xref: /linux/mm/debug_vm_pgtable.c (revision 3e93d5bbcbfc3808f83712c0701f9d4c148cc8ed)
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/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 <linux/vmalloc.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 #define RANDOM_NZVALUE	GENMASK(7, 0)
44 
45 struct pgtable_debug_args {
46 	struct mm_struct	*mm;
47 	struct vm_area_struct	*vma;
48 
49 	pgd_t			*pgdp;
50 	p4d_t			*p4dp;
51 	pud_t			*pudp;
52 	pmd_t			*pmdp;
53 	pte_t			*ptep;
54 
55 	p4d_t			*start_p4dp;
56 	pud_t			*start_pudp;
57 	pmd_t			*start_pmdp;
58 	pgtable_t		start_ptep;
59 
60 	unsigned long		vaddr;
61 	pgprot_t		page_prot;
62 	pgprot_t		page_prot_none;
63 
64 	bool			is_contiguous_page;
65 	unsigned long		pud_pfn;
66 	unsigned long		pmd_pfn;
67 	unsigned long		pte_pfn;
68 
69 	unsigned long		fixed_alignment;
70 	unsigned long		fixed_pgd_pfn;
71 	unsigned long		fixed_p4d_pfn;
72 	unsigned long		fixed_pud_pfn;
73 	unsigned long		fixed_pmd_pfn;
74 	unsigned long		fixed_pte_pfn;
75 
76 	swp_entry_t		swp_entry;
77 };
78 
pte_basic_tests(struct pgtable_debug_args * args,int idx)79 static void __init pte_basic_tests(struct pgtable_debug_args *args, int idx)
80 {
81 	pgprot_t prot = vm_get_page_prot(idx);
82 	pte_t pte = pfn_pte(args->fixed_pte_pfn, prot);
83 	unsigned long val = idx, *ptr = &val;
84 
85 	pr_debug("Validating PTE basic (%pGv)\n", ptr);
86 
87 	/*
88 	 * This test needs to be executed after the given page table entry
89 	 * is created with pfn_pte() to make sure that vm_get_page_prot(idx)
90 	 * does not have the dirty bit enabled from the beginning. This is
91 	 * important for platforms like arm64 where (!PTE_RDONLY) indicate
92 	 * dirty bit being set.
93 	 */
94 	WARN_ON(pte_dirty(pte_wrprotect(pte)));
95 
96 	WARN_ON(!pte_same(pte, pte));
97 	WARN_ON(!pte_young(pte_mkyoung(pte_mkold(pte))));
98 	WARN_ON(!pte_dirty(pte_mkdirty(pte_mkclean(pte))));
99 	WARN_ON(!pte_write(pte_mkwrite(pte_wrprotect(pte), args->vma)));
100 	WARN_ON(pte_young(pte_mkold(pte_mkyoung(pte))));
101 	WARN_ON(pte_dirty(pte_mkclean(pte_mkdirty(pte))));
102 	WARN_ON(pte_write(pte_wrprotect(pte_mkwrite(pte, args->vma))));
103 	WARN_ON(pte_dirty(pte_wrprotect(pte_mkclean(pte))));
104 	WARN_ON(!pte_dirty(pte_wrprotect(pte_mkdirty(pte))));
105 }
106 
pte_advanced_tests(struct pgtable_debug_args * args)107 static void __init pte_advanced_tests(struct pgtable_debug_args *args)
108 {
109 	struct page *page;
110 	pte_t pte;
111 
112 	/*
113 	 * Architectures optimize set_pte_at by avoiding TLB flush.
114 	 * This requires set_pte_at to be not used to update an
115 	 * existing pte entry. Clear pte before we do set_pte_at
116 	 *
117 	 * flush_dcache_page() is called after set_pte_at() to clear
118 	 * PG_arch_1 for the page on ARM64. The page flag isn't cleared
119 	 * when it's released and page allocation check will fail when
120 	 * the page is allocated again. For architectures other than ARM64,
121 	 * the unexpected overhead of cache flushing is acceptable.
122 	 */
123 	page = (args->pte_pfn != ULONG_MAX) ? pfn_to_page(args->pte_pfn) : NULL;
124 	if (!page)
125 		return;
126 
127 	pr_debug("Validating PTE advanced\n");
128 	if (WARN_ON(!args->ptep))
129 		return;
130 
131 	pte = pfn_pte(args->pte_pfn, args->page_prot);
132 	set_pte_at(args->mm, args->vaddr, args->ptep, pte);
133 	flush_dcache_page(page);
134 	ptep_set_wrprotect(args->mm, args->vaddr, args->ptep);
135 	pte = ptep_get(args->ptep);
136 	WARN_ON(pte_write(pte));
137 	ptep_get_and_clear(args->mm, args->vaddr, args->ptep);
138 	pte = ptep_get(args->ptep);
139 	WARN_ON(!pte_none(pte));
140 
141 	pte = pfn_pte(args->pte_pfn, args->page_prot);
142 	pte = pte_wrprotect(pte);
143 	pte = pte_mkclean(pte);
144 	set_pte_at(args->mm, args->vaddr, args->ptep, pte);
145 	flush_dcache_page(page);
146 	pte = pte_mkwrite(pte, args->vma);
147 	pte = pte_mkdirty(pte);
148 	ptep_set_access_flags(args->vma, args->vaddr, args->ptep, pte, 1);
149 	pte = ptep_get(args->ptep);
150 	WARN_ON(!(pte_write(pte) && pte_dirty(pte)));
151 	ptep_get_and_clear_full(args->mm, args->vaddr, args->ptep, 1);
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_mkyoung(pte);
157 	set_pte_at(args->mm, args->vaddr, args->ptep, pte);
158 	flush_dcache_page(page);
159 	ptep_test_and_clear_young(args->vma, args->vaddr, args->ptep);
160 	pte = ptep_get(args->ptep);
161 	WARN_ON(pte_young(pte));
162 
163 	ptep_get_and_clear_full(args->mm, args->vaddr, args->ptep, 1);
164 }
165 
166 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
pmd_basic_tests(struct pgtable_debug_args * args,int idx)167 static void __init pmd_basic_tests(struct pgtable_debug_args *args, int idx)
168 {
169 	pgprot_t prot = vm_get_page_prot(idx);
170 	unsigned long val = idx, *ptr = &val;
171 	pmd_t pmd;
172 
173 	if (!has_transparent_hugepage())
174 		return;
175 
176 	pr_debug("Validating PMD basic (%pGv)\n", ptr);
177 	pmd = pfn_pmd(args->fixed_pmd_pfn, prot);
178 
179 	/*
180 	 * This test needs to be executed after the given page table entry
181 	 * is created with pfn_pmd() to make sure that vm_get_page_prot(idx)
182 	 * does not have the dirty bit enabled from the beginning. This is
183 	 * important for platforms like arm64 where (!PTE_RDONLY) indicate
184 	 * dirty bit being set.
185 	 */
186 	WARN_ON(pmd_dirty(pmd_wrprotect(pmd)));
187 
188 
189 	WARN_ON(!pmd_same(pmd, pmd));
190 	WARN_ON(!pmd_young(pmd_mkyoung(pmd_mkold(pmd))));
191 	WARN_ON(!pmd_dirty(pmd_mkdirty(pmd_mkclean(pmd))));
192 	WARN_ON(!pmd_write(pmd_mkwrite(pmd_wrprotect(pmd), args->vma)));
193 	WARN_ON(pmd_young(pmd_mkold(pmd_mkyoung(pmd))));
194 	WARN_ON(pmd_dirty(pmd_mkclean(pmd_mkdirty(pmd))));
195 	WARN_ON(pmd_write(pmd_wrprotect(pmd_mkwrite(pmd, args->vma))));
196 	WARN_ON(pmd_dirty(pmd_wrprotect(pmd_mkclean(pmd))));
197 	WARN_ON(!pmd_dirty(pmd_wrprotect(pmd_mkdirty(pmd))));
198 	/*
199 	 * A huge page does not point to next level page table
200 	 * entry. Hence this must qualify as pmd_bad().
