]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
48c96a36 JK |
2 | #include <linux/debugfs.h> |
3 | #include <linux/mm.h> | |
4 | #include <linux/slab.h> | |
5 | #include <linux/uaccess.h> | |
57c8a661 | 6 | #include <linux/memblock.h> |
48c96a36 JK |
7 | #include <linux/stacktrace.h> |
8 | #include <linux/page_owner.h> | |
7dd80b8a | 9 | #include <linux/jump_label.h> |
7cd12b4a | 10 | #include <linux/migrate.h> |
f2ca0b55 | 11 | #include <linux/stackdepot.h> |
e2f612e6 | 12 | #include <linux/seq_file.h> |
9cc7e96a | 13 | #include <linux/sched/clock.h> |
f2ca0b55 | 14 | |
48c96a36 JK |
15 | #include "internal.h" |
16 | ||
f2ca0b55 JK |
17 | /* |
18 | * TODO: teach PAGE_OWNER_STACK_DEPTH (__dump_page_owner and save_stack) | |
19 | * to use off stack temporal storage | |
20 | */ | |
21 | #define PAGE_OWNER_STACK_DEPTH (16) | |
22 | ||
9300d8df | 23 | struct page_owner { |
6b4c54e3 AM |
24 | unsigned short order; |
25 | short last_migrate_reason; | |
9300d8df | 26 | gfp_t gfp_mask; |
9300d8df | 27 | depot_stack_handle_t handle; |
8974558f | 28 | depot_stack_handle_t free_handle; |
9cc7e96a LM |
29 | u64 ts_nsec; |
30 | pid_t pid; | |
9300d8df JK |
31 | }; |
32 | ||
0fe9a448 | 33 | static bool page_owner_enabled = false; |
7dd80b8a | 34 | DEFINE_STATIC_KEY_FALSE(page_owner_inited); |
48c96a36 | 35 | |
f2ca0b55 JK |
36 | static depot_stack_handle_t dummy_handle; |
37 | static depot_stack_handle_t failure_handle; | |
dab4ead1 | 38 | static depot_stack_handle_t early_handle; |
f2ca0b55 | 39 | |
61cf5feb JK |
40 | static void init_early_allocated_pages(void); |
41 | ||
1173194e | 42 | static int __init early_page_owner_param(char *buf) |
48c96a36 JK |
43 | { |
44 | if (!buf) | |
45 | return -EINVAL; | |
46 | ||
47 | if (strcmp(buf, "on") == 0) | |
0fe9a448 | 48 | page_owner_enabled = true; |
48c96a36 JK |
49 | |
50 | return 0; | |
51 | } | |
52 | early_param("page_owner", early_page_owner_param); | |
53 | ||
54 | static bool need_page_owner(void) | |
55 | { | |
0fe9a448 | 56 | return page_owner_enabled; |
48c96a36 JK |
57 | } |
58 | ||
dab4ead1 | 59 | static __always_inline depot_stack_handle_t create_dummy_stack(void) |
f2ca0b55 JK |
60 | { |
61 | unsigned long entries[4]; | |
af52bf6b | 62 | unsigned int nr_entries; |
f2ca0b55 | 63 | |
af52bf6b TG |
64 | nr_entries = stack_trace_save(entries, ARRAY_SIZE(entries), 0); |
65 | return stack_depot_save(entries, nr_entries, GFP_KERNEL); | |
f2ca0b55 JK |
66 | } |
67 | ||
dab4ead1 | 68 | static noinline void register_dummy_stack(void) |
f2ca0b55 | 69 | { |
dab4ead1 VB |
70 | dummy_handle = create_dummy_stack(); |
71 | } | |
f2ca0b55 | 72 | |
dab4ead1 VB |
73 | static noinline void register_failure_stack(void) |
74 | { | |
75 | failure_handle = create_dummy_stack(); | |
76 | } | |
f2ca0b55 | 77 | |
dab4ead1 VB |
78 | static noinline void register_early_stack(void) |
79 | { | |
80 | early_handle = create_dummy_stack(); | |
f2ca0b55 JK |
81 | } |
82 | ||
48c96a36 JK |
83 | static void init_page_owner(void) |
84 | { | |
0fe9a448 | 85 | if (!page_owner_enabled) |
48c96a36 JK |
86 | return; |
87 | ||
f2ca0b55 JK |
