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