| 1 | /* GNU/Linux on ARM target support. |
| 2 | Copyright 1999, 2000 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 "target.h" |
| 23 | #include "value.h" |
| 24 | #include "gdbtypes.h" |
| 25 | #include "floatformat.h" |
| 26 | |
| 27 | #ifdef GET_LONGJMP_TARGET |
| 28 | |
| 29 | /* Figure out where the longjmp will land. We expect that we have |
| 30 | just entered longjmp and haven't yet altered r0, r1, so the |
| 31 | arguments are still in the registers. (A1_REGNUM) points at the |
| 32 | jmp_buf structure from which we extract the pc (JB_PC) that we will |
| 33 | land at. The pc is copied into ADDR. This routine returns true on |
| 34 | success. */ |
| 35 | |
| 36 | #define LONGJMP_TARGET_SIZE sizeof(int) |
| 37 | #define JB_ELEMENT_SIZE sizeof(int) |
| 38 | #define JB_SL 18 |
| 39 | #define JB_FP 19 |
| 40 | #define JB_SP 20 |
| 41 | #define JB_PC 21 |
| 42 | |
| 43 | int |
| 44 | arm_get_longjmp_target (CORE_ADDR * pc) |
| 45 | { |
| 46 | CORE_ADDR jb_addr; |
| 47 | char buf[LONGJMP_TARGET_SIZE]; |
| 48 | |
| 49 | jb_addr = read_register (A1_REGNUM); |
| 50 | |
| 51 | if (target_read_memory (jb_addr + JB_PC * JB_ELEMENT_SIZE, buf, |
| 52 | LONGJMP_TARGET_SIZE)) |
| 53 | return 0; |
| 54 | |
| 55 | *pc = extract_address (buf, LONGJMP_TARGET_SIZE); |
| 56 | return 1; |
| 57 | } |
| 58 | |
| 59 | #endif /* GET_LONGJMP_TARGET */ |
| 60 | |
| 61 | /* Extract from an array REGBUF containing the (raw) register state |
| 62 | a function return value of type TYPE, and copy that, in virtual format, |
| 63 | into VALBUF. */ |
| 64 | |
| 65 | void |
| 66 | arm_linux_extract_return_value (struct type *type, |
| 67 | char regbuf[REGISTER_BYTES], |
| 68 | char *valbuf) |
| 69 | { |
| 70 | /* ScottB: This needs to be looked at to handle the different |
| 71 | floating point emulators on ARM Linux. Right now the code |
| 72 | assumes that fetch inferior registers does the right thing for |
| 73 | GDB. I suspect this won't handle NWFPE registers correctly, nor |
| 74 | will the default ARM version (arm_extract_return_value()). */ |
| 75 | |
| 76 | int regnum = (TYPE_CODE_FLT == TYPE_CODE (type)) ? F0_REGNUM : A1_REGNUM; |
| 77 | memcpy (valbuf, ®buf[REGISTER_BYTE (regnum)], TYPE_LENGTH (type)); |
| 78 | } |
| 79 | |
| 80 | /* Note: ScottB |
| 81 | |
| 82 | This function does not support passing parameters using the FPA |
| 83 | variant of the APCS. It passes any floating point arguments in the |
| 84 | general registers and/or on the stack. |
| 85 | |
| 86 | FIXME: This and arm_push_arguments should be merged. However this |
| 87 | function breaks on a little endian host, big endian target |
| 88 | using the COFF file format. ELF is ok. |
| 89 | |
| 90 | ScottB. */ |
| 91 | |
| 92 | /* Addresses for calling Thumb functions have the bit 0 set. |
| 93 | Here are some macros to test, set, or clear bit 0 of addresses. */ |
| 94 | #define IS_THUMB_ADDR(addr) ((addr) & 1) |
| 95 | #define MAKE_THUMB_ADDR(addr) ((addr) | 1) |
| 96 | #define UNMAKE_THUMB_ADDR(addr) ((addr) & ~1) |
| 97 | |
| 98 | CORE_ADDR |
| 99 | arm_linux_push_arguments (int nargs, value_ptr * args, CORE_ADDR sp, |
| 100 | int struct_return, CORE_ADDR struct_addr) |
| 101 | { |
| 102 | char *fp; |
| 103 | int argnum, argreg, nstack_size; |
| 104 | |
| 105 | /* Walk through the list of args and determine how large a temporary |
