| 1 | /* GNU/Linux/AArch64 specific low level interface, for the remote server for |
| 2 | GDB. |
| 3 | |
| 4 | Copyright (C) 2009-2020 Free Software Foundation, Inc. |
| 5 | Contributed by ARM Ltd. |
| 6 | |
| 7 | This file is part of GDB. |
| 8 | |
| 9 | This program is free software; you can redistribute it and/or modify |
| 10 | it under the terms of the GNU General Public License as published by |
| 11 | the Free Software Foundation; either version 3 of the License, or |
| 12 | (at your option) any later version. |
| 13 | |
| 14 | This program is distributed in the hope that it will be useful, |
| 15 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 16 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 17 | GNU General Public License for more details. |
| 18 | |
| 19 | You should have received a copy of the GNU General Public License |
| 20 | along with this program. If not, see <http://www.gnu.org/licenses/>. */ |
| 21 | |
| 22 | #include "server.h" |
| 23 | #include "linux-low.h" |
| 24 | #include "nat/aarch64-linux.h" |
| 25 | #include "nat/aarch64-linux-hw-point.h" |
| 26 | #include "arch/aarch64-insn.h" |
| 27 | #include "linux-aarch32-low.h" |
| 28 | #include "elf/common.h" |
| 29 | #include "ax.h" |
| 30 | #include "tracepoint.h" |
| 31 | #include "debug.h" |
| 32 | |
| 33 | #include <signal.h> |
| 34 | #include <sys/user.h> |
| 35 | #include "nat/gdb_ptrace.h" |
| 36 | #include <asm/ptrace.h> |
| 37 | #include <inttypes.h> |
| 38 | #include <endian.h> |
| 39 | #include <sys/uio.h> |
| 40 | |
| 41 | #include "gdb_proc_service.h" |
| 42 | #include "arch/aarch64.h" |
| 43 | #include "linux-aarch32-tdesc.h" |
| 44 | #include "linux-aarch64-tdesc.h" |
| 45 | #include "nat/aarch64-sve-linux-ptrace.h" |
| 46 | #include "tdesc.h" |
| 47 | |
| 48 | #ifdef HAVE_SYS_REG_H |
| 49 | #include <sys/reg.h> |
| 50 | #endif |
| 51 | |
| 52 | /* Linux target op definitions for the AArch64 architecture. */ |
| 53 | |
| 54 | class aarch64_target : public linux_process_target |
| 55 | { |
| 56 | public: |
| 57 | |
| 58 | const regs_info *get_regs_info () override; |
| 59 | |
| 60 | int breakpoint_kind_from_pc (CORE_ADDR *pcptr) override; |
| 61 | |
| 62 | int breakpoint_kind_from_current_state (CORE_ADDR *pcptr) override; |
| 63 | |
| 64 | const gdb_byte *sw_breakpoint_from_kind (int kind, int *size) override; |
| 65 | |
| 66 | bool supports_z_point_type (char z_type) override; |
| 67 | |
| 68 | bool supports_tracepoints () override; |
| 69 | |
| 70 | bool supports_fast_tracepoints () override; |
| 71 | |
| 72 | int install_fast_tracepoint_jump_pad |
| 73 | (CORE_ADDR tpoint, CORE_ADDR tpaddr, CORE_ADDR collector, |
| 74 | CORE_ADDR lockaddr, ULONGEST orig_size, CORE_ADDR *jump_entry, |
| 75 | CORE_ADDR *trampoline, ULONGEST *trampoline_size, |
| 76 | unsigned char *jjump_pad_insn, ULONGEST *jjump_pad_insn_size, |
| 77 | CORE_ADDR *adjusted_insn_addr, CORE_ADDR *adjusted_insn_addr_end, |
| 78 | char *err) override; |
| 79 | |
| 80 | int get_min_fast_tracepoint_insn_len () override; |
| 81 | |
| 82 | struct emit_ops *emit_ops () override; |
| 83 | |
| 84 | protected: |
| 85 | |
| 86 | void low_arch_setup () override; |
| 87 | |
| 88 | bool low_cannot_fetch_register (int regno) override; |
| 89 | |
| 90 | bool low_cannot_store_register (int regno) override; |
| 91 | |
| 92 | bool low_supports_breakpoints () override; |
| 93 | |
| 94 | CORE_ADDR low_get_pc (regcache *regcache) override; |
| 95 | |
| 96 | void low_set_pc (regcache *regcache, CORE_ADDR newpc) override; |
| 97 | |
| 98 | bool low_breakpoint_at (CORE_ADDR pc) override; |
| 99 | |
| 100 | int low_insert_point (raw_bkpt_type type, CORE_ADDR addr, |
| 101 | int size, raw_breakpoint *bp) override; |
| 102 | |
| 103 | int low_remove_point (raw_bkpt_type type, CORE_ADDR addr, |
| 104 | int size, raw_breakpoint *bp) override; |
| 105 | |
| 106 | bool low_stopped_by_watchpoint () override; |
| 107 | |
| 108 | CORE_ADDR low_stopped_data_address () override; |
| 109 | |
| 110 | bool low_siginfo_fixup (siginfo_t *native, gdb_byte *inf, |
| 111 | int direction) override; |
| 112 | |
| 113 | arch_process_info *low_new_process () override; |
| 114 | |
| 115 | void low_delete_process (arch_process_info *info) override; |
| 116 | |
| 117 | void low_new_thread (lwp_info *) override; |
| 118 | |
| 119 | void low_delete_thread (arch_lwp_info *) override; |
| 120 | |
| 121 | void low_new_fork (process_info *parent, process_info *child) override; |
| 122 | |
| 123 | void low_prepare_to_resume (lwp_info *lwp) override; |
| 124 | |
| 125 | int low_get_thread_area (int lwpid, CORE_ADDR *addrp) override; |
| 126 | |
| 127 | bool low_supports_range_stepping () override; |
| 128 | |
| 129 | bool low_supports_catch_syscall () override; |
| 130 | |
| 131 | void low_get_syscall_trapinfo (regcache *regcache, int *sysno) override; |
| 132 | }; |
| 133 | |
| 134 | /* The singleton target ops object. */ |
| 135 | |
| 136 | static aarch64_target the_aarch64_target; |
| 137 | |
| 138 | bool |
| 139 | aarch64_target::low_cannot_fetch_register (int regno) |
| 140 | { |
| 141 | gdb_assert_not_reached ("linux target op low_cannot_fetch_register " |
| 142 | "is not implemented by the target"); |
| 143 | } |
| 144 | |
| 145 | bool |
| 146 | aarch64_target::low_cannot_store_register (int regno) |
| 147 | { |
| 148 | gdb_assert_not_reached ("linux target op low_cannot_store_register " |
| 149 | "is not implemented by the target"); |
| 150 | } |
| 151 | |
| 152 | void |
| 153 | aarch64_target::low_prepare_to_resume (lwp_info *lwp) |
| 154 | { |
| 155 | aarch64_linux_prepare_to_resume (lwp); |
| 156 | } |
| 157 | |
| 158 | /* Per-process arch-specific data we want to keep. */ |
| 159 | |
| 160 | struct arch_process_info |
| 161 | { |
| 162 | /* Hardware breakpoint/watchpoint data. |
| 163 | The reason for them to be per-process rather than per-thread is |
| 164 | due to the lack of information in the gdbserver environment; |
| 165 | gdbserver is not told that whether a requested hardware |
| 166 | breakpoint/watchpoint is thread specific or not, so it has to set |
| 167 | each hw bp/wp for every thread in the current process. The |
| 168 | higher level bp/wp management in gdb will resume a thread if a hw |
| 169 | bp/wp trap is not expected for it. Since the hw bp/wp setting is |
| 170 | same for each thread, it is reasonable for the data to live here. |
| 171 | */ |
| 172 | struct aarch64_debug_reg_state debug_reg_state; |
| 173 | }; |
| 174 | |
| 175 | /* Return true if the size of register 0 is 8 byte. */ |
| 176 | |
| 177 | static int |
| 178 | is_64bit_tdesc (void) |
| 179 | { |
| 180 | struct regcache *regcache = get_thread_regcache (current_thread, 0); |
| 181 | |
| 182 | return register_size (regcache->tdesc, 0) == 8; |
| 183 | } |
| 184 | |
| 185 | /* Return true if the regcache contains the number of SVE registers. */ |
| 186 | |
| 187 | static bool |
| 188 | is_sve_tdesc (void) |
| 189 | { |
| 190 | struct regcache *regcache = get_thread_regcache (current_thread, 0); |
| 191 | |
| 192 | return tdesc_contains_feature (regcache->tdesc, "org.gnu.gdb.aarch64.sve"); |
| 193 | } |
| 194 | |
| 195 | static void |
| 196 | aarch64_fill_gregset (struct regcache *regcache, void *buf) |
| 197 | { |
| 198 | struct user_pt_regs *regset = (struct user_pt_regs *) buf; |
| 199 | int i; |
| 200 | |
| 201 | for (i = 0; i < AARCH64_X_REGS_NUM; i++) |
| 202 | collect_register (regcache, AARCH64_X0_REGNUM + i, ®set->regs[i]); |
| 203 | collect_register (regcache, AARCH64_SP_REGNUM, ®set->sp); |
| 204 | collect_register (regcache, AARCH64_PC_REGNUM, ®set->pc); |
| 205 | collect_register (regcache, AARCH64_CPSR_REGNUM, ®set->pstate); |
| 206 | } |
| 207 | |
| 208 | static void |
| 209 | aarch64_store_gregset (struct regcache *regcache, const void *buf) |
| 210 | { |
| 211 | const struct user_pt_regs *regset = (const struct user_pt_regs *) buf; |
| 212 | int i; |
| 213 | |
| 214 | for (i = 0; i < AARCH64_X_REGS_NUM; i++) |
| 215 | supply_register (regcache, AARCH64_X0_REGNUM + i, ®set->regs[i]); |
| 216 | supply_register (regcache, AARCH64_SP_REGNUM, ®set->sp); |
| 217 | supply_register (regcache, AARCH64_PC_REGNUM, ®set->pc); |
| 218 | supply_register (regcache, AARCH64_CPSR_REGNUM, ®set->pstate); |
| 219 | } |
| 220 | |
| 221 | static void |
| 222 | aarch64_fill_fpregset (struct regcache *regcache, void *buf) |
| 223 | { |
| 224 | struct user_fpsimd_state *regset = (struct user_fpsimd_state *) buf; |
| 225 | int i; |
| 226 | |
| 227 | for (i = 0; i < AARCH64_V_REGS_NUM; i++) |
| 228 | collect_register (regcache, AARCH64_V0_REGNUM + i, ®set->vregs[i]); |
| 229 | collect_register (regcache, AARCH64_FPSR_REGNUM, ®set->fpsr); |
| 230 | collect_register (regcache, AARCH64_FPCR_REGNUM, ®set->fpcr); |
| 231 | } |
| 232 | |
| 233 | static void |
| 234 | aarch64_store_fpregset (struct regcache *regcache, const void *buf) |
| 235 | { |
| 236 | const struct user_fpsimd_state *regset |
| 237 | = (const struct user_fpsimd_state *) buf; |
| 238 | int i; |
| 239 | |
| 240 | for (i = 0; i < AARCH64_V_REGS_NUM; i++) |
| 241 | supply_register (regcache, AARCH64_V0_REGNUM + i, ®set->vregs[i]); |
| 242 | supply_register (regcache, AARCH64_FPSR_REGNUM, ®set->fpsr); |
| 243 | supply_register (regcache, AARCH64_FPCR_REGNUM, ®set->fpcr); |
| 244 | } |
| 245 | |
| 246 | /* Store the pauth registers to regcache. */ |
| 247 | |
| 248 | static void |
| 249 | aarch64_store_pauthregset (struct regcache *regcache, const void *buf) |
| 250 | { |
| 251 | uint64_t *pauth_regset = (uint64_t *) buf; |
| 252 | int pauth_base = find_regno (regcache->tdesc, "pauth_dmask"); |
| 253 | |
| 254 | if (pauth_base == 0) |
| 255 | return; |
| 256 | |
| 257 | supply_register (regcache, AARCH64_PAUTH_DMASK_REGNUM (pauth_base), |
| 258 | &pauth_regset[0]); |
| 259 | supply_register (regcache, AARCH64_PAUTH_CMASK_REGNUM (pauth_base), |
| 260 | &pauth_regset[1]); |
| 261 | } |
| 262 | |
| 263 | bool |
| 264 | aarch64_target::low_supports_breakpoints () |
| 265 | { |
| 266 | return true; |
| 267 | } |
| 268 | |
| 269 | /* Implementation of linux target ops method "low_get_pc". */ |
| 270 | |
| 271 | CORE_ADDR |
| 272 | aarch64_target::low_get_pc (regcache *regcache) |
| 273 | { |
| 274 | if (register_size (regcache->tdesc, 0) == 8) |
| 275 | return linux_get_pc_64bit (regcache); |
| 276 | else |
| 277 | return linux_get_pc_32bit (regcache); |
| 278 | } |
| 279 | |
| 280 | /* Implementation of linux target ops method "low_set_pc". */ |
| 281 | |
| 282 | void |
| 283 | aarch64_target::low_set_pc (regcache *regcache, CORE_ADDR pc) |
| 284 | { |
| 285 | if (register_size (regcache->tdesc, 0) == 8) |
| 286 | linux_set_pc_64bit (regcache, pc); |
| 287 | else |
| 288 | linux_set_pc_32bit (regcache, pc); |
| 289 | } |
| 290 | |
| 291 | #define aarch64_breakpoint_len 4 |
| 292 | |
| 293 | /* AArch64 BRK software debug mode instruction. |
| 294 | This instruction needs to match gdb/aarch64-tdep.c |
| 295 | (aarch64_default_breakpoint). */ |
| 296 | static const gdb_byte aarch64_breakpoint[] = {0x00, 0x00, 0x20, 0xd4}; |
| 297 | |
| 298 | /* Implementation of linux target ops method "low_breakpoint_at". */ |
| 299 | |
| 300 | bool |
| 301 | aarch64_target::low_breakpoint_at (CORE_ADDR where) |
| 302 | { |
| 303 | if (is_64bit_tdesc ()) |
| 304 | { |
| 305 | gdb_byte insn[aarch64_breakpoint_len]; |
| 306 | |
| 307 | read_memory (where, (unsigned char *) &insn, aarch64_breakpoint_len); |
| 308 | if (memcmp (insn, aarch64_breakpoint, aarch64_breakpoint_len) == 0) |
| 309 | return true; |
| 310 | |
| 311 | return false; |
| 312 | } |
| 313 | else |
| 314 | return arm_breakpoint_at (where); |
| 315 | } |
| 316 | |
| 317 | static void |
| 318 | aarch64_init_debug_reg_state (struct aarch64_debug_reg_state *state) |
| 319 | { |
| 320 | int i; |
| 321 | |
| 322 | for (i = 0; i < AARCH64_HBP_MAX_NUM; ++i) |
| 323 | { |
| 324 | state->dr_addr_bp[i] = 0; |
| 325 | state->dr_ctrl_bp[i] = 0; |
| 326 | state->dr_ref_count_bp[i] = 0; |
| 327 | } |
| 328 | |
| 329 | for (i = 0; i < AARCH64_HWP_MAX_NUM; ++i) |
| 330 | { |
| 331 | state->dr_addr_wp[i] = 0; |
| 332 | state->dr_ctrl_wp[i] = 0; |
| 333 | state->dr_ref_count_wp[i] = 0; |
| 334 | } |
| 335 | } |
| 336 | |
| 337 | /* Return the pointer to the debug register state structure in the |
| 338 | current process' arch-specific data area. */ |
| 339 | |
| 340 | struct aarch64_debug_reg_state * |
| 341 | aarch64_get_debug_reg_state (pid_t pid) |
| 342 | { |
| 343 | struct process_info *proc = find_process_pid (pid); |
| 344 | |
| 345 | return &proc->priv->arch_private->debug_reg_state; |
| 346 | } |
| 347 | |
| 348 | /* Implementation of target ops method "supports_z_point_type". */ |
| 349 | |
| 350 | bool |
| 351 | aarch64_target::supports_z_point_type (char z_type) |
| 352 | { |
| 353 | switch (z_type) |
| 354 | { |
| 355 | case Z_PACKET_SW_BP: |
| 356 | case Z_PACKET_HW_BP: |
| 357 | case Z_PACKET_WRITE_WP: |
| 358 | case Z_PACKET_READ_WP: |
| 359 | case Z_PACKET_ACCESS_WP: |
| 360 | return true; |
| 361 | default: |
| 362 | return false; |
| 363 | } |
| 364 | } |
| 365 | |
| 366 | /* Implementation of linux target ops method "low_insert_point". |
| 367 | |
| 368 | It actually only records the info of the to-be-inserted bp/wp; |
| 369 | the actual insertion will happen when threads are resumed. */ |
| 370 | |
| 371 | int |
| 372 | aarch64_target::low_insert_point (raw_bkpt_type type, CORE_ADDR addr, |
| 373 | int len, raw_breakpoint *bp) |
| 374 | { |
| 375 | int ret; |
| 376 | enum target_hw_bp_type targ_type; |
| 377 | struct aarch64_debug_reg_state *state |
| 378 | = aarch64_get_debug_reg_state (pid_of (current_thread)); |
| 379 | |
| 380 | if (show_debug_regs) |
