1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * This kernel test validates architecture page table helpers and
4 * accessors and helps in verifying their continued compliance with
5 * expected generic MM semantics.
6 *
7 * Copyright (C) 2019 ARM Ltd.
8 *
9 * Author: Anshuman Khandual <anshuman.khandual@arm.com>
10 */
11#define pr_fmt(fmt) "debug_vm_pgtable: [%-25s]: " fmt, __func__
12
13#include <linux/gfp.h>
14#include <linux/highmem.h>
15#include <linux/hugetlb.h>
16#include <linux/kernel.h>
17#include <linux/kconfig.h>
18#include <linux/memblock.h>
19#include <linux/mm.h>
20#include <linux/mman.h>
21#include <linux/mm_types.h>
22#include <linux/module.h>
23#include <linux/pfn_t.h>
24#include <linux/printk.h>
25#include <linux/pgtable.h>
26#include <linux/random.h>
27#include <linux/spinlock.h>
28#include <linux/swap.h>
29#include <linux/swapops.h>
30#include <linux/start_kernel.h>
31#include <linux/sched/mm.h>
32#include <linux/io.h>
33
34#include <asm/cacheflush.h>
35#include <asm/pgalloc.h>
36#include <asm/tlbflush.h>
37
38/*
39 * Please refer Documentation/mm/arch_pgtable_helpers.rst for the semantics
40 * expectations that are being validated here. All future changes in here
41 * or the documentation need to be in sync.
42 *
43 * On s390 platform, the lower 4 bits are used to identify given page table
44 * entry type. But these bits might affect the ability to clear entries with
45 * pxx_clear() because of how dynamic page table folding works on s390. So
46 * while loading up the entries do not change the lower 4 bits. It does not
47 * have affect any other platform. Also avoid the 62nd bit on ppc64 that is
48 * used to mark a pte entry.
49 */
50#define S390_SKIP_MASK GENMASK(3, 0)
51#if __BITS_PER_LONG == 64
52#define PPC64_SKIP_MASK GENMASK(62, 62)
53#else
54#define PPC64_SKIP_MASK 0x0
55#endif
56#define ARCH_SKIP_MASK (S390_SKIP_MASK | PPC64_SKIP_MASK)
57#define RANDOM_ORVALUE (GENMASK(BITS_PER_LONG - 1, 0) & ~ARCH_SKIP_MASK)
58#define RANDOM_NZVALUE GENMASK(7, 0)
59
60struct pgtable_debug_args {
61 struct mm_struct *mm;
62 struct vm_area_struct *vma;
63
64 pgd_t *pgdp;
65 p4d_t *p4dp;
66 pud_t *pudp;
67 pmd_t *pmdp;
68 pte_t *ptep;
69
70 p4d_t *start_p4dp;
71 pud_t *start_pudp;
72 pmd_t *start_pmdp;
73 pgtable_t start_ptep;
74
75 unsigned long vaddr;
76 pgprot_t page_prot;
77 pgprot_t page_prot_none;
78
79 bool is_contiguous_page;
80 unsigned long pud_pfn;
81 unsigned long pmd_pfn;
82 unsigned long pte_pfn;
83
84 unsigned long fixed_alignment;
85 unsigned long fixed_pgd_pfn;
86 unsigned long fixed_p4d_pfn;
87 unsigned long fixed_pud_pfn;
88 unsigned long fixed_pmd_pfn;
89 unsigned long fixed_pte_pfn;
90};
91
92static void __init pte_basic_tests(struct pgtable_debug_args *args, int idx)
93{
94 pgprot_t prot = vm_get_page_prot(vm_flags: idx);
95 pte_t pte = pfn_pte(page_nr: args->fixed_pte_pfn, pgprot: prot);
96 unsigned long val = idx, *ptr = &val;
97
98 pr_debug("Validating PTE basic (%pGv)\n", ptr);
99
100 /*
101 * This test needs to be executed after the given page table entry
102 * is created with pfn_pte() to make sure that vm_get_page_prot(idx)
103 * does not have the dirty bit enabled from the beginning. This is
104 * important for platforms like arm64 where (!PTE_RDONLY) indicate
105 * dirty bit being set.
106 */
107 WARN_ON(pte_dirty(pte_wrprotect(pte)));
108
109 WARN_ON(!pte_same(pte, pte));
110 WARN_ON(!pte_young(pte_mkyoung(pte_mkold(pte))));
111 WARN_ON(!pte_dirty(pte_mkdirty(pte_mkclean(pte))));
112 WARN_ON(!pte_write(pte_mkwrite(pte_wrprotect(pte), args->vma)));
113 WARN_ON(pte_young(pte_mkold(pte_mkyoung(pte))));
114 WARN_ON(pte_dirty(pte_mkclean(pte_mkdirty(pte))));
115 WARN_ON(pte_write(pte_wrprotect(pte_mkwrite(pte, args->vma))));
116 WARN_ON(pte_dirty(pte_wrprotect(pte_mkclean(pte))));
117 WARN_ON(!pte_dirty(pte_wrprotect(pte_mkdirty(pte))));
118}
119
120static void __init pte_advanced_tests(struct pgtable_debug_args *args)
121{
122 struct page *page;
123 pte_t pte;
124
125 /*
126 * Architectures optimize set_pte_at by avoiding TLB flush.
127 * This requires set_pte_at to be not used to update an
128 * existing pte entry. Clear pte before we do set_pte_at
129 *
130 * flush_dcache_page() is called after set_pte_at() to clear
131 * PG_arch_1 for the page on ARM64. The page flag isn't cleared
132 * when it's released and page allocation check will fail when
133 * the page is allocated again. For architectures other than ARM64,
134 * the unexpected overhead of cache flushing is acceptable.
135 */
136 page = (args->pte_pfn != ULONG_MAX) ? pfn_to_page(args->pte_pfn) : NULL;
137 if (!page)
138 return;
139
140 pr_debug("Validating PTE advanced\n");
141 if (WARN_ON(!args->ptep))
142 return;
143
144 pte = pfn_pte(page_nr: args->pte_pfn, pgprot: args->page_prot);
145 set_pte_at(args->mm, args->vaddr, args->ptep, pte);
146 flush_dcache_page(page);
147 ptep_set_wrprotect(mm: args->mm, addr: args->vaddr, ptep: args->ptep);
148 pte = ptep_get(ptep: args->ptep);
149 WARN_ON(pte_write(pte));
150 ptep_get_and_clear(mm: args->mm, addr: args->vaddr, ptep: args->ptep);
151 pte = ptep_get(ptep: args->ptep);
152 WARN_ON(!pte_none(pte));
153
154 pte = pfn_pte(page_nr: args->pte_pfn, pgprot: args->page_prot);
155 pte = pte_wrprotect(pte);
156 pte = pte_mkclean(pte);
157 set_pte_at(args->mm, args->vaddr, args->ptep, pte);
158 flush_dcache_page(page);
159 pte = pte_mkwrite(pte, vma: args->vma);
160 pte = pte_mkdirty(pte);
161 ptep_set_access_flags(vma: args->vma, address: args->vaddr, ptep: args->ptep, entry: pte, dirty: 1);
162 pte = ptep_get(ptep: args->ptep);
163 WARN_ON(!(pte_write(pte) && pte_dirty(pte)));
164 ptep_get_and_clear_full(mm: args->mm, addr: args->vaddr, ptep: args->ptep, full: 1);
165 pte = ptep_get(ptep: args->ptep);
166 WARN_ON(!pte_none(pte));
167
168 pte = pfn_pte(page_nr: args->pte_pfn, pgprot: args->page_prot);
169 pte = pte_mkyoung(pte);
170 set_pte_at(args->mm, args->vaddr, args->ptep, pte);
171 flush_dcache_page(page);
172 ptep_test_and_clear_young(vma: args->vma, addr: args->vaddr, ptep: args->ptep);
173 pte = ptep_get(ptep: args->ptep);
174 WARN_ON(pte_young(pte));
175
176 ptep_get_and_clear_full(mm: args->mm, addr: args->vaddr, ptep: args->ptep, full: 1);
177}
178
179#ifdef CONFIG_TRANSPARENT_HUGEPAGE
180static void __init pmd_basic_tests(struct pgtable_debug_args *args, int idx)
181{
182 pgprot_t prot = vm_get_page_prot(vm_flags: idx);
183 unsigned long val = idx, *ptr = &val;
184 pmd_t pmd;
185
186 if (!has_transparent_hugepage())
187 return;
188
189 pr_debug("Validating PMD basic (%pGv)\n", ptr);
190 pmd = pfn_pmd(page_nr: args->fixed_pmd_pfn, pgprot: prot);
191
192 /*
193 * This test needs to be executed after the given page table entry
194 * is created with pfn_pmd() to make sure that vm_get_page_prot(idx)
195 * does not have the dirty bit enabled from the beginning. This is
196 * important for platforms like arm64 where (!PTE_RDONLY) indicate
197 * dirty bit being set.
