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