xref: /qemu/linux-user/elfload.c (revision d072cdf3)
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/disas.h"
18 
19 #ifdef _ARCH_PPC64
20 #undef ARCH_DLINFO
21 #undef ELF_PLATFORM
22 #undef ELF_HWCAP
23 #undef ELF_HWCAP2
24 #undef ELF_CLASS
25 #undef ELF_DATA
26 #undef ELF_ARCH
27 #endif
28 
29 #define ELF_OSABI   ELFOSABI_SYSV
30 
31 /* from personality.h */
32 
33 /*
34  * Flags for bug emulation.
35  *
36  * These occupy the top three bytes.
37  */
38 enum {
39     ADDR_NO_RANDOMIZE = 0x0040000,      /* disable randomization of VA space */
40     FDPIC_FUNCPTRS =    0x0080000,      /* userspace function ptrs point to
41                                            descriptors (signal handling) */
42     MMAP_PAGE_ZERO =    0x0100000,
43     ADDR_COMPAT_LAYOUT = 0x0200000,
44     READ_IMPLIES_EXEC = 0x0400000,
45     ADDR_LIMIT_32BIT =  0x0800000,
46     SHORT_INODE =       0x1000000,
47     WHOLE_SECONDS =     0x2000000,
48     STICKY_TIMEOUTS =   0x4000000,
49     ADDR_LIMIT_3GB =    0x8000000,
50 };
51 
52 /*
53  * Personality types.
54  *
55  * These go in the low byte.  Avoid using the top bit, it will
56  * conflict with error returns.
57  */
58 enum {
59     PER_LINUX =         0x0000,
60     PER_LINUX_32BIT =   0x0000 | ADDR_LIMIT_32BIT,
61     PER_LINUX_FDPIC =   0x0000 | FDPIC_FUNCPTRS,
62     PER_SVR4 =          0x0001 | STICKY_TIMEOUTS | MMAP_PAGE_ZERO,
63     PER_SVR3 =          0x0002 | STICKY_TIMEOUTS | SHORT_INODE,
64     PER_SCOSVR3 =       0x0003 | STICKY_TIMEOUTS | WHOLE_SECONDS | SHORT_INODE,
65     PER_OSR5 =          0x0003 | STICKY_TIMEOUTS | WHOLE_SECONDS,
66     PER_WYSEV386 =      0x0004 | STICKY_TIMEOUTS | SHORT_INODE,
67     PER_ISCR4 =         0x0005 | STICKY_TIMEOUTS,
68     PER_BSD =           0x0006,
69     PER_SUNOS =         0x0006 | STICKY_TIMEOUTS,
70     PER_XENIX =         0x0007 | STICKY_TIMEOUTS | SHORT_INODE,
71     PER_LINUX32 =       0x0008,
72     PER_LINUX32_3GB =   0x0008 | ADDR_LIMIT_3GB,
73     PER_IRIX32 =        0x0009 | STICKY_TIMEOUTS,/* IRIX5 32-bit */
74     PER_IRIXN32 =       0x000a | STICKY_TIMEOUTS,/* IRIX6 new 32-bit */
75     PER_IRIX64 =        0x000b | STICKY_TIMEOUTS,/* IRIX6 64-bit */
76     PER_RISCOS =        0x000c,
77     PER_SOLARIS =       0x000d | STICKY_TIMEOUTS,
78     PER_UW7 =           0x000e | STICKY_TIMEOUTS | MMAP_PAGE_ZERO,
79     PER_OSF4 =          0x000f,                  /* OSF/1 v4 */
80     PER_HPUX =          0x0010,
81     PER_MASK =          0x00ff,
82 };
83 
84 /*
85  * Return the base personality without flags.
86  */
87 #define personality(pers)       (pers & PER_MASK)
88 
89 /* this flag is uneffective under linux too, should be deleted */
90 #ifndef MAP_DENYWRITE
91 #define MAP_DENYWRITE 0
92 #endif
93 
94 /* should probably go in elf.h */
95 #ifndef ELIBBAD
96 #define ELIBBAD 80
97 #endif
98 
99 #ifdef TARGET_WORDS_BIGENDIAN
100 #define ELF_DATA        ELFDATA2MSB
101 #else
102 #define ELF_DATA        ELFDATA2LSB
103 #endif
104 
105 #ifdef TARGET_ABI_MIPSN32
106 typedef abi_ullong      target_elf_greg_t;
107 #define tswapreg(ptr)   tswap64(ptr)
108 #else
109 typedef abi_ulong       target_elf_greg_t;
110 #define tswapreg(ptr)   tswapal(ptr)
111 #endif
112 
113 #ifdef USE_UID16
114 typedef abi_ushort      target_uid_t;
115 typedef abi_ushort      target_gid_t;
116 #else
117 typedef abi_uint        target_uid_t;
118 typedef abi_uint        target_gid_t;
119 #endif
120 typedef abi_int         target_pid_t;
121 
122 #ifdef TARGET_I386
123 
124 #define ELF_PLATFORM get_elf_platform()
125 
126 static const char *get_elf_platform(void)
127 {
128     static char elf_platform[] = "i386";
129     int family = object_property_get_int(OBJECT(thread_cpu), "family", NULL);
130     if (family > 6)
131         family = 6;
132     if (family >= 3)
133         elf_platform[1] = '0' + family;
134     return elf_platform;
135 }
136 
137 #define ELF_HWCAP get_elf_hwcap()
138 
139 static uint32_t get_elf_hwcap(void)
140 {
141     X86CPU *cpu = X86_CPU(thread_cpu);
142 
143     return cpu->env.features[FEAT_1_EDX];
144 }
145 
146 #ifdef TARGET_X86_64
147 #define ELF_START_MMAP 0x2aaaaab000ULL
148 #define elf_check_arch(x) ( ((x) == ELF_ARCH) )
149 
150 #define ELF_CLASS      ELFCLASS64
151 #define ELF_ARCH       EM_X86_64
152 
153 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop)
154 {
155     regs->rax = 0;
156     regs->rsp = infop->start_stack;
157     regs->rip = infop->entry;
158 }
159 
160 #define ELF_NREG    27
161 typedef target_elf_greg_t  target_elf_gregset_t[ELF_NREG];
162 
163 /*
164  * Note that ELF_NREG should be 29 as there should be place for
165  * TRAPNO and ERR "registers" as well but linux doesn't dump
166  * those.
167  *
168  * See linux kernel: arch/x86/include/asm/elf.h
169  */
170 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUX86State *env)
171 {
172     (*regs)[0] = env->regs[15];
173     (*regs)[1] = env->regs[14];
174     (*regs)[2] = env->regs[13];
175     (*regs)[3] = env->regs[12];
176     (*regs)[4] = env->regs[R_EBP];
177     (*regs)[5] = env->regs[R_EBX];
178     (*regs)[6] = env->regs[11];
179     (*regs)[7] = env->regs[10];
180     (*regs)[8] = env->regs[9];
181     (*regs)[9] = env->regs[8];
182     (*regs)[10] = env->regs[R_EAX];
183     (*regs)[11] = env->regs[R_ECX];
184     (*regs)[12] = env->regs[R_EDX];
185     (*regs)[13] = env->regs[R_ESI];
186     (*regs)[14] = env->regs[R_EDI];
187     (*regs)[15] = env->regs[R_EAX]; /* XXX */
188     (*regs)[16] = env->eip;
189     (*regs)[17] = env->segs[R_CS].selector & 0xffff;
190     (*regs)[18] = env->eflags;
191     (*regs)[19] = env->regs[R_ESP];
192     (*regs)[20] = env->segs[R_SS].selector & 0xffff;
193     (*regs)[21] = env->segs[R_FS].selector & 0xffff;
194     (*regs)[22] = env->segs[R_GS].selector & 0xffff;
195     (*regs)[23] = env->segs[R_DS].selector & 0xffff;
196     (*regs)[24] = env->segs[R_ES].selector & 0xffff;
197     (*regs)[25] = env->segs[R_FS].selector & 0xffff;
198     (*regs)[26] = env->segs[R_GS].selector & 0xffff;
199 }
200 
201 #else
202 
203 #define ELF_START_MMAP 0x80000000
204 
205 /*
206  * This is used to ensure we don't load something for the wrong architecture.
207  */
208 #define elf_check_arch(x) ( ((x) == EM_386) || ((x) == EM_486) )
209 
210 /*
211  * These are used to set parameters in the core dumps.
212  */
213 #define ELF_CLASS       ELFCLASS32
214 #define ELF_ARCH        EM_386
215 
216 static inline void init_thread(struct target_pt_regs *regs,
217                                struct image_info *infop)
218 {
219     regs->esp = infop->start_stack;
220     regs->eip = infop->entry;
221 
222     /* SVR4/i386 ABI (pages 3-31, 3-32) says that when the program
223        starts %edx contains a pointer to a function which might be
224        registered using `atexit'.  This provides a mean for the
225        dynamic linker to call DT_FINI functions for shared libraries
226        that have been loaded before the code runs.
227 
228        A value of 0 tells we have no such handler.  */
229     regs->edx = 0;
230 }
231 
232 #define ELF_NREG    17
233 typedef target_elf_greg_t  target_elf_gregset_t[ELF_NREG];
234 
235 /*
236  * Note that ELF_NREG should be 19 as there should be place for
237  * TRAPNO and ERR "registers" as well but linux doesn't dump
238  * those.
239  *
240  * See linux kernel: arch/x86/include/asm/elf.h
241  */
242 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUX86State *env)
243 {
244     (*regs)[0] = env->regs[R_EBX];
245     (*regs)[1] = env->regs[R_ECX];
246     (*regs)[2] = env->regs[R_EDX];
247     (*regs)[3] = env->regs[R_ESI];
248     (*regs)[4] = env->regs[R_EDI];
249     (*regs)[5] = env->regs[R_EBP];
250     (*regs)[6] = env->regs[R_EAX];
251     (*regs)[7] = env->segs[R_DS].selector & 0xffff;
252     (*regs)[8] = env->segs[R_ES].selector & 0xffff;
253     (*regs)[9] = env->segs[R_FS].selector & 0xffff;
254     (*regs)[10] = env->segs[R_GS].selector & 0xffff;
255     (*regs)[11] = env->regs[R_EAX]; /* XXX */
256     (*regs)[12] = env->eip;
257     (*regs)[13] = env->segs[R_CS].selector & 0xffff;
258     (*regs)[14] = env->eflags;
259     (*regs)[15] = env->regs[R_ESP];
260     (*regs)[16] = env->segs[R_SS].selector & 0xffff;
261 }
262 #endif
263 
264 #define USE_ELF_CORE_DUMP
265 #define ELF_EXEC_PAGESIZE       4096
266 
267 #endif
268 
269 #ifdef TARGET_ARM
270 
271 #ifndef TARGET_AARCH64
272 /* 32 bit ARM definitions */
273 
274 #define ELF_START_MMAP 0x80000000
275 
276 #define elf_check_arch(x) ((x) == ELF_MACHINE)
277 
278 #define ELF_ARCH        ELF_MACHINE
279 #define ELF_CLASS       ELFCLASS32
280 
281 static inline void init_thread(struct target_pt_regs *regs,
282                                struct image_info *infop)
283 {
284     abi_long stack = infop->start_stack;
285     memset(regs, 0, sizeof(*regs));
286 
287     regs->ARM_cpsr = 0x10;
288     if (infop->entry & 1)
289         regs->ARM_cpsr |= CPSR_T;
290     regs->ARM_pc = infop->entry & 0xfffffffe;
291     regs->ARM_sp = infop->start_stack;
292     /* FIXME - what to for failure of get_user()? */
293     get_user_ual(regs->ARM_r2, stack + 8); /* envp */
294     get_user_ual(regs->ARM_r1, stack + 4); /* envp */
295     /* XXX: it seems that r0 is zeroed after ! */
296     regs->ARM_r0 = 0;
297     /* For uClinux PIC binaries.  */
298     /* XXX: Linux does this only on ARM with no MMU (do we care ?) */
299     regs->ARM_r10 = infop->start_data;
300 }
301 
302 #define ELF_NREG    18
303 typedef target_elf_greg_t  target_elf_gregset_t[ELF_NREG];
304 
305 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUARMState *env)
306 {
307     (*regs)[0] = tswapreg(env->regs[0]);
308     (*regs)[1] = tswapreg(env->regs[1]);
309     (*regs)[2] = tswapreg(env->regs[2]);
310     (*regs)[3] = tswapreg(env->regs[3]);
311     (*regs)[4] = tswapreg(env->regs[4]);
312     (*regs)[5] = tswapreg(env->regs[5]);
313     (*regs)[6] = tswapreg(env->regs[6]);
314     (*regs)[7] = tswapreg(env->regs[7]);
315     (*regs)[8] = tswapreg(env->regs[8]);
316     (*regs)[9] = tswapreg(env->regs[9]);
317     (*regs)[10] = tswapreg(env->regs[10]);
318     (*regs)[11] = tswapreg(env->regs[11]);
319     (*regs)[12] = tswapreg(env->regs[12]);
320     (*regs)[13] = tswapreg(env->regs[13]);
321     (*regs)[14] = tswapreg(env->regs[14]);
322     (*regs)[15] = tswapreg(env->regs[15]);
323 
324     (*regs)[16] = tswapreg(cpsr_read((CPUARMState *)env));
325     (*regs)[17] = tswapreg(env->regs[0]); /* XXX */
326 }
327 
328 #define USE_ELF_CORE_DUMP
329 #define ELF_EXEC_PAGESIZE       4096
330 
331 enum
332 {
333     ARM_HWCAP_ARM_SWP       = 1 << 0,
334     ARM_HWCAP_ARM_HALF      = 1 << 1,
335     ARM_HWCAP_ARM_THUMB     = 1 << 2,
336     ARM_HWCAP_ARM_26BIT     = 1 << 3,
337     ARM_HWCAP_ARM_FAST_MULT = 1 << 4,
338     ARM_HWCAP_ARM_FPA       = 1 << 5,
339     ARM_HWCAP_ARM_VFP       = 1 << 6,
340     ARM_HWCAP_ARM_EDSP      = 1 << 7,
341     ARM_HWCAP_ARM_JAVA      = 1 << 8,
342     ARM_HWCAP_ARM_IWMMXT    = 1 << 9,
343     ARM_HWCAP_ARM_CRUNCH    = 1 << 10,
344     ARM_HWCAP_ARM_THUMBEE   = 1 << 11,
345     ARM_HWCAP_ARM_NEON      = 1 << 12,
346     ARM_HWCAP_ARM_VFPv3     = 1 << 13,
347     ARM_HWCAP_ARM_VFPv3D16  = 1 << 14,
348     ARM_HWCAP_ARM_TLS       = 1 << 15,
349     ARM_HWCAP_ARM_VFPv4     = 1 << 16,
350     ARM_HWCAP_ARM_IDIVA     = 1 << 17,
351     ARM_HWCAP_ARM_IDIVT     = 1 << 18,
352     ARM_HWCAP_ARM_VFPD32    = 1 << 19,
353     ARM_HWCAP_ARM_LPAE      = 1 << 20,
354     ARM_HWCAP_ARM_EVTSTRM   = 1 << 21,
355 };
356 
357 enum {
358     ARM_HWCAP2_ARM_AES      = 1 << 0,
359     ARM_HWCAP2_ARM_PMULL    = 1 << 1,
360     ARM_HWCAP2_ARM_SHA1     = 1 << 2,
361     ARM_HWCAP2_ARM_SHA2     = 1 << 3,
362     ARM_HWCAP2_ARM_CRC32    = 1 << 4,
363 };
364 
365 /* The commpage only exists for 32 bit kernels */
366 
367 #define TARGET_HAS_VALIDATE_GUEST_SPACE
368 /* Return 1 if the proposed guest space is suitable for the guest.
369  * Return 0 if the proposed guest space isn't suitable, but another
370  * address space should be tried.
371  * Return -1 if there is no way the proposed guest space can be
372  * valid regardless of the base.
373  * The guest code may leave a page mapped and populate it if the
374  * address is suitable.
