Commit | Line | Data |
---|---|---|
8eb07b18 RG |
1 | /* |
2 | * Copyright 2016, Rashmica Gupta, IBM Corp. | |
3 | * | |
4 | * This traverses the kernel pagetables and dumps the | |
5 | * information about the used sections of memory to | |
6 | * /sys/kernel/debug/kernel_pagetables. | |
7 | * | |
8 | * Derived from the arm64 implementation: | |
9 | * Copyright (c) 2014, The Linux Foundation, Laura Abbott. | |
10 | * (C) Copyright 2008 Intel Corporation, Arjan van de Ven. | |
11 | * | |
12 | * This program is free software; you can redistribute it and/or | |
13 | * modify it under the terms of the GNU General Public License | |
14 | * as published by the Free Software Foundation; version 2 | |
15 | * of the License. | |
16 | */ | |
17 | #include <linux/debugfs.h> | |
18 | #include <linux/fs.h> | |
19 | #include <linux/io.h> | |
20 | #include <linux/mm.h> | |
21 | #include <linux/sched.h> | |
22 | #include <linux/seq_file.h> | |
23 | #include <asm/fixmap.h> | |
24 | #include <asm/pgtable.h> | |
25 | #include <linux/const.h> | |
26 | #include <asm/page.h> | |
27 | #include <asm/pgalloc.h> | |
28 | ||
29 | /* | |
30 | * To visualise what is happening, | |
31 | * | |
32 | * - PTRS_PER_P** = how many entries there are in the corresponding P** | |
33 | * - P**_SHIFT = how many bits of the address we use to index into the | |
34 | * corresponding P** | |
35 | * - P**_SIZE is how much memory we can access through the table - not the | |
36 | * size of the table itself. | |
37 | * P**={PGD, PUD, PMD, PTE} | |
38 | * | |
39 | * | |
40 | * Each entry of the PGD points to a PUD. Each entry of a PUD points to a | |
41 | * PMD. Each entry of a PMD points to a PTE. And every PTE entry points to | |
42 | * a page. | |
43 | * | |
44 | * In the case where there are only 3 levels, the PUD is folded into the | |
45 | * PGD: every PUD has only one entry which points to the PMD. | |
46 | * | |
47 | * The page dumper groups page table entries of the same type into a single | |
48 | * description. It uses pg_state to track the range information while | |
49 | * iterating over the PTE entries. When the continuity is broken it then | |
50 | * dumps out a description of the range - ie PTEs that are virtually contiguous | |
51 | * with the same PTE flags are chunked together. This is to make it clear how | |
52 | * different areas of the kernel virtual memory are used. | |
53 | * | |
54 | */ | |
55 | struct pg_state { | |
56 | struct seq_file *seq; | |
57 | const struct addr_marker *marker; | |
58 | unsigned long start_address; | |
59 | unsigned int level; | |
60 | u64 current_flags; | |
61 | }; | |
62 | ||
63 | struct addr_marker { | |
64 | unsigned long start_address; | |
65 | const char *name; | |
66 | }; | |
67 | ||
68 | static struct addr_marker address_markers[] = { | |
69 | { 0, "Start of kernel VM" }, | |
70 | { 0, "vmalloc() Area" }, | |
71 | { 0, "vmalloc() End" }, | |
72 | { 0, "isa I/O start" }, | |
73 | { 0, "isa I/O end" }, | |
74 | { 0, "phb I/O start" }, | |
75 | { 0, "phb I/O end" }, | |
76 | { 0, "I/O remap start" }, | |
77 | { 0, "I/O remap end" }, | |
78 | { 0, "vmemmap start" }, | |
79 | { -1, NULL }, | |
80 | }; | |
81 | ||
82 | struct flag_info { | |
83 | u64 mask; | |
84 | u64 val; | |
85 | const char *set; | |
86 | const char *clear; | |
87 | bool is_val; | |
88 | int shift; | |
89 | }; | |
90 | ||
91 | static const struct flag_info flag_array[] = { | |
92 | { | |
93 | #ifdef CONFIG_PPC_STD_MMU_64 | |
94 | .mask = _PAGE_PRIVILEGED, | |
95 | .val = 0, | |
