| 1 | /*> interp.c <*/ |
| 2 | /* Simulator for the MIPS architecture. |
| 3 | |
| 4 | This file is part of the MIPS sim |
| 5 | |
| 6 | THIS SOFTWARE IS NOT COPYRIGHTED |
| 7 | |
| 8 | Cygnus offers the following for use in the public domain. Cygnus |
| 9 | makes no warranty with regard to the software or it's performance |
| 10 | and the user accepts the software "AS IS" with all faults. |
| 11 | |
| 12 | CYGNUS DISCLAIMS ANY WARRANTIES, EXPRESS OR IMPLIED, WITH REGARD TO |
| 13 | THIS SOFTWARE INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF |
| 14 | MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. |
| 15 | |
| 16 | $Revision$ |
| 17 | $Author$ |
| 18 | $Date$ |
| 19 | |
| 20 | NOTEs: |
| 21 | |
| 22 | We only need to take account of the target endianness when moving data |
| 23 | between the simulator and the host. We do not need to worry about the |
| 24 | endianness of the host, since this sim code and GDB are executing in |
| 25 | the same process. |
| 26 | |
| 27 | The IDT monitor (found on the VR4300 board), seems to lie about |
| 28 | register contents. It seems to treat the registers as sign-extended |
| 29 | 32-bit values. This cause *REAL* problems when single-stepping 64-bit |
| 30 | code on the hardware. |
| 31 | |
| 32 | */ |
| 33 | |
| 34 | /* The TRACE and PROFILE manifests enable the provision of extra |
| 35 | features. If they are not defined then a simpler (quicker) |
| 36 | simulator is constructed without the required run-time checks, |
| 37 | etc. */ |
| 38 | #if 1 /* 0 to allow user build selection, 1 to force inclusion */ |
| 39 | #define TRACE (1) |
| 40 | #define PROFILE (1) |
| 41 | #endif |
| 42 | |
| 43 | #include "config.h" |
| 44 | |
| 45 | #include <stdio.h> |
| 46 | #include <stdarg.h> |
| 47 | #include <ansidecl.h> |
| 48 | #include <signal.h> |
| 49 | #include <ctype.h> |
| 50 | #include <limits.h> |
| 51 | #include <math.h> |
| 52 | #ifdef HAVE_STDLIB_H |
| 53 | #include <stdlib.h> |
| 54 | #endif |
| 55 | #ifdef HAVE_STRING_H |
| 56 | #include <string.h> |
| 57 | #else |
| 58 | #ifdef HAVE_STRINGS_H |
| 59 | #include <strings.h> |
| 60 | #endif |
| 61 | #endif |
| 62 | |
| 63 | #include "getopt.h" |
| 64 | #include "libiberty.h" |
| 65 | |
| 66 | #include "remote-sim.h" /* GDB simulator interface */ |
| 67 | #include "callback.h" /* GDB simulator callback interface */ |
| 68 | |
| 69 | #include "support.h" /* internal support manifests */ |
| 70 | |
| 71 | #include "sysdep.h" |
| 72 | |
| 73 | #ifndef SIGBUS |
| 74 | #define SIGBUS SIGSEGV |
| 75 | #endif |
| 76 | |
| 77 | /* Get the simulator engine description, without including the code: */ |
| 78 | #define SIM_MANIFESTS |
| 79 | #include "engine.c" |
| 80 | #undef SIM_MANIFESTS |
| 81 | |
| 82 | /* This variable holds the GDB view of the target endianness: */ |
| 83 | extern int target_byte_order; |
| 84 | |
| 85 | /* The following reserved instruction value is used when a simulator |
| 86 | trap is required. NOTE: Care must be taken, since this value may be |
| 87 | used in later revisions of the MIPS ISA. */ |
| 88 | #define RSVD_INSTRUCTION (0x7C000000) |
| 89 | #define RSVD_INSTRUCTION_AMASK (0x03FFFFFF) |
| 90 | |
| 91 | /* NOTE: These numbers depend on the processor architecture being |
| 92 | simulated: */ |
| 93 | #define Interrupt (0) |
| 94 | #define TLBModification (1) |
| 95 | #define TLBLoad (2) |
| 96 | #define TLBStore (3) |
| 97 | #define AddressLoad (4) |
| 98 | #define AddressStore (5) |
| 99 | #define InstructionFetch (6) |
| 100 | #define DataReference (7) |
| 101 | #define SystemCall (8) |
| 102 | #define BreakPoint (9) |
| 103 | #define ReservedInstruction (10) |
| 104 | #define CoProcessorUnusable (11) |
| 105 | #define IntegerOverflow (12) /* Arithmetic overflow (IDT monitor raises SIGFPE) */ |
| 106 | #define Trap (13) |
| 107 | #define FPE (15) |
| 108 | #define Watch (23) |
| 109 | |
| 110 | /* The following exception code is actually private to the simulator |
| 111 | world. It is *NOT* a processor feature, and is used to signal |
| 112 | run-time errors in the simulator. */ |
| 113 | #define SimulatorFault (0xFFFFFFFF) |
| 114 | |
| 115 | /* The following are generic to all versions of the MIPS architecture |
| 116 | to date: */ |
| 117 | /* Memory Access Types (for CCA): */ |
| 118 | #define Uncached (0) |
| 119 | #define CachedNoncoherent (1) |
| 120 | #define CachedCoherent (2) |
| 121 | #define Cached (3) |
| 122 | |
| 123 | #define isINSTRUCTION (1 == 0) /* FALSE */ |
| 124 | #define isDATA (1 == 1) /* TRUE */ |
| 125 | |
| 126 | #define isLOAD (1 == 0) /* FALSE */ |
| 127 | #define isSTORE (1 == 1) /* TRUE */ |
| 128 | |
| 129 | #define isREAL (1 == 0) /* FALSE */ |
| 130 | #define isRAW (1 == 1) /* TRUE */ |
| 131 | |
| 132 | #define isTARGET (1 == 0) /* FALSE */ |
| 133 | #define isHOST (1 == 1) /* TRUE */ |
| 134 | |
| 135 | /* The "AccessLength" specifications for Loads and Stores. NOTE: This |
| 136 | is the number of bytes minus 1. */ |
| 137 | #define AccessLength_BYTE (0) |
| 138 | #define AccessLength_HALFWORD (1) |
| 139 | #define AccessLength_TRIPLEBYTE (2) |
| 140 | #define AccessLength_WORD (3) |
| 141 | #define AccessLength_QUINTIBYTE (4) |
| 142 | #define AccessLength_SEXTIBYTE (5) |
| 143 | #define AccessLength_SEPTIBYTE (6) |
| 144 | #define AccessLength_DOUBLEWORD (7) |
| 145 | |
| 146 | #if defined(HASFPU) |
| 147 | /* FPU registers must be one of the following types. All other values |
| 148 | are reserved (and undefined). */ |
| 149 | typedef enum { |
| 150 | fmt_single = 0, |
| 151 | fmt_double = 1, |
| 152 | fmt_word = 4, |
| 153 | fmt_long = 5, |
| 154 | /* The following are well outside the normal acceptable format |
| 155 | range, and are used in the register status vector. */ |
| 156 | fmt_unknown = 0x10000000, |
| 157 | fmt_uninterpreted = 0x20000000, |
| 158 | } FP_formats; |
| 159 | #endif /* HASFPU */ |
| 160 | |
| 161 | /* NOTE: We cannot avoid globals, since the GDB "sim_" interface does |
| 162 | not allow a private variable to be passed around. This means that |
| 163 | simulators under GDB can only be single-threaded. However, it would |
| 164 | be possible for the simulators to be multi-threaded if GDB allowed |
| 165 | for a private pointer to be maintained. i.e. a general "void **ptr" |
| 166 | variable that GDB passed around in the argument list to all of |
| 167 | sim_xxx() routines. It could be initialised to NULL by GDB, and |
| 168 | then updated by sim_open() and used by the other sim_xxx() support |
| 169 | functions. This would allow new features in the simulator world, |
| 170 | like storing a context - continuing execution to gather a result, |
| 171 | and then going back to the point where the context was saved and |
| 172 | changing some state before continuing. i.e. the ability to perform |
| 173 | UNDOs on simulations. It would also allow the simulation of |
| 174 | shared-memory multi-processor systems. */ |
| 175 | |
| 176 | static host_callback *callback = NULL; /* handle onto the current callback structure */ |
| 177 | |
| 178 | /* This is nasty, since we have to rely on matching the register |
| 179 | numbers used by GDB. Unfortunately, depending on the MIPS target |
| 180 | GDB uses different register numbers. We cannot just include the |
| 181 | relevant "gdb/tm.h" link, since GDB may not be configured before |
| 182 | the sim world, and also the GDB header file requires too much other |
| 183 | state. */ |
| 184 | /* TODO: Sort out a scheme for *KNOWING* the mapping between real |
| 185 | registers, and the numbers that GDB uses. At the moment due to the |
| 186 | order that the tools are built, we cannot rely on a configured GDB |
| 187 | world whilst constructing the simulator. This means we have to |
| 188 | assume the GDB register number mapping. */ |
| 189 | #define LAST_EMBED_REGNUM (89) |
| 190 | |
| 191 | /* To keep this default simulator simple, and fast, we use a direct |
| 192 | vector of registers. The internal simulator engine then uses |
| 193 | manifests to access the correct slot. */ |
| 194 | static ut_reg registers[LAST_EMBED_REGNUM + 1]; |
| 195 | static int register_widths[LAST_EMBED_REGNUM + 1]; |
| 196 | |
| 197 | #define GPR (®isters[0]) |
| 198 | #if defined(HASFPU) |
| 199 | #define FGRIDX (38) |
| 200 | #define FGR (®isters[FGRIDX]) |
| 201 | #endif /* HASFPU */ |
| 202 | #define LO (registers[33]) |
| 203 | #define HI (registers[34]) |
| 204 | #define PC (registers[37]) |
| 205 | #define CAUSE (registers[36]) |
| 206 | #define SRIDX (32) |
| 207 | #define SR (registers[SRIDX]) /* CPU status register */ |
| 208 | #define FCR0IDX (71) |
| 209 | #define FCR0 (registers[FCR0IDX]) /* really a 32bit register */ |
| 210 | #define FCR31IDX (70) |
| 211 | #define FCR31 (registers[FCR31IDX]) /* really a 32bit register */ |
| 212 | #define FCSR (FCR31) |
| 213 | #define COCIDX (LAST_EMBED_REGNUM + 2) /* special case : outside the normal range */ |
| 214 | |
| 215 | /* The following are pseudonyms for standard registers */ |
| 216 | #define ZERO (registers[0]) |
| 217 | #define V0 (registers[2]) |
| 218 | #define A0 (registers[4]) |
| 219 | #define A1 (registers[5]) |
| 220 | #define A2 (registers[6]) |
| 221 | #define A3 (registers[7]) |
| 222 | #define SP (registers[29]) |
| 223 | #define RA (registers[31]) |
| 224 | |
| 225 | static ut_reg EPC = 0; /* Exception PC */ |
| 226 | |
| 227 | #if defined(HASFPU) |
| 228 | /* Keep the current format state for each register: */ |
| 229 | static FP_formats fpr_state[32]; |
| 230 | #endif /* HASFPU */ |
| 231 | |
| 232 | /* VR4300 CP0 configuration register: */ |
| 233 | static unsigned int CONFIG = 0; |
| 234 | |
| 235 | /* The following are internal simulator state variables: */ |
| 236 | static ut_reg IPC = 0; /* internal Instruction PC */ |
| 237 | static ut_reg DSPC = 0; /* delay-slot PC */ |
| 238 | |
| 239 | |
| 240 | /* TODO : these should be the bitmasks for these bits within the |
| 241 | status register. At the moment the following are VR4300 |
| 242 | bit-positions: */ |
| 243 | #define status_KSU_mask (0x3) /* mask for KSU bits */ |
| 244 | #define status_KSU_shift (3) /* shift for field */ |
| 245 | #define ksu_kernel (0x0) |
| 246 | #define ksu_supervisor (0x1) |
| 247 | #define ksu_user (0x2) |
| 248 | #define ksu_unknown (0x3) |
| 249 | |
| 250 | #define status_RE (1 << 25) /* Reverse Endian in user mode */ |
| 251 | #define status_FR (1 << 26) /* enables MIPS III additional FP registers */ |
| 252 | #define status_SR (1 << 20) /* soft reset or NMI */ |
| 253 | #define status_BEV (1 << 22) /* Location of general exception vectors */ |
| 254 | #define status_TS (1 << 21) /* TLB shutdown has occurred */ |
| 255 | #define status_ERL (1 << 2) /* Error level */ |
| 256 | #define status_RP (1 << 27) /* Reduced Power mode */ |
| 257 | |
| 258 | #define config_EP_mask (0xF) |
| 259 | #define config_EP_shift (27) |
| 260 | #define config_EP_D (0x0) |
| 261 | #define config_EP_DxxDxx (0x6) |
| 262 | |
| 263 | #define config_BE (1 << 15) |
| 264 | |
| 265 | #define cause_BD ((unsigned)1 << 31) /* Exception in branch delay slot */ |
| 266 | |
| 267 | #if defined(HASFPU) |
| 268 | /* Macro to update FPSR condition-code field. This is complicated by |
| 269 | the fact that there is a hole in the index range of the bits within |
| 270 | the FCSR register. Also, the number of bits visible depends on the |
| 271 | MIPS ISA version being supported. */ |
| 272 | #define SETFCC(cc,v) {\ |
| 273 | int bit = ((cc == 0) ? 23 : (24 + (cc)));\ |
| 274 | FCSR = ((FCSR & ~(1 << bit)) | ((v) << bit));\ |
| 275 | } |
| 276 | #define GETFCC(cc) (((((cc) == 0) ? (FCSR & (1 << 23)) : (FCSR & (1 << (24 + (cc))))) != 0) ? 1 : 0) |
| 277 | |
| 278 | /* This should be the COC1 value at the start of the preceding |
| 279 | instruction: */ |
| 280 | #define PREVCOC1() ((state & simPCOC1) ? 1 : 0) |
| 281 | #endif /* HASFPU */ |
| 282 | |
| 283 | /* Standard FCRS bits: */ |
| 284 | #define IR (0) /* Inexact Result */ |
| 285 | #define UF (1) /* UnderFlow */ |
| 286 | #define OF (2) /* OverFlow */ |
| 287 | #define DZ (3) /* Division by Zero */ |
| 288 | #define IO (4) /* Invalid Operation */ |
| 289 | #define UO (5) /* Unimplemented Operation */ |
| 290 | |
| 291 | /* Get masks for individual flags: */ |
| 292 | #if 1 /* SAFE version */ |
| 293 | #define FP_FLAGS(b) (((unsigned)(b) < 5) ? (1 << ((b) + 2)) : 0) |
| 294 | #define FP_ENABLE(b) (((unsigned)(b) < 5) ? (1 << ((b) + 7)) : 0) |
| 295 | #define FP_CAUSE(b) (((unsigned)(b) < 6) ? (1 << ((b) + 12)) : 0) |
| 296 | #else |
| 297 | #define FP_FLAGS(b) (1 << ((b) + 2)) |
| 298 | #define FP_ENABLE(b) (1 << ((b) + 7)) |
| 299 | #define FP_CAUSE(b) (1 << ((b) + 12)) |
| 300 | #endif |
| 301 | |
| 302 | #define FP_FS (1 << 24) /* MIPS III onwards : Flush to Zero */ |
| 303 | |
| 304 | #define FP_MASK_RM (0x3) |
| 305 | #define FP_SH_RM (0) |
| 306 | #define FP_RM_NEAREST (0) /* Round to nearest (Round) */ |
| 307 | #define FP_RM_TOZERO (1) /* Round to zero (Trunc) */ |
| 308 | #define FP_RM_TOPINF (2) /* Round to Plus infinity (Ceil) */ |
| 309 | #define FP_RM_TOMINF (3) /* Round to Minus infinity (Floor) */ |
| 310 | #define GETRM() (int)((FCSR >> FP_SH_RM) & FP_MASK_RM) |
| 311 | |
| 312 | /* Slots for delayed register updates. For the moment we just have a |
| 313 | fixed number of slots (rather than a more generic, dynamic |
| 314 | system). This keeps the simulator fast. However, we only allow for |
| 315 | the register update to be delayed for a single instruction |
| 316 | cycle. */ |
| 317 | #define PSLOTS (5) /* Maximum number of instruction cycles */ |
| 318 | static int pending_in; |
| 319 | static int pending_out; |
| 320 | static int pending_total; |
| 321 | static int pending_slot_count[PSLOTS]; |
| 322 | static int pending_slot_reg[PSLOTS]; |
| 323 | static ut_reg pending_slot_value[PSLOTS]; |
| 324 | |
| 325 | /* The following are not used for MIPS IV onwards: */ |
| 326 | #define PENDING_FILL(r,v) {\ |
| 327 | /* printf("DBG: FILL BEFORE pending_in = %d, pending_out = %d, pending_total = %d\n",pending_in,pending_out,pending_total); */\ |
| 328 | if (pending_slot_reg[pending_in] != (LAST_EMBED_REGNUM + 1))\ |
| 329 | sim_warning("Attempt to over-write pending value");\ |
| 330 | pending_slot_count[pending_in] = 2;\ |
| 331 | pending_slot_reg[pending_in] = (r);\ |
| 332 | pending_slot_value[pending_in] = (uword64)(v);\ |
| 333 | /*printf("DBG: FILL reg %d value = 0x%08X%08X\n",(r),WORD64HI(v),WORD64LO(v));*/\ |
| 334 | pending_total++;\ |
| 335 | pending_in++;\ |
| 336 | if (pending_in == PSLOTS)\ |
| 337 | pending_in = 0;\ |
| 338 | /*printf("DBG: FILL AFTER pending_in = %d, pending_out = %d, pending_total = %d\n",pending_in,pending_out,pending_total);*/\ |
| 339 | } |
| 340 | |
| 341 | static int LLBIT = 0; |
| 342 | /* LLBIT = Load-Linked bit. A bit of "virtual" state used by atomic |
| 343 | read-write instructions. It is set when a linked load occurs. It is |
| 344 | tested and cleared by the conditional store. It is cleared (during |
| 345 | other CPU operations) when a store to the location would no longer |
| 346 | be atomic. In particular, it is cleared by exception return |
| 347 | instructions. */ |
| 348 | |
| 349 | static int HIACCESS = 0; |
| 350 | static int LOACCESS = 0; |
| 351 | /* The HIACCESS and LOACCESS counts are used to ensure that |
| 352 | corruptions caused by using the HI or LO register to close to a |
| 353 | following operation are spotted. */ |
| 354 | static ut_reg HLPC = 0; |
| 355 | |
| 356 | /* TODO: The 4300 has interlocks so we should not need to warn of the possible over-write (CHECK THIS) */ |
| 357 | /* If either of the preceding two instructions have accessed the HI or |
| 358 | LO registers, then the values they see should be |
| 359 | undefined. However, to keep the simulator world simple, we just let |
| 360 | them use the value read and raise a warning to notify the user: */ |
| 361 | #define CHECKHILO(s) {\ |
| 362 | if ((HIACCESS != 0) || (LOACCESS != 0))\ |
| 363 | sim_warning("%s over-writing HI and LO registers values (PC = 0x%08X%08X HLPC = 0x%08X%08X)\n",(s),(unsigned int)(PC>>32),(unsigned int)(PC&0xFFFFFFFF),(unsigned int)(HLPC>>32),(unsigned int)(HLPC&0xFFFFFFFF));\ |
| 364 | } |
| 365 | |
| 366 | /* NOTE: We keep the following status flags as bit values (1 for true, |
| 367 | 0 for false). This allows them to be used in binary boolean |
| 368 | operations without worrying about what exactly the non-zero true |
| 369 | value is. */ |
| 370 | |
| 371 | /* UserMode */ |
| 372 | #define UserMode ((((SR & status_KSU_mask) >> status_KSU_shift) == ksu_user) ? 1 : 0) |
| 373 | |
| 374 | /* BigEndianMem */ |
| 375 | /* Hardware configuration. Affects endianness of LoadMemory and |
| 376 | StoreMemory and the endianness of Kernel and Supervisor mode |
| 377 | execution. The value is 0 for little-endian; 1 for big-endian. */ |
| 378 | #define BigEndianMem ((CONFIG & config_BE) ? 1 : 0) |
| 379 | /* NOTE: Problems will occur if the simulator memory model does not |
| 380 | match the host program expectation. i.e. if the host is writing |
| 381 | big-endian values to a little-endian memory model. */ |
| 382 | |
| 383 | /* ReverseEndian */ |
| 384 | /* This mode is selected if in User mode with the RE bit being set in |
| 385 | SR (Status Register). It reverses the endianness of load and store |
| 386 | instructions. */ |
| 387 | #define ReverseEndian (((SR & status_RE) && UserMode) ? 1 : 0) |
| 388 | |
| 389 | /* BigEndianCPU */ |
| 390 | /* The endianness for load and store instructions (0=little;1=big). In |
| 391 | User mode this endianness may be switched by setting the state_RE |
| 392 | bit in the SR register. Thus, BigEndianCPU may be computed as |
| 393 | (BigEndienMem EOR ReverseEndian). */ |
| 394 | #define BigEndianCPU (BigEndianMem ^ ReverseEndian) /* Already bits */ |
| 395 | |
| 396 | #if !defined(FASTSIM) || defined(PROFILE) |
| 397 | /* At the moment these values will be the same, since we do not have |
| 398 | access to the pipeline cycle count information from the simulator |
| 399 | engine. */ |
| 400 | static unsigned int instruction_fetches = 0; |
| 401 | static unsigned int instruction_fetch_overflow = 0; |
| 402 | static unsigned int pipeline_ticks = 0; |
| 403 | #endif |
| 404 | |
| 405 | /* Flags in the "state" variable: */ |
| 406 | #define simSTOP (1 << 0) /* 0 = execute; 1 = stop simulation */ |
| 407 | #define simSTEP (1 << 1) /* 0 = run; 1 = single-step */ |
| 408 | #define simHALTEX (1 << 2) /* 0 = run; 1 = halt on exception */ |
| 409 | #define simHALTIN (1 << 3) /* 0 = run; 1 = halt on interrupt */ |
| 410 | #define simTRACE (1 << 8) /* 0 = do nothing; 1 = trace address activity */ |
| 411 | #define simPROFILE (1 << 9) /* 0 = do nothing; 1 = gather profiling samples */ |
| 412 | #define simHOSTBE (1 << 10) /* 0 = little-endian; 1 = big-endian (host endianness) */ |
| 413 | /* Whilst simSTOP is not set, the simulator control loop should just |
| 414 | keep simulating instructions. The simSTEP flag is used to force |
| 415 | single-step execution. */ |
| 416 | #define simBE (1 << 16) /* 0 = little-endian; 1 = big-endian (target endianness) */ |
| 417 | #define simPCOC0 (1 << 17) /* COC[1] from current */ |
| 418 | #define simPCOC1 (1 << 18) /* COC[1] from previous */ |
| 419 | #define simDELAYSLOT (1 << 24) /* 0 = do nothing; 1 = delay slot entry exists */ |
| 420 | #define simSKIPNEXT (1 << 25) /* 0 = do nothing; 1 = skip instruction */ |
| 421 | #define simEXCEPTION (1 << 26) /* 0 = no exception; 1 = exception has occurred */ |
| 422 | #define simEXIT (1 << 27) /* 0 = do nothing; 1 = run-time exit() processing */ |
