xref: /linux/mm/debug_vm_pgtable.c (revision 24168c5e6dfbdd5b414f048f47f75d64533296ca)
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 #include <linux/vmalloc.h>
34 
35 #include <asm/cacheflush.h>
36 #include <asm/pgalloc.h>
37 #include <asm/tlbflush.h>
38 
39 /*
40  * Please refer Documentation/mm/arch_pgtable_helpers.rst for the semantics
41  * expectations that are being validated here. All future changes in here
42  * or the documentation need to be in sync.
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 struct pgtable_debug_args {
62 	struct mm_struct	*mm;
63 	struct vm_area_struct	*vma;
64 
65 	pgd_t			*pgdp;
66 	p4d_t			*p4dp;
67 	pud_t			*pudp;
68 	pmd_t			*pmdp;
69 	pte_t			*ptep;
70 
71 	p4d_t			*start_p4dp;
72 	pud_t			*start_pudp;
73 	pmd_t			*start_pmdp;
74 	pgtable_t		start_ptep;
75 
76 	unsigned long		vaddr;
77 	pgprot_t		page_prot;
78 	pgprot_t		page_prot_none;
79 
80 	bool			is_contiguous_page;
81 	unsigned long		pud_pfn;
82 	unsigned long		pmd_pfn;
83 	unsigned long		pte_pfn;
84 
85 	unsigned long		fixed_alignment;
86 	unsigned long		fixed_pgd_pfn;
87 	unsigned long		fixed_p4d_pfn;
88 	unsigned long		fixed_pud_pfn;
89 	unsigned long		fixed_pmd_pfn;
90 	unsigned long		fixed_pte_pfn;
91 };
92 
93 static void __init pte_basic_tests(struct pgtable_debug_args *args, int idx)
94 {
95 	pgprot_t prot = vm_get_page_prot(idx);
96 	pte_t pte = pfn_pte(args->fixed_pte_pfn, prot);
97 	unsigned long val = idx, *ptr = &val;
98 
99 	pr_debug("Validating PTE basic (%pGv)\n", ptr);
100 
101 	/*
102 	 * This test needs to be executed after the given page table entry
103 	 * is created with pfn_pte() to make sure that vm_get_page_prot(idx)
104 	 * does not have the dirty bit enabled from the beginning. This is
105 	 * important for platforms like arm64 where (!PTE_RDONLY) indicate
106 	 * dirty bit being set.
107 	 */
108 	WARN_ON(pte_dirty(pte_wrprotect(pte)));
109 
110 	WARN_ON(!pte_same(pte, pte));
111 	WARN_ON(!pte_young(pte_mkyoung(pte_mkold(pte))));
112 	WARN_ON(!pte_dirty(pte_mkdirty(pte_mkclean(pte))));
113 	WARN_ON(!pte_write(pte_mkwrite(pte_wrprotect(pte), args->vma)));
114 	WARN_ON(pte_young(pte_mkold(pte_mkyoung(pte))));
115 	WARN_ON(pte_dirty(pte_mkclean(pte_mkdirty(pte))));
116 	WARN_ON(pte_write(pte_wrprotect(pte_mkwrite(pte, args->vma))));
117 	WARN_ON(pte_dirty(pte_wrprotect(pte_mkclean(pte))));
118 	WARN_ON(!pte_dirty(pte_wrprotect(pte_mkdirty(pte))));
119 }
120 
121 static void __init pte_advanced_tests(struct pgtable_debug_args *args)
122 {
123 	struct page *page;
124 	pte_t pte;
125 
126 	/*
127 	 * Architectures optimize set_pte_at by avoiding TLB flush.
128 	 * This requires set_pte_at to be not used to update an
129 	 * existing pte entry. Clear pte before we do set_pte_at
130 	 *
131 	 * flush_dcache_page() is called after set_pte_at() to clear
132 	 * PG_arch_1 for the page on ARM64. The page flag isn't cleared
133 	 * when it's released and page allocation check will fail when
134 	 * the page is allocated again. For architectures other than ARM64,
135 	 * the unexpected overhead of cache flushing is acceptable.
136 	 */
137 	page = (args->pte_pfn != ULONG_MAX) ? pfn_to_page(args->pte_pfn) : NULL;
138 	if (!page)
139 		return;
140 
141 	pr_debug("Validating PTE advanced\n");
142 	if (WARN_ON(!args->ptep))
143 		return;
144 
145 	pte = pfn_pte(args->pte_pfn, args->page_prot);
146 	set_pte_at(args->mm, args->vaddr, args->ptep, pte);
147 	flush_dcache_page(page);
148 	ptep_set_wrprotect(args->mm, args->vaddr, args->ptep);
149 	pte = ptep_get(args->ptep);
150 	WARN_ON(pte_write(pte));
151 	ptep_get_and_clear(args->mm, args->vaddr, args->ptep);
152 	pte = ptep_get(args->ptep);
153 	WARN_ON(!pte_none(pte));
154 
155 	pte = pfn_pte(args->pte_pfn, args->page_prot);
156 	pte = pte_wrprotect(pte);
157 	pte = pte_mkclean(pte);
158 	set_pte_at(args->mm, args->vaddr, args->ptep, pte);
159 	flush_dcache_page(page);
160 	pte = pte_mkwrite(pte, args->vma);
161 	pte = pte_mkdirty(pte);
162 	ptep_set_access_flags(args->vma, args->vaddr, args->ptep, pte, 1);
163 	pte = ptep_get(args->ptep);
164 	WARN_ON(!(pte_write(pte) && pte_dirty(pte)));
165 	ptep_get_and_clear_full(args->mm, args->vaddr, args->ptep, 1);
166 	pte = ptep_get(args->ptep);
167 	WARN_ON(!pte_none(pte));
168 
169 	pte = pfn_pte(args->pte_pfn, args->page_prot);
170 	pte = pte_mkyoung(pte);
171 	set_pte_at(args->mm, args->vaddr, args->ptep, pte);
172 	flush_dcache_page(page);
173 	ptep_test_and_clear_young(args->vma, args->vaddr, args->ptep);
174 	pte = ptep_get(args->ptep);
175 	WARN_ON(pte_young(pte));
176 
177 	ptep_get_and_clear_full(args->mm, args->vaddr, args->ptep, 1);
178 }
179 
180 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
181 static void __init pmd_basic_tests(struct pgtable_debug_args *args, int idx)
182 {
183 	pgprot_t prot = vm_get_page_prot(idx);
184 	unsigned long val = idx, *ptr = &val;
185 	pmd_t pmd;
186 
187 	if (!has_transparent_hugepage())
188 		return;
189 
190 	pr_debug("Validating PMD basic (%pGv)\n", ptr);
191 	pmd = pfn_pmd(args->fixed_pmd_pfn, prot);
192 
193 	/*
194 	 * This test needs to be executed after the given page table entry
195 	 * is created with pfn_pmd() to make sure that vm_get_page_prot(idx)
196 	 * does not have the dirty bit enabled from the beginning. This is
197 	 * important for platforms like arm64 where (!PTE_RDONLY) indicate
198 	 * dirty bit being set.
