xref: /openbmc/linux/mm/debug_vm_pgtable.c (revision e15a5365)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * This kernel test validates architecture page table helpers and
4  * accessors and helps in verifying their continued compliance with
5  * expected generic MM semantics.
6  *
7  * Copyright (C) 2019 ARM Ltd.
8  *
9  * Author: Anshuman Khandual <anshuman.khandual@arm.com>
10  */
11 #define pr_fmt(fmt) "debug_vm_pgtable: [%-25s]: " fmt, __func__
12 
13 #include <linux/gfp.h>
14 #include <linux/highmem.h>
15 #include <linux/hugetlb.h>
16 #include <linux/kernel.h>
17 #include <linux/kconfig.h>
18 #include <linux/mm.h>
19 #include <linux/mman.h>
20 #include <linux/mm_types.h>
21 #include <linux/module.h>
22 #include <linux/pfn_t.h>
23 #include <linux/printk.h>
24 #include <linux/pgtable.h>
25 #include <linux/random.h>
26 #include <linux/spinlock.h>
27 #include <linux/swap.h>
28 #include <linux/swapops.h>
29 #include <linux/start_kernel.h>
30 #include <linux/sched/mm.h>
31 #include <linux/io.h>
32 #include <asm/pgalloc.h>
33 #include <asm/tlbflush.h>
34 
35 /*
36  * Please refer Documentation/vm/arch_pgtable_helpers.rst for the semantics
37  * expectations that are being validated here. All future changes in here
38  * or the documentation need to be in sync.
39  */
40 
41 #define VMFLAGS	(VM_READ|VM_WRITE|VM_EXEC)
42 
43 /*
44  * On s390 platform, the lower 4 bits are used to identify given page table
45  * entry type. But these bits might affect the ability to clear entries with
46  * pxx_clear() because of how dynamic page table folding works on s390. So
47  * while loading up the entries do not change the lower 4 bits. It does not
48  * have affect any other platform. Also avoid the 62nd bit on ppc64 that is
49  * used to mark a pte entry.
50  */
51 #define S390_SKIP_MASK		GENMASK(3, 0)
52 #if __BITS_PER_LONG == 64
53 #define PPC64_SKIP_MASK		GENMASK(62, 62)
54 #else
55 #define PPC64_SKIP_MASK		0x0
56 #endif
57 #define ARCH_SKIP_MASK (S390_SKIP_MASK | PPC64_SKIP_MASK)
58 #define RANDOM_ORVALUE (GENMASK(BITS_PER_LONG - 1, 0) & ~ARCH_SKIP_MASK)
59 #define RANDOM_NZVALUE	GENMASK(7, 0)
60 
61 static void __init pte_basic_tests(unsigned long pfn, pgprot_t prot)
62 {
63 	pte_t pte = pfn_pte(pfn, prot);
64 
65 	pr_debug("Validating PTE basic\n");
66 	WARN_ON(!pte_same(pte, pte));
67 	WARN_ON(!pte_young(pte_mkyoung(pte_mkold(pte))));
68 	WARN_ON(!pte_dirty(pte_mkdirty(pte_mkclean(pte))));
69 	WARN_ON(!pte_write(pte_mkwrite(pte_wrprotect(pte))));
70 	WARN_ON(pte_young(pte_mkold(pte_mkyoung(pte))));
71 	WARN_ON(pte_dirty(pte_mkclean(pte_mkdirty(pte))));
72 	WARN_ON(pte_write(pte_wrprotect(pte_mkwrite(pte))));
73 }
74 
75 static void __init pte_advanced_tests(struct mm_struct *mm,
76 				      struct vm_area_struct *vma, pte_t *ptep,
77 				      unsigned long pfn, unsigned long vaddr,
78 				      pgprot_t prot)
79 {
80 	pte_t pte = pfn_pte(pfn, prot);
81 
82 	/*
83 	 * Architectures optimize set_pte_at by avoiding TLB flush.
84 	 * This requires set_pte_at to be not used to update an
85 	 * existing pte entry. Clear pte before we do set_pte_at
86 	 */
87 
88 	pr_debug("Validating PTE advanced\n");
89 	pte = pfn_pte(pfn, prot);
90 	set_pte_at(mm, vaddr, ptep, pte);
91 	ptep_set_wrprotect(mm, vaddr, ptep);
92 	pte = ptep_get(ptep);
93 	WARN_ON(pte_write(pte));
94 	ptep_get_and_clear(mm, vaddr, ptep);
95 	pte = ptep_get(ptep);
96 	WARN_ON(!pte_none(pte));
97 
98 	pte = pfn_pte(pfn, prot);
99 	pte = pte_wrprotect(pte);
100 	pte = pte_mkclean(pte);
101 	set_pte_at(mm, vaddr, ptep, pte);
102 	pte = pte_mkwrite(pte);
103 	pte = pte_mkdirty(pte);
104 	ptep_set_access_flags(vma, vaddr, ptep, pte, 1);
105 	pte = ptep_get(ptep);
106 	WARN_ON(!(pte_write(pte) && pte_dirty(pte)));
107 	ptep_get_and_clear_full(mm, vaddr, ptep, 1);
108 	pte = ptep_get(ptep);
109 	WARN_ON(!pte_none(pte));
110 
111 	pte = pfn_pte(pfn, prot);
112 	pte = pte_mkyoung(pte);
113 	set_pte_at(mm, vaddr, ptep, pte);
114 	ptep_test_and_clear_young(vma, vaddr, ptep);
115 	pte = ptep_get(ptep);
116 	WARN_ON(pte_young(pte));
117 }
118 
119 static void __init pte_savedwrite_tests(unsigned long pfn, pgprot_t prot)
120 {
121 	pte_t pte = pfn_pte(pfn, prot);
122 
123 	if (!IS_ENABLED(CONFIG_NUMA_BALANCING))
124 		return;
125 
126 	pr_debug("Validating PTE saved write\n");
127 	WARN_ON(!pte_savedwrite(pte_mk_savedwrite(pte_clear_savedwrite(pte))));
128 	WARN_ON(pte_savedwrite(pte_clear_savedwrite(pte_mk_savedwrite(pte))));
129 }
130 
131 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
132 static void __init pmd_basic_tests(unsigned long pfn, pgprot_t prot)
133 {
134 	pmd_t pmd = pfn_pmd(pfn, prot);
135 
136 	if (!has_transparent_hugepage())
137 		return;
138 
139 	pr_debug("Validating PMD basic\n");
140 	WARN_ON(!pmd_same(pmd, pmd));
141 	WARN_ON(!pmd_young(pmd_mkyoung(pmd_mkold(pmd))));
142 	WARN_ON(!pmd_dirty(pmd_mkdirty(pmd_mkclean(pmd))));
143 	WARN_ON(!pmd_write(pmd_mkwrite(pmd_wrprotect(pmd))));
144 	WARN_ON(pmd_young(pmd_mkold(pmd_mkyoung(pmd))));
145 	WARN_ON(pmd_dirty(pmd_mkclean(pmd_mkdirty(pmd))));
146 	WARN_ON(pmd_write(pmd_wrprotect(pmd_mkwrite(pmd))));
147 	/*
148 	 * A huge page does not point to next level page table
149 	 * entry. Hence this must qualify as pmd_bad().
