?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
iscsi: fix races between task completion and abort
[qemu-kvm.git] / linux-user / elfload.c
blob819fdd515a554a00964d91315ba06266b1cb2d11
1 /* This is the Linux kernel elf-loading code, ported into user space */
2 #include <sys/time.h>
3 #include <sys/param.h>
4
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>
15
16 #include "qemu.h"
17 #include "disas.h"
18
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
27
28 #define ELF_OSABI ELFOSABI_SYSV
29
30 /* from personality.h */
31
32 /*
33 * Flags for bug emulation.
34 *
35 * These occupy the top three bytes.
36 */
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,
49 };
50
51 /*
52 * Personality types.
53 *
54 * These go in the low byte. Avoid using the top bit, it will
55 * conflict with error returns.
56 */
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,
81 };
82
83 /*
84 * Return the base personality without flags.
85 */
86 #define personality(pers) (pers & PER_MASK)
87
88 /* this flag is uneffective under linux too, should be deleted */
89 #ifndef MAP_DENYWRITE
90 #define MAP_DENYWRITE 0
91 #endif
92
93 /* should probably go in elf.h */
94 #ifndef ELIBBAD
95 #define ELIBBAD 80
96 #endif
97
98 #ifdef TARGET_WORDS_BIGENDIAN
99 #define ELF_DATA ELFDATA2MSB
100 #else
101 #define ELF_DATA ELFDATA2LSB
102 #endif
103
104 typedef target_ulong target_elf_greg_t;
105 #ifdef USE_UID16
106 typedef target_ushort target_uid_t;
107 typedef target_ushort target_gid_t;
108 #else
109 typedef target_uint target_uid_t;
110 typedef target_uint target_gid_t;
111 #endif
112 typedef target_int target_pid_t;
113
114 #ifdef TARGET_I386
115
116 #define ELF_PLATFORM get_elf_platform()
117
118 static const char *get_elf_platform(void)
119 {
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;
127 }
128
129 #define ELF_HWCAP get_elf_hwcap()
130
131 static uint32_t get_elf_hwcap(void)
132 {
133 return thread_env->cpuid_features;
134 }
135
136 #ifdef TARGET_X86_64
137 #define ELF_START_MMAP 0x2aaaaab000ULL
138 #define elf_check_arch(x) ( ((x) == ELF_ARCH) )
139
140 #define ELF_CLASS ELFCLASS64
141 #define ELF_ARCH EM_X86_64
142
143 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop)
144 {
145 regs->rax = 0;
146 regs->rsp = infop->start_stack;
147 regs->rip = infop->entry;
148 }
149
150 #define ELF_NREG 27
151 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
152
153 /*
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.
157 *
158 * See linux kernel: arch/x86/include/asm/elf.h
159 */
160 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUX86State *env)
161 {
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;
189 }
190
191 #else
192
193 #define ELF_START_MMAP 0x80000000
194
195 /*
196 * This is used to ensure we don't load something for the wrong architecture.
197 */
198 #define elf_check_arch(x) ( ((x) == EM_386) || ((x) == EM_486) )
199
200 /*
201 * These are used to set parameters in the core dumps.
202 */
203 #define ELF_CLASS ELFCLASS32
204 #define ELF_ARCH EM_386
205
206 static inline void init_thread(struct target_pt_regs *regs,
207 struct image_info *infop)
208 {
209 regs->esp = infop->start_stack;
210 regs->eip = infop->entry;
211
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.
217
218 A value of 0 tells we have no such handler. */
219 regs->edx = 0;
220 }
221
222 #define ELF_NREG 17
223 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
224
225 /*
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.
229 *
230 * See linux kernel: arch/x86/include/asm/elf.h
231 */
232 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUX86State *env)
233 {
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;
251 }
252 #endif
253
254 #define USE_ELF_CORE_DUMP
255 #define ELF_EXEC_PAGESIZE 4096
256
257 #endif
258
259 #ifdef TARGET_ARM
260
261 #define ELF_START_MMAP 0x80000000
262
263 #define elf_check_arch(x) ( (x) == EM_ARM )
264
265 #define ELF_CLASS ELFCLASS32
266 #define ELF_ARCH EM_ARM
267
268 static inline void init_thread(struct target_pt_regs *regs,
269 struct image_info *infop)
270 {
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;
286 }
287
288 #define ELF_NREG 18
289 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
290
291 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUARMState *env)
292 {
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]);
309
310 (*regs)[16] = tswapl(cpsr_read((CPUARMState *)env));
311 (*regs)[17] = tswapl(env->regs[0]); /* XXX */
312 }
313
314 #define USE_ELF_CORE_DUMP
315 #define ELF_EXEC_PAGESIZE 4096
316
317 enum
318 {
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,
333 };
334
335 #define TARGET_HAS_VALIDATE_GUEST_SPACE
336 /* Return 1 if the proposed guest space is suitable for the guest.
337 * Return 0 if the proposed guest space isn't suitable, but another
338 * address space should be tried.
339 * Return -1 if there is no way the proposed guest space can be
340 * valid regardless of the base.
341 * The guest code may leave a page mapped and populate it if the
342 * address is suitable.
343 */
344 static int validate_guest_space(unsigned long guest_base,
345 unsigned long guest_size)
346 {
347 unsigned long real_start, test_page_addr;
348
349 /* We need to check that we can force a fault on access to the
350 * commpage at 0xffff0fxx
351 */
352 test_page_addr = guest_base + (0xffff0f00 & qemu_host_page_mask);
353
354 /* If the commpage lies within the already allocated guest space,
355 * then there is no way we can allocate it.
356 */
357 if (test_page_addr >= guest_base
358 && test_page_addr <= (guest_base + guest_size)) {
359 return -1;
360 }
361
362 /* Note it needs to be writeable to let us initialise it */
363 real_start = (unsigned long)
364 mmap((void *)test_page_addr, qemu_host_page_size,
365 PROT_READ | PROT_WRITE,
366 MAP_ANONYMOUS | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
367
368 /* If we can't map it then try another address */
369 if (real_start == -1ul) {
370 return 0;
371 }
372
373 if (real_start != test_page_addr) {
374 /* OS didn't put the page where we asked - unmap and reject */
375 munmap((void *)real_start, qemu_host_page_size);
376 return 0;
377 }
378
379 /* Leave the page mapped
380 * Populate it (mmap should have left it all 0'd)
381 */
382
383 /* Kernel helper versions */
384 __put_user(5, (uint32_t *)g2h(0xffff0ffcul));
385
386 /* Now it's populated make it RO */
387 if (mprotect((void *)test_page_addr, qemu_host_page_size, PROT_READ)) {
388 perror("Protecting guest commpage");
389 exit(-1);
390 }
391
392 return 1; /* All good */
393 }
394
395
396 #define ELF_HWCAP get_elf_hwcap()
397
398 static uint32_t get_elf_hwcap(void)
399 {
400 CPUARMState *e = thread_env;
401 uint32_t hwcaps = 0;
402
403 hwcaps |= ARM_HWCAP_ARM_SWP;
404 hwcaps |= ARM_HWCAP_ARM_HALF;
405 hwcaps |= ARM_HWCAP_ARM_THUMB;
406 hwcaps |= ARM_HWCAP_ARM_FAST_MULT;
407 hwcaps |= ARM_HWCAP_ARM_FPA;
408
409 /* probe for the extra features */
410 #define GET_FEATURE(feat, hwcap) \
411 do {if (arm_feature(e, feat)) { hwcaps |= hwcap; } } while (0)
412 GET_FEATURE(ARM_FEATURE_VFP, ARM_HWCAP_ARM_VFP);
413 GET_FEATURE(ARM_FEATURE_IWMMXT, ARM_HWCAP_ARM_IWMMXT);
414 GET_FEATURE(ARM_FEATURE_THUMB2EE, ARM_HWCAP_ARM_THUMBEE);
415 GET_FEATURE(ARM_FEATURE_NEON, ARM_HWCAP_ARM_NEON);
416 GET_FEATURE(ARM_FEATURE_VFP3, ARM_HWCAP_ARM_VFPv3);
417 GET_FEATURE(ARM_FEATURE_VFP_FP16, ARM_HWCAP_ARM_VFPv3D16);
418 #undef GET_FEATURE
419
420 return hwcaps;
421 }
422
423 #endif
424
425 #ifdef TARGET_UNICORE32
426
427 #define ELF_START_MMAP 0x80000000
428
429 #define elf_check_arch(x) ((x) == EM_UNICORE32)
430
431 #define ELF_CLASS ELFCLASS32
432 #define ELF_DATA ELFDATA2LSB
433 #define ELF_ARCH EM_UNICORE32
434
435 static inline void init_thread(struct target_pt_regs *regs,
436 struct image_info *infop)
437 {
438 abi_long stack = infop->start_stack;
439 memset(regs, 0, sizeof(*regs));
440 regs->UC32_REG_asr = 0x10;
441 regs->UC32_REG_pc = infop->entry & 0xfffffffe;
442 regs->UC32_REG_sp = infop->start_stack;
443 /* FIXME - what to for failure of get_user()? */
444 get_user_ual(regs->UC32_REG_02, stack + 8); /* envp */
445 get_user_ual(regs->UC32_REG_01, stack + 4); /* envp */
446 /* XXX: it seems that r0 is zeroed after ! */
447 regs->UC32_REG_00 = 0;
448 }
449
450 #define ELF_NREG 34
451 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
452
453 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUUniCore32State *env)
454 {
455 (*regs)[0] = env->regs[0];
456 (*regs)[1] = env->regs[1];
457 (*regs)[2] = env->regs[2];
458 (*regs)[3] = env->regs[3];
459 (*regs)[4] = env->regs[4];
460 (*regs)[5] = env->regs[5];
461 (*regs)[6] = env->regs[6];
462 (*regs)[7] = env->regs[7];
463 (*regs)[8] = env->regs[8];
464 (*regs)[9] = env->regs[9];
465 (*regs)[10] = env->regs[10];
466 (*regs)[11] = env->regs[11];
467 (*regs)[12] = env->regs[12];
468 (*regs)[13] = env->regs[13];
469 (*regs)[14] = env->regs[14];
470 (*regs)[15] = env->regs[15];
471 (*regs)[16] = env->regs[16];
472 (*regs)[17] = env->regs[17];
473 (*regs)[18] = env->regs[18];
474 (*regs)[19] = env->regs[19];
475 (*regs)[20] = env->regs[20];
476 (*regs)[21] = env->regs[21];
477 (*regs)[22] = env->regs[22];
478 (*regs)[23] = env->regs[23];
479 (*regs)[24] = env->regs[24];
480 (*regs)[25] = env->regs[25];
481 (*regs)[26] = env->regs[26];
482 (*regs)[27] = env->regs[27];
483 (*regs)[28] = env->regs[28];
484 (*regs)[29] = env->regs[29];
485 (*regs)[30] = env->regs[30];
486 (*regs)[31] = env->regs[31];
487
488 (*regs)[32] = cpu_asr_read((CPUUniCore32State *)env);
489 (*regs)[33] = env->regs[0]; /* XXX */
490 }
491
492 #define USE_ELF_CORE_DUMP
493 #define ELF_EXEC_PAGESIZE 4096
494
495 #define ELF_HWCAP (UC32_HWCAP_CMOV | UC32_HWCAP_UCF64)
496
497 #endif
498
499 #ifdef TARGET_SPARC
500 #ifdef TARGET_SPARC64
501
502 #define ELF_START_MMAP 0x80000000
503 #define ELF_HWCAP (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \
504 | HWCAP_SPARC_MULDIV | HWCAP_SPARC_V9)
505 #ifndef TARGET_ABI32
506 #define elf_check_arch(x) ( (x) == EM_SPARCV9 || (x) == EM_SPARC32PLUS )
507 #else
508 #define elf_check_arch(x) ( (x) == EM_SPARC32PLUS || (x) == EM_SPARC )
509 #endif
510
511 #define ELF_CLASS ELFCLASS64
512 #define ELF_ARCH EM_SPARCV9
513
514 #define STACK_BIAS 2047
515
516 static inline void init_thread(struct target_pt_regs *regs,
517 struct image_info *infop)
518 {
519 #ifndef TARGET_ABI32
520 regs->tstate = 0;
521 #endif
522 regs->pc = infop->entry;
523 regs->npc = regs->pc + 4;
524 regs->y = 0;
525 #ifdef TARGET_ABI32
526 regs->u_regs[14] = infop->start_stack - 16 * 4;
527 #else
528 if (personality(infop->personality) == PER_LINUX32)
529 regs->u_regs[14] = infop->start_stack - 16 * 4;
530 else
531 regs->u_regs[14] = infop->start_stack - 16 * 8 - STACK_BIAS;
532 #endif
533 }
534
535 #else
536 #define ELF_START_MMAP 0x80000000
537 #define ELF_HWCAP (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \
538 | HWCAP_SPARC_MULDIV)
539 #define elf_check_arch(x) ( (x) == EM_SPARC )
540
541 #define ELF_CLASS ELFCLASS32
542 #define ELF_ARCH EM_SPARC
543
544 static inline void init_thread(struct target_pt_regs *regs,
545 struct image_info *infop)
546 {
547 regs->psr = 0;
548 regs->pc = infop->entry;
549 regs->npc = regs->pc + 4;
550 regs->y = 0;
551 regs->u_regs[14] = infop->start_stack - 16 * 4;
552 }
553
554 #endif
555 #endif
556
557 #ifdef TARGET_PPC
558
559 #define ELF_START_MMAP 0x80000000
560
561 #if defined(TARGET_PPC64) && !defined(TARGET_ABI32)
562
563 #define elf_check_arch(x) ( (x) == EM_PPC64 )
564
565 #define ELF_CLASS ELFCLASS64
566
567 #else
568
569 #define elf_check_arch(x) ( (x) == EM_PPC )
570
571 #define ELF_CLASS ELFCLASS32
572
573 #endif
574
575 #define ELF_ARCH EM_PPC
576
577 /* Feature masks for the Aux Vector Hardware Capabilities (AT_HWCAP).
