| 1 | /* |
| 2 | * Copyright (c) 2005-2012 Apple Inc. All rights reserved. |
| 3 | * |
| 4 | * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ |
| 5 | * |
| 6 | * This file contains Original Code and/or Modifications of Original Code |
| 7 | * as defined in and that are subject to the Apple Public Source License |
| 8 | * Version 2.0 (the 'License'). You may not use this file except in |
| 9 | * compliance with the License. The rights granted to you under the License |
| 10 | * may not be used to create, or enable the creation or redistribution of, |
| 11 | * unlawful or unlicensed copies of an Apple operating system, or to |
| 12 | * circumvent, violate, or enable the circumvention or violation of, any |
| 13 | * terms of an Apple operating system software license agreement. |
| 14 | * |
| 15 | * Please obtain a copy of the License at |
| 16 | * http://www.opensource.apple.com/apsl/ and read it before using this file. |
| 17 | * |
| 18 | * The Original Code and all software distributed under the License are |
| 19 | * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER |
| 20 | * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, |
| 21 | * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY, |
| 22 | * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT. |
| 23 | * Please see the License for the specific language governing rights and |
| 24 | * limitations under the License. |
| 25 | * |
| 26 | * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ |
| 27 | */ |
| 28 | |
| 29 | #define MACH__POSIX_C_SOURCE_PRIVATE 1 /* pulls in suitable savearea from mach/ppc/thread_status.h */ |
| 30 | #include <kern/thread.h> |
| 31 | #include <mach/thread_status.h> |
| 32 | |
| 33 | typedef x86_saved_state_t savearea_t; |
| 34 | |
| 35 | #include <stdarg.h> |
| 36 | #include <string.h> |
| 37 | #include <sys/malloc.h> |
| 38 | #include <sys/time.h> |
| 39 | #include <sys/systm.h> |
| 40 | #include <sys/proc.h> |
| 41 | #include <sys/proc_internal.h> |
| 42 | #include <sys/kauth.h> |
| 43 | #include <sys/dtrace.h> |
| 44 | #include <sys/dtrace_impl.h> |
| 45 | #include <libkern/OSAtomic.h> |
| 46 | #include <kern/thread_call.h> |
| 47 | #include <kern/task.h> |
| 48 | #include <kern/sched_prim.h> |
| 49 | #include <miscfs/devfs/devfs.h> |
| 50 | #include <mach/vm_param.h> |
| 51 | #include <machine/pal_routines.h> |
| 52 | #include <i386/mp.h> |
| 53 | |
| 54 | /* |
| 55 | * APPLE NOTE: The regmap is used to decode which 64bit uregs[] register |
| 56 | * is being accessed when passed the 32bit uregs[] constant (based on |
| 57 | * the reg.d translator file). The dtrace_getreg() is smart enough to handle |
| 58 | * the register mappings. The register set definitions are the same as |
| 59 | * those used by the fasttrap_getreg code. |
| 60 | */ |
| 61 | #include "fasttrap_regset.h" |
| 62 | static const uint8_t regmap[19] = { |
| 63 | REG_GS, /* GS */ |
| 64 | REG_FS, /* FS */ |
| 65 | REG_ES, /* ES */ |
| 66 | REG_DS, /* DS */ |
| 67 | REG_RDI, /* EDI */ |
| 68 | REG_RSI, /* ESI */ |
| 69 | REG_RBP, /* EBP, REG_FP */ |
| 70 | REG_RSP, /* ESP */ |
| 71 | REG_RBX, /* EBX */ |
| 72 | REG_RDX, /* EDX, REG_R1 */ |
| 73 | REG_RCX, /* ECX */ |
| 74 | REG_RAX, /* EAX, REG_R0 */ |
| 75 | REG_TRAPNO, /* TRAPNO */ |
| 76 | REG_ERR, /* ERR */ |
| 77 | REG_RIP, /* EIP, REG_PC */ |
| 78 | REG_CS, /* CS */ |
| 79 | REG_RFL, /* EFL, REG_PS */ |