201 	 */
202 	WARN_ON(!pmd_bad(pmd_mkhuge(pmd)));
203 }
204 
pmd_advanced_tests(struct pgtable_debug_args * args)205 static void __init pmd_advanced_tests(struct pgtable_debug_args *args)
206 {
207 	struct page *page;
208 	pmd_t pmd;
209 	unsigned long vaddr = args->vaddr;
210 
211 	if (!has_transparent_hugepage())
212 		return;
213 
214 	page = (args->pmd_pfn != ULONG_MAX) ? pfn_to_page(args->pmd_pfn) : NULL;
215 	if (!page)
216 		return;
217 
218 	/*
219 	 * flush_dcache_page() is called after set_pmd_at() to clear
220 	 * PG_arch_1 for the page on ARM64. The page flag isn't cleared
221 	 * when it's released and page allocation check will fail when
222 	 * the page is allocated again. For architectures other than ARM64,
223 	 * the unexpected overhead of cache flushing is acceptable.
224 	 */
225 	pr_debug("Validating PMD advanced\n");
226 	/* Align the address wrt HPAGE_PMD_SIZE */
227 	vaddr &= HPAGE_PMD_MASK;
228 
229 	pgtable_trans_huge_deposit(args->mm, args->pmdp, args->start_ptep);
230 
231 	pmd = pfn_pmd(args->pmd_pfn, args->page_prot);
232 	set_pmd_at(args->mm, vaddr, args->pmdp, pmd);
233 	flush_dcache_page(page);
234 	pmdp_set_wrprotect(args->mm, vaddr, args->pmdp);
235 	pmd = pmdp_get(args->pmdp);
236 	WARN_ON(pmd_write(pmd));
237 	pmdp_huge_get_and_clear(args->mm, vaddr, args->pmdp);
238 	pmd = pmdp_get(args->pmdp);
239 	WARN_ON(!pmd_none(pmd));
240 
241 	pmd = pfn_pmd(args->pmd_pfn, args->page_prot);
242 	pmd = pmd_wrprotect(pmd);
243 	pmd = pmd_mkclean(pmd);
244 	set_pmd_at(args->mm, vaddr, args->pmdp, pmd);
245 	flush_dcache_page(page);
246 	pmd = pmd_mkwrite(pmd, args->vma);
247 	pmd = pmd_mkdirty(pmd);
248 	pmdp_set_access_flags(args->vma, vaddr, args->pmdp, pmd, 1);
249 	pmd = pmdp_get(args->pmdp);
250 	WARN_ON(!(pmd_write(pmd) && pmd_dirty(pmd)));
251 	pmdp_huge_get_and_clear_full(args->vma, vaddr, args->pmdp, 1);
252 	pmd = pmdp_get(args->pmdp);
253 	WARN_ON(!pmd_none(pmd));
254 
255 	pmd = pmd_mkhuge(pfn_pmd(args->pmd_pfn, args->page_prot));
256 	pmd = pmd_mkyoung(pmd);
257 	set_pmd_at(args->mm, vaddr, args->pmdp, pmd);
258 	flush_dcache_page(page);
259 	pmdp_test_and_clear_young(args->vma, vaddr, args->pmdp);
260 	pmd = pmdp_get(args->pmdp);
261 	WARN_ON(pmd_young(pmd));
262 
263 	/*  Clear the pte entries  */
264 	pmdp_huge_get_and_clear(args->mm, vaddr, args->pmdp);
265 	pgtable_trans_huge_withdraw(args->mm, args->pmdp);
266 }
267 
pmd_leaf_tests(struct pgtable_debug_args * args)268 static void __init pmd_leaf_tests(struct pgtable_debug_args *args)
269 {
270 	pmd_t pmd;
271 
272 	if (!has_transparent_hugepage())
273 		return;
274 
275 	pr_debug("Validating PMD leaf\n");
276 	pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot);
277 
278 	/*
279 	 * PMD based THP is a leaf entry.
280 	 */
281 	pmd = pmd_mkhuge(pmd);
282 	WARN_ON(!pmd_leaf(pmd));
283 }
284 
285 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
pud_basic_tests(struct pgtable_debug_args * args,int idx)286 static void __init pud_basic_tests(struct pgtable_debug_args *args, int idx)
287 {
288 	pgprot_t prot = vm_get_page_prot(idx);
289 	unsigned long val = idx, *ptr = &val;
290 	pud_t pud;
291 
292 	if (!has_transparent_pud_hugepage())
293 		return;
294 
295 	pr_debug("Validating PUD basic (%pGv)\n", ptr);
296 	pud = pfn_pud(args->fixed_pud_pfn, prot);
297 
298 	/*
299 	 * This test needs to be executed after the given page table entry
300 	 * is created with pfn_pud() to make sure that vm_get_page_prot(idx)
301 	 * does not have the dirty bit enabled from the beginning. This is
302 	 * important for platforms like arm64 where (!PTE_RDONLY) indicate
303 	 * dirty bit being set.
304 	 */
305 	WARN_ON(pud_dirty(pud_wrprotect(pud)));
306 
307 	WARN_ON(!pud_same(pud, pud));
308 	WARN_ON(!pud_young(pud_mkyoung(pud_mkold(pud))));
309 	WARN_ON(!pud_dirty(pud_mkdirty(pud_mkclean(pud))));
310 	WARN_ON(pud_dirty(pud_mkclean(pud_mkdirty(pud))));
311 	WARN_ON(!pud_write(pud_mkwrite(pud_wrprotect(pud))));
312 	WARN_ON(pud_write(pud_wrprotect(pud_mkwrite(pud))));
313 	WARN_ON(pud_young(pud_mkold(pud_mkyoung(pud))));
314 	WARN_ON(pud_dirty(pud_wrprotect(pud_mkclean(pud))));
315 	WARN_ON(!pud_dirty(pud_wrprotect(pud_mkdirty(pud))));
316 
317 	if (mm_pmd_folded(args->mm))
318 		return;
319 
320 	/*
321 	 * A huge page does not point to next level page table
322 	 * entry. Hence this must qualify as pud_bad().
323 	 */
324 	WARN_ON(!pud_bad(pud_mkhuge(pud)));
325 }
326 
pud_advanced_tests(struct pgtable_debug_args * args)327 static void __init pud_advanced_tests(struct pgtable_debug_args *args)
328 {
329 	struct page *page;
330 	unsigned long vaddr = args->vaddr;
331 	pud_t pud;
332 
333 	if (!has_transparent_pud_hugepage())
334 		return;
335 
336 	page = (args->pud_pfn != ULONG_MAX) ? pfn_to_page(args->pud_pfn) : NULL;
337 	if (!page)
338 		return;
339 
340 	/*
341 	 * flush_dcache_page() is called after set_pud_at() to clear
342 	 * PG_arch_1 for the page on ARM64. The page flag isn't cleared
343 	 * when it's released and page allocation check will fail when
344 	 * the page is allocated again. For architectures other than ARM64,
345 	 * the unexpected overhead of cache flushing is acceptable.