| 423 | |
| 424 | static unsigned int state = 0; |
| 425 | static unsigned int rcexit = 0; /* _exit() reason code holder */ |
| 426 | |
| 427 | #define DELAYSLOT() {\ |
| 428 | if (state & simDELAYSLOT)\ |
| 429 | sim_warning("Delay slot already activated (branch in delay slot?)");\ |
| 430 | state |= simDELAYSLOT;\ |
| 431 | } |
| 432 | |
| 433 | #define NULLIFY() {\ |
| 434 | state &= ~simDELAYSLOT;\ |
| 435 | state |= simSKIPNEXT;\ |
| 436 | } |
| 437 | |
| 438 | #define K0BASE (0x80000000) |
| 439 | #define K0SIZE (0x20000000) |
| 440 | #define K1BASE (0xA0000000) |
| 441 | #define K1SIZE (0x20000000) |
| 442 | |
| 443 | /* Very simple memory model to start with: */ |
| 444 | static unsigned char *membank = NULL; |
| 445 | static ut_reg membank_base = K1BASE; |
| 446 | static unsigned membank_size = (1 << 20); /* (16 << 20); */ /* power-of-2 */ |
| 447 | |
| 448 | /* Simple run-time monitor support */ |
| 449 | static unsigned char *monitor = NULL; |
| 450 | static ut_reg monitor_base = 0xBFC00000; |
| 451 | static unsigned monitor_size = (1 << 11); /* power-of-2 */ |
| 452 | |
| 453 | static char *logfile = NULL; /* logging disabled by default */ |
| 454 | static FILE *logfh = NULL; |
| 455 | |
| 456 | #if defined(TRACE) |
| 457 | static char *tracefile = "trace.din"; /* default filename for trace log */ |
| 458 | static FILE *tracefh = NULL; |
| 459 | #endif /* TRACE */ |
| 460 | |
| 461 | #if defined(PROFILE) |
| 462 | static unsigned profile_frequency = 256; |
| 463 | static unsigned profile_nsamples = (128 << 10); |
| 464 | static unsigned short *profile_hist = NULL; |
| 465 | static ut_reg profile_minpc; |
| 466 | static ut_reg profile_maxpc; |
| 467 | static int profile_shift = 0; /* address shift amount */ |
| 468 | #endif /* PROFILE */ |
| 469 | |
| 470 | /* The following are used to provide shortcuts to the required version |
| 471 | of host<->target copying. This avoids run-time conditionals, which |
| 472 | would slow the simulator throughput. */ |
| 473 | typedef unsigned int (*fnptr_read_word) PARAMS((unsigned char *memory)); |
| 474 | typedef unsigned int (*fnptr_swap_word) PARAMS((unsigned int data)); |
| 475 | typedef uword64 (*fnptr_read_long) PARAMS((unsigned char *memory)); |
| 476 | typedef uword64 (*fnptr_swap_long) PARAMS((uword64 data)); |
| 477 | |
| 478 | static fnptr_read_word host_read_word; |
| 479 | static fnptr_read_long host_read_long; |
| 480 | static fnptr_swap_word host_swap_word; |
| 481 | static fnptr_swap_long host_swap_long; |
| 482 | |
| 483 | /*---------------------------------------------------------------------------*/ |
| 484 | /*-- GDB simulator interface ------------------------------------------------*/ |
| 485 | /*---------------------------------------------------------------------------*/ |
| 486 | |
| 487 | static void dotrace PARAMS((FILE *tracefh,int type,unsigned int address,int width,char *comment,...)); |
| 488 | static void sim_warning PARAMS((char *fmt,...)); |
| 489 | extern void sim_error PARAMS((char *fmt,...)); |
| 490 | static void ColdReset PARAMS((void)); |
| 491 | static int AddressTranslation PARAMS((uword64 vAddr,int IorD,int LorS,uword64 *pAddr,int *CCA,int host,int raw)); |
| 492 | static void StoreMemory PARAMS((int CCA,int AccessLength,uword64 MemElem,uword64 pAddr,uword64 vAddr,int raw)); |
| 493 | static uword64 LoadMemory PARAMS((int CCA,int AccessLength,uword64 pAddr,uword64 vAddr,int IorD,int raw)); |
| 494 | static void SignalException PARAMS((int exception,...)); |
| 495 | static void simulate PARAMS((void)); |
| 496 | static long getnum PARAMS((char *value)); |
| 497 | extern void sim_size PARAMS((unsigned int newsize)); |
| 498 | extern void sim_set_profile PARAMS((int frequency)); |
| 499 | static unsigned int power2 PARAMS((unsigned int value)); |
| 500 | |
| 501 | /*---------------------------------------------------------------------------*/ |
| 502 | |
| 503 | void |
| 504 | sim_open (args) |
| 505 | char *args; |
| 506 | { |
| 507 | if (callback == NULL) { |
| 508 | fprintf(stderr,"SIM Error: sim_open() called without callbacks attached\n"); |
| 509 | return; |
| 510 | } |
| 511 | |
| 512 | /* The following ensures that the standard file handles for stdin, |
| 513 | stdout and stderr are initialised: */ |
| 514 | callback->init(callback); |
| 515 | |
| 516 | state = 0; |
| 517 | CHECKSIM(); |
| 518 | if (state & simEXCEPTION) { |
| 519 | fprintf(stderr,"This simulator is not suitable for this host configuration\n"); |
| 520 | exit(1); |
| 521 | } |
| 522 | |
| 523 | { |
| 524 | int data = 0x12; |
| 525 | if (*((char *)&data) != 0x12) |
| 526 | state |= simHOSTBE; /* big-endian host */ |
| 527 | } |
| 528 | |
| 529 | #if defined(HASFPU) |
| 530 | /* Check that the host FPU conforms to IEEE 754-1985 for the SINGLE |
| 531 | and DOUBLE binary formats. This is a bit nasty, requiring that we |
| 532 | trust the explicit manifests held in the source: */ |
| 533 | { |
| 534 | unsigned int s[2]; |
| 535 | s[state & simHOSTBE ? 0 : 1] = 0x40805A5A; |
| 536 | s[state & simHOSTBE ? 1 : 0] = 0x00000000; |
| 537 | |
| 538 | /* TODO: We need to cope with the simulated target and the host |
| 539 | not having the same endianness. This will require the high and |
| 540 | low words of a (double) to be swapped when converting between |
| 541 | the host and the simulated target. */ |
| 542 | |
| 543 | if (((float)4.01102924346923828125 != *(float *)(s + ((state & simHOSTBE) ? 0 : 1))) || ((double)523.2939453125 != *(double *)s)) { |
| 544 | fprintf(stderr,"The host executing the simulator does not seem to have IEEE 754-1985 std FP\n"); |
| 545 | fprintf(stderr,"*(float *)s = %.20f (4.01102924346923828125)\n",*(float *)s); |
| 546 | fprintf(stderr,"*(double *)s = %.20f (523.2939453125)\n",*(double *)s); |
| 547 | exit(1); |
| 548 | } |
| 549 | } |
| 550 | #endif /* HASFPU */ |
| 551 | |
| 552 | /* This is NASTY, in that we are assuming the size of specific |
| 553 | registers: */ |
| 554 | { |
| 555 | int rn; |
| 556 | for (rn = 0; (rn < (LAST_EMBED_REGNUM + 1)); rn++) { |
| 557 | if (rn < 32) |
| 558 | register_widths[rn] = GPRLEN; |
| 559 | else if ((rn >= FGRIDX) && (rn < (FGRIDX + 32))) |
| 560 | register_widths[rn] = GPRLEN; |
| 561 | else if ((rn >= 33) && (rn <= 37)) |
| 562 | register_widths[rn] = GPRLEN; |
| 563 | else if ((rn == SRIDX) || (rn == FCR0IDX) || (rn == FCR31IDX) || ((rn >= 72) && (rn <= 89))) |
| 564 | register_widths[rn] = 32; |
| 565 | else |
| 566 | register_widths[rn] = 0; |
| 567 | } |
| 568 | } |
| 569 | |
| 570 | /* It would be good if we could select particular named MIPS |
| 571 | architecture simulators. However, having a pre-built, fixed |
| 572 | engine would mean including multiple engines. If the simulator is |
| 573 | changed to a run-time conditional version, then the ability to |
| 574 | select a particular architecture would be straightforward. */ |
| 575 | if (args != NULL) { |
| 576 | int c; |
| 577 | char *cline; |
| 578 | char **argv; |
| 579 | int argc; |
| 580 | static struct option cmdline[] = { |
| 581 | {"help", 0,0,'h'}, |
| 582 | {"log", 1,0,'l'}, |
| 583 | {"name", 1,0,'n'}, |
| 584 | {"profile", 0,0,'p'}, |
| 585 | {"size", 1,0,'s'}, |
| 586 | {"trace", 0,0,'t'}, |
| 587 | {"tracefile",1,0,'z'}, |
| 588 | {"frequency",1,0,'y'}, |
| 589 | {"samples", 1,0,'x'}, |
| 590 | {0, 0,0,0} |
| 591 | }; |
| 592 | |
| 593 | /* Unfortunately, getopt_long() is designed to be used with |
| 594 | vectors, where the first option is normally program name (and |
| 595 | ignored). We cheat by creating a dummy first argument, so that |
| 596 | we can use the standard argument processing. */ |
| 597 | #define DUMMYARG "simulator " |
| 598 | cline = (char *)malloc(strlen(args) + strlen(DUMMYARG) + 1); |
| 599 | if (cline == NULL) { |
| 600 | fprintf(stderr,"Failed to allocate memory for command line buffer\n"); |
| 601 | exit(1); |
| 602 | } |
| 603 | sprintf(cline,"%s%s",DUMMYARG,args); |
| 604 | argv = buildargv(cline); |
| 605 | for (argc = 0; argv[argc]; argc++); |
| 606 | |
| 607 | /* Unfortunately, getopt_long() assumes that it is ignoring the |
| 608 | first argument (normally the program name). This means it |
| 609 | ignores the first option on our "args" line. */ |
| 610 | optind = 0; /* Force reset of argument processing */ |
| 611 | while (1) { |
| 612 | int option_index = 0; |
| 613 | |
| 614 | c = getopt_long(argc,argv,"hn:s:tp",cmdline,&option_index); |
| 615 | if (c == -1) |
| 616 | break; |
| 617 | |
| 618 | switch (c) { |
| 619 | case 'h': |
| 620 | callback->printf_filtered(callback,"Usage:\n\t\ |
| 621 | target sim [-h] [--log=<file>] [--name=<model>] [--size=<amount>]"); |
| 622 | #if defined(TRACE) |
| 623 | callback->printf_filtered(callback," [-t [--tracefile=<name>]]"); |
| 624 | #endif /* TRACE */ |
| 625 | #if defined(PROFILE) |
| 626 | callback->printf_filtered(callback," [-p [--frequency=<count>] [--samples=<count>]]"); |
| 627 | #endif /* PROFILE */ |
| 628 | callback->printf_filtered(callback,"\n"); |
| 629 | break; |
| 630 | |
| 631 | case 'l': |
| 632 | if (optarg != NULL) { |
| 633 | char *tmp; |
| 634 | tmp = (char *)malloc(strlen(optarg) + 1); |
| 635 | if (tmp == NULL) |
| 636 | callback->printf_filtered(callback,"Failed to allocate buffer for logfile name \"%s\"\n",optarg); |
| 637 | else { |
| 638 | strcpy(tmp,optarg); |
| 639 | logfile = tmp; |
| 640 | } |
| 641 | } |
| 642 | break; |
| 643 | |
| 644 | case 'n': |
| 645 | callback->printf_filtered(callback,"Explicit model selection not yet available (Ignoring \"%s\")\n",optarg); |
| 646 | break; |
| 647 | |
| 648 | case 's': |
| 649 | membank_size = (unsigned)getnum(optarg); |
| 650 | break; |
| 651 | |
| 652 | case 't': |
| 653 | #if defined(TRACE) |
| 654 | /* Eventually the simTRACE flag could be treated as a toggle, to |
| 655 | allow external control of the program points being traced |
| 656 | (i.e. only from main onwards, excluding the run-time setup, |
| 657 | etc.). */ |
| 658 | state |= simTRACE; |
| 659 | #else /* !TRACE */ |
| 660 | fprintf(stderr,"\ |
| 661 | Simulator constructed without tracing support (for performance).\n\ |
| 662 | Re-compile simulator with \"-DTRACE\" to enable this option.\n"); |
| 663 | #endif /* !TRACE */ |
| 664 | break; |
| 665 | |
| 666 | case 'z': |
| 667 | #if defined(TRACE) |
| 668 | if (optarg != NULL) { |
| 669 | char *tmp; |
| 670 | tmp = (char *)malloc(strlen(optarg) + 1); |
| 671 | if (tmp == NULL) |
| 672 | callback->printf_filtered(callback,"Failed to allocate buffer for tracefile name \"%s\"\n",optarg); |
| 673 | else { |
| 674 | strcpy(tmp,optarg); |
| 675 | tracefile = tmp; |
| 676 | callback->printf_filtered(callback,"Placing trace information into file \"%s\"\n",tracefile); |
| 677 | } |
| 678 | } |
| 679 | #endif /* TRACE */ |
| 680 | break; |
| 681 | |
| 682 | case 'p': |
| 683 | #if defined(PROFILE) |
| 684 | state |= simPROFILE; |
| 685 | #else /* !PROFILE */ |
| 686 | fprintf(stderr,"\ |
| 687 | Simulator constructed without profiling support (for performance).\n\ |
| 688 | Re-compile simulator with \"-DPROFILE\" to enable this option.\n"); |
| 689 | #endif /* !PROFILE */ |
| 690 | break; |
| 691 | |
| 692 | case 'x': |
| 693 | #if defined(PROFILE) |
| 694 | profile_nsamples = (unsigned)getnum(optarg); |
| 695 | #endif /* PROFILE */ |
| 696 | break; |
| 697 | |
| 698 | case 'y': |
| 699 | #if defined(PROFILE) |
| 700 | sim_set_profile((int)getnum(optarg)); |
| 701 | #endif /* PROFILE */ |
| 702 | break; |
| 703 | |
| 704 | default: |
| 705 | callback->printf_filtered(callback,"Warning: Simulator getopt returned unrecognised code 0x%08X\n",c); |
| 706 | case '?': |
| 707 | break; |
| 708 | } |
| 709 | } |
| 710 | |
| 711 | #if 0 |
| 712 | if (optind < argc) { |
| 713 | callback->printf_filtered(callback,"Warning: Ignoring spurious non-option arguments "); |
| 714 | while (optind < argc) |
| 715 | callback->printf_filtered(callback,"\"%s\" ",argv[optind++]); |
| 716 | callback->printf_filtered(callback,"\n"); |
| 717 | } |
| 718 | #endif |
| 719 | |
| 720 | freeargv(argv); |
| 721 | } |
| 722 | |
| 723 | if (logfile != NULL) { |
| 724 | if (strcmp(logfile,"-") == 0) |
| 725 | logfh = stdout; |
| 726 | else { |
| 727 | logfh = fopen(logfile,"wb+"); |
| 728 | if (logfh == NULL) { |
| 729 | callback->printf_filtered(callback,"Failed to create file \"%s\", writing log information to stderr.\n",tracefile); |
| 730 | logfh = stderr; |
| 731 | } |
| 732 | } |
| 733 | } |
| 734 | |
| 735 | /* If the host has "mmap" available we could use it to provide a |
| 736 | very large virtual address space for the simulator, since memory |
| 737 | would only be allocated within the "mmap" space as it is |
| 738 | accessed. This can also be linked to the architecture specific |
| 739 | support, required to simulate the MMU. */ |
| 740 | sim_size(membank_size); |
| 741 | /* NOTE: The above will also have enabled any profiling state */ |
| 742 | |
| 743 | ColdReset(); |
| 744 | /* If we were providing a more complete I/O, co-processor or memory |
| 745 | simulation, we should perform any "device" initialisation at this |
| 746 | point. This can include pre-loading memory areas with particular |
| 747 | patterns (e.g. simulating ROM monitors). */ |
| 748 | |
| 749 | /* We can start writing to the memory, now that the processor has |
| 750 | been reset: */ |
| 751 | monitor = (unsigned char *)calloc(1,monitor_size); |
| 752 | if (!monitor) { |
| 753 | fprintf(stderr,"Not enough VM for monitor simulation (%d bytes)\n",monitor_size); |
| 754 | } else { |
| 755 | int loop; |
| 756 | /* Entry into the IDT monitor is via fixed address vectors, and |
| 757 | not using machine instructions. To avoid clashing with use of |
| 758 | the MIPS TRAP system, we place our own (simulator specific) |
| 759 | "undefined" instructions into the relevant vector slots. */ |
| 760 | for (loop = 0; (loop < monitor_size); loop += 4) { |
| 761 | uword64 vaddr = (monitor_base + loop); |
| 762 | uword64 paddr; |
| 763 | int cca; |
| 764 | if (AddressTranslation(vaddr,isDATA,isSTORE,&paddr,&cca,isTARGET,isRAW)) |
| 765 | StoreMemory(cca,AccessLength_WORD,(RSVD_INSTRUCTION | ((loop >> 2) & RSVD_INSTRUCTION_AMASK)),paddr,vaddr,isRAW); |
| 766 | } |
| 767 | /* The PMON monitor uses the same address space, but rather than |
| 768 | branching into it the address of a routine is loaded. We can |
| 769 | cheat for the moment, and direct the PMON routine to IDT style |
| 770 | instructions within the monitor space. This relies on the IDT |
| 771 | monitor not using the locations from 0xBFC00500 onwards as its |
| 772 | entry points.*/ |
| 773 | for (loop = 0; (loop < 24); loop++) |
| 774 | { |
| 775 | uword64 vaddr = (monitor_base + 0x500 + (loop * 4)); |
| 776 | uword64 paddr; |
| 777 | int cca; |
| 778 | unsigned int value = ((0x500 - 8) / 8); /* default UNDEFINED reason code */ |
| 779 | switch (loop) |
| 780 | { |
| 781 | case 0: /* read */ |
| 782 | value = 7; |
| 783 | break; |
| 784 | |
| 785 | case 1: /* write */ |
| 786 | value = 8; |
| 787 | break; |
| 788 | |
| 789 | case 2: /* open */ |
| 790 | value = 6; |
| 791 | break; |
| 792 | |
| 793 | case 3: /* close */ |
| 794 | value = 10; |
| 795 | break; |
| 796 | |
| 797 | case 5: /* printf */ |
| 798 | value = ((0x500 - 16) / 8); /* not an IDT reason code */ |
| 799 | break; |
| 800 | |
| 801 | case 8: /* cliexit */ |
| 802 | value = 17; |
| 803 | break; |
| 804 | } |
| 805 | value = (monitor_base + (value * 8)); |
| 806 | if (AddressTranslation(vaddr,isDATA,isSTORE,&paddr,&cca,isTARGET,isRAW)) |
| 807 | StoreMemory(cca,AccessLength_WORD,value,paddr,vaddr,isRAW); |
| 808 | else |
| 809 | sim_error("Failed to write to monitor space 0x%08X%08X",WORD64HI(vaddr),WORD64LO(vaddr)); |
| 810 | } |
| 811 | } |
| 812 | |
| 813 | #if defined(TRACE) |
| 814 | if (state & simTRACE) { |
| 815 | tracefh = fopen(tracefile,"wb+"); |
| 816 | if (tracefh == NULL) { |
| 817 | sim_warning("Failed to create file \"%s\", writing trace information to stderr.",tracefile); |
| 818 | tracefh = stderr; |
| 819 | } |
| 820 | } |
| 821 | #endif /* TRACE */ |
| 822 | |
| 823 | return; |
| 824 | } |
| 825 | |
| 826 | /* For the profile writing, we write the data in the host |
| 827 | endianness. This unfortunately means we are assuming that the |
| 828 | profile file we create is processed on the same host executing the |
| 829 | simulator. The gmon.out file format should either have an explicit |
| 830 | endianness, or a method of encoding the endianness in the file |
| 831 | header. */ |
| 832 | static int |
| 833 | writeout32(fh,val) |
| 834 | FILE *fh; |
| 835 | unsigned int val; |
| 836 | { |
| 837 | char buff[4]; |
| 838 | int res = 1; |
| 839 | |
| 840 | if (state & simHOSTBE) { |
| 841 | buff[3] = ((val >> 0) & 0xFF); |
| 842 | buff[2] = ((val >> 8) & 0xFF); |
| 843 | buff[1] = ((val >> 16) & 0xFF); |
| 844 | buff[0] = ((val >> 24) & 0xFF); |
| 845 | } else { |
| 846 | buff[0] = ((val >> 0) & 0xFF); |
| 847 | buff[1] = ((val >> 8) & 0xFF); |
| 848 | buff[2] = ((val >> 16) & 0xFF); |
| 849 | buff[3] = ((val >> 24) & 0xFF); |
| 850 | } |
| 851 | if (fwrite(buff,4,1,fh) != 1) { |
| 852 | sim_warning("Failed to write 4bytes to the profile file"); |
| 853 | res = 0; |
| 854 | } |
| 855 | return(res); |
| 856 | } |
| 857 | |
| 858 | static int |
| 859 | writeout16(fh,val) |
| 860 | FILE *fh; |
| 861 | unsigned short val; |
| 862 | { |
| 863 | char buff[2]; |
| 864 | int res = 1; |
| 865 | if (state & simHOSTBE) { |
| 866 | buff[1] = ((val >> 0) & 0xFF); |
| 867 | buff[0] = ((val >> 8) & 0xFF); |
| 868 | } else { |
| 869 | buff[0] = ((val >> 0) & 0xFF); |
| 870 | buff[1] = ((val >> 8) & 0xFF); |
| 871 | } |
| 872 | if (fwrite(buff,2,1,fh) != 1) { |
| 873 | sim_warning("Failed to write 2bytes to the profile file"); |
| 874 | res = 0; |
| 875 | } |
| 876 | return(res); |
| 877 | } |
| 878 | |
| 879 | void |
| 880 | sim_close (quitting) |
| 881 | int quitting; |
| 882 | { |
| 883 | #ifdef DEBUG |
| 884 | printf("DBG: sim_close: entered (quitting = %d)\n",quitting); |
| 885 | #endif |
| 886 | |
| 887 | /* Cannot assume sim_kill() has been called */ |
| 888 | /* "quitting" is non-zero if we cannot hang on errors */ |
| 889 | |
| 890 | /* Ensure that any resources allocated through the callback |
| 891 | mechanism are released: */ |
| 892 | callback->shutdown(callback); |
| 893 | |
| 894 | #if defined(PROFILE) |
| 895 | if ((state & simPROFILE) && (profile_hist != NULL)) { |
| 896 | unsigned short *p = profile_hist; |
| 897 | FILE *pf = fopen("gmon.out","wb"); |
| 898 | int loop; |
| 899 | |
| 900 | if (pf == NULL) |
| 901 | sim_warning("Failed to open \"gmon.out\" profile file"); |
| 902 | else { |
| 903 | int ok; |
| 904 | #ifdef DEBUG |
| 905 | printf("DBG: minpc = 0x%08X\n",(unsigned int)profile_minpc); |
| 906 | printf("DBG: maxpc = 0x%08X\n",(unsigned int)profile_maxpc); |
| 907 | #endif /* DEBUG */ |
| 908 | ok = writeout32(pf,(unsigned int)profile_minpc); |
| 909 | if (ok) |
| 910 | ok = writeout32(pf,(unsigned int)profile_maxpc); |
| 911 | if (ok) |
| 912 | ok = writeout32(pf,(profile_nsamples * 2) + 12); /* size of sample buffer (+ header) */ |
| 913 | #ifdef DEBUG |
| 914 | printf("DBG: nsamples = %d (size = 0x%08X)\n",profile_nsamples,((profile_nsamples * 2) + 12)); |
| 915 | #endif /* DEBUG */ |
| 916 | for (loop = 0; (ok && (loop < profile_nsamples)); loop++) { |
| 917 | ok = writeout16(pf,profile_hist[loop]); |
| 918 | if (!ok) |
| 919 | break; |
| 920 | } |
| 921 | |
| 922 | fclose(pf); |
| 923 | } |
| 924 | |
| 925 | free(profile_hist); |
| 926 | profile_hist = NULL; |
| 927 | state &= ~simPROFILE; |
