xref: /openbmc/qemu/linux-user/elfload.c (revision e7b79428)
1 /* This is the Linux kernel elf-loading code, ported into user space */
2 #include "qemu/osdep.h"
3 #include <sys/param.h>
4 
5 #include <sys/resource.h>
6 #include <sys/shm.h>
7 
8 #include "qemu.h"
9 #include "user-internals.h"
10 #include "signal-common.h"
11 #include "loader.h"
12 #include "user-mmap.h"
13 #include "disas/disas.h"
14 #include "qemu/bitops.h"
15 #include "qemu/path.h"
16 #include "qemu/queue.h"
17 #include "qemu/guest-random.h"
18 #include "qemu/units.h"
19 #include "qemu/selfmap.h"
20 #include "qapi/error.h"
21 #include "target_signal.h"
22 
23 #ifdef _ARCH_PPC64
24 #undef ARCH_DLINFO
25 #undef ELF_PLATFORM
26 #undef ELF_HWCAP
27 #undef ELF_HWCAP2
28 #undef ELF_CLASS
29 #undef ELF_DATA
30 #undef ELF_ARCH
31 #endif
32 
33 #define ELF_OSABI   ELFOSABI_SYSV
34 
35 /* from personality.h */
36 
37 /*
38  * Flags for bug emulation.
39  *
40  * These occupy the top three bytes.
41  */
42 enum {
43     ADDR_NO_RANDOMIZE = 0x0040000,      /* disable randomization of VA space */
44     FDPIC_FUNCPTRS =    0x0080000,      /* userspace function ptrs point to
45                                            descriptors (signal handling) */
46     MMAP_PAGE_ZERO =    0x0100000,
47     ADDR_COMPAT_LAYOUT = 0x0200000,
48     READ_IMPLIES_EXEC = 0x0400000,
49     ADDR_LIMIT_32BIT =  0x0800000,
50     SHORT_INODE =       0x1000000,
51     WHOLE_SECONDS =     0x2000000,
52     STICKY_TIMEOUTS =   0x4000000,
53     ADDR_LIMIT_3GB =    0x8000000,
54 };
55 
56 /*
57  * Personality types.
58  *
59  * These go in the low byte.  Avoid using the top bit, it will
60  * conflict with error returns.
61  */
62 enum {
63     PER_LINUX =         0x0000,
64     PER_LINUX_32BIT =   0x0000 | ADDR_LIMIT_32BIT,
65     PER_LINUX_FDPIC =   0x0000 | FDPIC_FUNCPTRS,
66     PER_SVR4 =          0x0001 | STICKY_TIMEOUTS | MMAP_PAGE_ZERO,
67     PER_SVR3 =          0x0002 | STICKY_TIMEOUTS | SHORT_INODE,
68     PER_SCOSVR3 =       0x0003 | STICKY_TIMEOUTS | WHOLE_SECONDS | SHORT_INODE,
69     PER_OSR5 =          0x0003 | STICKY_TIMEOUTS | WHOLE_SECONDS,
70     PER_WYSEV386 =      0x0004 | STICKY_TIMEOUTS | SHORT_INODE,
71     PER_ISCR4 =         0x0005 | STICKY_TIMEOUTS,
72     PER_BSD =           0x0006,
73     PER_SUNOS =         0x0006 | STICKY_TIMEOUTS,
74     PER_XENIX =         0x0007 | STICKY_TIMEOUTS | SHORT_INODE,
75     PER_LINUX32 =       0x0008,
76     PER_LINUX32_3GB =   0x0008 | ADDR_LIMIT_3GB,
77     PER_IRIX32 =        0x0009 | STICKY_TIMEOUTS,/* IRIX5 32-bit */
78     PER_IRIXN32 =       0x000a | STICKY_TIMEOUTS,/* IRIX6 new 32-bit */
79     PER_IRIX64 =        0x000b | STICKY_TIMEOUTS,/* IRIX6 64-bit */
80     PER_RISCOS =        0x000c,
81     PER_SOLARIS =       0x000d | STICKY_TIMEOUTS,
82     PER_UW7 =           0x000e | STICKY_TIMEOUTS | MMAP_PAGE_ZERO,
83     PER_OSF4 =          0x000f,                  /* OSF/1 v4 */
84     PER_HPUX =          0x0010,
85     PER_MASK =          0x00ff,
86 };
87 
88 /*
89  * Return the base personality without flags.
90  */
91 #define personality(pers)       (pers & PER_MASK)
92 
93 int info_is_fdpic(struct image_info *info)
94 {
95     return info->personality == PER_LINUX_FDPIC;
96 }
97 
98 /* this flag is uneffective under linux too, should be deleted */
99 #ifndef MAP_DENYWRITE
100 #define MAP_DENYWRITE 0
101 #endif
102 
103 /* should probably go in elf.h */
104 #ifndef ELIBBAD
105 #define ELIBBAD 80
106 #endif
107 
108 #ifdef TARGET_WORDS_BIGENDIAN
109 #define ELF_DATA        ELFDATA2MSB
110 #else
111 #define ELF_DATA        ELFDATA2LSB
112 #endif
113 
114 #ifdef TARGET_ABI_MIPSN32
115 typedef abi_ullong      target_elf_greg_t;
116 #define tswapreg(ptr)   tswap64(ptr)
117 #else
118 typedef abi_ulong       target_elf_greg_t;
119 #define tswapreg(ptr)   tswapal(ptr)
120 #endif
121 
122 #ifdef USE_UID16
123 typedef abi_ushort      target_uid_t;
124 typedef abi_ushort      target_gid_t;
125 #else
126 typedef abi_uint        target_uid_t;
127 typedef abi_uint        target_gid_t;
128 #endif
129 typedef abi_int         target_pid_t;
130 
131 #ifdef TARGET_I386
132 
133 #define ELF_PLATFORM get_elf_platform()
134 
135 static const char *get_elf_platform(void)
136 {
137     static char elf_platform[] = "i386";
138     int family = object_property_get_int(OBJECT(thread_cpu), "family", NULL);
139     if (family > 6)
140         family = 6;
141     if (family >= 3)
142         elf_platform[1] = '0' + family;
143     return elf_platform;
144 }
145 
146 #define ELF_HWCAP get_elf_hwcap()
147 
148 static uint32_t get_elf_hwcap(void)
149 {
150     X86CPU *cpu = X86_CPU(thread_cpu);
151 
152     return cpu->env.features[FEAT_1_EDX];
153 }
154 
155 #ifdef TARGET_X86_64
156 #define ELF_START_MMAP 0x2aaaaab000ULL
157 
158 #define ELF_CLASS      ELFCLASS64
159 #define ELF_ARCH       EM_X86_64
160 
161 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop)
162 {
163     regs->rax = 0;
164     regs->rsp = infop->start_stack;
165     regs->rip = infop->entry;
166 }
167 
168 #define ELF_NREG    27
169 typedef target_elf_greg_t  target_elf_gregset_t[ELF_NREG];
170 
171 /*
172  * Note that ELF_NREG should be 29 as there should be place for
173  * TRAPNO and ERR "registers" as well but linux doesn't dump
174  * those.
175  *
176  * See linux kernel: arch/x86/include/asm/elf.h
177  */
178 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUX86State *env)
179 {
180     (*regs)[0] = tswapreg(env->regs[15]);
181     (*regs)[1] = tswapreg(env->regs[14]);
182     (*regs)[2] = tswapreg(env->regs[13]);
183     (*regs)[3] = tswapreg(env->regs[12]);
184     (*regs)[4] = tswapreg(env->regs[R_EBP]);
185     (*regs)[5] = tswapreg(env->regs[R_EBX]);
186     (*regs)[6] = tswapreg(env->regs[11]);
187     (*regs)[7] = tswapreg(env->regs[10]);
188     (*regs)[8] = tswapreg(env->regs[9]);
189     (*regs)[9] = tswapreg(env->regs[8]);
190     (*regs)[10] = tswapreg(env->regs[R_EAX]);
191     (*regs)[11] = tswapreg(env->regs[R_ECX]);
192     (*regs)[12] = tswapreg(env->regs[R_EDX]);
193     (*regs)[13] = tswapreg(env->regs[R_ESI]);
194     (*regs)[14] = tswapreg(env->regs[R_EDI]);
195     (*regs)[15] = tswapreg(env->regs[R_EAX]); /* XXX */
196     (*regs)[16] = tswapreg(env->eip);
197     (*regs)[17] = tswapreg(env->segs[R_CS].selector & 0xffff);
198     (*regs)[18] = tswapreg(env->eflags);
199     (*regs)[19] = tswapreg(env->regs[R_ESP]);
200     (*regs)[20] = tswapreg(env->segs[R_SS].selector & 0xffff);
201     (*regs)[21] = tswapreg(env->segs[R_FS].selector & 0xffff);
202     (*regs)[22] = tswapreg(env->segs[R_GS].selector & 0xffff);
203     (*regs)[23] = tswapreg(env->segs[R_DS].selector & 0xffff);
204     (*regs)[24] = tswapreg(env->segs[R_ES].selector & 0xffff);
205     (*regs)[25] = tswapreg(env->segs[R_FS].selector & 0xffff);
206     (*regs)[26] = tswapreg(env->segs[R_GS].selector & 0xffff);
207 }
208 
209 #else
210 
211 #define ELF_START_MMAP 0x80000000
212 
213 /*
214  * This is used to ensure we don't load something for the wrong architecture.
215  */
216 #define elf_check_arch(x) ( ((x) == EM_386) || ((x) == EM_486) )
217 
218 /*
219  * These are used to set parameters in the core dumps.
220  */
221 #define ELF_CLASS       ELFCLASS32
222 #define ELF_ARCH        EM_386
223 
224 static inline void init_thread(struct target_pt_regs *regs,
225                                struct image_info *infop)
226 {
227     regs->esp = infop->start_stack;
228     regs->eip = infop->entry;
229 
230     /* SVR4/i386 ABI (pages 3-31, 3-32) says that when the program
231        starts %edx contains a pointer to a function which might be
232        registered using `atexit'.  This provides a mean for the
233        dynamic linker to call DT_FINI functions for shared libraries
234        that have been loaded before the code runs.
235 
236        A value of 0 tells we have no such handler.  */
237     regs->edx = 0;
238 }
239 
240 #define ELF_NREG    17
241 typedef target_elf_greg_t  target_elf_gregset_t[ELF_NREG];
242 
243 /*
244  * Note that ELF_NREG should be 19 as there should be place for
245  * TRAPNO and ERR "registers" as well but linux doesn't dump
246  * those.
247  *
248  * See linux kernel: arch/x86/include/asm/elf.h
249  */
250 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUX86State *env)
251 {
252     (*regs)[0] = tswapreg(env->regs[R_EBX]);
253     (*regs)[1] = tswapreg(env->regs[R_ECX]);
254     (*regs)[2] = tswapreg(env->regs[R_EDX]);
255     (*regs)[3] = tswapreg(env->regs[R_ESI]);
256     (*regs)[4] = tswapreg(env->regs[R_EDI]);
257     (*regs)[5] = tswapreg(env->regs[R_EBP]);
258     (*regs)[6] = tswapreg(env->regs[R_EAX]);
259     (*regs)[7] = tswapreg(env->segs[R_DS].selector & 0xffff);
260     (*regs)[8] = tswapreg(env->segs[R_ES].selector & 0xffff);
261     (*regs)[9] = tswapreg(env->segs[R_FS].selector & 0xffff);
262     (*regs)[10] = tswapreg(env->segs[R_GS].selector & 0xffff);
263     (*regs)[11] = tswapreg(env->regs[R_EAX]); /* XXX */
264     (*regs)[12] = tswapreg(env->eip);
265     (*regs)[13] = tswapreg(env->segs[R_CS].selector & 0xffff);
266     (*regs)[14] = tswapreg(env->eflags);
267     (*regs)[15] = tswapreg(env->regs[R_ESP]);
268     (*regs)[16] = tswapreg(env->segs[R_SS].selector & 0xffff);
269 }
270 #endif
271 
272 #define USE_ELF_CORE_DUMP
273 #define ELF_EXEC_PAGESIZE       4096
274 
275 #endif
276 
277 #ifdef TARGET_ARM
278 
279 #ifndef TARGET_AARCH64
280 /* 32 bit ARM definitions */
281 
282 #define ELF_START_MMAP 0x80000000
283 
284 #define ELF_ARCH        EM_ARM
285 #define ELF_CLASS       ELFCLASS32
286 
287 static inline void init_thread(struct target_pt_regs *regs,
288                                struct image_info *infop)
289 {
290     abi_long stack = infop->start_stack;
291     memset(regs, 0, sizeof(*regs));
292 
293     regs->uregs[16] = ARM_CPU_MODE_USR;
294     if (infop->entry & 1) {
295         regs->uregs[16] |= CPSR_T;
296     }
297     regs->uregs[15] = infop->entry & 0xfffffffe;
298     regs->uregs[13] = infop->start_stack;
299     /* FIXME - what to for failure of get_user()? */
300     get_user_ual(regs->uregs[2], stack + 8); /* envp */
301     get_user_ual(regs->uregs[1], stack + 4); /* envp */
302     /* XXX: it seems that r0 is zeroed after ! */
303     regs->uregs[0] = 0;
304     /* For uClinux PIC binaries.  */
305     /* XXX: Linux does this only on ARM with no MMU (do we care ?) */
306     regs->uregs[10] = infop->start_data;
307 
308     /* Support ARM FDPIC.  */
309     if (info_is_fdpic(infop)) {
310         /* As described in the ABI document, r7 points to the loadmap info
311          * prepared by the kernel. If an interpreter is needed, r8 points
312          * to the interpreter loadmap and r9 points to the interpreter
313          * PT_DYNAMIC info. If no interpreter is needed, r8 is zero, and
314          * r9 points to the main program PT_DYNAMIC info.
