xref: /linux/kernel/trace/fgraph.c (revision b22845487096247f5370c412376a472304627847)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Infrastructure to took into function calls and returns.
4  * Copyright (c) 2008-2009 Frederic Weisbecker <fweisbec@gmail.com>
5  * Mostly borrowed from function tracer which
6  * is Copyright (c) Steven Rostedt <srostedt@redhat.com>
7  *
8  * Highly modified by Steven Rostedt (VMware).
9  */
10 #include <linux/bits.h>
11 #include <linux/jump_label.h>
12 #include <linux/suspend.h>
13 #include <linux/ftrace.h>
14 #include <linux/static_call.h>
15 #include <linux/slab.h>
16 
17 #include <trace/events/sched.h>
18 
19 #include "ftrace_internal.h"
20 #include "trace.h"
21 
22 /*
23  * FGRAPH_FRAME_SIZE:	Size in bytes of the meta data on the shadow stack
24  * FGRAPH_FRAME_OFFSET:	Size in long words of the meta data frame
25  */
26 #define FGRAPH_FRAME_SIZE	sizeof(struct ftrace_ret_stack)
27 #define FGRAPH_FRAME_OFFSET	DIV_ROUND_UP(FGRAPH_FRAME_SIZE, sizeof(long))
28 
29 /*
30  * On entry to a function (via function_graph_enter()), a new fgraph frame
31  * (ftrace_ret_stack) is pushed onto the stack as well as a word that
32  * holds a bitmask and a type (called "bitmap"). The bitmap is defined as:
33  *
34  * bits:  0 -  9	offset in words from the previous ftrace_ret_stack
35  *
36  * bits: 10 - 11	Type of storage
37  *			  0 - reserved
38  *			  1 - bitmap of fgraph_array index
39  *			  2 - reserved data
40  *
41  * For type with "bitmap of fgraph_array index" (FGRAPH_TYPE_BITMAP):
42  *  bits: 12 - 27	The bitmap of fgraph_ops fgraph_array index
43  *			That is, it's a bitmask of 0-15 (16 bits)
44  *			where if a corresponding ops in the fgraph_array[]
45  *			expects a callback from the return of the function
46  *			it's corresponding bit will be set.
47  *
48  *
49  * The top of the ret_stack (when not empty) will always have a reference
50  * word that points to the last fgraph frame that was saved.
51  *
52  * For reserved data:
53  *  bits: 12 - 17	The size in words that is stored
54  *  bits: 18 - 23	The index of fgraph_array, which shows who is stored
55  *
56  * That is, at the end of function_graph_enter, if the first and forth
57  * fgraph_ops on the fgraph_array[] (index 0 and 3) needs their retfunc called
58  * on the return of the function being traced, and the forth fgraph_ops
59  * stored two words of data, this is what will be on the task's shadow
60  * ret_stack: (the stack grows upward)
61  *
62  *  ret_stack[SHADOW_STACK_OFFSET]
63  * | SHADOW_STACK_TASK_VARS(ret_stack)[15]      |
64  * ...
65  * | SHADOW_STACK_TASK_VARS(ret_stack)[0]       |
66  *  ret_stack[SHADOW_STACK_MAX_OFFSET]
67  * ...
68  * |                                            | <- task->curr_ret_stack
69  * +--------------------------------------------+
70  * | (3 << 12) | (3 << 10) | FGRAPH_FRAME_OFFSET|
71  * |         *or put another way*               |
72  * | (3 << FGRAPH_DATA_INDEX_SHIFT)| \          | This is for fgraph_ops[3].
73  * | ((2 - 1) << FGRAPH_DATA_SHIFT)| \          | The data size is 2 words.
74  * | (FGRAPH_TYPE_DATA << FGRAPH_TYPE_SHIFT)| \ |
75  * | (offset2:FGRAPH_FRAME_OFFSET+3)            | <- the offset2 is from here
76  * +--------------------------------------------+ ( It is 4 words from the ret_stack)
77  * |            STORED DATA WORD 2              |
78  * |            STORED DATA WORD 1              |
79  * +--------------------------------------------+
80  * | (9 << 12) | (1 << 10) | FGRAPH_FRAME_OFFSET|
81  * |         *or put another way*               |
82  * | (BIT(3)|BIT(0)) << FGRAPH_INDEX_SHIFT | \  |
83  * | FGRAPH_TYPE_BITMAP << FGRAPH_TYPE_SHIFT| \ |
84  * | (offset1:FGRAPH_FRAME_OFFSET)              | <- the offset1 is from here
85  * +--------------------------------------------+
86  * | struct ftrace_ret_stack                    |
87  * |   (stores the saved ret pointer)           | <- the offset points here
88  * +--------------------------------------------+
89  * |                 (X) | (N)                  | ( N words away from
90  * |                                            |   previous ret_stack)
91  * ...
92  * ret_stack[0]
93  *
94  * If a backtrace is required, and the real return pointer needs to be
95  * fetched, then it looks at the task's curr_ret_stack offset, if it
96  * is greater than zero (reserved, or right before popped), it would mask
97  * the value by FGRAPH_FRAME_OFFSET_MASK to get the offset of the
98  * ftrace_ret_stack structure stored on the shadow stack.
99  */
100 
101 /*
102  * The following is for the top word on the stack:
103  *
104  *   FGRAPH_FRAME_OFFSET (0-9) holds the offset delta to the fgraph frame
105  *   FGRAPH_TYPE (10-11) holds the type of word this is.
106  *     (RESERVED or BITMAP)
107  */
108 #define FGRAPH_FRAME_OFFSET_BITS	10
109 #define FGRAPH_FRAME_OFFSET_MASK	GENMASK(FGRAPH_FRAME_OFFSET_BITS - 1, 0)
110 
111 #define FGRAPH_TYPE_BITS	2
112 #define FGRAPH_TYPE_MASK	GENMASK(FGRAPH_TYPE_BITS - 1, 0)
113 #define FGRAPH_TYPE_SHIFT	FGRAPH_FRAME_OFFSET_BITS
114 
115 enum {
116 	FGRAPH_TYPE_RESERVED	= 0,
117 	FGRAPH_TYPE_BITMAP	= 1,
118 	FGRAPH_TYPE_DATA	= 2,
119 };
120 
121 /*
122  * For BITMAP type:
123  *   FGRAPH_INDEX (12-27) bits holding the gops index wanting return callback called
124  */
125 #define FGRAPH_INDEX_BITS	16
126 #define FGRAPH_INDEX_MASK	GENMASK(FGRAPH_INDEX_BITS - 1, 0)
127 #define FGRAPH_INDEX_SHIFT	(FGRAPH_TYPE_SHIFT + FGRAPH_TYPE_BITS)
128 
129 /*
130  * For DATA type:
131  *  FGRAPH_DATA (12-17) bits hold the size of data (in words)
132  *  FGRAPH_INDEX (18-23) bits hold the index for which gops->idx the data is for
133  *
134  * Note:
135  *  data_size == 0 means 1 word, and 31 (=2^5 - 1) means 32 words.