| 106 | stack is required. Need to take care here as structs may be |
| 107 | passed on the stack, and we have to to push them. */ |
| 108 | nstack_size = -4 * REGISTER_SIZE; /* Some arguments go into A1-A4. */ |
| 109 | |
| 110 | if (struct_return) /* The struct address goes in A1. */ |
| 111 | nstack_size += REGISTER_SIZE; |
| 112 | |
| 113 | /* Walk through the arguments and add their size to nstack_size. */ |
| 114 | for (argnum = 0; argnum < nargs; argnum++) |
| 115 | { |
| 116 | int len; |
| 117 | struct type *arg_type; |
| 118 | |
| 119 | arg_type = check_typedef (VALUE_TYPE (args[argnum])); |
| 120 | len = TYPE_LENGTH (arg_type); |
| 121 | |
| 122 | /* ANSI C code passes float arguments as integers, K&R code |
| 123 | passes float arguments as doubles. Correct for this here. */ |
| 124 | if (TYPE_CODE_FLT == TYPE_CODE (arg_type) && REGISTER_SIZE == len) |
| 125 | nstack_size += FP_REGISTER_VIRTUAL_SIZE; |
| 126 | else |
| 127 | nstack_size += len; |
| 128 | } |
| 129 | |
| 130 | /* Allocate room on the stack, and initialize our stack frame |
| 131 | pointer. */ |
| 132 | fp = NULL; |
| 133 | if (nstack_size > 0) |
| 134 | { |
| 135 | sp -= nstack_size; |
| 136 | fp = (char *) sp; |
| 137 | } |
| 138 | |
| 139 | /* Initialize the integer argument register pointer. */ |
| 140 | argreg = A1_REGNUM; |
| 141 | |
| 142 | /* The struct_return pointer occupies the first parameter passing |
| 143 | register. */ |
| 144 | if (struct_return) |
| 145 | write_register (argreg++, struct_addr); |
| 146 | |
| 147 | /* Process arguments from left to right. Store as many as allowed |
| 148 | in the parameter passing registers (A1-A4), and save the rest on |
| 149 | the temporary stack. */ |
| 150 | for (argnum = 0; argnum < nargs; argnum++) |
| 151 | { |
| 152 | int len; |
| 153 | char *val; |
| 154 | double dbl_arg; |
| 155 | CORE_ADDR regval; |
| 156 | enum type_code typecode; |
| 157 | struct type *arg_type, *target_type; |
| 158 | |
| 159 | arg_type = check_typedef (VALUE_TYPE (args[argnum])); |
| 160 | target_type = TYPE_TARGET_TYPE (arg_type); |
| 161 | len = TYPE_LENGTH (arg_type); |
| 162 | typecode = TYPE_CODE (arg_type); |
| 163 | val = (char *) VALUE_CONTENTS (args[argnum]); |
| 164 | |
| 165 | /* ANSI C code passes float arguments as integers, K&R code |
| 166 | passes float arguments as doubles. The .stabs record for |
| 167 | for ANSI prototype floating point arguments records the |
| 168 | type as FP_INTEGER, while a K&R style (no prototype) |
| 169 | .stabs records the type as FP_FLOAT. In this latter case |
| 170 | the compiler converts the float arguments to double before |
| 171 | calling the function. */ |
| 172 | if (TYPE_CODE_FLT == typecode && REGISTER_SIZE == len) |
| 173 | { |
| 174 | /* Float argument in buffer is in host format. Read it and |
| 175 | convert to DOUBLEST, and store it in target double. */ |
| 176 | DOUBLEST dblval; |
| 177 | |
| 178 | len = TARGET_DOUBLE_BIT / TARGET_CHAR_BIT; |
| 179 | floatformat_to_doublest (HOST_FLOAT_FORMAT, val, &dblval); |
| 180 | store_floating (&dbl_arg, len, dblval); |
| 181 | val = (char *) &dbl_arg; |
| 182 | } |
| 183 | |
| 184 | /* If the argument is a pointer to a function, and it is a Thumb |
| 185 | function, set the low bit of the pointer. */ |
| 186 | if (TYPE_CODE_PTR == typecode |
| 187 | && NULL != target_type |