315          */
316         regs->uregs[7] = infop->loadmap_addr;
317         if (infop->interpreter_loadmap_addr) {
318             /* Executable is dynamically loaded.  */
319             regs->uregs[8] = infop->interpreter_loadmap_addr;
320             regs->uregs[9] = infop->interpreter_pt_dynamic_addr;
321         } else {
322             regs->uregs[8] = 0;
323             regs->uregs[9] = infop->pt_dynamic_addr;
324         }
325     }
326 }
327 
328 #define ELF_NREG    18
329 typedef target_elf_greg_t  target_elf_gregset_t[ELF_NREG];
330 
331 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUARMState *env)
332 {
333     (*regs)[0] = tswapreg(env->regs[0]);
334     (*regs)[1] = tswapreg(env->regs[1]);
335     (*regs)[2] = tswapreg(env->regs[2]);
336     (*regs)[3] = tswapreg(env->regs[3]);
337     (*regs)[4] = tswapreg(env->regs[4]);
338     (*regs)[5] = tswapreg(env->regs[5]);
339     (*regs)[6] = tswapreg(env->regs[6]);
340     (*regs)[7] = tswapreg(env->regs[7]);
341     (*regs)[8] = tswapreg(env->regs[8]);
342     (*regs)[9] = tswapreg(env->regs[9]);
343     (*regs)[10] = tswapreg(env->regs[10]);
344     (*regs)[11] = tswapreg(env->regs[11]);
345     (*regs)[12] = tswapreg(env->regs[12]);
346     (*regs)[13] = tswapreg(env->regs[13]);
347     (*regs)[14] = tswapreg(env->regs[14]);
348     (*regs)[15] = tswapreg(env->regs[15]);
349 
350     (*regs)[16] = tswapreg(cpsr_read((CPUARMState *)env));
351     (*regs)[17] = tswapreg(env->regs[0]); /* XXX */
352 }
353 
354 #define USE_ELF_CORE_DUMP
355 #define ELF_EXEC_PAGESIZE       4096
356 
357 enum
358 {
359     ARM_HWCAP_ARM_SWP       = 1 << 0,
360     ARM_HWCAP_ARM_HALF      = 1 << 1,
361     ARM_HWCAP_ARM_THUMB     = 1 << 2,
362     ARM_HWCAP_ARM_26BIT     = 1 << 3,
363     ARM_HWCAP_ARM_FAST_MULT = 1 << 4,
364     ARM_HWCAP_ARM_FPA       = 1 << 5,
365     ARM_HWCAP_ARM_VFP       = 1 << 6,
366     ARM_HWCAP_ARM_EDSP      = 1 << 7,
367     ARM_HWCAP_ARM_JAVA      = 1 << 8,
368     ARM_HWCAP_ARM_IWMMXT    = 1 << 9,
369     ARM_HWCAP_ARM_CRUNCH    = 1 << 10,
370     ARM_HWCAP_ARM_THUMBEE   = 1 << 11,
371     ARM_HWCAP_ARM_NEON      = 1 << 12,
372     ARM_HWCAP_ARM_VFPv3     = 1 << 13,
373     ARM_HWCAP_ARM_VFPv3D16  = 1 << 14,
374     ARM_HWCAP_ARM_TLS       = 1 << 15,
375     ARM_HWCAP_ARM_VFPv4     = 1 << 16,
376     ARM_HWCAP_ARM_IDIVA     = 1 << 17,
377     ARM_HWCAP_ARM_IDIVT     = 1 << 18,
378     ARM_HWCAP_ARM_VFPD32    = 1 << 19,
379     ARM_HWCAP_ARM_LPAE      = 1 << 20,
380     ARM_HWCAP_ARM_EVTSTRM   = 1 << 21,
381 };
382 
383 enum {
384     ARM_HWCAP2_ARM_AES      = 1 << 0,
385     ARM_HWCAP2_ARM_PMULL    = 1 << 1,
386     ARM_HWCAP2_ARM_SHA1     = 1 << 2,
387     ARM_HWCAP2_ARM_SHA2     = 1 << 3,
388     ARM_HWCAP2_ARM_CRC32    = 1 << 4,
389 };
390 
391 /* The commpage only exists for 32 bit kernels */
392 
393 #define HI_COMMPAGE (intptr_t)0xffff0f00u
394 
395 static bool init_guest_commpage(void)
396 {
397     void *want = g2h_untagged(HI_COMMPAGE & -qemu_host_page_size);
398     void *addr = mmap(want, qemu_host_page_size, PROT_READ | PROT_WRITE,
399                       MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0);
400 
401     if (addr == MAP_FAILED) {
402         perror("Allocating guest commpage");
403         exit(EXIT_FAILURE);
404     }
405     if (addr != want) {
406         return false;
407     }
408 
409     /* Set kernel helper versions; rest of page is 0.  */
410     __put_user(5, (uint32_t *)g2h_untagged(0xffff0ffcu));
411 
412     if (mprotect(addr, qemu_host_page_size, PROT_READ)) {
413         perror("Protecting guest commpage");
414         exit(EXIT_FAILURE);
415     }
416     return true;
417 }
418 
419 #define ELF_HWCAP get_elf_hwcap()
420 #define ELF_HWCAP2 get_elf_hwcap2()
421 
422 static uint32_t get_elf_hwcap(void)
423 {
424     ARMCPU *cpu = ARM_CPU(thread_cpu);
425     uint32_t hwcaps = 0;
426 
427     hwcaps |= ARM_HWCAP_ARM_SWP;
428     hwcaps |= ARM_HWCAP_ARM_HALF;
429     hwcaps |= ARM_HWCAP_ARM_THUMB;
430     hwcaps |= ARM_HWCAP_ARM_FAST_MULT;
431 
432     /* probe for the extra features */
433 #define GET_FEATURE(feat, hwcap) \
434     do { if (arm_feature(&cpu->env, feat)) { hwcaps |= hwcap; } } while (0)
435 
436 #define GET_FEATURE_ID(feat, hwcap) \
437     do { if (cpu_isar_feature(feat, cpu)) { hwcaps |= hwcap; } } while (0)
438 
439     /* EDSP is in v5TE and above, but all our v5 CPUs are v5TE */
440     GET_FEATURE(ARM_FEATURE_V5, ARM_HWCAP_ARM_EDSP);
441     GET_FEATURE(ARM_FEATURE_IWMMXT, ARM_HWCAP_ARM_IWMMXT);
442     GET_FEATURE(ARM_FEATURE_THUMB2EE, ARM_HWCAP_ARM_THUMBEE);
443     GET_FEATURE(ARM_FEATURE_NEON, ARM_HWCAP_ARM_NEON);
444     GET_FEATURE(ARM_FEATURE_V6K, ARM_HWCAP_ARM_TLS);
445     GET_FEATURE(ARM_FEATURE_LPAE, ARM_HWCAP_ARM_LPAE);
446     GET_FEATURE_ID(aa32_arm_div, ARM_HWCAP_ARM_IDIVA);
447     GET_FEATURE_ID(aa32_thumb_div, ARM_HWCAP_ARM_IDIVT);
448     GET_FEATURE_ID(aa32_vfp, ARM_HWCAP_ARM_VFP);
449 
450     if (cpu_isar_feature(aa32_fpsp_v3, cpu) ||
451         cpu_isar_feature(aa32_fpdp_v3, cpu)) {
452         hwcaps |= ARM_HWCAP_ARM_VFPv3;
453         if (cpu_isar_feature(aa32_simd_r32, cpu)) {
454             hwcaps |= ARM_HWCAP_ARM_VFPD32;
455         } else {
456             hwcaps |= ARM_HWCAP_ARM_VFPv3D16;
457         }
458     }
459     GET_FEATURE_ID(aa32_simdfmac, ARM_HWCAP_ARM_VFPv4);
460 
461     return hwcaps;
462 }
463 
464 static uint32_t get_elf_hwcap2(void)
465 {
466     ARMCPU *cpu = ARM_CPU(thread_cpu);
467     uint32_t hwcaps = 0;
468 
469     GET_FEATURE_ID(aa32_aes, ARM_HWCAP2_ARM_AES);
470     GET_FEATURE_ID(aa32_pmull, ARM_HWCAP2_ARM_PMULL);
471     GET_FEATURE_ID(aa32_sha1, ARM_HWCAP2_ARM_SHA1);
472     GET_FEATURE_ID(aa32_sha2, ARM_HWCAP2_ARM_SHA2);
473     GET_FEATURE_ID(aa32_crc32, ARM_HWCAP2_ARM_CRC32);
474     return hwcaps;
475 }
476 
477 #undef GET_FEATURE
478 #undef GET_FEATURE_ID
479 
480 #define ELF_PLATFORM get_elf_platform()
481 
482 static const char *get_elf_platform(void)
483 {
484     CPUARMState *env = thread_cpu->env_ptr;
485 
486 #ifdef TARGET_WORDS_BIGENDIAN
487 # define END  "b"
488 #else
489 # define END  "l"
490 #endif
491 
492     if (arm_feature(env, ARM_FEATURE_V8)) {
493         return "v8" END;
494     } else if (arm_feature(env, ARM_FEATURE_V7)) {
495         if (arm_feature(env, ARM_FEATURE_M)) {
496             return "v7m" END;
497         } else {
498             return "v7" END;
499         }
500     } else if (arm_feature(env, ARM_FEATURE_V6)) {
501         return "v6" END;
502     } else if (arm_feature(env, ARM_FEATURE_V5)) {
503         return "v5" END;
504     } else {
505         return "v4" END;
506     }
507 
508 #undef END
509 }
510 
511 #else
512 /* 64 bit ARM definitions */
513 #define ELF_START_MMAP 0x80000000
514 
515 #define ELF_ARCH        EM_AARCH64
516 #define ELF_CLASS       ELFCLASS64
517 #ifdef TARGET_WORDS_BIGENDIAN
518 # define ELF_PLATFORM    "aarch64_be"
519 #else
520 # define ELF_PLATFORM    "aarch64"
521 #endif
522 
523 static inline void init_thread(struct target_pt_regs *regs,
524                                struct image_info *infop)
525 {
526     abi_long stack = infop->start_stack;
527     memset(regs, 0, sizeof(*regs));
528 
529     regs->pc = infop->entry & ~0x3ULL;
530     regs->sp = stack;
531 }
532 
533 #define ELF_NREG    34
534 typedef target_elf_greg_t  target_elf_gregset_t[ELF_NREG];
535 
536 static void elf_core_copy_regs(target_elf_gregset_t *regs,
537                                const CPUARMState *env)
538 {
539     int i;
540 
541     for (i = 0; i < 32; i++) {
542         (*regs)[i] = tswapreg(env->xregs[i]);
543     }
544     (*regs)[32] = tswapreg(env->pc);
545     (*regs)[33] = tswapreg(pstate_read((CPUARMState *)env));
546 }
547 
548 #define USE_ELF_CORE_DUMP
549 #define ELF_EXEC_PAGESIZE       4096
550 
551 enum {
552     ARM_HWCAP_A64_FP            = 1 << 0,
553     ARM_HWCAP_A64_ASIMD         = 1 << 1,
554     ARM_HWCAP_A64_EVTSTRM       = 1 << 2,
555     ARM_HWCAP_A64_AES           = 1 << 3,
556     ARM_HWCAP_A64_PMULL         = 1 << 4,
557     ARM_HWCAP_A64_SHA1          = 1 << 5,
558     ARM_HWCAP_A64_SHA2          = 1 << 6,
559     ARM_HWCAP_A64_CRC32         = 1 << 7,
560     ARM_HWCAP_A64_ATOMICS       = 1 << 8,
561     ARM_HWCAP_A64_FPHP          = 1 << 9,
562     ARM_HWCAP_A64_ASIMDHP       = 1 << 10,
563     ARM_HWCAP_A64_CPUID         = 1 << 11,
564     ARM_HWCAP_A64_ASIMDRDM      = 1 << 12,
565     ARM_HWCAP_A64_JSCVT         = 1 << 13,
566     ARM_HWCAP_A64_FCMA          = 1 << 14,
567     ARM_HWCAP_A64_LRCPC         = 1 << 15,
568     ARM_HWCAP_A64_DCPOP         = 1 << 16,
569     ARM_HWCAP_A64_SHA3          = 1 << 17,
570     ARM_HWCAP_A64_SM3           = 1 << 18,
571     ARM_HWCAP_A64_SM4           = 1 << 19,
572     ARM_HWCAP_A64_ASIMDDP       = 1 << 20,
573     ARM_HWCAP_A64_SHA512        = 1 << 21,
574     ARM_HWCAP_A64_SVE           = 1 << 22,
575     ARM_HWCAP_A64_ASIMDFHM      = 1 << 23,
576     ARM_HWCAP_A64_DIT           = 1 << 24,
577     ARM_HWCAP_A64_USCAT         = 1 << 25,
578     ARM_HWCAP_A64_ILRCPC        = 1 << 26,
579     ARM_HWCAP_A64_FLAGM         = 1 << 27,
580     ARM_HWCAP_A64_SSBS          = 1 << 28,
581     ARM_HWCAP_A64_SB            = 1 << 29,
582     ARM_HWCAP_A64_PACA          = 1 << 30,
583     ARM_HWCAP_A64_PACG          = 1UL << 31,
584 
585     ARM_HWCAP2_A64_DCPODP       = 1 << 0,
586     ARM_HWCAP2_A64_SVE2         = 1 << 1,
587     ARM_HWCAP2_A64_SVEAES       = 1 << 2,
588     ARM_HWCAP2_A64_SVEPMULL     = 1 << 3,
589     ARM_HWCAP2_A64_SVEBITPERM   = 1 << 4,
590     ARM_HWCAP2_A64_SVESHA3      = 1 << 5,
591     ARM_HWCAP2_A64_SVESM4       = 1 << 6,
592     ARM_HWCAP2_A64_FLAGM2       = 1 << 7,
593     ARM_HWCAP2_A64_FRINT        = 1 << 8,
594     ARM_HWCAP2_A64_SVEI8MM      = 1 << 9,
595     ARM_HWCAP2_A64_SVEF32MM     = 1 << 10,
596     ARM_HWCAP2_A64_SVEF64MM     = 1 << 11,
597     ARM_HWCAP2_A64_SVEBF16      = 1 << 12,
598     ARM_HWCAP2_A64_I8MM         = 1 << 13,
599     ARM_HWCAP2_A64_BF16         = 1 << 14,
600     ARM_HWCAP2_A64_DGH          = 1 << 15,
601     ARM_HWCAP2_A64_RNG          = 1 << 16,
602     ARM_HWCAP2_A64_BTI          = 1 << 17,
603     ARM_HWCAP2_A64_MTE          = 1 << 18,