578 See arch/powerpc/include/asm/cputable.h. */
579 enum {
580 QEMU_PPC_FEATURE_32 = 0x80000000,
581 QEMU_PPC_FEATURE_64 = 0x40000000,
582 QEMU_PPC_FEATURE_601_INSTR = 0x20000000,
583 QEMU_PPC_FEATURE_HAS_ALTIVEC = 0x10000000,
584 QEMU_PPC_FEATURE_HAS_FPU = 0x08000000,
585 QEMU_PPC_FEATURE_HAS_MMU = 0x04000000,
586 QEMU_PPC_FEATURE_HAS_4xxMAC = 0x02000000,
587 QEMU_PPC_FEATURE_UNIFIED_CACHE = 0x01000000,
588 QEMU_PPC_FEATURE_HAS_SPE = 0x00800000,
589 QEMU_PPC_FEATURE_HAS_EFP_SINGLE = 0x00400000,
590 QEMU_PPC_FEATURE_HAS_EFP_DOUBLE = 0x00200000,
591 QEMU_PPC_FEATURE_NO_TB = 0x00100000,
592 QEMU_PPC_FEATURE_POWER4 = 0x00080000,
593 QEMU_PPC_FEATURE_POWER5 = 0x00040000,
594 QEMU_PPC_FEATURE_POWER5_PLUS = 0x00020000,
595 QEMU_PPC_FEATURE_CELL = 0x00010000,
596 QEMU_PPC_FEATURE_BOOKE = 0x00008000,
597 QEMU_PPC_FEATURE_SMT = 0x00004000,
598 QEMU_PPC_FEATURE_ICACHE_SNOOP = 0x00002000,
599 QEMU_PPC_FEATURE_ARCH_2_05 = 0x00001000,
600 QEMU_PPC_FEATURE_PA6T = 0x00000800,
601 QEMU_PPC_FEATURE_HAS_DFP = 0x00000400,
602 QEMU_PPC_FEATURE_POWER6_EXT = 0x00000200,
603 QEMU_PPC_FEATURE_ARCH_2_06 = 0x00000100,
604 QEMU_PPC_FEATURE_HAS_VSX = 0x00000080,
605 QEMU_PPC_FEATURE_PSERIES_PERFMON_COMPAT = 0x00000040,
606
607 QEMU_PPC_FEATURE_TRUE_LE = 0x00000002,
608 QEMU_PPC_FEATURE_PPC_LE = 0x00000001,
609 };
610
611 #define ELF_HWCAP get_elf_hwcap()
612
613 static uint32_t get_elf_hwcap(void)
614 {
615 CPUPPCState *e = thread_env;
616 uint32_t features = 0;
617
618 /* We don't have to be terribly complete here; the high points are
619 Altivec/FP/SPE support. Anything else is just a bonus. */
620 #define GET_FEATURE(flag, feature) \
621 do {if (e->insns_flags & flag) features |= feature; } while(0)
622 GET_FEATURE(PPC_64B, QEMU_PPC_FEATURE_64);
623 GET_FEATURE(PPC_FLOAT, QEMU_PPC_FEATURE_HAS_FPU);
624 GET_FEATURE(PPC_ALTIVEC, QEMU_PPC_FEATURE_HAS_ALTIVEC);
625 GET_FEATURE(PPC_SPE, QEMU_PPC_FEATURE_HAS_SPE);
626 GET_FEATURE(PPC_SPE_SINGLE, QEMU_PPC_FEATURE_HAS_EFP_SINGLE);
627 GET_FEATURE(PPC_SPE_DOUBLE, QEMU_PPC_FEATURE_HAS_EFP_DOUBLE);
628 GET_FEATURE(PPC_BOOKE, QEMU_PPC_FEATURE_BOOKE);
629 GET_FEATURE(PPC_405_MAC, QEMU_PPC_FEATURE_HAS_4xxMAC);
630 #undef GET_FEATURE
631
632 return features;
633 }
634
635 /*
636 * The requirements here are:
637 * - keep the final alignment of sp (sp & 0xf)
638 * - make sure the 32-bit value at the first 16 byte aligned position of
639 * AUXV is greater than 16 for glibc compatibility.
640 * AT_IGNOREPPC is used for that.
641 * - for compatibility with glibc ARCH_DLINFO must always be defined on PPC,
642 * even if DLINFO_ARCH_ITEMS goes to zero or is undefined.
643 */
644 #define DLINFO_ARCH_ITEMS 5
645 #define ARCH_DLINFO \
646 do { \
647 NEW_AUX_ENT(AT_DCACHEBSIZE, 0x20); \
648 NEW_AUX_ENT(AT_ICACHEBSIZE, 0x20); \
649 NEW_AUX_ENT(AT_UCACHEBSIZE, 0); \
650 /* \
651 * Now handle glibc compatibility. \
652 */ \
653 NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC); \
654 NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC); \
655 } while (0)
656
657 static inline void init_thread(struct target_pt_regs *_regs, struct image_info *infop)
658 {
659 _regs->gpr[1] = infop->start_stack;
660 #if defined(TARGET_PPC64) && !defined(TARGET_ABI32)
661 _regs->gpr[2] = ldq_raw(infop->entry + 8) + infop->load_bias;
662 infop->entry = ldq_raw(infop->entry) + infop->load_bias;
663 #endif
664 _regs->nip = infop->entry;
665 }
666
667 /* See linux kernel: arch/powerpc/include/asm/elf.h. */
668 #define ELF_NREG 48
669 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
670
671 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUPPCState *env)
672 {
673 int i;
674 target_ulong ccr = 0;
675
676 for (i = 0; i < ARRAY_SIZE(env->gpr); i++) {
677 (*regs)[i] = tswapl(env->gpr[i]);
678 }
679
680 (*regs)[32] = tswapl(env->nip);
681 (*regs)[33] = tswapl(env->msr);
682 (*regs)[35] = tswapl(env->ctr);
683 (*regs)[36] = tswapl(env->lr);
684 (*regs)[37] = tswapl(env->xer);
685
686 for (i = 0; i < ARRAY_SIZE(env->crf); i++) {
687 ccr |= env->crf[i] << (32 - ((i + 1) * 4));
688 }
689 (*regs)[38] = tswapl(ccr);
690 }
691
692 #define USE_ELF_CORE_DUMP
693 #define ELF_EXEC_PAGESIZE 4096
694
695 #endif
696
697 #ifdef TARGET_MIPS
698
699 #define ELF_START_MMAP 0x80000000
700
701 #define elf_check_arch(x) ( (x) == EM_MIPS )
702
703 #ifdef TARGET_MIPS64
704 #define ELF_CLASS ELFCLASS64
705 #else
706 #define ELF_CLASS ELFCLASS32
707 #endif
708 #define ELF_ARCH EM_MIPS
709
710 static inline void init_thread(struct target_pt_regs *regs,
711 struct image_info *infop)
712 {
713 regs->cp0_status = 2 << CP0St_KSU;
714 regs->cp0_epc = infop->entry;
715 regs->regs[29] = infop->start_stack;
716 }
717
718 /* See linux kernel: arch/mips/include/asm/elf.h. */
719 #define ELF_NREG 45
720 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
721
722 /* See linux kernel: arch/mips/include/asm/reg.h. */
723 enum {
724 #ifdef TARGET_MIPS64
725 TARGET_EF_R0 = 0,
726 #else
727 TARGET_EF_R0 = 6,
728 #endif
729 TARGET_EF_R26 = TARGET_EF_R0 + 26,
730 TARGET_EF_R27 = TARGET_EF_R0 + 27,
731 TARGET_EF_LO = TARGET_EF_R0 + 32,
732 TARGET_EF_HI = TARGET_EF_R0 + 33,
733 TARGET_EF_CP0_EPC = TARGET_EF_R0 + 34,
734 TARGET_EF_CP0_BADVADDR = TARGET_EF_R0 + 35,
735 TARGET_EF_CP0_STATUS = TARGET_EF_R0 + 36,
736 TARGET_EF_CP0_CAUSE = TARGET_EF_R0 + 37
737 };
738
739 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */
740 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMIPSState *env)
741 {
742 int i;
743
744 for (i = 0; i < TARGET_EF_R0; i++) {
745 (*regs)[i] = 0;
746 }
747 (*regs)[TARGET_EF_R0] = 0;
748
749 for (i = 1; i < ARRAY_SIZE(env->active_tc.gpr); i++) {
750 (*regs)[TARGET_EF_R0 + i] = tswapl(env->active_tc.gpr[i]);
751 }
752
753 (*regs)[TARGET_EF_R26] = 0;
754 (*regs)[TARGET_EF_R27] = 0;
755 (*regs)[TARGET_EF_LO] = tswapl(env->active_tc.LO[0]);
756 (*regs)[TARGET_EF_HI] = tswapl(env->active_tc.HI[0]);
757 (*regs)[TARGET_EF_CP0_EPC] = tswapl(env->active_tc.PC);
758 (*regs)[TARGET_EF_CP0_BADVADDR] = tswapl(env->CP0_BadVAddr);
759 (*regs)[TARGET_EF_CP0_STATUS] = tswapl(env->CP0_Status);
760 (*regs)[TARGET_EF_CP0_CAUSE] = tswapl(env->CP0_Cause);
761 }
762
763 #define USE_ELF_CORE_DUMP
764 #define ELF_EXEC_PAGESIZE 4096
765
766 #endif /* TARGET_MIPS */
767
768 #ifdef TARGET_MICROBLAZE
769
770 #define ELF_START_MMAP 0x80000000
771
772 #define elf_check_arch(x) ( (x) == EM_MICROBLAZE || (x) == EM_MICROBLAZE_OLD)
773
774 #define ELF_CLASS ELFCLASS32
775 #define ELF_ARCH EM_MICROBLAZE
776
777 static inline void init_thread(struct target_pt_regs *regs,
778 struct image_info *infop)
779 {
780 regs->pc = infop->entry;
781 regs->r1 = infop->start_stack;
782
783 }
784
785 #define ELF_EXEC_PAGESIZE 4096
786
787 #define USE_ELF_CORE_DUMP
788 #define ELF_NREG 38
789 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
790
791 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */
792 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMBState *env)
793 {
794 int i, pos = 0;
795
796 for (i = 0; i < 32; i++) {
797 (*regs)[pos++] = tswapl(env->regs[i]);
798 }
799
800 for (i = 0; i < 6; i++) {
801 (*regs)[pos++] = tswapl(env->sregs[i]);
802 }
803 }
804
805 #endif /* TARGET_MICROBLAZE */
806
807 #ifdef TARGET_OPENRISC
808
809 #define ELF_START_MMAP 0x08000000
810
811 #define elf_check_arch(x) ((x) == EM_OPENRISC)
812
813 #define ELF_ARCH EM_OPENRISC
814 #define ELF_CLASS ELFCLASS32
815 #define ELF_DATA ELFDATA2MSB
816
817 static inline void init_thread(struct target_pt_regs *regs,
818 struct image_info *infop)
819 {
820 regs->pc = infop->entry;
821 regs->gpr[1] = infop->start_stack;
822 }
823
824 #define USE_ELF_CORE_DUMP
825 #define ELF_EXEC_PAGESIZE 8192
826
827 /* See linux kernel arch/openrisc/include/asm/elf.h. */
828 #define ELF_NREG 34 /* gprs and pc, sr */
829 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
830
831 static void elf_core_copy_regs(target_elf_gregset_t *regs,
832 const CPUOpenRISCState *env)
833 {
834 int i;
835
836 for (i = 0; i < 32; i++) {
837 (*regs)[i] = tswapl(env->gpr[i]);
838 }
839
840 (*regs)[32] = tswapl(env->pc);
841 (*regs)[33] = tswapl(env->sr);
842 }
843 #define ELF_HWCAP 0
844 #define ELF_PLATFORM NULL
845
846 #endif /* TARGET_OPENRISC */
847
848 #ifdef TARGET_SH4
849
850 #define ELF_START_MMAP 0x80000000
851
852 #define elf_check_arch(x) ( (x) == EM_SH )
853
854 #define ELF_CLASS ELFCLASS32
855 #define ELF_ARCH EM_SH
856
857 static inline void init_thread(struct target_pt_regs *regs,
858 struct image_info *infop)
859 {
860 /* Check other registers XXXXX */
861 regs->pc = infop->entry;
862 regs->regs[15] = infop->start_stack;
863 }
864
865 /* See linux kernel: arch/sh/include/asm/elf.h. */
866 #define ELF_NREG 23
867 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
868
869 /* See linux kernel: arch/sh/include/asm/ptrace.h. */
870 enum {
871 TARGET_REG_PC = 16,
872 TARGET_REG_PR = 17,
873 TARGET_REG_SR = 18,
874 TARGET_REG_GBR = 19,
875 TARGET_REG_MACH = 20,
876 TARGET_REG_MACL = 21,
877 TARGET_REG_SYSCALL = 22
878 };
879
880 static inline void elf_core_copy_regs(target_elf_gregset_t *regs,
881 const CPUSH4State *env)
882 {
883 int i;
884
885 for (i = 0; i < 16; i++) {
886 (*regs[i]) = tswapl(env->gregs[i]);
887 }
888
889 (*regs)[TARGET_REG_PC] = tswapl(env->pc);
890 (*regs)[TARGET_REG_PR] = tswapl(env->pr);
891 (*regs)[TARGET_REG_SR] = tswapl(env->sr);
892 (*regs)[TARGET_REG_GBR] = tswapl(env->gbr);
893 (*regs)[TARGET_REG_MACH] = tswapl(env->mach);
894 (*regs)[TARGET_REG_MACL] = tswapl(env->macl);
895 (*regs)[TARGET_REG_SYSCALL] = 0; /* FIXME */
896 }
897
898 #define USE_ELF_CORE_DUMP
899 #define ELF_EXEC_PAGESIZE 4096
900
901 #endif
902
903 #ifdef TARGET_CRIS
904
905 #define ELF_START_MMAP 0x80000000
906
907 #define elf_check_arch(x) ( (x) == EM_CRIS )
908
909 #define ELF_CLASS ELFCLASS32
910 #define ELF_ARCH EM_CRIS
911
912 static inline void init_thread(struct target_pt_regs *regs,
913 struct image_info *infop)
914 {
915 regs->erp = infop->entry;
916 }
917
918 #define ELF_EXEC_PAGESIZE 8192
919
920 #endif
921
922 #ifdef TARGET_M68K
923
924 #define ELF_START_MMAP 0x80000000
925
926 #define elf_check_arch(x) ( (x) == EM_68K )
927
928 #define ELF_CLASS ELFCLASS32
929 #define ELF_ARCH EM_68K
930
931 /* ??? Does this need to do anything?
