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> |
fcf89358 | 13 | #include <linux/memcontrol.h> |
9cc7e96a | 14 | #include <linux/sched/clock.h> |
f2ca0b55 | 15 | |
48c96a36 JK |
16 | #include "internal.h" |
17 | ||
f2ca0b55 JK |
18 | /* |
19 | * TODO: teach PAGE_OWNER_STACK_DEPTH (__dump_page_owner and save_stack) | |
20 | * to use off stack temporal storage | |
21 | */ | |
22 | #define PAGE_OWNER_STACK_DEPTH (16) | |
23 | ||
9300d8df | 24 | struct page_owner { |
6b4c54e3 AM |
25 | unsigned short order; |
26 | short last_migrate_reason; | |
9300d8df | 27 | gfp_t gfp_mask; |
9300d8df | 28 | depot_stack_handle_t handle; |
8974558f | 29 | depot_stack_handle_t free_handle; |
9cc7e96a | 30 | u64 ts_nsec; |
866b4852 | 31 | u64 free_ts_nsec; |
865ed6a3 | 32 | char comm[TASK_COMM_LEN]; |
9cc7e96a | 33 | pid_t pid; |
bf215eab | 34 | pid_t tgid; |
1b5c65b6 BS |
35 | pid_t free_pid; |
36 | pid_t free_tgid; | |
9300d8df JK |
37 | }; |
38 | ||
4bedfb31 OS |
39 | struct stack { |
40 | struct stack_record *stack_record; | |
41 | struct stack *next; | |
42 | }; | |
43 | static struct stack dummy_stack; | |
44 | static struct stack failure_stack; | |
45 | static struct stack *stack_list; | |
217b2119 | 46 | static DEFINE_SPINLOCK(stack_list_lock); |
4bedfb31 | 47 | |
3645b5ec | 48 | static bool page_owner_enabled __initdata; |
7dd80b8a | 49 | DEFINE_STATIC_KEY_FALSE(page_owner_inited); |
48c96a36 | 50 | |
f2ca0b55 JK |
51 | static depot_stack_handle_t dummy_handle; |
52 | static depot_stack_handle_t failure_handle; | |
dab4ead1 | 53 | static depot_stack_handle_t early_handle; |
f2ca0b55 | 54 | |
61cf5feb JK |
55 | static void init_early_allocated_pages(void); |
56 | ||
7844c014 OS |
57 | static inline void set_current_in_page_owner(void) |
58 | { | |
59 | /* | |
60 | * Avoid recursion. | |
61 | * | |
62 | * We might need to allocate more memory from page_owner code, so make | |
63 | * sure to signal it in order to avoid recursion. | |
64 | */ | |
65 | current->in_page_owner = 1; | |
66 | } | |
67 | ||
68 | static inline void unset_current_in_page_owner(void) | |
69 | { | |
70 | current->in_page_owner = 0; | |
71 | } | |
72 | ||
1173194e | 73 | static int __init early_page_owner_param(char *buf) |
48c96a36 | 74 | { |
a5f1783b VB |
75 | int ret = kstrtobool(buf, &page_owner_enabled); |
76 | ||
77 | if (page_owner_enabled) | |
1c0310ad | 78 | stack_depot_request_early_init(); |
a5f1783b VB |
79 | |
80 | return ret; | |
48c96a36 JK |
81 | } |
82 | early_param("page_owner", early_page_owner_param); | |
83 | ||
cab0a7c1 | 84 | static __init bool need_page_owner(void) |
48c96a36 | 85 | { |
0fe9a448 | 86 | return page_owner_enabled; |
48c96a36 JK |
87 | } |
88 | ||
dab4ead1 | 89 | static __always_inline depot_stack_handle_t create_dummy_stack(void) |
f2ca0b55 JK |
90 | { |
91 | unsigned long entries[4]; | |
af52bf6b | 92 | unsigned int nr_entries; |
f2ca0b55 | 93 | |
af52bf6b TG |
94 | nr_entries = stack_trace_save(entries, ARRAY_SIZE(entries), 0); |
95 | return stack_depot_save(entries, nr_entries, GFP_KERNEL); | |
f2ca0b55 JK |
96 | } |
97 | ||
dab4ead1 | 98 | static noinline void register_dummy_stack(void) |
f2ca0b55 | 99 | { |
dab4ead1 VB |
100 | dummy_handle = create_dummy_stack(); |
101 | } | |
f2ca0b55 | 102 | |
dab4ead1 VB |
103 | static noinline void register_failure_stack(void) |
104 | { | |
105 | failure_handle = create_dummy_stack(); | |
106 | } | |
f2ca0b55 | 107 | |
dab4ead1 VB |
108 | static noinline void register_early_stack(void) |
109 | { | |
110 | early_handle = create_dummy_stack(); | |
f2ca0b55 JK |
111 | } |
112 | ||
cab0a7c1 | 113 | static __init void init_page_owner(void) |
48c96a36 | 114 | { |
0fe9a448 | 115 | if (!page_owner_enabled) |
48c96a36 JK |
116 | return; |
117 | ||
f2ca0b55 JK |
118 | register_dummy_stack(); |
119 | register_failure_stack(); | |
dab4ead1 | 120 | register_early_stack(); |
61cf5feb | 121 | init_early_allocated_pages(); |
4bedfb31 OS |
122 | /* Initialize dummy and failure stacks and link them to stack_list */ |
123 | dummy_stack.stack_record = __stack_depot_get_stack_record(dummy_handle); | |
124 | failure_stack.stack_record = __stack_depot_get_stack_record(failure_handle); | |
84d6ac31 OS |
125 | if (dummy_stack.stack_record) |
126 | refcount_set(&dummy_stack.stack_record->count, 1); | |
127 | if (failure_stack.stack_record) | |
128 | refcount_set(&failure_stack.stack_record->count, 1); | |
4bedfb31 OS |
129 | dummy_stack.next = &failure_stack; |
130 | stack_list = &dummy_stack; | |
0b2cf0a4 | 131 | static_branch_enable(&page_owner_inited); |
48c96a36 JK |
132 | } |
133 | ||
