1 //: Labels are defined by ending names with a ':'. This layer will compute
  2 //: displacements for labels, and compute the offset for instructions using them.
  3 //:
  4 //: We won't check this, but our convention will be that jump targets will
  5 //: start with a '$', while functions will not. Function names will never be
  6 //: jumped to, and jump targets will never be called.
  7 
  8 //: We're introducing non-number names for the first time, so it's worth
  9 //: laying down some ground rules all transforms will follow, so things don't
 10 //: get too confusing:
 11 //:   - if it starts with a digit, it's treated as a number. If it can't be
 12 //:     parsed as hex it will raise an error.
 13 //:   - if it starts with '-' it's treated as a number.
 14 //:   - if it starts with '0x' it's treated as a number.
 15 //:   - if it's two characters long, it can't be a name. Either it's a hex
 16 //:     byte, or it raises an error.
 17 //: That's it. Names can start with any non-digit that isn't a dash. They can
 18 //: be a single character long. 'a' is not a hex number, it's a variable.
 19 //: Later layers may add more conventions partitioning the space of names. But
 20 //: the above rules will remain inviolate.
 21 
 22 :(before "End looks_like_hex_int(s) Detectors")
 23 if (SIZE(s) == 2) return true;
 24 
 25 :(scenarios transform)
 26 :(scenario pack_immediate_ignores_single_byte_nondigit_operand)
 27 % Hide_errors = true;
 28 == 0x1
 29 b9/copy  a/imm32
 30 +transform: packing instruction 'b9/copy a/imm32'
 31 # no change (we're just not printing metadata to the trace)
 32 +transform: instruction after packing: 'b9 a'
 33 
 34 :(scenario pack_immediate_ignores_3_hex_digit_operand)
 35 % Hide_errors = true;
 36 == 0x1
 37 b9/copy  aaa/imm32
 38 +transform: packing instruction 'b9/copy aaa/imm32'
 39 # no change (we're just not printing metadata to the trace)
 40 +transform: instruction after packing: 'b9 aaa'
 41 
 42 :(scenario pack_immediate_ignores_non_hex_operand)
 43 % Hide_errors = true;
 44 == 0x1
 45 b9/copy xxx/imm32
 46 +transform: packing instruction 'b9/copy xxx/imm32'
 47 # no change (we're just not printing metadata to the trace)
 48 +transform: instruction after packing: 'b9 xxx'
 49 
 50 //: a helper we'll find handy later
 51 :(code)
 52 void check_valid_name(const string& s) {
 53   if (s.empty()) {
 54     raise << "empty name!\n" << end();
 55     return;
 56   }
 57   if (s.at(0) == '-')
 58     raise << "'" << s << "' starts with '-', which can be confused with a negative number; use a different name\n" << end();
 59   if (s.substr(0, 2) == "0x") {
 60     raise << "'" << s << "' looks like a hex number; use a different name\n" << end();
 61     return;
 62   }
 63   if (isdigit(s.at(0)))
 64     raise << "'" << s << "' starts with a digit, and so can be confused with a negative number; use a different name.\n" << end();
 65   if (SIZE(s) == 2)
 66     raise << "'" << s << "' is two characters long which can look like raw hex bytes at a glance; use a different name\n" << end();
 67 }
 68 
 69 //: Now that that's done, let's start using names as labels.
