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; |
9300d8df JK |
35 | }; |
36 | ||
3645b5ec | 37 | static bool page_owner_enabled __initdata; |
7dd80b8a | 38 | DEFINE_STATIC_KEY_FALSE(page_owner_inited); |
48c96a36 | 39 | |
f2ca0b55 JK |
40 | static depot_stack_handle_t dummy_handle; |
41 | static depot_stack_handle_t failure_handle; | |
dab4ead1 | 42 | static depot_stack_handle_t early_handle; |
f2ca0b55 | 43 | |
61cf5feb JK |
44 | static void init_early_allocated_pages(void); |
45 | ||
1173194e | 46 | static int __init early_page_owner_param(char *buf) |
48c96a36 | 47 | { |
a5f1783b VB |
48 | int ret = kstrtobool(buf, &page_owner_enabled); |
49 | ||
50 | if (page_owner_enabled) | |
1c0310ad | 51 | stack_depot_request_early_init(); |
a5f1783b VB |
52 | |
53 | return ret; | |
48c96a36 JK |
54 | } |
55 | early_param("page_owner", early_page_owner_param); | |
56 | ||
cab0a7c1 | 57 | static __init bool need_page_owner(void) |
48c96a36 | 58 | { |
0fe9a448 | 59 | return page_owner_enabled; |
48c96a36 JK |
60 | } |
61 | ||
dab4ead1 | 62 | static __always_inline depot_stack_handle_t create_dummy_stack(void) |
f2ca0b55 JK |
63 | { |
64 | unsigned long entries[4]; | |
af52bf6b | 65 | unsigned int nr_entries; |
f2ca0b55 | 66 | |
af52bf6b TG |
67 | nr_entries = stack_trace_save(entries, ARRAY_SIZE(entries), 0); |
68 | return stack_depot_save(entries, nr_entries, GFP_KERNEL); | |
f2ca0b55 JK |
69 | } |
70 | ||
dab4ead1 | 71 | static noinline void register_dummy_stack(void) |
f2ca0b55 | 72 | { |
dab4ead1 VB |
73 | dummy_handle = create_dummy_stack(); |
74 | } | |
f2ca0b55 | 75 | |
dab4ead1 VB |
76 | static noinline void register_failure_stack(void) |
77 | { | |
78 | failure_handle = create_dummy_stack(); | |
79 | } | |
f2ca0b55 | 80 | |
dab4ead1 VB |
81 | static noinline void register_early_stack(void) |
82 | { | |
83 | early_handle = create_dummy_stack(); | |
f2ca0b55 JK |
84 | } |
85 | ||
cab0a7c1 | 86 | static __init void init_page_owner(void) |
48c96a36 | 87 | { |
0fe9a448 | 88 | if (!page_owner_enabled) |
48c96a36 JK |
89 | return; |
90 | ||
f2ca0b55 JK |
91 | register_dummy_stack(); |
92 | register_failure_stack(); | |
dab4ead1 | 93 | register_early_stack(); |
7dd80b8a | 94 | static_branch_enable(&page_owner_inited); |
61cf5feb | 95 | init_early_allocated_pages(); |
48c96a36 JK |
96 | } |
97 | ||
98 | struct page_ext_operations page_owner_ops = { | |
9300d8df | 99 | .size = sizeof(struct page_owner), |
48c96a36 JK |
100 | .need = need_page_owner, |
101 | .init = init_page_owner, | |
6189eb82 | 102 | .need_shared_flags = true, |
48c96a36 JK |
103 | }; |
104 | ||
9300d8df JK |
105 | static inline struct page_owner *get_page_owner(struct page_ext *page_ext) |
106 | { | |
1cac4c07 | 107 | return page_ext_data(page_ext, &page_owner_ops); |
9300d8df JK |
108 | } |
109 | ||
f2ca0b55 JK |
110 | static noinline depot_stack_handle_t save_stack(gfp_t flags) |
111 | { | |
112 | unsigned long entries[PAGE_OWNER_STACK_DEPTH]; | |
f2ca0b55 | 113 | depot_stack_handle_t handle; |
af52bf6b | 114 | unsigned int nr_entries; |
f2ca0b55 | 115 | |
f2ca0b55 | 116 | /* |
8e9b16c4 ST |
117 | * Avoid recursion. |
118 | * | |
119 | * Sometimes page metadata allocation tracking requires more | |
120 | * memory to be allocated: | |
121 | * - when new stack trace is saved to stack depot | |
122 | * - when backtrace itself is calculated (ia64) | |