150 	 */
151 	WARN_ON(!pmd_bad(pmd_mkhuge(pmd)));
152 }
153 
154 static void __init pmd_advanced_tests(struct mm_struct *mm,
155 				      struct vm_area_struct *vma, pmd_t *pmdp,
156 				      unsigned long pfn, unsigned long vaddr,
157 				      pgprot_t prot, pgtable_t pgtable)
158 {
159 	pmd_t pmd = pfn_pmd(pfn, prot);
160 
161 	if (!has_transparent_hugepage())
162 		return;
163 
164 	pr_debug("Validating PMD advanced\n");
165 	/* Align the address wrt HPAGE_PMD_SIZE */
166 	vaddr = (vaddr & HPAGE_PMD_MASK) + HPAGE_PMD_SIZE;
167 
168 	pgtable_trans_huge_deposit(mm, pmdp, pgtable);
169 
170 	pmd = pfn_pmd(pfn, prot);
171 	set_pmd_at(mm, vaddr, pmdp, pmd);
172 	pmdp_set_wrprotect(mm, vaddr, pmdp);
173 	pmd = READ_ONCE(*pmdp);
174 	WARN_ON(pmd_write(pmd));
175 	pmdp_huge_get_and_clear(mm, vaddr, pmdp);
176 	pmd = READ_ONCE(*pmdp);
177 	WARN_ON(!pmd_none(pmd));
178 
179 	pmd = pfn_pmd(pfn, prot);
180 	pmd = pmd_wrprotect(pmd);
181 	pmd = pmd_mkclean(pmd);
182 	set_pmd_at(mm, vaddr, pmdp, pmd);
183 	pmd = pmd_mkwrite(pmd);
184 	pmd = pmd_mkdirty(pmd);
185 	pmdp_set_access_flags(vma, vaddr, pmdp, pmd, 1);
186 	pmd = READ_ONCE(*pmdp);
187 	WARN_ON(!(pmd_write(pmd) && pmd_dirty(pmd)));
188 	pmdp_huge_get_and_clear_full(vma, vaddr, pmdp, 1);
189 	pmd = READ_ONCE(*pmdp);
190 	WARN_ON(!pmd_none(pmd));
191 
192 	pmd = pmd_mkhuge(pfn_pmd(pfn, prot));
193 	pmd = pmd_mkyoung(pmd);
194 	set_pmd_at(mm, vaddr, pmdp, pmd);
195 	pmdp_test_and_clear_young(vma, vaddr, pmdp);
196 	pmd = READ_ONCE(*pmdp);
197 	WARN_ON(pmd_young(pmd));
198 
199 	/*  Clear the pte entries  */
200 	pmdp_huge_get_and_clear(mm, vaddr, pmdp);
201 	pgtable = pgtable_trans_huge_withdraw(mm, pmdp);
202 }
203 
204 static void __init pmd_leaf_tests(unsigned long pfn, pgprot_t prot)
205 {
206 	pmd_t pmd = pfn_pmd(pfn, prot);
207 
208 	pr_debug("Validating PMD leaf\n");
209 	/*
210 	 * PMD based THP is a leaf entry.
211 	 */
212 	pmd = pmd_mkhuge(pmd);
213 	WARN_ON(!pmd_leaf(pmd));
214 }
215 
216 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
217 static void __init pmd_huge_tests(pmd_t *pmdp, unsigned long pfn, pgprot_t prot)
218 {
219 	pmd_t pmd;
220 
221 	if (!arch_ioremap_pmd_supported())
222 		return;
223 
224 	pr_debug("Validating PMD huge\n");
225 	/*
226 	 * X86 defined pmd_set_huge() verifies that the given
227 	 * PMD is not a populated non-leaf entry.
