1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * This kernel test validates architecture page table helpers and
4 * accessors and helps in verifying their continued compliance with
5 * expected generic MM semantics.
6 *
7 * Copyright (C) 2019 ARM Ltd.
8 *
9 * Author: Anshuman Khandual <anshuman.khandual@arm.com>
10 */
11 #define pr_fmt(fmt) "debug_vm_pgtable: [%-25s]: " fmt, __func__
12
13 #include <linux/gfp.h>
14 #include <linux/highmem.h>
15 #include <linux/hugetlb.h>
16 #include <linux/kernel.h>
17 #include <linux/kconfig.h>
18 #include <linux/mm.h>
19 #include <linux/mman.h>
20 #include <linux/mm_types.h>
21 #include <linux/module.h>
22 #include <linux/pfn_t.h>
23 #include <linux/printk.h>
24 #include <linux/pgtable.h>
25 #include <linux/random.h>
26 #include <linux/spinlock.h>
27 #include <linux/swap.h>
28 #include <linux/swapops.h>
29 #include <linux/start_kernel.h>
30 #include <linux/sched/mm.h>
31 #include <linux/io.h>
32
33 #include <asm/cacheflush.h>
34 #include <asm/pgalloc.h>
35 #include <asm/tlbflush.h>
36
37 /*
38 * Please refer Documentation/vm/arch_pgtable_helpers.rst for the semantics
39 * expectations that are being validated here. All future changes in here
40 * or the documentation need to be in sync.
41 */
42
43 #define VMFLAGS (VM_READ|VM_WRITE|VM_EXEC)
44
45 /*
46 * On s390 platform, the lower 4 bits are used to identify given page table
47 * entry type. But these bits might affect the ability to clear entries with
48 * pxx_clear() because of how dynamic page table folding works on s390. So
49 * while loading up the entries do not change the lower 4 bits. It does not
50 * have affect any other platform. Also avoid the 62nd bit on ppc64 that is
51 * used to mark a pte entry.
52 */
53 #define S390_SKIP_MASK GENMASK(3, 0)
54 #if __BITS_PER_LONG == 64
55 #define PPC64_SKIP_MASK GENMASK(62, 62)
56 #else
57 #define PPC64_SKIP_MASK 0x0
58 #endif
59 #define ARCH_SKIP_MASK (S390_SKIP_MASK | PPC64_SKIP_MASK)
60 #define RANDOM_ORVALUE (GENMASK(BITS_PER_LONG - 1, 0) & ~ARCH_SKIP_MASK)
61 #define RANDOM_NZVALUE GENMASK(7, 0)
62
63 struct pgtable_debug_args {
64 struct mm_struct *mm;
65 struct vm_area_struct *vma;
66
67 pgd_t *pgdp;
68 p4d_t *p4dp;
69 pud_t *pudp;
70 pmd_t *pmdp;
71 pte_t *ptep;
72
73 p4d_t *start_p4dp;
74 pud_t *start_pudp;
75 pmd_t *start_pmdp;
76 pgtable_t start_ptep;
77
78 unsigned long vaddr;
79 pgprot_t page_prot;
80 pgprot_t page_prot_none;
81
82 bool is_contiguous_page;
83 unsigned long pud_pfn;
84 unsigned long pmd_pfn;
85 unsigned long pte_pfn;
86
87 unsigned long fixed_pgd_pfn;
88 unsigned long fixed_p4d_pfn;
89 unsigned long fixed_pud_pfn;
90 unsigned long fixed_pmd_pfn;
91 unsigned long fixed_pte_pfn;
92 };
93
pte_basic_tests(struct pgtable_debug_args * args,int idx)94 static void __init pte_basic_tests(struct pgtable_debug_args *args, int idx)
95 {
96 pgprot_t prot = protection_map[idx];
97 pte_t pte = pfn_pte(args->fixed_pte_pfn, prot);
98 unsigned long val = idx, *ptr = &val;
99
100 pr_debug("Validating PTE basic (%pGv)\n", ptr);
101
102 /*
103 * This test needs to be executed after the given page table entry
104 * is created with pfn_pte() to make sure that protection_map[idx]
105 * does not have the dirty bit enabled from the beginning. This is
106 * important for platforms like arm64 where (!PTE_RDONLY) indicate
107 * dirty bit being set.
108 */
109 WARN_ON(pte_dirty(pte_wrprotect(pte)));
110
111 WARN_ON(!pte_same(pte, pte));
112 WARN_ON(!pte_young(pte_mkyoung(pte_mkold(pte))));
113 WARN_ON(!pte_dirty(pte_mkdirty(pte_mkclean(pte))));
114 WARN_ON(!pte_write(pte_mkwrite(pte_wrprotect(pte))));
115 WARN_ON(pte_young(pte_mkold(pte_mkyoung(pte))));
116 WARN_ON(pte_dirty(pte_mkclean(pte_mkdirty(pte))));
117 WARN_ON(pte_write(pte_wrprotect(pte_mkwrite(pte))));
118 WARN_ON(pte_dirty(pte_wrprotect(pte_mkclean(pte))));
119 WARN_ON(!pte_dirty(pte_wrprotect(pte_mkdirty(pte))));
120 }
121
pte_advanced_tests(struct pgtable_debug_args * args)122 static void __init pte_advanced_tests(struct pgtable_debug_args *args)
123 {
124 struct page *page;
125 pte_t pte;
126
127 /*
128 * Architectures optimize set_pte_at by avoiding TLB flush.
129 * This requires set_pte_at to be not used to update an
130 * existing pte entry. Clear pte before we do set_pte_at
131 *
132 * flush_dcache_page() is called after set_pte_at() to clear
133 * PG_arch_1 for the page on ARM64. The page flag isn't cleared
134 * when it's released and page allocation check will fail when
135 * the page is allocated again. For architectures other than ARM64,
136 * the unexpected overhead of cache flushing is acceptable.
