xref: /openbmc/qemu/linux-user/elfload.c (revision eab15862)
1 /* This is the Linux kernel elf-loading code, ported into user space */
2 #include "qemu/osdep.h"
3 #include <sys/param.h>
4 
5 #include <sys/resource.h>
6 
7 #include "qemu.h"
8 #include "disas/disas.h"
9 #include "qemu/path.h"
10 
11 #ifdef _ARCH_PPC64
12 #undef ARCH_DLINFO
13 #undef ELF_PLATFORM
14 #undef ELF_HWCAP
15 #undef ELF_HWCAP2
16 #undef ELF_CLASS
17 #undef ELF_DATA
18 #undef ELF_ARCH
19 #endif
20 
21 #define ELF_OSABI   ELFOSABI_SYSV
22 
23 /* from personality.h */
24 
25 /*
26  * Flags for bug emulation.
27  *
28  * These occupy the top three bytes.
29  */
30 enum {
31     ADDR_NO_RANDOMIZE = 0x0040000,      /* disable randomization of VA space */
32     FDPIC_FUNCPTRS =    0x0080000,      /* userspace function ptrs point to
33                                            descriptors (signal handling) */
34     MMAP_PAGE_ZERO =    0x0100000,
35     ADDR_COMPAT_LAYOUT = 0x0200000,
36     READ_IMPLIES_EXEC = 0x0400000,
37     ADDR_LIMIT_32BIT =  0x0800000,
38     SHORT_INODE =       0x1000000,
39     WHOLE_SECONDS =     0x2000000,
40     STICKY_TIMEOUTS =   0x4000000,
41     ADDR_LIMIT_3GB =    0x8000000,
42 };
43 
44 /*
45  * Personality types.
46  *
47  * These go in the low byte.  Avoid using the top bit, it will
48  * conflict with error returns.
49  */
50 enum {
51     PER_LINUX =         0x0000,
52     PER_LINUX_32BIT =   0x0000 | ADDR_LIMIT_32BIT,
53     PER_LINUX_FDPIC =   0x0000 | FDPIC_FUNCPTRS,
54     PER_SVR4 =          0x0001 | STICKY_TIMEOUTS | MMAP_PAGE_ZERO,
55     PER_SVR3 =          0x0002 | STICKY_TIMEOUTS | SHORT_INODE,
56     PER_SCOSVR3 =       0x0003 | STICKY_TIMEOUTS | WHOLE_SECONDS | SHORT_INODE,
57     PER_OSR5 =          0x0003 | STICKY_TIMEOUTS | WHOLE_SECONDS,
58     PER_WYSEV386 =      0x0004 | STICKY_TIMEOUTS | SHORT_INODE,
59     PER_ISCR4 =         0x0005 | STICKY_TIMEOUTS,
60     PER_BSD =           0x0006,
61     PER_SUNOS =         0x0006 | STICKY_TIMEOUTS,
62     PER_XENIX =         0x0007 | STICKY_TIMEOUTS | SHORT_INODE,
63     PER_LINUX32 =       0x0008,
64     PER_LINUX32_3GB =   0x0008 | ADDR_LIMIT_3GB,
65     PER_IRIX32 =        0x0009 | STICKY_TIMEOUTS,/* IRIX5 32-bit */
66     PER_IRIXN32 =       0x000a | STICKY_TIMEOUTS,/* IRIX6 new 32-bit */
67     PER_IRIX64 =        0x000b | STICKY_TIMEOUTS,/* IRIX6 64-bit */
68     PER_RISCOS =        0x000c,
69     PER_SOLARIS =       0x000d | STICKY_TIMEOUTS,
70     PER_UW7 =           0x000e | STICKY_TIMEOUTS | MMAP_PAGE_ZERO,
71     PER_OSF4 =          0x000f,                  /* OSF/1 v4 */
72     PER_HPUX =          0x0010,
73     PER_MASK =          0x00ff,
74 };
75 
76 /*
77  * Return the base personality without flags.
78  */
79 #define personality(pers)       (pers & PER_MASK)
80 
81 /* this flag is uneffective under linux too, should be deleted */
82 #ifndef MAP_DENYWRITE
83 #define MAP_DENYWRITE 0
84 #endif
85 
86 /* should probably go in elf.h */
87 #ifndef ELIBBAD
88 #define ELIBBAD 80
89 #endif
90 
91 #ifdef TARGET_WORDS_BIGENDIAN
92 #define ELF_DATA        ELFDATA2MSB
93 #else
94 #define ELF_DATA        ELFDATA2LSB
95 #endif
96 
97 #ifdef TARGET_ABI_MIPSN32
98 typedef abi_ullong      target_elf_greg_t;
99 #define tswapreg(ptr)   tswap64(ptr)
100 #else
101 typedef abi_ulong       target_elf_greg_t;
102 #define tswapreg(ptr)   tswapal(ptr)
103 #endif
104 
105 #ifdef USE_UID16
106 typedef abi_ushort      target_uid_t;
107 typedef abi_ushort      target_gid_t;
108 #else
109 typedef abi_uint        target_uid_t;
110 typedef abi_uint        target_gid_t;
111 #endif
112 typedef abi_int         target_pid_t;
113 
114 #ifdef TARGET_I386
115 
116 #define ELF_PLATFORM get_elf_platform()
117 
118 static const char *get_elf_platform(void)
119 {
120     static char elf_platform[] = "i386";
121     int family = object_property_get_int(OBJECT(thread_cpu), "family", NULL);
122     if (family > 6)
123         family = 6;
124     if (family >= 3)
125         elf_platform[1] = '0' + family;
126     return elf_platform;
127 }
128 
129 #define ELF_HWCAP get_elf_hwcap()
130 
131 static uint32_t get_elf_hwcap(void)
132 {
133     X86CPU *cpu = X86_CPU(thread_cpu);
134 
135     return cpu->env.features[FEAT_1_EDX];
136 }
137 
138 #ifdef TARGET_X86_64
139 #define ELF_START_MMAP 0x2aaaaab000ULL
140 
141 #define ELF_CLASS      ELFCLASS64
142 #define ELF_ARCH       EM_X86_64
143 
144 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop)
145 {
146     regs->rax = 0;
147     regs->rsp = infop->start_stack;
148     regs->rip = infop->entry;
149 }
150 
151 #define ELF_NREG    27
152 typedef target_elf_greg_t  target_elf_gregset_t[ELF_NREG];
153 
154 /*
155  * Note that ELF_NREG should be 29 as there should be place for
156  * TRAPNO and ERR "registers" as well but linux doesn't dump
157  * those.
158  *
159  * See linux kernel: arch/x86/include/asm/elf.h
160  */
161 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUX86State *env)
162 {
163     (*regs)[0] = env->regs[15];
164     (*regs)[1] = env->regs[14];
165     (*regs)[2] = env->regs[13];
166     (*regs)[3] = env->regs[12];
167     (*regs)[4] = env->regs[R_EBP];
168     (*regs)[5] = env->regs[R_EBX];
169     (*regs)[6] = env->regs[11];
170     (*regs)[7] = env->regs[10];
171     (*regs)[8] = env->regs[9];
172     (*regs)[9] = env->regs[8];
173     (*regs)[10] = env->regs[R_EAX];
174     (*regs)[11] = env->regs[R_ECX];
175     (*regs)[12] = env->regs[R_EDX];
176     (*regs)[13] = env->regs[R_ESI];
177     (*regs)[14] = env->regs[R_EDI];
178     (*regs)[15] = env->regs[R_EAX]; /* XXX */
179     (*regs)[16] = env->eip;
180     (*regs)[17] = env->segs[R_CS].selector & 0xffff;
181     (*regs)[18] = env->eflags;
182     (*regs)[19] = env->regs[R_ESP];
183     (*regs)[20] = env->segs[R_SS].selector & 0xffff;
184     (*regs)[21] = env->segs[R_FS].selector & 0xffff;
185     (*regs)[22] = env->segs[R_GS].selector & 0xffff;
186     (*regs)[23] = env->segs[R_DS].selector & 0xffff;
187     (*regs)[24] = env->segs[R_ES].selector & 0xffff;
188     (*regs)[25] = env->segs[R_FS].selector & 0xffff;
189     (*regs)[26] = env->segs[R_GS].selector & 0xffff;
190 }
191 
192 #else
193 
194 #define ELF_START_MMAP 0x80000000
195 
196 /*
197  * This is used to ensure we don't load something for the wrong architecture.
198  */
199 #define elf_check_arch(x) ( ((x) == EM_386) || ((x) == EM_486) )
200 
201 /*
202  * These are used to set parameters in the core dumps.
203  */
204 #define ELF_CLASS       ELFCLASS32
205 #define ELF_ARCH        EM_386
206 
207 static inline void init_thread(struct target_pt_regs *regs,
208                                struct image_info *infop)
209 {
210     regs->esp = infop->start_stack;
211     regs->eip = infop->entry;
212 
213     /* SVR4/i386 ABI (pages 3-31, 3-32) says that when the program
214        starts %edx contains a pointer to a function which might be
215        registered using `atexit'.  This provides a mean for the
216        dynamic linker to call DT_FINI functions for shared libraries
217        that have been loaded before the code runs.
218 
219        A value of 0 tells we have no such handler.  */
220     regs->edx = 0;
221 }
222 
223 #define ELF_NREG    17
224 typedef target_elf_greg_t  target_elf_gregset_t[ELF_NREG];
225 
226 /*
227  * Note that ELF_NREG should be 19 as there should be place for
228  * TRAPNO and ERR "registers" as well but linux doesn't dump
229  * those.
230  *
231  * See linux kernel: arch/x86/include/asm/elf.h
232  */
233 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUX86State *env)
234 {
235     (*regs)[0] = env->regs[R_EBX];
236     (*regs)[1] = env->regs[R_ECX];
237     (*regs)[2] = env->regs[R_EDX];
238     (*regs)[3] = env->regs[R_ESI];
239     (*regs)[4] = env->regs[R_EDI];
240     (*regs)[5] = env->regs[R_EBP];
241     (*regs)[6] = env->regs[R_EAX];
242     (*regs)[7] = env->segs[R_DS].selector & 0xffff;
243     (*regs)[8] = env->segs[R_ES].selector & 0xffff;
244     (*regs)[9] = env->segs[R_FS].selector & 0xffff;
245     (*regs)[10] = env->segs[R_GS].selector & 0xffff;
246     (*regs)[11] = env->regs[R_EAX]; /* XXX */
247     (*regs)[12] = env->eip;
248     (*regs)[13] = env->segs[R_CS].selector & 0xffff;
249     (*regs)[14] = env->eflags;
250     (*regs)[15] = env->regs[R_ESP];
251     (*regs)[16] = env->segs[R_SS].selector & 0xffff;
252 }
253 #endif
254 
255 #define USE_ELF_CORE_DUMP
256 #define ELF_EXEC_PAGESIZE       4096
257 
258 #endif
259 
260 #ifdef TARGET_ARM
261 
262 #ifndef TARGET_AARCH64
263 /* 32 bit ARM definitions */
264 
265 #define ELF_START_MMAP 0x80000000
266 
267 #define ELF_ARCH        EM_ARM
268 #define ELF_CLASS       ELFCLASS32
269 
270 static inline void init_thread(struct target_pt_regs *regs,
271                                struct image_info *infop)
272 {
273     abi_long stack = infop->start_stack;
274     memset(regs, 0, sizeof(*regs));
275 
276     regs->uregs[16] = ARM_CPU_MODE_USR;
277     if (infop->entry & 1) {
278         regs->uregs[16] |= CPSR_T;
279     }
280     regs->uregs[15] = infop->entry & 0xfffffffe;
281     regs->uregs[13] = infop->start_stack;
282     /* FIXME - what to for failure of get_user()? */
283     get_user_ual(regs->uregs[2], stack + 8); /* envp */
284     get_user_ual(regs->uregs[1], stack + 4); /* envp */
285     /* XXX: it seems that r0 is zeroed after ! */
286     regs->uregs[0] = 0;
287     /* For uClinux PIC binaries.  */
288     /* XXX: Linux does this only on ARM with no MMU (do we care ?) */
289     regs->uregs[10] = infop->start_data;
290 }
291 
292 #define ELF_NREG    18
293 typedef target_elf_greg_t  target_elf_gregset_t[ELF_NREG];
294 
295 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUARMState *env)
296 {
297     (*regs)[0] = tswapreg(env->regs[0]);
298     (*regs)[1] = tswapreg(env->regs[1]);
299     (*regs)[2] = tswapreg(env->regs[2]);
300     (*regs)[3] = tswapreg(env->regs[3]);
301     (*regs)[4] = tswapreg(env->regs[4]);
302     (*regs)[5] = tswapreg(env->regs[5]);
303     (*regs)[6] = tswapreg(env->regs[6]);
304     (*regs)[7] = tswapreg(env->regs[7]);
305     (*regs)[8] = tswapreg(env->regs[8]);
306     (*regs)[9] = tswapreg(env->regs[9]);
307     (*regs)[10] = tswapreg(env->regs[10]);
308     (*regs)[11] = tswapreg(env->regs[11]);
309     (*regs)[12] = tswapreg(env->regs[12]);
310     (*regs)[13] = tswapreg(env->regs[13]);
311     (*regs)[14] = tswapreg(env->regs[14]);
312     (*regs)[15] = tswapreg(env->regs[15]);
313 
314     (*regs)[16] = tswapreg(cpsr_read((CPUARMState *)env));
315     (*regs)[17] = tswapreg(env->regs[0]); /* XXX */
316 }
317 
318 #define USE_ELF_CORE_DUMP
319 #define ELF_EXEC_PAGESIZE       4096
320 
321 enum
322 {
323     ARM_HWCAP_ARM_SWP       = 1 << 0,
324     ARM_HWCAP_ARM_HALF      = 1 << 1,
325     ARM_HWCAP_ARM_THUMB     = 1 << 2,
326     ARM_HWCAP_ARM_26BIT     = 1 << 3,
327     ARM_HWCAP_ARM_FAST_MULT = 1 << 4,
328     ARM_HWCAP_ARM_FPA       = 1 << 5,
329     ARM_HWCAP_ARM_VFP       = 1 << 6,
330     ARM_HWCAP_ARM_EDSP      = 1 << 7,
331     ARM_HWCAP_ARM_JAVA      = 1 << 8,
332     ARM_HWCAP_ARM_IWMMXT    = 1 << 9,
333     ARM_HWCAP_ARM_CRUNCH    = 1 << 10,
334     ARM_HWCAP_ARM_THUMBEE   = 1 << 11,
335     ARM_HWCAP_ARM_NEON      = 1 << 12,
336     ARM_HWCAP_ARM_VFPv3     = 1 << 13,
337     ARM_HWCAP_ARM_VFPv3D16  = 1 << 14,
338     ARM_HWCAP_ARM_TLS       = 1 << 15,
339     ARM_HWCAP_ARM_VFPv4     = 1 << 16,
340     ARM_HWCAP_ARM_IDIVA     = 1 << 17,
341     ARM_HWCAP_ARM_IDIVT     = 1 << 18,
342     ARM_HWCAP_ARM_VFPD32    = 1 << 19,
343     ARM_HWCAP_ARM_LPAE      = 1 << 20,
344     ARM_HWCAP_ARM_EVTSTRM   = 1 << 21,
345 };
346 
347 enum {
348     ARM_HWCAP2_ARM_AES      = 1 << 0,
349     ARM_HWCAP2_ARM_PMULL    = 1 << 1,
350     ARM_HWCAP2_ARM_SHA1     = 1 << 2,
351     ARM_HWCAP2_ARM_SHA2     = 1 << 3,
352     ARM_HWCAP2_ARM_CRC32    = 1 << 4,
353 };
354 
355 /* The commpage only exists for 32 bit kernels */
356 
357 /* Return 1 if the proposed guest space is suitable for the guest.
358  * Return 0 if the proposed guest space isn't suitable, but another
359  * address space should be tried.
360  * Return -1 if there is no way the proposed guest space can be
361  * valid regardless of the base.
362  * The guest code may leave a page mapped and populate it if the
363  * address is suitable.
364  */
365 static int init_guest_commpage(unsigned long guest_base,
366                                unsigned long guest_size)
367 {
368     unsigned long real_start, test_page_addr;
369 
370     /* We need to check that we can force a fault on access to the
371      * commpage at 0xffff0fxx
372      */
373     test_page_addr = guest_base + (0xffff0f00 & qemu_host_page_mask);
374 
375     /* If the commpage lies within the already allocated guest space,
376      * then there is no way we can allocate it.
377      *
378      * You may be thinking that that this check is redundant because
379      * we already validated the guest size against MAX_RESERVED_VA;
380      * but if qemu_host_page_mask is unusually large, then
381      * test_page_addr may be lower.