604 };
605 
606 #define ELF_HWCAP   get_elf_hwcap()
607 #define ELF_HWCAP2  get_elf_hwcap2()
608 
609 #define GET_FEATURE_ID(feat, hwcap) \
610     do { if (cpu_isar_feature(feat, cpu)) { hwcaps |= hwcap; } } while (0)
611 
612 static uint32_t get_elf_hwcap(void)
613 {
614     ARMCPU *cpu = ARM_CPU(thread_cpu);
615     uint32_t hwcaps = 0;
616 
617     hwcaps |= ARM_HWCAP_A64_FP;
618     hwcaps |= ARM_HWCAP_A64_ASIMD;
619     hwcaps |= ARM_HWCAP_A64_CPUID;
620 
621     /* probe for the extra features */
622 
623     GET_FEATURE_ID(aa64_aes, ARM_HWCAP_A64_AES);
624     GET_FEATURE_ID(aa64_pmull, ARM_HWCAP_A64_PMULL);
625     GET_FEATURE_ID(aa64_sha1, ARM_HWCAP_A64_SHA1);
626     GET_FEATURE_ID(aa64_sha256, ARM_HWCAP_A64_SHA2);
627     GET_FEATURE_ID(aa64_sha512, ARM_HWCAP_A64_SHA512);
628     GET_FEATURE_ID(aa64_crc32, ARM_HWCAP_A64_CRC32);
629     GET_FEATURE_ID(aa64_sha3, ARM_HWCAP_A64_SHA3);
630     GET_FEATURE_ID(aa64_sm3, ARM_HWCAP_A64_SM3);
631     GET_FEATURE_ID(aa64_sm4, ARM_HWCAP_A64_SM4);
632     GET_FEATURE_ID(aa64_fp16, ARM_HWCAP_A64_FPHP | ARM_HWCAP_A64_ASIMDHP);
633     GET_FEATURE_ID(aa64_atomics, ARM_HWCAP_A64_ATOMICS);
634     GET_FEATURE_ID(aa64_rdm, ARM_HWCAP_A64_ASIMDRDM);
635     GET_FEATURE_ID(aa64_dp, ARM_HWCAP_A64_ASIMDDP);
636     GET_FEATURE_ID(aa64_fcma, ARM_HWCAP_A64_FCMA);
637     GET_FEATURE_ID(aa64_sve, ARM_HWCAP_A64_SVE);
638     GET_FEATURE_ID(aa64_pauth, ARM_HWCAP_A64_PACA | ARM_HWCAP_A64_PACG);
639     GET_FEATURE_ID(aa64_fhm, ARM_HWCAP_A64_ASIMDFHM);
640     GET_FEATURE_ID(aa64_jscvt, ARM_HWCAP_A64_JSCVT);
641     GET_FEATURE_ID(aa64_sb, ARM_HWCAP_A64_SB);
642     GET_FEATURE_ID(aa64_condm_4, ARM_HWCAP_A64_FLAGM);
643     GET_FEATURE_ID(aa64_dcpop, ARM_HWCAP_A64_DCPOP);
644     GET_FEATURE_ID(aa64_rcpc_8_3, ARM_HWCAP_A64_LRCPC);
645     GET_FEATURE_ID(aa64_rcpc_8_4, ARM_HWCAP_A64_ILRCPC);
646 
647     return hwcaps;
648 }
649 
650 static uint32_t get_elf_hwcap2(void)
651 {
652     ARMCPU *cpu = ARM_CPU(thread_cpu);
653     uint32_t hwcaps = 0;
654 
655     GET_FEATURE_ID(aa64_dcpodp, ARM_HWCAP2_A64_DCPODP);
656     GET_FEATURE_ID(aa64_sve2, ARM_HWCAP2_A64_SVE2);
657     GET_FEATURE_ID(aa64_sve2_aes, ARM_HWCAP2_A64_SVEAES);
658     GET_FEATURE_ID(aa64_sve2_pmull128, ARM_HWCAP2_A64_SVEPMULL);
659     GET_FEATURE_ID(aa64_sve2_bitperm, ARM_HWCAP2_A64_SVEBITPERM);
660     GET_FEATURE_ID(aa64_sve2_sha3, ARM_HWCAP2_A64_SVESHA3);
661     GET_FEATURE_ID(aa64_sve2_sm4, ARM_HWCAP2_A64_SVESM4);
662     GET_FEATURE_ID(aa64_condm_5, ARM_HWCAP2_A64_FLAGM2);
663     GET_FEATURE_ID(aa64_frint, ARM_HWCAP2_A64_FRINT);
664     GET_FEATURE_ID(aa64_sve_i8mm, ARM_HWCAP2_A64_SVEI8MM);
665     GET_FEATURE_ID(aa64_sve_f32mm, ARM_HWCAP2_A64_SVEF32MM);
666     GET_FEATURE_ID(aa64_sve_f64mm, ARM_HWCAP2_A64_SVEF64MM);
667     GET_FEATURE_ID(aa64_sve_bf16, ARM_HWCAP2_A64_SVEBF16);
668     GET_FEATURE_ID(aa64_i8mm, ARM_HWCAP2_A64_I8MM);
669     GET_FEATURE_ID(aa64_bf16, ARM_HWCAP2_A64_BF16);
670     GET_FEATURE_ID(aa64_rndr, ARM_HWCAP2_A64_RNG);
671     GET_FEATURE_ID(aa64_bti, ARM_HWCAP2_A64_BTI);
672     GET_FEATURE_ID(aa64_mte, ARM_HWCAP2_A64_MTE);
673 
674     return hwcaps;
675 }
676 
677 #undef GET_FEATURE_ID
678 
679 #endif /* not TARGET_AARCH64 */
680 #endif /* TARGET_ARM */
681 
682 #ifdef TARGET_SPARC
683 #ifdef TARGET_SPARC64
684 
685 #define ELF_START_MMAP 0x80000000
686 #define ELF_HWCAP  (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \
687                     | HWCAP_SPARC_MULDIV | HWCAP_SPARC_V9)
688 #ifndef TARGET_ABI32
689 #define elf_check_arch(x) ( (x) == EM_SPARCV9 || (x) == EM_SPARC32PLUS )
690 #else
691 #define elf_check_arch(x) ( (x) == EM_SPARC32PLUS || (x) == EM_SPARC )
692 #endif
693 
694 #define ELF_CLASS   ELFCLASS64
695 #define ELF_ARCH    EM_SPARCV9
696 #else
697 #define ELF_START_MMAP 0x80000000
698 #define ELF_HWCAP  (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \
699                     | HWCAP_SPARC_MULDIV)
700 #define ELF_CLASS   ELFCLASS32
701 #define ELF_ARCH    EM_SPARC
702 #endif /* TARGET_SPARC64 */
703 
704 static inline void init_thread(struct target_pt_regs *regs,
705                                struct image_info *infop)
706 {
707     /* Note that target_cpu_copy_regs does not read psr/tstate. */
708     regs->pc = infop->entry;
709     regs->npc = regs->pc + 4;
710     regs->y = 0;
711     regs->u_regs[14] = (infop->start_stack - 16 * sizeof(abi_ulong)
712                         - TARGET_STACK_BIAS);
713 }
714 #endif /* TARGET_SPARC */
715 
716 #ifdef TARGET_PPC
717 
718 #define ELF_MACHINE    PPC_ELF_MACHINE
719 #define ELF_START_MMAP 0x80000000
720 
721 #if defined(TARGET_PPC64)
722 
723 #define elf_check_arch(x) ( (x) == EM_PPC64 )
724 
725 #define ELF_CLASS       ELFCLASS64
726 
727 #else
728 
729 #define ELF_CLASS       ELFCLASS32
730 
731 #endif
732 
733 #define ELF_ARCH        EM_PPC
734 
735 /* Feature masks for the Aux Vector Hardware Capabilities (AT_HWCAP).
736    See arch/powerpc/include/asm/cputable.h.  */
737 enum {
738     QEMU_PPC_FEATURE_32 = 0x80000000,
739     QEMU_PPC_FEATURE_64 = 0x40000000,
740     QEMU_PPC_FEATURE_601_INSTR = 0x20000000,
741     QEMU_PPC_FEATURE_HAS_ALTIVEC = 0x10000000,
742     QEMU_PPC_FEATURE_HAS_FPU = 0x08000000,
743     QEMU_PPC_FEATURE_HAS_MMU = 0x04000000,
744     QEMU_PPC_FEATURE_HAS_4xxMAC = 0x02000000,
745     QEMU_PPC_FEATURE_UNIFIED_CACHE = 0x01000000,
746     QEMU_PPC_FEATURE_HAS_SPE = 0x00800000,
747     QEMU_PPC_FEATURE_HAS_EFP_SINGLE = 0x00400000,
748     QEMU_PPC_FEATURE_HAS_EFP_DOUBLE = 0x00200000,
749     QEMU_PPC_FEATURE_NO_TB = 0x00100000,
750     QEMU_PPC_FEATURE_POWER4 = 0x00080000,
751     QEMU_PPC_FEATURE_POWER5 = 0x00040000,
752     QEMU_PPC_FEATURE_POWER5_PLUS = 0x00020000,
753     QEMU_PPC_FEATURE_CELL = 0x00010000,
754     QEMU_PPC_FEATURE_BOOKE = 0x00008000,
755     QEMU_PPC_FEATURE_SMT = 0x00004000,
756     QEMU_PPC_FEATURE_ICACHE_SNOOP = 0x00002000,
757     QEMU_PPC_FEATURE_ARCH_2_05 = 0x00001000,
758     QEMU_PPC_FEATURE_PA6T = 0x00000800,
759     QEMU_PPC_FEATURE_HAS_DFP = 0x00000400,
760     QEMU_PPC_FEATURE_POWER6_EXT = 0x00000200,
761     QEMU_PPC_FEATURE_ARCH_2_06 = 0x00000100,
762     QEMU_PPC_FEATURE_HAS_VSX = 0x00000080,
763     QEMU_PPC_FEATURE_PSERIES_PERFMON_COMPAT = 0x00000040,
764 
765     QEMU_PPC_FEATURE_TRUE_LE = 0x00000002,
766     QEMU_PPC_FEATURE_PPC_LE = 0x00000001,
767 
768     /* Feature definitions in AT_HWCAP2.  */
769     QEMU_PPC_FEATURE2_ARCH_2_07 = 0x80000000, /* ISA 2.07 */
770     QEMU_PPC_FEATURE2_HAS_HTM = 0x40000000, /* Hardware Transactional Memory */
771     QEMU_PPC_FEATURE2_HAS_DSCR = 0x20000000, /* Data Stream Control Register */
772     QEMU_PPC_FEATURE2_HAS_EBB = 0x10000000, /* Event Base Branching */
773     QEMU_PPC_FEATURE2_HAS_ISEL = 0x08000000, /* Integer Select */
774     QEMU_PPC_FEATURE2_HAS_TAR = 0x04000000, /* Target Address Register */
775     QEMU_PPC_FEATURE2_VEC_CRYPTO = 0x02000000,
776     QEMU_PPC_FEATURE2_HTM_NOSC = 0x01000000,
777     QEMU_PPC_FEATURE2_ARCH_3_00 = 0x00800000, /* ISA 3.00 */
778     QEMU_PPC_FEATURE2_HAS_IEEE128 = 0x00400000, /* VSX IEEE Bin Float 128-bit */
779     QEMU_PPC_FEATURE2_DARN = 0x00200000, /* darn random number insn */
780     QEMU_PPC_FEATURE2_SCV = 0x00100000, /* scv syscall */
781     QEMU_PPC_FEATURE2_HTM_NO_SUSPEND = 0x00080000, /* TM w/o suspended state */
782 };
783 
784 #define ELF_HWCAP get_elf_hwcap()
785 
786 static uint32_t get_elf_hwcap(void)
787 {
788     PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);
789     uint32_t features = 0;
790 
791     /* We don't have to be terribly complete here; the high points are
792        Altivec/FP/SPE support.  Anything else is just a bonus.  */
793 #define GET_FEATURE(flag, feature)                                      \
794     do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0)
795 #define GET_FEATURE2(flags, feature) \
796     do { \
797         if ((cpu->env.insns_flags2 & flags) == flags) { \
798             features |= feature; \
799         } \
800     } while (0)
801     GET_FEATURE(PPC_64B, QEMU_PPC_FEATURE_64);
802     GET_FEATURE(PPC_FLOAT, QEMU_PPC_FEATURE_HAS_FPU);
803     GET_FEATURE(PPC_ALTIVEC, QEMU_PPC_FEATURE_HAS_ALTIVEC);
804     GET_FEATURE(PPC_SPE, QEMU_PPC_FEATURE_HAS_SPE);
805     GET_FEATURE(PPC_SPE_SINGLE, QEMU_PPC_FEATURE_HAS_EFP_SINGLE);
806     GET_FEATURE(PPC_SPE_DOUBLE, QEMU_PPC_FEATURE_HAS_EFP_DOUBLE);
807     GET_FEATURE(PPC_BOOKE, QEMU_PPC_FEATURE_BOOKE);
808     GET_FEATURE(PPC_405_MAC, QEMU_PPC_FEATURE_HAS_4xxMAC);
809     GET_FEATURE2(PPC2_DFP, QEMU_PPC_FEATURE_HAS_DFP);
810     GET_FEATURE2(PPC2_VSX, QEMU_PPC_FEATURE_HAS_VSX);
811     GET_FEATURE2((PPC2_PERM_ISA206 | PPC2_DIVE_ISA206 | PPC2_ATOMIC_ISA206 |
812                   PPC2_FP_CVT_ISA206 | PPC2_FP_TST_ISA206),
813                   QEMU_PPC_FEATURE_ARCH_2_06);
814 #undef GET_FEATURE
815 #undef GET_FEATURE2
816 
817     return features;
818 }
819 
820 #define ELF_HWCAP2 get_elf_hwcap2()
821 
822 static uint32_t get_elf_hwcap2(void)
823 {
824     PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);
825     uint32_t features = 0;
826 
827 #define GET_FEATURE(flag, feature)                                      \
828     do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0)
829 #define GET_FEATURE2(flag, feature)                                      \
830     do { if (cpu->env.insns_flags2 & flag) { features |= feature; } } while (0)
831 
832     GET_FEATURE(PPC_ISEL, QEMU_PPC_FEATURE2_HAS_ISEL);
833     GET_FEATURE2(PPC2_BCTAR_ISA207, QEMU_PPC_FEATURE2_HAS_TAR);
834     GET_FEATURE2((PPC2_BCTAR_ISA207 | PPC2_LSQ_ISA207 | PPC2_ALTIVEC_207 |
835                   PPC2_ISA207S), QEMU_PPC_FEATURE2_ARCH_2_07 |
836                   QEMU_PPC_FEATURE2_VEC_CRYPTO);
837     GET_FEATURE2(PPC2_ISA300, QEMU_PPC_FEATURE2_ARCH_3_00 |
838                  QEMU_PPC_FEATURE2_DARN | QEMU_PPC_FEATURE2_HAS_IEEE128);
839 
840 #undef GET_FEATURE
841 #undef GET_FEATURE2
842 
843     return features;
844 }
845 
846 /*
847  * The requirements here are:
848  * - keep the final alignment of sp (sp & 0xf)
849  * - make sure the 32-bit value at the first 16 byte aligned position of
850  *   AUXV is greater than 16 for glibc compatibility.
