apic: Don't iterate past last used apic
[qemu/ar7.git] / linux-user / elfload.c
blob33d776de41a69368854d85c17e4e6b181168faa0
1 /* This is the Linux kernel elf-loading code, ported into user space */
2 #include <sys/time.h>
3 #include <sys/param.h>
5 #include <stdio.h>
6 #include <sys/types.h>
7 #include <fcntl.h>
8 #include <errno.h>
9 #include <unistd.h>
10 #include <sys/mman.h>
11 #include <sys/resource.h>
12 #include <stdlib.h>
13 #include <string.h>
14 #include <time.h>
16 #include "qemu.h"
17 #include "disas.h"
19 #ifdef _ARCH_PPC64
20 #undef ARCH_DLINFO
21 #undef ELF_PLATFORM
22 #undef ELF_HWCAP
23 #undef ELF_CLASS
24 #undef ELF_DATA
25 #undef ELF_ARCH
26 #endif
28 #define ELF_OSABI ELFOSABI_SYSV
30 /* from personality.h */
33 * Flags for bug emulation.
35 * These occupy the top three bytes.
37 enum {
38 ADDR_NO_RANDOMIZE = 0x0040000, /* disable randomization of VA space */
39 FDPIC_FUNCPTRS = 0x0080000, /* userspace function ptrs point to
40 descriptors (signal handling) */
41 MMAP_PAGE_ZERO = 0x0100000,
42 ADDR_COMPAT_LAYOUT = 0x0200000,
43 READ_IMPLIES_EXEC = 0x0400000,
44 ADDR_LIMIT_32BIT = 0x0800000,
45 SHORT_INODE = 0x1000000,
46 WHOLE_SECONDS = 0x2000000,
47 STICKY_TIMEOUTS = 0x4000000,
48 ADDR_LIMIT_3GB = 0x8000000,
52 * Personality types.
54 * These go in the low byte. Avoid using the top bit, it will
55 * conflict with error returns.
57 enum {
58 PER_LINUX = 0x0000,
59 PER_LINUX_32BIT = 0x0000 | ADDR_LIMIT_32BIT,
60 PER_LINUX_FDPIC = 0x0000 | FDPIC_FUNCPTRS,
61 PER_SVR4 = 0x0001 | STICKY_TIMEOUTS | MMAP_PAGE_ZERO,
62 PER_SVR3 = 0x0002 | STICKY_TIMEOUTS | SHORT_INODE,
63 PER_SCOSVR3 = 0x0003 | STICKY_TIMEOUTS | WHOLE_SECONDS | SHORT_INODE,
64 PER_OSR5 = 0x0003 | STICKY_TIMEOUTS | WHOLE_SECONDS,
65 PER_WYSEV386 = 0x0004 | STICKY_TIMEOUTS | SHORT_INODE,
66 PER_ISCR4 = 0x0005 | STICKY_TIMEOUTS,
67 PER_BSD = 0x0006,
68 PER_SUNOS = 0x0006 | STICKY_TIMEOUTS,
69 PER_XENIX = 0x0007 | STICKY_TIMEOUTS | SHORT_INODE,
70 PER_LINUX32 = 0x0008,
71 PER_LINUX32_3GB = 0x0008 | ADDR_LIMIT_3GB,
72 PER_IRIX32 = 0x0009 | STICKY_TIMEOUTS,/* IRIX5 32-bit */
73 PER_IRIXN32 = 0x000a | STICKY_TIMEOUTS,/* IRIX6 new 32-bit */
74 PER_IRIX64 = 0x000b | STICKY_TIMEOUTS,/* IRIX6 64-bit */
75 PER_RISCOS = 0x000c,
76 PER_SOLARIS = 0x000d | STICKY_TIMEOUTS,
77 PER_UW7 = 0x000e | STICKY_TIMEOUTS | MMAP_PAGE_ZERO,
78 PER_OSF4 = 0x000f, /* OSF/1 v4 */
79 PER_HPUX = 0x0010,
80 PER_MASK = 0x00ff,
84 * Return the base personality without flags.
86 #define personality(pers) (pers & PER_MASK)
88 /* this flag is uneffective under linux too, should be deleted */
89 #ifndef MAP_DENYWRITE
90 #define MAP_DENYWRITE 0
91 #endif
93 /* should probably go in elf.h */
94 #ifndef ELIBBAD
95 #define ELIBBAD 80
96 #endif
98 #ifdef TARGET_WORDS_BIGENDIAN
99 #define ELF_DATA ELFDATA2MSB
100 #else
101 #define ELF_DATA ELFDATA2LSB
102 #endif
104 typedef target_ulong target_elf_greg_t;
105 #ifdef USE_UID16
106 typedef uint16_t target_uid_t;
107 typedef uint16_t target_gid_t;
108 #else
109 typedef uint32_t target_uid_t;
110 typedef uint32_t target_gid_t;
111 #endif
112 typedef int32_t target_pid_t;
114 #ifdef TARGET_I386
116 #define ELF_PLATFORM get_elf_platform()
118 static const char *get_elf_platform(void)
120 static char elf_platform[] = "i386";
121 int family = (thread_env->cpuid_version >> 8) & 0xff;
122 if (family > 6)
123 family = 6;
124 if (family >= 3)
125 elf_platform[1] = '0' + family;
126 return elf_platform;
129 #define ELF_HWCAP get_elf_hwcap()
131 static uint32_t get_elf_hwcap(void)
133 return thread_env->cpuid_features;
136 #ifdef TARGET_X86_64
137 #define ELF_START_MMAP 0x2aaaaab000ULL
138 #define elf_check_arch(x) ( ((x) == ELF_ARCH) )
140 #define ELF_CLASS ELFCLASS64
141 #define ELF_ARCH EM_X86_64
143 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop)
145 regs->rax = 0;
146 regs->rsp = infop->start_stack;
147 regs->rip = infop->entry;
150 #define ELF_NREG 27
151 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
154 * Note that ELF_NREG should be 29 as there should be place for
155 * TRAPNO and ERR "registers" as well but linux doesn't dump
156 * those.
158 * See linux kernel: arch/x86/include/asm/elf.h
160 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUState *env)
162 (*regs)[0] = env->regs[15];
163 (*regs)[1] = env->regs[14];
164 (*regs)[2] = env->regs[13];
165 (*regs)[3] = env->regs[12];
166 (*regs)[4] = env->regs[R_EBP];
167 (*regs)[5] = env->regs[R_EBX];
168 (*regs)[6] = env->regs[11];
169 (*regs)[7] = env->regs[10];
170 (*regs)[8] = env->regs[9];
171 (*regs)[9] = env->regs[8];
172 (*regs)[10] = env->regs[R_EAX];
173 (*regs)[11] = env->regs[R_ECX];
174 (*regs)[12] = env->regs[R_EDX];
175 (*regs)[13] = env->regs[R_ESI];
176 (*regs)[14] = env->regs[R_EDI];
177 (*regs)[15] = env->regs[R_EAX]; /* XXX */
178 (*regs)[16] = env->eip;
179 (*regs)[17] = env->segs[R_CS].selector & 0xffff;
180 (*regs)[18] = env->eflags;
181 (*regs)[19] = env->regs[R_ESP];
182 (*regs)[20] = env->segs[R_SS].selector & 0xffff;
183 (*regs)[21] = env->segs[R_FS].selector & 0xffff;
184 (*regs)[22] = env->segs[R_GS].selector & 0xffff;
185 (*regs)[23] = env->segs[R_DS].selector & 0xffff;
186 (*regs)[24] = env->segs[R_ES].selector & 0xffff;
187 (*regs)[25] = env->segs[R_FS].selector & 0xffff;
188 (*regs)[26] = env->segs[R_GS].selector & 0xffff;
191 #else
193 #define ELF_START_MMAP 0x80000000
196 * This is used to ensure we don't load something for the wrong architecture.
198 #define elf_check_arch(x) ( ((x) == EM_386) || ((x) == EM_486) )
201 * These are used to set parameters in the core dumps.
203 #define ELF_CLASS ELFCLASS32
204 #define ELF_ARCH EM_386
206 static inline void init_thread(struct target_pt_regs *regs,
207 struct image_info *infop)
209 regs->esp = infop->start_stack;
210 regs->eip = infop->entry;
212 /* SVR4/i386 ABI (pages 3-31, 3-32) says that when the program
213 starts %edx contains a pointer to a function which might be
214 registered using `atexit'. This provides a mean for the
215 dynamic linker to call DT_FINI functions for shared libraries
216 that have been loaded before the code runs.
218 A value of 0 tells we have no such handler. */
219 regs->edx = 0;
222 #define ELF_NREG 17
223 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
226 * Note that ELF_NREG should be 19 as there should be place for
227 * TRAPNO and ERR "registers" as well but linux doesn't dump
228 * those.