199 	 */
200 	WARN_ON(pmd_dirty(pmd_wrprotect(pmd)));
201 
202 
203 	WARN_ON(!pmd_same(pmd, pmd));
204 	WARN_ON(!pmd_young(pmd_mkyoung(pmd_mkold(pmd))));
205 	WARN_ON(!pmd_dirty(pmd_mkdirty(pmd_mkclean(pmd))));
206 	WARN_ON(!pmd_write(pmd_mkwrite(pmd_wrprotect(pmd), args->vma)));
207 	WARN_ON(pmd_young(pmd_mkold(pmd_mkyoung(pmd))));
208 	WARN_ON(pmd_dirty(pmd_mkclean(pmd_mkdirty(pmd))));
209 	WARN_ON(pmd_write(pmd_wrprotect(pmd_mkwrite(pmd, args->vma))));
210 	WARN_ON(pmd_dirty(pmd_wrprotect(pmd_mkclean(pmd))));
211 	WARN_ON(!pmd_dirty(pmd_wrprotect(pmd_mkdirty(pmd))));
212 	/*
213 	 * A huge page does not point to next level page table
214 	 * entry. Hence this must qualify as pmd_bad().
215 	 */
216 	WARN_ON(!pmd_bad(pmd_mkhuge(pmd)));
217 }
218 
219 static void __init pmd_advanced_tests(struct pgtable_debug_args *args)
220 {
221 	struct page *page;
222 	pmd_t pmd;
223 	unsigned long vaddr = args->vaddr;
224 
225 	if (!has_transparent_hugepage())
226 		return;
227 
228 	page = (args->pmd_pfn != ULONG_MAX) ? pfn_to_page(args->pmd_pfn) : NULL;
229 	if (!page)
230 		return;
231 
232 	/*
233 	 * flush_dcache_page() is called after set_pmd_at() to clear
234 	 * PG_arch_1 for the page on ARM64. The page flag isn't cleared
235 	 * when it's released and page allocation check will fail when
236 	 * the page is allocated again. For architectures other than ARM64,
237 	 * the unexpected overhead of cache flushing is acceptable.
238 	 */
239 	pr_debug("Validating PMD advanced\n");
240 	/* Align the address wrt HPAGE_PMD_SIZE */
241 	vaddr &= HPAGE_PMD_MASK;
242 
243 	pgtable_trans_huge_deposit(args->mm, args->pmdp, args->start_ptep);
244 
245 	pmd = pfn_pmd(args->pmd_pfn, args->page_prot);
246 	set_pmd_at(args->mm, vaddr, args->pmdp, pmd);
247 	flush_dcache_page(page);
248 	pmdp_set_wrprotect(args->mm, vaddr, args->pmdp);
249 	pmd = READ_ONCE(*args->pmdp);
250 	WARN_ON(pmd_write(pmd));
251 	pmdp_huge_get_and_clear(args->mm, vaddr, args->pmdp);
252 	pmd = READ_ONCE(*args->pmdp);
253 	WARN_ON(!pmd_none(pmd));
254 
255 	pmd = pfn_pmd(args->pmd_pfn, args->page_prot);
256 	pmd = pmd_wrprotect(pmd);
257 	pmd = pmd_mkclean(pmd);
258 	set_pmd_at(args->mm, vaddr, args->pmdp, pmd);
259 	flush_dcache_page(page);
260 	pmd = pmd_mkwrite(pmd, args->vma);
261 	pmd = pmd_mkdirty(pmd);
262 	pmdp_set_access_flags(args->vma, vaddr, args->pmdp, pmd, 1);
263 	pmd = READ_ONCE(*args->pmdp);
264 	WARN_ON(!(pmd_write(pmd) && pmd_dirty(pmd)));
265 	pmdp_huge_get_and_clear_full(args->vma, vaddr, args->pmdp, 1);
266 	pmd = READ_ONCE(*args->pmdp);
267 	WARN_ON(!pmd_none(pmd));
268 
269 	pmd = pmd_mkhuge(pfn_pmd(args->pmd_pfn, args->page_prot));
270 	pmd = pmd_mkyoung(pmd);
271 	set_pmd_at(args->mm, vaddr, args->pmdp, pmd);
272 	flush_dcache_page(page);
273 	pmdp_test_and_clear_young(args->vma, vaddr, args->pmdp);
274 	pmd = READ_ONCE(*args->pmdp);
275 	WARN_ON(pmd_young(pmd));
276 
277 	/*  Clear the pte entries  */
278 	pmdp_huge_get_and_clear(args->mm, vaddr, args->pmdp);
279 	pgtable_trans_huge_withdraw(args->mm, args->pmdp);
280 }
281 
282 static void __init pmd_leaf_tests(struct pgtable_debug_args *args)
283 {
284 	pmd_t pmd;
285 
286 	if (!has_transparent_hugepage())
287 		return;
288 
289 	pr_debug("Validating PMD leaf\n");
290 	pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot);
291 
292 	/*
293 	 * PMD based THP is a leaf entry.
294 	 */
295 	pmd = pmd_mkhuge(pmd);
296 	WARN_ON(!pmd_leaf(pmd));
297 }
298 
299 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
300 static void __init pud_basic_tests(struct pgtable_debug_args *args, int idx)
301 {
302 	pgprot_t prot = vm_get_page_prot(idx);
303 	unsigned long val = idx, *ptr = &val;
304 	pud_t pud;
305 
306 	if (!has_transparent_pud_hugepage())
307 		return;
308 
309 	pr_debug("Validating PUD basic (%pGv)\n", ptr);
310 	pud = pfn_pud(args->fixed_pud_pfn, prot);
311 
312 	/*
313 	 * This test needs to be executed after the given page table entry
314 	 * is created with pfn_pud() to make sure that vm_get_page_prot(idx)
315 	 * does not have the dirty bit enabled from the beginning. This is
316 	 * important for platforms like arm64 where (!PTE_RDONLY) indicate
317 	 * dirty bit being set.
318 	 */
319 	WARN_ON(pud_dirty(pud_wrprotect(pud)));
320 
321 	WARN_ON(!pud_same(pud, pud));
322 	WARN_ON(!pud_young(pud_mkyoung(pud_mkold(pud))));
323 	WARN_ON(!pud_dirty(pud_mkdirty(pud_mkclean(pud))));
324 	WARN_ON(pud_dirty(pud_mkclean(pud_mkdirty(pud))));
325 	WARN_ON(!pud_write(pud_mkwrite(pud_wrprotect(pud))));
326 	WARN_ON(pud_write(pud_wrprotect(pud_mkwrite(pud))));
327 	WARN_ON(pud_young(pud_mkold(pud_mkyoung(pud))));
328 	WARN_ON(pud_dirty(pud_wrprotect(pud_mkclean(pud))));
329 	WARN_ON(!pud_dirty(pud_wrprotect(pud_mkdirty(pud))));
330 
331 	if (mm_pmd_folded(args->mm))
332 		return;
333 
334 	/*
335 	 * A huge page does not point to next level page table
336 	 * entry. Hence this must qualify as pud_bad().
337 	 */
338 	WARN_ON(!pud_bad(pud_mkhuge(pud)));
339 }
340 
341 static void __init pud_advanced_tests(struct pgtable_debug_args *args)
342 {
343 	struct page *page;
344 	unsigned long vaddr = args->vaddr;
345 	pud_t pud;
346 
347 	if (!has_transparent_pud_hugepage())
348 		return;
349 
350 	page = (args->pud_pfn != ULONG_MAX) ? pfn_to_page(args->pud_pfn) : NULL;
351 	if (!page)
352 		return;
353 
354 	/*
355 	 * flush_dcache_page() is called after set_pud_at() to clear
356 	 * PG_arch_1 for the page on ARM64. The page flag isn't cleared
357 	 * when it's released and page allocation check will fail when
358 	 * the page is allocated again. For architectures other than ARM64,
359 	 * the unexpected overhead of cache flushing is acceptable.
