1 /* This is the Linux kernel elf-loading code, ported into user space */ 2 #include "qemu/osdep.h" 3 #include <sys/param.h> 4 5 #include <sys/resource.h> 6 #include <sys/shm.h> 7 8 #include "qemu.h" 9 #include "user-internals.h" 10 #include "signal-common.h" 11 #include "loader.h" 12 #include "user-mmap.h" 13 #include "disas/disas.h" 14 #include "qemu/bitops.h" 15 #include "qemu/path.h" 16 #include "qemu/queue.h" 17 #include "qemu/guest-random.h" 18 #include "qemu/units.h" 19 #include "qemu/selfmap.h" 20 #include "qapi/error.h" 21 #include "target_signal.h" 22 23 #ifdef _ARCH_PPC64 24 #undef ARCH_DLINFO 25 #undef ELF_PLATFORM 26 #undef ELF_HWCAP 27 #undef ELF_HWCAP2 28 #undef ELF_CLASS 29 #undef ELF_DATA 30 #undef ELF_ARCH 31 #endif 32 33 #define ELF_OSABI ELFOSABI_SYSV 34 35 /* from personality.h */ 36 37 /* 38 * Flags for bug emulation. 39 * 40 * These occupy the top three bytes. 41 */ 42 enum { 43 ADDR_NO_RANDOMIZE = 0x0040000, /* disable randomization of VA space */ 44 FDPIC_FUNCPTRS = 0x0080000, /* userspace function ptrs point to 45 descriptors (signal handling) */ 46 MMAP_PAGE_ZERO = 0x0100000, 47 ADDR_COMPAT_LAYOUT = 0x0200000, 48 READ_IMPLIES_EXEC = 0x0400000, 49 ADDR_LIMIT_32BIT = 0x0800000, 50 SHORT_INODE = 0x1000000, 51 WHOLE_SECONDS = 0x2000000, 52 STICKY_TIMEOUTS = 0x4000000, 53 ADDR_LIMIT_3GB = 0x8000000, 54 }; 55 56 /* 57 * Personality types. 58 * 59 * These go in the low byte. Avoid using the top bit, it will 60 * conflict with error returns. 61 */ 62 enum { 63 PER_LINUX = 0x0000, 64 PER_LINUX_32BIT = 0x0000 | ADDR_LIMIT_32BIT, 65 PER_LINUX_FDPIC = 0x0000 | FDPIC_FUNCPTRS, 66 PER_SVR4 = 0x0001 | STICKY_TIMEOUTS | MMAP_PAGE_ZERO, 67 PER_SVR3 = 0x0002 | STICKY_TIMEOUTS | SHORT_INODE, 68 PER_SCOSVR3 = 0x0003 | STICKY_TIMEOUTS | WHOLE_SECONDS | SHORT_INODE, 69 PER_OSR5 = 0x0003 | STICKY_TIMEOUTS | WHOLE_SECONDS, 70 PER_WYSEV386 = 0x0004 | STICKY_TIMEOUTS | SHORT_INODE, 71 PER_ISCR4 = 0x0005 | STICKY_TIMEOUTS, 72 PER_BSD = 0x0006, 73 PER_SUNOS = 0x0006 | STICKY_TIMEOUTS, 74 PER_XENIX = 0x0007 | STICKY_TIMEOUTS | SHORT_INODE, 75 PER_LINUX32 = 0x0008, 76 PER_LINUX32_3GB = 0x0008 | ADDR_LIMIT_3GB, 77 PER_IRIX32 = 0x0009 | STICKY_TIMEOUTS,/* IRIX5 32-bit */ 78 PER_IRIXN32 = 0x000a | STICKY_TIMEOUTS,/* IRIX6 new 32-bit */ 79 PER_IRIX64 = 0x000b | STICKY_TIMEOUTS,/* IRIX6 64-bit */ 80 PER_RISCOS = 0x000c, 81 PER_SOLARIS = 0x000d | STICKY_TIMEOUTS, 82 PER_UW7 = 0x000e | STICKY_TIMEOUTS | MMAP_PAGE_ZERO, 83 PER_OSF4 = 0x000f, /* OSF/1 v4 */ 84 PER_HPUX = 0x0010, 85 PER_MASK = 0x00ff, 86 }; 87 88 /* 89 * Return the base personality without flags. 90 */ 91 #define personality(pers) (pers & PER_MASK) 92 93 int info_is_fdpic(struct image_info *info) 94 { 95 return info->personality == PER_LINUX_FDPIC; 96 } 97 98 /* this flag is uneffective under linux too, should be deleted */ 99 #ifndef MAP_DENYWRITE 100 #define MAP_DENYWRITE 0 101 #endif 102 103 /* should probably go in elf.h */ 104 #ifndef ELIBBAD 105 #define ELIBBAD 80 106 #endif 107 108 #if TARGET_BIG_ENDIAN 109 #define ELF_DATA ELFDATA2MSB 110 #else 111 #define ELF_DATA ELFDATA2LSB 112 #endif 113 114 #ifdef TARGET_ABI_MIPSN32 115 typedef abi_ullong target_elf_greg_t; 116 #define tswapreg(ptr) tswap64(ptr) 117 #else 118 typedef abi_ulong target_elf_greg_t; 119 #define tswapreg(ptr) tswapal(ptr) 120 #endif 121 122 #ifdef USE_UID16 123 typedef abi_ushort target_uid_t; 124 typedef abi_ushort target_gid_t; 125 #else 126 typedef abi_uint target_uid_t; 127 typedef abi_uint target_gid_t; 128 #endif 129 typedef abi_int target_pid_t; 130 131 #ifdef TARGET_I386 132 133 #define ELF_HWCAP get_elf_hwcap() 134 135 static uint32_t get_elf_hwcap(void) 136 { 137 X86CPU *cpu = X86_CPU(thread_cpu); 138 139 return cpu->env.features[FEAT_1_EDX]; 140 } 141 142 #ifdef TARGET_X86_64 143 #define ELF_START_MMAP 0x2aaaaab000ULL 144 145 #define ELF_CLASS ELFCLASS64 146 #define ELF_ARCH EM_X86_64 147 148 #define ELF_PLATFORM "x86_64" 149 150 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop) 151 { 152 regs->rax = 0; 153 regs->rsp = infop->start_stack; 154 regs->rip = infop->entry; 155 } 156 157 #define ELF_NREG 27 158 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 159 160 /* 161 * Note that ELF_NREG should be 29 as there should be place for 162 * TRAPNO and ERR "registers" as well but linux doesn't dump 163 * those. 164 * 165 * See linux kernel: arch/x86/include/asm/elf.h 166 */ 167 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUX86State *env) 168 { 169 (*regs)[0] = tswapreg(env->regs[15]); 170 (*regs)[1] = tswapreg(env->regs[14]); 171 (*regs)[2] = tswapreg(env->regs[13]); 172 (*regs)[3] = tswapreg(env->regs[12]); 173 (*regs)[4] = tswapreg(env->regs[R_EBP]); 174 (*regs)[5] = tswapreg(env->regs[R_EBX]); 175 (*regs)[6] = tswapreg(env->regs[11]); 176 (*regs)[7] = tswapreg(env->regs[10]); 177 (*regs)[8] = tswapreg(env->regs[9]); 178 (*regs)[9] = tswapreg(env->regs[8]); 179 (*regs)[10] = tswapreg(env->regs[R_EAX]); 180 (*regs)[11] = tswapreg(env->regs[R_ECX]); 181 (*regs)[12] = tswapreg(env->regs[R_EDX]); 182 (*regs)[13] = tswapreg(env->regs[R_ESI]); 183 (*regs)[14] = tswapreg(env->regs[R_EDI]); 184 (*regs)[15] = tswapreg(env->regs[R_EAX]); /* XXX */ 185 (*regs)[16] = tswapreg(env->eip); 186 (*regs)[17] = tswapreg(env->segs[R_CS].selector & 0xffff); 187 (*regs)[18] = tswapreg(env->eflags); 188 (*regs)[19] = tswapreg(env->regs[R_ESP]); 189 (*regs)[20] = tswapreg(env->segs[R_SS].selector & 0xffff); 190 (*regs)[21] = tswapreg(env->segs[R_FS].selector & 0xffff); 191 (*regs)[22] = tswapreg(env->segs[R_GS].selector & 0xffff); 192 (*regs)[23] = tswapreg(env->segs[R_DS].selector & 0xffff); 193 (*regs)[24] = tswapreg(env->segs[R_ES].selector & 0xffff); 194 (*regs)[25] = tswapreg(env->segs[R_FS].selector & 0xffff); 195 (*regs)[26] = tswapreg(env->segs[R_GS].selector & 0xffff); 196 } 197 198 #else 199 200 #define ELF_START_MMAP 0x80000000 201 202 /* 203 * This is used to ensure we don't load something for the wrong architecture. 204 */ 205 #define elf_check_arch(x) ( ((x) == EM_386) || ((x) == EM_486) ) 206 207 /* 208 * These are used to set parameters in the core dumps. 209 */ 210 #define ELF_CLASS ELFCLASS32 211 #define ELF_ARCH EM_386 212 213 #define ELF_PLATFORM get_elf_platform() 214 215 static const char *get_elf_platform(void) 216 { 217 static char elf_platform[] = "i386"; 218 int family = object_property_get_int(OBJECT(thread_cpu), "family", NULL); 219 if (family > 6) { 220 family = 6; 221 } 222 if (family >= 3) { 223 elf_platform[1] = '0' + family; 224 } 225 return elf_platform; 226 } 227 228 static inline void init_thread(struct target_pt_regs *regs, 229 struct image_info *infop) 230 { 231 regs->esp = infop->start_stack; 232 regs->eip = infop->entry; 233 234 /* SVR4/i386 ABI (pages 3-31, 3-32) says that when the program 235 starts %edx contains a pointer to a function which might be 236 registered using `atexit'. This provides a mean for the 237 dynamic linker to call DT_FINI functions for shared libraries 238 that have been loaded before the code runs. 239 240 A value of 0 tells we have no such handler. */ 241 regs->edx = 0; 242 } 243 244 #define ELF_NREG 17 245 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 246 247 /* 248 * Note that ELF_NREG should be 19 as there should be place for 249 * TRAPNO and ERR "registers" as well but linux doesn't dump 250 * those. 251 * 252 * See linux kernel: arch/x86/include/asm/elf.h 253 */ 254 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUX86State *env) 255 { 256 (*regs)[0] = tswapreg(env->regs[R_EBX]); 257 (*regs)[1] = tswapreg(env->regs[R_ECX]); 258 (*regs)[2] = tswapreg(env->regs[R_EDX]); 259 (*regs)[3] = tswapreg(env->regs[R_ESI]); 260 (*regs)[4] = tswapreg(env->regs[R_EDI]); 261 (*regs)[5] = tswapreg(env->regs[R_EBP]); 262 (*regs)[6] = tswapreg(env->regs[R_EAX]); 263 (*regs)[7] = tswapreg(env->segs[R_DS].selector & 0xffff); 264 (*regs)[8] = tswapreg(env->segs[R_ES].selector & 0xffff); 265 (*regs)[9] = tswapreg(env->segs[R_FS].selector & 0xffff); 266 (*regs)[10] = tswapreg(env->segs[R_GS].selector & 0xffff); 267 (*regs)[11] = tswapreg(env->regs[R_EAX]); /* XXX */ 268 (*regs)[12] = tswapreg(env->eip); 269 (*regs)[13] = tswapreg(env->segs[R_CS].selector & 0xffff); 270 (*regs)[14] = tswapreg(env->eflags); 271 (*regs)[15] = tswapreg(env->regs[R_ESP]); 272 (*regs)[16] = tswapreg(env->segs[R_SS].selector & 0xffff); 273 } 274 #endif 275 276 #define USE_ELF_CORE_DUMP 277 #define ELF_EXEC_PAGESIZE 4096 278 279 #endif 280 281 #ifdef TARGET_ARM 282 283 #ifndef TARGET_AARCH64 284 /* 32 bit ARM definitions */ 285 286 #define ELF_START_MMAP 0x80000000 287 288 #define ELF_ARCH EM_ARM 289 #define ELF_CLASS ELFCLASS32 290 291 static inline void init_thread(struct target_pt_regs *regs, 292 struct image_info *infop) 293 { 294 abi_long stack = infop->start_stack; 295 memset(regs, 0, sizeof(*regs)); 296 297 regs->uregs[16] = ARM_CPU_MODE_USR; 298 if (infop->entry & 1) { 299 regs->uregs[16] |= CPSR_T; 300 } 301 regs->uregs[15] = infop->entry & 0xfffffffe; 302 regs->uregs[13] = infop->start_stack; 303 /* FIXME - what to for failure of get_user()? */ 304 get_user_ual(regs->uregs[2], stack + 8); /* envp */ 305 get_user_ual(regs->uregs[1], stack + 4); /* envp */ 306 /* XXX: it seems that r0 is zeroed after ! */ 307 regs->uregs[0] = 0; 308 /* For uClinux PIC binaries. */ 309 /* XXX: Linux does this only on ARM with no MMU (do we care ?) */ 310 regs->uregs[10] = infop->start_data; 311 312 /* Support ARM FDPIC. */ 313 if (info_is_fdpic(infop)) { 314 /* As described in the ABI document, r7 points to the loadmap info 315 * prepared by the kernel. If an interpreter is needed, r8 points 316 * to the interpreter loadmap and r9 points to the interpreter 317 * PT_DYNAMIC info. If no interpreter is needed, r8 is zero, and 318 * r9 points to the main program PT_DYNAMIC info. 319 */ 320 regs->uregs[7] = infop->loadmap_addr; 321 if (infop->interpreter_loadmap_addr) { 322 /* Executable is dynamically loaded. */ 323 regs->uregs[8] = infop->interpreter_loadmap_addr; 324 regs->uregs[9] = infop->interpreter_pt_dynamic_addr; 325 } else { 326 regs->uregs[8] = 0; 327 regs->uregs[9] = infop->pt_dynamic_addr; 328 } 329 } 330 } 331 332 #define ELF_NREG 18 333 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 334 335 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUARMState *env) 336 { 337 (*regs)[0] = tswapreg(env->regs[0]); 338 (*regs)[1] = tswapreg(env->regs[1]); 339 (*regs)[2] = tswapreg(env->regs[2]); 340 (*regs)[3] = tswapreg(env->regs[3]); 341 (*regs)[4] = tswapreg(env->regs[4]); 342 (*regs)[5] = tswapreg(env->regs[5]); 343 (*regs)[6] = tswapreg(env->regs[6]); 344 (*regs)[7] = tswapreg(env->regs[7]); 345 (*regs)[8] = tswapreg(env->regs[8]); 346 (*regs)[9] = tswapreg(env->regs[9]); 347 (*regs)[10] = tswapreg(env->regs[10]); 348 (*regs)[11] = tswapreg(env->regs[11]); 349 (*regs)[12] = tswapreg(env->regs[12]); 350 (*regs)[13] = tswapreg(env->regs[13]); 351 (*regs)[14] = tswapreg(env->regs[14]); 352 (*regs)[15] = tswapreg(env->regs[15]); 353 354 (*regs)[16] = tswapreg(cpsr_read((CPUARMState *)env)); 355 (*regs)[17] = tswapreg(env->regs[0]); /* XXX */ 356 } 357 358 #define USE_ELF_CORE_DUMP 359 #define ELF_EXEC_PAGESIZE 4096 360 361 enum 362 { 363 ARM_HWCAP_ARM_SWP = 1 << 0, 364 ARM_HWCAP_ARM_HALF = 1 << 1, 365 ARM_HWCAP_ARM_THUMB = 1 << 2, 366 ARM_HWCAP_ARM_26BIT = 1 << 3, 367 ARM_HWCAP_ARM_FAST_MULT = 1 << 4, 368 ARM_HWCAP_ARM_FPA = 1 << 5, 369 ARM_HWCAP_ARM_VFP = 1 << 6, 370 ARM_HWCAP_ARM_EDSP = 1 << 7, 371 ARM_HWCAP_ARM_JAVA = 1 << 8, 372 ARM_HWCAP_ARM_IWMMXT = 1 << 9, 373 ARM_HWCAP_ARM_CRUNCH = 1 << 10, 374 ARM_HWCAP_ARM_THUMBEE = 1 << 11, 375 ARM_HWCAP_ARM_NEON = 1 << 12, 376 ARM_HWCAP_ARM_VFPv3 = 1 << 13, 377 ARM_HWCAP_ARM_VFPv3D16 = 1 << 14, 378 ARM_HWCAP_ARM_TLS = 1 << 15, 379 ARM_HWCAP_ARM_VFPv4 = 1 << 16, 380 ARM_HWCAP_ARM_IDIVA = 1 << 17, 381 ARM_HWCAP_ARM_IDIVT = 1 << 18, 382 ARM_HWCAP_ARM_VFPD32 = 1 << 19, 383 ARM_HWCAP_ARM_LPAE = 1 << 20, 384 ARM_HWCAP_ARM_EVTSTRM = 1 << 21, 385 }; 386 387 enum { 388 ARM_HWCAP2_ARM_AES = 1 << 0, 389 ARM_HWCAP2_ARM_PMULL = 1 << 1, 390 ARM_HWCAP2_ARM_SHA1 = 1 << 2, 391 ARM_HWCAP2_ARM_SHA2 = 1 << 3, 392 ARM_HWCAP2_ARM_CRC32 = 1 << 4, 393 }; 394 395 /* The commpage only exists for 32 bit kernels */ 396 397 #define HI_COMMPAGE (intptr_t)0xffff0f00u 398 399 static bool init_guest_commpage(void) 400 { 401 void *want = g2h_untagged(HI_COMMPAGE & -qemu_host_page_size); 402 void *addr = mmap(want, qemu_host_page_size, PROT_READ | PROT_WRITE, 403 MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0); 404 405 if (addr == MAP_FAILED) { 406 perror("Allocating guest commpage"); 407 exit(EXIT_FAILURE); 408 } 409 if (addr != want) { 410 return false; 411 } 412 413 /* Set kernel helper versions; rest of page is 0. */ 414 __put_user(5, (uint32_t *)g2h_untagged(0xffff0ffcu)); 415 416 if (mprotect(addr, qemu_host_page_size, PROT_READ)) { 417 perror("Protecting guest commpage"); 418 exit(EXIT_FAILURE); 419 } 420 return true; 421 } 422 423 #define ELF_HWCAP get_elf_hwcap() 424 #define ELF_HWCAP2 get_elf_hwcap2() 425 426 static uint32_t get_elf_hwcap(void) 427 { 428 ARMCPU *cpu = ARM_CPU(thread_cpu); 429 uint32_t hwcaps = 0; 430 431 hwcaps |= ARM_HWCAP_ARM_SWP; 432 hwcaps |= ARM_HWCAP_ARM_HALF; 433 hwcaps |= ARM_HWCAP_ARM_THUMB; 434 hwcaps |= ARM_HWCAP_ARM_FAST_MULT; 435 436 /* probe for the extra features */ 437 #define GET_FEATURE(feat, hwcap) \ 438 do { if (arm_feature(&cpu->env, feat)) { hwcaps |= hwcap; } } while (0) 439 440 #define GET_FEATURE_ID(feat, hwcap) \ 441 do { if (cpu_isar_feature(feat, cpu)) { hwcaps |= hwcap; } } while (0) 442 443 /* EDSP is in v5TE and above, but all our v5 CPUs are v5TE */ 444 GET_FEATURE(ARM_FEATURE_V5, ARM_HWCAP_ARM_EDSP); 445 GET_FEATURE(ARM_FEATURE_IWMMXT, ARM_HWCAP_ARM_IWMMXT); 446 GET_FEATURE(ARM_FEATURE_THUMB2EE, ARM_HWCAP_ARM_THUMBEE); 447 GET_FEATURE(ARM_FEATURE_NEON, ARM_HWCAP_ARM_NEON); 448 GET_FEATURE(ARM_FEATURE_V6K, ARM_HWCAP_ARM_TLS); 449 GET_FEATURE(ARM_FEATURE_LPAE, ARM_HWCAP_ARM_LPAE); 450 GET_FEATURE_ID(aa32_arm_div, ARM_HWCAP_ARM_IDIVA); 451 GET_FEATURE_ID(aa32_thumb_div, ARM_HWCAP_ARM_IDIVT); 452 GET_FEATURE_ID(aa32_vfp, ARM_HWCAP_ARM_VFP); 453 454 if (cpu_isar_feature(aa32_fpsp_v3, cpu) || 455 cpu_isar_feature(aa32_fpdp_v3, cpu)) { 456 hwcaps |= ARM_HWCAP_ARM_VFPv3; 457 if (cpu_isar_feature(aa32_simd_r32, cpu)) { 458 hwcaps |= ARM_HWCAP_ARM_VFPD32; 459 } else { 460 hwcaps |= ARM_HWCAP_ARM_VFPv3D16; 461 } 462 } 463 GET_FEATURE_ID(aa32_simdfmac, ARM_HWCAP_ARM_VFPv4); 464 465 return hwcaps; 466 } 467 468 static uint32_t get_elf_hwcap2(void) 469 { 470 ARMCPU *cpu = ARM_CPU(thread_cpu); 471 uint32_t hwcaps = 0; 472 473 GET_FEATURE_ID(aa32_aes, ARM_HWCAP2_ARM_AES); 474 GET_FEATURE_ID(aa32_pmull, ARM_HWCAP2_ARM_PMULL); 475 GET_FEATURE_ID(aa32_sha1, ARM_HWCAP2_ARM_SHA1); 476 GET_FEATURE_ID(aa32_sha2, ARM_HWCAP2_ARM_SHA2); 477 GET_FEATURE_ID(aa32_crc32, ARM_HWCAP2_ARM_CRC32); 478 return hwcaps; 479 } 480 481 #undef GET_FEATURE 482 #undef GET_FEATURE_ID 483 484 #define ELF_PLATFORM get_elf_platform() 485 486 static const char *get_elf_platform(void) 487 { 488 CPUARMState *env = thread_cpu->env_ptr; 489 490 #if TARGET_BIG_ENDIAN 491 # define END "b" 492 #else 493 # define END "l" 494 #endif 495 496 if (arm_feature(env, ARM_FEATURE_V8)) { 497 return "v8" END; 498 } else if (arm_feature(env, ARM_FEATURE_V7)) { 499 if (arm_feature(env, ARM_FEATURE_M)) { 500 return "v7m" END; 501 } else { 502 return "v7" END; 503 } 504 } else if (arm_feature(env, ARM_FEATURE_V6)) { 505 return "v6" END; 506 } else if (arm_feature(env, ARM_FEATURE_V5)) { 507 return "v5" END; 508 } else { 509 return "v4" END; 510 } 511 512 #undef END 513 } 514 515 #else 516 /* 64 bit ARM definitions */ 517 #define ELF_START_MMAP 0x80000000 518 519 #define ELF_ARCH EM_AARCH64 520 #define ELF_CLASS ELFCLASS64 521 #if TARGET_BIG_ENDIAN 522 # define ELF_PLATFORM "aarch64_be" 523 #else 524 # define ELF_PLATFORM "aarch64" 525 #endif 526 527 static inline void init_thread(struct target_pt_regs *regs, 528 struct image_info *infop) 529 { 530 abi_long stack = infop->start_stack; 531 memset(regs, 0, sizeof(*regs)); 532 533 regs->pc = infop->entry & ~0x3ULL; 534 regs->sp = stack; 535 } 536 537 #define ELF_NREG 34 538 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 539 540 static void elf_core_copy_regs(target_elf_gregset_t *regs, 541 const CPUARMState *env) 542 { 543 int i; 544 545 for (i = 0; i < 32; i++) { 546 (*regs)[i] = tswapreg(env->xregs[i]); 547 } 548 (*regs)[32] = tswapreg(env->pc); 549 (*regs)[33] = tswapreg(pstate_read((CPUARMState *)env)); 550 } 551 552 #define USE_ELF_CORE_DUMP 553 #define ELF_EXEC_PAGESIZE 4096 554 555 enum { 556 ARM_HWCAP_A64_FP = 1 << 0, 557 