| 80 | REG_RSP, /* UESP, REG_SP */ |
| 81 | REG_SS /* SS */ |
| 82 | }; |
| 83 | |
| 84 | extern dtrace_id_t dtrace_probeid_error; /* special ERROR probe */ |
| 85 | |
| 86 | void |
| 87 | dtrace_probe_error(dtrace_state_t *state, dtrace_epid_t epid, int which, |
| 88 | int fltoffs, int fault, uint64_t illval) |
| 89 | { |
| 90 | /* |
| 91 | * For the case of the error probe firing lets |
| 92 | * stash away "illval" here, and special-case retrieving it in DIF_VARIABLE_ARG. |
| 93 | */ |
| 94 | state->dts_arg_error_illval = illval; |
| 95 | dtrace_probe( dtrace_probeid_error, (uint64_t)(uintptr_t)state, epid, which, fltoffs, fault ); |
| 96 | } |
| 97 | |
| 98 | /* |
| 99 | * Atomicity and synchronization |
| 100 | */ |
| 101 | void |
| 102 | dtrace_membar_producer(void) |
| 103 | { |
| 104 | __asm__ volatile("sfence" ); |
| 105 | } |
| 106 | |
| 107 | void |
| 108 | dtrace_membar_consumer(void) |
| 109 | { |
| 110 | __asm__ volatile("lfence" ); |
| 111 | } |
| 112 | |
| 113 | /* |
| 114 | * Interrupt manipulation |
| 115 | * XXX dtrace_getipl() can be called from probe context. |
| 116 | */ |
| 117 | int |
| 118 | dtrace_getipl(void) |
| 119 | { |
| 120 | /* |
| 121 | * XXX Drat, get_interrupt_level is MACH_KERNEL_PRIVATE |
| 122 | * in osfmk/kern/cpu_data.h |
| 123 | */ |
| 124 | /* return get_interrupt_level(); */ |
| 125 | return (ml_at_interrupt_context() ? 1: 0); |
| 126 | } |
| 127 | |
| 128 | /* |
| 129 | * MP coordination |
| 130 | */ |
| 131 | typedef struct xcArg { |
| 132 | processorid_t cpu; |
| 133 | dtrace_xcall_t f; |
| 134 | void *arg; |
| 135 | } xcArg_t; |
| 136 | |
| 137 | static void |
| 138 | xcRemote( void *foo ) |
| 139 | { |
| 140 | xcArg_t *pArg = (xcArg_t *)foo; |
| 141 | |
| 142 | if ( pArg->cpu == CPU->cpu_id || pArg->cpu == DTRACE_CPUALL ) { |
| 143 | (pArg->f)(pArg->arg); |
| 144 | } |
| 145 | } |
| 146 | |
| 147 | |
| 148 | /* |
| 149 | * dtrace_xcall() is not called from probe context. |
| 150 | */ |
| 151 | void |
| 152 | dtrace_xcall(processorid_t cpu, dtrace_xcall_t f, void *arg) |
| 153 | { |
| 154 | xcArg_t xcArg; |
| 155 | |
| 156 | xcArg.cpu = cpu; |
| 157 | xcArg.f = f; |
| 158 | xcArg.arg = arg; |
| 159 | |
| 160 | if (cpu == DTRACE_CPUALL) { |
| 161 | mp_cpus_call (CPUMASK_ALL, ASYNC, xcRemote, (void*)&xcArg); |
| 162 | } |
| 163 | else { |
| 164 | mp_cpus_call (cpu_to_cpumask((cpu_t)cpu), ASYNC, xcRemote, (void*)&xcArg); |
| 165 | } |
| 166 | } |
| 167 | |
| 168 | /* |
| 169 | * Initialization |
| 170 | */ |
| 171 | void |
| 172 | dtrace_isa_init(void) |
| 173 | { |
| 174 | return; |
| 175 | } |
| 176 | |
| 177 | /* |
| 178 | * Runtime and ABI |
| 179 | */ |
| 180 | uint64_t |
| 181 | dtrace_getreg(struct regs *savearea, uint_t reg) |
| 182 | { |
| 183 | boolean_t is64Bit = proc_is64bit(current_proc()); |
| 184 | x86_saved_state_t *regs = (x86_saved_state_t *)savearea; |
| 185 | |
| 186 | if (regs == NULL) { |
| 187 | DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); |
| 188 | return (0); |
| 189 | } |
| 190 | |
| 191 | if (is64Bit) { |
| 192 | if (reg <= SS) { |
| 193 | reg = regmap[reg]; |
| 194 | } else { |
| 195 | reg -= (SS + 1); |
| 196 | } |
| 197 | |
| 198 | switch (reg) { |
| 199 | case REG_RDI: |
| 200 | return (uint64_t)(regs->ss_64.rdi); |
| 201 | case REG_RSI: |
| 202 | return (uint64_t)(regs->ss_64.rsi); |
| 203 | case REG_RDX: |
| 204 | return (uint64_t)(regs->ss_64.rdx); |
