]>
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> |
f2ca0b55 | 13 | |
48c96a36 JK |
14 | #include "internal.h" |
15 | ||
f2ca0b55 JK |
16 | /* |
17 | * TODO: teach PAGE_OWNER_STACK_DEPTH (__dump_page_owner and save_stack) | |
18 | * to use off stack temporal storage | |
19 | */ | |
20 | #define PAGE_OWNER_STACK_DEPTH (16) | |
21 | ||
9300d8df | 22 | struct page_owner { |
6b4c54e3 AM |
23 | unsigned short order; |
24 | short last_migrate_reason; | |
9300d8df | 25 | gfp_t gfp_mask; |
9300d8df JK |
26 | depot_stack_handle_t handle; |
27 | }; | |
28 | ||
48c96a36 | 29 | static bool page_owner_disabled = true; |
7dd80b8a | 30 | DEFINE_STATIC_KEY_FALSE(page_owner_inited); |
48c96a36 | 31 | |
f2ca0b55 JK |
32 | static depot_stack_handle_t dummy_handle; |
33 | static depot_stack_handle_t failure_handle; | |
dab4ead1 | 34 | static depot_stack_handle_t early_handle; |
f2ca0b55 | 35 | |
61cf5feb JK |
36 | static void init_early_allocated_pages(void); |
37 | ||
1173194e | 38 | static int __init early_page_owner_param(char *buf) |
48c96a36 JK |
39 | { |
40 | if (!buf) | |
41 | return -EINVAL; | |
42 | ||
43 | if (strcmp(buf, "on") == 0) | |
44 | page_owner_disabled = false; | |
45 | ||
46 | return 0; | |
47 | } | |
48 | early_param("page_owner", early_page_owner_param); | |
49 | ||
50 | static bool need_page_owner(void) | |
51 | { | |
52 | if (page_owner_disabled) | |
53 | return false; | |
54 | ||
55 | return true; | |
56 | } | |
57 | ||
dab4ead1 | 58 | static __always_inline depot_stack_handle_t create_dummy_stack(void) |
f2ca0b55 JK |
59 | { |
60 | unsigned long entries[4]; | |
61 | struct stack_trace dummy; | |
62 | ||
63 | dummy.nr_entries = 0; | |
64 | dummy.max_entries = ARRAY_SIZE(entries); | |
65 | dummy.entries = &entries[0]; | |
66 | dummy.skip = 0; | |
67 | ||
68 | save_stack_trace(&dummy); | |
dab4ead1 | 69 | return depot_save_stack(&dummy, GFP_KERNEL); |
f2ca0b55 JK |
70 | } |
71 | ||
dab4ead1 | 72 | static noinline void register_dummy_stack(void) |
f2ca0b55 | 73 | { |
dab4ead1 VB |
74 | dummy_handle = create_dummy_stack(); |
75 | } | |
f2ca0b55 | 76 | |
dab4ead1 VB |
77 | static noinline void register_failure_stack(void) |
78 | { | |
79 | failure_handle = create_dummy_stack(); | |
80 | } | |
f2ca0b55 | 81 | |
dab4ead1 VB |
82 | static noinline void register_early_stack(void) |
83 | { | |
84 | early_handle = create_dummy_stack(); | |
f2ca0b55 JK |
85 | } |
86 | ||
48c96a36 JK |
87 | static void init_page_owner(void) |
88 | { | |
89 | if (page_owner_disabled) | |
90 | return; | |
91 | ||
f2ca0b55 JK |
92 | register_dummy_stack(); |
93 | register_failure_stack(); | |
dab4ead1 | 94 | register_early_stack(); |
7dd80b8a | 95 | static_branch_enable(&page_owner_inited); |
61cf5feb | 96 | init_early_allocated_pages(); |
48c96a36 JK |
97 | } |
98 | ||
99 | struct page_ext_operations page_owner_ops = { | |
9300d8df | 100 | .size = sizeof(struct page_owner), |
48c96a36 JK |
101 | .need = need_page_owner, |
102 | .init = init_page_owner, | |
103 | }; | |
104 | ||
9300d8df JK |
105 | static inline struct page_owner *get_page_owner(struct page_ext *page_ext) |
106 | { | |
107 | return (void *)page_ext + page_owner_ops.offset; | |
108 | } | |