| 928 | } |
| 929 | #endif /* PROFILE */ |
| 930 | |
| 931 | #if defined(TRACE) |
| 932 | if (tracefh != stderr) |
| 933 | fclose(tracefh); |
| 934 | state &= ~simTRACE; |
| 935 | #endif /* TRACE */ |
| 936 | |
| 937 | if (logfh != NULL && logfh != stdout && logfh != stderr) |
| 938 | fclose(logfh); |
| 939 | logfh = NULL; |
| 940 | |
| 941 | if (membank) |
| 942 | free(membank); /* cfree not available on all hosts */ |
| 943 | membank = NULL; |
| 944 | |
| 945 | return; |
| 946 | } |
| 947 | |
| 948 | void |
| 949 | sim_resume (step,signal) |
| 950 | int step, signal; |
| 951 | { |
| 952 | #ifdef DEBUG |
| 953 | printf("DBG: sim_resume entered: step = %d, signal = %d (membank = 0x%08X)\n",step,signal,membank); |
| 954 | #endif /* DEBUG */ |
| 955 | |
| 956 | if (step) |
| 957 | state |= simSTEP; /* execute only a single instruction */ |
| 958 | else |
| 959 | state &= ~(simSTOP | simSTEP); /* execute until event */ |
| 960 | |
| 961 | state |= (simHALTEX | simHALTIN); /* treat interrupt event as exception */ |
| 962 | |
| 963 | /* Start executing instructions from the current state (set |
| 964 | explicitly by register updates, or by sim_create_inferior): */ |
| 965 | |
| 966 | simulate(); |
| 967 | return; |
| 968 | } |
| 969 | |
| 970 | int |
| 971 | sim_write (addr,buffer,size) |
| 972 | SIM_ADDR addr; |
| 973 | unsigned char *buffer; |
| 974 | int size; |
| 975 | { |
| 976 | int index = size; |
| 977 | uword64 vaddr = (uword64)addr; |
| 978 | |
| 979 | /* Return the number of bytes written, or zero if error. */ |
| 980 | #ifdef DEBUG |
| 981 | callback->printf_filtered(callback,"sim_write(0x%08X%08X,buffer,%d);\n",WORD64HI(addr),WORD64LO(addr),size); |
| 982 | #endif |
| 983 | |
| 984 | /* We provide raw read and write routines, since we do not want to |
| 985 | count the GDB memory accesses in our statistics gathering. */ |
| 986 | |
| 987 | /* There is a lot of code duplication in the individual blocks |
| 988 | below, but the variables are declared locally to a block to give |
| 989 | the optimiser the best chance of improving the code. We have to |
| 990 | perform slow byte reads from the host memory, to ensure that we |
| 991 | get the data into the correct endianness for the (simulated) |
| 992 | target memory world. */ |
| 993 | |
| 994 | /* Mask count to get odd byte, odd halfword, and odd word out of the |
| 995 | way. We can then perform doubleword transfers to and from the |
| 996 | simulator memory for optimum performance. */ |
| 997 | if (index && (index & 1)) { |
| 998 | uword64 paddr; |
| 999 | int cca; |
| 1000 | if (AddressTranslation(vaddr,isDATA,isSTORE,&paddr,&cca,isTARGET,isRAW)) { |
| 1001 | uword64 value = ((uword64)(*buffer++)); |
| 1002 | StoreMemory(cca,AccessLength_BYTE,value,paddr,vaddr,isRAW); |
| 1003 | } |
| 1004 | vaddr++; |
| 1005 | index &= ~1; /* logical operations usually quicker than arithmetic on RISC systems */ |
| 1006 | } |
| 1007 | if (index && (index & 2)) { |
| 1008 | uword64 paddr; |
| 1009 | int cca; |
| 1010 | if (AddressTranslation(vaddr,isDATA,isSTORE,&paddr,&cca,isTARGET,isRAW)) { |
| 1011 | uword64 value; |
| 1012 | /* We need to perform the following magic to ensure that that |
| 1013 | bytes are written into same byte positions in the target memory |
| 1014 | world, regardless of the endianness of the host. */ |
| 1015 | if (BigEndianMem) { |
| 1016 | value = ((uword64)(*buffer++) << 8); |
| 1017 | value |= ((uword64)(*buffer++) << 0); |
| 1018 | } else { |
| 1019 | value = ((uword64)(*buffer++) << 0); |
| 1020 | value |= ((uword64)(*buffer++) << 8); |
| 1021 | } |
| 1022 | StoreMemory(cca,AccessLength_HALFWORD,value,paddr,vaddr,isRAW); |
| 1023 | } |
| 1024 | vaddr += 2; |
| 1025 | index &= ~2; |
| 1026 | } |
| 1027 | if (index && (index & 4)) { |
| 1028 | uword64 paddr; |
| 1029 | int cca; |
| 1030 | if (AddressTranslation(vaddr,isDATA,isSTORE,&paddr,&cca,isTARGET,isRAW)) { |
| 1031 | uword64 value; |
| 1032 | if (BigEndianMem) { |
| 1033 | value = ((uword64)(*buffer++) << 24); |
| 1034 | value |= ((uword64)(*buffer++) << 16); |
| 1035 | value |= ((uword64)(*buffer++) << 8); |
| 1036 | value |= ((uword64)(*buffer++) << 0); |
| 1037 | } else { |
| 1038 | value = ((uword64)(*buffer++) << 0); |
| 1039 | value |= ((uword64)(*buffer++) << 8); |
| 1040 | value |= ((uword64)(*buffer++) << 16); |
| 1041 | value |= ((uword64)(*buffer++) << 24); |
| 1042 | } |
| 1043 | StoreMemory(cca,AccessLength_WORD,value,paddr,vaddr,isRAW); |
| 1044 | } |
| 1045 | vaddr += 4; |
| 1046 | index &= ~4; |
| 1047 | } |
| 1048 | for (;index; index -= 8) { |
| 1049 | uword64 paddr; |
| 1050 | int cca; |
| 1051 | if (AddressTranslation(vaddr,isDATA,isSTORE,&paddr,&cca,isTARGET,isRAW)) { |
| 1052 | uword64 value; |
| 1053 | if (BigEndianMem) { |
| 1054 | value = ((uword64)(*buffer++) << 56); |
| 1055 | value |= ((uword64)(*buffer++) << 48); |
| 1056 | value |= ((uword64)(*buffer++) << 40); |
| 1057 | value |= ((uword64)(*buffer++) << 32); |
| 1058 | value |= ((uword64)(*buffer++) << 24); |
| 1059 | value |= ((uword64)(*buffer++) << 16); |
| 1060 | value |= ((uword64)(*buffer++) << 8); |
| 1061 | value |= ((uword64)(*buffer++) << 0); |
| 1062 | } else { |
| 1063 | value = ((uword64)(*buffer++) << 0); |
| 1064 | value |= ((uword64)(*buffer++) << 8); |
| 1065 | value |= ((uword64)(*buffer++) << 16); |
| 1066 | value |= ((uword64)(*buffer++) << 24); |
| 1067 | value |= ((uword64)(*buffer++) << 32); |
| 1068 | value |= ((uword64)(*buffer++) << 40); |
| 1069 | value |= ((uword64)(*buffer++) << 48); |
| 1070 | value |= ((uword64)(*buffer++) << 56); |
| 1071 | } |
| 1072 | StoreMemory(cca,AccessLength_DOUBLEWORD,value,paddr,vaddr,isRAW); |
| 1073 | } |
| 1074 | vaddr += 8; |
| 1075 | } |
| 1076 | |
| 1077 | return(size); |
| 1078 | } |
| 1079 | |
| 1080 | int |
| 1081 | sim_read (addr,buffer,size) |
| 1082 | SIM_ADDR addr; |
| 1083 | unsigned char *buffer; |
| 1084 | int size; |
| 1085 | { |
| 1086 | int index; |
| 1087 | |
| 1088 | /* Return the number of bytes read, or zero if error. */ |
| 1089 | #ifdef DEBUG |
| 1090 | callback->printf_filtered(callback,"sim_read(0x%08X%08X,buffer,%d);\n",WORD64HI(addr),WORD64LO(addr),size); |
| 1091 | #endif /* DEBUG */ |
| 1092 | |
| 1093 | /* TODO: Perform same optimisation as the sim_write() code |
| 1094 | above. NOTE: This will require a bit more work since we will need |
| 1095 | to ensure that the source physical address is doubleword aligned |
| 1096 | before, and then deal with trailing bytes. */ |
| 1097 | for (index = 0; (index < size); index++) { |
| 1098 | uword64 vaddr,paddr,value; |
| 1099 | int cca; |
| 1100 | vaddr = (uword64)addr + index; |
| 1101 | if (AddressTranslation(vaddr,isDATA,isLOAD,&paddr,&cca,isTARGET,isRAW)) { |
| 1102 | value = LoadMemory(cca,AccessLength_BYTE,paddr,vaddr,isDATA,isRAW); |
| 1103 | buffer[index] = (unsigned char)(value&0xFF); |
| 1104 | } else |
| 1105 | break; |
| 1106 | } |
| 1107 | |
| 1108 | return(index); |
| 1109 | } |
| 1110 | |
| 1111 | void |
| 1112 | sim_store_register (rn,memory) |
| 1113 | int rn; |
| 1114 | unsigned char *memory; |
| 1115 | { |
| 1116 | #ifdef DEBUG |
| 1117 | callback->printf_filtered(callback,"sim_store_register(%d,*memory=0x%08X%08X);\n",rn,*((unsigned int *)memory),*((unsigned int *)(memory + 4))); |
| 1118 | #endif /* DEBUG */ |
| 1119 | |
| 1120 | /* Unfortunately this suffers from the same problem as the register |
| 1121 | numbering one. We need to know what the width of each logical |
| 1122 | register number is for the architecture being simulated. */ |
| 1123 | if (register_widths[rn] == 0) |
| 1124 | sim_warning("Invalid register width for %d (register store ignored)",rn); |
| 1125 | else { |
| 1126 | if (register_widths[rn] == 32) |
| 1127 | registers[rn] = host_read_word(memory); |
| 1128 | else |
| 1129 | registers[rn] = host_read_long(memory); |
| 1130 | } |
| 1131 | |
| 1132 | return; |
| 1133 | } |
| 1134 | |
| 1135 | void |
| 1136 | sim_fetch_register (rn,memory) |
| 1137 | int rn; |
| 1138 | unsigned char *memory; |
| 1139 | { |
| 1140 | #ifdef DEBUG |
| 1141 | callback->printf_filtered(callback,"sim_fetch_register(%d=0x%08X%08X,mem) : place simulator registers into memory\n",rn,WORD64HI(registers[rn]),WORD64LO(registers[rn])); |
| 1142 | #endif /* DEBUG */ |
| 1143 | |
| 1144 | if (register_widths[rn] == 0) |
| 1145 | sim_warning("Invalid register width for %d (register fetch ignored)",rn); |
| 1146 | else { |
| 1147 | if (register_widths[rn] == 32) |
| 1148 | *((unsigned int *)memory) = host_swap_word(registers[rn] & 0xFFFFFFFF); |
| 1149 | else /* 64bit register */ |
| 1150 | *((uword64 *)memory) = host_swap_long(registers[rn]); |
| 1151 | } |
| 1152 | return; |
| 1153 | } |
| 1154 | |
| 1155 | void |
| 1156 | sim_stop_reason (reason,sigrc) |
| 1157 | enum sim_stop *reason; |
| 1158 | int *sigrc; |
| 1159 | { |
| 1160 | /* We can have "*reason = {sim_exited, sim_stopped, sim_signalled}", so |
| 1161 | sim_exited *sigrc = argument to exit() |
| 1162 | sim_stopped *sigrc = exception number |
| 1163 | sim_signalled *sigrc = signal number |
| 1164 | */ |
| 1165 | if (state & simEXCEPTION) { |
| 1166 | /* If "sim_signalled" is used, GDB expects normal SIGNAL numbers, |
| 1167 | and not the MIPS specific exception codes. */ |
| 1168 | #if 1 |
| 1169 | /* For some reason, sending GDB a sim_signalled reason cause it to |
| 1170 | terminate out. */ |
| 1171 | *reason = sim_stopped; |
| 1172 | #else |
| 1173 | *reason = sim_signalled; |
| 1174 | #endif |
| 1175 | switch ((CAUSE >> 2) & 0x1F) { |
| 1176 | case Interrupt: |
| 1177 | *sigrc = SIGINT; /* wrong type of interrupt, but it will do for the moment */ |
| 1178 | break; |
| 1179 | |
| 1180 | case TLBModification: |
| 1181 | case TLBLoad: |
| 1182 | case TLBStore: |
| 1183 | case AddressLoad: |
| 1184 | case AddressStore: |
| 1185 | case InstructionFetch: |
| 1186 | case DataReference: |
| 1187 | *sigrc = SIGBUS; |
| 1188 | break; |
| 1189 | |
| 1190 | case ReservedInstruction: |
| 1191 | case CoProcessorUnusable: |
| 1192 | *sigrc = SIGILL; |
| 1193 | break; |
| 1194 | |
| 1195 | case IntegerOverflow: |
| 1196 | case FPE: |
| 1197 | *sigrc = SIGFPE; |
| 1198 | break; |
| 1199 | |
| 1200 | case Trap: |
| 1201 | case Watch: |
| 1202 | case SystemCall: |
| 1203 | case BreakPoint: |
| 1204 | *sigrc = SIGTRAP; |
| 1205 | break; |
| 1206 | |
| 1207 | default : /* Unknown internal exception */ |
| 1208 | *sigrc = SIGQUIT; |
| 1209 | break; |
| 1210 | } |
| 1211 | } else if (state & simEXIT) { |
| 1212 | #if 0 |
| 1213 | printf("DBG: simEXIT (%d)\n",rcexit); |
| 1214 | #endif |
| 1215 | *reason = sim_exited; |
| 1216 | *sigrc = rcexit; |
| 1217 | } else { /* assume single-stepping */ |
| 1218 | *reason = sim_stopped; |
| 1219 | *sigrc = SIGTRAP; |
| 1220 | } |
| 1221 | state &= ~(simEXCEPTION | simEXIT); |
| 1222 | return; |
| 1223 | } |
| 1224 | |
| 1225 | void |
| 1226 | sim_info (verbose) |
| 1227 | int verbose; |
| 1228 | { |
| 1229 | /* Accessed from the GDB "info files" command: */ |
| 1230 | |
| 1231 | callback->printf_filtered(callback,"MIPS %d-bit simulator\n",(PROCESSOR_64BIT ? 64 : 32)); |
| 1232 | |
| 1233 | callback->printf_filtered(callback,"%s endian memory model\n",(BigEndianMem ? "Big" : "Little")); |
| 1234 | |
| 1235 | callback->printf_filtered(callback,"0x%08X bytes of memory at 0x%08X%08X\n",(unsigned int)membank_size,WORD64HI(membank_base),WORD64LO(membank_base)); |
| 1236 | |
| 1237 | #if !defined(FASTSIM) |
| 1238 | if (instruction_fetch_overflow != 0) |
| 1239 | callback->printf_filtered(callback,"Instruction fetches = 0x%08X%08X\n",instruction_fetch_overflow,instruction_fetches); |
| 1240 | else |
| 1241 | callback->printf_filtered(callback,"Instruction fetches = %d\n",instruction_fetches); |
| 1242 | callback->printf_filtered(callback,"Pipeline ticks = %d\n",pipeline_ticks); |
| 1243 | /* It would be a useful feature, if when performing multi-cycle |
| 1244 | simulations (rather than single-stepping) we keep the start and |
| 1245 | end times of the execution, so that we can give a performance |
| 1246 | figure for the simulator. */ |
| 1247 | #endif /* !FASTSIM */ |
| 1248 | |
| 1249 | /* print information pertaining to MIPS ISA and architecture being simulated */ |
| 1250 | /* things that may be interesting */ |
| 1251 | /* instructions executed - if available */ |
| 1252 | /* cycles executed - if available */ |
| 1253 | /* pipeline stalls - if available */ |
| 1254 | /* virtual time taken */ |
| 1255 | /* profiling size */ |
| 1256 | /* profiling frequency */ |
| 1257 | /* profile minpc */ |
| 1258 | /* profile maxpc */ |
| 1259 | |
| 1260 | return; |
| 1261 | } |
| 1262 | |
| 1263 | int |
| 1264 | sim_load (prog,from_tty) |
| 1265 | char *prog; |
| 1266 | int from_tty; |
| 1267 | { |
| 1268 | /* Return non-zero if the caller should handle the load. Zero if |
| 1269 | we have loaded the image. */ |
| 1270 | return(-1); |
| 1271 | } |
| 1272 | |
| 1273 | void |
| 1274 | sim_create_inferior (start_address,argv,env) |
| 1275 | SIM_ADDR start_address; |
| 1276 | char **argv; |
| 1277 | char **env; |
| 1278 | { |
| 1279 | #ifdef DEBUG |
| 1280 | printf("DBG: sim_create_inferior entered: start_address = 0x%08X\n",start_address); |
| 1281 | #endif /* DEBUG */ |
| 1282 | |
| 1283 | /* Prepare to execute the program to be simulated */ |
| 1284 | /* argv and env are NULL terminated lists of pointers */ |
| 1285 | |
| 1286 | #if 1 |
| 1287 | PC = (uword64)start_address; |
| 1288 | #else |
| 1289 | /* TODO: Sort this properly. SIM_ADDR may already be a 64bit value: */ |
| 1290 | PC = SIGNEXTEND(start_address,32); |
| 1291 | #endif |
| 1292 | /* NOTE: GDB normally sets the PC explicitly. However, this call is |
| 1293 | used by other clients of the simulator. */ |
| 1294 | |
| 1295 | if (argv || env) { |
| 1296 | #if 0 /* def DEBUG */ |
| 1297 | callback->printf_filtered(callback,"sim_create_inferior() : passed arguments ignored\n"); |
| 1298 | { |
| 1299 | char **cptr; |
| 1300 | for (cptr = argv; (cptr && *cptr); cptr++) |
| 1301 | printf("DBG: arg \"%s\"\n",*cptr); |
| 1302 | } |
| 1303 | #endif /* DEBUG */ |
| 1304 | /* We should really place the argv slot values into the argument |
| 1305 | registers, and onto the stack as required. However, this |
| 1306 | assumes that we have a stack defined, which is not necessarily |
| 1307 | true at the moment. */ |
| 1308 | } |
| 1309 | |
| 1310 | return; |
| 1311 | } |
| 1312 | |
| 1313 | void |
| 1314 | sim_kill () |
| 1315 | { |
| 1316 | #if 1 |
| 1317 | /* This routine should be for terminating any existing simulation |
| 1318 | thread. Since we are single-threaded only at the moment, this is |
| 1319 | not an issue. It should *NOT* be used to terminate the |
| 1320 | simulator. */ |
| 1321 | #else /* do *NOT* call sim_close */ |
| 1322 | sim_close(1); /* Do not hang on errors */ |
| 1323 | /* This would also be the point where any memory mapped areas used |
| 1324 | by the simulator should be released. */ |
| 1325 | #endif |
| 1326 | return; |
| 1327 | } |
| 1328 | |
| 1329 | int |
| 1330 | sim_get_quit_code () |
| 1331 | { |
| 1332 | /* The standard MIPS PCS (Procedure Calling Standard) uses V0(r2) as |
| 1333 | the function return value. However, it may be more correct for |
| 1334 | this to return the argument to the exit() function (if |
| 1335 | called). */ |
| 1336 | return(V0); |
| 1337 | } |
| 1338 | |
| 1339 | void |
| 1340 | sim_set_callbacks (p) |
| 1341 | host_callback *p; |
| 1342 | { |
| 1343 | callback = p; |
| 1344 | return; |
| 1345 | } |
| 1346 | |
| 1347 | typedef enum {e_terminate,e_help,e_setmemsize,e_reset} e_cmds; |
| 1348 | |
| 1349 | static struct t_sim_command { |
| 1350 | e_cmds id; |
| 1351 | const char *name; |
| 1352 | const char *help; |
| 1353 | } sim_commands[] = { |
| 1354 | {e_help, "help", ": Show MIPS simulator private commands"}, |
| 1355 | {e_setmemsize,"set-memory-size","<n> : Specify amount of memory simulated"}, |
| 1356 | {e_reset, "reset-system", ": Reset the simulated processor"}, |
| 1357 | {e_terminate, NULL} |
| 1358 | }; |
| 1359 | |
| 1360 | void |
| 1361 | sim_do_command (cmd) |
| 1362 | char *cmd; |
| 1363 | { |
| 1364 | struct t_sim_command *cptr; |
| 1365 | |
| 1366 | if (callback == NULL) { |
| 1367 | fprintf(stderr,"Simulator not enabled: \"target sim\" should be used to activate\n"); |
| 1368 | return; |
| 1369 | } |
| 1370 | |
| 1371 | if (!(cmd && *cmd != '\0')) |
| 1372 | cmd = "help"; |
| 1373 | |
| 1374 | /* NOTE: Accessed from the GDB "sim" commmand: */ |
| 1375 | for (cptr = sim_commands; cptr && cptr->name; cptr++) |
| 1376 | if (strncmp(cmd,cptr->name,strlen(cptr->name)) == 0) { |
| 1377 | cmd += strlen(cptr->name); |
| 1378 | switch (cptr->id) { |
| 1379 | case e_help: /* no arguments */ |
| 1380 | { /* no arguments */ |
| 1381 | struct t_sim_command *lptr; |
| 1382 | callback->printf_filtered(callback,"List of MIPS simulator commands:\n"); |
| 1383 | for (lptr = sim_commands; lptr->name; lptr++) |
| 1384 | callback->printf_filtered(callback,"%s %s\n",lptr->name,lptr->help); |
| 1385 | } |
| 1386 | break; |
| 1387 | |
| 1388 | case e_setmemsize: /* memory size argument */ |
| 1389 | { |
| 1390 | unsigned int newsize = (unsigned int)getnum(cmd); |
| 1391 | sim_size(newsize); |
| 1392 | } |
| 1393 | break; |
| 1394 | |
| 1395 | case e_reset: /* no arguments */ |
| 1396 | ColdReset(); |
| 1397 | /* NOTE: See the comments in sim_open() relating to device |
| 1398 | initialisation. */ |
| 1399 | break; |
| 1400 | |
| 1401 | default: |
| 1402 | callback->printf_filtered(callback,"FATAL: Matched \"%s\", but failed to match command id %d.\n",cmd,cptr->id); |
| 1403 | break; |
| 1404 | } |
| 1405 | break; |
| 1406 | } |
| 1407 | |
| 1408 | if (!(cptr->name)) |
| 1409 | callback->printf_filtered(callback,"Error: \"%s\" is not a valid MIPS simulator command.\n",cmd); |
| 1410 | |
| 1411 | return; |
| 1412 | } |
| 1413 | |
| 1414 | /*---------------------------------------------------------------------------*/ |
| 1415 | /* NOTE: The following routines do not seem to be used by GDB at the |
| 1416 | moment. However, they may be useful to the standalone simulator |
| 1417 | world. */ |
| 1418 | |
| 1419 | |
| 1420 | /* The profiling format is described in the "gmon_out.h" header file */ |
| 1421 | void |
| 1422 | sim_set_profile (n) |
| 1423 | int n; |
| 1424 | { |
| 1425 | #if defined(PROFILE) |
| 1426 | profile_frequency = n; |
| 1427 | state |= simPROFILE; |
| 1428 | #endif /* PROFILE */ |
| 1429 | return; |
| 1430 | } |
| 1431 | |
| 1432 | void |
| 1433 | sim_set_profile_size (n) |
| 1434 | int n; |
| 1435 | { |
| 1436 | #if defined(PROFILE) |
| 1437 | if (state & simPROFILE) { |
| 1438 | int bsize; |
| 1439 | |
| 1440 | /* Since we KNOW that the memory banks are a power-of-2 in size: */ |
| 1441 | profile_nsamples = power2(n); |
| 1442 | profile_minpc = membank_base; |
| 1443 | profile_maxpc = (membank_base + membank_size); |
| 1444 | |
| 1445 | /* Just in-case we are sampling every address: NOTE: The shift |
| 1446 | right of 2 is because we only have word-aligned PC addresses. */ |
| 1447 | if (profile_nsamples > (membank_size >> 2)) |
| 1448 | profile_nsamples = (membank_size >> 2); |
| 1449 | |
| 1450 | /* Since we are dealing with power-of-2 values: */ |
| 1451 | profile_shift = (((membank_size >> 2) / profile_nsamples) - 1); |
| 1452 | |
| 1453 | bsize = (profile_nsamples * sizeof(unsigned short)); |
| 1454 | if (profile_hist == NULL) |
| 1455 | profile_hist = (unsigned short *)calloc(64,(bsize / 64)); |
| 1456 | else |
| 1457 | profile_hist = (unsigned short *)realloc(profile_hist,bsize); |
| 1458 | if (profile_hist == NULL) { |
