| 1 | /* Target-dependent code for GNU/Linux i386. |
| 2 | |
| 3 | Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2007 |
| 4 | Free Software Foundation, Inc. |
| 5 | |
| 6 | This file is part of GDB. |
| 7 | |
| 8 | This program is free software; you can redistribute it and/or modify |
| 9 | it under the terms of the GNU General Public License as published by |
| 10 | the Free Software Foundation; either version 2 of the License, or |
| 11 | (at your option) any later version. |
| 12 | |
| 13 | This program is distributed in the hope that it will be useful, |
| 14 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 15 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 16 | GNU General Public License for more details. |
| 17 | |
| 18 | You should have received a copy of the GNU General Public License |
| 19 | along with this program; if not, write to the Free Software |
| 20 | Foundation, Inc., 51 Franklin Street, Fifth Floor, |
| 21 | Boston, MA 02110-1301, USA. */ |
| 22 | |
| 23 | #include "defs.h" |
| 24 | #include "gdbcore.h" |
| 25 | #include "frame.h" |
| 26 | #include "value.h" |
| 27 | #include "regcache.h" |
| 28 | #include "inferior.h" |
| 29 | #include "osabi.h" |
| 30 | #include "reggroups.h" |
| 31 | #include "dwarf2-frame.h" |
| 32 | #include "gdb_string.h" |
| 33 | |
| 34 | #include "i386-tdep.h" |
| 35 | #include "i386-linux-tdep.h" |
| 36 | #include "glibc-tdep.h" |
| 37 | #include "solib-svr4.h" |
| 38 | #include "symtab.h" |
| 39 | |
| 40 | /* Return the name of register REG. */ |
| 41 | |
| 42 | static const char * |
| 43 | i386_linux_register_name (int reg) |
| 44 | { |
| 45 | /* Deal with the extra "orig_eax" pseudo register. */ |
| 46 | if (reg == I386_LINUX_ORIG_EAX_REGNUM) |
| 47 | return "orig_eax"; |
| 48 | |
| 49 | return i386_register_name (reg); |
| 50 | } |
| 51 | |
| 52 | /* Return non-zero, when the register is in the corresponding register |
| 53 | group. Put the LINUX_ORIG_EAX register in the system group. */ |
| 54 | static int |
| 55 | i386_linux_register_reggroup_p (struct gdbarch *gdbarch, int regnum, |
| 56 | struct reggroup *group) |
| 57 | { |
| 58 | if (regnum == I386_LINUX_ORIG_EAX_REGNUM) |
| 59 | return (group == system_reggroup |
| 60 | || group == save_reggroup |
| 61 | || group == restore_reggroup); |
| 62 | return i386_register_reggroup_p (gdbarch, regnum, group); |
| 63 | } |
| 64 | |
| 65 | \f |
| 66 | /* Recognizing signal handler frames. */ |
| 67 | |
| 68 | /* GNU/Linux has two flavors of signals. Normal signal handlers, and |
| 69 | "realtime" (RT) signals. The RT signals can provide additional |
| 70 | information to the signal handler if the SA_SIGINFO flag is set |
| 71 | when establishing a signal handler using `sigaction'. It is not |
| 72 | unlikely that future versions of GNU/Linux will support SA_SIGINFO |
| 73 | for normal signals too. */ |
| 74 | |
| 75 | /* When the i386 Linux kernel calls a signal handler and the |
| 76 | SA_RESTORER flag isn't set, the return address points to a bit of |
| 77 | code on the stack. This function returns whether the PC appears to |