88 | register_dummy_stack(); |
89 | register_failure_stack(); | |
dab4ead1 | 90 | register_early_stack(); |
7dd80b8a | 91 | static_branch_enable(&page_owner_inited); |
61cf5feb | 92 | init_early_allocated_pages(); |
48c96a36 JK |
93 | } |
94 | ||
95 | struct page_ext_operations page_owner_ops = { | |
9300d8df | 96 | .size = sizeof(struct page_owner), |
48c96a36 JK |
97 | .need = need_page_owner, |
98 | .init = init_page_owner, | |
99 | }; | |
100 | ||
9300d8df JK |
101 | static inline struct page_owner *get_page_owner(struct page_ext *page_ext) |
102 | { | |
103 | return (void *)page_ext + page_owner_ops.offset; | |
104 | } | |
105 | ||
af52bf6b TG |
106 | static inline bool check_recursive_alloc(unsigned long *entries, |
107 | unsigned int nr_entries, | |
108 | unsigned long ip) | |
48c96a36 | 109 | { |
af52bf6b | 110 | unsigned int i; |
f2ca0b55 | 111 | |
af52bf6b TG |
112 | for (i = 0; i < nr_entries; i++) { |
113 | if (entries[i] == ip) | |
f2ca0b55 JK |
114 | return true; |
115 | } | |
f2ca0b55 JK |
116 | return false; |
117 | } | |
118 | ||
119 | static noinline depot_stack_handle_t save_stack(gfp_t flags) | |
120 | { | |
121 | unsigned long entries[PAGE_OWNER_STACK_DEPTH]; | |
f2ca0b55 | 122 | depot_stack_handle_t handle; |
af52bf6b | 123 | unsigned int nr_entries; |
f2ca0b55 | 124 | |
af52bf6b | 125 | nr_entries = stack_trace_save(entries, ARRAY_SIZE(entries), 2); |
f2ca0b55 JK |
126 | |
127 | /* | |
af52bf6b TG |
128 | * We need to check recursion here because our request to |
129 | * stackdepot could trigger memory allocation to save new | |
130 | * entry. New memory allocation would reach here and call | |
131 | * stack_depot_save_entries() again if we don't catch it. There is | |
132 | * still not enough memory in stackdepot so it would try to | |
133 | * allocate memory again and loop forever. | |
f2ca0b55 | 134 | */ |
af52bf6b | 135 | if (check_recursive_alloc(entries, nr_entries, _RET_IP_)) |
f2ca0b55 JK |
136 | return dummy_handle; |
137 | ||
af52bf6b | 138 | handle = stack_depot_save(entries, nr_entries, flags); |
f2ca0b55 JK |
139 | if (!handle) |
140 | handle = failure_handle; | |
141 | ||
142 | return handle; | |
143 | } | |
144 | ||
8974558f VB |
145 | void __reset_page_owner(struct page *page, unsigned int order) |
146 | { | |
147 | int i; | |
148 | struct page_ext *page_ext; | |
8974558f VB |
149 | depot_stack_handle_t handle = 0; |
150 | struct page_owner *page_owner; | |
151 | ||
0fe9a448 | 152 | handle = save_stack(GFP_NOWAIT | __GFP_NOWARN); |
8974558f | 153 | |
5556cfe8 VB |
154 | page_ext = lookup_page_ext(page); |
155 | if (unlikely(!page_ext)) | |
156 | return; | |
8974558f | 157 | for (i = 0; i < (1 << order); i++) { |
fdf3bf80 | 158 | __clear_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags); |
0fe9a448 VB |
159 | page_owner = get_page_owner(page_ext); |
160 | page_owner->free_handle = handle; | |
5556cfe8 | 161 | page_ext = page_ext_next(page_ext); |
8974558f VB |
162 | } |
163 | } | |
164 | ||
7e2f2a0c VB |
165 | static inline void __set_page_owner_handle(struct page *page, |
166 | struct page_ext *page_ext, depot_stack_handle_t handle, | |