109 | ||
48c96a36 JK |
110 | void __reset_page_owner(struct page *page, unsigned int order) |
111 | { | |
112 | int i; | |
113 | struct page_ext *page_ext; | |
114 | ||
115 | for (i = 0; i < (1 << order); i++) { | |
116 | page_ext = lookup_page_ext(page + i); | |
f86e4271 YS |
117 | if (unlikely(!page_ext)) |
118 | continue; | |
48c96a36 JK |
119 | __clear_bit(PAGE_EXT_OWNER, &page_ext->flags); |
120 | } | |
121 | } | |
122 | ||
f2ca0b55 JK |
123 | static inline bool check_recursive_alloc(struct stack_trace *trace, |
124 | unsigned long ip) | |
48c96a36 | 125 | { |
299815a4 | 126 | int i; |
f2ca0b55 JK |
127 | |
128 | if (!trace->nr_entries) | |
129 | return false; | |
130 | ||
299815a4 MS |
131 | for (i = 0; i < trace->nr_entries; i++) { |
132 | if (trace->entries[i] == ip) | |
f2ca0b55 JK |
133 | return true; |
134 | } | |
f86e4271 | 135 | |
f2ca0b55 JK |
136 | return false; |
137 | } | |
138 | ||
139 | static noinline depot_stack_handle_t save_stack(gfp_t flags) | |
140 | { | |
141 | unsigned long entries[PAGE_OWNER_STACK_DEPTH]; | |
94f759d6 SR |
142 | struct stack_trace trace = { |
143 | .nr_entries = 0, | |
f2ca0b55 JK |
144 | .entries = entries, |
145 | .max_entries = PAGE_OWNER_STACK_DEPTH, | |
5f48f0bd | 146 | .skip = 2 |
94f759d6 | 147 | }; |
f2ca0b55 JK |
148 | depot_stack_handle_t handle; |
149 | ||
150 | save_stack_trace(&trace); | |
f2ca0b55 JK |
151 | |
152 | /* | |
153 | * We need to check recursion here because our request to stackdepot | |
154 | * could trigger memory allocation to save new entry. New memory | |
155 | * allocation would reach here and call depot_save_stack() again | |
156 | * if we don't catch it. There is still not enough memory in stackdepot | |
157 | * so it would try to allocate memory again and loop forever. | |
158 | */ | |
159 | if (check_recursive_alloc(&trace, _RET_IP_)) | |
160 | return dummy_handle; | |
161 | ||
162 | handle = depot_save_stack(&trace, flags); | |
163 | if (!handle) | |
164 | handle = failure_handle; | |
165 | ||
166 | return handle; | |
167 | } | |
168 | ||
dab4ead1 VB |
169 | static inline void __set_page_owner_handle(struct page_ext *page_ext, |
170 | depot_stack_handle_t handle, unsigned int order, gfp_t gfp_mask) | |
f2ca0b55 | 171 | { |
9300d8df | 172 | struct page_owner *page_owner; |
48c96a36 | 173 | |
9300d8df | 174 | page_owner = get_page_owner(page_ext); |
dab4ead1 | 175 | page_owner->handle = handle; |
9300d8df JK |
176 | page_owner->order = order; |
177 | page_owner->gfp_mask = gfp_mask; | |
178 | page_owner->last_migrate_reason = -1; | |
48c96a36 JK |
179 | |
180 | __set_bit(PAGE_EXT_OWNER, &page_ext->flags); | |
181 | } | |
182 | ||
dab4ead1 VB |
183 | noinline void __set_page_owner(struct page *page, unsigned int order, |
184 | gfp_t gfp_mask) | |
185 | { | |
186 | struct page_ext *page_ext = lookup_page_ext(page); | |
187 | depot_stack_handle_t handle; | |
188 | ||
189 | if (unlikely(!page_ext)) | |
190 | return; | |
191 | ||
192 | handle = save_stack(gfp_mask); | |
193 | __set_page_owner_handle(page_ext, handle, order, gfp_mask); | |
194 | } | |
195 | ||
7cd12b4a VB |
196 | void __set_page_owner_migrate_reason(struct page *page, int reason) |
197 | { | |