346 	 */
347 	pr_debug("Validating PUD advanced\n");
348 	/* Align the address wrt HPAGE_PUD_SIZE */
349 	vaddr &= HPAGE_PUD_MASK;
350 
351 	pud = pfn_pud(args->pud_pfn, args->page_prot);
352 	set_pud_at(args->mm, vaddr, args->pudp, pud);
353 	flush_dcache_page(page);
354 	pudp_set_wrprotect(args->mm, vaddr, args->pudp);
355 	pud = pudp_get(args->pudp);
356 	WARN_ON(pud_write(pud));
357 
358 #ifndef __PAGETABLE_PMD_FOLDED
359 	pudp_huge_get_and_clear(args->mm, vaddr, args->pudp);
360 	pud = pudp_get(args->pudp);
361 	WARN_ON(!pud_none(pud));
362 #endif /* __PAGETABLE_PMD_FOLDED */
363 	pud = pfn_pud(args->pud_pfn, args->page_prot);
364 	pud = pud_wrprotect(pud);
365 	pud = pud_mkclean(pud);
366 	set_pud_at(args->mm, vaddr, args->pudp, pud);
367 	flush_dcache_page(page);
368 	pud = pud_mkwrite(pud);
369 	pud = pud_mkdirty(pud);
370 	pudp_set_access_flags(args->vma, vaddr, args->pudp, pud, 1);
371 	pud = pudp_get(args->pudp);
372 	WARN_ON(!(pud_write(pud) && pud_dirty(pud)));
373 
374 #ifndef __PAGETABLE_PMD_FOLDED
375 	pudp_huge_get_and_clear_full(args->vma, vaddr, args->pudp, 1);
376 	pud = pudp_get(args->pudp);
377 	WARN_ON(!pud_none(pud));
378 #endif /* __PAGETABLE_PMD_FOLDED */
379 
380 	pud = pfn_pud(args->pud_pfn, args->page_prot);
381 	pud = pud_mkyoung(pud);
382 	set_pud_at(args->mm, vaddr, args->pudp, pud);
383 	flush_dcache_page(page);
384 	pudp_test_and_clear_young(args->vma, vaddr, args->pudp);
385 	pud = pudp_get(args->pudp);
386 	WARN_ON(pud_young(pud));
387 
388 	pudp_huge_get_and_clear(args->mm, vaddr, args->pudp);
389 }
390 
pud_leaf_tests(struct pgtable_debug_args * args)391 static void __init pud_leaf_tests(struct pgtable_debug_args *args)
392 {
393 	pud_t pud;
394 
395 	if (!has_transparent_pud_hugepage())
396 		return;
397 
398 	pr_debug("Validating PUD leaf\n");
399 	pud = pfn_pud(args->fixed_pud_pfn, args->page_prot);
400 	/*
401 	 * PUD based THP is a leaf entry.
402 	 */
403 	pud = pud_mkhuge(pud);
404 	WARN_ON(!pud_leaf(pud));
405 }
406 #else  /* !CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
pud_basic_tests(struct pgtable_debug_args * args,int idx)407 static void __init pud_basic_tests(struct pgtable_debug_args *args, int idx) { }
pud_advanced_tests(struct pgtable_debug_args * args)408 static void __init pud_advanced_tests(struct pgtable_debug_args *args) { }
pud_leaf_tests(struct pgtable_debug_args * args)409 static void __init pud_leaf_tests(struct pgtable_debug_args *args) { }
410 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
411 #else  /* !CONFIG_TRANSPARENT_HUGEPAGE */
pmd_basic_tests(struct pgtable_debug_args * args,int idx)412 static void __init pmd_basic_tests(struct pgtable_debug_args *args, int idx) { }
pud_basic_tests(struct pgtable_debug_args * args,int idx)413 static void __init pud_basic_tests(struct pgtable_debug_args *args, int idx) { }
pmd_advanced_tests(struct pgtable_debug_args * args)414 static void __init pmd_advanced_tests(struct pgtable_debug_args *args) { }
pud_advanced_tests(struct pgtable_debug_args * args)415 static void __init pud_advanced_tests(struct pgtable_debug_args *args) { }
pmd_leaf_tests(struct pgtable_debug_args * args)416 static void __init pmd_leaf_tests(struct pgtable_debug_args *args) { }
pud_leaf_tests(struct pgtable_debug_args * args)417 static void __init pud_leaf_tests(struct pgtable_debug_args *args) { }
418 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
419 
420 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
pmd_huge_tests(struct pgtable_debug_args * args)421 static void __init pmd_huge_tests(struct pgtable_debug_args *args)
422 {
423 	pmd_t pmd;
424 
425 	if (!arch_vmap_pmd_supported(args->page_prot) ||
426 	    args->fixed_alignment < PMD_SIZE)
427 		return;
428 
429 	pr_debug("Validating PMD huge\n");
430 	/*
431 	 * X86 defined pmd_set_huge() verifies that the given
432 	 * PMD is not a populated non-leaf entry.
433 	 */
434 	WRITE_ONCE(*args->pmdp, __pmd(0));
435 	WARN_ON(!pmd_set_huge(args->pmdp, __pfn_to_phys(args->fixed_pmd_pfn), args->page_prot));
436 	WARN_ON(!pmd_clear_huge(args->pmdp));
437 	pmd = pmdp_get(args->pmdp);
438 	WARN_ON(!pmd_none(pmd));
439 }
440 
pud_huge_tests(struct pgtable_debug_args * args)441 static void __init pud_huge_tests(struct pgtable_debug_args *args)
442 {
443 	pud_t pud;
444 
445 	if (!arch_vmap_pud_supported(args->page_prot) ||
446 	    args->fixed_alignment < PUD_SIZE)
447 		return;
448 
449 	pr_debug("Validating PUD huge\n");
450 	/*
451 	 * X86 defined pud_set_huge() verifies that the given
452 	 * PUD is not a populated non-leaf entry.
453 	 */
454 	WRITE_ONCE(*args->pudp, __pud(0));
455 	WARN_ON(!pud_set_huge(args->pudp, __pfn_to_phys(args->fixed_pud_pfn), args->page_prot));
456 	WARN_ON(!pud_clear_huge(args->pudp));
457 	pud = pudp_get(args->pudp);
458 	WARN_ON(!pud_none(pud));
459 }
460 #else /* !CONFIG_HAVE_ARCH_HUGE_VMAP */
pmd_huge_tests(struct pgtable_debug_args * args)461 static void __init pmd_huge_tests(struct pgtable_debug_args *args) { }
pud_huge_tests(struct pgtable_debug_args * args)462 static void __init pud_huge_tests(struct pgtable_debug_args *args) { }
463 #endif /* CONFIG_HAVE_ARCH_HUGE_VMAP */
464 
p4d_basic_tests(struct pgtable_debug_args * args)465 static void __init p4d_basic_tests(struct pgtable_debug_args *args)
466 {
467 	p4d_t p4d;
468 
469 	pr_debug("Validating P4D basic\n");
470 	memset(&p4d, RANDOM_NZVALUE, sizeof(p4d_t));
471 	WARN_ON(!p4d_same(p4d, p4d));
472 }
473 
pgd_basic_tests(struct pgtable_debug_args * args)474 static void __init pgd_basic_tests(struct pgtable_debug_args *args)
475 {
476 	pgd_t pgd;
477 
478 	pr_debug("Validating PGD basic\n");
479 	memset(&pgd, RANDOM_NZVALUE, sizeof(pgd_t));
480 	WARN_ON(!pgd_same(pgd, pgd));
481 }
482 
483 #ifndef __PAGETABLE_PUD_FOLDED
pud_clear_tests(struct pgtable_debug_args * args)484 static void __init pud_clear_tests(struct pgtable_debug_args *args)
485 {
486 	pud_t pud = pudp_get(args->pudp);
487 
488 	if (mm_pmd_folded(args->mm))
489 		return;
490 
491 	pr_debug("Validating PUD clear\n");
492 	WARN_ON(pud_none(pud));
493 	pud_clear(args->pudp);
494 	pud = pudp_get(args->pudp);
495 	WARN_ON(!pud_none(pud));
496 }
497 
pud_populate_tests(struct pgtable_debug_args * args)498 static void __init pud_populate_tests(struct pgtable_debug_args *args)
499 {
500 	pud_t pud;
501 
502 	if (mm_pmd_folded(args->mm))
503 		return;
504 
505 	pr_debug("Validating PUD populate\n");
506 	/*
507 	 * This entry points to next level page table page.
