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 ARMCPU *cpu = ARM_CPU(thread_cpu); 428 abi_ptr commpage; 429 void *want; 430 void *addr; 431 432 /* 433 * M-profile allocates maximum of 2GB address space, so can never 434 * allocate the commpage. Skip it. 435 */ 436 if (arm_feature(&cpu->env, ARM_FEATURE_M)) { 437 return true; 438 } 439 440 commpage = HI_COMMPAGE & -qemu_host_page_size; 441 want = g2h_untagged(commpage); 442 addr = mmap(want, qemu_host_page_size, PROT_READ | PROT_WRITE, 443 MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0); 444 445 if (addr == MAP_FAILED) { 446 perror("Allocating guest commpage"); 447 exit(EXIT_FAILURE); 448 } 449 if (addr != want) { 450 return false; 451 } 452 453 /* Set kernel helper versions; rest of page is 0. */ 454 __put_user(5, (uint32_t *)g2h_untagged(0xffff0ffcu)); 455 456 if (mprotect(addr, qemu_host_page_size, PROT_READ)) { 457 perror("Protecting guest commpage"); 458 exit(EXIT_FAILURE); 459 } 460 461 page_set_flags(commpage, commpage | ~qemu_host_page_mask, 462 PAGE_READ | PAGE_EXEC | PAGE_VALID); 463 return true; 464 } 465 466 #define ELF_HWCAP get_elf_hwcap() 467 #define ELF_HWCAP2 get_elf_hwcap2() 468 469 static uint32_t get_elf_hwcap(void) 470 { 471 ARMCPU *cpu = ARM_CPU(thread_cpu); 472 uint32_t hwcaps = 0; 473 474 hwcaps |= ARM_HWCAP_ARM_SWP; 475 hwcaps |= ARM_HWCAP_ARM_HALF; 476 hwcaps |= ARM_HWCAP_ARM_THUMB; 477 hwcaps |= ARM_HWCAP_ARM_FAST_MULT; 478 479 /* probe for the extra features */ 480 #define GET_FEATURE(feat, hwcap) \ 481 do { if (arm_feature(&cpu->env, feat)) { hwcaps |= hwcap; } } while (0) 482 483 #define GET_FEATURE_ID(feat, hwcap) \ 484 do { if (cpu_isar_feature(feat, cpu)) { hwcaps |= hwcap; } } while (0) 485 486 /* EDSP is in v5TE and above, but all our v5 CPUs are v5TE */ 487 GET_FEATURE(ARM_FEATURE_V5, ARM_HWCAP_ARM_EDSP); 488 GET_FEATURE(ARM_FEATURE_IWMMXT, ARM_HWCAP_ARM_IWMMXT); 489 GET_FEATURE(ARM_FEATURE_THUMB2EE, ARM_HWCAP_ARM_THUMBEE); 490 GET_FEATURE(ARM_FEATURE_NEON, ARM_HWCAP_ARM_NEON); 491 GET_FEATURE(ARM_FEATURE_V6K, ARM_HWCAP_ARM_TLS); 492 GET_FEATURE(ARM_FEATURE_LPAE, ARM_HWCAP_ARM_LPAE); 493 GET_FEATURE_ID(aa32_arm_div, ARM_HWCAP_ARM_IDIVA); 494 GET_FEATURE_ID(aa32_thumb_div, ARM_HWCAP_ARM_IDIVT); 495 GET_FEATURE_ID(aa32_vfp, ARM_HWCAP_ARM_VFP); 496 497 if (cpu_isar_feature(aa32_fpsp_v3, cpu) || 498 cpu_isar_feature(aa32_fpdp_v3, cpu)) { 499 hwcaps |= ARM_HWCAP_ARM_VFPv3; 500 if (cpu_isar_feature(aa32_simd_r32, cpu)) { 501 hwcaps |= ARM_HWCAP_ARM_VFPD32; 502 } else { 503 hwcaps |= ARM_HWCAP_ARM_VFPv3D16; 504 } 505 } 506 GET_FEATURE_ID(aa32_simdfmac, ARM_HWCAP_ARM_VFPv4); 507 508 return hwcaps; 509 } 510 511 static uint32_t get_elf_hwcap2(void) 512 { 513 ARMCPU *cpu = ARM_CPU(thread_cpu); 514 uint32_t hwcaps = 0; 515 516 GET_FEATURE_ID(aa32_aes, ARM_HWCAP2_ARM_AES); 517 GET_FEATURE_ID(aa32_pmull, ARM_HWCAP2_ARM_PMULL); 518 GET_FEATURE_ID(aa32_sha1, ARM_HWCAP2_ARM_SHA1); 519 GET_FEATURE_ID(aa32_sha2, ARM_HWCAP2_ARM_SHA2); 520 GET_FEATURE_ID(aa32_crc32, ARM_HWCAP2_ARM_CRC32); 521 return hwcaps; 522 } 523 524 #undef GET_FEATURE 525 #undef GET_FEATURE_ID 526 527 #define ELF_PLATFORM get_elf_platform() 528 529 static const char *get_elf_platform(void) 530 { 531 CPUARMState *env = thread_cpu->env_ptr; 532 533 #if TARGET_BIG_ENDIAN 534 # define END "b" 535 #else 536 # define END "l" 537 #endif 538 539 if (arm_feature(env, ARM_FEATURE_V8)) { 540 return "v8" END; 541 } else if (arm_feature(env, ARM_FEATURE_V7)) { 542 if (arm_feature(env, ARM_FEATURE_M)) { 543 return "v7m" END; 544 } else { 545 return "v7" END; 546 } 547 } else if (arm_feature(env, ARM_FEATURE_V6)) { 548 return "v6" END; 549 } else if (arm_feature(env, ARM_FEATURE_V5)) { 550 return "v5" END; 551 } else { 552 return "v4" END; 553 } 554 555 #undef END 556 } 557 558 #else 559 /* 64 bit ARM definitions */ 560 #define ELF_START_MMAP 0x80000000 561 562 #define ELF_ARCH EM_AARCH64 563 #define ELF_CLASS ELFCLASS64 564 #if TARGET_BIG_ENDIAN 565 # define ELF_PLATFORM "aarch64_be" 566 #else 567 # define ELF_PLATFORM "aarch64" 568 #endif 569 570 static inline void init_thread(struct target_pt_regs *regs, 571 struct image_info *infop) 572 { 573 abi_long stack = infop->start_stack; 574 memset(regs, 0, sizeof(*regs)); 575 576 regs->pc = infop->entry & ~0x3ULL; 577 regs->sp = stack; 578 } 579 580 #define ELF_NREG 34 581 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 582 583 static void elf_core_copy_regs(target_elf_gregset_t *regs, 584 const CPUARMState *env) 585 { 586 int i; 587 588 for (i = 0; i < 32; i++) { 589 (*regs)[i] = tswapreg(env->xregs[i]); 590 } 591 (*regs)[32] = tswapreg(env->pc); 592 (*regs)[33] = tswapreg(pstate_read((CPUARMState *)env)); 593 } 594 595 #define USE_ELF_CORE_DUMP 596 #define ELF_EXEC_PAGESIZE 4096 597 598 enum { 599 ARM_HWCAP_A64_FP = 1 << 0, 600 ARM_HWCAP_A64_ASIMD = 1 << 1, 601 ARM_HWCAP_A64_EVTSTRM = 1 << 2, 602 ARM_HWCAP_A64_AES = 1 << 3, 603 ARM_HWCAP_A64_PMULL = 1 << 4, 604 ARM_HWCAP_A64_SHA1 = 1 << 5, 605 ARM_HWCAP_A64_SHA2 = 1 << 6, 606 ARM_HWCAP_A64_CRC32 = 1 << 7, 607 ARM_HWCAP_A64_ATOMICS = 1 << 8, 608 ARM_HWCAP_A64_FPHP = 1 << 9, 609 ARM_HWCAP_A64_ASIMDHP = 1 << 10, 610 ARM_HWCAP_A64_CPUID = 1 << 11, 611 ARM_HWCAP_A64_ASIMDRDM = 1 << 12, 612 ARM_HWCAP_A64_JSCVT = 1 << 13, 613 ARM_HWCAP_A64_FCMA = 1 << 14, 614 ARM_HWCAP_A64_LRCPC = 1 << 15, 615 ARM_HWCAP_A64_DCPOP = 1 << 16, 616 ARM_HWCAP_A64_SHA3 = 1 << 17, 617 ARM_HWCAP_A64_SM3 = 1 << 18, 618 ARM_HWCAP_A64_SM4 = 1 << 19, 619 ARM_HWCAP_A64_ASIMDDP = 1 << 20, 620 ARM_HWCAP_A64_SHA512 = 1 << 21, 621 ARM_HWCAP_A64_SVE = 1 << 22, 622 ARM_HWCAP_A64_ASIMDFHM = 1 << 23, 623 ARM_HWCAP_A64_DIT = 1 << 24, 624 ARM_HWCAP_A64_USCAT = 1 << 25, 625 ARM_HWCAP_A64_ILRCPC = 1 << 26, 626 ARM_HWCAP_A64_FLAGM = 1 << 27, 627 ARM_HWCAP_A64_SSBS = 1 << 28, 628 ARM_HWCAP_A64_SB = 1 << 29, 629 ARM_HWCAP_A64_PACA = 1 << 30, 630 ARM_HWCAP_A64_PACG = 1UL << 31, 631 632 ARM_HWCAP2_A64_DCPODP = 1 << 0, 633 ARM_HWCAP2_A64_SVE2 = 1 << 1, 634 ARM_HWCAP2_A64_SVEAES = 1 << 2, 635 ARM_HWCAP2_A64_SVEPMULL = 1 << 3, 636 ARM_HWCAP2_A64_SVEBITPERM = 1 << 4, 637 ARM_HWCAP2_A64_SVESHA3 = 1 << 5, 638 ARM_HWCAP2_A64_SVESM4 = 1 << 6, 639 ARM_HWCAP2_A64_FLAGM2 = 1 << 7, 640 ARM_HWCAP2_A64_FRINT = 1 << 8, 641 ARM_HWCAP2_A64_SVEI8MM = 1 << 9, 642 ARM_HWCAP2_A64_SVEF32MM = 1 << 10, 643 ARM_HWCAP2_A64_SVEF64MM = 1 << 11, 644 ARM_HWCAP2_A64_SVEBF16 = 1 << 12, 645 ARM_HWCAP2_A64_I8MM = 1 << 13, 646 ARM_HWCAP2_A64_BF16 = 1 << 14, 647 ARM_HWCAP2_A64_DGH = 1 << 15, 648 ARM_HWCAP2_A64_RNG = 1 << 16, 649 ARM_HWCAP2_A64_BTI = 1 << 17, 650 ARM_HWCAP2_A64_MTE = 1 << 18, 651 ARM_HWCAP2_A64_ECV = 1 << 19, 652 ARM_HWCAP2_A64_AFP = 1 << 20, 653 ARM_HWCAP2_A64_RPRES = 1 << 21, 654 ARM_HWCAP2_A64_MTE3 = 1 << 22, 655 ARM_HWCAP2_A64_SME = 1 << 23, 656 ARM_HWCAP2_A64_SME_I16I64 = 1 << 24, 657 ARM_HWCAP2_A64_SME_F64F64 = 1 << 25, 658 ARM_HWCAP2_A64_SME_I8I32 = 1 << 26, 659 ARM_HWCAP2_A64_SME_F16F32 = 1 << 27, 660 ARM_HWCAP2_A64_SME_B16F32 = 1 << 28, 661 ARM_HWCAP2_A64_SME_F32F32 = 1 << 29, 662 ARM_HWCAP2_A64_SME_FA64 = 1 << 30, 663 }; 664 665 #define ELF_HWCAP get_elf_hwcap() 666 #define ELF_HWCAP2 get_elf_hwcap2() 667 668 #define GET_FEATURE_ID(feat, hwcap) \ 669 do { if (cpu_isar_feature(feat, cpu)) { hwcaps |= hwcap; } } while (0) 670 671 static uint32_t get_elf_hwcap(void) 672 { 673 ARMCPU *cpu = ARM_CPU(thread_cpu); 674 uint32_t hwcaps = 0; 675 676 hwcaps |= ARM_HWCAP_A64_FP; 677 hwcaps |= ARM_HWCAP_A64_ASIMD; 678 hwcaps |= ARM_HWCAP_A64_CPUID; 679 680 /* probe for the extra features */ 681 682 GET_FEATURE_ID(aa64_aes, ARM_HWCAP_A64_AES); 683 GET_FEATURE_ID(aa64_pmull, ARM_HWCAP_A64_PMULL); 684 GET_FEATURE_ID(aa64_sha1, ARM_HWCAP_A64_SHA1); 685 GET_FEATURE_ID(aa64_sha256, ARM_HWCAP_A64_SHA2); 686 GET_FEATURE_ID(aa64_sha512, ARM_HWCAP_A64_SHA512); 687 GET_FEATURE_ID(aa64_crc32, ARM_HWCAP_A64_CRC32); 688 GET_FEATURE_ID(aa64_sha3, ARM_HWCAP_A64_SHA3); 689 GET_FEATURE_ID(aa64_sm3, ARM_HWCAP_A64_SM3); 690 GET_FEATURE_ID(aa64_sm4, ARM_HWCAP_A64_SM4); 691 GET_FEATURE_ID(aa64_fp16, ARM_HWCAP_A64_FPHP | ARM_HWCAP_A64_ASIMDHP); 692 GET_FEATURE_ID(aa64_atomics, ARM_HWCAP_A64_ATOMICS); 693 GET_FEATURE_ID(aa64_rdm, ARM_HWCAP_A64_ASIMDRDM); 694 GET_FEATURE_ID(aa64_dp, ARM_HWCAP_A64_ASIMDDP); 695 GET_FEATURE_ID(aa64_fcma, ARM_HWCAP_A64_FCMA); 696 GET_FEATURE_ID(aa64_sve, ARM_HWCAP_A64_SVE); 697 GET_FEATURE_ID(aa64_pauth, ARM_HWCAP_A64_PACA | ARM_HWCAP_A64_PACG); 698 GET_FEATURE_ID(aa64_fhm, ARM_HWCAP_A64_ASIMDFHM); 699 GET_FEATURE_ID(aa64_jscvt, ARM_HWCAP_A64_JSCVT); 700 GET_FEATURE_ID(aa64_sb, ARM_HWCAP_A64_SB); 701 GET_FEATURE_ID(aa64_condm_4, ARM_HWCAP_A64_FLAGM); 702 GET_FEATURE_ID(aa64_dcpop, ARM_HWCAP_A64_DCPOP); 703 GET_FEATURE_ID(aa64_rcpc_8_3, ARM_HWCAP_A64_LRCPC); 704 GET_FEATURE_ID(aa64_rcpc_8_4, ARM_HWCAP_A64_ILRCPC); 705 706 return hwcaps; 707 } 708 709 static uint32_t get_elf_hwcap2(void) 710 { 711 ARMCPU *cpu = ARM_CPU(thread_cpu); 712 uint32_t hwcaps = 0; 713 714 GET_FEATURE_ID(aa64_dcpodp, ARM_HWCAP2_A64_DCPODP); 715 GET_FEATURE_ID(aa64_sve2, ARM_HWCAP2_A64_SVE2); 716 GET_FEATURE_ID(aa64_sve2_aes, ARM_HWCAP2_A64_SVEAES); 717 GET_FEATURE_ID(aa64_sve2_pmull128, ARM_HWCAP2_A64_SVEPMULL); 718 GET_FEATURE_ID(aa64_sve2_bitperm, ARM_HWCAP2_A64_SVEBITPERM); 719 GET_FEATURE_ID(aa64_sve2_sha3, ARM_HWCAP2_A64_SVESHA3); 720 GET_FEATURE_ID(aa64_sve2_sm4, ARM_HWCAP2_A64_SVESM4); 721 GET_FEATURE_ID(aa64_condm_5, ARM_HWCAP2_A64_FLAGM2); 722 GET_FEATURE_ID(aa64_frint, ARM_HWCAP2_A64_FRINT); 723 GET_FEATURE_ID(aa64_sve_i8mm, ARM_HWCAP2_A64_SVEI8MM); 724 GET_FEATURE_ID(aa64_sve_f32mm, ARM_HWCAP2_A64_SVEF32MM); 725 GET_FEATURE_ID(aa64_sve_f64mm, ARM_HWCAP2_A64_SVEF64MM); 726 GET_FEATURE_ID(aa64_sve_bf16, ARM_HWCAP2_A64_SVEBF16); 727 GET_FEATURE_ID(aa64_i8mm, ARM_HWCAP2_A64_I8MM); 728 GET_FEATURE_ID(aa64_bf16, ARM_HWCAP2_A64_BF16); 729 GET_FEATURE_ID(aa64_rndr, ARM_HWCAP2_A64_RNG); 730 GET_FEATURE_ID(aa64_bti, ARM_HWCAP2_A64_BTI); 731 GET_FEATURE_ID(aa64_mte, ARM_HWCAP2_A64_MTE); 732 GET_FEATURE_ID(aa64_sme, (ARM_HWCAP2_A64_SME | 733 ARM_HWCAP2_A64_SME_F32F32 | 734 ARM_HWCAP2_A64_SME_B16F32 | 735 ARM_HWCAP2_A64_SME_F16F32 | 736 ARM_HWCAP2_A64_SME_I8I32)); 737 GET_FEATURE_ID(aa64_sme_f64f64, ARM_HWCAP2_A64_SME_F64F64); 738 GET_FEATURE_ID(aa64_sme_i16i64, ARM_HWCAP2_A64_SME_I16I64); 739 GET_FEATURE_ID(aa64_sme_fa64, ARM_HWCAP2_A64_SME_FA64); 740 741 return hwcaps; 742 } 743 744 #undef GET_FEATURE_ID 745 746 #endif /* not TARGET_AARCH64 */ 747 #endif /* TARGET_ARM */ 748 749 #ifdef TARGET_SPARC 750 #ifdef TARGET_SPARC64 751 752 #define ELF_START_MMAP 0x80000000 753 #define ELF_HWCAP (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \ 754 | HWCAP_SPARC_MULDIV | HWCAP_SPARC_V9) 755 #ifndef TARGET_ABI32 756 #define elf_check_arch(x) ( (x) == EM_SPARCV9 || (x) == EM_SPARC32PLUS ) 757 #else 758 #define elf_check_arch(x) ( (x) == EM_SPARC32PLUS || (x) == EM_SPARC ) 759 #endif 760 761 #define ELF_CLASS ELFCLASS64 762 #define ELF_ARCH EM_SPARCV9 763 #else 764 #define ELF_START_MMAP 0x80000000 765 #define ELF_HWCAP (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \ 766 | HWCAP_SPARC_MULDIV) 767 #define ELF_CLASS ELFCLASS32 768 #define ELF_ARCH EM_SPARC 769 #endif /* TARGET_SPARC64 */ 770 771 static inline void init_thread(struct target_pt_regs *regs, 772 struct image_info *infop) 773 { 774 /* Note that target_cpu_copy_regs does not read psr/tstate. */ 775 regs->pc = infop->entry; 776 regs->npc = regs->pc + 4; 777 regs->y = 0; 778 regs->u_regs[14] = (infop->start_stack - 16 * sizeof(abi_ulong) 779 - TARGET_STACK_BIAS); 780 } 781 #endif /* TARGET_SPARC */ 782 783 #ifdef TARGET_PPC 784 785 #define ELF_MACHINE PPC_ELF_MACHINE 786 #define ELF_START_MMAP 0x80000000 787 788 #if defined(TARGET_PPC64) 789 790 #define elf_check_arch(x) ( (x) == EM_PPC64 ) 791 792 #define ELF_CLASS ELFCLASS64 793 794 #else 795 796 #define ELF_CLASS ELFCLASS32 797 #define EXSTACK_DEFAULT true 798 799 #endif 800 801 #define ELF_ARCH EM_PPC 802 803 /* Feature masks for the Aux Vector Hardware Capabilities (AT_HWCAP). 804 See arch/powerpc/include/asm/cputable.h. */ 805 enum { 806 QEMU_PPC_FEATURE_32 = 0x80000000, 807 QEMU_PPC_FEATURE_64 = 0x40000000, 808 QEMU_PPC_FEATURE_601_INSTR = 0x20000000, 809 QEMU_PPC_FEATURE_HAS_ALTIVEC = 0x10000000, 810 QEMU_PPC_FEATURE_HAS_FPU = 0x08000000, 811 QEMU_PPC_FEATURE_HAS_MMU = 0x04000000, 812 QEMU_PPC_FEATURE_HAS_4xxMAC = 0x02000000, 813 QEMU_PPC_FEATURE_UNIFIED_CACHE = 0x01000000, 814 QEMU_PPC_FEATURE_HAS_SPE = 0x00800000, 815 QEMU_PPC_FEATURE_HAS_EFP_SINGLE = 0x00400000, 816 QEMU_PPC_FEATURE_HAS_EFP_DOUBLE = 0x00200000, 817 QEMU_PPC_FEATURE_NO_TB = 0x00100000, 818 QEMU_PPC_FEATURE_POWER4 = 0x00080000, 819 QEMU_PPC_FEATURE_POWER5 = 0x00040000, 820 QEMU_PPC_FEATURE_POWER5_PLUS = 0x00020000, 821 QEMU_PPC_FEATURE_CELL = 0x00010000, 822 QEMU_PPC_FEATURE_BOOKE = 0x00008000, 823 QEMU_PPC_FEATURE_SMT = 0x00004000, 824 QEMU_PPC_FEATURE_ICACHE_SNOOP = 0x00002000, 825 QEMU_PPC_FEATURE_ARCH_2_05 = 0x00001000, 826 QEMU_PPC_FEATURE_PA6T = 0x00000800, 827 QEMU_PPC_FEATURE_HAS_DFP = 0x00000400, 828 QEMU_PPC_FEATURE_POWER6_EXT = 0x00000200, 829 QEMU_PPC_FEATURE_ARCH_2_06 = 0x00000100, 830 QEMU_PPC_FEATURE_HAS_VSX = 0x00000080, 831 QEMU_PPC_FEATURE_PSERIES_PERFMON_COMPAT = 0x00000040, 832 833 QEMU_PPC_FEATURE_TRUE_LE = 0x00000002, 834 QEMU_PPC_FEATURE_PPC_LE = 0x00000001, 835 836 /* Feature definitions in AT_HWCAP2. */ 837 QEMU_PPC_FEATURE2_ARCH_2_07 = 0x80000000, /* ISA 2.07 */ 838 QEMU_PPC_FEATURE2_HAS_HTM = 0x40000000, /* Hardware Transactional Memory */ 839 QEMU_PPC_FEATURE2_HAS_DSCR = 0x20000000, /* Data Stream Control Register */ 840 QEMU_PPC_FEATURE2_HAS_EBB = 0x10000000, /* Event Base Branching */ 841 QEMU_PPC_FEATURE2_HAS_ISEL = 0x08000000, /* Integer Select */ 842 QEMU_PPC_FEATURE2_HAS_TAR = 0x04000000, /* Target Address Register */ 843 QEMU_PPC_FEATURE2_VEC_CRYPTO = 0x02000000, 844 QEMU_PPC_FEATURE2_HTM_NOSC = 0x01000000, 845 QEMU_PPC_FEATURE2_ARCH_3_00 = 0x00800000, /* ISA 3.00 */ 846 QEMU_PPC_FEATURE2_HAS_IEEE128 = 0x00400000, /* VSX IEEE Bin Float 128-bit */ 847 QEMU_PPC_FEATURE2_DARN = 0x00200000, /* darn random number insn */ 848 QEMU_PPC_FEATURE2_SCV = 0x00100000, /* scv syscall */ 849 QEMU_PPC_FEATURE2_HTM_NO_SUSPEND = 0x00080000, /* TM w/o suspended state */ 850 QEMU_PPC_FEATURE2_ARCH_3_1 = 0x00040000, /* ISA 3.1 */ 851 QEMU_PPC_FEATURE2_MMA = 0x00020000, /* Matrix-Multiply Assist */ 852 }; 853 854 #define ELF_HWCAP get_elf_hwcap() 855 856 static uint32_t get_elf_hwcap(void) 857 { 858 PowerPCCPU *cpu = POWERPC_CPU(thread_cpu); 859 uint32_t features = 0; 860 861 /* We don't have to be terribly complete here; the high points are 862 Altivec/FP/SPE support. Anything else is just a bonus. */ 863 #define GET_FEATURE(flag, feature) \ 864 do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0) 865 #define GET_FEATURE2(flags, feature) \ 866 do { \ 867 if ((cpu->env.insns_flags2 & flags) == flags) { \ 868 features |= feature; \ 869 } \ 870 } while (0) 871 GET_FEATURE(PPC_64B, QEMU_PPC_FEATURE_64); 872 GET_FEATURE(PPC_FLOAT, QEMU_PPC_FEATURE_HAS_FPU); 873 GET_FEATURE(PPC_ALTIVEC, QEMU_PPC_FEATURE_HAS_ALTIVEC); 874 GET_FEATURE(PPC_SPE, QEMU_PPC_FEATURE_HAS_SPE); 875 GET_FEATURE(PPC_SPE_SINGLE, QEMU_PPC_FEATURE_HAS_EFP_SINGLE); 876 GET_FEATURE(PPC_SPE_DOUBLE, QEMU_PPC_FEATURE_HAS_EFP_DOUBLE); 877 GET_FEATURE(PPC_BOOKE, QEMU_PPC_FEATURE_BOOKE); 878 GET_FEATURE(PPC_405_MAC, QEMU_PPC_FEATURE_HAS_4xxMAC); 879 GET_FEATURE2(PPC2_DFP, QEMU_PPC_FEATURE_HAS_DFP); 880 GET_FEATURE2(PPC2_VSX, QEMU_PPC_FEATURE_HAS_VSX); 881 GET_FEATURE2((PPC2_PERM_ISA206 | PPC2_DIVE_ISA206 | PPC2_ATOMIC_ISA206 | 882 PPC2_FP_CVT_ISA206 | PPC2_FP_TST_ISA206), 883 QEMU_PPC_FEATURE_ARCH_2_06); 884 #undef GET_FEATURE 885 #undef GET_FEATURE2 886 887 return features; 888 } 889 890 #define ELF_HWCAP2 get_elf_hwcap2() 891 892 static uint32_t get_elf_hwcap2(void) 893 { 894 PowerPCCPU *cpu = POWERPC_CPU(thread_cpu); 895 uint32_t features = 0; 896 897 #define GET_FEATURE(flag, feature) \ 898 do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0) 899 #define GET_FEATURE2(flag, feature) \ 900 do { if (cpu->env.insns_flags2 & flag) { features |= feature; } } while (0) 901 902 GET_FEATURE(PPC_ISEL, QEMU_PPC_FEATURE2_HAS_ISEL); 903 GET_FEATURE2(PPC2_BCTAR_ISA207, QEMU_PPC_FEATURE2_HAS_TAR); 904 GET_FEATURE2((PPC2_BCTAR_ISA207 | PPC2_LSQ_ISA207 | PPC2_ALTIVEC_207 | 905 PPC2_ISA207S), QEMU_PPC_FEATURE2_ARCH_2_07 | 906 QEMU_PPC_FEATURE2_VEC_CRYPTO); 907 GET_FEATURE2(PPC2_ISA300, QEMU_PPC_FEATURE2_ARCH_3_00 | 908 QEMU_PPC_FEATURE2_DARN | QEMU_PPC_FEATURE2_HAS_IEEE128); 909 GET_FEATURE2(PPC2_ISA310, QEMU_PPC_FEATURE2_ARCH_3_1 | 910 QEMU_PPC_FEATURE2_MMA); 911 912 #undef GET_FEATURE 913 #undef GET_FEATURE2 914 915 return features; 916 } 917 918 /* 919 * The requirements here are: 920 * - keep the final alignment of sp (sp & 0xf) 921 * - make sure the 32-bit value at the first 16 byte aligned position of 922 * AUXV is greater than 16 for glibc compatibility. 