ARM_HWCAP_A64_ASIMD = 1 << 1, 558 ARM_HWCAP_A64_EVTSTRM = 1 << 2, 559 ARM_HWCAP_A64_AES = 1 << 3, 560 ARM_HWCAP_A64_PMULL = 1 << 4, 561 ARM_HWCAP_A64_SHA1 = 1 << 5, 562 ARM_HWCAP_A64_SHA2 = 1 << 6, 563 ARM_HWCAP_A64_CRC32 = 1 << 7, 564 ARM_HWCAP_A64_ATOMICS = 1 << 8, 565 ARM_HWCAP_A64_FPHP = 1 << 9, 566 ARM_HWCAP_A64_ASIMDHP = 1 << 10, 567 ARM_HWCAP_A64_CPUID = 1 << 11, 568 ARM_HWCAP_A64_ASIMDRDM = 1 << 12, 569 ARM_HWCAP_A64_JSCVT = 1 << 13, 570 ARM_HWCAP_A64_FCMA = 1 << 14, 571 ARM_HWCAP_A64_LRCPC = 1 << 15, 572 ARM_HWCAP_A64_DCPOP = 1 << 16, 573 ARM_HWCAP_A64_SHA3 = 1 << 17, 574 ARM_HWCAP_A64_SM3 = 1 << 18, 575 ARM_HWCAP_A64_SM4 = 1 << 19, 576 ARM_HWCAP_A64_ASIMDDP = 1 << 20, 577 ARM_HWCAP_A64_SHA512 = 1 << 21, 578 ARM_HWCAP_A64_SVE = 1 << 22, 579 ARM_HWCAP_A64_ASIMDFHM = 1 << 23, 580 ARM_HWCAP_A64_DIT = 1 << 24, 581 ARM_HWCAP_A64_USCAT = 1 << 25, 582 ARM_HWCAP_A64_ILRCPC = 1 << 26, 583 ARM_HWCAP_A64_FLAGM = 1 << 27, 584 ARM_HWCAP_A64_SSBS = 1 << 28, 585 ARM_HWCAP_A64_SB = 1 << 29, 586 ARM_HWCAP_A64_PACA = 1 << 30, 587 ARM_HWCAP_A64_PACG = 1UL << 31, 588 589 ARM_HWCAP2_A64_DCPODP = 1 << 0, 590 ARM_HWCAP2_A64_SVE2 = 1 << 1, 591 ARM_HWCAP2_A64_SVEAES = 1 << 2, 592 ARM_HWCAP2_A64_SVEPMULL = 1 << 3, 593 ARM_HWCAP2_A64_SVEBITPERM = 1 << 4, 594 ARM_HWCAP2_A64_SVESHA3 = 1 << 5, 595 ARM_HWCAP2_A64_SVESM4 = 1 << 6, 596 ARM_HWCAP2_A64_FLAGM2 = 1 << 7, 597 ARM_HWCAP2_A64_FRINT = 1 << 8, 598 ARM_HWCAP2_A64_SVEI8MM = 1 << 9, 599 ARM_HWCAP2_A64_SVEF32MM = 1 << 10, 600 ARM_HWCAP2_A64_SVEF64MM = 1 << 11, 601 ARM_HWCAP2_A64_SVEBF16 = 1 << 12, 602 ARM_HWCAP2_A64_I8MM = 1 << 13, 603 ARM_HWCAP2_A64_BF16 = 1 << 14, 604 ARM_HWCAP2_A64_DGH = 1 << 15, 605 ARM_HWCAP2_A64_RNG = 1 << 16, 606 ARM_HWCAP2_A64_BTI = 1 << 17, 607 ARM_HWCAP2_A64_MTE = 1 << 18, 608 ARM_HWCAP2_A64_ECV = 1 << 19, 609 ARM_HWCAP2_A64_AFP = 1 << 20, 610 ARM_HWCAP2_A64_RPRES = 1 << 21, 611 ARM_HWCAP2_A64_MTE3 = 1 << 22, 612 ARM_HWCAP2_A64_SME = 1 << 23, 613 ARM_HWCAP2_A64_SME_I16I64 = 1 << 24, 614 ARM_HWCAP2_A64_SME_F64F64 = 1 << 25, 615 ARM_HWCAP2_A64_SME_I8I32 = 1 << 26, 616 ARM_HWCAP2_A64_SME_F16F32 = 1 << 27, 617 ARM_HWCAP2_A64_SME_B16F32 = 1 << 28, 618 ARM_HWCAP2_A64_SME_F32F32 = 1 << 29, 619 ARM_HWCAP2_A64_SME_FA64 = 1 << 30, 620 }; 621 622 #define ELF_HWCAP get_elf_hwcap() 623 #define ELF_HWCAP2 get_elf_hwcap2() 624 625 #define GET_FEATURE_ID(feat, hwcap) \ 626 do { if (cpu_isar_feature(feat, cpu)) { hwcaps |= hwcap; } } while (0) 627 628 static uint32_t get_elf_hwcap(void) 629 { 630 ARMCPU *cpu = ARM_CPU(thread_cpu); 631 uint32_t hwcaps = 0; 632 633 hwcaps |= ARM_HWCAP_A64_FP; 634 hwcaps |= ARM_HWCAP_A64_ASIMD; 635 hwcaps |= ARM_HWCAP_A64_CPUID; 636 637 /* probe for the extra features */ 638 639 GET_FEATURE_ID(aa64_aes, ARM_HWCAP_A64_AES); 640 GET_FEATURE_ID(aa64_pmull, ARM_HWCAP_A64_PMULL); 641 GET_FEATURE_ID(aa64_sha1, ARM_HWCAP_A64_SHA1); 642 GET_FEATURE_ID(aa64_sha256, ARM_HWCAP_A64_SHA2); 643 GET_FEATURE_ID(aa64_sha512, ARM_HWCAP_A64_SHA512); 644 GET_FEATURE_ID(aa64_crc32, ARM_HWCAP_A64_CRC32); 645 GET_FEATURE_ID(aa64_sha3, ARM_HWCAP_A64_SHA3); 646 GET_FEATURE_ID(aa64_sm3, ARM_HWCAP_A64_SM3); 647 GET_FEATURE_ID(aa64_sm4, ARM_HWCAP_A64_SM4); 648 GET_FEATURE_ID(aa64_fp16, ARM_HWCAP_A64_FPHP | ARM_HWCAP_A64_ASIMDHP); 649 GET_FEATURE_ID(aa64_atomics, ARM_HWCAP_A64_ATOMICS); 650 GET_FEATURE_ID(aa64_rdm, ARM_HWCAP_A64_ASIMDRDM); 651 GET_FEATURE_ID(aa64_dp, ARM_HWCAP_A64_ASIMDDP); 652 GET_FEATURE_ID(aa64_fcma, ARM_HWCAP_A64_FCMA); 653 GET_FEATURE_ID(aa64_sve, ARM_HWCAP_A64_SVE); 654 GET_FEATURE_ID(aa64_pauth, ARM_HWCAP_A64_PACA | ARM_HWCAP_A64_PACG); 655 GET_FEATURE_ID(aa64_fhm, ARM_HWCAP_A64_ASIMDFHM); 656 GET_FEATURE_ID(aa64_jscvt, ARM_HWCAP_A64_JSCVT); 657 GET_FEATURE_ID(aa64_sb, ARM_HWCAP_A64_SB); 658 GET_FEATURE_ID(aa64_condm_4, ARM_HWCAP_A64_FLAGM); 659 GET_FEATURE_ID(aa64_dcpop, ARM_HWCAP_A64_DCPOP); 660 GET_FEATURE_ID(aa64_rcpc_8_3, ARM_HWCAP_A64_LRCPC); 661 GET_FEATURE_ID(aa64_rcpc_8_4, ARM_HWCAP_A64_ILRCPC); 662 663 return hwcaps; 664 } 665 666 static uint32_t get_elf_hwcap2(void) 667 { 668 ARMCPU *cpu = ARM_CPU(thread_cpu); 669 uint32_t hwcaps = 0; 670 671 GET_FEATURE_ID(aa64_dcpodp, ARM_HWCAP2_A64_DCPODP); 672 GET_FEATURE_ID(aa64_sve2, ARM_HWCAP2_A64_SVE2); 673 GET_FEATURE_ID(aa64_sve2_aes, ARM_HWCAP2_A64_SVEAES); 674 GET_FEATURE_ID(aa64_sve2_pmull128, ARM_HWCAP2_A64_SVEPMULL); 675 GET_FEATURE_ID(aa64_sve2_bitperm, ARM_HWCAP2_A64_SVEBITPERM); 676 GET_FEATURE_ID(aa64_sve2_sha3, ARM_HWCAP2_A64_SVESHA3); 677 GET_FEATURE_ID(aa64_sve2_sm4, ARM_HWCAP2_A64_SVESM4); 678 GET_FEATURE_ID(aa64_condm_5, ARM_HWCAP2_A64_FLAGM2); 679 GET_FEATURE_ID(aa64_frint, ARM_HWCAP2_A64_FRINT); 680 GET_FEATURE_ID(aa64_sve_i8mm, ARM_HWCAP2_A64_SVEI8MM); 681 GET_FEATURE_ID(aa64_sve_f32mm, ARM_HWCAP2_A64_SVEF32MM); 682 GET_FEATURE_ID(aa64_sve_f64mm, ARM_HWCAP2_A64_SVEF64MM); 683 GET_FEATURE_ID(aa64_sve_bf16, ARM_HWCAP2_A64_SVEBF16); 684 GET_FEATURE_ID(aa64_i8mm, ARM_HWCAP2_A64_I8MM); 685 GET_FEATURE_ID(aa64_bf16, ARM_HWCAP2_A64_BF16); 686 GET_FEATURE_ID(aa64_rndr, ARM_HWCAP2_A64_RNG); 687 GET_FEATURE_ID(aa64_bti, ARM_HWCAP2_A64_BTI); 688 GET_FEATURE_ID(aa64_mte, ARM_HWCAP2_A64_MTE); 689 GET_FEATURE_ID(aa64_sme, (ARM_HWCAP2_A64_SME | 690 ARM_HWCAP2_A64_SME_F32F32 | 691 ARM_HWCAP2_A64_SME_B16F32 | 692 ARM_HWCAP2_A64_SME_F16F32 | 693 ARM_HWCAP2_A64_SME_I8I32)); 694 GET_FEATURE_ID(aa64_sme_f64f64, ARM_HWCAP2_A64_SME_F64F64); 695 GET_FEATURE_ID(aa64_sme_i16i64, ARM_HWCAP2_A64_SME_I16I64); 696 GET_FEATURE_ID(aa64_sme_fa64, ARM_HWCAP2_A64_SME_FA64); 697 698 return hwcaps; 699 } 700 701 #undef GET_FEATURE_ID 702 703 #endif /* not TARGET_AARCH64 */ 704 #endif /* TARGET_ARM */ 705 706 #ifdef TARGET_SPARC 707 #ifdef TARGET_SPARC64 708 709 #define ELF_START_MMAP 0x80000000 710 #define ELF_HWCAP (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \ 711 | HWCAP_SPARC_MULDIV | HWCAP_SPARC_V9) 712 #ifndef TARGET_ABI32 713 #define elf_check_arch(x) ( (x) == EM_SPARCV9 || (x) == EM_SPARC32PLUS ) 714 #else 715 #define elf_check_arch(x) ( (x) == EM_SPARC32PLUS || (x) == EM_SPARC ) 716 #endif 717 718 #define ELF_CLASS ELFCLASS64 719 #define ELF_ARCH EM_SPARCV9 720 #else 721 #define ELF_START_MMAP 0x80000000 722 #define ELF_HWCAP (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \ 723 | HWCAP_SPARC_MULDIV) 724 #define ELF_CLASS ELFCLASS32 725 #define ELF_ARCH EM_SPARC 726 #endif /* TARGET_SPARC64 */ 727 728 static inline void init_thread(struct target_pt_regs *regs, 729 struct image_info *infop) 730 { 731 /* Note that target_cpu_copy_regs does not read psr/tstate. */ 732 regs->pc = infop->entry; 733 regs->npc = regs->pc + 4; 734 regs->y = 0; 735 regs->u_regs[14] = (infop->start_stack - 16 * sizeof(abi_ulong) 736 - TARGET_STACK_BIAS); 737 } 738 #endif /* TARGET_SPARC */ 739 740 #ifdef TARGET_PPC 741 742 #define ELF_MACHINE PPC_ELF_MACHINE 743 #define ELF_START_MMAP 0x80000000 744 745 #if defined(TARGET_PPC64) 746 747 #define elf_check_arch(x) ( (x) == EM_PPC64 ) 748 749 #define ELF_CLASS ELFCLASS64 750 751 #else 752 753 #define ELF_CLASS ELFCLASS32 754 755 #endif 756 757 #define ELF_ARCH EM_PPC 758 759 /* Feature masks for the Aux Vector Hardware Capabilities (AT_HWCAP). 760 See arch/powerpc/include/asm/cputable.h. */ 761 enum { 762 QEMU_PPC_FEATURE_32 = 0x80000000, 763 QEMU_PPC_FEATURE_64 = 0x40000000, 764 QEMU_PPC_FEATURE_601_INSTR = 0x20000000, 765 QEMU_PPC_FEATURE_HAS_ALTIVEC = 0x10000000, 766 QEMU_PPC_FEATURE_HAS_FPU = 0x08000000, 767 QEMU_PPC_FEATURE_HAS_MMU = 0x04000000, 768 QEMU_PPC_FEATURE_HAS_4xxMAC = 0x02000000, 769 QEMU_PPC_FEATURE_UNIFIED_CACHE = 0x01000000, 770 QEMU_PPC_FEATURE_HAS_SPE = 0x00800000, 771 QEMU_PPC_FEATURE_HAS_EFP_SINGLE = 0x00400000, 772 QEMU_PPC_FEATURE_HAS_EFP_DOUBLE = 0x00200000, 773 QEMU_PPC_FEATURE_NO_TB = 0x00100000, 774 QEMU_PPC_FEATURE_POWER4 = 0x00080000, 775 QEMU_PPC_FEATURE_POWER5 = 0x00040000, 776 QEMU_PPC_FEATURE_POWER5_PLUS = 0x00020000, 777 QEMU_PPC_FEATURE_CELL = 0x00010000, 778 QEMU_PPC_FEATURE_BOOKE = 0x00008000, 779 QEMU_PPC_FEATURE_SMT = 0x00004000, 780 QEMU_PPC_FEATURE_ICACHE_SNOOP = 0x00002000, 781 QEMU_PPC_FEATURE_ARCH_2_05 = 0x00001000, 782 QEMU_PPC_FEATURE_PA6T = 0x00000800, 783 QEMU_PPC_FEATURE_HAS_DFP = 0x00000400, 784 QEMU_PPC_FEATURE_POWER6_EXT = 0x00000200, 785 QEMU_PPC_FEATURE_ARCH_2_06 = 0x00000100, 786 QEMU_PPC_FEATURE_HAS_VSX = 0x00000080, 787 QEMU_PPC_FEATURE_PSERIES_PERFMON_COMPAT = 0x00000040, 788 789 QEMU_PPC_FEATURE_TRUE_LE = 0x00000002, 790 QEMU_PPC_FEATURE_PPC_LE = 0x00000001, 791 792 /* Feature definitions in AT_HWCAP2. */ 793 QEMU_PPC_FEATURE2_ARCH_2_07 = 0x80000000, /* ISA 2.07 */ 794 QEMU_PPC_FEATURE2_HAS_HTM = 0x40000000, /* Hardware Transactional Memory */ 795 QEMU_PPC_FEATURE2_HAS_DSCR = 0x20000000, /* Data Stream Control Register */ 796 QEMU_PPC_FEATURE2_HAS_EBB = 0x10000000, /* Event Base Branching */ 797 QEMU_PPC_FEATURE2_HAS_ISEL = 0x08000000, /* Integer Select */ 798 QEMU_PPC_FEATURE2_HAS_TAR = 0x04000000, /* Target Address Register */ 799 QEMU_PPC_FEATURE2_VEC_CRYPTO = 0x02000000, 800 QEMU_PPC_FEATURE2_HTM_NOSC = 0x01000000, 801 QEMU_PPC_FEATURE2_ARCH_3_00 = 0x00800000, /* ISA 3.00 */ 802 QEMU_PPC_FEATURE2_HAS_IEEE128 = 0x00400000, /* VSX IEEE Bin Float 128-bit */ 803 QEMU_PPC_FEATURE2_DARN = 0x00200000, /* darn random number insn */ 804 QEMU_PPC_FEATURE2_SCV = 0x00100000, /* scv syscall */ 805 QEMU_PPC_FEATURE2_HTM_NO_SUSPEND = 0x00080000, /* TM w/o suspended state */ 806 QEMU_PPC_FEATURE2_ARCH_3_1 = 0x00040000, /* ISA 3.1 */ 807 QEMU_PPC_FEATURE2_MMA = 0x00020000, /* Matrix-Multiply Assist */ 808 }; 809 810 #define ELF_HWCAP get_elf_hwcap() 811 812 static uint32_t get_elf_hwcap(void) 813 { 814 PowerPCCPU *cpu = POWERPC_CPU(thread_cpu); 815 uint32_t features = 0; 816 817 /* We don't have to be terribly complete here; the high points are 818 Altivec/FP/SPE support. Anything else is just a bonus. */ 819 #define GET_FEATURE(flag, feature) \ 820 do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0) 821 #define GET_FEATURE2(flags, feature) \ 822 do { \ 823 if ((cpu->env.insns_flags2 & flags) == flags) { \ 824 features |= feature; \ 825 } \ 826 } while (0) 827 GET_FEATURE(PPC_64B, QEMU_PPC_FEATURE_64); 828 GET_FEATURE(PPC_FLOAT, QEMU_PPC_FEATURE_HAS_FPU); 829 GET_FEATURE(PPC_ALTIVEC, QEMU_PPC_FEATURE_HAS_ALTIVEC); 830 GET_FEATURE(PPC_SPE, QEMU_PPC_FEATURE_HAS_SPE); 831 GET_FEATURE(PPC_SPE_SINGLE, QEMU_PPC_FEATURE_HAS_EFP_SINGLE); 832 GET_FEATURE(PPC_SPE_DOUBLE, QEMU_PPC_FEATURE_HAS_EFP_DOUBLE); 833 GET_FEATURE(PPC_BOOKE, QEMU_PPC_FEATURE_BOOKE); 834 GET_FEATURE(PPC_405_MAC, QEMU_PPC_FEATURE_HAS_4xxMAC); 835 GET_FEATURE2(PPC2_DFP, QEMU_PPC_FEATURE_HAS_DFP); 836 GET_FEATURE2(PPC2_VSX, QEMU_PPC_FEATURE_HAS_VSX); 837 GET_FEATURE2((PPC2_PERM_ISA206 | PPC2_DIVE_ISA206 | PPC2_ATOMIC_ISA206 | 838 PPC2_FP_CVT_ISA206 | PPC2_FP_TST_ISA206), 839 QEMU_PPC_FEATURE_ARCH_2_06); 840 #undef GET_FEATURE 841 #undef GET_FEATURE2 842 843 return features; 844 } 845 846 #define ELF_HWCAP2 get_elf_hwcap2() 847 848 static uint32_t get_elf_hwcap2(void) 849 { 850 PowerPCCPU *cpu = POWERPC_CPU(thread_cpu); 851 uint32_t features = 0; 852 853 #define GET_FEATURE(flag, feature) \ 854 do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0) 855 #define GET_FEATURE2(flag, feature) \ 856 do { if (cpu->env.insns_flags2 & flag) { features |= feature; } } while (0) 857 858 GET_FEATURE(PPC_ISEL, QEMU_PPC_FEATURE2_HAS_ISEL); 859 GET_FEATURE2(PPC2_BCTAR_ISA207, QEMU_PPC_FEATURE2_HAS_TAR); 860 GET_FEATURE2((PPC2_BCTAR_ISA207 | PPC2_LSQ_ISA207 | PPC2_ALTIVEC_207 | 861 PPC2_ISA207S), QEMU_PPC_FEATURE2_ARCH_2_07 | 862 QEMU_PPC_FEATURE2_VEC_CRYPTO); 863 GET_FEATURE2(PPC2_ISA300, QEMU_PPC_FEATURE2_ARCH_3_00 | 864 QEMU_PPC_FEATURE2_DARN | QEMU_PPC_FEATURE2_HAS_IEEE128); 865 GET_FEATURE2(PPC2_ISA310, QEMU_PPC_FEATURE2_ARCH_3_1 | 866 QEMU_PPC_FEATURE2_MMA); 867 868 #undef GET_FEATURE 869 #undef GET_FEATURE2 870 871 return features; 872 } 873 874 /* 875 * The requirements here are: 876 * - keep the final alignment of sp (sp & 0xf) 877 * - make sure the 32-bit value at the first 16 byte aligned position of 878 * AUXV is greater than 16 for glibc compatibility. 879 * AT_IGNOREPPC is used for that. 880 * - for compatibility with glibc ARCH_DLINFO must always be defined on PPC, 881 * even if DLINFO_ARCH_ITEMS goes to zero or is undefined. 882 */ 883 #define DLINFO_ARCH_ITEMS 5 884 #define ARCH_DLINFO \ 885 do { \ 886 PowerPCCPU *cpu = POWERPC_CPU(thread_cpu); \ 887 /* \ 888 * Handle glibc compatibility: these magic entries must \ 889 * be at the lowest addresses in the final auxv. \ 890 */ \ 891 NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC); \ 892 NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC); \ 893 NEW_AUX_ENT(AT_DCACHEBSIZE, cpu->env.dcache_line_size); \ 894 NEW_AUX_ENT(AT_ICACHEBSIZE, cpu->env.icache_line_size); \ 895 NEW_AUX_ENT(AT_UCACHEBSIZE, 0); \ 896 } while (0) 897 898 static inline void init_thread(struct target_pt_regs *_regs, struct image_info *infop) 899 { 900 _regs->gpr[1] = infop->start_stack; 901 #if defined(TARGET_PPC64) 902 if (get_ppc64_abi(infop) < 2) { 903 uint64_t val; 904 get_user_u64(val, infop->entry + 8); 905 _regs->gpr[2] = val + infop->load_bias; 906 get_user_u64(val, infop->entry); 907 infop->entry = val + infop->load_bias; 908 } else { 909 _regs->gpr[12] = infop->entry; /* r12 set to global entry address */ 910 } 911 #endif 912 _regs->nip = infop->entry; 913 } 914 915 /* See linux kernel: arch/powerpc/include/asm/elf.h. */ 916 #define ELF_NREG 48 917 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 918 919 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUPPCState *env) 920 { 921 int i; 922 target_ulong ccr = 0; 923 924 for (i = 0; i < ARRAY_SIZE(env->gpr); i++) { 925 (*regs)[i] = tswapreg(env->gpr[i]); 926 } 927 928 (*regs)[32] = tswapreg(env->nip); 929 (*regs)[33] = tswapreg(env->msr); 930 (*regs)[35] = tswapreg(env->ctr); 931 (*regs)[36] = tswapreg(env->lr); 932 (*regs)[37] = tswapreg(cpu_read_xer(env)); 933 934 for (i = 0; i < ARRAY_SIZE(env->crf); i++) { 935 ccr |= env->crf[i] << (32 - ((i + 1) * 4)); 936 } 937 (*regs)[38] = tswapreg(ccr); 938 } 939 940 #define USE_ELF_CORE_DUMP 941 #define ELF_EXEC_PAGESIZE 4096 942 943 #endif 944 945 #ifdef TARGET_LOONGARCH64 946 947 #define ELF_START_MMAP 0x80000000 948 949 #define ELF_CLASS ELFCLASS64 950 #define ELF_ARCH EM_LOONGARCH 951 952 #define elf_check_arch(x) ((x) == EM_LOONGARCH) 953 954 static inline void init_thread(struct target_pt_regs *regs, 955 struct image_info *infop) 956 { 957 /*Set crmd PG,DA = 1,0 */ 958 regs->csr.crmd = 2 << 3; 959 regs->csr.era = infop->entry; 960 regs->regs[3] = infop->start_stack; 961 } 962 963 /* See linux kernel: arch/loongarch/include/asm/elf.h */ 964 #define ELF_NREG 45 965 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 966 967 enum { 968 TARGET_EF_R0 = 0, 969 TARGET_EF_CSR_ERA = TARGET_EF_R0 + 33, 970 TARGET_EF_CSR_BADV = TARGET_EF_R0 + 34, 971 }; 972 973 static void elf_core_copy_regs(target_elf_gregset_t *regs, 974 const CPULoongArchState *env) 975 { 976 int i; 977 978 (*regs)[TARGET_EF_R0] = 0; 979 980 for (i = 1; i < ARRAY_SIZE(env->gpr); i++) { 981 (*regs)[TARGET_EF_R0 + i] = tswapreg(env->gpr[i]); 982 } 983 984 (*regs)[TARGET_EF_CSR_ERA] = tswapreg(env->pc); 985 (*regs)[TARGET_EF_CSR_BADV] = tswapreg(env->CSR_BADV); 986 } 987 988 #define USE_ELF_CORE_DUMP 989 #define ELF_EXEC_PAGESIZE 4096 990 991 #define ELF_HWCAP get_elf_hwcap() 992 993 /* See arch/loongarch/include/uapi/asm/hwcap.h */ 994 enum { 995 HWCAP_LOONGARCH_CPUCFG = (1 << 0), 996 HWCAP_LOONGARCH_LAM = (1 << 1), 997 HWCAP_LOONGARCH_UAL = (1 << 2), 998 HWCAP_LOONGARCH_FPU = (1 << 3), 999 HWCAP_LOONGARCH_LSX = (1 << 4), 1000 HWCAP_LOONGARCH_LASX = (1 << 5), 1001 HWCAP_LOONGARCH_CRC32 = (1 << 6), 1002 HWCAP_LOONGARCH_COMPLEX = (1 << 7), 1003 HWCAP_LOONGARCH_CRYPTO = (1 << 8), 1004 HWCAP_LOONGARCH_LVZ = (1 << 9), 1005 HWCAP_LOONGARCH_LBT_X86 = (1 << 10), 1006 HWCAP_LOONGARCH_LBT_ARM = (1 << 11), 1007 HWCAP_LOONGARCH_LBT_MIPS = (1 << 12), 1008 }; 1009 1010 static uint32_t get_elf_hwcap(void) 1011 { 1012 LoongArchCPU *cpu = LOONGARCH_CPU(thread_cpu); 1013 uint32_t hwcaps = 0; 1014 1015 hwcaps |= HWCAP_LOONGARCH_CRC32; 1016 1017 if (FIELD_EX32(cpu->env.cpucfg[1], CPUCFG1, UAL)) { 1018 hwcaps |= HWCAP_LOONGARCH_UAL; 1019 } 1020 1021 if (FIELD_EX32(cpu->env.cpucfg[2], CPUCFG2, FP)) { 1022 hwcaps |= HWCAP_LOONGARCH_FPU; 1023 } 1024 1025 if (FIELD_EX32(cpu->env.cpucfg[2], CPUCFG2, LAM)) { 1026 hwcaps |= HWCAP_LOONGARCH_LAM; 1027 } 1028 1029 return hwcaps; 1030 } 1031 1032 #define ELF_PLATFORM "loongarch" 1033 1034 #endif /* TARGET_LOONGARCH64 */ 1035 1036 #ifdef TARGET_MIPS 1037 1038 #define ELF_START_MMAP 0x80000000 1039 1040 #ifdef TARGET_MIPS64 1041 #define ELF_CLASS ELFCLASS64 1042 #else 1043 #define ELF_CLASS ELFCLASS32 1044 #endif 1045 #define ELF_ARCH EM_MIPS 1046 1047 #ifdef TARGET_ABI_MIPSN32 1048 #define elf_check_abi(x) ((x) & EF_MIPS_ABI2) 1049 #else 1050 #define elf_check_abi(x) (!