375  */
376 static int validate_guest_space(unsigned long guest_base,
377                                 unsigned long guest_size)
378 {
379     unsigned long real_start, test_page_addr;
380 
381     /* We need to check that we can force a fault on access to the
382      * commpage at 0xffff0fxx
383      */
384     test_page_addr = guest_base + (0xffff0f00 & qemu_host_page_mask);
385 
386     /* If the commpage lies within the already allocated guest space,
387      * then there is no way we can allocate it.
388      */
389     if (test_page_addr >= guest_base
390         && test_page_addr <= (guest_base + guest_size)) {
391         return -1;
392     }
393 
394     /* Note it needs to be writeable to let us initialise it */
395     real_start = (unsigned long)
396                  mmap((void *)test_page_addr, qemu_host_page_size,
397                      PROT_READ | PROT_WRITE,
398                      MAP_ANONYMOUS | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
399 
400     /* If we can't map it then try another address */
401     if (real_start == -1ul) {
402         return 0;
403     }
404 
405     if (real_start != test_page_addr) {
406         /* OS didn't put the page where we asked - unmap and reject */
407         munmap((void *)real_start, qemu_host_page_size);
408         return 0;
409     }
410 
411     /* Leave the page mapped
412      * Populate it (mmap should have left it all 0'd)
413      */
414 
415     /* Kernel helper versions */
416     __put_user(5, (uint32_t *)g2h(0xffff0ffcul));
417 
418     /* Now it's populated make it RO */
419     if (mprotect((void *)test_page_addr, qemu_host_page_size, PROT_READ)) {
420         perror("Protecting guest commpage");
421         exit(-1);
422     }
423 
424     return 1; /* All good */
425 }
426 
427 #define ELF_HWCAP get_elf_hwcap()
428 #define ELF_HWCAP2 get_elf_hwcap2()
429 
430 static uint32_t get_elf_hwcap(void)
431 {
432     ARMCPU *cpu = ARM_CPU(thread_cpu);
433     uint32_t hwcaps = 0;
434 
435     hwcaps |= ARM_HWCAP_ARM_SWP;
436     hwcaps |= ARM_HWCAP_ARM_HALF;
437     hwcaps |= ARM_HWCAP_ARM_THUMB;
438     hwcaps |= ARM_HWCAP_ARM_FAST_MULT;
439 
440     /* probe for the extra features */
441 #define GET_FEATURE(feat, hwcap) \
442     do { if (arm_feature(&cpu->env, feat)) { hwcaps |= hwcap; } } while (0)
443     /* EDSP is in v5TE and above, but all our v5 CPUs are v5TE */
444     GET_FEATURE(ARM_FEATURE_V5, ARM_HWCAP_ARM_EDSP);
445     GET_FEATURE(ARM_FEATURE_VFP, ARM_HWCAP_ARM_VFP);
446     GET_FEATURE(ARM_FEATURE_IWMMXT, ARM_HWCAP_ARM_IWMMXT);
447     GET_FEATURE(ARM_FEATURE_THUMB2EE, ARM_HWCAP_ARM_THUMBEE);
448     GET_FEATURE(ARM_FEATURE_NEON, ARM_HWCAP_ARM_NEON);
449     GET_FEATURE(ARM_FEATURE_VFP3, ARM_HWCAP_ARM_VFPv3);
450     GET_FEATURE(ARM_FEATURE_V6K, ARM_HWCAP_ARM_TLS);
451     GET_FEATURE(ARM_FEATURE_VFP4, ARM_HWCAP_ARM_VFPv4);
452     GET_FEATURE(ARM_FEATURE_ARM_DIV, ARM_HWCAP_ARM_IDIVA);
453     GET_FEATURE(ARM_FEATURE_THUMB_DIV, ARM_HWCAP_ARM_IDIVT);
454     /* All QEMU's VFPv3 CPUs have 32 registers, see VFP_DREG in translate.c.
455      * Note that the ARM_HWCAP_ARM_VFPv3D16 bit is always the inverse of
456      * ARM_HWCAP_ARM_VFPD32 (and so always clear for QEMU); it is unrelated
457      * to our VFP_FP16 feature bit.
458      */
459     GET_FEATURE(ARM_FEATURE_VFP3, ARM_HWCAP_ARM_VFPD32);
460     GET_FEATURE(ARM_FEATURE_LPAE, ARM_HWCAP_ARM_LPAE);
461 
462     return hwcaps;
463 }
464 
465 static uint32_t get_elf_hwcap2(void)
466 {
467     ARMCPU *cpu = ARM_CPU(thread_cpu);
468     uint32_t hwcaps = 0;
469 
470     GET_FEATURE(ARM_FEATURE_V8_AES, ARM_HWCAP2_ARM_AES);
471     GET_FEATURE(ARM_FEATURE_V8_PMULL, ARM_HWCAP2_ARM_PMULL);
472     GET_FEATURE(ARM_FEATURE_V8_SHA1, ARM_HWCAP2_ARM_SHA1);
473     GET_FEATURE(ARM_FEATURE_V8_SHA256, ARM_HWCAP2_ARM_SHA2);
474     GET_FEATURE(ARM_FEATURE_CRC, ARM_HWCAP2_ARM_CRC32);
475     return hwcaps;
476 }
477 
478 #undef GET_FEATURE
479 
480 #else
481 /* 64 bit ARM definitions */
482 #define ELF_START_MMAP 0x80000000
483 
484 #define elf_check_arch(x) ((x) == ELF_MACHINE)
485 
486 #define ELF_ARCH        ELF_MACHINE
487 #define ELF_CLASS       ELFCLASS64
488 #define ELF_PLATFORM    "aarch64"
489 
490 static inline void init_thread(struct target_pt_regs *regs,
491                                struct image_info *infop)
492 {
493     abi_long stack = infop->start_stack;
494     memset(regs, 0, sizeof(*regs));
495 
496     regs->pc = infop->entry & ~0x3ULL;
497     regs->sp = stack;
498 }
499 
500 #define ELF_NREG    34
501 typedef target_elf_greg_t  target_elf_gregset_t[ELF_NREG];
502 
503 static void elf_core_copy_regs(target_elf_gregset_t *regs,
504                                const CPUARMState *env)
505 {
506     int i;
507 
508     for (i = 0; i < 32; i++) {
509         (*regs)[i] = tswapreg(env->xregs[i]);
510     }
511     (*regs)[32] = tswapreg(env->pc);
512     (*regs)[33] = tswapreg(pstate_read((CPUARMState *)env));
513 }
514 
515 #define USE_ELF_CORE_DUMP
516 #define ELF_EXEC_PAGESIZE       4096
517 
518 enum {
519     ARM_HWCAP_A64_FP            = 1 << 0,
520     ARM_HWCAP_A64_ASIMD         = 1 << 1,
521     ARM_HWCAP_A64_EVTSTRM       = 1 << 2,
522     ARM_HWCAP_A64_AES           = 1 << 3,
523     ARM_HWCAP_A64_PMULL         = 1 << 4,
524     ARM_HWCAP_A64_SHA1          = 1 << 5,
525     ARM_HWCAP_A64_SHA2          = 1 << 6,
526     ARM_HWCAP_A64_CRC32         = 1 << 7,
527 };
528 
529 #define ELF_HWCAP get_elf_hwcap()
530 
531 static uint32_t get_elf_hwcap(void)
532 {
533     ARMCPU *cpu = ARM_CPU(thread_cpu);
534     uint32_t hwcaps = 0;
535 
536     hwcaps |= ARM_HWCAP_A64_FP;
537     hwcaps |= ARM_HWCAP_A64_ASIMD;
538 
539     /* probe for the extra features */
540 #define GET_FEATURE(feat, hwcap) \
541     do { if (arm_feature(&cpu->env, feat)) { hwcaps |= hwcap; } } while (0)
542     GET_FEATURE(ARM_FEATURE_V8_AES, ARM_HWCAP_A64_AES);
543     GET_FEATURE(ARM_FEATURE_V8_PMULL, ARM_HWCAP_A64_PMULL);
544     GET_FEATURE(ARM_FEATURE_V8_SHA1, ARM_HWCAP_A64_SHA1);
545     GET_FEATURE(ARM_FEATURE_V8_SHA256, ARM_HWCAP_A64_SHA2);
546     GET_FEATURE(ARM_FEATURE_CRC, ARM_HWCAP_A64_CRC32);
547 #undef GET_FEATURE
548 
549     return hwcaps;
550 }
551 
552 #endif /* not TARGET_AARCH64 */
553 #endif /* TARGET_ARM */
554 
555 #ifdef TARGET_UNICORE32
556 
557 #define ELF_START_MMAP          0x80000000
558 
559 #define elf_check_arch(x)       ((x) == EM_UNICORE32)
560 
561 #define ELF_CLASS               ELFCLASS32
562 #define ELF_DATA                ELFDATA2LSB
563 #define ELF_ARCH                EM_UNICORE32
564 
565 static inline void init_thread(struct target_pt_regs *regs,
566         struct image_info *infop)
567 {
568     abi_long stack = infop->start_stack;
569     memset(regs, 0, sizeof(*regs));
570     regs->UC32_REG_asr = 0x10;
571     regs->UC32_REG_pc = infop->entry & 0xfffffffe;
572     regs->UC32_REG_sp = infop->start_stack;
573     /* FIXME - what to for failure of get_user()? */
574     get_user_ual(regs->UC32_REG_02, stack + 8); /* envp */
575     get_user_ual(regs->UC32_REG_01, stack + 4); /* envp */
576     /* XXX: it seems that r0 is zeroed after ! */
577     regs->UC32_REG_00 = 0;
578 }
579 
580 #define ELF_NREG    34
581 typedef target_elf_greg_t  target_elf_gregset_t[ELF_NREG];
582 
583 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUUniCore32State *env)
584 {
585     (*regs)[0] = env->regs[0];
586     (*regs)[1] = env->regs[1];
587     (*regs)[2] = env->regs[2];
588     (*regs)[3] = env->regs[3];
589     (*regs)[4] = env->regs[4];
590     (*regs)[5] = env->regs[5];
591     (*regs)[6] = env->regs[6];
592     (*regs)[7] = env->regs[7];
593     (*regs)[8] = env->regs[8];
594     (*regs)[9] = env->regs[9];
595     (*regs)[10] = env->regs[10];
596     (*regs)[11] = env->regs[11];
597     (*regs)[12] = env->regs[12];
598     (*regs)[13] = env->regs[13];
599     (*regs)[14] = env->regs[14];
600     (*regs)[15] = env->regs[15];
601     (*regs)[16] = env->regs[16];
602     (*regs)[17] = env->regs[17];
603     (*regs)[18] = env->regs[18];
604     (*regs)[19] = env->regs[19];
605     (*regs)[20] = env->regs[20];
606     (*regs)[21] = env->regs[21];
607     (*regs)[22] = env->regs[22];
608     (*regs)[23] = env->regs[23];
609     (*regs)[24] = env->regs[24];
610     (*regs)[25] = env->regs[25];
611     (*regs)[26] = env->regs[26];
612     (*regs)[27] = env->regs[27];
613     (*regs)[28] = env->regs[28];
614     (*regs)[29] = env->regs[29];
615     (*regs)[30] = env->regs[30];
616     (*regs)[31] = env->regs[31];
617 
618     (*regs)[32] = cpu_asr_read((CPUUniCore32State *)env);
619     (*regs)[33] = env->regs[0]; /* XXX */
620 }
621 
622 #define USE_ELF_CORE_DUMP
623 #define ELF_EXEC_PAGESIZE               4096
624 
625 #define ELF_HWCAP                       (UC32_HWCAP_CMOV | UC32_HWCAP_UCF64)
626 
627 #endif
628 
629 #ifdef TARGET_SPARC
630 #ifdef TARGET_SPARC64
631 
632 #define ELF_START_MMAP 0x80000000
633 #define ELF_HWCAP  (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \
634                     | HWCAP_SPARC_MULDIV | HWCAP_SPARC_V9)
635 #ifndef TARGET_ABI32
636 #define elf_check_arch(x) ( (x) == EM_SPARCV9 || (x) == EM_SPARC32PLUS )
637 #else
638 #define elf_check_arch(x) ( (x) == EM_SPARC32PLUS || (x) == EM_SPARC )
639 #endif
640 
641 #define ELF_CLASS   ELFCLASS64
642 #define ELF_ARCH    EM_SPARCV9
643 
644 #define STACK_BIAS              2047
645 
646 static inline void init_thread(struct target_pt_regs *regs,
647                                struct image_info *infop)
648 {
649 #ifndef TARGET_ABI32
650     regs->tstate = 0;
651 #endif
652     regs->pc = infop->entry;
653     regs->npc = regs->pc + 4;
654     regs->y = 0;
655 #ifdef TARGET_ABI32
656     regs->u_regs[14] = infop->start_stack - 16 * 4;
657 #else
658     if (personality(infop->personality) == PER_LINUX32)
659         regs->u_regs[14] = infop->start_stack - 16 * 4;
660     else
661         regs->u_regs[14] = infop->start_stack - 16 * 8 - STACK_BIAS;
662 #endif
663 }
664 
665 #else
666 #define ELF_START_MMAP 0x80000000
667 #define ELF_HWCAP  (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \
668                     | HWCAP_SPARC_MULDIV)
669 #define elf_check_arch(x) ( (x) == EM_SPARC )
670 
671 #define ELF_CLASS   ELFCLASS32
672 #define ELF_ARCH    EM_SPARC
673 
674 static inline void init_thread(struct target_pt_regs *regs,
675                                struct image_info *infop)
676 {
677     regs->psr = 0;
678     regs->pc = infop->entry;
679     regs->npc = regs->pc + 4;
680     regs->y = 0;
681     regs->u_regs[14] = infop->start_stack - 16 * 4;
682 }
683 
684 #endif
685 #endif
686 
687 #ifdef TARGET_PPC
688 
689 #define ELF_START_MMAP 0x80000000
690 
691 #if defined(TARGET_PPC64) && !defined(TARGET_ABI32)
692 
693 #define elf_check_arch(x) ( (x) == EM_PPC64 )
694 
695 #define ELF_CLASS       ELFCLASS64
696 
697 #else
698 
699 #define elf_check_arch(x) ( (x) == EM_PPC )
700 
701 #define ELF_CLASS       ELFCLASS32
702 
703 #endif
704 
705 #define ELF_ARCH        EM_PPC
706 
707 /* Feature masks for the Aux Vector Hardware Capabilities (AT_HWCAP).