230 * See linux kernel: arch/x86/include/asm/elf.h
232 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUState *env)
234 (*regs)[0] = env->regs[R_EBX];
235 (*regs)[1] = env->regs[R_ECX];
236 (*regs)[2] = env->regs[R_EDX];
237 (*regs)[3] = env->regs[R_ESI];
238 (*regs)[4] = env->regs[R_EDI];
239 (*regs)[5] = env->regs[R_EBP];
240 (*regs)[6] = env->regs[R_EAX];
241 (*regs)[7] = env->segs[R_DS].selector & 0xffff;
242 (*regs)[8] = env->segs[R_ES].selector & 0xffff;
243 (*regs)[9] = env->segs[R_FS].selector & 0xffff;
244 (*regs)[10] = env->segs[R_GS].selector & 0xffff;
245 (*regs)[11] = env->regs[R_EAX]; /* XXX */
246 (*regs)[12] = env->eip;
247 (*regs)[13] = env->segs[R_CS].selector & 0xffff;
248 (*regs)[14] = env->eflags;
249 (*regs)[15] = env->regs[R_ESP];
250 (*regs)[16] = env->segs[R_SS].selector & 0xffff;
252 #endif
254 #define USE_ELF_CORE_DUMP
255 #define ELF_EXEC_PAGESIZE 4096
257 #endif
259 #ifdef TARGET_ARM
261 #define ELF_START_MMAP 0x80000000
263 #define elf_check_arch(x) ( (x) == EM_ARM )
265 #define ELF_CLASS ELFCLASS32
266 #define ELF_ARCH EM_ARM
268 static inline void init_thread(struct target_pt_regs *regs,
269 struct image_info *infop)
271 abi_long stack = infop->start_stack;
272 memset(regs, 0, sizeof(*regs));
273 regs->ARM_cpsr = 0x10;
274 if (infop->entry & 1)
275 regs->ARM_cpsr |= CPSR_T;
276 regs->ARM_pc = infop->entry & 0xfffffffe;
277 regs->ARM_sp = infop->start_stack;
278 /* FIXME - what to for failure of get_user()? */
279 get_user_ual(regs->ARM_r2, stack + 8); /* envp */
280 get_user_ual(regs->ARM_r1, stack + 4); /* envp */
281 /* XXX: it seems that r0 is zeroed after ! */
282 regs->ARM_r0 = 0;
283 /* For uClinux PIC binaries. */
284 /* XXX: Linux does this only on ARM with no MMU (do we care ?) */
285 regs->ARM_r10 = infop->start_data;
288 #define ELF_NREG 18
289 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
291 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUState *env)
293 (*regs)[0] = tswapl(env->regs[0]);
294 (*regs)[1] = tswapl(env->regs[1]);
295 (*regs)[2] = tswapl(env->regs[2]);
296 (*regs)[3] = tswapl(env->regs[3]);
297 (*regs)[4] = tswapl(env->regs[4]);
298 (*regs)[5] = tswapl(env->regs[5]);
299 (*regs)[6] = tswapl(env->regs[6]);
300 (*regs)[7] = tswapl(env->regs[7]);
301 (*regs)[8] = tswapl(env->regs[8]);
302 (*regs)[9] = tswapl(env->regs[9]);
303 (*regs)[10] = tswapl(env->regs[10]);
304 (*regs)[11] = tswapl(env->regs[11]);
305 (*regs)[12] = tswapl(env->regs[12]);
306 (*regs)[13] = tswapl(env->regs[13]);
307 (*regs)[14] = tswapl(env->regs[14]);
308 (*regs)[15] = tswapl(env->regs[15]);
310 (*regs)[16] = tswapl(cpsr_read((CPUState *)env));
311 (*regs)[17] = tswapl(env->regs[0]); /* XXX */
314 #define USE_ELF_CORE_DUMP
315 #define ELF_EXEC_PAGESIZE 4096
317 enum
319 ARM_HWCAP_ARM_SWP = 1 << 0,
320 ARM_HWCAP_ARM_HALF = 1 << 1,
321 ARM_HWCAP_ARM_THUMB = 1 << 2,
322 ARM_HWCAP_ARM_26BIT = 1 << 3,
323 ARM_HWCAP_ARM_FAST_MULT = 1 << 4,
324 ARM_HWCAP_ARM_FPA = 1 << 5,
325 ARM_HWCAP_ARM_VFP = 1 << 6,
326 ARM_HWCAP_ARM_EDSP = 1 << 7,
327 ARM_HWCAP_ARM_JAVA = 1 << 8,
328 ARM_HWCAP_ARM_IWMMXT = 1 << 9,
329 ARM_HWCAP_ARM_THUMBEE = 1 << 10,
330 ARM_HWCAP_ARM_NEON = 1 << 11,
331 ARM_HWCAP_ARM_VFPv3 = 1 << 12,
332 ARM_HWCAP_ARM_VFPv3D16 = 1 << 13,
335 #define ELF_HWCAP (ARM_HWCAP_ARM_SWP | ARM_HWCAP_ARM_HALF \
336 | ARM_HWCAP_ARM_THUMB | ARM_HWCAP_ARM_FAST_MULT \
337 | ARM_HWCAP_ARM_FPA | ARM_HWCAP_ARM_VFP \
338 | ARM_HWCAP_ARM_NEON | ARM_HWCAP_ARM_VFPv3 )
340 #endif
342 #ifdef TARGET_SPARC
343 #ifdef TARGET_SPARC64
345 #define ELF_START_MMAP 0x80000000
347 #ifndef TARGET_ABI32
348 #define elf_check_arch(x) ( (x) == EM_SPARCV9 || (x) == EM_SPARC32PLUS )
349 #else
350 #define elf_check_arch(x) ( (x) == EM_SPARC32PLUS || (x) == EM_SPARC )
351 #endif
353 #define ELF_CLASS ELFCLASS64
354 #define ELF_ARCH EM_SPARCV9
356 #define STACK_BIAS 2047
358 static inline void init_thread(struct target_pt_regs *regs,
359 struct image_info *infop)
361 #ifndef TARGET_ABI32
362 regs->tstate = 0;
363 #endif
364 regs->pc = infop->entry;
365 regs->npc = regs->pc + 4;
366 regs->y = 0;
367 #ifdef TARGET_ABI32
368 regs->u_regs[14] = infop->start_stack - 16 * 4;
369 #else
370 if (personality(infop->personality) == PER_LINUX32)
371 regs->u_regs[14] = infop->start_stack - 16 * 4;
372 else
373 regs->u_regs[14] = infop->start_stack - 16 * 8 - STACK_BIAS;
374 #endif
377 #else
378 #define ELF_START_MMAP 0x80000000
380 #define elf_check_arch(x) ( (x) == EM_SPARC )
382 #define ELF_CLASS ELFCLASS32
383 #define ELF_ARCH EM_SPARC
385 static inline void init_thread(struct target_pt_regs *regs,
386 struct image_info *infop)
388 regs->psr = 0;
389 regs->pc = infop->entry;
390 regs->npc = regs->pc + 4;
391 regs->y = 0;
392 regs->u_regs[14] = infop->start_stack - 16 * 4;
395 #endif
396 #endif
398 #ifdef TARGET_PPC
400 #define ELF_START_MMAP 0x80000000
402 #if defined(TARGET_PPC64) && !defined(TARGET_ABI32)
404 #define elf_check_arch(x) ( (x) == EM_PPC64 )
406 #define ELF_CLASS ELFCLASS64
408 #else
410 #define elf_check_arch(x) ( (x) == EM_PPC )
412 #define ELF_CLASS ELFCLASS32
414 #endif
416 #define ELF_ARCH EM_PPC
418 /* Feature masks for the Aux Vector Hardware Capabilities (AT_HWCAP).
419 See arch/powerpc/include/asm/cputable.h. */
420 enum {
421 QEMU_PPC_FEATURE_32 = 0x80000000,
422 QEMU_PPC_FEATURE_64 = 0x40000000,
423 QEMU_PPC_FEATURE_601_INSTR = 0x20000000,
424 QEMU_PPC_FEATURE_HAS_ALTIVEC = 0x10000000,
425 QEMU_PPC_FEATURE_HAS_FPU = 0x08000000,
426 QEMU_PPC_FEATURE_HAS_MMU = 0x04000000,
427 QEMU_PPC_FEATURE_HAS_4xxMAC = 0x02000000,
428 QEMU_PPC_FEATURE_UNIFIED_CACHE = 0x01000000,
429 QEMU_PPC_FEATURE_HAS_SPE = 0x00800000,
430 QEMU_PPC_FEATURE_HAS_EFP_SINGLE = 0x00400000,
431 QEMU_PPC_FEATURE_HAS_EFP_DOUBLE = 0x00200000,
432 QEMU_PPC_FEATURE_NO_TB = 0x00100000,
433 QEMU_PPC_FEATURE_POWER4 = 0x00080000,
434 QEMU_PPC_FEATURE_POWER5 = 0x00040000,
435 QEMU_PPC_FEATURE_POWER5_PLUS = 0x00020000,
436 QEMU_PPC_FEATURE_CELL = 0x00010000,
437 QEMU_PPC_FEATURE_BOOKE = 0x00008000,
438 QEMU_PPC_FEATURE_SMT = 0x00004000,
439 QEMU_PPC_FEATURE_ICACHE_SNOOP = 0x00002000,
440 QEMU_PPC_FEATURE_ARCH_2_05 = 0x00001000,
441 QEMU_PPC_FEATURE_PA6T = 0x00000800,
442 QEMU_PPC_FEATURE_HAS_DFP = 0x00000400,
443 QEMU_PPC_FEATURE_POWER6_EXT = 0x00000200,
444 QEMU_PPC_FEATURE_ARCH_2_06 = 0x00000100,
445 QEMU_PPC_FEATURE_HAS_VSX = 0x00000080,
446 QEMU_PPC_FEATURE_PSERIES_PERFMON_COMPAT = 0x00000040,
448 QEMU_PPC_FEATURE_TRUE_LE = 0x00000002,
449 QEMU_PPC_FEATURE_PPC_LE = 0x00000001,
452 #define ELF_HWCAP get_elf_hwcap()
454 static uint32_t get_elf_hwcap(void)
456 CPUState *e = thread_env;
457 uint32_t features = 0;
459 /* We don't have to be terribly complete here; the high points are
460 Altivec/FP/SPE support. Anything else is just a bonus. */
461 #define GET_FEATURE(flag, feature) \
462 do {if (e->insns_flags & flag) features |= feature; } while(0)
463 GET_FEATURE(PPC_64B, QEMU_PPC_FEATURE_64);
464 GET_FEATURE(PPC_FLOAT, QEMU_PPC_FEATURE_HAS_FPU);
465 GET_FEATURE(PPC_ALTIVEC, QEMU_PPC_FEATURE_HAS_ALTIVEC);
466 GET_FEATURE(PPC_SPE, QEMU_PPC_FEATURE_HAS_SPE);
467 GET_FEATURE(PPC_SPE_SINGLE, QEMU_PPC_FEATURE_HAS_EFP_SINGLE);
468 GET_FEATURE(PPC_SPE_DOUBLE, QEMU_PPC_FEATURE_HAS_EFP_DOUBLE);
469 GET_FEATURE(PPC_BOOKE, QEMU_PPC_FEATURE_BOOKE);
470 GET_FEATURE(PPC_405_MAC, QEMU_PPC_FEATURE_HAS_4xxMAC);
471 #undef GET_FEATURE
473 return features;
477 * The requirements here are:
478 * - keep the final alignment of sp (sp & 0xf)
479 * - make sure the 32-bit value at the first 16 byte aligned position of
480 * AUXV is greater than 16 for glibc compatibility.
481 * AT_IGNOREPPC is used for that.
482 * - for compatibility with glibc ARCH_DLINFO must always be defined on PPC,
483 * even if DLINFO_ARCH_ITEMS goes to zero or is undefined.
