| 1 | /* Interface to prologue value handling for GDB. |
| 2 | Copyright 2003, 2004, 2005, 2007 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: |
| 18 | |
| 19 | Free Software Foundation, Inc. |
| 20 | 51 Franklin St - Fifth Floor |
| 21 | Boston, MA 02110-1301 |
| 22 | USA */ |
| 23 | |
| 24 | #ifndef PROLOGUE_VALUE_H |
| 25 | #define PROLOGUE_VALUE_H |
| 26 | |
| 27 | /* When we analyze a prologue, we're really doing 'abstract |
| 28 | interpretation' or 'pseudo-evaluation': running the function's code |
| 29 | in simulation, but using conservative approximations of the values |
| 30 | it would have when it actually runs. For example, if our function |
| 31 | starts with the instruction: |
| 32 | |
| 33 | addi r1, 42 # add 42 to r1 |
| 34 | |
| 35 | we don't know exactly what value will be in r1 after executing this |
| 36 | instruction, but we do know it'll be 42 greater than its original |
| 37 | value. |
| 38 | |
| 39 | If we then see an instruction like: |
| 40 | |
| 41 | addi r1, 22 # add 22 to r1 |
| 42 | |
| 43 | we still don't know what r1's value is, but again, we can say it is |
| 44 | now 64 greater than its original value. |
| 45 | |
| 46 | If the next instruction were: |
| 47 | |
| 48 | mov r2, r1 # set r2 to r1's value |
| 49 | |
| 50 | then we can say that r2's value is now the original value of r1 |
| 51 | plus 64. |
| 52 | |
| 53 | It's common for prologues to save registers on the stack, so we'll |
| 54 | need to track the values of stack frame slots, as well as the |
| 55 | registers. So after an instruction like this: |
| 56 | |
| 57 | mov (fp+4), r2 |
| 58 | |
| 59 | then we'd know that the stack slot four bytes above the frame |
| 60 | pointer holds the original value of r1 plus 64. |
| 61 | |
| 62 | And so on. |
| 63 | |
| 64 | Of course, this can only go so far before it gets unreasonable. If |
| 65 | we wanted to be able to say anything about the value of r1 after |
| 66 | the instruction: |
| 67 | |
| 68 | xor r1, r3 # exclusive-or r1 and r3, place result in r1 |
| 69 | |
| 70 | then things would get pretty complex. But remember, we're just |
| 71 | doing a conservative approximation; if exclusive-or instructions |
| 72 | aren't relevant to prologues, we can just say r1's value is now |
| 73 | 'unknown'. We can ignore things that are too complex, if that loss |
| 74 | of information is acceptable for our application. |
| 75 | |
| 76 | So when I say "conservative approximation" here, what I mean is an |
| 77 | approximation that is either accurate, or marked "unknown", but |
| 78 | never inaccurate. |
| 79 | |
| 80 | Once you've reached the current PC, or an instruction that you |
| 81 | don't know how to simulate, you stop. Now you can examine the |
| 82 | state of the registers and stack slots you've kept track of. |
| 83 | |
| 84 | - To see how large your stack frame is, just check the value of the |
| 85 | stack pointer register; if it's the original value of the SP |
| 86 | minus a constant, then that constant is the stack frame's size. |
| 87 | If the SP's value has been marked as 'unknown', then that means |
| 88 | the prologue has done something too complex for us to track, and |
| 89 | we don't know the frame size. |
| 90 | |
| 91 | - To see where we've saved the previous frame's registers, we just |
| 92 | search the values we've tracked --- stack slots, usually, but |
| 93 | registers, too, if you want --- for something equal to the |
| 94 | register's original value. If the ABI suggests a standard place |
| 95 | to save a given register, then we can check there first, but |
| 96 | really, anything that will get us back the original value will |
| 97 | probably work. |
| 98 | |
| 99 | Sure, this takes some work. But prologue analyzers aren't |
| 100 | quick-and-simple pattern patching to recognize a few fixed prologue |
| 101 | forms any more; they're big, hairy functions. Along with inferior |
| 102 | function calls, prologue analysis accounts for a substantial |
| 103 | portion of the time needed to stabilize a GDB port. So I think |
| 104 | it's worthwhile to look for an approach that will be easier to |
| 105 | understand and maintain. In the approach used here: |
| 106 | |
| 107 | - It's easier to see that the analyzer is correct: you just see |
