_debug_args *args)
0733 {
0734     pte_t pte = pfn_pte(args->fixed_pte_pfn, args->page_prot);
0735 
0736     pr_debug("Validating PTE devmap\n");
0737     WARN_ON(!pte_devmap(pte_mkdevmap(pte)));
0738 }
0739 
0740 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
0741 static void __init pmd_devmap_tests(struct pgtable_debug_args *args)
0742 {
0743     pmd_t pmd;
0744 
0745     if (!has_transparent_hugepage())
0746         return;
0747 
0748     pr_debug("Validating PMD devmap\n");
0749     pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot);
0750     WARN_ON(!pmd_devmap(pmd_mkdevmap(pmd)));
0751 }
0752 
0753 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
0754 static void __init pud_devmap_tests(struct pgtable_debug_args *args)
0755 {
0756     pud_t pud;
0757 
0758     if (!has_transparent_hugepage())
0759         return;
0760 
0761     pr_debug("Validating PUD devmap\n");
0762     pud = pfn_pud(args->fixed_pud_pfn, args->page_prot);
0763     WARN_ON(!pud_devmap(pud_mkdevmap(pud)));
0764 }
0765 #else  /* !CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
0766 static void __init pud_devmap_tests(struct pgtable_debug_args *args) { }
0767 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
0768 #else  /* CONFIG_TRANSPARENT_HUGEPAGE */
0769 static void __init pmd_devmap_tests(struct pgtable_debug_args *args) { }
0770 static void __init pud_devmap_tests(struct pgtable_debug_args *args) { }
0771 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
0772 #else
0773 static void __init pte_devmap_tests(struct pgtable_debug_args *args) { }
0774 static void __init pmd_devmap_tests(struct pgtable_debug_args *args) { }
0775 static void __init pud_devmap_tests(struct pgtable_debug_args *args) { }
0776 #endif /* CONFIG_ARCH_HAS_PTE_DEVMAP */
0777 
0778 static void __init pte_soft_dirty_tests(struct pgtable_debug_args *args)
0779 {
0780     pte_t pte = pfn_pte(args->fixed_pte_pfn, args->page_prot);
0781 
0782     if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY))
0783         return;
0784 
0785     pr_debug("Validating PTE soft dirty\n");
0786     WARN_ON(!pte_soft_dirty(pte_mksoft_dirty(pte)));
0787     WARN_ON(pte_soft_dirty(pte_clear_soft_dirty(pte)));
0788 }
0789 
0790 static void __init pte_swap_soft_dirty_tests(struct pgtable_debug_args *args)
0791 {
0792     pte_t pte = pfn_pte(args->fixed_pte_pfn, args->page_prot);
0793 
0794     if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY))
0795         return;
0796 
0797     pr_debug("Validating PTE swap soft dirty\n");
0798     WARN_ON(!pte_swp_soft_dirty(pte_swp_mksoft_dirty(pte)));
0799     WARN_ON(pte_swp_soft_dirty(pte_swp_clear_soft_dirty(pte)));
0800 }
0801 
0802 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
0803 static void __init pmd_soft_dirty_tests(struct pgtable_debug_args *args)
0804 {
0805     pmd_t pmd;
0806 
0807     if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY))
0808         return;
0809 
0810     if (!has_transparent_hugepage())
0811         return;
0812 
0813     pr_debug("Validating PMD soft dirty\n");
0814     pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot);
0815     WARN_ON(!pmd_soft_dirty(pmd_mksoft_dirty(pmd)));
0816     WARN_ON(pmd_soft_dirty(pmd_clear_soft_dirty(pmd)));
0817 }
0818 
0819 static void __init pmd_swap_soft_dirty_tests(struct pgtable_debug_args *args)
0820 {
0821     pmd_t pmd;
0822 
0823     if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) ||
0824         !IS_ENABLED(CONFIG_ARCH_ENABLE_THP_MIGRATION))
0825         return;
0826 
0827     if (!has_transparent_hugepage())
0828         return;
0829 
0830     pr_debug("Validating PMD swap soft dirty\n");
0831     pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot);
0832     WARN_ON(!pmd_swp_soft_dirty(pmd_swp_mksoft_dirty(pmd)));