708    See arch/powerpc/include/asm/cputable.h.  */
709 enum {
710     QEMU_PPC_FEATURE_32 = 0x80000000,
711     QEMU_PPC_FEATURE_64 = 0x40000000,
712     QEMU_PPC_FEATURE_601_INSTR = 0x20000000,
713     QEMU_PPC_FEATURE_HAS_ALTIVEC = 0x10000000,
714     QEMU_PPC_FEATURE_HAS_FPU = 0x08000000,
715     QEMU_PPC_FEATURE_HAS_MMU = 0x04000000,
716     QEMU_PPC_FEATURE_HAS_4xxMAC = 0x02000000,
717     QEMU_PPC_FEATURE_UNIFIED_CACHE = 0x01000000,
718     QEMU_PPC_FEATURE_HAS_SPE = 0x00800000,
719     QEMU_PPC_FEATURE_HAS_EFP_SINGLE = 0x00400000,
720     QEMU_PPC_FEATURE_HAS_EFP_DOUBLE = 0x00200000,
721     QEMU_PPC_FEATURE_NO_TB = 0x00100000,
722     QEMU_PPC_FEATURE_POWER4 = 0x00080000,
723     QEMU_PPC_FEATURE_POWER5 = 0x00040000,
724     QEMU_PPC_FEATURE_POWER5_PLUS = 0x00020000,
725     QEMU_PPC_FEATURE_CELL = 0x00010000,
726     QEMU_PPC_FEATURE_BOOKE = 0x00008000,
727     QEMU_PPC_FEATURE_SMT = 0x00004000,
728     QEMU_PPC_FEATURE_ICACHE_SNOOP = 0x00002000,
729     QEMU_PPC_FEATURE_ARCH_2_05 = 0x00001000,
730     QEMU_PPC_FEATURE_PA6T = 0x00000800,
731     QEMU_PPC_FEATURE_HAS_DFP = 0x00000400,
732     QEMU_PPC_FEATURE_POWER6_EXT = 0x00000200,
733     QEMU_PPC_FEATURE_ARCH_2_06 = 0x00000100,
734     QEMU_PPC_FEATURE_HAS_VSX = 0x00000080,
735     QEMU_PPC_FEATURE_PSERIES_PERFMON_COMPAT = 0x00000040,
736 
737     QEMU_PPC_FEATURE_TRUE_LE = 0x00000002,
738     QEMU_PPC_FEATURE_PPC_LE = 0x00000001,
739 
740     /* Feature definitions in AT_HWCAP2.  */
741     QEMU_PPC_FEATURE2_ARCH_2_07 = 0x80000000, /* ISA 2.07 */
742     QEMU_PPC_FEATURE2_HAS_HTM = 0x40000000, /* Hardware Transactional Memory */
743     QEMU_PPC_FEATURE2_HAS_DSCR = 0x20000000, /* Data Stream Control Register */
744     QEMU_PPC_FEATURE2_HAS_EBB = 0x10000000, /* Event Base Branching */
745     QEMU_PPC_FEATURE2_HAS_ISEL = 0x08000000, /* Integer Select */
746     QEMU_PPC_FEATURE2_HAS_TAR = 0x04000000, /* Target Address Register */
747 };
748 
749 #define ELF_HWCAP get_elf_hwcap()
750 
751 static uint32_t get_elf_hwcap(void)
752 {
753     PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);
754     uint32_t features = 0;
755 
756     /* We don't have to be terribly complete here; the high points are
757        Altivec/FP/SPE support.  Anything else is just a bonus.  */
758 #define GET_FEATURE(flag, feature)                                      \
759     do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0)
760 #define GET_FEATURE2(flag, feature)                                      \
761     do { if (cpu->env.insns_flags2 & flag) { features |= feature; } } while (0)
762     GET_FEATURE(PPC_64B, QEMU_PPC_FEATURE_64);
763     GET_FEATURE(PPC_FLOAT, QEMU_PPC_FEATURE_HAS_FPU);
764     GET_FEATURE(PPC_ALTIVEC, QEMU_PPC_FEATURE_HAS_ALTIVEC);
765     GET_FEATURE(PPC_SPE, QEMU_PPC_FEATURE_HAS_SPE);
766     GET_FEATURE(PPC_SPE_SINGLE, QEMU_PPC_FEATURE_HAS_EFP_SINGLE);
767     GET_FEATURE(PPC_SPE_DOUBLE, QEMU_PPC_FEATURE_HAS_EFP_DOUBLE);
768     GET_FEATURE(PPC_BOOKE, QEMU_PPC_FEATURE_BOOKE);
769     GET_FEATURE(PPC_405_MAC, QEMU_PPC_FEATURE_HAS_4xxMAC);
770     GET_FEATURE2(PPC2_DFP, QEMU_PPC_FEATURE_HAS_DFP);
771     GET_FEATURE2(PPC2_VSX, QEMU_PPC_FEATURE_HAS_VSX);
772     GET_FEATURE2((PPC2_PERM_ISA206 | PPC2_DIVE_ISA206 | PPC2_ATOMIC_ISA206 |
773                   PPC2_FP_CVT_ISA206 | PPC2_FP_TST_ISA206),
774                   QEMU_PPC_FEATURE_ARCH_2_06);
775 #undef GET_FEATURE
776 #undef GET_FEATURE2
777 
778     return features;
779 }
780 
781 #define ELF_HWCAP2 get_elf_hwcap2()
782 
783 static uint32_t get_elf_hwcap2(void)
784 {
785     PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);
786     uint32_t features = 0;
787 
788 #define GET_FEATURE(flag, feature)                                      \
789     do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0)
790 #define GET_FEATURE2(flag, feature)                                      \
791     do { if (cpu->env.insns_flags2 & flag) { features |= feature; } } while (0)
792 
793     GET_FEATURE(PPC_ISEL, QEMU_PPC_FEATURE2_HAS_ISEL);
794     GET_FEATURE2(PPC2_BCTAR_ISA207, QEMU_PPC_FEATURE2_HAS_TAR);
795     GET_FEATURE2((PPC2_BCTAR_ISA207 | PPC2_LSQ_ISA207 | PPC2_ALTIVEC_207 |
796                   PPC2_ISA207S), QEMU_PPC_FEATURE2_ARCH_2_07);
797 
798 #undef GET_FEATURE
799 #undef GET_FEATURE2
800 
801     return features;
802 }
803 
804 /*
805  * The requirements here are:
806  * - keep the final alignment of sp (sp & 0xf)
807  * - make sure the 32-bit value at the first 16 byte aligned position of
808  *   AUXV is greater than 16 for glibc compatibility.
809  *   AT_IGNOREPPC is used for that.
810  * - for compatibility with glibc ARCH_DLINFO must always be defined on PPC,
811  *   even if DLINFO_ARCH_ITEMS goes to zero or is undefined.
812  */
813 #define DLINFO_ARCH_ITEMS       5
814 #define ARCH_DLINFO                                     \
815     do {                                                \
816         PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);              \
817         NEW_AUX_ENT(AT_DCACHEBSIZE, cpu->env.dcache_line_size); \
818         NEW_AUX_ENT(AT_ICACHEBSIZE, cpu->env.icache_line_size); \
819         NEW_AUX_ENT(AT_UCACHEBSIZE, 0);                 \
820         /*                                              \
821          * Now handle glibc compatibility.              \
822          */                                             \
823         NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC);        \
824         NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC);        \
825     } while (0)
826 
827 static inline void init_thread(struct target_pt_regs *_regs, struct image_info *infop)
828 {
829     _regs->gpr[1] = infop->start_stack;
830 #if defined(TARGET_PPC64) && !defined(TARGET_ABI32)
831     if (get_ppc64_abi(infop) < 2) {
832         _regs->gpr[2] = ldq_raw(infop->entry + 8) + infop->load_bias;
833         infop->entry = ldq_raw(infop->entry) + infop->load_bias;
834     } else {
835         _regs->gpr[12] = infop->entry;  /* r12 set to global entry address */
836     }
837 #endif
838     _regs->nip = infop->entry;
839 }
840 
841 /* See linux kernel: arch/powerpc/include/asm/elf.h.  */
842 #define ELF_NREG 48
843 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
844 
845 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUPPCState *env)
846 {
847     int i;
848     target_ulong ccr = 0;
849 
850     for (i = 0; i < ARRAY_SIZE(env->gpr); i++) {
851         (*regs)[i] = tswapreg(env->gpr[i]);
852     }
853 
854     (*regs)[32] = tswapreg(env->nip);
855     (*regs)[33] = tswapreg(env->msr);
856     (*regs)[35] = tswapreg(env->ctr);
857     (*regs)[36] = tswapreg(env->lr);
858     (*regs)[37] = tswapreg(env->xer);
859 
860     for (i = 0; i < ARRAY_SIZE(env->crf); i++) {
861         ccr |= env->crf[i] << (32 - ((i + 1) * 4));
862     }
863     (*regs)[38] = tswapreg(ccr);
864 }
865 
866 #define USE_ELF_CORE_DUMP
867 #define ELF_EXEC_PAGESIZE       4096
868 
869 #endif
870 
871 #ifdef TARGET_MIPS
872 
873 #define ELF_START_MMAP 0x80000000
874 
875 #define elf_check_arch(x) ( (x) == EM_MIPS )
876 
877 #ifdef TARGET_MIPS64
878 #define ELF_CLASS   ELFCLASS64
879 #else
880 #define ELF_CLASS   ELFCLASS32
881 #endif
882 #define ELF_ARCH    EM_MIPS
883 
884 static inline void init_thread(struct target_pt_regs *regs,
885                                struct image_info *infop)
886 {
887     regs->cp0_status = 2 << CP0St_KSU;
888     regs->cp0_epc = infop->entry;
889     regs->regs[29] = infop->start_stack;
890 }
891 
892 /* See linux kernel: arch/mips/include/asm/elf.h.  */
893 #define ELF_NREG 45
894 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
895 
896 /* See linux kernel: arch/mips/include/asm/reg.h.  */
897 enum {
898 #ifdef TARGET_MIPS64
899     TARGET_EF_R0 = 0,
900 #else
901     TARGET_EF_R0 = 6,
902 #endif
903     TARGET_EF_R26 = TARGET_EF_R0 + 26,
904     TARGET_EF_R27 = TARGET_EF_R0 + 27,
905     TARGET_EF_LO = TARGET_EF_R0 + 32,
906     TARGET_EF_HI = TARGET_EF_R0 + 33,
907     TARGET_EF_CP0_EPC = TARGET_EF_R0 + 34,
908     TARGET_EF_CP0_BADVADDR = TARGET_EF_R0 + 35,
909     TARGET_EF_CP0_STATUS = TARGET_EF_R0 + 36,
910     TARGET_EF_CP0_CAUSE = TARGET_EF_R0 + 37
911 };
912 
913 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs.  */
914 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMIPSState *env)
915 {
916     int i;
917 
918     for (i = 0; i < TARGET_EF_R0; i++) {
919         (*regs)[i] = 0;
920     }
921     (*regs)[TARGET_EF_R0] = 0;
922 
923     for (i = 1; i < ARRAY_SIZE(env->active_tc.gpr); i++) {
924         (*regs)[TARGET_EF_R0 + i] = tswapreg(env->active_tc.gpr[i]);
925     }
926 
927     (*regs)[TARGET_EF_R26] = 0;
928     (*regs)[TARGET_EF_R27] = 0;
929     (*regs)[TARGET_EF_LO] = tswapreg(env->active_tc.LO[0]);
930     (*regs)[TARGET_EF_HI] = tswapreg(env->active_tc.HI[0]);
931     (*regs)[TARGET_EF_CP0_EPC] = tswapreg(env->active_tc.PC);
932     (*regs)[TARGET_EF_CP0_BADVADDR] = tswapreg(env->CP0_BadVAddr);
933     (*regs)[TARGET_EF_CP0_STATUS] = tswapreg(env->CP0_Status);
934     (*regs)[TARGET_EF_CP0_CAUSE] = tswapreg(env->CP0_Cause);
935 }
936 
937 #define USE_ELF_CORE_DUMP
938 #define ELF_EXEC_PAGESIZE        4096
939 
940 #endif /* TARGET_MIPS */
941 
942 #ifdef TARGET_MICROBLAZE
943 
944 #define ELF_START_MMAP 0x80000000
945 
946 #define elf_check_arch(x) ( (x) == EM_MICROBLAZE || (x) == EM_MICROBLAZE_OLD)
947 
948 #define ELF_CLASS   ELFCLASS32
949 #define ELF_ARCH    EM_MICROBLAZE
950 
951 static inline void init_thread(struct target_pt_regs *regs,
952                                struct image_info *infop)
953 {
954     regs->pc = infop->entry;
955     regs->r1 = infop->start_stack;
956 
957 }
958 
959 #define ELF_EXEC_PAGESIZE        4096
960 
961 #define USE_ELF_CORE_DUMP
962 #define ELF_NREG 38
963 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
964 
965 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs.  */
966 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMBState *env)
967 {
968     int i, pos = 0;
969 
970     for (i = 0; i < 32; i++) {
971         (*regs)[pos++] = tswapreg(env->regs[i]);
972     }
973 
974     for (i = 0; i < 6; i++) {
975         (*regs)[pos++] = tswapreg(env->sregs[i]);
976     }
977 }
978 
979 #endif /* TARGET_MICROBLAZE */
980 
981 #ifdef TARGET_OPENRISC
982 
983 #define ELF_START_MMAP 0x08000000
984 
985 #define elf_check_arch(x) ((x) == EM_OPENRISC)
986 
987 #define ELF_ARCH EM_OPENRISC
988 #define ELF_CLASS ELFCLASS32
989 #define ELF_DATA  ELFDATA2MSB
990 
991 static inline void init_thread(struct target_pt_regs *regs,
992                                struct image_info *infop)
993 {
994     regs->pc = infop->entry;
995     regs->gpr[1] = infop->start_stack;
996 }
997 
998 #define USE_ELF_CORE_DUMP
999 #define ELF_EXEC_PAGESIZE 8192
1000 
1001 /* See linux kernel arch/openrisc/include/asm/elf.h.  */
1002 #define ELF_NREG 34 /* gprs and pc, sr */
1003 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1004 
1005 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1006                                const CPUOpenRISCState *env)
1007 {
1008     int i;
1009 
1010     for (i = 0; i < 32; i++) {
1011         (*regs)[i] = tswapreg(env->gpr[i]);
1012     }
1013 
1014     (*regs)[32] = tswapreg(env->pc);
1015     (*regs)[33] = tswapreg(env->sr);
1016 }
1017 #define ELF_HWCAP 0
1018 #define ELF_PLATFORM NULL
1019 
1020 #endif /* TARGET_OPENRISC */
1021 
1022 #ifdef TARGET_SH4
1023 
1024 #define ELF_START_MMAP 0x80000000
1025 
1026 #define elf_check_arch(x) ( (x) == EM_SH )
1027 
1028 #define ELF_CLASS ELFCLASS32
1029 #define ELF_ARCH  EM_SH
1030 
1031 static inline void init_thread(struct target_pt_regs *regs,
1032                                struct image_info *infop)
1033 {
1034     /* Check other registers XXXXX */
1035     regs->pc = infop->entry;
1036     regs->regs[15] = infop->start_stack;
1037 }
1038 
1039 /* See linux kernel: arch/sh/include/asm/elf.h.  */
1040 #define ELF_NREG 23
1041 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1042 
1043 /* See linux kernel: arch/sh/include/asm/ptrace.h.  */
1044 enum {
1045     TARGET_REG_PC = 16,
1046     TARGET_REG_PR = 17,
1047     TARGET_REG_SR = 18,
1048     TARGET_REG_GBR = 19,
1049     TARGET_REG_MACH = 20,
1050     TARGET_REG_MACL = 21,
1051     TARGET_REG_SYSCALL = 22
1052 };
1053 
1054 static inline void elf_core_copy_regs(target_elf_gregset_t *regs,
1055                                       const CPUSH4State *env)
1056 {
1057     int i;
1058 
1059     for (i = 0; i < 16; i++) {
1060         (*regs[i]) = tswapreg(env->gregs[i]);
1061     }
1062 
1063     (*regs)[TARGET_REG_PC] = tswapreg(env->pc);
1064     (*regs)[TARGET_REG_PR] = tswapreg(env->pr);
1065     (*regs)[TARGET_REG_SR] = tswapreg(env->sr);
1066     (*regs)[TARGET_REG_GBR] = tswapreg(env->gbr);
1067     (*regs)[TARGET_REG_MACH] = tswapreg(env->mach);
1068     (*regs)[TARGET_REG_MACL] = tswapreg(env->macl);
1069     (*regs)[TARGET_REG_SYSCALL] = 0; /* FIXME */
1070 }
1071 
1072 #define USE_ELF_CORE_DUMP
1073 #define ELF_EXEC_PAGESIZE        4096
1074 
1075 #endif
1076 
1077 #ifdef TARGET_CRIS
1078 
1079 #define ELF_START_MMAP 0x80000000
1080 
1081 #define elf_check_arch(x) ( (x) == EM_CRIS )
1082 
1083 #define ELF_CLASS ELFCLASS32
1084 #define ELF_ARCH  EM_CRIS
1085 
1086 static inline void init_thread(struct target_pt_regs *regs,
1087                                struct image_info *infop)
1088 {
1089     regs->erp = infop->entry;
1090 }
1091 
1092 #define ELF_EXEC_PAGESIZE        8192
1093 
1094 #endif
1095 
1096 #ifdef TARGET_M68K
1097 
1098 #define ELF_START_MMAP 0x80000000
1099 
1100 #define elf_check_arch(x) ( (x) == EM_68K )
1101 
1102 #define ELF_CLASS       ELFCLASS32
1103 #define ELF_ARCH        EM_68K
1104 
1105 /* ??? Does this need to do anything?
