1 // SPDX-License-Identifier: GPL-2.0-only
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.
7 * Copyright (C) 2019 ARM Ltd.
9 * Author: Anshuman Khandual <anshuman.khandual@arm.com>
11 #define pr_fmt(fmt) "debug_vm_pgtable: [%-25s]: " fmt, __func__
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>
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>
32 #include <asm/pgalloc.h>
33 #include <asm/tlbflush.h>
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.
41 #define VMFLAGS (VM_READ|VM_WRITE|VM_EXEC)
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.
51 #define S390_SKIP_MASK GENMASK(3, 0)
52 #if __BITS_PER_LONG == 64
53 #define PPC64_SKIP_MASK GENMASK(62, 62)
55 #define PPC64_SKIP_MASK 0x0
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)
61 static void __init pte_basic_tests(unsigned long pfn, int idx)
63 pgprot_t prot = protection_map[idx];
64 pte_t pte = pfn_pte(pfn, prot);
65 unsigned long val = idx, *ptr = &val;
67 pr_debug("Validating PTE basic (%pGv)\n", ptr);
70 * This test needs to be executed after the given page table entry
71 * is created with pfn_pte() to make sure that protection_map[idx]
72 * does not have the dirty bit enabled from the beginning. This is
73 * important for platforms like arm64 where (!PTE_RDONLY) indicate
74 * dirty bit being set.
76 WARN_ON(pte_dirty(pte_wrprotect(pte)));
78 WARN_ON(!pte_same(pte, pte));
79 WARN_ON(!pte_young(pte_mkyoung(pte_mkold(pte))));
80 WARN_ON(!pte_dirty(pte_mkdirty(pte_mkclean(pte))));
81 WARN_ON(!pte_write(pte_mkwrite(pte_wrprotect(pte))));
82 WARN_ON(pte_young(pte_mkold(pte_mkyoung(pte))));
83 WARN_ON(pte_dirty(pte_mkclean(pte_mkdirty(pte))));
84 WARN_ON(pte_write(pte_wrprotect(pte_mkwrite(pte))));
85 WARN_ON(pte_dirty(pte_wrprotect(pte_mkclean(pte))));
86 WARN_ON(!pte_dirty(pte_wrprotect(pte_mkdirty(pte))));
89 static void __init pte_advanced_tests(struct mm_struct *mm,
90 struct vm_area_struct *vma, pte_t *ptep,
91 unsigned long pfn, unsigned long vaddr,
94 pte_t pte = pfn_pte(pfn, prot);
97 * Architectures optimize set_pte_at by avoiding TLB flush.
98 * This requires set_pte_at to be not used to update an
99 * existing pte entry. Clear pte before we do set_pte_at
102 pr_debug("Validating PTE advanced\n");
103 pte = pfn_pte(pfn, prot);
104 set_pte_at(mm, vaddr, ptep, pte);
105 ptep_set_wrprotect(mm, vaddr, ptep);
106 pte = ptep_get(ptep);
107 WARN_ON(pte_write(pte));
108 ptep_get_and_clear(mm, vaddr, ptep);
109 pte = ptep_get(ptep);
110 WARN_ON(!pte_none(pte));
112 pte = pfn_pte(pfn, prot);
113 pte = pte_wrprotect(pte);
114 pte = pte_mkclean(pte);
115 set_pte_at(mm, vaddr, ptep, pte);
116 pte = pte_mkwrite(pte);
117 pte = pte_mkdirty(pte);
118 ptep_set_access_flags(vma, vaddr, ptep, pte, 1);
119 pte = ptep_get(ptep);
120 WARN_ON(!(pte_write(pte) && pte_dirty(pte)));
121 ptep_get_and_clear_full(mm, vaddr, ptep, 1);
122 pte = ptep_get(ptep);
123 WARN_ON(!pte_none(pte));
125 pte = pfn_pte(pfn, prot);
126 pte = pte_mkyoung(pte);
127 set_pte_at(mm, vaddr, ptep, pte);
128 ptep_test_and_clear_young(vma, vaddr, ptep);
129 pte = ptep_get(ptep);
130 WARN_ON(pte_young(pte));
133 static void __init pte_savedwrite_tests(unsigned long pfn, pgprot_t prot)
135 pte_t pte = pfn_pte(pfn, prot);
137 if (!IS_ENABLED(CONFIG_NUMA_BALANCING))
140 pr_debug("Validating PTE saved write\n");
141 WARN_ON(!pte_savedwrite(pte_mk_savedwrite(pte_clear_savedwrite(pte))));
142 WARN_ON(pte_savedwrite(pte_clear_savedwrite(pte_mk_savedwrite(pte))));
145 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
146 static void __init pmd_basic_tests(unsigned long pfn, int idx)
148 pgprot_t prot = protection_map[idx];
149 unsigned long val = idx, *ptr = &val;
152 if (!has_transparent_hugepage())
155 pr_debug("Validating PMD basic (%pGv)\n", ptr);
156 pmd = pfn_pmd(pfn, prot);
159 * This test needs to be executed after the given page table entry
160 * is created with pfn_pmd() to make sure that protection_map[idx]
161 * does not have the dirty bit enabled from the beginning. This is
162 * important for platforms like arm64 where (!PTE_RDONLY) indicate
163 * dirty bit being set.