96 | #else | |
97 | .mask = _PAGE_USER, | |
98 | .val = _PAGE_USER, | |
99 | #endif | |
100 | .set = "user", | |
101 | .clear = " ", | |
102 | }, { | |
103 | .mask = _PAGE_RW, | |
104 | .val = _PAGE_RW, | |
105 | .set = "rw", | |
106 | .clear = "ro", | |
107 | }, { | |
108 | .mask = _PAGE_EXEC, | |
109 | .val = _PAGE_EXEC, | |
110 | .set = " X ", | |
111 | .clear = " ", | |
112 | }, { | |
113 | .mask = _PAGE_PTE, | |
114 | .val = _PAGE_PTE, | |
115 | .set = "pte", | |
116 | .clear = " ", | |
117 | }, { | |
118 | .mask = _PAGE_PRESENT, | |
119 | .val = _PAGE_PRESENT, | |
120 | .set = "present", | |
121 | .clear = " ", | |
122 | }, { | |
123 | #ifdef CONFIG_PPC_STD_MMU_64 | |
124 | .mask = H_PAGE_HASHPTE, | |
125 | .val = H_PAGE_HASHPTE, | |
126 | #else | |
127 | .mask = _PAGE_HASHPTE, | |
128 | .val = _PAGE_HASHPTE, | |
129 | #endif | |
130 | .set = "hpte", | |
131 | .clear = " ", | |
132 | }, { | |
133 | #ifndef CONFIG_PPC_STD_MMU_64 | |
134 | .mask = _PAGE_GUARDED, | |
135 | .val = _PAGE_GUARDED, | |
136 | .set = "guarded", | |
137 | .clear = " ", | |
138 | }, { | |
139 | #endif | |
140 | .mask = _PAGE_DIRTY, | |
141 | .val = _PAGE_DIRTY, | |
142 | .set = "dirty", | |
143 | .clear = " ", | |
144 | }, { | |
145 | .mask = _PAGE_ACCESSED, | |
146 | .val = _PAGE_ACCESSED, | |
147 | .set = "accessed", | |
148 | .clear = " ", | |
149 | }, { | |
150 | #ifndef CONFIG_PPC_STD_MMU_64 | |
151 | .mask = _PAGE_WRITETHRU, | |
152 | .val = _PAGE_WRITETHRU, | |
153 | .set = "write through", | |
154 | .clear = " ", | |
155 | }, { | |
156 | #endif | |
157 | .mask = _PAGE_NO_CACHE, | |
158 | .val = _PAGE_NO_CACHE, | |
159 | .set = "no cache", | |
160 | .clear = " ", | |
161 | }, { | |
162 | .mask = H_PAGE_BUSY, | |
163 | .val = H_PAGE_BUSY, | |
164 | .set = "busy", | |
165 | }, { | |
166 | #ifdef CONFIG_PPC_64K_PAGES | |
167 | .mask = H_PAGE_COMBO, | |
168 | .val = H_PAGE_COMBO, | |
169 | .set = "combo", | |
170 | }, { | |
171 | .mask = H_PAGE_4K_PFN, | |
172 | .val = H_PAGE_4K_PFN, | |
173 | .set = "4K_pfn", | |
174 | }, { | |
175 | #endif | |
176 | .mask = H_PAGE_F_GIX, | |
177 | .val = H_PAGE_F_GIX, | |
178 | .set = "f_gix", | |
179 | .is_val = true, | |
180 | .shift = H_PAGE_F_GIX_SHIFT, | |
181 | }, { | |
182 | .mask = H_PAGE_F_SECOND, | |
183 | .val = H_PAGE_F_SECOND, | |
184 | .set = "f_second", | |
185 | }, { | |
186 | .mask = _PAGE_SPECIAL, | |
187 | .val = _PAGE_SPECIAL, | |
188 | .set = "special", | |
189 | } | |
190 | }; | |
191 | ||
192 | struct pgtable_level { | |
193 | const struct flag_info *flag; | |
194 | size_t num; | |
195 | u64 mask; | |
196 | }; | |
197 | ||
198 | static struct pgtable_level pg_level[] = { | |
199 | { | |
200 | }, { /* pgd */ | |
201 | .flag = flag_array, | |
202 | .num = ARRAY_SIZE(flag_array), | |
203 | }, { /* pud */ | |
204 | .flag = flag_array, | |
205 | .num = ARRAY_SIZE(flag_array), | |
206 | }, { /* pmd */ | |
207 | .flag = flag_array, | |
208 | .num = ARRAY_SIZE(flag_array), | |
209 | }, { /* pte */ | |
210 | .flag = flag_array, | |
211 | .num = ARRAY_SIZE(flag_array), | |
212 | }, | |
213 | }; | |
214 | ||
215 | static void dump_flag_info(struct pg_state *st, const struct flag_info | |
216 | *flag, u64 pte, int num) | |
217 | { | |
218 | unsigned int i; | |
219 | ||
220 | for (i = 0; i < num; i++, flag++) { | |
221 | const char *s = NULL; | |
222 | u64 val; | |
223 | ||
224 | /* flag not defined so don't check it */ | |
225 | if (flag->mask == 0) | |
226 | continue; | |
227 | /* Some 'flags' are actually values */ | |
