Loading...
Note: File does not exist in v4.17.
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
33#include <asm/cacheflush.h>
34#include <asm/pgalloc.h>
35#include <asm/tlbflush.h>
36
37/*
38 * Please refer Documentation/mm/arch_pgtable_helpers.rst for the semantics
39 * expectations that are being validated here. All future changes in here
40 * or the documentation need to be in sync.
41 *
42 * On s390 platform, the lower 4 bits are used to identify given page table
43 * entry type. But these bits might affect the ability to clear entries with
44 * pxx_clear() because of how dynamic page table folding works on s390. So
45 * while loading up the entries do not change the lower 4 bits. It does not
46 * have affect any other platform. Also avoid the 62nd bit on ppc64 that is
47 * used to mark a pte entry.
48 */
49#define S390_SKIP_MASK GENMASK(3, 0)
50#if __BITS_PER_LONG == 64
51#define PPC64_SKIP_MASK GENMASK(62, 62)
52#else
53#define PPC64_SKIP_MASK 0x0
54#endif
55#define ARCH_SKIP_MASK (S390_SKIP_MASK | PPC64_SKIP_MASK)
56#define RANDOM_ORVALUE (GENMASK(BITS_PER_LONG - 1, 0) & ~ARCH_SKIP_MASK)
57#define RANDOM_NZVALUE GENMASK(7, 0)
58
59struct pgtable_debug_args {
60 struct mm_struct *mm;
61 struct vm_area_struct *vma;
62
63 pgd_t *pgdp;
64 p4d_t *p4dp;
65 pud_t *pudp;
66 pmd_t *pmdp;
67 pte_t *ptep;
68
69 p4d_t *start_p4dp;
70 pud_t *start_pudp;
71 pmd_t *start_pmdp;
72 pgtable_t start_ptep;
73
74 unsigned long vaddr;
75 pgprot_t page_prot;
76 pgprot_t page_prot_none;
77
78 bool is_contiguous_page;
79 unsigned long pud_pfn;
80 unsigned long pmd_pfn;
81 unsigned long pte_pfn;
82
83 unsigned long fixed_pgd_pfn;
84 unsigned long fixed_p4d_pfn;
85 unsigned long fixed_pud_pfn;
86 unsigned long fixed_pmd_pfn;
87 unsigned long fixed_pte_pfn;
88};
89
90static void __init pte_basic_tests(struct pgtable_debug_args *args, int idx)
91{
92 pgprot_t prot = vm_get_page_prot(idx);
93 pte_t pte = pfn_pte(args->fixed_pte_pfn, prot);
94 unsigned long val = idx, *ptr = &val;
95
96 pr_debug("Validating PTE basic (%pGv)\n", ptr);
97
98 /*
99 * This test needs to be executed after the given page table entry
100 * is created with pfn_pte() to make sure that vm_get_page_prot(idx)
101 * does not have the dirty bit enabled from the beginning. This is
102 * important for platforms like arm64 where (!PTE_RDONLY) indicate
103 * dirty bit being set.
104 */
105 WARN_ON(pte_dirty(pte_wrprotect(pte)));
106
107 WARN_ON(!pte_same(pte, pte));
108 WARN_ON(!pte_young(pte_mkyoung(pte_mkold(pte))));
109 WARN_ON(!pte_dirty(pte_mkdirty(pte_mkclean(pte))));
110 WARN_ON(!pte_write(pte_mkwrite(pte_wrprotect(pte))));
111 WARN_ON(pte_young(pte_mkold(pte_mkyoung(pte))));
112 WARN_ON(pte_dirty(pte_mkclean(pte_mkdirty(pte))));
113 WARN_ON(pte_write(pte_wrprotect(pte_mkwrite(pte))));
114 WARN_ON(pte_dirty(pte_wrprotect(pte_mkclean(pte))));
115 WARN_ON(!pte_dirty(pte_wrprotect(pte_mkdirty(pte))));
116}
117
118static void __init pte_advanced_tests(struct pgtable_debug_args *args)
119{
120 struct page *page;
121 pte_t pte;
122
123 /*
124 * Architectures optimize set_pte_at by avoiding TLB flush.
125 * This requires set_pte_at to be not used to update an
126 * existing pte entry. Clear pte before we do set_pte_at
127 *
128 * flush_dcache_page() is called after set_pte_at() to clear
129 * PG_arch_1 for the page on ARM64. The page flag isn't cleared
130 * when it's released and page allocation check will fail when
131 * the page is allocated again. For architectures other than ARM64,
132 * the unexpected overhead of cache flushing is acceptable.
133 */
134 page = (args->pte_pfn != ULONG_MAX) ? pfn_to_page(args->pte_pfn) : NULL;
135 if (!page)
136 return;
137
138 pr_debug("Validating PTE advanced\n");
139 pte = pfn_pte(args->pte_pfn, args->page_prot);
140 set_pte_at(args->mm, args->vaddr, args->ptep, pte);
141 flush_dcache_page(page);
142 ptep_set_wrprotect(args->mm, args->vaddr, args->ptep);
143 pte = ptep_get(args->ptep);
144 WARN_ON(pte_write(pte));
145 ptep_get_and_clear(args->mm, args->vaddr, args->ptep);
146 pte = ptep_get(args->ptep);
147 WARN_ON(!pte_none(pte));
148
149 pte = pfn_pte(args->pte_pfn, args->page_prot);
150 pte = pte_wrprotect(pte);
151 pte = pte_mkclean(pte);
152 set_pte_at(args->mm, args->vaddr, args->ptep, pte);
153 flush_dcache_page(page);
154 pte = pte_mkwrite(pte);
155 pte = pte_mkdirty(pte);
156 ptep_set_access_flags(args->vma, args->vaddr, args->ptep, pte, 1);
157 pte = ptep_get(args->ptep);
158 WARN_ON(!(pte_write(pte) && pte_dirty(pte)));
159 ptep_get_and_clear_full(args->mm, args->vaddr, args->ptep, 1);
160 pte = ptep_get(args->ptep);
161 WARN_ON(!pte_none(pte));
162
163 pte = pfn_pte(args->pte_pfn, args->page_prot);
164 pte = pte_mkyoung(pte);
165 set_pte_at(args->mm, args->vaddr, args->ptep, pte);
166 flush_dcache_page(page);
167 ptep_test_and_clear_young(args->vma, args->vaddr, args->ptep);
168 pte = ptep_get(args->ptep);
169 WARN_ON(pte_young(pte));
170
171 ptep_get_and_clear_full(args->mm, args->vaddr, args->ptep, 1);
172}
173
174#ifdef CONFIG_TRANSPARENT_HUGEPAGE
175static void __init pmd_basic_tests(struct pgtable_debug_args *args, int idx)
176{
177 pgprot_t prot = vm_get_page_prot(idx);
178 unsigned long val = idx, *ptr = &val;
179 pmd_t pmd;
180
181 if (!has_transparent_hugepage())
182 return;
183
184 pr_debug("Validating PMD basic (%pGv)\n", ptr);
185 pmd = pfn_pmd(args->fixed_pmd_pfn, prot);
186
187 /*
188 * This test needs to be executed after the given page table entry
189 * is created with pfn_pmd() to make sure that vm_get_page_prot(idx)
190 * does not have the dirty bit enabled from the beginning. This is
191 * important for platforms like arm64 where (!PTE_RDONLY) indicate
192 * dirty bit being set.