485 #define DLINFO_ARCH_ITEMS 5
486 #define ARCH_DLINFO \
487 do { \
488 NEW_AUX_ENT(AT_DCACHEBSIZE, 0x20); \
489 NEW_AUX_ENT(AT_ICACHEBSIZE, 0x20); \
490 NEW_AUX_ENT(AT_UCACHEBSIZE, 0); \
491 /* \
492 * Now handle glibc compatibility. \
493 */ \
494 NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC); \
495 NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC); \
496 } while (0)
498 static inline void init_thread(struct target_pt_regs *_regs, struct image_info *infop)
500 _regs->gpr[1] = infop->start_stack;
501 #if defined(TARGET_PPC64) && !defined(TARGET_ABI32)
502 _regs->gpr[2] = ldq_raw(infop->entry + 8) + infop->load_addr;
503 infop->entry = ldq_raw(infop->entry) + infop->load_addr;
504 #endif
505 _regs->nip = infop->entry;
508 /* See linux kernel: arch/powerpc/include/asm/elf.h. */
509 #define ELF_NREG 48
510 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
512 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUState *env)
514 int i;
515 target_ulong ccr = 0;
517 for (i = 0; i < ARRAY_SIZE(env->gpr); i++) {
518 (*regs)[i] = tswapl(env->gpr[i]);
521 (*regs)[32] = tswapl(env->nip);
522 (*regs)[33] = tswapl(env->msr);
523 (*regs)[35] = tswapl(env->ctr);
524 (*regs)[36] = tswapl(env->lr);
525 (*regs)[37] = tswapl(env->xer);
527 for (i = 0; i < ARRAY_SIZE(env->crf); i++) {
528 ccr |= env->crf[i] << (32 - ((i + 1) * 4));
530 (*regs)[38] = tswapl(ccr);
533 #define USE_ELF_CORE_DUMP
534 #define ELF_EXEC_PAGESIZE 4096
536 #endif
538 #ifdef TARGET_MIPS
540 #define ELF_START_MMAP 0x80000000
542 #define elf_check_arch(x) ( (x) == EM_MIPS )
544 #ifdef TARGET_MIPS64
545 #define ELF_CLASS ELFCLASS64
546 #else
547 #define ELF_CLASS ELFCLASS32
548 #endif
549 #define ELF_ARCH EM_MIPS
551 static inline void init_thread(struct target_pt_regs *regs,
552 struct image_info *infop)
554 regs->cp0_status = 2 << CP0St_KSU;
555 regs->cp0_epc = infop->entry;
556 regs->regs[29] = infop->start_stack;
559 /* See linux kernel: arch/mips/include/asm/elf.h. */
560 #define ELF_NREG 45
561 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
563 /* See linux kernel: arch/mips/include/asm/reg.h. */
564 enum {
565 #ifdef TARGET_MIPS64
566 TARGET_EF_R0 = 0,
567 #else
568 TARGET_EF_R0 = 6,
569 #endif
570 TARGET_EF_R26 = TARGET_EF_R0 + 26,
571 TARGET_EF_R27 = TARGET_EF_R0 + 27,
572 TARGET_EF_LO = TARGET_EF_R0 + 32,
573 TARGET_EF_HI = TARGET_EF_R0 + 33,
574 TARGET_EF_CP0_EPC = TARGET_EF_R0 + 34,
575 TARGET_EF_CP0_BADVADDR = TARGET_EF_R0 + 35,
576 TARGET_EF_CP0_STATUS = TARGET_EF_R0 + 36,
577 TARGET_EF_CP0_CAUSE = TARGET_EF_R0 + 37
580 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */
581 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUState *env)
583 int i;
585 for (i = 0; i < TARGET_EF_R0; i++) {
586 (*regs)[i] = 0;
588 (*regs)[TARGET_EF_R0] = 0;
590 for (i = 1; i < ARRAY_SIZE(env->active_tc.gpr); i++) {
591 (*regs)[TARGET_EF_R0 + i] = tswapl(env->active_tc.gpr[i]);
594 (*regs)[TARGET_EF_R26] = 0;
595 (*regs)[TARGET_EF_R27] = 0;
596 (*regs)[TARGET_EF_LO] = tswapl(env->active_tc.LO[0]);
597 (*regs)[TARGET_EF_HI] = tswapl(env->active_tc.HI[0]);
598 (*regs)[TARGET_EF_CP0_EPC] = tswapl(env->active_tc.PC);
599 (*regs)[TARGET_EF_CP0_BADVADDR] = tswapl(env->CP0_BadVAddr);
600 (*regs)[TARGET_EF_CP0_STATUS] = tswapl(env->CP0_Status);
601 (*regs)[TARGET_EF_CP0_CAUSE] = tswapl(env->CP0_Cause);
604 #define USE_ELF_CORE_DUMP
605 #define ELF_EXEC_PAGESIZE 4096
607 #endif /* TARGET_MIPS */
609 #ifdef TARGET_MICROBLAZE
611 #define ELF_START_MMAP 0x80000000
613 #define elf_check_arch(x) ( (x) == EM_MICROBLAZE || (x) == EM_MICROBLAZE_OLD)
615 #define ELF_CLASS ELFCLASS32
616 #define ELF_ARCH EM_MICROBLAZE
618 static inline void init_thread(struct target_pt_regs *regs,
619 struct image_info *infop)
621 regs->pc = infop->entry;
622 regs->r1 = infop->start_stack;
626 #define ELF_EXEC_PAGESIZE 4096
628 #define USE_ELF_CORE_DUMP
629 #define ELF_NREG 38
630 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
632 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */
633 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUState *env)
635 int i, pos = 0;
637 for (i = 0; i < 32; i++) {
638 (*regs)[pos++] = tswapl(env->regs[i]);
641 for (i = 0; i < 6; i++) {
642 (*regs)[pos++] = tswapl(env->sregs[i]);
646 #endif /* TARGET_MICROBLAZE */
648 #ifdef TARGET_SH4
650 #define ELF_START_MMAP 0x80000000
652 #define elf_check_arch(x) ( (x) == EM_SH )
654 #define ELF_CLASS ELFCLASS32
655 #define ELF_ARCH EM_SH
657 static inline void init_thread(struct target_pt_regs *regs,
658 struct image_info *infop)
660 /* Check other registers XXXXX */
661 regs->pc = infop->entry;
662 regs->regs[15] = infop->start_stack;
665 /* See linux kernel: arch/sh/include/asm/elf.h. */
666 #define ELF_NREG 23
667 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
669 /* See linux kernel: arch/sh/include/asm/ptrace.h. */
670 enum {
671 TARGET_REG_PC = 16,
672 TARGET_REG_PR = 17,
673 TARGET_REG_SR = 18,
674 TARGET_REG_GBR = 19,
675 TARGET_REG_MACH = 20,
676 TARGET_REG_MACL = 21,
677 TARGET_REG_SYSCALL = 22
680 static inline void elf_core_copy_regs(target_elf_gregset_t *regs,
681 const CPUState *env)
683 int i;
685 for (i = 0; i < 16; i++) {
686 (*regs[i]) = tswapl(env->gregs[i]);
689 (*regs)[TARGET_REG_PC] = tswapl(env->pc);
690 (*regs)[TARGET_REG_PR] = tswapl(env->pr);
691 (*regs)[TARGET_REG_SR] = tswapl(env->sr);
692 (*regs)[TARGET_REG_GBR] = tswapl(env->gbr);
693 (*regs)[TARGET_REG_MACH] = tswapl(env->mach);
694 (*regs)[TARGET_REG_MACL] = tswapl(env->macl);
695 (*regs)[TARGET_REG_SYSCALL] = 0; /* FIXME */
698 #define USE_ELF_CORE_DUMP
699 #define ELF_EXEC_PAGESIZE 4096
701 #endif
703 #ifdef TARGET_CRIS
705 #define ELF_START_MMAP 0x80000000
707 #define elf_check_arch(x) ( (x) == EM_CRIS )
709 #define ELF_CLASS ELFCLASS32
710 #define ELF_ARCH EM_CRIS
712 static inline void init_thread(struct target_pt_regs *regs,
713 struct image_info *infop)
715 regs->erp = infop->entry;
718 #define ELF_EXEC_PAGESIZE 8192
720 #endif
722 #ifdef TARGET_M68K
724 #define ELF_START_MMAP 0x80000000
726 #define elf_check_arch(x) ( (x) == EM_68K )
728 #define ELF_CLASS ELFCLASS32
729 #define ELF_ARCH EM_68K
731 /* ??? Does this need to do anything?
732 #define ELF_PLAT_INIT(_r) */
734 static inline void init_thread(struct target_pt_regs *regs,
735 struct image_info *infop)
737 regs->usp = infop->start_stack;
738 regs->sr = 0;
739 regs->pc = infop->entry;
742 /* See linux kernel: arch/m68k/include/asm/elf.h. */
743 #define ELF_NREG 20
744 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
746 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUState *env)
748 (*regs)[0] = tswapl(env->dregs[1]);
749 (*regs)[1] = tswapl(env->dregs[2]);
750 (*regs)[2] = tswapl(env->dregs[3]);
751 (*regs)[3] = tswapl(env->dregs[4]);
752 (*regs)[4] = tswapl(env->dregs[5]);
753 (*regs)[5] = tswapl(env->dregs[6]);
754 (*regs)[6] = tswapl(env->dregs[7]);
755 (*regs)[7] = tswapl(env->aregs[0]);
756 (*regs)[8] = tswapl(env->aregs[1]);
757 (*regs)[9] = tswapl(env->aregs[2]);
758 (*regs)[10] = tswapl(env->aregs[3]);
759 (*regs)[11] = tswapl(env->aregs[4]);
760 (*regs)[12] = tswapl(env->aregs[5]);
761 (*regs)[13] = tswapl(env->aregs[6]);
762 (*regs)[14] = tswapl(env->dregs[0]);
763 (*regs)[15] = tswapl(env->aregs[7]);
764 (*regs)[16] = tswapl(env->dregs[0]); /* FIXME: orig_d0 */
765 (*regs)[17] = tswapl(env->sr);
766 (*regs)[18] = tswapl(env->pc);
767 (*regs)[19] = 0; /* FIXME: regs->format | regs->vector */
770 #define USE_ELF_CORE_DUMP
771 #define ELF_EXEC_PAGESIZE 8192
773 #endif
775 #ifdef TARGET_ALPHA
777 #define ELF_START_MMAP (0x30000000000ULL)
779 #define elf_check_arch(x) ( (x) == ELF_ARCH )
781 #define ELF_CLASS ELFCLASS64
782 #define ELF_ARCH EM_ALPHA
784 static inline void init_thread(struct target_pt_regs *regs,
785 struct image_info *infop)
787 regs->pc = infop->entry;
788 regs->ps = 8;
789 regs->usp = infop->start_stack;
792 #define ELF_EXEC_PAGESIZE 8192
794 #endif /* TARGET_ALPHA */
796 #ifndef ELF_PLATFORM
797 #define ELF_PLATFORM (NULL)
798 #endif
800 #ifndef ELF_HWCAP
801 #define ELF_HWCAP 0
802 #endif
804 #ifdef TARGET_ABI32
805 #undef ELF_CLASS
806 #define ELF_CLASS ELFCLASS32
807 #undef bswaptls
808 #define bswaptls(ptr) bswap32s(ptr)
809 #endif
811 #include "elf.h"
813 struct exec
815 unsigned int a_info; /* Use macros N_MAGIC, etc for access */
816 unsigned int a_text; /* length of text, in bytes */
817 unsigned int a_data; /* length of data, in bytes */
818 unsigned int a_bss; /* length of uninitialized data area, in bytes */
819 unsigned int a_syms; /* length of symbol table data in file, in bytes */
820 unsigned int a_entry; /* start address */
821 unsigned int a_trsize; /* length of relocation info for text, in bytes */
822 unsigned int a_drsize; /* length of relocation info for data, in bytes */
826 #define N_MAGIC(exec) ((exec).a_info & 0xffff)
827 #define OMAGIC 0407
828 #define NMAGIC 0410
829 #define ZMAGIC 0413
830 #define QMAGIC 0314
832 /* Necessary parameters */
833 #define TARGET_ELF_EXEC_PAGESIZE TARGET_PAGE_SIZE
834 #define TARGET_ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(TARGET_ELF_EXEC_PAGESIZE-1))
835 #define TARGET_ELF_PAGEOFFSET(_v) ((_v) & (TARGET_ELF_EXEC_PAGESIZE-1))
837 #define DLINFO_ITEMS 12
839 static inline void memcpy_fromfs(void * to, const void * from, unsigned long n)
841 memcpy(to, from, n);
844 #ifdef BSWAP_NEEDED
845 static void bswap_ehdr(struct elfhdr *ehdr)
847 bswap16s(&ehdr->e_type); /* Object file type */
848 bswap16s(&ehdr->e_machine); /* Architecture */
849 bswap32s(&ehdr->e_version); /* Object file version */
850 bswaptls(&ehdr->e_entry); /* Entry point virtual address */
851 bswaptls(&ehdr->e_phoff); /* Program header table file offset */
852 bswaptls(&ehdr->e_shoff); /* Section header table file offset */
853 bswap32s(&ehdr->e_flags); /* Processor-specific flags */
854 bswap16s(&ehdr->e_ehsize); /* ELF header size in bytes */
855 bswap16s(&ehdr->e_phentsize); /* Program header table entry size */
856 bswap16s(&ehdr->e_phnum); /* Program header table entry count */
857 bswap16s(&ehdr->e_shentsize); /* Section header table entry size */
858 bswap16s(&ehdr->e_shnum); /* Section header table entry count */
859 bswap16s(&ehdr->e_shstrndx); /* Section header string table index */
862 static void bswap_phdr(struct elf_phdr *phdr, int phnum)
864 int i;
865 for (i = 0; i < phnum; ++i, ++phdr) {
866 bswap32s(&phdr->p_type); /* Segment type */
867 bswap32s(&phdr->p_flags); /* Segment flags */
868 bswaptls(&phdr->p_offset); /* Segment file offset */
869 bswaptls(&phdr->p_vaddr); /* Segment virtual address */
870 bswaptls(&phdr->p_paddr); /* Segment physical address */
871 bswaptls(&phdr->p_filesz); /* Segment size in file */
872 bswaptls(&phdr->p_memsz); /* Segment size in memory */
873 bswaptls(&phdr->p_align); /* Segment alignment */
877 static void bswap_shdr(struct elf_shdr *shdr, int shnum)
879 int i;
880 for (i = 0; i < shnum; ++i, ++shdr) {
881 bswap32s(&shdr->sh_name);
882 bswap32s(&shdr->sh_type);
883 bswaptls(&shdr->sh_flags);
884 bswaptls(&shdr->sh_addr);
885 bswaptls(&shdr->sh_offset);
886 bswaptls(&shdr->sh_size);
887 bswap32s(&shdr->sh_link);
888 bswap32s(&shdr->sh_info);
889 bswaptls(&shdr->sh_addralign);
890 bswaptls(&shdr->sh_entsize);
894 static void bswap_sym(struct elf_sym *sym)
896 bswap32s(&sym->st_name);
897 bswaptls(&sym->st_value);
898 bswaptls(&sym->st_size);
899 bswap16s(&sym->st_shndx);
901 #else
902 static inline void bswap_ehdr(struct elfhdr *ehdr) { }
903 static inline void bswap_phdr(struct elf_phdr *phdr, int phnum) { }
904 static inline void bswap_shdr(struct elf_shdr *shdr, int shnum) { }
905 static inline void bswap_sym(struct elf_sym *sym) { }
906 #endif
908 #ifdef USE_ELF_CORE_DUMP
909 static int elf_core_dump(int, const CPUState *);
910 #endif /* USE_ELF_CORE_DUMP */
911 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias);
913 /* Verify the portions of EHDR within E_IDENT for the target.
