1 /* This is the Linux kernel elf-loading code, ported into user space */ 2 #include "qemu/osdep.h" 3 #include <sys/param.h> 4 5 #include <sys/resource.h> 6 #include <sys/shm.h> 7 8 #include "qemu.h" 9 #include "user-internals.h" 10 #include "signal-common.h" 11 #include "loader.h" 12 #include "user-mmap.h" 13 #include "disas/disas.h" 14 #include "qemu/bitops.h" 15 #include "qemu/path.h" 16 #include "qemu/queue.h" 17 #include "qemu/guest-random.h" 18 #include "qemu/units.h" 19 #include "qemu/selfmap.h" 20 #include "qapi/error.h" 21 #include "qemu/error-report.h" 22 #include "target_signal.h" 23 #include "accel/tcg/debuginfo.h" 24 25 #ifdef _ARCH_PPC64 26 #undef ARCH_DLINFO 27 #undef ELF_PLATFORM 28 #undef ELF_HWCAP 29 #undef ELF_HWCAP2 30 #undef ELF_CLASS 31 #undef ELF_DATA 32 #undef ELF_ARCH 33 #endif 34 35 #define ELF_OSABI ELFOSABI_SYSV 36 37 /* from personality.h */ 38 39 /* 40 * Flags for bug emulation. 41 * 42 * These occupy the top three bytes. 43 */ 44 enum { 45 ADDR_NO_RANDOMIZE = 0x0040000, /* disable randomization of VA space */ 46 FDPIC_FUNCPTRS = 0x0080000, /* userspace function ptrs point to 47 descriptors (signal handling) */ 48 MMAP_PAGE_ZERO = 0x0100000, 49 ADDR_COMPAT_LAYOUT = 0x0200000, 50 READ_IMPLIES_EXEC = 0x0400000, 51 ADDR_LIMIT_32BIT = 0x0800000, 52 SHORT_INODE = 0x1000000, 53 WHOLE_SECONDS = 0x2000000, 54 STICKY_TIMEOUTS = 0x4000000, 55 ADDR_LIMIT_3GB = 0x8000000, 56 }; 57 58 /* 59 * Personality types. 60 * 61 * These go in the low byte. Avoid using the top bit, it will 62 * conflict with error returns. 63 */ 64 enum { 65 PER_LINUX = 0x0000, 66 PER_LINUX_32BIT = 0x0000 | ADDR_LIMIT_32BIT, 67 PER_LINUX_FDPIC = 0x0000 | FDPIC_FUNCPTRS, 68 PER_SVR4 = 0x0001 | STICKY_TIMEOUTS | MMAP_PAGE_ZERO, 69 PER_SVR3 = 0x0002 | STICKY_TIMEOUTS | SHORT_INODE, 70 PER_SCOSVR3 = 0x0003 | STICKY_TIMEOUTS | WHOLE_SECONDS | SHORT_INODE, 71 PER_OSR5 = 0x0003 | STICKY_TIMEOUTS | WHOLE_SECONDS, 72 PER_WYSEV386 = 0x0004 | STICKY_TIMEOUTS | SHORT_INODE, 73 PER_ISCR4 = 0x0005 | STICKY_TIMEOUTS, 74 PER_BSD = 0x0006, 75 PER_SUNOS = 0x0006 | STICKY_TIMEOUTS, 76 PER_XENIX = 0x0007 | STICKY_TIMEOUTS | SHORT_INODE, 77 PER_LINUX32 = 0x0008, 78 PER_LINUX32_3GB = 0x0008 | ADDR_LIMIT_3GB, 79 PER_IRIX32 = 0x0009 | STICKY_TIMEOUTS,/* IRIX5 32-bit */ 80 PER_IRIXN32 = 0x000a | STICKY_TIMEOUTS,/* IRIX6 new 32-bit */ 81 PER_IRIX64 = 0x000b | STICKY_TIMEOUTS,/* IRIX6 64-bit */ 82 PER_RISCOS = 0x000c, 83 PER_SOLARIS = 0x000d | STICKY_TIMEOUTS, 84 PER_UW7 = 0x000e | STICKY_TIMEOUTS | MMAP_PAGE_ZERO, 85 PER_OSF4 = 0x000f, /* OSF/1 v4 */ 86 PER_HPUX = 0x0010, 87 PER_MASK = 0x00ff, 88 }; 89 90 /* 91 * Return the base personality without flags. 92 */ 93 #define personality(pers) (pers & PER_MASK) 94 95 int info_is_fdpic(struct image_info *info) 96 { 97 return info->personality == PER_LINUX_FDPIC; 98 } 99 100 /* this flag is uneffective under linux too, should be deleted */ 101 #ifndef MAP_DENYWRITE 102 #define MAP_DENYWRITE 0 103 #endif 104 105 /* should probably go in elf.h */ 106 #ifndef ELIBBAD 107 #define ELIBBAD 80 108 #endif 109 110 #if TARGET_BIG_ENDIAN 111 #define ELF_DATA ELFDATA2MSB 112 #else 113 #define ELF_DATA ELFDATA2LSB 114 #endif 115 116 #ifdef TARGET_ABI_MIPSN32 117 typedef abi_ullong target_elf_greg_t; 118 #define tswapreg(ptr) tswap64(ptr) 119 #else 120 typedef abi_ulong target_elf_greg_t; 121 #define tswapreg(ptr) tswapal(ptr) 122 #endif 123 124 #ifdef USE_UID16 125 typedef abi_ushort target_uid_t; 126 typedef abi_ushort target_gid_t; 127 #else 128 typedef abi_uint target_uid_t; 129 typedef abi_uint target_gid_t; 130 #endif 131 typedef abi_int target_pid_t; 132 133 #ifdef TARGET_I386 134 135 #define ELF_HWCAP get_elf_hwcap() 136 137 static uint32_t get_elf_hwcap(void) 138 { 139 X86CPU *cpu = X86_CPU(thread_cpu); 140 141 return cpu->env.features[FEAT_1_EDX]; 142 } 143 144 #ifdef TARGET_X86_64 145 #define ELF_START_MMAP 0x2aaaaab000ULL 146 147 #define ELF_CLASS ELFCLASS64 148 #define ELF_ARCH EM_X86_64 149 150 #define ELF_PLATFORM "x86_64" 151 152 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop) 153 { 154 regs->rax = 0; 155 regs->rsp = infop->start_stack; 156 regs->rip = infop->entry; 157 } 158 159 #define ELF_NREG 27 160 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 161 162 /* 163 * Note that ELF_NREG should be 29 as there should be place for 164 * TRAPNO and ERR "registers" as well but linux doesn't dump 165 * those. 166 * 167 * See linux kernel: arch/x86/include/asm/elf.h 168 */ 169 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUX86State *env) 170 { 171 (*regs)[0] = tswapreg(env->regs[15]); 172 (*regs)[1] = tswapreg(env->regs[14]); 173 (*regs)[2] = tswapreg(env->regs[13]); 174 (*regs)[3] = tswapreg(env->regs[12]); 175 (*regs)[4] = tswapreg(env->regs[R_EBP]); 176 (*regs)[5] = tswapreg(env->regs[R_EBX]); 177 (*regs)[6] = tswapreg(env->regs[11]); 178 (*regs)[7] = tswapreg(env->regs[10]); 179 (*regs)[8] = tswapreg(env->regs[9]); 180 (*regs)[9] = tswapreg(env->regs[8]); 181 (*regs)[10] = tswapreg(env->regs[R_EAX]); 182 (*regs)[11] = tswapreg(env->regs[R_ECX]); 183 (*regs)[12] = tswapreg(env->regs[R_EDX]); 184 (*regs)[13] = tswapreg(env->regs[R_ESI]); 185 (*regs)[14] = tswapreg(env->regs[R_EDI]); 186 (*regs)[15] = tswapreg(env->regs[R_EAX]); /* XXX */ 187 (*regs)[16] = tswapreg(env->eip); 188 (*regs)[17] = tswapreg(env->segs[R_CS].selector & 0xffff); 189 (*regs)[18] = tswapreg(env->eflags); 190 (*regs)[19] = tswapreg(env->regs[R_ESP]); 191 (*regs)[20] = tswapreg(env->segs[R_SS].selector & 0xffff); 192 (*regs)[21] = tswapreg(env->segs[R_FS].selector & 0xffff); 193 (*regs)[22] = tswapreg(env->segs[R_GS].selector & 0xffff); 194 (*regs)[23] = tswapreg(env->segs[R_DS].selector & 0xffff); 195 (*regs)[24] = tswapreg(env->segs[R_ES].selector & 0xffff); 196 (*regs)[25] = tswapreg(env->segs[R_FS].selector & 0xffff); 197 (*regs)[26] = tswapreg(env->segs[R_GS].selector & 0xffff); 198 } 199 200 #if ULONG_MAX > UINT32_MAX 201 #define INIT_GUEST_COMMPAGE 202 static bool init_guest_commpage(void) 203 { 204 /* 205 * The vsyscall page is at a high negative address aka kernel space, 206 * which means that we cannot actually allocate it with target_mmap. 207 * We still should be able to use page_set_flags, unless the user 208 * has specified -R reserved_va, which would trigger an assert(). 209 */ 210 if (reserved_va != 0 && 211 TARGET_VSYSCALL_PAGE + TARGET_PAGE_SIZE - 1 > reserved_va) { 212 error_report("Cannot allocate vsyscall page"); 213 exit(EXIT_FAILURE); 214 } 215 page_set_flags(TARGET_VSYSCALL_PAGE, 216 TARGET_VSYSCALL_PAGE | ~TARGET_PAGE_MASK, 217 PAGE_EXEC | PAGE_VALID); 218 return true; 219 } 220 #endif 221 #else 222 223 #define ELF_START_MMAP 0x80000000 224 225 /* 226 * This is used to ensure we don't load something for the wrong architecture. 227 */ 228 #define elf_check_arch(x) ( ((x) == EM_386) || ((x) == EM_486) ) 229 230 /* 231 * These are used to set parameters in the core dumps. 232 */ 233 #define ELF_CLASS ELFCLASS32 234 #define ELF_ARCH EM_386 235 236 #define ELF_PLATFORM get_elf_platform() 237 #define EXSTACK_DEFAULT true 238 239 static const char *get_elf_platform(void) 240 { 241 static char elf_platform[] = "i386"; 242 int family = object_property_get_int(OBJECT(thread_cpu), "family", NULL); 243 if (family > 6) { 244 family = 6; 245 } 246 if (family >= 3) { 247 elf_platform[1] = '0' + family; 248 } 249 return elf_platform; 250 } 251 252 static inline void init_thread(struct target_pt_regs *regs, 253 struct image_info *infop) 254 { 255 regs->esp = infop->start_stack; 256 regs->eip = infop->entry; 257 258 /* SVR4/i386 ABI (pages 3-31, 3-32) says that when the program 259 starts %edx contains a pointer to a function which might be 260 registered using `atexit'. This provides a mean for the 261 dynamic linker to call DT_FINI functions for shared libraries 262 that have been loaded before the code runs. 263 264 A value of 0 tells we have no such handler. */ 265 regs->edx = 0; 266 } 267 268 #define ELF_NREG 17 269 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 270 271 /* 272 * Note that ELF_NREG should be 19 as there should be place for 273 * TRAPNO and ERR "registers" as well but linux doesn't dump 274 * those. 275 * 276 * See linux kernel: arch/x86/include/asm/elf.h 277 */ 278 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUX86State *env) 279 { 280 (*regs)[0] = tswapreg(env->regs[R_EBX]); 281 (*regs)[1] = tswapreg(env->regs[R_ECX]); 282 (*regs)[2] = tswapreg(env->regs[R_EDX]); 283 (*regs)[3] = tswapreg(env->regs[R_ESI]); 284 (*regs)[4] = tswapreg(env->regs[R_EDI]); 285 (*regs)[5] = tswapreg(env->regs[R_EBP]); 286 (*regs)[6] = tswapreg(env->regs[R_EAX]); 287 (*regs)[7] = tswapreg(env->segs[R_DS].selector & 0xffff); 288 (*regs)[8] = tswapreg(env->segs[R_ES].selector & 0xffff); 289 (*regs)[9] = tswapreg(env->segs[R_FS].selector & 0xffff); 290 (*regs)[10] = tswapreg(env->segs[R_GS].selector & 0xffff); 291 (*regs)[11] = tswapreg(env->regs[R_EAX]); /* XXX */ 292 (*regs)[12] = tswapreg(env->eip); 293 (*regs)[13] = tswapreg(env->segs[R_CS].selector & 0xffff); 294 (*regs)[14] = tswapreg(env->eflags); 295 (*regs)[15] = tswapreg(env->regs[R_ESP]); 296 (*regs)[16] = tswapreg(env->segs[R_SS].selector & 0xffff); 297 } 298 #endif 299 300 #define USE_ELF_CORE_DUMP 301 #define ELF_EXEC_PAGESIZE 4096 302 303 #endif 304 305 #ifdef TARGET_ARM 306 307 #ifndef TARGET_AARCH64 308 /* 32 bit ARM definitions */ 309 310 #define ELF_START_MMAP 0x80000000 311 312 #define ELF_ARCH EM_ARM 313 #define ELF_CLASS ELFCLASS32 314 #define EXSTACK_DEFAULT true 315 316 static inline void init_thread(struct target_pt_regs *regs, 317 struct image_info *infop) 318 { 319 abi_long stack = infop->start_stack; 320 memset(regs, 0, sizeof(*regs)); 321 322 regs->uregs[16] = ARM_CPU_MODE_USR; 323 if (infop->entry & 1) { 324 regs->uregs[16] |= CPSR_T; 325 } 326 regs->uregs[15] = infop->entry & 0xfffffffe; 327 regs->uregs[13] = infop->start_stack; 328 /* FIXME - what to for failure of get_user()? */ 329 get_user_ual(regs->uregs[2], stack + 8); /* envp */ 330 get_user_ual(regs->uregs[1], stack + 4); /* envp */ 331 /* XXX: it seems that r0 is zeroed after ! */ 332 regs->uregs[0] = 0; 333 /* For uClinux PIC binaries. */ 334 /* XXX: Linux does this only on ARM with no MMU (do we care ?) */ 335 regs->uregs[10] = infop->start_data; 336 337 /* Support ARM FDPIC. */ 338 if (info_is_fdpic(infop)) { 339 /* As described in the ABI document, r7 points to the loadmap info 340 * prepared by the kernel. If an interpreter is needed, r8 points 341 * to the interpreter loadmap and r9 points to the interpreter 342 * PT_DYNAMIC info. If no interpreter is needed, r8 is zero, and 343 * r9 points to the main program PT_DYNAMIC info. 344 */ 345 regs->uregs[7] = infop->loadmap_addr; 346 if (infop->interpreter_loadmap_addr) { 347 /* Executable is dynamically loaded. */ 348 regs->uregs[8] = infop->interpreter_loadmap_addr; 349 regs->uregs[9] = infop->interpreter_pt_dynamic_addr; 350 } else { 351 regs->uregs[8] = 0; 352 regs->uregs[9] = infop->pt_dynamic_addr; 353 } 354 } 355 } 356 357 #define ELF_NREG 18 358 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 359 360 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUARMState *env) 361 { 362 (*regs)[0] = tswapreg(env->regs[0]); 363 (*regs)[1] = tswapreg(env->regs[1]); 364 (*regs)[2] = tswapreg(env->regs[2]); 365 (*regs)[3] = tswapreg(env->regs[3]); 366 (*regs)[4] = tswapreg(env->regs[4]); 367 (*regs)[5] = tswapreg(env->regs[5]); 368 (*regs)[6] = tswapreg(env->regs[6]); 369 (*regs)[7] = tswapreg(env->regs[7]); 370 (*regs)[8] = tswapreg(env->regs[8]); 371 (*regs)[9] = tswapreg(env->regs[9]); 372 (*regs)[10] = tswapreg(env->regs[10]); 373 (*regs)[11] = tswapreg(env->regs[11]); 374 (*regs)[12] = tswapreg(env->regs[12]); 375 (*regs)[13] = tswapreg(env->regs[13]); 376 (*regs)[14] = tswapreg(env->regs[14]); 377 (*regs)[15] = tswapreg(env->regs[15]); 378 379 (*regs)[16] = tswapreg(cpsr_read((CPUARMState *)env)); 380 (*regs)[17] = tswapreg(env->regs[0]); /* XXX */ 381 } 382 383 #define USE_ELF_CORE_DUMP 384 #define ELF_EXEC_PAGESIZE 4096 385 386 enum 387 { 388 ARM_HWCAP_ARM_SWP = 1 << 0, 389 ARM_HWCAP_ARM_HALF = 1 << 1, 390 ARM_HWCAP_ARM_THUMB = 1 << 2, 391 ARM_HWCAP_ARM_26BIT = 1 << 3, 392 ARM_HWCAP_ARM_FAST_MULT = 1 << 4, 393 ARM_HWCAP_ARM_FPA = 1 << 5, 394 ARM_HWCAP_ARM_VFP = 1 << 6, 395 ARM_HWCAP_ARM_EDSP = 1 << 7, 396 ARM_HWCAP_ARM_JAVA = 1 << 8, 397 ARM_HWCAP_ARM_IWMMXT = 1 << 9, 398 ARM_HWCAP_ARM_CRUNCH = 1 << 10, 399 ARM_HWCAP_ARM_THUMBEE = 1 << 11, 400 ARM_HWCAP_ARM_NEON = 1 << 12, 401 ARM_HWCAP_ARM_VFPv3 = 1 << 13, 402 ARM_HWCAP_ARM_VFPv3D16 = 1 << 14, 403 ARM_HWCAP_ARM_TLS = 1 << 15, 404 ARM_HWCAP_ARM_VFPv4 = 1 << 16, 405 ARM_HWCAP_ARM_IDIVA = 1 << 17, 406 ARM_HWCAP_ARM_IDIVT = 1 << 18, 407 ARM_HWCAP_ARM_VFPD32 = 1 << 19, 408 ARM_HWCAP_ARM_LPAE = 1 << 20, 409 ARM_HWCAP_ARM_EVTSTRM = 1 << 21, 410 }; 411 412 enum { 413 ARM_HWCAP2_ARM_AES = 1 << 0, 414 ARM_HWCAP2_ARM_PMULL = 1 << 1, 415 ARM_HWCAP2_ARM_SHA1 = 1 << 2, 416 ARM_HWCAP2_ARM_SHA2 = 1 << 3, 417 ARM_HWCAP2_ARM_CRC32 = 1 << 4, 418 }; 419 420 /* The commpage only exists for 32 bit kernels */ 421 422 #define HI_COMMPAGE (intptr_t)0xffff0f00u 423 424 static bool init_guest_commpage(void) 425 { 426 ARMCPU *cpu = ARM_CPU(thread_cpu); 427 abi_ptr want = HI_COMMPAGE & TARGET_PAGE_MASK; 428 abi_ptr addr; 429 430 /* 431 * M-profile allocates maximum of 2GB address space, so can never 432 * allocate the commpage. Skip it. 433 */ 434 if (arm_feature(&cpu->env, ARM_FEATURE_M)) { 435 return true; 436 } 437 438 /* 439 * If reserved_va does not cover the commpage, we get an assert 440 * in page_set_flags. Produce an intelligent error instead. 441 */ 442 if (reserved_va != 0 && want + TARGET_PAGE_SIZE - 1 > reserved_va) { 443 error_report("Allocating guest commpage: -R 0x%" PRIx64 " too small", 444 (uint64_t)reserved_va + 1); 445 exit(EXIT_FAILURE); 446 } 447 448 addr = target_mmap(want, TARGET_PAGE_SIZE, PROT_READ | PROT_WRITE, 449 MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0); 450 451 if (addr == -1) { 452 perror("Allocating guest commpage"); 453 exit(EXIT_FAILURE); 454 } 455 if (addr != want) { 456 return false; 457 } 458 459 /* Set kernel helper versions; rest of page is 0. */ 460 put_user_u32(5, 0xffff0ffcu); 461 462 if (target_mprotect(addr, qemu_host_page_size, PROT_READ | PROT_EXEC)) { 463 perror("Protecting guest commpage"); 464 exit(EXIT_FAILURE); 465 } 466 return true; 467 } 468 469 #define ELF_HWCAP get_elf_hwcap() 470 #define ELF_HWCAP2 get_elf_hwcap2() 471 472 static uint32_t get_elf_hwcap(void) 473 { 474 ARMCPU *cpu = ARM_CPU(thread_cpu); 475 uint32_t hwcaps = 0; 476 477 hwcaps |= ARM_HWCAP_ARM_SWP; 478 hwcaps |= ARM_HWCAP_ARM_HALF; 479 hwcaps |= ARM_HWCAP_ARM_THUMB; 480 hwcaps |= ARM_HWCAP_ARM_FAST_MULT; 481 482 /* probe for the extra features */ 483 #define GET_FEATURE(feat, hwcap) \ 484 do { if (arm_feature(&cpu->env, feat)) { hwcaps |= hwcap; } } while (0) 485 486 #define GET_FEATURE_ID(feat, hwcap) \ 487 do { if (cpu_isar_feature(feat, cpu)) { hwcaps |= hwcap; } } while (0) 488 489 /* EDSP is in v5TE and above, but all our v5 CPUs are v5TE */ 490 GET_FEATURE(ARM_FEATURE_V5, ARM_HWCAP_ARM_EDSP); 491 GET_FEATURE(ARM_FEATURE_IWMMXT, ARM_HWCAP_ARM_IWMMXT); 492 GET_FEATURE(ARM_FEATURE_THUMB2EE, ARM_HWCAP_ARM_THUMBEE); 493 GET_FEATURE(ARM_FEATURE_NEON, ARM_HWCAP_ARM_NEON); 494 GET_FEATURE(ARM_FEATURE_V6K, ARM_HWCAP_ARM_TLS); 495 GET_FEATURE(ARM_FEATURE_LPAE, ARM_HWCAP_ARM_LPAE); 496 GET_FEATURE_ID(aa32_arm_div, ARM_HWCAP_ARM_IDIVA); 497 GET_FEATURE_ID(aa32_thumb_div, ARM_HWCAP_ARM_IDIVT); 498 GET_FEATURE_ID(aa32_vfp, ARM_HWCAP_ARM_VFP); 499 500 if (cpu_isar_feature(aa32_fpsp_v3, cpu) || 501 cpu_isar_feature(aa32_fpdp_v3, cpu)) { 502 hwcaps |= ARM_HWCAP_ARM_VFPv3; 503 if (cpu_isar_feature(aa32_simd_r32, cpu)) { 504 hwcaps |= ARM_HWCAP_ARM_VFPD32; 505 } else { 506 hwcaps |= ARM_HWCAP_ARM_VFPv3D16; 507 } 508 } 509 GET_FEATURE_ID(aa32_simdfmac, ARM_HWCAP_ARM_VFPv4); 510 511 return hwcaps; 512 } 513 514 static uint32_t get_elf_hwcap2(void) 515 { 516 ARMCPU *cpu = ARM_CPU(thread_cpu); 517 uint32_t hwcaps = 0; 518 519 GET_FEATURE_ID(aa32_aes, ARM_HWCAP2_ARM_AES); 520 GET_FEATURE_ID(aa32_pmull, ARM_HWCAP2_ARM_PMULL); 521 GET_FEATURE_ID(aa32_sha1, ARM_HWCAP2_ARM_SHA1); 522 GET_FEATURE_ID(aa32_sha2, ARM_HWCAP2_ARM_SHA2); 523 GET_FEATURE_ID(aa32_crc32, ARM_HWCAP2_ARM_CRC32); 524 return hwcaps; 525 } 526 527 #undef GET_FEATURE 528 #undef GET_FEATURE_ID 529 530 #define ELF_PLATFORM get_elf_platform() 531 532 static const char *get_elf_platform(void) 533 { 534 CPUARMState *env = thread_cpu->env_ptr; 535 536 #if TARGET_BIG_ENDIAN 537 # define END "b" 538 #else 539 # define END "l" 540 #endif 541 542 if (arm_feature(env, ARM_FEATURE_V8)) { 543 return "v8" END; 544 } else if (arm_feature(env, ARM_FEATURE_V7)) { 545 if (arm_feature(env, ARM_FEATURE_M)) { 546 return "v7m" END; 547 } else { 548 return "v7" END; 549 } 550 } else if (arm_feature(env, ARM_FEATURE_V6)) { 551 return "v6" END; 552 } else if (arm_feature(env, ARM_FEATURE_V5)) { 553 return "v5" END; 554 } else { 555 return "v4" END; 556 } 557 558 #undef END 559 } 560 561 #else 562 /* 64 bit ARM definitions */ 563 #define ELF_START_MMAP 0x80000000 564 565 #define ELF_ARCH EM_AARCH64 566 #define ELF_CLASS ELFCLASS64 567 #if TARGET_BIG_ENDIAN 568 # define ELF_PLATFORM "aarch64_be" 569 #else 570 # define ELF_PLATFORM "aarch64" 571 #endif 572 573 static inline void init_thread(struct target_pt_regs *regs, 574 struct image_info *infop) 575 { 576 abi_long stack = infop->start_stack; 577 memset(regs, 0, sizeof(*regs)); 578 579 regs->pc = infop->entry & ~0x3ULL; 580 regs->sp = stack; 581 } 582 583 #define ELF_NREG 34 584 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 585 586 static void elf_core_copy_regs(target_elf_gregset_t *regs, 587 const CPUARMState *env) 588 { 589 int i; 590 591 for (i = 0; i < 32; i++) { 592 (*regs)[i] = tswapreg(env->xregs[i]); 593 } 594 (*regs)[32] = tswapreg(env->pc); 595 (*regs)[33] = tswapreg(pstate_read((CPUARMState *)env)); 596 } 597 598 #define USE_ELF_CORE_DUMP 599 #define ELF_EXEC_PAGESIZE 4096 600 601 enum { 602 ARM_HWCAP_A64_FP = 1 << 0, 603 ARM_HWCAP_A64_ASIMD = 1 << 1, 604 ARM_HWCAP_A64_EVTSTRM = 1 << 2, 605 ARM_HWCAP_A64_AES = 1 << 3, 606 ARM_HWCAP_A64_PMULL = 1 << 4, 607 ARM_HWCAP_A64_SHA1 = 1 << 5, 608 ARM_HWCAP_A64_SHA2 = 1 << 6, 609 ARM_HWCAP_A64_CRC32 = 1 << 7, 610 ARM_HWCAP_A64_ATOMICS = 1 << 8, 611 ARM_HWCAP_A64_FPHP = 1 << 9, 612 ARM_HWCAP_A64_ASIMDHP = 1 << 10, 613 ARM_HWCAP_A64_CPUID = 1 << 11, 614 ARM_HWCAP_A64_ASIMDRDM = 1 << 12, 615 ARM_HWCAP_A64_JSCVT = 1 << 13, 616 ARM_HWCAP_A64_FCMA = 1 << 14, 617 ARM_HWCAP_A64_LRCPC = 1 << 15, 618 ARM_HWCAP_A64_DCPOP = 1 << 16, 619 ARM_HWCAP_A64_SHA3 = 1 << 17, 620 ARM_HWCAP_A64_SM3 = 1 << 18, 621 ARM_HWCAP_A64_SM4 = 1 << 19, 622 ARM_HWCAP_A64_ASIMDDP = 1 << 20, 623 ARM_HWCAP_A64_SHA512 = 1 << 21, 624 ARM_HWCAP_A64_SVE = 1 << 22, 625 ARM_HWCAP_A64_ASIMDFHM = 1 << 23, 626 ARM_HWCAP_A64_DIT = 1 << 24, 627 ARM_HWCAP_A64_USCAT = 1 << 25, 628 ARM_HWCAP_A64_ILRCPC = 1 << 26, 629 ARM_HWCAP_A64_FLAGM = 1 << 27, 630 ARM_HWCAP_A64_SSBS = 1 << 28, 631 ARM_HWCAP_A64_SB = 1 << 29, 632 ARM_HWCAP_A64_PACA = 1 << 30, 633 ARM_HWCAP_A64_PACG = 1UL << 31, 634 635 ARM_HWCAP2_A64_DCPODP = 1 << 0, 636 ARM_HWCAP2_A64_SVE2 = 1 << 1, 637 ARM_HWCAP2_A64_SVEAES = 1 << 2, 638 ARM_HWCAP2_A64_SVEPMULL = 1 << 3, 639 ARM_HWCAP2_A64_SVEBITPERM = 1 << 4, 640 ARM_HWCAP2_A64_SVESHA3 = 1 << 5, 641 ARM_HWCAP2_A64_SVESM4 = 1 << 6, 642 ARM_HWCAP2_A64_FLAGM2 = 1 << 7, 643 ARM_HWCAP2_A64_FRINT = 1 << 8, 644 ARM_HWCAP2_A64_SVEI8MM = 1 << 9, 645 ARM_HWCAP2_A64_SVEF32MM = 1 << 10, 646 ARM_HWCAP2_A64_SVEF64MM = 1 << 11, 647 ARM_HWCAP2_A64_SVEBF16 = 1 << 12, 648 ARM_HWCAP2_A64_I8MM = 1 << 13, 649 ARM_HWCAP2_A64_BF16 = 1 << 14, 650 ARM_HWCAP2_A64_DGH = 1 << 15, 651 ARM_HWCAP2_A64_RNG = 1 << 16, 652 ARM_HWCAP2_A64_BTI = 1 << 17, 653 ARM_HWCAP2_A64_MTE = 1 << 18, 654 ARM_HWCAP2_A64_ECV = 1 << 19, 655 ARM_HWCAP2_A64_AFP = 1 << 20, 656 ARM_HWCAP2_A64_RPRES = 1 << 21, 657 ARM_HWCAP2_A64_MTE3 = 1 << 22, 658 ARM_HWCAP2_A64_SME = 1 << 23, 659 ARM_HWCAP2_A64_SME_I16I64 = 1 << 24, 660 ARM_HWCAP2_A64_SME_F64F64 = 1 << 25, 661 ARM_HWCAP2_A64_SME_I8I32 = 1 << 26, 662 ARM_HWCAP2_A64_SME_F16F32 = 1 << 27, 663 ARM_HWCAP2_A64_SME_B16F32 = 1 << 28, 664 ARM_HWCAP2_A64_SME_F32F32 = 1 << 29, 665 ARM_HWCAP2_A64_SME_FA64 = 1 << 30, 666 }; 667 668 #define ELF_HWCAP get_elf_hwcap() 669 #define ELF_HWCAP2 get_elf_hwcap2() 670 671 #define GET_FEATURE_ID(feat, hwcap) \ 672 do { if (cpu_isar_feature(feat, cpu)) { hwcaps |= hwcap; } } while (0) 673 674 static uint32_t get_elf_hwcap(void) 675 { 676 ARMCPU *cpu = ARM_CPU(thread_cpu); 677 uint32_t hwcaps = 0; 678 679 hwcaps |= ARM_HWCAP_A64_FP; 680 hwcaps |= ARM_HWCAP_A64_ASIMD; 681 hwcaps |= ARM_HWCAP_A64_CPUID; 682 683 /* probe for the extra features */ 684 685 GET_FEATURE_ID(aa64_aes, ARM_HWCAP_A64_AES); 686 GET_FEATURE_ID(aa64_pmull, ARM_HWCAP_A64_PMULL); 687 GET_FEATURE_ID(aa64_sha1, ARM_HWCAP_A64_SHA1); 688 GET_FEATURE_ID(aa64_sha256, ARM_HWCAP_A64_SHA2); 689 GET_FEATURE_ID(aa64_sha512, ARM_HWCAP_A64_SHA512); 690 GET_FEATURE_ID(aa64_crc32, ARM_HWCAP_A64_CRC32); 691 GET_FEATURE_ID(aa64_sha3, ARM_HWCAP_A64_SHA3); 692 GET_FEATURE_ID(aa64_sm3, ARM_HWCAP_A64_SM3); 693 GET_FEATURE_ID(aa64_sm4, ARM_HWCAP_A64_SM4); 694 GET_FEATURE_ID(aa64_fp16, ARM_HWCAP_A64_FPHP | ARM_HWCAP_A64_ASIMDHP); 695 GET_FEATURE_ID(aa64_atomics, ARM_HWCAP_A64_ATOMICS); 696 GET_FEATURE_ID(aa64_rdm, ARM_HWCAP_A64_ASIMDRDM); 697 GET_FEATURE_ID(aa64_dp, ARM_HWCAP_A64_ASIMDDP); 698 GET_FEATURE_ID(aa64_fcma, ARM_HWCAP_A64_FCMA); 699 GET_FEATURE_ID(aa64_sve, ARM_HWCAP_A64_SVE); 700 GET_FEATURE_ID(aa64_pauth, ARM_HWCAP_A64_PACA | ARM_HWCAP_A64_PACG); 701 GET_FEATURE_ID(aa64_fhm, ARM_HWCAP_A64_ASIMDFHM); 702 GET_FEATURE_ID(aa64_jscvt, ARM_HWCAP_A64_JSCVT); 703 GET_FEATURE_ID(aa64_sb, ARM_HWCAP_A64_SB); 704 GET_FEATURE_ID(aa64_condm_4, ARM_HWCAP_A64_FLAGM); 705 GET_FEATURE_ID(aa64_dcpop, ARM_HWCAP_A64_DCPOP); 706 GET_FEATURE_ID(aa64_rcpc_8_3, ARM_HWCAP_A64_LRCPC); 707 GET_FEATURE_ID(aa64_rcpc_8_4, ARM_HWCAP_A64_ILRCPC); 708 709 return hwcaps; 710 } 711 712 static uint32_t get_elf_hwcap2(void) 713 { 714 ARMCPU *cpu = ARM_CPU(thread_cpu); 715 uint32_t hwcaps = 0; 716 717 GET_FEATURE_ID(aa64_dcpodp, ARM_HWCAP2_A64_DCPODP); 718 GET_FEATURE_ID(aa64_sve2, ARM_HWCAP2_A64_SVE2); 719 GET_FEATURE_ID(aa64_sve2_aes, ARM_HWCAP2_A64_SVEAES); 720 GET_FEATURE_ID(aa64_sve2_pmull128, ARM_HWCAP2_A64_SVEPMULL); 721 GET_FEATURE_ID(aa64_sve2_bitperm, ARM_HWCAP2_A64_SVEBITPERM); 722 GET_FEATURE_ID(aa64_sve2_sha3, ARM_HWCAP2_A64_SVESHA3); 723 GET_FEATURE_ID(aa64_sve2_sm4, ARM_HWCAP2_A64_SVESM4); 724 GET_FEATURE_ID(aa64_condm_5, ARM_HWCAP2_A64_FLAGM2); 725 GET_FEATURE_ID(aa64_frint, ARM_HWCAP2_A64_FRINT); 726 GET_FEATURE_ID(aa64_sve_i8mm, ARM_HWCAP2_A64_SVEI8MM); 727 GET_FEATURE_ID(aa64_sve_f32mm, ARM_HWCAP2_A64_SVEF32MM); 728 GET_FEATURE_ID(aa64_sve_f64mm, ARM_HWCAP2_A64_SVEF64MM); 729 GET_FEATURE_ID(aa64_sve_bf16, ARM_HWCAP2_A64_SVEBF16); 730 GET_FEATURE_ID(aa64_i8mm, ARM_HWCAP2_A64_I8MM); 731 GET_FEATURE_ID(aa64_bf16, ARM_HWCAP2_A64_BF16); 732 GET_FEATURE_ID(aa64_rndr, ARM_HWCAP2_A64_RNG); 733 GET_FEATURE_ID(aa64_bti, ARM_HWCAP2_A64_BTI); 734 GET_FEATURE_ID(aa64_mte, ARM_HWCAP2_A64_MTE); 735 GET_FEATURE_ID(aa64_sme, (ARM_HWCAP2_A64_SME | 736 ARM_HWCAP2_A64_SME_F32F32 | 737 ARM_HWCAP2_A64_SME_B16F32 | 738 ARM_HWCAP2_A64_SME_F16F32 | 739 ARM_HWCAP2_A64_SME_I8I32)); 740 GET_FEATURE_ID(aa64_sme_f64f64, ARM_HWCAP2_A64_SME_F64F64); 741 GET_FEATURE_ID(aa64_sme_i16i64, ARM_HWCAP2_A64_SME_I16I64); 742 GET_FEATURE_ID(aa64_sme_fa64, ARM_HWCAP2_A64_SME_FA64); 743 744 return hwcaps; 745 } 746 747 #undef GET_FEATURE_ID 748 749 #endif /* not TARGET_AARCH64 */ 750 #endif /* TARGET_ARM */ 751 752 #ifdef TARGET_SPARC 753 #ifdef TARGET_SPARC64 754 755 #define ELF_START_MMAP 0x80000000 756 #define ELF_HWCAP (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \ 757 | HWCAP_SPARC_MULDIV | HWCAP_SPARC_V9) 758 #ifndef TARGET_ABI32 759 #define elf_check_arch(x) ( (x) == EM_SPARCV9 || (x) == EM_SPARC32PLUS ) 760 #else 761 #define elf_check_arch(x) ( (x) == EM_SPARC32PLUS || (x) == EM_SPARC ) 762 #endif 763 764 #define ELF_CLASS ELFCLASS64 765 #define ELF_ARCH EM_SPARCV9 766 #else 767 #define ELF_START_MMAP 0x80000000 768 #define ELF_HWCAP (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \ 769 | HWCAP_SPARC_MULDIV) 770 #define ELF_CLASS ELFCLASS32 771 #define ELF_ARCH EM_SPARC 772 #endif /* TARGET_SPARC64 */ 773 774 static inline void init_thread(struct target_pt_regs *regs, 775 struct image_info *infop) 776 { 777 /* Note that target_cpu_copy_regs does not read psr/tstate. */ 778 regs->pc = infop->entry; 779 regs->npc = regs->pc + 4; 780 regs->y = 0; 781 regs->u_regs[14] = (infop->start_stack - 16 * sizeof(abi_ulong) 782 - TARGET_STACK_BIAS); 783 } 784 #endif /* TARGET_SPARC */ 785 786 #ifdef TARGET_PPC 787 788 #define ELF_MACHINE PPC_ELF_MACHINE 789 #define ELF_START_MMAP 0x80000000 790 791 #if defined(TARGET_PPC64) 792 793 #define elf_check_arch(x) ( (x) == EM_PPC64 ) 794 795 #define ELF_CLASS ELFCLASS64 796 797 #else 798 799 #define ELF_CLASS ELFCLASS32 800 #define EXSTACK_DEFAULT true 801 802 #endif 803 804 #define ELF_ARCH EM_PPC 805 806 /* Feature masks for the Aux Vector Hardware Capabilities (AT_HWCAP). 807 See arch/powerpc/include/asm/cputable.h. */ 808 enum { 809 QEMU_PPC_FEATURE_32 = 0x80000000, 810 QEMU_PPC_FEATURE_64 = 0x40000000, 811 QEMU_PPC_FEATURE_601_INSTR = 0x20000000, 812 QEMU_PPC_FEATURE_HAS_ALTIVEC = 0x10000000, 813 QEMU_PPC_FEATURE_HAS_FPU = 0x08000000, 814 QEMU_PPC_FEATURE_HAS_MMU = 0x04000000, 815 QEMU_PPC_FEATURE_HAS_4xxMAC = 0x02000000, 816 QEMU_PPC_FEATURE_UNIFIED_CACHE = 0x01000000, 817 QEMU_PPC_FEATURE_HAS_SPE = 0x00800000, 818 QEMU_PPC_FEATURE_HAS_EFP_SINGLE = 0x00400000, 819 QEMU_PPC_FEATURE_HAS_EFP_DOUBLE = 0x00200000, 820 QEMU_PPC_FEATURE_NO_TB = 0x00100000, 821 QEMU_PPC_FEATURE_POWER4 = 0x00080000, 822 QEMU_PPC_FEATURE_POWER5 = 0x00040000, 823 QEMU_PPC_FEATURE_POWER5_PLUS = 0x00020000, 824 QEMU_PPC_FEATURE_CELL = 0x00010000, 825 QEMU_PPC_FEATURE_BOOKE = 0x00008000, 826 QEMU_PPC_FEATURE_SMT = 0x00004000, 827 QEMU_PPC_FEATURE_ICACHE_SNOOP = 0x00002000, 828 QEMU_PPC_FEATURE_ARCH_2_05 = 0x00001000, 829 QEMU_PPC_FEATURE_PA6T = 0x00000800, 830 QEMU_PPC_FEATURE_HAS_DFP = 0x00000400, 831 QEMU_PPC_FEATURE_POWER6_EXT = 0x00000200, 832 QEMU_PPC_FEATURE_ARCH_2_06 = 0x00000100, 833 QEMU_PPC_FEATURE_HAS_VSX = 0x00000080, 834 QEMU_PPC_FEATURE_PSERIES_PERFMON_COMPAT = 0x00000040, 835 836 QEMU_PPC_FEATURE_TRUE_LE = 0x00000002, 837 QEMU_PPC_FEATURE_PPC_LE = 0x00000001, 838 839 /* Feature definitions in AT_HWCAP2. */ 840 QEMU_PPC_FEATURE2_ARCH_2_07 = 0x80000000, /* ISA 2.07 */ 841 QEMU_PPC_FEATURE2_HAS_HTM = 0x40000000, /* Hardware Transactional Memory */ 842 QEMU_PPC_FEATURE2_HAS_DSCR = 0x20000000, /* Data Stream Control Register */ 843 QEMU_PPC_FEATURE2_HAS_EBB = 0x10000000, /* Event Base Branching */ 844 QEMU_PPC_FEATURE2_HAS_ISEL = 0x08000000, /* Integer Select */ 845 QEMU_PPC_FEATURE2_HAS_TAR = 0x04000000, /* Target Address Register */ 846 QEMU_PPC_FEATURE2_VEC_CRYPTO = 0x02000000, 847 QEMU_PPC_FEATURE2_HTM_NOSC = 0x01000000, 848 QEMU_PPC_FEATURE2_ARCH_3_00 = 0x00800000, /* ISA 3.00 */ 849 QEMU_PPC_FEATURE2_HAS_IEEE128 = 0x00400000, /* VSX IEEE Bin Float 128-bit */ 850 QEMU_PPC_FEATURE2_DARN = 0x00200000, /* darn random number insn */ 851 QEMU_PPC_FEATURE2_SCV = 0x00100000, /* scv syscall */ 852 QEMU_PPC_FEATURE2_HTM_NO_SUSPEND = 0x00080000, /* TM w/o suspended state */ 853 QEMU_PPC_FEATURE2_ARCH_3_1 = 0x00040000, /* ISA 3.1 */ 854 QEMU_PPC_FEATURE2_MMA = 0x00020000, /* Matrix-Multiply Assist */ 855 }; 856 857 #define ELF_HWCAP get_elf_hwcap() 858 859 static uint32_t get_elf_hwcap(void) 860 { 861 PowerPCCPU *cpu = POWERPC_CPU(thread_cpu); 862 uint32_t features = 0; 863 864 /* We don't have to be terribly complete here; the high points are 865 Altivec/FP/SPE support. Anything else is just a bonus. */ 866 #define GET_FEATURE(flag, feature) \ 867 do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0) 868 #define GET_FEATURE2(flags, feature) \ 869 do { \ 870 if ((cpu->env.insns_flags2 & flags) == flags) { \ 871 features |= feature; \ 872 } \ 873 } while (0) 874 GET_FEATURE(PPC_64B, QEMU_PPC_FEATURE_64); 875 GET_FEATURE(PPC_FLOAT, QEMU_PPC_FEATURE_HAS_FPU); 876 GET_FEATURE(PPC_ALTIVEC, QEMU_PPC_FEATURE_HAS_ALTIVEC); 877 GET_FEATURE(PPC_SPE, QEMU_PPC_FEATURE_HAS_SPE); 878 GET_FEATURE(PPC_SPE_SINGLE, QEMU_PPC_FEATURE_HAS_EFP_SINGLE); 879 GET_FEATURE(PPC_SPE_DOUBLE, QEMU_PPC_FEATURE_HAS_EFP_DOUBLE); 880 GET_FEATURE(PPC_BOOKE, QEMU_PPC_FEATURE_BOOKE); 881 GET_FEATURE(PPC_405_MAC, QEMU_PPC_FEATURE_HAS_4xxMAC); 882 GET_FEATURE2(PPC2_DFP, QEMU_PPC_FEATURE_HAS_DFP); 883 GET_FEATURE2(PPC2_VSX, QEMU_PPC_FEATURE_HAS_VSX); 884 GET_FEATURE2((PPC2_PERM_ISA206 | PPC2_DIVE_ISA206 | PPC2_ATOMIC_ISA206 | 885 PPC2_FP_CVT_ISA206 | PPC2_FP_TST_ISA206), 886 QEMU_PPC_FEATURE_ARCH_2_06); 887 #undef GET_FEATURE 888 #undef GET_FEATURE2 889 890 return features; 891 } 892 893 #define ELF_HWCAP2 get_elf_hwcap2() 894 895 static uint32_t get_elf_hwcap2(void) 896 { 897 PowerPCCPU *cpu = POWERPC_CPU(thread_cpu); 898 uint32_t features = 0; 899 900 #define GET_FEATURE(flag, feature) \ 901 do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0) 902 #define GET_FEATURE2(flag, feature) \ 903 do { if (cpu->env.insns_flags2 & flag) { features |= feature; } } while (0) 904 905 GET_FEATURE(PPC_ISEL, QEMU_PPC_FEATURE2_HAS_ISEL); 906 GET_FEATURE2(PPC2_BCTAR_ISA207, QEMU_PPC_FEATURE2_HAS_TAR); 907 GET_FEATURE2((PPC2_BCTAR_ISA207 | PPC2_LSQ_ISA207 | PPC2_ALTIVEC_207 | 908 PPC2_ISA207S), QEMU_PPC_FEATURE2_ARCH_2_07 | 909 QEMU_PPC_FEATURE2_VEC_CRYPTO); 910 GET_FEATURE2(PPC2_ISA300, QEMU_PPC_FEATURE2_ARCH_3_00 | 911 QEMU_PPC_FEATURE2_DARN | QEMU_PPC_FEATURE2_HAS_IEEE128); 912 GET_FEATURE2(PPC2_ISA310, QEMU_PPC_FEATURE2_ARCH_3_1 | 913 QEMU_PPC_FEATURE2_MMA); 914 915 #undef GET_FEATURE 916 #undef GET_FEATURE2 917 918 return features; 919 } 920 921 /* 922 * The requirements here are: 923 * - keep the final alignment of sp (sp & 0xf) 924 * - make sure the 32-bit value at the first 16 byte aligned position of 925 * AUXV is greater than 16 for glibc compatibility. 926 * AT_IGNOREPPC is used for that. 927 * - for compatibility with glibc ARCH_DLINFO must always be defined on PPC, 928 * even if DLINFO_ARCH_ITEMS goes to zero or is undefined. 929 */ 930 #define DLINFO_ARCH_ITEMS 5 931 #define ARCH_DLINFO \ 932 do { \ 933 PowerPCCPU *cpu = POWERPC_CPU(thread_cpu); \ 934 /* \ 935 * Handle glibc compatibility: these magic entries must \ 936 * be at the lowest addresses in the final auxv. \ 937 */ \ 938 NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC); \ 939 NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC); \ 940 NEW_AUX_ENT(AT_DCACHEBSIZE, cpu->env.dcache_line_size); \ 941 NEW_AUX_ENT(AT_ICACHEBSIZE, cpu->env.icache_line_size); \ 942 NEW_AUX_ENT(AT_UCACHEBSIZE, 0); \ 943 } while (0) 944 945 static inline void init_thread(struct target_pt_regs *_regs, struct image_info *infop) 946 { 947 _regs->gpr[1] = infop->start_stack; 948 #if defined(TARGET_PPC64) 949 if (get_ppc64_abi(infop) < 2) { 950 uint64_t val; 951 get_user_u64(val, infop->entry + 8); 952 _regs->gpr[2] = val + infop->load_bias; 953 get_user_u64(val, infop->entry); 954 infop->entry = val + infop->load_bias; 955 } else { 956 _regs->gpr[12] = infop->entry; /* r12 set to global entry address */ 957 } 958 #endif 959 _regs->nip = infop->entry; 960 } 961 962 /* See linux kernel: arch/powerpc/include/asm/elf.h. */ 963 #define ELF_NREG 48 964 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 965 966 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUPPCState *env) 967 { 968 int i; 969 target_ulong ccr = 0; 970 971 for (i = 0; i < ARRAY_SIZE(env->gpr); i++) { 972 (*regs)[i] = tswapreg(env->gpr[i]); 973 } 974 975 (*regs)[32] = tswapreg(env->nip); 976 (*regs)[33] = tswapreg(env->msr); 977 (*regs)[35] = tswapreg(env->ctr); 978 (*regs)[36] = tswapreg(env->lr); 979 (*regs)[37] = tswapreg(cpu_read_xer(env)); 980 981 for (i = 0; i < ARRAY_SIZE(env->crf); i++) { 982 ccr |= env->crf[i] << (32 - ((i + 1) * 4)); 983 } 984 (*regs)[38] = tswapreg(ccr); 985 } 986 987 #define USE_ELF_CORE_DUMP 988 #define ELF_EXEC_PAGESIZE 4096 989 990 #endif 991 992 #ifdef TARGET_LOONGARCH64 993 994 #define ELF_START_MMAP 0x80000000 995 996 #define ELF_CLASS ELFCLASS64 997 #define ELF_ARCH EM_LOONGARCH 998 #define EXSTACK_DEFAULT true 999 1000 #define elf_check_arch(x) ((x) == EM_LOONGARCH) 1001 1002 static inline void init_thread(struct target_pt_regs *regs, 1003 struct image_info *infop) 1004 { 1005 /*Set crmd PG,DA = 1,0 */ 1006 regs->csr.crmd = 2 << 3; 1007 regs->csr.era = infop->entry; 1008 regs->regs[3] = infop->start_stack; 1009 } 1010 1011 /* See linux kernel: arch/loongarch/include/asm/elf.h */ 1012 #define ELF_NREG 45 1013 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1014 1015 enum { 1016 TARGET_EF_R0 = 0, 1017 TARGET_EF_CSR_ERA = TARGET_EF_R0 + 33, 1018 TARGET_EF_CSR_BADV = TARGET_EF_R0 + 34, 1019 }; 1020 1021 static void elf_core_copy_regs(target_elf_gregset_t *regs, 1022 const CPULoongArchState *env) 1023 { 1024 int i; 1025 1026 (*regs)[TARGET_EF_R0] = 0; 1027 1028 for (i = 1; i < ARRAY_SIZE(env->gpr); i++) { 1029 (*regs)[TARGET_EF_R0 + i] = tswapreg(env->gpr[i]); 1030 } 1031 1032 (*regs)[TARGET_EF_CSR_ERA] = tswapreg(env->pc); 1033 (*regs)[TARGET_EF_CSR_BADV] = tswapreg(env->CSR_BADV); 1034 } 1035 1036 #define USE_ELF_CORE_DUMP 1037 #define ELF_EXEC_PAGESIZE 4096 1038 1039 #define ELF_HWCAP get_elf_hwcap() 1040 1041 /* See arch/loongarch/include/uapi/asm/hwcap.h */ 1042 enum { 1043 HWCAP_LOONGARCH_CPUCFG = (1 << 0), 1044 HWCAP_LOONGARCH_LAM = (1 << 1), 1045 HWCAP_LOONGARCH_UAL = (1 << 2), 1046 HWCAP_LOONGARCH_FPU = (1 << 3), 1047 HWCAP_LOONGARCH_LSX = (1 << 4), 1048 HWCAP_LOONGARCH_LASX = (1 << 5), 1049 HWCAP_LOONGARCH_CRC32 = (1 << 6), 1050 HWCAP_LOONGARCH_COMPLEX = (1 << 7), 1051 HWCAP_LOONGARCH_CRYPTO = (1 << 8), 1052 HWCAP_LOONGARCH_LVZ = (1 << 9), 1053 HWCAP_LOONGARCH_LBT_X86 = (1 << 10), 1054 HWCAP_LOONGARCH_LBT_ARM = (1 << 11), 1055 HWCAP_LOONGARCH_LBT_MIPS = (1 << 12), 1056 }; 1057 1058 static uint32_t get_elf_hwcap(void) 1059 { 1060 LoongArchCPU *cpu = LOONGARCH_CPU(thread_cpu); 1061 uint32_t hwcaps = 0; 1062 1063 hwcaps |= HWCAP_LOONGARCH_CRC32; 1064 1065 if (FIELD_EX32(cpu->env.cpucfg[1], CPUCFG1, UAL)) { 1066 hwcaps |= HWCAP_LOONGARCH_UAL; 1067 } 1068 1069 if (FIELD_EX32(cpu->env.cpucfg[2], CPUCFG2, FP)) { 1070 hwcaps |= HWCAP_LOONGARCH_FPU; 1071 } 1072 1073 if (FIELD_EX32(cpu->env.cpucfg[2], CPUCFG2, LAM)) { 1074 hwcaps |= HWCAP_LOONGARCH_LAM; 1075 } 1076 1077 return hwcaps; 1078 } 1079 1080 #define ELF_PLATFORM "loongarch" 1081 1082 #endif /* TARGET_LOONGARCH64 */ 1083 1084 #ifdef TARGET_MIPS 1085 1086 #define ELF_START_MMAP 0x80000000 1087 1088 #ifdef TARGET_MIPS64 1089 #define ELF_CLASS ELFCLASS64 1090 #else 1091 #define ELF_CLASS ELFCLASS32 1092 #endif 1093 #define ELF_ARCH EM_MIPS 1094 #define EXSTACK_DEFAULT true 1095 1096 #ifdef TARGET_ABI_MIPSN32 1097 #define elf_check_abi(x) ((x) & EF_MIPS_ABI2) 1098 #else 1099 #define elf_check_abi(x) (!