1106    #define ELF_PLAT_INIT(_r) */
1107 
1108 static inline void init_thread(struct target_pt_regs *regs,
1109                                struct image_info *infop)
1110 {
1111     regs->usp = infop->start_stack;
1112     regs->sr = 0;
1113     regs->pc = infop->entry;
1114 }
1115 
1116 /* See linux kernel: arch/m68k/include/asm/elf.h.  */
1117 #define ELF_NREG 20
1118 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1119 
1120 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUM68KState *env)
1121 {
1122     (*regs)[0] = tswapreg(env->dregs[1]);
1123     (*regs)[1] = tswapreg(env->dregs[2]);
1124     (*regs)[2] = tswapreg(env->dregs[3]);
1125     (*regs)[3] = tswapreg(env->dregs[4]);
1126     (*regs)[4] = tswapreg(env->dregs[5]);
1127     (*regs)[5] = tswapreg(env->dregs[6]);
1128     (*regs)[6] = tswapreg(env->dregs[7]);
1129     (*regs)[7] = tswapreg(env->aregs[0]);
1130     (*regs)[8] = tswapreg(env->aregs[1]);
1131     (*regs)[9] = tswapreg(env->aregs[2]);
1132     (*regs)[10] = tswapreg(env->aregs[3]);
1133     (*regs)[11] = tswapreg(env->aregs[4]);
1134     (*regs)[12] = tswapreg(env->aregs[5]);
1135     (*regs)[13] = tswapreg(env->aregs[6]);
1136     (*regs)[14] = tswapreg(env->dregs[0]);
1137     (*regs)[15] = tswapreg(env->aregs[7]);
1138     (*regs)[16] = tswapreg(env->dregs[0]); /* FIXME: orig_d0 */
1139     (*regs)[17] = tswapreg(env->sr);
1140     (*regs)[18] = tswapreg(env->pc);
1141     (*regs)[19] = 0;  /* FIXME: regs->format | regs->vector */
1142 }
1143 
1144 #define USE_ELF_CORE_DUMP
1145 #define ELF_EXEC_PAGESIZE       8192
1146 
1147 #endif
1148 
1149 #ifdef TARGET_ALPHA
1150 
1151 #define ELF_START_MMAP (0x30000000000ULL)
1152 
1153 #define elf_check_arch(x) ( (x) == ELF_ARCH )
1154 
1155 #define ELF_CLASS      ELFCLASS64
1156 #define ELF_ARCH       EM_ALPHA
1157 
1158 static inline void init_thread(struct target_pt_regs *regs,
1159                                struct image_info *infop)
1160 {
1161     regs->pc = infop->entry;
1162     regs->ps = 8;
1163     regs->usp = infop->start_stack;
1164 }
1165 
1166 #define ELF_EXEC_PAGESIZE        8192
1167 
1168 #endif /* TARGET_ALPHA */
1169 
1170 #ifdef TARGET_S390X
1171 
1172 #define ELF_START_MMAP (0x20000000000ULL)
1173 
1174 #define elf_check_arch(x) ( (x) == ELF_ARCH )
1175 
1176 #define ELF_CLASS	ELFCLASS64
1177 #define ELF_DATA	ELFDATA2MSB
1178 #define ELF_ARCH	EM_S390
1179 
1180 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop)
1181 {
1182     regs->psw.addr = infop->entry;
1183     regs->psw.mask = PSW_MASK_64 | PSW_MASK_32;
1184     regs->gprs[15] = infop->start_stack;
1185 }
1186 
1187 #endif /* TARGET_S390X */
1188 
1189 #ifndef ELF_PLATFORM
1190 #define ELF_PLATFORM (NULL)
1191 #endif
1192 
1193 #ifndef ELF_HWCAP
1194 #define ELF_HWCAP 0
1195 #endif
1196 
1197 #ifdef TARGET_ABI32
1198 #undef ELF_CLASS
1199 #define ELF_CLASS ELFCLASS32
1200 #undef bswaptls
1201 #define bswaptls(ptr) bswap32s(ptr)
1202 #endif
1203 
1204 #include "elf.h"
1205 
1206 struct exec
1207 {
1208     unsigned int a_info;   /* Use macros N_MAGIC, etc for access */
1209     unsigned int a_text;   /* length of text, in bytes */
1210     unsigned int a_data;   /* length of data, in bytes */
1211     unsigned int a_bss;    /* length of uninitialized data area, in bytes */
1212     unsigned int a_syms;   /* length of symbol table data in file, in bytes */
1213     unsigned int a_entry;  /* start address */
1214     unsigned int a_trsize; /* length of relocation info for text, in bytes */
1215     unsigned int a_drsize; /* length of relocation info for data, in bytes */
1216 };
1217 
1218 
1219 #define N_MAGIC(exec) ((exec).a_info & 0xffff)
1220 #define OMAGIC 0407
1221 #define NMAGIC 0410
1222 #define ZMAGIC 0413
1223 #define QMAGIC 0314
1224 
1225 /* Necessary parameters */
1226 #define TARGET_ELF_EXEC_PAGESIZE TARGET_PAGE_SIZE
1227 #define TARGET_ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(TARGET_ELF_EXEC_PAGESIZE-1))
1228 #define TARGET_ELF_PAGEOFFSET(_v) ((_v) & (TARGET_ELF_EXEC_PAGESIZE-1))
1229 
1230 #define DLINFO_ITEMS 14
1231 
1232 static inline void memcpy_fromfs(void * to, const void * from, unsigned long n)
1233 {
1234     memcpy(to, from, n);
1235 }
1236 
1237 #ifdef BSWAP_NEEDED
1238 static void bswap_ehdr(struct elfhdr *ehdr)
1239 {
1240     bswap16s(&ehdr->e_type);            /* Object file type */
1241     bswap16s(&ehdr->e_machine);         /* Architecture */
1242     bswap32s(&ehdr->e_version);         /* Object file version */
1243     bswaptls(&ehdr->e_entry);           /* Entry point virtual address */
1244     bswaptls(&ehdr->e_phoff);           /* Program header table file offset */
1245     bswaptls(&ehdr->e_shoff);           /* Section header table file offset */
1246     bswap32s(&ehdr->e_flags);           /* Processor-specific flags */
1247     bswap16s(&ehdr->e_ehsize);          /* ELF header size in bytes */
1248     bswap16s(&ehdr->e_phentsize);       /* Program header table entry size */
1249     bswap16s(&ehdr->e_phnum);           /* Program header table entry count */
1250     bswap16s(&ehdr->e_shentsize);       /* Section header table entry size */
1251     bswap16s(&ehdr->e_shnum);           /* Section header table entry count */
1252     bswap16s(&ehdr->e_shstrndx);        /* Section header string table index */
1253 }
1254 
1255 static void bswap_phdr(struct elf_phdr *phdr, int phnum)
1256 {
1257     int i;
1258     for (i = 0; i < phnum; ++i, ++phdr) {
1259         bswap32s(&phdr->p_type);        /* Segment type */
1260         bswap32s(&phdr->p_flags);       /* Segment flags */
1261         bswaptls(&phdr->p_offset);      /* Segment file offset */
1262         bswaptls(&phdr->p_vaddr);       /* Segment virtual address */
1263         bswaptls(&phdr->p_paddr);       /* Segment physical address */
1264         bswaptls(&phdr->p_filesz);      /* Segment size in file */
1265         bswaptls(&phdr->p_memsz);       /* Segment size in memory */
1266         bswaptls(&phdr->p_align);       /* Segment alignment */
1267     }
1268 }
1269 
1270 static void bswap_shdr(struct elf_shdr *shdr, int shnum)
1271 {
1272     int i;
1273     for (i = 0; i < shnum; ++i, ++shdr) {
1274         bswap32s(&shdr->sh_name);
1275         bswap32s(&shdr->sh_type);
1276         bswaptls(&shdr->sh_flags);
1277         bswaptls(&shdr->sh_addr);
1278         bswaptls(&shdr->sh_offset);
1279         bswaptls(&shdr->sh_size);
1280         bswap32s(&shdr->sh_link);
1281         bswap32s(&shdr->sh_info);
1282         bswaptls(&shdr->sh_addralign);
1283         bswaptls(&shdr->sh_entsize);
1284     }
1285 }
1286 
1287 static void bswap_sym(struct elf_sym *sym)
1288 {
1289     bswap32s(&sym->st_name);
1290     bswaptls(&sym->st_value);
1291     bswaptls(&sym->st_size);
1292     bswap16s(&sym->st_shndx);
1293 }
1294 #else
1295 static inline void bswap_ehdr(struct elfhdr *ehdr) { }
1296 static inline void bswap_phdr(struct elf_phdr *phdr, int phnum) { }
1297 static inline void bswap_shdr(struct elf_shdr *shdr, int shnum) { }
1298 static inline void bswap_sym(struct elf_sym *sym) { }
1299 #endif
1300 
1301 #ifdef USE_ELF_CORE_DUMP
1302 static int elf_core_dump(int, const CPUArchState *);
1303 #endif /* USE_ELF_CORE_DUMP */
1304 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias);
1305 
1306 /* Verify the portions of EHDR within E_IDENT for the target.
1307    This can be performed before bswapping the entire header.  */
1308 static bool elf_check_ident(struct elfhdr *ehdr)
1309 {
1310     return (ehdr->e_ident[EI_MAG0] == ELFMAG0
1311             && ehdr->e_ident[EI_MAG1] == ELFMAG1
1312             && ehdr->e_ident[EI_MAG2] == ELFMAG2
1313             && ehdr->e_ident[EI_MAG3] == ELFMAG3
1314             && ehdr->e_ident[EI_CLASS] == ELF_CLASS
1315             && ehdr->e_ident[EI_DATA] == ELF_DATA
1316             && ehdr->e_ident[EI_VERSION] == EV_CURRENT);
1317 }
1318 
1319 /* Verify the portions of EHDR outside of E_IDENT for the target.
1320    This has to wait until after bswapping the header.  */
1321 static bool elf_check_ehdr(struct elfhdr *ehdr)
1322 {
1323     return (elf_check_arch(ehdr->e_machine)
1324             && ehdr->e_ehsize == sizeof(struct elfhdr)
1325             && ehdr->e_phentsize == sizeof(struct elf_phdr)
1326             && (ehdr->e_type == ET_EXEC || ehdr->e_type == ET_DYN));
1327 }
1328 
1329 /*
1330  * 'copy_elf_strings()' copies argument/envelope strings from user
1331  * memory to free pages in kernel mem. These are in a format ready
1332  * to be put directly into the top of new user memory.
1333  *
1334  */
1335 static abi_ulong copy_elf_strings(int argc,char ** argv, void **page,
1336                                   abi_ulong p)
1337 {
1338     char *tmp, *tmp1, *pag = NULL;
1339     int len, offset = 0;
1340 
1341     if (!p) {
1342         return 0;       /* bullet-proofing */
1343     }
1344     while (argc-- > 0) {
1345         tmp = argv[argc];
1346         if (!tmp) {
1347             fprintf(stderr, "VFS: argc is wrong");
1348             exit(-1);
1349         }
1350         tmp1 = tmp;
1351         while (*tmp++);
1352         len = tmp - tmp1;
1353         if (p < len) {  /* this shouldn't happen - 128kB */
1354             return 0;
1355         }
1356         while (len) {
1357             --p; --tmp; --len;
1358             if (--offset < 0) {
1359                 offset = p % TARGET_PAGE_SIZE;
1360                 pag = (char *)page[p/TARGET_PAGE_SIZE];
1361                 if (!pag) {
1362                     pag = g_try_malloc0(TARGET_PAGE_SIZE);
1363                     page[p/TARGET_PAGE_SIZE] = pag;
1364                     if (!pag)
1365                         return 0;
1366                 }
1367             }
1368             if (len == 0 || offset == 0) {
1369                 *(pag + offset) = *tmp;
1370             }
1371             else {
1372                 int bytes_to_copy = (len > offset) ? offset : len;
1373                 tmp -= bytes_to_copy;
1374                 p -= bytes_to_copy;
1375                 offset -= bytes_to_copy;
1376                 len -= bytes_to_copy;
1377                 memcpy_fromfs(pag + offset, tmp, bytes_to_copy + 1);
1378             }
1379         }
1380     }
1381     return p;
1382 }
1383 
1384 static abi_ulong setup_arg_pages(abi_ulong p, struct linux_binprm *bprm,
1385                                  struct image_info *info)
1386 {
1387     abi_ulong stack_base, size, error, guard;
1388     int i;
1389 
1390     /* Create enough stack to hold everything.  If we don't use
1391        it for args, we'll use it for something else.  */
1392     size = guest_stack_size;
1393     if (size < MAX_ARG_PAGES*TARGET_PAGE_SIZE) {
1394         size = MAX_ARG_PAGES*TARGET_PAGE_SIZE;
1395     }
1396     guard = TARGET_PAGE_SIZE;
1397     if (guard < qemu_real_host_page_size) {
1398         guard = qemu_real_host_page_size;
1399     }
1400 
1401     error = target_mmap(0, size + guard, PROT_READ | PROT_WRITE,
1402                         MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
1403     if (error == -1) {
1404         perror("mmap stack");
1405         exit(-1);
1406     }
1407 
1408     /* We reserve one extra page at the top of the stack as guard.  */
1409     target_mprotect(error, guard, PROT_NONE);
1410 
1411     info->stack_limit = error + guard;
1412     stack_base = info->stack_limit + size - MAX_ARG_PAGES*TARGET_PAGE_SIZE;
1413     p += stack_base;
1414 
1415     for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
1416         if (bprm->page[i]) {
1417             info->rss++;
1418             /* FIXME - check return value of memcpy_to_target() for failure */
1419             memcpy_to_target(stack_base, bprm->page[i], TARGET_PAGE_SIZE);
1420             g_free(bprm->page[i]);
1421         }
1422         stack_base += TARGET_PAGE_SIZE;
1423     }
1424     return p;
1425 }
1426 
1427 /* Map and zero the bss.  We need to explicitly zero any fractional pages
1428    after the data section (i.e. bss).  */
1429 static void zero_bss(abi_ulong elf_bss, abi_ulong last_bss, int prot)
1430 {
1431     uintptr_t host_start, host_map_start, host_end;
1432 
1433     last_bss = TARGET_PAGE_ALIGN(last_bss);
1434 
1435     /* ??? There is confusion between qemu_real_host_page_size and
1436        qemu_host_page_size here and elsewhere in target_mmap, which
1437        may lead to the end of the data section mapping from the file
1438        not being mapped.  At least there was an explicit test and
1439        comment for that here, suggesting that "the file size must
1440        be known".  The comment probably pre-dates the introduction
1441        of the fstat system call in target_mmap which does in fact
1442        find out the size.  What isn't clear is if the workaround
1443        here is still actually needed.  For now, continue with it,
1444        but merge it with the "normal" mmap that would allocate the bss.  */
1445 
1446     host_start = (uintptr_t) g2h(elf_bss);
1447     host_end = (uintptr_t) g2h(last_bss);
1448     host_map_start = (host_start + qemu_real_host_page_size - 1);
1449     host_map_start &= -qemu_real_host_page_size;
1450 
1451     if (host_map_start < host_end) {
1452         void *p = mmap((void *)host_map_start, host_end - host_map_start,
1453                        prot, MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
1454         if (p == MAP_FAILED) {
1455             perror("cannot mmap brk");
1456             exit(-1);
1457         }
1458     }
1459 
1460     /* Ensure that the bss page(s) are valid */
1461     if ((page_get_flags(last_bss-1) & prot) != prot) {
1462         page_set_flags(elf_bss & TARGET_PAGE_MASK, last_bss, prot | PAGE_VALID);
1463     }
1464 
1465     if (host_start < host_map_start) {
1466         memset((void *)host_start, 0, host_map_start - host_start);
1467     }
1468 }
1469 
1470 #ifdef CONFIG_USE_FDPIC
1471 static abi_ulong loader_build_fdpic_loadmap(struct image_info *info, abi_ulong sp)
1472 {
1473     uint16_t n;
1474     struct elf32_fdpic_loadseg *loadsegs = info->loadsegs;
1475 
1476     /* elf32_fdpic_loadseg */
1477     n = info->nsegs;
1478     while (n--) {
1479         sp -= 12;
1480         put_user_u32(loadsegs[n].addr, sp+0);
1481         put_user_u32(loadsegs[n].p_vaddr, sp+4);
1482         put_user_u32(loadsegs[n].p_memsz, sp+8);
1483     }
1484 
1485     /* elf32_fdpic_loadmap */
1486     sp -= 4;
1487     put_user_u16(0, sp+0); /* version */
1488     put_user_u16(info->nsegs, sp+2); /* nsegs */
1489 
1490     info->personality = PER_LINUX_FDPIC;
1491     info->loadmap_addr = sp;
1492 
1493     return sp;
1494 }
1495 #endif
1496 
1497 static abi_ulong create_elf_tables(abi_ulong p, int argc, int envc,
1498                                    struct elfhdr *exec,
1499                                    struct image_info *info,
1500                                    struct image_info *interp_info)
1501 {
1502     abi_ulong sp;
1503     abi_ulong sp_auxv;
1504     int size;
1505     int i;
1506     abi_ulong u_rand_bytes;
1507     uint8_t k_rand_bytes[16];
1508     abi_ulong u_platform;
1509     const char *k_platform;
1510     const int n = sizeof(elf_addr_t);
1511 
1512     sp = p;
1513 
1514 #ifdef CONFIG_USE_FDPIC
1515     /* Needs to be before we load the env/argc/... */
1516     if (elf_is_fdpic(exec)) {
1517         /* Need 4 byte alignment for these structs */
1518         sp &= ~3;
1519         sp = loader_build_fdpic_loadmap(info, sp);
1520         info->other_info = interp_info;
1521         if (interp_info) {
1522             interp_info->other_info = info;
1523             sp = loader_build_fdpic_loadmap(interp_info, sp);
1524         }
1525     }
1526 #endif
1527 
1528     u_platform = 0;
1529     k_platform = ELF_PLATFORM;
1530     if (k_platform) {
1531         size_t len = strlen(k_platform) + 1;
1532         sp -= (len + n - 1) & ~(n - 1);
1533         u_platform = sp;
1534         /* FIXME - check return value of memcpy_to_target() for failure */
1535         memcpy_to_target(sp, k_platform, len);
1536     }
1537 
1538     /*
1539      * Generate 16 random bytes for userspace PRNG seeding (not
1540      * cryptically secure but it's not the aim of QEMU).