134 | struct page_ext_operations page_owner_ops = { | |
9300d8df | 135 | .size = sizeof(struct page_owner), |
48c96a36 JK |
136 | .need = need_page_owner, |
137 | .init = init_page_owner, | |
6189eb82 | 138 | .need_shared_flags = true, |
48c96a36 JK |
139 | }; |
140 | ||
9300d8df JK |
141 | static inline struct page_owner *get_page_owner(struct page_ext *page_ext) |
142 | { | |
1cac4c07 | 143 | return page_ext_data(page_ext, &page_owner_ops); |
9300d8df JK |
144 | } |
145 | ||
f2ca0b55 JK |
146 | static noinline depot_stack_handle_t save_stack(gfp_t flags) |
147 | { | |
148 | unsigned long entries[PAGE_OWNER_STACK_DEPTH]; | |
f2ca0b55 | 149 | depot_stack_handle_t handle; |
af52bf6b | 150 | unsigned int nr_entries; |
f2ca0b55 | 151 | |
8e9b16c4 | 152 | if (current->in_page_owner) |
f2ca0b55 JK |
153 | return dummy_handle; |
154 | ||
7844c014 | 155 | set_current_in_page_owner(); |
8e9b16c4 | 156 | nr_entries = stack_trace_save(entries, ARRAY_SIZE(entries), 2); |
af52bf6b | 157 | handle = stack_depot_save(entries, nr_entries, flags); |
f2ca0b55 JK |
158 | if (!handle) |
159 | handle = failure_handle; | |
7844c014 | 160 | unset_current_in_page_owner(); |
f2ca0b55 JK |
161 | |
162 | return handle; | |
163 | } | |
164 | ||
217b2119 OS |
165 | static void add_stack_record_to_list(struct stack_record *stack_record, |
166 | gfp_t gfp_mask) | |
167 | { | |
168 | unsigned long flags; | |
169 | struct stack *stack; | |
170 | ||
7844c014 | 171 | set_current_in_page_owner(); |
99b80ac4 | 172 | stack = kmalloc(sizeof(*stack), gfp_nested_mask(gfp_mask)); |
7844c014 OS |
173 | if (!stack) { |
174 | unset_current_in_page_owner(); | |
217b2119 | 175 | return; |
7844c014 OS |
176 | } |
177 | unset_current_in_page_owner(); | |
217b2119 OS |
178 | |
179 | stack->stack_record = stack_record; | |
180 | stack->next = NULL; | |
181 | ||
182 | spin_lock_irqsave(&stack_list_lock, flags); | |
183 | stack->next = stack_list; | |
765973a0 OS |
184 | /* |
185 | * This pairs with smp_load_acquire() from function | |
186 | * stack_start(). This guarantees that stack_start() | |
187 | * will see an updated stack_list before starting to | |
188 | * traverse the list. | |
189 | */ | |
190 | smp_store_release(&stack_list, stack); | |
217b2119 OS |
191 | spin_unlock_irqrestore(&stack_list_lock, flags); |
192 | } | |
193 | ||
f5c12105 OS |
194 | static void inc_stack_record_count(depot_stack_handle_t handle, gfp_t gfp_mask, |
195 | int nr_base_pages) | |
217b2119 OS |
196 | { |
197 | struct stack_record *stack_record = __stack_depot_get_stack_record(handle); | |
198 | ||
199 | if (!stack_record) | |
200 | return; | |
201 | ||
202 | /* | |
203 | * New stack_record's that do not use STACK_DEPOT_FLAG_GET start | |
204 | * with REFCOUNT_SATURATED to catch spurious increments of their | |
205 | * refcount. | |
206 | * Since we do not use STACK_DEPOT_FLAG_GET API, let us | |
207 | * set a refcount of 1 ourselves. | |
208 | */ | |
209 | if (refcount_read(&stack_record->count) == REFCOUNT_SATURATED) { | |
210 | int old = REFCOUNT_SATURATED; | |
211 | ||
212 | if (atomic_try_cmpxchg_relaxed(&stack_record->count.refs, &old, 1)) | |
213 | /* Add the new stack_record to our list */ | |
214 | add_stack_record_to_list(stack_record, gfp_mask); | |
215 | } | |
f5c12105 | 216 | refcount_add(nr_base_pages, &stack_record->count); |
217b2119 OS |
217 | } |
218 | ||
f5c12105 OS |
219 | static void dec_stack_record_count(depot_stack_handle_t handle, |
220 | int nr_base_pages) | |
217b2119 OS |
221 | { |
222 | struct stack_record *stack_record = __stack_depot_get_stack_record(handle); | |
223 | ||
f5c12105 OS |
224 | if (!stack_record) |
225 | return; | |
226 | ||
227 | if (refcount_sub_and_test(nr_base_pages, &stack_record->count)) | |
228 | pr_warn("%s: refcount went to 0 for %u handle\n", __func__, | |
229 | handle); | |
217b2119 OS |
230 | } |
231 | ||
ea4b5b33 OS |
232 | static inline void __update_page_owner_handle(struct page_ext *page_ext, |
233 | depot_stack_handle_t handle, | |
234 | unsigned short order, | |
235 | gfp_t gfp_mask, | |
236 | short last_migrate_reason, u64 ts_nsec, | |
237 | pid_t pid, pid_t tgid, char *comm) | |
8974558f VB |
238 | { |
239 | int i; | |
ea4b5b33 OS |
240 | struct page_owner *page_owner; |
241 | ||
242 | for (i = 0; i < (1 << order); i++) { | |
243 | page_owner = get_page_owner(page_ext); | |
244 | page_owner->handle = handle; | |
245 | page_owner->order = order; | |
246 | page_owner->gfp_mask = gfp_mask; | |
247 | page_owner->last_migrate_reason = last_migrate_reason; | |
248 | page_owner->pid = pid; | |
249 | page_owner->tgid = tgid; | |