923 * AT_IGNOREPPC is used for that. 924 * - for compatibility with glibc ARCH_DLINFO must always be defined on PPC, 925 * even if DLINFO_ARCH_ITEMS goes to zero or is undefined. 926 */ 927 #define DLINFO_ARCH_ITEMS 5 928 #define ARCH_DLINFO \ 929 do { \ 930 PowerPCCPU *cpu = POWERPC_CPU(thread_cpu); \ 931 /* \ 932 * Handle glibc compatibility: these magic entries must \ 933 * be at the lowest addresses in the final auxv. \ 934 */ \ 935 NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC); \ 936 NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC); \ 937 NEW_AUX_ENT(AT_DCACHEBSIZE, cpu->env.dcache_line_size); \ 938 NEW_AUX_ENT(AT_ICACHEBSIZE, cpu->env.icache_line_size); \ 939 NEW_AUX_ENT(AT_UCACHEBSIZE, 0); \ 940 } while (0) 941 942 static inline void init_thread(struct target_pt_regs *_regs, struct image_info *infop) 943 { 944 _regs->gpr[1] = infop->start_stack; 945 #if defined(TARGET_PPC64) 946 if (get_ppc64_abi(infop) < 2) { 947 uint64_t val; 948 get_user_u64(val, infop->entry + 8); 949 _regs->gpr[2] = val + infop->load_bias; 950 get_user_u64(val, infop->entry); 951 infop->entry = val + infop->load_bias; 952 } else { 953 _regs->gpr[12] = infop->entry; /* r12 set to global entry address */ 954 } 955 #endif 956 _regs->nip = infop->entry; 957 } 958 959 /* See linux kernel: arch/powerpc/include/asm/elf.h. */ 960 #define ELF_NREG 48 961 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 962 963 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUPPCState *env) 964 { 965 int i; 966 target_ulong ccr = 0; 967 968 for (i = 0; i < ARRAY_SIZE(env->gpr); i++) { 969 (*regs)[i] = tswapreg(env->gpr[i]); 970 } 971 972 (*regs)[32] = tswapreg(env->nip); 973 (*regs)[33] = tswapreg(env->msr); 974 (*regs)[35] = tswapreg(env->ctr); 975 (*regs)[36] = tswapreg(env->lr); 976 (*regs)[37] = tswapreg(cpu_read_xer(env)); 977 978 ccr = ppc_get_cr(env); 979 (*regs)[38] = tswapreg(ccr); 980 } 981 982 #define USE_ELF_CORE_DUMP 983 #define ELF_EXEC_PAGESIZE 4096 984 985 #endif 986 987 #ifdef TARGET_LOONGARCH64 988 989 #define ELF_START_MMAP 0x80000000 990 991 #define ELF_CLASS ELFCLASS64 992 #define ELF_ARCH EM_LOONGARCH 993 #define EXSTACK_DEFAULT true 994 995 #define elf_check_arch(x) ((x) == EM_LOONGARCH) 996 997 static inline void init_thread(struct target_pt_regs *regs, 998 struct image_info *infop) 999 { 1000 /*Set crmd PG,DA = 1,0 */ 1001 regs->csr.crmd = 2 << 3; 1002 regs->csr.era = infop->entry; 1003 regs->regs[3] = infop->start_stack; 1004 } 1005 1006 /* See linux kernel: arch/loongarch/include/asm/elf.h */ 1007 #define ELF_NREG 45 1008 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1009 1010 enum { 1011 TARGET_EF_R0 = 0, 1012 TARGET_EF_CSR_ERA = TARGET_EF_R0 + 33, 1013 TARGET_EF_CSR_BADV = TARGET_EF_R0 + 34, 1014 }; 1015 1016 static void elf_core_copy_regs(target_elf_gregset_t *regs, 1017 const CPULoongArchState *env) 1018 { 1019 int i; 1020 1021 (*regs)[TARGET_EF_R0] = 0; 1022 1023 for (i = 1; i < ARRAY_SIZE(env->gpr); i++) { 1024 (*regs)[TARGET_EF_R0 + i] = tswapreg(env->gpr[i]); 1025 } 1026 1027 (*regs)[TARGET_EF_CSR_ERA] = tswapreg(env->pc); 1028 (*regs)[TARGET_EF_CSR_BADV] = tswapreg(env->CSR_BADV); 1029 } 1030 1031 #define USE_ELF_CORE_DUMP 1032 #define ELF_EXEC_PAGESIZE 4096 1033 1034 #define ELF_HWCAP get_elf_hwcap() 1035 1036 /* See arch/loongarch/include/uapi/asm/hwcap.h */ 1037 enum { 1038 HWCAP_LOONGARCH_CPUCFG = (1 << 0), 1039 HWCAP_LOONGARCH_LAM = (1 << 1), 1040 HWCAP_LOONGARCH_UAL = (1 << 2), 1041 HWCAP_LOONGARCH_FPU = (1 << 3), 1042 HWCAP_LOONGARCH_LSX = (1 << 4), 1043 HWCAP_LOONGARCH_LASX = (1 << 5), 1044 HWCAP_LOONGARCH_CRC32 = (1 << 6), 1045 HWCAP_LOONGARCH_COMPLEX = (1 << 7), 1046 HWCAP_LOONGARCH_CRYPTO = (1 << 8), 1047 HWCAP_LOONGARCH_LVZ = (1 << 9), 1048 HWCAP_LOONGARCH_LBT_X86 = (1 << 10), 1049 HWCAP_LOONGARCH_LBT_ARM = (1 << 11), 1050 HWCAP_LOONGARCH_LBT_MIPS = (1 << 12), 1051 }; 1052 1053 static uint32_t get_elf_hwcap(void) 1054 { 1055 LoongArchCPU *cpu = LOONGARCH_CPU(thread_cpu); 1056 uint32_t hwcaps = 0; 1057 1058 hwcaps |= HWCAP_LOONGARCH_CRC32; 1059 1060 if (FIELD_EX32(cpu->env.cpucfg[1], CPUCFG1, UAL)) { 1061 hwcaps |= HWCAP_LOONGARCH_UAL; 1062 } 1063 1064 if (FIELD_EX32(cpu->env.cpucfg[2], CPUCFG2, FP)) { 1065 hwcaps |= HWCAP_LOONGARCH_FPU; 1066 } 1067 1068 if (FIELD_EX32(cpu->env.cpucfg[2], CPUCFG2, LAM)) { 1069 hwcaps |= HWCAP_LOONGARCH_LAM; 1070 } 1071 1072 return hwcaps; 1073 } 1074 1075 #define ELF_PLATFORM "loongarch" 1076 1077 #endif /* TARGET_LOONGARCH64 */ 1078 1079 #ifdef TARGET_MIPS 1080 1081 #define ELF_START_MMAP 0x80000000 1082 1083 #ifdef TARGET_MIPS64 1084 #define ELF_CLASS ELFCLASS64 1085 #else 1086 #define ELF_CLASS ELFCLASS32 1087 #endif 1088 #define ELF_ARCH EM_MIPS 1089 #define EXSTACK_DEFAULT true 1090 1091 #ifdef TARGET_ABI_MIPSN32 1092 #define elf_check_abi(x) ((x) & EF_MIPS_ABI2) 1093 #else 1094 #define elf_check_abi(x) (!((x) & EF_MIPS_ABI2)) 1095 #endif 1096 1097 #define ELF_BASE_PLATFORM get_elf_base_platform() 1098 1099 #define MATCH_PLATFORM_INSN(_flags, _base_platform) \ 1100 do { if ((cpu->env.insn_flags & (_flags)) == _flags) \ 1101 { return _base_platform; } } while (0) 1102 1103 static const char *get_elf_base_platform(void) 1104 { 1105 MIPSCPU *cpu = MIPS_CPU(thread_cpu); 1106 1107 /* 64 bit ISAs goes first */ 1108 MATCH_PLATFORM_INSN(CPU_MIPS64R6, "mips64r6"); 1109 MATCH_PLATFORM_INSN(CPU_MIPS64R5, "mips64r5"); 1110 MATCH_PLATFORM_INSN(CPU_MIPS64R2, "mips64r2"); 1111 MATCH_PLATFORM_INSN(CPU_MIPS64R1, "mips64"); 1112 MATCH_PLATFORM_INSN(CPU_MIPS5, "mips5"); 1113 MATCH_PLATFORM_INSN(CPU_MIPS4, "mips4"); 1114 MATCH_PLATFORM_INSN(CPU_MIPS3, "mips3"); 1115 1116 /* 32 bit ISAs */ 1117 MATCH_PLATFORM_INSN(CPU_MIPS32R6, "mips32r6"); 1118 MATCH_PLATFORM_INSN(CPU_MIPS32R5, "mips32r5"); 1119 MATCH_PLATFORM_INSN(CPU_MIPS32R2, "mips32r2"); 1120 MATCH_PLATFORM_INSN(CPU_MIPS32R1, "mips32"); 1121 MATCH_PLATFORM_INSN(CPU_MIPS2, "mips2"); 1122 1123 /* Fallback */ 1124 return "mips"; 1125 } 1126 #undef MATCH_PLATFORM_INSN 1127 1128 static inline void init_thread(struct target_pt_regs *regs, 1129 struct image_info *infop) 1130 { 1131 regs->cp0_status = 2 << CP0St_KSU; 1132 regs->cp0_epc = infop->entry; 1133 regs->regs[29] = infop->start_stack; 1134 } 1135 1136 /* See linux kernel: arch/mips/include/asm/elf.h. */ 1137 #define ELF_NREG 45 1138 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1139 1140 /* See linux kernel: arch/mips/include/asm/reg.h. */ 1141 enum { 1142 #ifdef TARGET_MIPS64 1143 TARGET_EF_R0 = 0, 1144 #else 1145 TARGET_EF_R0 = 6, 1146 #endif 1147 TARGET_EF_R26 = TARGET_EF_R0 + 26, 1148 TARGET_EF_R27 = TARGET_EF_R0 + 27, 1149 TARGET_EF_LO = TARGET_EF_R0 + 32, 1150 TARGET_EF_HI = TARGET_EF_R0 + 33, 1151 TARGET_EF_CP0_EPC = TARGET_EF_R0 + 34, 1152 TARGET_EF_CP0_BADVADDR = TARGET_EF_R0 + 35, 1153 TARGET_EF_CP0_STATUS = TARGET_EF_R0 + 36, 1154 TARGET_EF_CP0_CAUSE = TARGET_EF_R0 + 37 1155 }; 1156 1157 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */ 1158 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMIPSState *env) 1159 { 1160 int i; 1161 1162 for (i = 0; i < TARGET_EF_R0; i++) { 1163 (*regs)[i] = 0; 1164 } 1165 (*regs)[TARGET_EF_R0] = 0; 1166 1167 for (i = 1; i < ARRAY_SIZE(env->active_tc.gpr); i++) { 1168 (*regs)[TARGET_EF_R0 + i] = tswapreg(env->active_tc.gpr[i]); 1169 } 1170 1171 (*regs)[TARGET_EF_R26] = 0; 1172 (*regs)[TARGET_EF_R27] = 0; 1173 (*regs)[TARGET_EF_LO] = tswapreg(env->active_tc.LO[0]); 1174 (*regs)[TARGET_EF_HI] = tswapreg(env->active_tc.HI[0]); 1175 (*regs)[TARGET_EF_CP0_EPC] = tswapreg(env->active_tc.PC); 1176 (*regs)[TARGET_EF_CP0_BADVADDR] = tswapreg(env->CP0_BadVAddr); 1177 (*regs)[TARGET_EF_CP0_STATUS] = tswapreg(env->CP0_Status); 1178 (*regs)[TARGET_EF_CP0_CAUSE] = tswapreg(env->CP0_Cause); 1179 } 1180 1181 #define USE_ELF_CORE_DUMP 1182 #define ELF_EXEC_PAGESIZE 4096 1183 1184 /* See arch/mips/include/uapi/asm/hwcap.h. */ 1185 enum { 1186 HWCAP_MIPS_R6 = (1 << 0), 1187 HWCAP_MIPS_MSA = (1 << 1), 1188 HWCAP_MIPS_CRC32 = (1 << 2), 1189 HWCAP_MIPS_MIPS16 = (1 << 3), 1190 HWCAP_MIPS_MDMX = (1 << 4), 1191 HWCAP_MIPS_MIPS3D = (1 << 5), 1192 HWCAP_MIPS_SMARTMIPS = (1 << 6), 1193 HWCAP_MIPS_DSP = (1 << 7), 1194 HWCAP_MIPS_DSP2 = (1 << 8), 1195 HWCAP_MIPS_DSP3 = (1 << 9), 1196 HWCAP_MIPS_MIPS16E2 = (1 << 10), 1197 HWCAP_LOONGSON_MMI = (1 << 11), 1198 HWCAP_LOONGSON_EXT = (1 << 12), 1199 HWCAP_LOONGSON_EXT2 = (1 << 13), 1200 HWCAP_LOONGSON_CPUCFG = (1 << 14), 1201 }; 1202 1203 #define ELF_HWCAP get_elf_hwcap() 1204 1205 #define GET_FEATURE_INSN(_flag, _hwcap) \ 1206 do { if (cpu->env.insn_flags & (_flag)) { hwcaps |= _hwcap; } } while (0) 1207 1208 #define GET_FEATURE_REG_SET(_reg, _mask, _hwcap) \ 1209 do { if (cpu->env._reg & (_mask)) { hwcaps |= _hwcap; } } while (0) 1210 1211 #define GET_FEATURE_REG_EQU(_reg, _start, _length, _val, _hwcap) \ 1212 do { \ 1213 if (extract32(cpu->env._reg, (_start), (_length)) == (_val)) { \ 1214 hwcaps |= _hwcap; \ 1215 } \ 1216 } while (0) 1217 1218 static uint32_t get_elf_hwcap(void) 1219 { 1220 MIPSCPU *cpu = MIPS_CPU(thread_cpu); 1221 uint32_t hwcaps = 0; 1222 1223 GET_FEATURE_REG_EQU(CP0_Config0, CP0C0_AR, CP0C0_AR_LENGTH, 1224 2, HWCAP_MIPS_R6); 1225 GET_FEATURE_REG_SET(CP0_Config3, 1 << CP0C3_MSAP, HWCAP_MIPS_MSA); 1226 GET_FEATURE_INSN(ASE_LMMI, HWCAP_LOONGSON_MMI); 1227 GET_FEATURE_INSN(ASE_LEXT, HWCAP_LOONGSON_EXT); 1228 1229 return hwcaps; 1230 } 1231 1232 #undef GET_FEATURE_REG_EQU 1233 #undef GET_FEATURE_REG_SET 1234 #undef GET_FEATURE_INSN 1235 1236 #endif /* TARGET_MIPS */ 1237 1238 #ifdef TARGET_MICROBLAZE 1239 1240 #define ELF_START_MMAP 0x80000000 1241 1242 #define elf_check_arch(x) ( (x) == EM_MICROBLAZE || (x) == EM_MICROBLAZE_OLD) 1243 1244 #define ELF_CLASS ELFCLASS32 1245 #define ELF_ARCH EM_MICROBLAZE 1246 1247 static inline void init_thread(struct target_pt_regs *regs, 1248 struct image_info *infop) 1249 { 1250 regs->pc = infop->entry; 1251 regs->r1 = infop->start_stack; 1252 1253 } 1254 1255 #define ELF_EXEC_PAGESIZE 4096 1256 1257 #define USE_ELF_CORE_DUMP 1258 #define ELF_NREG 38 1259 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1260 1261 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */ 1262 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMBState *env) 1263 { 1264 int i, pos = 0; 1265 1266 for (i = 0; i < 32; i++) { 1267 (*regs)[pos++] = tswapreg(env->regs[i]); 1268 } 1269 1270 (*regs)[pos++] = tswapreg(env->pc); 1271 (*regs)[pos++] = tswapreg(mb_cpu_read_msr(env)); 1272 (*regs)[pos++] = 0; 1273 (*regs)[pos++] = tswapreg(env->ear); 1274 (*regs)[pos++] = 0; 1275 (*regs)[pos++] = tswapreg(env->esr); 1276 } 1277 1278 #endif /* TARGET_MICROBLAZE */ 1279 1280 #ifdef TARGET_NIOS2 1281 1282 #define ELF_START_MMAP 0x80000000 1283 1284 #define elf_check_arch(x) ((x) == EM_ALTERA_NIOS2) 1285 1286 #define ELF_CLASS ELFCLASS32 1287 #define ELF_ARCH EM_ALTERA_NIOS2 1288 1289 static void init_thread(struct target_pt_regs *regs, struct image_info *infop) 1290 { 1291 regs->ea = infop->entry; 1292 regs->sp = infop->start_stack; 1293 } 1294 1295 #define LO_COMMPAGE TARGET_PAGE_SIZE 1296 1297 static bool init_guest_commpage(void) 1298 { 1299 static const uint8_t kuser_page[4 + 2 * 64] = { 1300 /* __kuser_helper_version */ 1301 [0x00] = 0x02, 0x00, 0x00, 0x00, 1302 1303 /* __kuser_cmpxchg */ 1304 [0x04] = 0x3a, 0x6c, 0x3b, 0x00, /* trap 16 */ 1305 0x3a, 0x28, 0x00, 0xf8, /* ret */ 1306 1307 /* __kuser_sigtramp */ 1308 [0x44] = 0xc4, 0x22, 0x80, 0x00, /* movi r2, __NR_rt_sigreturn */ 1309 0x3a, 0x68, 0x3b, 0x00, /* trap 0 */ 1310 }; 1311 1312 void *want = g2h_untagged(LO_COMMPAGE & -qemu_host_page_size); 1313 void *addr = mmap(want, qemu_host_page_size, PROT_READ | PROT_WRITE, 1314 MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0); 1315 1316 if (addr == MAP_FAILED) { 1317 perror("Allocating guest commpage"); 1318 exit(EXIT_FAILURE); 1319 } 1320 if (addr != want) { 1321 return false; 1322 } 1323 1324 memcpy(addr, kuser_page, sizeof(kuser_page)); 1325 1326 if (mprotect(addr, qemu_host_page_size, PROT_READ)) { 1327 perror("Protecting guest commpage"); 1328 exit(EXIT_FAILURE); 1329 } 1330 1331 page_set_flags(LO_COMMPAGE, LO_COMMPAGE | ~TARGET_PAGE_MASK, 1332 PAGE_READ | PAGE_EXEC | PAGE_VALID); 1333 return true; 1334 } 1335 1336 #define ELF_EXEC_PAGESIZE 4096 1337 1338 #define USE_ELF_CORE_DUMP 1339 #define ELF_NREG 49 1340 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1341 1342 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */ 1343 static void elf_core_copy_regs(target_elf_gregset_t *regs, 1344 const CPUNios2State *env) 1345 { 1346 int i; 1347 1348 (*regs)[0] = -1; 1349 for (i = 1; i < 8; i++) /* r0-r7 */ 1350 (*regs)[i] = tswapreg(env->regs[i + 7]); 1351 1352 for (i = 8; i < 16; i++) /* r8-r15 */ 1353 (*regs)[i] = tswapreg(env->regs[i - 8]); 1354 1355 for (i = 16; i < 24; i++) /* r16-r23 */ 1356 (*regs)[i] = tswapreg(env->regs[i + 7]); 1357 (*regs)[24] = -1; /* R_ET */ 1358 (*regs)[25] = -1; /* R_BT */ 1359 (*regs)[26] = tswapreg(env->regs[R_GP]); 1360 (*regs)[27] = tswapreg(env->regs[R_SP]); 1361 (*regs)[28] = tswapreg(env->regs[R_FP]); 1362 (*regs)[29] = tswapreg(env->regs[R_EA]); 1363 (*regs)[30] = -1; /* R_SSTATUS */ 1364 (*regs)[31] = tswapreg(env->regs[R_RA]); 1365 1366 (*regs)[32] = tswapreg(env->pc); 1367 1368 (*regs)[33] = -1; /* R_STATUS */ 1369 (*regs)[34] = tswapreg(env->regs[CR_ESTATUS]); 1370 1371 for (i = 35; i < 49; i++) /* ... */ 1372 (*regs)[i] = -1; 1373 } 1374 1375 #endif /* TARGET_NIOS2 */ 1376 1377 #ifdef TARGET_OPENRISC 1378 1379 #define ELF_START_MMAP 0x08000000 1380 1381 #define ELF_ARCH EM_OPENRISC 1382 #define ELF_CLASS ELFCLASS32 1383 #define ELF_DATA ELFDATA2MSB 1384 1385 static inline void init_thread(struct target_pt_regs *regs, 1386 struct image_info *infop) 1387 { 1388 regs->pc = infop->entry; 1389 regs->gpr[1] = infop->start_stack; 1390 } 1391 1392 #define USE_ELF_CORE_DUMP 1393 #define ELF_EXEC_PAGESIZE 8192 1394 1395 /* See linux kernel arch/openrisc/include/asm/elf.h. */ 1396 #define ELF_NREG 34 /* gprs and pc, sr */ 1397 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1398 1399 static void elf_core_copy_regs(target_elf_gregset_t *regs, 1400 const CPUOpenRISCState *env) 1401 { 1402 int i; 1403 1404 for (i = 0; i < 32; i++) { 1405 (*regs)[i] = tswapreg(cpu_get_gpr(env, i)); 1406 } 1407 (*regs)[32] = tswapreg(env->pc); 1408 (*regs)[33] = tswapreg(cpu_get_sr(env)); 1409 } 1410 #define ELF_HWCAP 0 1411 #define ELF_PLATFORM NULL 1412 1413 #endif /* TARGET_OPENRISC */ 1414 1415 #ifdef TARGET_SH4 1416 1417 #define ELF_START_MMAP 0x80000000 1418 1419 #define ELF_CLASS ELFCLASS32 1420 #define ELF_ARCH EM_SH 1421 1422 static inline void init_thread(struct target_pt_regs *regs, 1423 struct image_info *infop) 1424 { 1425 /* Check other registers XXXXX */ 1426 regs->pc = infop->entry; 1427 regs->regs[15] = infop->start_stack; 1428 } 1429 1430 /* See linux kernel: arch/sh/include/asm/elf.h. */ 1431 #define ELF_NREG 23 1432 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1433 1434 /* See linux kernel: arch/sh/include/asm/ptrace.h. */ 1435 enum { 1436 TARGET_REG_PC = 16, 1437 TARGET_REG_PR = 17, 1438 TARGET_REG_SR = 18, 1439 TARGET_REG_GBR = 19, 1440 TARGET_REG_MACH = 20, 1441 TARGET_REG_MACL = 21, 1442 TARGET_REG_SYSCALL = 22 1443 }; 1444 1445 static inline void elf_core_copy_regs(target_elf_gregset_t *regs, 1446 const CPUSH4State *env) 1447 { 1448 int i; 1449 1450 for (i = 0; i < 16; i++) { 1451 (*regs)[i] = tswapreg(env->gregs[i]); 1452 } 1453 1454 (*regs)[TARGET_REG_PC] = tswapreg(env->pc); 1455 (*regs)[TARGET_REG_PR] = tswapreg(env->pr); 1456 (*regs)[TARGET_REG_SR] = tswapreg(env->sr); 1457 (*regs)[TARGET_REG_GBR] = tswapreg(env->gbr); 1458 (*regs)[TARGET_REG_MACH] = tswapreg(env->mach); 1459 (*regs)[TARGET_REG_MACL] = tswapreg(env->macl); 1460 (*regs)[TARGET_REG_SYSCALL] = 0; /* FIXME */ 1461 } 1462 1463 #define USE_ELF_CORE_DUMP 1464 #define ELF_EXEC_PAGESIZE 4096 1465 1466 enum { 1467 SH_CPU_HAS_FPU = 0x0001, /* Hardware FPU support */ 1468 SH_CPU_HAS_P2_FLUSH_BUG = 0x0002, /* Need to flush the cache in P2 area */ 1469 SH_CPU_HAS_MMU_PAGE_ASSOC = 0x0004, /* SH3: TLB way selection bit support */ 1470 SH_CPU_HAS_DSP = 0x0008, /* SH-DSP: DSP support */ 1471 SH_CPU_HAS_PERF_COUNTER = 0x0010, /* Hardware performance counters */ 1472 SH_CPU_HAS_PTEA = 0x0020, /* PTEA register */ 1473 SH_CPU_HAS_LLSC = 0x0040, /* movli.l/movco.l */ 1474 SH_CPU_HAS_L2_CACHE = 0x0080, /* Secondary cache / URAM */ 1475 SH_CPU_HAS_OP32 = 0x0100, /* 32-bit instruction support */ 1476 SH_CPU_HAS_PTEAEX = 0x0200, /* PTE ASID Extension support */ 1477 }; 1478 1479 #define ELF_HWCAP get_elf_hwcap() 1480 1481 static uint32_t get_elf_hwcap(void) 1482 { 1483 SuperHCPU *cpu = SUPERH_CPU(thread_cpu); 1484 uint32_t hwcap = 0; 1485 1486 hwcap |= SH_CPU_HAS_FPU; 1487 1488 if (cpu->env.features & SH_FEATURE_SH4A) { 1489 hwcap |= SH_CPU_HAS_LLSC; 1490 } 1491 1492 return hwcap; 1493 } 1494 1495 #endif 1496 1497 #ifdef TARGET_CRIS 1498 1499 #define ELF_START_MMAP 0x80000000 1500 1501 #define ELF_CLASS ELFCLASS32 1502 #define ELF_ARCH EM_CRIS 1503 1504 static inline void init_thread(struct target_pt_regs *regs, 1505 struct image_info *infop) 1506 { 1507 regs->erp = infop->entry; 1508 } 1509 1510 #define ELF_EXEC_PAGESIZE 8192 1511 1512 #endif 1513 1514 #ifdef TARGET_M68K 1515 1516 #define ELF_START_MMAP 0x80000000 1517 1518 #define ELF_CLASS ELFCLASS32 1519 #define ELF_ARCH EM_68K 1520 1521 /* ??? Does this need to do anything? 1522 #define ELF_PLAT_INIT(_r) */ 1523 1524 static inline void init_thread(struct target_pt_regs *regs, 1525 struct image_info *infop) 1526 { 1527 regs->usp = infop->start_stack; 1528 regs->sr = 0; 1529 regs->pc = infop->entry; 1530 } 1531 1532 /* See linux kernel: arch/m68k/include/asm/elf.h. */ 1533 #define ELF_NREG 20 1534 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1535 1536 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUM68KState *env) 1537 { 1538 (*regs)[0] = tswapreg(env->dregs[1]); 1539 (*regs)[1] = tswapreg(env->dregs[2]); 1540 (*regs)[2] = tswapreg(env->dregs[3]); 1541 (*regs)[3] = tswapreg(env->dregs[4]); 1542 (*regs)[4] = tswapreg(env->dregs[5]); 1543 (*regs)[5] = tswapreg(env->dregs[6]); 1544 (*regs)[6] = tswapreg(env->dregs[7]); 1545 (*regs)[7] = tswapreg(env->aregs[0]); 1546 (*regs)[8] = tswapreg(env->aregs[1]); 1547 (*regs)[9] = tswapreg(env->aregs[2]); 1548 (*regs)[10] = tswapreg(env->aregs[3]); 1549 (*regs)[11] = tswapreg(env->aregs[4]); 1550 (*regs)[12] = tswapreg(env->aregs[5]); 1551 (*regs)[13] = tswapreg(env->aregs[6]); 1552 (*regs)[14] = tswapreg(env->dregs[0]); 1553 (*regs)[15] = tswapreg(env->aregs[7]); 1554 (*regs)[16] = tswapreg(env->dregs[0]); /* FIXME: orig_d0 */ 1555 (*regs)[17] = tswapreg(env->sr); 1556 (*regs)[18] = tswapreg(env->pc); 1557 (*regs)[19] = 0; /* FIXME: regs->format | regs->vector */ 1558 } 1559 1560 #define USE_ELF_CORE_DUMP 1561 #define ELF_EXEC_PAGESIZE 8192 1562 1563 #endif 1564 1565 #ifdef TARGET_ALPHA 1566 1567 #define ELF_START_MMAP (0x30000000000ULL) 1568 1569 #define ELF_CLASS ELFCLASS64 1570 #define ELF_ARCH EM_ALPHA 1571 1572 static inline void init_thread(struct target_pt_regs *regs, 1573 struct image_info *infop) 1574 { 1575 regs->pc = infop->entry; 1576 regs->ps = 8; 1577 regs->usp = infop->start_stack; 1578 } 1579 1580 #define ELF_EXEC_PAGESIZE 8192 1581 1582 #endif /* TARGET_ALPHA */ 1583 1584 #ifdef TARGET_S390X 1585 1586 #define ELF_START_MMAP (0x20000000000ULL) 1587 1588 #define ELF_CLASS ELFCLASS64 1589 #define ELF_DATA ELFDATA2MSB 1590 #define ELF_ARCH EM_S390 1591 1592 #include "elf.h" 1593 1594 #define ELF_HWCAP get_elf_hwcap() 1595 1596 #define GET_FEATURE(_feat, _hwcap) \ 1597 do { if (s390_has_feat(_feat)) { hwcap |= _hwcap; } } while (0) 1598 1599 uint32_t get_elf_hwcap(void) 1600 { 1601 /* 1602 * Let's assume we always have esan3 and zarch. 1603 * 31-bit processes can use 64-bit registers (high gprs). 1604 */ 1605 uint32_t hwcap = HWCAP_S390_ESAN3 | HWCAP_S390_ZARCH | HWCAP_S390_HIGH_GPRS; 1606 1607 GET_FEATURE(S390_FEAT_STFLE, HWCAP_S390_STFLE); 1608 GET_FEATURE(S390_FEAT_MSA, HWCAP_S390_MSA); 1609 GET_FEATURE(S390_FEAT_LONG_DISPLACEMENT, HWCAP_S390_LDISP); 1610 GET_FEATURE(S390_FEAT_EXTENDED_IMMEDIATE, HWCAP_S390_EIMM); 1611 if (s390_has_feat(S390_FEAT_EXTENDED_TRANSLATION_3) && 1612 s390_has_feat(S390_FEAT_ETF3_ENH)) { 1613 hwcap |= HWCAP_S390_ETF3EH; 1614 } 1615 GET_FEATURE(S390_FEAT_VECTOR, HWCAP_S390_VXRS); 1616 GET_FEATURE(S390_FEAT_VECTOR_ENH, HWCAP_S390_VXRS_EXT); 1617 1618 return hwcap; 1619 } 1620 1621 const char *elf_hwcap_str(uint32_t bit) 1622 { 1623 static const char *hwcap_str[] = { 1624 [HWCAP_S390_NR_ESAN3] = "esan3", 1625 [HWCAP_S390_NR_ZARCH] = "zarch", 1626 [HWCAP_S390_NR_STFLE] = "stfle", 1627 [HWCAP_S390_NR_MSA] = "msa", 1628 [HWCAP_S390_NR_LDISP] = "ldisp", 1629 [HWCAP_S390_NR_EIMM] = "eimm", 1630 [HWCAP_S390_NR_DFP] = "dfp", 1631 [HWCAP_S390_NR_HPAGE] = "edat", 1632 [HWCAP_S390_NR_ETF3EH] = "etf3eh", 1633 [HWCAP_S390_NR_HIGH_GPRS] = "highgprs", 1634 [HWCAP_S390_NR_TE] = "te", 1635 [HWCAP_S390_NR_VXRS] = "vx", 1636 [HWCAP_S390_NR_VXRS_BCD] = "vxd", 1637 [HWCAP_S390_NR_VXRS_EXT] = "vxe", 1638 [HWCAP_S390_NR_GS] = "gs", 1639 [HWCAP_S390_NR_VXRS_EXT2] = "vxe2", 1640 [HWCAP_S390_NR_VXRS_PDE] = "vxp", 1641 [HWCAP_S390_NR_SORT] = "sort", 1642 [HWCAP_S390_NR_DFLT] = "dflt", 1643 [HWCAP_S390_NR_NNPA] = "nnpa", 1644 [HWCAP_S390_NR_PCI_MIO] = "pcimio", 1645 [HWCAP_S390_NR_SIE] = "sie", 1646 }; 1647 1648 return bit < ARRAY_SIZE(hwcap_str) ? hwcap_str[bit] : NULL; 1649 } 1650 1651 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop) 1652 { 1653 regs->psw.addr = infop->entry; 1654 regs->psw.mask = PSW_MASK_DAT | PSW_MASK_IO | PSW_MASK_EXT | \ 1655 PSW_MASK_MCHECK | PSW_MASK_PSTATE | PSW_MASK_64 | \ 1656 PSW_MASK_32; 1657 regs->gprs[15] = infop->start_stack; 1658 } 1659 1660 /* See linux kernel: arch/s390/include/uapi/asm/ptrace.h (s390_regs). */ 1661 #define ELF_NREG 27 1662 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1663 1664 enum { 1665 TARGET_REG_PSWM = 0, 1666 TARGET_REG_PSWA = 1, 1667 TARGET_REG_GPRS = 2, 1668 TARGET_REG_ARS = 18, 1669 TARGET_REG_ORIG_R2 = 26, 1670 }; 1671 1672 static void elf_core_copy_regs(target_elf_gregset_t *regs, 1673 const CPUS390XState *env) 1674 { 1675 int i; 1676 uint32_t *aregs; 1677 1678 (*regs)[TARGET_REG_PSWM] = tswapreg(env->psw.mask); 1679 (*regs)[TARGET_REG_PSWA] = tswapreg(env->psw.addr); 1680 for (i = 0; i < 16; i++) { 1681 (*regs)[TARGET_REG_GPRS + i] = tswapreg(env->regs[i]); 1682 } 1683 aregs = (uint32_t *)&((*regs)[TARGET_REG_ARS]); 1684 for (i = 0; i < 16; i++) { 1685 aregs[i] = tswap32(env->aregs[i]); 1686 } 1687 (*regs)[TARGET_REG_ORIG_R2] = 0; 1688 } 1689 1690 #define USE_ELF_CORE_DUMP 1691 #define ELF_EXEC_PAGESIZE 4096 1692 1693 #endif /* TARGET_S390X */ 1694 1695 #ifdef TARGET_RISCV 1696 1697 #define ELF_START_MMAP 0x80000000 1698 #define ELF_ARCH EM_RISCV 1699 1700 #ifdef TARGET_RISCV32 1701 #define ELF_CLASS ELFCLASS32 1702 #else 1703 #define ELF_CLASS ELFCLASS64 1704 #endif 1705 1706 #define ELF_HWCAP get_elf_hwcap() 1707 1708 static uint32_t get_elf_hwcap(void) 1709 { 1710 #define MISA_BIT(EXT) (1 << (EXT - 'A')) 1711 RISCVCPU *cpu = RISCV_CPU(thread_cpu); 1712 uint32_t mask = MISA_BIT('I') | MISA_BIT('M') | MISA_BIT('A') 1713 | MISA_BIT('F') | MISA_BIT('D') | MISA_BIT('C'); 1714 1715 return cpu->env.misa_ext & mask; 1716 #undef MISA_BIT 1717 } 1718 1719 static inline void init_thread(struct target_pt_regs *regs, 1720 struct image_info *infop) 1721 { 1722 regs->sepc = infop->entry; 1723 regs->sp = infop->start_stack; 1724 } 1725 1726 #define ELF_EXEC_PAGESIZE 4096 1727 1728 #endif /* TARGET_RISCV */ 1729 1730 #ifdef TARGET_HPPA 1731 1732 #define ELF_START_MMAP 0x80000000 1733 #define ELF_CLASS ELFCLASS32 1734 #define ELF_ARCH EM_PARISC 1735 #define ELF_PLATFORM "PARISC" 1736 #define STACK_GROWS_DOWN 0 1737 #define STACK_ALIGNMENT 64 1738 1739 static inline void init_thread(struct target_pt_regs *regs, 1740 struct image_info *infop) 1741 { 1742 regs->iaoq[0] = infop->entry; 1743 regs->iaoq[1] = infop->entry + 4; 1744 regs->gr[23] = 0; 1745 regs->gr[24] = infop->argv; 1746 regs->gr[25] = infop->argc; 1747 /* The top-of-stack contains a linkage buffer. */ 1748 regs->gr[30] = infop->start_stack + 64; 1749 regs->gr[31] = infop->entry; 1750 } 1751 1752 #define LO_COMMPAGE 0 1753 1754 static bool init_guest_commpage(void) 1755 { 1756 void *want = g2h_untagged(LO_COMMPAGE); 1757 void *addr = mmap(want, qemu_host_page_size, PROT_NONE, 1758 MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0); 1759 1760 if (addr == MAP_FAILED) { 1761 perror("Allocating guest commpage"); 1762 exit(EXIT_FAILURE); 1763 } 1764 if (addr != want) { 1765 return false; 1766 } 1767 1768 /* 1769 * On Linux, page zero is normally marked execute only + gateway. 1770 * Normal read or write is supposed to fail (thus PROT_NONE above), 1771 * but specific offsets have kernel code mapped to raise permissions 1772 * and implement syscalls. Here, simply mark the page executable. 1773 * Special case the entry points during translation (see do_page_zero). 1774 */ 1775 page_set_flags(LO_COMMPAGE, LO_COMMPAGE | ~TARGET_PAGE_MASK, 1776 PAGE_EXEC | PAGE_VALID); 1777 return true; 1778 } 1779 1780 #endif /* TARGET_HPPA */ 1781 1782 #ifdef TARGET_XTENSA 1783 1784 #define ELF_START_MMAP 0x20000000 1785 1786 #define ELF_CLASS ELFCLASS32 1787 #define ELF_ARCH EM_XTENSA 1788 1789 static inline void init_thread(struct target_pt_regs *regs, 1790 struct image_info *infop) 1791 { 1792 regs->windowbase = 0; 1793 regs->windowstart = 1; 1794 regs->areg[1] = infop->start_stack; 1795 regs->pc = infop->entry; 1796 if (info_is_fdpic(infop)) { 1797 regs->areg[4] = infop->loadmap_addr; 1798 regs->areg[5] = infop->interpreter_loadmap_addr; 1799 if (infop->interpreter_loadmap_addr) { 1800 regs->areg[6] = infop->interpreter_pt_dynamic_addr; 1801 } else { 1802 regs->areg[6] = infop->pt_dynamic_addr; 1803 } 1804 } 1805 } 1806 1807 /* See linux kernel: arch/xtensa/include/asm/elf.h. */ 1808 #define ELF_NREG 128 1809 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG]; 1810 1811 enum { 1812 TARGET_REG_PC, 1813 TARGET_REG_PS, 1814 TARGET_REG_LBEG, 1815 TARGET_REG_LEND, 1816 TARGET_REG_LCOUNT, 1817 TARGET_REG_SAR, 1818 TARGET_REG_WINDOWSTART, 1819 TARGET_REG_WINDOWBASE, 1820 TARGET_REG_THREADPTR, 1821 TARGET_REG_AR0 = 64, 1822 }; 1823 1824 static void elf_core_copy_regs(target_elf_gregset_t *regs, 1825 const CPUXtensaState *env) 1826 { 1827 unsigned i; 1828 1829 (*regs)[TARGET_REG_PC] = tswapreg(env->pc); 1830 (*regs)[TARGET_REG_PS] = tswapreg(env->sregs[PS] & ~PS_EXCM); 1831 (*regs)[TARGET_REG_LBEG] = tswapreg(env->sregs[LBEG]); 1832 (*regs)[TARGET_REG_LEND] = tswapreg(env->sregs[LEND]); 1833 (*regs)[TARGET_REG_LCOUNT] = tswapreg(env->sregs[LCOUNT]); 1834 (*regs)[TARGET_REG_SAR] = tswapreg(env->sregs[SAR]); 1835 (*regs)[TARGET_REG_WINDOWSTART] = tswapreg(env->sregs[WINDOW_START]); 1836 (*regs)[TARGET_REG_WINDOWBASE] = tswapreg(env->sregs[WINDOW_BASE]); 1837 (*regs)[TARGET_REG_THREADPTR] = tswapreg(env->uregs[THREADPTR]); 1838 xtensa_sync_phys_from_window((CPUXtensaState *)env); 1839 for (i = 0; i < env->config->nareg; ++i) { 1840 (*regs)[TARGET_REG_AR0 + i] = tswapreg(env->phys_regs[i]); 1841 } 1842 } 1843 1844 #define USE_ELF_CORE_DUMP 1845 #define ELF_EXEC_PAGESIZE 4096 1846 1847 #endif /* TARGET_XTENSA */ 1848 1849 #ifdef TARGET_HEXAGON 1850 1851 #define ELF_START_MMAP 0x20000000 1852 1853 #define ELF_CLASS ELFCLASS32 1854 #define ELF_ARCH EM_HEXAGON 1855 1856 static inline void init_thread(struct target_pt_regs *regs, 1857 struct image_info *infop) 1858 { 1859 regs->sepc = infop->entry; 1860 regs->sp = infop->start_stack; 1861 } 1862 1863 #endif /* TARGET_HEXAGON */ 1864 1865 #ifndef ELF_BASE_PLATFORM 1866 #define ELF_BASE_PLATFORM (NULL) 1867 #endif 1868 1869 #ifndef ELF_PLATFORM 1870 #define ELF_PLATFORM (NULL) 1871 #endif 1872 1873 #ifndef ELF_MACHINE 1874 #define ELF_MACHINE ELF_ARCH 1875 #endif 1876 1877 #ifndef elf_check_arch 1878 #define elf_check_arch(x) ((x) == ELF_ARCH) 1879 #endif 1880 1881 #ifndef elf_check_abi 1882 #define elf_check_abi(x) (1) 1883 #endif 1884 1885 #ifndef ELF_HWCAP 1886 #define ELF_HWCAP 0 1887 #endif 1888 1889 #ifndef STACK_GROWS_DOWN 1890 #define STACK_GROWS_DOWN 1 1891 #endif 1892 1893 #ifndef STACK_ALIGNMENT 1894 #define STACK_ALIGNMENT 16 1895 #endif 1896 1897 #ifdef TARGET_ABI32 1898 #undef ELF_CLASS 1899 #define ELF_CLASS ELFCLASS32 1900 #undef bswaptls 1901 #define bswaptls(ptr) bswap32s(ptr) 1902 #endif 1903 1904 #ifndef EXSTACK_DEFAULT 1905 #define EXSTACK_DEFAULT false 1906 #endif 1907 1908 #include "elf.h" 1909 1910 /* We must delay the following stanzas until after "elf.h". */ 1911 #if defined(TARGET_AARCH64) 1912 1913 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz, 1914 const uint32_t *data, 1915 struct image_info *info, 1916 Error **errp) 1917 { 1918 if (pr_type == GNU_PROPERTY_AARCH64_FEATURE_1_AND) { 1919 if (pr_datasz != sizeof(uint32_t)) { 1920 error_setg(errp, "Ill-formed GNU_PROPERTY_AARCH64_FEATURE_1_AND"); 1921 return false; 1922 } 1923 /* We will extract GNU_PROPERTY_AARCH64_FEATURE_1_BTI later. */ 1924 info->note_flags = *data; 1925 } 1926 return true; 1927 } 1928 #define ARCH_USE_GNU_PROPERTY 1 1929 1930 #else 1931 1932 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz, 1933 const uint32_t *data, 1934 struct image_info *info, 1935 Error **errp) 1936 { 1937 g_assert_not_reached(); 1938 } 1939 #define ARCH_USE_GNU_PROPERTY 0 1940 1941 #endif 1942 1943 struct exec 1944 { 1945 unsigned int a_info; /* Use macros N_MAGIC, etc for access */ 1946 unsigned int a_text; /* length of text, in bytes */ 1947 unsigned int a_data; /* length of data, in bytes */ 1948 unsigned int a_bss; /* length of uninitialized data area, in bytes */ 1949 unsigned int a_syms; /* length of symbol table data in file, in bytes */ 1950 unsigned int a_entry; /* start address */ 1951 unsigned int a_trsize; /* length of relocation info for text, in bytes */ 1952 unsigned int a_drsize; /* length of relocation info for data, in bytes */ 1953 }; 1954 1955 1956 #define N_MAGIC(exec) ((exec).a_info & 0xffff) 1957 #define OMAGIC 0407 1958 #define NMAGIC 0410 1959 #define ZMAGIC 0413 1960 #define QMAGIC 0314 1961 1962 /* Necessary parameters */ 1963 #define TARGET_ELF_EXEC_PAGESIZE \ 1964 (((eppnt->p_align & ~qemu_host_page_mask) != 0) ? \ 1965 TARGET_PAGE_SIZE : MAX(qemu_host_page_size, TARGET_PAGE_SIZE)) 1966 #define TARGET_ELF_PAGELENGTH(_v) ROUND_UP((_v), TARGET_ELF_EXEC_PAGESIZE) 1967 #define TARGET_ELF_PAGESTART(_v) ((_v) & \ 1968 ~(abi_ulong)(TARGET_ELF_EXEC_PAGESIZE-1)) 