| 78 | be within this bit of code. |
| 79 | |
| 80 | The instruction sequence for normal signals is |
| 81 | pop %eax |
| 82 | mov $0x77, %eax |
| 83 | int $0x80 |
| 84 | or 0x58 0xb8 0x77 0x00 0x00 0x00 0xcd 0x80. |
| 85 | |
| 86 | Checking for the code sequence should be somewhat reliable, because |
| 87 | the effect is to call the system call sigreturn. This is unlikely |
| 88 | to occur anywhere other than in a signal trampoline. |
| 89 | |
| 90 | It kind of sucks that we have to read memory from the process in |
| 91 | order to identify a signal trampoline, but there doesn't seem to be |
| 92 | any other way. Therefore we only do the memory reads if no |
| 93 | function name could be identified, which should be the case since |
| 94 | the code is on the stack. |
| 95 | |
| 96 | Detection of signal trampolines for handlers that set the |
| 97 | SA_RESTORER flag is in general not possible. Unfortunately this is |
| 98 | what the GNU C Library has been doing for quite some time now. |
| 99 | However, as of version 2.1.2, the GNU C Library uses signal |
| 100 | trampolines (named __restore and __restore_rt) that are identical |
| 101 | to the ones used by the kernel. Therefore, these trampolines are |
| 102 | supported too. */ |
| 103 | |
| 104 | #define LINUX_SIGTRAMP_INSN0 0x58 /* pop %eax */ |
| 105 | #define LINUX_SIGTRAMP_OFFSET0 0 |
| 106 | #define LINUX_SIGTRAMP_INSN1 0xb8 /* mov $NNNN, %eax */ |
| 107 | #define LINUX_SIGTRAMP_OFFSET1 1 |
| 108 | #define LINUX_SIGTRAMP_INSN2 0xcd /* int */ |
| 109 | #define LINUX_SIGTRAMP_OFFSET2 6 |
| 110 | |
| 111 | static const gdb_byte linux_sigtramp_code[] = |
| 112 | { |
| 113 | LINUX_SIGTRAMP_INSN0, /* pop %eax */ |
| 114 | LINUX_SIGTRAMP_INSN1, 0x77, 0x00, 0x00, 0x00, /* mov $0x77, %eax */ |
| 115 | LINUX_SIGTRAMP_INSN2, 0x80 /* int $0x80 */ |
| 116 | }; |
| 117 | |
| 118 | #define LINUX_SIGTRAMP_LEN (sizeof linux_sigtramp_code) |
| 119 | |
| 120 | /* If NEXT_FRAME unwinds into a sigtramp routine, return the address |
| 121 | of the start of the routine. Otherwise, return 0. */ |
| 122 | |
| 123 | static CORE_ADDR |
| 124 | i386_linux_sigtramp_start (struct frame_info *next_frame) |
| 125 | { |
| 126 | CORE_ADDR pc = frame_pc_unwind (next_frame); |
| 127 | gdb_byte buf[LINUX_SIGTRAMP_LEN]; |
| 128 | |
| 129 | /* We only recognize a signal trampoline if PC is at the start of |
| 130 | one of the three instructions. We optimize for finding the PC at |
| 131 | the start, as will be the case when the trampoline is not the |
| 132 | first frame on the stack. We assume that in the case where the |
| 133 | PC is not at the start of the instruction sequence, there will be |
| 134 | a few trailing readable bytes on the stack. */ |
| 135 | |
| 136 | if (!safe_frame_unwind_memory (next_frame, pc, buf, LINUX_SIGTRAMP_LEN)) |
| 137 | return 0; |
| 138 | |
| 139 | if (buf[0] != LINUX_SIGTRAMP_INSN0) |
| 140 | { |
| 141 | int adjust; |
| 142 | |
| 143 | switch (buf[0]) |
| 144 | { |
| 145 | case LINUX_SIGTRAMP_INSN1: |