0833     WARN_ON(pmd_swp_soft_dirty(pmd_swp_clear_soft_dirty(pmd)));
0834 }
0835 #else  /* !CONFIG_TRANSPARENT_HUGEPAGE */
0836 static void __init pmd_soft_dirty_tests(struct pgtable_debug_args *args) { }
0837 static void __init pmd_swap_soft_dirty_tests(struct pgtable_debug_args *args) { }
0838 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
0839 
0840 static void __init pte_swap_exclusive_tests(struct pgtable_debug_args *args)
0841 {
0842 #ifdef __HAVE_ARCH_PTE_SWP_EXCLUSIVE
0843     pte_t pte = pfn_pte(args->fixed_pte_pfn, args->page_prot);
0844 
0845     pr_debug("Validating PTE swap exclusive\n");
0846     pte = pte_swp_mkexclusive(pte);
0847     WARN_ON(!pte_swp_exclusive(pte));
0848     pte = pte_swp_clear_exclusive(pte);
0849     WARN_ON(pte_swp_exclusive(pte));
0850 #endif /* __HAVE_ARCH_PTE_SWP_EXCLUSIVE */
0851 }
0852 
0853 static void __init pte_swap_tests(struct pgtable_debug_args *args)
0854 {
0855     swp_entry_t swp;
0856     pte_t pte;
0857 
0858     pr_debug("Validating PTE swap\n");
0859     pte = pfn_pte(args->fixed_pte_pfn, args->page_prot);
0860     swp = __pte_to_swp_entry(pte);
0861     pte = __swp_entry_to_pte(swp);
0862     WARN_ON(args->fixed_pte_pfn != pte_pfn(pte));
0863 }
0864 
0865 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
0866 static void __init pmd_swap_tests(struct pgtable_debug_args *args)
0867 {
0868     swp_entry_t swp;
0869     pmd_t pmd;
0870 
0871     if (!has_transparent_hugepage())
0872         return;
0873 
0874     pr_debug("Validating PMD swap\n");
0875     pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot);
0876     swp = __pmd_to_swp_entry(pmd);
0877     pmd = __swp_entry_to_pmd(swp);
0878     WARN_ON(args->fixed_pmd_pfn != pmd_pfn(pmd));
0879 }
0880 #else  /* !CONFIG_ARCH_ENABLE_THP_MIGRATION */
0881 static void __init pmd_swap_tests(struct pgtable_debug_args *args) { }
0882 #endif /* CONFIG_ARCH_ENABLE_THP_MIGRATION */
0883 
0884 static void __init swap_migration_tests(struct pgtable_debug_args *args)
0885 {
0886     struct page *page;
0887     swp_entry_t swp;
0888 
0889     if (!IS_ENABLED(CONFIG_MIGRATION))
0890         return;
0891 
0892     /*
0893      * swap_migration_tests() requires a dedicated page as it needs to
0894      * be locked before creating a migration entry from it. Locking the
0895      * page that actually maps kernel text ('start_kernel') can be real
0896      * problematic. Lets use the allocated page explicitly for this
0897      * purpose.
0898      */
0899     page = (args->pte_pfn != ULONG_MAX) ? pfn_to_page(args->pte_pfn) : NULL;
0900     if (!page)
0901         return;
0902 
0903     pr_debug("Validating swap migration\n");
0904 
0905     /*
0906      * make_[readable|writable]_migration_entry() expects given page to
0907      * be locked, otherwise it stumbles upon a BUG_ON().
0908      */
0909     __SetPageLocked(page);
0910     swp = make_writable_migration_entry(page_to_pfn(page));
0911     WARN_ON(!is_migration_entry(swp));
0912     WARN_ON(!is_writable_migration_entry(swp));
0913 
0914     swp = make_readable_migration_entry(swp_offset(swp));
0915     WARN_ON(!is_migration_entry(swp));
0916     WARN_ON(is_writable_migration_entry(swp));
0917 
0918     swp = make_readable_migration_entry(page_to_pfn(page));
0919     WARN_ON(!is_migration_entry(swp));
0920     WARN_ON(is_writable_migration_entry(swp));
0921     __ClearPageLocked(page);
0922 }
0923 
0924 #ifdef CONFIG_HUGETLB_PAGE
0925 static void __init hugetlb_basic_tests(struct pgtable_debug_args *args)
0926 {
0927     struct page *page;
0928     pte_t pte;
0929 
0930     pr_debug("Validating HugeTLB basic\n");
0931     /*
0932      * Accessing the page associated with the pfn is safe here,
0933      * as it was previously derived from a real kernel symbol.