1541      */
1542     srand((unsigned int) time(NULL));
1543     for (i = 0; i < 16; i++) {
1544         k_rand_bytes[i] = rand();
1545     }
1546     sp -= 16;
1547     u_rand_bytes = sp;
1548     /* FIXME - check return value of memcpy_to_target() for failure */
1549     memcpy_to_target(sp, k_rand_bytes, 16);
1550 
1551     /*
1552      * Force 16 byte _final_ alignment here for generality.
1553      */
1554     sp = sp &~ (abi_ulong)15;
1555     size = (DLINFO_ITEMS + 1) * 2;
1556     if (k_platform)
1557         size += 2;
1558 #ifdef DLINFO_ARCH_ITEMS
1559     size += DLINFO_ARCH_ITEMS * 2;
1560 #endif
1561 #ifdef ELF_HWCAP2
1562     size += 2;
1563 #endif
1564     size += envc + argc + 2;
1565     size += 1;  /* argc itself */
1566     size *= n;
1567     if (size & 15)
1568         sp -= 16 - (size & 15);
1569 
1570     /* This is correct because Linux defines
1571      * elf_addr_t as Elf32_Off / Elf64_Off
1572      */
1573 #define NEW_AUX_ENT(id, val) do {               \
1574         sp -= n; put_user_ual(val, sp);         \
1575         sp -= n; put_user_ual(id, sp);          \
1576     } while(0)
1577 
1578     sp_auxv = sp;
1579     NEW_AUX_ENT (AT_NULL, 0);
1580 
1581     /* There must be exactly DLINFO_ITEMS entries here.  */
1582     NEW_AUX_ENT(AT_PHDR, (abi_ulong)(info->load_addr + exec->e_phoff));
1583     NEW_AUX_ENT(AT_PHENT, (abi_ulong)(sizeof (struct elf_phdr)));
1584     NEW_AUX_ENT(AT_PHNUM, (abi_ulong)(exec->e_phnum));
1585     NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(MAX(TARGET_PAGE_SIZE, getpagesize())));
1586     NEW_AUX_ENT(AT_BASE, (abi_ulong)(interp_info ? interp_info->load_addr : 0));
1587     NEW_AUX_ENT(AT_FLAGS, (abi_ulong)0);
1588     NEW_AUX_ENT(AT_ENTRY, info->entry);
1589     NEW_AUX_ENT(AT_UID, (abi_ulong) getuid());
1590     NEW_AUX_ENT(AT_EUID, (abi_ulong) geteuid());
1591     NEW_AUX_ENT(AT_GID, (abi_ulong) getgid());
1592     NEW_AUX_ENT(AT_EGID, (abi_ulong) getegid());
1593     NEW_AUX_ENT(AT_HWCAP, (abi_ulong) ELF_HWCAP);
1594     NEW_AUX_ENT(AT_CLKTCK, (abi_ulong) sysconf(_SC_CLK_TCK));
1595     NEW_AUX_ENT(AT_RANDOM, (abi_ulong) u_rand_bytes);
1596 
1597 #ifdef ELF_HWCAP2
1598     NEW_AUX_ENT(AT_HWCAP2, (abi_ulong) ELF_HWCAP2);
1599 #endif
1600 
1601     if (k_platform)
1602         NEW_AUX_ENT(AT_PLATFORM, u_platform);
1603 #ifdef ARCH_DLINFO
1604     /*
1605      * ARCH_DLINFO must come last so platform specific code can enforce
1606      * special alignment requirements on the AUXV if necessary (eg. PPC).
1607      */
1608     ARCH_DLINFO;
1609 #endif
1610 #undef NEW_AUX_ENT
1611 
1612     info->saved_auxv = sp;
1613     info->auxv_len = sp_auxv - sp;
1614 
1615     sp = loader_build_argptr(envc, argc, sp, p, 0);
1616     /* Check the right amount of stack was allocated for auxvec, envp & argv. */
1617     assert(sp_auxv - sp == size);
1618     return sp;
1619 }
1620 
1621 #ifndef TARGET_HAS_VALIDATE_GUEST_SPACE
1622 /* If the guest doesn't have a validation function just agree */
1623 static int validate_guest_space(unsigned long guest_base,
1624                                 unsigned long guest_size)
1625 {
1626     return 1;
1627 }
1628 #endif
1629 
1630 unsigned long init_guest_space(unsigned long host_start,
1631                                unsigned long host_size,
1632                                unsigned long guest_start,
1633                                bool fixed)
1634 {
1635     unsigned long current_start, real_start;
1636     int flags;
1637 
1638     assert(host_start || host_size);
1639 
1640     /* If just a starting address is given, then just verify that
1641      * address.  */
1642     if (host_start && !host_size) {
1643         if (validate_guest_space(host_start, host_size) == 1) {
1644             return host_start;
1645         } else {
1646             return (unsigned long)-1;
1647         }
1648     }
1649 
1650     /* Setup the initial flags and start address.  */
1651     current_start = host_start & qemu_host_page_mask;
1652     flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE;
1653     if (fixed) {
1654         flags |= MAP_FIXED;
1655     }
1656 
1657     /* Otherwise, a non-zero size region of memory needs to be mapped
1658      * and validated.  */
1659     while (1) {
1660         unsigned long real_size = host_size;
1661 
1662         /* Do not use mmap_find_vma here because that is limited to the
1663          * guest address space.  We are going to make the
1664          * guest address space fit whatever we're given.
1665          */
1666         real_start = (unsigned long)
1667             mmap((void *)current_start, host_size, PROT_NONE, flags, -1, 0);
1668         if (real_start == (unsigned long)-1) {
1669             return (unsigned long)-1;
1670         }
1671 
1672         /* Ensure the address is properly aligned.  */
1673         if (real_start & ~qemu_host_page_mask) {
1674             munmap((void *)real_start, host_size);
1675             real_size = host_size + qemu_host_page_size;
1676             real_start = (unsigned long)
1677                 mmap((void *)real_start, real_size, PROT_NONE, flags, -1, 0);
1678             if (real_start == (unsigned long)-1) {
1679                 return (unsigned long)-1;
1680             }
1681             real_start = HOST_PAGE_ALIGN(real_start);
1682         }
1683 
1684         /* Check to see if the address is valid.  */
1685         if (!host_start || real_start == current_start) {
1686             int valid = validate_guest_space(real_start - guest_start,
1687                                              real_size);
1688             if (valid == 1) {
1689                 break;
1690             } else if (valid == -1) {
1691                 return (unsigned long)-1;
1692             }
1693             /* valid == 0, so try again. */
1694         }
1695 
1696         /* That address didn't work.  Unmap and try a different one.
1697          * The address the host picked because is typically right at
1698          * the top of the host address space and leaves the guest with
1699          * no usable address space.  Resort to a linear search.  We
1700          * already compensated for mmap_min_addr, so this should not
1701          * happen often.  Probably means we got unlucky and host
1702          * address space randomization put a shared library somewhere
1703          * inconvenient.
1704          */
1705         munmap((void *)real_start, host_size);
1706         current_start += qemu_host_page_size;
1707         if (host_start == current_start) {
1708             /* Theoretically possible if host doesn't have any suitably
1709              * aligned areas.  Normally the first mmap will fail.
1710              */
1711             return (unsigned long)-1;
1712         }
1713     }
1714 
1715     qemu_log("Reserved 0x%lx bytes of guest address space\n", host_size);
1716 
1717     return real_start;
1718 }
1719 
1720 static void probe_guest_base(const char *image_name,
1721                              abi_ulong loaddr, abi_ulong hiaddr)
1722 {
1723     /* Probe for a suitable guest base address, if the user has not set
1724      * it explicitly, and set guest_base appropriately.
1725      * In case of error we will print a suitable message and exit.
1726      */
1727 #if defined(CONFIG_USE_GUEST_BASE)
1728     const char *errmsg;
1729     if (!have_guest_base && !reserved_va) {
1730         unsigned long host_start, real_start, host_size;
1731 
1732         /* Round addresses to page boundaries.  */
1733         loaddr &= qemu_host_page_mask;
1734         hiaddr = HOST_PAGE_ALIGN(hiaddr);
1735 
1736         if (loaddr < mmap_min_addr) {
1737             host_start = HOST_PAGE_ALIGN(mmap_min_addr);
1738         } else {
1739             host_start = loaddr;
1740             if (host_start != loaddr) {
1741                 errmsg = "Address overflow loading ELF binary";
1742                 goto exit_errmsg;
1743             }
1744         }
1745         host_size = hiaddr - loaddr;
1746 
1747         /* Setup the initial guest memory space with ranges gleaned from
1748          * the ELF image that is being loaded.
1749          */
1750         real_start = init_guest_space(host_start, host_size, loaddr, false);
1751         if (real_start == (unsigned long)-1) {
1752             errmsg = "Unable to find space for application";
1753             goto exit_errmsg;
1754         }
1755         guest_base = real_start - loaddr;
1756 
1757         qemu_log("Relocating guest address space from 0x"
1758                  TARGET_ABI_FMT_lx " to 0x%lx\n",
1759                  loaddr, real_start);
1760     }
1761     return;
1762 
1763 exit_errmsg:
1764     fprintf(stderr, "%s: %s\n", image_name, errmsg);
1765     exit(-1);
1766 #endif
1767 }
1768 
1769 
1770 /* Load an ELF image into the address space.
1771 
1772    IMAGE_NAME is the filename of the image, to use in error messages.
1773    IMAGE_FD is the open file descriptor for the image.
1774 
1775    BPRM_BUF is a copy of the beginning of the file; this of course
1776    contains the elf file header at offset 0.  It is assumed that this
1777    buffer is sufficiently aligned to present no problems to the host
1778    in accessing data at aligned offsets within the buffer.