| 108 | whether the analyzer properly (albiet conservatively) simulates |
| 109 | the effect of each instruction. |
| 110 | |
| 111 | - It's easier to extend the analyzer: you can add support for new |
| 112 | instructions, and know that you haven't broken anything that |
| 113 | wasn't already broken before. |
| 114 | |
| 115 | - It's orthogonal: to gather new information, you don't need to |
| 116 | complicate the code for each instruction. As long as your domain |
| 117 | of conservative values is already detailed enough to tell you |
| 118 | what you need, then all the existing instruction simulations are |
| 119 | already gathering the right data for you. |
| 120 | |
| 121 | A 'struct prologue_value' is a conservative approximation of the |
| 122 | real value the register or stack slot will have. */ |
| 123 | |
| 124 | struct prologue_value { |
| 125 | |
| 126 | /* What sort of value is this? This determines the interpretation |
| 127 | of subsequent fields. */ |
| 128 | enum { |
| 129 | |
| 130 | /* We don't know anything about the value. This is also used for |
| 131 | values we could have kept track of, when doing so would have |
| 132 | been too complex and we don't want to bother. The bottom of |
| 133 | our lattice. */ |
| 134 | pvk_unknown, |
| 135 | |
| 136 | /* A known constant. K is its value. */ |
| 137 | pvk_constant, |
| 138 | |
| 139 | /* The value that register REG originally had *UPON ENTRY TO THE |
| 140 | FUNCTION*, plus K. If K is zero, this means, obviously, just |
| 141 | the value REG had upon entry to the function. REG is a GDB |
| 142 | register number. Before we start interpreting, we initialize |
| 143 | every register R to { pvk_register, R, 0 }. */ |
| 144 | pvk_register, |
| 145 | |
| 146 | } kind; |
| 147 | |
| 148 | /* The meanings of the following fields depend on 'kind'; see the |
| 149 | comments for the specific 'kind' values. */ |
| 150 | int reg; |
| 151 | CORE_ADDR k; |
| 152 | }; |
| 153 | |
| 154 | typedef struct prologue_value pv_t; |
| 155 | |
| 156 | |
| 157 | /* Return the unknown prologue value --- { pvk_unknown, ?, ? }. */ |
| 158 | pv_t pv_unknown (void); |
| 159 | |
| 160 | /* Return the prologue value representing the constant K. */ |
| 161 | pv_t pv_constant (CORE_ADDR k); |
| 162 | |
| 163 | /* Return the prologue value representing the original value of |
| 164 | register REG, plus the constant K. */ |
| 165 | pv_t pv_register (int reg, CORE_ADDR k); |
| 166 | |
| 167 | |
| 168 | /* Return conservative approximations of the results of the following |
| 169 | operations. */ |
| 170 | pv_t pv_add (pv_t a, pv_t b); /* a + b */ |
| 171 | pv_t pv_add_constant (pv_t v, CORE_ADDR k); /* a + k */ |
| 172 | pv_t pv_subtract (pv_t a, pv_t b); /* a - b */ |
| 173 | pv_t pv_logical_and (pv_t a, pv_t b); /* a & b */ |
| 174 | |
| 175 | |
| 176 | /* Return non-zero iff A and B are identical expressions. |
| 177 | |
| 178 | This is not the same as asking if the two values are equal; the |
| 179 | result of such a comparison would have to be a pv_boolean, and |
| 180 | asking whether two 'unknown' values were equal would give you |
| 181 | pv_maybe. Same for comparing, say, { pvk_register, R1, 0 } and { |
| 182 | pvk_register, R2, 0}. |
| 183 | |
| 184 | Instead, this function asks whether the two representations are the |
| 185 | same. */ |
| 186 | int pv_is_identical (pv_t a, pv_t b); |
| 187 | |
| 188 | |
| 189 | /* Return non-zero if A is known to be a constant. */ |
| 190 | int pv_is_constant (pv_t a); |
| 191 | |
| 192 | /* Return non-zero if A is the original value of register number R |
| 193 | plus some constant, zero otherwise. */ |
| 194 | int pv_is_register (pv_t a, int r); |
| 195 | |
| 196 | |
| 197 | /* Return non-zero if A is the original value of register R plus the |
| 198 | constant K. */ |
| 199 | int pv_is_register_k (pv_t a, int r, CORE_ADDR k); |
| 200 | |
| 201 | /* A conservative boolean type, including "maybe", when we can't |
| 202 | figure out whether something is true or not. */ |
| 203 | enum pv_boolean { |
| 204 | pv_maybe, |
| 205 | pv_definite_yes, |
| 206 | pv_definite_no, |
| 207 | }; |
| 208 | |
| 209 | |
| 210 | /* Decide whether a reference to SIZE bytes at ADDR refers exactly to |