193 */
194 WARN_ON(pmd_dirty(pmd_wrprotect(pmd)));
195
196
197 WARN_ON(!pmd_same(pmd, pmd));
198 WARN_ON(!pmd_young(pmd_mkyoung(pmd_mkold(pmd))));
199 WARN_ON(!pmd_dirty(pmd_mkdirty(pmd_mkclean(pmd))));
200 WARN_ON(!pmd_write(pmd_mkwrite(pmd_wrprotect(pmd))));
201 WARN_ON(pmd_young(pmd_mkold(pmd_mkyoung(pmd))));
202 WARN_ON(pmd_dirty(pmd_mkclean(pmd_mkdirty(pmd))));
203 WARN_ON(pmd_write(pmd_wrprotect(pmd_mkwrite(pmd))));
204 WARN_ON(pmd_dirty(pmd_wrprotect(pmd_mkclean(pmd))));
205 WARN_ON(!pmd_dirty(pmd_wrprotect(pmd_mkdirty(pmd))));
206 /*
207 * A huge page does not point to next level page table
208 * entry. Hence this must qualify as pmd_bad().
209 */
210 WARN_ON(!pmd_bad(pmd_mkhuge(pmd)));
211}
212
213static void __init pmd_advanced_tests(struct pgtable_debug_args *args)
214{
215 struct page *page;
216 pmd_t pmd;
217 unsigned long vaddr = args->vaddr;
218
219 if (!has_transparent_hugepage())
220 return;
221
222 page = (args->pmd_pfn != ULONG_MAX) ? pfn_to_page(args->pmd_pfn) : NULL;
223 if (!page)
224 return;
225
226 /*
227 * flush_dcache_page() is called after set_pmd_at() to clear
228 * PG_arch_1 for the page on ARM64. The page flag isn't cleared
229 * when it's released and page allocation check will fail when
230 * the page is allocated again. For architectures other than ARM64,
231 * the unexpected overhead of cache flushing is acceptable.
232 */
233 pr_debug("Validating PMD advanced\n");
234 /* Align the address wrt HPAGE_PMD_SIZE */
235 vaddr &= HPAGE_PMD_MASK;
236
237 pgtable_trans_huge_deposit(args->mm, args->pmdp, args->start_ptep);
238
239 pmd = pfn_pmd(args->pmd_pfn, args->page_prot);
240 set_pmd_at(args->mm, vaddr, args->pmdp, pmd);
241 flush_dcache_page(page);
242 pmdp_set_wrprotect(args->mm, vaddr, args->pmdp);
243 pmd = READ_ONCE(*args->pmdp);
244 WARN_ON(pmd_write(pmd));
245 pmdp_huge_get_and_clear(args->mm, vaddr, args->pmdp);
246 pmd = READ_ONCE(*args->pmdp);
247 WARN_ON(!pmd_none(pmd));
248
249 pmd = pfn_pmd(args->pmd_pfn, args->page_prot);
250 pmd = pmd_wrprotect(pmd);
251 pmd = pmd_mkclean(pmd);
252 set_pmd_at(args->mm, vaddr, args->pmdp, pmd);
253 flush_dcache_page(page);
254 pmd = pmd_mkwrite(pmd);
255 pmd = pmd_mkdirty(pmd);
256 pmdp_set_access_flags(args->vma, vaddr, args->pmdp, pmd, 1);
257 pmd = READ_ONCE(*args->pmdp);
258 WARN_ON(!(pmd_write(pmd) && pmd_dirty(pmd)));
259 pmdp_huge_get_and_clear_full(args->vma, vaddr, args->pmdp, 1);
260 pmd = READ_ONCE(*args->pmdp);
261 WARN_ON(!pmd_none(pmd));
262
263 pmd = pmd_mkhuge(pfn_pmd(args->pmd_pfn, args->page_prot));
264 pmd = pmd_mkyoung(pmd);
265 set_pmd_at(args->mm, vaddr, args->pmdp, pmd);
266 flush_dcache_page(page);
267 pmdp_test_and_clear_young(args->vma, vaddr, args->pmdp);
268 pmd = READ_ONCE(*args->pmdp);
269 WARN_ON(pmd_young(pmd));
270
271 /* Clear the pte entries */
272 pmdp_huge_get_and_clear(args->mm, vaddr, args->pmdp);
273 pgtable_trans_huge_withdraw(args->mm, args->pmdp);
274}
275
276static void __init pmd_leaf_tests(struct pgtable_debug_args *args)
277{
278 pmd_t pmd;
279
280 if (!has_transparent_hugepage())
281 return;
282
283 pr_debug("Validating PMD leaf\n");
284 pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot);
285
286 /*
287 * PMD based THP is a leaf entry.