136  */
137 #define FGRAPH_DATA_BITS	5
138 #define FGRAPH_DATA_MASK	GENMASK(FGRAPH_DATA_BITS - 1, 0)
139 #define FGRAPH_DATA_SHIFT	(FGRAPH_TYPE_SHIFT + FGRAPH_TYPE_BITS)
140 #define FGRAPH_MAX_DATA_SIZE (sizeof(long) * (1 << FGRAPH_DATA_BITS))
141 
142 #define FGRAPH_DATA_INDEX_BITS	4
143 #define FGRAPH_DATA_INDEX_MASK	GENMASK(FGRAPH_DATA_INDEX_BITS - 1, 0)
144 #define FGRAPH_DATA_INDEX_SHIFT	(FGRAPH_DATA_SHIFT + FGRAPH_DATA_BITS)
145 
146 #define FGRAPH_ARRAY_SIZE	FGRAPH_INDEX_BITS
147 
148 /*
149  * SHADOW_STACK_SIZE:	The size in bytes of the entire shadow stack
150  * SHADOW_STACK_OFFSET:	The size in long words of the shadow stack
151  * SHADOW_STACK_MAX_OFFSET: The max offset of the stack for a new frame to be added
152  */
153 #define SHADOW_STACK_SIZE	(4096)
154 #define SHADOW_STACK_OFFSET	(SHADOW_STACK_SIZE / sizeof(long))
155 /* Leave on a buffer at the end */
156 #define SHADOW_STACK_MAX_OFFSET				\
157 	(SHADOW_STACK_OFFSET - (FGRAPH_FRAME_OFFSET + 1 + FGRAPH_ARRAY_SIZE))
158 
159 /* RET_STACK():		Return the frame from a given @offset from task @t */
160 #define RET_STACK(t, offset) ((struct ftrace_ret_stack *)(&(t)->ret_stack[offset]))
161 
162 /*
163  * Each fgraph_ops has a reserved unsigned long at the end (top) of the
164  * ret_stack to store task specific state.
165  */
166 #define SHADOW_STACK_TASK_VARS(ret_stack) \
167 	((unsigned long *)(&(ret_stack)[SHADOW_STACK_OFFSET - FGRAPH_ARRAY_SIZE]))
168 
169 DEFINE_STATIC_KEY_FALSE(kill_ftrace_graph);
170 int ftrace_graph_active;
171 
172 static struct kmem_cache *fgraph_stack_cachep;
173 
174 static struct fgraph_ops *fgraph_array[FGRAPH_ARRAY_SIZE];
175 static unsigned long fgraph_array_bitmask;
176 
177 /* LRU index table for fgraph_array */
178 static int fgraph_lru_table[FGRAPH_ARRAY_SIZE];
179 static int fgraph_lru_next;
180 static int fgraph_lru_last;
181 
182 /* Initialize fgraph_lru_table with unused index */
183 static void fgraph_lru_init(void)
184 {
185 	int i;
186 
187 	for (i = 0; i < FGRAPH_ARRAY_SIZE; i++)
188 		fgraph_lru_table[i] = i;
189 }
190 
191 /* Release the used index to the LRU table */
192 static int fgraph_lru_release_index(int idx)
193 {
194 	if (idx < 0 || idx >= FGRAPH_ARRAY_SIZE ||
195 	    WARN_ON_ONCE(fgraph_lru_table[fgraph_lru_last] != -1))
196 		return -1;
197 
198 	fgraph_lru_table[fgraph_lru_last] = idx;
199 	fgraph_lru_last = (fgraph_lru_last + 1) % FGRAPH_ARRAY_SIZE;
200 
201 	clear_bit(idx, &fgraph_array_bitmask);
202 	return 0;
203 }
204 
205 /* Allocate a new index from LRU table */
206 static int fgraph_lru_alloc_index(void)
207 {
208 	int idx = fgraph_lru_table[fgraph_lru_next];
209 
210 	/* No id is available */
211 	if (idx == -1)
212 		return -1;
213 
214 	fgraph_lru_table[fgraph_lru_next] = -1;
215 	fgraph_lru_next = (fgraph_lru_next + 1) % FGRAPH_ARRAY_SIZE;
216 
217 	set_bit(idx, &fgraph_array_bitmask);
218 	return idx;
219 }
220 
221 /* Get the offset to the fgraph frame from a ret_stack value */
222 static inline int __get_offset(unsigned long val)
223 {
224 	return val & FGRAPH_FRAME_OFFSET_MASK;
225 }
226 
227 /* Get the type of word from a ret_stack value */
228 static inline int __get_type(unsigned long val)
229 {
230 	return (val >> FGRAPH_TYPE_SHIFT) & FGRAPH_TYPE_MASK;
231 }
232 
233 /* Get the data_index for a DATA type ret_stack word */
234 static inline int __get_data_index(unsigned long val)
235 {
236 	return (val >> FGRAPH_DATA_INDEX_SHIFT) & FGRAPH_DATA_INDEX_MASK;
237 }
238 
239 /* Get the data_size for a DATA type ret_stack word */
240 static inline int __get_data_size(unsigned long val)
241 {
242 	return ((val >> FGRAPH_DATA_SHIFT) & FGRAPH_DATA_MASK) + 1;
243 }
244 
245 /* Get the word from the ret_stack at @offset */
246 static inline unsigned long get_fgraph_entry(struct task_struct *t, int offset)
247 {
248 	return t->ret_stack[offset];
249 }
250 
251 /* Get the FRAME_OFFSET from the word from the @offset on ret_stack */
252 static inline int get_frame_offset(struct task_struct *t, int offset)
253 {
254 	return __get_offset(t->ret_stack[offset]);
255 }
256 
257 /* For BITMAP type: get the bitmask from the @offset at ret_stack */
258 static inline unsigned long
259 get_bitmap_bits(struct task_struct *t, int offset)
260 {
261 	return (t->ret_stack[offset] >> FGRAPH_INDEX_SHIFT) & FGRAPH_INDEX_MASK;
262 }
263 
264 /* Write the bitmap to the ret_stack at @offset (does index, offset and bitmask) */
265 static inline void
266 set_bitmap(struct task_struct *t, int offset, unsigned long bitmap)
267 {
268 	t->ret_stack[offset] = (bitmap << FGRAPH_INDEX_SHIFT) |
269 		(FGRAPH_TYPE_BITMAP << FGRAPH_TYPE_SHIFT) | FGRAPH_FRAME_OFFSET;
270 }
271 
272 /* For DATA type: get the data saved under the ret_stack word at @offset */
273 static inline void *get_data_type_data(struct task_struct *t, int offset)
274 {
275 	unsigned long val = t->ret_stack[offset];
276 
277 	if (__get_type(val) != FGRAPH_TYPE_DATA)
278 		return NULL;
279 	offset -= __get_data_size(val);
280 	return (void *)&t->ret_stack[offset];
281 }
282 
283 /* Create the ret_stack word for a DATA type */
284 static inline unsigned long make_data_type_val(int idx, int size, int offset)
285 {
286 	return (idx << FGRAPH_DATA_INDEX_SHIFT) |
287 		((size - 1) << FGRAPH_DATA_SHIFT) |
288 		(FGRAPH_TYPE_DATA << FGRAPH_TYPE_SHIFT) | offset;
289 }
290 
291 /* ftrace_graph_entry set to this to tell some archs to run function graph */
292 static int entry_run(struct ftrace_graph_ent *trace, struct fgraph_ops *ops,
293 		     struct ftrace_regs *fregs)
294 {
295 	return 0;
296 }
297 
298 /* ftrace_graph_return set to this to tell some archs to run function graph */
299 static void return_run(struct ftrace_graph_ret *trace, struct fgraph_ops *ops,
300 		       struct ftrace_regs *fregs)
301 {
302 }
303 
304 static void ret_stack_set_task_var(struct task_struct *t, int idx, long val)
305 {
306 	unsigned long *gvals = SHADOW_STACK_TASK_VARS(t->ret_stack);
307 
308 	gvals[idx] = val;
309 }
310 
311 static unsigned long *
312 ret_stack_get_task_var(struct task_struct *t, int idx)
313 {
314 	unsigned long *gvals = SHADOW_STACK_TASK_VARS(t->ret_stack);
315 
316 	return &gvals[idx];
317 }
318 
319 static void ret_stack_init_task_vars(unsigned long *ret_stack)
320 {
321 	unsigned long *gvals = SHADOW_STACK_TASK_VARS(ret_stack);
322 
323 	memset(gvals, 0, sizeof(*gvals) * FGRAPH_ARRAY_SIZE);
324 }
325 
326 /**
327  * fgraph_reserve_data - Reserve storage on the task's ret_stack
328  * @idx:	The index of fgraph_array
329  * @size_bytes: The size in bytes to reserve
330  *
331  * Reserves space of up to FGRAPH_MAX_DATA_SIZE bytes on the
332  * task's ret_stack shadow stack, for a given fgraph_ops during
333  * the entryfunc() call. If entryfunc() returns zero, the storage
334  * is discarded. An entryfunc() can only call this once per iteration.