137 */
138 page = (args->pte_pfn != ULONG_MAX) ? pfn_to_page(args->pte_pfn) : NULL;
139 if (!page)
140 return;
141
142 pr_debug("Validating PTE advanced\n");
143 pte = pfn_pte(args->pte_pfn, args->page_prot);
144 set_pte_at(args->mm, args->vaddr, args->ptep, pte);
145 flush_dcache_page(page);
146 ptep_set_wrprotect(args->mm, args->vaddr, args->ptep);
147 pte = ptep_get(args->ptep);
148 WARN_ON(pte_write(pte));
149 ptep_get_and_clear(args->mm, args->vaddr, args->ptep);
150 pte = ptep_get(args->ptep);
151 WARN_ON(!pte_none(pte));
152
153 pte = pfn_pte(args->pte_pfn, args->page_prot);
154 pte = pte_wrprotect(pte);
155 pte = pte_mkclean(pte);
156 set_pte_at(args->mm, args->vaddr, args->ptep, pte);
157 flush_dcache_page(page);
158 pte = pte_mkwrite(pte);
159 pte = pte_mkdirty(pte);
160 ptep_set_access_flags(args->vma, args->vaddr, args->ptep, pte, 1);
161 pte = ptep_get(args->ptep);
162 WARN_ON(!(pte_write(pte) && pte_dirty(pte)));
163 ptep_get_and_clear_full(args->mm, args->vaddr, args->ptep, 1);
164 pte = ptep_get(args->ptep);
165 WARN_ON(!pte_none(pte));
166
167 pte = pfn_pte(args->pte_pfn, args->page_prot);
168 pte = pte_mkyoung(pte);
169 set_pte_at(args->mm, args->vaddr, args->ptep, pte);
170 flush_dcache_page(page);
171 ptep_test_and_clear_young(args->vma, args->vaddr, args->ptep);
172 pte = ptep_get(args->ptep);
173 WARN_ON(pte_young(pte));
174
175 ptep_get_and_clear_full(args->mm, args->vaddr, args->ptep, 1);
176 }
177
pte_savedwrite_tests(struct pgtable_debug_args * args)178 static void __init pte_savedwrite_tests(struct pgtable_debug_args *args)
179 {
180 pte_t pte = pfn_pte(args->fixed_pte_pfn, args->page_prot_none);
181
182 if (!IS_ENABLED(CONFIG_NUMA_BALANCING))
183 return;
184
185 pr_debug("Validating PTE saved write\n");
186 WARN_ON(!pte_savedwrite(pte_mk_savedwrite(pte_clear_savedwrite(pte))));
187 WARN_ON(pte_savedwrite(pte_clear_savedwrite(pte_mk_savedwrite(pte))));
188 }
189
190 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
pmd_basic_tests(struct pgtable_debug_args * args,int idx)191 static void __init pmd_basic_tests(struct pgtable_debug_args *args, int idx)
192 {
193 pgprot_t prot = protection_map[idx];
194 unsigned long val = idx, *ptr = &val;
195 pmd_t pmd;
196
197 if (!has_transparent_hugepage())
198 return;
199
200 pr_debug("Validating PMD basic (%pGv)\n", ptr);
201 pmd = pfn_pmd(args->fixed_pmd_pfn, prot);
202
203 /*
204 * This test needs to be executed after the given page table entry
205 * is created with pfn_pmd() to make sure that protection_map[idx]
206 * does not have the dirty bit enabled from the beginning. This is
207 * important for platforms like arm64 where (!PTE_RDONLY) indicate
208 * dirty bit being set.
209 */
210 WARN_ON(pmd_dirty(pmd_wrprotect(pmd)));
211
212
213 WARN_ON(!pmd_same(pmd, pmd));
214 WARN_ON(!pmd_young(pmd_mkyoung(pmd_mkold(pmd))));
215 WARN_ON(!pmd_dirty(pmd_mkdirty(pmd_mkclean(pmd))));
216 WARN_ON(!pmd_write(pmd_mkwrite(pmd_wrprotect(pmd))));
217 WARN_ON(pmd_young(pmd_mkold(pmd_mkyoung(pmd))));
218 WARN_ON(pmd_dirty(pmd_mkclean(pmd_mkdirty(pmd))));
219 WARN_ON(pmd_write(pmd_wrprotect(pmd_mkwrite(pmd))));
220 WARN_ON(pmd_dirty(pmd_wrprotect(pmd_mkclean(pmd))));
221 WARN_ON(!pmd_dirty(pmd_wrprotect(pmd_mkdirty(pmd))));
222 /*
223 * A huge page does not point to next level page table
224 * entry. Hence this must qualify as pmd_bad().
225 */
226 WARN_ON(!pmd_bad(pmd_mkhuge(pmd)));
227 }
228
pmd_advanced_tests(struct pgtable_debug_args * args)229 static void __init pmd_advanced_tests(struct pgtable_debug_args *args)
230 {
231 struct page *page;
232 pmd_t pmd;
233 unsigned long vaddr = args->vaddr;
234
235 if (!has_transparent_hugepage())
236 return;
237
238 page = (args->pmd_pfn != ULONG_MAX) ? pfn_to_page(args->pmd_pfn) : NULL;
239 if (!page)
240 return;
241
242 /*
243 * flush_dcache_page() is called after set_pmd_at() to clear
244 * PG_arch_1 for the page on ARM64. The page flag isn't cleared
245 * when it's released and page allocation check will fail when
246 * the page is allocated again. For architectures other than ARM64,
247 * the unexpected overhead of cache flushing is acceptable.
248 */
249 pr_debug("Validating PMD advanced\n");
250 /* Align the address wrt HPAGE_PMD_SIZE */
251 vaddr &= HPAGE_PMD_MASK;
252
253 pgtable_trans_huge_deposit(args->mm, args->pmdp, args->start_ptep);
254
255 pmd = pfn_pmd(args->pmd_pfn, args->page_prot);
256 set_pmd_at(args->mm, vaddr, args->pmdp, pmd);
257 flush_dcache_page(page);
258 pmdp_set_wrprotect(args->mm, vaddr, args->pmdp);
259 pmd = READ_ONCE(*args->pmdp);
260 WARN_ON(pmd_write(pmd));
261 pmdp_huge_get_and_clear(args->mm, vaddr, args->pmdp);
262 pmd = READ_ONCE(*args->pmdp);
263 WARN_ON(!pmd_none(pmd));
264
265 pmd = pfn_pmd(args->pmd_pfn, args->page_prot);
266 pmd = pmd_wrprotect(pmd);
267 pmd = pmd_mkclean(pmd);
268 set_pmd_at(args->mm, vaddr, args->pmdp, pmd);
269 flush_dcache_page(page);
270 pmd = pmd_mkwrite(pmd);
271 pmd = pmd_mkdirty(pmd);
272 pmdp_set_access_flags(args->vma, vaddr, args->pmdp, pmd, 1);
273 pmd = READ_ONCE(*args->pmdp);
274 WARN_ON(!(pmd_write(pmd) && pmd_dirty(pmd)));
275 pmdp_huge_get_and_clear_full(args->vma, vaddr, args->pmdp, 1);
276 pmd = READ_ONCE(*args->pmdp);
277 WARN_ON(!pmd_none(pmd));
278
279 pmd = pmd_mkhuge(pfn_pmd(args->pmd_pfn, args->page_prot));
280 pmd = pmd_mkyoung(pmd);
281 set_pmd_at(args->mm, vaddr, args->pmdp, pmd);
282 flush_dcache_page(page);
283 pmdp_test_and_clear_young(args->vma, vaddr, args->pmdp);
284 pmd = READ_ONCE(*args->pmdp);
285 WARN_ON(pmd_young(pmd));
286
287 /* Clear the pte entries */
288 pmdp_huge_get_and_clear(args->mm, vaddr, args->pmdp);
289 pgtable_trans_huge_withdraw(args->mm, args->pmdp);
290 }
291
pmd_leaf_tests(struct pgtable_debug_args * args)292 static void __init pmd_leaf_tests(struct pgtable_debug_args *args)
293 {
294 pmd_t pmd;
295
296 if (!has_transparent_hugepage())
297 return;
298
299 pr_debug("Validating PMD leaf\n");
300 pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot);
301
302 /*
303 * PMD based THP is a leaf entry.