851  *   AT_IGNOREPPC is used for that.
852  * - for compatibility with glibc ARCH_DLINFO must always be defined on PPC,
853  *   even if DLINFO_ARCH_ITEMS goes to zero or is undefined.
854  */
855 #define DLINFO_ARCH_ITEMS       5
856 #define ARCH_DLINFO                                     \
857     do {                                                \
858         PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);              \
859         /*                                              \
860          * Handle glibc compatibility: these magic entries must \
861          * be at the lowest addresses in the final auxv.        \
862          */                                             \
863         NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC);        \
864         NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC);        \
865         NEW_AUX_ENT(AT_DCACHEBSIZE, cpu->env.dcache_line_size); \
866         NEW_AUX_ENT(AT_ICACHEBSIZE, cpu->env.icache_line_size); \
867         NEW_AUX_ENT(AT_UCACHEBSIZE, 0);                 \
868     } while (0)
869 
870 static inline void init_thread(struct target_pt_regs *_regs, struct image_info *infop)
871 {
872     _regs->gpr[1] = infop->start_stack;
873 #if defined(TARGET_PPC64)
874     if (get_ppc64_abi(infop) < 2) {
875         uint64_t val;
876         get_user_u64(val, infop->entry + 8);
877         _regs->gpr[2] = val + infop->load_bias;
878         get_user_u64(val, infop->entry);
879         infop->entry = val + infop->load_bias;
880     } else {
881         _regs->gpr[12] = infop->entry;  /* r12 set to global entry address */
882     }
883 #endif
884     _regs->nip = infop->entry;
885 }
886 
887 /* See linux kernel: arch/powerpc/include/asm/elf.h.  */
888 #define ELF_NREG 48
889 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
890 
891 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUPPCState *env)
892 {
893     int i;
894     target_ulong ccr = 0;
895 
896     for (i = 0; i < ARRAY_SIZE(env->gpr); i++) {
897         (*regs)[i] = tswapreg(env->gpr[i]);
898     }
899 
900     (*regs)[32] = tswapreg(env->nip);
901     (*regs)[33] = tswapreg(env->msr);
902     (*regs)[35] = tswapreg(env->ctr);
903     (*regs)[36] = tswapreg(env->lr);
904     (*regs)[37] = tswapreg(cpu_read_xer(env));
905 
906     for (i = 0; i < ARRAY_SIZE(env->crf); i++) {
907         ccr |= env->crf[i] << (32 - ((i + 1) * 4));
908     }
909     (*regs)[38] = tswapreg(ccr);
910 }
911 
912 #define USE_ELF_CORE_DUMP
913 #define ELF_EXEC_PAGESIZE       4096
914 
915 #endif
916 
917 #ifdef TARGET_MIPS
918 
919 #define ELF_START_MMAP 0x80000000
920 
921 #ifdef TARGET_MIPS64
922 #define ELF_CLASS   ELFCLASS64
923 #else
924 #define ELF_CLASS   ELFCLASS32
925 #endif
926 #define ELF_ARCH    EM_MIPS
927 
928 #ifdef TARGET_ABI_MIPSN32
929 #define elf_check_abi(x) ((x) & EF_MIPS_ABI2)
930 #else
931 #define elf_check_abi(x) (!((x) & EF_MIPS_ABI2))
932 #endif
933 
934 static inline void init_thread(struct target_pt_regs *regs,
935                                struct image_info *infop)
936 {
937     regs->cp0_status = 2 << CP0St_KSU;
938     regs->cp0_epc = infop->entry;
939     regs->regs[29] = infop->start_stack;
940 }
941 
942 /* See linux kernel: arch/mips/include/asm/elf.h.  */
943 #define ELF_NREG 45
944 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
945 
946 /* See linux kernel: arch/mips/include/asm/reg.h.  */
947 enum {
948 #ifdef TARGET_MIPS64
949     TARGET_EF_R0 = 0,
950 #else
951     TARGET_EF_R0 = 6,
952 #endif
953     TARGET_EF_R26 = TARGET_EF_R0 + 26,
954     TARGET_EF_R27 = TARGET_EF_R0 + 27,
955     TARGET_EF_LO = TARGET_EF_R0 + 32,
956     TARGET_EF_HI = TARGET_EF_R0 + 33,
957     TARGET_EF_CP0_EPC = TARGET_EF_R0 + 34,
958     TARGET_EF_CP0_BADVADDR = TARGET_EF_R0 + 35,
959     TARGET_EF_CP0_STATUS = TARGET_EF_R0 + 36,
960     TARGET_EF_CP0_CAUSE = TARGET_EF_R0 + 37
961 };
962 
963 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs.  */
964 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMIPSState *env)
965 {
966     int i;
967 
968     for (i = 0; i < TARGET_EF_R0; i++) {
969         (*regs)[i] = 0;
970     }
971     (*regs)[TARGET_EF_R0] = 0;
972 
973     for (i = 1; i < ARRAY_SIZE(env->active_tc.gpr); i++) {
974         (*regs)[TARGET_EF_R0 + i] = tswapreg(env->active_tc.gpr[i]);
975     }
976 
977     (*regs)[TARGET_EF_R26] = 0;
978     (*regs)[TARGET_EF_R27] = 0;
979     (*regs)[TARGET_EF_LO] = tswapreg(env->active_tc.LO[0]);
980     (*regs)[TARGET_EF_HI] = tswapreg(env->active_tc.HI[0]);
981     (*regs)[TARGET_EF_CP0_EPC] = tswapreg(env->active_tc.PC);
982     (*regs)[TARGET_EF_CP0_BADVADDR] = tswapreg(env->CP0_BadVAddr);
983     (*regs)[TARGET_EF_CP0_STATUS] = tswapreg(env->CP0_Status);
984     (*regs)[TARGET_EF_CP0_CAUSE] = tswapreg(env->CP0_Cause);
985 }
986 
987 #define USE_ELF_CORE_DUMP
988 #define ELF_EXEC_PAGESIZE        4096
989 
990 /* See arch/mips/include/uapi/asm/hwcap.h.  */
991 enum {
992     HWCAP_MIPS_R6           = (1 << 0),
993     HWCAP_MIPS_MSA          = (1 << 1),
994     HWCAP_MIPS_CRC32        = (1 << 2),
995     HWCAP_MIPS_MIPS16       = (1 << 3),
996     HWCAP_MIPS_MDMX         = (1 << 4),
997     HWCAP_MIPS_MIPS3D       = (1 << 5),
998     HWCAP_MIPS_SMARTMIPS    = (1 << 6),
999     HWCAP_MIPS_DSP          = (1 << 7),
1000     HWCAP_MIPS_DSP2         = (1 << 8),
1001     HWCAP_MIPS_DSP3         = (1 << 9),
1002     HWCAP_MIPS_MIPS16E2     = (1 << 10),
1003     HWCAP_LOONGSON_MMI      = (1 << 11),
1004     HWCAP_LOONGSON_EXT      = (1 << 12),
1005     HWCAP_LOONGSON_EXT2     = (1 << 13),
1006     HWCAP_LOONGSON_CPUCFG   = (1 << 14),
1007 };
1008 
1009 #define ELF_HWCAP get_elf_hwcap()
1010 
1011 #define GET_FEATURE_INSN(_flag, _hwcap) \
1012     do { if (cpu->env.insn_flags & (_flag)) { hwcaps |= _hwcap; } } while (0)
1013 
1014 #define GET_FEATURE_REG_SET(_reg, _mask, _hwcap) \
1015     do { if (cpu->env._reg & (_mask)) { hwcaps |= _hwcap; } } while (0)
1016 
1017 #define GET_FEATURE_REG_EQU(_reg, _start, _length, _val, _hwcap) \
1018     do { \
1019         if (extract32(cpu->env._reg, (_start), (_length)) == (_val)) { \
1020             hwcaps |= _hwcap; \
1021         } \
1022     } while (0)
1023 
1024 static uint32_t get_elf_hwcap(void)
1025 {
1026     MIPSCPU *cpu = MIPS_CPU(thread_cpu);
1027     uint32_t hwcaps = 0;
1028 
1029     GET_FEATURE_REG_EQU(CP0_Config0, CP0C0_AR, CP0C0_AR_LENGTH,
1030                         2, HWCAP_MIPS_R6);
1031     GET_FEATURE_REG_SET(CP0_Config3, 1 << CP0C3_MSAP, HWCAP_MIPS_MSA);
1032     GET_FEATURE_INSN(ASE_LMMI, HWCAP_LOONGSON_MMI);
1033     GET_FEATURE_INSN(ASE_LEXT, HWCAP_LOONGSON_EXT);
1034 
1035     return hwcaps;
1036 }
1037 
1038 #undef GET_FEATURE_REG_EQU
1039 #undef GET_FEATURE_REG_SET
1040 #undef GET_FEATURE_INSN
1041 
1042 #endif /* TARGET_MIPS */
1043 
1044 #ifdef TARGET_MICROBLAZE
1045 
1046 #define ELF_START_MMAP 0x80000000
1047 
1048 #define elf_check_arch(x) ( (x) == EM_MICROBLAZE || (x) == EM_MICROBLAZE_OLD)
1049 
1050 #define ELF_CLASS   ELFCLASS32
1051 #define ELF_ARCH    EM_MICROBLAZE
1052 
1053 static inline void init_thread(struct target_pt_regs *regs,
1054                                struct image_info *infop)
1055 {
1056     regs->pc = infop->entry;
1057     regs->r1 = infop->start_stack;
1058 
1059 }
1060 
1061 #define ELF_EXEC_PAGESIZE        4096
1062 
1063 #define USE_ELF_CORE_DUMP
1064 #define ELF_NREG 38
1065 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1066 
1067 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs.  */
1068 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMBState *env)
1069 {
1070     int i, pos = 0;
1071 
1072     for (i = 0; i < 32; i++) {
1073         (*regs)[pos++] = tswapreg(env->regs[i]);
1074     }
1075 
1076     (*regs)[pos++] = tswapreg(env->pc);
1077     (*regs)[pos++] = tswapreg(mb_cpu_read_msr(env));
1078     (*regs)[pos++] = 0;
1079     (*regs)[pos++] = tswapreg(env->ear);
1080     (*regs)[pos++] = 0;
1081     (*regs)[pos++] = tswapreg(env->esr);
1082 }
1083 
1084 #endif /* TARGET_MICROBLAZE */
1085 
1086 #ifdef TARGET_NIOS2
1087 
1088 #define ELF_START_MMAP 0x80000000
1089 
1090 #define elf_check_arch(x) ((x) == EM_ALTERA_NIOS2)
1091 
1092 #define ELF_CLASS   ELFCLASS32
1093 #define ELF_ARCH    EM_ALTERA_NIOS2
1094 
1095 static void init_thread(struct target_pt_regs *regs, struct image_info *infop)
1096 {
1097     regs->ea = infop->entry;
1098     regs->sp = infop->start_stack;
1099     regs->estatus = 0x3;
1100 }
1101 
1102 #define LO_COMMPAGE  TARGET_PAGE_SIZE
1103 
1104 static bool init_guest_commpage(void)
1105 {
1106     static const uint8_t kuser_page[4 + 2 * 64] = {
1107         /* __kuser_helper_version */
1108         [0x00] = 0x02, 0x00, 0x00, 0x00,
1109 
1110         /* __kuser_cmpxchg */
1111         [0x04] = 0x3a, 0x6c, 0x3b, 0x00,  /* trap 16 */
1112                  0x3a, 0x28, 0x00, 0xf8,  /* ret */
1113 
1114         /* __kuser_sigtramp */
1115         [0x44] = 0xc4, 0x22, 0x80, 0x00,  /* movi r2, __NR_rt_sigreturn */
1116                  0x3a, 0x68, 0x3b, 0x00,  /* trap 0 */
1117     };
1118 
1119     void *want = g2h_untagged(LO_COMMPAGE & -qemu_host_page_size);
1120     void *addr = mmap(want, qemu_host_page_size, PROT_READ | PROT_WRITE,
1121                       MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0);
1122 
1123     if (addr == MAP_FAILED) {
1124         perror("Allocating guest commpage");
1125         exit(EXIT_FAILURE);
1126     }
1127     if (addr != want) {
1128         return false;
1129     }
1130 
1131     memcpy(addr, kuser_page, sizeof(kuser_page));
1132 
1133     if (mprotect(addr, qemu_host_page_size, PROT_READ)) {
1134         perror("Protecting guest commpage");
1135         exit(EXIT_FAILURE);
1136     }
1137 
1138     page_set_flags(LO_COMMPAGE, LO_COMMPAGE + TARGET_PAGE_SIZE,
1139                    PAGE_READ | PAGE_EXEC | PAGE_VALID);
1140     return true;
1141 }
1142 
1143 #define ELF_EXEC_PAGESIZE        4096
1144 
1145 #define USE_ELF_CORE_DUMP
1146 #define ELF_NREG 49
1147 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1148 
1149 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs.  */
1150 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1151                                const CPUNios2State *env)
1152 {
1153     int i;
1154 
1155     (*regs)[0] = -1;
1156     for (i = 1; i < 8; i++)    /* r0-r7 */
1157         (*regs)[i] = tswapreg(env->regs[i + 7]);
1158 
1159     for (i = 8; i < 16; i++)   /* r8-r15 */
1160         (*regs)[i] = tswapreg(env->regs[i - 8]);
1161 
1162     for (i = 16; i < 24; i++)  /* r16-r23 */
1163         (*regs)[i] = tswapreg(env->regs[i + 7]);
1164     (*regs)[24] = -1;    /* R_ET */
1165     (*regs)[25] = -1;    /* R_BT */
1166     (*regs)[26] = tswapreg(env->regs[R_GP]);
1167     (*regs)[27] = tswapreg(env->regs[R_SP]);
1168     (*regs)[28] = tswapreg(env->regs[R_FP]);
1169     (*regs)[29] = tswapreg(env->regs[R_EA]);
1170     (*regs)[30] = -1;    /* R_SSTATUS */
1171     (*regs)[31] = tswapreg(env->regs[R_RA]);
1172 
1173     (*regs)[32] = tswapreg(env->regs[R_PC]);
1174 
1175     (*regs)[33] = -1; /* R_STATUS */
1176     (*regs)[34] = tswapreg(env->regs[CR_ESTATUS]);
1177 
1178     for (i = 35; i < 49; i++)    /* ... */
1179         (*regs)[i] = -1;
1180 }
1181 
1182 #endif /* TARGET_NIOS2 */
1183 
1184 #ifdef TARGET_OPENRISC
1185 
1186 #define ELF_START_MMAP 0x08000000
1187 
1188 #define ELF_ARCH EM_OPENRISC
1189 #define ELF_CLASS ELFCLASS32
1190 #define ELF_DATA  ELFDATA2MSB
1191 
1192 static inline void init_thread(struct target_pt_regs *regs,
1193                                struct image_info *infop)
1194 {
1195     regs->pc = infop->entry;
1196     regs->gpr[1] = infop->start_stack;
1197 }
1198 
1199 #define USE_ELF_CORE_DUMP
1200 #define ELF_EXEC_PAGESIZE 8192
1201 
1202 /* See linux kernel arch/openrisc/include/asm/elf.h.  */
1203 #define ELF_NREG 34 /* gprs and pc, sr */