| 205 | case REG_RCX: |
| 206 | return (uint64_t)(regs->ss_64.rcx); |
| 207 | case REG_R8: |
| 208 | return (uint64_t)(regs->ss_64.r8); |
| 209 | case REG_R9: |
| 210 | return (uint64_t)(regs->ss_64.r9); |
| 211 | case REG_RAX: |
| 212 | return (uint64_t)(regs->ss_64.rax); |
| 213 | case REG_RBX: |
| 214 | return (uint64_t)(regs->ss_64.rbx); |
| 215 | case REG_RBP: |
| 216 | return (uint64_t)(regs->ss_64.rbp); |
| 217 | case REG_R10: |
| 218 | return (uint64_t)(regs->ss_64.r10); |
| 219 | case REG_R11: |
| 220 | return (uint64_t)(regs->ss_64.r11); |
| 221 | case REG_R12: |
| 222 | return (uint64_t)(regs->ss_64.r12); |
| 223 | case REG_R13: |
| 224 | return (uint64_t)(regs->ss_64.r13); |
| 225 | case REG_R14: |
| 226 | return (uint64_t)(regs->ss_64.r14); |
| 227 | case REG_R15: |
| 228 | return (uint64_t)(regs->ss_64.r15); |
| 229 | case REG_FS: |
| 230 | return (uint64_t)(regs->ss_64.fs); |
| 231 | case REG_GS: |
| 232 | return (uint64_t)(regs->ss_64.gs); |
| 233 | case REG_TRAPNO: |
| 234 | return (uint64_t)(regs->ss_64.isf.trapno); |
| 235 | case REG_ERR: |
| 236 | return (uint64_t)(regs->ss_64.isf.err); |
| 237 | case REG_RIP: |
| 238 | return (uint64_t)(regs->ss_64.isf.rip); |
| 239 | case REG_CS: |
| 240 | return (uint64_t)(regs->ss_64.isf.cs); |
| 241 | case REG_SS: |
| 242 | return (uint64_t)(regs->ss_64.isf.ss); |
| 243 | case REG_RFL: |
| 244 | return (uint64_t)(regs->ss_64.isf.rflags); |
| 245 | case REG_RSP: |
| 246 | return (uint64_t)(regs->ss_64.isf.rsp); |
| 247 | case REG_DS: |
| 248 | case REG_ES: |
| 249 | default: |
| 250 | DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); |
| 251 | return (0); |
| 252 | } |
| 253 | |
| 254 | } else { /* is 32bit user */ |
| 255 | /* beyond register SS */ |
| 256 | if (reg > x86_SAVED_STATE32_COUNT - 1) { |
| 257 | DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); |
| 258 | return (0); |
| 259 | } |
| 260 | return (uint64_t)((unsigned int *)(&(regs->ss_32.gs)))[reg]; |
| 261 | } |
| 262 | } |
| 263 | |
| 264 | #define RETURN_OFFSET 4 |
| 265 | #define RETURN_OFFSET64 8 |
| 266 | |
| 267 | static int |
| 268 | dtrace_getustack_common(uint64_t *pcstack, int pcstack_limit, user_addr_t pc, |
| 269 | user_addr_t sp) |
| 270 | { |
| 271 | #if 0 |
| 272 | volatile uint16_t *flags = |
| 273 | (volatile uint16_t *)&cpu_core[CPU->cpu_id].cpuc_dtrace_flags; |
| 274 | |
| 275 | uintptr_t oldcontext = lwp->lwp_oldcontext; /* XXX signal stack crawl */ |
| 276 | size_t s1, s2; |
| 277 | #endif |
| 278 | int ret = 0; |
| 279 | boolean_t is64Bit = proc_is64bit(current_proc()); |
| 280 | |
| 281 | ASSERT(pcstack == NULL || pcstack_limit > 0); |
| 282 | |
| 283 | #if 0 /* XXX signal stack crawl */ |
| 284 | if (p->p_model == DATAMODEL_NATIVE) { |
| 285 | s1 = sizeof (struct frame) + 2 * sizeof (long); |
| 286 | s2 = s1 + sizeof (siginfo_t); |
| 287 | } else { |
| 288 | s1 = sizeof (struct frame32) + 3 * sizeof (int); |
| 289 | s2 = s1 + sizeof (siginfo32_t); |
| 290 | } |
| 291 | #endif |
| 292 | |
| 293 | while (pc != 0) { |
| 294 | ret++; |
| 295 | if (pcstack != NULL) { |
| 296 | *pcstack++ = (uint64_t)pc; |
| 297 | pcstack_limit--; |
| 298 | if (pcstack_limit <= 0) |
| 299 | break; |
| 300 | } |
| 301 | |
| 302 | if (sp == 0) |
| 303 | break; |
| 304 | |
| 305 | #if 0 /* XXX signal stack crawl */ |
| 306 | if (oldcontext == sp + s1 || oldcontext == sp + s2) { |