167 | unsigned int order, gfp_t gfp_mask) | |
f2ca0b55 | 168 | { |
9300d8df | 169 | struct page_owner *page_owner; |
7e2f2a0c | 170 | int i; |
48c96a36 | 171 | |
7e2f2a0c VB |
172 | for (i = 0; i < (1 << order); i++) { |
173 | page_owner = get_page_owner(page_ext); | |
174 | page_owner->handle = handle; | |
175 | page_owner->order = order; | |
176 | page_owner->gfp_mask = gfp_mask; | |
177 | page_owner->last_migrate_reason = -1; | |
9cc7e96a LM |
178 | page_owner->pid = current->pid; |
179 | page_owner->ts_nsec = local_clock(); | |
7e2f2a0c | 180 | __set_bit(PAGE_EXT_OWNER, &page_ext->flags); |
fdf3bf80 | 181 | __set_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags); |
48c96a36 | 182 | |
5556cfe8 | 183 | page_ext = page_ext_next(page_ext); |
7e2f2a0c | 184 | } |
48c96a36 JK |
185 | } |
186 | ||
dab4ead1 VB |
187 | noinline void __set_page_owner(struct page *page, unsigned int order, |
188 | gfp_t gfp_mask) | |
189 | { | |
190 | struct page_ext *page_ext = lookup_page_ext(page); | |
191 | depot_stack_handle_t handle; | |
192 | ||
193 | if (unlikely(!page_ext)) | |
194 | return; | |
195 | ||
196 | handle = save_stack(gfp_mask); | |
7e2f2a0c | 197 | __set_page_owner_handle(page, page_ext, handle, order, gfp_mask); |
dab4ead1 VB |
198 | } |
199 | ||
7cd12b4a VB |
200 | void __set_page_owner_migrate_reason(struct page *page, int reason) |
201 | { | |
202 | struct page_ext *page_ext = lookup_page_ext(page); | |
9300d8df JK |
203 | struct page_owner *page_owner; |
204 | ||
f86e4271 YS |
205 | if (unlikely(!page_ext)) |
206 | return; | |
7cd12b4a | 207 | |
9300d8df JK |
208 | page_owner = get_page_owner(page_ext); |
209 | page_owner->last_migrate_reason = reason; | |
7cd12b4a VB |
210 | } |
211 | ||
8fb156c9 | 212 | void __split_page_owner(struct page *page, unsigned int nr) |
e2cfc911 | 213 | { |
a9627bc5 | 214 | int i; |
e2cfc911 | 215 | struct page_ext *page_ext = lookup_page_ext(page); |
9300d8df | 216 | struct page_owner *page_owner; |
a9627bc5 | 217 | |
f86e4271 | 218 | if (unlikely(!page_ext)) |
a9627bc5 | 219 | return; |
e2cfc911 | 220 | |
8fb156c9 | 221 | for (i = 0; i < nr; i++) { |
7e2f2a0c VB |
222 | page_owner = get_page_owner(page_ext); |
223 | page_owner->order = 0; | |
5556cfe8 | 224 | page_ext = page_ext_next(page_ext); |
7e2f2a0c | 225 | } |
e2cfc911 JK |
226 | } |
227 | ||
d435edca VB |
228 | void __copy_page_owner(struct page *oldpage, struct page *newpage) |
229 | { | |
230 | struct page_ext *old_ext = lookup_page_ext(oldpage); | |
231 | struct page_ext *new_ext = lookup_page_ext(newpage); | |
9300d8df | 232 | struct page_owner *old_page_owner, *new_page_owner; |
d435edca | 233 | |
f86e4271 YS |
234 | if (unlikely(!old_ext || !new_ext)) |
235 | return; | |
236 | ||
9300d8df JK |
237 | old_page_owner = get_page_owner(old_ext); |
238 | new_page_owner = get_page_owner(new_ext); | |
239 | new_page_owner->order = old_page_owner->order; | |
240 | new_page_owner->gfp_mask = old_page_owner->gfp_mask; | |
241 | new_page_owner->last_migrate_reason = | |
242 | old_page_owner->last_migrate_reason; | |
243 | new_page_owner->handle = old_page_owner->handle; | |