((x) & EF_MIPS_ABI2)) 1051 #endif 1052 1053 static inline void init_thread(struct target_pt_regs *regs, 1054 struct image_info *infop) 1055 { 1056 regs->cp0_status = 2 << CP0St_KSU; 1057 regs->cp0_epc = infop->entry; 1058 regs->regs[29] = infop->start_stack; 1059 } 1060 1061 /* See linux kernel: arch/mips/include/asm/elf.h. */ 1062 #define ELF_NREG 45 1063 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1064 1065 /* See linux kernel: arch/mips/include/asm/reg.h. */ 1066 enum { 1067 #ifdef TARGET_MIPS64 1068 TARGET_EF_R0 = 0, 1069 #else 1070 TARGET_EF_R0 = 6, 1071 #endif 1072 TARGET_EF_R26 = TARGET_EF_R0 + 26, 1073 TARGET_EF_R27 = TARGET_EF_R0 + 27, 1074 TARGET_EF_LO = TARGET_EF_R0 + 32, 1075 TARGET_EF_HI = TARGET_EF_R0 + 33, 1076 TARGET_EF_CP0_EPC = TARGET_EF_R0 + 34, 1077 TARGET_EF_CP0_BADVADDR = TARGET_EF_R0 + 35, 1078 TARGET_EF_CP0_STATUS = TARGET_EF_R0 + 36, 1079 TARGET_EF_CP0_CAUSE = TARGET_EF_R0 + 37 1080 }; 1081 1082 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */ 1083 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMIPSState *env) 1084 { 1085 int i; 1086 1087 for (i = 0; i < TARGET_EF_R0; i++) { 1088 (*regs)[i] = 0; 1089 } 1090 (*regs)[TARGET_EF_R0] = 0; 1091 1092 for (i = 1; i < ARRAY_SIZE(env->active_tc.gpr); i++) { 1093 (*regs)[TARGET_EF_R0 + i] = tswapreg(env->active_tc.gpr[i]); 1094 } 1095 1096 (*regs)[TARGET_EF_R26] = 0; 1097 (*regs)[TARGET_EF_R27] = 0; 1098 (*regs)[TARGET_EF_LO] = tswapreg(env->active_tc.LO[0]); 1099 (*regs)[TARGET_EF_HI] = tswapreg(env->active_tc.HI[0]); 1100 (*regs)[TARGET_EF_CP0_EPC] = tswapreg(env->active_tc.PC); 1101 (*regs)[TARGET_EF_CP0_BADVADDR] = tswapreg(env->CP0_BadVAddr); 1102 (*regs)[TARGET_EF_CP0_STATUS] = tswapreg(env->CP0_Status); 1103 (*regs)[TARGET_EF_CP0_CAUSE] = tswapreg(env->CP0_Cause); 1104 } 1105 1106 #define USE_ELF_CORE_DUMP 1107 #define ELF_EXEC_PAGESIZE 4096 1108 1109 /* See arch/mips/include/uapi/asm/hwcap.h. */ 1110 enum { 1111 HWCAP_MIPS_R6 = (1 << 0), 1112 HWCAP_MIPS_MSA = (1 << 1), 1113 HWCAP_MIPS_CRC32 = (1 << 2), 1114 HWCAP_MIPS_MIPS16 = (1 << 3), 1115 HWCAP_MIPS_MDMX = (1 << 4), 1116 HWCAP_MIPS_MIPS3D = (1 << 5), 1117 HWCAP_MIPS_SMARTMIPS = (1 << 6), 1118 HWCAP_MIPS_DSP = (1 << 7), 1119 HWCAP_MIPS_DSP2 = (1 << 8), 1120 HWCAP_MIPS_DSP3 = (1 << 9), 1121 HWCAP_MIPS_MIPS16E2 = (1 << 10), 1122 HWCAP_LOONGSON_MMI = (1 << 11), 1123 HWCAP_LOONGSON_EXT = (1 << 12), 1124 HWCAP_LOONGSON_EXT2 = (1 << 13), 1125 HWCAP_LOONGSON_CPUCFG = (1 << 14), 1126 }; 1127 1128 #define ELF_HWCAP get_elf_hwcap() 1129 1130 #define GET_FEATURE_INSN(_flag, _hwcap) \ 1131 do { if (cpu->env.insn_flags & (_flag)) { hwcaps |= _hwcap; } } while (0) 1132 1133 #define GET_FEATURE_REG_SET(_reg, _mask, _hwcap) \ 1134 do { if (cpu->env._reg & (_mask)) { hwcaps |= _hwcap; } } while (0) 1135 1136 #define GET_FEATURE_REG_EQU(_reg, _start, _length, _val, _hwcap) \ 1137 do { \ 1138 if (extract32(cpu->env._reg, (_start), (_length)) == (_val)) { \ 1139 hwcaps |= _hwcap; \ 1140 } \ 1141 } while (0) 1142 1143 static uint32_t get_elf_hwcap(void) 1144 { 1145 MIPSCPU *cpu = MIPS_CPU(thread_cpu); 1146 uint32_t hwcaps = 0; 1147 1148 GET_FEATURE_REG_EQU(CP0_Config0, CP0C0_AR, CP0C0_AR_LENGTH, 1149 2, HWCAP_MIPS_R6); 1150 GET_FEATURE_REG_SET(CP0_Config3, 1 << CP0C3_MSAP, HWCAP_MIPS_MSA); 1151 GET_FEATURE_INSN(ASE_LMMI, HWCAP_LOONGSON_MMI); 1152 GET_FEATURE_INSN(ASE_LEXT, HWCAP_LOONGSON_EXT); 1153 1154 return hwcaps; 1155 } 1156 1157 #undef GET_FEATURE_REG_EQU 1158 #undef GET_FEATURE_REG_SET 1159 #undef GET_FEATURE_INSN 1160 1161 #endif /* TARGET_MIPS */ 1162 1163 #ifdef TARGET_MICROBLAZE 1164 1165 #define ELF_START_MMAP 0x80000000 1166 1167 #define elf_check_arch(x) ( (x) == EM_MICROBLAZE || (x) == EM_MICROBLAZE_OLD) 1168 1169 #define ELF_CLASS ELFCLASS32 1170 #define ELF_ARCH EM_MICROBLAZE 1171 1172 static inline void init_thread(struct target_pt_regs *regs, 1173 struct image_info *infop) 1174 { 1175 regs->pc = infop->entry; 1176 regs->r1 = infop->start_stack; 1177 1178 } 1179 1180 #define ELF_EXEC_PAGESIZE 4096 1181 1182 #define USE_ELF_CORE_DUMP 1183 #define ELF_NREG 38 1184 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1185 1186 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */ 1187 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMBState *env) 1188 { 1189 int i, pos = 0; 1190 1191 for (i = 0; i < 32; i++) { 1192 (*regs)[pos++] = tswapreg(env->regs[i]); 1193 } 1194 1195 (*regs)[pos++] = tswapreg(env->pc); 1196 (*regs)[pos++] = tswapreg(mb_cpu_read_msr(env)); 1197 (*regs)[pos++] = 0; 1198 (*regs)[pos++] = tswapreg(env->ear); 1199 (*regs)[pos++] = 0; 1200 (*regs)[pos++] = tswapreg(env->esr); 1201 } 1202 1203 #endif /* TARGET_MICROBLAZE */ 1204 1205 #ifdef TARGET_NIOS2 1206 1207 #define ELF_START_MMAP 0x80000000 1208 1209 #define elf_check_arch(x) ((x) == EM_ALTERA_NIOS2) 1210 1211 #define ELF_CLASS ELFCLASS32 1212 #define ELF_ARCH EM_ALTERA_NIOS2 1213 1214 static void init_thread(struct target_pt_regs *regs, struct image_info *infop) 1215 { 1216 regs->ea = infop->entry; 1217 regs->sp = infop->start_stack; 1218 } 1219 1220 #define LO_COMMPAGE TARGET_PAGE_SIZE 1221 1222 static bool init_guest_commpage(void) 1223 { 1224 static const uint8_t kuser_page[4 + 2 * 64] = { 1225 /* __kuser_helper_version */ 1226 [0x00] = 0x02, 0x00, 0x00, 0x00, 1227 1228 /* __kuser_cmpxchg */ 1229 [0x04] = 0x3a, 0x6c, 0x3b, 0x00, /* trap 16 */ 1230 0x3a, 0x28, 0x00, 0xf8, /* ret */ 1231 1232 /* __kuser_sigtramp */ 1233 [0x44] = 0xc4, 0x22, 0x80, 0x00, /* movi r2, __NR_rt_sigreturn */ 1234 0x3a, 0x68, 0x3b, 0x00, /* trap 0 */ 1235 }; 1236 1237 void *want = g2h_untagged(LO_COMMPAGE & -qemu_host_page_size); 1238 void *addr = mmap(want, qemu_host_page_size, PROT_READ | PROT_WRITE, 1239 MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0); 1240 1241 if (addr == MAP_FAILED) { 1242 perror("Allocating guest commpage"); 1243 exit(EXIT_FAILURE); 1244 } 1245 if (addr != want) { 1246 return false; 1247 } 1248 1249 memcpy(addr, kuser_page, sizeof(kuser_page)); 1250 1251 if (mprotect(addr, qemu_host_page_size, PROT_READ)) { 1252 perror("Protecting guest commpage"); 1253 exit(EXIT_FAILURE); 1254 } 1255 1256 page_set_flags(LO_COMMPAGE, LO_COMMPAGE + TARGET_PAGE_SIZE, 1257 PAGE_READ | PAGE_EXEC | PAGE_VALID); 1258 return true; 1259 } 1260 1261 #define ELF_EXEC_PAGESIZE 4096 1262 1263 #define USE_ELF_CORE_DUMP 1264 #define ELF_NREG 49 1265 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1266 1267 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */ 1268 static void elf_core_copy_regs(target_elf_gregset_t *regs, 1269 const CPUNios2State *env) 1270 { 1271 int i; 1272 1273 (*regs)[0] = -1; 1274 for (i = 1; i < 8; i++) /* r0-r7 */ 1275 (*regs)[i] = tswapreg(env->regs[i + 7]); 1276 1277 for (i = 8; i < 16; i++) /* r8-r15 */ 1278 (*regs)[i] = tswapreg(env->regs[i - 8]); 1279 1280 for (i = 16; i < 24; i++) /* r16-r23 */ 1281 (*regs)[i] = tswapreg(env->regs[i + 7]); 1282 (*regs)[24] = -1; /* R_ET */ 1283 (*regs)[25] = -1; /* R_BT */ 1284 (*regs)[26] = tswapreg(env->regs[R_GP]); 1285 (*regs)[27] = tswapreg(env->regs[R_SP]); 1286 (*regs)[28] = tswapreg(env->regs[R_FP]); 1287 (*regs)[29] = tswapreg(env->regs[R_EA]); 1288 (*regs)[30] = -1; /* R_SSTATUS */ 1289 (*regs)[31] = tswapreg(env->regs[R_RA]); 1290 1291 (*regs)[32] = tswapreg(env->pc); 1292 1293 (*regs)[33] = -1; /* R_STATUS */ 1294 (*regs)[34] = tswapreg(env->regs[CR_ESTATUS]); 1295 1296 for (i = 35; i < 49; i++) /* ... */ 1297 (*regs)[i] = -1; 1298 } 1299 1300 #endif /* TARGET_NIOS2 */ 1301 1302 #ifdef TARGET_OPENRISC 1303 1304 #define ELF_START_MMAP 0x08000000 1305 1306 #define ELF_ARCH EM_OPENRISC 1307 #define ELF_CLASS ELFCLASS32 1308 #define ELF_DATA ELFDATA2MSB 1309 1310 static inline void init_thread(struct target_pt_regs *regs, 1311 struct image_info *infop) 1312 { 1313 regs->pc = infop->entry; 1314 regs->gpr[1] = infop->start_stack; 1315 } 1316 1317 #define USE_ELF_CORE_DUMP 1318 #define ELF_EXEC_PAGESIZE 8192 1319 1320 /* See linux kernel arch/openrisc/include/asm/elf.h. */ 1321 #define ELF_NREG 34 /* gprs and pc, sr */ 1322 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1323 1324 static void elf_core_copy_regs(target_elf_gregset_t *regs, 1325 const CPUOpenRISCState *env) 1326 { 1327 int i; 1328 1329 for (i = 0; i < 32; i++) { 1330 (*regs)[i] = tswapreg(cpu_get_gpr(env, i)); 1331 } 1332 (*regs)[32] = tswapreg(env->pc); 1333 (*regs)[33] = tswapreg(cpu_get_sr(env)); 1334 } 1335 #define ELF_HWCAP 0 1336 #define ELF_PLATFORM NULL 1337 1338 #endif /* TARGET_OPENRISC */ 1339 1340 #ifdef TARGET_SH4 1341 1342 #define ELF_START_MMAP 0x80000000 1343 1344 #define ELF_CLASS ELFCLASS32 1345 #define ELF_ARCH EM_SH 1346 1347 static inline void init_thread(struct target_pt_regs *regs, 1348 struct image_info *infop) 1349 { 1350 /* Check other registers XXXXX */ 1351 regs->pc = infop->entry; 1352 regs->regs[15] = infop->start_stack; 1353 } 1354 1355 /* See linux kernel: arch/sh/include/asm/elf.h. */ 1356 #define ELF_NREG 23 1357 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1358 1359 /* See linux kernel: arch/sh/include/asm/ptrace.h. */ 1360 enum { 1361 TARGET_REG_PC = 16, 1362 TARGET_REG_PR = 17, 1363 TARGET_REG_SR = 18, 1364 TARGET_REG_GBR = 19, 1365 TARGET_REG_MACH = 20, 1366 TARGET_REG_MACL = 21, 1367 TARGET_REG_SYSCALL = 22 1368 }; 1369 1370 static inline void elf_core_copy_regs(target_elf_gregset_t *regs, 1371 const CPUSH4State *env) 1372 { 1373 int i; 1374 1375 for (i = 0; i < 16; i++) { 1376 (*regs)[i] = tswapreg(env->gregs[i]); 1377 } 1378 1379 (*regs)[TARGET_REG_PC] = tswapreg(env->pc); 1380 (*regs)[TARGET_REG_PR] = tswapreg(env->pr); 1381 (*regs)[TARGET_REG_SR] = tswapreg(env->sr); 1382 (*regs)[TARGET_REG_GBR] = tswapreg(env->gbr); 1383 (*regs)[TARGET_REG_MACH] = tswapreg(env->mach); 1384 (*regs)[TARGET_REG_MACL] = tswapreg(env->macl); 1385 (*regs)[TARGET_REG_SYSCALL] = 0; /* FIXME */ 1386 } 1387 1388 #define USE_ELF_CORE_DUMP 1389 #define ELF_EXEC_PAGESIZE 4096 1390 1391 enum { 1392 SH_CPU_HAS_FPU = 0x0001, /* Hardware FPU support */ 1393 SH_CPU_HAS_P2_FLUSH_BUG = 0x0002, /* Need to flush the cache in P2 area */ 1394 SH_CPU_HAS_MMU_PAGE_ASSOC = 0x0004, /* SH3: TLB way selection bit support */ 1395 SH_CPU_HAS_DSP = 0x0008, /* SH-DSP: DSP support */ 1396 SH_CPU_HAS_PERF_COUNTER = 0x0010, /* Hardware performance counters */ 1397 SH_CPU_HAS_PTEA = 0x0020, /* PTEA register */ 1398 SH_CPU_HAS_LLSC = 0x0040, /* movli.l/movco.l */ 1399 SH_CPU_HAS_L2_CACHE = 0x0080, /* Secondary cache / URAM */ 1400 SH_CPU_HAS_OP32 = 0x0100, /* 32-bit instruction support */ 1401 SH_CPU_HAS_PTEAEX = 0x0200, /* PTE ASID Extension support */ 1402 }; 1403 1404 #define ELF_HWCAP get_elf_hwcap() 1405 1406 static uint32_t get_elf_hwcap(void) 1407 { 1408 SuperHCPU *cpu = SUPERH_CPU(thread_cpu); 1409 uint32_t hwcap = 0; 1410 1411 hwcap |= SH_CPU_HAS_FPU; 1412 1413 if (cpu->env.features & SH_FEATURE_SH4A) { 1414 hwcap |= SH_CPU_HAS_LLSC; 1415 } 1416 1417 return hwcap; 1418 } 1419 1420 #endif 1421 1422 #ifdef TARGET_CRIS 1423 1424 #define ELF_START_MMAP 0x80000000 1425 1426 #define ELF_CLASS ELFCLASS32 1427 #define ELF_ARCH EM_CRIS 1428 1429 static inline void init_thread(struct target_pt_regs *regs, 1430 struct image_info *infop) 1431 { 1432 regs->erp = infop->entry; 1433 } 1434 1435 #define ELF_EXEC_PAGESIZE 8192 1436 1437 #endif 1438 1439 #ifdef TARGET_M68K 1440 1441 #define ELF_START_MMAP 0x80000000 1442 1443 #define ELF_CLASS ELFCLASS32 1444 #define ELF_ARCH EM_68K 1445 1446 /* ??? Does this need to do anything? 1447 #define ELF_PLAT_INIT(_r) */ 1448 1449 static inline void init_thread(struct target_pt_regs *regs, 1450 struct image_info *infop) 1451 { 1452 regs->usp = infop->start_stack; 1453 regs->sr = 0; 1454 regs->pc = infop->entry; 1455 } 1456 1457 /* See linux kernel: arch/m68k/include/asm/elf.h. */ 1458 #define ELF_NREG 20 1459 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1460 1461 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUM68KState *env) 1462 { 1463 (*regs)[0] = tswapreg(env->dregs[1]); 1464 (*regs)[1] = tswapreg(env->dregs[2]); 1465 (*regs)[2] = tswapreg(env->dregs[3]); 1466 (*regs)[3] = tswapreg(env->dregs[4]); 1467 (*regs)[4] = tswapreg(env->dregs[5]); 1468 (*regs)[5] = tswapreg(env->dregs[6]); 1469 (*regs)[6] = tswapreg(env->dregs[7]); 1470 (*regs)[7] = tswapreg(env->aregs[0]); 1471 (*regs)[8] = tswapreg(env->aregs[1]); 1472 (*regs)[9] = tswapreg(env->aregs[2]); 1473 (*regs)[10] = tswapreg(env->aregs[3]); 1474 (*regs)[11] = tswapreg(env->aregs[4]); 1475 (*regs)[12] = tswapreg(env->aregs[5]); 1476 (*regs)[13] = tswapreg(env->aregs[6]); 1477 (*regs)[14] = tswapreg(env->dregs[0]); 1478 (*regs)[15] = tswapreg(env->aregs[7]); 1479 (*regs)[16] = tswapreg(env->dregs[0]); /* FIXME: orig_d0 */ 1480 (*regs)[17] = tswapreg(env->sr); 1481 (*regs)[18] = tswapreg(env->pc); 1482 (*regs)[19] = 0; /* FIXME: regs->format | regs->vector */ 1483 } 1484 1485 #define USE_ELF_CORE_DUMP 1486 #define ELF_EXEC_PAGESIZE 8192 1487 1488 #endif 1489 1490 #ifdef TARGET_ALPHA 1491 1492 #define ELF_START_MMAP (0x30000000000ULL) 1493 1494 #define ELF_CLASS ELFCLASS64 1495 #define ELF_ARCH EM_ALPHA 1496 1497 static inline void init_thread(struct target_pt_regs *regs, 1498 struct image_info *infop) 1499 { 1500 regs->pc = infop->entry; 1501 regs->ps = 8; 1502 regs->usp = infop->start_stack; 1503 } 1504 1505 #define ELF_EXEC_PAGESIZE 8192 1506 1507 #endif /* TARGET_ALPHA */ 1508 1509 #ifdef TARGET_S390X 1510 1511 #define ELF_START_MMAP (0x20000000000ULL) 1512 1513 #define ELF_CLASS ELFCLASS64 1514 #define ELF_DATA ELFDATA2MSB 1515 #define ELF_ARCH EM_S390 1516 1517 #include "elf.h" 1518 1519 #define ELF_HWCAP get_elf_hwcap() 1520 1521 #define GET_FEATURE(_feat, _hwcap) \ 1522 do { if (s390_has_feat(_feat)) { hwcap |= _hwcap; } } while (0) 1523 1524 static uint32_t get_elf_hwcap(void) 1525 { 1526 /* 1527 * Let's assume we always have esan3 and zarch. 1528 * 31-bit processes can use 64-bit registers (high gprs). 1529 */ 1530 uint32_t hwcap = HWCAP_S390_ESAN3 | HWCAP_S390_ZARCH | HWCAP_S390_HIGH_GPRS; 1531 1532 GET_FEATURE(S390_FEAT_STFLE, HWCAP_S390_STFLE); 1533 GET_FEATURE(S390_FEAT_MSA, HWCAP_S390_MSA); 1534 GET_FEATURE(S390_FEAT_LONG_DISPLACEMENT, HWCAP_S390_LDISP); 1535 GET_FEATURE(S390_FEAT_EXTENDED_IMMEDIATE, HWCAP_S390_EIMM); 1536 if (s390_has_feat(S390_FEAT_EXTENDED_TRANSLATION_3) && 1537 s390_has_feat(S390_FEAT_ETF3_ENH)) { 1538 hwcap |= HWCAP_S390_ETF3EH; 1539 } 1540 GET_FEATURE(S390_FEAT_VECTOR, HWCAP_S390_VXRS); 1541 GET_FEATURE(S390_FEAT_VECTOR_ENH, HWCAP_S390_VXRS_EXT); 1542 1543 return hwcap; 1544 } 1545 1546 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop) 1547 { 1548 regs->psw.addr = infop->entry; 1549 regs->psw.mask = PSW_MASK_64 | PSW_MASK_32; 1550 regs->gprs[15] = infop->start_stack; 1551 } 1552 1553 /* See linux kernel: arch/s390/include/uapi/asm/ptrace.h (s390_regs). */ 1554 #define ELF_NREG 27 1555 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1556 1557 enum { 1558 TARGET_REG_PSWM = 0, 1559 TARGET_REG_PSWA = 1, 1560 TARGET_REG_GPRS = 2, 1561 TARGET_REG_ARS = 18, 1562 TARGET_REG_ORIG_R2 = 26, 1563 }; 1564 1565 static void elf_core_copy_regs(target_elf_gregset_t *regs, 1566 const CPUS390XState *env) 1567 { 1568 int i; 1569 uint32_t *aregs; 1570 1571 (*regs)[TARGET_REG_PSWM] = tswapreg(env->psw.mask); 1572 (*regs)[TARGET_REG_PSWA] = tswapreg(env->psw.addr); 1573 for (i = 0; i < 16; i++) { 1574 (*regs)[TARGET_REG_GPRS + i] = tswapreg(env->regs[i]); 1575 } 1576 aregs = (uint32_t *)&((*regs)[TARGET_REG_ARS]); 1577 for (i = 0; i < 16; i++) { 1578 aregs[i] = tswap32(env->aregs[i]); 1579 } 1580 (*regs)[TARGET_REG_ORIG_R2] = 0; 1581 } 1582 1583 #define USE_ELF_CORE_DUMP 1584 #define ELF_EXEC_PAGESIZE 4096 1585 1586 #endif /* TARGET_S390X */ 1587 1588 #ifdef TARGET_RISCV 1589 1590 #define ELF_START_MMAP 0x80000000 1591 #define ELF_ARCH EM_RISCV 1592 1593 #ifdef TARGET_RISCV32 1594 #define ELF_CLASS ELFCLASS32 1595 #else 1596 #define ELF_CLASS ELFCLASS64 1597 #endif 1598 1599 #define ELF_HWCAP get_elf_hwcap() 1600 1601 static uint32_t get_elf_hwcap(void) 1602 { 1603 #define MISA_BIT(EXT) (1 << (EXT - 'A')) 1604 RISCVCPU *cpu = RISCV_CPU(thread_cpu); 1605 uint32_t mask = MISA_BIT('I') | MISA_BIT('M') | MISA_BIT('A') 1606 | MISA_BIT('F') | MISA_BIT('D') | MISA_BIT('C'); 1607 1608 return cpu->env.misa_ext & mask; 1609 #undef MISA_BIT 1610 } 1611 1612 static inline void init_thread(struct target_pt_regs *regs, 1613 struct image_info *infop) 1614 { 1615 regs->sepc = infop->entry; 1616 regs->sp = infop->start_stack; 1617 } 