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 "disas/disas.h" 10 #include "qemu/path.h" 11 #include "qemu/queue.h" 12 #include "qemu/guest-random.h" 13 #include "qemu/units.h" 14 #include "qemu/selfmap.h" 15 16 #ifdef _ARCH_PPC64 17 #undef ARCH_DLINFO 18 #undef ELF_PLATFORM 19 #undef ELF_HWCAP 20 #undef ELF_HWCAP2 21 #undef ELF_CLASS 22 #undef ELF_DATA 23 #undef ELF_ARCH 24 #endif 25 26 #define ELF_OSABI ELFOSABI_SYSV 27 28 /* from personality.h */ 29 30 /* 31 * Flags for bug emulation. 32 * 33 * These occupy the top three bytes. 34 */ 35 enum { 36 ADDR_NO_RANDOMIZE = 0x0040000, /* disable randomization of VA space */ 37 FDPIC_FUNCPTRS = 0x0080000, /* userspace function ptrs point to 38 descriptors (signal handling) */ 39 MMAP_PAGE_ZERO = 0x0100000, 40 ADDR_COMPAT_LAYOUT = 0x0200000, 41 READ_IMPLIES_EXEC = 0x0400000, 42 ADDR_LIMIT_32BIT = 0x0800000, 43 SHORT_INODE = 0x1000000, 44 WHOLE_SECONDS = 0x2000000, 45 STICKY_TIMEOUTS = 0x4000000, 46 ADDR_LIMIT_3GB = 0x8000000, 47 }; 48 49 /* 50 * Personality types. 51 * 52 * These go in the low byte. Avoid using the top bit, it will 53 * conflict with error returns. 54 */ 55 enum { 56 PER_LINUX = 0x0000, 57 PER_LINUX_32BIT = 0x0000 | ADDR_LIMIT_32BIT, 58 PER_LINUX_FDPIC = 0x0000 | FDPIC_FUNCPTRS, 59 PER_SVR4 = 0x0001 | STICKY_TIMEOUTS | MMAP_PAGE_ZERO, 60 PER_SVR3 = 0x0002 | STICKY_TIMEOUTS | SHORT_INODE, 61 PER_SCOSVR3 = 0x0003 | STICKY_TIMEOUTS | WHOLE_SECONDS | SHORT_INODE, 62 PER_OSR5 = 0x0003 | STICKY_TIMEOUTS | WHOLE_SECONDS, 63 PER_WYSEV386 = 0x0004 | STICKY_TIMEOUTS | SHORT_INODE, 64 PER_ISCR4 = 0x0005 | STICKY_TIMEOUTS, 65 PER_BSD = 0x0006, 66 PER_SUNOS = 0x0006 | STICKY_TIMEOUTS, 67 PER_XENIX = 0x0007 | STICKY_TIMEOUTS | SHORT_INODE, 68 PER_LINUX32 = 0x0008, 69 PER_LINUX32_3GB = 0x0008 | ADDR_LIMIT_3GB, 70 PER_IRIX32 = 0x0009 | STICKY_TIMEOUTS,/* IRIX5 32-bit */ 71 PER_IRIXN32 = 0x000a | STICKY_TIMEOUTS,/* IRIX6 new 32-bit */ 72 PER_IRIX64 = 0x000b | STICKY_TIMEOUTS,/* IRIX6 64-bit */ 73 PER_RISCOS = 0x000c, 74 PER_SOLARIS = 0x000d | STICKY_TIMEOUTS, 75 PER_UW7 = 0x000e | STICKY_TIMEOUTS | MMAP_PAGE_ZERO, 76 PER_OSF4 = 0x000f, /* OSF/1 v4 */ 77 PER_HPUX = 0x0010, 78 PER_MASK = 0x00ff, 79 }; 80 81 /* 82 * Return the base personality without flags. 83 */ 84 #define personality(pers) (pers & PER_MASK) 85 86 int info_is_fdpic(struct image_info *info) 87 { 88 return info->personality == PER_LINUX_FDPIC; 89 } 90 91 /* this flag is uneffective under linux too, should be deleted */ 92 #ifndef MAP_DENYWRITE 93 #define MAP_DENYWRITE 0 94 #endif 95 96 /* should probably go in elf.h */ 97 #ifndef ELIBBAD 98 #define ELIBBAD 80 99 #endif 100 101 #ifdef TARGET_WORDS_BIGENDIAN 102 #define ELF_DATA ELFDATA2MSB 103 #else 104 #define ELF_DATA ELFDATA2LSB 105 #endif 106 107 #ifdef TARGET_ABI_MIPSN32 108 typedef abi_ullong target_elf_greg_t; 109 #define tswapreg(ptr) tswap64(ptr) 110 #else 111 typedef abi_ulong target_elf_greg_t; 112 #define tswapreg(ptr) tswapal(ptr) 113 #endif 114 115 #ifdef USE_UID16 116 typedef abi_ushort target_uid_t; 117 typedef abi_ushort target_gid_t; 118 #else 119 typedef abi_uint target_uid_t; 120 typedef abi_uint target_gid_t; 121 #endif 122 typedef abi_int target_pid_t; 123 124 #ifdef TARGET_I386 125 126 #define ELF_PLATFORM get_elf_platform() 127 128 static const char *get_elf_platform(void) 129 { 130 static char elf_platform[] = "i386"; 131 int family = object_property_get_int(OBJECT(thread_cpu), "family", NULL); 132 if (family > 6) 133 family = 6; 134 if (family >= 3) 135 elf_platform[1] = '0' + family; 136 return elf_platform; 137 } 138 139 #define ELF_HWCAP get_elf_hwcap() 140 141 static uint32_t get_elf_hwcap(void) 142 { 143 X86CPU *cpu = X86_CPU(thread_cpu); 144 145 return cpu->env.features[FEAT_1_EDX]; 146 } 147 148 #ifdef TARGET_X86_64 149 #define ELF_START_MMAP 0x2aaaaab000ULL 150 151 #define ELF_CLASS ELFCLASS64 152 #define ELF_ARCH EM_X86_64 153 154 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop) 155 { 156 regs->rax = 0; 157 regs->rsp = infop->start_stack; 158 regs->rip = infop->entry; 159 } 160 161 #define ELF_NREG 27 162 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 163 164 /* 165 * Note that ELF_NREG should be 29 as there should be place for 166 * TRAPNO and ERR "registers" as well but linux doesn't dump 167 * those. 168 * 169 * See linux kernel: arch/x86/include/asm/elf.h 170 */ 171 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUX86State *env) 172 { 173 (*regs)[0] = env->regs[15]; 174 (*regs)[1] = env->regs[14]; 175 (*regs)[2] = env->regs[13]; 176 (*regs)[3] = env->regs[12]; 177 (*regs)[4] = env->regs[R_EBP]; 178 (*regs)[5] = env->regs[R_EBX]; 179 (*regs)[6] = env->regs[11]; 180 (*regs)[7] = env->regs[10]; 181 (*regs)[8] = env->regs[9]; 182 (*regs)[9] = env->regs[8]; 183 (*regs)[10] = env->regs[R_EAX]; 184 (*regs)[11] = env->regs[R_ECX]; 185 (*regs)[12] = env->regs[R_EDX]; 186 (*regs)[13] = env->regs[R_ESI]; 187 (*regs)[14] = env->regs[R_EDI]; 188 (*regs)[15] = env->regs[R_EAX]; /* XXX */ 189 (*regs)[16] = env->eip; 190 (*regs)[17] = env->segs[R_CS].selector & 0xffff; 191 (*regs)[18] = env->eflags; 192 (*regs)[19] = env->regs[R_ESP]; 193 (*regs)[20] = env->segs[R_SS].selector & 0xffff; 194 (*regs)[21] = env->segs[R_FS].selector & 0xffff; 195 (*regs)[22] = env->segs[R_GS].selector & 0xffff; 196 (*regs)[23] = env->segs[R_DS].selector & 0xffff; 197 (*regs)[24] = env->segs[R_ES].selector & 0xffff; 198 (*regs)[25] = env->segs[R_FS].selector & 0xffff; 199 (*regs)[26] = env->segs[R_GS].selector & 0xffff; 200 } 201 202 #else 203 204 #define ELF_START_MMAP 0x80000000 205 206 /* 207 * This is used to ensure we don't load something for the wrong architecture. 208 */ 209 #define elf_check_arch(x) ( ((x) == EM_386) || ((x) == EM_486) ) 210 211 /* 212 * These are used to set parameters in the core dumps. 213 */ 214 #define ELF_CLASS ELFCLASS32 215 #define ELF_ARCH EM_386 216 217 static inline void init_thread(struct target_pt_regs *regs, 218 struct image_info *infop) 219 { 220 regs->esp = infop->start_stack; 221 regs->eip = infop->entry; 222 223 /* SVR4/i386 ABI (pages 3-31, 3-32) says that when the program 224 starts %edx contains a pointer to a function which might be 225 registered using `atexit'. This provides a mean for the 226 dynamic linker to call DT_FINI functions for shared libraries 227 that have been loaded before the code runs. 228 229 A value of 0 tells we have no such handler. */ 230 regs->edx = 0; 231 } 232 233 #define ELF_NREG 17 234 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 235 236 /* 237 * Note that ELF_NREG should be 19 as there should be place for 238 * TRAPNO and ERR "registers" as well but linux doesn't dump 239 * those. 240 * 241 * See linux kernel: arch/x86/include/asm/elf.h 242 */ 243 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUX86State *env) 244 { 245 (*regs)[0] = env->regs[R_EBX]; 246 (*regs)[1] = env->regs[R_ECX]; 247 (*regs)[2] = env->regs[R_EDX]; 248 (*regs)[3] = env->regs[R_ESI]; 249 (*regs)[4] = env->regs[R_EDI]; 250 (*regs)[5] = env->regs[R_EBP]; 251 (*regs)[6] = env->regs[R_EAX]; 252 (*regs)[7] = env->segs[R_DS].selector & 0xffff; 253 (*regs)[8] = env->segs[R_ES].selector & 0xffff; 254 (*regs)[9] = env->segs[R_FS].selector & 0xffff; 255 (*regs)[10] = env->segs[R_GS].selector & 0xffff; 256 (*regs)[11] = env->regs[R_EAX]; /* XXX */ 257 (*regs)[12] = env->eip; 258 (*regs)[13] = env->segs[R_CS].selector & 0xffff; 259 (*regs)[14] = env->eflags; 260 (*regs)[15] = env->regs[R_ESP]; 261 (*regs)[16] = env->segs[R_SS].selector & 0xffff; 262 } 263 #endif 264 265 #define USE_ELF_CORE_DUMP 266 #define ELF_EXEC_PAGESIZE 4096 267 268 #endif 269 270 #ifdef TARGET_ARM 271 272 #ifndef TARGET_AARCH64 273 /* 32 bit ARM definitions */ 274 275 #define ELF_START_MMAP 0x80000000 276 277 #define ELF_ARCH EM_ARM 278 #define ELF_CLASS ELFCLASS32 279 280 static inline void init_thread(struct target_pt_regs *regs, 281 struct image_info *infop) 282 { 283 abi_long stack = infop->start_stack; 284 memset(regs, 0, sizeof(*regs)); 285 286 regs->uregs[16] = ARM_CPU_MODE_USR; 287 if (infop->entry & 1) { 288 regs->uregs[16] |= CPSR_T; 289 } 290 regs->uregs[15] = infop->entry & 0xfffffffe; 291 regs->uregs[13] = infop->start_stack; 292 /* FIXME - what to for failure of get_user()? */ 293 get_user_ual(regs->uregs[2], stack + 8); /* envp */ 294 get_user_ual(regs->uregs[1], stack + 4); /* envp */ 295 /* XXX: it seems that r0 is zeroed after ! */ 296 regs->uregs[0] = 0; 297 /* For uClinux PIC binaries. */ 298 /* XXX: Linux does this only on ARM with no MMU (do we care ?) */ 299 regs->uregs[10] = infop->start_data; 300 301 /* Support ARM FDPIC. */ 302 if (info_is_fdpic(infop)) { 303 /* As described in the ABI document, r7 points to the loadmap info 304 * prepared by the kernel. If an interpreter is needed, r8 points 305 * to the interpreter loadmap and r9 points to the interpreter 306 * PT_DYNAMIC info. If no interpreter is needed, r8 is zero, and 307 * r9 points to the main program PT_DYNAMIC info. 308 */ 309 regs->uregs[7] = infop->loadmap_addr; 310 if (infop->interpreter_loadmap_addr) { 311 /* Executable is dynamically loaded. */ 312 regs->uregs[8] = infop->interpreter_loadmap_addr; 313 regs->uregs[9] = infop->interpreter_pt_dynamic_addr; 314 } else { 315 regs->uregs[8] = 0; 316 regs->uregs[9] = infop->pt_dynamic_addr; 317 } 318 } 319 } 320 321 #define ELF_NREG 18 322 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 323 324 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUARMState *env) 325 { 326 (*regs)[0] = tswapreg(env->regs[0]); 327 (*regs)[1] = tswapreg(env->regs[1]); 328 (*regs)[2] = tswapreg(env->regs[2]); 329 (*regs)[3] = tswapreg(env->regs[3]); 330 (*regs)[4] = tswapreg(env->regs[4]); 331 (*regs)[5] = tswapreg(env->regs[5]); 332 (*regs)[6] = tswapreg(env->regs[6]); 333 (*regs)[7] = tswapreg(env->regs[7]); 334 (*regs)[8] = tswapreg(env->regs[8]); 335 (*regs)[9] = tswapreg(env->regs[9]); 336 (*regs)[10] = tswapreg(env->regs[10]); 337 (*regs)[11] = tswapreg(env->regs[11]); 338 (*regs)[12] = tswapreg(env->regs[12]); 339 (*regs)[13] = tswapreg(env->regs[13]); 340 (*regs)[14] = tswapreg(env->regs[14]); 341 (*regs)[15] = tswapreg(env->regs[15]); 342 343 (*regs)[16] = tswapreg(cpsr_read((CPUARMState *)env)); 344 (*regs)[17] = tswapreg(env->regs[0]); /* XXX */ 345 } 346 347 #define USE_ELF_CORE_DUMP 348 #define ELF_EXEC_PAGESIZE 4096 349 350 enum 351 { 352 ARM_HWCAP_ARM_SWP = 1 << 0, 353 ARM_HWCAP_ARM_HALF = 1 << 1, 354 ARM_HWCAP_ARM_THUMB = 1 << 2, 355 ARM_HWCAP_ARM_26BIT = 1 << 3, 356 ARM_HWCAP_ARM_FAST_MULT = 1 << 4, 357 ARM_HWCAP_ARM_FPA = 1 << 5, 358 ARM_HWCAP_ARM_VFP = 1 << 6, 359 ARM_HWCAP_ARM_EDSP = 1 << 7, 360 ARM_HWCAP_ARM_JAVA = 1 << 8, 361 ARM_HWCAP_ARM_IWMMXT = 1 << 9, 362 ARM_HWCAP_ARM_CRUNCH = 1 << 10, 363 ARM_HWCAP_ARM_THUMBEE = 1 << 11, 364 ARM_HWCAP_ARM_NEON = 1 << 12, 365 ARM_HWCAP_ARM_VFPv3 = 1 << 13, 366 ARM_HWCAP_ARM_VFPv3D16 = 1 << 14, 367 ARM_HWCAP_ARM_TLS = 1 << 15, 368 ARM_HWCAP_ARM_VFPv4 = 1 << 16, 369 ARM_HWCAP_ARM_IDIVA = 1 << 17, 370 ARM_HWCAP_ARM_IDIVT = 1 << 18, 371 ARM_HWCAP_ARM_VFPD32 = 1 << 19, 372 ARM_HWCAP_ARM_LPAE = 1 << 20, 373 ARM_HWCAP_ARM_EVTSTRM = 1 << 21, 374 }; 375 376 enum { 377 ARM_HWCAP2_ARM_AES = 1 << 0, 378 ARM_HWCAP2_ARM_PMULL = 1 << 1, 379 ARM_HWCAP2_ARM_SHA1 = 1 << 2, 380 ARM_HWCAP2_ARM_SHA2 = 1 << 3, 381 ARM_HWCAP2_ARM_CRC32 = 1 << 4, 382 }; 383 384 /* The commpage only exists for 32 bit kernels */ 385 386 #define ARM_COMMPAGE (intptr_t)0xffff0f00u 387 388 static bool init_guest_commpage(void) 389 { 390 void *want = g2h(ARM_COMMPAGE & -qemu_host_page_size); 391 void *addr = mmap(want, qemu_host_page_size, PROT_READ | PROT_WRITE, 392 MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0); 393 394 if (addr == MAP_FAILED) { 395 perror("Allocating guest commpage"); 396 exit(EXIT_FAILURE); 397 } 398 if (addr != want) { 399 return false; 400 } 401 402 /* Set kernel helper versions; rest of page is 0. */ 403 __put_user(5, (uint32_t *)g2h(0xffff0ffcu)); 404 405 if (mprotect(addr, qemu_host_page_size, PROT_READ)) { 406 perror("Protecting guest commpage"); 407 exit(EXIT_FAILURE); 408 } 409 return true; 410 } 411 412 #define ELF_HWCAP get_elf_hwcap() 413 #define ELF_HWCAP2 get_elf_hwcap2() 414 415 static uint32_t get_elf_hwcap(void) 416 { 417 ARMCPU *cpu = ARM_CPU(thread_cpu); 418 uint32_t hwcaps = 0; 419 420 hwcaps |= ARM_HWCAP_ARM_SWP; 421 hwcaps |= ARM_HWCAP_ARM_HALF; 422 hwcaps |= ARM_HWCAP_ARM_THUMB; 423 hwcaps |= ARM_HWCAP_ARM_FAST_MULT; 424 425 /* probe for the extra features */ 426 #define GET_FEATURE(feat, hwcap) \ 427 do { if (arm_feature(&cpu->env, feat)) { hwcaps |= hwcap; } } while (0) 428 429 #define GET_FEATURE_ID(feat, hwcap) \ 430 do { if (cpu_isar_feature(feat, cpu)) { hwcaps |= hwcap; } } while (0) 431 432 /* EDSP is in v5TE and above, but all our v5 CPUs are v5TE */ 433 GET_FEATURE(ARM_FEATURE_V5, ARM_HWCAP_ARM_EDSP); 434 GET_FEATURE(ARM_FEATURE_IWMMXT, ARM_HWCAP_ARM_IWMMXT); 435 GET_FEATURE(ARM_FEATURE_THUMB2EE, ARM_HWCAP_ARM_THUMBEE); 436 GET_FEATURE(ARM_FEATURE_NEON, ARM_HWCAP_ARM_NEON); 437 GET_FEATURE(ARM_FEATURE_V6K, ARM_HWCAP_ARM_TLS); 438 GET_FEATURE(ARM_FEATURE_LPAE, ARM_HWCAP_ARM_LPAE); 439 GET_FEATURE_ID(aa32_arm_div, ARM_HWCAP_ARM_IDIVA); 440 GET_FEATURE_ID(aa32_thumb_div, ARM_HWCAP_ARM_IDIVT); 441 GET_FEATURE_ID(aa32_vfp, ARM_HWCAP_ARM_VFP); 442 443 if (cpu_isar_feature(aa32_fpsp_v3, cpu) || 444 cpu_isar_feature(aa32_fpdp_v3, cpu)) { 445 hwcaps |= ARM_HWCAP_ARM_VFPv3; 446 if (cpu_isar_feature(aa32_simd_r32, cpu)) { 447 hwcaps |= ARM_HWCAP_ARM_VFPD32; 448 } else { 449 hwcaps |= ARM_HWCAP_ARM_VFPv3D16; 450 } 451 } 452 GET_FEATURE_ID(aa32_simdfmac, ARM_HWCAP_ARM_VFPv4); 453 454 return hwcaps; 455 } 456 457 static uint32_t get_elf_hwcap2(void) 458 { 459 ARMCPU *cpu = ARM_CPU(thread_cpu); 460 uint32_t hwcaps = 0; 461 462 GET_FEATURE_ID(aa32_aes, ARM_HWCAP2_ARM_AES); 463 GET_FEATURE_ID(aa32_pmull, ARM_HWCAP2_ARM_PMULL); 464 GET_FEATURE_ID(aa32_sha1, ARM_HWCAP2_ARM_SHA1); 465 GET_FEATURE_ID(aa32_sha2, ARM_HWCAP2_ARM_SHA2); 466 GET_FEATURE_ID(aa32_crc32, ARM_HWCAP2_ARM_CRC32); 467 return hwcaps; 468 } 469 470 #undef GET_FEATURE 471 #undef GET_FEATURE_ID 472 473 #define ELF_PLATFORM get_elf_platform() 474 475 static const char *get_elf_platform(void) 476 { 477 CPUARMState *env = thread_cpu->env_ptr; 478 479 #ifdef TARGET_WORDS_BIGENDIAN 480 # define END "b" 481 #else 482 # define END "l" 483 #endif 484 485 if (arm_feature(env, ARM_FEATURE_V8)) { 486 return "v8" END; 487 } else if (arm_feature(env, ARM_FEATURE_V7)) { 488 if (arm_feature(env, ARM_FEATURE_M)) { 489 return "v7m" END; 490 } else { 491 return "v7" END; 492 } 493 } else if (arm_feature(env, ARM_FEATURE_V6)) { 494 return "v6" END; 495 } else if (arm_feature(env, ARM_FEATURE_V5)) { 496 return "v5" END; 497 } else { 498 return "v4" END; 499 } 500 501 #undef END 502 } 503 504 #else 505 /* 64 bit ARM definitions */ 506 #define ELF_START_MMAP 0x80000000 507 508 #define ELF_ARCH EM_AARCH64 509 #define ELF_CLASS ELFCLASS64 510 #ifdef TARGET_WORDS_BIGENDIAN 511 # define ELF_PLATFORM "aarch64_be" 512 #else 513 # define ELF_PLATFORM "aarch64" 514 #endif 515 516 static inline void init_thread(struct target_pt_regs *regs, 517 struct image_info *infop) 518 { 519 abi_long stack = infop->start_stack; 520 memset(regs, 0, sizeof(*regs)); 521 522 regs->pc = infop->entry & ~0x3ULL; 523 regs->sp = stack; 524 } 525 526 #define ELF_NREG 34 527 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 528 529 static void elf_core_copy_regs(target_elf_gregset_t *regs, 530 const CPUARMState *env) 531 { 532 int i; 533 534 for (i = 0; i < 32; i++) { 535 (*regs)[i] = tswapreg(env->xregs[i]); 536 } 537 (*regs)[32] = tswapreg(env->pc); 538 (*regs)[33] = tswapreg(pstate_read((CPUARMState *)env)); 539 } 540 541 #define USE_ELF_CORE_DUMP 542 #define ELF_EXEC_PAGESIZE 4096 543 544 enum { 545 ARM_HWCAP_A64_FP = 1 << 0, 546 ARM_HWCAP_A64_ASIMD = 1 << 1, 547 ARM_HWCAP_A64_EVTSTRM = 1 << 2, 548 ARM_HWCAP_A64_AES = 1 << 3, 549 ARM_HWCAP_A64_PMULL = 1 << 4, 550 ARM_HWCAP_A64_SHA1 = 1 << 5, 551 ARM_HWCAP_A64_SHA2 = 1 << 6, 552 ARM_HWCAP_A64_CRC32 = 1 << 7, 553 ARM_HWCAP_A64_ATOMICS = 1 << 8, 554 ARM_HWCAP_A64_FPHP = 1 << 9, 555 ARM_HWCAP_A64_ASIMDHP = 1 << 10, 556 ARM_HWCAP_A64_CPUID = 1 << 11, 557 ARM_HWCAP_A64_ASIMDRDM = 1 << 12, 558 ARM_HWCAP_A64_JSCVT = 1 << 13, 559 ARM_HWCAP_A64_FCMA = 1 << 14, 560 ARM_HWCAP_A64_LRCPC = 1 << 15, 561 ARM_HWCAP_A64_DCPOP = 1 << 16, 562 ARM_HWCAP_A64_SHA3 = 1 << 17, 563 ARM_HWCAP_A64_SM3 = 1 << 18, 564 ARM_HWCAP_A64_SM4 = 1 << 19, 565 ARM_HWCAP_A64_ASIMDDP = 1 << 20, 566 ARM_HWCAP_A64_SHA512 = 1 << 21, 567 ARM_HWCAP_A64_SVE = 1 << 22, 568 ARM_HWCAP_A64_ASIMDFHM = 1 << 23, 569 ARM_HWCAP_A64_DIT = 1 << 24, 570 ARM_HWCAP_A64_USCAT = 1 << 25, 571 ARM_HWCAP_A64_ILRCPC = 1 << 26, 572 ARM_HWCAP_A64_FLAGM = 1 << 27, 573 ARM_HWCAP_A64_SSBS = 1 << 28, 574 ARM_HWCAP_A64_SB = 1 << 29, 575 ARM_HWCAP_A64_PACA = 1 << 30, 576 ARM_HWCAP_A64_PACG = 1UL << 31, 577 578 ARM_HWCAP2_A64_DCPODP = 1 << 0, 579 ARM_HWCAP2_A64_SVE2 = 1 << 1, 580 ARM_HWCAP2_A64_SVEAES = 1 << 2, 581 ARM_HWCAP2_A64_SVEPMULL = 1 << 3, 582 ARM_HWCAP2_A64_SVEBITPERM = 1 << 4, 583 ARM_HWCAP2_A64_SVESHA3 = 1 << 5, 584 ARM_HWCAP2_A64_SVESM4 = 1 << 6, 585 ARM_HWCAP2_A64_FLAGM2 = 1 << 7, 586 ARM_HWCAP2_A64_FRINT = 1 << 8, 587 }; 588 589 #define ELF_HWCAP get_elf_hwcap() 590 #define ELF_HWCAP2 get_elf_hwcap2() 591 592 #define GET_FEATURE_ID(feat, hwcap) \ 593 do { if (cpu_isar_feature(feat, cpu)) { hwcaps |= hwcap; } } while (0) 594 595 static uint32_t get_elf_hwcap(void) 596 { 597 ARMCPU *cpu = ARM_CPU(thread_cpu); 598 uint32_t hwcaps = 0; 599 600 hwcaps |= ARM_HWCAP_A64_FP; 601 hwcaps |= ARM_HWCAP_A64_ASIMD; 602 hwcaps |= ARM_HWCAP_A64_CPUID; 603 604 /* probe for the extra features */ 605 606 GET_FEATURE_ID(aa64_aes, ARM_HWCAP_A64_AES); 607 GET_FEATURE_ID(aa64_pmull, ARM_HWCAP_A64_PMULL); 608 GET_FEATURE_ID(aa64_sha1, ARM_HWCAP_A64_SHA1); 609 GET_FEATURE_ID(aa64_sha256, ARM_HWCAP_A64_SHA2); 610 GET_FEATURE_ID(aa64_sha512, ARM_HWCAP_A64_SHA512); 611 GET_FEATURE_ID(aa64_crc32, ARM_HWCAP_A64_CRC32); 612 GET_FEATURE_ID(aa64_sha3, ARM_HWCAP_A64_SHA3); 613 GET_FEATURE_ID(aa64_sm3, ARM_HWCAP_A64_SM3); 614 GET_FEATURE_ID(aa64_sm4, ARM_HWCAP_A64_SM4); 615 GET_FEATURE_ID(aa64_fp16, ARM_HWCAP_A64_FPHP | ARM_HWCAP_A64_ASIMDHP); 616 GET_FEATURE_ID(aa64_atomics, ARM_HWCAP_A64_ATOMICS); 617 GET_FEATURE_ID(aa64_rdm, ARM_HWCAP_A64_ASIMDRDM); 618 GET_FEATURE_ID(aa64_dp, ARM_HWCAP_A64_ASIMDDP); 619 GET_FEATURE_ID(aa64_fcma, ARM_HWCAP_A64_FCMA); 620 GET_FEATURE_ID(aa64_sve, ARM_HWCAP_A64_SVE); 621 GET_FEATURE_ID(aa64_pauth, ARM_HWCAP_A64_PACA | ARM_HWCAP_A64_PACG); 622 GET_FEATURE_ID(aa64_fhm, ARM_HWCAP_A64_ASIMDFHM); 623 GET_FEATURE_ID(aa64_jscvt, ARM_HWCAP_A64_JSCVT); 624 GET_FEATURE_ID(aa64_sb, ARM_HWCAP_A64_SB); 625 GET_FEATURE_ID(aa64_condm_4, ARM_HWCAP_A64_FLAGM); 626 GET_FEATURE_ID(aa64_dcpop, ARM_HWCAP_A64_DCPOP); 627 GET_FEATURE_ID(aa64_rcpc_8_3, ARM_HWCAP_A64_LRCPC); 628 GET_FEATURE_ID(aa64_rcpc_8_4, ARM_HWCAP_A64_ILRCPC); 629 630 return hwcaps; 631 } 632 633 static uint32_t get_elf_hwcap2(void) 634 { 635 ARMCPU *cpu = ARM_CPU(thread_cpu); 636 uint32_t hwcaps = 0; 637 638 GET_FEATURE_ID(aa64_dcpodp, ARM_HWCAP2_A64_DCPODP); 639 GET_FEATURE_ID(aa64_condm_5, ARM_HWCAP2_A64_FLAGM2); 640 GET_FEATURE_ID(aa64_frint, ARM_HWCAP2_A64_FRINT); 641 642 return hwcaps; 643 } 644 645 #undef GET_FEATURE_ID 646 647 #endif /* not TARGET_AARCH64 */ 648 #endif /* TARGET_ARM */ 649 650 #ifdef TARGET_SPARC 651 #ifdef TARGET_SPARC64 652 653 #define ELF_START_MMAP 0x80000000 654 #define ELF_HWCAP (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \ 655 | HWCAP_SPARC_MULDIV | HWCAP_SPARC_V9) 656 #ifndef TARGET_ABI32 657 #define elf_check_arch(x) ( (x) == EM_SPARCV9 || (x) == EM_SPARC32PLUS ) 658 #else 659 #define elf_check_arch(x) ( (x) == EM_SPARC32PLUS || (x) == EM_SPARC ) 660 #endif 661 662 #define ELF_CLASS ELFCLASS64 663 #define ELF_ARCH EM_SPARCV9 664 665 #define STACK_BIAS 2047 666 667 static inline void init_thread(struct target_pt_regs *regs, 668 struct image_info *infop) 669 { 670 #ifndef TARGET_ABI32 671 regs->tstate = 0; 672 #endif 673 regs->pc = infop->entry; 674 regs->npc = regs->pc + 4; 675 regs->y = 0; 676 #ifdef TARGET_ABI32 677 regs->u_regs[14] = infop->start_stack - 16 * 4; 678 #else 679 if (personality(infop->personality) == PER_LINUX32) 680 regs->u_regs[14] = infop->start_stack - 16 * 4; 681 else 682 regs->u_regs[14] = infop->start_stack - 16 * 8 - STACK_BIAS; 683 #endif 684 } 685 686 #else 687 #define ELF_START_MMAP 0x80000000 688 #define ELF_HWCAP (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \ 689 | HWCAP_SPARC_MULDIV) 690 691 #define ELF_CLASS ELFCLASS32 692 #define ELF_ARCH EM_SPARC 693 694 static inline void init_thread(struct target_pt_regs *regs, 695 struct image_info *infop) 696 { 697 regs->psr = 0; 698 regs->pc = infop->entry; 699 regs->npc = regs->pc + 4; 700 regs->y = 0; 701 regs->u_regs[14] = infop->start_stack - 16 * 4; 702 } 703 704 #endif 705 #endif 706 707 #ifdef TARGET_PPC 708 709 #define ELF_MACHINE PPC_ELF_MACHINE 710 #define ELF_START_MMAP 0x80000000 711 712 #if defined(TARGET_PPC64) && !defined(TARGET_ABI32) 713 714 #define elf_check_arch(x) ( (x) == EM_PPC64 ) 715 716 #define ELF_CLASS ELFCLASS64 717 718 #else 719 720 #define ELF_CLASS ELFCLASS32 721 722 #endif 723 724 #define ELF_ARCH EM_PPC 725 726 /* Feature masks for the Aux Vector Hardware Capabilities (AT_HWCAP). 727 See arch/powerpc/include/asm/cputable.h. */ 728 enum { 729 QEMU_PPC_FEATURE_32 = 0x80000000, 730 QEMU_PPC_FEATURE_64 = 0x40000000, 731 QEMU_PPC_FEATURE_601_INSTR = 0x20000000, 732 QEMU_PPC_FEATURE_HAS_ALTIVEC = 0x10000000, 733 QEMU_PPC_FEATURE_HAS_FPU = 0x08000000, 734 QEMU_PPC_FEATURE_HAS_MMU = 0x04000000, 735 QEMU_PPC_FEATURE_HAS_4xxMAC = 0x02000000, 736 QEMU_PPC_FEATURE_UNIFIED_CACHE = 0x01000000, 737 QEMU_PPC_FEATURE_HAS_SPE = 0x00800000, 738 QEMU_PPC_FEATURE_HAS_EFP_SINGLE = 0x00400000, 739 QEMU_PPC_FEATURE_HAS_EFP_DOUBLE = 0x00200000, 740 QEMU_PPC_FEATURE_NO_TB = 0x00100000, 741 QEMU_PPC_FEATURE_POWER4 = 0x00080000, 742 QEMU_PPC_FEATURE_POWER5 = 0x00040000, 743 QEMU_PPC_FEATURE_POWER5_PLUS = 0x00020000, 744 QEMU_PPC_FEATURE_CELL = 0x00010000, 745 QEMU_PPC_FEATURE_BOOKE = 0x00008000, 746 QEMU_PPC_FEATURE_SMT = 0x00004000, 747 QEMU_PPC_FEATURE_ICACHE_SNOOP = 0x00002000, 748 QEMU_PPC_FEATURE_ARCH_2_05 = 0x00001000, 749 QEMU_PPC_FEATURE_PA6T = 0x00000800, 750 QEMU_PPC_FEATURE_HAS_DFP = 0x00000400, 751 QEMU_PPC_FEATURE_POWER6_EXT = 0x00000200, 752 QEMU_PPC_FEATURE_ARCH_2_06 = 0x00000100, 753 QEMU_PPC_FEATURE_HAS_VSX = 0x00000080, 754 QEMU_PPC_FEATURE_PSERIES_PERFMON_COMPAT = 0x00000040, 755 756 QEMU_PPC_FEATURE_TRUE_LE = 0x00000002, 757 QEMU_PPC_FEATURE_PPC_LE = 0x00000001, 758 759 /* Feature definitions in AT_HWCAP2. */ 760 QEMU_PPC_FEATURE2_ARCH_2_07 = 0x80000000, /* ISA 2.07 */ 761 QEMU_PPC_FEATURE2_HAS_HTM = 0x40000000, /* Hardware Transactional Memory */ 762 QEMU_PPC_FEATURE2_HAS_DSCR = 0x20000000, /* Data Stream Control Register */ 763 QEMU_PPC_FEATURE2_HAS_EBB = 0x10000000, /* Event Base Branching */ 764 QEMU_PPC_FEATURE2_HAS_ISEL = 0x08000000, /* Integer Select */ 765 QEMU_PPC_FEATURE2_HAS_TAR = 0x04000000, /* Target Address Register */ 766 QEMU_PPC_FEATURE2_VEC_CRYPTO = 0x02000000, 767 QEMU_PPC_FEATURE2_HTM_NOSC = 0x01000000, 768 QEMU_PPC_FEATURE2_ARCH_3_00 = 0x00800000, /* ISA 3.00 */ 769 QEMU_PPC_FEATURE2_HAS_IEEE128 = 0x00400000, /* VSX IEEE Bin Float 128-bit */ 770 QEMU_PPC_FEATURE2_DARN = 0x00200000, /* darn random number insn */ 771 QEMU_PPC_FEATURE2_SCV = 0x00100000, /* scv syscall */ 772 QEMU_PPC_FEATURE2_HTM_NO_SUSPEND = 0x00080000, /* TM w/o suspended state */ 773 }; 774 775 #define ELF_HWCAP get_elf_hwcap() 776 777 static uint32_t get_elf_hwcap(void) 778 { 779 PowerPCCPU *cpu = POWERPC_CPU(thread_cpu); 780 uint32_t features = 0; 781 782 /* We don't have to be terribly complete here; the high points are 783 Altivec/FP/SPE support. Anything else is just a bonus. */ 784 #define GET_FEATURE(flag, feature) \ 785 do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0) 786 #define GET_FEATURE2(flags, feature) \ 787 do { \ 788 if ((cpu->env.insns_flags2 & flags) == flags) { \ 789 features |= feature; \ 790 } \ 791 } while (0) 792 GET_FEATURE(PPC_64B, QEMU_PPC_FEATURE_64); 793 GET_FEATURE(PPC_FLOAT, QEMU_PPC_FEATURE_HAS_FPU); 794 GET_FEATURE(PPC_ALTIVEC, QEMU_PPC_FEATURE_HAS_ALTIVEC); 795 GET_FEATURE(PPC_SPE, QEMU_PPC_FEATURE_HAS_SPE); 796 GET_FEATURE(PPC_SPE_SINGLE, QEMU_PPC_FEATURE_HAS_EFP_SINGLE); 797 GET_FEATURE(PPC_SPE_DOUBLE, QEMU_PPC_FEATURE_HAS_EFP_DOUBLE); 798 GET_FEATURE(PPC_BOOKE, QEMU_PPC_FEATURE_BOOKE); 799 GET_FEATURE(PPC_405_MAC, QEMU_PPC_FEATURE_HAS_4xxMAC); 800 GET_FEATURE2(PPC2_DFP, QEMU_PPC_FEATURE_HAS_DFP); 801 GET_FEATURE2(PPC2_VSX, QEMU_PPC_FEATURE_HAS_VSX); 802 GET_FEATURE2((PPC2_PERM_ISA206 | PPC2_DIVE_ISA206 | PPC2_ATOMIC_ISA206 | 803 PPC2_FP_CVT_ISA206 | PPC2_FP_TST_ISA206), 804 QEMU_PPC_FEATURE_ARCH_2_06); 805 #undef GET_FEATURE 806 #undef GET_FEATURE2 807 808 return features; 809 } 810 811 #define ELF_HWCAP2 get_elf_hwcap2() 812 813 static uint32_t get_elf_hwcap2(void) 814 { 815 PowerPCCPU *cpu = POWERPC_CPU(thread_cpu); 816 uint32_t features = 0; 817 818 #define GET_FEATURE(flag, feature) \ 819 do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0) 820 #define GET_FEATURE2(flag, feature) \ 821 do { if (cpu->env.insns_flags2 & flag) { features |= feature; } } while (0) 822 823 GET_FEATURE(PPC_ISEL, QEMU_PPC_FEATURE2_HAS_ISEL); 824 GET_FEATURE2(PPC2_BCTAR_ISA207, QEMU_PPC_FEATURE2_HAS_TAR); 825 GET_FEATURE2((PPC2_BCTAR_ISA207 | PPC2_LSQ_ISA207 | PPC2_ALTIVEC_207 | 826 PPC2_ISA207S), QEMU_PPC_FEATURE2_ARCH_2_07 | 827 QEMU_PPC_FEATURE2_VEC_CRYPTO); 828 GET_FEATURE2(PPC2_ISA300, QEMU_PPC_FEATURE2_ARCH_3_00 | 829 QEMU_PPC_FEATURE2_DARN); 830 831 #undef GET_FEATURE 832 #undef GET_FEATURE2 833 834 return features; 835 } 836 837 /* 838 * The requirements here are: 839 * - keep the final alignment of sp (sp & 0xf) 840 * - make sure the 32-bit value at the first 16 byte aligned position of 841 * AUXV is greater than 16 for glibc compatibility. 842 * AT_IGNOREPPC is used for that. 843 * - for compatibility with glibc ARCH_DLINFO must always be defined on PPC, 844 * even if DLINFO_ARCH_ITEMS goes to zero or is undefined. 845 */ 846 #define DLINFO_ARCH_ITEMS 5 847 #define ARCH_DLINFO \ 848 do { \ 849 PowerPCCPU *cpu = POWERPC_CPU(thread_cpu); \ 850 /* \ 851 * Handle glibc compatibility: these magic entries must \ 852 * be at the lowest addresses in the final auxv. \ 853 */ \ 854 NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC); \ 855 NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC); \ 856 NEW_AUX_ENT(AT_DCACHEBSIZE, cpu->env.dcache_line_size); \ 857 NEW_AUX_ENT(AT_ICACHEBSIZE, cpu->env.icache_line_size); \ 858 NEW_AUX_ENT(AT_UCACHEBSIZE, 0); \ 859 } while (0) 860 861 static inline void init_thread(struct target_pt_regs *_regs, struct image_info *infop) 862 { 863 _regs->gpr[1] = infop->start_stack; 864 #if defined(TARGET_PPC64) && !defined(TARGET_ABI32) 865 if (get_ppc64_abi(infop) < 2) { 866 uint64_t val; 867 get_user_u64(val, infop->entry + 8); 868 _regs->gpr[2] = val + infop->load_bias; 869 get_user_u64(val, infop->entry); 870 infop->entry = val + infop->load_bias; 871 } else { 872 _regs->gpr[12] = infop->entry; /* r12 set to global entry address */ 873 } 874 #endif 875 _regs->nip = infop->entry; 876 } 877 878 /* See linux kernel: arch/powerpc/include/asm/elf.h. */ 879 #define ELF_NREG 48 880 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 881 882 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUPPCState *env) 883 { 884 int i; 885 target_ulong ccr = 0; 886 887 for (i = 0; i < ARRAY_SIZE(env->gpr); i++) { 888 (*regs)[i] = tswapreg(env->gpr[i]); 889 } 890 891 (*regs)[32] = tswapreg(env->nip); 892 (*regs)[33] = tswapreg(env->msr); 893 (*regs)[35] = tswapreg(env->ctr); 894 (*regs)[36] = tswapreg(env->lr); 895 (*regs)[37] = tswapreg(env->xer); 896 897 for (i = 0; i < ARRAY_SIZE(env->crf); i++) { 898 ccr |= env->crf[i] << (32 - ((i + 1) * 4)); 899 } 900 (*regs)[38] = tswapreg(ccr); 901 } 902 903 #define USE_ELF_CORE_DUMP 904 #define ELF_EXEC_PAGESIZE 4096 905 906 #endif 907 908 #ifdef TARGET_MIPS 909 910 #define ELF_START_MMAP 0x80000000 911 912 #ifdef TARGET_MIPS64 913 #define ELF_CLASS ELFCLASS64 914 #else 915 #define ELF_CLASS ELFCLASS32 916 #endif 917 #define ELF_ARCH EM_MIPS 918 919 #define elf_check_arch(x) ((x) == EM_MIPS || (x) == EM_NANOMIPS) 920 921 #ifdef TARGET_ABI_MIPSN32 922 #define elf_check_abi(x) ((x) & EF_MIPS_ABI2) 923 #else 924 #define elf_check_abi(x) (!