304 */
305 pmd = pmd_mkhuge(pmd);
306 WARN_ON(!pmd_leaf(pmd));
307 }
308
pmd_savedwrite_tests(struct pgtable_debug_args * args)309 static void __init pmd_savedwrite_tests(struct pgtable_debug_args *args)
310 {
311 pmd_t pmd;
312
313 if (!IS_ENABLED(CONFIG_NUMA_BALANCING))
314 return;
315
316 if (!has_transparent_hugepage())
317 return;
318
319 pr_debug("Validating PMD saved write\n");
320 pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot_none);
321 WARN_ON(!pmd_savedwrite(pmd_mk_savedwrite(pmd_clear_savedwrite(pmd))));
322 WARN_ON(pmd_savedwrite(pmd_clear_savedwrite(pmd_mk_savedwrite(pmd))));
323 }
324
325 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
pud_basic_tests(struct pgtable_debug_args * args,int idx)326 static void __init pud_basic_tests(struct pgtable_debug_args *args, int idx)
327 {
328 pgprot_t prot = protection_map[idx];
329 unsigned long val = idx, *ptr = &val;
330 pud_t pud;
331
332 if (!has_transparent_hugepage())
333 return;
334
335 pr_debug("Validating PUD basic (%pGv)\n", ptr);
336 pud = pfn_pud(args->fixed_pud_pfn, prot);
337
338 /*
339 * This test needs to be executed after the given page table entry
340 * is created with pfn_pud() to make sure that protection_map[idx]
341 * does not have the dirty bit enabled from the beginning. This is
342 * important for platforms like arm64 where (!PTE_RDONLY) indicate
343 * dirty bit being set.
344 */
345 WARN_ON(pud_dirty(pud_wrprotect(pud)));
346
347 WARN_ON(!pud_same(pud, pud));
348 WARN_ON(!pud_young(pud_mkyoung(pud_mkold(pud))));
349 WARN_ON(!pud_dirty(pud_mkdirty(pud_mkclean(pud))));
350 WARN_ON(pud_dirty(pud_mkclean(pud_mkdirty(pud))));
351 WARN_ON(!pud_write(pud_mkwrite(pud_wrprotect(pud))));
352 WARN_ON(pud_write(pud_wrprotect(pud_mkwrite(pud))));
353 WARN_ON(pud_young(pud_mkold(pud_mkyoung(pud))));
354 WARN_ON(pud_dirty(pud_wrprotect(pud_mkclean(pud))));
355 WARN_ON(!pud_dirty(pud_wrprotect(pud_mkdirty(pud))));
356
357 if (mm_pmd_folded(args->mm))
358 return;
359
360 /*
361 * A huge page does not point to next level page table
362 * entry. Hence this must qualify as pud_bad().
363 */
364 WARN_ON(!pud_bad(pud_mkhuge(pud)));
365 }
366
pud_advanced_tests(struct pgtable_debug_args * args)367 static void __init pud_advanced_tests(struct pgtable_debug_args *args)
368 {
369 struct page *page;
370 unsigned long vaddr = args->vaddr;
371 pud_t pud;
372
373 if (!has_transparent_hugepage())
374 return;
375
376 page = (args->pud_pfn != ULONG_MAX) ? pfn_to_page(args->pud_pfn) : NULL;
377 if (!page)
378 return;
379
380 /*
381 * flush_dcache_page() is called after set_pud_at() to clear
382 * PG_arch_1 for the page on ARM64. The page flag isn't cleared
383 * when it's released and page allocation check will fail when
384 * the page is allocated again. For architectures other than ARM64,
385 * the unexpected overhead of cache flushing is acceptable.
386 */
387 pr_debug("Validating PUD advanced\n");
388 /* Align the address wrt HPAGE_PUD_SIZE */
389 vaddr &= HPAGE_PUD_MASK;
390
391 pud = pfn_pud(args->pud_pfn, args->page_prot);
392 set_pud_at(args->mm, vaddr, args->pudp, pud);
393 flush_dcache_page(page);
394 pudp_set_wrprotect(args->mm, vaddr, args->pudp);
395 pud = READ_ONCE(*args->pudp);
396 WARN_ON(pud_write(pud));
397
398 #ifndef __PAGETABLE_PMD_FOLDED
399 pudp_huge_get_and_clear(args->mm, vaddr, args->pudp);
400 pud = READ_ONCE(*args->pudp);
401 WARN_ON(!pud_none(pud));
402 #endif /* __PAGETABLE_PMD_FOLDED */
403 pud = pfn_pud(args->pud_pfn, args->page_prot);
404 pud = pud_wrprotect(pud);
405 pud = pud_mkclean(pud);
406 set_pud_at(args->mm, vaddr, args->pudp, pud);
407 flush_dcache_page(page);
408 pud = pud_mkwrite(pud);
409 pud = pud_mkdirty(pud);
410 pudp_set_access_flags(args->vma, vaddr, args->pudp, pud, 1);
411 pud = READ_ONCE(*args->pudp);
412 WARN_ON(!(pud_write(pud) && pud_dirty(pud)));
413
414 #ifndef __PAGETABLE_PMD_FOLDED
415 pudp_huge_get_and_clear_full(args->mm, vaddr, args->pudp, 1);
416 pud = READ_ONCE(*args->pudp);
417 WARN_ON(!pud_none(pud));
418 #endif /* __PAGETABLE_PMD_FOLDED */
419
420 pud = pfn_pud(args->pud_pfn, args->page_prot);
421 pud = pud_mkyoung(pud);
422 set_pud_at(args->mm, vaddr, args->pudp, pud);
423 flush_dcache_page(page);
424 pudp_test_and_clear_young(args->vma, vaddr, args->pudp);
425 pud = READ_ONCE(*args->pudp);
426 WARN_ON(pud_young(pud));
427
428 pudp_huge_get_and_clear(args->mm, vaddr, args->pudp);
429 }
430
pud_leaf_tests(struct pgtable_debug_args * args)431 static void __init pud_leaf_tests(struct pgtable_debug_args *args)
432 {
433 pud_t pud;
434
435 if (!has_transparent_hugepage())
436 return;
437
438 pr_debug("Validating PUD leaf\n");
439 pud = pfn_pud(args->fixed_pud_pfn, args->page_prot);
440 /*
441 * PUD based THP is a leaf entry.
442 */
443 pud = pud_mkhuge(pud);
444 WARN_ON(!pud_leaf(pud));
445 }
446 #else /* !CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
pud_basic_tests(struct pgtable_debug_args * args,int idx)447 static void __init pud_basic_tests(struct pgtable_debug_args *args, int idx) { }
pud_advanced_tests(struct pgtable_debug_args * args)448 static void __init pud_advanced_tests(struct pgtable_debug_args *args) { }
pud_leaf_tests(struct pgtable_debug_args * args)449 static void __init pud_leaf_tests(struct pgtable_debug_args *args) { }
450 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
451 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
pmd_basic_tests(struct pgtable_debug_args * args,int idx)452 static void __init pmd_basic_tests(struct pgtable_debug_args *args, int idx) { }
pud_basic_tests(struct pgtable_debug_args * args,int idx)453 static void __init pud_basic_tests(struct pgtable_debug_args *args, int idx) { }
pmd_advanced_tests(struct pgtable_debug_args * args)454 static void __init pmd_advanced_tests(struct pgtable_debug_args *args) { }
pud_advanced_tests(struct pgtable_debug_args * args)455 static void __init pud_advanced_tests(struct pgtable_debug_args *args) { }
pmd_leaf_tests(struct pgtable_debug_args * args)456 static void __init pmd_leaf_tests(struct pgtable_debug_args *args) { }
pud_leaf_tests(struct pgtable_debug_args * args)457 static void __init pud_leaf_tests(struct pgtable_debug_args *args) { }
pmd_savedwrite_tests(struct pgtable_debug_args * args)458 static void __init pmd_savedwrite_tests(struct pgtable_debug_args *args) { }
459 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
460
461 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
pmd_huge_tests(struct pgtable_debug_args * args)462 static void __init pmd_huge_tests(struct pgtable_debug_args *args)
463 {
464 pmd_t pmd;
465
466 if (!arch_vmap_pmd_supported(args->page_prot))
467 return;
468
469 pr_debug("Validating PMD huge\n");
470 /*
471 * X86 defined pmd_set_huge() verifies that the given
472 * PMD is not a populated non-leaf entry.