360 	 */
361 	pr_debug("Validating PUD advanced\n");
362 	/* Align the address wrt HPAGE_PUD_SIZE */
363 	vaddr &= HPAGE_PUD_MASK;
364 
365 	pud = pfn_pud(args->pud_pfn, args->page_prot);
366 	/*
367 	 * Some architectures have debug checks to make sure
368 	 * huge pud mapping are only found with devmap entries
369 	 * For now test with only devmap entries.
370 	 */
371 	pud = pud_mkdevmap(pud);
372 	set_pud_at(args->mm, vaddr, args->pudp, pud);
373 	flush_dcache_page(page);
374 	pudp_set_wrprotect(args->mm, vaddr, args->pudp);
375 	pud = READ_ONCE(*args->pudp);
376 	WARN_ON(pud_write(pud));
377 
378 #ifndef __PAGETABLE_PMD_FOLDED
379 	pudp_huge_get_and_clear(args->mm, vaddr, args->pudp);
380 	pud = READ_ONCE(*args->pudp);
381 	WARN_ON(!pud_none(pud));
382 #endif /* __PAGETABLE_PMD_FOLDED */
383 	pud = pfn_pud(args->pud_pfn, args->page_prot);
384 	pud = pud_mkdevmap(pud);
385 	pud = pud_wrprotect(pud);
386 	pud = pud_mkclean(pud);
387 	set_pud_at(args->mm, vaddr, args->pudp, pud);
388 	flush_dcache_page(page);
389 	pud = pud_mkwrite(pud);
390 	pud = pud_mkdirty(pud);
391 	pudp_set_access_flags(args->vma, vaddr, args->pudp, pud, 1);
392 	pud = READ_ONCE(*args->pudp);
393 	WARN_ON(!(pud_write(pud) && pud_dirty(pud)));
394 
395 #ifndef __PAGETABLE_PMD_FOLDED
396 	pudp_huge_get_and_clear_full(args->vma, vaddr, args->pudp, 1);
397 	pud = READ_ONCE(*args->pudp);
398 	WARN_ON(!pud_none(pud));
399 #endif /* __PAGETABLE_PMD_FOLDED */
400 
401 	pud = pfn_pud(args->pud_pfn, args->page_prot);
402 	pud = pud_mkdevmap(pud);
403 	pud = pud_mkyoung(pud);
404 	set_pud_at(args->mm, vaddr, args->pudp, pud);
405 	flush_dcache_page(page);
406 	pudp_test_and_clear_young(args->vma, vaddr, args->pudp);
407 	pud = READ_ONCE(*args->pudp);
408 	WARN_ON(pud_young(pud));
409 
410 	pudp_huge_get_and_clear(args->mm, vaddr, args->pudp);
411 }
412 
413 static void __init pud_leaf_tests(struct pgtable_debug_args *args)
414 {
415 	pud_t pud;
416 
417 	if (!has_transparent_pud_hugepage())
418 		return;
419 
420 	pr_debug("Validating PUD leaf\n");
421 	pud = pfn_pud(args->fixed_pud_pfn, args->page_prot);
422 	/*
423 	 * PUD based THP is a leaf entry.
424 	 */
425 	pud = pud_mkhuge(pud);
426 	WARN_ON(!pud_leaf(pud));
427 }
428 #else  /* !CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
429 static void __init pud_basic_tests(struct pgtable_debug_args *args, int idx) { }
430 static void __init pud_advanced_tests(struct pgtable_debug_args *args) { }
431 static void __init pud_leaf_tests(struct pgtable_debug_args *args) { }
432 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
433 #else  /* !CONFIG_TRANSPARENT_HUGEPAGE */
434 static void __init pmd_basic_tests(struct pgtable_debug_args *args, int idx) { }
435 static void __init pud_basic_tests(struct pgtable_debug_args *args, int idx) { }
436 static void __init pmd_advanced_tests(struct pgtable_debug_args *args) { }
437 static void __init pud_advanced_tests(struct pgtable_debug_args *args) { }
438 static void __init pmd_leaf_tests(struct pgtable_debug_args *args) { }
439 static void __init pud_leaf_tests(struct pgtable_debug_args *args) { }
440 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
441 
442 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
443 static void __init pmd_huge_tests(struct pgtable_debug_args *args)
444 {
445 	pmd_t pmd;
446 
447 	if (!arch_vmap_pmd_supported(args->page_prot) ||
448 	    args->fixed_alignment < PMD_SIZE)
449 		return;
450 
451 	pr_debug("Validating PMD huge\n");
452 	/*
453 	 * X86 defined pmd_set_huge() verifies that the given
454 	 * PMD is not a populated non-leaf entry.
455 	 */
456 	WRITE_ONCE(*args->pmdp, __pmd(0));
457 	WARN_ON(!pmd_set_huge(args->pmdp, __pfn_to_phys(args->fixed_pmd_pfn), args->page_prot));
458 	WARN_ON(!pmd_clear_huge(args->pmdp));
459 	pmd = READ_ONCE(*args->pmdp);
460 	WARN_ON(!pmd_none(pmd));
461 }
462 
463 static void __init pud_huge_tests(struct pgtable_debug_args *args)
464 {
465 	pud_t pud;
466 
467 	if (!arch_vmap_pud_supported(args->page_prot) ||
468 	    args->fixed_alignment < PUD_SIZE)
469 		return;
470 
471 	pr_debug("Validating PUD huge\n");
472 	/*
473 	 * X86 defined pud_set_huge() verifies that the given
474 	 * PUD is not a populated non-leaf entry.
475 	 */
476 	WRITE_ONCE(*args->pudp, __pud(0));
477 	WARN_ON(!pud_set_huge(args->pudp, __pfn_to_phys(args->fixed_pud_pfn), args->page_prot));
478 	WARN_ON(!pud_clear_huge(args->pudp));
479 	pud = READ_ONCE(*args->pudp);
480 	WARN_ON(!pud_none(pud));
481 }
482 #else /* !CONFIG_HAVE_ARCH_HUGE_VMAP */
483 static void __init pmd_huge_tests(struct pgtable_debug_args *args) { }
484 static void __init pud_huge_tests(struct pgtable_debug_args *args) { }
485 #endif /* CONFIG_HAVE_ARCH_HUGE_VMAP */
486 
487 static void __init p4d_basic_tests(struct pgtable_debug_args *args)
488 {
489 	p4d_t p4d;
490 
491 	pr_debug("Validating P4D basic\n");
492 	memset(&p4d, RANDOM_NZVALUE, sizeof(p4d_t));
493 	WARN_ON(!p4d_same(p4d, p4d));
494 }
495 
496 static void __init pgd_basic_tests(struct pgtable_debug_args *args)
497 {
498 	pgd_t pgd;
499 
500 	pr_debug("Validating PGD basic\n");
501 	memset(&pgd, RANDOM_NZVALUE, sizeof(pgd_t));
502 	WARN_ON(!pgd_same(pgd, pgd));
503 }
504 
505 #ifndef __PAGETABLE_PUD_FOLDED
506 static void __init pud_clear_tests(struct pgtable_debug_args *args)
507 {
508 	pud_t pud = READ_ONCE(*args->pudp);
509 
510 	if (mm_pmd_folded(args->mm))
511 		return;
512 
513 	pr_debug("Validating PUD clear\n");
514 	pud = __pud(pud_val(pud) | RANDOM_ORVALUE);
515 	WRITE_ONCE(*args->pudp, pud);
516 	pud_clear(args->pudp);
517 	pud = READ_ONCE(*args->pudp);
518 	WARN_ON(!pud_none(pud));
519 }
520 
521 static void __init pud_populate_tests(struct pgtable_debug_args *args)
522 {
523 	pud_t pud;
524 
525 	if (mm_pmd_folded(args->mm))
526 		return;
527 
528 	pr_debug("Validating PUD populate\n");
529 	/*
530 	 * This entry points to next level page table page.