382      */
383     if (test_page_addr >= guest_base
384         && test_page_addr < (guest_base + guest_size)) {
385         return -1;
386     }
387 
388     /* Note it needs to be writeable to let us initialise it */
389     real_start = (unsigned long)
390                  mmap((void *)test_page_addr, qemu_host_page_size,
391                      PROT_READ | PROT_WRITE,
392                      MAP_ANONYMOUS | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
393 
394     /* If we can't map it then try another address */
395     if (real_start == -1ul) {
396         return 0;
397     }
398 
399     if (real_start != test_page_addr) {
400         /* OS didn't put the page where we asked - unmap and reject */
401         munmap((void *)real_start, qemu_host_page_size);
402         return 0;
403     }
404 
405     /* Leave the page mapped
406      * Populate it (mmap should have left it all 0'd)
407      */
408 
409     /* Kernel helper versions */
410     __put_user(5, (uint32_t *)g2h(0xffff0ffcul));
411 
412     /* Now it's populated make it RO */
413     if (mprotect((void *)test_page_addr, qemu_host_page_size, PROT_READ)) {
414         perror("Protecting guest commpage");
415         exit(-1);
416     }
417 
418     return 1; /* All good */
419 }
420 
421 #define ELF_HWCAP get_elf_hwcap()
422 #define ELF_HWCAP2 get_elf_hwcap2()
423 
424 static uint32_t get_elf_hwcap(void)
425 {
426     ARMCPU *cpu = ARM_CPU(thread_cpu);
427     uint32_t hwcaps = 0;
428 
429     hwcaps |= ARM_HWCAP_ARM_SWP;
430     hwcaps |= ARM_HWCAP_ARM_HALF;
431     hwcaps |= ARM_HWCAP_ARM_THUMB;
432     hwcaps |= ARM_HWCAP_ARM_FAST_MULT;
433 
434     /* probe for the extra features */
435 #define GET_FEATURE(feat, hwcap) \
436     do { if (arm_feature(&cpu->env, feat)) { hwcaps |= hwcap; } } while (0)
437     /* EDSP is in v5TE and above, but all our v5 CPUs are v5TE */
438     GET_FEATURE(ARM_FEATURE_V5, ARM_HWCAP_ARM_EDSP);
439     GET_FEATURE(ARM_FEATURE_VFP, ARM_HWCAP_ARM_VFP);
440     GET_FEATURE(ARM_FEATURE_IWMMXT, ARM_HWCAP_ARM_IWMMXT);
441     GET_FEATURE(ARM_FEATURE_THUMB2EE, ARM_HWCAP_ARM_THUMBEE);
442     GET_FEATURE(ARM_FEATURE_NEON, ARM_HWCAP_ARM_NEON);
443     GET_FEATURE(ARM_FEATURE_VFP3, ARM_HWCAP_ARM_VFPv3);
444     GET_FEATURE(ARM_FEATURE_V6K, ARM_HWCAP_ARM_TLS);
445     GET_FEATURE(ARM_FEATURE_VFP4, ARM_HWCAP_ARM_VFPv4);
446     GET_FEATURE(ARM_FEATURE_ARM_DIV, ARM_HWCAP_ARM_IDIVA);
447     GET_FEATURE(ARM_FEATURE_THUMB_DIV, ARM_HWCAP_ARM_IDIVT);
448     /* All QEMU's VFPv3 CPUs have 32 registers, see VFP_DREG in translate.c.
449      * Note that the ARM_HWCAP_ARM_VFPv3D16 bit is always the inverse of
450      * ARM_HWCAP_ARM_VFPD32 (and so always clear for QEMU); it is unrelated
451      * to our VFP_FP16 feature bit.
452      */
453     GET_FEATURE(ARM_FEATURE_VFP3, ARM_HWCAP_ARM_VFPD32);
454     GET_FEATURE(ARM_FEATURE_LPAE, ARM_HWCAP_ARM_LPAE);
455 
456     return hwcaps;
457 }
458 
459 static uint32_t get_elf_hwcap2(void)
460 {
461     ARMCPU *cpu = ARM_CPU(thread_cpu);
462     uint32_t hwcaps = 0;
463 
464     GET_FEATURE(ARM_FEATURE_V8_AES, ARM_HWCAP2_ARM_AES);
465     GET_FEATURE(ARM_FEATURE_V8_PMULL, ARM_HWCAP2_ARM_PMULL);
466     GET_FEATURE(ARM_FEATURE_V8_SHA1, ARM_HWCAP2_ARM_SHA1);
467     GET_FEATURE(ARM_FEATURE_V8_SHA256, ARM_HWCAP2_ARM_SHA2);
468     GET_FEATURE(ARM_FEATURE_CRC, ARM_HWCAP2_ARM_CRC32);
469     return hwcaps;
470 }
471 
472 #undef GET_FEATURE
473 
474 #else
475 /* 64 bit ARM definitions */
476 #define ELF_START_MMAP 0x80000000
477 
478 #define ELF_ARCH        EM_AARCH64
479 #define ELF_CLASS       ELFCLASS64
480 #define ELF_PLATFORM    "aarch64"
481 
482 static inline void init_thread(struct target_pt_regs *regs,
483                                struct image_info *infop)
484 {
485     abi_long stack = infop->start_stack;
486     memset(regs, 0, sizeof(*regs));
487 
488     regs->pc = infop->entry & ~0x3ULL;
489     regs->sp = stack;
490 }
491 
492 #define ELF_NREG    34
493 typedef target_elf_greg_t  target_elf_gregset_t[ELF_NREG];
494 
495 static void elf_core_copy_regs(target_elf_gregset_t *regs,
496                                const CPUARMState *env)
497 {
498     int i;
499 
500     for (i = 0; i < 32; i++) {
501         (*regs)[i] = tswapreg(env->xregs[i]);
502     }
503     (*regs)[32] = tswapreg(env->pc);
504     (*regs)[33] = tswapreg(pstate_read((CPUARMState *)env));
505 }
506 
507 #define USE_ELF_CORE_DUMP
508 #define ELF_EXEC_PAGESIZE       4096
509 
510 enum {
511     ARM_HWCAP_A64_FP            = 1 << 0,
512     ARM_HWCAP_A64_ASIMD         = 1 << 1,
513     ARM_HWCAP_A64_EVTSTRM       = 1 << 2,
514     ARM_HWCAP_A64_AES           = 1 << 3,
515     ARM_HWCAP_A64_PMULL         = 1 << 4,
516     ARM_HWCAP_A64_SHA1          = 1 << 5,
517     ARM_HWCAP_A64_SHA2          = 1 << 6,
518     ARM_HWCAP_A64_CRC32         = 1 << 7,
519     ARM_HWCAP_A64_ATOMICS       = 1 << 8,
520     ARM_HWCAP_A64_FPHP          = 1 << 9,
521     ARM_HWCAP_A64_ASIMDHP       = 1 << 10,
522     ARM_HWCAP_A64_CPUID         = 1 << 11,
523     ARM_HWCAP_A64_ASIMDRDM      = 1 << 12,
524     ARM_HWCAP_A64_JSCVT         = 1 << 13,
525     ARM_HWCAP_A64_FCMA          = 1 << 14,
526     ARM_HWCAP_A64_LRCPC         = 1 << 15,
527     ARM_HWCAP_A64_DCPOP         = 1 << 16,
528     ARM_HWCAP_A64_SHA3          = 1 << 17,
529     ARM_HWCAP_A64_SM3           = 1 << 18,
530     ARM_HWCAP_A64_SM4           = 1 << 19,
531     ARM_HWCAP_A64_ASIMDDP       = 1 << 20,
532     ARM_HWCAP_A64_SHA512        = 1 << 21,
533     ARM_HWCAP_A64_SVE           = 1 << 22,
534 };
535 
536 #define ELF_HWCAP get_elf_hwcap()
537 
538 static uint32_t get_elf_hwcap(void)
539 {
540     ARMCPU *cpu = ARM_CPU(thread_cpu);
541     uint32_t hwcaps = 0;
542 
543     hwcaps |= ARM_HWCAP_A64_FP;
544     hwcaps |= ARM_HWCAP_A64_ASIMD;
545 
546     /* probe for the extra features */
547 #define GET_FEATURE(feat, hwcap) \
548     do { if (arm_feature(&cpu->env, feat)) { hwcaps |= hwcap; } } while (0)
549     GET_FEATURE(ARM_FEATURE_V8_AES, ARM_HWCAP_A64_AES);
550     GET_FEATURE(ARM_FEATURE_V8_PMULL, ARM_HWCAP_A64_PMULL);
551     GET_FEATURE(ARM_FEATURE_V8_SHA1, ARM_HWCAP_A64_SHA1);
552     GET_FEATURE(ARM_FEATURE_V8_SHA256, ARM_HWCAP_A64_SHA2);
553     GET_FEATURE(ARM_FEATURE_CRC, ARM_HWCAP_A64_CRC32);
554     GET_FEATURE(ARM_FEATURE_V8_SHA3, ARM_HWCAP_A64_SHA3);
555     GET_FEATURE(ARM_FEATURE_V8_SM3, ARM_HWCAP_A64_SM3);
556     GET_FEATURE(ARM_FEATURE_V8_SM4, ARM_HWCAP_A64_SM4);
557     GET_FEATURE(ARM_FEATURE_V8_SHA512, ARM_HWCAP_A64_SHA512);
558     GET_FEATURE(ARM_FEATURE_V8_FP16,
559                 ARM_HWCAP_A64_FPHP | ARM_HWCAP_A64_ASIMDHP);
560     GET_FEATURE(ARM_FEATURE_V8_RDM, ARM_HWCAP_A64_ASIMDRDM);
561     GET_FEATURE(ARM_FEATURE_V8_FCMA, ARM_HWCAP_A64_FCMA);
562 #undef GET_FEATURE
563 
564     return hwcaps;
565 }
566 
567 #endif /* not TARGET_AARCH64 */
568 #endif /* TARGET_ARM */
569 
570 #ifdef TARGET_SPARC
571 #ifdef TARGET_SPARC64
572 
573 #define ELF_START_MMAP 0x80000000
574 #define ELF_HWCAP  (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \
575                     | HWCAP_SPARC_MULDIV | HWCAP_SPARC_V9)
576 #ifndef TARGET_ABI32
577 #define elf_check_arch(x) ( (x) == EM_SPARCV9 || (x) == EM_SPARC32PLUS )
578 #else
579 #define elf_check_arch(x) ( (x) == EM_SPARC32PLUS || (x) == EM_SPARC )
580 #endif
581 
582 #define ELF_CLASS   ELFCLASS64
583 #define ELF_ARCH    EM_SPARCV9
584 
585 #define STACK_BIAS              2047
586 
587 static inline void init_thread(struct target_pt_regs *regs,
588                                struct image_info *infop)
589 {
590 #ifndef TARGET_ABI32
591     regs->tstate = 0;
592 #endif
593     regs->pc = infop->entry;
594     regs->npc = regs->pc + 4;
595     regs->y = 0;
596 #ifdef TARGET_ABI32
597     regs->u_regs[14] = infop->start_stack - 16 * 4;
598 #else
599     if (personality(infop->personality) == PER_LINUX32)
600         regs->u_regs[14] = infop->start_stack - 16 * 4;
601     else
602         regs->u_regs[14] = infop->start_stack - 16 * 8 - STACK_BIAS;
603 #endif
604 }
605 
606 #else
607 #define ELF_START_MMAP 0x80000000
608 #define ELF_HWCAP  (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \
609                     | HWCAP_SPARC_MULDIV)
610 
611 #define ELF_CLASS   ELFCLASS32
612 #define ELF_ARCH    EM_SPARC
613 
614 static inline void init_thread(struct target_pt_regs *regs,
615                                struct image_info *infop)
616 {
617     regs->psr = 0;
618     regs->pc = infop->entry;
619     regs->npc = regs->pc + 4;
620     regs->y = 0;
621     regs->u_regs[14] = infop->start_stack - 16 * 4;
622 }
623 
624 #endif
625 #endif
626 
627 #ifdef TARGET_PPC
628 
629 #define ELF_MACHINE    PPC_ELF_MACHINE
630 #define ELF_START_MMAP 0x80000000
631 
632 #if defined(TARGET_PPC64) && !defined(TARGET_ABI32)
633 
634 #define elf_check_arch(x) ( (x) == EM_PPC64 )
635 
636 #define ELF_CLASS       ELFCLASS64
637 
638 #else
639 
640 #define ELF_CLASS       ELFCLASS32
641 
642 #endif
643 
644 #define ELF_ARCH        EM_PPC
645 
646 /* Feature masks for the Aux Vector Hardware Capabilities (AT_HWCAP).
647    See arch/powerpc/include/asm/cputable.h.  */
648 enum {
649     QEMU_PPC_FEATURE_32 = 0x80000000,
650     QEMU_PPC_FEATURE_64 = 0x40000000,
651     QEMU_PPC_FEATURE_601_INSTR = 0x20000000,
652     QEMU_PPC_FEATURE_HAS_ALTIVEC = 0x10000000,
653     QEMU_PPC_FEATURE_HAS_FPU = 0x08000000,
654     QEMU_PPC_FEATURE_HAS_MMU = 0x04000000,
655     QEMU_PPC_FEATURE_HAS_4xxMAC = 0x02000000,
656     QEMU_PPC_FEATURE_UNIFIED_CACHE = 0x01000000,
657     QEMU_PPC_FEATURE_HAS_SPE = 0x00800000,
658     QEMU_PPC_FEATURE_HAS_EFP_SINGLE = 0x00400000,
659     QEMU_PPC_FEATURE_HAS_EFP_DOUBLE = 0x00200000,
660     QEMU_PPC_FEATURE_NO_TB = 0x00100000,
661     QEMU_PPC_FEATURE_POWER4 = 0x00080000,
662     QEMU_PPC_FEATURE_POWER5 = 0x00040000,
663     QEMU_PPC_FEATURE_POWER5_PLUS = 0x00020000,
664     QEMU_PPC_FEATURE_CELL = 0x00010000,
665     QEMU_PPC_FEATURE_BOOKE = 0x00008000,
666     QEMU_PPC_FEATURE_SMT = 0x00004000,
667     QEMU_PPC_FEATURE_ICACHE_SNOOP = 0x00002000,
668     QEMU_PPC_FEATURE_ARCH_2_05 = 0x00001000,
669     QEMU_PPC_FEATURE_PA6T = 0x00000800,
670     QEMU_PPC_FEATURE_HAS_DFP = 0x00000400,
671     QEMU_PPC_FEATURE_POWER6_EXT = 0x00000200,
672     QEMU_PPC_FEATURE_ARCH_2_06 = 0x00000100,
673     QEMU_PPC_FEATURE_HAS_VSX = 0x00000080,
674     QEMU_PPC_FEATURE_PSERIES_PERFMON_COMPAT = 0x00000040,
675 
676     QEMU_PPC_FEATURE_TRUE_LE = 0x00000002,
677     QEMU_PPC_FEATURE_PPC_LE = 0x00000001,
678 
679     /* Feature definitions in AT_HWCAP2.  */
680     QEMU_PPC_FEATURE2_ARCH_2_07 = 0x80000000, /* ISA 2.07 */
681     QEMU_PPC_FEATURE2_HAS_HTM = 0x40000000, /* Hardware Transactional Memory */
682     QEMU_PPC_FEATURE2_HAS_DSCR = 0x20000000, /* Data Stream Control Register */
683     QEMU_PPC_FEATURE2_HAS_EBB = 0x10000000, /* Event Base Branching */
684     QEMU_PPC_FEATURE2_HAS_ISEL = 0x08000000, /* Integer Select */
685     QEMU_PPC_FEATURE2_HAS_TAR = 0x04000000, /* Target Address Register */
686 };
687 
688 #define ELF_HWCAP get_elf_hwcap()
689 
690 static uint32_t get_elf_hwcap(void)
691 {
692     PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);
693     uint32_t features = 0;
694 
695     /* We don't have to be terribly complete here; the high points are
696        Altivec/FP/SPE support.  Anything else is just a bonus.  */
697 #define GET_FEATURE(flag, feature)                                      \
698     do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0)
699 #define GET_FEATURE2(flags, feature) \
700     do { \
701         if ((cpu->env.insns_flags2 & flags) == flags) { \
702             features |= feature; \
703         } \
704     } while (0)
705     GET_FEATURE(PPC_64B, QEMU_PPC_FEATURE_64);
706     GET_FEATURE(PPC_FLOAT, QEMU_PPC_FEATURE_HAS_FPU);
707     GET_FEATURE(PPC_ALTIVEC, QEMU_PPC_FEATURE_HAS_ALTIVEC);
708     GET_FEATURE(PPC_SPE, QEMU_PPC_FEATURE_HAS_SPE);
709     GET_FEATURE(PPC_SPE_SINGLE, QEMU_PPC_FEATURE_HAS_EFP_SINGLE);
710     GET_FEATURE(PPC_SPE_DOUBLE, QEMU_PPC_FEATURE_HAS_EFP_DOUBLE);
711     GET_FEATURE(PPC_BOOKE, QEMU_PPC_FEATURE_BOOKE);
712     GET_FEATURE(PPC_405_MAC, QEMU_PPC_FEATURE_HAS_4xxMAC);
713     GET_FEATURE2(PPC2_DFP, QEMU_PPC_FEATURE_HAS_DFP);
714     GET_FEATURE2(PPC2_VSX, QEMU_PPC_FEATURE_HAS_VSX);
715     GET_FEATURE2((PPC2_PERM_ISA206 | PPC2_DIVE_ISA206 | PPC2_ATOMIC_ISA206 |
716                   PPC2_FP_CVT_ISA206 | PPC2_FP_TST_ISA206),
717                   QEMU_PPC_FEATURE_ARCH_2_06);
718 #undef GET_FEATURE
719 #undef GET_FEATURE2
720 
721     return features;
722 }
723 
724 #define ELF_HWCAP2 get_elf_hwcap2()
725 
726 static uint32_t get_elf_hwcap2(void)
727 {
728     PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);
729     uint32_t features = 0;
730 
731 #define GET_FEATURE(flag, feature)                                      \
732     do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0)
733 #define GET_FEATURE2(flag, feature)                                      \
734     do { if (cpu->env.insns_flags2 & flag) { features |= feature; } } while (0)
735 
736     GET_FEATURE(PPC_ISEL, QEMU_PPC_FEATURE2_HAS_ISEL);
737     GET_FEATURE2(PPC2_BCTAR_ISA207, QEMU_PPC_FEATURE2_HAS_TAR);
738     GET_FEATURE2((PPC2_BCTAR_ISA207 | PPC2_LSQ_ISA207 | PPC2_ALTIVEC_207 |
739                   PPC2_ISA207S), QEMU_PPC_FEATURE2_ARCH_2_07);
740 
741 #undef GET_FEATURE
742 #undef GET_FEATURE2
743 
744     return features;
745 }
746 
747 /*
748  * The requirements here are:
749  * - keep the final alignment of sp (sp & 0xf)
750  * - make sure the 32-bit value at the first 16 byte aligned position of
751  *   AUXV is greater than 16 for glibc compatibility.
752  *   AT_IGNOREPPC is used for that.