((x) & EF_MIPS_ABI2)) 925 #endif 926 927 static inline void init_thread(struct target_pt_regs *regs, 928 struct image_info *infop) 929 { 930 regs->cp0_status = 2 << CP0St_KSU; 931 regs->cp0_epc = infop->entry; 932 regs->regs[29] = infop->start_stack; 933 } 934 935 /* See linux kernel: arch/mips/include/asm/elf.h. */ 936 #define ELF_NREG 45 937 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 938 939 /* See linux kernel: arch/mips/include/asm/reg.h. */ 940 enum { 941 #ifdef TARGET_MIPS64 942 TARGET_EF_R0 = 0, 943 #else 944 TARGET_EF_R0 = 6, 945 #endif 946 TARGET_EF_R26 = TARGET_EF_R0 + 26, 947 TARGET_EF_R27 = TARGET_EF_R0 + 27, 948 TARGET_EF_LO = TARGET_EF_R0 + 32, 949 TARGET_EF_HI = TARGET_EF_R0 + 33, 950 TARGET_EF_CP0_EPC = TARGET_EF_R0 + 34, 951 TARGET_EF_CP0_BADVADDR = TARGET_EF_R0 + 35, 952 TARGET_EF_CP0_STATUS = TARGET_EF_R0 + 36, 953 TARGET_EF_CP0_CAUSE = TARGET_EF_R0 + 37 954 }; 955 956 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */ 957 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMIPSState *env) 958 { 959 int i; 960 961 for (i = 0; i < TARGET_EF_R0; i++) { 962 (*regs)[i] = 0; 963 } 964 (*regs)[TARGET_EF_R0] = 0; 965 966 for (i = 1; i < ARRAY_SIZE(env->active_tc.gpr); i++) { 967 (*regs)[TARGET_EF_R0 + i] = tswapreg(env->active_tc.gpr[i]); 968 } 969 970 (*regs)[TARGET_EF_R26] = 0; 971 (*regs)[TARGET_EF_R27] = 0; 972 (*regs)[TARGET_EF_LO] = tswapreg(env->active_tc.LO[0]); 973 (*regs)[TARGET_EF_HI] = tswapreg(env->active_tc.HI[0]); 974 (*regs)[TARGET_EF_CP0_EPC] = tswapreg(env->active_tc.PC); 975 (*regs)[TARGET_EF_CP0_BADVADDR] = tswapreg(env->CP0_BadVAddr); 976 (*regs)[TARGET_EF_CP0_STATUS] = tswapreg(env->CP0_Status); 977 (*regs)[TARGET_EF_CP0_CAUSE] = tswapreg(env->CP0_Cause); 978 } 979 980 #define USE_ELF_CORE_DUMP 981 #define ELF_EXEC_PAGESIZE 4096 982 983 /* See arch/mips/include/uapi/asm/hwcap.h. */ 984 enum { 985 HWCAP_MIPS_R6 = (1 << 0), 986 HWCAP_MIPS_MSA = (1 << 1), 987 }; 988 989 #define ELF_HWCAP get_elf_hwcap() 990 991 static uint32_t get_elf_hwcap(void) 992 { 993 MIPSCPU *cpu = MIPS_CPU(thread_cpu); 994 uint32_t hwcaps = 0; 995 996 #define GET_FEATURE(flag, hwcap) \ 997 do { if (cpu->env.insn_flags & (flag)) { hwcaps |= hwcap; } } while (0) 998 999 GET_FEATURE(ISA_MIPS32R6 | ISA_MIPS64R6, HWCAP_MIPS_R6); 1000 GET_FEATURE(ASE_MSA, HWCAP_MIPS_MSA); 1001 1002 #undef GET_FEATURE 1003 1004 return hwcaps; 1005 } 1006 1007 #endif /* TARGET_MIPS */ 1008 1009 #ifdef TARGET_MICROBLAZE 1010 1011 #define ELF_START_MMAP 0x80000000 1012 1013 #define elf_check_arch(x) ( (x) == EM_MICROBLAZE || (x) == EM_MICROBLAZE_OLD) 1014 1015 #define ELF_CLASS ELFCLASS32 1016 #define ELF_ARCH EM_MICROBLAZE 1017 1018 static inline void init_thread(struct target_pt_regs *regs, 1019 struct image_info *infop) 1020 { 1021 regs->pc = infop->entry; 1022 regs->r1 = infop->start_stack; 1023 1024 } 1025 1026 #define ELF_EXEC_PAGESIZE 4096 1027 1028 #define USE_ELF_CORE_DUMP 1029 #define ELF_NREG 38 1030 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1031 1032 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */ 1033 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMBState *env) 1034 { 1035 int i, pos = 0; 1036 1037 for (i = 0; i < 32; i++) { 1038 (*regs)[pos++] = tswapreg(env->regs[i]); 1039 } 1040 1041 (*regs)[pos++] = tswapreg(env->pc); 1042 (*regs)[pos++] = tswapreg(mb_cpu_read_msr(env)); 1043 (*regs)[pos++] = 0; 1044 (*regs)[pos++] = tswapreg(env->ear); 1045 (*regs)[pos++] = 0; 1046 (*regs)[pos++] = tswapreg(env->esr); 1047 } 1048 1049 #endif /* TARGET_MICROBLAZE */ 1050 1051 #ifdef TARGET_NIOS2 1052 1053 #define ELF_START_MMAP 0x80000000 1054 1055 #define elf_check_arch(x) ((x) == EM_ALTERA_NIOS2) 1056 1057 #define ELF_CLASS ELFCLASS32 1058 #define ELF_ARCH EM_ALTERA_NIOS2 1059 1060 static void init_thread(struct target_pt_regs *regs, struct image_info *infop) 1061 { 1062 regs->ea = infop->entry; 1063 regs->sp = infop->start_stack; 1064 regs->estatus = 0x3; 1065 } 1066 1067 #define ELF_EXEC_PAGESIZE 4096 1068 1069 #define USE_ELF_CORE_DUMP 1070 #define ELF_NREG 49 1071 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1072 1073 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */ 1074 static void elf_core_copy_regs(target_elf_gregset_t *regs, 1075 const CPUNios2State *env) 1076 { 1077 int i; 1078 1079 (*regs)[0] = -1; 1080 for (i = 1; i < 8; i++) /* r0-r7 */ 1081 (*regs)[i] = tswapreg(env->regs[i + 7]); 1082 1083 for (i = 8; i < 16; i++) /* r8-r15 */ 1084 (*regs)[i] = tswapreg(env->regs[i - 8]); 1085 1086 for (i = 16; i < 24; i++) /* r16-r23 */ 1087 (*regs)[i] = tswapreg(env->regs[i + 7]); 1088 (*regs)[24] = -1; /* R_ET */ 1089 (*regs)[25] = -1; /* R_BT */ 1090 (*regs)[26] = tswapreg(env->regs[R_GP]); 1091 (*regs)[27] = tswapreg(env->regs[R_SP]); 1092 (*regs)[28] = tswapreg(env->regs[R_FP]); 1093 (*regs)[29] = tswapreg(env->regs[R_EA]); 1094 (*regs)[30] = -1; /* R_SSTATUS */ 1095 (*regs)[31] = tswapreg(env->regs[R_RA]); 1096 1097 (*regs)[32] = tswapreg(env->regs[R_PC]); 1098 1099 (*regs)[33] = -1; /* R_STATUS */ 1100 (*regs)[34] = tswapreg(env->regs[CR_ESTATUS]); 1101 1102 for (i = 35; i < 49; i++) /* ... */ 1103 (*regs)[i] = -1; 1104 } 1105 1106 #endif /* TARGET_NIOS2 */ 1107 1108 #ifdef TARGET_OPENRISC 1109 1110 #define ELF_START_MMAP 0x08000000 1111 1112 #define ELF_ARCH EM_OPENRISC 1113 #define ELF_CLASS ELFCLASS32 1114 #define ELF_DATA ELFDATA2MSB 1115 1116 static inline void init_thread(struct target_pt_regs *regs, 1117 struct image_info *infop) 1118 { 1119 regs->pc = infop->entry; 1120 regs->gpr[1] = infop->start_stack; 1121 } 1122 1123 #define USE_ELF_CORE_DUMP 1124 #define ELF_EXEC_PAGESIZE 8192 1125 1126 /* See linux kernel arch/openrisc/include/asm/elf.h. */ 1127 #define ELF_NREG 34 /* gprs and pc, sr */ 1128 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1129 1130 static void elf_core_copy_regs(target_elf_gregset_t *regs, 1131 const CPUOpenRISCState *env) 1132 { 1133 int i; 1134 1135 for (i = 0; i < 32; i++) { 1136 (*regs)[i] = tswapreg(cpu_get_gpr(env, i)); 1137 } 1138 (*regs)[32] = tswapreg(env->pc); 1139 (*regs)[33] = tswapreg(cpu_get_sr(env)); 1140 } 1141 #define ELF_HWCAP 0 1142 #define ELF_PLATFORM NULL 1143 1144 #endif /* TARGET_OPENRISC */ 1145 1146 #ifdef TARGET_SH4 1147 1148 #define ELF_START_MMAP 0x80000000 1149 1150 #define ELF_CLASS ELFCLASS32 1151 #define ELF_ARCH EM_SH 1152 1153 static inline void init_thread(struct target_pt_regs *regs, 1154 struct image_info *infop) 1155 { 1156 /* Check other registers XXXXX */ 1157 regs->pc = infop->entry; 1158 regs->regs[15] = infop->start_stack; 1159 } 1160 1161 /* See linux kernel: arch/sh/include/asm/elf.h. */ 1162 #define ELF_NREG 23 1163 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1164 1165 /* See linux kernel: arch/sh/include/asm/ptrace.h. */ 1166 enum { 1167 TARGET_REG_PC = 16, 1168 TARGET_REG_PR = 17, 1169 TARGET_REG_SR = 18, 1170 TARGET_REG_GBR = 19, 1171 TARGET_REG_MACH = 20, 1172 TARGET_REG_MACL = 21, 1173 TARGET_REG_SYSCALL = 22 1174 }; 1175 1176 static inline void elf_core_copy_regs(target_elf_gregset_t *regs, 1177 const CPUSH4State *env) 1178 { 1179 int i; 1180 1181 for (i = 0; i < 16; i++) { 1182 (*regs)[i] = tswapreg(env->gregs[i]); 1183 } 1184 1185 (*regs)[TARGET_REG_PC] = tswapreg(env->pc); 1186 (*regs)[TARGET_REG_PR] = tswapreg(env->pr); 1187 (*regs)[TARGET_REG_SR] = tswapreg(env->sr); 1188 (*regs)[TARGET_REG_GBR] = tswapreg(env->gbr); 1189 (*regs)[TARGET_REG_MACH] = tswapreg(env->mach); 1190 (*regs)[TARGET_REG_MACL] = tswapreg(env->macl); 1191 (*regs)[TARGET_REG_SYSCALL] = 0; /* FIXME */ 1192 } 1193 1194 #define USE_ELF_CORE_DUMP 1195 #define ELF_EXEC_PAGESIZE 4096 1196 1197 enum { 1198 SH_CPU_HAS_FPU = 0x0001, /* Hardware FPU support */ 1199 SH_CPU_HAS_P2_FLUSH_BUG = 0x0002, /* Need to flush the cache in P2 area */ 1200 SH_CPU_HAS_MMU_PAGE_ASSOC = 0x0004, /* SH3: TLB way selection bit support */ 1201 SH_CPU_HAS_DSP = 0x0008, /* SH-DSP: DSP support */ 1202 SH_CPU_HAS_PERF_COUNTER = 0x0010, /* Hardware performance counters */ 1203 SH_CPU_HAS_PTEA = 0x0020, /* PTEA register */ 1204 SH_CPU_HAS_LLSC = 0x0040, /* movli.l/movco.l */ 1205 SH_CPU_HAS_L2_CACHE = 0x0080, /* Secondary cache / URAM */ 1206 SH_CPU_HAS_OP32 = 0x0100, /* 32-bit instruction support */ 1207 SH_CPU_HAS_PTEAEX = 0x0200, /* PTE ASID Extension support */ 1208 }; 1209 1210 #define ELF_HWCAP get_elf_hwcap() 1211 1212 static uint32_t get_elf_hwcap(void) 1213 { 1214 SuperHCPU *cpu = SUPERH_CPU(thread_cpu); 1215 uint32_t hwcap = 0; 1216 1217 hwcap |= SH_CPU_HAS_FPU; 1218 1219 if (cpu->env.features & SH_FEATURE_SH4A) { 1220 hwcap |= SH_CPU_HAS_LLSC; 1221 } 1222 1223 return hwcap; 1224 } 1225 1226 #endif 1227 1228 #ifdef TARGET_CRIS 1229 1230 #define ELF_START_MMAP 0x80000000 1231 1232 #define ELF_CLASS ELFCLASS32 1233 #define ELF_ARCH EM_CRIS 1234 1235 static inline void init_thread(struct target_pt_regs *regs, 1236 struct image_info *infop) 1237 { 1238 regs->erp = infop->entry; 1239 } 1240 1241 #define ELF_EXEC_PAGESIZE 8192 1242 1243 #endif 1244 1245 #ifdef TARGET_M68K 1246 1247 #define ELF_START_MMAP 0x80000000 1248 1249 #define ELF_CLASS ELFCLASS32 1250 #define ELF_ARCH EM_68K 1251 1252 /* ??? Does this need to do anything? 1253 #define ELF_PLAT_INIT(_r) */ 1254 1255 static inline void init_thread(struct target_pt_regs *regs, 1256 struct image_info *infop) 1257 { 1258 regs->usp = infop->start_stack; 1259 regs->sr = 0; 1260 regs->pc = infop->entry; 1261 } 1262 1263 /* See linux kernel: arch/m68k/include/asm/elf.h. */ 1264 #define ELF_NREG 20 1265 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1266 1267 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUM68KState *env) 1268 { 1269 (*regs)[0] = tswapreg(env->dregs[1]); 1270 (*regs)[1] = tswapreg(env->dregs[2]); 1271 (*regs)[2] = tswapreg(env->dregs[3]); 1272 (*regs)[3] = tswapreg(env->dregs[4]); 1273 (*regs)[4] = tswapreg(env->dregs[5]); 1274 (*regs)[5] = tswapreg(env->dregs[6]); 1275 (*regs)[6] = tswapreg(env->dregs[7]); 1276 (*regs)[7] = tswapreg(env->aregs[0]); 1277 (*regs)[8] = tswapreg(env->aregs[1]); 1278 (*regs)[9] = tswapreg(env->aregs[2]); 1279 (*regs)[10] = tswapreg(env->aregs[3]); 1280 (*regs)[11] = tswapreg(env->aregs[4]); 1281 (*regs)[12] = tswapreg(env->aregs[5]); 1282 (*regs)[13] = tswapreg(env->aregs[6]); 1283 (*regs)[14] = tswapreg(env->dregs[0]); 1284 (*regs)[15] = tswapreg(env->aregs[7]); 1285 (*regs)[16] = tswapreg(env->dregs[0]); /* FIXME: orig_d0 */ 1286 (*regs)[17] = tswapreg(env->sr); 1287 (*regs)[18] = tswapreg(env->pc); 1288 (*regs)[19] = 0; /* FIXME: regs->format | regs->vector */ 1289 } 1290 1291 #define USE_ELF_CORE_DUMP 1292 #define ELF_EXEC_PAGESIZE 8192 1293 1294 #endif 1295 1296 #ifdef TARGET_ALPHA 1297 1298 #define ELF_START_MMAP (0x30000000000ULL) 1299 1300 #define ELF_CLASS ELFCLASS64 1301 #define ELF_ARCH EM_ALPHA 1302 1303 static inline void init_thread(struct target_pt_regs *regs, 1304 struct image_info *infop) 1305 { 1306 regs->pc = infop->entry; 1307 regs->ps = 8; 1308 regs->usp = infop->start_stack; 1309 } 1310 1311 #define ELF_EXEC_PAGESIZE 8192 1312 1313 #endif /* TARGET_ALPHA */ 1314 1315 #ifdef TARGET_S390X 1316 1317 #define ELF_START_MMAP (0x20000000000ULL) 1318 1319 #define ELF_CLASS ELFCLASS64 1320 #define ELF_DATA ELFDATA2MSB 1321 #define ELF_ARCH EM_S390 1322 1323 #include "elf.h" 1324 1325 #define ELF_HWCAP get_elf_hwcap() 1326 1327 #define GET_FEATURE(_feat, _hwcap) \ 1328 do { if (s390_has_feat(_feat)) { hwcap |= _hwcap; } } while (0) 1329 1330 static uint32_t get_elf_hwcap(void) 1331 { 1332 /* 1333 * Let's assume we always have esan3 and zarch. 1334 * 31-bit processes can use 64-bit registers (high gprs). 1335 */ 1336 uint32_t hwcap = HWCAP_S390_ESAN3 | HWCAP_S390_ZARCH | HWCAP_S390_HIGH_GPRS; 1337 1338 GET_FEATURE(S390_FEAT_STFLE, HWCAP_S390_STFLE); 1339 GET_FEATURE(S390_FEAT_MSA, HWCAP_S390_MSA); 1340 GET_FEATURE(S390_FEAT_LONG_DISPLACEMENT, HWCAP_S390_LDISP); 1341 GET_FEATURE(S390_FEAT_EXTENDED_IMMEDIATE, HWCAP_S390_EIMM); 1342 if (s390_has_feat(S390_FEAT_EXTENDED_TRANSLATION_3) && 1343 s390_has_feat(S390_FEAT_ETF3_ENH)) { 1344 hwcap |= HWCAP_S390_ETF3EH; 1345 } 1346 GET_FEATURE(S390_FEAT_VECTOR, HWCAP_S390_VXRS); 1347 1348 return hwcap; 1349 } 1350 1351 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop) 1352 { 1353 regs->psw.addr = infop->entry; 1354 regs->psw.mask = PSW_MASK_64 | PSW_MASK_32; 1355 regs->gprs[15] = infop->start_stack; 1356 } 1357 1358 #endif /* TARGET_S390X */ 1359 1360 #ifdef TARGET_TILEGX 1361 1362 /* 42 bits real used address, a half for user mode */ 1363 #define ELF_START_MMAP (0x00000020000000000ULL) 1364 1365 #define elf_check_arch(x) ((x) == EM_TILEGX) 1366 1367 #define ELF_CLASS ELFCLASS64 1368 #define ELF_DATA ELFDATA2LSB 1369 #define ELF_ARCH EM_TILEGX 1370 1371 static inline void init_thread(struct target_pt_regs *regs, 1372 struct image_info *infop) 1373 { 1374 regs->pc = infop->entry; 1375 regs->sp = infop->start_stack; 1376 1377 } 1378 1379 #define ELF_EXEC_PAGESIZE 65536 /* TILE-Gx page size is 64KB */ 1380 1381 #endif /* TARGET_TILEGX */ 1382 1383 #ifdef TARGET_RISCV 1384 1385 #define ELF_START_MMAP 0x80000000 1386 #define ELF_ARCH EM_RISCV 1387 1388 #ifdef TARGET_RISCV32 1389 #define ELF_CLASS ELFCLASS32 1390 #else 1391 #define ELF_CLASS ELFCLASS64 1392 #endif 1393 1394 static inline void init_thread(struct target_pt_regs *regs, 1395 struct image_info *infop) 1396 { 1397 regs->sepc = infop->entry; 1398 regs->sp = infop->start_stack; 1399 } 1400 1401 #define ELF_EXEC_PAGESIZE 4096 1402 1403 #endif /* TARGET_RISCV */ 1404 1405 #ifdef TARGET_HPPA 1406 1407 #define ELF_START_MMAP 0x80000000 1408 #define ELF_CLASS ELFCLASS32 1409 #define ELF_ARCH EM_PARISC 1410 #define ELF_PLATFORM "PARISC" 1411 #define STACK_GROWS_DOWN 0 1412 #define STACK_ALIGNMENT 64 1413 1414 static inline void init_thread(struct target_pt_regs *regs, 1415 struct image_info *infop) 1416 { 1417 regs->iaoq[0] = infop->entry; 1418 regs->iaoq[1] = infop->entry + 4; 1419 regs->gr[23] = 0; 1420 regs->gr[24] = infop->arg_start; 1421 regs->gr[25] = (infop->arg_end - infop->arg_start) / sizeof(abi_ulong); 1422 /* The top-of-stack contains a linkage buffer. */ 1423 regs->gr[30] = infop->start_stack + 64; 1424 regs->gr[31] = infop->entry; 1425 } 1426 1427 #endif /* TARGET_HPPA */ 1428 1429 #ifdef TARGET_XTENSA 1430 1431 #define ELF_START_MMAP 0x20000000 1432 1433 #define ELF_CLASS ELFCLASS32 1434 #define ELF_ARCH EM_XTENSA 1435 1436 static inline void init_thread(struct target_pt_regs *regs, 1437 struct image_info *infop) 1438 { 1439 regs->windowbase = 0; 1440 regs->windowstart = 1; 1441 regs->areg[1] = infop->start_stack; 1442 regs->pc = infop->entry; 1443 } 1444 1445 /* See linux kernel: arch/xtensa/include/asm/elf.h. */ 1446 #define ELF_NREG 128 1447 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1448 1449 enum { 1450 TARGET_REG_PC, 1451 TARGET_REG_PS, 1452 TARGET_REG_LBEG, 1453 TARGET_REG_LEND, 1454 TARGET_REG_LCOUNT, 1455 TARGET_REG_SAR, 1456 TARGET_REG_WINDOWSTART, 1457 TARGET_REG_WINDOWBASE, 1458 TARGET_REG_THREADPTR, 1459 TARGET_REG_AR0 = 64, 1460 }; 1461 1462 static void elf_core_copy_regs(target_elf_gregset_t *regs, 1463 const CPUXtensaState *env) 1464 { 1465 unsigned i; 1466 1467 (*regs)[TARGET_REG_PC] = tswapreg(env->pc); 1468 (*regs)[TARGET_REG_PS] = tswapreg(env->sregs[PS] & ~PS_EXCM); 1469 (*regs)[TARGET_REG_LBEG] = tswapreg(env->sregs[LBEG]); 1470 (*regs)[TARGET_REG_LEND] = tswapreg(env->sregs[LEND]); 1471 (*regs)[TARGET_REG_LCOUNT] = tswapreg(env->sregs[LCOUNT]); 