((x) & EF_MIPS_ABI2)) 1100 #endif 1101 1102 #define ELF_BASE_PLATFORM get_elf_base_platform() 1103 1104 #define MATCH_PLATFORM_INSN(_flags, _base_platform) \ 1105 do { if ((cpu->env.insn_flags & (_flags)) == _flags) \ 1106 { return _base_platform; } } while (0) 1107 1108 static const char *get_elf_base_platform(void) 1109 { 1110 MIPSCPU *cpu = MIPS_CPU(thread_cpu); 1111 1112 /* 64 bit ISAs goes first */ 1113 MATCH_PLATFORM_INSN(CPU_MIPS64R6, "mips64r6"); 1114 MATCH_PLATFORM_INSN(CPU_MIPS64R5, "mips64r5"); 1115 MATCH_PLATFORM_INSN(CPU_MIPS64R2, "mips64r2"); 1116 MATCH_PLATFORM_INSN(CPU_MIPS64R1, "mips64"); 1117 MATCH_PLATFORM_INSN(CPU_MIPS5, "mips5"); 1118 MATCH_PLATFORM_INSN(CPU_MIPS4, "mips4"); 1119 MATCH_PLATFORM_INSN(CPU_MIPS3, "mips3"); 1120 1121 /* 32 bit ISAs */ 1122 MATCH_PLATFORM_INSN(CPU_MIPS32R6, "mips32r6"); 1123 MATCH_PLATFORM_INSN(CPU_MIPS32R5, "mips32r5"); 1124 MATCH_PLATFORM_INSN(CPU_MIPS32R2, "mips32r2"); 1125 MATCH_PLATFORM_INSN(CPU_MIPS32R1, "mips32"); 1126 MATCH_PLATFORM_INSN(CPU_MIPS2, "mips2"); 1127 1128 /* Fallback */ 1129 return "mips"; 1130 } 1131 #undef MATCH_PLATFORM_INSN 1132 1133 static inline void init_thread(struct target_pt_regs *regs, 1134 struct image_info *infop) 1135 { 1136 regs->cp0_status = 2 << CP0St_KSU; 1137 regs->cp0_epc = infop->entry; 1138 regs->regs[29] = infop->start_stack; 1139 } 1140 1141 /* See linux kernel: arch/mips/include/asm/elf.h. */ 1142 #define ELF_NREG 45 1143 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1144 1145 /* See linux kernel: arch/mips/include/asm/reg.h. */ 1146 enum { 1147 #ifdef TARGET_MIPS64 1148 TARGET_EF_R0 = 0, 1149 #else 1150 TARGET_EF_R0 = 6, 1151 #endif 1152 TARGET_EF_R26 = TARGET_EF_R0 + 26, 1153 TARGET_EF_R27 = TARGET_EF_R0 + 27, 1154 TARGET_EF_LO = TARGET_EF_R0 + 32, 1155 TARGET_EF_HI = TARGET_EF_R0 + 33, 1156 TARGET_EF_CP0_EPC = TARGET_EF_R0 + 34, 1157 TARGET_EF_CP0_BADVADDR = TARGET_EF_R0 + 35, 1158 TARGET_EF_CP0_STATUS = TARGET_EF_R0 + 36, 1159 TARGET_EF_CP0_CAUSE = TARGET_EF_R0 + 37 1160 }; 1161 1162 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */ 1163 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMIPSState *env) 1164 { 1165 int i; 1166 1167 for (i = 0; i < TARGET_EF_R0; i++) { 1168 (*regs)[i] = 0; 1169 } 1170 (*regs)[TARGET_EF_R0] = 0; 1171 1172 for (i = 1; i < ARRAY_SIZE(env->active_tc.gpr); i++) { 1173 (*regs)[TARGET_EF_R0 + i] = tswapreg(env->active_tc.gpr[i]); 1174 } 1175 1176 (*regs)[TARGET_EF_R26] = 0; 1177 (*regs)[TARGET_EF_R27] = 0; 1178 (*regs)[TARGET_EF_LO] = tswapreg(env->active_tc.LO[0]); 1179 (*regs)[TARGET_EF_HI] = tswapreg(env->active_tc.HI[0]); 1180 (*regs)[TARGET_EF_CP0_EPC] = tswapreg(env->active_tc.PC); 1181 (*regs)[TARGET_EF_CP0_BADVADDR] = tswapreg(env->CP0_BadVAddr); 1182 (*regs)[TARGET_EF_CP0_STATUS] = tswapreg(env->CP0_Status); 1183 (*regs)[TARGET_EF_CP0_CAUSE] = tswapreg(env->CP0_Cause); 1184 } 1185 1186 #define USE_ELF_CORE_DUMP 1187 #define ELF_EXEC_PAGESIZE 4096 1188 1189 /* See arch/mips/include/uapi/asm/hwcap.h. */ 1190 enum { 1191 HWCAP_MIPS_R6 = (1 << 0), 1192 HWCAP_MIPS_MSA = (1 << 1), 1193 HWCAP_MIPS_CRC32 = (1 << 2), 1194 HWCAP_MIPS_MIPS16 = (1 << 3), 1195 HWCAP_MIPS_MDMX = (1 << 4), 1196 HWCAP_MIPS_MIPS3D = (1 << 5), 1197 HWCAP_MIPS_SMARTMIPS = (1 << 6), 1198 HWCAP_MIPS_DSP = (1 << 7), 1199 HWCAP_MIPS_DSP2 = (1 << 8), 1200 HWCAP_MIPS_DSP3 = (1 << 9), 1201 HWCAP_MIPS_MIPS16E2 = (1 << 10), 1202 HWCAP_LOONGSON_MMI = (1 << 11), 1203 HWCAP_LOONGSON_EXT = (1 << 12), 1204 HWCAP_LOONGSON_EXT2 = (1 << 13), 1205 HWCAP_LOONGSON_CPUCFG = (1 << 14), 1206 }; 1207 1208 #define ELF_HWCAP get_elf_hwcap() 1209 1210 #define GET_FEATURE_INSN(_flag, _hwcap) \ 1211 do { if (cpu->env.insn_flags & (_flag)) { hwcaps |= _hwcap; } } while (0) 1212 1213 #define GET_FEATURE_REG_SET(_reg, _mask, _hwcap) \ 1214 do { if (cpu->env._reg & (_mask)) { hwcaps |= _hwcap; } } while (0) 1215 1216 #define GET_FEATURE_REG_EQU(_reg, _start, _length, _val, _hwcap) \ 1217 do { \ 1218 if (extract32(cpu->env._reg, (_start), (_length)) == (_val)) { \ 1219 hwcaps |= _hwcap; \ 1220 } \ 1221 } while (0) 1222 1223 static uint32_t get_elf_hwcap(void) 1224 { 1225 MIPSCPU *cpu = MIPS_CPU(thread_cpu); 1226 uint32_t hwcaps = 0; 1227 1228 GET_FEATURE_REG_EQU(CP0_Config0, CP0C0_AR, CP0C0_AR_LENGTH, 1229 2, HWCAP_MIPS_R6); 1230 GET_FEATURE_REG_SET(CP0_Config3, 1 << CP0C3_MSAP, HWCAP_MIPS_MSA); 1231 GET_FEATURE_INSN(ASE_LMMI, HWCAP_LOONGSON_MMI); 1232 GET_FEATURE_INSN(ASE_LEXT, HWCAP_LOONGSON_EXT); 1233 1234 return hwcaps; 1235 } 1236 1237 #undef GET_FEATURE_REG_EQU 1238 #undef GET_FEATURE_REG_SET 1239 #undef GET_FEATURE_INSN 1240 1241 #endif /* TARGET_MIPS */ 1242 1243 #ifdef TARGET_MICROBLAZE 1244 1245 #define ELF_START_MMAP 0x80000000 1246 1247 #define elf_check_arch(x) ( (x) == EM_MICROBLAZE || (x) == EM_MICROBLAZE_OLD) 1248 1249 #define ELF_CLASS ELFCLASS32 1250 #define ELF_ARCH EM_MICROBLAZE 1251 1252 static inline void init_thread(struct target_pt_regs *regs, 1253 struct image_info *infop) 1254 { 1255 regs->pc = infop->entry; 1256 regs->r1 = infop->start_stack; 1257 1258 } 1259 1260 #define ELF_EXEC_PAGESIZE 4096 1261 1262 #define USE_ELF_CORE_DUMP 1263 #define ELF_NREG 38 1264 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1265 1266 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */ 1267 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMBState *env) 1268 { 1269 int i, pos = 0; 1270 1271 for (i = 0; i < 32; i++) { 1272 (*regs)[pos++] = tswapreg(env->regs[i]); 1273 } 1274 1275 (*regs)[pos++] = tswapreg(env->pc); 1276 (*regs)[pos++] = tswapreg(mb_cpu_read_msr(env)); 1277 (*regs)[pos++] = 0; 1278 (*regs)[pos++] = tswapreg(env->ear); 1279 (*regs)[pos++] = 0; 1280 (*regs)[pos++] = tswapreg(env->esr); 1281 } 1282 1283 #endif /* TARGET_MICROBLAZE */ 1284 1285 #ifdef TARGET_NIOS2 1286 1287 #define ELF_START_MMAP 0x80000000 1288 1289 #define elf_check_arch(x) ((x) == EM_ALTERA_NIOS2) 1290 1291 #define ELF_CLASS ELFCLASS32 1292 #define ELF_ARCH EM_ALTERA_NIOS2 1293 1294 static void init_thread(struct target_pt_regs *regs, struct image_info *infop) 1295 { 1296 regs->ea = infop->entry; 1297 regs->sp = infop->start_stack; 1298 } 1299 1300 #define LO_COMMPAGE TARGET_PAGE_SIZE 1301 1302 static bool init_guest_commpage(void) 1303 { 1304 static const uint8_t kuser_page[4 + 2 * 64] = { 1305 /* __kuser_helper_version */ 1306 [0x00] = 0x02, 0x00, 0x00, 0x00, 1307 1308 /* __kuser_cmpxchg */ 1309 [0x04] = 0x3a, 0x6c, 0x3b, 0x00, /* trap 16 */ 1310 0x3a, 0x28, 0x00, 0xf8, /* ret */ 1311 1312 /* __kuser_sigtramp */ 1313 [0x44] = 0xc4, 0x22, 0x80, 0x00, /* movi r2, __NR_rt_sigreturn */ 1314 0x3a, 0x68, 0x3b, 0x00, /* trap 0 */ 1315 }; 1316 1317 void *want = g2h_untagged(LO_COMMPAGE & -qemu_host_page_size); 1318 void *addr = mmap(want, qemu_host_page_size, PROT_READ | PROT_WRITE, 1319 MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0); 1320 1321 if (addr == MAP_FAILED) { 1322 perror("Allocating guest commpage"); 1323 exit(EXIT_FAILURE); 1324 } 1325 if (addr != want) { 1326 return false; 1327 } 1328 1329 memcpy(addr, kuser_page, sizeof(kuser_page)); 1330 1331 if (mprotect(addr, qemu_host_page_size, PROT_READ)) { 1332 perror("Protecting guest commpage"); 1333 exit(EXIT_FAILURE); 1334 } 1335 1336 page_set_flags(LO_COMMPAGE, LO_COMMPAGE | ~TARGET_PAGE_MASK, 1337 PAGE_READ | PAGE_EXEC | PAGE_VALID); 1338 return true; 1339 } 1340 1341 #define ELF_EXEC_PAGESIZE 4096 1342 1343 #define USE_ELF_CORE_DUMP 1344 #define ELF_NREG 49 1345 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1346 1347 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */ 1348 static void elf_core_copy_regs(target_elf_gregset_t *regs, 1349 const CPUNios2State *env) 1350 { 1351 int i; 1352 1353 (*regs)[0] = -1; 1354 for (i = 1; i < 8; i++) /* r0-r7 */ 1355 (*regs)[i] = tswapreg(env->regs[i + 7]); 1356 1357 for (i = 8; i < 16; i++) /* r8-r15 */ 1358 (*regs)[i] = tswapreg(env->regs[i - 8]); 1359 1360 for (i = 16; i < 24; i++) /* r16-r23 */ 1361 (*regs)[i] = tswapreg(env->regs[i + 7]); 1362 (*regs)[24] = -1; /* R_ET */ 1363 (*regs)[25] = -1; /* R_BT */ 1364 (*regs)[26] = tswapreg(env->regs[R_GP]); 1365 (*regs)[27] = tswapreg(env->regs[R_SP]); 1366 (*regs)[28] = tswapreg(env->regs[R_FP]); 1367 (*regs)[29] = tswapreg(env->regs[R_EA]); 1368 (*regs)[30] = -1; /* R_SSTATUS */ 1369 (*regs)[31] = tswapreg(env->regs[R_RA]); 1370 1371 (*regs)[32] = tswapreg(env->pc); 1372 1373 (*regs)[33] = -1; /* R_STATUS */ 1374 (*regs)[34] = tswapreg(env->regs[CR_ESTATUS]); 1375 1376 for (i = 35; i < 49; i++) /* ... */ 1377 (*regs)[i] = -1; 1378 } 1379 1380 #endif /* TARGET_NIOS2 */ 1381 1382 #ifdef TARGET_OPENRISC 1383 1384 #define ELF_START_MMAP 0x08000000 1385 1386 #define ELF_ARCH EM_OPENRISC 1387 #define ELF_CLASS ELFCLASS32 1388 #define ELF_DATA ELFDATA2MSB 1389 1390 static inline void init_thread(struct target_pt_regs *regs, 1391 struct image_info *infop) 1392 { 1393 regs->pc = infop->entry; 1394 regs->gpr[1] = infop->start_stack; 1395 } 1396 1397 #define USE_ELF_CORE_DUMP 1398 #define ELF_EXEC_PAGESIZE 8192 1399 1400 /* See linux kernel arch/openrisc/include/asm/elf.h. */ 1401 #define ELF_NREG 34 /* gprs and pc, sr */ 1402 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1403 1404 static void elf_core_copy_regs(target_elf_gregset_t *regs, 1405 const CPUOpenRISCState *env) 1406 { 1407 int i; 1408 1409 for (i = 0; i < 32; i++) { 1410 (*regs)[i] = tswapreg(cpu_get_gpr(env, i)); 1411 } 1412 (*regs)[32] = tswapreg(env->pc); 1413 (*regs)[33] = tswapreg(cpu_get_sr(env)); 1414 } 1415 #define ELF_HWCAP 0 1416 #define ELF_PLATFORM NULL 1417 1418 #endif /* TARGET_OPENRISC */ 1419 1420 #ifdef TARGET_SH4 1421 1422 #define ELF_START_MMAP 0x80000000 1423 1424 #define ELF_CLASS ELFCLASS32 1425 #define ELF_ARCH EM_SH 1426 1427 static inline void init_thread(struct target_pt_regs *regs, 1428 struct image_info *infop) 1429 { 1430 /* Check other registers XXXXX */ 1431 regs->pc = infop->entry; 1432 regs->regs[15] = infop->start_stack; 1433 } 1434 1435 /* See linux kernel: arch/sh/include/asm/elf.h. */ 1436 #define ELF_NREG 23 1437 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1438 1439 /* See linux kernel: arch/sh/include/asm/ptrace.h. */ 1440 enum { 1441 TARGET_REG_PC = 16, 1442 TARGET_REG_PR = 17, 1443 TARGET_REG_SR = 18, 1444 TARGET_REG_GBR = 19, 1445 TARGET_REG_MACH = 20, 1446 TARGET_REG_MACL = 21, 1447 TARGET_REG_SYSCALL = 22 1448 }; 1449 1450 static inline void elf_core_copy_regs(target_elf_gregset_t *regs, 1451 const CPUSH4State *env) 1452 { 1453 int i; 1454 1455 for (i = 0; i < 16; i++) { 1456 (*regs)[i] = tswapreg(env->gregs[i]); 1457 } 1458 1459 (*regs)[TARGET_REG_PC] = tswapreg(env->pc); 1460 (*regs)[TARGET_REG_PR] = tswapreg(env->pr); 1461 (*regs)[TARGET_REG_SR] = tswapreg(env->sr); 1462 (*regs)[TARGET_REG_GBR] = tswapreg(env->gbr); 1463 (*regs)[TARGET_REG_MACH] = tswapreg(env->mach); 1464 (*regs)[TARGET_REG_MACL] = tswapreg(env->macl); 1465 (*regs)[TARGET_REG_SYSCALL] = 0; /* FIXME */ 1466 } 1467 1468 #define USE_ELF_CORE_DUMP 1469 #define ELF_EXEC_PAGESIZE 4096 1470 1471 enum { 1472 SH_CPU_HAS_FPU = 0x0001, /* Hardware FPU support */ 1473 SH_CPU_HAS_P2_FLUSH_BUG = 0x0002, /* Need to flush the cache in P2 area */ 1474 SH_CPU_HAS_MMU_PAGE_ASSOC = 0x0004, /* SH3: TLB way selection bit support */ 1475 SH_CPU_HAS_DSP = 0x0008, /* SH-DSP: DSP support */ 1476 SH_CPU_HAS_PERF_COUNTER = 0x0010, /* Hardware performance counters */ 1477 SH_CPU_HAS_PTEA = 0x0020, /* PTEA register */ 1478 SH_CPU_HAS_LLSC = 0x0040, /* movli.l/movco.l */ 1479 SH_CPU_HAS_L2_CACHE = 0x0080, /* Secondary cache / URAM */ 1480 SH_CPU_HAS_OP32 = 0x0100, /* 32-bit instruction support */ 1481 SH_CPU_HAS_PTEAEX = 0x0200, /* PTE ASID Extension support */ 1482 }; 1483 1484 #define ELF_HWCAP get_elf_hwcap() 1485 1486 static uint32_t get_elf_hwcap(void) 1487 { 1488 SuperHCPU *cpu = SUPERH_CPU(thread_cpu); 1489 uint32_t hwcap = 0; 1490 1491 hwcap |= SH_CPU_HAS_FPU; 1492 1493 if (cpu->env.features & SH_FEATURE_SH4A) { 1494 hwcap |= SH_CPU_HAS_LLSC; 1495 } 1496 1497 return hwcap; 1498 } 1499 1500 #endif 1501 1502 #ifdef TARGET_CRIS 1503 1504 #define ELF_START_MMAP 0x80000000 1505 1506 #define ELF_CLASS ELFCLASS32 1507 #define ELF_ARCH EM_CRIS 1508 1509 static inline void init_thread(struct target_pt_regs *regs, 1510 struct image_info *infop) 1511 { 1512 regs->erp = infop->entry; 1513 } 1514 1515 #define ELF_EXEC_PAGESIZE 8192 1516 1517 #endif 1518 1519 #ifdef TARGET_M68K 1520 1521 #define ELF_START_MMAP 0x80000000 1522 1523 #define ELF_CLASS ELFCLASS32 1524 #define ELF_ARCH EM_68K 1525 1526 /* ??? Does this need to do anything? 1527 #define ELF_PLAT_INIT(_r) */ 1528 1529 static inline void init_thread(struct target_pt_regs *regs, 1530 struct image_info *infop) 1531 { 1532 regs->usp = infop->start_stack; 1533 regs->sr = 0; 1534 regs->pc = infop->entry; 1535 } 1536 1537 /* See linux kernel: arch/m68k/include/asm/elf.h. */ 1538 #define ELF_NREG 20 1539 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1540 1541 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUM68KState *env) 1542 { 1543 (*regs)[0] = tswapreg(env->dregs[1]); 1544 (*regs)[1] = tswapreg(env->dregs[2]); 1545 (*regs)[2] = tswapreg(env->dregs[3]); 1546 (*regs)[3] = tswapreg(env->dregs[4]); 1547 (*regs)[4] = tswapreg(env->dregs[5]); 1548 (*regs)[5] = tswapreg(env->dregs[6]); 1549 (*regs)[6] = tswapreg(env->dregs[7]); 1550 (*regs)[7] = tswapreg(env->aregs[0]); 1551 (*regs)[8] = tswapreg(env->aregs[1]); 1552 (*regs)[9] = tswapreg(env->aregs[2]); 1553 (*regs)[10] = tswapreg(env->aregs[3]); 1554 (*regs)[11] = tswapreg(env->aregs[4]); 1555 (*regs)[12] = tswapreg(env->aregs[5]); 1556 (*regs)[13] = tswapreg(env->aregs[6]); 1557 (*regs)[14] = tswapreg(env->dregs[0]); 1558 (*regs)[15] = tswapreg(env->aregs[7]); 1559 (*regs)[16] = tswapreg(env->dregs[0]); /* FIXME: orig_d0 */ 1560 (*regs)[17] = tswapreg(env->sr); 1561 (*regs)[18] = tswapreg(env->pc); 1562 (*regs)[19] = 0; /* FIXME: regs->format | regs->vector */ 1563 } 1564 1565 #define USE_ELF_CORE_DUMP 1566 #define ELF_EXEC_PAGESIZE 8192 1567 1568 #endif 1569 1570 #ifdef TARGET_ALPHA 1571 1572 #define ELF_START_MMAP (0x30000000000ULL) 1573 1574 #define ELF_CLASS ELFCLASS64 1575 #define ELF_ARCH EM_ALPHA 1576 1577 static inline void init_thread(struct target_pt_regs *regs, 1578 struct image_info *infop) 1579 { 1580 regs->pc = infop->entry; 1581 regs->ps = 8; 1582 regs->usp = infop->start_stack; 1583 } 1584 1585 #define ELF_EXEC_PAGESIZE 8192 1586 1587 #endif /* TARGET_ALPHA */ 1588 1589 #ifdef TARGET_S390X 1590 1591 #define ELF_START_MMAP (0x20000000000ULL) 1592 1593 #define ELF_CLASS ELFCLASS64 1594 #define ELF_DATA ELFDATA2MSB 1595 #define ELF_ARCH EM_S390 1596 1597 #include "elf.h" 1598 1599 #define ELF_HWCAP get_elf_hwcap() 1600 1601 #define GET_FEATURE(_feat, _hwcap) \ 1602 do { if (s390_has_feat(_feat)) { hwcap |= _hwcap; } } while (0) 1603 1604 static uint32_t get_elf_hwcap(void) 1605 { 1606 /* 1607 * Let's assume we always have esan3 and zarch. 1608 * 31-bit processes can use 64-bit registers (high gprs). 1609 */ 1610 uint32_t hwcap = HWCAP_S390_ESAN3 | HWCAP_S390_ZARCH | HWCAP_S390_HIGH_GPRS; 1611 1612 GET_FEATURE(S390_FEAT_STFLE, HWCAP_S390_STFLE); 1613 GET_FEATURE(S390_FEAT_MSA, HWCAP_S390_MSA); 1614 GET_FEATURE(S390_FEAT_LONG_DISPLACEMENT, HWCAP_S390_LDISP); 1615 GET_FEATURE(S390_FEAT_EXTENDED_IMMEDIATE, HWCAP_S390_EIMM); 1616 if (s390_has_feat(S390_FEAT_EXTENDED_TRANSLATION_3) && 1617 s390_has_feat(S390_FEAT_ETF3_ENH)) { 1618 hwcap |= HWCAP_S390_ETF3EH; 1619 } 1620 GET_FEATURE(S390_FEAT_VECTOR, HWCAP_S390_VXRS); 1621 GET_FEATURE(S390_FEAT_VECTOR_ENH, HWCAP_S390_VXRS_EXT); 1622 1623 return hwcap; 1624 } 1625 1626 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop) 1627 { 1628 regs->psw.addr = infop->entry; 1629 regs->psw.mask = PSW_MASK_64 | PSW_MASK_32; 1630 regs->gprs[15] = infop->start_stack; 1631 } 1632 1633 /* See linux kernel: arch/s390/include/uapi/asm/ptrace.h (s390_regs). */ 1634 #define ELF_NREG 27 1635 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1636 1637 enum { 1638 TARGET_REG_PSWM = 0, 1639 TARGET_REG_PSWA = 1, 1640 TARGET_REG_GPRS = 2, 1641 TARGET_REG_ARS = 18, 1642 TARGET_REG_ORIG_R2 = 26, 1643 }; 1644 1645 static void elf_core_copy_regs(target_elf_gregset_t *regs, 1646 const CPUS390XState *env) 1647 { 1648 int i; 1649 uint32_t *aregs; 1650 1651 (*regs)[TARGET_REG_PSWM] = tswapreg(env->psw.mask); 1652 (*regs)[TARGET_REG_PSWA] = tswapreg(env->psw.addr); 1653 for (i = 0; i < 16; i++) { 1654 (*regs)[TARGET_REG_GPRS + i] = tswapreg(env->regs[i]); 1655 } 1656 aregs = (uint32_t *)&((*regs)[TARGET_REG_ARS]); 1657 for (i = 0; i < 16; i++) { 1658 aregs[i] = tswap32(env->aregs[i]); 1659 } 1660 (*regs)[TARGET_REG_ORIG_R2] = 0; 1661 } 1662 1663 #define USE_ELF_CORE_DUMP 1664 #define ELF_EXEC_PAGESIZE 4096 1665 1666 #endif /* TARGET_S390X */ 1667 1668 #ifdef TARGET_RISCV 1669 1670 #define ELF_START_MMAP 0x80000000 1671 #define ELF_ARCH EM_RISCV 1672 1673 #ifdef TARGET_RISCV32 1674 #define ELF_CLASS ELFCLASS32 1675 #else 1676 #define ELF_CLASS ELFCLASS64 1677 #endif 1678 1679 #define ELF_HWCAP get_elf_hwcap() 1680 1681 static uint32_t get_elf_hwcap(void) 1682 { 1683 #define MISA_BIT(EXT) (1 << (EXT - 'A')) 1684 RISCVCPU *cpu = RISCV_CPU(thread_cpu); 1685 uint32_t mask = MISA_BIT('I') | MISA_BIT('M') | MISA_BIT('A') 1686 | MISA_BIT('F') | MISA_BIT('D') | MISA_BIT('C'); 1687 1688 return cpu->env.misa_ext & mask; 1689 #undef MISA_BIT 1690 } 1691 1692 static inline void init_thread(struct target_pt_regs *regs, 1693 struct image_info *infop) 1694 { 1695 regs->sepc = infop->entry; 1696 regs->sp = infop->start_stack; 1697 } 1698 1699 #define ELF_EXEC_PAGESIZE 4096 1700 1701 #endif /* TARGET_RISCV */ 1702 1703 #ifdef TARGET_HPPA 1704 1705 #define ELF_START_MMAP 0x80000000 1706 #define ELF_CLASS ELFCLASS32 1707 #define ELF_ARCH EM_PARISC 1708 #define ELF_PLATFORM "PARISC" 1709 #define STACK_GROWS_DOWN 0 1710 #define STACK_ALIGNMENT 64 1711 1712 static inline void init_thread(struct target_pt_regs *regs, 1713 struct image_info *infop) 1714 { 1715 regs->iaoq[0] = infop->entry; 1716 regs->iaoq[1] = infop->entry + 4; 1717 regs->gr[23] = 0; 1718 regs->gr[24] = infop->argv; 1719 regs->gr[25] = infop->argc; 1720 /* The top-of-stack contains a linkage buffer. */ 1721 regs->gr[30] = infop->start_stack + 64; 1722 regs->gr[31] = infop->entry; 1723 } 1724 1725 #define LO_COMMPAGE 0 1726 1727 static bool init_guest_commpage(void) 1728 { 1729 void *want = g2h_untagged(LO_COMMPAGE); 1730 void *addr = mmap(want, qemu_host_page_size, PROT_NONE, 1731 MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0); 1732 1733 if (addr == MAP_FAILED) { 1734 perror("Allocating guest commpage"); 1735 exit(EXIT_FAILURE); 1736 } 1737 if (addr != want) { 1738 return false; 1739 } 1740 1741 /* 1742 * On Linux, page zero is normally marked execute only + gateway. 