1618 1619 #define ELF_EXEC_PAGESIZE 4096 1620 1621 #endif /* TARGET_RISCV */ 1622 1623 #ifdef TARGET_HPPA 1624 1625 #define ELF_START_MMAP 0x80000000 1626 #define ELF_CLASS ELFCLASS32 1627 #define ELF_ARCH EM_PARISC 1628 #define ELF_PLATFORM "PARISC" 1629 #define STACK_GROWS_DOWN 0 1630 #define STACK_ALIGNMENT 64 1631 1632 static inline void init_thread(struct target_pt_regs *regs, 1633 struct image_info *infop) 1634 { 1635 regs->iaoq[0] = infop->entry; 1636 regs->iaoq[1] = infop->entry + 4; 1637 regs->gr[23] = 0; 1638 regs->gr[24] = infop->argv; 1639 regs->gr[25] = infop->argc; 1640 /* The top-of-stack contains a linkage buffer. */ 1641 regs->gr[30] = infop->start_stack + 64; 1642 regs->gr[31] = infop->entry; 1643 } 1644 1645 #endif /* TARGET_HPPA */ 1646 1647 #ifdef TARGET_XTENSA 1648 1649 #define ELF_START_MMAP 0x20000000 1650 1651 #define ELF_CLASS ELFCLASS32 1652 #define ELF_ARCH EM_XTENSA 1653 1654 static inline void init_thread(struct target_pt_regs *regs, 1655 struct image_info *infop) 1656 { 1657 regs->windowbase = 0; 1658 regs->windowstart = 1; 1659 regs->areg[1] = infop->start_stack; 1660 regs->pc = infop->entry; 1661 } 1662 1663 /* See linux kernel: arch/xtensa/include/asm/elf.h. */ 1664 #define ELF_NREG 128 1665 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1666 1667 enum { 1668 TARGET_REG_PC, 1669 TARGET_REG_PS, 1670 TARGET_REG_LBEG, 1671 TARGET_REG_LEND, 1672 TARGET_REG_LCOUNT, 1673 TARGET_REG_SAR, 1674 TARGET_REG_WINDOWSTART, 1675 TARGET_REG_WINDOWBASE, 1676 TARGET_REG_THREADPTR, 1677 TARGET_REG_AR0 = 64, 1678 }; 1679 1680 static void elf_core_copy_regs(target_elf_gregset_t *regs, 1681 const CPUXtensaState *env) 1682 { 1683 unsigned i; 1684 1685 (*regs)[TARGET_REG_PC] = tswapreg(env->pc); 1686 (*regs)[TARGET_REG_PS] = tswapreg(env->sregs[PS] & ~PS_EXCM); 1687 (*regs)[TARGET_REG_LBEG] = tswapreg(env->sregs[LBEG]); 1688 (*regs)[TARGET_REG_LEND] = tswapreg(env->sregs[LEND]); 1689 (*regs)[TARGET_REG_LCOUNT] = tswapreg(env->sregs[LCOUNT]); 1690 (*regs)[TARGET_REG_SAR] = tswapreg(env->sregs[SAR]); 1691 (*regs)[TARGET_REG_WINDOWSTART] = tswapreg(env->sregs[WINDOW_START]); 1692 (*regs)[TARGET_REG_WINDOWBASE] = tswapreg(env->sregs[WINDOW_BASE]); 1693 (*regs)[TARGET_REG_THREADPTR] = tswapreg(env->uregs[THREADPTR]); 1694 xtensa_sync_phys_from_window((CPUXtensaState *)env); 1695 for (i = 0; i < env->config->nareg; ++i) { 1696 (*regs)[TARGET_REG_AR0 + i] = tswapreg(env->phys_regs[i]); 1697 } 1698 } 1699 1700 #define USE_ELF_CORE_DUMP 1701 #define ELF_EXEC_PAGESIZE 4096 1702 1703 #endif /* TARGET_XTENSA */ 1704 1705 #ifdef TARGET_HEXAGON 1706 1707 #define ELF_START_MMAP 0x20000000 1708 1709 #define ELF_CLASS ELFCLASS32 1710 #define ELF_ARCH EM_HEXAGON 1711 1712 static inline void init_thread(struct target_pt_regs *regs, 1713 struct image_info *infop) 1714 { 1715 regs->sepc = infop->entry; 1716 regs->sp = infop->start_stack; 1717 } 1718 1719 #endif /* TARGET_HEXAGON */ 1720 1721 #ifndef ELF_PLATFORM 1722 #define ELF_PLATFORM (NULL) 1723 #endif 1724 1725 #ifndef ELF_MACHINE 1726 #define ELF_MACHINE ELF_ARCH 1727 #endif 1728 1729 #ifndef elf_check_arch 1730 #define elf_check_arch(x) ((x) == ELF_ARCH) 1731 #endif 1732 1733 #ifndef elf_check_abi 1734 #define elf_check_abi(x) (1) 1735 #endif 1736 1737 #ifndef ELF_HWCAP 1738 #define ELF_HWCAP 0 1739 #endif 1740 1741 #ifndef STACK_GROWS_DOWN 1742 #define STACK_GROWS_DOWN 1 1743 #endif 1744 1745 #ifndef STACK_ALIGNMENT 1746 #define STACK_ALIGNMENT 16 1747 #endif 1748 1749 #ifdef TARGET_ABI32 1750 #undef ELF_CLASS 1751 #define ELF_CLASS ELFCLASS32 1752 #undef bswaptls 1753 #define bswaptls(ptr) bswap32s(ptr) 1754 #endif 1755 1756 #include "elf.h" 1757 1758 /* We must delay the following stanzas until after "elf.h". */ 1759 #if defined(TARGET_AARCH64) 1760 1761 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz, 1762 const uint32_t *data, 1763 struct image_info *info, 1764 Error **errp) 1765 { 1766 if (pr_type == GNU_PROPERTY_AARCH64_FEATURE_1_AND) { 1767 if (pr_datasz != sizeof(uint32_t)) { 1768 error_setg(errp, "Ill-formed GNU_PROPERTY_AARCH64_FEATURE_1_AND"); 1769 return false; 1770 } 1771 /* We will extract GNU_PROPERTY_AARCH64_FEATURE_1_BTI later. */ 1772 info->note_flags = *data; 1773 } 1774 return true; 1775 } 1776 #define ARCH_USE_GNU_PROPERTY 1 1777 1778 #else 1779 1780 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz, 1781 const uint32_t *data, 1782 struct image_info *info, 1783 Error **errp) 1784 { 1785 g_assert_not_reached(); 1786 } 1787 #define ARCH_USE_GNU_PROPERTY 0 1788 1789 #endif 1790 1791 struct exec 1792 { 1793 unsigned int a_info; /* Use macros N_MAGIC, etc for access */ 1794 unsigned int a_text; /* length of text, in bytes */ 1795 unsigned int a_data; /* length of data, in bytes */ 1796 unsigned int a_bss; /* length of uninitialized data area, in bytes */ 1797 unsigned int a_syms; /* length of symbol table data in file, in bytes */ 1798 unsigned int a_entry; /* start address */ 1799 unsigned int a_trsize; /* length of relocation info for text, in bytes */ 1800 unsigned int a_drsize; /* length of relocation info for data, in bytes */ 1801 }; 1802 1803 1804 #define N_MAGIC(exec) ((exec).a_info & 0xffff) 1805 #define OMAGIC 0407 1806 #define NMAGIC 0410 1807 #define ZMAGIC 0413 1808 #define QMAGIC 0314 1809 1810 /* Necessary parameters */ 1811 #define TARGET_ELF_EXEC_PAGESIZE \ 1812 (((eppnt->p_align & ~qemu_host_page_mask) != 0) ? \ 1813 TARGET_PAGE_SIZE : MAX(qemu_host_page_size, TARGET_PAGE_SIZE)) 1814 #define TARGET_ELF_PAGELENGTH(_v) ROUND_UP((_v), TARGET_ELF_EXEC_PAGESIZE) 1815 #define TARGET_ELF_PAGESTART(_v) ((_v) & \ 1816 ~(abi_ulong)(TARGET_ELF_EXEC_PAGESIZE-1)) 1817 #define TARGET_ELF_PAGEOFFSET(_v) ((_v) & (TARGET_ELF_EXEC_PAGESIZE-1)) 1818 1819 #define DLINFO_ITEMS 16 1820 1821 static inline void memcpy_fromfs(void * to, const void * from, unsigned long n) 1822 { 1823 memcpy(to, from, n); 1824 } 1825 1826 #ifdef BSWAP_NEEDED 1827 static void bswap_ehdr(struct elfhdr *ehdr) 1828 { 1829 bswap16s(&ehdr->e_type); /* Object file type */ 1830 bswap16s(&ehdr->e_machine); /* Architecture */ 1831 bswap32s(&ehdr->e_version); /* Object file version */ 1832 bswaptls(&ehdr->e_entry); /* Entry point virtual address */ 1833 bswaptls(&ehdr->e_phoff); /* Program header table file offset */ 1834 bswaptls(&ehdr->e_shoff); /* Section header table file offset */ 1835 bswap32s(&ehdr->e_flags); /* Processor-specific flags */ 1836 bswap16s(&ehdr->e_ehsize); /* ELF header size in bytes */ 1837 bswap16s(&ehdr->e_phentsize); /* Program header table entry size */ 1838 bswap16s(&ehdr->e_phnum); /* Program header table entry count */ 1839 bswap16s(&ehdr->e_shentsize); /* Section header table entry size */ 1840 bswap16s(&ehdr->e_shnum); /* Section header table entry count */ 1841 bswap16s(&ehdr->e_shstrndx); /* Section header string table index */ 1842 } 1843 1844 static void bswap_phdr(struct elf_phdr *phdr, int phnum) 1845 { 1846 int i; 1847 for (i = 0; i < phnum; ++i, ++phdr) { 1848 bswap32s(&phdr->p_type); /* Segment type */ 1849 bswap32s(&phdr->p_flags); /* Segment flags */ 1850 bswaptls(&phdr->p_offset); /* Segment file offset */ 1851 bswaptls(&phdr->p_vaddr); /* Segment virtual address */ 1852 bswaptls(&phdr->p_paddr); /* Segment physical address */ 1853 bswaptls(&phdr->p_filesz); /* Segment size in file */ 1854 bswaptls(&phdr->p_memsz); /* Segment size in memory */ 1855 bswaptls(&phdr->p_align); /* Segment alignment */ 1856 } 1857 } 1858 1859 static void bswap_shdr(struct elf_shdr *shdr, int shnum) 1860 { 1861 int i; 1862 for (i = 0; i < shnum; ++i, ++shdr) { 1863 bswap32s(&shdr->sh_name); 1864 bswap32s(&shdr->sh_type); 1865 bswaptls(&shdr->sh_flags); 1866 bswaptls(&shdr->sh_addr); 1867 bswaptls(&shdr->sh_offset); 1868 bswaptls(&shdr->sh_size); 1869 bswap32s(&shdr->sh_link); 1870 bswap32s(&shdr->sh_info); 1871 bswaptls(&shdr->sh_addralign); 1872 bswaptls(&shdr->sh_entsize); 1873 } 1874 } 1875 1876 static void bswap_sym(struct elf_sym *sym) 1877 { 1878 bswap32s(&sym->st_name); 1879 bswaptls(&sym->st_value); 1880 bswaptls(&sym->st_size); 1881 bswap16s(&sym->st_shndx); 1882 } 1883 1884 #ifdef TARGET_MIPS 1885 static void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags) 1886 { 1887 bswap16s(&abiflags->version); 1888 bswap32s(&abiflags->ases); 1889 bswap32s(&abiflags->isa_ext); 1890 bswap32s(&abiflags->flags1); 1891 bswap32s(&abiflags->flags2); 1892 } 1893 #endif 1894 #else 1895 static inline void bswap_ehdr(struct elfhdr *ehdr) { } 1896 static inline void bswap_phdr(struct elf_phdr *phdr, int phnum) { } 1897 static inline void bswap_shdr(struct elf_shdr *shdr, int shnum) { } 1898 static inline void bswap_sym(struct elf_sym *sym) { } 1899 #ifdef TARGET_MIPS 1900 static inline void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags) { } 1901 #endif 1902 #endif 1903 1904 #ifdef USE_ELF_CORE_DUMP 1905 static int elf_core_dump(int, const CPUArchState *); 1906 #endif /* USE_ELF_CORE_DUMP */ 1907 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias); 1908 1909 /* Verify the portions of EHDR within E_IDENT for the target. 1910 This can be performed before bswapping the entire header. */ 1911 static bool elf_check_ident(struct elfhdr *ehdr) 1912 { 1913 return (ehdr->e_ident[EI_MAG0] == ELFMAG0 1914 && ehdr->e_ident[EI_MAG1] == ELFMAG1 1915 && ehdr->e_ident[EI_MAG2] == ELFMAG2 1916 && ehdr->e_ident[EI_MAG3] == ELFMAG3 1917 && ehdr->e_ident[EI_CLASS] == ELF_CLASS 1918 && ehdr->e_ident[EI_DATA] == ELF_DATA 1919 && ehdr->e_ident[EI_VERSION] == EV_CURRENT); 1920 } 1921 1922 /* Verify the portions of EHDR outside of E_IDENT for the target. 1923 This has to wait until after bswapping the header. */ 1924 static bool elf_check_ehdr(struct elfhdr *ehdr) 1925 { 1926 return (elf_check_arch(ehdr->e_machine) 1927 && elf_check_abi(ehdr->e_flags) 1928 && ehdr->e_ehsize == sizeof(struct elfhdr) 1929 && ehdr->e_phentsize == sizeof(struct elf_phdr) 1930 && (ehdr->e_type == ET_EXEC || ehdr->e_type == ET_DYN)); 1931 } 1932 1933 /* 1934 * 'copy_elf_strings()' copies argument/envelope strings from user 1935 * memory to free pages in kernel mem. These are in a format ready 1936 * to be put directly into the top of new user memory. 1937 * 1938 */ 1939 static abi_ulong copy_elf_strings(int argc, char **argv, char *scratch, 1940 abi_ulong p, abi_ulong stack_limit) 1941 { 1942 char *tmp; 1943 int len, i; 1944 abi_ulong top = p; 1945 1946 if (!p) { 1947 return 0; /* bullet-proofing */ 1948 } 1949 1950 if (STACK_GROWS_DOWN) { 1951 int offset = ((p - 1) % TARGET_PAGE_SIZE) + 1; 1952 for (i = argc - 1; i >= 0; --i) { 1953 tmp = argv[i]; 1954 if (!tmp) { 1955 fprintf(stderr, "VFS: argc is wrong"); 1956 exit(-1); 1957 } 1958 len = strlen(tmp) + 1; 1959 tmp += len; 1960 1961 if (len > (p - stack_limit)) { 1962 return 0; 1963 } 1964 while (len) { 1965 int bytes_to_copy = (len > offset) ? offset : len; 1966 tmp -= bytes_to_copy; 1967 p -= bytes_to_copy; 1968 offset -= bytes_to_copy; 1969 len -= bytes_to_copy; 1970 1971 memcpy_fromfs(scratch + offset, tmp, bytes_to_copy); 1972 1973 if (offset == 0) { 1974 memcpy_to_target(p, scratch, top - p); 1975 top = p; 1976 offset = TARGET_PAGE_SIZE; 1977 } 1978 } 1979 } 1980 if (p != top) { 1981 memcpy_to_target(p, scratch + offset, top - p); 1982 } 1983 } else { 1984 int remaining = TARGET_PAGE_SIZE - (p % TARGET_PAGE_SIZE); 1985 for (i = 0; i < argc; ++i) { 1986 tmp = argv[i]; 1987 if (!tmp) { 1988 fprintf(stderr, "VFS: argc is wrong"); 1989 exit(-1); 1990 } 1991 len = strlen(tmp) + 1; 1992 if (len > (stack_limit - p)) { 1993 return 0; 1994 } 1995 while (len) { 1996 int bytes_to_copy = (len > remaining) ? remaining : len; 1997 1998 memcpy_fromfs(scratch + (p - top), tmp, bytes_to_copy); 1999 2000 tmp += bytes_to_copy; 2001 remaining -= bytes_to_copy; 2002 p += bytes_to_copy; 2003 len -= bytes_to_copy; 2004 2005 if (remaining == 0) { 2006 memcpy_to_target(top, scratch, p - top); 2007 top = p; 2008 remaining = TARGET_PAGE_SIZE; 2009 } 2010 } 2011 } 2012 if (p != top) { 2013 memcpy_to_target(top, scratch, p - top); 2014 } 2015 } 2016 2017 return p; 2018 } 2019 2020 /* Older linux kernels provide up to MAX_ARG_PAGES (default: 32) of 2021 * argument/environment space. Newer kernels (>2.6.33) allow more, 2022 * dependent on stack size, but guarantee at least 32 pages for 2023 * backwards compatibility. 2024 */ 2025 #define STACK_LOWER_LIMIT (32 * TARGET_PAGE_SIZE) 2026 2027 static abi_ulong setup_arg_pages(struct linux_binprm *bprm, 2028 struct image_info *info) 2029 { 2030 abi_ulong size, error, guard; 2031 2032 size = guest_stack_size; 2033 if (size < STACK_LOWER_LIMIT) { 2034 size = STACK_LOWER_LIMIT; 2035 } 2036 guard = TARGET_PAGE_SIZE; 2037 if (guard < qemu_real_host_page_size()) { 2038 guard = qemu_real_host_page_size(); 2039 } 2040 2041 error = target_mmap(0, size + guard, PROT_READ | PROT_WRITE, 2042 MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); 2043 if (error == -1) { 2044 perror("mmap stack"); 2045 exit(-1); 2046 } 2047 2048 /* We reserve one extra page at the top of the stack as guard. */ 2049 if (STACK_GROWS_DOWN) { 2050 target_mprotect(error, guard, PROT_NONE); 2051 info->stack_limit = error + guard; 2052 return info->stack_limit + size - sizeof(void *); 2053 } else { 2054 target_mprotect(error + size, guard, PROT_NONE); 2055 info->stack_limit = error + size; 2056 return error; 2057 } 2058 } 2059 2060 /* Map and zero the bss. We need to explicitly zero any fractional pages 2061 after the data section (i.e. bss). */ 2062 static void zero_bss(abi_ulong elf_bss, abi_ulong last_bss, int prot) 2063 { 2064 uintptr_t host_start, host_map_start, host_end; 2065 2066 last_bss = TARGET_PAGE_ALIGN(last_bss); 2067 2068 /* ??? There is confusion between qemu_real_host_page_size and 2069 qemu_host_page_size here and elsewhere in target_mmap, which 2070 may lead to the end of the data section mapping from the file 2071 not being mapped. At least there was an explicit test and 2072 comment for that here, suggesting that "the file size must 2073 be known". The comment probably pre-dates the introduction 2074 of the fstat system call in target_mmap which does in fact 2075 find out the size. What isn't clear is if the workaround 2076 here is still actually needed. For now, continue with it, 2077 but merge it with the "normal" mmap that would allocate the bss. */ 2078 2079 host_start = (uintptr_t) g2h_untagged(elf_bss); 2080 host_end = (uintptr_t) g2h_untagged(last_bss); 2081 host_map_start = REAL_HOST_PAGE_ALIGN(host_start); 2082 2083 if (host_map_start < host_end) { 2084 void *p = mmap((void *)host_map_start, host_end - host_map_start, 2085 prot, MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); 2086 if (p == MAP_FAILED) { 2087 perror("cannot mmap brk"); 2088 exit(-1); 2089 } 2090 } 2091 2092 /* Ensure that the bss page(s) are valid */ 2093 if ((page_get_flags(last_bss-1) & prot) != prot) { 2094 page_set_flags(elf_bss & TARGET_PAGE_MASK, last_bss, prot | PAGE_VALID); 2095 } 2096 2097 if (host_start < host_map_start) { 2098 memset((void *)host_start, 0, host_map_start - host_start); 2099 } 2100 } 2101 2102 #ifdef TARGET_ARM 2103 static int elf_is_fdpic(struct elfhdr *exec) 2104 { 2105 return exec->e_ident[EI_OSABI] == ELFOSABI_ARM_FDPIC; 2106 } 2107 #else 2108 /* Default implementation, always false. */ 2109 static int elf_is_fdpic(struct elfhdr *exec) 2110 { 2111 return 0; 2112 } 2113 #endif 2114 2115 static abi_ulong loader_build_fdpic_loadmap(struct image_info *info, abi_ulong sp) 2116 { 2117 uint16_t n; 2118 struct elf32_fdpic_loadseg *loadsegs = info->loadsegs; 2119 2120 /* elf32_fdpic_loadseg */ 2121 n = info->nsegs; 2122 while (n--) { 2123 sp -= 12; 2124 put_user_u32(loadsegs[n].addr, sp+0); 2125 put_user_u32(loadsegs[n].p_vaddr, sp+4); 2126 put_user_u32(loadsegs[n].p_memsz, sp+8); 2127 } 2128 2129 /* elf32_fdpic_loadmap */ 2130 sp -= 4; 2131 put_user_u16(0, sp+0); /* version */ 2132 put_user_u16(info->nsegs, sp+2); /* nsegs */ 2133 2134 info->personality = PER_LINUX_FDPIC; 2135 info->loadmap_addr = sp; 2136 2137 return sp; 2138 } 2139 2140 static abi_ulong create_elf_tables(abi_ulong p, int argc, int envc, 2141 struct elfhdr *exec, 2142 struct image_info *info, 2143 struct image_info *interp_info) 2144 { 2145 abi_ulong sp; 2146 abi_ulong u_argc, u_argv, u_envp, u_auxv; 2147 int size; 2148 int i; 2149 abi_ulong u_rand_bytes; 2150 uint8_t k_rand_bytes[16]; 2151 abi_ulong u_platform; 2152 const char *k_platform; 2153 const int n = sizeof(elf_addr_t); 2154 2155 sp = p; 2156 2157 /* Needs to be before we load the env/argc/... */ 2158 if (elf_is_fdpic(exec)) { 2159 /* Need 4 byte alignment for these structs */ 2160 sp &= ~3; 2161 sp = loader_build_fdpic_loadmap(info, sp); 2162 info->other_info = interp_info; 2163 if (interp_info) { 2164 interp_info->other_info = info; 2165 sp = loader_build_fdpic_loadmap(interp_info, sp); 2166 info->interpreter_loadmap_addr = interp_info->loadmap_addr; 2167 info->interpreter_pt_dynamic_addr = interp_info->pt_dynamic_addr; 2168 } else { 2169 info->interpreter_loadmap_addr = 0; 2170 info->interpreter_pt_dynamic_addr = 0; 2171 } 2172 } 2173 2174 u_platform = 0; 2175 k_platform = ELF_PLATFORM; 2176 if (k_platform) { 2177 size_t len = strlen(k_platform) + 1; 2178 if (STACK_GROWS_DOWN) { 2179 sp -= (len + n - 1) & ~(n - 1); 2180 u_platform = sp; 2181 /* FIXME - check return value of memcpy_to_target() for failure */ 2182 memcpy_to_target(sp, k_platform, len); 2183 } else { 2184 memcpy_to_target(sp, k_platform, len); 2185 u_platform = sp; 2186 sp += len + 1; 2187 } 2188 } 2189 2190 /* Provide 16 byte alignment for the PRNG, and basic alignment for 2191 * the argv and envp pointers. 