f2ca0b55 | 123 | */ |
8e9b16c4 | 124 | if (current->in_page_owner) |
f2ca0b55 | 125 | return dummy_handle; |
8e9b16c4 | 126 | current->in_page_owner = 1; |
f2ca0b55 | 127 | |
8e9b16c4 | 128 | nr_entries = stack_trace_save(entries, ARRAY_SIZE(entries), 2); |
af52bf6b | 129 | handle = stack_depot_save(entries, nr_entries, flags); |
f2ca0b55 JK |
130 | if (!handle) |
131 | handle = failure_handle; | |
132 | ||
8e9b16c4 | 133 | current->in_page_owner = 0; |
f2ca0b55 JK |
134 | return handle; |
135 | } | |
136 | ||
0093de69 | 137 | void __reset_page_owner(struct page *page, unsigned short order) |
8974558f VB |
138 | { |
139 | int i; | |
140 | struct page_ext *page_ext; | |
fab765c2 | 141 | depot_stack_handle_t handle; |
8974558f | 142 | struct page_owner *page_owner; |
866b4852 | 143 | u64 free_ts_nsec = local_clock(); |
8974558f | 144 | |
b1d5488a | 145 | page_ext = page_ext_get(page); |
5556cfe8 VB |
146 | if (unlikely(!page_ext)) |
147 | return; | |
fab765c2 ST |
148 | |
149 | handle = save_stack(GFP_NOWAIT | __GFP_NOWARN); | |
8974558f | 150 | for (i = 0; i < (1 << order); i++) { |
fdf3bf80 | 151 | __clear_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags); |
0fe9a448 VB |
152 | page_owner = get_page_owner(page_ext); |
153 | page_owner->free_handle = handle; | |
866b4852 | 154 | page_owner->free_ts_nsec = free_ts_nsec; |
5556cfe8 | 155 | page_ext = page_ext_next(page_ext); |
8974558f | 156 | } |
b1d5488a | 157 | page_ext_put(page_ext); |
8974558f VB |
158 | } |
159 | ||
64ea78d2 | 160 | static inline void __set_page_owner_handle(struct page_ext *page_ext, |
161 | depot_stack_handle_t handle, | |
0093de69 | 162 | unsigned short order, gfp_t gfp_mask) |
f2ca0b55 | 163 | { |
9300d8df | 164 | struct page_owner *page_owner; |
7e2f2a0c | 165 | int i; |
05a42199 | 166 | u64 ts_nsec = local_clock(); |
48c96a36 | 167 | |
7e2f2a0c VB |
168 | for (i = 0; i < (1 << order); i++) { |
169 | page_owner = get_page_owner(page_ext); | |
170 | page_owner->handle = handle; | |
171 | page_owner->order = order; | |
172 | page_owner->gfp_mask = gfp_mask; | |
173 | page_owner->last_migrate_reason = -1; | |
9cc7e96a | 174 | page_owner->pid = current->pid; |
bf215eab | 175 | page_owner->tgid = current->tgid; |
05a42199 | 176 | page_owner->ts_nsec = ts_nsec; |
cd8c1fd8 | 177 | strscpy(page_owner->comm, current->comm, |
865ed6a3 | 178 | sizeof(page_owner->comm)); |
7e2f2a0c | 179 | __set_bit(PAGE_EXT_OWNER, &page_ext->flags); |
fdf3bf80 | 180 | __set_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags); |
48c96a36 | 181 | |
5556cfe8 | 182 | page_ext = page_ext_next(page_ext); |
7e2f2a0c | 183 | } |
48c96a36 JK |
184 | } |
185 | ||
0093de69 | 186 | noinline void __set_page_owner(struct page *page, unsigned short order, |
dab4ead1 VB |
187 | gfp_t gfp_mask) |
188 | { | |
b1d5488a | 189 | struct page_ext *page_ext; |
dab4ead1 VB |
190 | depot_stack_handle_t handle; |
191 | ||
b1d5488a CTK |
192 | handle = save_stack(gfp_mask); |
193 | ||
194 | page_ext = page_ext_get(page); | |
dab4ead1 VB |
195 | if (unlikely(!page_ext)) |
196 | return; | |
64ea78d2 | 197 | __set_page_owner_handle(page_ext, handle, order, gfp_mask); |
b1d5488a | 198 | page_ext_put(page_ext); |
dab4ead1 VB |
199 | } |
200 | ||
7cd12b4a VB |
201 | void __set_page_owner_migrate_reason(struct page *page, int reason) |
202 | { | |
