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 "qemu/lockable.h" 21 #include "qapi/error.h" 22 #include "qemu/error-report.h" 23 #include "target_signal.h" 24 #include "accel/tcg/debuginfo.h" 25 26 #ifdef _ARCH_PPC64 27 #undef ARCH_DLINFO 28 #undef ELF_PLATFORM 29 #undef ELF_HWCAP 30 #undef ELF_HWCAP2 31 #undef ELF_CLASS 32 #undef ELF_DATA 33 #undef ELF_ARCH 34 #endif 35 36 #define ELF_OSABI ELFOSABI_SYSV 37 38 /* from personality.h */ 39 40 /* 41 * Flags for bug emulation. 42 * 43 * These occupy the top three bytes. 44 */ 45 enum { 46 ADDR_NO_RANDOMIZE = 0x0040000, /* disable randomization of VA space */ 47 FDPIC_FUNCPTRS = 0x0080000, /* userspace function ptrs point to 48 descriptors (signal handling) */ 49 MMAP_PAGE_ZERO = 0x0100000, 50 ADDR_COMPAT_LAYOUT = 0x0200000, 51 READ_IMPLIES_EXEC = 0x0400000, 52 ADDR_LIMIT_32BIT = 0x0800000, 53 SHORT_INODE = 0x1000000, 54 WHOLE_SECONDS = 0x2000000, 55 STICKY_TIMEOUTS = 0x4000000, 56 ADDR_LIMIT_3GB = 0x8000000, 57 }; 58 59 /* 60 * Personality types. 61 * 62 * These go in the low byte. Avoid using the top bit, it will 63 * conflict with error returns. 64 */ 65 enum { 66 PER_LINUX = 0x0000, 67 PER_LINUX_32BIT = 0x0000 | ADDR_LIMIT_32BIT, 68 PER_LINUX_FDPIC = 0x0000 | FDPIC_FUNCPTRS, 69 PER_SVR4 = 0x0001 | STICKY_TIMEOUTS | MMAP_PAGE_ZERO, 70 PER_SVR3 = 0x0002 | STICKY_TIMEOUTS | SHORT_INODE, 71 PER_SCOSVR3 = 0x0003 | STICKY_TIMEOUTS | WHOLE_SECONDS | SHORT_INODE, 72 PER_OSR5 = 0x0003 | STICKY_TIMEOUTS | WHOLE_SECONDS, 73 PER_WYSEV386 = 0x0004 | STICKY_TIMEOUTS | SHORT_INODE, 74 PER_ISCR4 = 0x0005 | STICKY_TIMEOUTS, 75 PER_BSD = 0x0006, 76 PER_SUNOS = 0x0006 | STICKY_TIMEOUTS, 77 PER_XENIX = 0x0007 | STICKY_TIMEOUTS | SHORT_INODE, 78 PER_LINUX32 = 0x0008, 79 PER_LINUX32_3GB = 0x0008 | ADDR_LIMIT_3GB, 80 PER_IRIX32 = 0x0009 | STICKY_TIMEOUTS,/* IRIX5 32-bit */ 81 PER_IRIXN32 = 0x000a | STICKY_TIMEOUTS,/* IRIX6 new 32-bit */ 82 PER_IRIX64 = 0x000b | STICKY_TIMEOUTS,/* IRIX6 64-bit */ 83 PER_RISCOS = 0x000c, 84 PER_SOLARIS = 0x000d | STICKY_TIMEOUTS, 85 PER_UW7 = 0x000e | STICKY_TIMEOUTS | MMAP_PAGE_ZERO, 86 PER_OSF4 = 0x000f, /* OSF/1 v4 */ 87 PER_HPUX = 0x0010, 88 PER_MASK = 0x00ff, 89 }; 90 91 /* 92 * Return the base personality without flags. 93 */ 94 #define personality(pers) (pers & PER_MASK) 95 96 int info_is_fdpic(struct image_info *info) 97 { 98 return info->personality == PER_LINUX_FDPIC; 99 } 100 101 /* this flag is uneffective under linux too, should be deleted */ 102 #ifndef MAP_DENYWRITE 103 #define MAP_DENYWRITE 0 104 #endif 105 106 /* should probably go in elf.h */ 107 #ifndef ELIBBAD 108 #define ELIBBAD 80 109 #endif 110 111 #if TARGET_BIG_ENDIAN 112 #define ELF_DATA ELFDATA2MSB 113 #else 114 #define ELF_DATA ELFDATA2LSB 115 #endif 116 117 #ifdef TARGET_ABI_MIPSN32 118 typedef abi_ullong target_elf_greg_t; 119 #define tswapreg(ptr) tswap64(ptr) 120 #else 121 typedef abi_ulong target_elf_greg_t; 122 #define tswapreg(ptr) tswapal(ptr) 123 #endif 124 125 #ifdef USE_UID16 126 typedef abi_ushort target_uid_t; 127 typedef abi_ushort target_gid_t; 128 #else 129 typedef abi_uint target_uid_t; 130 typedef abi_uint target_gid_t; 131 #endif 132 typedef abi_int target_pid_t; 133 134 #ifdef TARGET_I386 135 136 #define ELF_HWCAP get_elf_hwcap() 137 138 static uint32_t get_elf_hwcap(void) 139 { 140 X86CPU *cpu = X86_CPU(thread_cpu); 141 142 return cpu->env.features[FEAT_1_EDX]; 143 } 144 145 #ifdef TARGET_X86_64 146 #define ELF_CLASS ELFCLASS64 147 #define ELF_ARCH EM_X86_64 148 149 #define ELF_PLATFORM "x86_64" 150 151 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop) 152 { 153 regs->rax = 0; 154 regs->rsp = infop->start_stack; 155 regs->rip = infop->entry; 156 } 157 158 #define ELF_NREG 27 159 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 160 161 /* 162 * Note that ELF_NREG should be 29 as there should be place for 163 * TRAPNO and ERR "registers" as well but linux doesn't dump 164 * those. 165 * 166 * See linux kernel: arch/x86/include/asm/elf.h 167 */ 168 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUX86State *env) 169 { 170 (*regs)[0] = tswapreg(env->regs[15]); 171 (*regs)[1] = tswapreg(env->regs[14]); 172 (*regs)[2] = tswapreg(env->regs[13]); 173 (*regs)[3] = tswapreg(env->regs[12]); 174 (*regs)[4] = tswapreg(env->regs[R_EBP]); 175 (*regs)[5] = tswapreg(env->regs[R_EBX]); 176 (*regs)[6] = tswapreg(env->regs[11]); 177 (*regs)[7] = tswapreg(env->regs[10]); 178 (*regs)[8] = tswapreg(env->regs[9]); 179 (*regs)[9] = tswapreg(env->regs[8]); 180 (*regs)[10] = tswapreg(env->regs[R_EAX]); 181 (*regs)[11] = tswapreg(env->regs[R_ECX]); 182 (*regs)[12] = tswapreg(env->regs[R_EDX]); 183 (*regs)[13] = tswapreg(env->regs[R_ESI]); 184 (*regs)[14] = tswapreg(env->regs[R_EDI]); 185 (*regs)[15] = tswapreg(env->regs[R_EAX]); /* XXX */ 186 (*regs)[16] = tswapreg(env->eip); 187 (*regs)[17] = tswapreg(env->segs[R_CS].selector & 0xffff); 188 (*regs)[18] = tswapreg(env->eflags); 189 (*regs)[19] = tswapreg(env->regs[R_ESP]); 190 (*regs)[20] = tswapreg(env->segs[R_SS].selector & 0xffff); 191 (*regs)[21] = tswapreg(env->segs[R_FS].selector & 0xffff); 192 (*regs)[22] = tswapreg(env->segs[R_GS].selector & 0xffff); 193 (*regs)[23] = tswapreg(env->segs[R_DS].selector & 0xffff); 194 (*regs)[24] = tswapreg(env->segs[R_ES].selector & 0xffff); 195 (*regs)[25] = tswapreg(env->segs[R_FS].selector & 0xffff); 196 (*regs)[26] = tswapreg(env->segs[R_GS].selector & 0xffff); 197 } 198 199 #if ULONG_MAX > UINT32_MAX 200 #define INIT_GUEST_COMMPAGE 201 static bool init_guest_commpage(void) 202 { 203 /* 204 * The vsyscall page is at a high negative address aka kernel space, 205 * which means that we cannot actually allocate it with target_mmap. 206 * We still should be able to use page_set_flags, unless the user 207 * has specified -R reserved_va, which would trigger an assert(). 208 */ 209 if (reserved_va != 0 && 210 TARGET_VSYSCALL_PAGE + TARGET_PAGE_SIZE - 1 > reserved_va) { 211 error_report("Cannot allocate vsyscall page"); 212 exit(EXIT_FAILURE); 213 } 214 page_set_flags(TARGET_VSYSCALL_PAGE, 215 TARGET_VSYSCALL_PAGE | ~TARGET_PAGE_MASK, 216 PAGE_EXEC | PAGE_VALID); 217 return true; 218 } 219 #endif 220 #else 221 222 /* 223 * This is used to ensure we don't load something for the wrong architecture. 224 */ 225 #define elf_check_arch(x) ( ((x) == EM_386) || ((x) == EM_486) ) 226 227 /* 228 * These are used to set parameters in the core dumps. 229 */ 230 #define ELF_CLASS ELFCLASS32 231 #define ELF_ARCH EM_386 232 233 #define ELF_PLATFORM get_elf_platform() 234 #define EXSTACK_DEFAULT true 235 236 static const char *get_elf_platform(void) 237 { 238 static char elf_platform[] = "i386"; 239 int family = object_property_get_int(OBJECT(thread_cpu), "family", NULL); 240 if (family > 6) { 241 family = 6; 242 } 243 if (family >= 3) { 244 elf_platform[1] = '0' + family; 245 } 246 return elf_platform; 247 } 248 249 static inline void init_thread(struct target_pt_regs *regs, 250 struct image_info *infop) 251 { 252 regs->esp = infop->start_stack; 253 regs->eip = infop->entry; 254 255 /* SVR4/i386 ABI (pages 3-31, 3-32) says that when the program 256 starts %edx contains a pointer to a function which might be 257 registered using `atexit'. This provides a mean for the 258 dynamic linker to call DT_FINI functions for shared libraries 259 that have been loaded before the code runs. 260 261 A value of 0 tells we have no such handler. */ 262 regs->edx = 0; 263 } 264 265 #define ELF_NREG 17 266 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 267 268 /* 269 * Note that ELF_NREG should be 19 as there should be place for 270 * TRAPNO and ERR "registers" as well but linux doesn't dump 271 * those. 272 * 273 * See linux kernel: arch/x86/include/asm/elf.h 274 */ 275 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUX86State *env) 276 { 277 (*regs)[0] = tswapreg(env->regs[R_EBX]); 278 (*regs)[1] = tswapreg(env->regs[R_ECX]); 279 (*regs)[2] = tswapreg(env->regs[R_EDX]); 280 (*regs)[3] = tswapreg(env->regs[R_ESI]); 281 (*regs)[4] = tswapreg(env->regs[R_EDI]); 282 (*regs)[5] = tswapreg(env->regs[R_EBP]); 283 (*regs)[6] = tswapreg(env->regs[R_EAX]); 284 (*regs)[7] = tswapreg(env->segs[R_DS].selector & 0xffff); 285 (*regs)[8] = tswapreg(env->segs[R_ES].selector & 0xffff); 286 (*regs)[9] = tswapreg(env->segs[R_FS].selector & 0xffff); 287 (*regs)[10] = tswapreg(env->segs[R_GS].selector & 0xffff); 288 (*regs)[11] = tswapreg(env->regs[R_EAX]); /* XXX */ 289 (*regs)[12] = tswapreg(env->eip); 290 (*regs)[13] = tswapreg(env->segs[R_CS].selector & 0xffff); 291 (*regs)[14] = tswapreg(env->eflags); 292 (*regs)[15] = tswapreg(env->regs[R_ESP]); 293 (*regs)[16] = tswapreg(env->segs[R_SS].selector & 0xffff); 294 } 295 #endif 296 297 #define USE_ELF_CORE_DUMP 298 #define ELF_EXEC_PAGESIZE 4096 299 300 #endif 301 302 #ifdef TARGET_ARM 303 304 #ifndef TARGET_AARCH64 305 /* 32 bit ARM definitions */ 306 307 #define ELF_ARCH EM_ARM 308 #define ELF_CLASS ELFCLASS32 309 #define EXSTACK_DEFAULT true 310 311 static inline void init_thread(struct target_pt_regs *regs, 312 struct image_info *infop) 313 { 314 abi_long stack = infop->start_stack; 315 memset(regs, 0, sizeof(*regs)); 316 317 regs->uregs[16] = ARM_CPU_MODE_USR; 318 if (infop->entry & 1) { 319 regs->uregs[16] |= CPSR_T; 320 } 321 regs->uregs[15] = infop->entry & 0xfffffffe; 322 regs->uregs[13] = infop->start_stack; 323 /* FIXME - what to for failure of get_user()? */ 324 get_user_ual(regs->uregs[2], stack + 8); /* envp */ 325 get_user_ual(regs->uregs[1], stack + 4); /* envp */ 326 /* XXX: it seems that r0 is zeroed after ! */ 327 regs->uregs[0] = 0; 328 /* For uClinux PIC binaries. */ 329 /* XXX: Linux does this only on ARM with no MMU (do we care ?) */ 330 regs->uregs[10] = infop->start_data; 331 332 /* Support ARM FDPIC. */ 333 if (info_is_fdpic(infop)) { 334 /* As described in the ABI document, r7 points to the loadmap info 335 * prepared by the kernel. If an interpreter is needed, r8 points 336 * to the interpreter loadmap and r9 points to the interpreter 337 * PT_DYNAMIC info. If no interpreter is needed, r8 is zero, and 338 * r9 points to the main program PT_DYNAMIC info. 339 */ 340 regs->uregs[7] = infop->loadmap_addr; 341 if (infop->interpreter_loadmap_addr) { 342 /* Executable is dynamically loaded. */ 343 regs->uregs[8] = infop->interpreter_loadmap_addr; 344 regs->uregs[9] = infop->interpreter_pt_dynamic_addr; 345 } else { 346 regs->uregs[8] = 0; 347 regs->uregs[9] = infop->pt_dynamic_addr; 348 } 349 } 350 } 351 352 #define ELF_NREG 18 353 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 354 355 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUARMState *env) 356 { 357 (*regs)[0] = tswapreg(env->regs[0]); 358 (*regs)[1] = tswapreg(env->regs[1]); 359 (*regs)[2] = tswapreg(env->regs[2]); 360 (*regs)[3] = tswapreg(env->regs[3]); 361 (*regs)[4] = tswapreg(env->regs[4]); 362 (*regs)[5] = tswapreg(env->regs[5]); 363 (*regs)[6] = tswapreg(env->regs[6]); 364 (*regs)[7] = tswapreg(env->regs[7]); 365 (*regs)[8] = tswapreg(env->regs[8]); 366 (*regs)[9] = tswapreg(env->regs[9]); 367 (*regs)[10] = tswapreg(env->regs[10]); 368 (*regs)[11] = tswapreg(env->regs[11]); 369 (*regs)[12] = tswapreg(env->regs[12]); 370 (*regs)[13] = tswapreg(env->regs[13]); 371 (*regs)[14] = tswapreg(env->regs[14]); 372 (*regs)[15] = tswapreg(env->regs[15]); 373 374 (*regs)[16] = tswapreg(cpsr_read((CPUARMState *)env)); 375 (*regs)[17] = tswapreg(env->regs[0]); /* XXX */ 376 } 377 378 #define USE_ELF_CORE_DUMP 379 #define ELF_EXEC_PAGESIZE 4096 380 381 enum 382 { 383 ARM_HWCAP_ARM_SWP = 1 << 0, 384 ARM_HWCAP_ARM_HALF = 1 << 1, 385 ARM_HWCAP_ARM_THUMB = 1 << 2, 386 ARM_HWCAP_ARM_26BIT = 1 << 3, 387 ARM_HWCAP_ARM_FAST_MULT = 1 << 4, 388 ARM_HWCAP_ARM_FPA = 1 << 5, 389 ARM_HWCAP_ARM_VFP = 1 << 6, 390 ARM_HWCAP_ARM_EDSP = 1 << 7, 391 ARM_HWCAP_ARM_JAVA = 1 << 8, 392 ARM_HWCAP_ARM_IWMMXT = 1 << 9, 393 ARM_HWCAP_ARM_CRUNCH = 1 << 10, 394 ARM_HWCAP_ARM_THUMBEE = 1 << 11, 395 ARM_HWCAP_ARM_NEON = 1 << 12, 396 ARM_HWCAP_ARM_VFPv3 = 1 << 13, 397 ARM_HWCAP_ARM_VFPv3D16 = 1 << 14, 398 ARM_HWCAP_ARM_TLS = 1 << 15, 399 ARM_HWCAP_ARM_VFPv4 = 1 << 16, 400 ARM_HWCAP_ARM_IDIVA = 1 << 17, 401 ARM_HWCAP_ARM_IDIVT = 1 << 18, 402 ARM_HWCAP_ARM_VFPD32 = 1 << 19, 403 ARM_HWCAP_ARM_LPAE = 1 << 20, 404 ARM_HWCAP_ARM_EVTSTRM = 1 << 21, 405 }; 406 407 enum { 408 ARM_HWCAP2_ARM_AES = 1 << 0, 409 ARM_HWCAP2_ARM_PMULL = 1 << 1, 410 ARM_HWCAP2_ARM_SHA1 = 1 << 2, 411 ARM_HWCAP2_ARM_SHA2 = 1 << 3, 412 ARM_HWCAP2_ARM_CRC32 = 1 << 4, 413 }; 414 415 /* The commpage only exists for 32 bit kernels */ 416 417 #define HI_COMMPAGE (intptr_t)0xffff0f00u 418 419 static bool init_guest_commpage(void) 420 { 421 ARMCPU *cpu = ARM_CPU(thread_cpu); 422 abi_ptr commpage; 423 void *want; 424 void *addr; 425 426 /* 427 * M-profile allocates maximum of 2GB address space, so can never 428 * allocate the commpage. Skip it. 429 */ 430 if (arm_feature(&cpu->env, ARM_FEATURE_M)) { 431 return true; 432 } 433 434 commpage = HI_COMMPAGE & -qemu_host_page_size; 435 want = g2h_untagged(commpage); 436 addr = mmap(want, qemu_host_page_size, PROT_READ | PROT_WRITE, 437 MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0); 438 439 if (addr == MAP_FAILED) { 440 perror("Allocating guest commpage"); 441 exit(EXIT_FAILURE); 442 } 443 if (addr != want) { 444 return false; 445 } 446 447 /* Set kernel helper versions; rest of page is 0. */ 448 __put_user(5, (uint32_t *)g2h_untagged(0xffff0ffcu)); 449 450 if (mprotect(addr, qemu_host_page_size, PROT_READ)) { 451 perror("Protecting guest commpage"); 452 exit(EXIT_FAILURE); 453 } 454 455 page_set_flags(commpage, commpage | ~qemu_host_page_mask, 456 PAGE_READ | PAGE_EXEC | PAGE_VALID); 457 return true; 458 } 459 460 #define ELF_HWCAP get_elf_hwcap() 461 #define ELF_HWCAP2 get_elf_hwcap2() 462 463 uint32_t get_elf_hwcap(void) 464 { 465 ARMCPU *cpu = ARM_CPU(thread_cpu); 466 uint32_t hwcaps = 0; 467 468 hwcaps |= ARM_HWCAP_ARM_SWP; 469 hwcaps |= ARM_HWCAP_ARM_HALF; 470 hwcaps |= ARM_HWCAP_ARM_THUMB; 471 hwcaps |= ARM_HWCAP_ARM_FAST_MULT; 472 473 /* probe for the extra features */ 474 #define GET_FEATURE(feat, hwcap) \ 475 do { if (arm_feature(&cpu->env, feat)) { hwcaps |= hwcap; } } while (0) 476 477 #define GET_FEATURE_ID(feat, hwcap) \ 478 do { if (cpu_isar_feature(feat, cpu)) { hwcaps |= hwcap; } } while (0) 479 480 /* EDSP is in v5TE and above, but all our v5 CPUs are v5TE */ 481 GET_FEATURE(ARM_FEATURE_V5, ARM_HWCAP_ARM_EDSP); 482 GET_FEATURE(ARM_FEATURE_IWMMXT, ARM_HWCAP_ARM_IWMMXT); 483 GET_FEATURE(ARM_FEATURE_THUMB2EE, ARM_HWCAP_ARM_THUMBEE); 484 GET_FEATURE(ARM_FEATURE_NEON, ARM_HWCAP_ARM_NEON); 485 GET_FEATURE(ARM_FEATURE_V6K, ARM_HWCAP_ARM_TLS); 486 GET_FEATURE(ARM_FEATURE_LPAE, ARM_HWCAP_ARM_LPAE); 487 GET_FEATURE_ID(aa32_arm_div, ARM_HWCAP_ARM_IDIVA); 488 GET_FEATURE_ID(aa32_thumb_div, ARM_HWCAP_ARM_IDIVT); 489 GET_FEATURE_ID(aa32_vfp, ARM_HWCAP_ARM_VFP); 490 491 if (cpu_isar_feature(aa32_fpsp_v3, cpu) || 492 cpu_isar_feature(aa32_fpdp_v3, cpu)) { 493 hwcaps |= ARM_HWCAP_ARM_VFPv3; 494 if (cpu_isar_feature(aa32_simd_r32, cpu)) { 495 hwcaps |= ARM_HWCAP_ARM_VFPD32; 496 } else { 497 hwcaps |= ARM_HWCAP_ARM_VFPv3D16; 498 } 499 } 500 GET_FEATURE_ID(aa32_simdfmac, ARM_HWCAP_ARM_VFPv4); 501 502 return hwcaps; 503 } 504 505 uint32_t get_elf_hwcap2(void) 506 { 507 ARMCPU *cpu = ARM_CPU(thread_cpu); 508 uint32_t hwcaps = 0; 509 510 GET_FEATURE_ID(aa32_aes, ARM_HWCAP2_ARM_AES); 511 GET_FEATURE_ID(aa32_pmull, ARM_HWCAP2_ARM_PMULL); 512 GET_FEATURE_ID(aa32_sha1, ARM_HWCAP2_ARM_SHA1); 513 GET_FEATURE_ID(aa32_sha2, ARM_HWCAP2_ARM_SHA2); 514 GET_FEATURE_ID(aa32_crc32, ARM_HWCAP2_ARM_CRC32); 515 return hwcaps; 516 } 517 518 const char *elf_hwcap_str(uint32_t bit) 519 { 520 static const char *hwcap_str[] = { 521 [__builtin_ctz(ARM_HWCAP_ARM_SWP )] = "swp", 522 [__builtin_ctz(ARM_HWCAP_ARM_HALF )] = "half", 523 [__builtin_ctz(ARM_HWCAP_ARM_THUMB )] = "thumb", 524 [__builtin_ctz(ARM_HWCAP_ARM_26BIT )] = "26bit", 525 [__builtin_ctz(ARM_HWCAP_ARM_FAST_MULT)] = "fast_mult", 526 [__builtin_ctz(ARM_HWCAP_ARM_FPA )] = "fpa", 527 [__builtin_ctz(ARM_HWCAP_ARM_VFP )] = "vfp", 528 [__builtin_ctz(ARM_HWCAP_ARM_EDSP )] = "edsp", 529 [__builtin_ctz(ARM_HWCAP_ARM_JAVA )] = "java", 530 [__builtin_ctz(ARM_HWCAP_ARM_IWMMXT )] = "iwmmxt", 531 [__builtin_ctz(ARM_HWCAP_ARM_CRUNCH )] = "crunch", 532 [__builtin_ctz(ARM_HWCAP_ARM_THUMBEE )] = "thumbee", 533 [__builtin_ctz(ARM_HWCAP_ARM_NEON )] = "neon", 534 [__builtin_ctz(ARM_HWCAP_ARM_VFPv3 )] = "vfpv3", 535 [__builtin_ctz(ARM_HWCAP_ARM_VFPv3D16 )] = "vfpv3d16", 536 [__builtin_ctz(ARM_HWCAP_ARM_TLS )] = "tls", 537 [__builtin_ctz(ARM_HWCAP_ARM_VFPv4 )] = "vfpv4", 538 [__builtin_ctz(ARM_HWCAP_ARM_IDIVA )] = "idiva", 539 [__builtin_ctz(ARM_HWCAP_ARM_IDIVT )] = "idivt", 540 [__builtin_ctz(ARM_HWCAP_ARM_VFPD32 )] = "vfpd32", 541 [__builtin_ctz(ARM_HWCAP_ARM_LPAE )] = "lpae", 542 [__builtin_ctz(ARM_HWCAP_ARM_EVTSTRM )] = "evtstrm", 543 }; 544 545 return bit < ARRAY_SIZE(hwcap_str) ? hwcap_str[bit] : NULL; 546 } 547 548 const char *elf_hwcap2_str(uint32_t bit) 549 { 550 static const char *hwcap_str[] = { 551 [__builtin_ctz(ARM_HWCAP2_ARM_AES )] = "aes", 552 [__builtin_ctz(ARM_HWCAP2_ARM_PMULL)] = "pmull", 553 [__builtin_ctz(ARM_HWCAP2_ARM_SHA1 )] = "sha1", 554 [__builtin_ctz(ARM_HWCAP2_ARM_SHA2 )] = "sha2", 555 [__builtin_ctz(ARM_HWCAP2_ARM_CRC32)] = "crc32", 556 }; 557 558 return bit < ARRAY_SIZE(hwcap_str) ? hwcap_str[bit] : NULL; 559 } 560 561 #undef GET_FEATURE 562 #undef GET_FEATURE_ID 563 564 #define ELF_PLATFORM get_elf_platform() 565 566 static const char *get_elf_platform(void) 567 { 568 CPUARMState *env = thread_cpu->env_ptr; 569 570 #if TARGET_BIG_ENDIAN 571 # define END "b" 572 #else 573 # define END "l" 574 #endif 575 576 if (arm_feature(env, ARM_FEATURE_V8)) { 577 return "v8" END; 