228 	 */
229 	WRITE_ONCE(*pmdp, __pmd(0));
230 	WARN_ON(!pmd_set_huge(pmdp, __pfn_to_phys(pfn), prot));
231 	WARN_ON(!pmd_clear_huge(pmdp));
232 	pmd = READ_ONCE(*pmdp);
233 	WARN_ON(!pmd_none(pmd));
234 }
235 #else /* CONFIG_HAVE_ARCH_HUGE_VMAP */
236 static void __init pmd_huge_tests(pmd_t *pmdp, unsigned long pfn, pgprot_t prot) { }
237 #endif /* CONFIG_HAVE_ARCH_HUGE_VMAP */
238 
239 static void __init pmd_savedwrite_tests(unsigned long pfn, pgprot_t prot)
240 {
241 	pmd_t pmd = pfn_pmd(pfn, prot);
242 
243 	if (!IS_ENABLED(CONFIG_NUMA_BALANCING))
244 		return;
245 
246 	pr_debug("Validating PMD saved write\n");
247 	WARN_ON(!pmd_savedwrite(pmd_mk_savedwrite(pmd_clear_savedwrite(pmd))));
248 	WARN_ON(pmd_savedwrite(pmd_clear_savedwrite(pmd_mk_savedwrite(pmd))));
249 }
250 
251 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
252 static void __init pud_basic_tests(unsigned long pfn, pgprot_t prot)
253 {
254 	pud_t pud = pfn_pud(pfn, prot);
255 
256 	if (!has_transparent_hugepage())
257 		return;
258 
259 	pr_debug("Validating PUD basic\n");
260 	WARN_ON(!pud_same(pud, pud));
261 	WARN_ON(!pud_young(pud_mkyoung(pud_mkold(pud))));
262 	WARN_ON(!pud_write(pud_mkwrite(pud_wrprotect(pud))));
263 	WARN_ON(pud_write(pud_wrprotect(pud_mkwrite(pud))));
264 	WARN_ON(pud_young(pud_mkold(pud_mkyoung(pud))));
265 
266 	if (mm_pmd_folded(mm))
267 		return;
268 
269 	/*
270 	 * A huge page does not point to next level page table
271 	 * entry. Hence this must qualify as pud_bad().
272 	 */
273 	WARN_ON(!pud_bad(pud_mkhuge(pud)));
274 }
275 
276 static void __init pud_advanced_tests(struct mm_struct *mm,
277 				      struct vm_area_struct *vma, pud_t *pudp,
278 				      unsigned long pfn, unsigned long vaddr,
279 				      pgprot_t prot)
280 {
281 	pud_t pud = pfn_pud(pfn, prot);
282 
283 	if (!has_transparent_hugepage())
284 		return;
285 
286 	pr_debug("Validating PUD advanced\n");
287 	/* Align the address wrt HPAGE_PUD_SIZE */
288 	vaddr = (vaddr & HPAGE_PUD_MASK) + HPAGE_PUD_SIZE;
289 
290 	set_pud_at(mm, vaddr, pudp, pud);
291 	pudp_set_wrprotect(mm, vaddr, pudp);
292 	pud = READ_ONCE(*pudp);
293 	WARN_ON(pud_write(pud));
294 
295 #ifndef __PAGETABLE_PMD_FOLDED
296 	pudp_huge_get_and_clear(mm, vaddr, pudp);
297 	pud = READ_ONCE(*pudp);
298 	WARN_ON(!pud_none(pud));
299 #endif /* __PAGETABLE_PMD_FOLDED */
300 	pud = pfn_pud(pfn, prot);
301 	pud = pud_wrprotect(pud);
302 	pud = pud_mkclean(pud);
303 	set_pud_at(mm, vaddr, pudp, pud);
304 	pud = pud_mkwrite(pud);
305 	pud = pud_mkdirty(pud);
306 	pudp_set_access_flags(vma, vaddr, pudp, pud, 1);
307 	pud = READ_ONCE(*pudp);
308 	WARN_ON(!(pud_write(pud) && pud_dirty(pud)));
309 
310 #ifndef __PAGETABLE_PMD_FOLDED
311 	pudp_huge_get_and_clear_full(mm, vaddr, pudp, 1);
312 	pud = READ_ONCE(*pudp);
313 	WARN_ON(!pud_none(pud));
314 #endif /* __PAGETABLE_PMD_FOLDED */
315 
316 	pud = pfn_pud(pfn, prot);
317 	pud = pud_mkyoung(pud);
318 	set_pud_at(mm, vaddr, pudp, pud);
319 	pudp_test_and_clear_young(vma, vaddr, pudp);
320 	pud = READ_ONCE(*pudp);
321 	WARN_ON(pud_young(pud));
322 
323 	pudp_huge_get_and_clear(mm, vaddr, pudp);
324 }
325 
326 static void __init pud_leaf_tests(unsigned long pfn, pgprot_t prot)
327 {
328 	pud_t pud = pfn_pud(pfn, prot);
329 
330 	pr_debug("Validating PUD leaf\n");
331 	/*
332 	 * PUD based THP is a leaf entry.
333 	 */
334 	pud = pud_mkhuge(pud);
335 	WARN_ON(!pud_leaf(pud));
336 }
337 
338 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
339 static void __init pud_huge_tests(pud_t *pudp, unsigned long pfn, pgprot_t prot)
340 {
341 	pud_t pud;
342 
343 	if (!arch_ioremap_pud_supported())
344 		return;
345 
346 	pr_debug("Validating PUD huge\n");
347 	/*
348 	 * X86 defined pud_set_huge() verifies that the given
349 	 * PUD is not a populated non-leaf entry.