198 | struct page_ext *page_ext = lookup_page_ext(page); | |
9300d8df JK |
199 | struct page_owner *page_owner; |
200 | ||
f86e4271 YS |
201 | if (unlikely(!page_ext)) |
202 | return; | |
7cd12b4a | 203 | |
9300d8df JK |
204 | page_owner = get_page_owner(page_ext); |
205 | page_owner->last_migrate_reason = reason; | |
7cd12b4a VB |
206 | } |
207 | ||
a9627bc5 | 208 | void __split_page_owner(struct page *page, unsigned int order) |
e2cfc911 | 209 | { |
a9627bc5 | 210 | int i; |
e2cfc911 | 211 | struct page_ext *page_ext = lookup_page_ext(page); |
9300d8df | 212 | struct page_owner *page_owner; |
a9627bc5 | 213 | |
f86e4271 | 214 | if (unlikely(!page_ext)) |
a9627bc5 | 215 | return; |
e2cfc911 | 216 | |
9300d8df JK |
217 | page_owner = get_page_owner(page_ext); |
218 | page_owner->order = 0; | |
a9627bc5 JK |
219 | for (i = 1; i < (1 << order); i++) |
220 | __copy_page_owner(page, page + i); | |
e2cfc911 JK |
221 | } |
222 | ||
d435edca VB |
223 | void __copy_page_owner(struct page *oldpage, struct page *newpage) |
224 | { | |
225 | struct page_ext *old_ext = lookup_page_ext(oldpage); | |
226 | struct page_ext *new_ext = lookup_page_ext(newpage); | |
9300d8df | 227 | struct page_owner *old_page_owner, *new_page_owner; |
d435edca | 228 | |
f86e4271 YS |
229 | if (unlikely(!old_ext || !new_ext)) |
230 | return; | |
231 | ||
9300d8df JK |
232 | old_page_owner = get_page_owner(old_ext); |
233 | new_page_owner = get_page_owner(new_ext); | |
234 | new_page_owner->order = old_page_owner->order; | |
235 | new_page_owner->gfp_mask = old_page_owner->gfp_mask; | |
236 | new_page_owner->last_migrate_reason = | |
237 | old_page_owner->last_migrate_reason; | |
238 | new_page_owner->handle = old_page_owner->handle; | |
d435edca VB |
239 | |
240 | /* | |
241 | * We don't clear the bit on the oldpage as it's going to be freed | |
242 | * after migration. Until then, the info can be useful in case of | |
243 | * a bug, and the overal stats will be off a bit only temporarily. | |
244 | * Also, migrate_misplaced_transhuge_page() can still fail the | |
245 | * migration and then we want the oldpage to retain the info. But | |
246 | * in that case we also don't need to explicitly clear the info from | |
247 | * the new page, which will be freed. | |
248 | */ | |
249 | __set_bit(PAGE_EXT_OWNER, &new_ext->flags); | |
250 | } | |
251 | ||
e2f612e6 JK |
252 | void pagetypeinfo_showmixedcount_print(struct seq_file *m, |
253 | pg_data_t *pgdat, struct zone *zone) | |
254 | { | |
255 | struct page *page; | |
256 | struct page_ext *page_ext; | |
9300d8df | 257 | struct page_owner *page_owner; |
e2f612e6 JK |
258 | unsigned long pfn = zone->zone_start_pfn, block_end_pfn; |
259 | unsigned long end_pfn = pfn + zone->spanned_pages; | |
260 | unsigned long count[MIGRATE_TYPES] = { 0, }; | |
261 | int pageblock_mt, page_mt; | |
262 | int i; | |
263 | ||
264 | /* Scan block by block. First and last block may be incomplete */ | |
265 | pfn = zone->zone_start_pfn; | |
266 | ||
267 | /* | |
268 | * Walk the zone in pageblock_nr_pages steps. If a page block spans | |
269 | * a zone boundary, it will be double counted between zones. This does | |