932 #define ELF_PLAT_INIT(_r) */
933
934 static inline void init_thread(struct target_pt_regs *regs,
935 struct image_info *infop)
936 {
937 regs->usp = infop->start_stack;
938 regs->sr = 0;
939 regs->pc = infop->entry;
940 }
941
942 /* See linux kernel: arch/m68k/include/asm/elf.h. */
943 #define ELF_NREG 20
944 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
945
946 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUM68KState *env)
947 {
948 (*regs)[0] = tswapl(env->dregs[1]);
949 (*regs)[1] = tswapl(env->dregs[2]);
950 (*regs)[2] = tswapl(env->dregs[3]);
951 (*regs)[3] = tswapl(env->dregs[4]);
952 (*regs)[4] = tswapl(env->dregs[5]);
953 (*regs)[5] = tswapl(env->dregs[6]);
954 (*regs)[6] = tswapl(env->dregs[7]);
955 (*regs)[7] = tswapl(env->aregs[0]);
956 (*regs)[8] = tswapl(env->aregs[1]);
957 (*regs)[9] = tswapl(env->aregs[2]);
958 (*regs)[10] = tswapl(env->aregs[3]);
959 (*regs)[11] = tswapl(env->aregs[4]);
960 (*regs)[12] = tswapl(env->aregs[5]);
961 (*regs)[13] = tswapl(env->aregs[6]);
962 (*regs)[14] = tswapl(env->dregs[0]);
963 (*regs)[15] = tswapl(env->aregs[7]);
964 (*regs)[16] = tswapl(env->dregs[0]); /* FIXME: orig_d0 */
965 (*regs)[17] = tswapl(env->sr);
966 (*regs)[18] = tswapl(env->pc);
967 (*regs)[19] = 0; /* FIXME: regs->format | regs->vector */
968 }
969
970 #define USE_ELF_CORE_DUMP
971 #define ELF_EXEC_PAGESIZE 8192
972
973 #endif
974
975 #ifdef TARGET_ALPHA
976
977 #define ELF_START_MMAP (0x30000000000ULL)
978
979 #define elf_check_arch(x) ( (x) == ELF_ARCH )
980
981 #define ELF_CLASS ELFCLASS64
982 #define ELF_ARCH EM_ALPHA
983
984 static inline void init_thread(struct target_pt_regs *regs,
985 struct image_info *infop)
986 {
987 regs->pc = infop->entry;
988 regs->ps = 8;
989 regs->usp = infop->start_stack;
990 }
991
992 #define ELF_EXEC_PAGESIZE 8192
993
994 #endif /* TARGET_ALPHA */
995
996 #ifdef TARGET_S390X
997
998 #define ELF_START_MMAP (0x20000000000ULL)
999
1000 #define elf_check_arch(x) ( (x) == ELF_ARCH )
1001
1002 #define ELF_CLASS ELFCLASS64
1003 #define ELF_DATA ELFDATA2MSB
1004 #define ELF_ARCH EM_S390
1005
1006 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop)
1007 {
1008 regs->psw.addr = infop->entry;
1009 regs->psw.mask = PSW_MASK_64 | PSW_MASK_32;
1010 regs->gprs[15] = infop->start_stack;
1011 }
1012
1013 #endif /* TARGET_S390X */
1014
1015 #ifndef ELF_PLATFORM
1016 #define ELF_PLATFORM (NULL)
1017 #endif
1018
1019 #ifndef ELF_HWCAP
1020 #define ELF_HWCAP 0
1021 #endif
1022
1023 #ifdef TARGET_ABI32
1024 #undef ELF_CLASS
1025 #define ELF_CLASS ELFCLASS32
1026 #undef bswaptls
1027 #define bswaptls(ptr) bswap32s(ptr)
1028 #endif
1029
1030 #include "elf.h"
1031
1032 struct exec
1033 {
1034 unsigned int a_info; /* Use macros N_MAGIC, etc for access */
1035 unsigned int a_text; /* length of text, in bytes */
1036 unsigned int a_data; /* length of data, in bytes */
1037 unsigned int a_bss; /* length of uninitialized data area, in bytes */
1038 unsigned int a_syms; /* length of symbol table data in file, in bytes */
1039 unsigned int a_entry; /* start address */
1040 unsigned int a_trsize; /* length of relocation info for text, in bytes */
1041 unsigned int a_drsize; /* length of relocation info for data, in bytes */
1042 };
1043
1044
1045 #define N_MAGIC(exec) ((exec).a_info & 0xffff)
1046 #define OMAGIC 0407
1047 #define NMAGIC 0410
1048 #define ZMAGIC 0413
1049 #define QMAGIC 0314
1050
1051 /* Necessary parameters */
1052 #define TARGET_ELF_EXEC_PAGESIZE TARGET_PAGE_SIZE
1053 #define TARGET_ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(TARGET_ELF_EXEC_PAGESIZE-1))
1054 #define TARGET_ELF_PAGEOFFSET(_v) ((_v) & (TARGET_ELF_EXEC_PAGESIZE-1))
1055
1056 #define DLINFO_ITEMS 13
1057
1058 static inline void memcpy_fromfs(void * to, const void * from, unsigned long n)
1059 {
1060 memcpy(to, from, n);
1061 }
1062
1063 #ifdef BSWAP_NEEDED
1064 static void bswap_ehdr(struct elfhdr *ehdr)
1065 {
1066 bswap16s(&ehdr->e_type); /* Object file type */
1067 bswap16s(&ehdr->e_machine); /* Architecture */
1068 bswap32s(&ehdr->e_version); /* Object file version */
1069 bswaptls(&ehdr->e_entry); /* Entry point virtual address */
1070 bswaptls(&ehdr->e_phoff); /* Program header table file offset */
1071 bswaptls(&ehdr->e_shoff); /* Section header table file offset */
1072 bswap32s(&ehdr->e_flags); /* Processor-specific flags */
1073 bswap16s(&ehdr->e_ehsize); /* ELF header size in bytes */
1074 bswap16s(&ehdr->e_phentsize); /* Program header table entry size */
1075 bswap16s(&ehdr->e_phnum); /* Program header table entry count */
1076 bswap16s(&ehdr->e_shentsize); /* Section header table entry size */
1077 bswap16s(&ehdr->e_shnum); /* Section header table entry count */
1078 bswap16s(&ehdr->e_shstrndx); /* Section header string table index */
1079 }
1080
1081 static void bswap_phdr(struct elf_phdr *phdr, int phnum)
1082 {
1083 int i;
1084 for (i = 0; i < phnum; ++i, ++phdr) {
1085 bswap32s(&phdr->p_type); /* Segment type */
1086 bswap32s(&phdr->p_flags); /* Segment flags */
1087 bswaptls(&phdr->p_offset); /* Segment file offset */
1088 bswaptls(&phdr->p_vaddr); /* Segment virtual address */
1089 bswaptls(&phdr->p_paddr); /* Segment physical address */
1090 bswaptls(&phdr->p_filesz); /* Segment size in file */
1091 bswaptls(&phdr->p_memsz); /* Segment size in memory */
1092 bswaptls(&phdr->p_align); /* Segment alignment */
1093 }
1094 }
1095
1096 static void bswap_shdr(struct elf_shdr *shdr, int shnum)
1097 {
1098 int i;
1099 for (i = 0; i < shnum; ++i, ++shdr) {
1100 bswap32s(&shdr->sh_name);
1101 bswap32s(&shdr->sh_type);
1102 bswaptls(&shdr->sh_flags);
1103 bswaptls(&shdr->sh_addr);
1104 bswaptls(&shdr->sh_offset);
1105 bswaptls(&shdr->sh_size);
1106 bswap32s(&shdr->sh_link);
1107 bswap32s(&shdr->sh_info);
1108 bswaptls(&shdr->sh_addralign);
1109 bswaptls(&shdr->sh_entsize);
1110 }
1111 }
1112
1113 static void bswap_sym(struct elf_sym *sym)
1114 {
1115 bswap32s(&sym->st_name);
1116 bswaptls(&sym->st_value);
1117 bswaptls(&sym->st_size);
1118 bswap16s(&sym->st_shndx);
1119 }
1120 #else
1121 static inline void bswap_ehdr(struct elfhdr *ehdr) { }
1122 static inline void bswap_phdr(struct elf_phdr *phdr, int phnum) { }
1123 static inline void bswap_shdr(struct elf_shdr *shdr, int shnum) { }
1124 static inline void bswap_sym(struct elf_sym *sym) { }
1125 #endif
1126
1127 #ifdef USE_ELF_CORE_DUMP
1128 static int elf_core_dump(int, const CPUArchState *);
1129 #endif /* USE_ELF_CORE_DUMP */
1130 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias);
1131
1132 /* Verify the portions of EHDR within E_IDENT for the target.
1133 This can be performed before bswapping the entire header. */
1134 static bool elf_check_ident(struct elfhdr *ehdr)
1135 {
1136 return (ehdr->e_ident[EI_MAG0] == ELFMAG0
1137 && ehdr->e_ident[EI_MAG1] == ELFMAG1
1138 && ehdr->e_ident[EI_MAG2] == ELFMAG2
1139 && ehdr->e_ident[EI_MAG3] == ELFMAG3
1140 && ehdr->e_ident[EI_CLASS] == ELF_CLASS
1141 && ehdr->e_ident[EI_DATA] == ELF_DATA
1142 && ehdr->e_ident[EI_VERSION] == EV_CURRENT);
1143 }
1144
1145 /* Verify the portions of EHDR outside of E_IDENT for the target.
1146 This has to wait until after bswapping the header. */
1147 static bool elf_check_ehdr(struct elfhdr *ehdr)
1148 {
1149 return (elf_check_arch(ehdr->e_machine)
1150 && ehdr->e_ehsize == sizeof(struct elfhdr)
1151 && ehdr->e_phentsize == sizeof(struct elf_phdr)
1152 && ehdr->e_shentsize == sizeof(struct elf_shdr)
1153 && (ehdr->e_type == ET_EXEC || ehdr->e_type == ET_DYN));
1154 }
1155
1156 /*
1157 * 'copy_elf_strings()' copies argument/envelope strings from user
1158 * memory to free pages in kernel mem. These are in a format ready
1159 * to be put directly into the top of new user memory.