288 */
289 pmd = pmd_mkhuge(pmd);
290 WARN_ON(!pmd_leaf(pmd));
291}
292
293#ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
294static void __init pud_basic_tests(struct pgtable_debug_args *args, int idx)
295{
296 pgprot_t prot = vm_get_page_prot(idx);
297 unsigned long val = idx, *ptr = &val;
298 pud_t pud;
299
300 if (!has_transparent_hugepage())
301 return;
302
303 pr_debug("Validating PUD basic (%pGv)\n", ptr);
304 pud = pfn_pud(args->fixed_pud_pfn, prot);
305
306 /*
307 * This test needs to be executed after the given page table entry
308 * is created with pfn_pud() to make sure that vm_get_page_prot(idx)
309 * does not have the dirty bit enabled from the beginning. This is
310 * important for platforms like arm64 where (!PTE_RDONLY) indicate
311 * dirty bit being set.
312 */
313 WARN_ON(pud_dirty(pud_wrprotect(pud)));
314
315 WARN_ON(!pud_same(pud, pud));
316 WARN_ON(!pud_young(pud_mkyoung(pud_mkold(pud))));
317 WARN_ON(!pud_dirty(pud_mkdirty(pud_mkclean(pud))));
318 WARN_ON(pud_dirty(pud_mkclean(pud_mkdirty(pud))));
319 WARN_ON(!pud_write(pud_mkwrite(pud_wrprotect(pud))));
320 WARN_ON(pud_write(pud_wrprotect(pud_mkwrite(pud))));
321 WARN_ON(pud_young(pud_mkold(pud_mkyoung(pud))));
322 WARN_ON(pud_dirty(pud_wrprotect(pud_mkclean(pud))));
323 WARN_ON(!pud_dirty(pud_wrprotect(pud_mkdirty(pud))));
324
325 if (mm_pmd_folded(args->mm))
326 return;
327
328 /*
329 * A huge page does not point to next level page table
330 * entry. Hence this must qualify as pud_bad().
331 */
332 WARN_ON(!pud_bad(pud_mkhuge(pud)));
333}
334
335static void __init pud_advanced_tests(struct pgtable_debug_args *args)
336{
337 struct page *page;
338 unsigned long vaddr = args->vaddr;
339 pud_t pud;
340
341 if (!has_transparent_hugepage())
342 return;
343
344 page = (args->pud_pfn != ULONG_MAX) ? pfn_to_page(args->pud_pfn) : NULL;
345 if (!page)
346 return;
347
348 /*
349 * flush_dcache_page() is called after set_pud_at() to clear
350 * PG_arch_1 for the page on ARM64. The page flag isn't cleared
351 * when it's released and page allocation check will fail when
352 * the page is allocated again. For architectures other than ARM64,
353 * the unexpected overhead of cache flushing is acceptable.
354 */
355 pr_debug("Validating PUD advanced\n");
356 /* Align the address wrt HPAGE_PUD_SIZE */
357 vaddr &= HPAGE_PUD_MASK;
358
359 pud = pfn_pud(args->pud_pfn, args->page_prot);
360 set_pud_at(args->mm, vaddr, args->pudp, pud);
361 flush_dcache_page(page);
362 pudp_set_wrprotect(args->mm, vaddr, args->pudp);
363 pud = READ_ONCE(*args->pudp);
364 WARN_ON(pud_write(pud));
365
366#ifndef __PAGETABLE_PMD_FOLDED
367 pudp_huge_get_and_clear(args->mm, vaddr, args->pudp);
368 pud = READ_ONCE(*args->pudp);
369 WARN_ON(!pud_none(pud));
370#endif /* __PAGETABLE_PMD_FOLDED */
371 pud = pfn_pud(args->pud_pfn, args->page_prot);
372 pud = pud_wrprotect(pud);
373 pud = pud_mkclean(pud);
374 set_pud_at(args->mm, vaddr, args->pudp, pud);
375 flush_dcache_page(page);
376 pud = pud_mkwrite(pud);
377 pud = pud_mkdirty(pud);
378 pudp_set_access_flags(args->vma, vaddr, args->pudp, pud, 1);
379 pud = READ_ONCE(*args->pudp);
380 WARN_ON(!(pud_write(pud) && pud_dirty(pud)));
381
382#ifndef __PAGETABLE_PMD_FOLDED
383 pudp_huge_get_and_clear_full(args->mm, vaddr, args->pudp, 1);
384 pud = READ_ONCE(*args->pudp);
385 WARN_ON(!pud_none(pud));
386#endif /* __PAGETABLE_PMD_FOLDED */
387
388 pud = pfn_pud(args->pud_pfn, args->page_prot);
389 pud = pud_mkyoung(pud);
390 set_pud_at(args->mm, vaddr, args->pudp, pud);
391 flush_dcache_page(page);
392 pudp_test_and_clear_young(args->vma, vaddr, args->pudp);
393 pud = READ_ONCE(*args->pudp);
394 WARN_ON(pud_young(pud));
395
396 pudp_huge_get_and_clear(args->mm, vaddr, args->pudp);
397}
398
399static void __init pud_leaf_tests(struct pgtable_debug_args *args)
400{
401 pud_t pud;
402
403 if (!has_transparent_hugepage())
404 return;
405
406 pr_debug("Validating PUD leaf\n");
407 pud = pfn_pud(args->fixed_pud_pfn, args->page_prot);
408 /*
409 * PUD based THP is a leaf entry.