350 	 */
351 	WRITE_ONCE(*pudp, __pud(0));
352 	WARN_ON(!pud_set_huge(pudp, __pfn_to_phys(pfn), prot));
353 	WARN_ON(!pud_clear_huge(pudp));
354 	pud = READ_ONCE(*pudp);
355 	WARN_ON(!pud_none(pud));
356 }
357 #else /* !CONFIG_HAVE_ARCH_HUGE_VMAP */
358 static void __init pud_huge_tests(pud_t *pudp, unsigned long pfn, pgprot_t prot) { }
359 #endif /* !CONFIG_HAVE_ARCH_HUGE_VMAP */
360 
361 #else  /* !CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
362 static void __init pud_basic_tests(unsigned long pfn, pgprot_t prot) { }
363 static void __init pud_advanced_tests(struct mm_struct *mm,
364 				      struct vm_area_struct *vma, pud_t *pudp,
365 				      unsigned long pfn, unsigned long vaddr,
366 				      pgprot_t prot)
367 {
368 }
369 static void __init pud_leaf_tests(unsigned long pfn, pgprot_t prot) { }
370 static void __init pud_huge_tests(pud_t *pudp, unsigned long pfn, pgprot_t prot)
371 {
372 }
373 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
374 #else  /* !CONFIG_TRANSPARENT_HUGEPAGE */
375 static void __init pmd_basic_tests(unsigned long pfn, pgprot_t prot) { }
376 static void __init pud_basic_tests(unsigned long pfn, pgprot_t prot) { }
377 static void __init pmd_advanced_tests(struct mm_struct *mm,
378 				      struct vm_area_struct *vma, pmd_t *pmdp,
379 				      unsigned long pfn, unsigned long vaddr,
380 				      pgprot_t prot, pgtable_t pgtable)
381 {
382 }
383 static void __init pud_advanced_tests(struct mm_struct *mm,
384 				      struct vm_area_struct *vma, pud_t *pudp,
385 				      unsigned long pfn, unsigned long vaddr,
386 				      pgprot_t prot)
387 {
388 }
389 static void __init pmd_leaf_tests(unsigned long pfn, pgprot_t prot) { }
390 static void __init pud_leaf_tests(unsigned long pfn, pgprot_t prot) { }
391 static void __init pmd_huge_tests(pmd_t *pmdp, unsigned long pfn, pgprot_t prot)
392 {
393 }
394 static void __init pud_huge_tests(pud_t *pudp, unsigned long pfn, pgprot_t prot)
395 {
396 }
397 static void __init pmd_savedwrite_tests(unsigned long pfn, pgprot_t prot) { }
398 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
399 
400 static void __init p4d_basic_tests(unsigned long pfn, pgprot_t prot)
401 {
402 	p4d_t p4d;
403 
404 	pr_debug("Validating P4D basic\n");
405 	memset(&p4d, RANDOM_NZVALUE, sizeof(p4d_t));
406 	WARN_ON(!p4d_same(p4d, p4d));
407 }
408 
409 static void __init pgd_basic_tests(unsigned long pfn, pgprot_t prot)
410 {
411 	pgd_t pgd;
412 
413 	pr_debug("Validating PGD basic\n");
414 	memset(&pgd, RANDOM_NZVALUE, sizeof(pgd_t));
415 	WARN_ON(!pgd_same(pgd, pgd));
416 }
417 
418 #ifndef __PAGETABLE_PUD_FOLDED
419 static void __init pud_clear_tests(struct mm_struct *mm, pud_t *pudp)
420 {
421 	pud_t pud = READ_ONCE(*pudp);
422 
423 	if (mm_pmd_folded(mm))
424 		return;
425 
426 	pr_debug("Validating PUD clear\n");
427 	pud = __pud(pud_val(pud) | RANDOM_ORVALUE);
428 	WRITE_ONCE(*pudp, pud);
429 	pud_clear(pudp);
430 	pud = READ_ONCE(*pudp);
431 	WARN_ON(!pud_none(pud));
432 }
433 
434 static void __init pud_populate_tests(struct mm_struct *mm, pud_t *pudp,
435 				      pmd_t *pmdp)
436 {
437 	pud_t pud;
438 
439 	if (mm_pmd_folded(mm))
440 		return;
441 
442 	pr_debug("Validating PUD populate\n");
443 	/*
444 	 * This entry points to next level page table page.
445 	 * Hence this must not qualify as pud_bad().
446 	 */
447 	pud_populate(mm, pudp, pmdp);
448 	pud = READ_ONCE(*pudp);
449 	WARN_ON(pud_bad(pud));
450 }
451 #else  /* !__PAGETABLE_PUD_FOLDED */
452 static void __init pud_clear_tests(struct mm_struct *mm, pud_t *pudp) { }
453 static void __init pud_populate_tests(struct mm_struct *mm, pud_t *pudp,
454 				      pmd_t *pmdp)
455 {
456 }
457 #endif /* PAGETABLE_PUD_FOLDED */
458 
459 #ifndef __PAGETABLE_P4D_FOLDED
460 static void __init p4d_clear_tests(struct mm_struct *mm, p4d_t *p4dp)
461 {
462 	p4d_t p4d = READ_ONCE(*p4dp);
463 
464 	if (mm_pud_folded(mm))
465 		return;
466 
467 	pr_debug("Validating P4D clear\n");
468 	p4d = __p4d(p4d_val(p4d) | RANDOM_ORVALUE);
469 	WRITE_ONCE(*p4dp, p4d);
470 	p4d_clear(p4dp);
471 	p4d = READ_ONCE(*p4dp);
472 	WARN_ON(!p4d_none(p4d));
473 }
474 
475 static void __init p4d_populate_tests(struct mm_struct *mm, p4d_t *p4dp,
476 				      pud_t *pudp)
477 {
478 	p4d_t p4d;
479 
480 	if (mm_pud_folded(mm))
481 		return;
482 
483 	pr_debug("Validating P4D populate\n");
484 	/*
485 	 * This entry points to next level page table page.
486 	 * Hence this must not qualify as p4d_bad().
487 	 */
488 	pud_clear(pudp);
489 	p4d_clear(p4dp);
490 	p4d_populate(mm, p4dp, pudp);
491 	p4d = READ_ONCE(*p4dp);
492 	WARN_ON(p4d_bad(p4d));
493 }
494 
495 static void __init pgd_clear_tests(struct mm_struct *mm, pgd_t *pgdp)
496 {
497 	pgd_t pgd = READ_ONCE(*pgdp);
498 
499 	if (mm_p4d_folded(mm))
500 		return;
501 
502 	pr_debug("Validating PGD clear\n");
503 	pgd = __pgd(pgd_val(pgd) | RANDOM_ORVALUE);
504 	WRITE_ONCE(*pgdp, pgd);
505 	pgd_clear(pgdp);
506 	pgd = READ_ONCE(*pgdp);
507 	WARN_ON(!pgd_none(pgd));
508 }
509 
510 static void __init pgd_populate_tests(struct mm_struct *mm, pgd_t *pgdp,
511 				      p4d_t *p4dp)
512 {
513 	pgd_t pgd;
514 
515 	if (mm_p4d_folded(mm))
516 		return;
517 
518 	pr_debug("Validating PGD populate\n");
519 	/*
520 	 * This entry points to next level page table page.