| 146 | adjust = LINUX_SIGTRAMP_OFFSET1; |
| 147 | break; |
| 148 | case LINUX_SIGTRAMP_INSN2: |
| 149 | adjust = LINUX_SIGTRAMP_OFFSET2; |
| 150 | break; |
| 151 | default: |
| 152 | return 0; |
| 153 | } |
| 154 | |
| 155 | pc -= adjust; |
| 156 | |
| 157 | if (!safe_frame_unwind_memory (next_frame, pc, buf, LINUX_SIGTRAMP_LEN)) |
| 158 | return 0; |
| 159 | } |
| 160 | |
| 161 | if (memcmp (buf, linux_sigtramp_code, LINUX_SIGTRAMP_LEN) != 0) |
| 162 | return 0; |
| 163 | |
| 164 | return pc; |
| 165 | } |
| 166 | |
| 167 | /* This function does the same for RT signals. Here the instruction |
| 168 | sequence is |
| 169 | mov $0xad, %eax |
| 170 | int $0x80 |
| 171 | or 0xb8 0xad 0x00 0x00 0x00 0xcd 0x80. |
| 172 | |
| 173 | The effect is to call the system call rt_sigreturn. */ |
| 174 | |
| 175 | #define LINUX_RT_SIGTRAMP_INSN0 0xb8 /* mov $NNNN, %eax */ |
| 176 | #define LINUX_RT_SIGTRAMP_OFFSET0 0 |
| 177 | #define LINUX_RT_SIGTRAMP_INSN1 0xcd /* int */ |
| 178 | #define LINUX_RT_SIGTRAMP_OFFSET1 5 |
| 179 | |
| 180 | static const gdb_byte linux_rt_sigtramp_code[] = |
| 181 | { |
| 182 | LINUX_RT_SIGTRAMP_INSN0, 0xad, 0x00, 0x00, 0x00, /* mov $0xad, %eax */ |
| 183 | LINUX_RT_SIGTRAMP_INSN1, 0x80 /* int $0x80 */ |
| 184 | }; |
| 185 | |
| 186 | #define LINUX_RT_SIGTRAMP_LEN (sizeof linux_rt_sigtramp_code) |
| 187 | |
| 188 | /* If NEXT_FRAME unwinds into an RT sigtramp routine, return the |
| 189 | address of the start of the routine. Otherwise, return 0. */ |
| 190 | |
| 191 | static CORE_ADDR |
| 192 | i386_linux_rt_sigtramp_start (struct frame_info *next_frame) |
| 193 | { |
| 194 | CORE_ADDR pc = frame_pc_unwind (next_frame); |
| 195 | gdb_byte buf[LINUX_RT_SIGTRAMP_LEN]; |
| 196 | |
| 197 | /* We only recognize a signal trampoline if PC is at the start of |
| 198 | one of the two instructions. We optimize for finding the PC at |
| 199 | the start, as will be the case when the trampoline is not the |
| 200 | first frame on the stack. We assume that in the case where the |
| 201 | PC is not at the start of the instruction sequence, there will be |
| 202 | a few trailing readable bytes on the stack. */ |
| 203 | |
| 204 | if (!safe_frame_unwind_memory (next_frame, pc, buf, LINUX_RT_SIGTRAMP_LEN)) |
| 205 | return 0; |
| 206 | |
| 207 | if (buf[0] != LINUX_RT_SIGTRAMP_INSN0) |
| 208 | { |
| 209 | if (buf[0] != LINUX_RT_SIGTRAMP_INSN1) |
| 210 | return 0; |
| 211 | |
| 212 | pc -= LINUX_RT_SIGTRAMP_OFFSET1; |
| 213 | |
| 214 | if (!safe_frame_unwind_memory (next_frame, pc, buf, |
| 215 | LINUX_RT_SIGTRAMP_LEN)) |
| 216 | return 0; |
| 217 | } |
| 218 | |
| 219 | if (memcmp (buf, linux_rt_sigtramp_code, LINUX_RT_SIGTRAMP_LEN) != 0) |
| 220 | return 0; |
| 221 | |
| 222 | return pc; |
| 223 | } |
| 224 | |
| 225 | /* Return whether the frame preceding NEXT_FRAME corresponds to a |
| 226 | GNU/Linux sigtramp routine. */ |
| 227 | |
| 228 | static int |
| 229 | i386_linux_sigtramp_p (struct frame_info *next_frame) |
| 230 | { |
| 231 | CORE_ADDR pc = frame_pc_unwind (next_frame); |
| 232 | char *name; |
| 233 | |