531 	 * Hence this must not qualify as pud_bad().
532 	 */
533 	pud_populate(args->mm, args->pudp, args->start_pmdp);
534 	pud = READ_ONCE(*args->pudp);
535 	WARN_ON(pud_bad(pud));
536 }
537 #else  /* !__PAGETABLE_PUD_FOLDED */
538 static void __init pud_clear_tests(struct pgtable_debug_args *args) { }
539 static void __init pud_populate_tests(struct pgtable_debug_args *args) { }
540 #endif /* PAGETABLE_PUD_FOLDED */
541 
542 #ifndef __PAGETABLE_P4D_FOLDED
543 static void __init p4d_clear_tests(struct pgtable_debug_args *args)
544 {
545 	p4d_t p4d = READ_ONCE(*args->p4dp);
546 
547 	if (mm_pud_folded(args->mm))
548 		return;
549 
550 	pr_debug("Validating P4D clear\n");
551 	p4d = __p4d(p4d_val(p4d) | RANDOM_ORVALUE);
552 	WRITE_ONCE(*args->p4dp, p4d);
553 	p4d_clear(args->p4dp);
554 	p4d = READ_ONCE(*args->p4dp);
555 	WARN_ON(!p4d_none(p4d));
556 }
557 
558 static void __init p4d_populate_tests(struct pgtable_debug_args *args)
559 {
560 	p4d_t p4d;
561 
562 	if (mm_pud_folded(args->mm))
563 		return;
564 
565 	pr_debug("Validating P4D populate\n");
566 	/*
567 	 * This entry points to next level page table page.
568 	 * Hence this must not qualify as p4d_bad().
569 	 */
570 	pud_clear(args->pudp);
571 	p4d_clear(args->p4dp);
572 	p4d_populate(args->mm, args->p4dp, args->start_pudp);
573 	p4d = READ_ONCE(*args->p4dp);
574 	WARN_ON(p4d_bad(p4d));
575 }
576 
577 static void __init pgd_clear_tests(struct pgtable_debug_args *args)
578 {
579 	pgd_t pgd = READ_ONCE(*(args->pgdp));
580 
581 	if (mm_p4d_folded(args->mm))
582 		return;
583 
584 	pr_debug("Validating PGD clear\n");
585 	pgd = __pgd(pgd_val(pgd) | RANDOM_ORVALUE);
586 	WRITE_ONCE(*args->pgdp, pgd);
587 	pgd_clear(args->pgdp);
588 	pgd = READ_ONCE(*args->pgdp);
589 	WARN_ON(!pgd_none(pgd));
590 }
591 
592 static void __init pgd_populate_tests(struct pgtable_debug_args *args)
593 {
594 	pgd_t pgd;
595 
596 	if (mm_p4d_folded(args->mm))
597 		return;
598 
599 	pr_debug("Validating PGD populate\n");
600 	/*
601 	 * This entry points to next level page table page.
602 	 * Hence this must not qualify as pgd_bad().
603 	 */
604 	p4d_clear(args->p4dp);
605 	pgd_clear(args->pgdp);
606 	pgd_populate(args->mm, args->pgdp, args->start_p4dp);
607 	pgd = READ_ONCE(*args->pgdp);
608 	WARN_ON(pgd_bad(pgd));
609 }
610 #else  /* !__PAGETABLE_P4D_FOLDED */
611 static void __init p4d_clear_tests(struct pgtable_debug_args *args) { }
612 static void __init pgd_clear_tests(struct pgtable_debug_args *args) { }
613 static void __init p4d_populate_tests(struct pgtable_debug_args *args) { }
614 static void __init pgd_populate_tests(struct pgtable_debug_args *args) { }
615 #endif /* PAGETABLE_P4D_FOLDED */
616 
617 static void __init pte_clear_tests(struct pgtable_debug_args *args)
618 {
619 	struct page *page;
620 	pte_t pte = pfn_pte(args->pte_pfn, args->page_prot);
621 
622 	page = (args->pte_pfn != ULONG_MAX) ? pfn_to_page(args->pte_pfn) : NULL;
623 	if (!page)
624 		return;
625 
626 	/*
627 	 * flush_dcache_page() is called after set_pte_at() to clear
628 	 * PG_arch_1 for the page on ARM64. The page flag isn't cleared
629 	 * when it's released and page allocation check will fail when
630 	 * the page is allocated again. For architectures other than ARM64,
631 	 * the unexpected overhead of cache flushing is acceptable.
632 	 */
633 	pr_debug("Validating PTE clear\n");
634 	if (WARN_ON(!args->ptep))
635 		return;
636 
637 #ifndef CONFIG_RISCV
638 	pte = __pte(pte_val(pte) | RANDOM_ORVALUE);
639 #endif
640 	set_pte_at(args->mm, args->vaddr, args->ptep, pte);
641 	flush_dcache_page(page);
642 	barrier();
643 	ptep_clear(args->mm, args->vaddr, args->ptep);
644 	pte = ptep_get(args->ptep);
645 	WARN_ON(!pte_none(pte));
646 }
647 
648 static void __init pmd_clear_tests(struct pgtable_debug_args *args)
649 {
650 	pmd_t pmd = READ_ONCE(*args->pmdp);
651 
652 	pr_debug("Validating PMD clear\n");
653 	pmd = __pmd(pmd_val(pmd) | RANDOM_ORVALUE);
654 	WRITE_ONCE(*args->pmdp, pmd);
655 	pmd_clear(args->pmdp);
656 	pmd = READ_ONCE(*args->pmdp);
657 	WARN_ON(!pmd_none(pmd));
658 }
659 
660 static void __init pmd_populate_tests(struct pgtable_debug_args *args)
661 {
662 	pmd_t pmd;
663 
664 	pr_debug("Validating PMD populate\n");
665 	/*
666 	 * This entry points to next level page table page.
667 	 * Hence this must not qualify as pmd_bad().