335  * The fgraph_ops retfunc() can retrieve this stored data with
336  * fgraph_retrieve_data().
337  *
338  * Returns: On success, a pointer to the data on the stack.
339  *   Otherwise, NULL if there's not enough space left on the
340  *   ret_stack for the data, or if fgraph_reserve_data() was called
341  *   more than once for a single entryfunc() call.
342  */
343 void *fgraph_reserve_data(int idx, int size_bytes)
344 {
345 	unsigned long val;
346 	void *data;
347 	int curr_ret_stack = current->curr_ret_stack;
348 	int data_size;
349 
350 	if (size_bytes > FGRAPH_MAX_DATA_SIZE)
351 		return NULL;
352 
353 	/* Convert the data size to number of longs. */
354 	data_size = (size_bytes + sizeof(long) - 1) >> (sizeof(long) == 4 ? 2 : 3);
355 
356 	val = get_fgraph_entry(current, curr_ret_stack - 1);
357 	data = &current->ret_stack[curr_ret_stack];
358 
359 	curr_ret_stack += data_size + 1;
360 	if (unlikely(curr_ret_stack >= SHADOW_STACK_MAX_OFFSET))
361 		return NULL;
362 
363 	val = make_data_type_val(idx, data_size, __get_offset(val) + data_size + 1);
364 
365 	/* Set the last word to be reserved */
366 	current->ret_stack[curr_ret_stack - 1] = val;
367 
368 	/* Make sure interrupts see this */
369 	barrier();
370 	current->curr_ret_stack = curr_ret_stack;
371 	/* Again sync with interrupts, and reset reserve */
372 	current->ret_stack[curr_ret_stack - 1] = val;
373 
374 	return data;
375 }
376 
377 /**
378  * fgraph_retrieve_data - Retrieve stored data from fgraph_reserve_data()
379  * @idx:	the index of fgraph_array (fgraph_ops::idx)
380  * @size_bytes: pointer to retrieved data size.
381  *
382  * This is to be called by a fgraph_ops retfunc(), to retrieve data that
383  * was stored by the fgraph_ops entryfunc() on the function entry.
384  * That is, this will retrieve the data that was reserved on the
385  * entry of the function that corresponds to the exit of the function
386  * that the fgraph_ops retfunc() is called on.
387  *
388  * Returns: The stored data from fgraph_reserve_data() called by the
389  *    matching entryfunc() for the retfunc() this is called from.
390  *   Or NULL if there was nothing stored.
391  */
392 void *fgraph_retrieve_data(int idx, int *size_bytes)
393 {
394 	return fgraph_retrieve_parent_data(idx, size_bytes, 0);
395 }
396 
397 /**
398  * fgraph_get_task_var - retrieve a task specific state variable
399  * @gops: The ftrace_ops that owns the task specific variable
400  *
401  * Every registered fgraph_ops has a task state variable
402  * reserved on the task's ret_stack. This function returns the
403  * address to that variable.
404  *
405  * Returns the address to the fgraph_ops @gops tasks specific
406  * unsigned long variable.
407  */
408 unsigned long *fgraph_get_task_var(struct fgraph_ops *gops)
409 {
410 	return ret_stack_get_task_var(current, gops->idx);
411 }
412 
413 /*
414  * @offset: The offset into @t->ret_stack to find the ret_stack entry
415  * @frame_offset: Where to place the offset into @t->ret_stack of that entry
416  *
417  * Returns a pointer to the previous ret_stack below @offset or NULL
418  *   when it reaches the bottom of the stack.
419  *
420  * Calling this with:
421  *
422  *   offset = task->curr_ret_stack;
423  *   do {
424  *	ret_stack = get_ret_stack(task, offset, &offset);
425  *   } while (ret_stack);
426  *
427  * Will iterate through all the ret_stack entries from curr_ret_stack
428  * down to the first one.
429  */
430 static inline struct ftrace_ret_stack *
431 get_ret_stack(struct task_struct *t, int offset, int *frame_offset)
432 {
433 	int offs;
434 
435 	BUILD_BUG_ON(FGRAPH_FRAME_SIZE % sizeof(long));
436 
437 	if (unlikely(offset <= 0))
438 		return NULL;
439 
440 	offs = get_frame_offset(t, --offset);
441 	if (WARN_ON_ONCE(offs <= 0 || offs > offset))
442 		return NULL;
443 
444 	offset -= offs;
445 
446 	*frame_offset = offset;
447 	return RET_STACK(t, offset);
448 }
449 
450 /**
451  * fgraph_retrieve_parent_data - get data from a parent function
452  * @idx: The index into the fgraph_array (fgraph_ops::idx)
453  * @size_bytes: A pointer to retrieved data size
454  * @depth: The depth to find the parent (0 is the current function)
455  *
456  * This is similar to fgraph_retrieve_data() but can be used to retrieve
457  * data from a parent caller function.
458  *
459  * Return: a pointer to the specified parent data or NULL if not found
460  */
461 void *fgraph_retrieve_parent_data(int idx, int *size_bytes, int depth)
462 {
463 	struct ftrace_ret_stack *ret_stack = NULL;
464 	int offset = current->curr_ret_stack;
465 	unsigned long val;
466 
467 	if (offset <= 0)
468 		return NULL;
469 
470 	for (;;) {
471 		int next_offset;
472 
473 		ret_stack = get_ret_stack(current, offset, &next_offset);
474 		if (!ret_stack || --depth < 0)
475 			break;
476 		offset = next_offset;
477 	}
478 
479 	if (!ret_stack)
480 		return NULL;
481 
482 	offset--;
483 
484 	val = get_fgraph_entry(current, offset);
485 	while (__get_type(val) == FGRAPH_TYPE_DATA) {
486 		if (__get_data_index(val) == idx)
487 			goto found;
488 		offset -= __get_data_size(val) + 1;
489 		val = get_fgraph_entry(current, offset);
490 	}
491 	return NULL;
492 found:
493 	if (size_bytes)
494 		*size_bytes = __get_data_size(val) * sizeof(long);
495 	return get_data_type_data(current, offset);
496 }
497 
498 #ifdef CONFIG_DYNAMIC_FTRACE
499 /*
500  * archs can override this function if they must do something
501  * to enable hook for graph tracer.