753  * - for compatibility with glibc ARCH_DLINFO must always be defined on PPC,
754  *   even if DLINFO_ARCH_ITEMS goes to zero or is undefined.
755  */
756 #define DLINFO_ARCH_ITEMS       5
757 #define ARCH_DLINFO                                     \
758     do {                                                \
759         PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);              \
760         /*                                              \
761          * Handle glibc compatibility: these magic entries must \
762          * be at the lowest addresses in the final auxv.        \
763          */                                             \
764         NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC);        \
765         NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC);        \
766         NEW_AUX_ENT(AT_DCACHEBSIZE, cpu->env.dcache_line_size); \
767         NEW_AUX_ENT(AT_ICACHEBSIZE, cpu->env.icache_line_size); \
768         NEW_AUX_ENT(AT_UCACHEBSIZE, 0);                 \
769     } while (0)
770 
771 static inline void init_thread(struct target_pt_regs *_regs, struct image_info *infop)
772 {
773     _regs->gpr[1] = infop->start_stack;
774 #if defined(TARGET_PPC64) && !defined(TARGET_ABI32)
775     if (get_ppc64_abi(infop) < 2) {
776         uint64_t val;
777         get_user_u64(val, infop->entry + 8);
778         _regs->gpr[2] = val + infop->load_bias;
779         get_user_u64(val, infop->entry);
780         infop->entry = val + infop->load_bias;
781     } else {
782         _regs->gpr[12] = infop->entry;  /* r12 set to global entry address */
783     }
784 #endif
785     _regs->nip = infop->entry;
786 }
787 
788 /* See linux kernel: arch/powerpc/include/asm/elf.h.  */
789 #define ELF_NREG 48
790 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
791 
792 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUPPCState *env)
793 {
794     int i;
795     target_ulong ccr = 0;
796 
797     for (i = 0; i < ARRAY_SIZE(env->gpr); i++) {
798         (*regs)[i] = tswapreg(env->gpr[i]);
799     }
800 
801     (*regs)[32] = tswapreg(env->nip);
802     (*regs)[33] = tswapreg(env->msr);
803     (*regs)[35] = tswapreg(env->ctr);
804     (*regs)[36] = tswapreg(env->lr);
805     (*regs)[37] = tswapreg(env->xer);
806 
807     for (i = 0; i < ARRAY_SIZE(env->crf); i++) {
808         ccr |= env->crf[i] << (32 - ((i + 1) * 4));
809     }
810     (*regs)[38] = tswapreg(ccr);
811 }
812 
813 #define USE_ELF_CORE_DUMP
814 #define ELF_EXEC_PAGESIZE       4096
815 
816 #endif
817 
818 #ifdef TARGET_MIPS
819 
820 #define ELF_START_MMAP 0x80000000
821 
822 #ifdef TARGET_MIPS64
823 #define ELF_CLASS   ELFCLASS64
824 #else
825 #define ELF_CLASS   ELFCLASS32
826 #endif
827 #define ELF_ARCH    EM_MIPS
828 
829 static inline void init_thread(struct target_pt_regs *regs,
830                                struct image_info *infop)
831 {
832     regs->cp0_status = 2 << CP0St_KSU;
833     regs->cp0_epc = infop->entry;
834     regs->regs[29] = infop->start_stack;
835 }
836 
837 /* See linux kernel: arch/mips/include/asm/elf.h.  */
838 #define ELF_NREG 45
839 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
840 
841 /* See linux kernel: arch/mips/include/asm/reg.h.  */
842 enum {
843 #ifdef TARGET_MIPS64
844     TARGET_EF_R0 = 0,
845 #else
846     TARGET_EF_R0 = 6,
847 #endif
848     TARGET_EF_R26 = TARGET_EF_R0 + 26,
849     TARGET_EF_R27 = TARGET_EF_R0 + 27,
850     TARGET_EF_LO = TARGET_EF_R0 + 32,
851     TARGET_EF_HI = TARGET_EF_R0 + 33,
852     TARGET_EF_CP0_EPC = TARGET_EF_R0 + 34,
853     TARGET_EF_CP0_BADVADDR = TARGET_EF_R0 + 35,
854     TARGET_EF_CP0_STATUS = TARGET_EF_R0 + 36,
855     TARGET_EF_CP0_CAUSE = TARGET_EF_R0 + 37
856 };
857 
858 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs.  */
859 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMIPSState *env)
860 {
861     int i;
862 
863     for (i = 0; i < TARGET_EF_R0; i++) {
864         (*regs)[i] = 0;
865     }
866     (*regs)[TARGET_EF_R0] = 0;
867 
868     for (i = 1; i < ARRAY_SIZE(env->active_tc.gpr); i++) {
869         (*regs)[TARGET_EF_R0 + i] = tswapreg(env->active_tc.gpr[i]);
870     }
871 
872     (*regs)[TARGET_EF_R26] = 0;
873     (*regs)[TARGET_EF_R27] = 0;
874     (*regs)[TARGET_EF_LO] = tswapreg(env->active_tc.LO[0]);
875     (*regs)[TARGET_EF_HI] = tswapreg(env->active_tc.HI[0]);
876     (*regs)[TARGET_EF_CP0_EPC] = tswapreg(env->active_tc.PC);
877     (*regs)[TARGET_EF_CP0_BADVADDR] = tswapreg(env->CP0_BadVAddr);
878     (*regs)[TARGET_EF_CP0_STATUS] = tswapreg(env->CP0_Status);
879     (*regs)[TARGET_EF_CP0_CAUSE] = tswapreg(env->CP0_Cause);
880 }
881 
882 #define USE_ELF_CORE_DUMP
883 #define ELF_EXEC_PAGESIZE        4096
884 
885 #endif /* TARGET_MIPS */
886 
887 #ifdef TARGET_MICROBLAZE
888 
889 #define ELF_START_MMAP 0x80000000
890 
891 #define elf_check_arch(x) ( (x) == EM_MICROBLAZE || (x) == EM_MICROBLAZE_OLD)
892 
893 #define ELF_CLASS   ELFCLASS32
894 #define ELF_ARCH    EM_MICROBLAZE
895 
896 static inline void init_thread(struct target_pt_regs *regs,
897                                struct image_info *infop)
898 {
899     regs->pc = infop->entry;
900     regs->r1 = infop->start_stack;
901 
902 }
903 
904 #define ELF_EXEC_PAGESIZE        4096
905 
906 #define USE_ELF_CORE_DUMP
907 #define ELF_NREG 38
908 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
909 
910 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs.  */
911 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMBState *env)
912 {
913     int i, pos = 0;
914 
915     for (i = 0; i < 32; i++) {
916         (*regs)[pos++] = tswapreg(env->regs[i]);
917     }
918 
919     for (i = 0; i < 6; i++) {
920         (*regs)[pos++] = tswapreg(env->sregs[i]);
921     }
922 }
923 
924 #endif /* TARGET_MICROBLAZE */
925 
926 #ifdef TARGET_NIOS2
927 
928 #define ELF_START_MMAP 0x80000000
929 
930 #define elf_check_arch(x) ((x) == EM_ALTERA_NIOS2)
931 
932 #define ELF_CLASS   ELFCLASS32
933 #define ELF_ARCH    EM_ALTERA_NIOS2
934 
935 static void init_thread(struct target_pt_regs *regs, struct image_info *infop)
936 {
937     regs->ea = infop->entry;
938     regs->sp = infop->start_stack;
939     regs->estatus = 0x3;
940 }
941 
942 #define ELF_EXEC_PAGESIZE        4096
943 
944 #define USE_ELF_CORE_DUMP
945 #define ELF_NREG 49
946 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
947 
948 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs.  */
949 static void elf_core_copy_regs(target_elf_gregset_t *regs,
950                                const CPUNios2State *env)
951 {
952     int i;
953 
954     (*regs)[0] = -1;
955     for (i = 1; i < 8; i++)    /* r0-r7 */
956         (*regs)[i] = tswapreg(env->regs[i + 7]);
957 
958     for (i = 8; i < 16; i++)   /* r8-r15 */
959         (*regs)[i] = tswapreg(env->regs[i - 8]);
960 
961     for (i = 16; i < 24; i++)  /* r16-r23 */
962         (*regs)[i] = tswapreg(env->regs[i + 7]);
963     (*regs)[24] = -1;    /* R_ET */
964     (*regs)[25] = -1;    /* R_BT */
965     (*regs)[26] = tswapreg(env->regs[R_GP]);
966     (*regs)[27] = tswapreg(env->regs[R_SP]);
967     (*regs)[28] = tswapreg(env->regs[R_FP]);
968     (*regs)[29] = tswapreg(env->regs[R_EA]);
969     (*regs)[30] = -1;    /* R_SSTATUS */
970     (*regs)[31] = tswapreg(env->regs[R_RA]);
971 
972     (*regs)[32] = tswapreg(env->regs[R_PC]);
973 
974     (*regs)[33] = -1; /* R_STATUS */
975     (*regs)[34] = tswapreg(env->regs[CR_ESTATUS]);
976 
977     for (i = 35; i < 49; i++)    /* ... */
978         (*regs)[i] = -1;
979 }
980 
981 #endif /* TARGET_NIOS2 */
982 
983 #ifdef TARGET_OPENRISC
984 
985 #define ELF_START_MMAP 0x08000000
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(cpu_get_gpr(env, i));
1012     }
1013     (*regs)[32] = tswapreg(env->pc);
1014     (*regs)[33] = tswapreg(cpu_get_sr(env));
1015 }
1016 #define ELF_HWCAP 0
1017 #define ELF_PLATFORM NULL
1018 
1019 #endif /* TARGET_OPENRISC */
1020 
1021 #ifdef TARGET_SH4
1022 
1023 #define ELF_START_MMAP 0x80000000
1024 
1025 #define ELF_CLASS ELFCLASS32
1026 #define ELF_ARCH  EM_SH
1027 
1028 static inline void init_thread(struct target_pt_regs *regs,
1029                                struct image_info *infop)
1030 {
1031     /* Check other registers XXXXX */
1032     regs->pc = infop->entry;
1033     regs->regs[15] = infop->start_stack;
1034 }
1035 
1036 /* See linux kernel: arch/sh/include/asm/elf.h.  */
1037 #define ELF_NREG 23
1038 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1039 
1040 /* See linux kernel: arch/sh/include/asm/ptrace.h.  */
1041 enum {
1042     TARGET_REG_PC = 16,
1043     TARGET_REG_PR = 17,
1044     TARGET_REG_SR = 18,
1045     TARGET_REG_GBR = 19,
1046     TARGET_REG_MACH = 20,
1047     TARGET_REG_MACL = 21,
1048     TARGET_REG_SYSCALL = 22
1049 };
1050 
1051 static inline void elf_core_copy_regs(target_elf_gregset_t *regs,
1052                                       const CPUSH4State *env)
1053 {
1054     int i;
1055 
1056     for (i = 0; i < 16; i++) {
1057         (*regs)[i] = tswapreg(env->gregs[i]);
1058     }
1059 
1060     (*regs)[TARGET_REG_PC] = tswapreg(env->pc);
1061     (*regs)[TARGET_REG_PR] = tswapreg(env->pr);
1062     (*regs)[TARGET_REG_SR] = tswapreg(env->sr);
1063     (*regs)[TARGET_REG_GBR] = tswapreg(env->gbr);
1064     (*regs)[TARGET_REG_MACH] = tswapreg(env->mach);
1065     (*regs)[TARGET_REG_MACL] = tswapreg(env->macl);
1066     (*regs)[TARGET_REG_SYSCALL] = 0; /* FIXME */
1067 }
1068 
1069 #define USE_ELF_CORE_DUMP
1070 #define ELF_EXEC_PAGESIZE        4096
1071 
1072 enum {
1073     SH_CPU_HAS_FPU            = 0x0001, /* Hardware FPU support */
1074     SH_CPU_HAS_P2_FLUSH_BUG   = 0x0002, /* Need to flush the cache in P2 area */
1075     SH_CPU_HAS_MMU_PAGE_ASSOC = 0x0004, /* SH3: TLB way selection bit support */
1076     SH_CPU_HAS_DSP            = 0x0008, /* SH-DSP: DSP support */
1077     SH_CPU_HAS_PERF_COUNTER   = 0x0010, /* Hardware performance counters */
1078     SH_CPU_HAS_PTEA           = 0x0020, /* PTEA register */
1079     SH_CPU_HAS_LLSC           = 0x0040, /* movli.l/movco.l */
1080     SH_CPU_HAS_L2_CACHE       = 0x0080, /* Secondary cache / URAM */
1081     SH_CPU_HAS_OP32           = 0x0100, /* 32-bit instruction support */
1082     SH_CPU_HAS_PTEAEX         = 0x0200, /* PTE ASID Extension support */
1083 };
1084 
1085 #define ELF_HWCAP get_elf_hwcap()
1086 
1087 static uint32_t get_elf_hwcap(void)
1088 {
1089     SuperHCPU *cpu = SUPERH_CPU(thread_cpu);
1090     uint32_t hwcap = 0;
1091 
1092     hwcap |= SH_CPU_HAS_FPU;
1093 
1094     if (cpu->env.features & SH_FEATURE_SH4A) {
1095         hwcap |= SH_CPU_HAS_LLSC;
1096     }
1097 
1098     return hwcap;
1099 }
1100 
1101 #endif
1102 
1103 #ifdef TARGET_CRIS
1104 
1105 #define ELF_START_MMAP 0x80000000
1106 
1107 #define ELF_CLASS ELFCLASS32
1108 #define ELF_ARCH  EM_CRIS
1109 
1110 static inline void init_thread(struct target_pt_regs *regs,
1111                                struct image_info *infop)
1112 {
1113     regs->erp = infop->entry;
1114 }
1115 
1116 #define ELF_EXEC_PAGESIZE        8192
1117 
1118 #endif
1119 
1120 #ifdef TARGET_M68K
1121 
1122 #define ELF_START_MMAP 0x80000000
1123 
1124 #define ELF_CLASS       ELFCLASS32
1125 #define ELF_ARCH        EM_68K
1126 
1127 /* ??? Does this need to do anything?