| 188 | && TYPE_CODE_FUNC == TYPE_CODE (target_type)) |
| 189 | { |
| 190 | CORE_ADDR regval = extract_address (val, len); |
| 191 | if (arm_pc_is_thumb (regval)) |
| 192 | store_address (val, len, MAKE_THUMB_ADDR (regval)); |
| 193 | } |
| 194 | |
| 195 | /* Copy the argument to general registers or the stack in |
| 196 | register-sized pieces. Large arguments are split between |
| 197 | registers and stack. */ |
| 198 | while (len > 0) |
| 199 | { |
| 200 | int partial_len = len < REGISTER_SIZE ? len : REGISTER_SIZE; |
| 201 | |
| 202 | if (argreg <= ARM_LAST_ARG_REGNUM) |
| 203 | { |
| 204 | /* It's an argument being passed in a general register. */ |
| 205 | regval = extract_address (val, partial_len); |
| 206 | write_register (argreg++, regval); |
| 207 | } |
| 208 | else |
| 209 | { |
| 210 | /* Push the arguments onto the stack. */ |
| 211 | write_memory ((CORE_ADDR) fp, val, REGISTER_SIZE); |
| 212 | fp += REGISTER_SIZE; |
| 213 | } |
| 214 | |
| 215 | len -= partial_len; |
| 216 | val += partial_len; |
| 217 | } |
| 218 | } |
| 219 | |
| 220 | /* Return adjusted stack pointer. */ |
| 221 | return sp; |
| 222 | } |
| 223 | |
| 224 | /* |
| 225 | Dynamic Linking on ARM Linux |
| 226 | ---------------------------- |
| 227 | |
| 228 | Note: PLT = procedure linkage table |
| 229 | GOT = global offset table |
| 230 | |
| 231 | As much as possible, ELF dynamic linking defers the resolution of |
| 232 | jump/call addresses until the last minute. The technique used is |
| 233 | inspired by the i386 ELF design, and is based on the following |
| 234 | constraints. |
| 235 | |
| 236 | 1) The calling technique should not force a change in the assembly |
| 237 | code produced for apps; it MAY cause changes in the way assembly |
| 238 | code is produced for position independent code (i.e. shared |
| 239 | libraries). |
| 240 | |
| 241 | 2) The technique must be such that all executable areas must not be |
| 242 | modified; and any modified areas must not be executed. |
| 243 | |
| 244 | To do this, there are three steps involved in a typical jump: |
| 245 | |
| 246 | 1) in the code |
| 247 | 2) through the PLT |
| 248 | 3) using a pointer from the GOT |
| 249 | |
| 250 | When the executable or library is first loaded, each GOT entry is |
| 251 | initialized to point to the code which implements dynamic name |
| 252 | resolution and code finding. This is normally a function in the |
| 253 | program interpreter (on ARM Linux this is usually ld-linux.so.2, |
| 254 | but it does not have to be). On the first invocation, the function |
| 255 | is located and the GOT entry is replaced with the real function |
| 256 | address. Subsequent calls go through steps 1, 2 and 3 and end up |
| 257 | calling the real code. |
| 258 | |
| 259 | 1) In the code: |
| 260 | |
| 261 | b function_call |
| 262 | bl function_call |
| 263 | |
| 264 | This is typical ARM code using the 26 bit relative branch or branch |
| 265 | and link instructions. The target of the instruction |
| 266 | (function_call is usually the address of the function to be called. |
| 267 | In position independent code, the target of the instruction is |
| 268 | actually an entry in the PLT when calling functions in a shared |
| 269 | library. Note that this call is identical to a normal function |
| 270 | call, only the target differs. |
| 271 | |