410 */
411 pud = pud_mkhuge(pud);
412 WARN_ON(!pud_leaf(pud));
413}
414#else /* !CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
415static void __init pud_basic_tests(struct pgtable_debug_args *args, int idx) { }
416static void __init pud_advanced_tests(struct pgtable_debug_args *args) { }
417static void __init pud_leaf_tests(struct pgtable_debug_args *args) { }
418#endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
419#else /* !CONFIG_TRANSPARENT_HUGEPAGE */
420static void __init pmd_basic_tests(struct pgtable_debug_args *args, int idx) { }
421static void __init pud_basic_tests(struct pgtable_debug_args *args, int idx) { }
422static void __init pmd_advanced_tests(struct pgtable_debug_args *args) { }
423static void __init pud_advanced_tests(struct pgtable_debug_args *args) { }
424static void __init pmd_leaf_tests(struct pgtable_debug_args *args) { }
425static void __init pud_leaf_tests(struct pgtable_debug_args *args) { }
426#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
427
428#ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
429static void __init pmd_huge_tests(struct pgtable_debug_args *args)
430{
431 pmd_t pmd;
432
433 if (!arch_vmap_pmd_supported(args->page_prot))
434 return;
435
436 pr_debug("Validating PMD huge\n");
437 /*
438 * X86 defined pmd_set_huge() verifies that the given
439 * PMD is not a populated non-leaf entry.
440 */
441 WRITE_ONCE(*args->pmdp, __pmd(0));
442 WARN_ON(!pmd_set_huge(args->pmdp, __pfn_to_phys(args->fixed_pmd_pfn), args->page_prot));
443 WARN_ON(!pmd_clear_huge(args->pmdp));
444 pmd = READ_ONCE(*args->pmdp);
445 WARN_ON(!pmd_none(pmd));
446}
447
448static void __init pud_huge_tests(struct pgtable_debug_args *args)
449{
450 pud_t pud;
451
452 if (!arch_vmap_pud_supported(args->page_prot))
453 return;
454
455 pr_debug("Validating PUD huge\n");
456 /*
457 * X86 defined pud_set_huge() verifies that the given
458 * PUD is not a populated non-leaf entry.
459 */
460 WRITE_ONCE(*args->pudp, __pud(0));
461 WARN_ON(!pud_set_huge(args->pudp, __pfn_to_phys(args->fixed_pud_pfn), args->page_prot));
462 WARN_ON(!pud_clear_huge(args->pudp));
463 pud = READ_ONCE(*args->pudp);
464 WARN_ON(!pud_none(pud));
465}
466#else /* !CONFIG_HAVE_ARCH_HUGE_VMAP */
467static void __init pmd_huge_tests(struct pgtable_debug_args *args) { }
468static void __init pud_huge_tests(struct pgtable_debug_args *args) { }
469#endif /* CONFIG_HAVE_ARCH_HUGE_VMAP */
470
471static void __init p4d_basic_tests(struct pgtable_debug_args *args)
472{
473 p4d_t p4d;
474
475 pr_debug("Validating P4D basic\n");
476 memset(&p4d, RANDOM_NZVALUE, sizeof(p4d_t));
477 WARN_ON(!p4d_same(p4d, p4d));
478}
479
480static void __init pgd_basic_tests(struct pgtable_debug_args *args)
481{
482 pgd_t pgd;
483
484 pr_debug("Validating PGD basic\n");
485 memset(&pgd, RANDOM_NZVALUE, sizeof(pgd_t));
486 WARN_ON(!pgd_same(pgd, pgd));
487}
488
489#ifndef __PAGETABLE_PUD_FOLDED
490static void __init pud_clear_tests(struct pgtable_debug_args *args)
491{
492 pud_t pud = READ_ONCE(*args->pudp);
493
494 if (mm_pmd_folded(args->mm))
495 return;
496
497 pr_debug("Validating PUD clear\n");
498 pud = __pud(pud_val(pud) | RANDOM_ORVALUE);
499 WRITE_ONCE(*args->pudp, pud);
500 pud_clear(args->pudp);
501 pud = READ_ONCE(*args->pudp);
502 WARN_ON(!pud_none(pud));
503}
504
505static void __init pud_populate_tests(struct pgtable_debug_args *args)
506{
507 pud_t pud;
508
509 if (mm_pmd_folded(args->mm))
510 return;
511
512 pr_debug("Validating PUD populate\n");
513 /*
514 * This entry points to next level page table page.
515 * Hence this must not qualify as pud_bad().
516 */
517 pud_populate(args->mm, args->pudp, args->start_pmdp);
518 pud = READ_ONCE(*args->pudp);
519 WARN_ON(pud_bad(pud));
520}
521#else /* !__PAGETABLE_PUD_FOLDED */
522static void __init pud_clear_tests(struct pgtable_debug_args *args) { }
523static void __init pud_populate_tests(struct pgtable_debug_args *args) { }
524#endif /* PAGETABLE_PUD_FOLDED */
525
526#ifndef __PAGETABLE_P4D_FOLDED
527static void __init p4d_clear_tests(struct pgtable_debug_args *args)
528{
529 p4d_t p4d = READ_ONCE(*args->p4dp);
530
531 if (mm_pud_folded(args->mm))
532 return;
533
534 pr_debug("Validating P4D clear\n");
535 p4d = __p4d(p4d_val(p4d) | RANDOM_ORVALUE);
536 WRITE_ONCE(*args->p4dp, p4d);
537 p4d_clear(args->p4dp);
538 p4d = READ_ONCE(*args->p4dp);
539 WARN_ON(!p4d_none(p4d));
540}
541
542static void __init p4d_populate_tests(struct pgtable_debug_args *args)
543{
544 p4d_t p4d;
545
546 if (mm_pud_folded(args->mm))
547 return;
548
549 pr_debug("Validating P4D populate\n");
550 /*
551 * This entry points to next level page table page.
552 * Hence this must not qualify as p4d_bad().
553 */
554 pud_clear(args->pudp);
555 p4d_clear(args->p4dp);
556 p4d_populate(args->mm, args->p4dp, args->start_pudp);
557 p4d = READ_ONCE(*args->p4dp);
558 WARN_ON(p4d_bad(p4d));
559}
560
561static void __init pgd_clear_tests(struct pgtable_debug_args *args)
562{
563 pgd_t pgd = READ_ONCE(*(args->pgdp));
564
565 if (mm_p4d_folded(args->mm))
566 return;
567
568 pr_debug("Validating PGD clear\n");
569 pgd = __pgd(pgd_val(pgd) | RANDOM_ORVALUE);
570 WRITE_ONCE(*args->pgdp, pgd);
571 pgd_clear(args->pgdp);
572 pgd = READ_ONCE(*args->pgdp);
573 WARN_ON(!pgd_none(pgd));
574}
575
576static void __init pgd_populate_tests(struct pgtable_debug_args *args)
577{
578 pgd_t pgd;
579
580 if (mm_p4d_folded(args->mm))
581 return;
582
583 pr_debug("Validating PGD populate\n");
584 /*
585 * This entry points to next level page table page.