1743 * Normal read or write is supposed to fail (thus PROT_NONE above), 1744 * but specific offsets have kernel code mapped to raise permissions 1745 * and implement syscalls. Here, simply mark the page executable. 1746 * Special case the entry points during translation (see do_page_zero). 1747 */ 1748 page_set_flags(LO_COMMPAGE, LO_COMMPAGE | ~TARGET_PAGE_MASK, 1749 PAGE_EXEC | PAGE_VALID); 1750 return true; 1751 } 1752 1753 #endif /* TARGET_HPPA */ 1754 1755 #ifdef TARGET_XTENSA 1756 1757 #define ELF_START_MMAP 0x20000000 1758 1759 #define ELF_CLASS ELFCLASS32 1760 #define ELF_ARCH EM_XTENSA 1761 1762 static inline void init_thread(struct target_pt_regs *regs, 1763 struct image_info *infop) 1764 { 1765 regs->windowbase = 0; 1766 regs->windowstart = 1; 1767 regs->areg[1] = infop->start_stack; 1768 regs->pc = infop->entry; 1769 if (info_is_fdpic(infop)) { 1770 regs->areg[4] = infop->loadmap_addr; 1771 regs->areg[5] = infop->interpreter_loadmap_addr; 1772 if (infop->interpreter_loadmap_addr) { 1773 regs->areg[6] = infop->interpreter_pt_dynamic_addr; 1774 } else { 1775 regs->areg[6] = infop->pt_dynamic_addr; 1776 } 1777 } 1778 } 1779 1780 /* See linux kernel: arch/xtensa/include/asm/elf.h. */ 1781 #define ELF_NREG 128 1782 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1783 1784 enum { 1785 TARGET_REG_PC, 1786 TARGET_REG_PS, 1787 TARGET_REG_LBEG, 1788 TARGET_REG_LEND, 1789 TARGET_REG_LCOUNT, 1790 TARGET_REG_SAR, 1791 TARGET_REG_WINDOWSTART, 1792 TARGET_REG_WINDOWBASE, 1793 TARGET_REG_THREADPTR, 1794 TARGET_REG_AR0 = 64, 1795 }; 1796 1797 static void elf_core_copy_regs(target_elf_gregset_t *regs, 1798 const CPUXtensaState *env) 1799 { 1800 unsigned i; 1801 1802 (*regs)[TARGET_REG_PC] = tswapreg(env->pc); 1803 (*regs)[TARGET_REG_PS] = tswapreg(env->sregs[PS] & ~PS_EXCM); 1804 (*regs)[TARGET_REG_LBEG] = tswapreg(env->sregs[LBEG]); 1805 (*regs)[TARGET_REG_LEND] = tswapreg(env->sregs[LEND]); 1806 (*regs)[TARGET_REG_LCOUNT] = tswapreg(env->sregs[LCOUNT]); 1807 (*regs)[TARGET_REG_SAR] = tswapreg(env->sregs[SAR]); 1808 (*regs)[TARGET_REG_WINDOWSTART] = tswapreg(env->sregs[WINDOW_START]); 1809 (*regs)[TARGET_REG_WINDOWBASE] = tswapreg(env->sregs[WINDOW_BASE]); 1810 (*regs)[TARGET_REG_THREADPTR] = tswapreg(env->uregs[THREADPTR]); 1811 xtensa_sync_phys_from_window((CPUXtensaState *)env); 1812 for (i = 0; i < env->config->nareg; ++i) { 1813 (*regs)[TARGET_REG_AR0 + i] = tswapreg(env->phys_regs[i]); 1814 } 1815 } 1816 1817 #define USE_ELF_CORE_DUMP 1818 #define ELF_EXEC_PAGESIZE 4096 1819 1820 #endif /* TARGET_XTENSA */ 1821 1822 #ifdef TARGET_HEXAGON 1823 1824 #define ELF_START_MMAP 0x20000000 1825 1826 #define ELF_CLASS ELFCLASS32 1827 #define ELF_ARCH EM_HEXAGON 1828 1829 static inline void init_thread(struct target_pt_regs *regs, 1830 struct image_info *infop) 1831 { 1832 regs->sepc = infop->entry; 1833 regs->sp = infop->start_stack; 1834 } 1835 1836 #endif /* TARGET_HEXAGON */ 1837 1838 #ifndef ELF_BASE_PLATFORM 1839 #define ELF_BASE_PLATFORM (NULL) 1840 #endif 1841 1842 #ifndef ELF_PLATFORM 1843 #define ELF_PLATFORM (NULL) 1844 #endif 1845 1846 #ifndef ELF_MACHINE 1847 #define ELF_MACHINE ELF_ARCH 1848 #endif 1849 1850 #ifndef elf_check_arch 1851 #define elf_check_arch(x) ((x) == ELF_ARCH) 1852 #endif 1853 1854 #ifndef elf_check_abi 1855 #define elf_check_abi(x) (1) 1856 #endif 1857 1858 #ifndef ELF_HWCAP 1859 #define ELF_HWCAP 0 1860 #endif 1861 1862 #ifndef STACK_GROWS_DOWN 1863 #define STACK_GROWS_DOWN 1 1864 #endif 1865 1866 #ifndef STACK_ALIGNMENT 1867 #define STACK_ALIGNMENT 16 1868 #endif 1869 1870 #ifdef TARGET_ABI32 1871 #undef ELF_CLASS 1872 #define ELF_CLASS ELFCLASS32 1873 #undef bswaptls 1874 #define bswaptls(ptr) bswap32s(ptr) 1875 #endif 1876 1877 #ifndef EXSTACK_DEFAULT 1878 #define EXSTACK_DEFAULT false 1879 #endif 1880 1881 #include "elf.h" 1882 1883 /* We must delay the following stanzas until after "elf.h". */ 1884 #if defined(TARGET_AARCH64) 1885 1886 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz, 1887 const uint32_t *data, 1888 struct image_info *info, 1889 Error **errp) 1890 { 1891 if (pr_type == GNU_PROPERTY_AARCH64_FEATURE_1_AND) { 1892 if (pr_datasz != sizeof(uint32_t)) { 1893 error_setg(errp, "Ill-formed GNU_PROPERTY_AARCH64_FEATURE_1_AND"); 1894 return false; 1895 } 1896 /* We will extract GNU_PROPERTY_AARCH64_FEATURE_1_BTI later. */ 1897 info->note_flags = *data; 1898 } 1899 return true; 1900 } 1901 #define ARCH_USE_GNU_PROPERTY 1 1902 1903 #else 1904 1905 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz, 1906 const uint32_t *data, 1907 struct image_info *info, 1908 Error **errp) 1909 { 1910 g_assert_not_reached(); 1911 } 1912 #define ARCH_USE_GNU_PROPERTY 0 1913 1914 #endif 1915 1916 struct exec 1917 { 1918 unsigned int a_info; /* Use macros N_MAGIC, etc for access */ 1919 unsigned int a_text; /* length of text, in bytes */ 1920 unsigned int a_data; /* length of data, in bytes */ 1921 unsigned int a_bss; /* length of uninitialized data area, in bytes */ 1922 unsigned int a_syms; /* length of symbol table data in file, in bytes */ 1923 unsigned int a_entry; /* start address */ 1924 unsigned int a_trsize; /* length of relocation info for text, in bytes */ 1925 unsigned int a_drsize; /* length of relocation info for data, in bytes */ 1926 }; 1927 1928 1929 #define N_MAGIC(exec) ((exec).a_info & 0xffff) 1930 #define OMAGIC 0407 1931 #define NMAGIC 0410 1932 #define ZMAGIC 0413 1933 #define QMAGIC 0314 1934 1935 /* Necessary parameters */ 1936 #define TARGET_ELF_EXEC_PAGESIZE \ 1937 (((eppnt->p_align & ~qemu_host_page_mask) != 0) ? \ 1938 TARGET_PAGE_SIZE : MAX(qemu_host_page_size, TARGET_PAGE_SIZE)) 1939 #define TARGET_ELF_PAGELENGTH(_v) ROUND_UP((_v), TARGET_ELF_EXEC_PAGESIZE) 1940 #define TARGET_ELF_PAGESTART(_v) ((_v) & \ 1941 ~(abi_ulong)(TARGET_ELF_EXEC_PAGESIZE-1)) 1942 #define TARGET_ELF_PAGEOFFSET(_v) ((_v) & (TARGET_ELF_EXEC_PAGESIZE-1)) 1943 1944 #define DLINFO_ITEMS 16 1945 1946 static inline void memcpy_fromfs(void * to, const void * from, unsigned long n) 1947 { 1948 memcpy(to, from, n); 1949 } 1950 1951 #ifdef BSWAP_NEEDED 1952 static void bswap_ehdr(struct elfhdr *ehdr) 1953 { 1954 bswap16s(&ehdr->e_type); /* Object file type */ 1955 bswap16s(&ehdr->e_machine); /* Architecture */ 1956 bswap32s(&ehdr->e_version); /* Object file version */ 1957 bswaptls(&ehdr->e_entry); /* Entry point virtual address */ 1958 bswaptls(&ehdr->e_phoff); /* Program header table file offset */ 1959 bswaptls(&ehdr->e_shoff); /* Section header table file offset */ 1960 bswap32s(&ehdr->e_flags); /* Processor-specific flags */ 1961 bswap16s(&ehdr->e_ehsize); /* ELF header size in bytes */ 1962 bswap16s(&ehdr->e_phentsize); /* Program header table entry size */ 1963 bswap16s(&ehdr->e_phnum); /* Program header table entry count */ 1964 bswap16s(&ehdr->e_shentsize); /* Section header table entry size */ 1965 bswap16s(&ehdr->e_shnum); /* Section header table entry count */ 1966 bswap16s(&ehdr->e_shstrndx); /* Section header string table index */ 1967 } 1968 1969 static void bswap_phdr(struct elf_phdr *phdr, int phnum) 1970 { 1971 int i; 1972 for (i = 0; i < phnum; ++i, ++phdr) { 1973 bswap32s(&phdr->p_type); /* Segment type */ 1974 bswap32s(&phdr->p_flags); /* Segment flags */ 1975 bswaptls(&phdr->p_offset); /* Segment file offset */ 1976 bswaptls(&phdr->p_vaddr); /* Segment virtual address */ 1977 bswaptls(&phdr->p_paddr); /* Segment physical address */ 1978 bswaptls(&phdr->p_filesz); /* Segment size in file */ 1979 bswaptls(&phdr->p_memsz); /* Segment size in memory */ 1980 bswaptls(&phdr->p_align); /* Segment alignment */ 1981 } 1982 } 1983 1984 static void bswap_shdr(struct elf_shdr *shdr, int shnum) 1985 { 1986 int i; 1987 for (i = 0; i < shnum; ++i, ++shdr) { 1988 bswap32s(&shdr->sh_name); 1989 bswap32s(&shdr->sh_type); 1990 bswaptls(&shdr->sh_flags); 1991 bswaptls(&shdr->sh_addr); 1992 bswaptls(&shdr->sh_offset); 1993 bswaptls(&shdr->sh_size); 1994 bswap32s(&shdr->sh_link); 1995 bswap32s(&shdr->sh_info); 1996 bswaptls(&shdr->sh_addralign); 1997 bswaptls(&shdr->sh_entsize); 1998 } 1999 } 2000 2001 static void bswap_sym(struct elf_sym *sym) 2002 { 2003 bswap32s(&sym->st_name); 2004 bswaptls(&sym->st_value); 2005 bswaptls(&sym->st_size); 2006 bswap16s(&sym->st_shndx); 2007 } 2008 2009 #ifdef TARGET_MIPS 2010 static void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags) 2011 { 2012 bswap16s(&abiflags->version); 2013 bswap32s(&abiflags->ases); 2014 bswap32s(&abiflags->isa_ext); 2015 bswap32s(&abiflags->flags1); 2016 bswap32s(&abiflags->flags2); 2017 } 2018 #endif 2019 #else 2020 static inline void bswap_ehdr(struct elfhdr *ehdr) { } 2021 static inline void bswap_phdr(struct elf_phdr *phdr, int phnum) { } 2022 static inline void bswap_shdr(struct elf_shdr *shdr, int shnum) { } 2023 static inline void bswap_sym(struct elf_sym *sym) { } 2024 #ifdef TARGET_MIPS 2025 static inline void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags) { } 2026 #endif 2027 #endif 2028 2029 #ifdef USE_ELF_CORE_DUMP 2030 static int elf_core_dump(int, const CPUArchState *); 2031 #endif /* USE_ELF_CORE_DUMP */ 2032 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias); 2033 2034 /* Verify the portions of EHDR within E_IDENT for the target. 2035 This can be performed before bswapping the entire header. */ 2036 static bool elf_check_ident(struct elfhdr *ehdr) 2037 { 2038 return (ehdr->e_ident[EI_MAG0] == ELFMAG0 2039 && ehdr->e_ident[EI_MAG1] == ELFMAG1 2040 && ehdr->e_ident[EI_MAG2] == ELFMAG2 2041 && ehdr->e_ident[EI_MAG3] == ELFMAG3 2042 && ehdr->e_ident[EI_CLASS] == ELF_CLASS 2043 && ehdr->e_ident[EI_DATA] == ELF_DATA 2044 && ehdr->e_ident[EI_VERSION] == EV_CURRENT); 2045 } 2046 2047 /* Verify the portions of EHDR outside of E_IDENT for the target. 2048 This has to wait until after bswapping the header. */ 2049 static bool elf_check_ehdr(struct elfhdr *ehdr) 2050 { 2051 return (elf_check_arch(ehdr->e_machine) 2052 && elf_check_abi(ehdr->e_flags) 2053 && ehdr->e_ehsize == sizeof(struct elfhdr) 2054 && ehdr->e_phentsize == sizeof(struct elf_phdr) 2055 && (ehdr->e_type == ET_EXEC || ehdr->e_type == ET_DYN)); 2056 } 2057 2058 /* 2059 * 'copy_elf_strings()' copies argument/envelope strings from user 2060 * memory to free pages in kernel mem. These are in a format ready 2061 * to be put directly into the top of new user memory. 2062 * 2063 */ 2064 static abi_ulong copy_elf_strings(int argc, char **argv, char *scratch, 2065 abi_ulong p, abi_ulong stack_limit) 2066 { 2067 char *tmp; 2068 int len, i; 2069 abi_ulong top = p; 2070 2071 if (!p) { 2072 return 0; /* bullet-proofing */ 2073 } 2074 2075 if (STACK_GROWS_DOWN) { 2076 int offset = ((p - 1) % TARGET_PAGE_SIZE) + 1; 2077 for (i = argc - 1; i >= 0; --i) { 2078 tmp = argv[i]; 2079 if (!tmp) { 2080 fprintf(stderr, "VFS: argc is wrong"); 2081 exit(-1); 2082 } 2083 len = strlen(tmp) + 1; 2084 tmp += len; 2085 2086 if (len > (p - stack_limit)) { 2087 return 0; 2088 } 2089 while (len) { 2090 int bytes_to_copy = (len > offset) ? offset : len; 2091 tmp -= bytes_to_copy; 2092 p -= bytes_to_copy; 2093 offset -= bytes_to_copy; 2094 len -= bytes_to_copy; 2095 2096 memcpy_fromfs(scratch + offset, tmp, bytes_to_copy); 2097 2098 if (offset == 0) { 2099 memcpy_to_target(p, scratch, top - p); 2100 top = p; 2101 offset = TARGET_PAGE_SIZE; 2102 } 2103 } 2104 } 2105 if (p != top) { 2106 memcpy_to_target(p, scratch + offset, top - p); 2107 } 2108 } else { 2109 int remaining = TARGET_PAGE_SIZE - (p % TARGET_PAGE_SIZE); 2110 for (i = 0; i < argc; ++i) { 2111 tmp = argv[i]; 2112 if (!tmp) { 2113 fprintf(stderr, "VFS: argc is wrong"); 2114 exit(-1); 2115 } 2116 len = strlen(tmp) + 1; 2117 if (len > (stack_limit - p)) { 2118 return 0; 2119 } 2120 while (len) { 2121 int bytes_to_copy = (len > remaining) ? remaining : len; 2122 2123 memcpy_fromfs(scratch + (p - top), tmp, bytes_to_copy); 2124 2125 tmp += bytes_to_copy; 2126 remaining -= bytes_to_copy; 2127 p += bytes_to_copy; 2128 len -= bytes_to_copy; 2129 2130 if (remaining == 0) { 2131 memcpy_to_target(top, scratch, p - top); 2132 top = p; 2133 remaining = TARGET_PAGE_SIZE; 2134 } 2135 } 2136 } 2137 if (p != top) { 2138 memcpy_to_target(top, scratch, p - top); 2139 } 2140 } 2141 2142 return p; 2143 } 2144 2145 /* Older linux kernels provide up to MAX_ARG_PAGES (default: 32) of 2146 * argument/environment space. Newer kernels (>2.6.33) allow more, 2147 * dependent on stack size, but guarantee at least 32 pages for 2148 * backwards compatibility. 2149 */ 2150 #define STACK_LOWER_LIMIT (32 * TARGET_PAGE_SIZE) 2151 2152 static abi_ulong setup_arg_pages(struct linux_binprm *bprm, 2153 struct image_info *info) 2154 { 2155 abi_ulong size, error, guard; 2156 int prot; 2157 2158 size = guest_stack_size; 2159 if (size < STACK_LOWER_LIMIT) { 2160 size = STACK_LOWER_LIMIT; 2161 } 2162 2163 if (STACK_GROWS_DOWN) { 2164 guard = TARGET_PAGE_SIZE; 2165 if (guard < qemu_real_host_page_size()) { 2166 guard = qemu_real_host_page_size(); 2167 } 2168 } else { 2169 /* no guard page for hppa target where stack grows upwards. */ 2170 guard = 0; 2171 } 2172 2173 prot = PROT_READ | PROT_WRITE; 2174 if (info->exec_stack) { 2175 prot |= PROT_EXEC; 2176 } 2177 error = target_mmap(0, size + guard, prot, 2178 MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); 2179 if (error == -1) { 2180 perror("mmap stack"); 2181 exit(-1); 2182 } 2183 2184 /* We reserve one extra page at the top of the stack as guard. */ 2185 if (STACK_GROWS_DOWN) { 2186 target_mprotect(error, guard, PROT_NONE); 2187 info->stack_limit = error + guard; 2188 return info->stack_limit + size - sizeof(void *); 2189 } else { 2190 info->stack_limit = error + size; 2191 return error; 2192 } 2193 } 2194 2195 /* Map and zero the bss. We need to explicitly zero any fractional pages 2196 after the data section (i.e. bss). */ 2197 static void zero_bss(abi_ulong elf_bss, abi_ulong last_bss, int prot) 2198 { 2199 uintptr_t host_start, host_map_start, host_end; 2200 2201 last_bss = TARGET_PAGE_ALIGN(last_bss); 2202 2203 /* ??? There is confusion between qemu_real_host_page_size and 2204 qemu_host_page_size here and elsewhere in target_mmap, which 2205 may lead to the end of the data section mapping from the file 2206 not being mapped. At least there was an explicit test and 2207 comment for that here, suggesting that "the file size must 2208 be known". The comment probably pre-dates the introduction 2209 of the fstat system call in target_mmap which does in fact 2210 find out the size. What isn't clear is if the workaround 2211 here is still actually needed. For now, continue with it, 2212 but merge it with the "normal" mmap that would allocate the bss. */ 2213 2214 host_start = (uintptr_t) g2h_untagged(elf_bss); 2215 host_end = (uintptr_t) g2h_untagged(last_bss); 2216 host_map_start = REAL_HOST_PAGE_ALIGN(host_start); 2217 2218 if (host_map_start < host_end) { 2219 void *p = mmap((void *)host_map_start, host_end - host_map_start, 2220 prot, MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); 2221 if (p == MAP_FAILED) { 2222 perror("cannot mmap brk"); 2223 exit(-1); 2224 } 2225 } 2226 2227 /* Ensure that the bss page(s) are valid */ 2228 if ((page_get_flags(last_bss-1) & prot) != prot) { 2229 page_set_flags(elf_bss & TARGET_PAGE_MASK, last_bss - 1, 2230 prot | PAGE_VALID); 2231 } 2232 2233 if (host_start < host_map_start) { 2234 memset((void *)host_start, 0, host_map_start - host_start); 2235 } 2236 } 2237 2238 #if defined(TARGET_ARM) 2239 