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 | ||
3645b5ec | 39 | static bool page_owner_enabled __initdata; |
7dd80b8a | 40 | DEFINE_STATIC_KEY_FALSE(page_owner_inited); |
48c96a36 | 41 | |
f2ca0b55 JK |
42 | static depot_stack_handle_t dummy_handle; |
43 | static depot_stack_handle_t failure_handle; | |
dab4ead1 | 44 | static depot_stack_handle_t early_handle; |
f2ca0b55 | 45 | |
61cf5feb JK |
46 | static void init_early_allocated_pages(void); |
47 | ||
1173194e | 48 | static int __init early_page_owner_param(char *buf) |
48c96a36 | 49 | { |
a5f1783b VB |
50 | int ret = kstrtobool(buf, &page_owner_enabled); |
51 | ||
52 | if (page_owner_enabled) | |
1c0310ad | 53 | stack_depot_request_early_init(); |
a5f1783b VB |
54 | |
55 | return ret; | |
48c96a36 JK |
56 | } |
57 | early_param("page_owner", early_page_owner_param); | |
58 | ||
cab0a7c1 | 59 | static __init bool need_page_owner(void) |
48c96a36 | 60 | { |
0fe9a448 | 61 | return page_owner_enabled; |
48c96a36 JK |
62 | } |
63 | ||
dab4ead1 | 64 | static __always_inline depot_stack_handle_t create_dummy_stack(void) |
f2ca0b55 JK |
65 | { |
66 | unsigned long entries[4]; | |
af52bf6b | 67 | unsigned int nr_entries; |
f2ca0b55 | 68 | |
af52bf6b TG |
69 | nr_entries = stack_trace_save(entries, ARRAY_SIZE(entries), 0); |
70 | return stack_depot_save(entries, nr_entries, GFP_KERNEL); | |
f2ca0b55 JK |
71 | } |
72 | ||
dab4ead1 | 73 | static noinline void register_dummy_stack(void) |
f2ca0b55 | 74 | { |
dab4ead1 VB |
75 | dummy_handle = create_dummy_stack(); |
76 | } | |
f2ca0b55 | 77 | |
dab4ead1 VB |
78 | static noinline void register_failure_stack(void) |
79 | { | |
80 | failure_handle = create_dummy_stack(); | |
81 | } | |
f2ca0b55 | 82 | |
dab4ead1 VB |
83 | static noinline void register_early_stack(void) |
84 | { | |
85 | early_handle = create_dummy_stack(); | |
f2ca0b55 JK |
86 | } |
87 | ||
cab0a7c1 | 88 | static __init void init_page_owner(void) |
48c96a36 | 89 | { |
0fe9a448 | 90 | if (!page_owner_enabled) |
48c96a36 JK |
91 | return; |
92 | ||
f2ca0b55 JK |
93 | register_dummy_stack(); |
94 | register_failure_stack(); | |
dab4ead1 | 95 | register_early_stack(); |
7dd80b8a | 96 | static_branch_enable(&page_owner_inited); |
61cf5feb | 97 | init_early_allocated_pages(); |
48c96a36 JK |
98 | } |
99 | ||
100 | struct page_ext_operations page_owner_ops = { | |
9300d8df | 101 | .size = sizeof(struct page_owner), |
48c96a36 JK |
102 | .need = need_page_owner, |
103 | .init = init_page_owner, | |
6189eb82 | 104 | .need_shared_flags = true, |
48c96a36 JK |
105 | }; |
106 | ||
9300d8df JK |
107 | static inline struct page_owner *get_page_owner(struct page_ext *page_ext) |
108 | { | |
1cac4c07 | 109 | return page_ext_data(page_ext, &page_owner_ops); |
9300d8df JK |
110 | } |
111 | ||
f2ca0b55 JK |
112 | static noinline depot_stack_handle_t save_stack(gfp_t flags) |
113 | { | |
114 | unsigned long entries[PAGE_OWNER_STACK_DEPTH]; | |
f2ca0b55 | 115 | depot_stack_handle_t handle; |
af52bf6b | 116 | unsigned int nr_entries; |
f2ca0b55 | 117 | |
f2ca0b55 | 118 | /* |
8e9b16c4 ST |
119 | * Avoid recursion. |
120 | * | |
121 | * Sometimes page metadata allocation tracking requires more | |
122 | * memory to be allocated: | |
123 | * - when new stack trace is saved to stack depot | |
f2ca0b55 | 124 | */ |
8e9b16c4 | 125 | if (current->in_page_owner) |
f2ca0b55 | 126 | return dummy_handle; |