270 | * not matter as the mixed block count will still be correct | |
271 | */ | |
272 | for (; pfn < end_pfn; ) { | |
273 | if (!pfn_valid(pfn)) { | |
274 | pfn = ALIGN(pfn + 1, MAX_ORDER_NR_PAGES); | |
275 | continue; | |
276 | } | |
277 | ||
278 | block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); | |
279 | block_end_pfn = min(block_end_pfn, end_pfn); | |
280 | ||
281 | page = pfn_to_page(pfn); | |
282 | pageblock_mt = get_pageblock_migratetype(page); | |
283 | ||
284 | for (; pfn < block_end_pfn; pfn++) { | |
285 | if (!pfn_valid_within(pfn)) | |
286 | continue; | |
287 | ||
288 | page = pfn_to_page(pfn); | |
289 | ||
290 | if (page_zone(page) != zone) | |
291 | continue; | |
292 | ||
293 | if (PageBuddy(page)) { | |
727c080f VM |
294 | unsigned long freepage_order; |
295 | ||
296 | freepage_order = page_order_unsafe(page); | |
297 | if (freepage_order < MAX_ORDER) | |
298 | pfn += (1UL << freepage_order) - 1; | |
e2f612e6 JK |
299 | continue; |
300 | } | |
301 | ||
302 | if (PageReserved(page)) | |
303 | continue; | |
304 | ||
305 | page_ext = lookup_page_ext(page); | |
306 | if (unlikely(!page_ext)) | |
307 | continue; | |
308 | ||
309 | if (!test_bit(PAGE_EXT_OWNER, &page_ext->flags)) | |
310 | continue; | |
311 | ||
9300d8df JK |
312 | page_owner = get_page_owner(page_ext); |
313 | page_mt = gfpflags_to_migratetype( | |
314 | page_owner->gfp_mask); | |
e2f612e6 JK |
315 | if (pageblock_mt != page_mt) { |
316 | if (is_migrate_cma(pageblock_mt)) | |
317 | count[MIGRATE_MOVABLE]++; | |
318 | else | |
319 | count[pageblock_mt]++; | |
320 | ||
321 | pfn = block_end_pfn; | |
322 | break; | |
323 | } | |
9300d8df | 324 | pfn += (1UL << page_owner->order) - 1; |
e2f612e6 JK |
325 | } |
326 | } | |
327 | ||
328 | /* Print counts */ | |
329 | seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name); | |
330 | for (i = 0; i < MIGRATE_TYPES; i++) | |
331 | seq_printf(m, "%12lu ", count[i]); | |
332 | seq_putc(m, '\n'); | |
333 | } | |
334 | ||
48c96a36 JK |
335 | static ssize_t |
336 | print_page_owner(char __user *buf, size_t count, unsigned long pfn, | |
9300d8df | 337 | struct page *page, struct page_owner *page_owner, |
f2ca0b55 | 338 | depot_stack_handle_t handle) |
48c96a36 JK |
339 | { |
340 | int ret; | |
341 | int pageblock_mt, page_mt; | |
342 | char *kbuf; | |
f2ca0b55 | 343 | unsigned long entries[PAGE_OWNER_STACK_DEPTH]; |
94f759d6 | 344 | struct stack_trace trace = { |
f2ca0b55 JK |
345 | .nr_entries = 0, |
346 | .entries = entries, | |
347 | .max_entries = PAGE_OWNER_STACK_DEPTH, | |
348 | .skip = 0 | |
94f759d6 | 349 | }; |
48c96a36 | 350 | |
c8f61cfc | 351 | count = min_t(size_t, count, PAGE_SIZE); |
48c96a36 JK |
352 | kbuf = kmalloc(count, GFP_KERNEL); |
353 | if (!kbuf) | |
354 | return -ENOMEM; | |
355 | ||
356 | ret = snprintf(kbuf, count, | |
60f30350 | 357 | "Page allocated via order %u, mask %#x(%pGg)\n", |
9300d8df JK |
358 | page_owner->order, page_owner->gfp_mask, |
359 | &page_owner->gfp_mask); | |
48c96a36 JK |
360 | |
361 | if (ret >= count) | |
362 | goto err; | |
363 | ||
364 | /* Print information relevant to grouping pages by mobility */ | |
0b423ca2 | 365 | pageblock_mt = get_pageblock_migratetype(page); |