b1d5488a | 203 | struct page_ext *page_ext = page_ext_get(page); |
9300d8df JK |
204 | struct page_owner *page_owner; |
205 | ||
f86e4271 YS |
206 | if (unlikely(!page_ext)) |
207 | return; | |
7cd12b4a | 208 | |
9300d8df JK |
209 | page_owner = get_page_owner(page_ext); |
210 | page_owner->last_migrate_reason = reason; | |
b1d5488a | 211 | page_ext_put(page_ext); |
7cd12b4a VB |
212 | } |
213 | ||
8fb156c9 | 214 | void __split_page_owner(struct page *page, unsigned int nr) |
e2cfc911 | 215 | { |
a9627bc5 | 216 | int i; |
b1d5488a | 217 | struct page_ext *page_ext = page_ext_get(page); |
9300d8df | 218 | struct page_owner *page_owner; |
a9627bc5 | 219 | |
f86e4271 | 220 | if (unlikely(!page_ext)) |
a9627bc5 | 221 | return; |
e2cfc911 | 222 | |
8fb156c9 | 223 | for (i = 0; i < nr; i++) { |
7e2f2a0c VB |
224 | page_owner = get_page_owner(page_ext); |
225 | page_owner->order = 0; | |
5556cfe8 | 226 | page_ext = page_ext_next(page_ext); |
7e2f2a0c | 227 | } |
b1d5488a | 228 | page_ext_put(page_ext); |
e2cfc911 JK |
229 | } |
230 | ||
19138349 | 231 | void __folio_copy_owner(struct folio *newfolio, struct folio *old) |
d435edca | 232 | { |
b1d5488a CTK |
233 | struct page_ext *old_ext; |
234 | struct page_ext *new_ext; | |
9300d8df | 235 | struct page_owner *old_page_owner, *new_page_owner; |
d435edca | 236 | |
b1d5488a CTK |
237 | old_ext = page_ext_get(&old->page); |
238 | if (unlikely(!old_ext)) | |
f86e4271 YS |
239 | return; |
240 | ||
b1d5488a CTK |
241 | new_ext = page_ext_get(&newfolio->page); |
242 | if (unlikely(!new_ext)) { | |
243 | page_ext_put(old_ext); | |
244 | return; | |
245 | } | |
246 | ||
9300d8df JK |
247 | old_page_owner = get_page_owner(old_ext); |
248 | new_page_owner = get_page_owner(new_ext); | |
249 | new_page_owner->order = old_page_owner->order; | |
250 | new_page_owner->gfp_mask = old_page_owner->gfp_mask; | |
251 | new_page_owner->last_migrate_reason = | |
252 | old_page_owner->last_migrate_reason; | |
253 | new_page_owner->handle = old_page_owner->handle; | |
9cc7e96a | 254 | new_page_owner->pid = old_page_owner->pid; |
bf215eab | 255 | new_page_owner->tgid = old_page_owner->tgid; |
9cc7e96a | 256 | new_page_owner->ts_nsec = old_page_owner->ts_nsec; |
866b4852 | 257 | new_page_owner->free_ts_nsec = old_page_owner->ts_nsec; |
865ed6a3 | 258 | strcpy(new_page_owner->comm, old_page_owner->comm); |
d435edca VB |
259 | |
260 | /* | |
19138349 | 261 | * We don't clear the bit on the old folio as it's going to be freed |
d435edca | 262 | * after migration. Until then, the info can be useful in case of |
f0953a1b | 263 | * a bug, and the overall stats will be off a bit only temporarily. |
d435edca | 264 | * Also, migrate_misplaced_transhuge_page() can still fail the |
19138349 | 265 | * migration and then we want the old folio to retain the info. But |
d435edca VB |
266 | * in that case we also don't need to explicitly clear the info from |
267 | * the new page, which will be freed. | |
268 | */ | |
269 | __set_bit(PAGE_EXT_OWNER, &new_ext->flags); | |
fdf3bf80 | 270 | __set_bit(PAGE_EXT_OWNER_ALLOCATED, &new_ext->flags); |
b1d5488a CTK |
271 | page_ext_put(new_ext); |
272 | page_ext_put(old_ext); | |
d435edca VB |
273 | } |
274 | ||
e2f612e6 JK |
275 | void pagetypeinfo_showmixedcount_print(struct seq_file *m, |
276 | pg_data_t *pgdat, struct zone *zone) | |
277 | { | |
278 | struct page *page; | |