1969 #define TARGET_ELF_PAGEOFFSET(_v) ((_v) & (TARGET_ELF_EXEC_PAGESIZE-1)) 1970 1971 #define DLINFO_ITEMS 16 1972 1973 static inline void memcpy_fromfs(void * to, const void * from, unsigned long n) 1974 { 1975 memcpy(to, from, n); 1976 } 1977 1978 #ifdef BSWAP_NEEDED 1979 static void bswap_ehdr(struct elfhdr *ehdr) 1980 { 1981 bswap16s(&ehdr->e_type); /* Object file type */ 1982 bswap16s(&ehdr->e_machine); /* Architecture */ 1983 bswap32s(&ehdr->e_version); /* Object file version */ 1984 bswaptls(&ehdr->e_entry); /* Entry point virtual address */ 1985 bswaptls(&ehdr->e_phoff); /* Program header table file offset */ 1986 bswaptls(&ehdr->e_shoff); /* Section header table file offset */ 1987 bswap32s(&ehdr->e_flags); /* Processor-specific flags */ 1988 bswap16s(&ehdr->e_ehsize); /* ELF header size in bytes */ 1989 bswap16s(&ehdr->e_phentsize); /* Program header table entry size */ 1990 bswap16s(&ehdr->e_phnum); /* Program header table entry count */ 1991 bswap16s(&ehdr->e_shentsize); /* Section header table entry size */ 1992 bswap16s(&ehdr->e_shnum); /* Section header table entry count */ 1993 bswap16s(&ehdr->e_shstrndx); /* Section header string table index */ 1994 } 1995 1996 static void bswap_phdr(struct elf_phdr *phdr, int phnum) 1997 { 1998 int i; 1999 for (i = 0; i < phnum; ++i, ++phdr) { 2000 bswap32s(&phdr->p_type); /* Segment type */ 2001 bswap32s(&phdr->p_flags); /* Segment flags */ 2002 bswaptls(&phdr->p_offset); /* Segment file offset */ 2003 bswaptls(&phdr->p_vaddr); /* Segment virtual address */ 2004 bswaptls(&phdr->p_paddr); /* Segment physical address */ 2005 bswaptls(&phdr->p_filesz); /* Segment size in file */ 2006 bswaptls(&phdr->p_memsz); /* Segment size in memory */ 2007 bswaptls(&phdr->p_align); /* Segment alignment */ 2008 } 2009 } 2010 2011 static void bswap_shdr(struct elf_shdr *shdr, int shnum) 2012 { 2013 int i; 2014 for (i = 0; i < shnum; ++i, ++shdr) { 2015 bswap32s(&shdr->sh_name); 2016 bswap32s(&shdr->sh_type); 2017 bswaptls(&shdr->sh_flags); 2018 bswaptls(&shdr->sh_addr); 2019 bswaptls(&shdr->sh_offset); 2020 bswaptls(&shdr->sh_size); 2021 bswap32s(&shdr->sh_link); 2022 bswap32s(&shdr->sh_info); 2023 bswaptls(&shdr->sh_addralign); 2024 bswaptls(&shdr->sh_entsize); 2025 } 2026 } 2027 2028 static void bswap_sym(struct elf_sym *sym) 2029 { 2030 bswap32s(&sym->st_name); 2031 bswaptls(&sym->st_value); 2032 bswaptls(&sym->st_size); 2033 bswap16s(&sym->st_shndx); 2034 } 2035 2036 #ifdef TARGET_MIPS 2037 static void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags) 2038 { 2039 bswap16s(&abiflags->version); 2040 bswap32s(&abiflags->ases); 2041 bswap32s(&abiflags->isa_ext); 2042 bswap32s(&abiflags->flags1); 2043 bswap32s(&abiflags->flags2); 2044 } 2045 #endif 2046 #else 2047 static inline void bswap_ehdr(struct elfhdr *ehdr) { } 2048 static inline void bswap_phdr(struct elf_phdr *phdr, int phnum) { } 2049 static inline void bswap_shdr(struct elf_shdr *shdr, int shnum) { } 2050 static inline void bswap_sym(struct elf_sym *sym) { } 2051 #ifdef TARGET_MIPS 2052 static inline void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags) { } 2053 #endif 2054 #endif 2055 2056 #ifdef USE_ELF_CORE_DUMP 2057 static int elf_core_dump(int, const CPUArchState *); 2058 #endif /* USE_ELF_CORE_DUMP */ 2059 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias); 2060 2061 /* Verify the portions of EHDR within E_IDENT for the target. 2062 This can be performed before bswapping the entire header. */ 2063 static bool elf_check_ident(struct elfhdr *ehdr) 2064 { 2065 return (ehdr->e_ident[EI_MAG0] == ELFMAG0 2066 && ehdr->e_ident[EI_MAG1] == ELFMAG1 2067 && ehdr->e_ident[EI_MAG2] == ELFMAG2 2068 && ehdr->e_ident[EI_MAG3] == ELFMAG3 2069 && ehdr->e_ident[EI_CLASS] == ELF_CLASS 2070 && ehdr->e_ident[EI_DATA] == ELF_DATA 2071 && ehdr->e_ident[EI_VERSION] == EV_CURRENT); 2072 } 2073 2074 /* Verify the portions of EHDR outside of E_IDENT for the target. 2075 This has to wait until after bswapping the header. */ 2076 static bool elf_check_ehdr(struct elfhdr *ehdr) 2077 { 2078 return (elf_check_arch(ehdr->e_machine) 2079 && elf_check_abi(ehdr->e_flags) 2080 && ehdr->e_ehsize == sizeof(struct elfhdr) 2081 && ehdr->e_phentsize == sizeof(struct elf_phdr) 2082 && (ehdr->e_type == ET_EXEC || ehdr->e_type == ET_DYN)); 2083 } 2084 2085 /* 2086 * 'copy_elf_strings()' copies argument/envelope strings from user 2087 * memory to free pages in kernel mem. These are in a format ready 2088 * to be put directly into the top of new user memory. 2089 * 2090 */ 2091 static abi_ulong copy_elf_strings(int argc, char **argv, char *scratch, 2092 abi_ulong p, abi_ulong stack_limit) 2093 { 2094 char *tmp; 2095 int len, i; 2096 abi_ulong top = p; 2097 2098 if (!p) { 2099 return 0; /* bullet-proofing */ 2100 } 2101 2102 if (STACK_GROWS_DOWN) { 2103 int offset = ((p - 1) % TARGET_PAGE_SIZE) + 1; 2104 for (i = argc - 1; i >= 0; --i) { 2105 tmp = argv[i]; 2106 if (!tmp) { 2107 fprintf(stderr, "VFS: argc is wrong"); 2108 exit(-1); 2109 } 2110 len = strlen(tmp) + 1; 2111 tmp += len; 2112 2113 if (len > (p - stack_limit)) { 2114 return 0; 2115 } 2116 while (len) { 2117 int bytes_to_copy = (len > offset) ? offset : len; 2118 tmp -= bytes_to_copy; 2119 p -= bytes_to_copy; 2120 offset -= bytes_to_copy; 2121 len -= bytes_to_copy; 2122 2123 memcpy_fromfs(scratch + offset, tmp, bytes_to_copy); 2124 2125 if (offset == 0) { 2126 memcpy_to_target(p, scratch, top - p); 2127 top = p; 2128 offset = TARGET_PAGE_SIZE; 2129 } 2130 } 2131 } 2132 if (p != top) { 2133 memcpy_to_target(p, scratch + offset, top - p); 2134 } 2135 } else { 2136 int remaining = TARGET_PAGE_SIZE - (p % TARGET_PAGE_SIZE); 2137 for (i = 0; i < argc; ++i) { 2138 tmp = argv[i]; 2139 if (!tmp) { 2140 fprintf(stderr, "VFS: argc is wrong"); 2141 exit(-1); 2142 } 2143 len = strlen(tmp) + 1; 2144 if (len > (stack_limit - p)) { 2145 return 0; 2146 } 2147 while (len) { 2148 int bytes_to_copy = (len > remaining) ? remaining : len; 2149 2150 memcpy_fromfs(scratch + (p - top), tmp, bytes_to_copy); 2151 2152 tmp += bytes_to_copy; 2153 remaining -= bytes_to_copy; 2154 p += bytes_to_copy; 2155 len -= bytes_to_copy; 2156 2157 if (remaining == 0) { 2158 memcpy_to_target(top, scratch, p - top); 2159 top = p; 2160 remaining = TARGET_PAGE_SIZE; 2161 } 2162 } 2163 } 2164 if (p != top) { 2165 memcpy_to_target(top, scratch, p - top); 2166 } 2167 } 2168 2169 return p; 2170 } 2171 2172 /* Older linux kernels provide up to MAX_ARG_PAGES (default: 32) of 2173 * argument/environment space. Newer kernels (>2.6.33) allow more, 2174 * dependent on stack size, but guarantee at least 32 pages for 2175 * backwards compatibility. 2176 */ 2177 #define STACK_LOWER_LIMIT (32 * TARGET_PAGE_SIZE) 2178 2179 static abi_ulong setup_arg_pages(struct linux_binprm *bprm, 2180 struct image_info *info) 2181 { 2182 abi_ulong size, error, guard; 2183 int prot; 2184 2185 size = guest_stack_size; 2186 if (size < STACK_LOWER_LIMIT) { 2187 size = STACK_LOWER_LIMIT; 2188 } 2189 2190 if (STACK_GROWS_DOWN) { 2191 guard = TARGET_PAGE_SIZE; 2192 if (guard < qemu_real_host_page_size()) { 2193 guard = qemu_real_host_page_size(); 2194 } 2195 } else { 2196 /* no guard page for hppa target where stack grows upwards. */ 2197 guard = 0; 2198 } 2199 2200 prot = PROT_READ | PROT_WRITE; 2201 if (info->exec_stack) { 2202 prot |= PROT_EXEC; 2203 } 2204 error = target_mmap(0, size + guard, prot, 2205 MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); 2206 if (error == -1) { 2207 perror("mmap stack"); 2208 exit(-1); 2209 } 2210 2211 /* We reserve one extra page at the top of the stack as guard. */ 2212 if (STACK_GROWS_DOWN) { 2213 target_mprotect(error, guard, PROT_NONE); 2214 info->stack_limit = error + guard; 2215 return info->stack_limit + size - sizeof(void *); 2216 } else { 2217 info->stack_limit = error + size; 2218 return error; 2219 } 2220 } 2221 2222 /* Map and zero the bss. We need to explicitly zero any fractional pages 2223 after the data section (i.e. bss). */ 2224 static void zero_bss(abi_ulong elf_bss, abi_ulong last_bss, int prot) 2225 { 2226 uintptr_t host_start, host_map_start, host_end; 2227 2228 last_bss = TARGET_PAGE_ALIGN(last_bss); 2229 2230 /* ??? There is confusion between qemu_real_host_page_size and 2231 qemu_host_page_size here and elsewhere in target_mmap, which 2232 may lead to the end of the data section mapping from the file 2233 not being mapped. At least there was an explicit test and 2234 comment for that here, suggesting that "the file size must 2235 be known". The comment probably pre-dates the introduction 2236 of the fstat system call in target_mmap which does in fact 2237 find out the size. What isn't clear is if the workaround 2238 here is still actually needed. For now, continue with it, 2239 but merge it with the "normal" mmap that would allocate the bss. */ 2240 2241 host_start = (uintptr_t) g2h_untagged(elf_bss); 2242 host_end = (uintptr_t) g2h_untagged(last_bss); 2243 host_map_start = REAL_HOST_PAGE_ALIGN(host_start); 2244 2245 if (host_map_start < host_end) { 2246 void *p = mmap((void *)host_map_start, host_end - host_map_start, 2247 prot, MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); 2248 if (p == MAP_FAILED) { 2249 perror("cannot mmap brk"); 2250 exit(-1); 2251 } 2252 } 2253 2254 /* Ensure that the bss page(s) are valid */ 2255 if ((page_get_flags(last_bss-1) & prot) != prot) { 2256 page_set_flags(elf_bss & TARGET_PAGE_MASK, last_bss - 1, 2257 prot | PAGE_VALID); 2258 } 2259 2260 if (host_start < host_map_start) { 2261 memset((void *)host_start, 0, host_map_start - host_start); 2262 } 2263 } 2264 2265 #if defined(TARGET_ARM) 2266 static int elf_is_fdpic(struct elfhdr *exec) 2267 { 2268 return exec->e_ident[EI_OSABI] == ELFOSABI_ARM_FDPIC; 2269 } 2270 #elif defined(TARGET_XTENSA) 2271 static int elf_is_fdpic(struct elfhdr *exec) 2272 { 2273 return exec->e_ident[EI_OSABI] == ELFOSABI_XTENSA_FDPIC; 2274 } 2275 #else 2276 /* Default implementation, always false. */ 2277 static int elf_is_fdpic(struct elfhdr *exec) 2278 { 2279 return 0; 2280 } 2281 #endif 2282 2283 static abi_ulong loader_build_fdpic_loadmap(struct image_info *info, abi_ulong sp) 2284 { 2285 uint16_t n; 2286 struct elf32_fdpic_loadseg *loadsegs = info->loadsegs; 2287 2288 /* elf32_fdpic_loadseg */ 2289 n = info->nsegs; 2290 while (n--) { 2291 sp -= 12; 2292 put_user_u32(loadsegs[n].addr, sp+0); 2293 put_user_u32(loadsegs[n].p_vaddr, sp+4); 2294 put_user_u32(loadsegs[n].p_memsz, sp+8); 2295 } 2296 2297 /* elf32_fdpic_loadmap */ 2298 sp -= 4; 2299 put_user_u16(0, sp+0); /* version */ 2300 put_user_u16(info->nsegs, sp+2); /* nsegs */ 2301 2302 info->personality = PER_LINUX_FDPIC; 2303 info->loadmap_addr = sp; 2304 2305 return sp; 2306 } 2307 2308 static abi_ulong create_elf_tables(abi_ulong p, int argc, int envc, 2309 struct elfhdr *exec, 2310 struct image_info *info, 2311 struct image_info *interp_info) 2312 { 2313 abi_ulong sp; 2314 abi_ulong u_argc, u_argv, u_envp, u_auxv; 2315 int size; 2316 int i; 2317 abi_ulong u_rand_bytes; 2318 uint8_t k_rand_bytes[16]; 2319 abi_ulong u_platform, u_base_platform; 2320 const char *k_platform, *k_base_platform; 2321 const int n = sizeof(elf_addr_t); 2322 2323 sp = p; 2324 2325 /* Needs to be before we load the env/argc/... */ 2326 if (elf_is_fdpic(exec)) { 2327 /* Need 4 byte alignment for these structs */ 2328 sp &= ~3; 2329 sp = loader_build_fdpic_loadmap(info, sp); 2330 info->other_info = interp_info; 2331 if (interp_info) { 2332 interp_info->other_info = info; 2333 sp = loader_build_fdpic_loadmap(interp_info, sp); 2334 info->interpreter_loadmap_addr = interp_info->loadmap_addr; 2335 info->interpreter_pt_dynamic_addr = interp_info->pt_dynamic_addr; 2336 } else { 2337 info->interpreter_loadmap_addr = 0; 2338 info->interpreter_pt_dynamic_addr = 0; 2339 } 2340 } 2341 2342 u_base_platform = 0; 2343 k_base_platform = ELF_BASE_PLATFORM; 2344 if (k_base_platform) { 2345 size_t len = strlen(k_base_platform) + 1; 2346 if (STACK_GROWS_DOWN) { 2347 sp -= (len + n - 1) & ~(n - 1); 2348 u_base_platform = sp; 2349 /* FIXME - check return value of memcpy_to_target() for failure */ 2350 memcpy_to_target(sp, k_base_platform, len); 2351 } else { 2352 memcpy_to_target(sp, k_base_platform, len); 2353 u_base_platform = sp; 2354 sp += len + 1; 2355 } 2356 } 2357 2358 u_platform = 0; 2359 k_platform = ELF_PLATFORM; 2360 if (k_platform) { 2361 size_t len = strlen(k_platform) + 1; 2362 if (STACK_GROWS_DOWN) { 2363 sp -= (len + n - 1) & ~(n - 1); 2364 u_platform = sp; 2365 /* FIXME - check return value of memcpy_to_target() for failure */ 2366 memcpy_to_target(sp, k_platform, len); 2367 } else { 2368 memcpy_to_target(sp, k_platform, len); 2369 u_platform = sp; 2370 sp += len + 1; 2371 } 2372 } 2373 2374 /* Provide 16 byte alignment for the PRNG, and basic alignment for 2375 * the argv and envp pointers. 2376 */ 2377 if (STACK_GROWS_DOWN) { 2378 sp = QEMU_ALIGN_DOWN(sp, 16); 2379 } else { 2380 sp = QEMU_ALIGN_UP(sp, 16); 2381 } 2382 2383 /* 2384 * Generate 16 random bytes for userspace PRNG seeding. 2385 */ 2386 qemu_guest_getrandom_nofail(k_rand_bytes, sizeof(k_rand_bytes)); 2387 if (STACK_GROWS_DOWN) { 2388 sp -= 16; 2389 u_rand_bytes = sp; 2390 /* FIXME - check return value of memcpy_to_target() for failure */ 2391 memcpy_to_target(sp, k_rand_bytes, 16); 2392 } else { 2393 memcpy_to_target(sp, k_rand_bytes, 16); 2394 u_rand_bytes = sp; 2395 sp += 16; 2396 } 2397 2398 size = (DLINFO_ITEMS + 1) * 2; 2399 if (k_base_platform) 2400 size += 2; 2401 if (k_platform) 2402 size += 2; 2403 #ifdef DLINFO_ARCH_ITEMS 2404 size += DLINFO_ARCH_ITEMS * 2; 2405 #endif 2406 #ifdef ELF_HWCAP2 2407 size += 2; 2408 #endif 2409 info->auxv_len = size * n; 2410 2411 size += envc + argc + 2; 2412 size += 1; /* argc itself */ 2413 size *= n; 2414 2415 /* Allocate space and finalize stack alignment for entry now. */ 2416 if (STACK_GROWS_DOWN) { 2417 u_argc = QEMU_ALIGN_DOWN(sp - size, STACK_ALIGNMENT); 2418 sp = u_argc; 2419 } else { 2420 u_argc = sp; 2421 sp = QEMU_ALIGN_UP(sp + size, STACK_ALIGNMENT); 2422 } 2423 2424 u_argv = u_argc + n; 2425 u_envp = u_argv + (argc + 1) * n; 2426 u_auxv = u_envp + (envc + 1) * n; 2427 info->saved_auxv = u_auxv; 2428 info->argc = argc; 2429 info->envc = envc; 2430 info->argv = u_argv; 2431 info->envp = u_envp; 2432 2433 /* This is correct because Linux defines 2434 * elf_addr_t as Elf32_Off / Elf64_Off 2435 */ 2436 #define NEW_AUX_ENT(id, val) do { \ 2437 put_user_ual(id, u_auxv); u_auxv += n; \ 2438 put_user_ual(val, u_auxv); u_auxv += n; \ 2439 } while(0) 2440 2441 #ifdef ARCH_DLINFO 2442 /* 2443 * ARCH_DLINFO must come first so platform specific code can enforce 2444 * special alignment requirements on the AUXV if necessary (eg. PPC). 2445 */ 2446 ARCH_DLINFO; 2447 #endif 2448 /* There must be exactly DLINFO_ITEMS entries here, or the assert 2449 * on info->auxv_len will trigger. 