228 | if (flag->is_val) { | |
229 | val = pte & flag->val; | |
230 | if (flag->shift) | |
231 | val = val >> flag->shift; | |
232 | seq_printf(st->seq, " %s:%llx", flag->set, val); | |
233 | } else { | |
234 | if ((pte & flag->mask) == flag->val) | |
235 | s = flag->set; | |
236 | else | |
237 | s = flag->clear; | |
238 | if (s) | |
239 | seq_printf(st->seq, " %s", s); | |
240 | } | |
241 | st->current_flags &= ~flag->mask; | |
242 | } | |
243 | if (st->current_flags != 0) | |
244 | seq_printf(st->seq, " unknown flags:%llx", st->current_flags); | |
245 | } | |
246 | ||
247 | static void dump_addr(struct pg_state *st, unsigned long addr) | |
248 | { | |
249 | static const char units[] = "KMGTPE"; | |
250 | const char *unit = units; | |
251 | unsigned long delta; | |
252 | ||
253 | seq_printf(st->seq, "0x%016lx-0x%016lx ", st->start_address, addr-1); | |
254 | delta = (addr - st->start_address) >> 10; | |
255 | /* Work out what appropriate unit to use */ | |
256 | while (!(delta & 1023) && unit[1]) { | |
257 | delta >>= 10; | |
258 | unit++; | |
259 | } | |
260 | seq_printf(st->seq, "%9lu%c", delta, *unit); | |
261 | ||
262 | } | |
263 | ||
264 | static void note_page(struct pg_state *st, unsigned long addr, | |
265 | unsigned int level, u64 val) | |
266 | { | |
267 | u64 flag = val & pg_level[level].mask; | |
268 | /* At first no level is set */ | |
269 | if (!st->level) { | |
270 | st->level = level; | |
271 | st->current_flags = flag; | |
272 | st->start_address = addr; | |
273 | seq_printf(st->seq, "---[ %s ]---\n", st->marker->name); | |
274 | /* | |
275 | * Dump the section of virtual memory when: | |
276 | * - the PTE flags from one entry to the next differs. | |
277 | * - we change levels in the tree. | |
278 | * - the address is in a different section of memory and is thus | |
279 | * used for a different purpose, regardless of the flags. | |
280 | */ | |
281 | } else if (flag != st->current_flags || level != st->level || | |
282 | addr >= st->marker[1].start_address) { | |
283 | ||
284 | /* Check the PTE flags */ | |
285 | if (st->current_flags) { | |
286 | dump_addr(st, addr); | |
287 | ||
288 | /* Dump all the flags */ | |
289 | if (pg_level[st->level].flag) | |
290 | dump_flag_info(st, pg_level[st->level].flag, | |
291 | st->current_flags, | |
292 | pg_level[st->level].num); | |
293 | ||
294 | seq_puts(st->seq, "\n"); | |
295 | } | |
296 | ||
297 | /* | |
298 | * Address indicates we have passed the end of the | |
299 | * current section of virtual memory | |
300 | */ | |
301 | while (addr >= st->marker[1].start_address) { | |
302 | st->marker++; | |
303 | seq_printf(st->seq, "---[ %s ]---\n", st->marker->name); | |
304 | } | |
305 | st->start_address = addr; | |
306 | st->current_flags = flag; | |
307 | st->level = level; | |
308 | } | |
309 | } | |
310 | ||
311 | static void walk_pte(struct pg_state *st, pmd_t *pmd, unsigned long start) | |
312 | { | |
313 | pte_t *pte = pte_offset_kernel(pmd, 0); | |
314 | unsigned long addr; | |
315 | unsigned int i; | |
316 | ||
317 | for (i = 0; i < PTRS_PER_PTE; i++, pte++) { | |
318 | addr = start + i * PAGE_SIZE; | |
319 | note_page(st, addr, 4, pte_val(*pte)); | |
320 | ||
321 | } | |
322 | } | |
323 | ||
324 | static void walk_pmd(struct pg_state *st, pud_t *pud, unsigned long start) | |
325 | { | |
326 | pmd_t *pmd = pmd_offset(pud, 0); | |
327 | unsigned long addr; | |
328 | unsigned int i; | |
329 | ||
330 | for (i = 0; i < PTRS_PER_PMD; i++, pmd++) { | |
331 | addr = start + i * PMD_SIZE; | |