| 1459 | sim_warning("Failed to allocate VM for profiling buffer (0x%08X bytes)",bsize); |
| 1460 | state &= ~simPROFILE; |
| 1461 | } |
| 1462 | } |
| 1463 | #endif /* PROFILE */ |
| 1464 | |
| 1465 | return; |
| 1466 | } |
| 1467 | |
| 1468 | void |
| 1469 | sim_size(newsize) |
| 1470 | unsigned int newsize; |
| 1471 | { |
| 1472 | char *new; |
| 1473 | /* Used by "run", and internally, to set the simulated memory size */ |
| 1474 | if (newsize == 0) { |
| 1475 | callback->printf_filtered(callback,"Zero not valid: Memory size still 0x%08X bytes\n",membank_size); |
| 1476 | return; |
| 1477 | } |
| 1478 | newsize = power2(newsize); |
| 1479 | if (membank == NULL) |
| 1480 | new = (char *)calloc(64,(membank_size / 64)); |
| 1481 | else |
| 1482 | new = (char *)realloc(membank,newsize); |
| 1483 | if (new == NULL) { |
| 1484 | if (membank == NULL) |
| 1485 | sim_error("Not enough VM for simulation memory of 0x%08X bytes",membank_size); |
| 1486 | else |
| 1487 | sim_warning("Failed to resize memory (still 0x%08X bytes)",membank_size); |
| 1488 | } else { |
| 1489 | membank_size = (unsigned)newsize; |
| 1490 | membank = new; |
| 1491 | #if defined(PROFILE) |
| 1492 | /* Ensure that we sample across the new memory range */ |
| 1493 | sim_set_profile_size(profile_nsamples); |
| 1494 | #endif /* PROFILE */ |
| 1495 | } |
| 1496 | |
| 1497 | return; |
| 1498 | } |
| 1499 | |
| 1500 | int |
| 1501 | sim_trace() |
| 1502 | { |
| 1503 | /* This routine is called by the "run" program, when detailed |
| 1504 | execution information is required. Rather than executing a single |
| 1505 | instruction, and looping around externally... we just start |
| 1506 | simulating, returning TRUE when the simulator stops (for whatever |
| 1507 | reason). */ |
| 1508 | |
| 1509 | #if defined(TRACE) |
| 1510 | /* Ensure tracing is enabled, if available */ |
| 1511 | if (tracefh != NULL) |
| 1512 | state |= simTRACE; |
| 1513 | #endif /* TRACE */ |
| 1514 | |
| 1515 | state &= ~(simSTOP | simSTEP); /* execute until event */ |
| 1516 | state |= (simHALTEX | simHALTIN); /* treat interrupt event as exception */ |
| 1517 | /* Start executing instructions from the current state (set |
| 1518 | explicitly by register updates, or by sim_create_inferior): */ |
| 1519 | simulate(); |
| 1520 | |
| 1521 | return(1); |
| 1522 | } |
| 1523 | |
| 1524 | /*---------------------------------------------------------------------------*/ |
| 1525 | /*-- Private simulator support interface ------------------------------------*/ |
| 1526 | /*---------------------------------------------------------------------------*/ |
| 1527 | |
| 1528 | /* Simple monitor interface (currently setup for the IDT and PMON monitors) */ |
| 1529 | static void |
| 1530 | sim_monitor(reason) |
| 1531 | unsigned int reason; |
| 1532 | { |
| 1533 | /* The IDT monitor actually allows two instructions per vector |
| 1534 | slot. However, the simulator currently causes a trap on each |
| 1535 | individual instruction. We cheat, and lose the bottom bit. */ |
| 1536 | reason >>= 1; |
| 1537 | |
| 1538 | /* The following callback functions are available, however the |
| 1539 | monitor we are simulating does not make use of them: get_errno, |
| 1540 | isatty, lseek, rename, system, time and unlink */ |
| 1541 | switch (reason) { |
| 1542 | case 6: /* int open(char *path,int flags) */ |
| 1543 | { |
| 1544 | const char *ptr; |
| 1545 | uword64 paddr; |
| 1546 | int cca; |
| 1547 | if (AddressTranslation(A0,isDATA,isLOAD,&paddr,&cca,isHOST,isREAL)) |
| 1548 | V0 = callback->open(callback,(char *)((int)paddr),(int)A1); |
| 1549 | else |
| 1550 | sim_error("Attempt to pass pointer that does not reference simulated memory"); |
| 1551 | } |
| 1552 | break; |
| 1553 | |
| 1554 | case 7: /* int read(int file,char *ptr,int len) */ |
| 1555 | { |
| 1556 | const char *ptr; |
| 1557 | uword64 paddr; |
| 1558 | int cca; |
| 1559 | if (AddressTranslation(A1,isDATA,isLOAD,&paddr,&cca,isHOST,isREAL)) |
| 1560 | V0 = callback->read(callback,(int)A0,(char *)((int)paddr),(int)A2); |
| 1561 | else |
| 1562 | sim_error("Attempt to pass pointer that does not reference simulated memory"); |
| 1563 | } |
| 1564 | break; |
| 1565 | |
| 1566 | case 8: /* int write(int file,char *ptr,int len) */ |
| 1567 | { |
| 1568 | const char *ptr; |
| 1569 | uword64 paddr; |
| 1570 | int cca; |
| 1571 | if (AddressTranslation(A1,isDATA,isLOAD,&paddr,&cca,isHOST,isREAL)) |
| 1572 | V0 = callback->write(callback,(int)A0,(const char *)((int)paddr),(int)A2); |
| 1573 | else |
| 1574 | sim_error("Attempt to pass pointer that does not reference simulated memory"); |
| 1575 | } |
| 1576 | break; |
| 1577 | |
| 1578 | case 10: /* int close(int file) */ |
| 1579 | V0 = callback->close(callback,(int)A0); |
| 1580 | break; |
| 1581 | |
| 1582 | case 11: /* char inbyte(void) */ |
| 1583 | { |
| 1584 | char tmp; |
| 1585 | if (callback->read_stdin(callback,&tmp,sizeof(char)) != sizeof(char)) { |
| 1586 | sim_error("Invalid return from character read"); |
| 1587 | V0 = -1; |
| 1588 | } |
| 1589 | else |
| 1590 | V0 = tmp; |
| 1591 | } |
| 1592 | break; |
| 1593 | |
| 1594 | case 12: /* void outbyte(char chr) : write a byte to "stdout" */ |
| 1595 | { |
| 1596 | char tmp = (char)(A0 & 0xFF); |
| 1597 | callback->write_stdout(callback,&tmp,sizeof(char)); |
| 1598 | } |
| 1599 | break; |
| 1600 | |
| 1601 | case 17: /* void _exit() */ |
| 1602 | sim_warning("sim_monitor(17): _exit(int reason) to be coded"); |
| 1603 | state |= (simSTOP | simEXIT); /* stop executing code */ |
| 1604 | rcexit = (unsigned int)(A0 & 0xFFFFFFFF); |
| 1605 | break; |
| 1606 | |
| 1607 | case 55: /* void get_mem_info(unsigned int *ptr) */ |
| 1608 | /* in: A0 = pointer to three word memory location */ |
| 1609 | /* out: [A0 + 0] = size */ |
| 1610 | /* [A0 + 4] = instruction cache size */ |
| 1611 | /* [A0 + 8] = data cache size */ |
| 1612 | { |
| 1613 | uword64 vaddr = A0; |
| 1614 | uword64 paddr, value; |
| 1615 | int cca; |
| 1616 | int failed = 0; |
| 1617 | |
| 1618 | /* NOTE: We use RAW memory writes here, but since we are not |
| 1619 | gathering statistics for the monitor calls we are simulating, |
| 1620 | it is not an issue. */ |
| 1621 | |
| 1622 | /* Memory size */ |
| 1623 | if (AddressTranslation(vaddr,isDATA,isSTORE,&paddr,&cca,isTARGET,isREAL)) { |
| 1624 | value = (uword64)membank_size; |
| 1625 | StoreMemory(cca,AccessLength_WORD,value,paddr,vaddr,isRAW); |
| 1626 | /* We re-do the address translations, in-case the block |
| 1627 | overlaps a memory boundary: */ |
| 1628 | value = 0; |
| 1629 | vaddr += (AccessLength_WORD + 1); |
| 1630 | if (AddressTranslation(vaddr,isDATA,isSTORE,&paddr,&cca,isTARGET,isREAL)) { |
| 1631 | StoreMemory(cca,AccessLength_WORD,value,paddr,vaddr,isRAW); |
| 1632 | vaddr += (AccessLength_WORD + 1); |
| 1633 | if (AddressTranslation(vaddr,isDATA,isSTORE,&paddr,&cca,isTARGET,isREAL)) |
| 1634 | StoreMemory(cca,AccessLength_WORD,value,paddr,vaddr,isRAW); |
| 1635 | else |
| 1636 | failed = -1; |
| 1637 | } else |
| 1638 | failed = -1; |
| 1639 | } else |
| 1640 | failed = -1; |
| 1641 | |
| 1642 | if (failed) |
| 1643 | sim_error("Invalid pointer passed into monitor call"); |
| 1644 | } |
| 1645 | break; |
| 1646 | |
| 1647 | case 158 : /* PMON printf */ |
| 1648 | /* in: A0 = pointer to format string */ |
| 1649 | /* A1 = optional argument 1 */ |
| 1650 | /* A2 = optional argument 2 */ |
| 1651 | /* A3 = optional argument 3 */ |
| 1652 | /* out: void */ |
| 1653 | /* The following is based on the PMON printf source */ |
| 1654 | { |
| 1655 | uword64 paddr; |
| 1656 | int cca; |
| 1657 | /* This isn't the quickest way, since we call the host print |
| 1658 | routine for every character almost. But it does avoid |
| 1659 | having to allocate and manage a temporary string buffer. */ |
| 1660 | if (AddressTranslation(A0,isDATA,isLOAD,&paddr,&cca,isHOST,isREAL)) { |
| 1661 | char *s = (char *)((int)paddr); |
| 1662 | ut_reg *ap = &A1; /* 1st argument */ |
| 1663 | /* TODO: Include check that we only use three arguments (A1, A2 and A3) */ |
| 1664 | for (; *s;) { |
| 1665 | if (*s == '%') { |
| 1666 | char tmp[40]; |
| 1667 | enum {FMT_RJUST, FMT_LJUST, FMT_RJUST0, FMT_CENTER} fmt = FMT_RJUST; |
| 1668 | int width = 0, trunc = 0, haddot = 0, longlong = 0; |
| 1669 | int base = 10; |
| 1670 | s++; |
| 1671 | for (; *s; s++) { |
| 1672 | if (strchr ("dobxXulscefg%", *s)) |
| 1673 | break; |
| 1674 | else if (*s == '-') |
| 1675 | fmt = FMT_LJUST; |
| 1676 | else if (*s == '0') |
| 1677 | fmt = FMT_RJUST0; |
| 1678 | else if (*s == '~') |
| 1679 | fmt = FMT_CENTER; |
| 1680 | else if (*s == '*') { |
| 1681 | if (haddot) |
| 1682 | trunc = (int)*ap++; |
| 1683 | else |
| 1684 | width = (int)*ap++; |
| 1685 | } else if (*s >= '1' && *s <= '9') { |
| 1686 | char *t; |
| 1687 | unsigned int n; |
| 1688 | for (t = s; isdigit (*s); s++); |
| 1689 | strncpy (tmp, t, s - t); |
| 1690 | tmp[s - t] = '\0'; |
| 1691 | n = (unsigned int)strtol(tmp,NULL,10); |
| 1692 | if (haddot) |
| 1693 | trunc = n; |
| 1694 | else |
| 1695 | width = n; |
| 1696 | s--; |
| 1697 | } else if (*s == '.') |
| 1698 | haddot = 1; |
| 1699 | } |
| 1700 | if (*s == '%') { |
| 1701 | callback->printf_filtered(callback,"%%"); |
| 1702 | } else if (*s == 's') { |
| 1703 | if ((int)*ap != 0) { |
| 1704 | if (AddressTranslation(*ap++,isDATA,isLOAD,&paddr,&cca,isHOST,isREAL)) { |
| 1705 | char *p = (char *)((int)paddr);; |
| 1706 | callback->printf_filtered(callback,p); |
| 1707 | } else { |
| 1708 | ap++; |
| 1709 | sim_error("Attempt to pass pointer that does not reference simulated memory"); |
| 1710 | } |
| 1711 | } |
| 1712 | else |
| 1713 | callback->printf_filtered(callback,"(null)"); |
| 1714 | } else if (*s == 'c') { |
| 1715 | int n = (int)*ap++; |
| 1716 | callback->printf_filtered(callback,"%c",n); |
| 1717 | } else { |
| 1718 | if (*s == 'l') { |
| 1719 | if (*++s == 'l') { |
| 1720 | longlong = 1; |
| 1721 | ++s; |
| 1722 | } |
| 1723 | } |
| 1724 | if (strchr ("dobxXu", *s)) { |
| 1725 | long long lv = (long long)*ap++; |
| 1726 | if (*s == 'b') |
| 1727 | callback->printf_filtered(callback,"<binary not supported>"); |
| 1728 | else { |
| 1729 | sprintf(tmp,"%%%s%c",longlong ? "ll" : "",*s); |
| 1730 | if (longlong) |
| 1731 | callback->printf_filtered(callback,tmp,lv); |
| 1732 | else |
| 1733 | callback->printf_filtered(callback,tmp,(int)lv); |
| 1734 | } |
| 1735 | } else if (strchr ("eEfgG", *s)) { |
| 1736 | double dbl = (double)*ap++; |
| 1737 | sprintf(tmp,"%%%d.%d%c",width,trunc,*s); |
| 1738 | callback->printf_filtered(callback,tmp,dbl); |
| 1739 | trunc = 0; |
| 1740 | } |
| 1741 | } |
| 1742 | s++; |
| 1743 | } else |
| 1744 | callback->printf_filtered(callback,"%c",*s++); |
| 1745 | } |
| 1746 | } else |
| 1747 | sim_error("Attempt to pass pointer that does not reference simulated memory"); |
| 1748 | } |
| 1749 | break; |
| 1750 | |
| 1751 | default: |
| 1752 | sim_warning("TODO: sim_monitor(%d) : PC = 0x%08X%08X",reason,WORD64HI(IPC),WORD64LO(IPC)); |
| 1753 | sim_warning("(Arguments : A0 = 0x%08X%08X : A1 = 0x%08X%08X : A2 = 0x%08X%08X : A3 = 0x%08X%08X)",WORD64HI(A0),WORD64LO(A0),WORD64HI(A1),WORD64LO(A1),WORD64HI(A2),WORD64LO(A2),WORD64HI(A3),WORD64LO(A3)); |
| 1754 | break; |
| 1755 | } |
| 1756 | return; |
| 1757 | } |
| 1758 | |
| 1759 | void |
| 1760 | sim_warning(fmt) |
| 1761 | char *fmt; |
| 1762 | { |
| 1763 | va_list ap; |
| 1764 | va_start(ap,fmt); |
| 1765 | if (logfh != NULL) { |
| 1766 | #if 1 |
| 1767 | fprintf(logfh,"SIM Warning: "); |
| 1768 | fprintf(logfh,fmt,ap); |
| 1769 | fprintf(logfh,"\n"); |
| 1770 | #else /* we should provide a method of routing log messages to the simulator output stream */ |
| 1771 | callback->printf_filtered(callback,"SIM Warning: "); |
| 1772 | callback->printf_filtered(callback,fmt,ap); |
| 1773 | #endif |
| 1774 | } |
| 1775 | va_end(ap); |
| 1776 | SignalException(SimulatorFault,""); |
| 1777 | return; |
| 1778 | } |
| 1779 | |
| 1780 | void |
| 1781 | sim_error(fmt) |
| 1782 | char *fmt; |
| 1783 | { |
| 1784 | va_list ap; |
| 1785 | va_start(ap,fmt); |
| 1786 | callback->printf_filtered(callback,"SIM Error: "); |
| 1787 | callback->printf_filtered(callback,fmt,ap); |
| 1788 | va_end(ap); |
| 1789 | SignalException(SimulatorFault,""); |
| 1790 | return; |
| 1791 | } |
| 1792 | |
| 1793 | static unsigned int |
| 1794 | power2(value) |
| 1795 | unsigned int value; |
| 1796 | { |
| 1797 | int loop,tmp; |
| 1798 | |
| 1799 | /* Round *UP* to the nearest power-of-2 if not already one */ |
| 1800 | if (value != (value & ~(value - 1))) { |
| 1801 | for (tmp = value, loop = 0; (tmp != 0); loop++) |
| 1802 | tmp >>= 1; |
| 1803 | value = (1 << loop); |
| 1804 | } |
| 1805 | |
| 1806 | return(value); |
| 1807 | } |
| 1808 | |
| 1809 | static long |
| 1810 | getnum(value) |
| 1811 | char *value; |
| 1812 | { |
| 1813 | long num; |
| 1814 | char *end; |
| 1815 | |
| 1816 | num = strtol(value,&end,10); |
| 1817 | if (end == value) |
| 1818 | callback->printf_filtered(callback,"Warning: Invalid number \"%s\" ignored, using zero\n",value); |
| 1819 | else { |
| 1820 | if (*end && ((tolower(*end) == 'k') || (tolower(*end) == 'm'))) { |
| 1821 | if (tolower(*end) == 'k') |
| 1822 | num *= (1 << 10); |
| 1823 | else |
| 1824 | num *= (1 << 20); |
| 1825 | end++; |
| 1826 | } |
| 1827 | if (*end) |
| 1828 | callback->printf_filtered(callback,"Warning: Spurious characters \"%s\" at end of number ignored\n",end); |
| 1829 | } |
| 1830 | |
| 1831 | return(num); |
| 1832 | } |
| 1833 | |
| 1834 | /*-- trace support ----------------------------------------------------------*/ |
| 1835 | |
| 1836 | /* The TRACE support is provided (if required) in the memory accessing |
| 1837 | routines. Since we are also providing the architecture specific |
| 1838 | features, the architecture simulation code can also deal with |
| 1839 | notifying the TRACE world of cache flushes, etc. Similarly we do |
| 1840 | not need to provide profiling support in the simulator engine, |
| 1841 | since we can sample in the instruction fetch control loop. By |
| 1842 | defining the TRACE manifest, we add tracing as a run-time |
| 1843 | option. */ |
| 1844 | |
| 1845 | #if defined(TRACE) |
| 1846 | /* Tracing by default produces "din" format (as required by |
| 1847 | dineroIII). Each line of such a trace file *MUST* have a din label |
| 1848 | and address field. The rest of the line is ignored, so comments can |
| 1849 | be included if desired. The first field is the label which must be |
| 1850 | one of the following values: |
| 1851 | |
| 1852 | 0 read data |
| 1853 | 1 write data |
| 1854 | 2 instruction fetch |
| 1855 | 3 escape record (treated as unknown access type) |
| 1856 | 4 escape record (causes cache flush) |
| 1857 | |
| 1858 | The address field is a 32bit (lower-case) hexadecimal address |
| 1859 | value. The address should *NOT* be preceded by "0x". |
| 1860 | |
| 1861 | The size of the memory transfer is not important when dealing with |
| 1862 | cache lines (as long as no more than a cache line can be |
| 1863 | transferred in a single operation :-), however more information |
| 1864 | could be given following the dineroIII requirement to allow more |
| 1865 | complete memory and cache simulators to provide better |
| 1866 | results. i.e. the University of Pisa has a cache simulator that can |
| 1867 | also take bus size and speed as (variable) inputs to calculate |
| 1868 | complete system performance (a much more useful ability when trying |
| 1869 | to construct an end product, rather than a processor). They |
| 1870 | currently have an ARM version of their tool called ChARM. */ |
| 1871 | |
| 1872 | static |
| 1873 | void dotrace(tracefh,type,address,width,comment) |
| 1874 | FILE *tracefh; |
| 1875 | int type; |
| 1876 | unsigned int address; |
| 1877 | int width; |
| 1878 | char *comment; |
| 1879 | { |
| 1880 | if (state & simTRACE) { |
| 1881 | va_list ap; |
| 1882 | fprintf(tracefh,"%d %08x ; width %d ; ",type,address,width); |
| 1883 | va_start(ap,comment); |
| 1884 | fprintf(tracefh,comment,ap); |
| 1885 | va_end(ap); |
| 1886 | fprintf(tracefh,"\n"); |
| 1887 | } |
| 1888 | /* NOTE: Since the "din" format will only accept 32bit addresses, and |
| 1889 | we may be generating 64bit ones, we should put the hi-32bits of the |
| 1890 | address into the comment field. */ |
| 1891 | |
| 1892 | /* TODO: Provide a buffer for the trace lines. We can then avoid |
| 1893 | performing writes until the buffer is filled, or the file is |
| 1894 | being closed. */ |
| 1895 | |
| 1896 | /* NOTE: We could consider adding a comment field to the "din" file |
| 1897 | produced using type 3 markers (unknown access). This would then |
| 1898 | allow information about the program that the "din" is for, and |
| 1899 | the MIPs world that was being simulated, to be placed into the |
| 1900 | trace file. */ |
| 1901 | |
| 1902 | return; |
| 1903 | } |
| 1904 | #endif /* TRACE */ |
| 1905 | |
| 1906 | /*---------------------------------------------------------------------------*/ |
| 1907 | /*-- host<->target transfers ------------------------------------------------*/ |
| 1908 | /*---------------------------------------------------------------------------*/ |
| 1909 | /* The following routines allow conditionals to be avoided during the |
| 1910 | simulation, at the cost of increasing the image and source size. */ |
| 1911 | |
| 1912 | static unsigned int |
| 1913 | xfer_direct_word(memory) |
| 1914 | unsigned char *memory; |
| 1915 | { |
| 1916 | return *((unsigned int *)memory); |
| 1917 | } |
| 1918 | |
| 1919 | static uword64 |
| 1920 | xfer_direct_long(memory) |
| 1921 | unsigned char *memory; |
| 1922 | { |
| 1923 | return *((uword64 *)memory); |
| 1924 | } |
| 1925 | |
| 1926 | static unsigned int |
| 1927 | swap_direct_word(data) |
| 1928 | unsigned int data; |
| 1929 | { |
| 1930 | return data; |
| 1931 | } |
| 1932 | |
| 1933 | static uword64 |
| 1934 | swap_direct_long(data) |
| 1935 | uword64 data; |
| 1936 | { |
| 1937 | return data; |
| 1938 | } |
| 1939 | |
| 1940 | static unsigned int |
| 1941 | xfer_big_word(memory) |
| 1942 | unsigned char *memory; |
| 1943 | { |
| 1944 | return ((memory[0] << 24) | (memory[1] << 16) | (memory[2] << 8) | memory[3]); |
| 1945 | } |
| 1946 | |
| 1947 | static uword64 |
| 1948 | xfer_big_long(memory) |
| 1949 | unsigned char *memory; |
| 1950 | { |
| 1951 | return (((uword64)memory[0] << 56) | ((uword64)memory[1] << 48) |
| 1952 | | ((uword64)memory[2] << 40) | ((uword64)memory[3] << 32) |
| 1953 | | (memory[4] << 24) | (memory[5] << 16) | (memory[6] << 8) | memory[7]); |
| 1954 | } |
| 1955 | |
| 1956 | static unsigned int |
| 1957 | xfer_little_word(memory) |
| 1958 | unsigned char *memory; |
| 1959 | { |
| 1960 | return ((memory[3] << 24) | (memory[2] << 16) | (memory[1] << 8) | memory[0]); |
| 1961 | } |
| 1962 | |
| 1963 | static uword64 |
| 1964 | xfer_little_long(memory) |
| 1965 | unsigned char *memory; |
| 1966 | { |
| 1967 | return (((uword64)memory[7] << 56) | ((uword64)memory[6] << 48) |
| 1968 | | ((uword64)memory[5] << 40) | ((uword64)memory[4] << 32) |
| 1969 | | (memory[3] << 24) | (memory[2] << 16) | (memory[1] << 8) | memory[0]); |
| 1970 | } |
| 1971 | |
| 1972 | static unsigned int |
| 1973 | swap_word(data) |
| 1974 | unsigned int data; |
| 1975 | { |
| 1976 | unsigned int result; |
| 1977 | result = data ^ ((data << 16) | (data >> 16)); |
| 1978 | result &= ~0x00FF0000; |
| 1979 | data = (data << 24) | (data >> 8); |
| 1980 | return data ^ (result >> 8); |
| 1981 | } |
| 1982 | |
| 1983 | static uword64 |
| 1984 | swap_long(data) |
