1 // SPDX-License-Identifier: GPL-2.0 1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/debugfs.h> 2 #include <linux/debugfs.h>
3 #include <linux/mm.h> 3 #include <linux/mm.h>
4 #include <linux/slab.h> 4 #include <linux/slab.h>
5 #include <linux/uaccess.h> 5 #include <linux/uaccess.h>
6 #include <linux/memblock.h> 6 #include <linux/memblock.h>
7 #include <linux/stacktrace.h> 7 #include <linux/stacktrace.h>
8 #include <linux/page_owner.h> 8 #include <linux/page_owner.h>
9 #include <linux/jump_label.h> 9 #include <linux/jump_label.h>
10 #include <linux/migrate.h> 10 #include <linux/migrate.h>
11 #include <linux/stackdepot.h> 11 #include <linux/stackdepot.h>
12 #include <linux/seq_file.h> 12 #include <linux/seq_file.h>
13 #include <linux/memcontrol.h> 13 #include <linux/memcontrol.h>
14 #include <linux/sched/clock.h> 14 #include <linux/sched/clock.h>
15 15
16 #include "internal.h" 16 #include "internal.h"
17 17
18 /* 18 /*
19 * TODO: teach PAGE_OWNER_STACK_DEPTH (__dump_ 19 * TODO: teach PAGE_OWNER_STACK_DEPTH (__dump_page_owner and save_stack)
20 * to use off stack temporal storage 20 * to use off stack temporal storage
21 */ 21 */
22 #define PAGE_OWNER_STACK_DEPTH (16) 22 #define PAGE_OWNER_STACK_DEPTH (16)
23 23
24 struct page_owner { 24 struct page_owner {
25 unsigned short order; 25 unsigned short order;
26 short last_migrate_reason; 26 short last_migrate_reason;
27 gfp_t gfp_mask; 27 gfp_t gfp_mask;
28 depot_stack_handle_t handle; 28 depot_stack_handle_t handle;
29 depot_stack_handle_t free_handle; 29 depot_stack_handle_t free_handle;
30 u64 ts_nsec; 30 u64 ts_nsec;
31 u64 free_ts_nsec; 31 u64 free_ts_nsec;
32 char comm[TASK_COMM_LEN]; 32 char comm[TASK_COMM_LEN];
33 pid_t pid; 33 pid_t pid;
34 pid_t tgid; 34 pid_t tgid;
35 pid_t free_pid; 35 pid_t free_pid;
36 pid_t free_tgid; 36 pid_t free_tgid;
37 }; 37 };
38 38
39 struct stack { 39 struct stack {
40 struct stack_record *stack_record; 40 struct stack_record *stack_record;
41 struct stack *next; 41 struct stack *next;
42 }; 42 };
43 static struct stack dummy_stack; 43 static struct stack dummy_stack;
44 static struct stack failure_stack; 44 static struct stack failure_stack;
45 static struct stack *stack_list; 45 static struct stack *stack_list;
46 static DEFINE_SPINLOCK(stack_list_lock); 46 static DEFINE_SPINLOCK(stack_list_lock);
47 47
48 static bool page_owner_enabled __initdata; 48 static bool page_owner_enabled __initdata;
49 DEFINE_STATIC_KEY_FALSE(page_owner_inited); 49 DEFINE_STATIC_KEY_FALSE(page_owner_inited);
50 50
51 static depot_stack_handle_t dummy_handle; 51 static depot_stack_handle_t dummy_handle;
52 static depot_stack_handle_t failure_handle; 52 static depot_stack_handle_t failure_handle;
53 static depot_stack_handle_t early_handle; 53 static depot_stack_handle_t early_handle;
54 54
55 static void init_early_allocated_pages(void); 55 static void init_early_allocated_pages(void);
56 56
57 static inline void set_current_in_page_owner(v 57 static inline void set_current_in_page_owner(void)
58 { 58 {
59 /* 59 /*
60 * Avoid recursion. 60 * Avoid recursion.
61 * 61 *
62 * We might need to allocate more memo 62 * We might need to allocate more memory from page_owner code, so make
63 * sure to signal it in order to avoid 63 * sure to signal it in order to avoid recursion.
64 */ 64 */
65 current->in_page_owner = 1; 65 current->in_page_owner = 1;
66 } 66 }
67 67
68 static inline void unset_current_in_page_owner 68 static inline void unset_current_in_page_owner(void)
69 { 69 {
70 current->in_page_owner = 0; 70 current->in_page_owner = 0;
71 } 71 }
72 72
73 static int __init early_page_owner_param(char 73 static int __init early_page_owner_param(char *buf)
74 { 74 {
75 int ret = kstrtobool(buf, &page_owner_ 75 int ret = kstrtobool(buf, &page_owner_enabled);
76 76
77 if (page_owner_enabled) 77 if (page_owner_enabled)
78 stack_depot_request_early_init 78 stack_depot_request_early_init();
79 79
80 return ret; 80 return ret;
81 } 81 }
82 early_param("page_owner", early_page_owner_par 82 early_param("page_owner", early_page_owner_param);
83 83
84 static __init bool need_page_owner(void) 84 static __init bool need_page_owner(void)
85 { 85 {
86 return page_owner_enabled; 86 return page_owner_enabled;
87 } 87 }
88 88
89 static __always_inline depot_stack_handle_t cr 89 static __always_inline depot_stack_handle_t create_dummy_stack(void)
90 { 90 {
91 unsigned long entries[4]; 91 unsigned long entries[4];
92 unsigned int nr_entries; 92 unsigned int nr_entries;
93 93
94 nr_entries = stack_trace_save(entries, 94 nr_entries = stack_trace_save(entries, ARRAY_SIZE(entries), 0);
95 return stack_depot_save(entries, nr_en 95 return stack_depot_save(entries, nr_entries, GFP_KERNEL);
96 } 96 }
97 97
98 static noinline void register_dummy_stack(void 98 static noinline void register_dummy_stack(void)
99 { 99 {
100 dummy_handle = create_dummy_stack(); 100 dummy_handle = create_dummy_stack();
101 } 101 }
102 102
103 static noinline void register_failure_stack(vo 103 static noinline void register_failure_stack(void)
104 { 104 {
105 failure_handle = create_dummy_stack(); 105 failure_handle = create_dummy_stack();
106 } 106 }
107 107
108 static noinline void register_early_stack(void 108 static noinline void register_early_stack(void)
109 { 109 {
110 early_handle = create_dummy_stack(); 110 early_handle = create_dummy_stack();
111 } 111 }
112 112
113 static __init void init_page_owner(void) 113 static __init void init_page_owner(void)