668 	 */
669 	pmd_populate(args->mm, args->pmdp, args->start_ptep);
670 	pmd = READ_ONCE(*args->pmdp);
671 	WARN_ON(pmd_bad(pmd));
672 }
673 
674 static void __init pte_special_tests(struct pgtable_debug_args *args)
675 {
676 	pte_t pte = pfn_pte(args->fixed_pte_pfn, args->page_prot);
677 
678 	if (!IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL))
679 		return;
680 
681 	pr_debug("Validating PTE special\n");
682 	WARN_ON(!pte_special(pte_mkspecial(pte)));
683 }
684 
685 static void __init pte_protnone_tests(struct pgtable_debug_args *args)
686 {
687 	pte_t pte = pfn_pte(args->fixed_pte_pfn, args->page_prot_none);
688 
689 	if (!IS_ENABLED(CONFIG_NUMA_BALANCING))
690 		return;
691 
692 	pr_debug("Validating PTE protnone\n");
693 	WARN_ON(!pte_protnone(pte));
694 	WARN_ON(!pte_present(pte));
695 }
696 
697 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
698 static void __init pmd_protnone_tests(struct pgtable_debug_args *args)
699 {
700 	pmd_t pmd;
701 
702 	if (!IS_ENABLED(CONFIG_NUMA_BALANCING))
703 		return;
704 
705 	if (!has_transparent_hugepage())
706 		return;
707 
708 	pr_debug("Validating PMD protnone\n");
709 	pmd = pmd_mkhuge(pfn_pmd(args->fixed_pmd_pfn, args->page_prot_none));
710 	WARN_ON(!pmd_protnone(pmd));
711 	WARN_ON(!pmd_present(pmd));
712 }
713 #else  /* !CONFIG_TRANSPARENT_HUGEPAGE */
714 static void __init pmd_protnone_tests(struct pgtable_debug_args *args) { }
715 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
716 
717 #ifdef CONFIG_ARCH_HAS_PTE_DEVMAP
718 static void __init pte_devmap_tests(struct pgtable_debug_args *args)
719 {
720 	pte_t pte = pfn_pte(args->fixed_pte_pfn, args->page_prot);
721 
722 	pr_debug("Validating PTE devmap\n");
723 	WARN_ON(!pte_devmap(pte_mkdevmap(pte)));
724 }
725 
726 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
727 static void __init pmd_devmap_tests(struct pgtable_debug_args *args)
728 {
729 	pmd_t pmd;
730 
731 	if (!has_transparent_hugepage())
732 		return;
733 
734 	pr_debug("Validating PMD devmap\n");
735 	pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot);
736 	WARN_ON(!pmd_devmap(pmd_mkdevmap(pmd)));
737 }
738 
739 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
740 static void __init pud_devmap_tests(struct pgtable_debug_args *args)
741 {
742 	pud_t pud;
743 
744 	if (!has_transparent_pud_hugepage())
745 		return;
746 
747 	pr_debug("Validating PUD devmap\n");
748 	pud = pfn_pud(args->fixed_pud_pfn, args->page_prot);
749 	WARN_ON(!pud_devmap(pud_mkdevmap(pud)));
750 }
751 #else  /* !CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
752 static void __init pud_devmap_tests(struct pgtable_debug_args *args) { }
753 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
754 #else  /* CONFIG_TRANSPARENT_HUGEPAGE */
755 static void __init pmd_devmap_tests(struct pgtable_debug_args *args) { }
756 static void __init pud_devmap_tests(struct pgtable_debug_args *args) { }
757 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
758 #else
759 static void __init pte_devmap_tests(struct pgtable_debug_args *args) { }
760 static void __init pmd_devmap_tests(struct pgtable_debug_args *args) { }
761 static void __init pud_devmap_tests(struct pgtable_debug_args *args) { }
762 #endif /* CONFIG_ARCH_HAS_PTE_DEVMAP */
763 
764 static void __init pte_soft_dirty_tests(struct pgtable_debug_args *args)
765 {
766 	pte_t pte = pfn_pte(args->fixed_pte_pfn, args->page_prot);
767 
768 	if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY))
769 		return;
770 
771 	pr_debug("Validating PTE soft dirty\n");
772 	WARN_ON(!pte_soft_dirty(pte_mksoft_dirty(pte)));
773 	WARN_ON(pte_soft_dirty(pte_clear_soft_dirty(pte)));
774 }
775 
776 static void __init pte_swap_soft_dirty_tests(struct pgtable_debug_args *args)
777 {
778 	pte_t pte = pfn_pte(args->fixed_pte_pfn, args->page_prot);
779 
780 	if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY))
781 		return;
782 
783 	pr_debug("Validating PTE swap soft dirty\n");
784 	WARN_ON(!pte_swp_soft_dirty(pte_swp_mksoft_dirty(pte)));
785 	WARN_ON(pte_swp_soft_dirty(pte_swp_clear_soft_dirty(pte)));
786 }
787 
788 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
789 static void __init pmd_soft_dirty_tests(struct pgtable_debug_args *args)
790 {
791 	pmd_t pmd;
792 
793 	if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY))
794 		return;
795 
796 	if (!has_transparent_hugepage())
797 		return;
798 
799 	pr_debug("Validating PMD soft dirty\n");
800 	pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot);
801 	WARN_ON(!pmd_soft_dirty(pmd_mksoft_dirty(pmd)));
802 	WARN_ON(pmd_soft_dirty(pmd_clear_soft_dirty(pmd)));
803 }
804 
805 static void __init pmd_swap_soft_dirty_tests(struct pgtable_debug_args *args)
806 {
807 	pmd_t pmd;
808 
809 	if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) ||
810 		!IS_ENABLED(CONFIG_ARCH_ENABLE_THP_MIGRATION))
811 		return;
812 
813 	if (!has_transparent_hugepage())
814 		return;
815 
816 	pr_debug("Validating PMD swap soft dirty\n");
817 	pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot);
818 	WARN_ON(!pmd_swp_soft_dirty(pmd_swp_mksoft_dirty(pmd)));
819 	WARN_ON(pmd_swp_soft_dirty(pmd_swp_clear_soft_dirty(pmd)));
820 }
821 #else  /* !CONFIG_TRANSPARENT_HUGEPAGE */
822 static void __init pmd_soft_dirty_tests(struct pgtable_debug_args *args) { }
823 static void __init pmd_swap_soft_dirty_tests(struct pgtable_debug_args *args) { }
824 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
825 
826 static void __init pte_swap_exclusive_tests(struct pgtable_debug_args *args)
827 {
828 	unsigned long max_swap_offset;
829 	swp_entry_t entry, entry2;
830 	pte_t pte;
831 
832 	pr_debug("Validating PTE swap exclusive\n");
833 
834 	/* See generic_max_swapfile_size(): probe the maximum offset */
835 	max_swap_offset = swp_offset(pte_to_swp_entry(swp_entry_to_pte(swp_entry(0, ~0UL))));
836 
837 	/* Create a swp entry with all possible bits set */
838 	entry = swp_entry((1 << MAX_SWAPFILES_SHIFT) - 1, max_swap_offset);
839 
840 	pte = swp_entry_to_pte(entry);
841 	WARN_ON(pte_swp_exclusive(pte));
842 	WARN_ON(!is_swap_pte(pte));
843 	entry2 = pte_to_swp_entry(pte);
844 	WARN_ON(memcmp(&entry, &entry2, sizeof(entry)));
845 
846 	pte = pte_swp_mkexclusive(pte);
847 	WARN_ON(!pte_swp_exclusive(pte));
848 	WARN_ON(!is_swap_pte(pte));
849 	WARN_ON(pte_swp_soft_dirty(pte));
850 	entry2 = pte_to_swp_entry(pte);
851 	WARN_ON(memcmp(&entry, &entry2, sizeof(entry)));
852 
853 	pte = pte_swp_clear_exclusive(pte);
854 	WARN_ON(pte_swp_exclusive(pte));
855 	WARN_ON(!is_swap_pte(pte));
856 	entry2 = pte_to_swp_entry(pte);
857 	WARN_ON(memcmp(&entry, &entry2, sizeof(entry)));
858 }
859 
860 static void __init pte_swap_tests(struct pgtable_debug_args *args)
861 {
862 	swp_entry_t swp;
863 	pte_t pte;
864 
865 	pr_debug("Validating PTE swap\n");
866 	pte = pfn_pte(args->fixed_pte_pfn, args->page_prot);
867 	swp = __pte_to_swp_entry(pte);
868 	pte = __swp_entry_to_pte(swp);
869 	WARN_ON(args->fixed_pte_pfn != pte_pfn(pte));
870 }
871 
872 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
873 static void __init pmd_swap_tests(struct pgtable_debug_args *args)
874 {
875 	swp_entry_t swp;
876 	pmd_t pmd;
877 
878 	if (!has_transparent_hugepage())
879 		return;
880 
881 	pr_debug("Validating PMD swap\n");
882 	pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot);
883 	swp = __pmd_to_swp_entry(pmd);
884 	pmd = __swp_entry_to_pmd(swp);
885 	WARN_ON(args->fixed_pmd_pfn != pmd_pfn(pmd));
886 }
887 #else  /* !CONFIG_ARCH_ENABLE_THP_MIGRATION */
888 static void __init pmd_swap_tests(struct pgtable_debug_args *args) { }
889 #endif /* CONFIG_ARCH_ENABLE_THP_MIGRATION */
890 
891 static void __init swap_migration_tests(struct pgtable_debug_args *args)
892 {
893 	struct page *page;
894 	swp_entry_t swp;
895 
896 	if (!IS_ENABLED(CONFIG_MIGRATION))
897 		return;
898 
899 	/*
900 	 * swap_migration_tests() requires a dedicated page as it needs to
901 	 * be locked before creating a migration entry from it. Locking the
902 	 * page that actually maps kernel text ('start_kernel') can be real
903 	 * problematic. Lets use the allocated page explicitly for this
904 	 * purpose.