1160 *
1161 */
1162 static abi_ulong copy_elf_strings(int argc,char ** argv, void **page,
1163 abi_ulong p)
1164 {
1165 char *tmp, *tmp1, *pag = NULL;
1166 int len, offset = 0;
1167
1168 if (!p) {
1169 return 0; /* bullet-proofing */
1170 }
1171 while (argc-- > 0) {
1172 tmp = argv[argc];
1173 if (!tmp) {
1174 fprintf(stderr, "VFS: argc is wrong");
1175 exit(-1);
1176 }
1177 tmp1 = tmp;
1178 while (*tmp++);
1179 len = tmp - tmp1;
1180 if (p < len) { /* this shouldn't happen - 128kB */
1181 return 0;
1182 }
1183 while (len) {
1184 --p; --tmp; --len;
1185 if (--offset < 0) {
1186 offset = p % TARGET_PAGE_SIZE;
1187 pag = (char *)page[p/TARGET_PAGE_SIZE];
1188 if (!pag) {
1189 pag = g_try_malloc0(TARGET_PAGE_SIZE);
1190 page[p/TARGET_PAGE_SIZE] = pag;
1191 if (!pag)
1192 return 0;
1193 }
1194 }
1195 if (len == 0 || offset == 0) {
1196 *(pag + offset) = *tmp;
1197 }
1198 else {
1199 int bytes_to_copy = (len > offset) ? offset : len;
1200 tmp -= bytes_to_copy;
1201 p -= bytes_to_copy;
1202 offset -= bytes_to_copy;
1203 len -= bytes_to_copy;
1204 memcpy_fromfs(pag + offset, tmp, bytes_to_copy + 1);
1205 }
1206 }
1207 }
1208 return p;
1209 }
1210
1211 static abi_ulong setup_arg_pages(abi_ulong p, struct linux_binprm *bprm,
1212 struct image_info *info)
1213 {
1214 abi_ulong stack_base, size, error, guard;
1215 int i;
1216
1217 /* Create enough stack to hold everything. If we don't use
1218 it for args, we'll use it for something else. */
1219 size = guest_stack_size;
1220 if (size < MAX_ARG_PAGES*TARGET_PAGE_SIZE) {
1221 size = MAX_ARG_PAGES*TARGET_PAGE_SIZE;
1222 }
1223 guard = TARGET_PAGE_SIZE;
1224 if (guard < qemu_real_host_page_size) {
1225 guard = qemu_real_host_page_size;
1226 }
1227
1228 error = target_mmap(0, size + guard, PROT_READ | PROT_WRITE,
1229 MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
1230 if (error == -1) {
1231 perror("mmap stack");
1232 exit(-1);
1233 }
1234
1235 /* We reserve one extra page at the top of the stack as guard. */
1236 target_mprotect(error, guard, PROT_NONE);
1237
1238 info->stack_limit = error + guard;
1239 stack_base = info->stack_limit + size - MAX_ARG_PAGES*TARGET_PAGE_SIZE;
1240 p += stack_base;
1241
1242 for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
1243 if (bprm->page[i]) {
1244 info->rss++;
1245 /* FIXME - check return value of memcpy_to_target() for failure */
1246 memcpy_to_target(stack_base, bprm->page[i], TARGET_PAGE_SIZE);
1247 g_free(bprm->page[i]);
1248 }
1249 stack_base += TARGET_PAGE_SIZE;
1250 }
1251 return p;
1252 }
1253
1254 /* Map and zero the bss. We need to explicitly zero any fractional pages
1255 after the data section (i.e. bss). */
1256 static void zero_bss(abi_ulong elf_bss, abi_ulong last_bss, int prot)
1257 {
1258 uintptr_t host_start, host_map_start, host_end;
1259
1260 last_bss = TARGET_PAGE_ALIGN(last_bss);
1261
1262 /* ??? There is confusion between qemu_real_host_page_size and
1263 qemu_host_page_size here and elsewhere in target_mmap, which
1264 may lead to the end of the data section mapping from the file
1265 not being mapped. At least there was an explicit test and
1266 comment for that here, suggesting that "the file size must
1267 be known". The comment probably pre-dates the introduction
1268 of the fstat system call in target_mmap which does in fact
1269 find out the size. What isn't clear is if the workaround
1270 here is still actually needed. For now, continue with it,
1271 but merge it with the "normal" mmap that would allocate the bss. */
1272
1273 host_start = (uintptr_t) g2h(elf_bss);
1274 host_end = (uintptr_t) g2h(last_bss);
1275 host_map_start = (host_start + qemu_real_host_page_size - 1);
1276 host_map_start &= -qemu_real_host_page_size;
1277
1278 if (host_map_start < host_end) {
1279 void *p = mmap((void *)host_map_start, host_end - host_map_start,
1280 prot, MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
1281 if (p == MAP_FAILED) {
1282 perror("cannot mmap brk");
1283 exit(-1);
1284 }
1285
1286 /* Since we didn't use target_mmap, make sure to record
1287 the validity of the pages with qemu. */
1288 page_set_flags(elf_bss & TARGET_PAGE_MASK, last_bss, prot|PAGE_VALID);
1289 }
1290
1291 if (host_start < host_map_start) {
1292 memset((void *)host_start, 0, host_map_start - host_start);
1293 }
1294 }
1295
1296 #ifdef CONFIG_USE_FDPIC
1297 static abi_ulong loader_build_fdpic_loadmap(struct image_info *info, abi_ulong sp)
1298 {
1299 uint16_t n;
1300 struct elf32_fdpic_loadseg *loadsegs = info->loadsegs;
1301
1302 /* elf32_fdpic_loadseg */
1303 n = info->nsegs;
1304 while (n--) {
1305 sp -= 12;
1306 put_user_u32(loadsegs[n].addr, sp+0);
1307 put_user_u32(loadsegs[n].p_vaddr, sp+4);
1308 put_user_u32(loadsegs[n].p_memsz, sp+8);
1309 }
1310
1311 /* elf32_fdpic_loadmap */
1312 sp -= 4;
1313 put_user_u16(0, sp+0); /* version */
1314 put_user_u16(info->nsegs, sp+2); /* nsegs */
1315
1316 info->personality = PER_LINUX_FDPIC;
1317 info->loadmap_addr = sp;
1318
1319 return sp;
1320 }
1321 #endif
1322
1323 static abi_ulong create_elf_tables(abi_ulong p, int argc, int envc,
1324 struct elfhdr *exec,
1325 struct image_info *info,
1326 struct image_info *interp_info)
1327 {
1328 abi_ulong sp;
1329 abi_ulong sp_auxv;
1330 int size;
1331 int i;
1332 abi_ulong u_rand_bytes;
1333 uint8_t k_rand_bytes[16];
1334 abi_ulong u_platform;
1335 const char *k_platform;
1336 const int n = sizeof(elf_addr_t);
1337
1338 sp = p;
1339
1340 #ifdef CONFIG_USE_FDPIC
1341 /* Needs to be before we load the env/argc/... */
1342 if (elf_is_fdpic(exec)) {
1343 /* Need 4 byte alignment for these structs */
1344 sp &= ~3;
1345 sp = loader_build_fdpic_loadmap(info, sp);
1346 info->other_info = interp_info;
1347 if (interp_info) {
1348 interp_info->other_info = info;
1349 sp = loader_build_fdpic_loadmap(interp_info, sp);
1350 }
1351 }
1352 #endif
1353
1354 u_platform = 0;
1355 k_platform = ELF_PLATFORM;
1356 if (k_platform) {
1357 size_t len = strlen(k_platform) + 1;
1358 sp -= (len + n - 1) & ~(n - 1);
1359 u_platform = sp;
1360 /* FIXME - check return value of memcpy_to_target() for failure */
1361 memcpy_to_target(sp, k_platform, len);
1362 }
1363
1364 /*
1365 * Generate 16 random bytes for userspace PRNG seeding (not
1366 * cryptically secure but it's not the aim of QEMU).
1367 */
1368 srand((unsigned int) time(NULL));
1369 for (i = 0; i < 16; i++) {
1370 k_rand_bytes[i] = rand();
1371 }
1372 sp -= 16;
1373 u_rand_bytes = sp;
1374 /* FIXME - check return value of memcpy_to_target() for failure */
1375 memcpy_to_target(sp, k_rand_bytes, 16);
1376
1377 /*
1378 * Force 16 byte _final_ alignment here for generality.
1379 */
1380 sp = sp &~ (abi_ulong)15;
1381 size = (DLINFO_ITEMS + 1) * 2;
1382 if (k_platform)
1383 size += 2;
1384 #ifdef DLINFO_ARCH_ITEMS
1385 size += DLINFO_ARCH_ITEMS * 2;
1386 #endif
1387 size += envc + argc + 2;
1388 size += 1; /* argc itself */
1389 size *= n;
1390 if (size & 15)
1391 sp -= 16 - (size & 15);
1392
1393 /* This is correct because Linux defines
1394 * elf_addr_t as Elf32_Off / Elf64_Off
1395 */
1396 #define NEW_AUX_ENT(id, val) do { \
1397 sp -= n; put_user_ual(val, sp); \
1398 sp -= n; put_user_ual(id, sp); \
1399 } while(0)
1400
1401 sp_auxv = sp;
1402 NEW_AUX_ENT (AT_NULL, 0);
1403
1404 /* There must be exactly DLINFO_ITEMS entries here. */
1405 NEW_AUX_ENT(AT_PHDR, (abi_ulong)(info->load_addr + exec->e_phoff));
1406 NEW_AUX_ENT(AT_PHENT, (abi_ulong)(sizeof (struct elf_phdr)));
1407 NEW_AUX_ENT(AT_PHNUM, (abi_ulong)(exec->e_phnum));
1408 NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(TARGET_PAGE_SIZE));
1409 NEW_AUX_ENT(AT_BASE, (abi_ulong)(interp_info ? interp_info->load_addr : 0));
1410 NEW_AUX_ENT(AT_FLAGS, (abi_ulong)0);
1411 NEW_AUX_ENT(AT_ENTRY, info->entry);
1412 NEW_AUX_ENT(AT_UID, (abi_ulong) getuid());
1413 NEW_AUX_ENT(AT_EUID, (abi_ulong) geteuid());
1414 NEW_AUX_ENT(AT_GID, (abi_ulong) getgid());
1415 NEW_AUX_ENT(AT_EGID, (abi_ulong) getegid());
1416 NEW_AUX_ENT(AT_HWCAP, (abi_ulong) ELF_HWCAP);
1417 NEW_AUX_ENT(AT_CLKTCK, (abi_ulong) sysconf(_SC_CLK_TCK));
1418 NEW_AUX_ENT(AT_RANDOM, (abi_ulong) u_rand_bytes);
1419
1420 if (k_platform)
1421 NEW_AUX_ENT(AT_PLATFORM, u_platform);
1422 #ifdef ARCH_DLINFO
1423 /*
1424 * ARCH_DLINFO must come last so platform specific code can enforce
1425 * special alignment requirements on the AUXV if necessary (eg. PPC).
1426 */
1427 ARCH_DLINFO;
1428 #endif
1429 #undef NEW_AUX_ENT
1430
1431 info->saved_auxv = sp;
1432 info->auxv_len = sp_auxv - sp;
1433
1434 sp = loader_build_argptr(envc, argc, sp, p, 0);
1435 return sp;
1436 }
1437
1438 #ifndef TARGET_HAS_VALIDATE_GUEST_SPACE
1439 /* If the guest doesn't have a validation function just agree */
1440 static int validate_guest_space(unsigned long guest_base,
1441 unsigned long guest_size)
1442 {
1443 return 1;
1444 }
1445 #endif
1446
1447 unsigned long init_guest_space(unsigned long host_start,
1448 unsigned long host_size,
1449 unsigned long guest_start,
1450 bool fixed)
1451 {
1452 unsigned long current_start, real_start;
1453 int flags;
1454
1455 assert(host_start || host_size);
1456
1457 /* If just a starting address is given, then just verify that
1458 * address. */
1459 if (host_start && !host_size) {
1460 if (validate_guest_space(host_start, host_size) == 1) {
1461 return host_start;
1462 } else {
1463 return (unsigned long)-1;
1464 }
1465 }
1466
1467 /* Setup the initial flags and start address. */
1468 current_start = host_start & qemu_host_page_mask;
1469 flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE;
1470 if (fixed) {
1471 flags |= MAP_FIXED;
1472 }
1473
1474 /* Otherwise, a non-zero size region of memory needs to be mapped
1475 * and validated. */
1476 while (1) {
1477 unsigned long real_size = host_size;
1478
1479 /* Do not use mmap_find_vma here because that is limited to the
1480 * guest address space. We are going to make the
1481 * guest address space fit whatever we're given.
1482 */
1483 real_start = (unsigned long)
1484 mmap((void *)current_start, host_size, PROT_NONE, flags, -1, 0);
1485 if (real_start == (unsigned long)-1) {
1486 return (unsigned long)-1;
1487 }
1488
1489 /* Ensure the address is properly aligned. */
1490 if (real_start & ~qemu_host_page_mask) {
1491 munmap((void *)real_start, host_size);
1492 real_size = host_size + qemu_host_page_size;
1493 real_start = (unsigned long)
1494 mmap((void *)real_start, real_size, PROT_NONE, flags, -1, 0);
1495 if (real_start == (unsigned long)-1) {
1496 return (unsigned long)-1;
1497 }
1498 real_start = HOST_PAGE_ALIGN(real_start);
1499 }
1500
1501 /* Check to see if the address is valid. */
1502 if (!host_start || real_start == current_start) {
1503 int valid = validate_guest_space(real_start - guest_start,
1504 real_size);
1505 if (valid == 1) {
1506 break;
1507 } else if (valid == -1) {
1508 return (unsigned long)-1;
1509 }
1510 /* valid == 0, so try again. */
1511 }
1512
1513 /* That address didn't work. Unmap and try a different one.
1514 * The address the host picked because is typically right at
1515 * the top of the host address space and leaves the guest with
1516 * no usable address space. Resort to a linear search. We
1517 * already compensated for mmap_min_addr, so this should not
1518 * happen often. Probably means we got unlucky and host
1519 * address space randomization put a shared library somewhere
1520 * inconvenient.