8e9b16c4 | 127 | current->in_page_owner = 1; |
f2ca0b55 | 128 | |
8e9b16c4 | 129 | nr_entries = stack_trace_save(entries, ARRAY_SIZE(entries), 2); |
af52bf6b | 130 | handle = stack_depot_save(entries, nr_entries, flags); |
f2ca0b55 JK |
131 | if (!handle) |
132 | handle = failure_handle; | |
133 | ||
8e9b16c4 | 134 | current->in_page_owner = 0; |
f2ca0b55 JK |
135 | return handle; |
136 | } | |
137 | ||
0093de69 | 138 | void __reset_page_owner(struct page *page, unsigned short order) |
8974558f VB |
139 | { |
140 | int i; | |
141 | struct page_ext *page_ext; | |
fab765c2 | 142 | depot_stack_handle_t handle; |
8974558f | 143 | struct page_owner *page_owner; |
866b4852 | 144 | u64 free_ts_nsec = local_clock(); |
8974558f | 145 | |
b1d5488a | 146 | page_ext = page_ext_get(page); |
5556cfe8 VB |
147 | if (unlikely(!page_ext)) |
148 | return; | |
fab765c2 ST |
149 | |
150 | handle = save_stack(GFP_NOWAIT | __GFP_NOWARN); | |
8974558f | 151 | for (i = 0; i < (1 << order); i++) { |
fdf3bf80 | 152 | __clear_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags); |
0fe9a448 VB |
153 | page_owner = get_page_owner(page_ext); |
154 | page_owner->free_handle = handle; | |
866b4852 | 155 | page_owner->free_ts_nsec = free_ts_nsec; |
1b5c65b6 BS |
156 | page_owner->free_pid = current->pid; |
157 | page_owner->free_tgid = current->tgid; | |
5556cfe8 | 158 | page_ext = page_ext_next(page_ext); |
8974558f | 159 | } |
b1d5488a | 160 | page_ext_put(page_ext); |
8974558f VB |
161 | } |
162 | ||
64ea78d2 | 163 | static inline void __set_page_owner_handle(struct page_ext *page_ext, |
164 | depot_stack_handle_t handle, | |
0093de69 | 165 | unsigned short order, gfp_t gfp_mask) |
f2ca0b55 | 166 | { |
9300d8df | 167 | struct page_owner *page_owner; |
7e2f2a0c | 168 | int i; |
05a42199 | 169 | u64 ts_nsec = local_clock(); |
48c96a36 | 170 | |
7e2f2a0c VB |
171 | for (i = 0; i < (1 << order); i++) { |
172 | page_owner = get_page_owner(page_ext); | |
173 | page_owner->handle = handle; | |
174 | page_owner->order = order; | |
175 | page_owner->gfp_mask = gfp_mask; | |
176 | page_owner->last_migrate_reason = -1; | |
9cc7e96a | 177 | page_owner->pid = current->pid; |
bf215eab | 178 | page_owner->tgid = current->tgid; |
05a42199 | 179 | page_owner->ts_nsec = ts_nsec; |
cd8c1fd8 | 180 | strscpy(page_owner->comm, current->comm, |
865ed6a3 | 181 | sizeof(page_owner->comm)); |
7e2f2a0c | 182 | __set_bit(PAGE_EXT_OWNER, &page_ext->flags); |
fdf3bf80 | 183 | __set_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags); |
48c96a36 | 184 | |
5556cfe8 | 185 | page_ext = page_ext_next(page_ext); |
7e2f2a0c | 186 | } |
48c96a36 JK |
187 | } |
188 | ||
0093de69 | 189 | noinline void __set_page_owner(struct page *page, unsigned short order, |
dab4ead1 VB |
190 | gfp_t gfp_mask) |
191 | { | |
b1d5488a | 192 | struct page_ext *page_ext; |
dab4ead1 VB |
193 | depot_stack_handle_t handle; |
194 | ||
b1d5488a CTK |
195 | handle = save_stack(gfp_mask); |
196 | ||
197 | page_ext = page_ext_get(page); | |
dab4ead1 VB |
198 | if (unlikely(!page_ext)) |
199 | return; | |
64ea78d2 | 200 | __set_page_owner_handle(page_ext, handle, order, gfp_mask); |
b1d5488a | 201 | page_ext_put(page_ext); |
dab4ead1 VB |
202 | } |
203 | ||
7cd12b4a VB |
204 | void __set_page_owner_migrate_reason(struct page *page, int reason) |
205 | { | |
b1d5488a | 206 | struct page_ext *page_ext = page_ext_get(page); |
