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 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 const char *elf_hwcap_str(uint32_t bit) 1609 { 1610 static const char *hwcap_str[] = { 1611 [HWCAP_S390_ESAN3] = "esan3", 1612 [HWCAP_S390_ZARCH] = "zarch", 1613 [HWCAP_S390_STFLE] = "stfle", 1614 [HWCAP_S390_MSA] = "msa", 1615 [HWCAP_S390_LDISP] = "ldisp", 1616 [HWCAP_S390_EIMM] = "eimm", 1617 [HWCAP_S390_DFP] = "dfp", 1618 [HWCAP_S390_HPAGE] = "edat", 1619 [HWCAP_S390_ETF3EH] = "etf3eh", 1620 [HWCAP_S390_HIGH_GPRS] = "highgprs", 1621 [HWCAP_S390_TE] = "te", 1622 [HWCAP_S390_VXRS] = "vx", 1623 [HWCAP_S390_VXRS_BCD] = "vxd", 1624 [HWCAP_S390_VXRS_EXT] = "vxe", 1625 [HWCAP_S390_GS] = "gs", 1626 [HWCAP_S390_VXRS_EXT2] = "vxe2", 1627 [HWCAP_S390_VXRS_PDE] = "vxp", 1628 [HWCAP_S390_SORT] = "sort", 1629 [HWCAP_S390_DFLT] = "dflt", 1630 }; 1631 1632 return bit < ARRAY_SIZE(hwcap_str) ? hwcap_str[bit] : NULL; 1633 } 1634 1635 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop) 1636 { 1637 regs->psw.addr = infop->entry; 1638 regs->psw.mask = PSW_MASK_DAT | PSW_MASK_IO | PSW_MASK_EXT | \ 1639 PSW_MASK_MCHECK | PSW_MASK_PSTATE | PSW_MASK_64 | \ 1640 PSW_MASK_32; 1641 regs->gprs[15] = infop->start_stack; 1642 } 1643 1644 /* See linux kernel: arch/s390/include/uapi/asm/ptrace.h (s390_regs). */ 1645 #define ELF_NREG 27 1646 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1647 1648 enum { 1649 TARGET_REG_PSWM = 0, 1650 TARGET_REG_PSWA = 1, 1651 TARGET_REG_GPRS = 2, 1652 TARGET_REG_ARS = 18, 1653 TARGET_REG_ORIG_R2 = 26, 1654 }; 1655 1656 static void elf_core_copy_regs(target_elf_gregset_t *regs, 1657 const CPUS390XState *env) 1658 { 1659 int i; 1660 uint32_t *aregs; 1661 1662 (*regs)[TARGET_REG_PSWM] = tswapreg(env->psw.mask); 1663 (*regs)[TARGET_REG_PSWA] = tswapreg(env->psw.addr); 1664 for (i = 0; i < 16; i++) { 1665 (*regs)[TARGET_REG_GPRS + i] = tswapreg(env->regs[i]); 1666 } 1667 aregs = (uint32_t *)&((*regs)[TARGET_REG_ARS]); 1668 for (i = 0; i < 16; i++) { 1669 aregs[i] = tswap32(env->aregs[i]); 1670 } 1671 (*regs)[TARGET_REG_ORIG_R2] = 0; 1672 } 1673 1674 #define USE_ELF_CORE_DUMP 1675 #define ELF_EXEC_PAGESIZE 4096 1676 1677 #endif /* TARGET_S390X */ 1678 1679 #ifdef TARGET_RISCV 1680 1681 #define ELF_START_MMAP 0x80000000 1682 #define ELF_ARCH EM_RISCV 1683 1684 #ifdef TARGET_RISCV32 1685 #define ELF_CLASS ELFCLASS32 1686 #else 1687 #define ELF_CLASS ELFCLASS64 1688 #endif 1689 1690 #define ELF_HWCAP get_elf_hwcap() 1691 1692 static uint32_t get_elf_hwcap(void) 1693 { 1694 #define MISA_BIT(EXT) (1 << (EXT - 'A')) 1695 RISCVCPU *cpu = RISCV_CPU(thread_cpu); 1696 uint32_t mask = MISA_BIT('I') | MISA_BIT('M') | MISA_BIT('A') 1697 | MISA_BIT('F') | MISA_BIT('D') | MISA_BIT('C'); 1698 1699 return cpu->env.misa_ext & mask; 1700 #undef MISA_BIT 1701 } 1702 1703 static inline void init_thread(struct target_pt_regs *regs, 1704 struct image_info *infop) 1705 { 1706 regs->sepc = infop->entry; 1707 regs->sp = infop->start_stack; 1708 } 1709 1710 #define ELF_EXEC_PAGESIZE 4096 1711 1712 #endif /* TARGET_RISCV */ 1713 1714 #ifdef TARGET_HPPA 1715 1716 #define ELF_START_MMAP 0x80000000 1717 #define ELF_CLASS ELFCLASS32 1718 #define ELF_ARCH EM_PARISC 1719 #define ELF_PLATFORM "PARISC" 1720 #define STACK_GROWS_DOWN 0 1721 #define STACK_ALIGNMENT 64 1722 1723 static inline void init_thread(struct target_pt_regs *regs, 1724 struct image_info *infop) 1725 { 1726 regs->iaoq[0] = infop->entry; 1727 regs->iaoq[1] = infop->entry + 4; 1728 regs->gr[23] = 0; 1729 regs->gr[24] = infop->argv; 1730 regs->gr[25] = infop->argc; 1731 /* The top-of-stack contains a linkage buffer. */ 1732 regs->gr[30] = infop->start_stack + 64; 1733 regs->gr[31] = infop->entry; 1734 } 1735 1736 #define LO_COMMPAGE 0 1737 1738 static bool init_guest_commpage(void) 1739 { 1740 void *want = g2h_untagged(LO_COMMPAGE); 1741 void *addr = mmap(want, qemu_host_page_size, PROT_NONE, 1742 MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0); 1743 1744 if (addr == MAP_FAILED) { 1745 perror("Allocating guest commpage"); 1746 exit(EXIT_FAILURE); 1747 } 1748 if (addr != want) { 1749 return false; 1750 } 1751 1752 /* 1753 * On Linux, page zero is normally marked execute only + gateway. 1754 * Normal read or write is supposed to fail (thus PROT_NONE above), 1755 * but specific offsets have kernel code mapped to raise permissions 1756 * and implement syscalls. Here, simply mark the page executable. 1757 * Special case the entry points during translation (see do_page_zero). 1758 */ 1759 page_set_flags(LO_COMMPAGE, LO_COMMPAGE | ~TARGET_PAGE_MASK, 1760 PAGE_EXEC | PAGE_VALID); 1761 return true; 1762 } 1763 1764 #endif /* TARGET_HPPA */ 1765 1766 #ifdef TARGET_XTENSA 1767 1768 #define ELF_START_MMAP 0x20000000 1769 1770 #define ELF_CLASS ELFCLASS32 1771 #define ELF_ARCH EM_XTENSA 1772 1773 static inline void init_thread(struct target_pt_regs *regs, 1774 struct image_info *infop) 1775 { 1776 regs->windowbase = 0; 1777 regs->windowstart = 1; 1778 regs->areg[1] = infop->start_stack; 1779 regs->pc = infop->entry; 1780 if (info_is_fdpic(infop)) { 1781 regs->areg[4] = infop->loadmap_addr; 1782 regs->areg[5] = infop->interpreter_loadmap_addr; 1783 if (infop->interpreter_loadmap_addr) { 1784 regs->areg[6] = infop->interpreter_pt_dynamic_addr; 1785 } else { 1786 regs->areg[6] = infop->pt_dynamic_addr; 1787 } 1788 } 1789 } 1790 1791 /* See linux kernel: arch/xtensa/include/asm/elf.h. */ 1792 #define ELF_NREG 128 1793 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1794 1795 enum { 1796 TARGET_REG_PC, 1797 TARGET_REG_PS, 1798 TARGET_REG_LBEG, 1799 TARGET_REG_LEND, 1800 TARGET_REG_LCOUNT, 1801 TARGET_REG_SAR, 1802 TARGET_REG_WINDOWSTART, 1803 TARGET_REG_WINDOWBASE, 1804 TARGET_REG_THREADPTR, 1805 TARGET_REG_AR0 = 64, 1806 }; 1807 1808 static void elf_core_copy_regs(target_elf_gregset_t *regs, 1809 const CPUXtensaState *env) 1810 { 1811 unsigned i; 1812 1813 (*regs)[TARGET_REG_PC] = tswapreg(env->pc); 1814 (*regs)[TARGET_REG_PS] = tswapreg(env->sregs[PS] & ~PS_EXCM); 1815 (*regs)[TARGET_REG_LBEG] = tswapreg(env->sregs[LBEG]); 1816 (*regs)[TARGET_REG_LEND] = tswapreg(env->sregs[LEND]); 1817 (*regs)[TARGET_REG_LCOUNT] = tswapreg(env->sregs[LCOUNT]); 1818 (*regs)[TARGET_REG_SAR] = tswapreg(env->sregs[SAR]); 1819 (*regs)[TARGET_REG_WINDOWSTART] = tswapreg(env->sregs[WINDOW_START]); 1820 (*regs)[TARGET_REG_WINDOWBASE] = tswapreg(env->sregs[WINDOW_BASE]); 1821 (*regs)[TARGET_REG_THREADPTR] = tswapreg(env->uregs[THREADPTR]); 1822 xtensa_sync_phys_from_window((CPUXtensaState *)env); 1823 for (i = 0; i < env->config->nareg; ++i) { 1824 (*regs)[TARGET_REG_AR0 + i] = tswapreg(env->phys_regs[i]); 1825 } 1826 } 1827 1828 #define USE_ELF_CORE_DUMP 1829 #define ELF_EXEC_PAGESIZE 4096 1830 1831 #endif /* TARGET_XTENSA */ 1832 1833 #ifdef TARGET_HEXAGON 1834 1835 #define ELF_START_MMAP 0x20000000 1836 1837 #define ELF_CLASS ELFCLASS32 1838 #define ELF_ARCH EM_HEXAGON 1839 1840 static inline void init_thread(struct target_pt_regs *regs, 1841 struct image_info *infop) 1842 { 1843 regs->sepc = infop->entry; 1844 regs->sp = infop->start_stack; 1845 } 1846 1847 #endif /* TARGET_HEXAGON */ 1848 1849 #ifndef ELF_BASE_PLATFORM 1850 #define ELF_BASE_PLATFORM (NULL) 1851 #endif 1852 1853 #ifndef ELF_PLATFORM 1854 #define ELF_PLATFORM (NULL) 1855 #endif 1856 1857 #ifndef ELF_MACHINE 1858 #define ELF_MACHINE ELF_ARCH 1859 #endif 1860 1861 #ifndef elf_check_arch 1862 #define elf_check_arch(x) ((x) == ELF_ARCH) 1863 #endif 1864 1865 #ifndef elf_check_abi 1866 #define elf_check_abi(x) (1) 1867 #endif 1868 1869 #ifndef ELF_HWCAP 1870 #define ELF_HWCAP 0 1871 #endif 1872 1873 #ifndef STACK_GROWS_DOWN 1874 #define STACK_GROWS_DOWN 1 1875 #endif 1876 1877 #ifndef STACK_ALIGNMENT 1878 #define STACK_ALIGNMENT 16 1879 #endif 1880 1881 #ifdef TARGET_ABI32 1882 #undef ELF_CLASS 1883 #define ELF_CLASS ELFCLASS32 1884 #undef bswaptls 1885 #define bswaptls(ptr) bswap32s(ptr) 1886 #endif 1887 1888 #ifndef EXSTACK_DEFAULT 1889 #define EXSTACK_DEFAULT false 1890 #endif 1891 1892 #include "elf.h" 1893 1894 /* We must delay the following stanzas until after "elf.h". */ 1895 #if defined(TARGET_AARCH64) 1896 1897 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz, 1898 const uint32_t *data, 1899 struct image_info *info, 1900 Error **errp) 1901 { 1902 if (pr_type == GNU_PROPERTY_AARCH64_FEATURE_1_AND) { 1903 if (pr_datasz != sizeof(uint32_t)) { 1904 error_setg(errp, "Ill-formed GNU_PROPERTY_AARCH64_FEATURE_1_AND"); 1905 return false; 1906 } 1907 /* We will extract GNU_PROPERTY_AARCH64_FEATURE_1_BTI later. */ 1908 info->note_flags = *data; 1909 } 1910 return true; 1911 } 1912 #define ARCH_USE_GNU_PROPERTY 1 1913 1914 #else 1915 1916 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz, 1917 const uint32_t *data, 1918 struct image_info *info, 1919 Error **errp) 1920 { 1921 g_assert_not_reached(); 1922 } 1923 #define ARCH_USE_GNU_PROPERTY 0 1924 1925 #endif 1926 1927 struct exec 1928 { 1929 unsigned int a_info; /* Use macros N_MAGIC, etc for access */ 1930 unsigned int a_text; /* length of text, in bytes */ 1931 unsigned int a_data; /* length of data, in bytes */ 1932 unsigned int a_bss; /* length of uninitialized data area, in bytes */ 1933 unsigned int a_syms; /* length of symbol table data in file, in bytes */ 1934 unsigned int a_entry; /* start address */ 1935 unsigned int a_trsize; /* length of relocation info for text, in bytes */ 1936 unsigned int a_drsize; /* length of relocation info for data, in bytes */ 1937 }; 1938 1939 1940 #define N_MAGIC(exec) ((exec).a_info & 0xffff) 1941 #define OMAGIC 0407 1942 #define NMAGIC 0410 1943 #define ZMAGIC 0413 1944 #define QMAGIC 0314 1945 1946 /* Necessary parameters */ 1947 #define TARGET_ELF_EXEC_PAGESIZE \ 1948 (((eppnt->p_align & ~qemu_host_page_mask) != 0) ? \ 1949 TARGET_PAGE_SIZE : MAX(qemu_host_page_size, TARGET_PAGE_SIZE)) 1950 #define TARGET_ELF_PAGELENGTH(_v) ROUND_UP((_v), TARGET_ELF_EXEC_PAGESIZE) 1951 #define TARGET_ELF_PAGESTART(_v) ((_v) & \ 1952 ~(abi_ulong)(TARGET_ELF_EXEC_PAGESIZE-1)) 1953 #define TARGET_ELF_PAGEOFFSET(_v) ((_v) & (TARGET_ELF_EXEC_PAGESIZE-1)) 1954 1955 #define DLINFO_ITEMS 16 1956 1957 static inline void memcpy_fromfs(void * to, const void * from, unsigned long n) 1958 { 1959 memcpy(to, from, n); 1960 } 1961 1962 #ifdef BSWAP_NEEDED 1963 static void bswap_ehdr(struct elfhdr *ehdr) 1964 { 1965 bswap16s(&ehdr->e_type); /* Object file type */ 1966 bswap16s(&ehdr->e_machine); /* Architecture */ 1967 bswap32s(&ehdr->e_version); /* Object file version */ 1968 bswaptls(&ehdr->e_entry); /* Entry point virtual address */ 1969 bswaptls(&ehdr->e_phoff); /* Program header table file offset */ 1970 bswaptls(&ehdr->e_shoff); /* Section header table file offset */ 1971 bswap32s(&ehdr->e_flags); /* Processor-specific flags */ 1972 bswap16s(&ehdr->e_ehsize); /* ELF header size in bytes */ 1973 bswap16s(&ehdr->e_phentsize); /* Program header table entry size */ 1974 bswap16s(&ehdr->e_phnum); /* Program header table entry count */ 1975 bswap16s(&ehdr->e_shentsize); /* Section header table entry size */ 1976 bswap16s(&ehdr->e_shnum); /* Section header table entry count */ 