198 */
199 WARN_ON(pmd_dirty(pmd_wrprotect(pmd)));
200
201
202 WARN_ON(!pmd_same(pmd, pmd));
203 WARN_ON(!pmd_young(pmd_mkyoung(pmd_mkold(pmd))));
204 WARN_ON(!pmd_dirty(pmd_mkdirty(pmd_mkclean(pmd))));
205 WARN_ON(!pmd_write(pmd_mkwrite(pmd_wrprotect(pmd), args->vma)));
206 WARN_ON(pmd_young(pmd_mkold(pmd_mkyoung(pmd))));
207 WARN_ON(pmd_dirty(pmd_mkclean(pmd_mkdirty(pmd))));
208 WARN_ON(pmd_write(pmd_wrprotect(pmd_mkwrite(pmd, args->vma))));
209 WARN_ON(pmd_dirty(pmd_wrprotect(pmd_mkclean(pmd))));
210 WARN_ON(!pmd_dirty(pmd_wrprotect(pmd_mkdirty(pmd))));
211 /*
212 * A huge page does not point to next level page table
213 * entry. Hence this must qualify as pmd_bad().
214 */
215 WARN_ON(!pmd_bad(pmd_mkhuge(pmd)));
216}
217
218static void __init pmd_advanced_tests(struct pgtable_debug_args *args)
219{
220 struct page *page;
221 pmd_t pmd;
222 unsigned long vaddr = args->vaddr;
223
224 if (!has_transparent_hugepage())
225 return;
226
227 page = (args->pmd_pfn != ULONG_MAX) ? pfn_to_page(args->pmd_pfn) : NULL;
228 if (!page)
229 return;
230
231 /*
232 * flush_dcache_page() is called after set_pmd_at() to clear
233 * PG_arch_1 for the page on ARM64. The page flag isn't cleared
234 * when it's released and page allocation check will fail when
235 * the page is allocated again. For architectures other than ARM64,
236 * the unexpected overhead of cache flushing is acceptable.
237 */
238 pr_debug("Validating PMD advanced\n");
239 /* Align the address wrt HPAGE_PMD_SIZE */
240 vaddr &= HPAGE_PMD_MASK;
241
242 pgtable_trans_huge_deposit(mm: args->mm, pmdp: args->pmdp, pgtable: args->start_ptep);
243
244 pmd = pfn_pmd(page_nr: args->pmd_pfn, pgprot: args->page_prot);
245 set_pmd_at(mm: args->mm, addr: vaddr, pmdp: args->pmdp, pmd);
246 flush_dcache_page(page);
247 pmdp_set_wrprotect(mm: args->mm, addr: vaddr, pmdp: args->pmdp);
248 pmd = READ_ONCE(*args->pmdp);
249 WARN_ON(pmd_write(pmd));
250 pmdp_huge_get_and_clear(mm: args->mm, addr: vaddr, pmdp: args->pmdp);
251 pmd = READ_ONCE(*args->pmdp);
252 WARN_ON(!pmd_none(pmd));
253
254 pmd = pfn_pmd(page_nr: args->pmd_pfn, pgprot: args->page_prot);
255 pmd = pmd_wrprotect(pmd);
256 pmd = pmd_mkclean(pmd);
257 set_pmd_at(mm: args->mm, addr: vaddr, pmdp: args->pmdp, pmd);
258 flush_dcache_page(page);
259 pmd = pmd_mkwrite(pmd, vma: args->vma);
260 pmd = pmd_mkdirty(pmd);
261 pmdp_set_access_flags(vma: args->vma, address: vaddr, pmdp: args->pmdp, entry: pmd, dirty: 1);
262 pmd = READ_ONCE(*args->pmdp);
263 WARN_ON(!(pmd_write(pmd) && pmd_dirty(pmd)));
264 pmdp_huge_get_and_clear_full(vma: args->vma, address: vaddr, pmdp: args->pmdp, full: 1);
265 pmd = READ_ONCE(*args->pmdp);
266 WARN_ON(!pmd_none(pmd));
267
268 pmd = pmd_mkhuge(pmd: pfn_pmd(page_nr: args->pmd_pfn, pgprot: args->page_prot));
269 pmd = pmd_mkyoung(pmd);
270 set_pmd_at(mm: args->mm, addr: vaddr, pmdp: args->pmdp, pmd);
271 flush_dcache_page(page);
272 pmdp_test_and_clear_young(vma: args->vma, addr: vaddr, pmdp: args->pmdp);
273 pmd = READ_ONCE(*args->pmdp);
274 WARN_ON(pmd_young(pmd));
275
276 /* Clear the pte entries */
277 pmdp_huge_get_and_clear(mm: args->mm, addr: vaddr, pmdp: args->pmdp);
278 pgtable_trans_huge_withdraw(mm: args->mm, pmdp: args->pmdp);
279}
280
281static void __init pmd_leaf_tests(struct pgtable_debug_args *args)
282{
283 pmd_t pmd;
284
285 if (!has_transparent_hugepage())
286 return;
287
288 pr_debug("Validating PMD leaf\n");
289 pmd = pfn_pmd(page_nr: args->fixed_pmd_pfn, pgprot: args->page_prot);
290
291 /*
292 * PMD based THP is a leaf entry.
293 */
294 pmd = pmd_mkhuge(pmd);
295 WARN_ON(!pmd_leaf(pmd));
296}
297
298#ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
299static void __init pud_basic_tests(struct pgtable_debug_args *args, int idx)
300{
301 pgprot_t prot = vm_get_page_prot(vm_flags: idx);
302 unsigned long val = idx, *ptr = &val;
303 pud_t pud;
304
305 if (!has_transparent_pud_hugepage())
306 return;
307
308 pr_debug("Validating PUD basic (%pGv)\n", ptr);
309 pud = pfn_pud(page_nr: args->fixed_pud_pfn, pgprot: prot);
310
311 /*
312 * This test needs to be executed after the given page table entry
313 * is created with pfn_pud() to make sure that vm_get_page_prot(idx)
314 * does not have the dirty bit enabled from the beginning. This is
315 * important for platforms like arm64 where (!PTE_RDONLY) indicate
316 * dirty bit being set.