914 This can be performed before bswapping the entire header. */
915 static bool elf_check_ident(struct elfhdr *ehdr)
917 return (ehdr->e_ident[EI_MAG0] == ELFMAG0
918 && ehdr->e_ident[EI_MAG1] == ELFMAG1
919 && ehdr->e_ident[EI_MAG2] == ELFMAG2
920 && ehdr->e_ident[EI_MAG3] == ELFMAG3
921 && ehdr->e_ident[EI_CLASS] == ELF_CLASS
922 && ehdr->e_ident[EI_DATA] == ELF_DATA
923 && ehdr->e_ident[EI_VERSION] == EV_CURRENT);
926 /* Verify the portions of EHDR outside of E_IDENT for the target.
927 This has to wait until after bswapping the header. */
928 static bool elf_check_ehdr(struct elfhdr *ehdr)
930 return (elf_check_arch(ehdr->e_machine)
931 && ehdr->e_ehsize == sizeof(struct elfhdr)
932 && ehdr->e_phentsize == sizeof(struct elf_phdr)
933 && ehdr->e_shentsize == sizeof(struct elf_shdr)
934 && (ehdr->e_type == ET_EXEC || ehdr->e_type == ET_DYN));
938 * 'copy_elf_strings()' copies argument/envelope strings from user
939 * memory to free pages in kernel mem. These are in a format ready
940 * to be put directly into the top of new user memory.
943 static abi_ulong copy_elf_strings(int argc,char ** argv, void **page,
944 abi_ulong p)
946 char *tmp, *tmp1, *pag = NULL;
947 int len, offset = 0;
949 if (!p) {
950 return 0; /* bullet-proofing */
952 while (argc-- > 0) {
953 tmp = argv[argc];
954 if (!tmp) {
955 fprintf(stderr, "VFS: argc is wrong");
956 exit(-1);
958 tmp1 = tmp;
959 while (*tmp++);
960 len = tmp - tmp1;
961 if (p < len) { /* this shouldn't happen - 128kB */
962 return 0;
964 while (len) {
965 --p; --tmp; --len;
966 if (--offset < 0) {
967 offset = p % TARGET_PAGE_SIZE;
968 pag = (char *)page[p/TARGET_PAGE_SIZE];
969 if (!pag) {
970 pag = (char *)malloc(TARGET_PAGE_SIZE);
971 memset(pag, 0, TARGET_PAGE_SIZE);
972 page[p/TARGET_PAGE_SIZE] = pag;
973 if (!pag)
974 return 0;
977 if (len == 0 || offset == 0) {
978 *(pag + offset) = *tmp;
980 else {
981 int bytes_to_copy = (len > offset) ? offset : len;
982 tmp -= bytes_to_copy;
983 p -= bytes_to_copy;
984 offset -= bytes_to_copy;
985 len -= bytes_to_copy;
986 memcpy_fromfs(pag + offset, tmp, bytes_to_copy + 1);
990 return p;
993 static abi_ulong setup_arg_pages(abi_ulong p, struct linux_binprm *bprm,
994 struct image_info *info)
996 abi_ulong stack_base, size, error, guard;
997 int i;
999 /* Create enough stack to hold everything. If we don't use
1000 it for args, we'll use it for something else. */
1001 size = guest_stack_size;
1002 if (size < MAX_ARG_PAGES*TARGET_PAGE_SIZE) {
1003 size = MAX_ARG_PAGES*TARGET_PAGE_SIZE;
1005 guard = TARGET_PAGE_SIZE;
1006 if (guard < qemu_real_host_page_size) {
1007 guard = qemu_real_host_page_size;
1010 error = target_mmap(0, size + guard, PROT_READ | PROT_WRITE,
1011 MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
1012 if (error == -1) {
1013 perror("mmap stack");
1014 exit(-1);
1017 /* We reserve one extra page at the top of the stack as guard. */
1018 target_mprotect(error, guard, PROT_NONE);
1020 info->stack_limit = error + guard;
1021 stack_base = info->stack_limit + size - MAX_ARG_PAGES*TARGET_PAGE_SIZE;
1022 p += stack_base;
1024 for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
1025 if (bprm->page[i]) {
1026 info->rss++;
1027 /* FIXME - check return value of memcpy_to_target() for failure */
1028 memcpy_to_target(stack_base, bprm->page[i], TARGET_PAGE_SIZE);
1029 free(bprm->page[i]);
1031 stack_base += TARGET_PAGE_SIZE;
1033 return p;
1036 /* Map and zero the bss. We need to explicitly zero any fractional pages
1037 after the data section (i.e. bss). */
1038 static void zero_bss(abi_ulong elf_bss, abi_ulong last_bss, int prot)
1040 uintptr_t host_start, host_map_start, host_end;
1042 last_bss = TARGET_PAGE_ALIGN(last_bss);
1044 /* ??? There is confusion between qemu_real_host_page_size and
1045 qemu_host_page_size here and elsewhere in target_mmap, which
1046 may lead to the end of the data section mapping from the file
1047 not being mapped. At least there was an explicit test and
1048 comment for that here, suggesting that "the file size must
1049 be known". The comment probably pre-dates the introduction
1050 of the fstat system call in target_mmap which does in fact
1051 find out the size. What isn't clear is if the workaround
1052 here is still actually needed. For now, continue with it,
1053 but merge it with the "normal" mmap that would allocate the bss. */
1055 host_start = (uintptr_t) g2h(elf_bss);
1056 host_end = (uintptr_t) g2h(last_bss);
1057 host_map_start = (host_start + qemu_real_host_page_size - 1);
1058 host_map_start &= -qemu_real_host_page_size;
1060 if (host_map_start < host_end) {
1061 void *p = mmap((void *)host_map_start, host_end - host_map_start,
1062 prot, MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
1063 if (p == MAP_FAILED) {
1064 perror("cannot mmap brk");
1065 exit(-1);
1068 /* Since we didn't use target_mmap, make sure to record
1069 the validity of the pages with qemu. */
1070 page_set_flags(elf_bss & TARGET_PAGE_MASK, last_bss, prot|PAGE_VALID);
1073 if (host_start < host_map_start) {
1074 memset((void *)host_start, 0, host_map_start - host_start);
1078 static abi_ulong create_elf_tables(abi_ulong p, int argc, int envc,
1079 struct elfhdr *exec,
1080 struct image_info *info,
1081 struct image_info *interp_info)
1083 abi_ulong sp;
1084 int size;
1085 abi_ulong u_platform;
1086 const char *k_platform;
1087 const int n = sizeof(elf_addr_t);
1089 sp = p;
1090 u_platform = 0;
1091 k_platform = ELF_PLATFORM;
1092 if (k_platform) {
1093 size_t len = strlen(k_platform) + 1;
1094 sp -= (len + n - 1) & ~(n - 1);
1095 u_platform = sp;
1096 /* FIXME - check return value of memcpy_to_target() for failure */
1097 memcpy_to_target(sp, k_platform, len);
1100 * Force 16 byte _final_ alignment here for generality.
1102 sp = sp &~ (abi_ulong)15;
1103 size = (DLINFO_ITEMS + 1) * 2;
1104 if (k_platform)
1105 size += 2;
1106 #ifdef DLINFO_ARCH_ITEMS
1107 size += DLINFO_ARCH_ITEMS * 2;
1108 #endif
1109 size += envc + argc + 2;
1110 size += 1; /* argc itself */
1111 size *= n;
1112 if (size & 15)
1113 sp -= 16 - (size & 15);
1115 /* This is correct because Linux defines
1116 * elf_addr_t as Elf32_Off / Elf64_Off
1118 #define NEW_AUX_ENT(id, val) do { \
1119 sp -= n; put_user_ual(val, sp); \
1120 sp -= n; put_user_ual(id, sp); \
1121 } while(0)
1123 NEW_AUX_ENT (AT_NULL, 0);
1125 /* There must be exactly DLINFO_ITEMS entries here. */
1126 NEW_AUX_ENT(AT_PHDR, (abi_ulong)(info->load_addr + exec->e_phoff));
1127 NEW_AUX_ENT(AT_PHENT, (abi_ulong)(sizeof (struct elf_phdr)));
1128 NEW_AUX_ENT(AT_PHNUM, (abi_ulong)(exec->e_phnum));
1129 NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(TARGET_PAGE_SIZE));
1130 NEW_AUX_ENT(AT_BASE, (abi_ulong)(interp_info ? interp_info->load_addr : 0));
1131 NEW_AUX_ENT(AT_FLAGS, (abi_ulong)0);
1132 NEW_AUX_ENT(AT_ENTRY, info->entry);
1133 NEW_AUX_ENT(AT_UID, (abi_ulong) getuid());
1134 NEW_AUX_ENT(AT_EUID, (abi_ulong) geteuid());
1135 NEW_AUX_ENT(AT_GID, (abi_ulong) getgid());
1136 NEW_AUX_ENT(AT_EGID, (abi_ulong) getegid());
1137 NEW_AUX_ENT(AT_HWCAP, (abi_ulong) ELF_HWCAP);
1138 NEW_AUX_ENT(AT_CLKTCK, (abi_ulong) sysconf(_SC_CLK_TCK));
1139 if (k_platform)
1140 NEW_AUX_ENT(AT_PLATFORM, u_platform);
1141 #ifdef ARCH_DLINFO
1143 * ARCH_DLINFO must come last so platform specific code can enforce
1144 * special alignment requirements on the AUXV if necessary (eg. PPC).
1146 ARCH_DLINFO;
1147 #endif
1148 #undef NEW_AUX_ENT
1150 info->saved_auxv = sp;
1152 sp = loader_build_argptr(envc, argc, sp, p, 0);
1153 return sp;
1156 /* Load an ELF image into the address space.
1158 IMAGE_NAME is the filename of the image, to use in error messages.
1159 IMAGE_FD is the open file descriptor for the image.
1161 BPRM_BUF is a copy of the beginning of the file; this of course
1162 contains the elf file header at offset 0. It is assumed that this
1163 buffer is sufficiently aligned to present no problems to the host
1164 in accessing data at aligned offsets within the buffer.