502  */
503 int __weak ftrace_enable_ftrace_graph_caller(void)
504 {
505 	return 0;
506 }
507 
508 /*
509  * archs can override this function if they must do something
510  * to disable hook for graph tracer.
511  */
512 int __weak ftrace_disable_ftrace_graph_caller(void)
513 {
514 	return 0;
515 }
516 #endif
517 
518 int ftrace_graph_entry_stub(struct ftrace_graph_ent *trace,
519 			    struct fgraph_ops *gops,
520 			    struct ftrace_regs *fregs)
521 {
522 	return 0;
523 }
524 
525 static void ftrace_graph_ret_stub(struct ftrace_graph_ret *trace,
526 				  struct fgraph_ops *gops,
527 				  struct ftrace_regs *fregs)
528 {
529 }
530 
531 static struct fgraph_ops fgraph_stub = {
532 	.entryfunc = ftrace_graph_entry_stub,
533 	.retfunc = ftrace_graph_ret_stub,
534 };
535 
536 static struct fgraph_ops *fgraph_direct_gops = &fgraph_stub;
537 DEFINE_STATIC_CALL(fgraph_func, ftrace_graph_entry_stub);
538 DEFINE_STATIC_CALL(fgraph_retfunc, ftrace_graph_ret_stub);
539 #if FGRAPH_NO_DIRECT
540 static DEFINE_STATIC_KEY_FALSE(fgraph_do_direct);
541 #else
542 static DEFINE_STATIC_KEY_TRUE(fgraph_do_direct);
543 #endif
544 
545 /**
546  * ftrace_graph_stop - set to permanently disable function graph tracing
547  *
548  * In case of an error int function graph tracing, this is called
549  * to try to keep function graph tracing from causing any more harm.
550  * Usually this is pretty severe and this is called to try to at least
551  * get a warning out to the user.
552  */
553 void ftrace_graph_stop(void)
554 {
555 	static_branch_enable(&kill_ftrace_graph);
556 }
557 
558 /* Add a function return address to the trace stack on thread info.*/
559 static int
560 ftrace_push_return_trace(unsigned long ret, unsigned long func,
561 			 unsigned long frame_pointer, unsigned long *retp,
562 			 int fgraph_idx)
563 {
564 	struct ftrace_ret_stack *ret_stack;
565 	unsigned long val;
566 	int offset;
567 
568 	if (unlikely(ftrace_graph_is_dead()))
569 		return -EBUSY;
570 
571 	if (!current->ret_stack)
572 		return -EBUSY;
573 
574 	BUILD_BUG_ON(SHADOW_STACK_SIZE % sizeof(long));
575 
576 	/* Set val to "reserved" with the delta to the new fgraph frame */
577 	val = (FGRAPH_TYPE_RESERVED << FGRAPH_TYPE_SHIFT) | FGRAPH_FRAME_OFFSET;
578 
579 	/*
580 	 * We must make sure the ret_stack is tested before we read
581 	 * anything else.
582 	 */
583 	smp_rmb();
584 
585 	/*
586 	 * Check if there's room on the shadow stack to fit a fraph frame
587 	 * and a bitmap word.
588 	 */
589 	if (current->curr_ret_stack + FGRAPH_FRAME_OFFSET + 1 >= SHADOW_STACK_MAX_OFFSET) {
590 		atomic_inc(&current->trace_overrun);
591 		return -EBUSY;
592 	}
593 
594 	offset = READ_ONCE(current->curr_ret_stack);
595 	ret_stack = RET_STACK(current, offset);
596 	offset += FGRAPH_FRAME_OFFSET;
597 
598 	/* ret offset = FGRAPH_FRAME_OFFSET ; type = reserved */
599 	current->ret_stack[offset] = val;
600 	ret_stack->ret = ret;
601 	/*
602 	 * The unwinders expect curr_ret_stack to point to either zero
603 	 * or an offset where to find the next ret_stack. Even though the
604 	 * ret stack might be bogus, we want to write the ret and the
605 	 * offset to find the ret_stack before we increment the stack point.
606 	 * If an interrupt comes in now before we increment the curr_ret_stack
607 	 * it may blow away what we wrote. But that's fine, because the
608 	 * offset will still be correct (even though the 'ret' won't be).
609 	 * What we worry about is the offset being correct after we increment
610 	 * the curr_ret_stack and before we update that offset, as if an
611 	 * interrupt comes in and does an unwind stack dump, it will need
612 	 * at least a correct offset!
613 	 */
614 	barrier();
615 	WRITE_ONCE(current->curr_ret_stack, offset + 1);
616 	/*
617 	 * This next barrier is to ensure that an interrupt coming in
618 	 * will not corrupt what we are about to write.
619 	 */
620 	barrier();
621 
622 	/* Still keep it reserved even if an interrupt came in */
623 	current->ret_stack[offset] = val;
624 
625 	ret_stack->ret = ret;
626 	ret_stack->func = func;
627 #ifdef HAVE_FUNCTION_GRAPH_FP_TEST
628 	ret_stack->fp = frame_pointer;
629 #endif
630 	ret_stack->retp = retp;
631 	return offset;
632 }
633 
634 /*
635  * Not all archs define MCOUNT_INSN_SIZE which is used to look for direct
636  * functions. But those archs currently don't support direct functions
637  * anyway, and ftrace_find_rec_direct() is just a stub for them.
638  * Define MCOUNT_INSN_SIZE to keep those archs compiling.
639  */
640 #ifndef MCOUNT_INSN_SIZE
641 /* Make sure this only works without direct calls */
642 # ifdef CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS
643 #  error MCOUNT_INSN_SIZE not defined with direct calls enabled
644 # endif
645 # define MCOUNT_INSN_SIZE 0
646 #endif
647 
648 /* If the caller does not use ftrace, call this function. */
649 int function_graph_enter_regs(unsigned long ret, unsigned long func,
650 			      unsigned long frame_pointer, unsigned long *retp,
651 			      struct ftrace_regs *fregs)
652 {
653 	struct ftrace_graph_ent trace;
654 	unsigned long bitmap = 0;
655 	int offset;
656 	int bit;
657 	int i;
658 
659 	bit = ftrace_test_recursion_trylock(func, ret);
660 	if (bit < 0)
661 		return -EBUSY;
662 
663 	trace.func = func;
664 	trace.depth = ++current->curr_ret_depth;
665 
666 	offset = ftrace_push_return_trace(ret, func, frame_pointer, retp, 0);
667 	if (offset < 0)
668 		goto out;
669 
670 #ifdef CONFIG_HAVE_STATIC_CALL
671 	if (static_branch_likely(&fgraph_do_direct)) {
672 		int save_curr_ret_stack = current->curr_ret_stack;
673 
674 		if (static_call(fgraph_func)(&trace, fgraph_direct_gops, fregs))
675 			bitmap |= BIT(fgraph_direct_gops->idx);
676 		else
677 			/* Clear out any saved storage */
678 			current->curr_ret_stack = save_curr_ret_stack;
679 	} else
680 #endif
681 	{
682 		for_each_set_bit(i, &fgraph_array_bitmask,
683 					 sizeof(fgraph_array_bitmask) * BITS_PER_BYTE) {
684 			struct fgraph_ops *gops = READ_ONCE(fgraph_array[i]);
685 			int save_curr_ret_stack;
686 
687 			if (gops == &fgraph_stub)
688 				continue;
689 
690 			save_curr_ret_stack = current->curr_ret_stack;
691 			if (ftrace_ops_test(&gops->ops, func, NULL) &&
692 			    gops->entryfunc(&trace, gops, fregs))
693 				bitmap |= BIT(i);
694 			else
695 				/* Clear out any saved storage */
696 				current->curr_ret_stack = save_curr_ret_stack;
697 		}
698 	}
699 
700 	if (!bitmap)
701 		goto out_ret;
702 
703 	/*
704 	 * Since this function uses fgraph_idx = 0 as a tail-call checking
705 	 * flag, set that bit always.