578 } else if (arm_feature(env, ARM_FEATURE_V7)) { 579 if (arm_feature(env, ARM_FEATURE_M)) { 580 return "v7m" END; 581 } else { 582 return "v7" END; 583 } 584 } else if (arm_feature(env, ARM_FEATURE_V6)) { 585 return "v6" END; 586 } else if (arm_feature(env, ARM_FEATURE_V5)) { 587 return "v5" END; 588 } else { 589 return "v4" END; 590 } 591 592 #undef END 593 } 594 595 #else 596 /* 64 bit ARM definitions */ 597 598 #define ELF_ARCH EM_AARCH64 599 #define ELF_CLASS ELFCLASS64 600 #if TARGET_BIG_ENDIAN 601 # define ELF_PLATFORM "aarch64_be" 602 #else 603 # define ELF_PLATFORM "aarch64" 604 #endif 605 606 static inline void init_thread(struct target_pt_regs *regs, 607 struct image_info *infop) 608 { 609 abi_long stack = infop->start_stack; 610 memset(regs, 0, sizeof(*regs)); 611 612 regs->pc = infop->entry & ~0x3ULL; 613 regs->sp = stack; 614 } 615 616 #define ELF_NREG 34 617 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 618 619 static void elf_core_copy_regs(target_elf_gregset_t *regs, 620 const CPUARMState *env) 621 { 622 int i; 623 624 for (i = 0; i < 32; i++) { 625 (*regs)[i] = tswapreg(env->xregs[i]); 626 } 627 (*regs)[32] = tswapreg(env->pc); 628 (*regs)[33] = tswapreg(pstate_read((CPUARMState *)env)); 629 } 630 631 #define USE_ELF_CORE_DUMP 632 #define ELF_EXEC_PAGESIZE 4096 633 634 enum { 635 ARM_HWCAP_A64_FP = 1 << 0, 636 ARM_HWCAP_A64_ASIMD = 1 << 1, 637 ARM_HWCAP_A64_EVTSTRM = 1 << 2, 638 ARM_HWCAP_A64_AES = 1 << 3, 639 ARM_HWCAP_A64_PMULL = 1 << 4, 640 ARM_HWCAP_A64_SHA1 = 1 << 5, 641 ARM_HWCAP_A64_SHA2 = 1 << 6, 642 ARM_HWCAP_A64_CRC32 = 1 << 7, 643 ARM_HWCAP_A64_ATOMICS = 1 << 8, 644 ARM_HWCAP_A64_FPHP = 1 << 9, 645 ARM_HWCAP_A64_ASIMDHP = 1 << 10, 646 ARM_HWCAP_A64_CPUID = 1 << 11, 647 ARM_HWCAP_A64_ASIMDRDM = 1 << 12, 648 ARM_HWCAP_A64_JSCVT = 1 << 13, 649 ARM_HWCAP_A64_FCMA = 1 << 14, 650 ARM_HWCAP_A64_LRCPC = 1 << 15, 651 ARM_HWCAP_A64_DCPOP = 1 << 16, 652 ARM_HWCAP_A64_SHA3 = 1 << 17, 653 ARM_HWCAP_A64_SM3 = 1 << 18, 654 ARM_HWCAP_A64_SM4 = 1 << 19, 655 ARM_HWCAP_A64_ASIMDDP = 1 << 20, 656 ARM_HWCAP_A64_SHA512 = 1 << 21, 657 ARM_HWCAP_A64_SVE = 1 << 22, 658 ARM_HWCAP_A64_ASIMDFHM = 1 << 23, 659 ARM_HWCAP_A64_DIT = 1 << 24, 660 ARM_HWCAP_A64_USCAT = 1 << 25, 661 ARM_HWCAP_A64_ILRCPC = 1 << 26, 662 ARM_HWCAP_A64_FLAGM = 1 << 27, 663 ARM_HWCAP_A64_SSBS = 1 << 28, 664 ARM_HWCAP_A64_SB = 1 << 29, 665 ARM_HWCAP_A64_PACA = 1 << 30, 666 ARM_HWCAP_A64_PACG = 1UL << 31, 667 668 ARM_HWCAP2_A64_DCPODP = 1 << 0, 669 ARM_HWCAP2_A64_SVE2 = 1 << 1, 670 ARM_HWCAP2_A64_SVEAES = 1 << 2, 671 ARM_HWCAP2_A64_SVEPMULL = 1 << 3, 672 ARM_HWCAP2_A64_SVEBITPERM = 1 << 4, 673 ARM_HWCAP2_A64_SVESHA3 = 1 << 5, 674 ARM_HWCAP2_A64_SVESM4 = 1 << 6, 675 ARM_HWCAP2_A64_FLAGM2 = 1 << 7, 676 ARM_HWCAP2_A64_FRINT = 1 << 8, 677 ARM_HWCAP2_A64_SVEI8MM = 1 << 9, 678 ARM_HWCAP2_A64_SVEF32MM = 1 << 10, 679 ARM_HWCAP2_A64_SVEF64MM = 1 << 11, 680 ARM_HWCAP2_A64_SVEBF16 = 1 << 12, 681 ARM_HWCAP2_A64_I8MM = 1 << 13, 682 ARM_HWCAP2_A64_BF16 = 1 << 14, 683 ARM_HWCAP2_A64_DGH = 1 << 15, 684 ARM_HWCAP2_A64_RNG = 1 << 16, 685 ARM_HWCAP2_A64_BTI = 1 << 17, 686 ARM_HWCAP2_A64_MTE = 1 << 18, 687 ARM_HWCAP2_A64_ECV = 1 << 19, 688 ARM_HWCAP2_A64_AFP = 1 << 20, 689 ARM_HWCAP2_A64_RPRES = 1 << 21, 690 ARM_HWCAP2_A64_MTE3 = 1 << 22, 691 ARM_HWCAP2_A64_SME = 1 << 23, 692 ARM_HWCAP2_A64_SME_I16I64 = 1 << 24, 693 ARM_HWCAP2_A64_SME_F64F64 = 1 << 25, 694 ARM_HWCAP2_A64_SME_I8I32 = 1 << 26, 695 ARM_HWCAP2_A64_SME_F16F32 = 1 << 27, 696 ARM_HWCAP2_A64_SME_B16F32 = 1 << 28, 697 ARM_HWCAP2_A64_SME_F32F32 = 1 << 29, 698 ARM_HWCAP2_A64_SME_FA64 = 1 << 30, 699 }; 700 701 #define ELF_HWCAP get_elf_hwcap() 702 #define ELF_HWCAP2 get_elf_hwcap2() 703 704 #define GET_FEATURE_ID(feat, hwcap) \ 705 do { if (cpu_isar_feature(feat, cpu)) { hwcaps |= hwcap; } } while (0) 706 707 uint32_t get_elf_hwcap(void) 708 { 709 ARMCPU *cpu = ARM_CPU(thread_cpu); 710 uint32_t hwcaps = 0; 711 712 hwcaps |= ARM_HWCAP_A64_FP; 713 hwcaps |= ARM_HWCAP_A64_ASIMD; 714 hwcaps |= ARM_HWCAP_A64_CPUID; 715 716 /* probe for the extra features */ 717 718 GET_FEATURE_ID(aa64_aes, ARM_HWCAP_A64_AES); 719 GET_FEATURE_ID(aa64_pmull, ARM_HWCAP_A64_PMULL); 720 GET_FEATURE_ID(aa64_sha1, ARM_HWCAP_A64_SHA1); 721 GET_FEATURE_ID(aa64_sha256, ARM_HWCAP_A64_SHA2); 722 GET_FEATURE_ID(aa64_sha512, ARM_HWCAP_A64_SHA512); 723 GET_FEATURE_ID(aa64_crc32, ARM_HWCAP_A64_CRC32); 724 GET_FEATURE_ID(aa64_sha3, ARM_HWCAP_A64_SHA3); 725 GET_FEATURE_ID(aa64_sm3, ARM_HWCAP_A64_SM3); 726 GET_FEATURE_ID(aa64_sm4, ARM_HWCAP_A64_SM4); 727 GET_FEATURE_ID(aa64_fp16, ARM_HWCAP_A64_FPHP | ARM_HWCAP_A64_ASIMDHP); 728 GET_FEATURE_ID(aa64_atomics, ARM_HWCAP_A64_ATOMICS); 729 GET_FEATURE_ID(aa64_rdm, ARM_HWCAP_A64_ASIMDRDM); 730 GET_FEATURE_ID(aa64_dp, ARM_HWCAP_A64_ASIMDDP); 731 GET_FEATURE_ID(aa64_fcma, ARM_HWCAP_A64_FCMA); 732 GET_FEATURE_ID(aa64_sve, ARM_HWCAP_A64_SVE); 733 GET_FEATURE_ID(aa64_pauth, ARM_HWCAP_A64_PACA | ARM_HWCAP_A64_PACG); 734 GET_FEATURE_ID(aa64_fhm, ARM_HWCAP_A64_ASIMDFHM); 735 GET_FEATURE_ID(aa64_jscvt, ARM_HWCAP_A64_JSCVT); 736 GET_FEATURE_ID(aa64_sb, ARM_HWCAP_A64_SB); 737 GET_FEATURE_ID(aa64_condm_4, ARM_HWCAP_A64_FLAGM); 738 GET_FEATURE_ID(aa64_dcpop, ARM_HWCAP_A64_DCPOP); 739 GET_FEATURE_ID(aa64_rcpc_8_3, ARM_HWCAP_A64_LRCPC); 740 GET_FEATURE_ID(aa64_rcpc_8_4, ARM_HWCAP_A64_ILRCPC); 741 742 return hwcaps; 743 } 744 745 uint32_t get_elf_hwcap2(void) 746 { 747 ARMCPU *cpu = ARM_CPU(thread_cpu); 748 uint32_t hwcaps = 0; 749 750 GET_FEATURE_ID(aa64_dcpodp, ARM_HWCAP2_A64_DCPODP); 751 GET_FEATURE_ID(aa64_sve2, ARM_HWCAP2_A64_SVE2); 752 GET_FEATURE_ID(aa64_sve2_aes, ARM_HWCAP2_A64_SVEAES); 753 GET_FEATURE_ID(aa64_sve2_pmull128, ARM_HWCAP2_A64_SVEPMULL); 754 GET_FEATURE_ID(aa64_sve2_bitperm, ARM_HWCAP2_A64_SVEBITPERM); 755 GET_FEATURE_ID(aa64_sve2_sha3, ARM_HWCAP2_A64_SVESHA3); 756 GET_FEATURE_ID(aa64_sve2_sm4, ARM_HWCAP2_A64_SVESM4); 757 GET_FEATURE_ID(aa64_condm_5, ARM_HWCAP2_A64_FLAGM2); 758 GET_FEATURE_ID(aa64_frint, ARM_HWCAP2_A64_FRINT); 759 GET_FEATURE_ID(aa64_sve_i8mm, ARM_HWCAP2_A64_SVEI8MM); 760 GET_FEATURE_ID(aa64_sve_f32mm, ARM_HWCAP2_A64_SVEF32MM); 761 GET_FEATURE_ID(aa64_sve_f64mm, ARM_HWCAP2_A64_SVEF64MM); 762 GET_FEATURE_ID(aa64_sve_bf16, ARM_HWCAP2_A64_SVEBF16); 763 GET_FEATURE_ID(aa64_i8mm, ARM_HWCAP2_A64_I8MM); 764 GET_FEATURE_ID(aa64_bf16, ARM_HWCAP2_A64_BF16); 765 GET_FEATURE_ID(aa64_rndr, ARM_HWCAP2_A64_RNG); 766 GET_FEATURE_ID(aa64_bti, ARM_HWCAP2_A64_BTI); 767 GET_FEATURE_ID(aa64_mte, ARM_HWCAP2_A64_MTE); 768 GET_FEATURE_ID(aa64_sme, (ARM_HWCAP2_A64_SME | 769 ARM_HWCAP2_A64_SME_F32F32 | 770 ARM_HWCAP2_A64_SME_B16F32 | 771 ARM_HWCAP2_A64_SME_F16F32 | 772 ARM_HWCAP2_A64_SME_I8I32)); 773 GET_FEATURE_ID(aa64_sme_f64f64, ARM_HWCAP2_A64_SME_F64F64); 774 GET_FEATURE_ID(aa64_sme_i16i64, ARM_HWCAP2_A64_SME_I16I64); 775 GET_FEATURE_ID(aa64_sme_fa64, ARM_HWCAP2_A64_SME_FA64); 776 777 return hwcaps; 778 } 779 780 const char *elf_hwcap_str(uint32_t bit) 781 { 782 static const char *hwcap_str[] = { 783 [__builtin_ctz(ARM_HWCAP_A64_FP )] = "fp", 784 [__builtin_ctz(ARM_HWCAP_A64_ASIMD )] = "asimd", 785 [__builtin_ctz(ARM_HWCAP_A64_EVTSTRM )] = "evtstrm", 786 [__builtin_ctz(ARM_HWCAP_A64_AES )] = "aes", 787 [__builtin_ctz(ARM_HWCAP_A64_PMULL )] = "pmull", 788 [__builtin_ctz(ARM_HWCAP_A64_SHA1 )] = "sha1", 789 [__builtin_ctz(ARM_HWCAP_A64_SHA2 )] = "sha2", 790 [__builtin_ctz(ARM_HWCAP_A64_CRC32 )] = "crc32", 791 [__builtin_ctz(ARM_HWCAP_A64_ATOMICS )] = "atomics", 792 [__builtin_ctz(ARM_HWCAP_A64_FPHP )] = "fphp", 793 [__builtin_ctz(ARM_HWCAP_A64_ASIMDHP )] = "asimdhp", 794 [__builtin_ctz(ARM_HWCAP_A64_CPUID )] = "cpuid", 795 [__builtin_ctz(ARM_HWCAP_A64_ASIMDRDM)] = "asimdrdm", 796 [__builtin_ctz(ARM_HWCAP_A64_JSCVT )] = "jscvt", 797 [__builtin_ctz(ARM_HWCAP_A64_FCMA )] = "fcma", 798 [__builtin_ctz(ARM_HWCAP_A64_LRCPC )] = "lrcpc", 799 [__builtin_ctz(ARM_HWCAP_A64_DCPOP )] = "dcpop", 800 [__builtin_ctz(ARM_HWCAP_A64_SHA3 )] = "sha3", 801 [__builtin_ctz(ARM_HWCAP_A64_SM3 )] = "sm3", 802 [__builtin_ctz(ARM_HWCAP_A64_SM4 )] = "sm4", 803 [__builtin_ctz(ARM_HWCAP_A64_ASIMDDP )] = "asimddp", 804 [__builtin_ctz(ARM_HWCAP_A64_SHA512 )] = "sha512", 805 [__builtin_ctz(ARM_HWCAP_A64_SVE )] = "sve", 806 [__builtin_ctz(ARM_HWCAP_A64_ASIMDFHM)] = "asimdfhm", 807 [__builtin_ctz(ARM_HWCAP_A64_DIT )] = "dit", 808 [__builtin_ctz(ARM_HWCAP_A64_USCAT )] = "uscat", 809 [__builtin_ctz(ARM_HWCAP_A64_ILRCPC )] = "ilrcpc", 810 [__builtin_ctz(ARM_HWCAP_A64_FLAGM )] = "flagm", 811 [__builtin_ctz(ARM_HWCAP_A64_SSBS )] = "ssbs", 812 [__builtin_ctz(ARM_HWCAP_A64_SB )] = "sb", 813 [__builtin_ctz(ARM_HWCAP_A64_PACA )] = "paca", 814 [__builtin_ctz(ARM_HWCAP_A64_PACG )] = "pacg", 815 }; 816 817 return bit < ARRAY_SIZE(hwcap_str) ? hwcap_str[bit] : NULL; 818 } 819 820 const char *elf_hwcap2_str(uint32_t bit) 821 { 822 static const char *hwcap_str[] = { 823 [__builtin_ctz(ARM_HWCAP2_A64_DCPODP )] = "dcpodp", 824 [__builtin_ctz(ARM_HWCAP2_A64_SVE2 )] = "sve2", 825 [__builtin_ctz(ARM_HWCAP2_A64_SVEAES )] = "sveaes", 826 [__builtin_ctz(ARM_HWCAP2_A64_SVEPMULL )] = "svepmull", 827 [__builtin_ctz(ARM_HWCAP2_A64_SVEBITPERM )] = "svebitperm", 828 [__builtin_ctz(ARM_HWCAP2_A64_SVESHA3 )] = "svesha3", 829 [__builtin_ctz(ARM_HWCAP2_A64_SVESM4 )] = "svesm4", 830 [__builtin_ctz(ARM_HWCAP2_A64_FLAGM2 )] = "flagm2", 831 [__builtin_ctz(ARM_HWCAP2_A64_FRINT )] = "frint", 832 [__builtin_ctz(ARM_HWCAP2_A64_SVEI8MM )] = "svei8mm", 833 [__builtin_ctz(ARM_HWCAP2_A64_SVEF32MM )] = "svef32mm", 834 [__builtin_ctz(ARM_HWCAP2_A64_SVEF64MM )] = "svef64mm", 835 [__builtin_ctz(ARM_HWCAP2_A64_SVEBF16 )] = "svebf16", 836 [__builtin_ctz(ARM_HWCAP2_A64_I8MM )] = "i8mm", 837 [__builtin_ctz(ARM_HWCAP2_A64_BF16 )] = "bf16", 838 [__builtin_ctz(ARM_HWCAP2_A64_DGH )] = "dgh", 839 [__builtin_ctz(ARM_HWCAP2_A64_RNG )] = "rng", 840 [__builtin_ctz(ARM_HWCAP2_A64_BTI )] = "bti", 841 [__builtin_ctz(ARM_HWCAP2_A64_MTE )] = "mte", 842 [__builtin_ctz(ARM_HWCAP2_A64_ECV )] = "ecv", 843 [__builtin_ctz(ARM_HWCAP2_A64_AFP )] = "afp", 844 [__builtin_ctz(ARM_HWCAP2_A64_RPRES )] = "rpres", 845 [__builtin_ctz(ARM_HWCAP2_A64_MTE3 )] = "mte3", 846 [__builtin_ctz(ARM_HWCAP2_A64_SME )] = "sme", 847 [__builtin_ctz(ARM_HWCAP2_A64_SME_I16I64 )] = "sme_i16i64", 848 [__builtin_ctz(ARM_HWCAP2_A64_SME_F64F64 )] = "sme_f64f64", 849 [__builtin_ctz(ARM_HWCAP2_A64_SME_I8I32 )] = "sme_i8i32", 850 [__builtin_ctz(ARM_HWCAP2_A64_SME_F16F32 )] = "sme_f16f32", 851 [__builtin_ctz(ARM_HWCAP2_A64_SME_B16F32 )] = "sme_b16f32", 852 [__builtin_ctz(ARM_HWCAP2_A64_SME_F32F32 )] = "sme_f32f32", 853 [__builtin_ctz(ARM_HWCAP2_A64_SME_FA64 )] = "sme_fa64", 854 }; 855 856 return bit < ARRAY_SIZE(hwcap_str) ? hwcap_str[bit] : NULL; 857 } 858 859 #undef GET_FEATURE_ID 860 861 #endif /* not TARGET_AARCH64 */ 862 #endif /* TARGET_ARM */ 863 864 #ifdef TARGET_SPARC 865 #ifdef TARGET_SPARC64 866 867 #define ELF_HWCAP (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \ 868 | HWCAP_SPARC_MULDIV | HWCAP_SPARC_V9) 869 #ifndef TARGET_ABI32 870 #define elf_check_arch(x) ( (x) == EM_SPARCV9 || (x) == EM_SPARC32PLUS ) 871 #else 872 #define elf_check_arch(x) ( (x) == EM_SPARC32PLUS || (x) == EM_SPARC ) 873 #endif 874 875 #define ELF_CLASS ELFCLASS64 876 #define ELF_ARCH EM_SPARCV9 877 #else 878 #define ELF_HWCAP (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \ 879 | HWCAP_SPARC_MULDIV) 880 #define ELF_CLASS ELFCLASS32 881 #define ELF_ARCH EM_SPARC 882 #endif /* TARGET_SPARC64 */ 883 884 static inline void init_thread(struct target_pt_regs *regs, 885 struct image_info *infop) 886 { 887 /* Note that target_cpu_copy_regs does not read psr/tstate. */ 888 regs->pc = infop->entry; 889 regs->npc = regs->pc + 4; 890 regs->y = 0; 891 regs->u_regs[14] = (infop->start_stack - 16 * sizeof(abi_ulong) 892 - TARGET_STACK_BIAS); 893 } 894 #endif /* TARGET_SPARC */ 895 896 #ifdef TARGET_PPC 897 898 #define ELF_MACHINE PPC_ELF_MACHINE 899 900 #if defined(TARGET_PPC64) 901 902 #define elf_check_arch(x) ( (x) == EM_PPC64 ) 903 904 #define ELF_CLASS ELFCLASS64 905 906 #else 907 908 #define ELF_CLASS ELFCLASS32 909 #define EXSTACK_DEFAULT true 910 911 #endif 912 913 #define ELF_ARCH EM_PPC 914 915 /* Feature masks for the Aux Vector Hardware Capabilities (AT_HWCAP). 916 See arch/powerpc/include/asm/cputable.h. */ 917 enum { 918 QEMU_PPC_FEATURE_32 = 0x80000000, 919 QEMU_PPC_FEATURE_64 = 0x40000000, 920 QEMU_PPC_FEATURE_601_INSTR = 0x20000000, 921 QEMU_PPC_FEATURE_HAS_ALTIVEC = 0x10000000, 922 QEMU_PPC_FEATURE_HAS_FPU = 0x08000000, 923 QEMU_PPC_FEATURE_HAS_MMU = 0x04000000, 924 QEMU_PPC_FEATURE_HAS_4xxMAC = 0x02000000, 925 QEMU_PPC_FEATURE_UNIFIED_CACHE = 0x01000000, 926 QEMU_PPC_FEATURE_HAS_SPE = 0x00800000, 927 QEMU_PPC_FEATURE_HAS_EFP_SINGLE = 0x00400000, 928 QEMU_PPC_FEATURE_HAS_EFP_DOUBLE = 0x00200000, 929 QEMU_PPC_FEATURE_NO_TB = 0x00100000, 930 QEMU_PPC_FEATURE_POWER4 = 0x00080000, 931 QEMU_PPC_FEATURE_POWER5 = 0x00040000, 932 QEMU_PPC_FEATURE_POWER5_PLUS = 0x00020000, 933 QEMU_PPC_FEATURE_CELL = 0x00010000, 934 QEMU_PPC_FEATURE_BOOKE = 0x00008000, 935 QEMU_PPC_FEATURE_SMT = 0x00004000, 936 QEMU_PPC_FEATURE_ICACHE_SNOOP = 0x00002000, 937 QEMU_PPC_FEATURE_ARCH_2_05 = 0x00001000, 938 QEMU_PPC_FEATURE_PA6T = 0x00000800, 939 QEMU_PPC_FEATURE_HAS_DFP = 0x00000400, 940 QEMU_PPC_FEATURE_POWER6_EXT = 0x00000200, 941 QEMU_PPC_FEATURE_ARCH_2_06 = 0x00000100, 942 QEMU_PPC_FEATURE_HAS_VSX = 0x00000080, 943 QEMU_PPC_FEATURE_PSERIES_PERFMON_COMPAT = 0x00000040, 944 945 QEMU_PPC_FEATURE_TRUE_LE = 0x00000002, 946 QEMU_PPC_FEATURE_PPC_LE = 0x00000001, 947 948 /* Feature definitions in AT_HWCAP2. */ 949 QEMU_PPC_FEATURE2_ARCH_2_07 = 0x80000000, /* ISA 2.07 */ 950 QEMU_PPC_FEATURE2_HAS_HTM = 0x40000000, /* Hardware Transactional Memory */ 951 QEMU_PPC_FEATURE2_HAS_DSCR = 0x20000000, /* Data Stream Control Register */ 952 QEMU_PPC_FEATURE2_HAS_EBB = 0x10000000, /* Event Base Branching */ 953 QEMU_PPC_FEATURE2_HAS_ISEL = 0x08000000, /* Integer Select */ 954 QEMU_PPC_FEATURE2_HAS_TAR = 0x04000000, /* Target Address Register */ 955 QEMU_PPC_FEATURE2_VEC_CRYPTO = 0x02000000, 956 QEMU_PPC_FEATURE2_HTM_NOSC = 0x01000000, 957 QEMU_PPC_FEATURE2_ARCH_3_00 = 0x00800000, /* ISA 3.00 */ 958 QEMU_PPC_FEATURE2_HAS_IEEE128 = 0x00400000, /* VSX IEEE Bin Float 128-bit */ 959 QEMU_PPC_FEATURE2_DARN = 0x00200000, /* darn random number insn */ 960 QEMU_PPC_FEATURE2_SCV = 0x00100000, /* scv syscall */ 961 QEMU_PPC_FEATURE2_HTM_NO_SUSPEND = 0x00080000, /* TM w/o suspended state */ 962 QEMU_PPC_FEATURE2_ARCH_3_1 = 0x00040000, /* ISA 3.1 */ 963 QEMU_PPC_FEATURE2_MMA = 0x00020000, /* Matrix-Multiply Assist */ 964 }; 965 966 #define ELF_HWCAP get_elf_hwcap() 967 968 static uint32_t get_elf_hwcap(void) 969 { 970 PowerPCCPU *cpu = POWERPC_CPU(thread_cpu); 971 uint32_t features = 0; 972 973 /* We don't have to be terribly complete here; the high points are 974 Altivec/FP/SPE support. Anything else is just a bonus. */ 975 #define GET_FEATURE(flag, feature) \ 976 do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0) 977 #define GET_FEATURE2(flags, feature) \ 978 do { \ 979 if ((cpu->env.insns_flags2 & flags) == flags) { \ 980 features |= feature; \ 981 } \ 982 } while (0) 983 GET_FEATURE(PPC_64B, QEMU_PPC_FEATURE_64); 984 GET_FEATURE(PPC_FLOAT, QEMU_PPC_FEATURE_HAS_FPU); 985 GET_FEATURE(PPC_ALTIVEC, QEMU_PPC_FEATURE_HAS_ALTIVEC); 986 GET_FEATURE(PPC_SPE, QEMU_PPC_FEATURE_HAS_SPE); 987 GET_FEATURE(PPC_SPE_SINGLE, QEMU_PPC_FEATURE_HAS_EFP_SINGLE); 988 GET_FEATURE(PPC_SPE_DOUBLE, QEMU_PPC_FEATURE_HAS_EFP_DOUBLE); 989 GET_FEATURE(PPC_BOOKE, QEMU_PPC_FEATURE_BOOKE); 990 GET_FEATURE(PPC_405_MAC, QEMU_PPC_FEATURE_HAS_4xxMAC); 991 GET_FEATURE2(PPC2_DFP, QEMU_PPC_FEATURE_HAS_DFP); 992 GET_FEATURE2(PPC2_VSX, QEMU_PPC_FEATURE_HAS_VSX); 993 GET_FEATURE2((PPC2_PERM_ISA206 | PPC2_DIVE_ISA206 | PPC2_ATOMIC_ISA206 | 994 PPC2_FP_CVT_ISA206 | PPC2_FP_TST_ISA206), 995 QEMU_PPC_FEATURE_ARCH_2_06); 996 #undef GET_FEATURE 997 #undef GET_FEATURE2 998 999 return features; 1000 } 1001 1002 #define ELF_HWCAP2 get_elf_hwcap2() 1003 1004 static uint32_t get_elf_hwcap2(void) 1005 { 1006 PowerPCCPU *cpu = POWERPC_CPU(thread_cpu); 1007 uint32_t features = 0; 1008 1009 #define GET_FEATURE(flag, feature) \ 1010 do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0) 1011 #define GET_FEATURE2(flag, feature) \ 1012 do { if (cpu->env.insns_flags2 & flag) { features |= feature; } } while (0) 1013 1014 GET_FEATURE(PPC_ISEL, QEMU_PPC_FEATURE2_HAS_ISEL); 1015 GET_FEATURE2(PPC2_BCTAR_ISA207, QEMU_PPC_FEATURE2_HAS_TAR); 1016 GET_FEATURE2((PPC2_BCTAR_ISA207 | PPC2_LSQ_ISA207 | PPC2_ALTIVEC_207 | 1017 PPC2_ISA207S), QEMU_PPC_FEATURE2_ARCH_2_07 | 1018 QEMU_PPC_FEATURE2_VEC_CRYPTO); 1019 GET_FEATURE2(PPC2_ISA300, QEMU_PPC_FEATURE2_ARCH_3_00 | 1020 QEMU_PPC_FEATURE2_DARN | QEMU_PPC_FEATURE2_HAS_IEEE128); 1021 GET_FEATURE2(PPC2_ISA310, QEMU_PPC_FEATURE2_ARCH_3_1 | 1022 QEMU_PPC_FEATURE2_MMA); 1023 1024 #undef GET_FEATURE 1025 #undef GET_FEATURE2 1026 1027 return features; 1028 } 1029 1030 /* 1031 * The requirements here are: 1032 * - keep the final alignment of sp (sp & 0xf) 1033 * - make sure the 32-bit value at the first 16 byte aligned position of 1034 * AUXV is greater than 16 for glibc compatibility. 1035 * AT_IGNOREPPC is used for that. 1036 * - for compatibility with glibc ARCH_DLINFO must always be defined on PPC, 1037 * even if DLINFO_ARCH_ITEMS goes to zero or is undefined. 