1166 On return: INFO values will be filled in, as necessary or available. */
1168 static void load_elf_image(const char *image_name, int image_fd,
1169 struct image_info *info, char **pinterp_name,
1170 char bprm_buf[BPRM_BUF_SIZE])
1172 struct elfhdr *ehdr = (struct elfhdr *)bprm_buf;
1173 struct elf_phdr *phdr;
1174 abi_ulong load_addr, load_bias, loaddr, hiaddr, error;
1175 int i, retval;
1176 const char *errmsg;
1178 /* First of all, some simple consistency checks */
1179 errmsg = "Invalid ELF image for this architecture";
1180 if (!elf_check_ident(ehdr)) {
1181 goto exit_errmsg;
1183 bswap_ehdr(ehdr);
1184 if (!elf_check_ehdr(ehdr)) {
1185 goto exit_errmsg;
1188 i = ehdr->e_phnum * sizeof(struct elf_phdr);
1189 if (ehdr->e_phoff + i <= BPRM_BUF_SIZE) {
1190 phdr = (struct elf_phdr *)(bprm_buf + ehdr->e_phoff);
1191 } else {
1192 phdr = (struct elf_phdr *) alloca(i);
1193 retval = pread(image_fd, phdr, i, ehdr->e_phoff);
1194 if (retval != i) {
1195 goto exit_read;
1198 bswap_phdr(phdr, ehdr->e_phnum);
1200 /* Find the maximum size of the image and allocate an appropriate
1201 amount of memory to handle that. */
1202 loaddr = -1, hiaddr = 0;
1203 for (i = 0; i < ehdr->e_phnum; ++i) {
1204 if (phdr[i].p_type == PT_LOAD) {
1205 abi_ulong a = phdr[i].p_vaddr;
1206 if (a < loaddr) {
1207 loaddr = a;
1209 a += phdr[i].p_memsz;
1210 if (a > hiaddr) {
1211 hiaddr = a;
1216 load_addr = loaddr;
1217 if (ehdr->e_type == ET_DYN) {
1218 /* The image indicates that it can be loaded anywhere. Find a
1219 location that can hold the memory space required. If the
1220 image is pre-linked, LOADDR will be non-zero. Since we do
1221 not supply MAP_FIXED here we'll use that address if and
1222 only if it remains available. */
1223 load_addr = target_mmap(loaddr, hiaddr - loaddr, PROT_NONE,
1224 MAP_PRIVATE | MAP_ANON | MAP_NORESERVE,
1225 -1, 0);
1226 if (load_addr == -1) {
1227 goto exit_perror;
1229 } else if (pinterp_name != NULL) {
1230 /* This is the main executable. Make sure that the low
1231 address does not conflict with MMAP_MIN_ADDR or the
1232 QEMU application itself. */
1233 #if defined(CONFIG_USE_GUEST_BASE)
1235 * In case where user has not explicitly set the guest_base, we
1236 * probe here that should we set it automatically.
1238 if (!have_guest_base && !reserved_va) {
1239 unsigned long host_start, real_start, host_size;
1241 /* Round addresses to page boundaries. */
1242 loaddr &= qemu_host_page_mask;
1243 hiaddr = HOST_PAGE_ALIGN(hiaddr);
1245 if (loaddr < mmap_min_addr) {
1246 host_start = HOST_PAGE_ALIGN(mmap_min_addr);
1247 } else {
1248 host_start = loaddr;
1249 if (host_start != loaddr) {
1250 errmsg = "Address overflow loading ELF binary";
1251 goto exit_errmsg;
1254 host_size = hiaddr - loaddr;
1255 while (1) {
1256 /* Do not use mmap_find_vma here because that is limited to the
1257 guest address space. We are going to make the
1258 guest address space fit whatever we're given. */
1259 real_start = (unsigned long)
1260 mmap((void *)host_start, host_size, PROT_NONE,
1261 MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE, -1, 0);
1262 if (real_start == (unsigned long)-1) {
1263 goto exit_perror;
1265 if (real_start == host_start) {
1266 break;
1268 /* That address didn't work. Unmap and try a different one.
1269 The address the host picked because is typically right at
1270 the top of the host address space and leaves the guest with
1271 no usable address space. Resort to a linear search. We
1272 already compensated for mmap_min_addr, so this should not
1273 happen often. Probably means we got unlucky and host
1274 address space randomization put a shared library somewhere
1275 inconvenient. */
1276 munmap((void *)real_start, host_size);
1277 host_start += qemu_host_page_size;
1278 if (host_start == loaddr) {
1279 /* Theoretically possible if host doesn't have any suitably
1280 aligned areas. Normally the first mmap will fail. */
1281 errmsg = "Unable to find space for application";
1282 goto exit_errmsg;
1285 qemu_log("Relocating guest address space from 0x"
1286 TARGET_ABI_FMT_lx " to 0x%lx\n", loaddr, real_start);
1287 guest_base = real_start - loaddr;
1289 #endif
1291 load_bias = load_addr - loaddr;
1293 info->load_bias = load_bias;
1294 info->load_addr = load_addr;
1295 info->entry = ehdr->e_entry + load_bias;
1296 info->start_code = -1;
1297 info->end_code = 0;
1298 info->start_data = -1;
1299 info->end_data = 0;
1300 info->brk = 0;
1302 for (i = 0; i < ehdr->e_phnum; i++) {
1303 struct elf_phdr *eppnt = phdr + i;
1304 if (eppnt->p_type == PT_LOAD) {
1305 abi_ulong vaddr, vaddr_po, vaddr_ps, vaddr_ef, vaddr_em;
1306 int elf_prot = 0;
1308 if (eppnt->p_flags & PF_R) elf_prot = PROT_READ;
1309 if (eppnt->p_flags & PF_W) elf_prot |= PROT_WRITE;
1310 if (eppnt->p_flags & PF_X) elf_prot |= PROT_EXEC;
1312 vaddr = load_bias + eppnt->p_vaddr;
1313 vaddr_po = TARGET_ELF_PAGEOFFSET(vaddr);
1314 vaddr_ps = TARGET_ELF_PAGESTART(vaddr);
1316 error = target_mmap(vaddr_ps, eppnt->p_filesz + vaddr_po,
1317 elf_prot, MAP_PRIVATE | MAP_FIXED,
1318 image_fd, eppnt->p_offset - vaddr_po);
1319 if (error == -1) {
1320 goto exit_perror;
1323 vaddr_ef = vaddr + eppnt->p_filesz;
1324 vaddr_em = vaddr + eppnt->p_memsz;
1326 /* If the load segment requests extra zeros (e.g. bss), map it. */
1327 if (vaddr_ef < vaddr_em) {
1328 zero_bss(vaddr_ef, vaddr_em, elf_prot);
1331 /* Find the full program boundaries. */
1332 if (elf_prot & PROT_EXEC) {
1333 if (vaddr < info->start_code) {
1334 info->start_code = vaddr;
1336 if (vaddr_ef > info->end_code) {
1337 info->end_code = vaddr_ef;
1340 if (elf_prot & PROT_WRITE) {
1341 if (vaddr < info->start_data) {
1342 info->start_data = vaddr;
1344 if (vaddr_ef > info->end_data) {
1345 info->end_data = vaddr_ef;
1347 if (vaddr_em > info->brk) {
1348 info->brk = vaddr_em;
1351 } else if (eppnt->p_type == PT_INTERP && pinterp_name) {
1352 char *interp_name;
1354 if (*pinterp_name) {
1355 errmsg = "Multiple PT_INTERP entries";
1356 goto exit_errmsg;
1358 interp_name = malloc(eppnt->p_filesz);
1359 if (!interp_name) {
1360 goto exit_perror;
1363 if (eppnt->p_offset + eppnt->p_filesz <= BPRM_BUF_SIZE) {
1364 memcpy(interp_name, bprm_buf + eppnt->p_offset,
1365 eppnt->p_filesz);
1366 } else {
1367 retval = pread(image_fd, interp_name, eppnt->p_filesz,
1368 eppnt->p_offset);
1369 if (retval != eppnt->p_filesz) {
1370 goto exit_perror;
1373 if (interp_name[eppnt->p_filesz - 1] != 0) {
1374 errmsg = "Invalid PT_INTERP entry";
1375 goto exit_errmsg;
1377 *pinterp_name = interp_name;
1381 if (info->end_data == 0) {
1382 info->start_data = info->end_code;
1383 info->end_data = info->end_code;
1384 info->brk = info->end_code;
1387 if (qemu_log_enabled()) {
1388 load_symbols(ehdr, image_fd, load_bias);
1391 close(image_fd);
1392 return;
1394 exit_read:
1395 if (retval >= 0) {
1396 errmsg = "Incomplete read of file header";
1397 goto exit_errmsg;
1399 exit_perror:
1400 errmsg = strerror(errno);
1401 exit_errmsg:
1402 fprintf(stderr, "%s: %s\n", image_name, errmsg);
1403 exit(-1);
1406 static void load_elf_interp(const char *filename, struct image_info *info,
1407 char bprm_buf[BPRM_BUF_SIZE])
1409 int fd, retval;
1411 fd = open(path(filename), O_RDONLY);
1412 if (fd < 0) {
1413 goto exit_perror;
1416 retval = read(fd, bprm_buf, BPRM_BUF_SIZE);
1417 if (retval < 0) {
1418 goto exit_perror;
1420 if (retval < BPRM_BUF_SIZE) {
1421 memset(bprm_buf + retval, 0, BPRM_BUF_SIZE - retval);
1424 load_elf_image(filename, fd, info, NULL, bprm_buf);
1425 return;
1427 exit_perror:
1428 fprintf(stderr, "%s: %s\n", filename, strerror(errno));
1429 exit(-1);
1432 static int symfind(const void *s0, const void *s1)
1434 struct elf_sym *key = (struct elf_sym *)s0;
1435 struct elf_sym *sym = (struct elf_sym *)s1;
1436 int result = 0;
1437 if (key->st_value < sym->st_value) {
1438 result = -1;
1439 } else if (key->st_value >= sym->st_value + sym->st_size) {
1440 result = 1;
1442 return result;
1445 static const char *lookup_symbolxx(struct syminfo *s, target_ulong orig_addr)
1447 #if ELF_CLASS == ELFCLASS32
1448 struct elf_sym *syms = s->disas_symtab.elf32;
1449 #else
1450 struct elf_sym *syms = s->disas_symtab.elf64;
1451 #endif
1453 // binary search
1454 struct elf_sym key;
1455 struct elf_sym *sym;
1457 key.st_value = orig_addr;
1459 sym = bsearch(&key, syms, s->disas_num_syms, sizeof(*syms), symfind);
1460 if (sym != NULL) {
1461 return s->disas_strtab + sym->st_name;
1464 return "";
1467 /* FIXME: This should use elf_ops.h */
1468 static int symcmp(const void *s0, const void *s1)
1470 struct elf_sym *sym0 = (struct elf_sym *)s0;
1471 struct elf_sym *sym1 = (struct elf_sym *)s1;
1472 return (sym0->st_value < sym1->st_value)
1473 ? -1
1474 : ((sym0->st_value > sym1->st_value) ? 1 : 0);
1477 /* Best attempt to load symbols from this ELF object. */
1478 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias)
1480 int i, shnum, nsyms, sym_idx = 0, str_idx = 0;
1481 struct elf_shdr *shdr;
1482 char *strings;
1483 struct syminfo *s;
1484 struct elf_sym *syms;
1486 shnum = hdr->e_shnum;
1487 i = shnum * sizeof(struct elf_shdr);
1488 shdr = (struct elf_shdr *)alloca(i);
1489 if (pread(fd, shdr, i, hdr->e_shoff) != i) {
1490 return;
1493 bswap_shdr(shdr, shnum);
1494 for (i = 0; i < shnum; ++i) {
1495 if (shdr[i].sh_type == SHT_SYMTAB) {
1496 sym_idx = i;
1497 str_idx = shdr[i].sh_link;
1498 goto found;
1502 /* There will be no symbol table if the file was stripped. */
1503 return;
1505 found:
1506 /* Now know where the strtab and symtab are. Snarf them. */
1507 s = malloc(sizeof(*s));
1508 if (!s) {
1509 return;
1512 i = shdr[str_idx].sh_size;
1513 s->disas_strtab = strings = malloc(i);
1514 if (!strings || pread(fd, strings, i, shdr[str_idx].sh_offset) != i) {
1515 free(s);
1516 free(strings);
1517 return;
1520 i = shdr[sym_idx].sh_size;
1521 syms = malloc(i);
1522 if (!syms || pread(fd, syms, i, shdr[sym_idx].sh_offset) != i) {
1523 free(s);
1524 free(strings);
1525 free(syms);
1526 return;
1529 nsyms = i / sizeof(struct elf_sym);
1530 for (i = 0; i < nsyms; ) {
1531 bswap_sym(syms + i);
1532 /* Throw away entries which we do not need. */
1533 if (syms[i].st_shndx == SHN_UNDEF
1534 || syms[i].st_shndx >= SHN_LORESERVE
1535 || ELF_ST_TYPE(syms[i].st_info) != STT_FUNC) {
1536 if (i < --nsyms) {
1537 syms[i] = syms[nsyms];
1539 } else {
1540 #if defined(TARGET_ARM) || defined (TARGET_MIPS)
1541 /* The bottom address bit marks a Thumb or MIPS16 symbol. */
1542 syms[i].st_value &= ~(target_ulong)1;
1543 #endif
1544 syms[i].st_value += load_bias;
1545 i++;
1549 /* Attempt to free the storage associated with the local symbols
1550 that we threw away. Whether or not this has any effect on the
1551 memory allocation depends on the malloc implementation and how
1552 many symbols we managed to discard. */
1553 syms = realloc(syms, nsyms * sizeof(*syms));
1554 if (syms == NULL) {
1555 free(s);
1556 free(strings);
1557 return;
1560 qsort(syms, nsyms, sizeof(*syms), symcmp);
1562 s->disas_num_syms = nsyms;
1563 #if ELF_CLASS == ELFCLASS32
1564 s->disas_symtab.elf32 = syms;
1565 #else
1566 s->disas_symtab.elf64 = syms;
1567 #endif
1568 s->lookup_symbol = lookup_symbolxx;
1569 s->next = syminfos;
1570 syminfos = s;
1573 int load_elf_binary(struct linux_binprm * bprm, struct target_pt_regs * regs,
1574 struct image_info * info)
1576 struct image_info interp_info;
1577 struct elfhdr elf_ex;
1578 char *elf_interpreter = NULL;
1580 info->start_mmap = (abi_ulong)ELF_START_MMAP;
1581 info->mmap = 0;
1582 info->rss = 0;
1584 load_elf_image(bprm->filename, bprm->fd, info,
1585 &elf_interpreter, bprm->buf);
1587 /* ??? We need a copy of the elf header for passing to create_elf_tables.