1977 bswap16s(&ehdr->e_shstrndx); /* Section header string table index */ 1978 } 1979 1980 static void bswap_phdr(struct elf_phdr *phdr, int phnum) 1981 { 1982 int i; 1983 for (i = 0; i < phnum; ++i, ++phdr) { 1984 bswap32s(&phdr->p_type); /* Segment type */ 1985 bswap32s(&phdr->p_flags); /* Segment flags */ 1986 bswaptls(&phdr->p_offset); /* Segment file offset */ 1987 bswaptls(&phdr->p_vaddr); /* Segment virtual address */ 1988 bswaptls(&phdr->p_paddr); /* Segment physical address */ 1989 bswaptls(&phdr->p_filesz); /* Segment size in file */ 1990 bswaptls(&phdr->p_memsz); /* Segment size in memory */ 1991 bswaptls(&phdr->p_align); /* Segment alignment */ 1992 } 1993 } 1994 1995 static void bswap_shdr(struct elf_shdr *shdr, int shnum) 1996 { 1997 int i; 1998 for (i = 0; i < shnum; ++i, ++shdr) { 1999 bswap32s(&shdr->sh_name); 2000 bswap32s(&shdr->sh_type); 2001 bswaptls(&shdr->sh_flags); 2002 bswaptls(&shdr->sh_addr); 2003 bswaptls(&shdr->sh_offset); 2004 bswaptls(&shdr->sh_size); 2005 bswap32s(&shdr->sh_link); 2006 bswap32s(&shdr->sh_info); 2007 bswaptls(&shdr->sh_addralign); 2008 bswaptls(&shdr->sh_entsize); 2009 } 2010 } 2011 2012 static void bswap_sym(struct elf_sym *sym) 2013 { 2014 bswap32s(&sym->st_name); 2015 bswaptls(&sym->st_value); 2016 bswaptls(&sym->st_size); 2017 bswap16s(&sym->st_shndx); 2018 } 2019 2020 #ifdef TARGET_MIPS 2021 static void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags) 2022 { 2023 bswap16s(&abiflags->version); 2024 bswap32s(&abiflags->ases); 2025 bswap32s(&abiflags->isa_ext); 2026 bswap32s(&abiflags->flags1); 2027 bswap32s(&abiflags->flags2); 2028 } 2029 #endif 2030 #else 2031 static inline void bswap_ehdr(struct elfhdr *ehdr) { } 2032 static inline void bswap_phdr(struct elf_phdr *phdr, int phnum) { } 2033 static inline void bswap_shdr(struct elf_shdr *shdr, int shnum) { } 2034 static inline void bswap_sym(struct elf_sym *sym) { } 2035 #ifdef TARGET_MIPS 2036 static inline void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags) { } 2037 #endif 2038 #endif 2039 2040 #ifdef USE_ELF_CORE_DUMP 2041 static int elf_core_dump(int, const CPUArchState *); 2042 #endif /* USE_ELF_CORE_DUMP */ 2043 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias); 2044 2045 /* Verify the portions of EHDR within E_IDENT for the target. 2046 This can be performed before bswapping the entire header. */ 2047 static bool elf_check_ident(struct elfhdr *ehdr) 2048 { 2049 return (ehdr->e_ident[EI_MAG0] == ELFMAG0 2050 && ehdr->e_ident[EI_MAG1] == ELFMAG1 2051 && ehdr->e_ident[EI_MAG2] == ELFMAG2 2052 && ehdr->e_ident[EI_MAG3] == ELFMAG3 2053 && ehdr->e_ident[EI_CLASS] == ELF_CLASS 2054 && ehdr->e_ident[EI_DATA] == ELF_DATA 2055 && ehdr->e_ident[EI_VERSION] == EV_CURRENT); 2056 } 2057 2058 /* Verify the portions of EHDR outside of E_IDENT for the target. 2059 This has to wait until after bswapping the header. */ 2060 static bool elf_check_ehdr(struct elfhdr *ehdr) 2061 { 2062 return (elf_check_arch(ehdr->e_machine) 2063 && elf_check_abi(ehdr->e_flags) 2064 && ehdr->e_ehsize == sizeof(struct elfhdr) 2065 && ehdr->e_phentsize == sizeof(struct elf_phdr) 2066 && (ehdr->e_type == ET_EXEC || ehdr->e_type == ET_DYN)); 2067 } 2068 2069 /* 2070 * 'copy_elf_strings()' copies argument/envelope strings from user 2071 * memory to free pages in kernel mem. These are in a format ready 2072 * to be put directly into the top of new user memory. 2073 * 2074 */ 2075 static abi_ulong copy_elf_strings(int argc, char **argv, char *scratch, 2076 abi_ulong p, abi_ulong stack_limit) 2077 { 2078 char *tmp; 2079 int len, i; 2080 abi_ulong top = p; 2081 2082 if (!p) { 2083 return 0; /* bullet-proofing */ 2084 } 2085 2086 if (STACK_GROWS_DOWN) { 2087 int offset = ((p - 1) % TARGET_PAGE_SIZE) + 1; 2088 for (i = argc - 1; i >= 0; --i) { 2089 tmp = argv[i]; 2090 if (!tmp) { 2091 fprintf(stderr, "VFS: argc is wrong"); 2092 exit(-1); 2093 } 2094 len = strlen(tmp) + 1; 2095 tmp += len; 2096 2097 if (len > (p - stack_limit)) { 2098 return 0; 2099 } 2100 while (len) { 2101 int bytes_to_copy = (len > offset) ? offset : len; 2102 tmp -= bytes_to_copy; 2103 p -= bytes_to_copy; 2104 offset -= bytes_to_copy; 2105 len -= bytes_to_copy; 2106 2107 memcpy_fromfs(scratch + offset, tmp, bytes_to_copy); 2108 2109 if (offset == 0) { 2110 memcpy_to_target(p, scratch, top - p); 2111 top = p; 2112 offset = TARGET_PAGE_SIZE; 2113 } 2114 } 2115 } 2116 if (p != top) { 2117 memcpy_to_target(p, scratch + offset, top - p); 2118 } 2119 } else { 2120 int remaining = TARGET_PAGE_SIZE - (p % TARGET_PAGE_SIZE); 2121 for (i = 0; i < argc; ++i) { 2122 tmp = argv[i]; 2123 if (!tmp) { 2124 fprintf(stderr, "VFS: argc is wrong"); 2125 exit(-1); 2126 } 2127 len = strlen(tmp) + 1; 2128 if (len > (stack_limit - p)) { 2129 return 0; 2130 } 2131 while (len) { 2132 int bytes_to_copy = (len > remaining) ? remaining : len; 2133 2134 memcpy_fromfs(scratch + (p - top), tmp, bytes_to_copy); 2135 2136 tmp += bytes_to_copy; 2137 remaining -= bytes_to_copy; 2138 p += bytes_to_copy; 2139 len -= bytes_to_copy; 2140 2141 if (remaining == 0) { 2142 memcpy_to_target(top, scratch, p - top); 2143 top = p; 2144 remaining = TARGET_PAGE_SIZE; 2145 } 2146 } 2147 } 2148 if (p != top) { 2149 memcpy_to_target(top, scratch, p - top); 2150 } 2151 } 2152 2153 return p; 2154 } 2155 2156 /* Older linux kernels provide up to MAX_ARG_PAGES (default: 32) of 2157 * argument/environment space. Newer kernels (>2.6.33) allow more, 2158 * dependent on stack size, but guarantee at least 32 pages for 2159 * backwards compatibility. 2160 */ 2161 #define STACK_LOWER_LIMIT (32 * TARGET_PAGE_SIZE) 2162 2163 static abi_ulong setup_arg_pages(struct linux_binprm *bprm, 2164 struct image_info *info) 2165 { 2166 abi_ulong size, error, guard; 2167 int prot; 2168 2169 size = guest_stack_size; 2170 if (size < STACK_LOWER_LIMIT) { 2171 size = STACK_LOWER_LIMIT; 2172 } 2173 2174 if (STACK_GROWS_DOWN) { 2175 guard = TARGET_PAGE_SIZE; 2176 if (guard < qemu_real_host_page_size()) { 2177 guard = qemu_real_host_page_size(); 2178 } 2179 } else { 2180 /* no guard page for hppa target where stack grows upwards. */ 2181 guard = 0; 2182 } 2183 2184 prot = PROT_READ | PROT_WRITE; 2185 if (info->exec_stack) { 2186 prot |= PROT_EXEC; 2187 } 2188 error = target_mmap(0, size + guard, prot, 2189 MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); 2190 if (error == -1) { 2191 perror("mmap stack"); 2192 exit(-1); 2193 } 2194 2195 /* We reserve one extra page at the top of the stack as guard. */ 2196 if (STACK_GROWS_DOWN) { 2197 target_mprotect(error, guard, PROT_NONE); 2198 info->stack_limit = error + guard; 2199 return info->stack_limit + size - sizeof(void *); 2200 } else { 2201 info->stack_limit = error + size; 2202 return error; 2203 } 2204 } 2205 2206 /* Map and zero the bss. We need to explicitly zero any fractional pages 2207 after the data section (i.e. bss). */ 2208 static void zero_bss(abi_ulong elf_bss, abi_ulong last_bss, int prot) 2209 { 2210 uintptr_t host_start, host_map_start, host_end; 2211 2212 last_bss = TARGET_PAGE_ALIGN(last_bss); 2213 2214 /* ??? There is confusion between qemu_real_host_page_size and 2215 qemu_host_page_size here and elsewhere in target_mmap, which 2216 may lead to the end of the data section mapping from the file 2217 not being mapped. At least there was an explicit test and 2218 comment for that here, suggesting that "the file size must 2219 be known". The comment probably pre-dates the introduction 2220 of the fstat system call in target_mmap which does in fact 2221 find out the size. What isn't clear is if the workaround 2222 here is still actually needed. For now, continue with it, 2223 but merge it with the "normal" mmap that would allocate the bss. */ 2224 2225 host_start = (uintptr_t) g2h_untagged(elf_bss); 2226 host_end = (uintptr_t) g2h_untagged(last_bss); 2227 host_map_start = REAL_HOST_PAGE_ALIGN(host_start); 2228 2229 if (host_map_start < host_end) { 2230 void *p = mmap((void *)host_map_start, host_end - host_map_start, 2231 prot, MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); 2232 if (p == MAP_FAILED) { 2233 perror("cannot mmap brk"); 2234 exit(-1); 2235 } 2236 } 2237 2238 /* Ensure that the bss page(s) are valid */ 2239 if ((page_get_flags(last_bss-1) & prot) != prot) { 2240 page_set_flags(elf_bss & TARGET_PAGE_MASK, last_bss - 1, 2241 prot | PAGE_VALID); 2242 } 2243 2244 if (host_start < host_map_start) { 2245 memset((void *)host_start, 0, host_map_start - host_start); 2246 } 2247 } 2248 2249 #if defined(TARGET_ARM) 2250 static int elf_is_fdpic(struct elfhdr *exec) 2251 { 2252 return exec->e_ident[EI_OSABI] == ELFOSABI_ARM_FDPIC; 2253 } 2254 #elif defined(TARGET_XTENSA) 2255 static int elf_is_fdpic(struct elfhdr *exec) 2256 { 2257 return exec->e_ident[EI_OSABI] == ELFOSABI_XTENSA_FDPIC; 2258 } 2259 #else 2260 /* Default implementation, always false. */ 2261 static int elf_is_fdpic(struct elfhdr *exec) 2262 { 2263 return 0; 2264 } 2265 #endif 2266 2267 static abi_ulong loader_build_fdpic_loadmap(struct image_info *info, abi_ulong sp) 2268 { 2269 uint16_t n; 2270 struct elf32_fdpic_loadseg *loadsegs = info->loadsegs; 2271 2272 /* elf32_fdpic_loadseg */ 2273 n = info->nsegs; 2274 while (n--) { 2275 sp -= 12; 2276 put_user_u32(loadsegs[n].addr, sp+0); 2277 put_user_u32(loadsegs[n].p_vaddr, sp+4); 2278 put_user_u32(loadsegs[n].p_memsz, sp+8); 2279 } 2280 2281 /* elf32_fdpic_loadmap */ 2282 sp -= 4; 2283 put_user_u16(0, sp+0); /* version */ 2284 put_user_u16(info->nsegs, sp+2); /* nsegs */ 2285 2286 info->personality = PER_LINUX_FDPIC; 2287 info->loadmap_addr = sp; 2288 2289 return sp; 2290 } 2291 2292 static abi_ulong create_elf_tables(abi_ulong p, int argc, int envc, 2293 struct elfhdr *exec, 2294 struct image_info *info, 2295 struct image_info *interp_info) 2296 { 2297 abi_ulong sp; 2298 abi_ulong u_argc, u_argv, u_envp, u_auxv; 2299 int size; 2300 int i; 2301 abi_ulong u_rand_bytes; 2302 uint8_t k_rand_bytes[16]; 2303 abi_ulong u_platform, u_base_platform; 2304 const char *k_platform, *k_base_platform; 2305 const int n = sizeof(elf_addr_t); 2306 2307 sp = p; 2308 2309 /* Needs to be before we load the env/argc/... */ 2310 if (elf_is_fdpic(exec)) { 2311 /* Need 4 byte alignment for these structs */ 2312 sp &= ~3; 2313 sp = loader_build_fdpic_loadmap(info, sp); 2314 info->other_info = interp_info; 2315 if (interp_info) { 2316 interp_info->other_info = info; 2317 sp = loader_build_fdpic_loadmap(interp_info, sp); 2318 info->interpreter_loadmap_addr = interp_info->loadmap_addr; 2319 info->interpreter_pt_dynamic_addr = interp_info->pt_dynamic_addr; 2320 } else { 2321 info->interpreter_loadmap_addr = 0; 2322 info->interpreter_pt_dynamic_addr = 0; 2323 } 2324 } 2325 2326 u_base_platform = 0; 2327 k_base_platform = ELF_BASE_PLATFORM; 2328 if (k_base_platform) { 2329 size_t len = strlen(k_base_platform) + 1; 2330 if (STACK_GROWS_DOWN) { 2331 sp -= (len + n - 1) & ~(n - 1); 2332 u_base_platform = sp; 2333 /* FIXME - check return value of memcpy_to_target() for failure */ 2334 memcpy_to_target(sp, k_base_platform, len); 2335 } else { 2336 memcpy_to_target(sp, k_base_platform, len); 2337 u_base_platform = sp; 2338 sp += len + 1; 2339 } 2340 } 2341 2342 u_platform = 0; 2343 k_platform = ELF_PLATFORM; 2344 if (k_platform) { 2345 size_t len = strlen(k_platform) + 1; 2346 if (STACK_GROWS_DOWN) { 2347 sp -= (len + n - 1) & ~(n - 1); 2348 u_platform = sp; 2349 /* FIXME - check return value of memcpy_to_target() for failure */ 2350 memcpy_to_target(sp, k_platform, len); 2351 } else { 2352 memcpy_to_target(sp, k_platform, len); 2353 u_platform = sp; 2354 sp += len + 1; 2355 } 2356 } 2357 2358 /* Provide 16 byte alignment for the PRNG, and basic alignment for 2359 * the argv and envp pointers. 