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