114 { 114 {
115 if (!page_owner_enabled) 115 if (!page_owner_enabled)
116 return; 116 return;
117 117
118 register_dummy_stack(); 118 register_dummy_stack();
119 register_failure_stack(); 119 register_failure_stack();
120 register_early_stack(); 120 register_early_stack();
121 init_early_allocated_pages(); 121 init_early_allocated_pages();
122 /* Initialize dummy and failure stacks 122 /* Initialize dummy and failure stacks and link them to stack_list */
123 dummy_stack.stack_record = __stack_dep 123 dummy_stack.stack_record = __stack_depot_get_stack_record(dummy_handle);
124 failure_stack.stack_record = __stack_d 124 failure_stack.stack_record = __stack_depot_get_stack_record(failure_handle);
125 if (dummy_stack.stack_record) 125 if (dummy_stack.stack_record)
126 refcount_set(&dummy_stack.stac 126 refcount_set(&dummy_stack.stack_record->count, 1);
127 if (failure_stack.stack_record) 127 if (failure_stack.stack_record)
128 refcount_set(&failure_stack.st 128 refcount_set(&failure_stack.stack_record->count, 1);
129 dummy_stack.next = &failure_stack; 129 dummy_stack.next = &failure_stack;
130 stack_list = &dummy_stack; 130 stack_list = &dummy_stack;
131 static_branch_enable(&page_owner_inite 131 static_branch_enable(&page_owner_inited);
132 } 132 }
133 133
134 struct page_ext_operations page_owner_ops = { 134 struct page_ext_operations page_owner_ops = {
135 .size = sizeof(struct page_owner), 135 .size = sizeof(struct page_owner),
136 .need = need_page_owner, 136 .need = need_page_owner,
137 .init = init_page_owner, 137 .init = init_page_owner,
138 .need_shared_flags = true, 138 .need_shared_flags = true,
139 }; 139 };
140 140
141 static inline struct page_owner *get_page_owne 141 static inline struct page_owner *get_page_owner(struct page_ext *page_ext)
142 { 142 {
143 return page_ext_data(page_ext, &page_o 143 return page_ext_data(page_ext, &page_owner_ops);
144 } 144 }
145 145
146 static noinline depot_stack_handle_t save_stac 146 static noinline depot_stack_handle_t save_stack(gfp_t flags)
147 { 147 {
148 unsigned long entries[PAGE_OWNER_STACK 148 unsigned long entries[PAGE_OWNER_STACK_DEPTH];
149 depot_stack_handle_t handle; 149 depot_stack_handle_t handle;
150 unsigned int nr_entries; 150 unsigned int nr_entries;
151 151
152 if (current->in_page_owner) 152 if (current->in_page_owner)
153 return dummy_handle; 153 return dummy_handle;
154 154
155 set_current_in_page_owner(); 155 set_current_in_page_owner();
156 nr_entries = stack_trace_save(entries, 156 nr_entries = stack_trace_save(entries, ARRAY_SIZE(entries), 2);
157 handle = stack_depot_save(entries, nr_ 157 handle = stack_depot_save(entries, nr_entries, flags);
158 if (!handle) 158 if (!handle)
159 handle = failure_handle; 159 handle = failure_handle;
160 unset_current_in_page_owner(); 160 unset_current_in_page_owner();
161 161
162 return handle; 162 return handle;
163 } 163 }
164 164
165 static void add_stack_record_to_list(struct st 165 static void add_stack_record_to_list(struct stack_record *stack_record,
166 gfp_t gfp 166 gfp_t gfp_mask)
167 { 167 {
168 unsigned long flags; 168 unsigned long flags;
169 struct stack *stack; 169 struct stack *stack;
170 170
>> 171 /* Filter gfp_mask the same way stackdepot does, for consistency */
>> 172 gfp_mask &= ~GFP_ZONEMASK;
>> 173 gfp_mask &= (GFP_ATOMIC | GFP_KERNEL | __GFP_NOLOCKDEP);
>> 174 gfp_mask |= __GFP_NOWARN;
>> 175
171 set_current_in_page_owner(); 176 set_current_in_page_owner();
172 stack = kmalloc(sizeof(*stack), gfp_ne !! 177 stack = kmalloc(sizeof(*stack), gfp_mask);
173 if (!stack) { 178 if (!stack) {
174 unset_current_in_page_owner(); 179 unset_current_in_page_owner();
175 return; 180 return;
176 } 181 }
177 unset_current_in_page_owner(); 182 unset_current_in_page_owner();
178 183
179 stack->stack_record = stack_record; 184 stack->stack_record = stack_record;
180 stack->next = NULL; 185 stack->next = NULL;
181 186
182 spin_lock_irqsave(&stack_list_lock, fl 187 spin_lock_irqsave(&stack_list_lock, flags);
183 stack->next = stack_list; 188 stack->next = stack_list;
184 /* 189 /*
185 * This pairs with smp_load_acquire() 190 * This pairs with smp_load_acquire() from function
186 * stack_start(). This guarantees that 191 * stack_start(). This guarantees that stack_start()
187 * will see an updated stack_list befo 192 * will see an updated stack_list before starting to
188 * traverse the list. 193 * traverse the list.
189 */ 194 */
190 smp_store_release(&stack_list, stack); 195 smp_store_release(&stack_list, stack);
191 spin_unlock_irqrestore(&stack_list_loc 196 spin_unlock_irqrestore(&stack_list_lock, flags);
192 } 197 }
193 198
194 static void inc_stack_record_count(depot_stack 199 static void inc_stack_record_count(depot_stack_handle_t handle, gfp_t gfp_mask,
195 int nr_base 200 int nr_base_pages)
196 { 201 {
197 struct stack_record *stack_record = __ 202 struct stack_record *stack_record = __stack_depot_get_stack_record(handle);
198 203
199 if (!stack_record) 204 if (!stack_record)
200 return; 205 return;
201 206
202 /* 207 /*
203 * New stack_record's that do not use 208 * New stack_record's that do not use STACK_DEPOT_FLAG_GET start
204 * with REFCOUNT_SATURATED to catch sp 209 * with REFCOUNT_SATURATED to catch spurious increments of their
205 * refcount. 210 * refcount.
206 * Since we do not use STACK_DEPOT_FLA 211 * Since we do not use STACK_DEPOT_FLAG_GET API, let us
207 * set a refcount of 1 ourselves. 212 * set a refcount of 1 ourselves.
