| 1 | /* |
| 2 | * Code for replacing ftrace calls with jumps. |
| 3 | * |
| 4 | * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com> |
| 5 | * |
| 6 | * Thanks goes to Ingo Molnar, for suggesting the idea. |
| 7 | * Mathieu Desnoyers, for suggesting postponing the modifications. |
| 8 | * Arjan van de Ven, for keeping me straight, and explaining to me |
| 9 | * the dangers of modifying code on the run. |
| 10 | */ |
| 11 | |
| 12 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
| 13 | |
| 14 | #include <linux/spinlock.h> |
| 15 | #include <linux/hardirq.h> |
| 16 | #include <linux/uaccess.h> |
| 17 | #include <linux/ftrace.h> |
| 18 | #include <linux/percpu.h> |
| 19 | #include <linux/sched.h> |
| 20 | #include <linux/init.h> |
| 21 | #include <linux/list.h> |
| 22 | #include <linux/module.h> |
| 23 | |
| 24 | #include <trace/syscall.h> |
| 25 | |
| 26 | #include <asm/cacheflush.h> |
| 27 | #include <asm/kprobes.h> |
| 28 | #include <asm/ftrace.h> |
| 29 | #include <asm/nops.h> |
| 30 | |
| 31 | #ifdef CONFIG_DYNAMIC_FTRACE |
| 32 | |
| 33 | int ftrace_arch_code_modify_prepare(void) |
| 34 | { |
| 35 | set_kernel_text_rw(); |
| 36 | set_all_modules_text_rw(); |
| 37 | return 0; |
| 38 | } |
| 39 | |
| 40 | int ftrace_arch_code_modify_post_process(void) |
| 41 | { |
| 42 | set_all_modules_text_ro(); |
| 43 | set_kernel_text_ro(); |
| 44 | return 0; |
| 45 | } |
| 46 | |
| 47 | union ftrace_code_union { |
| 48 | char code[MCOUNT_INSN_SIZE]; |
| 49 | struct { |
| 50 | char e8; |
| 51 | int offset; |
| 52 | } __attribute__((packed)); |
| 53 | }; |
| 54 | |
| 55 | static int ftrace_calc_offset(long ip, long addr) |
| 56 | { |
| 57 | return (int)(addr - ip); |
| 58 | } |
| 59 | |
| 60 | static unsigned char *ftrace_call_replace(unsigned long ip, unsigned long addr) |
| 61 | { |
| 62 | static union ftrace_code_union calc; |
| 63 | |
| 64 | calc.e8 = 0xe8; |
| 65 | calc.offset = ftrace_calc_offset(ip + MCOUNT_INSN_SIZE, addr); |
| 66 | |
| 67 | /* |
| 68 | * No locking needed, this must be called via kstop_machine |
| 69 | * which in essence is like running on a uniprocessor machine. |
| 70 | */ |
| 71 | return calc.code; |
| 72 | } |
| 73 | |
| 74 | static inline int |
| 75 | within(unsigned long addr, unsigned long start, unsigned long end) |
| 76 | { |
| 77 | return addr >= start && addr < end; |
| 78 | } |
| 79 | |
| 80 | static unsigned long text_ip_addr(unsigned long ip) |
| 81 | { |
| 82 | /* |
| 83 | * On x86_64, kernel text mappings are mapped read-only with |
| 84 | * CONFIG_DEBUG_RODATA. So we use the kernel identity mapping instead |
| 85 | * of the kernel text mapping to modify the kernel text. |
| 86 | * |
| 87 | * For 32bit kernels, these mappings are same and we can use |
| 88 | * kernel identity mapping to modify code. |
| 89 | */ |
| 90 | if (within(ip, (unsigned long)_text, (unsigned long)_etext)) |
| 91 | ip = (unsigned long)__va(__pa_symbol(ip)); |
| 92 | |
| 93 | return ip; |
| 94 | } |
| 95 | |
| 96 | static const unsigned char *ftrace_nop_replace(void) |
| 97 | { |
| 98 | return ideal_nops[NOP_ATOMIC5]; |
| 99 | } |
| 100 | |
| 101 | static int |
| 102 | ftrace_modify_code_direct(unsigned long ip, unsigned const char *old_code, |