586 * Hence this must not qualify as pgd_bad().
587 */
588 p4d_clear(args->p4dp);
589 pgd_clear(args->pgdp);
590 pgd_populate(args->mm, args->pgdp, args->start_p4dp);
591 pgd = READ_ONCE(*args->pgdp);
592 WARN_ON(pgd_bad(pgd));
593}
594#else /* !__PAGETABLE_P4D_FOLDED */
595static void __init p4d_clear_tests(struct pgtable_debug_args *args) { }
596static void __init pgd_clear_tests(struct pgtable_debug_args *args) { }
597static void __init p4d_populate_tests(struct pgtable_debug_args *args) { }
598static void __init pgd_populate_tests(struct pgtable_debug_args *args) { }
599#endif /* PAGETABLE_P4D_FOLDED */
600
601static void __init pte_clear_tests(struct pgtable_debug_args *args)
602{
603 struct page *page;
604 pte_t pte = pfn_pte(args->pte_pfn, args->page_prot);
605
606 page = (args->pte_pfn != ULONG_MAX) ? pfn_to_page(args->pte_pfn) : NULL;
607 if (!page)
608 return;
609
610 /*
611 * flush_dcache_page() is called after set_pte_at() to clear
612 * PG_arch_1 for the page on ARM64. The page flag isn't cleared
613 * when it's released and page allocation check will fail when
614 * the page is allocated again. For architectures other than ARM64,
615 * the unexpected overhead of cache flushing is acceptable.
616 */
617 pr_debug("Validating PTE clear\n");
618#ifndef CONFIG_RISCV
619 pte = __pte(pte_val(pte) | RANDOM_ORVALUE);
620#endif
621 set_pte_at(args->mm, args->vaddr, args->ptep, pte);
622 flush_dcache_page(page);
623 barrier();
624 ptep_clear(args->mm, args->vaddr, args->ptep);
625 pte = ptep_get(args->ptep);
626 WARN_ON(!pte_none(pte));
627}
628
629static void __init pmd_clear_tests(struct pgtable_debug_args *args)
630{
631 pmd_t pmd = READ_ONCE(*args->pmdp);
632
633 pr_debug("Validating PMD clear\n");
634 pmd = __pmd(pmd_val(pmd) | RANDOM_ORVALUE);
635 WRITE_ONCE(*args->pmdp, pmd);
636 pmd_clear(args->pmdp);
637 pmd = READ_ONCE(*args->pmdp);
638 WARN_ON(!pmd_none(pmd));
639}
640
641static void __init pmd_populate_tests(struct pgtable_debug_args *args)
642{
643 pmd_t pmd;
644
645 pr_debug("Validating PMD populate\n");
646 /*
647 * This entry points to next level page table page.
648 * Hence this must not qualify as pmd_bad().
649 */
650 pmd_populate(args->mm, args->pmdp, args->start_ptep);
651 pmd = READ_ONCE(*args->pmdp);
652 WARN_ON(pmd_bad(pmd));
653}
654
655static void __init pte_special_tests(struct pgtable_debug_args *args)
656{
657 pte_t pte = pfn_pte(args->fixed_pte_pfn, args->page_prot);
658
659 if (!IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL))
660 return;
661
662 pr_debug("Validating PTE special\n");
663 WARN_ON(!pte_special(pte_mkspecial(pte)));
664}
665
666static void __init pte_protnone_tests(struct pgtable_debug_args *args)
667{
668 pte_t pte = pfn_pte(args->fixed_pte_pfn, args->page_prot_none);
669
670 if (!IS_ENABLED(CONFIG_NUMA_BALANCING))
671 return;
672
673 pr_debug("Validating PTE protnone\n");
674 WARN_ON(!pte_protnone(pte));
675 WARN_ON(!pte_present(pte));
676}
677
678#ifdef CONFIG_TRANSPARENT_HUGEPAGE
679static void __init pmd_protnone_tests(struct pgtable_debug_args *args)
680{
681 pmd_t pmd;
682
683 if (!IS_ENABLED(CONFIG_NUMA_BALANCING))
684 return;
685
686 if (!has_transparent_hugepage())
687 return;
688
689 pr_debug("Validating PMD protnone\n");
690 pmd = pmd_mkhuge(pfn_pmd(args->fixed_pmd_pfn, args->page_prot_none));
691 WARN_ON(!pmd_protnone(pmd));
692 WARN_ON(!pmd_present(pmd));
693}
694#else /* !CONFIG_TRANSPARENT_HUGEPAGE */
695static void __init pmd_protnone_tests(struct pgtable_debug_args *args) { }
696#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
697
698#ifdef CONFIG_ARCH_HAS_PTE_DEVMAP
699static void __init pte_devmap_tests(struct pgtable_debug_args *args)
700{
701 pte_t pte = pfn_pte(args->fixed_pte_pfn, args->page_prot);
702
703 pr_debug("Validating PTE devmap\n");
704 WARN_ON(!pte_devmap(pte_mkdevmap(pte)));
705}
706
707#ifdef CONFIG_TRANSPARENT_HUGEPAGE
708static void __init pmd_devmap_tests(struct pgtable_debug_args *args)
709{
710 pmd_t pmd;
711
712 if (!has_transparent_hugepage())
713 return;
714
715 pr_debug("Validating PMD devmap\n");
716 pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot);
717 WARN_ON(!pmd_devmap(pmd_mkdevmap(pmd)));
718}
719
720#ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
721static void __init pud_devmap_tests(struct pgtable_debug_args *args)
722{
723 pud_t pud;
724
725 if (!has_transparent_hugepage())
726 return;
727
728 pr_debug("Validating PUD devmap\n");
729 pud = pfn_pud(args->fixed_pud_pfn, args->page_prot);
730 WARN_ON(!pud_devmap(pud_mkdevmap(pud)));
731}