473 */
474 WRITE_ONCE(*args->pmdp, __pmd(0));
475 WARN_ON(!pmd_set_huge(args->pmdp, __pfn_to_phys(args->fixed_pmd_pfn), args->page_prot));
476 WARN_ON(!pmd_clear_huge(args->pmdp));
477 pmd = READ_ONCE(*args->pmdp);
478 WARN_ON(!pmd_none(pmd));
479 }
480
pud_huge_tests(struct pgtable_debug_args * args)481 static void __init pud_huge_tests(struct pgtable_debug_args *args)
482 {
483 pud_t pud;
484
485 if (!arch_vmap_pud_supported(args->page_prot))
486 return;
487
488 pr_debug("Validating PUD huge\n");
489 /*
490 * X86 defined pud_set_huge() verifies that the given
491 * PUD is not a populated non-leaf entry.
492 */
493 WRITE_ONCE(*args->pudp, __pud(0));
494 WARN_ON(!pud_set_huge(args->pudp, __pfn_to_phys(args->fixed_pud_pfn), args->page_prot));
495 WARN_ON(!pud_clear_huge(args->pudp));
496 pud = READ_ONCE(*args->pudp);
497 WARN_ON(!pud_none(pud));
498 }
499 #else /* !CONFIG_HAVE_ARCH_HUGE_VMAP */
pmd_huge_tests(struct pgtable_debug_args * args)500 static void __init pmd_huge_tests(struct pgtable_debug_args *args) { }
pud_huge_tests(struct pgtable_debug_args * args)501 static void __init pud_huge_tests(struct pgtable_debug_args *args) { }
502 #endif /* CONFIG_HAVE_ARCH_HUGE_VMAP */
503
p4d_basic_tests(struct pgtable_debug_args * args)504 static void __init p4d_basic_tests(struct pgtable_debug_args *args)
505 {
506 p4d_t p4d;
507
508 pr_debug("Validating P4D basic\n");
509 memset(&p4d, RANDOM_NZVALUE, sizeof(p4d_t));
510 WARN_ON(!p4d_same(p4d, p4d));
511 }
512
pgd_basic_tests(struct pgtable_debug_args * args)513 static void __init pgd_basic_tests(struct pgtable_debug_args *args)
514 {
515 pgd_t pgd;
516
517 pr_debug("Validating PGD basic\n");
518 memset(&pgd, RANDOM_NZVALUE, sizeof(pgd_t));
519 WARN_ON(!pgd_same(pgd, pgd));
520 }
521
522 #ifndef __PAGETABLE_PUD_FOLDED
pud_clear_tests(struct pgtable_debug_args * args)523 static void __init pud_clear_tests(struct pgtable_debug_args *args)
524 {
525 pud_t pud = READ_ONCE(*args->pudp);
526
527 if (mm_pmd_folded(args->mm))
528 return;
529
530 pr_debug("Validating PUD clear\n");
531 pud = __pud(pud_val(pud) | RANDOM_ORVALUE);
532 WRITE_ONCE(*args->pudp, pud);
533 pud_clear(args->pudp);
534 pud = READ_ONCE(*args->pudp);
535 WARN_ON(!pud_none(pud));
536 }
537
pud_populate_tests(struct pgtable_debug_args * args)538 static void __init pud_populate_tests(struct pgtable_debug_args *args)
539 {
540 pud_t pud;
541
542 if (mm_pmd_folded(args->mm))
543 return;
544
545 pr_debug("Validating PUD populate\n");
546 /*
547 * This entry points to next level page table page.
548 * Hence this must not qualify as pud_bad().
549 */
550 pud_populate(args->mm, args->pudp, args->start_pmdp);
551 pud = READ_ONCE(*args->pudp);
552 WARN_ON(pud_bad(pud));
553 }
554 #else /* !__PAGETABLE_PUD_FOLDED */
pud_clear_tests(struct pgtable_debug_args * args)555 static void __init pud_clear_tests(struct pgtable_debug_args *args) { }
pud_populate_tests(struct pgtable_debug_args * args)556 static void __init pud_populate_tests(struct pgtable_debug_args *args) { }
557 #endif /* PAGETABLE_PUD_FOLDED */
558
559 #ifndef __PAGETABLE_P4D_FOLDED
p4d_clear_tests(struct pgtable_debug_args * args)560 static void __init p4d_clear_tests(struct pgtable_debug_args *args)
561 {
562 p4d_t p4d = READ_ONCE(*args->p4dp);
563
564 if (mm_pud_folded(args->mm))
565 return;
566
567 pr_debug("Validating P4D clear\n");
568 p4d = __p4d(p4d_val(p4d) | RANDOM_ORVALUE);
569 WRITE_ONCE(*args->p4dp, p4d);
570 p4d_clear(args->p4dp);
571 p4d = READ_ONCE(*args->p4dp);
572 WARN_ON(!p4d_none(p4d));
573 }
574
p4d_populate_tests(struct pgtable_debug_args * args)575 static void __init p4d_populate_tests(struct pgtable_debug_args *args)
576 {
577 p4d_t p4d;
578
579 if (mm_pud_folded(args->mm))
580 return;
581
582 pr_debug("Validating P4D populate\n");
583 /*
584 * This entry points to next level page table page.
585 * Hence this must not qualify as p4d_bad().
586 */
587 pud_clear(args->pudp);
588 p4d_clear(args->p4dp);
589 p4d_populate(args->mm, args->p4dp, args->start_pudp);
590 p4d = READ_ONCE(*args->p4dp);
591 WARN_ON(p4d_bad(p4d));
592 }
593
pgd_clear_tests(struct pgtable_debug_args * args)594 static void __init pgd_clear_tests(struct pgtable_debug_args *args)
595 {
596 pgd_t pgd = READ_ONCE(*(args->pgdp));
597
598 if (mm_p4d_folded(args->mm))
599 return;
600
601 pr_debug("Validating PGD clear\n");
602 pgd = __pgd(pgd_val(pgd) | RANDOM_ORVALUE);
603 WRITE_ONCE(*args->pgdp, pgd);
604 pgd_clear(args->pgdp);
605 pgd = READ_ONCE(*args->pgdp);
606 WARN_ON(!pgd_none(pgd));
607 }
608
pgd_populate_tests(struct pgtable_debug_args * args)609 static void __init pgd_populate_tests(struct pgtable_debug_args *args)
610 {
611 pgd_t pgd;
612
613 if (mm_p4d_folded(args->mm))
614 return;
615
616 pr_debug("Validating PGD populate\n");
617 /*
618 * This entry points to next level page table page.
619 * Hence this must not qualify as pgd_bad().