1204 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1205 
1206 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1207                                const CPUOpenRISCState *env)
1208 {
1209     int i;
1210 
1211     for (i = 0; i < 32; i++) {
1212         (*regs)[i] = tswapreg(cpu_get_gpr(env, i));
1213     }
1214     (*regs)[32] = tswapreg(env->pc);
1215     (*regs)[33] = tswapreg(cpu_get_sr(env));
1216 }
1217 #define ELF_HWCAP 0
1218 #define ELF_PLATFORM NULL
1219 
1220 #endif /* TARGET_OPENRISC */
1221 
1222 #ifdef TARGET_SH4
1223 
1224 #define ELF_START_MMAP 0x80000000
1225 
1226 #define ELF_CLASS ELFCLASS32
1227 #define ELF_ARCH  EM_SH
1228 
1229 static inline void init_thread(struct target_pt_regs *regs,
1230                                struct image_info *infop)
1231 {
1232     /* Check other registers XXXXX */
1233     regs->pc = infop->entry;
1234     regs->regs[15] = infop->start_stack;
1235 }
1236 
1237 /* See linux kernel: arch/sh/include/asm/elf.h.  */
1238 #define ELF_NREG 23
1239 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1240 
1241 /* See linux kernel: arch/sh/include/asm/ptrace.h.  */
1242 enum {
1243     TARGET_REG_PC = 16,
1244     TARGET_REG_PR = 17,
1245     TARGET_REG_SR = 18,
1246     TARGET_REG_GBR = 19,
1247     TARGET_REG_MACH = 20,
1248     TARGET_REG_MACL = 21,
1249     TARGET_REG_SYSCALL = 22
1250 };
1251 
1252 static inline void elf_core_copy_regs(target_elf_gregset_t *regs,
1253                                       const CPUSH4State *env)
1254 {
1255     int i;
1256 
1257     for (i = 0; i < 16; i++) {
1258         (*regs)[i] = tswapreg(env->gregs[i]);
1259     }
1260 
1261     (*regs)[TARGET_REG_PC] = tswapreg(env->pc);
1262     (*regs)[TARGET_REG_PR] = tswapreg(env->pr);
1263     (*regs)[TARGET_REG_SR] = tswapreg(env->sr);
1264     (*regs)[TARGET_REG_GBR] = tswapreg(env->gbr);
1265     (*regs)[TARGET_REG_MACH] = tswapreg(env->mach);
1266     (*regs)[TARGET_REG_MACL] = tswapreg(env->macl);
1267     (*regs)[TARGET_REG_SYSCALL] = 0; /* FIXME */
1268 }
1269 
1270 #define USE_ELF_CORE_DUMP
1271 #define ELF_EXEC_PAGESIZE        4096
1272 
1273 enum {
1274     SH_CPU_HAS_FPU            = 0x0001, /* Hardware FPU support */
1275     SH_CPU_HAS_P2_FLUSH_BUG   = 0x0002, /* Need to flush the cache in P2 area */
1276     SH_CPU_HAS_MMU_PAGE_ASSOC = 0x0004, /* SH3: TLB way selection bit support */
1277     SH_CPU_HAS_DSP            = 0x0008, /* SH-DSP: DSP support */
1278     SH_CPU_HAS_PERF_COUNTER   = 0x0010, /* Hardware performance counters */
1279     SH_CPU_HAS_PTEA           = 0x0020, /* PTEA register */
1280     SH_CPU_HAS_LLSC           = 0x0040, /* movli.l/movco.l */
1281     SH_CPU_HAS_L2_CACHE       = 0x0080, /* Secondary cache / URAM */
1282     SH_CPU_HAS_OP32           = 0x0100, /* 32-bit instruction support */
1283     SH_CPU_HAS_PTEAEX         = 0x0200, /* PTE ASID Extension support */
1284 };
1285 
1286 #define ELF_HWCAP get_elf_hwcap()
1287 
1288 static uint32_t get_elf_hwcap(void)
1289 {
1290     SuperHCPU *cpu = SUPERH_CPU(thread_cpu);
1291     uint32_t hwcap = 0;
1292 
1293     hwcap |= SH_CPU_HAS_FPU;
1294 
1295     if (cpu->env.features & SH_FEATURE_SH4A) {
1296         hwcap |= SH_CPU_HAS_LLSC;
1297     }
1298 
1299     return hwcap;
1300 }
1301 
1302 #endif
1303 
1304 #ifdef TARGET_CRIS
1305 
1306 #define ELF_START_MMAP 0x80000000
1307 
1308 #define ELF_CLASS ELFCLASS32
1309 #define ELF_ARCH  EM_CRIS
1310 
1311 static inline void init_thread(struct target_pt_regs *regs,
1312                                struct image_info *infop)
1313 {
1314     regs->erp = infop->entry;
1315 }
1316 
1317 #define ELF_EXEC_PAGESIZE        8192
1318 
1319 #endif
1320 
1321 #ifdef TARGET_M68K
1322 
1323 #define ELF_START_MMAP 0x80000000
1324 
1325 #define ELF_CLASS       ELFCLASS32
1326 #define ELF_ARCH        EM_68K
1327 
1328 /* ??? Does this need to do anything?
1329    #define ELF_PLAT_INIT(_r) */
1330 
1331 static inline void init_thread(struct target_pt_regs *regs,
1332                                struct image_info *infop)
1333 {
1334     regs->usp = infop->start_stack;
1335     regs->sr = 0;
1336     regs->pc = infop->entry;
1337 }
1338 
1339 /* See linux kernel: arch/m68k/include/asm/elf.h.  */
1340 #define ELF_NREG 20
1341 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1342 
1343 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUM68KState *env)
1344 {
1345     (*regs)[0] = tswapreg(env->dregs[1]);
1346     (*regs)[1] = tswapreg(env->dregs[2]);
1347     (*regs)[2] = tswapreg(env->dregs[3]);
1348     (*regs)[3] = tswapreg(env->dregs[4]);
1349     (*regs)[4] = tswapreg(env->dregs[5]);
1350     (*regs)[5] = tswapreg(env->dregs[6]);
1351     (*regs)[6] = tswapreg(env->dregs[7]);
1352     (*regs)[7] = tswapreg(env->aregs[0]);
1353     (*regs)[8] = tswapreg(env->aregs[1]);
1354     (*regs)[9] = tswapreg(env->aregs[2]);
1355     (*regs)[10] = tswapreg(env->aregs[3]);
1356     (*regs)[11] = tswapreg(env->aregs[4]);
1357     (*regs)[12] = tswapreg(env->aregs[5]);
1358     (*regs)[13] = tswapreg(env->aregs[6]);
1359     (*regs)[14] = tswapreg(env->dregs[0]);
1360     (*regs)[15] = tswapreg(env->aregs[7]);
1361     (*regs)[16] = tswapreg(env->dregs[0]); /* FIXME: orig_d0 */
1362     (*regs)[17] = tswapreg(env->sr);
1363     (*regs)[18] = tswapreg(env->pc);
1364     (*regs)[19] = 0;  /* FIXME: regs->format | regs->vector */
1365 }
1366 
1367 #define USE_ELF_CORE_DUMP
1368 #define ELF_EXEC_PAGESIZE       8192
1369 
1370 #endif
1371 
1372 #ifdef TARGET_ALPHA
1373 
1374 #define ELF_START_MMAP (0x30000000000ULL)
1375 
1376 #define ELF_CLASS      ELFCLASS64
1377 #define ELF_ARCH       EM_ALPHA
1378 
1379 static inline void init_thread(struct target_pt_regs *regs,
1380                                struct image_info *infop)
1381 {
1382     regs->pc = infop->entry;
1383     regs->ps = 8;
1384     regs->usp = infop->start_stack;
1385 }
1386 
1387 #define ELF_EXEC_PAGESIZE        8192
1388 
1389 #endif /* TARGET_ALPHA */
1390 
1391 #ifdef TARGET_S390X
1392 
1393 #define ELF_START_MMAP (0x20000000000ULL)
1394 
1395 #define ELF_CLASS	ELFCLASS64
1396 #define ELF_DATA	ELFDATA2MSB
1397 #define ELF_ARCH	EM_S390
1398 
1399 #include "elf.h"
1400 
1401 #define ELF_HWCAP get_elf_hwcap()
1402 
1403 #define GET_FEATURE(_feat, _hwcap) \
1404     do { if (s390_has_feat(_feat)) { hwcap |= _hwcap; } } while (0)
1405 
1406 static uint32_t get_elf_hwcap(void)
1407 {
1408     /*
1409      * Let's assume we always have esan3 and zarch.
1410      * 31-bit processes can use 64-bit registers (high gprs).
1411      */
1412     uint32_t hwcap = HWCAP_S390_ESAN3 | HWCAP_S390_ZARCH | HWCAP_S390_HIGH_GPRS;
1413 
1414     GET_FEATURE(S390_FEAT_STFLE, HWCAP_S390_STFLE);
1415     GET_FEATURE(S390_FEAT_MSA, HWCAP_S390_MSA);
1416     GET_FEATURE(S390_FEAT_LONG_DISPLACEMENT, HWCAP_S390_LDISP);
1417     GET_FEATURE(S390_FEAT_EXTENDED_IMMEDIATE, HWCAP_S390_EIMM);
1418     if (s390_has_feat(S390_FEAT_EXTENDED_TRANSLATION_3) &&
1419         s390_has_feat(S390_FEAT_ETF3_ENH)) {
1420         hwcap |= HWCAP_S390_ETF3EH;
1421     }
1422     GET_FEATURE(S390_FEAT_VECTOR, HWCAP_S390_VXRS);
1423     GET_FEATURE(S390_FEAT_VECTOR_ENH, HWCAP_S390_VXRS_EXT);
1424 
1425     return hwcap;
1426 }
1427 
1428 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop)
1429 {
1430     regs->psw.addr = infop->entry;
1431     regs->psw.mask = PSW_MASK_64 | PSW_MASK_32;
1432     regs->gprs[15] = infop->start_stack;
1433 }
1434 
1435 /* See linux kernel: arch/s390/include/uapi/asm/ptrace.h (s390_regs).  */
1436 #define ELF_NREG 27
1437 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1438 
1439 enum {
1440     TARGET_REG_PSWM = 0,
1441     TARGET_REG_PSWA = 1,
1442     TARGET_REG_GPRS = 2,
1443     TARGET_REG_ARS = 18,
1444     TARGET_REG_ORIG_R2 = 26,
1445 };
1446 
1447 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1448                                const CPUS390XState *env)
1449 {
1450     int i;
1451     uint32_t *aregs;
1452 
1453     (*regs)[TARGET_REG_PSWM] = tswapreg(env->psw.mask);
1454     (*regs)[TARGET_REG_PSWA] = tswapreg(env->psw.addr);
1455     for (i = 0; i < 16; i++) {
1456         (*regs)[TARGET_REG_GPRS + i] = tswapreg(env->regs[i]);
1457     }
1458     aregs = (uint32_t *)&((*regs)[TARGET_REG_ARS]);
1459     for (i = 0; i < 16; i++) {
1460         aregs[i] = tswap32(env->aregs[i]);
1461     }
1462     (*regs)[TARGET_REG_ORIG_R2] = 0;
1463 }
1464 
1465 #define USE_ELF_CORE_DUMP
1466 #define ELF_EXEC_PAGESIZE 4096
1467 
1468 #endif /* TARGET_S390X */
1469 
1470 #ifdef TARGET_RISCV
1471 
1472 #define ELF_START_MMAP 0x80000000
1473 #define ELF_ARCH  EM_RISCV
1474 
1475 #ifdef TARGET_RISCV32
1476 #define ELF_CLASS ELFCLASS32
1477 #else
1478 #define ELF_CLASS ELFCLASS64
1479 #endif
1480 
1481 #define ELF_HWCAP get_elf_hwcap()
1482 
1483 static uint32_t get_elf_hwcap(void)
1484 {
1485 #define MISA_BIT(EXT) (1 << (EXT - 'A'))
1486     RISCVCPU *cpu = RISCV_CPU(thread_cpu);
1487     uint32_t mask = MISA_BIT('I') | MISA_BIT('M') | MISA_BIT('A')
1488                     | MISA_BIT('F') | MISA_BIT('D') | MISA_BIT('C');
1489 
1490     return cpu->env.misa_ext & mask;
1491 #undef MISA_BIT
1492 }
1493 
1494 static inline void init_thread(struct target_pt_regs *regs,
1495                                struct image_info *infop)
1496 {
1497     regs->sepc = infop->entry;
1498     regs->sp = infop->start_stack;
1499 }
1500 
1501 #define ELF_EXEC_PAGESIZE 4096
1502 
1503 #endif /* TARGET_RISCV */
1504 
1505 #ifdef TARGET_HPPA
1506 
1507 #define ELF_START_MMAP  0x80000000
1508 #define ELF_CLASS       ELFCLASS32
1509 #define ELF_ARCH        EM_PARISC
1510 #define ELF_PLATFORM    "PARISC"
1511 #define STACK_GROWS_DOWN 0
1512 #define STACK_ALIGNMENT  64
1513 
1514 static inline void init_thread(struct target_pt_regs *regs,
1515                                struct image_info *infop)
1516 {
1517     regs->iaoq[0] = infop->entry;
1518     regs->iaoq[1] = infop->entry + 4;
1519     regs->gr[23] = 0;
1520     regs->gr[24] = infop->arg_start;
1521     regs->gr[25] = (infop->arg_end - infop->arg_start) / sizeof(abi_ulong);
1522     /* The top-of-stack contains a linkage buffer.  */
1523     regs->gr[30] = infop->start_stack + 64;
1524     regs->gr[31] = infop->entry;
1525 }
1526 
1527 #endif /* TARGET_HPPA */
1528 
1529 #ifdef TARGET_XTENSA
1530 
1531 #define ELF_START_MMAP 0x20000000
1532 
1533 #define ELF_CLASS       ELFCLASS32
1534 #define ELF_ARCH        EM_XTENSA
1535 
1536 static inline void init_thread(struct target_pt_regs *regs,
1537                                struct image_info *infop)
1538 {
1539     regs->windowbase = 0;
1540     regs->windowstart = 1;
1541     regs->areg[1] = infop->start_stack;
1542     regs->pc = infop->entry;
1543 }
1544 
1545 /* See linux kernel: arch/xtensa/include/asm/elf.h.  */
1546 #define ELF_NREG 128
1547 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1548 
1549 enum {
1550     TARGET_REG_PC,
1551     TARGET_REG_PS,
1552     TARGET_REG_LBEG,
1553     TARGET_REG_LEND,
1554     TARGET_REG_LCOUNT,
1555     TARGET_REG_SAR,
1556     TARGET_REG_WINDOWSTART,
1557     TARGET_REG_WINDOWBASE,
1558     TARGET_REG_THREADPTR,
1559     TARGET_REG_AR0 = 64,
1560 };
1561 
1562 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1563                                const CPUXtensaState *env)
1564 {
1565     unsigned i;
1566 
1567     (*regs)[TARGET_REG_PC] = tswapreg(env->pc);
1568     (*regs)[TARGET_REG_PS] = tswapreg(env->sregs[PS] & ~PS_EXCM);
1569     (*regs)[TARGET_REG_LBEG] = tswapreg(env->sregs[LBEG]);
1570     (*regs)[TARGET_REG_LEND] = tswapreg(env->sregs[LEND]);
1571     (*regs)[TARGET_REG_LCOUNT] = tswapreg(env->sregs[LCOUNT]);
1572     (*regs)[TARGET_REG_SAR] = tswapreg(env->sregs[SAR]);
1573     (*regs)[TARGET_REG_WINDOWSTART] = tswapreg(env->sregs[WINDOW_START]);
1574     (*regs)[TARGET_REG_WINDOWBASE] = tswapreg(env->sregs[WINDOW_BASE]);
1575     (*regs)[TARGET_REG_THREADPTR] = tswapreg(env->uregs[THREADPTR]);