| 381 | fprintf (stderr, "insert_point on entry (addr=0x%08lx, len=%d)\n", |
| 382 | (unsigned long) addr, len); |
| 383 | |
| 384 | /* Determine the type from the raw breakpoint type. */ |
| 385 | targ_type = raw_bkpt_type_to_target_hw_bp_type (type); |
| 386 | |
| 387 | if (targ_type != hw_execute) |
| 388 | { |
| 389 | if (aarch64_linux_region_ok_for_watchpoint (addr, len)) |
| 390 | ret = aarch64_handle_watchpoint (targ_type, addr, len, |
| 391 | 1 /* is_insert */, state); |
| 392 | else |
| 393 | ret = -1; |
| 394 | } |
| 395 | else |
| 396 | { |
| 397 | if (len == 3) |
| 398 | { |
| 399 | /* LEN is 3 means the breakpoint is set on a 32-bit thumb |
| 400 | instruction. Set it to 2 to correctly encode length bit |
| 401 | mask in hardware/watchpoint control register. */ |
| 402 | len = 2; |
| 403 | } |
| 404 | ret = aarch64_handle_breakpoint (targ_type, addr, len, |
| 405 | 1 /* is_insert */, state); |
| 406 | } |
| 407 | |
| 408 | if (show_debug_regs) |
| 409 | aarch64_show_debug_reg_state (state, "insert_point", addr, len, |
| 410 | targ_type); |
| 411 | |
| 412 | return ret; |
| 413 | } |
| 414 | |
| 415 | /* Implementation of linux target ops method "low_remove_point". |
| 416 | |
| 417 | It actually only records the info of the to-be-removed bp/wp, |
| 418 | the actual removal will be done when threads are resumed. */ |
| 419 | |
| 420 | int |
| 421 | aarch64_target::low_remove_point (raw_bkpt_type type, CORE_ADDR addr, |
| 422 | int len, raw_breakpoint *bp) |
| 423 | { |
| 424 | int ret; |
| 425 | enum target_hw_bp_type targ_type; |
| 426 | struct aarch64_debug_reg_state *state |
| 427 | = aarch64_get_debug_reg_state (pid_of (current_thread)); |
| 428 | |
| 429 | if (show_debug_regs) |
| 430 | fprintf (stderr, "remove_point on entry (addr=0x%08lx, len=%d)\n", |
| 431 | (unsigned long) addr, len); |
| 432 | |
| 433 | /* Determine the type from the raw breakpoint type. */ |
| 434 | targ_type = raw_bkpt_type_to_target_hw_bp_type (type); |
| 435 | |
| 436 | /* Set up state pointers. */ |
| 437 | if (targ_type != hw_execute) |
| 438 | ret = |
| 439 | aarch64_handle_watchpoint (targ_type, addr, len, 0 /* is_insert */, |
| 440 | state); |
| 441 | else |
| 442 | { |
| 443 | if (len == 3) |
| 444 | { |
| 445 | /* LEN is 3 means the breakpoint is set on a 32-bit thumb |
| 446 | instruction. Set it to 2 to correctly encode length bit |
| 447 | mask in hardware/watchpoint control register. */ |
| 448 | len = 2; |
| 449 | } |
| 450 | ret = aarch64_handle_breakpoint (targ_type, addr, len, |
| 451 | 0 /* is_insert */, state); |
| 452 | } |
| 453 | |
| 454 | if (show_debug_regs) |
| 455 | aarch64_show_debug_reg_state (state, "remove_point", addr, len, |
| 456 | targ_type); |
| 457 | |
| 458 | return ret; |
| 459 | } |
| 460 | |
| 461 | /* Implementation of linux target ops method "low_stopped_data_address". */ |
| 462 | |
| 463 | CORE_ADDR |
| 464 | aarch64_target::low_stopped_data_address () |
| 465 | { |
| 466 | siginfo_t siginfo; |
| 467 | int pid, i; |
| 468 | struct aarch64_debug_reg_state *state; |
| 469 | |
| 470 | pid = lwpid_of (current_thread); |
| 471 | |
| 472 | /* Get the siginfo. */ |
| 473 | if (ptrace (PTRACE_GETSIGINFO, pid, NULL, &siginfo) != 0) |
| 474 | return (CORE_ADDR) 0; |
| 475 | |
| 476 | /* Need to be a hardware breakpoint/watchpoint trap. */ |
| 477 | if (siginfo.si_signo != SIGTRAP |
| 478 | || (siginfo.si_code & 0xffff) != 0x0004 /* TRAP_HWBKPT */) |
| 479 | return (CORE_ADDR) 0; |
| 480 | |
| 481 | /* Check if the address matches any watched address. */ |
| 482 | state = aarch64_get_debug_reg_state (pid_of (current_thread)); |
| 483 | for (i = aarch64_num_wp_regs - 1; i >= 0; --i) |
| 484 | { |
| 485 | const unsigned int offset |
| 486 | = aarch64_watchpoint_offset (state->dr_ctrl_wp[i]); |
| 487 | const unsigned int len = aarch64_watchpoint_length (state->dr_ctrl_wp[i]); |
| 488 | const CORE_ADDR addr_trap = (CORE_ADDR) siginfo.si_addr; |
| 489 | const CORE_ADDR addr_watch = state->dr_addr_wp[i] + offset; |
| 490 | const CORE_ADDR addr_watch_aligned = align_down (state->dr_addr_wp[i], 8); |
| 491 | const CORE_ADDR addr_orig = state->dr_addr_orig_wp[i]; |
| 492 | |
| 493 | if (state->dr_ref_count_wp[i] |
| 494 | && DR_CONTROL_ENABLED (state->dr_ctrl_wp[i]) |
| 495 | && addr_trap >= addr_watch_aligned |
| 496 | && addr_trap < addr_watch + len) |
| 497 | { |
| 498 | /* ADDR_TRAP reports the first address of the memory range |
| 499 | accessed by the CPU, regardless of what was the memory |
| 500 | range watched. Thus, a large CPU access that straddles |
| 501 | the ADDR_WATCH..ADDR_WATCH+LEN range may result in an |
| 502 | ADDR_TRAP that is lower than the |
| 503 | ADDR_WATCH..ADDR_WATCH+LEN range. E.g.: |
| 504 | |
| 505 | addr: | 4 | 5 | 6 | 7 | 8 | |
| 506 | |---- range watched ----| |
| 507 | |----------- range accessed ------------| |
| 508 | |
| 509 | In this case, ADDR_TRAP will be 4. |
| 510 | |
| 511 | To match a watchpoint known to GDB core, we must never |
| 512 | report *ADDR_P outside of any ADDR_WATCH..ADDR_WATCH+LEN |
| 513 | range. ADDR_WATCH <= ADDR_TRAP < ADDR_ORIG is a false |
| 514 | positive on kernels older than 4.10. See PR |
| 515 | external/20207. */ |
| 516 | return addr_orig; |
| 517 | } |
| 518 | } |
| 519 | |
| 520 | return (CORE_ADDR) 0; |
| 521 | } |
| 522 | |
| 523 | /* Implementation of linux target ops method "low_stopped_by_watchpoint". */ |
| 524 | |
| 525 | bool |
| 526 | aarch64_target::low_stopped_by_watchpoint () |
| 527 | { |
| 528 | return (low_stopped_data_address () != 0); |
| 529 | } |
| 530 | |
| 531 | /* Fetch the thread-local storage pointer for libthread_db. */ |
| 532 | |
| 533 | ps_err_e |
| 534 | ps_get_thread_area (struct ps_prochandle *ph, |
| 535 | lwpid_t lwpid, int idx, void **base) |
| 536 | { |
| 537 | return aarch64_ps_get_thread_area (ph, lwpid, idx, base, |
| 538 | is_64bit_tdesc ()); |
| 539 | } |
| 540 | |
| 541 | /* Implementation of linux target ops method "low_siginfo_fixup". */ |
| 542 | |
| 543 | bool |
| 544 | aarch64_target::low_siginfo_fixup (siginfo_t *native, gdb_byte *inf, |
| 545 | int direction) |
| 546 | { |
| 547 | /* Is the inferior 32-bit? If so, then fixup the siginfo object. */ |
| 548 | if (!is_64bit_tdesc ()) |
| 549 | { |
| 550 | if (direction == 0) |
| 551 | aarch64_compat_siginfo_from_siginfo ((struct compat_siginfo *) inf, |
| 552 | native); |
| 553 | else |
| 554 | aarch64_siginfo_from_compat_siginfo (native, |
| 555 | (struct compat_siginfo *) inf); |
| 556 | |
| 557 | return true; |
| 558 | } |
| 559 | |
| 560 | return false; |
| 561 | } |
| 562 | |
| 563 | /* Implementation of linux target ops method "low_new_process". */ |
| 564 | |
| 565 | arch_process_info * |
| 566 | aarch64_target::low_new_process () |
| 567 | { |
| 568 | struct arch_process_info *info = XCNEW (struct arch_process_info); |
| 569 | |
| 570 | aarch64_init_debug_reg_state (&info->debug_reg_state); |
| 571 | |
| 572 | return info; |
| 573 | } |
| 574 | |
| 575 | /* Implementation of linux target ops method "low_delete_process". */ |
| 576 | |
| 577 | void |
| 578 | aarch64_target::low_delete_process (arch_process_info *info) |
| 579 | { |
| 580 | xfree (info); |
| 581 | } |
| 582 | |
| 583 | void |
| 584 | aarch64_target::low_new_thread (lwp_info *lwp) |
| 585 | { |
| 586 | aarch64_linux_new_thread (lwp); |
| 587 | } |
| 588 | |
| 589 | void |
| 590 | aarch64_target::low_delete_thread (arch_lwp_info *arch_lwp) |
| 591 | { |
| 592 | aarch64_linux_delete_thread (arch_lwp); |
| 593 | } |
| 594 | |
| 595 | /* Implementation of linux target ops method "low_new_fork". */ |
| 596 | |
| 597 | void |
| 598 | aarch64_target::low_new_fork (process_info *parent, |
| 599 | process_info *child) |
| 600 | { |
| 601 | /* These are allocated by linux_add_process. */ |
| 602 | gdb_assert (parent->priv != NULL |
| 603 | && parent->priv->arch_private != NULL); |
| 604 | gdb_assert (child->priv != NULL |
| 605 | && child->priv->arch_private != NULL); |
| 606 | |
| 607 | /* Linux kernel before 2.6.33 commit |
| 608 | 72f674d203cd230426437cdcf7dd6f681dad8b0d |
| 609 | will inherit hardware debug registers from parent |
| 610 | on fork/vfork/clone. Newer Linux kernels create such tasks with |
| 611 | zeroed debug registers. |
| 612 | |
| 613 | GDB core assumes the child inherits the watchpoints/hw |
| 614 | breakpoints of the parent, and will remove them all from the |
| 615 | forked off process. Copy the debug registers mirrors into the |
| 616 | new process so that all breakpoints and watchpoints can be |
| 617 | removed together. The debug registers mirror will become zeroed |
| 618 | in the end before detaching the forked off process, thus making |
| 619 | this compatible with older Linux kernels too. */ |
| 620 | |
| 621 | *child->priv->arch_private = *parent->priv->arch_private; |
| 622 | } |
| 623 | |
| 624 | /* Matches HWCAP_PACA in kernel header arch/arm64/include/uapi/asm/hwcap.h. */ |
| 625 | #define AARCH64_HWCAP_PACA (1 << 30) |
| 626 | |
| 627 | /* Implementation of linux target ops method "low_arch_setup". */ |
| 628 | |
| 629 | void |
| 630 | aarch64_target::low_arch_setup () |
| 631 | { |
| 632 | unsigned int machine; |
| 633 | int is_elf64; |
| 634 | int tid; |
| 635 | |
| 636 | tid = lwpid_of (current_thread); |
| 637 | |
| 638 | is_elf64 = linux_pid_exe_is_elf_64_file (tid, &machine); |
| 639 | |
| 640 | if (is_elf64) |
| 641 | { |
| 642 | uint64_t vq = aarch64_sve_get_vq (tid); |
| 643 | unsigned long hwcap = linux_get_hwcap (8); |
| 644 | bool pauth_p = hwcap & AARCH64_HWCAP_PACA; |
| 645 | |
| 646 | current_process ()->tdesc = aarch64_linux_read_description (vq, pauth_p); |
| 647 | } |
| 648 | else |
| 649 | current_process ()->tdesc = aarch32_linux_read_description (); |
| 650 | |
| 651 | aarch64_linux_get_debug_reg_capacity (lwpid_of (current_thread)); |
| 652 | } |
| 653 | |
| 654 | /* Wrapper for aarch64_sve_regs_copy_to_reg_buf. */ |
| 655 | |
| 656 | static void |
| 657 | aarch64_sve_regs_copy_to_regcache (struct regcache *regcache, const void *buf) |
| 658 | { |
| 659 | return aarch64_sve_regs_copy_to_reg_buf (regcache, buf); |
| 660 | } |
| 661 | |
| 662 | /* Wrapper for aarch64_sve_regs_copy_from_reg_buf. */ |
| 663 | |
| 664 | static void |
| 665 | aarch64_sve_regs_copy_from_regcache (struct regcache *regcache, void *buf) |
| 666 | { |
| 667 | return aarch64_sve_regs_copy_from_reg_buf (regcache, buf); |
| 668 | } |
| 669 | |
| 670 | static struct regset_info aarch64_regsets[] = |
| 671 | { |
| 672 | { PTRACE_GETREGSET, PTRACE_SETREGSET, NT_PRSTATUS, |
| 673 | sizeof (struct user_pt_regs), GENERAL_REGS, |
| 674 | aarch64_fill_gregset, aarch64_store_gregset }, |
| 675 | { PTRACE_GETREGSET, PTRACE_SETREGSET, NT_FPREGSET, |
| 676 | sizeof (struct user_fpsimd_state), FP_REGS, |
| 677 | aarch64_fill_fpregset, aarch64_store_fpregset |
| 678 | }, |
| 679 | { PTRACE_GETREGSET, PTRACE_SETREGSET, NT_ARM_PAC_MASK, |
| 680 | AARCH64_PAUTH_REGS_SIZE, OPTIONAL_REGS, |
| 681 | NULL, aarch64_store_pauthregset }, |
| 682 | NULL_REGSET |
| 683 | }; |
| 684 | |
| 685 | static struct regsets_info aarch64_regsets_info = |
| 686 | { |
| 687 | aarch64_regsets, /* regsets */ |
| 688 | 0, /* num_regsets */ |
| 689 | NULL, /* disabled_regsets */ |
| 690 | }; |
| 691 | |
| 692 | static struct regs_info regs_info_aarch64 = |
| 693 | { |
| 694 | NULL, /* regset_bitmap */ |
| 695 | NULL, /* usrregs */ |
| 696 | &aarch64_regsets_info, |
| 697 | }; |
| 698 | |
| 699 | static struct regset_info aarch64_sve_regsets[] = |
| 700 | { |
| 701 | { PTRACE_GETREGSET, PTRACE_SETREGSET, NT_PRSTATUS, |
| 702 | sizeof (struct user_pt_regs), GENERAL_REGS, |
| 703 | aarch64_fill_gregset, aarch64_store_gregset }, |
| 704 | { PTRACE_GETREGSET, PTRACE_SETREGSET, NT_ARM_SVE, |
| 705 | SVE_PT_SIZE (AARCH64_MAX_SVE_VQ, SVE_PT_REGS_SVE), EXTENDED_REGS, |
| 706 | aarch64_sve_regs_copy_from_regcache, aarch64_sve_regs_copy_to_regcache |
| 707 | }, |
| 708 | { PTRACE_GETREGSET, PTRACE_SETREGSET, NT_ARM_PAC_MASK, |
| 709 | AARCH64_PAUTH_REGS_SIZE, OPTIONAL_REGS, |
| 710 | NULL, aarch64_store_pauthregset }, |
| 711 | NULL_REGSET |
| 712 | }; |
| 713 | |
| 714 | static struct regsets_info aarch64_sve_regsets_info = |
| 715 | { |
| 716 | aarch64_sve_regsets, /* regsets. */ |
| 717 | 0, /* num_regsets. */ |
| 718 | NULL, /* disabled_regsets. */ |
| 719 | }; |
| 720 | |
| 721 | static struct regs_info regs_info_aarch64_sve = |
| 722 | { |
| 723 | NULL, /* regset_bitmap. */ |
| 724 | NULL, /* usrregs. */ |
| 725 | &aarch64_sve_regsets_info, |
| 726 | }; |
| 727 | |
| 728 | /* Implementation of linux target ops method "get_regs_info". */ |
| 729 | |
| 730 | const regs_info * |
| 731 | aarch64_target::get_regs_info () |
| 732 | { |
| 733 | if (!is_64bit_tdesc ()) |
| 734 | return ®s_info_aarch32; |
| 735 | |
| 736 | if (is_sve_tdesc ()) |
| 737 | return ®s_info_aarch64_sve; |
| 738 | |
| 739 | return ®s_info_aarch64; |
| 740 | } |
| 741 | |
| 742 | /* Implementation of target ops method "supports_tracepoints". */ |
| 743 | |
| 744 | bool |
| 745 | aarch64_target::supports_tracepoints () |
| 746 | { |
| 747 | if (current_thread == NULL) |
| 748 | return true; |
| 749 | else |
| 750 | { |
| 751 | /* We don't support tracepoints on aarch32 now. */ |
| 752 | return is_64bit_tdesc (); |
| 753 | } |
| 754 | } |
| 755 | |
| 756 | /* Implementation of linux target ops method "low_get_thread_area". */ |
| 757 | |
| 758 | int |
| 759 | aarch64_target::low_get_thread_area (int lwpid, CORE_ADDR *addrp) |
| 760 | { |
| 761 | struct iovec iovec; |
| 762 | uint64_t reg; |
| 763 | |
| 764 | iovec.iov_base = ® |
| 765 | iovec.iov_len = sizeof (reg); |
| 766 | |
| 767 | if (ptrace (PTRACE_GETREGSET, lwpid, NT_ARM_TLS, &iovec) != 0) |