9300d8df JK |
207 | struct page_owner *page_owner; |
208 | ||
f86e4271 YS |
209 | if (unlikely(!page_ext)) |
210 | return; | |
7cd12b4a | 211 | |
9300d8df JK |
212 | page_owner = get_page_owner(page_ext); |
213 | page_owner->last_migrate_reason = reason; | |
b1d5488a | 214 | page_ext_put(page_ext); |
7cd12b4a VB |
215 | } |
216 | ||
8fb156c9 | 217 | void __split_page_owner(struct page *page, unsigned int nr) |
e2cfc911 | 218 | { |
a9627bc5 | 219 | int i; |
b1d5488a | 220 | struct page_ext *page_ext = page_ext_get(page); |
9300d8df | 221 | struct page_owner *page_owner; |
a9627bc5 | 222 | |
f86e4271 | 223 | if (unlikely(!page_ext)) |
a9627bc5 | 224 | return; |
e2cfc911 | 225 | |
8fb156c9 | 226 | for (i = 0; i < nr; i++) { |
7e2f2a0c VB |
227 | page_owner = get_page_owner(page_ext); |
228 | page_owner->order = 0; | |
5556cfe8 | 229 | page_ext = page_ext_next(page_ext); |
7e2f2a0c | 230 | } |
b1d5488a | 231 | page_ext_put(page_ext); |
e2cfc911 JK |
232 | } |
233 | ||
19138349 | 234 | void __folio_copy_owner(struct folio *newfolio, struct folio *old) |
d435edca | 235 | { |
b1d5488a CTK |
236 | struct page_ext *old_ext; |
237 | struct page_ext *new_ext; | |
9300d8df | 238 | struct page_owner *old_page_owner, *new_page_owner; |
d435edca | 239 | |
b1d5488a CTK |
240 | old_ext = page_ext_get(&old->page); |
241 | if (unlikely(!old_ext)) | |
f86e4271 YS |
242 | return; |
243 | ||
b1d5488a CTK |
244 | new_ext = page_ext_get(&newfolio->page); |
245 | if (unlikely(!new_ext)) { | |
246 | page_ext_put(old_ext); | |
247 | return; | |
248 | } | |
249 | ||
9300d8df JK |
250 | old_page_owner = get_page_owner(old_ext); |
251 | new_page_owner = get_page_owner(new_ext); | |
252 | new_page_owner->order = old_page_owner->order; | |
253 | new_page_owner->gfp_mask = old_page_owner->gfp_mask; | |
254 | new_page_owner->last_migrate_reason = | |
255 | old_page_owner->last_migrate_reason; | |
256 | new_page_owner->handle = old_page_owner->handle; | |
9cc7e96a | 257 | new_page_owner->pid = old_page_owner->pid; |
bf215eab | 258 | new_page_owner->tgid = old_page_owner->tgid; |
1b5c65b6 BS |
259 | new_page_owner->free_pid = old_page_owner->free_pid; |
260 | new_page_owner->free_tgid = old_page_owner->free_tgid; | |
9cc7e96a | 261 | new_page_owner->ts_nsec = old_page_owner->ts_nsec; |
866b4852 | 262 | new_page_owner->free_ts_nsec = old_page_owner->ts_nsec; |
865ed6a3 | 263 | strcpy(new_page_owner->comm, old_page_owner->comm); |
d435edca VB |
264 | |
265 | /* | |
19138349 | 266 | * We don't clear the bit on the old folio as it's going to be freed |
d435edca | 267 | * after migration. Until then, the info can be useful in case of |
f0953a1b | 268 | * a bug, and the overall stats will be off a bit only temporarily. |
d435edca | 269 | * Also, migrate_misplaced_transhuge_page() can still fail the |
19138349 | 270 | * migration and then we want the old folio to retain the info. But |
d435edca VB |
271 | * in that case we also don't need to explicitly clear the info from |
272 | * the new page, which will be freed. | |
273 | */ | |
274 | __set_bit(PAGE_EXT_OWNER, &new_ext->flags); | |
fdf3bf80 | 275 | __set_bit(PAGE_EXT_OWNER_ALLOCATED, &new_ext->flags); |
b1d5488a CTK |
276 | page_ext_put(new_ext); |
277 | page_ext_put(old_ext); | |
d435edca VB |
278 | } |
279 | ||
e2f612e6 JK |
280 | void pagetypeinfo_showmixedcount_print(struct seq_file *m, |