| 307 | if (p->p_model == DATAMODEL_NATIVE) { |
| 308 | ucontext_t *ucp = (ucontext_t *)oldcontext; |
| 309 | greg_t *gregs = ucp->uc_mcontext.gregs; |
| 310 | |
| 311 | sp = dtrace_fulword(&gregs[REG_FP]); |
| 312 | pc = dtrace_fulword(&gregs[REG_PC]); |
| 313 | |
| 314 | oldcontext = dtrace_fulword(&ucp->uc_link); |
| 315 | } else { |
| 316 | ucontext32_t *ucp = (ucontext32_t *)oldcontext; |
| 317 | greg32_t *gregs = ucp->uc_mcontext.gregs; |
| 318 | |
| 319 | sp = dtrace_fuword32(&gregs[EBP]); |
| 320 | pc = dtrace_fuword32(&gregs[EIP]); |
| 321 | |
| 322 | oldcontext = dtrace_fuword32(&ucp->uc_link); |
| 323 | } |
| 324 | } |
| 325 | else |
| 326 | #endif |
| 327 | { |
| 328 | if (is64Bit) { |
| 329 | pc = dtrace_fuword64((sp + RETURN_OFFSET64)); |
| 330 | sp = dtrace_fuword64(sp); |
| 331 | } else { |
| 332 | pc = dtrace_fuword32((sp + RETURN_OFFSET)); |
| 333 | sp = dtrace_fuword32(sp); |
| 334 | } |
| 335 | } |
| 336 | |
| 337 | #if 0 /* XXX */ |
| 338 | /* |
| 339 | * This is totally bogus: if we faulted, we're going to clear |
| 340 | * the fault and break. This is to deal with the apparently |
| 341 | * broken Java stacks on x86. |
| 342 | */ |
| 343 | if (*flags & CPU_DTRACE_FAULT) { |
| 344 | *flags &= ~CPU_DTRACE_FAULT; |
| 345 | break; |
| 346 | } |
| 347 | #endif |
| 348 | } |
| 349 | |
| 350 | return (ret); |
| 351 | } |
| 352 | |
| 353 | |
| 354 | /* |
| 355 | * The return value indicates if we've modified the stack. |
| 356 | */ |
| 357 | static int |
| 358 | dtrace_adjust_stack(uint64_t **pcstack, int *pcstack_limit, user_addr_t *pc, |
| 359 | user_addr_t sp) |
| 360 | { |
| 361 | int64_t missing_tos; |
| 362 | int rc = 0; |
| 363 | boolean_t is64Bit = proc_is64bit(current_proc()); |
| 364 | |
| 365 | ASSERT(pc != NULL); |
| 366 | |
| 367 | if (DTRACE_CPUFLAG_ISSET(CPU_DTRACE_ENTRY)) { |
| 368 | /* |
| 369 | * If we found ourselves in an entry probe, the frame pointer has not |
| 370 | * yet been pushed (that happens in the |
| 371 | * function prologue). The best approach is to |
| 372 | * add the current pc as a missing top of stack, |
| 373 | * and back the pc up to the caller, which is stored at the |
| 374 | * current stack pointer address since the call |
| 375 | * instruction puts it there right before |
| 376 | * the branch. |
| 377 | */ |
| 378 | |
| 379 | missing_tos = *pc; |
| 380 | |
| 381 | if (is64Bit) |
| 382 | *pc = dtrace_fuword64(sp); |
| 383 | else |
| 384 | *pc = dtrace_fuword32(sp); |
| 385 | } else { |
| 386 | /* |
| 387 | * We might have a top of stack override, in which case we just |
| 388 | * add that frame without question to the top. This |
| 389 | * happens in return probes where you have a valid |
| 390 | * frame pointer, but it's for the callers frame |
| 391 | * and you'd like to add the pc of the return site |
| 392 | * to the frame. |
| 393 | */ |
| 394 | missing_tos = cpu_core[CPU->cpu_id].cpuc_missing_tos; |
| 395 | } |
| 396 | |
| 397 | if (missing_tos != 0) { |
| 398 | if (pcstack != NULL && pcstack_limit != NULL) { |
| 399 | /* |
| 400 | * If the missing top of stack has been filled out, then |
| 401 | * we add it and adjust the size. |
| 402 | */ |
| 403 | *(*pcstack)++ = missing_tos; |
| 404 | (*pcstack_limit)--; |
| 405 | } |
| 406 | /* |
| 407 | * return 1 because we would have changed the |
| 408 | * stack whether or not it was passed in. This |
| 409 | * ensures the stack count is correct |