250 | page_owner->ts_nsec = ts_nsec; | |
251 | strscpy(page_owner->comm, comm, | |
252 | sizeof(page_owner->comm)); | |
253 | __set_bit(PAGE_EXT_OWNER, &page_ext->flags); | |
254 | __set_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags); | |
255 | page_ext = page_ext_next(page_ext); | |
256 | } | |
257 | } | |
258 | ||
259 | static inline void __update_page_owner_free_handle(struct page_ext *page_ext, | |
260 | depot_stack_handle_t handle, | |
261 | unsigned short order, | |
262 | pid_t pid, pid_t tgid, | |
263 | u64 free_ts_nsec) | |
264 | { | |
265 | int i; | |
266 | struct page_owner *page_owner; | |
267 | ||
268 | for (i = 0; i < (1 << order); i++) { | |
269 | page_owner = get_page_owner(page_ext); | |
270 | /* Only __reset_page_owner() wants to clear the bit */ | |
271 | if (handle) { | |
272 | __clear_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags); | |
273 | page_owner->free_handle = handle; | |
274 | } | |
275 | page_owner->free_ts_nsec = free_ts_nsec; | |
276 | page_owner->free_pid = current->pid; | |
277 | page_owner->free_tgid = current->tgid; | |
278 | page_ext = page_ext_next(page_ext); | |
279 | } | |
280 | } | |
281 | ||
282 | void __reset_page_owner(struct page *page, unsigned short order) | |
283 | { | |
8974558f | 284 | struct page_ext *page_ext; |
fab765c2 | 285 | depot_stack_handle_t handle; |
217b2119 | 286 | depot_stack_handle_t alloc_handle; |
8974558f | 287 | struct page_owner *page_owner; |
866b4852 | 288 | u64 free_ts_nsec = local_clock(); |
8974558f | 289 | |
b1d5488a | 290 | page_ext = page_ext_get(page); |
5556cfe8 VB |
291 | if (unlikely(!page_ext)) |
292 | return; | |
fab765c2 | 293 | |
217b2119 OS |
294 | page_owner = get_page_owner(page_ext); |
295 | alloc_handle = page_owner->handle; | |
296 | ||
fab765c2 | 297 | handle = save_stack(GFP_NOWAIT | __GFP_NOWARN); |
ea4b5b33 OS |
298 | __update_page_owner_free_handle(page_ext, handle, order, current->pid, |
299 | current->tgid, free_ts_nsec); | |
b1d5488a | 300 | page_ext_put(page_ext); |
ea4b5b33 | 301 | |
217b2119 OS |
302 | if (alloc_handle != early_handle) |
303 | /* | |
304 | * early_handle is being set as a handle for all those | |
305 | * early allocated pages. See init_pages_in_zone(). | |
306 | * Since their refcount is not being incremented because | |
307 | * the machinery is not ready yet, we cannot decrement | |
308 | * their refcount either. | |
309 | */ | |
f5c12105 | 310 | dec_stack_record_count(alloc_handle, 1 << order); |
8974558f VB |
311 | } |
312 | ||
0093de69 | 313 | noinline void __set_page_owner(struct page *page, unsigned short order, |
dab4ead1 VB |
314 | gfp_t gfp_mask) |
315 | { | |
b1d5488a | 316 | struct page_ext *page_ext; |
ea4b5b33 | 317 | u64 ts_nsec = local_clock(); |
dab4ead1 VB |
318 | depot_stack_handle_t handle; |
319 | ||
b1d5488a CTK |
320 | handle = save_stack(gfp_mask); |
321 | ||
322 | page_ext = page_ext_get(page); | |
dab4ead1 VB |
323 | if (unlikely(!page_ext)) |
324 | return; | |
ea4b5b33 | 325 | __update_page_owner_handle(page_ext, handle, order, gfp_mask, -1, |
e7af4014 | 326 | ts_nsec, current->pid, current->tgid, |
ea4b5b33 | 327 | current->comm); |
b1d5488a | 328 | page_ext_put(page_ext); |
f5c12105 | 329 | inc_stack_record_count(handle, gfp_mask, 1 << order); |
dab4ead1 VB |
330 | } |
331 | ||
7cd12b4a VB |
332 | void __set_page_owner_migrate_reason(struct page *page, int reason) |
333 | { | |
b1d5488a | 334 | struct page_ext *page_ext = page_ext_get(page); |
9300d8df JK |
335 | struct page_owner *page_owner; |
336 | ||
f86e4271 YS |
337 | if (unlikely(!page_ext)) |
338 | return; | |
7cd12b4a | 339 | |
9300d8df JK |
340 | page_owner = get_page_owner(page_ext); |
341 | page_owner->last_migrate_reason = reason; | |
b1d5488a | 342 | page_ext_put(page_ext); |
7cd12b4a VB |
343 | } |
344 | ||
46d44d09 | 345 | void __split_page_owner(struct page *page, int old_order, int new_order) |
e2cfc911 | 346 | { |
a9627bc5 | 347 | int i; |
b1d5488a | 348 | struct page_ext *page_ext = page_ext_get(page); |
9300d8df | 349 | struct page_owner *page_owner; |
a9627bc5 | 350 | |
f86e4271 | 351 | if (unlikely(!page_ext)) |
a9627bc5 | 352 | return; |
e2cfc911 | 353 | |
46d44d09 | 354 | for (i = 0; i < (1 << old_order); i++) { |
7e2f2a0c | 355 | page_owner = get_page_owner(page_ext); |
46d44d09 | 356 | page_owner->order = new_order; |
5556cfe8 | 357 | page_ext = page_ext_next(page_ext); |
7e2f2a0c | 358 | } |
b1d5488a | 359 | page_ext_put(page_ext); |
e2cfc911 JK |
360 | } |
361 | ||
19138349 | 362 | void __folio_copy_owner(struct folio *newfolio, struct folio *old) |