static int elf_is_fdpic(struct elfhdr *exec) 2240 { 2241 return exec->e_ident[EI_OSABI] == ELFOSABI_ARM_FDPIC; 2242 } 2243 #elif defined(TARGET_XTENSA) 2244 static int elf_is_fdpic(struct elfhdr *exec) 2245 { 2246 return exec->e_ident[EI_OSABI] == ELFOSABI_XTENSA_FDPIC; 2247 } 2248 #else 2249 /* Default implementation, always false. */ 2250 static int elf_is_fdpic(struct elfhdr *exec) 2251 { 2252 return 0; 2253 } 2254 #endif 2255 2256 static abi_ulong loader_build_fdpic_loadmap(struct image_info *info, abi_ulong sp) 2257 { 2258 uint16_t n; 2259 struct elf32_fdpic_loadseg *loadsegs = info->loadsegs; 2260 2261 /* elf32_fdpic_loadseg */ 2262 n = info->nsegs; 2263 while (n--) { 2264 sp -= 12; 2265 put_user_u32(loadsegs[n].addr, sp+0); 2266 put_user_u32(loadsegs[n].p_vaddr, sp+4); 2267 put_user_u32(loadsegs[n].p_memsz, sp+8); 2268 } 2269 2270 /* elf32_fdpic_loadmap */ 2271 sp -= 4; 2272 put_user_u16(0, sp+0); /* version */ 2273 put_user_u16(info->nsegs, sp+2); /* nsegs */ 2274 2275 info->personality = PER_LINUX_FDPIC; 2276 info->loadmap_addr = sp; 2277 2278 return sp; 2279 } 2280 2281 static abi_ulong create_elf_tables(abi_ulong p, int argc, int envc, 2282 struct elfhdr *exec, 2283 struct image_info *info, 2284 struct image_info *interp_info) 2285 { 2286 abi_ulong sp; 2287 abi_ulong u_argc, u_argv, u_envp, u_auxv; 2288 int size; 2289 int i; 2290 abi_ulong u_rand_bytes; 2291 uint8_t k_rand_bytes[16]; 2292 abi_ulong u_platform, u_base_platform; 2293 const char *k_platform, *k_base_platform; 2294 const int n = sizeof(elf_addr_t); 2295 2296 sp = p; 2297 2298 /* Needs to be before we load the env/argc/... */ 2299 if (elf_is_fdpic(exec)) { 2300 /* Need 4 byte alignment for these structs */ 2301 sp &= ~3; 2302 sp = loader_build_fdpic_loadmap(info, sp); 2303 info->other_info = interp_info; 2304 if (interp_info) { 2305 interp_info->other_info = info; 2306 sp = loader_build_fdpic_loadmap(interp_info, sp); 2307 info->interpreter_loadmap_addr = interp_info->loadmap_addr; 2308 info->interpreter_pt_dynamic_addr = interp_info->pt_dynamic_addr; 2309 } else { 2310 info->interpreter_loadmap_addr = 0; 2311 info->interpreter_pt_dynamic_addr = 0; 2312 } 2313 } 2314 2315 u_base_platform = 0; 2316 k_base_platform = ELF_BASE_PLATFORM; 2317 if (k_base_platform) { 2318 size_t len = strlen(k_base_platform) + 1; 2319 if (STACK_GROWS_DOWN) { 2320 sp -= (len + n - 1) & ~(n - 1); 2321 u_base_platform = sp; 2322 /* FIXME - check return value of memcpy_to_target() for failure */ 2323 memcpy_to_target(sp, k_base_platform, len); 2324 } else { 2325 memcpy_to_target(sp, k_base_platform, len); 2326 u_base_platform = sp; 2327 sp += len + 1; 2328 } 2329 } 2330 2331 u_platform = 0; 2332 k_platform = ELF_PLATFORM; 2333 if (k_platform) { 2334 size_t len = strlen(k_platform) + 1; 2335 if (STACK_GROWS_DOWN) { 2336 sp -= (len + n - 1) & ~(n - 1); 2337 u_platform = sp; 2338 /* FIXME - check return value of memcpy_to_target() for failure */ 2339 memcpy_to_target(sp, k_platform, len); 2340 } else { 2341 memcpy_to_target(sp, k_platform, len); 2342 u_platform = sp; 2343 sp += len + 1; 2344 } 2345 } 2346 2347 /* Provide 16 byte alignment for the PRNG, and basic alignment for 2348 * the argv and envp pointers. 2349 */ 2350 if (STACK_GROWS_DOWN) { 2351 sp = QEMU_ALIGN_DOWN(sp, 16); 2352 } else { 2353 sp = QEMU_ALIGN_UP(sp, 16); 2354 } 2355 2356 /* 2357 * Generate 16 random bytes for userspace PRNG seeding. 2358 */ 2359 qemu_guest_getrandom_nofail(k_rand_bytes, sizeof(k_rand_bytes)); 2360 if (STACK_GROWS_DOWN) { 2361 sp -= 16; 2362 u_rand_bytes = sp; 2363 /* FIXME - check return value of memcpy_to_target() for failure */ 2364 memcpy_to_target(sp, k_rand_bytes, 16); 2365 } else { 2366 memcpy_to_target(sp, k_rand_bytes, 16); 2367 u_rand_bytes = sp; 2368 sp += 16; 2369 } 2370 2371 size = (DLINFO_ITEMS + 1) * 2; 2372 if (k_base_platform) 2373 size += 2; 2374 if (k_platform) 2375 size += 2; 2376 #ifdef DLINFO_ARCH_ITEMS 2377 size += DLINFO_ARCH_ITEMS * 2; 2378 #endif 2379 #ifdef ELF_HWCAP2 2380 size += 2; 2381 #endif 2382 info->auxv_len = size * n; 2383 2384 size += envc + argc + 2; 2385 size += 1; /* argc itself */ 2386 size *= n; 2387 2388 /* Allocate space and finalize stack alignment for entry now. */ 2389 if (STACK_GROWS_DOWN) { 2390 u_argc = QEMU_ALIGN_DOWN(sp - size, STACK_ALIGNMENT); 2391 sp = u_argc; 2392 } else { 2393 u_argc = sp; 2394 sp = QEMU_ALIGN_UP(sp + size, STACK_ALIGNMENT); 2395 } 2396 2397 u_argv = u_argc + n; 2398 u_envp = u_argv + (argc + 1) * n; 2399 u_auxv = u_envp + (envc + 1) * n; 2400 info->saved_auxv = u_auxv; 2401 info->argc = argc; 2402 info->envc = envc; 2403 info->argv = u_argv; 2404 info->envp = u_envp; 2405 2406 /* This is correct because Linux defines 2407 * elf_addr_t as Elf32_Off / Elf64_Off 2408 */ 2409 #define NEW_AUX_ENT(id, val) do { \ 2410 put_user_ual(id, u_auxv); u_auxv += n; \ 2411 put_user_ual(val, u_auxv); u_auxv += n; \ 2412 } while(0) 2413 2414 #ifdef ARCH_DLINFO 2415 /* 2416 * ARCH_DLINFO must come first so platform specific code can enforce 2417 * special alignment requirements on the AUXV if necessary (eg. PPC). 2418 */ 2419 ARCH_DLINFO; 2420 #endif 2421 /* There must be exactly DLINFO_ITEMS entries here, or the assert 2422 * on info->auxv_len will trigger. 2423 */ 2424 NEW_AUX_ENT(AT_PHDR, (abi_ulong)(info->load_addr + exec->e_phoff)); 2425 NEW_AUX_ENT(AT_PHENT, (abi_ulong)(sizeof (struct elf_phdr))); 2426 NEW_AUX_ENT(AT_PHNUM, (abi_ulong)(exec->e_phnum)); 2427 if ((info->alignment & ~qemu_host_page_mask) != 0) { 2428 /* Target doesn't support host page size alignment */ 2429 NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(TARGET_PAGE_SIZE)); 2430 } else { 2431 NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(MAX(TARGET_PAGE_SIZE, 2432 qemu_host_page_size))); 2433 } 2434 NEW_AUX_ENT(AT_BASE, (abi_ulong)(interp_info ? interp_info->load_addr : 0)); 2435 NEW_AUX_ENT(AT_FLAGS, (abi_ulong)0); 2436 NEW_AUX_ENT(AT_ENTRY, info->entry); 2437 NEW_AUX_ENT(AT_UID, (abi_ulong) getuid()); 2438 NEW_AUX_ENT(AT_EUID, (abi_ulong) geteuid()); 2439 NEW_AUX_ENT(AT_GID, (abi_ulong) getgid()); 2440 NEW_AUX_ENT(AT_EGID, (abi_ulong) getegid()); 2441 NEW_AUX_ENT(AT_HWCAP, (abi_ulong) ELF_HWCAP); 2442 NEW_AUX_ENT(AT_CLKTCK, (abi_ulong) sysconf(_SC_CLK_TCK)); 2443 NEW_AUX_ENT(AT_RANDOM, (abi_ulong) u_rand_bytes); 2444 NEW_AUX_ENT(AT_SECURE, (abi_ulong) qemu_getauxval(AT_SECURE)); 2445 NEW_AUX_ENT(AT_EXECFN, info->file_string); 2446 2447 #ifdef ELF_HWCAP2 2448 NEW_AUX_ENT(AT_HWCAP2, (abi_ulong) ELF_HWCAP2); 2449 #endif 2450 2451 if (u_base_platform) { 2452 NEW_AUX_ENT(AT_BASE_PLATFORM, u_base_platform); 2453 } 2454 if (u_platform) { 2455 NEW_AUX_ENT(AT_PLATFORM, u_platform); 2456 } 2457 NEW_AUX_ENT (AT_NULL, 0); 2458 #undef NEW_AUX_ENT 2459 2460 /* Check that our initial calculation of the auxv length matches how much 2461 * we actually put into it. 2462 */ 2463 assert(info->auxv_len == u_auxv - info->saved_auxv); 2464 2465 put_user_ual(argc, u_argc); 2466 2467 p = info->arg_strings; 2468 for (i = 0; i < argc; ++i) { 2469 put_user_ual(p, u_argv); 2470 u_argv += n; 2471 p += target_strlen(p) + 1; 2472 } 2473 put_user_ual(0, u_argv); 2474 2475 p = info->env_strings; 2476 for (i = 0; i < envc; ++i) { 2477 put_user_ual(p, u_envp); 2478 u_envp += n; 2479 p += target_strlen(p) + 1; 2480 } 2481 put_user_ual(0, u_envp); 2482 2483 return sp; 2484 } 2485 2486 #if defined(HI_COMMPAGE) 2487 #define LO_COMMPAGE -1 2488 #elif defined(LO_COMMPAGE) 2489 #define HI_COMMPAGE 0 2490 #else 2491 #define HI_COMMPAGE 0 2492 #define LO_COMMPAGE -1 2493 #ifndef INIT_GUEST_COMMPAGE 2494 #define init_guest_commpage() true 2495 #endif 2496 #endif 2497 2498 static void pgb_fail_in_use(const char *image_name) 2499 { 2500 error_report("%s: requires virtual address space that is in use " 2501 "(omit the -B option or choose a different value)", 2502 image_name); 2503 exit(EXIT_FAILURE); 2504 } 2505 2506 static void pgb_have_guest_base(const char *image_name, abi_ulong guest_loaddr, 2507 abi_ulong guest_hiaddr, long align) 2508 { 2509 const int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE; 2510 void *addr, *test; 2511 2512 if (!QEMU_IS_ALIGNED(guest_base, align)) { 2513 fprintf(stderr, "Requested guest base %p does not satisfy " 2514 "host minimum alignment (0x%lx)\n", 2515 (void *)guest_base, align); 2516 exit(EXIT_FAILURE); 2517 } 2518 2519 /* Sanity check the guest binary. */ 2520 if (reserved_va) { 2521 if (guest_hiaddr > reserved_va) { 2522 error_report("%s: requires more than reserved virtual " 2523 "address space (0x%" PRIx64 " > 0x%lx)", 2524 image_name, (uint64_t)guest_hiaddr, reserved_va); 2525 exit(EXIT_FAILURE); 2526 } 2527 } else { 2528 #if HOST_LONG_BITS < TARGET_ABI_BITS 2529 if ((guest_hiaddr - guest_base) > ~(uintptr_t)0) { 2530 error_report("%s: requires more virtual address space " 2531 "than the host can provide (0x%" PRIx64 ")", 2532 image_name, (uint64_t)guest_hiaddr + 1 - guest_base); 2533 exit(EXIT_FAILURE); 2534 } 2535 #endif 2536 } 2537 2538 /* 2539 * Expand the allocation to the entire reserved_va. 2540 * Exclude the mmap_min_addr hole. 2541 */ 2542 if (reserved_va) { 2543 guest_loaddr = (guest_base >= mmap_min_addr ? 0 2544 : mmap_min_addr - guest_base); 2545 guest_hiaddr = reserved_va; 2546 } 2547 2548 /* Reserve the address space for the binary, or reserved_va. */ 2549 test = g2h_untagged(guest_loaddr); 2550 addr = mmap(test, guest_hiaddr - guest_loaddr + 1, PROT_NONE, flags, -1, 0); 2551 if (test != addr) { 2552 pgb_fail_in_use(image_name); 2553 } 2554 qemu_log_mask(CPU_LOG_PAGE, 2555 "%s: base @ %p for %" PRIu64 " bytes\n", 2556 __func__, addr, (uint64_t)guest_hiaddr - guest_loaddr + 1); 2557 } 2558 2559 /** 2560 * pgd_find_hole_fallback: potential mmap address 2561 * @guest_size: size of available space 2562 * @brk: location of break 2563 * @align: memory alignment 2564 * 2565 * This is a fallback method for finding a hole in the host address 2566 * space if we don't have the benefit of being able to access 2567 * /proc/self/map. It can potentially take a very long time as we can 2568 * only dumbly iterate up the host address space seeing if the 2569 * allocation would work. 2570 */ 2571 static uintptr_t pgd_find_hole_fallback(uintptr_t guest_size, uintptr_t brk, 2572 long align, uintptr_t offset) 2573 { 2574 uintptr_t base; 2575 2576 /* Start (aligned) at the bottom and work our way up */ 2577 base = ROUND_UP(mmap_min_addr, align); 2578 2579 while (true) { 2580 uintptr_t align_start, end; 2581 align_start = ROUND_UP(base, align); 2582 end = align_start + guest_size + offset; 2583 2584 /* if brk is anywhere in the range give ourselves some room to grow. */ 2585 if (align_start <= brk && brk < end) { 2586 base = brk + (16 * MiB); 2587 continue; 2588 } else if (align_start + guest_size < align_start) { 2589 /* we have run out of space */ 2590 return -1; 2591 } else { 2592 int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE | 2593 MAP_FIXED_NOREPLACE; 2594 void * mmap_start = mmap((void *) align_start, guest_size, 2595 PROT_NONE, flags, -1, 0); 2596 if (mmap_start != MAP_FAILED) { 2597 munmap(mmap_start, guest_size); 2598 if (mmap_start == (void *) align_start) { 2599 qemu_log_mask(CPU_LOG_PAGE, 2600 "%s: base @ %p for %" PRIdPTR" bytes\n", 2601 __func__, mmap_start + offset, guest_size); 2602 return (uintptr_t) mmap_start + offset; 2603 } 2604 } 2605 base += qemu_host_page_size; 2606 } 2607 } 2608 } 2609 2610 /* Return value for guest_base, or -1 if no hole found. */ 2611 static uintptr_t pgb_find_hole(uintptr_t guest_loaddr, uintptr_t guest_size, 2612 long align, uintptr_t offset) 2613 { 2614 GSList *maps, *iter; 2615 uintptr_t this_start, this_end, next_start, brk; 2616 intptr_t ret = -1; 2617 2618 assert(QEMU_IS_ALIGNED(guest_loaddr, align)); 2619 2620 maps = read_self_maps(); 2621 2622 /* Read brk after we've read the maps, which will malloc. */ 2623 brk = (uintptr_t)sbrk(0); 2624 2625 if (!maps) { 2626 return pgd_find_hole_fallback(guest_size, brk, align, offset); 2627 } 2628 2629 /* The first hole is before the first map entry. */ 2630 this_start = mmap_min_addr; 2631 2632 for (iter = maps; iter; 2633 this_start = next_start, iter = g_slist_next(iter)) { 2634 uintptr_t align_start, hole_size; 2635 2636 this_end = ((MapInfo *)iter->data)->start; 2637 next_start = ((MapInfo *)iter->data)->end; 2638 align_start = ROUND_UP(this_start + offset, align); 2639 2640 /* Skip holes that are too small. */ 2641 if (align_start >= this_end) { 2642 continue; 2643 } 2644 hole_size = this_end - align_start; 2645 if (hole_size < guest_size) { 2646 continue; 2647 } 2648 2649 /* If this hole contains brk, give ourselves some room to grow. */ 2650 if (this_start <= brk && brk < this_end) { 2651 hole_size -= guest_size; 2652 if (sizeof(uintptr_t) == 8 && hole_size >= 1 * GiB) { 2653 align_start += 1 * GiB; 2654 } else if (hole_size >= 16 * MiB) { 2655 align_start += 16 * MiB; 2656 } else { 2657 align_start = (this_end - guest_size) & -align; 2658 if (align_start < this_start) { 2659 continue; 2660 } 2661 } 2662 } 2663 2664 /* Record the lowest successful match. */ 2665 if (ret < 0) { 2666 ret = align_start; 2667 } 2668 /* If this hole contains the identity map, select it. */ 2669 if (align_start <= guest_loaddr && 2670 guest_loaddr + guest_size <= this_end) { 2671 ret = 0; 2672 } 2673 /* If this hole ends above the identity map, stop looking. */ 2674 if (this_end >= guest_loaddr) { 2675 break; 2676 } 2677 } 2678 free_self_maps(maps); 2679 2680 if (ret != -1) { 2681 qemu_log_mask(CPU_LOG_PAGE, "%s: base @ %" PRIxPTR 2682 " for %" PRIuPTR " bytes\n", 2683 __func__, ret, guest_size); 2684 } 2685 2686 return ret; 2687 } 2688 2689 static void pgb_static(const char *image_name, abi_ulong orig_loaddr, 2690 abi_ulong orig_hiaddr, long align) 2691 { 2692 uintptr_t loaddr = orig_loaddr; 2693 uintptr_t hiaddr = orig_hiaddr; 2694 uintptr_t offset = 0; 2695 uintptr_t addr; 2696 2697 if (hiaddr != orig_hiaddr) { 2698 error_report("%s: requires virtual address space that the " 2699 "host cannot provide (0x%" PRIx64 ")", 2700 image_name, (uint64_t)orig_hiaddr + 1); 2701 exit(EXIT_FAILURE); 2702 } 2703 2704 loaddr &= -align; 2705 if (HI_COMMPAGE) { 2706 /* 2707 * Extend the allocation to include the commpage. 2708 * For a 64-bit host, this is just 4GiB; for a 32-bit host we 2709 * need to ensure there is space bellow the guest_base so we 2710 * can map the commpage in the place needed when the address 2711 * arithmetic wraps around. 2712 */ 2713 if (sizeof(uintptr_t) == 8 || loaddr >= 0x80000000u) { 2714 hiaddr = UINT32_MAX; 2715 } else { 2716 offset = -(HI_COMMPAGE & -align); 2717 } 2718 } else if (LO_COMMPAGE != -1) { 2719 loaddr = MIN(loaddr, LO_COMMPAGE & -align); 2720 } 2721 2722 addr = pgb_find_hole(loaddr, hiaddr - loaddr + 1, align, offset); 2723 if (addr == -1) { 2724 /* 2725 * If HI_COMMPAGE, there *might* be a non-consecutive allocation 2726 * that can satisfy both. But as the normal arm32 link base address 2727 * is ~32k, and we extend down to include the commpage, making the 2728 * overhead only ~96k, this is unlikely. 2729 */ 2730 error_report("%s: Unable to allocate %#zx bytes of " 2731 "virtual address space", image_name, 2732 (size_t)(hiaddr - loaddr)); 2733 exit(EXIT_FAILURE); 2734 } 2735 2736 guest_base = addr; 2737 2738 qemu_log_mask(CPU_LOG_PAGE, "%s: base @ %"PRIxPTR" for %" PRIuPTR" bytes\n", 2739 __func__, addr, hiaddr - loaddr); 2740 } 2741 2742 static void pgb_dynamic(const char *image_name, long align) 2743 { 2744 /* 2745 * The executable is dynamic and does not require a fixed address. 2746 * All we need is a commpage that satisfies align. 2747 * If we do not need a commpage, leave guest_base == 0. 2748 */ 2749 if (HI_COMMPAGE) { 2750 uintptr_t addr, commpage; 2751 2752 /* 64-bit hosts should have used reserved_va. */ 2753 assert(sizeof(uintptr_t) == 4); 2754 2755 /* 2756 * By putting the commpage at the first hole, that puts guest_base 2757 * just above that, and maximises the positive guest addresses. 2758 */ 2759 commpage = HI_COMMPAGE & -align; 2760 addr = pgb_find_hole(commpage, -commpage, align, 0); 2761 assert(addr != -1); 2762 guest_base = addr; 2763 } 2764 } 2765 2766 static void pgb_reserved_va(const char *image_name, abi_ulong guest_loaddr, 2767 abi_ulong guest_hiaddr, long align) 2768 { 2769 int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE; 2770 void *addr, *test; 2771 2772 if (guest_hiaddr > reserved_va) { 2773 error_report("%s: requires more than reserved virtual " 2774 "address space (0x%" PRIx64 " > 0x%lx)", 2775 image_name, (uint64_t)guest_hiaddr, reserved_va); 2776 exit(EXIT_FAILURE); 2777 } 2778 2779 /* Widen the "image" to the entire reserved address space. */ 2780 pgb_static(image_name, 0, reserved_va, align); 2781 2782 /* osdep.h defines this as 0 if it's missing */ 2783 flags |= MAP_FIXED_NOREPLACE; 2784 2785 /* Reserve the memory on the host. */ 2786 assert(guest_base != 0); 2787 test = g2h_untagged(0); 2788 addr = mmap(test, reserved_va + 1, PROT_NONE, flags, -1, 0); 2789 if (addr == MAP_FAILED || addr != test) { 2790 error_report("Unable to reserve 0x%lx bytes of virtual address " 2791 "space at %p (%s) for use as guest address space (check your " 2792 "virtual memory ulimit setting, min_mmap_addr or reserve less " 2793 "using -R option)", reserved_va + 1, test, strerror(errno)); 2794 exit(EXIT_FAILURE); 2795 } 2796 2797 qemu_log_mask(CPU_LOG_PAGE, "%s: base @ %p for %lu bytes\n", 2798 __func__, addr, reserved_va + 1); 2799 } 2800 2801 void probe_guest_base(const char *image_name, abi_ulong guest_loaddr, 2802 abi_ulong guest_hiaddr) 2803 { 2804 /* In order to use host shmat, we must be able to honor SHMLBA. */ 2805 uintptr_t align = MAX(SHMLBA, qemu_host_page_size); 2806 2807 if (have_guest_base) { 2808 pgb_have_guest_base(image_name, guest_loaddr, guest_hiaddr, align); 2809 } else if (reserved_va) { 2810 pgb_reserved_va(image_name, guest_loaddr, guest_hiaddr, align); 2811 } else if (guest_loaddr) { 2812 pgb_static(image_name, guest_loaddr, guest_hiaddr, align); 2813 } else { 2814 pgb_dynamic(image_name, align); 2815 } 2816 2817 /* Reserve and initialize the commpage. */ 2818 if (!init_guest_commpage()) { 2819 /* 2820 * With have_guest_base, the user has selected the address and 2821 * we are trying to work with that. Otherwise, we have selected 2822 * free space and init_guest_commpage must succeeded. 