| 1 | /* Target-dependent code for Linux running on i386's, for GDB. |
| 2 | Copyright 2000, 2001 Free Software Foundation, Inc. |
| 3 | |
| 4 | This file is part of GDB. |
| 5 | |
| 6 | This program is free software; you can redistribute it and/or modify |
| 7 | it under the terms of the GNU General Public License as published by |
| 8 | the Free Software Foundation; either version 2 of the License, or |
| 9 | (at your option) any later version. |
| 10 | |
| 11 | This program is distributed in the hope that it will be useful, |
| 12 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 13 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 14 | GNU General Public License for more details. |
| 15 | |
| 16 | You should have received a copy of the GNU General Public License |
| 17 | along with this program; if not, write to the Free Software |
| 18 | Foundation, Inc., 59 Temple Place - Suite 330, |
| 19 | Boston, MA 02111-1307, USA. */ |
| 20 | |
| 21 | #include "defs.h" |
| 22 | #include "gdbcore.h" |
| 23 | #include "frame.h" |
| 24 | #include "value.h" |
| 25 | #include "regcache.h" |
| 26 | |
| 27 | /* For i386_linux_skip_solib_resolver. */ |
| 28 | #include "symtab.h" |
| 29 | #include "symfile.h" |
| 30 | #include "objfiles.h" |
| 31 | |
| 32 | #include "solib-svr4.h" /* For struct link_map_offsets. */ |
| 33 | |
| 34 | \f |
| 35 | /* Recognizing signal handler frames. */ |
| 36 | |
| 37 | /* Linux has two flavors of signals. Normal signal handlers, and |
| 38 | "realtime" (RT) signals. The RT signals can provide additional |
| 39 | information to the signal handler if the SA_SIGINFO flag is set |
| 40 | when establishing a signal handler using `sigaction'. It is not |
| 41 | unlikely that future versions of Linux will support SA_SIGINFO for |
| 42 | normal signals too. */ |
| 43 | |
| 44 | /* When the i386 Linux kernel calls a signal handler and the |
| 45 | SA_RESTORER flag isn't set, the return address points to a bit of |
| 46 | code on the stack. This function returns whether the PC appears to |
| 47 | be within this bit of code. |
| 48 | |
| 49 | The instruction sequence for normal signals is |
| 50 | pop %eax |
| 51 | mov $0x77,%eax |
| 52 | int $0x80 |
| 53 | or 0x58 0xb8 0x77 0x00 0x00 0x00 0xcd 0x80. |
| 54 | |
| 55 | Checking for the code sequence should be somewhat reliable, because |
| 56 | the effect is to call the system call sigreturn. This is unlikely |
| 57 | to occur anywhere other than a signal trampoline. |
| 58 | |
| 59 | It kind of sucks that we have to read memory from the process in |
| 60 | order to identify a signal trampoline, but there doesn't seem to be |
| 61 | any other way. The IN_SIGTRAMP macro in tm-linux.h arranges to |
| 62 | only call us if no function name could be identified, which should |
| 63 | be the case since the code is on the stack. |
| 64 | |
| 65 | Detection of signal trampolines for handlers that set the |
| 66 | SA_RESTORER flag is in general not possible. Unfortunately this is |
| 67 | what the GNU C Library has been doing for quite some time now. |