1472 (*regs)[TARGET_REG_SAR] = tswapreg(env->sregs[SAR]); 1473 (*regs)[TARGET_REG_WINDOWSTART] = tswapreg(env->sregs[WINDOW_START]); 1474 (*regs)[TARGET_REG_WINDOWBASE] = tswapreg(env->sregs[WINDOW_BASE]); 1475 (*regs)[TARGET_REG_THREADPTR] = tswapreg(env->uregs[THREADPTR]); 1476 xtensa_sync_phys_from_window((CPUXtensaState *)env); 1477 for (i = 0; i < env->config->nareg; ++i) { 1478 (*regs)[TARGET_REG_AR0 + i] = tswapreg(env->phys_regs[i]); 1479 } 1480 } 1481 1482 #define USE_ELF_CORE_DUMP 1483 #define ELF_EXEC_PAGESIZE 4096 1484 1485 #endif /* TARGET_XTENSA */ 1486 1487 #ifndef ELF_PLATFORM 1488 #define ELF_PLATFORM (NULL) 1489 #endif 1490 1491 #ifndef ELF_MACHINE 1492 #define ELF_MACHINE ELF_ARCH 1493 #endif 1494 1495 #ifndef elf_check_arch 1496 #define elf_check_arch(x) ((x) == ELF_ARCH) 1497 #endif 1498 1499 #ifndef elf_check_abi 1500 #define elf_check_abi(x) (1) 1501 #endif 1502 1503 #ifndef ELF_HWCAP 1504 #define ELF_HWCAP 0 1505 #endif 1506 1507 #ifndef STACK_GROWS_DOWN 1508 #define STACK_GROWS_DOWN 1 1509 #endif 1510 1511 #ifndef STACK_ALIGNMENT 1512 #define STACK_ALIGNMENT 16 1513 #endif 1514 1515 #ifdef TARGET_ABI32 1516 #undef ELF_CLASS 1517 #define ELF_CLASS ELFCLASS32 1518 #undef bswaptls 1519 #define bswaptls(ptr) bswap32s(ptr) 1520 #endif 1521 1522 #include "elf.h" 1523 1524 struct exec 1525 { 1526 unsigned int a_info; /* Use macros N_MAGIC, etc for access */ 1527 unsigned int a_text; /* length of text, in bytes */ 1528 unsigned int a_data; /* length of data, in bytes */ 1529 unsigned int a_bss; /* length of uninitialized data area, in bytes */ 1530 unsigned int a_syms; /* length of symbol table data in file, in bytes */ 1531 unsigned int a_entry; /* start address */ 1532 unsigned int a_trsize; /* length of relocation info for text, in bytes */ 1533 unsigned int a_drsize; /* length of relocation info for data, in bytes */ 1534 }; 1535 1536 1537 #define N_MAGIC(exec) ((exec).a_info & 0xffff) 1538 #define OMAGIC 0407 1539 #define NMAGIC 0410 1540 #define ZMAGIC 0413 1541 #define QMAGIC 0314 1542 1543 /* Necessary parameters */ 1544 #define TARGET_ELF_EXEC_PAGESIZE \ 1545 (((eppnt->p_align & ~qemu_host_page_mask) != 0) ? \ 1546 TARGET_PAGE_SIZE : MAX(qemu_host_page_size, TARGET_PAGE_SIZE)) 1547 #define TARGET_ELF_PAGELENGTH(_v) ROUND_UP((_v), TARGET_ELF_EXEC_PAGESIZE) 1548 #define TARGET_ELF_PAGESTART(_v) ((_v) & \ 1549 ~(abi_ulong)(TARGET_ELF_EXEC_PAGESIZE-1)) 1550 #define TARGET_ELF_PAGEOFFSET(_v) ((_v) & (TARGET_ELF_EXEC_PAGESIZE-1)) 1551 1552 #define DLINFO_ITEMS 16 1553 1554 static inline void memcpy_fromfs(void * to, const void * from, unsigned long n) 1555 { 1556 memcpy(to, from, n); 1557 } 1558 1559 #ifdef BSWAP_NEEDED 1560 static void bswap_ehdr(struct elfhdr *ehdr) 1561 { 1562 bswap16s(&ehdr->e_type); /* Object file type */ 1563 bswap16s(&ehdr->e_machine); /* Architecture */ 1564 bswap32s(&ehdr->e_version); /* Object file version */ 1565 bswaptls(&ehdr->e_entry); /* Entry point virtual address */ 1566 bswaptls(&ehdr->e_phoff); /* Program header table file offset */ 1567 bswaptls(&ehdr->e_shoff); /* Section header table file offset */ 1568 bswap32s(&ehdr->e_flags); /* Processor-specific flags */ 1569 bswap16s(&ehdr->e_ehsize); /* ELF header size in bytes */ 1570 bswap16s(&ehdr->e_phentsize); /* Program header table entry size */ 1571 bswap16s(&ehdr->e_phnum); /* Program header table entry count */ 1572 bswap16s(&ehdr->e_shentsize); /* Section header table entry size */ 1573 bswap16s(&ehdr->e_shnum); /* Section header table entry count */ 1574 bswap16s(&ehdr->e_shstrndx); /* Section header string table index */ 1575 } 1576 1577 static void bswap_phdr(struct elf_phdr *phdr, int phnum) 1578 { 1579 int i; 1580 for (i = 0; i < phnum; ++i, ++phdr) { 1581 bswap32s(&phdr->p_type); /* Segment type */ 1582 bswap32s(&phdr->p_flags); /* Segment flags */ 1583 bswaptls(&phdr->p_offset); /* Segment file offset */ 1584 bswaptls(&phdr->p_vaddr); /* Segment virtual address */ 1585 bswaptls(&phdr->p_paddr); /* Segment physical address */ 1586 bswaptls(&phdr->p_filesz); /* Segment size in file */ 1587 bswaptls(&phdr->p_memsz); /* Segment size in memory */ 1588 bswaptls(&phdr->p_align); /* Segment alignment */ 1589 } 1590 } 1591 1592 static void bswap_shdr(struct elf_shdr *shdr, int shnum) 1593 { 1594 int i; 1595 for (i = 0; i < shnum; ++i, ++shdr) { 1596 bswap32s(&shdr->sh_name); 1597 bswap32s(&shdr->sh_type); 1598 bswaptls(&shdr->sh_flags); 1599 bswaptls(&shdr->sh_addr); 1600 bswaptls(&shdr->sh_offset); 1601 bswaptls(&shdr->sh_size); 1602 bswap32s(&shdr->sh_link); 1603 bswap32s(&shdr->sh_info); 1604 bswaptls(&shdr->sh_addralign); 1605 bswaptls(&shdr->sh_entsize); 1606 } 1607 } 1608 1609 static void bswap_sym(struct elf_sym *sym) 1610 { 1611 bswap32s(&sym->st_name); 1612 bswaptls(&sym->st_value); 1613 bswaptls(&sym->st_size); 1614 bswap16s(&sym->st_shndx); 1615 } 1616 1617 #ifdef TARGET_MIPS 1618 static void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags) 1619 { 1620 bswap16s(&abiflags->version); 1621 bswap32s(&abiflags->ases); 1622 bswap32s(&abiflags->isa_ext); 1623 bswap32s(&abiflags->flags1); 1624 bswap32s(&abiflags->flags2); 1625 } 1626 #endif 1627 #else 1628 static inline void bswap_ehdr(struct elfhdr *ehdr) { } 1629 static inline void bswap_phdr(struct elf_phdr *phdr, int phnum) { } 1630 static inline void bswap_shdr(struct elf_shdr *shdr, int shnum) { } 1631 static inline void bswap_sym(struct elf_sym *sym) { } 1632 #ifdef TARGET_MIPS 1633 static inline void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags) { } 1634 #endif 1635 #endif 1636 1637 #ifdef USE_ELF_CORE_DUMP 1638 static int elf_core_dump(int, const CPUArchState *); 1639 #endif /* USE_ELF_CORE_DUMP */ 1640 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias); 1641 1642 /* Verify the portions of EHDR within E_IDENT for the target. 1643 This can be performed before bswapping the entire header. */ 1644 static bool elf_check_ident(struct elfhdr *ehdr) 1645 { 1646 return (ehdr->e_ident[EI_MAG0] == ELFMAG0 1647 && ehdr->e_ident[EI_MAG1] == ELFMAG1 1648 && ehdr->e_ident[EI_MAG2] == ELFMAG2 1649 && ehdr->e_ident[EI_MAG3] == ELFMAG3 1650 && ehdr->e_ident[EI_CLASS] == ELF_CLASS 1651 && ehdr->e_ident[EI_DATA] == ELF_DATA 1652 && ehdr->e_ident[EI_VERSION] == EV_CURRENT); 1653 } 1654 1655 /* Verify the portions of EHDR outside of E_IDENT for the target. 1656 This has to wait until after bswapping the header. */ 1657 static bool elf_check_ehdr(struct elfhdr *ehdr) 1658 { 1659 return (elf_check_arch(ehdr->e_machine) 1660 && elf_check_abi(ehdr->e_flags) 1661 && ehdr->e_ehsize == sizeof(struct elfhdr) 1662 && ehdr->e_phentsize == sizeof(struct elf_phdr) 1663 && (ehdr->e_type == ET_EXEC || ehdr->e_type == ET_DYN)); 1664 } 1665 1666 /* 1667 * 'copy_elf_strings()' copies argument/envelope strings from user 1668 * memory to free pages in kernel mem. These are in a format ready 1669 * to be put directly into the top of new user memory. 1670 * 1671 */ 1672 static abi_ulong copy_elf_strings(int argc, char **argv, char *scratch, 1673 abi_ulong p, abi_ulong stack_limit) 1674 { 1675 char *tmp; 1676 int len, i; 1677 abi_ulong top = p; 1678 1679 if (!p) { 1680 return 0; /* bullet-proofing */ 1681 } 1682 1683 if (STACK_GROWS_DOWN) { 1684 int offset = ((p - 1) % TARGET_PAGE_SIZE) + 1; 1685 for (i = argc - 1; i >= 0; --i) { 1686 tmp = argv[i]; 1687 if (!tmp) { 1688 fprintf(stderr, "VFS: argc is wrong"); 1689 exit(-1); 1690 } 1691 len = strlen(tmp) + 1; 1692 tmp += len; 1693 1694 if (len > (p - stack_limit)) { 1695 return 0; 1696 } 1697 while (len) { 1698 int bytes_to_copy = (len > offset) ? offset : len; 1699 tmp -= bytes_to_copy; 1700 p -= bytes_to_copy; 1701 offset -= bytes_to_copy; 1702 len -= bytes_to_copy; 1703 1704 memcpy_fromfs(scratch + offset, tmp, bytes_to_copy); 1705 1706 if (offset == 0) { 1707 memcpy_to_target(p, scratch, top - p); 1708 top = p; 1709 offset = TARGET_PAGE_SIZE; 1710 } 1711 } 1712 } 1713 if (p != top) { 1714 memcpy_to_target(p, scratch + offset, top - p); 1715 } 1716 } else { 1717 int remaining = TARGET_PAGE_SIZE - (p % TARGET_PAGE_SIZE); 1718 for (i = 0; i < argc; ++i) { 1719 tmp = argv[i]; 1720 if (!tmp) { 1721 fprintf(stderr, "VFS: argc is wrong"); 1722 exit(-1); 1723 } 1724 len = strlen(tmp) + 1; 1725 if (len > (stack_limit - p)) { 1726 return 0; 1727 } 1728 while (len) { 1729 int bytes_to_copy = (len > remaining) ? remaining : len; 1730 1731 memcpy_fromfs(scratch + (p - top), tmp, bytes_to_copy); 1732 1733 tmp += bytes_to_copy; 1734 remaining -= bytes_to_copy; 1735 p += bytes_to_copy; 1736 len -= bytes_to_copy; 1737 1738 if (remaining == 0) { 1739 memcpy_to_target(top, scratch, p - top); 1740 top = p; 1741 remaining = TARGET_PAGE_SIZE; 1742 } 1743 } 1744 } 1745 if (p != top) { 1746 memcpy_to_target(top, scratch, p - top); 1747 } 1748 } 1749 1750 return p; 1751 } 1752 1753 /* Older linux kernels provide up to MAX_ARG_PAGES (default: 32) of 1754 * argument/environment space. Newer kernels (>2.6.33) allow more, 1755 * dependent on stack size, but guarantee at least 32 pages for 1756 * backwards compatibility. 1757 */ 1758 #define STACK_LOWER_LIMIT (32 * TARGET_PAGE_SIZE) 1759 1760 static abi_ulong setup_arg_pages(struct linux_binprm *bprm, 1761 struct image_info *info) 1762 { 1763 abi_ulong size, error, guard; 1764 1765 size = guest_stack_size; 1766 if (size < STACK_LOWER_LIMIT) { 1767 size = STACK_LOWER_LIMIT; 1768 } 1769 guard = TARGET_PAGE_SIZE; 1770 if (guard < qemu_real_host_page_size) { 1771 guard = qemu_real_host_page_size; 1772 } 1773 1774 error = target_mmap(0, size + guard, PROT_READ | PROT_WRITE, 1775 MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); 1776 if (error == -1) { 1777 perror("mmap stack"); 1778 exit(-1); 1779 } 1780 1781 /* We reserve one extra page at the top of the stack as guard. */ 1782 if (STACK_GROWS_DOWN) { 1783 target_mprotect(error, guard, PROT_NONE); 1784 info->stack_limit = error + guard; 1785 return info->stack_limit + size - sizeof(void *); 1786 } else { 1787 target_mprotect(error + size, guard, PROT_NONE); 1788 info->stack_limit = error + size; 1789 return error; 1790 } 1791 } 1792 1793 /* Map and zero the bss. We need to explicitly zero any fractional pages 1794 after the data section (i.e. bss). */ 1795 static void zero_bss(abi_ulong elf_bss, abi_ulong last_bss, int prot) 1796 { 1797 uintptr_t host_start, host_map_start, host_end; 1798 1799 last_bss = TARGET_PAGE_ALIGN(last_bss); 1800 1801 /* ??? There is confusion between qemu_real_host_page_size and 1802 qemu_host_page_size here and elsewhere in target_mmap, which 1803 may lead to the end of the data section mapping from the file 1804 not being mapped. At least there was an explicit test and 1805 comment for that here, suggesting that "the file size must 1806 be known". The comment probably pre-dates the introduction 1807 of the fstat system call in target_mmap which does in fact 1808 find out the size. What isn't clear is if the workaround 1809 here is still actually needed. For now, continue with it, 1810 but merge it with the "normal" mmap that would allocate the bss. */ 1811 1812 host_start = (uintptr_t) g2h(elf_bss); 1813 host_end = (uintptr_t) g2h(last_bss); 1814 host_map_start = REAL_HOST_PAGE_ALIGN(host_start); 1815 1816 if (host_map_start < host_end) { 1817 void *p = mmap((void *)host_map_start, host_end - host_map_start, 1818 prot, MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); 1819 if (p == MAP_FAILED) { 1820 perror("cannot mmap brk"); 1821 exit(-1); 1822 } 1823 } 1824 1825 /* Ensure that the bss page(s) are valid */ 1826 if ((page_get_flags(last_bss-1) & prot) != prot) { 1827 page_set_flags(elf_bss & TARGET_PAGE_MASK, last_bss, prot | PAGE_VALID); 1828 } 1829 1830 if (host_start < host_map_start) { 1831 memset((void *)host_start, 0, host_map_start - host_start); 1832 } 1833 } 1834 1835 #ifdef TARGET_ARM 1836 static int elf_is_fdpic(struct elfhdr *exec) 1837 { 1838 return exec->e_ident[EI_OSABI] == ELFOSABI_ARM_FDPIC; 1839 } 1840 #else 1841 /* Default implementation, always false. */ 1842 static int elf_is_fdpic(struct elfhdr *exec) 1843 { 1844 return 0; 1845 } 1846 #endif 1847 1848 static abi_ulong loader_build_fdpic_loadmap(struct image_info *info, abi_ulong sp) 1849 { 1850 uint16_t n; 1851 struct elf32_fdpic_loadseg *loadsegs = info->loadsegs; 1852 1853 /* elf32_fdpic_loadseg */ 1854 n = info->nsegs; 1855 while (n--) { 1856 sp -= 12; 1857 put_user_u32(loadsegs[n].addr, sp+0); 1858 put_user_u32(loadsegs[n].p_vaddr, sp+4); 1859 put_user_u32(loadsegs[n].p_memsz, sp+8); 1860 } 1861 1862 /* elf32_fdpic_loadmap */ 1863 sp -= 4; 1864 put_user_u16(0, sp+0); /* version */ 1865 put_user_u16(info->nsegs, sp+2); /* nsegs */ 1866 1867 info->personality = PER_LINUX_FDPIC; 1868 info->loadmap_addr = sp; 1869 1870 return sp; 1871 } 1872 1873 static abi_ulong create_elf_tables(abi_ulong p, int argc, int envc, 1874 struct elfhdr *exec, 1875 struct image_info *info, 1876 struct image_info *interp_info) 1877 { 1878 abi_ulong sp; 1879 abi_ulong u_argc, u_argv, u_envp, u_auxv; 1880 int size; 1881 int i; 1882 abi_ulong u_rand_bytes; 1883 uint8_t k_rand_bytes[16]; 1884 abi_ulong u_platform; 1885 const char *k_platform; 1886 const int n = sizeof(elf_addr_t); 1887 1888 sp = p; 1889 1890 /* Needs to be before we load the env/argc/... */ 1891 if (elf_is_fdpic(exec)) { 1892 /* Need 4 byte alignment for these structs */ 1893 sp &= ~3; 1894 sp = loader_build_fdpic_loadmap(info, sp); 1895 info->other_info = interp_info; 1896 if (interp_info) { 1897 interp_info->other_info = info; 1898 sp = loader_build_fdpic_loadmap(interp_info, sp); 1899 info->interpreter_loadmap_addr = interp_info->loadmap_addr; 1900 info->interpreter_pt_dynamic_addr = interp_info->pt_dynamic_addr; 1901 } else { 1902 info->interpreter_loadmap_addr = 0; 1903 info->interpreter_pt_dynamic_addr = 0; 1904 } 1905 } 1906 1907 u_platform = 0; 1908 k_platform = ELF_PLATFORM; 1909 if (k_platform) { 1910 size_t len = strlen(k_platform) + 1; 1911 if (STACK_GROWS_DOWN) { 1912 sp -= (len + n - 1) & ~(n - 1); 1913 u_platform = sp; 1914 /* FIXME - check return value of memcpy_to_target() for failure */ 1915 memcpy_to_target(sp, k_platform, len); 1916 } else { 1917 memcpy_to_target(sp, k_platform, len); 1918 u_platform = sp; 1919 sp += len + 1; 1920 } 1921 } 1922 1923 /* Provide 16 byte alignment for the PRNG, and basic alignment for 1924 * the argv and envp pointers. 1925 */ 1926 if (STACK_GROWS_DOWN) { 1927 sp = QEMU_ALIGN_DOWN(sp, 16); 1928 } else { 1929 sp = QEMU_ALIGN_UP(sp, 16); 1930 } 1931 1932 /* 1933 * Generate 16 random bytes for userspace PRNG seeding. 