2450 */ 2451 NEW_AUX_ENT(AT_PHDR, (abi_ulong)(info->load_addr + exec->e_phoff)); 2452 NEW_AUX_ENT(AT_PHENT, (abi_ulong)(sizeof (struct elf_phdr))); 2453 NEW_AUX_ENT(AT_PHNUM, (abi_ulong)(exec->e_phnum)); 2454 if ((info->alignment & ~qemu_host_page_mask) != 0) { 2455 /* Target doesn't support host page size alignment */ 2456 NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(TARGET_PAGE_SIZE)); 2457 } else { 2458 NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(MAX(TARGET_PAGE_SIZE, 2459 qemu_host_page_size))); 2460 } 2461 NEW_AUX_ENT(AT_BASE, (abi_ulong)(interp_info ? interp_info->load_addr : 0)); 2462 NEW_AUX_ENT(AT_FLAGS, (abi_ulong)0); 2463 NEW_AUX_ENT(AT_ENTRY, info->entry); 2464 NEW_AUX_ENT(AT_UID, (abi_ulong) getuid()); 2465 NEW_AUX_ENT(AT_EUID, (abi_ulong) geteuid()); 2466 NEW_AUX_ENT(AT_GID, (abi_ulong) getgid()); 2467 NEW_AUX_ENT(AT_EGID, (abi_ulong) getegid()); 2468 NEW_AUX_ENT(AT_HWCAP, (abi_ulong) ELF_HWCAP); 2469 NEW_AUX_ENT(AT_CLKTCK, (abi_ulong) sysconf(_SC_CLK_TCK)); 2470 NEW_AUX_ENT(AT_RANDOM, (abi_ulong) u_rand_bytes); 2471 NEW_AUX_ENT(AT_SECURE, (abi_ulong) qemu_getauxval(AT_SECURE)); 2472 NEW_AUX_ENT(AT_EXECFN, info->file_string); 2473 2474 #ifdef ELF_HWCAP2 2475 NEW_AUX_ENT(AT_HWCAP2, (abi_ulong) ELF_HWCAP2); 2476 #endif 2477 2478 if (u_base_platform) { 2479 NEW_AUX_ENT(AT_BASE_PLATFORM, u_base_platform); 2480 } 2481 if (u_platform) { 2482 NEW_AUX_ENT(AT_PLATFORM, u_platform); 2483 } 2484 NEW_AUX_ENT (AT_NULL, 0); 2485 #undef NEW_AUX_ENT 2486 2487 /* Check that our initial calculation of the auxv length matches how much 2488 * we actually put into it. 2489 */ 2490 assert(info->auxv_len == u_auxv - info->saved_auxv); 2491 2492 put_user_ual(argc, u_argc); 2493 2494 p = info->arg_strings; 2495 for (i = 0; i < argc; ++i) { 2496 put_user_ual(p, u_argv); 2497 u_argv += n; 2498 p += target_strlen(p) + 1; 2499 } 2500 put_user_ual(0, u_argv); 2501 2502 p = info->env_strings; 2503 for (i = 0; i < envc; ++i) { 2504 put_user_ual(p, u_envp); 2505 u_envp += n; 2506 p += target_strlen(p) + 1; 2507 } 2508 put_user_ual(0, u_envp); 2509 2510 return sp; 2511 } 2512 2513 #if defined(HI_COMMPAGE) 2514 #define LO_COMMPAGE -1 2515 #elif defined(LO_COMMPAGE) 2516 #define HI_COMMPAGE 0 2517 #else 2518 #define HI_COMMPAGE 0 2519 #define LO_COMMPAGE -1 2520 #ifndef INIT_GUEST_COMMPAGE 2521 #define init_guest_commpage() true 2522 #endif 2523 #endif 2524 2525 static void pgb_fail_in_use(const char *image_name) 2526 { 2527 error_report("%s: requires virtual address space that is in use " 2528 "(omit the -B option or choose a different value)", 2529 image_name); 2530 exit(EXIT_FAILURE); 2531 } 2532 2533 static void pgb_have_guest_base(const char *image_name, abi_ulong guest_loaddr, 2534 abi_ulong guest_hiaddr, long align) 2535 { 2536 const int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE; 2537 void *addr, *test; 2538 2539 if (!QEMU_IS_ALIGNED(guest_base, align)) { 2540 fprintf(stderr, "Requested guest base %p does not satisfy " 2541 "host minimum alignment (0x%lx)\n", 2542 (void *)guest_base, align); 2543 exit(EXIT_FAILURE); 2544 } 2545 2546 /* Sanity check the guest binary. */ 2547 if (reserved_va) { 2548 if (guest_hiaddr > reserved_va) { 2549 error_report("%s: requires more than reserved virtual " 2550 "address space (0x%" PRIx64 " > 0x%lx)", 2551 image_name, (uint64_t)guest_hiaddr, reserved_va); 2552 exit(EXIT_FAILURE); 2553 } 2554 } else { 2555 #if HOST_LONG_BITS < TARGET_ABI_BITS 2556 if ((guest_hiaddr - guest_base) > ~(uintptr_t)0) { 2557 error_report("%s: requires more virtual address space " 2558 "than the host can provide (0x%" PRIx64 ")", 2559 image_name, (uint64_t)guest_hiaddr + 1 - guest_base); 2560 exit(EXIT_FAILURE); 2561 } 2562 #endif 2563 } 2564 2565 /* 2566 * Expand the allocation to the entire reserved_va. 2567 * Exclude the mmap_min_addr hole. 2568 */ 2569 if (reserved_va) { 2570 guest_loaddr = (guest_base >= mmap_min_addr ? 0 2571 : mmap_min_addr - guest_base); 2572 guest_hiaddr = reserved_va; 2573 } 2574 2575 /* Reserve the address space for the binary, or reserved_va. */ 2576 test = g2h_untagged(guest_loaddr); 2577 addr = mmap(test, guest_hiaddr - guest_loaddr + 1, PROT_NONE, flags, -1, 0); 2578 if (test != addr) { 2579 pgb_fail_in_use(image_name); 2580 } 2581 qemu_log_mask(CPU_LOG_PAGE, 2582 "%s: base @ %p for %" PRIu64 " bytes\n", 2583 __func__, addr, (uint64_t)guest_hiaddr - guest_loaddr + 1); 2584 } 2585 2586 /** 2587 * pgd_find_hole_fallback: potential mmap address 2588 * @guest_size: size of available space 2589 * @brk: location of break 2590 * @align: memory alignment 2591 * 2592 * This is a fallback method for finding a hole in the host address 2593 * space if we don't have the benefit of being able to access 2594 * /proc/self/map. It can potentially take a very long time as we can 2595 * only dumbly iterate up the host address space seeing if the 2596 * allocation would work. 2597 */ 2598 static uintptr_t pgd_find_hole_fallback(uintptr_t guest_size, uintptr_t brk, 2599 long align, uintptr_t offset) 2600 { 2601 uintptr_t base; 2602 2603 /* Start (aligned) at the bottom and work our way up */ 2604 base = ROUND_UP(mmap_min_addr, align); 2605 2606 while (true) { 2607 uintptr_t align_start, end; 2608 align_start = ROUND_UP(base, align); 2609 end = align_start + guest_size + offset; 2610 2611 /* if brk is anywhere in the range give ourselves some room to grow. */ 2612 if (align_start <= brk && brk < end) { 2613 base = brk + (16 * MiB); 2614 continue; 2615 } else if (align_start + guest_size < align_start) { 2616 /* we have run out of space */ 2617 return -1; 2618 } else { 2619 int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE | 2620 MAP_FIXED_NOREPLACE; 2621 void * mmap_start = mmap((void *) align_start, guest_size, 2622 PROT_NONE, flags, -1, 0); 2623 if (mmap_start != MAP_FAILED) { 2624 munmap(mmap_start, guest_size); 2625 if (mmap_start == (void *) align_start) { 2626 qemu_log_mask(CPU_LOG_PAGE, 2627 "%s: base @ %p for %" PRIdPTR" bytes\n", 2628 __func__, mmap_start + offset, guest_size); 2629 return (uintptr_t) mmap_start + offset; 2630 } 2631 } 2632 base += qemu_host_page_size; 2633 } 2634 } 2635 } 2636 2637 /* Return value for guest_base, or -1 if no hole found. */ 2638 static uintptr_t pgb_find_hole(uintptr_t guest_loaddr, uintptr_t guest_size, 2639 long align, uintptr_t offset) 2640 { 2641 GSList *maps, *iter; 2642 uintptr_t this_start, this_end, next_start, brk; 2643 intptr_t ret = -1; 2644 2645 assert(QEMU_IS_ALIGNED(guest_loaddr, align)); 2646 2647 maps = read_self_maps(); 2648 2649 /* Read brk after we've read the maps, which will malloc. */ 2650 brk = (uintptr_t)sbrk(0); 2651 2652 if (!maps) { 2653 return pgd_find_hole_fallback(guest_size, brk, align, offset); 2654 } 2655 2656 /* The first hole is before the first map entry. */ 2657 this_start = mmap_min_addr; 2658 2659 for (iter = maps; iter; 2660 this_start = next_start, iter = g_slist_next(iter)) { 2661 uintptr_t align_start, hole_size; 2662 2663 this_end = ((MapInfo *)iter->data)->start; 2664 next_start = ((MapInfo *)iter->data)->end; 2665 align_start = ROUND_UP(this_start + offset, align); 2666 2667 /* Skip holes that are too small. */ 2668 if (align_start >= this_end) { 2669 continue; 2670 } 2671 hole_size = this_end - align_start; 2672 if (hole_size < guest_size) { 2673 continue; 2674 } 2675 2676 /* If this hole contains brk, give ourselves some room to grow. */ 2677 if (this_start <= brk && brk < this_end) { 2678 hole_size -= guest_size; 2679 if (sizeof(uintptr_t) == 8 && hole_size >= 1 * GiB) { 2680 align_start += 1 * GiB; 2681 } else if (hole_size >= 16 * MiB) { 2682 align_start += 16 * MiB; 2683 } else { 2684 align_start = (this_end - guest_size) & -align; 2685 if (align_start < this_start) { 2686 continue; 2687 } 2688 } 2689 } 2690 2691 /* Record the lowest successful match. */ 2692 if (ret < 0) { 2693 ret = align_start; 2694 } 2695 /* If this hole contains the identity map, select it. */ 2696 if (align_start <= guest_loaddr && 2697 guest_loaddr + guest_size <= this_end) { 2698 ret = 0; 2699 } 2700 /* If this hole ends above the identity map, stop looking. */ 2701 if (this_end >= guest_loaddr) { 2702 break; 2703 } 2704 } 2705 free_self_maps(maps); 2706 2707 if (ret != -1) { 2708 qemu_log_mask(CPU_LOG_PAGE, "%s: base @ %" PRIxPTR 2709 " for %" PRIuPTR " bytes\n", 2710 __func__, ret, guest_size); 2711 } 2712 2713 return ret; 2714 } 2715 2716 static void pgb_static(const char *image_name, abi_ulong orig_loaddr, 2717 abi_ulong orig_hiaddr, long align) 2718 { 2719 uintptr_t loaddr = orig_loaddr; 2720 uintptr_t hiaddr = orig_hiaddr; 2721 uintptr_t offset = 0; 2722 uintptr_t addr; 2723 2724 if (hiaddr != orig_hiaddr) { 2725 error_report("%s: requires virtual address space that the " 2726 "host cannot provide (0x%" PRIx64 ")", 2727 image_name, (uint64_t)orig_hiaddr + 1); 2728 exit(EXIT_FAILURE); 2729 } 2730 2731 loaddr &= -align; 2732 if (HI_COMMPAGE) { 2733 /* 2734 * Extend the allocation to include the commpage. 2735 * For a 64-bit host, this is just 4GiB; for a 32-bit host we 2736 * need to ensure there is space bellow the guest_base so we 2737 * can map the commpage in the place needed when the address 2738 * arithmetic wraps around. 2739 */ 2740 if (sizeof(uintptr_t) == 8 || loaddr >= 0x80000000u) { 2741 hiaddr = UINT32_MAX; 2742 } else { 2743 offset = -(HI_COMMPAGE & -align); 2744 } 2745 } else if (LO_COMMPAGE != -1) { 2746 loaddr = MIN(loaddr, LO_COMMPAGE & -align); 2747 } 2748 2749 addr = pgb_find_hole(loaddr, hiaddr - loaddr + 1, align, offset); 2750 if (addr == -1) { 2751 /* 2752 * If HI_COMMPAGE, there *might* be a non-consecutive allocation 2753 * that can satisfy both. But as the normal arm32 link base address 2754 * is ~32k, and we extend down to include the commpage, making the 2755 * overhead only ~96k, this is unlikely. 2756 */ 2757 error_report("%s: Unable to allocate %#zx bytes of " 2758 "virtual address space", image_name, 2759 (size_t)(hiaddr - loaddr)); 2760 exit(EXIT_FAILURE); 2761 } 2762 2763 guest_base = addr; 2764 2765 qemu_log_mask(CPU_LOG_PAGE, "%s: base @ %"PRIxPTR" for %" PRIuPTR" bytes\n", 2766 __func__, addr, hiaddr - loaddr); 2767 } 2768 2769 static void pgb_dynamic(const char *image_name, long align) 2770 { 2771 /* 2772 * The executable is dynamic and does not require a fixed address. 2773 * All we need is a commpage that satisfies align. 2774 * If we do not need a commpage, leave guest_base == 0. 2775 */ 2776 if (HI_COMMPAGE) { 2777 uintptr_t addr, commpage; 2778 2779 /* 64-bit hosts should have used reserved_va. */ 2780 assert(sizeof(uintptr_t) == 4); 2781 2782 /* 2783 * By putting the commpage at the first hole, that puts guest_base 2784 * just above that, and maximises the positive guest addresses. 2785 */ 2786 commpage = HI_COMMPAGE & -align; 2787 addr = pgb_find_hole(commpage, -commpage, align, 0); 2788 assert(addr != -1); 2789 guest_base = addr; 2790 } 2791 } 2792 2793 static void pgb_reserved_va(const char *image_name, abi_ulong guest_loaddr, 2794 abi_ulong guest_hiaddr, long align) 2795 { 2796 int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE; 2797 void *addr, *test; 2798 2799 if (guest_hiaddr > reserved_va) { 2800 error_report("%s: requires more than reserved virtual " 2801 "address space (0x%" PRIx64 " > 0x%lx)", 2802 image_name, (uint64_t)guest_hiaddr, reserved_va); 2803 exit(EXIT_FAILURE); 2804 } 2805 2806 /* Widen the "image" to the entire reserved address space. */ 2807 pgb_static(image_name, 0, reserved_va, align); 2808 2809 /* osdep.h defines this as 0 if it's missing */ 2810 flags |= MAP_FIXED_NOREPLACE; 2811 2812 /* Reserve the memory on the host. */ 2813 assert(guest_base != 0); 2814 test = g2h_untagged(0); 2815 addr = mmap(test, reserved_va + 1, PROT_NONE, flags, -1, 0); 2816 if (addr == MAP_FAILED || addr != test) { 2817 error_report("Unable to reserve 0x%lx bytes of virtual address " 2818 "space at %p (%s) for use as guest address space (check your " 2819 "virtual memory ulimit setting, mmap_min_addr or reserve less " 2820 "using qemu-user's -R option)", 2821 reserved_va + 1, test, strerror(errno)); 2822 exit(EXIT_FAILURE); 2823 } 2824 2825 qemu_log_mask(CPU_LOG_PAGE, "%s: base @ %p for %lu bytes\n", 2826 __func__, addr, reserved_va + 1); 2827 } 2828 2829 void probe_guest_base(const char *image_name, abi_ulong guest_loaddr, 2830 abi_ulong guest_hiaddr) 2831 { 2832 /* In order to use host shmat, we must be able to honor SHMLBA. */ 2833 uintptr_t align = MAX(SHMLBA, qemu_host_page_size); 2834 2835 if (have_guest_base) { 2836 pgb_have_guest_base(image_name, guest_loaddr, guest_hiaddr, align); 2837 } else if (reserved_va) { 2838 pgb_reserved_va(image_name, guest_loaddr, guest_hiaddr, align); 2839 } else if (guest_loaddr) { 2840 pgb_static(image_name, guest_loaddr, guest_hiaddr, align); 2841 } else { 2842 pgb_dynamic(image_name, align); 2843 } 2844 2845 /* Reserve and initialize the commpage. */ 2846 if (!init_guest_commpage()) { 2847 /* 2848 * With have_guest_base, the user has selected the address and 2849 * we are trying to work with that. Otherwise, we have selected 2850 * free space and init_guest_commpage must succeeded. 2851 */ 2852 assert(have_guest_base); 2853 pgb_fail_in_use(image_name); 2854 } 2855 2856 assert(QEMU_IS_ALIGNED(guest_base, align)); 2857 qemu_log_mask(CPU_LOG_PAGE, "Locating guest address space " 2858 "@ 0x%" PRIx64 "\n", (uint64_t)guest_base); 2859 } 2860 2861 enum { 2862 /* The string "GNU\0" as a magic number. */ 2863 GNU0_MAGIC = const_le32('G' | 'N' << 8 | 'U' << 16), 2864 NOTE_DATA_SZ = 1 * KiB, 2865 NOTE_NAME_SZ = 4, 2866 ELF_GNU_PROPERTY_ALIGN = ELF_CLASS == ELFCLASS32 ? 4 : 8, 2867 }; 2868 2869 /* 2870 * Process a single gnu_property entry. 2871 * Return false for error. 2872 */ 2873 static bool parse_elf_property(const uint32_t *data, int *off, int datasz, 2874 struct image_info *info, bool have_prev_type, 2875 uint32_t *prev_type, Error **errp) 2876 { 2877 uint32_t pr_type, pr_datasz, step; 2878 2879 if (*off > datasz || !QEMU_IS_ALIGNED(*off, ELF_GNU_PROPERTY_ALIGN)) { 2880 goto error_data; 2881 } 2882 datasz -= *off; 2883 data += *off / sizeof(uint32_t); 2884 2885 if (datasz < 2 * sizeof(uint32_t)) { 2886 goto error_data; 2887 } 2888 pr_type = data[0]; 2889 pr_datasz = data[1]; 2890 data += 2; 2891 datasz -= 2 * sizeof(uint32_t); 2892 step = ROUND_UP(pr_datasz, ELF_GNU_PROPERTY_ALIGN); 2893 if (step > datasz) { 2894 goto error_data; 2895 } 2896 2897 /* Properties are supposed to be unique and sorted on pr_type. */ 2898 if (have_prev_type && pr_type <= *prev_type) { 2899 if (pr_type == *prev_type) { 2900 error_setg(errp, "Duplicate property in PT_GNU_PROPERTY"); 2901 } else { 2902 error_setg(errp, "Unsorted property in PT_GNU_PROPERTY"); 2903 } 2904 return false; 2905 } 2906 *prev_type = pr_type; 2907 2908 if (!arch_parse_elf_property(pr_type, pr_datasz, data, info, errp)) { 2909 return false; 2910 } 2911 2912 *off += 2 * sizeof(uint32_t) + step; 2913 return true; 2914 2915 error_data: 2916 error_setg(errp, "Ill-formed property in PT_GNU_PROPERTY"); 2917 return false; 2918 } 2919 2920 /* Process NT_GNU_PROPERTY_TYPE_0. */ 2921 static bool parse_elf_properties(int image_fd, 2922 struct image_info *info, 2923 const struct elf_phdr *phdr, 2924 char bprm_buf[BPRM_BUF_SIZE], 2925 Error **errp) 2926 { 2927 union { 2928 struct elf_note nhdr; 2929 uint32_t data[NOTE_DATA_SZ / sizeof(uint32_t)]; 2930 } note; 2931 2932 int n, off, datasz; 2933 bool have_prev_type; 2934 uint32_t prev_type; 2935 2936 /* Unless the arch requires properties, ignore them. */ 2937 if (!ARCH_USE_GNU_PROPERTY) { 2938 return true; 2939 } 2940 2941 /* If the properties are crazy large, that's too bad. */ 2942 n = phdr->p_filesz; 2943 if (n > sizeof(note)) { 2944 error_setg(errp, "PT_GNU_PROPERTY too large"); 2945 return false; 2946 } 2947 if (n < sizeof(note.nhdr)) { 2948 error_setg(errp, "PT_GNU_PROPERTY too small"); 2949 return false; 2950 } 2951 2952 if (phdr->p_offset + n <= BPRM_BUF_SIZE) { 2953 memcpy(¬e, bprm_buf + phdr->p_offset, n); 2954 } else { 2955 ssize_t len = pread(image_fd, ¬e, n, phdr->p_offset); 2956 if (len != n) { 2957 error_setg_errno(errp, errno, "Error reading file header"); 2958 return false; 2959 } 2960 } 2961 2962 /* 2963 * The contents of a valid PT_GNU_PROPERTY is a sequence 2964 * of uint32_t -- swap them all now. 2965 */ 2966 #ifdef BSWAP_NEEDED 2967 for (int i = 0; i < n / 4; i++) { 2968 bswap32s(note.data + i); 2969 } 2970 #endif 2971 2972 /* 2973 * Note that nhdr is 3 words, and that the "name" described by namesz 2974 * immediately follows nhdr and is thus at the 4th word. Further, all 2975 * of the inputs to the kernel's round_up are multiples of 4. 