1128    #define ELF_PLAT_INIT(_r) */
1129 
1130 static inline void init_thread(struct target_pt_regs *regs,
1131                                struct image_info *infop)
1132 {
1133     regs->usp = infop->start_stack;
1134     regs->sr = 0;
1135     regs->pc = infop->entry;
1136 }
1137 
1138 /* See linux kernel: arch/m68k/include/asm/elf.h.  */
1139 #define ELF_NREG 20
1140 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1141 
1142 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUM68KState *env)
1143 {
1144     (*regs)[0] = tswapreg(env->dregs[1]);
1145     (*regs)[1] = tswapreg(env->dregs[2]);
1146     (*regs)[2] = tswapreg(env->dregs[3]);
1147     (*regs)[3] = tswapreg(env->dregs[4]);
1148     (*regs)[4] = tswapreg(env->dregs[5]);
1149     (*regs)[5] = tswapreg(env->dregs[6]);
1150     (*regs)[6] = tswapreg(env->dregs[7]);
1151     (*regs)[7] = tswapreg(env->aregs[0]);
1152     (*regs)[8] = tswapreg(env->aregs[1]);
1153     (*regs)[9] = tswapreg(env->aregs[2]);
1154     (*regs)[10] = tswapreg(env->aregs[3]);
1155     (*regs)[11] = tswapreg(env->aregs[4]);
1156     (*regs)[12] = tswapreg(env->aregs[5]);
1157     (*regs)[13] = tswapreg(env->aregs[6]);
1158     (*regs)[14] = tswapreg(env->dregs[0]);
1159     (*regs)[15] = tswapreg(env->aregs[7]);
1160     (*regs)[16] = tswapreg(env->dregs[0]); /* FIXME: orig_d0 */
1161     (*regs)[17] = tswapreg(env->sr);
1162     (*regs)[18] = tswapreg(env->pc);
1163     (*regs)[19] = 0;  /* FIXME: regs->format | regs->vector */
1164 }
1165 
1166 #define USE_ELF_CORE_DUMP
1167 #define ELF_EXEC_PAGESIZE       8192
1168 
1169 #endif
1170 
1171 #ifdef TARGET_ALPHA
1172 
1173 #define ELF_START_MMAP (0x30000000000ULL)
1174 
1175 #define ELF_CLASS      ELFCLASS64
1176 #define ELF_ARCH       EM_ALPHA
1177 
1178 static inline void init_thread(struct target_pt_regs *regs,
1179                                struct image_info *infop)
1180 {
1181     regs->pc = infop->entry;
1182     regs->ps = 8;
1183     regs->usp = infop->start_stack;
1184 }
1185 
1186 #define ELF_EXEC_PAGESIZE        8192
1187 
1188 #endif /* TARGET_ALPHA */
1189 
1190 #ifdef TARGET_S390X
1191 
1192 #define ELF_START_MMAP (0x20000000000ULL)
1193 
1194 #define ELF_CLASS	ELFCLASS64
1195 #define ELF_DATA	ELFDATA2MSB
1196 #define ELF_ARCH	EM_S390
1197 
1198 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop)
1199 {
1200     regs->psw.addr = infop->entry;
1201     regs->psw.mask = PSW_MASK_64 | PSW_MASK_32;
1202     regs->gprs[15] = infop->start_stack;
1203 }
1204 
1205 #endif /* TARGET_S390X */
1206 
1207 #ifdef TARGET_TILEGX
1208 
1209 /* 42 bits real used address, a half for user mode */
1210 #define ELF_START_MMAP (0x00000020000000000ULL)
1211 
1212 #define elf_check_arch(x) ((x) == EM_TILEGX)
1213 
1214 #define ELF_CLASS   ELFCLASS64
1215 #define ELF_DATA    ELFDATA2LSB
1216 #define ELF_ARCH    EM_TILEGX
1217 
1218 static inline void init_thread(struct target_pt_regs *regs,
1219                                struct image_info *infop)
1220 {
1221     regs->pc = infop->entry;
1222     regs->sp = infop->start_stack;
1223 
1224 }
1225 
1226 #define ELF_EXEC_PAGESIZE        65536 /* TILE-Gx page size is 64KB */
1227 
1228 #endif /* TARGET_TILEGX */
1229 
1230 #ifdef TARGET_RISCV
1231 
1232 #define ELF_START_MMAP 0x80000000
1233 #define ELF_ARCH  EM_RISCV
1234 
1235 #ifdef TARGET_RISCV32
1236 #define ELF_CLASS ELFCLASS32
1237 #else
1238 #define ELF_CLASS ELFCLASS64
1239 #endif
1240 
1241 static inline void init_thread(struct target_pt_regs *regs,
1242                                struct image_info *infop)
1243 {
1244     regs->sepc = infop->entry;
1245     regs->sp = infop->start_stack;
1246 }
1247 
1248 #define ELF_EXEC_PAGESIZE 4096
1249 
1250 #endif /* TARGET_RISCV */
1251 
1252 #ifdef TARGET_HPPA
1253 
1254 #define ELF_START_MMAP  0x80000000
1255 #define ELF_CLASS       ELFCLASS32
1256 #define ELF_ARCH        EM_PARISC
1257 #define ELF_PLATFORM    "PARISC"
1258 #define STACK_GROWS_DOWN 0
1259 #define STACK_ALIGNMENT  64
1260 
1261 static inline void init_thread(struct target_pt_regs *regs,
1262                                struct image_info *infop)
1263 {
1264     regs->iaoq[0] = infop->entry;
1265     regs->iaoq[1] = infop->entry + 4;
1266     regs->gr[23] = 0;
1267     regs->gr[24] = infop->arg_start;
1268     regs->gr[25] = (infop->arg_end - infop->arg_start) / sizeof(abi_ulong);
1269     /* The top-of-stack contains a linkage buffer.  */
1270     regs->gr[30] = infop->start_stack + 64;
1271     regs->gr[31] = infop->entry;
1272 }
1273 
1274 #endif /* TARGET_HPPA */
1275 
1276 #ifdef TARGET_XTENSA
1277 
1278 #define ELF_START_MMAP 0x20000000
1279 
1280 #define ELF_CLASS       ELFCLASS32
1281 #define ELF_ARCH        EM_XTENSA
1282 
1283 static inline void init_thread(struct target_pt_regs *regs,
1284                                struct image_info *infop)
1285 {
1286     regs->windowbase = 0;
1287     regs->windowstart = 1;
1288     regs->areg[1] = infop->start_stack;
1289     regs->pc = infop->entry;
1290 }
1291 
1292 /* See linux kernel: arch/xtensa/include/asm/elf.h.  */
1293 #define ELF_NREG 128
1294 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1295 
1296 enum {
1297     TARGET_REG_PC,
1298     TARGET_REG_PS,
1299     TARGET_REG_LBEG,
1300     TARGET_REG_LEND,
1301     TARGET_REG_LCOUNT,
1302     TARGET_REG_SAR,
1303     TARGET_REG_WINDOWSTART,
1304     TARGET_REG_WINDOWBASE,
1305     TARGET_REG_THREADPTR,
1306     TARGET_REG_AR0 = 64,
1307 };
1308 
1309 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1310                                const CPUXtensaState *env)
1311 {
1312     unsigned i;
1313 
1314     (*regs)[TARGET_REG_PC] = tswapreg(env->pc);
1315     (*regs)[TARGET_REG_PS] = tswapreg(env->sregs[PS] & ~PS_EXCM);
1316     (*regs)[TARGET_REG_LBEG] = tswapreg(env->sregs[LBEG]);
1317     (*regs)[TARGET_REG_LEND] = tswapreg(env->sregs[LEND]);
1318     (*regs)[TARGET_REG_LCOUNT] = tswapreg(env->sregs[LCOUNT]);
1319     (*regs)[TARGET_REG_SAR] = tswapreg(env->sregs[SAR]);
1320     (*regs)[TARGET_REG_WINDOWSTART] = tswapreg(env->sregs[WINDOW_START]);
1321     (*regs)[TARGET_REG_WINDOWBASE] = tswapreg(env->sregs[WINDOW_BASE]);
1322     (*regs)[TARGET_REG_THREADPTR] = tswapreg(env->uregs[THREADPTR]);
1323     xtensa_sync_phys_from_window((CPUXtensaState *)env);
1324     for (i = 0; i < env->config->nareg; ++i) {
1325         (*regs)[TARGET_REG_AR0 + i] = tswapreg(env->phys_regs[i]);
1326     }
1327 }
1328 
1329 #define USE_ELF_CORE_DUMP
1330 #define ELF_EXEC_PAGESIZE       4096
1331 
1332 #endif /* TARGET_XTENSA */
1333 
1334 #ifndef ELF_PLATFORM
1335 #define ELF_PLATFORM (NULL)
1336 #endif
1337 
1338 #ifndef ELF_MACHINE
1339 #define ELF_MACHINE ELF_ARCH
1340 #endif
1341 
1342 #ifndef elf_check_arch
1343 #define elf_check_arch(x) ((x) == ELF_ARCH)
1344 #endif
1345 
1346 #ifndef ELF_HWCAP
1347 #define ELF_HWCAP 0
1348 #endif
1349 
1350 #ifndef STACK_GROWS_DOWN
1351 #define STACK_GROWS_DOWN 1
1352 #endif
1353 
1354 #ifndef STACK_ALIGNMENT
1355 #define STACK_ALIGNMENT 16
1356 #endif
1357 
1358 #ifdef TARGET_ABI32
1359 #undef ELF_CLASS
1360 #define ELF_CLASS ELFCLASS32
1361 #undef bswaptls
1362 #define bswaptls(ptr) bswap32s(ptr)
1363 #endif
1364 
1365 #include "elf.h"
1366 
1367 struct exec
1368 {
1369     unsigned int a_info;   /* Use macros N_MAGIC, etc for access */
1370     unsigned int a_text;   /* length of text, in bytes */
1371     unsigned int a_data;   /* length of data, in bytes */
1372     unsigned int a_bss;    /* length of uninitialized data area, in bytes */
1373     unsigned int a_syms;   /* length of symbol table data in file, in bytes */
1374     unsigned int a_entry;  /* start address */
1375     unsigned int a_trsize; /* length of relocation info for text, in bytes */
1376     unsigned int a_drsize; /* length of relocation info for data, in bytes */
1377 };
1378 
1379 
1380 #define N_MAGIC(exec) ((exec).a_info & 0xffff)
1381 #define OMAGIC 0407
1382 #define NMAGIC 0410
1383 #define ZMAGIC 0413
1384 #define QMAGIC 0314
1385 
1386 /* Necessary parameters */
1387 #define TARGET_ELF_EXEC_PAGESIZE TARGET_PAGE_SIZE
1388 #define TARGET_ELF_PAGESTART(_v) ((_v) & \
1389                                  ~(abi_ulong)(TARGET_ELF_EXEC_PAGESIZE-1))
1390 #define TARGET_ELF_PAGEOFFSET(_v) ((_v) & (TARGET_ELF_EXEC_PAGESIZE-1))
1391 
1392 #define DLINFO_ITEMS 15
1393 
1394 static inline void memcpy_fromfs(void * to, const void * from, unsigned long n)
1395 {
1396     memcpy(to, from, n);
1397 }
1398 
1399 #ifdef BSWAP_NEEDED
1400 static void bswap_ehdr(struct elfhdr *ehdr)
1401 {
1402     bswap16s(&ehdr->e_type);            /* Object file type */
1403     bswap16s(&ehdr->e_machine);         /* Architecture */
1404     bswap32s(&ehdr->e_version);         /* Object file version */
1405     bswaptls(&ehdr->e_entry);           /* Entry point virtual address */
1406     bswaptls(&ehdr->e_phoff);           /* Program header table file offset */
1407     bswaptls(&ehdr->e_shoff);           /* Section header table file offset */
1408     bswap32s(&ehdr->e_flags);           /* Processor-specific flags */
1409     bswap16s(&ehdr->e_ehsize);          /* ELF header size in bytes */
1410     bswap16s(&ehdr->e_phentsize);       /* Program header table entry size */
1411     bswap16s(&ehdr->e_phnum);           /* Program header table entry count */
1412     bswap16s(&ehdr->e_shentsize);       /* Section header table entry size */
1413     bswap16s(&ehdr->e_shnum);           /* Section header table entry count */
1414     bswap16s(&ehdr->e_shstrndx);        /* Section header string table index */
1415 }
1416 
1417 static void bswap_phdr(struct elf_phdr *phdr, int phnum)
1418 {
1419     int i;
1420     for (i = 0; i < phnum; ++i, ++phdr) {
1421         bswap32s(&phdr->p_type);        /* Segment type */
1422         bswap32s(&phdr->p_flags);       /* Segment flags */
1423         bswaptls(&phdr->p_offset);      /* Segment file offset */
1424         bswaptls(&phdr->p_vaddr);       /* Segment virtual address */
1425         bswaptls(&phdr->p_paddr);       /* Segment physical address */
1426         bswaptls(&phdr->p_filesz);      /* Segment size in file */
1427         bswaptls(&phdr->p_memsz);       /* Segment size in memory */
1428         bswaptls(&phdr->p_align);       /* Segment alignment */
1429     }
1430 }
1431 
1432 static void bswap_shdr(struct elf_shdr *shdr, int shnum)
1433 {
1434     int i;
1435     for (i = 0; i < shnum; ++i, ++shdr) {
1436         bswap32s(&shdr->sh_name);
1437         bswap32s(&shdr->sh_type);
1438         bswaptls(&shdr->sh_flags);
1439         bswaptls(&shdr->sh_addr);
1440         bswaptls(&shdr->sh_offset);
1441         bswaptls(&shdr->sh_size);
1442         bswap32s(&shdr->sh_link);
1443         bswap32s(&shdr->sh_info);
1444         bswaptls(&shdr->sh_addralign);
1445         bswaptls(&shdr->sh_entsize);
1446     }
1447 }
1448 
1449 static void bswap_sym(struct elf_sym *sym)
1450 {
1451     bswap32s(&sym->st_name);
1452     bswaptls(&sym->st_value);
1453     bswaptls(&sym->st_size);
1454     bswap16s(&sym->st_shndx);
1455 }
1456 #else
1457 static inline void bswap_ehdr(struct elfhdr *ehdr) { }
1458 static inline void bswap_phdr(struct elf_phdr *phdr, int phnum) { }
1459 static inline void bswap_shdr(struct elf_shdr *shdr, int shnum) { }
1460 static inline void bswap_sym(struct elf_sym *sym) { }
1461 #endif
1462 
1463 #ifdef USE_ELF_CORE_DUMP
1464 static int elf_core_dump(int, const CPUArchState *);
1465 #endif /* USE_ELF_CORE_DUMP */
1466 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias);
1467 
1468 /* Verify the portions of EHDR within E_IDENT for the target.
1469    This can be performed before bswapping the entire header.  */
1470 static bool elf_check_ident(struct elfhdr *ehdr)
1471 {
1472     return (ehdr->e_ident[EI_MAG0] == ELFMAG0
1473             && ehdr->e_ident[EI_MAG1] == ELFMAG1
1474             && ehdr->e_ident[EI_MAG2] == ELFMAG2
1475             && ehdr->e_ident[EI_MAG3] == ELFMAG3
1476             && ehdr->e_ident[EI_CLASS] == ELF_CLASS
1477             && ehdr->e_ident[EI_DATA] == ELF_DATA
1478             && ehdr->e_ident[EI_VERSION] == EV_CURRENT);
1479 }
1480 
1481 /* Verify the portions of EHDR outside of E_IDENT for the target.
1482    This has to wait until after bswapping the header.  */
1483 static bool elf_check_ehdr(struct elfhdr *ehdr)
1484 {
1485     return (elf_check_arch(ehdr->e_machine)
1486             && ehdr->e_ehsize == sizeof(struct elfhdr)
1487             && ehdr->e_phentsize == sizeof(struct elf_phdr)
1488             && (ehdr->e_type == ET_EXEC || ehdr->e_type == ET_DYN));
1489 }
1490 
1491 /*
1492  * 'copy_elf_strings()' copies argument/envelope strings from user
1493  * memory to free pages in kernel mem. These are in a format ready
1494  * to be put directly into the top of new user memory.
1495  *
1496  */
1497 static abi_ulong copy_elf_strings(int argc, char **argv, char *scratch,
1498                                   abi_ulong p, abi_ulong stack_limit)
1499 {
1500     char *tmp;
1501     int len, i;
1502     abi_ulong top = p;
1503 
1504     if (!p) {
1505         return 0;       /* bullet-proofing */
1506     }
1507 
1508     if (STACK_GROWS_DOWN) {
1509         int offset = ((p - 1) % TARGET_PAGE_SIZE) + 1;
1510         for (i = argc - 1; i >= 0; --i) {
1511             tmp = argv[i];
1512             if (!tmp) {
1513                 fprintf(stderr, "VFS: argc is wrong");
1514                 exit(-1);
1515             }
1516             len = strlen(tmp) + 1;
1517             tmp += len;
1518 
1519             if (len > (p - stack_limit)) {
1520                 return 0;
1521             }
1522             while (len) {
1523                 int bytes_to_copy = (len > offset) ? offset : len;
1524                 tmp -= bytes_to_copy;
1525                 p -= bytes_to_copy;
1526                 offset -= bytes_to_copy;
1527                 len -= bytes_to_copy;
1528 
1529                 memcpy_fromfs(scratch + offset, tmp, bytes_to_copy);
1530 
1531                 if (offset == 0) {
1532                     memcpy_to_target(p, scratch, top - p);
1533                     top = p;
1534                     offset = TARGET_PAGE_SIZE;
1535                 }
1536             }
1537         }
1538         if (p != top) {
1539             memcpy_to_target(p, scratch + offset, top - p);
1540         }
1541     } else {
1542         int remaining = TARGET_PAGE_SIZE - (p % TARGET_PAGE_SIZE);
1543         for (i = 0; i < argc; ++i) {
1544             tmp = argv[i];
1545             if (!tmp) {
1546                 fprintf(stderr, "VFS: argc is wrong");
1547                 exit(-1);
1548             }
1549             len = strlen(tmp) + 1;
1550             if (len > (stack_limit - p)) {
1551                 return 0;
1552             }
1553             while (len) {
1554                 int bytes_to_copy = (len > remaining) ? remaining : len;
1555 
1556                 memcpy_fromfs(scratch + (p - top), tmp, bytes_to_copy);
1557 
1558                 tmp += bytes_to_copy;
1559                 remaining -= bytes_to_copy;
1560                 p += bytes_to_copy;
1561                 len -= bytes_to_copy;
1562 
1563                 if (remaining == 0) {
1564                     memcpy_to_target(top, scratch, p - top);
1565                     top = p;
1566                     remaining = TARGET_PAGE_SIZE;
1567                 }
1568             }
1569         }
1570         if (p != top) {
1571             memcpy_to_target(top, scratch, p - top);
1572         }
1573     }
1574 
1575     return p;
1576 }
1577 
1578 /* Older linux kernels provide up to MAX_ARG_PAGES (default: 32) of
1579  * argument/environment space. Newer kernels (>2.6.33) allow more,
1580  * dependent on stack size, but guarantee at least 32 pages for
1581  * backwards compatibility.