317 */
318 WARN_ON(pud_dirty(pud_wrprotect(pud)));
319
320 WARN_ON(!pud_same(pud, pud));
321 WARN_ON(!pud_young(pud_mkyoung(pud_mkold(pud))));
322 WARN_ON(!pud_dirty(pud_mkdirty(pud_mkclean(pud))));
323 WARN_ON(pud_dirty(pud_mkclean(pud_mkdirty(pud))));
324 WARN_ON(!pud_write(pud_mkwrite(pud_wrprotect(pud))));
325 WARN_ON(pud_write(pud_wrprotect(pud_mkwrite(pud))));
326 WARN_ON(pud_young(pud_mkold(pud_mkyoung(pud))));
327 WARN_ON(pud_dirty(pud_wrprotect(pud_mkclean(pud))));
328 WARN_ON(!pud_dirty(pud_wrprotect(pud_mkdirty(pud))));
329
330 if (mm_pmd_folded(args->mm))
331 return;
332
333 /*
334 * A huge page does not point to next level page table
335 * entry. Hence this must qualify as pud_bad().
336 */
337 WARN_ON(!pud_bad(pud_mkhuge(pud)));
338}
339
340static void __init pud_advanced_tests(struct pgtable_debug_args *args)
341{
342 struct page *page;
343 unsigned long vaddr = args->vaddr;
344 pud_t pud;
345
346 if (!has_transparent_pud_hugepage())
347 return;
348
349 page = (args->pud_pfn != ULONG_MAX) ? pfn_to_page(args->pud_pfn) : NULL;
350 if (!page)
351 return;
352
353 /*
354 * flush_dcache_page() is called after set_pud_at() to clear
355 * PG_arch_1 for the page on ARM64. The page flag isn't cleared
356 * when it's released and page allocation check will fail when
357 * the page is allocated again. For architectures other than ARM64,
358 * the unexpected overhead of cache flushing is acceptable.
359 */
360 pr_debug("Validating PUD advanced\n");
361 /* Align the address wrt HPAGE_PUD_SIZE */
362 vaddr &= HPAGE_PUD_MASK;
363
364 pud = pfn_pud(page_nr: args->pud_pfn, pgprot: args->page_prot);
365 set_pud_at(mm: args->mm, addr: vaddr, pudp: args->pudp, pud);
366 flush_dcache_page(page);
367 pudp_set_wrprotect(mm: args->mm, address: vaddr, pudp: args->pudp);
368 pud = READ_ONCE(*args->pudp);
369 WARN_ON(pud_write(pud));
370
371#ifndef __PAGETABLE_PMD_FOLDED
372 pudp_huge_get_and_clear(mm: args->mm, addr: vaddr, pudp: args->pudp);
373 pud = READ_ONCE(*args->pudp);
374 WARN_ON(!pud_none(pud));
375#endif /* __PAGETABLE_PMD_FOLDED */
376 pud = pfn_pud(page_nr: args->pud_pfn, pgprot: args->page_prot);
377 pud = pud_wrprotect(pud);
378 pud = pud_mkclean(pud);
379 set_pud_at(mm: args->mm, addr: vaddr, pudp: args->pudp, pud);
380 flush_dcache_page(page);
381 pud = pud_mkwrite(pud);
382 pud = pud_mkdirty(pud);
383 pudp_set_access_flags(vma: args->vma, address: vaddr, pudp: args->pudp, entry: pud, dirty: 1);
384 pud = READ_ONCE(*args->pudp);
385 WARN_ON(!(pud_write(pud) && pud_dirty(pud)));
386
387#ifndef __PAGETABLE_PMD_FOLDED
388 pudp_huge_get_and_clear_full(vma: args->vma, address: vaddr, pudp: args->pudp, full: 1);
389 pud = READ_ONCE(*args->pudp);
390 WARN_ON(!pud_none(pud));
391#endif /* __PAGETABLE_PMD_FOLDED */
392
393 pud = pfn_pud(page_nr: args->pud_pfn, pgprot: args->page_prot);
394 pud = pud_mkyoung(pud);
395 set_pud_at(mm: args->mm, addr: vaddr, pudp: args->pudp, pud);
396 flush_dcache_page(page);
397 pudp_test_and_clear_young(vma: args->vma, addr: vaddr, pudp: args->pudp);
398 pud = READ_ONCE(*args->pudp);
399 WARN_ON(pud_young(pud));
400
401 pudp_huge_get_and_clear(mm: args->mm, addr: vaddr, pudp: args->pudp);
402}
403
404static void __init pud_leaf_tests(struct pgtable_debug_args *args)
405{
406 pud_t pud;
407
408 if (!has_transparent_pud_hugepage())
409 return;
410
411 pr_debug("Validating PUD leaf\n");
412 pud = pfn_pud(page_nr: args->fixed_pud_pfn, pgprot: args->page_prot);
413 /*
414 * PUD based THP is a leaf entry.
415 */
416 pud = pud_mkhuge(pud);
417 WARN_ON(!pud_leaf(pud));
418}
419#else /* !CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
420static void __init pud_basic_tests(struct pgtable_debug_args *args, int idx) { }
421static void __init pud_advanced_tests(struct pgtable_debug_args *args) { }
422static void __init pud_leaf_tests(struct pgtable_debug_args *args) { }
423#endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
424#else /* !CONFIG_TRANSPARENT_HUGEPAGE */
425static void __init pmd_basic_tests(struct pgtable_debug_args *args, int idx) { }
426static void __init pud_basic_tests(struct pgtable_debug_args *args, int idx) { }
427static void __init pmd_advanced_tests(struct pgtable_debug_args *args) { }
428static void __init pud_advanced_tests(struct pgtable_debug_args *args) { }
429static void __init pmd_leaf_tests(struct pgtable_debug_args *args) { }
430static void __init pud_leaf_tests(struct pgtable_debug_args *args) { }
431#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
432
433#ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
434static void __init pmd_huge_tests(struct pgtable_debug_args *args)
435{
436 pmd_t pmd;
437
438 if (!arch_vmap_pmd_supported(prot: args->page_prot) ||
439 args->fixed_alignment < PMD_SIZE)
440 return;
441
442 pr_debug("Validating PMD huge\n");
443 /*
444 * X86 defined pmd_set_huge() verifies that the given
445 * PMD is not a populated non-leaf entry.
446 */
447 WRITE_ONCE(*args->pmdp, __pmd(0));
448 WARN_ON(!pmd_set_huge(args->pmdp, __pfn_to_phys(args->fixed_pmd_pfn), args->page_prot));
449 WARN_ON(!pmd_clear_huge(args->pmdp));
450 pmd = READ_ONCE(*args->pmdp);
451 WARN_ON(!pmd_none(pmd));
452}
453
454static void __init pud_huge_tests(struct pgtable_debug_args *args)
455{
456 pud_t pud;
457
458 if (!arch_vmap_pud_supported(prot: args->page_prot) ||
459 args->fixed_alignment < PUD_SIZE)
460 return;
461
462 pr_debug("Validating PUD huge\n");
463 /*
464 * X86 defined pud_set_huge() verifies that the given
465 * PUD is not a populated non-leaf entry.