165 WARN_ON(pmd_dirty(pmd_wrprotect(pmd)));
168 WARN_ON(!pmd_same(pmd, pmd));
169 WARN_ON(!pmd_young(pmd_mkyoung(pmd_mkold(pmd))));
170 WARN_ON(!pmd_dirty(pmd_mkdirty(pmd_mkclean(pmd))));
171 WARN_ON(!pmd_write(pmd_mkwrite(pmd_wrprotect(pmd))));
172 WARN_ON(pmd_young(pmd_mkold(pmd_mkyoung(pmd))));
173 WARN_ON(pmd_dirty(pmd_mkclean(pmd_mkdirty(pmd))));
174 WARN_ON(pmd_write(pmd_wrprotect(pmd_mkwrite(pmd))));
175 WARN_ON(pmd_dirty(pmd_wrprotect(pmd_mkclean(pmd))));
176 WARN_ON(!pmd_dirty(pmd_wrprotect(pmd_mkdirty(pmd))));
178 * A huge page does not point to next level page table
179 * entry. Hence this must qualify as pmd_bad().
181 WARN_ON(!pmd_bad(pmd_mkhuge(pmd)));
184 static void __init pmd_advanced_tests(struct mm_struct *mm,
185 struct vm_area_struct *vma, pmd_t *pmdp,
186 unsigned long pfn, unsigned long vaddr,
187 pgprot_t prot, pgtable_t pgtable)
191 if (!has_transparent_hugepage())
194 pr_debug("Validating PMD advanced\n");
195 /* Align the address wrt HPAGE_PMD_SIZE */
196 vaddr &= HPAGE_PMD_MASK;
198 pgtable_trans_huge_deposit(mm, pmdp, pgtable);
200 pmd = pfn_pmd(pfn, prot);
201 set_pmd_at(mm, vaddr, pmdp, pmd);
202 pmdp_set_wrprotect(mm, vaddr, pmdp);
203 pmd = READ_ONCE(*pmdp);
204 WARN_ON(pmd_write(pmd));
205 pmdp_huge_get_and_clear(mm, vaddr, pmdp);
206 pmd = READ_ONCE(*pmdp);
207 WARN_ON(!pmd_none(pmd));
209 pmd = pfn_pmd(pfn, prot);
210 pmd = pmd_wrprotect(pmd);
211 pmd = pmd_mkclean(pmd);
212 set_pmd_at(mm, vaddr, pmdp, pmd);
213 pmd = pmd_mkwrite(pmd);
214 pmd = pmd_mkdirty(pmd);
215 pmdp_set_access_flags(vma, vaddr, pmdp, pmd, 1);
216 pmd = READ_ONCE(*pmdp);
217 WARN_ON(!(pmd_write(pmd) && pmd_dirty(pmd)));
218 pmdp_huge_get_and_clear_full(vma, vaddr, pmdp, 1);
219 pmd = READ_ONCE(*pmdp);
220 WARN_ON(!pmd_none(pmd));
222 pmd = pmd_mkhuge(pfn_pmd(pfn, prot));
223 pmd = pmd_mkyoung(pmd);
224 set_pmd_at(mm, vaddr, pmdp, pmd);
225 pmdp_test_and_clear_young(vma, vaddr, pmdp);
226 pmd = READ_ONCE(*pmdp);
227 WARN_ON(pmd_young(pmd));
229 /* Clear the pte entries */
230 pmdp_huge_get_and_clear(mm, vaddr, pmdp);
231 pgtable = pgtable_trans_huge_withdraw(mm, pmdp);
234 static void __init pmd_leaf_tests(unsigned long pfn, pgprot_t prot)
238 if (!has_transparent_hugepage())
241 pr_debug("Validating PMD leaf\n");
242 pmd = pfn_pmd(pfn, prot);
245 * PMD based THP is a leaf entry.
247 pmd = pmd_mkhuge(pmd);
248 WARN_ON(!pmd_leaf(pmd));
251 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
252 static void __init pmd_huge_tests(pmd_t *pmdp, unsigned long pfn, pgprot_t prot)
256 if (!arch_ioremap_pmd_supported())
259 pr_debug("Validating PMD huge\n");
261 * X86 defined pmd_set_huge() verifies that the given
262 * PMD is not a populated non-leaf entry.