2360 */ 2361 if (STACK_GROWS_DOWN) { 2362 sp = QEMU_ALIGN_DOWN(sp, 16); 2363 } else { 2364 sp = QEMU_ALIGN_UP(sp, 16); 2365 } 2366 2367 /* 2368 * Generate 16 random bytes for userspace PRNG seeding. 2369 */ 2370 qemu_guest_getrandom_nofail(k_rand_bytes, sizeof(k_rand_bytes)); 2371 if (STACK_GROWS_DOWN) { 2372 sp -= 16; 2373 u_rand_bytes = sp; 2374 /* FIXME - check return value of memcpy_to_target() for failure */ 2375 memcpy_to_target(sp, k_rand_bytes, 16); 2376 } else { 2377 memcpy_to_target(sp, k_rand_bytes, 16); 2378 u_rand_bytes = sp; 2379 sp += 16; 2380 } 2381 2382 size = (DLINFO_ITEMS + 1) * 2; 2383 if (k_base_platform) 2384 size += 2; 2385 if (k_platform) 2386 size += 2; 2387 #ifdef DLINFO_ARCH_ITEMS 2388 size += DLINFO_ARCH_ITEMS * 2; 2389 #endif 2390 #ifdef ELF_HWCAP2 2391 size += 2; 2392 #endif 2393 info->auxv_len = size * n; 2394 2395 size += envc + argc + 2; 2396 size += 1; /* argc itself */ 2397 size *= n; 2398 2399 /* Allocate space and finalize stack alignment for entry now. */ 2400 if (STACK_GROWS_DOWN) { 2401 u_argc = QEMU_ALIGN_DOWN(sp - size, STACK_ALIGNMENT); 2402 sp = u_argc; 2403 } else { 2404 u_argc = sp; 2405 sp = QEMU_ALIGN_UP(sp + size, STACK_ALIGNMENT); 2406 } 2407 2408 u_argv = u_argc + n; 2409 u_envp = u_argv + (argc + 1) * n; 2410 u_auxv = u_envp + (envc + 1) * n; 2411 info->saved_auxv = u_auxv; 2412 info->argc = argc; 2413 info->envc = envc; 2414 info->argv = u_argv; 2415 info->envp = u_envp; 2416 2417 /* This is correct because Linux defines 2418 * elf_addr_t as Elf32_Off / Elf64_Off 2419 */ 2420 #define NEW_AUX_ENT(id, val) do { \ 2421 put_user_ual(id, u_auxv); u_auxv += n; \ 2422 put_user_ual(val, u_auxv); u_auxv += n; \ 2423 } while(0) 2424 2425 #ifdef ARCH_DLINFO 2426 /* 2427 * ARCH_DLINFO must come first so platform specific code can enforce 2428 * special alignment requirements on the AUXV if necessary (eg. PPC). 2429 */ 2430 ARCH_DLINFO; 2431 #endif 2432 /* There must be exactly DLINFO_ITEMS entries here, or the assert 2433 * on info->auxv_len will trigger. 2434 */ 2435 NEW_AUX_ENT(AT_PHDR, (abi_ulong)(info->load_addr + exec->e_phoff)); 2436 NEW_AUX_ENT(AT_PHENT, (abi_ulong)(sizeof (struct elf_phdr))); 2437 NEW_AUX_ENT(AT_PHNUM, (abi_ulong)(exec->e_phnum)); 2438 if ((info->alignment & ~qemu_host_page_mask) != 0) { 2439 /* Target doesn't support host page size alignment */ 2440 NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(TARGET_PAGE_SIZE)); 2441 } else { 2442 NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(MAX(TARGET_PAGE_SIZE, 2443 qemu_host_page_size))); 2444 } 2445 NEW_AUX_ENT(AT_BASE, (abi_ulong)(interp_info ? interp_info->load_addr : 0)); 2446 NEW_AUX_ENT(AT_FLAGS, (abi_ulong)0); 2447 NEW_AUX_ENT(AT_ENTRY, info->entry); 2448 NEW_AUX_ENT(AT_UID, (abi_ulong) getuid()); 2449 NEW_AUX_ENT(AT_EUID, (abi_ulong) geteuid()); 2450 NEW_AUX_ENT(AT_GID, (abi_ulong) getgid()); 2451 NEW_AUX_ENT(AT_EGID, (abi_ulong) getegid()); 2452 NEW_AUX_ENT(AT_HWCAP, (abi_ulong) ELF_HWCAP); 2453 NEW_AUX_ENT(AT_CLKTCK, (abi_ulong) sysconf(_SC_CLK_TCK)); 2454 NEW_AUX_ENT(AT_RANDOM, (abi_ulong) u_rand_bytes); 2455 NEW_AUX_ENT(AT_SECURE, (abi_ulong) qemu_getauxval(AT_SECURE)); 2456 NEW_AUX_ENT(AT_EXECFN, info->file_string); 2457 2458 #ifdef ELF_HWCAP2 2459 NEW_AUX_ENT(AT_HWCAP2, (abi_ulong) ELF_HWCAP2); 2460 #endif 2461 2462 if (u_base_platform) { 2463 NEW_AUX_ENT(AT_BASE_PLATFORM, u_base_platform); 2464 } 2465 if (u_platform) { 2466 NEW_AUX_ENT(AT_PLATFORM, u_platform); 2467 } 2468 NEW_AUX_ENT (AT_NULL, 0); 2469 #undef NEW_AUX_ENT 2470 2471 /* Check that our initial calculation of the auxv length matches how much 2472 * we actually put into it. 2473 */ 2474 assert(info->auxv_len == u_auxv - info->saved_auxv); 2475 2476 put_user_ual(argc, u_argc); 2477 2478 p = info->arg_strings; 2479 for (i = 0; i < argc; ++i) { 2480 put_user_ual(p, u_argv); 2481 u_argv += n; 2482 p += target_strlen(p) + 1; 2483 } 2484 put_user_ual(0, u_argv); 2485 2486 p = info->env_strings; 2487 for (i = 0; i < envc; ++i) { 2488 put_user_ual(p, u_envp); 2489 u_envp += n; 2490 p += target_strlen(p) + 1; 2491 } 2492 put_user_ual(0, u_envp); 2493 2494 return sp; 2495 } 2496 2497 #if defined(HI_COMMPAGE) 2498 #define LO_COMMPAGE -1 2499 #elif defined(LO_COMMPAGE) 2500 #define HI_COMMPAGE 0 2501 #else 2502 #define HI_COMMPAGE 0 2503 #define LO_COMMPAGE -1 2504 #ifndef INIT_GUEST_COMMPAGE 2505 #define init_guest_commpage() true 2506 #endif 2507 #endif 2508 2509 static void pgb_fail_in_use(const char *image_name) 2510 { 2511 error_report("%s: requires virtual address space that is in use " 2512 "(omit the -B option or choose a different value)", 2513 image_name); 2514 exit(EXIT_FAILURE); 2515 } 2516 2517 static void pgb_have_guest_base(const char *image_name, abi_ulong guest_loaddr, 2518 abi_ulong guest_hiaddr, long align) 2519 { 2520 const int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE; 2521 void *addr, *test; 2522 2523 if (!QEMU_IS_ALIGNED(guest_base, align)) { 2524 fprintf(stderr, "Requested guest base %p does not satisfy " 2525 "host minimum alignment (0x%lx)\n", 2526 (void *)guest_base, align); 2527 exit(EXIT_FAILURE); 2528 } 2529 2530 /* Sanity check the guest binary. */ 2531 if (reserved_va) { 2532 if (guest_hiaddr > reserved_va) { 2533 error_report("%s: requires more than reserved virtual " 2534 "address space (0x%" PRIx64 " > 0x%lx)", 2535 image_name, (uint64_t)guest_hiaddr, reserved_va); 2536 exit(EXIT_FAILURE); 2537 } 2538 } else { 2539 #if HOST_LONG_BITS < TARGET_ABI_BITS 2540 if ((guest_hiaddr - guest_base) > ~(uintptr_t)0) { 2541 error_report("%s: requires more virtual address space " 2542 "than the host can provide (0x%" PRIx64 ")", 2543 image_name, (uint64_t)guest_hiaddr + 1 - guest_base); 2544 exit(EXIT_FAILURE); 2545 } 2546 #endif 2547 } 2548 2549 /* 2550 * Expand the allocation to the entire reserved_va. 2551 * Exclude the mmap_min_addr hole. 2552 */ 2553 if (reserved_va) { 2554 guest_loaddr = (guest_base >= mmap_min_addr ? 0 2555 : mmap_min_addr - guest_base); 2556 guest_hiaddr = reserved_va; 2557 } 2558 2559 /* Reserve the address space for the binary, or reserved_va. */ 2560 test = g2h_untagged(guest_loaddr); 2561 addr = mmap(test, guest_hiaddr - guest_loaddr + 1, PROT_NONE, flags, -1, 0); 2562 if (test != addr) { 2563 pgb_fail_in_use(image_name); 2564 } 2565 qemu_log_mask(CPU_LOG_PAGE, 2566 "%s: base @ %p for %" PRIu64 " bytes\n", 2567 __func__, addr, (uint64_t)guest_hiaddr - guest_loaddr + 1); 2568 } 2569 2570 /** 2571 * pgd_find_hole_fallback: potential mmap address 2572 * @guest_size: size of available space 2573 * @brk: location of break 2574 * @align: memory alignment 2575 * 2576 * This is a fallback method for finding a hole in the host address 2577 * space if we don't have the benefit of being able to access 2578 * /proc/self/map. It can potentially take a very long time as we can 2579 * only dumbly iterate up the host address space seeing if the 2580 * allocation would work. 2581 */ 2582 static uintptr_t pgd_find_hole_fallback(uintptr_t guest_size, uintptr_t brk, 2583 long align, uintptr_t offset) 2584 { 2585 uintptr_t base; 2586 2587 /* Start (aligned) at the bottom and work our way up */ 2588 base = ROUND_UP(mmap_min_addr, align); 2589 2590 while (true) { 2591 uintptr_t align_start, end; 2592 align_start = ROUND_UP(base, align); 2593 end = align_start + guest_size + offset; 2594 2595 /* if brk is anywhere in the range give ourselves some room to grow. */ 2596 if (align_start <= brk && brk < end) { 2597 base = brk + (16 * MiB); 2598 continue; 2599 } else if (align_start + guest_size < align_start) { 2600 /* we have run out of space */ 2601 return -1; 2602 } else { 2603 int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE | 2604 MAP_FIXED_NOREPLACE; 2605 void * mmap_start = mmap((void *) align_start, guest_size, 2606 PROT_NONE, flags, -1, 0); 2607 if (mmap_start != MAP_FAILED) { 2608 munmap(mmap_start, guest_size); 2609 if (mmap_start == (void *) align_start) { 2610 qemu_log_mask(CPU_LOG_PAGE, 2611 "%s: base @ %p for %" PRIdPTR" bytes\n", 2612 __func__, mmap_start + offset, guest_size); 2613 return (uintptr_t) mmap_start + offset; 2614 } 2615 } 2616 base += qemu_host_page_size; 2617 } 2618 } 2619 } 2620 2621 /* Return value for guest_base, or -1 if no hole found. */ 2622 static uintptr_t pgb_find_hole(uintptr_t guest_loaddr, uintptr_t guest_size, 2623 long align, uintptr_t offset) 2624 { 2625 GSList *maps, *iter; 2626 uintptr_t this_start, this_end, next_start, brk; 2627 intptr_t ret = -1; 2628 2629 assert(QEMU_IS_ALIGNED(guest_loaddr, align)); 2630 2631 maps = read_self_maps(); 2632 2633 /* Read brk after we've read the maps, which will malloc. */ 2634 brk = (uintptr_t)sbrk(0); 2635 2636 if (!maps) { 2637 return pgd_find_hole_fallback(guest_size, brk, align, offset); 2638 } 2639 2640 /* The first hole is before the first map entry. */ 2641 this_start = mmap_min_addr; 2642 2643 for (iter = maps; iter; 2644 this_start = next_start, iter = g_slist_next(iter)) { 2645 uintptr_t align_start, hole_size; 2646 2647 this_end = ((MapInfo *)iter->data)->start; 2648 next_start = ((MapInfo *)iter->data)->end; 2649 align_start = ROUND_UP(this_start + offset, align); 2650 2651 /* Skip holes that are too small. */ 2652 if (align_start >= this_end) { 2653 continue; 2654 } 2655 hole_size = this_end - align_start; 2656 if (hole_size < guest_size) { 2657 continue; 2658 } 2659 2660 /* If this hole contains brk, give ourselves some room to grow. */ 2661 if (this_start <= brk && brk < this_end) { 2662 hole_size -= guest_size; 2663 if (sizeof(uintptr_t) == 8 && hole_size >= 1 * GiB) { 2664 align_start += 1 * GiB; 2665 } else if (hole_size >= 16 * MiB) { 2666 align_start += 16 * MiB; 2667 } else { 2668 align_start = (this_end - guest_size) & -align; 2669 if (align_start < this_start) { 2670 continue; 2671 } 2672 } 2673 } 2674 2675 /* Record the lowest successful match. */ 2676 if (ret < 0) { 2677 ret = align_start; 2678 } 2679 /* If this hole contains the identity map, select it. */ 2680 if (align_start <= guest_loaddr && 2681 guest_loaddr + guest_size <= this_end) { 2682 ret = 0; 2683 } 2684 /* If this hole ends above the identity map, stop looking. */ 2685 if (this_end >= guest_loaddr) { 2686 break; 2687 } 2688 } 2689 free_self_maps(maps); 2690 2691 if (ret != -1) { 2692 qemu_log_mask(CPU_LOG_PAGE, "%s: base @ %" PRIxPTR 2693 " for %" PRIuPTR " bytes\n", 2694 __func__, ret, guest_size); 2695 } 2696 2697 return ret; 2698 } 2699 2700 static void pgb_static(const char *image_name, abi_ulong orig_loaddr, 2701 abi_ulong orig_hiaddr, long align) 2702 { 2703 uintptr_t loaddr = orig_loaddr; 2704 uintptr_t hiaddr = orig_hiaddr; 2705 uintptr_t offset = 0; 2706 uintptr_t addr; 2707 2708 if (hiaddr != orig_hiaddr) { 2709 error_report("%s: requires virtual address space that the " 2710 "host cannot provide (0x%" PRIx64 ")", 2711 image_name, (uint64_t)orig_hiaddr + 1); 2712 exit(EXIT_FAILURE); 2713 } 2714 2715 loaddr &= -align; 2716 if (HI_COMMPAGE) { 2717 /* 2718 * Extend the allocation to include the commpage. 