2976 */ 2977 if (note.nhdr.n_type != NT_GNU_PROPERTY_TYPE_0 || 2978 note.nhdr.n_namesz != NOTE_NAME_SZ || 2979 note.data[3] != GNU0_MAGIC) { 2980 error_setg(errp, "Invalid note in PT_GNU_PROPERTY"); 2981 return false; 2982 } 2983 off = sizeof(note.nhdr) + NOTE_NAME_SZ; 2984 2985 datasz = note.nhdr.n_descsz + off; 2986 if (datasz > n) { 2987 error_setg(errp, "Invalid note size in PT_GNU_PROPERTY"); 2988 return false; 2989 } 2990 2991 have_prev_type = false; 2992 prev_type = 0; 2993 while (1) { 2994 if (off == datasz) { 2995 return true; /* end, exit ok */ 2996 } 2997 if (!parse_elf_property(note.data, &off, datasz, info, 2998 have_prev_type, &prev_type, errp)) { 2999 return false; 3000 } 3001 have_prev_type = true; 3002 } 3003 } 3004 3005 /* Load an ELF image into the address space. 3006 3007 IMAGE_NAME is the filename of the image, to use in error messages. 3008 IMAGE_FD is the open file descriptor for the image. 3009 3010 BPRM_BUF is a copy of the beginning of the file; this of course 3011 contains the elf file header at offset 0. It is assumed that this 3012 buffer is sufficiently aligned to present no problems to the host 3013 in accessing data at aligned offsets within the buffer. 3014 3015 On return: INFO values will be filled in, as necessary or available. */ 3016 3017 static void load_elf_image(const char *image_name, int image_fd, 3018 struct image_info *info, char **pinterp_name, 3019 char bprm_buf[BPRM_BUF_SIZE]) 3020 { 3021 struct elfhdr *ehdr = (struct elfhdr *)bprm_buf; 3022 struct elf_phdr *phdr; 3023 abi_ulong load_addr, load_bias, loaddr, hiaddr, error; 3024 int i, retval, prot_exec; 3025 Error *err = NULL; 3026 3027 /* First of all, some simple consistency checks */ 3028 if (!elf_check_ident(ehdr)) { 3029 error_setg(&err, "Invalid ELF image for this architecture"); 3030 goto exit_errmsg; 3031 } 3032 bswap_ehdr(ehdr); 3033 if (!elf_check_ehdr(ehdr)) { 3034 error_setg(&err, "Invalid ELF image for this architecture"); 3035 goto exit_errmsg; 3036 } 3037 3038 i = ehdr->e_phnum * sizeof(struct elf_phdr); 3039 if (ehdr->e_phoff + i <= BPRM_BUF_SIZE) { 3040 phdr = (struct elf_phdr *)(bprm_buf + ehdr->e_phoff); 3041 } else { 3042 phdr = (struct elf_phdr *) alloca(i); 3043 retval = pread(image_fd, phdr, i, ehdr->e_phoff); 3044 if (retval != i) { 3045 goto exit_read; 3046 } 3047 } 3048 bswap_phdr(phdr, ehdr->e_phnum); 3049 3050 info->nsegs = 0; 3051 info->pt_dynamic_addr = 0; 3052 3053 mmap_lock(); 3054 3055 /* 3056 * Find the maximum size of the image and allocate an appropriate 3057 * amount of memory to handle that. Locate the interpreter, if any. 3058 */ 3059 loaddr = -1, hiaddr = 0; 3060 info->alignment = 0; 3061 info->exec_stack = EXSTACK_DEFAULT; 3062 for (i = 0; i < ehdr->e_phnum; ++i) { 3063 struct elf_phdr *eppnt = phdr + i; 3064 if (eppnt->p_type == PT_LOAD) { 3065 abi_ulong a = eppnt->p_vaddr - eppnt->p_offset; 3066 if (a < loaddr) { 3067 loaddr = a; 3068 } 3069 a = eppnt->p_vaddr + eppnt->p_memsz - 1; 3070 if (a > hiaddr) { 3071 hiaddr = a; 3072 } 3073 ++info->nsegs; 3074 info->alignment |= eppnt->p_align; 3075 } else if (eppnt->p_type == PT_INTERP && pinterp_name) { 3076 g_autofree char *interp_name = NULL; 3077 3078 if (*pinterp_name) { 3079 error_setg(&err, "Multiple PT_INTERP entries"); 3080 goto exit_errmsg; 3081 } 3082 3083 interp_name = g_malloc(eppnt->p_filesz); 3084 3085 if (eppnt->p_offset + eppnt->p_filesz <= BPRM_BUF_SIZE) { 3086 memcpy(interp_name, bprm_buf + eppnt->p_offset, 3087 eppnt->p_filesz); 3088 } else { 3089 retval = pread(image_fd, interp_name, eppnt->p_filesz, 3090 eppnt->p_offset); 3091 if (retval != eppnt->p_filesz) { 3092 goto exit_read; 3093 } 3094 } 3095 if (interp_name[eppnt->p_filesz - 1] != 0) { 3096 error_setg(&err, "Invalid PT_INTERP entry"); 3097 goto exit_errmsg; 3098 } 3099 *pinterp_name = g_steal_pointer(&interp_name); 3100 } else if (eppnt->p_type == PT_GNU_PROPERTY) { 3101 if (!parse_elf_properties(image_fd, info, eppnt, bprm_buf, &err)) { 3102 goto exit_errmsg; 3103 } 3104 } else if (eppnt->p_type == PT_GNU_STACK) { 3105 info->exec_stack = eppnt->p_flags & PF_X; 3106 } 3107 } 3108 3109 if (pinterp_name != NULL) { 3110 /* 3111 * This is the main executable. 3112 * 3113 * Reserve extra space for brk. 3114 * We hold on to this space while placing the interpreter 3115 * and the stack, lest they be placed immediately after 3116 * the data segment and block allocation from the brk. 3117 * 3118 * 16MB is chosen as "large enough" without being so large as 3119 * to allow the result to not fit with a 32-bit guest on a 3120 * 32-bit host. However some 64 bit guests (e.g. s390x) 3121 * attempt to place their heap further ahead and currently 3122 * nothing stops them smashing into QEMUs address space. 3123 */ 3124 #if TARGET_LONG_BITS == 64 3125 info->reserve_brk = 32 * MiB; 3126 #else 3127 info->reserve_brk = 16 * MiB; 3128 #endif 3129 hiaddr += info->reserve_brk; 3130 3131 if (ehdr->e_type == ET_EXEC) { 3132 /* 3133 * Make sure that the low address does not conflict with 3134 * MMAP_MIN_ADDR or the QEMU application itself. 3135 */ 3136 probe_guest_base(image_name, loaddr, hiaddr); 3137 } else { 3138 /* 3139 * The binary is dynamic, but we still need to 3140 * select guest_base. In this case we pass a size. 3141 */ 3142 probe_guest_base(image_name, 0, hiaddr - loaddr); 3143 } 3144 } 3145 3146 /* 3147 * Reserve address space for all of this. 3148 * 3149 * In the case of ET_EXEC, we supply MAP_FIXED so that we get 3150 * exactly the address range that is required. 3151 * 3152 * Otherwise this is ET_DYN, and we are searching for a location 3153 * that can hold the memory space required. If the image is 3154 * pre-linked, LOADDR will be non-zero, and the kernel should 3155 * honor that address if it happens to be free. 3156 * 3157 * In both cases, we will overwrite pages in this range with mappings 3158 * from the executable. 3159 */ 3160 load_addr = target_mmap(loaddr, (size_t)hiaddr - loaddr + 1, PROT_NONE, 3161 MAP_PRIVATE | MAP_ANON | MAP_NORESERVE | 3162 (ehdr->e_type == ET_EXEC ? MAP_FIXED : 0), 3163 -1, 0); 3164 if (load_addr == -1) { 3165 goto exit_mmap; 3166 } 3167 load_bias = load_addr - loaddr; 3168 3169 if (elf_is_fdpic(ehdr)) { 3170 struct elf32_fdpic_loadseg *loadsegs = info->loadsegs = 3171 g_malloc(sizeof(*loadsegs) * info->nsegs); 3172 3173 for (i = 0; i < ehdr->e_phnum; ++i) { 3174 switch (phdr[i].p_type) { 3175 case PT_DYNAMIC: 3176 info->pt_dynamic_addr = phdr[i].p_vaddr + load_bias; 3177 break; 3178 case PT_LOAD: 3179 loadsegs->addr = phdr[i].p_vaddr + load_bias; 3180 loadsegs->p_vaddr = phdr[i].p_vaddr; 3181 loadsegs->p_memsz = phdr[i].p_memsz; 3182 ++loadsegs; 3183 break; 3184 } 3185 } 3186 } 3187 3188 info->load_bias = load_bias; 3189 info->code_offset = load_bias; 3190 info->data_offset = load_bias; 3191 info->load_addr = load_addr; 3192 info->entry = ehdr->e_entry + load_bias; 3193 info->start_code = -1; 3194 info->end_code = 0; 3195 info->start_data = -1; 3196 info->end_data = 0; 3197 info->brk = 0; 3198 info->elf_flags = ehdr->e_flags; 3199 3200 prot_exec = PROT_EXEC; 3201 #ifdef TARGET_AARCH64 3202 /* 3203 * If the BTI feature is present, this indicates that the executable 3204 * pages of the startup binary should be mapped with PROT_BTI, so that 3205 * branch targets are enforced. 3206 * 3207 * The startup binary is either the interpreter or the static executable. 3208 * The interpreter is responsible for all pages of a dynamic executable. 3209 * 3210 * Elf notes are backward compatible to older cpus. 3211 * Do not enable BTI unless it is supported. 3212 */ 3213 if ((info->note_flags & GNU_PROPERTY_AARCH64_FEATURE_1_BTI) 3214 && (pinterp_name == NULL || *pinterp_name == 0) 3215 && cpu_isar_feature(aa64_bti, ARM_CPU(thread_cpu))) { 3216 prot_exec |= TARGET_PROT_BTI; 3217 } 3218 #endif 3219 3220 for (i = 0; i < ehdr->e_phnum; i++) { 3221 struct elf_phdr *eppnt = phdr + i; 3222 if (eppnt->p_type == PT_LOAD) { 3223 abi_ulong vaddr, vaddr_po, vaddr_ps, vaddr_ef, vaddr_em, vaddr_len; 3224 int elf_prot = 0; 3225 3226 if (eppnt->p_flags & PF_R) { 3227 elf_prot |= PROT_READ; 3228 } 3229 if (eppnt->p_flags & PF_W) { 3230 elf_prot |= PROT_WRITE; 3231 } 3232 if (eppnt->p_flags & PF_X) { 3233 elf_prot |= prot_exec; 3234 } 3235 3236 vaddr = load_bias + eppnt->p_vaddr; 3237 vaddr_po = TARGET_ELF_PAGEOFFSET(vaddr); 3238 vaddr_ps = TARGET_ELF_PAGESTART(vaddr); 3239 3240 vaddr_ef = vaddr + eppnt->p_filesz; 3241 vaddr_em = vaddr + eppnt->p_memsz; 3242 3243 /* 3244 * Some segments may be completely empty, with a non-zero p_memsz 3245 * but no backing file segment. 3246 */ 3247 if (eppnt->p_filesz != 0) { 3248 vaddr_len = TARGET_ELF_PAGELENGTH(eppnt->p_filesz + vaddr_po); 3249 error = target_mmap(vaddr_ps, vaddr_len, elf_prot, 3250 MAP_PRIVATE | MAP_FIXED, 3251 image_fd, eppnt->p_offset - vaddr_po); 3252 3253 if (error == -1) { 3254 goto exit_mmap; 3255 } 3256 3257 /* 3258 * If the load segment requests extra zeros (e.g. bss), map it. 3259 */ 3260 if (eppnt->p_filesz < eppnt->p_memsz) { 3261 zero_bss(vaddr_ef, vaddr_em, elf_prot); 3262 } 3263 } else if (eppnt->p_memsz != 0) { 3264 vaddr_len = TARGET_ELF_PAGELENGTH(eppnt->p_memsz + vaddr_po); 3265 error = target_mmap(vaddr_ps, vaddr_len, elf_prot, 3266 MAP_PRIVATE | MAP_FIXED | MAP_ANONYMOUS, 3267 -1, 0); 3268 3269 if (error == -1) { 3270 goto exit_mmap; 3271 } 3272 } 3273 3274 /* Find the full program boundaries. */ 3275 if (elf_prot & PROT_EXEC) { 3276 if (vaddr < info->start_code) { 3277 info->start_code = vaddr; 3278 } 3279 if (vaddr_ef > info->end_code) { 3280 info->end_code = vaddr_ef; 3281 } 3282 } 3283 if (elf_prot & PROT_WRITE) { 3284 if (vaddr < info->start_data) { 3285 info->start_data = vaddr; 3286 } 3287 if (vaddr_ef > info->end_data) { 3288 info->end_data = vaddr_ef; 3289 } 3290 } 3291 if (vaddr_em > info->brk) { 3292 info->brk = vaddr_em; 3293 } 3294 #ifdef TARGET_MIPS 3295 } else if (eppnt->p_type == PT_MIPS_ABIFLAGS) { 3296 Mips_elf_abiflags_v0 abiflags; 3297 if (eppnt->p_filesz < sizeof(Mips_elf_abiflags_v0)) { 3298 error_setg(&err, "Invalid PT_MIPS_ABIFLAGS entry"); 3299 goto exit_errmsg; 3300 } 3301 if (eppnt->p_offset + eppnt->p_filesz <= BPRM_BUF_SIZE) { 3302 memcpy(&abiflags, bprm_buf + eppnt->p_offset, 3303 sizeof(Mips_elf_abiflags_v0)); 3304 } else { 3305 retval = pread(image_fd, &abiflags, sizeof(Mips_elf_abiflags_v0), 3306 eppnt->p_offset); 3307 if (retval != sizeof(Mips_elf_abiflags_v0)) { 3308 goto exit_read; 3309 } 3310 } 3311 bswap_mips_abiflags(&abiflags); 3312 info->fp_abi = abiflags.fp_abi; 3313 #endif 3314 } 3315 } 3316 3317 if (info->end_data == 0) { 3318 info->start_data = info->end_code; 3319 info->end_data = info->end_code; 3320 } 3321 3322 if (qemu_log_enabled()) { 3323 load_symbols(ehdr, image_fd, load_bias); 3324 } 3325 3326 debuginfo_report_elf(image_name, image_fd, load_bias); 3327 3328 mmap_unlock(); 3329 3330 close(image_fd); 3331 return; 3332 3333 exit_read: 3334 if (retval >= 0) { 3335 error_setg(&err, "Incomplete read of file header"); 3336 } else { 3337 error_setg_errno(&err, errno, "Error reading file header"); 3338 } 3339 goto exit_errmsg; 3340 exit_mmap: 3341 error_setg_errno(&err, errno, "Error mapping file"); 3342 goto exit_errmsg; 3343 exit_errmsg: 3344 error_reportf_err(err, "%s: ", image_name); 3345 exit(-1); 3346 } 3347 3348 static void load_elf_interp(const char *filename, struct image_info *info, 3349 char bprm_buf[BPRM_BUF_SIZE]) 3350 { 3351 int fd, retval; 3352 Error *err = NULL; 3353 3354 fd = open(path(filename), O_RDONLY); 3355 if (fd < 0) { 3356 error_setg_file_open(&err, errno, filename); 3357 error_report_err(err); 3358 exit(-1); 3359 } 3360 3361 retval = read(fd, bprm_buf, BPRM_BUF_SIZE); 3362 if (retval < 0) { 3363 error_setg_errno(&err, errno, "Error reading file header"); 3364 error_reportf_err(err, "%s: ", filename); 3365 exit(-1); 3366 } 3367 3368 if (retval < BPRM_BUF_SIZE) { 3369 memset(bprm_buf + retval, 0, BPRM_BUF_SIZE - retval); 3370 } 3371 3372 load_elf_image(filename, fd, info, NULL, bprm_buf); 3373 } 3374 3375 static int symfind(const void *s0, const void *s1) 3376 { 3377 struct elf_sym *sym = (struct elf_sym *)s1; 3378 __typeof(sym->st_value) addr = *(uint64_t *)s0; 3379 int result = 0; 3380 3381 if (addr < sym->st_value) { 3382 result = -1; 3383 } else if (addr >= sym->st_value + sym->st_size) { 3384 result = 1; 3385 } 3386 return result; 3387 } 3388 3389 static const char *lookup_symbolxx(struct syminfo *s, uint64_t orig_addr) 3390 { 3391 #if ELF_CLASS == ELFCLASS32 3392 struct elf_sym *syms = s->disas_symtab.elf32; 3393 #else 3394 struct elf_sym *syms = s->disas_symtab.elf64; 3395 #endif 3396 3397 // binary search 3398 struct elf_sym *sym; 3399 3400 sym = bsearch(&orig_addr, syms, s->disas_num_syms, sizeof(*syms), symfind); 3401 if (sym != NULL) { 3402 return s->disas_strtab + sym->st_name; 3403 } 3404 3405 return ""; 3406 } 3407 3408 /* FIXME: This should use elf_ops.h */ 3409 static int symcmp(const void *s0, const void *s1) 3410 { 3411 struct elf_sym *sym0 = (struct elf_sym *)s0; 3412 struct elf_sym *sym1 = (struct elf_sym *)s1; 3413 return (sym0->st_value < sym1->st_value) 3414 ? -1 3415 : ((sym0->st_value > sym1->st_value) ? 1 : 0); 3416 } 3417 3418 /* Best attempt to load symbols from this ELF object. */ 3419 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias) 3420 { 3421 int i, shnum, nsyms, sym_idx = 0, str_idx = 0; 3422 uint64_t segsz; 3423 struct elf_shdr *shdr; 3424 char *strings = NULL; 3425 struct syminfo *s = NULL; 3426 struct elf_sym *new_syms, *syms = NULL; 3427 3428 shnum = hdr->e_shnum; 3429 i = shnum * sizeof(struct elf_shdr); 3430 shdr = (struct elf_shdr *)alloca(i); 3431 if (pread(fd, shdr, i, hdr->e_shoff) != i) { 3432 return; 3433 } 3434 3435 bswap_shdr(shdr, shnum); 3436 for (i = 0; i < shnum; ++i) { 3437 if (shdr[i].sh_type == SHT_SYMTAB) { 3438 sym_idx = i; 3439 str_idx = shdr[i].sh_link; 3440 goto found; 3441 } 3442 } 3443 3444 /* There will be no symbol table if the file was stripped. */ 3445 return; 3446 3447 found: 3448 /* Now know where the strtab and symtab are. Snarf them. */ 3449 s = g_try_new(struct syminfo, 1); 3450 if (!s) { 3451 goto give_up; 3452 } 3453 3454 segsz = shdr[str_idx].sh_size; 3455 s->disas_strtab = strings = g_try_malloc(segsz); 3456 if (!strings || 3457 pread(fd, strings, segsz, shdr[str_idx].sh_offset) != segsz) { 3458 goto give_up; 3459 } 3460 3461 segsz = shdr[sym_idx].sh_size; 3462 syms = g_try_malloc(segsz); 3463 if (!syms || pread(fd, syms, segsz, shdr[sym_idx].sh_offset) != segsz) { 3464 goto give_up; 3465 } 3466 3467 if (segsz / sizeof(struct elf_sym) > INT_MAX) { 3468 /* Implausibly large symbol table: give up rather than ploughing 3469 * on with the number of symbols calculation overflowing 3470 */ 3471 goto give_up; 3472 } 3473 nsyms = segsz / sizeof(struct elf_sym); 3474 for (i = 0; i < nsyms; ) { 3475 bswap_sym(syms + i); 3476 /* Throw away entries which we do not need. */ 3477 if (syms[i].st_shndx == SHN_UNDEF 3478 || syms[i].st_shndx >= SHN_LORESERVE 3479 || ELF_ST_TYPE(syms[i].st_info) != STT_FUNC) { 3480 if (i < --nsyms) { 3481 syms[i] = syms[nsyms]; 3482 } 3483 } else { 3484 #if defined(TARGET_ARM) || defined (TARGET_MIPS) 3485 /* The bottom address bit marks a Thumb or MIPS16 symbol. */ 3486 syms[i].st_value &= ~(target_ulong)1; 3487 #endif 3488 syms[i].st_value += load_bias; 3489 i++; 3490 } 3491 } 3492 3493 /* No "useful" symbol. */ 3494 if (nsyms == 0) { 3495 goto give_up; 3496 } 3497 3498 /* Attempt to free the storage associated with the local symbols 3499 that we threw away. Whether or not this has any effect on the 3500 memory allocation depends on the malloc implementation and how 3501 many symbols we managed to discard. */ 3502 new_syms = g_try_renew(struct elf_sym, syms, nsyms); 3503 if (new_syms == NULL) { 3504 goto give_up; 3505 } 3506 syms = new_syms; 3507 3508 qsort(syms, nsyms, sizeof(*syms), symcmp); 3509 3510 s->disas_num_syms = nsyms; 3511 #if ELF_CLASS == ELFCLASS32 3512 s->disas_symtab.elf32 = syms; 3513 #else 3514 s->disas_symtab.elf64 = syms; 3515 #endif 3516 s->lookup_symbol = lookup_symbolxx; 3517 s->next = syminfos; 3518 syminfos = s; 3519 3520 return; 3521 3522 give_up: 3523 g_free(s); 3524 g_free(strings); 3525 g_free(syms); 3526 } 3527 3528 uint32_t get_elf_eflags(int fd) 3529 { 3530 struct elfhdr ehdr; 3531 off_t offset; 3532 int ret; 3533 3534 /* Read ELF header */ 3535 offset = lseek(fd, 0, SEEK_SET); 3536 if (offset == (off_t) -1) { 3537 return 0; 3538 } 3539 ret = read(fd, &ehdr, sizeof(ehdr)); 3540 if (ret < sizeof(ehdr)) { 3541 return 0; 3542 } 3543 offset = lseek(fd, offset, SEEK_SET); 3544 if (offset == (off_t) -1) { 3545 return 0; 3546 } 3547 3548 /* Check ELF signature */ 3549 if (!elf_check_ident(&ehdr)) { 3550 return 0; 3551 } 3552 3553 /* check header */ 3554 bswap_ehdr(&ehdr); 3555 if (!elf_check_ehdr(&ehdr)) { 3556 return 0; 3557 } 3558 3559 /* return architecture id */ 3560 return ehdr.e_flags; 3561 } 3562 3563 int load_elf_binary(struct linux_binprm *bprm, struct image_info *info) 3564 { 3565 struct image_info interp_info; 3566 struct elfhdr elf_ex; 3567 char *elf_interpreter = NULL; 3568 char *scratch; 3569 3570 memset(&interp_info, 0, sizeof(interp_info)); 3571 #ifdef TARGET_MIPS 3572 interp_info.fp_abi = MIPS_ABI_FP_UNKNOWN; 3573 #endif 3574 3575 info->start_mmap = (abi_ulong)ELF_START_MMAP; 3576 3577 load_elf_image(bprm->filename, bprm->fd, info, 3578 &elf_interpreter, bprm->buf); 3579 3580 /* ??? We need a copy of the elf header for passing to create_elf_tables. 