0934      */
0935     page = pfn_to_page(args->fixed_pmd_pfn);
0936     pte = mk_huge_pte(page, args->page_prot);
0937 
0938     WARN_ON(!huge_pte_dirty(huge_pte_mkdirty(pte)));
0939     WARN_ON(!huge_pte_write(huge_pte_mkwrite(huge_pte_wrprotect(pte))));
0940     WARN_ON(huge_pte_write(huge_pte_wrprotect(huge_pte_mkwrite(pte))));
0941 
0942 #ifdef CONFIG_ARCH_WANT_GENERAL_HUGETLB
0943     pte = pfn_pte(args->fixed_pmd_pfn, args->page_prot);
0944 
0945     WARN_ON(!pte_huge(pte_mkhuge(pte)));
0946 #endif /* CONFIG_ARCH_WANT_GENERAL_HUGETLB */
0947 }
0948 #else  /* !CONFIG_HUGETLB_PAGE */
0949 static void __init hugetlb_basic_tests(struct pgtable_debug_args *args) { }
0950 #endif /* CONFIG_HUGETLB_PAGE */
0951 
0952 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
0953 static void __init pmd_thp_tests(struct pgtable_debug_args *args)
0954 {
0955     pmd_t pmd;
0956 
0957     if (!has_transparent_hugepage())
0958         return;
0959 
0960     pr_debug("Validating PMD based THP\n");
0961     /*
0962      * pmd_trans_huge() and pmd_present() must return positive after
0963      * MMU invalidation with pmd_mkinvalid(). This behavior is an
0964      * optimization for transparent huge page. pmd_trans_huge() must
0965      * be true if pmd_page() returns a valid THP to avoid taking the
0966      * pmd_lock when others walk over non transhuge pmds (i.e. there
0967      * are no THP allocated). Especially when splitting a THP and
0968      * removing the present bit from the pmd, pmd_trans_huge() still
0969      * needs to return true. pmd_present() should be true whenever
0970      * pmd_trans_huge() returns true.
0971      */
0972     pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot);
0973     WARN_ON(!pmd_trans_huge(pmd_mkhuge(pmd)));
0974 
0975 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
0976     WARN_ON(!pmd_trans_huge(pmd_mkinvalid(pmd_mkhuge(pmd))));
0977     WARN_ON(!pmd_present(pmd_mkinvalid(pmd_mkhuge(pmd))));
0978 #endif /* __HAVE_ARCH_PMDP_INVALIDATE */
0979 }
0980 
0981 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
0982 static void __init pud_thp_tests(struct pgtable_debug_args *args)
0983 {
0984     pud_t pud;
0985 
0986     if (!has_transparent_hugepage())
0987         return;
0988 
0989     pr_debug("Validating PUD based THP\n");
0990     pud = pfn_pud(args->fixed_pud_pfn, args->page_prot);
0991     WARN_ON(!pud_trans_huge(pud_mkhuge(pud)));
0992 
0993     /*
0994      * pud_mkinvalid() has been dropped for now. Enable back
0995      * these tests when it comes back with a modified pud_present().
0996      *
0997      * WARN_ON(!pud_trans_huge(pud_mkinvalid(pud_mkhuge(pud))));
0998      * WARN_ON(!pud_present(pud_mkinvalid(pud_mkhuge(pud))));
0999      */
1000 }
1001 #else  /* !CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
1002 static void __init pud_thp_tests(struct pgtable_debug_args *args) { }
1003 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
1004 #else  /* !CONFIG_TRANSPARENT_HUGEPAGE */
1005 static void __init pmd_thp_tests(struct pgtable_debug_args *args) { }
1006 static void __init pud_thp_tests(struct pgtable_debug_args *args) { }
1007 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1008 
1009 static unsigned long __init get_random_vaddr(void)
1010 {
1011     unsigned long random_vaddr, random_pages, total_user_pages;
1012 
1013     total_user_pages = (TASK_SIZE - FIRST_USER_ADDRESS) / PAGE_SIZE;
1014 
1015     random_pages = get_random_long() % total_user_pages;
1016     random_vaddr = FIRST_USER_ADDRESS + random_pages * PAGE_SIZE;
1017 
1018     return random_vaddr;
1019 }
1020 
1021 static void __init destroy_args(struct pgtable_debug_args *args)
1022 {
1023     struct page *page = NULL;
1024 
1025     /* Free (huge) page */
1026     if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
1027         IS_ENABLED(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD) &&
1028         has_transparent_hugepage() &&
1029         args->pud_pfn != ULONG_MAX) {
1030         if (args->is_contiguous_page) {