2719 * For a 64-bit host, this is just 4GiB; for a 32-bit host we 2720 * need to ensure there is space bellow the guest_base so we 2721 * can map the commpage in the place needed when the address 2722 * arithmetic wraps around. 2723 */ 2724 if (sizeof(uintptr_t) == 8 || loaddr >= 0x80000000u) { 2725 hiaddr = UINT32_MAX; 2726 } else { 2727 offset = -(HI_COMMPAGE & -align); 2728 } 2729 } else if (LO_COMMPAGE != -1) { 2730 loaddr = MIN(loaddr, LO_COMMPAGE & -align); 2731 } 2732 2733 addr = pgb_find_hole(loaddr, hiaddr - loaddr + 1, align, offset); 2734 if (addr == -1) { 2735 /* 2736 * If HI_COMMPAGE, there *might* be a non-consecutive allocation 2737 * that can satisfy both. But as the normal arm32 link base address 2738 * is ~32k, and we extend down to include the commpage, making the 2739 * overhead only ~96k, this is unlikely. 2740 */ 2741 error_report("%s: Unable to allocate %#zx bytes of " 2742 "virtual address space", image_name, 2743 (size_t)(hiaddr - loaddr)); 2744 exit(EXIT_FAILURE); 2745 } 2746 2747 guest_base = addr; 2748 2749 qemu_log_mask(CPU_LOG_PAGE, "%s: base @ %"PRIxPTR" for %" PRIuPTR" bytes\n", 2750 __func__, addr, hiaddr - loaddr); 2751 } 2752 2753 static void pgb_dynamic(const char *image_name, long align) 2754 { 2755 /* 2756 * The executable is dynamic and does not require a fixed address. 2757 * All we need is a commpage that satisfies align. 2758 * If we do not need a commpage, leave guest_base == 0. 2759 */ 2760 if (HI_COMMPAGE) { 2761 uintptr_t addr, commpage; 2762 2763 /* 64-bit hosts should have used reserved_va. */ 2764 assert(sizeof(uintptr_t) == 4); 2765 2766 /* 2767 * By putting the commpage at the first hole, that puts guest_base 2768 * just above that, and maximises the positive guest addresses. 2769 */ 2770 commpage = HI_COMMPAGE & -align; 2771 addr = pgb_find_hole(commpage, -commpage, align, 0); 2772 assert(addr != -1); 2773 guest_base = addr; 2774 } 2775 } 2776 2777 static void pgb_reserved_va(const char *image_name, abi_ulong guest_loaddr, 2778 abi_ulong guest_hiaddr, long align) 2779 { 2780 int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE; 2781 void *addr, *test; 2782 2783 if (guest_hiaddr > reserved_va) { 2784 error_report("%s: requires more than reserved virtual " 2785 "address space (0x%" PRIx64 " > 0x%lx)", 2786 image_name, (uint64_t)guest_hiaddr, reserved_va); 2787 exit(EXIT_FAILURE); 2788 } 2789 2790 /* Widen the "image" to the entire reserved address space. */ 2791 pgb_static(image_name, 0, reserved_va, align); 2792 2793 /* osdep.h defines this as 0 if it's missing */ 2794 flags |= MAP_FIXED_NOREPLACE; 2795 2796 /* Reserve the memory on the host. */ 2797 assert(guest_base != 0); 2798 test = g2h_untagged(0); 2799 addr = mmap(test, reserved_va + 1, PROT_NONE, flags, -1, 0); 2800 if (addr == MAP_FAILED || addr != test) { 2801 error_report("Unable to reserve 0x%lx bytes of virtual address " 2802 "space at %p (%s) for use as guest address space (check your " 2803 "virtual memory ulimit setting, mmap_min_addr or reserve less " 2804 "using qemu-user's -R option)", 2805 reserved_va + 1, test, strerror(errno)); 2806 exit(EXIT_FAILURE); 2807 } 2808 2809 qemu_log_mask(CPU_LOG_PAGE, "%s: base @ %p for %lu bytes\n", 2810 __func__, addr, reserved_va + 1); 2811 } 2812 2813 void probe_guest_base(const char *image_name, abi_ulong guest_loaddr, 2814 abi_ulong guest_hiaddr) 2815 { 2816 /* In order to use host shmat, we must be able to honor SHMLBA. */ 2817 uintptr_t align = MAX(SHMLBA, qemu_host_page_size); 2818 2819 if (have_guest_base) { 2820 pgb_have_guest_base(image_name, guest_loaddr, guest_hiaddr, align); 2821 } else if (reserved_va) { 2822 pgb_reserved_va(image_name, guest_loaddr, guest_hiaddr, align); 2823 } else if (guest_loaddr) { 2824 pgb_static(image_name, guest_loaddr, guest_hiaddr, align); 2825 } else { 2826 pgb_dynamic(image_name, align); 2827 } 2828 2829 /* Reserve and initialize the commpage. */ 2830 if (!init_guest_commpage()) { 2831 /* 2832 * With have_guest_base, the user has selected the address and 2833 * we are trying to work with that. Otherwise, we have selected 2834 * free space and init_guest_commpage must succeeded. 2835 */ 2836 assert(have_guest_base); 2837 pgb_fail_in_use(image_name); 2838 } 2839 2840 assert(QEMU_IS_ALIGNED(guest_base, align)); 2841 qemu_log_mask(CPU_LOG_PAGE, "Locating guest address space " 2842 "@ 0x%" PRIx64 "\n", (uint64_t)guest_base); 2843 } 2844 2845 enum { 2846 /* The string "GNU\0" as a magic number. */ 2847 GNU0_MAGIC = const_le32('G' | 'N' << 8 | 'U' << 16), 2848 NOTE_DATA_SZ = 1 * KiB, 2849 NOTE_NAME_SZ = 4, 2850 ELF_GNU_PROPERTY_ALIGN = ELF_CLASS == ELFCLASS32 ? 4 : 8, 2851 }; 2852 2853 /* 2854 * Process a single gnu_property entry. 2855 * Return false for error. 2856 */ 2857 static bool parse_elf_property(const uint32_t *data, int *off, int datasz, 2858 struct image_info *info, bool have_prev_type, 2859 uint32_t *prev_type, Error **errp) 2860 { 2861 uint32_t pr_type, pr_datasz, step; 2862 2863 if (*off > datasz || !QEMU_IS_ALIGNED(*off, ELF_GNU_PROPERTY_ALIGN)) { 2864 goto error_data; 2865 } 2866 datasz -= *off; 2867 data += *off / sizeof(uint32_t); 2868 2869 if (datasz < 2 * sizeof(uint32_t)) { 2870 goto error_data; 2871 } 2872 pr_type = data[0]; 2873 pr_datasz = data[1]; 2874 data += 2; 2875 datasz -= 2 * sizeof(uint32_t); 2876 step = ROUND_UP(pr_datasz, ELF_GNU_PROPERTY_ALIGN); 2877 if (step > datasz) { 2878 goto error_data; 2879 } 2880 2881 /* Properties are supposed to be unique and sorted on pr_type. */ 2882 if (have_prev_type && pr_type <= *prev_type) { 2883 if (pr_type == *prev_type) { 2884 error_setg(errp, "Duplicate property in PT_GNU_PROPERTY"); 2885 } else { 2886 error_setg(errp, "Unsorted property in PT_GNU_PROPERTY"); 2887 } 2888 return false; 2889 } 2890 *prev_type = pr_type; 2891 2892 if (!arch_parse_elf_property(pr_type, pr_datasz, data, info, errp)) { 2893 return false; 2894 } 2895 2896 *off += 2 * sizeof(uint32_t) + step; 2897 return true; 2898 2899 error_data: 2900 error_setg(errp, "Ill-formed property in PT_GNU_PROPERTY"); 2901 return false; 2902 } 2903 2904 /* Process NT_GNU_PROPERTY_TYPE_0. */ 2905 static bool parse_elf_properties(int image_fd, 2906 struct image_info *info, 2907 const struct elf_phdr *phdr, 2908 char bprm_buf[BPRM_BUF_SIZE], 2909 Error **errp) 2910 { 2911 union { 2912 struct elf_note nhdr; 2913 uint32_t data[NOTE_DATA_SZ / sizeof(uint32_t)]; 2914 } note; 2915 2916 int n, off, datasz; 2917 bool have_prev_type; 2918 uint32_t prev_type; 2919 2920 /* Unless the arch requires properties, ignore them. */ 2921 if (!ARCH_USE_GNU_PROPERTY) { 2922 return true; 2923 } 2924 2925 /* If the properties are crazy large, that's too bad. */ 2926 n = phdr->p_filesz; 2927 if (n > sizeof(note)) { 2928 error_setg(errp, "PT_GNU_PROPERTY too large"); 2929 return false; 2930 } 2931 if (n < sizeof(note.nhdr)) { 2932 error_setg(errp, "PT_GNU_PROPERTY too small"); 2933 return false; 2934 } 2935 2936 if (phdr->p_offset + n <= BPRM_BUF_SIZE) { 2937 memcpy(¬e, bprm_buf + phdr->p_offset, n); 2938 } else { 2939 ssize_t len = pread(image_fd, ¬e, n, phdr->p_offset); 2940 if (len != n) { 2941 error_setg_errno(errp, errno, "Error reading file header"); 2942 return false; 2943 } 2944 } 2945 2946 /* 2947 * The contents of a valid PT_GNU_PROPERTY is a sequence 2948 * of uint32_t -- swap them all now. 2949 */ 2950 #ifdef BSWAP_NEEDED 2951 for (int i = 0; i < n / 4; i++) { 2952 bswap32s(note.data + i); 2953 } 2954 #endif 2955 2956 /* 2957 * Note that nhdr is 3 words, and that the "name" described by namesz 2958 * immediately follows nhdr and is thus at the 4th word. Further, all 2959 * of the inputs to the kernel's round_up are multiples of 4. 2960 */ 2961 if (note.nhdr.n_type != NT_GNU_PROPERTY_TYPE_0 || 2962 note.nhdr.n_namesz != NOTE_NAME_SZ || 2963 note.data[3] != GNU0_MAGIC) { 2964 error_setg(errp, "Invalid note in PT_GNU_PROPERTY"); 2965 return false; 2966 } 2967 off = sizeof(note.nhdr) + NOTE_NAME_SZ; 2968 2969 datasz = note.nhdr.n_descsz + off; 2970 if (datasz > n) { 2971 error_setg(errp, "Invalid note size in PT_GNU_PROPERTY"); 2972 return false; 2973 } 2974 2975 have_prev_type = false; 2976 prev_type = 0; 2977 while (1) { 2978 if (off == datasz) { 2979 return true; /* end, exit ok */ 2980 } 2981 if (!parse_elf_property(note.data, &off, datasz, info, 2982 have_prev_type, &prev_type, errp)) { 2983 return false; 2984 } 2985 have_prev_type = true; 2986 } 2987 } 2988 2989 /* Load an ELF image into the address space. 2990 2991 IMAGE_NAME is the filename of the image, to use in error messages. 2992 IMAGE_FD is the open file descriptor for the image. 2993 2994 BPRM_BUF is a copy of the beginning of the file; this of course 2995 contains the elf file header at offset 0. It is assumed that this 2996 buffer is sufficiently aligned to present no problems to the host 2997 in accessing data at aligned offsets within the buffer. 2998 2999 On return: INFO values will be filled in, as necessary or available. */ 3000 3001 static void load_elf_image(const char *image_name, int image_fd, 3002 struct image_info *info, char **pinterp_name, 3003 char bprm_buf[BPRM_BUF_SIZE]) 3004 { 3005 struct elfhdr *ehdr = (struct elfhdr *)bprm_buf; 3006 struct elf_phdr *phdr; 3007 abi_ulong load_addr, load_bias, loaddr, hiaddr, error; 3008 int i, retval, prot_exec; 3009 Error *err = NULL; 3010 3011 /* First of all, some simple consistency checks */ 3012 if (!elf_check_ident(ehdr)) { 3013 error_setg(&err, "Invalid ELF image for this architecture"); 3014 goto exit_errmsg; 3015 } 3016 bswap_ehdr(ehdr); 3017 if (!elf_check_ehdr(ehdr)) { 3018 error_setg(&err, "Invalid ELF image for this architecture"); 3019 goto exit_errmsg; 3020 } 3021 3022 i = ehdr->e_phnum * sizeof(struct elf_phdr); 3023 if (ehdr->e_phoff + i <= BPRM_BUF_SIZE) { 3024 phdr = (struct elf_phdr *)(bprm_buf + ehdr->e_phoff); 3025 } else { 3026 phdr = (struct elf_phdr *) alloca(i); 3027 retval = pread(image_fd, phdr, i, ehdr->e_phoff); 3028 if (retval != i) { 3029 goto exit_read; 3030 } 3031 } 3032 bswap_phdr(phdr, ehdr->e_phnum); 3033 3034 info->nsegs = 0; 3035 info->pt_dynamic_addr = 0; 3036 3037 mmap_lock(); 3038 3039 /* 3040 * Find the maximum size of the image and allocate an appropriate 3041 * amount of memory to handle that. Locate the interpreter, if any. 3042 */ 3043 loaddr = -1, hiaddr = 0; 3044 info->alignment = 0; 3045 info->exec_stack = EXSTACK_DEFAULT; 3046 for (i = 0; i < ehdr->e_phnum; ++i) { 3047 struct elf_phdr *eppnt = phdr + i; 3048 if (eppnt->p_type == PT_LOAD) { 3049 abi_ulong a = eppnt->p_vaddr - eppnt->p_offset; 3050 if (a < loaddr) { 3051 loaddr = a; 3052 } 3053 a = eppnt->p_vaddr + eppnt->p_memsz - 1; 3054 if (a > hiaddr) { 3055 hiaddr = a; 3056 } 3057 ++info->nsegs; 3058 info->alignment |= eppnt->p_align; 3059 } else if (eppnt->p_type == PT_INTERP && pinterp_name) { 3060 g_autofree char *interp_name = NULL; 3061 3062 if (*pinterp_name) { 3063 error_setg(&err, "Multiple PT_INTERP entries"); 3064 goto exit_errmsg; 3065 } 3066 3067 interp_name = g_malloc(eppnt->p_filesz); 3068 3069 if (eppnt->p_offset + eppnt->p_filesz <= BPRM_BUF_SIZE) { 3070 memcpy(interp_name, bprm_buf + eppnt->p_offset, 3071 eppnt->p_filesz); 3072 } else { 3073 retval = pread(image_fd, interp_name, eppnt->p_filesz, 3074 eppnt->p_offset); 3075 if (retval != eppnt->p_filesz) { 3076 goto exit_read; 3077 } 3078 } 3079 if (interp_name[eppnt->p_filesz - 1] != 0) { 3080 error_setg(&err, "Invalid PT_INTERP entry"); 3081 goto exit_errmsg; 3082 } 3083 *pinterp_name = g_steal_pointer(&interp_name); 3084 } else if (eppnt->p_type == PT_GNU_PROPERTY) { 3085 if (!parse_elf_properties(image_fd, info, eppnt, bprm_buf, &err)) { 3086 goto exit_errmsg; 3087 } 3088 } else if (eppnt->p_type == PT_GNU_STACK) { 3089 info->exec_stack = eppnt->p_flags & PF_X; 3090 } 3091 } 3092 3093 if (pinterp_name != NULL) { 3094 /* 3095 * This is the main executable. 