264 WRITE_ONCE(*pmdp, __pmd(0));
265 WARN_ON(!pmd_set_huge(pmdp, __pfn_to_phys(pfn), prot));
266 WARN_ON(!pmd_clear_huge(pmdp));
267 pmd = READ_ONCE(*pmdp);
268 WARN_ON(!pmd_none(pmd));
270 #else /* CONFIG_HAVE_ARCH_HUGE_VMAP */
271 static void __init pmd_huge_tests(pmd_t *pmdp, unsigned long pfn, pgprot_t prot) { }
272 #endif /* CONFIG_HAVE_ARCH_HUGE_VMAP */
274 static void __init pmd_savedwrite_tests(unsigned long pfn, pgprot_t prot)
278 if (!IS_ENABLED(CONFIG_NUMA_BALANCING))
281 if (!has_transparent_hugepage())
284 pr_debug("Validating PMD saved write\n");
285 pmd = pfn_pmd(pfn, prot);
286 WARN_ON(!pmd_savedwrite(pmd_mk_savedwrite(pmd_clear_savedwrite(pmd))));
287 WARN_ON(pmd_savedwrite(pmd_clear_savedwrite(pmd_mk_savedwrite(pmd))));
290 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
291 static void __init pud_basic_tests(struct mm_struct *mm, unsigned long pfn, int idx)
293 pgprot_t prot = protection_map[idx];
294 unsigned long val = idx, *ptr = &val;
297 if (!has_transparent_hugepage())
300 pr_debug("Validating PUD basic (%pGv)\n", ptr);
301 pud = pfn_pud(pfn, prot);
304 * This test needs to be executed after the given page table entry
305 * is created with pfn_pud() to make sure that protection_map[idx]
306 * does not have the dirty bit enabled from the beginning. This is
307 * important for platforms like arm64 where (!PTE_RDONLY) indicate
308 * dirty bit being set.
310 WARN_ON(pud_dirty(pud_wrprotect(pud)));
312 WARN_ON(!pud_same(pud, pud));
313 WARN_ON(!pud_young(pud_mkyoung(pud_mkold(pud))));
314 WARN_ON(!pud_dirty(pud_mkdirty(pud_mkclean(pud))));
315 WARN_ON(pud_dirty(pud_mkclean(pud_mkdirty(pud))));
316 WARN_ON(!pud_write(pud_mkwrite(pud_wrprotect(pud))));
317 WARN_ON(pud_write(pud_wrprotect(pud_mkwrite(pud))));
318 WARN_ON(pud_young(pud_mkold(pud_mkyoung(pud))));
319 WARN_ON(pud_dirty(pud_wrprotect(pud_mkclean(pud))));
320 WARN_ON(!pud_dirty(pud_wrprotect(pud_mkdirty(pud))));
322 if (mm_pmd_folded(mm))
326 * A huge page does not point to next level page table
327 * entry. Hence this must qualify as pud_bad().
329 WARN_ON(!pud_bad(pud_mkhuge(pud)));
332 static void __init pud_advanced_tests(struct mm_struct *mm,
333 struct vm_area_struct *vma, pud_t *pudp,
334 unsigned long pfn, unsigned long vaddr,
339 if (!has_transparent_hugepage())
342 pr_debug("Validating PUD advanced\n");
343 /* Align the address wrt HPAGE_PUD_SIZE */
344 vaddr &= HPAGE_PUD_MASK;
346 pud = pfn_pud(pfn, prot);
347 set_pud_at(mm, vaddr, pudp, pud);
348 pudp_set_wrprotect(mm, vaddr, pudp);
349 pud = READ_ONCE(*pudp);
350 WARN_ON(pud_write(pud));
352 #ifndef __PAGETABLE_PMD_FOLDED
353 pudp_huge_get_and_clear(mm, vaddr, pudp);
354 pud = READ_ONCE(*pudp);
355 WARN_ON(!pud_none(pud));
356 #endif /* __PAGETABLE_PMD_FOLDED */
357 pud = pfn_pud(pfn, prot);
358 pud = pud_wrprotect(pud);
359 pud = pud_mkclean(pud);
360 set_pud_at(mm, vaddr, pudp, pud);
361 pud = pud_mkwrite(pud);
362 pud = pud_mkdirty(pud);
363 pudp_set_access_flags(vma, vaddr, pudp, pud, 1);
364 pud = READ_ONCE(*pudp);
365 WARN_ON(!(pud_write(pud) && pud_dirty(pud)));
367 #ifndef __PAGETABLE_PMD_FOLDED
368 pudp_huge_get_and_clear_full(mm, vaddr, pudp, 1);
369 pud = READ_ONCE(*pudp);
370 WARN_ON(!pud_none(pud));
371 #endif /* __PAGETABLE_PMD_FOLDED */
373 pud = pfn_pud(pfn, prot);
374 pud = pud_mkyoung(pud);
375 set_pud_at(mm, vaddr, pudp, pud);
376 pudp_test_and_clear_young(vma, vaddr, pudp);
377 pud = READ_ONCE(*pudp);
378 WARN_ON(pud_young(pud));
380 pudp_huge_get_and_clear(mm, vaddr, pudp);
383 static void __init pud_leaf_tests(unsigned long pfn, pgprot_t prot)
387 if (!has_transparent_hugepage())
390 pr_debug("Validating PUD leaf\n");
391 pud = pfn_pud(pfn, prot);
393 * PUD based THP is a leaf entry.
395 pud = pud_mkhuge(pud);
396 WARN_ON(!pud_leaf(pud));
399 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
400 static void __init pud_huge_tests(pud_t *pudp, unsigned long pfn, pgprot_t prot)
404 if (!arch_ioremap_pud_supported())
407 pr_debug("Validating PUD huge\n");
409 * X86 defined pud_set_huge() verifies that the given
410 * PUD is not a populated non-leaf entry.