d435edca | 363 | { |
718b1f33 | 364 | int i; |
b1d5488a CTK |
365 | struct page_ext *old_ext; |
366 | struct page_ext *new_ext; | |
ea4b5b33 | 367 | struct page_owner *old_page_owner; |
718b1f33 OS |
368 | struct page_owner *new_page_owner; |
369 | depot_stack_handle_t migrate_handle; | |
d435edca | 370 | |
b1d5488a CTK |
371 | old_ext = page_ext_get(&old->page); |
372 | if (unlikely(!old_ext)) | |
f86e4271 YS |
373 | return; |
374 | ||
b1d5488a CTK |
375 | new_ext = page_ext_get(&newfolio->page); |
376 | if (unlikely(!new_ext)) { | |
377 | page_ext_put(old_ext); | |
378 | return; | |
379 | } | |
380 | ||
9300d8df | 381 | old_page_owner = get_page_owner(old_ext); |
718b1f33 OS |
382 | new_page_owner = get_page_owner(new_ext); |
383 | migrate_handle = new_page_owner->handle; | |
ea4b5b33 OS |
384 | __update_page_owner_handle(new_ext, old_page_owner->handle, |
385 | old_page_owner->order, old_page_owner->gfp_mask, | |
386 | old_page_owner->last_migrate_reason, | |
387 | old_page_owner->ts_nsec, old_page_owner->pid, | |
388 | old_page_owner->tgid, old_page_owner->comm); | |
d435edca | 389 | /* |
ea4b5b33 OS |
390 | * Do not proactively clear PAGE_EXT_OWNER{_ALLOCATED} bits as the folio |
391 | * will be freed after migration. Keep them until then as they may be | |
392 | * useful. | |
d435edca | 393 | */ |
ea4b5b33 OS |
394 | __update_page_owner_free_handle(new_ext, 0, old_page_owner->order, |
395 | old_page_owner->free_pid, | |
396 | old_page_owner->free_tgid, | |
397 | old_page_owner->free_ts_nsec); | |
718b1f33 OS |
398 | /* |
399 | * We linked the original stack to the new folio, we need to do the same | |
400 | * for the new one and the old folio otherwise there will be an imbalance | |
401 | * when subtracting those pages from the stack. | |
402 | */ | |
403 | for (i = 0; i < (1 << new_page_owner->order); i++) { | |
404 | old_page_owner->handle = migrate_handle; | |
405 | old_ext = page_ext_next(old_ext); | |
406 | old_page_owner = get_page_owner(old_ext); | |
407 | } | |
ea4b5b33 | 408 | |
b1d5488a CTK |
409 | page_ext_put(new_ext); |
410 | page_ext_put(old_ext); | |
d435edca VB |
411 | } |
412 | ||
e2f612e6 JK |
413 | void pagetypeinfo_showmixedcount_print(struct seq_file *m, |
414 | pg_data_t *pgdat, struct zone *zone) | |
415 | { | |
416 | struct page *page; | |
417 | struct page_ext *page_ext; | |
9300d8df | 418 | struct page_owner *page_owner; |
1d2cae8e ML |
419 | unsigned long pfn, block_end_pfn; |
420 | unsigned long end_pfn = zone_end_pfn(zone); | |
e2f612e6 JK |
421 | unsigned long count[MIGRATE_TYPES] = { 0, }; |
422 | int pageblock_mt, page_mt; | |
423 | int i; | |
424 | ||
425 | /* Scan block by block. First and last block may be incomplete */ | |
426 | pfn = zone->zone_start_pfn; | |
427 | ||
428 | /* | |
429 | * Walk the zone in pageblock_nr_pages steps. If a page block spans | |
430 | * a zone boundary, it will be double counted between zones. This does | |
431 | * not matter as the mixed block count will still be correct | |
432 | */ | |
433 | for (; pfn < end_pfn; ) { | |
a26ee565 QC |
434 | page = pfn_to_online_page(pfn); |
435 | if (!page) { | |
e2f612e6 JK |
436 | pfn = ALIGN(pfn + 1, MAX_ORDER_NR_PAGES); |
437 | continue; | |
438 | } | |
439 | ||
4f9bc69a | 440 | block_end_pfn = pageblock_end_pfn(pfn); |
e2f612e6 JK |
441 | block_end_pfn = min(block_end_pfn, end_pfn); |
442 | ||
e2f612e6 JK |
443 | pageblock_mt = get_pageblock_migratetype(page); |
444 | ||
445 | for (; pfn < block_end_pfn; pfn++) { | |
a26ee565 | 446 | /* The pageblock is online, no need to recheck. */ |
e2f612e6 JK |
447 | page = pfn_to_page(pfn); |
448 | ||
449 | if (page_zone(page) != zone) | |
450 | continue; | |
451 | ||
452 | if (PageBuddy(page)) { | |
727c080f VM |
453 | unsigned long freepage_order; |
454 | ||
ab130f91 | 455 | freepage_order = buddy_order_unsafe(page); |
5e0a760b | 456 | if (freepage_order <= MAX_PAGE_ORDER) |
727c080f | 457 | pfn += (1UL << freepage_order) - 1; |
e2f612e6 JK |
458 | continue; |
459 | } | |
460 | ||
461 | if (PageReserved(page)) | |
462 | continue; | |
463 | ||
b1d5488a | 464 | page_ext = page_ext_get(page); |
e2f612e6 JK |
465 | if (unlikely(!page_ext)) |
466 | continue; | |
467 | ||
fdf3bf80 | 468 | if (!test_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags)) |
b1d5488a | 469 | goto ext_put_continue; |
e2f612e6 | 470 | |
9300d8df | 471 | page_owner = get_page_owner(page_ext); |
01c0bfe0 | 472 | page_mt = gfp_migratetype(page_owner->gfp_mask); |
e2f612e6 JK |
473 | if (pageblock_mt != page_mt) { |
474 | if (is_migrate_cma(pageblock_mt)) | |
475 | count[MIGRATE_MOVABLE]++; | |