| 68 | However, as of version 2.1.2, the GNU C Library uses signal |
| 69 | trampolines (named __restore and __restore_rt) that are identical |
| 70 | to the ones used by the kernel. Therefore, these trampolines are |
| 71 | supported too. */ |
| 72 | |
| 73 | #define LINUX_SIGTRAMP_INSN0 (0x58) /* pop %eax */ |
| 74 | #define LINUX_SIGTRAMP_OFFSET0 (0) |
| 75 | #define LINUX_SIGTRAMP_INSN1 (0xb8) /* mov $NNNN,%eax */ |
| 76 | #define LINUX_SIGTRAMP_OFFSET1 (1) |
| 77 | #define LINUX_SIGTRAMP_INSN2 (0xcd) /* int */ |
| 78 | #define LINUX_SIGTRAMP_OFFSET2 (6) |
| 79 | |
| 80 | static const unsigned char linux_sigtramp_code[] = |
| 81 | { |
| 82 | LINUX_SIGTRAMP_INSN0, /* pop %eax */ |
| 83 | LINUX_SIGTRAMP_INSN1, 0x77, 0x00, 0x00, 0x00, /* mov $0x77,%eax */ |
| 84 | LINUX_SIGTRAMP_INSN2, 0x80 /* int $0x80 */ |
| 85 | }; |
| 86 | |
| 87 | #define LINUX_SIGTRAMP_LEN (sizeof linux_sigtramp_code) |
| 88 | |
| 89 | /* If PC is in a sigtramp routine, return the address of the start of |
| 90 | the routine. Otherwise, return 0. */ |
| 91 | |
| 92 | static CORE_ADDR |
| 93 | i386_linux_sigtramp_start (CORE_ADDR pc) |
| 94 | { |
| 95 | unsigned char buf[LINUX_SIGTRAMP_LEN]; |
| 96 | |
| 97 | /* We only recognize a signal trampoline if PC is at the start of |
| 98 | one of the three instructions. We optimize for finding the PC at |
| 99 | the start, as will be the case when the trampoline is not the |
| 100 | first frame on the stack. We assume that in the case where the |
| 101 | PC is not at the start of the instruction sequence, there will be |
| 102 | a few trailing readable bytes on the stack. */ |
| 103 | |
| 104 | if (read_memory_nobpt (pc, (char *) buf, LINUX_SIGTRAMP_LEN) != 0) |
| 105 | return 0; |
| 106 | |
| 107 | if (buf[0] != LINUX_SIGTRAMP_INSN0) |
| 108 | { |
| 109 | int adjust; |
| 110 | |
| 111 | switch (buf[0]) |
| 112 | { |
| 113 | case LINUX_SIGTRAMP_INSN1: |
| 114 | adjust = LINUX_SIGTRAMP_OFFSET1; |
| 115 | break; |
| 116 | case LINUX_SIGTRAMP_INSN2: |
| 117 | adjust = LINUX_SIGTRAMP_OFFSET2; |
| 118 | break; |
| 119 | default: |
| 120 | return 0; |
| 121 | } |
| 122 | |
| 123 | pc -= adjust; |
| 124 | |
| 125 | if (read_memory_nobpt (pc, (char *) buf, LINUX_SIGTRAMP_LEN) != 0) |
| 126 | return 0; |
| 127 | } |
| 128 | |
| 129 | if (memcmp (buf, linux_sigtramp_code, LINUX_SIGTRAMP_LEN) != 0) |
| 130 | return 0; |
| 131 | |
| 132 | return pc; |
| 133 | } |
| 134 | |
| 135 | /* This function does the same for RT signals. Here the instruction |
| 136 | sequence is |
| 137 | mov $0xad,%eax |
| 138 | int $0x80 |
| 139 | or 0xb8 0xad 0x00 0x00 0x00 0xcd 0x80. |
| 140 | |
| 141 | The effect is to call the system call rt_sigreturn. */ |
| 142 | |
| 143 | #define LINUX_RT_SIGTRAMP_INSN0 (0xb8) /* mov $NNNN,%eax */ |
| 144 | #define LINUX_RT_SIGTRAMP_OFFSET0 (0) |
| 145 | #define LINUX_RT_SIGTRAMP_INSN1 (0xcd) /* int */ |
| 146 | #define LINUX_RT_SIGTRAMP_OFFSET1 (5) |
| 147 | |
| 148 | static const unsigned char linux_rt_sigtramp_code[] = |
| 149 | { |