279 | struct page_ext *page_ext; | |
9300d8df | 280 | struct page_owner *page_owner; |
1d2cae8e ML |
281 | unsigned long pfn, block_end_pfn; |
282 | unsigned long end_pfn = zone_end_pfn(zone); | |
e2f612e6 JK |
283 | unsigned long count[MIGRATE_TYPES] = { 0, }; |
284 | int pageblock_mt, page_mt; | |
285 | int i; | |
286 | ||
287 | /* Scan block by block. First and last block may be incomplete */ | |
288 | pfn = zone->zone_start_pfn; | |
289 | ||
290 | /* | |
291 | * Walk the zone in pageblock_nr_pages steps. If a page block spans | |
292 | * a zone boundary, it will be double counted between zones. This does | |
293 | * not matter as the mixed block count will still be correct | |
294 | */ | |
295 | for (; pfn < end_pfn; ) { | |
a26ee565 QC |
296 | page = pfn_to_online_page(pfn); |
297 | if (!page) { | |
e2f612e6 JK |
298 | pfn = ALIGN(pfn + 1, MAX_ORDER_NR_PAGES); |
299 | continue; | |
300 | } | |
301 | ||
4f9bc69a | 302 | block_end_pfn = pageblock_end_pfn(pfn); |
e2f612e6 JK |
303 | block_end_pfn = min(block_end_pfn, end_pfn); |
304 | ||
e2f612e6 JK |
305 | pageblock_mt = get_pageblock_migratetype(page); |
306 | ||
307 | for (; pfn < block_end_pfn; pfn++) { | |
a26ee565 | 308 | /* The pageblock is online, no need to recheck. */ |
e2f612e6 JK |
309 | page = pfn_to_page(pfn); |
310 | ||
311 | if (page_zone(page) != zone) | |
312 | continue; | |
313 | ||
314 | if (PageBuddy(page)) { | |
727c080f VM |
315 | unsigned long freepage_order; |
316 | ||
ab130f91 | 317 | freepage_order = buddy_order_unsafe(page); |
23baf831 | 318 | if (freepage_order <= MAX_ORDER) |
727c080f | 319 | pfn += (1UL << freepage_order) - 1; |
e2f612e6 JK |
320 | continue; |
321 | } | |
322 | ||
323 | if (PageReserved(page)) | |
324 | continue; | |
325 | ||
b1d5488a | 326 | page_ext = page_ext_get(page); |
e2f612e6 JK |
327 | if (unlikely(!page_ext)) |
328 | continue; | |
329 | ||
fdf3bf80 | 330 | if (!test_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags)) |
b1d5488a | 331 | goto ext_put_continue; |
e2f612e6 | 332 | |
9300d8df | 333 | page_owner = get_page_owner(page_ext); |
01c0bfe0 | 334 | page_mt = gfp_migratetype(page_owner->gfp_mask); |
e2f612e6 JK |
335 | if (pageblock_mt != page_mt) { |
336 | if (is_migrate_cma(pageblock_mt)) | |
337 | count[MIGRATE_MOVABLE]++; | |
338 | else | |
339 | count[pageblock_mt]++; | |
340 | ||
341 | pfn = block_end_pfn; | |
b1d5488a | 342 | page_ext_put(page_ext); |
e2f612e6 JK |
343 | break; |
344 | } | |
9300d8df | 345 | pfn += (1UL << page_owner->order) - 1; |
b1d5488a CTK |
346 | ext_put_continue: |
347 | page_ext_put(page_ext); | |
e2f612e6 JK |
348 | } |
349 | } | |
350 | ||
351 | /* Print counts */ | |
352 | seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name); | |
353 | for (i = 0; i < MIGRATE_TYPES; i++) | |
354 | seq_printf(m, "%12lu ", count[i]); | |
355 | seq_putc(m, '\n'); | |
356 | } | |
357 | ||
fcf89358 WL |
358 | /* |
359 | * Looking for memcg information and print it out | |
360 | */ | |
361 | static inline int print_page_owner_memcg(char *kbuf, size_t count, int ret, | |
362 | struct page *page) | |
363 | { | |
364 | #ifdef CONFIG_MEMCG | |
365 | unsigned long memcg_data; | |
366 | struct mem_cgroup *memcg; | |
367 | bool online; | |
368 | char name[80]; | |
369 | ||
370 | rcu_read_lock(); | |
371 | memcg_data = READ_ONCE(page->memcg_data); | |
372 | if (!memcg_data) | |