1038 */ 1039 #define DLINFO_ARCH_ITEMS 5 1040 #define ARCH_DLINFO \ 1041 do { \ 1042 PowerPCCPU *cpu = POWERPC_CPU(thread_cpu); \ 1043 /* \ 1044 * Handle glibc compatibility: these magic entries must \ 1045 * be at the lowest addresses in the final auxv. \ 1046 */ \ 1047 NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC); \ 1048 NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC); \ 1049 NEW_AUX_ENT(AT_DCACHEBSIZE, cpu->env.dcache_line_size); \ 1050 NEW_AUX_ENT(AT_ICACHEBSIZE, cpu->env.icache_line_size); \ 1051 NEW_AUX_ENT(AT_UCACHEBSIZE, 0); \ 1052 } while (0) 1053 1054 static inline void init_thread(struct target_pt_regs *_regs, struct image_info *infop) 1055 { 1056 _regs->gpr[1] = infop->start_stack; 1057 #if defined(TARGET_PPC64) 1058 if (get_ppc64_abi(infop) < 2) { 1059 uint64_t val; 1060 get_user_u64(val, infop->entry + 8); 1061 _regs->gpr[2] = val + infop->load_bias; 1062 get_user_u64(val, infop->entry); 1063 infop->entry = val + infop->load_bias; 1064 } else { 1065 _regs->gpr[12] = infop->entry; /* r12 set to global entry address */ 1066 } 1067 #endif 1068 _regs->nip = infop->entry; 1069 } 1070 1071 /* See linux kernel: arch/powerpc/include/asm/elf.h. */ 1072 #define ELF_NREG 48 1073 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1074 1075 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUPPCState *env) 1076 { 1077 int i; 1078 target_ulong ccr = 0; 1079 1080 for (i = 0; i < ARRAY_SIZE(env->gpr); i++) { 1081 (*regs)[i] = tswapreg(env->gpr[i]); 1082 } 1083 1084 (*regs)[32] = tswapreg(env->nip); 1085 (*regs)[33] = tswapreg(env->msr); 1086 (*regs)[35] = tswapreg(env->ctr); 1087 (*regs)[36] = tswapreg(env->lr); 1088 (*regs)[37] = tswapreg(cpu_read_xer(env)); 1089 1090 ccr = ppc_get_cr(env); 1091 (*regs)[38] = tswapreg(ccr); 1092 } 1093 1094 #define USE_ELF_CORE_DUMP 1095 #define ELF_EXEC_PAGESIZE 4096 1096 1097 #endif 1098 1099 #ifdef TARGET_LOONGARCH64 1100 1101 #define ELF_CLASS ELFCLASS64 1102 #define ELF_ARCH EM_LOONGARCH 1103 #define EXSTACK_DEFAULT true 1104 1105 #define elf_check_arch(x) ((x) == EM_LOONGARCH) 1106 1107 static inline void init_thread(struct target_pt_regs *regs, 1108 struct image_info *infop) 1109 { 1110 /*Set crmd PG,DA = 1,0 */ 1111 regs->csr.crmd = 2 << 3; 1112 regs->csr.era = infop->entry; 1113 regs->regs[3] = infop->start_stack; 1114 } 1115 1116 /* See linux kernel: arch/loongarch/include/asm/elf.h */ 1117 #define ELF_NREG 45 1118 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1119 1120 enum { 1121 TARGET_EF_R0 = 0, 1122 TARGET_EF_CSR_ERA = TARGET_EF_R0 + 33, 1123 TARGET_EF_CSR_BADV = TARGET_EF_R0 + 34, 1124 }; 1125 1126 static void elf_core_copy_regs(target_elf_gregset_t *regs, 1127 const CPULoongArchState *env) 1128 { 1129 int i; 1130 1131 (*regs)[TARGET_EF_R0] = 0; 1132 1133 for (i = 1; i < ARRAY_SIZE(env->gpr); i++) { 1134 (*regs)[TARGET_EF_R0 + i] = tswapreg(env->gpr[i]); 1135 } 1136 1137 (*regs)[TARGET_EF_CSR_ERA] = tswapreg(env->pc); 1138 (*regs)[TARGET_EF_CSR_BADV] = tswapreg(env->CSR_BADV); 1139 } 1140 1141 #define USE_ELF_CORE_DUMP 1142 #define ELF_EXEC_PAGESIZE 4096 1143 1144 #define ELF_HWCAP get_elf_hwcap() 1145 1146 /* See arch/loongarch/include/uapi/asm/hwcap.h */ 1147 enum { 1148 HWCAP_LOONGARCH_CPUCFG = (1 << 0), 1149 HWCAP_LOONGARCH_LAM = (1 << 1), 1150 HWCAP_LOONGARCH_UAL = (1 << 2), 1151 HWCAP_LOONGARCH_FPU = (1 << 3), 1152 HWCAP_LOONGARCH_LSX = (1 << 4), 1153 HWCAP_LOONGARCH_LASX = (1 << 5), 1154 HWCAP_LOONGARCH_CRC32 = (1 << 6), 1155 HWCAP_LOONGARCH_COMPLEX = (1 << 7), 1156 HWCAP_LOONGARCH_CRYPTO = (1 << 8), 1157 HWCAP_LOONGARCH_LVZ = (1 << 9), 1158 HWCAP_LOONGARCH_LBT_X86 = (1 << 10), 1159 HWCAP_LOONGARCH_LBT_ARM = (1 << 11), 1160 HWCAP_LOONGARCH_LBT_MIPS = (1 << 12), 1161 }; 1162 1163 static uint32_t get_elf_hwcap(void) 1164 { 1165 LoongArchCPU *cpu = LOONGARCH_CPU(thread_cpu); 1166 uint32_t hwcaps = 0; 1167 1168 hwcaps |= HWCAP_LOONGARCH_CRC32; 1169 1170 if (FIELD_EX32(cpu->env.cpucfg[1], CPUCFG1, UAL)) { 1171 hwcaps |= HWCAP_LOONGARCH_UAL; 1172 } 1173 1174 if (FIELD_EX32(cpu->env.cpucfg[2], CPUCFG2, FP)) { 1175 hwcaps |= HWCAP_LOONGARCH_FPU; 1176 } 1177 1178 if (FIELD_EX32(cpu->env.cpucfg[2], CPUCFG2, LAM)) { 1179 hwcaps |= HWCAP_LOONGARCH_LAM; 1180 } 1181 1182 return hwcaps; 1183 } 1184 1185 #define ELF_PLATFORM "loongarch" 1186 1187 #endif /* TARGET_LOONGARCH64 */ 1188 1189 #ifdef TARGET_MIPS 1190 1191 #ifdef TARGET_MIPS64 1192 #define ELF_CLASS ELFCLASS64 1193 #else 1194 #define ELF_CLASS ELFCLASS32 1195 #endif 1196 #define ELF_ARCH EM_MIPS 1197 #define EXSTACK_DEFAULT true 1198 1199 #ifdef TARGET_ABI_MIPSN32 1200 #define elf_check_abi(x) ((x) & EF_MIPS_ABI2) 1201 #else 1202 #define elf_check_abi(x) (!((x) & EF_MIPS_ABI2)) 1203 #endif 1204 1205 #define ELF_BASE_PLATFORM get_elf_base_platform() 1206 1207 #define MATCH_PLATFORM_INSN(_flags, _base_platform) \ 1208 do { if ((cpu->env.insn_flags & (_flags)) == _flags) \ 1209 { return _base_platform; } } while (0) 1210 1211 static const char *get_elf_base_platform(void) 1212 { 1213 MIPSCPU *cpu = MIPS_CPU(thread_cpu); 1214 1215 /* 64 bit ISAs goes first */ 1216 MATCH_PLATFORM_INSN(CPU_MIPS64R6, "mips64r6"); 1217 MATCH_PLATFORM_INSN(CPU_MIPS64R5, "mips64r5"); 1218 MATCH_PLATFORM_INSN(CPU_MIPS64R2, "mips64r2"); 1219 MATCH_PLATFORM_INSN(CPU_MIPS64R1, "mips64"); 1220 MATCH_PLATFORM_INSN(CPU_MIPS5, "mips5"); 1221 MATCH_PLATFORM_INSN(CPU_MIPS4, "mips4"); 1222 MATCH_PLATFORM_INSN(CPU_MIPS3, "mips3"); 1223 1224 /* 32 bit ISAs */ 1225 MATCH_PLATFORM_INSN(CPU_MIPS32R6, "mips32r6"); 1226 MATCH_PLATFORM_INSN(CPU_MIPS32R5, "mips32r5"); 1227 MATCH_PLATFORM_INSN(CPU_MIPS32R2, "mips32r2"); 1228 MATCH_PLATFORM_INSN(CPU_MIPS32R1, "mips32"); 1229 MATCH_PLATFORM_INSN(CPU_MIPS2, "mips2"); 1230 1231 /* Fallback */ 1232 return "mips"; 1233 } 1234 #undef MATCH_PLATFORM_INSN 1235 1236 static inline void init_thread(struct target_pt_regs *regs, 1237 struct image_info *infop) 1238 { 1239 regs->cp0_status = 2 << CP0St_KSU; 1240 regs->cp0_epc = infop->entry; 1241 regs->regs[29] = infop->start_stack; 1242 } 1243 1244 /* See linux kernel: arch/mips/include/asm/elf.h. */ 1245 #define ELF_NREG 45 1246 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1247 1248 /* See linux kernel: arch/mips/include/asm/reg.h. */ 1249 enum { 1250 #ifdef TARGET_MIPS64 1251 TARGET_EF_R0 = 0, 1252 #else 1253 TARGET_EF_R0 = 6, 1254 #endif 1255 TARGET_EF_R26 = TARGET_EF_R0 + 26, 1256 TARGET_EF_R27 = TARGET_EF_R0 + 27, 1257 TARGET_EF_LO = TARGET_EF_R0 + 32, 1258 TARGET_EF_HI = TARGET_EF_R0 + 33, 1259 TARGET_EF_CP0_EPC = TARGET_EF_R0 + 34, 1260 TARGET_EF_CP0_BADVADDR = TARGET_EF_R0 + 35, 1261 TARGET_EF_CP0_STATUS = TARGET_EF_R0 + 36, 1262 TARGET_EF_CP0_CAUSE = TARGET_EF_R0 + 37 1263 }; 1264 1265 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */ 1266 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMIPSState *env) 1267 { 1268 int i; 1269 1270 for (i = 0; i < TARGET_EF_R0; i++) { 1271 (*regs)[i] = 0; 1272 } 1273 (*regs)[TARGET_EF_R0] = 0; 1274 1275 for (i = 1; i < ARRAY_SIZE(env->active_tc.gpr); i++) { 1276 (*regs)[TARGET_EF_R0 + i] = tswapreg(env->active_tc.gpr[i]); 1277 } 1278 1279 (*regs)[TARGET_EF_R26] = 0; 1280 (*regs)[TARGET_EF_R27] = 0; 1281 (*regs)[TARGET_EF_LO] = tswapreg(env->active_tc.LO[0]); 1282 (*regs)[TARGET_EF_HI] = tswapreg(env->active_tc.HI[0]); 1283 (*regs)[TARGET_EF_CP0_EPC] = tswapreg(env->active_tc.PC); 1284 (*regs)[TARGET_EF_CP0_BADVADDR] = tswapreg(env->CP0_BadVAddr); 1285 (*regs)[TARGET_EF_CP0_STATUS] = tswapreg(env->CP0_Status); 1286 (*regs)[TARGET_EF_CP0_CAUSE] = tswapreg(env->CP0_Cause); 1287 } 1288 1289 #define USE_ELF_CORE_DUMP 1290 #define ELF_EXEC_PAGESIZE 4096 1291 1292 /* See arch/mips/include/uapi/asm/hwcap.h. */ 1293 enum { 1294 HWCAP_MIPS_R6 = (1 << 0), 1295 HWCAP_MIPS_MSA = (1 << 1), 1296 HWCAP_MIPS_CRC32 = (1 << 2), 1297 HWCAP_MIPS_MIPS16 = (1 << 3), 1298 HWCAP_MIPS_MDMX = (1 << 4), 1299 HWCAP_MIPS_MIPS3D = (1 << 5), 1300 HWCAP_MIPS_SMARTMIPS = (1 << 6), 1301 HWCAP_MIPS_DSP = (1 << 7), 1302 HWCAP_MIPS_DSP2 = (1 << 8), 1303 HWCAP_MIPS_DSP3 = (1 << 9), 1304 HWCAP_MIPS_MIPS16E2 = (1 << 10), 1305 HWCAP_LOONGSON_MMI = (1 << 11), 1306 HWCAP_LOONGSON_EXT = (1 << 12), 1307 HWCAP_LOONGSON_EXT2 = (1 << 13), 1308 HWCAP_LOONGSON_CPUCFG = (1 << 14), 1309 }; 1310 1311 #define ELF_HWCAP get_elf_hwcap() 1312 1313 #define GET_FEATURE_INSN(_flag, _hwcap) \ 1314 do { if (cpu->env.insn_flags & (_flag)) { hwcaps |= _hwcap; } } while (0) 1315 1316 #define GET_FEATURE_REG_SET(_reg, _mask, _hwcap) \ 1317 do { if (cpu->env._reg & (_mask)) { hwcaps |= _hwcap; } } while (0) 1318 1319 #define GET_FEATURE_REG_EQU(_reg, _start, _length, _val, _hwcap) \ 1320 do { \ 1321 if (extract32(cpu->env._reg, (_start), (_length)) == (_val)) { \ 1322 hwcaps |= _hwcap; \ 1323 } \ 1324 } while (0) 1325 1326 static uint32_t get_elf_hwcap(void) 1327 { 1328 MIPSCPU *cpu = MIPS_CPU(thread_cpu); 1329 uint32_t hwcaps = 0; 1330 1331 GET_FEATURE_REG_EQU(CP0_Config0, CP0C0_AR, CP0C0_AR_LENGTH, 1332 2, HWCAP_MIPS_R6); 1333 GET_FEATURE_REG_SET(CP0_Config3, 1 << CP0C3_MSAP, HWCAP_MIPS_MSA); 1334 GET_FEATURE_INSN(ASE_LMMI, HWCAP_LOONGSON_MMI); 1335 GET_FEATURE_INSN(ASE_LEXT, HWCAP_LOONGSON_EXT); 1336 1337 return hwcaps; 1338 } 1339 1340 #undef GET_FEATURE_REG_EQU 1341 #undef GET_FEATURE_REG_SET 1342 #undef GET_FEATURE_INSN 1343 1344 #endif /* TARGET_MIPS */ 1345 1346 #ifdef TARGET_MICROBLAZE 1347 1348 #define elf_check_arch(x) ( (x) == EM_MICROBLAZE || (x) == EM_MICROBLAZE_OLD) 1349 1350 #define ELF_CLASS ELFCLASS32 1351 #define ELF_ARCH EM_MICROBLAZE 1352 1353 static inline void init_thread(struct target_pt_regs *regs, 1354 struct image_info *infop) 1355 { 1356 regs->pc = infop->entry; 1357 regs->r1 = infop->start_stack; 1358 1359 } 1360 1361 #define ELF_EXEC_PAGESIZE 4096 1362 1363 #define USE_ELF_CORE_DUMP 1364 #define ELF_NREG 38 1365 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1366 1367 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */ 1368 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMBState *env) 1369 { 1370 int i, pos = 0; 1371 1372 for (i = 0; i < 32; i++) { 1373 (*regs)[pos++] = tswapreg(env->regs[i]); 1374 } 1375 1376 (*regs)[pos++] = tswapreg(env->pc); 1377 (*regs)[pos++] = tswapreg(mb_cpu_read_msr(env)); 1378 (*regs)[pos++] = 0; 1379 (*regs)[pos++] = tswapreg(env->ear); 1380 (*regs)[pos++] = 0; 1381 (*regs)[pos++] = tswapreg(env->esr); 1382 } 1383 1384 #endif /* TARGET_MICROBLAZE */ 1385 1386 #ifdef TARGET_NIOS2 1387 1388 #define elf_check_arch(x) ((x) == EM_ALTERA_NIOS2) 1389 1390 #define ELF_CLASS ELFCLASS32 1391 #define ELF_ARCH EM_ALTERA_NIOS2 1392 1393 static void init_thread(struct target_pt_regs *regs, struct image_info *infop) 1394 { 1395 regs->ea = infop->entry; 1396 regs->sp = infop->start_stack; 1397 } 1398 1399 #define LO_COMMPAGE TARGET_PAGE_SIZE 1400 1401 static bool init_guest_commpage(void) 1402 { 1403 static const uint8_t kuser_page[4 + 2 * 64] = { 1404 /* __kuser_helper_version */ 1405 [0x00] = 0x02, 0x00, 0x00, 0x00, 1406 1407 /* __kuser_cmpxchg */ 1408 [0x04] = 0x3a, 0x6c, 0x3b, 0x00, /* trap 16 */ 1409 0x3a, 0x28, 0x00, 0xf8, /* ret */ 1410 1411 /* __kuser_sigtramp */ 1412 [0x44] = 0xc4, 0x22, 0x80, 0x00, /* movi r2, __NR_rt_sigreturn */ 1413 0x3a, 0x68, 0x3b, 0x00, /* trap 0 */ 1414 }; 1415 1416 void *want = g2h_untagged(LO_COMMPAGE & -qemu_host_page_size); 1417 void *addr = mmap(want, qemu_host_page_size, PROT_READ | PROT_WRITE, 1418 MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0); 1419 1420 if (addr == MAP_FAILED) { 1421 perror("Allocating guest commpage"); 1422 exit(EXIT_FAILURE); 1423 } 1424 if (addr != want) { 1425 return false; 1426 } 1427 1428 memcpy(addr, kuser_page, sizeof(kuser_page)); 1429 1430 if (mprotect(addr, qemu_host_page_size, PROT_READ)) { 1431 perror("Protecting guest commpage"); 1432 exit(EXIT_FAILURE); 1433 } 1434 1435 page_set_flags(LO_COMMPAGE, LO_COMMPAGE | ~TARGET_PAGE_MASK, 1436 PAGE_READ | PAGE_EXEC | PAGE_VALID); 1437 return true; 1438 } 1439 1440 #define ELF_EXEC_PAGESIZE 4096 1441 1442 #define USE_ELF_CORE_DUMP 1443 #define ELF_NREG 49 1444 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1445 1446 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */ 1447 static void elf_core_copy_regs(target_elf_gregset_t *regs, 1448 const CPUNios2State *env) 1449 { 1450 int i; 1451 1452 (*regs)[0] = -1; 1453 for (i = 1; i < 8; i++) /* r0-r7 */ 1454 (*regs)[i] = tswapreg(env->regs[i + 7]); 1455 1456 for (i = 8; i < 16; i++) /* r8-r15 */ 1457 (*regs)[i] = tswapreg(env->regs[i - 8]); 1458 1459 for (i = 16; i < 24; i++) /* r16-r23 */ 1460 (*regs)[i] = tswapreg(env->regs[i + 7]); 1461 (*regs)[24] = -1; /* R_ET */ 1462 (*regs)[25] = -1; /* R_BT */ 1463 (*regs)[26] = tswapreg(env->regs[R_GP]); 1464 (*regs)[27] = tswapreg(env->regs[R_SP]); 1465 (*regs)[28] = tswapreg(env->regs[R_FP]); 1466 (*regs)[29] = tswapreg(env->regs[R_EA]); 1467 (*regs)[30] = -1; /* R_SSTATUS */ 1468 (*regs)[31] = tswapreg(env->regs[R_RA]); 1469 1470 (*regs)[32] = tswapreg(env->pc); 1471 1472 (*regs)[33] = -1; /* R_STATUS */ 1473 (*regs)[34] = tswapreg(env->regs[CR_ESTATUS]); 1474 1475 for (i = 35; i < 49; i++) /* ... */ 1476 (*regs)[i] = -1; 1477 } 1478 1479 #endif /* TARGET_NIOS2 */ 1480 1481 #ifdef TARGET_OPENRISC 1482 1483 #define ELF_ARCH EM_OPENRISC 1484 #define ELF_CLASS ELFCLASS32 1485 #define ELF_DATA ELFDATA2MSB 1486 1487 static inline void init_thread(struct target_pt_regs *regs, 1488 struct image_info *infop) 1489 { 1490 regs->pc = infop->entry; 1491 regs->gpr[1] = infop->start_stack; 1492 } 1493 1494 #define USE_ELF_CORE_DUMP 1495 #define ELF_EXEC_PAGESIZE 8192 1496 1497 /* See linux kernel arch/openrisc/include/asm/elf.h. */ 1498 #define ELF_NREG 34 /* gprs and pc, sr */ 1499 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1500 1501 static void elf_core_copy_regs(target_elf_gregset_t *regs, 1502 const CPUOpenRISCState *env) 1503 { 1504 int i; 1505 1506 for (i = 0; i < 32; i++) { 1507 (*regs)[i] = tswapreg(cpu_get_gpr(env, i)); 1508 } 1509 (*regs)[32] = tswapreg(env->pc); 1510 (*regs)[33] = tswapreg(cpu_get_sr(env)); 1511 } 1512 #define ELF_HWCAP 0 1513 #define ELF_PLATFORM NULL 1514 1515 #endif /* TARGET_OPENRISC */ 1516 1517 #ifdef TARGET_SH4 1518 1519 #define ELF_CLASS ELFCLASS32 1520 #define ELF_ARCH EM_SH 1521 1522 static inline void init_thread(struct target_pt_regs *regs, 1523 struct image_info *infop) 1524 { 1525 /* Check other registers XXXXX */ 1526 regs->pc = infop->entry; 1527 regs->regs[15] = infop->start_stack; 1528 } 1529 1530 /* See linux kernel: arch/sh/include/asm/elf.h. */ 1531 #define ELF_NREG 23 1532 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1533 1534 /* See linux kernel: arch/sh/include/asm/ptrace.h. */ 1535 enum { 1536 TARGET_REG_PC = 16, 1537 TARGET_REG_PR = 17, 1538 TARGET_REG_SR = 18, 1539 TARGET_REG_GBR = 19, 1540 TARGET_REG_MACH = 20, 1541 TARGET_REG_MACL = 21, 1542 TARGET_REG_SYSCALL = 22 1543 }; 1544 1545 static inline void elf_core_copy_regs(target_elf_gregset_t *regs, 1546 const CPUSH4State *env) 1547 { 1548 int i; 1549 1550 for (i = 0; i < 16; i++) { 1551 (*regs)[i] = tswapreg(env->gregs[i]); 1552 } 1553 1554 (*regs)[TARGET_REG_PC] = tswapreg(env->pc); 1555 (*regs)[TARGET_REG_PR] = tswapreg(env->pr); 1556 (*regs)[TARGET_REG_SR] = tswapreg(env->sr); 1557 (*regs)[TARGET_REG_GBR] = tswapreg(env->gbr); 1558 (*regs)[TARGET_REG_MACH] = tswapreg(env->mach); 1559 (*regs)[TARGET_REG_MACL] = tswapreg(env->macl); 1560 (*regs)[TARGET_REG_SYSCALL] = 0; /* FIXME */ 1561 } 1562 1563 #define USE_ELF_CORE_DUMP 1564 #define ELF_EXEC_PAGESIZE 4096 1565 1566 enum { 1567 SH_CPU_HAS_FPU = 0x0001, /* Hardware FPU support */ 1568 SH_CPU_HAS_P2_FLUSH_BUG = 0x0002, /* Need to flush the cache in P2 area */ 1569 SH_CPU_HAS_MMU_PAGE_ASSOC = 0x0004, /* SH3: TLB way selection bit support */ 1570 SH_CPU_HAS_DSP = 0x0008, /* SH-DSP: DSP support */ 1571 SH_CPU_HAS_PERF_COUNTER = 0x0010, /* Hardware performance counters */ 1572 SH_CPU_HAS_PTEA = 0x0020, /* PTEA register */ 1573 SH_CPU_HAS_LLSC = 0x0040, /* movli.l/movco.l */ 1574 SH_CPU_HAS_L2_CACHE = 0x0080, /* Secondary cache / URAM */ 1575 SH_CPU_HAS_OP32 = 0x0100, /* 32-bit instruction support */ 1576 SH_CPU_HAS_PTEAEX = 0x0200, /* PTE ASID Extension support */ 1577 }; 1578 1579 #define ELF_HWCAP get_elf_hwcap() 1580 1581 static uint32_t get_elf_hwcap(void) 1582 { 1583 SuperHCPU *cpu = SUPERH_CPU(thread_cpu); 1584 uint32_t hwcap = 0; 1585 1586 hwcap |= SH_CPU_HAS_FPU; 1587 1588 if (cpu->env.features & SH_FEATURE_SH4A) { 1589 hwcap |= SH_CPU_HAS_LLSC; 1590 } 1591 1592 return hwcap; 1593 } 1594 1595 #endif 1596 1597 #ifdef TARGET_CRIS 1598 1599 #define ELF_CLASS ELFCLASS32 1600 #define ELF_ARCH EM_CRIS 1601 1602 static inline void init_thread(struct target_pt_regs *regs, 1603 struct image_info *infop) 1604 { 1605 regs->erp = infop->entry; 1606 } 1607 1608 #define ELF_EXEC_PAGESIZE 8192 1609 1610 #endif 1611 1612 #ifdef TARGET_M68K 1613 1614 #define ELF_CLASS ELFCLASS32 1615 #define ELF_ARCH EM_68K 1616 1617 /* ??? Does this need to do anything? 1618 #define ELF_PLAT_INIT(_r) */ 1619 1620 static inline void init_thread(struct target_pt_regs *regs, 1621 struct image_info *infop) 1622 { 1623 regs->usp = infop->start_stack; 1624 regs->sr = 0; 1625 regs->pc = infop->entry; 1626 } 1627 1628 /* See linux kernel: arch/m68k/include/asm/elf.h. */ 1629 #define ELF_NREG 20 1630 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1631 1632 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUM68KState *env) 1633 { 1634 (*regs)[0] = tswapreg(env->dregs[1]); 1635 (*regs)[1] = tswapreg(env->dregs[2]); 1636 (*regs)[2] = tswapreg(env->dregs[3]); 1637 (*regs)[3] = tswapreg(env->dregs[4]); 1638 (*regs)[4] = tswapreg(env->dregs[5]); 1639 (*regs)[5] = tswapreg(env->dregs[6]); 1640 (*regs)[6] = tswapreg(env->dregs[7]); 1641 (*regs)[7] = tswapreg(env->aregs[0]); 1642 (*regs)[8] = tswapreg(env->aregs[1]); 1643 (*regs)[9] = tswapreg(env->aregs[2]); 1644 (*regs)[10] = tswapreg(env->aregs[3]); 1645 (*regs)[11] = tswapreg(env->aregs[4]); 1646 (*regs)[12] = tswapreg(env->aregs[5]); 1647 (*regs)[13] = tswapreg(env->aregs[6]); 1648 (*regs)[14] = tswapreg(env->dregs[0]); 1649 (*regs)[15] = tswapreg(env->aregs[7]); 1650 (*regs)[16] = tswapreg(env->dregs[0]); /* FIXME: orig_d0 */ 1651 (*regs)[17] = tswapreg(env->sr); 1652 (*regs)[18] = tswapreg(env->pc); 1653 (*regs)[19] = 0; /* FIXME: regs->format | regs->vector */ 1654 } 1655 1656 #define USE_ELF_CORE_DUMP 1657 #define ELF_EXEC_PAGESIZE 8192 1658 1659 #endif 1660 1661 #ifdef TARGET_ALPHA 1662 1663 #define ELF_CLASS ELFCLASS64 1664 #define ELF_ARCH EM_ALPHA 1665 1666 static inline void init_thread(struct target_pt_regs *regs, 1667 struct image_info *infop) 1668 { 1669 regs->pc = infop->entry; 1670 regs->ps = 8; 1671 regs->usp = infop->start_stack; 1672 } 1673 1674 #define ELF_EXEC_PAGESIZE 8192 1675 1676 #endif /* TARGET_ALPHA */ 1677 1678 #ifdef TARGET_S390X 1679 1680 #define ELF_CLASS ELFCLASS64 1681 #define ELF_DATA ELFDATA2MSB 1682 #define ELF_ARCH EM_S390 1683 1684 #include "elf.h" 1685 1686 #define ELF_HWCAP get_elf_hwcap() 1687 1688 #define GET_FEATURE(_feat, _hwcap) \ 1689 do { if (s390_has_feat(_feat)) { hwcap |= _hwcap; } } while (0) 1690 1691 uint32_t get_elf_hwcap(void) 1692 { 1693 /* 1694 * Let's assume we always have esan3 and zarch. 