1031             free_contig_range(args->pud_pfn,
1032                       (1 << (HPAGE_PUD_SHIFT - PAGE_SHIFT)));
1033         } else {
1034             page = pfn_to_page(args->pud_pfn);
1035             __free_pages(page, HPAGE_PUD_SHIFT - PAGE_SHIFT);
1036         }
1037 
1038         args->pud_pfn = ULONG_MAX;
1039         args->pmd_pfn = ULONG_MAX;
1040         args->pte_pfn = ULONG_MAX;
1041     }
1042 
1043     if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
1044         has_transparent_hugepage() &&
1045         args->pmd_pfn != ULONG_MAX) {
1046         if (args->is_contiguous_page) {
1047             free_contig_range(args->pmd_pfn, (1 << HPAGE_PMD_ORDER));
1048         } else {
1049             page = pfn_to_page(args->pmd_pfn);
1050             __free_pages(page, HPAGE_PMD_ORDER);
1051         }
1052 
1053         args->pmd_pfn = ULONG_MAX;
1054         args->pte_pfn = ULONG_MAX;
1055     }
1056 
1057     if (args->pte_pfn != ULONG_MAX) {
1058         page = pfn_to_page(args->pte_pfn);
1059         __free_pages(page, 0);
1060 
1061         args->pte_pfn = ULONG_MAX;
1062     }
1063 
1064     /* Free page table entries */
1065     if (args->start_ptep) {
1066         pte_free(args->mm, args->start_ptep);
1067         mm_dec_nr_ptes(args->mm);
1068     }
1069 
1070     if (args->start_pmdp) {
1071         pmd_free(args->mm, args->start_pmdp);
1072         mm_dec_nr_pmds(args->mm);
1073     }
1074 
1075     if (args->start_pudp) {
1076         pud_free(args->mm, args->start_pudp);
1077         mm_dec_nr_puds(args->mm);
1078     }
1079 
1080     if (args->start_p4dp)
1081         p4d_free(args->mm, args->start_p4dp);
1082 
1083     /* Free vma and mm struct */
1084     if (args->vma)
1085         vm_area_free(args->vma);
1086 
1087     if (args->mm)
1088         mmdrop(args->mm);
1089 }
1090 
1091 static struct page * __init
1092 debug_vm_pgtable_alloc_huge_page(struct pgtable_debug_args *args, int order)
1093 {
1094     struct page *page = NULL;
1095 
1096 #ifdef CONFIG_CONTIG_ALLOC
1097     if (order >= MAX_ORDER) {
1098         page = alloc_contig_pages((1 << order), GFP_KERNEL,
1099                       first_online_node, NULL);
1100         if (page) {
1101             args->is_contiguous_page = true;
1102             return page;
1103         }
1104     }
1105 #endif
1106 
1107     if (order < MAX_ORDER)
1108         page = alloc_pages(GFP_KERNEL, order);
1109 
1110     return page;
1111 }
1112 
1113 static int __init init_args(struct pgtable_debug_args *args)
1114 {
1115     struct page *page = NULL;
1116     phys_addr_t phys;
1117     int ret = 0;
1118 
1119     /*
1120      * Initialize the debugging data.
1121      *
1122      * vm_get_page_prot(VM_NONE) or vm_get_page_prot(VM_SHARED|VM_NONE)
1123      * will help create page table entries with PROT_NONE permission as
1124      * required for pxx_protnone_tests().
1125      */
1126     memset(args, 0, sizeof(*args));
1127     args->vaddr              = get_random_vaddr();
1128     args->page_prot          = vm_get_page_prot(VMFLAGS);
1129     args->page_prot_none     = vm_get_page_prot(VM_NONE);
1130     args->is_contiguous_page = false;
1131     args->pud_pfn            = ULONG_MAX;
1132     args->pmd_pfn            = ULONG_MAX;
1133     args->pte_pfn            = ULONG_MAX;
1134     args->fixed_pgd_pfn      = ULONG_MAX;
1135     args->fixed_p4d_pfn      = ULONG_MAX;
1136     args->fixed_pud_pfn      = ULONG_MAX;
1137     args->fixed_pmd_pfn      = ULONG_MAX;
1138     args->fixed_pte_pfn      = ULONG_MAX;
1139 
1140     /* Allocate mm and vma */
1141     args->mm = mm_alloc();
1142     if (!args->mm) {
1143         pr_err("Failed to allocate mm struct\n");
1144         ret = -ENOMEM;
1145         goto error;
1146     }
1147 
1148     args->vma = vm_area_alloc(args->mm);
1149     if (!args->vma) {
1150         pr_err("Failed to allocate vma\n");
1151         ret = -ENOMEM;
1152         goto error;
1153     }
1154 
1155     /*
1156      * Allocate page table entries. They will be modified in the tests.