1934 */ 1935 qemu_guest_getrandom_nofail(k_rand_bytes, sizeof(k_rand_bytes)); 1936 if (STACK_GROWS_DOWN) { 1937 sp -= 16; 1938 u_rand_bytes = sp; 1939 /* FIXME - check return value of memcpy_to_target() for failure */ 1940 memcpy_to_target(sp, k_rand_bytes, 16); 1941 } else { 1942 memcpy_to_target(sp, k_rand_bytes, 16); 1943 u_rand_bytes = sp; 1944 sp += 16; 1945 } 1946 1947 size = (DLINFO_ITEMS + 1) * 2; 1948 if (k_platform) 1949 size += 2; 1950 #ifdef DLINFO_ARCH_ITEMS 1951 size += DLINFO_ARCH_ITEMS * 2; 1952 #endif 1953 #ifdef ELF_HWCAP2 1954 size += 2; 1955 #endif 1956 info->auxv_len = size * n; 1957 1958 size += envc + argc + 2; 1959 size += 1; /* argc itself */ 1960 size *= n; 1961 1962 /* Allocate space and finalize stack alignment for entry now. */ 1963 if (STACK_GROWS_DOWN) { 1964 u_argc = QEMU_ALIGN_DOWN(sp - size, STACK_ALIGNMENT); 1965 sp = u_argc; 1966 } else { 1967 u_argc = sp; 1968 sp = QEMU_ALIGN_UP(sp + size, STACK_ALIGNMENT); 1969 } 1970 1971 u_argv = u_argc + n; 1972 u_envp = u_argv + (argc + 1) * n; 1973 u_auxv = u_envp + (envc + 1) * n; 1974 info->saved_auxv = u_auxv; 1975 info->arg_start = u_argv; 1976 info->arg_end = u_argv + argc * n; 1977 1978 /* This is correct because Linux defines 1979 * elf_addr_t as Elf32_Off / Elf64_Off 1980 */ 1981 #define NEW_AUX_ENT(id, val) do { \ 1982 put_user_ual(id, u_auxv); u_auxv += n; \ 1983 put_user_ual(val, u_auxv); u_auxv += n; \ 1984 } while(0) 1985 1986 #ifdef ARCH_DLINFO 1987 /* 1988 * ARCH_DLINFO must come first so platform specific code can enforce 1989 * special alignment requirements on the AUXV if necessary (eg. PPC). 1990 */ 1991 ARCH_DLINFO; 1992 #endif 1993 /* There must be exactly DLINFO_ITEMS entries here, or the assert 1994 * on info->auxv_len will trigger. 1995 */ 1996 NEW_AUX_ENT(AT_PHDR, (abi_ulong)(info->load_addr + exec->e_phoff)); 1997 NEW_AUX_ENT(AT_PHENT, (abi_ulong)(sizeof (struct elf_phdr))); 1998 NEW_AUX_ENT(AT_PHNUM, (abi_ulong)(exec->e_phnum)); 1999 if ((info->alignment & ~qemu_host_page_mask) != 0) { 2000 /* Target doesn't support host page size alignment */ 2001 NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(TARGET_PAGE_SIZE)); 2002 } else { 2003 NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(MAX(TARGET_PAGE_SIZE, 2004 qemu_host_page_size))); 2005 } 2006 NEW_AUX_ENT(AT_BASE, (abi_ulong)(interp_info ? interp_info->load_addr : 0)); 2007 NEW_AUX_ENT(AT_FLAGS, (abi_ulong)0); 2008 NEW_AUX_ENT(AT_ENTRY, info->entry); 2009 NEW_AUX_ENT(AT_UID, (abi_ulong) getuid()); 2010 NEW_AUX_ENT(AT_EUID, (abi_ulong) geteuid()); 2011 NEW_AUX_ENT(AT_GID, (abi_ulong) getgid()); 2012 NEW_AUX_ENT(AT_EGID, (abi_ulong) getegid()); 2013 NEW_AUX_ENT(AT_HWCAP, (abi_ulong) ELF_HWCAP); 2014 NEW_AUX_ENT(AT_CLKTCK, (abi_ulong) sysconf(_SC_CLK_TCK)); 2015 NEW_AUX_ENT(AT_RANDOM, (abi_ulong) u_rand_bytes); 2016 NEW_AUX_ENT(AT_SECURE, (abi_ulong) qemu_getauxval(AT_SECURE)); 2017 NEW_AUX_ENT(AT_EXECFN, info->file_string); 2018 2019 #ifdef ELF_HWCAP2 2020 NEW_AUX_ENT(AT_HWCAP2, (abi_ulong) ELF_HWCAP2); 2021 #endif 2022 2023 if (u_platform) { 2024 NEW_AUX_ENT(AT_PLATFORM, u_platform); 2025 } 2026 NEW_AUX_ENT (AT_NULL, 0); 2027 #undef NEW_AUX_ENT 2028 2029 /* Check that our initial calculation of the auxv length matches how much 2030 * we actually put into it. 2031 */ 2032 assert(info->auxv_len == u_auxv - info->saved_auxv); 2033 2034 put_user_ual(argc, u_argc); 2035 2036 p = info->arg_strings; 2037 for (i = 0; i < argc; ++i) { 2038 put_user_ual(p, u_argv); 2039 u_argv += n; 2040 p += target_strlen(p) + 1; 2041 } 2042 put_user_ual(0, u_argv); 2043 2044 p = info->env_strings; 2045 for (i = 0; i < envc; ++i) { 2046 put_user_ual(p, u_envp); 2047 u_envp += n; 2048 p += target_strlen(p) + 1; 2049 } 2050 put_user_ual(0, u_envp); 2051 2052 return sp; 2053 } 2054 2055 #ifndef ARM_COMMPAGE 2056 #define ARM_COMMPAGE 0 2057 #define init_guest_commpage() true 2058 #endif 2059 2060 static void pgb_fail_in_use(const char *image_name) 2061 { 2062 error_report("%s: requires virtual address space that is in use " 2063 "(omit the -B option or choose a different value)", 2064 image_name); 2065 exit(EXIT_FAILURE); 2066 } 2067 2068 static void pgb_have_guest_base(const char *image_name, abi_ulong guest_loaddr, 2069 abi_ulong guest_hiaddr, long align) 2070 { 2071 const int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE; 2072 void *addr, *test; 2073 2074 if (!QEMU_IS_ALIGNED(guest_base, align)) { 2075 fprintf(stderr, "Requested guest base 0x%lx does not satisfy " 2076 "host minimum alignment (0x%lx)\n", 2077 guest_base, align); 2078 exit(EXIT_FAILURE); 2079 } 2080 2081 /* Sanity check the guest binary. */ 2082 if (reserved_va) { 2083 if (guest_hiaddr > reserved_va) { 2084 error_report("%s: requires more than reserved virtual " 2085 "address space (0x%" PRIx64 " > 0x%lx)", 2086 image_name, (uint64_t)guest_hiaddr, reserved_va); 2087 exit(EXIT_FAILURE); 2088 } 2089 } else { 2090 #if HOST_LONG_BITS < TARGET_ABI_BITS 2091 if ((guest_hiaddr - guest_base) > ~(uintptr_t)0) { 2092 error_report("%s: requires more virtual address space " 2093 "than the host can provide (0x%" PRIx64 ")", 2094 image_name, (uint64_t)guest_hiaddr - guest_base); 2095 exit(EXIT_FAILURE); 2096 } 2097 #endif 2098 } 2099 2100 /* 2101 * Expand the allocation to the entire reserved_va. 2102 * Exclude the mmap_min_addr hole. 2103 */ 2104 if (reserved_va) { 2105 guest_loaddr = (guest_base >= mmap_min_addr ? 0 2106 : mmap_min_addr - guest_base); 2107 guest_hiaddr = reserved_va; 2108 } 2109 2110 /* Reserve the address space for the binary, or reserved_va. */ 2111 test = g2h(guest_loaddr); 2112 addr = mmap(test, guest_hiaddr - guest_loaddr, PROT_NONE, flags, -1, 0); 2113 if (test != addr) { 2114 pgb_fail_in_use(image_name); 2115 } 2116 } 2117 2118 /** 2119 * pgd_find_hole_fallback: potential mmap address 2120 * @guest_size: size of available space 2121 * @brk: location of break 2122 * @align: memory alignment 2123 * 2124 * This is a fallback method for finding a hole in the host address 2125 * space if we don't have the benefit of being able to access 2126 * /proc/self/map. It can potentially take a very long time as we can 2127 * only dumbly iterate up the host address space seeing if the 2128 * allocation would work. 2129 */ 2130 static uintptr_t pgd_find_hole_fallback(uintptr_t guest_size, uintptr_t brk, 2131 long align, uintptr_t offset) 2132 { 2133 uintptr_t base; 2134 2135 /* Start (aligned) at the bottom and work our way up */ 2136 base = ROUND_UP(mmap_min_addr, align); 2137 2138 while (true) { 2139 uintptr_t align_start, end; 2140 align_start = ROUND_UP(base, align); 2141 end = align_start + guest_size + offset; 2142 2143 /* if brk is anywhere in the range give ourselves some room to grow. */ 2144 if (align_start <= brk && brk < end) { 2145 base = brk + (16 * MiB); 2146 continue; 2147 } else if (align_start + guest_size < align_start) { 2148 /* we have run out of space */ 2149 return -1; 2150 } else { 2151 int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE | 2152 MAP_FIXED_NOREPLACE; 2153 void * mmap_start = mmap((void *) align_start, guest_size, 2154 PROT_NONE, flags, -1, 0); 2155 if (mmap_start != MAP_FAILED) { 2156 munmap((void *) align_start, guest_size); 2157 if (MAP_FIXED_NOREPLACE || mmap_start == (void *) align_start) { 2158 return (uintptr_t) mmap_start + offset; 2159 } 2160 } 2161 base += qemu_host_page_size; 2162 } 2163 } 2164 } 2165 2166 /* Return value for guest_base, or -1 if no hole found. */ 2167 static uintptr_t pgb_find_hole(uintptr_t guest_loaddr, uintptr_t guest_size, 2168 long align, uintptr_t offset) 2169 { 2170 GSList *maps, *iter; 2171 uintptr_t this_start, this_end, next_start, brk; 2172 intptr_t ret = -1; 2173 2174 assert(QEMU_IS_ALIGNED(guest_loaddr, align)); 2175 2176 maps = read_self_maps(); 2177 2178 /* Read brk after we've read the maps, which will malloc. */ 2179 brk = (uintptr_t)sbrk(0); 2180 2181 if (!maps) { 2182 return pgd_find_hole_fallback(guest_size, brk, align, offset); 2183 } 2184 2185 /* The first hole is before the first map entry. */ 2186 this_start = mmap_min_addr; 2187 2188 for (iter = maps; iter; 2189 this_start = next_start, iter = g_slist_next(iter)) { 2190 uintptr_t align_start, hole_size; 2191 2192 this_end = ((MapInfo *)iter->data)->start; 2193 next_start = ((MapInfo *)iter->data)->end; 2194 align_start = ROUND_UP(this_start + offset, align); 2195 2196 /* Skip holes that are too small. */ 2197 if (align_start >= this_end) { 2198 continue; 2199 } 2200 hole_size = this_end - align_start; 2201 if (hole_size < guest_size) { 2202 continue; 2203 } 2204 2205 /* If this hole contains brk, give ourselves some room to grow. */ 2206 if (this_start <= brk && brk < this_end) { 2207 hole_size -= guest_size; 2208 if (sizeof(uintptr_t) == 8 && hole_size >= 1 * GiB) { 2209 align_start += 1 * GiB; 2210 } else if (hole_size >= 16 * MiB) { 2211 align_start += 16 * MiB; 2212 } else { 2213 align_start = (this_end - guest_size) & -align; 2214 if (align_start < this_start) { 2215 continue; 2216 } 2217 } 2218 } 2219 2220 /* Record the lowest successful match. */ 2221 if (ret < 0) { 2222 ret = align_start - guest_loaddr; 2223 } 2224 /* If this hole contains the identity map, select it. */ 2225 if (align_start <= guest_loaddr && 2226 guest_loaddr + guest_size <= this_end) { 2227 ret = 0; 2228 } 2229 /* If this hole ends above the identity map, stop looking. */ 2230 if (this_end >= guest_loaddr) { 2231 break; 2232 } 2233 } 2234 free_self_maps(maps); 2235 2236 return ret; 2237 } 2238 2239 static void pgb_static(const char *image_name, abi_ulong orig_loaddr, 2240 abi_ulong orig_hiaddr, long align) 2241 { 2242 uintptr_t loaddr = orig_loaddr; 2243 uintptr_t hiaddr = orig_hiaddr; 2244 uintptr_t offset = 0; 2245 uintptr_t addr; 2246 2247 if (hiaddr != orig_hiaddr) { 2248 error_report("%s: requires virtual address space that the " 2249 "host cannot provide (0x%" PRIx64 ")", 2250 image_name, (uint64_t)orig_hiaddr); 2251 exit(EXIT_FAILURE); 2252 } 2253 2254 loaddr &= -align; 2255 if (ARM_COMMPAGE) { 2256 /* 2257 * Extend the allocation to include the commpage. 2258 * For a 64-bit host, this is just 4GiB; for a 32-bit host we 2259 * need to ensure there is space bellow the guest_base so we 2260 * can map the commpage in the place needed when the address 2261 * arithmetic wraps around. 2262 */ 2263 if (sizeof(uintptr_t) == 8 || loaddr >= 0x80000000u) { 2264 hiaddr = (uintptr_t) 4 << 30; 2265 } else { 2266 offset = -(ARM_COMMPAGE & -align); 2267 } 2268 } 2269 2270 addr = pgb_find_hole(loaddr, hiaddr - loaddr, align, offset); 2271 if (addr == -1) { 2272 /* 2273 * If ARM_COMMPAGE, there *might* be a non-consecutive allocation 2274 * that can satisfy both. But as the normal arm32 link base address 2275 * is ~32k, and we extend down to include the commpage, making the 2276 * overhead only ~96k, this is unlikely. 2277 */ 2278 error_report("%s: Unable to allocate %#zx bytes of " 2279 "virtual address space", image_name, 2280 (size_t)(hiaddr - loaddr)); 2281 exit(EXIT_FAILURE); 2282 } 2283 2284 guest_base = addr; 2285 } 2286 2287 static void pgb_dynamic(const char *image_name, long align) 2288 { 2289 /* 2290 * The executable is dynamic and does not require a fixed address. 2291 * All we need is a commpage that satisfies align. 2292 * If we do not need a commpage, leave guest_base == 0. 2293 */ 2294 if (ARM_COMMPAGE) { 2295 uintptr_t addr, commpage; 2296 2297 /* 64-bit hosts should have used reserved_va. */ 2298 assert(sizeof(uintptr_t) == 4); 2299 2300 /* 2301 * By putting the commpage at the first hole, that puts guest_base 2302 * just above that, and maximises the positive guest addresses. 2303 */ 2304 commpage = ARM_COMMPAGE & -align; 2305 addr = pgb_find_hole(commpage, -commpage, align, 0); 2306 assert(addr != -1); 2307 guest_base = addr; 2308 } 2309 } 2310 2311 static void pgb_reserved_va(const char *image_name, abi_ulong guest_loaddr, 2312 abi_ulong guest_hiaddr, long align) 2313 { 2314 int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE; 2315 void *addr, *test; 2316 2317 if (guest_hiaddr > reserved_va) { 2318 error_report("%s: requires more than reserved virtual " 2319 "address space (0x%" PRIx64 " > 0x%lx)", 2320 image_name, (uint64_t)guest_hiaddr, reserved_va); 2321 exit(EXIT_FAILURE); 2322 } 2323 2324 /* Widen the "image" to the entire reserved address space. */ 2325 pgb_static(image_name, 0, reserved_va, align); 2326 2327 /* osdep.h defines this as 0 if it's missing */ 2328 flags |= MAP_FIXED_NOREPLACE; 2329 2330 /* Reserve the memory on the host. */ 2331 assert(guest_base != 0); 2332 test = g2h(0); 2333 addr = mmap(test, reserved_va, PROT_NONE, flags, -1, 0); 2334 if (addr == MAP_FAILED) { 2335 error_report("Unable to reserve 0x%lx bytes of virtual address " 2336 "space (%s) for use as guest address space (check your " 2337 "virtual memory ulimit setting or reserve less " 2338 "using -R option)", reserved_va, strerror(errno)); 2339 exit(EXIT_FAILURE); 2340 } 2341 assert(addr == test); 2342 } 2343 2344 void probe_guest_base(const char *image_name, abi_ulong guest_loaddr, 2345 abi_ulong guest_hiaddr) 2346 { 2347 /* In order to use host shmat, we must be able to honor SHMLBA. */ 2348 uintptr_t align = MAX(SHMLBA, qemu_host_page_size); 2349 2350 if (have_guest_base) { 2351 pgb_have_guest_base(image_name, guest_loaddr, guest_hiaddr, align); 2352 } else if (reserved_va) { 2353 pgb_reserved_va(image_name, guest_loaddr, guest_hiaddr, align); 2354 } else if (guest_loaddr) { 2355 pgb_static(image_name, guest_loaddr, guest_hiaddr, align); 2356 } else { 2357 pgb_dynamic(image_name, align); 2358 } 2359 2360 /* Reserve and initialize the commpage. */ 2361 if (!init_guest_commpage()) { 2362 /* 2363 * With have_guest_base, the user has selected the address and 2364 * we are trying to work with that. Otherwise, we have selected 2365 * free space and init_guest_commpage must succeeded. 2366 */ 2367 assert(have_guest_base); 2368 pgb_fail_in_use(image_name); 2369 } 2370 2371 assert(QEMU_IS_ALIGNED(guest_base, align)); 2372 qemu_log_mask(CPU_LOG_PAGE, "Locating guest address space " 2373 "@ 0x%" PRIx64 "\n", (uint64_t)guest_base); 2374 } 2375 2376 /* Load an ELF image into the address space. 2377 2378 IMAGE_NAME is the filename of the image, to use in error messages. 2379 IMAGE_FD is the open file descriptor for the image. 2380 2381 BPRM_BUF is a copy of the beginning of the file; this of course 2382 contains the elf file header at offset 0. It is assumed that this 2383 buffer is sufficiently aligned to present no problems to the host 2384 in accessing data at aligned offsets within the buffer. 