1576     xtensa_sync_phys_from_window((CPUXtensaState *)env);
1577     for (i = 0; i < env->config->nareg; ++i) {
1578         (*regs)[TARGET_REG_AR0 + i] = tswapreg(env->phys_regs[i]);
1579     }
1580 }
1581 
1582 #define USE_ELF_CORE_DUMP
1583 #define ELF_EXEC_PAGESIZE       4096
1584 
1585 #endif /* TARGET_XTENSA */
1586 
1587 #ifdef TARGET_HEXAGON
1588 
1589 #define ELF_START_MMAP 0x20000000
1590 
1591 #define ELF_CLASS       ELFCLASS32
1592 #define ELF_ARCH        EM_HEXAGON
1593 
1594 static inline void init_thread(struct target_pt_regs *regs,
1595                                struct image_info *infop)
1596 {
1597     regs->sepc = infop->entry;
1598     regs->sp = infop->start_stack;
1599 }
1600 
1601 #endif /* TARGET_HEXAGON */
1602 
1603 #ifndef ELF_PLATFORM
1604 #define ELF_PLATFORM (NULL)
1605 #endif
1606 
1607 #ifndef ELF_MACHINE
1608 #define ELF_MACHINE ELF_ARCH
1609 #endif
1610 
1611 #ifndef elf_check_arch
1612 #define elf_check_arch(x) ((x) == ELF_ARCH)
1613 #endif
1614 
1615 #ifndef elf_check_abi
1616 #define elf_check_abi(x) (1)
1617 #endif
1618 
1619 #ifndef ELF_HWCAP
1620 #define ELF_HWCAP 0
1621 #endif
1622 
1623 #ifndef STACK_GROWS_DOWN
1624 #define STACK_GROWS_DOWN 1
1625 #endif
1626 
1627 #ifndef STACK_ALIGNMENT
1628 #define STACK_ALIGNMENT 16
1629 #endif
1630 
1631 #ifdef TARGET_ABI32
1632 #undef ELF_CLASS
1633 #define ELF_CLASS ELFCLASS32
1634 #undef bswaptls
1635 #define bswaptls(ptr) bswap32s(ptr)
1636 #endif
1637 
1638 #include "elf.h"
1639 
1640 /* We must delay the following stanzas until after "elf.h". */
1641 #if defined(TARGET_AARCH64)
1642 
1643 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz,
1644                                     const uint32_t *data,
1645                                     struct image_info *info,
1646                                     Error **errp)
1647 {
1648     if (pr_type == GNU_PROPERTY_AARCH64_FEATURE_1_AND) {
1649         if (pr_datasz != sizeof(uint32_t)) {
1650             error_setg(errp, "Ill-formed GNU_PROPERTY_AARCH64_FEATURE_1_AND");
1651             return false;
1652         }
1653         /* We will extract GNU_PROPERTY_AARCH64_FEATURE_1_BTI later. */
1654         info->note_flags = *data;
1655     }
1656     return true;
1657 }
1658 #define ARCH_USE_GNU_PROPERTY 1
1659 
1660 #else
1661 
1662 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz,
1663                                     const uint32_t *data,
1664                                     struct image_info *info,
1665                                     Error **errp)
1666 {
1667     g_assert_not_reached();
1668 }
1669 #define ARCH_USE_GNU_PROPERTY 0
1670 
1671 #endif
1672 
1673 struct exec
1674 {
1675     unsigned int a_info;   /* Use macros N_MAGIC, etc for access */
1676     unsigned int a_text;   /* length of text, in bytes */
1677     unsigned int a_data;   /* length of data, in bytes */
1678     unsigned int a_bss;    /* length of uninitialized data area, in bytes */
1679     unsigned int a_syms;   /* length of symbol table data in file, in bytes */
1680     unsigned int a_entry;  /* start address */
1681     unsigned int a_trsize; /* length of relocation info for text, in bytes */
1682     unsigned int a_drsize; /* length of relocation info for data, in bytes */
1683 };
1684 
1685 
1686 #define N_MAGIC(exec) ((exec).a_info & 0xffff)
1687 #define OMAGIC 0407
1688 #define NMAGIC 0410
1689 #define ZMAGIC 0413
1690 #define QMAGIC 0314
1691 
1692 /* Necessary parameters */
1693 #define TARGET_ELF_EXEC_PAGESIZE \
1694         (((eppnt->p_align & ~qemu_host_page_mask) != 0) ? \
1695          TARGET_PAGE_SIZE : MAX(qemu_host_page_size, TARGET_PAGE_SIZE))
1696 #define TARGET_ELF_PAGELENGTH(_v) ROUND_UP((_v), TARGET_ELF_EXEC_PAGESIZE)
1697 #define TARGET_ELF_PAGESTART(_v) ((_v) & \
1698                                  ~(abi_ulong)(TARGET_ELF_EXEC_PAGESIZE-1))
1699 #define TARGET_ELF_PAGEOFFSET(_v) ((_v) & (TARGET_ELF_EXEC_PAGESIZE-1))
1700 
1701 #define DLINFO_ITEMS 16
1702 
1703 static inline void memcpy_fromfs(void * to, const void * from, unsigned long n)
1704 {
1705     memcpy(to, from, n);
1706 }
1707 
1708 #ifdef BSWAP_NEEDED
1709 static void bswap_ehdr(struct elfhdr *ehdr)
1710 {
1711     bswap16s(&ehdr->e_type);            /* Object file type */
1712     bswap16s(&ehdr->e_machine);         /* Architecture */
1713     bswap32s(&ehdr->e_version);         /* Object file version */
1714     bswaptls(&ehdr->e_entry);           /* Entry point virtual address */
1715     bswaptls(&ehdr->e_phoff);           /* Program header table file offset */
1716     bswaptls(&ehdr->e_shoff);           /* Section header table file offset */
1717     bswap32s(&ehdr->e_flags);           /* Processor-specific flags */
1718     bswap16s(&ehdr->e_ehsize);          /* ELF header size in bytes */
1719     bswap16s(&ehdr->e_phentsize);       /* Program header table entry size */
1720     bswap16s(&ehdr->e_phnum);           /* Program header table entry count */
1721     bswap16s(&ehdr->e_shentsize);       /* Section header table entry size */
1722     bswap16s(&ehdr->e_shnum);           /* Section header table entry count */
1723     bswap16s(&ehdr->e_shstrndx);        /* Section header string table index */
1724 }
1725 
1726 static void bswap_phdr(struct elf_phdr *phdr, int phnum)
1727 {
1728     int i;
1729     for (i = 0; i < phnum; ++i, ++phdr) {
1730         bswap32s(&phdr->p_type);        /* Segment type */
1731         bswap32s(&phdr->p_flags);       /* Segment flags */
1732         bswaptls(&phdr->p_offset);      /* Segment file offset */
1733         bswaptls(&phdr->p_vaddr);       /* Segment virtual address */
1734         bswaptls(&phdr->p_paddr);       /* Segment physical address */
1735         bswaptls(&phdr->p_filesz);      /* Segment size in file */
1736         bswaptls(&phdr->p_memsz);       /* Segment size in memory */
1737         bswaptls(&phdr->p_align);       /* Segment alignment */
1738     }
1739 }
1740 
1741 static void bswap_shdr(struct elf_shdr *shdr, int shnum)
1742 {
1743     int i;
1744     for (i = 0; i < shnum; ++i, ++shdr) {
1745         bswap32s(&shdr->sh_name);
1746         bswap32s(&shdr->sh_type);
1747         bswaptls(&shdr->sh_flags);
1748         bswaptls(&shdr->sh_addr);
1749         bswaptls(&shdr->sh_offset);
1750         bswaptls(&shdr->sh_size);
1751         bswap32s(&shdr->sh_link);
1752         bswap32s(&shdr->sh_info);
1753         bswaptls(&shdr->sh_addralign);
1754         bswaptls(&shdr->sh_entsize);
1755     }
1756 }
1757 
1758 static void bswap_sym(struct elf_sym *sym)
1759 {
1760     bswap32s(&sym->st_name);
1761     bswaptls(&sym->st_value);
1762     bswaptls(&sym->st_size);
1763     bswap16s(&sym->st_shndx);
1764 }
1765 
1766 #ifdef TARGET_MIPS
1767 static void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags)
1768 {
1769     bswap16s(&abiflags->version);
1770     bswap32s(&abiflags->ases);
1771     bswap32s(&abiflags->isa_ext);
1772     bswap32s(&abiflags->flags1);
1773     bswap32s(&abiflags->flags2);
1774 }
1775 #endif
1776 #else
1777 static inline void bswap_ehdr(struct elfhdr *ehdr) { }
1778 static inline void bswap_phdr(struct elf_phdr *phdr, int phnum) { }
1779 static inline void bswap_shdr(struct elf_shdr *shdr, int shnum) { }
1780 static inline void bswap_sym(struct elf_sym *sym) { }
1781 #ifdef TARGET_MIPS
1782 static inline void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags) { }
1783 #endif
1784 #endif
1785 
1786 #ifdef USE_ELF_CORE_DUMP
1787 static int elf_core_dump(int, const CPUArchState *);
1788 #endif /* USE_ELF_CORE_DUMP */
1789 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias);
1790 
1791 /* Verify the portions of EHDR within E_IDENT for the target.
1792    This can be performed before bswapping the entire header.  */
1793 static bool elf_check_ident(struct elfhdr *ehdr)
1794 {
1795     return (ehdr->e_ident[EI_MAG0] == ELFMAG0
1796             && ehdr->e_ident[EI_MAG1] == ELFMAG1
1797             && ehdr->e_ident[EI_MAG2] == ELFMAG2
1798             && ehdr->e_ident[EI_MAG3] == ELFMAG3
1799             && ehdr->e_ident[EI_CLASS] == ELF_CLASS
1800             && ehdr->e_ident[EI_DATA] == ELF_DATA
1801             && ehdr->e_ident[EI_VERSION] == EV_CURRENT);
1802 }
1803 
1804 /* Verify the portions of EHDR outside of E_IDENT for the target.
1805    This has to wait until after bswapping the header.  */
1806 static bool elf_check_ehdr(struct elfhdr *ehdr)
1807 {
1808     return (elf_check_arch(ehdr->e_machine)
1809             && elf_check_abi(ehdr->e_flags)
1810             && ehdr->e_ehsize == sizeof(struct elfhdr)
1811             && ehdr->e_phentsize == sizeof(struct elf_phdr)
1812             && (ehdr->e_type == ET_EXEC || ehdr->e_type == ET_DYN));
1813 }
1814 
1815 /*
1816  * 'copy_elf_strings()' copies argument/envelope strings from user
1817  * memory to free pages in kernel mem. These are in a format ready
1818  * to be put directly into the top of new user memory.
1819  *
1820  */
1821 static abi_ulong copy_elf_strings(int argc, char **argv, char *scratch,
1822                                   abi_ulong p, abi_ulong stack_limit)
1823 {
1824     char *tmp;
1825     int len, i;
1826     abi_ulong top = p;
1827 
1828     if (!p) {
1829         return 0;       /* bullet-proofing */
1830     }
1831 
1832     if (STACK_GROWS_DOWN) {
1833         int offset = ((p - 1) % TARGET_PAGE_SIZE) + 1;
1834         for (i = argc - 1; i >= 0; --i) {
1835             tmp = argv[i];
1836             if (!tmp) {
1837                 fprintf(stderr, "VFS: argc is wrong");
1838                 exit(-1);
1839             }
1840             len = strlen(tmp) + 1;
1841             tmp += len;
1842 
1843             if (len > (p - stack_limit)) {
1844                 return 0;
1845             }
1846             while (len) {
1847                 int bytes_to_copy = (len > offset) ? offset : len;
1848                 tmp -= bytes_to_copy;
1849                 p -= bytes_to_copy;
1850                 offset -= bytes_to_copy;
1851                 len -= bytes_to_copy;
1852 
1853                 memcpy_fromfs(scratch + offset, tmp, bytes_to_copy);
1854 
1855                 if (offset == 0) {
1856                     memcpy_to_target(p, scratch, top - p);
1857                     top = p;
1858                     offset = TARGET_PAGE_SIZE;
1859                 }
1860             }
1861         }
1862         if (p != top) {
1863             memcpy_to_target(p, scratch + offset, top - p);
1864         }
1865     } else {
1866         int remaining = TARGET_PAGE_SIZE - (p % TARGET_PAGE_SIZE);
1867         for (i = 0; i < argc; ++i) {
1868             tmp = argv[i];
1869             if (!tmp) {
1870                 fprintf(stderr, "VFS: argc is wrong");
1871                 exit(-1);
1872             }
1873             len = strlen(tmp) + 1;
1874             if (len > (stack_limit - p)) {
1875                 return 0;
1876             }
1877             while (len) {
1878                 int bytes_to_copy = (len > remaining) ? remaining : len;
1879 
1880                 memcpy_fromfs(scratch + (p - top), tmp, bytes_to_copy);
1881 
1882                 tmp += bytes_to_copy;
1883                 remaining -= bytes_to_copy;
1884                 p += bytes_to_copy;
1885                 len -= bytes_to_copy;
1886 
1887                 if (remaining == 0) {
1888                     memcpy_to_target(top, scratch, p - top);
1889                     top = p;
1890                     remaining = TARGET_PAGE_SIZE;
1891                 }
1892             }
1893         }
1894         if (p != top) {
1895             memcpy_to_target(top, scratch, p - top);
1896         }
1897     }
1898 
1899     return p;
1900 }
1901 
1902 /* Older linux kernels provide up to MAX_ARG_PAGES (default: 32) of
1903  * argument/environment space. Newer kernels (>2.6.33) allow more,
1904  * dependent on stack size, but guarantee at least 32 pages for
1905  * backwards compatibility.