| 103 | unsigned const char *new_code) |
| 104 | { |
| 105 | unsigned char replaced[MCOUNT_INSN_SIZE]; |
| 106 | |
| 107 | /* |
| 108 | * Note: Due to modules and __init, code can |
| 109 | * disappear and change, we need to protect against faulting |
| 110 | * as well as code changing. We do this by using the |
| 111 | * probe_kernel_* functions. |
| 112 | * |
| 113 | * No real locking needed, this code is run through |
| 114 | * kstop_machine, or before SMP starts. |
| 115 | */ |
| 116 | |
| 117 | /* read the text we want to modify */ |
| 118 | if (probe_kernel_read(replaced, (void *)ip, MCOUNT_INSN_SIZE)) |
| 119 | return -EFAULT; |
| 120 | |
| 121 | /* Make sure it is what we expect it to be */ |
| 122 | if (memcmp(replaced, old_code, MCOUNT_INSN_SIZE) != 0) |
| 123 | return -EINVAL; |
| 124 | |
| 125 | ip = text_ip_addr(ip); |
| 126 | |
| 127 | /* replace the text with the new text */ |
| 128 | if (probe_kernel_write((void *)ip, new_code, MCOUNT_INSN_SIZE)) |
| 129 | return -EPERM; |
| 130 | |
| 131 | sync_core(); |
| 132 | |
| 133 | return 0; |
| 134 | } |
| 135 | |
| 136 | int ftrace_make_nop(struct module *mod, |
| 137 | struct dyn_ftrace *rec, unsigned long addr) |
| 138 | { |
| 139 | unsigned const char *new, *old; |
| 140 | unsigned long ip = rec->ip; |
| 141 | |
| 142 | old = ftrace_call_replace(ip, addr); |
| 143 | new = ftrace_nop_replace(); |
| 144 | |
| 145 | /* |
| 146 | * On boot up, and when modules are loaded, the MCOUNT_ADDR |
| 147 | * is converted to a nop, and will never become MCOUNT_ADDR |
| 148 | * again. This code is either running before SMP (on boot up) |
| 149 | * or before the code will ever be executed (module load). |
| 150 | * We do not want to use the breakpoint version in this case, |
| 151 | * just modify the code directly. |
| 152 | */ |
| 153 | if (addr == MCOUNT_ADDR) |
| 154 | return ftrace_modify_code_direct(rec->ip, old, new); |
| 155 | |
| 156 | /* Normal cases use add_brk_on_nop */ |
| 157 | WARN_ONCE(1, "invalid use of ftrace_make_nop"); |
| 158 | return -EINVAL; |
| 159 | } |
| 160 | |
| 161 | int ftrace_make_call(struct dyn_ftrace *rec, unsigned long addr) |
| 162 | { |
| 163 | unsigned const char *new, *old; |
| 164 | unsigned long ip = rec->ip; |
| 165 | |
| 166 | old = ftrace_nop_replace(); |
| 167 | new = ftrace_call_replace(ip, addr); |
| 168 | |
| 169 | /* Should only be called when module is loaded */ |
| 170 | return ftrace_modify_code_direct(rec->ip, old, new); |
| 171 | } |
| 172 | |
| 173 | /* |
| 174 | * The modifying_ftrace_code is used to tell the breakpoint |
| 175 | * handler to call ftrace_int3_handler(). If it fails to |
| 176 | * call this handler for a breakpoint added by ftrace, then |
| 177 | * the kernel may crash. |
| 178 | * |
| 179 | * As atomic_writes on x86 do not need a barrier, we do not |
| 180 | * need to add smp_mb()s for this to work. It is also considered |
| 181 | * that we can not read the modifying_ftrace_code before |
| 182 | * executing the breakpoint. That would be quite remarkable if |
| 183 | * it could do that. Here's the flow that is required: |
| 184 | * |
| 185 | * CPU-0 CPU-1 |
| 186 | * |
| 187 | * atomic_inc(mfc); |
| 188 | * write int3s |
| 189 | * <trap-int3> // implicit (r)mb |