332 | if (!pmd_none(*pmd)) | |
333 | /* pmd exists */ | |
334 | walk_pte(st, pmd, addr); | |
335 | else | |
336 | note_page(st, addr, 3, pmd_val(*pmd)); | |
337 | } | |
338 | } | |
339 | ||
340 | static void walk_pud(struct pg_state *st, pgd_t *pgd, unsigned long start) | |
341 | { | |
342 | pud_t *pud = pud_offset(pgd, 0); | |
343 | unsigned long addr; | |
344 | unsigned int i; | |
345 | ||
346 | for (i = 0; i < PTRS_PER_PUD; i++, pud++) { | |
347 | addr = start + i * PUD_SIZE; | |
348 | if (!pud_none(*pud)) | |
349 | /* pud exists */ | |
350 | walk_pmd(st, pud, addr); | |
351 | else | |
352 | note_page(st, addr, 2, pud_val(*pud)); | |
353 | } | |
354 | } | |
355 | ||
356 | static void walk_pagetables(struct pg_state *st) | |
357 | { | |
358 | pgd_t *pgd = pgd_offset_k(0UL); | |
359 | unsigned int i; | |
360 | unsigned long addr; | |
361 | ||
362 | /* | |
363 | * Traverse the linux pagetable structure and dump pages that are in | |
364 | * the hash pagetable. | |
365 | */ | |
366 | for (i = 0; i < PTRS_PER_PGD; i++, pgd++) { | |
367 | addr = KERN_VIRT_START + i * PGDIR_SIZE; | |
368 | if (!pgd_none(*pgd)) | |
369 | /* pgd exists */ | |
370 | walk_pud(st, pgd, addr); | |
371 | else | |
372 | note_page(st, addr, 1, pgd_val(*pgd)); | |
373 | } | |
374 | } | |
375 | ||
376 | static void populate_markers(void) | |
377 | { | |
378 | address_markers[0].start_address = PAGE_OFFSET; | |
379 | address_markers[1].start_address = VMALLOC_START; | |
380 | address_markers[2].start_address = VMALLOC_END; | |
381 | address_markers[3].start_address = ISA_IO_BASE; | |
382 | address_markers[4].start_address = ISA_IO_END; | |
383 | address_markers[5].start_address = PHB_IO_BASE; | |
384 | address_markers[6].start_address = PHB_IO_END; | |
385 | address_markers[7].start_address = IOREMAP_BASE; | |
386 | address_markers[8].start_address = IOREMAP_END; | |
387 | #ifdef CONFIG_PPC_STD_MMU_64 | |
388 | address_markers[9].start_address = H_VMEMMAP_BASE; | |
389 | #else | |
390 | address_markers[9].start_address = VMEMMAP_BASE; | |
391 | #endif | |
392 | } | |
393 | ||
394 | static int ptdump_show(struct seq_file *m, void *v) | |
395 | { | |
396 | struct pg_state st = { | |
397 | .seq = m, | |
398 | .start_address = KERN_VIRT_START, | |
399 | .marker = address_markers, | |
400 | }; | |
401 | /* Traverse kernel page tables */ | |
402 | walk_pagetables(&st); | |
403 | note_page(&st, 0, 0, 0); | |
404 | return 0; | |
405 | } | |
406 | ||
407 | ||
408 | static int ptdump_open(struct inode *inode, struct file *file) | |
409 | { | |
410 | return single_open(file, ptdump_show, NULL); | |
411 | } | |
412 | ||
413 | static const struct file_operations ptdump_fops = { | |
414 | .open = ptdump_open, | |
415 | .read = seq_read, | |
416 | .llseek = seq_lseek, | |
417 | .release = single_release, | |
418 | }; | |
419 | ||
420 | static void build_pgtable_complete_mask(void) | |
421 | { | |
422 | unsigned int i, j; | |
423 | ||
424 | for (i = 0; i < ARRAY_SIZE(pg_level); i++) | |
425 | if (pg_level[i].flag) | |
426 | for (j = 0; j < pg_level[i].num; j++) | |
427 | pg_level[i].mask |= pg_level[i].flag[j].mask; | |
428 | } | |
429 | ||
430 | static int ptdump_init(void) | |
431 | { | |
432 | struct dentry *debugfs_file; | |
433 | ||
434 | populate_markers(); | |
435 | build_pgtable_complete_mask(); | |
436 | debugfs_file = debugfs_create_file("kernel_pagetables", 0400, NULL, | |
437 | NULL, &ptdump_fops); | |
438 | return debugfs_file ? 0 : -ENOMEM; | |
439 | } | |
440 | device_initcall(ptdump_init); |