208 */ 213 */
209 if (refcount_read(&stack_record->count 214 if (refcount_read(&stack_record->count) == REFCOUNT_SATURATED) {
210 int old = REFCOUNT_SATURATED; 215 int old = REFCOUNT_SATURATED;
211 216
212 if (atomic_try_cmpxchg_relaxed 217 if (atomic_try_cmpxchg_relaxed(&stack_record->count.refs, &old, 1))
213 /* Add the new stack_r 218 /* Add the new stack_record to our list */
214 add_stack_record_to_li 219 add_stack_record_to_list(stack_record, gfp_mask);
215 } 220 }
216 refcount_add(nr_base_pages, &stack_rec 221 refcount_add(nr_base_pages, &stack_record->count);
217 } 222 }
218 223
219 static void dec_stack_record_count(depot_stack 224 static void dec_stack_record_count(depot_stack_handle_t handle,
220 int nr_base 225 int nr_base_pages)
221 { 226 {
222 struct stack_record *stack_record = __ 227 struct stack_record *stack_record = __stack_depot_get_stack_record(handle);
223 228
224 if (!stack_record) 229 if (!stack_record)
225 return; 230 return;
226 231
227 if (refcount_sub_and_test(nr_base_page 232 if (refcount_sub_and_test(nr_base_pages, &stack_record->count))
228 pr_warn("%s: refcount went to 233 pr_warn("%s: refcount went to 0 for %u handle\n", __func__,
229 handle); 234 handle);
230 } 235 }
231 236
232 static inline void __update_page_owner_handle( 237 static inline void __update_page_owner_handle(struct page_ext *page_ext,
233 238 depot_stack_handle_t handle,
234 239 unsigned short order,
235 240 gfp_t gfp_mask,
236 241 short last_migrate_reason, u64 ts_nsec,
237 242 pid_t pid, pid_t tgid, char *comm)
238 { 243 {
239 int i; 244 int i;
240 struct page_owner *page_owner; 245 struct page_owner *page_owner;
241 246
242 for (i = 0; i < (1 << order); i++) { 247 for (i = 0; i < (1 << order); i++) {
243 page_owner = get_page_owner(pa 248 page_owner = get_page_owner(page_ext);
244 page_owner->handle = handle; 249 page_owner->handle = handle;
245 page_owner->order = order; 250 page_owner->order = order;
246 page_owner->gfp_mask = gfp_mas 251 page_owner->gfp_mask = gfp_mask;
247 page_owner->last_migrate_reaso 252 page_owner->last_migrate_reason = last_migrate_reason;
248 page_owner->pid = pid; 253 page_owner->pid = pid;
249 page_owner->tgid = tgid; 254 page_owner->tgid = tgid;
250 page_owner->ts_nsec = ts_nsec; 255 page_owner->ts_nsec = ts_nsec;
251 strscpy(page_owner->comm, comm 256 strscpy(page_owner->comm, comm,
252 sizeof(page_owner->com 257 sizeof(page_owner->comm));
253 __set_bit(PAGE_EXT_OWNER, &pag 258 __set_bit(PAGE_EXT_OWNER, &page_ext->flags);
254 __set_bit(PAGE_EXT_OWNER_ALLOC 259 __set_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags);
255 page_ext = page_ext_next(page_ 260 page_ext = page_ext_next(page_ext);
256 } 261 }
257 } 262 }
258 263
259 static inline void __update_page_owner_free_ha 264 static inline void __update_page_owner_free_handle(struct page_ext *page_ext,
260 265 depot_stack_handle_t handle,
261 266 unsigned short order,
262 267 pid_t pid, pid_t tgid,
263 268 u64 free_ts_nsec)
264 { 269 {
265 int i; 270 int i;
266 struct page_owner *page_owner; 271 struct page_owner *page_owner;
267 272
268 for (i = 0; i < (1 << order); i++) { 273 for (i = 0; i < (1 << order); i++) {
269 page_owner = get_page_owner(pa 274 page_owner = get_page_owner(page_ext);
270 /* Only __reset_page_owner() w 275 /* Only __reset_page_owner() wants to clear the bit */
271 if (handle) { 276 if (handle) {
272 __clear_bit(PAGE_EXT_O 277 __clear_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags);
273 page_owner->free_handl 278 page_owner->free_handle = handle;
274 } 279 }
275 page_owner->free_ts_nsec = fre 280 page_owner->free_ts_nsec = free_ts_nsec;
276 page_owner->free_pid = current 281 page_owner->free_pid = current->pid;
277 page_owner->free_tgid = curren 282 page_owner->free_tgid = current->tgid;
278 page_ext = page_ext_next(page_ 283 page_ext = page_ext_next(page_ext);
279 } 284 }
280 } 285 }
281 286
282 void __reset_page_owner(struct page *page, uns 287 void __reset_page_owner(struct page *page, unsigned short order)
283 { 288 {
284 struct page_ext *page_ext; 289 struct page_ext *page_ext;
285 depot_stack_handle_t handle; 290 depot_stack_handle_t handle;
286 depot_stack_handle_t alloc_handle; 291 depot_stack_handle_t alloc_handle;
287 struct page_owner *page_owner; 292 struct page_owner *page_owner;
288 u64 free_ts_nsec = local_clock(); 293 u64 free_ts_nsec = local_clock();
289 294
290 page_ext = page_ext_get(page); 295 page_ext = page_ext_get(page);
291 if (unlikely(!page_ext)) 296 if (unlikely(!page_ext))
292 return; 297 return;
293 298
294 page_owner = get_page_owner(page_ext); 299 page_owner = get_page_owner(page_ext);
295 alloc_handle = page_owner->handle; 300 alloc_handle = page_owner->handle;
296 301
297 handle = save_stack(GFP_NOWAIT | __GFP 302 handle = save_stack(GFP_NOWAIT | __GFP_NOWARN);
298 __update_page_owner_free_handle(page_e 303 __update_page_owner_free_handle(page_ext, handle, order, current->pid,
299 curren 304 current->tgid, free_ts_nsec);
300 page_ext_put(page_ext); 305 page_ext_put(page_ext);
301 306
302 if (alloc_handle != early_handle) 307 if (alloc_handle != early_handle)
303 /* 308 /*
304 * early_handle is being set a 309 * early_handle is being set as a handle for all those
305 * early allocated pages. See 310 * early allocated pages. See init_pages_in_zone().
306 * Since their refcount is not 311 * Since their refcount is not being incremented because
307 * the machinery is not ready 312 * the machinery is not ready yet, we cannot decrement
308 * their refcount either. 313 * their refcount either.
309 */ 314 */
310 dec_stack_record_count(alloc_h 315 dec_stack_record_count(alloc_handle, 1 << order);
311 } 316 }
312 317
313 noinline void __set_page_owner(struct page *pa 318 noinline void __set_page_owner(struct page *page, unsigned short order,
314 gfp_t 319 gfp_t gfp_mask)
315 { 320 {
316 struct page_ext *page_ext; 321 struct page_ext *page_ext;
317 u64 ts_nsec = local_clock(); 322 u64 ts_nsec = local_clock();
318 depot_stack_handle_t handle; 323 depot_stack_handle_t handle;
319 324
320 handle = save_stack(gfp_mask); 325 handle = save_stack(gfp_mask);
321 326
322 page_ext = page_ext_get(page); 327 page_ext = page_ext_get(page);
323 if (unlikely(!page_ext)) 328 if (unlikely(!page_ext))
324 return; 329 return;
325 __update_page_owner_handle(page_ext, h 330 __update_page_owner_handle(page_ext, handle, order, gfp_mask, -1,
326 ts_nsec, cu 331 ts_nsec, current->pid, current->tgid,