2192 */ 2193 if (STACK_GROWS_DOWN) { 2194 sp = QEMU_ALIGN_DOWN(sp, 16); 2195 } else { 2196 sp = QEMU_ALIGN_UP(sp, 16); 2197 } 2198 2199 /* 2200 * Generate 16 random bytes for userspace PRNG seeding. 2201 */ 2202 qemu_guest_getrandom_nofail(k_rand_bytes, sizeof(k_rand_bytes)); 2203 if (STACK_GROWS_DOWN) { 2204 sp -= 16; 2205 u_rand_bytes = sp; 2206 /* FIXME - check return value of memcpy_to_target() for failure */ 2207 memcpy_to_target(sp, k_rand_bytes, 16); 2208 } else { 2209 memcpy_to_target(sp, k_rand_bytes, 16); 2210 u_rand_bytes = sp; 2211 sp += 16; 2212 } 2213 2214 size = (DLINFO_ITEMS + 1) * 2; 2215 if (k_platform) 2216 size += 2; 2217 #ifdef DLINFO_ARCH_ITEMS 2218 size += DLINFO_ARCH_ITEMS * 2; 2219 #endif 2220 #ifdef ELF_HWCAP2 2221 size += 2; 2222 #endif 2223 info->auxv_len = size * n; 2224 2225 size += envc + argc + 2; 2226 size += 1; /* argc itself */ 2227 size *= n; 2228 2229 /* Allocate space and finalize stack alignment for entry now. */ 2230 if (STACK_GROWS_DOWN) { 2231 u_argc = QEMU_ALIGN_DOWN(sp - size, STACK_ALIGNMENT); 2232 sp = u_argc; 2233 } else { 2234 u_argc = sp; 2235 sp = QEMU_ALIGN_UP(sp + size, STACK_ALIGNMENT); 2236 } 2237 2238 u_argv = u_argc + n; 2239 u_envp = u_argv + (argc + 1) * n; 2240 u_auxv = u_envp + (envc + 1) * n; 2241 info->saved_auxv = u_auxv; 2242 info->argc = argc; 2243 info->envc = envc; 2244 info->argv = u_argv; 2245 info->envp = u_envp; 2246 2247 /* This is correct because Linux defines 2248 * elf_addr_t as Elf32_Off / Elf64_Off 2249 */ 2250 #define NEW_AUX_ENT(id, val) do { \ 2251 put_user_ual(id, u_auxv); u_auxv += n; \ 2252 put_user_ual(val, u_auxv); u_auxv += n; \ 2253 } while(0) 2254 2255 #ifdef ARCH_DLINFO 2256 /* 2257 * ARCH_DLINFO must come first so platform specific code can enforce 2258 * special alignment requirements on the AUXV if necessary (eg. PPC). 2259 */ 2260 ARCH_DLINFO; 2261 #endif 2262 /* There must be exactly DLINFO_ITEMS entries here, or the assert 2263 * on info->auxv_len will trigger. 2264 */ 2265 NEW_AUX_ENT(AT_PHDR, (abi_ulong)(info->load_addr + exec->e_phoff)); 2266 NEW_AUX_ENT(AT_PHENT, (abi_ulong)(sizeof (struct elf_phdr))); 2267 NEW_AUX_ENT(AT_PHNUM, (abi_ulong)(exec->e_phnum)); 2268 if ((info->alignment & ~qemu_host_page_mask) != 0) { 2269 /* Target doesn't support host page size alignment */ 2270 NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(TARGET_PAGE_SIZE)); 2271 } else { 2272 NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(MAX(TARGET_PAGE_SIZE, 2273 qemu_host_page_size))); 2274 } 2275 NEW_AUX_ENT(AT_BASE, (abi_ulong)(interp_info ? interp_info->load_addr : 0)); 2276 NEW_AUX_ENT(AT_FLAGS, (abi_ulong)0); 2277 NEW_AUX_ENT(AT_ENTRY, info->entry); 2278 NEW_AUX_ENT(AT_UID, (abi_ulong) getuid()); 2279 NEW_AUX_ENT(AT_EUID, (abi_ulong) geteuid()); 2280 NEW_AUX_ENT(AT_GID, (abi_ulong) getgid()); 2281 NEW_AUX_ENT(AT_EGID, (abi_ulong) getegid()); 2282 NEW_AUX_ENT(AT_HWCAP, (abi_ulong) ELF_HWCAP); 2283 NEW_AUX_ENT(AT_CLKTCK, (abi_ulong) sysconf(_SC_CLK_TCK)); 2284 NEW_AUX_ENT(AT_RANDOM, (abi_ulong) u_rand_bytes); 2285 NEW_AUX_ENT(AT_SECURE, (abi_ulong) qemu_getauxval(AT_SECURE)); 2286 NEW_AUX_ENT(AT_EXECFN, info->file_string); 2287 2288 #ifdef ELF_HWCAP2 2289 NEW_AUX_ENT(AT_HWCAP2, (abi_ulong) ELF_HWCAP2); 2290 #endif 2291 2292 if (u_platform) { 2293 NEW_AUX_ENT(AT_PLATFORM, u_platform); 2294 } 2295 NEW_AUX_ENT (AT_NULL, 0); 2296 #undef NEW_AUX_ENT 2297 2298 /* Check that our initial calculation of the auxv length matches how much 2299 * we actually put into it. 2300 */ 2301 assert(info->auxv_len == u_auxv - info->saved_auxv); 2302 2303 put_user_ual(argc, u_argc); 2304 2305 p = info->arg_strings; 2306 for (i = 0; i < argc; ++i) { 2307 put_user_ual(p, u_argv); 2308 u_argv += n; 2309 p += target_strlen(p) + 1; 2310 } 2311 put_user_ual(0, u_argv); 2312 2313 p = info->env_strings; 2314 for (i = 0; i < envc; ++i) { 2315 put_user_ual(p, u_envp); 2316 u_envp += n; 2317 p += target_strlen(p) + 1; 2318 } 2319 put_user_ual(0, u_envp); 2320 2321 return sp; 2322 } 2323 2324 #if defined(HI_COMMPAGE) 2325 #define LO_COMMPAGE 0 2326 #elif defined(LO_COMMPAGE) 2327 #define HI_COMMPAGE 0 2328 #else 2329 #define HI_COMMPAGE 0 2330 #define LO_COMMPAGE 0 2331 #define init_guest_commpage() true 2332 #endif 2333 2334 static void pgb_fail_in_use(const char *image_name) 2335 { 2336 error_report("%s: requires virtual address space that is in use " 2337 "(omit the -B option or choose a different value)", 2338 image_name); 2339 exit(EXIT_FAILURE); 2340 } 2341 2342 static void pgb_have_guest_base(const char *image_name, abi_ulong guest_loaddr, 2343 abi_ulong guest_hiaddr, long align) 2344 { 2345 const int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE; 2346 void *addr, *test; 2347 2348 if (!QEMU_IS_ALIGNED(guest_base, align)) { 2349 fprintf(stderr, "Requested guest base %p does not satisfy " 2350 "host minimum alignment (0x%lx)\n", 2351 (void *)guest_base, align); 2352 exit(EXIT_FAILURE); 2353 } 2354 2355 /* Sanity check the guest binary. */ 2356 if (reserved_va) { 2357 if (guest_hiaddr > reserved_va) { 2358 error_report("%s: requires more than reserved virtual " 2359 "address space (0x%" PRIx64 " > 0x%lx)", 2360 image_name, (uint64_t)guest_hiaddr, reserved_va); 2361 exit(EXIT_FAILURE); 2362 } 2363 } else { 2364 #if HOST_LONG_BITS < TARGET_ABI_BITS 2365 if ((guest_hiaddr - guest_base) > ~(uintptr_t)0) { 2366 error_report("%s: requires more virtual address space " 2367 "than the host can provide (0x%" PRIx64 ")", 2368 image_name, (uint64_t)guest_hiaddr - guest_base); 2369 exit(EXIT_FAILURE); 2370 } 2371 #endif 2372 } 2373 2374 /* 2375 * Expand the allocation to the entire reserved_va. 2376 * Exclude the mmap_min_addr hole. 2377 */ 2378 if (reserved_va) { 2379 guest_loaddr = (guest_base >= mmap_min_addr ? 0 2380 : mmap_min_addr - guest_base); 2381 guest_hiaddr = reserved_va; 2382 } 2383 2384 /* Reserve the address space for the binary, or reserved_va. */ 2385 test = g2h_untagged(guest_loaddr); 2386 addr = mmap(test, guest_hiaddr - guest_loaddr, PROT_NONE, flags, -1, 0); 2387 if (test != addr) { 2388 pgb_fail_in_use(image_name); 2389 } 2390 qemu_log_mask(CPU_LOG_PAGE, 2391 "%s: base @ %p for " TARGET_ABI_FMT_ld " bytes\n", 2392 __func__, addr, guest_hiaddr - guest_loaddr); 2393 } 2394 2395 /** 2396 * pgd_find_hole_fallback: potential mmap address 2397 * @guest_size: size of available space 2398 * @brk: location of break 2399 * @align: memory alignment 2400 * 2401 * This is a fallback method for finding a hole in the host address 2402 * space if we don't have the benefit of being able to access 2403 * /proc/self/map. It can potentially take a very long time as we can 2404 * only dumbly iterate up the host address space seeing if the 2405 * allocation would work. 2406 */ 2407 static uintptr_t pgd_find_hole_fallback(uintptr_t guest_size, uintptr_t brk, 2408 long align, uintptr_t offset) 2409 { 2410 uintptr_t base; 2411 2412 /* Start (aligned) at the bottom and work our way up */ 2413 base = ROUND_UP(mmap_min_addr, align); 2414 2415 while (true) { 2416 uintptr_t align_start, end; 2417 align_start = ROUND_UP(base, align); 2418 end = align_start + guest_size + offset; 2419 2420 /* if brk is anywhere in the range give ourselves some room to grow. */ 2421 if (align_start <= brk && brk < end) { 2422 base = brk + (16 * MiB); 2423 continue; 2424 } else if (align_start + guest_size < align_start) { 2425 /* we have run out of space */ 2426 return -1; 2427 } else { 2428 int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE | 2429 MAP_FIXED_NOREPLACE; 2430 void * mmap_start = mmap((void *) align_start, guest_size, 2431 PROT_NONE, flags, -1, 0); 2432 if (mmap_start != MAP_FAILED) { 2433 munmap(mmap_start, guest_size); 2434 if (mmap_start == (void *) align_start) { 2435 qemu_log_mask(CPU_LOG_PAGE, 2436 "%s: base @ %p for %" PRIdPTR" bytes\n", 2437 __func__, mmap_start + offset, guest_size); 2438 return (uintptr_t) mmap_start + offset; 2439 } 2440 } 2441 base += qemu_host_page_size; 2442 } 2443 } 2444 } 2445 2446 /* Return value for guest_base, or -1 if no hole found. */ 2447 static uintptr_t pgb_find_hole(uintptr_t guest_loaddr, uintptr_t guest_size, 2448 long align, uintptr_t offset) 2449 { 2450 GSList *maps, *iter; 2451 uintptr_t this_start, this_end, next_start, brk; 2452 intptr_t ret = -1; 2453 2454 assert(QEMU_IS_ALIGNED(guest_loaddr, align)); 2455 2456 maps = read_self_maps(); 2457 2458 /* Read brk after we've read the maps, which will malloc. */ 2459 brk = (uintptr_t)sbrk(0); 2460 2461 if (!maps) { 2462 return pgd_find_hole_fallback(guest_size, brk, align, offset); 2463 } 2464 2465 /* The first hole is before the first map entry. */ 2466 this_start = mmap_min_addr; 2467 2468 for (iter = maps; iter; 2469 this_start = next_start, iter = g_slist_next(iter)) { 2470 uintptr_t align_start, hole_size; 2471 2472 this_end = ((MapInfo *)iter->data)->start; 2473 next_start = ((MapInfo *)iter->data)->end; 2474 align_start = ROUND_UP(this_start + offset, align); 2475 2476 /* Skip holes that are too small. */ 2477 if (align_start >= this_end) { 2478 continue; 2479 } 2480 hole_size = this_end - align_start; 2481 if (hole_size < guest_size) { 2482 continue; 2483 } 2484 2485 /* If this hole contains brk, give ourselves some room to grow. */ 2486 if (this_start <= brk && brk < this_end) { 2487 hole_size -= guest_size; 2488 if (sizeof(uintptr_t) == 8 && hole_size >= 1 * GiB) { 2489 align_start += 1 * GiB; 2490 } else if (hole_size >= 16 * MiB) { 2491 align_start += 16 * MiB; 2492 } else { 2493 align_start = (this_end - guest_size) & -align; 2494 if (align_start < this_start) { 2495 continue; 2496 } 2497 } 2498 } 2499 2500 /* Record the lowest successful match. */ 2501 if (ret < 0) { 2502 ret = align_start; 2503 } 2504 /* If this hole contains the identity map, select it. */ 2505 if (align_start <= guest_loaddr && 2506 guest_loaddr + guest_size <= this_end) { 2507 ret = 0; 2508 } 2509 /* If this hole ends above the identity map, stop looking. */ 2510 if (this_end >= guest_loaddr) { 2511 break; 2512 } 2513 } 2514 free_self_maps(maps); 2515 2516 if (ret != -1) { 2517 qemu_log_mask(CPU_LOG_PAGE, "%s: base @ %" PRIxPTR 2518 " for %" PRIuPTR " bytes\n", 2519 __func__, ret, guest_size); 2520 } 2521 2522 return ret; 2523 } 2524 2525 static void pgb_static(const char *image_name, abi_ulong orig_loaddr, 2526 abi_ulong orig_hiaddr, long align) 2527 { 2528 uintptr_t loaddr = orig_loaddr; 2529 uintptr_t hiaddr = orig_hiaddr; 2530 uintptr_t offset = 0; 2531 uintptr_t addr; 2532 2533 if (hiaddr != orig_hiaddr) { 2534 error_report("%s: requires virtual address space that the " 2535 "host cannot provide (0x%" PRIx64 ")", 2536 image_name, (uint64_t)orig_hiaddr); 2537 exit(EXIT_FAILURE); 2538 } 2539 2540 loaddr &= -align; 2541 if (HI_COMMPAGE) { 2542 /* 2543 * Extend the allocation to include the commpage. 2544 * For a 64-bit host, this is just 4GiB; for a 32-bit host we 2545 * need to ensure there is space bellow the guest_base so we 2546 * can map the commpage in the place needed when the address 2547 * arithmetic wraps around. 2548 */ 2549 if (sizeof(uintptr_t) == 8 || loaddr >= 0x80000000u) { 2550 hiaddr = (uintptr_t) 4 << 30; 2551 } else { 2552 offset = -(HI_COMMPAGE & -align); 2553 } 2554 } else if (LO_COMMPAGE != 0) { 2555 loaddr = MIN(loaddr, LO_COMMPAGE & -align); 2556 } 2557 2558 addr = pgb_find_hole(loaddr, hiaddr - loaddr, align, offset); 2559 if (addr == -1) { 2560 /* 2561 * If HI_COMMPAGE, there *might* be a non-consecutive allocation 2562 * that can satisfy both. But as the normal arm32 link base address 2563 * is ~32k, and we extend down to include the commpage, making the 2564 * overhead only ~96k, this is unlikely. 2565 */ 2566 error_report("%s: Unable to allocate %#zx bytes of " 2567 "virtual address space", image_name, 2568 (size_t)(hiaddr - loaddr)); 2569 exit(EXIT_FAILURE); 2570 } 2571 2572 guest_base = addr; 2573 2574 qemu_log_mask(CPU_LOG_PAGE, "%s: base @ %"PRIxPTR" for %" PRIuPTR" bytes\n", 2575 __func__, addr, hiaddr - loaddr); 2576 } 2577 2578 static void pgb_dynamic(const char *image_name, long align) 2579 { 2580 /* 2581 * The executable is dynamic and does not require a fixed address. 2582 * All we need is a commpage that satisfies align. 2583 * If we do not need a commpage, leave guest_base == 0. 2584 */ 2585 if (HI_COMMPAGE) { 2586 uintptr_t addr, commpage; 2587 2588 /* 64-bit hosts should have used reserved_va. */ 2589 assert(sizeof(uintptr_t) == 4); 2590 2591 /* 2592 * By putting the commpage at the first hole, that puts guest_base 2593 * just above that, and maximises the positive guest addresses. 2594 */ 2595 commpage = HI_COMMPAGE & -align; 2596 addr = pgb_find_hole(commpage, -commpage, align, 0); 2597 assert(addr != -1); 2598 guest_base = addr; 2599 } 2600 } 2601 2602 static void pgb_reserved_va(const char *image_name, abi_ulong guest_loaddr, 2603 abi_ulong guest_hiaddr, long align) 2604 { 2605 int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE; 2606 void *addr, *test; 2607 2608 if (guest_hiaddr > reserved_va) { 2609 error_report("%s: requires more than reserved virtual " 2610 "address space (0x%" PRIx64 " > 0x%lx)", 2611 image_name, (uint64_t)guest_hiaddr, reserved_va); 2612 exit(EXIT_FAILURE); 2613 } 2614 2615 /* Widen the "image" to the entire reserved address space. */ 2616 pgb_static(image_name, 0, reserved_va, align); 2617 2618 /* osdep.h defines this as 0 if it's missing */ 2619 flags |= MAP_FIXED_NOREPLACE; 2620 2621 /* Reserve the memory on the host. */ 2622 assert(guest_base != 0); 2623 test = g2h_untagged(0); 2624 addr = mmap(test, reserved_va, PROT_NONE, flags, -1, 0); 2625 if (addr == MAP_FAILED || addr != test) { 2626 error_report("Unable to reserve 0x%lx bytes of virtual address " 2627 "space at %p (%s) for use as guest address space (check your " 2628 "virtual memory ulimit setting, min_mmap_addr or reserve less " 2629 "using -R option)", reserved_va, test, strerror(errno)); 2630 exit(EXIT_FAILURE); 2631 } 2632 2633 qemu_log_mask(CPU_LOG_PAGE, "%s: base @ %p for %lu bytes\n", 2634 __func__, addr, reserved_va); 2635 } 2636 2637 void probe_guest_base(const char *image_name, abi_ulong guest_loaddr, 2638 abi_ulong guest_hiaddr) 2639 { 2640 /* In order to use host shmat, we must be able to honor SHMLBA. */ 2641 uintptr_t align = MAX(SHMLBA, qemu_host_page_size); 2642 2643 if (have_guest_base) { 2644 pgb_have_guest_base(image_name, guest_loaddr, guest_hiaddr, align); 2645 } else if (reserved_va) { 2646 pgb_reserved_va(image_name, guest_loaddr, guest_hiaddr, align); 2647 } else if (guest_loaddr) { 2648 pgb_static(image_name, guest_loaddr, guest_hiaddr, align); 2649 } else { 2650 pgb_dynamic(image_name, align); 2651 } 2652 2653 /* Reserve and initialize the commpage. */ 2654 if (!init_guest_commpage()) { 2655 /* 2656 * With have_guest_base, the user has selected the address and 2657 * we are trying to work with that. Otherwise, we have selected 2658 * free space and init_guest_commpage must succeeded. 2659 */ 2660 assert(have_guest_base); 2661 pgb_fail_in_use(image_name); 2662 } 2663 2664 assert(QEMU_IS_ALIGNED(guest_base, align)); 2665 qemu_log_mask(CPU_LOG_PAGE, "Locating guest address space " 2666 "@ 0x%" PRIx64 "\n", (uint64_t)guest_base); 2667 } 2668 2669 enum { 2670 /* The string "GNU\0" as a magic number. */ 2671 GNU0_MAGIC = const_le32('G' | 'N' << 8 | 'U' << 16), 2672 NOTE_DATA_SZ = 1 * KiB, 2673 NOTE_NAME_SZ = 4, 2674 ELF_GNU_PROPERTY_ALIGN = ELF_CLASS == ELFCLASS32 ? 4 : 8, 2675 }; 2676 2677 /* 2678 * Process a single gnu_property entry. 2679 * Return false for error. 