905 	 */
906 	page = (args->pte_pfn != ULONG_MAX) ? pfn_to_page(args->pte_pfn) : NULL;
907 	if (!page)
908 		return;
909 
910 	pr_debug("Validating swap migration\n");
911 
912 	/*
913 	 * make_[readable|writable]_migration_entry() expects given page to
914 	 * be locked, otherwise it stumbles upon a BUG_ON().
915 	 */
916 	__SetPageLocked(page);
917 	swp = make_writable_migration_entry(page_to_pfn(page));
918 	WARN_ON(!is_migration_entry(swp));
919 	WARN_ON(!is_writable_migration_entry(swp));
920 
921 	swp = make_readable_migration_entry(swp_offset(swp));
922 	WARN_ON(!is_migration_entry(swp));
923 	WARN_ON(is_writable_migration_entry(swp));
924 
925 	swp = make_readable_migration_entry(page_to_pfn(page));
926 	WARN_ON(!is_migration_entry(swp));
927 	WARN_ON(is_writable_migration_entry(swp));
928 	__ClearPageLocked(page);
929 }
930 
931 #ifdef CONFIG_HUGETLB_PAGE
932 static void __init hugetlb_basic_tests(struct pgtable_debug_args *args)
933 {
934 	struct page *page;
935 	pte_t pte;
936 
937 	pr_debug("Validating HugeTLB basic\n");
938 	/*
939 	 * Accessing the page associated with the pfn is safe here,
940 	 * as it was previously derived from a real kernel symbol.
941 	 */
942 	page = pfn_to_page(args->fixed_pmd_pfn);
943 	pte = mk_huge_pte(page, args->page_prot);
944 
945 	WARN_ON(!huge_pte_dirty(huge_pte_mkdirty(pte)));
946 	WARN_ON(!huge_pte_write(huge_pte_mkwrite(huge_pte_wrprotect(pte))));
947 	WARN_ON(huge_pte_write(huge_pte_wrprotect(huge_pte_mkwrite(pte))));
948 
949 #ifdef CONFIG_ARCH_WANT_GENERAL_HUGETLB
950 	pte = pfn_pte(args->fixed_pmd_pfn, args->page_prot);
951 
952 	WARN_ON(!pte_huge(arch_make_huge_pte(pte, PMD_SHIFT, VM_ACCESS_FLAGS)));
953 #endif /* CONFIG_ARCH_WANT_GENERAL_HUGETLB */
954 }
955 #else  /* !CONFIG_HUGETLB_PAGE */
956 static void __init hugetlb_basic_tests(struct pgtable_debug_args *args) { }
957 #endif /* CONFIG_HUGETLB_PAGE */
958 
959 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
960 static void __init pmd_thp_tests(struct pgtable_debug_args *args)
961 {
962 	pmd_t pmd;
963 
964 	if (!has_transparent_hugepage())
965 		return;
966 
967 	pr_debug("Validating PMD based THP\n");
968 	/*
969 	 * pmd_trans_huge() and pmd_present() must return positive after
970 	 * MMU invalidation with pmd_mkinvalid(). This behavior is an
971 	 * optimization for transparent huge page. pmd_trans_huge() must
972 	 * be true if pmd_page() returns a valid THP to avoid taking the
973 	 * pmd_lock when others walk over non transhuge pmds (i.e. there
974 	 * are no THP allocated). Especially when splitting a THP and
975 	 * removing the present bit from the pmd, pmd_trans_huge() still
976 	 * needs to return true. pmd_present() should be true whenever
977 	 * pmd_trans_huge() returns true.
978 	 */
979 	pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot);
980 	WARN_ON(!pmd_trans_huge(pmd_mkhuge(pmd)));
981 
982 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
983 	WARN_ON(!pmd_trans_huge(pmd_mkinvalid(pmd_mkhuge(pmd))));
984 	WARN_ON(!pmd_present(pmd_mkinvalid(pmd_mkhuge(pmd))));
985 	WARN_ON(!pmd_leaf(pmd_mkinvalid(pmd_mkhuge(pmd))));
986 #endif /* __HAVE_ARCH_PMDP_INVALIDATE */
987 }
988 
989 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
990 static void __init pud_thp_tests(struct pgtable_debug_args *args)
991 {
992 	pud_t pud;
993 
994 	if (!has_transparent_pud_hugepage())
995 		return;
996 
997 	pr_debug("Validating PUD based THP\n");
998 	pud = pfn_pud(args->fixed_pud_pfn, args->page_prot);
999 	WARN_ON(!pud_trans_huge(pud_mkhuge(pud)));
1000 
1001 	/*
1002 	 * pud_mkinvalid() has been dropped for now. Enable back
1003 	 * these tests when it comes back with a modified pud_present().