3096 * 3097 * Reserve extra space for brk. 3098 * We hold on to this space while placing the interpreter 3099 * and the stack, lest they be placed immediately after 3100 * the data segment and block allocation from the brk. 3101 * 3102 * 16MB is chosen as "large enough" without being so large as 3103 * to allow the result to not fit with a 32-bit guest on a 3104 * 32-bit host. However some 64 bit guests (e.g. s390x) 3105 * attempt to place their heap further ahead and currently 3106 * nothing stops them smashing into QEMUs address space. 3107 */ 3108 #if TARGET_LONG_BITS == 64 3109 info->reserve_brk = 32 * MiB; 3110 #else 3111 info->reserve_brk = 16 * MiB; 3112 #endif 3113 hiaddr += info->reserve_brk; 3114 3115 if (ehdr->e_type == ET_EXEC) { 3116 /* 3117 * Make sure that the low address does not conflict with 3118 * MMAP_MIN_ADDR or the QEMU application itself. 3119 */ 3120 probe_guest_base(image_name, loaddr, hiaddr); 3121 } else { 3122 /* 3123 * The binary is dynamic, but we still need to 3124 * select guest_base. In this case we pass a size. 3125 */ 3126 probe_guest_base(image_name, 0, hiaddr - loaddr); 3127 } 3128 } 3129 3130 /* 3131 * Reserve address space for all of this. 3132 * 3133 * In the case of ET_EXEC, we supply MAP_FIXED so that we get 3134 * exactly the address range that is required. 3135 * 3136 * Otherwise this is ET_DYN, and we are searching for a location 3137 * that can hold the memory space required. If the image is 3138 * pre-linked, LOADDR will be non-zero, and the kernel should 3139 * honor that address if it happens to be free. 3140 * 3141 * In both cases, we will overwrite pages in this range with mappings 3142 * from the executable. 3143 */ 3144 load_addr = target_mmap(loaddr, (size_t)hiaddr - loaddr + 1, PROT_NONE, 3145 MAP_PRIVATE | MAP_ANON | MAP_NORESERVE | 3146 (ehdr->e_type == ET_EXEC ? MAP_FIXED : 0), 3147 -1, 0); 3148 if (load_addr == -1) { 3149 goto exit_mmap; 3150 } 3151 load_bias = load_addr - loaddr; 3152 3153 if (elf_is_fdpic(ehdr)) { 3154 struct elf32_fdpic_loadseg *loadsegs = info->loadsegs = 3155 g_malloc(sizeof(*loadsegs) * info->nsegs); 3156 3157 for (i = 0; i < ehdr->e_phnum; ++i) { 3158 switch (phdr[i].p_type) { 3159 case PT_DYNAMIC: 3160 info->pt_dynamic_addr = phdr[i].p_vaddr + load_bias; 3161 break; 3162 case PT_LOAD: 3163 loadsegs->addr = phdr[i].p_vaddr + load_bias; 3164 loadsegs->p_vaddr = phdr[i].p_vaddr; 3165 loadsegs->p_memsz = phdr[i].p_memsz; 3166 ++loadsegs; 3167 break; 3168 } 3169 } 3170 } 3171 3172 info->load_bias = load_bias; 3173 info->code_offset = load_bias; 3174 info->data_offset = load_bias; 3175 info->load_addr = load_addr; 3176 info->entry = ehdr->e_entry + load_bias; 3177 info->start_code = -1; 3178 info->end_code = 0; 3179 info->start_data = -1; 3180 info->end_data = 0; 3181 info->brk = 0; 3182 info->elf_flags = ehdr->e_flags; 3183 3184 prot_exec = PROT_EXEC; 3185 #ifdef TARGET_AARCH64 3186 /* 3187 * If the BTI feature is present, this indicates that the executable 3188 * pages of the startup binary should be mapped with PROT_BTI, so that 3189 * branch targets are enforced. 3190 * 3191 * The startup binary is either the interpreter or the static executable. 3192 * The interpreter is responsible for all pages of a dynamic executable. 3193 * 3194 * Elf notes are backward compatible to older cpus. 3195 * Do not enable BTI unless it is supported. 3196 */ 3197 if ((info->note_flags & GNU_PROPERTY_AARCH64_FEATURE_1_BTI) 3198 && (pinterp_name == NULL || *pinterp_name == 0) 3199 && cpu_isar_feature(aa64_bti, ARM_CPU(thread_cpu))) { 3200 prot_exec |= TARGET_PROT_BTI; 3201 } 3202 #endif 3203 3204 for (i = 0; i < ehdr->e_phnum; i++) { 3205 struct elf_phdr *eppnt = phdr + i; 3206 if (eppnt->p_type == PT_LOAD) { 3207 abi_ulong vaddr, vaddr_po, vaddr_ps, vaddr_ef, vaddr_em, vaddr_len; 3208 int elf_prot = 0; 3209 3210 if (eppnt->p_flags & PF_R) { 3211 elf_prot |= PROT_READ; 3212 } 3213 if (eppnt->p_flags & PF_W) { 3214 elf_prot |= PROT_WRITE; 3215 } 3216 if (eppnt->p_flags & PF_X) { 3217 elf_prot |= prot_exec; 3218 } 3219 3220 vaddr = load_bias + eppnt->p_vaddr; 3221 vaddr_po = TARGET_ELF_PAGEOFFSET(vaddr); 3222 vaddr_ps = TARGET_ELF_PAGESTART(vaddr); 3223 3224 vaddr_ef = vaddr + eppnt->p_filesz; 3225 vaddr_em = vaddr + eppnt->p_memsz; 3226 3227 /* 3228 * Some segments may be completely empty, with a non-zero p_memsz 3229 * but no backing file segment. 3230 */ 3231 if (eppnt->p_filesz != 0) { 3232 vaddr_len = TARGET_ELF_PAGELENGTH(eppnt->p_filesz + vaddr_po); 3233 error = target_mmap(vaddr_ps, vaddr_len, elf_prot, 3234 MAP_PRIVATE | MAP_FIXED, 3235 image_fd, eppnt->p_offset - vaddr_po); 3236 3237 if (error == -1) { 3238 goto exit_mmap; 3239 } 3240 3241 /* 3242 * If the load segment requests extra zeros (e.g. bss), map it. 3243 */ 3244 if (eppnt->p_filesz < eppnt->p_memsz) { 3245 zero_bss(vaddr_ef, vaddr_em, elf_prot); 3246 } 3247 } else if (eppnt->p_memsz != 0) { 3248 vaddr_len = TARGET_ELF_PAGELENGTH(eppnt->p_memsz + vaddr_po); 3249 error = target_mmap(vaddr_ps, vaddr_len, elf_prot, 3250 MAP_PRIVATE | MAP_FIXED | MAP_ANONYMOUS, 3251 -1, 0); 3252 3253 if (error == -1) { 3254 goto exit_mmap; 3255 } 3256 } 3257 3258 /* Find the full program boundaries. */ 3259 if (elf_prot & PROT_EXEC) { 3260 if (vaddr < info->start_code) { 3261 info->start_code = vaddr; 3262 } 3263 if (vaddr_ef > info->end_code) { 3264 info->end_code = vaddr_ef; 3265 } 3266 } 3267 if (elf_prot & PROT_WRITE) { 3268 if (vaddr < info->start_data) { 3269 info->start_data = vaddr; 3270 } 3271 if (vaddr_ef > info->end_data) { 3272 info->end_data = vaddr_ef; 3273 } 3274 } 3275 if (vaddr_em > info->brk) { 3276 info->brk = vaddr_em; 3277 } 3278 #ifdef TARGET_MIPS 3279 } else if (eppnt->p_type == PT_MIPS_ABIFLAGS) { 3280 Mips_elf_abiflags_v0 abiflags; 3281 if (eppnt->p_filesz < sizeof(Mips_elf_abiflags_v0)) { 3282 error_setg(&err, "Invalid PT_MIPS_ABIFLAGS entry"); 3283 goto exit_errmsg; 3284 } 3285 if (eppnt->p_offset + eppnt->p_filesz <= BPRM_BUF_SIZE) { 3286 memcpy(&abiflags, bprm_buf + eppnt->p_offset, 3287 sizeof(Mips_elf_abiflags_v0)); 3288 } else { 3289 retval = pread(image_fd, &abiflags, sizeof(Mips_elf_abiflags_v0), 3290 eppnt->p_offset); 3291 if (retval != sizeof(Mips_elf_abiflags_v0)) { 3292 goto exit_read; 3293 } 3294 } 3295 bswap_mips_abiflags(&abiflags); 3296 info->fp_abi = abiflags.fp_abi; 3297 #endif 3298 } 3299 } 3300 3301 if (info->end_data == 0) { 3302 info->start_data = info->end_code; 3303 info->end_data = info->end_code; 3304 } 3305 3306 if (qemu_log_enabled()) { 3307 load_symbols(ehdr, image_fd, load_bias); 3308 } 3309 3310 debuginfo_report_elf(image_name, image_fd, load_bias); 3311 3312 mmap_unlock(); 3313 3314 close(image_fd); 3315 return; 3316 3317 exit_read: 3318 if (retval >= 0) { 3319 error_setg(&err, "Incomplete read of file header"); 3320 } else { 3321 error_setg_errno(&err, errno, "Error reading file header"); 3322 } 3323 goto exit_errmsg; 3324 exit_mmap: 3325 error_setg_errno(&err, errno, "Error mapping file"); 3326 goto exit_errmsg; 3327 exit_errmsg: 3328 error_reportf_err(err, "%s: ", image_name); 3329 exit(-1); 3330 } 3331 3332 static void load_elf_interp(const char *filename, struct image_info *info, 3333 char bprm_buf[BPRM_BUF_SIZE]) 3334 { 3335 int fd, retval; 3336 Error *err = NULL; 3337 3338 fd = open(path(filename), O_RDONLY); 3339 if (fd < 0) { 3340 error_setg_file_open(&err, errno, filename); 3341 error_report_err(err); 3342 exit(-1); 3343 } 3344 3345 retval = read(fd, bprm_buf, BPRM_BUF_SIZE); 3346 if (retval < 0) { 3347 error_setg_errno(&err, errno, "Error reading file header"); 3348 error_reportf_err(err, "%s: ", filename); 3349 exit(-1); 3350 } 3351 3352 if (retval < BPRM_BUF_SIZE) { 3353 memset(bprm_buf + retval, 0, BPRM_BUF_SIZE - retval); 3354 } 3355 3356 load_elf_image(filename, fd, info, NULL, bprm_buf); 3357 } 3358 3359 static int symfind(const void *s0, const void *s1) 3360 { 3361 struct elf_sym *sym = (struct elf_sym *)s1; 3362 __typeof(sym->st_value) addr = *(uint64_t *)s0; 3363 int result = 0; 3364 3365 if (addr < sym->st_value) { 3366 result = -1; 3367 } else if (addr >= sym->st_value + sym->st_size) { 3368 result = 1; 3369 } 3370 return result; 3371 } 3372 3373 static const char *lookup_symbolxx(struct syminfo *s, uint64_t orig_addr) 3374 { 3375 #if ELF_CLASS == ELFCLASS32 3376 struct elf_sym *syms = s->disas_symtab.elf32; 3377 #else 3378 struct elf_sym *syms = s->disas_symtab.elf64; 3379 #endif 3380 3381 // binary search 3382 struct elf_sym *sym; 3383 3384 sym = bsearch(&orig_addr, syms, s->disas_num_syms, sizeof(*syms), symfind); 3385 if (sym != NULL) { 3386 return s->disas_strtab + sym->st_name; 3387 } 3388 3389 return ""; 3390 } 3391 3392 /* FIXME: This should use elf_ops.h */ 3393 static int symcmp(const void *s0, const void *s1) 3394 { 3395 struct elf_sym *sym0 = (struct elf_sym *)s0; 3396 struct elf_sym *sym1 = (struct elf_sym *)s1; 3397 return (sym0->st_value < sym1->st_value) 3398 ? -1 3399 : ((sym0->st_value > sym1->st_value) ? 