1588 If we do nothing, we'll have overwritten this when we re-use bprm->buf
1589 when we load the interpreter. */
1590 elf_ex = *(struct elfhdr *)bprm->buf;
1592 bprm->p = copy_elf_strings(1, &bprm->filename, bprm->page, bprm->p);
1593 bprm->p = copy_elf_strings(bprm->envc,bprm->envp,bprm->page,bprm->p);
1594 bprm->p = copy_elf_strings(bprm->argc,bprm->argv,bprm->page,bprm->p);
1595 if (!bprm->p) {
1596 fprintf(stderr, "%s: %s\n", bprm->filename, strerror(E2BIG));
1597 exit(-1);
1600 /* Do this so that we can load the interpreter, if need be. We will
1601 change some of these later */
1602 bprm->p = setup_arg_pages(bprm->p, bprm, info);
1604 if (elf_interpreter) {
1605 load_elf_interp(elf_interpreter, &interp_info, bprm->buf);
1607 /* If the program interpreter is one of these two, then assume
1608 an iBCS2 image. Otherwise assume a native linux image. */
1610 if (strcmp(elf_interpreter, "/usr/lib/libc.so.1") == 0
1611 || strcmp(elf_interpreter, "/usr/lib/ld.so.1") == 0) {
1612 info->personality = PER_SVR4;
1614 /* Why this, you ask??? Well SVr4 maps page 0 as read-only,
1615 and some applications "depend" upon this behavior. Since
1616 we do not have the power to recompile these, we emulate
1617 the SVr4 behavior. Sigh. */
1618 target_mmap(0, qemu_host_page_size, PROT_READ | PROT_EXEC,
1619 MAP_FIXED | MAP_PRIVATE, -1, 0);
1623 bprm->p = create_elf_tables(bprm->p, bprm->argc, bprm->envc, &elf_ex,
1624 info, (elf_interpreter ? &interp_info : NULL));
1625 info->start_stack = bprm->p;
1627 /* If we have an interpreter, set that as the program's entry point.
1628 Copy the load_addr as well, to help PPC64 interpret the entry
1629 point as a function descriptor. Do this after creating elf tables
1630 so that we copy the original program entry point into the AUXV. */
1631 if (elf_interpreter) {
1632 info->load_addr = interp_info.load_addr;
1633 info->entry = interp_info.entry;
1634 free(elf_interpreter);
1637 #ifdef USE_ELF_CORE_DUMP
1638 bprm->core_dump = &elf_core_dump;
1639 #endif
1641 return 0;
1644 #ifdef USE_ELF_CORE_DUMP
1646 * Definitions to generate Intel SVR4-like core files.
1647 * These mostly have the same names as the SVR4 types with "target_elf_"
1648 * tacked on the front to prevent clashes with linux definitions,
1649 * and the typedef forms have been avoided. This is mostly like
1650 * the SVR4 structure, but more Linuxy, with things that Linux does
1651 * not support and which gdb doesn't really use excluded.
1653 * Fields we don't dump (their contents is zero) in linux-user qemu
1654 * are marked with XXX.
1656 * Core dump code is copied from linux kernel (fs/binfmt_elf.c).
1658 * Porting ELF coredump for target is (quite) simple process. First you
1659 * define USE_ELF_CORE_DUMP in target ELF code (where init_thread() for
1660 * the target resides):
1662 * #define USE_ELF_CORE_DUMP
1664 * Next you define type of register set used for dumping. ELF specification
1665 * says that it needs to be array of elf_greg_t that has size of ELF_NREG.
1667 * typedef <target_regtype> target_elf_greg_t;
1668 * #define ELF_NREG <number of registers>
1669 * typedef taret_elf_greg_t target_elf_gregset_t[ELF_NREG];
1671 * Last step is to implement target specific function that copies registers
1672 * from given cpu into just specified register set. Prototype is:
1674 * static void elf_core_copy_regs(taret_elf_gregset_t *regs,
1675 * const CPUState *env);
1677 * Parameters:
1678 * regs - copy register values into here (allocated and zeroed by caller)
1679 * env - copy registers from here
1681 * Example for ARM target is provided in this file.
1684 /* An ELF note in memory */
1685 struct memelfnote {
1686 const char *name;
1687 size_t namesz;
1688 size_t namesz_rounded;
1689 int type;
1690 size_t datasz;
1691 void *data;
1692 size_t notesz;
1695 struct target_elf_siginfo {
1696 int si_signo; /* signal number */
1697 int si_code; /* extra code */
1698 int si_errno; /* errno */
1701 struct target_elf_prstatus {
1702 struct target_elf_siginfo pr_info; /* Info associated with signal */
1703 short pr_cursig; /* Current signal */
1704 target_ulong pr_sigpend; /* XXX */
1705 target_ulong pr_sighold; /* XXX */
1706 target_pid_t pr_pid;
1707 target_pid_t pr_ppid;
1708 target_pid_t pr_pgrp;
1709 target_pid_t pr_sid;
1710 struct target_timeval pr_utime; /* XXX User time */
1711 struct target_timeval pr_stime; /* XXX System time */
1712 struct target_timeval pr_cutime; /* XXX Cumulative user time */
1713 struct target_timeval pr_cstime; /* XXX Cumulative system time */
1714 target_elf_gregset_t pr_reg; /* GP registers */
1715 int pr_fpvalid; /* XXX */
1718 #define ELF_PRARGSZ (80) /* Number of chars for args */
1720 struct target_elf_prpsinfo {
1721 char pr_state; /* numeric process state */
1722 char pr_sname; /* char for pr_state */
1723 char pr_zomb; /* zombie */
1724 char pr_nice; /* nice val */
1725 target_ulong pr_flag; /* flags */
1726 target_uid_t pr_uid;
1727 target_gid_t pr_gid;
1728 target_pid_t pr_pid, pr_ppid, pr_pgrp, pr_sid;
1729 /* Lots missing */
1730 char pr_fname[16]; /* filename of executable */
1731 char pr_psargs[ELF_PRARGSZ]; /* initial part of arg list */
1734 /* Here is the structure in which status of each thread is captured. */
1735 struct elf_thread_status {
1736 QTAILQ_ENTRY(elf_thread_status) ets_link;
1737 struct target_elf_prstatus prstatus; /* NT_PRSTATUS */
1738 #if 0
1739 elf_fpregset_t fpu; /* NT_PRFPREG */
1740 struct task_struct *thread;
1741 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */
1742 #endif
1743 struct memelfnote notes[1];
1744 int num_notes;
1747 struct elf_note_info {
1748 struct memelfnote *notes;
1749 struct target_elf_prstatus *prstatus; /* NT_PRSTATUS */
1750 struct target_elf_prpsinfo *psinfo; /* NT_PRPSINFO */
1752 QTAILQ_HEAD(thread_list_head, elf_thread_status) thread_list;
1753 #if 0
1755 * Current version of ELF coredump doesn't support
1756 * dumping fp regs etc.
1758 elf_fpregset_t *fpu;
1759 elf_fpxregset_t *xfpu;
1760 int thread_status_size;
1761 #endif
1762 int notes_size;
1763 int numnote;
1766 struct vm_area_struct {
1767 abi_ulong vma_start; /* start vaddr of memory region */
1768 abi_ulong vma_end; /* end vaddr of memory region */
1769 abi_ulong vma_flags; /* protection etc. flags for the region */
1770 QTAILQ_ENTRY(vm_area_struct) vma_link;
1773 struct mm_struct {
1774 QTAILQ_HEAD(, vm_area_struct) mm_mmap;
1775 int mm_count; /* number of mappings */
1778 static struct mm_struct *vma_init(void);
1779 static void vma_delete(struct mm_struct *);
1780 static int vma_add_mapping(struct mm_struct *, abi_ulong,
1781 abi_ulong, abi_ulong);
1782 static int vma_get_mapping_count(const struct mm_struct *);
1783 static struct vm_area_struct *vma_first(const struct mm_struct *);
1784 static struct vm_area_struct *vma_next(struct vm_area_struct *);
1785 static abi_ulong vma_dump_size(const struct vm_area_struct *);
1786 static int vma_walker(void *priv, abi_ulong start, abi_ulong end,
1787 unsigned long flags);
1789 static void fill_elf_header(struct elfhdr *, int, uint16_t, uint32_t);
1790 static void fill_note(struct memelfnote *, const char *, int,
1791 unsigned int, void *);
1792 static void fill_prstatus(struct target_elf_prstatus *, const TaskState *, int);
1793 static int fill_psinfo(struct target_elf_prpsinfo *, const TaskState *);
1794 static void fill_auxv_note(struct memelfnote *, const TaskState *);
1795 static void fill_elf_note_phdr(struct elf_phdr *, int, off_t);
1796 static size_t note_size(const struct memelfnote *);
1797 static void free_note_info(struct elf_note_info *);
1798 static int fill_note_info(struct elf_note_info *, long, const CPUState *);
1799 static void fill_thread_info(struct elf_note_info *, const CPUState *);
1800 static int core_dump_filename(const TaskState *, char *, size_t);
1802 static int dump_write(int, const void *, size_t);
1803 static int write_note(struct memelfnote *, int);
1804 static int write_note_info(struct elf_note_info *, int);
1806 #ifdef BSWAP_NEEDED
1807 static void bswap_prstatus(struct target_elf_prstatus *prstatus)
1809 prstatus->pr_info.si_signo = tswapl(prstatus->pr_info.si_signo);
1810 prstatus->pr_info.si_code = tswapl(prstatus->pr_info.si_code);
1811 prstatus->pr_info.si_errno = tswapl(prstatus->pr_info.si_errno);
1812 prstatus->pr_cursig = tswap16(prstatus->pr_cursig);
1813 prstatus->pr_sigpend = tswapl(prstatus->pr_sigpend);
1814 prstatus->pr_sighold = tswapl(prstatus->pr_sighold);
1815 prstatus->pr_pid = tswap32(prstatus->pr_pid);
1816 prstatus->pr_ppid = tswap32(prstatus->pr_ppid);
1817 prstatus->pr_pgrp = tswap32(prstatus->pr_pgrp);
1818 prstatus->pr_sid = tswap32(prstatus->pr_sid);
1819 /* cpu times are not filled, so we skip them */
1820 /* regs should be in correct format already */
1821 prstatus->pr_fpvalid = tswap32(prstatus->pr_fpvalid);
1824 static void bswap_psinfo(struct target_elf_prpsinfo *psinfo)
1826 psinfo->pr_flag = tswapl(psinfo->pr_flag);
1827 psinfo->pr_uid = tswap16(psinfo->pr_uid);
1828 psinfo->pr_gid = tswap16(psinfo->pr_gid);
1829 psinfo->pr_pid = tswap32(psinfo->pr_pid);
1830 psinfo->pr_ppid = tswap32(psinfo->pr_ppid);
1831 psinfo->pr_pgrp = tswap32(psinfo->pr_pgrp);
1832 psinfo->pr_sid = tswap32(psinfo->pr_sid);
1835 static void bswap_note(struct elf_note *en)
1837 bswap32s(&en->n_namesz);
1838 bswap32s(&en->n_descsz);
1839 bswap32s(&en->n_type);
1841 #else
1842 static inline void bswap_prstatus(struct target_elf_prstatus *p) { }
1843 static inline void bswap_psinfo(struct target_elf_prpsinfo *p) {}
1844 static inline void bswap_note(struct elf_note *en) { }
1845 #endif /* BSWAP_NEEDED */
1848 * Minimal support for linux memory regions. These are needed
1849 * when we are finding out what memory exactly belongs to
1850 * emulated process. No locks needed here, as long as
1851 * thread that received the signal is stopped.
