riscv/
.clang-format
570 B
.gitignore
5 B
.gitsigners
112 B
LICENSE
1.1 KiB
Makefile
1.1 KiB
README
3.3 KiB
color.h
567 B
emulator.c
79.7 KiB
io.c
1.1 KiB
io.h
444 B
jit.c
32.6 KiB
jit.h
5.0 KiB
riscv.c
12.0 KiB
riscv.h
12.0 KiB
types.h
1.0 KiB
emulator.c
raw
| 1 | #include <errno.h> |
| 2 | #include <fcntl.h> |
| 3 | #include <limits.h> |
| 4 | #include <stdint.h> |
| 5 | #include <stdio.h> |
| 6 | #include <stdlib.h> |
| 7 | #include <string.h> |
| 8 | #include <sys/ioctl.h> |
| 9 | #include <termios.h> |
| 10 | #include <unistd.h> |
| 11 | |
| 12 | #include "color.h" |
| 13 | #include "io.h" |
| 14 | #include "jit.h" |
| 15 | #include "riscv.h" |
| 16 | #include "riscv/debug.h" |
| 17 | #include "types.h" |
| 18 | |
| 19 | #ifndef PATH_MAX |
| 20 | #define PATH_MAX 4096 |
| 21 | #endif |
| 22 | |
| 23 | /* Define BINARY for `bail` and `assert` functions. */ |
| 24 | #undef BINARY |
| 25 | #define BINARY "emulator" |
| 26 | |
| 27 | #undef assert |
| 28 | #define assert(condition) \ |
| 29 | ((condition) ? (void)0 : _assert_failed(#condition, __FILE__, __LINE__)) |
| 30 | |
| 31 | static inline __attribute__((noreturn)) void _assert_failed( |
| 32 | const char *condition, const char *file, int line |
| 33 | ) { |
| 34 | fprintf(stderr, "%s:%d: assertion `%s` failed\n", file, line, condition); |
| 35 | abort(); |
| 36 | } |
| 37 | |
| 38 | /* Maximum physical memory size (384MB). The runtime can choose any size up to |
| 39 | * this value with `-memory-size=...`. */ |
| 40 | #define MEMORY_SIZE (384 * 1024 * 1024) |
| 41 | /* Default physical memory size (128MB). */ |
| 42 | #define DEFAULT_MEMORY_SIZE (128 * 1024 * 1024) |
| 43 | /* Program memory size (4MB), reserved at start of memory for program code. */ |
| 44 | #define PROGRAM_SIZE (4 * 1024 * 1024) |
| 45 | /* Writable data region base. */ |
| 46 | #define DATA_RW_OFFSET 0xFFFFF0 |
| 47 | /* Data memory starts at writable data region. */ |
| 48 | #define DATA_MEMORY_START DATA_RW_OFFSET |
| 49 | /* Default data memory size. */ |
| 50 | #define DEFAULT_DATA_MEMORY_SIZE (DEFAULT_MEMORY_SIZE - DATA_MEMORY_START) |
| 51 | /* Default stack size (256KB), allocated at the end of memory. */ |
| 52 | #define DEFAULT_STACK_SIZE (256 * 1024) |
| 53 | /* Maximum instructions to show in the TUI. */ |
| 54 | #define MAX_INSTR_DISPLAY 40 |
| 55 | /* Stack words to display in the TUI. */ |
| 56 | #define STACK_DISPLAY_WORDS 32 |
| 57 | /* Maximum number of CPU state snapshots to store for undo. */ |
| 58 | #define MAX_SNAPSHOTS 64 |
| 59 | /* Maximum open files in guest runtime. */ |
| 60 | #define MAX_OPEN_FILES 32 |
| 61 | /* Number of history entries to print when reporting faults. */ |
| 62 | #define FAULT_TRACE_DEPTH 8 |
| 63 | /* Instruction trace depth for headless tracing. */ |
| 64 | #define TRACE_HISTORY 64 |
| 65 | /* Maximum number of steps executed in headless mode before timing out. */ |
| 66 | #define HEADLESS_MAX_STEPS ((u64)1000000000000) |
| 67 | /* Height of header and footer in rows. */ |
| 68 | #define HEADER_HEIGHT 2 |
| 69 | #define FOOTER_HEIGHT 3 |
| 70 | /* Read-only data offset. */ |
| 71 | #define DATA_RO_OFFSET 0x10000 |
| 72 | /* TTY escape codes. */ |
| 73 | #define TTY_CLEAR "\033[2J\033[H" |
| 74 | #define TTY_GOTO_RC "\033[%d;%dH" |
| 75 | /* Exit code returned on EBREAK. */ |
| 76 | #define EBREAK_EXIT_CODE 133 |
| 77 | |
| 78 | /* Registers displayed in the TUI, in order. */ |
| 79 | static const reg_t registers_displayed[] = { |
| 80 | SP, FP, RA, A0, A1, A2, A3, A4, A5, A6, A7, T0, T1, T2, T3, T4, T5, T6 |
| 81 | }; |
| 82 | |
| 83 | /* Display mode for immediates and values. */ |
| 84 | enum display { DISPLAY_HEX, DISPLAY_DEC }; |
| 85 | |
| 86 | /* Debug info entry mapping PC to source location. */ |
| 87 | struct debug_entry { |
| 88 | u32 pc; |
| 89 | u32 offset; |
| 90 | char file[PATH_MAX]; |
| 91 | }; |
| 92 | |
| 93 | /* Debug info table. */ |
| 94 | struct debug_info { |
| 95 | struct debug_entry *entries; |
| 96 | size_t count; |
| 97 | size_t capacity; |
| 98 | }; |
| 99 | |
| 100 | /* Global debug info. */ |
| 101 | static struct debug_info g_debug = { 0 }; |
| 102 | |
| 103 | /* CPU state. */ |
| 104 | struct cpu { |
| 105 | u64 regs[REGISTERS]; |
| 106 | u32 pc; /* Program counter. */ |
| 107 | u32 programsize; /* Size of loaded program. */ |
| 108 | instr_t *program; /* Program instructions. */ |
| 109 | bool running; /* Execution status. */ |
| 110 | bool faulted; /* There was a fault in execution. */ |
| 111 | bool ebreak; /* Program terminated via EBREAK. */ |
| 112 | reg_t modified; /* Index of the last modified register. */ |
| 113 | }; |
| 114 | |
| 115 | /* Snapshot of CPU and memory state for reversing execution. */ |
| 116 | struct snapshot { |
| 117 | struct cpu cpu; /* Copy of CPU state. */ |
| 118 | u8 memory[MEMORY_SIZE]; /* Copy of memory. */ |
| 119 | }; |
| 120 | |
| 121 | /* Circular buffer for snapshots. */ |
| 122 | struct snapshot_buffer { |
| 123 | struct snapshot snapshots[MAX_SNAPSHOTS]; |
| 124 | int head; /* Index of most recent snapshot. */ |
| 125 | int count; |
| 126 | }; |
| 127 | |
| 128 | /* CPU memory. */ |
| 129 | static u8 memory[MEMORY_SIZE]; |
| 130 | |
| 131 | /* Loaded section sizes, used for bounds checking and diagnostics. */ |
| 132 | static u32 program_base = 0; |
| 133 | static u32 program_bytes = 0; |
| 134 | static u32 rodata_bytes = 0; |
| 135 | static u32 data_bytes = 0; |
| 136 | |
| 137 | /* Snapshot buffer. */ |
| 138 | static struct snapshot_buffer snapshots; |
| 139 | |
| 140 | /* File descriptor table for guest file operations. */ |
| 141 | static int guest_fds[MAX_OPEN_FILES]; |
| 142 | |
| 143 | /* Initialize the guest file descriptor table. */ |
| 144 | static void guest_fd_table_init(void) { |
| 145 | for (int i = 0; i < MAX_OPEN_FILES; i++) { |
| 146 | guest_fds[i] = -1; |
| 147 | } |
| 148 | } |
| 149 | |
| 150 | /* Add a host file descriptor to the guest table. */ |
| 151 | static int guest_fd_table_add(int host_fd) { |
| 152 | /* Start at 3 to skip stdin/stdout/stderr. */ |
| 153 | for (int i = 3; i < MAX_OPEN_FILES; i++) { |
| 154 | if (guest_fds[i] == -1) { |
| 155 | guest_fds[i] = host_fd; |
| 156 | return i; |
| 157 | } |
| 158 | } |
| 159 | return -1; |
| 160 | } |
| 161 | |
| 162 | /* Get the host fd for a guest file descriptor. */ |
| 163 | static int guest_fd_table_get(int guest_fd) { |
| 164 | if (guest_fd < 0 || guest_fd >= MAX_OPEN_FILES) { |
| 165 | return -1; |
| 166 | } |
| 167 | /* Standard streams map directly. */ |
| 168 | if (guest_fd < 3) { |
| 169 | return guest_fd; |
| 170 | } |
| 171 | return guest_fds[guest_fd]; |
| 172 | } |
| 173 | |
| 174 | /* Remove a file descriptor from the guest table. */ |
| 175 | static void guest_fd_table_remove(int guest_fd) { |
| 176 | if (guest_fd >= 3 && guest_fd < MAX_OPEN_FILES) { |
| 177 | guest_fds[guest_fd] = -1; |
| 178 | } |
| 179 | } |
| 180 | |
| 181 | /* Single entry in the instruction trace ring buffer. */ |
| 182 | struct trace_entry { |
| 183 | u32 pc; |
| 184 | instr_t instr; |
| 185 | u64 regs[REGISTERS]; |
| 186 | }; |
| 187 | |
| 188 | /* Circular buffer of recent instruction traces. */ |
| 189 | struct trace_ring { |
| 190 | struct trace_entry entries[TRACE_HISTORY]; |
| 191 | int head; |
| 192 | int count; |
| 193 | }; |
| 194 | |
| 195 | /* Headless-mode instruction trace buffer. */ |
| 196 | static struct trace_ring headless_trace = { .head = -1, .count = 0 }; |
| 197 | |
| 198 | /* Terminal dimensions in rows and columns. */ |
| 199 | struct termsize { |
| 200 | int rows; |
| 201 | int cols; |
| 202 | }; |
| 203 | |
| 204 | /* Forward declarations. */ |
| 205 | static void ui_render_instructions( |
| 206 | struct cpu *, int col, int width, int height |
| 207 | ); |
| 208 | static void ui_render_registers( |
| 209 | struct cpu *, enum display, int col, int height |
| 210 | ); |
| 211 | static void ui_render_stack(struct cpu *, enum display, int col, int height); |
| 212 | static void ui_render(struct cpu *, enum display); |
| 213 | static void cpu_execute(struct cpu *, enum display, bool headless); |
| 214 | static void emit_fault_diagnostics(struct cpu *, u32 pc); |
| 215 | |
| 216 | /* Emulator runtime options, populated from CLI flags. */ |
| 217 | struct emulator_options { |
| 218 | bool stack_guard; |
| 219 | u32 stack_size; |
| 220 | bool debug_enabled; |
| 221 | bool trace_headless; |
| 222 | bool trace_enabled; |
| 223 | bool trace_print_instructions; |
| 224 | u32 trace_depth; |
| 225 | u64 headless_max_steps; |
| 226 | u32 memory_size; |
| 227 | u32 data_memory_size; |
| 228 | bool watch_enabled; |
| 229 | u32 watch_addr; |
| 230 | u32 watch_size; |
| 231 | u32 watch_arm_pc; |
| 232 | bool watch_zero_only; |
| 233 | u32 watch_skip; |
| 234 | bool watch_backtrace; |
| 235 | u32 watch_backtrace_depth; |
| 236 | bool validate_memory; |
| 237 | bool count_instructions; |
| 238 | bool jit_disabled; |
| 239 | }; |
| 240 | |
| 241 | /* Global emulator options. */ |
| 242 | static struct emulator_options g_opts = { |
| 243 | .stack_guard = true, |
| 244 | .stack_size = DEFAULT_STACK_SIZE, |
| 245 | .debug_enabled = false, |
| 246 | .trace_headless = false, |
| 247 | .trace_enabled = false, |
| 248 | .trace_print_instructions = false, |
| 249 | .trace_depth = 32, |
| 250 | .headless_max_steps = HEADLESS_MAX_STEPS, |
| 251 | .memory_size = DEFAULT_MEMORY_SIZE, |
| 252 | .data_memory_size = DEFAULT_DATA_MEMORY_SIZE, |
| 253 | .watch_enabled = false, |
| 254 | .watch_addr = 0, |
| 255 | .watch_size = 0, |
| 256 | .watch_arm_pc = 0, |
| 257 | .watch_zero_only = false, |
| 258 | .watch_skip = 0, |
| 259 | .watch_backtrace = false, |
| 260 | .watch_backtrace_depth = 8, |
| 261 | .validate_memory = true, |
| 262 | .count_instructions = false, |
| 263 | .jit_disabled = false, |
| 264 | }; |
| 265 | |
| 266 | static void dump_watch_context(struct cpu *, u32 addr, u32 size, u32 value); |
| 267 | |
| 268 | /* Return true if the given address range overlaps the watched region. */ |
| 269 | static inline bool watch_hit(u32 addr, u32 size) { |
| 270 | if (!g_opts.watch_enabled) |
| 271 | return false; |
| 272 | u32 start = g_opts.watch_addr; |
| 273 | u32 end = start + (g_opts.watch_size ? g_opts.watch_size : 1); |
| 274 | return addr < end && (addr + size) > start; |
| 275 | } |
| 276 | |
| 277 | /* Check a store against the memory watchpoint and halt on a hit. */ |
| 278 | static inline void watch_store(struct cpu *cpu, u32 addr, u32 size, u32 value) { |
| 279 | if (!watch_hit(addr, size)) |
| 280 | return; |
| 281 | if (g_opts.watch_arm_pc && cpu && cpu->pc < g_opts.watch_arm_pc) |
| 282 | return; |
| 283 | if (g_opts.watch_zero_only && value != 0) |
| 284 | return; |
| 285 | if (g_opts.watch_skip > 0) { |
| 286 | g_opts.watch_skip--; |
| 287 | return; |
| 288 | } |
| 289 | fprintf( |
| 290 | stderr, |
| 291 | "[WATCH] pc=%08x addr=%08x size=%u value=%08x\n", |
| 292 | cpu ? cpu->pc : 0, |
| 293 | addr, |
| 294 | size, |
| 295 | value |
| 296 | ); |
| 297 | dump_watch_context(cpu, addr, size, value); |
| 298 | if (cpu) { |
| 299 | cpu->running = false; |
| 300 | cpu->faulted = true; |
| 301 | cpu->ebreak = true; |
| 302 | } |
| 303 | } |