| 410 | */ |
| 411 | rc = 1; |
| 412 | } |
| 413 | return rc; |
| 414 | } |
| 415 | |
| 416 | void |
| 417 | dtrace_getupcstack(uint64_t *pcstack, int pcstack_limit) |
| 418 | { |
| 419 | thread_t thread = current_thread(); |
| 420 | x86_saved_state_t *regs; |
| 421 | user_addr_t pc, sp, fp; |
| 422 | volatile uint16_t *flags = |
| 423 | (volatile uint16_t *)&cpu_core[CPU->cpu_id].cpuc_dtrace_flags; |
| 424 | int n; |
| 425 | boolean_t is64Bit = proc_is64bit(current_proc()); |
| 426 | |
| 427 | if (*flags & CPU_DTRACE_FAULT) |
| 428 | return; |
| 429 | |
| 430 | if (pcstack_limit <= 0) |
| 431 | return; |
| 432 | |
| 433 | /* |
| 434 | * If there's no user context we still need to zero the stack. |
| 435 | */ |
| 436 | if (thread == NULL) |
| 437 | goto zero; |
| 438 | |
| 439 | pal_register_cache_state(thread, VALID); |
| 440 | regs = (x86_saved_state_t *)find_user_regs(thread); |
| 441 | if (regs == NULL) |
| 442 | goto zero; |
| 443 | |
| 444 | *pcstack++ = (uint64_t)dtrace_proc_selfpid(); |
| 445 | pcstack_limit--; |
| 446 | |
| 447 | if (pcstack_limit <= 0) |
| 448 | return; |
| 449 | |
| 450 | if (is64Bit) { |
| 451 | pc = regs->ss_64.isf.rip; |
| 452 | sp = regs->ss_64.isf.rsp; |
| 453 | fp = regs->ss_64.rbp; |
| 454 | } else { |
| 455 | pc = regs->ss_32.eip; |
| 456 | sp = regs->ss_32.uesp; |
| 457 | fp = regs->ss_32.ebp; |
| 458 | } |
| 459 | |
| 460 | /* |
| 461 | * The return value indicates if we've modified the stack. |
| 462 | * Since there is nothing else to fix up in either case, |
| 463 | * we can safely ignore it here. |
| 464 | */ |
| 465 | (void)dtrace_adjust_stack(&pcstack, &pcstack_limit, &pc, sp); |
| 466 | |
| 467 | if(pcstack_limit <= 0) |
| 468 | return; |
| 469 | |
| 470 | /* |
| 471 | * Note that unlike ppc, the x86 code does not use |
| 472 | * CPU_DTRACE_USTACK_FP. This is because x86 always |
| 473 | * traces from the fp, even in syscall/profile/fbt |
| 474 | * providers. |
| 475 | */ |
| 476 | n = dtrace_getustack_common(pcstack, pcstack_limit, pc, fp); |
| 477 | ASSERT(n >= 0); |
| 478 | ASSERT(n <= pcstack_limit); |
| 479 | |
| 480 | pcstack += n; |
| 481 | pcstack_limit -= n; |
| 482 | |
| 483 | zero: |
| 484 | while (pcstack_limit-- > 0) |
| 485 | *pcstack++ = 0; |
| 486 | } |
| 487 | |
| 488 | int |
| 489 | dtrace_getustackdepth(void) |
| 490 | { |
| 491 | thread_t thread = current_thread(); |
| 492 | x86_saved_state_t *regs; |
| 493 | user_addr_t pc, sp, fp; |
| 494 | int n = 0; |
| 495 | boolean_t is64Bit = proc_is64bit(current_proc()); |
| 496 | |
| 497 | if (thread == NULL) |
| 498 | return 0; |
| 499 | |
| 500 | if (DTRACE_CPUFLAG_ISSET(CPU_DTRACE_FAULT)) |
| 501 | return (-1); |
| 502 | |
| 503 | pal_register_cache_state(thread, VALID); |
| 504 | regs = (x86_saved_state_t *)find_user_regs(thread); |
| 505 | if (regs == NULL) |
| 506 | return 0; |
| 507 | |
| 508 | if (is64Bit) { |
| 509 | pc = regs->ss_64.isf.rip; |
| 510 | sp = regs->ss_64.isf.rsp; |
| 511 | fp = regs->ss_64.rbp; |
| 512 | } else { |
| 513 | pc = regs->ss_32.eip; |
| 514 | sp = regs->ss_32.uesp; |
| 515 | fp = regs->ss_32.ebp; |
| 516 | } |
| 517 | |
| 518 | if (dtrace_adjust_stack(NULL, NULL, &pc, sp) == 1) { |
| 519 | /* |
| 520 | * we would have adjusted the stack if we had |
| 521 | * supplied one (that is what rc == 1 means). |
| 522 | * Also, as a side effect, the pc might have |
| 523 | * been fixed up, which is good for calling |