281 | pg_data_t *pgdat, struct zone *zone) | |
282 | { | |
283 | struct page *page; | |
284 | struct page_ext *page_ext; | |
9300d8df | 285 | struct page_owner *page_owner; |
1d2cae8e ML |
286 | unsigned long pfn, block_end_pfn; |
287 | unsigned long end_pfn = zone_end_pfn(zone); | |
e2f612e6 JK |
288 | unsigned long count[MIGRATE_TYPES] = { 0, }; |
289 | int pageblock_mt, page_mt; | |
290 | int i; | |
291 | ||
292 | /* Scan block by block. First and last block may be incomplete */ | |
293 | pfn = zone->zone_start_pfn; | |
294 | ||
295 | /* | |
296 | * Walk the zone in pageblock_nr_pages steps. If a page block spans | |
297 | * a zone boundary, it will be double counted between zones. This does | |
298 | * not matter as the mixed block count will still be correct | |
299 | */ | |
300 | for (; pfn < end_pfn; ) { | |
a26ee565 QC |
301 | page = pfn_to_online_page(pfn); |
302 | if (!page) { | |
e2f612e6 JK |
303 | pfn = ALIGN(pfn + 1, MAX_ORDER_NR_PAGES); |
304 | continue; | |
305 | } | |
306 | ||
4f9bc69a | 307 | block_end_pfn = pageblock_end_pfn(pfn); |
e2f612e6 JK |
308 | block_end_pfn = min(block_end_pfn, end_pfn); |
309 | ||
e2f612e6 JK |
310 | pageblock_mt = get_pageblock_migratetype(page); |
311 | ||
312 | for (; pfn < block_end_pfn; pfn++) { | |
a26ee565 | 313 | /* The pageblock is online, no need to recheck. */ |
e2f612e6 JK |
314 | page = pfn_to_page(pfn); |
315 | ||
316 | if (page_zone(page) != zone) | |
317 | continue; | |
318 | ||
319 | if (PageBuddy(page)) { | |
727c080f VM |
320 | unsigned long freepage_order; |
321 | ||
ab130f91 | 322 | freepage_order = buddy_order_unsafe(page); |
5e0a760b | 323 | if (freepage_order <= MAX_PAGE_ORDER) |
727c080f | 324 | pfn += (1UL << freepage_order) - 1; |
e2f612e6 JK |
325 | continue; |
326 | } | |
327 | ||
328 | if (PageReserved(page)) | |
329 | continue; | |
330 | ||
b1d5488a | 331 | page_ext = page_ext_get(page); |
e2f612e6 JK |
332 | if (unlikely(!page_ext)) |
333 | continue; | |
334 | ||
fdf3bf80 | 335 | if (!test_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags)) |
b1d5488a | 336 | goto ext_put_continue; |
e2f612e6 | 337 | |
9300d8df | 338 | page_owner = get_page_owner(page_ext); |
01c0bfe0 | 339 | page_mt = gfp_migratetype(page_owner->gfp_mask); |
e2f612e6 JK |
340 | if (pageblock_mt != page_mt) { |
341 | if (is_migrate_cma(pageblock_mt)) | |
342 | count[MIGRATE_MOVABLE]++; | |
343 | else | |
344 | count[pageblock_mt]++; | |
345 | ||
346 | pfn = block_end_pfn; | |
b1d5488a | 347 | page_ext_put(page_ext); |
e2f612e6 JK |
348 | break; |
349 | } | |
9300d8df | 350 | pfn += (1UL << page_owner->order) - 1; |
b1d5488a CTK |
351 | ext_put_continue: |
352 | page_ext_put(page_ext); | |
e2f612e6 JK |
353 | } |
354 | } | |
355 | ||
356 | /* Print counts */ | |
357 | seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name); | |
358 | for (i = 0; i < MIGRATE_TYPES; i++) | |
359 | seq_printf(m, "%12lu ", count[i]); | |
360 | seq_putc(m, '\n'); | |
361 | } | |
362 | ||
fcf89358 WL |
363 | /* |
364 | * Looking for memcg information and print it out | |
365 | */ | |
366 | static inline int print_page_owner_memcg(char *kbuf, size_t count, int ret, | |
367 | struct page *page) | |
368 | { | |
369 | #ifdef CONFIG_MEMCG | |
370 | unsigned long memcg_data; | |
371 | struct mem_cgroup *memcg; | |
372 | bool online; | |
373 | char name[80]; | |
374 | ||
375 | rcu_read_lock(); | |