1582  */
1583 #define STACK_LOWER_LIMIT (32 * TARGET_PAGE_SIZE)
1584 
1585 static abi_ulong setup_arg_pages(struct linux_binprm *bprm,
1586                                  struct image_info *info)
1587 {
1588     abi_ulong size, error, guard;
1589 
1590     size = guest_stack_size;
1591     if (size < STACK_LOWER_LIMIT) {
1592         size = STACK_LOWER_LIMIT;
1593     }
1594     guard = TARGET_PAGE_SIZE;
1595     if (guard < qemu_real_host_page_size) {
1596         guard = qemu_real_host_page_size;
1597     }
1598 
1599     error = target_mmap(0, size + guard, PROT_READ | PROT_WRITE,
1600                         MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
1601     if (error == -1) {
1602         perror("mmap stack");
1603         exit(-1);
1604     }
1605 
1606     /* We reserve one extra page at the top of the stack as guard.  */
1607     if (STACK_GROWS_DOWN) {
1608         target_mprotect(error, guard, PROT_NONE);
1609         info->stack_limit = error + guard;
1610         return info->stack_limit + size - sizeof(void *);
1611     } else {
1612         target_mprotect(error + size, guard, PROT_NONE);
1613         info->stack_limit = error + size;
1614         return error;
1615     }
1616 }
1617 
1618 /* Map and zero the bss.  We need to explicitly zero any fractional pages
1619    after the data section (i.e. bss).  */
1620 static void zero_bss(abi_ulong elf_bss, abi_ulong last_bss, int prot)
1621 {
1622     uintptr_t host_start, host_map_start, host_end;
1623 
1624     last_bss = TARGET_PAGE_ALIGN(last_bss);
1625 
1626     /* ??? There is confusion between qemu_real_host_page_size and
1627        qemu_host_page_size here and elsewhere in target_mmap, which
1628        may lead to the end of the data section mapping from the file
1629        not being mapped.  At least there was an explicit test and
1630        comment for that here, suggesting that "the file size must
1631        be known".  The comment probably pre-dates the introduction
1632        of the fstat system call in target_mmap which does in fact
1633        find out the size.  What isn't clear is if the workaround
1634        here is still actually needed.  For now, continue with it,
1635        but merge it with the "normal" mmap that would allocate the bss.  */
1636 
1637     host_start = (uintptr_t) g2h(elf_bss);
1638     host_end = (uintptr_t) g2h(last_bss);
1639     host_map_start = REAL_HOST_PAGE_ALIGN(host_start);
1640 
1641     if (host_map_start < host_end) {
1642         void *p = mmap((void *)host_map_start, host_end - host_map_start,
1643                        prot, MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
1644         if (p == MAP_FAILED) {
1645             perror("cannot mmap brk");
1646             exit(-1);
1647         }
1648     }
1649 
1650     /* Ensure that the bss page(s) are valid */
1651     if ((page_get_flags(last_bss-1) & prot) != prot) {
1652         page_set_flags(elf_bss & TARGET_PAGE_MASK, last_bss, prot | PAGE_VALID);
1653     }
1654 
1655     if (host_start < host_map_start) {
1656         memset((void *)host_start, 0, host_map_start - host_start);
1657     }
1658 }
1659 
1660 #ifdef CONFIG_USE_FDPIC
1661 static abi_ulong loader_build_fdpic_loadmap(struct image_info *info, abi_ulong sp)
1662 {
1663     uint16_t n;
1664     struct elf32_fdpic_loadseg *loadsegs = info->loadsegs;
1665 
1666     /* elf32_fdpic_loadseg */
1667     n = info->nsegs;
1668     while (n--) {
1669         sp -= 12;
1670         put_user_u32(loadsegs[n].addr, sp+0);
1671         put_user_u32(loadsegs[n].p_vaddr, sp+4);
1672         put_user_u32(loadsegs[n].p_memsz, sp+8);
1673     }
1674 
1675     /* elf32_fdpic_loadmap */
1676     sp -= 4;
1677     put_user_u16(0, sp+0); /* version */
1678     put_user_u16(info->nsegs, sp+2); /* nsegs */
1679 
1680     info->personality = PER_LINUX_FDPIC;
1681     info->loadmap_addr = sp;
1682 
1683     return sp;
1684 }
1685 #endif
1686 
1687 static abi_ulong create_elf_tables(abi_ulong p, int argc, int envc,
1688                                    struct elfhdr *exec,
1689                                    struct image_info *info,
1690                                    struct image_info *interp_info)
1691 {
1692     abi_ulong sp;
1693     abi_ulong u_argc, u_argv, u_envp, u_auxv;
1694     int size;
1695     int i;
1696     abi_ulong u_rand_bytes;
1697     uint8_t k_rand_bytes[16];
1698     abi_ulong u_platform;
1699     const char *k_platform;
1700     const int n = sizeof(elf_addr_t);
1701 
1702     sp = p;
1703 
1704 #ifdef CONFIG_USE_FDPIC
1705     /* Needs to be before we load the env/argc/... */
1706     if (elf_is_fdpic(exec)) {
1707         /* Need 4 byte alignment for these structs */
1708         sp &= ~3;
1709         sp = loader_build_fdpic_loadmap(info, sp);
1710         info->other_info = interp_info;
1711         if (interp_info) {
1712             interp_info->other_info = info;
1713             sp = loader_build_fdpic_loadmap(interp_info, sp);
1714         }
1715     }
1716 #endif
1717 
1718     u_platform = 0;
1719     k_platform = ELF_PLATFORM;
1720     if (k_platform) {
1721         size_t len = strlen(k_platform) + 1;
1722         if (STACK_GROWS_DOWN) {
1723             sp -= (len + n - 1) & ~(n - 1);
1724             u_platform = sp;
1725             /* FIXME - check return value of memcpy_to_target() for failure */
1726             memcpy_to_target(sp, k_platform, len);
1727         } else {
1728             memcpy_to_target(sp, k_platform, len);
1729             u_platform = sp;
1730             sp += len + 1;
1731         }
1732     }
1733 
1734     /* Provide 16 byte alignment for the PRNG, and basic alignment for
1735      * the argv and envp pointers.
1736      */
1737     if (STACK_GROWS_DOWN) {
1738         sp = QEMU_ALIGN_DOWN(sp, 16);
1739     } else {
1740         sp = QEMU_ALIGN_UP(sp, 16);
1741     }
1742 
1743     /*
1744      * Generate 16 random bytes for userspace PRNG seeding (not
1745      * cryptically secure but it's not the aim of QEMU).
1746      */
1747     for (i = 0; i < 16; i++) {
1748         k_rand_bytes[i] = rand();
1749     }
1750     if (STACK_GROWS_DOWN) {
1751         sp -= 16;
1752         u_rand_bytes = sp;
1753         /* FIXME - check return value of memcpy_to_target() for failure */
1754         memcpy_to_target(sp, k_rand_bytes, 16);
1755     } else {
1756         memcpy_to_target(sp, k_rand_bytes, 16);
1757         u_rand_bytes = sp;
1758         sp += 16;
1759     }
1760 
1761     size = (DLINFO_ITEMS + 1) * 2;
1762     if (k_platform)
1763         size += 2;
1764 #ifdef DLINFO_ARCH_ITEMS
1765     size += DLINFO_ARCH_ITEMS * 2;
1766 #endif
1767 #ifdef ELF_HWCAP2
1768     size += 2;
1769 #endif
1770     info->auxv_len = size * n;
1771 
1772     size += envc + argc + 2;
1773     size += 1;  /* argc itself */
1774     size *= n;
1775 
1776     /* Allocate space and finalize stack alignment for entry now.  */
1777     if (STACK_GROWS_DOWN) {
1778         u_argc = QEMU_ALIGN_DOWN(sp - size, STACK_ALIGNMENT);
1779         sp = u_argc;
1780     } else {
1781         u_argc = sp;
1782         sp = QEMU_ALIGN_UP(sp + size, STACK_ALIGNMENT);
1783     }
1784 
1785     u_argv = u_argc + n;
1786     u_envp = u_argv + (argc + 1) * n;
1787     u_auxv = u_envp + (envc + 1) * n;
1788     info->saved_auxv = u_auxv;
1789     info->arg_start = u_argv;
1790     info->arg_end = u_argv + argc * n;
1791 
1792     /* This is correct because Linux defines
1793      * elf_addr_t as Elf32_Off / Elf64_Off
1794      */
1795 #define NEW_AUX_ENT(id, val) do {               \
1796         put_user_ual(id, u_auxv);  u_auxv += n; \
1797         put_user_ual(val, u_auxv); u_auxv += n; \
1798     } while(0)
1799 
1800 #ifdef ARCH_DLINFO
1801     /*
1802      * ARCH_DLINFO must come first so platform specific code can enforce
1803      * special alignment requirements on the AUXV if necessary (eg. PPC).
1804      */
1805     ARCH_DLINFO;
1806 #endif
1807     /* There must be exactly DLINFO_ITEMS entries here, or the assert
1808      * on info->auxv_len will trigger.
1809      */
1810     NEW_AUX_ENT(AT_PHDR, (abi_ulong)(info->load_addr + exec->e_phoff));
1811     NEW_AUX_ENT(AT_PHENT, (abi_ulong)(sizeof (struct elf_phdr)));
1812     NEW_AUX_ENT(AT_PHNUM, (abi_ulong)(exec->e_phnum));
1813     NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(MAX(TARGET_PAGE_SIZE, getpagesize())));
1814     NEW_AUX_ENT(AT_BASE, (abi_ulong)(interp_info ? interp_info->load_addr : 0));
1815     NEW_AUX_ENT(AT_FLAGS, (abi_ulong)0);
1816     NEW_AUX_ENT(AT_ENTRY, info->entry);
1817     NEW_AUX_ENT(AT_UID, (abi_ulong) getuid());
1818     NEW_AUX_ENT(AT_EUID, (abi_ulong) geteuid());
1819     NEW_AUX_ENT(AT_GID, (abi_ulong) getgid());
1820     NEW_AUX_ENT(AT_EGID, (abi_ulong) getegid());
1821     NEW_AUX_ENT(AT_HWCAP, (abi_ulong) ELF_HWCAP);
1822     NEW_AUX_ENT(AT_CLKTCK, (abi_ulong) sysconf(_SC_CLK_TCK));
1823     NEW_AUX_ENT(AT_RANDOM, (abi_ulong) u_rand_bytes);
1824     NEW_AUX_ENT(AT_SECURE, (abi_ulong) qemu_getauxval(AT_SECURE));
1825 
1826 #ifdef ELF_HWCAP2
1827     NEW_AUX_ENT(AT_HWCAP2, (abi_ulong) ELF_HWCAP2);
1828 #endif
1829 
1830     if (u_platform) {
1831         NEW_AUX_ENT(AT_PLATFORM, u_platform);
1832     }
1833     NEW_AUX_ENT (AT_NULL, 0);
1834 #undef NEW_AUX_ENT
1835 
1836     /* Check that our initial calculation of the auxv length matches how much
1837      * we actually put into it.
1838      */
1839     assert(info->auxv_len == u_auxv - info->saved_auxv);
1840 
1841     put_user_ual(argc, u_argc);
1842 
1843     p = info->arg_strings;
1844     for (i = 0; i < argc; ++i) {
1845         put_user_ual(p, u_argv);
1846         u_argv += n;
1847         p += target_strlen(p) + 1;
1848     }
1849     put_user_ual(0, u_argv);
1850 
1851     p = info->env_strings;
1852     for (i = 0; i < envc; ++i) {
1853         put_user_ual(p, u_envp);
1854         u_envp += n;
1855         p += target_strlen(p) + 1;
1856     }
1857     put_user_ual(0, u_envp);
1858 
1859     return sp;
1860 }
1861 
1862 unsigned long init_guest_space(unsigned long host_start,
1863                                unsigned long host_size,
1864                                unsigned long guest_start,
1865                                bool fixed)
1866 {
1867     unsigned long current_start, aligned_start;
1868     int flags;
1869 
1870     assert(host_start || host_size);
1871 
1872     /* If just a starting address is given, then just verify that
1873      * address.  */
1874     if (host_start && !host_size) {
1875 #if defined(TARGET_ARM) && !defined(TARGET_AARCH64)
1876         if (init_guest_commpage(host_start, host_size) != 1) {
1877             return (unsigned long)-1;
1878         }
1879 #endif
1880         return host_start;
1881     }
1882 
1883     /* Setup the initial flags and start address.  */
1884     current_start = host_start & qemu_host_page_mask;
1885     flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE;
1886     if (fixed) {
1887         flags |= MAP_FIXED;
1888     }
1889 
1890     /* Otherwise, a non-zero size region of memory needs to be mapped
1891      * and validated.  */
1892 
1893 #if defined(TARGET_ARM) && !defined(TARGET_AARCH64)
1894     /* On 32-bit ARM, we need to map not just the usable memory, but
1895      * also the commpage.  Try to find a suitable place by allocating
1896      * a big chunk for all of it.  If host_start, then the naive
1897      * strategy probably does good enough.
1898      */
1899     if (!host_start) {
1900         unsigned long guest_full_size, host_full_size, real_start;
1901 
1902         guest_full_size =
1903             (0xffff0f00 & qemu_host_page_mask) + qemu_host_page_size;
1904         host_full_size = guest_full_size - guest_start;
1905         real_start = (unsigned long)
1906             mmap(NULL, host_full_size, PROT_NONE, flags, -1, 0);
1907         if (real_start == (unsigned long)-1) {
1908             if (host_size < host_full_size - qemu_host_page_size) {
1909                 /* We failed to map a continous segment, but we're
1910                  * allowed to have a gap between the usable memory and
1911                  * the commpage where other things can be mapped.
1912                  * This sparseness gives us more flexibility to find
1913                  * an address range.
1914                  */
1915                 goto naive;
1916             }
1917             return (unsigned long)-1;
1918         }
1919         munmap((void *)real_start, host_full_size);
1920         if (real_start & ~qemu_host_page_mask) {
1921             /* The same thing again, but with an extra qemu_host_page_size
1922              * so that we can shift around alignment.
1923              */
1924             unsigned long real_size = host_full_size + qemu_host_page_size;
1925             real_start = (unsigned long)
1926                 mmap(NULL, real_size, PROT_NONE, flags, -1, 0);
1927             if (real_start == (unsigned long)-1) {
1928                 if (host_size < host_full_size - qemu_host_page_size) {
1929                     goto naive;
1930                 }
1931                 return (unsigned long)-1;
1932             }
1933             munmap((void *)real_start, real_size);
1934             real_start = HOST_PAGE_ALIGN(real_start);
1935         }
1936         current_start = real_start;
1937     }
1938  naive:
1939 #endif
1940 
1941     while (1) {
1942         unsigned long real_start, real_size, aligned_size;
1943         aligned_size = real_size = host_size;
1944 
1945         /* Do not use mmap_find_vma here because that is limited to the
1946          * guest address space.  We are going to make the
1947          * guest address space fit whatever we're given.
1948          */
1949         real_start = (unsigned long)
1950             mmap((void *)current_start, host_size, PROT_NONE, flags, -1, 0);
1951         if (real_start == (unsigned long)-1) {
1952             return (unsigned long)-1;
1953         }
1954 
1955         /* Check to see if the address is valid.  */
1956         if (host_start && real_start != current_start) {
1957             goto try_again;
1958         }
1959 
1960         /* Ensure the address is properly aligned.  */
1961         if (real_start & ~qemu_host_page_mask) {
1962             /* Ideally, we adjust like
1963              *
1964              *    pages: [  ][  ][  ][  ][  ]
1965              *      old:   [   real   ]
1966              *             [ aligned  ]
1967              *      new:   [     real     ]
1968              *               [ aligned  ]
1969              *
1970              * But if there is something else mapped right after it,
1971              * then obviously it won't have room to grow, and the
1972              * kernel will put the new larger real someplace else with
1973              * unknown alignment (if we made it to here, then
1974              * fixed=false).  Which is why we grow real by a full page
1975              * size, instead of by part of one; so that even if we get
1976              * moved, we can still guarantee alignment.  But this does
1977              * mean that there is a padding of < 1 page both before
1978              * and after the aligned range; the "after" could could
1979              * cause problems for ARM emulation where it could butt in
1980              * to where we need to put the commpage.
1981              */
1982             munmap((void *)real_start, host_size);
1983             real_size = aligned_size + qemu_host_page_size;
1984             real_start = (unsigned long)
1985                 mmap((void *)real_start, real_size, PROT_NONE, flags, -1, 0);
1986             if (real_start == (unsigned long)-1) {
1987                 return (unsigned long)-1;
1988             }
1989             aligned_start = HOST_PAGE_ALIGN(real_start);
1990         } else {
1991             aligned_start = real_start;
1992         }
1993 
1994 #if defined(TARGET_ARM) && !defined(TARGET_AARCH64)
1995         /* On 32-bit ARM, we need to also be able to map the commpage.  */
1996         int valid = init_guest_commpage(aligned_start - guest_start,
1997                                         aligned_size + guest_start);
1998         if (valid == -1) {
1999             munmap((void *)real_start, real_size);
2000             return (unsigned long)-1;
2001         } else if (valid == 0) {
2002             goto try_again;
2003         }
2004 #endif
2005 
2006         /* If nothing has said `return -1` or `goto try_again` yet,
2007          * then the address we have is good.
2008          */
2009         break;
2010 
2011     try_again:
2012         /* That address didn't work.  Unmap and try a different one.
2013          * The address the host picked because is typically right at
2014          * the top of the host address space and leaves the guest with
2015          * no usable address space.  Resort to a linear search.  We
2016          * already compensated for mmap_min_addr, so this should not
2017          * happen often.  Probably means we got unlucky and host
2018          * address space randomization put a shared library somewhere
2019          * inconvenient.
2020          *
2021          * This is probably a good strategy if host_start, but is
2022          * probably a bad strategy if not, which means we got here
2023          * because of trouble with ARM commpage setup.
2024          */
2025         munmap((void *)real_start, real_size);
2026         current_start += qemu_host_page_size;
2027         if (host_start == current_start) {
2028             /* Theoretically possible if host doesn't have any suitably
2029              * aligned areas.  Normally the first mmap will fail.
2030              */
2031             return (unsigned long)-1;
2032         }
2033     }
2034 
2035     qemu_log_mask(CPU_LOG_PAGE, "Reserved 0x%lx bytes of guest address space\n", host_size);
2036 
2037     return aligned_start;
2038 }
2039 
2040 static void probe_guest_base(const char *image_name,
2041                              abi_ulong loaddr, abi_ulong hiaddr)
2042 {
2043     /* Probe for a suitable guest base address, if the user has not set
2044      * it explicitly, and set guest_base appropriately.
2045      * In case of error we will print a suitable message and exit.
2046      */
2047     const char *errmsg;
2048     if (!have_guest_base && !reserved_va) {
2049         unsigned long host_start, real_start, host_size;
2050 
2051         /* Round addresses to page boundaries.  */
2052         loaddr &= qemu_host_page_mask;
2053         hiaddr = HOST_PAGE_ALIGN(hiaddr);
2054 
2055         if (loaddr < mmap_min_addr) {
2056             host_start = HOST_PAGE_ALIGN(mmap_min_addr);
2057         } else {
2058             host_start = loaddr;
2059             if (host_start != loaddr) {
2060                 errmsg = "Address overflow loading ELF binary";
2061                 goto exit_errmsg;
2062             }
2063         }
2064         host_size = hiaddr - loaddr;
2065 
2066         /* Setup the initial guest memory space with ranges gleaned from
2067          * the ELF image that is being loaded.
2068          */
2069         real_start = init_guest_space(host_start, host_size, loaddr, false);
2070         if (real_start == (unsigned long)-1) {
2071             errmsg = "Unable to find space for application";
2072             goto exit_errmsg;
2073         }
2074         guest_base = real_start - loaddr;
2075 
2076         qemu_log_mask(CPU_LOG_PAGE, "Relocating guest address space from 0x"
2077                       TARGET_ABI_FMT_lx " to 0x%lx\n",
2078                       loaddr, real_start);
2079     }
2080     return;
2081 
2082 exit_errmsg:
2083     fprintf(stderr, "%s: %s\n", image_name, errmsg);
2084     exit(-1);
2085 }
2086 
2087 
2088 /* Load an ELF image into the address space.
2089 
2090    IMAGE_NAME is the filename of the image, to use in error messages.
2091    IMAGE_FD is the open file descriptor for the image.
2092 
2093    BPRM_BUF is a copy of the beginning of the file; this of course
2094    contains the elf file header at offset 0.  It is assumed that this
2095    buffer is sufficiently aligned to present no problems to the host
2096    in accessing data at aligned offsets within the buffer.