| 524 | * in to dtrace_getustack_common. |
| 525 | */ |
| 526 | n++; |
| 527 | } |
| 528 | |
| 529 | /* |
| 530 | * Note that unlike ppc, the x86 code does not use |
| 531 | * CPU_DTRACE_USTACK_FP. This is because x86 always |
| 532 | * traces from the fp, even in syscall/profile/fbt |
| 533 | * providers. |
| 534 | */ |
| 535 | |
| 536 | n += dtrace_getustack_common(NULL, 0, pc, fp); |
| 537 | |
| 538 | return (n); |
| 539 | } |
| 540 | |
| 541 | void |
| 542 | dtrace_getufpstack(uint64_t *pcstack, uint64_t *fpstack, int pcstack_limit) |
| 543 | { |
| 544 | thread_t thread = current_thread(); |
| 545 | savearea_t *regs; |
| 546 | user_addr_t pc, sp; |
| 547 | volatile uint16_t *flags = |
| 548 | (volatile uint16_t *)&cpu_core[CPU->cpu_id].cpuc_dtrace_flags; |
| 549 | #if 0 |
| 550 | uintptr_t oldcontext; |
| 551 | size_t s1, s2; |
| 552 | #endif |
| 553 | boolean_t is64Bit = proc_is64bit(current_proc()); |
| 554 | |
| 555 | if (*flags & CPU_DTRACE_FAULT) |
| 556 | return; |
| 557 | |
| 558 | if (pcstack_limit <= 0) |
| 559 | return; |
| 560 | |
| 561 | /* |
| 562 | * If there's no user context we still need to zero the stack. |
| 563 | */ |
| 564 | if (thread == NULL) |
| 565 | goto zero; |
| 566 | |
| 567 | regs = (savearea_t *)find_user_regs(thread); |
| 568 | if (regs == NULL) |
| 569 | goto zero; |
| 570 | |
| 571 | *pcstack++ = (uint64_t)dtrace_proc_selfpid(); |
| 572 | pcstack_limit--; |
| 573 | |
| 574 | if (pcstack_limit <= 0) |
| 575 | return; |
| 576 | |
| 577 | pc = regs->ss_32.eip; |
| 578 | sp = regs->ss_32.ebp; |
| 579 | |
| 580 | #if 0 /* XXX signal stack crawl */ |
| 581 | oldcontext = lwp->lwp_oldcontext; |
| 582 | |
| 583 | if (p->p_model == DATAMODEL_NATIVE) { |
| 584 | s1 = sizeof (struct frame) + 2 * sizeof (long); |
| 585 | s2 = s1 + sizeof (siginfo_t); |
| 586 | } else { |
| 587 | s1 = sizeof (struct frame32) + 3 * sizeof (int); |
| 588 | s2 = s1 + sizeof (siginfo32_t); |
| 589 | } |
| 590 | #endif |
| 591 | |
| 592 | if(dtrace_adjust_stack(&pcstack, &pcstack_limit, &pc, sp) == 1) { |
| 593 | /* |
| 594 | * we made a change. |
| 595 | */ |
| 596 | *fpstack++ = 0; |
| 597 | if (pcstack_limit <= 0) |
| 598 | return; |
| 599 | } |
| 600 | |
| 601 | while (pc != 0) { |
| 602 | *pcstack++ = (uint64_t)pc; |
| 603 | *fpstack++ = sp; |
| 604 | pcstack_limit--; |
| 605 | if (pcstack_limit <= 0) |
| 606 | break; |
| 607 | |
| 608 | if (sp == 0) |
| 609 | break; |
| 610 | |
| 611 | #if 0 /* XXX signal stack crawl */ |
| 612 | if (oldcontext == sp + s1 || oldcontext == sp + s2) { |
| 613 | if (p->p_model == DATAMODEL_NATIVE) { |
| 614 | ucontext_t *ucp = (ucontext_t *)oldcontext; |
| 615 | greg_t *gregs = ucp->uc_mcontext.gregs; |
| 616 | |
| 617 | sp = dtrace_fulword(&gregs[REG_FP]); |
| 618 | pc = dtrace_fulword(&gregs[REG_PC]); |
| 619 | |
| 620 | oldcontext = dtrace_fulword(&ucp->uc_link); |
| 621 | } else { |
| 622 | ucontext_t *ucp = (ucontext_t *)oldcontext; |
| 623 | greg_t *gregs = ucp->uc_mcontext.gregs; |
| 624 | |
| 625 | sp = dtrace_fuword32(&gregs[EBP]); |
| 626 | pc = dtrace_fuword32(&gregs[EIP]); |
| 627 | |
| 628 | oldcontext = dtrace_fuword32(&ucp->uc_link); |
| 629 | } |
| 630 | } |
| 631 | else |
| 632 | #endif |
| 633 | { |
| 634 | if (is64Bit) { |
| 635 | pc = dtrace_fuword64((sp + RETURN_OFFSET64)); |
| 636 | sp = dtrace_fuword64(sp); |
| 637 | } else { |
| 638 | pc = dtrace_fuword32((sp + RETURN_OFFSET)); |