2823 */ 2824 assert(have_guest_base); 2825 pgb_fail_in_use(image_name); 2826 } 2827 2828 assert(QEMU_IS_ALIGNED(guest_base, align)); 2829 qemu_log_mask(CPU_LOG_PAGE, "Locating guest address space " 2830 "@ 0x%" PRIx64 "\n", (uint64_t)guest_base); 2831 } 2832 2833 enum { 2834 /* The string "GNU\0" as a magic number. */ 2835 GNU0_MAGIC = const_le32('G' | 'N' << 8 | 'U' << 16), 2836 NOTE_DATA_SZ = 1 * KiB, 2837 NOTE_NAME_SZ = 4, 2838 ELF_GNU_PROPERTY_ALIGN = ELF_CLASS == ELFCLASS32 ? 4 : 8, 2839 }; 2840 2841 /* 2842 * Process a single gnu_property entry. 2843 * Return false for error. 2844 */ 2845 static bool parse_elf_property(const uint32_t *data, int *off, int datasz, 2846 struct image_info *info, bool have_prev_type, 2847 uint32_t *prev_type, Error **errp) 2848 { 2849 uint32_t pr_type, pr_datasz, step; 2850 2851 if (*off > datasz || !QEMU_IS_ALIGNED(*off, ELF_GNU_PROPERTY_ALIGN)) { 2852 goto error_data; 2853 } 2854 datasz -= *off; 2855 data += *off / sizeof(uint32_t); 2856 2857 if (datasz < 2 * sizeof(uint32_t)) { 2858 goto error_data; 2859 } 2860 pr_type = data[0]; 2861 pr_datasz = data[1]; 2862 data += 2; 2863 datasz -= 2 * sizeof(uint32_t); 2864 step = ROUND_UP(pr_datasz, ELF_GNU_PROPERTY_ALIGN); 2865 if (step > datasz) { 2866 goto error_data; 2867 } 2868 2869 /* Properties are supposed to be unique and sorted on pr_type. */ 2870 if (have_prev_type && pr_type <= *prev_type) { 2871 if (pr_type == *prev_type) { 2872 error_setg(errp, "Duplicate property in PT_GNU_PROPERTY"); 2873 } else { 2874 error_setg(errp, "Unsorted property in PT_GNU_PROPERTY"); 2875 } 2876 return false; 2877 } 2878 *prev_type = pr_type; 2879 2880 if (!arch_parse_elf_property(pr_type, pr_datasz, data, info, errp)) { 2881 return false; 2882 } 2883 2884 *off += 2 * sizeof(uint32_t) + step; 2885 return true; 2886 2887 error_data: 2888 error_setg(errp, "Ill-formed property in PT_GNU_PROPERTY"); 2889 return false; 2890 } 2891 2892 /* Process NT_GNU_PROPERTY_TYPE_0. */ 2893 static bool parse_elf_properties(int image_fd, 2894 struct image_info *info, 2895 const struct elf_phdr *phdr, 2896 char bprm_buf[BPRM_BUF_SIZE], 2897 Error **errp) 2898 { 2899 union { 2900 struct elf_note nhdr; 2901 uint32_t data[NOTE_DATA_SZ / sizeof(uint32_t)]; 2902 } note; 2903 2904 int n, off, datasz; 2905 bool have_prev_type; 2906 uint32_t prev_type; 2907 2908 /* Unless the arch requires properties, ignore them. */ 2909 if (!ARCH_USE_GNU_PROPERTY) { 2910 return true; 2911 } 2912 2913 /* If the properties are crazy large, that's too bad. */ 2914 n = phdr->p_filesz; 2915 if (n > sizeof(note)) { 2916 error_setg(errp, "PT_GNU_PROPERTY too large"); 2917 return false; 2918 } 2919 if (n < sizeof(note.nhdr)) { 2920 error_setg(errp, "PT_GNU_PROPERTY too small"); 2921 return false; 2922 } 2923 2924 if (phdr->p_offset + n <= BPRM_BUF_SIZE) { 2925 memcpy(¬e, bprm_buf + phdr->p_offset, n); 2926 } else { 2927 ssize_t len = pread(image_fd, ¬e, n, phdr->p_offset); 2928 if (len != n) { 2929 error_setg_errno(errp, errno, "Error reading file header"); 2930 return false; 2931 } 2932 } 2933 2934 /* 2935 * The contents of a valid PT_GNU_PROPERTY is a sequence 2936 * of uint32_t -- swap them all now. 2937 */ 2938 #ifdef BSWAP_NEEDED 2939 for (int i = 0; i < n / 4; i++) { 2940 bswap32s(note.data + i); 2941 } 2942 #endif 2943 2944 /* 2945 * Note that nhdr is 3 words, and that the "name" described by namesz 2946 * immediately follows nhdr and is thus at the 4th word. Further, all 2947 * of the inputs to the kernel's round_up are multiples of 4. 2948 */ 2949 if (note.nhdr.n_type != NT_GNU_PROPERTY_TYPE_0 || 2950 note.nhdr.n_namesz != NOTE_NAME_SZ || 2951 note.data[3] != GNU0_MAGIC) { 2952 error_setg(errp, "Invalid note in PT_GNU_PROPERTY"); 2953 return false; 2954 } 2955 off = sizeof(note.nhdr) + NOTE_NAME_SZ; 2956 2957 datasz = note.nhdr.n_descsz + off; 2958 if (datasz > n) { 2959 error_setg(errp, "Invalid note size in PT_GNU_PROPERTY"); 2960 return false; 2961 } 2962 2963 have_prev_type = false; 2964 prev_type = 0; 2965 while (1) { 2966 if (off == datasz) { 2967 return true; /* end, exit ok */ 2968 } 2969 if (!parse_elf_property(note.data, &off, datasz, info, 2970 have_prev_type, &prev_type, errp)) { 2971 return false; 2972 } 2973 have_prev_type = true; 2974 } 2975 } 2976 2977 /* Load an ELF image into the address space. 2978 2979 IMAGE_NAME is the filename of the image, to use in error messages. 2980 IMAGE_FD is the open file descriptor for the image. 2981 2982 BPRM_BUF is a copy of the beginning of the file; this of course 2983 contains the elf file header at offset 0. It is assumed that this 2984 buffer is sufficiently aligned to present no problems to the host 2985 in accessing data at aligned offsets within the buffer. 2986 2987 On return: INFO values will be filled in, as necessary or available. */ 2988 2989 static void load_elf_image(const char *image_name, int image_fd, 2990 struct image_info *info, char **pinterp_name, 2991 char bprm_buf[BPRM_BUF_SIZE]) 2992 { 2993 struct elfhdr *ehdr = (struct elfhdr *)bprm_buf; 2994 struct elf_phdr *phdr; 2995 abi_ulong load_addr, load_bias, loaddr, hiaddr, error; 2996 int i, retval, prot_exec; 2997 Error *err = NULL; 2998 2999 /* First of all, some simple consistency checks */ 3000 if (!elf_check_ident(ehdr)) { 3001 error_setg(&err, "Invalid ELF image for this architecture"); 3002 goto exit_errmsg; 3003 } 3004 bswap_ehdr(ehdr); 3005 if (!elf_check_ehdr(ehdr)) { 3006 error_setg(&err, "Invalid ELF image for this architecture"); 3007 goto exit_errmsg; 3008 } 3009 3010 i = ehdr->e_phnum * sizeof(struct elf_phdr); 3011 if (ehdr->e_phoff + i <= BPRM_BUF_SIZE) { 3012 phdr = (struct elf_phdr *)(bprm_buf + ehdr->e_phoff); 3013 } else { 3014 phdr = (struct elf_phdr *) alloca(i); 3015 retval = pread(image_fd, phdr, i, ehdr->e_phoff); 3016 if (retval != i) { 3017 goto exit_read; 3018 } 3019 } 3020 bswap_phdr(phdr, ehdr->e_phnum); 3021 3022 info->nsegs = 0; 3023 info->pt_dynamic_addr = 0; 3024 3025 mmap_lock(); 3026 3027 /* 3028 * Find the maximum size of the image and allocate an appropriate 3029 * amount of memory to handle that. Locate the interpreter, if any. 3030 */ 3031 loaddr = -1, hiaddr = 0; 3032 info->alignment = 0; 3033 info->exec_stack = EXSTACK_DEFAULT; 3034 for (i = 0; i < ehdr->e_phnum; ++i) { 3035 struct elf_phdr *eppnt = phdr + i; 3036 if (eppnt->p_type == PT_LOAD) { 3037 abi_ulong a = eppnt->p_vaddr - eppnt->p_offset; 3038 if (a < loaddr) { 3039 loaddr = a; 3040 } 3041 a = eppnt->p_vaddr + eppnt->p_memsz - 1; 3042 if (a > hiaddr) { 3043 hiaddr = a; 3044 } 3045 ++info->nsegs; 3046 info->alignment |= eppnt->p_align; 3047 } else if (eppnt->p_type == PT_INTERP && pinterp_name) { 3048 g_autofree char *interp_name = NULL; 3049 3050 if (*pinterp_name) { 3051 error_setg(&err, "Multiple PT_INTERP entries"); 3052 goto exit_errmsg; 3053 } 3054 3055 interp_name = g_malloc(eppnt->p_filesz); 3056 3057 if (eppnt->p_offset + eppnt->p_filesz <= BPRM_BUF_SIZE) { 3058 memcpy(interp_name, bprm_buf + eppnt->p_offset, 3059 eppnt->p_filesz); 3060 } else { 3061 retval = pread(image_fd, interp_name, eppnt->p_filesz, 3062 eppnt->p_offset); 3063 if (retval != eppnt->p_filesz) { 3064 goto exit_read; 3065 } 3066 } 3067 if (interp_name[eppnt->p_filesz - 1] != 0) { 3068 error_setg(&err, "Invalid PT_INTERP entry"); 3069 goto exit_errmsg; 3070 } 3071 *pinterp_name = g_steal_pointer(&interp_name); 3072 } else if (eppnt->p_type == PT_GNU_PROPERTY) { 3073 if (!parse_elf_properties(image_fd, info, eppnt, bprm_buf, &err)) { 3074 goto exit_errmsg; 3075 } 3076 } else if (eppnt->p_type == PT_GNU_STACK) { 3077 info->exec_stack = eppnt->p_flags & PF_X; 3078 } 3079 } 3080 3081 if (pinterp_name != NULL) { 3082 /* 3083 * This is the main executable. 3084 * 3085 * Reserve extra space for brk. 3086 * We hold on to this space while placing the interpreter 3087 * and the stack, lest they be placed immediately after 3088 * the data segment and block allocation from the brk. 3089 * 3090 * 16MB is chosen as "large enough" without being so large as 3091 * to allow the result to not fit with a 32-bit guest on a 3092 * 32-bit host. However some 64 bit guests (e.g. s390x) 3093 * attempt to place their heap further ahead and currently 3094 * nothing stops them smashing into QEMUs address space. 3095 */ 3096 #if TARGET_LONG_BITS == 64 3097 info->reserve_brk = 32 * MiB; 3098 #else 3099 info->reserve_brk = 16 * MiB; 3100 #endif 3101 hiaddr += info->reserve_brk; 3102 3103 if (ehdr->e_type == ET_EXEC) { 3104 /* 3105 * Make sure that the low address does not conflict with 3106 * MMAP_MIN_ADDR or the QEMU application itself. 3107 */ 3108 probe_guest_base(image_name, loaddr, hiaddr); 3109 } else { 3110 /* 3111 * The binary is dynamic, but we still need to 3112 * select guest_base. In this case we pass a size. 3113 */ 3114 probe_guest_base(image_name, 0, hiaddr - loaddr); 3115 } 3116 } 3117 3118 /* 3119 * Reserve address space for all of this. 3120 * 3121 * In the case of ET_EXEC, we supply MAP_FIXED so that we get 3122 * exactly the address range that is required. 3123 * 3124 * Otherwise this is ET_DYN, and we are searching for a location 3125 * that can hold the memory space required. If the image is 3126 * pre-linked, LOADDR will be non-zero, and the kernel should 3127 * honor that address if it happens to be free. 3128 * 3129 * In both cases, we will overwrite pages in this range with mappings 3130 * from the executable. 3131 */ 3132 load_addr = target_mmap(loaddr, (size_t)hiaddr - loaddr + 1, PROT_NONE, 3133 MAP_PRIVATE | MAP_ANON | MAP_NORESERVE | 3134 (ehdr->e_type == ET_EXEC ? MAP_FIXED : 0), 3135 -1, 0); 3136 if (load_addr == -1) { 3137 goto exit_mmap; 3138 } 3139 load_bias = load_addr - loaddr; 3140 3141 if (elf_is_fdpic(ehdr)) { 3142 struct elf32_fdpic_loadseg *loadsegs = info->loadsegs = 3143 g_malloc(sizeof(*loadsegs) * info->nsegs); 3144 3145 for (i = 0; i < ehdr->e_phnum; ++i) { 3146 switch (phdr[i].p_type) { 3147 case PT_DYNAMIC: 3148 info->pt_dynamic_addr = phdr[i].p_vaddr + load_bias; 3149 break; 3150 case PT_LOAD: 3151 loadsegs->addr = phdr[i].p_vaddr + load_bias; 3152 loadsegs->p_vaddr = phdr[i].p_vaddr; 3153 loadsegs->p_memsz = phdr[i].p_memsz; 3154 ++loadsegs; 3155 break; 3156 } 3157 } 3158 } 3159 3160 info->load_bias = load_bias; 3161 info->code_offset = load_bias; 3162 info->data_offset = load_bias; 3163 info->load_addr = load_addr; 3164 info->entry = ehdr->e_entry + load_bias; 3165 info->start_code = -1; 3166 info->end_code = 0; 3167 info->start_data = -1; 3168 info->end_data = 0; 3169 info->brk = 0; 3170 info->elf_flags = ehdr->e_flags; 3171 3172 prot_exec = PROT_EXEC; 3173 #ifdef TARGET_AARCH64 3174 /* 3175 * If the BTI feature is present, this indicates that the executable 3176 * pages of the startup binary should be mapped with PROT_BTI, so that 3177 * branch targets are enforced. 3178 * 3179 * The startup binary is either the interpreter or the static executable. 3180 * The interpreter is responsible for all pages of a dynamic executable. 3181 * 3182 * Elf notes are backward compatible to older cpus. 3183 * Do not enable BTI unless it is supported. 3184 */ 3185 if ((info->note_flags & GNU_PROPERTY_AARCH64_FEATURE_1_BTI) 3186 && (pinterp_name == NULL || *pinterp_name == 0) 3187 && cpu_isar_feature(aa64_bti, ARM_CPU(thread_cpu))) { 3188 prot_exec |= TARGET_PROT_BTI; 3189 } 3190 #endif 3191 3192 for (i = 0; i < ehdr->e_phnum; i++) { 3193 struct elf_phdr *eppnt = phdr + i; 3194 if (eppnt->p_type == PT_LOAD) { 3195 abi_ulong vaddr, vaddr_po, vaddr_ps, vaddr_ef, vaddr_em, vaddr_len; 3196 int elf_prot = 0; 3197 3198 if (eppnt->p_flags & PF_R) { 3199 elf_prot |= PROT_READ; 3200 } 3201 if (eppnt->p_flags & PF_W) { 3202 elf_prot |= PROT_WRITE; 3203 } 3204 if (eppnt->p_flags & PF_X) { 3205 elf_prot |= prot_exec; 3206 } 3207 3208 vaddr = load_bias + eppnt->p_vaddr; 3209 vaddr_po = TARGET_ELF_PAGEOFFSET(vaddr); 3210 vaddr_ps = TARGET_ELF_PAGESTART(vaddr); 3211 3212 vaddr_ef = vaddr + eppnt->p_filesz; 3213 vaddr_em = vaddr + eppnt->p_memsz; 3214 3215 /* 3216 * Some segments may be completely empty, with a non-zero p_memsz 3217 * but no backing file segment. 3218 */ 3219 if (eppnt->p_filesz != 0) { 3220 vaddr_len = TARGET_ELF_PAGELENGTH(eppnt->p_filesz + vaddr_po); 3221 error = target_mmap(vaddr_ps, vaddr_len, elf_prot, 3222 MAP_PRIVATE | MAP_FIXED, 3223 image_fd, eppnt->p_offset - vaddr_po); 3224 3225 if (error == -1) { 3226 goto exit_mmap; 3227 } 3228 3229 /* 3230 * If the load segment requests extra zeros (e.g. bss), map it. 3231 */ 3232 if (eppnt->p_filesz < eppnt->p_memsz) { 3233 zero_bss(vaddr_ef, vaddr_em, elf_prot); 3234 } 3235 } else if (eppnt->p_memsz != 0) { 3236 vaddr_len = TARGET_ELF_PAGELENGTH(eppnt->p_memsz + vaddr_po); 3237 error = target_mmap(vaddr_ps, vaddr_len, elf_prot, 3238 MAP_PRIVATE | MAP_FIXED | MAP_ANONYMOUS, 3239 -1, 0); 3240 3241 if (error == -1) { 3242 goto exit_mmap; 3243 } 3244 } 3245 3246 /* Find the full program boundaries. */ 3247 if (elf_prot & PROT_EXEC) { 3248 if (vaddr < info->start_code) { 3249 info->start_code = vaddr; 3250 } 3251 if (vaddr_ef > info->end_code) { 3252 info->end_code = vaddr_ef; 3253 } 3254 } 3255 if (elf_prot & PROT_WRITE) { 3256 if (vaddr < info->start_data) { 3257 info->start_data = vaddr; 3258 } 3259 if (vaddr_ef > info->end_data) { 3260 info->end_data = vaddr_ef; 3261 } 3262 } 3263 if (vaddr_em > info->brk) { 3264 info->brk = vaddr_em; 3265 } 3266 #ifdef TARGET_MIPS 3267 } else if (eppnt->p_type == PT_MIPS_ABIFLAGS) { 3268 Mips_elf_abiflags_v0 abiflags; 3269 if (eppnt->p_filesz < sizeof(Mips_elf_abiflags_v0)) { 3270 error_setg(&err, "Invalid PT_MIPS_ABIFLAGS entry"); 3271 goto exit_errmsg; 3272 } 3273 if (eppnt->p_offset + eppnt->p_filesz <= BPRM_BUF_SIZE) { 3274 memcpy(&abiflags, bprm_buf + eppnt->p_offset, 3275 sizeof(Mips_elf_abiflags_v0)); 3276 } else { 3277 retval = pread(image_fd, &abiflags, sizeof(Mips_elf_abiflags_v0), 3278 eppnt->p_offset); 3279 if (retval != sizeof(Mips_elf_abiflags_v0)) { 3280 goto exit_read; 3281 } 3282 } 3283 bswap_mips_abiflags(&abiflags); 3284 info->fp_abi = abiflags.fp_abi; 3285 #endif 3286 } 3287 } 3288 3289 if (info->end_data == 0) { 3290 info->start_data = info->end_code; 3291 info->end_data = info->end_code; 3292 } 3293 3294 if (qemu_log_enabled()) { 3295 load_symbols(ehdr, image_fd, load_bias); 3296 } 3297 3298 debuginfo_report_elf(image_name, image_fd, load_bias); 3299 3300 mmap_unlock(); 3301 3302 close(image_fd); 3303 return; 3304 3305 exit_read: 3306 if (retval >= 0) { 3307 error_setg(&err, "Incomplete read of file header"); 3308 } else { 3309 error_setg_errno(&err, errno, "Error reading file header"); 3310 } 3311 goto exit_errmsg; 3312 exit_mmap: 3313 error_setg_errno(&err, errno, "Error mapping file"); 3314 goto exit_errmsg; 3315 exit_errmsg: 3316 error_reportf_err(err, "%s: ", image_name); 3317 exit(-1); 3318 } 3319 3320 static void load_elf_interp(const char *filename, struct image_info *info, 3321 char bprm_buf[BPRM_BUF_SIZE]) 3322 { 3323 int fd, retval; 3324 Error *err = NULL; 3325 3326 fd = open(path(filename), O_RDONLY); 3327 if (fd < 0) { 3328 error_setg_file_open(&err, errno, filename); 3329 error_report_err(err); 3330 exit(-1); 3331 } 3332 3333 retval = read(fd, bprm_buf, BPRM_BUF_SIZE); 3334 if (retval < 0) { 3335 error_setg_errno(&err, errno, "Error reading file header"); 3336 error_reportf_err(err, "%s: ", filename); 3337 exit(-1); 3338 } 3339 3340 if (retval < BPRM_BUF_SIZE) { 3341 memset(bprm_buf + retval, 0, BPRM_BUF_SIZE - retval); 3342 } 3343 3344 load_elf_image(filename, fd, info, NULL, bprm_buf); 3345 } 3346 3347 static int symfind(const void *s0, const void *s1) 3348 { 3349 target_ulong addr = *(target_ulong *)s0; 3350 struct elf_sym *sym = (struct elf_sym *)s1; 3351 int result = 0; 3352 if (addr < sym->st_value) { 3353 result = -1; 3354 } else if (addr >= sym->st_value + sym->st_size) { 3355 result = 1; 3356 } 3357 return result; 3358 } 3359 3360 static const char *lookup_symbolxx(struct syminfo *s, target_ulong orig_addr) 3361 { 3362 #if ELF_CLASS == ELFCLASS32 3363 struct elf_sym *syms = s->disas_symtab.elf32; 3364 #else 3365 struct elf_sym *syms = s->disas_symtab.elf64; 3366 #endif 3367 3368 // binary search 3369 struct elf_sym *sym; 3370 3371 sym = bsearch(&orig_addr, syms, s->disas_num_syms, sizeof(*syms), symfind); 3372 if (sym != NULL) { 3373 return s->disas_strtab + sym->st_name; 3374 } 3375 3376 return ""; 3377 } 3378 3379 /* FIXME: This should use elf_ops.h */ 3380 static int symcmp(const void *s0, const void *s1) 3381 { 3382 struct elf_sym *sym0 = (struct elf_sym *)s0; 3383 struct elf_sym *sym1 = (struct elf_sym *)s1; 3384 return (sym0->st_value < sym1->st_value) 3385 ? -1 3386 : ((sym0->st_value > sym1->st_value) ? 