| 211 | an element of an array. The array starts at ARRAY_ADDR, and has |
| 212 | ARRAY_LEN values of ELT_SIZE bytes each. If ADDR definitely does |
| 213 | refer to an array element, set *I to the index of the referenced |
| 214 | element in the array, and return pv_definite_yes. If it definitely |
| 215 | doesn't, return pv_definite_no. If we can't tell, return pv_maybe. |
| 216 | |
| 217 | If the reference does touch the array, but doesn't fall exactly on |
| 218 | an element boundary, or doesn't refer to the whole element, return |
| 219 | pv_maybe. */ |
| 220 | enum pv_boolean pv_is_array_ref (pv_t addr, CORE_ADDR size, |
| 221 | pv_t array_addr, CORE_ADDR array_len, |
| 222 | CORE_ADDR elt_size, |
| 223 | int *i); |
| 224 | |
| 225 | |
| 226 | /* A 'struct pv_area' keeps track of values stored in a particular |
| 227 | region of memory. */ |
| 228 | struct pv_area; |
| 229 | |
| 230 | /* Create a new area, tracking stores relative to the original value |
| 231 | of BASE_REG. If BASE_REG is SP, then this effectively records the |
| 232 | contents of the stack frame: the original value of the SP is the |
| 233 | frame's CFA, or some constant offset from it. |
| 234 | |
| 235 | Stores to constant addresses, unknown addresses, or to addresses |
| 236 | relative to registers other than BASE_REG will trash this area; see |
| 237 | pv_area_store_would_trash. */ |
| 238 | struct pv_area *make_pv_area (int base_reg); |
| 239 | |
| 240 | /* Free AREA. */ |
| 241 | void free_pv_area (struct pv_area *area); |
| 242 | |
| 243 | |
| 244 | /* Register a cleanup to free AREA. */ |
| 245 | struct cleanup *make_cleanup_free_pv_area (struct pv_area *area); |
| 246 | |
| 247 | |
| 248 | /* Store the SIZE-byte value VALUE at ADDR in AREA. |
| 249 | |
| 250 | If ADDR is not relative to the same base register we used in |
| 251 | creating AREA, then we can't tell which values here the stored |
| 252 | value might overlap, and we'll have to mark everything as |
| 253 | unknown. */ |
| 254 | void pv_area_store (struct pv_area *area, |
| 255 | pv_t addr, |
| 256 | CORE_ADDR size, |
| 257 | pv_t value); |
| 258 | |
| 259 | /* Return the SIZE-byte value at ADDR in AREA. This may return |
| 260 | pv_unknown (). */ |
| 261 | pv_t pv_area_fetch (struct pv_area *area, pv_t addr, CORE_ADDR size); |
| 262 | |
| 263 | /* Return true if storing to address ADDR in AREA would force us to |
| 264 | mark the contents of the entire area as unknown. This could happen |
| 265 | if, say, ADDR is unknown, since we could be storing anywhere. Or, |
| 266 | it could happen if ADDR is relative to a different register than |
| 267 | the other stores base register, since we don't know the relative |
| 268 | values of the two registers. |
| 269 | |
| 270 | If you've reached such a store, it may be better to simply stop the |
| 271 | prologue analysis, and return the information you've gathered, |
| 272 | instead of losing all that information, most of which is probably |
| 273 | okay. */ |
| 274 | int pv_area_store_would_trash (struct pv_area *area, pv_t addr); |
| 275 | |
| 276 | |
| 277 | /* Search AREA for the original value of REGISTER. If we can't find |
| 278 | it, return zero; if we can find it, return a non-zero value, and if |
| 279 | OFFSET_P is non-zero, set *OFFSET_P to the register's offset within |
| 280 | AREA. GDBARCH is the architecture of which REGISTER is a member. |
| 281 | |
| 282 | In the worst case, this takes time proportional to the number of |
| 283 | items stored in AREA. If you plan to gather a lot of information |
| 284 | about registers saved in AREA, consider calling pv_area_scan |
| 285 | instead, and collecting all your information in one pass. */ |
| 286 | int pv_area_find_reg (struct pv_area *area, |
| 287 | struct gdbarch *gdbarch, |
| 288 | int reg, |
| 289 | CORE_ADDR *offset_p); |
| 290 | |
| 291 | |
| 292 | /* For every part of AREA whose value we know, apply FUNC to CLOSURE, |
| 293 | the value's address, its size, and the value itself. */ |
| 294 | void pv_area_scan (struct pv_area *area, |
| 295 | void (*func) (void *closure, |
| 296 | pv_t addr, |
| 297 | CORE_ADDR size, |
| 298 | pv_t value), |
| 299 | void *closure); |
| 300 | |
| 301 | |
| 302 | #endif /* PROLOGUE_VALUE_H */ |