521 	 * Hence this must not qualify as pgd_bad().
522 	 */
523 	p4d_clear(p4dp);
524 	pgd_clear(pgdp);
525 	pgd_populate(mm, pgdp, p4dp);
526 	pgd = READ_ONCE(*pgdp);
527 	WARN_ON(pgd_bad(pgd));
528 }
529 #else  /* !__PAGETABLE_P4D_FOLDED */
530 static void __init p4d_clear_tests(struct mm_struct *mm, p4d_t *p4dp) { }
531 static void __init pgd_clear_tests(struct mm_struct *mm, pgd_t *pgdp) { }
532 static void __init p4d_populate_tests(struct mm_struct *mm, p4d_t *p4dp,
533 				      pud_t *pudp)
534 {
535 }
536 static void __init pgd_populate_tests(struct mm_struct *mm, pgd_t *pgdp,
537 				      p4d_t *p4dp)
538 {
539 }
540 #endif /* PAGETABLE_P4D_FOLDED */
541 
542 static void __init pte_clear_tests(struct mm_struct *mm, pte_t *ptep,
543 				   unsigned long pfn, unsigned long vaddr,
544 				   pgprot_t prot)
545 {
546 	pte_t pte = pfn_pte(pfn, prot);
547 
548 	pr_debug("Validating PTE clear\n");
549 #ifndef CONFIG_RISCV
550 	pte = __pte(pte_val(pte) | RANDOM_ORVALUE);
551 #endif
552 	set_pte_at(mm, vaddr, ptep, pte);
553 	barrier();
554 	pte_clear(mm, vaddr, ptep);
555 	pte = ptep_get(ptep);
556 	WARN_ON(!pte_none(pte));
557 }
558 
559 static void __init pmd_clear_tests(struct mm_struct *mm, pmd_t *pmdp)
560 {
561 	pmd_t pmd = READ_ONCE(*pmdp);
562 
563 	pr_debug("Validating PMD clear\n");
564 	pmd = __pmd(pmd_val(pmd) | RANDOM_ORVALUE);
565 	WRITE_ONCE(*pmdp, pmd);
566 	pmd_clear(pmdp);
567 	pmd = READ_ONCE(*pmdp);
568 	WARN_ON(!pmd_none(pmd));
569 }
570 
571 static void __init pmd_populate_tests(struct mm_struct *mm, pmd_t *pmdp,
572 				      pgtable_t pgtable)
573 {
574 	pmd_t pmd;
575 
576 	pr_debug("Validating PMD populate\n");
577 	/*
578 	 * This entry points to next level page table page.
579 	 * Hence this must not qualify as pmd_bad().
580 	 */
581 	pmd_populate(mm, pmdp, pgtable);
582 	pmd = READ_ONCE(*pmdp);
583 	WARN_ON(pmd_bad(pmd));
584 }
585 
586 static void __init pte_special_tests(unsigned long pfn, pgprot_t prot)
587 {
588 	pte_t pte = pfn_pte(pfn, prot);
589 
590 	if (!IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL))
591 		return;
592 
593 	pr_debug("Validating PTE special\n");
594 	WARN_ON(!pte_special(pte_mkspecial(pte)));
595 }
596 
597 static void __init pte_protnone_tests(unsigned long pfn, pgprot_t prot)
598 {
599 	pte_t pte = pfn_pte(pfn, prot);
600 
601 	if (!IS_ENABLED(CONFIG_NUMA_BALANCING))
602 		return;
603 
604 	pr_debug("Validating PTE protnone\n");
605 	WARN_ON(!pte_protnone(pte));
606 	WARN_ON(!pte_present(pte));
607 }
608 
609 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
610 static void __init pmd_protnone_tests(unsigned long pfn, pgprot_t prot)
611 {
612 	pmd_t pmd = pmd_mkhuge(pfn_pmd(pfn, prot));
613 
614 	if (!IS_ENABLED(CONFIG_NUMA_BALANCING))
615 		return;
616 
617 	pr_debug("Validating PMD protnone\n");
618 	WARN_ON(!pmd_protnone(pmd));
619 	WARN_ON(!pmd_present(pmd));
620 }
621 #else  /* !CONFIG_TRANSPARENT_HUGEPAGE */
622 static void __init pmd_protnone_tests(unsigned long pfn, pgprot_t prot) { }
623 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
624 
625 #ifdef CONFIG_ARCH_HAS_PTE_DEVMAP
626 static void __init pte_devmap_tests(unsigned long pfn, pgprot_t prot)
627 {
628 	pte_t pte = pfn_pte(pfn, prot);
629 
630 	pr_debug("Validating PTE devmap\n");
631 	WARN_ON(!pte_devmap(pte_mkdevmap(pte)));
632 }
633 
634 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
635 static void __init pmd_devmap_tests(unsigned long pfn, pgprot_t prot)
636 {
637 	pmd_t pmd = pfn_pmd(pfn, prot);
638 
639 	pr_debug("Validating PMD devmap\n");
640 	WARN_ON(!pmd_devmap(pmd_mkdevmap(pmd)));
641 }
642 
643 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
644 static void __init pud_devmap_tests(unsigned long pfn, pgprot_t prot)
645 {
646 	pud_t pud = pfn_pud(pfn, prot);
647 
648 	pr_debug("Validating PUD devmap\n");
649 	WARN_ON(!pud_devmap(pud_mkdevmap(pud)));
650 }
651 #else  /* !CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
652 static void __init pud_devmap_tests(unsigned long pfn, pgprot_t prot) { }
653 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
654 #else  /* CONFIG_TRANSPARENT_HUGEPAGE */
655 static void __init pmd_devmap_tests(unsigned long pfn, pgprot_t prot) { }
656 static void __init pud_devmap_tests(unsigned long pfn, pgprot_t prot) { }
657 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
658 #else
659 static void __init pte_devmap_tests(unsigned long pfn, pgprot_t prot) { }
660 static void __init pmd_devmap_tests(unsigned long pfn, pgprot_t prot) { }