1854 static struct mm_struct *vma_init(void)
1856 struct mm_struct *mm;
1858 if ((mm = qemu_malloc(sizeof (*mm))) == NULL)
1859 return (NULL);
1861 mm->mm_count = 0;
1862 QTAILQ_INIT(&mm->mm_mmap);
1864 return (mm);
1867 static void vma_delete(struct mm_struct *mm)
1869 struct vm_area_struct *vma;
1871 while ((vma = vma_first(mm)) != NULL) {
1872 QTAILQ_REMOVE(&mm->mm_mmap, vma, vma_link);
1873 qemu_free(vma);
1875 qemu_free(mm);
1878 static int vma_add_mapping(struct mm_struct *mm, abi_ulong start,
1879 abi_ulong end, abi_ulong flags)
1881 struct vm_area_struct *vma;
1883 if ((vma = qemu_mallocz(sizeof (*vma))) == NULL)
1884 return (-1);
1886 vma->vma_start = start;
1887 vma->vma_end = end;
1888 vma->vma_flags = flags;
1890 QTAILQ_INSERT_TAIL(&mm->mm_mmap, vma, vma_link);
1891 mm->mm_count++;
1893 return (0);
1896 static struct vm_area_struct *vma_first(const struct mm_struct *mm)
1898 return (QTAILQ_FIRST(&mm->mm_mmap));
1901 static struct vm_area_struct *vma_next(struct vm_area_struct *vma)
1903 return (QTAILQ_NEXT(vma, vma_link));
1906 static int vma_get_mapping_count(const struct mm_struct *mm)
1908 return (mm->mm_count);
1912 * Calculate file (dump) size of given memory region.
1914 static abi_ulong vma_dump_size(const struct vm_area_struct *vma)
1916 /* if we cannot even read the first page, skip it */
1917 if (!access_ok(VERIFY_READ, vma->vma_start, TARGET_PAGE_SIZE))
1918 return (0);
1921 * Usually we don't dump executable pages as they contain
1922 * non-writable code that debugger can read directly from
1923 * target library etc. However, thread stacks are marked
1924 * also executable so we read in first page of given region
1925 * and check whether it contains elf header. If there is
1926 * no elf header, we dump it.
1928 if (vma->vma_flags & PROT_EXEC) {
1929 char page[TARGET_PAGE_SIZE];
1931 copy_from_user(page, vma->vma_start, sizeof (page));
1932 if ((page[EI_MAG0] == ELFMAG0) &&
1933 (page[EI_MAG1] == ELFMAG1) &&
1934 (page[EI_MAG2] == ELFMAG2) &&
1935 (page[EI_MAG3] == ELFMAG3)) {
1937 * Mappings are possibly from ELF binary. Don't dump
1938 * them.
1940 return (0);
1944 return (vma->vma_end - vma->vma_start);
1947 static int vma_walker(void *priv, abi_ulong start, abi_ulong end,
1948 unsigned long flags)
1950 struct mm_struct *mm = (struct mm_struct *)priv;
1952 vma_add_mapping(mm, start, end, flags);
1953 return (0);
1956 static void fill_note(struct memelfnote *note, const char *name, int type,
1957 unsigned int sz, void *data)
1959 unsigned int namesz;
1961 namesz = strlen(name) + 1;
1962 note->name = name;
1963 note->namesz = namesz;
1964 note->namesz_rounded = roundup(namesz, sizeof (int32_t));
1965 note->type = type;
1966 note->datasz = roundup(sz, sizeof (int32_t));;
1967 note->data = data;
1970 * We calculate rounded up note size here as specified by
1971 * ELF document.
1973 note->notesz = sizeof (struct elf_note) +
1974 note->namesz_rounded + note->datasz;
1977 static void fill_elf_header(struct elfhdr *elf, int segs, uint16_t machine,
1978 uint32_t flags)
1980 (void) memset(elf, 0, sizeof(*elf));
1982 (void) memcpy(elf->e_ident, ELFMAG, SELFMAG);
1983 elf->e_ident[EI_CLASS] = ELF_CLASS;
1984 elf->e_ident[EI_DATA] = ELF_DATA;
1985 elf->e_ident[EI_VERSION] = EV_CURRENT;
1986 elf->e_ident[EI_OSABI] = ELF_OSABI;
1988 elf->e_type = ET_CORE;
1989 elf->e_machine = machine;
1990 elf->e_version = EV_CURRENT;
1991 elf->e_phoff = sizeof(struct elfhdr);
1992 elf->e_flags = flags;
1993 elf->e_ehsize = sizeof(struct elfhdr);
1994 elf->e_phentsize = sizeof(struct elf_phdr);
1995 elf->e_phnum = segs;
1997 bswap_ehdr(elf);
2000 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, off_t offset)
2002 phdr->p_type = PT_NOTE;
2003 phdr->p_offset = offset;
2004 phdr->p_vaddr = 0;
2005 phdr->p_paddr = 0;
2006 phdr->p_filesz = sz;
2007 phdr->p_memsz = 0;
2008 phdr->p_flags = 0;
2009 phdr->p_align = 0;
2011 bswap_phdr(phdr, 1);
2014 static size_t note_size(const struct memelfnote *note)
2016 return (note->notesz);
2019 static void fill_prstatus(struct target_elf_prstatus *prstatus,
2020 const TaskState *ts, int signr)
2022 (void) memset(prstatus, 0, sizeof (*prstatus));
2023 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
2024 prstatus->pr_pid = ts->ts_tid;
2025 prstatus->pr_ppid = getppid();
2026 prstatus->pr_pgrp = getpgrp();
2027 prstatus->pr_sid = getsid(0);
2029 bswap_prstatus(prstatus);
2032 static int fill_psinfo(struct target_elf_prpsinfo *psinfo, const TaskState *ts)
2034 char *filename, *base_filename;
2035 unsigned int i, len;
2037 (void) memset(psinfo, 0, sizeof (*psinfo));
2039 len = ts->info->arg_end - ts->info->arg_start;
2040 if (len >= ELF_PRARGSZ)
2041 len = ELF_PRARGSZ - 1;
2042 if (copy_from_user(&psinfo->pr_psargs, ts->info->arg_start, len))
2043 return -EFAULT;
2044 for (i = 0; i < len; i++)
2045 if (psinfo->pr_psargs[i] == 0)
2046 psinfo->pr_psargs[i] = ' ';
2047 psinfo->pr_psargs[len] = 0;
2049 psinfo->pr_pid = getpid();
2050 psinfo->pr_ppid = getppid();
2051 psinfo->pr_pgrp = getpgrp();
2052 psinfo->pr_sid = getsid(0);
2053 psinfo->pr_uid = getuid();
2054 psinfo->pr_gid = getgid();
2056 filename = strdup(ts->bprm->filename);
2057 base_filename = strdup(basename(filename));
2058 (void) strncpy(psinfo->pr_fname, base_filename,
2059 sizeof(psinfo->pr_fname));
2060 free(base_filename);
2061 free(filename);
2063 bswap_psinfo(psinfo);
2064 return (0);
2067 static void fill_auxv_note(struct memelfnote *note, const TaskState *ts)
2069 elf_addr_t auxv = (elf_addr_t)ts->info->saved_auxv;
2070 elf_addr_t orig_auxv = auxv;
2071 abi_ulong val;
2072 void *ptr;
2073 int i, len;
2076 * Auxiliary vector is stored in target process stack. It contains
2077 * {type, value} pairs that we need to dump into note. This is not
2078 * strictly necessary but we do it here for sake of completeness.
2081 /* find out lenght of the vector, AT_NULL is terminator */
2082 i = len = 0;
2083 do {
2084 get_user_ual(val, auxv);
2085 i += 2;
2086 auxv += 2 * sizeof (elf_addr_t);
2087 } while (val != AT_NULL);
2088 len = i * sizeof (elf_addr_t);
2090 /* read in whole auxv vector and copy it to memelfnote */
2091 ptr = lock_user(VERIFY_READ, orig_auxv, len, 0);
2092 if (ptr != NULL) {
2093 fill_note(note, "CORE", NT_AUXV, len, ptr);
2094 unlock_user(ptr, auxv, len);
2099 * Constructs name of coredump file. We have following convention
2100 * for the name:
2101 * qemu_<basename-of-target-binary>_<date>-<time>_<pid>.core
2103 * Returns 0 in case of success, -1 otherwise (errno is set).
2105 static int core_dump_filename(const TaskState *ts, char *buf,
2106 size_t bufsize)
2108 char timestamp[64];
2109 char *filename = NULL;
2110 char *base_filename = NULL;
2111 struct timeval tv;
2112 struct tm tm;
2114 assert(bufsize >= PATH_MAX);
2116 if (gettimeofday(&tv, NULL) < 0) {
2117 (void) fprintf(stderr, "unable to get current timestamp: %s",
2118 strerror(errno));
2119 return (-1);
2122 filename = strdup(ts->bprm->filename);
2123 base_filename = strdup(basename(filename));
2124 (void) strftime(timestamp, sizeof (timestamp), "%Y%m%d-%H%M%S",
2125 localtime_r(&tv.tv_sec, &tm));
2126 (void) snprintf(buf, bufsize, "qemu_%s_%s_%d.core",
2127 base_filename, timestamp, (int)getpid());
2128 free(base_filename);
2129 free(filename);
2131 return (0);
2134 static int dump_write(int fd, const void *ptr, size_t size)
2136 const char *bufp = (const char *)ptr;
2137 ssize_t bytes_written, bytes_left;
2138 struct rlimit dumpsize;
2139 off_t pos;
2141 bytes_written = 0;
2142 getrlimit(RLIMIT_CORE, &dumpsize);
2143 if ((pos = lseek(fd, 0, SEEK_CUR))==-1) {
2144 if (errno == ESPIPE) { /* not a seekable stream */
2145 bytes_left = size;
2146 } else {
2147 return pos;
2149 } else {
2150 if (dumpsize.rlim_cur <= pos) {
2151 return -1;
2152 } else if (dumpsize.rlim_cur == RLIM_INFINITY) {
2153 bytes_left = size;
2154 } else {
2155 size_t limit_left=dumpsize.rlim_cur - pos;
2156 bytes_left = limit_left >= size ? size : limit_left ;
2161 * In normal conditions, single write(2) should do but
2162 * in case of socket etc. this mechanism is more portable.