1004 	 *
1005 	 * WARN_ON(!pud_trans_huge(pud_mkinvalid(pud_mkhuge(pud))));
1006 	 * WARN_ON(!pud_present(pud_mkinvalid(pud_mkhuge(pud))));
1007 	 */
1008 }
1009 #else  /* !CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
1010 static void __init pud_thp_tests(struct pgtable_debug_args *args) { }
1011 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
1012 #else  /* !CONFIG_TRANSPARENT_HUGEPAGE */
1013 static void __init pmd_thp_tests(struct pgtable_debug_args *args) { }
1014 static void __init pud_thp_tests(struct pgtable_debug_args *args) { }
1015 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1016 
1017 static unsigned long __init get_random_vaddr(void)
1018 {
1019 	unsigned long random_vaddr, random_pages, total_user_pages;
1020 
1021 	total_user_pages = (TASK_SIZE - FIRST_USER_ADDRESS) / PAGE_SIZE;
1022 
1023 	random_pages = get_random_long() % total_user_pages;
1024 	random_vaddr = FIRST_USER_ADDRESS + random_pages * PAGE_SIZE;
1025 
1026 	return random_vaddr;
1027 }
1028 
1029 static void __init destroy_args(struct pgtable_debug_args *args)
1030 {
1031 	struct page *page = NULL;
1032 
1033 	/* Free (huge) page */
1034 	if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
1035 	    has_transparent_pud_hugepage() &&
1036 	    args->pud_pfn != ULONG_MAX) {
1037 		if (args->is_contiguous_page) {
1038 			free_contig_range(args->pud_pfn,
1039 					  (1 << (HPAGE_PUD_SHIFT - PAGE_SHIFT)));
1040 		} else {
1041 			page = pfn_to_page(args->pud_pfn);
1042 			__free_pages(page, HPAGE_PUD_SHIFT - PAGE_SHIFT);
1043 		}
1044 
1045 		args->pud_pfn = ULONG_MAX;
1046 		args->pmd_pfn = ULONG_MAX;
1047 		args->pte_pfn = ULONG_MAX;
1048 	}
1049 
1050 	if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
1051 	    has_transparent_hugepage() &&
1052 	    args->pmd_pfn != ULONG_MAX) {
1053 		if (args->is_contiguous_page) {
1054 			free_contig_range(args->pmd_pfn, (1 << HPAGE_PMD_ORDER));
1055 		} else {
1056 			page = pfn_to_page(args->pmd_pfn);
1057 			__free_pages(page, HPAGE_PMD_ORDER);
1058 		}
1059 
1060 		args->pmd_pfn = ULONG_MAX;
1061 		args->pte_pfn = ULONG_MAX;
1062 	}
1063 
1064 	if (args->pte_pfn != ULONG_MAX) {
1065 		page = pfn_to_page(args->pte_pfn);
1066 		__free_page(page);
1067 
1068 		args->pte_pfn = ULONG_MAX;
1069 	}
1070 
1071 	/* Free page table entries */
1072 	if (args->start_ptep) {
1073 		pte_free(args->mm, args->start_ptep);
1074 		mm_dec_nr_ptes(args->mm);
1075 	}
1076 
1077 	if (args->start_pmdp) {
1078 		pmd_free(args->mm, args->start_pmdp);
1079 		mm_dec_nr_pmds(args->mm);
1080 	}
1081 
1082 	if (args->start_pudp) {
1083 		pud_free(args->mm, args->start_pudp);
1084 		mm_dec_nr_puds(args->mm);
1085 	}
1086 
1087 	if (args->start_p4dp)
1088 		p4d_free(args->mm, args->start_p4dp);
1089 
1090 	/* Free vma and mm struct */
1091 	if (args->vma)
1092 		vm_area_free(args->vma);
1093 
1094 	if (args->mm)
1095 		mmdrop(args->mm);
1096 }
1097 
1098 static struct page * __init
1099 debug_vm_pgtable_alloc_huge_page(struct pgtable_debug_args *args, int order)
1100 {
1101 	struct page *page = NULL;
1102 
1103 #ifdef CONFIG_CONTIG_ALLOC
1104 	if (order > MAX_PAGE_ORDER) {
1105 		page = alloc_contig_pages((1 << order), GFP_KERNEL,
1106 					  first_online_node, NULL);
1107 		if (page) {
1108 			args->is_contiguous_page = true;
1109 			return page;
1110 		}
1111 	}
1112 #endif
1113 
1114 	if (order <= MAX_PAGE_ORDER)
1115 		page = alloc_pages(GFP_KERNEL, order);
1116 
1117 	return page;
1118 }
1119 
1120 /*
1121  * Check if a physical memory range described by <pstart, pend> contains
1122  * an area that is of size psize, and aligned to psize.
1123  *
1124  * Don't use address 0, an all-zeroes physical address might mask bugs, and
1125  * it's not used on x86.
1126  */
1127 static void  __init phys_align_check(phys_addr_t pstart,
1128 				     phys_addr_t pend, unsigned long psize,
1129 				     phys_addr_t *physp, unsigned long *alignp)
1130 {
1131 	phys_addr_t aligned_start, aligned_end;
1132 
1133 	if (pstart == 0)
1134 		pstart = PAGE_SIZE;
1135 
1136 	aligned_start = ALIGN(pstart, psize);
1137 	aligned_end = aligned_start + psize;
1138 
1139 	if (aligned_end > aligned_start && aligned_end <= pend) {
1140 		*alignp = psize;
1141 		*physp = aligned_start;
1142 	}
1143 }
1144 
1145 static void __init init_fixed_pfns(struct pgtable_debug_args *args)
1146 {
1147 	u64 idx;
1148 	phys_addr_t phys, pstart, pend;
1149 
1150 	/*
1151 	 * Initialize the fixed pfns. To do this, try to find a
1152 	 * valid physical range, preferably aligned to PUD_SIZE,
1153 	 * but settling for aligned to PMD_SIZE as a fallback. If
1154 	 * neither of those is found, use the physical address of
1155 	 * the start_kernel symbol.
1156 	 *
1157 	 * The memory doesn't need to be allocated, it just needs to exist
1158 	 * as usable memory. It won't be touched.
1159 	 *
1160 	 * The alignment is recorded, and can be checked to see if we
1161 	 * can run the tests that require an actual valid physical
1162 	 * address range on some architectures ({pmd,pud}_huge_test
1163 	 * on x86).
1164 	 */
1165 
1166 	phys = __pa_symbol(&start_kernel);
1167 	args->fixed_alignment = PAGE_SIZE;
1168 
1169 	for_each_mem_range(idx, &pstart, &pend) {
1170 		/* First check for a PUD-aligned area */
1171 		phys_align_check(pstart, pend, PUD_SIZE, &phys,
1172 				 &args->fixed_alignment);
1173 
1174 		/* If a PUD-aligned area is found, we're done */
1175 		if (args->fixed_alignment == PUD_SIZE)
1176 			break;
1177 
1178 		/*
1179 		 * If no PMD-aligned area found yet, check for one,
1180 		 * but continue the loop to look for a PUD-aligned area.
1181 		 */
1182 		if (args->fixed_alignment < PMD_SIZE)
1183 			phys_align_check(pstart, pend, PMD_SIZE, &phys,
1184 					 &args->fixed_alignment);
1185 	}
1186 
1187 	args->fixed_pgd_pfn = __phys_to_pfn(phys & PGDIR_MASK);
1188 	args->fixed_p4d_pfn = __phys_to_pfn(phys & P4D_MASK);
1189 	args->fixed_pud_pfn = __phys_to_pfn(phys & PUD_MASK);
1190 	args->fixed_pmd_pfn = __phys_to_pfn(phys & PMD_MASK);
1191 	args->fixed_pte_pfn = __phys_to_pfn(phys & PAGE_MASK);
1192 	WARN_ON(!pfn_valid(args->fixed_pte_pfn));
1193 }
1194 
1195 
1196 static int __init init_args(struct pgtable_debug_args *args)
1197 {
1198 	struct page *page = NULL;
1199 	int ret = 0;
1200 
1201 	/*
1202 	 * Initialize the debugging data.
1203 	 *
1204 	 * vm_get_page_prot(VM_NONE) or vm_get_page_prot(VM_SHARED|VM_NONE)
1205 	 * will help create page table entries with PROT_NONE permission as
1206 	 * required for pxx_protnone_tests().
1207 	 */
1208 	memset(args, 0, sizeof(*args));
1209 	args->vaddr              = get_random_vaddr();
1210 	args->page_prot          = vm_get_page_prot(VM_ACCESS_FLAGS);
1211 	args->page_prot_none     = vm_get_page_prot(VM_NONE);
1212 	args->is_contiguous_page = false;
1213 	args->pud_pfn            = ULONG_MAX;
1214 	args->pmd_pfn            = ULONG_MAX;
1215 	args->pte_pfn            = ULONG_MAX;
1216 	args->fixed_pgd_pfn      = ULONG_MAX;
1217 	args->fixed_p4d_pfn      = ULONG_MAX;
1218 	args->fixed_pud_pfn      = ULONG_MAX;
1219 	args->fixed_pmd_pfn      = ULONG_MAX;
1220 	args->fixed_pte_pfn      = ULONG_MAX;
1221 
1222 	/* Allocate mm and vma */
1223 	args->mm = mm_alloc();
1224 	if (!args->mm) {
1225 		pr_err("Failed to allocate mm struct\n");
1226 		ret = -ENOMEM;
1227 		goto error;
1228 	}
1229 
1230 	args->vma = vm_area_alloc(args->mm);
1231 	if (!args->vma) {
1232 		pr_err("Failed to allocate vma\n");
1233 		ret = -ENOMEM;
1234 		goto error;
1235 	}
1236 
1237 	/*
1238 	 * Allocate page table entries. They will be modified in the tests.