| 1985 | uword64 data; |
| 1986 | { |
| 1987 | unsigned int tmphi = WORD64HI(data); |
| 1988 | unsigned int tmplo = WORD64LO(data); |
| 1989 | tmphi = swap_word(tmphi); |
| 1990 | tmplo = swap_word(tmplo); |
| 1991 | /* Now swap the HI and LO parts */ |
| 1992 | return SET64LO(tmphi) | SET64HI(tmplo); |
| 1993 | } |
| 1994 | |
| 1995 | /*---------------------------------------------------------------------------*/ |
| 1996 | /*-- simulator engine -------------------------------------------------------*/ |
| 1997 | /*---------------------------------------------------------------------------*/ |
| 1998 | |
| 1999 | static void |
| 2000 | ColdReset() |
| 2001 | { |
| 2002 | /* RESET: Fixed PC address: */ |
| 2003 | PC = (((uword64)0xFFFFFFFF<<32) | 0xBFC00000); |
| 2004 | /* The reset vector address is in the unmapped, uncached memory space. */ |
| 2005 | |
| 2006 | SR &= ~(status_SR | status_TS | status_RP); |
| 2007 | SR |= (status_ERL | status_BEV); |
| 2008 | /* VR4300 starts in Big-Endian mode */ |
| 2009 | CONFIG &= ~(config_EP_mask << config_EP_shift); |
| 2010 | CONFIG |= ((config_EP_D << config_EP_shift) | config_BE); |
| 2011 | /* TODO: The VR4300 CONFIG register is not modelled fully at the moment */ |
| 2012 | |
| 2013 | #if defined(HASFPU) && (GPRLEN == (64)) |
| 2014 | /* Cheat and allow access to the complete register set immediately: */ |
| 2015 | SR |= status_FR; /* 64bit registers */ |
| 2016 | #endif /* HASFPU and 64bit FP registers */ |
| 2017 | |
| 2018 | /* Ensure that any instructions with pending register updates are |
| 2019 | cleared: */ |
| 2020 | { |
| 2021 | int loop; |
| 2022 | for (loop = 0; (loop < PSLOTS); loop++) |
| 2023 | pending_slot_reg[loop] = (LAST_EMBED_REGNUM + 1); |
| 2024 | pending_in = pending_out = pending_total = 0; |
| 2025 | } |
| 2026 | |
| 2027 | #if defined(HASFPU) |
| 2028 | /* Initialise the FPU registers to the unknown state */ |
| 2029 | { |
| 2030 | int rn; |
| 2031 | for (rn = 0; (rn < 32); rn++) |
| 2032 | fpr_state[rn] = fmt_uninterpreted; |
| 2033 | } |
| 2034 | #endif /* HASFPU */ |
| 2035 | |
| 2036 | /* In reality this check should be performed at various points |
| 2037 | within the simulation, since it is possible to change the |
| 2038 | endianness of user programs. However, we perform the check here |
| 2039 | to ensure that the start-of-day values agree: */ |
| 2040 | state |= (BigEndianCPU ? simBE : 0); |
| 2041 | if ((target_byte_order == 1234) != !(state & simBE)) { |
| 2042 | fprintf(stderr,"ColdReset: GDB (%s) and simulator (%s) do not agree on target endianness\n", |
| 2043 | target_byte_order == 1234 ? "little" : "big", |
| 2044 | state & simBE ? "big" : "little"); |
| 2045 | exit(1); |
| 2046 | } |
| 2047 | |
| 2048 | if (!(state & simHOSTBE) == !(state & simBE)) { |
| 2049 | host_read_word = xfer_direct_word; |
| 2050 | host_read_long = xfer_direct_long; |
| 2051 | host_swap_word = swap_direct_word; |
| 2052 | host_swap_long = swap_direct_long; |
| 2053 | } else if (state & simHOSTBE) { |
| 2054 | host_read_word = xfer_little_word; |
| 2055 | host_read_long = xfer_little_long; |
| 2056 | host_swap_word = swap_word; |
| 2057 | host_swap_long = swap_long; |
| 2058 | } else { /* HOST little-endian */ |
| 2059 | host_read_word = xfer_big_word; |
| 2060 | host_read_long = xfer_big_long; |
| 2061 | host_swap_word = swap_word; |
| 2062 | host_swap_long = swap_long; |
| 2063 | } |
| 2064 | |
| 2065 | return; |
| 2066 | } |
| 2067 | |
| 2068 | /* Description from page A-22 of the "MIPS IV Instruction Set" manual (revision 3.1) */ |
| 2069 | /* Translate a virtual address to a physical address and cache |
| 2070 | coherence algorithm describing the mechanism used to resolve the |
| 2071 | memory reference. Given the virtual address vAddr, and whether the |
| 2072 | reference is to Instructions ot Data (IorD), find the corresponding |
| 2073 | physical address (pAddr) and the cache coherence algorithm (CCA) |
| 2074 | used to resolve the reference. If the virtual address is in one of |
| 2075 | the unmapped address spaces the physical address and the CCA are |
| 2076 | determined directly by the virtual address. If the virtual address |
| 2077 | is in one of the mapped address spaces then the TLB is used to |
| 2078 | determine the physical address and access type; if the required |
| 2079 | translation is not present in the TLB or the desired access is not |
| 2080 | permitted the function fails and an exception is taken. |
| 2081 | |
| 2082 | NOTE: This function is extended to return an exception state. This, |
| 2083 | along with the exception generation is used to notify whether a |
| 2084 | valid address translation occured */ |
| 2085 | |
| 2086 | static int |
| 2087 | AddressTranslation(vAddr,IorD,LorS,pAddr,CCA,host,raw) |
| 2088 | uword64 vAddr; |
| 2089 | int IorD; |
| 2090 | int LorS; |
| 2091 | uword64 *pAddr; |
| 2092 | int *CCA; |
| 2093 | int host; |
| 2094 | int raw; |
| 2095 | { |
| 2096 | int res = -1; /* TRUE : Assume good return */ |
| 2097 | |
| 2098 | #ifdef DEBUG |
| 2099 | callback->printf_filtered(callback,"AddressTranslation(0x%08X%08X,%s,%s,...);\n",WORD64HI(vAddr),WORD64LO(vAddr),(IorD ? "isDATA" : "isINSTRUCTION"),(LorS ? "iSTORE" : "isLOAD")); |
| 2100 | #endif |
| 2101 | |
| 2102 | /* Check that the address is valid for this memory model */ |
| 2103 | |
| 2104 | /* For a simple (flat) memory model, we simply pass virtual |
| 2105 | addressess through (mostly) unchanged. */ |
| 2106 | vAddr &= 0xFFFFFFFF; |
| 2107 | |
| 2108 | /* Treat the kernel memory spaces identically for the moment: */ |
| 2109 | if ((membank_base == K1BASE) && (vAddr >= K0BASE) && (vAddr < (K0BASE + K0SIZE))) |
| 2110 | vAddr += (K1BASE - K0BASE); |
| 2111 | |
| 2112 | /* Also assume that the K1BASE memory wraps. This is required to |
| 2113 | allow the PMON run-time __sizemem() routine to function (without |
| 2114 | having to provide exception simulation). NOTE: A kludge to work |
| 2115 | around the fact that the monitor memory is currently held in the |
| 2116 | K1BASE space. */ |
| 2117 | if (((vAddr < monitor_base) || (vAddr >= (monitor_base + monitor_size))) && (vAddr >= K1BASE && vAddr < (K1BASE + K1SIZE))) |
| 2118 | vAddr = (K1BASE | (vAddr & (membank_size - 1))); |
| 2119 | |
| 2120 | *pAddr = vAddr; /* default for isTARGET */ |
| 2121 | *CCA = Uncached; /* not used for isHOST */ |
| 2122 | |
| 2123 | /* NOTE: This is a duplicate of the code that appears in the |
| 2124 | LoadMemory and StoreMemory functions. They should be merged into |
| 2125 | a single function (that can be in-lined if required). */ |
| 2126 | if ((vAddr >= membank_base) && (vAddr < (membank_base + membank_size))) { |
| 2127 | if (host) |
| 2128 | *pAddr = (int)&membank[((unsigned int)(vAddr - membank_base) & (membank_size - 1))]; |
| 2129 | } else if ((vAddr >= monitor_base) && (vAddr < (monitor_base + monitor_size))) { |
| 2130 | if (host) |
| 2131 | *pAddr = (int)&monitor[((unsigned int)(vAddr - monitor_base) & (monitor_size - 1))]; |
| 2132 | } else { |
| 2133 | #if 1 /* def DEBUG */ |
| 2134 | sim_warning("Failed: AddressTranslation(0x%08X%08X,%s,%s,...) IPC = 0x%08X%08X",WORD64HI(vAddr),WORD64LO(vAddr),(IorD ? "isDATA" : "isINSTRUCTION"),(LorS ? "isSTORE" : "isLOAD"),WORD64HI(IPC),WORD64LO(IPC)); |
| 2135 | #endif /* DEBUG */ |
| 2136 | res = 0; /* AddressTranslation has failed */ |
| 2137 | *pAddr = -1; |
| 2138 | if (!raw) /* only generate exceptions on real memory transfers */ |
| 2139 | SignalException((LorS == isSTORE) ? AddressStore : AddressLoad); |
| 2140 | else |
| 2141 | sim_warning("AddressTranslation for %s %s from 0x%08X%08X failed",(IorD ? "data" : "instruction"),(LorS ? "store" : "load"),WORD64HI(vAddr),WORD64LO(vAddr)); |
| 2142 | } |
| 2143 | |
| 2144 | return(res); |
| 2145 | } |
| 2146 | |
| 2147 | /* Description from page A-23 of the "MIPS IV Instruction Set" manual (revision 3.1) */ |
| 2148 | /* Prefetch data from memory. Prefetch is an advisory instruction for |
| 2149 | which an implementation specific action is taken. The action taken |
| 2150 | may increase performance, but must not change the meaning of the |
| 2151 | program, or alter architecturally-visible state. */ |
| 2152 | static void |
| 2153 | Prefetch(CCA,pAddr,vAddr,DATA,hint) |
| 2154 | int CCA; |
| 2155 | uword64 pAddr; |
| 2156 | uword64 vAddr; |
| 2157 | int DATA; |
| 2158 | int hint; |
| 2159 | { |
| 2160 | #ifdef DEBUG |
| 2161 | callback->printf_filtered(callback,"Prefetch(%d,0x%08X%08X,0x%08X%08X,%d,%d);\n",CCA,WORD64HI(pAddr),WORD64LO(pAddr),WORD64HI(vAddr),WORD64LO(vAddr),DATA,hint); |
| 2162 | #endif /* DEBUG */ |
| 2163 | |
| 2164 | /* For our simple memory model we do nothing */ |
| 2165 | return; |
| 2166 | } |
| 2167 | |
| 2168 | /* Description from page A-22 of the "MIPS IV Instruction Set" manual (revision 3.1) */ |
| 2169 | /* Load a value from memory. Use the cache and main memory as |
| 2170 | specified in the Cache Coherence Algorithm (CCA) and the sort of |
| 2171 | access (IorD) to find the contents of AccessLength memory bytes |
| 2172 | starting at physical location pAddr. The data is returned in the |
| 2173 | fixed width naturally-aligned memory element (MemElem). The |
| 2174 | low-order two (or three) bits of the address and the AccessLength |
| 2175 | indicate which of the bytes within MemElem needs to be given to the |
| 2176 | processor. If the memory access type of the reference is uncached |
| 2177 | then only the referenced bytes are read from memory and valid |
| 2178 | within the memory element. If the access type is cached, and the |
| 2179 | data is not present in cache, an implementation specific size and |
| 2180 | alignment block of memory is read and loaded into the cache to |
| 2181 | satisfy a load reference. At a minimum, the block is the entire |
| 2182 | memory element. */ |
| 2183 | static uword64 |
| 2184 | LoadMemory(CCA,AccessLength,pAddr,vAddr,IorD,raw) |
| 2185 | int CCA; |
| 2186 | int AccessLength; |
| 2187 | uword64 pAddr; |
| 2188 | uword64 vAddr; |
| 2189 | int IorD; |
| 2190 | int raw; |
| 2191 | { |
| 2192 | uword64 value; |
| 2193 | |
| 2194 | #ifdef DEBUG |
| 2195 | if (membank == NULL) |
| 2196 | callback->printf_filtered(callback,"DBG: LoadMemory(%d,%d,0x%08X%08X,0x%08X%08X,%s,%s)\n",CCA,AccessLength,WORD64HI(pAddr),WORD64LO(pAddr),WORD64HI(vAddr),WORD64LO(vAddr),(IorD ? "isDATA" : "isINSTRUCTION"),(raw ? "isRAW" : "isREAL")); |
| 2197 | #endif /* DEBUG */ |
| 2198 | |
| 2199 | #if defined(WARN_MEM) |
| 2200 | if (CCA != uncached) |
| 2201 | sim_warning("LoadMemory CCA (%d) is not uncached (currently all accesses treated as cached)",CCA); |
| 2202 | |
| 2203 | if (((pAddr & LOADDRMASK) + AccessLength) > LOADDRMASK) { |
| 2204 | /* In reality this should be a Bus Error */ |
| 2205 | sim_error("AccessLength of %d would extend over %dbit aligned boundary for physical address 0x%08X%08X\n",AccessLength,(LOADDRMASK + 1)<<2,WORD64HI(pAddr),WORD64LO(pAddr)); |
| 2206 | } |
| 2207 | #endif /* WARN_MEM */ |
| 2208 | |
| 2209 | /* Decide which physical memory locations are being dealt with. At |
| 2210 | this point we should be able to split the pAddr bits into the |
| 2211 | relevant address map being simulated. If the "raw" variable is |
| 2212 | set, the memory read being performed should *NOT* update any I/O |
| 2213 | state or affect the CPU state. This also includes avoiding |
| 2214 | affecting statistics gathering. */ |
| 2215 | |
| 2216 | /* If instruction fetch then we need to check that the two lo-order |
| 2217 | bits are zero, otherwise raise a InstructionFetch exception: */ |
| 2218 | if ((IorD == isINSTRUCTION) && ((pAddr & 0x3) != 0)) |
| 2219 | SignalException(InstructionFetch); |
| 2220 | else { |
| 2221 | unsigned int index; |
| 2222 | unsigned char *mem = NULL; |
| 2223 | |
| 2224 | #if defined(TRACE) |
| 2225 | if (!raw) |
| 2226 | dotrace(tracefh,((IorD == isDATA) ? 0 : 2),(unsigned int)(pAddr&0xFFFFFFFF),(AccessLength + 1),"load%s",((IorD == isDATA) ? "" : " instruction")); |
| 2227 | #endif /* TRACE */ |
| 2228 | |
| 2229 | /* NOTE: Quicker methods of decoding the address space can be used |
| 2230 | when a real memory map is being simulated (i.e. using hi-order |
| 2231 | address bits to select device). */ |
| 2232 | if ((pAddr >= membank_base) && (pAddr < (membank_base + membank_size))) { |
| 2233 | index = ((unsigned int)(pAddr - membank_base) & (membank_size - 1)); |
| 2234 | mem = membank; |
| 2235 | } else if ((pAddr >= monitor_base) && (pAddr < (monitor_base + monitor_size))) { |
| 2236 | index = ((unsigned int)(pAddr - monitor_base) & (monitor_size - 1)); |
| 2237 | mem = monitor; |
| 2238 | } |
| 2239 | if (mem == NULL) |
| 2240 | sim_error("Simulator memory not found for physical address 0x%08X%08X\n",WORD64HI(pAddr),WORD64LO(pAddr)); |
| 2241 | else { |
| 2242 | /* If we obtained the endianness of the host, and it is the same |
| 2243 | as the target memory system we can optimise the memory |
| 2244 | accesses. However, without that information we must perform |
| 2245 | slow transfer, and hope that the compiler optimisation will |
| 2246 | merge successive loads. */ |
| 2247 | value = 0; /* no data loaded yet */ |
| 2248 | |
| 2249 | /* In reality we should always be loading a doubleword value (or |
| 2250 | word value in 32bit memory worlds). The external code then |
| 2251 | extracts the required bytes. However, to keep performance |
| 2252 | high we only load the required bytes into the relevant |
| 2253 | slots. */ |
| 2254 | if (BigEndianMem) |
| 2255 | switch (AccessLength) { /* big-endian memory */ |
| 2256 | case AccessLength_DOUBLEWORD : |
| 2257 | value |= ((uword64)mem[index++] << 56); |
| 2258 | case AccessLength_SEPTIBYTE : |
| 2259 | value |= ((uword64)mem[index++] << 48); |
| 2260 | case AccessLength_SEXTIBYTE : |
| 2261 | value |= ((uword64)mem[index++] << 40); |
| 2262 | case AccessLength_QUINTIBYTE : |
| 2263 | value |= ((uword64)mem[index++] << 32); |
| 2264 | case AccessLength_WORD : |
| 2265 | value |= ((unsigned int)mem[index++] << 24); |
| 2266 | case AccessLength_TRIPLEBYTE : |
| 2267 | value |= ((unsigned int)mem[index++] << 16); |
| 2268 | case AccessLength_HALFWORD : |
| 2269 | value |= ((unsigned int)mem[index++] << 8); |
| 2270 | case AccessLength_BYTE : |
| 2271 | value |= mem[index]; |
| 2272 | break; |
| 2273 | } |
| 2274 | else { |
| 2275 | index += (AccessLength + 1); |
| 2276 | switch (AccessLength) { /* little-endian memory */ |
| 2277 | case AccessLength_DOUBLEWORD : |
| 2278 | value |= ((uword64)mem[--index] << 56); |
| 2279 | case AccessLength_SEPTIBYTE : |
| 2280 | value |= ((uword64)mem[--index] << 48); |
| 2281 | case AccessLength_SEXTIBYTE : |
| 2282 | value |= ((uword64)mem[--index] << 40); |
| 2283 | case AccessLength_QUINTIBYTE : |
| 2284 | value |= ((uword64)mem[--index] << 32); |
| 2285 | case AccessLength_WORD : |
| 2286 | value |= ((uword64)mem[--index] << 24); |
| 2287 | case AccessLength_TRIPLEBYTE : |
| 2288 | value |= ((uword64)mem[--index] << 16); |
| 2289 | case AccessLength_HALFWORD : |
| 2290 | value |= ((uword64)mem[--index] << 8); |
| 2291 | case AccessLength_BYTE : |
| 2292 | value |= ((uword64)mem[--index] << 0); |
| 2293 | break; |
| 2294 | } |
| 2295 | } |
| 2296 | |
| 2297 | #ifdef DEBUG |
| 2298 | printf("DBG: LoadMemory() : (offset %d) : value = 0x%08X%08X\n",(int)(pAddr & LOADDRMASK),WORD64HI(value),WORD64LO(value)); |
| 2299 | #endif /* DEBUG */ |
| 2300 | |
| 2301 | /* TODO: We could try and avoid the shifts when dealing with raw |
| 2302 | memory accesses. This would mean updating the LoadMemory and |
| 2303 | StoreMemory routines to avoid shifting the data before |
| 2304 | returning or using it. */ |
| 2305 | if (!raw) { /* do nothing for raw accessess */ |
| 2306 | if (BigEndianMem) |
| 2307 | value <<= (((7 - (pAddr & LOADDRMASK)) - AccessLength) * 8); |
| 2308 | else /* little-endian only needs to be shifted up to the correct byte offset */ |
| 2309 | value <<= ((pAddr & LOADDRMASK) * 8); |
| 2310 | } |
| 2311 | |
| 2312 | #ifdef DEBUG |
| 2313 | printf("DBG: LoadMemory() : shifted value = 0x%08X%08X\n",WORD64HI(value),WORD64LO(value)); |
| 2314 | #endif /* DEBUG */ |
| 2315 | } |
| 2316 | } |
| 2317 | |
| 2318 | return(value); |
| 2319 | } |
| 2320 | |
| 2321 | /* Description from page A-23 of the "MIPS IV Instruction Set" manual (revision 3.1) */ |
| 2322 | /* Store a value to memory. The specified data is stored into the |
| 2323 | physical location pAddr using the memory hierarchy (data caches and |
| 2324 | main memory) as specified by the Cache Coherence Algorithm |
| 2325 | (CCA). The MemElem contains the data for an aligned, fixed-width |
| 2326 | memory element (word for 32-bit processors, doubleword for 64-bit |
| 2327 | processors), though only the bytes that will actually be stored to |
| 2328 | memory need to be valid. The low-order two (or three) bits of pAddr |
| 2329 | and the AccessLength field indicates which of the bytes within the |
| 2330 | MemElem data should actually be stored; only these bytes in memory |
| 2331 | will be changed. */ |
| 2332 | static void |
| 2333 | StoreMemory(CCA,AccessLength,MemElem,pAddr,vAddr,raw) |
| 2334 | int CCA; |
| 2335 | int AccessLength; |
| 2336 | uword64 MemElem; |
| 2337 | uword64 pAddr; |
| 2338 | uword64 vAddr; |
| 2339 | int raw; |
| 2340 | { |
| 2341 | #ifdef DEBUG |
| 2342 | callback->printf_filtered(callback,"DBG: StoreMemory(%d,%d,0x%08X%08X,0x%08X%08X,0x%08X%08X,%s)\n",CCA,AccessLength,WORD64HI(MemElem),WORD64LO(MemElem),WORD64HI(pAddr),WORD64LO(pAddr),WORD64HI(vAddr),WORD64LO(vAddr),(raw ? "isRAW" : "isREAL")); |
| 2343 | #endif /* DEBUG */ |
| 2344 | |
| 2345 | #if defined(WARN_MEM) |
| 2346 | if (CCA != uncached) |
| 2347 | sim_warning("StoreMemory CCA (%d) is not uncached (currently all accesses treated as cached)",CCA); |
| 2348 | |
| 2349 | if (((pAddr & LOADDRMASK) + AccessLength) > LOADDRMASK) |
| 2350 | sim_error("AccessLength of %d would extend over %dbit aligned boundary for physical address 0x%08X%08X\n",AccessLength,(LOADDRMASK + 1)<<2,WORD64HI(pAddr),WORD64LO(pAddr)); |
| 2351 | #endif /* WARN_MEM */ |
| 2352 | |
| 2353 | #if defined(TRACE) |
| 2354 | if (!raw) |
| 2355 | dotrace(tracefh,1,(unsigned int)(pAddr&0xFFFFFFFF),(AccessLength + 1),"store"); |
| 2356 | #endif /* TRACE */ |
| 2357 | |
| 2358 | /* See the comments in the LoadMemory routine about optimising |
| 2359 | memory accesses. Also if we wanted to make the simulator smaller, |
| 2360 | we could merge a lot of this code with the LoadMemory |
| 2361 | routine. However, this would slow the simulator down with |
| 2362 | run-time conditionals. */ |
| 2363 | { |
| 2364 | unsigned int index; |
| 2365 | unsigned char *mem = NULL; |
| 2366 | |
| 2367 | if ((pAddr >= membank_base) && (pAddr < (membank_base + membank_size))) { |
| 2368 | index = ((unsigned int)(pAddr - membank_base) & (membank_size - 1)); |
| 2369 | mem = membank; |
| 2370 | } else if ((pAddr >= monitor_base) && (pAddr < (monitor_base + monitor_size))) { |
| 2371 | index = ((unsigned int)(pAddr - monitor_base) & (monitor_size - 1)); |
| 2372 | mem = monitor; |
| 2373 | } |
| 2374 | |
| 2375 | if (mem == NULL) |
| 2376 | sim_error("Simulator memory not found for physical address 0x%08X%08X\n",WORD64HI(pAddr),WORD64LO(pAddr)); |
| 2377 | else { |
| 2378 | int shift = 0; |
| 2379 | |
| 2380 | #ifdef DEBUG |
| 2381 | printf("DBG: StoreMemory: offset = %d MemElem = 0x%08X%08X\n",(unsigned int)(pAddr & LOADDRMASK),WORD64HI(MemElem),WORD64LO(MemElem)); |
| 2382 | #endif /* DEBUG */ |
| 2383 | |
| 2384 | if (BigEndianMem) { |
| 2385 | if (raw) |