| 190 | * if (atomic_read(mfc)) |
| 191 | * call ftrace_int3_handler() |
| 192 | * |
| 193 | * Then when we are finished: |
| 194 | * |
| 195 | * atomic_dec(mfc); |
| 196 | * |
| 197 | * If we hit a breakpoint that was not set by ftrace, it does not |
| 198 | * matter if ftrace_int3_handler() is called or not. It will |
| 199 | * simply be ignored. But it is crucial that a ftrace nop/caller |
| 200 | * breakpoint is handled. No other user should ever place a |
| 201 | * breakpoint on an ftrace nop/caller location. It must only |
| 202 | * be done by this code. |
| 203 | */ |
| 204 | atomic_t modifying_ftrace_code __read_mostly; |
| 205 | |
| 206 | static int |
| 207 | ftrace_modify_code(unsigned long ip, unsigned const char *old_code, |
| 208 | unsigned const char *new_code); |
| 209 | |
| 210 | /* |
| 211 | * Should never be called: |
| 212 | * As it is only called by __ftrace_replace_code() which is called by |
| 213 | * ftrace_replace_code() that x86 overrides, and by ftrace_update_code() |
| 214 | * which is called to turn mcount into nops or nops into function calls |
| 215 | * but not to convert a function from not using regs to one that uses |
| 216 | * regs, which ftrace_modify_call() is for. |
| 217 | */ |
| 218 | int ftrace_modify_call(struct dyn_ftrace *rec, unsigned long old_addr, |
| 219 | unsigned long addr) |
| 220 | { |
| 221 | WARN_ON(1); |
| 222 | return -EINVAL; |
| 223 | } |
| 224 | |
| 225 | static unsigned long ftrace_update_func; |
| 226 | |
| 227 | static int update_ftrace_func(unsigned long ip, void *new) |
| 228 | { |
| 229 | unsigned char old[MCOUNT_INSN_SIZE]; |
| 230 | int ret; |
| 231 | |
| 232 | memcpy(old, (void *)ip, MCOUNT_INSN_SIZE); |
| 233 | |
| 234 | ftrace_update_func = ip; |
| 235 | /* Make sure the breakpoints see the ftrace_update_func update */ |
| 236 | smp_wmb(); |
| 237 | |
| 238 | /* See comment above by declaration of modifying_ftrace_code */ |
| 239 | atomic_inc(&modifying_ftrace_code); |
| 240 | |
| 241 | ret = ftrace_modify_code(ip, old, new); |
| 242 | |
| 243 | atomic_dec(&modifying_ftrace_code); |
| 244 | |
| 245 | return ret; |
| 246 | } |
| 247 | |
| 248 | int ftrace_update_ftrace_func(ftrace_func_t func) |
| 249 | { |
| 250 | unsigned long ip = (unsigned long)(&ftrace_call); |
| 251 | unsigned char *new; |
| 252 | int ret; |
| 253 | |
| 254 | new = ftrace_call_replace(ip, (unsigned long)func); |
| 255 | ret = update_ftrace_func(ip, new); |
| 256 | |
| 257 | /* Also update the regs callback function */ |
| 258 | if (!ret) { |
| 259 | ip = (unsigned long)(&ftrace_regs_call); |
| 260 | new = ftrace_call_replace(ip, (unsigned long)func); |
| 261 | ret = update_ftrace_func(ip, new); |
| 262 | } |
| 263 | |
| 264 | return ret; |
| 265 | } |
| 266 | |
| 267 | static int is_ftrace_caller(unsigned long ip) |
| 268 | { |
| 269 | if (ip == ftrace_update_func) |
| 270 | return 1; |
| 271 | |
| 272 | return 0; |
| 273 | } |
| 274 | |
| 275 | /* |
| 276 | * A breakpoint was added to the code address we are about to |
| 277 | * modify, and this is the handle that will just skip over it. |
| 278 | * We are either changing a nop into a trace call, or a trace |
| 279 | * call to a nop. While the change is taking place, we treat |