620 */
621 p4d_clear(args->p4dp);
622 pgd_clear(args->pgdp);
623 pgd_populate(args->mm, args->pgdp, args->start_p4dp);
624 pgd = READ_ONCE(*args->pgdp);
625 WARN_ON(pgd_bad(pgd));
626 }
627 #else /* !__PAGETABLE_P4D_FOLDED */
p4d_clear_tests(struct pgtable_debug_args * args)628 static void __init p4d_clear_tests(struct pgtable_debug_args *args) { }
pgd_clear_tests(struct pgtable_debug_args * args)629 static void __init pgd_clear_tests(struct pgtable_debug_args *args) { }
p4d_populate_tests(struct pgtable_debug_args * args)630 static void __init p4d_populate_tests(struct pgtable_debug_args *args) { }
pgd_populate_tests(struct pgtable_debug_args * args)631 static void __init pgd_populate_tests(struct pgtable_debug_args *args) { }
632 #endif /* PAGETABLE_P4D_FOLDED */
633
pte_clear_tests(struct pgtable_debug_args * args)634 static void __init pte_clear_tests(struct pgtable_debug_args *args)
635 {
636 struct page *page;
637 pte_t pte = pfn_pte(args->pte_pfn, args->page_prot);
638
639 page = (args->pte_pfn != ULONG_MAX) ? pfn_to_page(args->pte_pfn) : NULL;
640 if (!page)
641 return;
642
643 /*
644 * flush_dcache_page() is called after set_pte_at() to clear
645 * PG_arch_1 for the page on ARM64. The page flag isn't cleared
646 * when it's released and page allocation check will fail when
647 * the page is allocated again. For architectures other than ARM64,
648 * the unexpected overhead of cache flushing is acceptable.
649 */
650 pr_debug("Validating PTE clear\n");
651 #ifndef CONFIG_RISCV
652 pte = __pte(pte_val(pte) | RANDOM_ORVALUE);
653 #endif
654 set_pte_at(args->mm, args->vaddr, args->ptep, pte);
655 flush_dcache_page(page);
656 barrier();
657 pte_clear(args->mm, args->vaddr, args->ptep);
658 pte = ptep_get(args->ptep);
659 WARN_ON(!pte_none(pte));
660 }
661
pmd_clear_tests(struct pgtable_debug_args * args)662 static void __init pmd_clear_tests(struct pgtable_debug_args *args)
663 {
664 pmd_t pmd = READ_ONCE(*args->pmdp);
665
666 pr_debug("Validating PMD clear\n");
667 pmd = __pmd(pmd_val(pmd) | RANDOM_ORVALUE);
668 WRITE_ONCE(*args->pmdp, pmd);
669 pmd_clear(args->pmdp);
670 pmd = READ_ONCE(*args->pmdp);
671 WARN_ON(!pmd_none(pmd));
672 }
673
pmd_populate_tests(struct pgtable_debug_args * args)674 static void __init pmd_populate_tests(struct pgtable_debug_args *args)
675 {
676 pmd_t pmd;
677
678 pr_debug("Validating PMD populate\n");
679 /*
680 * This entry points to next level page table page.
681 * Hence this must not qualify as pmd_bad().
682 */
683 pmd_populate(args->mm, args->pmdp, args->start_ptep);
684 pmd = READ_ONCE(*args->pmdp);
685 WARN_ON(pmd_bad(pmd));
686 }
687
pte_special_tests(struct pgtable_debug_args * args)688 static void __init pte_special_tests(struct pgtable_debug_args *args)
689 {
690 pte_t pte = pfn_pte(args->fixed_pte_pfn, args->page_prot);
691
692 if (!IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL))
693 return;
694
695 pr_debug("Validating PTE special\n");
696 WARN_ON(!pte_special(pte_mkspecial(pte)));
697 }
698
pte_protnone_tests(struct pgtable_debug_args * args)699 static void __init pte_protnone_tests(struct pgtable_debug_args *args)
700 {
701 pte_t pte = pfn_pte(args->fixed_pte_pfn, args->page_prot_none);
702
703 if (!IS_ENABLED(CONFIG_NUMA_BALANCING))
704 return;
705
706 pr_debug("Validating PTE protnone\n");
707 WARN_ON(!pte_protnone(pte));
708 WARN_ON(!pte_present(pte));
709 }
710
711 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
pmd_protnone_tests(struct pgtable_debug_args * args)712 static void __init pmd_protnone_tests(struct pgtable_debug_args *args)
713 {
714 pmd_t pmd;
715
716 if (!IS_ENABLED(CONFIG_NUMA_BALANCING))
717 return;
718
719 if (!has_transparent_hugepage())
720 return;
721
722 pr_debug("Validating PMD protnone\n");
723 pmd = pmd_mkhuge(pfn_pmd(args->fixed_pmd_pfn, args->page_prot_none));
724 WARN_ON(!pmd_protnone(pmd));
725 WARN_ON(!pmd_present(pmd));
726 }
727 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
pmd_protnone_tests(struct pgtable_debug_args * args)728 static void __init pmd_protnone_tests(struct pgtable_debug_args *args) { }
729 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
730
731 #ifdef CONFIG_ARCH_HAS_PTE_DEVMAP
pte_devmap_tests(struct pgtable_debug_args * args)732 static void __init pte_devmap_tests(struct pgtable_debug_args *args)
733 {
734 pte_t pte = pfn_pte(args->fixed_pte_pfn, args->page_prot);
735
736 pr_debug("Validating PTE devmap\n");
737 WARN_ON(!pte_devmap(pte_mkdevmap(pte)));
738 }
739
740 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
pmd_devmap_tests(struct pgtable_debug_args * args)741 static void __init pmd_devmap_tests(struct pgtable_debug_args *args)
742 {
743 pmd_t pmd;
744
745 if (!has_transparent_hugepage())
746 return;
747
748 pr_debug("Validating PMD devmap\n");
749 pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot);
750 WARN_ON(!pmd_devmap(pmd_mkdevmap(pmd)));
751 }
752
753 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
pud_devmap_tests(struct pgtable_debug_args * args)754 static void __init pud_devmap_tests(struct pgtable_debug_args *args)
755 {
756 pud_t pud;
757
758 if (!has_transparent_hugepage())
759 return;
760
761 pr_debug("Validating PUD devmap\n");
762 pud = pfn_pud(args->fixed_pud_pfn, args->page_prot);
763 WARN_ON(!pud_devmap(pud_mkdevmap(pud)));
764 }
765 #else /* !CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
pud_devmap_tests(struct pgtable_debug_args * args)766 static void __init pud_devmap_tests(struct pgtable_debug_args *args) { }
767 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
768 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
pmd_devmap_tests(struct pgtable_debug_args * args)769 static void __init pmd_devmap_tests(struct pgtable_debug_args *args) { }
pud_devmap_tests(struct pgtable_debug_args * args)770 static void __init pud_devmap_tests(struct pgtable_debug_args *args) { }
771 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
772 #else
pte_devmap_tests(struct pgtable_debug_args * args)773 static void __init pte_devmap_tests(struct pgtable_debug_args *args) { }
pmd_devmap_tests(struct pgtable_debug_args * args)774 static void __init pmd_devmap_tests(struct pgtable_debug_args *args) { }
pud_devmap_tests(struct pgtable_debug_args * args)775 static void __init pud_devmap_tests(struct pgtable_debug_args *args) { }
776 #endif /* CONFIG_ARCH_HAS_PTE_DEVMAP */
777
pte_soft_dirty_tests(struct pgtable_debug_args * args)778 static void __init pte_soft_dirty_tests(struct pgtable_debug_args *args)
779 {
780 pte_t pte = pfn_pte(args->fixed_pte_pfn, args->page_prot);
781
782 if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY))
783 return;