508 	 * Hence this must not qualify as pud_bad().
509 	 */
510 	pud_populate(args->mm, args->pudp, args->start_pmdp);
511 	pud = pudp_get(args->pudp);
512 	WARN_ON(pud_bad(pud));
513 }
514 #else  /* !__PAGETABLE_PUD_FOLDED */
pud_clear_tests(struct pgtable_debug_args * args)515 static void __init pud_clear_tests(struct pgtable_debug_args *args) { }
pud_populate_tests(struct pgtable_debug_args * args)516 static void __init pud_populate_tests(struct pgtable_debug_args *args) { }
517 #endif /* PAGETABLE_PUD_FOLDED */
518 
519 #ifndef __PAGETABLE_P4D_FOLDED
p4d_clear_tests(struct pgtable_debug_args * args)520 static void __init p4d_clear_tests(struct pgtable_debug_args *args)
521 {
522 	p4d_t p4d = p4dp_get(args->p4dp);
523 
524 	if (mm_pud_folded(args->mm))
525 		return;
526 
527 	pr_debug("Validating P4D clear\n");
528 	WARN_ON(p4d_none(p4d));
529 	p4d_clear(args->p4dp);
530 	p4d = p4dp_get(args->p4dp);
531 	WARN_ON(!p4d_none(p4d));
532 }
533 
p4d_populate_tests(struct pgtable_debug_args * args)534 static void __init p4d_populate_tests(struct pgtable_debug_args *args)
535 {
536 	p4d_t p4d;
537 
538 	if (mm_pud_folded(args->mm))
539 		return;
540 
541 	pr_debug("Validating P4D populate\n");
542 	/*
543 	 * This entry points to next level page table page.
544 	 * Hence this must not qualify as p4d_bad().
545 	 */
546 	pud_clear(args->pudp);
547 	p4d_clear(args->p4dp);
548 	p4d_populate(args->mm, args->p4dp, args->start_pudp);
549 	p4d = p4dp_get(args->p4dp);
550 	WARN_ON(p4d_bad(p4d));
551 }
552 
pgd_clear_tests(struct pgtable_debug_args * args)553 static void __init pgd_clear_tests(struct pgtable_debug_args *args)
554 {
555 	pgd_t pgd = pgdp_get(args->pgdp);
556 
557 	if (mm_p4d_folded(args->mm))
558 		return;
559 
560 	pr_debug("Validating PGD clear\n");
561 	WARN_ON(pgd_none(pgd));
562 	pgd_clear(args->pgdp);
563 	pgd = pgdp_get(args->pgdp);
564 	WARN_ON(!pgd_none(pgd));
565 }
566 
pgd_populate_tests(struct pgtable_debug_args * args)567 static void __init pgd_populate_tests(struct pgtable_debug_args *args)
568 {
569 	pgd_t pgd;
570 
571 	if (mm_p4d_folded(args->mm))
572 		return;
573 
574 	pr_debug("Validating PGD populate\n");
575 	/*
576 	 * This entry points to next level page table page.
577 	 * Hence this must not qualify as pgd_bad().
578 	 */
579 	p4d_clear(args->p4dp);
580 	pgd_clear(args->pgdp);
581 	pgd_populate(args->mm, args->pgdp, args->start_p4dp);
582 	pgd = pgdp_get(args->pgdp);
583 	WARN_ON(pgd_bad(pgd));
584 }
585 #else  /* !__PAGETABLE_P4D_FOLDED */
p4d_clear_tests(struct pgtable_debug_args * args)586 static void __init p4d_clear_tests(struct pgtable_debug_args *args) { }
pgd_clear_tests(struct pgtable_debug_args * args)587 static void __init pgd_clear_tests(struct pgtable_debug_args *args) { }
p4d_populate_tests(struct pgtable_debug_args * args)588 static void __init p4d_populate_tests(struct pgtable_debug_args *args) { }
pgd_populate_tests(struct pgtable_debug_args * args)589 static void __init pgd_populate_tests(struct pgtable_debug_args *args) { }
590 #endif /* PAGETABLE_P4D_FOLDED */
591 
pte_clear_tests(struct pgtable_debug_args * args)592 static void __init pte_clear_tests(struct pgtable_debug_args *args)
593 {
594 	struct page *page;
595 	pte_t pte = pfn_pte(args->pte_pfn, args->page_prot);
596 
597 	page = (args->pte_pfn != ULONG_MAX) ? pfn_to_page(args->pte_pfn) : NULL;
598 	if (!page)
599 		return;
600 
601 	/*
602 	 * flush_dcache_page() is called after set_pte_at() to clear
603 	 * PG_arch_1 for the page on ARM64. The page flag isn't cleared
604 	 * when it's released and page allocation check will fail when
605 	 * the page is allocated again. For architectures other than ARM64,
606 	 * the unexpected overhead of cache flushing is acceptable.
607 	 */
608 	pr_debug("Validating PTE clear\n");
609 	if (WARN_ON(!args->ptep))
610 		return;
611 
612 	set_pte_at(args->mm, args->vaddr, args->ptep, pte);
613 	WARN_ON(pte_none(pte));
614 	flush_dcache_page(page);
615 	barrier();
616 	ptep_clear(args->mm, args->vaddr, args->ptep);
617 	pte = ptep_get(args->ptep);
618 	WARN_ON(!pte_none(pte));
619 }
620 
pmd_clear_tests(struct pgtable_debug_args * args)621 static void __init pmd_clear_tests(struct pgtable_debug_args *args)
622 {
623 	pmd_t pmd = pmdp_get(args->pmdp);
624 
625 	pr_debug("Validating PMD clear\n");
626 	WARN_ON(pmd_none(pmd));
627 	pmd_clear(args->pmdp);
628 	pmd = pmdp_get(args->pmdp);
629 	WARN_ON(!pmd_none(pmd));
630 }
631 
pmd_populate_tests(struct pgtable_debug_args * args)632 static void __init pmd_populate_tests(struct pgtable_debug_args *args)
633 {
634 	pmd_t pmd;
635 
636 	pr_debug("Validating PMD populate\n");
637 	/*
638 	 * This entry points to next level page table page.
639 	 * Hence this must not qualify as pmd_bad().