376 | memcg_data = READ_ONCE(page->memcg_data); | |
377 | if (!memcg_data) | |
378 | goto out_unlock; | |
379 | ||
380 | if (memcg_data & MEMCG_DATA_OBJCGS) | |
381 | ret += scnprintf(kbuf + ret, count - ret, | |
382 | "Slab cache page\n"); | |
383 | ||
384 | memcg = page_memcg_check(page); | |
385 | if (!memcg) | |
386 | goto out_unlock; | |
387 | ||
388 | online = (memcg->css.flags & CSS_ONLINE); | |
389 | cgroup_name(memcg->css.cgroup, name, sizeof(name)); | |
390 | ret += scnprintf(kbuf + ret, count - ret, | |
391 | "Charged %sto %smemcg %s\n", | |
392 | PageMemcgKmem(page) ? "(via objcg) " : "", | |
393 | online ? "" : "offline ", | |
394 | name); | |
395 | out_unlock: | |
396 | rcu_read_unlock(); | |
397 | #endif /* CONFIG_MEMCG */ | |
398 | ||
399 | return ret; | |
400 | } | |
401 | ||
48c96a36 JK |
402 | static ssize_t |
403 | print_page_owner(char __user *buf, size_t count, unsigned long pfn, | |
9300d8df | 404 | struct page *page, struct page_owner *page_owner, |
f2ca0b55 | 405 | depot_stack_handle_t handle) |
48c96a36 | 406 | { |
af52bf6b | 407 | int ret, pageblock_mt, page_mt; |
48c96a36 JK |
408 | char *kbuf; |
409 | ||
c8f61cfc | 410 | count = min_t(size_t, count, PAGE_SIZE); |
48c96a36 JK |
411 | kbuf = kmalloc(count, GFP_KERNEL); |
412 | if (!kbuf) | |
413 | return -ENOMEM; | |
414 | ||
3ebc4397 | 415 | ret = scnprintf(kbuf, count, |
b459f090 | 416 | "Page allocated via order %u, mask %#x(%pGg), pid %d, tgid %d (%s), ts %llu ns\n", |
9300d8df | 417 | page_owner->order, page_owner->gfp_mask, |
9cc7e96a | 418 | &page_owner->gfp_mask, page_owner->pid, |
bf215eab | 419 | page_owner->tgid, page_owner->comm, |
b459f090 | 420 | page_owner->ts_nsec); |
48c96a36 | 421 | |
48c96a36 | 422 | /* Print information relevant to grouping pages by mobility */ |
0b423ca2 | 423 | pageblock_mt = get_pageblock_migratetype(page); |
01c0bfe0 | 424 | page_mt = gfp_migratetype(page_owner->gfp_mask); |
3ebc4397 | 425 | ret += scnprintf(kbuf + ret, count - ret, |
399fd496 | 426 | "PFN 0x%lx type %s Block %lu type %s Flags %pGp\n", |
48c96a36 | 427 | pfn, |
60f30350 | 428 | migratetype_names[page_mt], |
48c96a36 | 429 | pfn >> pageblock_order, |
60f30350 | 430 | migratetype_names[pageblock_mt], |
23efd080 | 431 | &page->flags); |
48c96a36 | 432 | |
0f68d45e | 433 | ret += stack_depot_snprint(handle, kbuf + ret, count - ret, 0); |
48c96a36 JK |
434 | if (ret >= count) |
435 | goto err; | |
436 | ||
9300d8df | 437 | if (page_owner->last_migrate_reason != -1) { |
3ebc4397 | 438 | ret += scnprintf(kbuf + ret, count - ret, |
7cd12b4a | 439 | "Page has been migrated, last migrate reason: %s\n", |
9300d8df | 440 | migrate_reason_names[page_owner->last_migrate_reason]); |
7cd12b4a VB |
441 | } |
442 | ||
fcf89358 WL |
443 | ret = print_page_owner_memcg(kbuf, count, ret, page); |
444 | ||
48c96a36 JK |
445 | ret += snprintf(kbuf + ret, count - ret, "\n"); |
446 | if (ret >= count) | |
447 | goto err; | |
448 | ||
449 | if (copy_to_user(buf, kbuf, ret)) | |
450 | ret = -EFAULT; | |
451 | ||
452 | kfree(kbuf); | |
453 | return ret; | |
454 | ||
455 | err: | |
456 | kfree(kbuf); | |
457 | return -ENOMEM; | |
458 | } | |
459 | ||
8bf6f451 | 460 | void __dump_page_owner(const struct page *page) |
4e462112 | 461 | { |
b1d5488a | 462 | struct page_ext *page_ext = page_ext_get((void *)page); |
9300d8df | 463 | struct page_owner *page_owner; |
f2ca0b55 | 464 | depot_stack_handle_t handle; |