2385 2386 On return: INFO values will be filled in, as necessary or available. */ 2387 2388 static void load_elf_image(const char *image_name, int image_fd, 2389 struct image_info *info, char **pinterp_name, 2390 char bprm_buf[BPRM_BUF_SIZE]) 2391 { 2392 struct elfhdr *ehdr = (struct elfhdr *)bprm_buf; 2393 struct elf_phdr *phdr; 2394 abi_ulong load_addr, load_bias, loaddr, hiaddr, error; 2395 int i, retval; 2396 const char *errmsg; 2397 2398 /* First of all, some simple consistency checks */ 2399 errmsg = "Invalid ELF image for this architecture"; 2400 if (!elf_check_ident(ehdr)) { 2401 goto exit_errmsg; 2402 } 2403 bswap_ehdr(ehdr); 2404 if (!elf_check_ehdr(ehdr)) { 2405 goto exit_errmsg; 2406 } 2407 2408 i = ehdr->e_phnum * sizeof(struct elf_phdr); 2409 if (ehdr->e_phoff + i <= BPRM_BUF_SIZE) { 2410 phdr = (struct elf_phdr *)(bprm_buf + ehdr->e_phoff); 2411 } else { 2412 phdr = (struct elf_phdr *) alloca(i); 2413 retval = pread(image_fd, phdr, i, ehdr->e_phoff); 2414 if (retval != i) { 2415 goto exit_read; 2416 } 2417 } 2418 bswap_phdr(phdr, ehdr->e_phnum); 2419 2420 info->nsegs = 0; 2421 info->pt_dynamic_addr = 0; 2422 2423 mmap_lock(); 2424 2425 /* Find the maximum size of the image and allocate an appropriate 2426 amount of memory to handle that. */ 2427 loaddr = -1, hiaddr = 0; 2428 info->alignment = 0; 2429 for (i = 0; i < ehdr->e_phnum; ++i) { 2430 if (phdr[i].p_type == PT_LOAD) { 2431 abi_ulong a = phdr[i].p_vaddr - phdr[i].p_offset; 2432 if (a < loaddr) { 2433 loaddr = a; 2434 } 2435 a = phdr[i].p_vaddr + phdr[i].p_memsz; 2436 if (a > hiaddr) { 2437 hiaddr = a; 2438 } 2439 ++info->nsegs; 2440 info->alignment |= phdr[i].p_align; 2441 } 2442 } 2443 2444 if (pinterp_name != NULL) { 2445 /* 2446 * This is the main executable. 2447 * 2448 * Reserve extra space for brk. 2449 * We hold on to this space while placing the interpreter 2450 * and the stack, lest they be placed immediately after 2451 * the data segment and block allocation from the brk. 2452 * 2453 * 16MB is chosen as "large enough" without being so large 2454 * as to allow the result to not fit with a 32-bit guest on 2455 * a 32-bit host. 2456 */ 2457 info->reserve_brk = 16 * MiB; 2458 hiaddr += info->reserve_brk; 2459 2460 if (ehdr->e_type == ET_EXEC) { 2461 /* 2462 * Make sure that the low address does not conflict with 2463 * MMAP_MIN_ADDR or the QEMU application itself. 2464 */ 2465 probe_guest_base(image_name, loaddr, hiaddr); 2466 } else { 2467 /* 2468 * The binary is dynamic, but we still need to 2469 * select guest_base. In this case we pass a size. 2470 */ 2471 probe_guest_base(image_name, 0, hiaddr - loaddr); 2472 } 2473 } 2474 2475 /* 2476 * Reserve address space for all of this. 2477 * 2478 * In the case of ET_EXEC, we supply MAP_FIXED so that we get 2479 * exactly the address range that is required. 2480 * 2481 * Otherwise this is ET_DYN, and we are searching for a location 2482 * that can hold the memory space required. If the image is 2483 * pre-linked, LOADDR will be non-zero, and the kernel should 2484 * honor that address if it happens to be free. 2485 * 2486 * In both cases, we will overwrite pages in this range with mappings 2487 * from the executable. 2488 */ 2489 load_addr = target_mmap(loaddr, hiaddr - loaddr, PROT_NONE, 2490 MAP_PRIVATE | MAP_ANON | MAP_NORESERVE | 2491 (ehdr->e_type == ET_EXEC ? MAP_FIXED : 0), 2492 -1, 0); 2493 if (load_addr == -1) { 2494 goto exit_perror; 2495 } 2496 load_bias = load_addr - loaddr; 2497 2498 if (elf_is_fdpic(ehdr)) { 2499 struct elf32_fdpic_loadseg *loadsegs = info->loadsegs = 2500 g_malloc(sizeof(*loadsegs) * info->nsegs); 2501 2502 for (i = 0; i < ehdr->e_phnum; ++i) { 2503 switch (phdr[i].p_type) { 2504 case PT_DYNAMIC: 2505 info->pt_dynamic_addr = phdr[i].p_vaddr + load_bias; 2506 break; 2507 case PT_LOAD: 2508 loadsegs->addr = phdr[i].p_vaddr + load_bias; 2509 loadsegs->p_vaddr = phdr[i].p_vaddr; 2510 loadsegs->p_memsz = phdr[i].p_memsz; 2511 ++loadsegs; 2512 break; 2513 } 2514 } 2515 } 2516 2517 info->load_bias = load_bias; 2518 info->code_offset = load_bias; 2519 info->data_offset = load_bias; 2520 info->load_addr = load_addr; 2521 info->entry = ehdr->e_entry + load_bias; 2522 info->start_code = -1; 2523 info->end_code = 0; 2524 info->start_data = -1; 2525 info->end_data = 0; 2526 info->brk = 0; 2527 info->elf_flags = ehdr->e_flags; 2528 2529 for (i = 0; i < ehdr->e_phnum; i++) { 2530 struct elf_phdr *eppnt = phdr + i; 2531 if (eppnt->p_type == PT_LOAD) { 2532 abi_ulong vaddr, vaddr_po, vaddr_ps, vaddr_ef, vaddr_em, vaddr_len; 2533 int elf_prot = 0; 2534 2535 if (eppnt->p_flags & PF_R) elf_prot = PROT_READ; 2536 if (eppnt->p_flags & PF_W) elf_prot |= PROT_WRITE; 2537 if (eppnt->p_flags & PF_X) elf_prot |= PROT_EXEC; 2538 2539 vaddr = load_bias + eppnt->p_vaddr; 2540 vaddr_po = TARGET_ELF_PAGEOFFSET(vaddr); 2541 vaddr_ps = TARGET_ELF_PAGESTART(vaddr); 2542 vaddr_len = TARGET_ELF_PAGELENGTH(eppnt->p_filesz + vaddr_po); 2543 2544 /* 2545 * Some segments may be completely empty without any backing file 2546 * segment, in that case just let zero_bss allocate an empty buffer 2547 * for it. 2548 */ 2549 if (eppnt->p_filesz != 0) { 2550 error = target_mmap(vaddr_ps, vaddr_len, elf_prot, 2551 MAP_PRIVATE | MAP_FIXED, 2552 image_fd, eppnt->p_offset - vaddr_po); 2553 2554 if (error == -1) { 2555 goto exit_perror; 2556 } 2557 } 2558 2559 vaddr_ef = vaddr + eppnt->p_filesz; 2560 vaddr_em = vaddr + eppnt->p_memsz; 2561 2562 /* If the load segment requests extra zeros (e.g. bss), map it. */ 2563 if (vaddr_ef < vaddr_em) { 2564 zero_bss(vaddr_ef, vaddr_em, elf_prot); 2565 } 2566 2567 /* Find the full program boundaries. */ 2568 if (elf_prot & PROT_EXEC) { 2569 if (vaddr < info->start_code) { 2570 info->start_code = vaddr; 2571 } 2572 if (vaddr_ef > info->end_code) { 2573 info->end_code = vaddr_ef; 2574 } 2575 } 2576 if (elf_prot & PROT_WRITE) { 2577 if (vaddr < info->start_data) { 2578 info->start_data = vaddr; 2579 } 2580 if (vaddr_ef > info->end_data) { 2581 info->end_data = vaddr_ef; 2582 } 2583 } 2584 if (vaddr_em > info->brk) { 2585 info->brk = vaddr_em; 2586 } 2587 } else if (eppnt->p_type == PT_INTERP && pinterp_name) { 2588 char *interp_name; 2589 2590 if (*pinterp_name) { 2591 errmsg = "Multiple PT_INTERP entries"; 2592 goto exit_errmsg; 2593 } 2594 interp_name = malloc(eppnt->p_filesz); 2595 if (!interp_name) { 2596 goto exit_perror; 2597 } 2598 2599 if (eppnt->p_offset + eppnt->p_filesz <= BPRM_BUF_SIZE) { 2600 memcpy(interp_name, bprm_buf + eppnt->p_offset, 2601 eppnt->p_filesz); 2602 } else { 2603 retval = pread(image_fd, interp_name, eppnt->p_filesz, 2604 eppnt->p_offset); 2605 if (retval != eppnt->p_filesz) { 2606 goto exit_perror; 2607 } 2608 } 2609 if (interp_name[eppnt->p_filesz - 1] != 0) { 2610 errmsg = "Invalid PT_INTERP entry"; 2611 goto exit_errmsg; 2612 } 2613 *pinterp_name = interp_name; 2614 #ifdef TARGET_MIPS 2615 } else if (eppnt->p_type == PT_MIPS_ABIFLAGS) { 2616 Mips_elf_abiflags_v0 abiflags; 2617 if (eppnt->p_filesz < sizeof(Mips_elf_abiflags_v0)) { 2618 errmsg = "Invalid PT_MIPS_ABIFLAGS entry"; 2619 goto exit_errmsg; 2620 } 2621 if (eppnt->p_offset + eppnt->p_filesz <= BPRM_BUF_SIZE) { 2622 memcpy(&abiflags, bprm_buf + eppnt->p_offset, 2623 sizeof(Mips_elf_abiflags_v0)); 2624 } else { 2625 retval = pread(image_fd, &abiflags, sizeof(Mips_elf_abiflags_v0), 2626 eppnt->p_offset); 2627 if (retval != sizeof(Mips_elf_abiflags_v0)) { 2628 goto exit_perror; 2629 } 2630 } 2631 bswap_mips_abiflags(&abiflags); 2632 info->fp_abi = abiflags.fp_abi; 2633 #endif 2634 } 2635 } 2636 2637 if (info->end_data == 0) { 2638 info->start_data = info->end_code; 2639 info->end_data = info->end_code; 2640 } 2641 2642 if (qemu_log_enabled()) { 2643 load_symbols(ehdr, image_fd, load_bias); 2644 } 2645 2646 mmap_unlock(); 2647 2648 close(image_fd); 2649 return; 2650 2651 exit_read: 2652 if (retval >= 0) { 2653 errmsg = "Incomplete read of file header"; 2654 goto exit_errmsg; 2655 } 2656 exit_perror: 2657 errmsg = strerror(errno); 2658 exit_errmsg: 2659 fprintf(stderr, "%s: %s\n", image_name, errmsg); 2660 exit(-1); 2661 } 2662 2663 static void load_elf_interp(const char *filename, struct image_info *info, 2664 char bprm_buf[BPRM_BUF_SIZE]) 2665 { 2666 int fd, retval; 2667 2668 fd = open(path(filename), O_RDONLY); 2669 if (fd < 0) { 2670 goto exit_perror; 2671 } 2672 2673 retval = read(fd, bprm_buf, BPRM_BUF_SIZE); 2674 if (retval < 0) { 2675 goto exit_perror; 2676 } 2677 if (retval < BPRM_BUF_SIZE) { 2678 memset(bprm_buf + retval, 0, BPRM_BUF_SIZE - retval); 2679 } 2680 2681 load_elf_image(filename, fd, info, NULL, bprm_buf); 2682 return; 2683 2684 exit_perror: 2685 fprintf(stderr, "%s: %s\n", filename, strerror(errno)); 2686 exit(-1); 2687 } 2688 2689 static int symfind(const void *s0, const void *s1) 2690 { 2691 target_ulong addr = *(target_ulong *)s0; 2692 struct elf_sym *sym = (struct elf_sym *)s1; 2693 int result = 0; 2694 if (addr < sym->st_value) { 2695 result = -1; 2696 } else if (addr >= sym->st_value + sym->st_size) { 2697 result = 1; 2698 } 2699 return result; 2700 } 2701 2702 static const char *lookup_symbolxx(struct syminfo *s, target_ulong orig_addr) 2703 { 2704 #if ELF_CLASS == ELFCLASS32 2705 struct elf_sym *syms = s->disas_symtab.elf32; 2706 #else 2707 struct elf_sym *syms = s->disas_symtab.elf64; 2708 #endif 2709 2710 // binary search 2711 struct elf_sym *sym; 2712 2713 sym = bsearch(&orig_addr, syms, s->disas_num_syms, sizeof(*syms), symfind); 2714 if (sym != NULL) { 2715 return s->disas_strtab + sym->st_name; 2716 } 2717 2718 return ""; 2719 } 2720 2721 /* FIXME: This should use elf_ops.h */ 2722 static int symcmp(const void *s0, const void *s1) 2723 { 2724 struct elf_sym *sym0 = (struct elf_sym *)s0; 2725 struct elf_sym *sym1 = (struct elf_sym *)s1; 2726 return (sym0->st_value < sym1->st_value) 2727 ? -1 2728 : ((sym0->st_value > sym1->st_value) ? 1 : 0); 2729 } 2730 2731 /* Best attempt to load symbols from this ELF object. */ 2732 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias) 2733 { 2734 int i, shnum, nsyms, sym_idx = 0, str_idx = 0; 2735 uint64_t segsz; 2736 struct elf_shdr *shdr; 2737 char *strings = NULL; 2738 struct syminfo *s = NULL; 2739 struct elf_sym *new_syms, *syms = NULL; 2740 2741 shnum = hdr->e_shnum; 2742 i = shnum * sizeof(struct elf_shdr); 2743 shdr = (struct elf_shdr *)alloca(i); 2744 if (pread(fd, shdr, i, hdr->e_shoff) != i) { 2745 return; 2746 } 2747 2748 bswap_shdr(shdr, shnum); 2749 for (i = 0; i < shnum; ++i) { 2750 if (shdr[i].sh_type == SHT_SYMTAB) { 2751 sym_idx = i; 2752 str_idx = shdr[i].sh_link; 2753 goto found; 2754 } 2755 } 2756 2757 /* There will be no symbol table if the file was stripped. */ 2758 return; 2759 2760 found: 2761 /* Now know where the strtab and symtab are. Snarf them. */ 2762 s = g_try_new(struct syminfo, 1); 2763 if (!s) { 2764 goto give_up; 2765 } 2766 2767 segsz = shdr[str_idx].sh_size; 2768 s->disas_strtab = strings = g_try_malloc(segsz); 2769 if (!strings || 2770 pread(fd, strings, segsz, shdr[str_idx].sh_offset) != segsz) { 2771 goto give_up; 2772 } 2773 2774 segsz = shdr[sym_idx].sh_size; 2775 syms = g_try_malloc(segsz); 2776 if (!syms || pread(fd, syms, segsz, shdr[sym_idx].sh_offset) != segsz) { 2777 goto give_up; 2778 } 2779 2780 if (segsz / sizeof(struct elf_sym) > INT_MAX) { 2781 /* Implausibly large symbol table: give up rather than ploughing 2782 * on with the number of symbols calculation overflowing 2783 */ 2784 goto give_up; 2785 } 2786 nsyms = segsz / sizeof(struct elf_sym); 2787 for (i = 0; i < nsyms; ) { 2788 bswap_sym(syms + i); 2789 /* Throw away entries which we do not need. */ 2790 if (syms[i].st_shndx == SHN_UNDEF 2791 || syms[i].st_shndx >= SHN_LORESERVE 2792 || ELF_ST_TYPE(syms[i].st_info) != STT_FUNC) { 2793 if (i < --nsyms) { 2794 syms[i] = syms[nsyms]; 2795 } 2796 } else { 2797 #if defined(TARGET_ARM) || defined (TARGET_MIPS) 2798 /* The bottom address bit marks a Thumb or MIPS16 symbol. */ 2799 syms[i].st_value &= ~(target_ulong)1; 2800 #endif 2801 syms[i].st_value += load_bias; 2802 i++; 2803 } 2804 } 2805 2806 /* No "useful" symbol. */ 2807 if (nsyms == 0) { 2808 goto give_up; 2809 } 2810 2811 /* Attempt to free the storage associated with the local symbols 2812 that we threw away. Whether or not this has any effect on the 2813 memory allocation depends on the malloc implementation and how 2814 many symbols we managed to discard. */ 2815 new_syms = g_try_renew(struct elf_sym, syms, nsyms); 2816 if (new_syms == NULL) { 2817 goto give_up; 2818 } 2819 syms = new_syms; 2820 2821 qsort(syms, nsyms, sizeof(*syms), symcmp); 2822 2823 s->disas_num_syms = nsyms; 2824 #if ELF_CLASS == ELFCLASS32 2825 s->disas_symtab.elf32 = syms; 2826 #else 2827 s->disas_symtab.elf64 = syms; 2828 #endif 2829 s->lookup_symbol = lookup_symbolxx; 2830 s->next = syminfos; 2831 syminfos = s; 2832 2833 return; 2834 2835 give_up: 2836 g_free(s); 2837 g_free(strings); 2838 g_free(syms); 2839 } 2840 2841 uint32_t get_elf_eflags(int fd) 2842 { 2843 struct elfhdr ehdr; 2844 off_t offset; 2845 int ret; 2846 2847 /* Read ELF header */ 2848 offset = lseek(fd, 0, SEEK_SET); 2849 if (offset == (off_t) -1) { 2850 return 0; 2851 } 2852 ret = read(fd, &ehdr, sizeof(ehdr)); 2853 if (ret < sizeof(ehdr)) { 2854 return 0; 2855 } 2856 offset = lseek(fd, offset, SEEK_SET); 2857 if (offset == (off_t) -1) { 2858 return 0; 2859 } 2860 2861 /* Check ELF signature */ 2862 if (!elf_check_ident(&ehdr)) { 2863 return 0; 2864 } 2865 2866 /* check header */ 2867 bswap_ehdr(&ehdr); 2868 if (!elf_check_ehdr(&ehdr)) { 2869 return 0; 2870 } 2871 2872 /* return architecture id */ 2873 return ehdr.e_flags; 2874 } 2875 2876 int load_elf_binary(struct linux_binprm *bprm, struct image_info *info) 2877 { 2878 struct image_info interp_info; 2879 struct elfhdr elf_ex; 2880 char *elf_interpreter = NULL; 2881 char *scratch; 2882 2883 memset(&interp_info, 0, sizeof(interp_info)); 2884 #ifdef TARGET_MIPS 2885 interp_info.fp_abi = MIPS_ABI_FP_UNKNOWN; 2886 #endif 2887 2888 info->start_mmap = (abi_ulong)ELF_START_MMAP; 2889 2890 load_elf_image(bprm->filename, bprm->fd, info, 2891 &elf_interpreter, bprm->buf); 2892 2893 /* ??? We need a copy of the elf header for passing to create_elf_tables. 