327 current->co 332 current->comm);
328 page_ext_put(page_ext); 333 page_ext_put(page_ext);
329 inc_stack_record_count(handle, gfp_mas 334 inc_stack_record_count(handle, gfp_mask, 1 << order);
330 } 335 }
331 336
332 void __set_page_owner_migrate_reason(struct pa 337 void __set_page_owner_migrate_reason(struct page *page, int reason)
333 { 338 {
334 struct page_ext *page_ext = page_ext_g 339 struct page_ext *page_ext = page_ext_get(page);
335 struct page_owner *page_owner; 340 struct page_owner *page_owner;
336 341
337 if (unlikely(!page_ext)) 342 if (unlikely(!page_ext))
338 return; 343 return;
339 344
340 page_owner = get_page_owner(page_ext); 345 page_owner = get_page_owner(page_ext);
341 page_owner->last_migrate_reason = reas 346 page_owner->last_migrate_reason = reason;
342 page_ext_put(page_ext); 347 page_ext_put(page_ext);
343 } 348 }
344 349
345 void __split_page_owner(struct page *page, int 350 void __split_page_owner(struct page *page, int old_order, int new_order)
346 { 351 {
347 int i; 352 int i;
348 struct page_ext *page_ext = page_ext_g 353 struct page_ext *page_ext = page_ext_get(page);
349 struct page_owner *page_owner; 354 struct page_owner *page_owner;
350 355
351 if (unlikely(!page_ext)) 356 if (unlikely(!page_ext))
352 return; 357 return;
353 358
354 for (i = 0; i < (1 << old_order); i++) 359 for (i = 0; i < (1 << old_order); i++) {
355 page_owner = get_page_owner(pa 360 page_owner = get_page_owner(page_ext);
356 page_owner->order = new_order; 361 page_owner->order = new_order;
357 page_ext = page_ext_next(page_ 362 page_ext = page_ext_next(page_ext);
358 } 363 }
359 page_ext_put(page_ext); 364 page_ext_put(page_ext);
360 } 365 }
361 366
362 void __folio_copy_owner(struct folio *newfolio 367 void __folio_copy_owner(struct folio *newfolio, struct folio *old)
363 { 368 {
364 int i; 369 int i;
365 struct page_ext *old_ext; 370 struct page_ext *old_ext;
366 struct page_ext *new_ext; 371 struct page_ext *new_ext;
367 struct page_owner *old_page_owner; 372 struct page_owner *old_page_owner;
368 struct page_owner *new_page_owner; 373 struct page_owner *new_page_owner;
369 depot_stack_handle_t migrate_handle; 374 depot_stack_handle_t migrate_handle;
370 375
371 old_ext = page_ext_get(&old->page); 376 old_ext = page_ext_get(&old->page);
372 if (unlikely(!old_ext)) 377 if (unlikely(!old_ext))
373 return; 378 return;
374 379
375 new_ext = page_ext_get(&newfolio->page 380 new_ext = page_ext_get(&newfolio->page);
376 if (unlikely(!new_ext)) { 381 if (unlikely(!new_ext)) {
377 page_ext_put(old_ext); 382 page_ext_put(old_ext);
378 return; 383 return;
379 } 384 }
380 385
381 old_page_owner = get_page_owner(old_ex 386 old_page_owner = get_page_owner(old_ext);
382 new_page_owner = get_page_owner(new_ex 387 new_page_owner = get_page_owner(new_ext);
383 migrate_handle = new_page_owner->handl 388 migrate_handle = new_page_owner->handle;
384 __update_page_owner_handle(new_ext, ol 389 __update_page_owner_handle(new_ext, old_page_owner->handle,
385 old_page_ow 390 old_page_owner->order, old_page_owner->gfp_mask,
386 old_page_ow 391 old_page_owner->last_migrate_reason,
387 old_page_ow 392 old_page_owner->ts_nsec, old_page_owner->pid,
388 old_page_ow 393 old_page_owner->tgid, old_page_owner->comm);
389 /* 394 /*
390 * Do not proactively clear PAGE_EXT_O 395 * Do not proactively clear PAGE_EXT_OWNER{_ALLOCATED} bits as the folio
391 * will be freed after migration. Keep 396 * will be freed after migration. Keep them until then as they may be
392 * useful. 397 * useful.
393 */ 398 */
394 __update_page_owner_free_handle(new_ex 399 __update_page_owner_free_handle(new_ext, 0, old_page_owner->order,
395 old_pa 400 old_page_owner->free_pid,
396 old_pa 401 old_page_owner->free_tgid,
397 old_pa 402 old_page_owner->free_ts_nsec);
398 /* 403 /*
399 * We linked the original stack to the 404 * We linked the original stack to the new folio, we need to do the same
400 * for the new one and the old folio o 405 * for the new one and the old folio otherwise there will be an imbalance
401 * when subtracting those pages from t 406 * when subtracting those pages from the stack.
402 */ 407 */
403 for (i = 0; i < (1 << new_page_owner-> 408 for (i = 0; i < (1 << new_page_owner->order); i++) {
404 old_page_owner->handle = migra 409 old_page_owner->handle = migrate_handle;
405 old_ext = page_ext_next(old_ex 410 old_ext = page_ext_next(old_ext);
406 old_page_owner = get_page_owne 411 old_page_owner = get_page_owner(old_ext);
407 } 412 }
408 413
409 page_ext_put(new_ext); 414 page_ext_put(new_ext);
410 page_ext_put(old_ext); 415 page_ext_put(old_ext);
411 } 416 }
412 417
413 void pagetypeinfo_showmixedcount_print(struct 418 void pagetypeinfo_showmixedcount_print(struct seq_file *m,
414 pg_data 419 pg_data_t *pgdat, struct zone *zone)
415 { 420 {
416 struct page *page; 421 struct page *page;
417 struct page_ext *page_ext; 422 struct page_ext *page_ext;
418 struct page_owner *page_owner; 423 struct page_owner *page_owner;
419 unsigned long pfn, block_end_pfn; 424 unsigned long pfn, block_end_pfn;
420 unsigned long end_pfn = zone_end_pfn(z 425 unsigned long end_pfn = zone_end_pfn(zone);
421 unsigned long count[MIGRATE_TYPES] = { 426 unsigned long count[MIGRATE_TYPES] = { 0, };
422 int pageblock_mt, page_mt; 427 int pageblock_mt, page_mt;
423 int i; 428 int i;
424 429
425 /* Scan block by block. First and last 430 /* Scan block by block. First and last block may be incomplete */
426 pfn = zone->zone_start_pfn; 431 pfn = zone->zone_start_pfn;
427 432
428 /* 433 /*
429 * Walk the zone in pageblock_nr_pages 434 * Walk the zone in pageblock_nr_pages steps. If a page block spans
430 * a zone boundary, it will be double 435 * a zone boundary, it will be double counted between zones. This does
431 * not matter as the mixed block count 436 * not matter as the mixed block count will still be correct
432 */ 437 */
433 for (; pfn < end_pfn; ) { 438 for (; pfn < end_pfn; ) {
434 page = pfn_to_online_page(pfn) 439 page = pfn_to_online_page(pfn);
435 if (!page) { 440 if (!page) {
436 pfn = ALIGN(pfn + 1, M 441 pfn = ALIGN(pfn + 1, MAX_ORDER_NR_PAGES);
437 continue; 442 continue;
438 } 443 }
439 444
440 block_end_pfn = pageblock_end_ 445 block_end_pfn = pageblock_end_pfn(pfn);
441 block_end_pfn = min(block_end_ 446 block_end_pfn = min(block_end_pfn, end_pfn);
442 447
443 pageblock_mt = get_pageblock_m 448 pageblock_mt = get_pageblock_migratetype(page);