732#else /* !CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
733static void __init pud_devmap_tests(struct pgtable_debug_args *args) { }
734#endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
735#else /* CONFIG_TRANSPARENT_HUGEPAGE */
736static void __init pmd_devmap_tests(struct pgtable_debug_args *args) { }
737static void __init pud_devmap_tests(struct pgtable_debug_args *args) { }
738#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
739#else
740static void __init pte_devmap_tests(struct pgtable_debug_args *args) { }
741static void __init pmd_devmap_tests(struct pgtable_debug_args *args) { }
742static void __init pud_devmap_tests(struct pgtable_debug_args *args) { }
743#endif /* CONFIG_ARCH_HAS_PTE_DEVMAP */
744
745static void __init pte_soft_dirty_tests(struct pgtable_debug_args *args)
746{
747 pte_t pte = pfn_pte(args->fixed_pte_pfn, args->page_prot);
748
749 if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY))
750 return;
751
752 pr_debug("Validating PTE soft dirty\n");
753 WARN_ON(!pte_soft_dirty(pte_mksoft_dirty(pte)));
754 WARN_ON(pte_soft_dirty(pte_clear_soft_dirty(pte)));
755}
756
757static void __init pte_swap_soft_dirty_tests(struct pgtable_debug_args *args)
758{
759 pte_t pte = pfn_pte(args->fixed_pte_pfn, args->page_prot);
760
761 if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY))
762 return;
763
764 pr_debug("Validating PTE swap soft dirty\n");
765 WARN_ON(!pte_swp_soft_dirty(pte_swp_mksoft_dirty(pte)));
766 WARN_ON(pte_swp_soft_dirty(pte_swp_clear_soft_dirty(pte)));
767}
768
769#ifdef CONFIG_TRANSPARENT_HUGEPAGE
770static void __init pmd_soft_dirty_tests(struct pgtable_debug_args *args)
771{
772 pmd_t pmd;
773
774 if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY))
775 return;
776
777 if (!has_transparent_hugepage())
778 return;
779
780 pr_debug("Validating PMD soft dirty\n");
781 pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot);
782 WARN_ON(!pmd_soft_dirty(pmd_mksoft_dirty(pmd)));
783 WARN_ON(pmd_soft_dirty(pmd_clear_soft_dirty(pmd)));
784}
785
786static void __init pmd_swap_soft_dirty_tests(struct pgtable_debug_args *args)
787{
788 pmd_t pmd;
789
790 if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) ||
791 !IS_ENABLED(CONFIG_ARCH_ENABLE_THP_MIGRATION))
792 return;
793
794 if (!has_transparent_hugepage())
795 return;
796
797 pr_debug("Validating PMD swap soft dirty\n");
798 pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot);
799 WARN_ON(!pmd_swp_soft_dirty(pmd_swp_mksoft_dirty(pmd)));
800 WARN_ON(pmd_swp_soft_dirty(pmd_swp_clear_soft_dirty(pmd)));
801}
802#else /* !CONFIG_TRANSPARENT_HUGEPAGE */
803static void __init pmd_soft_dirty_tests(struct pgtable_debug_args *args) { }
804static void __init pmd_swap_soft_dirty_tests(struct pgtable_debug_args *args) { }
805#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
806
807static void __init pte_swap_exclusive_tests(struct pgtable_debug_args *args)
808{
809#ifdef __HAVE_ARCH_PTE_SWP_EXCLUSIVE
810 pte_t pte = pfn_pte(args->fixed_pte_pfn, args->page_prot);
811
812 pr_debug("Validating PTE swap exclusive\n");
813 pte = pte_swp_mkexclusive(pte);
814 WARN_ON(!pte_swp_exclusive(pte));
815 pte = pte_swp_clear_exclusive(pte);
816 WARN_ON(pte_swp_exclusive(pte));
817#endif /* __HAVE_ARCH_PTE_SWP_EXCLUSIVE */
818}
819
820static void __init pte_swap_tests(struct pgtable_debug_args *args)
821{
822 swp_entry_t swp;
823 pte_t pte;
824
825 pr_debug("Validating PTE swap\n");
826 pte = pfn_pte(args->fixed_pte_pfn, args->page_prot);
827 swp = __pte_to_swp_entry(pte);
828 pte = __swp_entry_to_pte(swp);
829 WARN_ON(args->fixed_pte_pfn != pte_pfn(pte));
830}
831
832#ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
833static void __init pmd_swap_tests(struct pgtable_debug_args *args)
834{
835 swp_entry_t swp;
836 pmd_t pmd;
837
838 if (!has_transparent_hugepage())
839 return;
840
841 pr_debug("Validating PMD swap\n");
842 pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot);
843 swp = __pmd_to_swp_entry(pmd);
844 pmd = __swp_entry_to_pmd(swp);
845 WARN_ON(args->fixed_pmd_pfn != pmd_pfn(pmd));
846}
847#else /* !CONFIG_ARCH_ENABLE_THP_MIGRATION */
848static void __init pmd_swap_tests(struct pgtable_debug_args *args) { }
849#endif /* CONFIG_ARCH_ENABLE_THP_MIGRATION */
850
851static void __init swap_migration_tests(struct pgtable_debug_args *args)
852{
853 struct page *page;
854 swp_entry_t swp;
855
856 if (!IS_ENABLED(CONFIG_MIGRATION))
857 return;
858
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 use the allocated page explicitly for this
864 * purpose.