2097 
2098    On return: INFO values will be filled in, as necessary or available.  */
2099 
2100 static void load_elf_image(const char *image_name, int image_fd,
2101                            struct image_info *info, char **pinterp_name,
2102                            char bprm_buf[BPRM_BUF_SIZE])
2103 {
2104     struct elfhdr *ehdr = (struct elfhdr *)bprm_buf;
2105     struct elf_phdr *phdr;
2106     abi_ulong load_addr, load_bias, loaddr, hiaddr, error;
2107     int i, retval;
2108     const char *errmsg;
2109 
2110     /* First of all, some simple consistency checks */
2111     errmsg = "Invalid ELF image for this architecture";
2112     if (!elf_check_ident(ehdr)) {
2113         goto exit_errmsg;
2114     }
2115     bswap_ehdr(ehdr);
2116     if (!elf_check_ehdr(ehdr)) {
2117         goto exit_errmsg;
2118     }
2119 
2120     i = ehdr->e_phnum * sizeof(struct elf_phdr);
2121     if (ehdr->e_phoff + i <= BPRM_BUF_SIZE) {
2122         phdr = (struct elf_phdr *)(bprm_buf + ehdr->e_phoff);
2123     } else {
2124         phdr = (struct elf_phdr *) alloca(i);
2125         retval = pread(image_fd, phdr, i, ehdr->e_phoff);
2126         if (retval != i) {
2127             goto exit_read;
2128         }
2129     }
2130     bswap_phdr(phdr, ehdr->e_phnum);
2131 
2132 #ifdef CONFIG_USE_FDPIC
2133     info->nsegs = 0;
2134     info->pt_dynamic_addr = 0;
2135 #endif
2136 
2137     mmap_lock();
2138 
2139     /* Find the maximum size of the image and allocate an appropriate
2140        amount of memory to handle that.  */
2141     loaddr = -1, hiaddr = 0;
2142     for (i = 0; i < ehdr->e_phnum; ++i) {
2143         if (phdr[i].p_type == PT_LOAD) {
2144             abi_ulong a = phdr[i].p_vaddr - phdr[i].p_offset;
2145             if (a < loaddr) {
2146                 loaddr = a;
2147             }
2148             a = phdr[i].p_vaddr + phdr[i].p_memsz;
2149             if (a > hiaddr) {
2150                 hiaddr = a;
2151             }
2152 #ifdef CONFIG_USE_FDPIC
2153             ++info->nsegs;
2154 #endif
2155         }
2156     }
2157 
2158     load_addr = loaddr;
2159     if (ehdr->e_type == ET_DYN) {
2160         /* The image indicates that it can be loaded anywhere.  Find a
2161            location that can hold the memory space required.  If the
2162            image is pre-linked, LOADDR will be non-zero.  Since we do
2163            not supply MAP_FIXED here we'll use that address if and
2164            only if it remains available.  */
2165         load_addr = target_mmap(loaddr, hiaddr - loaddr, PROT_NONE,
2166                                 MAP_PRIVATE | MAP_ANON | MAP_NORESERVE,
2167                                 -1, 0);
2168         if (load_addr == -1) {
2169             goto exit_perror;
2170         }
2171     } else if (pinterp_name != NULL) {
2172         /* This is the main executable.  Make sure that the low
2173            address does not conflict with MMAP_MIN_ADDR or the
2174            QEMU application itself.  */
2175         probe_guest_base(image_name, loaddr, hiaddr);
2176     }
2177     load_bias = load_addr - loaddr;
2178 
2179 #ifdef CONFIG_USE_FDPIC
2180     {
2181         struct elf32_fdpic_loadseg *loadsegs = info->loadsegs =
2182             g_malloc(sizeof(*loadsegs) * info->nsegs);
2183 
2184         for (i = 0; i < ehdr->e_phnum; ++i) {
2185             switch (phdr[i].p_type) {
2186             case PT_DYNAMIC:
2187                 info->pt_dynamic_addr = phdr[i].p_vaddr + load_bias;
2188                 break;
2189             case PT_LOAD:
2190                 loadsegs->addr = phdr[i].p_vaddr + load_bias;
2191                 loadsegs->p_vaddr = phdr[i].p_vaddr;
2192                 loadsegs->p_memsz = phdr[i].p_memsz;
2193                 ++loadsegs;
2194                 break;
2195             }
2196         }
2197     }
2198 #endif
2199 
2200     info->load_bias = load_bias;
2201     info->load_addr = load_addr;
2202     info->entry = ehdr->e_entry + load_bias;
2203     info->start_code = -1;
2204     info->end_code = 0;
2205     info->start_data = -1;
2206     info->end_data = 0;
2207     info->brk = 0;
2208     info->elf_flags = ehdr->e_flags;
2209 
2210     for (i = 0; i < ehdr->e_phnum; i++) {
2211         struct elf_phdr *eppnt = phdr + i;
2212         if (eppnt->p_type == PT_LOAD) {
2213             abi_ulong vaddr, vaddr_po, vaddr_ps, vaddr_ef, vaddr_em;
2214             int elf_prot = 0;
2215 
2216             if (eppnt->p_flags & PF_R) elf_prot =  PROT_READ;
2217             if (eppnt->p_flags & PF_W) elf_prot |= PROT_WRITE;
2218             if (eppnt->p_flags & PF_X) elf_prot |= PROT_EXEC;
2219 
2220             vaddr = load_bias + eppnt->p_vaddr;
2221             vaddr_po = TARGET_ELF_PAGEOFFSET(vaddr);
2222             vaddr_ps = TARGET_ELF_PAGESTART(vaddr);
2223 
2224             error = target_mmap(vaddr_ps, eppnt->p_filesz + vaddr_po,
2225                                 elf_prot, MAP_PRIVATE | MAP_FIXED,
2226                                 image_fd, eppnt->p_offset - vaddr_po);
2227             if (error == -1) {
2228                 goto exit_perror;
2229             }
2230 
2231             vaddr_ef = vaddr + eppnt->p_filesz;
2232             vaddr_em = vaddr + eppnt->p_memsz;
2233 
2234             /* If the load segment requests extra zeros (e.g. bss), map it.  */
2235             if (vaddr_ef < vaddr_em) {
2236                 zero_bss(vaddr_ef, vaddr_em, elf_prot);
2237             }
2238 
2239             /* Find the full program boundaries.  */
2240             if (elf_prot & PROT_EXEC) {
2241                 if (vaddr < info->start_code) {
2242                     info->start_code = vaddr;
2243                 }
2244                 if (vaddr_ef > info->end_code) {
2245                     info->end_code = vaddr_ef;
2246                 }
2247             }
2248             if (elf_prot & PROT_WRITE) {
2249                 if (vaddr < info->start_data) {
2250                     info->start_data = vaddr;
2251                 }
2252                 if (vaddr_ef > info->end_data) {
2253                     info->end_data = vaddr_ef;
2254                 }
2255                 if (vaddr_em > info->brk) {
2256                     info->brk = vaddr_em;
2257                 }
2258             }
2259         } else if (eppnt->p_type == PT_INTERP && pinterp_name) {
2260             char *interp_name;
2261 
2262             if (*pinterp_name) {
2263                 errmsg = "Multiple PT_INTERP entries";
2264                 goto exit_errmsg;
2265             }
2266             interp_name = malloc(eppnt->p_filesz);
2267             if (!interp_name) {
2268                 goto exit_perror;
2269             }
2270 
2271             if (eppnt->p_offset + eppnt->p_filesz <= BPRM_BUF_SIZE) {
2272                 memcpy(interp_name, bprm_buf + eppnt->p_offset,
2273                        eppnt->p_filesz);
2274             } else {
2275                 retval = pread(image_fd, interp_name, eppnt->p_filesz,
2276                                eppnt->p_offset);
2277                 if (retval != eppnt->p_filesz) {
2278                     goto exit_perror;
2279                 }
2280             }
2281             if (interp_name[eppnt->p_filesz - 1] != 0) {
2282                 errmsg = "Invalid PT_INTERP entry";
2283                 goto exit_errmsg;
2284             }
2285             *pinterp_name = interp_name;
2286         }
2287     }
2288 
2289     if (info->end_data == 0) {
2290         info->start_data = info->end_code;
2291         info->end_data = info->end_code;
2292         info->brk = info->end_code;
2293     }
2294 
2295     if (qemu_log_enabled()) {
2296         load_symbols(ehdr, image_fd, load_bias);
2297     }
2298 
2299     mmap_unlock();
2300 
2301     close(image_fd);
2302     return;
2303 
2304  exit_read:
2305     if (retval >= 0) {
2306         errmsg = "Incomplete read of file header";
2307         goto exit_errmsg;
2308     }
2309  exit_perror:
2310     errmsg = strerror(errno);
2311  exit_errmsg:
2312     fprintf(stderr, "%s: %s\n", image_name, errmsg);
2313     exit(-1);
2314 }
2315 
2316 static void load_elf_interp(const char *filename, struct image_info *info,
2317                             char bprm_buf[BPRM_BUF_SIZE])
2318 {
2319     int fd, retval;
2320 
2321     fd = open(path(filename), O_RDONLY);
2322     if (fd < 0) {
2323         goto exit_perror;
2324     }
2325 
2326     retval = read(fd, bprm_buf, BPRM_BUF_SIZE);
2327     if (retval < 0) {
2328         goto exit_perror;
2329     }
2330     if (retval < BPRM_BUF_SIZE) {
2331         memset(bprm_buf + retval, 0, BPRM_BUF_SIZE - retval);
2332     }
2333 
2334     load_elf_image(filename, fd, info, NULL, bprm_buf);
2335     return;
2336 
2337  exit_perror:
2338     fprintf(stderr, "%s: %s\n", filename, strerror(errno));
2339     exit(-1);
2340 }
2341 
2342 static int symfind(const void *s0, const void *s1)
2343 {
2344     target_ulong addr = *(target_ulong *)s0;
2345     struct elf_sym *sym = (struct elf_sym *)s1;
2346     int result = 0;
2347     if (addr < sym->st_value) {
2348         result = -1;
2349     } else if (addr >= sym->st_value + sym->st_size) {
2350         result = 1;
2351     }
2352     return result;
2353 }
2354 
2355 static const char *lookup_symbolxx(struct syminfo *s, target_ulong orig_addr)
2356 {
2357 #if ELF_CLASS == ELFCLASS32
2358     struct elf_sym *syms = s->disas_symtab.elf32;
2359 #else
2360     struct elf_sym *syms = s->disas_symtab.elf64;
2361 #endif
2362 
2363     // binary search
2364     struct elf_sym *sym;
2365 
2366     sym = bsearch(&orig_addr, syms, s->disas_num_syms, sizeof(*syms), symfind);
2367     if (sym != NULL) {
2368         return s->disas_strtab + sym->st_name;
2369     }
2370 
2371     return "";
2372 }
2373 
2374 /* FIXME: This should use elf_ops.h  */
2375 static int symcmp(const void *s0, const void *s1)
2376 {
2377     struct elf_sym *sym0 = (struct elf_sym *)s0;
2378     struct elf_sym *sym1 = (struct elf_sym *)s1;
2379     return (sym0->st_value < sym1->st_value)
2380         ? -1
2381         : ((sym0->st_value > sym1->st_value) ? 1 : 0);
2382 }
2383 
2384 /* Best attempt to load symbols from this ELF object. */
2385 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias)
2386 {
2387     int i, shnum, nsyms, sym_idx = 0, str_idx = 0;
2388     uint64_t segsz;
2389     struct elf_shdr *shdr;
2390     char *strings = NULL;
2391     struct syminfo *s = NULL;
2392     struct elf_sym *new_syms, *syms = NULL;
2393 
2394     shnum = hdr->e_shnum;
2395     i = shnum * sizeof(struct elf_shdr);
2396     shdr = (struct elf_shdr *)alloca(i);
2397     if (pread(fd, shdr, i, hdr->e_shoff) != i) {
2398         return;
2399     }
2400 
2401     bswap_shdr(shdr, shnum);
2402     for (i = 0; i < shnum; ++i) {
2403         if (shdr[i].sh_type == SHT_SYMTAB) {
2404             sym_idx = i;
2405             str_idx = shdr[i].sh_link;
2406             goto found;
2407         }
2408     }
2409 
2410     /* There will be no symbol table if the file was stripped.  */
2411     return;
2412 
2413  found:
2414     /* Now know where the strtab and symtab are.  Snarf them.  */
2415     s = g_try_new(struct syminfo, 1);
2416     if (!s) {
2417         goto give_up;
2418     }
2419 
2420     segsz = shdr[str_idx].sh_size;
2421     s->disas_strtab = strings = g_try_malloc(segsz);
2422     if (!strings ||
2423         pread(fd, strings, segsz, shdr[str_idx].sh_offset) != segsz) {
2424         goto give_up;
2425     }
2426 
2427     segsz = shdr[sym_idx].sh_size;
2428     syms = g_try_malloc(segsz);
2429     if (!syms || pread(fd, syms, segsz, shdr[sym_idx].sh_offset) != segsz) {
2430         goto give_up;
2431     }
2432 
2433     if (segsz / sizeof(struct elf_sym) > INT_MAX) {
2434         /* Implausibly large symbol table: give up rather than ploughing
2435          * on with the number of symbols calculation overflowing
2436          */
2437         goto give_up;
2438     }
2439     nsyms = segsz / sizeof(struct elf_sym);
2440     for (i = 0; i < nsyms; ) {
2441         bswap_sym(syms + i);
2442         /* Throw away entries which we do not need.  */
2443         if (syms[i].st_shndx == SHN_UNDEF
2444             || syms[i].st_shndx >= SHN_LORESERVE
2445             || ELF_ST_TYPE(syms[i].st_info) != STT_FUNC) {
2446             if (i < --nsyms) {
2447                 syms[i] = syms[nsyms];
2448             }
2449         } else {
2450 #if defined(TARGET_ARM) || defined (TARGET_MIPS)
2451             /* The bottom address bit marks a Thumb or MIPS16 symbol.  */
2452             syms[i].st_value &= ~(target_ulong)1;
2453 #endif
2454             syms[i].st_value += load_bias;
2455             i++;
2456         }
2457     }
2458 
2459     /* No "useful" symbol.  */
2460     if (nsyms == 0) {
2461         goto give_up;
2462     }
2463 
2464     /* Attempt to free the storage associated with the local symbols
2465        that we threw away.  Whether or not this has any effect on the
2466        memory allocation depends on the malloc implementation and how
2467        many symbols we managed to discard.  */
2468     new_syms = g_try_renew(struct elf_sym, syms, nsyms);
2469     if (new_syms == NULL) {
2470         goto give_up;
2471     }
2472     syms = new_syms;
2473 
2474     qsort(syms, nsyms, sizeof(*syms), symcmp);
2475 
2476     s->disas_num_syms = nsyms;
2477 #if ELF_CLASS == ELFCLASS32
2478     s->disas_symtab.elf32 = syms;
2479 #else
2480     s->disas_symtab.elf64 = syms;
2481 #endif
2482     s->lookup_symbol = lookup_symbolxx;
2483     s->next = syminfos;
2484     syminfos = s;
2485 
2486     return;
2487 
2488 give_up:
2489     g_free(s);
2490     g_free(strings);
2491     g_free(syms);
2492 }
2493 
2494 uint32_t get_elf_eflags(int fd)
2495 {
2496     struct elfhdr ehdr;
2497     off_t offset;
2498     int ret;
2499 
2500     /* Read ELF header */
2501     offset = lseek(fd, 0, SEEK_SET);
2502     if (offset == (off_t) -1) {
2503         return 0;
2504     }
2505     ret = read(fd, &ehdr, sizeof(ehdr));
2506     if (ret < sizeof(ehdr)) {
2507         return 0;
2508     }
2509     offset = lseek(fd, offset, SEEK_SET);
2510     if (offset == (off_t) -1) {
2511         return 0;
2512     }
2513 
2514     /* Check ELF signature */
2515     if (!elf_check_ident(&ehdr)) {
2516         return 0;
2517     }
2518 
2519     /* check header */
2520     bswap_ehdr(&ehdr);
2521     if (!elf_check_ehdr(&ehdr)) {
2522         return 0;
2523     }
2524 
2525     /* return architecture id */
2526     return ehdr.e_flags;
2527 }
2528 
2529 int load_elf_binary(struct linux_binprm *bprm, struct image_info *info)
2530 {
2531     struct image_info interp_info;
2532     struct elfhdr elf_ex;
2533     char *elf_interpreter = NULL;
2534     char *scratch;
2535 
2536     info->start_mmap = (abi_ulong)ELF_START_MMAP;
2537 
2538     load_elf_image(bprm->filename, bprm->fd, info,
2539                    &elf_interpreter, bprm->buf);
2540 
2541     /* ??? We need a copy of the elf header for passing to create_elf_tables.