| 280 | * it just like it was a nop. |
| 281 | */ |
| 282 | int ftrace_int3_handler(struct pt_regs *regs) |
| 283 | { |
| 284 | unsigned long ip; |
| 285 | |
| 286 | if (WARN_ON_ONCE(!regs)) |
| 287 | return 0; |
| 288 | |
| 289 | ip = regs->ip - 1; |
| 290 | if (!ftrace_location(ip) && !is_ftrace_caller(ip)) |
| 291 | return 0; |
| 292 | |
| 293 | regs->ip += MCOUNT_INSN_SIZE - 1; |
| 294 | |
| 295 | return 1; |
| 296 | } |
| 297 | |
| 298 | static int ftrace_write(unsigned long ip, const char *val, int size) |
| 299 | { |
| 300 | /* |
| 301 | * On x86_64, kernel text mappings are mapped read-only with |
| 302 | * CONFIG_DEBUG_RODATA. So we use the kernel identity mapping instead |
| 303 | * of the kernel text mapping to modify the kernel text. |
| 304 | * |
| 305 | * For 32bit kernels, these mappings are same and we can use |
| 306 | * kernel identity mapping to modify code. |
| 307 | */ |
| 308 | if (within(ip, (unsigned long)_text, (unsigned long)_etext)) |
| 309 | ip = (unsigned long)__va(__pa_symbol(ip)); |
| 310 | |
| 311 | return probe_kernel_write((void *)ip, val, size); |
| 312 | } |
| 313 | |
| 314 | static int add_break(unsigned long ip, const char *old) |
| 315 | { |
| 316 | unsigned char replaced[MCOUNT_INSN_SIZE]; |
| 317 | unsigned char brk = BREAKPOINT_INSTRUCTION; |
| 318 | |
| 319 | if (probe_kernel_read(replaced, (void *)ip, MCOUNT_INSN_SIZE)) |
| 320 | return -EFAULT; |
| 321 | |
| 322 | /* Make sure it is what we expect it to be */ |
| 323 | if (memcmp(replaced, old, MCOUNT_INSN_SIZE) != 0) |
| 324 | return -EINVAL; |
| 325 | |
| 326 | if (ftrace_write(ip, &brk, 1)) |
| 327 | return -EPERM; |
| 328 | |
| 329 | return 0; |
| 330 | } |
| 331 | |
| 332 | static int add_brk_on_call(struct dyn_ftrace *rec, unsigned long addr) |
| 333 | { |
| 334 | unsigned const char *old; |
| 335 | unsigned long ip = rec->ip; |
| 336 | |
| 337 | old = ftrace_call_replace(ip, addr); |
| 338 | |
| 339 | return add_break(rec->ip, old); |
| 340 | } |
| 341 | |
| 342 | |
| 343 | static int add_brk_on_nop(struct dyn_ftrace *rec) |
| 344 | { |
| 345 | unsigned const char *old; |
| 346 | |
| 347 | old = ftrace_nop_replace(); |
| 348 | |
| 349 | return add_break(rec->ip, old); |
| 350 | } |
| 351 | |
| 352 | /* |
| 353 | * If the record has the FTRACE_FL_REGS set, that means that it |
| 354 | * wants to convert to a callback that saves all regs. If FTRACE_FL_REGS |
| 355 | * is not not set, then it wants to convert to the normal callback. |
| 356 | */ |
| 357 | static unsigned long get_ftrace_addr(struct dyn_ftrace *rec) |
| 358 | { |
| 359 | if (rec->flags & FTRACE_FL_REGS) |
| 360 | return (unsigned long)FTRACE_REGS_ADDR; |
| 361 | else |
| 362 | return (unsigned long)FTRACE_ADDR; |
| 363 | } |
| 364 | |
| 365 | /* |
| 366 | * The FTRACE_FL_REGS_EN is set when the record already points to |
| 367 | * a function that saves all the regs. Basically the '_EN' version |
| 368 | * represents the current state of the function. |
| 369 | */ |
| 370 | static unsigned long get_ftrace_old_addr(struct dyn_ftrace *rec) |
| 371 | { |
| 372 | if (rec->flags & FTRACE_FL_REGS_EN) |
| 373 | return (unsigned long)FTRACE_REGS_ADDR; |
| 374 | else |
| 375 | return (unsigned long)FTRACE_ADDR; |
| 376 | } |