 70 
 71 :(scenario map_label)
 72 == 0x1
 73 loop:
 74   05  0x0d0c0b0a/imm32
 75 +transform: label 'loop' is at address 1
 76 
 77 :(before "End Level-2 Transforms")
 78 Transform.push_back(rewrite_labels);
 79 :(code)
 80 void rewrite_labels(program& p) {
 81   trace(99, "transform") << "-- rewrite labels" << end();
 82   if (p.segments.empty()) return;
 83   segment& code = p.segments.at(0);
 84   map<string, int32_t> byte_index;  // values are unsigned, but we're going to do subtractions on them so they need to fit in 31 bits
 85   compute_byte_indices_for_labels(code, byte_index);
 86   if (trace_contains_errors()) return;
 87   drop_labels(code);
 88   if (trace_contains_errors()) return;
 89   replace_labels_with_displacements(code, byte_index);
 90 }
 91 
 92 void compute_byte_indices_for_labels(const segment& code, map<string, int32_t>& byte_index) {
 93   int current_byte = 0;
 94   for (int i = 0;  i < SIZE(code.lines);  ++i) {
 95     const line& inst = code.lines.at(i);
 96     for (int j = 0;  j < SIZE(inst.words);  ++j) {
 97       const word& curr = inst.words.at(j);
 98       // hack: if we have any operand metadata left after previous transforms,
 99       // deduce its size
100       // Maybe we should just move this transform to before instruction
101       // packing, and deduce the size of *all* operands. But then we'll also
102       // have to deal with bitfields.
103       if (has_operand_metadata(curr, "disp32") || has_operand_metadata(curr, "imm32")) {
104         if (*curr.data.rbegin() == ':')
105           raise << "'" << to_string(inst) << "': don't use ':' when jumping to labels\n" << end();
106         current_byte += 4;
107       }
108       else if (has_operand_metadata(curr, "disp16")) {
109         if (*curr.data.rbegin() == ':')
110           raise << "'" << to_string(inst) << "': don't use ':' when jumping to labels\n" << end();
111         current_byte += 2;
112       }
113       // automatically handle /disp8 and /imm8 here
114       else if (*curr.data.rbegin() != ':') {
115         ++current_byte;
116       }
117       else {
118         string label = drop_last(curr.data);
119         // ensure labels look sufficiently different from raw hex
120         check_valid_name(label);
121         if (trace_contains_errors()) return;
122         if (contains_any_operand_metadata(curr))
123           raise << "'" << to_string(inst) << "': label definition (':') not allowed in operand\n" << end();
124         if (j > 0)
125           raise << "'" << to_string(inst) << "': labels can only be the first word in a line.\n" << end();
126         if (Map_file.is_open())
127           Map_file << "0x" << HEXWORD << (code.start + current_byte) << ' ' << label << '\n';
128         if (contains_key(byte_index, label)) {
129           raise << "duplicate label '" << label << "'\n" << end();
130           return;
131         }
132         put(byte_index, label, current_byte);
133         trace(99, "transform") << "label '" << label << "' is at address " << (current_byte+code.start) << end();
134         // no modifying current_byte; label definitions won't be in the final binary
135       }
136     }
137   }
138 }
139 
140 :(before "End Globals")
141 bool Dump_map = false;  // currently used only by 'subx translate'
142 ofstream Map_file;
143 :(before "End Commandline Options")
144 else if (is_equal(*arg, "--map")) {
145   Dump_map = true;
146   // End --map Settings
147 }
148 //: wait to open "map" for writing until we're sure we aren't trying to read it
149 :(after "Begin subx translate")
150 if (Dump_map)
151   Map_file.open("map");
152 :(before "End subx translate")
153 if (Dump_map)
154   Map_file.close();
155 
156 :(code)
157 void drop_labels(segment& code) {
158   for (int i = 0;  i < SIZE(code.lines);  ++i) {
159     line& inst = code.lines.at(i);
160     vector<word>::iterator new_end = remove_if(inst.words.begin(), inst.words.end(), is_label);
161     inst.words.erase(new_end, inst.words.end());
162   }
163 }
164 
165 bool is_label(const word& w) {
166   return *w.data.rbegin() == ':';
167 }
168 
169 void replace_labels_with_displacements(segment& code, const map<string, int32_t>& byte_index) {