9cc7e96a LM |
244 | new_page_owner->pid = old_page_owner->pid; |
245 | new_page_owner->ts_nsec = old_page_owner->ts_nsec; | |
d435edca VB |
246 | |
247 | /* | |
248 | * We don't clear the bit on the oldpage as it's going to be freed | |
249 | * after migration. Until then, the info can be useful in case of | |
250 | * a bug, and the overal stats will be off a bit only temporarily. | |
251 | * Also, migrate_misplaced_transhuge_page() can still fail the | |
252 | * migration and then we want the oldpage to retain the info. But | |
253 | * in that case we also don't need to explicitly clear the info from | |
254 | * the new page, which will be freed. | |
255 | */ | |
256 | __set_bit(PAGE_EXT_OWNER, &new_ext->flags); | |
fdf3bf80 | 257 | __set_bit(PAGE_EXT_OWNER_ALLOCATED, &new_ext->flags); |
d435edca VB |
258 | } |
259 | ||
e2f612e6 JK |
260 | void pagetypeinfo_showmixedcount_print(struct seq_file *m, |
261 | pg_data_t *pgdat, struct zone *zone) | |
262 | { | |
263 | struct page *page; | |
264 | struct page_ext *page_ext; | |
9300d8df | 265 | struct page_owner *page_owner; |
1d2cae8e ML |
266 | unsigned long pfn, block_end_pfn; |
267 | unsigned long end_pfn = zone_end_pfn(zone); | |
e2f612e6 JK |
268 | unsigned long count[MIGRATE_TYPES] = { 0, }; |
269 | int pageblock_mt, page_mt; | |
270 | int i; | |
271 | ||
272 | /* Scan block by block. First and last block may be incomplete */ | |
273 | pfn = zone->zone_start_pfn; | |
274 | ||
275 | /* | |
276 | * Walk the zone in pageblock_nr_pages steps. If a page block spans | |
277 | * a zone boundary, it will be double counted between zones. This does | |
278 | * not matter as the mixed block count will still be correct | |
279 | */ | |
280 | for (; pfn < end_pfn; ) { | |
a26ee565 QC |
281 | page = pfn_to_online_page(pfn); |
282 | if (!page) { | |
e2f612e6 JK |
283 | pfn = ALIGN(pfn + 1, MAX_ORDER_NR_PAGES); |
284 | continue; | |
285 | } | |
286 | ||
287 | block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); | |
288 | block_end_pfn = min(block_end_pfn, end_pfn); | |
289 | ||
e2f612e6 JK |
290 | pageblock_mt = get_pageblock_migratetype(page); |
291 | ||
292 | for (; pfn < block_end_pfn; pfn++) { | |
293 | if (!pfn_valid_within(pfn)) | |
294 | continue; | |
295 | ||
a26ee565 | 296 | /* The pageblock is online, no need to recheck. */ |
e2f612e6 JK |
297 | page = pfn_to_page(pfn); |
298 | ||
299 | if (page_zone(page) != zone) | |
300 | continue; | |
301 | ||
302 | if (PageBuddy(page)) { | |
727c080f VM |
303 | unsigned long freepage_order; |
304 | ||
ab130f91 | 305 | freepage_order = buddy_order_unsafe(page); |
727c080f VM |
306 | if (freepage_order < MAX_ORDER) |
307 | pfn += (1UL << freepage_order) - 1; | |
e2f612e6 JK |
308 | continue; |
309 | } | |
310 | ||
311 | if (PageReserved(page)) | |
312 | continue; | |
313 | ||
314 | page_ext = lookup_page_ext(page); | |
315 | if (unlikely(!page_ext)) | |
316 | continue; | |
317 | ||
fdf3bf80 | 318 | if (!test_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags)) |
e2f612e6 JK |
319 | continue; |
320 | ||
9300d8df | 321 | page_owner = get_page_owner(page_ext); |