784
785 pr_debug("Validating PTE soft dirty\n");
786 WARN_ON(!pte_soft_dirty(pte_mksoft_dirty(pte)));
787 WARN_ON(pte_soft_dirty(pte_clear_soft_dirty(pte)));
788 }
789
pte_swap_soft_dirty_tests(struct pgtable_debug_args * args)790 static void __init pte_swap_soft_dirty_tests(struct pgtable_debug_args *args)
791 {
792 pte_t pte = pfn_pte(args->fixed_pte_pfn, args->page_prot);
793
794 if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY))
795 return;
796
797 pr_debug("Validating PTE swap soft dirty\n");
798 WARN_ON(!pte_swp_soft_dirty(pte_swp_mksoft_dirty(pte)));
799 WARN_ON(pte_swp_soft_dirty(pte_swp_clear_soft_dirty(pte)));
800 }
801
802 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
pmd_soft_dirty_tests(struct pgtable_debug_args * args)803 static void __init pmd_soft_dirty_tests(struct pgtable_debug_args *args)
804 {
805 pmd_t pmd;
806
807 if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY))
808 return;
809
810 if (!has_transparent_hugepage())
811 return;
812
813 pr_debug("Validating PMD soft dirty\n");
814 pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot);
815 WARN_ON(!pmd_soft_dirty(pmd_mksoft_dirty(pmd)));
816 WARN_ON(pmd_soft_dirty(pmd_clear_soft_dirty(pmd)));
817 }
818
pmd_swap_soft_dirty_tests(struct pgtable_debug_args * args)819 static void __init pmd_swap_soft_dirty_tests(struct pgtable_debug_args *args)
820 {
821 pmd_t pmd;
822
823 if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) ||
824 !IS_ENABLED(CONFIG_ARCH_ENABLE_THP_MIGRATION))
825 return;
826
827 if (!has_transparent_hugepage())
828 return;
829
830 pr_debug("Validating PMD swap soft dirty\n");
831 pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot);
832 WARN_ON(!pmd_swp_soft_dirty(pmd_swp_mksoft_dirty(pmd)));
833 WARN_ON(pmd_swp_soft_dirty(pmd_swp_clear_soft_dirty(pmd)));
834 }
835 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
pmd_soft_dirty_tests(struct pgtable_debug_args * args)836 static void __init pmd_soft_dirty_tests(struct pgtable_debug_args *args) { }
pmd_swap_soft_dirty_tests(struct pgtable_debug_args * args)837 static void __init pmd_swap_soft_dirty_tests(struct pgtable_debug_args *args) { }
838 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
839
pte_swap_tests(struct pgtable_debug_args * args)840 static void __init pte_swap_tests(struct pgtable_debug_args *args)
841 {
842 swp_entry_t swp;
843 pte_t pte;
844
845 pr_debug("Validating PTE swap\n");
846 pte = pfn_pte(args->fixed_pte_pfn, args->page_prot);
847 swp = __pte_to_swp_entry(pte);
848 pte = __swp_entry_to_pte(swp);
849 WARN_ON(args->fixed_pte_pfn != pte_pfn(pte));
850 }
851
852 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
pmd_swap_tests(struct pgtable_debug_args * args)853 static void __init pmd_swap_tests(struct pgtable_debug_args *args)
854 {
855 swp_entry_t swp;
856 pmd_t pmd;
857
858 if (!has_transparent_hugepage())
859 return;
860
861 pr_debug("Validating PMD swap\n");
862 pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot);
863 swp = __pmd_to_swp_entry(pmd);
864 pmd = __swp_entry_to_pmd(swp);
865 WARN_ON(args->fixed_pmd_pfn != pmd_pfn(pmd));
866 }
867 #else /* !CONFIG_ARCH_ENABLE_THP_MIGRATION */
pmd_swap_tests(struct pgtable_debug_args * args)868 static void __init pmd_swap_tests(struct pgtable_debug_args *args) { }
869 #endif /* CONFIG_ARCH_ENABLE_THP_MIGRATION */
870
swap_migration_tests(struct pgtable_debug_args * args)871 static void __init swap_migration_tests(struct pgtable_debug_args *args)
872 {
873 struct page *page;
874 swp_entry_t swp;
875
876 if (!IS_ENABLED(CONFIG_MIGRATION))
877 return;
878
879 /*
880 * swap_migration_tests() requires a dedicated page as it needs to
881 * be locked before creating a migration entry from it. Locking the
882 * page that actually maps kernel text ('start_kernel') can be real
883 * problematic. Lets use the allocated page explicitly for this
884 * purpose.
885 */
886 page = (args->pte_pfn != ULONG_MAX) ? pfn_to_page(args->pte_pfn) : NULL;
887 if (!page)
888 return;
889
890 pr_debug("Validating swap migration\n");
891
892 /*
893 * make_migration_entry() expects given page to be
894 * locked, otherwise it stumbles upon a BUG_ON().
895 */
896 __SetPageLocked(page);
897 swp = make_writable_migration_entry(page_to_pfn(page));
898 WARN_ON(!is_migration_entry(swp));
899 WARN_ON(!is_writable_migration_entry(swp));
900
901 swp = make_readable_migration_entry(swp_offset(swp));
902 WARN_ON(!is_migration_entry(swp));
903 WARN_ON(is_writable_migration_entry(swp));
904
905 swp = make_readable_migration_entry(page_to_pfn(page));
906 WARN_ON(!is_migration_entry(swp));
907 WARN_ON(is_writable_migration_entry(swp));
908 __ClearPageLocked(page);
909 }
910
911 #ifdef CONFIG_HUGETLB_PAGE
hugetlb_basic_tests(struct pgtable_debug_args * args)912 static void __init hugetlb_basic_tests(struct pgtable_debug_args *args)
913 {
914 struct page *page;
915 pte_t pte;
916
917 pr_debug("Validating HugeTLB basic\n");
918 /*
919 * Accessing the page associated with the pfn is safe here,
920 * as it was previously derived from a real kernel symbol.
921 */
922 page = pfn_to_page(args->fixed_pmd_pfn);
923 pte = mk_huge_pte(page, args->page_prot);
924
925 WARN_ON(!huge_pte_dirty(huge_pte_mkdirty(pte)));
926 WARN_ON(!huge_pte_write(huge_pte_mkwrite(huge_pte_wrprotect(pte))));
927 WARN_ON(huge_pte_write(huge_pte_wrprotect(huge_pte_mkwrite(pte))));
928
929 #ifdef CONFIG_ARCH_WANT_GENERAL_HUGETLB
930 pte = pfn_pte(args->fixed_pmd_pfn, args->page_prot);
931
932 WARN_ON(!pte_huge(pte_mkhuge(pte)));
933 #endif /* CONFIG_ARCH_WANT_GENERAL_HUGETLB */
934 }
935 #else /* !CONFIG_HUGETLB_PAGE */
hugetlb_basic_tests(struct pgtable_debug_args * args)936 static void __init hugetlb_basic_tests(struct pgtable_debug_args *args) { }
937 #endif /* CONFIG_HUGETLB_PAGE */
938
939 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
pmd_thp_tests(struct pgtable_debug_args * args)940 static void __init pmd_thp_tests(struct pgtable_debug_args *args)
941 {
942 pmd_t pmd;
943
944 if (!has_transparent_hugepage())
945 return;
946
947 pr_debug("Validating PMD based THP\n");
948 /*
949 * pmd_trans_huge() and pmd_present() must return positive after
950 * MMU invalidation with pmd_mkinvalid(). This behavior is an
951 * optimization for transparent huge page. pmd_trans_huge() must
952 * be true if pmd_page() returns a valid THP to avoid taking the
953 * pmd_lock when others walk over non transhuge pmds (i.e. there
954 * are no THP allocated). Especially when splitting a THP and
955 * removing the present bit from the pmd, pmd_trans_huge() still
956 * needs to return true. pmd_present() should be true whenever
957 * pmd_trans_huge() returns true.