865 */
866 page = (args->pte_pfn != ULONG_MAX) ? pfn_to_page(args->pte_pfn) : NULL;
867 if (!page)
868 return;
869
870 pr_debug("Validating swap migration\n");
871
872 /*
873 * make_[readable|writable]_migration_entry() expects given page to
874 * be 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}
890
891#ifdef CONFIG_HUGETLB_PAGE
892static void __init hugetlb_basic_tests(struct pgtable_debug_args *args)
893{
894 struct page *page;
895 pte_t pte;
896
897 pr_debug("Validating HugeTLB basic\n");
898 /*
899 * Accessing the page associated with the pfn is safe here,
900 * as it was previously derived from a real kernel symbol.
901 */
902 page = pfn_to_page(args->fixed_pmd_pfn);
903 pte = mk_huge_pte(page, args->page_prot);
904
905 WARN_ON(!huge_pte_dirty(huge_pte_mkdirty(pte)));
906 WARN_ON(!huge_pte_write(huge_pte_mkwrite(huge_pte_wrprotect(pte))));
907 WARN_ON(huge_pte_write(huge_pte_wrprotect(huge_pte_mkwrite(pte))));
908
909#ifdef CONFIG_ARCH_WANT_GENERAL_HUGETLB
910 pte = pfn_pte(args->fixed_pmd_pfn, args->page_prot);
911
912 WARN_ON(!pte_huge(pte_mkhuge(pte)));
913#endif /* CONFIG_ARCH_WANT_GENERAL_HUGETLB */
914}
915#else /* !CONFIG_HUGETLB_PAGE */
916static void __init hugetlb_basic_tests(struct pgtable_debug_args *args) { }
917#endif /* CONFIG_HUGETLB_PAGE */
918
919#ifdef CONFIG_TRANSPARENT_HUGEPAGE
920static void __init pmd_thp_tests(struct pgtable_debug_args *args)
921{
922 pmd_t pmd;
923
924 if (!has_transparent_hugepage())
925 return;
926
927 pr_debug("Validating PMD based THP\n");
928 /*
929 * pmd_trans_huge() and pmd_present() must return positive after
930 * MMU invalidation with pmd_mkinvalid(). This behavior is an
931 * optimization for transparent huge page. pmd_trans_huge() must
932 * be true if pmd_page() returns a valid THP to avoid taking the
933 * pmd_lock when others walk over non transhuge pmds (i.e. there
934 * are no THP allocated). Especially when splitting a THP and
935 * removing the present bit from the pmd, pmd_trans_huge() still
936 * needs to return true. pmd_present() should be true whenever
937 * pmd_trans_huge() returns true.
938 */
939 pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot);
940 WARN_ON(!pmd_trans_huge(pmd_mkhuge(pmd)));
941
942#ifndef __HAVE_ARCH_PMDP_INVALIDATE
943 WARN_ON(!pmd_trans_huge(pmd_mkinvalid(pmd_mkhuge(pmd))));
944 WARN_ON(!pmd_present(pmd_mkinvalid(pmd_mkhuge(pmd))));
945#endif /* __HAVE_ARCH_PMDP_INVALIDATE */
946}
947
948#ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
949static void __init pud_thp_tests(struct pgtable_debug_args *args)
950{
951 pud_t pud;
952
953 if (!has_transparent_hugepage())
954 return;
955
956 pr_debug("Validating PUD based THP\n");
957 pud = pfn_pud(args->fixed_pud_pfn, args->page_prot);
958 WARN_ON(!pud_trans_huge(pud_mkhuge(pud)));
959
960 /*
961 * pud_mkinvalid() has been dropped for now. Enable back
962 * these tests when it comes back with a modified pud_present().
963 *
964 * WARN_ON(!pud_trans_huge(pud_mkinvalid(pud_mkhuge(pud))));
965 * WARN_ON(!pud_present(pud_mkinvalid(pud_mkhuge(pud))));
966 */
967}
968#else /* !CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
969static void __init pud_thp_tests(struct pgtable_debug_args *args) { }
970#endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
971#else /* !CONFIG_TRANSPARENT_HUGEPAGE */
972static void __init pmd_thp_tests(struct pgtable_debug_args *args) { }
973static void __init pud_thp_tests(struct pgtable_debug_args *args) { }
974#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
975
976static unsigned long __init get_random_vaddr(void)
977{
978 unsigned long random_vaddr, random_pages, total_user_pages;
979
980 total_user_pages = (TASK_SIZE - FIRST_USER_ADDRESS) / PAGE_SIZE;
981
982 random_pages = get_random_long() % total_user_pages;
983 random_vaddr = FIRST_USER_ADDRESS + random_pages * PAGE_SIZE;
984
985 return random_vaddr;
986}
987
988static void __init destroy_args(struct pgtable_debug_args *args)
989{
990 struct page *page = NULL;
991
992 /* Free (huge) page */
993 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
994 IS_ENABLED(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD) &&
995 has_transparent_hugepage() &&
996 args->pud_pfn != ULONG_MAX) {
997 if (args->is_contiguous_page) {
998 free_contig_range(args->pud_pfn,
999 (1 << (HPAGE_PUD_SHIFT - PAGE_SHIFT)));
1000 } else {
1001 page = pfn_to_page(args->pud_pfn);
1002 __free_pages(page, HPAGE_PUD_SHIFT - PAGE_SHIFT);
1003 }
1004
1005 args->pud_pfn = ULONG_MAX;
1006 args->pmd_pfn = ULONG_MAX;
1007 args->pte_pfn = ULONG_MAX;
1008 }
1009
1010 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
1011 has_transparent_hugepage() &&
1012 args->pmd_pfn != ULONG_MAX) {
1013 if (args->is_contiguous_page) {
1014 free_contig_range(args->pmd_pfn, (1 << HPAGE_PMD_ORDER));
1015 } else {
1016 page = pfn_to_page(args->pmd_pfn);
1017 __free_pages(page, HPAGE_PMD_ORDER);