| 150 | LINUX_RT_SIGTRAMP_INSN0, 0xad, 0x00, 0x00, 0x00, /* mov $0xad,%eax */ |
| 151 | LINUX_RT_SIGTRAMP_INSN1, 0x80 /* int $0x80 */ |
| 152 | }; |
| 153 | |
| 154 | #define LINUX_RT_SIGTRAMP_LEN (sizeof linux_rt_sigtramp_code) |
| 155 | |
| 156 | /* If PC is in a RT sigtramp routine, return the address of the start |
| 157 | of the routine. Otherwise, return 0. */ |
| 158 | |
| 159 | static CORE_ADDR |
| 160 | i386_linux_rt_sigtramp_start (CORE_ADDR pc) |
| 161 | { |
| 162 | unsigned char buf[LINUX_RT_SIGTRAMP_LEN]; |
| 163 | |
| 164 | /* We only recognize a signal trampoline if PC is at the start of |
| 165 | one of the two instructions. We optimize for finding the PC at |
| 166 | the start, as will be the case when the trampoline is not the |
| 167 | first frame on the stack. We assume that in the case where the |
| 168 | PC is not at the start of the instruction sequence, there will be |
| 169 | a few trailing readable bytes on the stack. */ |
| 170 | |
| 171 | if (read_memory_nobpt (pc, (char *) buf, LINUX_RT_SIGTRAMP_LEN) != 0) |
| 172 | return 0; |
| 173 | |
| 174 | if (buf[0] != LINUX_RT_SIGTRAMP_INSN0) |
| 175 | { |
| 176 | if (buf[0] != LINUX_RT_SIGTRAMP_INSN1) |
| 177 | return 0; |
| 178 | |
| 179 | pc -= LINUX_RT_SIGTRAMP_OFFSET1; |
| 180 | |
| 181 | if (read_memory_nobpt (pc, (char *) buf, LINUX_RT_SIGTRAMP_LEN) != 0) |
| 182 | return 0; |
| 183 | } |
| 184 | |
| 185 | if (memcmp (buf, linux_rt_sigtramp_code, LINUX_RT_SIGTRAMP_LEN) != 0) |
| 186 | return 0; |
| 187 | |
| 188 | return pc; |
| 189 | } |
| 190 | |
| 191 | /* Return whether PC is in a Linux sigtramp routine. */ |
| 192 | |
| 193 | int |
| 194 | i386_linux_in_sigtramp (CORE_ADDR pc, char *name) |
| 195 | { |
| 196 | if (name) |
| 197 | return STREQ ("__restore", name) || STREQ ("__restore_rt", name); |
| 198 | |
| 199 | return (i386_linux_sigtramp_start (pc) != 0 |
| 200 | || i386_linux_rt_sigtramp_start (pc) != 0); |
| 201 | } |
| 202 | |
| 203 | /* Assuming FRAME is for a Linux sigtramp routine, return the address |
| 204 | of the associated sigcontext structure. */ |
| 205 | |
| 206 | CORE_ADDR |
| 207 | i386_linux_sigcontext_addr (struct frame_info *frame) |
| 208 | { |
| 209 | CORE_ADDR pc; |
| 210 | |
| 211 | pc = i386_linux_sigtramp_start (frame->pc); |
| 212 | if (pc) |
| 213 | { |
| 214 | CORE_ADDR sp; |
| 215 | |
| 216 | if (frame->next) |
| 217 | /* If this isn't the top frame, the next frame must be for the |
| 218 | signal handler itself. The sigcontext structure lives on |
| 219 | the stack, right after the signum argument. */ |
| 220 | return frame->next->frame + 12; |
| 221 | |
| 222 | /* This is the top frame. We'll have to find the address of the |
| 223 | sigcontext structure by looking at the stack pointer. Keep |
| 224 | in mind that the first instruction of the sigtramp code is |
| 225 | "pop %eax". If the PC is at this instruction, adjust the |
| 226 | returned value accordingly. */ |
| 227 | sp = read_register (SP_REGNUM); |
| 228 | if (pc == frame->pc) |
| 229 | return sp + 4; |
| 230 | return sp; |
| 231 | } |