412 WRITE_ONCE(*pudp, __pud(0));
413 WARN_ON(!pud_set_huge(pudp, __pfn_to_phys(pfn), prot));
414 WARN_ON(!pud_clear_huge(pudp));
415 pud = READ_ONCE(*pudp);
416 WARN_ON(!pud_none(pud));
418 #else /* !CONFIG_HAVE_ARCH_HUGE_VMAP */
419 static void __init pud_huge_tests(pud_t *pudp, unsigned long pfn, pgprot_t prot) { }
420 #endif /* !CONFIG_HAVE_ARCH_HUGE_VMAP */
422 #else /* !CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
423 static void __init pud_basic_tests(struct mm_struct *mm, unsigned long pfn, int idx) { }
424 static void __init pud_advanced_tests(struct mm_struct *mm,
425 struct vm_area_struct *vma, pud_t *pudp,
426 unsigned long pfn, unsigned long vaddr,
430 static void __init pud_leaf_tests(unsigned long pfn, pgprot_t prot) { }
431 static void __init pud_huge_tests(pud_t *pudp, unsigned long pfn, pgprot_t prot)
434 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
435 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
436 static void __init pmd_basic_tests(unsigned long pfn, int idx) { }
437 static void __init pud_basic_tests(struct mm_struct *mm, unsigned long pfn, int idx) { }
438 static void __init pmd_advanced_tests(struct mm_struct *mm,
439 struct vm_area_struct *vma, pmd_t *pmdp,
440 unsigned long pfn, unsigned long vaddr,
441 pgprot_t prot, pgtable_t pgtable)
444 static void __init pud_advanced_tests(struct mm_struct *mm,
445 struct vm_area_struct *vma, pud_t *pudp,
446 unsigned long pfn, unsigned long vaddr,
450 static void __init pmd_leaf_tests(unsigned long pfn, pgprot_t prot) { }
451 static void __init pud_leaf_tests(unsigned long pfn, pgprot_t prot) { }
452 static void __init pmd_huge_tests(pmd_t *pmdp, unsigned long pfn, pgprot_t prot)
455 static void __init pud_huge_tests(pud_t *pudp, unsigned long pfn, pgprot_t prot)
458 static void __init pmd_savedwrite_tests(unsigned long pfn, pgprot_t prot) { }
459 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
461 static void __init p4d_basic_tests(unsigned long pfn, pgprot_t prot)
465 pr_debug("Validating P4D basic\n");
466 memset(&p4d, RANDOM_NZVALUE, sizeof(p4d_t));
467 WARN_ON(!p4d_same(p4d, p4d));
470 static void __init pgd_basic_tests(unsigned long pfn, pgprot_t prot)
474 pr_debug("Validating PGD basic\n");
475 memset(&pgd, RANDOM_NZVALUE, sizeof(pgd_t));
476 WARN_ON(!pgd_same(pgd, pgd));
479 #ifndef __PAGETABLE_PUD_FOLDED
480 static void __init pud_clear_tests(struct mm_struct *mm, pud_t *pudp)
482 pud_t pud = READ_ONCE(*pudp);
484 if (mm_pmd_folded(mm))
487 pr_debug("Validating PUD clear\n");
488 pud = __pud(pud_val(pud) | RANDOM_ORVALUE);
489 WRITE_ONCE(*pudp, pud);
491 pud = READ_ONCE(*pudp);
492 WARN_ON(!pud_none(pud));
495 static void __init pud_populate_tests(struct mm_struct *mm, pud_t *pudp,
500 if (mm_pmd_folded(mm))
503 pr_debug("Validating PUD populate\n");
505 * This entry points to next level page table page.
506 * Hence this must not qualify as pud_bad().
508 pud_populate(mm, pudp, pmdp);
509 pud = READ_ONCE(*pudp);
510 WARN_ON(pud_bad(pud));
512 #else /* !__PAGETABLE_PUD_FOLDED */
513 static void __init pud_clear_tests(struct mm_struct *mm, pud_t *pudp) { }
514 static void __init pud_populate_tests(struct mm_struct *mm, pud_t *pudp,
518 #endif /* PAGETABLE_PUD_FOLDED */
520 #ifndef __PAGETABLE_P4D_FOLDED
521 static void __init p4d_clear_tests(struct mm_struct *mm, p4d_t *p4dp)
523 p4d_t p4d = READ_ONCE(*p4dp);
525 if (mm_pud_folded(mm))
528 pr_debug("Validating P4D clear\n");
529 p4d = __p4d(p4d_val(p4d) | RANDOM_ORVALUE);
530 WRITE_ONCE(*p4dp, p4d);
532 p4d = READ_ONCE(*p4dp);
533 WARN_ON(!p4d_none(p4d));
536 static void __init p4d_populate_tests(struct mm_struct *mm, p4d_t *p4dp,
541 if (mm_pud_folded(mm))
544 pr_debug("Validating P4D populate\n");
546 * This entry points to next level page table page.
547 * Hence this must not qualify as p4d_bad().