706 	 */
707 	set_bitmap(current, offset, bitmap | BIT(0));
708 	ftrace_test_recursion_unlock(bit);
709 	return 0;
710  out_ret:
711 	current->curr_ret_stack -= FGRAPH_FRAME_OFFSET + 1;
712  out:
713 	current->curr_ret_depth--;
714 	ftrace_test_recursion_unlock(bit);
715 	return -EBUSY;
716 }
717 
718 /* Retrieve a function return address to the trace stack on thread info.*/
719 static struct ftrace_ret_stack *
720 ftrace_pop_return_trace(struct ftrace_graph_ret *trace, unsigned long *ret,
721 			unsigned long frame_pointer, int *offset)
722 {
723 	struct ftrace_ret_stack *ret_stack;
724 
725 	ret_stack = get_ret_stack(current, current->curr_ret_stack, offset);
726 
727 	if (unlikely(!ret_stack)) {
728 		ftrace_graph_stop();
729 		WARN(1, "Bad function graph ret_stack pointer: %d",
730 		     current->curr_ret_stack);
731 		/* Might as well panic, otherwise we have no where to go */
732 		*ret = (unsigned long)panic;
733 		return NULL;
734 	}
735 
736 #ifdef HAVE_FUNCTION_GRAPH_FP_TEST
737 	/*
738 	 * The arch may choose to record the frame pointer used
739 	 * and check it here to make sure that it is what we expect it
740 	 * to be. If gcc does not set the place holder of the return
741 	 * address in the frame pointer, and does a copy instead, then
742 	 * the function graph trace will fail. This test detects this
743 	 * case.
744 	 *
745 	 * Currently, x86_32 with optimize for size (-Os) makes the latest
746 	 * gcc do the above.
747 	 *
748 	 * Note, -mfentry does not use frame pointers, and this test
749 	 *  is not needed if CC_USING_FENTRY is set.
750 	 */
751 	if (unlikely(ret_stack->fp != frame_pointer)) {
752 		ftrace_graph_stop();
753 		WARN(1, "Bad frame pointer: expected %lx, received %lx\n"
754 		     "  from func %ps return to %lx\n",
755 		     ret_stack->fp,
756 		     frame_pointer,
757 		     (void *)ret_stack->func,
758 		     ret_stack->ret);
759 		*ret = (unsigned long)panic;
760 		return NULL;
761 	}
762 #endif
763 
764 	*offset += FGRAPH_FRAME_OFFSET;
765 	*ret = ret_stack->ret;
766 	trace->func = ret_stack->func;
767 	trace->overrun = atomic_read(&current->trace_overrun);
768 	trace->depth = current->curr_ret_depth;
769 	/*
770 	 * We still want to trace interrupts coming in if
771 	 * max_depth is set to 1. Make sure the decrement is
772 	 * seen before ftrace_graph_return.
773 	 */
774 	barrier();
775 
776 	return ret_stack;
777 }
778 
779 /*
780  * Hibernation protection.
781  * The state of the current task is too much unstable during
782  * suspend/restore to disk. We want to protect against that.
783  */
784 static int
785 ftrace_suspend_notifier_call(struct notifier_block *bl, unsigned long state,
786 							void *unused)
787 {
788 	switch (state) {
789 	case PM_HIBERNATION_PREPARE:
790 		pause_graph_tracing();
791 		break;
792 
793 	case PM_POST_HIBERNATION:
794 		unpause_graph_tracing();
795 		break;
796 	}
797 	return NOTIFY_DONE;
798 }
799 
800 static struct notifier_block ftrace_suspend_notifier = {
801 	.notifier_call = ftrace_suspend_notifier_call,
802 };
803 
804 /*
805  * Send the trace to the ring-buffer.
806  * @return the original return address.
807  */
808 static inline unsigned long
809 __ftrace_return_to_handler(struct ftrace_regs *fregs, unsigned long frame_pointer)
810 {
811 	struct ftrace_ret_stack *ret_stack;
812 	struct ftrace_graph_ret trace;
813 	unsigned long bitmap;
814 	unsigned long ret;
815 	int offset;
816 	int bit;
817 	int i;
818 
819 	ret_stack = ftrace_pop_return_trace(&trace, &ret, frame_pointer, &offset);
820 
821 	if (unlikely(!ret_stack)) {
822 		ftrace_graph_stop();
823 		WARN_ON(1);
824 		/* Might as well panic. What else to do? */
825 		return (unsigned long)panic;
826 	}
827 
828 	if (fregs)
829 		ftrace_regs_set_instruction_pointer(fregs, ret);
830 
831 	bit = ftrace_test_recursion_trylock(trace.func, ret);
832 	/*
833 	 * This can fail because ftrace_test_recursion_trylock() allows one nest
834 	 * call. If we are already in a nested call, then we don't probe this and
835 	 * just return the original return address.
836 	 */
837 	if (unlikely(bit < 0))
838 		goto out;
839 
840 #ifdef CONFIG_FUNCTION_GRAPH_RETVAL
841 	trace.retval = ftrace_regs_get_return_value(fregs);
842 #endif
843 
844 	bitmap = get_bitmap_bits(current, offset);
845 
846 #ifdef CONFIG_HAVE_STATIC_CALL
847 	if (!FGRAPH_NO_DIRECT && static_branch_likely(&fgraph_do_direct)) {
848 		if (test_bit(fgraph_direct_gops->idx, &bitmap))
849 			static_call(fgraph_retfunc)(&trace, fgraph_direct_gops, fregs);
850 	} else
851 #endif
852 	{
853 		for_each_set_bit(i, &bitmap, sizeof(bitmap) * BITS_PER_BYTE) {
854 			struct fgraph_ops *gops = READ_ONCE(fgraph_array[i]);
855 
856 			if (gops == &fgraph_stub)
857 				continue;
858 
859 			gops->retfunc(&trace, gops, fregs);
860 		}
861 	}
862 
863 	ftrace_test_recursion_unlock(bit);
864 out:
865 	/*
866 	 * The ftrace_graph_return() may still access the current
867 	 * ret_stack structure, we need to make sure the update of
868 	 * curr_ret_stack is after that.
869 	 */
870 	barrier();
871 	current->curr_ret_stack = offset - FGRAPH_FRAME_OFFSET;
872 
873 	current->curr_ret_depth--;
874 	return ret;
875 }
876 
877 /*
878  * After all architectures have selected HAVE_FUNCTION_GRAPH_FREGS, we can
879  * leave only ftrace_return_to_handler(fregs).
880  */
881 #ifdef CONFIG_HAVE_FUNCTION_GRAPH_FREGS
882 unsigned long ftrace_return_to_handler(struct ftrace_regs *fregs)
883 {
884 	return __ftrace_return_to_handler(fregs,
885 				ftrace_regs_get_frame_pointer(fregs));
886 }
887 #else
888 unsigned long ftrace_return_to_handler(unsigned long frame_pointer)
889 {
890 	return __ftrace_return_to_handler(NULL, frame_pointer);
891 }
892 #endif
893 
894 /**
895  * ftrace_graph_get_ret_stack - return the entry of the shadow stack
896  * @task: The task to read the shadow stack from.