1779 
1780    On return: INFO values will be filled in, as necessary or available.  */
1781 
1782 static void load_elf_image(const char *image_name, int image_fd,
1783                            struct image_info *info, char **pinterp_name,
1784                            char bprm_buf[BPRM_BUF_SIZE])
1785 {
1786     struct elfhdr *ehdr = (struct elfhdr *)bprm_buf;
1787     struct elf_phdr *phdr;
1788     abi_ulong load_addr, load_bias, loaddr, hiaddr, error;
1789     int i, retval;
1790     const char *errmsg;
1791 
1792     /* First of all, some simple consistency checks */
1793     errmsg = "Invalid ELF image for this architecture";
1794     if (!elf_check_ident(ehdr)) {
1795         goto exit_errmsg;
1796     }
1797     bswap_ehdr(ehdr);
1798     if (!elf_check_ehdr(ehdr)) {
1799         goto exit_errmsg;
1800     }
1801 
1802     i = ehdr->e_phnum * sizeof(struct elf_phdr);
1803     if (ehdr->e_phoff + i <= BPRM_BUF_SIZE) {
1804         phdr = (struct elf_phdr *)(bprm_buf + ehdr->e_phoff);
1805     } else {
1806         phdr = (struct elf_phdr *) alloca(i);
1807         retval = pread(image_fd, phdr, i, ehdr->e_phoff);
1808         if (retval != i) {
1809             goto exit_read;
1810         }
1811     }
1812     bswap_phdr(phdr, ehdr->e_phnum);
1813 
1814 #ifdef CONFIG_USE_FDPIC
1815     info->nsegs = 0;
1816     info->pt_dynamic_addr = 0;
1817 #endif
1818 
1819     /* Find the maximum size of the image and allocate an appropriate
1820        amount of memory to handle that.  */
1821     loaddr = -1, hiaddr = 0;
1822     for (i = 0; i < ehdr->e_phnum; ++i) {
1823         if (phdr[i].p_type == PT_LOAD) {
1824             abi_ulong a = phdr[i].p_vaddr;
1825             if (a < loaddr) {
1826                 loaddr = a;
1827             }
1828             a += phdr[i].p_memsz;
1829             if (a > hiaddr) {
1830                 hiaddr = a;
1831             }
1832 #ifdef CONFIG_USE_FDPIC
1833             ++info->nsegs;
1834 #endif
1835         }
1836     }
1837 
1838     load_addr = loaddr;
1839     if (ehdr->e_type == ET_DYN) {
1840         /* The image indicates that it can be loaded anywhere.  Find a
1841            location that can hold the memory space required.  If the
1842            image is pre-linked, LOADDR will be non-zero.  Since we do
1843            not supply MAP_FIXED here we'll use that address if and
1844            only if it remains available.  */
1845         load_addr = target_mmap(loaddr, hiaddr - loaddr, PROT_NONE,
1846                                 MAP_PRIVATE | MAP_ANON | MAP_NORESERVE,
1847                                 -1, 0);
1848         if (load_addr == -1) {
1849             goto exit_perror;
1850         }
1851     } else if (pinterp_name != NULL) {
1852         /* This is the main executable.  Make sure that the low
1853            address does not conflict with MMAP_MIN_ADDR or the
1854            QEMU application itself.  */
1855         probe_guest_base(image_name, loaddr, hiaddr);
1856     }
1857     load_bias = load_addr - loaddr;
1858 
1859 #ifdef CONFIG_USE_FDPIC
1860     {
1861         struct elf32_fdpic_loadseg *loadsegs = info->loadsegs =
1862             g_malloc(sizeof(*loadsegs) * info->nsegs);
1863 
1864         for (i = 0; i < ehdr->e_phnum; ++i) {
1865             switch (phdr[i].p_type) {
1866             case PT_DYNAMIC:
1867                 info->pt_dynamic_addr = phdr[i].p_vaddr + load_bias;
1868                 break;
1869             case PT_LOAD:
1870                 loadsegs->addr = phdr[i].p_vaddr + load_bias;
1871                 loadsegs->p_vaddr = phdr[i].p_vaddr;
1872                 loadsegs->p_memsz = phdr[i].p_memsz;
1873                 ++loadsegs;
1874                 break;
1875             }
1876         }
1877     }
1878 #endif
1879 
1880     info->load_bias = load_bias;
1881     info->load_addr = load_addr;
1882     info->entry = ehdr->e_entry + load_bias;
1883     info->start_code = -1;
1884     info->end_code = 0;
1885     info->start_data = -1;
1886     info->end_data = 0;
1887     info->brk = 0;
1888     info->elf_flags = ehdr->e_flags;
1889 
1890     for (i = 0; i < ehdr->e_phnum; i++) {
1891         struct elf_phdr *eppnt = phdr + i;
1892         if (eppnt->p_type == PT_LOAD) {
1893             abi_ulong vaddr, vaddr_po, vaddr_ps, vaddr_ef, vaddr_em;
1894             int elf_prot = 0;
1895 
1896             if (eppnt->p_flags & PF_R) elf_prot =  PROT_READ;
1897             if (eppnt->p_flags & PF_W) elf_prot |= PROT_WRITE;
1898             if (eppnt->p_flags & PF_X) elf_prot |= PROT_EXEC;
1899 
1900             vaddr = load_bias + eppnt->p_vaddr;
1901             vaddr_po = TARGET_ELF_PAGEOFFSET(vaddr);
1902             vaddr_ps = TARGET_ELF_PAGESTART(vaddr);
1903 
1904             error = target_mmap(vaddr_ps, eppnt->p_filesz + vaddr_po,
1905                                 elf_prot, MAP_PRIVATE | MAP_FIXED,
1906                                 image_fd, eppnt->p_offset - vaddr_po);
1907             if (error == -1) {
1908                 goto exit_perror;
1909             }
1910 
1911             vaddr_ef = vaddr + eppnt->p_filesz;
1912             vaddr_em = vaddr + eppnt->p_memsz;
1913 
1914             /* If the load segment requests extra zeros (e.g. bss), map it.  */
1915             if (vaddr_ef < vaddr_em) {
1916                 zero_bss(vaddr_ef, vaddr_em, elf_prot);
1917             }
1918 
1919             /* Find the full program boundaries.  */
1920             if (elf_prot & PROT_EXEC) {
1921                 if (vaddr < info->start_code) {
1922                     info->start_code = vaddr;
1923                 }
1924                 if (vaddr_ef > info->end_code) {
1925                     info->end_code = vaddr_ef;
1926                 }
1927             }
1928             if (elf_prot & PROT_WRITE) {
1929                 if (vaddr < info->start_data) {
1930                     info->start_data = vaddr;
1931                 }
1932                 if (vaddr_ef > info->end_data) {
1933                     info->end_data = vaddr_ef;
1934                 }
1935                 if (vaddr_em > info->brk) {
1936                     info->brk = vaddr_em;
1937                 }
1938             }
1939         } else if (eppnt->p_type == PT_INTERP && pinterp_name) {
1940             char *interp_name;
1941 
1942             if (*pinterp_name) {
1943                 errmsg = "Multiple PT_INTERP entries";
1944                 goto exit_errmsg;
1945             }
1946             interp_name = malloc(eppnt->p_filesz);
1947             if (!interp_name) {
1948                 goto exit_perror;
1949             }
1950 
1951             if (eppnt->p_offset + eppnt->p_filesz <= BPRM_BUF_SIZE) {
1952                 memcpy(interp_name, bprm_buf + eppnt->p_offset,
1953                        eppnt->p_filesz);
1954             } else {
1955                 retval = pread(image_fd, interp_name, eppnt->p_filesz,
1956                                eppnt->p_offset);
1957                 if (retval != eppnt->p_filesz) {
1958                     goto exit_perror;
1959                 }
1960             }
1961             if (interp_name[eppnt->p_filesz - 1] != 0) {
1962                 errmsg = "Invalid PT_INTERP entry";
1963                 goto exit_errmsg;
1964             }
1965             *pinterp_name = interp_name;
1966         }
1967     }
1968 
1969     if (info->end_data == 0) {
1970         info->start_data = info->end_code;
1971         info->end_data = info->end_code;
1972         info->brk = info->end_code;
1973     }
1974 
1975     if (qemu_log_enabled()) {
1976         load_symbols(ehdr, image_fd, load_bias);
1977     }
1978 
1979     close(image_fd);
1980     return;
1981 
1982  exit_read:
1983     if (retval >= 0) {
1984         errmsg = "Incomplete read of file header";
1985         goto exit_errmsg;
1986     }
1987  exit_perror:
1988     errmsg = strerror(errno);
1989  exit_errmsg:
1990     fprintf(stderr, "%s: %s\n", image_name, errmsg);
1991     exit(-1);
1992 }
1993 
1994 static void load_elf_interp(const char *filename, struct image_info *info,
1995                             char bprm_buf[BPRM_BUF_SIZE])
1996 {
1997     int fd, retval;
1998 
1999     fd = open(path(filename), O_RDONLY);
2000     if (fd < 0) {
2001         goto exit_perror;
2002     }
2003 
2004     retval = read(fd, bprm_buf, BPRM_BUF_SIZE);
2005     if (retval < 0) {
2006         goto exit_perror;
2007     }
2008     if (retval < BPRM_BUF_SIZE) {
2009         memset(bprm_buf + retval, 0, BPRM_BUF_SIZE - retval);
2010     }
2011 
2012     load_elf_image(filename, fd, info, NULL, bprm_buf);
2013     return;
2014 
2015  exit_perror:
2016     fprintf(stderr, "%s: %s\n", filename, strerror(errno));
2017     exit(-1);
2018 }
2019 
2020 static int symfind(const void *s0, const void *s1)
2021 {
2022     target_ulong addr = *(target_ulong *)s0;
2023     struct elf_sym *sym = (struct elf_sym *)s1;
2024     int result = 0;
2025     if (addr < sym->st_value) {
2026         result = -1;
2027     } else if (addr >= sym->st_value + sym->st_size) {
2028         result = 1;
2029     }
2030     return result;
2031 }
2032 
2033 static const char *lookup_symbolxx(struct syminfo *s, target_ulong orig_addr)
2034 {
2035 #if ELF_CLASS == ELFCLASS32
2036     struct elf_sym *syms = s->disas_symtab.elf32;
2037 #else
2038     struct elf_sym *syms = s->disas_symtab.elf64;
2039 #endif
2040 
2041     // binary search
2042     struct elf_sym *sym;
2043 
2044     sym = bsearch(&orig_addr, syms, s->disas_num_syms, sizeof(*syms), symfind);
2045     if (sym != NULL) {
2046         return s->disas_strtab + sym->st_name;
2047     }
2048 
2049     return "";
2050 }
2051 
2052 /* FIXME: This should use elf_ops.h  */
2053 static int symcmp(const void *s0, const void *s1)
2054 {
2055     struct elf_sym *sym0 = (struct elf_sym *)s0;
2056     struct elf_sym *sym1 = (struct elf_sym *)s1;
2057     return (sym0->st_value < sym1->st_value)
2058         ? -1
2059         : ((sym0->st_value > sym1->st_value) ? 1 : 0);
2060 }
2061 
2062 /* Best attempt to load symbols from this ELF object. */
2063 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias)
2064 {
2065     int i, shnum, nsyms, sym_idx = 0, str_idx = 0;
2066     struct elf_shdr *shdr;
2067     char *strings = NULL;
2068     struct syminfo *s = NULL;
2069     struct elf_sym *new_syms, *syms = NULL;
2070 
2071     shnum = hdr->e_shnum;
2072     i = shnum * sizeof(struct elf_shdr);
2073     shdr = (struct elf_shdr *)alloca(i);
2074     if (pread(fd, shdr, i, hdr->e_shoff) != i) {
2075         return;
2076     }
2077 
2078     bswap_shdr(shdr, shnum);
2079     for (i = 0; i < shnum; ++i) {
2080         if (shdr[i].sh_type == SHT_SYMTAB) {
2081             sym_idx = i;
2082             str_idx = shdr[i].sh_link;
2083             goto found;
2084         }
2085     }
2086 
2087     /* There will be no symbol table if the file was stripped.  */
2088     return;
2089 
2090  found:
2091     /* Now know where the strtab and symtab are.  Snarf them.  */
2092     s = malloc(sizeof(*s));
2093     if (!s) {
2094         goto give_up;
2095     }
2096 
2097     i = shdr[str_idx].sh_size;
2098     s->disas_strtab = strings = malloc(i);
2099     if (!strings || pread(fd, strings, i, shdr[str_idx].sh_offset) != i) {
2100         goto give_up;
2101     }
2102 
2103     i = shdr[sym_idx].sh_size;
2104     syms = malloc(i);
2105     if (!syms || pread(fd, syms, i, shdr[sym_idx].sh_offset) != i) {
2106         goto give_up;
2107     }
2108 
2109     nsyms = i / sizeof(struct elf_sym);
2110     for (i = 0; i < nsyms; ) {
2111         bswap_sym(syms + i);
2112         /* Throw away entries which we do not need.  */
2113         if (syms[i].st_shndx == SHN_UNDEF
2114             || syms[i].st_shndx >= SHN_LORESERVE
2115             || ELF_ST_TYPE(syms[i].st_info) != STT_FUNC) {
2116             if (i < --nsyms) {
2117                 syms[i] = syms[nsyms];
2118             }
2119         } else {
2120 #if defined(TARGET_ARM) || defined (TARGET_MIPS)
2121             /* The bottom address bit marks a Thumb or MIPS16 symbol.  */
2122             syms[i].st_value &= ~(target_ulong)1;
2123 #endif
2124             syms[i].st_value += load_bias;
2125             i++;
2126         }
2127     }
2128 
2129     /* No "useful" symbol.  */
2130     if (nsyms == 0) {
2131         goto give_up;
2132     }
2133 
2134     /* Attempt to free the storage associated with the local symbols
2135        that we threw away.  Whether or not this has any effect on the
2136        memory allocation depends on the malloc implementation and how
2137        many symbols we managed to discard.  */
2138     new_syms = realloc(syms, nsyms * sizeof(*syms));
2139     if (new_syms == NULL) {
2140         goto give_up;
2141     }
2142     syms = new_syms;
2143 
2144     qsort(syms, nsyms, sizeof(*syms), symcmp);
2145 
2146     s->disas_num_syms = nsyms;
2147 #if ELF_CLASS == ELFCLASS32
2148     s->disas_symtab.elf32 = syms;
2149 #else
2150     s->disas_symtab.elf64 = syms;
2151 #endif
2152     s->lookup_symbol = lookup_symbolxx;
2153     s->next = syminfos;
2154     syminfos = s;
2155 
2156     return;
2157 
2158 give_up:
2159     free(s);
2160     free(strings);
2161     free(syms);
2162 }
2163 
2164 int load_elf_binary(struct linux_binprm *bprm, struct image_info *info)
2165 {
2166     struct image_info interp_info;
2167     struct elfhdr elf_ex;
2168     char *elf_interpreter = NULL;
2169 
2170     info->start_mmap = (abi_ulong)ELF_START_MMAP;
2171     info->mmap = 0;
2172     info->rss = 0;
2173 
2174     load_elf_image(bprm->filename, bprm->fd, info,
2175                    &elf_interpreter, bprm->buf);
2176 
2177     /* ??? We need a copy of the elf header for passing to create_elf_tables.
2178        If we do nothing, we'll have overwritten this when we re-use bprm->buf
2179        when we load the interpreter.  */
2180     elf_ex = *(struct elfhdr *)bprm->buf;
2181 
2182     bprm->p = copy_elf_strings(1, &bprm->filename, bprm->page, bprm->p);
2183     bprm->p = copy_elf_strings(bprm->envc,bprm->envp,bprm->page,bprm->p);
2184     bprm->p = copy_elf_strings(bprm->argc,bprm->argv,bprm->page,bprm->p);
2185     if (!bprm->p) {
2186         fprintf(stderr, "%s: %s\n", bprm->filename, strerror(E2BIG));
2187         exit(-1);
2188     }
2189 
2190     /* Do this so that we can load the interpreter, if need be.  We will
2191        change some of these later */
2192     bprm->p = setup_arg_pages(bprm->p, bprm, info);
2193 
2194     if (elf_interpreter) {
2195         load_elf_interp(elf_interpreter, &interp_info, bprm->buf);
2196 
2197         /* If the program interpreter is one of these two, then assume
2198            an iBCS2 image.  Otherwise assume a native linux image.  */
2199 
2200         if (strcmp(elf_interpreter, "/usr/lib/libc.so.1") == 0
2201             || strcmp(elf_interpreter, "/usr/lib/ld.so.1") == 0) {
2202             info->personality = PER_SVR4;
2203 
2204             /* Why this, you ask???  Well SVr4 maps page 0 as read-only,
2205                and some applications "depend" upon this behavior.  Since
2206                we do not have the power to recompile these, we emulate
2207                the SVr4 behavior.  Sigh.  */
2208             target_mmap(0, qemu_host_page_size, PROT_READ | PROT_EXEC,
2209                         MAP_FIXED | MAP_PRIVATE, -1, 0);
2210         }
2211     }
2212 
2213     bprm->p = create_elf_tables(bprm->p, bprm->argc, bprm->envc, &elf_ex,
2214                                 info, (elf_interpreter ? &interp_info : NULL));
2215     info->start_stack = bprm->p;
2216 
2217     /* If we have an interpreter, set that as the program's entry point.
2218        Copy the load_bias as well, to help PPC64 interpret the entry
2219        point as a function descriptor.  Do this after creating elf tables
2220        so that we copy the original program entry point into the AUXV.  */
2221     if (elf_interpreter) {
2222         info->load_bias = interp_info.load_bias;
2223         info->entry = interp_info.entry;
2224         free(elf_interpreter);
2225     }
2226 
2227 #ifdef USE_ELF_CORE_DUMP
2228     bprm->core_dump = &elf_core_dump;
2229 #endif
2230 
2231     return 0;
2232 }
2233 
2234 #ifdef USE_ELF_CORE_DUMP
2235 /*
2236  * Definitions to generate Intel SVR4-like core files.
2237  * These mostly have the same names as the SVR4 types with "target_elf_"
2238  * tacked on the front to prevent clashes with linux definitions,
2239  * and the typedef forms have been avoided.  This is mostly like
2240  * the SVR4 structure, but more Linuxy, with things that Linux does
2241  * not support and which gdb doesn't really use excluded.
2242  *
2243  * Fields we don't dump (their contents is zero) in linux-user qemu
2244  * are marked with XXX.
2245  *
2246  * Core dump code is copied from linux kernel (fs/binfmt_elf.c).
2247  *
2248  * Porting ELF coredump for target is (quite) simple process.  First you
2249  * define USE_ELF_CORE_DUMP in target ELF code (where init_thread() for
2250  * the target resides):
2251  *
2252  * #define USE_ELF_CORE_DUMP
2253  *
2254  * Next you define type of register set used for dumping.  ELF specification
2255  * says that it needs to be array of elf_greg_t that has size of ELF_NREG.
2256  *
2257  * typedef <target_regtype> target_elf_greg_t;
2258  * #define ELF_NREG <number of registers>
2259  * typedef taret_elf_greg_t target_elf_gregset_t[ELF_NREG];
2260  *
2261  * Last step is to implement target specific function that copies registers
2262  * from given cpu into just specified register set.  Prototype is:
2263  *
2264  * static void elf_core_copy_regs(taret_elf_gregset_t *regs,
2265  *                                const CPUArchState *env);
2266  *
2267  * Parameters:
2268  *     regs - copy register values into here (allocated and zeroed by caller)
2269  *     env - copy registers from here
2270  *
2271  * Example for ARM target is provided in this file.