170   int32_t byte_index_next_instruction_starts_at = 0;
171   for (int i = 0;  i < SIZE(code.lines);  ++i) {
172     line& inst = code.lines.at(i);
173     byte_index_next_instruction_starts_at += num_bytes(inst);
174     line new_inst;
175     for (int j = 0;  j < SIZE(inst.words);  ++j) {
176       const word& curr = inst.words.at(j);
177       if (contains_key(byte_index, curr.data)) {
178         int32_t displacement = static_cast<int32_t>(get(byte_index, curr.data)) - byte_index_next_instruction_starts_at;
179         if (has_operand_metadata(curr, "disp8")) {
180           if (displacement > 0xff || displacement < -0x7f)
181             raise << "'" << to_string(inst) << "': label too far away for displacement " << std::hex << displacement << " to fit in 8 bits\n" << end();
182           else
183             emit_hex_bytes(new_inst, displacement, 1);
184         }
185         else if (has_operand_metadata(curr, "disp16")) {
186           if (displacement > 0xffff || displacement < -0x7fff)
187             raise << "'" << to_string(inst) << "': label too far away for displacement " << std::hex << displacement << " to fit in 16 bits\n" << end();
188           else
189             emit_hex_bytes(new_inst, displacement, 2);
190         }
191         else if (has_operand_metadata(curr, "disp32")) {
192           emit_hex_bytes(new_inst, displacement, 4);
193         }
194       }
195       else {
196         new_inst.words.push_back(curr);
197       }
198     }
199     inst.words.swap(new_inst.words);
200     trace(99, "transform") << "instruction after transform: '" << data_to_string(inst) << "'" << end();
201   }
202 }
203 
204 string data_to_string(const line& inst) {
205   ostringstream out;
206   for (int i = 0;  i < SIZE(inst.words);  ++i) {
207     if (i > 0) out << ' ';
208     out << inst.words.at(i).data;
209   }
210   return out.str();
211 }
212 
213 string drop_last(const string& s) {
214   return string(s.begin(), --s.end());
215 }
216 
217 //: Label definitions must be the first word on a line. No jumping inside
218 //: instructions.
219 //: They should also be the only word on a line.
220 //: However, you can absolutely have multiple labels map to the same address,
221 //: as long as they're on separate lines.
222 
223 :(scenario multiple_labels_at)
224 == 0x1
225 # address 1
226 loop:
227  $loop2:
228 # address 1 (labels take up no space)
229     05  0x0d0c0b0a/imm32
230 # address 6
231     eb  $loop2/disp8
232 # address 8
233     eb  $loop3/disp8
234 # address 0xa
235  $loop3:
236 +transform: label 'loop' is at address 1
237 +transform: label '$loop2' is at address 1
238 +transform: label '$loop3' is at address a
239 # first jump is to -7
240 +transform: instruction after transform: 'eb f9'
241 # second jump is to 0 (fall through)
242 +transform: instruction after transform: 'eb 00'
243 
244 :(scenario duplicate_label)
245 % Hide_errors = true;
246 == 0x1
247 loop:
248 loop:
249     05  0x0d0c0b0a/imm32
250 +error: duplicate label 'loop'
251 
252 :(scenario label_too_short)
253 % Hide_errors = true;
254 == 0x1
255 xz:
256   05  0x0d0c0b0a/imm32
257 +error: 'xz' is two characters long which can look like raw hex bytes at a glance; use a different name
258 
259 :(scenario label_hex)
260 % Hide_errors = true;
261 == 0x1
262 0xab:
263   05  0x0d0c0b0a/imm32
264 +error: '0xab' looks like a hex number; use a different name
265 
266 :(scenario label_negative_hex)
267 % Hide_errors = true;
268 == 0x1
269  -a:  # indent to avoid looking like a trace_should_not_contain command for this scenario
270     05  0x0d0c0b0a/imm32
271 +error: '-a' starts with '-', which can be confused with a negative number; use a different name
272 
273 //: now that we have labels, we need to adjust segment size computation to
274 //: ignore them.
275 
276 :(scenario segment_size_ignores_labels)
277 == code  # 0x09000074
278   05/add  0x0d0c0b0a/imm32  # 5 bytes
279 foo:                      # 0 bytes
280 == data  # 0x0a000079
281 bar:
282   00
283 +transform: segment 1 begins at address 0x0a000079
284 
285 :(before "End size_of(word w) Special-cases")
286 else if (is_label(w))
287   return 0;