| 768 | return -1; |
| 769 | |
| 770 | *addrp = reg; |
| 771 | |
| 772 | return 0; |
| 773 | } |
| 774 | |
| 775 | bool |
| 776 | aarch64_target::low_supports_catch_syscall () |
| 777 | { |
| 778 | return true; |
| 779 | } |
| 780 | |
| 781 | /* Implementation of linux target ops method "low_get_syscall_trapinfo". */ |
| 782 | |
| 783 | void |
| 784 | aarch64_target::low_get_syscall_trapinfo (regcache *regcache, int *sysno) |
| 785 | { |
| 786 | int use_64bit = register_size (regcache->tdesc, 0) == 8; |
| 787 | |
| 788 | if (use_64bit) |
| 789 | { |
| 790 | long l_sysno; |
| 791 | |
| 792 | collect_register_by_name (regcache, "x8", &l_sysno); |
| 793 | *sysno = (int) l_sysno; |
| 794 | } |
| 795 | else |
| 796 | collect_register_by_name (regcache, "r7", sysno); |
| 797 | } |
| 798 | |
| 799 | /* List of condition codes that we need. */ |
| 800 | |
| 801 | enum aarch64_condition_codes |
| 802 | { |
| 803 | EQ = 0x0, |
| 804 | NE = 0x1, |
| 805 | LO = 0x3, |
| 806 | GE = 0xa, |
| 807 | LT = 0xb, |
| 808 | GT = 0xc, |
| 809 | LE = 0xd, |
| 810 | }; |
| 811 | |
| 812 | enum aarch64_operand_type |
| 813 | { |
| 814 | OPERAND_IMMEDIATE, |
| 815 | OPERAND_REGISTER, |
| 816 | }; |
| 817 | |
| 818 | /* Representation of an operand. At this time, it only supports register |
| 819 | and immediate types. */ |
| 820 | |
| 821 | struct aarch64_operand |
| 822 | { |
| 823 | /* Type of the operand. */ |
| 824 | enum aarch64_operand_type type; |
| 825 | |
| 826 | /* Value of the operand according to the type. */ |
| 827 | union |
| 828 | { |
| 829 | uint32_t imm; |
| 830 | struct aarch64_register reg; |
| 831 | }; |
| 832 | }; |
| 833 | |
| 834 | /* List of registers that we are currently using, we can add more here as |
| 835 | we need to use them. */ |
| 836 | |
| 837 | /* General purpose scratch registers (64 bit). */ |
| 838 | static const struct aarch64_register x0 = { 0, 1 }; |
| 839 | static const struct aarch64_register x1 = { 1, 1 }; |
| 840 | static const struct aarch64_register x2 = { 2, 1 }; |
| 841 | static const struct aarch64_register x3 = { 3, 1 }; |
| 842 | static const struct aarch64_register x4 = { 4, 1 }; |
| 843 | |
| 844 | /* General purpose scratch registers (32 bit). */ |
| 845 | static const struct aarch64_register w0 = { 0, 0 }; |
| 846 | static const struct aarch64_register w2 = { 2, 0 }; |
| 847 | |
| 848 | /* Intra-procedure scratch registers. */ |
| 849 | static const struct aarch64_register ip0 = { 16, 1 }; |
| 850 | |
| 851 | /* Special purpose registers. */ |
| 852 | static const struct aarch64_register fp = { 29, 1 }; |
| 853 | static const struct aarch64_register lr = { 30, 1 }; |
| 854 | static const struct aarch64_register sp = { 31, 1 }; |
| 855 | static const struct aarch64_register xzr = { 31, 1 }; |
| 856 | |
| 857 | /* Dynamically allocate a new register. If we know the register |
| 858 | statically, we should make it a global as above instead of using this |
| 859 | helper function. */ |
| 860 | |
| 861 | static struct aarch64_register |
| 862 | aarch64_register (unsigned num, int is64) |
| 863 | { |
| 864 | return (struct aarch64_register) { num, is64 }; |
| 865 | } |
| 866 | |
| 867 | /* Helper function to create a register operand, for instructions with |
| 868 | different types of operands. |
| 869 | |
| 870 | For example: |
| 871 | p += emit_mov (p, x0, register_operand (x1)); */ |
| 872 | |
| 873 | static struct aarch64_operand |
| 874 | register_operand (struct aarch64_register reg) |
| 875 | { |
| 876 | struct aarch64_operand operand; |
| 877 | |
| 878 | operand.type = OPERAND_REGISTER; |
| 879 | operand.reg = reg; |
| 880 | |
| 881 | return operand; |
| 882 | } |
| 883 | |
| 884 | /* Helper function to create an immediate operand, for instructions with |
| 885 | different types of operands. |
| 886 | |
| 887 | For example: |
| 888 | p += emit_mov (p, x0, immediate_operand (12)); */ |
| 889 | |
| 890 | static struct aarch64_operand |
| 891 | immediate_operand (uint32_t imm) |
| 892 | { |
| 893 | struct aarch64_operand operand; |
| 894 | |
| 895 | operand.type = OPERAND_IMMEDIATE; |
| 896 | operand.imm = imm; |
| 897 | |
| 898 | return operand; |
| 899 | } |
| 900 | |
| 901 | /* Helper function to create an offset memory operand. |
| 902 | |
| 903 | For example: |
| 904 | p += emit_ldr (p, x0, sp, offset_memory_operand (16)); */ |
| 905 | |
| 906 | static struct aarch64_memory_operand |
| 907 | offset_memory_operand (int32_t offset) |
| 908 | { |
| 909 | return (struct aarch64_memory_operand) { MEMORY_OPERAND_OFFSET, offset }; |
| 910 | } |
| 911 | |
| 912 | /* Helper function to create a pre-index memory operand. |
| 913 | |
| 914 | For example: |
| 915 | p += emit_ldr (p, x0, sp, preindex_memory_operand (16)); */ |
| 916 | |
| 917 | static struct aarch64_memory_operand |
| 918 | preindex_memory_operand (int32_t index) |
| 919 | { |
| 920 | return (struct aarch64_memory_operand) { MEMORY_OPERAND_PREINDEX, index }; |
| 921 | } |
| 922 | |
| 923 | /* Helper function to create a post-index memory operand. |
| 924 | |
| 925 | For example: |
| 926 | p += emit_ldr (p, x0, sp, postindex_memory_operand (16)); */ |
| 927 | |
| 928 | static struct aarch64_memory_operand |
| 929 | postindex_memory_operand (int32_t index) |
| 930 | { |
| 931 | return (struct aarch64_memory_operand) { MEMORY_OPERAND_POSTINDEX, index }; |
| 932 | } |
| 933 | |
| 934 | /* System control registers. These special registers can be written and |
| 935 | read with the MRS and MSR instructions. |
| 936 | |
| 937 | - NZCV: Condition flags. GDB refers to this register under the CPSR |
| 938 | name. |
| 939 | - FPSR: Floating-point status register. |
| 940 | - FPCR: Floating-point control registers. |
| 941 | - TPIDR_EL0: Software thread ID register. */ |
| 942 | |
| 943 | enum aarch64_system_control_registers |
| 944 | { |
| 945 | /* op0 op1 crn crm op2 */ |
| 946 | NZCV = (0x1 << 14) | (0x3 << 11) | (0x4 << 7) | (0x2 << 3) | 0x0, |
| 947 | FPSR = (0x1 << 14) | (0x3 << 11) | (0x4 << 7) | (0x4 << 3) | 0x1, |
| 948 | FPCR = (0x1 << 14) | (0x3 << 11) | (0x4 << 7) | (0x4 << 3) | 0x0, |
| 949 | TPIDR_EL0 = (0x1 << 14) | (0x3 << 11) | (0xd << 7) | (0x0 << 3) | 0x2 |
| 950 | }; |
| 951 | |
| 952 | /* Write a BLR instruction into *BUF. |
| 953 | |
| 954 | BLR rn |
| 955 | |
| 956 | RN is the register to branch to. */ |
| 957 | |
| 958 | static int |
| 959 | emit_blr (uint32_t *buf, struct aarch64_register rn) |
| 960 | { |
| 961 | return aarch64_emit_insn (buf, BLR | ENCODE (rn.num, 5, 5)); |
| 962 | } |
| 963 | |
| 964 | /* Write a RET instruction into *BUF. |
| 965 | |
| 966 | RET xn |
| 967 | |
| 968 | RN is the register to branch to. */ |
| 969 | |
| 970 | static int |
| 971 | emit_ret (uint32_t *buf, struct aarch64_register rn) |
| 972 | { |
| 973 | return aarch64_emit_insn (buf, RET | ENCODE (rn.num, 5, 5)); |
| 974 | } |
| 975 | |
| 976 | static int |
| 977 | emit_load_store_pair (uint32_t *buf, enum aarch64_opcodes opcode, |
| 978 | struct aarch64_register rt, |
| 979 | struct aarch64_register rt2, |
| 980 | struct aarch64_register rn, |
| 981 | struct aarch64_memory_operand operand) |
| 982 | { |
| 983 | uint32_t opc; |
| 984 | uint32_t pre_index; |
| 985 | uint32_t write_back; |
| 986 | |
| 987 | if (rt.is64) |
| 988 | opc = ENCODE (2, 2, 30); |
| 989 | else |
| 990 | opc = ENCODE (0, 2, 30); |
| 991 | |
| 992 | switch (operand.type) |
| 993 | { |
| 994 | case MEMORY_OPERAND_OFFSET: |
| 995 | { |
| 996 | pre_index = ENCODE (1, 1, 24); |
| 997 | write_back = ENCODE (0, 1, 23); |
| 998 | break; |
| 999 | } |
| 1000 | case MEMORY_OPERAND_POSTINDEX: |
| 1001 | { |
| 1002 | pre_index = ENCODE (0, 1, 24); |
| 1003 | write_back = ENCODE (1, 1, 23); |
| 1004 | break; |
| 1005 | } |
| 1006 | case MEMORY_OPERAND_PREINDEX: |
| 1007 | { |
| 1008 | pre_index = ENCODE (1, 1, 24); |
| 1009 | write_back = ENCODE (1, 1, 23); |
| 1010 | break; |
| 1011 | } |
| 1012 | default: |
| 1013 | return 0; |
| 1014 | } |
| 1015 | |
| 1016 | return aarch64_emit_insn (buf, opcode | opc | pre_index | write_back |
| 1017 | | ENCODE (operand.index >> 3, 7, 15) |
| 1018 | | ENCODE (rt2.num, 5, 10) |
| 1019 | | ENCODE (rn.num, 5, 5) | ENCODE (rt.num, 5, 0)); |
| 1020 | } |
| 1021 | |
| 1022 | /* Write a STP instruction into *BUF. |
| 1023 | |
| 1024 | STP rt, rt2, [rn, #offset] |
| 1025 | STP rt, rt2, [rn, #index]! |
| 1026 | STP rt, rt2, [rn], #index |
| 1027 | |
| 1028 | RT and RT2 are the registers to store. |
| 1029 | RN is the base address register. |
| 1030 | OFFSET is the immediate to add to the base address. It is limited to a |
| 1031 | -512 .. 504 range (7 bits << 3). */ |
| 1032 | |
| 1033 | static int |
| 1034 | emit_stp (uint32_t *buf, struct aarch64_register rt, |
| 1035 | struct aarch64_register rt2, struct aarch64_register rn, |
| 1036 | struct aarch64_memory_operand operand) |
| 1037 | { |
| 1038 | return emit_load_store_pair (buf, STP, rt, rt2, rn, operand); |
| 1039 | } |
| 1040 | |
| 1041 | /* Write a LDP instruction into *BUF. |
| 1042 | |
| 1043 | LDP rt, rt2, [rn, #offset] |
| 1044 | LDP rt, rt2, [rn, #index]! |
| 1045 | LDP rt, rt2, [rn], #index |
| 1046 | |
| 1047 | RT and RT2 are the registers to store. |
| 1048 | RN is the base address register. |
| 1049 | OFFSET is the immediate to add to the base address. It is limited to a |
| 1050 | -512 .. 504 range (7 bits << 3). */ |
| 1051 | |
| 1052 | static int |
| 1053 | emit_ldp (uint32_t *buf, struct aarch64_register rt, |
| 1054 | struct aarch64_register rt2, struct aarch64_register rn, |
| 1055 | struct aarch64_memory_operand operand) |
| 1056 | { |
| 1057 | return emit_load_store_pair (buf, LDP, rt, rt2, rn, operand); |
| 1058 | } |
| 1059 | |
| 1060 | /* Write a LDP (SIMD&VFP) instruction using Q registers into *BUF. |
| 1061 | |
| 1062 | LDP qt, qt2, [rn, #offset] |
| 1063 | |
| 1064 | RT and RT2 are the Q registers to store. |
| 1065 | RN is the base address register. |
| 1066 | OFFSET is the immediate to add to the base address. It is limited to |
| 1067 | -1024 .. 1008 range (7 bits << 4). */ |
| 1068 | |
| 1069 | static int |
| 1070 | emit_ldp_q_offset (uint32_t *buf, unsigned rt, unsigned rt2, |
| 1071 | struct aarch64_register rn, int32_t offset) |
| 1072 | { |
| 1073 | uint32_t opc = ENCODE (2, 2, 30); |
| 1074 | uint32_t pre_index = ENCODE (1, 1, 24); |
| 1075 | |
| 1076 | return aarch64_emit_insn (buf, LDP_SIMD_VFP | opc | pre_index |
| 1077 | | ENCODE (offset >> 4, 7, 15) |
| 1078 | | ENCODE (rt2, 5, 10) |
| 1079 | | ENCODE (rn.num, 5, 5) | ENCODE (rt, 5, 0)); |
| 1080 | } |
| 1081 | |
| 1082 | /* Write a STP (SIMD&VFP) instruction using Q registers into *BUF. |
| 1083 | |
| 1084 | STP qt, qt2, [rn, #offset] |
| 1085 | |
| 1086 | RT and RT2 are the Q registers to store. |
| 1087 | RN is the base address register. |
| 1088 | OFFSET is the immediate to add to the base address. It is limited to |
| 1089 | -1024 .. 1008 range (7 bits << 4). */ |
| 1090 | |
| 1091 | static int |
| 1092 | emit_stp_q_offset (uint32_t *buf, unsigned rt, unsigned rt2, |
| 1093 | struct aarch64_register rn, int32_t offset) |
| 1094 | { |
| 1095 | uint32_t opc = ENCODE (2, 2, 30); |
| 1096 | uint32_t pre_index = ENCODE (1, 1, 24); |
| 1097 | |
| 1098 | return aarch64_emit_insn (buf, STP_SIMD_VFP | opc | pre_index |
| 1099 | | ENCODE (offset >> 4, 7, 15) |
| 1100 | | ENCODE (rt2, 5, 10) |
| 1101 | | ENCODE (rn.num, 5, 5) | ENCODE (rt, 5, 0)); |
| 1102 | } |
| 1103 | |
| 1104 | /* Write a LDRH instruction into *BUF. |
| 1105 | |
| 1106 | LDRH wt, [xn, #offset] |
| 1107 | LDRH wt, [xn, #index]! |
| 1108 | LDRH wt, [xn], #index |
| 1109 | |
| 1110 | RT is the register to store. |
| 1111 | RN is the base address register. |
| 1112 | OFFSET is the immediate to add to the base address. It is limited to |
| 1113 | 0 .. 32760 range (12 bits << 3). */ |
| 1114 | |
| 1115 | static int |
| 1116 | emit_ldrh (uint32_t *buf, struct aarch64_register rt, |
| 1117 | struct aarch64_register rn, |
| 1118 | struct aarch64_memory_operand operand) |
| 1119 | { |
| 1120 | return aarch64_emit_load_store (buf, 1, LDR, rt, rn, operand); |
| 1121 | } |
| 1122 | |
| 1123 | /* Write a LDRB instruction into *BUF. |
| 1124 | |
| 1125 | LDRB wt, [xn, #offset] |
| 1126 | LDRB wt, [xn, #index]! |
| 1127 | LDRB wt, [xn], #index |
| 1128 | |
| 1129 | RT is the register to store. |
| 1130 | RN is the base address register. |
| 1131 | OFFSET is the immediate to add to the base address. It is limited to |
| 1132 | 0 .. 32760 range (12 bits << 3). */ |
| 1133 | |
| 1134 | static int |
| 1135 | emit_ldrb (uint32_t *buf, struct aarch64_register rt, |
| 1136 | struct aarch64_register rn, |
| 1137 | struct aarch64_memory_operand operand) |
| 1138 | { |
| 1139 | return aarch64_emit_load_store (buf, 0, LDR, rt, rn, operand); |
| 1140 | } |
| 1141 | |
| 1142 | |
| 1143 | |
| 1144 | /* Write a STR instruction into *BUF. |
| 1145 | |
| 1146 | STR rt, [rn, #offset] |
| 1147 | STR rt, [rn, #index]! |
| 1148 | STR rt, [rn], #index |
| 1149 | |
| 1150 | RT is the register to store. |
| 1151 | RN is the base address register. |
| 1152 | OFFSET is the immediate to add to the base address. It is limited to |
| 1153 | 0 .. 32760 range (12 bits << 3). */ |
| 1154 | |
| 1155 | static int |
| 1156 | emit_str (uint32_t *buf, struct aarch64_register rt, |
| 1157 | struct aarch64_register rn, |
| 1158 | struct aarch64_memory_operand operand) |
| 1159 | { |
| 1160 | return aarch64_emit_load_store (buf, rt.is64 ? 3 : 2, STR, rt, rn, operand); |
| 1161 | } |
| 1162 | |
| 1163 | /* Helper function emitting an exclusive load or store instruction. */ |
| 1164 | |
| 1165 | static int |
| 1166 | emit_load_store_exclusive (uint32_t *buf, uint32_t size, |