551 p4d_populate(mm, p4dp, pudp);
552 p4d = READ_ONCE(*p4dp);
553 WARN_ON(p4d_bad(p4d));
556 static void __init pgd_clear_tests(struct mm_struct *mm, pgd_t *pgdp)
558 pgd_t pgd = READ_ONCE(*pgdp);
560 if (mm_p4d_folded(mm))
563 pr_debug("Validating PGD clear\n");
564 pgd = __pgd(pgd_val(pgd) | RANDOM_ORVALUE);
565 WRITE_ONCE(*pgdp, pgd);
567 pgd = READ_ONCE(*pgdp);
568 WARN_ON(!pgd_none(pgd));
571 static void __init pgd_populate_tests(struct mm_struct *mm, pgd_t *pgdp,
576 if (mm_p4d_folded(mm))
579 pr_debug("Validating PGD populate\n");
581 * This entry points to next level page table page.
582 * Hence this must not qualify as pgd_bad().
586 pgd_populate(mm, pgdp, p4dp);
587 pgd = READ_ONCE(*pgdp);
588 WARN_ON(pgd_bad(pgd));
590 #else /* !__PAGETABLE_P4D_FOLDED */
591 static void __init p4d_clear_tests(struct mm_struct *mm, p4d_t *p4dp) { }
592 static void __init pgd_clear_tests(struct mm_struct *mm, pgd_t *pgdp) { }
593 static void __init p4d_populate_tests(struct mm_struct *mm, p4d_t *p4dp,
597 static void __init pgd_populate_tests(struct mm_struct *mm, pgd_t *pgdp,
601 #endif /* PAGETABLE_P4D_FOLDED */
603 static void __init pte_clear_tests(struct mm_struct *mm, pte_t *ptep,
604 unsigned long pfn, unsigned long vaddr,
607 pte_t pte = pfn_pte(pfn, prot);
609 pr_debug("Validating PTE clear\n");
611 pte = __pte(pte_val(pte) | RANDOM_ORVALUE);
613 set_pte_at(mm, vaddr, ptep, pte);
615 pte_clear(mm, vaddr, ptep);
616 pte = ptep_get(ptep);
617 WARN_ON(!pte_none(pte));
620 static void __init pmd_clear_tests(struct mm_struct *mm, pmd_t *pmdp)
622 pmd_t pmd = READ_ONCE(*pmdp);
624 pr_debug("Validating PMD clear\n");
625 pmd = __pmd(pmd_val(pmd) | RANDOM_ORVALUE);
626 WRITE_ONCE(*pmdp, pmd);
628 pmd = READ_ONCE(*pmdp);
629 WARN_ON(!pmd_none(pmd));
632 static void __init pmd_populate_tests(struct mm_struct *mm, pmd_t *pmdp,
637 pr_debug("Validating PMD populate\n");
639 * This entry points to next level page table page.
640 * Hence this must not qualify as pmd_bad().
642 pmd_populate(mm, pmdp, pgtable);
643 pmd = READ_ONCE(*pmdp);
644 WARN_ON(pmd_bad(pmd));
647 static void __init pte_special_tests(unsigned long pfn, pgprot_t prot)
649 pte_t pte = pfn_pte(pfn, prot);
651 if (!IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL))
654 pr_debug("Validating PTE special\n");
655 WARN_ON(!pte_special(pte_mkspecial(pte)));
658 static void __init pte_protnone_tests(unsigned long pfn, pgprot_t prot)
660 pte_t pte = pfn_pte(pfn, prot);
662 if (!IS_ENABLED(CONFIG_NUMA_BALANCING))
665 pr_debug("Validating PTE protnone\n");
666 WARN_ON(!pte_protnone(pte));
667 WARN_ON(!pte_present(pte));
670 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
671 static void __init pmd_protnone_tests(unsigned long pfn, pgprot_t prot)
675 if (!IS_ENABLED(CONFIG_NUMA_BALANCING))
678 if (!has_transparent_hugepage())
681 pr_debug("Validating PMD protnone\n");
682 pmd = pmd_mkhuge(pfn_pmd(pfn, prot));
683 WARN_ON(!pmd_protnone(pmd));
684 WARN_ON(!pmd_present(pmd));
686 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
687 static void __init pmd_protnone_tests(unsigned long pfn, pgprot_t prot) { }
688 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
690 #ifdef CONFIG_ARCH_HAS_PTE_DEVMAP
691 static void __init pte_devmap_tests(unsigned long pfn, pgprot_t prot)
693 pte_t pte = pfn_pte(pfn, prot);
695 pr_debug("Validating PTE devmap\n");
696 WARN_ON(!pte_devmap(pte_mkdevmap(pte)));
699 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
700 static void __init pmd_devmap_tests(unsigned long pfn, pgprot_t prot)
704 if (!has_transparent_hugepage())
707 pr_debug("Validating PMD devmap\n");
708 pmd = pfn_pmd(pfn, prot);
709 WARN_ON(!pmd_devmap(pmd_mkdevmap(pmd)));
712 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
713 static void __init pud_devmap_tests(unsigned long pfn, pgprot_t prot)
717 if (!has_transparent_hugepage())
720 pr_debug("Validating PUD devmap\n");
721 pud = pfn_pud(pfn, prot);
722 WARN_ON(!pud_devmap(pud_mkdevmap(pud)));
724 #else /* !CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
725 static void __init pud_devmap_tests(unsigned long pfn, pgprot_t prot) { }
726 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
727 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
728 static void __init pmd_devmap_tests(unsigned long pfn, pgprot_t prot) { }