| 2386 | shift = ((7 - AccessLength) * 8); |
| 2387 | else /* real memory access */ |
| 2388 | shift = ((pAddr & LOADDRMASK) * 8); |
| 2389 | MemElem <<= shift; |
| 2390 | } else { |
| 2391 | /* no need to shift raw little-endian data */ |
| 2392 | if (!raw) |
| 2393 | MemElem >>= ((pAddr & LOADDRMASK) * 8); |
| 2394 | } |
| 2395 | |
| 2396 | #ifdef DEBUG |
| 2397 | printf("DBG: StoreMemory: shift = %d MemElem = 0x%08X%08X\n",shift,WORD64HI(MemElem),WORD64LO(MemElem)); |
| 2398 | #endif /* DEBUG */ |
| 2399 | |
| 2400 | if (BigEndianMem) { |
| 2401 | switch (AccessLength) { /* big-endian memory */ |
| 2402 | case AccessLength_DOUBLEWORD : |
| 2403 | mem[index++] = (unsigned char)(MemElem >> 56); |
| 2404 | MemElem <<= 8; |
| 2405 | case AccessLength_SEPTIBYTE : |
| 2406 | mem[index++] = (unsigned char)(MemElem >> 56); |
| 2407 | MemElem <<= 8; |
| 2408 | case AccessLength_SEXTIBYTE : |
| 2409 | mem[index++] = (unsigned char)(MemElem >> 56); |
| 2410 | MemElem <<= 8; |
| 2411 | case AccessLength_QUINTIBYTE : |
| 2412 | mem[index++] = (unsigned char)(MemElem >> 56); |
| 2413 | MemElem <<= 8; |
| 2414 | case AccessLength_WORD : |
| 2415 | mem[index++] = (unsigned char)(MemElem >> 56); |
| 2416 | MemElem <<= 8; |
| 2417 | case AccessLength_TRIPLEBYTE : |
| 2418 | mem[index++] = (unsigned char)(MemElem >> 56); |
| 2419 | MemElem <<= 8; |
| 2420 | case AccessLength_HALFWORD : |
| 2421 | mem[index++] = (unsigned char)(MemElem >> 56); |
| 2422 | MemElem <<= 8; |
| 2423 | case AccessLength_BYTE : |
| 2424 | mem[index++] = (unsigned char)(MemElem >> 56); |
| 2425 | break; |
| 2426 | } |
| 2427 | } else { |
| 2428 | index += (AccessLength + 1); |
| 2429 | switch (AccessLength) { /* little-endian memory */ |
| 2430 | case AccessLength_DOUBLEWORD : |
| 2431 | mem[--index] = (unsigned char)(MemElem >> 56); |
| 2432 | case AccessLength_SEPTIBYTE : |
| 2433 | mem[--index] = (unsigned char)(MemElem >> 48); |
| 2434 | case AccessLength_SEXTIBYTE : |
| 2435 | mem[--index] = (unsigned char)(MemElem >> 40); |
| 2436 | case AccessLength_QUINTIBYTE : |
| 2437 | mem[--index] = (unsigned char)(MemElem >> 32); |
| 2438 | case AccessLength_WORD : |
| 2439 | mem[--index] = (unsigned char)(MemElem >> 24); |
| 2440 | case AccessLength_TRIPLEBYTE : |
| 2441 | mem[--index] = (unsigned char)(MemElem >> 16); |
| 2442 | case AccessLength_HALFWORD : |
| 2443 | mem[--index] = (unsigned char)(MemElem >> 8); |
| 2444 | case AccessLength_BYTE : |
| 2445 | mem[--index] = (unsigned char)(MemElem >> 0); |
| 2446 | break; |
| 2447 | } |
| 2448 | } |
| 2449 | } |
| 2450 | } |
| 2451 | |
| 2452 | return; |
| 2453 | } |
| 2454 | |
| 2455 | /* Description from page A-26 of the "MIPS IV Instruction Set" manual (revision 3.1) */ |
| 2456 | /* Order loads and stores to synchronise shared memory. Perform the |
| 2457 | action necessary to make the effects of groups of synchronizable |
| 2458 | loads and stores indicated by stype occur in the same order for all |
| 2459 | processors. */ |
| 2460 | static void |
| 2461 | SyncOperation(stype) |
| 2462 | int stype; |
| 2463 | { |
| 2464 | #ifdef DEBUG |
| 2465 | callback->printf_filtered(callback,"SyncOperation(%d) : TODO\n",stype); |
| 2466 | #endif /* DEBUG */ |
| 2467 | return; |
| 2468 | } |
| 2469 | |
| 2470 | /* Description from page A-26 of the "MIPS IV Instruction Set" manual (revision 3.1) */ |
| 2471 | /* Signal an exception condition. This will result in an exception |
| 2472 | that aborts the instruction. The instruction operation pseudocode |
| 2473 | will never see a return from this function call. */ |
| 2474 | static void |
| 2475 | SignalException(exception) |
| 2476 | int exception; |
| 2477 | { |
| 2478 | /* Ensure that any active atomic read/modify/write operation will fail: */ |
| 2479 | LLBIT = 0; |
| 2480 | |
| 2481 | switch (exception) { |
| 2482 | /* TODO: For testing purposes I have been ignoring TRAPs. In |
| 2483 | reality we should either simulate them, or allow the user to |
| 2484 | ignore them at run-time. */ |
| 2485 | case Trap : |
| 2486 | sim_warning("Ignoring instruction TRAP (PC 0x%08X%08X)",WORD64HI(IPC),WORD64LO(IPC)); |
| 2487 | break; |
| 2488 | |
| 2489 | case ReservedInstruction : |
| 2490 | { |
| 2491 | va_list ap; |
| 2492 | unsigned int instruction; |
| 2493 | va_start(ap,exception); |
| 2494 | instruction = va_arg(ap,unsigned int); |
| 2495 | va_end(ap); |
| 2496 | /* Provide simple monitor support using ReservedInstruction |
| 2497 | exceptions. The following code simulates the fixed vector |
| 2498 | entry points into the IDT monitor by causing a simulator |
| 2499 | trap, performing the monitor operation, and returning to |
| 2500 | the address held in the $ra register (standard PCS return |
| 2501 | address). This means we only need to pre-load the vector |
| 2502 | space with suitable instruction values. For systems were |
| 2503 | actual trap instructions are used, we would not need to |
| 2504 | perform this magic. */ |
| 2505 | if ((instruction & ~RSVD_INSTRUCTION_AMASK) == RSVD_INSTRUCTION) { |
| 2506 | sim_monitor(instruction & RSVD_INSTRUCTION_AMASK); |
| 2507 | PC = RA; /* simulate the return from the vector entry */ |
| 2508 | /* NOTE: This assumes that a branch-and-link style |
| 2509 | instruction was used to enter the vector (which is the |
| 2510 | case with the current IDT monitor). */ |
| 2511 | break; /* out of the switch statement */ |
| 2512 | } /* else fall through to normal exception processing */ |
| 2513 | sim_warning("ReservedInstruction 0x%08X at IPC = 0x%08X%08X",instruction,WORD64HI(IPC),WORD64LO(IPC)); |
| 2514 | } |
| 2515 | |
| 2516 | default: |
| 2517 | #if 1 /* def DEBUG */ |
| 2518 | if (exception != BreakPoint) |
| 2519 | callback->printf_filtered(callback,"DBG: SignalException(%d) IPC = 0x%08X%08X\n",exception,WORD64HI(IPC),WORD64LO(IPC)); |
| 2520 | #endif /* DEBUG */ |
| 2521 | /* Store exception code into current exception id variable (used |
| 2522 | by exit code): */ |
| 2523 | |
| 2524 | /* TODO: If not simulating exceptions then stop the simulator |
| 2525 | execution. At the moment we always stop the simulation. */ |
| 2526 | #if 1 /* bodge to allow exit() code to be returned, by assuming that a breakpoint exception after a monitor exit() call should be silent */ |
| 2527 | /* further bodged since the standard libgloss/mips world doesn't use the _exit() monitor call, it just uses a break instruction */ |
| 2528 | if (exception == BreakPoint /* && state & simEXIT */) |
| 2529 | { |
| 2530 | state |= simSTOP; |
| 2531 | #if 1 /* since the _exit() monitor call may not be called */ |
| 2532 | state |= simEXIT; |
| 2533 | rcexit = (unsigned int)(A0 & 0xFFFFFFFF); |
| 2534 | #endif |
| 2535 | } |
| 2536 | else |
| 2537 | state |= (simSTOP | simEXCEPTION); |
| 2538 | #else |
| 2539 | state |= (simSTOP | simEXCEPTION); |
| 2540 | #endif |
| 2541 | CAUSE = (exception << 2); |
| 2542 | if (state & simDELAYSLOT) { |
| 2543 | CAUSE |= cause_BD; |
| 2544 | EPC = (IPC - 4); /* reference the branch instruction */ |
| 2545 | } else |
| 2546 | EPC = IPC; |
| 2547 | /* The following is so that the simulator will continue from the |
| 2548 | exception address on breakpoint operations. */ |
| 2549 | PC = EPC; |
| 2550 | break; |
| 2551 | |
| 2552 | case SimulatorFault: |
| 2553 | { |
| 2554 | va_list ap; |
| 2555 | char *msg; |
| 2556 | va_start(ap,exception); |
| 2557 | msg = va_arg(ap,char *); |
| 2558 | fprintf(stderr,"FATAL: Simulator error \"%s\"\n",msg); |
| 2559 | va_end(ap); |
| 2560 | } |
| 2561 | exit(1); |
| 2562 | } |
| 2563 | |
| 2564 | return; |
| 2565 | } |
| 2566 | |
| 2567 | #if defined(WARN_RESULT) |
| 2568 | /* Description from page A-26 of the "MIPS IV Instruction Set" manual (revision 3.1) */ |
| 2569 | /* This function indicates that the result of the operation is |
| 2570 | undefined. However, this should not affect the instruction |
| 2571 | stream. All that is meant to happen is that the destination |
| 2572 | register is set to an undefined result. To keep the simulator |
| 2573 | simple, we just don't bother updating the destination register, so |
| 2574 | the overall result will be undefined. If desired we can stop the |
| 2575 | simulator by raising a pseudo-exception. */ |
| 2576 | static void |
| 2577 | UndefinedResult() |
| 2578 | { |
| 2579 | sim_warning("UndefinedResult: IPC = 0x%08X%08X",WORD64HI(IPC),WORD64LO(IPC)); |
| 2580 | #if 0 /* Disabled for the moment, since it actually happens a lot at the moment. */ |
| 2581 | state |= simSTOP; |
| 2582 | #endif |
| 2583 | return; |
| 2584 | } |
| 2585 | #endif /* WARN_RESULT */ |
| 2586 | |
| 2587 | static void |
| 2588 | CacheOp(op,pAddr,vAddr,instruction) |
| 2589 | int op; |
| 2590 | uword64 pAddr; |
| 2591 | uword64 vAddr; |
| 2592 | unsigned int instruction; |
| 2593 | { |
| 2594 | #if 1 /* stop warning message being displayed (we should really just remove the code) */ |
| 2595 | static int icache_warning = 1; |
| 2596 | static int dcache_warning = 1; |
| 2597 | #else |
| 2598 | static int icache_warning = 0; |
| 2599 | static int dcache_warning = 0; |
| 2600 | #endif |
| 2601 | |
| 2602 | /* If CP0 is not useable (User or Supervisor mode) and the CP0 |
| 2603 | enable bit in the Status Register is clear - a coprocessor |
| 2604 | unusable exception is taken. */ |
| 2605 | #if 0 |
| 2606 | callback->printf_filtered(callback,"TODO: Cache availability checking (PC = 0x%08X%08X)\n",WORD64HI(IPC),WORD64LO(IPC)); |
| 2607 | #endif |
| 2608 | |
| 2609 | switch (op & 0x3) { |
| 2610 | case 0: /* instruction cache */ |
| 2611 | switch (op >> 2) { |
| 2612 | case 0: /* Index Invalidate */ |
| 2613 | case 1: /* Index Load Tag */ |
| 2614 | case 2: /* Index Store Tag */ |
| 2615 | case 4: /* Hit Invalidate */ |
| 2616 | case 5: /* Fill */ |
| 2617 | case 6: /* Hit Writeback */ |
| 2618 | if (!icache_warning) |
| 2619 | { |
| 2620 | sim_warning("Instruction CACHE operation %d to be coded",(op >> 2)); |
| 2621 | icache_warning = 1; |
| 2622 | } |
| 2623 | break; |
| 2624 | |
| 2625 | default: |
| 2626 | SignalException(ReservedInstruction,instruction); |
| 2627 | break; |
| 2628 | } |
| 2629 | break; |
| 2630 | |
| 2631 | case 1: /* data cache */ |
| 2632 | switch (op >> 2) { |
| 2633 | case 0: /* Index Writeback Invalidate */ |
| 2634 | case 1: /* Index Load Tag */ |
| 2635 | case 2: /* Index Store Tag */ |
| 2636 | case 3: /* Create Dirty */ |
| 2637 | case 4: /* Hit Invalidate */ |
| 2638 | case 5: /* Hit Writeback Invalidate */ |
| 2639 | case 6: /* Hit Writeback */ |
| 2640 | if (!dcache_warning) |
| 2641 | { |
| 2642 | sim_warning("Data CACHE operation %d to be coded",(op >> 2)); |
| 2643 | dcache_warning = 1; |
| 2644 | } |
| 2645 | break; |
| 2646 | |
| 2647 | default: |
| 2648 | SignalException(ReservedInstruction,instruction); |
| 2649 | break; |
| 2650 | } |
| 2651 | break; |
| 2652 | |
| 2653 | default: /* unrecognised cache ID */ |
| 2654 | SignalException(ReservedInstruction,instruction); |
| 2655 | break; |
| 2656 | } |
| 2657 | |
| 2658 | return; |
| 2659 | } |
| 2660 | |
| 2661 | /*-- FPU support routines ---------------------------------------------------*/ |
| 2662 | |
| 2663 | #if defined(HASFPU) /* Only needed when building FPU aware simulators */ |
| 2664 | |
| 2665 | #if 1 |
| 2666 | #define SizeFGR() (GPRLEN) |
| 2667 | #else |
| 2668 | /* They depend on the CPU being simulated */ |
| 2669 | #define SizeFGR() ((PROCESSOR_64BIT && ((SR & status_FR) == 1)) ? 64 : 32) |
| 2670 | #endif |
| 2671 | |
| 2672 | /* Numbers are held in normalized form. The SINGLE and DOUBLE binary |
| 2673 | formats conform to ANSI/IEEE Std 754-1985. */ |
| 2674 | /* SINGLE precision floating: |
| 2675 | * seeeeeeeefffffffffffffffffffffff |
| 2676 | * s = 1bit = sign |
| 2677 | * e = 8bits = exponent |
| 2678 | * f = 23bits = fraction |
| 2679 | */ |
| 2680 | /* SINGLE precision fixed: |
| 2681 | * siiiiiiiiiiiiiiiiiiiiiiiiiiiiiii |
| 2682 | * s = 1bit = sign |
| 2683 | * i = 31bits = integer |
| 2684 | */ |
| 2685 | /* DOUBLE precision floating: |
| 2686 | * seeeeeeeeeeeffffffffffffffffffffffffffffffffffffffffffffffffffff |
| 2687 | * s = 1bit = sign |
| 2688 | * e = 11bits = exponent |
| 2689 | * f = 52bits = fraction |
| 2690 | */ |
| 2691 | /* DOUBLE precision fixed: |
| 2692 | * siiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiii |
| 2693 | * s = 1bit = sign |
| 2694 | * i = 63bits = integer |
| 2695 | */ |
| 2696 | |
| 2697 | /* Extract sign-bit: */ |
| 2698 | #define FP_S_s(v) (((v) & ((unsigned)1 << 31)) ? 1 : 0) |
| 2699 | #define FP_D_s(v) (((v) & ((uword64)1 << 63)) ? 1 : 0) |
| 2700 | /* Extract biased exponent: */ |
| 2701 | #define FP_S_be(v) (((v) >> 23) & 0xFF) |
| 2702 | #define FP_D_be(v) (((v) >> 52) & 0x7FF) |
| 2703 | /* Extract unbiased Exponent: */ |
| 2704 | #define FP_S_e(v) (FP_S_be(v) - 0x7F) |
| 2705 | #define FP_D_e(v) (FP_D_be(v) - 0x3FF) |
| 2706 | /* Extract complete fraction field: */ |
| 2707 | #define FP_S_f(v) ((v) & ~((unsigned)0x1FF << 23)) |
| 2708 | #define FP_D_f(v) ((v) & ~((uword64)0xFFF << 52)) |
| 2709 | /* Extract numbered fraction bit: */ |
| 2710 | #define FP_S_fb(b,v) (((v) & (1 << (23 - (b)))) ? 1 : 0) |
| 2711 | #define FP_D_fb(b,v) (((v) & (1 << (52 - (b)))) ? 1 : 0) |
| 2712 | |
| 2713 | /* Explicit QNaN values used when value required: */ |
| 2714 | #define FPQNaN_SINGLE (0x7FBFFFFF) |
| 2715 | #define FPQNaN_WORD (0x7FFFFFFF) |
| 2716 | #define FPQNaN_DOUBLE (((uword64)0x7FF7FFFF << 32) | 0xFFFFFFFF) |
| 2717 | #define FPQNaN_LONG (((uword64)0x7FFFFFFF << 32) | 0xFFFFFFFF) |
| 2718 | |
| 2719 | /* Explicit Infinity values used when required: */ |
| 2720 | #define FPINF_SINGLE (0x7F800000) |
| 2721 | #define FPINF_DOUBLE (((uword64)0x7FF00000 << 32) | 0x00000000) |
| 2722 | |
| 2723 | #if 1 /* def DEBUG */ |
| 2724 | #define RMMODE(v) (((v) == FP_RM_NEAREST) ? "Round" : (((v) == FP_RM_TOZERO) ? "Trunc" : (((v) == FP_RM_TOPINF) ? "Ceil" : "Floor"))) |
| 2725 | #define DOFMT(v) (((v) == fmt_single) ? "single" : (((v) == fmt_double) ? "double" : (((v) == fmt_word) ? "word" : (((v) == fmt_long) ? "long" : (((v) == fmt_unknown) ? "<unknown>" : (((v) == fmt_uninterpreted) ? "<uninterpreted>" : "<format error>")))))) |
| 2726 | #endif /* DEBUG */ |
| 2727 | |
| 2728 | static uword64 |
| 2729 | ValueFPR(fpr,fmt) |
| 2730 | int fpr; |
| 2731 | FP_formats fmt; |
| 2732 | { |
| 2733 | uword64 value; |
| 2734 | int err = 0; |
| 2735 | |
| 2736 | /* Treat unused register values, as fixed-point 64bit values: */ |
| 2737 | if ((fmt == fmt_uninterpreted) || (fmt == fmt_unknown)) |
| 2738 | #if 1 |
| 2739 | /* If request to read data as "uninterpreted", then use the current |
| 2740 | encoding: */ |
| 2741 | fmt = fpr_state[fpr]; |
| 2742 | #else |
| 2743 | fmt = fmt_long; |
| 2744 | #endif |
| 2745 | |
| 2746 | /* For values not yet accessed, set to the desired format: */ |
| 2747 | if (fpr_state[fpr] == fmt_uninterpreted) { |
| 2748 | fpr_state[fpr] = fmt; |
| 2749 | #ifdef DEBUG |
| 2750 | printf("DBG: Register %d was fmt_uninterpreted. Now %s\n",fpr,DOFMT(fmt)); |
| 2751 | #endif /* DEBUG */ |
| 2752 | } |
| 2753 | if (fmt != fpr_state[fpr]) { |
| 2754 | sim_warning("FPR %d (format %s) being accessed with format %s - setting to unknown (PC = 0x%08X%08X)",fpr,DOFMT(fpr_state[fpr]),DOFMT(fmt),WORD64HI(IPC),WORD64LO(IPC)); |
| 2755 | fpr_state[fpr] = fmt_unknown; |
| 2756 | } |
| 2757 | |
| 2758 | if (fpr_state[fpr] == fmt_unknown) { |
| 2759 | /* Set QNaN value: */ |
| 2760 | switch (fmt) { |
| 2761 | case fmt_single: |
| 2762 | value = FPQNaN_SINGLE; |
| 2763 | break; |
| 2764 | |
| 2765 | case fmt_double: |
| 2766 | value = FPQNaN_DOUBLE; |
| 2767 | break; |
| 2768 | |
| 2769 | case fmt_word: |
| 2770 | value = FPQNaN_WORD; |
| 2771 | break; |
| 2772 | |
| 2773 | case fmt_long: |
| 2774 | value = FPQNaN_LONG; |
| 2775 | break; |
| 2776 | |
| 2777 | default: |
| 2778 | err = -1; |
| 2779 | break; |
| 2780 | } |
| 2781 | } else if (SizeFGR() == 64) { |
| 2782 | switch (fmt) { |
| 2783 | case fmt_single: |
| 2784 | case fmt_word: |
| 2785 | value = (FGR[fpr] & 0xFFFFFFFF); |
| 2786 | break; |
| 2787 | |
| 2788 | case fmt_uninterpreted: |
| 2789 | case fmt_double: |
| 2790 | case fmt_long: |
| 2791 | value = FGR[fpr]; |
| 2792 | break; |
| 2793 | |
| 2794 | default : |
| 2795 | err = -1; |
| 2796 | break; |
| 2797 | } |
| 2798 | } else if ((fpr & 1) == 0) { /* even registers only */ |
| 2799 | switch (fmt) { |
| 2800 | case fmt_single: |
| 2801 | case fmt_word: |
| 2802 | value = (FGR[fpr] & 0xFFFFFFFF); |
| 2803 | break; |
| 2804 | |
| 2805 | case fmt_uninterpreted: |
| 2806 | case fmt_double: |
| 2807 | case fmt_long: |
| 2808 | value = ((((uword64)FGR[fpr+1]) << 32) | (FGR[fpr] & 0xFFFFFFFF)); |
| 2809 | break; |
| 2810 | |
| 2811 | default : |
| 2812 | err = -1; |
| 2813 | break; |
| 2814 | } |
| 2815 | } |
| 2816 | |
| 2817 | if (err) |
| 2818 | SignalException(SimulatorFault,"Unrecognised FP format in ValueFPR()"); |
| 2819 | |
| 2820 | #ifdef DEBUG |
| 2821 | printf("DBG: ValueFPR: fpr = %d, fmt = %s, value = 0x%08X%08X : PC = 0x%08X%08X : SizeFGR() = %d\n",fpr,DOFMT(fmt),WORD64HI(value),WORD64LO(value),WORD64HI(IPC),WORD64LO(IPC),SizeFGR()); |
| 2822 | #endif /* DEBUG */ |
| 2823 | |
| 2824 | return(value); |
| 2825 | } |
| 2826 | |
| 2827 | static void |
| 2828 | StoreFPR(fpr,fmt,value) |
| 2829 | int fpr; |
| 2830 | FP_formats fmt; |
| 2831 | uword64 value; |
| 2832 | { |
| 2833 | int err = 0; |
| 2834 | |
| 2835 | #ifdef DEBUG |
| 2836 | printf("DBG: StoreFPR: fpr = %d, fmt = %s, value = 0x%08X%08X : PC = 0x%08X%08X : SizeFGR() = %d\n",fpr,DOFMT(fmt),WORD64HI(value),WORD64LO(value),WORD64HI(IPC),WORD64LO(IPC),SizeFGR()); |
| 2837 | #endif /* DEBUG */ |
| 2838 | |
| 2839 | if (SizeFGR() == 64) { |
| 2840 | switch (fmt) { |
| 2841 | case fmt_single : |
| 2842 | case fmt_word : |
| 2843 | FGR[fpr] = (((uword64)0xDEADC0DE << 32) | (value & 0xFFFFFFFF)); |
| 2844 | fpr_state[fpr] = fmt; |
| 2845 | break; |
| 2846 | |
| 2847 | case fmt_uninterpreted: |
| 2848 | case fmt_double : |
| 2849 | case fmt_long : |
| 2850 | FGR[fpr] = value; |
| 2851 | fpr_state[fpr] = fmt; |
| 2852 | break; |
| 2853 | |
| 2854 | default : |
| 2855 | fpr_state[fpr] = fmt_unknown; |
| 2856 | err = -1; |
| 2857 | break; |
| 2858 | } |
| 2859 | } else if ((fpr & 1) == 0) { /* even register number only */ |
| 2860 | switch (fmt) { |
| 2861 | case fmt_single : |
| 2862 | case fmt_word : |
| 2863 | FGR[fpr+1] = 0xDEADC0DE; |
| 2864 | FGR[fpr] = (value & 0xFFFFFFFF); |
| 2865 | fpr_state[fpr + 1] = fmt; |
| 2866 | fpr_state[fpr] = fmt; |
| 2867 | break; |
| 2868 | |
| 2869 | case fmt_uninterpreted: |
| 2870 | case fmt_double : |
| 2871 | case fmt_long : |
| 2872 | FGR[fpr+1] = (value >> 32); |
| 2873 | FGR[fpr] = (value & 0xFFFFFFFF); |
| 2874 | fpr_state[fpr + 1] = fmt; |
| 2875 | fpr_state[fpr] = fmt; |
| 2876 | break; |
| 2877 | |
| 2878 | default : |
| 2879 | fpr_state[fpr] = fmt_unknown; |
| 2880 | err = -1; |
| 2881 | break; |
| 2882 | } |
| 2883 | } |
| 2884 | #if defined(WARN_RESULT) |
| 2885 | else |
| 2886 | UndefinedResult(); |
| 2887 | #endif /* WARN_RESULT */ |
| 2888 | |
| 2889 | if (err) |
| 2890 | SignalException(SimulatorFault,"Unrecognised FP format in StoreFPR()"); |
| 2891 | |
| 2892 | #ifdef DEBUG |
| 2893 | printf("DBG: StoreFPR: fpr[%d] = 0x%08X%08X (format %s)\n",fpr,WORD64HI(FGR[fpr]),WORD64LO(FGR[fpr]),DOFMT(fmt)); |
| 2894 | #endif /* DEBUG */ |
| 2895 | |
| 2896 | return; |
| 2897 | } |
| 2898 | |
| 2899 | static int |
| 2900 | NaN(op,fmt) |
| 2901 | uword64 op; |
| 2902 | FP_formats fmt; |
| 2903 | { |
| 2904 | int boolean = 0; |
| 2905 | |
| 2906 | /* Check if (((E - bias) == (E_max + 1)) && (fraction != 0)). We |