373 | goto out_unlock; | |
374 | ||
375 | if (memcg_data & MEMCG_DATA_OBJCGS) | |
376 | ret += scnprintf(kbuf + ret, count - ret, | |
377 | "Slab cache page\n"); | |
378 | ||
379 | memcg = page_memcg_check(page); | |
380 | if (!memcg) | |
381 | goto out_unlock; | |
382 | ||
383 | online = (memcg->css.flags & CSS_ONLINE); | |
384 | cgroup_name(memcg->css.cgroup, name, sizeof(name)); | |
385 | ret += scnprintf(kbuf + ret, count - ret, | |
386 | "Charged %sto %smemcg %s\n", | |
387 | PageMemcgKmem(page) ? "(via objcg) " : "", | |
388 | online ? "" : "offline ", | |
389 | name); | |
390 | out_unlock: | |
391 | rcu_read_unlock(); | |
392 | #endif /* CONFIG_MEMCG */ | |
393 | ||
394 | return ret; | |
395 | } | |
396 | ||
48c96a36 JK |
397 | static ssize_t |
398 | print_page_owner(char __user *buf, size_t count, unsigned long pfn, | |
9300d8df | 399 | struct page *page, struct page_owner *page_owner, |
f2ca0b55 | 400 | depot_stack_handle_t handle) |
48c96a36 | 401 | { |
af52bf6b | 402 | int ret, pageblock_mt, page_mt; |
48c96a36 JK |
403 | char *kbuf; |
404 | ||
c8f61cfc | 405 | count = min_t(size_t, count, PAGE_SIZE); |
48c96a36 JK |
406 | kbuf = kmalloc(count, GFP_KERNEL); |
407 | if (!kbuf) | |
408 | return -ENOMEM; | |
409 | ||
3ebc4397 | 410 | ret = scnprintf(kbuf, count, |
bf215eab | 411 | "Page allocated via order %u, mask %#x(%pGg), pid %d, tgid %d (%s), ts %llu ns, free_ts %llu ns\n", |
9300d8df | 412 | page_owner->order, page_owner->gfp_mask, |
9cc7e96a | 413 | &page_owner->gfp_mask, page_owner->pid, |
bf215eab YC |
414 | page_owner->tgid, page_owner->comm, |
415 | page_owner->ts_nsec, page_owner->free_ts_nsec); | |
48c96a36 | 416 | |
48c96a36 | 417 | /* Print information relevant to grouping pages by mobility */ |
0b423ca2 | 418 | pageblock_mt = get_pageblock_migratetype(page); |
01c0bfe0 | 419 | page_mt = gfp_migratetype(page_owner->gfp_mask); |
3ebc4397 | 420 | ret += scnprintf(kbuf + ret, count - ret, |
399fd496 | 421 | "PFN 0x%lx type %s Block %lu type %s Flags %pGp\n", |
48c96a36 | 422 | pfn, |
60f30350 | 423 | migratetype_names[page_mt], |
48c96a36 | 424 | pfn >> pageblock_order, |
60f30350 | 425 | migratetype_names[pageblock_mt], |
23efd080 | 426 | &page->flags); |
48c96a36 | 427 | |
0f68d45e | 428 | ret += stack_depot_snprint(handle, kbuf + ret, count - ret, 0); |
48c96a36 JK |
429 | if (ret >= count) |
430 | goto err; | |
431 | ||
9300d8df | 432 | if (page_owner->last_migrate_reason != -1) { |
3ebc4397 | 433 | ret += scnprintf(kbuf + ret, count - ret, |
7cd12b4a | 434 | "Page has been migrated, last migrate reason: %s\n", |
9300d8df | 435 | migrate_reason_names[page_owner->last_migrate_reason]); |
7cd12b4a VB |
436 | } |
437 | ||
fcf89358 WL |
438 | ret = print_page_owner_memcg(kbuf, count, ret, page); |
439 | ||
48c96a36 JK |
440 | ret += snprintf(kbuf + ret, count - ret, "\n"); |
441 | if (ret >= count) | |
442 | goto err; | |
443 | ||
444 | if (copy_to_user(buf, kbuf, ret)) | |
445 | ret = -EFAULT; | |
446 | ||
447 | kfree(kbuf); | |
448 | return ret; | |
449 | ||
450 | err: | |
451 | kfree(kbuf); | |
452 | return -ENOMEM; | |
453 | } | |
454 | ||
8bf6f451 | 455 | void __dump_page_owner(const struct page *page) |
4e462112 | 456 | { |
b1d5488a | 457 | struct page_ext *page_ext = page_ext_get((void *)page); |
9300d8df | 458 | struct page_owner *page_owner; |
f2ca0b55 | 459 | depot_stack_handle_t handle; |