729 static void __init pud_devmap_tests(unsigned long pfn, pgprot_t prot) { }
730 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
732 static void __init pte_devmap_tests(unsigned long pfn, pgprot_t prot) { }
733 static void __init pmd_devmap_tests(unsigned long pfn, pgprot_t prot) { }
734 static void __init pud_devmap_tests(unsigned long pfn, pgprot_t prot) { }
735 #endif /* CONFIG_ARCH_HAS_PTE_DEVMAP */
737 static void __init pte_soft_dirty_tests(unsigned long pfn, pgprot_t prot)
739 pte_t pte = pfn_pte(pfn, prot);
741 if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY))
744 pr_debug("Validating PTE soft dirty\n");
745 WARN_ON(!pte_soft_dirty(pte_mksoft_dirty(pte)));
746 WARN_ON(pte_soft_dirty(pte_clear_soft_dirty(pte)));
749 static void __init pte_swap_soft_dirty_tests(unsigned long pfn, pgprot_t prot)
751 pte_t pte = pfn_pte(pfn, prot);
753 if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY))
756 pr_debug("Validating PTE swap soft dirty\n");
757 WARN_ON(!pte_swp_soft_dirty(pte_swp_mksoft_dirty(pte)));
758 WARN_ON(pte_swp_soft_dirty(pte_swp_clear_soft_dirty(pte)));
761 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
762 static void __init pmd_soft_dirty_tests(unsigned long pfn, pgprot_t prot)
766 if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY))
769 if (!has_transparent_hugepage())
772 pr_debug("Validating PMD soft dirty\n");
773 pmd = pfn_pmd(pfn, prot);
774 WARN_ON(!pmd_soft_dirty(pmd_mksoft_dirty(pmd)));
775 WARN_ON(pmd_soft_dirty(pmd_clear_soft_dirty(pmd)));
778 static void __init pmd_swap_soft_dirty_tests(unsigned long pfn, pgprot_t prot)
782 if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) ||
783 !IS_ENABLED(CONFIG_ARCH_ENABLE_THP_MIGRATION))
786 if (!has_transparent_hugepage())
789 pr_debug("Validating PMD swap soft dirty\n");
790 pmd = pfn_pmd(pfn, prot);
791 WARN_ON(!pmd_swp_soft_dirty(pmd_swp_mksoft_dirty(pmd)));
792 WARN_ON(pmd_swp_soft_dirty(pmd_swp_clear_soft_dirty(pmd)));
794 #else /* !CONFIG_ARCH_HAS_PTE_DEVMAP */
795 static void __init pmd_soft_dirty_tests(unsigned long pfn, pgprot_t prot) { }
796 static void __init pmd_swap_soft_dirty_tests(unsigned long pfn, pgprot_t prot)
799 #endif /* CONFIG_ARCH_HAS_PTE_DEVMAP */
801 static void __init pte_swap_tests(unsigned long pfn, pgprot_t prot)
806 pr_debug("Validating PTE swap\n");
807 pte = pfn_pte(pfn, prot);
808 swp = __pte_to_swp_entry(pte);
809 pte = __swp_entry_to_pte(swp);
810 WARN_ON(pfn != pte_pfn(pte));
813 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
814 static void __init pmd_swap_tests(unsigned long pfn, pgprot_t prot)
819 if (!has_transparent_hugepage())
822 pr_debug("Validating PMD swap\n");
823 pmd = pfn_pmd(pfn, prot);
824 swp = __pmd_to_swp_entry(pmd);
825 pmd = __swp_entry_to_pmd(swp);
826 WARN_ON(pfn != pmd_pfn(pmd));
828 #else /* !CONFIG_ARCH_ENABLE_THP_MIGRATION */
829 static void __init pmd_swap_tests(unsigned long pfn, pgprot_t prot) { }
830 #endif /* CONFIG_ARCH_ENABLE_THP_MIGRATION */
832 static void __init swap_migration_tests(void)
837 if (!IS_ENABLED(CONFIG_MIGRATION))
840 pr_debug("Validating swap migration\n");
842 * swap_migration_tests() requires a dedicated page as it needs to
843 * be locked before creating a migration entry from it. Locking the
844 * page that actually maps kernel text ('start_kernel') can be real
845 * problematic. Lets allocate a dedicated page explicitly for this
846 * purpose that will be freed subsequently.
848 page = alloc_page(GFP_KERNEL);
850 pr_err("page allocation failed\n");
855 * make_migration_entry() expects given page to be
856 * locked, otherwise it stumbles upon a BUG_ON().
858 __SetPageLocked(page);
859 swp = make_migration_entry(page, 1);
860 WARN_ON(!is_migration_entry(swp));
861 WARN_ON(!is_write_migration_entry(swp));
863 make_migration_entry_read(&swp);
864 WARN_ON(!is_migration_entry(swp));
865 WARN_ON(is_write_migration_entry(swp));
867 swp = make_migration_entry(page, 0);
868 WARN_ON(!is_migration_entry(swp));
869 WARN_ON(is_write_migration_entry(swp));
870 __ClearPageLocked(page);
874 #ifdef CONFIG_HUGETLB_PAGE
875 static void __init hugetlb_basic_tests(unsigned long pfn, pgprot_t prot)
880 pr_debug("Validating HugeTLB basic\n");
882 * Accessing the page associated with the pfn is safe here,
883 * as it was previously derived from a real kernel symbol.