1157      * Lets save the page table entries so that they can be released
1158      * when the tests are completed.
1159      */
1160     args->pgdp = pgd_offset(args->mm, args->vaddr);
1161     args->p4dp = p4d_alloc(args->mm, args->pgdp, args->vaddr);
1162     if (!args->p4dp) {
1163         pr_err("Failed to allocate p4d entries\n");
1164         ret = -ENOMEM;
1165         goto error;
1166     }
1167     args->start_p4dp = p4d_offset(args->pgdp, 0UL);
1168     WARN_ON(!args->start_p4dp);
1169 
1170     args->pudp = pud_alloc(args->mm, args->p4dp, args->vaddr);
1171     if (!args->pudp) {
1172         pr_err("Failed to allocate pud entries\n");
1173         ret = -ENOMEM;
1174         goto error;
1175     }
1176     args->start_pudp = pud_offset(args->p4dp, 0UL);
1177     WARN_ON(!args->start_pudp);
1178 
1179     args->pmdp = pmd_alloc(args->mm, args->pudp, args->vaddr);
1180     if (!args->pmdp) {
1181         pr_err("Failed to allocate pmd entries\n");
1182         ret = -ENOMEM;
1183         goto error;
1184     }
1185     args->start_pmdp = pmd_offset(args->pudp, 0UL);
1186     WARN_ON(!args->start_pmdp);
1187 
1188     if (pte_alloc(args->mm, args->pmdp)) {
1189         pr_err("Failed to allocate pte entries\n");
1190         ret = -ENOMEM;
1191         goto error;
1192     }
1193     args->start_ptep = pmd_pgtable(READ_ONCE(*args->pmdp));
1194     WARN_ON(!args->start_ptep);
1195 
1196     /*
1197      * PFN for mapping at PTE level is determined from a standard kernel
1198      * text symbol. But pfns for higher page table levels are derived by
1199      * masking lower bits of this real pfn. These derived pfns might not
1200      * exist on the platform but that does not really matter as pfn_pxx()
1201      * helpers will still create appropriate entries for the test. This
1202      * helps avoid large memory block allocations to be used for mapping
1203      * at higher page table levels in some of the tests.
1204      */
1205     phys = __pa_symbol(&start_kernel);
1206     args->fixed_pgd_pfn = __phys_to_pfn(phys & PGDIR_MASK);
1207     args->fixed_p4d_pfn = __phys_to_pfn(phys & P4D_MASK);
1208     args->fixed_pud_pfn = __phys_to_pfn(phys & PUD_MASK);
1209     args->fixed_pmd_pfn = __phys_to_pfn(phys & PMD_MASK);
1210     args->fixed_pte_pfn = __phys_to_pfn(phys & PAGE_MASK);
1211     WARN_ON(!pfn_valid(args->fixed_pte_pfn));
1212 
1213     /*
1214      * Allocate (huge) pages because some of the tests need to access
1215      * the data in the pages. The corresponding tests will be skipped
1216      * if we fail to allocate (huge) pages.
1217      */
1218     if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
1219         IS_ENABLED(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD) &&
1220         has_transparent_hugepage()) {
1221         page = debug_vm_pgtable_alloc_huge_page(args,
1222                 HPAGE_PUD_SHIFT - PAGE_SHIFT);
1223         if (page) {
1224             args->pud_pfn = page_to_pfn(page);
1225             args->pmd_pfn = args->pud_pfn;
1226             args->pte_pfn = args->pud_pfn;
1227             return 0;
1228         }
1229     }
1230 
1231     if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
1232         has_transparent_hugepage()) {
1233         page = debug_vm_pgtable_alloc_huge_page(args, HPAGE_PMD_ORDER);
1234         if (page) {
1235             args->pmd_pfn = page_to_pfn(page);
1236             args->pte_pfn = args->pmd_pfn;
1237             return 0;
1238         }
1239     }
1240 
1241     page = alloc_pages(GFP_KERNEL, 0);
1242     if (page)
1243         args->pte_pfn = page_to_pfn(page);
1244 
1245     return 0;
1246 
1247 error:
1248     destroy_args(args);
1249     return ret;
1250 }
1251 
1252 static int __init debug_vm_pgtable(void)
1253 {
1254     struct pgtable_debug_args args;
1255     spinlock_t *ptl = NULL;
1256     int idx, ret;
1257 
1258     pr_info("Validating architecture page table helpers\n");
1259     ret = init_args(&args);
1260     if (ret)
1261         return ret;
1262 
1263     /*
1264      * Iterate over each possible vm_flags to make sure that all
1265      * the basic page table transformation validations just hold
1266      * true irrespective of the starting protection value for a
1267      * given page table entry.