9300d8df | 366 | page_mt = gfpflags_to_migratetype(page_owner->gfp_mask); |
48c96a36 | 367 | ret += snprintf(kbuf + ret, count - ret, |
60f30350 | 368 | "PFN %lu type %s Block %lu type %s Flags %#lx(%pGp)\n", |
48c96a36 | 369 | pfn, |
60f30350 | 370 | migratetype_names[page_mt], |
48c96a36 | 371 | pfn >> pageblock_order, |
60f30350 VB |
372 | migratetype_names[pageblock_mt], |
373 | page->flags, &page->flags); | |
48c96a36 JK |
374 | |
375 | if (ret >= count) | |
376 | goto err; | |
377 | ||
f2ca0b55 | 378 | depot_fetch_stack(handle, &trace); |
94f759d6 | 379 | ret += snprint_stack_trace(kbuf + ret, count - ret, &trace, 0); |
48c96a36 JK |
380 | if (ret >= count) |
381 | goto err; | |
382 | ||
9300d8df | 383 | if (page_owner->last_migrate_reason != -1) { |
7cd12b4a VB |
384 | ret += snprintf(kbuf + ret, count - ret, |
385 | "Page has been migrated, last migrate reason: %s\n", | |
9300d8df | 386 | migrate_reason_names[page_owner->last_migrate_reason]); |
7cd12b4a VB |
387 | if (ret >= count) |
388 | goto err; | |
389 | } | |
390 | ||
48c96a36 JK |
391 | ret += snprintf(kbuf + ret, count - ret, "\n"); |
392 | if (ret >= count) | |
393 | goto err; | |
394 | ||
395 | if (copy_to_user(buf, kbuf, ret)) | |
396 | ret = -EFAULT; | |
397 | ||
398 | kfree(kbuf); | |
399 | return ret; | |
400 | ||
401 | err: | |
402 | kfree(kbuf); | |
403 | return -ENOMEM; | |
404 | } | |
405 | ||
4e462112 VB |
406 | void __dump_page_owner(struct page *page) |
407 | { | |
408 | struct page_ext *page_ext = lookup_page_ext(page); | |
9300d8df | 409 | struct page_owner *page_owner; |
f2ca0b55 | 410 | unsigned long entries[PAGE_OWNER_STACK_DEPTH]; |
4e462112 | 411 | struct stack_trace trace = { |
f2ca0b55 JK |
412 | .nr_entries = 0, |
413 | .entries = entries, | |
414 | .max_entries = PAGE_OWNER_STACK_DEPTH, | |
415 | .skip = 0 | |
4e462112 | 416 | }; |
f2ca0b55 | 417 | depot_stack_handle_t handle; |
8285027f SM |
418 | gfp_t gfp_mask; |
419 | int mt; | |
4e462112 | 420 | |
f86e4271 YS |
421 | if (unlikely(!page_ext)) { |
422 | pr_alert("There is not page extension available.\n"); | |
423 | return; | |
424 | } | |
9300d8df JK |
425 | |
426 | page_owner = get_page_owner(page_ext); | |
427 | gfp_mask = page_owner->gfp_mask; | |
8285027f | 428 | mt = gfpflags_to_migratetype(gfp_mask); |
f86e4271 | 429 | |
4e462112 VB |
430 | if (!test_bit(PAGE_EXT_OWNER, &page_ext->flags)) { |
431 | pr_alert("page_owner info is not active (free page?)\n"); | |
432 | return; | |
433 | } | |
434 | ||
9300d8df | 435 | handle = READ_ONCE(page_owner->handle); |
f2ca0b55 JK |
436 | if (!handle) { |
437 | pr_alert("page_owner info is not active (free page?)\n"); | |
438 | return; | |
439 | } | |
440 | ||
441 | depot_fetch_stack(handle, &trace); | |
756a025f | 442 | pr_alert("page allocated via order %u, migratetype %s, gfp_mask %#x(%pGg)\n", |
9300d8df | 443 | page_owner->order, migratetype_names[mt], gfp_mask, &gfp_mask); |
4e462112 VB |
444 | print_stack_trace(&trace, 0); |
445 | ||
9300d8df | 446 | if (page_owner->last_migrate_reason != -1) |
4e462112 | 447 | pr_alert("page has been migrated, last migrate reason: %s\n", |
9300d8df | 448 | migrate_reason_names[page_owner->last_migrate_reason]); |
4e462112 VB |