01c0bfe0 | 322 | page_mt = gfp_migratetype(page_owner->gfp_mask); |
e2f612e6 JK |
323 | if (pageblock_mt != page_mt) { |
324 | if (is_migrate_cma(pageblock_mt)) | |
325 | count[MIGRATE_MOVABLE]++; | |
326 | else | |
327 | count[pageblock_mt]++; | |
328 | ||
329 | pfn = block_end_pfn; | |
330 | break; | |
331 | } | |
9300d8df | 332 | pfn += (1UL << page_owner->order) - 1; |
e2f612e6 JK |
333 | } |
334 | } | |
335 | ||
336 | /* Print counts */ | |
337 | seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name); | |
338 | for (i = 0; i < MIGRATE_TYPES; i++) | |
339 | seq_printf(m, "%12lu ", count[i]); | |
340 | seq_putc(m, '\n'); | |
341 | } | |
342 | ||
48c96a36 JK |
343 | static ssize_t |
344 | print_page_owner(char __user *buf, size_t count, unsigned long pfn, | |
9300d8df | 345 | struct page *page, struct page_owner *page_owner, |
f2ca0b55 | 346 | depot_stack_handle_t handle) |
48c96a36 | 347 | { |
af52bf6b TG |
348 | int ret, pageblock_mt, page_mt; |
349 | unsigned long *entries; | |
350 | unsigned int nr_entries; | |
48c96a36 JK |
351 | char *kbuf; |
352 | ||
c8f61cfc | 353 | count = min_t(size_t, count, PAGE_SIZE); |
48c96a36 JK |
354 | kbuf = kmalloc(count, GFP_KERNEL); |
355 | if (!kbuf) | |
356 | return -ENOMEM; | |
357 | ||
358 | ret = snprintf(kbuf, count, | |
9cc7e96a | 359 | "Page allocated via order %u, mask %#x(%pGg), pid %d, ts %llu ns\n", |
9300d8df | 360 | page_owner->order, page_owner->gfp_mask, |
9cc7e96a LM |
361 | &page_owner->gfp_mask, page_owner->pid, |
362 | page_owner->ts_nsec); | |
48c96a36 JK |
363 | |
364 | if (ret >= count) | |
365 | goto err; | |
366 | ||
367 | /* Print information relevant to grouping pages by mobility */ | |
0b423ca2 | 368 | pageblock_mt = get_pageblock_migratetype(page); |
01c0bfe0 | 369 | page_mt = gfp_migratetype(page_owner->gfp_mask); |
48c96a36 | 370 | ret += snprintf(kbuf + ret, count - ret, |
60f30350 | 371 | "PFN %lu type %s Block %lu type %s Flags %#lx(%pGp)\n", |
48c96a36 | 372 | pfn, |
60f30350 | 373 | migratetype_names[page_mt], |
48c96a36 | 374 | pfn >> pageblock_order, |
60f30350 VB |
375 | migratetype_names[pageblock_mt], |
376 | page->flags, &page->flags); | |
48c96a36 JK |
377 | |
378 | if (ret >= count) | |
379 | goto err; | |
380 | ||
af52bf6b TG |
381 | nr_entries = stack_depot_fetch(handle, &entries); |
382 | ret += stack_trace_snprint(kbuf + ret, count - ret, entries, nr_entries, 0); | |
48c96a36 JK |
383 | if (ret >= count) |
384 | goto err; | |
385 | ||
9300d8df | 386 | if (page_owner->last_migrate_reason != -1) { |
7cd12b4a VB |
387 | ret += snprintf(kbuf + ret, count - ret, |
388 | "Page has been migrated, last migrate reason: %s\n", | |
9300d8df | 389 | migrate_reason_names[page_owner->last_migrate_reason]); |
7cd12b4a VB |
390 | if (ret >= count) |
391 | goto err; | |
392 | } | |
393 | ||
48c96a36 JK |
394 | ret += snprintf(kbuf + ret, count - ret, "\n"); |
395 | if (ret >= count) | |
396 | goto err; | |
397 | ||
398 | if (copy_to_user(buf, kbuf, ret)) | |
399 | ret = -EFAULT; | |
400 | ||
401 | kfree(kbuf); | |
402 | return ret; | |
403 | ||
404 | err: | |
405 | kfree(kbuf); | |
406 | return -ENOMEM; | |