| 304 | |
| 305 | /* Fixed stack guard zone size. */ |
| 306 | #define STACK_GUARD_BYTES 16 |
| 307 | |
| 308 | /* Clamp and align stack size to a valid range. */ |
| 309 | static inline u32 sanitize_stack_bytes(u32 bytes) { |
| 310 | if (bytes < WORD_SIZE) |
| 311 | bytes = WORD_SIZE; |
| 312 | bytes = (u32)align((i32)bytes, WORD_SIZE); |
| 313 | |
| 314 | /* Keep at least one word for guard computations. */ |
| 315 | if (bytes >= g_opts.memory_size) |
| 316 | bytes = g_opts.memory_size - WORD_SIZE; |
| 317 | |
| 318 | return bytes; |
| 319 | } |
| 320 | |
| 321 | /* Return the active stack guard size, or 0 if guards are disabled. */ |
| 322 | static inline u32 stack_guard_bytes(void) { |
| 323 | if (!g_opts.stack_guard) |
| 324 | return 0; |
| 325 | return STACK_GUARD_BYTES; |
| 326 | } |
| 327 | |
| 328 | /* Return the configured stack size. */ |
| 329 | static inline u32 stack_size(void) { |
| 330 | return g_opts.stack_size; |
| 331 | } |
| 332 | |
| 333 | /* Return the highest addressable word-aligned memory address. */ |
| 334 | static inline u32 memory_top(void) { |
| 335 | return g_opts.memory_size - WORD_SIZE; |
| 336 | } |
| 337 | |
| 338 | /* Return the lowest address in the stack region. */ |
| 339 | static inline u32 stack_bottom(void) { |
| 340 | return memory_top() - stack_size() + WORD_SIZE; |
| 341 | } |
| 342 | |
| 343 | /* Return the highest usable stack address (inside the guard zone). */ |
| 344 | static inline u32 stack_usable_top(void) { |
| 345 | u32 guard = stack_guard_bytes(); |
| 346 | u32 size = stack_size(); |
| 347 | if (guard >= size) |
| 348 | guard = size - WORD_SIZE; |
| 349 | return memory_top() - guard; |
| 350 | } |
| 351 | |
| 352 | /* Return the lowest usable stack address (inside the guard zone). */ |
| 353 | static inline u32 stack_usable_bottom(void) { |
| 354 | u32 guard = stack_guard_bytes(); |
| 355 | u32 size = stack_size(); |
| 356 | if (guard >= size) |
| 357 | guard = size - WORD_SIZE; |
| 358 | return stack_bottom() + guard; |
| 359 | } |
| 360 | |
| 361 | /* Return true if addr falls within the stack region. */ |
| 362 | static inline bool stack_contains(u32 addr) { |
| 363 | return addr >= stack_bottom() && addr <= memory_top(); |
| 364 | } |
| 365 | |
| 366 | /* Return true if the range [start, end] overlaps a stack guard zone. */ |
| 367 | static inline bool stack_guard_overlaps(u32 guard, u32 start, u32 end) { |
| 368 | if (guard == 0) |
| 369 | return false; |
| 370 | |
| 371 | u32 low_guard_end = stack_bottom() + guard - 1; |
| 372 | u32 high_guard_start = memory_top() - guard + 1; |
| 373 | |
| 374 | return (start <= low_guard_end && end >= stack_bottom()) || |
| 375 | (end >= high_guard_start && start <= memory_top()); |
| 376 | } |
| 377 | |
| 378 | /* Return true if addr falls inside a stack guard zone. */ |
| 379 | static inline bool stack_guard_contains(u32 guard, u32 addr) { |
| 380 | return stack_guard_overlaps(guard, addr, addr); |
| 381 | } |
| 382 | |
| 383 | /* Load a 16-bit value from memory in little-endian byte order. */ |
| 384 | static inline u16 memory_load_u16(u32 addr) { |
| 385 | return (u16)(memory[addr] | (memory[addr + 1] << 8)); |
| 386 | } |
| 387 | |
| 388 | /* Load a 32-bit value from memory in little-endian byte order. */ |
| 389 | static inline u32 memory_load_u32(u32 addr) { |
| 390 | return memory[addr] | (memory[addr + 1] << 8) | (memory[addr + 2] << 16) | |
| 391 | (memory[addr + 3] << 24); |
| 392 | } |
| 393 | |
| 394 | /* Load a 64-bit value from memory in little-endian byte order. */ |
| 395 | static inline u64 memory_load_u64(u32 addr) { |
| 396 | return (u64)memory[addr] | ((u64)memory[addr + 1] << 8) | |
| 397 | ((u64)memory[addr + 2] << 16) | ((u64)memory[addr + 3] << 24) | |
| 398 | ((u64)memory[addr + 4] << 32) | ((u64)memory[addr + 5] << 40) | |
| 399 | ((u64)memory[addr + 6] << 48) | ((u64)memory[addr + 7] << 56); |
| 400 | } |
| 401 | |
| 402 | /* Store a byte to memory. */ |
| 403 | static inline void memory_store_u8(u32 addr, u8 value) { |
| 404 | memory[addr] = value; |
| 405 | } |
| 406 | |
| 407 | /* Store a 16-bit value to memory in little-endian byte order. */ |
| 408 | static inline void memory_store_u16(u32 addr, u16 value) { |
| 409 | memory[addr] = (u8)(value & 0xFF); |
| 410 | memory[addr + 1] = (u8)((value >> 8) & 0xFF); |
| 411 | } |
| 412 | |
| 413 | /* Store a 32-bit value to memory in little-endian byte order. */ |
| 414 | static inline void memory_store_u32(u32 addr, u32 value) { |
| 415 | assert(addr + 3 < g_opts.memory_size); |
| 416 | memory[addr] = (u8)(value & 0xFF); |
| 417 | memory[addr + 1] = (u8)((value >> 8) & 0xFF); |
| 418 | memory[addr + 2] = (u8)((value >> 16) & 0xFF); |
| 419 | memory[addr + 3] = (u8)((value >> 24) & 0xFF); |
| 420 | } |
| 421 | |
| 422 | /* Store a 64-bit value to memory in little-endian byte order. */ |
| 423 | static inline void memory_store_u64(u32 addr, u64 value) { |
| 424 | assert(addr + 7 < g_opts.memory_size); |
| 425 | memory_store_u32(addr, (u32)(value & 0xFFFFFFFF)); |
| 426 | memory_store_u32(addr + 4, (u32)(value >> 32)); |
| 427 | } |
| 428 | |
| 429 | /* Load a 32-bit word from memory, returning false if out of bounds. */ |
| 430 | static inline bool load_word_safe(u32 addr, u32 *out) { |
| 431 | if (addr > g_opts.memory_size - WORD_SIZE) |
| 432 | return false; |
| 433 | *out = memory_load_u32(addr); |
| 434 | return true; |
| 435 | } |
| 436 | |
| 437 | /* Dump register state and backtrace when a watchpoint fires. */ |
| 438 | static void dump_watch_context(struct cpu *cpu, u32 addr, u32 size, u32 value) { |
| 439 | if (!g_opts.watch_backtrace || !cpu) |
| 440 | return; |
| 441 | |
| 442 | (void)addr; |
| 443 | (void)size; |
| 444 | (void)value; |
| 445 | |
| 446 | fprintf( |
| 447 | stderr, |
| 448 | " regs: SP=%08x FP=%08x RA=%08x A0=%08x A1=%08x A2=%08x " |
| 449 | "A3=%08x\n", |
| 450 | (u32)cpu->regs[SP], |
| 451 | (u32)cpu->regs[FP], |
| 452 | (u32)cpu->regs[RA], |
| 453 | (u32)cpu->regs[A0], |
| 454 | (u32)cpu->regs[A1], |
| 455 | (u32)cpu->regs[A2], |
| 456 | (u32)cpu->regs[A3] |
| 457 | ); |
| 458 | |
| 459 | u64 fp64 = cpu->regs[FP]; |
| 460 | u32 fp = (fp64 <= (u64)UINT32_MAX) ? (u32)fp64 : 0; |
| 461 | u32 pc = cpu->pc; |
| 462 | |
| 463 | fprintf( |
| 464 | stderr, " backtrace (depth %u):\n", g_opts.watch_backtrace_depth |
| 465 | ); |
| 466 | |
| 467 | for (u32 depth = 0; depth < g_opts.watch_backtrace_depth; depth++) { |
| 468 | bool has_frame = stack_contains(fp) && fp >= (2 * WORD_SIZE); |
| 469 | u32 saved_ra = 0; |
| 470 | u32 prev_fp = 0; |
| 471 | |
| 472 | if (has_frame) { |
| 473 | has_frame = load_word_safe(fp - WORD_SIZE, &saved_ra) && |
| 474 | load_word_safe(fp - 2 * WORD_SIZE, &prev_fp) && |
| 475 | stack_contains(prev_fp); |
| 476 | } |
| 477 | |
| 478 | fprintf( |
| 479 | stderr, |
| 480 | " #%u pc=%08x fp=%08x ra=%08x%s\n", |
| 481 | depth, |
| 482 | pc, |
| 483 | fp, |
| 484 | has_frame ? saved_ra : 0, |
| 485 | has_frame ? "" : " (?)" |
| 486 | ); |
| 487 | |
| 488 | if (!has_frame || prev_fp == fp || prev_fp == 0) |
| 489 | break; |
| 490 | |
| 491 | pc = saved_ra; |
| 492 | fp = prev_fp; |
| 493 | } |
| 494 | } |
| 495 | |
| 496 | /* Print usage information and return 1. */ |
| 497 | static int usage(const char *prog) { |
| 498 | fprintf( |
| 499 | stderr, |
| 500 | "usage: %s [-run] [-no-guard-stack]" |
| 501 | " [-stack-size=KB] [-no-validate] [-debug]" |
| 502 | " [-trace|-trace-headless] [-trace-depth=n] [-trace-instructions]" |
| 503 | " [-max-steps=n] [-memory-size=KB] [-data-size=KB]" |
| 504 | " [-watch=addr] [-watch-size=bytes] [-watch-arm-pc=addr]" |
| 505 | " [-watch-zero-only] [-watch-skip=n]" |
| 506 | " [-watch-backtrace] [-watch-bt-depth=n]" |
| 507 | " [-count-instructions] [-no-jit]" |
| 508 | " <file.bin> [program args...]\n", |
| 509 | prog |
| 510 | ); |
| 511 | return 1; |
| 512 | } |
| 513 | |
| 514 | /* Configuration parsed from CLI flags prior to launching the emulator. */ |
| 515 | struct cli_config { |
| 516 | bool headless; |
| 517 | const char *program_path; |
| 518 | int arg_index; |
| 519 | }; |
| 520 | |
| 521 | /* Parse a string as an unsigned 32-bit integer. Returns false on error. */ |
| 522 | static bool parse_u32(const char *str, const char *label, int base, u32 *out) { |
| 523 | char *end = NULL; |
| 524 | errno = 0; |
| 525 | unsigned long val = strtoul(str, &end, base); |
| 526 | if (errno != 0 || end == str || *end != '\0') { |
| 527 | fprintf(stderr, "invalid %s '%s'; expected integer\n", label, str); |
| 528 | return false; |
| 529 | } |
| 530 | if (val > UINT32_MAX) |
| 531 | val = UINT32_MAX; |
| 532 | *out = (u32)val; |
| 533 | return true; |
| 534 | } |
| 535 | |
| 536 | /* Parse a string as an unsigned 64-bit integer. Returns false on error. */ |
| 537 | static bool parse_u64(const char *str, const char *label, u64 *out) { |
| 538 | char *end = NULL; |
| 539 | errno = 0; |
| 540 | unsigned long long val = strtoull(str, &end, 10); |
| 541 | if (errno != 0 || end == str || *end != '\0') { |
| 542 | fprintf(stderr, "invalid %s '%s'; expected integer\n", label, str); |
| 543 | return false; |
| 544 | } |
| 545 | *out = (u64)val; |
| 546 | return true; |
| 547 | } |
| 548 | |
| 549 | /* Parse and validate the physical memory size passed to -memory-size=. */ |
| 550 | static bool parse_memory_size_value(const char *value) { |
| 551 | u64 parsed; |
| 552 | if (!parse_u64(value, "memory size", &parsed)) |
| 553 | return false; |
| 554 | u64 bytes = parsed * 1024; |
| 555 | if (bytes <= (u64)(DATA_MEMORY_START + WORD_SIZE)) { |
| 556 | fprintf( |
| 557 | stderr, |
| 558 | "memory size too small; minimum is %u KB\n", |
| 559 | (DATA_MEMORY_START + WORD_SIZE + 1024) / 1024 |
| 560 | ); |
| 561 | return false; |
| 562 | } |
| 563 | if (bytes > (u64)MEMORY_SIZE) { |
| 564 | fprintf( |
| 565 | stderr, |
| 566 | "memory size too large; maximum is %u KB (recompile emulator " |
| 567 | "to increase)\n", |
| 568 | MEMORY_SIZE / 1024 |
| 569 | ); |
| 570 | return false; |
| 571 | } |
| 572 | g_opts.memory_size = (u32)bytes; |
| 573 | return true; |
| 574 | } |
| 575 | |
| 576 | /* Parse and validate the depth passed to -trace-depth=. */ |
| 577 | static bool parse_trace_depth_value(const char *value) { |
| 578 | u32 parsed; |
| 579 | if (!parse_u32(value, "trace depth", 10, &parsed)) |
| 580 | return false; |
| 581 | if (parsed == 0) { |
| 582 | fprintf(stderr, "trace depth must be greater than zero\n"); |
| 583 | return false; |
| 584 | } |
| 585 | if (parsed > TRACE_HISTORY) |
| 586 | parsed = TRACE_HISTORY; |
| 587 | g_opts.trace_depth = parsed; |
| 588 | return true; |
| 589 | } |
| 590 | |
| 591 | /* Parse and validate the step limit passed to -max-steps=. */ |
| 592 | static bool parse_max_steps_value(const char *value) { |
| 593 | u64 parsed; |
| 594 | if (!parse_u64(value, "max steps", &parsed)) |
| 595 | return false; |
| 596 | if (parsed == 0) { |
| 597 | fprintf(stderr, "max steps must be greater than zero\n"); |
| 598 | return false; |
| 599 | } |
| 600 | g_opts.headless_max_steps = parsed; |
| 601 | return true; |
| 602 | } |
| 603 | |
| 604 | /* Parse and validate the stack size passed to -stack-size=. */ |