| 234 | find_pc_partial_function (pc, &name, NULL, NULL); |
| 235 | |
| 236 | /* If we have NAME, we can optimize the search. The trampolines are |
| 237 | named __restore and __restore_rt. However, they aren't dynamically |
| 238 | exported from the shared C library, so the trampoline may appear to |
| 239 | be part of the preceding function. This should always be sigaction, |
| 240 | __sigaction, or __libc_sigaction (all aliases to the same function). */ |
| 241 | if (name == NULL || strstr (name, "sigaction") != NULL) |
| 242 | return (i386_linux_sigtramp_start (next_frame) != 0 |
| 243 | || i386_linux_rt_sigtramp_start (next_frame) != 0); |
| 244 | |
| 245 | return (strcmp ("__restore", name) == 0 |
| 246 | || strcmp ("__restore_rt", name) == 0); |
| 247 | } |
| 248 | |
| 249 | /* Return one if the unwound PC from NEXT_FRAME is in a signal trampoline |
| 250 | which may have DWARF-2 CFI. */ |
| 251 | |
| 252 | static int |
| 253 | i386_linux_dwarf_signal_frame_p (struct gdbarch *gdbarch, |
| 254 | struct frame_info *next_frame) |
| 255 | { |
| 256 | CORE_ADDR pc = frame_pc_unwind (next_frame); |
| 257 | char *name; |
| 258 | |
| 259 | find_pc_partial_function (pc, &name, NULL, NULL); |
| 260 | |
| 261 | /* If a vsyscall DSO is in use, the signal trampolines may have these |
| 262 | names. */ |
| 263 | if (name && (strcmp (name, "__kernel_sigreturn") == 0 |
| 264 | || strcmp (name, "__kernel_rt_sigreturn") == 0)) |
| 265 | return 1; |
| 266 | |
| 267 | return 0; |
| 268 | } |
| 269 | |
| 270 | /* Offset to struct sigcontext in ucontext, from <asm/ucontext.h>. */ |
| 271 | #define I386_LINUX_UCONTEXT_SIGCONTEXT_OFFSET 20 |
| 272 | |
| 273 | /* Assuming NEXT_FRAME is a frame following a GNU/Linux sigtramp |
| 274 | routine, return the address of the associated sigcontext structure. */ |
| 275 | |
| 276 | static CORE_ADDR |
| 277 | i386_linux_sigcontext_addr (struct frame_info *next_frame) |
| 278 | { |
| 279 | CORE_ADDR pc; |
| 280 | CORE_ADDR sp; |
| 281 | gdb_byte buf[4]; |
| 282 | |
| 283 | frame_unwind_register (next_frame, I386_ESP_REGNUM, buf); |
| 284 | sp = extract_unsigned_integer (buf, 4); |
| 285 | |
| 286 | pc = i386_linux_sigtramp_start (next_frame); |
| 287 | if (pc) |
| 288 | { |
| 289 | /* The sigcontext structure lives on the stack, right after |
| 290 | the signum argument. We determine the address of the |
| 291 | sigcontext structure by looking at the frame's stack |
| 292 | pointer. Keep in mind that the first instruction of the |
| 293 | sigtramp code is "pop %eax". If the PC is after this |
| 294 | instruction, adjust the returned value accordingly. */ |
| 295 | if (pc == frame_pc_unwind (next_frame)) |
| 296 | return sp + 4; |
| 297 | return sp; |
| 298 | } |
| 299 | |
| 300 | pc = i386_linux_rt_sigtramp_start (next_frame); |
| 301 | if (pc) |
| 302 | { |
| 303 | CORE_ADDR ucontext_addr; |
| 304 | |
| 305 | /* The sigcontext structure is part of the user context. A |
| 306 | pointer to the user context is passed as the third argument |
| 307 | to the signal handler. */ |