958 */
959 pmd = pfn_pmd(args->fixed_pmd_pfn, args->page_prot);
960 WARN_ON(!pmd_trans_huge(pmd_mkhuge(pmd)));
961
962 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
963 WARN_ON(!pmd_trans_huge(pmd_mkinvalid(pmd_mkhuge(pmd))));
964 WARN_ON(!pmd_present(pmd_mkinvalid(pmd_mkhuge(pmd))));
965 #endif /* __HAVE_ARCH_PMDP_INVALIDATE */
966 }
967
968 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
pud_thp_tests(struct pgtable_debug_args * args)969 static void __init pud_thp_tests(struct pgtable_debug_args *args)
970 {
971 pud_t pud;
972
973 if (!has_transparent_hugepage())
974 return;
975
976 pr_debug("Validating PUD based THP\n");
977 pud = pfn_pud(args->fixed_pud_pfn, args->page_prot);
978 WARN_ON(!pud_trans_huge(pud_mkhuge(pud)));
979
980 /*
981 * pud_mkinvalid() has been dropped for now. Enable back
982 * these tests when it comes back with a modified pud_present().
983 *
984 * WARN_ON(!pud_trans_huge(pud_mkinvalid(pud_mkhuge(pud))));
985 * WARN_ON(!pud_present(pud_mkinvalid(pud_mkhuge(pud))));
986 */
987 }
988 #else /* !CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
pud_thp_tests(struct pgtable_debug_args * args)989 static void __init pud_thp_tests(struct pgtable_debug_args *args) { }
990 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
991 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
pmd_thp_tests(struct pgtable_debug_args * args)992 static void __init pmd_thp_tests(struct pgtable_debug_args *args) { }
pud_thp_tests(struct pgtable_debug_args * args)993 static void __init pud_thp_tests(struct pgtable_debug_args *args) { }
994 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
995
get_random_vaddr(void)996 static unsigned long __init get_random_vaddr(void)
997 {
998 unsigned long random_vaddr, random_pages, total_user_pages;
999
1000 total_user_pages = (TASK_SIZE - FIRST_USER_ADDRESS) / PAGE_SIZE;
1001
1002 random_pages = get_random_long() % total_user_pages;
1003 random_vaddr = FIRST_USER_ADDRESS + random_pages * PAGE_SIZE;
1004
1005 return random_vaddr;
1006 }
1007
destroy_args(struct pgtable_debug_args * args)1008 static void __init destroy_args(struct pgtable_debug_args *args)
1009 {
1010 struct page *page = NULL;
1011
1012 /* Free (huge) page */
1013 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
1014 IS_ENABLED(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD) &&
1015 has_transparent_hugepage() &&
1016 args->pud_pfn != ULONG_MAX) {
1017 if (args->is_contiguous_page) {
1018 free_contig_range(args->pud_pfn,
1019 (1 << (HPAGE_PUD_SHIFT - PAGE_SHIFT)));
1020 } else {
1021 page = pfn_to_page(args->pud_pfn);
1022 __free_pages(page, HPAGE_PUD_SHIFT - PAGE_SHIFT);
1023 }
1024
1025 args->pud_pfn = ULONG_MAX;
1026 args->pmd_pfn = ULONG_MAX;
1027 args->pte_pfn = ULONG_MAX;
1028 }
1029
1030 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
1031 has_transparent_hugepage() &&
1032 args->pmd_pfn != ULONG_MAX) {
1033 if (args->is_contiguous_page) {
1034 free_contig_range(args->pmd_pfn, (1 << HPAGE_PMD_ORDER));
1035 } else {
1036 page = pfn_to_page(args->pmd_pfn);
1037 __free_pages(page, HPAGE_PMD_ORDER);
1038 }
1039
1040 args->pmd_pfn = ULONG_MAX;
1041 args->pte_pfn = ULONG_MAX;
1042 }
1043
1044 if (args->pte_pfn != ULONG_MAX) {
1045 page = pfn_to_page(args->pte_pfn);
1046 __free_pages(page, 0);
1047
1048 args->pte_pfn = ULONG_MAX;
1049 }
1050
1051 /* Free page table entries */
1052 if (args->start_ptep) {
1053 pte_free(args->mm, args->start_ptep);
1054 mm_dec_nr_ptes(args->mm);
1055 }
1056
1057 if (args->start_pmdp) {
1058 pmd_free(args->mm, args->start_pmdp);
1059 mm_dec_nr_pmds(args->mm);
1060 }
1061
1062 if (args->start_pudp) {
1063 pud_free(args->mm, args->start_pudp);
1064 mm_dec_nr_puds(args->mm);
1065 }
1066
1067 if (args->start_p4dp)
1068 p4d_free(args->mm, args->start_p4dp);
1069
1070 /* Free vma and mm struct */
1071 if (args->vma)
1072 vm_area_free(args->vma);
1073
1074 if (args->mm)
1075 mmdrop(args->mm);
1076 }
1077
1078 static struct page * __init
debug_vm_pgtable_alloc_huge_page(struct pgtable_debug_args * args,int order)1079 debug_vm_pgtable_alloc_huge_page(struct pgtable_debug_args *args, int order)
1080 {
1081 struct page *page = NULL;
1082
1083 #ifdef CONFIG_CONTIG_ALLOC
1084 if (order >= MAX_ORDER) {
1085 page = alloc_contig_pages((1 << order), GFP_KERNEL,
1086 first_online_node, NULL);
1087 if (page) {
1088 args->is_contiguous_page = true;
1089 return page;
1090 }
1091 }
1092 #endif
1093
1094 if (order < MAX_ORDER)
1095 page = alloc_pages(GFP_KERNEL, order);
1096
1097 return page;
1098 }
1099
init_args(struct pgtable_debug_args * args)1100 static int __init init_args(struct pgtable_debug_args *args)
1101 {
1102 struct page *page = NULL;
1103 phys_addr_t phys;
1104 int ret = 0;
1105
1106 /*
1107 * Initialize the debugging data.
1108 *
1109 * __P000 (or even __S000) will help create page table entries with
1110 * PROT_NONE permission as required for pxx_protnone_tests().
1111 */
1112 memset(args, 0, sizeof(*args));
1113 args->vaddr = get_random_vaddr();
1114 args->page_prot = vm_get_page_prot(VMFLAGS);
1115 args->page_prot_none = __P000;
1116 args->is_contiguous_page = false;
1117 args->pud_pfn = ULONG_MAX;
1118 args->pmd_pfn = ULONG_MAX;
1119 args->pte_pfn = ULONG_MAX;
1120 args->fixed_pgd_pfn = ULONG_MAX;
1121 args->fixed_p4d_pfn = ULONG_MAX;
1122 args->fixed_pud_pfn = ULONG_MAX;
1123 args->fixed_pmd_pfn = ULONG_MAX;
1124 args->fixed_pte_pfn = ULONG_MAX;
1125
1126 /* Allocate mm and vma */
1127 args->mm = mm_alloc();
1128 if (!args->mm) {
1129 pr_err("Failed to allocate mm struct\n");
1130 ret = -ENOMEM;
1131 goto error;
1132 }
1133
1134 args->vma = vm_area_alloc(args->mm);
1135 if (!args->vma) {
1136 pr_err("Failed to allocate vma\n");
1137 ret = -ENOMEM;
1138 goto error;
1139 }
1140
1141 /*
1142 * Allocate page table entries. They will be modified in the tests.
1143 * Lets save the page table entries so that they can be released
1144 * when the tests are completed.