| 605 | static bool parse_stack_size_value(const char *value) { |
| 606 | u64 parsed; |
| 607 | if (!parse_u64(value, "stack size", &parsed)) |
| 608 | return false; |
| 609 | if (parsed == 0) { |
| 610 | fprintf(stderr, "stack size must be greater than zero\n"); |
| 611 | return false; |
| 612 | } |
| 613 | u64 bytes = parsed * 1024; |
| 614 | if (bytes >= MEMORY_SIZE) { |
| 615 | fprintf( |
| 616 | stderr, |
| 617 | "stack size too large; maximum is %u KB\n", |
| 618 | MEMORY_SIZE / 1024 |
| 619 | ); |
| 620 | return false; |
| 621 | } |
| 622 | g_opts.stack_size = sanitize_stack_bytes((u32)bytes); |
| 623 | return true; |
| 624 | } |
| 625 | |
| 626 | /* Parse and validate the data size passed to -data-size=. */ |
| 627 | static bool parse_data_size_value(const char *value) { |
| 628 | u64 parsed; |
| 629 | if (!parse_u64(value, "data size", &parsed)) |
| 630 | return false; |
| 631 | if (parsed == 0) { |
| 632 | fprintf(stderr, "data size must be greater than zero\n"); |
| 633 | return false; |
| 634 | } |
| 635 | u64 bytes = parsed * 1024; |
| 636 | if (bytes > (u64)MEMORY_SIZE) { |
| 637 | fprintf( |
| 638 | stderr, |
| 639 | "data size too large; maximum is %u KB\n", |
| 640 | MEMORY_SIZE / 1024 |
| 641 | ); |
| 642 | return false; |
| 643 | } |
| 644 | g_opts.data_memory_size = (u32)bytes; |
| 645 | return true; |
| 646 | } |
| 647 | |
| 648 | /* Validate that the stack fits within available memory. */ |
| 649 | static bool validate_memory_layout(void) { |
| 650 | if (g_opts.stack_size >= g_opts.memory_size) { |
| 651 | fprintf( |
| 652 | stderr, |
| 653 | "stack size (%u) must be smaller than memory size (%u)\n", |
| 654 | g_opts.stack_size, |
| 655 | g_opts.memory_size |
| 656 | ); |
| 657 | return false; |
| 658 | } |
| 659 | return true; |
| 660 | } |
| 661 | |
| 662 | /* Parse emulator CLI arguments, returning the selected mode and file path. */ |
| 663 | static bool parse_cli_args(int argc, char *argv[], struct cli_config *cfg) { |
| 664 | bool headless = false; |
| 665 | int argi = 1; |
| 666 | |
| 667 | while (argi < argc) { |
| 668 | const char *arg = argv[argi]; |
| 669 | |
| 670 | if (strcmp(arg, "--") == 0) { |
| 671 | argi++; |
| 672 | break; |
| 673 | } |
| 674 | if (arg[0] != '-') |
| 675 | break; |
| 676 | |
| 677 | if (strcmp(arg, "-run") == 0) { |
| 678 | headless = true; |
| 679 | argi++; |
| 680 | continue; |
| 681 | } |
| 682 | |
| 683 | if (strncmp(arg, "-stack-size=", 12) == 0) { |
| 684 | if (!parse_stack_size_value(arg + 12)) |
| 685 | return false; |
| 686 | argi++; |
| 687 | continue; |
| 688 | } |
| 689 | if (strcmp(arg, "-no-guard-stack") == 0) { |
| 690 | g_opts.stack_guard = false; |
| 691 | argi++; |
| 692 | continue; |
| 693 | } |
| 694 | if (strcmp(arg, "-no-validate") == 0) { |
| 695 | g_opts.validate_memory = false; |
| 696 | argi++; |
| 697 | continue; |
| 698 | } |
| 699 | if (strcmp(arg, "-debug") == 0) { |
| 700 | g_opts.debug_enabled = true; |
| 701 | argi++; |
| 702 | continue; |
| 703 | } |
| 704 | if (strcmp(arg, "-trace") == 0 || strcmp(arg, "-trace-headless") == 0) { |
| 705 | g_opts.trace_enabled = true; |
| 706 | argi++; |
| 707 | continue; |
| 708 | } |
| 709 | if (strcmp(arg, "-trace-instructions") == 0) { |
| 710 | g_opts.trace_print_instructions = true; |
| 711 | argi++; |
| 712 | continue; |
| 713 | } |
| 714 | if (strncmp(arg, "-trace-depth=", 13) == 0) { |
| 715 | if (!parse_trace_depth_value(arg + 13)) |
| 716 | return false; |
| 717 | argi++; |
| 718 | continue; |
| 719 | } |
| 720 | if (strncmp(arg, "-max-steps=", 11) == 0) { |
| 721 | if (!parse_max_steps_value(arg + 11)) |
| 722 | return false; |
| 723 | argi++; |
| 724 | continue; |
| 725 | } |
| 726 | if (strncmp(arg, "-memory-size=", 13) == 0) { |
| 727 | if (!parse_memory_size_value(arg + 13)) |
| 728 | return false; |
| 729 | argi++; |
| 730 | continue; |
| 731 | } |
| 732 | if (strncmp(arg, "-data-size=", 11) == 0) { |
| 733 | if (!parse_data_size_value(arg + 11)) |
| 734 | return false; |
| 735 | argi++; |
| 736 | continue; |
| 737 | } |
| 738 | if (strncmp(arg, "-watch=", 7) == 0) { |
| 739 | if (!parse_u32(arg + 7, "watch address", 0, &g_opts.watch_addr)) |
| 740 | return false; |
| 741 | g_opts.watch_enabled = true; |
| 742 | argi++; |
| 743 | continue; |
| 744 | } |
| 745 | if (strncmp(arg, "-watch-size=", 12) == 0) { |
| 746 | if (!parse_u32(arg + 12, "watch size", 0, &g_opts.watch_size)) |
| 747 | return false; |
| 748 | if (g_opts.watch_size == 0) { |
| 749 | fprintf(stderr, "watch size must be greater than zero\n"); |
| 750 | return false; |
| 751 | } |
| 752 | argi++; |
| 753 | continue; |
| 754 | } |
| 755 | if (strcmp(arg, "-watch-zero-only") == 0) { |
| 756 | g_opts.watch_zero_only = true; |
| 757 | argi++; |
| 758 | continue; |
| 759 | } |
| 760 | if (strncmp(arg, "-watch-skip=", 12) == 0) { |
| 761 | if (!parse_u32(arg + 12, "watch skip", 0, &g_opts.watch_skip)) |
| 762 | return false; |
| 763 | argi++; |
| 764 | continue; |
| 765 | } |
| 766 | if (strcmp(arg, "-watch-disable") == 0) { |
| 767 | g_opts.watch_enabled = false; |
| 768 | argi++; |
| 769 | continue; |
| 770 | } |
| 771 | if (strncmp(arg, "-watch-arm-pc=", 14) == 0) { |
| 772 | if (!parse_u32(arg + 14, "watch arm pc", 0, &g_opts.watch_arm_pc)) |
| 773 | return false; |
| 774 | argi++; |
| 775 | continue; |
| 776 | } |
| 777 | if (strcmp(arg, "-watch-backtrace") == 0) { |
| 778 | g_opts.watch_backtrace = true; |
| 779 | argi++; |
| 780 | continue; |
| 781 | } |
| 782 | if (strncmp(arg, "-watch-bt-depth=", 16) == 0) { |
| 783 | u32 depth; |
| 784 | if (!parse_u32(arg + 16, "watch backtrace depth", 0, &depth)) |
| 785 | return false; |
| 786 | if (depth == 0) { |
| 787 | fprintf( |
| 788 | stderr, "watch backtrace depth must be greater than zero\n" |
| 789 | ); |
| 790 | return false; |
| 791 | } |
| 792 | g_opts.watch_backtrace = true; |
| 793 | g_opts.watch_backtrace_depth = depth; |
| 794 | argi++; |
| 795 | continue; |
| 796 | } |
| 797 | if (strcmp(arg, "-count-instructions") == 0) { |
| 798 | g_opts.count_instructions = true; |
| 799 | argi++; |
| 800 | continue; |
| 801 | } |
| 802 | if (strcmp(arg, "-no-jit") == 0) { |
| 803 | g_opts.jit_disabled = true; |
| 804 | argi++; |
| 805 | continue; |
| 806 | } |
| 807 | usage(argv[0]); |
| 808 | |
| 809 | return false; |
| 810 | } |
| 811 | if (argi >= argc) { |
| 812 | usage(argv[0]); |
| 813 | return false; |
| 814 | } |
| 815 | cfg->program_path = argv[argi++]; |
| 816 | cfg->arg_index = argi; |
| 817 | cfg->headless = headless; |
| 818 | if (g_opts.watch_enabled && g_opts.watch_size == 0) |
| 819 | g_opts.watch_size = 4; |
| 820 | if (!validate_memory_layout()) |
| 821 | return false; |
| 822 | |
| 823 | g_opts.stack_size = sanitize_stack_bytes(g_opts.stack_size); |
| 824 | |
| 825 | return true; |
| 826 | } |
| 827 | |
| 828 | /* Validate a load or store against memory bounds and stack guards. */ |
| 829 | static bool validate_memory_access( |
| 830 | struct cpu *cpu, |
| 831 | u64 addr, |
| 832 | u32 size, |
| 833 | reg_t base_reg, |
| 834 | const char *op, |
| 835 | bool is_store |
| 836 | ) { |
| 837 | /* Skip validation for performance if disabled. */ |
| 838 | if (!g_opts.validate_memory) |
| 839 | return true; |
| 840 | |
| 841 | if (size == 0) |
| 842 | size = 1; |
| 843 | |
| 844 | const char *kind = is_store ? "store" : "load"; |
| 845 | |
| 846 | u64 span_end = addr + (u64)size; |
| 847 | if (addr > (u64)g_opts.memory_size || span_end > (u64)g_opts.memory_size) { |
| 848 | printf( |
| 849 | "Memory %s out of bounds at PC=%08x: addr=%016llx size=%u (%s)\n", |
| 850 | kind, |
| 851 | cpu->pc, |
| 852 | (unsigned long long)addr, |
| 853 | size, |
| 854 | op |
| 855 | ); |
| 856 | cpu->running = false; |
| 857 | emit_fault_diagnostics(cpu, cpu->pc); |
| 858 | return false; |
| 859 | } |
| 860 | |
| 861 | u32 addr32 = (u32)addr; |
| 862 | u32 end = (u32)(span_end - 1); |
| 863 | |
| 864 | if (addr32 < DATA_MEMORY_START) { |
| 865 | if (is_store) { |
| 866 | printf( |
| 867 | "Read-only memory store at PC=%08x: addr=%08x size=%u (%s)\n", |
| 868 | cpu->pc, |
| 869 | addr32, |
| 870 | size, |
| 871 | op |
| 872 | ); |
| 873 | cpu->running = false; |
| 874 | emit_fault_diagnostics(cpu, cpu->pc); |
| 875 | return false; |
| 876 | } |
| 877 | return true; |
| 878 | } |
| 879 | |
| 880 | u32 guard = stack_guard_bytes(); |
| 881 | u64 base_val = cpu->regs[base_reg]; |
| 882 | bool base_in_stack = |
| 883 | base_val <= (u64)UINT32_MAX && stack_contains((u32)base_val); |
| 884 | bool start_in_stack = stack_contains(addr32); |
| 885 | bool end_in_stack = stack_contains(end); |
| 886 | |
| 887 | if (base_in_stack || start_in_stack || end_in_stack) { |
| 888 | u32 bottom = stack_bottom(); |
| 889 | if (addr32 < bottom || end > memory_top()) { |
| 890 | printf( |
| 891 | "Stack %s out of bounds at PC=%08x: base=%s (0x%08x) addr=%08x " |
| 892 | "size=%u (%s)\n", |
| 893 | kind, |
| 894 | cpu->pc, |
| 895 | reg_names[base_reg], |
| 896 | (u32)cpu->regs[base_reg], |
| 897 | addr32, |
| 898 | size, |
| 899 | op |
| 900 | ); |
| 901 | cpu->running = false; |
| 902 | emit_fault_diagnostics(cpu, cpu->pc); |
| 903 | return false; |
| 904 | } |
| 905 | if (stack_guard_overlaps(guard, addr32, end)) { |
| 906 | printf( |
| 907 | "Stack guard %s violation at PC=%08x: base=%s (0x%08x) " |
| 908 | "addr=%08x " |
| 909 | "size=%u guard=%u (%s)\n", |
| 910 | kind, |
| 911 | cpu->pc, |
| 912 | reg_names[base_reg], |
| 913 | (u32)cpu->regs[base_reg], |
| 914 | addr32, |
| 915 | size, |
| 916 | guard, |
| 917 | op |
| 918 | ); |
| 919 | cpu->running = false; |
| 920 | emit_fault_diagnostics(cpu, cpu->pc); |
| 921 | return false; |
| 922 | } |
| 923 | } |
| 924 | return true; |
| 925 | } |
| 926 | |
| 927 | /* Validate that a register holds a valid stack address. */ |
| 928 | static bool validate_stack_register( |
| 929 | struct cpu *cpu, reg_t reg, const char *label, u32 pc, bool optional |
| 930 | ) { |
| 931 | /* Skip validation for performance if disabled. */ |
| 932 | if (!g_opts.validate_memory) |
| 933 | return true; |
| 934 | |
| 935 | u64 value = cpu->regs[reg]; |
| 936 | |
| 937 | if (optional && value == 0) |
| 938 | return true; |
| 939 | |
| 940 | /* Detect addresses with upper bits set -- these can never be valid |
| 941 | * stack addresses in the emulator's physical memory. */ |
| 942 | if (value > (u64)UINT32_MAX || (u32)value < stack_bottom() || |
| 943 | (u32)value > memory_top()) { |
| 944 | printf( |
| 945 | "%s (%s) out of stack bounds at PC=%08x: value=%016llx\n", |
| 946 | label, |
| 947 | reg_names[reg], |
| 948 | pc, |
| 949 | (unsigned long long)value |
| 950 | ); |
| 951 | cpu->running = false; |
| 952 | emit_fault_diagnostics(cpu, pc); |
| 953 | return false; |
| 954 | } |
| 955 | |
| 956 | u32 guard = stack_guard_bytes(); |
| 957 | |
| 958 | if (stack_guard_contains(guard, (u32)value)) { |
| 959 | printf( |
| 960 | "Stack guard triggered by %s (%s) at PC=%08x: value=%08x " |
| 961 | "guard=%u\n", |
| 962 | label, |
| 963 | reg_names[reg], |
| 964 | pc, |
| 965 | (u32)value, |
| 966 | guard |
| 967 | ); |
| 968 | cpu->running = false; |
| 969 | emit_fault_diagnostics(cpu, pc); |
| 970 | return false; |