2272  */
2273 
2274 /* An ELF note in memory */
2275 struct memelfnote {
2276     const char *name;
2277     size_t     namesz;
2278     size_t     namesz_rounded;
2279     int        type;
2280     size_t     datasz;
2281     size_t     datasz_rounded;
2282     void       *data;
2283     size_t     notesz;
2284 };
2285 
2286 struct target_elf_siginfo {
2287     abi_int    si_signo; /* signal number */
2288     abi_int    si_code;  /* extra code */
2289     abi_int    si_errno; /* errno */
2290 };
2291 
2292 struct target_elf_prstatus {
2293     struct target_elf_siginfo pr_info;      /* Info associated with signal */
2294     abi_short          pr_cursig;    /* Current signal */
2295     abi_ulong          pr_sigpend;   /* XXX */
2296     abi_ulong          pr_sighold;   /* XXX */
2297     target_pid_t       pr_pid;
2298     target_pid_t       pr_ppid;
2299     target_pid_t       pr_pgrp;
2300     target_pid_t       pr_sid;
2301     struct target_timeval pr_utime;  /* XXX User time */
2302     struct target_timeval pr_stime;  /* XXX System time */
2303     struct target_timeval pr_cutime; /* XXX Cumulative user time */
2304     struct target_timeval pr_cstime; /* XXX Cumulative system time */
2305     target_elf_gregset_t      pr_reg;       /* GP registers */
2306     abi_int            pr_fpvalid;   /* XXX */
2307 };
2308 
2309 #define ELF_PRARGSZ     (80) /* Number of chars for args */
2310 
2311 struct target_elf_prpsinfo {
2312     char         pr_state;       /* numeric process state */
2313     char         pr_sname;       /* char for pr_state */
2314     char         pr_zomb;        /* zombie */
2315     char         pr_nice;        /* nice val */
2316     abi_ulong    pr_flag;        /* flags */
2317     target_uid_t pr_uid;
2318     target_gid_t pr_gid;
2319     target_pid_t pr_pid, pr_ppid, pr_pgrp, pr_sid;
2320     /* Lots missing */
2321     char    pr_fname[16];           /* filename of executable */
2322     char    pr_psargs[ELF_PRARGSZ]; /* initial part of arg list */
2323 };
2324 
2325 /* Here is the structure in which status of each thread is captured. */
2326 struct elf_thread_status {
2327     QTAILQ_ENTRY(elf_thread_status)  ets_link;
2328     struct target_elf_prstatus prstatus;   /* NT_PRSTATUS */
2329 #if 0
2330     elf_fpregset_t fpu;             /* NT_PRFPREG */
2331     struct task_struct *thread;
2332     elf_fpxregset_t xfpu;           /* ELF_CORE_XFPREG_TYPE */
2333 #endif
2334     struct memelfnote notes[1];
2335     int num_notes;
2336 };
2337 
2338 struct elf_note_info {
2339     struct memelfnote   *notes;
2340     struct target_elf_prstatus *prstatus;  /* NT_PRSTATUS */
2341     struct target_elf_prpsinfo *psinfo;    /* NT_PRPSINFO */
2342 
2343     QTAILQ_HEAD(thread_list_head, elf_thread_status) thread_list;
2344 #if 0
2345     /*
2346      * Current version of ELF coredump doesn't support
2347      * dumping fp regs etc.
2348      */
2349     elf_fpregset_t *fpu;
2350     elf_fpxregset_t *xfpu;
2351     int thread_status_size;
2352 #endif
2353     int notes_size;
2354     int numnote;
2355 };
2356 
2357 struct vm_area_struct {
2358     abi_ulong   vma_start;  /* start vaddr of memory region */
2359     abi_ulong   vma_end;    /* end vaddr of memory region */
2360     abi_ulong   vma_flags;  /* protection etc. flags for the region */
2361     QTAILQ_ENTRY(vm_area_struct) vma_link;
2362 };
2363 
2364 struct mm_struct {
2365     QTAILQ_HEAD(, vm_area_struct) mm_mmap;
2366     int mm_count;           /* number of mappings */
2367 };
2368 
2369 static struct mm_struct *vma_init(void);
2370 static void vma_delete(struct mm_struct *);
2371 static int vma_add_mapping(struct mm_struct *, abi_ulong,
2372                            abi_ulong, abi_ulong);
2373 static int vma_get_mapping_count(const struct mm_struct *);
2374 static struct vm_area_struct *vma_first(const struct mm_struct *);
2375 static struct vm_area_struct *vma_next(struct vm_area_struct *);
2376 static abi_ulong vma_dump_size(const struct vm_area_struct *);
2377 static int vma_walker(void *priv, abi_ulong start, abi_ulong end,
2378                       unsigned long flags);
2379 
2380 static void fill_elf_header(struct elfhdr *, int, uint16_t, uint32_t);
2381 static void fill_note(struct memelfnote *, const char *, int,
2382                       unsigned int, void *);
2383 static void fill_prstatus(struct target_elf_prstatus *, const TaskState *, int);
2384 static int fill_psinfo(struct target_elf_prpsinfo *, const TaskState *);
2385 static void fill_auxv_note(struct memelfnote *, const TaskState *);
2386 static void fill_elf_note_phdr(struct elf_phdr *, int, off_t);
2387 static size_t note_size(const struct memelfnote *);
2388 static void free_note_info(struct elf_note_info *);
2389 static int fill_note_info(struct elf_note_info *, long, const CPUArchState *);
2390 static void fill_thread_info(struct elf_note_info *, const CPUArchState *);
2391 static int core_dump_filename(const TaskState *, char *, size_t);
2392 
2393 static int dump_write(int, const void *, size_t);
2394 static int write_note(struct memelfnote *, int);
2395 static int write_note_info(struct elf_note_info *, int);
2396 
2397 #ifdef BSWAP_NEEDED
2398 static void bswap_prstatus(struct target_elf_prstatus *prstatus)
2399 {
2400     prstatus->pr_info.si_signo = tswap32(prstatus->pr_info.si_signo);
2401     prstatus->pr_info.si_code = tswap32(prstatus->pr_info.si_code);
2402     prstatus->pr_info.si_errno = tswap32(prstatus->pr_info.si_errno);
2403     prstatus->pr_cursig = tswap16(prstatus->pr_cursig);
2404     prstatus->pr_sigpend = tswapal(prstatus->pr_sigpend);
2405     prstatus->pr_sighold = tswapal(prstatus->pr_sighold);
2406     prstatus->pr_pid = tswap32(prstatus->pr_pid);
2407     prstatus->pr_ppid = tswap32(prstatus->pr_ppid);
2408     prstatus->pr_pgrp = tswap32(prstatus->pr_pgrp);
2409     prstatus->pr_sid = tswap32(prstatus->pr_sid);
2410     /* cpu times are not filled, so we skip them */
2411     /* regs should be in correct format already */
2412     prstatus->pr_fpvalid = tswap32(prstatus->pr_fpvalid);
2413 }
2414 
2415 static void bswap_psinfo(struct target_elf_prpsinfo *psinfo)
2416 {
2417     psinfo->pr_flag = tswapal(psinfo->pr_flag);
2418     psinfo->pr_uid = tswap16(psinfo->pr_uid);
2419     psinfo->pr_gid = tswap16(psinfo->pr_gid);
2420     psinfo->pr_pid = tswap32(psinfo->pr_pid);
2421     psinfo->pr_ppid = tswap32(psinfo->pr_ppid);
2422     psinfo->pr_pgrp = tswap32(psinfo->pr_pgrp);
2423     psinfo->pr_sid = tswap32(psinfo->pr_sid);
2424 }
2425 
2426 static void bswap_note(struct elf_note *en)
2427 {
2428     bswap32s(&en->n_namesz);
2429     bswap32s(&en->n_descsz);
2430     bswap32s(&en->n_type);
2431 }
2432 #else
2433 static inline void bswap_prstatus(struct target_elf_prstatus *p) { }
2434 static inline void bswap_psinfo(struct target_elf_prpsinfo *p) {}
2435 static inline void bswap_note(struct elf_note *en) { }
2436 #endif /* BSWAP_NEEDED */
2437 
2438 /*
2439  * Minimal support for linux memory regions.  These are needed
2440  * when we are finding out what memory exactly belongs to
2441  * emulated process.  No locks needed here, as long as
2442  * thread that received the signal is stopped.
2443  */
2444 
2445 static struct mm_struct *vma_init(void)
2446 {
2447     struct mm_struct *mm;
2448 
2449     if ((mm = g_malloc(sizeof (*mm))) == NULL)
2450         return (NULL);
2451 
2452     mm->mm_count = 0;
2453     QTAILQ_INIT(&mm->mm_mmap);
2454 
2455     return (mm);
2456 }
2457 
2458 static void vma_delete(struct mm_struct *mm)
2459 {
2460     struct vm_area_struct *vma;
2461 
2462     while ((vma = vma_first(mm)) != NULL) {
2463         QTAILQ_REMOVE(&mm->mm_mmap, vma, vma_link);
2464         g_free(vma);
2465     }
2466     g_free(mm);
2467 }
2468 
2469 static int vma_add_mapping(struct mm_struct *mm, abi_ulong start,
2470                            abi_ulong end, abi_ulong flags)
2471 {
2472     struct vm_area_struct *vma;
2473 
2474     if ((vma = g_malloc0(sizeof (*vma))) == NULL)
2475         return (-1);
2476 
2477     vma->vma_start = start;
2478     vma->vma_end = end;
2479     vma->vma_flags = flags;
2480 
2481     QTAILQ_INSERT_TAIL(&mm->mm_mmap, vma, vma_link);
2482     mm->mm_count++;
2483 
2484     return (0);
2485 }
2486 
2487 static struct vm_area_struct *vma_first(const struct mm_struct *mm)
2488 {
2489     return (QTAILQ_FIRST(&mm->mm_mmap));
2490 }
2491 
2492 static struct vm_area_struct *vma_next(struct vm_area_struct *vma)
2493 {
2494     return (QTAILQ_NEXT(vma, vma_link));
2495 }
2496 
2497 static int vma_get_mapping_count(const struct mm_struct *mm)
2498 {
2499     return (mm->mm_count);
2500 }
2501 
2502 /*
2503  * Calculate file (dump) size of given memory region.
2504  */
2505 static abi_ulong vma_dump_size(const struct vm_area_struct *vma)
2506 {
2507     /* if we cannot even read the first page, skip it */
2508     if (!access_ok(VERIFY_READ, vma->vma_start, TARGET_PAGE_SIZE))
2509         return (0);
2510 
2511     /*
2512      * Usually we don't dump executable pages as they contain
2513      * non-writable code that debugger can read directly from
2514      * target library etc.  However, thread stacks are marked
2515      * also executable so we read in first page of given region
2516      * and check whether it contains elf header.  If there is
2517      * no elf header, we dump it.
2518      */
2519     if (vma->vma_flags & PROT_EXEC) {
2520         char page[TARGET_PAGE_SIZE];
2521 
2522         copy_from_user(page, vma->vma_start, sizeof (page));
2523         if ((page[EI_MAG0] == ELFMAG0) &&
2524             (page[EI_MAG1] == ELFMAG1) &&
2525             (page[EI_MAG2] == ELFMAG2) &&
2526             (page[EI_MAG3] == ELFMAG3)) {
2527             /*
2528              * Mappings are possibly from ELF binary.  Don't dump
2529              * them.
2530              */
2531             return (0);
2532         }
2533     }
2534 
2535     return (vma->vma_end - vma->vma_start);
2536 }
2537 
2538 static int vma_walker(void *priv, abi_ulong start, abi_ulong end,
2539                       unsigned long flags)
2540 {
2541     struct mm_struct *mm = (struct mm_struct *)priv;
2542 
2543     vma_add_mapping(mm, start, end, flags);
2544     return (0);
2545 }
2546 
2547 static void fill_note(struct memelfnote *note, const char *name, int type,
2548                       unsigned int sz, void *data)
2549 {
2550     unsigned int namesz;
2551 
2552     namesz = strlen(name) + 1;
2553     note->name = name;
2554     note->namesz = namesz;
2555     note->namesz_rounded = roundup(namesz, sizeof (int32_t));
2556     note->type = type;
2557     note->datasz = sz;
2558     note->datasz_rounded = roundup(sz, sizeof (int32_t));
2559 
2560     note->data = data;
2561 
2562     /*
2563      * We calculate rounded up note size here as specified by
2564      * ELF document.
2565      */
2566     note->notesz = sizeof (struct elf_note) +
2567         note->namesz_rounded + note->datasz_rounded;
2568 }
2569 
2570 static void fill_elf_header(struct elfhdr *elf, int segs, uint16_t machine,
2571                             uint32_t flags)
2572 {
2573     (void) memset(elf, 0, sizeof(*elf));
2574 
2575     (void) memcpy(elf->e_ident, ELFMAG, SELFMAG);
2576     elf->e_ident[EI_CLASS] = ELF_CLASS;
2577     elf->e_ident[EI_DATA] = ELF_DATA;
2578     elf->e_ident[EI_VERSION] = EV_CURRENT;
2579     elf->e_ident[EI_OSABI] = ELF_OSABI;
2580 
2581     elf->e_type = ET_CORE;
2582     elf->e_machine = machine;
2583     elf->e_version = EV_CURRENT;
2584     elf->e_phoff = sizeof(struct elfhdr);
2585     elf->e_flags = flags;
2586     elf->e_ehsize = sizeof(struct elfhdr);
2587     elf->e_phentsize = sizeof(struct elf_phdr);
2588     elf->e_phnum = segs;
2589 
2590     bswap_ehdr(elf);
2591 }
2592 
2593 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, off_t offset)
2594 {
2595     phdr->p_type = PT_NOTE;
2596     phdr->p_offset = offset;
2597     phdr->p_vaddr = 0;
2598     phdr->p_paddr = 0;
2599     phdr->p_filesz = sz;
2600     phdr->p_memsz = 0;
2601     phdr->p_flags = 0;
2602     phdr->p_align = 0;
2603 
2604     bswap_phdr(phdr, 1);
2605 }
2606 
2607 static size_t note_size(const struct memelfnote *note)
2608 {
2609     return (note->notesz);
2610 }
2611 
2612 static void fill_prstatus(struct target_elf_prstatus *prstatus,
2613                           const TaskState *ts, int signr)
2614 {
2615     (void) memset(prstatus, 0, sizeof (*prstatus));
2616     prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
2617     prstatus->pr_pid = ts->ts_tid;
2618     prstatus->pr_ppid = getppid();
2619     prstatus->pr_pgrp = getpgrp();
2620     prstatus->pr_sid = getsid(0);
2621 
2622     bswap_prstatus(prstatus);
2623 }
2624 
2625 static int fill_psinfo(struct target_elf_prpsinfo *psinfo, const TaskState *ts)
2626 {
2627     char *base_filename;
2628     unsigned int i, len;
2629 
2630     (void) memset(psinfo, 0, sizeof (*psinfo));
2631 
2632     len = ts->info->arg_end - ts->info->arg_start;
2633     if (len >= ELF_PRARGSZ)
2634         len = ELF_PRARGSZ - 1;
2635     if (copy_from_user(&psinfo->pr_psargs, ts->info->arg_start, len))
2636         return -EFAULT;
2637     for (i = 0; i < len; i++)
2638         if (psinfo->pr_psargs[i] == 0)
2639             psinfo->pr_psargs[i] = ' ';
2640     psinfo->pr_psargs[len] = 0;
2641 
2642     psinfo->pr_pid = getpid();
2643     psinfo->pr_ppid = getppid();
2644     psinfo->pr_pgrp = getpgrp();
2645     psinfo->pr_sid = getsid(0);
2646     psinfo->pr_uid = getuid();
2647     psinfo->pr_gid = getgid();
2648 
2649     base_filename = g_path_get_basename(ts->bprm->filename);
2650     /*
2651      * Using strncpy here is fine: at max-length,
2652      * this field is not NUL-terminated.