444 449
445 for (; pfn < block_end_pfn; pf 450 for (; pfn < block_end_pfn; pfn++) {
446 /* The pageblock is on 451 /* The pageblock is online, no need to recheck. */
447 page = pfn_to_page(pfn 452 page = pfn_to_page(pfn);
448 453
449 if (page_zone(page) != 454 if (page_zone(page) != zone)
450 continue; 455 continue;
451 456
452 if (PageBuddy(page)) { 457 if (PageBuddy(page)) {
453 unsigned long 458 unsigned long freepage_order;
454 459
455 freepage_order 460 freepage_order = buddy_order_unsafe(page);
456 if (freepage_o 461 if (freepage_order <= MAX_PAGE_ORDER)
457 pfn += 462 pfn += (1UL << freepage_order) - 1;
458 continue; 463 continue;
459 } 464 }
460 465
461 if (PageReserved(page) 466 if (PageReserved(page))
462 continue; 467 continue;
463 468
464 page_ext = page_ext_ge 469 page_ext = page_ext_get(page);
465 if (unlikely(!page_ext 470 if (unlikely(!page_ext))
466 continue; 471 continue;
467 472
468 if (!test_bit(PAGE_EXT 473 if (!test_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags))
469 goto ext_put_c 474 goto ext_put_continue;
470 475
471 page_owner = get_page_ 476 page_owner = get_page_owner(page_ext);
472 page_mt = gfp_migratet 477 page_mt = gfp_migratetype(page_owner->gfp_mask);
473 if (pageblock_mt != pa 478 if (pageblock_mt != page_mt) {
474 if (is_migrate 479 if (is_migrate_cma(pageblock_mt))
475 count[ 480 count[MIGRATE_MOVABLE]++;
476 else 481 else
477 count[ 482 count[pageblock_mt]++;
478 483
479 pfn = block_en 484 pfn = block_end_pfn;
480 page_ext_put(p 485 page_ext_put(page_ext);
481 break; 486 break;
482 } 487 }
483 pfn += (1UL << page_ow 488 pfn += (1UL << page_owner->order) - 1;
484 ext_put_continue: 489 ext_put_continue:
485 page_ext_put(page_ext) 490 page_ext_put(page_ext);
486 } 491 }
487 } 492 }
488 493
489 /* Print counts */ 494 /* Print counts */
490 seq_printf(m, "Node %d, zone %8s ", pg 495 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
491 for (i = 0; i < MIGRATE_TYPES; i++) 496 for (i = 0; i < MIGRATE_TYPES; i++)
492 seq_printf(m, "%12lu ", count[ 497 seq_printf(m, "%12lu ", count[i]);
493 seq_putc(m, '\n'); 498 seq_putc(m, '\n');
494 } 499 }
495 500
496 /* 501 /*
497 * Looking for memcg information and print it 502 * Looking for memcg information and print it out
498 */ 503 */
499 static inline int print_page_owner_memcg(char 504 static inline int print_page_owner_memcg(char *kbuf, size_t count, int ret,
500 struc 505 struct page *page)
501 { 506 {
502 #ifdef CONFIG_MEMCG 507 #ifdef CONFIG_MEMCG
503 unsigned long memcg_data; 508 unsigned long memcg_data;
504 struct mem_cgroup *memcg; 509 struct mem_cgroup *memcg;
505 bool online; 510 bool online;
506 char name[80]; 511 char name[80];
507 512
508 rcu_read_lock(); 513 rcu_read_lock();
509 memcg_data = READ_ONCE(page->memcg_dat 514 memcg_data = READ_ONCE(page->memcg_data);
510 if (!memcg_data) 515 if (!memcg_data)
511 goto out_unlock; 516 goto out_unlock;
512 517
513 if (memcg_data & MEMCG_DATA_OBJEXTS) !! 518 if (memcg_data & MEMCG_DATA_OBJCGS)
514 ret += scnprintf(kbuf + ret, c 519 ret += scnprintf(kbuf + ret, count - ret,
515 "Slab cache pa 520 "Slab cache page\n");
516 521
517 memcg = page_memcg_check(page); 522 memcg = page_memcg_check(page);
518 if (!memcg) 523 if (!memcg)
519 goto out_unlock; 524 goto out_unlock;
520 525
521 online = (memcg->css.flags & CSS_ONLIN 526 online = (memcg->css.flags & CSS_ONLINE);
522 cgroup_name(memcg->css.cgroup, name, s 527 cgroup_name(memcg->css.cgroup, name, sizeof(name));
523 ret += scnprintf(kbuf + ret, count - r 528 ret += scnprintf(kbuf + ret, count - ret,
524 "Charged %sto %smemcg 529 "Charged %sto %smemcg %s\n",
525 PageMemcgKmem(page) ? 530 PageMemcgKmem(page) ? "(via objcg) " : "",
526 online ? "" : "offline 531 online ? "" : "offline ",
527 name); 532 name);
528 out_unlock: 533 out_unlock:
529 rcu_read_unlock(); 534 rcu_read_unlock();
530 #endif /* CONFIG_MEMCG */ 535 #endif /* CONFIG_MEMCG */
531 536
532 return ret; 537 return ret;
533 } 538 }
534 539
535 static ssize_t 540 static ssize_t
536 print_page_owner(char __user *buf, size_t coun 541 print_page_owner(char __user *buf, size_t count, unsigned long pfn,
537 struct page *page, struct page 542 struct page *page, struct page_owner *page_owner,
538 depot_stack_handle_t handle) 543 depot_stack_handle_t handle)
539 { 544 {
540 int ret, pageblock_mt, page_mt; 545 int ret, pageblock_mt, page_mt;
541 char *kbuf; 546 char *kbuf;
542 547
543 count = min_t(size_t, count, PAGE_SIZE 548 count = min_t(size_t, count, PAGE_SIZE);
544 kbuf = kmalloc(count, GFP_KERNEL); 549 kbuf = kmalloc(count, GFP_KERNEL);
545 if (!kbuf) 550 if (!kbuf)
546 return -ENOMEM; 551 return -ENOMEM;
547 552
548 ret = scnprintf(kbuf, count, 553 ret = scnprintf(kbuf, count,
549 "Page allocated via or 554 "Page allocated via order %u, mask %#x(%pGg), pid %d, tgid %d (%s), ts %llu ns\n",
550 page_owner->order, pag 555 page_owner->order, page_owner->gfp_mask,
551 &page_owner->gfp_mask, 556 &page_owner->gfp_mask, page_owner->pid,
552 page_owner->tgid, page 557 page_owner->tgid, page_owner->comm,
553 page_owner->ts_nsec); 558 page_owner->ts_nsec);
554 559
555 /* Print information relevant to group 560 /* Print information relevant to grouping pages by mobility */
556 pageblock_mt = get_pageblock_migratety 561 pageblock_mt = get_pageblock_migratetype(page);
557 page_mt = gfp_migratetype(page_owner- 562 page_mt = gfp_migratetype(page_owner->gfp_mask);
558 ret += scnprintf(kbuf + ret, count - r 563 ret += scnprintf(kbuf + ret, count - ret,
559 "PFN 0x%lx type %s Blo 564 "PFN 0x%lx type %s Block %lu type %s Flags %pGp\n",
560 pfn, 565 pfn,
561 migratetype_names[page 566 migratetype_names[page_mt],
562 pfn >> pageblock_order 567 pfn >> pageblock_order,
563 migratetype_names[page 568 migratetype_names[pageblock_mt],
564 &page->flags); 569 &page->flags);
565 570
566 ret += stack_depot_snprint(handle, kbu 571 ret += stack_depot_snprint(handle, kbuf + ret, count - ret, 0);
567 if (ret >= count) 572 if (ret >= count)
568 goto err; 573 goto err;
569 574
570 if (page_owner->last_migrate_reason != 575 if (page_owner->last_migrate_reason != -1) {
571 ret += scnprintf(kbuf + ret, c 576 ret += scnprintf(kbuf + ret, count - ret,
572 "Page has been migrate 577 "Page has been migrated, last migrate reason: %s\n",
573 migrate_reason_names[p 578 migrate_reason_names[page_owner->last_migrate_reason]);
574 } 579 }
575 580