3581 If we do nothing, we'll have overwritten this when we re-use bprm->buf 3582 when we load the interpreter. */ 3583 elf_ex = *(struct elfhdr *)bprm->buf; 3584 3585 /* Do this so that we can load the interpreter, if need be. We will 3586 change some of these later */ 3587 bprm->p = setup_arg_pages(bprm, info); 3588 3589 scratch = g_new0(char, TARGET_PAGE_SIZE); 3590 if (STACK_GROWS_DOWN) { 3591 bprm->p = copy_elf_strings(1, &bprm->filename, scratch, 3592 bprm->p, info->stack_limit); 3593 info->file_string = bprm->p; 3594 bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch, 3595 bprm->p, info->stack_limit); 3596 info->env_strings = bprm->p; 3597 bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch, 3598 bprm->p, info->stack_limit); 3599 info->arg_strings = bprm->p; 3600 } else { 3601 info->arg_strings = bprm->p; 3602 bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch, 3603 bprm->p, info->stack_limit); 3604 info->env_strings = bprm->p; 3605 bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch, 3606 bprm->p, info->stack_limit); 3607 info->file_string = bprm->p; 3608 bprm->p = copy_elf_strings(1, &bprm->filename, scratch, 3609 bprm->p, info->stack_limit); 3610 } 3611 3612 g_free(scratch); 3613 3614 if (!bprm->p) { 3615 fprintf(stderr, "%s: %s\n", bprm->filename, strerror(E2BIG)); 3616 exit(-1); 3617 } 3618 3619 if (elf_interpreter) { 3620 load_elf_interp(elf_interpreter, &interp_info, bprm->buf); 3621 3622 /* If the program interpreter is one of these two, then assume 3623 an iBCS2 image. Otherwise assume a native linux image. */ 3624 3625 if (strcmp(elf_interpreter, "/usr/lib/libc.so.1") == 0 3626 || strcmp(elf_interpreter, "/usr/lib/ld.so.1") == 0) { 3627 info->personality = PER_SVR4; 3628 3629 /* Why this, you ask??? Well SVr4 maps page 0 as read-only, 3630 and some applications "depend" upon this behavior. Since 3631 we do not have the power to recompile these, we emulate 3632 the SVr4 behavior. Sigh. */ 3633 target_mmap(0, qemu_host_page_size, PROT_READ | PROT_EXEC, 3634 MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); 3635 } 3636 #ifdef TARGET_MIPS 3637 info->interp_fp_abi = interp_info.fp_abi; 3638 #endif 3639 } 3640 3641 /* 3642 * TODO: load a vdso, which would also contain the signal trampolines. 3643 * Otherwise, allocate a private page to hold them. 3644 */ 3645 if (TARGET_ARCH_HAS_SIGTRAMP_PAGE) { 3646 abi_long tramp_page = target_mmap(0, TARGET_PAGE_SIZE, 3647 PROT_READ | PROT_WRITE, 3648 MAP_PRIVATE | MAP_ANON, -1, 0); 3649 if (tramp_page == -1) { 3650 return -errno; 3651 } 3652 3653 setup_sigtramp(tramp_page); 3654 target_mprotect(tramp_page, TARGET_PAGE_SIZE, PROT_READ | PROT_EXEC); 3655 } 3656 3657 bprm->p = create_elf_tables(bprm->p, bprm->argc, bprm->envc, &elf_ex, 3658 info, (elf_interpreter ? &interp_info : NULL)); 3659 info->start_stack = bprm->p; 3660 3661 /* If we have an interpreter, set that as the program's entry point. 3662 Copy the load_bias as well, to help PPC64 interpret the entry 3663 point as a function descriptor. Do this after creating elf tables 3664 so that we copy the original program entry point into the AUXV. */ 3665 if (elf_interpreter) { 3666 info->load_bias = interp_info.load_bias; 3667 info->entry = interp_info.entry; 3668 g_free(elf_interpreter); 3669 } 3670 3671 #ifdef USE_ELF_CORE_DUMP 3672 bprm->core_dump = &elf_core_dump; 3673 #endif 3674 3675 /* 3676 * If we reserved extra space for brk, release it now. 3677 * The implementation of do_brk in syscalls.c expects to be able 3678 * to mmap pages in this space. 3679 */ 3680 if (info->reserve_brk) { 3681 abi_ulong start_brk = HOST_PAGE_ALIGN(info->brk); 3682 abi_ulong end_brk = HOST_PAGE_ALIGN(info->brk + info->reserve_brk); 3683 target_munmap(start_brk, end_brk - start_brk); 3684 } 3685 3686 return 0; 3687 } 3688 3689 #ifdef USE_ELF_CORE_DUMP 3690 /* 3691 * Definitions to generate Intel SVR4-like core files. 3692 * These mostly have the same names as the SVR4 types with "target_elf_" 3693 * tacked on the front to prevent clashes with linux definitions, 3694 * and the typedef forms have been avoided. This is mostly like 3695 * the SVR4 structure, but more Linuxy, with things that Linux does 3696 * not support and which gdb doesn't really use excluded. 3697 * 3698 * Fields we don't dump (their contents is zero) in linux-user qemu 3699 * are marked with XXX. 3700 * 3701 * Core dump code is copied from linux kernel (fs/binfmt_elf.c). 3702 * 3703 * Porting ELF coredump for target is (quite) simple process. First you 3704 * define USE_ELF_CORE_DUMP in target ELF code (where init_thread() for 3705 * the target resides): 3706 * 3707 * #define USE_ELF_CORE_DUMP 3708 * 3709 * Next you define type of register set used for dumping. ELF specification 3710 * says that it needs to be array of elf_greg_t that has size of ELF_NREG. 3711 * 3712 * typedef <target_regtype> target_elf_greg_t; 3713 * #define ELF_NREG <number of registers> 3714 * typedef taret_elf_greg_t target_elf_gregset_t[ELF_NREG]; 3715 * 3716 * Last step is to implement target specific function that copies registers 3717 * from given cpu into just specified register set. Prototype is: 3718 * 3719 * static void elf_core_copy_regs(taret_elf_gregset_t *regs, 3720 * const CPUArchState *env); 3721 * 3722 * Parameters: 3723 * regs - copy register values into here (allocated and zeroed by caller) 3724 * env - copy registers from here 3725 * 3726 * Example for ARM target is provided in this file. 3727 */ 3728 3729 /* An ELF note in memory */ 3730 struct memelfnote { 3731 const char *name; 3732 size_t namesz; 3733 size_t namesz_rounded; 3734 int type; 3735 size_t datasz; 3736 size_t datasz_rounded; 3737 void *data; 3738 size_t notesz; 3739 }; 3740 3741 struct target_elf_siginfo { 3742 abi_int si_signo; /* signal number */ 3743 abi_int si_code; /* extra code */ 3744 abi_int si_errno; /* errno */ 3745 }; 3746 3747 struct target_elf_prstatus { 3748 struct target_elf_siginfo pr_info; /* Info associated with signal */ 3749 abi_short pr_cursig; /* Current signal */ 3750 abi_ulong pr_sigpend; /* XXX */ 3751 abi_ulong pr_sighold; /* XXX */ 3752 target_pid_t pr_pid; 3753 target_pid_t pr_ppid; 3754 target_pid_t pr_pgrp; 3755 target_pid_t pr_sid; 3756 struct target_timeval pr_utime; /* XXX User time */ 3757 struct target_timeval pr_stime; /* XXX System time */ 3758 struct target_timeval pr_cutime; /* XXX Cumulative user time */ 3759 struct target_timeval pr_cstime; /* XXX Cumulative system time */ 3760 target_elf_gregset_t pr_reg; /* GP registers */ 3761 abi_int pr_fpvalid; /* XXX */ 3762 }; 3763 3764 #define ELF_PRARGSZ (80) /* Number of chars for args */ 3765 3766 struct target_elf_prpsinfo { 3767 char pr_state; /* numeric process state */ 3768 char pr_sname; /* char for pr_state */ 3769 char pr_zomb; /* zombie */ 3770 char pr_nice; /* nice val */ 3771 abi_ulong pr_flag; /* flags */ 3772 target_uid_t pr_uid; 3773 target_gid_t pr_gid; 3774 target_pid_t pr_pid, pr_ppid, pr_pgrp, pr_sid; 3775 /* Lots missing */ 3776 char pr_fname[16] QEMU_NONSTRING; /* filename of executable */ 3777 char pr_psargs[ELF_PRARGSZ]; /* initial part of arg list */ 3778 }; 3779 3780 /* Here is the structure in which status of each thread is captured. */ 3781 struct elf_thread_status { 3782 QTAILQ_ENTRY(elf_thread_status) ets_link; 3783 struct target_elf_prstatus prstatus; /* NT_PRSTATUS */ 3784 #if 0 3785 elf_fpregset_t fpu; /* NT_PRFPREG */ 3786 struct task_struct *thread; 3787 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */ 3788 #endif 3789 struct memelfnote notes[1]; 3790 int num_notes; 3791 }; 3792 3793 struct elf_note_info { 3794 struct memelfnote *notes; 3795 struct target_elf_prstatus *prstatus; /* NT_PRSTATUS */ 3796 struct target_elf_prpsinfo *psinfo; /* NT_PRPSINFO */ 3797 3798 QTAILQ_HEAD(, elf_thread_status) thread_list; 3799 #if 0 3800 /* 3801 * Current version of ELF coredump doesn't support 3802 * dumping fp regs etc. 3803 */ 3804 elf_fpregset_t *fpu; 3805 elf_fpxregset_t *xfpu; 3806 int thread_status_size; 3807 #endif 3808 int notes_size; 3809 int numnote; 3810 }; 3811 3812 struct vm_area_struct { 3813 target_ulong vma_start; /* start vaddr of memory region */ 3814 target_ulong vma_end; /* end vaddr of memory region */ 3815 abi_ulong vma_flags; /* protection etc. flags for the region */ 3816 QTAILQ_ENTRY(vm_area_struct) vma_link; 3817 }; 3818 3819 struct mm_struct { 3820 QTAILQ_HEAD(, vm_area_struct) mm_mmap; 3821 int mm_count; /* number of mappings */ 3822 }; 3823 3824 static struct mm_struct *vma_init(void); 3825 static void vma_delete(struct mm_struct *); 3826 static int vma_add_mapping(struct mm_struct *, target_ulong, 3827 target_ulong, abi_ulong); 3828 static int vma_get_mapping_count(const struct mm_struct *); 3829 static struct vm_area_struct *vma_first(const struct mm_struct *); 3830 static struct vm_area_struct *vma_next(struct vm_area_struct *); 3831 static abi_ulong vma_dump_size(const struct vm_area_struct *); 3832 static int vma_walker(void *priv, target_ulong start, target_ulong end, 3833 unsigned long flags); 3834 3835 static void fill_elf_header(struct elfhdr *, int, uint16_t, uint32_t); 3836 static void fill_note(struct memelfnote *, const char *, int, 3837 unsigned int, void *); 3838 static void fill_prstatus(struct target_elf_prstatus *, const TaskState *, int); 3839 static int fill_psinfo(struct target_elf_prpsinfo *, const TaskState *); 3840 static void fill_auxv_note(struct memelfnote *, const TaskState *); 3841 static void fill_elf_note_phdr(struct elf_phdr *, int, off_t); 3842 static size_t note_size(const struct memelfnote *); 3843 static void free_note_info(struct elf_note_info *); 3844 static int fill_note_info(struct elf_note_info *, long, const CPUArchState *); 3845 static void fill_thread_info(struct elf_note_info *, const CPUArchState *); 3846 3847 static int dump_write(int, const void *, size_t); 3848 static int write_note(struct memelfnote *, int); 3849 static int write_note_info(struct elf_note_info *, int); 3850 3851 #ifdef BSWAP_NEEDED 3852 static void bswap_prstatus(struct target_elf_prstatus *prstatus) 3853 { 3854 prstatus->pr_info.si_signo = tswap32(prstatus->pr_info.si_signo); 3855 prstatus->pr_info.si_code = tswap32(prstatus->pr_info.si_code); 3856 prstatus->pr_info.si_errno = tswap32(prstatus->pr_info.si_errno); 3857 prstatus->pr_cursig = tswap16(prstatus->pr_cursig); 3858 prstatus->pr_sigpend = tswapal(prstatus->pr_sigpend); 3859 prstatus->pr_sighold = tswapal(prstatus->pr_sighold); 3860 prstatus->pr_pid = tswap32(prstatus->pr_pid); 3861 prstatus->pr_ppid = tswap32(prstatus->pr_ppid); 3862 prstatus->pr_pgrp = tswap32(prstatus->pr_pgrp); 3863 prstatus->pr_sid = tswap32(prstatus->pr_sid); 3864 /* cpu times are not filled, so we skip them */ 3865 /* regs should be in correct format already */ 3866 prstatus->pr_fpvalid = tswap32(prstatus->pr_fpvalid); 3867 } 3868 3869 static void bswap_psinfo(struct target_elf_prpsinfo *psinfo) 3870 { 3871 psinfo->pr_flag = tswapal(psinfo->pr_flag); 3872 psinfo->pr_uid = tswap16(psinfo->pr_uid); 3873 psinfo->pr_gid = tswap16(psinfo->pr_gid); 3874 psinfo->pr_pid = tswap32(psinfo->pr_pid); 3875 psinfo->pr_ppid = tswap32(psinfo->pr_ppid); 3876 psinfo->pr_pgrp = tswap32(psinfo->pr_pgrp); 3877 psinfo->pr_sid = tswap32(psinfo->pr_sid); 3878 } 3879 3880 static void bswap_note(struct elf_note *en) 3881 { 3882 bswap32s(&en->n_namesz); 3883 bswap32s(&en->n_descsz); 3884 bswap32s(&en->n_type); 3885 } 3886 #else 3887 static inline void bswap_prstatus(struct target_elf_prstatus *p) { } 3888 static inline void bswap_psinfo(struct target_elf_prpsinfo *p) {} 3889 static inline void bswap_note(struct elf_note *en) { } 3890 #endif /* BSWAP_NEEDED */ 3891 3892 /* 3893 * Minimal support for linux memory regions. These are needed 3894 * when we are finding out what memory exactly belongs to 3895 * emulated process. No locks needed here, as long as 3896 * thread that received the signal is stopped. 3897 */ 3898 3899 static struct mm_struct *vma_init(void) 3900 { 3901 struct mm_struct *mm; 3902 3903 if ((mm = g_malloc(sizeof (*mm))) == NULL) 3904 return (NULL); 3905 3906 mm->mm_count = 0; 3907 QTAILQ_INIT(&mm->mm_mmap); 3908 3909 return (mm); 3910 } 3911 3912 static void vma_delete(struct mm_struct *mm) 3913 { 3914 struct vm_area_struct *vma; 3915 3916 while ((vma = vma_first(mm)) != NULL) { 3917 QTAILQ_REMOVE(&mm->mm_mmap, vma, vma_link); 3918 g_free(vma); 3919 } 3920 g_free(mm); 3921 } 3922 3923 static int vma_add_mapping(struct mm_struct *mm, target_ulong start, 3924 target_ulong end, abi_ulong flags) 3925 { 3926 struct vm_area_struct *vma; 3927 3928 if ((vma = g_malloc0(sizeof (*vma))) == NULL) 3929 return (-1); 3930 3931 vma->vma_start = start; 3932 vma->vma_end = end; 3933 vma->vma_flags = flags; 3934 3935 QTAILQ_INSERT_TAIL(&mm->mm_mmap, vma, vma_link); 3936 mm->mm_count++; 3937 3938 return (0); 3939 } 3940 3941 static struct vm_area_struct *vma_first(const struct mm_struct *mm) 3942 { 3943 return (QTAILQ_FIRST(&mm->mm_mmap)); 3944 } 3945 3946 static struct vm_area_struct *vma_next(struct vm_area_struct *vma) 3947 { 3948 return (QTAILQ_NEXT(vma, vma_link)); 3949 } 3950 3951 static int vma_get_mapping_count(const struct mm_struct *mm) 3952 { 3953 return (mm->mm_count); 3954 } 3955 3956 /* 3957 * Calculate file (dump) size of given memory region. 