466 */
467 WRITE_ONCE(*args->pudp, __pud(0));
468 WARN_ON(!pud_set_huge(args->pudp, __pfn_to_phys(args->fixed_pud_pfn), args->page_prot));
469 WARN_ON(!pud_clear_huge(args->pudp));
470 pud = READ_ONCE(*args->pudp);
471 WARN_ON(!pud_none(pud));
472}
473#else /* !CONFIG_HAVE_ARCH_HUGE_VMAP */
474static void __init pmd_huge_tests(struct pgtable_debug_args *args) { }
475static void __init pud_huge_tests(struct pgtable_debug_args *args) { }
476#endif /* CONFIG_HAVE_ARCH_HUGE_VMAP */
477
478static void __init p4d_basic_tests(struct pgtable_debug_args *args)
479{
480 p4d_t p4d;
481
482 pr_debug("Validating P4D basic\n");
483 memset(&p4d, RANDOM_NZVALUE, sizeof(p4d_t));
484 WARN_ON(!p4d_same(p4d, p4d));
485}
486
487static void __init pgd_basic_tests(struct pgtable_debug_args *args)
488{
489 pgd_t pgd;
490
491 pr_debug("Validating PGD basic\n");
492 memset(&pgd, RANDOM_NZVALUE, sizeof(pgd_t));
493 WARN_ON(!pgd_same(pgd, pgd));
494}
495
496#ifndef __PAGETABLE_PUD_FOLDED
497static void __init pud_clear_tests(struct pgtable_debug_args *args)
498{
499 pud_t pud = READ_ONCE(*args->pudp);
500
501 if (mm_pmd_folded(args->mm))
502 return;
503
504 pr_debug("Validating PUD clear\n");
505 pud = __pud(val: pud_val(pud) | RANDOM_ORVALUE);
506 WRITE_ONCE(*args->pudp, pud);
507 pud_clear(pudp: args->pudp);
508 pud = READ_ONCE(*args->pudp);
509 WARN_ON(!pud_none(pud));
510}
511
512static void __init pud_populate_tests(struct pgtable_debug_args *args)
513{
514 pud_t pud;
515
516 if (mm_pmd_folded(args->mm))
517 return;
518
519 pr_debug("Validating PUD populate\n");
520 /*
521 * This entry points to next level page table page.
522 * Hence this must not qualify as pud_bad().
523 */
524 pud_populate(mm: args->mm, pud: args->pudp, pmd: args->start_pmdp);
525 pud = READ_ONCE(*args->pudp);
526 WARN_ON(pud_bad(pud));
527}
528#else /* !__PAGETABLE_PUD_FOLDED */
529static void __init pud_clear_tests(struct pgtable_debug_args *args) { }
530static void __init pud_populate_tests(struct pgtable_debug_args *args) { }
531#endif /* PAGETABLE_PUD_FOLDED */
532
533#ifndef __PAGETABLE_P4D_FOLDED
534static void __init p4d_clear_tests(struct pgtable_debug_args *args)
535{
536 p4d_t p4d = READ_ONCE(*args->p4dp);
537
538 if (mm_pud_folded(args->mm))
539 return;
540
541 pr_debug("Validating P4D clear\n");
542 p4d = __p4d(val: p4d_val(p4d) | RANDOM_ORVALUE);
543 WRITE_ONCE(*args->p4dp, p4d);
544 p4d_clear(p4dp: args->p4dp);
545 p4d = READ_ONCE(*args->p4dp);
546 WARN_ON(!p4d_none(p4d));
547}
548
549static void __init p4d_populate_tests(struct pgtable_debug_args *args)
550{
551 p4d_t p4d;
552
553 if (mm_pud_folded(args->mm))
554 return;
555
556 pr_debug("Validating P4D populate\n");
557 /*
558 * This entry points to next level page table page.
559 * Hence this must not qualify as p4d_bad().
560 */
561 pud_clear(pudp: args->pudp);
562 p4d_clear(p4dp: args->p4dp);
563 p4d_populate(mm: args->mm, p4d: args->p4dp, pud: args->start_pudp);
564 p4d = READ_ONCE(*args->p4dp);
565 WARN_ON(p4d_bad(p4d));
566}
567
568static void __init pgd_clear_tests(struct pgtable_debug_args *args)
569{
570 pgd_t pgd = READ_ONCE(*(args->pgdp));
571
572 if (mm_p4d_folded(mm: args->mm))
573 return;
574
575 pr_debug("Validating PGD clear\n");
576 pgd = __pgd(val: pgd_val(pgd) | RANDOM_ORVALUE);
577 WRITE_ONCE(*args->pgdp, pgd);
578 pgd_clear(args->pgdp);
579 pgd = READ_ONCE(*args->pgdp);
580 WARN_ON(!pgd_none(pgd));
581}
582
583static void __init pgd_populate_tests(struct pgtable_debug_args *args)
584{
585 pgd_t pgd;
586
587 if (mm_p4d_folded(mm: args->mm))
588 return;
589
590 pr_debug("Validating PGD populate\n");
591 /*
592 * This entry points to next level page table page.
593 * Hence this must not qualify as pgd_bad().
594 */
595 p4d_clear(p4dp: args->p4dp);
596 pgd_clear(args->pgdp);
597 pgd_populate(mm: args->mm, pgd: args->pgdp, p4d: args->start_p4dp);
598 pgd = READ_ONCE(*args->pgdp);
599 WARN_ON(pgd_bad(pgd));
600}
601#else /* !__PAGETABLE_P4D_FOLDED */
602static void __init p4d_clear_tests(struct pgtable_debug_args *args) { }
603static void __init pgd_clear_tests(struct pgtable_debug_args *args) { }
604static void __init p4d_populate_tests(struct pgtable_debug_args *args) { }
605static void __init pgd_populate_tests(struct pgtable_debug_args *args) { }
606#endif /* PAGETABLE_P4D_FOLDED */
607
608static void __init pte_clear_tests(struct pgtable_debug_args *args)
609{
610 struct page *page;
611 pte_t pte = pfn_pte(page_nr: args->pte_pfn, pgprot: args->page_prot);
612
613 page = (args->pte_pfn != ULONG_MAX) ? pfn_to_page(args->pte_pfn) : NULL;
614 if (!page)
615 return;
616
617 /*
618 * flush_dcache_page() is called after set_pte_at() to clear
619 * PG_arch_1 for the page on ARM64. The page flag isn't cleared
620 * when it's released and page allocation check will fail when
621 * the page is allocated again. For architectures other than ARM64,
622 * the unexpected overhead of cache flushing is acceptable.
623 */
624 pr_debug("Validating PTE clear\n");
625 if (WARN_ON(!args->ptep))
626 return;
627
628#ifndef CONFIG_RISCV
629 pte = __pte(val: pte_val(pte) | RANDOM_ORVALUE);
630#endif
631 set_pte_at(args->mm, args->vaddr, args->ptep, pte);
632 flush_dcache_page(page);
633 barrier();
634 ptep_clear(mm: args->mm, addr: args->vaddr, ptep: args->ptep);
635 pte = ptep_get(ptep: args->ptep);
636 WARN_ON(!pte_none(pte));
637}
638
639static void __init pmd_clear_tests(struct pgtable_debug_args *args)
640{
641 pmd_t pmd = READ_ONCE(*args->pmdp);
642
643 pr_debug("Validating PMD clear\n");
644 pmd = __pmd(val: pmd_val(pmd) | RANDOM_ORVALUE);
645 WRITE_ONCE(*args->pmdp, pmd);
646 pmd_clear(pmdp: args->pmdp);
647 pmd = READ_ONCE(*args->pmdp);
648 WARN_ON(!pmd_none(pmd));
649}
650
651static void __init pmd_populate_tests(struct pgtable_debug_args *args)
652{
653 pmd_t pmd;
654
655 pr_debug("Validating PMD populate\n");
656 /*
657 * This entry points to next level page table page.
658 * Hence this must not qualify as pmd_bad().