2653      */
2654     (void) strncpy(psinfo->pr_fname, base_filename,
2655                    sizeof(psinfo->pr_fname));
2656 
2657     g_free(base_filename);
2658     bswap_psinfo(psinfo);
2659     return (0);
2660 }
2661 
2662 static void fill_auxv_note(struct memelfnote *note, const TaskState *ts)
2663 {
2664     elf_addr_t auxv = (elf_addr_t)ts->info->saved_auxv;
2665     elf_addr_t orig_auxv = auxv;
2666     void *ptr;
2667     int len = ts->info->auxv_len;
2668 
2669     /*
2670      * Auxiliary vector is stored in target process stack.  It contains
2671      * {type, value} pairs that we need to dump into note.  This is not
2672      * strictly necessary but we do it here for sake of completeness.
2673      */
2674 
2675     /* read in whole auxv vector and copy it to memelfnote */
2676     ptr = lock_user(VERIFY_READ, orig_auxv, len, 0);
2677     if (ptr != NULL) {
2678         fill_note(note, "CORE", NT_AUXV, len, ptr);
2679         unlock_user(ptr, auxv, len);
2680     }
2681 }
2682 
2683 /*
2684  * Constructs name of coredump file.  We have following convention
2685  * for the name:
2686  *     qemu_<basename-of-target-binary>_<date>-<time>_<pid>.core
2687  *
2688  * Returns 0 in case of success, -1 otherwise (errno is set).
2689  */
2690 static int core_dump_filename(const TaskState *ts, char *buf,
2691                               size_t bufsize)
2692 {
2693     char timestamp[64];
2694     char *filename = NULL;
2695     char *base_filename = NULL;
2696     struct timeval tv;
2697     struct tm tm;
2698 
2699     assert(bufsize >= PATH_MAX);
2700 
2701     if (gettimeofday(&tv, NULL) < 0) {
2702         (void) fprintf(stderr, "unable to get current timestamp: %s",
2703                        strerror(errno));
2704         return (-1);
2705     }
2706 
2707     filename = strdup(ts->bprm->filename);
2708     base_filename = strdup(basename(filename));
2709     (void) strftime(timestamp, sizeof (timestamp), "%Y%m%d-%H%M%S",
2710                     localtime_r(&tv.tv_sec, &tm));
2711     (void) snprintf(buf, bufsize, "qemu_%s_%s_%d.core",
2712                     base_filename, timestamp, (int)getpid());
2713     free(base_filename);
2714     free(filename);
2715 
2716     return (0);
2717 }
2718 
2719 static int dump_write(int fd, const void *ptr, size_t size)
2720 {
2721     const char *bufp = (const char *)ptr;
2722     ssize_t bytes_written, bytes_left;
2723     struct rlimit dumpsize;
2724     off_t pos;
2725 
2726     bytes_written = 0;
2727     getrlimit(RLIMIT_CORE, &dumpsize);
2728     if ((pos = lseek(fd, 0, SEEK_CUR))==-1) {
2729         if (errno == ESPIPE) { /* not a seekable stream */
2730             bytes_left = size;
2731         } else {
2732             return pos;
2733         }
2734     } else {
2735         if (dumpsize.rlim_cur <= pos) {
2736             return -1;
2737         } else if (dumpsize.rlim_cur == RLIM_INFINITY) {
2738             bytes_left = size;
2739         } else {
2740             size_t limit_left=dumpsize.rlim_cur - pos;
2741             bytes_left = limit_left >= size ? size : limit_left ;
2742         }
2743     }
2744 
2745     /*
2746      * In normal conditions, single write(2) should do but
2747      * in case of socket etc. this mechanism is more portable.
2748      */
2749     do {
2750         bytes_written = write(fd, bufp, bytes_left);
2751         if (bytes_written < 0) {
2752             if (errno == EINTR)
2753                 continue;
2754             return (-1);
2755         } else if (bytes_written == 0) { /* eof */
2756             return (-1);
2757         }
2758         bufp += bytes_written;
2759         bytes_left -= bytes_written;
2760     } while (bytes_left > 0);
2761 
2762     return (0);
2763 }
2764 
2765 static int write_note(struct memelfnote *men, int fd)
2766 {
2767     struct elf_note en;
2768 
2769     en.n_namesz = men->namesz;
2770     en.n_type = men->type;
2771     en.n_descsz = men->datasz;
2772 
2773     bswap_note(&en);
2774 
2775     if (dump_write(fd, &en, sizeof(en)) != 0)
2776         return (-1);
2777     if (dump_write(fd, men->name, men->namesz_rounded) != 0)
2778         return (-1);
2779     if (dump_write(fd, men->data, men->datasz_rounded) != 0)
2780         return (-1);
2781 
2782     return (0);
2783 }
2784 
2785 static void fill_thread_info(struct elf_note_info *info, const CPUArchState *env)
2786 {
2787     CPUState *cpu = ENV_GET_CPU((CPUArchState *)env);
2788     TaskState *ts = (TaskState *)cpu->opaque;
2789     struct elf_thread_status *ets;
2790 
2791     ets = g_malloc0(sizeof (*ets));
2792     ets->num_notes = 1; /* only prstatus is dumped */
2793     fill_prstatus(&ets->prstatus, ts, 0);
2794     elf_core_copy_regs(&ets->prstatus.pr_reg, env);
2795     fill_note(&ets->notes[0], "CORE", NT_PRSTATUS, sizeof (ets->prstatus),
2796               &ets->prstatus);
2797 
2798     QTAILQ_INSERT_TAIL(&info->thread_list, ets, ets_link);
2799 
2800     info->notes_size += note_size(&ets->notes[0]);
2801 }
2802 
2803 static void init_note_info(struct elf_note_info *info)
2804 {
2805     /* Initialize the elf_note_info structure so that it is at
2806      * least safe to call free_note_info() on it. Must be
2807      * called before calling fill_note_info().
2808      */
2809     memset(info, 0, sizeof (*info));
2810     QTAILQ_INIT(&info->thread_list);
2811 }
2812 
2813 static int fill_note_info(struct elf_note_info *info,
2814                           long signr, const CPUArchState *env)
2815 {
2816 #define NUMNOTES 3
2817     CPUState *cpu = ENV_GET_CPU((CPUArchState *)env);
2818     TaskState *ts = (TaskState *)cpu->opaque;
2819     int i;
2820 
2821     info->notes = g_malloc0(NUMNOTES * sizeof (struct memelfnote));
2822     if (info->notes == NULL)
2823         return (-ENOMEM);
2824     info->prstatus = g_malloc0(sizeof (*info->prstatus));
2825     if (info->prstatus == NULL)
2826         return (-ENOMEM);
2827     info->psinfo = g_malloc0(sizeof (*info->psinfo));
2828     if (info->prstatus == NULL)
2829         return (-ENOMEM);
2830 
2831     /*
2832      * First fill in status (and registers) of current thread
2833      * including process info & aux vector.
2834      */
2835     fill_prstatus(info->prstatus, ts, signr);
2836     elf_core_copy_regs(&info->prstatus->pr_reg, env);
2837     fill_note(&info->notes[0], "CORE", NT_PRSTATUS,
2838               sizeof (*info->prstatus), info->prstatus);
2839     fill_psinfo(info->psinfo, ts);
2840     fill_note(&info->notes[1], "CORE", NT_PRPSINFO,
2841               sizeof (*info->psinfo), info->psinfo);
2842     fill_auxv_note(&info->notes[2], ts);
2843     info->numnote = 3;
2844 
2845     info->notes_size = 0;
2846     for (i = 0; i < info->numnote; i++)
2847         info->notes_size += note_size(&info->notes[i]);
2848 
2849     /* read and fill status of all threads */
2850     cpu_list_lock();
2851     CPU_FOREACH(cpu) {
2852         if (cpu == thread_cpu) {
2853             continue;
2854         }
2855         fill_thread_info(info, (CPUArchState *)cpu->env_ptr);
2856     }
2857     cpu_list_unlock();
2858 
2859     return (0);
2860 }
2861 
2862 static void free_note_info(struct elf_note_info *info)
2863 {
2864     struct elf_thread_status *ets;
2865 
2866     while (!QTAILQ_EMPTY(&info->thread_list)) {
2867         ets = QTAILQ_FIRST(&info->thread_list);
2868         QTAILQ_REMOVE(&info->thread_list, ets, ets_link);
2869         g_free(ets);
2870     }
2871 
2872     g_free(info->prstatus);
2873     g_free(info->psinfo);
2874     g_free(info->notes);
2875 }
2876 
2877 static int write_note_info(struct elf_note_info *info, int fd)
2878 {
2879     struct elf_thread_status *ets;
2880     int i, error = 0;
2881 
2882     /* write prstatus, psinfo and auxv for current thread */
2883     for (i = 0; i < info->numnote; i++)
2884         if ((error = write_note(&info->notes[i], fd)) != 0)
2885             return (error);
2886 
2887     /* write prstatus for each thread */
2888     for (ets = info->thread_list.tqh_first; ets != NULL;
2889          ets = ets->ets_link.tqe_next) {
2890         if ((error = write_note(&ets->notes[0], fd)) != 0)
2891             return (error);
2892     }
2893 
2894     return (0);
2895 }
2896 
2897 /*
2898  * Write out ELF coredump.
2899  *
2900  * See documentation of ELF object file format in:
2901  * http://www.caldera.com/developers/devspecs/gabi41.pdf
2902  *
2903  * Coredump format in linux is following:
2904  *
2905  * 0   +----------------------+         \
2906  *     | ELF header           | ET_CORE  |
2907  *     +----------------------+          |
2908  *     | ELF program headers  |          |--- headers
2909  *     | - NOTE section       |          |
2910  *     | - PT_LOAD sections   |          |
2911  *     +----------------------+         /
2912  *     | NOTEs:               |
2913  *     | - NT_PRSTATUS        |
2914  *     | - NT_PRSINFO         |
2915  *     | - NT_AUXV            |
2916  *     +----------------------+ <-- aligned to target page
2917  *     | Process memory dump  |
2918  *     :                      :
2919  *     .                      .
2920  *     :                      :
2921  *     |                      |
2922  *     +----------------------+
2923  *
2924  * NT_PRSTATUS -> struct elf_prstatus (per thread)
2925  * NT_PRSINFO  -> struct elf_prpsinfo
2926  * NT_AUXV is array of { type, value } pairs (see fill_auxv_note()).
2927  *
2928  * Format follows System V format as close as possible.  Current
2929  * version limitations are as follows:
2930  *     - no floating point registers are dumped
2931  *
2932  * Function returns 0 in case of success, negative errno otherwise.
2933  *
2934  * TODO: make this work also during runtime: it should be
2935  * possible to force coredump from running process and then
2936  * continue processing.  For example qemu could set up SIGUSR2
2937  * handler (provided that target process haven't registered
2938  * handler for that) that does the dump when signal is received.
2939  */
2940 static int elf_core_dump(int signr, const CPUArchState *env)
2941 {
2942     const CPUState *cpu = ENV_GET_CPU((CPUArchState *)env);
2943     const TaskState *ts = (const TaskState *)cpu->opaque;
2944     struct vm_area_struct *vma = NULL;
2945     char corefile[PATH_MAX];
2946     struct elf_note_info info;
2947     struct elfhdr elf;
2948     struct elf_phdr phdr;
2949     struct rlimit dumpsize;
2950     struct mm_struct *mm = NULL;
2951     off_t offset = 0, data_offset = 0;
2952     int segs = 0;
2953     int fd = -1;
2954 
2955     init_note_info(&info);
2956 
2957     errno = 0;
2958     getrlimit(RLIMIT_CORE, &dumpsize);
2959     if (dumpsize.rlim_cur == 0)
2960         return 0;
2961 
2962     if (core_dump_filename(ts, corefile, sizeof (corefile)) < 0)
2963         return (-errno);
2964 
2965     if ((fd = open(corefile, O_WRONLY | O_CREAT,
2966                    S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH)) < 0)
2967         return (-errno);
2968 
2969     /*
2970      * Walk through target process memory mappings and
2971      * set up structure containing this information.  After
2972      * this point vma_xxx functions can be used.
2973      */
2974     if ((mm = vma_init()) == NULL)
2975         goto out;
2976 
2977     walk_memory_regions(mm, vma_walker);
2978     segs = vma_get_mapping_count(mm);
2979 
2980     /*
2981      * Construct valid coredump ELF header.  We also
2982      * add one more segment for notes.
2983      */
2984     fill_elf_header(&elf, segs + 1, ELF_MACHINE, 0);
2985     if (dump_write(fd, &elf, sizeof (elf)) != 0)
2986         goto out;
2987 
2988     /* fill in in-memory version of notes */
2989     if (fill_note_info(&info, signr, env) < 0)
2990         goto out;
2991 
2992     offset += sizeof (elf);                             /* elf header */
2993     offset += (segs + 1) * sizeof (struct elf_phdr);    /* program headers */
2994 
2995     /* write out notes program header */
2996     fill_elf_note_phdr(&phdr, info.notes_size, offset);
2997 
2998     offset += info.notes_size;
2999     if (dump_write(fd, &phdr, sizeof (phdr)) != 0)
3000         goto out;
3001 
3002     /*
3003      * ELF specification wants data to start at page boundary so
3004      * we align it here.
3005      */
3006     data_offset = offset = roundup(offset, ELF_EXEC_PAGESIZE);
3007 
3008     /*
3009      * Write program headers for memory regions mapped in
3010      * the target process.
3011      */
3012     for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) {
3013         (void) memset(&phdr, 0, sizeof (phdr));
3014 
3015         phdr.p_type = PT_LOAD;
3016         phdr.p_offset = offset;
3017         phdr.p_vaddr = vma->vma_start;
3018         phdr.p_paddr = 0;
3019         phdr.p_filesz = vma_dump_size(vma);
3020         offset += phdr.p_filesz;
3021         phdr.p_memsz = vma->vma_end - vma->vma_start;
3022         phdr.p_flags = vma->vma_flags & PROT_READ ? PF_R : 0;
3023         if (vma->vma_flags & PROT_WRITE)
3024             phdr.p_flags |= PF_W;
3025         if (vma->vma_flags & PROT_EXEC)
3026             phdr.p_flags |= PF_X;
3027         phdr.p_align = ELF_EXEC_PAGESIZE;
3028 
3029         bswap_phdr(&phdr, 1);
3030         dump_write(fd, &phdr, sizeof (phdr));
3031     }
3032 
3033     /*
3034      * Next we write notes just after program headers.  No
3035      * alignment needed here.
3036      */
3037     if (write_note_info(&info, fd) < 0)
3038         goto out;
3039 
3040     /* align data to page boundary */
3041     if (lseek(fd, data_offset, SEEK_SET) != data_offset)
3042         goto out;
3043 
3044     /*
3045      * Finally we can dump process memory into corefile as well.
3046      */
3047     for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) {
3048         abi_ulong addr;
3049         abi_ulong end;
3050 
3051         end = vma->vma_start + vma_dump_size(vma);
3052 
3053         for (addr = vma->vma_start; addr < end;
3054              addr += TARGET_PAGE_SIZE) {
3055             char page[TARGET_PAGE_SIZE];
3056             int error;
3057 
3058             /*
3059              *  Read in page from target process memory and
3060              *  write it to coredump file.
3061              */
3062             error = copy_from_user(page, addr, sizeof (page));
3063             if (error != 0) {
3064                 (void) fprintf(stderr, "unable to dump " TARGET_ABI_FMT_lx "\n",
3065                                addr);
3066                 errno = -error;
3067                 goto out;
3068             }
3069             if (dump_write(fd, page, TARGET_PAGE_SIZE) < 0)
3070                 goto out;
3071         }
3072     }
3073 
3074  out:
3075     free_note_info(&info);
3076     if (mm != NULL)
3077         vma_delete(mm);
3078     (void) close(fd);
3079 
3080     if (errno != 0)
3081         return (-errno);
3082     return (0);
3083 }
3084 #endif /* USE_ELF_CORE_DUMP */
3085 
3086 void do_init_thread(struct target_pt_regs *regs, struct image_info *infop)
3087 {
3088     init_thread(regs, infop);
3089 }
3090