407 | } | |
408 | ||
4e462112 VB |
409 | void __dump_page_owner(struct page *page) |
410 | { | |
411 | struct page_ext *page_ext = lookup_page_ext(page); | |
9300d8df | 412 | struct page_owner *page_owner; |
f2ca0b55 | 413 | depot_stack_handle_t handle; |
af52bf6b TG |
414 | unsigned long *entries; |
415 | unsigned int nr_entries; | |
8285027f SM |
416 | gfp_t gfp_mask; |
417 | int mt; | |
4e462112 | 418 | |
f86e4271 YS |
419 | if (unlikely(!page_ext)) { |
420 | pr_alert("There is not page extension available.\n"); | |
421 | return; | |
422 | } | |
9300d8df JK |
423 | |
424 | page_owner = get_page_owner(page_ext); | |
425 | gfp_mask = page_owner->gfp_mask; | |
01c0bfe0 | 426 | mt = gfp_migratetype(gfp_mask); |
f86e4271 | 427 | |
4e462112 | 428 | if (!test_bit(PAGE_EXT_OWNER, &page_ext->flags)) { |
37389167 | 429 | pr_alert("page_owner info is not present (never set?)\n"); |
4e462112 VB |
430 | return; |
431 | } | |
432 | ||
fdf3bf80 | 433 | if (test_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags)) |
37389167 VB |
434 | pr_alert("page_owner tracks the page as allocated\n"); |
435 | else | |
436 | pr_alert("page_owner tracks the page as freed\n"); | |
437 | ||
9cc7e96a LM |
438 | pr_alert("page last allocated via order %u, migratetype %s, gfp_mask %#x(%pGg), pid %d, ts %llu\n", |
439 | page_owner->order, migratetype_names[mt], gfp_mask, &gfp_mask, | |
440 | page_owner->pid, page_owner->ts_nsec); | |
37389167 | 441 | |
9300d8df | 442 | handle = READ_ONCE(page_owner->handle); |
f2ca0b55 | 443 | if (!handle) { |
37389167 VB |
444 | pr_alert("page_owner allocation stack trace missing\n"); |
445 | } else { | |
446 | nr_entries = stack_depot_fetch(handle, &entries); | |
447 | stack_trace_print(entries, nr_entries, 0); | |
f2ca0b55 JK |
448 | } |
449 | ||
8974558f VB |
450 | handle = READ_ONCE(page_owner->free_handle); |
451 | if (!handle) { | |
452 | pr_alert("page_owner free stack trace missing\n"); | |
453 | } else { | |
454 | nr_entries = stack_depot_fetch(handle, &entries); | |
455 | pr_alert("page last free stack trace:\n"); | |
456 | stack_trace_print(entries, nr_entries, 0); | |
457 | } | |
8974558f | 458 | |
9300d8df | 459 | if (page_owner->last_migrate_reason != -1) |
4e462112 | 460 | pr_alert("page has been migrated, last migrate reason: %s\n", |
9300d8df | 461 | migrate_reason_names[page_owner->last_migrate_reason]); |
4e462112 VB |
462 | } |
463 | ||
48c96a36 JK |
464 | static ssize_t |
465 | read_page_owner(struct file *file, char __user *buf, size_t count, loff_t *ppos) | |
466 | { | |
467 | unsigned long pfn; | |
468 | struct page *page; | |
469 | struct page_ext *page_ext; | |
9300d8df | 470 | struct page_owner *page_owner; |
f2ca0b55 | 471 | depot_stack_handle_t handle; |
48c96a36 | 472 | |
7dd80b8a | 473 | if (!static_branch_unlikely(&page_owner_inited)) |
48c96a36 JK |
474 | return -EINVAL; |
475 | ||
476 | page = NULL; | |
477 | pfn = min_low_pfn + *ppos; | |
478 | ||
479 | /* Find a valid PFN or the start of a MAX_ORDER_NR_PAGES area */ | |
480 | while (!pfn_valid(pfn) && (pfn & (MAX_ORDER_NR_PAGES - 1)) != 0) | |
481 | pfn++; | |