1906  */
1907 #define STACK_LOWER_LIMIT (32 * TARGET_PAGE_SIZE)
1908 
1909 static abi_ulong setup_arg_pages(struct linux_binprm *bprm,
1910                                  struct image_info *info)
1911 {
1912     abi_ulong size, error, guard;
1913 
1914     size = guest_stack_size;
1915     if (size < STACK_LOWER_LIMIT) {
1916         size = STACK_LOWER_LIMIT;
1917     }
1918     guard = TARGET_PAGE_SIZE;
1919     if (guard < qemu_real_host_page_size) {
1920         guard = qemu_real_host_page_size;
1921     }
1922 
1923     error = target_mmap(0, size + guard, PROT_READ | PROT_WRITE,
1924                         MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
1925     if (error == -1) {
1926         perror("mmap stack");
1927         exit(-1);
1928     }
1929 
1930     /* We reserve one extra page at the top of the stack as guard.  */
1931     if (STACK_GROWS_DOWN) {
1932         target_mprotect(error, guard, PROT_NONE);
1933         info->stack_limit = error + guard;
1934         return info->stack_limit + size - sizeof(void *);
1935     } else {
1936         target_mprotect(error + size, guard, PROT_NONE);
1937         info->stack_limit = error + size;
1938         return error;
1939     }
1940 }
1941 
1942 /* Map and zero the bss.  We need to explicitly zero any fractional pages
1943    after the data section (i.e. bss).  */
1944 static void zero_bss(abi_ulong elf_bss, abi_ulong last_bss, int prot)
1945 {
1946     uintptr_t host_start, host_map_start, host_end;
1947 
1948     last_bss = TARGET_PAGE_ALIGN(last_bss);
1949 
1950     /* ??? There is confusion between qemu_real_host_page_size and
1951        qemu_host_page_size here and elsewhere in target_mmap, which
1952        may lead to the end of the data section mapping from the file
1953        not being mapped.  At least there was an explicit test and
1954        comment for that here, suggesting that "the file size must
1955        be known".  The comment probably pre-dates the introduction
1956        of the fstat system call in target_mmap which does in fact
1957        find out the size.  What isn't clear is if the workaround
1958        here is still actually needed.  For now, continue with it,
1959        but merge it with the "normal" mmap that would allocate the bss.  */
1960 
1961     host_start = (uintptr_t) g2h_untagged(elf_bss);
1962     host_end = (uintptr_t) g2h_untagged(last_bss);
1963     host_map_start = REAL_HOST_PAGE_ALIGN(host_start);
1964 
1965     if (host_map_start < host_end) {
1966         void *p = mmap((void *)host_map_start, host_end - host_map_start,
1967                        prot, MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
1968         if (p == MAP_FAILED) {
1969             perror("cannot mmap brk");
1970             exit(-1);
1971         }
1972     }
1973 
1974     /* Ensure that the bss page(s) are valid */
1975     if ((page_get_flags(last_bss-1) & prot) != prot) {
1976         page_set_flags(elf_bss & TARGET_PAGE_MASK, last_bss, prot | PAGE_VALID);
1977     }
1978 
1979     if (host_start < host_map_start) {
1980         memset((void *)host_start, 0, host_map_start - host_start);
1981     }
1982 }
1983 
1984 #ifdef TARGET_ARM
1985 static int elf_is_fdpic(struct elfhdr *exec)
1986 {
1987     return exec->e_ident[EI_OSABI] == ELFOSABI_ARM_FDPIC;
1988 }
1989 #else
1990 /* Default implementation, always false.  */
1991 static int elf_is_fdpic(struct elfhdr *exec)
1992 {
1993     return 0;
1994 }
1995 #endif
1996 
1997 static abi_ulong loader_build_fdpic_loadmap(struct image_info *info, abi_ulong sp)
1998 {
1999     uint16_t n;
2000     struct elf32_fdpic_loadseg *loadsegs = info->loadsegs;
2001 
2002     /* elf32_fdpic_loadseg */
2003     n = info->nsegs;
2004     while (n--) {
2005         sp -= 12;
2006         put_user_u32(loadsegs[n].addr, sp+0);
2007         put_user_u32(loadsegs[n].p_vaddr, sp+4);
2008         put_user_u32(loadsegs[n].p_memsz, sp+8);
2009     }
2010 
2011     /* elf32_fdpic_loadmap */
2012     sp -= 4;
2013     put_user_u16(0, sp+0); /* version */
2014     put_user_u16(info->nsegs, sp+2); /* nsegs */
2015 
2016     info->personality = PER_LINUX_FDPIC;
2017     info->loadmap_addr = sp;
2018 
2019     return sp;
2020 }
2021 
2022 static abi_ulong create_elf_tables(abi_ulong p, int argc, int envc,
2023                                    struct elfhdr *exec,
2024                                    struct image_info *info,
2025                                    struct image_info *interp_info)
2026 {
2027     abi_ulong sp;
2028     abi_ulong u_argc, u_argv, u_envp, u_auxv;
2029     int size;
2030     int i;
2031     abi_ulong u_rand_bytes;
2032     uint8_t k_rand_bytes[16];
2033     abi_ulong u_platform;
2034     const char *k_platform;
2035     const int n = sizeof(elf_addr_t);
2036 
2037     sp = p;
2038 
2039     /* Needs to be before we load the env/argc/... */
2040     if (elf_is_fdpic(exec)) {
2041         /* Need 4 byte alignment for these structs */
2042         sp &= ~3;
2043         sp = loader_build_fdpic_loadmap(info, sp);
2044         info->other_info = interp_info;
2045         if (interp_info) {
2046             interp_info->other_info = info;
2047             sp = loader_build_fdpic_loadmap(interp_info, sp);
2048             info->interpreter_loadmap_addr = interp_info->loadmap_addr;
2049             info->interpreter_pt_dynamic_addr = interp_info->pt_dynamic_addr;
2050         } else {
2051             info->interpreter_loadmap_addr = 0;
2052             info->interpreter_pt_dynamic_addr = 0;
2053         }
2054     }
2055 
2056     u_platform = 0;
2057     k_platform = ELF_PLATFORM;
2058     if (k_platform) {
2059         size_t len = strlen(k_platform) + 1;
2060         if (STACK_GROWS_DOWN) {
2061             sp -= (len + n - 1) & ~(n - 1);
2062             u_platform = sp;
2063             /* FIXME - check return value of memcpy_to_target() for failure */
2064             memcpy_to_target(sp, k_platform, len);
2065         } else {
2066             memcpy_to_target(sp, k_platform, len);
2067             u_platform = sp;
2068             sp += len + 1;
2069         }
2070     }
2071 
2072     /* Provide 16 byte alignment for the PRNG, and basic alignment for
2073      * the argv and envp pointers.
2074      */
2075     if (STACK_GROWS_DOWN) {
2076         sp = QEMU_ALIGN_DOWN(sp, 16);
2077     } else {
2078         sp = QEMU_ALIGN_UP(sp, 16);
2079     }
2080 
2081     /*
2082      * Generate 16 random bytes for userspace PRNG seeding.
2083      */
2084     qemu_guest_getrandom_nofail(k_rand_bytes, sizeof(k_rand_bytes));
2085     if (STACK_GROWS_DOWN) {
2086         sp -= 16;
2087         u_rand_bytes = sp;
2088         /* FIXME - check return value of memcpy_to_target() for failure */
2089         memcpy_to_target(sp, k_rand_bytes, 16);
2090     } else {
2091         memcpy_to_target(sp, k_rand_bytes, 16);
2092         u_rand_bytes = sp;
2093         sp += 16;
2094     }
2095 
2096     size = (DLINFO_ITEMS + 1) * 2;
2097     if (k_platform)
2098         size += 2;
2099 #ifdef DLINFO_ARCH_ITEMS
2100     size += DLINFO_ARCH_ITEMS * 2;
2101 #endif
2102 #ifdef ELF_HWCAP2
2103     size += 2;
2104 #endif
2105     info->auxv_len = size * n;
2106 
2107     size += envc + argc + 2;
2108     size += 1;  /* argc itself */
2109     size *= n;
2110 
2111     /* Allocate space and finalize stack alignment for entry now.  */
2112     if (STACK_GROWS_DOWN) {
2113         u_argc = QEMU_ALIGN_DOWN(sp - size, STACK_ALIGNMENT);
2114         sp = u_argc;
2115     } else {
2116         u_argc = sp;
2117         sp = QEMU_ALIGN_UP(sp + size, STACK_ALIGNMENT);
2118     }
2119 
2120     u_argv = u_argc + n;
2121     u_envp = u_argv + (argc + 1) * n;
2122     u_auxv = u_envp + (envc + 1) * n;
2123     info->saved_auxv = u_auxv;
2124     info->arg_start = u_argv;
2125     info->arg_end = u_argv + argc * n;
2126 
2127     /* This is correct because Linux defines
2128      * elf_addr_t as Elf32_Off / Elf64_Off
2129      */
2130 #define NEW_AUX_ENT(id, val) do {               \
2131         put_user_ual(id, u_auxv);  u_auxv += n; \
2132         put_user_ual(val, u_auxv); u_auxv += n; \
2133     } while(0)
2134 
2135 #ifdef ARCH_DLINFO
2136     /*
2137      * ARCH_DLINFO must come first so platform specific code can enforce
2138      * special alignment requirements on the AUXV if necessary (eg. PPC).
2139      */
2140     ARCH_DLINFO;
2141 #endif
2142     /* There must be exactly DLINFO_ITEMS entries here, or the assert
2143      * on info->auxv_len will trigger.
2144      */
2145     NEW_AUX_ENT(AT_PHDR, (abi_ulong)(info->load_addr + exec->e_phoff));
2146     NEW_AUX_ENT(AT_PHENT, (abi_ulong)(sizeof (struct elf_phdr)));
2147     NEW_AUX_ENT(AT_PHNUM, (abi_ulong)(exec->e_phnum));
2148     if ((info->alignment & ~qemu_host_page_mask) != 0) {
2149         /* Target doesn't support host page size alignment */
2150         NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(TARGET_PAGE_SIZE));
2151     } else {
2152         NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(MAX(TARGET_PAGE_SIZE,
2153                                                qemu_host_page_size)));
2154     }
2155     NEW_AUX_ENT(AT_BASE, (abi_ulong)(interp_info ? interp_info->load_addr : 0));
2156     NEW_AUX_ENT(AT_FLAGS, (abi_ulong)0);
2157     NEW_AUX_ENT(AT_ENTRY, info->entry);
2158     NEW_AUX_ENT(AT_UID, (abi_ulong) getuid());
2159     NEW_AUX_ENT(AT_EUID, (abi_ulong) geteuid());
2160     NEW_AUX_ENT(AT_GID, (abi_ulong) getgid());
2161     NEW_AUX_ENT(AT_EGID, (abi_ulong) getegid());
2162     NEW_AUX_ENT(AT_HWCAP, (abi_ulong) ELF_HWCAP);
2163     NEW_AUX_ENT(AT_CLKTCK, (abi_ulong) sysconf(_SC_CLK_TCK));
2164     NEW_AUX_ENT(AT_RANDOM, (abi_ulong) u_rand_bytes);
2165     NEW_AUX_ENT(AT_SECURE, (abi_ulong) qemu_getauxval(AT_SECURE));
2166     NEW_AUX_ENT(AT_EXECFN, info->file_string);
2167 
2168 #ifdef ELF_HWCAP2
2169     NEW_AUX_ENT(AT_HWCAP2, (abi_ulong) ELF_HWCAP2);
2170 #endif
2171 
2172     if (u_platform) {
2173         NEW_AUX_ENT(AT_PLATFORM, u_platform);
2174     }
2175     NEW_AUX_ENT (AT_NULL, 0);
2176 #undef NEW_AUX_ENT
2177 
2178     /* Check that our initial calculation of the auxv length matches how much
2179      * we actually put into it.
2180      */
2181     assert(info->auxv_len == u_auxv - info->saved_auxv);
2182 
2183     put_user_ual(argc, u_argc);
2184 
2185     p = info->arg_strings;
2186     for (i = 0; i < argc; ++i) {
2187         put_user_ual(p, u_argv);
2188         u_argv += n;
2189         p += target_strlen(p) + 1;
2190     }
2191     put_user_ual(0, u_argv);
2192 
2193     p = info->env_strings;
2194     for (i = 0; i < envc; ++i) {
2195         put_user_ual(p, u_envp);
2196         u_envp += n;
2197         p += target_strlen(p) + 1;
2198     }
2199     put_user_ual(0, u_envp);
2200 
2201     return sp;
2202 }
2203 
2204 #if defined(HI_COMMPAGE)
2205 #define LO_COMMPAGE 0
2206 #elif defined(LO_COMMPAGE)
2207 #define HI_COMMPAGE 0
2208 #else
2209 #define HI_COMMPAGE 0
2210 #define LO_COMMPAGE 0
2211 #define init_guest_commpage() true
2212 #endif
2213 
2214 static void pgb_fail_in_use(const char *image_name)
2215 {
2216     error_report("%s: requires virtual address space that is in use "
2217                  "(omit the -B option or choose a different value)",
2218                  image_name);
2219     exit(EXIT_FAILURE);
2220 }
2221 
2222 static void pgb_have_guest_base(const char *image_name, abi_ulong guest_loaddr,
2223                                 abi_ulong guest_hiaddr, long align)
2224 {
2225     const int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE;
2226     void *addr, *test;
2227 
2228     if (!QEMU_IS_ALIGNED(guest_base, align)) {
2229         fprintf(stderr, "Requested guest base %p does not satisfy "
2230                 "host minimum alignment (0x%lx)\n",
2231                 (void *)guest_base, align);
2232         exit(EXIT_FAILURE);
2233     }
2234 
2235     /* Sanity check the guest binary. */
2236     if (reserved_va) {
2237         if (guest_hiaddr > reserved_va) {
2238             error_report("%s: requires more than reserved virtual "
2239                          "address space (0x%" PRIx64 " > 0x%lx)",
2240                          image_name, (uint64_t)guest_hiaddr, reserved_va);
2241             exit(EXIT_FAILURE);
2242         }
2243     } else {
2244 #if HOST_LONG_BITS < TARGET_ABI_BITS
2245         if ((guest_hiaddr - guest_base) > ~(uintptr_t)0) {
2246             error_report("%s: requires more virtual address space "
2247                          "than the host can provide (0x%" PRIx64 ")",
2248                          image_name, (uint64_t)guest_hiaddr - guest_base);
2249             exit(EXIT_FAILURE);
2250         }
2251 #endif
2252     }
2253 
2254     /*
2255      * Expand the allocation to the entire reserved_va.