8285027f SM |
460 | gfp_t gfp_mask; |
461 | int mt; | |
4e462112 | 462 | |
f86e4271 YS |
463 | if (unlikely(!page_ext)) { |
464 | pr_alert("There is not page extension available.\n"); | |
465 | return; | |
466 | } | |
9300d8df JK |
467 | |
468 | page_owner = get_page_owner(page_ext); | |
469 | gfp_mask = page_owner->gfp_mask; | |
01c0bfe0 | 470 | mt = gfp_migratetype(gfp_mask); |
f86e4271 | 471 | |
4e462112 | 472 | if (!test_bit(PAGE_EXT_OWNER, &page_ext->flags)) { |
37389167 | 473 | pr_alert("page_owner info is not present (never set?)\n"); |
b1d5488a | 474 | page_ext_put(page_ext); |
4e462112 VB |
475 | return; |
476 | } | |
477 | ||
fdf3bf80 | 478 | if (test_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags)) |
37389167 VB |
479 | pr_alert("page_owner tracks the page as allocated\n"); |
480 | else | |
481 | pr_alert("page_owner tracks the page as freed\n"); | |
482 | ||
bf215eab | 483 | 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 | 484 | page_owner->order, migratetype_names[mt], gfp_mask, &gfp_mask, |
bf215eab YC |
485 | page_owner->pid, page_owner->tgid, page_owner->comm, |
486 | page_owner->ts_nsec, page_owner->free_ts_nsec); | |
37389167 | 487 | |
9300d8df | 488 | handle = READ_ONCE(page_owner->handle); |
505be481 | 489 | if (!handle) |
37389167 | 490 | pr_alert("page_owner allocation stack trace missing\n"); |
505be481 IK |
491 | else |
492 | stack_depot_print(handle); | |
f2ca0b55 | 493 | |
8974558f VB |
494 | handle = READ_ONCE(page_owner->free_handle); |
495 | if (!handle) { | |
496 | pr_alert("page_owner free stack trace missing\n"); | |
497 | } else { | |
8974558f | 498 | pr_alert("page last free stack trace:\n"); |
505be481 | 499 | stack_depot_print(handle); |
8974558f | 500 | } |
8974558f | 501 | |
9300d8df | 502 | if (page_owner->last_migrate_reason != -1) |
4e462112 | 503 | pr_alert("page has been migrated, last migrate reason: %s\n", |
9300d8df | 504 | migrate_reason_names[page_owner->last_migrate_reason]); |
b1d5488a | 505 | page_ext_put(page_ext); |
4e462112 VB |
506 | } |
507 | ||
48c96a36 JK |
508 | static ssize_t |
509 | read_page_owner(struct file *file, char __user *buf, size_t count, loff_t *ppos) | |
510 | { | |
511 | unsigned long pfn; | |
512 | struct page *page; | |
513 | struct page_ext *page_ext; | |
9300d8df | 514 | struct page_owner *page_owner; |
f2ca0b55 | 515 | depot_stack_handle_t handle; |
48c96a36 | 516 | |
7dd80b8a | 517 | if (!static_branch_unlikely(&page_owner_inited)) |
48c96a36 JK |
518 | return -EINVAL; |
519 | ||
520 | page = NULL; | |
8f0efa81 KL |
521 | if (*ppos == 0) |
522 | pfn = min_low_pfn; | |
523 | else | |
524 | pfn = *ppos; | |
48c96a36 JK |
525 | /* Find a valid PFN or the start of a MAX_ORDER_NR_PAGES area */ |
526 | while (!pfn_valid(pfn) && (pfn & (MAX_ORDER_NR_PAGES - 1)) != 0) | |
527 | pfn++; | |
528 | ||
48c96a36 JK |
529 | /* Find an allocated page */ |
530 | for (; pfn < max_pfn; pfn++) { | |
b1d5488a CTK |
531 | /* |
532 | * This temporary page_owner is required so | |
533 | * that we can avoid the context switches while holding | |
534 | * the rcu lock and copying the page owner information to | |
535 | * user through copy_to_user() or GFP_KERNEL allocations. | |
536 | */ | |
537 | struct page_owner page_owner_tmp; | |
538 | ||
48c96a36 JK |
539 | /* |
540 | * If the new page is in a new MAX_ORDER_NR_PAGES area, | |