449 | } |
450 | ||
48c96a36 JK |
451 | static ssize_t |
452 | read_page_owner(struct file *file, char __user *buf, size_t count, loff_t *ppos) | |
453 | { | |
454 | unsigned long pfn; | |
455 | struct page *page; | |
456 | struct page_ext *page_ext; | |
9300d8df | 457 | struct page_owner *page_owner; |
f2ca0b55 | 458 | depot_stack_handle_t handle; |
48c96a36 | 459 | |
7dd80b8a | 460 | if (!static_branch_unlikely(&page_owner_inited)) |
48c96a36 JK |
461 | return -EINVAL; |
462 | ||
463 | page = NULL; | |
464 | pfn = min_low_pfn + *ppos; | |
465 | ||
466 | /* Find a valid PFN or the start of a MAX_ORDER_NR_PAGES area */ | |
467 | while (!pfn_valid(pfn) && (pfn & (MAX_ORDER_NR_PAGES - 1)) != 0) | |
468 | pfn++; | |
469 | ||
470 | drain_all_pages(NULL); | |
471 | ||
472 | /* Find an allocated page */ | |
473 | for (; pfn < max_pfn; pfn++) { | |
474 | /* | |
475 | * If the new page is in a new MAX_ORDER_NR_PAGES area, | |
476 | * validate the area as existing, skip it if not | |
477 | */ | |
478 | if ((pfn & (MAX_ORDER_NR_PAGES - 1)) == 0 && !pfn_valid(pfn)) { | |
479 | pfn += MAX_ORDER_NR_PAGES - 1; | |
480 | continue; | |
481 | } | |
482 | ||
483 | /* Check for holes within a MAX_ORDER area */ | |
484 | if (!pfn_valid_within(pfn)) | |
485 | continue; | |
486 | ||
487 | page = pfn_to_page(pfn); | |
488 | if (PageBuddy(page)) { | |
489 | unsigned long freepage_order = page_order_unsafe(page); | |
490 | ||
491 | if (freepage_order < MAX_ORDER) | |
492 | pfn += (1UL << freepage_order) - 1; | |
493 | continue; | |
494 | } | |
495 | ||
496 | page_ext = lookup_page_ext(page); | |
f86e4271 YS |
497 | if (unlikely(!page_ext)) |
498 | continue; | |
48c96a36 JK |
499 | |
500 | /* | |
61cf5feb JK |
501 | * Some pages could be missed by concurrent allocation or free, |
502 | * because we don't hold the zone lock. | |
48c96a36 JK |
503 | */ |
504 | if (!test_bit(PAGE_EXT_OWNER, &page_ext->flags)) | |
505 | continue; | |
506 | ||
9300d8df JK |
507 | page_owner = get_page_owner(page_ext); |
508 | ||
f2ca0b55 JK |
509 | /* |
510 | * Access to page_ext->handle isn't synchronous so we should | |
511 | * be careful to access it. | |
512 | */ | |
9300d8df | 513 | handle = READ_ONCE(page_owner->handle); |
f2ca0b55 JK |
514 | if (!handle) |
515 | continue; | |
516 | ||
48c96a36 JK |
517 | /* Record the next PFN to read in the file offset */ |
518 | *ppos = (pfn - min_low_pfn) + 1; | |
519 | ||
f2ca0b55 | 520 | return print_page_owner(buf, count, pfn, page, |
9300d8df | 521 | page_owner, handle); |
48c96a36 JK |
522 | } |
523 | ||
524 | return 0; | |
525 | } | |
526 | ||
61cf5feb JK |
527 | static void init_pages_in_zone(pg_data_t *pgdat, struct zone *zone) |
528 | { | |
6787c1da OS |
529 | unsigned long pfn = zone->zone_start_pfn; |
530 | unsigned long end_pfn = zone_end_pfn(zone); | |
61cf5feb JK |
531 | unsigned long count = 0; |
532 | ||
61cf5feb JK |
533 | /* |
534 | * Walk the zone in pageblock_nr_pages steps. If a page block spans | |
535 | * a zone boundary, it will be double counted between zones. This does | |
536 | * not matter as the mixed block count will still be correct | |
537 | */ | |
538 | for (; pfn < end_pfn; ) { | |
6787c1da OS |