1695 * 31-bit processes can use 64-bit registers (high gprs). 1696 */ 1697 uint32_t hwcap = HWCAP_S390_ESAN3 | HWCAP_S390_ZARCH | HWCAP_S390_HIGH_GPRS; 1698 1699 GET_FEATURE(S390_FEAT_STFLE, HWCAP_S390_STFLE); 1700 GET_FEATURE(S390_FEAT_MSA, HWCAP_S390_MSA); 1701 GET_FEATURE(S390_FEAT_LONG_DISPLACEMENT, HWCAP_S390_LDISP); 1702 GET_FEATURE(S390_FEAT_EXTENDED_IMMEDIATE, HWCAP_S390_EIMM); 1703 if (s390_has_feat(S390_FEAT_EXTENDED_TRANSLATION_3) && 1704 s390_has_feat(S390_FEAT_ETF3_ENH)) { 1705 hwcap |= HWCAP_S390_ETF3EH; 1706 } 1707 GET_FEATURE(S390_FEAT_VECTOR, HWCAP_S390_VXRS); 1708 GET_FEATURE(S390_FEAT_VECTOR_ENH, HWCAP_S390_VXRS_EXT); 1709 GET_FEATURE(S390_FEAT_VECTOR_ENH2, HWCAP_S390_VXRS_EXT2); 1710 1711 return hwcap; 1712 } 1713 1714 const char *elf_hwcap_str(uint32_t bit) 1715 { 1716 static const char *hwcap_str[] = { 1717 [HWCAP_S390_NR_ESAN3] = "esan3", 1718 [HWCAP_S390_NR_ZARCH] = "zarch", 1719 [HWCAP_S390_NR_STFLE] = "stfle", 1720 [HWCAP_S390_NR_MSA] = "msa", 1721 [HWCAP_S390_NR_LDISP] = "ldisp", 1722 [HWCAP_S390_NR_EIMM] = "eimm", 1723 [HWCAP_S390_NR_DFP] = "dfp", 1724 [HWCAP_S390_NR_HPAGE] = "edat", 1725 [HWCAP_S390_NR_ETF3EH] = "etf3eh", 1726 [HWCAP_S390_NR_HIGH_GPRS] = "highgprs", 1727 [HWCAP_S390_NR_TE] = "te", 1728 [HWCAP_S390_NR_VXRS] = "vx", 1729 [HWCAP_S390_NR_VXRS_BCD] = "vxd", 1730 [HWCAP_S390_NR_VXRS_EXT] = "vxe", 1731 [HWCAP_S390_NR_GS] = "gs", 1732 [HWCAP_S390_NR_VXRS_EXT2] = "vxe2", 1733 [HWCAP_S390_NR_VXRS_PDE] = "vxp", 1734 [HWCAP_S390_NR_SORT] = "sort", 1735 [HWCAP_S390_NR_DFLT] = "dflt", 1736 [HWCAP_S390_NR_NNPA] = "nnpa", 1737 [HWCAP_S390_NR_PCI_MIO] = "pcimio", 1738 [HWCAP_S390_NR_SIE] = "sie", 1739 }; 1740 1741 return bit < ARRAY_SIZE(hwcap_str) ? hwcap_str[bit] : NULL; 1742 } 1743 1744 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop) 1745 { 1746 regs->psw.addr = infop->entry; 1747 regs->psw.mask = PSW_MASK_DAT | PSW_MASK_IO | PSW_MASK_EXT | \ 1748 PSW_MASK_MCHECK | PSW_MASK_PSTATE | PSW_MASK_64 | \ 1749 PSW_MASK_32; 1750 regs->gprs[15] = infop->start_stack; 1751 } 1752 1753 /* See linux kernel: arch/s390/include/uapi/asm/ptrace.h (s390_regs). */ 1754 #define ELF_NREG 27 1755 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1756 1757 enum { 1758 TARGET_REG_PSWM = 0, 1759 TARGET_REG_PSWA = 1, 1760 TARGET_REG_GPRS = 2, 1761 TARGET_REG_ARS = 18, 1762 TARGET_REG_ORIG_R2 = 26, 1763 }; 1764 1765 static void elf_core_copy_regs(target_elf_gregset_t *regs, 1766 const CPUS390XState *env) 1767 { 1768 int i; 1769 uint32_t *aregs; 1770 1771 (*regs)[TARGET_REG_PSWM] = tswapreg(env->psw.mask); 1772 (*regs)[TARGET_REG_PSWA] = tswapreg(env->psw.addr); 1773 for (i = 0; i < 16; i++) { 1774 (*regs)[TARGET_REG_GPRS + i] = tswapreg(env->regs[i]); 1775 } 1776 aregs = (uint32_t *)&((*regs)[TARGET_REG_ARS]); 1777 for (i = 0; i < 16; i++) { 1778 aregs[i] = tswap32(env->aregs[i]); 1779 } 1780 (*regs)[TARGET_REG_ORIG_R2] = 0; 1781 } 1782 1783 #define USE_ELF_CORE_DUMP 1784 #define ELF_EXEC_PAGESIZE 4096 1785 1786 #endif /* TARGET_S390X */ 1787 1788 #ifdef TARGET_RISCV 1789 1790 #define ELF_ARCH EM_RISCV 1791 1792 #ifdef TARGET_RISCV32 1793 #define ELF_CLASS ELFCLASS32 1794 #else 1795 #define ELF_CLASS ELFCLASS64 1796 #endif 1797 1798 #define ELF_HWCAP get_elf_hwcap() 1799 1800 static uint32_t get_elf_hwcap(void) 1801 { 1802 #define MISA_BIT(EXT) (1 << (EXT - 'A')) 1803 RISCVCPU *cpu = RISCV_CPU(thread_cpu); 1804 uint32_t mask = MISA_BIT('I') | MISA_BIT('M') | MISA_BIT('A') 1805 | MISA_BIT('F') | MISA_BIT('D') | MISA_BIT('C') 1806 | MISA_BIT('V'); 1807 1808 return cpu->env.misa_ext & mask; 1809 #undef MISA_BIT 1810 } 1811 1812 static inline void init_thread(struct target_pt_regs *regs, 1813 struct image_info *infop) 1814 { 1815 regs->sepc = infop->entry; 1816 regs->sp = infop->start_stack; 1817 } 1818 1819 #define ELF_EXEC_PAGESIZE 4096 1820 1821 #endif /* TARGET_RISCV */ 1822 1823 #ifdef TARGET_HPPA 1824 1825 #define ELF_CLASS ELFCLASS32 1826 #define ELF_ARCH EM_PARISC 1827 #define ELF_PLATFORM "PARISC" 1828 #define STACK_GROWS_DOWN 0 1829 #define STACK_ALIGNMENT 64 1830 1831 static inline void init_thread(struct target_pt_regs *regs, 1832 struct image_info *infop) 1833 { 1834 regs->iaoq[0] = infop->entry; 1835 regs->iaoq[1] = infop->entry + 4; 1836 regs->gr[23] = 0; 1837 regs->gr[24] = infop->argv; 1838 regs->gr[25] = infop->argc; 1839 /* The top-of-stack contains a linkage buffer. */ 1840 regs->gr[30] = infop->start_stack + 64; 1841 regs->gr[31] = infop->entry; 1842 } 1843 1844 #define LO_COMMPAGE 0 1845 1846 static bool init_guest_commpage(void) 1847 { 1848 void *want = g2h_untagged(LO_COMMPAGE); 1849 void *addr = mmap(want, qemu_host_page_size, PROT_NONE, 1850 MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0); 1851 1852 if (addr == MAP_FAILED) { 1853 perror("Allocating guest commpage"); 1854 exit(EXIT_FAILURE); 1855 } 1856 if (addr != want) { 1857 return false; 1858 } 1859 1860 /* 1861 * On Linux, page zero is normally marked execute only + gateway. 1862 * Normal read or write is supposed to fail (thus PROT_NONE above), 1863 * but specific offsets have kernel code mapped to raise permissions 1864 * and implement syscalls. Here, simply mark the page executable. 1865 * Special case the entry points during translation (see do_page_zero). 1866 */ 1867 page_set_flags(LO_COMMPAGE, LO_COMMPAGE | ~TARGET_PAGE_MASK, 1868 PAGE_EXEC | PAGE_VALID); 1869 return true; 1870 } 1871 1872 #endif /* TARGET_HPPA */ 1873 1874 #ifdef TARGET_XTENSA 1875 1876 #define ELF_CLASS ELFCLASS32 1877 #define ELF_ARCH EM_XTENSA 1878 1879 static inline void init_thread(struct target_pt_regs *regs, 1880 struct image_info *infop) 1881 { 1882 regs->windowbase = 0; 1883 regs->windowstart = 1; 1884 regs->areg[1] = infop->start_stack; 1885 regs->pc = infop->entry; 1886 if (info_is_fdpic(infop)) { 1887 regs->areg[4] = infop->loadmap_addr; 1888 regs->areg[5] = infop->interpreter_loadmap_addr; 1889 if (infop->interpreter_loadmap_addr) { 1890 regs->areg[6] = infop->interpreter_pt_dynamic_addr; 1891 } else { 1892 regs->areg[6] = infop->pt_dynamic_addr; 1893 } 1894 } 1895 } 1896 1897 /* See linux kernel: arch/xtensa/include/asm/elf.h. */ 1898 #define ELF_NREG 128 1899 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1900 1901 enum { 1902 TARGET_REG_PC, 1903 TARGET_REG_PS, 1904 TARGET_REG_LBEG, 1905 TARGET_REG_LEND, 1906 TARGET_REG_LCOUNT, 1907 TARGET_REG_SAR, 1908 TARGET_REG_WINDOWSTART, 1909 TARGET_REG_WINDOWBASE, 1910 TARGET_REG_THREADPTR, 1911 TARGET_REG_AR0 = 64, 1912 }; 1913 1914 static void elf_core_copy_regs(target_elf_gregset_t *regs, 1915 const CPUXtensaState *env) 1916 { 1917 unsigned i; 1918 1919 (*regs)[TARGET_REG_PC] = tswapreg(env->pc); 1920 (*regs)[TARGET_REG_PS] = tswapreg(env->sregs[PS] & ~PS_EXCM); 1921 (*regs)[TARGET_REG_LBEG] = tswapreg(env->sregs[LBEG]); 1922 (*regs)[TARGET_REG_LEND] = tswapreg(env->sregs[LEND]); 1923 (*regs)[TARGET_REG_LCOUNT] = tswapreg(env->sregs[LCOUNT]); 1924 (*regs)[TARGET_REG_SAR] = tswapreg(env->sregs[SAR]); 1925 (*regs)[TARGET_REG_WINDOWSTART] = tswapreg(env->sregs[WINDOW_START]); 1926 (*regs)[TARGET_REG_WINDOWBASE] = tswapreg(env->sregs[WINDOW_BASE]); 1927 (*regs)[TARGET_REG_THREADPTR] = tswapreg(env->uregs[THREADPTR]); 1928 xtensa_sync_phys_from_window((CPUXtensaState *)env); 1929 for (i = 0; i < env->config->nareg; ++i) { 1930 (*regs)[TARGET_REG_AR0 + i] = tswapreg(env->phys_regs[i]); 1931 } 1932 } 1933 1934 #define USE_ELF_CORE_DUMP 1935 #define ELF_EXEC_PAGESIZE 4096 1936 1937 #endif /* TARGET_XTENSA */ 1938 1939 #ifdef TARGET_HEXAGON 1940 1941 #define ELF_CLASS ELFCLASS32 1942 #define ELF_ARCH EM_HEXAGON 1943 1944 static inline void init_thread(struct target_pt_regs *regs, 1945 struct image_info *infop) 1946 { 1947 regs->sepc = infop->entry; 1948 regs->sp = infop->start_stack; 1949 } 1950 1951 #endif /* TARGET_HEXAGON */ 1952 1953 #ifndef ELF_BASE_PLATFORM 1954 #define ELF_BASE_PLATFORM (NULL) 1955 #endif 1956 1957 #ifndef ELF_PLATFORM 1958 #define ELF_PLATFORM (NULL) 1959 #endif 1960 1961 #ifndef ELF_MACHINE 1962 #define ELF_MACHINE ELF_ARCH 1963 #endif 1964 1965 #ifndef elf_check_arch 1966 #define elf_check_arch(x) ((x) == ELF_ARCH) 1967 #endif 1968 1969 #ifndef elf_check_abi 1970 #define elf_check_abi(x) (1) 1971 #endif 1972 1973 #ifndef ELF_HWCAP 1974 #define ELF_HWCAP 0 1975 #endif 1976 1977 #ifndef STACK_GROWS_DOWN 1978 #define STACK_GROWS_DOWN 1 1979 #endif 1980 1981 #ifndef STACK_ALIGNMENT 1982 #define STACK_ALIGNMENT 16 1983 #endif 1984 1985 #ifdef TARGET_ABI32 1986 #undef ELF_CLASS 1987 #define ELF_CLASS ELFCLASS32 1988 #undef bswaptls 1989 #define bswaptls(ptr) bswap32s(ptr) 1990 #endif 1991 1992 #ifndef EXSTACK_DEFAULT 1993 #define EXSTACK_DEFAULT false 1994 #endif 1995 1996 #include "elf.h" 1997 1998 /* We must delay the following stanzas until after "elf.h". */ 1999 #if defined(TARGET_AARCH64) 2000 2001 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz, 2002 const uint32_t *data, 2003 struct image_info *info, 2004 Error **errp) 2005 { 2006 if (pr_type == GNU_PROPERTY_AARCH64_FEATURE_1_AND) { 2007 if (pr_datasz != sizeof(uint32_t)) { 2008 error_setg(errp, "Ill-formed GNU_PROPERTY_AARCH64_FEATURE_1_AND"); 2009 return false; 2010 } 2011 /* We will extract GNU_PROPERTY_AARCH64_FEATURE_1_BTI later. */ 2012 info->note_flags = *data; 2013 } 2014 return true; 2015 } 2016 #define ARCH_USE_GNU_PROPERTY 1 2017 2018 #else 2019 2020 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz, 2021 const uint32_t *data, 2022 struct image_info *info, 2023 Error **errp) 2024 { 2025 g_assert_not_reached(); 2026 } 2027 #define ARCH_USE_GNU_PROPERTY 0 2028 2029 #endif 2030 2031 struct exec 2032 { 2033 unsigned int a_info; /* Use macros N_MAGIC, etc for access */ 2034 unsigned int a_text; /* length of text, in bytes */ 2035 unsigned int a_data; /* length of data, in bytes */ 2036 unsigned int a_bss; /* length of uninitialized data area, in bytes */ 2037 unsigned int a_syms; /* length of symbol table data in file, in bytes */ 2038 unsigned int a_entry; /* start address */ 2039 unsigned int a_trsize; /* length of relocation info for text, in bytes */ 2040 unsigned int a_drsize; /* length of relocation info for data, in bytes */ 2041 }; 2042 2043 2044 #define N_MAGIC(exec) ((exec).a_info & 0xffff) 2045 #define OMAGIC 0407 2046 #define NMAGIC 0410 2047 #define ZMAGIC 0413 2048 #define QMAGIC 0314 2049 2050 #define DLINFO_ITEMS 16 2051 2052 static inline void memcpy_fromfs(void * to, const void * from, unsigned long n) 2053 { 2054 memcpy(to, from, n); 2055 } 2056 2057 #ifdef BSWAP_NEEDED 2058 static void bswap_ehdr(struct elfhdr *ehdr) 2059 { 2060 bswap16s(&ehdr->e_type); /* Object file type */ 2061 bswap16s(&ehdr->e_machine); /* Architecture */ 2062 bswap32s(&ehdr->e_version); /* Object file version */ 2063 bswaptls(&ehdr->e_entry); /* Entry point virtual address */ 2064 bswaptls(&ehdr->e_phoff); /* Program header table file offset */ 2065 bswaptls(&ehdr->e_shoff); /* Section header table file offset */ 2066 bswap32s(&ehdr->e_flags); /* Processor-specific flags */ 2067 bswap16s(&ehdr->e_ehsize); /* ELF header size in bytes */ 2068 bswap16s(&ehdr->e_phentsize); /* Program header table entry size */ 2069 bswap16s(&ehdr->e_phnum); /* Program header table entry count */ 2070 bswap16s(&ehdr->e_shentsize); /* Section header table entry size */ 2071 bswap16s(&ehdr->e_shnum); /* Section header table entry count */ 2072 bswap16s(&ehdr->e_shstrndx); /* Section header string table index */ 2073 } 2074 2075 static void bswap_phdr(struct elf_phdr *phdr, int phnum) 2076 { 2077 int i; 2078 for (i = 0; i < phnum; ++i, ++phdr) { 2079 bswap32s(&phdr->p_type); /* Segment type */ 2080 bswap32s(&phdr->p_flags); /* Segment flags */ 2081 bswaptls(&phdr->p_offset); /* Segment file offset */ 2082 bswaptls(&phdr->p_vaddr); /* Segment virtual address */ 2083 bswaptls(&phdr->p_paddr); /* Segment physical address */ 2084 bswaptls(&phdr->p_filesz); /* Segment size in file */ 2085 bswaptls(&phdr->p_memsz); /* Segment size in memory */ 2086 bswaptls(&phdr->p_align); /* Segment alignment */ 2087 } 2088 } 2089 2090 static void bswap_shdr(struct elf_shdr *shdr, int shnum) 2091 { 2092 int i; 2093 for (i = 0; i < shnum; ++i, ++shdr) { 2094 bswap32s(&shdr->sh_name); 2095 bswap32s(&shdr->sh_type); 2096 bswaptls(&shdr->sh_flags); 2097 bswaptls(&shdr->sh_addr); 2098 bswaptls(&shdr->sh_offset); 2099 bswaptls(&shdr->sh_size); 2100 bswap32s(&shdr->sh_link); 2101 bswap32s(&shdr->sh_info); 2102 bswaptls(&shdr->sh_addralign); 2103 bswaptls(&shdr->sh_entsize); 2104 } 2105 } 2106 2107 static void bswap_sym(struct elf_sym *sym) 2108 { 2109 bswap32s(&sym->st_name); 2110 bswaptls(&sym->st_value); 2111 bswaptls(&sym->st_size); 2112 bswap16s(&sym->st_shndx); 2113 } 2114 2115 #ifdef TARGET_MIPS 2116 static void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags) 2117 { 2118 bswap16s(&abiflags->version); 2119 bswap32s(&abiflags->ases); 2120 bswap32s(&abiflags->isa_ext); 2121 bswap32s(&abiflags->flags1); 2122 bswap32s(&abiflags->flags2); 2123 } 2124 #endif 2125 #else 2126 static inline void bswap_ehdr(struct elfhdr *ehdr) { } 2127 static inline void bswap_phdr(struct elf_phdr *phdr, int phnum) { } 2128 static inline void bswap_shdr(struct elf_shdr *shdr, int shnum) { } 2129 static inline void bswap_sym(struct elf_sym *sym) { } 2130 #ifdef TARGET_MIPS 2131 static inline void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags) { } 2132 #endif 2133 #endif 2134 2135 #ifdef USE_ELF_CORE_DUMP 2136 static int elf_core_dump(int, const CPUArchState *); 2137 #endif /* USE_ELF_CORE_DUMP */ 2138 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias); 2139 2140 /* Verify the portions of EHDR within E_IDENT for the target. 2141 This can be performed before bswapping the entire header. */ 2142 static bool elf_check_ident(struct elfhdr *ehdr) 2143 { 2144 return (ehdr->e_ident[EI_MAG0] == ELFMAG0 2145 && ehdr->e_ident[EI_MAG1] == ELFMAG1 2146 && ehdr->e_ident[EI_MAG2] == ELFMAG2 2147 && ehdr->e_ident[EI_MAG3] == ELFMAG3 2148 && ehdr->e_ident[EI_CLASS] == ELF_CLASS 2149 && ehdr->e_ident[EI_DATA] == ELF_DATA 2150 && ehdr->e_ident[EI_VERSION] == EV_CURRENT); 2151 } 2152 2153 /* Verify the portions of EHDR outside of E_IDENT for the target. 2154 This has to wait until after bswapping the header. */ 2155 static bool elf_check_ehdr(struct elfhdr *ehdr) 2156 { 2157 return (elf_check_arch(ehdr->e_machine) 2158 && elf_check_abi(ehdr->e_flags) 2159 && ehdr->e_ehsize == sizeof(struct elfhdr) 2160 && ehdr->e_phentsize == sizeof(struct elf_phdr) 2161 && (ehdr->e_type == ET_EXEC || ehdr->e_type == ET_DYN)); 2162 } 2163 2164 /* 2165 * 'copy_elf_strings()' copies argument/envelope strings from user 2166 * memory to free pages in kernel mem. These are in a format ready 2167 * to be put directly into the top of new user memory. 2168 * 2169 */ 2170 static abi_ulong copy_elf_strings(int argc, char **argv, char *scratch, 2171 abi_ulong p, abi_ulong stack_limit) 2172 { 2173 char *tmp; 2174 int len, i; 2175 abi_ulong top = p; 2176 2177 if (!p) { 2178 return 0; /* bullet-proofing */ 2179 } 2180 2181 if (STACK_GROWS_DOWN) { 2182 int offset = ((p - 1) % TARGET_PAGE_SIZE) + 1; 2183 for (i = argc - 1; i >= 0; --i) { 2184 tmp = argv[i]; 2185 if (!tmp) { 2186 fprintf(stderr, "VFS: argc is wrong"); 2187 exit(-1); 2188 } 2189 len = strlen(tmp) + 1; 2190 tmp += len; 2191 2192 if (len > (p - stack_limit)) { 2193 return 0; 2194 } 2195 while (len) { 2196 int bytes_to_copy = (len > offset) ? offset : len; 2197 tmp -= bytes_to_copy; 2198 p -= bytes_to_copy; 2199 offset -= bytes_to_copy; 2200 len -= bytes_to_copy; 2201 2202 memcpy_fromfs(scratch + offset, tmp, bytes_to_copy); 2203 2204 if (offset == 0) { 2205 memcpy_to_target(p, scratch, top - p); 2206 top = p; 2207 offset = TARGET_PAGE_SIZE; 2208 } 2209 } 2210 } 2211 if (p != top) { 2212 memcpy_to_target(p, scratch + offset, top - p); 2213 } 2214 } else { 2215 int remaining = TARGET_PAGE_SIZE - (p % TARGET_PAGE_SIZE); 2216 for (i = 0; i < argc; ++i) { 2217 tmp = argv[i]; 2218 if (!tmp) { 2219 fprintf(stderr, "VFS: argc is wrong"); 2220 exit(-1); 2221 } 2222 len = strlen(tmp) + 1; 2223 if (len > (stack_limit - p)) { 2224 return 0; 2225 } 2226 while (len) { 2227 int bytes_to_copy = (len > remaining) ? remaining : len; 2228 2229 memcpy_fromfs(scratch + (p - top), tmp, bytes_to_copy); 2230 2231 tmp += bytes_to_copy; 2232 remaining -= bytes_to_copy; 2233 p += bytes_to_copy; 2234 len -= bytes_to_copy; 2235 2236 if (remaining == 0) { 2237 memcpy_to_target(top, scratch, p - top); 2238 top = p; 2239 remaining = TARGET_PAGE_SIZE; 2240 } 2241 } 2242 } 2243 if (p != top) { 2244 memcpy_to_target(top, scratch, p - top); 2245 } 2246 } 2247 2248 return p; 2249 } 2250 2251 /* Older linux kernels provide up to MAX_ARG_PAGES (default: 32) of 2252 * argument/environment space. Newer kernels (>2.6.33) allow more, 2253 * dependent on stack size, but guarantee at least 32 pages for 2254 * backwards compatibility. 