482 | ||
483 | drain_all_pages(NULL); | |
484 | ||
485 | /* Find an allocated page */ | |
486 | for (; pfn < max_pfn; pfn++) { | |
487 | /* | |
488 | * If the new page is in a new MAX_ORDER_NR_PAGES area, | |
489 | * validate the area as existing, skip it if not | |
490 | */ | |
491 | if ((pfn & (MAX_ORDER_NR_PAGES - 1)) == 0 && !pfn_valid(pfn)) { | |
492 | pfn += MAX_ORDER_NR_PAGES - 1; | |
493 | continue; | |
494 | } | |
495 | ||
496 | /* Check for holes within a MAX_ORDER area */ | |
497 | if (!pfn_valid_within(pfn)) | |
498 | continue; | |
499 | ||
500 | page = pfn_to_page(pfn); | |
501 | if (PageBuddy(page)) { | |
ab130f91 | 502 | unsigned long freepage_order = buddy_order_unsafe(page); |
48c96a36 JK |
503 | |
504 | if (freepage_order < MAX_ORDER) | |
505 | pfn += (1UL << freepage_order) - 1; | |
506 | continue; | |
507 | } | |
508 | ||
509 | page_ext = lookup_page_ext(page); | |
f86e4271 YS |
510 | if (unlikely(!page_ext)) |
511 | continue; | |
48c96a36 JK |
512 | |
513 | /* | |
61cf5feb JK |
514 | * Some pages could be missed by concurrent allocation or free, |
515 | * because we don't hold the zone lock. | |
48c96a36 JK |
516 | */ |
517 | if (!test_bit(PAGE_EXT_OWNER, &page_ext->flags)) | |
518 | continue; | |
519 | ||
37389167 VB |
520 | /* |
521 | * Although we do have the info about past allocation of free | |
522 | * pages, it's not relevant for current memory usage. | |
523 | */ | |
fdf3bf80 | 524 | if (!test_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags)) |
37389167 VB |
525 | continue; |
526 | ||
9300d8df JK |
527 | page_owner = get_page_owner(page_ext); |
528 | ||
7e2f2a0c VB |
529 | /* |
530 | * Don't print "tail" pages of high-order allocations as that | |
531 | * would inflate the stats. | |
532 | */ | |
533 | if (!IS_ALIGNED(pfn, 1 << page_owner->order)) | |
534 | continue; | |
535 | ||
f2ca0b55 JK |
536 | /* |
537 | * Access to page_ext->handle isn't synchronous so we should | |
538 | * be careful to access it. | |
539 | */ | |
9300d8df | 540 | handle = READ_ONCE(page_owner->handle); |
f2ca0b55 JK |
541 | if (!handle) |
542 | continue; | |
543 | ||
48c96a36 JK |
544 | /* Record the next PFN to read in the file offset */ |
545 | *ppos = (pfn - min_low_pfn) + 1; | |
546 | ||
f2ca0b55 | 547 | return print_page_owner(buf, count, pfn, page, |
9300d8df | 548 | page_owner, handle); |
48c96a36 JK |
549 | } |
550 | ||
551 | return 0; | |
552 | } | |
553 | ||
61cf5feb JK |
554 | static void init_pages_in_zone(pg_data_t *pgdat, struct zone *zone) |
555 | { | |
6787c1da OS |
556 | unsigned long pfn = zone->zone_start_pfn; |
557 | unsigned long end_pfn = zone_end_pfn(zone); | |
61cf5feb JK |
558 | unsigned long count = 0; |
559 | ||
61cf5feb JK |
560 | /* |
561 | * Walk the zone in pageblock_nr_pages steps. If a page block spans | |
562 | * a zone boundary, it will be double counted between zones. This does | |
563 | * not matter as the mixed block count will still be correct | |
564 | */ | |
565 | for (; pfn < end_pfn; ) { | |
6787c1da OS |
566 | unsigned long block_end_pfn; |
567 | ||
61cf5feb JK |
568 | if (!pfn_valid(pfn)) { |