539 | unsigned long block_end_pfn; |
540 | ||
61cf5feb JK |
541 | if (!pfn_valid(pfn)) { |
542 | pfn = ALIGN(pfn + 1, MAX_ORDER_NR_PAGES); | |
543 | continue; | |
544 | } | |
545 | ||
546 | block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); | |
547 | block_end_pfn = min(block_end_pfn, end_pfn); | |
548 | ||
61cf5feb | 549 | for (; pfn < block_end_pfn; pfn++) { |
6787c1da OS |
550 | struct page *page; |
551 | struct page_ext *page_ext; | |
552 | ||
61cf5feb JK |
553 | if (!pfn_valid_within(pfn)) |
554 | continue; | |
555 | ||
556 | page = pfn_to_page(pfn); | |
557 | ||
9d43f5ae JK |
558 | if (page_zone(page) != zone) |
559 | continue; | |
560 | ||
61cf5feb | 561 | /* |
10903027 VB |
562 | * To avoid having to grab zone->lock, be a little |
563 | * careful when reading buddy page order. The only | |
564 | * danger is that we skip too much and potentially miss | |
565 | * some early allocated pages, which is better than | |
566 | * heavy lock contention. | |
61cf5feb JK |
567 | */ |
568 | if (PageBuddy(page)) { | |
10903027 VB |
569 | unsigned long order = page_order_unsafe(page); |
570 | ||
571 | if (order > 0 && order < MAX_ORDER) | |
572 | pfn += (1UL << order) - 1; | |
61cf5feb JK |
573 | continue; |
574 | } | |
575 | ||
576 | if (PageReserved(page)) | |
577 | continue; | |
578 | ||
579 | page_ext = lookup_page_ext(page); | |
f86e4271 YS |
580 | if (unlikely(!page_ext)) |
581 | continue; | |
61cf5feb | 582 | |
dab4ead1 | 583 | /* Maybe overlapping zone */ |
61cf5feb JK |
584 | if (test_bit(PAGE_EXT_OWNER, &page_ext->flags)) |
585 | continue; | |
586 | ||
587 | /* Found early allocated page */ | |
dab4ead1 | 588 | __set_page_owner_handle(page_ext, early_handle, 0, 0); |
61cf5feb JK |
589 | count++; |
590 | } | |
10903027 | 591 | cond_resched(); |
61cf5feb JK |
592 | } |
593 | ||
594 | pr_info("Node %d, zone %8s: page owner found early allocated %lu pages\n", | |
595 | pgdat->node_id, zone->name, count); | |
596 | } | |
597 | ||
598 | static void init_zones_in_node(pg_data_t *pgdat) | |
599 | { | |
600 | struct zone *zone; | |
601 | struct zone *node_zones = pgdat->node_zones; | |
61cf5feb JK |
602 | |
603 | for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) { | |
604 | if (!populated_zone(zone)) | |
605 | continue; | |
606 | ||
61cf5feb | 607 | init_pages_in_zone(pgdat, zone); |
61cf5feb JK |
608 | } |
609 | } | |
610 | ||
611 | static void init_early_allocated_pages(void) | |
612 | { | |
613 | pg_data_t *pgdat; | |
614 | ||
61cf5feb JK |
615 | for_each_online_pgdat(pgdat) |
616 | init_zones_in_node(pgdat); | |
617 | } | |
618 | ||
48c96a36 JK |
619 | static const struct file_operations proc_page_owner_operations = { |
620 | .read = read_page_owner, | |
621 | }; | |
622 | ||
623 | static int __init pageowner_init(void) | |
624 | { | |
7dd80b8a | 625 | if (!static_branch_unlikely(&page_owner_inited)) { |
48c96a36 JK |
626 | pr_info("page_owner is disabled\n"); |
627 | return 0; | |
628 | } | |
629 | ||
d9f7979c GKH |
630 | debugfs_create_file("page_owner", 0400, NULL, NULL, |
631 | &proc_page_owner_operations); | |
48c96a36 | 632 | |
d9f7979c | 633 | return 0; |
48c96a36 | 634 | } |
44c5af96 | 635 | late_initcall(pageowner_init) |