1 : 0); 3400 } 3401 3402 /* Best attempt to load symbols from this ELF object. */ 3403 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias) 3404 { 3405 int i, shnum, nsyms, sym_idx = 0, str_idx = 0; 3406 uint64_t segsz; 3407 struct elf_shdr *shdr; 3408 char *strings = NULL; 3409 struct syminfo *s = NULL; 3410 struct elf_sym *new_syms, *syms = NULL; 3411 3412 shnum = hdr->e_shnum; 3413 i = shnum * sizeof(struct elf_shdr); 3414 shdr = (struct elf_shdr *)alloca(i); 3415 if (pread(fd, shdr, i, hdr->e_shoff) != i) { 3416 return; 3417 } 3418 3419 bswap_shdr(shdr, shnum); 3420 for (i = 0; i < shnum; ++i) { 3421 if (shdr[i].sh_type == SHT_SYMTAB) { 3422 sym_idx = i; 3423 str_idx = shdr[i].sh_link; 3424 goto found; 3425 } 3426 } 3427 3428 /* There will be no symbol table if the file was stripped. */ 3429 return; 3430 3431 found: 3432 /* Now know where the strtab and symtab are. Snarf them. */ 3433 s = g_try_new(struct syminfo, 1); 3434 if (!s) { 3435 goto give_up; 3436 } 3437 3438 segsz = shdr[str_idx].sh_size; 3439 s->disas_strtab = strings = g_try_malloc(segsz); 3440 if (!strings || 3441 pread(fd, strings, segsz, shdr[str_idx].sh_offset) != segsz) { 3442 goto give_up; 3443 } 3444 3445 segsz = shdr[sym_idx].sh_size; 3446 syms = g_try_malloc(segsz); 3447 if (!syms || pread(fd, syms, segsz, shdr[sym_idx].sh_offset) != segsz) { 3448 goto give_up; 3449 } 3450 3451 if (segsz / sizeof(struct elf_sym) > INT_MAX) { 3452 /* Implausibly large symbol table: give up rather than ploughing 3453 * on with the number of symbols calculation overflowing 3454 */ 3455 goto give_up; 3456 } 3457 nsyms = segsz / sizeof(struct elf_sym); 3458 for (i = 0; i < nsyms; ) { 3459 bswap_sym(syms + i); 3460 /* Throw away entries which we do not need. */ 3461 if (syms[i].st_shndx == SHN_UNDEF 3462 || syms[i].st_shndx >= SHN_LORESERVE 3463 || ELF_ST_TYPE(syms[i].st_info) != STT_FUNC) { 3464 if (i < --nsyms) { 3465 syms[i] = syms[nsyms]; 3466 } 3467 } else { 3468 #if defined(TARGET_ARM) || defined (TARGET_MIPS) 3469 /* The bottom address bit marks a Thumb or MIPS16 symbol. */ 3470 syms[i].st_value &= ~(target_ulong)1; 3471 #endif 3472 syms[i].st_value += load_bias; 3473 i++; 3474 } 3475 } 3476 3477 /* No "useful" symbol. */ 3478 if (nsyms == 0) { 3479 goto give_up; 3480 } 3481 3482 /* Attempt to free the storage associated with the local symbols 3483 that we threw away. Whether or not this has any effect on the 3484 memory allocation depends on the malloc implementation and how 3485 many symbols we managed to discard. */ 3486 new_syms = g_try_renew(struct elf_sym, syms, nsyms); 3487 if (new_syms == NULL) { 3488 goto give_up; 3489 } 3490 syms = new_syms; 3491 3492 qsort(syms, nsyms, sizeof(*syms), symcmp); 3493 3494 s->disas_num_syms = nsyms; 3495 #if ELF_CLASS == ELFCLASS32 3496 s->disas_symtab.elf32 = syms; 3497 #else 3498 s->disas_symtab.elf64 = syms; 3499 #endif 3500 s->lookup_symbol = lookup_symbolxx; 3501 s->next = syminfos; 3502 syminfos = s; 3503 3504 return; 3505 3506 give_up: 3507 g_free(s); 3508 g_free(strings); 3509 g_free(syms); 3510 } 3511 3512 uint32_t get_elf_eflags(int fd) 3513 { 3514 struct elfhdr ehdr; 3515 off_t offset; 3516 int ret; 3517 3518 /* Read ELF header */ 3519 offset = lseek(fd, 0, SEEK_SET); 3520 if (offset == (off_t) -1) { 3521 return 0; 3522 } 3523 ret = read(fd, &ehdr, sizeof(ehdr)); 3524 if (ret < sizeof(ehdr)) { 3525 return 0; 3526 } 3527 offset = lseek(fd, offset, SEEK_SET); 3528 if (offset == (off_t) -1) { 3529 return 0; 3530 } 3531 3532 /* Check ELF signature */ 3533 if (!elf_check_ident(&ehdr)) { 3534 return 0; 3535 } 3536 3537 /* check header */ 3538 bswap_ehdr(&ehdr); 3539 if (!elf_check_ehdr(&ehdr)) { 3540 return 0; 3541 } 3542 3543 /* return architecture id */ 3544 return ehdr.e_flags; 3545 } 3546 3547 int load_elf_binary(struct linux_binprm *bprm, struct image_info *info) 3548 { 3549 struct image_info interp_info; 3550 struct elfhdr elf_ex; 3551 char *elf_interpreter = NULL; 3552 char *scratch; 3553 3554 memset(&interp_info, 0, sizeof(interp_info)); 3555 #ifdef TARGET_MIPS 3556 interp_info.fp_abi = MIPS_ABI_FP_UNKNOWN; 3557 #endif 3558 3559 info->start_mmap = (abi_ulong)ELF_START_MMAP; 3560 3561 load_elf_image(bprm->filename, bprm->fd, info, 3562 &elf_interpreter, bprm->buf); 3563 3564 /* ??? We need a copy of the elf header for passing to create_elf_tables. 3565 If we do nothing, we'll have overwritten this when we re-use bprm->buf 3566 when we load the interpreter. */ 3567 elf_ex = *(struct elfhdr *)bprm->buf; 3568 3569 /* Do this so that we can load the interpreter, if need be. We will 3570 change some of these later */ 3571 bprm->p = setup_arg_pages(bprm, info); 3572 3573 scratch = g_new0(char, TARGET_PAGE_SIZE); 3574 if (STACK_GROWS_DOWN) { 3575 bprm->p = copy_elf_strings(1, &bprm->filename, scratch, 3576 bprm->p, info->stack_limit); 3577 info->file_string = bprm->p; 3578 bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch, 3579 bprm->p, info->stack_limit); 3580 info->env_strings = bprm->p; 3581 bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch, 3582 bprm->p, info->stack_limit); 3583 info->arg_strings = bprm->p; 3584 } else { 3585 info->arg_strings = bprm->p; 3586 bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch, 3587 bprm->p, info->stack_limit); 3588 info->env_strings = bprm->p; 3589 bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch, 3590 bprm->p, info->stack_limit); 3591 info->file_string = bprm->p; 3592 bprm->p = copy_elf_strings(1, &bprm->filename, scratch, 3593 bprm->p, info->stack_limit); 3594 } 3595 3596 g_free(scratch); 3597 3598 if (!bprm->p) { 3599 fprintf(stderr, "%s: %s\n", bprm->filename, strerror(E2BIG)); 3600 exit(-1); 3601 } 3602 3603 if (elf_interpreter) { 3604 load_elf_interp(elf_interpreter, &interp_info, bprm->buf); 3605 3606 /* If the program interpreter is one of these two, then assume 3607 an iBCS2 image. Otherwise assume a native linux image. */ 3608 3609 if (strcmp(elf_interpreter, "/usr/lib/libc.so.1") == 0 3610 || strcmp(elf_interpreter, "/usr/lib/ld.so.1") == 0) { 3611 info->personality = PER_SVR4; 3612 3613 /* Why this, you ask??? Well SVr4 maps page 0 as read-only, 3614 and some applications "depend" upon this behavior. Since 3615 we do not have the power to recompile these, we emulate 3616 the SVr4 behavior. Sigh. */ 3617 target_mmap(0, qemu_host_page_size, PROT_READ | PROT_EXEC, 3618 MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); 3619 } 3620 #ifdef TARGET_MIPS 3621 info->interp_fp_abi = interp_info.fp_abi; 3622 #endif 3623 } 3624 3625 /* 3626 * TODO: load a vdso, which would also contain the signal trampolines. 3627 * Otherwise, allocate a private page to hold them. 3628 */ 3629 if (TARGET_ARCH_HAS_SIGTRAMP_PAGE) { 3630 abi_long tramp_page = target_mmap(0, TARGET_PAGE_SIZE, 3631 PROT_READ | PROT_WRITE, 3632 MAP_PRIVATE | MAP_ANON, -1, 0); 3633 if (tramp_page == -1) { 3634 return -errno; 3635 } 3636 3637 setup_sigtramp(tramp_page); 3638 target_mprotect(tramp_page, TARGET_PAGE_SIZE, PROT_READ | PROT_EXEC); 3639 } 3640 3641 bprm->p = create_elf_tables(bprm->p, bprm->argc, bprm->envc, &elf_ex, 3642 info, (elf_interpreter ? &interp_info : NULL)); 3643 info->start_stack = bprm->p; 3644 3645 /* If we have an interpreter, set that as the program's entry point. 3646 Copy the load_bias as well, to help PPC64 interpret the entry 3647 point as a function descriptor. Do this after creating elf tables 3648 so that we copy the original program entry point into the AUXV. */ 3649 if (elf_interpreter) { 3650 info->load_bias = interp_info.load_bias; 3651 info->entry = interp_info.entry; 3652 g_free(elf_interpreter); 3653 } 3654 3655 #ifdef USE_ELF_CORE_DUMP 3656 bprm->core_dump = &elf_core_dump; 3657 #endif 3658 3659 /* 3660 * If we reserved extra space for brk, release it now. 3661 * The implementation of do_brk in syscalls.c expects to be able 3662 * to mmap pages in this space. 3663 */ 3664 if (info->reserve_brk) { 3665 abi_ulong start_brk = HOST_PAGE_ALIGN(info->brk); 3666 abi_ulong end_brk = HOST_PAGE_ALIGN(info->brk + info->reserve_brk); 3667 target_munmap(start_brk, end_brk - start_brk); 3668 } 3669 3670 return 0; 3671 } 3672 3673 #ifdef USE_ELF_CORE_DUMP 3674 /* 3675 * Definitions to generate Intel SVR4-like core files. 3676 * These mostly have the same names as the SVR4 types with "target_elf_" 3677 * tacked on the front to prevent clashes with linux definitions, 3678 * and the typedef forms have been avoided. This is mostly like 3679 * the SVR4 structure, but more Linuxy, with things that Linux does 3680 * not support and which gdb doesn't really use excluded. 3681 * 3682 * Fields we don't dump (their contents is zero) in linux-user qemu 3683 * are marked with XXX. 3684 * 3685 * Core dump code is copied from linux kernel (fs/binfmt_elf.c). 3686 * 3687 * Porting ELF coredump for target is (quite) simple process. First you 3688 * define USE_ELF_CORE_DUMP in target ELF code (where init_thread() for 3689 * the target resides): 3690 * 3691 * #define USE_ELF_CORE_DUMP 3692 * 3693 * Next you define type of register set used for dumping. ELF specification 3694 * says that it needs to be array of elf_greg_t that has size of ELF_NREG. 3695 * 3696 * typedef <target_regtype> target_elf_greg_t; 3697 * #define ELF_NREG <number of registers> 3698 * typedef taret_elf_greg_t target_elf_gregset_t[ELF_NREG]; 3699 * 3700 * Last step is to implement target specific function that copies registers 3701 * from given cpu into just specified register set. Prototype is: 3702 * 3703 * static void elf_core_copy_regs(taret_elf_gregset_t *regs, 3704 * const CPUArchState *env); 3705 * 3706 * Parameters: 3707 * regs - copy register values into here (allocated and zeroed by caller) 3708 * env - copy registers from here 3709 * 3710 * Example for ARM target is provided in this file. 3711 */ 3712 3713 /* An ELF note in memory */ 3714 struct memelfnote { 3715 const char *name; 3716 size_t namesz; 3717 size_t namesz_rounded; 3718 int type; 3719 size_t datasz; 3720 size_t datasz_rounded; 3721 void *data; 3722 size_t notesz; 3723 }; 3724 3725 struct target_elf_siginfo { 3726 abi_int si_signo; /* signal number */ 3727 abi_int si_code; /* extra code */ 3728 abi_int si_errno; /* errno */ 3729 }; 3730 3731 struct target_elf_prstatus { 3732 struct target_elf_siginfo pr_info; /* Info associated with signal */ 3733 abi_short pr_cursig; /* Current signal */ 3734 abi_ulong pr_sigpend; /* XXX */ 3735 abi_ulong pr_sighold; /* XXX */ 3736 target_pid_t pr_pid; 3737 target_pid_t pr_ppid; 3738 target_pid_t pr_pgrp; 3739 target_pid_t pr_sid; 3740 struct target_timeval pr_utime; /* XXX User time */ 3741 struct target_timeval pr_stime; /* XXX System time */ 3742 struct target_timeval pr_cutime; /* XXX Cumulative user time */ 3743 struct target_timeval pr_cstime; /* XXX Cumulative system time */ 3744 target_elf_gregset_t pr_reg; /* GP registers */ 3745 abi_int pr_fpvalid; /* XXX */ 3746 }; 3747 3748 #define ELF_PRARGSZ (80) /* Number of chars for args */ 3749 3750 struct target_elf_prpsinfo { 3751 char pr_state; /* numeric process state */ 3752 char pr_sname; /* char for pr_state */ 3753 char pr_zomb; /* zombie */ 3754 char pr_nice; /* nice val */ 3755 abi_ulong pr_flag; /* flags */ 3756 target_uid_t pr_uid; 3757 target_gid_t pr_gid; 3758 target_pid_t pr_pid, pr_ppid, pr_pgrp, pr_sid; 3759 /* Lots missing */ 3760 char pr_fname[16] QEMU_NONSTRING; /* filename of executable */ 3761 char pr_psargs[ELF_PRARGSZ]; /* initial part of arg list */ 3762 }; 3763 3764 /* Here is the structure in which status of each thread is captured. */ 3765 struct elf_thread_status { 3766 QTAILQ_ENTRY(elf_thread_status) ets_link; 3767 struct target_elf_prstatus prstatus; /* NT_PRSTATUS */ 3768 #if 0 3769 elf_fpregset_t fpu; /* NT_PRFPREG */ 3770 struct task_struct *thread; 3771 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */ 3772 #endif 3773 struct memelfnote notes[1]; 3774 int num_notes; 3775 }; 3776 3777 struct elf_note_info { 3778 struct memelfnote *notes; 3779 struct target_elf_prstatus *prstatus; /* NT_PRSTATUS */ 3780 struct target_elf_prpsinfo *psinfo; /* NT_PRPSINFO */ 3781 3782 QTAILQ_HEAD(, elf_thread_status) thread_list; 3783 #if 0 3784 /* 3785 * Current version of ELF coredump doesn't support 3786 * dumping fp regs etc. 3787 */ 3788 elf_fpregset_t *fpu; 3789 elf_fpxregset_t *xfpu; 3790 int thread_status_size; 3791 #endif 3792 int notes_size; 3793 int numnote; 3794 }; 3795 3796 struct vm_area_struct { 3797 target_ulong vma_start; /* start vaddr of memory region */ 3798 target_ulong vma_end; /* end vaddr of memory region */ 3799 abi_ulong vma_flags; /* protection etc. flags for the region */ 3800 QTAILQ_ENTRY(vm_area_struct) vma_link; 3801 }; 3802 3803 struct mm_struct { 3804 QTAILQ_HEAD(, vm_area_struct) mm_mmap; 3805 int mm_count; /* number of mappings */ 3806 }; 3807 3808 static struct mm_struct *vma_init(void); 3809 static void vma_delete(struct mm_struct *); 3810 static int vma_add_mapping(struct mm_struct *, target_ulong, 3811 target_ulong, abi_ulong); 3812 static int vma_get_mapping_count(const struct mm_struct *); 3813 static struct vm_area_struct *vma_first(const struct mm_struct *); 3814 static struct vm_area_struct *vma_next(struct vm_area_struct *); 3815 static abi_ulong vma_dump_size(const struct vm_area_struct *); 3816 static int vma_walker(void *priv, target_ulong start, target_ulong end, 3817 unsigned long flags); 3818 3819 static void fill_elf_header(struct elfhdr *, int, uint16_t, uint32_t); 3820 static void fill_note(struct memelfnote *, const char *, int, 3821 unsigned int, void *); 3822 static void fill_prstatus(struct target_elf_prstatus *, const TaskState *, int); 3823 static int fill_psinfo(struct target_elf_prpsinfo *, const TaskState *); 3824 static void fill_auxv_note(struct memelfnote *, const TaskState *); 3825 static void fill_elf_note_phdr(struct elf_phdr *, int, off_t); 3826 static size_t note_size(const struct memelfnote *); 3827 static void free_note_info(struct elf_note_info *); 3828 static int fill_note_info(struct elf_note_info *, long, const CPUArchState *); 3829 static void fill_thread_info(struct elf_note_info *, const CPUArchState *); 3830 3831 static int dump_write(int, const void *, size_t); 3832 static int write_note(struct memelfnote *, int); 3833 static int write_note_info(struct elf_note_info *, int); 3834 3835 #ifdef BSWAP_NEEDED 3836 static void bswap_prstatus(struct target_elf_prstatus *prstatus) 3837 { 3838 prstatus->pr_info.si_signo = tswap32(prstatus->pr_info.si_signo); 3839 prstatus->pr_info.si_code = tswap32(prstatus->pr_info.si_code); 3840 prstatus->pr_info.si_errno = tswap32(prstatus->pr_info.si_errno); 3841 prstatus->pr_cursig = tswap16(prstatus->pr_cursig); 3842 prstatus->pr_sigpend = tswapal(prstatus->pr_sigpend); 3843 prstatus->pr_sighold = tswapal(prstatus->pr_sighold); 3844 prstatus->pr_pid = tswap32(prstatus->pr_pid); 3845 prstatus->pr_ppid = tswap32(prstatus->pr_ppid); 3846 prstatus->pr_pgrp = tswap32(prstatus->pr_pgrp); 3847 prstatus->pr_sid = tswap32(prstatus->pr_sid); 3848 /* cpu times are not filled, so we skip them */ 3849 /* regs should be in correct format already */ 3850 prstatus->pr_fpvalid = tswap32(prstatus->pr_fpvalid); 3851 } 3852 3853 static void bswap_psinfo(struct target_elf_prpsinfo *psinfo) 3854 { 3855 psinfo->pr_flag = tswapal(psinfo->pr_flag); 3856 psinfo->pr_uid = tswap16(psinfo->pr_uid); 3857 psinfo->pr_gid = tswap16(psinfo->pr_gid); 3858 psinfo->pr_pid = tswap32(psinfo->pr_pid); 3859 psinfo->pr_ppid = tswap32(psinfo->pr_ppid); 3860 psinfo->pr_pgrp = tswap32(psinfo->pr_pgrp); 3861 psinfo->pr_sid = tswap32(psinfo->pr_sid); 3862 } 3863 3864 static void bswap_note(struct elf_note *en) 3865 { 3866 bswap32s(&en->n_namesz); 3867 bswap32s(&en->n_descsz); 3868 bswap32s(&en->n_type); 3869 } 3870 #else 3871 static inline void bswap_prstatus(struct target_elf_prstatus *p) { } 3872 static inline void bswap_psinfo(struct target_elf_prpsinfo *p) {} 3873 static inline void bswap_note(struct elf_note *en) { } 3874 #endif /* BSWAP_NEEDED */ 3875 3876 /* 3877 * Minimal support for linux memory regions. These are needed 3878 * when we are finding out what memory exactly belongs to 3879 * emulated process. No locks needed here, as long as 3880 * thread that received the signal is stopped. 3881 */ 3882 3883 static struct mm_struct *vma_init(void) 3884 { 3885 struct mm_struct *mm; 3886 3887 if ((mm = g_malloc(sizeof (*mm))) == NULL) 3888 return (NULL); 3889 3890 mm->mm_count = 0; 3891 QTAILQ_INIT(&mm->mm_mmap); 3892 3893 return (mm); 3894 } 3895 3896 static void vma_delete(struct mm_struct *mm) 3897 { 3898 struct vm_area_struct *vma; 3899 3900 while ((vma = vma_first(mm)) != NULL) { 3901 QTAILQ_REMOVE(&mm->mm_mmap, vma, vma_link); 3902 g_free(vma); 3903 } 3904 g_free(mm); 3905 } 3906 3907 static int vma_add_mapping(struct mm_struct *mm, target_ulong start, 3908 target_ulong end, abi_ulong flags) 3909 { 3910 struct vm_area_struct *vma; 3911 3912 if ((vma = g_malloc0(sizeof (*vma))) == NULL) 3913 return (-1); 3914 3915 vma->vma_start = start; 3916 vma->vma_end = end; 3917 vma->vma_flags = flags; 3918 3919 QTAILQ_INSERT_TAIL(&mm->mm_mmap, vma, vma_link); 3920 mm->mm_count++; 3921 3922 return (0); 3923 } 3924 3925 static struct vm_area_struct *vma_first(const struct mm_struct *mm) 3926 { 3927 return (QTAILQ_FIRST(&mm->mm_mmap)); 3928 } 3929 3930 static struct vm_area_struct *vma_next(struct vm_area_struct *vma) 3931 { 3932 return (QTAILQ_NEXT(vma, vma_link)); 3933 } 3934 3935 static int vma_get_mapping_count(const struct mm_struct *mm) 3936 { 3937 return (mm->mm_count); 3938 } 3939 3940 /* 3941 * Calculate file (dump) size of given memory region. 3942 */ 3943 static abi_ulong vma_dump_size(const struct vm_area_struct *vma) 3944 { 3945 /* if we cannot even read the first page, skip it */ 3946 if (!access_ok_untagged(VERIFY_READ, vma->vma_start, TARGET_PAGE_SIZE)) 3947 return (0); 3948 3949 /* 3950 * Usually we don't dump executable pages as they contain 3951 * non-writable code that debugger can read directly from 3952 * target library etc. However, thread stacks are marked 3953 * also executable so we read in first page of given region 3954 * and check whether it contains elf header. If there is 3955 * no elf header, we dump it. 3956 */ 3957 if (vma->vma_flags & PROT_EXEC) { 3958 char page[TARGET_PAGE_SIZE]; 3959 3960 if (copy_from_user(page, vma->vma_start, sizeof (page))) { 3961 return 0; 3962 } 3963 if ((page[EI_MAG0] == ELFMAG0) && 3964 (page[EI_MAG1] == ELFMAG1) && 3965 (page[EI_MAG2] == ELFMAG2) && 3966 (page[EI_MAG3] == ELFMAG3)) { 3967 /* 3968 * Mappings are possibly from ELF binary. Don't dump 3969 * them. 3970 */ 3971 return (0); 3972 } 3973 } 3974 3975 return (vma->vma_end - vma->vma_start); 3976 } 3977 3978 static int vma_walker(void *priv, target_ulong start, target_ulong end, 3979 unsigned long flags) 3980 { 3981 struct mm_struct *mm = (struct mm_struct *)priv; 3982 3983 vma_add_mapping(mm, start, end, flags); 3984 return (0); 3985 } 3986 3987 static void fill_note(struct memelfnote *note, const char *name, int type, 3988 unsigned int sz, void *data) 3989 { 3990 unsigned int namesz; 3991 3992 namesz = strlen(name) + 1; 3993 note->name = name; 3994 note->namesz = namesz; 3995 note->namesz_rounded = roundup(namesz, sizeof (int32_t)); 3996 note->type = type; 3997 note->datasz = sz; 3998 note->datasz_rounded = roundup(sz, sizeof (int32_t)); 3999 4000 note->data = data; 4001 4002 /* 4003 * We calculate rounded up note size here as specified by 4004 * ELF document. 4005 */ 4006 note->notesz = sizeof (struct elf_note) + 4007 note->namesz_rounded + note->datasz_rounded; 4008 } 4009 4010 static void fill_elf_header(struct elfhdr *elf, int segs, uint16_t machine, 4011 uint32_t flags) 4012 { 4013 (void) memset(elf, 0, sizeof(*elf)); 4014 4015 (void) memcpy(elf->e_ident, ELFMAG, SELFMAG); 4016 elf->e_ident[EI_CLASS] = ELF_CLASS; 4017 elf->e_ident[EI_DATA] = ELF_DATA; 4018 elf->e_ident[EI_VERSION] = EV_CURRENT; 4019 elf->e_ident[EI_OSABI] = ELF_OSABI; 4020 4021 elf->e_type = ET_CORE; 4022 elf->e_machine = machine; 4023 elf->e_version = EV_CURRENT; 4024 elf->e_phoff = sizeof(struct elfhdr); 4025 elf->e_flags = flags; 4026 elf->e_ehsize = sizeof(struct elfhdr); 4027 elf->e_phentsize = sizeof(struct elf_phdr); 4028 elf->e_phnum = segs; 4029 4030 bswap_ehdr(elf); 4031 } 4032 4033 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, off_t offset) 4034 { 4035 phdr->p_type = PT_NOTE; 4036 phdr->p_offset = offset; 4037 phdr->p_vaddr = 0; 4038 phdr->p_paddr = 0; 4039 phdr->p_filesz = sz; 4040 phdr->p_memsz = 0; 4041 phdr->p_flags = 0; 4042 phdr->p_align = 0; 4043 4044 bswap_phdr(phdr, 1); 4045 } 4046 4047 static size_t note_size(const struct memelfnote *note) 4048 { 4049 return (note->notesz); 4050 } 4051 4052 static void fill_prstatus(struct target_elf_prstatus *prstatus, 4053 const TaskState *ts, int signr) 4054 { 4055 (void) memset(prstatus, 0, sizeof (*prstatus)); 4056 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr; 4057 prstatus->pr_pid = ts->ts_tid; 4058 prstatus->pr_ppid = getppid(); 4059 prstatus->pr_pgrp = getpgrp(); 4060 prstatus->pr_sid = getsid(0); 4061 4062 bswap_prstatus(prstatus); 4063 } 4064 4065 static int fill_psinfo(struct target_elf_prpsinfo *psinfo, const TaskState *ts) 4066 { 4067 char *base_filename; 4068 unsigned int i, len; 4069 4070 (void) memset(psinfo, 0, sizeof (*psinfo)); 4071 4072 len = ts->info->env_strings - ts->info->arg_strings; 4073 if (len >= ELF_PRARGSZ) 4074 len = ELF_PRARGSZ - 1; 4075 if (copy_from_user(&psinfo->pr_psargs, ts->info->arg_strings, len)) { 4076 return -EFAULT; 4077 } 4078 for (i = 0; i < len; i++) 4079 if (psinfo->pr_psargs[i] == 0) 4080 psinfo->pr_psargs[i] = ' '; 4081 psinfo->pr_psargs[len] = 0; 4082 4083 psinfo->pr_pid = getpid(); 4084 psinfo->pr_ppid = getppid(); 4085 psinfo->pr_pgrp = getpgrp(); 4086 psinfo->pr_sid = getsid(0); 4087 psinfo->pr_uid = getuid(); 4088 psinfo->pr_gid = getgid(); 4089 4090 base_filename = g_path_get_basename(ts->bprm->filename); 4091 /* 4092 * Using strncpy here is fine: at max-length, 4093 * this field is not NUL-terminated. 4094 */ 4095 (void) strncpy(psinfo->pr_fname, base_filename, 4096 sizeof(psinfo->pr_fname)); 4097 4098 g_free(base_filename); 4099 bswap_psinfo(psinfo); 4100 return (0); 4101 } 4102 4103 static void fill_auxv_note(struct memelfnote *note, const TaskState *ts) 4104 { 4105 elf_addr_t auxv = (elf_addr_t)ts->info->saved_auxv; 4106 elf_addr_t orig_auxv = auxv; 4107 void *ptr; 4108 int len = ts->info->auxv_len; 4109 4110 /* 4111 * Auxiliary vector is stored in target process stack. It contains 4112 * {type, value} pairs that we need to dump into note. This is not 4113 * strictly necessary but we do it here for sake of completeness. 