| 377 | |
| 378 | static int add_breakpoints(struct dyn_ftrace *rec, int enable) |
| 379 | { |
| 380 | unsigned long ftrace_addr; |
| 381 | int ret; |
| 382 | |
| 383 | ret = ftrace_test_record(rec, enable); |
| 384 | |
| 385 | ftrace_addr = get_ftrace_addr(rec); |
| 386 | |
| 387 | switch (ret) { |
| 388 | case FTRACE_UPDATE_IGNORE: |
| 389 | return 0; |
| 390 | |
| 391 | case FTRACE_UPDATE_MAKE_CALL: |
| 392 | /* converting nop to call */ |
| 393 | return add_brk_on_nop(rec); |
| 394 | |
| 395 | case FTRACE_UPDATE_MODIFY_CALL_REGS: |
| 396 | case FTRACE_UPDATE_MODIFY_CALL: |
| 397 | ftrace_addr = get_ftrace_old_addr(rec); |
| 398 | /* fall through */ |
| 399 | case FTRACE_UPDATE_MAKE_NOP: |
| 400 | /* converting a call to a nop */ |
| 401 | return add_brk_on_call(rec, ftrace_addr); |
| 402 | } |
| 403 | return 0; |
| 404 | } |
| 405 | |
| 406 | /* |
| 407 | * On error, we need to remove breakpoints. This needs to |
| 408 | * be done caefully. If the address does not currently have a |
| 409 | * breakpoint, we know we are done. Otherwise, we look at the |
| 410 | * remaining 4 bytes of the instruction. If it matches a nop |
| 411 | * we replace the breakpoint with the nop. Otherwise we replace |
| 412 | * it with the call instruction. |
| 413 | */ |
| 414 | static int remove_breakpoint(struct dyn_ftrace *rec) |
| 415 | { |
| 416 | unsigned char ins[MCOUNT_INSN_SIZE]; |
| 417 | unsigned char brk = BREAKPOINT_INSTRUCTION; |
| 418 | const unsigned char *nop; |
| 419 | unsigned long ftrace_addr; |
| 420 | unsigned long ip = rec->ip; |
| 421 | |
| 422 | /* If we fail the read, just give up */ |
| 423 | if (probe_kernel_read(ins, (void *)ip, MCOUNT_INSN_SIZE)) |
| 424 | return -EFAULT; |
| 425 | |
| 426 | /* If this does not have a breakpoint, we are done */ |
| 427 | if (ins[0] != brk) |
| 428 | return -1; |
| 429 | |
| 430 | nop = ftrace_nop_replace(); |
| 431 | |
| 432 | /* |
| 433 | * If the last 4 bytes of the instruction do not match |
| 434 | * a nop, then we assume that this is a call to ftrace_addr. |
| 435 | */ |
| 436 | if (memcmp(&ins[1], &nop[1], MCOUNT_INSN_SIZE - 1) != 0) { |
| 437 | /* |
| 438 | * For extra paranoidism, we check if the breakpoint is on |
| 439 | * a call that would actually jump to the ftrace_addr. |
| 440 | * If not, don't touch the breakpoint, we make just create |
| 441 | * a disaster. |
| 442 | */ |
| 443 | ftrace_addr = get_ftrace_addr(rec); |
| 444 | nop = ftrace_call_replace(ip, ftrace_addr); |
| 445 | |
| 446 | if (memcmp(&ins[1], &nop[1], MCOUNT_INSN_SIZE - 1) == 0) |
| 447 | goto update; |
| 448 | |
| 449 | /* Check both ftrace_addr and ftrace_old_addr */ |
| 450 | ftrace_addr = get_ftrace_old_addr(rec); |
| 451 | nop = ftrace_call_replace(ip, ftrace_addr); |
| 452 | |
| 453 | if (memcmp(&ins[1], &nop[1], MCOUNT_INSN_SIZE - 1) != 0) |
| 454 | return -EINVAL; |
| 455 | } |
| 456 | |
| 457 | update: |
| 458 | return probe_kernel_write((void *)ip, &nop[0], 1); |
| 459 | } |
| 460 | |
| 461 | static int add_update_code(unsigned long ip, unsigned const char *new) |
| 462 | { |
| 463 | /* skip breakpoint */ |
| 464 | ip++; |
| 465 | new++; |
| 466 | if (ftrace_write(ip, new, MCOUNT_INSN_SIZE - 1)) |
| 467 | return -EPERM; |
| 468 | return 0; |