1521 */
1522 munmap((void *)real_start, host_size);
1523 current_start += qemu_host_page_size;
1524 if (host_start == current_start) {
1525 /* Theoretically possible if host doesn't have any suitably
1526 * aligned areas. Normally the first mmap will fail.
1527 */
1528 return (unsigned long)-1;
1529 }
1530 }
1531
1532 qemu_log("Reserved 0x%lx bytes of guest address space\n", host_size);
1533
1534 return real_start;
1535 }
1536
1537 static void probe_guest_base(const char *image_name,
1538 abi_ulong loaddr, abi_ulong hiaddr)
1539 {
1540 /* Probe for a suitable guest base address, if the user has not set
1541 * it explicitly, and set guest_base appropriately.
1542 * In case of error we will print a suitable message and exit.
1543 */
1544 #if defined(CONFIG_USE_GUEST_BASE)
1545 const char *errmsg;
1546 if (!have_guest_base && !reserved_va) {
1547 unsigned long host_start, real_start, host_size;
1548
1549 /* Round addresses to page boundaries. */
1550 loaddr &= qemu_host_page_mask;
1551 hiaddr = HOST_PAGE_ALIGN(hiaddr);
1552
1553 if (loaddr < mmap_min_addr) {
1554 host_start = HOST_PAGE_ALIGN(mmap_min_addr);
1555 } else {
1556 host_start = loaddr;
1557 if (host_start != loaddr) {
1558 errmsg = "Address overflow loading ELF binary";
1559 goto exit_errmsg;
1560 }
1561 }
1562 host_size = hiaddr - loaddr;
1563
1564 /* Setup the initial guest memory space with ranges gleaned from
1565 * the ELF image that is being loaded.
1566 */
1567 real_start = init_guest_space(host_start, host_size, loaddr, false);
1568 if (real_start == (unsigned long)-1) {
1569 errmsg = "Unable to find space for application";
1570 goto exit_errmsg;
1571 }
1572 guest_base = real_start - loaddr;
1573
1574 qemu_log("Relocating guest address space from 0x"
1575 TARGET_ABI_FMT_lx " to 0x%lx\n",
1576 loaddr, real_start);
1577 }
1578 return;
1579
1580 exit_errmsg:
1581 fprintf(stderr, "%s: %s\n", image_name, errmsg);
1582 exit(-1);
1583 #endif
1584 }
1585
1586
1587 /* Load an ELF image into the address space.
1588
1589 IMAGE_NAME is the filename of the image, to use in error messages.
1590 IMAGE_FD is the open file descriptor for the image.
1591
1592 BPRM_BUF is a copy of the beginning of the file; this of course
1593 contains the elf file header at offset 0. It is assumed that this
1594 buffer is sufficiently aligned to present no problems to the host
1595 in accessing data at aligned offsets within the buffer.
1596
1597 On return: INFO values will be filled in, as necessary or available. */
1598
1599 static void load_elf_image(const char *image_name, int image_fd,
1600 struct image_info *info, char **pinterp_name,
1601 char bprm_buf[BPRM_BUF_SIZE])
1602 {
1603 struct elfhdr *ehdr = (struct elfhdr *)bprm_buf;
1604 struct elf_phdr *phdr;
1605 abi_ulong load_addr, load_bias, loaddr, hiaddr, error;
1606 int i, retval;
1607 const char *errmsg;
1608
1609 /* First of all, some simple consistency checks */
1610 errmsg = "Invalid ELF image for this architecture";
1611 if (!elf_check_ident(ehdr)) {
1612 goto exit_errmsg;
1613 }
1614 bswap_ehdr(ehdr);
1615 if (!elf_check_ehdr(ehdr)) {
1616 goto exit_errmsg;
1617 }
1618
1619 i = ehdr->e_phnum * sizeof(struct elf_phdr);
1620 if (ehdr->e_phoff + i <= BPRM_BUF_SIZE) {
1621 phdr = (struct elf_phdr *)(bprm_buf + ehdr->e_phoff);
1622 } else {
1623 phdr = (struct elf_phdr *) alloca(i);
1624 retval = pread(image_fd, phdr, i, ehdr->e_phoff);
1625 if (retval != i) {
1626 goto exit_read;
1627 }
1628 }
1629 bswap_phdr(phdr, ehdr->e_phnum);
1630
1631 #ifdef CONFIG_USE_FDPIC
1632 info->nsegs = 0;
1633 info->pt_dynamic_addr = 0;
1634 #endif
1635
1636 /* Find the maximum size of the image and allocate an appropriate
1637 amount of memory to handle that. */
1638 loaddr = -1, hiaddr = 0;
1639 for (i = 0; i < ehdr->e_phnum; ++i) {
1640 if (phdr[i].p_type == PT_LOAD) {
1641 abi_ulong a = phdr[i].p_vaddr;
1642 if (a < loaddr) {
1643 loaddr = a;
1644 }
1645 a += phdr[i].p_memsz;
1646 if (a > hiaddr) {
1647 hiaddr = a;
1648 }
1649 #ifdef CONFIG_USE_FDPIC
1650 ++info->nsegs;
1651 #endif
1652 }
1653 }
1654
1655 load_addr = loaddr;
1656 if (ehdr->e_type == ET_DYN) {
1657 /* The image indicates that it can be loaded anywhere. Find a
1658 location that can hold the memory space required. If the
1659 image is pre-linked, LOADDR will be non-zero. Since we do
1660 not supply MAP_FIXED here we'll use that address if and
1661 only if it remains available. */
1662 load_addr = target_mmap(loaddr, hiaddr - loaddr, PROT_NONE,
1663 MAP_PRIVATE | MAP_ANON | MAP_NORESERVE,
1664 -1, 0);
1665 if (load_addr == -1) {
1666 goto exit_perror;
1667 }
1668 } else if (pinterp_name != NULL) {
1669 /* This is the main executable. Make sure that the low
1670 address does not conflict with MMAP_MIN_ADDR or the
1671 QEMU application itself. */
1672 probe_guest_base(image_name, loaddr, hiaddr);
1673 }
1674 load_bias = load_addr - loaddr;
1675
1676 #ifdef CONFIG_USE_FDPIC
1677 {
1678 struct elf32_fdpic_loadseg *loadsegs = info->loadsegs =
1679 g_malloc(sizeof(*loadsegs) * info->nsegs);
1680
1681 for (i = 0; i < ehdr->e_phnum; ++i) {
1682 switch (phdr[i].p_type) {
1683 case PT_DYNAMIC:
1684 info->pt_dynamic_addr = phdr[i].p_vaddr + load_bias;
1685 break;
1686 case PT_LOAD:
1687 loadsegs->addr = phdr[i].p_vaddr + load_bias;
1688 loadsegs->p_vaddr = phdr[i].p_vaddr;
1689 loadsegs->p_memsz = phdr[i].p_memsz;
1690 ++loadsegs;
1691 break;
1692 }
1693 }
1694 }
1695 #endif
1696
1697 info->load_bias = load_bias;
1698 info->load_addr = load_addr;
1699 info->entry = ehdr->e_entry + load_bias;
1700 info->start_code = -1;
1701 info->end_code = 0;
1702 info->start_data = -1;
1703 info->end_data = 0;
1704 info->brk = 0;
1705 info->elf_flags = ehdr->e_flags;
1706
1707 for (i = 0; i < ehdr->e_phnum; i++) {
1708 struct elf_phdr *eppnt = phdr + i;
1709 if (eppnt->p_type == PT_LOAD) {
1710 abi_ulong vaddr, vaddr_po, vaddr_ps, vaddr_ef, vaddr_em;
1711 int elf_prot = 0;
1712
1713 if (eppnt->p_flags & PF_R) elf_prot = PROT_READ;
1714 if (eppnt->p_flags & PF_W) elf_prot |= PROT_WRITE;
1715 if (eppnt->p_flags & PF_X) elf_prot |= PROT_EXEC;
1716
1717 vaddr = load_bias + eppnt->p_vaddr;
1718 vaddr_po = TARGET_ELF_PAGEOFFSET(vaddr);
1719 vaddr_ps = TARGET_ELF_PAGESTART(vaddr);
1720
1721 error = target_mmap(vaddr_ps, eppnt->p_filesz + vaddr_po,
1722 elf_prot, MAP_PRIVATE | MAP_FIXED,
1723 image_fd, eppnt->p_offset - vaddr_po);
1724 if (error == -1) {
1725 goto exit_perror;
1726 }
1727
1728 vaddr_ef = vaddr + eppnt->p_filesz;
1729 vaddr_em = vaddr + eppnt->p_memsz;
1730
1731 /* If the load segment requests extra zeros (e.g. bss), map it. */
1732 if (vaddr_ef < vaddr_em) {
1733 zero_bss(vaddr_ef, vaddr_em, elf_prot);
1734 }
1735
1736 /* Find the full program boundaries. */
1737 if (elf_prot & PROT_EXEC) {
1738 if (vaddr < info->start_code) {
1739 info->start_code = vaddr;
1740 }
1741 if (vaddr_ef > info->end_code) {
1742 info->end_code = vaddr_ef;
1743 }
1744 }
1745 if (elf_prot & PROT_WRITE) {
1746 if (vaddr < info->start_data) {
1747 info->start_data = vaddr;
1748 }
1749 if (vaddr_ef > info->end_data) {
1750 info->end_data = vaddr_ef;
1751 }
1752 if (vaddr_em > info->brk) {
1753 info->brk = vaddr_em;
1754 }
1755 }
1756 } else if (eppnt->p_type == PT_INTERP && pinterp_name) {
1757 char *interp_name;
1758
1759 if (*pinterp_name) {
1760 errmsg = "Multiple PT_INTERP entries";
1761 goto exit_errmsg;
1762 }
1763 interp_name = malloc(eppnt->p_filesz);
1764 if (!interp_name) {
1765 goto exit_perror;
1766 }
1767
1768 if (eppnt->p_offset + eppnt->p_filesz <= BPRM_BUF_SIZE) {
1769 memcpy(interp_name, bprm_buf + eppnt->p_offset,
1770 eppnt->p_filesz);
1771 } else {
1772 retval = pread(image_fd, interp_name, eppnt->p_filesz,
1773 eppnt->p_offset);
1774 if (retval != eppnt->p_filesz) {
1775 goto exit_perror;
1776 }
1777 }
1778 if (interp_name[eppnt->p_filesz - 1] != 0) {
1779 errmsg = "Invalid PT_INTERP entry";
1780 goto exit_errmsg;
1781 }
1782 *pinterp_name = interp_name;
1783 }
1784 }
1785
1786 if (info->end_data == 0) {
1787 info->start_data = info->end_code;
1788 info->end_data = info->end_code;
1789 info->brk = info->end_code;
1790 }
1791
1792 if (qemu_log_enabled()) {
1793 load_symbols(ehdr, image_fd, load_bias);
1794 }
1795
1796 close(image_fd);
1797 return;
1798
1799 exit_read:
1800 if (retval >= 0) {
1801 errmsg = "Incomplete read of file header";
1802 goto exit_errmsg;
1803 }
1804 exit_perror:
1805 errmsg = strerror(errno);
1806 exit_errmsg:
1807 fprintf(stderr, "%s: %s\n", image_name, errmsg);
1808 exit(-1);
1809 }
1810
1811 static void load_elf_interp(const char *filename, struct image_info *info,
1812 char bprm_buf[BPRM_BUF_SIZE])
1813 {
1814 int fd, retval;
1815
1816 fd = open(path(filename), O_RDONLY);
1817 if (fd < 0) {
1818 goto exit_perror;
1819 }
1820
1821 retval = read(fd, bprm_buf, BPRM_BUF_SIZE);
1822 if (retval < 0) {
1823 goto exit_perror;
1824 }
1825 if (retval < BPRM_BUF_SIZE) {
1826 memset(bprm_buf + retval, 0, BPRM_BUF_SIZE - retval);
1827 }
1828
1829 load_elf_image(filename, fd, info, NULL, bprm_buf);
1830 return;
1831
1832 exit_perror:
1833 fprintf(stderr, "%s: %s\n", filename, strerror(errno));
1834 exit(-1);
1835 }
1836
1837 static int symfind(const void *s0, const void *s1)
1838 {
1839 target_ulong addr = *(target_ulong *)s0;
1840 struct elf_sym *sym = (struct elf_sym *)s1;
1841 int result = 0;
1842 if (addr < sym->st_value) {
1843 result = -1;
1844 } else if (addr >= sym->st_value + sym->st_size) {
1845 result = 1;
1846 }
1847 return result;
1848 }
1849
1850 static const char *lookup_symbolxx(struct syminfo *s, target_ulong orig_addr)
1851 {
1852 #if ELF_CLASS == ELFCLASS32
1853 struct elf_sym *syms = s->disas_symtab.elf32;
1854 #else
1855 struct elf_sym *syms = s->disas_symtab.elf64;
1856 #endif
1857
1858 // binary search
1859 struct elf_sym *sym;
1860
1861 sym = bsearch(&orig_addr, syms, s->disas_num_syms, sizeof(*syms), symfind);
1862 if (sym != NULL) {
