1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 * 21 * Portions Copyright 2006-2008 John Birrell jb@freebsd.org 22 * 23 * $FreeBSD$ 24 * 25 */ 26 27 /* 28 * Copyright 2006 Sun Microsystems, Inc. All rights reserved. 29 * Use is subject to license terms. 30 */ 31 32 #include <sys/cdefs.h> 33 #include <sys/param.h> 34 35 #include <sys/dtrace.h> 36 37 #include <machine/cpufunc.h> 38 #include <machine/md_var.h> 39 40 #include "fbt.h" 41 42 #define FBT_PUSHL_EBP 0x55 43 #define FBT_MOVL_ESP_EBP0_V0 0x8b 44 #define FBT_MOVL_ESP_EBP1_V0 0xec 45 #define FBT_MOVL_ESP_EBP0_V1 0x89 46 #define FBT_MOVL_ESP_EBP1_V1 0xe5 47 #define FBT_REX_RSP_RBP 0x48 48 49 #define FBT_POPL_EBP 0x5d 50 #define FBT_RET 0xc3 51 #define FBT_RET_IMM16 0xc2 52 #define FBT_LEAVE 0xc9 53 54 #ifdef __amd64__ 55 #define FBT_PATCHVAL 0xcc 56 #else 57 #define FBT_PATCHVAL 0xf0 58 #endif 59 60 #define FBT_ENTRY "entry" 61 #define FBT_RETURN "return" 62 63 int 64 fbt_invop(uintptr_t addr, struct trapframe *frame, uintptr_t rval) 65 { 66 solaris_cpu_t *cpu; 67 uintptr_t *stack; 68 uintptr_t arg0, arg1, arg2, arg3, arg4; 69 fbt_probe_t *fbt; 70 71 #ifdef __amd64__ 72 stack = (uintptr_t *)frame->tf_rsp; 73 #else 74 /* Skip hardware-saved registers. */ 75 stack = (uintptr_t *)frame->tf_isp + 3; 76 #endif 77 78 cpu = &solaris_cpu[curcpu]; 79 fbt = fbt_probetab[FBT_ADDR2NDX(addr)]; 80 for (; fbt != NULL; fbt = fbt->fbtp_hashnext) { 81 if ((uintptr_t)fbt->fbtp_patchpoint == addr) { 82 if (fbt->fbtp_roffset == 0) { 83 #ifdef __amd64__ 84 /* fbt->fbtp_rval == DTRACE_INVOP_PUSHQ_RBP */ 85 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); 86 cpu->cpu_dtrace_caller = stack[0]; 87 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT | 88 CPU_DTRACE_BADADDR); 89 90 arg0 = frame->tf_rdi; 91 arg1 = frame->tf_rsi; 92 arg2 = frame->tf_rdx; 93 arg3 = frame->tf_rcx; 94 arg4 = frame->tf_r8; 95 #else 96 int i = 0; 97 98 /* 99 * When accessing the arguments on the stack, 100 * we must protect against accessing beyond 101 * the stack. We can safely set NOFAULT here 102 * -- we know that interrupts are already 103 * disabled. 104 */ 105 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); 106 cpu->cpu_dtrace_caller = stack[i++]; 107 arg0 = stack[i++]; 108 arg1 = stack[i++]; 109 arg2 = stack[i++]; 110 arg3 = stack[i++]; 111 arg4 = stack[i++]; 112 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT | 113 CPU_DTRACE_BADADDR); 114 #endif 115 116 dtrace_probe(fbt->fbtp_id, arg0, arg1, 117 arg2, arg3, arg4); 118 119 cpu->cpu_dtrace_caller = 0; 120 } else { 121 #ifdef __amd64__ 122 /* 123 * On amd64, we instrument the ret, not the 124 * leave. We therefore need to set the caller 125 * to ensure that the top frame of a stack() 126 * action is correct. 127 */ 128 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); 129 cpu->cpu_dtrace_caller = stack[0]; 130 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT | 131 CPU_DTRACE_BADADDR); 132 #endif 133 134 dtrace_probe(fbt->fbtp_id, fbt->fbtp_roffset, 135 rval, 0, 0, 0); 136 cpu->cpu_dtrace_caller = 0; 137 } 138 139 return (fbt->fbtp_rval); 140 } 141 } 142 143 return (0); 144 } 145 146 void 147 fbt_patch_tracepoint(fbt_probe_t *fbt, fbt_patchval_t val) 148 { 149 register_t intr; 150 bool old_wp; 151 152 intr = intr_disable(); 153 old_wp = disable_wp(); 154 *fbt->fbtp_patchpoint = val; 155 restore_wp(old_wp); 156 intr_restore(intr); 157 } 158 159 int 160 fbt_provide_module_function(linker_file_t lf, int symindx, 161 linker_symval_t *symval, void *opaque) 162 { 163 char *modname = opaque; 164 const char *name = symval->name; 165 fbt_probe_t *fbt, *retfbt; 166 int j; 167 int size; 168 uint8_t *instr, *limit; 169 170 if (fbt_excluded(name)) 171 return (0); 172 173 /* 174 * trap_check() is a wrapper for DTrace's fault handler, so we don't 175 * want to be able to instrument it. 176 */ 177 if (strcmp(name, "trap_check") == 0) 178 return (0); 179 180 size = symval->size; 181 182 instr = (uint8_t *) symval->value; 183 limit = (uint8_t *) symval->value + symval->size; 184 185 #ifdef __amd64__ 186 while (instr < limit) { 187 if (*instr == FBT_PUSHL_EBP) 188 break; 189 190 if ((size = dtrace_instr_size(instr)) <= 0) 191 break; 192 193 instr += size; 194 } 195 196 if (instr >= limit || *instr != FBT_PUSHL_EBP) { 197 /* 198 * We either don't save the frame pointer in this 199 * function, or we ran into some disassembly 200 * screw-up. Either way, we bail. 201 */ 202 return (0); 203 } 204 #else 205 if (instr[0] != FBT_PUSHL_EBP) 206 return (0); 207 208 if (!