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