2680 */ 2681 static bool parse_elf_property(const uint32_t *data, int *off, int datasz, 2682 struct image_info *info, bool have_prev_type, 2683 uint32_t *prev_type, Error **errp) 2684 { 2685 uint32_t pr_type, pr_datasz, step; 2686 2687 if (*off > datasz || !QEMU_IS_ALIGNED(*off, ELF_GNU_PROPERTY_ALIGN)) { 2688 goto error_data; 2689 } 2690 datasz -= *off; 2691 data += *off / sizeof(uint32_t); 2692 2693 if (datasz < 2 * sizeof(uint32_t)) { 2694 goto error_data; 2695 } 2696 pr_type = data[0]; 2697 pr_datasz = data[1]; 2698 data += 2; 2699 datasz -= 2 * sizeof(uint32_t); 2700 step = ROUND_UP(pr_datasz, ELF_GNU_PROPERTY_ALIGN); 2701 if (step > datasz) { 2702 goto error_data; 2703 } 2704 2705 /* Properties are supposed to be unique and sorted on pr_type. */ 2706 if (have_prev_type && pr_type <= *prev_type) { 2707 if (pr_type == *prev_type) { 2708 error_setg(errp, "Duplicate property in PT_GNU_PROPERTY"); 2709 } else { 2710 error_setg(errp, "Unsorted property in PT_GNU_PROPERTY"); 2711 } 2712 return false; 2713 } 2714 *prev_type = pr_type; 2715 2716 if (!arch_parse_elf_property(pr_type, pr_datasz, data, info, errp)) { 2717 return false; 2718 } 2719 2720 *off += 2 * sizeof(uint32_t) + step; 2721 return true; 2722 2723 error_data: 2724 error_setg(errp, "Ill-formed property in PT_GNU_PROPERTY"); 2725 return false; 2726 } 2727 2728 /* Process NT_GNU_PROPERTY_TYPE_0. */ 2729 static bool parse_elf_properties(int image_fd, 2730 struct image_info *info, 2731 const struct elf_phdr *phdr, 2732 char bprm_buf[BPRM_BUF_SIZE], 2733 Error **errp) 2734 { 2735 union { 2736 struct elf_note nhdr; 2737 uint32_t data[NOTE_DATA_SZ / sizeof(uint32_t)]; 2738 } note; 2739 2740 int n, off, datasz; 2741 bool have_prev_type; 2742 uint32_t prev_type; 2743 2744 /* Unless the arch requires properties, ignore them. */ 2745 if (!ARCH_USE_GNU_PROPERTY) { 2746 return true; 2747 } 2748 2749 /* If the properties are crazy large, that's too bad. */ 2750 n = phdr->p_filesz; 2751 if (n > sizeof(note)) { 2752 error_setg(errp, "PT_GNU_PROPERTY too large"); 2753 return false; 2754 } 2755 if (n < sizeof(note.nhdr)) { 2756 error_setg(errp, "PT_GNU_PROPERTY too small"); 2757 return false; 2758 } 2759 2760 if (phdr->p_offset + n <= BPRM_BUF_SIZE) { 2761 memcpy(¬e, bprm_buf + phdr->p_offset, n); 2762 } else { 2763 ssize_t len = pread(image_fd, ¬e, n, phdr->p_offset); 2764 if (len != n) { 2765 error_setg_errno(errp, errno, "Error reading file header"); 2766 return false; 2767 } 2768 } 2769 2770 /* 2771 * The contents of a valid PT_GNU_PROPERTY is a sequence 2772 * of uint32_t -- swap them all now. 2773 */ 2774 #ifdef BSWAP_NEEDED 2775 for (int i = 0; i < n / 4; i++) { 2776 bswap32s(note.data + i); 2777 } 2778 #endif 2779 2780 /* 2781 * Note that nhdr is 3 words, and that the "name" described by namesz 2782 * immediately follows nhdr and is thus at the 4th word. Further, all 2783 * of the inputs to the kernel's round_up are multiples of 4. 2784 */ 2785 if (note.nhdr.n_type != NT_GNU_PROPERTY_TYPE_0 || 2786 note.nhdr.n_namesz != NOTE_NAME_SZ || 2787 note.data[3] != GNU0_MAGIC) { 2788 error_setg(errp, "Invalid note in PT_GNU_PROPERTY"); 2789 return false; 2790 } 2791 off = sizeof(note.nhdr) + NOTE_NAME_SZ; 2792 2793 datasz = note.nhdr.n_descsz + off; 2794 if (datasz > n) { 2795 error_setg(errp, "Invalid note size in PT_GNU_PROPERTY"); 2796 return false; 2797 } 2798 2799 have_prev_type = false; 2800 prev_type = 0; 2801 while (1) { 2802 if (off == datasz) { 2803 return true; /* end, exit ok */ 2804 } 2805 if (!parse_elf_property(note.data, &off, datasz, info, 2806 have_prev_type, &prev_type, errp)) { 2807 return false; 2808 } 2809 have_prev_type = true; 2810 } 2811 } 2812 2813 /* Load an ELF image into the address space. 2814 2815 IMAGE_NAME is the filename of the image, to use in error messages. 2816 IMAGE_FD is the open file descriptor for the image. 2817 2818 BPRM_BUF is a copy of the beginning of the file; this of course 2819 contains the elf file header at offset 0. It is assumed that this 2820 buffer is sufficiently aligned to present no problems to the host 2821 in accessing data at aligned offsets within the buffer. 2822 2823 On return: INFO values will be filled in, as necessary or available. */ 2824 2825 static void load_elf_image(const char *image_name, int image_fd, 2826 struct image_info *info, char **pinterp_name, 2827 char bprm_buf[BPRM_BUF_SIZE]) 2828 { 2829 struct elfhdr *ehdr = (struct elfhdr *)bprm_buf; 2830 struct elf_phdr *phdr; 2831 abi_ulong load_addr, load_bias, loaddr, hiaddr, error; 2832 int i, retval, prot_exec; 2833 Error *err = NULL; 2834 2835 /* First of all, some simple consistency checks */ 2836 if (!elf_check_ident(ehdr)) { 2837 error_setg(&err, "Invalid ELF image for this architecture"); 2838 goto exit_errmsg; 2839 } 2840 bswap_ehdr(ehdr); 2841 if (!elf_check_ehdr(ehdr)) { 2842 error_setg(&err, "Invalid ELF image for this architecture"); 2843 goto exit_errmsg; 2844 } 2845 2846 i = ehdr->e_phnum * sizeof(struct elf_phdr); 2847 if (ehdr->e_phoff + i <= BPRM_BUF_SIZE) { 2848 phdr = (struct elf_phdr *)(bprm_buf + ehdr->e_phoff); 2849 } else { 2850 phdr = (struct elf_phdr *) alloca(i); 2851 retval = pread(image_fd, phdr, i, ehdr->e_phoff); 2852 if (retval != i) { 2853 goto exit_read; 2854 } 2855 } 2856 bswap_phdr(phdr, ehdr->e_phnum); 2857 2858 info->nsegs = 0; 2859 info->pt_dynamic_addr = 0; 2860 2861 mmap_lock(); 2862 2863 /* 2864 * Find the maximum size of the image and allocate an appropriate 2865 * amount of memory to handle that. Locate the interpreter, if any. 2866 */ 2867 loaddr = -1, hiaddr = 0; 2868 info->alignment = 0; 2869 for (i = 0; i < ehdr->e_phnum; ++i) { 2870 struct elf_phdr *eppnt = phdr + i; 2871 if (eppnt->p_type == PT_LOAD) { 2872 abi_ulong a = eppnt->p_vaddr - eppnt->p_offset; 2873 if (a < loaddr) { 2874 loaddr = a; 2875 } 2876 a = eppnt->p_vaddr + eppnt->p_memsz; 2877 if (a > hiaddr) { 2878 hiaddr = a; 2879 } 2880 ++info->nsegs; 2881 info->alignment |= eppnt->p_align; 2882 } else if (eppnt->p_type == PT_INTERP && pinterp_name) { 2883 g_autofree char *interp_name = NULL; 2884 2885 if (*pinterp_name) { 2886 error_setg(&err, "Multiple PT_INTERP entries"); 2887 goto exit_errmsg; 2888 } 2889 2890 interp_name = g_malloc(eppnt->p_filesz); 2891 2892 if (eppnt->p_offset + eppnt->p_filesz <= BPRM_BUF_SIZE) { 2893 memcpy(interp_name, bprm_buf + eppnt->p_offset, 2894 eppnt->p_filesz); 2895 } else { 2896 retval = pread(image_fd, interp_name, eppnt->p_filesz, 2897 eppnt->p_offset); 2898 if (retval != eppnt->p_filesz) { 2899 goto exit_read; 2900 } 2901 } 2902 if (interp_name[eppnt->p_filesz - 1] != 0) { 2903 error_setg(&err, "Invalid PT_INTERP entry"); 2904 goto exit_errmsg; 2905 } 2906 *pinterp_name = g_steal_pointer(&interp_name); 2907 } else if (eppnt->p_type == PT_GNU_PROPERTY) { 2908 if (!parse_elf_properties(image_fd, info, eppnt, bprm_buf, &err)) { 2909 goto exit_errmsg; 2910 } 2911 } 2912 } 2913 2914 if (pinterp_name != NULL) { 2915 /* 2916 * This is the main executable. 2917 * 2918 * Reserve extra space for brk. 2919 * We hold on to this space while placing the interpreter 2920 * and the stack, lest they be placed immediately after 2921 * the data segment and block allocation from the brk. 2922 * 2923 * 16MB is chosen as "large enough" without being so large as 2924 * to allow the result to not fit with a 32-bit guest on a 2925 * 32-bit host. However some 64 bit guests (e.g. s390x) 2926 * attempt to place their heap further ahead and currently 2927 * nothing stops them smashing into QEMUs address space. 2928 */ 2929 #if TARGET_LONG_BITS == 64 2930 info->reserve_brk = 32 * MiB; 2931 #else 2932 info->reserve_brk = 16 * MiB; 2933 #endif 2934 hiaddr += info->reserve_brk; 2935 2936 if (ehdr->e_type == ET_EXEC) { 2937 /* 2938 * Make sure that the low address does not conflict with 2939 * MMAP_MIN_ADDR or the QEMU application itself. 2940 */ 2941 probe_guest_base(image_name, loaddr, hiaddr); 2942 } else { 2943 /* 2944 * The binary is dynamic, but we still need to 2945 * select guest_base. In this case we pass a size. 2946 */ 2947 probe_guest_base(image_name, 0, hiaddr - loaddr); 2948 } 2949 } 2950 2951 /* 2952 * Reserve address space for all of this. 2953 * 2954 * In the case of ET_EXEC, we supply MAP_FIXED so that we get 2955 * exactly the address range that is required. 2956 * 2957 * Otherwise this is ET_DYN, and we are searching for a location 2958 * that can hold the memory space required. If the image is 2959 * pre-linked, LOADDR will be non-zero, and the kernel should 2960 * honor that address if it happens to be free. 2961 * 2962 * In both cases, we will overwrite pages in this range with mappings 2963 * from the executable. 2964 */ 2965 load_addr = target_mmap(loaddr, hiaddr - loaddr, PROT_NONE, 2966 MAP_PRIVATE | MAP_ANON | MAP_NORESERVE | 2967 (ehdr->e_type == ET_EXEC ? MAP_FIXED : 0), 2968 -1, 0); 2969 if (load_addr == -1) { 2970 goto exit_mmap; 2971 } 2972 load_bias = load_addr - loaddr; 2973 2974 if (elf_is_fdpic(ehdr)) { 2975 struct elf32_fdpic_loadseg *loadsegs = info->loadsegs = 2976 g_malloc(sizeof(*loadsegs) * info->nsegs); 2977 2978 for (i = 0; i < ehdr->e_phnum; ++i) { 2979 switch (phdr[i].p_type) { 2980 case PT_DYNAMIC: 2981 info->pt_dynamic_addr = phdr[i].p_vaddr + load_bias; 2982 break; 2983 case PT_LOAD: 2984 loadsegs->addr = phdr[i].p_vaddr + load_bias; 2985 loadsegs->p_vaddr = phdr[i].p_vaddr; 2986 loadsegs->p_memsz = phdr[i].p_memsz; 2987 ++loadsegs; 2988 break; 2989 } 2990 } 2991 } 2992 2993 info->load_bias = load_bias; 2994 info->code_offset = load_bias; 2995 info->data_offset = load_bias; 2996 info->load_addr = load_addr; 2997 info->entry = ehdr->e_entry + load_bias; 2998 info->start_code = -1; 2999 info->end_code = 0; 3000 info->start_data = -1; 3001 info->end_data = 0; 3002 info->brk = 0; 3003 info->elf_flags = ehdr->e_flags; 3004 3005 prot_exec = PROT_EXEC; 3006 #ifdef TARGET_AARCH64 3007 /* 3008 * If the BTI feature is present, this indicates that the executable 3009 * pages of the startup binary should be mapped with PROT_BTI, so that 3010 * branch targets are enforced. 3011 * 3012 * The startup binary is either the interpreter or the static executable. 3013 * The interpreter is responsible for all pages of a dynamic executable. 3014 * 3015 * Elf notes are backward compatible to older cpus. 3016 * Do not enable BTI unless it is supported. 3017 */ 3018 if ((info->note_flags & GNU_PROPERTY_AARCH64_FEATURE_1_BTI) 3019 && (pinterp_name == NULL || *pinterp_name == 0) 3020 && cpu_isar_feature(aa64_bti, ARM_CPU(thread_cpu))) { 3021 prot_exec |= TARGET_PROT_BTI; 3022 } 3023 #endif 3024 3025 for (i = 0; i < ehdr->e_phnum; i++) { 3026 struct elf_phdr *eppnt = phdr + i; 3027 if (eppnt->p_type == PT_LOAD) { 3028 abi_ulong vaddr, vaddr_po, vaddr_ps, vaddr_ef, vaddr_em, vaddr_len; 3029 int elf_prot = 0; 3030 3031 if (eppnt->p_flags & PF_R) { 3032 elf_prot |= PROT_READ; 3033 } 3034 if (eppnt->p_flags & PF_W) { 3035 elf_prot |= PROT_WRITE; 3036 } 3037 if (eppnt->p_flags & PF_X) { 3038 elf_prot |= prot_exec; 3039 } 3040 3041 vaddr = load_bias + eppnt->p_vaddr; 3042 vaddr_po = TARGET_ELF_PAGEOFFSET(vaddr); 3043 vaddr_ps = TARGET_ELF_PAGESTART(vaddr); 3044 3045 vaddr_ef = vaddr + eppnt->p_filesz; 3046 vaddr_em = vaddr + eppnt->p_memsz; 3047 3048 /* 3049 * Some segments may be completely empty, with a non-zero p_memsz 3050 * but no backing file segment. 3051 */ 3052 if (eppnt->p_filesz != 0) { 3053 vaddr_len = TARGET_ELF_PAGELENGTH(eppnt->p_filesz + vaddr_po); 3054 error = target_mmap(vaddr_ps, vaddr_len, elf_prot, 3055 MAP_PRIVATE | MAP_FIXED, 3056 image_fd, eppnt->p_offset - vaddr_po); 3057 3058 if (error == -1) { 3059 goto exit_mmap; 3060 } 3061 3062 /* 3063 * If the load segment requests extra zeros (e.g. bss), map it. 3064 */ 3065 if (eppnt->p_filesz < eppnt->p_memsz) { 3066 zero_bss(vaddr_ef, vaddr_em, elf_prot); 3067 } 3068 } else if (eppnt->p_memsz != 0) { 3069 vaddr_len = TARGET_ELF_PAGELENGTH(eppnt->p_memsz + vaddr_po); 3070 error = target_mmap(vaddr_ps, vaddr_len, elf_prot, 3071 MAP_PRIVATE | MAP_FIXED | MAP_ANONYMOUS, 3072 -1, 0); 3073 3074 if (error == -1) { 3075 goto exit_mmap; 3076 } 3077 } 3078 3079 /* Find the full program boundaries. */ 3080 if (elf_prot & PROT_EXEC) { 3081 if (vaddr < info->start_code) { 3082 info->start_code = vaddr; 3083 } 3084 if (vaddr_ef > info->end_code) { 3085 info->end_code = vaddr_ef; 3086 } 3087 } 3088 if (elf_prot & PROT_WRITE) { 3089 if (vaddr < info->start_data) { 3090 info->start_data = vaddr; 3091 } 3092 if (vaddr_ef > info->end_data) { 3093 info->end_data = vaddr_ef; 3094 } 3095 } 3096 if (vaddr_em > info->brk) { 3097 info->brk = vaddr_em; 3098 } 3099 #ifdef TARGET_MIPS 3100 } else if (eppnt->p_type == PT_MIPS_ABIFLAGS) { 3101 Mips_elf_abiflags_v0 abiflags; 3102 if (eppnt->p_filesz < sizeof(Mips_elf_abiflags_v0)) { 3103 error_setg(&err, "Invalid PT_MIPS_ABIFLAGS entry"); 3104 goto exit_errmsg; 3105 } 3106 if (eppnt->p_offset + eppnt->p_filesz <= BPRM_BUF_SIZE) { 3107 memcpy(&abiflags, bprm_buf + eppnt->p_offset, 3108 sizeof(Mips_elf_abiflags_v0)); 3109 } else { 3110 retval = pread(image_fd, &abiflags, sizeof(Mips_elf_abiflags_v0), 3111 eppnt->p_offset); 3112 if (retval != sizeof(Mips_elf_abiflags_v0)) { 3113 goto exit_read; 3114 } 3115 } 3116 bswap_mips_abiflags(&abiflags); 3117 info->fp_abi = abiflags.fp_abi; 3118 #endif 3119 } 3120 } 3121 3122 if (info->end_data == 0) { 3123 info->start_data = info->end_code; 3124 info->end_data = info->end_code; 3125 } 3126 3127 if (qemu_log_enabled()) { 3128 load_symbols(ehdr, image_fd, load_bias); 3129 } 3130 3131 mmap_unlock(); 3132 3133 close(image_fd); 3134 return; 3135 3136 exit_read: 3137 if (retval >= 0) { 3138 error_setg(&err, "Incomplete read of file header"); 3139 } else { 3140 error_setg_errno(&err, errno, "Error reading file header"); 3141 } 3142 goto exit_errmsg; 3143 exit_mmap: 3144 error_setg_errno(&err, errno, "Error mapping file"); 3145 goto exit_errmsg; 3146 exit_errmsg: 3147 error_reportf_err(err, "%s: ", image_name); 3148 exit(-1); 3149 } 3150 3151 static void load_elf_interp(const char *filename, struct image_info *info, 3152 char bprm_buf[BPRM_BUF_SIZE]) 3153 { 3154 int fd, retval; 3155 Error *err = NULL; 3156 3157 fd = open(path(filename), O_RDONLY); 3158 if (fd < 0) { 3159 error_setg_file_open(&err, errno, filename); 3160 error_report_err(err); 3161 exit(-1); 3162 } 3163 3164 retval = read(fd, bprm_buf, BPRM_BUF_SIZE); 3165 if (retval < 0) { 3166 error_setg_errno(&err, errno, "Error reading file header"); 3167 error_reportf_err(err, "%s: ", filename); 3168 exit(-1); 3169 } 3170 3171 if (retval < BPRM_BUF_SIZE) { 3172 memset(bprm_buf + retval, 0, BPRM_BUF_SIZE - retval); 3173 } 3174 3175 load_elf_image(filename, fd, info, NULL, bprm_buf); 3176 } 3177 3178 static int symfind(const void *s0, const void *s1) 3179 { 3180 target_ulong addr = *(target_ulong *)s0; 3181 struct elf_sym *sym = (struct elf_sym *)s1; 3182 int result = 0; 3183 if (addr < sym->st_value) { 3184 result = -1; 3185 } else if (addr >= sym->st_value + sym->st_size) { 3186 result = 1; 3187 } 3188 return result; 3189 } 3190 3191 static const char *lookup_symbolxx(struct syminfo *s, target_ulong orig_addr) 3192 { 3193 #if ELF_CLASS == ELFCLASS32 3194 struct elf_sym *syms = s->disas_symtab.elf32; 3195 #else 3196 struct elf_sym *syms = s->disas_symtab.elf64; 3197 #endif 3198 3199 // binary search 3200 struct elf_sym *sym; 3201 3202 sym = bsearch(&orig_addr, syms, s->disas_num_syms, sizeof(*syms), symfind); 3203 if (sym != NULL) { 3204 return s->disas_strtab + sym->st_name; 3205 } 3206 3207 return ""; 3208 } 3209 3210 /* FIXME: This should use elf_ops.h */ 3211 static int symcmp(const void *s0, const void *s1) 3212 { 3213 struct elf_sym *sym0 = (struct elf_sym *)s0; 3214 struct elf_sym *sym1 = (struct elf_sym *)s1; 3215 return (sym0->st_value < sym1->st_value) 3216 ? -1 3217 : ((sym0->st_value > sym1->st_value) ? 