2255 */ 2256 #define STACK_LOWER_LIMIT (32 * TARGET_PAGE_SIZE) 2257 2258 static abi_ulong setup_arg_pages(struct linux_binprm *bprm, 2259 struct image_info *info) 2260 { 2261 abi_ulong size, error, guard; 2262 int prot; 2263 2264 size = guest_stack_size; 2265 if (size < STACK_LOWER_LIMIT) { 2266 size = STACK_LOWER_LIMIT; 2267 } 2268 2269 if (STACK_GROWS_DOWN) { 2270 guard = TARGET_PAGE_SIZE; 2271 if (guard < qemu_real_host_page_size()) { 2272 guard = qemu_real_host_page_size(); 2273 } 2274 } else { 2275 /* no guard page for hppa target where stack grows upwards. */ 2276 guard = 0; 2277 } 2278 2279 prot = PROT_READ | PROT_WRITE; 2280 if (info->exec_stack) { 2281 prot |= PROT_EXEC; 2282 } 2283 error = target_mmap(0, size + guard, prot, 2284 MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); 2285 if (error == -1) { 2286 perror("mmap stack"); 2287 exit(-1); 2288 } 2289 2290 /* We reserve one extra page at the top of the stack as guard. */ 2291 if (STACK_GROWS_DOWN) { 2292 target_mprotect(error, guard, PROT_NONE); 2293 info->stack_limit = error + guard; 2294 return info->stack_limit + size - sizeof(void *); 2295 } else { 2296 info->stack_limit = error + size; 2297 return error; 2298 } 2299 } 2300 2301 /** 2302 * zero_bss: 2303 * 2304 * Map and zero the bss. We need to explicitly zero any fractional pages 2305 * after the data section (i.e. bss). Return false on mapping failure. 2306 */ 2307 static bool zero_bss(abi_ulong start_bss, abi_ulong end_bss, int prot) 2308 { 2309 abi_ulong align_bss; 2310 2311 align_bss = TARGET_PAGE_ALIGN(start_bss); 2312 end_bss = TARGET_PAGE_ALIGN(end_bss); 2313 2314 if (start_bss < align_bss) { 2315 int flags = page_get_flags(start_bss); 2316 2317 if (!(flags & PAGE_VALID)) { 2318 /* Map the start of the bss. */ 2319 align_bss -= TARGET_PAGE_SIZE; 2320 } else if (flags & PAGE_WRITE) { 2321 /* The page is already mapped writable. */ 2322 memset(g2h_untagged(start_bss), 0, align_bss - start_bss); 2323 } else { 2324 /* Read-only zeros? */ 2325 g_assert_not_reached(); 2326 } 2327 } 2328 2329 return align_bss >= end_bss || 2330 target_mmap(align_bss, end_bss - align_bss, prot, 2331 MAP_FIXED | MAP_PRIVATE | MAP_ANON, -1, 0) != -1; 2332 } 2333 2334 #if defined(TARGET_ARM) 2335 static int elf_is_fdpic(struct elfhdr *exec) 2336 { 2337 return exec->e_ident[EI_OSABI] == ELFOSABI_ARM_FDPIC; 2338 } 2339 #elif defined(TARGET_XTENSA) 2340 static int elf_is_fdpic(struct elfhdr *exec) 2341 { 2342 return exec->e_ident[EI_OSABI] == ELFOSABI_XTENSA_FDPIC; 2343 } 2344 #else 2345 /* Default implementation, always false. */ 2346 static int elf_is_fdpic(struct elfhdr *exec) 2347 { 2348 return 0; 2349 } 2350 #endif 2351 2352 static abi_ulong loader_build_fdpic_loadmap(struct image_info *info, abi_ulong sp) 2353 { 2354 uint16_t n; 2355 struct elf32_fdpic_loadseg *loadsegs = info->loadsegs; 2356 2357 /* elf32_fdpic_loadseg */ 2358 n = info->nsegs; 2359 while (n--) { 2360 sp -= 12; 2361 put_user_u32(loadsegs[n].addr, sp+0); 2362 put_user_u32(loadsegs[n].p_vaddr, sp+4); 2363 put_user_u32(loadsegs[n].p_memsz, sp+8); 2364 } 2365 2366 /* elf32_fdpic_loadmap */ 2367 sp -= 4; 2368 put_user_u16(0, sp+0); /* version */ 2369 put_user_u16(info->nsegs, sp+2); /* nsegs */ 2370 2371 info->personality = PER_LINUX_FDPIC; 2372 info->loadmap_addr = sp; 2373 2374 return sp; 2375 } 2376 2377 static abi_ulong create_elf_tables(abi_ulong p, int argc, int envc, 2378 struct elfhdr *exec, 2379 struct image_info *info, 2380 struct image_info *interp_info) 2381 { 2382 abi_ulong sp; 2383 abi_ulong u_argc, u_argv, u_envp, u_auxv; 2384 int size; 2385 int i; 2386 abi_ulong u_rand_bytes; 2387 uint8_t k_rand_bytes[16]; 2388 abi_ulong u_platform, u_base_platform; 2389 const char *k_platform, *k_base_platform; 2390 const int n = sizeof(elf_addr_t); 2391 2392 sp = p; 2393 2394 /* Needs to be before we load the env/argc/... */ 2395 if (elf_is_fdpic(exec)) { 2396 /* Need 4 byte alignment for these structs */ 2397 sp &= ~3; 2398 sp = loader_build_fdpic_loadmap(info, sp); 2399 info->other_info = interp_info; 2400 if (interp_info) { 2401 interp_info->other_info = info; 2402 sp = loader_build_fdpic_loadmap(interp_info, sp); 2403 info->interpreter_loadmap_addr = interp_info->loadmap_addr; 2404 info->interpreter_pt_dynamic_addr = interp_info->pt_dynamic_addr; 2405 } else { 2406 info->interpreter_loadmap_addr = 0; 2407 info->interpreter_pt_dynamic_addr = 0; 2408 } 2409 } 2410 2411 u_base_platform = 0; 2412 k_base_platform = ELF_BASE_PLATFORM; 2413 if (k_base_platform) { 2414 size_t len = strlen(k_base_platform) + 1; 2415 if (STACK_GROWS_DOWN) { 2416 sp -= (len + n - 1) & ~(n - 1); 2417 u_base_platform = sp; 2418 /* FIXME - check return value of memcpy_to_target() for failure */ 2419 memcpy_to_target(sp, k_base_platform, len); 2420 } else { 2421 memcpy_to_target(sp, k_base_platform, len); 2422 u_base_platform = sp; 2423 sp += len + 1; 2424 } 2425 } 2426 2427 u_platform = 0; 2428 k_platform = ELF_PLATFORM; 2429 if (k_platform) { 2430 size_t len = strlen(k_platform) + 1; 2431 if (STACK_GROWS_DOWN) { 2432 sp -= (len + n - 1) & ~(n - 1); 2433 u_platform = sp; 2434 /* FIXME - check return value of memcpy_to_target() for failure */ 2435 memcpy_to_target(sp, k_platform, len); 2436 } else { 2437 memcpy_to_target(sp, k_platform, len); 2438 u_platform = sp; 2439 sp += len + 1; 2440 } 2441 } 2442 2443 /* Provide 16 byte alignment for the PRNG, and basic alignment for 2444 * the argv and envp pointers. 2445 */ 2446 if (STACK_GROWS_DOWN) { 2447 sp = QEMU_ALIGN_DOWN(sp, 16); 2448 } else { 2449 sp = QEMU_ALIGN_UP(sp, 16); 2450 } 2451 2452 /* 2453 * Generate 16 random bytes for userspace PRNG seeding. 2454 */ 2455 qemu_guest_getrandom_nofail(k_rand_bytes, sizeof(k_rand_bytes)); 2456 if (STACK_GROWS_DOWN) { 2457 sp -= 16; 2458 u_rand_bytes = sp; 2459 /* FIXME - check return value of memcpy_to_target() for failure */ 2460 memcpy_to_target(sp, k_rand_bytes, 16); 2461 } else { 2462 memcpy_to_target(sp, k_rand_bytes, 16); 2463 u_rand_bytes = sp; 2464 sp += 16; 2465 } 2466 2467 size = (DLINFO_ITEMS + 1) * 2; 2468 if (k_base_platform) 2469 size += 2; 2470 if (k_platform) 2471 size += 2; 2472 #ifdef DLINFO_ARCH_ITEMS 2473 size += DLINFO_ARCH_ITEMS * 2; 2474 #endif 2475 #ifdef ELF_HWCAP2 2476 size += 2; 2477 #endif 2478 info->auxv_len = size * n; 2479 2480 size += envc + argc + 2; 2481 size += 1; /* argc itself */ 2482 size *= n; 2483 2484 /* Allocate space and finalize stack alignment for entry now. */ 2485 if (STACK_GROWS_DOWN) { 2486 u_argc = QEMU_ALIGN_DOWN(sp - size, STACK_ALIGNMENT); 2487 sp = u_argc; 2488 } else { 2489 u_argc = sp; 2490 sp = QEMU_ALIGN_UP(sp + size, STACK_ALIGNMENT); 2491 } 2492 2493 u_argv = u_argc + n; 2494 u_envp = u_argv + (argc + 1) * n; 2495 u_auxv = u_envp + (envc + 1) * n; 2496 info->saved_auxv = u_auxv; 2497 info->argc = argc; 2498 info->envc = envc; 2499 info->argv = u_argv; 2500 info->envp = u_envp; 2501 2502 /* This is correct because Linux defines 2503 * elf_addr_t as Elf32_Off / Elf64_Off 2504 */ 2505 #define NEW_AUX_ENT(id, val) do { \ 2506 put_user_ual(id, u_auxv); u_auxv += n; \ 2507 put_user_ual(val, u_auxv); u_auxv += n; \ 2508 } while(0) 2509 2510 #ifdef ARCH_DLINFO 2511 /* 2512 * ARCH_DLINFO must come first so platform specific code can enforce 2513 * special alignment requirements on the AUXV if necessary (eg. PPC). 2514 */ 2515 ARCH_DLINFO; 2516 #endif 2517 /* There must be exactly DLINFO_ITEMS entries here, or the assert 2518 * on info->auxv_len will trigger. 2519 */ 2520 NEW_AUX_ENT(AT_PHDR, (abi_ulong)(info->load_addr + exec->e_phoff)); 2521 NEW_AUX_ENT(AT_PHENT, (abi_ulong)(sizeof (struct elf_phdr))); 2522 NEW_AUX_ENT(AT_PHNUM, (abi_ulong)(exec->e_phnum)); 2523 if ((info->alignment & ~qemu_host_page_mask) != 0) { 2524 /* Target doesn't support host page size alignment */ 2525 NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(TARGET_PAGE_SIZE)); 2526 } else { 2527 NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(MAX(TARGET_PAGE_SIZE, 2528 qemu_host_page_size))); 2529 } 2530 NEW_AUX_ENT(AT_BASE, (abi_ulong)(interp_info ? interp_info->load_addr : 0)); 2531 NEW_AUX_ENT(AT_FLAGS, (abi_ulong)0); 2532 NEW_AUX_ENT(AT_ENTRY, info->entry); 2533 NEW_AUX_ENT(AT_UID, (abi_ulong) getuid()); 2534 NEW_AUX_ENT(AT_EUID, (abi_ulong) geteuid()); 2535 NEW_AUX_ENT(AT_GID, (abi_ulong) getgid()); 2536 NEW_AUX_ENT(AT_EGID, (abi_ulong) getegid()); 2537 NEW_AUX_ENT(AT_HWCAP, (abi_ulong) ELF_HWCAP); 2538 NEW_AUX_ENT(AT_CLKTCK, (abi_ulong) sysconf(_SC_CLK_TCK)); 2539 NEW_AUX_ENT(AT_RANDOM, (abi_ulong) u_rand_bytes); 2540 NEW_AUX_ENT(AT_SECURE, (abi_ulong) qemu_getauxval(AT_SECURE)); 2541 NEW_AUX_ENT(AT_EXECFN, info->file_string); 2542 2543 #ifdef ELF_HWCAP2 2544 NEW_AUX_ENT(AT_HWCAP2, (abi_ulong) ELF_HWCAP2); 2545 #endif 2546 2547 if (u_base_platform) { 2548 NEW_AUX_ENT(AT_BASE_PLATFORM, u_base_platform); 2549 } 2550 if (u_platform) { 2551 NEW_AUX_ENT(AT_PLATFORM, u_platform); 2552 } 2553 NEW_AUX_ENT (AT_NULL, 0); 2554 #undef NEW_AUX_ENT 2555 2556 /* Check that our initial calculation of the auxv length matches how much 2557 * we actually put into it. 2558 */ 2559 assert(info->auxv_len == u_auxv - info->saved_auxv); 2560 2561 put_user_ual(argc, u_argc); 2562 2563 p = info->arg_strings; 2564 for (i = 0; i < argc; ++i) { 2565 put_user_ual(p, u_argv); 2566 u_argv += n; 2567 p += target_strlen(p) + 1; 2568 } 2569 put_user_ual(0, u_argv); 2570 2571 p = info->env_strings; 2572 for (i = 0; i < envc; ++i) { 2573 put_user_ual(p, u_envp); 2574 u_envp += n; 2575 p += target_strlen(p) + 1; 2576 } 2577 put_user_ual(0, u_envp); 2578 2579 return sp; 2580 } 2581 2582 #if defined(HI_COMMPAGE) 2583 #define LO_COMMPAGE -1 2584 #elif defined(LO_COMMPAGE) 2585 #define HI_COMMPAGE 0 2586 #else 2587 #define HI_COMMPAGE 0 2588 #define LO_COMMPAGE -1 2589 #ifndef INIT_GUEST_COMMPAGE 2590 #define init_guest_commpage() true 2591 #endif 2592 #endif 2593 2594 /** 2595 * pgb_try_mmap: 2596 * @addr: host start address 2597 * @addr_last: host last address 2598 * @keep: do not unmap the probe region 2599 * 2600 * Return 1 if [@addr, @addr_last] is not mapped in the host, 2601 * return 0 if it is not available to map, and -1 on mmap error. 2602 * If @keep, the region is left mapped on success, otherwise unmapped. 2603 */ 2604 static int pgb_try_mmap(uintptr_t addr, uintptr_t addr_last, bool keep) 2605 { 2606 size_t size = addr_last - addr + 1; 2607 void *p = mmap((void *)addr, size, PROT_NONE, 2608 MAP_ANONYMOUS | MAP_PRIVATE | 2609 MAP_NORESERVE | MAP_FIXED_NOREPLACE, -1, 0); 2610 int ret; 2611 2612 if (p == MAP_FAILED) { 2613 return errno == EEXIST ? 0 : -1; 2614 } 2615 ret = p == (void *)addr; 2616 if (!keep || !ret) { 2617 munmap(p, size); 2618 } 2619 return ret; 2620 } 2621 2622 /** 2623 * pgb_try_mmap_skip_brk(uintptr_t addr, uintptr_t size, uintptr_t brk) 2624 * @addr: host address 2625 * @addr_last: host last address 2626 * @brk: host brk 2627 * 2628 * Like pgb_try_mmap, but additionally reserve some memory following brk. 2629 */ 2630 static int pgb_try_mmap_skip_brk(uintptr_t addr, uintptr_t addr_last, 2631 uintptr_t brk, bool keep) 2632 { 2633 uintptr_t brk_last = brk + 16 * MiB - 1; 2634 2635 /* Do not map anything close to the host brk. */ 2636 if (addr <= brk_last && brk <= addr_last) { 2637 return 0; 2638 } 2639 return pgb_try_mmap(addr, addr_last, keep); 2640 } 2641 2642 /** 2643 * pgb_try_mmap_set: 2644 * @ga: set of guest addrs 2645 * @base: guest_base 2646 * @brk: host brk 2647 * 2648 * Return true if all @ga can be mapped by the host at @base. 2649 * On success, retain the mapping at index 0 for reserved_va. 2650 */ 2651 2652 typedef struct PGBAddrs { 2653 uintptr_t bounds[3][2]; /* start/last pairs */ 2654 int nbounds; 2655 } PGBAddrs; 2656 2657 static bool pgb_try_mmap_set(const PGBAddrs *ga, uintptr_t base, uintptr_t brk) 2658 { 2659 for (int i = ga->nbounds - 1; i >= 0; --i) { 2660 if (pgb_try_mmap_skip_brk(ga->bounds[i][0] + base, 2661 ga->bounds[i][1] + base, 2662 brk, i == 0 && reserved_va) <= 0) { 2663 return false; 2664 } 2665 } 2666 return true; 2667 } 2668 2669 /** 2670 * pgb_addr_set: 2671 * @ga: output set of guest addrs 2672 * @guest_loaddr: guest image low address 2673 * @guest_loaddr: guest image high address 2674 * @identity: create for identity mapping 2675 * 2676 * Fill in @ga with the image, COMMPAGE and NULL page. 2677 */ 2678 static bool pgb_addr_set(PGBAddrs *ga, abi_ulong guest_loaddr, 2679 abi_ulong guest_hiaddr, bool try_identity) 2680 { 2681 int n; 2682 2683 /* 2684 * With a low commpage, or a guest mapped very low, 2685 * we may not be able to use the identity map. 2686 */ 2687 if (try_identity) { 2688 if (LO_COMMPAGE != -1 && LO_COMMPAGE < mmap_min_addr) { 2689 return false; 2690 } 2691 if (guest_loaddr != 0 && guest_loaddr < mmap_min_addr) { 2692 return false; 2693 } 2694 } 2695 2696 memset(ga, 0, sizeof(*ga)); 2697 n = 0; 2698 2699 if (reserved_va) { 2700 ga->bounds[n][0] = try_identity ? mmap_min_addr : 0; 2701 ga->bounds[n][1] = reserved_va; 2702 n++; 2703 /* LO_COMMPAGE and NULL handled by reserving from 0. */ 2704 } else { 2705 /* Add any LO_COMMPAGE or NULL page. */ 2706 if (LO_COMMPAGE != -1) { 2707 ga->bounds[n][0] = 0; 2708 ga->bounds[n][1] = LO_COMMPAGE + TARGET_PAGE_SIZE - 1; 2709 n++; 2710 } else if (!try_identity) { 2711 ga->bounds[n][0] = 0; 2712 ga->bounds[n][1] = TARGET_PAGE_SIZE - 1; 2713 n++; 2714 } 2715 2716 /* Add the guest image for ET_EXEC. */ 2717 if (guest_loaddr) { 2718 ga->bounds[n][0] = guest_loaddr; 2719 ga->bounds[n][1] = guest_hiaddr; 2720 n++; 2721 } 2722 } 2723 2724 /* 2725 * Temporarily disable 2726 * "comparison is always false due to limited range of data type" 2727 * due to comparison between unsigned and (possible) 0. 2728 */ 2729 #pragma GCC diagnostic push 2730 #pragma GCC diagnostic ignored "-Wtype-limits" 2731 2732 /* Add any HI_COMMPAGE not covered by reserved_va. */ 2733 if (reserved_va < HI_COMMPAGE) { 2734 ga->bounds[n][0] = HI_COMMPAGE & qemu_host_page_mask; 2735 ga->bounds[n][1] = HI_COMMPAGE + TARGET_PAGE_SIZE - 1; 2736 n++; 2737 } 2738 2739 #pragma GCC diagnostic pop 2740 2741 ga->nbounds = n; 2742 return true; 2743 } 2744 2745 static void pgb_fail_in_use(const char *image_name) 2746 { 2747 error_report("%s: requires virtual address space that is in use " 2748 "(omit the -B option or choose a different value)", 2749 image_name); 2750 exit(EXIT_FAILURE); 2751 } 2752 2753 static void pgb_fixed(const char *image_name, uintptr_t guest_loaddr, 2754 uintptr_t guest_hiaddr, uintptr_t align) 2755 { 2756 PGBAddrs ga; 2757 uintptr_t brk = (uintptr_t)sbrk(0); 2758 2759 if (!QEMU_IS_ALIGNED(guest_base, align)) { 2760 fprintf(stderr, "Requested guest base %p does not satisfy " 2761 "host minimum alignment (0x%" PRIxPTR ")\n", 2762 (void *)guest_base, align); 2763 exit(EXIT_FAILURE); 2764 } 2765 2766 if (!pgb_addr_set(&ga, guest_loaddr, guest_hiaddr, !guest_base) 2767 || !pgb_try_mmap_set(&ga, guest_base, brk)) { 2768 pgb_fail_in_use(image_name); 2769 } 2770 } 2771 2772 /** 2773 * pgb_find_fallback: 2774 * 2775 * This is a fallback method for finding holes in the host address space 2776 * if we don't have the benefit of being able to access /proc/self/map. 2777 * It can potentially take a very long time as we can only dumbly iterate 2778 * up the host address space seeing if the allocation would work. 2779 */ 2780 static uintptr_t pgb_find_fallback(const PGBAddrs *ga, uintptr_t align, 2781 uintptr_t brk) 2782 { 2783 /* TODO: come up with a better estimate of how much to skip. */ 2784 uintptr_t skip = sizeof(uintptr_t) == 4 ? MiB : GiB; 2785 2786 for (uintptr_t base = skip; ; base += skip) { 2787 base = ROUND_UP(base, align); 2788 if (pgb_try_mmap_set(ga, base, brk)) { 2789 return base; 2790 } 2791 if (base >= -skip) { 2792 return -1; 2793 } 2794 } 2795 } 2796 2797 static uintptr_t pgb_try_itree(const PGBAddrs *ga, uintptr_t base, 2798 IntervalTreeRoot *root) 2799 { 2800 for (int i = ga->nbounds - 1; i >= 0; --i) { 2801 uintptr_t s = base + ga->bounds[i][0]; 2802 uintptr_t l = base + ga->bounds[i][1]; 2803 IntervalTreeNode *n; 2804 2805 if (l < s) { 2806 /* Wraparound. Skip to advance S to mmap_min_addr. */ 2807 return mmap_min_addr - s; 2808 } 2809 2810 n = interval_tree_iter_first(root, s, l); 2811 if (n != NULL) { 2812 /* Conflict. Skip to advance S to LAST + 1. */ 2813 return n->last - s + 1; 2814 } 2815 } 2816 return 0; /* success */ 2817 } 2818 2819 static uintptr_t pgb_find_itree(const PGBAddrs *ga, IntervalTreeRoot *root, 2820 uintptr_t align, uintptr_t brk) 2821 { 2822 uintptr_t last = mmap_min_addr; 2823 uintptr_t base, skip; 2824 2825 while (true) { 2826 base = ROUND_UP(last, align); 2827 if (base < last) { 2828 return -1; 2829 } 2830 2831 skip = pgb_try_itree(ga, base, root); 2832 if (skip == 0) { 2833 break; 2834 } 2835 2836 last = base + skip; 2837 if (last < base) { 2838 return -1; 2839 } 2840 } 2841 2842 /* 2843 * We've chosen 'base' based on holes in the interval tree, 2844 * but we don't yet know if it is a valid host address. 2845 * Because it is the first matching hole, if the host addresses 2846 * are invalid we know there are no further matches. 