| 971 | } |
| 972 | return true; |
| 973 | } |
| 974 | |
| 975 | /* Toggle stack guarding in the TUI and re-validate live stack registers. */ |
| 976 | static void toggle_stack_guard(struct cpu *cpu) { |
| 977 | g_opts.stack_guard = !g_opts.stack_guard; |
| 978 | |
| 979 | printf( |
| 980 | "\nStack guard %s (%u bytes)\n", |
| 981 | g_opts.stack_guard ? "enabled" : "disabled", |
| 982 | STACK_GUARD_BYTES |
| 983 | ); |
| 984 | |
| 985 | if (g_opts.stack_guard && cpu->running) { |
| 986 | validate_stack_register(cpu, SP, "SP", cpu->pc, false); |
| 987 | if (cpu->running) { |
| 988 | validate_stack_register(cpu, FP, "FP", cpu->pc, true); |
| 989 | } |
| 990 | } |
| 991 | } |
| 992 | |
| 993 | /* Get terminal dimensions. */ |
| 994 | static struct termsize termsize(void) { |
| 995 | struct winsize w; |
| 996 | struct termsize size = { 24, 80 }; /* Default fallback. */ |
| 997 | |
| 998 | if (ioctl(STDOUT_FILENO, TIOCGWINSZ, &w) != -1) { |
| 999 | size.rows = w.ws_row; |
| 1000 | size.cols = w.ws_col; |
| 1001 | } |
| 1002 | return size; |
| 1003 | } |
| 1004 | |
| 1005 | /* Take a snapshot of the current CPU and memory state. */ |
| 1006 | static void snapshot_save(struct cpu *cpu) { |
| 1007 | int nexti = (snapshots.head + 1 + MAX_SNAPSHOTS) % MAX_SNAPSHOTS; |
| 1008 | |
| 1009 | memcpy(&snapshots.snapshots[nexti].cpu, cpu, sizeof(struct cpu)); |
| 1010 | memcpy(snapshots.snapshots[nexti].memory, memory, g_opts.memory_size); |
| 1011 | |
| 1012 | /* Fix the program pointer to reference the snapshot's own memory. */ |
| 1013 | snapshots.snapshots[nexti].cpu.program = |
| 1014 | (instr_t *)snapshots.snapshots[nexti].memory; |
| 1015 | |
| 1016 | snapshots.head = nexti; |
| 1017 | if (snapshots.count < MAX_SNAPSHOTS) |
| 1018 | snapshots.count++; |
| 1019 | } |
| 1020 | |
| 1021 | /* Restore the most recent snapshot, returning false if none remain. */ |
| 1022 | static bool snapshot_restore(struct cpu *cpu) { |
| 1023 | if (snapshots.count <= 1) |
| 1024 | return false; |
| 1025 | |
| 1026 | snapshots.head = (snapshots.head + MAX_SNAPSHOTS - 1) % MAX_SNAPSHOTS; |
| 1027 | snapshots.count--; |
| 1028 | |
| 1029 | int previ = snapshots.head; |
| 1030 | memcpy(cpu, &snapshots.snapshots[previ].cpu, sizeof(struct cpu)); |
| 1031 | memcpy(memory, snapshots.snapshots[previ].memory, g_opts.memory_size); |
| 1032 | |
| 1033 | /* Fix the program pointer to reference the live memory buffer. */ |
| 1034 | cpu->program = (instr_t *)memory; |
| 1035 | |
| 1036 | return true; |
| 1037 | } |
| 1038 | |
| 1039 | /* Initialize the snapshot buffer with an initial snapshot. */ |
| 1040 | static void snapshot_init(struct cpu *cpu) { |
| 1041 | snapshots.head = -1; |
| 1042 | snapshots.count = 0; |
| 1043 | snapshot_save(cpu); |
| 1044 | } |
| 1045 | |
| 1046 | /* Reset the headless instruction trace buffer. */ |
| 1047 | static void trace_reset(void) { |
| 1048 | headless_trace.head = -1; |
| 1049 | headless_trace.count = 0; |
| 1050 | } |
| 1051 | |
| 1052 | /* Record the current instruction into the trace ring buffer. */ |
| 1053 | static void trace_record(struct cpu *cpu, instr_t ins) { |
| 1054 | if (!g_opts.trace_enabled || !g_opts.trace_headless) |
| 1055 | return; |
| 1056 | |
| 1057 | int next = (headless_trace.head + 1 + TRACE_HISTORY) % TRACE_HISTORY; |
| 1058 | |
| 1059 | headless_trace.head = next; |
| 1060 | headless_trace.entries[next].pc = cpu->pc; |
| 1061 | headless_trace.entries[next].instr = ins; |
| 1062 | memcpy(headless_trace.entries[next].regs, cpu->regs, sizeof(cpu->regs)); |
| 1063 | if (headless_trace.count < TRACE_HISTORY) |
| 1064 | headless_trace.count++; |
| 1065 | } |
| 1066 | |
| 1067 | /* Dump the headless instruction trace to stdout. */ |
| 1068 | static bool trace_dump(u32 fault_pc) { |
| 1069 | if (!g_opts.trace_enabled || !g_opts.trace_headless || |
| 1070 | headless_trace.count == 0) |
| 1071 | return false; |
| 1072 | |
| 1073 | int limit = (int)g_opts.trace_depth; |
| 1074 | if (limit <= 0) |
| 1075 | limit = FAULT_TRACE_DEPTH; |
| 1076 | if (limit > TRACE_HISTORY) |
| 1077 | limit = TRACE_HISTORY; |
| 1078 | if (limit > headless_trace.count) |
| 1079 | limit = headless_trace.count; |
| 1080 | |
| 1081 | printf("Headless trace (newest first):\n"); |
| 1082 | for (int i = 0; i < limit; i++) { |
| 1083 | int idx = (headless_trace.head - i + TRACE_HISTORY) % TRACE_HISTORY; |
| 1084 | struct trace_entry *entry = &headless_trace.entries[idx]; |
| 1085 | char istr[MAX_INSTR_STR_LEN] = { 0 }; |
| 1086 | |
| 1087 | sprint_instr(entry->instr, istr, true); |
| 1088 | |
| 1089 | printf( |
| 1090 | " [%d] PC=%08x %s%s\n", |
| 1091 | i, |
| 1092 | entry->pc, |
| 1093 | istr, |
| 1094 | (entry->pc == fault_pc) ? " <-- fault" : "" |
| 1095 | ); |
| 1096 | printf( |
| 1097 | " SP=%08x FP=%08x RA=%08x A0=%08x A1=%08x A2=%08x\n", |
| 1098 | (u32)entry->regs[SP], |
| 1099 | (u32)entry->regs[FP], |
| 1100 | (u32)entry->regs[RA], |
| 1101 | (u32)entry->regs[A0], |
| 1102 | (u32)entry->regs[A1], |
| 1103 | (u32)entry->regs[A2] |
| 1104 | ); |
| 1105 | } |
| 1106 | return true; |
| 1107 | } |
| 1108 | |
| 1109 | /* Dump recent snapshot history to stdout for fault diagnostics. */ |
| 1110 | static bool snapshot_dump_history(struct cpu *cpu, u32 fault_pc) { |
| 1111 | (void)cpu; |
| 1112 | if (snapshots.count == 0) |
| 1113 | return false; |
| 1114 | |
| 1115 | int limit = FAULT_TRACE_DEPTH; |
| 1116 | if (limit > snapshots.count) |
| 1117 | limit = snapshots.count; |
| 1118 | |
| 1119 | printf("Snapshot history (newest first):\n"); |
| 1120 | for (int i = 0; i < limit; i++) { |
| 1121 | int idx = (snapshots.head - i + MAX_SNAPSHOTS) % MAX_SNAPSHOTS; |
| 1122 | struct snapshot *snap = &snapshots.snapshots[idx]; |
| 1123 | u32 next_pc = snap->cpu.pc; |
| 1124 | u32 exec_pc = next_pc >= INSTR_SIZE ? next_pc - INSTR_SIZE : next_pc; |
| 1125 | char istr[MAX_INSTR_STR_LEN] = { 0 }; |
| 1126 | u32 instr_index = exec_pc / INSTR_SIZE; |
| 1127 | |
| 1128 | if (instr_index < snap->cpu.programsize) { |
| 1129 | sprint_instr(snap->cpu.program[instr_index], istr, true); |
| 1130 | } else { |
| 1131 | snprintf(istr, sizeof(istr), "<pc %08x>", exec_pc); |
| 1132 | } |
| 1133 | |
| 1134 | printf( |
| 1135 | " [%d] PC next=%08x prev=%08x %s%s\n", |
| 1136 | i, |
| 1137 | next_pc, |
| 1138 | exec_pc, |
| 1139 | istr, |
| 1140 | (exec_pc == fault_pc || next_pc == fault_pc) ? " <-- fault" : "" |
| 1141 | ); |
| 1142 | printf( |
| 1143 | " SP=%08x FP=%08x RA=%08x A0=%08x\n", |
| 1144 | (u32)snap->cpu.regs[SP], |
| 1145 | (u32)snap->cpu.regs[FP], |
| 1146 | (u32)snap->cpu.regs[RA], |
| 1147 | (u32)snap->cpu.regs[A0] |
| 1148 | ); |
| 1149 | } |
| 1150 | return true; |
| 1151 | } |
| 1152 | |
| 1153 | /* Emit runtime fault diagnostics including trace and snapshot history. */ |
| 1154 | static void emit_fault_diagnostics(struct cpu *cpu, u32 pc) { |
| 1155 | if (cpu->faulted) |
| 1156 | return; |
| 1157 | |
| 1158 | cpu->faulted = true; |
| 1159 | |
| 1160 | printf("\n--- runtime fault diagnostics ---\n"); |
| 1161 | bool printed = false; |
| 1162 | |
| 1163 | printed |= trace_dump(pc); |
| 1164 | printed |= snapshot_dump_history(cpu, pc); |
| 1165 | |
| 1166 | if (!printed) { |
| 1167 | printf("No trace data available.\n"); |
| 1168 | } |
| 1169 | printf("--- end diagnostics ---\n"); |
| 1170 | fflush(stdout); |
| 1171 | } |
| 1172 | |
| 1173 | /* Return true if the CPU's PC is outside the loaded program bounds. */ |
| 1174 | static inline bool cpu_out_of_bounds(struct cpu *cpu) { |
| 1175 | if (!g_opts.validate_memory) |
| 1176 | return false; |
| 1177 | if (program_bytes == 0) |
| 1178 | return true; |
| 1179 | if (cpu->pc < program_base) |
| 1180 | return true; |
| 1181 | return (cpu->pc - program_base) >= program_bytes; |
| 1182 | } |
| 1183 | |
| 1184 | /* Last executed PC, used for detecting branches/jumps in trace mode. */ |
| 1185 | static u32 last_executed_pc = 0; |
| 1186 | |
| 1187 | /* Reset CPU state (keeping program loaded). */ |
| 1188 | static void cpu_reset(struct cpu *cpu) { |
| 1189 | trace_reset(); |
| 1190 | memset(cpu->regs, 0, sizeof(cpu->regs)); |
| 1191 | |
| 1192 | /* Set SP to the top of the usable stack, aligned to 16 bytes |
| 1193 | * as required by the RISC-V ABI. */ |
| 1194 | cpu->regs[SP] = stack_usable_top() & ~0xF; |
| 1195 | cpu->pc = program_base; |
| 1196 | cpu->running = true; |
| 1197 | cpu->faulted = false; |
| 1198 | cpu->ebreak = false; |
| 1199 | cpu->modified = ZERO; |
| 1200 | last_executed_pc = 0; |
| 1201 | } |
| 1202 | |
| 1203 | /* Initialize CPU and memory to a clean state. */ |
| 1204 | static void cpu_init(struct cpu *cpu) { |
| 1205 | memset(memory, 0, g_opts.memory_size); |
| 1206 | cpu->program = (instr_t *)memory; |
| 1207 | cpu->programsize = 0; |
| 1208 | trace_reset(); |
| 1209 | guest_fd_table_init(); |
| 1210 | cpu_reset(cpu); |
| 1211 | } |
| 1212 | |
| 1213 | /* Open a file via the openat syscall (56). */ |
| 1214 | static i32 ecall_openat(u32 pathname_addr, i32 flags) { |
| 1215 | if (pathname_addr >= g_opts.memory_size) |
| 1216 | return -1; |
| 1217 | |
| 1218 | /* Find the null terminator to validate the string is in bounds. */ |
| 1219 | u32 path_end = pathname_addr; |
| 1220 | while (path_end < g_opts.memory_size && memory[path_end] != 0) |
| 1221 | path_end++; |
| 1222 | if (path_end >= g_opts.memory_size) |
| 1223 | return -1; |
| 1224 | |
| 1225 | i32 host_fd = open((const char *)&memory[pathname_addr], flags, 0644); |
| 1226 | if (host_fd < 0) |
| 1227 | return -1; |
| 1228 | |
| 1229 | i32 guest_fd = guest_fd_table_add(host_fd); |
| 1230 | if (guest_fd < 0) { |
| 1231 | close(host_fd); |
| 1232 | return -1; |
| 1233 | } |
| 1234 | return guest_fd; |
| 1235 | } |
| 1236 | |
| 1237 | /* Close a file descriptor via the close syscall (57). */ |
| 1238 | static i32 ecall_close(i32 guest_fd) { |
| 1239 | /* Don't close standard streams. */ |
| 1240 | if (guest_fd < 3) |
| 1241 | return 0; |
| 1242 | |
| 1243 | i32 host_fd = guest_fd_table_get(guest_fd); |
| 1244 | if (host_fd >= 0) { |
| 1245 | i32 result = close(host_fd); |
| 1246 | guest_fd_table_remove(guest_fd); |
| 1247 | return result; |
| 1248 | } |
| 1249 | return -1; |
| 1250 | } |
| 1251 | |
| 1252 | /* Load a binary section from disk into emulator memory at the given offset. */ |
| 1253 | static u32 load_section( |
| 1254 | const char *filepath, |
| 1255 | const char *suffix, |
| 1256 | u32 offset, |
| 1257 | u32 limit, |
| 1258 | const char *label |
| 1259 | ) { |
| 1260 | char path[PATH_MAX]; |
| 1261 | snprintf(path, sizeof(path), "%s.%s", filepath, suffix); |
| 1262 | |
| 1263 | FILE *file = fopen(path, "rb"); |
| 1264 | if (!file) |
| 1265 | return 0; |
| 1266 | |
| 1267 | if (fseek(file, 0, SEEK_END) != 0) { |
| 1268 | fclose(file); |
| 1269 | bail("failed to seek %s section", label); |
| 1270 | } |
| 1271 | long size = ftell(file); |
| 1272 | if (size < 0) { |
| 1273 | fclose(file); |
| 1274 | bail("failed to determine size of %s section", label); |
| 1275 | } |
| 1276 | if (fseek(file, 0, SEEK_SET) != 0) { |
| 1277 | fclose(file); |
| 1278 | bail("failed to rewind %s section", label); |
| 1279 | } |
| 1280 | if (size == 0) { |
| 1281 | fclose(file); |