1863 return s->disas_strtab + sym->st_name;
1864 }
1865
1866 return "";
1867 }
1868
1869 /* FIXME: This should use elf_ops.h */
1870 static int symcmp(const void *s0, const void *s1)
1871 {
1872 struct elf_sym *sym0 = (struct elf_sym *)s0;
1873 struct elf_sym *sym1 = (struct elf_sym *)s1;
1874 return (sym0->st_value < sym1->st_value)
1875 ? -1
1876 : ((sym0->st_value > sym1->st_value) ? 1 : 0);
1877 }
1878
1879 /* Best attempt to load symbols from this ELF object. */
1880 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias)
1881 {
1882 int i, shnum, nsyms, sym_idx = 0, str_idx = 0;
1883 struct elf_shdr *shdr;
1884 char *strings = NULL;
1885 struct syminfo *s = NULL;
1886 struct elf_sym *new_syms, *syms = NULL;
1887
1888 shnum = hdr->e_shnum;
1889 i = shnum * sizeof(struct elf_shdr);
1890 shdr = (struct elf_shdr *)alloca(i);
1891 if (pread(fd, shdr, i, hdr->e_shoff) != i) {
1892 return;
1893 }
1894
1895 bswap_shdr(shdr, shnum);
1896 for (i = 0; i < shnum; ++i) {
1897 if (shdr[i].sh_type == SHT_SYMTAB) {
1898 sym_idx = i;
1899 str_idx = shdr[i].sh_link;
1900 goto found;
1901 }
1902 }
1903
1904 /* There will be no symbol table if the file was stripped. */
1905 return;
1906
1907 found:
1908 /* Now know where the strtab and symtab are. Snarf them. */
1909 s = malloc(sizeof(*s));
1910 if (!s) {
1911 goto give_up;
1912 }
1913
1914 i = shdr[str_idx].sh_size;
1915 s->disas_strtab = strings = malloc(i);
1916 if (!strings || pread(fd, strings, i, shdr[str_idx].sh_offset) != i) {
1917 goto give_up;
1918 }
1919
1920 i = shdr[sym_idx].sh_size;
1921 syms = malloc(i);
1922 if (!syms || pread(fd, syms, i, shdr[sym_idx].sh_offset) != i) {
1923 goto give_up;
1924 }
1925
1926 nsyms = i / sizeof(struct elf_sym);
1927 for (i = 0; i < nsyms; ) {
1928 bswap_sym(syms + i);
1929 /* Throw away entries which we do not need. */
1930 if (syms[i].st_shndx == SHN_UNDEF
1931 || syms[i].st_shndx >= SHN_LORESERVE
1932 || ELF_ST_TYPE(syms[i].st_info) != STT_FUNC) {
1933 if (i < --nsyms) {
1934 syms[i] = syms[nsyms];
1935 }
1936 } else {
1937 #if defined(TARGET_ARM) || defined (TARGET_MIPS)
1938 /* The bottom address bit marks a Thumb or MIPS16 symbol. */
1939 syms[i].st_value &= ~(target_ulong)1;
1940 #endif
1941 syms[i].st_value += load_bias;
1942 i++;
1943 }
1944 }
1945
1946 /* No "useful" symbol. */
1947 if (nsyms == 0) {
1948 goto give_up;
1949 }
1950
1951 /* Attempt to free the storage associated with the local symbols
1952 that we threw away. Whether or not this has any effect on the
1953 memory allocation depends on the malloc implementation and how
1954 many symbols we managed to discard. */
1955 new_syms = realloc(syms, nsyms * sizeof(*syms));
1956 if (new_syms == NULL) {
1957 goto give_up;
1958 }
1959 syms = new_syms;
1960
1961 qsort(syms, nsyms, sizeof(*syms), symcmp);
1962
1963 s->disas_num_syms = nsyms;
1964 #if ELF_CLASS == ELFCLASS32
1965 s->disas_symtab.elf32 = syms;
1966 #else
1967 s->disas_symtab.elf64 = syms;
1968 #endif
1969 s->lookup_symbol = lookup_symbolxx;
1970 s->next = syminfos;
1971 syminfos = s;
1972
1973 return;
1974
1975 give_up:
1976 free(s);
1977 free(strings);
1978 free(syms);
1979 }
1980
1981 int load_elf_binary(struct linux_binprm * bprm, struct target_pt_regs * regs,
1982 struct image_info * info)
1983 {
1984 struct image_info interp_info;
1985 struct elfhdr elf_ex;
1986 char *elf_interpreter = NULL;
1987
1988 info->start_mmap = (abi_ulong)ELF_START_MMAP;
1989 info->mmap = 0;
1990 info->rss = 0;
1991
1992 load_elf_image(bprm->filename, bprm->fd, info,
1993 &elf_interpreter, bprm->buf);
1994
1995 /* ??? We need a copy of the elf header for passing to create_elf_tables.
1996 If we do nothing, we'll have overwritten this when we re-use bprm->buf
1997 when we load the interpreter. */
1998 elf_ex = *(struct elfhdr *)bprm->buf;
1999
2000 bprm->p = copy_elf_strings(1, &bprm->filename, bprm->page, bprm->p);
2001 bprm->p = copy_elf_strings(bprm->envc,bprm->envp,bprm->page,bprm->p);
2002 bprm->p = copy_elf_strings(bprm->argc,bprm->argv,bprm->page,bprm->p);
2003 if (!bprm->p) {
2004 fprintf(stderr, "%s: %s\n", bprm->filename, strerror(E2BIG));
2005 exit(-1);
2006 }
2007
2008 /* Do this so that we can load the interpreter, if need be. We will
2009 change some of these later */
2010 bprm->p = setup_arg_pages(bprm->p, bprm, info);
2011
2012 if (elf_interpreter) {
2013 load_elf_interp(elf_interpreter, &interp_info, bprm->buf);
2014
2015 /* If the program interpreter is one of these two, then assume
2016 an iBCS2 image. Otherwise assume a native linux image. */
2017
2018 if (strcmp(elf_interpreter, "/usr/lib/libc.so.1") == 0
2019 || strcmp(elf_interpreter, "/usr/lib/ld.so.1") == 0) {
2020 info->personality = PER_SVR4;
2021
2022 /* Why this, you ask??? Well SVr4 maps page 0 as read-only,
2023 and some applications "depend" upon this behavior. Since
2024 we do not have the power to recompile these, we emulate
2025 the SVr4 behavior. Sigh. */
2026 target_mmap(0, qemu_host_page_size, PROT_READ | PROT_EXEC,
2027 MAP_FIXED | MAP_PRIVATE, -1, 0);
2028 }
2029 }
2030
2031 bprm->p = create_elf_tables(bprm->p, bprm->argc, bprm->envc, &elf_ex,
2032 info, (elf_interpreter ? &interp_info : NULL));
2033 info->start_stack = bprm->p;
2034
2035 /* If we have an interpreter, set that as the program's entry point.
2036 Copy the load_bias as well, to help PPC64 interpret the entry
2037 point as a function descriptor. Do this after creating elf tables
2038 so that we copy the original program entry point into the AUXV. */
2039 if (elf_interpreter) {
2040 info->load_bias = interp_info.load_bias;
2041 info->entry = interp_info.entry;
2042 free(elf_interpreter);
2043 }
2044
2045 #ifdef USE_ELF_CORE_DUMP
2046 bprm->core_dump = &elf_core_dump;
2047 #endif
2048
2049 return 0;
2050 }
2051
2052 #ifdef USE_ELF_CORE_DUMP
2053 /*
2054 * Definitions to generate Intel SVR4-like core files.
2055 * These mostly have the same names as the SVR4 types with "target_elf_"
2056 * tacked on the front to prevent clashes with linux definitions,
2057 * and the typedef forms have been avoided. This is mostly like
2058 * the SVR4 structure, but more Linuxy, with things that Linux does
2059 * not support and which gdb doesn't really use excluded.
2060 *
2061 * Fields we don't dump (their contents is zero) in linux-user qemu
2062 * are marked with XXX.
2063 *
2064 * Core dump code is copied from linux kernel (fs/binfmt_elf.c).
2065 *
2066 * Porting ELF coredump for target is (quite) simple process. First you
2067 * define USE_ELF_CORE_DUMP in target ELF code (where init_thread() for
2068 * the target resides):
2069 *
2070 * #define USE_ELF_CORE_DUMP
2071 *
2072 * Next you define type of register set used for dumping. ELF specification
2073 * says that it needs to be array of elf_greg_t that has size of ELF_NREG.
2074 *
2075 * typedef <target_regtype> target_elf_greg_t;
2076 * #define ELF_NREG <number of registers>
2077 * typedef taret_elf_greg_t target_elf_gregset_t[ELF_NREG];
2078 *
2079 * Last step is to implement target specific function that copies registers
2080 * from given cpu into just specified register set. Prototype is:
2081 *
2082 * static void elf_core_copy_regs(taret_elf_gregset_t *regs,
2083 * const CPUArchState *env);
2084 *
2085 * Parameters:
2086 * regs - copy register values into here (allocated and zeroed by caller)
2087 * env - copy registers from here
2088 *
2089 * Example for ARM target is provided in this file.
2090 */
2091
2092 /* An ELF note in memory */
2093 struct memelfnote {
2094 const char *name;
2095 size_t namesz;
2096 size_t namesz_rounded;
2097 int type;
2098 size_t datasz;
2099 size_t datasz_rounded;
2100 void *data;
2101 size_t notesz;
2102 };
2103
2104 struct target_elf_siginfo {
2105 target_int si_signo; /* signal number */
2106 target_int si_code; /* extra code */
2107 target_int si_errno; /* errno */
2108 };
2109
2110 struct target_elf_prstatus {
2111 struct target_elf_siginfo pr_info; /* Info associated with signal */
2112 target_short pr_cursig; /* Current signal */
2113 target_ulong pr_sigpend; /* XXX */
2114 target_ulong pr_sighold; /* XXX */
2115 target_pid_t pr_pid;
2116 target_pid_t pr_ppid;
2117 target_pid_t pr_pgrp;
2118 target_pid_t pr_sid;
2119 struct target_timeval pr_utime; /* XXX User time */
2120 struct target_timeval pr_stime; /* XXX System time */
2121 struct target_timeval pr_cutime; /* XXX Cumulative user time */
2122 struct target_timeval pr_cstime; /* XXX Cumulative system time */
2123 target_elf_gregset_t pr_reg; /* GP registers */
2124 target_int pr_fpvalid; /* XXX */
2125 };
2126
2127 #define ELF_PRARGSZ (80) /* Number of chars for args */
2128
2129 struct target_elf_prpsinfo {
2130 char pr_state; /* numeric process state */
2131 char pr_sname; /* char for pr_state */
2132 char pr_zomb; /* zombie */
2133 char pr_nice; /* nice val */
2134 target_ulong pr_flag; /* flags */
2135 target_uid_t pr_uid;
2136 target_gid_t pr_gid;
2137 target_pid_t pr_pid, pr_ppid, pr_pgrp, pr_sid;
2138 /* Lots missing */
2139 char pr_fname[16]; /* filename of executable */
2140 char pr_psargs[ELF_PRARGSZ]; /* initial part of arg list */
2141 };
2142
2143 /* Here is the structure in which status of each thread is captured. */
2144 struct elf_thread_status {
2145 QTAILQ_ENTRY(elf_thread_status) ets_link;
2146 struct target_elf_prstatus prstatus; /* NT_PRSTATUS */
2147 #if 0
2148 elf_fpregset_t fpu; /* NT_PRFPREG */
2149 struct task_struct *thread;
2150 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */
2151 #endif
2152 struct memelfnote notes[1];
2153 int num_notes;
2154 };
2155
2156 struct elf_note_info {
2157 struct memelfnote *notes;
2158 struct target_elf_prstatus *prstatus; /* NT_PRSTATUS */
2159 struct target_elf_prpsinfo *psinfo; /* NT_PRPSINFO */
2160
2161 QTAILQ_HEAD(thread_list_head, elf_thread_status) thread_list;
2162 #if 0
2163 /*
2164 * Current version of ELF coredump doesn't support
2165 * dumping fp regs etc.