576 ret = print_page_owner_memcg(kbuf, cou 581 ret = print_page_owner_memcg(kbuf, count, ret, page);
577 582
578 ret += snprintf(kbuf + ret, count - re 583 ret += snprintf(kbuf + ret, count - ret, "\n");
579 if (ret >= count) 584 if (ret >= count)
580 goto err; 585 goto err;
581 586
582 if (copy_to_user(buf, kbuf, ret)) 587 if (copy_to_user(buf, kbuf, ret))
583 ret = -EFAULT; 588 ret = -EFAULT;
584 589
585 kfree(kbuf); 590 kfree(kbuf);
586 return ret; 591 return ret;
587 592
588 err: 593 err:
589 kfree(kbuf); 594 kfree(kbuf);
590 return -ENOMEM; 595 return -ENOMEM;
591 } 596 }
592 597
593 void __dump_page_owner(const struct page *page 598 void __dump_page_owner(const struct page *page)
594 { 599 {
595 struct page_ext *page_ext = page_ext_g 600 struct page_ext *page_ext = page_ext_get((void *)page);
596 struct page_owner *page_owner; 601 struct page_owner *page_owner;
597 depot_stack_handle_t handle; 602 depot_stack_handle_t handle;
598 gfp_t gfp_mask; 603 gfp_t gfp_mask;
599 int mt; 604 int mt;
600 605
601 if (unlikely(!page_ext)) { 606 if (unlikely(!page_ext)) {
602 pr_alert("There is not page ex 607 pr_alert("There is not page extension available.\n");
603 return; 608 return;
604 } 609 }
605 610
606 page_owner = get_page_owner(page_ext); 611 page_owner = get_page_owner(page_ext);
607 gfp_mask = page_owner->gfp_mask; 612 gfp_mask = page_owner->gfp_mask;
608 mt = gfp_migratetype(gfp_mask); 613 mt = gfp_migratetype(gfp_mask);
609 614
610 if (!test_bit(PAGE_EXT_OWNER, &page_ex 615 if (!test_bit(PAGE_EXT_OWNER, &page_ext->flags)) {
611 pr_alert("page_owner info is n 616 pr_alert("page_owner info is not present (never set?)\n");
612 page_ext_put(page_ext); 617 page_ext_put(page_ext);
613 return; 618 return;
614 } 619 }
615 620
616 if (test_bit(PAGE_EXT_OWNER_ALLOCATED, 621 if (test_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags))
617 pr_alert("page_owner tracks th 622 pr_alert("page_owner tracks the page as allocated\n");
618 else 623 else
619 pr_alert("page_owner tracks th 624 pr_alert("page_owner tracks the page as freed\n");
620 625
621 pr_alert("page last allocated via orde 626 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",
622 page_owner->order, migratetyp 627 page_owner->order, migratetype_names[mt], gfp_mask, &gfp_mask,
623 page_owner->pid, page_owner-> 628 page_owner->pid, page_owner->tgid, page_owner->comm,
624 page_owner->ts_nsec, page_own 629 page_owner->ts_nsec, page_owner->free_ts_nsec);
625 630
626 handle = READ_ONCE(page_owner->handle) 631 handle = READ_ONCE(page_owner->handle);
627 if (!handle) 632 if (!handle)
628 pr_alert("page_owner allocatio 633 pr_alert("page_owner allocation stack trace missing\n");
629 else 634 else
630 stack_depot_print(handle); 635 stack_depot_print(handle);
631 636
632 handle = READ_ONCE(page_owner->free_ha 637 handle = READ_ONCE(page_owner->free_handle);
633 if (!handle) { 638 if (!handle) {
634 pr_alert("page_owner free stac 639 pr_alert("page_owner free stack trace missing\n");
635 } else { 640 } else {
636 pr_alert("page last free pid % 641 pr_alert("page last free pid %d tgid %d stack trace:\n",
637 page_owner->free_pid 642 page_owner->free_pid, page_owner->free_tgid);
638 stack_depot_print(handle); 643 stack_depot_print(handle);
639 } 644 }
640 645
641 if (page_owner->last_migrate_reason != 646 if (page_owner->last_migrate_reason != -1)
642 pr_alert("page has been migrat 647 pr_alert("page has been migrated, last migrate reason: %s\n",
643 migrate_reason_names[p 648 migrate_reason_names[page_owner->last_migrate_reason]);
644 page_ext_put(page_ext); 649 page_ext_put(page_ext);
645 } 650 }
646 651
647 static ssize_t 652 static ssize_t
648 read_page_owner(struct file *file, char __user 653 read_page_owner(struct file *file, char __user *buf, size_t count, loff_t *ppos)
649 { 654 {
650 unsigned long pfn; 655 unsigned long pfn;
651 struct page *page; 656 struct page *page;
652 struct page_ext *page_ext; 657 struct page_ext *page_ext;
653 struct page_owner *page_owner; 658 struct page_owner *page_owner;
654 depot_stack_handle_t handle; 659 depot_stack_handle_t handle;
655 660
656 if (!static_branch_unlikely(&page_owne 661 if (!static_branch_unlikely(&page_owner_inited))
657 return -EINVAL; 662 return -EINVAL;
658 663
659 page = NULL; 664 page = NULL;
660 if (*ppos == 0) 665 if (*ppos == 0)
661 pfn = min_low_pfn; 666 pfn = min_low_pfn;
662 else 667 else
663 pfn = *ppos; 668 pfn = *ppos;
664 /* Find a valid PFN or the start of a 669 /* Find a valid PFN or the start of a MAX_ORDER_NR_PAGES area */
665 while (!pfn_valid(pfn) && (pfn & (MAX_ 670 while (!pfn_valid(pfn) && (pfn & (MAX_ORDER_NR_PAGES - 1)) != 0)
666 pfn++; 671 pfn++;
667 672
668 /* Find an allocated page */ 673 /* Find an allocated page */
669 for (; pfn < max_pfn; pfn++) { 674 for (; pfn < max_pfn; pfn++) {
670 /* 675 /*
671 * This temporary page_owner i 676 * This temporary page_owner is required so
672 * that we can avoid the conte 677 * that we can avoid the context switches while holding
673 * the rcu lock and copying th 678 * the rcu lock and copying the page owner information to
674 * user through copy_to_user() 679 * user through copy_to_user() or GFP_KERNEL allocations.
675 */ 680 */
676 struct page_owner page_owner_t 681 struct page_owner page_owner_tmp;
677 682
678 /* 683 /*
679 * If the new page is in a new 684 * If the new page is in a new MAX_ORDER_NR_PAGES area,
680 * validate the area as existi 685 * validate the area as existing, skip it if not
681 */ 686 */
682 if ((pfn & (MAX_ORDER_NR_PAGES 687 if ((pfn & (MAX_ORDER_NR_PAGES - 1)) == 0 && !pfn_valid(pfn)) {
683 pfn += MAX_ORDER_NR_PA 688 pfn += MAX_ORDER_NR_PAGES - 1;
684 continue; 689 continue;
685 } 690 }
686 691
687 page = pfn_to_page(pfn); 692 page = pfn_to_page(pfn);
688 if (PageBuddy(page)) { 693 if (PageBuddy(page)) {
689 unsigned long freepage 694 unsigned long freepage_order = buddy_order_unsafe(page);
690 695
691 if (freepage_order <= 696 if (freepage_order <= MAX_PAGE_ORDER)
692 pfn += (1UL << 697 pfn += (1UL << freepage_order) - 1;
693 continue; 698 continue;
694 } 699 }
695 700
696 page_ext = page_ext_get(page); 701 page_ext = page_ext_get(page);
697 if (unlikely(!page_ext)) 702 if (unlikely(!page_ext))
698 continue; 703 continue;
699 704
700 /* 705 /*
701 * Some pages could be missed 706 * Some pages could be missed by concurrent allocation or free,
702 * because we don't hold the z 707 * because we don't hold the zone lock.