2164 do {
2165 bytes_written = write(fd, bufp, bytes_left);
2166 if (bytes_written < 0) {
2167 if (errno == EINTR)
2168 continue;
2169 return (-1);
2170 } else if (bytes_written == 0) { /* eof */
2171 return (-1);
2173 bufp += bytes_written;
2174 bytes_left -= bytes_written;
2175 } while (bytes_left > 0);
2177 return (0);
2180 static int write_note(struct memelfnote *men, int fd)
2182 struct elf_note en;
2184 en.n_namesz = men->namesz;
2185 en.n_type = men->type;
2186 en.n_descsz = men->datasz;
2188 bswap_note(&en);
2190 if (dump_write(fd, &en, sizeof(en)) != 0)
2191 return (-1);
2192 if (dump_write(fd, men->name, men->namesz_rounded) != 0)
2193 return (-1);
2194 if (dump_write(fd, men->data, men->datasz) != 0)
2195 return (-1);
2197 return (0);
2200 static void fill_thread_info(struct elf_note_info *info, const CPUState *env)
2202 TaskState *ts = (TaskState *)env->opaque;
2203 struct elf_thread_status *ets;
2205 ets = qemu_mallocz(sizeof (*ets));
2206 ets->num_notes = 1; /* only prstatus is dumped */
2207 fill_prstatus(&ets->prstatus, ts, 0);
2208 elf_core_copy_regs(&ets->prstatus.pr_reg, env);
2209 fill_note(&ets->notes[0], "CORE", NT_PRSTATUS, sizeof (ets->prstatus),
2210 &ets->prstatus);
2212 QTAILQ_INSERT_TAIL(&info->thread_list, ets, ets_link);
2214 info->notes_size += note_size(&ets->notes[0]);
2217 static int fill_note_info(struct elf_note_info *info,
2218 long signr, const CPUState *env)
2220 #define NUMNOTES 3
2221 CPUState *cpu = NULL;
2222 TaskState *ts = (TaskState *)env->opaque;
2223 int i;
2225 (void) memset(info, 0, sizeof (*info));
2227 QTAILQ_INIT(&info->thread_list);
2229 info->notes = qemu_mallocz(NUMNOTES * sizeof (struct memelfnote));
2230 if (info->notes == NULL)
2231 return (-ENOMEM);
2232 info->prstatus = qemu_mallocz(sizeof (*info->prstatus));
2233 if (info->prstatus == NULL)
2234 return (-ENOMEM);
2235 info->psinfo = qemu_mallocz(sizeof (*info->psinfo));
2236 if (info->prstatus == NULL)
2237 return (-ENOMEM);
2240 * First fill in status (and registers) of current thread
2241 * including process info & aux vector.
2243 fill_prstatus(info->prstatus, ts, signr);
2244 elf_core_copy_regs(&info->prstatus->pr_reg, env);
2245 fill_note(&info->notes[0], "CORE", NT_PRSTATUS,
2246 sizeof (*info->prstatus), info->prstatus);
2247 fill_psinfo(info->psinfo, ts);
2248 fill_note(&info->notes[1], "CORE", NT_PRPSINFO,
2249 sizeof (*info->psinfo), info->psinfo);
2250 fill_auxv_note(&info->notes[2], ts);
2251 info->numnote = 3;
2253 info->notes_size = 0;
2254 for (i = 0; i < info->numnote; i++)
2255 info->notes_size += note_size(&info->notes[i]);
2257 /* read and fill status of all threads */
2258 cpu_list_lock();
2259 for (cpu = first_cpu; cpu != NULL; cpu = cpu->next_cpu) {
2260 if (cpu == thread_env)
2261 continue;
2262 fill_thread_info(info, cpu);
2264 cpu_list_unlock();
2266 return (0);
2269 static void free_note_info(struct elf_note_info *info)
2271 struct elf_thread_status *ets;
2273 while (!QTAILQ_EMPTY(&info->thread_list)) {
2274 ets = QTAILQ_FIRST(&info->thread_list);
2275 QTAILQ_REMOVE(&info->thread_list, ets, ets_link);
2276 qemu_free(ets);
2279 qemu_free(info->prstatus);
2280 qemu_free(info->psinfo);
2281 qemu_free(info->notes);
2284 static int write_note_info(struct elf_note_info *info, int fd)
2286 struct elf_thread_status *ets;
2287 int i, error = 0;
2289 /* write prstatus, psinfo and auxv for current thread */
2290 for (i = 0; i < info->numnote; i++)
2291 if ((error = write_note(&info->notes[i], fd)) != 0)
2292 return (error);
2294 /* write prstatus for each thread */
2295 for (ets = info->thread_list.tqh_first; ets != NULL;
2296 ets = ets->ets_link.tqe_next) {
2297 if ((error = write_note(&ets->notes[0], fd)) != 0)
2298 return (error);
2301 return (0);
2305 * Write out ELF coredump.
2307 * See documentation of ELF object file format in:
2308 * http://www.caldera.com/developers/devspecs/gabi41.pdf
2310 * Coredump format in linux is following:
2312 * 0 +----------------------+ \
2313 * | ELF header | ET_CORE |
2314 * +----------------------+ |
2315 * | ELF program headers | |--- headers
2316 * | - NOTE section | |
2317 * | - PT_LOAD sections | |
2318 * +----------------------+ /
2319 * | NOTEs: |
2320 * | - NT_PRSTATUS |
2321 * | - NT_PRSINFO |
2322 * | - NT_AUXV |
2323 * +----------------------+ <-- aligned to target page
2324 * | Process memory dump |
2325 * : :
2326 * . .
2327 * : :
2328 * | |
2329 * +----------------------+
2331 * NT_PRSTATUS -> struct elf_prstatus (per thread)
2332 * NT_PRSINFO -> struct elf_prpsinfo
2333 * NT_AUXV is array of { type, value } pairs (see fill_auxv_note()).
2335 * Format follows System V format as close as possible. Current
2336 * version limitations are as follows:
2337 * - no floating point registers are dumped
2339 * Function returns 0 in case of success, negative errno otherwise.
2341 * TODO: make this work also during runtime: it should be
2342 * possible to force coredump from running process and then
2343 * continue processing. For example qemu could set up SIGUSR2
2344 * handler (provided that target process haven't registered
2345 * handler for that) that does the dump when signal is received.
2347 static int elf_core_dump(int signr, const CPUState *env)
2349 const TaskState *ts = (const TaskState *)env->opaque;
2350 struct vm_area_struct *vma = NULL;
2351 char corefile[PATH_MAX];
2352 struct elf_note_info info;
2353 struct elfhdr elf;
2354 struct elf_phdr phdr;
2355 struct rlimit dumpsize;
2356 struct mm_struct *mm = NULL;
2357 off_t offset = 0, data_offset = 0;
2358 int segs = 0;
2359 int fd = -1;
2361 errno = 0;
2362 getrlimit(RLIMIT_CORE, &dumpsize);
2363 if (dumpsize.rlim_cur == 0)
2364 return 0;
2366 if (core_dump_filename(ts, corefile, sizeof (corefile)) < 0)
2367 return (-errno);
2369 if ((fd = open(corefile, O_WRONLY | O_CREAT,
2370 S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH)) < 0)
2371 return (-errno);
2374 * Walk through target process memory mappings and
2375 * set up structure containing this information. After
2376 * this point vma_xxx functions can be used.
2378 if ((mm = vma_init()) == NULL)
2379 goto out;
2381 walk_memory_regions(mm, vma_walker);
2382 segs = vma_get_mapping_count(mm);
2385 * Construct valid coredump ELF header. We also
2386 * add one more segment for notes.
2388 fill_elf_header(&elf, segs + 1, ELF_MACHINE, 0);
2389 if (dump_write(fd, &elf, sizeof (elf)) != 0)
2390 goto out;
2392 /* fill in in-memory version of notes */
2393 if (fill_note_info(&info, signr, env) < 0)
2394 goto out;
2396 offset += sizeof (elf); /* elf header */
2397 offset += (segs + 1) * sizeof (struct elf_phdr); /* program headers */
2399 /* write out notes program header */
2400 fill_elf_note_phdr(&phdr, info.notes_size, offset);
2402 offset += info.notes_size;
2403 if (dump_write(fd, &phdr, sizeof (phdr)) != 0)
2404 goto out;
2407 * ELF specification wants data to start at page boundary so
2408 * we align it here.
2410 offset = roundup(offset, ELF_EXEC_PAGESIZE);
2413 * Write program headers for memory regions mapped in
2414 * the target process.
2416 for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) {
2417 (void) memset(&phdr, 0, sizeof (phdr));
2419 phdr.p_type = PT_LOAD;
2420 phdr.p_offset = offset;
2421 phdr.p_vaddr = vma->vma_start;
2422 phdr.p_paddr = 0;
2423 phdr.p_filesz = vma_dump_size(vma);
2424 offset += phdr.p_filesz;
2425 phdr.p_memsz = vma->vma_end - vma->vma_start;
2426 phdr.p_flags = vma->vma_flags & PROT_READ ? PF_R : 0;
2427 if (vma->vma_flags & PROT_WRITE)
2428 phdr.p_flags |= PF_W;
2429 if (vma->vma_flags & PROT_EXEC)
2430 phdr.p_flags |= PF_X;
2431 phdr.p_align = ELF_EXEC_PAGESIZE;
2433 dump_write(fd, &phdr, sizeof (phdr));
2437 * Next we write notes just after program headers. No
2438 * alignment needed here.
2440 if (write_note_info(&info, fd) < 0)
2441 goto out;
2443 /* align data to page boundary */
2444 data_offset = lseek(fd, 0, SEEK_CUR);
2445 data_offset = TARGET_PAGE_ALIGN(data_offset);
2446 if (lseek(fd, data_offset, SEEK_SET) != data_offset)
2447 goto out;
2450 * Finally we can dump process memory into corefile as well.
2452 for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) {
2453 abi_ulong addr;
2454 abi_ulong end;
2456 end = vma->vma_start + vma_dump_size(vma);
2458 for (addr = vma->vma_start; addr < end;
2459 addr += TARGET_PAGE_SIZE) {
2460 char page[TARGET_PAGE_SIZE];
2461 int error;
2464 * Read in page from target process memory and
2465 * write it to coredump file.
2467 error = copy_from_user(page, addr, sizeof (page));
2468 if (error != 0) {
2469 (void) fprintf(stderr, "unable to dump " TARGET_ABI_FMT_lx "\n",
2470 addr);
2471 errno = -error;
2472 goto out;
2474 if (dump_write(fd, page, TARGET_PAGE_SIZE) < 0)
2475 goto out;
2479 out:
2480 free_note_info(&info);
2481 if (mm != NULL)
2482 vma_delete(mm);
2483 (void) close(fd);
2485 if (errno != 0)
2486 return (-errno);
2487 return (0);
2489 #endif /* USE_ELF_CORE_DUMP */
2491 void do_init_thread(struct target_pt_regs *regs, struct image_info *infop)
2493 init_thread(regs, infop);