541 | * validate the area as existing, skip it if not | |
542 | */ | |
543 | if ((pfn & (MAX_ORDER_NR_PAGES - 1)) == 0 && !pfn_valid(pfn)) { | |
544 | pfn += MAX_ORDER_NR_PAGES - 1; | |
545 | continue; | |
546 | } | |
547 | ||
48c96a36 JK |
548 | page = pfn_to_page(pfn); |
549 | if (PageBuddy(page)) { | |
ab130f91 | 550 | unsigned long freepage_order = buddy_order_unsafe(page); |
48c96a36 | 551 | |
23baf831 | 552 | if (freepage_order <= MAX_ORDER) |
48c96a36 JK |
553 | pfn += (1UL << freepage_order) - 1; |
554 | continue; | |
555 | } | |
556 | ||
b1d5488a | 557 | page_ext = page_ext_get(page); |
f86e4271 YS |
558 | if (unlikely(!page_ext)) |
559 | continue; | |
48c96a36 JK |
560 | |
561 | /* | |
61cf5feb JK |
562 | * Some pages could be missed by concurrent allocation or free, |
563 | * because we don't hold the zone lock. | |
48c96a36 JK |
564 | */ |
565 | if (!test_bit(PAGE_EXT_OWNER, &page_ext->flags)) | |
b1d5488a | 566 | goto ext_put_continue; |
48c96a36 | 567 | |
37389167 VB |
568 | /* |
569 | * Although we do have the info about past allocation of free | |
570 | * pages, it's not relevant for current memory usage. | |
571 | */ | |
fdf3bf80 | 572 | if (!test_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags)) |
b1d5488a | 573 | goto ext_put_continue; |
37389167 | 574 | |
9300d8df JK |
575 | page_owner = get_page_owner(page_ext); |
576 | ||
7e2f2a0c VB |
577 | /* |
578 | * Don't print "tail" pages of high-order allocations as that | |
579 | * would inflate the stats. | |
580 | */ | |
581 | if (!IS_ALIGNED(pfn, 1 << page_owner->order)) | |
b1d5488a | 582 | goto ext_put_continue; |
7e2f2a0c | 583 | |
f2ca0b55 JK |
584 | /* |
585 | * Access to page_ext->handle isn't synchronous so we should | |
586 | * be careful to access it. | |
587 | */ | |
9300d8df | 588 | handle = READ_ONCE(page_owner->handle); |
f2ca0b55 | 589 | if (!handle) |
b1d5488a | 590 | goto ext_put_continue; |
f2ca0b55 | 591 | |
48c96a36 | 592 | /* Record the next PFN to read in the file offset */ |
8f0efa81 | 593 | *ppos = pfn + 1; |
48c96a36 | 594 | |
b1d5488a CTK |
595 | page_owner_tmp = *page_owner; |
596 | page_ext_put(page_ext); | |
f2ca0b55 | 597 | return print_page_owner(buf, count, pfn, page, |
b1d5488a CTK |
598 | &page_owner_tmp, handle); |
599 | ext_put_continue: | |
600 | page_ext_put(page_ext); | |
48c96a36 JK |
601 | } |
602 | ||
603 | return 0; | |
604 | } | |
605 | ||
8f0efa81 KL |
606 | static loff_t lseek_page_owner(struct file *file, loff_t offset, int orig) |
607 | { | |
608 | switch (orig) { | |
609 | case SEEK_SET: | |
610 | file->f_pos = offset; | |
611 | break; | |
612 | case SEEK_CUR: | |
613 | file->f_pos += offset; | |
614 | break; | |
615 | default: | |
616 | return -EINVAL; | |
617 | } | |
618 | return file->f_pos; | |
619 | } | |
620 | ||
61cf5feb JK |
621 | static void init_pages_in_zone(pg_data_t *pgdat, struct zone *zone) |
622 | { | |
6787c1da OS |
623 | unsigned long pfn = zone->zone_start_pfn; |
624 | unsigned long end_pfn = zone_end_pfn(zone); | |
61cf5feb JK |
625 | unsigned long count = 0; |
626 | ||
61cf5feb JK |
627 | /* |
628 | * Walk the zone in pageblock_nr_pages steps. If a page block spans | |
629 | * a zone boundary, it will be double counted between zones. This does | |
630 | * not matter as the mixed block count will still be correct | |
631 | */ | |