2166 */
2167 elf_fpregset_t *fpu;
2168 elf_fpxregset_t *xfpu;
2169 int thread_status_size;
2170 #endif
2171 int notes_size;
2172 int numnote;
2173 };
2174
2175 struct vm_area_struct {
2176 abi_ulong vma_start; /* start vaddr of memory region */
2177 abi_ulong vma_end; /* end vaddr of memory region */
2178 abi_ulong vma_flags; /* protection etc. flags for the region */
2179 QTAILQ_ENTRY(vm_area_struct) vma_link;
2180 };
2181
2182 struct mm_struct {
2183 QTAILQ_HEAD(, vm_area_struct) mm_mmap;
2184 int mm_count; /* number of mappings */
2185 };
2186
2187 static struct mm_struct *vma_init(void);
2188 static void vma_delete(struct mm_struct *);
2189 static int vma_add_mapping(struct mm_struct *, abi_ulong,
2190 abi_ulong, abi_ulong);
2191 static int vma_get_mapping_count(const struct mm_struct *);
2192 static struct vm_area_struct *vma_first(const struct mm_struct *);
2193 static struct vm_area_struct *vma_next(struct vm_area_struct *);
2194 static abi_ulong vma_dump_size(const struct vm_area_struct *);
2195 static int vma_walker(void *priv, abi_ulong start, abi_ulong end,
2196 unsigned long flags);
2197
2198 static void fill_elf_header(struct elfhdr *, int, uint16_t, uint32_t);
2199 static void fill_note(struct memelfnote *, const char *, int,
2200 unsigned int, void *);
2201 static void fill_prstatus(struct target_elf_prstatus *, const TaskState *, int);
2202 static int fill_psinfo(struct target_elf_prpsinfo *, const TaskState *);
2203 static void fill_auxv_note(struct memelfnote *, const TaskState *);
2204 static void fill_elf_note_phdr(struct elf_phdr *, int, off_t);
2205 static size_t note_size(const struct memelfnote *);
2206 static void free_note_info(struct elf_note_info *);
2207 static int fill_note_info(struct elf_note_info *, long, const CPUArchState *);
2208 static void fill_thread_info(struct elf_note_info *, const CPUArchState *);
2209 static int core_dump_filename(const TaskState *, char *, size_t);
2210
2211 static int dump_write(int, const void *, size_t);
2212 static int write_note(struct memelfnote *, int);
2213 static int write_note_info(struct elf_note_info *, int);
2214
2215 #ifdef BSWAP_NEEDED
2216 static void bswap_prstatus(struct target_elf_prstatus *prstatus)
2217 {
2218 prstatus->pr_info.si_signo = tswapl(prstatus->pr_info.si_signo);
2219 prstatus->pr_info.si_code = tswapl(prstatus->pr_info.si_code);
2220 prstatus->pr_info.si_errno = tswapl(prstatus->pr_info.si_errno);
2221 prstatus->pr_cursig = tswap16(prstatus->pr_cursig);
2222 prstatus->pr_sigpend = tswapl(prstatus->pr_sigpend);
2223 prstatus->pr_sighold = tswapl(prstatus->pr_sighold);
2224 prstatus->pr_pid = tswap32(prstatus->pr_pid);
2225 prstatus->pr_ppid = tswap32(prstatus->pr_ppid);
2226 prstatus->pr_pgrp = tswap32(prstatus->pr_pgrp);
2227 prstatus->pr_sid = tswap32(prstatus->pr_sid);
2228 /* cpu times are not filled, so we skip them */
2229 /* regs should be in correct format already */
2230 prstatus->pr_fpvalid = tswap32(prstatus->pr_fpvalid);
2231 }
2232
2233 static void bswap_psinfo(struct target_elf_prpsinfo *psinfo)
2234 {
2235 psinfo->pr_flag = tswapl(psinfo->pr_flag);
2236 psinfo->pr_uid = tswap16(psinfo->pr_uid);
2237 psinfo->pr_gid = tswap16(psinfo->pr_gid);
2238 psinfo->pr_pid = tswap32(psinfo->pr_pid);
2239 psinfo->pr_ppid = tswap32(psinfo->pr_ppid);
2240 psinfo->pr_pgrp = tswap32(psinfo->pr_pgrp);
2241 psinfo->pr_sid = tswap32(psinfo->pr_sid);
2242 }
2243
2244 static void bswap_note(struct elf_note *en)
2245 {
2246 bswap32s(&en->n_namesz);
2247 bswap32s(&en->n_descsz);
2248 bswap32s(&en->n_type);
2249 }
2250 #else
2251 static inline void bswap_prstatus(struct target_elf_prstatus *p) { }
2252 static inline void bswap_psinfo(struct target_elf_prpsinfo *p) {}
2253 static inline void bswap_note(struct elf_note *en) { }
2254 #endif /* BSWAP_NEEDED */
2255
2256 /*
2257 * Minimal support for linux memory regions. These are needed
2258 * when we are finding out what memory exactly belongs to
2259 * emulated process. No locks needed here, as long as
2260 * thread that received the signal is stopped.
2261 */
2262
2263 static struct mm_struct *vma_init(void)
2264 {
2265 struct mm_struct *mm;
2266
2267 if ((mm = g_malloc(sizeof (*mm))) == NULL)
2268 return (NULL);
2269
2270 mm->mm_count = 0;
2271 QTAILQ_INIT(&mm->mm_mmap);
2272
2273 return (mm);
2274 }
2275
2276 static void vma_delete(struct mm_struct *mm)
2277 {
2278 struct vm_area_struct *vma;
2279
2280 while ((vma = vma_first(mm)) != NULL) {
2281 QTAILQ_REMOVE(&mm->mm_mmap, vma, vma_link);
2282 g_free(vma);
2283 }
2284 g_free(mm);
2285 }
2286
2287 static int vma_add_mapping(struct mm_struct *mm, abi_ulong start,
2288 abi_ulong end, abi_ulong flags)
2289 {
2290 struct vm_area_struct *vma;
2291
2292 if ((vma = g_malloc0(sizeof (*vma))) == NULL)
2293 return (-1);
2294
2295 vma->vma_start = start;
2296 vma->vma_end = end;
2297 vma->vma_flags = flags;
2298
2299 QTAILQ_INSERT_TAIL(&mm->mm_mmap, vma, vma_link);
2300 mm->mm_count++;
2301
2302 return (0);
2303 }
2304
2305 static struct vm_area_struct *vma_first(const struct mm_struct *mm)
2306 {
2307 return (QTAILQ_FIRST(&mm->mm_mmap));
2308 }
2309
2310 static struct vm_area_struct *vma_next(struct vm_area_struct *vma)
2311 {
2312 return (QTAILQ_NEXT(vma, vma_link));
2313 }
2314
2315 static int vma_get_mapping_count(const struct mm_struct *mm)
2316 {
2317 return (mm->mm_count);
2318 }
2319
2320 /*
2321 * Calculate file (dump) size of given memory region.
2322 */
2323 static abi_ulong vma_dump_size(const struct vm_area_struct *vma)
2324 {
2325 /* if we cannot even read the first page, skip it */
2326 if (!access_ok(VERIFY_READ, vma->vma_start, TARGET_PAGE_SIZE))
2327 return (0);
2328
2329 /*
2330 * Usually we don't dump executable pages as they contain
2331 * non-writable code that debugger can read directly from
2332 * target library etc. However, thread stacks are marked
2333 * also executable so we read in first page of given region
2334 * and check whether it contains elf header. If there is
2335 * no elf header, we dump it.
2336 */
2337 if (vma->vma_flags & PROT_EXEC) {
2338 char page[TARGET_PAGE_SIZE];
2339
2340 copy_from_user(page, vma->vma_start, sizeof (page));
2341 if ((page[EI_MAG0] == ELFMAG0) &&
2342 (page[EI_MAG1] == ELFMAG1) &&
2343 (page[EI_MAG2] == ELFMAG2) &&
2344 (page[EI_MAG3] == ELFMAG3)) {
2345 /*
2346 * Mappings are possibly from ELF binary. Don't dump
2347 * them.
2348 */
2349 return (0);
2350 }
2351 }
2352
2353 return (vma->vma_end - vma->vma_start);
2354 }
2355
2356 static int vma_walker(void *priv, abi_ulong start, abi_ulong end,
2357 unsigned long flags)
2358 {
2359 struct mm_struct *mm = (struct mm_struct *)priv;
2360
2361 vma_add_mapping(mm, start, end, flags);
2362 return (0);
2363 }
2364
2365 static void fill_note(struct memelfnote *note, const char *name, int type,
2366 unsigned int sz, void *data)
2367 {
2368 unsigned int namesz;
2369
2370 namesz = strlen(name) + 1;
2371 note->name = name;
2372 note->namesz = namesz;
2373 note->namesz_rounded = roundup(namesz, sizeof (int32_t));
2374 note->type = type;
2375 note->datasz = sz;
2376 note->datasz_rounded = roundup(sz, sizeof (int32_t));
2377
2378 note->data = data;
2379
2380 /*
2381 * We calculate rounded up note size here as specified by
2382 * ELF document.
2383 */
2384 note->notesz = sizeof (struct elf_note) +
2385 note->namesz_rounded + note->datasz_rounded;
2386 }
2387
2388 static void fill_elf_header(struct elfhdr *elf, int segs, uint16_t machine,
2389 uint32_t flags)
2390 {
2391 (void) memset(elf, 0, sizeof(*elf));
2392
2393 (void) memcpy(elf->e_ident, ELFMAG, SELFMAG);
2394 elf->e_ident[EI_CLASS] = ELF_CLASS;
2395 elf->e_ident[EI_DATA] = ELF_DATA;
2396 elf->e_ident[EI_VERSION] = EV_CURRENT;
2397 elf->e_ident[EI_OSABI] = ELF_OSABI;
2398
2399 elf->e_type = ET_CORE;
2400 elf->e_machine = machine;
2401 elf->e_version = EV_CURRENT;
2402 elf->e_phoff = sizeof(struct elfhdr);
2403 elf->e_flags = flags;
2404 elf->e_ehsize = sizeof(struct elfhdr);
2405 elf->e_phentsize = sizeof(struct elf_phdr);
2406 elf->e_phnum = segs;
2407
2408 bswap_ehdr(elf);
2409 }
2410
2411 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, off_t offset)
2412 {
2413 phdr->p_type = PT_NOTE;
2414 phdr->p_offset = offset;
2415 phdr->p_vaddr = 0;
2416 phdr->p_paddr = 0;
2417 phdr->p_filesz = sz;
2418 phdr->p_memsz = 0;
2419 phdr->p_flags = 0;
2420 phdr->p_align = 0;
2421
2422 bswap_phdr(phdr, 1);
2423 }
2424
2425 static size_t note_size(const struct memelfnote *note)
2426 {
2427 return (note->notesz);
2428 }
2429
2430 static void fill_prstatus(struct target_elf_prstatus *prstatus,
2431 const TaskState *ts, int signr)
2432 {
2433 (void) memset(prstatus, 0, sizeof (*prstatus));
2434 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
2435 prstatus->pr_pid = ts->ts_tid;
2436 prstatus->pr_ppid = getppid();
2437 prstatus->pr_pgrp = getpgrp();
2438 prstatus->pr_sid = getsid(0);
2439
2440 bswap_prstatus(prstatus);
2441 }
2442
2443 static int fill_psinfo(struct target_elf_prpsinfo *psinfo, const TaskState *ts)
2444 {
2445 char *filename, *base_filename;
2446 unsigned int i, len;
2447
2448 (void) memset(psinfo, 0, sizeof (*psinfo));
2449
2450 len = ts->info->arg_end - ts->info->arg_start;
2451 if (len >= ELF_PRARGSZ)
2452 len = ELF_PRARGSZ - 1;
2453 if (copy_from_user(&psinfo->pr_psargs, ts->info->arg_start, len))
2454 return -EFAULT;
2455 for (i = 0; i < len; i++)
2456 if (psinfo->pr_psargs[i] == 0)
2457 psinfo->pr_psargs[i] = ' ';
2458 psinfo->pr_psargs[len] = 0;
2459
2460 psinfo->pr_pid = getpid();
2461 psinfo->pr_ppid = getppid();
2462 psinfo->pr_pgrp = getpgrp();
2463 psinfo->pr_sid = getsid(0);
2464 psinfo->pr_uid = getuid();
2465 psinfo->pr_gid = getgid();
2466
2467 filename = strdup(ts->bprm->filename);
2468 base_filename = strdup(basename(filename));
2469 (void) strncpy(psinfo->pr_fname, base_filename,
2470 sizeof(psinfo->pr_fname));
2471 free(base_filename);
2472 free(filename);
2473
2474 bswap_psinfo(psinfo);
2475 return (0);
2476 }
2477
2478 static void fill_auxv_note(struct memelfnote *note, const TaskState *ts)
2479 {
2480 elf_addr_t auxv = (elf_addr_t)ts->info->saved_auxv;
2481 elf_addr_t orig_auxv = auxv;
2482 void *ptr;
2483 int len = ts->info->auxv_len;
2484
2485 /*
2486 * Auxiliary vector is stored in target process stack. It contains
2487 * {type, value} pairs that we need to dump into note. This is not
2488 * strictly necessary but we do it here for sake of completeness.
2489 */
2490
2491 /* read in whole auxv vector and copy it to memelfnote */
2492 ptr = lock_user(VERIFY_READ, orig_auxv, len, 0);
2493 if (ptr != NULL) {
2494 fill_note(note, "CORE", NT_AUXV, len, ptr);
2495 unlock_user(ptr, auxv, len);
2496 }
2497 }
2498
2499 /*
2500 * Constructs name of coredump file. We have following convention
2501 * for the name:
2502 * qemu_<basename-of-target-binary>_<date>-<time>_<pid>.core
2503 *
2504 * Returns 0 in case of success, -1 otherwise (errno is set).