476 | else | |
477 | count[pageblock_mt]++; | |
478 | ||
479 | pfn = block_end_pfn; | |
b1d5488a | 480 | page_ext_put(page_ext); |
e2f612e6 JK |
481 | break; |
482 | } | |
9300d8df | 483 | pfn += (1UL << page_owner->order) - 1; |
b1d5488a CTK |
484 | ext_put_continue: |
485 | page_ext_put(page_ext); | |
e2f612e6 JK |
486 | } |
487 | } | |
488 | ||
489 | /* Print counts */ | |
490 | seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name); | |
491 | for (i = 0; i < MIGRATE_TYPES; i++) | |
492 | seq_printf(m, "%12lu ", count[i]); | |
493 | seq_putc(m, '\n'); | |
494 | } | |
495 | ||
fcf89358 WL |
496 | /* |
497 | * Looking for memcg information and print it out | |
498 | */ | |
499 | static inline int print_page_owner_memcg(char *kbuf, size_t count, int ret, | |
500 | struct page *page) | |
501 | { | |
502 | #ifdef CONFIG_MEMCG | |
503 | unsigned long memcg_data; | |
504 | struct mem_cgroup *memcg; | |
505 | bool online; | |
506 | char name[80]; | |
507 | ||
508 | rcu_read_lock(); | |
509 | memcg_data = READ_ONCE(page->memcg_data); | |
510 | if (!memcg_data) | |
511 | goto out_unlock; | |
512 | ||
21c690a3 | 513 | if (memcg_data & MEMCG_DATA_OBJEXTS) |
fcf89358 WL |
514 | ret += scnprintf(kbuf + ret, count - ret, |
515 | "Slab cache page\n"); | |
516 | ||
517 | memcg = page_memcg_check(page); | |
518 | if (!memcg) | |
519 | goto out_unlock; | |
520 | ||
521 | online = (memcg->css.flags & CSS_ONLINE); | |
522 | cgroup_name(memcg->css.cgroup, name, sizeof(name)); | |
523 | ret += scnprintf(kbuf + ret, count - ret, | |
524 | "Charged %sto %smemcg %s\n", | |
525 | PageMemcgKmem(page) ? "(via objcg) " : "", | |
526 | online ? "" : "offline ", | |
527 | name); | |
528 | out_unlock: | |
529 | rcu_read_unlock(); | |
530 | #endif /* CONFIG_MEMCG */ | |
531 | ||
532 | return ret; | |
533 | } | |
534 | ||
48c96a36 JK |
535 | static ssize_t |
536 | print_page_owner(char __user *buf, size_t count, unsigned long pfn, | |
9300d8df | 537 | struct page *page, struct page_owner *page_owner, |
f2ca0b55 | 538 | depot_stack_handle_t handle) |
48c96a36 | 539 | { |
af52bf6b | 540 | int ret, pageblock_mt, page_mt; |
48c96a36 JK |
541 | char *kbuf; |
542 | ||
c8f61cfc | 543 | count = min_t(size_t, count, PAGE_SIZE); |
48c96a36 JK |
544 | kbuf = kmalloc(count, GFP_KERNEL); |
545 | if (!kbuf) | |
546 | return -ENOMEM; | |
547 | ||
3ebc4397 | 548 | ret = scnprintf(kbuf, count, |
b459f090 | 549 | "Page allocated via order %u, mask %#x(%pGg), pid %d, tgid %d (%s), ts %llu ns\n", |
9300d8df | 550 | page_owner->order, page_owner->gfp_mask, |
9cc7e96a | 551 | &page_owner->gfp_mask, page_owner->pid, |
bf215eab | 552 | page_owner->tgid, page_owner->comm, |
b459f090 | 553 | page_owner->ts_nsec); |
48c96a36 | 554 | |
48c96a36 | 555 | /* Print information relevant to grouping pages by mobility */ |
0b423ca2 | 556 | pageblock_mt = get_pageblock_migratetype(page); |
01c0bfe0 | 557 | page_mt = gfp_migratetype(page_owner->gfp_mask); |
3ebc4397 | 558 | ret += scnprintf(kbuf + ret, count - ret, |
399fd496 | 559 | "PFN 0x%lx type %s Block %lu type %s Flags %pGp\n", |
48c96a36 | 560 | pfn, |
60f30350 | 561 | migratetype_names[page_mt], |
48c96a36 | 562 | pfn >> pageblock_order, |
60f30350 | 563 | migratetype_names[pageblock_mt], |
23efd080 | 564 | &page->flags); |
48c96a36 | 565 | |
0f68d45e | 566 | ret += stack_depot_snprint(handle, kbuf + ret, count - ret, 0); |
48c96a36 JK |
567 | if (ret >= count) |
568 | goto err; | |
569 | ||
9300d8df | 570 | if (page_owner->last_migrate_reason != -1) { |
3ebc4397 | 571 | ret += scnprintf(kbuf + ret, count - ret, |
7cd12b4a | 572 | "Page has been migrated, last migrate reason: %s\n", |
9300d8df | 573 | migrate_reason_names[page_owner->last_migrate_reason]); |
7cd12b4a VB |
574 | } |
575 | ||
fcf89358 WL |
576 | ret = print_page_owner_memcg(kbuf, count, ret, page); |
577 | ||
48c96a36 JK |
578 | ret += snprintf(kbuf + ret, count - ret, "\n"); |
579 | if (ret >= count) | |
580 | goto err; | |
581 | ||
582 | if (copy_to_user(buf, kbuf, ret)) | |
583 | ret = -EFAULT; | |
584 | ||
585 | kfree(kbuf); | |
586 | return ret; | |
587 | ||
588 | err: | |
589 | kfree(kbuf); | |
590 | return -ENOMEM; | |
591 | } | |
592 | ||
8bf6f451 | 593 | void __dump_page_owner(const struct page *page) |
4e462112 | 594 | { |
b1d5488a | 595 | struct page_ext *page_ext = page_ext_get((void *)page); |
9300d8df | 596 | struct page_owner *page_owner; |
f2ca0b55 | 597 | depot_stack_handle_t handle; |
8285027f SM |
598 | gfp_t gfp_mask; |
599 | int mt; | |
4e462112 | 600 | |
f86e4271 YS |
601 | if (unlikely(!page_ext)) { |