661 static void __init pud_devmap_tests(unsigned long pfn, pgprot_t prot) { }
662 #endif /* CONFIG_ARCH_HAS_PTE_DEVMAP */
663 
664 static void __init pte_soft_dirty_tests(unsigned long pfn, pgprot_t prot)
665 {
666 	pte_t pte = pfn_pte(pfn, prot);
667 
668 	if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY))
669 		return;
670 
671 	pr_debug("Validating PTE soft dirty\n");
672 	WARN_ON(!pte_soft_dirty(pte_mksoft_dirty(pte)));
673 	WARN_ON(pte_soft_dirty(pte_clear_soft_dirty(pte)));
674 }
675 
676 static void __init pte_swap_soft_dirty_tests(unsigned long pfn, pgprot_t prot)
677 {
678 	pte_t pte = pfn_pte(pfn, prot);
679 
680 	if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY))
681 		return;
682 
683 	pr_debug("Validating PTE swap soft dirty\n");
684 	WARN_ON(!pte_swp_soft_dirty(pte_swp_mksoft_dirty(pte)));
685 	WARN_ON(pte_swp_soft_dirty(pte_swp_clear_soft_dirty(pte)));
686 }
687 
688 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
689 static void __init pmd_soft_dirty_tests(unsigned long pfn, pgprot_t prot)
690 {
691 	pmd_t pmd = pfn_pmd(pfn, prot);
692 
693 	if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY))
694 		return;
695 
696 	pr_debug("Validating PMD soft dirty\n");
697 	WARN_ON(!pmd_soft_dirty(pmd_mksoft_dirty(pmd)));
698 	WARN_ON(pmd_soft_dirty(pmd_clear_soft_dirty(pmd)));
699 }
700 
701 static void __init pmd_swap_soft_dirty_tests(unsigned long pfn, pgprot_t prot)
702 {
703 	pmd_t pmd = pfn_pmd(pfn, prot);
704 
705 	if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) ||
706 		!IS_ENABLED(CONFIG_ARCH_ENABLE_THP_MIGRATION))
707 		return;
708 
709 	pr_debug("Validating PMD swap soft dirty\n");
710 	WARN_ON(!pmd_swp_soft_dirty(pmd_swp_mksoft_dirty(pmd)));
711 	WARN_ON(pmd_swp_soft_dirty(pmd_swp_clear_soft_dirty(pmd)));
712 }
713 #else  /* !CONFIG_ARCH_HAS_PTE_DEVMAP */
714 static void __init pmd_soft_dirty_tests(unsigned long pfn, pgprot_t prot) { }
715 static void __init pmd_swap_soft_dirty_tests(unsigned long pfn, pgprot_t prot)
716 {
717 }
718 #endif /* CONFIG_ARCH_HAS_PTE_DEVMAP */
719 
720 static void __init pte_swap_tests(unsigned long pfn, pgprot_t prot)
721 {
722 	swp_entry_t swp;
723 	pte_t pte;
724 
725 	pr_debug("Validating PTE swap\n");
726 	pte = pfn_pte(pfn, prot);
727 	swp = __pte_to_swp_entry(pte);
728 	pte = __swp_entry_to_pte(swp);
729 	WARN_ON(pfn != pte_pfn(pte));
730 }
731 
732 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
733 static void __init pmd_swap_tests(unsigned long pfn, pgprot_t prot)
734 {
735 	swp_entry_t swp;
736 	pmd_t pmd;
737 
738 	pr_debug("Validating PMD swap\n");
739 	pmd = pfn_pmd(pfn, prot);
740 	swp = __pmd_to_swp_entry(pmd);
741 	pmd = __swp_entry_to_pmd(swp);
742 	WARN_ON(pfn != pmd_pfn(pmd));
743 }
744 #else  /* !CONFIG_ARCH_ENABLE_THP_MIGRATION */
745 static void __init pmd_swap_tests(unsigned long pfn, pgprot_t prot) { }
746 #endif /* CONFIG_ARCH_ENABLE_THP_MIGRATION */
747 
748 static void __init swap_migration_tests(void)
749 {
750 	struct page *page;
751 	swp_entry_t swp;
752 
753 	if (!IS_ENABLED(CONFIG_MIGRATION))
754 		return;
755 
756 	pr_debug("Validating swap migration\n");
757 	/*
758 	 * swap_migration_tests() requires a dedicated page as it needs to
759 	 * be locked before creating a migration entry from it. Locking the
760 	 * page that actually maps kernel text ('start_kernel') can be real
761 	 * problematic. Lets allocate a dedicated page explicitly for this
762 	 * purpose that will be freed subsequently.
763 	 */
764 	page = alloc_page(GFP_KERNEL);
765 	if (!page) {
766 		pr_err("page allocation failed\n");
767 		return;
768 	}
769 
770 	/*
771 	 * make_migration_entry() expects given page to be
772 	 * locked, otherwise it stumbles upon a BUG_ON().
773 	 */
774 	__SetPageLocked(page);
775 	swp = make_migration_entry(page, 1);
776 	WARN_ON(!is_migration_entry(swp));
777 	WARN_ON(!is_write_migration_entry(swp));
778 
779 	make_migration_entry_read(&swp);
780 	WARN_ON(!is_migration_entry(swp));
781 	WARN_ON(is_write_migration_entry(swp));
782 
783 	swp = make_migration_entry(page, 0);
784 	WARN_ON(!is_migration_entry(swp));
785 	WARN_ON(is_write_migration_entry(swp));
786 	__ClearPageLocked(page);
787 	__free_page(page);
788 }
789 
790 #ifdef CONFIG_HUGETLB_PAGE
791 static void __init hugetlb_basic_tests(unsigned long pfn, pgprot_t prot)
792 {
793 	struct page *page;
794 	pte_t pte;
795 
796 	pr_debug("Validating HugeTLB basic\n");
797 	/*
798 	 * Accessing the page associated with the pfn is safe here,
799 	 * as it was previously derived from a real kernel symbol.