897  * @idx: Index down the shadow stack
898  *
899  * Return the ret_struct on the shadow stack of the @task at the
900  * call graph at @idx starting with zero. If @idx is zero, it
901  * will return the last saved ret_stack entry. If it is greater than
902  * zero, it will return the corresponding ret_stack for the depth
903  * of saved return addresses.
904  */
905 struct ftrace_ret_stack *
906 ftrace_graph_get_ret_stack(struct task_struct *task, int idx)
907 {
908 	struct ftrace_ret_stack *ret_stack = NULL;
909 	int offset = task->curr_ret_stack;
910 
911 	if (offset < 0)
912 		return NULL;
913 
914 	do {
915 		ret_stack = get_ret_stack(task, offset, &offset);
916 	} while (ret_stack && --idx >= 0);
917 
918 	return ret_stack;
919 }
920 
921 /**
922  * ftrace_graph_top_ret_addr - return the top return address in the shadow stack
923  * @task: The task to read the shadow stack from.
924  *
925  * Return the first return address on the shadow stack of the @task, which is
926  * not the fgraph's return_to_handler.
927  */
928 unsigned long ftrace_graph_top_ret_addr(struct task_struct *task)
929 {
930 	unsigned long return_handler = (unsigned long)dereference_kernel_function_descriptor(return_to_handler);
931 	struct ftrace_ret_stack *ret_stack = NULL;
932 	int offset = task->curr_ret_stack;
933 
934 	if (offset < 0)
935 		return 0;
936 
937 	do {
938 		ret_stack = get_ret_stack(task, offset, &offset);
939 	} while (ret_stack && ret_stack->ret == return_handler);
940 
941 	return ret_stack ? ret_stack->ret : 0;
942 }
943 
944 /**
945  * ftrace_graph_ret_addr - return the original value of the return address
946  * @task: The task the unwinder is being executed on
947  * @idx: An initialized pointer to the next stack index to use
948  * @ret: The current return address (likely pointing to return_handler)
949  * @retp: The address on the stack of the current return location
950  *
951  * This function can be called by stack unwinding code to convert a found stack
952  * return address (@ret) to its original value, in case the function graph
953  * tracer has modified it to be 'return_to_handler'.  If the address hasn't
954  * been modified, the unchanged value of @ret is returned.
955  *
956  * @idx holds the last index used to know where to start from. It should be
957  * initialized to zero for the first iteration as that will mean to start
958  * at the top of the shadow stack. If the location is found, this pointer
959  * will be assigned that location so that if called again, it will continue
960  * where it left off.
961  *
962  * @retp is a pointer to the return address on the stack.
963  */
964 unsigned long ftrace_graph_ret_addr(struct task_struct *task, int *idx,
965 				    unsigned long ret, unsigned long *retp)
966 {
967 	struct ftrace_ret_stack *ret_stack;
968 	unsigned long return_handler = (unsigned long)dereference_kernel_function_descriptor(return_to_handler);
969 	int i;
970 
971 	if (ret != return_handler)
972 		return ret;
973 
974 	if (!idx)
975 		return ret;
976 
977 	i = *idx ? : task->curr_ret_stack;
978 	while (i > 0) {
979 		ret_stack = get_ret_stack(task, i, &i);
980 		if (!ret_stack)
981 			break;
982 		/*
983 		 * For the tail-call, there would be 2 or more ftrace_ret_stacks on
984 		 * the ret_stack, which records "return_to_handler" as the return
985 		 * address except for the last one.
986 		 * But on the real stack, there should be 1 entry because tail-call
987 		 * reuses the return address on the stack and jump to the next function.
988 		 * Thus we will continue to find real return address.
989 		 */
990 		if (ret_stack->retp == retp &&
991 		    ret_stack->ret != return_handler) {
992 			*idx = i;
993 			return ret_stack->ret;
994 		}
995 	}
996 
997 	return ret;
998 }
999 
1000 static struct ftrace_ops graph_ops = {
1001 	.func			= ftrace_graph_func,
1002 	.flags			= FTRACE_OPS_GRAPH_STUB,
1003 #ifdef FTRACE_GRAPH_TRAMP_ADDR
1004 	.trampoline		= FTRACE_GRAPH_TRAMP_ADDR,
1005 	/* trampoline_size is only needed for dynamically allocated tramps */
1006 #endif
1007 };
1008 
1009 void fgraph_init_ops(struct ftrace_ops *dst_ops,
1010 		     struct ftrace_ops *src_ops)
1011 {
1012 	dst_ops->flags = FTRACE_OPS_FL_PID | FTRACE_OPS_GRAPH_STUB;
1013 
1014 #ifdef CONFIG_DYNAMIC_FTRACE
1015 	if (src_ops) {
1016 		dst_ops->func_hash = &src_ops->local_hash;
1017 		mutex_init(&dst_ops->local_hash.regex_lock);
1018 		INIT_LIST_HEAD(&dst_ops->subop_list);
1019 		dst_ops->flags |= FTRACE_OPS_FL_INITIALIZED;
1020 		dst_ops->private = src_ops->private;
1021 	}
1022 #endif
1023 }
1024 
1025 /*
1026  * Simply points to ftrace_stub, but with the proper protocol.
1027  * Defined by the linker script in linux/vmlinux.lds.h
1028  */
1029 void ftrace_stub_graph(struct ftrace_graph_ret *trace, struct fgraph_ops *gops,
1030 		       struct ftrace_regs *fregs);
1031 
1032 /* The callbacks that hook a function */
1033 trace_func_graph_ret_t ftrace_graph_return = ftrace_stub_graph;
1034 trace_func_graph_ent_t ftrace_graph_entry = ftrace_graph_entry_stub;
1035 
1036 /* Try to assign a return stack array on FTRACE_RETSTACK_ALLOC_SIZE tasks. */
1037 static int alloc_retstack_tasklist(unsigned long **ret_stack_list)
1038 {
1039 	int i;
1040 	int ret = 0;
1041 	int start = 0, end = FTRACE_RETSTACK_ALLOC_SIZE;
1042 	struct task_struct *g, *t;
1043 
1044 	if (WARN_ON_ONCE(!fgraph_stack_cachep))
1045 		return -ENOMEM;
1046 
1047 	for (i = 0; i < FTRACE_RETSTACK_ALLOC_SIZE; i++) {
1048 		ret_stack_list[i] = kmem_cache_alloc(fgraph_stack_cachep, GFP_KERNEL);
1049 		if (!ret_stack_list[i]) {
1050 			start = 0;
1051 			end = i;
1052 			ret = -ENOMEM;
1053 			goto free;
1054 		}
1055 	}
1056 
1057 	rcu_read_lock();
1058 	for_each_process_thread(g, t) {
1059 		if (start == end) {
1060 			ret = -EAGAIN;
1061 			goto unlock;
1062 		}
1063 
1064 		if (t->ret_stack == NULL) {
1065 			atomic_set(&t->trace_overrun, 0);
1066 			ret_stack_init_task_vars(ret_stack_list[start]);
1067 			t->curr_ret_stack = 0;
1068 			t->curr_ret_depth = -1;
1069 			/* Make sure the tasks see the 0 first: */
1070 			smp_wmb();
1071 			t->ret_stack = ret_stack_list[start++];
1072 		}
1073 	}
1074 
1075 unlock:
1076 	rcu_read_unlock();
1077 free:
1078 	for (i = start; i < end; i++)
1079 		kmem_cache_free(fgraph_stack_cachep, ret_stack_list[i]);
1080 	return ret;
1081 }
1082 
1083 static void
1084 ftrace_graph_probe_sched_switch(void *ignore, bool preempt,
1085 				struct task_struct *prev,
1086 				struct task_struct *next,
1087 				unsigned int prev_state)
1088 {
1089 	unsigned long long timestamp;
1090 
1091 	/*
1092 	 * Does the user want to count the time a function was asleep.