| 1282 | return 0; |
| 1283 | } |
| 1284 | |
| 1285 | u32 u_size = (u32)size; |
| 1286 | u64 end = (u64)offset + (u64)u_size; |
| 1287 | if (end > (u64)limit) { |
| 1288 | fclose(file); |
| 1289 | u32 max_size = limit - offset; |
| 1290 | bail( |
| 1291 | "%s section too large for emulator memory: required %u bytes, max " |
| 1292 | "%u bytes", |
| 1293 | label, |
| 1294 | u_size, |
| 1295 | max_size |
| 1296 | ); |
| 1297 | } |
| 1298 | if (end > (u64)g_opts.memory_size) { |
| 1299 | fclose(file); |
| 1300 | u32 max_size = g_opts.memory_size - offset; |
| 1301 | bail( |
| 1302 | "%s section exceeds physical memory: required %u bytes at offset " |
| 1303 | "%u, " |
| 1304 | "but only %u bytes available (total memory=%u, use " |
| 1305 | "-memory-size=... or recompile emulator with a larger " |
| 1306 | "MEMORY_SIZE)", |
| 1307 | label, |
| 1308 | u_size, |
| 1309 | offset, |
| 1310 | max_size, |
| 1311 | g_opts.memory_size |
| 1312 | ); |
| 1313 | } |
| 1314 | size_t read = fread(&memory[offset], 1, u_size, file); |
| 1315 | fclose(file); |
| 1316 | |
| 1317 | if (read != u_size) { |
| 1318 | bail( |
| 1319 | "could not read entire %s section: read %zu bytes, expected %u " |
| 1320 | "bytes (offset=%u, limit=%u)", |
| 1321 | label, |
| 1322 | read, |
| 1323 | u_size, |
| 1324 | offset, |
| 1325 | limit |
| 1326 | ); |
| 1327 | } |
| 1328 | return u_size; |
| 1329 | } |
| 1330 | |
| 1331 | /* Prepare the environment block (argv) on the guest stack. */ |
| 1332 | static void prepare_env(struct cpu *cpu, int argc, char **argv) { |
| 1333 | if (argc < 0 || argv == NULL) |
| 1334 | argc = 0; |
| 1335 | |
| 1336 | usize bytes = 0; |
| 1337 | for (int i = 0; i < argc; i++) |
| 1338 | bytes += strlen(argv[i]) + 1; /* Include terminating NUL. */ |
| 1339 | |
| 1340 | /* In RV64, slices are 16 bytes (8-byte ptr + 4-byte len + 4 padding). */ |
| 1341 | u32 slice_size = 16; |
| 1342 | u32 slice_array_size = (u32)argc * slice_size; |
| 1343 | u32 base_size = slice_size + slice_array_size; |
| 1344 | u32 total_size = align(base_size + (u32)bytes, 16); |
| 1345 | |
| 1346 | /* Place the env block below the current stack pointer so it doesn't |
| 1347 | * overlap with uninitialized static data. The .rw.data file only |
| 1348 | * contains initialized statics; undefined statics occupy memory after |
| 1349 | * the loaded data but are not in the file. Placing the env block in |
| 1350 | * the data region would clobber those zero-initialized areas. */ |
| 1351 | u32 sp = (u32)cpu->regs[SP]; |
| 1352 | u32 env_addr = (sp - total_size) & ~0xFu; |
| 1353 | |
| 1354 | if (env_addr <= DATA_MEMORY_START + data_bytes) |
| 1355 | bail("not enough memory to prepare environment block"); |
| 1356 | |
| 1357 | /* Move SP below the env block so the program's stack doesn't overwrite it. |
| 1358 | */ |
| 1359 | cpu->regs[SP] = env_addr; |
| 1360 | |
| 1361 | u32 slices_addr = env_addr + slice_size; |
| 1362 | u32 strings_addr = slices_addr + (argc > 0 ? slice_array_size : 0); |
| 1363 | |
| 1364 | /* Write the Env slice header. */ |
| 1365 | memory_store_u64(env_addr, argc > 0 ? slices_addr : 0); |
| 1366 | memory_store_u32(env_addr + 8, (u32)argc); |
| 1367 | |
| 1368 | /* Copy argument strings and populate slices. */ |
| 1369 | u32 curr = strings_addr; |
| 1370 | for (int i = 0; i < argc; i++) { |
| 1371 | size_t len = strlen(argv[i]); |
| 1372 | if (curr + len >= g_opts.memory_size) |
| 1373 | bail("environment string does not fit in emulator memory"); |
| 1374 | |
| 1375 | memcpy(&memory[curr], argv[i], len); |
| 1376 | memory[curr + len] = 0; /* Null-terminate for syscall compatibility. */ |
| 1377 | |
| 1378 | u32 slice_entry = slices_addr + (u32)i * slice_size; |
| 1379 | memory_store_u64(slice_entry, curr); |
| 1380 | memory_store_u32(slice_entry + 8, (u32)len); |
| 1381 | |
| 1382 | curr += (u32)len + 1; |
| 1383 | } |
| 1384 | cpu->regs[A0] = env_addr; |
| 1385 | cpu->regs[A1] = env_addr; |
| 1386 | } |
| 1387 | |
| 1388 | /* Load debug information from the .debug file. */ |
| 1389 | static void debug_load(const char *program_path) { |
| 1390 | char debugpath[PATH_MAX]; |
| 1391 | snprintf(debugpath, sizeof(debugpath), "%s.debug", program_path); |
| 1392 | |
| 1393 | FILE *file = fopen(debugpath, "rb"); |
| 1394 | if (!file) |
| 1395 | return; /* Debug file is optional. */ |
| 1396 | |
| 1397 | g_debug.capacity = 64; |
| 1398 | g_debug.entries = malloc(sizeof(struct debug_entry) * g_debug.capacity); |
| 1399 | g_debug.count = 0; |
| 1400 | |
| 1401 | if (!g_debug.entries) { |
| 1402 | fclose(file); |
| 1403 | return; |
| 1404 | } |
| 1405 | while (!feof(file)) { |
| 1406 | struct debug_entry entry; |
| 1407 | |
| 1408 | if (fread(&entry.pc, sizeof(u32), 1, file) != 1) |
| 1409 | break; |
| 1410 | if (fread(&entry.offset, sizeof(u32), 1, file) != 1) |
| 1411 | break; |
| 1412 | |
| 1413 | /* Read null-terminated file path. */ |
| 1414 | size_t i = 0; |
| 1415 | int c; |
| 1416 | while (i < PATH_MAX - 1 && (c = fgetc(file)) != EOF && c != '\0') { |
| 1417 | entry.file[i++] = (char)c; |
| 1418 | } |
| 1419 | entry.file[i] = '\0'; |
| 1420 | |
| 1421 | if (c == EOF && i == 0) |
| 1422 | break; |
| 1423 | |
| 1424 | /* Grow array if needed. */ |
| 1425 | if (g_debug.count >= g_debug.capacity) { |
| 1426 | g_debug.capacity *= 2; |
| 1427 | g_debug.entries = realloc( |
| 1428 | g_debug.entries, sizeof(struct debug_entry) * g_debug.capacity |
| 1429 | ); |
| 1430 | if (!g_debug.entries) { |
| 1431 | fclose(file); |
| 1432 | return; |
| 1433 | } |
| 1434 | } |
| 1435 | g_debug.entries[g_debug.count++] = entry; |
| 1436 | } |
| 1437 | fclose(file); |
| 1438 | } |
| 1439 | |
| 1440 | /* Look up source location for a given PC. */ |
| 1441 | static struct debug_entry *debug_lookup(u32 pc) { |
| 1442 | struct debug_entry *best = NULL; |
| 1443 | |
| 1444 | for (size_t i = 0; i < g_debug.count; i++) { |
| 1445 | if (g_debug.entries[i].pc == pc) { |
| 1446 | return &g_debug.entries[i]; |
| 1447 | } |
| 1448 | /* Track the closest entry at or before this PC. */ |
| 1449 | if (g_debug.entries[i].pc <= pc) { |
| 1450 | best = &g_debug.entries[i]; |
| 1451 | } |
| 1452 | } |
| 1453 | return best; |
| 1454 | } |
| 1455 | |
| 1456 | /* Compute the line number from a file path and byte offset. */ |
| 1457 | static int line_from_offset(const char *filepath, u32 offset) { |
| 1458 | FILE *file = fopen(filepath, "r"); |
| 1459 | if (!file) |
| 1460 | return 0; |
| 1461 | |
| 1462 | u32 line = 1; |
| 1463 | for (u32 i = 0; i < offset; i++) { |
| 1464 | int c = fgetc(file); |
| 1465 | if (c == EOF) |
| 1466 | break; |
| 1467 | if (c == '\n') |
| 1468 | line++; |
| 1469 | } |
| 1470 | fclose(file); |
| 1471 | |
| 1472 | return line; |
| 1473 | } |
| 1474 | |
| 1475 | /* Load the program binary and data sections into memory. */ |
| 1476 | static void program_init(struct cpu *cpu, const char *filepath) { |
| 1477 | program_bytes = 0; |
| 1478 | data_bytes = 0; |
| 1479 | rodata_bytes = load_section( |
| 1480 | filepath, "ro.data", DATA_RO_OFFSET, DATA_MEMORY_START, "ro.data" |
| 1481 | ); |
| 1482 | program_base = align(DATA_RO_OFFSET + rodata_bytes, WORD_SIZE); |
| 1483 | |
| 1484 | if (g_opts.debug_enabled) |
| 1485 | debug_load(filepath); |
| 1486 | |
| 1487 | FILE *file = fopen(filepath, "rb"); |
| 1488 | if (!file) |
| 1489 | bail("failed to open file '%s'", filepath); |
| 1490 | if (fseek(file, 0, SEEK_END) != 0) { |
| 1491 | fclose(file); |
| 1492 | bail("failed to seek program '%s'", filepath); |
| 1493 | } |
| 1494 | long size = ftell(file); |
| 1495 | if (size <= 0 || size > PROGRAM_SIZE) { |
| 1496 | fclose(file); |
| 1497 | bail( |
| 1498 | "invalid file size: %ld; maximum program size is %d bytes", |
| 1499 | size, |
| 1500 | PROGRAM_SIZE |
| 1501 | ); |
| 1502 | } |
| 1503 | if (program_base + (u32)size > DATA_MEMORY_START) { |
| 1504 | fclose(file); |
| 1505 | bail("text section exceeds available program memory"); |
| 1506 | } |
| 1507 | if (fseek(file, 0, SEEK_SET) != 0) { |
| 1508 | fclose(file); |
| 1509 | bail("failed to rewind program '%s'", filepath); |
| 1510 | } |
| 1511 | |
| 1512 | usize read = fread(&memory[program_base], 1, size, file); |
| 1513 | fclose(file); |
| 1514 | if (read != (size_t)size) |
| 1515 | bail("could not read entire file"); |
| 1516 | |
| 1517 | program_bytes = (u32)size; |
| 1518 | cpu->programsize = (u32)size / sizeof(instr_t); |
| 1519 | |
| 1520 | u32 data_limit = DATA_MEMORY_START + g_opts.data_memory_size; |
| 1521 | if (data_limit > g_opts.memory_size) |
| 1522 | data_limit = g_opts.memory_size; |
| 1523 | |
| 1524 | data_bytes = load_section( |
| 1525 | filepath, "rw.data", DATA_MEMORY_START, data_limit, "rw.data" |
| 1526 | ); |
| 1527 | cpu->pc = program_base; |
| 1528 | } |
| 1529 | |
| 1530 | /* Execute a single instruction. */ |
| 1531 | static void cpu_execute(struct cpu *cpu, enum display display, bool headless) { |
| 1532 | if (cpu_out_of_bounds(cpu)) { |
| 1533 | cpu->running = false; |
| 1534 | emit_fault_diagnostics(cpu, cpu->pc); |
| 1535 | if (headless) { |
| 1536 | fprintf(stderr, "program is out of bounds\n"); |
| 1537 | return; |
| 1538 | } |
| 1539 | bail("program is out of bounds"); |
| 1540 | } |
| 1541 | |
| 1542 | u32 executed_pc = cpu->pc; |
| 1543 | instr_t ins = cpu->program[cpu->pc / sizeof(instr_t)]; |
| 1544 | u32 pc_next = cpu->pc + INSTR_SIZE; |
| 1545 | u32 opcode = ins.r.opcode; |
| 1546 | |
| 1547 | cpu->modified = ZERO; |
| 1548 | trace_record(cpu, ins); |
| 1549 | |
| 1550 | /* Print instruction if tracing is enabled in headless mode. |
| 1551 | * Skip NOPs (addi x0, x0, 0 = 0x00000013). */ |
| 1552 | if (headless && g_opts.trace_print_instructions && ins.raw != 0x00000013) { |
| 1553 | /* Print ellipsis if we jumped to a non-sequential instruction. */ |
| 1554 | if (last_executed_pc != 0 && |
| 1555 | executed_pc != last_executed_pc + INSTR_SIZE) { |
| 1556 | printf("%s :%s\n", COLOR_GREY, COLOR_RESET); |
| 1557 | } |
| 1558 | |
| 1559 | char istr[MAX_INSTR_STR_LEN] = { 0 }; |
| 1560 | int len = sprint_instr(ins, istr, true); |
| 1561 | int padding = INSTR_STR_LEN - len; |
| 1562 | if (padding < 0) |
| 1563 | padding = 0; |
| 1564 | printf( |
| 1565 | "%s%08x%s %s%-*s%s", |
| 1566 | COLOR_GREY, |
| 1567 | executed_pc, |
| 1568 | COLOR_RESET, |
| 1569 | istr, |
| 1570 | padding, |
| 1571 | "", |
| 1572 | COLOR_GREY |
| 1573 | ); |
| 1574 | |
| 1575 | /* Print all non-zero registers. */ |
| 1576 | bool first = true; |
| 1577 | for (int i = 0; i < REGISTERS; i++) { |
| 1578 | if (cpu->regs[i] != 0) { |
| 1579 | if (!first) |
| 1580 | printf(" "); |
| 1581 | printf("%s=%08x", reg_names[i], (u32)cpu->regs[i]); |
| 1582 | first = false; |
| 1583 | } |
| 1584 | } |
| 1585 | printf("%s\n", COLOR_RESET); |
| 1586 | } |
| 1587 | |
| 1588 | switch (opcode) { |
| 1589 | case OP_LUI: |
| 1590 | if (ins.u.rd != 0) { |
| 1591 | u32 lui_val = ins.u.imm_31_12 << 12; |
| 1592 | cpu->regs[ins.u.rd] = |
| 1593 | (u64)(i64)(i32)lui_val; /* RV64: sign-extend to 64 bits. */ |
| 1594 | cpu->modified = ins.u.rd; |
| 1595 | } |
| 1596 | break; |
| 1597 | |
| 1598 | case OP_AUIPC: |
| 1599 | if (ins.u.rd != 0) { |