| 1167 | enum aarch64_opcodes opcode, |
| 1168 | struct aarch64_register rs, |
| 1169 | struct aarch64_register rt, |
| 1170 | struct aarch64_register rt2, |
| 1171 | struct aarch64_register rn) |
| 1172 | { |
| 1173 | return aarch64_emit_insn (buf, opcode | ENCODE (size, 2, 30) |
| 1174 | | ENCODE (rs.num, 5, 16) | ENCODE (rt2.num, 5, 10) |
| 1175 | | ENCODE (rn.num, 5, 5) | ENCODE (rt.num, 5, 0)); |
| 1176 | } |
| 1177 | |
| 1178 | /* Write a LAXR instruction into *BUF. |
| 1179 | |
| 1180 | LDAXR rt, [xn] |
| 1181 | |
| 1182 | RT is the destination register. |
| 1183 | RN is the base address register. */ |
| 1184 | |
| 1185 | static int |
| 1186 | emit_ldaxr (uint32_t *buf, struct aarch64_register rt, |
| 1187 | struct aarch64_register rn) |
| 1188 | { |
| 1189 | return emit_load_store_exclusive (buf, rt.is64 ? 3 : 2, LDAXR, xzr, rt, |
| 1190 | xzr, rn); |
| 1191 | } |
| 1192 | |
| 1193 | /* Write a STXR instruction into *BUF. |
| 1194 | |
| 1195 | STXR ws, rt, [xn] |
| 1196 | |
| 1197 | RS is the result register, it indicates if the store succeeded or not. |
| 1198 | RT is the destination register. |
| 1199 | RN is the base address register. */ |
| 1200 | |
| 1201 | static int |
| 1202 | emit_stxr (uint32_t *buf, struct aarch64_register rs, |
| 1203 | struct aarch64_register rt, struct aarch64_register rn) |
| 1204 | { |
| 1205 | return emit_load_store_exclusive (buf, rt.is64 ? 3 : 2, STXR, rs, rt, |
| 1206 | xzr, rn); |
| 1207 | } |
| 1208 | |
| 1209 | /* Write a STLR instruction into *BUF. |
| 1210 | |
| 1211 | STLR rt, [xn] |
| 1212 | |
| 1213 | RT is the register to store. |
| 1214 | RN is the base address register. */ |
| 1215 | |
| 1216 | static int |
| 1217 | emit_stlr (uint32_t *buf, struct aarch64_register rt, |
| 1218 | struct aarch64_register rn) |
| 1219 | { |
| 1220 | return emit_load_store_exclusive (buf, rt.is64 ? 3 : 2, STLR, xzr, rt, |
| 1221 | xzr, rn); |
| 1222 | } |
| 1223 | |
| 1224 | /* Helper function for data processing instructions with register sources. */ |
| 1225 | |
| 1226 | static int |
| 1227 | emit_data_processing_reg (uint32_t *buf, uint32_t opcode, |
| 1228 | struct aarch64_register rd, |
| 1229 | struct aarch64_register rn, |
| 1230 | struct aarch64_register rm) |
| 1231 | { |
| 1232 | uint32_t size = ENCODE (rd.is64, 1, 31); |
| 1233 | |
| 1234 | return aarch64_emit_insn (buf, opcode | size | ENCODE (rm.num, 5, 16) |
| 1235 | | ENCODE (rn.num, 5, 5) | ENCODE (rd.num, 5, 0)); |
| 1236 | } |
| 1237 | |
| 1238 | /* Helper function for data processing instructions taking either a register |
| 1239 | or an immediate. */ |
| 1240 | |
| 1241 | static int |
| 1242 | emit_data_processing (uint32_t *buf, enum aarch64_opcodes opcode, |
| 1243 | struct aarch64_register rd, |
| 1244 | struct aarch64_register rn, |
| 1245 | struct aarch64_operand operand) |
| 1246 | { |
| 1247 | uint32_t size = ENCODE (rd.is64, 1, 31); |
| 1248 | /* The opcode is different for register and immediate source operands. */ |
| 1249 | uint32_t operand_opcode; |
| 1250 | |
| 1251 | if (operand.type == OPERAND_IMMEDIATE) |
| 1252 | { |
| 1253 | /* xxx1 000x xxxx xxxx xxxx xxxx xxxx xxxx */ |
| 1254 | operand_opcode = ENCODE (8, 4, 25); |
| 1255 | |
| 1256 | return aarch64_emit_insn (buf, opcode | operand_opcode | size |
| 1257 | | ENCODE (operand.imm, 12, 10) |
| 1258 | | ENCODE (rn.num, 5, 5) |
| 1259 | | ENCODE (rd.num, 5, 0)); |
| 1260 | } |
| 1261 | else |
| 1262 | { |
| 1263 | /* xxx0 101x xxxx xxxx xxxx xxxx xxxx xxxx */ |
| 1264 | operand_opcode = ENCODE (5, 4, 25); |
| 1265 | |
| 1266 | return emit_data_processing_reg (buf, opcode | operand_opcode, rd, |
| 1267 | rn, operand.reg); |
| 1268 | } |
| 1269 | } |
| 1270 | |
| 1271 | /* Write an ADD instruction into *BUF. |
| 1272 | |
| 1273 | ADD rd, rn, #imm |
| 1274 | ADD rd, rn, rm |
| 1275 | |
| 1276 | This function handles both an immediate and register add. |
| 1277 | |
| 1278 | RD is the destination register. |
| 1279 | RN is the input register. |
| 1280 | OPERAND is the source operand, either of type OPERAND_IMMEDIATE or |
| 1281 | OPERAND_REGISTER. */ |
| 1282 | |
| 1283 | static int |
| 1284 | emit_add (uint32_t *buf, struct aarch64_register rd, |
| 1285 | struct aarch64_register rn, struct aarch64_operand operand) |
| 1286 | { |
| 1287 | return emit_data_processing (buf, ADD, rd, rn, operand); |
| 1288 | } |
| 1289 | |
| 1290 | /* Write a SUB instruction into *BUF. |
| 1291 | |
| 1292 | SUB rd, rn, #imm |
| 1293 | SUB rd, rn, rm |
| 1294 | |
| 1295 | This function handles both an immediate and register sub. |
| 1296 | |
| 1297 | RD is the destination register. |
| 1298 | RN is the input register. |
| 1299 | IMM is the immediate to substract to RN. */ |
| 1300 | |
| 1301 | static int |
| 1302 | emit_sub (uint32_t *buf, struct aarch64_register rd, |
| 1303 | struct aarch64_register rn, struct aarch64_operand operand) |
| 1304 | { |
| 1305 | return emit_data_processing (buf, SUB, rd, rn, operand); |
| 1306 | } |
| 1307 | |
| 1308 | /* Write a MOV instruction into *BUF. |
| 1309 | |
| 1310 | MOV rd, #imm |
| 1311 | MOV rd, rm |
| 1312 | |
| 1313 | This function handles both a wide immediate move and a register move, |
| 1314 | with the condition that the source register is not xzr. xzr and the |
| 1315 | stack pointer share the same encoding and this function only supports |
| 1316 | the stack pointer. |
| 1317 | |
| 1318 | RD is the destination register. |
| 1319 | OPERAND is the source operand, either of type OPERAND_IMMEDIATE or |
| 1320 | OPERAND_REGISTER. */ |
| 1321 | |
| 1322 | static int |
| 1323 | emit_mov (uint32_t *buf, struct aarch64_register rd, |
| 1324 | struct aarch64_operand operand) |
| 1325 | { |
| 1326 | if (operand.type == OPERAND_IMMEDIATE) |
| 1327 | { |
| 1328 | uint32_t size = ENCODE (rd.is64, 1, 31); |
| 1329 | /* Do not shift the immediate. */ |
| 1330 | uint32_t shift = ENCODE (0, 2, 21); |
| 1331 | |
| 1332 | return aarch64_emit_insn (buf, MOV | size | shift |
| 1333 | | ENCODE (operand.imm, 16, 5) |
| 1334 | | ENCODE (rd.num, 5, 0)); |
| 1335 | } |
| 1336 | else |
| 1337 | return emit_add (buf, rd, operand.reg, immediate_operand (0)); |
| 1338 | } |
| 1339 | |
| 1340 | /* Write a MOVK instruction into *BUF. |
| 1341 | |
| 1342 | MOVK rd, #imm, lsl #shift |
| 1343 | |
| 1344 | RD is the destination register. |
| 1345 | IMM is the immediate. |
| 1346 | SHIFT is the logical shift left to apply to IMM. */ |
| 1347 | |
| 1348 | static int |
| 1349 | emit_movk (uint32_t *buf, struct aarch64_register rd, uint32_t imm, |
| 1350 | unsigned shift) |
| 1351 | { |
| 1352 | uint32_t size = ENCODE (rd.is64, 1, 31); |
| 1353 | |
| 1354 | return aarch64_emit_insn (buf, MOVK | size | ENCODE (shift, 2, 21) | |
| 1355 | ENCODE (imm, 16, 5) | ENCODE (rd.num, 5, 0)); |
| 1356 | } |
| 1357 | |
| 1358 | /* Write instructions into *BUF in order to move ADDR into a register. |
| 1359 | ADDR can be a 64-bit value. |
| 1360 | |
| 1361 | This function will emit a series of MOV and MOVK instructions, such as: |
| 1362 | |
| 1363 | MOV xd, #(addr) |
| 1364 | MOVK xd, #(addr >> 16), lsl #16 |
| 1365 | MOVK xd, #(addr >> 32), lsl #32 |
| 1366 | MOVK xd, #(addr >> 48), lsl #48 */ |
| 1367 | |
| 1368 | static int |
| 1369 | emit_mov_addr (uint32_t *buf, struct aarch64_register rd, CORE_ADDR addr) |
| 1370 | { |
| 1371 | uint32_t *p = buf; |
| 1372 | |
| 1373 | /* The MOV (wide immediate) instruction clears to top bits of the |
| 1374 | register. */ |
| 1375 | p += emit_mov (p, rd, immediate_operand (addr & 0xffff)); |
| 1376 | |
| 1377 | if ((addr >> 16) != 0) |
| 1378 | p += emit_movk (p, rd, (addr >> 16) & 0xffff, 1); |
| 1379 | else |
| 1380 | return p - buf; |
| 1381 | |
| 1382 | if ((addr >> 32) != 0) |
| 1383 | p += emit_movk (p, rd, (addr >> 32) & 0xffff, 2); |
| 1384 | else |
| 1385 | return p - buf; |
| 1386 | |
| 1387 | if ((addr >> 48) != 0) |
| 1388 | p += emit_movk (p, rd, (addr >> 48) & 0xffff, 3); |
| 1389 | |
| 1390 | return p - buf; |
| 1391 | } |
| 1392 | |
| 1393 | /* Write a SUBS instruction into *BUF. |
| 1394 | |
| 1395 | SUBS rd, rn, rm |
| 1396 | |
| 1397 | This instruction update the condition flags. |
| 1398 | |
| 1399 | RD is the destination register. |
| 1400 | RN and RM are the source registers. */ |
| 1401 | |
| 1402 | static int |
| 1403 | emit_subs (uint32_t *buf, struct aarch64_register rd, |
| 1404 | struct aarch64_register rn, struct aarch64_operand operand) |
| 1405 | { |
| 1406 | return emit_data_processing (buf, SUBS, rd, rn, operand); |
| 1407 | } |
| 1408 | |
| 1409 | /* Write a CMP instruction into *BUF. |
| 1410 | |
| 1411 | CMP rn, rm |
| 1412 | |
| 1413 | This instruction is an alias of SUBS xzr, rn, rm. |
| 1414 | |
| 1415 | RN and RM are the registers to compare. */ |
| 1416 | |
| 1417 | static int |
| 1418 | emit_cmp (uint32_t *buf, struct aarch64_register rn, |
| 1419 | struct aarch64_operand operand) |
| 1420 | { |
| 1421 | return emit_subs (buf, xzr, rn, operand); |
| 1422 | } |
| 1423 | |
| 1424 | /* Write a AND instruction into *BUF. |
| 1425 | |
| 1426 | AND rd, rn, rm |
| 1427 | |
| 1428 | RD is the destination register. |
| 1429 | RN and RM are the source registers. */ |
| 1430 | |
| 1431 | static int |
| 1432 | emit_and (uint32_t *buf, struct aarch64_register rd, |
| 1433 | struct aarch64_register rn, struct aarch64_register rm) |
| 1434 | { |
| 1435 | return emit_data_processing_reg (buf, AND, rd, rn, rm); |
| 1436 | } |
| 1437 | |
| 1438 | /* Write a ORR instruction into *BUF. |
| 1439 | |
| 1440 | ORR rd, rn, rm |
| 1441 | |
| 1442 | RD is the destination register. |
| 1443 | RN and RM are the source registers. */ |
| 1444 | |
| 1445 | static int |
| 1446 | emit_orr (uint32_t *buf, struct aarch64_register rd, |
| 1447 | struct aarch64_register rn, struct aarch64_register rm) |
| 1448 | { |
| 1449 | return emit_data_processing_reg (buf, ORR, rd, rn, rm); |
| 1450 | } |
| 1451 | |
| 1452 | /* Write a ORN instruction into *BUF. |
| 1453 | |
| 1454 | ORN rd, rn, rm |
| 1455 | |
| 1456 | RD is the destination register. |
| 1457 | RN and RM are the source registers. */ |
| 1458 | |
| 1459 | static int |
| 1460 | emit_orn (uint32_t *buf, struct aarch64_register rd, |
| 1461 | struct aarch64_register rn, struct aarch64_register rm) |
| 1462 | { |
| 1463 | return emit_data_processing_reg (buf, ORN, rd, rn, rm); |
| 1464 | } |
| 1465 | |
| 1466 | /* Write a EOR instruction into *BUF. |
| 1467 | |
| 1468 | EOR rd, rn, rm |
| 1469 | |
| 1470 | RD is the destination register. |
| 1471 | RN and RM are the source registers. */ |
| 1472 | |
| 1473 | static int |
| 1474 | emit_eor (uint32_t *buf, struct aarch64_register rd, |
| 1475 | struct aarch64_register rn, struct aarch64_register rm) |
| 1476 | { |
| 1477 | return emit_data_processing_reg (buf, EOR, rd, rn, rm); |
| 1478 | } |
| 1479 | |
| 1480 | /* Write a MVN instruction into *BUF. |
| 1481 | |
| 1482 | MVN rd, rm |
| 1483 | |
| 1484 | This is an alias for ORN rd, xzr, rm. |
| 1485 | |
| 1486 | RD is the destination register. |
| 1487 | RM is the source register. */ |
| 1488 | |
| 1489 | static int |
| 1490 | emit_mvn (uint32_t *buf, struct aarch64_register rd, |
| 1491 | struct aarch64_register rm) |
| 1492 | { |
| 1493 | return emit_orn (buf, rd, xzr, rm); |
| 1494 | } |
| 1495 | |
| 1496 | /* Write a LSLV instruction into *BUF. |
| 1497 | |
| 1498 | LSLV rd, rn, rm |
| 1499 | |
| 1500 | RD is the destination register. |
| 1501 | RN and RM are the source registers. */ |
| 1502 | |
| 1503 | static int |
| 1504 | emit_lslv (uint32_t *buf, struct aarch64_register rd, |
| 1505 | struct aarch64_register rn, struct aarch64_register rm) |
| 1506 | { |
| 1507 | return emit_data_processing_reg (buf, LSLV, rd, rn, rm); |
| 1508 | } |
| 1509 | |
| 1510 | /* Write a LSRV instruction into *BUF. |
| 1511 | |
| 1512 | LSRV rd, rn, rm |
| 1513 | |
| 1514 | RD is the destination register. |
| 1515 | RN and RM are the source registers. */ |
| 1516 | |
| 1517 | static int |
| 1518 | emit_lsrv (uint32_t *buf, struct aarch64_register rd, |
| 1519 | struct aarch64_register rn, struct aarch64_register rm) |
| 1520 | { |
| 1521 | return emit_data_processing_reg (buf, LSRV, rd, rn, rm); |
| 1522 | } |
| 1523 | |
| 1524 | /* Write a ASRV instruction into *BUF. |
| 1525 | |
| 1526 | ASRV rd, rn, rm |
| 1527 | |
| 1528 | RD is the destination register. |
| 1529 | RN and RM are the source registers. */ |
| 1530 | |
| 1531 | static int |
| 1532 | emit_asrv (uint32_t *buf, struct aarch64_register rd, |
| 1533 | struct aarch64_register rn, struct aarch64_register rm) |
| 1534 | { |
| 1535 | return emit_data_processing_reg (buf, ASRV, rd, rn, rm); |
| 1536 | } |
| 1537 | |
| 1538 | /* Write a MUL instruction into *BUF. |
| 1539 | |
| 1540 | MUL rd, rn, rm |
| 1541 | |
| 1542 | RD is the destination register. |
| 1543 | RN and RM are the source registers. */ |
| 1544 | |
| 1545 | static int |
| 1546 | emit_mul (uint32_t *buf, struct aarch64_register rd, |
| 1547 | struct aarch64_register rn, struct aarch64_register rm) |
| 1548 | { |
| 1549 | return emit_data_processing_reg (buf, MUL, rd, rn, rm); |
| 1550 | } |
| 1551 | |
| 1552 | /* Write a MRS instruction into *BUF. The register size is 64-bit. |
| 1553 | |
| 1554 | MRS xt, system_reg |
| 1555 | |
| 1556 | RT is the destination register. |
| 1557 | SYSTEM_REG is special purpose register to read. */ |
| 1558 | |
| 1559 | static int |
| 1560 | emit_mrs (uint32_t *buf, struct aarch64_register rt, |
| 1561 | enum aarch64_system_control_registers system_reg) |
| 1562 | { |
| 1563 | return aarch64_emit_insn (buf, MRS | ENCODE (system_reg, 15, 5) |
| 1564 | | ENCODE (rt.num, 5, 0)); |
| 1565 | } |
| 1566 | |
| 1567 | /* Write a MSR instruction into *BUF. The register size is 64-bit. |