| 639 | sp = dtrace_fuword32(sp); |
| 640 | } |
| 641 | } |
| 642 | |
| 643 | #if 0 /* XXX */ |
| 644 | /* |
| 645 | * This is totally bogus: if we faulted, we're going to clear |
| 646 | * the fault and break. This is to deal with the apparently |
| 647 | * broken Java stacks on x86. |
| 648 | */ |
| 649 | if (*flags & CPU_DTRACE_FAULT) { |
| 650 | *flags &= ~CPU_DTRACE_FAULT; |
| 651 | break; |
| 652 | } |
| 653 | #endif |
| 654 | } |
| 655 | |
| 656 | zero: |
| 657 | while (pcstack_limit-- > 0) |
| 658 | *pcstack++ = 0; |
| 659 | } |
| 660 | |
| 661 | void |
| 662 | dtrace_getpcstack(pc_t *pcstack, int pcstack_limit, int aframes, |
| 663 | uint32_t *intrpc) |
| 664 | { |
| 665 | struct frame *fp = (struct frame *)__builtin_frame_address(0); |
| 666 | struct frame *nextfp, *minfp, *stacktop; |
| 667 | int depth = 0; |
| 668 | int last = 0; |
| 669 | uintptr_t pc; |
| 670 | uintptr_t caller = CPU->cpu_dtrace_caller; |
| 671 | int on_intr; |
| 672 | |
| 673 | if ((on_intr = CPU_ON_INTR(CPU)) != 0) |
| 674 | stacktop = (struct frame *)dtrace_get_cpu_int_stack_top(); |
| 675 | else |
| 676 | stacktop = (struct frame *)(dtrace_get_kernel_stack(current_thread()) + kernel_stack_size); |
| 677 | |
| 678 | minfp = fp; |
| 679 | |
| 680 | aframes++; |
| 681 | |
| 682 | if (intrpc != NULL && depth < pcstack_limit) |
| 683 | pcstack[depth++] = (pc_t)intrpc; |
| 684 | |
| 685 | while (depth < pcstack_limit) { |
| 686 | nextfp = *(struct frame **)fp; |
| 687 | pc = *(uintptr_t *)(((uintptr_t)fp) + RETURN_OFFSET64); |
| 688 | |
| 689 | if (nextfp <= minfp || nextfp >= stacktop) { |
| 690 | if (on_intr) { |
| 691 | /* |
| 692 | * Hop from interrupt stack to thread stack. |
| 693 | */ |
| 694 | vm_offset_t kstack_base = dtrace_get_kernel_stack(current_thread()); |
| 695 | |
| 696 | minfp = (struct frame *)kstack_base; |
| 697 | stacktop = (struct frame *)(kstack_base + kernel_stack_size); |
| 698 | |
| 699 | on_intr = 0; |
| 700 | continue; |
| 701 | } |
| 702 | /* |
| 703 | * This is the last frame we can process; indicate |
| 704 | * that we should return after processing this frame. |
| 705 | */ |
| 706 | last = 1; |
| 707 | } |
| 708 | |
| 709 | if (aframes > 0) { |
| 710 | if (--aframes == 0 && caller != 0) { |
| 711 | /* |
| 712 | * We've just run out of artificial frames, |
| 713 | * and we have a valid caller -- fill it in |
| 714 | * now. |
| 715 | */ |
| 716 | ASSERT(depth < pcstack_limit); |
| 717 | pcstack[depth++] = (pc_t)caller; |
| 718 | caller = 0; |
| 719 | } |
| 720 | } else { |
| 721 | if (depth < pcstack_limit) |
| 722 | pcstack[depth++] = (pc_t)pc; |
| 723 | } |
| 724 | |
| 725 | if (last) { |
| 726 | while (depth < pcstack_limit) |
| 727 | pcstack[depth++] = 0; |
| 728 | return; |
| 729 | } |
| 730 | |
| 731 | fp = nextfp; |
| 732 | minfp = fp; |
| 733 | } |
| 734 | } |
| 735 | |
| 736 | struct frame { |
| 737 | struct frame *backchain; |
| 738 | uintptr_t retaddr; |
| 739 | }; |
| 740 | |
| 741 | uint64_t |
| 742 | dtrace_getarg(int arg, int aframes, dtrace_mstate_t *mstate, dtrace_vstate_t *vstate) |
| 743 | { |
| 744 | uint64_t val = 0; |
| 745 | struct frame *fp = (struct frame *)__builtin_frame_address(0); |
| 746 | uintptr_t *stack; |
| 747 | uintptr_t pc; |
| 748 | int i; |
| 749 | |
| 750 | |
| 751 | /* |
| 752 | * A total of 6 arguments are passed via registers; any argument with |