2894 If we do nothing, we'll have overwritten this when we re-use bprm->buf 2895 when we load the interpreter. */ 2896 elf_ex = *(struct elfhdr *)bprm->buf; 2897 2898 /* Do this so that we can load the interpreter, if need be. We will 2899 change some of these later */ 2900 bprm->p = setup_arg_pages(bprm, info); 2901 2902 scratch = g_new0(char, TARGET_PAGE_SIZE); 2903 if (STACK_GROWS_DOWN) { 2904 bprm->p = copy_elf_strings(1, &bprm->filename, scratch, 2905 bprm->p, info->stack_limit); 2906 info->file_string = bprm->p; 2907 bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch, 2908 bprm->p, info->stack_limit); 2909 info->env_strings = bprm->p; 2910 bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch, 2911 bprm->p, info->stack_limit); 2912 info->arg_strings = bprm->p; 2913 } else { 2914 info->arg_strings = bprm->p; 2915 bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch, 2916 bprm->p, info->stack_limit); 2917 info->env_strings = bprm->p; 2918 bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch, 2919 bprm->p, info->stack_limit); 2920 info->file_string = bprm->p; 2921 bprm->p = copy_elf_strings(1, &bprm->filename, scratch, 2922 bprm->p, info->stack_limit); 2923 } 2924 2925 g_free(scratch); 2926 2927 if (!bprm->p) { 2928 fprintf(stderr, "%s: %s\n", bprm->filename, strerror(E2BIG)); 2929 exit(-1); 2930 } 2931 2932 if (elf_interpreter) { 2933 load_elf_interp(elf_interpreter, &interp_info, bprm->buf); 2934 2935 /* If the program interpreter is one of these two, then assume 2936 an iBCS2 image. Otherwise assume a native linux image. */ 2937 2938 if (strcmp(elf_interpreter, "/usr/lib/libc.so.1") == 0 2939 || strcmp(elf_interpreter, "/usr/lib/ld.so.1") == 0) { 2940 info->personality = PER_SVR4; 2941 2942 /* Why this, you ask??? Well SVr4 maps page 0 as read-only, 2943 and some applications "depend" upon this behavior. Since 2944 we do not have the power to recompile these, we emulate 2945 the SVr4 behavior. Sigh. */ 2946 target_mmap(0, qemu_host_page_size, PROT_READ | PROT_EXEC, 2947 MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); 2948 } 2949 #ifdef TARGET_MIPS 2950 info->interp_fp_abi = interp_info.fp_abi; 2951 #endif 2952 } 2953 2954 bprm->p = create_elf_tables(bprm->p, bprm->argc, bprm->envc, &elf_ex, 2955 info, (elf_interpreter ? &interp_info : NULL)); 2956 info->start_stack = bprm->p; 2957 2958 /* If we have an interpreter, set that as the program's entry point. 2959 Copy the load_bias as well, to help PPC64 interpret the entry 2960 point as a function descriptor. Do this after creating elf tables 2961 so that we copy the original program entry point into the AUXV. */ 2962 if (elf_interpreter) { 2963 info->load_bias = interp_info.load_bias; 2964 info->entry = interp_info.entry; 2965 free(elf_interpreter); 2966 } 2967 2968 #ifdef USE_ELF_CORE_DUMP 2969 bprm->core_dump = &elf_core_dump; 2970 #endif 2971 2972 /* 2973 * If we reserved extra space for brk, release it now. 2974 * The implementation of do_brk in syscalls.c expects to be able 2975 * to mmap pages in this space. 2976 */ 2977 if (info->reserve_brk) { 2978 abi_ulong start_brk = HOST_PAGE_ALIGN(info->brk); 2979 abi_ulong end_brk = HOST_PAGE_ALIGN(info->brk + info->reserve_brk); 2980 target_munmap(start_brk, end_brk - start_brk); 2981 } 2982 2983 return 0; 2984 } 2985 2986 #ifdef USE_ELF_CORE_DUMP 2987 /* 2988 * Definitions to generate Intel SVR4-like core files. 2989 * These mostly have the same names as the SVR4 types with "target_elf_" 2990 * tacked on the front to prevent clashes with linux definitions, 2991 * and the typedef forms have been avoided. This is mostly like 2992 * the SVR4 structure, but more Linuxy, with things that Linux does 2993 * not support and which gdb doesn't really use excluded. 2994 * 2995 * Fields we don't dump (their contents is zero) in linux-user qemu 2996 * are marked with XXX. 2997 * 2998 * Core dump code is copied from linux kernel (fs/binfmt_elf.c). 2999 * 3000 * Porting ELF coredump for target is (quite) simple process. First you 3001 * define USE_ELF_CORE_DUMP in target ELF code (where init_thread() for 3002 * the target resides): 3003 * 3004 * #define USE_ELF_CORE_DUMP 3005 * 3006 * Next you define type of register set used for dumping. ELF specification 3007 * says that it needs to be array of elf_greg_t that has size of ELF_NREG. 3008 * 3009 * typedef <target_regtype> target_elf_greg_t; 3010 * #define ELF_NREG <number of registers> 3011 * typedef taret_elf_greg_t target_elf_gregset_t[ELF_NREG]; 3012 * 3013 * Last step is to implement target specific function that copies registers 3014 * from given cpu into just specified register set. Prototype is: 3015 * 3016 * static void elf_core_copy_regs(taret_elf_gregset_t *regs, 3017 * const CPUArchState *env); 3018 * 3019 * Parameters: 3020 * regs - copy register values into here (allocated and zeroed by caller) 3021 * env - copy registers from here 3022 * 3023 * Example for ARM target is provided in this file. 3024 */ 3025 3026 /* An ELF note in memory */ 3027 struct memelfnote { 3028 const char *name; 3029 size_t namesz; 3030 size_t namesz_rounded; 3031 int type; 3032 size_t datasz; 3033 size_t datasz_rounded; 3034 void *data; 3035 size_t notesz; 3036 }; 3037 3038 struct target_elf_siginfo { 3039 abi_int si_signo; /* signal number */ 3040 abi_int si_code; /* extra code */ 3041 abi_int si_errno; /* errno */ 3042 }; 3043 3044 struct target_elf_prstatus { 3045 struct target_elf_siginfo pr_info; /* Info associated with signal */ 3046 abi_short pr_cursig; /* Current signal */ 3047 abi_ulong pr_sigpend; /* XXX */ 3048 abi_ulong pr_sighold; /* XXX */ 3049 target_pid_t pr_pid; 3050 target_pid_t pr_ppid; 3051 target_pid_t pr_pgrp; 3052 target_pid_t pr_sid; 3053 struct target_timeval pr_utime; /* XXX User time */ 3054 struct target_timeval pr_stime; /* XXX System time */ 3055 struct target_timeval pr_cutime; /* XXX Cumulative user time */ 3056 struct target_timeval pr_cstime; /* XXX Cumulative system time */ 3057 target_elf_gregset_t pr_reg; /* GP registers */ 3058 abi_int pr_fpvalid; /* XXX */ 3059 }; 3060 3061 #define ELF_PRARGSZ (80) /* Number of chars for args */ 3062 3063 struct target_elf_prpsinfo { 3064 char pr_state; /* numeric process state */ 3065 char pr_sname; /* char for pr_state */ 3066 char pr_zomb; /* zombie */ 3067 char pr_nice; /* nice val */ 3068 abi_ulong pr_flag; /* flags */ 3069 target_uid_t pr_uid; 3070 target_gid_t pr_gid; 3071 target_pid_t pr_pid, pr_ppid, pr_pgrp, pr_sid; 3072 /* Lots missing */ 3073 char pr_fname[16] QEMU_NONSTRING; /* filename of executable */ 3074 char pr_psargs[ELF_PRARGSZ]; /* initial part of arg list */ 3075 }; 3076 3077 /* Here is the structure in which status of each thread is captured. */ 3078 struct elf_thread_status { 3079 QTAILQ_ENTRY(elf_thread_status) ets_link; 3080 struct target_elf_prstatus prstatus; /* NT_PRSTATUS */ 3081 #if 0 3082 elf_fpregset_t fpu; /* NT_PRFPREG */ 3083 struct task_struct *thread; 3084 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */ 3085 #endif 3086 struct memelfnote notes[1]; 3087 int num_notes; 3088 }; 3089 3090 struct elf_note_info { 3091 struct memelfnote *notes; 3092 struct target_elf_prstatus *prstatus; /* NT_PRSTATUS */ 3093 struct target_elf_prpsinfo *psinfo; /* NT_PRPSINFO */ 3094 3095 QTAILQ_HEAD(, elf_thread_status) thread_list; 3096 #if 0 3097 /* 3098 * Current version of ELF coredump doesn't support 3099 * dumping fp regs etc. 3100 */ 3101 elf_fpregset_t *fpu; 3102 elf_fpxregset_t *xfpu; 3103 int thread_status_size; 3104 #endif 3105 int notes_size; 3106 int numnote; 3107 }; 3108 3109 struct vm_area_struct { 3110 target_ulong vma_start; /* start vaddr of memory region */ 3111 target_ulong vma_end; /* end vaddr of memory region */ 3112 abi_ulong vma_flags; /* protection etc. flags for the region */ 3113 QTAILQ_ENTRY(vm_area_struct) vma_link; 3114 }; 3115 3116 struct mm_struct { 3117 QTAILQ_HEAD(, vm_area_struct) mm_mmap; 3118 int mm_count; /* number of mappings */ 3119 }; 3120 3121 static struct mm_struct *vma_init(void); 3122 static void vma_delete(struct mm_struct *); 3123 static int vma_add_mapping(struct mm_struct *, target_ulong, 3124 target_ulong, abi_ulong); 3125 static int vma_get_mapping_count(const struct mm_struct *); 3126 static struct vm_area_struct *vma_first(const struct mm_struct *); 3127 static struct vm_area_struct *vma_next(struct vm_area_struct *); 3128 static abi_ulong vma_dump_size(const struct vm_area_struct *); 3129 static int vma_walker(void *priv, target_ulong start, target_ulong end, 3130 unsigned long flags); 3131 3132 static void fill_elf_header(struct elfhdr *, int, uint16_t, uint32_t); 3133 static void fill_note(struct memelfnote *, const char *, int, 3134 unsigned int, void *); 3135 static void fill_prstatus(struct target_elf_prstatus *, const TaskState *, int); 3136 static int fill_psinfo(struct target_elf_prpsinfo *, const TaskState *); 3137 static void fill_auxv_note(struct memelfnote *, const TaskState *); 3138 static void fill_elf_note_phdr(struct elf_phdr *, int, off_t); 3139 static size_t note_size(const struct memelfnote *); 3140 static void free_note_info(struct elf_note_info *); 3141 static int fill_note_info(struct elf_note_info *, long, const CPUArchState *); 3142 static void fill_thread_info(struct elf_note_info *, const CPUArchState *); 3143 static int core_dump_filename(const TaskState *, char *, size_t); 3144 3145 static int dump_write(int, const void *, size_t); 3146 static int write_note(struct memelfnote *, int); 3147 static int write_note_info(struct elf_note_info *, int); 3148 3149 #ifdef BSWAP_NEEDED 3150 static void bswap_prstatus(struct target_elf_prstatus *prstatus) 3151 { 3152 prstatus->pr_info.si_signo = tswap32(prstatus->pr_info.si_signo); 3153 prstatus->pr_info.si_code = tswap32(prstatus->pr_info.si_code); 3154 prstatus->pr_info.si_errno = tswap32(prstatus->pr_info.si_errno); 3155 prstatus->pr_cursig = tswap16(prstatus->pr_cursig); 3156 prstatus->pr_sigpend = tswapal(prstatus->pr_sigpend); 3157 prstatus->pr_sighold = tswapal(prstatus->pr_sighold); 3158 prstatus->pr_pid = tswap32(prstatus->pr_pid); 3159 prstatus->pr_ppid = tswap32(prstatus->pr_ppid); 3160 prstatus->pr_pgrp = tswap32(prstatus->pr_pgrp); 3161 prstatus->pr_sid = tswap32(prstatus->pr_sid); 3162 /* cpu times are not filled, so we skip them */ 3163 /* regs should be in correct format already */ 3164 prstatus->pr_fpvalid = tswap32(prstatus->pr_fpvalid); 3165 } 3166 3167 static void bswap_psinfo(struct target_elf_prpsinfo *psinfo) 3168 { 3169 psinfo->pr_flag = tswapal(psinfo->pr_flag); 3170 psinfo->pr_uid = tswap16(psinfo->pr_uid); 3171 psinfo->pr_gid = tswap16(psinfo->pr_gid); 3172 psinfo->pr_pid = tswap32(psinfo->pr_pid); 3173 psinfo->pr_ppid = tswap32(psinfo->pr_ppid); 3174 psinfo->pr_pgrp = tswap32(psinfo->pr_pgrp); 3175 psinfo->pr_sid = tswap32(psinfo->pr_sid); 3176 } 3177 3178 static void bswap_note(struct elf_note *en) 3179 { 3180 bswap32s(&en->n_namesz); 3181 bswap32s(&en->n_descsz); 3182 bswap32s(&en->n_type); 3183 } 3184 #else 3185 static inline void bswap_prstatus(struct target_elf_prstatus *p) { } 3186 static inline void bswap_psinfo(struct target_elf_prpsinfo *p) {} 3187 static inline void bswap_note(struct elf_note *en) { } 3188 #endif /* BSWAP_NEEDED */ 3189 3190 /* 3191 * Minimal support for linux memory regions. These are needed 3192 * when we are finding out what memory exactly belongs to 3193 * emulated process. No locks needed here, as long as 3194 * thread that received the signal is stopped. 3195 */ 3196 3197 static struct mm_struct *vma_init(void) 3198 { 3199 struct mm_struct *mm; 3200 3201 if ((mm = g_malloc(sizeof (*mm))) == NULL) 3202 return (NULL); 3203 3204 mm->mm_count = 0; 3205 QTAILQ_INIT(&mm->mm_mmap); 3206 3207 return (mm); 3208 } 3209 3210 static void vma_delete(struct mm_struct *mm) 3211 { 3212 struct vm_area_struct *vma; 3213 3214 while ((vma = vma_first(mm)) != NULL) { 3215 QTAILQ_REMOVE(&mm->mm_mmap, vma, vma_link); 3216 g_free(vma); 3217 } 3218 g_free(mm); 3219 } 3220 3221 static int vma_add_mapping(struct mm_struct *mm, target_ulong start, 3222 target_ulong end, abi_ulong flags) 3223 { 3224 struct vm_area_struct *vma; 3225 3226 if ((vma = g_malloc0(sizeof (*vma))) == NULL) 3227 return (-1); 3228 3229 vma->vma_start = start; 3230 vma->vma_end = end; 3231 vma->vma_flags = flags; 3232 3233 QTAILQ_INSERT_TAIL(&mm->mm_mmap, vma, vma_link); 3234 mm->mm_count++; 3235 3236 return (0); 3237 } 3238 3239 static struct vm_area_struct *vma_first(const struct mm_struct *mm) 3240 { 3241 return (QTAILQ_FIRST(&mm->mm_mmap)); 3242 } 3243 3244 static struct vm_area_struct *vma_next(struct vm_area_struct *vma) 3245 { 3246 return (QTAILQ_NEXT(vma, vma_link)); 3247 } 3248 3249 static int vma_get_mapping_count(const struct mm_struct *mm) 3250 { 3251 return (mm->mm_count); 3252 } 3253 3254 /* 3255 * Calculate file (dump) size of given memory region. 3256 */ 3257 static abi_ulong vma_dump_size(const struct vm_area_struct *vma) 3258 { 3259 /* if we cannot even read the first page, skip it */ 3260 if (!access_ok(VERIFY_READ, vma->vma_start, TARGET_PAGE_SIZE)) 3261 return (0); 3262 3263 /* 3264 * Usually we don't dump executable pages as they contain 3265 * non-writable code that debugger can read directly from 3266 * target library etc. However, thread stacks are marked 3267 * also executable so we read in first page of given region 3268 * and check whether it contains elf header. If there is 3269 * no elf header, we dump it. 3270 */ 3271 if (vma->vma_flags & PROT_EXEC) { 3272 char page[TARGET_PAGE_SIZE]; 3273 3274 copy_from_user(page, vma->vma_start, sizeof (page)); 3275 if ((page[EI_MAG0] == ELFMAG0) && 3276 (page[EI_MAG1] == ELFMAG1) && 3277 (page[EI_MAG2] == ELFMAG2) && 3278 (page[EI_MAG3] == ELFMAG3)) { 3279 /* 3280 * Mappings are possibly from ELF binary. Don't dump 3281 * them. 3282 */ 3283 return (0); 3284 } 3285 } 3286 3287 return (vma->vma_end - vma->vma_start); 3288 } 3289 3290 static int vma_walker(void *priv, target_ulong start, target_ulong end, 3291 unsigned long flags) 3292 { 3293 struct mm_struct *mm = (struct mm_struct *)priv; 3294 3295 vma_add_mapping(mm, start, end, flags); 3296 return (0); 3297 } 3298 3299 static void fill_note(struct memelfnote *note, const char *name, int type, 3300 unsigned int sz, void *data) 3301 { 3302 unsigned int namesz; 3303 3304 namesz = strlen(name) + 1; 3305 note->name = name; 3306 note->namesz = namesz; 3307 note->namesz_rounded = roundup(namesz, sizeof (int32_t)); 3308 note->type = type; 3309 note->datasz = sz; 3310 note->datasz_rounded = roundup(sz, sizeof (int32_t)); 3311 3312 note->data = data; 3313 3314 /* 3315 * We calculate rounded up note size here as specified by 3316 * ELF document. 