2847 */ 2848 return pgb_try_mmap_set(ga, base, brk) ? base : -1; 2849 } 2850 2851 static void pgb_dynamic(const char *image_name, uintptr_t guest_loaddr, 2852 uintptr_t guest_hiaddr, uintptr_t align) 2853 { 2854 IntervalTreeRoot *root; 2855 uintptr_t brk, ret; 2856 PGBAddrs ga; 2857 2858 assert(QEMU_IS_ALIGNED(guest_loaddr, align)); 2859 2860 /* Try the identity map first. */ 2861 if (pgb_addr_set(&ga, guest_loaddr, guest_hiaddr, true)) { 2862 brk = (uintptr_t)sbrk(0); 2863 if (pgb_try_mmap_set(&ga, 0, brk)) { 2864 guest_base = 0; 2865 return; 2866 } 2867 } 2868 2869 /* 2870 * Rebuild the address set for non-identity map. 2871 * This differs in the mapping of the guest NULL page. 2872 */ 2873 pgb_addr_set(&ga, guest_loaddr, guest_hiaddr, false); 2874 2875 root = read_self_maps(); 2876 2877 /* Read brk after we've read the maps, which will malloc. */ 2878 brk = (uintptr_t)sbrk(0); 2879 2880 if (!root) { 2881 ret = pgb_find_fallback(&ga, align, brk); 2882 } else { 2883 /* 2884 * Reserve the area close to the host brk. 2885 * This will be freed with the rest of the tree. 2886 */ 2887 IntervalTreeNode *b = g_new0(IntervalTreeNode, 1); 2888 b->start = brk; 2889 b->last = brk + 16 * MiB - 1; 2890 interval_tree_insert(b, root); 2891 2892 ret = pgb_find_itree(&ga, root, align, brk); 2893 free_self_maps(root); 2894 } 2895 2896 if (ret == -1) { 2897 int w = TARGET_LONG_BITS / 4; 2898 2899 error_report("%s: Unable to find a guest_base to satisfy all " 2900 "guest address mapping requirements", image_name); 2901 2902 for (int i = 0; i < ga.nbounds; ++i) { 2903 error_printf(" %0*" PRIx64 "-%0*" PRIx64 "\n", 2904 w, (uint64_t)ga.bounds[i][0], 2905 w, (uint64_t)ga.bounds[i][1]); 2906 } 2907 exit(EXIT_FAILURE); 2908 } 2909 guest_base = ret; 2910 } 2911 2912 void probe_guest_base(const char *image_name, abi_ulong guest_loaddr, 2913 abi_ulong guest_hiaddr) 2914 { 2915 /* In order to use host shmat, we must be able to honor SHMLBA. */ 2916 uintptr_t align = MAX(SHMLBA, qemu_host_page_size); 2917 2918 /* Sanity check the guest binary. */ 2919 if (reserved_va) { 2920 if (guest_hiaddr > reserved_va) { 2921 error_report("%s: requires more than reserved virtual " 2922 "address space (0x%" PRIx64 " > 0x%lx)", 2923 image_name, (uint64_t)guest_hiaddr, reserved_va); 2924 exit(EXIT_FAILURE); 2925 } 2926 } else { 2927 if (guest_hiaddr != (uintptr_t)guest_hiaddr) { 2928 error_report("%s: requires more virtual address space " 2929 "than the host can provide (0x%" PRIx64 ")", 2930 image_name, (uint64_t)guest_hiaddr + 1); 2931 exit(EXIT_FAILURE); 2932 } 2933 } 2934 2935 if (have_guest_base) { 2936 pgb_fixed(image_name, guest_loaddr, guest_hiaddr, align); 2937 } else { 2938 pgb_dynamic(image_name, guest_loaddr, guest_hiaddr, align); 2939 } 2940 2941 /* Reserve and initialize the commpage. */ 2942 if (!init_guest_commpage()) { 2943 /* We have already probed for the commpage being free. */ 2944 g_assert_not_reached(); 2945 } 2946 2947 assert(QEMU_IS_ALIGNED(guest_base, align)); 2948 qemu_log_mask(CPU_LOG_PAGE, "Locating guest address space " 2949 "@ 0x%" PRIx64 "\n", (uint64_t)guest_base); 2950 } 2951 2952 enum { 2953 /* The string "GNU\0" as a magic number. */ 2954 GNU0_MAGIC = const_le32('G' | 'N' << 8 | 'U' << 16), 2955 NOTE_DATA_SZ = 1 * KiB, 2956 NOTE_NAME_SZ = 4, 2957 ELF_GNU_PROPERTY_ALIGN = ELF_CLASS == ELFCLASS32 ? 4 : 8, 2958 }; 2959 2960 /* 2961 * Process a single gnu_property entry. 2962 * Return false for error. 2963 */ 2964 static bool parse_elf_property(const uint32_t *data, int *off, int datasz, 2965 struct image_info *info, bool have_prev_type, 2966 uint32_t *prev_type, Error **errp) 2967 { 2968 uint32_t pr_type, pr_datasz, step; 2969 2970 if (*off > datasz || !QEMU_IS_ALIGNED(*off, ELF_GNU_PROPERTY_ALIGN)) { 2971 goto error_data; 2972 } 2973 datasz -= *off; 2974 data += *off / sizeof(uint32_t); 2975 2976 if (datasz < 2 * sizeof(uint32_t)) { 2977 goto error_data; 2978 } 2979 pr_type = data[0]; 2980 pr_datasz = data[1]; 2981 data += 2; 2982 datasz -= 2 * sizeof(uint32_t); 2983 step = ROUND_UP(pr_datasz, ELF_GNU_PROPERTY_ALIGN); 2984 if (step > datasz) { 2985 goto error_data; 2986 } 2987 2988 /* Properties are supposed to be unique and sorted on pr_type. */ 2989 if (have_prev_type && pr_type <= *prev_type) { 2990 if (pr_type == *prev_type) { 2991 error_setg(errp, "Duplicate property in PT_GNU_PROPERTY"); 2992 } else { 2993 error_setg(errp, "Unsorted property in PT_GNU_PROPERTY"); 2994 } 2995 return false; 2996 } 2997 *prev_type = pr_type; 2998 2999 if (!arch_parse_elf_property(pr_type, pr_datasz, data, info, errp)) { 3000 return false; 3001 } 3002 3003 *off += 2 * sizeof(uint32_t) + step; 3004 return true; 3005 3006 error_data: 3007 error_setg(errp, "Ill-formed property in PT_GNU_PROPERTY"); 3008 return false; 3009 } 3010 3011 /* Process NT_GNU_PROPERTY_TYPE_0. */ 3012 static bool parse_elf_properties(int image_fd, 3013 struct image_info *info, 3014 const struct elf_phdr *phdr, 3015 char bprm_buf[BPRM_BUF_SIZE], 3016 Error **errp) 3017 { 3018 union { 3019 struct elf_note nhdr; 3020 uint32_t data[NOTE_DATA_SZ / sizeof(uint32_t)]; 3021 } note; 3022 3023 int n, off, datasz; 3024 bool have_prev_type; 3025 uint32_t prev_type; 3026 3027 /* Unless the arch requires properties, ignore them. */ 3028 if (!ARCH_USE_GNU_PROPERTY) { 3029 return true; 3030 } 3031 3032 /* If the properties are crazy large, that's too bad. */ 3033 n = phdr->p_filesz; 3034 if (n > sizeof(note)) { 3035 error_setg(errp, "PT_GNU_PROPERTY too large"); 3036 return false; 3037 } 3038 if (n < sizeof(note.nhdr)) { 3039 error_setg(errp, "PT_GNU_PROPERTY too small"); 3040 return false; 3041 } 3042 3043 if (phdr->p_offset + n <= BPRM_BUF_SIZE) { 3044 memcpy(¬e, bprm_buf + phdr->p_offset, n); 3045 } else { 3046 ssize_t len = pread(image_fd, ¬e, n, phdr->p_offset); 3047 if (len != n) { 3048 error_setg_errno(errp, errno, "Error reading file header"); 3049 return false; 3050 } 3051 } 3052 3053 /* 3054 * The contents of a valid PT_GNU_PROPERTY is a sequence 3055 * of uint32_t -- swap them all now. 3056 */ 3057 #ifdef BSWAP_NEEDED 3058 for (int i = 0; i < n / 4; i++) { 3059 bswap32s(note.data + i); 3060 } 3061 #endif 3062 3063 /* 3064 * Note that nhdr is 3 words, and that the "name" described by namesz 3065 * immediately follows nhdr and is thus at the 4th word. Further, all 3066 * of the inputs to the kernel's round_up are multiples of 4. 3067 */ 3068 if (note.nhdr.n_type != NT_GNU_PROPERTY_TYPE_0 || 3069 note.nhdr.n_namesz != NOTE_NAME_SZ || 3070 note.data[3] != GNU0_MAGIC) { 3071 error_setg(errp, "Invalid note in PT_GNU_PROPERTY"); 3072 return false; 3073 } 3074 off = sizeof(note.nhdr) + NOTE_NAME_SZ; 3075 3076 datasz = note.nhdr.n_descsz + off; 3077 if (datasz > n) { 3078 error_setg(errp, "Invalid note size in PT_GNU_PROPERTY"); 3079 return false; 3080 } 3081 3082 have_prev_type = false; 3083 prev_type = 0; 3084 while (1) { 3085 if (off == datasz) { 3086 return true; /* end, exit ok */ 3087 } 3088 if (!parse_elf_property(note.data, &off, datasz, info, 3089 have_prev_type, &prev_type, errp)) { 3090 return false; 3091 } 3092 have_prev_type = true; 3093 } 3094 } 3095 3096 /* Load an ELF image into the address space. 3097 3098 IMAGE_NAME is the filename of the image, to use in error messages. 3099 IMAGE_FD is the open file descriptor for the image. 3100 3101 BPRM_BUF is a copy of the beginning of the file; this of course 3102 contains the elf file header at offset 0. It is assumed that this 3103 buffer is sufficiently aligned to present no problems to the host 3104 in accessing data at aligned offsets within the buffer. 3105 3106 On return: INFO values will be filled in, as necessary or available. */ 3107 3108 static void load_elf_image(const char *image_name, int image_fd, 3109 struct image_info *info, char **pinterp_name, 3110 char bprm_buf[BPRM_BUF_SIZE]) 3111 { 3112 struct elfhdr *ehdr = (struct elfhdr *)bprm_buf; 3113 struct elf_phdr *phdr; 3114 abi_ulong load_addr, load_bias, loaddr, hiaddr, error; 3115 int i, retval, prot_exec; 3116 Error *err = NULL; 3117 3118 /* First of all, some simple consistency checks */ 3119 if (!elf_check_ident(ehdr)) { 3120 error_setg(&err, "Invalid ELF image for this architecture"); 3121 goto exit_errmsg; 3122 } 3123 bswap_ehdr(ehdr); 3124 if (!elf_check_ehdr(ehdr)) { 3125 error_setg(&err, "Invalid ELF image for this architecture"); 3126 goto exit_errmsg; 3127 } 3128 3129 i = ehdr->e_phnum * sizeof(struct elf_phdr); 3130 if (ehdr->e_phoff + i <= BPRM_BUF_SIZE) { 3131 phdr = (struct elf_phdr *)(bprm_buf + ehdr->e_phoff); 3132 } else { 3133 phdr = (struct elf_phdr *) alloca(i); 3134 retval = pread(image_fd, phdr, i, ehdr->e_phoff); 3135 if (retval != i) { 3136 goto exit_read; 3137 } 3138 } 3139 bswap_phdr(phdr, ehdr->e_phnum); 3140 3141 info->nsegs = 0; 3142 info->pt_dynamic_addr = 0; 3143 3144 mmap_lock(); 3145 3146 /* 3147 * Find the maximum size of the image and allocate an appropriate 3148 * amount of memory to handle that. Locate the interpreter, if any. 3149 */ 3150 loaddr = -1, hiaddr = 0; 3151 info->alignment = 0; 3152 info->exec_stack = EXSTACK_DEFAULT; 3153 for (i = 0; i < ehdr->e_phnum; ++i) { 3154 struct elf_phdr *eppnt = phdr + i; 3155 if (eppnt->p_type == PT_LOAD) { 3156 abi_ulong a = eppnt->p_vaddr - eppnt->p_offset; 3157 if (a < loaddr) { 3158 loaddr = a; 3159 } 3160 a = eppnt->p_vaddr + eppnt->p_memsz - 1; 3161 if (a > hiaddr) { 3162 hiaddr = a; 3163 } 3164 ++info->nsegs; 3165 info->alignment |= eppnt->p_align; 3166 } else if (eppnt->p_type == PT_INTERP && pinterp_name) { 3167 g_autofree char *interp_name = NULL; 3168 3169 if (*pinterp_name) { 3170 error_setg(&err, "Multiple PT_INTERP entries"); 3171 goto exit_errmsg; 3172 } 3173 3174 interp_name = g_malloc(eppnt->p_filesz); 3175 3176 if (eppnt->p_offset + eppnt->p_filesz <= BPRM_BUF_SIZE) { 3177 memcpy(interp_name, bprm_buf + eppnt->p_offset, 3178 eppnt->p_filesz); 3179 } else { 3180 retval = pread(image_fd, interp_name, eppnt->p_filesz, 3181 eppnt->p_offset); 3182 if (retval != eppnt->p_filesz) { 3183 goto exit_read; 3184 } 3185 } 3186 if (interp_name[eppnt->p_filesz - 1] != 0) { 3187 error_setg(&err, "Invalid PT_INTERP entry"); 3188 goto exit_errmsg; 3189 } 3190 *pinterp_name = g_steal_pointer(&interp_name); 3191 } else if (eppnt->p_type == PT_GNU_PROPERTY) { 3192 if (!parse_elf_properties(image_fd, info, eppnt, bprm_buf, &err)) { 3193 goto exit_errmsg; 3194 } 3195 } else if (eppnt->p_type == PT_GNU_STACK) { 3196 info->exec_stack = eppnt->p_flags & PF_X; 3197 } 3198 } 3199 3200 load_addr = loaddr; 3201 3202 if (pinterp_name != NULL) { 3203 if (ehdr->e_type == ET_EXEC) { 3204 /* 3205 * Make sure that the low address does not conflict with 3206 * MMAP_MIN_ADDR or the QEMU application itself. 3207 */ 3208 probe_guest_base(image_name, loaddr, hiaddr); 3209 } else { 3210 abi_ulong align; 3211 3212 /* 3213 * The binary is dynamic, but we still need to 3214 * select guest_base. In this case we pass a size. 3215 */ 3216 probe_guest_base(image_name, 0, hiaddr - loaddr); 3217 3218 /* 3219 * Avoid collision with the loader by providing a different 3220 * default load address. 3221 */ 3222 load_addr += elf_et_dyn_base; 3223 3224 /* 3225 * TODO: Better support for mmap alignment is desirable. 3226 * Since we do not have complete control over the guest 3227 * address space, we prefer the kernel to choose some address 3228 * rather than force the use of LOAD_ADDR via MAP_FIXED. 3229 * But without MAP_FIXED we cannot guarantee alignment, 3230 * only suggest it. 3231 */ 3232 align = pow2ceil(info->alignment); 3233 if (align) { 3234 load_addr &= -align; 3235 } 3236 } 3237 } 3238 3239 /* 3240 * Reserve address space for all of this. 3241 * 3242 * In the case of ET_EXEC, we supply MAP_FIXED_NOREPLACE so that we get 3243 * exactly the address range that is required. Without reserved_va, 3244 * the guest address space is not isolated. We have attempted to avoid 3245 * conflict with the host program itself via probe_guest_base, but using 3246 * MAP_FIXED_NOREPLACE instead of MAP_FIXED provides an extra check. 3247 * 3248 * Otherwise this is ET_DYN, and we are searching for a location 3249 * that can hold the memory space required. If the image is 3250 * pre-linked, LOAD_ADDR will be non-zero, and the kernel should 3251 * honor that address if it happens to be free. 3252 * 3253 * In both cases, we will overwrite pages in this range with mappings 3254 * from the executable. 3255 */ 3256 load_addr = target_mmap(load_addr, (size_t)hiaddr - loaddr + 1, PROT_NONE, 3257 MAP_PRIVATE | MAP_ANON | MAP_NORESERVE | 3258 (ehdr->e_type == ET_EXEC ? MAP_FIXED_NOREPLACE : 0), 3259 -1, 0); 3260 if (load_addr == -1) { 3261 goto exit_mmap; 3262 } 3263 load_bias = load_addr - loaddr; 3264 3265 if (elf_is_fdpic(ehdr)) { 3266 struct elf32_fdpic_loadseg *loadsegs = info->loadsegs = 3267 g_malloc(sizeof(*loadsegs) * info->nsegs); 3268 3269 for (i = 0; i < ehdr->e_phnum; ++i) { 3270 switch (phdr[i].p_type) { 3271 case PT_DYNAMIC: 3272 info->pt_dynamic_addr = phdr[i].p_vaddr + load_bias; 3273 break; 3274 case PT_LOAD: 3275 loadsegs->addr = phdr[i].p_vaddr + load_bias; 3276 loadsegs->p_vaddr = phdr[i].p_vaddr; 3277 loadsegs->p_memsz = phdr[i].p_memsz; 3278 ++loadsegs; 3279 break; 3280 } 3281 } 3282 } 3283 3284 info->load_bias = load_bias; 3285 info->code_offset = load_bias; 3286 info->data_offset = load_bias; 3287 info->load_addr = load_addr; 3288 info->entry = ehdr->e_entry + load_bias; 3289 info->start_code = -1; 3290 info->end_code = 0; 3291 info->start_data = -1; 3292 info->end_data = 0; 3293 /* Usual start for brk is after all sections of the main executable. */ 3294 info->brk = TARGET_PAGE_ALIGN(hiaddr + load_bias); 3295 info->elf_flags = ehdr->e_flags; 3296 3297 prot_exec = PROT_EXEC; 3298 #ifdef TARGET_AARCH64 3299 /* 3300 * If the BTI feature is present, this indicates that the executable 3301 * pages of the startup binary should be mapped with PROT_BTI, so that 3302 * branch targets are enforced. 3303 * 3304 * The startup binary is either the interpreter or the static executable. 3305 * The interpreter is responsible for all pages of a dynamic executable. 3306 * 3307 * Elf notes are backward compatible to older cpus. 3308 * Do not enable BTI unless it is supported. 3309 */ 3310 if ((info->note_flags & GNU_PROPERTY_AARCH64_FEATURE_1_BTI) 3311 && (pinterp_name == NULL || *pinterp_name == 0) 3312 && cpu_isar_feature(aa64_bti, ARM_CPU(thread_cpu))) { 3313 prot_exec |= TARGET_PROT_BTI; 3314 } 3315 #endif 3316 3317 for (i = 0; i < ehdr->e_phnum; i++) { 3318 struct elf_phdr *eppnt = phdr + i; 3319 if (eppnt->p_type == PT_LOAD) { 3320 abi_ulong vaddr, vaddr_po, vaddr_ps, vaddr_ef, vaddr_em; 3321 int elf_prot = 0; 3322 3323 if (eppnt->p_flags & PF_R) { 3324 elf_prot |= PROT_READ; 3325 } 3326 if (eppnt->p_flags & PF_W) { 3327 elf_prot |= PROT_WRITE; 3328 } 3329 if (eppnt->p_flags & PF_X) { 3330 elf_prot |= prot_exec; 3331 } 3332 3333 vaddr = load_bias + eppnt->p_vaddr; 3334 vaddr_po = vaddr & ~TARGET_PAGE_MASK; 3335 vaddr_ps = vaddr & TARGET_PAGE_MASK; 3336 3337 vaddr_ef = vaddr + eppnt->p_filesz; 3338 vaddr_em = vaddr + eppnt->p_memsz; 3339 3340 /* 3341 * Some segments may be completely empty, with a non-zero p_memsz 3342 * but no backing file segment. 3343 */ 3344 if (eppnt->p_filesz != 0) { 3345 error = target_mmap(vaddr_ps, eppnt->p_filesz + vaddr_po, 3346 elf_prot, MAP_PRIVATE | MAP_FIXED, 3347 image_fd, eppnt->p_offset - vaddr_po); 3348 if (error == -1) { 3349 goto exit_mmap; 3350 } 3351 } 3352 3353 /* If the load segment requests extra zeros (e.g. bss), map it. */ 3354 if (vaddr_ef < vaddr_em && 3355 !zero_bss(vaddr_ef, vaddr_em, elf_prot)) { 3356 goto exit_mmap; 3357 } 3358 3359 /* Find the full program boundaries. */ 3360 if (elf_prot & PROT_EXEC) { 3361 if (vaddr < info->start_code) { 3362 info->start_code = vaddr; 3363 } 3364 if (vaddr_ef > info->end_code) { 3365 info->end_code = vaddr_ef; 3366 } 3367 } 3368 if (elf_prot & PROT_WRITE) { 3369 if (vaddr < info->start_data) { 3370 info->start_data = vaddr; 3371 } 3372 if (vaddr_ef > info->end_data) { 3373 info->end_data = vaddr_ef; 3374 } 3375 } 3376 #ifdef TARGET_MIPS 3377 } else if (eppnt->p_type == PT_MIPS_ABIFLAGS) { 3378 Mips_elf_abiflags_v0 abiflags; 3379 if (eppnt->p_filesz < sizeof(Mips_elf_abiflags_v0)) { 3380 error_setg(&err, "Invalid PT_MIPS_ABIFLAGS entry"); 3381 goto exit_errmsg; 3382 } 3383 if (eppnt->p_offset + eppnt->p_filesz <= BPRM_BUF_SIZE) { 3384 memcpy(&abiflags, bprm_buf + eppnt->p_offset, 3385 sizeof(Mips_elf_abiflags_v0)); 3386 } else { 3387 retval = pread(image_fd, &abiflags, sizeof(Mips_elf_abiflags_v0), 3388 eppnt->p_offset); 3389 if (retval != sizeof(Mips_elf_abiflags_v0)) { 3390 goto exit_read; 3391 } 3392 } 3393 bswap_mips_abiflags(&abiflags); 3394 info->fp_abi = abiflags.fp_abi; 3395 #endif 3396 } 3397 } 3398 3399 if (info->end_data == 0) { 3400 info->start_data = info->end_code; 3401 info->end_data = info->end_code; 3402 } 3403 3404 if (qemu_log_enabled()) { 3405 load_symbols(ehdr, image_fd, load_bias); 3406 } 3407 3408 debuginfo_report_elf(image_name, image_fd, load_bias); 3409 3410 mmap_unlock(); 3411 3412 close(image_fd); 3413 return; 3414 3415 exit_read: 3416 if (retval >= 0) { 3417 error_setg(&err, "Incomplete read of file header"); 3418 } else { 3419 error_setg_errno(&err, errno, "Error reading file header"); 3420 } 3421 goto exit_errmsg; 3422 exit_mmap: 3423 error_setg_errno(&err, errno, "Error mapping file"); 3424 goto exit_errmsg; 3425 exit_errmsg: 3426 error_reportf_err(err, "%s: ", image_name); 3427 exit(-1); 3428 } 3429 3430 static void load_elf_interp(const char *filename, struct image_info *info, 3431 char bprm_buf[BPRM_BUF_SIZE]) 3432 { 3433 int fd, retval; 3434 Error *err = NULL; 3435 3436 fd = open(path(filename), O_RDONLY); 3437 if (fd < 0) { 3438 error_setg_file_open(&err, errno, filename); 3439 error_report_err(err); 3440 exit(-1); 3441 } 3442 3443 retval = read(fd, bprm_buf, BPRM_BUF_SIZE); 3444 if (retval < 0) { 3445 error_setg_errno(&err, errno, "Error reading file header"); 3446 error_reportf_err(err, "%s: ", filename); 3447 exit(-1); 3448 } 3449 3450 if (retval < BPRM_BUF_SIZE) { 3451 memset(bprm_buf + retval, 0, BPRM_BUF_SIZE - retval); 3452 } 3453 3454 load_elf_image(filename, fd, info, NULL, bprm_buf); 3455 } 3456 3457 static int symfind(const void *s0, const void *s1) 3458 { 3459 struct elf_sym *sym = (struct elf_sym *)s1; 3460 __typeof(sym->st_value) addr = *(uint64_t *)s0; 3461 int result = 0; 3462 3463 if (addr < sym->st_value) { 3464 result = -1; 3465 } else if (addr >= sym->st_value + sym->st_size) { 3466 result = 1; 3467 } 3468 return result; 3469 } 3470 3471 static const char *lookup_symbolxx(struct syminfo *s, uint64_t orig_addr) 3472 { 3473 #if ELF_CLASS == ELFCLASS32 3474 struct elf_sym *syms = s->disas_symtab.elf32; 3475 #else 3476 struct elf_sym *syms = s->disas_symtab.elf64; 3477 #endif 3478 3479 // binary search 3480 struct elf_sym *sym; 3481 3482 sym = bsearch(&orig_addr, syms, s->disas_num_syms, sizeof(*syms), symfind); 3483 if (sym != NULL) { 3484 return s->disas_strtab + sym->st_name; 3485 } 3486 3487 return ""; 3488 } 3489 3490 /* FIXME: This should use elf_ops.h */ 3491 static int symcmp(const void *s0, const void *s1) 3492 { 3493 struct elf_sym *sym0 = (struct elf_sym *)s0; 3494 struct elf_sym *sym1 = (struct elf_sym *)s1; 3495 return (sym0->st_value < sym1->st_value) 3496 ? -1 3497 : ((sym0->st_value > sym1->st_value) ? 