1093 	 * If so, do not update the time stamps.
1094 	 */
1095 	if (!fgraph_no_sleep_time)
1096 		return;
1097 
1098 	timestamp = trace_clock_local();
1099 
1100 	prev->ftrace_timestamp = timestamp;
1101 
1102 	/* only process tasks that we timestamped */
1103 	if (!next->ftrace_timestamp)
1104 		return;
1105 
1106 	next->ftrace_sleeptime += timestamp - next->ftrace_timestamp;
1107 }
1108 
1109 static DEFINE_PER_CPU(unsigned long *, idle_ret_stack);
1110 
1111 static void
1112 graph_init_task(struct task_struct *t, unsigned long *ret_stack)
1113 {
1114 	atomic_set(&t->trace_overrun, 0);
1115 	ret_stack_init_task_vars(ret_stack);
1116 	t->ftrace_timestamp = 0;
1117 	t->curr_ret_stack = 0;
1118 	t->curr_ret_depth = -1;
1119 	/* make curr_ret_stack visible before we add the ret_stack */
1120 	smp_wmb();
1121 	t->ret_stack = ret_stack;
1122 }
1123 
1124 /*
1125  * Allocate a return stack for the idle task. May be the first
1126  * time through, or it may be done by CPU hotplug online.
1127  */
1128 void ftrace_graph_init_idle_task(struct task_struct *t, int cpu)
1129 {
1130 	t->curr_ret_stack = 0;
1131 	t->curr_ret_depth = -1;
1132 	/*
1133 	 * The idle task has no parent, it either has its own
1134 	 * stack or no stack at all.
1135 	 */
1136 	if (t->ret_stack)
1137 		WARN_ON(t->ret_stack != per_cpu(idle_ret_stack, cpu));
1138 
1139 	if (ftrace_graph_active) {
1140 		unsigned long *ret_stack;
1141 
1142 		if (WARN_ON_ONCE(!fgraph_stack_cachep))
1143 			return;
1144 
1145 		ret_stack = per_cpu(idle_ret_stack, cpu);
1146 		if (!ret_stack) {
1147 			ret_stack = kmem_cache_alloc(fgraph_stack_cachep, GFP_KERNEL);
1148 			if (!ret_stack)
1149 				return;
1150 			per_cpu(idle_ret_stack, cpu) = ret_stack;
1151 		}
1152 		graph_init_task(t, ret_stack);
1153 	}
1154 }
1155 
1156 /* Allocate a return stack for newly created task */
1157 void ftrace_graph_init_task(struct task_struct *t)
1158 {
1159 	/* Make sure we do not use the parent ret_stack */
1160 	t->ret_stack = NULL;
1161 	t->curr_ret_stack = 0;
1162 	t->curr_ret_depth = -1;
1163 
1164 	if (ftrace_graph_active) {
1165 		unsigned long *ret_stack;
1166 
1167 		if (WARN_ON_ONCE(!fgraph_stack_cachep))
1168 			return;
1169 
1170 		ret_stack = kmem_cache_alloc(fgraph_stack_cachep, GFP_KERNEL);
1171 		if (!ret_stack)
1172 			return;
1173 		graph_init_task(t, ret_stack);
1174 	}
1175 }
1176 
1177 void ftrace_graph_exit_task(struct task_struct *t)
1178 {
1179 	unsigned long *ret_stack = t->ret_stack;
1180 
1181 	t->ret_stack = NULL;
1182 	/* NULL must become visible to IRQs before we free it: */
1183 	barrier();
1184 
1185 	if (ret_stack) {
1186 		if (WARN_ON_ONCE(!fgraph_stack_cachep))
1187 			return;
1188 		kmem_cache_free(fgraph_stack_cachep, ret_stack);
1189 	}
1190 }
1191 
1192 #ifdef CONFIG_DYNAMIC_FTRACE
1193 static int fgraph_pid_func(struct ftrace_graph_ent *trace,
1194 			   struct fgraph_ops *gops,
1195 			   struct ftrace_regs *fregs)
1196 {
1197 	struct trace_array *tr = gops->ops.private;
1198 	int pid;
1199 
1200 	if (tr) {
1201 		pid = this_cpu_read(tr->array_buffer.data->ftrace_ignore_pid);
1202 		if (pid == FTRACE_PID_IGNORE)
1203 			return 0;
1204 		if (pid != FTRACE_PID_TRACE &&
1205 		    pid != current->pid)
1206 			return 0;
1207 	}
1208 
1209 	return gops->saved_func(trace, gops, fregs);
1210 }
1211 
1212 void fgraph_update_pid_func(void)
1213 {
1214 	struct fgraph_ops *gops;
1215 	struct ftrace_ops *op;
1216 
1217 	if (!(graph_ops.flags & FTRACE_OPS_FL_INITIALIZED))
1218 		return;
1219 
1220 	list_for_each_entry(op, &graph_ops.subop_list, list) {
1221 		if (op->flags & FTRACE_OPS_FL_PID) {
1222 			gops = container_of(op, struct fgraph_ops, ops);
1223 			gops->entryfunc = ftrace_pids_enabled(op) ?