3317 */ 3318 note->notesz = sizeof (struct elf_note) + 3319 note->namesz_rounded + note->datasz_rounded; 3320 } 3321 3322 static void fill_elf_header(struct elfhdr *elf, int segs, uint16_t machine, 3323 uint32_t flags) 3324 { 3325 (void) memset(elf, 0, sizeof(*elf)); 3326 3327 (void) memcpy(elf->e_ident, ELFMAG, SELFMAG); 3328 elf->e_ident[EI_CLASS] = ELF_CLASS; 3329 elf->e_ident[EI_DATA] = ELF_DATA; 3330 elf->e_ident[EI_VERSION] = EV_CURRENT; 3331 elf->e_ident[EI_OSABI] = ELF_OSABI; 3332 3333 elf->e_type = ET_CORE; 3334 elf->e_machine = machine; 3335 elf->e_version = EV_CURRENT; 3336 elf->e_phoff = sizeof(struct elfhdr); 3337 elf->e_flags = flags; 3338 elf->e_ehsize = sizeof(struct elfhdr); 3339 elf->e_phentsize = sizeof(struct elf_phdr); 3340 elf->e_phnum = segs; 3341 3342 bswap_ehdr(elf); 3343 } 3344 3345 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, off_t offset) 3346 { 3347 phdr->p_type = PT_NOTE; 3348 phdr->p_offset = offset; 3349 phdr->p_vaddr = 0; 3350 phdr->p_paddr = 0; 3351 phdr->p_filesz = sz; 3352 phdr->p_memsz = 0; 3353 phdr->p_flags = 0; 3354 phdr->p_align = 0; 3355 3356 bswap_phdr(phdr, 1); 3357 } 3358 3359 static size_t note_size(const struct memelfnote *note) 3360 { 3361 return (note->notesz); 3362 } 3363 3364 static void fill_prstatus(struct target_elf_prstatus *prstatus, 3365 const TaskState *ts, int signr) 3366 { 3367 (void) memset(prstatus, 0, sizeof (*prstatus)); 3368 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr; 3369 prstatus->pr_pid = ts->ts_tid; 3370 prstatus->pr_ppid = getppid(); 3371 prstatus->pr_pgrp = getpgrp(); 3372 prstatus->pr_sid = getsid(0); 3373 3374 bswap_prstatus(prstatus); 3375 } 3376 3377 static int fill_psinfo(struct target_elf_prpsinfo *psinfo, const TaskState *ts) 3378 { 3379 char *base_filename; 3380 unsigned int i, len; 3381 3382 (void) memset(psinfo, 0, sizeof (*psinfo)); 3383 3384 len = ts->info->arg_end - ts->info->arg_start; 3385 if (len >= ELF_PRARGSZ) 3386 len = ELF_PRARGSZ - 1; 3387 if (copy_from_user(&psinfo->pr_psargs, ts->info->arg_start, len)) 3388 return -EFAULT; 3389 for (i = 0; i < len; i++) 3390 if (psinfo->pr_psargs[i] == 0) 3391 psinfo->pr_psargs[i] = ' '; 3392 psinfo->pr_psargs[len] = 0; 3393 3394 psinfo->pr_pid = getpid(); 3395 psinfo->pr_ppid = getppid(); 3396 psinfo->pr_pgrp = getpgrp(); 3397 psinfo->pr_sid = getsid(0); 3398 psinfo->pr_uid = getuid(); 3399 psinfo->pr_gid = getgid(); 3400 3401 base_filename = g_path_get_basename(ts->bprm->filename); 3402 /* 3403 * Using strncpy here is fine: at max-length, 3404 * this field is not NUL-terminated. 3405 */ 3406 (void) strncpy(psinfo->pr_fname, base_filename, 3407 sizeof(psinfo->pr_fname)); 3408 3409 g_free(base_filename); 3410 bswap_psinfo(psinfo); 3411 return (0); 3412 } 3413 3414 static void fill_auxv_note(struct memelfnote *note, const TaskState *ts) 3415 { 3416 elf_addr_t auxv = (elf_addr_t)ts->info->saved_auxv; 3417 elf_addr_t orig_auxv = auxv; 3418 void *ptr; 3419 int len = ts->info->auxv_len; 3420 3421 /* 3422 * Auxiliary vector is stored in target process stack. It contains 3423 * {type, value} pairs that we need to dump into note. This is not 3424 * strictly necessary but we do it here for sake of completeness. 3425 */ 3426 3427 /* read in whole auxv vector and copy it to memelfnote */ 3428 ptr = lock_user(VERIFY_READ, orig_auxv, len, 0); 3429 if (ptr != NULL) { 3430 fill_note(note, "CORE", NT_AUXV, len, ptr); 3431 unlock_user(ptr, auxv, len); 3432 } 3433 } 3434 3435 /* 3436 * Constructs name of coredump file. We have following convention 3437 * for the name: 3438 * qemu_<basename-of-target-binary>_<date>-<time>_<pid>.core 3439 * 3440 * Returns 0 in case of success, -1 otherwise (errno is set). 3441 */ 3442 static int core_dump_filename(const TaskState *ts, char *buf, 3443 size_t bufsize) 3444 { 3445 char timestamp[64]; 3446 char *base_filename = NULL; 3447 struct timeval tv; 3448 struct tm tm; 3449 3450 assert(bufsize >= PATH_MAX); 3451 3452 if (gettimeofday(&tv, NULL) < 0) { 3453 (void) fprintf(stderr, "unable to get current timestamp: %s", 3454 strerror(errno)); 3455 return (-1); 3456 } 3457 3458 base_filename = g_path_get_basename(ts->bprm->filename); 3459 (void) strftime(timestamp, sizeof (timestamp), "%Y%m%d-%H%M%S", 3460 localtime_r(&tv.tv_sec, &tm)); 3461 (void) snprintf(buf, bufsize, "qemu_%s_%s_%d.core", 3462 base_filename, timestamp, (int)getpid()); 3463 g_free(base_filename); 3464 3465 return (0); 3466 } 3467 3468 static int dump_write(int fd, const void *ptr, size_t size) 3469 { 3470 const char *bufp = (const char *)ptr; 3471 ssize_t bytes_written, bytes_left; 3472 struct rlimit dumpsize; 3473 off_t pos; 3474 3475 bytes_written = 0; 3476 getrlimit(RLIMIT_CORE, &dumpsize); 3477 if ((pos = lseek(fd, 0, SEEK_CUR))==-1) { 3478 if (errno == ESPIPE) { /* not a seekable stream */ 3479 bytes_left = size; 3480 } else { 3481 return pos; 3482 } 3483 } else { 3484 if (dumpsize.rlim_cur <= pos) { 3485 return -1; 3486 } else if (dumpsize.rlim_cur == RLIM_INFINITY) { 3487 bytes_left = size; 3488 } else { 3489 size_t limit_left=dumpsize.rlim_cur - pos; 3490 bytes_left = limit_left >= size ? size : limit_left ; 3491 } 3492 } 3493 3494 /* 3495 * In normal conditions, single write(2) should do but 3496 * in case of socket etc. this mechanism is more portable. 3497 */ 3498 do { 3499 bytes_written = write(fd, bufp, bytes_left); 3500 if (bytes_written < 0) { 3501 if (errno == EINTR) 3502 continue; 3503 return (-1); 3504 } else if (bytes_written == 0) { /* eof */ 3505 return (-1); 3506 } 3507 bufp += bytes_written; 3508 bytes_left -= bytes_written; 3509 } while (bytes_left > 0); 3510 3511 return (0); 3512 } 3513 3514 static int write_note(struct memelfnote *men, int fd) 3515 { 3516 struct elf_note en; 3517 3518 en.n_namesz = men->namesz; 3519 en.n_type = men->type; 3520 en.n_descsz = men->datasz; 3521 3522 bswap_note(&en); 3523 3524 if (dump_write(fd, &en, sizeof(en)) != 0) 3525 return (-1); 3526 if (dump_write(fd, men->name, men->namesz_rounded) != 0) 3527 return (-1); 3528 if (dump_write(fd, men->data, men->datasz_rounded) != 0) 3529 return (-1); 3530 3531 return (0); 3532 } 3533 3534 static void fill_thread_info(struct elf_note_info *info, const CPUArchState *env) 3535 { 3536 CPUState *cpu = env_cpu((CPUArchState *)env); 3537 TaskState *ts = (TaskState *)cpu->opaque; 3538 struct elf_thread_status *ets; 3539 3540 ets = g_malloc0(sizeof (*ets)); 3541 ets->num_notes = 1; /* only prstatus is dumped */ 3542 fill_prstatus(&ets->prstatus, ts, 0); 3543 elf_core_copy_regs(&ets->prstatus.pr_reg, env); 3544 fill_note(&ets->notes[0], "CORE", NT_PRSTATUS, sizeof (ets->prstatus), 3545 &ets->prstatus); 3546 3547 QTAILQ_INSERT_TAIL(&info->thread_list, ets, ets_link); 3548 3549 info->notes_size += note_size(&ets->notes[0]); 3550 } 3551 3552 static void init_note_info(struct elf_note_info *info) 3553 { 3554 /* Initialize the elf_note_info structure so that it is at 3555 * least safe to call free_note_info() on it. Must be 3556 * called before calling fill_note_info(). 3557 */ 3558 memset(info, 0, sizeof (*info)); 3559 QTAILQ_INIT(&info->thread_list); 3560 } 3561 3562 static int fill_note_info(struct elf_note_info *info, 3563 long signr, const CPUArchState *env) 3564 { 3565 #define NUMNOTES 3 3566 CPUState *cpu = env_cpu((CPUArchState *)env); 3567 TaskState *ts = (TaskState *)cpu->opaque; 3568 int i; 3569 3570 info->notes = g_new0(struct memelfnote, NUMNOTES); 3571 if (info->notes == NULL) 3572 return (-ENOMEM); 3573 info->prstatus = g_malloc0(sizeof (*info->prstatus)); 3574 if (info->prstatus == NULL) 3575 return (-ENOMEM); 3576 info->psinfo = g_malloc0(sizeof (*info->psinfo)); 3577 if (info->prstatus == NULL) 3578 return (-ENOMEM); 3579 3580 /* 3581 * First fill in status (and registers) of current thread 3582 * including process info & aux vector. 3583 */ 3584 fill_prstatus(info->prstatus, ts, signr); 3585 elf_core_copy_regs(&info->prstatus->pr_reg, env); 3586 fill_note(&info->notes[0], "CORE", NT_PRSTATUS, 3587 sizeof (*info->prstatus), info->prstatus); 3588 fill_psinfo(info->psinfo, ts); 3589 fill_note(&info->notes[1], "CORE", NT_PRPSINFO, 3590 sizeof (*info->psinfo), info->psinfo); 3591 fill_auxv_note(&info->notes[2], ts); 3592 info->numnote = 3; 3593 3594 info->notes_size = 0; 3595 for (i = 0; i < info->numnote; i++) 3596 info->notes_size += note_size(&info->notes[i]); 3597 3598 /* read and fill status of all threads */ 3599 cpu_list_lock(); 3600 CPU_FOREACH(cpu) { 3601 if (cpu == thread_cpu) { 3602 continue; 3603 } 3604 fill_thread_info(info, (CPUArchState *)cpu->env_ptr); 3605 } 3606 cpu_list_unlock(); 3607 3608 return (0); 3609 } 3610 3611 static void free_note_info(struct elf_note_info *info) 3612 { 3613 struct elf_thread_status *ets; 3614 3615 while (!QTAILQ_EMPTY(&info->thread_list)) { 3616 ets = QTAILQ_FIRST(&info->thread_list); 3617 QTAILQ_REMOVE(&info->thread_list, ets, ets_link); 3618 g_free(ets); 3619 } 3620 3621 g_free(info->prstatus); 3622 g_free(info->psinfo); 3623 g_free(info->notes); 3624 } 3625 3626 static int write_note_info(struct elf_note_info *info, int fd) 3627 { 3628 struct elf_thread_status *ets; 3629 int i, error = 0; 3630 3631 /* write prstatus, psinfo and auxv for current thread */ 3632 for (i = 0; i < info->numnote; i++) 3633 if ((error = write_note(&info->notes[i], fd)) != 0) 3634 return (error); 3635 3636 /* write prstatus for each thread */ 3637 QTAILQ_FOREACH(ets, &info->thread_list, ets_link) { 3638 if ((error = write_note(&ets->notes[0], fd)) != 0) 3639 return (error); 3640 } 3641 3642 return (0); 3643 } 3644 3645 /* 3646 * Write out ELF coredump. 3647 * 3648 * See documentation of ELF object file format in: 3649 * http://www.caldera.com/developers/devspecs/gabi41.pdf 3650 * 3651 * Coredump format in linux is following: 3652 * 3653 * 0 +----------------------+ \ 3654 * | ELF header | ET_CORE | 3655 * +----------------------+ | 3656 * | ELF program headers | |--- headers 3657 * | - NOTE section | | 3658 * | - PT_LOAD sections | | 3659 * +----------------------+ / 3660 * | NOTEs: | 3661 * | - NT_PRSTATUS | 3662 * | - NT_PRSINFO | 3663 * | - NT_AUXV | 3664 * +----------------------+ <-- aligned to target page 3665 * | Process memory dump | 3666 * : : 3667 * . . 3668 * : : 3669 * | | 3670 * +----------------------+ 3671 * 3672 * NT_PRSTATUS -> struct elf_prstatus (per thread) 3673 * NT_PRSINFO -> struct elf_prpsinfo 3674 * NT_AUXV is array of { type, value } pairs (see fill_auxv_note()). 3675 * 3676 * Format follows System V format as close as possible. Current 3677 * version limitations are as follows: 3678 * - no floating point registers are dumped 3679 * 3680 * Function returns 0 in case of success, negative errno otherwise. 3681 * 3682 * TODO: make this work also during runtime: it should be 3683 * possible to force coredump from running process and then 3684 * continue processing. For example qemu could set up SIGUSR2 3685 * handler (provided that target process haven't registered 3686 * handler for that) that does the dump when signal is received. 3687 */ 3688 static int elf_core_dump(int signr, const CPUArchState *env) 3689 { 3690 const CPUState *cpu = env_cpu((CPUArchState *)env); 3691 const TaskState *ts = (const TaskState *)cpu->opaque; 3692 struct vm_area_struct *vma = NULL; 3693 char corefile[PATH_MAX]; 3694 struct elf_note_info info; 3695 struct elfhdr elf; 3696 struct elf_phdr phdr; 3697 struct rlimit dumpsize; 3698 struct mm_struct *mm = NULL; 3699 off_t offset = 0, data_offset = 0; 3700 int segs = 0; 3701 int fd = -1; 3702 3703 init_note_info(&info); 3704 3705 errno = 0; 3706 getrlimit(RLIMIT_CORE, &dumpsize); 3707 if (dumpsize.rlim_cur == 0) 3708 return 0; 3709 3710 if (core_dump_filename(ts, corefile, sizeof (corefile)) < 0) 3711 return (-errno); 3712 3713 if ((fd = open(corefile, O_WRONLY | O_CREAT, 3714 S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH)) < 0) 3715 return (-errno); 3716 3717 /* 3718 * Walk through target process memory mappings and 3719 * set up structure containing this information. After 3720 * this point vma_xxx functions can be used. 3721 */ 3722 if ((mm = vma_init()) == NULL) 3723 goto out; 3724 3725 walk_memory_regions(mm, vma_walker); 3726 segs = vma_get_mapping_count(mm); 3727 3728 /* 3729 * Construct valid coredump ELF header. We also 3730 * add one more segment for notes. 3731 */ 3732 fill_elf_header(&elf, segs + 1, ELF_MACHINE, 0); 3733 if (dump_write(fd, &elf, sizeof (elf)) != 0) 3734 goto out; 3735 3736 /* fill in the in-memory version of notes */ 3737 if (fill_note_info(&info, signr, env) < 0) 3738 goto out; 3739 3740 offset += sizeof (elf); /* elf header */ 3741 offset += (segs + 1) * sizeof (struct elf_phdr); /* program headers */ 3742 3743 /* write out notes program header */ 3744 fill_elf_note_phdr(&phdr, info.notes_size, offset); 3745 3746 offset += info.notes_size; 3747 if (dump_write(fd, &phdr, sizeof (phdr)) != 0) 3748 goto out; 3749 3750 /* 3751 * ELF specification wants data to start at page boundary so 3752 * we align it here. 3753 */ 3754 data_offset = offset = roundup(offset, ELF_EXEC_PAGESIZE); 3755 3756 /* 3757 * Write program headers for memory regions mapped in 3758 * the target process. 3759 */ 3760 for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) { 3761 (void) memset(&phdr, 0, sizeof (phdr)); 3762 3763 phdr.p_type = PT_LOAD; 3764 phdr.p_offset = offset; 3765 phdr.p_vaddr = vma->vma_start; 3766 phdr.p_paddr = 0; 3767 phdr.p_filesz = vma_dump_size(vma); 3768 offset += phdr.p_filesz; 3769 phdr.p_memsz = vma->vma_end - vma->vma_start; 3770 phdr.p_flags = vma->vma_flags & PROT_READ ? PF_R : 0; 3771 if (vma->vma_flags & PROT_WRITE) 3772 phdr.p_flags |= PF_W; 3773 if (vma->vma_flags & PROT_EXEC) 3774 phdr.p_flags |= PF_X; 3775 phdr.p_align = ELF_EXEC_PAGESIZE; 3776 3777 bswap_phdr(&phdr, 1); 3778 if (dump_write(fd, &phdr, sizeof(phdr)) != 0) { 3779 goto out; 3780 } 3781 } 3782 3783 /* 3784 * Next we write notes just after program headers. No 3785 * alignment needed here. 3786 */ 3787 if (write_note_info(&info, fd) < 0) 3788 goto out; 3789 3790 /* align data to page boundary */ 3791 if (lseek(fd, data_offset, SEEK_SET) != data_offset) 3792 goto out; 3793 3794 /* 3795 * Finally we can dump process memory into corefile as well. 3796 */ 3797 for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) { 3798 abi_ulong addr; 3799 abi_ulong end; 3800 3801 end = vma->vma_start + vma_dump_size(vma); 3802 3803 for (addr = vma->vma_start; addr < end; 3804 addr += TARGET_PAGE_SIZE) { 3805 char page[TARGET_PAGE_SIZE]; 3806 int error; 3807 3808 /* 3809 * Read in page from target process memory and 3810 * write it to coredump file. 3811 */ 3812 error = copy_from_user(page, addr, sizeof (page)); 3813 if (error != 0) { 3814 (void) fprintf(stderr, "unable to dump " TARGET_ABI_FMT_lx "\n", 3815 addr); 3816 errno = -error; 3817 goto out; 3818 } 3819 if (dump_write(fd, page, TARGET_PAGE_SIZE) < 0) 3820 goto out; 3821 } 3822 } 3823 3824 out: 3825 free_note_info(&info); 3826 if (mm != NULL) 3827 vma_delete(mm); 3828 (void) close(fd); 3829 3830 if (errno != 0) 3831 return (-errno); 3832 return (0); 3833 } 3834 #endif /* USE_ELF_CORE_DUMP */ 3835 3836 void do_init_thread(struct target_pt_regs *regs, struct image_info *infop) 3837 { 3838 init_thread(regs, infop); 3839 } 3840