| 308 | read_memory (sp + 8, buf, 4); |
| 309 | ucontext_addr = extract_unsigned_integer (buf, 4); |
| 310 | return ucontext_addr + I386_LINUX_UCONTEXT_SIGCONTEXT_OFFSET; |
| 311 | } |
| 312 | |
| 313 | error (_("Couldn't recognize signal trampoline.")); |
| 314 | return 0; |
| 315 | } |
| 316 | |
| 317 | /* Set the program counter for process PTID to PC. */ |
| 318 | |
| 319 | static void |
| 320 | i386_linux_write_pc (struct regcache *regcache, CORE_ADDR pc) |
| 321 | { |
| 322 | regcache_cooked_write_unsigned (regcache, I386_EIP_REGNUM, pc); |
| 323 | |
| 324 | /* We must be careful with modifying the program counter. If we |
| 325 | just interrupted a system call, the kernel might try to restart |
| 326 | it when we resume the inferior. On restarting the system call, |
| 327 | the kernel will try backing up the program counter even though it |
| 328 | no longer points at the system call. This typically results in a |
| 329 | SIGSEGV or SIGILL. We can prevent this by writing `-1' in the |
| 330 | "orig_eax" pseudo-register. |
| 331 | |
| 332 | Note that "orig_eax" is saved when setting up a dummy call frame. |
| 333 | This means that it is properly restored when that frame is |
| 334 | popped, and that the interrupted system call will be restarted |
| 335 | when we resume the inferior on return from a function call from |
| 336 | within GDB. In all other cases the system call will not be |
| 337 | restarted. */ |
| 338 | regcache_cooked_write_unsigned (regcache, I386_LINUX_ORIG_EAX_REGNUM, -1); |
| 339 | } |
| 340 | \f |
| 341 | |
| 342 | /* The register sets used in GNU/Linux ELF core-dumps are identical to |
| 343 | the register sets in `struct user' that are used for a.out |
| 344 | core-dumps. These are also used by ptrace(2). The corresponding |
| 345 | types are `elf_gregset_t' for the general-purpose registers (with |
| 346 | `elf_greg_t' the type of a single GP register) and `elf_fpregset_t' |
| 347 | for the floating-point registers. |
| 348 | |
| 349 | Those types used to be available under the names `gregset_t' and |
| 350 | `fpregset_t' too, and GDB used those names in the past. But those |
| 351 | names are now used for the register sets used in the `mcontext_t' |
| 352 | type, which have a different size and layout. */ |
| 353 | |
| 354 | /* Mapping between the general-purpose registers in `struct user' |
| 355 | format and GDB's register cache layout. */ |
| 356 | |
| 357 | /* From <sys/reg.h>. */ |
| 358 | static int i386_linux_gregset_reg_offset[] = |
| 359 | { |
| 360 | 6 * 4, /* %eax */ |
| 361 | 1 * 4, /* %ecx */ |
| 362 | 2 * 4, /* %edx */ |
| 363 | 0 * 4, /* %ebx */ |
| 364 | 15 * 4, /* %esp */ |
| 365 | 5 * 4, /* %ebp */ |
| 366 | 3 * 4, /* %esi */ |
| 367 | 4 * 4, /* %edi */ |
| 368 | 12 * 4, /* %eip */ |
| 369 | 14 * 4, /* %eflags */ |
| 370 | 13 * 4, /* %cs */ |
| 371 | 16 * 4, /* %ss */ |
| 372 | 7 * 4, /* %ds */ |
| 373 | 8 * 4, /* %es */ |
| 374 | 9 * 4, /* %fs */ |
| 375 | 10 * 4, /* %gs */ |
| 376 | -1, -1, -1, -1, -1, -1, -1, -1, |
| 377 | -1, -1, -1, -1, -1, -1, -1, -1, |