1 : 0); 3387 } 3388 3389 /* Best attempt to load symbols from this ELF object. */ 3390 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias) 3391 { 3392 int i, shnum, nsyms, sym_idx = 0, str_idx = 0; 3393 uint64_t segsz; 3394 struct elf_shdr *shdr; 3395 char *strings = NULL; 3396 struct syminfo *s = NULL; 3397 struct elf_sym *new_syms, *syms = NULL; 3398 3399 shnum = hdr->e_shnum; 3400 i = shnum * sizeof(struct elf_shdr); 3401 shdr = (struct elf_shdr *)alloca(i); 3402 if (pread(fd, shdr, i, hdr->e_shoff) != i) { 3403 return; 3404 } 3405 3406 bswap_shdr(shdr, shnum); 3407 for (i = 0; i < shnum; ++i) { 3408 if (shdr[i].sh_type == SHT_SYMTAB) { 3409 sym_idx = i; 3410 str_idx = shdr[i].sh_link; 3411 goto found; 3412 } 3413 } 3414 3415 /* There will be no symbol table if the file was stripped. */ 3416 return; 3417 3418 found: 3419 /* Now know where the strtab and symtab are. Snarf them. */ 3420 s = g_try_new(struct syminfo, 1); 3421 if (!s) { 3422 goto give_up; 3423 } 3424 3425 segsz = shdr[str_idx].sh_size; 3426 s->disas_strtab = strings = g_try_malloc(segsz); 3427 if (!strings || 3428 pread(fd, strings, segsz, shdr[str_idx].sh_offset) != segsz) { 3429 goto give_up; 3430 } 3431 3432 segsz = shdr[sym_idx].sh_size; 3433 syms = g_try_malloc(segsz); 3434 if (!syms || pread(fd, syms, segsz, shdr[sym_idx].sh_offset) != segsz) { 3435 goto give_up; 3436 } 3437 3438 if (segsz / sizeof(struct elf_sym) > INT_MAX) { 3439 /* Implausibly large symbol table: give up rather than ploughing 3440 * on with the number of symbols calculation overflowing 3441 */ 3442 goto give_up; 3443 } 3444 nsyms = segsz / sizeof(struct elf_sym); 3445 for (i = 0; i < nsyms; ) { 3446 bswap_sym(syms + i); 3447 /* Throw away entries which we do not need. */ 3448 if (syms[i].st_shndx == SHN_UNDEF 3449 || syms[i].st_shndx >= SHN_LORESERVE 3450 || ELF_ST_TYPE(syms[i].st_info) != STT_FUNC) { 3451 if (i < --nsyms) { 3452 syms[i] = syms[nsyms]; 3453 } 3454 } else { 3455 #if defined(TARGET_ARM) || defined (TARGET_MIPS) 3456 /* The bottom address bit marks a Thumb or MIPS16 symbol. */ 3457 syms[i].st_value &= ~(target_ulong)1; 3458 #endif 3459 syms[i].st_value += load_bias; 3460 i++; 3461 } 3462 } 3463 3464 /* No "useful" symbol. */ 3465 if (nsyms == 0) { 3466 goto give_up; 3467 } 3468 3469 /* Attempt to free the storage associated with the local symbols 3470 that we threw away. Whether or not this has any effect on the 3471 memory allocation depends on the malloc implementation and how 3472 many symbols we managed to discard. */ 3473 new_syms = g_try_renew(struct elf_sym, syms, nsyms); 3474 if (new_syms == NULL) { 3475 goto give_up; 3476 } 3477 syms = new_syms; 3478 3479 qsort(syms, nsyms, sizeof(*syms), symcmp); 3480 3481 s->disas_num_syms = nsyms; 3482 #if ELF_CLASS == ELFCLASS32 3483 s->disas_symtab.elf32 = syms; 3484 #else 3485 s->disas_symtab.elf64 = syms; 3486 #endif 3487 s->lookup_symbol = lookup_symbolxx; 3488 s->next = syminfos; 3489 syminfos = s; 3490 3491 return; 3492 3493 give_up: 3494 g_free(s); 3495 g_free(strings); 3496 g_free(syms); 3497 } 3498 3499 uint32_t get_elf_eflags(int fd) 3500 { 3501 struct elfhdr ehdr; 3502 off_t offset; 3503 int ret; 3504 3505 /* Read ELF header */ 3506 offset = lseek(fd, 0, SEEK_SET); 3507 if (offset == (off_t) -1) { 3508 return 0; 3509 } 3510 ret = read(fd, &ehdr, sizeof(ehdr)); 3511 if (ret < sizeof(ehdr)) { 3512 return 0; 3513 } 3514 offset = lseek(fd, offset, SEEK_SET); 3515 if (offset == (off_t) -1) { 3516 return 0; 3517 } 3518 3519 /* Check ELF signature */ 3520 if (!elf_check_ident(&ehdr)) { 3521 return 0; 3522 } 3523 3524 /* check header */ 3525 bswap_ehdr(&ehdr); 3526 if (!elf_check_ehdr(&ehdr)) { 3527 return 0; 3528 } 3529 3530 /* return architecture id */ 3531 return ehdr.e_flags; 3532 } 3533 3534 int load_elf_binary(struct linux_binprm *bprm, struct image_info *info) 3535 { 3536 struct image_info interp_info; 3537 struct elfhdr elf_ex; 3538 char *elf_interpreter = NULL; 3539 char *scratch; 3540 3541 memset(&interp_info, 0, sizeof(interp_info)); 3542 #ifdef TARGET_MIPS 3543 interp_info.fp_abi = MIPS_ABI_FP_UNKNOWN; 3544 #endif 3545 3546 info->start_mmap = (abi_ulong)ELF_START_MMAP; 3547 3548 load_elf_image(bprm->filename, bprm->fd, info, 3549 &elf_interpreter, bprm->buf); 3550 3551 /* ??? We need a copy of the elf header for passing to create_elf_tables. 3552 If we do nothing, we'll have overwritten this when we re-use bprm->buf 3553 when we load the interpreter. */ 3554 elf_ex = *(struct elfhdr *)bprm->buf; 3555 3556 /* Do this so that we can load the interpreter, if need be. We will 3557 change some of these later */ 3558 bprm->p = setup_arg_pages(bprm, info); 3559 3560 scratch = g_new0(char, TARGET_PAGE_SIZE); 3561 if (STACK_GROWS_DOWN) { 3562 bprm->p = copy_elf_strings(1, &bprm->filename, scratch, 3563 bprm->p, info->stack_limit); 3564 info->file_string = bprm->p; 3565 bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch, 3566 bprm->p, info->stack_limit); 3567 info->env_strings = bprm->p; 3568 bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch, 3569 bprm->p, info->stack_limit); 3570 info->arg_strings = bprm->p; 3571 } else { 3572 info->arg_strings = bprm->p; 3573 bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch, 3574 bprm->p, info->stack_limit); 3575 info->env_strings = bprm->p; 3576 bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch, 3577 bprm->p, info->stack_limit); 3578 info->file_string = bprm->p; 3579 bprm->p = copy_elf_strings(1, &bprm->filename, scratch, 3580 bprm->p, info->stack_limit); 3581 } 3582 3583 g_free(scratch); 3584 3585 if (!bprm->p) { 3586 fprintf(stderr, "%s: %s\n", bprm->filename, strerror(E2BIG)); 3587 exit(-1); 3588 } 3589 3590 if (elf_interpreter) { 3591 load_elf_interp(elf_interpreter, &interp_info, bprm->buf); 3592 3593 /* If the program interpreter is one of these two, then assume 3594 an iBCS2 image. Otherwise assume a native linux image. */ 3595 3596 if (strcmp(elf_interpreter, "/usr/lib/libc.so.1") == 0 3597 || strcmp(elf_interpreter, "/usr/lib/ld.so.1") == 0) { 3598 info->personality = PER_SVR4; 3599 3600 /* Why this, you ask??? Well SVr4 maps page 0 as read-only, 3601 and some applications "depend" upon this behavior. Since 3602 we do not have the power to recompile these, we emulate 3603 the SVr4 behavior. Sigh. */ 3604 target_mmap(0, qemu_host_page_size, PROT_READ | PROT_EXEC, 3605 MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); 3606 } 3607 #ifdef TARGET_MIPS 3608 info->interp_fp_abi = interp_info.fp_abi; 3609 #endif 3610 } 3611 3612 /* 3613 * TODO: load a vdso, which would also contain the signal trampolines. 3614 * Otherwise, allocate a private page to hold them. 3615 */ 3616 if (TARGET_ARCH_HAS_SIGTRAMP_PAGE) { 3617 abi_long tramp_page = target_mmap(0, TARGET_PAGE_SIZE, 3618 PROT_READ | PROT_WRITE, 3619 MAP_PRIVATE | MAP_ANON, -1, 0); 3620 if (tramp_page == -1) { 3621 return -errno; 3622 } 3623 3624 setup_sigtramp(tramp_page); 3625 target_mprotect(tramp_page, TARGET_PAGE_SIZE, PROT_READ | PROT_EXEC); 3626 } 3627 3628 bprm->p = create_elf_tables(bprm->p, bprm->argc, bprm->envc, &elf_ex, 3629 info, (elf_interpreter ? &interp_info : NULL)); 3630 info->start_stack = bprm->p; 3631 3632 /* If we have an interpreter, set that as the program's entry point. 3633 Copy the load_bias as well, to help PPC64 interpret the entry 3634 point as a function descriptor. Do this after creating elf tables 3635 so that we copy the original program entry point into the AUXV. */ 3636 if (elf_interpreter) { 3637 info->load_bias = interp_info.load_bias; 3638 info->entry = interp_info.entry; 3639 g_free(elf_interpreter); 3640 } 3641 3642 #ifdef USE_ELF_CORE_DUMP 3643 bprm->core_dump = &elf_core_dump; 3644 #endif 3645 3646 /* 3647 * If we reserved extra space for brk, release it now. 3648 * The implementation of do_brk in syscalls.c expects to be able 3649 * to mmap pages in this space. 3650 */ 3651 if (info->reserve_brk) { 3652 abi_ulong start_brk = HOST_PAGE_ALIGN(info->brk); 3653 abi_ulong end_brk = HOST_PAGE_ALIGN(info->brk + info->reserve_brk); 3654 target_munmap(start_brk, end_brk - start_brk); 3655 } 3656 3657 return 0; 3658 } 3659 3660 #ifdef USE_ELF_CORE_DUMP 3661 /* 3662 * Definitions to generate Intel SVR4-like core files. 3663 * These mostly have the same names as the SVR4 types with "target_elf_" 3664 * tacked on the front to prevent clashes with linux definitions, 3665 * and the typedef forms have been avoided. This is mostly like 3666 * the SVR4 structure, but more Linuxy, with things that Linux does 3667 * not support and which gdb doesn't really use excluded. 3668 * 3669 * Fields we don't dump (their contents is zero) in linux-user qemu 3670 * are marked with XXX. 3671 * 3672 * Core dump code is copied from linux kernel (fs/binfmt_elf.c). 3673 * 3674 * Porting ELF coredump for target is (quite) simple process. First you 3675 * define USE_ELF_CORE_DUMP in target ELF code (where init_thread() for 3676 * the target resides): 3677 * 3678 * #define USE_ELF_CORE_DUMP 3679 * 3680 * Next you define type of register set used for dumping. ELF specification 3681 * says that it needs to be array of elf_greg_t that has size of ELF_NREG. 3682 * 3683 * typedef <target_regtype> target_elf_greg_t; 3684 * #define ELF_NREG <number of registers> 3685 * typedef taret_elf_greg_t target_elf_gregset_t[ELF_NREG]; 3686 * 3687 * Last step is to implement target specific function that copies registers 3688 * from given cpu into just specified register set. Prototype is: 3689 * 3690 * static void elf_core_copy_regs(taret_elf_gregset_t *regs, 3691 * const CPUArchState *env); 3692 * 3693 * Parameters: 3694 * regs - copy register values into here (allocated and zeroed by caller) 3695 * env - copy registers from here 3696 * 3697 * Example for ARM target is provided in this file. 3698 */ 3699 3700 /* An ELF note in memory */ 3701 struct memelfnote { 3702 const char *name; 3703 size_t namesz; 3704 size_t namesz_rounded; 3705 int type; 3706 size_t datasz; 3707 size_t datasz_rounded; 3708 void *data; 3709 size_t notesz; 3710 }; 3711 3712 struct target_elf_siginfo { 3713 abi_int si_signo; /* signal number */ 3714 abi_int si_code; /* extra code */ 3715 abi_int si_errno; /* errno */ 3716 }; 3717 3718 struct target_elf_prstatus { 3719 struct target_elf_siginfo pr_info; /* Info associated with signal */ 3720 abi_short pr_cursig; /* Current signal */ 3721 abi_ulong pr_sigpend; /* XXX */ 3722 abi_ulong pr_sighold; /* XXX */ 3723 target_pid_t pr_pid; 3724 target_pid_t pr_ppid; 3725 target_pid_t pr_pgrp; 3726 target_pid_t pr_sid; 3727 struct target_timeval pr_utime; /* XXX User time */ 3728 struct target_timeval pr_stime; /* XXX System time */ 3729 struct target_timeval pr_cutime; /* XXX Cumulative user time */ 3730 struct target_timeval pr_cstime; /* XXX Cumulative system time */ 3731 target_elf_gregset_t pr_reg; /* GP registers */ 3732 abi_int pr_fpvalid; /* XXX */ 3733 }; 3734 3735 #define ELF_PRARGSZ (80) /* Number of chars for args */ 3736 3737 struct target_elf_prpsinfo { 3738 char pr_state; /* numeric process state */ 3739 char pr_sname; /* char for pr_state */ 3740 char pr_zomb; /* zombie */ 3741 char pr_nice; /* nice val */ 3742 abi_ulong pr_flag; /* flags */ 3743 target_uid_t pr_uid; 3744 target_gid_t pr_gid; 3745 target_pid_t pr_pid, pr_ppid, pr_pgrp, pr_sid; 3746 /* Lots missing */ 3747 char pr_fname[16] QEMU_NONSTRING; /* filename of executable */ 3748 char pr_psargs[ELF_PRARGSZ]; /* initial part of arg list */ 3749 }; 3750 3751 /* Here is the structure in which status of each thread is captured. */ 3752 struct elf_thread_status { 3753 QTAILQ_ENTRY(elf_thread_status) ets_link; 3754 struct target_elf_prstatus prstatus; /* NT_PRSTATUS */ 3755 #if 0 3756 elf_fpregset_t fpu; /* NT_PRFPREG */ 3757 struct task_struct *thread; 3758 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */ 3759 #endif 3760 struct memelfnote notes[1]; 3761 int num_notes; 3762 }; 3763 3764 struct elf_note_info { 3765 struct memelfnote *notes; 3766 struct target_elf_prstatus *prstatus; /* NT_PRSTATUS */ 3767 struct target_elf_prpsinfo *psinfo; /* NT_PRPSINFO */ 3768 3769 QTAILQ_HEAD(, elf_thread_status) thread_list; 3770 #if 0 3771 /* 3772 * Current version of ELF coredump doesn't support 3773 * dumping fp regs etc. 3774 */ 3775 elf_fpregset_t *fpu; 3776 elf_fpxregset_t *xfpu; 3777 int thread_status_size; 3778 #endif 3779 int notes_size; 3780 int numnote; 3781 }; 3782 3783 struct vm_area_struct { 3784 target_ulong vma_start; /* start vaddr of memory region */ 3785 target_ulong vma_end; /* end vaddr of memory region */ 3786 abi_ulong vma_flags; /* protection etc. flags for the region */ 3787 QTAILQ_ENTRY(vm_area_struct) vma_link; 3788 }; 3789 3790 struct mm_struct { 3791 QTAILQ_HEAD(, vm_area_struct) mm_mmap; 3792 int mm_count; /* number of mappings */ 3793 }; 3794 3795 static struct mm_struct *vma_init(void); 3796 static void vma_delete(struct mm_struct *); 3797 static int vma_add_mapping(struct mm_struct *, target_ulong, 3798 target_ulong, abi_ulong); 3799 static int vma_get_mapping_count(const struct mm_struct *); 3800 static struct vm_area_struct *vma_first(const struct mm_struct *); 3801 static struct vm_area_struct *vma_next(struct vm_area_struct *); 3802 static abi_ulong vma_dump_size(const struct vm_area_struct *); 3803 static int vma_walker(void *priv, target_ulong start, target_ulong end, 3804 unsigned long flags); 3805 3806 static void fill_elf_header(struct elfhdr *, int, uint16_t, uint32_t); 3807 static void fill_note(struct memelfnote *, const char *, int, 3808 unsigned int, void *); 3809 static void fill_prstatus(struct target_elf_prstatus *, const TaskState *, int); 3810 static int fill_psinfo(struct target_elf_prpsinfo *, const TaskState *); 3811 static void fill_auxv_note(struct memelfnote *, const TaskState *); 3812 static void fill_elf_note_phdr(struct elf_phdr *, int, off_t); 3813 static size_t note_size(const struct memelfnote *); 3814 static void free_note_info(struct elf_note_info *); 3815 static int fill_note_info(struct elf_note_info *, long, const CPUArchState *); 3816 static void fill_thread_info(struct elf_note_info *, const CPUArchState *); 3817 3818 static int dump_write(int, const void *, size_t); 3819 static int write_note(struct memelfnote *, int); 3820 static int write_note_info(struct elf_note_info *, int); 3821 3822 #ifdef BSWAP_NEEDED 3823 static void bswap_prstatus(struct target_elf_prstatus *prstatus) 3824 { 3825 prstatus->pr_info.si_signo = tswap32(prstatus->pr_info.si_signo); 3826 prstatus->pr_info.si_code = tswap32(prstatus->pr_info.si_code); 3827 prstatus->pr_info.si_errno = tswap32(prstatus->pr_info.si_errno); 3828 prstatus->pr_cursig = tswap16(prstatus->pr_cursig); 3829 prstatus->pr_sigpend = tswapal(prstatus->pr_sigpend); 3830 prstatus->pr_sighold = tswapal(prstatus->pr_sighold); 3831 prstatus->pr_pid = tswap32(prstatus->pr_pid); 3832 prstatus->pr_ppid = tswap32(prstatus->pr_ppid); 3833 prstatus->pr_pgrp = tswap32(prstatus->pr_pgrp); 3834 prstatus->pr_sid = tswap32(prstatus->pr_sid); 3835 /* cpu times are not filled, so we skip them */ 3836 /* regs should be in correct format already */ 3837 prstatus->pr_fpvalid = tswap32(prstatus->pr_fpvalid); 3838 } 3839 3840 static void bswap_psinfo(struct target_elf_prpsinfo *psinfo) 3841 { 3842 psinfo->pr_flag = tswapal(psinfo->pr_flag); 3843 psinfo->pr_uid = tswap16(psinfo->pr_uid); 3844 psinfo->pr_gid = tswap16(psinfo->pr_gid); 3845 psinfo->pr_pid = tswap32(psinfo->pr_pid); 3846 psinfo->pr_ppid = tswap32(psinfo->pr_ppid); 3847 psinfo->pr_pgrp = tswap32(psinfo->pr_pgrp); 3848 psinfo->pr_sid = tswap32(psinfo->pr_sid); 3849 } 3850 3851 static void bswap_note(struct elf_note *en) 3852 { 3853 bswap32s(&en->n_namesz); 3854 bswap32s(&en->n_descsz); 3855 bswap32s(&en->n_type); 3856 } 3857 #else 3858 static inline void bswap_prstatus(struct target_elf_prstatus *p) { } 3859 static inline void bswap_psinfo(struct target_elf_prpsinfo *p) {} 3860 static inline void bswap_note(struct elf_note *en) { } 3861 #endif /* BSWAP_NEEDED */ 3862 3863 /* 3864 * Minimal support for linux memory regions. These are needed 3865 * when we are finding out what memory exactly belongs to 3866 * emulated process. No locks needed here, as long as 3867 * thread that received the signal is stopped. 3868 */ 3869 3870 static struct mm_struct *vma_init(void) 3871 { 3872 struct mm_struct *mm; 3873 3874 if ((mm = g_malloc(sizeof (*mm))) == NULL) 3875 return (NULL); 3876 3877 mm->mm_count = 0; 3878 QTAILQ_INIT(&mm->mm_mmap); 3879 3880 return (mm); 3881 } 3882 3883 static void vma_delete(struct mm_struct *mm) 3884 { 3885 struct vm_area_struct *vma; 3886 3887 while ((vma = vma_first(mm)) != NULL) { 3888 QTAILQ_REMOVE(&mm->mm_mmap, vma, vma_link); 3889 g_free(vma); 3890 } 3891 g_free(mm); 3892 } 3893 3894 static int vma_add_mapping(struct mm_struct *mm, target_ulong start, 3895 target_ulong end, abi_ulong flags) 3896 { 3897 struct vm_area_struct *vma; 3898 3899 if ((vma = g_malloc0(sizeof (*vma))) == NULL) 3900 return (-1); 3901 3902 vma->vma_start = start; 3903 vma->vma_end = end; 3904 vma->vma_flags = flags; 3905 3906 QTAILQ_INSERT_TAIL(&mm->mm_mmap, vma, vma_link); 3907 mm->mm_count++; 3908 3909 return (0); 3910 } 3911 3912 static struct vm_area_struct *vma_first(const struct mm_struct *mm) 3913 { 3914 return (QTAILQ_FIRST(&mm->mm_mmap)); 3915 } 3916 3917 static struct vm_area_struct *vma_next(struct vm_area_struct *vma) 3918 { 3919 return (QTAILQ_NEXT(vma, vma_link)); 3920 } 3921 3922 static int vma_get_mapping_count(const struct mm_struct *mm) 3923 { 3924 return (mm->mm_count); 3925 } 3926 3927 /* 3928 * Calculate file (dump) size of given memory region. 