| 1600 | u32 auipc_val = ins.u.imm_31_12 << 12; |
| 1601 | cpu->regs[ins.u.rd] = |
| 1602 | cpu->pc + |
| 1603 | (u64)(i64)(i32)auipc_val; /* RV64: sign-extend offset. */ |
| 1604 | cpu->modified = (reg_t)ins.u.rd; |
| 1605 | } |
| 1606 | break; |
| 1607 | |
| 1608 | case OP_JAL: { |
| 1609 | i32 imm = get_j_imm(ins); |
| 1610 | if (ins.j.rd != 0) { |
| 1611 | cpu->regs[ins.j.rd] = pc_next; |
| 1612 | cpu->modified = (reg_t)ins.j.rd; |
| 1613 | } |
| 1614 | pc_next = cpu->pc + imm; |
| 1615 | break; |
| 1616 | } |
| 1617 | |
| 1618 | case OP_JALR: { |
| 1619 | i32 imm = get_i_imm(ins); |
| 1620 | if (ins.i.rd != 0) { |
| 1621 | cpu->regs[ins.i.rd] = pc_next; |
| 1622 | cpu->modified = (reg_t)ins.i.rd; |
| 1623 | } |
| 1624 | /* Calculate target address in full 64-bit precision. */ |
| 1625 | u64 jalr_target = (cpu->regs[ins.i.rs1] + (i64)imm) & ~(u64)1; |
| 1626 | /* Check if this is a RET instruction (jalr x0, ra, 0). */ |
| 1627 | if (ins.i.rd == 0 && ins.i.rs1 == 1 && imm == 0 && jalr_target == 0) { |
| 1628 | cpu->running = false; |
| 1629 | if (!headless) { |
| 1630 | ui_render(cpu, display); |
| 1631 | |
| 1632 | printf( |
| 1633 | "\n%sProgram terminated with return value %d (0x%08x)%s ", |
| 1634 | COLOR_BOLD_GREEN, |
| 1635 | (i32)cpu->regs[A0], |
| 1636 | (u32)cpu->regs[A0], |
| 1637 | COLOR_RESET |
| 1638 | ); |
| 1639 | } |
| 1640 | } else { |
| 1641 | pc_next = (u32)jalr_target; |
| 1642 | } |
| 1643 | break; |
| 1644 | } |
| 1645 | |
| 1646 | case OP_BRANCH: { |
| 1647 | bool jump = false; |
| 1648 | i32 imm = get_b_imm(ins); |
| 1649 | |
| 1650 | switch (ins.b.funct3) { |
| 1651 | case FUNCT3_BYTE: /* beq. */ |
| 1652 | jump = (cpu->regs[ins.b.rs1] == cpu->regs[ins.b.rs2]); |
| 1653 | break; |
| 1654 | case FUNCT3_HALF: /* bne. */ |
| 1655 | jump = (cpu->regs[ins.b.rs1] != cpu->regs[ins.b.rs2]); |
| 1656 | break; |
| 1657 | case FUNCT3_BYTE_U: /* blt. */ |
| 1658 | jump = ((i64)cpu->regs[ins.b.rs1] < (i64)cpu->regs[ins.b.rs2]); |
| 1659 | break; |
| 1660 | case FUNCT3_HALF_U: /* bge. */ |
| 1661 | jump = ((i64)cpu->regs[ins.b.rs1] >= (i64)cpu->regs[ins.b.rs2]); |
| 1662 | break; |
| 1663 | case FUNCT3_OR: /* bltu. */ |
| 1664 | jump = (cpu->regs[ins.b.rs1] < cpu->regs[ins.b.rs2]); |
| 1665 | break; |
| 1666 | case FUNCT3_AND: /* bgeu. */ |
| 1667 | jump = (cpu->regs[ins.b.rs1] >= cpu->regs[ins.b.rs2]); |
| 1668 | break; |
| 1669 | } |
| 1670 | if (jump) { |
| 1671 | pc_next = cpu->pc + imm; |
| 1672 | } |
| 1673 | break; |
| 1674 | } |
| 1675 | |
| 1676 | case OP_LOAD: { |
| 1677 | i32 imm = get_i_imm(ins); |
| 1678 | u64 addr = cpu->regs[ins.i.rs1] + (i64)imm; |
| 1679 | |
| 1680 | if (ins.i.rd == ZERO) |
| 1681 | break; |
| 1682 | |
| 1683 | cpu->modified = (reg_t)ins.i.rd; |
| 1684 | bool fault = false; |
| 1685 | |
| 1686 | switch (ins.i.funct3) { |
| 1687 | case FUNCT3_BYTE: /* lb. */ |
| 1688 | if (!validate_memory_access(cpu, addr, 1, ins.i.rs1, "lb", false)) { |
| 1689 | fault = true; |
| 1690 | break; |
| 1691 | } |
| 1692 | /* sign_extend returns i32; on RV64 we sign-extend to 64 bits. */ |
| 1693 | cpu->regs[ins.i.rd] = (u64)(i64)sign_extend(memory[addr], 8); |
| 1694 | break; |
| 1695 | case FUNCT3_HALF: /* lh. */ |
| 1696 | if (!validate_memory_access(cpu, addr, 2, ins.i.rs1, "lh", false)) { |
| 1697 | fault = true; |
| 1698 | break; |
| 1699 | } |
| 1700 | cpu->regs[ins.i.rd] = |
| 1701 | (u64)(i64)sign_extend(memory_load_u16(addr), 16); |
| 1702 | break; |
| 1703 | case FUNCT3_WORD: /* lw. */ |
| 1704 | if (!validate_memory_access(cpu, addr, 4, ins.i.rs1, "lw", false)) { |
| 1705 | fault = true; |
| 1706 | break; |
| 1707 | } |
| 1708 | /* RV64: lw sign-extends the 32-bit value to 64 bits. */ |
| 1709 | cpu->regs[ins.i.rd] = (u64)(i64)(i32)memory_load_u32(addr); |
| 1710 | break; |
| 1711 | case 0x6: /* lwu (RV64). */ |
| 1712 | if (!validate_memory_access( |
| 1713 | cpu, addr, 4, ins.i.rs1, "lwu", false |
| 1714 | )) { |
| 1715 | fault = true; |
| 1716 | break; |
| 1717 | } |
| 1718 | cpu->regs[ins.i.rd] = (u64)memory_load_u32(addr); |
| 1719 | break; |
| 1720 | case FUNCT3_BYTE_U: /* lbu. */ |
| 1721 | if (!validate_memory_access( |
| 1722 | cpu, addr, 1, ins.i.rs1, "lbu", false |
| 1723 | )) { |
| 1724 | fault = true; |
| 1725 | break; |
| 1726 | } |
| 1727 | cpu->regs[ins.i.rd] = memory[addr]; |
| 1728 | break; |
| 1729 | case FUNCT3_HALF_U: /* lhu. */ |
| 1730 | if (!validate_memory_access( |
| 1731 | cpu, addr, 2, ins.i.rs1, "lhu", false |
| 1732 | )) { |
| 1733 | fault = true; |
| 1734 | break; |
| 1735 | } |
| 1736 | cpu->regs[ins.i.rd] = memory_load_u16(addr); |
| 1737 | break; |
| 1738 | case 0x3: /* ld (RV64). */ |
| 1739 | if (!validate_memory_access(cpu, addr, 8, ins.i.rs1, "ld", false)) { |
| 1740 | fault = true; |
| 1741 | break; |
| 1742 | } |
| 1743 | cpu->regs[ins.i.rd] = memory_load_u64(addr); |
| 1744 | break; |
| 1745 | } |
| 1746 | if (fault || !cpu->running) |
| 1747 | break; |
| 1748 | break; |
| 1749 | } |
| 1750 | |
| 1751 | case OP_STORE: { |
| 1752 | i32 imm = get_s_imm(ins); |
| 1753 | u64 addr = cpu->regs[ins.s.rs1] + (i64)imm; |
| 1754 | |
| 1755 | switch (ins.s.funct3) { |
| 1756 | case FUNCT3_BYTE: /* sb. */ |
| 1757 | if (!validate_memory_access(cpu, addr, 1, ins.s.rs1, "sb", true)) |
| 1758 | break; |
| 1759 | watch_store(cpu, (u32)addr, 1, (u32)cpu->regs[ins.s.rs2]); |
| 1760 | memory_store_u8(addr, (u8)cpu->regs[ins.s.rs2]); |
| 1761 | break; |
| 1762 | case FUNCT3_HALF: /* sh. */ |
| 1763 | if (!validate_memory_access(cpu, addr, 2, ins.s.rs1, "sh", true)) |
| 1764 | break; |
| 1765 | watch_store(cpu, (u32)addr, 2, (u32)cpu->regs[ins.s.rs2]); |
| 1766 | memory_store_u16(addr, (u16)cpu->regs[ins.s.rs2]); |
| 1767 | break; |
| 1768 | case FUNCT3_WORD: /* sw. */ |
| 1769 | if (!validate_memory_access(cpu, addr, 4, ins.s.rs1, "sw", true)) |
| 1770 | break; |
| 1771 | watch_store(cpu, (u32)addr, 4, (u32)cpu->regs[ins.s.rs2]); |
| 1772 | memory_store_u32(addr, (u32)cpu->regs[ins.s.rs2]); |
| 1773 | break; |
| 1774 | case 0x3: /* sd (RV64). */ |
| 1775 | if (!validate_memory_access(cpu, addr, 8, ins.s.rs1, "sd", true)) |
| 1776 | break; |
| 1777 | watch_store(cpu, (u32)addr, 8, (u32)cpu->regs[ins.s.rs2]); |
| 1778 | memory_store_u64(addr, cpu->regs[ins.s.rs2]); |
| 1779 | break; |
| 1780 | } |
| 1781 | break; |
| 1782 | } |
| 1783 | |
| 1784 | case OP_IMM: { |
| 1785 | i32 imm = get_i_imm(ins); |
| 1786 | u32 shamt_mask = 0x3F; /* RV64: 6-bit shift amounts. */ |
| 1787 | |
| 1788 | if (ins.i.rd == ZERO) |
| 1789 | break; |
| 1790 | |
| 1791 | cpu->modified = (reg_t)ins.i.rd; |
| 1792 | |
| 1793 | switch (ins.i.funct3) { |
| 1794 | case FUNCT3_ADD: /* addi. */ |
| 1795 | cpu->regs[ins.i.rd] = cpu->regs[ins.i.rs1] + imm; |
| 1796 | break; |
| 1797 | case FUNCT3_SLL: /* slli. */ |
| 1798 | cpu->regs[ins.i.rd] = cpu->regs[ins.i.rs1] << (imm & shamt_mask); |
| 1799 | break; |
| 1800 | case FUNCT3_SLT: /* slti. */ |
| 1801 | cpu->regs[ins.i.rd] = |
| 1802 | ((i64)cpu->regs[ins.i.rs1] < (i64)imm) ? 1 : 0; |
| 1803 | break; |
| 1804 | case FUNCT3_SLTU: /* sltiu. */ |
| 1805 | cpu->regs[ins.i.rd] = |
| 1806 | (cpu->regs[ins.i.rs1] < (u64)(i64)imm) ? 1 : 0; |
| 1807 | break; |
| 1808 | case FUNCT3_XOR: /* xori. */ |
| 1809 | cpu->regs[ins.i.rd] = cpu->regs[ins.i.rs1] ^ imm; |
| 1810 | break; |
| 1811 | case FUNCT3_SRL: /* srli/srai. */ |
| 1812 | if ((imm & 0x400) == 0) { |
| 1813 | /* srli -- logical right shift. */ |
| 1814 | cpu->regs[ins.i.rd] = |
| 1815 | cpu->regs[ins.i.rs1] >> (imm & shamt_mask); |
| 1816 | } else { |
| 1817 | /* srai -- arithmetic right shift. */ |
| 1818 | cpu->regs[ins.i.rd] = |
| 1819 | (u64)((i64)cpu->regs[ins.i.rs1] >> (imm & shamt_mask)); |
| 1820 | } |
| 1821 | break; |
| 1822 | case FUNCT3_OR: /* ori. */ |
| 1823 | cpu->regs[ins.i.rd] = cpu->regs[ins.i.rs1] | imm; |
| 1824 | break; |
| 1825 | case FUNCT3_AND: /* andi. */ |
| 1826 | cpu->regs[ins.i.rd] = cpu->regs[ins.i.rs1] & imm; |
| 1827 | break; |
| 1828 | } |
| 1829 | break; |
| 1830 | } |
| 1831 | |
| 1832 | case OP_IMM_32: { |
| 1833 | /* RV64I: 32-bit immediate operations (ADDIW, SLLIW, SRLIW, SRAIW). |
| 1834 | * These operate on the lower 32 bits and sign-extend the result. */ |
| 1835 | i32 imm = get_i_imm(ins); |
| 1836 | |
| 1837 | if (ins.i.rd == ZERO) |
| 1838 | break; |
| 1839 | |
| 1840 | cpu->modified = (reg_t)ins.i.rd; |
| 1841 | |
| 1842 | switch (ins.i.funct3) { |
| 1843 | case FUNCT3_ADD: { /* addiw. */ |
| 1844 | i32 result = (i32)cpu->regs[ins.i.rs1] + imm; |
| 1845 | cpu->regs[ins.i.rd] = (u64)(i64)result; |
| 1846 | break; |
| 1847 | } |
| 1848 | case FUNCT3_SLL: { /* slliw. */ |
| 1849 | i32 result = (i32)((u32)cpu->regs[ins.i.rs1] << (imm & 0x1F)); |
| 1850 | cpu->regs[ins.i.rd] = (u64)(i64)result; |
| 1851 | break; |
| 1852 | } |
| 1853 | case FUNCT3_SRL: { /* srliw/sraiw. */ |
| 1854 | if ((imm & 0x400) == 0) { |
| 1855 | /* srliw -- logical right shift, then sign-extend. */ |
| 1856 | i32 result = (i32)((u32)cpu->regs[ins.i.rs1] >> (imm & 0x1F)); |
| 1857 | cpu->regs[ins.i.rd] = (u64)(i64)result; |
| 1858 | } else { |
| 1859 | /* sraiw -- arithmetic right shift, then sign-extend. */ |
| 1860 | i32 result = (i32)cpu->regs[ins.i.rs1] >> (imm & 0x1F); |
| 1861 | cpu->regs[ins.i.rd] = (u64)(i64)result; |
| 1862 | } |
| 1863 | break; |
| 1864 | } |
| 1865 | } |
| 1866 | break; |
| 1867 | } |
| 1868 | |
| 1869 | case OP_OP: { |
| 1870 | if (ins.r.rd == ZERO) |
| 1871 | break; |
| 1872 | |
| 1873 | cpu->modified = (reg_t)ins.r.rd; |
| 1874 | |
| 1875 | switch (ins.r.funct7) { |
| 1876 | case FUNCT7_NORMAL: { |
| 1877 | u32 shamt_mask = 0x3F; |
| 1878 | switch (ins.r.funct3) { |
| 1879 | case FUNCT3_ADD: /* add. */ |
| 1880 | cpu->regs[ins.r.rd] = |
| 1881 | cpu->regs[ins.r.rs1] + cpu->regs[ins.r.rs2]; |
| 1882 | break; |
| 1883 | case FUNCT3_SLL: /* sll. */ |
| 1884 | cpu->regs[ins.r.rd] = cpu->regs[ins.r.rs1] |
| 1885 | << (cpu->regs[ins.r.rs2] & shamt_mask); |
| 1886 | break; |
| 1887 | case FUNCT3_SLT: /* slt. */ |
| 1888 | cpu->regs[ins.r.rd] = |
| 1889 | ((i64)cpu->regs[ins.r.rs1] < (i64)cpu->regs[ins.r.rs2]) ? 1 |
| 1890 | : 0; |
| 1891 | break; |
| 1892 | case FUNCT3_SLTU: /* sltu. */ |
| 1893 | cpu->regs[ins.r.rd] = |
| 1894 | (cpu->regs[ins.r.rs1] < cpu->regs[ins.r.rs2]) ? 1 : 0; |
| 1895 | break; |
| 1896 | case FUNCT3_XOR: /* xor. */ |
| 1897 | cpu->regs[ins.r.rd] = |
| 1898 | cpu->regs[ins.r.rs1] ^ cpu->regs[ins.r.rs2]; |
| 1899 | break; |
| 1900 | case FUNCT3_SRL: /* srl. */ |
| 1901 | cpu->regs[ins.r.rd] = |
| 1902 | cpu->regs[ins.r.rs1] >> (cpu->regs[ins.r.rs2] & shamt_mask); |
| 1903 | break; |
| 1904 | case FUNCT3_OR: /* or. */ |
| 1905 | cpu->regs[ins.r.rd] = |
| 1906 | cpu->regs[ins.r.rs1] | cpu->regs[ins.r.rs2]; |
| 1907 | break; |
| 1908 | case FUNCT3_AND: /* and. */ |
| 1909 | cpu->regs[ins.r.rd] = |
| 1910 | cpu->regs[ins.r.rs1] & cpu->regs[ins.r.rs2]; |
| 1911 | break; |
| 1912 | } |
| 1913 | break; |
| 1914 | } |
| 1915 | |
| 1916 | case FUNCT7_SUB: |
| 1917 | switch (ins.r.funct3) { |
| 1918 | case FUNCT3_ADD: /* sub. */ |
| 1919 | cpu->regs[ins.r.rd] = |
| 1920 | cpu->regs[ins.r.rs1] - cpu->regs[ins.r.rs2]; |
| 1921 | break; |
| 1922 | case FUNCT3_SRL: /* sra. */ |