1268      *
1269      * Protection based vm_flags combinatins are always linear
1270      * and increasing i.e starting from VM_NONE and going upto
1271      * (VM_SHARED | READ | WRITE | EXEC).
1272      */
1273 #define VM_FLAGS_START  (VM_NONE)
1274 #define VM_FLAGS_END    (VM_SHARED | VM_EXEC | VM_WRITE | VM_READ)
1275 
1276     for (idx = VM_FLAGS_START; idx <= VM_FLAGS_END; idx++) {
1277         pte_basic_tests(&args, idx);
1278         pmd_basic_tests(&args, idx);
1279         pud_basic_tests(&args, idx);
1280     }
1281 
1282     /*
1283      * Both P4D and PGD level tests are very basic which do not
1284      * involve creating page table entries from the protection
1285      * value and the given pfn. Hence just keep them out from
1286      * the above iteration for now to save some test execution
1287      * time.
1288      */
1289     p4d_basic_tests(&args);
1290     pgd_basic_tests(&args);
1291 
1292     pmd_leaf_tests(&args);
1293     pud_leaf_tests(&args);
1294 
1295     pte_savedwrite_tests(&args);
1296     pmd_savedwrite_tests(&args);
1297 
1298     pte_special_tests(&args);
1299     pte_protnone_tests(&args);
1300     pmd_protnone_tests(&args);
1301 
1302     pte_devmap_tests(&args);
1303     pmd_devmap_tests(&args);
1304     pud_devmap_tests(&args);
1305 
1306     pte_soft_dirty_tests(&args);
1307     pmd_soft_dirty_tests(&args);
1308     pte_swap_soft_dirty_tests(&args);
1309     pmd_swap_soft_dirty_tests(&args);
1310 
1311     pte_swap_exclusive_tests(&args);
1312 
1313     pte_swap_tests(&args);
1314     pmd_swap_tests(&args);
1315 
1316     swap_migration_tests(&args);
1317 
1318     pmd_thp_tests(&args);
1319     pud_thp_tests(&args);
1320 
1321     hugetlb_basic_tests(&args);
1322 
1323     /*
1324      * Page table modifying tests. They need to hold
1325      * proper page table lock.
1326      */
1327 
1328     args.ptep = pte_offset_map_lock(args.mm, args.pmdp, args.vaddr, &ptl);
1329     pte_clear_tests(&args);
1330     pte_advanced_tests(&args);
1331     pte_unmap_unlock(args.ptep, ptl);
1332 
1333     ptl = pmd_lock(args.mm, args.pmdp);
1334     pmd_clear_tests(&args);
1335     pmd_advanced_tests(&args);
1336     pmd_huge_tests(&args);
1337     pmd_populate_tests(&args);
1338     spin_unlock(ptl);
1339 
1340     ptl = pud_lock(args.mm, args.pudp);
1341     pud_clear_tests(&args);
1342     pud_advanced_tests(&args);
1343     pud_huge_tests(&args);
1344     pud_populate_tests(&args);
1345     spin_unlock(ptl);
1346 
1347     spin_lock(&(args.mm->page_table_lock));
1348     p4d_clear_tests(&args);
1349     pgd_clear_tests(&args);
1350     p4d_populate_tests(&args);
1351     pgd_populate_tests(&args);
1352     spin_unlock(&(args.mm->page_table_lock));
1353 
1354     destroy_args(&args);
1355     return 0;
1356 }
1357 late_initcall(debug_vm_pgtable);
[ Source navigation ] [ Diff markup ] [ Identifier search ] [ general search ]

This page was automatically generated by the 2.1.0 LXR engine.

The LXR team
Valid CSS 2.1! Valid HTML 4.01!