800 	 */
801 	page = pfn_to_page(pfn);
802 	pte = mk_huge_pte(page, prot);
803 
804 	WARN_ON(!huge_pte_dirty(huge_pte_mkdirty(pte)));
805 	WARN_ON(!huge_pte_write(huge_pte_mkwrite(huge_pte_wrprotect(pte))));
806 	WARN_ON(huge_pte_write(huge_pte_wrprotect(huge_pte_mkwrite(pte))));
807 
808 #ifdef CONFIG_ARCH_WANT_GENERAL_HUGETLB
809 	pte = pfn_pte(pfn, prot);
810 
811 	WARN_ON(!pte_huge(pte_mkhuge(pte)));
812 #endif /* CONFIG_ARCH_WANT_GENERAL_HUGETLB */
813 }
814 #else  /* !CONFIG_HUGETLB_PAGE */
815 static void __init hugetlb_basic_tests(unsigned long pfn, pgprot_t prot) { }
816 #endif /* CONFIG_HUGETLB_PAGE */
817 
818 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
819 static void __init pmd_thp_tests(unsigned long pfn, pgprot_t prot)
820 {
821 	pmd_t pmd;
822 
823 	if (!has_transparent_hugepage())
824 		return;
825 
826 	pr_debug("Validating PMD based THP\n");
827 	/*
828 	 * pmd_trans_huge() and pmd_present() must return positive after
829 	 * MMU invalidation with pmd_mkinvalid(). This behavior is an
830 	 * optimization for transparent huge page. pmd_trans_huge() must
831 	 * be true if pmd_page() returns a valid THP to avoid taking the
832 	 * pmd_lock when others walk over non transhuge pmds (i.e. there
833 	 * are no THP allocated). Especially when splitting a THP and
834 	 * removing the present bit from the pmd, pmd_trans_huge() still
835 	 * needs to return true. pmd_present() should be true whenever
836 	 * pmd_trans_huge() returns true.
837 	 */
838 	pmd = pfn_pmd(pfn, prot);
839 	WARN_ON(!pmd_trans_huge(pmd_mkhuge(pmd)));
840 
841 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
842 	WARN_ON(!pmd_trans_huge(pmd_mkinvalid(pmd_mkhuge(pmd))));
843 	WARN_ON(!pmd_present(pmd_mkinvalid(pmd_mkhuge(pmd))));
844 #endif /* __HAVE_ARCH_PMDP_INVALIDATE */
845 }
846 
847 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
848 static void __init pud_thp_tests(unsigned long pfn, pgprot_t prot)
849 {
850 	pud_t pud;
851 
852 	if (!has_transparent_hugepage())
853 		return;
854 
855 	pr_debug("Validating PUD based THP\n");
856 	pud = pfn_pud(pfn, prot);
857 	WARN_ON(!pud_trans_huge(pud_mkhuge(pud)));
858 
859 	/*
860 	 * pud_mkinvalid() has been dropped for now. Enable back
861 	 * these tests when it comes back with a modified pud_present().
862 	 *
863 	 * WARN_ON(!pud_trans_huge(pud_mkinvalid(pud_mkhuge(pud))));
864 	 * WARN_ON(!pud_present(pud_mkinvalid(pud_mkhuge(pud))));
865 	 */
866 }
867 #else  /* !CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
868 static void __init pud_thp_tests(unsigned long pfn, pgprot_t prot) { }
869 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
870 #else  /* !CONFIG_TRANSPARENT_HUGEPAGE */
871 static void __init pmd_thp_tests(unsigned long pfn, pgprot_t prot) { }
872 static void __init pud_thp_tests(unsigned long pfn, pgprot_t prot) { }
873 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
874 
875 static unsigned long __init get_random_vaddr(void)
876 {
877 	unsigned long random_vaddr, random_pages, total_user_pages;
878 
879 	total_user_pages = (TASK_SIZE - FIRST_USER_ADDRESS) / PAGE_SIZE;
880 
881 	random_pages = get_random_long() % total_user_pages;
882 	random_vaddr = FIRST_USER_ADDRESS + random_pages * PAGE_SIZE;
883 
884 	return random_vaddr;
885 }
886 
887 static int __init debug_vm_pgtable(void)
888 {
889 	struct vm_area_struct *vma;
890 	struct mm_struct *mm;
891 	pgd_t *pgdp;
892 	p4d_t *p4dp, *saved_p4dp;
893 	pud_t *pudp, *saved_pudp;
894 	pmd_t *pmdp, *saved_pmdp, pmd;
895 	pte_t *ptep;
896 	pgtable_t saved_ptep;
897 	pgprot_t prot, protnone;
898 	phys_addr_t paddr;
899 	unsigned long vaddr, pte_aligned, pmd_aligned;
900 	unsigned long pud_aligned, p4d_aligned, pgd_aligned;
901 	spinlock_t *ptl = NULL;
902 
903 	pr_info("Validating architecture page table helpers\n");
904 	prot = vm_get_page_prot(VMFLAGS);
905 	vaddr = get_random_vaddr();
906 	mm = mm_alloc();
907 	if (!mm) {
908 		pr_err("mm_struct allocation failed\n");
909 		return 1;
910 	}
911 
912 	/*
913 	 * __P000 (or even __S000) will help create page table entries with
914 	 * PROT_NONE permission as required for pxx_protnone_tests().