1 : 0); 3218 } 3219 3220 /* Best attempt to load symbols from this ELF object. */ 3221 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias) 3222 { 3223 int i, shnum, nsyms, sym_idx = 0, str_idx = 0; 3224 uint64_t segsz; 3225 struct elf_shdr *shdr; 3226 char *strings = NULL; 3227 struct syminfo *s = NULL; 3228 struct elf_sym *new_syms, *syms = NULL; 3229 3230 shnum = hdr->e_shnum; 3231 i = shnum * sizeof(struct elf_shdr); 3232 shdr = (struct elf_shdr *)alloca(i); 3233 if (pread(fd, shdr, i, hdr->e_shoff) != i) { 3234 return; 3235 } 3236 3237 bswap_shdr(shdr, shnum); 3238 for (i = 0; i < shnum; ++i) { 3239 if (shdr[i].sh_type == SHT_SYMTAB) { 3240 sym_idx = i; 3241 str_idx = shdr[i].sh_link; 3242 goto found; 3243 } 3244 } 3245 3246 /* There will be no symbol table if the file was stripped. */ 3247 return; 3248 3249 found: 3250 /* Now know where the strtab and symtab are. Snarf them. */ 3251 s = g_try_new(struct syminfo, 1); 3252 if (!s) { 3253 goto give_up; 3254 } 3255 3256 segsz = shdr[str_idx].sh_size; 3257 s->disas_strtab = strings = g_try_malloc(segsz); 3258 if (!strings || 3259 pread(fd, strings, segsz, shdr[str_idx].sh_offset) != segsz) { 3260 goto give_up; 3261 } 3262 3263 segsz = shdr[sym_idx].sh_size; 3264 syms = g_try_malloc(segsz); 3265 if (!syms || pread(fd, syms, segsz, shdr[sym_idx].sh_offset) != segsz) { 3266 goto give_up; 3267 } 3268 3269 if (segsz / sizeof(struct elf_sym) > INT_MAX) { 3270 /* Implausibly large symbol table: give up rather than ploughing 3271 * on with the number of symbols calculation overflowing 3272 */ 3273 goto give_up; 3274 } 3275 nsyms = segsz / sizeof(struct elf_sym); 3276 for (i = 0; i < nsyms; ) { 3277 bswap_sym(syms + i); 3278 /* Throw away entries which we do not need. */ 3279 if (syms[i].st_shndx == SHN_UNDEF 3280 || syms[i].st_shndx >= SHN_LORESERVE 3281 || ELF_ST_TYPE(syms[i].st_info) != STT_FUNC) { 3282 if (i < --nsyms) { 3283 syms[i] = syms[nsyms]; 3284 } 3285 } else { 3286 #if defined(TARGET_ARM) || defined (TARGET_MIPS) 3287 /* The bottom address bit marks a Thumb or MIPS16 symbol. */ 3288 syms[i].st_value &= ~(target_ulong)1; 3289 #endif 3290 syms[i].st_value += load_bias; 3291 i++; 3292 } 3293 } 3294 3295 /* No "useful" symbol. */ 3296 if (nsyms == 0) { 3297 goto give_up; 3298 } 3299 3300 /* Attempt to free the storage associated with the local symbols 3301 that we threw away. Whether or not this has any effect on the 3302 memory allocation depends on the malloc implementation and how 3303 many symbols we managed to discard. */ 3304 new_syms = g_try_renew(struct elf_sym, syms, nsyms); 3305 if (new_syms == NULL) { 3306 goto give_up; 3307 } 3308 syms = new_syms; 3309 3310 qsort(syms, nsyms, sizeof(*syms), symcmp); 3311 3312 s->disas_num_syms = nsyms; 3313 #if ELF_CLASS == ELFCLASS32 3314 s->disas_symtab.elf32 = syms; 3315 #else 3316 s->disas_symtab.elf64 = syms; 3317 #endif 3318 s->lookup_symbol = lookup_symbolxx; 3319 s->next = syminfos; 3320 syminfos = s; 3321 3322 return; 3323 3324 give_up: 3325 g_free(s); 3326 g_free(strings); 3327 g_free(syms); 3328 } 3329 3330 uint32_t get_elf_eflags(int fd) 3331 { 3332 struct elfhdr ehdr; 3333 off_t offset; 3334 int ret; 3335 3336 /* Read ELF header */ 3337 offset = lseek(fd, 0, SEEK_SET); 3338 if (offset == (off_t) -1) { 3339 return 0; 3340 } 3341 ret = read(fd, &ehdr, sizeof(ehdr)); 3342 if (ret < sizeof(ehdr)) { 3343 return 0; 3344 } 3345 offset = lseek(fd, offset, SEEK_SET); 3346 if (offset == (off_t) -1) { 3347 return 0; 3348 } 3349 3350 /* Check ELF signature */ 3351 if (!elf_check_ident(&ehdr)) { 3352 return 0; 3353 } 3354 3355 /* check header */ 3356 bswap_ehdr(&ehdr); 3357 if (!elf_check_ehdr(&ehdr)) { 3358 return 0; 3359 } 3360 3361 /* return architecture id */ 3362 return ehdr.e_flags; 3363 } 3364 3365 int load_elf_binary(struct linux_binprm *bprm, struct image_info *info) 3366 { 3367 struct image_info interp_info; 3368 struct elfhdr elf_ex; 3369 char *elf_interpreter = NULL; 3370 char *scratch; 3371 3372 memset(&interp_info, 0, sizeof(interp_info)); 3373 #ifdef TARGET_MIPS 3374 interp_info.fp_abi = MIPS_ABI_FP_UNKNOWN; 3375 #endif 3376 3377 info->start_mmap = (abi_ulong)ELF_START_MMAP; 3378 3379 load_elf_image(bprm->filename, bprm->fd, info, 3380 &elf_interpreter, bprm->buf); 3381 3382 /* ??? We need a copy of the elf header for passing to create_elf_tables. 3383 If we do nothing, we'll have overwritten this when we re-use bprm->buf 3384 when we load the interpreter. */ 3385 elf_ex = *(struct elfhdr *)bprm->buf; 3386 3387 /* Do this so that we can load the interpreter, if need be. We will 3388 change some of these later */ 3389 bprm->p = setup_arg_pages(bprm, info); 3390 3391 scratch = g_new0(char, TARGET_PAGE_SIZE); 3392 if (STACK_GROWS_DOWN) { 3393 bprm->p = copy_elf_strings(1, &bprm->filename, scratch, 3394 bprm->p, info->stack_limit); 3395 info->file_string = bprm->p; 3396 bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch, 3397 bprm->p, info->stack_limit); 3398 info->env_strings = bprm->p; 3399 bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch, 3400 bprm->p, info->stack_limit); 3401 info->arg_strings = bprm->p; 3402 } else { 3403 info->arg_strings = bprm->p; 3404 bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch, 3405 bprm->p, info->stack_limit); 3406 info->env_strings = bprm->p; 3407 bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch, 3408 bprm->p, info->stack_limit); 3409 info->file_string = bprm->p; 3410 bprm->p = copy_elf_strings(1, &bprm->filename, scratch, 3411 bprm->p, info->stack_limit); 3412 } 3413 3414 g_free(scratch); 3415 3416 if (!bprm->p) { 3417 fprintf(stderr, "%s: %s\n", bprm->filename, strerror(E2BIG)); 3418 exit(-1); 3419 } 3420 3421 if (elf_interpreter) { 3422 load_elf_interp(elf_interpreter, &interp_info, bprm->buf); 3423 3424 /* If the program interpreter is one of these two, then assume 3425 an iBCS2 image. Otherwise assume a native linux image. */ 3426 3427 if (strcmp(elf_interpreter, "/usr/lib/libc.so.1") == 0 3428 || strcmp(elf_interpreter, "/usr/lib/ld.so.1") == 0) { 3429 info->personality = PER_SVR4; 3430 3431 /* Why this, you ask??? Well SVr4 maps page 0 as read-only, 3432 and some applications "depend" upon this behavior. Since 3433 we do not have the power to recompile these, we emulate 3434 the SVr4 behavior. Sigh. */ 3435 target_mmap(0, qemu_host_page_size, PROT_READ | PROT_EXEC, 3436 MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); 3437 } 3438 #ifdef TARGET_MIPS 3439 info->interp_fp_abi = interp_info.fp_abi; 3440 #endif 3441 } 3442 3443 /* 3444 * TODO: load a vdso, which would also contain the signal trampolines. 3445 * Otherwise, allocate a private page to hold them. 3446 */ 3447 if (TARGET_ARCH_HAS_SIGTRAMP_PAGE) { 3448 abi_long tramp_page = target_mmap(0, TARGET_PAGE_SIZE, 3449 PROT_READ | PROT_WRITE, 3450 MAP_PRIVATE | MAP_ANON, -1, 0); 3451 if (tramp_page == -1) { 3452 return -errno; 3453 } 3454 3455 setup_sigtramp(tramp_page); 3456 target_mprotect(tramp_page, TARGET_PAGE_SIZE, PROT_READ | PROT_EXEC); 3457 } 3458 3459 bprm->p = create_elf_tables(bprm->p, bprm->argc, bprm->envc, &elf_ex, 3460 info, (elf_interpreter ? &interp_info : NULL)); 3461 info->start_stack = bprm->p; 3462 3463 /* If we have an interpreter, set that as the program's entry point. 3464 Copy the load_bias as well, to help PPC64 interpret the entry 3465 point as a function descriptor. Do this after creating elf tables 3466 so that we copy the original program entry point into the AUXV. */ 3467 if (elf_interpreter) { 3468 info->load_bias = interp_info.load_bias; 3469 info->entry = interp_info.entry; 3470 g_free(elf_interpreter); 3471 } 3472 3473 #ifdef USE_ELF_CORE_DUMP 3474 bprm->core_dump = &elf_core_dump; 3475 #endif 3476 3477 /* 3478 * If we reserved extra space for brk, release it now. 3479 * The implementation of do_brk in syscalls.c expects to be able 3480 * to mmap pages in this space. 3481 */ 3482 if (info->reserve_brk) { 3483 abi_ulong start_brk = HOST_PAGE_ALIGN(info->brk); 3484 abi_ulong end_brk = HOST_PAGE_ALIGN(info->brk + info->reserve_brk); 3485 target_munmap(start_brk, end_brk - start_brk); 3486 } 3487 3488 return 0; 3489 } 3490 3491 #ifdef USE_ELF_CORE_DUMP 3492 /* 3493 * Definitions to generate Intel SVR4-like core files. 3494 * These mostly have the same names as the SVR4 types with "target_elf_" 3495 * tacked on the front to prevent clashes with linux definitions, 3496 * and the typedef forms have been avoided. This is mostly like 3497 * the SVR4 structure, but more Linuxy, with things that Linux does 3498 * not support and which gdb doesn't really use excluded. 3499 * 3500 * Fields we don't dump (their contents is zero) in linux-user qemu 3501 * are marked with XXX. 3502 * 3503 * Core dump code is copied from linux kernel (fs/binfmt_elf.c). 3504 * 3505 * Porting ELF coredump for target is (quite) simple process. First you 3506 * define USE_ELF_CORE_DUMP in target ELF code (where init_thread() for 3507 * the target resides): 3508 * 3509 * #define USE_ELF_CORE_DUMP 3510 * 3511 * Next you define type of register set used for dumping. ELF specification 3512 * says that it needs to be array of elf_greg_t that has size of ELF_NREG. 3513 * 3514 * typedef <target_regtype> target_elf_greg_t; 3515 * #define ELF_NREG <number of registers> 3516 * typedef taret_elf_greg_t target_elf_gregset_t[ELF_NREG]; 3517 * 3518 * Last step is to implement target specific function that copies registers 3519 * from given cpu into just specified register set. Prototype is: 3520 * 3521 * static void elf_core_copy_regs(taret_elf_gregset_t *regs, 3522 * const CPUArchState *env); 3523 * 3524 * Parameters: 3525 * regs - copy register values into here (allocated and zeroed by caller) 3526 * env - copy registers from here 3527 * 3528 * Example for ARM target is provided in this file. 3529 */ 3530 3531 /* An ELF note in memory */ 3532 struct memelfnote { 3533 const char *name; 3534 size_t namesz; 3535 size_t namesz_rounded; 3536 int type; 3537 size_t datasz; 3538 size_t datasz_rounded; 3539 void *data; 3540 size_t notesz; 3541 }; 3542 3543 struct target_elf_siginfo { 3544 abi_int si_signo; /* signal number */ 3545 abi_int si_code; /* extra code */ 3546 abi_int si_errno; /* errno */ 3547 }; 3548 3549 struct target_elf_prstatus { 3550 struct target_elf_siginfo pr_info; /* Info associated with signal */ 3551 abi_short pr_cursig; /* Current signal */ 3552 abi_ulong pr_sigpend; /* XXX */ 3553 abi_ulong pr_sighold; /* XXX */ 3554 target_pid_t pr_pid; 3555 target_pid_t pr_ppid; 3556 target_pid_t pr_pgrp; 3557 target_pid_t pr_sid; 3558 struct target_timeval pr_utime; /* XXX User time */ 3559 struct target_timeval pr_stime; /* XXX System time */ 3560 struct target_timeval pr_cutime; /* XXX Cumulative user time */ 3561 struct target_timeval pr_cstime; /* XXX Cumulative system time */ 3562 target_elf_gregset_t pr_reg; /* GP registers */ 3563 abi_int pr_fpvalid; /* XXX */ 3564 }; 3565 3566 #define ELF_PRARGSZ (80) /* Number of chars for args */ 3567 3568 struct target_elf_prpsinfo { 3569 char pr_state; /* numeric process state */ 3570 char pr_sname; /* char for pr_state */ 3571 char pr_zomb; /* zombie */ 3572 char pr_nice; /* nice val */ 3573 abi_ulong pr_flag; /* flags */ 3574 target_uid_t pr_uid; 3575 target_gid_t pr_gid; 3576 target_pid_t pr_pid, pr_ppid, pr_pgrp, pr_sid; 3577 /* Lots missing */ 3578 char pr_fname[16] QEMU_NONSTRING; /* filename of executable */ 3579 char pr_psargs[ELF_PRARGSZ]; /* initial part of arg list */ 3580 }; 3581 3582 /* Here is the structure in which status of each thread is captured. */ 3583 struct elf_thread_status { 3584 QTAILQ_ENTRY(elf_thread_status) ets_link; 3585 struct target_elf_prstatus prstatus; /* NT_PRSTATUS */ 3586 #if 0 3587 elf_fpregset_t fpu; /* NT_PRFPREG */ 3588 struct task_struct *thread; 3589 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */ 3590 #endif 3591 struct memelfnote notes[1]; 3592 int num_notes; 3593 }; 3594 3595 struct elf_note_info { 3596 struct memelfnote *notes; 3597 struct target_elf_prstatus *prstatus; /* NT_PRSTATUS */ 3598 struct target_elf_prpsinfo *psinfo; /* NT_PRPSINFO */ 3599 3600 QTAILQ_HEAD(, elf_thread_status) thread_list; 3601 #if 0 3602 /* 3603 * Current version of ELF coredump doesn't support 3604 * dumping fp regs etc. 3605 */ 3606 elf_fpregset_t *fpu; 3607 elf_fpxregset_t *xfpu; 3608 int thread_status_size; 3609 #endif 3610 int notes_size; 3611 int numnote; 3612 }; 3613 3614 struct vm_area_struct { 3615 target_ulong vma_start; /* start vaddr of memory region */ 3616 target_ulong vma_end; /* end vaddr of memory region */ 3617 abi_ulong vma_flags; /* protection etc. flags for the region */ 3618 QTAILQ_ENTRY(vm_area_struct) vma_link; 3619 }; 3620 3621 struct mm_struct { 3622 QTAILQ_HEAD(, vm_area_struct) mm_mmap; 3623 int mm_count; /* number of mappings */ 3624 }; 3625 3626 static struct mm_struct *vma_init(void); 3627 static void vma_delete(struct mm_struct *); 3628 static int vma_add_mapping(struct mm_struct *, target_ulong, 3629 target_ulong, abi_ulong); 3630 static int vma_get_mapping_count(const struct mm_struct *); 3631 static struct vm_area_struct *vma_first(const struct mm_struct *); 3632 static struct vm_area_struct *vma_next(struct vm_area_struct *); 3633 static abi_ulong vma_dump_size(const struct vm_area_struct *); 3634 static int vma_walker(void *priv, target_ulong start, target_ulong end, 3635 unsigned long flags); 3636 3637 static void fill_elf_header(struct elfhdr *, int, uint16_t, uint32_t); 3638 static void fill_note(struct memelfnote *, const char *, int, 3639 unsigned int, void *); 3640 static void fill_prstatus(struct target_elf_prstatus *, const TaskState *, int); 3641 static int fill_psinfo(struct target_elf_prpsinfo *, const TaskState *); 3642 static void fill_auxv_note(struct memelfnote *, const TaskState *); 3643 static void fill_elf_note_phdr(struct elf_phdr *, int, off_t); 3644 static size_t note_size(const struct memelfnote *); 3645 static void free_note_info(struct elf_note_info *); 3646 static int fill_note_info(struct elf_note_info *, long, const CPUArchState *); 3647 static void fill_thread_info(struct elf_note_info *, const CPUArchState *); 3648 3649 static int dump_write(int, const void *, size_t); 3650 static int write_note(struct memelfnote *, int); 3651 static int write_note_info(struct elf_note_info *, int); 3652 3653 #ifdef BSWAP_NEEDED 3654 static void bswap_prstatus(struct target_elf_prstatus *prstatus) 3655 { 3656 prstatus->pr_info.si_signo = tswap32(prstatus->pr_info.si_signo); 3657 prstatus->pr_info.si_code = tswap32(prstatus->pr_info.si_code); 3658 prstatus->pr_info.si_errno = tswap32(prstatus->pr_info.si_errno); 3659 prstatus->pr_cursig = tswap16(prstatus->pr_cursig); 3660 prstatus->pr_sigpend = tswapal(prstatus->pr_sigpend); 3661 prstatus->pr_sighold = tswapal(prstatus->pr_sighold); 3662 prstatus->pr_pid = tswap32(prstatus->pr_pid); 3663 prstatus->pr_ppid = tswap32(prstatus->pr_ppid); 3664 prstatus->pr_pgrp = tswap32(prstatus->pr_pgrp); 3665 prstatus->pr_sid = tswap32(prstatus->pr_sid); 3666 /* cpu times are not filled, so we skip them */ 3667 /* regs should be in correct format already */ 3668 prstatus->pr_fpvalid = tswap32(prstatus->pr_fpvalid); 3669 } 3670 3671 static void bswap_psinfo(struct target_elf_prpsinfo *psinfo) 3672 { 3673 psinfo->pr_flag = tswapal(psinfo->pr_flag); 3674 psinfo->pr_uid = tswap16(psinfo->pr_uid); 3675 psinfo->pr_gid = tswap16(psinfo->pr_gid); 3676 psinfo->pr_pid = tswap32(psinfo->pr_pid); 3677 psinfo->pr_ppid = tswap32(psinfo->pr_ppid); 3678 psinfo->pr_pgrp = tswap32(psinfo->pr_pgrp); 3679 psinfo->pr_sid = tswap32(psinfo->pr_sid); 3680 } 3681 3682 static void bswap_note(struct elf_note *en) 3683 { 3684 bswap32s(&en->n_namesz); 3685 bswap32s(&en->n_descsz); 3686 bswap32s(&en->n_type); 3687 } 3688 #else 3689 static inline void bswap_prstatus(struct target_elf_prstatus *p) { } 3690 static inline void bswap_psinfo(struct target_elf_prpsinfo *p) {} 3691 static inline void bswap_note(struct elf_note *en) { } 3692 #endif /* BSWAP_NEEDED */ 3693 3694 /* 3695 * Minimal support for linux memory regions. These are needed 3696 * when we are finding out what memory exactly belongs to 3697 * emulated process. No locks needed here, as long as 3698 * thread that received the signal is stopped. 3699 */ 3700 3701 static struct mm_struct *vma_init(void) 3702 { 3703 struct mm_struct *mm; 3704 3705 if ((mm = g_malloc(sizeof (*mm))) == NULL) 3706 return (NULL); 3707 3708 mm->mm_count = 0; 3709 QTAILQ_INIT(&mm->mm_mmap); 3710 3711 return (mm); 3712 } 3713 3714 static void vma_delete(struct mm_struct *mm) 3715 { 3716 struct vm_area_struct *vma; 3717 3718 while ((vma = vma_first(mm)) != NULL) { 3719 QTAILQ_REMOVE(&mm->mm_mmap, vma, vma_link); 3720 g_free(vma); 3721 } 3722 g_free(mm); 3723 } 3724 3725 static int vma_add_mapping(struct mm_struct *mm, target_ulong start, 3726 target_ulong end, abi_ulong flags) 3727 { 3728 struct vm_area_struct *vma; 3729 3730 if ((vma = g_malloc0(sizeof (*vma))) == NULL) 3731 return (-1); 3732 3733 vma->vma_start = start; 3734 vma->vma_end = end; 3735 vma->vma_flags = flags; 3736 3737 QTAILQ_INSERT_TAIL(&mm->mm_mmap, vma, vma_link); 3738 mm->mm_count++; 3739 3740 return (0); 3741 } 3742 3743 static struct vm_area_struct *vma_first(const struct mm_struct *mm) 3744 { 3745 return (QTAILQ_FIRST(&mm->mm_mmap)); 3746 } 3747 3748 static struct vm_area_struct *vma_next(struct vm_area_struct *vma) 3749 { 3750 return (QTAILQ_NEXT(vma, vma_link)); 3751 } 3752 3753 static int vma_get_mapping_count(const struct mm_struct *mm) 3754 { 3755 return (mm->mm_count); 3756 } 3757 3758 /* 3759 * Calculate file (dump) size of given memory region. 