(instr[1] == FBT_MOVL_ESP_EBP0_V0 && 209 instr[2] == FBT_MOVL_ESP_EBP1_V0) && 210 !(instr[1] == FBT_MOVL_ESP_EBP0_V1 && 211 instr[2] == FBT_MOVL_ESP_EBP1_V1)) 212 return (0); 213 #endif 214 215 fbt = malloc(sizeof (fbt_probe_t), M_FBT, M_WAITOK | M_ZERO); 216 fbt->fbtp_name = name; 217 fbt->fbtp_id = dtrace_probe_create(fbt_id, modname, 218 name, FBT_ENTRY, 3, fbt); 219 fbt->fbtp_patchpoint = instr; 220 fbt->fbtp_ctl = lf; 221 fbt->fbtp_loadcnt = lf->loadcnt; 222 fbt->fbtp_rval = DTRACE_INVOP_PUSHL_EBP; 223 fbt->fbtp_savedval = *instr; 224 fbt->fbtp_patchval = FBT_PATCHVAL; 225 fbt->fbtp_symindx = symindx; 226 227 fbt->fbtp_hashnext = fbt_probetab[FBT_ADDR2NDX(instr)]; 228 fbt_probetab[FBT_ADDR2NDX(instr)] = fbt; 229 230 lf->fbt_nentries++; 231 232 retfbt = NULL; 233 again: 234 if (instr >= limit) 235 return (0); 236 237 /* 238 * If this disassembly fails, then we've likely walked off into 239 * a jump table or some other unsuitable area. Bail out of the 240 * disassembly now. 241 */ 242 if ((size = dtrace_instr_size(instr)) <= 0) 243 return (0); 244 245 #ifdef __amd64__ 246 /* 247 * We only instrument "ret" on amd64 -- we don't yet instrument 248 * ret imm16, largely because the compiler doesn't seem to 249 * (yet) emit them in the kernel... 250 */ 251 if (*instr != FBT_RET) { 252 instr += size; 253 goto again; 254 } 255 #else 256 if (!(size == 1 && 257 (*instr == FBT_POPL_EBP || *instr == FBT_LEAVE) && 258 (*(instr + 1) == FBT_RET || 259 *(instr + 1) == FBT_RET_IMM16))) { 260 instr += size; 261 goto again; 262 } 263 #endif 264 265 /* 266 * We (desperately) want to avoid erroneously instrumenting a 267 * jump table, especially given that our markers are pretty 268 * short: two bytes on x86, and just one byte on amd64. To 269 * determine if we're looking at a true instruction sequence 270 * or an inline jump table that happens to contain the same 271 * byte sequences, we resort to some heuristic sleeze: we 272 * treat this instruction as being contained within a pointer, 273 * and see if that pointer points to within the body of the 274 * function. If it does, we refuse to instrument it. 275 */ 276 for (j = 0; j < sizeof (uintptr_t); j++) { 277 caddr_t check = (caddr_t) instr - j; 278 uint8_t *ptr; 279 280 if (check < symval->value) 281 break; 282 283 if (check + sizeof (caddr_t) > (caddr_t)limit) 284 continue; 285 286 ptr = *(uint8_t **)check; 287 288 if (ptr >= (uint8_t *) symval->value && ptr < limit) { 289 instr += size; 290 goto again; 291 } 292 } 293 294 /* 295 * We have a winner! 296 */ 297 fbt = malloc(sizeof (fbt_probe_t), M_FBT, M_WAITOK | M_ZERO); 298 fbt->fbtp_name = name; 299 300 if (retfbt == NULL) { 301 fbt->fbtp_id = dtrace_probe_create(fbt_id, modname, 302 name, FBT_RETURN, 3, fbt); 303 } else { 304 retfbt->fbtp_next = fbt; 305 fbt->fbtp_id = retfbt->fbtp_id; 306 } 307 308 retfbt = fbt; 309 fbt->fbtp_patchpoint = instr; 310 fbt->fbtp_ctl = lf; 311 fbt->fbtp_loadcnt = lf->loadcnt; 312 fbt->fbtp_symindx = symindx; 313 314 #ifndef __amd64__ 315 if (*instr == FBT_POPL_EBP) { 316 fbt->fbtp_rval = DTRACE_INVOP_POPL_EBP; 317 } else { 318 ASSERT(*instr == FBT_LEAVE); 319 fbt->fbtp_rval = DTRACE_INVOP_LEAVE; 320 } 321 fbt->fbtp_roffset = 322 (uintptr_t)(instr - (uint8_t *) symval->value) + 1; 323 324 #else 325 ASSERT(*instr == FBT_RET); 326 fbt->fbtp_rval = DTRACE_INVOP_RET; 327 fbt->fbtp_roffset = 328 (uintptr_t)(instr - (uint8_t *) symval->value); 329 #endif 330 331 fbt->fbtp_savedval = *instr; 332 fbt->fbtp_patchval = FBT_PATCHVAL; 333 fbt->fbtp_hashnext = fbt_probetab[FBT_ADDR2NDX(instr)]; 334 fbt_probetab[FBT_ADDR2NDX(instr)] = fbt; 335 336 lf->fbt_nentries++; 337 338 instr += size; 339 goto again; 340 } 341