703 */ 708 */
704 if (!test_bit(PAGE_EXT_OWNER, 709 if (!test_bit(PAGE_EXT_OWNER, &page_ext->flags))
705 goto ext_put_continue; 710 goto ext_put_continue;
706 711
707 /* 712 /*
708 * Although we do have the inf 713 * Although we do have the info about past allocation of free
709 * pages, it's not relevant fo 714 * pages, it's not relevant for current memory usage.
710 */ 715 */
711 if (!test_bit(PAGE_EXT_OWNER_A 716 if (!test_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags))
712 goto ext_put_continue; 717 goto ext_put_continue;
713 718
714 page_owner = get_page_owner(pa 719 page_owner = get_page_owner(page_ext);
715 720
716 /* 721 /*
717 * Don't print "tail" pages of 722 * Don't print "tail" pages of high-order allocations as that
718 * would inflate the stats. 723 * would inflate the stats.
719 */ 724 */
720 if (!IS_ALIGNED(pfn, 1 << page 725 if (!IS_ALIGNED(pfn, 1 << page_owner->order))
721 goto ext_put_continue; 726 goto ext_put_continue;
722 727
723 /* 728 /*
724 * Access to page_ext->handle 729 * Access to page_ext->handle isn't synchronous so we should
725 * be careful to access it. 730 * be careful to access it.
726 */ 731 */
727 handle = READ_ONCE(page_owner- 732 handle = READ_ONCE(page_owner->handle);
728 if (!handle) 733 if (!handle)
729 goto ext_put_continue; 734 goto ext_put_continue;
730 735
731 /* Record the next PFN to read 736 /* Record the next PFN to read in the file offset */
732 *ppos = pfn + 1; 737 *ppos = pfn + 1;
733 738
734 page_owner_tmp = *page_owner; 739 page_owner_tmp = *page_owner;
735 page_ext_put(page_ext); 740 page_ext_put(page_ext);
736 return print_page_owner(buf, c 741 return print_page_owner(buf, count, pfn, page,
737 &page_owner_tm 742 &page_owner_tmp, handle);
738 ext_put_continue: 743 ext_put_continue:
739 page_ext_put(page_ext); 744 page_ext_put(page_ext);
740 } 745 }
741 746
742 return 0; 747 return 0;
743 } 748 }
744 749
745 static loff_t lseek_page_owner(struct file *fi 750 static loff_t lseek_page_owner(struct file *file, loff_t offset, int orig)
746 { 751 {
747 switch (orig) { 752 switch (orig) {
748 case SEEK_SET: 753 case SEEK_SET:
749 file->f_pos = offset; 754 file->f_pos = offset;
750 break; 755 break;
751 case SEEK_CUR: 756 case SEEK_CUR:
752 file->f_pos += offset; 757 file->f_pos += offset;
753 break; 758 break;
754 default: 759 default:
755 return -EINVAL; 760 return -EINVAL;
756 } 761 }
757 return file->f_pos; 762 return file->f_pos;
758 } 763 }
759 764
760 static void init_pages_in_zone(pg_data_t *pgda 765 static void init_pages_in_zone(pg_data_t *pgdat, struct zone *zone)
761 { 766 {
762 unsigned long pfn = zone->zone_start_p 767 unsigned long pfn = zone->zone_start_pfn;
763 unsigned long end_pfn = zone_end_pfn(z 768 unsigned long end_pfn = zone_end_pfn(zone);
764 unsigned long count = 0; 769 unsigned long count = 0;
765 770
766 /* 771 /*
767 * Walk the zone in pageblock_nr_pages 772 * Walk the zone in pageblock_nr_pages steps. If a page block spans
768 * a zone boundary, it will be double 773 * a zone boundary, it will be double counted between zones. This does
769 * not matter as the mixed block count 774 * not matter as the mixed block count will still be correct
770 */ 775 */
771 for (; pfn < end_pfn; ) { 776 for (; pfn < end_pfn; ) {
772 unsigned long block_end_pfn; 777 unsigned long block_end_pfn;
773 778
774 if (!pfn_valid(pfn)) { 779 if (!pfn_valid(pfn)) {
775 pfn = ALIGN(pfn + 1, M 780 pfn = ALIGN(pfn + 1, MAX_ORDER_NR_PAGES);
776 continue; 781 continue;
777 } 782 }
778 783
779 block_end_pfn = pageblock_end_ 784 block_end_pfn = pageblock_end_pfn(pfn);
780 block_end_pfn = min(block_end_ 785 block_end_pfn = min(block_end_pfn, end_pfn);
781 786
782 for (; pfn < block_end_pfn; pf 787 for (; pfn < block_end_pfn; pfn++) {
783 struct page *page = pf 788 struct page *page = pfn_to_page(pfn);
784 struct page_ext *page_ 789 struct page_ext *page_ext;
785 790
786 if (page_zone(page) != 791 if (page_zone(page) != zone)
787 continue; 792 continue;
788 793
789 /* 794 /*
790 * To avoid having to 795 * To avoid having to grab zone->lock, be a little
791 * careful when readin 796 * careful when reading buddy page order. The only
792 * danger is that we s 797 * danger is that we skip too much and potentially miss
793 * some early allocate 798 * some early allocated pages, which is better than
794 * heavy lock contenti 799 * heavy lock contention.
795 */ 800 */
796 if (PageBuddy(page)) { 801 if (PageBuddy(page)) {
797 unsigned long 802 unsigned long order = buddy_order_unsafe(page);
798 803
799 if (order > 0 804 if (order > 0 && order <= MAX_PAGE_ORDER)
800 pfn += 805 pfn += (1UL << order) - 1;
801 continue; 806 continue;
802 } 807 }
803 808
804 if (PageReserved(page) 809 if (PageReserved(page))
805 continue; 810 continue;
806 811
807 page_ext = page_ext_ge 812 page_ext = page_ext_get(page);
808 if (unlikely(!page_ext 813 if (unlikely(!page_ext))
809 continue; 814 continue;
810 815
811 /* Maybe overlapping z 816 /* Maybe overlapping zone */
812 if (test_bit(PAGE_EXT_ 817 if (test_bit(PAGE_EXT_OWNER, &page_ext->flags))
813 goto ext_put_c 818 goto ext_put_continue;
814 819
815 /* Found early allocat 820 /* Found early allocated page */
816 __update_page_owner_ha 821 __update_page_owner_handle(page_ext, early_handle, 0, 0,
817 822 -1, local_clock(), current->pid,
818 823 current->tgid, current->comm);
819 count++; 824 count++;
820 ext_put_continue: 825 ext_put_continue:
821 page_ext_put(page_ext) 826 page_ext_put(page_ext);
822 } 827 }
823 cond_resched(); 828 cond_resched();
824 } 829 }
825 830
826 pr_info("Node %d, zone %8s: page owner 831 pr_info("Node %d, zone %8s: page owner found early allocated %lu pages\n",
827 pgdat->node_id, zone->name, co 832 pgdat->node_id, zone->name, count);
828 } 833 }
829 834
830 static void init_zones_in_node(pg_data_t *pgda 835 static void init_zones_in_node(pg_data_t *pgdat)
831 { 836 {
832 struct zone *zone; 837 struct zone *zone;
833 struct zone *node_zones = pgdat->node_ 838 struct zone *node_zones = pgdat->node_zones;
834 839
835 for (zone = node_zones; zone - node_zo 840 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
836 if (!populated_zone(zone)) 841 if (!populated_zone(zone))
837 continue; 842 continue;
838 843
839 init_pages_in_zone(pgdat, zone 844 init_pages_in_zone(pgdat, zone);
840 } 845 }
841 } 846 }
842 847
843 static void init_early_allocated_pages(void) 848 static void init_early_allocated_pages(void)
844 { 849 {
845 pg_data_t *pgdat; 850 pg_data_t *pgdat;
846 851
847 for_each_online_pgdat(pgdat) 852 for_each_online_pgdat(pgdat)
848 init_zones_in_node(pgdat); 853 init_zones_in_node(pgdat);
849 } 854 }
850 855
851 static const struct file_operations proc_page_ 856 static const struct file_operations proc_page_owner_operations = {
852 .read = read_page_owner, 857 .read = read_page_owner,
853 .llseek = lseek_page_owner, 858 .llseek = lseek_page_owner,
854 }; 859 };
855 860
856 static void *stack_start(struct seq_file *m, l 861 static void *stack_start(struct seq_file *m, loff_t *ppos)
857 { 862 {
858 struct stack *stack; 863 struct stack *stack;
859 864
860 if (*ppos == -1UL) 865 if (*ppos == -1UL)
861 return NULL; 866 return NULL;
862 867
863 if (!*ppos) { 868 if (!*ppos) {
864 /* 869 /*
865 * This pairs with smp_store_r 870 * This pairs with smp_store_release() from function
866 * add_stack_record_to_list(), 871 * add_stack_record_to_list(), so we get a consistent
867 * value of stack_list. 872 * value of stack_list.