602 | pr_alert("There is not page extension available.\n"); | |
603 | return; | |
604 | } | |
9300d8df JK |
605 | |
606 | page_owner = get_page_owner(page_ext); | |
607 | gfp_mask = page_owner->gfp_mask; | |
01c0bfe0 | 608 | mt = gfp_migratetype(gfp_mask); |
f86e4271 | 609 | |
4e462112 | 610 | if (!test_bit(PAGE_EXT_OWNER, &page_ext->flags)) { |
37389167 | 611 | pr_alert("page_owner info is not present (never set?)\n"); |
b1d5488a | 612 | page_ext_put(page_ext); |
4e462112 VB |
613 | return; |
614 | } | |
615 | ||
fdf3bf80 | 616 | if (test_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags)) |
37389167 VB |
617 | pr_alert("page_owner tracks the page as allocated\n"); |
618 | else | |
619 | pr_alert("page_owner tracks the page as freed\n"); | |
620 | ||
bf215eab | 621 | pr_alert("page last allocated via order %u, migratetype %s, gfp_mask %#x(%pGg), pid %d, tgid %d (%s), ts %llu, free_ts %llu\n", |
9cc7e96a | 622 | page_owner->order, migratetype_names[mt], gfp_mask, &gfp_mask, |
bf215eab YC |
623 | page_owner->pid, page_owner->tgid, page_owner->comm, |
624 | page_owner->ts_nsec, page_owner->free_ts_nsec); | |
37389167 | 625 | |
9300d8df | 626 | handle = READ_ONCE(page_owner->handle); |
505be481 | 627 | if (!handle) |
37389167 | 628 | pr_alert("page_owner allocation stack trace missing\n"); |
505be481 IK |
629 | else |
630 | stack_depot_print(handle); | |
f2ca0b55 | 631 | |
8974558f VB |
632 | handle = READ_ONCE(page_owner->free_handle); |
633 | if (!handle) { | |
634 | pr_alert("page_owner free stack trace missing\n"); | |
635 | } else { | |
1b5c65b6 BS |
636 | pr_alert("page last free pid %d tgid %d stack trace:\n", |
637 | page_owner->free_pid, page_owner->free_tgid); | |
505be481 | 638 | stack_depot_print(handle); |
8974558f | 639 | } |
8974558f | 640 | |
9300d8df | 641 | if (page_owner->last_migrate_reason != -1) |
4e462112 | 642 | pr_alert("page has been migrated, last migrate reason: %s\n", |
9300d8df | 643 | migrate_reason_names[page_owner->last_migrate_reason]); |
b1d5488a | 644 | page_ext_put(page_ext); |
4e462112 VB |
645 | } |
646 | ||
48c96a36 JK |
647 | static ssize_t |
648 | read_page_owner(struct file *file, char __user *buf, size_t count, loff_t *ppos) | |
649 | { | |
650 | unsigned long pfn; | |
651 | struct page *page; | |
652 | struct page_ext *page_ext; | |
9300d8df | 653 | struct page_owner *page_owner; |
f2ca0b55 | 654 | depot_stack_handle_t handle; |
48c96a36 | 655 | |
7dd80b8a | 656 | if (!static_branch_unlikely(&page_owner_inited)) |
48c96a36 JK |
657 | return -EINVAL; |
658 | ||
659 | page = NULL; | |
8f0efa81 KL |
660 | if (*ppos == 0) |
661 | pfn = min_low_pfn; | |
662 | else | |
663 | pfn = *ppos; | |
48c96a36 JK |
664 | /* Find a valid PFN or the start of a MAX_ORDER_NR_PAGES area */ |
665 | while (!pfn_valid(pfn) && (pfn & (MAX_ORDER_NR_PAGES - 1)) != 0) | |
666 | pfn++; | |
667 | ||
48c96a36 JK |
668 | /* Find an allocated page */ |
669 | for (; pfn < max_pfn; pfn++) { | |
b1d5488a CTK |
670 | /* |
671 | * This temporary page_owner is required so | |
672 | * that we can avoid the context switches while holding | |
673 | * the rcu lock and copying the page owner information to | |
674 | * user through copy_to_user() or GFP_KERNEL allocations. | |
675 | */ | |
676 | struct page_owner page_owner_tmp; | |
677 | ||
48c96a36 JK |
678 | /* |
679 | * If the new page is in a new MAX_ORDER_NR_PAGES area, | |
680 | * validate the area as existing, skip it if not | |
681 | */ | |
682 | if ((pfn & (MAX_ORDER_NR_PAGES - 1)) == 0 && !pfn_valid(pfn)) { | |
683 | pfn += MAX_ORDER_NR_PAGES - 1; | |
684 | continue; | |
685 | } | |
686 | ||
48c96a36 JK |
687 | page = pfn_to_page(pfn); |
688 | if (PageBuddy(page)) { | |
ab130f91 | 689 | unsigned long freepage_order = buddy_order_unsafe(page); |
48c96a36 | 690 | |
5e0a760b | 691 | if (freepage_order <= MAX_PAGE_ORDER) |
48c96a36 JK |
692 | pfn += (1UL << freepage_order) - 1; |
693 | continue; | |
694 | } | |
695 | ||
b1d5488a | 696 | page_ext = page_ext_get(page); |
f86e4271 YS |
697 | if (unlikely(!page_ext)) |
698 | continue; | |
48c96a36 JK |
699 | |
700 | /* | |
61cf5feb JK |
701 | * Some pages could be missed by concurrent allocation or free, |
702 | * because we don't hold the zone lock. | |
48c96a36 JK |
703 | */ |
704 | if (!test_bit(PAGE_EXT_OWNER, &page_ext->flags)) | |
b1d5488a | 705 | goto ext_put_continue; |
48c96a36 | 706 | |
37389167 VB |
707 | /* |
708 | * Although we do have the info about past allocation of free | |
709 | * pages, it's not relevant for current memory usage. | |
710 | */ | |
fdf3bf80 | 711 | if (!test_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags)) |