| 232 | |
| 233 | pc = i386_linux_rt_sigtramp_start (frame->pc); |
| 234 | if (pc) |
| 235 | { |
| 236 | if (frame->next) |
| 237 | /* If this isn't the top frame, the next frame must be for the |
| 238 | signal handler itself. The sigcontext structure is part of |
| 239 | the user context. A pointer to the user context is passed |
| 240 | as the third argument to the signal handler. */ |
| 241 | return read_memory_integer (frame->next->frame + 16, 4) + 20; |
| 242 | |
| 243 | /* This is the top frame. Again, use the stack pointer to find |
| 244 | the address of the sigcontext structure. */ |
| 245 | return read_memory_integer (read_register (SP_REGNUM) + 8, 4) + 20; |
| 246 | } |
| 247 | |
| 248 | error ("Couldn't recognize signal trampoline."); |
| 249 | return 0; |
| 250 | } |
| 251 | |
| 252 | /* Offset to saved PC in sigcontext, from <asm/sigcontext.h>. */ |
| 253 | #define LINUX_SIGCONTEXT_PC_OFFSET (56) |
| 254 | |
| 255 | /* Assuming FRAME is for a Linux sigtramp routine, return the saved |
| 256 | program counter. */ |
| 257 | |
| 258 | static CORE_ADDR |
| 259 | i386_linux_sigtramp_saved_pc (struct frame_info *frame) |
| 260 | { |
| 261 | CORE_ADDR addr; |
| 262 | addr = i386_linux_sigcontext_addr (frame); |
| 263 | return read_memory_integer (addr + LINUX_SIGCONTEXT_PC_OFFSET, 4); |
| 264 | } |
| 265 | |
| 266 | /* Offset to saved SP in sigcontext, from <asm/sigcontext.h>. */ |
| 267 | #define LINUX_SIGCONTEXT_SP_OFFSET (28) |
| 268 | |
| 269 | /* Assuming FRAME is for a Linux sigtramp routine, return the saved |
| 270 | stack pointer. */ |
| 271 | |
| 272 | static CORE_ADDR |
| 273 | i386_linux_sigtramp_saved_sp (struct frame_info *frame) |
| 274 | { |
| 275 | CORE_ADDR addr; |
| 276 | addr = i386_linux_sigcontext_addr (frame); |
| 277 | return read_memory_integer (addr + LINUX_SIGCONTEXT_SP_OFFSET, 4); |
| 278 | } |
| 279 | |
| 280 | /* Signal trampolines don't have a meaningful frame. As in |
| 281 | "i386/tm-i386.h", the frame pointer value we use is actually the |
| 282 | frame pointer of the calling frame -- that is, the frame which was |
| 283 | in progress when the signal trampoline was entered. GDB mostly |
| 284 | treats this frame pointer value as a magic cookie. We detect the |
| 285 | case of a signal trampoline by looking at the SIGNAL_HANDLER_CALLER |
| 286 | field, which is set based on IN_SIGTRAMP. |
| 287 | |
| 288 | When a signal trampoline is invoked from a frameless function, we |
| 289 | essentially have two frameless functions in a row. In this case, |
| 290 | we use the same magic cookie for three frames in a row. We detect |
| 291 | this case by seeing whether the next frame has |
| 292 | SIGNAL_HANDLER_CALLER set, and, if it does, checking whether the |
| 293 | current frame is actually frameless. In this case, we need to get |
| 294 | the PC by looking at the SP register value stored in the signal |
| 295 | context. |
| 296 | |
| 297 | This should work in most cases except in horrible situations where |
| 298 | a signal occurs just as we enter a function but before the frame |