1 : 0); 3498 } 3499 3500 /* Best attempt to load symbols from this ELF object. */ 3501 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias) 3502 { 3503 int i, shnum, nsyms, sym_idx = 0, str_idx = 0; 3504 uint64_t segsz; 3505 struct elf_shdr *shdr; 3506 char *strings = NULL; 3507 struct syminfo *s = NULL; 3508 struct elf_sym *new_syms, *syms = NULL; 3509 3510 shnum = hdr->e_shnum; 3511 i = shnum * sizeof(struct elf_shdr); 3512 shdr = (struct elf_shdr *)alloca(i); 3513 if (pread(fd, shdr, i, hdr->e_shoff) != i) { 3514 return; 3515 } 3516 3517 bswap_shdr(shdr, shnum); 3518 for (i = 0; i < shnum; ++i) { 3519 if (shdr[i].sh_type == SHT_SYMTAB) { 3520 sym_idx = i; 3521 str_idx = shdr[i].sh_link; 3522 goto found; 3523 } 3524 } 3525 3526 /* There will be no symbol table if the file was stripped. */ 3527 return; 3528 3529 found: 3530 /* Now know where the strtab and symtab are. Snarf them. */ 3531 s = g_try_new(struct syminfo, 1); 3532 if (!s) { 3533 goto give_up; 3534 } 3535 3536 segsz = shdr[str_idx].sh_size; 3537 s->disas_strtab = strings = g_try_malloc(segsz); 3538 if (!strings || 3539 pread(fd, strings, segsz, shdr[str_idx].sh_offset) != segsz) { 3540 goto give_up; 3541 } 3542 3543 segsz = shdr[sym_idx].sh_size; 3544 syms = g_try_malloc(segsz); 3545 if (!syms || pread(fd, syms, segsz, shdr[sym_idx].sh_offset) != segsz) { 3546 goto give_up; 3547 } 3548 3549 if (segsz / sizeof(struct elf_sym) > INT_MAX) { 3550 /* Implausibly large symbol table: give up rather than ploughing 3551 * on with the number of symbols calculation overflowing 3552 */ 3553 goto give_up; 3554 } 3555 nsyms = segsz / sizeof(struct elf_sym); 3556 for (i = 0; i < nsyms; ) { 3557 bswap_sym(syms + i); 3558 /* Throw away entries which we do not need. */ 3559 if (syms[i].st_shndx == SHN_UNDEF 3560 || syms[i].st_shndx >= SHN_LORESERVE 3561 || ELF_ST_TYPE(syms[i].st_info) != STT_FUNC) { 3562 if (i < --nsyms) { 3563 syms[i] = syms[nsyms]; 3564 } 3565 } else { 3566 #if defined(TARGET_ARM) || defined (TARGET_MIPS) 3567 /* The bottom address bit marks a Thumb or MIPS16 symbol. */ 3568 syms[i].st_value &= ~(target_ulong)1; 3569 #endif 3570 syms[i].st_value += load_bias; 3571 i++; 3572 } 3573 } 3574 3575 /* No "useful" symbol. */ 3576 if (nsyms == 0) { 3577 goto give_up; 3578 } 3579 3580 /* Attempt to free the storage associated with the local symbols 3581 that we threw away. Whether or not this has any effect on the 3582 memory allocation depends on the malloc implementation and how 3583 many symbols we managed to discard. */ 3584 new_syms = g_try_renew(struct elf_sym, syms, nsyms); 3585 if (new_syms == NULL) { 3586 goto give_up; 3587 } 3588 syms = new_syms; 3589 3590 qsort(syms, nsyms, sizeof(*syms), symcmp); 3591 3592 s->disas_num_syms = nsyms; 3593 #if ELF_CLASS == ELFCLASS32 3594 s->disas_symtab.elf32 = syms; 3595 #else 3596 s->disas_symtab.elf64 = syms; 3597 #endif 3598 s->lookup_symbol = lookup_symbolxx; 3599 s->next = syminfos; 3600 syminfos = s; 3601 3602 return; 3603 3604 give_up: 3605 g_free(s); 3606 g_free(strings); 3607 g_free(syms); 3608 } 3609 3610 uint32_t get_elf_eflags(int fd) 3611 { 3612 struct elfhdr ehdr; 3613 off_t offset; 3614 int ret; 3615 3616 /* Read ELF header */ 3617 offset = lseek(fd, 0, SEEK_SET); 3618 if (offset == (off_t) -1) { 3619 return 0; 3620 } 3621 ret = read(fd, &ehdr, sizeof(ehdr)); 3622 if (ret < sizeof(ehdr)) { 3623 return 0; 3624 } 3625 offset = lseek(fd, offset, SEEK_SET); 3626 if (offset == (off_t) -1) { 3627 return 0; 3628 } 3629 3630 /* Check ELF signature */ 3631 if (!elf_check_ident(&ehdr)) { 3632 return 0; 3633 } 3634 3635 /* check header */ 3636 bswap_ehdr(&ehdr); 3637 if (!elf_check_ehdr(&ehdr)) { 3638 return 0; 3639 } 3640 3641 /* return architecture id */ 3642 return ehdr.e_flags; 3643 } 3644 3645 int load_elf_binary(struct linux_binprm *bprm, struct image_info *info) 3646 { 3647 struct image_info interp_info; 3648 struct elfhdr elf_ex; 3649 char *elf_interpreter = NULL; 3650 char *scratch; 3651 3652 memset(&interp_info, 0, sizeof(interp_info)); 3653 #ifdef TARGET_MIPS 3654 interp_info.fp_abi = MIPS_ABI_FP_UNKNOWN; 3655 #endif 3656 3657 load_elf_image(bprm->filename, bprm->fd, info, 3658 &elf_interpreter, bprm->buf); 3659 3660 /* ??? We need a copy of the elf header for passing to create_elf_tables. 3661 If we do nothing, we'll have overwritten this when we re-use bprm->buf 3662 when we load the interpreter. */ 3663 elf_ex = *(struct elfhdr *)bprm->buf; 3664 3665 /* Do this so that we can load the interpreter, if need be. We will 3666 change some of these later */ 3667 bprm->p = setup_arg_pages(bprm, info); 3668 3669 scratch = g_new0(char, TARGET_PAGE_SIZE); 3670 if (STACK_GROWS_DOWN) { 3671 bprm->p = copy_elf_strings(1, &bprm->filename, scratch, 3672 bprm->p, info->stack_limit); 3673 info->file_string = bprm->p; 3674 bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch, 3675 bprm->p, info->stack_limit); 3676 info->env_strings = bprm->p; 3677 bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch, 3678 bprm->p, info->stack_limit); 3679 info->arg_strings = bprm->p; 3680 } else { 3681 info->arg_strings = bprm->p; 3682 bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch, 3683 bprm->p, info->stack_limit); 3684 info->env_strings = bprm->p; 3685 bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch, 3686 bprm->p, info->stack_limit); 3687 info->file_string = bprm->p; 3688 bprm->p = copy_elf_strings(1, &bprm->filename, scratch, 3689 bprm->p, info->stack_limit); 3690 } 3691 3692 g_free(scratch); 3693 3694 if (!bprm->p) { 3695 fprintf(stderr, "%s: %s\n", bprm->filename, strerror(E2BIG)); 3696 exit(-1); 3697 } 3698 3699 if (elf_interpreter) { 3700 load_elf_interp(elf_interpreter, &interp_info, bprm->buf); 3701 3702 /* 3703 * While unusual because of ELF_ET_DYN_BASE, if we are unlucky 3704 * with the mappings the interpreter can be loaded above but 3705 * near the main executable, which can leave very little room 3706 * for the heap. 3707 * If the current brk has less than 16MB, use the end of the 3708 * interpreter. 3709 */ 3710 if (interp_info.brk > info->brk && 3711 interp_info.load_bias - info->brk < 16 * MiB) { 3712 info->brk = interp_info.brk; 3713 } 3714 3715 /* If the program interpreter is one of these two, then assume 3716 an iBCS2 image. Otherwise assume a native linux image. */ 3717 3718 if (strcmp(elf_interpreter, "/usr/lib/libc.so.1") == 0 3719 || strcmp(elf_interpreter, "/usr/lib/ld.so.1") == 0) { 3720 info->personality = PER_SVR4; 3721 3722 /* Why this, you ask??? Well SVr4 maps page 0 as read-only, 3723 and some applications "depend" upon this behavior. Since 3724 we do not have the power to recompile these, we emulate 3725 the SVr4 behavior. Sigh. */ 3726 target_mmap(0, qemu_host_page_size, PROT_READ | PROT_EXEC, 3727 MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); 3728 } 3729 #ifdef TARGET_MIPS 3730 info->interp_fp_abi = interp_info.fp_abi; 3731 #endif 3732 } 3733 3734 /* 3735 * TODO: load a vdso, which would also contain the signal trampolines. 3736 * Otherwise, allocate a private page to hold them. 3737 */ 3738 if (TARGET_ARCH_HAS_SIGTRAMP_PAGE) { 3739 abi_long tramp_page = target_mmap(0, TARGET_PAGE_SIZE, 3740 PROT_READ | PROT_WRITE, 3741 MAP_PRIVATE | MAP_ANON, -1, 0); 3742 if (tramp_page == -1) { 3743 return -errno; 3744 } 3745 3746 setup_sigtramp(tramp_page); 3747 target_mprotect(tramp_page, TARGET_PAGE_SIZE, PROT_READ | PROT_EXEC); 3748 } 3749 3750 bprm->p = create_elf_tables(bprm->p, bprm->argc, bprm->envc, &elf_ex, 3751 info, (elf_interpreter ? &interp_info : NULL)); 3752 info->start_stack = bprm->p; 3753 3754 /* If we have an interpreter, set that as the program's entry point. 3755 Copy the load_bias as well, to help PPC64 interpret the entry 3756 point as a function descriptor. Do this after creating elf tables 3757 so that we copy the original program entry point into the AUXV. */ 3758 if (elf_interpreter) { 3759 info->load_bias = interp_info.load_bias; 3760 info->entry = interp_info.entry; 3761 g_free(elf_interpreter); 3762 } 3763 3764 #ifdef USE_ELF_CORE_DUMP 3765 bprm->core_dump = &elf_core_dump; 3766 #endif 3767 3768 return 0; 3769 } 3770 3771 #ifdef USE_ELF_CORE_DUMP 3772 /* 3773 * Definitions to generate Intel SVR4-like core files. 3774 * These mostly have the same names as the SVR4 types with "target_elf_" 3775 * tacked on the front to prevent clashes with linux definitions, 3776 * and the typedef forms have been avoided. This is mostly like 3777 * the SVR4 structure, but more Linuxy, with things that Linux does 3778 * not support and which gdb doesn't really use excluded. 3779 * 3780 * Fields we don't dump (their contents is zero) in linux-user qemu 3781 * are marked with XXX. 3782 * 3783 * Core dump code is copied from linux kernel (fs/binfmt_elf.c). 3784 * 3785 * Porting ELF coredump for target is (quite) simple process. First you 3786 * define USE_ELF_CORE_DUMP in target ELF code (where init_thread() for 3787 * the target resides): 3788 * 3789 * #define USE_ELF_CORE_DUMP 3790 * 3791 * Next you define type of register set used for dumping. ELF specification 3792 * says that it needs to be array of elf_greg_t that has size of ELF_NREG. 3793 * 3794 * typedef <target_regtype> target_elf_greg_t; 3795 * #define ELF_NREG <number of registers> 3796 * typedef taret_elf_greg_t target_elf_gregset_t[ELF_NREG]; 3797 * 3798 * Last step is to implement target specific function that copies registers 3799 * from given cpu into just specified register set. Prototype is: 3800 * 3801 * static void elf_core_copy_regs(taret_elf_gregset_t *regs, 3802 * const CPUArchState *env); 3803 * 3804 * Parameters: 3805 * regs - copy register values into here (allocated and zeroed by caller) 3806 * env - copy registers from here 3807 * 3808 * Example for ARM target is provided in this file. 3809 */ 3810 3811 /* An ELF note in memory */ 3812 struct memelfnote { 3813 const char *name; 3814 size_t namesz; 3815 size_t namesz_rounded; 3816 int type; 3817 size_t datasz; 3818 size_t datasz_rounded; 3819 void *data; 3820 size_t notesz; 3821 }; 3822 3823 struct target_elf_siginfo { 3824 abi_int si_signo; /* signal number */ 3825 abi_int si_code; /* extra code */ 3826 abi_int si_errno; /* errno */ 3827 }; 3828 3829 struct target_elf_prstatus { 3830 struct target_elf_siginfo pr_info; /* Info associated with signal */ 3831 abi_short pr_cursig; /* Current signal */ 3832 abi_ulong pr_sigpend; /* XXX */ 3833 abi_ulong pr_sighold; /* XXX */ 3834 target_pid_t pr_pid; 3835 target_pid_t pr_ppid; 3836 target_pid_t pr_pgrp; 3837 target_pid_t pr_sid; 3838 struct target_timeval pr_utime; /* XXX User time */ 3839 struct target_timeval pr_stime; /* XXX System time */ 3840 struct target_timeval pr_cutime; /* XXX Cumulative user time */ 3841 struct target_timeval pr_cstime; /* XXX Cumulative system time */ 3842 target_elf_gregset_t pr_reg; /* GP registers */ 3843 abi_int pr_fpvalid; /* XXX */ 3844 }; 3845 3846 #define ELF_PRARGSZ (80) /* Number of chars for args */ 3847 3848 struct target_elf_prpsinfo { 3849 char pr_state; /* numeric process state */ 3850 char pr_sname; /* char for pr_state */ 3851 char pr_zomb; /* zombie */ 3852 char pr_nice; /* nice val */ 3853 abi_ulong pr_flag; /* flags */ 3854 target_uid_t pr_uid; 3855 target_gid_t pr_gid; 3856 target_pid_t pr_pid, pr_ppid, pr_pgrp, pr_sid; 3857 /* Lots missing */ 3858 char pr_fname[16] QEMU_NONSTRING; /* filename of executable */ 3859 char pr_psargs[ELF_PRARGSZ]; /* initial part of arg list */ 3860 }; 3861 3862 /* Here is the structure in which status of each thread is captured. */ 3863 struct elf_thread_status { 3864 QTAILQ_ENTRY(elf_thread_status) ets_link; 3865 struct target_elf_prstatus prstatus; /* NT_PRSTATUS */ 3866 #if 0 3867 elf_fpregset_t fpu; /* NT_PRFPREG */ 3868 struct task_struct *thread; 3869 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */ 3870 #endif 3871 struct memelfnote notes[1]; 3872 int num_notes; 3873 }; 3874 3875 struct elf_note_info { 3876 struct memelfnote *notes; 3877 struct target_elf_prstatus *prstatus; /* NT_PRSTATUS */ 3878 struct target_elf_prpsinfo *psinfo; /* NT_PRPSINFO */ 3879 3880 QTAILQ_HEAD(, elf_thread_status) thread_list; 3881 #if 0 3882 /* 3883 * Current version of ELF coredump doesn't support 3884 * dumping fp regs etc. 3885 */ 3886 elf_fpregset_t *fpu; 3887 elf_fpxregset_t *xfpu; 3888 int thread_status_size; 3889 #endif 3890 int notes_size; 3891 int numnote; 3892 }; 3893 3894 struct vm_area_struct { 3895 target_ulong vma_start; /* start vaddr of memory region */ 3896 target_ulong vma_end; /* end vaddr of memory region */ 3897 abi_ulong vma_flags; /* protection etc. flags for the region */ 3898 QTAILQ_ENTRY(vm_area_struct) vma_link; 3899 }; 3900 3901 struct mm_struct { 3902 QTAILQ_HEAD(, vm_area_struct) mm_mmap; 3903 int mm_count; /* number of mappings */ 3904 }; 3905 3906 static struct mm_struct *vma_init(void); 3907 static void vma_delete(struct mm_struct *); 3908 static int vma_add_mapping(struct mm_struct *, target_ulong, 3909 target_ulong, abi_ulong); 3910 static int vma_get_mapping_count(const struct mm_struct *); 3911 static struct vm_area_struct *vma_first(const struct mm_struct *); 3912 static struct vm_area_struct *vma_next(struct vm_area_struct *); 3913 static abi_ulong vma_dump_size(const struct vm_area_struct *); 3914 static int vma_walker(void *priv, target_ulong start, target_ulong end, 3915 unsigned long flags); 3916 3917 static void fill_elf_header(struct elfhdr *, int, uint16_t, uint32_t); 3918 static void fill_note(struct memelfnote *, const char *, int, 3919 unsigned int, void *); 3920 static void fill_prstatus(struct target_elf_prstatus *, const TaskState *, int); 3921 static int fill_psinfo(struct target_elf_prpsinfo *, const TaskState *); 3922 static void fill_auxv_note(struct memelfnote *, const TaskState *); 3923 static void fill_elf_note_phdr(struct elf_phdr *, int, off_t); 3924 static size_t note_size(const struct memelfnote *); 3925 static void free_note_info(struct elf_note_info *); 3926 static int fill_note_info(struct elf_note_info *, long, const CPUArchState *); 3927 static void fill_thread_info(struct elf_note_info *, const CPUArchState *); 3928 3929 static int dump_write(int, const void *, size_t); 3930 static int write_note(struct memelfnote *, int); 3931 static int write_note_info(struct elf_note_info *, int); 3932 3933 #ifdef BSWAP_NEEDED 3934 static void bswap_prstatus(struct target_elf_prstatus *prstatus) 3935 { 3936 prstatus->pr_info.si_signo = tswap32(prstatus->pr_info.si_signo); 3937 prstatus->pr_info.si_code = tswap32(prstatus->pr_info.si_code); 3938 prstatus->pr_info.si_errno = tswap32(prstatus->pr_info.si_errno); 3939 prstatus->pr_cursig = tswap16(prstatus->pr_cursig); 3940 prstatus->pr_sigpend = tswapal(prstatus->pr_sigpend); 3941 prstatus->pr_sighold = tswapal(prstatus->pr_sighold); 3942 prstatus->pr_pid = tswap32(prstatus->pr_pid); 3943 prstatus->pr_ppid = tswap32(prstatus->pr_ppid); 3944 prstatus->pr_pgrp = tswap32(prstatus->pr_pgrp); 3945 prstatus->pr_sid = tswap32(prstatus->pr_sid); 3946 /* cpu times are not filled, so we skip them */ 3947 /* regs should be in correct format already */ 3948 prstatus->pr_fpvalid = tswap32(prstatus->pr_fpvalid); 3949 } 3950 3951 static void bswap_psinfo(struct target_elf_prpsinfo *psinfo) 3952 { 3953 psinfo->pr_flag = tswapal(psinfo->pr_flag); 3954 psinfo->pr_uid = tswap16(psinfo->pr_uid); 3955 psinfo->pr_gid = tswap16(psinfo->pr_gid); 3956 psinfo->pr_pid = tswap32(psinfo->pr_pid); 3957 psinfo->pr_ppid = tswap32(psinfo->pr_ppid); 3958 psinfo->pr_pgrp = tswap32(psinfo->pr_pgrp); 3959 psinfo->pr_sid = tswap32(psinfo->pr_sid); 3960 } 3961 3962 static void bswap_note(struct elf_note *en) 3963 { 3964 bswap32s(&en->n_namesz); 3965 bswap32s(&en->n_descsz); 3966 bswap32s(&en->n_type); 3967 } 3968 #else 3969 static inline void bswap_prstatus(struct target_elf_prstatus *p) { } 3970 static inline void bswap_psinfo(struct target_elf_prpsinfo *p) {} 3971 static inline void bswap_note(struct elf_note *en) { } 3972 #endif /* BSWAP_NEEDED */ 3973 3974 /* 3975 * Minimal support for linux memory regions. These are needed 3976 * when we are finding out what memory exactly belongs to 3977 * emulated process. No locks needed here, as long as 3978 * thread that received the signal is stopped. 