640 	 */
641 	pmd_populate(args->mm, args->pmdp, args->start_ptep);
642 	pmd = pmdp_get(args->pmdp);
643 	WARN_ON(pmd_bad(pmd));
644 }
645 
pte_special_tests(struct pgtable_debug_args * args)646 static void __init pte_special_tests(struct pgtable_debug_args *args)
647 {
648 	pte_t pte = pfn_pte(args->fixed_pte_pfn, args->page_prot);
649 
650 	if (!IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL))
651 		return;
652 
653 	pr_debug("Validating PTE special\n");
654 	WARN_ON(!pte_special(pte_mkspecial(pte)));
655 }
656 
pte_protnone_tests(struct pgtable_debug_args * args)657 static void __init pte_protnone_tests(struct pgtable_debug_args *args)
658 {
659 	pte_t pte = pfn_pte(args->fixed_pte_pfn, args->page_prot_none);
660 
661 	if (!IS_ENABLED(CONFIG_NUMA_BALANCING))
662 		return;
663 
664 	pr_debug("Validating PTE protnone\n");
665 	WARN_ON(!pte_protnone(pte));
666 	WARN_ON(!pte_present(pte));
667 }
668 
669 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
pmd_protnone_tests(struct pgtable_debug_args * args)670 static void __init pmd_protnone_tests(struct pgtable_debug_args *args)
671 {
672 	pmd_t pmd;
673 
674 	if (!IS_ENABLED(CONFIG_NUMA_BALANCING))
675 		return;
676 
677 	if (!has_transparent_hugepage())
678 		return;
679 
680 	pr_debug("Validating PMD protnone\n");
681 	pmd = pmd_mkhuge(pfn_pmd(args->fixed_pmd_pfn, args->page_prot_none));
682 	WARN_ON(!pmd_protnone(pmd));
683 	WARN_ON(!pmd_present(pmd));
684 }
685 #else  /* !CONFIG_TRANSPARENT_HUGEPAGE */
pmd_protnone_tests(struct pgtable_debug_args * args)686 static void __init pmd_protnone_tests(struct pgtable_debug_args *args) { }
687 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
688 
pte_soft_dirty_tests(struct pgtable_debug_args * args)689 static void __init pte_soft_dirty_tests(struct pgtable_debug_args *args)
690 {
691 	pte_t pte = pfn_pte(args->fixed_pte_pfn, args->page_prot);
692 
693 	if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY))
694 		return;
695 
696 	pr_debug("Validating PTE soft dirty\n");
697 	WARN_ON(!pte_soft_dirty(pte_mksoft_dirty(pte)));
698 	WARN_ON(pte_soft_dirty(pte_clear_soft_dirty(pte)));
699 }
700 
pte_swap_soft_dirty_tests(struct pgtable_debug_args * args)701 static void __init pte_swap_soft_dirty_tests(struct pgtable_debug_args *args)
702 {
703 	pte_t pte;
704 
705 	if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY))
706 		return;
707 
708 	pr_debug("Validating PTE swap soft dirty\n");
709 	pte = swp_entry_to_pte(args->swp_entry);
710 	WARN_ON(!is_swap_pte(pte));
711 
712 	WARN_ON(!pte_swp_soft_dirty(pte_swp_mksoft_dirty(pte)));
713 	WARN_ON(pte_swp_soft_dirty(pte_swp_clear_soft_dirty(pte)));
714 }
715 
716 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
pmd_soft_dirty_tests(struct pgtable_debug_args * args)717 static void __init pmd_soft_dirty_tests(struct pgtable_debug_args *args)
718 {
719 	pmd_t pmd;
720 
721 	if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY))
722 		return;
723 
724 	if (!has_transparent_hugepage())
725 		return;
726 
727 	pr_debug("Validating PMD soft dirty\n");
728 	pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot);
729 	WARN_ON(!pmd_soft_dirty(pmd_mksoft_dirty(pmd)));
730 	WARN_ON(pmd_soft_dirty(pmd_clear_soft_dirty(pmd)));
731 }
732 
pmd_swap_soft_dirty_tests(struct pgtable_debug_args * args)733 static void __init pmd_swap_soft_dirty_tests(struct pgtable_debug_args *args)
734 {
735 	pmd_t pmd;
736 
737 	if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) ||
738 		!IS_ENABLED(CONFIG_ARCH_ENABLE_THP_MIGRATION))
739 		return;
740 
741 	if (!has_transparent_hugepage())
742 		return;
743 
744 	pr_debug("Validating PMD swap soft dirty\n");
745 	pmd = swp_entry_to_pmd(args->swp_entry);
746 	WARN_ON(!is_swap_pmd(pmd));
747 
748 	WARN_ON(!pmd_swp_soft_dirty(pmd_swp_mksoft_dirty(pmd)));
749 	WARN_ON(pmd_swp_soft_dirty(pmd_swp_clear_soft_dirty(pmd)));
750 }
751 #else  /* !CONFIG_TRANSPARENT_HUGEPAGE */
pmd_soft_dirty_tests(struct pgtable_debug_args * args)752 static void __init pmd_soft_dirty_tests(struct pgtable_debug_args *args) { }
pmd_swap_soft_dirty_tests(struct pgtable_debug_args * args)753 static void __init pmd_swap_soft_dirty_tests(struct pgtable_debug_args *args) { }
754 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
755 
pte_swap_exclusive_tests(struct pgtable_debug_args * args)756 static void __init pte_swap_exclusive_tests(struct pgtable_debug_args *args)
757 {
758 	swp_entry_t entry, entry2;
759 	pte_t pte;
760 
761 	pr_debug("Validating PTE swap exclusive\n");
762 	entry = args->swp_entry;
763 
764 	pte = swp_entry_to_pte(entry);
765 	WARN_ON(pte_swp_exclusive(pte));
766 	WARN_ON(!is_swap_pte(pte));
767 	entry2 = pte_to_swp_entry(pte);
768 	WARN_ON(memcmp(&entry, &entry2, sizeof(entry)));
769 
770 	pte = pte_swp_mkexclusive(pte);
771 	WARN_ON(!pte_swp_exclusive(pte));
772 	WARN_ON(!is_swap_pte(pte));
773 	WARN_ON(pte_swp_soft_dirty(pte));
774 	entry2 = pte_to_swp_entry(pte);
775 	WARN_ON(memcmp(&entry, &entry2, sizeof(entry)));
776 
777 	pte = pte_swp_clear_exclusive(pte);
778 	WARN_ON(pte_swp_exclusive(pte));
779 	WARN_ON(!is_swap_pte(pte));
780 	entry2 = pte_to_swp_entry(pte);
781 	WARN_ON(memcmp(&entry, &entry2, sizeof(entry)));
782 }
783 
pte_swap_tests(struct pgtable_debug_args * args)784 static void __init pte_swap_tests(struct pgtable_debug_args *args)
785 {
786 	swp_entry_t arch_entry;
787 	pte_t pte1, pte2;
788 
789 	pr_debug("Validating PTE swap\n");
790 	pte1 = swp_entry_to_pte(args->swp_entry);
791 	WARN_ON(!is_swap_pte(pte1));
792 
793 	arch_entry = __pte_to_swp_entry(pte1);
794 	pte2 = __swp_entry_to_pte(arch_entry);
795 	WARN_ON(memcmp(&pte1, &pte2, sizeof(pte1)));
796 }
797 
798 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
pmd_swap_tests(struct pgtable_debug_args * args)799 static void __init pmd_swap_tests(struct pgtable_debug_args *args)
800 {
801 	swp_entry_t arch_entry;
802 	pmd_t pmd1, pmd2;
803 
804 	if (!has_transparent_hugepage())
805 		return;
806 
807 	pr_debug("Validating PMD swap\n");
808 	pmd1 = swp_entry_to_pmd(args->swp_entry);
809 	WARN_ON(!is_swap_pmd(pmd1));
810 
811 	arch_entry = __pmd_to_swp_entry(pmd1);
812 	pmd2 = __swp_entry_to_pmd(arch_entry);
813 	WARN_ON(memcmp(&pmd1, &pmd2, sizeof(pmd1)));
814 }
815 #else  /* !CONFIG_ARCH_ENABLE_THP_MIGRATION */
pmd_swap_tests(struct pgtable_debug_args * args)816 static void __init pmd_swap_tests(struct pgtable_debug_args *args) { }
817 #endif /* CONFIG_ARCH_ENABLE_THP_MIGRATION */
818 
swap_migration_tests(struct pgtable_debug_args * args)819 static void __init swap_migration_tests(struct pgtable_debug_args *args)
820 {
821 	struct page *page;
822 	swp_entry_t swp;
823 
824 	if (!IS_ENABLED(CONFIG_MIGRATION))
825 		return;
826 
827 	/*
828 	 * swap_migration_tests() requires a dedicated page as it needs to
829 	 * be locked before creating a migration entry from it. Locking the
830 	 * page that actually maps kernel text ('start_kernel') can be real
831 	 * problematic. Lets use the allocated page explicitly for this
832 	 * purpose.