8285027f SM |
465 | gfp_t gfp_mask; |
466 | int mt; | |
4e462112 | 467 | |
f86e4271 YS |
468 | if (unlikely(!page_ext)) { |
469 | pr_alert("There is not page extension available.\n"); | |
470 | return; | |
471 | } | |
9300d8df JK |
472 | |
473 | page_owner = get_page_owner(page_ext); | |
474 | gfp_mask = page_owner->gfp_mask; | |
01c0bfe0 | 475 | mt = gfp_migratetype(gfp_mask); |
f86e4271 | 476 | |
4e462112 | 477 | if (!test_bit(PAGE_EXT_OWNER, &page_ext->flags)) { |
37389167 | 478 | pr_alert("page_owner info is not present (never set?)\n"); |
b1d5488a | 479 | page_ext_put(page_ext); |
4e462112 VB |
480 | return; |
481 | } | |
482 | ||
fdf3bf80 | 483 | if (test_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags)) |
37389167 VB |
484 | pr_alert("page_owner tracks the page as allocated\n"); |
485 | else | |
486 | pr_alert("page_owner tracks the page as freed\n"); | |
487 | ||
bf215eab | 488 | 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 | 489 | page_owner->order, migratetype_names[mt], gfp_mask, &gfp_mask, |
bf215eab YC |
490 | page_owner->pid, page_owner->tgid, page_owner->comm, |
491 | page_owner->ts_nsec, page_owner->free_ts_nsec); | |
37389167 | 492 | |
9300d8df | 493 | handle = READ_ONCE(page_owner->handle); |
505be481 | 494 | if (!handle) |
37389167 | 495 | pr_alert("page_owner allocation stack trace missing\n"); |
505be481 IK |
496 | else |
497 | stack_depot_print(handle); | |
f2ca0b55 | 498 | |
8974558f VB |
499 | handle = READ_ONCE(page_owner->free_handle); |
500 | if (!handle) { | |
501 | pr_alert("page_owner free stack trace missing\n"); | |
502 | } else { | |
1b5c65b6 BS |
503 | pr_alert("page last free pid %d tgid %d stack trace:\n", |
504 | page_owner->free_pid, page_owner->free_tgid); | |
505be481 | 505 | stack_depot_print(handle); |
8974558f | 506 | } |
8974558f | 507 | |
9300d8df | 508 | if (page_owner->last_migrate_reason != -1) |
4e462112 | 509 | pr_alert("page has been migrated, last migrate reason: %s\n", |
9300d8df | 510 | migrate_reason_names[page_owner->last_migrate_reason]); |
b1d5488a | 511 | page_ext_put(page_ext); |
4e462112 VB |
512 | } |
513 | ||
48c96a36 JK |
514 | static ssize_t |
515 | read_page_owner(struct file *file, char __user *buf, size_t count, loff_t *ppos) | |
516 | { | |
517 | unsigned long pfn; | |
518 | struct page *page; | |
519 | struct page_ext *page_ext; | |
9300d8df | 520 | struct page_owner *page_owner; |
f2ca0b55 | 521 | depot_stack_handle_t handle; |
48c96a36 | 522 | |
7dd80b8a | 523 | if (!static_branch_unlikely(&page_owner_inited)) |
48c96a36 JK |
524 | return -EINVAL; |
525 | ||
526 | page = NULL; | |
8f0efa81 KL |
527 | if (*ppos == 0) |
528 | pfn = min_low_pfn; | |
529 | else | |
530 | pfn = *ppos; | |
48c96a36 JK |
531 | /* Find a valid PFN or the start of a MAX_ORDER_NR_PAGES area */ |
532 | while (!pfn_valid(pfn) && (pfn & (MAX_ORDER_NR_PAGES - 1)) != 0) | |
533 | pfn++; | |
534 | ||
48c96a36 JK |
535 | /* Find an allocated page */ |
536 | for (; pfn < max_pfn; pfn++) { | |
b1d5488a CTK |
537 | /* |
538 | * This temporary page_owner is required so | |
539 | * that we can avoid the context switches while holding | |
540 | * the rcu lock and copying the page owner information to | |
541 | * user through copy_to_user() or GFP_KERNEL allocations. | |
542 | */ | |
543 | struct page_owner page_owner_tmp; | |
544 | ||
48c96a36 JK |
545 | /* |