b1d5488a | 712 | goto ext_put_continue; |
37389167 | 713 | |
9300d8df JK |
714 | page_owner = get_page_owner(page_ext); |
715 | ||
7e2f2a0c VB |
716 | /* |
717 | * Don't print "tail" pages of high-order allocations as that | |
718 | * would inflate the stats. | |
719 | */ | |
720 | if (!IS_ALIGNED(pfn, 1 << page_owner->order)) | |
b1d5488a | 721 | goto ext_put_continue; |
7e2f2a0c | 722 | |
f2ca0b55 JK |
723 | /* |
724 | * Access to page_ext->handle isn't synchronous so we should | |
725 | * be careful to access it. | |
726 | */ | |
9300d8df | 727 | handle = READ_ONCE(page_owner->handle); |
f2ca0b55 | 728 | if (!handle) |
b1d5488a | 729 | goto ext_put_continue; |
f2ca0b55 | 730 | |
48c96a36 | 731 | /* Record the next PFN to read in the file offset */ |
8f0efa81 | 732 | *ppos = pfn + 1; |
48c96a36 | 733 | |
b1d5488a CTK |
734 | page_owner_tmp = *page_owner; |
735 | page_ext_put(page_ext); | |
f2ca0b55 | 736 | return print_page_owner(buf, count, pfn, page, |
b1d5488a CTK |
737 | &page_owner_tmp, handle); |
738 | ext_put_continue: | |
739 | page_ext_put(page_ext); | |
48c96a36 JK |
740 | } |
741 | ||
742 | return 0; | |
743 | } | |
744 | ||
8f0efa81 KL |
745 | static loff_t lseek_page_owner(struct file *file, loff_t offset, int orig) |
746 | { | |
747 | switch (orig) { | |
748 | case SEEK_SET: | |
749 | file->f_pos = offset; | |
750 | break; | |
751 | case SEEK_CUR: | |
752 | file->f_pos += offset; | |
753 | break; | |
754 | default: | |
755 | return -EINVAL; | |
756 | } | |
757 | return file->f_pos; | |
758 | } | |
759 | ||
61cf5feb JK |
760 | static void init_pages_in_zone(pg_data_t *pgdat, struct zone *zone) |
761 | { | |
6787c1da OS |
762 | unsigned long pfn = zone->zone_start_pfn; |
763 | unsigned long end_pfn = zone_end_pfn(zone); | |
61cf5feb JK |
764 | unsigned long count = 0; |
765 | ||
61cf5feb JK |
766 | /* |
767 | * Walk the zone in pageblock_nr_pages steps. If a page block spans | |
768 | * a zone boundary, it will be double counted between zones. This does | |
769 | * not matter as the mixed block count will still be correct | |
770 | */ | |
771 | for (; pfn < end_pfn; ) { | |
6787c1da OS |
772 | unsigned long block_end_pfn; |
773 | ||
61cf5feb JK |
774 | if (!pfn_valid(pfn)) { |
775 | pfn = ALIGN(pfn + 1, MAX_ORDER_NR_PAGES); | |
776 | continue; | |
777 | } | |
778 | ||
4f9bc69a | 779 | block_end_pfn = pageblock_end_pfn(pfn); |
61cf5feb JK |
780 | block_end_pfn = min(block_end_pfn, end_pfn); |
781 | ||
61cf5feb | 782 | for (; pfn < block_end_pfn; pfn++) { |
859a85dd | 783 | struct page *page = pfn_to_page(pfn); |
6787c1da OS |
784 | struct page_ext *page_ext; |
785 | ||
9d43f5ae JK |
786 | if (page_zone(page) != zone) |
787 | continue; | |
788 | ||
61cf5feb | 789 | /* |
10903027 VB |
790 | * To avoid having to grab zone->lock, be a little |
791 | * careful when reading buddy page order. The only | |
792 | * danger is that we skip too much and potentially miss | |
793 | * some early allocated pages, which is better than | |
794 | * heavy lock contention. | |
61cf5feb JK |
795 | */ |
796 | if (PageBuddy(page)) { | |
ab130f91 | 797 | unsigned long order = buddy_order_unsafe(page); |
10903027 | 798 | |
5e0a760b | 799 | if (order > 0 && order <= MAX_PAGE_ORDER) |
10903027 | 800 | pfn += (1UL << order) - 1; |
61cf5feb JK |
801 | continue; |
802 | } | |
803 | ||
804 | if (PageReserved(page)) | |
805 | continue; | |
806 | ||
b1d5488a | 807 | page_ext = page_ext_get(page); |
f86e4271 YS |
808 | if (unlikely(!page_ext)) |
809 | continue; | |
61cf5feb | 810 | |
dab4ead1 | 811 | /* Maybe overlapping zone */ |
61cf5feb | 812 | if (test_bit(PAGE_EXT_OWNER, &page_ext->flags)) |
b1d5488a | 813 | goto ext_put_continue; |
61cf5feb JK |
814 | |
815 | /* Found early allocated page */ | |
ea4b5b33 OS |
816 | __update_page_owner_handle(page_ext, early_handle, 0, 0, |
817 | -1, local_clock(), current->pid, | |
818 | current->tgid, current->comm); | |
61cf5feb | 819 | count++; |
b1d5488a CTK |
820 | ext_put_continue: |
821 | page_ext_put(page_ext); | |
61cf5feb | 822 | } |
10903027 | 823 | cond_resched(); |
61cf5feb JK |
824 | } |
825 | ||
826 | pr_info("Node %d, zone %8s: page owner found early allocated %lu pages\n", | |
827 | pgdat->node_id, zone->name, count); | |
828 | } | |
829 | ||
830 | static void init_zones_in_node(pg_data_t *pgdat) | |
831 | { | |
832 | struct zone *zone; | |
833 | struct zone *node_zones = pgdat->node_zones; | |
61cf5feb JK |
834 | |
835 | for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) { | |
836 | if (!populated_zone(zone)) | |
837 | continue; | |
838 | ||