833 	 */
834 	page = (args->pte_pfn != ULONG_MAX) ? pfn_to_page(args->pte_pfn) : NULL;
835 	if (!page)
836 		return;
837 
838 	pr_debug("Validating swap migration\n");
839 
840 	/*
841 	 * make_[readable|writable]_migration_entry() expects given page to
842 	 * be locked, otherwise it stumbles upon a BUG_ON().
843 	 */
844 	__SetPageLocked(page);
845 	swp = make_writable_migration_entry(page_to_pfn(page));
846 	WARN_ON(!is_migration_entry(swp));
847 	WARN_ON(!is_writable_migration_entry(swp));
848 
849 	swp = make_readable_migration_entry(swp_offset(swp));
850 	WARN_ON(!is_migration_entry(swp));
851 	WARN_ON(is_writable_migration_entry(swp));
852 
853 	swp = make_readable_migration_entry(page_to_pfn(page));
854 	WARN_ON(!is_migration_entry(swp));
855 	WARN_ON(is_writable_migration_entry(swp));
856 	__ClearPageLocked(page);
857 }
858 
859 #ifdef CONFIG_HUGETLB_PAGE
hugetlb_basic_tests(struct pgtable_debug_args * args)860 static void __init hugetlb_basic_tests(struct pgtable_debug_args *args)
861 {
862 	pte_t pte;
863 
864 	pr_debug("Validating HugeTLB basic\n");
865 	pte = pfn_pte(args->fixed_pmd_pfn, args->page_prot);
866 	pte = arch_make_huge_pte(pte, PMD_SHIFT, VM_ACCESS_FLAGS);
867 
868 #ifdef CONFIG_ARCH_WANT_GENERAL_HUGETLB
869 	WARN_ON(!pte_huge(pte));
870 #endif
871 	WARN_ON(!huge_pte_dirty(huge_pte_mkdirty(pte)));
872 	WARN_ON(!huge_pte_write(huge_pte_mkwrite(huge_pte_wrprotect(pte))));
873 	WARN_ON(huge_pte_write(huge_pte_wrprotect(huge_pte_mkwrite(pte))));
874 }
875 #else  /* !CONFIG_HUGETLB_PAGE */
hugetlb_basic_tests(struct pgtable_debug_args * args)876 static void __init hugetlb_basic_tests(struct pgtable_debug_args *args) { }
877 #endif /* CONFIG_HUGETLB_PAGE */
878 
879 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
pmd_thp_tests(struct pgtable_debug_args * args)880 static void __init pmd_thp_tests(struct pgtable_debug_args *args)
881 {
882 	pmd_t pmd;
883 
884 	if (!has_transparent_hugepage())
885 		return;
886 
887 	pr_debug("Validating PMD based THP\n");
888 	/*
889 	 * pmd_trans_huge() and pmd_present() must return positive after
890 	 * MMU invalidation with pmd_mkinvalid(). This behavior is an
891 	 * optimization for transparent huge page. pmd_trans_huge() must
892 	 * be true if pmd_page() returns a valid THP to avoid taking the
893 	 * pmd_lock when others walk over non transhuge pmds (i.e. there
894 	 * are no THP allocated). Especially when splitting a THP and
895 	 * removing the present bit from the pmd, pmd_trans_huge() still
896 	 * needs to return true. pmd_present() should be true whenever
897 	 * pmd_trans_huge() returns true.
898 	 */
899 	pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot);
900 	WARN_ON(!pmd_trans_huge(pmd_mkhuge(pmd)));
901 
902 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
903 	WARN_ON(!pmd_trans_huge(pmd_mkinvalid(pmd_mkhuge(pmd))));
904 	WARN_ON(!pmd_present(pmd_mkinvalid(pmd_mkhuge(pmd))));
905 	WARN_ON(!pmd_leaf(pmd_mkinvalid(pmd_mkhuge(pmd))));
906 #endif /* __HAVE_ARCH_PMDP_INVALIDATE */
907 }
908 
909 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
pud_thp_tests(struct pgtable_debug_args * args)910 static void __init pud_thp_tests(struct pgtable_debug_args *args)
911 {
912 	pud_t pud;
913 
914 	if (!has_transparent_pud_hugepage())
915 		return;
916 
917 	pr_debug("Validating PUD based THP\n");
918 	pud = pfn_pud(args->fixed_pud_pfn, args->page_prot);
919 	WARN_ON(!pud_trans_huge(pud_mkhuge(pud)));
920 
921 	/*
922 	 * pud_mkinvalid() has been dropped for now. Enable back
923 	 * these tests when it comes back with a modified pud_present().
924 	 *
925 	 * WARN_ON(!pud_trans_huge(pud_mkinvalid(pud_mkhuge(pud))));
926 	 * WARN_ON(!pud_present(pud_mkinvalid(pud_mkhuge(pud))));
927 	 */
928 }
929 #else  /* !CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
pud_thp_tests(struct pgtable_debug_args * args)930 static void __init pud_thp_tests(struct pgtable_debug_args *args) { }
931 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
932 #else  /* !CONFIG_TRANSPARENT_HUGEPAGE */
pmd_thp_tests(struct pgtable_debug_args * args)933 static void __init pmd_thp_tests(struct pgtable_debug_args *args) { }
pud_thp_tests(struct pgtable_debug_args * args)934 static void __init pud_thp_tests(struct pgtable_debug_args *args) { }
935 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
936 
get_random_vaddr(void)937 static unsigned long __init get_random_vaddr(void)
938 {
939 	unsigned long random_vaddr, random_pages, total_user_pages;
940 
941 	total_user_pages = (TASK_SIZE - FIRST_USER_ADDRESS) / PAGE_SIZE;
942 
943 	random_pages = get_random_long() % total_user_pages;
944 	random_vaddr = FIRST_USER_ADDRESS + random_pages * PAGE_SIZE;
945 
946 	return random_vaddr;
947 }
948 
destroy_args(struct pgtable_debug_args * args)949 static void __init destroy_args(struct pgtable_debug_args *args)
950 {
951 	struct page *page = NULL;
952 
953 	/* Free (huge) page */
954 	if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
955 	    has_transparent_pud_hugepage() &&
956 	    args->pud_pfn != ULONG_MAX) {
957 		if (args->is_contiguous_page) {
958 			free_contig_range(args->pud_pfn,
959 					  (1 << (HPAGE_PUD_SHIFT - PAGE_SHIFT)));
960 		} else {
961 			page = pfn_to_page(args->pud_pfn);
962 			__free_pages(page, HPAGE_PUD_SHIFT - PAGE_SHIFT);
963 		}
964 
965 		args->pud_pfn = ULONG_MAX;
966 		args->pmd_pfn = ULONG_MAX;
967 		args->pte_pfn = ULONG_MAX;
968 	}
969 
970 	if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
971 	    has_transparent_hugepage() &&
972 	    args->pmd_pfn != ULONG_MAX) {
973 		if (args->is_contiguous_page) {
974 			free_contig_range(args->pmd_pfn, (1 << HPAGE_PMD_ORDER));
975 		} else {
976 			page = pfn_to_page(args->pmd_pfn);
977 			__free_pages(page, HPAGE_PMD_ORDER);
978 		}
979 
980 		args->pmd_pfn = ULONG_MAX;
981 		args->pte_pfn = ULONG_MAX;
982 	}
983 