885 page = pfn_to_page(pfn);
886 pte = mk_huge_pte(page, prot);
888 WARN_ON(!huge_pte_dirty(huge_pte_mkdirty(pte)));
889 WARN_ON(!huge_pte_write(huge_pte_mkwrite(huge_pte_wrprotect(pte))));
890 WARN_ON(huge_pte_write(huge_pte_wrprotect(huge_pte_mkwrite(pte))));
892 #ifdef CONFIG_ARCH_WANT_GENERAL_HUGETLB
893 pte = pfn_pte(pfn, prot);
895 WARN_ON(!pte_huge(pte_mkhuge(pte)));
896 #endif /* CONFIG_ARCH_WANT_GENERAL_HUGETLB */
898 #else /* !CONFIG_HUGETLB_PAGE */
899 static void __init hugetlb_basic_tests(unsigned long pfn, pgprot_t prot) { }
900 #endif /* CONFIG_HUGETLB_PAGE */
902 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
903 static void __init pmd_thp_tests(unsigned long pfn, pgprot_t prot)
907 if (!has_transparent_hugepage())
910 pr_debug("Validating PMD based THP\n");
912 * pmd_trans_huge() and pmd_present() must return positive after
913 * MMU invalidation with pmd_mkinvalid(). This behavior is an
914 * optimization for transparent huge page. pmd_trans_huge() must
915 * be true if pmd_page() returns a valid THP to avoid taking the
916 * pmd_lock when others walk over non transhuge pmds (i.e. there
917 * are no THP allocated). Especially when splitting a THP and
918 * removing the present bit from the pmd, pmd_trans_huge() still
919 * needs to return true. pmd_present() should be true whenever
920 * pmd_trans_huge() returns true.
922 pmd = pfn_pmd(pfn, prot);
923 WARN_ON(!pmd_trans_huge(pmd_mkhuge(pmd)));
925 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
926 WARN_ON(!pmd_trans_huge(pmd_mkinvalid(pmd_mkhuge(pmd))));
927 WARN_ON(!pmd_present(pmd_mkinvalid(pmd_mkhuge(pmd))));
928 #endif /* __HAVE_ARCH_PMDP_INVALIDATE */
931 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
932 static void __init pud_thp_tests(unsigned long pfn, pgprot_t prot)
936 if (!has_transparent_hugepage())
939 pr_debug("Validating PUD based THP\n");
940 pud = pfn_pud(pfn, prot);
941 WARN_ON(!pud_trans_huge(pud_mkhuge(pud)));
944 * pud_mkinvalid() has been dropped for now. Enable back
945 * these tests when it comes back with a modified pud_present().
947 * WARN_ON(!pud_trans_huge(pud_mkinvalid(pud_mkhuge(pud))));
948 * WARN_ON(!pud_present(pud_mkinvalid(pud_mkhuge(pud))));
951 #else /* !CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
952 static void __init pud_thp_tests(unsigned long pfn, pgprot_t prot) { }
953 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
954 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
955 static void __init pmd_thp_tests(unsigned long pfn, pgprot_t prot) { }
956 static void __init pud_thp_tests(unsigned long pfn, pgprot_t prot) { }
957 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
959 static unsigned long __init get_random_vaddr(void)
961 unsigned long random_vaddr, random_pages, total_user_pages;
963 total_user_pages = (TASK_SIZE - FIRST_USER_ADDRESS) / PAGE_SIZE;
965 random_pages = get_random_long() % total_user_pages;
966 random_vaddr = FIRST_USER_ADDRESS + random_pages * PAGE_SIZE;
971 static int __init debug_vm_pgtable(void)
973 struct vm_area_struct *vma;
974 struct mm_struct *mm;
976 p4d_t *p4dp, *saved_p4dp;
977 pud_t *pudp, *saved_pudp;
978 pmd_t *pmdp, *saved_pmdp, pmd;
980 pgtable_t saved_ptep;
981 pgprot_t prot, protnone;
983 unsigned long vaddr, pte_aligned, pmd_aligned;
984 unsigned long pud_aligned, p4d_aligned, pgd_aligned;
985 spinlock_t *ptl = NULL;
988 pr_info("Validating architecture page table helpers\n");
989 prot = vm_get_page_prot(VMFLAGS);
990 vaddr = get_random_vaddr();
993 pr_err("mm_struct allocation failed\n");
998 * __P000 (or even __S000) will help create page table entries with
999 * PROT_NONE permission as required for pxx_protnone_tests().
1003 vma = vm_area_alloc(mm);
1005 pr_err("vma allocation failed\n");
1010 * PFN for mapping at PTE level is determined from a standard kernel
1011 * text symbol. But pfns for higher page table levels are derived by
1012 * masking lower bits of this real pfn. These derived pfns might not
1013 * exist on the platform but that does not really matter as pfn_pxx()
1014 * helpers will still create appropriate entries for the test. This
1015 * helps avoid large memory block allocations to be used for mapping
1016 * at higher page table levels.