| 299 | has been set up. */ |
| 300 | |
| 301 | #define FRAMELESS_SIGNAL(frame) \ |
| 302 | ((frame)->next != NULL \ |
| 303 | && (frame)->next->signal_handler_caller \ |
| 304 | && frameless_look_for_prologue (frame)) |
| 305 | |
| 306 | CORE_ADDR |
| 307 | i386_linux_frame_chain (struct frame_info *frame) |
| 308 | { |
| 309 | if (frame->signal_handler_caller || FRAMELESS_SIGNAL (frame)) |
| 310 | return frame->frame; |
| 311 | |
| 312 | if (! inside_entry_file (frame->pc)) |
| 313 | return read_memory_unsigned_integer (frame->frame, 4); |
| 314 | |
| 315 | return 0; |
| 316 | } |
| 317 | |
| 318 | /* Return the saved program counter for FRAME. */ |
| 319 | |
| 320 | CORE_ADDR |
| 321 | i386_linux_frame_saved_pc (struct frame_info *frame) |
| 322 | { |
| 323 | if (frame->signal_handler_caller) |
| 324 | return i386_linux_sigtramp_saved_pc (frame); |
| 325 | |
| 326 | if (FRAMELESS_SIGNAL (frame)) |
| 327 | { |
| 328 | CORE_ADDR sp = i386_linux_sigtramp_saved_sp (frame->next); |
| 329 | return read_memory_unsigned_integer (sp, 4); |
| 330 | } |
| 331 | |
| 332 | return read_memory_unsigned_integer (frame->frame + 4, 4); |
| 333 | } |
| 334 | |
| 335 | /* Immediately after a function call, return the saved pc. */ |
| 336 | |
| 337 | CORE_ADDR |
| 338 | i386_linux_saved_pc_after_call (struct frame_info *frame) |
| 339 | { |
| 340 | if (frame->signal_handler_caller) |
| 341 | return i386_linux_sigtramp_saved_pc (frame); |
| 342 | |
| 343 | return read_memory_unsigned_integer (read_register (SP_REGNUM), 4); |
| 344 | } |
| 345 | \f |
| 346 | |
| 347 | /* Calling functions in shared libraries. */ |
| 348 | /* Find the minimal symbol named NAME, and return both the minsym |
| 349 | struct and its objfile. This probably ought to be in minsym.c, but |
| 350 | everything there is trying to deal with things like C++ and |
| 351 | SOFUN_ADDRESS_MAYBE_TURQUOISE, ... Since this is so simple, it may |
| 352 | be considered too special-purpose for general consumption. */ |
| 353 | |
| 354 | static struct minimal_symbol * |
| 355 | find_minsym_and_objfile (char *name, struct objfile **objfile_p) |
| 356 | { |
| 357 | struct objfile *objfile; |
| 358 | |
| 359 | ALL_OBJFILES (objfile) |
| 360 | { |
| 361 | struct minimal_symbol *msym; |
| 362 | |
| 363 | ALL_OBJFILE_MSYMBOLS (objfile, msym) |
| 364 | { |
| 365 | if (SYMBOL_NAME (msym) |
| 366 | && STREQ (SYMBOL_NAME (msym), name)) |
| 367 | { |
| 368 | *objfile_p = objfile; |
| 369 | return msym; |
| 370 | } |
| 371 | } |
| 372 | } |
| 373 | |
| 374 | return 0; |
| 375 | } |
| 376 | |
| 377 | static CORE_ADDR |
| 378 | skip_hurd_resolver (CORE_ADDR pc) |
| 379 | { |
| 380 | /* The HURD dynamic linker is part of the GNU C library, so many |
| 381 | GNU/Linux distributions use it. (All ELF versions, as far as I |
| 382 | know.) An unresolved PLT entry points to "_dl_runtime_resolve", |
| 383 | which calls "fixup" to patch the PLT, and then passes control to |
| 384 | the function. |
| 385 | |
| 386 | We look for the symbol `_dl_runtime_resolve', and find `fixup' in |