| 1568 | |
| 1569 | MSR system_reg, xt |
| 1570 | |
| 1571 | SYSTEM_REG is special purpose register to write. |
| 1572 | RT is the input register. */ |
| 1573 | |
| 1574 | static int |
| 1575 | emit_msr (uint32_t *buf, enum aarch64_system_control_registers system_reg, |
| 1576 | struct aarch64_register rt) |
| 1577 | { |
| 1578 | return aarch64_emit_insn (buf, MSR | ENCODE (system_reg, 15, 5) |
| 1579 | | ENCODE (rt.num, 5, 0)); |
| 1580 | } |
| 1581 | |
| 1582 | /* Write a SEVL instruction into *BUF. |
| 1583 | |
| 1584 | This is a hint instruction telling the hardware to trigger an event. */ |
| 1585 | |
| 1586 | static int |
| 1587 | emit_sevl (uint32_t *buf) |
| 1588 | { |
| 1589 | return aarch64_emit_insn (buf, SEVL); |
| 1590 | } |
| 1591 | |
| 1592 | /* Write a WFE instruction into *BUF. |
| 1593 | |
| 1594 | This is a hint instruction telling the hardware to wait for an event. */ |
| 1595 | |
| 1596 | static int |
| 1597 | emit_wfe (uint32_t *buf) |
| 1598 | { |
| 1599 | return aarch64_emit_insn (buf, WFE); |
| 1600 | } |
| 1601 | |
| 1602 | /* Write a SBFM instruction into *BUF. |
| 1603 | |
| 1604 | SBFM rd, rn, #immr, #imms |
| 1605 | |
| 1606 | This instruction moves the bits from #immr to #imms into the |
| 1607 | destination, sign extending the result. |
| 1608 | |
| 1609 | RD is the destination register. |
| 1610 | RN is the source register. |
| 1611 | IMMR is the bit number to start at (least significant bit). |
| 1612 | IMMS is the bit number to stop at (most significant bit). */ |
| 1613 | |
| 1614 | static int |
| 1615 | emit_sbfm (uint32_t *buf, struct aarch64_register rd, |
| 1616 | struct aarch64_register rn, uint32_t immr, uint32_t imms) |
| 1617 | { |
| 1618 | uint32_t size = ENCODE (rd.is64, 1, 31); |
| 1619 | uint32_t n = ENCODE (rd.is64, 1, 22); |
| 1620 | |
| 1621 | return aarch64_emit_insn (buf, SBFM | size | n | ENCODE (immr, 6, 16) |
| 1622 | | ENCODE (imms, 6, 10) | ENCODE (rn.num, 5, 5) |
| 1623 | | ENCODE (rd.num, 5, 0)); |
| 1624 | } |
| 1625 | |
| 1626 | /* Write a SBFX instruction into *BUF. |
| 1627 | |
| 1628 | SBFX rd, rn, #lsb, #width |
| 1629 | |
| 1630 | This instruction moves #width bits from #lsb into the destination, sign |
| 1631 | extending the result. This is an alias for: |
| 1632 | |
| 1633 | SBFM rd, rn, #lsb, #(lsb + width - 1) |
| 1634 | |
| 1635 | RD is the destination register. |
| 1636 | RN is the source register. |
| 1637 | LSB is the bit number to start at (least significant bit). |
| 1638 | WIDTH is the number of bits to move. */ |
| 1639 | |
| 1640 | static int |
| 1641 | emit_sbfx (uint32_t *buf, struct aarch64_register rd, |
| 1642 | struct aarch64_register rn, uint32_t lsb, uint32_t width) |
| 1643 | { |
| 1644 | return emit_sbfm (buf, rd, rn, lsb, lsb + width - 1); |
| 1645 | } |
| 1646 | |
| 1647 | /* Write a UBFM instruction into *BUF. |
| 1648 | |
| 1649 | UBFM rd, rn, #immr, #imms |
| 1650 | |
| 1651 | This instruction moves the bits from #immr to #imms into the |
| 1652 | destination, extending the result with zeros. |
| 1653 | |
| 1654 | RD is the destination register. |
| 1655 | RN is the source register. |
| 1656 | IMMR is the bit number to start at (least significant bit). |
| 1657 | IMMS is the bit number to stop at (most significant bit). */ |
| 1658 | |
| 1659 | static int |
| 1660 | emit_ubfm (uint32_t *buf, struct aarch64_register rd, |
| 1661 | struct aarch64_register rn, uint32_t immr, uint32_t imms) |
| 1662 | { |
| 1663 | uint32_t size = ENCODE (rd.is64, 1, 31); |
| 1664 | uint32_t n = ENCODE (rd.is64, 1, 22); |
| 1665 | |
| 1666 | return aarch64_emit_insn (buf, UBFM | size | n | ENCODE (immr, 6, 16) |
| 1667 | | ENCODE (imms, 6, 10) | ENCODE (rn.num, 5, 5) |
| 1668 | | ENCODE (rd.num, 5, 0)); |
| 1669 | } |
| 1670 | |
| 1671 | /* Write a UBFX instruction into *BUF. |
| 1672 | |
| 1673 | UBFX rd, rn, #lsb, #width |
| 1674 | |
| 1675 | This instruction moves #width bits from #lsb into the destination, |
| 1676 | extending the result with zeros. This is an alias for: |
| 1677 | |
| 1678 | UBFM rd, rn, #lsb, #(lsb + width - 1) |
| 1679 | |
| 1680 | RD is the destination register. |
| 1681 | RN is the source register. |
| 1682 | LSB is the bit number to start at (least significant bit). |
| 1683 | WIDTH is the number of bits to move. */ |
| 1684 | |
| 1685 | static int |
| 1686 | emit_ubfx (uint32_t *buf, struct aarch64_register rd, |
| 1687 | struct aarch64_register rn, uint32_t lsb, uint32_t width) |
| 1688 | { |
| 1689 | return emit_ubfm (buf, rd, rn, lsb, lsb + width - 1); |
| 1690 | } |
| 1691 | |
| 1692 | /* Write a CSINC instruction into *BUF. |
| 1693 | |
| 1694 | CSINC rd, rn, rm, cond |
| 1695 | |
| 1696 | This instruction conditionally increments rn or rm and places the result |
| 1697 | in rd. rn is chosen is the condition is true. |
| 1698 | |
| 1699 | RD is the destination register. |
| 1700 | RN and RM are the source registers. |
| 1701 | COND is the encoded condition. */ |
| 1702 | |
| 1703 | static int |
| 1704 | emit_csinc (uint32_t *buf, struct aarch64_register rd, |
| 1705 | struct aarch64_register rn, struct aarch64_register rm, |
| 1706 | unsigned cond) |
| 1707 | { |
| 1708 | uint32_t size = ENCODE (rd.is64, 1, 31); |
| 1709 | |
| 1710 | return aarch64_emit_insn (buf, CSINC | size | ENCODE (rm.num, 5, 16) |
| 1711 | | ENCODE (cond, 4, 12) | ENCODE (rn.num, 5, 5) |
| 1712 | | ENCODE (rd.num, 5, 0)); |
| 1713 | } |
| 1714 | |
| 1715 | /* Write a CSET instruction into *BUF. |
| 1716 | |
| 1717 | CSET rd, cond |
| 1718 | |
| 1719 | This instruction conditionally write 1 or 0 in the destination register. |
| 1720 | 1 is written if the condition is true. This is an alias for: |
| 1721 | |
| 1722 | CSINC rd, xzr, xzr, !cond |
| 1723 | |
| 1724 | Note that the condition needs to be inverted. |
| 1725 | |
| 1726 | RD is the destination register. |
| 1727 | RN and RM are the source registers. |
| 1728 | COND is the encoded condition. */ |
| 1729 | |
| 1730 | static int |
| 1731 | emit_cset (uint32_t *buf, struct aarch64_register rd, unsigned cond) |
| 1732 | { |
| 1733 | /* The least significant bit of the condition needs toggling in order to |
| 1734 | invert it. */ |
| 1735 | return emit_csinc (buf, rd, xzr, xzr, cond ^ 0x1); |
| 1736 | } |
| 1737 | |
| 1738 | /* Write LEN instructions from BUF into the inferior memory at *TO. |
| 1739 | |
| 1740 | Note instructions are always little endian on AArch64, unlike data. */ |
| 1741 | |
| 1742 | static void |
| 1743 | append_insns (CORE_ADDR *to, size_t len, const uint32_t *buf) |
| 1744 | { |
| 1745 | size_t byte_len = len * sizeof (uint32_t); |
| 1746 | #if (__BYTE_ORDER == __BIG_ENDIAN) |
| 1747 | uint32_t *le_buf = (uint32_t *) xmalloc (byte_len); |
| 1748 | size_t i; |
| 1749 | |
| 1750 | for (i = 0; i < len; i++) |
| 1751 | le_buf[i] = htole32 (buf[i]); |
| 1752 | |
| 1753 | target_write_memory (*to, (const unsigned char *) le_buf, byte_len); |
| 1754 | |
| 1755 | xfree (le_buf); |
| 1756 | #else |
| 1757 | target_write_memory (*to, (const unsigned char *) buf, byte_len); |
| 1758 | #endif |
| 1759 | |
| 1760 | *to += byte_len; |
| 1761 | } |
| 1762 | |
| 1763 | /* Sub-class of struct aarch64_insn_data, store information of |
| 1764 | instruction relocation for fast tracepoint. Visitor can |
| 1765 | relocate an instruction from BASE.INSN_ADDR to NEW_ADDR and save |
| 1766 | the relocated instructions in buffer pointed by INSN_PTR. */ |
| 1767 | |
| 1768 | struct aarch64_insn_relocation_data |
| 1769 | { |
| 1770 | struct aarch64_insn_data base; |
| 1771 | |
| 1772 | /* The new address the instruction is relocated to. */ |
| 1773 | CORE_ADDR new_addr; |
| 1774 | /* Pointer to the buffer of relocated instruction(s). */ |
| 1775 | uint32_t *insn_ptr; |
| 1776 | }; |
| 1777 | |
| 1778 | /* Implementation of aarch64_insn_visitor method "b". */ |
| 1779 | |
| 1780 | static void |
| 1781 | aarch64_ftrace_insn_reloc_b (const int is_bl, const int32_t offset, |
| 1782 | struct aarch64_insn_data *data) |
| 1783 | { |
| 1784 | struct aarch64_insn_relocation_data *insn_reloc |
| 1785 | = (struct aarch64_insn_relocation_data *) data; |
| 1786 | int64_t new_offset |
| 1787 | = insn_reloc->base.insn_addr - insn_reloc->new_addr + offset; |
| 1788 | |
| 1789 | if (can_encode_int32 (new_offset, 28)) |
| 1790 | insn_reloc->insn_ptr += emit_b (insn_reloc->insn_ptr, is_bl, new_offset); |
| 1791 | } |
| 1792 | |
| 1793 | /* Implementation of aarch64_insn_visitor method "b_cond". */ |
| 1794 | |
| 1795 | static void |
| 1796 | aarch64_ftrace_insn_reloc_b_cond (const unsigned cond, const int32_t offset, |
| 1797 | struct aarch64_insn_data *data) |
| 1798 | { |
| 1799 | struct aarch64_insn_relocation_data *insn_reloc |
| 1800 | = (struct aarch64_insn_relocation_data *) data; |
| 1801 | int64_t new_offset |
| 1802 | = insn_reloc->base.insn_addr - insn_reloc->new_addr + offset; |
| 1803 | |
| 1804 | if (can_encode_int32 (new_offset, 21)) |
| 1805 | { |
| 1806 | insn_reloc->insn_ptr += emit_bcond (insn_reloc->insn_ptr, cond, |
| 1807 | new_offset); |
| 1808 | } |
| 1809 | else if (can_encode_int32 (new_offset, 28)) |
| 1810 | { |
| 1811 | /* The offset is out of range for a conditional branch |
| 1812 | instruction but not for a unconditional branch. We can use |
| 1813 | the following instructions instead: |
| 1814 | |
| 1815 | B.COND TAKEN ; If cond is true, then jump to TAKEN. |
| 1816 | B NOT_TAKEN ; Else jump over TAKEN and continue. |
| 1817 | TAKEN: |
| 1818 | B #(offset - 8) |
| 1819 | NOT_TAKEN: |
| 1820 | |
| 1821 | */ |
| 1822 | |
| 1823 | insn_reloc->insn_ptr += emit_bcond (insn_reloc->insn_ptr, cond, 8); |
| 1824 | insn_reloc->insn_ptr += emit_b (insn_reloc->insn_ptr, 0, 8); |
| 1825 | insn_reloc->insn_ptr += emit_b (insn_reloc->insn_ptr, 0, new_offset - 8); |
| 1826 | } |
| 1827 | } |
| 1828 | |
| 1829 | /* Implementation of aarch64_insn_visitor method "cb". */ |
| 1830 | |
| 1831 | static void |
| 1832 | aarch64_ftrace_insn_reloc_cb (const int32_t offset, const int is_cbnz, |
| 1833 | const unsigned rn, int is64, |
| 1834 | struct aarch64_insn_data *data) |
| 1835 | { |
| 1836 | struct aarch64_insn_relocation_data *insn_reloc |
| 1837 | = (struct aarch64_insn_relocation_data *) data; |
| 1838 | int64_t new_offset |
| 1839 | = insn_reloc->base.insn_addr - insn_reloc->new_addr + offset; |
| 1840 | |
| 1841 | if (can_encode_int32 (new_offset, 21)) |
| 1842 | { |
| 1843 | insn_reloc->insn_ptr += emit_cb (insn_reloc->insn_ptr, is_cbnz, |
| 1844 | aarch64_register (rn, is64), new_offset); |
| 1845 | } |
| 1846 | else if (can_encode_int32 (new_offset, 28)) |
| 1847 | { |
| 1848 | /* The offset is out of range for a compare and branch |
| 1849 | instruction but not for a unconditional branch. We can use |
| 1850 | the following instructions instead: |
| 1851 | |
| 1852 | CBZ xn, TAKEN ; xn == 0, then jump to TAKEN. |
| 1853 | B NOT_TAKEN ; Else jump over TAKEN and continue. |
| 1854 | TAKEN: |
| 1855 | B #(offset - 8) |
| 1856 | NOT_TAKEN: |
| 1857 | |
| 1858 | */ |
| 1859 | insn_reloc->insn_ptr += emit_cb (insn_reloc->insn_ptr, is_cbnz, |
| 1860 | aarch64_register (rn, is64), 8); |
| 1861 | insn_reloc->insn_ptr += emit_b (insn_reloc->insn_ptr, 0, 8); |
| 1862 | insn_reloc->insn_ptr += emit_b (insn_reloc->insn_ptr, 0, new_offset - 8); |
| 1863 | } |
| 1864 | } |
| 1865 | |
| 1866 | /* Implementation of aarch64_insn_visitor method "tb". */ |
| 1867 | |
| 1868 | static void |
| 1869 | aarch64_ftrace_insn_reloc_tb (const int32_t offset, int is_tbnz, |
| 1870 | const unsigned rt, unsigned bit, |
| 1871 | struct aarch64_insn_data *data) |
| 1872 | { |
| 1873 | struct aarch64_insn_relocation_data *insn_reloc |
| 1874 | = (struct aarch64_insn_relocation_data *) data; |
| 1875 | int64_t new_offset |
| 1876 | = insn_reloc->base.insn_addr - insn_reloc->new_addr + offset; |
| 1877 | |
| 1878 | if (can_encode_int32 (new_offset, 16)) |
| 1879 | { |
| 1880 | insn_reloc->insn_ptr += emit_tb (insn_reloc->insn_ptr, is_tbnz, bit, |
| 1881 | aarch64_register (rt, 1), new_offset); |
| 1882 | } |
| 1883 | else if (can_encode_int32 (new_offset, 28)) |
| 1884 | { |
| 1885 | /* The offset is out of range for a test bit and branch |
| 1886 | instruction but not for a unconditional branch. We can use |
| 1887 | the following instructions instead: |
| 1888 | |
| 1889 | TBZ xn, #bit, TAKEN ; xn[bit] == 0, then jump to TAKEN. |
| 1890 | B NOT_TAKEN ; Else jump over TAKEN and continue. |
| 1891 | TAKEN: |
| 1892 | B #(offset - 8) |
| 1893 | NOT_TAKEN: |
| 1894 | |
| 1895 | */ |
| 1896 | insn_reloc->insn_ptr += emit_tb (insn_reloc->insn_ptr, is_tbnz, bit, |
| 1897 | aarch64_register (rt, 1), 8); |
| 1898 | insn_reloc->insn_ptr += emit_b (insn_reloc->insn_ptr, 0, 8); |
| 1899 | insn_reloc->insn_ptr += emit_b (insn_reloc->insn_ptr, 0, |
| 1900 | new_offset - 8); |
| 1901 | } |
| 1902 | } |
| 1903 | |
| 1904 | /* Implementation of aarch64_insn_visitor method "adr". */ |
| 1905 | |
| 1906 | static void |
| 1907 | aarch64_ftrace_insn_reloc_adr (const int32_t offset, const unsigned rd, |
| 1908 | const int is_adrp, |
| 1909 | struct aarch64_insn_data *data) |
| 1910 | { |
| 1911 | struct aarch64_insn_relocation_data *insn_reloc |
| 1912 | = (struct aarch64_insn_relocation_data *) data; |
| 1913 | /* We know exactly the address the ADR{P,} instruction will compute. |
| 1914 | We can just write it to the destination register. */ |
| 1915 | CORE_ADDR address = data->insn_addr + offset; |
| 1916 | |
| 1917 | if (is_adrp) |
| 1918 | { |
| 1919 | /* Clear the lower 12 bits of the offset to get the 4K page. */ |
| 1920 | insn_reloc->insn_ptr += emit_mov_addr (insn_reloc->insn_ptr, |
| 1921 | aarch64_register (rd, 1), |
| 1922 | address & ~0xfff); |
