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