915 	 */
916 	protnone = __P000;
917 
918 	vma = vm_area_alloc(mm);
919 	if (!vma) {
920 		pr_err("vma allocation failed\n");
921 		return 1;
922 	}
923 
924 	/*
925 	 * PFN for mapping at PTE level is determined from a standard kernel
926 	 * text symbol. But pfns for higher page table levels are derived by
927 	 * masking lower bits of this real pfn. These derived pfns might not
928 	 * exist on the platform but that does not really matter as pfn_pxx()
929 	 * helpers will still create appropriate entries for the test. This
930 	 * helps avoid large memory block allocations to be used for mapping
931 	 * at higher page table levels.
932 	 */
933 	paddr = __pa_symbol(&start_kernel);
934 
935 	pte_aligned = (paddr & PAGE_MASK) >> PAGE_SHIFT;
936 	pmd_aligned = (paddr & PMD_MASK) >> PAGE_SHIFT;
937 	pud_aligned = (paddr & PUD_MASK) >> PAGE_SHIFT;
938 	p4d_aligned = (paddr & P4D_MASK) >> PAGE_SHIFT;
939 	pgd_aligned = (paddr & PGDIR_MASK) >> PAGE_SHIFT;
940 	WARN_ON(!pfn_valid(pte_aligned));
941 
942 	pgdp = pgd_offset(mm, vaddr);
943 	p4dp = p4d_alloc(mm, pgdp, vaddr);
944 	pudp = pud_alloc(mm, p4dp, vaddr);
945 	pmdp = pmd_alloc(mm, pudp, vaddr);
946 	/*
947 	 * Allocate pgtable_t
948 	 */
949 	if (pte_alloc(mm, pmdp)) {
950 		pr_err("pgtable allocation failed\n");
951 		return 1;
952 	}
953 
954 	/*
955 	 * Save all the page table page addresses as the page table
956 	 * entries will be used for testing with random or garbage
957 	 * values. These saved addresses will be used for freeing
958 	 * page table pages.
959 	 */
960 	pmd = READ_ONCE(*pmdp);
961 	saved_p4dp = p4d_offset(pgdp, 0UL);
962 	saved_pudp = pud_offset(p4dp, 0UL);
963 	saved_pmdp = pmd_offset(pudp, 0UL);
964 	saved_ptep = pmd_pgtable(pmd);
965 
966 	pte_basic_tests(pte_aligned, prot);
967 	pmd_basic_tests(pmd_aligned, prot);
968 	pud_basic_tests(pud_aligned, prot);
969 	p4d_basic_tests(p4d_aligned, prot);
970 	pgd_basic_tests(pgd_aligned, prot);
971 
972 	pmd_leaf_tests(pmd_aligned, prot);
973 	pud_leaf_tests(pud_aligned, prot);
974 
975 	pte_savedwrite_tests(pte_aligned, protnone);
976 	pmd_savedwrite_tests(pmd_aligned, protnone);
977 
978 	pte_special_tests(pte_aligned, prot);
979 	pte_protnone_tests(pte_aligned, protnone);
980 	pmd_protnone_tests(pmd_aligned, protnone);
981 
982 	pte_devmap_tests(pte_aligned, prot);
983 	pmd_devmap_tests(pmd_aligned, prot);
984 	pud_devmap_tests(pud_aligned, prot);
985 
986 	pte_soft_dirty_tests(pte_aligned, prot);
987 	pmd_soft_dirty_tests(pmd_aligned, prot);
988 	pte_swap_soft_dirty_tests(pte_aligned, prot);
989 	pmd_swap_soft_dirty_tests(pmd_aligned, prot);
990 
991 	pte_swap_tests(pte_aligned, prot);
992 	pmd_swap_tests(pmd_aligned, prot);
993 
994 	swap_migration_tests();
995 
996 	pmd_thp_tests(pmd_aligned, prot);
997 	pud_thp_tests(pud_aligned, prot);
998 
999 	hugetlb_basic_tests(pte_aligned, prot);
1000 
1001 	/*
1002 	 * Page table modifying tests. They need to hold
1003 	 * proper page table lock.
1004 	 */
1005 
1006 	ptep = pte_offset_map_lock(mm, pmdp, vaddr, &ptl);
1007 	pte_clear_tests(mm, ptep, pte_aligned, vaddr, prot);
1008 	pte_advanced_tests(mm, vma, ptep, pte_aligned, vaddr, prot);
1009 	pte_unmap_unlock(ptep, ptl);
1010 
1011 	ptl = pmd_lock(mm, pmdp);
1012 	pmd_clear_tests(mm, pmdp);
1013 	pmd_advanced_tests(mm, vma, pmdp, pmd_aligned, vaddr, prot, saved_ptep);
1014 	pmd_huge_tests(pmdp, pmd_aligned, prot);
1015 	pmd_populate_tests(mm, pmdp, saved_ptep);
1016 	spin_unlock(ptl);
1017 
1018 	ptl = pud_lock(mm, pudp);
1019 	pud_clear_tests(mm, pudp);
1020 	pud_advanced_tests(mm, vma, pudp, pud_aligned, vaddr, prot);
1021 	pud_huge_tests(pudp, pud_aligned, prot);
1022 	pud_populate_tests(mm, pudp, saved_pmdp);
1023 	spin_unlock(ptl);
1024 
1025 	spin_lock(&mm->page_table_lock);
1026 	p4d_clear_tests(mm, p4dp);
1027 	pgd_clear_tests(mm, pgdp);
1028 	p4d_populate_tests(mm, p4dp, saved_pudp);
1029 	pgd_populate_tests(mm, pgdp, saved_p4dp);
1030 	spin_unlock(&mm->page_table_lock);
1031 
1032 	p4d_free(mm, saved_p4dp);
1033 	pud_free(mm, saved_pudp);
1034 	pmd_free(mm, saved_pmdp);
1035 	pte_free(mm, saved_ptep);
1036 
1037 	vm_area_free(vma);
1038 	mm_dec_nr_puds(mm);
1039 	mm_dec_nr_pmds(mm);
1040 	mm_dec_nr_ptes(mm);
1041 	mmdrop(mm);
1042 	return 0;
1043 }
1044 late_initcall(debug_vm_pgtable);
1045