| 753 | * index of 5 or lower is therefore in a register. |
| 754 | */ |
| 755 | int inreg = 5; |
| 756 | |
| 757 | for (i = 1; i <= aframes; i++) { |
| 758 | fp = fp->backchain; |
| 759 | pc = fp->retaddr; |
| 760 | |
| 761 | if (dtrace_invop_callsite_pre != NULL |
| 762 | && pc > (uintptr_t)dtrace_invop_callsite_pre |
| 763 | && pc <= (uintptr_t)dtrace_invop_callsite_post) { |
| 764 | /* |
| 765 | * In the case of x86_64, we will use the pointer to the |
| 766 | * save area structure that was pushed when we took the |
| 767 | * trap. To get this structure, we must increment |
| 768 | * beyond the frame structure. If the |
| 769 | * argument that we're seeking is passed on the stack, |
| 770 | * we'll pull the true stack pointer out of the saved |
| 771 | * registers and decrement our argument by the number |
| 772 | * of arguments passed in registers; if the argument |
| 773 | * we're seeking is passed in regsiters, we can just |
| 774 | * load it directly. |
| 775 | */ |
| 776 | |
| 777 | /* fp points to frame of dtrace_invop() activation. */ |
| 778 | fp = fp->backchain; /* to fbt_perfcallback() activation. */ |
| 779 | fp = fp->backchain; /* to kernel_trap() activation. */ |
| 780 | fp = fp->backchain; /* to trap_from_kernel() activation. */ |
| 781 | |
| 782 | x86_saved_state_t *tagged_regs = (x86_saved_state_t *)&fp[1]; |
| 783 | x86_saved_state64_t *saved_state = saved_state64(tagged_regs); |
| 784 | |
| 785 | if (arg <= inreg) { |
| 786 | stack = (uintptr_t *)(void*)&saved_state->rdi; |
| 787 | } else { |
| 788 | fp = (struct frame *)(saved_state->isf.rsp); |
| 789 | stack = (uintptr_t *)&fp[1]; /* Find marshalled |
| 790 | arguments */ |
| 791 | arg -= inreg + 1; |
| 792 | } |
| 793 | goto load; |
| 794 | } |
| 795 | } |
| 796 | |
| 797 | /* |
| 798 | * We know that we did not come through a trap to get into |
| 799 | * dtrace_probe() -- We arrive here when the provider has |
| 800 | * called dtrace_probe() directly. |
| 801 | * The probe ID is the first argument to dtrace_probe(). |
| 802 | * We must advance beyond that to get the argX. |
| 803 | */ |
| 804 | arg++; /* Advance past probeID */ |
| 805 | |
| 806 | if (arg <= inreg) { |
| 807 | /* |
| 808 | * This shouldn't happen. If the argument is passed in a |
| 809 | * register then it should have been, well, passed in a |
| 810 | * register... |
| 811 | */ |
| 812 | DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); |
| 813 | return (0); |
| 814 | } |
| 815 | |
| 816 | arg -= (inreg + 1); |
| 817 | stack = (uintptr_t *)&fp[1]; /* Find marshalled arguments */ |
| 818 | |
| 819 | load: |
| 820 | if (dtrace_canload((uint64_t)(stack + arg), sizeof(uint64_t), |
| 821 | mstate, vstate)) { |
| 822 | /* dtrace_probe arguments arg0 ... arg4 are 64bits wide */ |
| 823 | val = dtrace_load64((uint64_t)(stack + arg)); |
| 824 | } |
| 825 | |
| 826 | return (val); |
| 827 | } |
| 828 | |
| 829 | /* |
| 830 | * Load/Store Safety |
| 831 | */ |
| 832 | void |
| 833 | dtrace_toxic_ranges(void (*func)(uintptr_t base, uintptr_t limit)) |
| 834 | { |
| 835 | /* |
| 836 | * "base" is the smallest toxic address in the range, "limit" is the first |
| 837 | * VALID address greater than "base". |
| 838 | */ |
| 839 | func(0x0, VM_MIN_KERNEL_AND_KEXT_ADDRESS); |
| 840 | if (VM_MAX_KERNEL_ADDRESS < ~(uintptr_t)0) |
| 841 | func(VM_MAX_KERNEL_ADDRESS + 1, ~(uintptr_t)0); |
| 842 | } |
| 843 | |
| 844 | |