546 | * If the new page is in a new MAX_ORDER_NR_PAGES area, | |
547 | * validate the area as existing, skip it if not | |
548 | */ | |
549 | if ((pfn & (MAX_ORDER_NR_PAGES - 1)) == 0 && !pfn_valid(pfn)) { | |
550 | pfn += MAX_ORDER_NR_PAGES - 1; | |
551 | continue; | |
552 | } | |
553 | ||
48c96a36 JK |
554 | page = pfn_to_page(pfn); |
555 | if (PageBuddy(page)) { | |
ab130f91 | 556 | unsigned long freepage_order = buddy_order_unsafe(page); |
48c96a36 | 557 | |
5e0a760b | 558 | if (freepage_order <= MAX_PAGE_ORDER) |
48c96a36 JK |
559 | pfn += (1UL << freepage_order) - 1; |
560 | continue; | |
561 | } | |
562 | ||
b1d5488a | 563 | page_ext = page_ext_get(page); |
f86e4271 YS |
564 | if (unlikely(!page_ext)) |
565 | continue; | |
48c96a36 JK |
566 | |
567 | /* | |
61cf5feb JK |
568 | * Some pages could be missed by concurrent allocation or free, |
569 | * because we don't hold the zone lock. | |
48c96a36 JK |
570 | */ |
571 | if (!test_bit(PAGE_EXT_OWNER, &page_ext->flags)) | |
b1d5488a | 572 | goto ext_put_continue; |
48c96a36 | 573 | |
37389167 VB |
574 | /* |
575 | * Although we do have the info about past allocation of free | |
576 | * pages, it's not relevant for current memory usage. | |
577 | */ | |
fdf3bf80 | 578 | if (!test_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags)) |
b1d5488a | 579 | goto ext_put_continue; |
37389167 | 580 | |
9300d8df JK |
581 | page_owner = get_page_owner(page_ext); |
582 | ||
7e2f2a0c VB |
583 | /* |
584 | * Don't print "tail" pages of high-order allocations as that | |
585 | * would inflate the stats. | |
586 | */ | |
587 | if (!IS_ALIGNED(pfn, 1 << page_owner->order)) | |
b1d5488a | 588 | goto ext_put_continue; |
7e2f2a0c | 589 | |
f2ca0b55 JK |
590 | /* |
591 | * Access to page_ext->handle isn't synchronous so we should | |
592 | * be careful to access it. | |
593 | */ | |
9300d8df | 594 | handle = READ_ONCE(page_owner->handle); |
f2ca0b55 | 595 | if (!handle) |
b1d5488a | 596 | goto ext_put_continue; |
f2ca0b55 | 597 | |
48c96a36 | 598 | /* Record the next PFN to read in the file offset */ |
8f0efa81 | 599 | *ppos = pfn + 1; |
48c96a36 | 600 | |
b1d5488a CTK |
601 | page_owner_tmp = *page_owner; |
602 | page_ext_put(page_ext); | |
f2ca0b55 | 603 | return print_page_owner(buf, count, pfn, page, |
b1d5488a CTK |
604 | &page_owner_tmp, handle); |
605 | ext_put_continue: | |
606 | page_ext_put(page_ext); | |
48c96a36 JK |
607 | } |
608 | ||
609 | return 0; | |
610 | } | |
611 | ||
8f0efa81 KL |
612 | static loff_t lseek_page_owner(struct file *file, loff_t offset, int orig) |
613 | { | |
614 | switch (orig) { | |
615 | case SEEK_SET: | |
616 | file->f_pos = offset; | |
617 | break; | |
618 | case SEEK_CUR: | |
619 | file->f_pos += offset; | |
620 | break; | |
621 | default: | |
622 | return -EINVAL; | |
623 | } | |
624 | return file->f_pos; | |
625 | } | |
626 | ||
61cf5feb JK |
627 | static void init_pages_in_zone(pg_data_t *pgdat, struct zone *zone) |
628 | { | |
6787c1da OS |
629 | unsigned long pfn = zone->zone_start_pfn; |
630 | unsigned long end_pfn = zone_end_pfn(zone); | |
61cf5feb JK |
631 | unsigned long count = 0; |
632 | ||
61cf5feb JK |
633 | /* |
634 | * Walk the zone in pageblock_nr_pages steps. If a page block spans | |
635 | * a zone boundary, it will be double counted between zones. This does | |
636 | * not matter as the mixed block count will still be correct | |
637 | */ | |