2542        If we do nothing, we'll have overwritten this when we re-use bprm->buf
2543        when we load the interpreter.  */
2544     elf_ex = *(struct elfhdr *)bprm->buf;
2545 
2546     /* Do this so that we can load the interpreter, if need be.  We will
2547        change some of these later */
2548     bprm->p = setup_arg_pages(bprm, info);
2549 
2550     scratch = g_new0(char, TARGET_PAGE_SIZE);
2551     if (STACK_GROWS_DOWN) {
2552         bprm->p = copy_elf_strings(1, &bprm->filename, scratch,
2553                                    bprm->p, info->stack_limit);
2554         info->file_string = bprm->p;
2555         bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch,
2556                                    bprm->p, info->stack_limit);
2557         info->env_strings = bprm->p;
2558         bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch,
2559                                    bprm->p, info->stack_limit);
2560         info->arg_strings = bprm->p;
2561     } else {
2562         info->arg_strings = bprm->p;
2563         bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch,
2564                                    bprm->p, info->stack_limit);
2565         info->env_strings = bprm->p;
2566         bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch,
2567                                    bprm->p, info->stack_limit);
2568         info->file_string = bprm->p;
2569         bprm->p = copy_elf_strings(1, &bprm->filename, scratch,
2570                                    bprm->p, info->stack_limit);
2571     }
2572 
2573     g_free(scratch);
2574 
2575     if (!bprm->p) {
2576         fprintf(stderr, "%s: %s\n", bprm->filename, strerror(E2BIG));
2577         exit(-1);
2578     }
2579 
2580     if (elf_interpreter) {
2581         load_elf_interp(elf_interpreter, &interp_info, bprm->buf);
2582 
2583         /* If the program interpreter is one of these two, then assume
2584            an iBCS2 image.  Otherwise assume a native linux image.  */
2585 
2586         if (strcmp(elf_interpreter, "/usr/lib/libc.so.1") == 0
2587             || strcmp(elf_interpreter, "/usr/lib/ld.so.1") == 0) {
2588             info->personality = PER_SVR4;
2589 
2590             /* Why this, you ask???  Well SVr4 maps page 0 as read-only,
2591                and some applications "depend" upon this behavior.  Since
2592                we do not have the power to recompile these, we emulate
2593                the SVr4 behavior.  Sigh.  */
2594             target_mmap(0, qemu_host_page_size, PROT_READ | PROT_EXEC,
2595                         MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
2596         }
2597     }
2598 
2599     bprm->p = create_elf_tables(bprm->p, bprm->argc, bprm->envc, &elf_ex,
2600                                 info, (elf_interpreter ? &interp_info : NULL));
2601     info->start_stack = bprm->p;
2602 
2603     /* If we have an interpreter, set that as the program's entry point.
2604        Copy the load_bias as well, to help PPC64 interpret the entry
2605        point as a function descriptor.  Do this after creating elf tables
2606        so that we copy the original program entry point into the AUXV.  */
2607     if (elf_interpreter) {
2608         info->load_bias = interp_info.load_bias;
2609         info->entry = interp_info.entry;
2610         free(elf_interpreter);
2611     }
2612 
2613 #ifdef USE_ELF_CORE_DUMP
2614     bprm->core_dump = &elf_core_dump;
2615 #endif
2616 
2617     return 0;
2618 }
2619 
2620 #ifdef USE_ELF_CORE_DUMP
2621 /*
2622  * Definitions to generate Intel SVR4-like core files.
2623  * These mostly have the same names as the SVR4 types with "target_elf_"
2624  * tacked on the front to prevent clashes with linux definitions,
2625  * and the typedef forms have been avoided.  This is mostly like
2626  * the SVR4 structure, but more Linuxy, with things that Linux does
2627  * not support and which gdb doesn't really use excluded.
2628  *
2629  * Fields we don't dump (their contents is zero) in linux-user qemu
2630  * are marked with XXX.
2631  *
2632  * Core dump code is copied from linux kernel (fs/binfmt_elf.c).
2633  *
2634  * Porting ELF coredump for target is (quite) simple process.  First you
2635  * define USE_ELF_CORE_DUMP in target ELF code (where init_thread() for
2636  * the target resides):
2637  *
2638  * #define USE_ELF_CORE_DUMP
2639  *
2640  * Next you define type of register set used for dumping.  ELF specification
2641  * says that it needs to be array of elf_greg_t that has size of ELF_NREG.
2642  *
2643  * typedef <target_regtype> target_elf_greg_t;
2644  * #define ELF_NREG <number of registers>
2645  * typedef taret_elf_greg_t target_elf_gregset_t[ELF_NREG];
2646  *
2647  * Last step is to implement target specific function that copies registers
2648  * from given cpu into just specified register set.  Prototype is:
2649  *
2650  * static void elf_core_copy_regs(taret_elf_gregset_t *regs,
2651  *                                const CPUArchState *env);
2652  *
2653  * Parameters:
2654  *     regs - copy register values into here (allocated and zeroed by caller)
2655  *     env - copy registers from here
2656  *
2657  * Example for ARM target is provided in this file.
2658  */
2659 
2660 /* An ELF note in memory */
2661 struct memelfnote {
2662     const char *name;
2663     size_t     namesz;
2664     size_t     namesz_rounded;
2665     int        type;
2666     size_t     datasz;
2667     size_t     datasz_rounded;
2668     void       *data;
2669     size_t     notesz;
2670 };
2671 
2672 struct target_elf_siginfo {
2673     abi_int    si_signo; /* signal number */
2674     abi_int    si_code;  /* extra code */
2675     abi_int    si_errno; /* errno */
2676 };
2677 
2678 struct target_elf_prstatus {
2679     struct target_elf_siginfo pr_info;      /* Info associated with signal */
2680     abi_short          pr_cursig;    /* Current signal */
2681     abi_ulong          pr_sigpend;   /* XXX */
2682     abi_ulong          pr_sighold;   /* XXX */
2683     target_pid_t       pr_pid;
2684     target_pid_t       pr_ppid;
2685     target_pid_t       pr_pgrp;
2686     target_pid_t       pr_sid;
2687     struct target_timeval pr_utime;  /* XXX User time */
2688     struct target_timeval pr_stime;  /* XXX System time */
2689     struct target_timeval pr_cutime; /* XXX Cumulative user time */
2690     struct target_timeval pr_cstime; /* XXX Cumulative system time */
2691     target_elf_gregset_t      pr_reg;       /* GP registers */
2692     abi_int            pr_fpvalid;   /* XXX */
2693 };
2694 
2695 #define ELF_PRARGSZ     (80) /* Number of chars for args */
2696 
2697 struct target_elf_prpsinfo {
2698     char         pr_state;       /* numeric process state */
2699     char         pr_sname;       /* char for pr_state */
2700     char         pr_zomb;        /* zombie */
2701     char         pr_nice;        /* nice val */
2702     abi_ulong    pr_flag;        /* flags */
2703     target_uid_t pr_uid;
2704     target_gid_t pr_gid;
2705     target_pid_t pr_pid, pr_ppid, pr_pgrp, pr_sid;
2706     /* Lots missing */
2707     char    pr_fname[16];           /* filename of executable */
2708     char    pr_psargs[ELF_PRARGSZ]; /* initial part of arg list */
2709 };
2710 
2711 /* Here is the structure in which status of each thread is captured. */
2712 struct elf_thread_status {
2713     QTAILQ_ENTRY(elf_thread_status)  ets_link;
2714     struct target_elf_prstatus prstatus;   /* NT_PRSTATUS */
2715 #if 0
2716     elf_fpregset_t fpu;             /* NT_PRFPREG */
2717     struct task_struct *thread;
2718     elf_fpxregset_t xfpu;           /* ELF_CORE_XFPREG_TYPE */
2719 #endif
2720     struct memelfnote notes[1];
2721     int num_notes;
2722 };
2723 
2724 struct elf_note_info {
2725     struct memelfnote   *notes;
2726     struct target_elf_prstatus *prstatus;  /* NT_PRSTATUS */
2727     struct target_elf_prpsinfo *psinfo;    /* NT_PRPSINFO */
2728 
2729     QTAILQ_HEAD(thread_list_head, elf_thread_status) thread_list;
2730 #if 0
2731     /*
2732      * Current version of ELF coredump doesn't support
2733      * dumping fp regs etc.
2734      */
2735     elf_fpregset_t *fpu;
2736     elf_fpxregset_t *xfpu;
2737     int thread_status_size;
2738 #endif
2739     int notes_size;
2740     int numnote;
2741 };
2742 
2743 struct vm_area_struct {
2744     target_ulong   vma_start;  /* start vaddr of memory region */
2745     target_ulong   vma_end;    /* end vaddr of memory region */
2746     abi_ulong      vma_flags;  /* protection etc. flags for the region */
2747     QTAILQ_ENTRY(vm_area_struct) vma_link;
2748 };
2749 
2750 struct mm_struct {
2751     QTAILQ_HEAD(, vm_area_struct) mm_mmap;
2752     int mm_count;           /* number of mappings */
2753 };
2754 
2755 static struct mm_struct *vma_init(void);
2756 static void vma_delete(struct mm_struct *);
2757 static int vma_add_mapping(struct mm_struct *, target_ulong,
2758                            target_ulong, abi_ulong);
2759 static int vma_get_mapping_count(const struct mm_struct *);
2760 static struct vm_area_struct *vma_first(const struct mm_struct *);
2761 static struct vm_area_struct *vma_next(struct vm_area_struct *);
2762 static abi_ulong vma_dump_size(const struct vm_area_struct *);
2763 static int vma_walker(void *priv, target_ulong start, target_ulong end,
2764                       unsigned long flags);
2765 
2766 static void fill_elf_header(struct elfhdr *, int, uint16_t, uint32_t);
2767 static void fill_note(struct memelfnote *, const char *, int,
2768                       unsigned int, void *);
2769 static void fill_prstatus(struct target_elf_prstatus *, const TaskState *, int);
2770 static int fill_psinfo(struct target_elf_prpsinfo *, const TaskState *);
2771 static void fill_auxv_note(struct memelfnote *, const TaskState *);
2772 static void fill_elf_note_phdr(struct elf_phdr *, int, off_t);
2773 static size_t note_size(const struct memelfnote *);
2774 static void free_note_info(struct elf_note_info *);
2775 static int fill_note_info(struct elf_note_info *, long, const CPUArchState *);
2776 static void fill_thread_info(struct elf_note_info *, const CPUArchState *);
2777 static int core_dump_filename(const TaskState *, char *, size_t);
2778 
2779 static int dump_write(int, const void *, size_t);
2780 static int write_note(struct memelfnote *, int);
2781 static int write_note_info(struct elf_note_info *, int);
2782 
2783 #ifdef BSWAP_NEEDED
2784 static void bswap_prstatus(struct target_elf_prstatus *prstatus)
2785 {
2786     prstatus->pr_info.si_signo = tswap32(prstatus->pr_info.si_signo);
2787     prstatus->pr_info.si_code = tswap32(prstatus->pr_info.si_code);
2788     prstatus->pr_info.si_errno = tswap32(prstatus->pr_info.si_errno);
2789     prstatus->pr_cursig = tswap16(prstatus->pr_cursig);
2790     prstatus->pr_sigpend = tswapal(prstatus->pr_sigpend);
2791     prstatus->pr_sighold = tswapal(prstatus->pr_sighold);
2792     prstatus->pr_pid = tswap32(prstatus->pr_pid);
2793     prstatus->pr_ppid = tswap32(prstatus->pr_ppid);
2794     prstatus->pr_pgrp = tswap32(prstatus->pr_pgrp);
2795     prstatus->pr_sid = tswap32(prstatus->pr_sid);
2796     /* cpu times are not filled, so we skip them */
2797     /* regs should be in correct format already */
2798     prstatus->pr_fpvalid = tswap32(prstatus->pr_fpvalid);
2799 }
2800 
2801 static void bswap_psinfo(struct target_elf_prpsinfo *psinfo)
2802 {
2803     psinfo->pr_flag = tswapal(psinfo->pr_flag);
2804     psinfo->pr_uid = tswap16(psinfo->pr_uid);
2805     psinfo->pr_gid = tswap16(psinfo->pr_gid);
2806     psinfo->pr_pid = tswap32(psinfo->pr_pid);
2807     psinfo->pr_ppid = tswap32(psinfo->pr_ppid);
2808     psinfo->pr_pgrp = tswap32(psinfo->pr_pgrp);
2809     psinfo->pr_sid = tswap32(psinfo->pr_sid);
2810 }
2811 
2812 static void bswap_note(struct elf_note *en)
2813 {
2814     bswap32s(&en->n_namesz);
2815     bswap32s(&en->n_descsz);
2816     bswap32s(&en->n_type);
2817 }
2818 #else
2819 static inline void bswap_prstatus(struct target_elf_prstatus *p) { }
2820 static inline void bswap_psinfo(struct target_elf_prpsinfo *p) {}
2821 static inline void bswap_note(struct elf_note *en) { }
2822 #endif /* BSWAP_NEEDED */
2823 
2824 /*
2825  * Minimal support for linux memory regions.  These are needed
2826  * when we are finding out what memory exactly belongs to
2827  * emulated process.  No locks needed here, as long as
2828  * thread that received the signal is stopped.
2829  */
2830 
2831 static struct mm_struct *vma_init(void)
2832 {
2833     struct mm_struct *mm;
2834 
2835     if ((mm = g_malloc(sizeof (*mm))) == NULL)
2836         return (NULL);
2837 
2838     mm->mm_count = 0;
2839     QTAILQ_INIT(&mm->mm_mmap);
2840 
2841     return (mm);
2842 }
2843 
2844 static void vma_delete(struct mm_struct *mm)
2845 {
2846     struct vm_area_struct *vma;
2847 
2848     while ((vma = vma_first(mm)) != NULL) {
2849         QTAILQ_REMOVE(&mm->mm_mmap, vma, vma_link);
2850         g_free(vma);
2851     }
2852     g_free(mm);
2853 }
2854 
2855 static int vma_add_mapping(struct mm_struct *mm, target_ulong start,
2856                            target_ulong end, abi_ulong flags)
2857 {
2858     struct vm_area_struct *vma;
2859 
2860     if ((vma = g_malloc0(sizeof (*vma))) == NULL)
2861         return (-1);
2862 
2863     vma->vma_start = start;
2864     vma->vma_end = end;
2865     vma->vma_flags = flags;
2866 
2867     QTAILQ_INSERT_TAIL(&mm->mm_mmap, vma, vma_link);
2868     mm->mm_count++;
2869 
2870     return (0);
2871 }
2872 
2873 static struct vm_area_struct *vma_first(const struct mm_struct *mm)
2874 {
2875     return (QTAILQ_FIRST(&mm->mm_mmap));
2876 }
2877 
2878 static struct vm_area_struct *vma_next(struct vm_area_struct *vma)
2879 {
2880     return (QTAILQ_NEXT(vma, vma_link));
2881 }
2882 
2883 static int vma_get_mapping_count(const struct mm_struct *mm)
2884 {
2885     return (mm->mm_count);
2886 }
2887 
2888 /*
2889  * Calculate file (dump) size of given memory region.
2890  */
2891 static abi_ulong vma_dump_size(const struct vm_area_struct *vma)
2892 {
2893     /* if we cannot even read the first page, skip it */
2894     if (!access_ok(VERIFY_READ, vma->vma_start, TARGET_PAGE_SIZE))
2895         return (0);
2896 
2897     /*
2898      * Usually we don't dump executable pages as they contain
2899      * non-writable code that debugger can read directly from
2900      * target library etc.  However, thread stacks are marked
2901      * also executable so we read in first page of given region
2902      * and check whether it contains elf header.  If there is
2903      * no elf header, we dump it.
2904      */
2905     if (vma->vma_flags & PROT_EXEC) {
2906         char page[TARGET_PAGE_SIZE];
2907 
2908         copy_from_user(page, vma->vma_start, sizeof (page));
2909         if ((page[EI_MAG0] == ELFMAG0) &&
2910             (page[EI_MAG1] == ELFMAG1) &&
2911             (page[EI_MAG2] == ELFMAG2) &&
2912             (page[EI_MAG3] == ELFMAG3)) {
2913             /*
2914              * Mappings are possibly from ELF binary.  Don't dump
2915              * them.
2916              */
2917             return (0);
2918         }
2919     }
2920 
2921     return (vma->vma_end - vma->vma_start);
2922 }
2923 
2924 static int vma_walker(void *priv, target_ulong start, target_ulong end,
2925                       unsigned long flags)
2926 {
2927     struct mm_struct *mm = (struct mm_struct *)priv;
2928 
2929     vma_add_mapping(mm, start, end, flags);
2930     return (0);
2931 }
2932 
2933 static void fill_note(struct memelfnote *note, const char *name, int type,
2934                       unsigned int sz, void *data)
2935 {
2936     unsigned int namesz;
2937 
2938     namesz = strlen(name) + 1;
2939     note->name = name;
2940     note->namesz = namesz;
2941     note->namesz_rounded = roundup(namesz, sizeof (int32_t));
2942     note->type = type;
2943     note->datasz = sz;
2944     note->datasz_rounded = roundup(sz, sizeof (int32_t));
2945 
2946     note->data = data;
2947 
2948     /*
2949      * We calculate rounded up note size here as specified by
2950      * ELF document.