| 1923 | } |
| 1924 | else |
| 1925 | insn_reloc->insn_ptr += emit_mov_addr (insn_reloc->insn_ptr, |
| 1926 | aarch64_register (rd, 1), address); |
| 1927 | } |
| 1928 | |
| 1929 | /* Implementation of aarch64_insn_visitor method "ldr_literal". */ |
| 1930 | |
| 1931 | static void |
| 1932 | aarch64_ftrace_insn_reloc_ldr_literal (const int32_t offset, const int is_sw, |
| 1933 | const unsigned rt, const int is64, |
| 1934 | struct aarch64_insn_data *data) |
| 1935 | { |
| 1936 | struct aarch64_insn_relocation_data *insn_reloc |
| 1937 | = (struct aarch64_insn_relocation_data *) data; |
| 1938 | CORE_ADDR address = data->insn_addr + offset; |
| 1939 | |
| 1940 | insn_reloc->insn_ptr += emit_mov_addr (insn_reloc->insn_ptr, |
| 1941 | aarch64_register (rt, 1), address); |
| 1942 | |
| 1943 | /* We know exactly what address to load from, and what register we |
| 1944 | can use: |
| 1945 | |
| 1946 | MOV xd, #(oldloc + offset) |
| 1947 | MOVK xd, #((oldloc + offset) >> 16), lsl #16 |
| 1948 | ... |
| 1949 | |
| 1950 | LDR xd, [xd] ; or LDRSW xd, [xd] |
| 1951 | |
| 1952 | */ |
| 1953 | |
| 1954 | if (is_sw) |
| 1955 | insn_reloc->insn_ptr += emit_ldrsw (insn_reloc->insn_ptr, |
| 1956 | aarch64_register (rt, 1), |
| 1957 | aarch64_register (rt, 1), |
| 1958 | offset_memory_operand (0)); |
| 1959 | else |
| 1960 | insn_reloc->insn_ptr += emit_ldr (insn_reloc->insn_ptr, |
| 1961 | aarch64_register (rt, is64), |
| 1962 | aarch64_register (rt, 1), |
| 1963 | offset_memory_operand (0)); |
| 1964 | } |
| 1965 | |
| 1966 | /* Implementation of aarch64_insn_visitor method "others". */ |
| 1967 | |
| 1968 | static void |
| 1969 | aarch64_ftrace_insn_reloc_others (const uint32_t insn, |
| 1970 | struct aarch64_insn_data *data) |
| 1971 | { |
| 1972 | struct aarch64_insn_relocation_data *insn_reloc |
| 1973 | = (struct aarch64_insn_relocation_data *) data; |
| 1974 | |
| 1975 | /* The instruction is not PC relative. Just re-emit it at the new |
| 1976 | location. */ |
| 1977 | insn_reloc->insn_ptr += aarch64_emit_insn (insn_reloc->insn_ptr, insn); |
| 1978 | } |
| 1979 | |
| 1980 | static const struct aarch64_insn_visitor visitor = |
| 1981 | { |
| 1982 | aarch64_ftrace_insn_reloc_b, |
| 1983 | aarch64_ftrace_insn_reloc_b_cond, |
| 1984 | aarch64_ftrace_insn_reloc_cb, |
| 1985 | aarch64_ftrace_insn_reloc_tb, |
| 1986 | aarch64_ftrace_insn_reloc_adr, |
| 1987 | aarch64_ftrace_insn_reloc_ldr_literal, |
| 1988 | aarch64_ftrace_insn_reloc_others, |
| 1989 | }; |
| 1990 | |
| 1991 | bool |
| 1992 | aarch64_target::supports_fast_tracepoints () |
| 1993 | { |
| 1994 | return true; |
| 1995 | } |
| 1996 | |
| 1997 | /* Implementation of target ops method |
| 1998 | "install_fast_tracepoint_jump_pad". */ |
| 1999 | |
| 2000 | int |
| 2001 | aarch64_target::install_fast_tracepoint_jump_pad |
| 2002 | (CORE_ADDR tpoint, CORE_ADDR tpaddr, CORE_ADDR collector, |
| 2003 | CORE_ADDR lockaddr, ULONGEST orig_size, CORE_ADDR *jump_entry, |
| 2004 | CORE_ADDR *trampoline, ULONGEST *trampoline_size, |
| 2005 | unsigned char *jjump_pad_insn, ULONGEST *jjump_pad_insn_size, |
| 2006 | CORE_ADDR *adjusted_insn_addr, CORE_ADDR *adjusted_insn_addr_end, |
| 2007 | char *err) |
| 2008 | { |
| 2009 | uint32_t buf[256]; |
| 2010 | uint32_t *p = buf; |
| 2011 | int64_t offset; |
| 2012 | int i; |
| 2013 | uint32_t insn; |
| 2014 | CORE_ADDR buildaddr = *jump_entry; |
| 2015 | struct aarch64_insn_relocation_data insn_data; |
| 2016 | |
| 2017 | /* We need to save the current state on the stack both to restore it |
| 2018 | later and to collect register values when the tracepoint is hit. |
| 2019 | |
| 2020 | The saved registers are pushed in a layout that needs to be in sync |
| 2021 | with aarch64_ft_collect_regmap (see linux-aarch64-ipa.c). Later on |
| 2022 | the supply_fast_tracepoint_registers function will fill in the |
| 2023 | register cache from a pointer to saved registers on the stack we build |
| 2024 | here. |
| 2025 | |
| 2026 | For simplicity, we set the size of each cell on the stack to 16 bytes. |
| 2027 | This way one cell can hold any register type, from system registers |
| 2028 | to the 128 bit SIMD&FP registers. Furthermore, the stack pointer |
| 2029 | has to be 16 bytes aligned anyway. |
| 2030 | |
| 2031 | Note that the CPSR register does not exist on AArch64. Instead we |
| 2032 | can access system bits describing the process state with the |
| 2033 | MRS/MSR instructions, namely the condition flags. We save them as |
| 2034 | if they are part of a CPSR register because that's how GDB |
| 2035 | interprets these system bits. At the moment, only the condition |
| 2036 | flags are saved in CPSR (NZCV). |
| 2037 | |
| 2038 | Stack layout, each cell is 16 bytes (descending): |
| 2039 | |
| 2040 | High *-------- SIMD&FP registers from 31 down to 0. --------* |
| 2041 | | q31 | |
| 2042 | . . |
| 2043 | . . 32 cells |
| 2044 | . . |
| 2045 | | q0 | |
| 2046 | *---- General purpose registers from 30 down to 0. ----* |
| 2047 | | x30 | |
| 2048 | . . |
| 2049 | . . 31 cells |
| 2050 | . . |
| 2051 | | x0 | |
| 2052 | *------------- Special purpose registers. -------------* |
| 2053 | | SP | |
| 2054 | | PC | |
| 2055 | | CPSR (NZCV) | 5 cells |
| 2056 | | FPSR | |
| 2057 | | FPCR | <- SP + 16 |
| 2058 | *------------- collecting_t object --------------------* |
| 2059 | | TPIDR_EL0 | struct tracepoint * | |
| 2060 | Low *------------------------------------------------------* |
| 2061 | |
| 2062 | After this stack is set up, we issue a call to the collector, passing |
| 2063 | it the saved registers at (SP + 16). */ |
| 2064 | |
| 2065 | /* Push SIMD&FP registers on the stack: |
| 2066 | |
| 2067 | SUB sp, sp, #(32 * 16) |
| 2068 | |
| 2069 | STP q30, q31, [sp, #(30 * 16)] |
| 2070 | ... |
| 2071 | STP q0, q1, [sp] |
| 2072 | |
| 2073 | */ |
| 2074 | p += emit_sub (p, sp, sp, immediate_operand (32 * 16)); |
| 2075 | for (i = 30; i >= 0; i -= 2) |
| 2076 | p += emit_stp_q_offset (p, i, i + 1, sp, i * 16); |
| 2077 | |
| 2078 | /* Push general purpose registers on the stack. Note that we do not need |
| 2079 | to push x31 as it represents the xzr register and not the stack |
| 2080 | pointer in a STR instruction. |
| 2081 | |
| 2082 | SUB sp, sp, #(31 * 16) |
| 2083 | |
| 2084 | STR x30, [sp, #(30 * 16)] |
| 2085 | ... |
| 2086 | STR x0, [sp] |
| 2087 | |
| 2088 | */ |
| 2089 | p += emit_sub (p, sp, sp, immediate_operand (31 * 16)); |
| 2090 | for (i = 30; i >= 0; i -= 1) |
| 2091 | p += emit_str (p, aarch64_register (i, 1), sp, |
| 2092 | offset_memory_operand (i * 16)); |
| 2093 | |
| 2094 | /* Make space for 5 more cells. |
| 2095 | |
| 2096 | SUB sp, sp, #(5 * 16) |
| 2097 | |
| 2098 | */ |
| 2099 | p += emit_sub (p, sp, sp, immediate_operand (5 * 16)); |
| 2100 | |
| 2101 | |
| 2102 | /* Save SP: |
| 2103 | |
| 2104 | ADD x4, sp, #((32 + 31 + 5) * 16) |
| 2105 | STR x4, [sp, #(4 * 16)] |
| 2106 | |
| 2107 | */ |
| 2108 | p += emit_add (p, x4, sp, immediate_operand ((32 + 31 + 5) * 16)); |
| 2109 | p += emit_str (p, x4, sp, offset_memory_operand (4 * 16)); |
| 2110 | |
| 2111 | /* Save PC (tracepoint address): |
| 2112 | |
| 2113 | MOV x3, #(tpaddr) |
| 2114 | ... |
| 2115 | |
| 2116 | STR x3, [sp, #(3 * 16)] |
| 2117 | |
| 2118 | */ |
| 2119 | |
| 2120 | p += emit_mov_addr (p, x3, tpaddr); |
| 2121 | p += emit_str (p, x3, sp, offset_memory_operand (3 * 16)); |
| 2122 | |
| 2123 | /* Save CPSR (NZCV), FPSR and FPCR: |
| 2124 | |
| 2125 | MRS x2, nzcv |
| 2126 | MRS x1, fpsr |
| 2127 | MRS x0, fpcr |
| 2128 | |
| 2129 | STR x2, [sp, #(2 * 16)] |
| 2130 | STR x1, [sp, #(1 * 16)] |
| 2131 | STR x0, [sp, #(0 * 16)] |
| 2132 | |
| 2133 | */ |
| 2134 | p += emit_mrs (p, x2, NZCV); |
| 2135 | p += emit_mrs (p, x1, FPSR); |
| 2136 | p += emit_mrs (p, x0, FPCR); |
| 2137 | p += emit_str (p, x2, sp, offset_memory_operand (2 * 16)); |
| 2138 | p += emit_str (p, x1, sp, offset_memory_operand (1 * 16)); |
| 2139 | p += emit_str (p, x0, sp, offset_memory_operand (0 * 16)); |
| 2140 | |
| 2141 | /* Push the collecting_t object. It consist of the address of the |
| 2142 | tracepoint and an ID for the current thread. We get the latter by |
| 2143 | reading the tpidr_el0 system register. It corresponds to the |
| 2144 | NT_ARM_TLS register accessible with ptrace. |
| 2145 | |
| 2146 | MOV x0, #(tpoint) |
| 2147 | ... |
| 2148 | |
| 2149 | MRS x1, tpidr_el0 |
| 2150 | |
| 2151 | STP x0, x1, [sp, #-16]! |
| 2152 | |
| 2153 | */ |
| 2154 | |
| 2155 | p += emit_mov_addr (p, x0, tpoint); |
| 2156 | p += emit_mrs (p, x1, TPIDR_EL0); |
| 2157 | p += emit_stp (p, x0, x1, sp, preindex_memory_operand (-16)); |
| 2158 | |
| 2159 | /* Spin-lock: |
| 2160 | |
| 2161 | The shared memory for the lock is at lockaddr. It will hold zero |
| 2162 | if no-one is holding the lock, otherwise it contains the address of |
| 2163 | the collecting_t object on the stack of the thread which acquired it. |
| 2164 | |
| 2165 | At this stage, the stack pointer points to this thread's collecting_t |
| 2166 | object. |
| 2167 | |
| 2168 | We use the following registers: |
| 2169 | - x0: Address of the lock. |
| 2170 | - x1: Pointer to collecting_t object. |
| 2171 | - x2: Scratch register. |
| 2172 | |
| 2173 | MOV x0, #(lockaddr) |
| 2174 | ... |
| 2175 | MOV x1, sp |
| 2176 | |
| 2177 | ; Trigger an event local to this core. So the following WFE |
| 2178 | ; instruction is ignored. |
| 2179 | SEVL |
| 2180 | again: |
| 2181 | ; Wait for an event. The event is triggered by either the SEVL |
| 2182 | ; or STLR instructions (store release). |
| 2183 | WFE |
| 2184 | |
| 2185 | ; Atomically read at lockaddr. This marks the memory location as |
| 2186 | ; exclusive. This instruction also has memory constraints which |
| 2187 | ; make sure all previous data reads and writes are done before |
| 2188 | ; executing it. |
| 2189 | LDAXR x2, [x0] |
| 2190 | |
| 2191 | ; Try again if another thread holds the lock. |
| 2192 | CBNZ x2, again |
| 2193 | |
| 2194 | ; We can lock it! Write the address of the collecting_t object. |
| 2195 | ; This instruction will fail if the memory location is not marked |
| 2196 | ; as exclusive anymore. If it succeeds, it will remove the |
| 2197 | ; exclusive mark on the memory location. This way, if another |
| 2198 | ; thread executes this instruction before us, we will fail and try |
| 2199 | ; all over again. |
| 2200 | STXR w2, x1, [x0] |
| 2201 | CBNZ w2, again |
| 2202 | |
| 2203 | */ |
| 2204 | |
| 2205 | p += emit_mov_addr (p, x0, lockaddr); |
| 2206 | p += emit_mov (p, x1, register_operand (sp)); |
| 2207 | |
| 2208 | p += emit_sevl (p); |
| 2209 | p += emit_wfe (p); |
| 2210 | p += emit_ldaxr (p, x2, x0); |
| 2211 | p += emit_cb (p, 1, w2, -2 * 4); |
| 2212 | p += emit_stxr (p, w2, x1, x0); |
| 2213 | p += emit_cb (p, 1, x2, -4 * 4); |
| 2214 | |
| 2215 | /* Call collector (struct tracepoint *, unsigned char *): |
| 2216 | |
| 2217 | MOV x0, #(tpoint) |
| 2218 | ... |
| 2219 | |
| 2220 | ; Saved registers start after the collecting_t object. |
| 2221 | ADD x1, sp, #16 |
| 2222 | |
| 2223 | ; We use an intra-procedure-call scratch register. |
| 2224 | MOV ip0, #(collector) |
| 2225 | ... |
| 2226 | |
| 2227 | ; And call back to C! |
| 2228 | BLR ip0 |
| 2229 | |
| 2230 | */ |
| 2231 | |
| 2232 | p += emit_mov_addr (p, x0, tpoint); |
| 2233 | p += emit_add (p, x1, sp, immediate_operand (16)); |
| 2234 | |
| 2235 | p += emit_mov_addr (p, ip0, collector); |
| 2236 | p += emit_blr (p, ip0); |
| 2237 | |
| 2238 | /* Release the lock. |
| 2239 | |
| 2240 | MOV x0, #(lockaddr) |
| 2241 | ... |
| 2242 | |
| 2243 | ; This instruction is a normal store with memory ordering |
| 2244 | ; constraints. Thanks to this we do not have to put a data |
| 2245 | ; barrier instruction to make sure all data read and writes are done |
| 2246 | ; before this instruction is executed. Furthermore, this instruction |
| 2247 | ; will trigger an event, letting other threads know they can grab |
| 2248 | ; the lock. |
| 2249 | STLR xzr, [x0] |
| 2250 | |
| 2251 | */ |
| 2252 | p += emit_mov_addr (p, x0, lockaddr); |
| 2253 | p += emit_stlr (p, xzr, x0); |
| 2254 | |
| 2255 | /* Free collecting_t object: |
| 2256 | |
| 2257 | ADD sp, sp, #16 |
| 2258 | |
| 2259 | */ |
| 2260 | p += emit_add (p, sp, sp, immediate_operand (16)); |
| 2261 | |
| 2262 | /* Restore CPSR (NZCV), FPSR and FPCR. And free all special purpose |
| 2263 | registers from the stack. |
| 2264 | |
| 2265 | LDR x2, [sp, #(2 * 16)] |
| 2266 | LDR x1, [sp, #(1 * 16)] |
| 2267 | LDR x0, [sp, #(0 * 16)] |
| 2268 | |
| 2269 | MSR NZCV, x2 |
| 2270 | MSR FPSR, x1 |
| 2271 | MSR FPCR, x0 |
| 2272 | |
| 2273 | ADD sp, sp #(5 * 16) |
| 2274 | |
| 2275 | */ |
| 2276 | p += emit_ldr (p, x2, sp, offset_memory_operand (2 * 16)); |
| 2277 | p += emit_ldr (p, x1, sp, offset_memory_operand (1 * 16)); |
| 2278 | p += emit_ldr (p, x0, sp, offset_memory_operand (0 * 16)); |
| 2279 | p += emit_msr (p, NZCV, x2); |
| 2280 | p += emit_msr (p, FPSR, x1); |
| 2281 | p += emit_msr (p, FPCR, x0); |
| 2282 | |
| 2283 | p += emit_add (p, sp, sp, immediate_operand (5 * 16)); |
| 2284 | |
| 2285 | /* Pop general purpose registers: |
| 2286 | |
| 2287 | LDR x0, [sp] |
| 2288 | ... |
| 2289 | LDR x30, [sp, #(30 * 16)] |
| 2290 | |
| 2291 | ADD sp, sp, #(31 * 16) |
| 2292 | |
| 2293 | */ |
| 2294 | for (i = 0; i <= 30; i += 1) |
| 2295 | p += emit_ldr (p, aarch64_register (i, 1), sp, |
| 2296 | offset_memory_operand (i * 16)); |
| 2297 | p += emit_add (p, sp, sp, immediate_operand (31 * 16)); |
| 2298 | |
| 2299 | /* Pop SIMD&FP registers: |
| 2300 | |
| 2301 | LDP q0, q1, [sp] |
| 2302 | ... |
| 2303 | LDP q30, q31, [sp, #(30 * 16)] |
| 2304 | |
| 2305 | ADD sp, sp, #(32 * 16) |