659 */
660 pmd_populate(mm: args->mm, pmd: args->pmdp, pte: args->start_ptep);
661 pmd = READ_ONCE(*args->pmdp);
662 WARN_ON(pmd_bad(pmd));
663}
664
665static void __init pte_special_tests(struct pgtable_debug_args *args)
666{
667 pte_t pte = pfn_pte(page_nr: args->fixed_pte_pfn, pgprot: args->page_prot);
668
669 if (!IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL))
670 return;
671
672 pr_debug("Validating PTE special\n");
673 WARN_ON(!pte_special(pte_mkspecial(pte)));
674}
675
676static void __init pte_protnone_tests(struct pgtable_debug_args *args)
677{
678 pte_t pte = pfn_pte(page_nr: args->fixed_pte_pfn, pgprot: args->page_prot_none);
679
680 if (!IS_ENABLED(CONFIG_NUMA_BALANCING))
681 return;
682
683 pr_debug("Validating PTE protnone\n");
684 WARN_ON(!pte_protnone(pte));
685 WARN_ON(!pte_present(pte));
686}
687
688#ifdef CONFIG_TRANSPARENT_HUGEPAGE
689static void __init pmd_protnone_tests(struct pgtable_debug_args *args)
690{
691 pmd_t pmd;
692
693 if (!IS_ENABLED(CONFIG_NUMA_BALANCING))
694 return;
695
696 if (!has_transparent_hugepage())
697 return;
698
699 pr_debug("Validating PMD protnone\n");
700 pmd = pmd_mkhuge(pmd: pfn_pmd(page_nr: args->fixed_pmd_pfn, pgprot: args->page_prot_none));
701 WARN_ON(!pmd_protnone(pmd));
702 WARN_ON(!pmd_present(pmd));
703}
704#else /* !CONFIG_TRANSPARENT_HUGEPAGE */
705static void __init pmd_protnone_tests(struct pgtable_debug_args *args) { }
706#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
707
708#ifdef CONFIG_ARCH_HAS_PTE_DEVMAP
709static void __init pte_devmap_tests(struct pgtable_debug_args *args)
710{
711 pte_t pte = pfn_pte(page_nr: args->fixed_pte_pfn, pgprot: args->page_prot);
712
713 pr_debug("Validating PTE devmap\n");
714 WARN_ON(!pte_devmap(pte_mkdevmap(pte)));
715}
716
717#ifdef CONFIG_TRANSPARENT_HUGEPAGE
718static void __init pmd_devmap_tests(struct pgtable_debug_args *args)
719{
720 pmd_t pmd;
721
722 if (!has_transparent_hugepage())
723 return;
724
725 pr_debug("Validating PMD devmap\n");
726 pmd = pfn_pmd(page_nr: args->fixed_pmd_pfn, pgprot: args->page_prot);
727 WARN_ON(!pmd_devmap(pmd_mkdevmap(pmd)));
728}
729
730#ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
731static void __init pud_devmap_tests(struct pgtable_debug_args *args)
732{
733 pud_t pud;
734
735 if (!has_transparent_pud_hugepage())
736 return;
737
738 pr_debug("Validating PUD devmap\n");
739 pud = pfn_pud(page_nr: args->fixed_pud_pfn, pgprot: args->page_prot);
740 WARN_ON(!pud_devmap(pud_mkdevmap(pud)));
741}
742#else /* !CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
743static void __init pud_devmap_tests(struct pgtable_debug_args *args) { }
744#endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
745#else /* CONFIG_TRANSPARENT_HUGEPAGE */
746static void __init pmd_devmap_tests(struct pgtable_debug_args *args) { }
747static void __init pud_devmap_tests(struct pgtable_debug_args *args) { }
748#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
749#else
750static void __init pte_devmap_tests(struct pgtable_debug_args *args) { }
751static void __init pmd_devmap_tests(struct pgtable_debug_args *args) { }
752static void __init pud_devmap_tests(struct pgtable_debug_args *args) { }
753#endif /* CONFIG_ARCH_HAS_PTE_DEVMAP */
754
755static void __init pte_soft_dirty_tests(struct pgtable_debug_args *args)
756{
757 pte_t pte = pfn_pte(page_nr: args->fixed_pte_pfn, pgprot: args->page_prot);
758
759 if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY))
760 return;
761
762 pr_debug("Validating PTE soft dirty\n");
763 WARN_ON(!pte_soft_dirty(pte_mksoft_dirty(pte)));
764 WARN_ON(pte_soft_dirty(pte_clear_soft_dirty(pte)));
765}
766
767static void __init pte_swap_soft_dirty_tests(struct pgtable_debug_args *args)
768{
769 pte_t pte = pfn_pte(page_nr: args->fixed_pte_pfn, pgprot: args->page_prot);
770
771 if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY))
772 return;
773
774 pr_debug("Validating PTE swap soft dirty\n");
775 WARN_ON(!pte_swp_soft_dirty(pte_swp_mksoft_dirty(pte)));
776 WARN_ON(pte_swp_soft_dirty(pte_swp_clear_soft_dirty(pte)));
777}
778
779#ifdef CONFIG_TRANSPARENT_HUGEPAGE
780static void __init pmd_soft_dirty_tests(struct pgtable_debug_args *args)
781{
782 pmd_t pmd;
783
784 if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY))
785 return;
786
787 if (!has_transparent_hugepage())
788 return;
789
790 pr_debug("Validating PMD soft dirty\n");
791 pmd = pfn_pmd(page_nr: args->fixed_pmd_pfn, pgprot: args->page_prot);
792 WARN_ON(!pmd_soft_dirty(pmd_mksoft_dirty(pmd)));
793 WARN_ON(pmd_soft_dirty(pmd_clear_soft_dirty(pmd)));
794}
795
796static void __init pmd_swap_soft_dirty_tests(struct pgtable_debug_args *args)
797{
798 pmd_t pmd;
799
800 if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) ||
801 !IS_ENABLED(CONFIG_ARCH_ENABLE_THP_MIGRATION))
802 return;
803
804 if (!has_transparent_hugepage())
805 return;
806
807 pr_debug("Validating PMD swap soft dirty\n");
808 pmd = pfn_pmd(page_nr: args->fixed_pmd_pfn, pgprot: args->page_prot);
809 WARN_ON(!pmd_swp_soft_dirty(pmd_swp_mksoft_dirty(pmd)));
810 WARN_ON(pmd_swp_soft_dirty(pmd_swp_clear_soft_dirty(pmd)));
811}
812#else /* !CONFIG_TRANSPARENT_HUGEPAGE */
813static void __init pmd_soft_dirty_tests(struct pgtable_debug_args *args) { }
814static void __init pmd_swap_soft_dirty_tests(struct pgtable_debug_args *args) { }
815#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
816
817static void __init pte_swap_exclusive_tests(struct pgtable_debug_args *args)
818{
819 unsigned long max_swap_offset;
820 swp_entry_t entry, entry2;
821 pte_t pte;
822
823 pr_debug("Validating PTE swap exclusive\n");
824
825 /* See generic_max_swapfile_size(): probe the maximum offset */
826 max_swap_offset = swp_offset(entry: pte_to_swp_entry(pte: swp_entry_to_pte(entry: swp_entry(type: 0, offset: ~0UL))));
827
828 /* Create a swp entry with all possible bits set */
829 entry = swp_entry(type: (1 << MAX_SWAPFILES_SHIFT) - 1, offset: max_swap_offset);
830
831 pte = swp_entry_to_pte(entry);
832 WARN_ON(pte_swp_exclusive(pte));
833 WARN_ON(!is_swap_pte(pte));
834 entry2 = pte_to_swp_entry(pte);
835 WARN_ON(memcmp(&entry, &entry2, sizeof(entry)));
836
837 pte = pte_swp_mkexclusive(pte);
838 WARN_ON(!pte_swp_exclusive(pte));
839 WARN_ON(!is_swap_pte(pte));
840 WARN_ON(pte_swp_soft_dirty(pte));
841 entry2 = pte_to_swp_entry(pte);
842 WARN_ON(memcmp(&entry, &entry2, sizeof(entry)));
843
844 pte = pte_swp_clear_exclusive(pte);
845 WARN_ON(pte_swp_exclusive(pte));
846 WARN_ON(!is_swap_pte(pte));
847 entry2 = pte_to_swp_entry(pte);
848 WARN_ON(memcmp(&entry, &entry2, sizeof(entry)));
849}
850
851static void __init pte_swap_tests(struct pgtable_debug_args *args)
852{
853 swp_entry_t swp;
854 pte_t pte;
855
856 pr_debug("Validating PTE swap\n");
857 pte = pfn_pte(page_nr: args->fixed_pte_pfn, pgprot: args->page_prot);
858 swp = __pte_to_swp_entry(pte);
859 pte = __swp_entry_to_pte(swp);
860 WARN_ON(args->fixed_pte_pfn != pte_pfn(pte));
861}
862
863#ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
864static void __init pmd_swap_tests(struct pgtable_debug_args *args)
865{
866 swp_entry_t swp;
867 pmd_t pmd;
868
869 if (!has_transparent_hugepage())
870 return;
871
872 pr_debug("Validating PMD swap\n");
873 pmd = pfn_pmd(page_nr: args->fixed_pmd_pfn, pgprot: args->page_prot);
874 swp = __pmd_to_swp_entry(pmd);
875 pmd = __swp_entry_to_pmd(swp);
876 WARN_ON(args->fixed_pmd_pfn != pmd_pfn(pmd));
877}
878#else /* !CONFIG_ARCH_ENABLE_THP_MIGRATION */
879static void __init pmd_swap_tests(struct pgtable_debug_args *args) { }
880#endif /* CONFIG_ARCH_ENABLE_THP_MIGRATION */
881
882static void __init swap_migration_tests(struct pgtable_debug_args *args)
883{
884 struct page *page;
885 swp_entry_t swp;
886
887 if (!IS_ENABLED(CONFIG_MIGRATION))
888 return;
889
890 /*
891 * swap_migration_tests() requires a dedicated page as it needs to
892 * be locked before creating a migration entry from it. Locking the
893 * page that actually maps kernel text ('start_kernel') can be real
894 * problematic. Lets use the allocated page explicitly for this
895 * purpose.