3958 */ 3959 static abi_ulong vma_dump_size(const struct vm_area_struct *vma) 3960 { 3961 /* if we cannot even read the first page, skip it */ 3962 if (!access_ok_untagged(VERIFY_READ, vma->vma_start, TARGET_PAGE_SIZE)) 3963 return (0); 3964 3965 /* 3966 * Usually we don't dump executable pages as they contain 3967 * non-writable code that debugger can read directly from 3968 * target library etc. However, thread stacks are marked 3969 * also executable so we read in first page of given region 3970 * and check whether it contains elf header. If there is 3971 * no elf header, we dump it. 3972 */ 3973 if (vma->vma_flags & PROT_EXEC) { 3974 char page[TARGET_PAGE_SIZE]; 3975 3976 if (copy_from_user(page, vma->vma_start, sizeof (page))) { 3977 return 0; 3978 } 3979 if ((page[EI_MAG0] == ELFMAG0) && 3980 (page[EI_MAG1] == ELFMAG1) && 3981 (page[EI_MAG2] == ELFMAG2) && 3982 (page[EI_MAG3] == ELFMAG3)) { 3983 /* 3984 * Mappings are possibly from ELF binary. Don't dump 3985 * them. 3986 */ 3987 return (0); 3988 } 3989 } 3990 3991 return (vma->vma_end - vma->vma_start); 3992 } 3993 3994 static int vma_walker(void *priv, target_ulong start, target_ulong end, 3995 unsigned long flags) 3996 { 3997 struct mm_struct *mm = (struct mm_struct *)priv; 3998 3999 vma_add_mapping(mm, start, end, flags); 4000 return (0); 4001 } 4002 4003 static void fill_note(struct memelfnote *note, const char *name, int type, 4004 unsigned int sz, void *data) 4005 { 4006 unsigned int namesz; 4007 4008 namesz = strlen(name) + 1; 4009 note->name = name; 4010 note->namesz = namesz; 4011 note->namesz_rounded = roundup(namesz, sizeof (int32_t)); 4012 note->type = type; 4013 note->datasz = sz; 4014 note->datasz_rounded = roundup(sz, sizeof (int32_t)); 4015 4016 note->data = data; 4017 4018 /* 4019 * We calculate rounded up note size here as specified by 4020 * ELF document. 4021 */ 4022 note->notesz = sizeof (struct elf_note) + 4023 note->namesz_rounded + note->datasz_rounded; 4024 } 4025 4026 static void fill_elf_header(struct elfhdr *elf, int segs, uint16_t machine, 4027 uint32_t flags) 4028 { 4029 (void) memset(elf, 0, sizeof(*elf)); 4030 4031 (void) memcpy(elf->e_ident, ELFMAG, SELFMAG); 4032 elf->e_ident[EI_CLASS] = ELF_CLASS; 4033 elf->e_ident[EI_DATA] = ELF_DATA; 4034 elf->e_ident[EI_VERSION] = EV_CURRENT; 4035 elf->e_ident[EI_OSABI] = ELF_OSABI; 4036 4037 elf->e_type = ET_CORE; 4038 elf->e_machine = machine; 4039 elf->e_version = EV_CURRENT; 4040 elf->e_phoff = sizeof(struct elfhdr); 4041 elf->e_flags = flags; 4042 elf->e_ehsize = sizeof(struct elfhdr); 4043 elf->e_phentsize = sizeof(struct elf_phdr); 4044 elf->e_phnum = segs; 4045 4046 bswap_ehdr(elf); 4047 } 4048 4049 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, off_t offset) 4050 { 4051 phdr->p_type = PT_NOTE; 4052 phdr->p_offset = offset; 4053 phdr->p_vaddr = 0; 4054 phdr->p_paddr = 0; 4055 phdr->p_filesz = sz; 4056 phdr->p_memsz = 0; 4057 phdr->p_flags = 0; 4058 phdr->p_align = 0; 4059 4060 bswap_phdr(phdr, 1); 4061 } 4062 4063 static size_t note_size(const struct memelfnote *note) 4064 { 4065 return (note->notesz); 4066 } 4067 4068 static void fill_prstatus(struct target_elf_prstatus *prstatus, 4069 const TaskState *ts, int signr) 4070 { 4071 (void) memset(prstatus, 0, sizeof (*prstatus)); 4072 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr; 4073 prstatus->pr_pid = ts->ts_tid; 4074 prstatus->pr_ppid = getppid(); 4075 prstatus->pr_pgrp = getpgrp(); 4076 prstatus->pr_sid = getsid(0); 4077 4078 bswap_prstatus(prstatus); 4079 } 4080 4081 static int fill_psinfo(struct target_elf_prpsinfo *psinfo, const TaskState *ts) 4082 { 4083 char *base_filename; 4084 unsigned int i, len; 4085 4086 (void) memset(psinfo, 0, sizeof (*psinfo)); 4087 4088 len = ts->info->env_strings - ts->info->arg_strings; 4089 if (len >= ELF_PRARGSZ) 4090 len = ELF_PRARGSZ - 1; 4091 if (copy_from_user(&psinfo->pr_psargs, ts->info->arg_strings, len)) { 4092 return -EFAULT; 4093 } 4094 for (i = 0; i < len; i++) 4095 if (psinfo->pr_psargs[i] == 0) 4096 psinfo->pr_psargs[i] = ' '; 4097 psinfo->pr_psargs[len] = 0; 4098 4099 psinfo->pr_pid = getpid(); 4100 psinfo->pr_ppid = getppid(); 4101 psinfo->pr_pgrp = getpgrp(); 4102 psinfo->pr_sid = getsid(0); 4103 psinfo->pr_uid = getuid(); 4104 psinfo->pr_gid = getgid(); 4105 4106 base_filename = g_path_get_basename(ts->bprm->filename); 4107 /* 4108 * Using strncpy here is fine: at max-length, 4109 * this field is not NUL-terminated. 4110 */ 4111 (void) strncpy(psinfo->pr_fname, base_filename, 4112 sizeof(psinfo->pr_fname)); 4113 4114 g_free(base_filename); 4115 bswap_psinfo(psinfo); 4116 return (0); 4117 } 4118 4119 static void fill_auxv_note(struct memelfnote *note, const TaskState *ts) 4120 { 4121 elf_addr_t auxv = (elf_addr_t)ts->info->saved_auxv; 4122 elf_addr_t orig_auxv = auxv; 4123 void *ptr; 4124 int len = ts->info->auxv_len; 4125 4126 /* 4127 * Auxiliary vector is stored in target process stack. It contains 4128 * {type, value} pairs that we need to dump into note. This is not 4129 * strictly necessary but we do it here for sake of completeness. 4130 */ 4131 4132 /* read in whole auxv vector and copy it to memelfnote */ 4133 ptr = lock_user(VERIFY_READ, orig_auxv, len, 0); 4134 if (ptr != NULL) { 4135 fill_note(note, "CORE", NT_AUXV, len, ptr); 4136 unlock_user(ptr, auxv, len); 4137 } 4138 } 4139 4140 /* 4141 * Constructs name of coredump file. We have following convention 4142 * for the name: 4143 * qemu_<basename-of-target-binary>_<date>-<time>_<pid>.core 4144 * 4145 * Returns the filename 4146 */ 4147 static char *core_dump_filename(const TaskState *ts) 4148 { 4149 g_autoptr(GDateTime) now = g_date_time_new_now_local(); 4150 g_autofree char *nowstr = g_date_time_format(now, "%Y%m%d-%H%M%S"); 4151 g_autofree char *base_filename = g_path_get_basename(ts->bprm->filename); 4152 4153 return g_strdup_printf("qemu_%s_%s_%d.core", 4154 base_filename, nowstr, (int)getpid()); 4155 } 4156 4157 static int dump_write(int fd, const void *ptr, size_t size) 4158 { 4159 const char *bufp = (const char *)ptr; 4160 ssize_t bytes_written, bytes_left; 4161 struct rlimit dumpsize; 4162 off_t pos; 4163 4164 bytes_written = 0; 4165 getrlimit(RLIMIT_CORE, &dumpsize); 4166 if ((pos = lseek(fd, 0, SEEK_CUR))==-1) { 4167 if (errno == ESPIPE) { /* not a seekable stream */ 4168 bytes_left = size; 4169 } else { 4170 return pos; 4171 } 4172 } else { 4173 if (dumpsize.rlim_cur <= pos) { 4174 return -1; 4175 } else if (dumpsize.rlim_cur == RLIM_INFINITY) { 4176 bytes_left = size; 4177 } else { 4178 size_t limit_left=dumpsize.rlim_cur - pos; 4179 bytes_left = limit_left >= size ? size : limit_left ; 4180 } 4181 } 4182 4183 /* 4184 * In normal conditions, single write(2) should do but 4185 * in case of socket etc. this mechanism is more portable. 4186 */ 4187 do { 4188 bytes_written = write(fd, bufp, bytes_left); 4189 if (bytes_written < 0) { 4190 if (errno == EINTR) 4191 continue; 4192 return (-1); 4193 } else if (bytes_written == 0) { /* eof */ 4194 return (-1); 4195 } 4196 bufp += bytes_written; 4197 bytes_left -= bytes_written; 4198 } while (bytes_left > 0); 4199 4200 return (0); 4201 } 4202 4203 static int write_note(struct memelfnote *men, int fd) 4204 { 4205 struct elf_note en; 4206 4207 en.n_namesz = men->namesz; 4208 en.n_type = men->type; 4209 en.n_descsz = men->datasz; 4210 4211 bswap_note(&en); 4212 4213 if (dump_write(fd, &en, sizeof(en)) != 0) 4214 return (-1); 4215 if (dump_write(fd, men->name, men->namesz_rounded) != 0) 4216 return (-1); 4217 if (dump_write(fd, men->data, men->datasz_rounded) != 0) 4218 return (-1); 4219 4220 return (0); 4221 } 4222 4223 static void fill_thread_info(struct elf_note_info *info, const CPUArchState *env) 4224 { 4225 CPUState *cpu = env_cpu((CPUArchState *)env); 4226 TaskState *ts = (TaskState *)cpu->opaque; 4227 struct elf_thread_status *ets; 4228 4229 ets = g_malloc0(sizeof (*ets)); 4230 ets->num_notes = 1; /* only prstatus is dumped */ 4231 fill_prstatus(&ets->prstatus, ts, 0); 4232 elf_core_copy_regs(&ets->prstatus.pr_reg, env); 4233 fill_note(&ets->notes[0], "CORE", NT_PRSTATUS, sizeof (ets->prstatus), 4234 &ets->prstatus); 4235 4236 QTAILQ_INSERT_TAIL(&info->thread_list, ets, ets_link); 4237 4238 info->notes_size += note_size(&ets->notes[0]); 4239 } 4240 4241 static void init_note_info(struct elf_note_info *info) 4242 { 4243 /* Initialize the elf_note_info structure so that it is at 4244 * least safe to call free_note_info() on it. Must be 4245 * called before calling fill_note_info(). 4246 */ 4247 memset(info, 0, sizeof (*info)); 4248 QTAILQ_INIT(&info->thread_list); 4249 } 4250 4251 static int fill_note_info(struct elf_note_info *info, 4252 long signr, const CPUArchState *env) 4253 { 4254 #define NUMNOTES 3 4255 CPUState *cpu = env_cpu((CPUArchState *)env); 4256 TaskState *ts = (TaskState *)cpu->opaque; 4257 int i; 4258 4259 info->notes = g_new0(struct memelfnote, NUMNOTES); 4260 if (info->notes == NULL) 4261 return (-ENOMEM); 4262 info->prstatus = g_malloc0(sizeof (*info->prstatus)); 4263 if (info->prstatus == NULL) 4264 return (-ENOMEM); 4265 info->psinfo = g_malloc0(sizeof (*info->psinfo)); 4266 if (info->prstatus == NULL) 4267 return (-ENOMEM); 4268 4269 /* 4270 * First fill in status (and registers) of current thread 4271 * including process info & aux vector. 4272 */ 4273 fill_prstatus(info->prstatus, ts, signr); 4274 elf_core_copy_regs(&info->prstatus->pr_reg, env); 4275 fill_note(&info->notes[0], "CORE", NT_PRSTATUS, 4276 sizeof (*info->prstatus), info->prstatus); 4277 fill_psinfo(info->psinfo, ts); 4278 fill_note(&info->notes[1], "CORE", NT_PRPSINFO, 4279 sizeof (*info->psinfo), info->psinfo); 4280 fill_auxv_note(&info->notes[2], ts); 4281 info->numnote = 3; 4282 4283 info->notes_size = 0; 4284 for (i = 0; i < info->numnote; i++) 4285 info->notes_size += note_size(&info->notes[i]); 4286 4287 /* read and fill status of all threads */ 4288 WITH_QEMU_LOCK_GUARD(&qemu_cpu_list_lock) { 4289 CPU_FOREACH(cpu) { 4290 if (cpu == thread_cpu) { 4291 continue; 4292 } 4293 fill_thread_info(info, cpu->env_ptr); 4294 } 4295 } 4296 4297 return (0); 4298 } 4299 4300 static void free_note_info(struct elf_note_info *info) 4301 { 4302 struct elf_thread_status *ets; 4303 4304 while (!QTAILQ_EMPTY(&info->thread_list)) { 4305 ets = QTAILQ_FIRST(&info->thread_list); 4306 QTAILQ_REMOVE(&info->thread_list, ets, ets_link); 4307 g_free(ets); 4308 } 4309 4310 g_free(info->prstatus); 4311 g_free(info->psinfo); 4312 g_free(info->notes); 4313 } 4314 4315 static int write_note_info(struct elf_note_info *info, int fd) 4316 { 4317 struct elf_thread_status *ets; 4318 int i, error = 0; 4319 4320 /* write prstatus, psinfo and auxv for current thread */ 4321 for (i = 0; i < info->numnote; i++) 4322 if ((error = write_note(&info->notes[i], fd)) != 0) 4323 return (error); 4324 4325 /* write prstatus for each thread */ 4326 QTAILQ_FOREACH(ets, &info->thread_list, ets_link) { 4327 if ((error = write_note(&ets->notes[0], fd)) != 0) 4328 return (error); 4329 } 4330 4331 return (0); 4332 } 4333 4334 /* 4335 * Write out ELF coredump. 4336 * 4337 * See documentation of ELF object file format in: 4338 * http://www.caldera.com/developers/devspecs/gabi41.pdf 4339 * 4340 * Coredump format in linux is following: 4341 * 4342 * 0 +----------------------+ \ 4343 * | ELF header | ET_CORE | 4344 * +----------------------+ | 4345 * | ELF program headers | |--- headers 4346 * | - NOTE section | | 4347 * | - PT_LOAD sections | | 4348 * +----------------------+ / 4349 * | NOTEs: | 4350 * | - NT_PRSTATUS | 4351 * | - NT_PRSINFO | 4352 * | - NT_AUXV | 4353 * +----------------------+ <-- aligned to target page 4354 * | Process memory dump | 4355 * : : 4356 * . . 4357 * : : 4358 * | | 4359 * +----------------------+ 4360 * 4361 * NT_PRSTATUS -> struct elf_prstatus (per thread) 4362 * NT_PRSINFO -> struct elf_prpsinfo 4363 * NT_AUXV is array of { type, value } pairs (see fill_auxv_note()). 4364 * 4365 * Format follows System V format as close as possible. Current 4366 * version limitations are as follows: 4367 * - no floating point registers are dumped 4368 * 4369 * Function returns 0 in case of success, negative errno otherwise. 4370 * 4371 * TODO: make this work also during runtime: it should be 4372 * possible to force coredump from running process and then 4373 * continue processing. For example qemu could set up SIGUSR2 4374 * handler (provided that target process haven't registered 4375 * handler for that) that does the dump when signal is received. 4376 */ 4377 static int elf_core_dump(int signr, const CPUArchState *env) 4378 { 4379 const CPUState *cpu = env_cpu((CPUArchState *)env); 4380 const TaskState *ts = (const TaskState *)cpu->opaque; 4381 struct vm_area_struct *vma = NULL; 4382 g_autofree char *corefile = NULL; 4383 struct elf_note_info info; 4384 struct elfhdr elf; 4385 struct elf_phdr phdr; 4386 struct rlimit dumpsize; 4387 struct mm_struct *mm = NULL; 4388 off_t offset = 0, data_offset = 0; 4389 int segs = 0; 4390 int fd = -1; 4391 4392 init_note_info(&info); 4393 4394 errno = 0; 4395 getrlimit(RLIMIT_CORE, &dumpsize); 4396 if (dumpsize.rlim_cur == 0) 4397 return 0; 4398 4399 corefile = core_dump_filename(ts); 4400 4401 if ((fd = open(corefile, O_WRONLY | O_CREAT, 4402 S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH)) < 0) 4403 return (-errno); 4404 4405 /* 4406 * Walk through target process memory mappings and 4407 * set up structure containing this information. After 4408 * this point vma_xxx functions can be used. 4409 */ 4410 if ((mm = vma_init()) == NULL) 4411 goto out; 4412 4413 walk_memory_regions(mm, vma_walker); 4414 segs = vma_get_mapping_count(mm); 4415 4416 /* 4417 * Construct valid coredump ELF header. We also 4418 * add one more segment for notes. 4419 */ 4420 fill_elf_header(&elf, segs + 1, ELF_MACHINE, 0); 4421 if (dump_write(fd, &elf, sizeof (elf)) != 0) 4422 goto out; 4423 4424 /* fill in the in-memory version of notes */ 4425 if (fill_note_info(&info, signr, env) < 0) 4426 goto out; 4427 4428 offset += sizeof (elf); /* elf header */ 4429 offset += (segs + 1) * sizeof (struct elf_phdr); /* program headers */ 4430 4431 /* write out notes program header */ 4432 fill_elf_note_phdr(&phdr, info.notes_size, offset); 4433 4434 offset += info.notes_size; 4435 if (dump_write(fd, &phdr, sizeof (phdr)) != 0) 4436 goto out; 4437 4438 /* 4439 * ELF specification wants data to start at page boundary so 4440 * we align it here. 4441 */ 4442 data_offset = offset = roundup(offset, ELF_EXEC_PAGESIZE); 4443 4444 /* 4445 * Write program headers for memory regions mapped in 4446 * the target process. 4447 */ 4448 for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) { 4449 (void) memset(&phdr, 0, sizeof (phdr)); 4450 4451 phdr.p_type = PT_LOAD; 4452 phdr.p_offset = offset; 4453 phdr.p_vaddr = vma->vma_start; 4454 phdr.p_paddr = 0; 4455 phdr.p_filesz = vma_dump_size(vma); 4456 offset += phdr.p_filesz; 4457 phdr.p_memsz = vma->vma_end - vma->vma_start; 4458 phdr.p_flags = vma->vma_flags & PROT_READ ? PF_R : 0; 4459 if (vma->vma_flags & PROT_WRITE) 4460 phdr.p_flags |= PF_W; 4461 if (vma->vma_flags & PROT_EXEC) 4462 phdr.p_flags |= PF_X; 4463 phdr.p_align = ELF_EXEC_PAGESIZE; 4464 4465 bswap_phdr(&phdr, 1); 4466 if (dump_write(fd, &phdr, sizeof(phdr)) != 0) { 4467 goto out; 4468 } 4469 } 4470 4471 /* 4472 * Next we write notes just after program headers. No 4473 * alignment needed here. 4474 */ 4475 if (write_note_info(&info, fd) < 0) 4476 goto out; 4477 4478 /* align data to page boundary */ 4479 if (lseek(fd, data_offset, SEEK_SET) != data_offset) 4480 goto out; 4481 4482 /* 4483 * Finally we can dump process memory into corefile as well. 4484 */ 4485 for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) { 4486 abi_ulong addr; 4487 abi_ulong end; 4488 4489 end = vma->vma_start + vma_dump_size(vma); 4490 4491 for (addr = vma->vma_start; addr < end; 4492 addr += TARGET_PAGE_SIZE) { 4493 char page[TARGET_PAGE_SIZE]; 4494 int error; 4495 4496 /* 4497 * Read in page from target process memory and 4498 * write it to coredump file. 4499 */ 4500 error = copy_from_user(page, addr, sizeof (page)); 4501 if (error != 0) { 4502 (void) fprintf(stderr, "unable to dump " TARGET_ABI_FMT_lx "\n", 4503 addr); 4504 errno = -error; 4505 goto out; 4506 } 4507 if (dump_write(fd, page, TARGET_PAGE_SIZE) < 0) 4508 goto out; 4509 } 4510 } 4511 4512 out: 4513 free_note_info(&info); 4514 if (mm != NULL) 4515 vma_delete(mm); 4516 (void) close(fd); 4517 4518 if (errno != 0) 4519 return (-errno); 4520 return (0); 4521 } 4522 #endif /* USE_ELF_CORE_DUMP */ 4523 4524 void do_init_thread(struct target_pt_regs *regs, struct image_info *infop) 4525 { 4526 init_thread(regs, infop); 4527 } 4528