61cf5feb | 839 | init_pages_in_zone(pgdat, zone); |
61cf5feb JK |
840 | } |
841 | } | |
842 | ||
843 | static void init_early_allocated_pages(void) | |
844 | { | |
845 | pg_data_t *pgdat; | |
846 | ||
61cf5feb JK |
847 | for_each_online_pgdat(pgdat) |
848 | init_zones_in_node(pgdat); | |
849 | } | |
850 | ||
48c96a36 JK |
851 | static const struct file_operations proc_page_owner_operations = { |
852 | .read = read_page_owner, | |
8f0efa81 | 853 | .llseek = lseek_page_owner, |
48c96a36 JK |
854 | }; |
855 | ||
765973a0 OS |
856 | static void *stack_start(struct seq_file *m, loff_t *ppos) |
857 | { | |
858 | struct stack *stack; | |
859 | ||
860 | if (*ppos == -1UL) | |
861 | return NULL; | |
862 | ||
863 | if (!*ppos) { | |
864 | /* | |
865 | * This pairs with smp_store_release() from function | |
866 | * add_stack_record_to_list(), so we get a consistent | |
867 | * value of stack_list. | |
868 | */ | |
869 | stack = smp_load_acquire(&stack_list); | |
74017458 | 870 | m->private = stack; |
765973a0 OS |
871 | } else { |
872 | stack = m->private; | |
765973a0 OS |
873 | } |
874 | ||
765973a0 OS |
875 | return stack; |
876 | } | |
877 | ||
878 | static void *stack_next(struct seq_file *m, void *v, loff_t *ppos) | |
879 | { | |
880 | struct stack *stack = v; | |
881 | ||
882 | stack = stack->next; | |
883 | *ppos = stack ? *ppos + 1 : -1UL; | |
884 | m->private = stack; | |
885 | ||
886 | return stack; | |
887 | } | |
888 | ||
f5c12105 | 889 | static unsigned long page_owner_pages_threshold; |
05bb6f4e | 890 | |
765973a0 OS |
891 | static int stack_print(struct seq_file *m, void *v) |
892 | { | |
f5c12105 | 893 | int i, nr_base_pages; |
765973a0 OS |
894 | struct stack *stack = v; |
895 | unsigned long *entries; | |
896 | unsigned long nr_entries; | |
897 | struct stack_record *stack_record = stack->stack_record; | |
898 | ||
84d6ac31 OS |
899 | if (!stack->stack_record) |
900 | return 0; | |
901 | ||
765973a0 OS |
902 | nr_entries = stack_record->size; |
903 | entries = stack_record->entries; | |
f5c12105 | 904 | nr_base_pages = refcount_read(&stack_record->count) - 1; |
765973a0 | 905 | |
f5c12105 | 906 | if (nr_base_pages < 1 || nr_base_pages < page_owner_pages_threshold) |
765973a0 OS |
907 | return 0; |
908 | ||
909 | for (i = 0; i < nr_entries; i++) | |
910 | seq_printf(m, " %pS\n", (void *)entries[i]); | |
f5c12105 | 911 | seq_printf(m, "nr_base_pages: %d\n\n", nr_base_pages); |
765973a0 OS |
912 | |
913 | return 0; | |
914 | } | |
915 | ||
916 | static void stack_stop(struct seq_file *m, void *v) | |
917 | { | |
918 | } | |
919 | ||
920 | static const struct seq_operations page_owner_stack_op = { | |
921 | .start = stack_start, | |
922 | .next = stack_next, | |
923 | .stop = stack_stop, | |
924 | .show = stack_print | |
925 | }; | |
926 | ||
927 | static int page_owner_stack_open(struct inode *inode, struct file *file) | |
928 | { | |
929 | return seq_open_private(file, &page_owner_stack_op, 0); | |
930 | } | |
931 | ||
932 | static const struct file_operations page_owner_stack_operations = { | |
933 | .open = page_owner_stack_open, | |
934 | .read = seq_read, | |
935 | .llseek = seq_lseek, | |
936 | .release = seq_release, | |
937 | }; | |
938 | ||
05bb6f4e OS |
939 | static int page_owner_threshold_get(void *data, u64 *val) |
940 | { | |
f5c12105 | 941 | *val = READ_ONCE(page_owner_pages_threshold); |
05bb6f4e OS |
942 | return 0; |
943 | } | |
944 | ||
945 | static int page_owner_threshold_set(void *data, u64 val) | |
946 | { | |
f5c12105 | 947 | WRITE_ONCE(page_owner_pages_threshold, val); |
05bb6f4e OS |
948 | return 0; |
949 | } | |
950 | ||
951 | DEFINE_SIMPLE_ATTRIBUTE(proc_page_owner_threshold, &page_owner_threshold_get, | |
952 | &page_owner_threshold_set, "%llu"); | |
953 | ||
954 | ||
48c96a36 JK |
955 | static int __init pageowner_init(void) |
956 | { | |
765973a0 OS |
957 | struct dentry *dir; |
958 | ||
7dd80b8a | 959 | if (!static_branch_unlikely(&page_owner_inited)) { |
48c96a36 JK |
960 | pr_info("page_owner is disabled\n"); |
961 | return 0; | |
962 | } | |
963 | ||
d9f7979c GKH |
964 | debugfs_create_file("page_owner", 0400, NULL, NULL, |
965 | &proc_page_owner_operations); | |
765973a0 OS |
966 | dir = debugfs_create_dir("page_owner_stacks", NULL); |
967 | debugfs_create_file("show_stacks", 0400, dir, NULL, | |
968 | &page_owner_stack_operations); | |
05bb6f4e OS |
969 | debugfs_create_file("count_threshold", 0600, dir, NULL, |
970 | &proc_page_owner_threshold); | |
48c96a36 | 971 | |
d9f7979c | 972 | return 0; |
48c96a36 | 973 | } |
44c5af96 | 974 | late_initcall(pageowner_init) |