1239 	 * Lets save the page table entries so that they can be released
1240 	 * when the tests are completed.
1241 	 */
1242 	args->pgdp = pgd_offset(args->mm, args->vaddr);
1243 	args->p4dp = p4d_alloc(args->mm, args->pgdp, args->vaddr);
1244 	if (!args->p4dp) {
1245 		pr_err("Failed to allocate p4d entries\n");
1246 		ret = -ENOMEM;
1247 		goto error;
1248 	}
1249 	args->start_p4dp = p4d_offset(args->pgdp, 0UL);
1250 	WARN_ON(!args->start_p4dp);
1251 
1252 	args->pudp = pud_alloc(args->mm, args->p4dp, args->vaddr);
1253 	if (!args->pudp) {
1254 		pr_err("Failed to allocate pud entries\n");
1255 		ret = -ENOMEM;
1256 		goto error;
1257 	}
1258 	args->start_pudp = pud_offset(args->p4dp, 0UL);
1259 	WARN_ON(!args->start_pudp);
1260 
1261 	args->pmdp = pmd_alloc(args->mm, args->pudp, args->vaddr);
1262 	if (!args->pmdp) {
1263 		pr_err("Failed to allocate pmd entries\n");
1264 		ret = -ENOMEM;
1265 		goto error;
1266 	}
1267 	args->start_pmdp = pmd_offset(args->pudp, 0UL);
1268 	WARN_ON(!args->start_pmdp);
1269 
1270 	if (pte_alloc(args->mm, args->pmdp)) {
1271 		pr_err("Failed to allocate pte entries\n");
1272 		ret = -ENOMEM;
1273 		goto error;
1274 	}
1275 	args->start_ptep = pmd_pgtable(READ_ONCE(*args->pmdp));
1276 	WARN_ON(!args->start_ptep);
1277 
1278 	init_fixed_pfns(args);
1279 
1280 	/*
1281 	 * Allocate (huge) pages because some of the tests need to access
1282 	 * the data in the pages. The corresponding tests will be skipped
1283 	 * if we fail to allocate (huge) pages.
1284 	 */
1285 	if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
1286 	    has_transparent_pud_hugepage()) {
1287 		page = debug_vm_pgtable_alloc_huge_page(args,
1288 				HPAGE_PUD_SHIFT - PAGE_SHIFT);
1289 		if (page) {
1290 			args->pud_pfn = page_to_pfn(page);
1291 			args->pmd_pfn = args->pud_pfn;
1292 			args->pte_pfn = args->pud_pfn;
1293 			return 0;
1294 		}
1295 	}
1296 
1297 	if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
1298 	    has_transparent_hugepage()) {
1299 		page = debug_vm_pgtable_alloc_huge_page(args, HPAGE_PMD_ORDER);
1300 		if (page) {
1301 			args->pmd_pfn = page_to_pfn(page);
1302 			args->pte_pfn = args->pmd_pfn;
1303 			return 0;
1304 		}
1305 	}
1306 
1307 	page = alloc_page(GFP_KERNEL);
1308 	if (page)
1309 		args->pte_pfn = page_to_pfn(page);
1310 
1311 	return 0;
1312 
1313 error:
1314 	destroy_args(args);
1315 	return ret;
1316 }
1317 
1318 static int __init debug_vm_pgtable(void)
1319 {
1320 	struct pgtable_debug_args args;
1321 	spinlock_t *ptl = NULL;
1322 	int idx, ret;
1323 
1324 	pr_info("Validating architecture page table helpers\n");
1325 	ret = init_args(&args);
1326 	if (ret)
1327 		return ret;
1328 
1329 	/*
1330 	 * Iterate over each possible vm_flags to make sure that all
1331 	 * the basic page table transformation validations just hold
1332 	 * true irrespective of the starting protection value for a
1333 	 * given page table entry.
1334 	 *
1335 	 * Protection based vm_flags combinations are always linear
1336 	 * and increasing i.e starting from VM_NONE and going up to
1337 	 * (VM_SHARED | READ | WRITE | EXEC).
1338 	 */
1339 #define VM_FLAGS_START	(VM_NONE)
1340 #define VM_FLAGS_END	(VM_SHARED | VM_EXEC | VM_WRITE | VM_READ)
1341 
1342 	for (idx = VM_FLAGS_START; idx <= VM_FLAGS_END; idx++) {
1343 		pte_basic_tests(&args, idx);
1344 		pmd_basic_tests(&args, idx);
1345 		pud_basic_tests(&args, idx);
1346 	}
1347 
1348 	/*
1349 	 * Both P4D and PGD level tests are very basic which do not
1350 	 * involve creating page table entries from the protection
1351 	 * value and the given pfn. Hence just keep them out from
1352 	 * the above iteration for now to save some test execution
1353 	 * time.
1354 	 */
1355 	p4d_basic_tests(&args);
1356 	pgd_basic_tests(&args);
1357 
1358 	pmd_leaf_tests(&args);
1359 	pud_leaf_tests(&args);
1360 
1361 	pte_special_tests(&args);
1362 	pte_protnone_tests(&args);
1363 	pmd_protnone_tests(&args);
1364 
1365 	pte_devmap_tests(&args);
1366 	pmd_devmap_tests(&args);
1367 	pud_devmap_tests(&args);
1368 
1369 	pte_soft_dirty_tests(&args);
1370 	pmd_soft_dirty_tests(&args);
1371 	pte_swap_soft_dirty_tests(&args);
1372 	pmd_swap_soft_dirty_tests(&args);
1373 
1374 	pte_swap_exclusive_tests(&args);
1375 
1376 	pte_swap_tests(&args);
1377 	pmd_swap_tests(&args);
1378 
1379 	swap_migration_tests(&args);
1380 
1381 	pmd_thp_tests(&args);
1382 	pud_thp_tests(&args);
1383 
1384 	hugetlb_basic_tests(&args);
1385 
1386 	/*
1387 	 * Page table modifying tests. They need to hold
1388 	 * proper page table lock.
1389 	 */
1390 
1391 	args.ptep = pte_offset_map_lock(args.mm, args.pmdp, args.vaddr, &ptl);
1392 	pte_clear_tests(&args);
1393 	pte_advanced_tests(&args);
1394 	if (args.ptep)
1395 		pte_unmap_unlock(args.ptep, ptl);
1396 
1397 	ptl = pmd_lock(args.mm, args.pmdp);
1398 	pmd_clear_tests(&args);
1399 	pmd_advanced_tests(&args);
1400 	pmd_huge_tests(&args);
1401 	pmd_populate_tests(&args);
1402 	spin_unlock(ptl);
1403 
1404 	ptl = pud_lock(args.mm, args.pudp);
1405 	pud_clear_tests(&args);
1406 	pud_advanced_tests(&args);
1407 	pud_huge_tests(&args);
1408 	pud_populate_tests(&args);
1409 	spin_unlock(ptl);
1410 
1411 	spin_lock(&(args.mm->page_table_lock));
1412 	p4d_clear_tests(&args);
1413 	pgd_clear_tests(&args);
1414 	p4d_populate_tests(&args);
1415 	pgd_populate_tests(&args);
1416 	spin_unlock(&(args.mm->page_table_lock));
1417 
1418 	destroy_args(&args);
1419 	return 0;
1420 }
1421 late_initcall(debug_vm_pgtable);
1422