984 	if (args->pte_pfn != ULONG_MAX) {
985 		page = pfn_to_page(args->pte_pfn);
986 		__free_page(page);
987 
988 		args->pte_pfn = ULONG_MAX;
989 	}
990 
991 	/* Free page table entries */
992 	if (args->start_ptep) {
993 		pmd_clear(args->pmdp);
994 		pte_free(args->mm, args->start_ptep);
995 		mm_dec_nr_ptes(args->mm);
996 	}
997 
998 	if (args->start_pmdp) {
999 		pud_clear(args->pudp);
1000 		pmd_free(args->mm, args->start_pmdp);
1001 		mm_dec_nr_pmds(args->mm);
1002 	}
1003 
1004 	if (args->start_pudp) {
1005 		p4d_clear(args->p4dp);
1006 		pud_free(args->mm, args->start_pudp);
1007 		mm_dec_nr_puds(args->mm);
1008 	}
1009 
1010 	if (args->start_p4dp) {
1011 		pgd_clear(args->pgdp);
1012 		p4d_free(args->mm, args->start_p4dp);
1013 	}
1014 
1015 	/* Free vma and mm struct */
1016 	if (args->vma)
1017 		vm_area_free(args->vma);
1018 
1019 	if (args->mm)
1020 		mmput(args->mm);
1021 }
1022 
1023 static struct page * __init
debug_vm_pgtable_alloc_huge_page(struct pgtable_debug_args * args,int order)1024 debug_vm_pgtable_alloc_huge_page(struct pgtable_debug_args *args, int order)
1025 {
1026 	struct page *page = NULL;
1027 
1028 #ifdef CONFIG_CONTIG_ALLOC
1029 	if (order > MAX_PAGE_ORDER) {
1030 		page = alloc_contig_pages((1 << order), GFP_KERNEL,
1031 					  first_online_node, NULL);
1032 		if (page) {
1033 			args->is_contiguous_page = true;
1034 			return page;
1035 		}
1036 	}
1037 #endif
1038 
1039 	if (order <= MAX_PAGE_ORDER)
1040 		page = alloc_pages(GFP_KERNEL, order);
1041 
1042 	return page;
1043 }
1044 
1045 /*
1046  * Check if a physical memory range described by <pstart, pend> contains
1047  * an area that is of size psize, and aligned to psize.
1048  *
1049  * Don't use address 0, an all-zeroes physical address might mask bugs, and
1050  * it's not used on x86.
1051  */
phys_align_check(phys_addr_t pstart,phys_addr_t pend,unsigned long psize,phys_addr_t * physp,unsigned long * alignp)1052 static void  __init phys_align_check(phys_addr_t pstart,
1053 				     phys_addr_t pend, unsigned long psize,
1054 				     phys_addr_t *physp, unsigned long *alignp)
1055 {
1056 	phys_addr_t aligned_start, aligned_end;
1057 
1058 	if (pstart == 0)
1059 		pstart = PAGE_SIZE;
1060 
1061 	aligned_start = ALIGN(pstart, psize);
1062 	aligned_end = aligned_start + psize;
1063 
1064 	if (aligned_end > aligned_start && aligned_end <= pend) {
1065 		*alignp = psize;
1066 		*physp = aligned_start;
1067 	}
1068 }
1069 
init_fixed_pfns(struct pgtable_debug_args * args)1070 static void __init init_fixed_pfns(struct pgtable_debug_args *args)
1071 {
1072 	u64 idx;
1073 	phys_addr_t phys, pstart, pend;
1074 
1075 	/*
1076 	 * Initialize the fixed pfns. To do this, try to find a
1077 	 * valid physical range, preferably aligned to PUD_SIZE,
1078 	 * but settling for aligned to PMD_SIZE as a fallback. If
1079 	 * neither of those is found, use the physical address of
1080 	 * the start_kernel symbol.
1081 	 *
1082 	 * The memory doesn't need to be allocated, it just needs to exist
1083 	 * as usable memory. It won't be touched.
1084 	 *
1085 	 * The alignment is recorded, and can be checked to see if we
1086 	 * can run the tests that require an actual valid physical
1087 	 * address range on some architectures ({pmd,pud}_huge_test
1088 	 * on x86).
1089 	 */
1090 
1091 	phys = __pa_symbol(&start_kernel);
1092 	args->fixed_alignment = PAGE_SIZE;
1093 
1094 	for_each_mem_range(idx, &pstart, &pend) {
1095 		/* First check for a PUD-aligned area */
1096 		phys_align_check(pstart, pend, PUD_SIZE, &phys,
1097 				 &args->fixed_alignment);
1098 
1099 		/* If a PUD-aligned area is found, we're done */
1100 		if (args->fixed_alignment == PUD_SIZE)
1101 			break;
1102 
1103 		/*
1104 		 * If no PMD-aligned area found yet, check for one,
1105 		 * but continue the loop to look for a PUD-aligned area.
1106 		 */
1107 		if (args->fixed_alignment < PMD_SIZE)
1108 			phys_align_check(pstart, pend, PMD_SIZE, &phys,
1109 					 &args->fixed_alignment);
1110 	}
1111 
1112 	args->fixed_pgd_pfn = __phys_to_pfn(phys & PGDIR_MASK);
1113 	args->fixed_p4d_pfn = __phys_to_pfn(phys & P4D_MASK);
1114 	args->fixed_pud_pfn = __phys_to_pfn(phys & PUD_MASK);
1115 	args->fixed_pmd_pfn = __phys_to_pfn(phys & PMD_MASK);
1116 	args->fixed_pte_pfn = __phys_to_pfn(phys & PAGE_MASK);
1117 	WARN_ON(!pfn_valid(args->fixed_pte_pfn));
1118 }
1119 
1120 
init_args(struct pgtable_debug_args * args)1121 static int __init init_args(struct pgtable_debug_args *args)
1122 {
1123 	unsigned long max_swap_offset;
1124 	struct page *page = NULL;
1125 	int ret = 0;
1126 
1127 	/*
1128 	 * Initialize the debugging data.
1129 	 *
1130 	 * vm_get_page_prot(VM_NONE) or vm_get_page_prot(VM_SHARED|VM_NONE)
1131 	 * will help create page table entries with PROT_NONE permission as
1132 	 * required for pxx_protnone_tests().
1133 	 */
1134 	memset(args, 0, sizeof(*args));
1135 	args->vaddr              = get_random_vaddr();
1136 	args->page_prot          = vm_get_page_prot(VM_ACCESS_FLAGS);
1137 	args->page_prot_none     = vm_get_page_prot(VM_NONE);
1138 	args->is_contiguous_page = false;
1139 	args->pud_pfn            = ULONG_MAX;
1140 	args->pmd_pfn            = ULONG_MAX;
1141 	args->pte_pfn            = ULONG_MAX;
1142 	args->fixed_pgd_pfn      = ULONG_MAX;
1143 	args->fixed_p4d_pfn      = ULONG_MAX;
1144 	args->fixed_pud_pfn      = ULONG_MAX;
1145 	args->fixed_pmd_pfn      = ULONG_MAX;
1146 	args->fixed_pte_pfn      = ULONG_MAX;
1147 
1148 	/* Allocate mm and vma */
1149 	args->mm = mm_alloc();
1150 	if (!args->mm) {
1151 		pr_err("Failed to allocate mm struct\n");
1152 		ret = -ENOMEM;
1153 		goto error;
1154 	}
1155 
1156 	args->vma = vm_area_alloc(args->mm);
1157 	if (!args->vma) {
1158 		pr_err("Failed to allocate vma\n");
1159 		ret = -ENOMEM;
1160 		goto error;
1161 	}
1162 
1163 	/*
1164 	 * Allocate page table entries. They will be modified in the tests.