896 */
897 page = (args->pte_pfn != ULONG_MAX) ? pfn_to_page(args->pte_pfn) : NULL;
898 if (!page)
899 return;
900
901 pr_debug("Validating swap migration\n");
902
903 /*
904 * make_[readable|writable]_migration_entry() expects given page to
905 * be locked, otherwise it stumbles upon a BUG_ON().
906 */
907 __SetPageLocked(page);
908 swp = make_writable_migration_entry(page_to_pfn(page));
909 WARN_ON(!is_migration_entry(swp));
910 WARN_ON(!is_writable_migration_entry(swp));
911
912 swp = make_readable_migration_entry(offset: swp_offset(entry: swp));
913 WARN_ON(!is_migration_entry(swp));
914 WARN_ON(is_writable_migration_entry(swp));
915
916 swp = make_readable_migration_entry(page_to_pfn(page));
917 WARN_ON(!is_migration_entry(swp));
918 WARN_ON(is_writable_migration_entry(swp));
919 __ClearPageLocked(page);
920}
921
922#ifdef CONFIG_HUGETLB_PAGE
923static void __init hugetlb_basic_tests(struct pgtable_debug_args *args)
924{
925 struct page *page;
926 pte_t pte;
927
928 pr_debug("Validating HugeTLB basic\n");
929 /*
930 * Accessing the page associated with the pfn is safe here,
931 * as it was previously derived from a real kernel symbol.
932 */
933 page = pfn_to_page(args->fixed_pmd_pfn);
934 pte = mk_huge_pte(page, pgprot: args->page_prot);
935
936 WARN_ON(!huge_pte_dirty(huge_pte_mkdirty(pte)));
937 WARN_ON(!huge_pte_write(huge_pte_mkwrite(huge_pte_wrprotect(pte))));
938 WARN_ON(huge_pte_write(huge_pte_wrprotect(huge_pte_mkwrite(pte))));
939
940#ifdef CONFIG_ARCH_WANT_GENERAL_HUGETLB
941 pte = pfn_pte(page_nr: args->fixed_pmd_pfn, pgprot: args->page_prot);
942
943 WARN_ON(!pte_huge(arch_make_huge_pte(pte, PMD_SHIFT, VM_ACCESS_FLAGS)));
944#endif /* CONFIG_ARCH_WANT_GENERAL_HUGETLB */
945}
946#else /* !CONFIG_HUGETLB_PAGE */
947static void __init hugetlb_basic_tests(struct pgtable_debug_args *args) { }
948#endif /* CONFIG_HUGETLB_PAGE */
949
950#ifdef CONFIG_TRANSPARENT_HUGEPAGE
951static void __init pmd_thp_tests(struct pgtable_debug_args *args)
952{
953 pmd_t pmd;
954
955 if (!has_transparent_hugepage())
956 return;
957
958 pr_debug("Validating PMD based THP\n");
959 /*
960 * pmd_trans_huge() and pmd_present() must return positive after
961 * MMU invalidation with pmd_mkinvalid(). This behavior is an
962 * optimization for transparent huge page. pmd_trans_huge() must
963 * be true if pmd_page() returns a valid THP to avoid taking the
964 * pmd_lock when others walk over non transhuge pmds (i.e. there
965 * are no THP allocated). Especially when splitting a THP and
966 * removing the present bit from the pmd, pmd_trans_huge() still
967 * needs to return true. pmd_present() should be true whenever
968 * pmd_trans_huge() returns true.
969 */
970 pmd = pfn_pmd(page_nr: args->fixed_pmd_pfn, pgprot: args->page_prot);
971 WARN_ON(!pmd_trans_huge(pmd_mkhuge(pmd)));
972
973#ifndef __HAVE_ARCH_PMDP_INVALIDATE
974 WARN_ON(!pmd_trans_huge(pmd_mkinvalid(pmd_mkhuge(pmd))));
975 WARN_ON(!pmd_present(pmd_mkinvalid(pmd_mkhuge(pmd))));
976#endif /* __HAVE_ARCH_PMDP_INVALIDATE */
977}
978
979#ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
980static void __init pud_thp_tests(struct pgtable_debug_args *args)
981{
982 pud_t pud;
983
984 if (!has_transparent_pud_hugepage())
985 return;
986
987 pr_debug("Validating PUD based THP\n");
988 pud = pfn_pud(page_nr: args->fixed_pud_pfn, pgprot: args->page_prot);
989 WARN_ON(!pud_trans_huge(pud_mkhuge(pud)));
990
991 /*
992 * pud_mkinvalid() has been dropped for now. Enable back
993 * these tests when it comes back with a modified pud_present().