3929 */ 3930 static abi_ulong vma_dump_size(const struct vm_area_struct *vma) 3931 { 3932 /* if we cannot even read the first page, skip it */ 3933 if (!access_ok_untagged(VERIFY_READ, vma->vma_start, TARGET_PAGE_SIZE)) 3934 return (0); 3935 3936 /* 3937 * Usually we don't dump executable pages as they contain 3938 * non-writable code that debugger can read directly from 3939 * target library etc. However, thread stacks are marked 3940 * also executable so we read in first page of given region 3941 * and check whether it contains elf header. If there is 3942 * no elf header, we dump it. 3943 */ 3944 if (vma->vma_flags & PROT_EXEC) { 3945 char page[TARGET_PAGE_SIZE]; 3946 3947 if (copy_from_user(page, vma->vma_start, sizeof (page))) { 3948 return 0; 3949 } 3950 if ((page[EI_MAG0] == ELFMAG0) && 3951 (page[EI_MAG1] == ELFMAG1) && 3952 (page[EI_MAG2] == ELFMAG2) && 3953 (page[EI_MAG3] == ELFMAG3)) { 3954 /* 3955 * Mappings are possibly from ELF binary. Don't dump 3956 * them. 3957 */ 3958 return (0); 3959 } 3960 } 3961 3962 return (vma->vma_end - vma->vma_start); 3963 } 3964 3965 static int vma_walker(void *priv, target_ulong start, target_ulong end, 3966 unsigned long flags) 3967 { 3968 struct mm_struct *mm = (struct mm_struct *)priv; 3969 3970 vma_add_mapping(mm, start, end, flags); 3971 return (0); 3972 } 3973 3974 static void fill_note(struct memelfnote *note, const char *name, int type, 3975 unsigned int sz, void *data) 3976 { 3977 unsigned int namesz; 3978 3979 namesz = strlen(name) + 1; 3980 note->name = name; 3981 note->namesz = namesz; 3982 note->namesz_rounded = roundup(namesz, sizeof (int32_t)); 3983 note->type = type; 3984 note->datasz = sz; 3985 note->datasz_rounded = roundup(sz, sizeof (int32_t)); 3986 3987 note->data = data; 3988 3989 /* 3990 * We calculate rounded up note size here as specified by 3991 * ELF document. 3992 */ 3993 note->notesz = sizeof (struct elf_note) + 3994 note->namesz_rounded + note->datasz_rounded; 3995 } 3996 3997 static void fill_elf_header(struct elfhdr *elf, int segs, uint16_t machine, 3998 uint32_t flags) 3999 { 4000 (void) memset(elf, 0, sizeof(*elf)); 4001 4002 (void) memcpy(elf->e_ident, ELFMAG, SELFMAG); 4003 elf->e_ident[EI_CLASS] = ELF_CLASS; 4004 elf->e_ident[EI_DATA] = ELF_DATA; 4005 elf->e_ident[EI_VERSION] = EV_CURRENT; 4006 elf->e_ident[EI_OSABI] = ELF_OSABI; 4007 4008 elf->e_type = ET_CORE; 4009 elf->e_machine = machine; 4010 elf->e_version = EV_CURRENT; 4011 elf->e_phoff = sizeof(struct elfhdr); 4012 elf->e_flags = flags; 4013 elf->e_ehsize = sizeof(struct elfhdr); 4014 elf->e_phentsize = sizeof(struct elf_phdr); 4015 elf->e_phnum = segs; 4016 4017 bswap_ehdr(elf); 4018 } 4019 4020 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, off_t offset) 4021 { 4022 phdr->p_type = PT_NOTE; 4023 phdr->p_offset = offset; 4024 phdr->p_vaddr = 0; 4025 phdr->p_paddr = 0; 4026 phdr->p_filesz = sz; 4027 phdr->p_memsz = 0; 4028 phdr->p_flags = 0; 4029 phdr->p_align = 0; 4030 4031 bswap_phdr(phdr, 1); 4032 } 4033 4034 static size_t note_size(const struct memelfnote *note) 4035 { 4036 return (note->notesz); 4037 } 4038 4039 static void fill_prstatus(struct target_elf_prstatus *prstatus, 4040 const TaskState *ts, int signr) 4041 { 4042 (void) memset(prstatus, 0, sizeof (*prstatus)); 4043 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr; 4044 prstatus->pr_pid = ts->ts_tid; 4045 prstatus->pr_ppid = getppid(); 4046 prstatus->pr_pgrp = getpgrp(); 4047 prstatus->pr_sid = getsid(0); 4048 4049 bswap_prstatus(prstatus); 4050 } 4051 4052 static int fill_psinfo(struct target_elf_prpsinfo *psinfo, const TaskState *ts) 4053 { 4054 char *base_filename; 4055 unsigned int i, len; 4056 4057 (void) memset(psinfo, 0, sizeof (*psinfo)); 4058 4059 len = ts->info->env_strings - ts->info->arg_strings; 4060 if (len >= ELF_PRARGSZ) 4061 len = ELF_PRARGSZ - 1; 4062 if (copy_from_user(&psinfo->pr_psargs, ts->info->arg_strings, len)) { 4063 return -EFAULT; 4064 } 4065 for (i = 0; i < len; i++) 4066 if (psinfo->pr_psargs[i] == 0) 4067 psinfo->pr_psargs[i] = ' '; 4068 psinfo->pr_psargs[len] = 0; 4069 4070 psinfo->pr_pid = getpid(); 4071 psinfo->pr_ppid = getppid(); 4072 psinfo->pr_pgrp = getpgrp(); 4073 psinfo->pr_sid = getsid(0); 4074 psinfo->pr_uid = getuid(); 4075 psinfo->pr_gid = getgid(); 4076 4077 base_filename = g_path_get_basename(ts->bprm->filename); 4078 /* 4079 * Using strncpy here is fine: at max-length, 4080 * this field is not NUL-terminated. 4081 */ 4082 (void) strncpy(psinfo->pr_fname, base_filename, 4083 sizeof(psinfo->pr_fname)); 4084 4085 g_free(base_filename); 4086 bswap_psinfo(psinfo); 4087 return (0); 4088 } 4089 4090 static void fill_auxv_note(struct memelfnote *note, const TaskState *ts) 4091 { 4092 elf_addr_t auxv = (elf_addr_t)ts->info->saved_auxv; 4093 elf_addr_t orig_auxv = auxv; 4094 void *ptr; 4095 int len = ts->info->auxv_len; 4096 4097 /* 4098 * Auxiliary vector is stored in target process stack. It contains 4099 * {type, value} pairs that we need to dump into note. This is not 4100 * strictly necessary but we do it here for sake of completeness. 4101 */ 4102 4103 /* read in whole auxv vector and copy it to memelfnote */ 4104 ptr = lock_user(VERIFY_READ, orig_auxv, len, 0); 4105 if (ptr != NULL) { 4106 fill_note(note, "CORE", NT_AUXV, len, ptr); 4107 unlock_user(ptr, auxv, len); 4108 } 4109 } 4110 4111 /* 4112 * Constructs name of coredump file. We have following convention 4113 * for the name: 4114 * qemu_<basename-of-target-binary>_<date>-<time>_<pid>.core 4115 * 4116 * Returns the filename 4117 */ 4118 static char *core_dump_filename(const TaskState *ts) 4119 { 4120 g_autoptr(GDateTime) now = g_date_time_new_now_local(); 4121 g_autofree char *nowstr = g_date_time_format(now, "%Y%m%d-%H%M%S"); 4122 g_autofree char *base_filename = g_path_get_basename(ts->bprm->filename); 4123 4124 return g_strdup_printf("qemu_%s_%s_%d.core", 4125 base_filename, nowstr, (int)getpid()); 4126 } 4127 4128 static int dump_write(int fd, const void *ptr, size_t size) 4129 { 4130 const char *bufp = (const char *)ptr; 4131 ssize_t bytes_written, bytes_left; 4132 struct rlimit dumpsize; 4133 off_t pos; 4134 4135 bytes_written = 0; 4136 getrlimit(RLIMIT_CORE, &dumpsize); 4137 if ((pos = lseek(fd, 0, SEEK_CUR))==-1) { 4138 if (errno == ESPIPE) { /* not a seekable stream */ 4139 bytes_left = size; 4140 } else { 4141 return pos; 4142 } 4143 } else { 4144 if (dumpsize.rlim_cur <= pos) { 4145 return -1; 4146 } else if (dumpsize.rlim_cur == RLIM_INFINITY) { 4147 bytes_left = size; 4148 } else { 4149 size_t limit_left=dumpsize.rlim_cur - pos; 4150 bytes_left = limit_left >= size ? size : limit_left ; 4151 } 4152 } 4153 4154 /* 4155 * In normal conditions, single write(2) should do but 4156 * in case of socket etc. this mechanism is more portable. 4157 */ 4158 do { 4159 bytes_written = write(fd, bufp, bytes_left); 4160 if (bytes_written < 0) { 4161 if (errno == EINTR) 4162 continue; 4163 return (-1); 4164 } else if (bytes_written == 0) { /* eof */ 4165 return (-1); 4166 } 4167 bufp += bytes_written; 4168 bytes_left -= bytes_written; 4169 } while (bytes_left > 0); 4170 4171 return (0); 4172 } 4173 4174 static int write_note(struct memelfnote *men, int fd) 4175 { 4176 struct elf_note en; 4177 4178 en.n_namesz = men->namesz; 4179 en.n_type = men->type; 4180 en.n_descsz = men->datasz; 4181 4182 bswap_note(&en); 4183 4184 if (dump_write(fd, &en, sizeof(en)) != 0) 4185 return (-1); 4186 if (dump_write(fd, men->name, men->namesz_rounded) != 0) 4187 return (-1); 4188 if (dump_write(fd, men->data, men->datasz_rounded) != 0) 4189 return (-1); 4190 4191 return (0); 4192 } 4193 4194 static void fill_thread_info(struct elf_note_info *info, const CPUArchState *env) 4195 { 4196 CPUState *cpu = env_cpu((CPUArchState *)env); 4197 TaskState *ts = (TaskState *)cpu->opaque; 4198 struct elf_thread_status *ets; 4199 4200 ets = g_malloc0(sizeof (*ets)); 4201 ets->num_notes = 1; /* only prstatus is dumped */ 4202 fill_prstatus(&ets->prstatus, ts, 0); 4203 elf_core_copy_regs(&ets->prstatus.pr_reg, env); 4204 fill_note(&ets->notes[0], "CORE", NT_PRSTATUS, sizeof (ets->prstatus), 4205 &ets->prstatus); 4206 4207 QTAILQ_INSERT_TAIL(&info->thread_list, ets, ets_link); 4208 4209 info->notes_size += note_size(&ets->notes[0]); 4210 } 4211 4212 static void init_note_info(struct elf_note_info *info) 4213 { 4214 /* Initialize the elf_note_info structure so that it is at 4215 * least safe to call free_note_info() on it. Must be 4216 * called before calling fill_note_info(). 4217 */ 4218 memset(info, 0, sizeof (*info)); 4219 QTAILQ_INIT(&info->thread_list); 4220 } 4221 4222 static int fill_note_info(struct elf_note_info *info, 4223 long signr, const CPUArchState *env) 4224 { 4225 #define NUMNOTES 3 4226 CPUState *cpu = env_cpu((CPUArchState *)env); 4227 TaskState *ts = (TaskState *)cpu->opaque; 4228 int i; 4229 4230 info->notes = g_new0(struct memelfnote, NUMNOTES); 4231 if (info->notes == NULL) 4232 return (-ENOMEM); 4233 info->prstatus = g_malloc0(sizeof (*info->prstatus)); 4234 if (info->prstatus == NULL) 4235 return (-ENOMEM); 4236 info->psinfo = g_malloc0(sizeof (*info->psinfo)); 4237 if (info->prstatus == NULL) 4238 return (-ENOMEM); 4239 4240 /* 4241 * First fill in status (and registers) of current thread 4242 * including process info & aux vector. 4243 */ 4244 fill_prstatus(info->prstatus, ts, signr); 4245 elf_core_copy_regs(&info->prstatus->pr_reg, env); 4246 fill_note(&info->notes[0], "CORE", NT_PRSTATUS, 4247 sizeof (*info->prstatus), info->prstatus); 4248 fill_psinfo(info->psinfo, ts); 4249 fill_note(&info->notes[1], "CORE", NT_PRPSINFO, 4250 sizeof (*info->psinfo), info->psinfo); 4251 fill_auxv_note(&info->notes[2], ts); 4252 info->numnote = 3; 4253 4254 info->notes_size = 0; 4255 for (i = 0; i < info->numnote; i++) 4256 info->notes_size += note_size(&info->notes[i]); 4257 4258 /* read and fill status of all threads */ 4259 cpu_list_lock(); 4260 CPU_FOREACH(cpu) { 4261 if (cpu == thread_cpu) { 4262 continue; 4263 } 4264 fill_thread_info(info, cpu->env_ptr); 4265 } 4266 cpu_list_unlock(); 4267 4268 return (0); 4269 } 4270 4271 static void free_note_info(struct elf_note_info *info) 4272 { 4273 struct elf_thread_status *ets; 4274 4275 while (!QTAILQ_EMPTY(&info->thread_list)) { 4276 ets = QTAILQ_FIRST(&info->thread_list); 4277 QTAILQ_REMOVE(&info->thread_list, ets, ets_link); 4278 g_free(ets); 4279 } 4280 4281 g_free(info->prstatus); 4282 g_free(info->psinfo); 4283 g_free(info->notes); 4284 } 4285 4286 static int write_note_info(struct elf_note_info *info, int fd) 4287 { 4288 struct elf_thread_status *ets; 4289 int i, error = 0; 4290 4291 /* write prstatus, psinfo and auxv for current thread */ 4292 for (i = 0; i < info->numnote; i++) 4293 if ((error = write_note(&info->notes[i], fd)) != 0) 4294 return (error); 4295 4296 /* write prstatus for each thread */ 4297 QTAILQ_FOREACH(ets, &info->thread_list, ets_link) { 4298 if ((error = write_note(&ets->notes[0], fd)) != 0) 4299 return (error); 4300 } 4301 4302 return (0); 4303 } 4304 4305 /* 4306 * Write out ELF coredump. 4307 * 4308 * See documentation of ELF object file format in: 4309 * http://www.caldera.com/developers/devspecs/gabi41.pdf 4310 * 4311 * Coredump format in linux is following: 4312 * 4313 * 0 +----------------------+ \ 4314 * | ELF header | ET_CORE | 4315 * +----------------------+ | 4316 * | ELF program headers | |--- headers 4317 * | - NOTE section | | 4318 * | - PT_LOAD sections | | 4319 * +----------------------+ / 4320 * | NOTEs: | 4321 * | - NT_PRSTATUS | 4322 * | - NT_PRSINFO | 4323 * | - NT_AUXV | 4324 * +----------------------+ <-- aligned to target page 4325 * | Process memory dump | 4326 * : : 4327 * . . 4328 * : : 4329 * | | 4330 * +----------------------+ 4331 * 4332 * NT_PRSTATUS -> struct elf_prstatus (per thread) 4333 * NT_PRSINFO -> struct elf_prpsinfo 4334 * NT_AUXV is array of { type, value } pairs (see fill_auxv_note()). 4335 * 4336 * Format follows System V format as close as possible. Current 4337 * version limitations are as follows: 4338 * - no floating point registers are dumped 4339 * 4340 * Function returns 0 in case of success, negative errno otherwise. 4341 * 4342 * TODO: make this work also during runtime: it should be 4343 * possible to force coredump from running process and then 4344 * continue processing. For example qemu could set up SIGUSR2 4345 * handler (provided that target process haven't registered 4346 * handler for that) that does the dump when signal is received. 4347 */ 4348 static int elf_core_dump(int signr, const CPUArchState *env) 4349 { 4350 const CPUState *cpu = env_cpu((CPUArchState *)env); 4351 const TaskState *ts = (const TaskState *)cpu->opaque; 4352 struct vm_area_struct *vma = NULL; 4353 g_autofree char *corefile = NULL; 4354 struct elf_note_info info; 4355 struct elfhdr elf; 4356 struct elf_phdr phdr; 4357 struct rlimit dumpsize; 4358 struct mm_struct *mm = NULL; 4359 off_t offset = 0, data_offset = 0; 4360 int segs = 0; 4361 int fd = -1; 4362 4363 init_note_info(&info); 4364 4365 errno = 0; 4366 getrlimit(RLIMIT_CORE, &dumpsize); 4367 if (dumpsize.rlim_cur == 0) 4368 return 0; 4369 4370 corefile = core_dump_filename(ts); 4371 4372 if ((fd = open(corefile, O_WRONLY | O_CREAT, 4373 S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH)) < 0) 4374 return (-errno); 4375 4376 /* 4377 * Walk through target process memory mappings and 4378 * set up structure containing this information. After 4379 * this point vma_xxx functions can be used. 4380 */ 4381 if ((mm = vma_init()) == NULL) 4382 goto out; 4383 4384 walk_memory_regions(mm, vma_walker); 4385 segs = vma_get_mapping_count(mm); 4386 4387 /* 4388 * Construct valid coredump ELF header. We also 4389 * add one more segment for notes. 4390 */ 4391 fill_elf_header(&elf, segs + 1, ELF_MACHINE, 0); 4392 if (dump_write(fd, &elf, sizeof (elf)) != 0) 4393 goto out; 4394 4395 /* fill in the in-memory version of notes */ 4396 if (fill_note_info(&info, signr, env) < 0) 4397 goto out; 4398 4399 offset += sizeof (elf); /* elf header */ 4400 offset += (segs + 1) * sizeof (struct elf_phdr); /* program headers */ 4401 4402 /* write out notes program header */ 4403 fill_elf_note_phdr(&phdr, info.notes_size, offset); 4404 4405 offset += info.notes_size; 4406 if (dump_write(fd, &phdr, sizeof (phdr)) != 0) 4407 goto out; 4408 4409 /* 4410 * ELF specification wants data to start at page boundary so 4411 * we align it here. 4412 */ 4413 data_offset = offset = roundup(offset, ELF_EXEC_PAGESIZE); 4414 4415 /* 4416 * Write program headers for memory regions mapped in 4417 * the target process. 4418 */ 4419 for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) { 4420 (void) memset(&phdr, 0, sizeof (phdr)); 4421 4422 phdr.p_type = PT_LOAD; 4423 phdr.p_offset = offset; 4424 phdr.p_vaddr = vma->vma_start; 4425 phdr.p_paddr = 0; 4426 phdr.p_filesz = vma_dump_size(vma); 4427 offset += phdr.p_filesz; 4428 phdr.p_memsz = vma->vma_end - vma->vma_start; 4429 phdr.p_flags = vma->vma_flags & PROT_READ ? PF_R : 0; 4430 if (vma->vma_flags & PROT_WRITE) 4431 phdr.p_flags |= PF_W; 4432 if (vma->vma_flags & PROT_EXEC) 4433 phdr.p_flags |= PF_X; 4434 phdr.p_align = ELF_EXEC_PAGESIZE; 4435 4436 bswap_phdr(&phdr, 1); 4437 if (dump_write(fd, &phdr, sizeof(phdr)) != 0) { 4438 goto out; 4439 } 4440 } 4441 4442 /* 4443 * Next we write notes just after program headers. No 4444 * alignment needed here. 4445 */ 4446 if (write_note_info(&info, fd) < 0) 4447 goto out; 4448 4449 /* align data to page boundary */ 4450 if (lseek(fd, data_offset, SEEK_SET) != data_offset) 4451 goto out; 4452 4453 /* 4454 * Finally we can dump process memory into corefile as well. 4455 */ 4456 for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) { 4457 abi_ulong addr; 4458 abi_ulong end; 4459 4460 end = vma->vma_start + vma_dump_size(vma); 4461 4462 for (addr = vma->vma_start; addr < end; 4463 addr += TARGET_PAGE_SIZE) { 4464 char page[TARGET_PAGE_SIZE]; 4465 int error; 4466 4467 /* 4468 * Read in page from target process memory and 4469 * write it to coredump file. 4470 */ 4471 error = copy_from_user(page, addr, sizeof (page)); 4472 if (error != 0) { 4473 (void) fprintf(stderr, "unable to dump " TARGET_ABI_FMT_lx "\n", 4474 addr); 4475 errno = -error; 4476 goto out; 4477 } 4478 if (dump_write(fd, page, TARGET_PAGE_SIZE) < 0) 4479 goto out; 4480 } 4481 } 4482 4483 out: 4484 free_note_info(&info); 4485 if (mm != NULL) 4486 vma_delete(mm); 4487 (void) close(fd); 4488 4489 if (errno != 0) 4490 return (-errno); 4491 return (0); 4492 } 4493 #endif /* USE_ELF_CORE_DUMP */ 4494 4495 void do_init_thread(struct target_pt_regs *regs, struct image_info *infop) 4496 { 4497 init_thread(regs, infop); 4498 } 4499