3979 */ 3980 3981 static struct mm_struct *vma_init(void) 3982 { 3983 struct mm_struct *mm; 3984 3985 if ((mm = g_malloc(sizeof (*mm))) == NULL) 3986 return (NULL); 3987 3988 mm->mm_count = 0; 3989 QTAILQ_INIT(&mm->mm_mmap); 3990 3991 return (mm); 3992 } 3993 3994 static void vma_delete(struct mm_struct *mm) 3995 { 3996 struct vm_area_struct *vma; 3997 3998 while ((vma = vma_first(mm)) != NULL) { 3999 QTAILQ_REMOVE(&mm->mm_mmap, vma, vma_link); 4000 g_free(vma); 4001 } 4002 g_free(mm); 4003 } 4004 4005 static int vma_add_mapping(struct mm_struct *mm, target_ulong start, 4006 target_ulong end, abi_ulong flags) 4007 { 4008 struct vm_area_struct *vma; 4009 4010 if ((vma = g_malloc0(sizeof (*vma))) == NULL) 4011 return (-1); 4012 4013 vma->vma_start = start; 4014 vma->vma_end = end; 4015 vma->vma_flags = flags; 4016 4017 QTAILQ_INSERT_TAIL(&mm->mm_mmap, vma, vma_link); 4018 mm->mm_count++; 4019 4020 return (0); 4021 } 4022 4023 static struct vm_area_struct *vma_first(const struct mm_struct *mm) 4024 { 4025 return (QTAILQ_FIRST(&mm->mm_mmap)); 4026 } 4027 4028 static struct vm_area_struct *vma_next(struct vm_area_struct *vma) 4029 { 4030 return (QTAILQ_NEXT(vma, vma_link)); 4031 } 4032 4033 static int vma_get_mapping_count(const struct mm_struct *mm) 4034 { 4035 return (mm->mm_count); 4036 } 4037 4038 /* 4039 * Calculate file (dump) size of given memory region. 4040 */ 4041 static abi_ulong vma_dump_size(const struct vm_area_struct *vma) 4042 { 4043 /* if we cannot even read the first page, skip it */ 4044 if (!access_ok_untagged(VERIFY_READ, vma->vma_start, TARGET_PAGE_SIZE)) 4045 return (0); 4046 4047 /* 4048 * Usually we don't dump executable pages as they contain 4049 * non-writable code that debugger can read directly from 4050 * target library etc. However, thread stacks are marked 4051 * also executable so we read in first page of given region 4052 * and check whether it contains elf header. If there is 4053 * no elf header, we dump it. 4054 */ 4055 if (vma->vma_flags & PROT_EXEC) { 4056 char page[TARGET_PAGE_SIZE]; 4057 4058 if (copy_from_user(page, vma->vma_start, sizeof (page))) { 4059 return 0; 4060 } 4061 if ((page[EI_MAG0] == ELFMAG0) && 4062 (page[EI_MAG1] == ELFMAG1) && 4063 (page[EI_MAG2] == ELFMAG2) && 4064 (page[EI_MAG3] == ELFMAG3)) { 4065 /* 4066 * Mappings are possibly from ELF binary. Don't dump 4067 * them. 4068 */ 4069 return (0); 4070 } 4071 } 4072 4073 return (vma->vma_end - vma->vma_start); 4074 } 4075 4076 static int vma_walker(void *priv, target_ulong start, target_ulong end, 4077 unsigned long flags) 4078 { 4079 struct mm_struct *mm = (struct mm_struct *)priv; 4080 4081 vma_add_mapping(mm, start, end, flags); 4082 return (0); 4083 } 4084 4085 static void fill_note(struct memelfnote *note, const char *name, int type, 4086 unsigned int sz, void *data) 4087 { 4088 unsigned int namesz; 4089 4090 namesz = strlen(name) + 1; 4091 note->name = name; 4092 note->namesz = namesz; 4093 note->namesz_rounded = roundup(namesz, sizeof (int32_t)); 4094 note->type = type; 4095 note->datasz = sz; 4096 note->datasz_rounded = roundup(sz, sizeof (int32_t)); 4097 4098 note->data = data; 4099 4100 /* 4101 * We calculate rounded up note size here as specified by 4102 * ELF document. 4103 */ 4104 note->notesz = sizeof (struct elf_note) + 4105 note->namesz_rounded + note->datasz_rounded; 4106 } 4107 4108 static void fill_elf_header(struct elfhdr *elf, int segs, uint16_t machine, 4109 uint32_t flags) 4110 { 4111 (void) memset(elf, 0, sizeof(*elf)); 4112 4113 (void) memcpy(elf->e_ident, ELFMAG, SELFMAG); 4114 elf->e_ident[EI_CLASS] = ELF_CLASS; 4115 elf->e_ident[EI_DATA] = ELF_DATA; 4116 elf->e_ident[EI_VERSION] = EV_CURRENT; 4117 elf->e_ident[EI_OSABI] = ELF_OSABI; 4118 4119 elf->e_type = ET_CORE; 4120 elf->e_machine = machine; 4121 elf->e_version = EV_CURRENT; 4122 elf->e_phoff = sizeof(struct elfhdr); 4123 elf->e_flags = flags; 4124 elf->e_ehsize = sizeof(struct elfhdr); 4125 elf->e_phentsize = sizeof(struct elf_phdr); 4126 elf->e_phnum = segs; 4127 4128 bswap_ehdr(elf); 4129 } 4130 4131 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, off_t offset) 4132 { 4133 phdr->p_type = PT_NOTE; 4134 phdr->p_offset = offset; 4135 phdr->p_vaddr = 0; 4136 phdr->p_paddr = 0; 4137 phdr->p_filesz = sz; 4138 phdr->p_memsz = 0; 4139 phdr->p_flags = 0; 4140 phdr->p_align = 0; 4141 4142 bswap_phdr(phdr, 1); 4143 } 4144 4145 static size_t note_size(const struct memelfnote *note) 4146 { 4147 return (note->notesz); 4148 } 4149 4150 static void fill_prstatus(struct target_elf_prstatus *prstatus, 4151 const TaskState *ts, int signr) 4152 { 4153 (void) memset(prstatus, 0, sizeof (*prstatus)); 4154 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr; 4155 prstatus->pr_pid = ts->ts_tid; 4156 prstatus->pr_ppid = getppid(); 4157 prstatus->pr_pgrp = getpgrp(); 4158 prstatus->pr_sid = getsid(0); 4159 4160 bswap_prstatus(prstatus); 4161 } 4162 4163 static int fill_psinfo(struct target_elf_prpsinfo *psinfo, const TaskState *ts) 4164 { 4165 char *base_filename; 4166 unsigned int i, len; 4167 4168 (void) memset(psinfo, 0, sizeof (*psinfo)); 4169 4170 len = ts->info->env_strings - ts->info->arg_strings; 4171 if (len >= ELF_PRARGSZ) 4172 len = ELF_PRARGSZ - 1; 4173 if (copy_from_user(&psinfo->pr_psargs, ts->info->arg_strings, len)) { 4174 return -EFAULT; 4175 } 4176 for (i = 0; i < len; i++) 4177 if (psinfo->pr_psargs[i] == 0) 4178 psinfo->pr_psargs[i] = ' '; 4179 psinfo->pr_psargs[len] = 0; 4180 4181 psinfo->pr_pid = getpid(); 4182 psinfo->pr_ppid = getppid(); 4183 psinfo->pr_pgrp = getpgrp(); 4184 psinfo->pr_sid = getsid(0); 4185 psinfo->pr_uid = getuid(); 4186 psinfo->pr_gid = getgid(); 4187 4188 base_filename = g_path_get_basename(ts->bprm->filename); 4189 /* 4190 * Using strncpy here is fine: at max-length, 4191 * this field is not NUL-terminated. 4192 */ 4193 (void) strncpy(psinfo->pr_fname, base_filename, 4194 sizeof(psinfo->pr_fname)); 4195 4196 g_free(base_filename); 4197 bswap_psinfo(psinfo); 4198 return (0); 4199 } 4200 4201 static void fill_auxv_note(struct memelfnote *note, const TaskState *ts) 4202 { 4203 elf_addr_t auxv = (elf_addr_t)ts->info->saved_auxv; 4204 elf_addr_t orig_auxv = auxv; 4205 void *ptr; 4206 int len = ts->info->auxv_len; 4207 4208 /* 4209 * Auxiliary vector is stored in target process stack. It contains 4210 * {type, value} pairs that we need to dump into note. This is not 4211 * strictly necessary but we do it here for sake of completeness. 4212 */ 4213 4214 /* read in whole auxv vector and copy it to memelfnote */ 4215 ptr = lock_user(VERIFY_READ, orig_auxv, len, 0); 4216 if (ptr != NULL) { 4217 fill_note(note, "CORE", NT_AUXV, len, ptr); 4218 unlock_user(ptr, auxv, len); 4219 } 4220 } 4221 4222 /* 4223 * Constructs name of coredump file. We have following convention 4224 * for the name: 4225 * qemu_<basename-of-target-binary>_<date>-<time>_<pid>.core 4226 * 4227 * Returns the filename 4228 */ 4229 static char *core_dump_filename(const TaskState *ts) 4230 { 4231 g_autoptr(GDateTime) now = g_date_time_new_now_local(); 4232 g_autofree char *nowstr = g_date_time_format(now, "%Y%m%d-%H%M%S"); 4233 g_autofree char *base_filename = g_path_get_basename(ts->bprm->filename); 4234 4235 return g_strdup_printf("qemu_%s_%s_%d.core", 4236 base_filename, nowstr, (int)getpid()); 4237 } 4238 4239 static int dump_write(int fd, const void *ptr, size_t size) 4240 { 4241 const char *bufp = (const char *)ptr; 4242 ssize_t bytes_written, bytes_left; 4243 struct rlimit dumpsize; 4244 off_t pos; 4245 4246 bytes_written = 0; 4247 getrlimit(RLIMIT_CORE, &dumpsize); 4248 if ((pos = lseek(fd, 0, SEEK_CUR))==-1) { 4249 if (errno == ESPIPE) { /* not a seekable stream */ 4250 bytes_left = size; 4251 } else { 4252 return pos; 4253 } 4254 } else { 4255 if (dumpsize.rlim_cur <= pos) { 4256 return -1; 4257 } else if (dumpsize.rlim_cur == RLIM_INFINITY) { 4258 bytes_left = size; 4259 } else { 4260 size_t limit_left=dumpsize.rlim_cur - pos; 4261 bytes_left = limit_left >= size ? size : limit_left ; 4262 } 4263 } 4264 4265 /* 4266 * In normal conditions, single write(2) should do but 4267 * in case of socket etc. this mechanism is more portable. 4268 */ 4269 do { 4270 bytes_written = write(fd, bufp, bytes_left); 4271 if (bytes_written < 0) { 4272 if (errno == EINTR) 4273 continue; 4274 return (-1); 4275 } else if (bytes_written == 0) { /* eof */ 4276 return (-1); 4277 } 4278 bufp += bytes_written; 4279 bytes_left -= bytes_written; 4280 } while (bytes_left > 0); 4281 4282 return (0); 4283 } 4284 4285 static int write_note(struct memelfnote *men, int fd) 4286 { 4287 struct elf_note en; 4288 4289 en.n_namesz = men->namesz; 4290 en.n_type = men->type; 4291 en.n_descsz = men->datasz; 4292 4293 bswap_note(&en); 4294 4295 if (dump_write(fd, &en, sizeof(en)) != 0) 4296 return (-1); 4297 if (dump_write(fd, men->name, men->namesz_rounded) != 0) 4298 return (-1); 4299 if (dump_write(fd, men->data, men->datasz_rounded) != 0) 4300 return (-1); 4301 4302 return (0); 4303 } 4304 4305 static void fill_thread_info(struct elf_note_info *info, const CPUArchState *env) 4306 { 4307 CPUState *cpu = env_cpu((CPUArchState *)env); 4308 TaskState *ts = (TaskState *)cpu->opaque; 4309 struct elf_thread_status *ets; 4310 4311 ets = g_malloc0(sizeof (*ets)); 4312 ets->num_notes = 1; /* only prstatus is dumped */ 4313 fill_prstatus(&ets->prstatus, ts, 0); 4314 elf_core_copy_regs(&ets->prstatus.pr_reg, env); 4315 fill_note(&ets->notes[0], "CORE", NT_PRSTATUS, sizeof (ets->prstatus), 4316 &ets->prstatus); 4317 4318 QTAILQ_INSERT_TAIL(&info->thread_list, ets, ets_link); 4319 4320 info->notes_size += note_size(&ets->notes[0]); 4321 } 4322 4323 static void init_note_info(struct elf_note_info *info) 4324 { 4325 /* Initialize the elf_note_info structure so that it is at 4326 * least safe to call free_note_info() on it. Must be 4327 * called before calling fill_note_info(). 4328 */ 4329 memset(info, 0, sizeof (*info)); 4330 QTAILQ_INIT(&info->thread_list); 4331 } 4332 4333 static int fill_note_info(struct elf_note_info *info, 4334 long signr, const CPUArchState *env) 4335 { 4336 #define NUMNOTES 3 4337 CPUState *cpu = env_cpu((CPUArchState *)env); 4338 TaskState *ts = (TaskState *)cpu->opaque; 4339 int i; 4340 4341 info->notes = g_new0(struct memelfnote, NUMNOTES); 4342 if (info->notes == NULL) 4343 return (-ENOMEM); 4344 info->prstatus = g_malloc0(sizeof (*info->prstatus)); 4345 if (info->prstatus == NULL) 4346 return (-ENOMEM); 4347 info->psinfo = g_malloc0(sizeof (*info->psinfo)); 4348 if (info->prstatus == NULL) 4349 return (-ENOMEM); 4350 4351 /* 4352 * First fill in status (and registers) of current thread 4353 * including process info & aux vector. 4354 */ 4355 fill_prstatus(info->prstatus, ts, signr); 4356 elf_core_copy_regs(&info->prstatus->pr_reg, env); 4357 fill_note(&info->notes[0], "CORE", NT_PRSTATUS, 4358 sizeof (*info->prstatus), info->prstatus); 4359 fill_psinfo(info->psinfo, ts); 4360 fill_note(&info->notes[1], "CORE", NT_PRPSINFO, 4361 sizeof (*info->psinfo), info->psinfo); 4362 fill_auxv_note(&info->notes[2], ts); 4363 info->numnote = 3; 4364 4365 info->notes_size = 0; 4366 for (i = 0; i < info->numnote; i++) 4367 info->notes_size += note_size(&info->notes[i]); 4368 4369 /* read and fill status of all threads */ 4370 WITH_QEMU_LOCK_GUARD(&qemu_cpu_list_lock) { 4371 CPU_FOREACH(cpu) { 4372 if (cpu == thread_cpu) { 4373 continue; 4374 } 4375 fill_thread_info(info, cpu->env_ptr); 4376 } 4377 } 4378 4379 return (0); 4380 } 4381 4382 static void free_note_info(struct elf_note_info *info) 4383 { 4384 struct elf_thread_status *ets; 4385 4386 while (!QTAILQ_EMPTY(&info->thread_list)) { 4387 ets = QTAILQ_FIRST(&info->thread_list); 4388 QTAILQ_REMOVE(&info->thread_list, ets, ets_link); 4389 g_free(ets); 4390 } 4391 4392 g_free(info->prstatus); 4393 g_free(info->psinfo); 4394 g_free(info->notes); 4395 } 4396 4397 static int write_note_info(struct elf_note_info *info, int fd) 4398 { 4399 struct elf_thread_status *ets; 4400 int i, error = 0; 4401 4402 /* write prstatus, psinfo and auxv for current thread */ 4403 for (i = 0; i < info->numnote; i++) 4404 if ((error = write_note(&info->notes[i], fd)) != 0) 4405 return (error); 4406 4407 /* write prstatus for each thread */ 4408 QTAILQ_FOREACH(ets, &info->thread_list, ets_link) { 4409 if ((error = write_note(&ets->notes[0], fd)) != 0) 4410 return (error); 4411 } 4412 4413 return (0); 4414 } 4415 4416 /* 4417 * Write out ELF coredump. 4418 * 4419 * See documentation of ELF object file format in: 4420 * http://www.caldera.com/developers/devspecs/gabi41.pdf 4421 * 4422 * Coredump format in linux is following: 4423 * 4424 * 0 +----------------------+ \ 4425 * | ELF header | ET_CORE | 4426 * +----------------------+ | 4427 * | ELF program headers | |--- headers 4428 * | - NOTE section | | 4429 * | - PT_LOAD sections | | 4430 * +----------------------+ / 4431 * | NOTEs: | 4432 * | - NT_PRSTATUS | 4433 * | - NT_PRSINFO | 4434 * | - NT_AUXV | 4435 * +----------------------+ <-- aligned to target page 4436 * | Process memory dump | 4437 * : : 4438 * . . 4439 * : : 4440 * | | 4441 * +----------------------+ 4442 * 4443 * NT_PRSTATUS -> struct elf_prstatus (per thread) 4444 * NT_PRSINFO -> struct elf_prpsinfo 4445 * NT_AUXV is array of { type, value } pairs (see fill_auxv_note()). 4446 * 4447 * Format follows System V format as close as possible. Current 4448 * version limitations are as follows: 4449 * - no floating point registers are dumped 4450 * 4451 * Function returns 0 in case of success, negative errno otherwise. 4452 * 4453 * TODO: make this work also during runtime: it should be 4454 * possible to force coredump from running process and then 4455 * continue processing. For example qemu could set up SIGUSR2 4456 * handler (provided that target process haven't registered 4457 * handler for that) that does the dump when signal is received. 4458 */ 4459 static int elf_core_dump(int signr, const CPUArchState *env) 4460 { 4461 const CPUState *cpu = env_cpu((CPUArchState *)env); 4462 const TaskState *ts = (const TaskState *)cpu->opaque; 4463 struct vm_area_struct *vma = NULL; 4464 g_autofree char *corefile = NULL; 4465 struct elf_note_info info; 4466 struct elfhdr elf; 4467 struct elf_phdr phdr; 4468 struct rlimit dumpsize; 4469 struct mm_struct *mm = NULL; 4470 off_t offset = 0, data_offset = 0; 4471 int segs = 0; 4472 int fd = -1; 4473 4474 init_note_info(&info); 4475 4476 errno = 0; 4477 getrlimit(RLIMIT_CORE, &dumpsize); 4478 if (dumpsize.rlim_cur == 0) 4479 return 0; 4480 4481 corefile = core_dump_filename(ts); 4482 4483 if ((fd = open(corefile, O_WRONLY | O_CREAT, 4484 S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH)) < 0) 4485 return (-errno); 4486 4487 /* 4488 * Walk through target process memory mappings and 4489 * set up structure containing this information. After 4490 * this point vma_xxx functions can be used. 4491 */ 4492 if ((mm = vma_init()) == NULL) 4493 goto out; 4494 4495 walk_memory_regions(mm, vma_walker); 4496 segs = vma_get_mapping_count(mm); 4497 4498 /* 4499 * Construct valid coredump ELF header. We also 4500 * add one more segment for notes. 4501 */ 4502 fill_elf_header(&elf, segs + 1, ELF_MACHINE, 0); 4503 if (dump_write(fd, &elf, sizeof (elf)) != 0) 4504 goto out; 4505 4506 /* fill in the in-memory version of notes */ 4507 if (fill_note_info(&info, signr, env) < 0) 4508 goto out; 4509 4510 offset += sizeof (elf); /* elf header */ 4511 offset += (segs + 1) * sizeof (struct elf_phdr); /* program headers */ 4512 4513 /* write out notes program header */ 4514 fill_elf_note_phdr(&phdr, info.notes_size, offset); 4515 4516 offset += info.notes_size; 4517 if (dump_write(fd, &phdr, sizeof (phdr)) != 0) 4518 goto out; 4519 4520 /* 4521 * ELF specification wants data to start at page boundary so 4522 * we align it here. 4523 */ 4524 data_offset = offset = roundup(offset, ELF_EXEC_PAGESIZE); 4525 4526 /* 4527 * Write program headers for memory regions mapped in 4528 * the target process. 4529 */ 4530 for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) { 4531 (void) memset(&phdr, 0, sizeof (phdr)); 4532 4533 phdr.p_type = PT_LOAD; 4534 phdr.p_offset = offset; 4535 phdr.p_vaddr = vma->vma_start; 4536 phdr.p_paddr = 0; 4537 phdr.p_filesz = vma_dump_size(vma); 4538 offset += phdr.p_filesz; 4539 phdr.p_memsz = vma->vma_end - vma->vma_start; 4540 phdr.p_flags = vma->vma_flags & PROT_READ ? PF_R : 0; 4541 if (vma->vma_flags & PROT_WRITE) 4542 phdr.p_flags |= PF_W; 4543 if (vma->vma_flags & PROT_EXEC) 4544 phdr.p_flags |= PF_X; 4545 phdr.p_align = ELF_EXEC_PAGESIZE; 4546 4547 bswap_phdr(&phdr, 1); 4548 if (dump_write(fd, &phdr, sizeof(phdr)) != 0) { 4549 goto out; 4550 } 4551 } 4552 4553 /* 4554 * Next we write notes just after program headers. No 4555 * alignment needed here. 4556 */ 4557 if (write_note_info(&info, fd) < 0) 4558 goto out; 4559 4560 /* align data to page boundary */ 4561 if (lseek(fd, data_offset, SEEK_SET) != data_offset) 4562 goto out; 4563 4564 /* 4565 * Finally we can dump process memory into corefile as well. 4566 */ 4567 for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) { 4568 abi_ulong addr; 4569 abi_ulong end; 4570 4571 end = vma->vma_start + vma_dump_size(vma); 4572 4573 for (addr = vma->vma_start; addr < end; 4574 addr += TARGET_PAGE_SIZE) { 4575 char page[TARGET_PAGE_SIZE]; 4576 int error; 4577 4578 /* 4579 * Read in page from target process memory and 4580 * write it to coredump file. 4581 */ 4582 error = copy_from_user(page, addr, sizeof (page)); 4583 if (error != 0) { 4584 (void) fprintf(stderr, "unable to dump " TARGET_ABI_FMT_lx "\n", 4585 addr); 4586 errno = -error; 4587 goto out; 4588 } 4589 if (dump_write(fd, page, TARGET_PAGE_SIZE) < 0) 4590 goto out; 4591 } 4592 } 4593 4594 out: 4595 free_note_info(&info); 4596 if (mm != NULL) 4597 vma_delete(mm); 4598 (void) close(fd); 4599 4600 if (errno != 0) 4601 return (-errno); 4602 return (0); 4603 } 4604 #endif /* USE_ELF_CORE_DUMP */ 4605 4606 void do_init_thread(struct target_pt_regs *regs, struct image_info *infop) 4607 { 4608 init_thread(regs, infop); 4609 } 4610