| 2306 | |
| 2307 | */ |
| 2308 | for (i = 0; i <= 30; i += 2) |
| 2309 | p += emit_ldp_q_offset (p, i, i + 1, sp, i * 16); |
| 2310 | p += emit_add (p, sp, sp, immediate_operand (32 * 16)); |
| 2311 | |
| 2312 | /* Write the code into the inferior memory. */ |
| 2313 | append_insns (&buildaddr, p - buf, buf); |
| 2314 | |
| 2315 | /* Now emit the relocated instruction. */ |
| 2316 | *adjusted_insn_addr = buildaddr; |
| 2317 | target_read_uint32 (tpaddr, &insn); |
| 2318 | |
| 2319 | insn_data.base.insn_addr = tpaddr; |
| 2320 | insn_data.new_addr = buildaddr; |
| 2321 | insn_data.insn_ptr = buf; |
| 2322 | |
| 2323 | aarch64_relocate_instruction (insn, &visitor, |
| 2324 | (struct aarch64_insn_data *) &insn_data); |
| 2325 | |
| 2326 | /* We may not have been able to relocate the instruction. */ |
| 2327 | if (insn_data.insn_ptr == buf) |
| 2328 | { |
| 2329 | sprintf (err, |
| 2330 | "E.Could not relocate instruction from %s to %s.", |
| 2331 | core_addr_to_string_nz (tpaddr), |
| 2332 | core_addr_to_string_nz (buildaddr)); |
| 2333 | return 1; |
| 2334 | } |
| 2335 | else |
| 2336 | append_insns (&buildaddr, insn_data.insn_ptr - buf, buf); |
| 2337 | *adjusted_insn_addr_end = buildaddr; |
| 2338 | |
| 2339 | /* Go back to the start of the buffer. */ |
| 2340 | p = buf; |
| 2341 | |
| 2342 | /* Emit a branch back from the jump pad. */ |
| 2343 | offset = (tpaddr + orig_size - buildaddr); |
| 2344 | if (!can_encode_int32 (offset, 28)) |
| 2345 | { |
| 2346 | sprintf (err, |
| 2347 | "E.Jump back from jump pad too far from tracepoint " |
| 2348 | "(offset 0x%" PRIx64 " cannot be encoded in 28 bits).", |
| 2349 | offset); |
| 2350 | return 1; |
| 2351 | } |
| 2352 | |
| 2353 | p += emit_b (p, 0, offset); |
| 2354 | append_insns (&buildaddr, p - buf, buf); |
| 2355 | |
| 2356 | /* Give the caller a branch instruction into the jump pad. */ |
| 2357 | offset = (*jump_entry - tpaddr); |
| 2358 | if (!can_encode_int32 (offset, 28)) |
| 2359 | { |
| 2360 | sprintf (err, |
| 2361 | "E.Jump pad too far from tracepoint " |
| 2362 | "(offset 0x%" PRIx64 " cannot be encoded in 28 bits).", |
| 2363 | offset); |
| 2364 | return 1; |
| 2365 | } |
| 2366 | |
| 2367 | emit_b ((uint32_t *) jjump_pad_insn, 0, offset); |
| 2368 | *jjump_pad_insn_size = 4; |
| 2369 | |
| 2370 | /* Return the end address of our pad. */ |
| 2371 | *jump_entry = buildaddr; |
| 2372 | |
| 2373 | return 0; |
| 2374 | } |
| 2375 | |
| 2376 | /* Helper function writing LEN instructions from START into |
| 2377 | current_insn_ptr. */ |
| 2378 | |
| 2379 | static void |
| 2380 | emit_ops_insns (const uint32_t *start, int len) |
| 2381 | { |
| 2382 | CORE_ADDR buildaddr = current_insn_ptr; |
| 2383 | |
| 2384 | if (debug_threads) |
| 2385 | debug_printf ("Adding %d instrucions at %s\n", |
| 2386 | len, paddress (buildaddr)); |
| 2387 | |
| 2388 | append_insns (&buildaddr, len, start); |
| 2389 | current_insn_ptr = buildaddr; |
| 2390 | } |
| 2391 | |
| 2392 | /* Pop a register from the stack. */ |
| 2393 | |
| 2394 | static int |
| 2395 | emit_pop (uint32_t *buf, struct aarch64_register rt) |
| 2396 | { |
| 2397 | return emit_ldr (buf, rt, sp, postindex_memory_operand (1 * 16)); |
| 2398 | } |
| 2399 | |
| 2400 | /* Push a register on the stack. */ |
| 2401 | |
| 2402 | static int |
| 2403 | emit_push (uint32_t *buf, struct aarch64_register rt) |
| 2404 | { |
| 2405 | return emit_str (buf, rt, sp, preindex_memory_operand (-1 * 16)); |
| 2406 | } |
| 2407 | |
| 2408 | /* Implementation of emit_ops method "emit_prologue". */ |
| 2409 | |
| 2410 | static void |
| 2411 | aarch64_emit_prologue (void) |
| 2412 | { |
| 2413 | uint32_t buf[16]; |
| 2414 | uint32_t *p = buf; |
| 2415 | |
| 2416 | /* This function emit a prologue for the following function prototype: |
| 2417 | |
| 2418 | enum eval_result_type f (unsigned char *regs, |
| 2419 | ULONGEST *value); |
| 2420 | |
| 2421 | The first argument is a buffer of raw registers. The second |
| 2422 | argument is the result of |
| 2423 | evaluating the expression, which will be set to whatever is on top of |
| 2424 | the stack at the end. |
| 2425 | |
| 2426 | The stack set up by the prologue is as such: |
| 2427 | |
| 2428 | High *------------------------------------------------------* |
| 2429 | | LR | |
| 2430 | | FP | <- FP |
| 2431 | | x1 (ULONGEST *value) | |
| 2432 | | x0 (unsigned char *regs) | |
| 2433 | Low *------------------------------------------------------* |
| 2434 | |
| 2435 | As we are implementing a stack machine, each opcode can expand the |
| 2436 | stack so we never know how far we are from the data saved by this |
| 2437 | prologue. In order to be able refer to value and regs later, we save |
| 2438 | the current stack pointer in the frame pointer. This way, it is not |
| 2439 | clobbered when calling C functions. |
| 2440 | |
| 2441 | Finally, throughout every operation, we are using register x0 as the |
| 2442 | top of the stack, and x1 as a scratch register. */ |
| 2443 | |
| 2444 | p += emit_stp (p, x0, x1, sp, preindex_memory_operand (-2 * 16)); |
| 2445 | p += emit_str (p, lr, sp, offset_memory_operand (3 * 8)); |
| 2446 | p += emit_str (p, fp, sp, offset_memory_operand (2 * 8)); |
| 2447 | |
| 2448 | p += emit_add (p, fp, sp, immediate_operand (2 * 8)); |
| 2449 | |
| 2450 | |
| 2451 | emit_ops_insns (buf, p - buf); |
| 2452 | } |
| 2453 | |
| 2454 | /* Implementation of emit_ops method "emit_epilogue". */ |
| 2455 | |
| 2456 | static void |
| 2457 | aarch64_emit_epilogue (void) |
| 2458 | { |
| 2459 | uint32_t buf[16]; |
| 2460 | uint32_t *p = buf; |
| 2461 | |
| 2462 | /* Store the result of the expression (x0) in *value. */ |
| 2463 | p += emit_sub (p, x1, fp, immediate_operand (1 * 8)); |
| 2464 | p += emit_ldr (p, x1, x1, offset_memory_operand (0)); |
| 2465 | p += emit_str (p, x0, x1, offset_memory_operand (0)); |
| 2466 | |
| 2467 | /* Restore the previous state. */ |
| 2468 | p += emit_add (p, sp, fp, immediate_operand (2 * 8)); |
| 2469 | p += emit_ldp (p, fp, lr, fp, offset_memory_operand (0)); |
| 2470 | |
| 2471 | /* Return expr_eval_no_error. */ |
| 2472 | p += emit_mov (p, x0, immediate_operand (expr_eval_no_error)); |
| 2473 | p += emit_ret (p, lr); |
| 2474 | |
| 2475 | emit_ops_insns (buf, p - buf); |
| 2476 | } |
| 2477 | |
| 2478 | /* Implementation of emit_ops method "emit_add". */ |
| 2479 | |
| 2480 | static void |
| 2481 | aarch64_emit_add (void) |
| 2482 | { |
| 2483 | uint32_t buf[16]; |
| 2484 | uint32_t *p = buf; |
| 2485 | |
| 2486 | p += emit_pop (p, x1); |
| 2487 | p += emit_add (p, x0, x1, register_operand (x0)); |
| 2488 | |
| 2489 | emit_ops_insns (buf, p - buf); |
| 2490 | } |
| 2491 | |
| 2492 | /* Implementation of emit_ops method "emit_sub". */ |
| 2493 | |
| 2494 | static void |
| 2495 | aarch64_emit_sub (void) |
| 2496 | { |
| 2497 | uint32_t buf[16]; |
| 2498 | uint32_t *p = buf; |
| 2499 | |
| 2500 | p += emit_pop (p, x1); |
| 2501 | p += emit_sub (p, x0, x1, register_operand (x0)); |
| 2502 | |
| 2503 | emit_ops_insns (buf, p - buf); |
| 2504 | } |
| 2505 | |
| 2506 | /* Implementation of emit_ops method "emit_mul". */ |
| 2507 | |
| 2508 | static void |
| 2509 | aarch64_emit_mul (void) |
| 2510 | { |
| 2511 | uint32_t buf[16]; |
| 2512 | uint32_t *p = buf; |
| 2513 | |
| 2514 | p += emit_pop (p, x1); |
| 2515 | p += emit_mul (p, x0, x1, x0); |
| 2516 | |
| 2517 | emit_ops_insns (buf, p - buf); |
| 2518 | } |
| 2519 | |
| 2520 | /* Implementation of emit_ops method "emit_lsh". */ |
| 2521 | |
| 2522 | static void |
| 2523 | aarch64_emit_lsh (void) |
| 2524 | { |
| 2525 | uint32_t buf[16]; |
| 2526 | uint32_t *p = buf; |
| 2527 | |
| 2528 | p += emit_pop (p, x1); |
| 2529 | p += emit_lslv (p, x0, x1, x0); |
| 2530 | |
| 2531 | emit_ops_insns (buf, p - buf); |
| 2532 | } |
| 2533 | |
| 2534 | /* Implementation of emit_ops method "emit_rsh_signed". */ |
| 2535 | |
| 2536 | static void |
| 2537 | aarch64_emit_rsh_signed (void) |
| 2538 | { |
| 2539 | uint32_t buf[16]; |
| 2540 | uint32_t *p = buf; |
| 2541 | |
| 2542 | p += emit_pop (p, x1); |
| 2543 | p += emit_asrv (p, x0, x1, x0); |
| 2544 | |
| 2545 | emit_ops_insns (buf, p - buf); |
| 2546 | } |
| 2547 | |
| 2548 | /* Implementation of emit_ops method "emit_rsh_unsigned". */ |
| 2549 | |
| 2550 | static void |
| 2551 | aarch64_emit_rsh_unsigned (void) |
| 2552 | { |
| 2553 | uint32_t buf[16]; |
| 2554 | uint32_t *p = buf; |
| 2555 | |
| 2556 | p += emit_pop (p, x1); |
| 2557 | p += emit_lsrv (p, x0, x1, x0); |
| 2558 | |
| 2559 | emit_ops_insns (buf, p - buf); |
| 2560 | } |
| 2561 | |
| 2562 | /* Implementation of emit_ops method "emit_ext". */ |
| 2563 | |
| 2564 | static void |
| 2565 | aarch64_emit_ext (int arg) |
| 2566 | { |
| 2567 | uint32_t buf[16]; |
| 2568 | uint32_t *p = buf; |
| 2569 | |
| 2570 | p += emit_sbfx (p, x0, x0, 0, arg); |
| 2571 | |
| 2572 | emit_ops_insns (buf, p - buf); |
| 2573 | } |
| 2574 | |
| 2575 | /* Implementation of emit_ops method "emit_log_not". */ |
| 2576 | |
| 2577 | static void |
| 2578 | aarch64_emit_log_not (void) |
| 2579 | { |
| 2580 | uint32_t buf[16]; |
| 2581 | uint32_t *p = buf; |
| 2582 | |
| 2583 | /* If the top of the stack is 0, replace it with 1. Else replace it with |
| 2584 | 0. */ |
| 2585 | |
| 2586 | p += emit_cmp (p, x0, immediate_operand (0)); |
| 2587 | p += emit_cset (p, x0, EQ); |
| 2588 | |
| 2589 | emit_ops_insns (buf, p - buf); |
| 2590 | } |
| 2591 | |
| 2592 | /* Implementation of emit_ops method "emit_bit_and". */ |
| 2593 | |
| 2594 | static void |
| 2595 | aarch64_emit_bit_and (void) |
| 2596 | { |
| 2597 | uint32_t buf[16]; |
| 2598 | uint32_t *p = buf; |
| 2599 | |
| 2600 | p += emit_pop (p, x1); |
| 2601 | p += emit_and (p, x0, x0, x1); |
| 2602 | |
| 2603 | emit_ops_insns (buf, p - buf); |
| 2604 | } |
| 2605 | |
| 2606 | /* Implementation of emit_ops method "emit_bit_or". */ |
| 2607 | |
| 2608 | static void |
| 2609 | aarch64_emit_bit_or (void) |
| 2610 | { |
| 2611 | uint32_t buf[16]; |
| 2612 | uint32_t *p = buf; |
| 2613 | |
| 2614 | p += emit_pop (p, x1); |
| 2615 | p += emit_orr (p, x0, x0, x1); |
| 2616 | |
| 2617 | emit_ops_insns (buf, p - buf); |
| 2618 | } |
| 2619 | |
| 2620 | /* Implementation of emit_ops method "emit_bit_xor". */ |
| 2621 | |
| 2622 | static void |
| 2623 | aarch64_emit_bit_xor (void) |
| 2624 | { |
| 2625 | uint32_t buf[16]; |
| 2626 | uint32_t *p = buf; |
| 2627 | |
| 2628 | p += emit_pop (p, x1); |
| 2629 | p += emit_eor (p, x0, x0, x1); |
| 2630 | |
| 2631 | emit_ops_insns (buf, p - buf); |
| 2632 | } |
| 2633 | |
| 2634 | /* Implementation of emit_ops method "emit_bit_not". */ |
| 2635 | |
| 2636 | static void |
| 2637 | aarch64_emit_bit_not (void) |
| 2638 | { |
| 2639 | uint32_t buf[16]; |
| 2640 | uint32_t *p = buf; |
| 2641 | |
| 2642 | p += emit_mvn (p, x0, x0); |
| 2643 | |
| 2644 | emit_ops_insns (buf, p - buf); |
| 2645 | } |
| 2646 | |
| 2647 | /* Implementation of emit_ops method "emit_equal". */ |
| 2648 | |
| 2649 | static void |
| 2650 | aarch64_emit_equal (void) |
| 2651 | { |
| 2652 | uint32_t buf[16]; |
| 2653 | uint32_t *p = buf; |
| 2654 | |
| 2655 | p += emit_pop (p, x1); |
| 2656 | p += emit_cmp (p, x0, register_operand (x1)); |
| 2657 | p += emit_cset (p, x0, EQ); |
| 2658 | |
| 2659 | emit_ops_insns (buf, p - buf); |
| 2660 | } |
| 2661 | |
| 2662 | /* Implementation of emit_ops method "emit_less_signed". */ |
| 2663 | |
| 2664 | static void |
| 2665 | aarch64_emit_less_signed (void) |
| 2666 | { |
| 2667 | uint32_t buf[16]; |
| 2668 | uint32_t *p = buf; |
| 2669 | |
| 2670 | p += emit_pop (p, x1); |
| 2671 | p += emit_cmp (p, x1, register_operand (x0)); |
| 2672 | p += emit_cset (p, x0, LT); |
| 2673 | |
| 2674 | emit_ops_insns (buf, p - buf); |
| 2675 | } |
| 2676 | |
| 2677 | /* Implementation of emit_ops method "emit_less_unsigned". */ |
| 2678 | |
| 2679 | static void |
| 2680 | aarch64_emit_less_unsigned (void) |
| 2681 | { |
| 2682 | uint32_t buf[16]; |
| 2683 | uint32_t *p = buf; |
| 2684 | |
| 2685 | p += emit_pop (p, x1); |
| 2686 | p += emit_cmp (p, x1, register_operand (x0)); |
| 2687 | p += emit_cset (p, x0, LO); |
| 2688 | |
| 2689 | emit_ops_insns (buf, p - buf); |
| 2690 | } |
| 2691 | |
| 2692 | /* Implementation of emit_ops method "emit_ref". */ |
| 2693 | |
| 2694 | static void |
| 2695 | aarch64_emit_ref (int size) |
| 2696 | { |
| 2697 | uint32_t buf[16]; |
| 2698 | uint32_t *p = buf; |
| 2699 | |
| 2700 | switch (size) |
| 2701 | { |
| 2702 | case 1: |
| 2703 | p += emit_ldrb (p, w0, x0, offset_memory_operand (0)); |
| 2704 | break; |
| 2705 | case 2: |
| 2706 | p += emit_ldrh (p, w0, x0, offset_memory_operand (0)); |
| 2707 | break; |
| 2708 | case 4: |
| 2709 | p += emit_ldr (p, w0, x0, offset_memory_operand (0)); |
| 2710 | break; |
| 2711 | case 8: |
| 2712 | p += emit_ldr (p, x0, x0, offset_memory_operand (0)); |
| 2713 | break; |
| 2714 | default: |
| 2715 | /* Unknown size, bail on compilation. */ |
| 2716 | emit_error = 1; |
| 2717 | break; |
| 2718 | } |
| 2719 | |
| 2720 | emit_ops_insns (buf, p - buf); |
| 2721 | } |
| 2722 | |
| 2723 | /* Implementation of emit_ops method "emit_if_goto". */ |
| 2724 | |
| 2725 | static void |
| 2726 | aarch64_emit_if_goto (int *offset_p, int *size_p) |
| 2727 | { |
| 2728 | uint32_t buf[16]; |
| 2729 | uint32_t *p = buf; |
| 2730 | |
| 2731 | /* The Z flag is set or cleared here. */ |
| 2732 | p += emit_cmp (p, x0, immediate_operand (0)); |
| 2733 | /* This instruction must not change the Z flag. */ |
| 2734 | p += emit_pop (p, x0); |
| 2735 | /* Branch over the next instruction if x0 == 0. */ |
| 2736 | p += emit_bcond (p, EQ, 8); |
| 2737 | |
| 2738 | /* The NOP instruction will be patched with an unconditional branch. */ |