4114 */ 4115 4116 /* read in whole auxv vector and copy it to memelfnote */ 4117 ptr = lock_user(VERIFY_READ, orig_auxv, len, 0); 4118 if (ptr != NULL) { 4119 fill_note(note, "CORE", NT_AUXV, len, ptr); 4120 unlock_user(ptr, auxv, len); 4121 } 4122 } 4123 4124 /* 4125 * Constructs name of coredump file. We have following convention 4126 * for the name: 4127 * qemu_<basename-of-target-binary>_<date>-<time>_<pid>.core 4128 * 4129 * Returns the filename 4130 */ 4131 static char *core_dump_filename(const TaskState *ts) 4132 { 4133 g_autoptr(GDateTime) now = g_date_time_new_now_local(); 4134 g_autofree char *nowstr = g_date_time_format(now, "%Y%m%d-%H%M%S"); 4135 g_autofree char *base_filename = g_path_get_basename(ts->bprm->filename); 4136 4137 return g_strdup_printf("qemu_%s_%s_%d.core", 4138 base_filename, nowstr, (int)getpid()); 4139 } 4140 4141 static int dump_write(int fd, const void *ptr, size_t size) 4142 { 4143 const char *bufp = (const char *)ptr; 4144 ssize_t bytes_written, bytes_left; 4145 struct rlimit dumpsize; 4146 off_t pos; 4147 4148 bytes_written = 0; 4149 getrlimit(RLIMIT_CORE, &dumpsize); 4150 if ((pos = lseek(fd, 0, SEEK_CUR))==-1) { 4151 if (errno == ESPIPE) { /* not a seekable stream */ 4152 bytes_left = size; 4153 } else { 4154 return pos; 4155 } 4156 } else { 4157 if (dumpsize.rlim_cur <= pos) { 4158 return -1; 4159 } else if (dumpsize.rlim_cur == RLIM_INFINITY) { 4160 bytes_left = size; 4161 } else { 4162 size_t limit_left=dumpsize.rlim_cur - pos; 4163 bytes_left = limit_left >= size ? size : limit_left ; 4164 } 4165 } 4166 4167 /* 4168 * In normal conditions, single write(2) should do but 4169 * in case of socket etc. this mechanism is more portable. 4170 */ 4171 do { 4172 bytes_written = write(fd, bufp, bytes_left); 4173 if (bytes_written < 0) { 4174 if (errno == EINTR) 4175 continue; 4176 return (-1); 4177 } else if (bytes_written == 0) { /* eof */ 4178 return (-1); 4179 } 4180 bufp += bytes_written; 4181 bytes_left -= bytes_written; 4182 } while (bytes_left > 0); 4183 4184 return (0); 4185 } 4186 4187 static int write_note(struct memelfnote *men, int fd) 4188 { 4189 struct elf_note en; 4190 4191 en.n_namesz = men->namesz; 4192 en.n_type = men->type; 4193 en.n_descsz = men->datasz; 4194 4195 bswap_note(&en); 4196 4197 if (dump_write(fd, &en, sizeof(en)) != 0) 4198 return (-1); 4199 if (dump_write(fd, men->name, men->namesz_rounded) != 0) 4200 return (-1); 4201 if (dump_write(fd, men->data, men->datasz_rounded) != 0) 4202 return (-1); 4203 4204 return (0); 4205 } 4206 4207 static void fill_thread_info(struct elf_note_info *info, const CPUArchState *env) 4208 { 4209 CPUState *cpu = env_cpu((CPUArchState *)env); 4210 TaskState *ts = (TaskState *)cpu->opaque; 4211 struct elf_thread_status *ets; 4212 4213 ets = g_malloc0(sizeof (*ets)); 4214 ets->num_notes = 1; /* only prstatus is dumped */ 4215 fill_prstatus(&ets->prstatus, ts, 0); 4216 elf_core_copy_regs(&ets->prstatus.pr_reg, env); 4217 fill_note(&ets->notes[0], "CORE", NT_PRSTATUS, sizeof (ets->prstatus), 4218 &ets->prstatus); 4219 4220 QTAILQ_INSERT_TAIL(&info->thread_list, ets, ets_link); 4221 4222 info->notes_size += note_size(&ets->notes[0]); 4223 } 4224 4225 static void init_note_info(struct elf_note_info *info) 4226 { 4227 /* Initialize the elf_note_info structure so that it is at 4228 * least safe to call free_note_info() on it. Must be 4229 * called before calling fill_note_info(). 4230 */ 4231 memset(info, 0, sizeof (*info)); 4232 QTAILQ_INIT(&info->thread_list); 4233 } 4234 4235 static int fill_note_info(struct elf_note_info *info, 4236 long signr, const CPUArchState *env) 4237 { 4238 #define NUMNOTES 3 4239 CPUState *cpu = env_cpu((CPUArchState *)env); 4240 TaskState *ts = (TaskState *)cpu->opaque; 4241 int i; 4242 4243 info->notes = g_new0(struct memelfnote, NUMNOTES); 4244 if (info->notes == NULL) 4245 return (-ENOMEM); 4246 info->prstatus = g_malloc0(sizeof (*info->prstatus)); 4247 if (info->prstatus == NULL) 4248 return (-ENOMEM); 4249 info->psinfo = g_malloc0(sizeof (*info->psinfo)); 4250 if (info->prstatus == NULL) 4251 return (-ENOMEM); 4252 4253 /* 4254 * First fill in status (and registers) of current thread 4255 * including process info & aux vector. 4256 */ 4257 fill_prstatus(info->prstatus, ts, signr); 4258 elf_core_copy_regs(&info->prstatus->pr_reg, env); 4259 fill_note(&info->notes[0], "CORE", NT_PRSTATUS, 4260 sizeof (*info->prstatus), info->prstatus); 4261 fill_psinfo(info->psinfo, ts); 4262 fill_note(&info->notes[1], "CORE", NT_PRPSINFO, 4263 sizeof (*info->psinfo), info->psinfo); 4264 fill_auxv_note(&info->notes[2], ts); 4265 info->numnote = 3; 4266 4267 info->notes_size = 0; 4268 for (i = 0; i < info->numnote; i++) 4269 info->notes_size += note_size(&info->notes[i]); 4270 4271 /* read and fill status of all threads */ 4272 WITH_QEMU_LOCK_GUARD(&qemu_cpu_list_lock) { 4273 CPU_FOREACH(cpu) { 4274 if (cpu == thread_cpu) { 4275 continue; 4276 } 4277 fill_thread_info(info, cpu->env_ptr); 4278 } 4279 } 4280 4281 return (0); 4282 } 4283 4284 static void free_note_info(struct elf_note_info *info) 4285 { 4286 struct elf_thread_status *ets; 4287 4288 while (!QTAILQ_EMPTY(&info->thread_list)) { 4289 ets = QTAILQ_FIRST(&info->thread_list); 4290 QTAILQ_REMOVE(&info->thread_list, ets, ets_link); 4291 g_free(ets); 4292 } 4293 4294 g_free(info->prstatus); 4295 g_free(info->psinfo); 4296 g_free(info->notes); 4297 } 4298 4299 static int write_note_info(struct elf_note_info *info, int fd) 4300 { 4301 struct elf_thread_status *ets; 4302 int i, error = 0; 4303 4304 /* write prstatus, psinfo and auxv for current thread */ 4305 for (i = 0; i < info->numnote; i++) 4306 if ((error = write_note(&info->notes[i], fd)) != 0) 4307 return (error); 4308 4309 /* write prstatus for each thread */ 4310 QTAILQ_FOREACH(ets, &info->thread_list, ets_link) { 4311 if ((error = write_note(&ets->notes[0], fd)) != 0) 4312 return (error); 4313 } 4314 4315 return (0); 4316 } 4317 4318 /* 4319 * Write out ELF coredump. 4320 * 4321 * See documentation of ELF object file format in: 4322 * http://www.caldera.com/developers/devspecs/gabi41.pdf 4323 * 4324 * Coredump format in linux is following: 4325 * 4326 * 0 +----------------------+ \ 4327 * | ELF header | ET_CORE | 4328 * +----------------------+ | 4329 * | ELF program headers | |--- headers 4330 * | - NOTE section | | 4331 * | - PT_LOAD sections | | 4332 * +----------------------+ / 4333 * | NOTEs: | 4334 * | - NT_PRSTATUS | 4335 * | - NT_PRSINFO | 4336 * | - NT_AUXV | 4337 * +----------------------+ <-- aligned to target page 4338 * | Process memory dump | 4339 * : : 4340 * . . 4341 * : : 4342 * | | 4343 * +----------------------+ 4344 * 4345 * NT_PRSTATUS -> struct elf_prstatus (per thread) 4346 * NT_PRSINFO -> struct elf_prpsinfo 4347 * NT_AUXV is array of { type, value } pairs (see fill_auxv_note()). 4348 * 4349 * Format follows System V format as close as possible. Current 4350 * version limitations are as follows: 4351 * - no floating point registers are dumped 4352 * 4353 * Function returns 0 in case of success, negative errno otherwise. 4354 * 4355 * TODO: make this work also during runtime: it should be 4356 * possible to force coredump from running process and then 4357 * continue processing. For example qemu could set up SIGUSR2 4358 * handler (provided that target process haven't registered 4359 * handler for that) that does the dump when signal is received. 4360 */ 4361 static int elf_core_dump(int signr, const CPUArchState *env) 4362 { 4363 const CPUState *cpu = env_cpu((CPUArchState *)env); 4364 const TaskState *ts = (const TaskState *)cpu->opaque; 4365 struct vm_area_struct *vma = NULL; 4366 g_autofree char *corefile = NULL; 4367 struct elf_note_info info; 4368 struct elfhdr elf; 4369 struct elf_phdr phdr; 4370 struct rlimit dumpsize; 4371 struct mm_struct *mm = NULL; 4372 off_t offset = 0, data_offset = 0; 4373 int segs = 0; 4374 int fd = -1; 4375 4376 init_note_info(&info); 4377 4378 errno = 0; 4379 getrlimit(RLIMIT_CORE, &dumpsize); 4380 if (dumpsize.rlim_cur == 0) 4381 return 0; 4382 4383 corefile = core_dump_filename(ts); 4384 4385 if ((fd = open(corefile, O_WRONLY | O_CREAT, 4386 S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH)) < 0) 4387 return (-errno); 4388 4389 /* 4390 * Walk through target process memory mappings and 4391 * set up structure containing this information. After 4392 * this point vma_xxx functions can be used. 4393 */ 4394 if ((mm = vma_init()) == NULL) 4395 goto out; 4396 4397 walk_memory_regions(mm, vma_walker); 4398 segs = vma_get_mapping_count(mm); 4399 4400 /* 4401 * Construct valid coredump ELF header. We also 4402 * add one more segment for notes. 4403 */ 4404 fill_elf_header(&elf, segs + 1, ELF_MACHINE, 0); 4405 if (dump_write(fd, &elf, sizeof (elf)) != 0) 4406 goto out; 4407 4408 /* fill in the in-memory version of notes */ 4409 if (fill_note_info(&info, signr, env) < 0) 4410 goto out; 4411 4412 offset += sizeof (elf); /* elf header */ 4413 offset += (segs + 1) * sizeof (struct elf_phdr); /* program headers */ 4414 4415 /* write out notes program header */ 4416 fill_elf_note_phdr(&phdr, info.notes_size, offset); 4417 4418 offset += info.notes_size; 4419 if (dump_write(fd, &phdr, sizeof (phdr)) != 0) 4420 goto out; 4421 4422 /* 4423 * ELF specification wants data to start at page boundary so 4424 * we align it here. 4425 */ 4426 data_offset = offset = roundup(offset, ELF_EXEC_PAGESIZE); 4427 4428 /* 4429 * Write program headers for memory regions mapped in 4430 * the target process. 4431 */ 4432 for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) { 4433 (void) memset(&phdr, 0, sizeof (phdr)); 4434 4435 phdr.p_type = PT_LOAD; 4436 phdr.p_offset = offset; 4437 phdr.p_vaddr = vma->vma_start; 4438 phdr.p_paddr = 0; 4439 phdr.p_filesz = vma_dump_size(vma); 4440 offset += phdr.p_filesz; 4441 phdr.p_memsz = vma->vma_end - vma->vma_start; 4442 phdr.p_flags = vma->vma_flags & PROT_READ ? PF_R : 0; 4443 if (vma->vma_flags & PROT_WRITE) 4444 phdr.p_flags |= PF_W; 4445 if (vma->vma_flags & PROT_EXEC) 4446 phdr.p_flags |= PF_X; 4447 phdr.p_align = ELF_EXEC_PAGESIZE; 4448 4449 bswap_phdr(&phdr, 1); 4450 if (dump_write(fd, &phdr, sizeof(phdr)) != 0) { 4451 goto out; 4452 } 4453 } 4454 4455 /* 4456 * Next we write notes just after program headers. No 4457 * alignment needed here. 4458 */ 4459 if (write_note_info(&info, fd) < 0) 4460 goto out; 4461 4462 /* align data to page boundary */ 4463 if (lseek(fd, data_offset, SEEK_SET) != data_offset) 4464 goto out; 4465 4466 /* 4467 * Finally we can dump process memory into corefile as well. 4468 */ 4469 for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) { 4470 abi_ulong addr; 4471 abi_ulong end; 4472 4473 end = vma->vma_start + vma_dump_size(vma); 4474 4475 for (addr = vma->vma_start; addr < end; 4476 addr += TARGET_PAGE_SIZE) { 4477 char page[TARGET_PAGE_SIZE]; 4478 int error; 4479 4480 /* 4481 * Read in page from target process memory and 4482 * write it to coredump file. 4483 */ 4484 error = copy_from_user(page, addr, sizeof (page)); 4485 if (error != 0) { 4486 (void) fprintf(stderr, "unable to dump " TARGET_ABI_FMT_lx "\n", 4487 addr); 4488 errno = -error; 4489 goto out; 4490 } 4491 if (dump_write(fd, page, TARGET_PAGE_SIZE) < 0) 4492 goto out; 4493 } 4494 } 4495 4496 out: 4497 free_note_info(&info); 4498 if (mm != NULL) 4499 vma_delete(mm); 4500 (void) close(fd); 4501 4502 if (errno != 0) 4503 return (-errno); 4504 return (0); 4505 } 4506 #endif /* USE_ELF_CORE_DUMP */ 4507 4508 void do_init_thread(struct target_pt_regs *regs, struct image_info *infop) 4509 { 4510 init_thread(regs, infop); 4511 } 4512