569 | pfn = ALIGN(pfn + 1, MAX_ORDER_NR_PAGES); | |
570 | continue; | |
571 | } | |
572 | ||
573 | block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); | |
574 | block_end_pfn = min(block_end_pfn, end_pfn); | |
575 | ||
61cf5feb | 576 | for (; pfn < block_end_pfn; pfn++) { |
6787c1da OS |
577 | struct page *page; |
578 | struct page_ext *page_ext; | |
579 | ||
61cf5feb JK |
580 | if (!pfn_valid_within(pfn)) |
581 | continue; | |
582 | ||
583 | page = pfn_to_page(pfn); | |
584 | ||
9d43f5ae JK |
585 | if (page_zone(page) != zone) |
586 | continue; | |
587 | ||
61cf5feb | 588 | /* |
10903027 VB |
589 | * To avoid having to grab zone->lock, be a little |
590 | * careful when reading buddy page order. The only | |
591 | * danger is that we skip too much and potentially miss | |
592 | * some early allocated pages, which is better than | |
593 | * heavy lock contention. | |
61cf5feb JK |
594 | */ |
595 | if (PageBuddy(page)) { | |
ab130f91 | 596 | unsigned long order = buddy_order_unsafe(page); |
10903027 VB |
597 | |
598 | if (order > 0 && order < MAX_ORDER) | |
599 | pfn += (1UL << order) - 1; | |
61cf5feb JK |
600 | continue; |
601 | } | |
602 | ||
603 | if (PageReserved(page)) | |
604 | continue; | |
605 | ||
606 | page_ext = lookup_page_ext(page); | |
f86e4271 YS |
607 | if (unlikely(!page_ext)) |
608 | continue; | |
61cf5feb | 609 | |
dab4ead1 | 610 | /* Maybe overlapping zone */ |
61cf5feb JK |
611 | if (test_bit(PAGE_EXT_OWNER, &page_ext->flags)) |
612 | continue; | |
613 | ||
614 | /* Found early allocated page */ | |
7e2f2a0c VB |
615 | __set_page_owner_handle(page, page_ext, early_handle, |
616 | 0, 0); | |
61cf5feb JK |
617 | count++; |
618 | } | |
10903027 | 619 | cond_resched(); |
61cf5feb JK |
620 | } |
621 | ||
622 | pr_info("Node %d, zone %8s: page owner found early allocated %lu pages\n", | |
623 | pgdat->node_id, zone->name, count); | |
624 | } | |
625 | ||
626 | static void init_zones_in_node(pg_data_t *pgdat) | |
627 | { | |
628 | struct zone *zone; | |
629 | struct zone *node_zones = pgdat->node_zones; | |
61cf5feb JK |
630 | |
631 | for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) { | |
632 | if (!populated_zone(zone)) | |
633 | continue; | |
634 | ||
61cf5feb | 635 | init_pages_in_zone(pgdat, zone); |
61cf5feb JK |
636 | } |
637 | } | |
638 | ||
639 | static void init_early_allocated_pages(void) | |
640 | { | |
641 | pg_data_t *pgdat; | |
642 | ||
61cf5feb JK |
643 | for_each_online_pgdat(pgdat) |
644 | init_zones_in_node(pgdat); | |
645 | } | |
646 | ||
48c96a36 JK |
647 | static const struct file_operations proc_page_owner_operations = { |
648 | .read = read_page_owner, | |
649 | }; | |
650 | ||
651 | static int __init pageowner_init(void) | |
652 | { | |
7dd80b8a | 653 | if (!static_branch_unlikely(&page_owner_inited)) { |
48c96a36 JK |
654 | pr_info("page_owner is disabled\n"); |
655 | return 0; | |
656 | } | |
657 | ||
d9f7979c GKH |
658 | debugfs_create_file("page_owner", 0400, NULL, NULL, |
659 | &proc_page_owner_operations); | |
48c96a36 | 660 | |
d9f7979c | 661 | return 0; |
48c96a36 | 662 | } |
44c5af96 | 663 | late_initcall(pageowner_init) |