| 2739 | if (offset_p) |
| 2740 | *offset_p = (p - buf) * 4; |
| 2741 | if (size_p) |
| 2742 | *size_p = 4; |
| 2743 | p += emit_nop (p); |
| 2744 | |
| 2745 | emit_ops_insns (buf, p - buf); |
| 2746 | } |
| 2747 | |
| 2748 | /* Implementation of emit_ops method "emit_goto". */ |
| 2749 | |
| 2750 | static void |
| 2751 | aarch64_emit_goto (int *offset_p, int *size_p) |
| 2752 | { |
| 2753 | uint32_t buf[16]; |
| 2754 | uint32_t *p = buf; |
| 2755 | |
| 2756 | /* The NOP instruction will be patched with an unconditional branch. */ |
| 2757 | if (offset_p) |
| 2758 | *offset_p = 0; |
| 2759 | if (size_p) |
| 2760 | *size_p = 4; |
| 2761 | p += emit_nop (p); |
| 2762 | |
| 2763 | emit_ops_insns (buf, p - buf); |
| 2764 | } |
| 2765 | |
| 2766 | /* Implementation of emit_ops method "write_goto_address". */ |
| 2767 | |
| 2768 | static void |
| 2769 | aarch64_write_goto_address (CORE_ADDR from, CORE_ADDR to, int size) |
| 2770 | { |
| 2771 | uint32_t insn; |
| 2772 | |
| 2773 | emit_b (&insn, 0, to - from); |
| 2774 | append_insns (&from, 1, &insn); |
| 2775 | } |
| 2776 | |
| 2777 | /* Implementation of emit_ops method "emit_const". */ |
| 2778 | |
| 2779 | static void |
| 2780 | aarch64_emit_const (LONGEST num) |
| 2781 | { |
| 2782 | uint32_t buf[16]; |
| 2783 | uint32_t *p = buf; |
| 2784 | |
| 2785 | p += emit_mov_addr (p, x0, num); |
| 2786 | |
| 2787 | emit_ops_insns (buf, p - buf); |
| 2788 | } |
| 2789 | |
| 2790 | /* Implementation of emit_ops method "emit_call". */ |
| 2791 | |
| 2792 | static void |
| 2793 | aarch64_emit_call (CORE_ADDR fn) |
| 2794 | { |
| 2795 | uint32_t buf[16]; |
| 2796 | uint32_t *p = buf; |
| 2797 | |
| 2798 | p += emit_mov_addr (p, ip0, fn); |
| 2799 | p += emit_blr (p, ip0); |
| 2800 | |
| 2801 | emit_ops_insns (buf, p - buf); |
| 2802 | } |
| 2803 | |
| 2804 | /* Implementation of emit_ops method "emit_reg". */ |
| 2805 | |
| 2806 | static void |
| 2807 | aarch64_emit_reg (int reg) |
| 2808 | { |
| 2809 | uint32_t buf[16]; |
| 2810 | uint32_t *p = buf; |
| 2811 | |
| 2812 | /* Set x0 to unsigned char *regs. */ |
| 2813 | p += emit_sub (p, x0, fp, immediate_operand (2 * 8)); |
| 2814 | p += emit_ldr (p, x0, x0, offset_memory_operand (0)); |
| 2815 | p += emit_mov (p, x1, immediate_operand (reg)); |
| 2816 | |
| 2817 | emit_ops_insns (buf, p - buf); |
| 2818 | |
| 2819 | aarch64_emit_call (get_raw_reg_func_addr ()); |
| 2820 | } |
| 2821 | |
| 2822 | /* Implementation of emit_ops method "emit_pop". */ |
| 2823 | |
| 2824 | static void |
| 2825 | aarch64_emit_pop (void) |
| 2826 | { |
| 2827 | uint32_t buf[16]; |
| 2828 | uint32_t *p = buf; |
| 2829 | |
| 2830 | p += emit_pop (p, x0); |
| 2831 | |
| 2832 | emit_ops_insns (buf, p - buf); |
| 2833 | } |
| 2834 | |
| 2835 | /* Implementation of emit_ops method "emit_stack_flush". */ |
| 2836 | |
| 2837 | static void |
| 2838 | aarch64_emit_stack_flush (void) |
| 2839 | { |
| 2840 | uint32_t buf[16]; |
| 2841 | uint32_t *p = buf; |
| 2842 | |
| 2843 | p += emit_push (p, x0); |
| 2844 | |
| 2845 | emit_ops_insns (buf, p - buf); |
| 2846 | } |
| 2847 | |
| 2848 | /* Implementation of emit_ops method "emit_zero_ext". */ |
| 2849 | |
| 2850 | static void |
| 2851 | aarch64_emit_zero_ext (int arg) |
| 2852 | { |
| 2853 | uint32_t buf[16]; |
| 2854 | uint32_t *p = buf; |
| 2855 | |
| 2856 | p += emit_ubfx (p, x0, x0, 0, arg); |
| 2857 | |
| 2858 | emit_ops_insns (buf, p - buf); |
| 2859 | } |
| 2860 | |
| 2861 | /* Implementation of emit_ops method "emit_swap". */ |
| 2862 | |
| 2863 | static void |
| 2864 | aarch64_emit_swap (void) |
| 2865 | { |
| 2866 | uint32_t buf[16]; |
| 2867 | uint32_t *p = buf; |
| 2868 | |
| 2869 | p += emit_ldr (p, x1, sp, offset_memory_operand (0 * 16)); |
| 2870 | p += emit_str (p, x0, sp, offset_memory_operand (0 * 16)); |
| 2871 | p += emit_mov (p, x0, register_operand (x1)); |
| 2872 | |
| 2873 | emit_ops_insns (buf, p - buf); |
| 2874 | } |
| 2875 | |
| 2876 | /* Implementation of emit_ops method "emit_stack_adjust". */ |
| 2877 | |
| 2878 | static void |
| 2879 | aarch64_emit_stack_adjust (int n) |
| 2880 | { |
| 2881 | /* This is not needed with our design. */ |
| 2882 | uint32_t buf[16]; |
| 2883 | uint32_t *p = buf; |
| 2884 | |
| 2885 | p += emit_add (p, sp, sp, immediate_operand (n * 16)); |
| 2886 | |
| 2887 | emit_ops_insns (buf, p - buf); |
| 2888 | } |
| 2889 | |
| 2890 | /* Implementation of emit_ops method "emit_int_call_1". */ |
| 2891 | |
| 2892 | static void |
| 2893 | aarch64_emit_int_call_1 (CORE_ADDR fn, int arg1) |
| 2894 | { |
| 2895 | uint32_t buf[16]; |
| 2896 | uint32_t *p = buf; |
| 2897 | |
| 2898 | p += emit_mov (p, x0, immediate_operand (arg1)); |
| 2899 | |
| 2900 | emit_ops_insns (buf, p - buf); |
| 2901 | |
| 2902 | aarch64_emit_call (fn); |
| 2903 | } |
| 2904 | |
| 2905 | /* Implementation of emit_ops method "emit_void_call_2". */ |
| 2906 | |
| 2907 | static void |
| 2908 | aarch64_emit_void_call_2 (CORE_ADDR fn, int arg1) |
| 2909 | { |
| 2910 | uint32_t buf[16]; |
| 2911 | uint32_t *p = buf; |
| 2912 | |
| 2913 | /* Push x0 on the stack. */ |
| 2914 | aarch64_emit_stack_flush (); |
| 2915 | |
| 2916 | /* Setup arguments for the function call: |
| 2917 | |
| 2918 | x0: arg1 |
| 2919 | x1: top of the stack |
| 2920 | |
| 2921 | MOV x1, x0 |
| 2922 | MOV x0, #arg1 */ |
| 2923 | |
| 2924 | p += emit_mov (p, x1, register_operand (x0)); |
| 2925 | p += emit_mov (p, x0, immediate_operand (arg1)); |
| 2926 | |
| 2927 | emit_ops_insns (buf, p - buf); |
| 2928 | |
| 2929 | aarch64_emit_call (fn); |
| 2930 | |
| 2931 | /* Restore x0. */ |
| 2932 | aarch64_emit_pop (); |
| 2933 | } |
| 2934 | |
| 2935 | /* Implementation of emit_ops method "emit_eq_goto". */ |
| 2936 | |
| 2937 | static void |
| 2938 | aarch64_emit_eq_goto (int *offset_p, int *size_p) |
| 2939 | { |
| 2940 | uint32_t buf[16]; |
| 2941 | uint32_t *p = buf; |
| 2942 | |
| 2943 | p += emit_pop (p, x1); |
| 2944 | p += emit_cmp (p, x1, register_operand (x0)); |
| 2945 | /* Branch over the next instruction if x0 != x1. */ |
| 2946 | p += emit_bcond (p, NE, 8); |
| 2947 | /* The NOP instruction will be patched with an unconditional branch. */ |
| 2948 | if (offset_p) |
| 2949 | *offset_p = (p - buf) * 4; |
| 2950 | if (size_p) |
| 2951 | *size_p = 4; |
| 2952 | p += emit_nop (p); |
| 2953 | |
| 2954 | emit_ops_insns (buf, p - buf); |
| 2955 | } |
| 2956 | |
| 2957 | /* Implementation of emit_ops method "emit_ne_goto". */ |
| 2958 | |
| 2959 | static void |
| 2960 | aarch64_emit_ne_goto (int *offset_p, int *size_p) |
| 2961 | { |
| 2962 | uint32_t buf[16]; |
| 2963 | uint32_t *p = buf; |
| 2964 | |
| 2965 | p += emit_pop (p, x1); |
| 2966 | p += emit_cmp (p, x1, register_operand (x0)); |
| 2967 | /* Branch over the next instruction if x0 == x1. */ |
| 2968 | p += emit_bcond (p, EQ, 8); |
| 2969 | /* The NOP instruction will be patched with an unconditional branch. */ |
| 2970 | if (offset_p) |
| 2971 | *offset_p = (p - buf) * 4; |
| 2972 | if (size_p) |
| 2973 | *size_p = 4; |
| 2974 | p += emit_nop (p); |
| 2975 | |
| 2976 | emit_ops_insns (buf, p - buf); |
| 2977 | } |
| 2978 | |
| 2979 | /* Implementation of emit_ops method "emit_lt_goto". */ |
| 2980 | |
| 2981 | static void |
| 2982 | aarch64_emit_lt_goto (int *offset_p, int *size_p) |
| 2983 | { |
| 2984 | uint32_t buf[16]; |
| 2985 | uint32_t *p = buf; |
| 2986 | |
| 2987 | p += emit_pop (p, x1); |
| 2988 | p += emit_cmp (p, x1, register_operand (x0)); |
| 2989 | /* Branch over the next instruction if x0 >= x1. */ |
| 2990 | p += emit_bcond (p, GE, 8); |
| 2991 | /* The NOP instruction will be patched with an unconditional branch. */ |
| 2992 | if (offset_p) |
| 2993 | *offset_p = (p - buf) * 4; |
| 2994 | if (size_p) |
| 2995 | *size_p = 4; |
| 2996 | p += emit_nop (p); |
| 2997 | |
| 2998 | emit_ops_insns (buf, p - buf); |
| 2999 | } |
| 3000 | |
| 3001 | /* Implementation of emit_ops method "emit_le_goto". */ |
| 3002 | |
| 3003 | static void |
| 3004 | aarch64_emit_le_goto (int *offset_p, int *size_p) |
| 3005 | { |
| 3006 | uint32_t buf[16]; |
| 3007 | uint32_t *p = buf; |
| 3008 | |
| 3009 | p += emit_pop (p, x1); |
| 3010 | p += emit_cmp (p, x1, register_operand (x0)); |
| 3011 | /* Branch over the next instruction if x0 > x1. */ |
| 3012 | p += emit_bcond (p, GT, 8); |
| 3013 | /* The NOP instruction will be patched with an unconditional branch. */ |
| 3014 | if (offset_p) |
| 3015 | *offset_p = (p - buf) * 4; |
| 3016 | if (size_p) |
| 3017 | *size_p = 4; |
| 3018 | p += emit_nop (p); |
| 3019 | |
| 3020 | emit_ops_insns (buf, p - buf); |
| 3021 | } |
| 3022 | |
| 3023 | /* Implementation of emit_ops method "emit_gt_goto". */ |
| 3024 | |
| 3025 | static void |
| 3026 | aarch64_emit_gt_goto (int *offset_p, int *size_p) |
| 3027 | { |
| 3028 | uint32_t buf[16]; |
| 3029 | uint32_t *p = buf; |
| 3030 | |
| 3031 | p += emit_pop (p, x1); |
| 3032 | p += emit_cmp (p, x1, register_operand (x0)); |
| 3033 | /* Branch over the next instruction if x0 <= x1. */ |
| 3034 | p += emit_bcond (p, LE, 8); |
| 3035 | /* The NOP instruction will be patched with an unconditional branch. */ |
| 3036 | if (offset_p) |
| 3037 | *offset_p = (p - buf) * 4; |
| 3038 | if (size_p) |
| 3039 | *size_p = 4; |
| 3040 | p += emit_nop (p); |
| 3041 | |
| 3042 | emit_ops_insns (buf, p - buf); |
| 3043 | } |
| 3044 | |
| 3045 | /* Implementation of emit_ops method "emit_ge_got". */ |
| 3046 | |
| 3047 | static void |
| 3048 | aarch64_emit_ge_got (int *offset_p, int *size_p) |
| 3049 | { |
| 3050 | uint32_t buf[16]; |
| 3051 | uint32_t *p = buf; |
| 3052 | |
| 3053 | p += emit_pop (p, x1); |
| 3054 | p += emit_cmp (p, x1, register_operand (x0)); |
| 3055 | /* Branch over the next instruction if x0 <= x1. */ |
| 3056 | p += emit_bcond (p, LT, 8); |
| 3057 | /* The NOP instruction will be patched with an unconditional branch. */ |
| 3058 | if (offset_p) |
| 3059 | *offset_p = (p - buf) * 4; |
| 3060 | if (size_p) |
| 3061 | *size_p = 4; |
| 3062 | p += emit_nop (p); |
| 3063 | |
| 3064 | emit_ops_insns (buf, p - buf); |
| 3065 | } |
| 3066 | |
| 3067 | static struct emit_ops aarch64_emit_ops_impl = |
| 3068 | { |
| 3069 | aarch64_emit_prologue, |
| 3070 | aarch64_emit_epilogue, |
| 3071 | aarch64_emit_add, |
| 3072 | aarch64_emit_sub, |
| 3073 | aarch64_emit_mul, |
| 3074 | aarch64_emit_lsh, |
| 3075 | aarch64_emit_rsh_signed, |
| 3076 | aarch64_emit_rsh_unsigned, |
| 3077 | aarch64_emit_ext, |
| 3078 | aarch64_emit_log_not, |
| 3079 | aarch64_emit_bit_and, |
| 3080 | aarch64_emit_bit_or, |
| 3081 | aarch64_emit_bit_xor, |
| 3082 | aarch64_emit_bit_not, |
| 3083 | aarch64_emit_equal, |
| 3084 | aarch64_emit_less_signed, |
| 3085 | aarch64_emit_less_unsigned, |
| 3086 | aarch64_emit_ref, |
| 3087 | aarch64_emit_if_goto, |
| 3088 | aarch64_emit_goto, |
| 3089 | aarch64_write_goto_address, |
| 3090 | aarch64_emit_const, |
| 3091 | aarch64_emit_call, |
| 3092 | aarch64_emit_reg, |
| 3093 | aarch64_emit_pop, |
| 3094 | aarch64_emit_stack_flush, |
| 3095 | aarch64_emit_zero_ext, |
| 3096 | aarch64_emit_swap, |
| 3097 | aarch64_emit_stack_adjust, |
| 3098 | aarch64_emit_int_call_1, |
| 3099 | aarch64_emit_void_call_2, |
| 3100 | aarch64_emit_eq_goto, |
| 3101 | aarch64_emit_ne_goto, |
| 3102 | aarch64_emit_lt_goto, |
| 3103 | aarch64_emit_le_goto, |
| 3104 | aarch64_emit_gt_goto, |
| 3105 | aarch64_emit_ge_got, |
| 3106 | }; |
| 3107 | |
| 3108 | /* Implementation of target ops method "emit_ops". */ |
| 3109 | |
| 3110 | emit_ops * |
| 3111 | aarch64_target::emit_ops () |
| 3112 | { |
| 3113 | return &aarch64_emit_ops_impl; |
| 3114 | } |
| 3115 | |
| 3116 | /* Implementation of target ops method |
| 3117 | "get_min_fast_tracepoint_insn_len". */ |
| 3118 | |
| 3119 | int |
| 3120 | aarch64_target::get_min_fast_tracepoint_insn_len () |
| 3121 | { |
| 3122 | return 4; |
| 3123 | } |
| 3124 | |
| 3125 | /* Implementation of linux target ops method "low_supports_range_stepping". */ |
| 3126 | |
| 3127 | bool |
| 3128 | aarch64_target::low_supports_range_stepping () |
| 3129 | { |
| 3130 | return true; |
| 3131 | } |
| 3132 | |
| 3133 | /* Implementation of target ops method "sw_breakpoint_from_kind". */ |
| 3134 | |
| 3135 | const gdb_byte * |
| 3136 | aarch64_target::sw_breakpoint_from_kind (int kind, int *size) |
| 3137 | { |
| 3138 | if (is_64bit_tdesc ()) |
| 3139 | { |
| 3140 | *size = aarch64_breakpoint_len; |
| 3141 | return aarch64_breakpoint; |
| 3142 | } |
| 3143 | else |
| 3144 | return arm_sw_breakpoint_from_kind (kind, size); |
| 3145 | } |
| 3146 | |
| 3147 | /* Implementation of target ops method "breakpoint_kind_from_pc". */ |
| 3148 | |
| 3149 | int |
| 3150 | aarch64_target::breakpoint_kind_from_pc (CORE_ADDR *pcptr) |
| 3151 | { |
| 3152 | if (is_64bit_tdesc ()) |
| 3153 | return aarch64_breakpoint_len; |
| 3154 | else |
| 3155 | return arm_breakpoint_kind_from_pc (pcptr); |
| 3156 | } |
| 3157 | |
| 3158 | /* Implementation of the target ops method |
| 3159 | "breakpoint_kind_from_current_state". */ |
| 3160 | |
| 3161 | int |
| 3162 | aarch64_target::breakpoint_kind_from_current_state (CORE_ADDR *pcptr) |
| 3163 | { |
| 3164 | if (is_64bit_tdesc ()) |
| 3165 | return aarch64_breakpoint_len; |
| 3166 | else |
| 3167 | return arm_breakpoint_kind_from_current_state (pcptr); |
| 3168 | } |
| 3169 | |
| 3170 | /* The linux target ops object. */ |
| 3171 | |
| 3172 | linux_process_target *the_linux_target = &the_aarch64_target; |
| 3173 | |
| 3174 | void |
| 3175 | initialize_low_arch (void) |
| 3176 | { |
| 3177 | initialize_low_arch_aarch32 (); |
| 3178 | |
| 3179 | initialize_regsets_info (&aarch64_regsets_info); |
| 3180 | initialize_regsets_info (&aarch64_sve_regsets_info); |
| 3181 | } |