| 378 | -1, -1, -1, -1, -1, -1, -1, -1, |
| 379 | -1, |
| 380 | 11 * 4 /* "orig_eax" */ |
| 381 | }; |
| 382 | |
| 383 | /* Mapping between the general-purpose registers in `struct |
| 384 | sigcontext' format and GDB's register cache layout. */ |
| 385 | |
| 386 | /* From <asm/sigcontext.h>. */ |
| 387 | static int i386_linux_sc_reg_offset[] = |
| 388 | { |
| 389 | 11 * 4, /* %eax */ |
| 390 | 10 * 4, /* %ecx */ |
| 391 | 9 * 4, /* %edx */ |
| 392 | 8 * 4, /* %ebx */ |
| 393 | 7 * 4, /* %esp */ |
| 394 | 6 * 4, /* %ebp */ |
| 395 | 5 * 4, /* %esi */ |
| 396 | 4 * 4, /* %edi */ |
| 397 | 14 * 4, /* %eip */ |
| 398 | 16 * 4, /* %eflags */ |
| 399 | 15 * 4, /* %cs */ |
| 400 | 18 * 4, /* %ss */ |
| 401 | 3 * 4, /* %ds */ |
| 402 | 2 * 4, /* %es */ |
| 403 | 1 * 4, /* %fs */ |
| 404 | 0 * 4 /* %gs */ |
| 405 | }; |
| 406 | |
| 407 | static void |
| 408 | i386_linux_init_abi (struct gdbarch_info info, struct gdbarch *gdbarch) |
| 409 | { |
| 410 | struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch); |
| 411 | |
| 412 | /* GNU/Linux uses ELF. */ |
| 413 | i386_elf_init_abi (info, gdbarch); |
| 414 | |
| 415 | /* Since we have the extra "orig_eax" register on GNU/Linux, we have |
| 416 | to adjust a few things. */ |
| 417 | |
| 418 | set_gdbarch_write_pc (gdbarch, i386_linux_write_pc); |
| 419 | set_gdbarch_num_regs (gdbarch, I386_LINUX_NUM_REGS); |
| 420 | set_gdbarch_register_name (gdbarch, i386_linux_register_name); |
| 421 | set_gdbarch_register_reggroup_p (gdbarch, i386_linux_register_reggroup_p); |
| 422 | |
| 423 | tdep->gregset_reg_offset = i386_linux_gregset_reg_offset; |
| 424 | tdep->gregset_num_regs = ARRAY_SIZE (i386_linux_gregset_reg_offset); |
| 425 | tdep->sizeof_gregset = 17 * 4; |
| 426 | |
| 427 | tdep->jb_pc_offset = 20; /* From <bits/setjmp.h>. */ |
| 428 | |
| 429 | tdep->sigtramp_p = i386_linux_sigtramp_p; |
| 430 | tdep->sigcontext_addr = i386_linux_sigcontext_addr; |
| 431 | tdep->sc_reg_offset = i386_linux_sc_reg_offset; |
| 432 | tdep->sc_num_regs = ARRAY_SIZE (i386_linux_sc_reg_offset); |
| 433 | |
| 434 | /* GNU/Linux uses SVR4-style shared libraries. */ |
| 435 | set_gdbarch_skip_trampoline_code (gdbarch, find_solib_trampoline_target); |
| 436 | set_solib_svr4_fetch_link_map_offsets |
| 437 | (gdbarch, svr4_ilp32_fetch_link_map_offsets); |
| 438 | |
| 439 | /* GNU/Linux uses the dynamic linker included in the GNU C Library. */ |
| 440 | set_gdbarch_skip_solib_resolver (gdbarch, glibc_skip_solib_resolver); |
| 441 | |
| 442 | dwarf2_frame_set_signal_frame_p (gdbarch, i386_linux_dwarf_signal_frame_p); |
| 443 | |
| 444 | /* Enable TLS support. */ |
| 445 | set_gdbarch_fetch_tls_load_module_address (gdbarch, |
| 446 | svr4_fetch_objfile_link_map); |
| 447 | } |
| 448 | |
| 449 | /* Provide a prototype to silence -Wmissing-prototypes. */ |
| 450 | extern void _initialize_i386_linux_tdep (void); |
| 451 | |
| 452 | void |
| 453 | _initialize_i386_linux_tdep (void) |
| 454 | { |
| 455 | gdbarch_register_osabi (bfd_arch_i386, 0, GDB_OSABI_LINUX, |
| 456 | i386_linux_init_abi); |
| 457 | } |