| 469 | } |
| 470 | |
| 471 | static int add_update_call(struct dyn_ftrace *rec, unsigned long addr) |
| 472 | { |
| 473 | unsigned long ip = rec->ip; |
| 474 | unsigned const char *new; |
| 475 | |
| 476 | new = ftrace_call_replace(ip, addr); |
| 477 | return add_update_code(ip, new); |
| 478 | } |
| 479 | |
| 480 | static int add_update_nop(struct dyn_ftrace *rec) |
| 481 | { |
| 482 | unsigned long ip = rec->ip; |
| 483 | unsigned const char *new; |
| 484 | |
| 485 | new = ftrace_nop_replace(); |
| 486 | return add_update_code(ip, new); |
| 487 | } |
| 488 | |
| 489 | static int add_update(struct dyn_ftrace *rec, int enable) |
| 490 | { |
| 491 | unsigned long ftrace_addr; |
| 492 | int ret; |
| 493 | |
| 494 | ret = ftrace_test_record(rec, enable); |
| 495 | |
| 496 | ftrace_addr = get_ftrace_addr(rec); |
| 497 | |
| 498 | switch (ret) { |
| 499 | case FTRACE_UPDATE_IGNORE: |
| 500 | return 0; |
| 501 | |
| 502 | case FTRACE_UPDATE_MODIFY_CALL_REGS: |
| 503 | case FTRACE_UPDATE_MODIFY_CALL: |
| 504 | case FTRACE_UPDATE_MAKE_CALL: |
| 505 | /* converting nop to call */ |
| 506 | return add_update_call(rec, ftrace_addr); |
| 507 | |
| 508 | case FTRACE_UPDATE_MAKE_NOP: |
| 509 | /* converting a call to a nop */ |
| 510 | return add_update_nop(rec); |
| 511 | } |
| 512 | |
| 513 | return 0; |
| 514 | } |
| 515 | |
| 516 | static int finish_update_call(struct dyn_ftrace *rec, unsigned long addr) |
| 517 | { |
| 518 | unsigned long ip = rec->ip; |
| 519 | unsigned const char *new; |
| 520 | |
| 521 | new = ftrace_call_replace(ip, addr); |
| 522 | |
| 523 | if (ftrace_write(ip, new, 1)) |
| 524 | return -EPERM; |
| 525 | |
| 526 | return 0; |
| 527 | } |
| 528 | |
| 529 | static int finish_update_nop(struct dyn_ftrace *rec) |
| 530 | { |
| 531 | unsigned long ip = rec->ip; |
| 532 | unsigned const char *new; |
| 533 | |
| 534 | new = ftrace_nop_replace(); |
| 535 | |
| 536 | if (ftrace_write(ip, new, 1)) |
| 537 | return -EPERM; |
| 538 | return 0; |
| 539 | } |
| 540 | |
| 541 | static int finish_update(struct dyn_ftrace *rec, int enable) |
| 542 | { |
| 543 | unsigned long ftrace_addr; |
| 544 | int ret; |
| 545 | |
| 546 | ret = ftrace_update_record(rec, enable); |
| 547 | |
| 548 | ftrace_addr = get_ftrace_addr(rec); |
| 549 | |
| 550 | switch (ret) { |
| 551 | case FTRACE_UPDATE_IGNORE: |
| 552 | return 0; |
| 553 | |
| 554 | case FTRACE_UPDATE_MODIFY_CALL_REGS: |
| 555 | case FTRACE_UPDATE_MODIFY_CALL: |
| 556 | case FTRACE_UPDATE_MAKE_CALL: |
| 557 | /* converting nop to call */ |
| 558 | return finish_update_call(rec, ftrace_addr); |
| 559 | |
| 560 | case FTRACE_UPDATE_MAKE_NOP: |
| 561 | /* converting a call to a nop */ |
| 562 | return finish_update_nop(rec); |
| 563 | } |
| 564 | |
| 565 | return 0; |
| 566 | } |
| 567 | |
| 568 | static void do_sync_core(void *data) |
| 569 | { |
| 570 | sync_core(); |
| 571 | } |
| 572 | |
| 573 | static void run_sync(void) |
| 574 | { |
| 575 | int enable_irqs = irqs_disabled(); |
| 576 | |
| 577 | /* We may be called with interrupts disbled (on bootup). */ |
| 578 | if (enable_irqs) |
| 579 | local_irq_enable(); |
| 580 | on_each_cpu(do_sync_core, NULL, 1); |
| 581 | if (enable_irqs) |
| 582 | local_irq_disable(); |