2505 */
2506 static int core_dump_filename(const TaskState *ts, char *buf,
2507 size_t bufsize)
2508 {
2509 char timestamp[64];
2510 char *filename = NULL;
2511 char *base_filename = NULL;
2512 struct timeval tv;
2513 struct tm tm;
2514
2515 assert(bufsize >= PATH_MAX);
2516
2517 if (gettimeofday(&tv, NULL) < 0) {
2518 (void) fprintf(stderr, "unable to get current timestamp: %s",
2519 strerror(errno));
2520 return (-1);
2521 }
2522
2523 filename = strdup(ts->bprm->filename);
2524 base_filename = strdup(basename(filename));
2525 (void) strftime(timestamp, sizeof (timestamp), "%Y%m%d-%H%M%S",
2526 localtime_r(&tv.tv_sec, &tm));
2527 (void) snprintf(buf, bufsize, "qemu_%s_%s_%d.core",
2528 base_filename, timestamp, (int)getpid());
2529 free(base_filename);
2530 free(filename);
2531
2532 return (0);
2533 }
2534
2535 static int dump_write(int fd, const void *ptr, size_t size)
2536 {
2537 const char *bufp = (const char *)ptr;
2538 ssize_t bytes_written, bytes_left;
2539 struct rlimit dumpsize;
2540 off_t pos;
2541
2542 bytes_written = 0;
2543 getrlimit(RLIMIT_CORE, &dumpsize);
2544 if ((pos = lseek(fd, 0, SEEK_CUR))==-1) {
2545 if (errno == ESPIPE) { /* not a seekable stream */
2546 bytes_left = size;
2547 } else {
2548 return pos;
2549 }
2550 } else {
2551 if (dumpsize.rlim_cur <= pos) {
2552 return -1;
2553 } else if (dumpsize.rlim_cur == RLIM_INFINITY) {
2554 bytes_left = size;
2555 } else {
2556 size_t limit_left=dumpsize.rlim_cur - pos;
2557 bytes_left = limit_left >= size ? size : limit_left ;
2558 }
2559 }
2560
2561 /*
2562 * In normal conditions, single write(2) should do but
2563 * in case of socket etc. this mechanism is more portable.
2564 */
2565 do {
2566 bytes_written = write(fd, bufp, bytes_left);
2567 if (bytes_written < 0) {
2568 if (errno == EINTR)
2569 continue;
2570 return (-1);
2571 } else if (bytes_written == 0) { /* eof */
2572 return (-1);
2573 }
2574 bufp += bytes_written;
2575 bytes_left -= bytes_written;
2576 } while (bytes_left > 0);
2577
2578 return (0);
2579 }
2580
2581 static int write_note(struct memelfnote *men, int fd)
2582 {
2583 struct elf_note en;
2584
2585 en.n_namesz = men->namesz;
2586 en.n_type = men->type;
2587 en.n_descsz = men->datasz;
2588
2589 bswap_note(&en);
2590
2591 if (dump_write(fd, &en, sizeof(en)) != 0)
2592 return (-1);
2593 if (dump_write(fd, men->name, men->namesz_rounded) != 0)
2594 return (-1);
2595 if (dump_write(fd, men->data, men->datasz_rounded) != 0)
2596 return (-1);
2597
2598 return (0);
2599 }
2600
2601 static void fill_thread_info(struct elf_note_info *info, const CPUArchState *env)
2602 {
2603 TaskState *ts = (TaskState *)env->opaque;
2604 struct elf_thread_status *ets;
2605
2606 ets = g_malloc0(sizeof (*ets));
2607 ets->num_notes = 1; /* only prstatus is dumped */
2608 fill_prstatus(&ets->prstatus, ts, 0);
2609 elf_core_copy_regs(&ets->prstatus.pr_reg, env);
2610 fill_note(&ets->notes[0], "CORE", NT_PRSTATUS, sizeof (ets->prstatus),
2611 &ets->prstatus);
2612
2613 QTAILQ_INSERT_TAIL(&info->thread_list, ets, ets_link);
2614
2615 info->notes_size += note_size(&ets->notes[0]);
2616 }
2617
2618 static int fill_note_info(struct elf_note_info *info,
2619 long signr, const CPUArchState *env)
2620 {
2621 #define NUMNOTES 3
2622 CPUArchState *cpu = NULL;
2623 TaskState *ts = (TaskState *)env->opaque;
2624 int i;
2625
2626 (void) memset(info, 0, sizeof (*info));
2627
2628 QTAILQ_INIT(&info->thread_list);
2629
2630 info->notes = g_malloc0(NUMNOTES * sizeof (struct memelfnote));
2631 if (info->notes == NULL)
2632 return (-ENOMEM);
2633 info->prstatus = g_malloc0(sizeof (*info->prstatus));
2634 if (info->prstatus == NULL)
2635 return (-ENOMEM);
2636 info->psinfo = g_malloc0(sizeof (*info->psinfo));
2637 if (info->prstatus == NULL)
2638 return (-ENOMEM);
2639
2640 /*
2641 * First fill in status (and registers) of current thread
2642 * including process info & aux vector.
2643 */
2644 fill_prstatus(info->prstatus, ts, signr);
2645 elf_core_copy_regs(&info->prstatus->pr_reg, env);
2646 fill_note(&info->notes[0], "CORE", NT_PRSTATUS,
2647 sizeof (*info->prstatus), info->prstatus);
2648 fill_psinfo(info->psinfo, ts);
2649 fill_note(&info->notes[1], "CORE", NT_PRPSINFO,
2650 sizeof (*info->psinfo), info->psinfo);
2651 fill_auxv_note(&info->notes[2], ts);
2652 info->numnote = 3;
2653
2654 info->notes_size = 0;
2655 for (i = 0; i < info->numnote; i++)
2656 info->notes_size += note_size(&info->notes[i]);
2657
2658 /* read and fill status of all threads */
2659 cpu_list_lock();
2660 for (cpu = first_cpu; cpu != NULL; cpu = cpu->next_cpu) {
2661 if (cpu == thread_env)
2662 continue;
2663 fill_thread_info(info, cpu);
2664 }
2665 cpu_list_unlock();
2666
2667 return (0);
2668 }
2669
2670 static void free_note_info(struct elf_note_info *info)
2671 {
2672 struct elf_thread_status *ets;
2673
2674 while (!QTAILQ_EMPTY(&info->thread_list)) {
2675 ets = QTAILQ_FIRST(&info->thread_list);
2676 QTAILQ_REMOVE(&info->thread_list, ets, ets_link);
2677 g_free(ets);
2678 }
2679
2680 g_free(info->prstatus);
2681 g_free(info->psinfo);
2682 g_free(info->notes);
2683 }
2684
2685 static int write_note_info(struct elf_note_info *info, int fd)
2686 {
2687 struct elf_thread_status *ets;
2688 int i, error = 0;
2689
2690 /* write prstatus, psinfo and auxv for current thread */
2691 for (i = 0; i < info->numnote; i++)
2692 if ((error = write_note(&info->notes[i], fd)) != 0)
2693 return (error);
2694
2695 /* write prstatus for each thread */
2696 for (ets = info->thread_list.tqh_first; ets != NULL;
2697 ets = ets->ets_link.tqe_next) {
2698 if ((error = write_note(&ets->notes[0], fd)) != 0)
2699 return (error);
2700 }
2701
2702 return (0);
2703 }
2704
2705 /*
2706 * Write out ELF coredump.
2707 *
2708 * See documentation of ELF object file format in:
2709 * http://www.caldera.com/developers/devspecs/gabi41.pdf
2710 *
2711 * Coredump format in linux is following:
2712 *
2713 * 0 +----------------------+ \
2714 * | ELF header | ET_CORE |
2715 * +----------------------+ |
2716 * | ELF program headers | |--- headers
2717 * | - NOTE section | |
2718 * | - PT_LOAD sections | |
2719 * +----------------------+ /
2720 * | NOTEs: |
2721 * | - NT_PRSTATUS |
2722 * | - NT_PRSINFO |
2723 * | - NT_AUXV |
2724 * +----------------------+ <-- aligned to target page
2725 * | Process memory dump |
2726 * : :
2727 * . .
2728 * : :
2729 * | |
2730 * +----------------------+
2731 *
2732 * NT_PRSTATUS -> struct elf_prstatus (per thread)
2733 * NT_PRSINFO -> struct elf_prpsinfo
2734 * NT_AUXV is array of { type, value } pairs (see fill_auxv_note()).
2735 *
2736 * Format follows System V format as close as possible. Current
2737 * version limitations are as follows:
2738 * - no floating point registers are dumped
2739 *
2740 * Function returns 0 in case of success, negative errno otherwise.
2741 *
2742 * TODO: make this work also during runtime: it should be
2743 * possible to force coredump from running process and then
2744 * continue processing. For example qemu could set up SIGUSR2
2745 * handler (provided that target process haven't registered
2746 * handler for that) that does the dump when signal is received.
2747 */
2748 static int elf_core_dump(int signr, const CPUArchState *env)
2749 {
2750 const TaskState *ts = (const TaskState *)env->opaque;
2751 struct vm_area_struct *vma = NULL;
2752 char corefile[PATH_MAX];
2753 struct elf_note_info info;
2754 struct elfhdr elf;
2755 struct elf_phdr phdr;
2756 struct rlimit dumpsize;
2757 struct mm_struct *mm = NULL;
2758 off_t offset = 0, data_offset = 0;
2759 int segs = 0;
2760 int fd = -1;
2761
2762 errno = 0;
2763 getrlimit(RLIMIT_CORE, &dumpsize);
2764 if (dumpsize.rlim_cur == 0)
2765 return 0;
2766
2767 if (core_dump_filename(ts, corefile, sizeof (corefile)) < 0)
2768 return (-errno);
2769
2770 if ((fd = open(corefile, O_WRONLY | O_CREAT,
2771 S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH)) < 0)
2772 return (-errno);
2773
2774 /*
2775 * Walk through target process memory mappings and
2776 * set up structure containing this information. After
2777 * this point vma_xxx functions can be used.
2778 */
2779 if ((mm = vma_init()) == NULL)
2780 goto out;
2781
2782 walk_memory_regions(mm, vma_walker);
2783 segs = vma_get_mapping_count(mm);
2784
2785 /*
2786 * Construct valid coredump ELF header. We also
2787 * add one more segment for notes.
2788 */
2789 fill_elf_header(&elf, segs + 1, ELF_MACHINE, 0);
2790 if (dump_write(fd, &elf, sizeof (elf)) != 0)
2791 goto out;
2792
2793 /* fill in in-memory version of notes */
2794 if (fill_note_info(&info, signr, env) < 0)
2795 goto out;
2796
2797 offset += sizeof (elf); /* elf header */
2798 offset += (segs + 1) * sizeof (struct elf_phdr); /* program headers */
2799
2800 /* write out notes program header */
2801 fill_elf_note_phdr(&phdr, info.notes_size, offset);
2802
2803 offset += info.notes_size;
2804 if (dump_write(fd, &phdr, sizeof (phdr)) != 0)
2805 goto out;
2806
2807 /*
2808 * ELF specification wants data to start at page boundary so
2809 * we align it here.
2810 */
2811 data_offset = offset = roundup(offset, ELF_EXEC_PAGESIZE);
2812
2813 /*
2814 * Write program headers for memory regions mapped in
2815 * the target process.
2816 */
2817 for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) {
2818 (void) memset(&phdr, 0, sizeof (phdr));
2819
2820 phdr.p_type = PT_LOAD;
2821 phdr.p_offset = offset;
2822 phdr.p_vaddr = vma->vma_start;
2823 phdr.p_paddr = 0;
2824 phdr.p_filesz = vma_dump_size(vma);
2825 offset += phdr.p_filesz;
2826 phdr.p_memsz = vma->vma_end - vma->vma_start;
2827 phdr.p_flags = vma->vma_flags & PROT_READ ? PF_R : 0;
2828 if (vma->vma_flags & PROT_WRITE)
2829 phdr.p_flags |= PF_W;
2830 if (vma->vma_flags & PROT_EXEC)
2831 phdr.p_flags |= PF_X;
2832 phdr.p_align = ELF_EXEC_PAGESIZE;
2833
2834 bswap_phdr(&phdr, 1);
2835 dump_write(fd, &phdr, sizeof (phdr));
2836 }
2837
2838 /*
2839 * Next we write notes just after program headers. No
2840 * alignment needed here.
2841 */
2842 if (write_note_info(&info, fd) < 0)
2843 goto out;
2844
2845 /* align data to page boundary */
2846 if (lseek(fd, data_offset, SEEK_SET) != data_offset)
2847 goto out;
2848
2849 /*
2850 * Finally we can dump process memory into corefile as well.
2851 */
2852 for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) {
2853 abi_ulong addr;
2854 abi_ulong end;
2855
2856 end = vma->vma_start + vma_dump_size(vma);
2857
2858 for (addr = vma->vma_start; addr < end;
2859 addr += TARGET_PAGE_SIZE) {
2860 char page[TARGET_PAGE_SIZE];
2861 int error;
2862
2863 /*
2864 * Read in page from target process memory and
2865 * write it to coredump file.
2866 */
2867 error = copy_from_user(page, addr, sizeof (page));
2868 if (error != 0) {
2869 (void) fprintf(stderr, "unable to dump " TARGET_ABI_FMT_lx "\n",
2870 addr);
2871 errno = -error;
2872 goto out;
2873 }
2874 if (dump_write(fd, page, TARGET_PAGE_SIZE) < 0)
2875 goto out;
2876 }
2877 }
2878
2879 out:
2880 free_note_info(&info);
2881 if (mm != NULL)
2882 vma_delete(mm);
2883 (void) close(fd);
2884
2885 if (errno != 0)
2886 return (-errno);
2887 return (0);
2888 }
2889 #endif /* USE_ELF_CORE_DUMP */
2890
2891 void do_init_thread(struct target_pt_regs *regs, struct image_info *infop)
2892 {
2893 init_thread(regs, infop);
2894 }
QEMU modified to work with kvm