1224 				fgraph_pid_func : gops->saved_func;
1225 			if (ftrace_graph_active == 1)
1226 				static_call_update(fgraph_func, gops->entryfunc);
1227 		}
1228 	}
1229 }
1230 #endif
1231 
1232 /* Allocate a return stack for each task */
1233 static int start_graph_tracing(void)
1234 {
1235 	unsigned long **ret_stack_list;
1236 	int ret, cpu;
1237 
1238 	ret_stack_list = kcalloc(FTRACE_RETSTACK_ALLOC_SIZE,
1239 				 sizeof(*ret_stack_list), GFP_KERNEL);
1240 
1241 	if (!ret_stack_list)
1242 		return -ENOMEM;
1243 
1244 	/* The cpu_boot init_task->ret_stack will never be freed */
1245 	for_each_online_cpu(cpu) {
1246 		if (!idle_task(cpu)->ret_stack)
1247 			ftrace_graph_init_idle_task(idle_task(cpu), cpu);
1248 	}
1249 
1250 	do {
1251 		ret = alloc_retstack_tasklist(ret_stack_list);
1252 	} while (ret == -EAGAIN);
1253 
1254 	if (!ret) {
1255 		ret = register_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
1256 		if (ret)
1257 			pr_info("ftrace_graph: Couldn't activate tracepoint"
1258 				" probe to kernel_sched_switch\n");
1259 	}
1260 
1261 	kfree(ret_stack_list);
1262 	return ret;
1263 }
1264 
1265 static void init_task_vars(int idx)
1266 {
1267 	struct task_struct *g, *t;
1268 	int cpu;
1269 
1270 	for_each_online_cpu(cpu) {
1271 		if (idle_task(cpu)->ret_stack)
1272 			ret_stack_set_task_var(idle_task(cpu), idx, 0);
1273 	}
1274 
1275 	read_lock(&tasklist_lock);
1276 	for_each_process_thread(g, t) {
1277 		if (t->ret_stack)
1278 			ret_stack_set_task_var(t, idx, 0);
1279 	}
1280 	read_unlock(&tasklist_lock);
1281 }
1282 
1283 static void ftrace_graph_enable_direct(bool enable_branch, struct fgraph_ops *gops)
1284 {
1285 	trace_func_graph_ent_t func = NULL;
1286 	trace_func_graph_ret_t retfunc = NULL;
1287 	int i;
1288 
1289 	if (FGRAPH_NO_DIRECT)
1290 		return;
1291 
1292 	if (gops) {
1293 		func = gops->entryfunc;
1294 		retfunc = gops->retfunc;
1295 		fgraph_direct_gops = gops;
1296 	} else {
1297 		for_each_set_bit(i, &fgraph_array_bitmask,
1298 				 sizeof(fgraph_array_bitmask) * BITS_PER_BYTE) {
1299 			func = fgraph_array[i]->entryfunc;
1300 			retfunc = fgraph_array[i]->retfunc;
1301 			fgraph_direct_gops = fgraph_array[i];
1302 		}
1303 	}
1304 	if (WARN_ON_ONCE(!func))
1305 		return;
1306 
1307 	static_call_update(fgraph_func, func);
1308 	static_call_update(fgraph_retfunc, retfunc);
1309 	if (enable_branch)
1310 		static_branch_enable(&fgraph_do_direct);
1311 }
1312 
1313 static void ftrace_graph_disable_direct(bool disable_branch)
1314 {
1315 	if (FGRAPH_NO_DIRECT)
1316 		return;
1317 
1318 	if (disable_branch)
1319 		static_branch_disable(&fgraph_do_direct);
1320 	static_call_update(fgraph_func, ftrace_graph_entry_stub);
1321 	static_call_update(fgraph_retfunc, ftrace_graph_ret_stub);
1322 	fgraph_direct_gops = &fgraph_stub;
1323 }
1324 
1325 /* The cpu_boot init_task->ret_stack will never be freed */
1326 static int fgraph_cpu_init(unsigned int cpu)
1327 {
1328 	if (!idle_task(cpu)->ret_stack)
1329 		ftrace_graph_init_idle_task(idle_task(cpu), cpu);
1330 	return 0;
1331 }
1332 
1333 int register_ftrace_graph(struct fgraph_ops *gops)
1334 {
1335 	static bool fgraph_initialized;
1336 	int command = 0;
1337 	int ret = 0;
1338 	int i = -1;
1339 
1340 	if (WARN_ONCE(gops->ops.flags & FTRACE_OPS_FL_GRAPH,
1341 		      "function graph ops registered again"))
1342 		return -EBUSY;
1343 
1344 	guard(mutex)(&ftrace_lock);
1345 
1346 	if (!fgraph_stack_cachep) {
1347 		fgraph_stack_cachep = kmem_cache_create("fgraph_stack",
1348 							SHADOW_STACK_SIZE,
1349 							SHADOW_STACK_SIZE, 0, NULL);
1350 		if (!fgraph_stack_cachep)
1351 			return -ENOMEM;
1352 	}
1353 
1354 	if (!fgraph_initialized) {
1355 		ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "fgraph:online",
1356 					fgraph_cpu_init, NULL);
1357 		if (ret < 0) {
1358 			pr_warn("fgraph: Error to init cpu hotplug support\n");
1359 			return ret;
1360 		}
1361 		fgraph_initialized = true;
1362 		ret = 0;
1363 	}
1364 
1365 	if (!fgraph_array[0]) {
1366 		/* The array must always have real data on it */
1367 		for (i = 0; i < FGRAPH_ARRAY_SIZE; i++)
1368 			fgraph_array[i] = &fgraph_stub;
1369 		fgraph_lru_init();
1370 	}
1371 
1372 	i = fgraph_lru_alloc_index();
1373 	if (i < 0 || WARN_ON_ONCE(fgraph_array[i] != &fgraph_stub))
1374 		return -ENOSPC;
1375 	gops->idx = i;
1376 
1377 	ftrace_graph_active++;
1378 
1379 	/* Always save the function, and reset at unregistering */
1380 	gops->saved_func = gops->entryfunc;
1381 #ifdef CONFIG_DYNAMIC_FTRACE
1382 	if (ftrace_pids_enabled(&gops->ops))
1383 		gops->entryfunc = fgraph_pid_func;
1384 #endif
1385 
1386 	if (ftrace_graph_active == 2)
1387 		ftrace_graph_disable_direct(true);
1388 
1389 	if (ftrace_graph_active == 1) {
1390 		ftrace_graph_enable_direct(false, gops);
1391 		register_pm_notifier(&ftrace_suspend_notifier);
1392 		ret = start_graph_tracing();
1393 		if (ret)
1394 			goto error;
1395 		/*
1396 		 * Some archs just test to see if these are not
1397 		 * the default function
1398 		 */
1399 		ftrace_graph_return = return_run;
1400 		ftrace_graph_entry = entry_run;
1401 		command = FTRACE_START_FUNC_RET;
1402 	} else {
1403 		init_task_vars(gops->idx);
1404 	}
1405 
1406 	gops->ops.flags |= FTRACE_OPS_FL_GRAPH;
1407 
1408 	ret = ftrace_startup_subops(&graph_ops, &gops->ops, command);
1409 	if (!ret)
1410 		fgraph_array[i] = gops;
1411 
1412 error:
1413 	if (ret) {
1414 		ftrace_graph_active--;
1415 		gops->saved_func = NULL;
1416 		fgraph_lru_release_index(i);
1417 		if (!ftrace_graph_active)
1418 			unregister_pm_notifier(&ftrace_suspend_notifier);
1419 	}
1420 	return ret;
1421 }
1422 
1423 void unregister_ftrace_graph(struct fgraph_ops *gops)
1424 {
1425 	int command = 0;
1426 
1427 	if (WARN_ONCE(!(gops->ops.flags & FTRACE_OPS_FL_GRAPH),
1428 		      "function graph ops unregistered without registering"))
1429 		return;
1430 
1431 	guard(mutex)(&ftrace_lock);
1432 
1433 	if (unlikely(!ftrace_graph_active))
1434 		goto out;
1435 
1436 	if (unlikely(gops->idx < 0 || gops->idx >= FGRAPH_ARRAY_SIZE ||
1437 		     fgraph_array[gops->idx] != gops))
1438 		goto out;
1439 
1440 	if (fgraph_lru_release_index(gops->idx) < 0)
1441 		goto out;
1442 
1443 	fgraph_array[gops->idx] = &fgraph_stub;
1444 
1445 	ftrace_graph_active--;
1446 
1447 	if (!ftrace_graph_active)
1448 		command = FTRACE_STOP_FUNC_RET;
1449 
1450 	ftrace_shutdown_subops(&graph_ops, &gops->ops, command);
1451 
1452 	if (ftrace_graph_active == 1)
1453 		ftrace_graph_enable_direct(true, NULL);
1454 	else if (!ftrace_graph_active)
1455 		ftrace_graph_disable_direct(false);
1456 
1457 	if (!ftrace_graph_active) {
1458 		ftrace_graph_return = ftrace_stub_graph;
1459 		ftrace_graph_entry = ftrace_graph_entry_stub;
1460 		unregister_pm_notifier(&ftrace_suspend_notifier);
1461 		unregister_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
1462 	}
1463 	gops->saved_func = NULL;
1464  out:
1465 	gops->ops.flags &= ~FTRACE_OPS_FL_GRAPH;
1466 }
1467