2951      */
2952     note->notesz = sizeof (struct elf_note) +
2953         note->namesz_rounded + note->datasz_rounded;
2954 }
2955 
2956 static void fill_elf_header(struct elfhdr *elf, int segs, uint16_t machine,
2957                             uint32_t flags)
2958 {
2959     (void) memset(elf, 0, sizeof(*elf));
2960 
2961     (void) memcpy(elf->e_ident, ELFMAG, SELFMAG);
2962     elf->e_ident[EI_CLASS] = ELF_CLASS;
2963     elf->e_ident[EI_DATA] = ELF_DATA;
2964     elf->e_ident[EI_VERSION] = EV_CURRENT;
2965     elf->e_ident[EI_OSABI] = ELF_OSABI;
2966 
2967     elf->e_type = ET_CORE;
2968     elf->e_machine = machine;
2969     elf->e_version = EV_CURRENT;
2970     elf->e_phoff = sizeof(struct elfhdr);
2971     elf->e_flags = flags;
2972     elf->e_ehsize = sizeof(struct elfhdr);
2973     elf->e_phentsize = sizeof(struct elf_phdr);
2974     elf->e_phnum = segs;
2975 
2976     bswap_ehdr(elf);
2977 }
2978 
2979 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, off_t offset)
2980 {
2981     phdr->p_type = PT_NOTE;
2982     phdr->p_offset = offset;
2983     phdr->p_vaddr = 0;
2984     phdr->p_paddr = 0;
2985     phdr->p_filesz = sz;
2986     phdr->p_memsz = 0;
2987     phdr->p_flags = 0;
2988     phdr->p_align = 0;
2989 
2990     bswap_phdr(phdr, 1);
2991 }
2992 
2993 static size_t note_size(const struct memelfnote *note)
2994 {
2995     return (note->notesz);
2996 }
2997 
2998 static void fill_prstatus(struct target_elf_prstatus *prstatus,
2999                           const TaskState *ts, int signr)
3000 {
3001     (void) memset(prstatus, 0, sizeof (*prstatus));
3002     prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
3003     prstatus->pr_pid = ts->ts_tid;
3004     prstatus->pr_ppid = getppid();
3005     prstatus->pr_pgrp = getpgrp();
3006     prstatus->pr_sid = getsid(0);
3007 
3008     bswap_prstatus(prstatus);
3009 }
3010 
3011 static int fill_psinfo(struct target_elf_prpsinfo *psinfo, const TaskState *ts)
3012 {
3013     char *base_filename;
3014     unsigned int i, len;
3015 
3016     (void) memset(psinfo, 0, sizeof (*psinfo));
3017 
3018     len = ts->info->arg_end - ts->info->arg_start;
3019     if (len >= ELF_PRARGSZ)
3020         len = ELF_PRARGSZ - 1;
3021     if (copy_from_user(&psinfo->pr_psargs, ts->info->arg_start, len))
3022         return -EFAULT;
3023     for (i = 0; i < len; i++)
3024         if (psinfo->pr_psargs[i] == 0)
3025             psinfo->pr_psargs[i] = ' ';
3026     psinfo->pr_psargs[len] = 0;
3027 
3028     psinfo->pr_pid = getpid();
3029     psinfo->pr_ppid = getppid();
3030     psinfo->pr_pgrp = getpgrp();
3031     psinfo->pr_sid = getsid(0);
3032     psinfo->pr_uid = getuid();
3033     psinfo->pr_gid = getgid();
3034 
3035     base_filename = g_path_get_basename(ts->bprm->filename);
3036     /*
3037      * Using strncpy here is fine: at max-length,
3038      * this field is not NUL-terminated.
3039      */
3040     (void) strncpy(psinfo->pr_fname, base_filename,
3041                    sizeof(psinfo->pr_fname));
3042 
3043     g_free(base_filename);
3044     bswap_psinfo(psinfo);
3045     return (0);
3046 }
3047 
3048 static void fill_auxv_note(struct memelfnote *note, const TaskState *ts)
3049 {
3050     elf_addr_t auxv = (elf_addr_t)ts->info->saved_auxv;
3051     elf_addr_t orig_auxv = auxv;
3052     void *ptr;
3053     int len = ts->info->auxv_len;
3054 
3055     /*
3056      * Auxiliary vector is stored in target process stack.  It contains
3057      * {type, value} pairs that we need to dump into note.  This is not
3058      * strictly necessary but we do it here for sake of completeness.
3059      */
3060 
3061     /* read in whole auxv vector and copy it to memelfnote */
3062     ptr = lock_user(VERIFY_READ, orig_auxv, len, 0);
3063     if (ptr != NULL) {
3064         fill_note(note, "CORE", NT_AUXV, len, ptr);
3065         unlock_user(ptr, auxv, len);
3066     }
3067 }
3068 
3069 /*
3070  * Constructs name of coredump file.  We have following convention
3071  * for the name:
3072  *     qemu_<basename-of-target-binary>_<date>-<time>_<pid>.core
3073  *
3074  * Returns 0 in case of success, -1 otherwise (errno is set).
3075  */
3076 static int core_dump_filename(const TaskState *ts, char *buf,
3077                               size_t bufsize)
3078 {
3079     char timestamp[64];
3080     char *base_filename = NULL;
3081     struct timeval tv;
3082     struct tm tm;
3083 
3084     assert(bufsize >= PATH_MAX);
3085 
3086     if (gettimeofday(&tv, NULL) < 0) {
3087         (void) fprintf(stderr, "unable to get current timestamp: %s",
3088                        strerror(errno));
3089         return (-1);
3090     }
3091 
3092     base_filename = g_path_get_basename(ts->bprm->filename);
3093     (void) strftime(timestamp, sizeof (timestamp), "%Y%m%d-%H%M%S",
3094                     localtime_r(&tv.tv_sec, &tm));
3095     (void) snprintf(buf, bufsize, "qemu_%s_%s_%d.core",
3096                     base_filename, timestamp, (int)getpid());
3097     g_free(base_filename);
3098 
3099     return (0);
3100 }
3101 
3102 static int dump_write(int fd, const void *ptr, size_t size)
3103 {
3104     const char *bufp = (const char *)ptr;
3105     ssize_t bytes_written, bytes_left;
3106     struct rlimit dumpsize;
3107     off_t pos;
3108 
3109     bytes_written = 0;
3110     getrlimit(RLIMIT_CORE, &dumpsize);
3111     if ((pos = lseek(fd, 0, SEEK_CUR))==-1) {
3112         if (errno == ESPIPE) { /* not a seekable stream */
3113             bytes_left = size;
3114         } else {
3115             return pos;
3116         }
3117     } else {
3118         if (dumpsize.rlim_cur <= pos) {
3119             return -1;
3120         } else if (dumpsize.rlim_cur == RLIM_INFINITY) {
3121             bytes_left = size;
3122         } else {
3123             size_t limit_left=dumpsize.rlim_cur - pos;
3124             bytes_left = limit_left >= size ? size : limit_left ;
3125         }
3126     }
3127 
3128     /*
3129      * In normal conditions, single write(2) should do but
3130      * in case of socket etc. this mechanism is more portable.
3131      */
3132     do {
3133         bytes_written = write(fd, bufp, bytes_left);
3134         if (bytes_written < 0) {
3135             if (errno == EINTR)
3136                 continue;
3137             return (-1);
3138         } else if (bytes_written == 0) { /* eof */
3139             return (-1);
3140         }
3141         bufp += bytes_written;
3142         bytes_left -= bytes_written;
3143     } while (bytes_left > 0);
3144 
3145     return (0);
3146 }
3147 
3148 static int write_note(struct memelfnote *men, int fd)
3149 {
3150     struct elf_note en;
3151 
3152     en.n_namesz = men->namesz;
3153     en.n_type = men->type;
3154     en.n_descsz = men->datasz;
3155 
3156     bswap_note(&en);
3157 
3158     if (dump_write(fd, &en, sizeof(en)) != 0)
3159         return (-1);
3160     if (dump_write(fd, men->name, men->namesz_rounded) != 0)
3161         return (-1);
3162     if (dump_write(fd, men->data, men->datasz_rounded) != 0)
3163         return (-1);
3164 
3165     return (0);
3166 }
3167 
3168 static void fill_thread_info(struct elf_note_info *info, const CPUArchState *env)
3169 {
3170     CPUState *cpu = ENV_GET_CPU((CPUArchState *)env);
3171     TaskState *ts = (TaskState *)cpu->opaque;
3172     struct elf_thread_status *ets;
3173 
3174     ets = g_malloc0(sizeof (*ets));
3175     ets->num_notes = 1; /* only prstatus is dumped */
3176     fill_prstatus(&ets->prstatus, ts, 0);
3177     elf_core_copy_regs(&ets->prstatus.pr_reg, env);
3178     fill_note(&ets->notes[0], "CORE", NT_PRSTATUS, sizeof (ets->prstatus),
3179               &ets->prstatus);
3180 
3181     QTAILQ_INSERT_TAIL(&info->thread_list, ets, ets_link);
3182 
3183     info->notes_size += note_size(&ets->notes[0]);
3184 }
3185 
3186 static void init_note_info(struct elf_note_info *info)
3187 {
3188     /* Initialize the elf_note_info structure so that it is at
3189      * least safe to call free_note_info() on it. Must be
3190      * called before calling fill_note_info().
3191      */
3192     memset(info, 0, sizeof (*info));
3193     QTAILQ_INIT(&info->thread_list);
3194 }
3195 
3196 static int fill_note_info(struct elf_note_info *info,
3197                           long signr, const CPUArchState *env)
3198 {
3199 #define NUMNOTES 3
3200     CPUState *cpu = ENV_GET_CPU((CPUArchState *)env);
3201     TaskState *ts = (TaskState *)cpu->opaque;
3202     int i;
3203 
3204     info->notes = g_new0(struct memelfnote, NUMNOTES);
3205     if (info->notes == NULL)
3206         return (-ENOMEM);
3207     info->prstatus = g_malloc0(sizeof (*info->prstatus));
3208     if (info->prstatus == NULL)
3209         return (-ENOMEM);
3210     info->psinfo = g_malloc0(sizeof (*info->psinfo));
3211     if (info->prstatus == NULL)
3212         return (-ENOMEM);
3213 
3214     /*
3215      * First fill in status (and registers) of current thread
3216      * including process info & aux vector.
3217      */
3218     fill_prstatus(info->prstatus, ts, signr);
3219     elf_core_copy_regs(&info->prstatus->pr_reg, env);
3220     fill_note(&info->notes[0], "CORE", NT_PRSTATUS,
3221               sizeof (*info->prstatus), info->prstatus);
3222     fill_psinfo(info->psinfo, ts);
3223     fill_note(&info->notes[1], "CORE", NT_PRPSINFO,
3224               sizeof (*info->psinfo), info->psinfo);
3225     fill_auxv_note(&info->notes[2], ts);
3226     info->numnote = 3;
3227 
3228     info->notes_size = 0;
3229     for (i = 0; i < info->numnote; i++)
3230         info->notes_size += note_size(&info->notes[i]);
3231 
3232     /* read and fill status of all threads */
3233     cpu_list_lock();
3234     CPU_FOREACH(cpu) {
3235         if (cpu == thread_cpu) {
3236             continue;
3237         }
3238         fill_thread_info(info, (CPUArchState *)cpu->env_ptr);
3239     }
3240     cpu_list_unlock();
3241 
3242     return (0);
3243 }
3244 
3245 static void free_note_info(struct elf_note_info *info)
3246 {
3247     struct elf_thread_status *ets;
3248 
3249     while (!QTAILQ_EMPTY(&info->thread_list)) {
3250         ets = QTAILQ_FIRST(&info->thread_list);
3251         QTAILQ_REMOVE(&info->thread_list, ets, ets_link);
3252         g_free(ets);
3253     }
3254 
3255     g_free(info->prstatus);
3256     g_free(info->psinfo);
3257     g_free(info->notes);
3258 }
3259 
3260 static int write_note_info(struct elf_note_info *info, int fd)
3261 {
3262     struct elf_thread_status *ets;
3263     int i, error = 0;
3264 
3265     /* write prstatus, psinfo and auxv for current thread */
3266     for (i = 0; i < info->numnote; i++)
3267         if ((error = write_note(&info->notes[i], fd)) != 0)
3268             return (error);
3269 
3270     /* write prstatus for each thread */
3271     QTAILQ_FOREACH(ets, &info->thread_list, ets_link) {
3272         if ((error = write_note(&ets->notes[0], fd)) != 0)
3273             return (error);
3274     }
3275 
3276     return (0);
3277 }
3278 
3279 /*
3280  * Write out ELF coredump.
3281  *
3282  * See documentation of ELF object file format in:
3283  * http://www.caldera.com/developers/devspecs/gabi41.pdf
3284  *
3285  * Coredump format in linux is following:
3286  *
3287  * 0   +----------------------+         \
3288  *     | ELF header           | ET_CORE  |
3289  *     +----------------------+          |
3290  *     | ELF program headers  |          |--- headers
3291  *     | - NOTE section       |          |
3292  *     | - PT_LOAD sections   |          |
3293  *     +----------------------+         /
3294  *     | NOTEs:               |
3295  *     | - NT_PRSTATUS        |
3296  *     | - NT_PRSINFO         |
3297  *     | - NT_AUXV            |
3298  *     +----------------------+ <-- aligned to target page
3299  *     | Process memory dump  |
3300  *     :                      :
3301  *     .                      .
3302  *     :                      :
3303  *     |                      |
3304  *     +----------------------+
3305  *
3306  * NT_PRSTATUS -> struct elf_prstatus (per thread)
3307  * NT_PRSINFO  -> struct elf_prpsinfo
3308  * NT_AUXV is array of { type, value } pairs (see fill_auxv_note()).
3309  *
3310  * Format follows System V format as close as possible.  Current
3311  * version limitations are as follows:
3312  *     - no floating point registers are dumped
3313  *
3314  * Function returns 0 in case of success, negative errno otherwise.
3315  *
3316  * TODO: make this work also during runtime: it should be
3317  * possible to force coredump from running process and then
3318  * continue processing.  For example qemu could set up SIGUSR2
3319  * handler (provided that target process haven't registered
3320  * handler for that) that does the dump when signal is received.
3321  */
3322 static int elf_core_dump(int signr, const CPUArchState *env)
3323 {
3324     const CPUState *cpu = ENV_GET_CPU((CPUArchState *)env);
3325     const TaskState *ts = (const TaskState *)cpu->opaque;
3326     struct vm_area_struct *vma = NULL;
3327     char corefile[PATH_MAX];
3328     struct elf_note_info info;
3329     struct elfhdr elf;
3330     struct elf_phdr phdr;
3331     struct rlimit dumpsize;
3332     struct mm_struct *mm = NULL;
3333     off_t offset = 0, data_offset = 0;
3334     int segs = 0;
3335     int fd = -1;
3336 
3337     init_note_info(&info);
3338 
3339     errno = 0;
3340     getrlimit(RLIMIT_CORE, &dumpsize);
3341     if (dumpsize.rlim_cur == 0)
3342         return 0;
3343 
3344     if (core_dump_filename(ts, corefile, sizeof (corefile)) < 0)
3345         return (-errno);
3346 
3347     if ((fd = open(corefile, O_WRONLY | O_CREAT,
3348                    S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH)) < 0)
3349         return (-errno);
3350 
3351     /*
3352      * Walk through target process memory mappings and
3353      * set up structure containing this information.  After
3354      * this point vma_xxx functions can be used.
3355      */
3356     if ((mm = vma_init()) == NULL)
3357         goto out;
3358 
3359     walk_memory_regions(mm, vma_walker);
3360     segs = vma_get_mapping_count(mm);
3361 
3362     /*
3363      * Construct valid coredump ELF header.  We also
3364      * add one more segment for notes.
3365      */
3366     fill_elf_header(&elf, segs + 1, ELF_MACHINE, 0);
3367     if (dump_write(fd, &elf, sizeof (elf)) != 0)
3368         goto out;
3369 
3370     /* fill in the in-memory version of notes */
3371     if (fill_note_info(&info, signr, env) < 0)
3372         goto out;
3373 
3374     offset += sizeof (elf);                             /* elf header */
3375     offset += (segs + 1) * sizeof (struct elf_phdr);    /* program headers */
3376 
3377     /* write out notes program header */
3378     fill_elf_note_phdr(&phdr, info.notes_size, offset);
3379 
3380     offset += info.notes_size;
3381     if (dump_write(fd, &phdr, sizeof (phdr)) != 0)
3382         goto out;
3383 
3384     /*
3385      * ELF specification wants data to start at page boundary so
3386      * we align it here.
3387      */
3388     data_offset = offset = roundup(offset, ELF_EXEC_PAGESIZE);
3389 
3390     /*
3391      * Write program headers for memory regions mapped in
3392      * the target process.
3393      */
3394     for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) {
3395         (void) memset(&phdr, 0, sizeof (phdr));
3396 
3397         phdr.p_type = PT_LOAD;
3398         phdr.p_offset = offset;
3399         phdr.p_vaddr = vma->vma_start;
3400         phdr.p_paddr = 0;
3401         phdr.p_filesz = vma_dump_size(vma);
3402         offset += phdr.p_filesz;
3403         phdr.p_memsz = vma->vma_end - vma->vma_start;
3404         phdr.p_flags = vma->vma_flags & PROT_READ ? PF_R : 0;
3405         if (vma->vma_flags & PROT_WRITE)
3406             phdr.p_flags |= PF_W;
3407         if (vma->vma_flags & PROT_EXEC)
3408             phdr.p_flags |= PF_X;
3409         phdr.p_align = ELF_EXEC_PAGESIZE;
3410 
3411         bswap_phdr(&phdr, 1);
3412         if (dump_write(fd, &phdr, sizeof(phdr)) != 0) {
3413             goto out;
3414         }
3415     }
3416 
3417     /*
3418      * Next we write notes just after program headers.  No
3419      * alignment needed here.
3420      */
3421     if (write_note_info(&info, fd) < 0)
3422         goto out;
3423 
3424     /* align data to page boundary */
3425     if (lseek(fd, data_offset, SEEK_SET) != data_offset)
3426         goto out;
3427 
3428     /*
3429      * Finally we can dump process memory into corefile as well.
3430      */
3431     for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) {
3432         abi_ulong addr;
3433         abi_ulong end;
3434 
3435         end = vma->vma_start + vma_dump_size(vma);
3436 
3437         for (addr = vma->vma_start; addr < end;
3438              addr += TARGET_PAGE_SIZE) {
3439             char page[TARGET_PAGE_SIZE];
3440             int error;
3441 
3442             /*
3443              *  Read in page from target process memory and
3444              *  write it to coredump file.
3445              */
3446             error = copy_from_user(page, addr, sizeof (page));
3447             if (error != 0) {
3448                 (void) fprintf(stderr, "unable to dump " TARGET_ABI_FMT_lx "\n",
3449                                addr);
3450                 errno = -error;
3451                 goto out;
3452             }
3453             if (dump_write(fd, page, TARGET_PAGE_SIZE) < 0)
3454                 goto out;
3455         }
3456     }
3457 
3458  out:
3459     free_note_info(&info);
3460     if (mm != NULL)
3461         vma_delete(mm);
3462     (void) close(fd);
3463 
3464     if (errno != 0)
3465         return (-errno);
3466     return (0);
3467 }
3468 #endif /* USE_ELF_CORE_DUMP */
3469 
3470 void do_init_thread(struct target_pt_regs *regs, struct image_info *infop)
3471 {
3472     init_thread(regs, infop);
3473 }
3474