| 1923 | cpu->regs[ins.r.rd] = (u64)((i64)cpu->regs[ins.r.rs1] >> |
| 1924 | (cpu->regs[ins.r.rs2] & 0x3F)); |
| 1925 | break; |
| 1926 | } |
| 1927 | break; |
| 1928 | |
| 1929 | case FUNCT7_MUL: |
| 1930 | switch (ins.r.funct3) { |
| 1931 | case FUNCT3_ADD: /* mul. */ |
| 1932 | cpu->regs[ins.r.rd] = |
| 1933 | cpu->regs[ins.r.rs1] * cpu->regs[ins.r.rs2]; |
| 1934 | break; |
| 1935 | case FUNCT3_XOR: /* div. */ |
| 1936 | if (cpu->regs[ins.r.rs2] != 0) { |
| 1937 | cpu->regs[ins.r.rd] = (u64)((i64)cpu->regs[ins.r.rs1] / |
| 1938 | (i64)cpu->regs[ins.r.rs2]); |
| 1939 | } else { |
| 1940 | cpu->regs[ins.r.rd] = (u64)-1; /* Division by zero. */ |
| 1941 | } |
| 1942 | break; |
| 1943 | case FUNCT3_SRL: /* divu. */ |
| 1944 | if (cpu->regs[ins.r.rs2] != 0) { |
| 1945 | cpu->regs[ins.r.rd] = |
| 1946 | cpu->regs[ins.r.rs1] / cpu->regs[ins.r.rs2]; |
| 1947 | } else { |
| 1948 | cpu->regs[ins.r.rd] = (u64)-1; /* Division by zero. */ |
| 1949 | } |
| 1950 | break; |
| 1951 | case FUNCT3_OR: /* rem. */ |
| 1952 | if (cpu->regs[ins.r.rs2] != 0) { |
| 1953 | cpu->regs[ins.r.rd] = (u64)((i64)cpu->regs[ins.r.rs1] % |
| 1954 | (i64)cpu->regs[ins.r.rs2]); |
| 1955 | } else { |
| 1956 | cpu->regs[ins.r.rd] = cpu->regs[ins.r.rs1]; |
| 1957 | } |
| 1958 | break; |
| 1959 | case FUNCT3_AND: /* remu. */ |
| 1960 | if (cpu->regs[ins.r.rs2] != 0) { |
| 1961 | cpu->regs[ins.r.rd] = |
| 1962 | cpu->regs[ins.r.rs1] % cpu->regs[ins.r.rs2]; |
| 1963 | } else { |
| 1964 | cpu->regs[ins.r.rd] = cpu->regs[ins.r.rs1]; |
| 1965 | } |
| 1966 | break; |
| 1967 | } |
| 1968 | break; |
| 1969 | } |
| 1970 | break; |
| 1971 | } |
| 1972 | |
| 1973 | case OP_OP_32: { |
| 1974 | /* RV64I: 32-bit register-register operations (ADDW, SUBW, SLLW, SRLW, |
| 1975 | * SRAW, MULW, DIVW, DIVUW, REMW, REMUW). These operate on the lower 32 |
| 1976 | * bits and sign-extend the result to 64 bits. */ |
| 1977 | if (ins.r.rd == ZERO) |
| 1978 | break; |
| 1979 | |
| 1980 | cpu->modified = (reg_t)ins.r.rd; |
| 1981 | u32 rs1_32 = (u32)cpu->regs[ins.r.rs1]; |
| 1982 | u32 rs2_32 = (u32)cpu->regs[ins.r.rs2]; |
| 1983 | |
| 1984 | switch (ins.r.funct7) { |
| 1985 | case FUNCT7_NORMAL: |
| 1986 | switch (ins.r.funct3) { |
| 1987 | case FUNCT3_ADD: { /* addw. */ |
| 1988 | i32 result = (i32)(rs1_32 + rs2_32); |
| 1989 | cpu->regs[ins.r.rd] = (u64)(i64)result; |
| 1990 | break; |
| 1991 | } |
| 1992 | case FUNCT3_SLL: { /* sllw. */ |
| 1993 | i32 result = (i32)(rs1_32 << (rs2_32 & 0x1F)); |
| 1994 | cpu->regs[ins.r.rd] = (u64)(i64)result; |
| 1995 | break; |
| 1996 | } |
| 1997 | case FUNCT3_SRL: { /* srlw. */ |
| 1998 | i32 result = (i32)(rs1_32 >> (rs2_32 & 0x1F)); |
| 1999 | cpu->regs[ins.r.rd] = (u64)(i64)result; |
| 2000 | break; |
| 2001 | } |
| 2002 | } |
| 2003 | break; |
| 2004 | |
| 2005 | case FUNCT7_SUB: |
| 2006 | switch (ins.r.funct3) { |
| 2007 | case FUNCT3_ADD: { /* subw. */ |
| 2008 | i32 result = (i32)(rs1_32 - rs2_32); |
| 2009 | cpu->regs[ins.r.rd] = (u64)(i64)result; |
| 2010 | break; |
| 2011 | } |
| 2012 | case FUNCT3_SRL: { /* sraw. */ |
| 2013 | i32 result = (i32)rs1_32 >> (rs2_32 & 0x1F); |
| 2014 | cpu->regs[ins.r.rd] = (u64)(i64)result; |
| 2015 | break; |
| 2016 | } |
| 2017 | } |
| 2018 | break; |
| 2019 | |
| 2020 | case FUNCT7_MUL: |
| 2021 | switch (ins.r.funct3) { |
| 2022 | case FUNCT3_ADD: { /* mulw. */ |
| 2023 | i32 result = (i32)(rs1_32 * rs2_32); |
| 2024 | cpu->regs[ins.r.rd] = (u64)(i64)result; |
| 2025 | break; |
| 2026 | } |
| 2027 | case FUNCT3_XOR: { /* divw. */ |
| 2028 | if (rs2_32 != 0) { |
| 2029 | i32 result = (i32)rs1_32 / (i32)rs2_32; |
| 2030 | cpu->regs[ins.r.rd] = (u64)(i64)result; |
| 2031 | } else { |
| 2032 | cpu->regs[ins.r.rd] = (u64)(i64)(i32)-1; |
| 2033 | } |
| 2034 | break; |
| 2035 | } |
| 2036 | case FUNCT3_SRL: { /* divuw. */ |
| 2037 | if (rs2_32 != 0) { |
| 2038 | i32 result = (i32)(rs1_32 / rs2_32); |
| 2039 | cpu->regs[ins.r.rd] = (u64)(i64)result; |
| 2040 | } else { |
| 2041 | cpu->regs[ins.r.rd] = (u64)(i64)(i32)-1; |
| 2042 | } |
| 2043 | break; |
| 2044 | } |
| 2045 | case FUNCT3_OR: { /* remw. */ |
| 2046 | if (rs2_32 != 0) { |
| 2047 | i32 result = (i32)rs1_32 % (i32)rs2_32; |
| 2048 | cpu->regs[ins.r.rd] = (u64)(i64)result; |
| 2049 | } else { |
| 2050 | cpu->regs[ins.r.rd] = (u64)(i64)(i32)rs1_32; |
| 2051 | } |
| 2052 | break; |
| 2053 | } |
| 2054 | case FUNCT3_AND: { /* remuw. */ |
| 2055 | if (rs2_32 != 0) { |
| 2056 | i32 result = (i32)(rs1_32 % rs2_32); |
| 2057 | cpu->regs[ins.r.rd] = (u64)(i64)result; |
| 2058 | } else { |
| 2059 | cpu->regs[ins.r.rd] = (u64)(i64)(i32)rs1_32; |
| 2060 | } |
| 2061 | break; |
| 2062 | } |
| 2063 | } |
| 2064 | break; |
| 2065 | } |
| 2066 | break; |
| 2067 | } |
| 2068 | |
| 2069 | case OP_SYSTEM: { |
| 2070 | u32 funct12 = ins.i.imm_11_0; |
| 2071 | |
| 2072 | if (funct12 == 0) { |
| 2073 | u32 syscall_num = (u32)cpu->regs[A7]; |
| 2074 | |
| 2075 | switch (syscall_num) { |
| 2076 | case 64: { /* write. */ |
| 2077 | int guest_fd = (int)cpu->regs[A0]; |
| 2078 | u64 addr = cpu->regs[A1]; |
| 2079 | u64 count = cpu->regs[A2]; |
| 2080 | |
| 2081 | if (addr + count > g_opts.memory_size || |
| 2082 | addr > (u64)g_opts.memory_size) { |
| 2083 | printf( |
| 2084 | "sys_write out of bounds: addr=%016llx len=%llu\n", |
| 2085 | (unsigned long long)addr, |
| 2086 | (unsigned long long)count |
| 2087 | ); |
| 2088 | cpu->running = false; |
| 2089 | emit_fault_diagnostics(cpu, executed_pc); |
| 2090 | break; |
| 2091 | } |
| 2092 | ssize_t written = 0; |
| 2093 | int host_fd = guest_fd_table_get(guest_fd); |
| 2094 | |
| 2095 | if (host_fd >= 0 && count > 0) { |
| 2096 | written = write(host_fd, &memory[(u32)addr], (u32)count); |
| 2097 | if (written < 0) { |
| 2098 | written = 0; |
| 2099 | } |
| 2100 | } |
| 2101 | cpu->regs[A0] = (u64)written; |
| 2102 | break; |
| 2103 | } |
| 2104 | case 63: { /* read. */ |
| 2105 | int guest_fd = (int)cpu->regs[A0]; |
| 2106 | u64 addr = cpu->regs[A1]; |
| 2107 | u64 count = cpu->regs[A2]; |
| 2108 | |
| 2109 | if (addr + count > g_opts.memory_size || |
| 2110 | addr > (u64)g_opts.memory_size) { |
| 2111 | printf( |
| 2112 | "sys_read out of bounds: addr=%016llx len=%llu\n", |
| 2113 | (unsigned long long)addr, |
| 2114 | (unsigned long long)count |
| 2115 | ); |
| 2116 | cpu->running = false; |
| 2117 | emit_fault_diagnostics(cpu, executed_pc); |
| 2118 | break; |
| 2119 | } |
| 2120 | ssize_t read_bytes = 0; |
| 2121 | int host_fd = guest_fd_table_get(guest_fd); |
| 2122 | |
| 2123 | if (host_fd >= 0 && count > 0) { |
| 2124 | read_bytes = read(host_fd, &memory[(u32)addr], (u32)count); |
| 2125 | if (read_bytes < 0) { |
| 2126 | read_bytes = 0; |
| 2127 | } |
| 2128 | } |
| 2129 | cpu->regs[A0] = (u64)read_bytes; |
| 2130 | break; |
| 2131 | } |
| 2132 | case 93: { /* exit. */ |
| 2133 | cpu->running = false; |
| 2134 | break; |
| 2135 | } |
| 2136 | case 56: { /* openat. */ |
| 2137 | u64 pathname_addr = cpu->regs[A1]; |
| 2138 | i32 flags = (i32)cpu->regs[A2]; |
| 2139 | if (pathname_addr > (u64)g_opts.memory_size) { |
| 2140 | cpu->regs[A0] = (u64)(i64)(i32)-1; |
| 2141 | break; |
| 2142 | } |
| 2143 | cpu->regs[A0] = |
| 2144 | (u64)(i64)(i32)ecall_openat((u32)pathname_addr, flags); |
| 2145 | break; |
| 2146 | } |
| 2147 | case 57: { /* close. */ |
| 2148 | i32 guest_fd = (i32)cpu->regs[A0]; |
| 2149 | cpu->regs[A0] = (u64)(i64)ecall_close(guest_fd); |
| 2150 | break; |
| 2151 | } |
| 2152 | default: |
| 2153 | cpu->regs[A0] = (u32)syscall_num; |
| 2154 | break; |
| 2155 | } |
| 2156 | } else if (funct12 == 1) { |
| 2157 | /* Look up source location for this EBREAK. PC in the debug |
| 2158 | * file is relative to program start, so subtract base. */ |
| 2159 | u32 relative_pc = executed_pc - program_base; |
| 2160 | struct debug_entry *entry = debug_lookup(relative_pc); |
| 2161 | |
| 2162 | printf("\n%sRuntime error (EBREAK)%s", COLOR_BOLD_RED, COLOR_RESET); |
| 2163 | if (entry) { |
| 2164 | u32 line = line_from_offset(entry->file, entry->offset); |
| 2165 | printf( |
| 2166 | " at %s%s:%d%s", COLOR_CYAN, entry->file, line, COLOR_RESET |
| 2167 | ); |
| 2168 | } |
| 2169 | printf("\n"); |
| 2170 | |
| 2171 | cpu->running = false; |
| 2172 | cpu->regs[A0] = EBREAK_EXIT_CODE; |
| 2173 | cpu->ebreak = true; |
| 2174 | emit_fault_diagnostics(cpu, executed_pc); |
| 2175 | cpu->faulted = false; |
| 2176 | } else { |
| 2177 | printf( |
| 2178 | "\n%sUnknown system instruction (imm=%08x)%s\n", |
| 2179 | COLOR_BOLD_RED, |
| 2180 | funct12, |
| 2181 | COLOR_RESET |
| 2182 | ); |
| 2183 | cpu->running = false; |
| 2184 | emit_fault_diagnostics(cpu, executed_pc); |
| 2185 | } |
| 2186 | break; |
| 2187 | } |
| 2188 | |
| 2189 | case OP_FENCE: |
| 2190 | /* Memory barriers are not implemented. */ |
| 2191 | break; |
| 2192 | |
| 2193 | default: |
| 2194 | printf("Unknown opcode %02x at PC=%08x\n", opcode, cpu->pc); |
| 2195 | cpu->running = false; |
| 2196 | emit_fault_diagnostics(cpu, executed_pc); |
| 2197 | break; |
| 2198 | } |
| 2199 | /* Register x0 is hardwired to zero. */ |
| 2200 | cpu->regs[ZERO] = 0; |
| 2201 | cpu->pc = pc_next; |
| 2202 | |
| 2203 | if (cpu->running) { |
| 2204 | validate_stack_register(cpu, SP, "SP", executed_pc, false); |
| 2205 | if (cpu->running) |
| 2206 | validate_stack_register(cpu, FP, "FP", executed_pc, true); |
| 2207 | } |
| 2208 | |
| 2209 | /* Track last executed PC for trace mode. */ |
| 2210 | if (headless && g_opts.trace_print_instructions) |
| 2211 | last_executed_pc = executed_pc; |
| 2212 | } |
| 2213 | |
| 2214 | /* Render the instructions column. */ |
| 2215 | static void ui_render_instructions( |
| 2216 | struct cpu *cpu, int col, int width, int height |
| 2217 | ) { |
| 2218 | int row = 1; |
| 2219 | |
| 2220 | u32 program_start_idx = program_base / INSTR_SIZE; |
| 2221 | u32 program_end_idx = program_start_idx + cpu->programsize; |
| 2222 | |
| 2223 | /* Calculate PC index in program. */ |
| 2224 | u32 pc_idx = cpu->pc / sizeof(instr_t); |
| 2225 | if (pc_idx < program_start_idx || pc_idx >= program_end_idx) |
| 2226 | pc_idx = program_start_idx; |
| 2227 | |
| 2228 | /* Calculate first instruction to display, centering PC if possible. */ |
| 2229 | i32 progstart = (i32)pc_idx - height / 2; |
| 2230 | i32 min_start = (i32)program_start_idx; |
| 2231 | i32 max_start = (i32)program_end_idx - height; |
| 2232 | |
| 2233 | if (max_start < min_start) |
| 2234 | max_start = min_start; |
| 2235 | if (progstart < min_start) |
| 2236 | progstart = min_start; |
| 2237 | if (progstart > max_start) |
| 2238 | progstart = max_start; |
| 2239 | |
| 2240 | printf(TTY_GOTO_RC, row++, col); |
| 2241 | printf(" INSTRUCTIONS"); |
| 2242 | |
| 2243 | for (int i = 0; i < height; i++) { |
| 2244 | u32 idx = (u32)progstart + i; |
| 2245 | if (idx >= program_end_idx) |
| 2246 | break; |
| 2247 | |
| 2248 | printf(TTY_GOTO_RC, row + i + 1, col); |
| 2249 | |
| 2250 | char istr[MAX_INSTR_STR_LEN] = { 0 }; |
| 2251 | int len = sprint_instr(cpu->program[idx], istr, true); |
| 2252 | |
| 2253 | if (idx == pc_idx) { /* Highlight current instruction. */ |
| 2254 | printf("%s>%s %04x: ", COLOR_GREEN, COLOR_RESET, idx * INSTR_SIZE); |
| 2255 | } else { |