| 583 | } |
| 584 | |
| 585 | void ftrace_replace_code(int enable) |
| 586 | { |
| 587 | struct ftrace_rec_iter *iter; |
| 588 | struct dyn_ftrace *rec; |
| 589 | const char *report = "adding breakpoints"; |
| 590 | int count = 0; |
| 591 | int ret; |
| 592 | |
| 593 | for_ftrace_rec_iter(iter) { |
| 594 | rec = ftrace_rec_iter_record(iter); |
| 595 | |
| 596 | ret = add_breakpoints(rec, enable); |
| 597 | if (ret) |
| 598 | goto remove_breakpoints; |
| 599 | count++; |
| 600 | } |
| 601 | |
| 602 | run_sync(); |
| 603 | |
| 604 | report = "updating code"; |
| 605 | |
| 606 | for_ftrace_rec_iter(iter) { |
| 607 | rec = ftrace_rec_iter_record(iter); |
| 608 | |
| 609 | ret = add_update(rec, enable); |
| 610 | if (ret) |
| 611 | goto remove_breakpoints; |
| 612 | } |
| 613 | |
| 614 | run_sync(); |
| 615 | |
| 616 | report = "removing breakpoints"; |
| 617 | |
| 618 | for_ftrace_rec_iter(iter) { |
| 619 | rec = ftrace_rec_iter_record(iter); |
| 620 | |
| 621 | ret = finish_update(rec, enable); |
| 622 | if (ret) |
| 623 | goto remove_breakpoints; |
| 624 | } |
| 625 | |
| 626 | run_sync(); |
| 627 | |
| 628 | return; |
| 629 | |
| 630 | remove_breakpoints: |
| 631 | ftrace_bug(ret, rec ? rec->ip : 0); |
| 632 | printk(KERN_WARNING "Failed on %s (%d):\n", report, count); |
| 633 | for_ftrace_rec_iter(iter) { |
| 634 | rec = ftrace_rec_iter_record(iter); |
| 635 | remove_breakpoint(rec); |
| 636 | } |
| 637 | } |
| 638 | |
| 639 | static int |
| 640 | ftrace_modify_code(unsigned long ip, unsigned const char *old_code, |
| 641 | unsigned const char *new_code) |
| 642 | { |
| 643 | int ret; |
| 644 | |
| 645 | ret = add_break(ip, old_code); |
| 646 | if (ret) |
| 647 | goto out; |
| 648 | |
| 649 | run_sync(); |
| 650 | |
| 651 | ret = add_update_code(ip, new_code); |
| 652 | if (ret) |
| 653 | goto fail_update; |
| 654 | |
| 655 | run_sync(); |
| 656 | |
| 657 | ret = ftrace_write(ip, new_code, 1); |
| 658 | if (ret) { |
| 659 | ret = -EPERM; |
| 660 | goto out; |
| 661 | } |
| 662 | out: |
| 663 | run_sync(); |
| 664 | return ret; |
| 665 | |
| 666 | fail_update: |
| 667 | probe_kernel_write((void *)ip, &old_code[0], 1); |
| 668 | goto out; |
| 669 | } |
| 670 | |
| 671 | void arch_ftrace_update_code(int command) |
| 672 | { |
| 673 | /* See comment above by declaration of modifying_ftrace_code */ |
| 674 | atomic_inc(&modifying_ftrace_code); |
| 675 | |
| 676 | ftrace_modify_all_code(command); |
| 677 | |
| 678 | atomic_dec(&modifying_ftrace_code); |
| 679 | } |
| 680 | |
| 681 | int __init ftrace_dyn_arch_init(void) |
| 682 | { |
| 683 | return 0; |
| 684 | } |
| 685 | #endif |
| 686 | |
| 687 | #ifdef CONFIG_FUNCTION_GRAPH_TRACER |
| 688 | |
| 689 | #ifdef CONFIG_DYNAMIC_FTRACE |
| 690 | extern void ftrace_graph_call(void); |
| 691 | |
| 692 | static unsigned char *ftrace_jmp_replace(unsigned long ip, unsigned long addr) |
| 693 | { |
| 694 | static union ftrace_code_union calc; |
| 695 | |
| 696 | /* Jmp not a call (ignore the .e8) */ |
| 697 | calc.e8 = 0xe9; |
| 698 | calc.offset = ftrace_calc_offset(ip + MCOUNT_INSN_SIZE, addr); |
| 699 | |
| 700 | /* |
| 701 | * ftrace external locks synchronize the access to the static variable. |
| 702 | */ |
| 703 | return calc.code; |