1018 paddr = __pa_symbol(&start_kernel);
1020 pte_aligned = (paddr & PAGE_MASK) >> PAGE_SHIFT;
1021 pmd_aligned = (paddr & PMD_MASK) >> PAGE_SHIFT;
1022 pud_aligned = (paddr & PUD_MASK) >> PAGE_SHIFT;
1023 p4d_aligned = (paddr & P4D_MASK) >> PAGE_SHIFT;
1024 pgd_aligned = (paddr & PGDIR_MASK) >> PAGE_SHIFT;
1025 WARN_ON(!pfn_valid(pte_aligned));
1027 pgdp = pgd_offset(mm, vaddr);
1028 p4dp = p4d_alloc(mm, pgdp, vaddr);
1029 pudp = pud_alloc(mm, p4dp, vaddr);
1030 pmdp = pmd_alloc(mm, pudp, vaddr);
1032 * Allocate pgtable_t
1034 if (pte_alloc(mm, pmdp)) {
1035 pr_err("pgtable allocation failed\n");
1040 * Save all the page table page addresses as the page table
1041 * entries will be used for testing with random or garbage
1042 * values. These saved addresses will be used for freeing
1045 pmd = READ_ONCE(*pmdp);
1046 saved_p4dp = p4d_offset(pgdp, 0UL);
1047 saved_pudp = pud_offset(p4dp, 0UL);
1048 saved_pmdp = pmd_offset(pudp, 0UL);
1049 saved_ptep = pmd_pgtable(pmd);
1052 * Iterate over the protection_map[] to make sure that all
1053 * the basic page table transformation validations just hold
1054 * true irrespective of the starting protection value for a
1055 * given page table entry.
1057 for (idx = 0; idx < ARRAY_SIZE(protection_map); idx++) {
1058 pte_basic_tests(pte_aligned, idx);
1059 pmd_basic_tests(pmd_aligned, idx);
1060 pud_basic_tests(mm, pud_aligned, idx);
1064 * Both P4D and PGD level tests are very basic which do not
1065 * involve creating page table entries from the protection
1066 * value and the given pfn. Hence just keep them out from
1067 * the above iteration for now to save some test execution
1070 p4d_basic_tests(p4d_aligned, prot);
1071 pgd_basic_tests(pgd_aligned, prot);
1073 pmd_leaf_tests(pmd_aligned, prot);
1074 pud_leaf_tests(pud_aligned, prot);
1076 pte_savedwrite_tests(pte_aligned, protnone);
1077 pmd_savedwrite_tests(pmd_aligned, protnone);
1079 pte_special_tests(pte_aligned, prot);
1080 pte_protnone_tests(pte_aligned, protnone);
1081 pmd_protnone_tests(pmd_aligned, protnone);
1083 pte_devmap_tests(pte_aligned, prot);
1084 pmd_devmap_tests(pmd_aligned, prot);
1085 pud_devmap_tests(pud_aligned, prot);
1087 pte_soft_dirty_tests(pte_aligned, prot);
1088 pmd_soft_dirty_tests(pmd_aligned, prot);
1089 pte_swap_soft_dirty_tests(pte_aligned, prot);
1090 pmd_swap_soft_dirty_tests(pmd_aligned, prot);
1092 pte_swap_tests(pte_aligned, prot);
1093 pmd_swap_tests(pmd_aligned, prot);
1095 swap_migration_tests();
1097 pmd_thp_tests(pmd_aligned, prot);
1098 pud_thp_tests(pud_aligned, prot);
1100 hugetlb_basic_tests(pte_aligned, prot);
1103 * Page table modifying tests. They need to hold
1104 * proper page table lock.
1107 ptep = pte_offset_map_lock(mm, pmdp, vaddr, &ptl);
1108 pte_clear_tests(mm, ptep, pte_aligned, vaddr, prot);
1109 pte_advanced_tests(mm, vma, ptep, pte_aligned, vaddr, prot);
1110 pte_unmap_unlock(ptep, ptl);
1112 ptl = pmd_lock(mm, pmdp);
1113 pmd_clear_tests(mm, pmdp);
1114 pmd_advanced_tests(mm, vma, pmdp, pmd_aligned, vaddr, prot, saved_ptep);
1115 pmd_huge_tests(pmdp, pmd_aligned, prot);
1116 pmd_populate_tests(mm, pmdp, saved_ptep);
1119 ptl = pud_lock(mm, pudp);
1120 pud_clear_tests(mm, pudp);
1121 pud_advanced_tests(mm, vma, pudp, pud_aligned, vaddr, prot);
1122 pud_huge_tests(pudp, pud_aligned, prot);
1123 pud_populate_tests(mm, pudp, saved_pmdp);
1126 spin_lock(&mm->page_table_lock);
1127 p4d_clear_tests(mm, p4dp);
1128 pgd_clear_tests(mm, pgdp);
1129 p4d_populate_tests(mm, p4dp, saved_pudp);
1130 pgd_populate_tests(mm, pgdp, saved_p4dp);
1131 spin_unlock(&mm->page_table_lock);
1133 p4d_free(mm, saved_p4dp);
1134 pud_free(mm, saved_pudp);
1135 pmd_free(mm, saved_pmdp);
1136 pte_free(mm, saved_ptep);
1145 late_initcall(debug_vm_pgtable);