994 *
995 * WARN_ON(!pud_trans_huge(pud_mkinvalid(pud_mkhuge(pud))));
996 * WARN_ON(!pud_present(pud_mkinvalid(pud_mkhuge(pud))));
997 */
998}
999#else /* !CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
1000static void __init pud_thp_tests(struct pgtable_debug_args *args) { }
1001#endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
1002#else /* !CONFIG_TRANSPARENT_HUGEPAGE */
1003static void __init pmd_thp_tests(struct pgtable_debug_args *args) { }
1004static void __init pud_thp_tests(struct pgtable_debug_args *args) { }
1005#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1006
1007static unsigned long __init get_random_vaddr(void)
1008{
1009 unsigned long random_vaddr, random_pages, total_user_pages;
1010
1011 total_user_pages = (TASK_SIZE - FIRST_USER_ADDRESS) / PAGE_SIZE;
1012
1013 random_pages = get_random_long() % total_user_pages;
1014 random_vaddr = FIRST_USER_ADDRESS + random_pages * PAGE_SIZE;
1015
1016 return random_vaddr;
1017}
1018
1019static void __init destroy_args(struct pgtable_debug_args *args)
1020{
1021 struct page *page = NULL;
1022
1023 /* Free (huge) page */
1024 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
1025 has_transparent_pud_hugepage() &&
1026 args->pud_pfn != ULONG_MAX) {
1027 if (args->is_contiguous_page) {
1028 free_contig_range(pfn: args->pud_pfn,
1029 nr_pages: (1 << (HPAGE_PUD_SHIFT - PAGE_SHIFT)));
1030 } else {
1031 page = pfn_to_page(args->pud_pfn);
1032 __free_pages(page, HPAGE_PUD_SHIFT - PAGE_SHIFT);
1033 }
1034
1035 args->pud_pfn = ULONG_MAX;
1036 args->pmd_pfn = ULONG_MAX;
1037 args->pte_pfn = ULONG_MAX;
1038 }
1039
1040 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
1041 has_transparent_hugepage() &&
1042 args->pmd_pfn != ULONG_MAX) {
1043 if (args->is_contiguous_page) {
1044 free_contig_range(pfn: args->pmd_pfn, nr_pages: (1 << HPAGE_PMD_ORDER));
1045 } else {
1046 page = pfn_to_page(args->pmd_pfn);
1047 __free_pages(page, HPAGE_PMD_ORDER);
1048 }
1049
1050 args->pmd_pfn = ULONG_MAX;
1051 args->pte_pfn = ULONG_MAX;
1052 }
1053
1054 if (args->pte_pfn != ULONG_MAX) {
1055 page = pfn_to_page(args->pte_pfn);
1056 __free_page(page);
1057
1058 args->pte_pfn = ULONG_MAX;
1059 }
1060
1061 /* Free page table entries */
1062 if (args->start_ptep) {
1063 pte_free(mm: args->mm, pte_page: args->start_ptep);
1064 mm_dec_nr_ptes(mm: args->mm);
1065 }
1066
1067 if (args->start_pmdp) {
1068 pmd_free(mm: args->mm, pmd: args->start_pmdp);
1069 mm_dec_nr_pmds(mm: args->mm);
1070 }
1071
1072 if (args->start_pudp) {
1073 pud_free(mm: args->mm, pud: args->start_pudp);
1074 mm_dec_nr_puds(mm: args->mm);
1075 }
1076
1077 if (args->start_p4dp)
1078 p4d_free(mm: args->mm, p4d: args->start_p4dp);
1079
1080 /* Free vma and mm struct */
1081 if (args->vma)
1082 vm_area_free(args->vma);
1083
1084 if (args->mm)
1085 mmdrop(mm: args->mm);
1086}
1087
1088static struct page * __init
1089debug_vm_pgtable_alloc_huge_page(struct pgtable_debug_args *args, int order)
1090{
1091 struct page *page = NULL;
1092
1093#ifdef CONFIG_CONTIG_ALLOC
1094 if (order > MAX_ORDER) {
1095 page = alloc_contig_pages(nr_pages: (1 << order), GFP_KERNEL,
1096 first_online_node, NULL);
1097 if (page) {
1098 args->is_contiguous_page = true;
1099 return page;
1100 }
1101 }
1102#endif
1103
1104 if (order <= MAX_ORDER)
1105 page = alloc_pages(GFP_KERNEL, order);
1106
1107 return page;
1108}
1109
1110/*
1111 * Check if a physical memory range described by <pstart, pend> contains
1112 * an area that is of size psize, and aligned to psize.
1113 *
1114 * Don't use address 0, an all-zeroes physical address might mask bugs, and
1115 * it's not used on x86.
1116 */
1117static void __init phys_align_check(phys_addr_t pstart,
1118 phys_addr_t pend, unsigned long psize,
1119 phys_addr_t *physp, unsigned long *alignp)
1120{
1121 phys_addr_t aligned_start, aligned_end;
1122
1123 if (pstart == 0)
1124 pstart = PAGE_SIZE;
1125
1126 aligned_start = ALIGN(pstart, psize);
1127 aligned_end = aligned_start + psize;
1128
1129 if (aligned_end > aligned_start && aligned_end <= pend) {
1130 *alignp = psize;
1131 *physp = aligned_start;
1132 }
1133}
1134
1135static void __init init_fixed_pfns(struct pgtable_debug_args *args)
1136{
1137 u64 idx;
1138 phys_addr_t phys, pstart, pend;
1139
1140 /*
1141 * Initialize the fixed pfns. To do this, try to find a
1142 * valid physical range, preferably aligned to PUD_SIZE,
1143 * but settling for aligned to PMD_SIZE as a fallback. If
1144 * neither of those is found, use the physical address of
1145 * the start_kernel symbol.
1146 *
1147 * The memory doesn't need to be allocated, it just needs to exist
1148 * as usable memory. It won't be touched.
1149 *
1150 * The alignment is recorded, and can be checked to see if we
1151 * can run the tests that require an actual valid physical
1152 * address range on some architectures ({pmd,pud}_huge_test
1153 * on x86).
1154 */
1155
1156 phys = __pa_symbol(&start_kernel);
1157 args->fixed_alignment = PAGE_SIZE;
1158
1159 for_each_mem_range(idx, &pstart, &pend) {
1160 /* First check for a PUD-aligned area */
1161 phys_align_check(pstart, pend, PUD_SIZE, physp: &phys,
1162 alignp: &args->fixed_alignment);
1163
1164 /* If a PUD-aligned area is found, we're done */
1165 if (args->fixed_alignment == PUD_SIZE)
1166 break;
1167
1168 /*
1169 * If no PMD-aligned area found yet, check for one,
1170 * but continue the loop to look for a PUD-aligned area.
1171 */
1172 if (args->fixed_alignment < PMD_SIZE)
1173 phys_align_check(pstart, pend, PMD_SIZE, physp: &phys,
1174 alignp: &args->fixed_alignment);
1175 }
1176
1177 args->fixed_pgd_pfn = __phys_to_pfn(phys & PGDIR_MASK);
1178 args->fixed_p4d_pfn = __phys_to_pfn(phys & P4D_MASK);
1179 args->fixed_pud_pfn = __phys_to_pfn(phys & PUD_MASK);
1180 args->fixed_pmd_pfn = __phys_to_pfn(phys & PMD_MASK);
1181 args->fixed_pte_pfn = __phys_to_pfn(phys & PAGE_MASK);
1182 WARN_ON(!pfn_valid(args->fixed_pte_pfn));
1183}
1184
1185
1186static int __init init_args(struct pgtable_debug_args *args)
1187{
1188 struct page *page = NULL;
1189 int ret = 0;
1190
1191 /*
1192 * Initialize the debugging data.
1193 *
1194 * vm_get_page_prot(VM_NONE) or vm_get_page_prot(VM_SHARED|VM_NONE)
1195 * will help create page table entries with PROT_NONE permission as
1196 * required for pxx_protnone_tests().
1197 */
1198 memset(args, 0, sizeof(*args));
1199 args->vaddr = get_random_vaddr();
1200 args->page_prot = vm_get_page_prot(VM_ACCESS_FLAGS);
1201 args->page_prot_none = vm_get_page_prot(VM_NONE);
1202 args->is_contiguous_page = false;
1203 args->pud_pfn = ULONG_MAX;
1204 args->pmd_pfn = ULONG_MAX;
1205 args->pte_pfn = ULONG_MAX;
1206 args->fixed_pgd_pfn = ULONG_MAX;
1207 args->fixed_p4d_pfn = ULONG_MAX;
1208 args->fixed_pud_pfn = ULONG_MAX;
1209 args->fixed_pmd_pfn = ULONG_MAX;
1210 args->fixed_pte_pfn = ULONG_MAX;
1211
1212 /* Allocate mm and vma */
1213 args->mm = mm_alloc();
1214 if (!args->mm) {
1215 pr_err("Failed to allocate mm struct\n");
1216 ret = -ENOMEM;
1217 goto error;
1218 }
1219
1220 args->vma = vm_area_alloc(args->mm);
1221 if (!args->vma) {
1222 pr_err("Failed to allocate vma\n");
1223 ret = -ENOMEM;
1224 goto error;
1225 }
1226
1227 /*
1228 * Allocate page table entries. They will be modified in the tests.