| 2907 | know that the exponent field is biased... we we cheat and avoid |
| 2908 | removing the bias value. */ |
| 2909 | switch (fmt) { |
| 2910 | case fmt_single: |
| 2911 | boolean = ((FP_S_be(op) == 0xFF) && (FP_S_f(op) != 0)); |
| 2912 | /* We could use "FP_S_fb(1,op)" to ascertain whether we are |
| 2913 | dealing with a SNaN or QNaN */ |
| 2914 | break; |
| 2915 | case fmt_double: |
| 2916 | boolean = ((FP_D_be(op) == 0x7FF) && (FP_D_f(op) != 0)); |
| 2917 | /* We could use "FP_S_fb(1,op)" to ascertain whether we are |
| 2918 | dealing with a SNaN or QNaN */ |
| 2919 | break; |
| 2920 | case fmt_word: |
| 2921 | boolean = (op == FPQNaN_WORD); |
| 2922 | break; |
| 2923 | case fmt_long: |
| 2924 | boolean = (op == FPQNaN_LONG); |
| 2925 | break; |
| 2926 | } |
| 2927 | |
| 2928 | #ifdef DEBUG |
| 2929 | printf("DBG: NaN: returning %d for 0x%08X%08X (format = %s)\n",boolean,WORD64HI(op),WORD64LO(op),DOFMT(fmt)); |
| 2930 | #endif /* DEBUG */ |
| 2931 | |
| 2932 | return(boolean); |
| 2933 | } |
| 2934 | |
| 2935 | static int |
| 2936 | Infinity(op,fmt) |
| 2937 | uword64 op; |
| 2938 | FP_formats fmt; |
| 2939 | { |
| 2940 | int boolean = 0; |
| 2941 | |
| 2942 | #ifdef DEBUG |
| 2943 | printf("DBG: Infinity: format %s 0x%08X%08X (PC = 0x%08X%08X)\n",DOFMT(fmt),WORD64HI(op),WORD64LO(op),WORD64HI(IPC),WORD64LO(IPC)); |
| 2944 | #endif /* DEBUG */ |
| 2945 | |
| 2946 | /* Check if (((E - bias) == (E_max + 1)) && (fraction == 0)). We |
| 2947 | know that the exponent field is biased... we we cheat and avoid |
| 2948 | removing the bias value. */ |
| 2949 | switch (fmt) { |
| 2950 | case fmt_single: |
| 2951 | boolean = ((FP_S_be(op) == 0xFF) && (FP_S_f(op) == 0)); |
| 2952 | break; |
| 2953 | case fmt_double: |
| 2954 | boolean = ((FP_D_be(op) == 0x7FF) && (FP_D_f(op) == 0)); |
| 2955 | break; |
| 2956 | default: |
| 2957 | printf("DBG: TODO: unrecognised format (%s) for Infinity check\n",DOFMT(fmt)); |
| 2958 | break; |
| 2959 | } |
| 2960 | |
| 2961 | #ifdef DEBUG |
| 2962 | printf("DBG: Infinity: returning %d for 0x%08X%08X (format = %s)\n",boolean,WORD64HI(op),WORD64LO(op),DOFMT(fmt)); |
| 2963 | #endif /* DEBUG */ |
| 2964 | |
| 2965 | return(boolean); |
| 2966 | } |
| 2967 | |
| 2968 | static int |
| 2969 | Less(op1,op2,fmt) |
| 2970 | uword64 op1; |
| 2971 | uword64 op2; |
| 2972 | FP_formats fmt; |
| 2973 | { |
| 2974 | int boolean = 0; |
| 2975 | |
| 2976 | /* Argument checking already performed by the FPCOMPARE code */ |
| 2977 | |
| 2978 | #ifdef DEBUG |
| 2979 | printf("DBG: Less: %s: op1 = 0x%08X%08X : op2 = 0x%08X%08X\n",DOFMT(fmt),WORD64HI(op1),WORD64LO(op1),WORD64HI(op2),WORD64LO(op2)); |
| 2980 | #endif /* DEBUG */ |
| 2981 | |
| 2982 | /* The format type should already have been checked: */ |
| 2983 | switch (fmt) { |
| 2984 | case fmt_single: |
| 2985 | { |
| 2986 | unsigned int wop1 = (unsigned int)op1; |
| 2987 | unsigned int wop2 = (unsigned int)op2; |
| 2988 | boolean = (*(float *)&wop1 < *(float *)&wop2); |
| 2989 | } |
| 2990 | break; |
| 2991 | case fmt_double: |
| 2992 | boolean = (*(double *)&op1 < *(double *)&op2); |
| 2993 | break; |
| 2994 | } |
| 2995 | |
| 2996 | #ifdef DEBUG |
| 2997 | printf("DBG: Less: returning %d (format = %s)\n",boolean,DOFMT(fmt)); |
| 2998 | #endif /* DEBUG */ |
| 2999 | |
| 3000 | return(boolean); |
| 3001 | } |
| 3002 | |
| 3003 | static int |
| 3004 | Equal(op1,op2,fmt) |
| 3005 | uword64 op1; |
| 3006 | uword64 op2; |
| 3007 | FP_formats fmt; |
| 3008 | { |
| 3009 | int boolean = 0; |
| 3010 | |
| 3011 | /* Argument checking already performed by the FPCOMPARE code */ |
| 3012 | |
| 3013 | #ifdef DEBUG |
| 3014 | printf("DBG: Equal: %s: op1 = 0x%08X%08X : op2 = 0x%08X%08X\n",DOFMT(fmt),WORD64HI(op1),WORD64LO(op1),WORD64HI(op2),WORD64LO(op2)); |
| 3015 | #endif /* DEBUG */ |
| 3016 | |
| 3017 | /* The format type should already have been checked: */ |
| 3018 | switch (fmt) { |
| 3019 | case fmt_single: |
| 3020 | boolean = ((op1 & 0xFFFFFFFF) == (op2 & 0xFFFFFFFF)); |
| 3021 | break; |
| 3022 | case fmt_double: |
| 3023 | boolean = (op1 == op2); |
| 3024 | break; |
| 3025 | } |
| 3026 | |
| 3027 | #ifdef DEBUG |
| 3028 | printf("DBG: Equal: returning %d (format = %s)\n",boolean,DOFMT(fmt)); |
| 3029 | #endif /* DEBUG */ |
| 3030 | |
| 3031 | return(boolean); |
| 3032 | } |
| 3033 | |
| 3034 | static uword64 |
| 3035 | AbsoluteValue(op,fmt) |
| 3036 | uword64 op; |
| 3037 | FP_formats fmt; |
| 3038 | { |
| 3039 | uword64 result; |
| 3040 | |
| 3041 | #ifdef DEBUG |
| 3042 | printf("DBG: AbsoluteValue: %s: op = 0x%08X%08X\n",DOFMT(fmt),WORD64HI(op),WORD64LO(op)); |
| 3043 | #endif /* DEBUG */ |
| 3044 | |
| 3045 | /* The format type should already have been checked: */ |
| 3046 | switch (fmt) { |
| 3047 | case fmt_single: |
| 3048 | { |
| 3049 | unsigned int wop = (unsigned int)op; |
| 3050 | float tmp = ((float)fabs((double)*(float *)&wop)); |
| 3051 | result = (uword64)*(unsigned int *)&tmp; |
| 3052 | } |
| 3053 | break; |
| 3054 | case fmt_double: |
| 3055 | { |
| 3056 | double tmp = (fabs(*(double *)&op)); |
| 3057 | result = *(uword64 *)&tmp; |
| 3058 | } |
| 3059 | } |
| 3060 | |
| 3061 | return(result); |
| 3062 | } |
| 3063 | |
| 3064 | static uword64 |
| 3065 | Negate(op,fmt) |
| 3066 | uword64 op; |
| 3067 | FP_formats fmt; |
| 3068 | { |
| 3069 | uword64 result; |
| 3070 | |
| 3071 | #ifdef DEBUG |
| 3072 | printf("DBG: Negate: %s: op = 0x%08X%08X\n",DOFMT(fmt),WORD64HI(op),WORD64LO(op)); |
| 3073 | #endif /* DEBUG */ |
| 3074 | |
| 3075 | /* The format type should already have been checked: */ |
| 3076 | switch (fmt) { |
| 3077 | case fmt_single: |
| 3078 | { |
| 3079 | unsigned int wop = (unsigned int)op; |
| 3080 | float tmp = ((float)0.0 - *(float *)&wop); |
| 3081 | result = (uword64)*(unsigned int *)&tmp; |
| 3082 | } |
| 3083 | break; |
| 3084 | case fmt_double: |
| 3085 | { |
| 3086 | double tmp = ((double)0.0 - *(double *)&op); |
| 3087 | result = *(uword64 *)&tmp; |
| 3088 | } |
| 3089 | break; |
| 3090 | } |
| 3091 | |
| 3092 | return(result); |
| 3093 | } |
| 3094 | |
| 3095 | static uword64 |
| 3096 | Add(op1,op2,fmt) |
| 3097 | uword64 op1; |
| 3098 | uword64 op2; |
| 3099 | FP_formats fmt; |
| 3100 | { |
| 3101 | uword64 result; |
| 3102 | |
| 3103 | #ifdef DEBUG |
| 3104 | printf("DBG: Add: %s: op1 = 0x%08X%08X : op2 = 0x%08X%08X\n",DOFMT(fmt),WORD64HI(op1),WORD64LO(op1),WORD64HI(op2),WORD64LO(op2)); |
| 3105 | #endif /* DEBUG */ |
| 3106 | |
| 3107 | /* The registers must specify FPRs valid for operands of type |
| 3108 | "fmt". If they are not valid, the result is undefined. */ |
| 3109 | |
| 3110 | /* The format type should already have been checked: */ |
| 3111 | switch (fmt) { |
| 3112 | case fmt_single: |
| 3113 | { |
| 3114 | unsigned int wop1 = (unsigned int)op1; |
| 3115 | unsigned int wop2 = (unsigned int)op2; |
| 3116 | float tmp = (*(float *)&wop1 + *(float *)&wop2); |
| 3117 | result = (uword64)*(unsigned int *)&tmp; |
| 3118 | } |
| 3119 | break; |
| 3120 | case fmt_double: |
| 3121 | { |
| 3122 | double tmp = (*(double *)&op1 + *(double *)&op2); |
| 3123 | result = *(uword64 *)&tmp; |
| 3124 | } |
| 3125 | break; |
| 3126 | } |
| 3127 | |
| 3128 | #ifdef DEBUG |
| 3129 | printf("DBG: Add: returning 0x%08X%08X (format = %s)\n",WORD64HI(result),WORD64LO(result),DOFMT(fmt)); |
| 3130 | #endif /* DEBUG */ |
| 3131 | |
| 3132 | return(result); |
| 3133 | } |
| 3134 | |
| 3135 | static uword64 |
| 3136 | Sub(op1,op2,fmt) |
| 3137 | uword64 op1; |
| 3138 | uword64 op2; |
| 3139 | FP_formats fmt; |
| 3140 | { |
| 3141 | uword64 result; |
| 3142 | |
| 3143 | #ifdef DEBUG |
| 3144 | printf("DBG: Sub: %s: op1 = 0x%08X%08X : op2 = 0x%08X%08X\n",DOFMT(fmt),WORD64HI(op1),WORD64LO(op1),WORD64HI(op2),WORD64LO(op2)); |
| 3145 | #endif /* DEBUG */ |
| 3146 | |
| 3147 | /* The registers must specify FPRs valid for operands of type |
| 3148 | "fmt". If they are not valid, the result is undefined. */ |
| 3149 | |
| 3150 | /* The format type should already have been checked: */ |
| 3151 | switch (fmt) { |
| 3152 | case fmt_single: |
| 3153 | { |
| 3154 | unsigned int wop1 = (unsigned int)op1; |
| 3155 | unsigned int wop2 = (unsigned int)op2; |
| 3156 | float tmp = (*(float *)&wop1 - *(float *)&wop2); |
| 3157 | result = (uword64)*(unsigned int *)&tmp; |
| 3158 | } |
| 3159 | break; |
| 3160 | case fmt_double: |
| 3161 | { |
| 3162 | double tmp = (*(double *)&op1 - *(double *)&op2); |
| 3163 | result = *(uword64 *)&tmp; |
| 3164 | } |
| 3165 | break; |
| 3166 | } |
| 3167 | |
| 3168 | #ifdef DEBUG |
| 3169 | printf("DBG: Sub: returning 0x%08X%08X (format = %s)\n",WORD64HI(result),WORD64LO(result),DOFMT(fmt)); |
| 3170 | #endif /* DEBUG */ |
| 3171 | |
| 3172 | return(result); |
| 3173 | } |
| 3174 | |
| 3175 | static uword64 |
| 3176 | Multiply(op1,op2,fmt) |
| 3177 | uword64 op1; |
| 3178 | uword64 op2; |
| 3179 | FP_formats fmt; |
| 3180 | { |
| 3181 | uword64 result; |
| 3182 | |
| 3183 | #ifdef DEBUG |
| 3184 | printf("DBG: Multiply: %s: op1 = 0x%08X%08X : op2 = 0x%08X%08X\n",DOFMT(fmt),WORD64HI(op1),WORD64LO(op1),WORD64HI(op2),WORD64LO(op2)); |
| 3185 | #endif /* DEBUG */ |
| 3186 | |
| 3187 | /* The registers must specify FPRs valid for operands of type |
| 3188 | "fmt". If they are not valid, the result is undefined. */ |
| 3189 | |
| 3190 | /* The format type should already have been checked: */ |
| 3191 | switch (fmt) { |
| 3192 | case fmt_single: |
| 3193 | { |
| 3194 | unsigned int wop1 = (unsigned int)op1; |
| 3195 | unsigned int wop2 = (unsigned int)op2; |
| 3196 | float tmp = (*(float *)&wop1 * *(float *)&wop2); |
| 3197 | result = (uword64)*(unsigned int *)&tmp; |
| 3198 | } |
| 3199 | break; |
| 3200 | case fmt_double: |
| 3201 | { |
| 3202 | double tmp = (*(double *)&op1 * *(double *)&op2); |
| 3203 | result = *(uword64 *)&tmp; |
| 3204 | } |
| 3205 | break; |
| 3206 | } |
| 3207 | |
| 3208 | #ifdef DEBUG |
| 3209 | printf("DBG: Multiply: returning 0x%08X%08X (format = %s)\n",WORD64HI(result),WORD64LO(result),DOFMT(fmt)); |
| 3210 | #endif /* DEBUG */ |
| 3211 | |
| 3212 | return(result); |
| 3213 | } |
| 3214 | |
| 3215 | static uword64 |
| 3216 | Divide(op1,op2,fmt) |
| 3217 | uword64 op1; |
| 3218 | uword64 op2; |
| 3219 | FP_formats fmt; |
| 3220 | { |
| 3221 | uword64 result; |
| 3222 | |
| 3223 | #ifdef DEBUG |
| 3224 | printf("DBG: Divide: %s: op1 = 0x%08X%08X : op2 = 0x%08X%08X\n",DOFMT(fmt),WORD64HI(op1),WORD64LO(op1),WORD64HI(op2),WORD64LO(op2)); |
| 3225 | #endif /* DEBUG */ |
| 3226 | |
| 3227 | /* The registers must specify FPRs valid for operands of type |
| 3228 | "fmt". If they are not valid, the result is undefined. */ |
| 3229 | |
| 3230 | /* The format type should already have been checked: */ |
| 3231 | switch (fmt) { |
| 3232 | case fmt_single: |
| 3233 | { |
| 3234 | unsigned int wop1 = (unsigned int)op1; |
| 3235 | unsigned int wop2 = (unsigned int)op2; |
| 3236 | float tmp = (*(float *)&wop1 / *(float *)&wop2); |
| 3237 | result = (uword64)*(unsigned int *)&tmp; |
| 3238 | } |
| 3239 | break; |
| 3240 | case fmt_double: |
| 3241 | { |
| 3242 | double tmp = (*(double *)&op1 / *(double *)&op2); |
| 3243 | result = *(uword64 *)&tmp; |
| 3244 | } |
| 3245 | break; |
| 3246 | } |
| 3247 | |
| 3248 | #ifdef DEBUG |
| 3249 | printf("DBG: Divide: returning 0x%08X%08X (format = %s)\n",WORD64HI(result),WORD64LO(result),DOFMT(fmt)); |
| 3250 | #endif /* DEBUG */ |
| 3251 | |
| 3252 | return(result); |
| 3253 | } |
| 3254 | |
| 3255 | static uword64 |
| 3256 | Recip(op,fmt) |
| 3257 | uword64 op; |
| 3258 | FP_formats fmt; |
| 3259 | { |
| 3260 | uword64 result; |
| 3261 | |
| 3262 | #ifdef DEBUG |
| 3263 | printf("DBG: Recip: %s: op = 0x%08X%08X\n",DOFMT(fmt),WORD64HI(op),WORD64LO(op)); |
| 3264 | #endif /* DEBUG */ |
| 3265 | |
| 3266 | /* The registers must specify FPRs valid for operands of type |
| 3267 | "fmt". If they are not valid, the result is undefined. */ |
| 3268 | |
| 3269 | /* The format type should already have been checked: */ |
| 3270 | switch (fmt) { |
| 3271 | case fmt_single: |
| 3272 | { |
| 3273 | unsigned int wop = (unsigned int)op; |
| 3274 | float tmp = ((float)1.0 / *(float *)&wop); |
| 3275 | result = (uword64)*(unsigned int *)&tmp; |
| 3276 | } |
| 3277 | break; |
| 3278 | case fmt_double: |
| 3279 | { |
| 3280 | double tmp = ((double)1.0 / *(double *)&op); |
| 3281 | result = *(uword64 *)&tmp; |
| 3282 | } |
| 3283 | break; |
| 3284 | } |
| 3285 | |
| 3286 | #ifdef DEBUG |
| 3287 | printf("DBG: Recip: returning 0x%08X%08X (format = %s)\n",WORD64HI(result),WORD64LO(result),DOFMT(fmt)); |
| 3288 | #endif /* DEBUG */ |
| 3289 | |
| 3290 | return(result); |
| 3291 | } |
| 3292 | |
| 3293 | static uword64 |
| 3294 | SquareRoot(op,fmt) |
| 3295 | uword64 op; |
| 3296 | FP_formats fmt; |
| 3297 | { |
| 3298 | uword64 result; |
| 3299 | |
| 3300 | #ifdef DEBUG |
| 3301 | printf("DBG: SquareRoot: %s: op = 0x%08X%08X\n",DOFMT(fmt),WORD64HI(op),WORD64LO(op)); |
| 3302 | #endif /* DEBUG */ |
| 3303 | |
| 3304 | /* The registers must specify FPRs valid for operands of type |
| 3305 | "fmt". If they are not valid, the result is undefined. */ |
| 3306 | |
| 3307 | /* The format type should already have been checked: */ |
| 3308 | switch (fmt) { |
| 3309 | case fmt_single: |
| 3310 | { |
| 3311 | unsigned int wop = (unsigned int)op; |
| 3312 | float tmp = ((float)sqrt((double)*(float *)&wop)); |
| 3313 | result = (uword64)*(unsigned int *)&tmp; |
| 3314 | } |
| 3315 | break; |
| 3316 | case fmt_double: |
| 3317 | { |
| 3318 | double tmp = (sqrt(*(double *)&op)); |
| 3319 | result = *(uword64 *)&tmp; |
| 3320 | } |
| 3321 | break; |
| 3322 | } |
| 3323 | |
| 3324 | #ifdef DEBUG |
| 3325 | printf("DBG: SquareRoot: returning 0x%08X%08X (format = %s)\n",WORD64HI(result),WORD64LO(result),DOFMT(fmt)); |
| 3326 | #endif /* DEBUG */ |
| 3327 | |
| 3328 | return(result); |
| 3329 | } |
| 3330 | |
| 3331 | static uword64 |
| 3332 | Convert(rm,op,from,to) |
| 3333 | int rm; |
| 3334 | uword64 op; |
| 3335 | FP_formats from; |
| 3336 | FP_formats to; |
| 3337 | { |
| 3338 | uword64 result; |
| 3339 | |
| 3340 | #ifdef DEBUG |
| 3341 | printf("DBG: Convert: mode %s : op 0x%08X%08X : from %s : to %s : (PC = 0x%08X%08X)\n",RMMODE(rm),WORD64HI(op),WORD64LO(op),DOFMT(from),DOFMT(to),WORD64HI(IPC),WORD64LO(IPC)); |
| 3342 | #endif /* DEBUG */ |
| 3343 | |
| 3344 | /* The value "op" is converted to the destination format, rounding |
| 3345 | using mode "rm". When the destination is a fixed-point format, |
| 3346 | then a source value of Infinity, NaN or one which would round to |
| 3347 | an integer outside the fixed point range then an IEEE Invalid |
| 3348 | Operation condition is raised. */ |
| 3349 | switch (to) { |
| 3350 | case fmt_single: |
| 3351 | { |
| 3352 | float tmp; |
| 3353 | switch (from) { |
| 3354 | case fmt_double: |
| 3355 | tmp = (float)(*(double *)&op); |
| 3356 | break; |
| 3357 | |
| 3358 | case fmt_word: |
| 3359 | tmp = (float)((int)(op & 0xFFFFFFFF)); |
| 3360 | break; |
| 3361 | |
| 3362 | case fmt_long: |
| 3363 | tmp = (float)((int)op); |
| 3364 | break; |
| 3365 | } |
| 3366 | |
| 3367 | switch (rm) { |
| 3368 | case FP_RM_NEAREST: |
| 3369 | /* Round result to nearest representable value. When two |
| 3370 | representable values are equally near, round to the value |
| 3371 | that has a least significant bit of zero (i.e. is even). */ |
| 3372 | #ifdef HAVE_ANINT |
| 3373 | tmp = (float)anint((double)tmp); |
| 3374 | #else |
| 3375 | /* TODO: Provide round-to-nearest */ |
| 3376 | #endif |
| 3377 | break; |
| 3378 | |
| 3379 | case FP_RM_TOZERO: |
| 3380 | /* Round result to the value closest to, and not greater in |
| 3381 | magnitude than, the result. */ |
| 3382 | #ifdef HAVE_AINT |
| 3383 | tmp = (float)aint((double)tmp); |
| 3384 | #else |
| 3385 | /* TODO: Provide round-to-zero */ |
| 3386 | #endif |
| 3387 | break; |
| 3388 | |
| 3389 | case FP_RM_TOPINF: |
| 3390 | /* Round result to the value closest to, and not less than, |
| 3391 | the result. */ |
| 3392 | tmp = (float)ceil((double)tmp); |
| 3393 | break; |
| 3394 | |
| 3395 | case FP_RM_TOMINF: |
| 3396 | /* Round result to the value closest to, and not greater than, |
| 3397 | the result. */ |
| 3398 | tmp = (float)floor((double)tmp); |
| 3399 | break; |
| 3400 | } |
| 3401 | result = (uword64)*(unsigned int *)&tmp; |
| 3402 | } |
| 3403 | break; |
| 3404 | |
| 3405 | case fmt_double: |
| 3406 | { |
| 3407 | double tmp; |
| 3408 | word64 xxx; |
| 3409 | |
| 3410 | switch (from) { |
| 3411 | case fmt_single: |
| 3412 | { |
| 3413 | unsigned int wop = (unsigned int)op; |
| 3414 | tmp = (double)(*(float *)&wop); |
| 3415 | } |
| 3416 | break; |
| 3417 | |
| 3418 | case fmt_word: |
| 3419 | xxx = SIGNEXTEND((op & 0xFFFFFFFF),32); |
| 3420 | tmp = xxx; |
| 3421 | break; |
| 3422 | |
| 3423 | case fmt_long: |
| 3424 | tmp = (double)((word64)op); |
| 3425 | break; |
| 3426 | } |
| 3427 | |
| 3428 | switch (rm) { |
| 3429 | case FP_RM_NEAREST: |
| 3430 | #ifdef HAVE_ANINT |
| 3431 | tmp = anint(*(double *)&tmp); |
| 3432 | #else |
| 3433 | /* TODO: Provide round-to-nearest */ |
| 3434 | #endif |
| 3435 | break; |
| 3436 | |
| 3437 | case FP_RM_TOZERO: |
| 3438 | #ifdef HAVE_AINT |
| 3439 | tmp = aint(*(double *)&tmp); |
| 3440 | #else |
| 3441 | /* TODO: Provide round-to-zero */ |
| 3442 | #endif |
| 3443 | break; |
| 3444 | |
| 3445 | case FP_RM_TOPINF: |
| 3446 | tmp = ceil(*(double *)&tmp); |
| 3447 | break; |
| 3448 | |
| 3449 | case FP_RM_TOMINF: |
| 3450 | tmp = floor(*(double *)&tmp); |
| 3451 | break; |
| 3452 | } |
| 3453 | result = *(uword64 *)&tmp; |
| 3454 | } |
| 3455 | break; |
| 3456 | |
| 3457 | case fmt_word: |
| 3458 | case fmt_long: |
| 3459 | if (Infinity(op,from) || NaN(op,from) || (1 == 0/*TODO: check range */)) { |
| 3460 | printf("DBG: TODO: update FCSR\n"); |
| 3461 | SignalException(FPE); |
| 3462 | } else { |
| 3463 | if (to == fmt_word) { |
| 3464 | unsigned int tmp; |
| 3465 | switch (from) { |
| 3466 | case fmt_single: |
| 3467 | { |
| 3468 | unsigned int wop = (unsigned int)op; |
| 3469 | tmp = (unsigned int)*((float *)&wop); |
| 3470 | } |
| 3471 | break; |
| 3472 | case fmt_double: |
| 3473 | tmp = (unsigned int)*((double *)&op); |
| 3474 | #ifdef DEBUG |
| 3475 | printf("DBG: from double %.30f (0x%08X%08X) to word: 0x%08X\n",*((double *)&op),WORD64HI(op),WORD64LO(op),tmp); |
| 3476 | #endif /* DEBUG */ |
| 3477 | break; |
| 3478 | } |
| 3479 | result = (uword64)tmp; |
| 3480 | } else { /* fmt_long */ |
| 3481 | switch (from) { |
| 3482 | case fmt_single: |
| 3483 | { |
| 3484 | unsigned int wop = (unsigned int)op; |
| 3485 | result = (uword64)*((float *)&wop); |
| 3486 | } |
| 3487 | break; |
| 3488 | case fmt_double: |
| 3489 | result = (uword64)*((double *)&op); |
| 3490 | break; |
| 3491 | } |
| 3492 | } |
| 3493 | } |
| 3494 | break; |
| 3495 | } |
| 3496 | |
| 3497 | #ifdef DEBUG |
| 3498 | printf("DBG: Convert: returning 0x%08X%08X (to format = %s)\n",WORD64HI(result),WORD64LO(result),DOFMT(to)); |
| 3499 | #endif /* DEBUG */ |
| 3500 | |
| 3501 | return(result); |
| 3502 | } |
| 3503 | #endif /* HASFPU */ |
| 3504 | |
| 3505 | /*-- co-processor support routines ------------------------------------------*/ |
| 3506 | |
| 3507 | static int |
| 3508 | CoProcPresent(coproc_number) |
| 3509 | unsigned int coproc_number; |
| 3510 | { |
| 3511 | /* Return TRUE if simulator provides a model for the given co-processor number */ |
| 3512 | return(0); |
| 3513 | } |
| 3514 | |
| 3515 | static void |
| 3516 | COP_LW(coproc_num,coproc_reg,memword) |
| 3517 | int coproc_num, coproc_reg; |
| 3518 | unsigned int memword; |