632 | for (; pfn < end_pfn; ) { | |
6787c1da OS |
633 | unsigned long block_end_pfn; |
634 | ||
61cf5feb JK |
635 | if (!pfn_valid(pfn)) { |
636 | pfn = ALIGN(pfn + 1, MAX_ORDER_NR_PAGES); | |
637 | continue; | |
638 | } | |
639 | ||
4f9bc69a | 640 | block_end_pfn = pageblock_end_pfn(pfn); |
61cf5feb JK |
641 | block_end_pfn = min(block_end_pfn, end_pfn); |
642 | ||
61cf5feb | 643 | for (; pfn < block_end_pfn; pfn++) { |
859a85dd | 644 | struct page *page = pfn_to_page(pfn); |
6787c1da OS |
645 | struct page_ext *page_ext; |
646 | ||
9d43f5ae JK |
647 | if (page_zone(page) != zone) |
648 | continue; | |
649 | ||
61cf5feb | 650 | /* |
10903027 VB |
651 | * To avoid having to grab zone->lock, be a little |
652 | * careful when reading buddy page order. The only | |
653 | * danger is that we skip too much and potentially miss | |
654 | * some early allocated pages, which is better than | |
655 | * heavy lock contention. | |
61cf5feb JK |
656 | */ |
657 | if (PageBuddy(page)) { | |
ab130f91 | 658 | unsigned long order = buddy_order_unsafe(page); |
10903027 | 659 | |
23baf831 | 660 | if (order > 0 && order <= MAX_ORDER) |
10903027 | 661 | pfn += (1UL << order) - 1; |
61cf5feb JK |
662 | continue; |
663 | } | |
664 | ||
665 | if (PageReserved(page)) | |
666 | continue; | |
667 | ||
b1d5488a | 668 | page_ext = page_ext_get(page); |
f86e4271 YS |
669 | if (unlikely(!page_ext)) |
670 | continue; | |
61cf5feb | 671 | |
dab4ead1 | 672 | /* Maybe overlapping zone */ |
61cf5feb | 673 | if (test_bit(PAGE_EXT_OWNER, &page_ext->flags)) |
b1d5488a | 674 | goto ext_put_continue; |
61cf5feb JK |
675 | |
676 | /* Found early allocated page */ | |
64ea78d2 | 677 | __set_page_owner_handle(page_ext, early_handle, |
7e2f2a0c | 678 | 0, 0); |
61cf5feb | 679 | count++; |
b1d5488a CTK |
680 | ext_put_continue: |
681 | page_ext_put(page_ext); | |
61cf5feb | 682 | } |
10903027 | 683 | cond_resched(); |
61cf5feb JK |
684 | } |
685 | ||
686 | pr_info("Node %d, zone %8s: page owner found early allocated %lu pages\n", | |
687 | pgdat->node_id, zone->name, count); | |
688 | } | |
689 | ||
690 | static void init_zones_in_node(pg_data_t *pgdat) | |
691 | { | |
692 | struct zone *zone; | |
693 | struct zone *node_zones = pgdat->node_zones; | |
61cf5feb JK |
694 | |
695 | for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) { | |
696 | if (!populated_zone(zone)) | |
697 | continue; | |
698 | ||
61cf5feb | 699 | init_pages_in_zone(pgdat, zone); |
61cf5feb JK |
700 | } |
701 | } | |
702 | ||
703 | static void init_early_allocated_pages(void) | |
704 | { | |
705 | pg_data_t *pgdat; | |
706 | ||
61cf5feb JK |
707 | for_each_online_pgdat(pgdat) |
708 | init_zones_in_node(pgdat); | |
709 | } | |
710 | ||
48c96a36 JK |
711 | static const struct file_operations proc_page_owner_operations = { |
712 | .read = read_page_owner, | |
8f0efa81 | 713 | .llseek = lseek_page_owner, |
48c96a36 JK |
714 | }; |
715 | ||
716 | static int __init pageowner_init(void) | |
717 | { | |
7dd80b8a | 718 | if (!static_branch_unlikely(&page_owner_inited)) { |
48c96a36 JK |
719 | pr_info("page_owner is disabled\n"); |
720 | return 0; | |
721 | } | |
722 | ||
d9f7979c GKH |
723 | debugfs_create_file("page_owner", 0400, NULL, NULL, |
724 | &proc_page_owner_operations); | |
48c96a36 | 725 | |
d9f7979c | 726 | return 0; |
48c96a36 | 727 | } |
44c5af96 | 728 | late_initcall(pageowner_init) |