868 */ 873 */
869 stack = smp_load_acquire(&stac 874 stack = smp_load_acquire(&stack_list);
870 m->private = stack; 875 m->private = stack;
871 } else { 876 } else {
872 stack = m->private; 877 stack = m->private;
873 } 878 }
874 879
875 return stack; 880 return stack;
876 } 881 }
877 882
878 static void *stack_next(struct seq_file *m, vo 883 static void *stack_next(struct seq_file *m, void *v, loff_t *ppos)
879 { 884 {
880 struct stack *stack = v; 885 struct stack *stack = v;
881 886
882 stack = stack->next; 887 stack = stack->next;
883 *ppos = stack ? *ppos + 1 : -1UL; 888 *ppos = stack ? *ppos + 1 : -1UL;
884 m->private = stack; 889 m->private = stack;
885 890
886 return stack; 891 return stack;
887 } 892 }
888 893
889 static unsigned long page_owner_pages_threshol 894 static unsigned long page_owner_pages_threshold;
890 895
891 static int stack_print(struct seq_file *m, voi 896 static int stack_print(struct seq_file *m, void *v)
892 { 897 {
893 int i, nr_base_pages; 898 int i, nr_base_pages;
894 struct stack *stack = v; 899 struct stack *stack = v;
895 unsigned long *entries; 900 unsigned long *entries;
896 unsigned long nr_entries; 901 unsigned long nr_entries;
897 struct stack_record *stack_record = st 902 struct stack_record *stack_record = stack->stack_record;
898 903
899 if (!stack->stack_record) 904 if (!stack->stack_record)
900 return 0; 905 return 0;
901 906
902 nr_entries = stack_record->size; 907 nr_entries = stack_record->size;
903 entries = stack_record->entries; 908 entries = stack_record->entries;
904 nr_base_pages = refcount_read(&stack_r 909 nr_base_pages = refcount_read(&stack_record->count) - 1;
905 910
906 if (nr_base_pages < 1 || nr_base_pages 911 if (nr_base_pages < 1 || nr_base_pages < page_owner_pages_threshold)
907 return 0; 912 return 0;
908 913
909 for (i = 0; i < nr_entries; i++) 914 for (i = 0; i < nr_entries; i++)
910 seq_printf(m, " %pS\n", (void 915 seq_printf(m, " %pS\n", (void *)entries[i]);
911 seq_printf(m, "nr_base_pages: %d\n\n", 916 seq_printf(m, "nr_base_pages: %d\n\n", nr_base_pages);
912 917
913 return 0; 918 return 0;
914 } 919 }
915 920
916 static void stack_stop(struct seq_file *m, voi 921 static void stack_stop(struct seq_file *m, void *v)
917 { 922 {
918 } 923 }
919 924
920 static const struct seq_operations page_owner_ 925 static const struct seq_operations page_owner_stack_op = {
921 .start = stack_start, 926 .start = stack_start,
922 .next = stack_next, 927 .next = stack_next,
923 .stop = stack_stop, 928 .stop = stack_stop,
924 .show = stack_print 929 .show = stack_print
925 }; 930 };
926 931
927 static int page_owner_stack_open(struct inode 932 static int page_owner_stack_open(struct inode *inode, struct file *file)
928 { 933 {
929 return seq_open_private(file, &page_ow 934 return seq_open_private(file, &page_owner_stack_op, 0);
930 } 935 }
931 936
932 static const struct file_operations page_owner 937 static const struct file_operations page_owner_stack_operations = {
933 .open = page_owner_stack_ope 938 .open = page_owner_stack_open,
934 .read = seq_read, 939 .read = seq_read,
935 .llseek = seq_lseek, 940 .llseek = seq_lseek,
936 .release = seq_release, 941 .release = seq_release,
937 }; 942 };
938 943
939 static int page_owner_threshold_get(void *data 944 static int page_owner_threshold_get(void *data, u64 *val)
940 { 945 {
941 *val = READ_ONCE(page_owner_pages_thre 946 *val = READ_ONCE(page_owner_pages_threshold);
942 return 0; 947 return 0;
943 } 948 }
944 949
945 static int page_owner_threshold_set(void *data 950 static int page_owner_threshold_set(void *data, u64 val)
946 { 951 {
947 WRITE_ONCE(page_owner_pages_threshold, 952 WRITE_ONCE(page_owner_pages_threshold, val);
948 return 0; 953 return 0;
949 } 954 }
950 955
951 DEFINE_SIMPLE_ATTRIBUTE(proc_page_owner_thresh 956 DEFINE_SIMPLE_ATTRIBUTE(proc_page_owner_threshold, &page_owner_threshold_get,
952 &page_owner_threshold_ 957 &page_owner_threshold_set, "%llu");
953 958
954 959
955 static int __init pageowner_init(void) 960 static int __init pageowner_init(void)
956 { 961 {
957 struct dentry *dir; 962 struct dentry *dir;
958 963
959 if (!static_branch_unlikely(&page_owne 964 if (!static_branch_unlikely(&page_owner_inited)) {
960 pr_info("page_owner is disable 965 pr_info("page_owner is disabled\n");
961 return 0; 966 return 0;
962 } 967 }
963 968
964 debugfs_create_file("page_owner", 0400 969 debugfs_create_file("page_owner", 0400, NULL, NULL,
965 &proc_page_owner_o 970 &proc_page_owner_operations);
966 dir = debugfs_create_dir("page_owner_s 971 dir = debugfs_create_dir("page_owner_stacks", NULL);
967 debugfs_create_file("show_stacks", 040 972 debugfs_create_file("show_stacks", 0400, dir, NULL,
968 &page_owner_stack_ 973 &page_owner_stack_operations);
969 debugfs_create_file("count_threshold", 974 debugfs_create_file("count_threshold", 0600, dir, NULL,
970 &proc_page_owner_t 975 &proc_page_owner_threshold);
971 976
972 return 0; 977 return 0;
973 } 978 }
974 late_initcall(pageowner_init) 979 late_initcall(pageowner_init)
975 980
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.