638 | for (; pfn < end_pfn; ) { | |
6787c1da OS |
639 | unsigned long block_end_pfn; |
640 | ||
61cf5feb JK |
641 | if (!pfn_valid(pfn)) { |
642 | pfn = ALIGN(pfn + 1, MAX_ORDER_NR_PAGES); | |
643 | continue; | |
644 | } | |
645 | ||
4f9bc69a | 646 | block_end_pfn = pageblock_end_pfn(pfn); |
61cf5feb JK |
647 | block_end_pfn = min(block_end_pfn, end_pfn); |
648 | ||
61cf5feb | 649 | for (; pfn < block_end_pfn; pfn++) { |
859a85dd | 650 | struct page *page = pfn_to_page(pfn); |
6787c1da OS |
651 | struct page_ext *page_ext; |
652 | ||
9d43f5ae JK |
653 | if (page_zone(page) != zone) |
654 | continue; | |
655 | ||
61cf5feb | 656 | /* |
10903027 VB |
657 | * To avoid having to grab zone->lock, be a little |
658 | * careful when reading buddy page order. The only | |
659 | * danger is that we skip too much and potentially miss | |
660 | * some early allocated pages, which is better than | |
661 | * heavy lock contention. | |
61cf5feb JK |
662 | */ |
663 | if (PageBuddy(page)) { | |
ab130f91 | 664 | unsigned long order = buddy_order_unsafe(page); |
10903027 | 665 | |
5e0a760b | 666 | if (order > 0 && order <= MAX_PAGE_ORDER) |
10903027 | 667 | pfn += (1UL << order) - 1; |
61cf5feb JK |
668 | continue; |
669 | } | |
670 | ||
671 | if (PageReserved(page)) | |
672 | continue; | |
673 | ||
b1d5488a | 674 | page_ext = page_ext_get(page); |
f86e4271 YS |
675 | if (unlikely(!page_ext)) |
676 | continue; | |
61cf5feb | 677 | |
dab4ead1 | 678 | /* Maybe overlapping zone */ |
61cf5feb | 679 | if (test_bit(PAGE_EXT_OWNER, &page_ext->flags)) |
b1d5488a | 680 | goto ext_put_continue; |
61cf5feb JK |
681 | |
682 | /* Found early allocated page */ | |
64ea78d2 | 683 | __set_page_owner_handle(page_ext, early_handle, |
7e2f2a0c | 684 | 0, 0); |
61cf5feb | 685 | count++; |
b1d5488a CTK |
686 | ext_put_continue: |
687 | page_ext_put(page_ext); | |
61cf5feb | 688 | } |
10903027 | 689 | cond_resched(); |
61cf5feb JK |
690 | } |
691 | ||
692 | pr_info("Node %d, zone %8s: page owner found early allocated %lu pages\n", | |
693 | pgdat->node_id, zone->name, count); | |
694 | } | |
695 | ||
696 | static void init_zones_in_node(pg_data_t *pgdat) | |
697 | { | |
698 | struct zone *zone; | |
699 | struct zone *node_zones = pgdat->node_zones; | |
61cf5feb JK |
700 | |
701 | for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) { | |
702 | if (!populated_zone(zone)) | |
703 | continue; | |
704 | ||
61cf5feb | 705 | init_pages_in_zone(pgdat, zone); |
61cf5feb JK |
706 | } |
707 | } | |
708 | ||
709 | static void init_early_allocated_pages(void) | |
710 | { | |
711 | pg_data_t *pgdat; | |
712 | ||
61cf5feb JK |
713 | for_each_online_pgdat(pgdat) |
714 | init_zones_in_node(pgdat); | |
715 | } | |
716 | ||
48c96a36 JK |
717 | static const struct file_operations proc_page_owner_operations = { |
718 | .read = read_page_owner, | |
8f0efa81 | 719 | .llseek = lseek_page_owner, |
48c96a36 JK |
720 | }; |
721 | ||
722 | static int __init pageowner_init(void) | |
723 | { | |
7dd80b8a | 724 | if (!static_branch_unlikely(&page_owner_inited)) { |
48c96a36 JK |
725 | pr_info("page_owner is disabled\n"); |
726 | return 0; | |
727 | } | |
728 | ||
d9f7979c GKH |
729 | debugfs_create_file("page_owner", 0400, NULL, NULL, |
730 | &proc_page_owner_operations); | |
48c96a36 | 731 | |
d9f7979c | 732 | return 0; |
48c96a36 | 733 | } |
44c5af96 | 734 | late_initcall(pageowner_init) |