1145 */
1146 args->pgdp = pgd_offset(args->mm, args->vaddr);
1147 args->p4dp = p4d_alloc(args->mm, args->pgdp, args->vaddr);
1148 if (!args->p4dp) {
1149 pr_err("Failed to allocate p4d entries\n");
1150 ret = -ENOMEM;
1151 goto error;
1152 }
1153 args->start_p4dp = p4d_offset(args->pgdp, 0UL);
1154 WARN_ON(!args->start_p4dp);
1155
1156 args->pudp = pud_alloc(args->mm, args->p4dp, args->vaddr);
1157 if (!args->pudp) {
1158 pr_err("Failed to allocate pud entries\n");
1159 ret = -ENOMEM;
1160 goto error;
1161 }
1162 args->start_pudp = pud_offset(args->p4dp, 0UL);
1163 WARN_ON(!args->start_pudp);
1164
1165 args->pmdp = pmd_alloc(args->mm, args->pudp, args->vaddr);
1166 if (!args->pmdp) {
1167 pr_err("Failed to allocate pmd entries\n");
1168 ret = -ENOMEM;
1169 goto error;
1170 }
1171 args->start_pmdp = pmd_offset(args->pudp, 0UL);
1172 WARN_ON(!args->start_pmdp);
1173
1174 if (pte_alloc(args->mm, args->pmdp)) {
1175 pr_err("Failed to allocate pte entries\n");
1176 ret = -ENOMEM;
1177 goto error;
1178 }
1179 args->start_ptep = pmd_pgtable(READ_ONCE(*args->pmdp));
1180 WARN_ON(!args->start_ptep);
1181
1182 /*
1183 * PFN for mapping at PTE level is determined from a standard kernel
1184 * text symbol. But pfns for higher page table levels are derived by
1185 * masking lower bits of this real pfn. These derived pfns might not
1186 * exist on the platform but that does not really matter as pfn_pxx()
1187 * helpers will still create appropriate entries for the test. This
1188 * helps avoid large memory block allocations to be used for mapping
1189 * at higher page table levels in some of the tests.
1190 */
1191 phys = __pa_symbol(&start_kernel);
1192 args->fixed_pgd_pfn = __phys_to_pfn(phys & PGDIR_MASK);
1193 args->fixed_p4d_pfn = __phys_to_pfn(phys & P4D_MASK);
1194 args->fixed_pud_pfn = __phys_to_pfn(phys & PUD_MASK);
1195 args->fixed_pmd_pfn = __phys_to_pfn(phys & PMD_MASK);
1196 args->fixed_pte_pfn = __phys_to_pfn(phys & PAGE_MASK);
1197 WARN_ON(!pfn_valid(args->fixed_pte_pfn));
1198
1199 /*
1200 * Allocate (huge) pages because some of the tests need to access
1201 * the data in the pages. The corresponding tests will be skipped
1202 * if we fail to allocate (huge) pages.
1203 */
1204 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
1205 IS_ENABLED(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD) &&
1206 has_transparent_hugepage()) {
1207 page = debug_vm_pgtable_alloc_huge_page(args,
1208 HPAGE_PUD_SHIFT - PAGE_SHIFT);
1209 if (page) {
1210 args->pud_pfn = page_to_pfn(page);
1211 args->pmd_pfn = args->pud_pfn;
1212 args->pte_pfn = args->pud_pfn;
1213 return 0;
1214 }
1215 }
1216
1217 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
1218 has_transparent_hugepage()) {
1219 page = debug_vm_pgtable_alloc_huge_page(args, HPAGE_PMD_ORDER);
1220 if (page) {
1221 args->pmd_pfn = page_to_pfn(page);
1222 args->pte_pfn = args->pmd_pfn;
1223 return 0;
1224 }
1225 }
1226
1227 page = alloc_pages(GFP_KERNEL, 0);
1228 if (page)
1229 args->pte_pfn = page_to_pfn(page);
1230
1231 return 0;
1232
1233 error:
1234 destroy_args(args);
1235 return ret;
1236 }
1237
debug_vm_pgtable(void)1238 static int __init debug_vm_pgtable(void)
1239 {
1240 struct pgtable_debug_args args;
1241 spinlock_t *ptl = NULL;
1242 int idx, ret;
1243
1244 pr_info("Validating architecture page table helpers\n");
1245 ret = init_args(&args);
1246 if (ret)
1247 return ret;
1248
1249 /*
1250 * Iterate over the protection_map[] to make sure that all
1251 * the basic page table transformation validations just hold
1252 * true irrespective of the starting protection value for a
1253 * given page table entry.
1254 */
1255 for (idx = 0; idx < ARRAY_SIZE(protection_map); idx++) {
1256 pte_basic_tests(&args, idx);
1257 pmd_basic_tests(&args, idx);
1258 pud_basic_tests(&args, idx);
1259 }
1260
1261 /*
1262 * Both P4D and PGD level tests are very basic which do not
1263 * involve creating page table entries from the protection
1264 * value and the given pfn. Hence just keep them out from
1265 * the above iteration for now to save some test execution
1266 * time.
1267 */
1268 p4d_basic_tests(&args);
1269 pgd_basic_tests(&args);
1270
1271 pmd_leaf_tests(&args);
1272 pud_leaf_tests(&args);
1273
1274 pte_savedwrite_tests(&args);
1275 pmd_savedwrite_tests(&args);
1276
1277 pte_special_tests(&args);
1278 pte_protnone_tests(&args);
1279 pmd_protnone_tests(&args);
1280
1281 pte_devmap_tests(&args);
1282 pmd_devmap_tests(&args);
1283 pud_devmap_tests(&args);
1284
1285 pte_soft_dirty_tests(&args);
1286 pmd_soft_dirty_tests(&args);
1287 pte_swap_soft_dirty_tests(&args);
1288 pmd_swap_soft_dirty_tests(&args);
1289
1290 pte_swap_tests(&args);
1291 pmd_swap_tests(&args);
1292
1293 swap_migration_tests(&args);
1294
1295 pmd_thp_tests(&args);
1296 pud_thp_tests(&args);
1297
1298 hugetlb_basic_tests(&args);
1299
1300 /*
1301 * Page table modifying tests. They need to hold
1302 * proper page table lock.
1303 */
1304
1305 args.ptep = pte_offset_map_lock(args.mm, args.pmdp, args.vaddr, &ptl);
1306 pte_clear_tests(&args);
1307 pte_advanced_tests(&args);
1308 pte_unmap_unlock(args.ptep, ptl);
1309
1310 ptl = pmd_lock(args.mm, args.pmdp);
1311 pmd_clear_tests(&args);
1312 pmd_advanced_tests(&args);
1313 pmd_huge_tests(&args);
1314 pmd_populate_tests(&args);
1315 spin_unlock(ptl);
1316
1317 ptl = pud_lock(args.mm, args.pudp);
1318 pud_clear_tests(&args);
1319 pud_advanced_tests(&args);
1320 pud_huge_tests(&args);
1321 pud_populate_tests(&args);
1322 spin_unlock(ptl);
1323
1324 spin_lock(&(args.mm->page_table_lock));
1325 p4d_clear_tests(&args);
1326 pgd_clear_tests(&args);
1327 p4d_populate_tests(&args);
1328 pgd_populate_tests(&args);
1329 spin_unlock(&(args.mm->page_table_lock));
1330
1331 destroy_args(&args);
1332 return 0;
1333 }
1334 late_initcall(debug_vm_pgtable);
1335