| 3519 | { |
| 3520 | switch (coproc_num) { |
| 3521 | #if defined(HASFPU) |
| 3522 | case 1: |
| 3523 | #ifdef DEBUG |
| 3524 | printf("DBG: COP_LW: memword = 0x%08X (uword64)memword = 0x%08X%08X\n",memword,WORD64HI(memword),WORD64LO(memword)); |
| 3525 | #endif |
| 3526 | StoreFPR(coproc_reg,fmt_uninterpreted,(uword64)memword); |
| 3527 | break; |
| 3528 | #endif /* HASFPU */ |
| 3529 | |
| 3530 | default: |
| 3531 | #if 0 /* this should be controlled by a configuration option */ |
| 3532 | callback->printf_filtered(callback,"COP_LW(%d,%d,0x%08X) at IPC = 0x%08X%08X : TODO (architecture specific)\n",coproc_num,coproc_reg,memword,WORD64HI(IPC),WORD64LO(IPC)); |
| 3533 | #endif |
| 3534 | break; |
| 3535 | } |
| 3536 | |
| 3537 | return; |
| 3538 | } |
| 3539 | |
| 3540 | static void |
| 3541 | COP_LD(coproc_num,coproc_reg,memword) |
| 3542 | int coproc_num, coproc_reg; |
| 3543 | uword64 memword; |
| 3544 | { |
| 3545 | switch (coproc_num) { |
| 3546 | #if defined(HASFPU) |
| 3547 | case 1: |
| 3548 | StoreFPR(coproc_reg,fmt_uninterpreted,memword); |
| 3549 | break; |
| 3550 | #endif /* HASFPU */ |
| 3551 | |
| 3552 | default: |
| 3553 | #if 0 /* this message should be controlled by a configuration option */ |
| 3554 | callback->printf_filtered(callback,"COP_LD(%d,%d,0x%08X%08X) at IPC = 0x%08X%08X : TODO (architecture specific)\n",coproc_num,coproc_reg,WORD64HI(memword),WORD64LO(memword),WORD64HI(IPC),WORD64LO(IPC)); |
| 3555 | #endif |
| 3556 | break; |
| 3557 | } |
| 3558 | |
| 3559 | return; |
| 3560 | } |
| 3561 | |
| 3562 | static unsigned int |
| 3563 | COP_SW(coproc_num,coproc_reg) |
| 3564 | int coproc_num, coproc_reg; |
| 3565 | { |
| 3566 | unsigned int value = 0; |
| 3567 | switch (coproc_num) { |
| 3568 | #if defined(HASFPU) |
| 3569 | case 1: |
| 3570 | #if 1 |
| 3571 | value = (unsigned int)ValueFPR(coproc_reg,fmt_uninterpreted); |
| 3572 | #else |
| 3573 | #if 1 |
| 3574 | value = (unsigned int)ValueFPR(coproc_reg,fpr_state[coproc_reg]); |
| 3575 | #else |
| 3576 | #ifdef DEBUG |
| 3577 | printf("DBG: COP_SW: reg in format %s (will be accessing as single)\n",DOFMT(fpr_state[coproc_reg])); |
| 3578 | #endif /* DEBUG */ |
| 3579 | value = (unsigned int)ValueFPR(coproc_reg,fmt_single); |
| 3580 | #endif |
| 3581 | #endif |
| 3582 | break; |
| 3583 | #endif /* HASFPU */ |
| 3584 | |
| 3585 | default: |
| 3586 | #if 0 /* should be controlled by configuration option */ |
| 3587 | callback->printf_filtered(callback,"COP_SW(%d,%d) at IPC = 0x%08X%08X : TODO (architecture specific)\n",coproc_num,coproc_reg,WORD64HI(IPC),WORD64LO(IPC)); |
| 3588 | #endif |
| 3589 | break; |
| 3590 | } |
| 3591 | |
| 3592 | return(value); |
| 3593 | } |
| 3594 | |
| 3595 | static uword64 |
| 3596 | COP_SD(coproc_num,coproc_reg) |
| 3597 | int coproc_num, coproc_reg; |
| 3598 | { |
| 3599 | uword64 value = 0; |
| 3600 | switch (coproc_num) { |
| 3601 | #if defined(HASFPU) |
| 3602 | case 1: |
| 3603 | #if 1 |
| 3604 | value = ValueFPR(coproc_reg,fmt_uninterpreted); |
| 3605 | #else |
| 3606 | #if 1 |
| 3607 | value = ValueFPR(coproc_reg,fpr_state[coproc_reg]); |
| 3608 | #else |
| 3609 | #ifdef DEBUG |
| 3610 | printf("DBG: COP_SD: reg in format %s (will be accessing as double)\n",DOFMT(fpr_state[coproc_reg])); |
| 3611 | #endif /* DEBUG */ |
| 3612 | value = ValueFPR(coproc_reg,fmt_double); |
| 3613 | #endif |
| 3614 | #endif |
| 3615 | break; |
| 3616 | #endif /* HASFPU */ |
| 3617 | |
| 3618 | default: |
| 3619 | #if 0 /* should be controlled by configuration option */ |
| 3620 | callback->printf_filtered(callback,"COP_SD(%d,%d) at IPC = 0x%08X%08X : TODO (architecture specific)\n",coproc_num,coproc_reg,WORD64HI(IPC),WORD64LO(IPC)); |
| 3621 | #endif |
| 3622 | break; |
| 3623 | } |
| 3624 | |
| 3625 | return(value); |
| 3626 | } |
| 3627 | |
| 3628 | static void |
| 3629 | decode_coproc(instruction) |
| 3630 | unsigned int instruction; |
| 3631 | { |
| 3632 | int coprocnum = ((instruction >> 26) & 3); |
| 3633 | |
| 3634 | switch (coprocnum) { |
| 3635 | case 0: /* standard CPU control and cache registers */ |
| 3636 | { |
| 3637 | /* NOTEs: |
| 3638 | Standard CP0 registers |
| 3639 | 0 = Index R4000 VR4100 VR4300 |
| 3640 | 1 = Random R4000 VR4100 VR4300 |
| 3641 | 2 = EntryLo0 R4000 VR4100 VR4300 |
| 3642 | 3 = EntryLo1 R4000 VR4100 VR4300 |
| 3643 | 4 = Context R4000 VR4100 VR4300 |
| 3644 | 5 = PageMask R4000 VR4100 VR4300 |
| 3645 | 6 = Wired R4000 VR4100 VR4300 |
| 3646 | 8 = BadVAddr R4000 VR4100 VR4300 |
| 3647 | 9 = Count R4000 VR4100 VR4300 |
| 3648 | 10 = EntryHi R4000 VR4100 VR4300 |
| 3649 | 11 = Compare R4000 VR4100 VR4300 |
| 3650 | 12 = SR R4000 VR4100 VR4300 |
| 3651 | 13 = Cause R4000 VR4100 VR4300 |
| 3652 | 14 = EPC R4000 VR4100 VR4300 |
| 3653 | 15 = PRId R4000 VR4100 VR4300 |
| 3654 | 16 = Config R4000 VR4100 VR4300 |
| 3655 | 17 = LLAddr R4000 VR4100 VR4300 |
| 3656 | 18 = WatchLo R4000 VR4100 VR4300 |
| 3657 | 19 = WatchHi R4000 VR4100 VR4300 |
| 3658 | 20 = XContext R4000 VR4100 VR4300 |
| 3659 | 26 = PErr or ECC R4000 VR4100 VR4300 |
| 3660 | 27 = CacheErr R4000 VR4100 |
| 3661 | 28 = TagLo R4000 VR4100 VR4300 |
| 3662 | 29 = TagHi R4000 VR4100 VR4300 |
| 3663 | 30 = ErrorEPC R4000 VR4100 VR4300 |
| 3664 | */ |
| 3665 | int code = ((instruction >> 21) & 0x1F); |
| 3666 | /* R4000 Users Manual (second edition) lists the following CP0 |
| 3667 | instructions: |
| 3668 | DMFC0 Doubleword Move From CP0 (VR4100 = 01000000001tttttddddd00000000000) |
| 3669 | DMTC0 Doubleword Move To CP0 (VR4100 = 01000000101tttttddddd00000000000) |
| 3670 | MFC0 word Move From CP0 (VR4100 = 01000000000tttttddddd00000000000) |
| 3671 | MTC0 word Move To CP0 (VR4100 = 01000000100tttttddddd00000000000) |
| 3672 | TLBR Read Indexed TLB Entry (VR4100 = 01000010000000000000000000000001) |
| 3673 | TLBWI Write Indexed TLB Entry (VR4100 = 01000010000000000000000000000010) |
| 3674 | TLBWR Write Random TLB Entry (VR4100 = 01000010000000000000000000000110) |
| 3675 | TLBP Probe TLB for Matching Entry (VR4100 = 01000010000000000000000000001000) |
| 3676 | CACHE Cache operation (VR4100 = 101111bbbbbpppppiiiiiiiiiiiiiiii) |
| 3677 | ERET Exception return (VR4100 = 01000010000000000000000000011000) |
| 3678 | */ |
| 3679 | if (((code == 0x00) || (code == 0x04)) && ((instruction & 0x7FF) == 0)) { |
| 3680 | int rt = ((instruction >> 16) & 0x1F); |
| 3681 | int rd = ((instruction >> 11) & 0x1F); |
| 3682 | if (code == 0x00) { /* MF : move from */ |
| 3683 | #if 0 /* message should be controlled by configuration option */ |
| 3684 | callback->printf_filtered(callback,"Warning: MFC0 %d,%d not handled yet (architecture specific)\n",rt,rd); |
| 3685 | #endif |
| 3686 | GPR[rt] = 0xDEADC0DE; /* CPR[0,rd] */ |
| 3687 | } else { /* MT : move to */ |
| 3688 | /* CPR[0,rd] = GPR[rt]; */ |
| 3689 | #if 0 /* should be controlled by configuration option */ |
| 3690 | callback->printf_filtered(callback,"Warning: MTC0 %d,%d not handled yet (architecture specific)\n",rt,rd); |
| 3691 | #endif |
| 3692 | } |
| 3693 | } else |
| 3694 | sim_warning("Unrecognised COP0 instruction 0x%08X at IPC = 0x%08X%08X : No handler present",instruction,WORD64HI(IPC),WORD64LO(IPC)); |
| 3695 | /* TODO: When executing an ERET or RFE instruction we should |
| 3696 | clear LLBIT, to ensure that any out-standing atomic |
| 3697 | read/modify/write sequence fails. */ |
| 3698 | } |
| 3699 | break; |
| 3700 | |
| 3701 | case 2: /* undefined co-processor */ |
| 3702 | sim_warning("COP2 instruction 0x%08X at IPC = 0x%08X%08X : No handler present",instruction,WORD64HI(IPC),WORD64LO(IPC)); |
| 3703 | break; |
| 3704 | |
| 3705 | case 1: /* should not occur (FPU co-processor) */ |
| 3706 | case 3: /* should not occur (FPU co-processor) */ |
| 3707 | SignalException(ReservedInstruction,instruction); |
| 3708 | break; |
| 3709 | } |
| 3710 | |
| 3711 | return; |
| 3712 | } |
| 3713 | |
| 3714 | /*-- instruction simulation -------------------------------------------------*/ |
| 3715 | |
| 3716 | static void |
| 3717 | simulate () |
| 3718 | { |
| 3719 | unsigned int pipeline_count = 1; |
| 3720 | |
| 3721 | #ifdef DEBUG |
| 3722 | if (membank == NULL) { |
| 3723 | printf("DBG: simulate() entered with no memory\n"); |
| 3724 | exit(1); |
| 3725 | } |
| 3726 | #endif /* DEBUG */ |
| 3727 | |
| 3728 | #if 0 /* Disabled to check that everything works OK */ |
| 3729 | /* The VR4300 seems to sign-extend the PC on its first |
| 3730 | access. However, this may just be because it is currently |
| 3731 | configured in 32bit mode. However... */ |
| 3732 | PC = SIGNEXTEND(PC,32); |
| 3733 | #endif |
| 3734 | |
| 3735 | /* main controlling loop */ |
| 3736 | do { |
| 3737 | /* Fetch the next instruction from the simulator memory: */ |
| 3738 | uword64 vaddr = (uword64)PC; |
| 3739 | uword64 paddr; |
| 3740 | int cca; |
| 3741 | unsigned int instruction; |
| 3742 | int dsstate = (state & simDELAYSLOT); |
| 3743 | |
| 3744 | #ifdef DEBUG |
| 3745 | { |
| 3746 | printf("DBG: state = 0x%08X :",state); |
| 3747 | if (state & simSTOP) printf(" simSTOP"); |
| 3748 | if (state & simSTEP) printf(" simSTEP"); |
| 3749 | if (state & simHALTEX) printf(" simHALTEX"); |
| 3750 | if (state & simHALTIN) printf(" simHALTIN"); |
| 3751 | if (state & simBE) printf(" simBE"); |
| 3752 | } |
| 3753 | #endif /* DEBUG */ |
| 3754 | |
| 3755 | #ifdef DEBUG |
| 3756 | if (dsstate) |
| 3757 | callback->printf_filtered(callback,"DBG: DSPC = 0x%08X%08X\n",WORD64HI(DSPC),WORD64LO(DSPC)); |
| 3758 | #endif /* DEBUG */ |
| 3759 | |
| 3760 | if (AddressTranslation(PC,isINSTRUCTION,isLOAD,&paddr,&cca,isTARGET,isREAL)) { /* Copy the action of the LW instruction */ |
| 3761 | unsigned int reverse = (ReverseEndian ? (LOADDRMASK >> 2) : 0); |
| 3762 | unsigned int bigend = (BigEndianCPU ? (LOADDRMASK >> 2) : 0); |
| 3763 | uword64 value; |
| 3764 | unsigned int byte; |
| 3765 | paddr = ((paddr & ~LOADDRMASK) | ((paddr & LOADDRMASK) ^ (reverse << 2))); |
| 3766 | value = LoadMemory(cca,AccessLength_WORD,paddr,vaddr,isINSTRUCTION,isREAL); |
| 3767 | byte = ((vaddr & LOADDRMASK) ^ (bigend << 2)); |
| 3768 | instruction = ((value >> (8 * byte)) & 0xFFFFFFFF); |
| 3769 | } else { |
| 3770 | fprintf(stderr,"Cannot translate address for PC = 0x%08X%08X failed\n",WORD64HI(PC),WORD64LO(PC)); |
| 3771 | exit(1); |
| 3772 | } |
| 3773 | |
| 3774 | #ifdef DEBUG |
| 3775 | callback->printf_filtered(callback,"DBG: fetched 0x%08X from PC = 0x%08X%08X\n",instruction,WORD64HI(PC),WORD64LO(PC)); |
| 3776 | #endif /* DEBUG */ |
| 3777 | |
| 3778 | #if !defined(FASTSIM) || defined(PROFILE) |
| 3779 | instruction_fetches++; |
| 3780 | /* Since we increment above, the value should only ever be zero if |
| 3781 | we have just overflowed: */ |
| 3782 | if (instruction_fetches == 0) |
| 3783 | instruction_fetch_overflow++; |
| 3784 | #if defined(PROFILE) |
| 3785 | if ((state & simPROFILE) && ((instruction_fetches % profile_frequency) == 0) && profile_hist) { |
| 3786 | int n = ((unsigned int)(PC - profile_minpc) >> (profile_shift + 2)); |
| 3787 | if (n < profile_nsamples) { |
| 3788 | /* NOTE: The counts for the profiling bins are only 16bits wide */ |
| 3789 | if (profile_hist[n] != USHRT_MAX) |
| 3790 | (profile_hist[n])++; |
| 3791 | } |
| 3792 | } |
| 3793 | #endif /* PROFILE */ |
| 3794 | #endif /* !FASTSIM && PROFILE */ |
| 3795 | |
| 3796 | IPC = PC; /* copy PC for this instruction */ |
| 3797 | /* This is required by exception processing, to ensure that we can |
| 3798 | cope with exceptions in the delay slots of branches that may |
| 3799 | already have changed the PC. */ |
| 3800 | PC += 4; /* increment ready for the next fetch */ |
| 3801 | /* NOTE: If we perform a delay slot change to the PC, this |
| 3802 | increment is not requuired. However, it would make the |
| 3803 | simulator more complicated to try and avoid this small hit. */ |
| 3804 | |
| 3805 | /* Currently this code provides a simple model. For more |
| 3806 | complicated models we could perform exception status checks at |
| 3807 | this point, and set the simSTOP state as required. This could |
| 3808 | also include processing any hardware interrupts raised by any |
| 3809 | I/O model attached to the simulator context. |
| 3810 | |
| 3811 | Support for "asynchronous" I/O events within the simulated world |
| 3812 | could be providing by managing a counter, and calling a I/O |
| 3813 | specific handler when a particular threshold is reached. On most |
| 3814 | architectures a decrement and check for zero operation is |
| 3815 | usually quicker than an increment and compare. However, the |
| 3816 | process of managing a known value decrement to zero, is higher |
| 3817 | than the cost of using an explicit value UINT_MAX into the |
| 3818 | future. Which system is used will depend on how complicated the |
| 3819 | I/O model is, and how much it is likely to affect the simulator |
| 3820 | bandwidth. |
| 3821 | |
| 3822 | If events need to be scheduled further in the future than |
| 3823 | UINT_MAX event ticks, then the I/O model should just provide its |
| 3824 | own counter, triggered from the event system. */ |
| 3825 | |
| 3826 | /* MIPS pipeline ticks. To allow for future support where the |
| 3827 | pipeline hit of individual instructions is known, this control |
| 3828 | loop manages a "pipeline_count" variable. It is initialised to |
| 3829 | 1 (one), and will only be changed by the simulator engine when |
| 3830 | executing an instruction. If the engine does not have access to |
| 3831 | pipeline cycle count information then all instructions will be |
| 3832 | treated as using a single cycle. NOTE: A standard system is not |
| 3833 | provided by the default simulator because different MIPS |
| 3834 | architectures have different cycle counts for the same |
| 3835 | instructions. */ |
| 3836 | |
| 3837 | #if defined(HASFPU) |
| 3838 | /* Set previous flag, depending on current: */ |
| 3839 | if (state & simPCOC0) |
| 3840 | state |= simPCOC1; |
| 3841 | else |
| 3842 | state &= ~simPCOC1; |
| 3843 | /* and update the current value: */ |
| 3844 | if (GETFCC(0)) |
| 3845 | state |= simPCOC0; |
| 3846 | else |
| 3847 | state &= ~simPCOC0; |
| 3848 | #endif /* HASFPU */ |
| 3849 | |
| 3850 | /* NOTE: For multi-context simulation environments the "instruction" |
| 3851 | variable should be local to this routine. */ |
| 3852 | |
| 3853 | /* Shorthand accesses for engine. Note: If we wanted to use global |
| 3854 | variables (and a single-threaded simulator engine), then we can |
| 3855 | create the actual variables with these names. */ |
| 3856 | |
| 3857 | if (!(state & simSKIPNEXT)) { |
| 3858 | /* Include the simulator engine */ |
| 3859 | #include "engine.c" |
| 3860 | #if ((GPRLEN == 64) && !PROCESSOR_64BIT) || ((GPRLEN == 32) && PROCESSOR_64BIT) |
| 3861 | #error "Mismatch between run-time simulator code and simulation engine" |
| 3862 | #endif |
| 3863 | |
| 3864 | #if defined(WARN_LOHI) |
| 3865 | /* Decrement the HI/LO validity ticks */ |
| 3866 | if (HIACCESS > 0) |
| 3867 | HIACCESS--; |
| 3868 | if (LOACCESS > 0) |
| 3869 | LOACCESS--; |
| 3870 | #endif /* WARN_LOHI */ |
| 3871 | |
| 3872 | #if defined(WARN_ZERO) |
| 3873 | /* For certain MIPS architectures, GPR[0] is hardwired to zero. We |
| 3874 | should check for it being changed. It is better doing it here, |
| 3875 | than within the simulator, since it will help keep the simulator |
| 3876 | small. */ |
| 3877 | if (ZERO != 0) { |
| 3878 | sim_warning("The ZERO register has been updated with 0x%08X%08X (PC = 0x%08X%08X) (reset back to zero)",WORD64HI(ZERO),WORD64LO(ZERO),WORD64HI(IPC),WORD64LO(IPC)); |
| 3879 | ZERO = 0; /* reset back to zero before next instruction */ |
| 3880 | } |
| 3881 | #endif /* WARN_ZERO */ |
| 3882 | } else /* simSKIPNEXT check */ |
| 3883 | state &= ~simSKIPNEXT; |
| 3884 | |
| 3885 | /* If the delay slot was active before the instruction is |
| 3886 | executed, then update the PC to its new value: */ |
| 3887 | if (dsstate) { |
| 3888 | #ifdef DEBUG |
| 3889 | printf("DBG: dsstate set before instruction execution - updating PC to 0x%08X%08X\n",WORD64HI(DSPC),WORD64LO(DSPC)); |
| 3890 | #endif /* DEBUG */ |
| 3891 | PC = DSPC; |
| 3892 | state &= ~simDELAYSLOT; |
| 3893 | } |
| 3894 | |
| 3895 | if (MIPSISA < 4) { /* The following is only required on pre MIPS IV processors: */ |
| 3896 | /* Deal with pending register updates: */ |
| 3897 | #ifdef DEBUG |
| 3898 | printf("DBG: EMPTY BEFORE pending_in = %d, pending_out = %d, pending_total = %d\n",pending_in,pending_out,pending_total); |
| 3899 | #endif /* DEBUG */ |
| 3900 | if (pending_out != pending_in) { |
| 3901 | int loop; |
| 3902 | int index = pending_out; |
| 3903 | int total = pending_total; |
| 3904 | if (pending_total == 0) { |
| 3905 | fprintf(stderr,"FATAL: Mis-match on pending update pointers\n"); |
| 3906 | exit(1); |
| 3907 | } |
| 3908 | for (loop = 0; (loop < total); loop++) { |
| 3909 | #ifdef DEBUG |
| 3910 | printf("DBG: BEFORE index = %d, loop = %d\n",index,loop); |
| 3911 | #endif /* DEBUG */ |
| 3912 | if (pending_slot_reg[index] != (LAST_EMBED_REGNUM + 1)) { |
| 3913 | #ifdef DEBUG |
| 3914 | printf("pending_slot_count[%d] = %d\n",index,pending_slot_count[index]); |
| 3915 | #endif /* DEBUG */ |
| 3916 | if (--(pending_slot_count[index]) == 0) { |
| 3917 | #ifdef DEBUG |
| 3918 | printf("pending_slot_reg[%d] = %d\n",index,pending_slot_reg[index]); |
| 3919 | printf("pending_slot_value[%d] = 0x%08X%08X\n",index,WORD64HI(pending_slot_value[index]),WORD64LO(pending_slot_value[index])); |
| 3920 | #endif /* DEBUG */ |
| 3921 | if (pending_slot_reg[index] == COCIDX) { |
| 3922 | SETFCC(0,((FCR31 & (1 << 23)) ? 1 : 0)); |
| 3923 | } else { |
| 3924 | registers[pending_slot_reg[index]] = pending_slot_value[index]; |
| 3925 | #if defined(HASFPU) |
| 3926 | /* The only time we have PENDING updates to FPU |
| 3927 | registers, is when performing binary transfers. This |
| 3928 | means we should update the register type field. */ |
| 3929 | if ((pending_slot_reg[index] >= FGRIDX) && (pending_slot_reg[index] < (FGRIDX + 32))) |
| 3930 | fpr_state[pending_slot_reg[index]] = fmt_uninterpreted; |
| 3931 | #endif /* HASFPU */ |
| 3932 | } |
| 3933 | #ifdef DEBUG |
| 3934 | printf("registers[%d] = 0x%08X%08X\n",pending_slot_reg[index],WORD64HI(registers[pending_slot_reg[index]]),WORD64LO(registers[pending_slot_reg[index]])); |
| 3935 | #endif /* DEBUG */ |
| 3936 | pending_slot_reg[index] = (LAST_EMBED_REGNUM + 1); |
| 3937 | pending_out++; |
| 3938 | if (pending_out == PSLOTS) |
| 3939 | pending_out = 0; |
| 3940 | pending_total--; |
| 3941 | } |
| 3942 | } |
| 3943 | #ifdef DEBUG |
| 3944 | printf("DBG: AFTER index = %d, loop = %d\n",index,loop); |
| 3945 | #endif /* DEBUG */ |
| 3946 | index++; |
| 3947 | if (index == PSLOTS) |
| 3948 | index = 0; |
| 3949 | } |
| 3950 | } |
| 3951 | #ifdef DEBUG |
| 3952 | printf("DBG: EMPTY AFTER pending_in = %d, pending_out = %d, pending_total = %d\n",pending_in,pending_out,pending_total); |
| 3953 | #endif /* DEBUG */ |
| 3954 | } |
| 3955 | |
| 3956 | #if !defined(FASTSIM) |
| 3957 | pipeline_ticks += pipeline_count; |
| 3958 | #endif /* FASTSIM */ |
| 3959 | |
| 3960 | if (state & simSTEP) |
| 3961 | state |= simSTOP; |
| 3962 | } while (!(state & simSTOP)); |
| 3963 | |
| 3964 | #ifdef DEBUG |
| 3965 | if (membank == NULL) { |
| 3966 | printf("DBG: simulate() LEAVING with no memory\n"); |
| 3967 | exit(1); |
| 3968 | } |
| 3969 | #endif /* DEBUG */ |
| 3970 | |
| 3971 | return; |
| 3972 | } |
| 3973 | |
| 3974 | /*---------------------------------------------------------------------------*/ |
| 3975 | /*> EOF interp.c <*/ |