1229 * Lets save the page table entries so that they can be released
1230 * when the tests are completed.
1231 */
1232 args->pgdp = pgd_offset(args->mm, args->vaddr);
1233 args->p4dp = p4d_alloc(mm: args->mm, pgd: args->pgdp, address: args->vaddr);
1234 if (!args->p4dp) {
1235 pr_err("Failed to allocate p4d entries\n");
1236 ret = -ENOMEM;
1237 goto error;
1238 }
1239 args->start_p4dp = p4d_offset(pgd: args->pgdp, address: 0UL);
1240 WARN_ON(!args->start_p4dp);
1241
1242 args->pudp = pud_alloc(mm: args->mm, p4d: args->p4dp, address: args->vaddr);
1243 if (!args->pudp) {
1244 pr_err("Failed to allocate pud entries\n");
1245 ret = -ENOMEM;
1246 goto error;
1247 }
1248 args->start_pudp = pud_offset(p4d: args->p4dp, address: 0UL);
1249 WARN_ON(!args->start_pudp);
1250
1251 args->pmdp = pmd_alloc(mm: args->mm, pud: args->pudp, address: args->vaddr);
1252 if (!args->pmdp) {
1253 pr_err("Failed to allocate pmd entries\n");
1254 ret = -ENOMEM;
1255 goto error;
1256 }
1257 args->start_pmdp = pmd_offset(pud: args->pudp, address: 0UL);
1258 WARN_ON(!args->start_pmdp);
1259
1260 if (pte_alloc(args->mm, args->pmdp)) {
1261 pr_err("Failed to allocate pte entries\n");
1262 ret = -ENOMEM;
1263 goto error;
1264 }
1265 args->start_ptep = pmd_pgtable(READ_ONCE(*args->pmdp));
1266 WARN_ON(!args->start_ptep);
1267
1268 init_fixed_pfns(args);
1269
1270 /*
1271 * Allocate (huge) pages because some of the tests need to access
1272 * the data in the pages. The corresponding tests will be skipped
1273 * if we fail to allocate (huge) pages.
1274 */
1275 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
1276 has_transparent_pud_hugepage()) {
1277 page = debug_vm_pgtable_alloc_huge_page(args,
1278 HPAGE_PUD_SHIFT - PAGE_SHIFT);
1279 if (page) {
1280 args->pud_pfn = page_to_pfn(page);
1281 args->pmd_pfn = args->pud_pfn;
1282 args->pte_pfn = args->pud_pfn;
1283 return 0;
1284 }
1285 }
1286
1287 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
1288 has_transparent_hugepage()) {
1289 page = debug_vm_pgtable_alloc_huge_page(args, HPAGE_PMD_ORDER);
1290 if (page) {
1291 args->pmd_pfn = page_to_pfn(page);
1292 args->pte_pfn = args->pmd_pfn;
1293 return 0;
1294 }
1295 }
1296
1297 page = alloc_page(GFP_KERNEL);
1298 if (page)
1299 args->pte_pfn = page_to_pfn(page);
1300
1301 return 0;
1302
1303error:
1304 destroy_args(args);
1305 return ret;
1306}
1307
1308static int __init debug_vm_pgtable(void)
1309{
1310 struct pgtable_debug_args args;
1311 spinlock_t *ptl = NULL;
1312 int idx, ret;
1313
1314 pr_info("Validating architecture page table helpers\n");
1315 ret = init_args(args: &args);
1316 if (ret)
1317 return ret;
1318
1319 /*
1320 * Iterate over each possible vm_flags to make sure that all
1321 * the basic page table transformation validations just hold
1322 * true irrespective of the starting protection value for a
1323 * given page table entry.
1324 *
1325 * Protection based vm_flags combinations are always linear
1326 * and increasing i.e starting from VM_NONE and going up to
1327 * (VM_SHARED | READ | WRITE | EXEC).
1328 */
1329#define VM_FLAGS_START (VM_NONE)
1330#define VM_FLAGS_END (VM_SHARED | VM_EXEC | VM_WRITE | VM_READ)
1331
1332 for (idx = VM_FLAGS_START; idx <= VM_FLAGS_END; idx++) {
1333 pte_basic_tests(args: &args, idx);
1334 pmd_basic_tests(args: &args, idx);
1335 pud_basic_tests(args: &args, idx);
1336 }
1337
1338 /*
1339 * Both P4D and PGD level tests are very basic which do not
1340 * involve creating page table entries from the protection
1341 * value and the given pfn. Hence just keep them out from
1342 * the above iteration for now to save some test execution
1343 * time.
1344 */
1345 p4d_basic_tests(args: &args);
1346 pgd_basic_tests(args: &args);
1347
1348 pmd_leaf_tests(args: &args);
1349 pud_leaf_tests(args: &args);
1350
1351 pte_special_tests(args: &args);
1352 pte_protnone_tests(args: &args);
1353 pmd_protnone_tests(args: &args);
1354
1355 pte_devmap_tests(args: &args);
1356 pmd_devmap_tests(args: &args);
1357 pud_devmap_tests(args: &args);
1358
1359 pte_soft_dirty_tests(args: &args);
1360 pmd_soft_dirty_tests(args: &args);
1361 pte_swap_soft_dirty_tests(args: &args);
1362 pmd_swap_soft_dirty_tests(args: &args);
1363
1364 pte_swap_exclusive_tests(args: &args);
1365
1366 pte_swap_tests(args: &args);
1367 pmd_swap_tests(args: &args);
1368
1369 swap_migration_tests(args: &args);
1370
1371 pmd_thp_tests(args: &args);
1372 pud_thp_tests(args: &args);
1373
1374 hugetlb_basic_tests(args: &args);
1375
1376 /*
1377 * Page table modifying tests. They need to hold
1378 * proper page table lock.
1379 */
1380
1381 args.ptep = pte_offset_map_lock(mm: args.mm, pmd: args.pmdp, addr: args.vaddr, ptlp: &ptl);
1382 pte_clear_tests(args: &args);
1383 pte_advanced_tests(args: &args);
1384 if (args.ptep)
1385 pte_unmap_unlock(args.ptep, ptl);
1386
1387 ptl = pmd_lock(mm: args.mm, pmd: args.pmdp);
1388 pmd_clear_tests(args: &args);
1389 pmd_advanced_tests(args: &args);
1390 pmd_huge_tests(args: &args);
1391 pmd_populate_tests(args: &args);
1392 spin_unlock(lock: ptl);
1393
1394 ptl = pud_lock(mm: args.mm, pud: args.pudp);
1395 pud_clear_tests(args: &args);
1396 pud_advanced_tests(args: &args);
1397 pud_huge_tests(args: &args);
1398 pud_populate_tests(args: &args);
1399 spin_unlock(lock: ptl);
1400
1401 spin_lock(lock: &(args.mm->page_table_lock));
1402 p4d_clear_tests(args: &args);
1403 pgd_clear_tests(args: &args);
1404 p4d_populate_tests(args: &args);
1405 pgd_populate_tests(args: &args);
1406 spin_unlock(lock: &(args.mm->page_table_lock));
1407
1408 destroy_args(args: &args);
1409 return 0;
1410}
1411late_initcall(debug_vm_pgtable);
1412

source code of linux/mm/debug_vm_pgtable.c