3760 */ 3761 static abi_ulong vma_dump_size(const struct vm_area_struct *vma) 3762 { 3763 /* if we cannot even read the first page, skip it */ 3764 if (!access_ok_untagged(VERIFY_READ, vma->vma_start, TARGET_PAGE_SIZE)) 3765 return (0); 3766 3767 /* 3768 * Usually we don't dump executable pages as they contain 3769 * non-writable code that debugger can read directly from 3770 * target library etc. However, thread stacks are marked 3771 * also executable so we read in first page of given region 3772 * and check whether it contains elf header. If there is 3773 * no elf header, we dump it. 3774 */ 3775 if (vma->vma_flags & PROT_EXEC) { 3776 char page[TARGET_PAGE_SIZE]; 3777 3778 if (copy_from_user(page, vma->vma_start, sizeof (page))) { 3779 return 0; 3780 } 3781 if ((page[EI_MAG0] == ELFMAG0) && 3782 (page[EI_MAG1] == ELFMAG1) && 3783 (page[EI_MAG2] == ELFMAG2) && 3784 (page[EI_MAG3] == ELFMAG3)) { 3785 /* 3786 * Mappings are possibly from ELF binary. Don't dump 3787 * them. 3788 */ 3789 return (0); 3790 } 3791 } 3792 3793 return (vma->vma_end - vma->vma_start); 3794 } 3795 3796 static int vma_walker(void *priv, target_ulong start, target_ulong end, 3797 unsigned long flags) 3798 { 3799 struct mm_struct *mm = (struct mm_struct *)priv; 3800 3801 vma_add_mapping(mm, start, end, flags); 3802 return (0); 3803 } 3804 3805 static void fill_note(struct memelfnote *note, const char *name, int type, 3806 unsigned int sz, void *data) 3807 { 3808 unsigned int namesz; 3809 3810 namesz = strlen(name) + 1; 3811 note->name = name; 3812 note->namesz = namesz; 3813 note->namesz_rounded = roundup(namesz, sizeof (int32_t)); 3814 note->type = type; 3815 note->datasz = sz; 3816 note->datasz_rounded = roundup(sz, sizeof (int32_t)); 3817 3818 note->data = data; 3819 3820 /* 3821 * We calculate rounded up note size here as specified by 3822 * ELF document. 3823 */ 3824 note->notesz = sizeof (struct elf_note) + 3825 note->namesz_rounded + note->datasz_rounded; 3826 } 3827 3828 static void fill_elf_header(struct elfhdr *elf, int segs, uint16_t machine, 3829 uint32_t flags) 3830 { 3831 (void) memset(elf, 0, sizeof(*elf)); 3832 3833 (void) memcpy(elf->e_ident, ELFMAG, SELFMAG); 3834 elf->e_ident[EI_CLASS] = ELF_CLASS; 3835 elf->e_ident[EI_DATA] = ELF_DATA; 3836 elf->e_ident[EI_VERSION] = EV_CURRENT; 3837 elf->e_ident[EI_OSABI] = ELF_OSABI; 3838 3839 elf->e_type = ET_CORE; 3840 elf->e_machine = machine; 3841 elf->e_version = EV_CURRENT; 3842 elf->e_phoff = sizeof(struct elfhdr); 3843 elf->e_flags = flags; 3844 elf->e_ehsize = sizeof(struct elfhdr); 3845 elf->e_phentsize = sizeof(struct elf_phdr); 3846 elf->e_phnum = segs; 3847 3848 bswap_ehdr(elf); 3849 } 3850 3851 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, off_t offset) 3852 { 3853 phdr->p_type = PT_NOTE; 3854 phdr->p_offset = offset; 3855 phdr->p_vaddr = 0; 3856 phdr->p_paddr = 0; 3857 phdr->p_filesz = sz; 3858 phdr->p_memsz = 0; 3859 phdr->p_flags = 0; 3860 phdr->p_align = 0; 3861 3862 bswap_phdr(phdr, 1); 3863 } 3864 3865 static size_t note_size(const struct memelfnote *note) 3866 { 3867 return (note->notesz); 3868 } 3869 3870 static void fill_prstatus(struct target_elf_prstatus *prstatus, 3871 const TaskState *ts, int signr) 3872 { 3873 (void) memset(prstatus, 0, sizeof (*prstatus)); 3874 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr; 3875 prstatus->pr_pid = ts->ts_tid; 3876 prstatus->pr_ppid = getppid(); 3877 prstatus->pr_pgrp = getpgrp(); 3878 prstatus->pr_sid = getsid(0); 3879 3880 bswap_prstatus(prstatus); 3881 } 3882 3883 static int fill_psinfo(struct target_elf_prpsinfo *psinfo, const TaskState *ts) 3884 { 3885 char *base_filename; 3886 unsigned int i, len; 3887 3888 (void) memset(psinfo, 0, sizeof (*psinfo)); 3889 3890 len = ts->info->env_strings - ts->info->arg_strings; 3891 if (len >= ELF_PRARGSZ) 3892 len = ELF_PRARGSZ - 1; 3893 if (copy_from_user(&psinfo->pr_psargs, ts->info->arg_strings, len)) { 3894 return -EFAULT; 3895 } 3896 for (i = 0; i < len; i++) 3897 if (psinfo->pr_psargs[i] == 0) 3898 psinfo->pr_psargs[i] = ' '; 3899 psinfo->pr_psargs[len] = 0; 3900 3901 psinfo->pr_pid = getpid(); 3902 psinfo->pr_ppid = getppid(); 3903 psinfo->pr_pgrp = getpgrp(); 3904 psinfo->pr_sid = getsid(0); 3905 psinfo->pr_uid = getuid(); 3906 psinfo->pr_gid = getgid(); 3907 3908 base_filename = g_path_get_basename(ts->bprm->filename); 3909 /* 3910 * Using strncpy here is fine: at max-length, 3911 * this field is not NUL-terminated. 3912 */ 3913 (void) strncpy(psinfo->pr_fname, base_filename, 3914 sizeof(psinfo->pr_fname)); 3915 3916 g_free(base_filename); 3917 bswap_psinfo(psinfo); 3918 return (0); 3919 } 3920 3921 static void fill_auxv_note(struct memelfnote *note, const TaskState *ts) 3922 { 3923 elf_addr_t auxv = (elf_addr_t)ts->info->saved_auxv; 3924 elf_addr_t orig_auxv = auxv; 3925 void *ptr; 3926 int len = ts->info->auxv_len; 3927 3928 /* 3929 * Auxiliary vector is stored in target process stack. It contains 3930 * {type, value} pairs that we need to dump into note. This is not 3931 * strictly necessary but we do it here for sake of completeness. 3932 */ 3933 3934 /* read in whole auxv vector and copy it to memelfnote */ 3935 ptr = lock_user(VERIFY_READ, orig_auxv, len, 0); 3936 if (ptr != NULL) { 3937 fill_note(note, "CORE", NT_AUXV, len, ptr); 3938 unlock_user(ptr, auxv, len); 3939 } 3940 } 3941 3942 /* 3943 * Constructs name of coredump file. We have following convention 3944 * for the name: 3945 * qemu_<basename-of-target-binary>_<date>-<time>_<pid>.core 3946 * 3947 * Returns the filename 3948 */ 3949 static char *core_dump_filename(const TaskState *ts) 3950 { 3951 g_autoptr(GDateTime) now = g_date_time_new_now_local(); 3952 g_autofree char *nowstr = g_date_time_format(now, "%Y%m%d-%H%M%S"); 3953 g_autofree char *base_filename = g_path_get_basename(ts->bprm->filename); 3954 3955 return g_strdup_printf("qemu_%s_%s_%d.core", 3956 base_filename, nowstr, (int)getpid()); 3957 } 3958 3959 static int dump_write(int fd, const void *ptr, size_t size) 3960 { 3961 const char *bufp = (const char *)ptr; 3962 ssize_t bytes_written, bytes_left; 3963 struct rlimit dumpsize; 3964 off_t pos; 3965 3966 bytes_written = 0; 3967 getrlimit(RLIMIT_CORE, &dumpsize); 3968 if ((pos = lseek(fd, 0, SEEK_CUR))==-1) { 3969 if (errno == ESPIPE) { /* not a seekable stream */ 3970 bytes_left = size; 3971 } else { 3972 return pos; 3973 } 3974 } else { 3975 if (dumpsize.rlim_cur <= pos) { 3976 return -1; 3977 } else if (dumpsize.rlim_cur == RLIM_INFINITY) { 3978 bytes_left = size; 3979 } else { 3980 size_t limit_left=dumpsize.rlim_cur - pos; 3981 bytes_left = limit_left >= size ? size : limit_left ; 3982 } 3983 } 3984 3985 /* 3986 * In normal conditions, single write(2) should do but 3987 * in case of socket etc. this mechanism is more portable. 3988 */ 3989 do { 3990 bytes_written = write(fd, bufp, bytes_left); 3991 if (bytes_written < 0) { 3992 if (errno == EINTR) 3993 continue; 3994 return (-1); 3995 } else if (bytes_written == 0) { /* eof */ 3996 return (-1); 3997 } 3998 bufp += bytes_written; 3999 bytes_left -= bytes_written; 4000 } while (bytes_left > 0); 4001 4002 return (0); 4003 } 4004 4005 static int write_note(struct memelfnote *men, int fd) 4006 { 4007 struct elf_note en; 4008 4009 en.n_namesz = men->namesz; 4010 en.n_type = men->type; 4011 en.n_descsz = men->datasz; 4012 4013 bswap_note(&en); 4014 4015 if (dump_write(fd, &en, sizeof(en)) != 0) 4016 return (-1); 4017 if (dump_write(fd, men->name, men->namesz_rounded) != 0) 4018 return (-1); 4019 if (dump_write(fd, men->data, men->datasz_rounded) != 0) 4020 return (-1); 4021 4022 return (0); 4023 } 4024 4025 static void fill_thread_info(struct elf_note_info *info, const CPUArchState *env) 4026 { 4027 CPUState *cpu = env_cpu((CPUArchState *)env); 4028 TaskState *ts = (TaskState *)cpu->opaque; 4029 struct elf_thread_status *ets; 4030 4031 ets = g_malloc0(sizeof (*ets)); 4032 ets->num_notes = 1; /* only prstatus is dumped */ 4033 fill_prstatus(&ets->prstatus, ts, 0); 4034 elf_core_copy_regs(&ets->prstatus.pr_reg, env); 4035 fill_note(&ets->notes[0], "CORE", NT_PRSTATUS, sizeof (ets->prstatus), 4036 &ets->prstatus); 4037 4038 QTAILQ_INSERT_TAIL(&info->thread_list, ets, ets_link); 4039 4040 info->notes_size += note_size(&ets->notes[0]); 4041 } 4042 4043 static void init_note_info(struct elf_note_info *info) 4044 { 4045 /* Initialize the elf_note_info structure so that it is at 4046 * least safe to call free_note_info() on it. Must be 4047 * called before calling fill_note_info(). 4048 */ 4049 memset(info, 0, sizeof (*info)); 4050 QTAILQ_INIT(&info->thread_list); 4051 } 4052 4053 static int fill_note_info(struct elf_note_info *info, 4054 long signr, const CPUArchState *env) 4055 { 4056 #define NUMNOTES 3 4057 CPUState *cpu = env_cpu((CPUArchState *)env); 4058 TaskState *ts = (TaskState *)cpu->opaque; 4059 int i; 4060 4061 info->notes = g_new0(struct memelfnote, NUMNOTES); 4062 if (info->notes == NULL) 4063 return (-ENOMEM); 4064 info->prstatus = g_malloc0(sizeof (*info->prstatus)); 4065 if (info->prstatus == NULL) 4066 return (-ENOMEM); 4067 info->psinfo = g_malloc0(sizeof (*info->psinfo)); 4068 if (info->prstatus == NULL) 4069 return (-ENOMEM); 4070 4071 /* 4072 * First fill in status (and registers) of current thread 4073 * including process info & aux vector. 4074 */ 4075 fill_prstatus(info->prstatus, ts, signr); 4076 elf_core_copy_regs(&info->prstatus->pr_reg, env); 4077 fill_note(&info->notes[0], "CORE", NT_PRSTATUS, 4078 sizeof (*info->prstatus), info->prstatus); 4079 fill_psinfo(info->psinfo, ts); 4080 fill_note(&info->notes[1], "CORE", NT_PRPSINFO, 4081 sizeof (*info->psinfo), info->psinfo); 4082 fill_auxv_note(&info->notes[2], ts); 4083 info->numnote = 3; 4084 4085 info->notes_size = 0; 4086 for (i = 0; i < info->numnote; i++) 4087 info->notes_size += note_size(&info->notes[i]); 4088 4089 /* read and fill status of all threads */ 4090 cpu_list_lock(); 4091 CPU_FOREACH(cpu) { 4092 if (cpu == thread_cpu) { 4093 continue; 4094 } 4095 fill_thread_info(info, cpu->env_ptr); 4096 } 4097 cpu_list_unlock(); 4098 4099 return (0); 4100 } 4101 4102 static void free_note_info(struct elf_note_info *info) 4103 { 4104 struct elf_thread_status *ets; 4105 4106 while (!QTAILQ_EMPTY(&info->thread_list)) { 4107 ets = QTAILQ_FIRST(&info->thread_list); 4108 QTAILQ_REMOVE(&info->thread_list, ets, ets_link); 4109 g_free(ets); 4110 } 4111 4112 g_free(info->prstatus); 4113 g_free(info->psinfo); 4114 g_free(info->notes); 4115 } 4116 4117 static int write_note_info(struct elf_note_info *info, int fd) 4118 { 4119 struct elf_thread_status *ets; 4120 int i, error = 0; 4121 4122 /* write prstatus, psinfo and auxv for current thread */ 4123 for (i = 0; i < info->numnote; i++) 4124 if ((error = write_note(&info->notes[i], fd)) != 0) 4125 return (error); 4126 4127 /* write prstatus for each thread */ 4128 QTAILQ_FOREACH(ets, &info->thread_list, ets_link) { 4129 if ((error = write_note(&ets->notes[0], fd)) != 0) 4130 return (error); 4131 } 4132 4133 return (0); 4134 } 4135 4136 /* 4137 * Write out ELF coredump. 4138 * 4139 * See documentation of ELF object file format in: 4140 * http://www.caldera.com/developers/devspecs/gabi41.pdf 4141 * 4142 * Coredump format in linux is following: 4143 * 4144 * 0 +----------------------+ \ 4145 * | ELF header | ET_CORE | 4146 * +----------------------+ | 4147 * | ELF program headers | |--- headers 4148 * | - NOTE section | | 4149 * | - PT_LOAD sections | | 4150 * +----------------------+ / 4151 * | NOTEs: | 4152 * | - NT_PRSTATUS | 4153 * | - NT_PRSINFO | 4154 * | - NT_AUXV | 4155 * +----------------------+ <-- aligned to target page 4156 * | Process memory dump | 4157 * : : 4158 * . . 4159 * : : 4160 * | | 4161 * +----------------------+ 4162 * 4163 * NT_PRSTATUS -> struct elf_prstatus (per thread) 4164 * NT_PRSINFO -> struct elf_prpsinfo 4165 * NT_AUXV is array of { type, value } pairs (see fill_auxv_note()). 4166 * 4167 * Format follows System V format as close as possible. Current 4168 * version limitations are as follows: 4169 * - no floating point registers are dumped 4170 * 4171 * Function returns 0 in case of success, negative errno otherwise. 4172 * 4173 * TODO: make this work also during runtime: it should be 4174 * possible to force coredump from running process and then 4175 * continue processing. For example qemu could set up SIGUSR2 4176 * handler (provided that target process haven't registered 4177 * handler for that) that does the dump when signal is received. 4178 */ 4179 static int elf_core_dump(int signr, const CPUArchState *env) 4180 { 4181 const CPUState *cpu = env_cpu((CPUArchState *)env); 4182 const TaskState *ts = (const TaskState *)cpu->opaque; 4183 struct vm_area_struct *vma = NULL; 4184 g_autofree char *corefile = NULL; 4185 struct elf_note_info info; 4186 struct elfhdr elf; 4187 struct elf_phdr phdr; 4188 struct rlimit dumpsize; 4189 struct mm_struct *mm = NULL; 4190 off_t offset = 0, data_offset = 0; 4191 int segs = 0; 4192 int fd = -1; 4193 4194 init_note_info(&info); 4195 4196 errno = 0; 4197 getrlimit(RLIMIT_CORE, &dumpsize); 4198 if (dumpsize.rlim_cur == 0) 4199 return 0; 4200 4201 corefile = core_dump_filename(ts); 4202 4203 if ((fd = open(corefile, O_WRONLY | O_CREAT, 4204 S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH)) < 0) 4205 return (-errno); 4206 4207 /* 4208 * Walk through target process memory mappings and 4209 * set up structure containing this information. After 4210 * this point vma_xxx functions can be used. 4211 */ 4212 if ((mm = vma_init()) == NULL) 4213 goto out; 4214 4215 walk_memory_regions(mm, vma_walker); 4216 segs = vma_get_mapping_count(mm); 4217 4218 /* 4219 * Construct valid coredump ELF header. We also 4220 * add one more segment for notes. 4221 */ 4222 fill_elf_header(&elf, segs + 1, ELF_MACHINE, 0); 4223 if (dump_write(fd, &elf, sizeof (elf)) != 0) 4224 goto out; 4225 4226 /* fill in the in-memory version of notes */ 4227 if (fill_note_info(&info, signr, env) < 0) 4228 goto out; 4229 4230 offset += sizeof (elf); /* elf header */ 4231 offset += (segs + 1) * sizeof (struct elf_phdr); /* program headers */ 4232 4233 /* write out notes program header */ 4234 fill_elf_note_phdr(&phdr, info.notes_size, offset); 4235 4236 offset += info.notes_size; 4237 if (dump_write(fd, &phdr, sizeof (phdr)) != 0) 4238 goto out; 4239 4240 /* 4241 * ELF specification wants data to start at page boundary so 4242 * we align it here. 4243 */ 4244 data_offset = offset = roundup(offset, ELF_EXEC_PAGESIZE); 4245 4246 /* 4247 * Write program headers for memory regions mapped in 4248 * the target process. 4249 */ 4250 for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) { 4251 (void) memset(&phdr, 0, sizeof (phdr)); 4252 4253 phdr.p_type = PT_LOAD; 4254 phdr.p_offset = offset; 4255 phdr.p_vaddr = vma->vma_start; 4256 phdr.p_paddr = 0; 4257 phdr.p_filesz = vma_dump_size(vma); 4258 offset += phdr.p_filesz; 4259 phdr.p_memsz = vma->vma_end - vma->vma_start; 4260 phdr.p_flags = vma->vma_flags & PROT_READ ? PF_R : 0; 4261 if (vma->vma_flags & PROT_WRITE) 4262 phdr.p_flags |= PF_W; 4263 if (vma->vma_flags & PROT_EXEC) 4264 phdr.p_flags |= PF_X; 4265 phdr.p_align = ELF_EXEC_PAGESIZE; 4266 4267 bswap_phdr(&phdr, 1); 4268 if (dump_write(fd, &phdr, sizeof(phdr)) != 0) { 4269 goto out; 4270 } 4271 } 4272 4273 /* 4274 * Next we write notes just after program headers. No 4275 * alignment needed here. 4276 */ 4277 if (write_note_info(&info, fd) < 0) 4278 goto out; 4279 4280 /* align data to page boundary */ 4281 if (lseek(fd, data_offset, SEEK_SET) != data_offset) 4282 goto out; 4283 4284 /* 4285 * Finally we can dump process memory into corefile as well. 4286 */ 4287 for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) { 4288 abi_ulong addr; 4289 abi_ulong end; 4290 4291 end = vma->vma_start + vma_dump_size(vma); 4292 4293 for (addr = vma->vma_start; addr < end; 4294 addr += TARGET_PAGE_SIZE) { 4295 char page[TARGET_PAGE_SIZE]; 4296 int error; 4297 4298 /* 4299 * Read in page from target process memory and 4300 * write it to coredump file. 4301 */ 4302 error = copy_from_user(page, addr, sizeof (page)); 4303 if (error != 0) { 4304 (void) fprintf(stderr, "unable to dump " TARGET_ABI_FMT_lx "\n", 4305 addr); 4306 errno = -error; 4307 goto out; 4308 } 4309 if (dump_write(fd, page, TARGET_PAGE_SIZE) < 0) 4310 goto out; 4311 } 4312 } 4313 4314 out: 4315 free_note_info(&info); 4316 if (mm != NULL) 4317 vma_delete(mm); 4318 (void) close(fd); 4319 4320 if (errno != 0) 4321 return (-errno); 4322 return (0); 4323 } 4324 #endif /* USE_ELF_CORE_DUMP */ 4325 4326 void do_init_thread(struct target_pt_regs *regs, struct image_info *infop) 4327 { 4328 init_thread(regs, infop); 4329 } 4330