| 704 | } |
| 705 | |
| 706 | static int ftrace_mod_jmp(unsigned long ip, void *func) |
| 707 | { |
| 708 | unsigned char *new; |
| 709 | |
| 710 | new = ftrace_jmp_replace(ip, (unsigned long)func); |
| 711 | |
| 712 | return update_ftrace_func(ip, new); |
| 713 | } |
| 714 | |
| 715 | int ftrace_enable_ftrace_graph_caller(void) |
| 716 | { |
| 717 | unsigned long ip = (unsigned long)(&ftrace_graph_call); |
| 718 | |
| 719 | return ftrace_mod_jmp(ip, &ftrace_graph_caller); |
| 720 | } |
| 721 | |
| 722 | int ftrace_disable_ftrace_graph_caller(void) |
| 723 | { |
| 724 | unsigned long ip = (unsigned long)(&ftrace_graph_call); |
| 725 | |
| 726 | return ftrace_mod_jmp(ip, &ftrace_stub); |
| 727 | } |
| 728 | |
| 729 | #endif /* !CONFIG_DYNAMIC_FTRACE */ |
| 730 | |
| 731 | /* |
| 732 | * Hook the return address and push it in the stack of return addrs |
| 733 | * in current thread info. |
| 734 | */ |
| 735 | void prepare_ftrace_return(unsigned long *parent, unsigned long self_addr, |
| 736 | unsigned long frame_pointer) |
| 737 | { |
| 738 | unsigned long old; |
| 739 | int faulted; |
| 740 | struct ftrace_graph_ent trace; |
| 741 | unsigned long return_hooker = (unsigned long) |
| 742 | &return_to_handler; |
| 743 | |
| 744 | /* |
| 745 | * When resuming from suspend-to-ram, this function can be indirectly |
| 746 | * called from early CPU startup code while the CPU is in real mode, |
| 747 | * which would fail miserably. Make sure the stack pointer is a |
| 748 | * virtual address. |
| 749 | * |
| 750 | * This check isn't as accurate as virt_addr_valid(), but it should be |
| 751 | * good enough for this purpose, and it's fast. |
| 752 | */ |
| 753 | if (unlikely((long)__builtin_frame_address(0) >= 0)) |
| 754 | return; |
| 755 | |
| 756 | if (unlikely(atomic_read(¤t->tracing_graph_pause))) |
| 757 | return; |
| 758 | |
| 759 | /* |
| 760 | * Protect against fault, even if it shouldn't |
| 761 | * happen. This tool is too much intrusive to |
| 762 | * ignore such a protection. |
| 763 | */ |
| 764 | asm volatile( |
| 765 | "1: " _ASM_MOV " (%[parent]), %[old]\n" |
| 766 | "2: " _ASM_MOV " %[return_hooker], (%[parent])\n" |
| 767 | " movl $0, %[faulted]\n" |
| 768 | "3:\n" |
| 769 | |
| 770 | ".section .fixup, \"ax\"\n" |
| 771 | "4: movl $1, %[faulted]\n" |
| 772 | " jmp 3b\n" |
| 773 | ".previous\n" |
| 774 | |
| 775 | _ASM_EXTABLE(1b, 4b) |
| 776 | _ASM_EXTABLE(2b, 4b) |
| 777 | |
| 778 | : [old] "=&r" (old), [faulted] "=r" (faulted) |
| 779 | : [parent] "r" (parent), [return_hooker] "r" (return_hooker) |
| 780 | : "memory" |
| 781 | ); |
| 782 | |
| 783 | if (unlikely(faulted)) { |
| 784 | ftrace_graph_stop(); |
| 785 | WARN_ON(1); |
| 786 | return; |
| 787 | } |
| 788 | |
| 789 | trace.func = self_addr; |
| 790 | trace.depth = current->curr_ret_stack + 1; |
| 791 | |
| 792 | /* Only trace if the calling function expects to */ |
| 793 | if (!ftrace_graph_entry(&trace)) { |
| 794 | *parent = old; |
| 795 | return; |
| 796 | } |
| 797 | |
| 798 | if (ftrace_push_return_trace(old, self_addr, &trace.depth, |
| 799 | frame_pointer) == -EBUSY) { |
| 800 | *parent = old; |
| 801 | return; |
| 802 | } |
| 803 | } |
| 804 | #endif /* CONFIG_FUNCTION_GRAPH_TRACER */ |