| 2256 | printf(" %s%04x:%s ", COLOR_GREY, idx * INSTR_SIZE, COLOR_RESET); |
| 2257 | } |
| 2258 | printf("%s", istr); |
| 2259 | printf("%-*s", width - len - 8, ""); |
| 2260 | } |
| 2261 | } |
| 2262 | |
| 2263 | /* Render the registers column. */ |
| 2264 | static void ui_render_registers( |
| 2265 | struct cpu *cpu, enum display display, int col, int height |
| 2266 | ) { |
| 2267 | int row = 1; |
| 2268 | |
| 2269 | printf(TTY_GOTO_RC, row++, col); |
| 2270 | printf("REGISTERS"); |
| 2271 | |
| 2272 | int reg_count = |
| 2273 | sizeof(registers_displayed) / sizeof(registers_displayed[0]); |
| 2274 | if (reg_count > height) |
| 2275 | reg_count = height; |
| 2276 | |
| 2277 | for (int i = 0; i < reg_count; i++) { |
| 2278 | printf(TTY_GOTO_RC, row + i + 1, col); |
| 2279 | |
| 2280 | reg_t r = registers_displayed[i]; |
| 2281 | const char *reg_color = |
| 2282 | (r == cpu->modified) ? COLOR_BOLD_BLUE : COLOR_BLUE; |
| 2283 | u64 reg_value = cpu->regs[r]; |
| 2284 | bool is_stack_addr = |
| 2285 | reg_value <= (u64)UINT32_MAX && stack_contains((u32)reg_value); |
| 2286 | |
| 2287 | /* Always show registers that contain stack addresses in hex. */ |
| 2288 | printf("%s%-2s%s = ", COLOR_GREEN, reg_names[r], COLOR_RESET); |
| 2289 | if (display == DISPLAY_HEX || is_stack_addr) { |
| 2290 | printf("%s0x%08x%s", reg_color, (u32)reg_value, COLOR_RESET); |
| 2291 | } else { |
| 2292 | printf("%s%-10d%s", reg_color, (i32)reg_value, COLOR_RESET); |
| 2293 | } |
| 2294 | } |
| 2295 | } |
| 2296 | |
| 2297 | /* Render the stack column. */ |
| 2298 | static void ui_render_stack( |
| 2299 | struct cpu *cpu, enum display display, int col, int height |
| 2300 | ) { |
| 2301 | int row = 1; |
| 2302 | |
| 2303 | printf(TTY_GOTO_RC, row++, col); |
| 2304 | printf(" STACK FRAME"); |
| 2305 | |
| 2306 | assert(cpu->regs[SP] <= memory_top() && cpu->regs[FP] <= memory_top()); |
| 2307 | |
| 2308 | u32 fp = (u32)cpu->regs[FP]; |
| 2309 | u32 sp = (u32)cpu->regs[SP]; |
| 2310 | u32 rows = (u32)height; |
| 2311 | if (rows > STACK_DISPLAY_WORDS) |
| 2312 | rows = STACK_DISPLAY_WORDS; |
| 2313 | if (rows == 0) |
| 2314 | return; |
| 2315 | |
| 2316 | u32 top = stack_usable_top(); |
| 2317 | u32 bottom = stack_usable_bottom(); |
| 2318 | if (sp > top) |
| 2319 | sp = top; |
| 2320 | if (fp > top) |
| 2321 | fp = top; |
| 2322 | |
| 2323 | u32 used_bytes = (top >= sp) ? (top - sp) : 0; |
| 2324 | u32 total_words = (used_bytes / WORD_SIZE) + 1; |
| 2325 | u32 frame_words = total_words; |
| 2326 | if (frame_words > rows) |
| 2327 | frame_words = rows; |
| 2328 | if (frame_words == 0) |
| 2329 | return; |
| 2330 | |
| 2331 | u32 start; |
| 2332 | if (frame_words == total_words) { |
| 2333 | start = top; |
| 2334 | } else { |
| 2335 | start = sp + (frame_words - 1) * WORD_SIZE; |
| 2336 | if (start > top) |
| 2337 | start = top; |
| 2338 | } |
| 2339 | |
| 2340 | if (start < bottom) |
| 2341 | start = bottom; |
| 2342 | |
| 2343 | u32 addr = start; |
| 2344 | i32 offset = (i32)(start - sp); |
| 2345 | |
| 2346 | for (u32 i = 0; i < frame_words; i++) { |
| 2347 | if (addr < bottom) |
| 2348 | break; |
| 2349 | |
| 2350 | assert(addr <= memory_top()); |
| 2351 | printf(TTY_GOTO_RC, row + i + 1, col); |
| 2352 | |
| 2353 | /* Mark SP and FP positions. */ |
| 2354 | const char *marker = " "; |
| 2355 | |
| 2356 | if (addr == sp) { |
| 2357 | marker = "sp"; |
| 2358 | } else if (addr == fp) { |
| 2359 | marker = "fp"; |
| 2360 | } |
| 2361 | u32 word = memory_load_u32(addr); |
| 2362 | |
| 2363 | char offset_buf[6]; |
| 2364 | if (addr == sp) { |
| 2365 | memcpy(offset_buf, " ", 5); |
| 2366 | } else { |
| 2367 | snprintf(offset_buf, sizeof(offset_buf), "%+4d", offset); |
| 2368 | } |
| 2369 | |
| 2370 | printf( |
| 2371 | "%s%s %s%s%s %08x: ", |
| 2372 | COLOR_GREEN, |
| 2373 | marker, |
| 2374 | COLOR_GREY, |
| 2375 | offset_buf, |
| 2376 | COLOR_RESET, |
| 2377 | addr |
| 2378 | ); |
| 2379 | bool is_stack_addr = stack_contains(word); |
| 2380 | |
| 2381 | if (display == DISPLAY_HEX || is_stack_addr) { |
| 2382 | printf("%s0x%08x%s", COLOR_BLUE, word, COLOR_RESET); |
| 2383 | } else { |
| 2384 | printf("%s%-10d%s", COLOR_BLUE, (i32)word, COLOR_RESET); |
| 2385 | } |
| 2386 | if (addr < WORD_SIZE) |
| 2387 | break; |
| 2388 | |
| 2389 | addr -= WORD_SIZE; |
| 2390 | offset -= WORD_SIZE; |
| 2391 | } |
| 2392 | } |
| 2393 | |
| 2394 | /* Render the full debugger TUI. */ |
| 2395 | static void ui_render(struct cpu *cpu, enum display display) { |
| 2396 | printf(TTY_CLEAR); |
| 2397 | |
| 2398 | struct termsize tsize = termsize(); |
| 2399 | |
| 2400 | /* Enforce a minimum display size. */ |
| 2401 | if (tsize.cols < 60) |
| 2402 | tsize.cols = 60; |
| 2403 | if (tsize.rows < 15) |
| 2404 | tsize.rows = 15; |
| 2405 | |
| 2406 | /* Column layout: 40% instructions, 20% registers, rest for stack. */ |
| 2407 | int instr_width = (tsize.cols * 2) / 5; |
| 2408 | int reg_width = tsize.cols / 5; |
| 2409 | |
| 2410 | int instr_col = 1; |
| 2411 | int reg_col = instr_col + instr_width + 2; |
| 2412 | int stack_col = reg_col + reg_width + 2; |
| 2413 | |
| 2414 | int display_height = tsize.rows - FOOTER_HEIGHT - HEADER_HEIGHT; |
| 2415 | if (display_height > MAX_INSTR_DISPLAY) |
| 2416 | display_height = MAX_INSTR_DISPLAY; |
| 2417 | if (display_height <= 0) |
| 2418 | display_height = 1; |
| 2419 | |
| 2420 | ui_render_instructions(cpu, instr_col, instr_width, display_height); |
| 2421 | ui_render_registers(cpu, display, reg_col, display_height); |
| 2422 | ui_render_stack(cpu, display, stack_col, display_height); |
| 2423 | |
| 2424 | printf(TTY_GOTO_RC, display_height + FOOTER_HEIGHT, 1); |
| 2425 | printf( |
| 2426 | "%sPress `j` to step forward, `k` to step backward, `q` to quit,\n" |
| 2427 | "`d` to toggle decimal display, `r` to reset program.%s ", |
| 2428 | COLOR_GREY, |
| 2429 | COLOR_RESET |
| 2430 | ); |
| 2431 | } |
| 2432 | |
| 2433 | /* Set up the terminal for interactive mode, saving the original settings. */ |
| 2434 | static void term_init(struct termios *oldterm) { |
| 2435 | struct termios term; |
| 2436 | |
| 2437 | tcgetattr(STDIN_FILENO, oldterm); |
| 2438 | term = *oldterm; |
| 2439 | term.c_lflag &= ~(ICANON | ECHO); |
| 2440 | tcsetattr(STDIN_FILENO, TCSANOW, &term); |
| 2441 | } |
| 2442 | |
| 2443 | /* Restore terminal settings. */ |
| 2444 | static void term_restore(struct termios *old) { |
| 2445 | tcsetattr(STDIN_FILENO, TCSANOW, old); |
| 2446 | } |
| 2447 | |
| 2448 | int main(int argc, char *argv[]) { |
| 2449 | struct cpu cpu; |
| 2450 | enum display display = DISPLAY_DEC; |
| 2451 | struct cli_config cli = { 0 }; |
| 2452 | |
| 2453 | if (!parse_cli_args(argc, argv, &cli)) |
| 2454 | return 1; |
| 2455 | |
| 2456 | bool headless = cli.headless; |
| 2457 | |
| 2458 | g_opts.trace_headless = headless; |
| 2459 | |
| 2460 | cpu_init(&cpu); |
| 2461 | program_init(&cpu, cli.program_path); |
| 2462 | int prog_argc = argc - cli.arg_index; |
| 2463 | char **prog_argv = &argv[cli.arg_index]; |
| 2464 | prepare_env(&cpu, prog_argc, prog_argv); |
| 2465 | |
| 2466 | if (headless) { |
| 2467 | u64 max_steps = g_opts.headless_max_steps; |
| 2468 | u64 steps = 0; |
| 2469 | |
| 2470 | /* Try to initialise the JIT for headless mode. Falls back to the |
| 2471 | * interpreter automatically when JIT is disabled, unavailable, |
| 2472 | * or when the code cache fills up. */ |
| 2473 | static struct jit_state jit; |
| 2474 | bool use_jit = false; |
| 2475 | |
| 2476 | if (!g_opts.jit_disabled && !g_opts.trace_enabled && |
| 2477 | !g_opts.trace_print_instructions && !g_opts.watch_enabled) { |
| 2478 | use_jit = jit_init(&jit); |
| 2479 | } |
| 2480 | |
| 2481 | if (use_jit) { |
| 2482 | /* ---- JIT execution loop ---- */ |
| 2483 | while (cpu.running && steps < max_steps) { |
| 2484 | struct jit_block *block = jit_get_block( |
| 2485 | &jit, cpu.pc, memory, program_base, program_bytes |
| 2486 | ); |
| 2487 | if (!block) { |
| 2488 | /* Cache full or compilation error -- fall back to |
| 2489 | * interpreter for remainder. */ |
| 2490 | while (cpu.running && steps++ < max_steps) { |
| 2491 | cpu_execute(&cpu, display, true); |
| 2492 | } |
| 2493 | break; |
| 2494 | } |
| 2495 | u32 next_pc = 0; |
| 2496 | int exit_reason = |
| 2497 | jit_exec_block(block, cpu.regs, memory, &next_pc); |
| 2498 | steps += block->insn_count; |
| 2499 | jit.blocks_executed++; |
| 2500 | jit.insns_executed += block->insn_count; |
| 2501 | |
| 2502 | switch (exit_reason) { |
| 2503 | case JIT_EXIT_BRANCH: |
| 2504 | case JIT_EXIT_CHAIN: |
| 2505 | cpu.pc = next_pc; |
| 2506 | break; |
| 2507 | |
| 2508 | case JIT_EXIT_RET: |
| 2509 | cpu.running = false; |
| 2510 | break; |
| 2511 | |
| 2512 | default: |
| 2513 | /* ECALL, EBREAK, FAULT -- interpreter handles it. |
| 2514 | * The instruction is already counted above, |
| 2515 | * so don't increment steps again. */ |
| 2516 | cpu.pc = next_pc; |
| 2517 | cpu_execute(&cpu, display, true); |
| 2518 | break; |
| 2519 | } |
| 2520 | } |
| 2521 | jit_destroy(&jit); |
| 2522 | } else { |
| 2523 | /* ---- Interpreter-only loop ---- */ |
| 2524 | while (cpu.running && steps++ < max_steps) { |
| 2525 | cpu_execute(&cpu, display, true); |
| 2526 | } |
| 2527 | } |
| 2528 | |
| 2529 | if (cpu.running) { |
| 2530 | fprintf( |
| 2531 | stderr, |
| 2532 | "program did not terminate within %zu steps\n", |
| 2533 | (size_t)max_steps |
| 2534 | ); |
| 2535 | return -1; |
| 2536 | } |
| 2537 | if (cpu.faulted) { |
| 2538 | fprintf(stderr, "program terminated due to runtime fault\n"); |
| 2539 | return -1; |
| 2540 | } |
| 2541 | if (g_opts.count_instructions) { |
| 2542 | fprintf( |
| 2543 | stderr, |
| 2544 | "Processed %llu instructions\n", |
| 2545 | (unsigned long long)steps |
| 2546 | ); |
| 2547 | } |
| 2548 | return (int)cpu.regs[A0]; |
| 2549 | } |
| 2550 | struct termios oldterm; |
| 2551 | term_init(&oldterm); |
| 2552 | snapshot_init(&cpu); |
| 2553 | |
| 2554 | for (;;) { |
| 2555 | if (cpu.running) |
| 2556 | ui_render(&cpu, display); |
| 2557 | |
| 2558 | int ch = getchar(); |
| 2559 | |
| 2560 | if (ch == 'q' || ch == 'Q') { |
| 2561 | printf("\n"); |
| 2562 | break; |
| 2563 | } else if (ch == 'd' || ch == 'D') { /* Toggle display mode. */ |
| 2564 | display = (display == DISPLAY_HEX) ? DISPLAY_DEC : DISPLAY_HEX; |
| 2565 | } else if (ch == 'r' || ch == 'R') { /* Reset program and state. */ |
| 2566 | cpu_reset(&cpu); |
| 2567 | snapshot_init(&cpu); |
| 2568 | } else if (ch == 'g' || ch == 'G') { /* Toggle stack guard. */ |
| 2569 | toggle_stack_guard(&cpu); |
| 2570 | } else if (ch == 'j' && cpu.running) { /* Step forward. */ |
| 2571 | cpu_execute(&cpu, display, false); |
| 2572 | snapshot_save(&cpu); |
| 2573 | } else if (ch == 'k' && cpu.pc > 0) { /* Step backward. */ |
| 2574 | bool restored = snapshot_restore(&cpu); |
| 2575 | |
| 2576 | if (restored) { |
| 2577 | cpu.running = true; |
| 2578 | } else { |
| 2579 | printf( |
| 2580 | "\n%sNo more history to go back to.%s\n", |
| 2581 | COLOR_BOLD_RED, |
| 2582 | COLOR_RESET |
| 2583 | ); |
| 2584 | } |
| 2585 | } |
| 2586 | } |
| 2587 | term_restore(&oldterm); |
| 2588 | |
| 2589 | return 0; |
| 2590 | } |