1165 	 * Lets save the page table entries so that they can be released
1166 	 * when the tests are completed.
1167 	 */
1168 	args->pgdp = pgd_offset(args->mm, args->vaddr);
1169 	args->p4dp = p4d_alloc(args->mm, args->pgdp, args->vaddr);
1170 	if (!args->p4dp) {
1171 		pr_err("Failed to allocate p4d entries\n");
1172 		ret = -ENOMEM;
1173 		goto error;
1174 	}
1175 	args->start_p4dp = p4d_offset(args->pgdp, 0UL);
1176 	WARN_ON(!args->start_p4dp);
1177 
1178 	args->pudp = pud_alloc(args->mm, args->p4dp, args->vaddr);
1179 	if (!args->pudp) {
1180 		pr_err("Failed to allocate pud entries\n");
1181 		ret = -ENOMEM;
1182 		goto error;
1183 	}
1184 	args->start_pudp = pud_offset(args->p4dp, 0UL);
1185 	WARN_ON(!args->start_pudp);
1186 
1187 	args->pmdp = pmd_alloc(args->mm, args->pudp, args->vaddr);
1188 	if (!args->pmdp) {
1189 		pr_err("Failed to allocate pmd entries\n");
1190 		ret = -ENOMEM;
1191 		goto error;
1192 	}
1193 	args->start_pmdp = pmd_offset(args->pudp, 0UL);
1194 	WARN_ON(!args->start_pmdp);
1195 
1196 	if (pte_alloc(args->mm, args->pmdp)) {
1197 		pr_err("Failed to allocate pte entries\n");
1198 		ret = -ENOMEM;
1199 		goto error;
1200 	}
1201 	args->start_ptep = pmd_pgtable(pmdp_get(args->pmdp));
1202 	WARN_ON(!args->start_ptep);
1203 
1204 	init_fixed_pfns(args);
1205 
1206 	/* See generic_max_swapfile_size(): probe the maximum offset */
1207 	max_swap_offset = swp_offset(pte_to_swp_entry(swp_entry_to_pte(swp_entry(0, ~0UL))));
1208 	/* Create a swp entry with all possible bits set */
1209 	args->swp_entry = swp_entry((1 << MAX_SWAPFILES_SHIFT) - 1, max_swap_offset);
1210 
1211 	/*
1212 	 * Allocate (huge) pages because some of the tests need to access
1213 	 * the data in the pages. The corresponding tests will be skipped
1214 	 * if we fail to allocate (huge) pages.
1215 	 */
1216 	if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
1217 	    has_transparent_pud_hugepage()) {
1218 		page = debug_vm_pgtable_alloc_huge_page(args,
1219 				HPAGE_PUD_SHIFT - PAGE_SHIFT);
1220 		if (page) {
1221 			args->pud_pfn = page_to_pfn(page);
1222 			args->pmd_pfn = args->pud_pfn;
1223 			args->pte_pfn = args->pud_pfn;
1224 			return 0;
1225 		}
1226 	}
1227 
1228 	if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
1229 	    has_transparent_hugepage()) {
1230 		page = debug_vm_pgtable_alloc_huge_page(args, HPAGE_PMD_ORDER);
1231 		if (page) {
1232 			args->pmd_pfn = page_to_pfn(page);
1233 			args->pte_pfn = args->pmd_pfn;
1234 			return 0;
1235 		}
1236 	}
1237 
1238 	page = alloc_page(GFP_KERNEL);
1239 	if (page)
1240 		args->pte_pfn = page_to_pfn(page);
1241 
1242 	return 0;
1243 
1244 error:
1245 	destroy_args(args);
1246 	return ret;
1247 }
1248 
debug_vm_pgtable(void)1249 static int __init debug_vm_pgtable(void)
1250 {
1251 	struct pgtable_debug_args args;
1252 	spinlock_t *ptl = NULL;
1253 	int idx, ret;
1254 
1255 	pr_info("Validating architecture page table helpers\n");
1256 	ret = init_args(&args);
1257 	if (ret)
1258 		return ret;
1259 
1260 	/*
1261 	 * Iterate over each possible vm_flags to make sure that all
1262 	 * the basic page table transformation validations just hold
1263 	 * true irrespective of the starting protection value for a
1264 	 * given page table entry.
1265 	 *
1266 	 * Protection based vm_flags combinations are always linear
1267 	 * and increasing i.e starting from VM_NONE and going up to
1268 	 * (VM_SHARED | READ | WRITE | EXEC).
1269 	 */
1270 #define VM_FLAGS_START	(VM_NONE)
1271 #define VM_FLAGS_END	(VM_SHARED | VM_EXEC | VM_WRITE | VM_READ)
1272 
1273 	for (idx = VM_FLAGS_START; idx <= VM_FLAGS_END; idx++) {
1274 		pte_basic_tests(&args, idx);
1275 		pmd_basic_tests(&args, idx);
1276 		pud_basic_tests(&args, idx);
1277 	}
1278 
1279 	/*
1280 	 * Both P4D and PGD level tests are very basic which do not
1281 	 * involve creating page table entries from the protection
1282 	 * value and the given pfn. Hence just keep them out from
1283 	 * the above iteration for now to save some test execution
1284 	 * time.
1285 	 */
1286 	p4d_basic_tests(&args);
1287 	pgd_basic_tests(&args);
1288 
1289 	pmd_leaf_tests(&args);
1290 	pud_leaf_tests(&args);
1291 
1292 	pte_special_tests(&args);
1293 	pte_protnone_tests(&args);
1294 	pmd_protnone_tests(&args);
1295 
1296 	pte_soft_dirty_tests(&args);
1297 	pmd_soft_dirty_tests(&args);
1298 	pte_swap_soft_dirty_tests(&args);
1299 	pmd_swap_soft_dirty_tests(&args);
1300 
1301 	pte_swap_exclusive_tests(&args);
1302 
1303 	pte_swap_tests(&args);
1304 	pmd_swap_tests(&args);
1305 
1306 	swap_migration_tests(&args);
1307 
1308 	pmd_thp_tests(&args);
1309 	pud_thp_tests(&args);
1310 
1311 	hugetlb_basic_tests(&args);
1312 
1313 	/*
1314 	 * Page table modifying tests. They need to hold
1315 	 * proper page table lock.
1316 	 */
1317 
1318 	args.ptep = pte_offset_map_lock(args.mm, args.pmdp, args.vaddr, &ptl);
1319 	pte_clear_tests(&args);
1320 	pte_advanced_tests(&args);
1321 	if (args.ptep)
1322 		pte_unmap_unlock(args.ptep, ptl);
1323 
1324 	ptl = pmd_lock(args.mm, args.pmdp);
1325 	pmd_clear_tests(&args);
1326 	pmd_advanced_tests(&args);
1327 	pmd_huge_tests(&args);
1328 	pmd_populate_tests(&args);
1329 	spin_unlock(ptl);
1330 
1331 	ptl = pud_lock(args.mm, args.pudp);
1332 	pud_clear_tests(&args);
1333 	pud_advanced_tests(&args);
1334 	pud_huge_tests(&args);
1335 	pud_populate_tests(&args);
1336 	spin_unlock(ptl);
1337 
1338 	spin_lock(&(args.mm->page_table_lock));
1339 	p4d_clear_tests(&args);
1340 	pgd_clear_tests(&args);
1341 	p4d_populate_tests(&args);
1342 	pgd_populate_tests(&args);
1343 	spin_unlock(&(args.mm->page_table_lock));
1344 
1345 	destroy_args(&args);
1346 	return 0;
1347 }
1348 late_initcall(debug_vm_pgtable);
1349