1018 }
1019
1020 args->pmd_pfn = ULONG_MAX;
1021 args->pte_pfn = ULONG_MAX;
1022 }
1023
1024 if (args->pte_pfn != ULONG_MAX) {
1025 page = pfn_to_page(args->pte_pfn);
1026 __free_pages(page, 0);
1027
1028 args->pte_pfn = ULONG_MAX;
1029 }
1030
1031 /* Free page table entries */
1032 if (args->start_ptep) {
1033 pte_free(args->mm, args->start_ptep);
1034 mm_dec_nr_ptes(args->mm);
1035 }
1036
1037 if (args->start_pmdp) {
1038 pmd_free(args->mm, args->start_pmdp);
1039 mm_dec_nr_pmds(args->mm);
1040 }
1041
1042 if (args->start_pudp) {
1043 pud_free(args->mm, args->start_pudp);
1044 mm_dec_nr_puds(args->mm);
1045 }
1046
1047 if (args->start_p4dp)
1048 p4d_free(args->mm, args->start_p4dp);
1049
1050 /* Free vma and mm struct */
1051 if (args->vma)
1052 vm_area_free(args->vma);
1053
1054 if (args->mm)
1055 mmdrop(args->mm);
1056}
1057
1058static struct page * __init
1059debug_vm_pgtable_alloc_huge_page(struct pgtable_debug_args *args, int order)
1060{
1061 struct page *page = NULL;
1062
1063#ifdef CONFIG_CONTIG_ALLOC
1064 if (order >= MAX_ORDER) {
1065 page = alloc_contig_pages((1 << order), GFP_KERNEL,
1066 first_online_node, NULL);
1067 if (page) {
1068 args->is_contiguous_page = true;
1069 return page;
1070 }
1071 }
1072#endif
1073
1074 if (order < MAX_ORDER)
1075 page = alloc_pages(GFP_KERNEL, order);
1076
1077 return page;
1078}
1079
1080static int __init init_args(struct pgtable_debug_args *args)
1081{
1082 struct page *page = NULL;
1083 phys_addr_t phys;
1084 int ret = 0;
1085
1086 /*
1087 * Initialize the debugging data.
1088 *
1089 * vm_get_page_prot(VM_NONE) or vm_get_page_prot(VM_SHARED|VM_NONE)
1090 * will help create page table entries with PROT_NONE permission as
1091 * 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(VM_ACCESS_FLAGS);
1096 args->page_prot_none = vm_get_page_prot(VM_NONE);
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
1214error:
1215 destroy_args(args);
1216 return ret;
1217}
1218
1219static int __init debug_vm_pgtable(void)
1220{
1221 struct pgtable_debug_args args;
1222 spinlock_t *ptl = NULL;
1223 int idx, ret;
1224
1225 pr_info("Validating architecture page table helpers\n");
1226 ret = init_args(&args);
1227 if (ret)
1228 return ret;
1229
1230 /*
1231 * Iterate over each possible vm_flags to make sure that all
1232 * the basic page table transformation validations just hold
1233 * true irrespective of the starting protection value for a
1234 * given page table entry.
1235 *
1236 * Protection based vm_flags combinatins are always linear
1237 * and increasing i.e starting from VM_NONE and going upto
1238 * (VM_SHARED | READ | WRITE | EXEC).
1239 */
1240#define VM_FLAGS_START (VM_NONE)
1241#define VM_FLAGS_END (VM_SHARED | VM_EXEC | VM_WRITE | VM_READ)
1242
1243 for (idx = VM_FLAGS_START; idx <= VM_FLAGS_END; idx++) {
1244 pte_basic_tests(&args, idx);
1245 pmd_basic_tests(&args, idx);
1246 pud_basic_tests(&args, idx);
1247 }
1248
1249 /*
1250 * Both P4D and PGD level tests are very basic which do not
1251 * involve creating page table entries from the protection
1252 * value and the given pfn. Hence just keep them out from
1253 * the above iteration for now to save some test execution
1254 * time.
1255 */
1256 p4d_basic_tests(&args);
1257 pgd_basic_tests(&args);
1258
1259 pmd_leaf_tests(&args);
1260 pud_leaf_tests(&args);
1261
1262 pte_special_tests(&args);
1263 pte_protnone_tests(&args);
1264 pmd_protnone_tests(&args);
1265
1266 pte_devmap_tests(&args);
1267 pmd_devmap_tests(&args);
1268 pud_devmap_tests(&args);
1269
1270 pte_soft_dirty_tests(&args);
1271 pmd_soft_dirty_tests(&args);
1272 pte_swap_soft_dirty_tests(&args);
1273 pmd_swap_soft_dirty_tests(&args);
1274
1275 pte_swap_exclusive_tests(&args);
1276
1277 pte_swap_tests(&args);
1278 pmd_swap_tests(&args);
1279
1280 swap_migration_tests(&args);
1281
1282 pmd_thp_tests(&args);
1283 pud_thp_tests(&args);
1284
1285 hugetlb_basic_tests(&args);
1286
1287 /*
1288 * Page table modifying tests. They need to hold
1289 * proper page table lock.
1290 */
1291
1292 args.ptep = pte_offset_map_lock(args.mm, args.pmdp, args.vaddr, &ptl);
1293 pte_clear_tests(&args);
1294 pte_advanced_tests(&args);
1295 pte_unmap_unlock(args.ptep, ptl);
1296
1297 ptl = pmd_lock(args.mm, args.pmdp);
1298 pmd_clear_tests(&args);
1299 pmd_advanced_tests(&args);
1300 pmd_huge_tests(&args);
1301 pmd_populate_tests(&args);
1302 spin_unlock(ptl);
1303
1304 ptl = pud_lock(args.mm, args.pudp);
1305 pud_clear_tests(&args);
1306 pud_advanced_tests(&args);
1307 pud_huge_tests(&args);
1308 pud_populate_tests(&args);
1309 spin_unlock(ptl);
1310
1311 spin_lock(&(args.mm->page_table_lock));
1312 p4d_clear_tests(&args);
1313 pgd_clear_tests(&args);
1314 p4d_populate_tests(&args);
1315 pgd_populate_tests(&args);
1316 spin_unlock(&(args.mm->page_table_lock));
1317
1318 destroy_args(&args);
1319 return 0;
1320}
1321late_initcall(debug_vm_pgtable);