| 387 | the same objfile. If we are at the entry point of `fixup', then |
| 388 | we set a breakpoint at the return address (at the top of the |
| 389 | stack), and continue. |
| 390 | |
| 391 | It's kind of gross to do all these checks every time we're |
| 392 | called, since they don't change once the executable has gotten |
| 393 | started. But this is only a temporary hack --- upcoming versions |
| 394 | of Linux will provide a portable, efficient interface for |
| 395 | debugging programs that use shared libraries. */ |
| 396 | |
| 397 | struct objfile *objfile; |
| 398 | struct minimal_symbol *resolver |
| 399 | = find_minsym_and_objfile ("_dl_runtime_resolve", &objfile); |
| 400 | |
| 401 | if (resolver) |
| 402 | { |
| 403 | struct minimal_symbol *fixup |
| 404 | = lookup_minimal_symbol ("fixup", 0, objfile); |
| 405 | |
| 406 | if (fixup && SYMBOL_VALUE_ADDRESS (fixup) == pc) |
| 407 | return (SAVED_PC_AFTER_CALL (get_current_frame ())); |
| 408 | } |
| 409 | |
| 410 | return 0; |
| 411 | } |
| 412 | |
| 413 | /* See the comments for SKIP_SOLIB_RESOLVER at the top of infrun.c. |
| 414 | This function: |
| 415 | 1) decides whether a PLT has sent us into the linker to resolve |
| 416 | a function reference, and |
| 417 | 2) if so, tells us where to set a temporary breakpoint that will |
| 418 | trigger when the dynamic linker is done. */ |
| 419 | |
| 420 | CORE_ADDR |
| 421 | i386_linux_skip_solib_resolver (CORE_ADDR pc) |
| 422 | { |
| 423 | CORE_ADDR result; |
| 424 | |
| 425 | /* Plug in functions for other kinds of resolvers here. */ |
| 426 | result = skip_hurd_resolver (pc); |
| 427 | if (result) |
| 428 | return result; |
| 429 | |
| 430 | return 0; |
| 431 | } |
| 432 | |
| 433 | /* Fetch (and possibly build) an appropriate link_map_offsets |
| 434 | structure for native Linux/x86 targets using the struct offsets |
| 435 | defined in link.h (but without actual reference to that file). |
| 436 | |
| 437 | This makes it possible to access Linux/x86 shared libraries from a |
| 438 | GDB that was not built on an Linux/x86 host (for cross debugging). */ |
| 439 | |
| 440 | struct link_map_offsets * |
| 441 | i386_linux_svr4_fetch_link_map_offsets (void) |
| 442 | { |
| 443 | static struct link_map_offsets lmo; |
| 444 | static struct link_map_offsets *lmp = NULL; |
| 445 | |
| 446 | if (lmp == NULL) |
| 447 | { |
| 448 | lmp = &lmo; |
| 449 | |
| 450 | lmo.r_debug_size = 8; /* The actual size is 20 bytes, but |
| 451 | this is all we need. */ |
| 452 | lmo.r_map_offset = 4; |
| 453 | lmo.r_map_size = 4; |
| 454 | |
| 455 | lmo.link_map_size = 20; /* The actual size is 552 bytes, but |
| 456 | this is all we need. */ |
| 457 | lmo.l_addr_offset = 0; |
| 458 | lmo.l_addr_size = 4; |
| 459 | |
| 460 | lmo.l_name_offset = 4; |
| 461 | lmo.l_name_size = 4; |
| 462 | |
| 463 | lmo.l_next_offset = 12; |
| 464 | lmo.l_next_size = 4; |
| 465 | |
| 466 | lmo.l_prev_offset = 16; |
| 467 | lmo.l_prev_size = 4; |
| 468 | } |
| 469 | |
| 470 | return lmp; |
| 471 | } |