| 272 | 2) In the PLT: |
| 273 | |
| 274 | The PLT is a synthetic area, created by the linker. It exists in |
| 275 | both executables and libraries. It is an array of stubs, one per |
| 276 | imported function call. It looks like this: |
| 277 | |
| 278 | PLT[0]: |
| 279 | str lr, [sp, #-4]! @push the return address (lr) |
| 280 | ldr lr, [pc, #16] @load from 6 words ahead |
| 281 | add lr, pc, lr @form an address for GOT[0] |
| 282 | ldr pc, [lr, #8]! @jump to the contents of that addr |
| 283 | |
| 284 | The return address (lr) is pushed on the stack and used for |
| 285 | calculations. The load on the second line loads the lr with |
| 286 | &GOT[3] - . - 20. The addition on the third leaves: |
| 287 | |
| 288 | lr = (&GOT[3] - . - 20) + (. + 8) |
| 289 | lr = (&GOT[3] - 12) |
| 290 | lr = &GOT[0] |
| 291 | |
| 292 | On the fourth line, the pc and lr are both updated, so that: |
| 293 | |
| 294 | pc = GOT[2] |
| 295 | lr = &GOT[0] + 8 |
| 296 | = &GOT[2] |
| 297 | |
| 298 | NOTE: PLT[0] borrows an offset .word from PLT[1]. This is a little |
| 299 | "tight", but allows us to keep all the PLT entries the same size. |
| 300 | |
| 301 | PLT[n+1]: |
| 302 | ldr ip, [pc, #4] @load offset from gotoff |
| 303 | add ip, pc, ip @add the offset to the pc |
| 304 | ldr pc, [ip] @jump to that address |
| 305 | gotoff: .word GOT[n+3] - . |
| 306 | |
| 307 | The load on the first line, gets an offset from the fourth word of |
| 308 | the PLT entry. The add on the second line makes ip = &GOT[n+3], |
| 309 | which contains either a pointer to PLT[0] (the fixup trampoline) or |
| 310 | a pointer to the actual code. |
| 311 | |
| 312 | 3) In the GOT: |
| 313 | |
| 314 | The GOT contains helper pointers for both code (PLT) fixups and |
| 315 | data fixups. The first 3 entries of the GOT are special. The next |
| 316 | M entries (where M is the number of entries in the PLT) belong to |
| 317 | the PLT fixups. The next D (all remaining) entries belong to |
| 318 | various data fixups. The actual size of the GOT is 3 + M + D. |
| 319 | |
| 320 | The GOT is also a synthetic area, created by the linker. It exists |
| 321 | in both executables and libraries. When the GOT is first |
| 322 | initialized , all the GOT entries relating to PLT fixups are |
| 323 | pointing to code back at PLT[0]. |
| 324 | |
| 325 | The special entries in the GOT are: |
| 326 | |
| 327 | GOT[0] = linked list pointer used by the dynamic loader |
| 328 | GOT[1] = pointer to the reloc table for this module |
| 329 | GOT[2] = pointer to the fixup/resolver code |
| 330 | |
| 331 | The first invocation of function call comes through and uses the |
| 332 | fixup/resolver code. On the entry to the fixup/resolver code: |
| 333 | |
| 334 | ip = &GOT[n+3] |
| 335 | lr = &GOT[2] |
| 336 | stack[0] = return address (lr) of the function call |
| 337 | [r0, r1, r2, r3] are still the arguments to the function call |
| 338 | |
| 339 | This is enough information for the fixup/resolver code to work |
| 340 | with. Before the fixup/resolver code returns, it actually calls |
| 341 | the requested function and repairs &GOT[n+3]. */ |
| 342 | |
| 343 | CORE_ADDR |
| 344 | arm_skip_solib_resolver (CORE_ADDR pc) |
| 345 | { |
| 346 | /* FIXME */ |
| 347 | return 0; |
| 348 | } |
| 349 | |
| 350 | void |
| 351 | _initialize_arm_linux_tdep (void) |
| 352 | { |
| 353 | } |