2256      * Exclude the mmap_min_addr hole.
2257      */
2258     if (reserved_va) {
2259         guest_loaddr = (guest_base >= mmap_min_addr ? 0
2260                         : mmap_min_addr - guest_base);
2261         guest_hiaddr = reserved_va;
2262     }
2263 
2264     /* Reserve the address space for the binary, or reserved_va. */
2265     test = g2h_untagged(guest_loaddr);
2266     addr = mmap(test, guest_hiaddr - guest_loaddr, PROT_NONE, flags, -1, 0);
2267     if (test != addr) {
2268         pgb_fail_in_use(image_name);
2269     }
2270     qemu_log_mask(CPU_LOG_PAGE,
2271                   "%s: base @ %p for " TARGET_ABI_FMT_ld " bytes\n",
2272                   __func__, addr, guest_hiaddr - guest_loaddr);
2273 }
2274 
2275 /**
2276  * pgd_find_hole_fallback: potential mmap address
2277  * @guest_size: size of available space
2278  * @brk: location of break
2279  * @align: memory alignment
2280  *
2281  * This is a fallback method for finding a hole in the host address
2282  * space if we don't have the benefit of being able to access
2283  * /proc/self/map. It can potentially take a very long time as we can
2284  * only dumbly iterate up the host address space seeing if the
2285  * allocation would work.
2286  */
2287 static uintptr_t pgd_find_hole_fallback(uintptr_t guest_size, uintptr_t brk,
2288                                         long align, uintptr_t offset)
2289 {
2290     uintptr_t base;
2291 
2292     /* Start (aligned) at the bottom and work our way up */
2293     base = ROUND_UP(mmap_min_addr, align);
2294 
2295     while (true) {
2296         uintptr_t align_start, end;
2297         align_start = ROUND_UP(base, align);
2298         end = align_start + guest_size + offset;
2299 
2300         /* if brk is anywhere in the range give ourselves some room to grow. */
2301         if (align_start <= brk && brk < end) {
2302             base = brk + (16 * MiB);
2303             continue;
2304         } else if (align_start + guest_size < align_start) {
2305             /* we have run out of space */
2306             return -1;
2307         } else {
2308             int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE |
2309                 MAP_FIXED_NOREPLACE;
2310             void * mmap_start = mmap((void *) align_start, guest_size,
2311                                      PROT_NONE, flags, -1, 0);
2312             if (mmap_start != MAP_FAILED) {
2313                 munmap(mmap_start, guest_size);
2314                 if (mmap_start == (void *) align_start) {
2315                     qemu_log_mask(CPU_LOG_PAGE,
2316                                   "%s: base @ %p for %" PRIdPTR" bytes\n",
2317                                   __func__, mmap_start + offset, guest_size);
2318                     return (uintptr_t) mmap_start + offset;
2319                 }
2320             }
2321             base += qemu_host_page_size;
2322         }
2323     }
2324 }
2325 
2326 /* Return value for guest_base, or -1 if no hole found. */
2327 static uintptr_t pgb_find_hole(uintptr_t guest_loaddr, uintptr_t guest_size,
2328                                long align, uintptr_t offset)
2329 {
2330     GSList *maps, *iter;
2331     uintptr_t this_start, this_end, next_start, brk;
2332     intptr_t ret = -1;
2333 
2334     assert(QEMU_IS_ALIGNED(guest_loaddr, align));
2335 
2336     maps = read_self_maps();
2337 
2338     /* Read brk after we've read the maps, which will malloc. */
2339     brk = (uintptr_t)sbrk(0);
2340 
2341     if (!maps) {
2342         return pgd_find_hole_fallback(guest_size, brk, align, offset);
2343     }
2344 
2345     /* The first hole is before the first map entry. */
2346     this_start = mmap_min_addr;
2347 
2348     for (iter = maps; iter;
2349          this_start = next_start, iter = g_slist_next(iter)) {
2350         uintptr_t align_start, hole_size;
2351 
2352         this_end = ((MapInfo *)iter->data)->start;
2353         next_start = ((MapInfo *)iter->data)->end;
2354         align_start = ROUND_UP(this_start + offset, align);
2355 
2356         /* Skip holes that are too small. */
2357         if (align_start >= this_end) {
2358             continue;
2359         }
2360         hole_size = this_end - align_start;
2361         if (hole_size < guest_size) {
2362             continue;
2363         }
2364 
2365         /* If this hole contains brk, give ourselves some room to grow. */
2366         if (this_start <= brk && brk < this_end) {
2367             hole_size -= guest_size;
2368             if (sizeof(uintptr_t) == 8 && hole_size >= 1 * GiB) {
2369                 align_start += 1 * GiB;
2370             } else if (hole_size >= 16 * MiB) {
2371                 align_start += 16 * MiB;
2372             } else {
2373                 align_start = (this_end - guest_size) & -align;
2374                 if (align_start < this_start) {
2375                     continue;
2376                 }
2377             }
2378         }
2379 
2380         /* Record the lowest successful match. */
2381         if (ret < 0) {
2382             ret = align_start;
2383         }
2384         /* If this hole contains the identity map, select it. */
2385         if (align_start <= guest_loaddr &&
2386             guest_loaddr + guest_size <= this_end) {
2387             ret = 0;
2388         }
2389         /* If this hole ends above the identity map, stop looking. */
2390         if (this_end >= guest_loaddr) {
2391             break;
2392         }
2393     }
2394     free_self_maps(maps);
2395 
2396     if (ret != -1) {
2397         qemu_log_mask(CPU_LOG_PAGE, "%s: base @ %" PRIxPTR
2398                       " for %" PRIuPTR " bytes\n",
2399                       __func__, ret, guest_size);
2400     }
2401 
2402     return ret;
2403 }
2404 
2405 static void pgb_static(const char *image_name, abi_ulong orig_loaddr,
2406                        abi_ulong orig_hiaddr, long align)
2407 {
2408     uintptr_t loaddr = orig_loaddr;
2409     uintptr_t hiaddr = orig_hiaddr;
2410     uintptr_t offset = 0;
2411     uintptr_t addr;
2412 
2413     if (hiaddr != orig_hiaddr) {
2414         error_report("%s: requires virtual address space that the "
2415                      "host cannot provide (0x%" PRIx64 ")",
2416                      image_name, (uint64_t)orig_hiaddr);
2417         exit(EXIT_FAILURE);
2418     }
2419 
2420     loaddr &= -align;
2421     if (HI_COMMPAGE) {
2422         /*
2423          * Extend the allocation to include the commpage.
2424          * For a 64-bit host, this is just 4GiB; for a 32-bit host we
2425          * need to ensure there is space bellow the guest_base so we
2426          * can map the commpage in the place needed when the address
2427          * arithmetic wraps around.
2428          */
2429         if (sizeof(uintptr_t) == 8 || loaddr >= 0x80000000u) {
2430             hiaddr = (uintptr_t) 4 << 30;
2431         } else {
2432             offset = -(HI_COMMPAGE & -align);
2433         }
2434     } else if (LO_COMMPAGE != 0) {
2435         loaddr = MIN(loaddr, LO_COMMPAGE & -align);
2436     }
2437 
2438     addr = pgb_find_hole(loaddr, hiaddr - loaddr, align, offset);
2439     if (addr == -1) {
2440         /*
2441          * If HI_COMMPAGE, there *might* be a non-consecutive allocation
2442          * that can satisfy both.  But as the normal arm32 link base address
2443          * is ~32k, and we extend down to include the commpage, making the
2444          * overhead only ~96k, this is unlikely.
2445          */
2446         error_report("%s: Unable to allocate %#zx bytes of "
2447                      "virtual address space", image_name,
2448                      (size_t)(hiaddr - loaddr));
2449         exit(EXIT_FAILURE);
2450     }
2451 
2452     guest_base = addr;
2453 
2454     qemu_log_mask(CPU_LOG_PAGE, "%s: base @ %"PRIxPTR" for %" PRIuPTR" bytes\n",
2455                   __func__, addr, hiaddr - loaddr);
2456 }
2457 
2458 static void pgb_dynamic(const char *image_name, long align)
2459 {
2460     /*
2461      * The executable is dynamic and does not require a fixed address.
2462      * All we need is a commpage that satisfies align.
2463      * If we do not need a commpage, leave guest_base == 0.
2464      */
2465     if (HI_COMMPAGE) {
2466         uintptr_t addr, commpage;
2467 
2468         /* 64-bit hosts should have used reserved_va. */
2469         assert(sizeof(uintptr_t) == 4);
2470 
2471         /*
2472          * By putting the commpage at the first hole, that puts guest_base
2473          * just above that, and maximises the positive guest addresses.
2474          */
2475         commpage = HI_COMMPAGE & -align;
2476         addr = pgb_find_hole(commpage, -commpage, align, 0);
2477         assert(addr != -1);
2478         guest_base = addr;
2479     }
2480 }
2481 
2482 static void pgb_reserved_va(const char *image_name, abi_ulong guest_loaddr,
2483                             abi_ulong guest_hiaddr, long align)
2484 {
2485     int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE;
2486     void *addr, *test;
2487 
2488     if (guest_hiaddr > reserved_va) {
2489         error_report("%s: requires more than reserved virtual "
2490                      "address space (0x%" PRIx64 " > 0x%lx)",
2491                      image_name, (uint64_t)guest_hiaddr, reserved_va);
2492         exit(EXIT_FAILURE);
2493     }
2494 
2495     /* Widen the "image" to the entire reserved address space. */
2496     pgb_static(image_name, 0, reserved_va, align);
2497 
2498     /* osdep.h defines this as 0 if it's missing */
2499     flags |= MAP_FIXED_NOREPLACE;
2500 
2501     /* Reserve the memory on the host. */
2502     assert(guest_base != 0);
2503     test = g2h_untagged(0);
2504     addr = mmap(test, reserved_va, PROT_NONE, flags, -1, 0);
2505     if (addr == MAP_FAILED || addr != test) {
2506         error_report("Unable to reserve 0x%lx bytes of virtual address "
2507                      "space at %p (%s) for use as guest address space (check your"
2508                      "virtual memory ulimit setting, min_mmap_addr or reserve less "
2509                      "using -R option)", reserved_va, test, strerror(errno));
2510         exit(EXIT_FAILURE);
2511     }
2512 
2513     qemu_log_mask(CPU_LOG_PAGE, "%s: base @ %p for %lu bytes\n",
2514                   __func__, addr, reserved_va);
2515 }
2516 
2517 void probe_guest_base(const char *image_name, abi_ulong guest_loaddr,
2518                       abi_ulong guest_hiaddr)
2519 {
2520     /* In order to use host shmat, we must be able to honor SHMLBA.  */
2521     uintptr_t align = MAX(SHMLBA, qemu_host_page_size);
2522 
2523     if (have_guest_base) {
2524         pgb_have_guest_base(image_name, guest_loaddr, guest_hiaddr, align);
2525     } else if (reserved_va) {
2526         pgb_reserved_va(image_name, guest_loaddr, guest_hiaddr, align);
2527     } else if (guest_loaddr) {
2528         pgb_static(image_name, guest_loaddr, guest_hiaddr, align);
2529     } else {
2530         pgb_dynamic(image_name, align);
2531     }
2532 
2533     /* Reserve and initialize the commpage. */
2534     if (!init_guest_commpage()) {
2535         /*
2536          * With have_guest_base, the user has selected the address and
2537          * we are trying to work with that.  Otherwise, we have selected
2538          * free space and init_guest_commpage must succeeded.
2539          */
2540         assert(have_guest_base);
2541         pgb_fail_in_use(image_name);
2542     }
2543 
2544     assert(QEMU_IS_ALIGNED(guest_base, align));
2545     qemu_log_mask(CPU_LOG_PAGE, "Locating guest address space "
2546                   "@ 0x%" PRIx64 "\n", (uint64_t)guest_base);
2547 }
2548 
2549 enum {
2550     /* The string "GNU\0" as a magic number. */
2551     GNU0_MAGIC = const_le32('G' | 'N' << 8 | 'U' << 16),
2552     NOTE_DATA_SZ = 1 * KiB,
2553     NOTE_NAME_SZ = 4,
2554     ELF_GNU_PROPERTY_ALIGN = ELF_CLASS == ELFCLASS32 ? 4 : 8,
2555 };
2556 
2557 /*
2558  * Process a single gnu_property entry.
2559  * Return false for error.
2560  */
2561 static bool parse_elf_property(const uint32_t *data, int *off, int datasz,
2562                                struct image_info *info, bool have_prev_type,
2563                                uint32_t *prev_type, Error **errp)
2564 {
2565     uint32_t pr_type, pr_datasz, step;
2566 
2567     if (*off > datasz || !QEMU_IS_ALIGNED(*off, ELF_GNU_PROPERTY_ALIGN)) {
2568         goto error_data;
2569     }
2570     datasz -= *off;
2571     data += *off / sizeof(uint32_t);
2572 
2573     if (datasz < 2 * sizeof(uint32_t)) {
2574         goto error_data;
2575     }
2576     pr_type = data[0];
2577     pr_datasz = data[1];
2578     data += 2;
2579     datasz -= 2 * sizeof(uint32_t);
2580     step = ROUND_UP(pr_datasz, ELF_GNU_PROPERTY_ALIGN);
2581     if (step > datasz) {
2582         goto error_data;
2583     }
2584 
2585     /* Properties are supposed to be unique and sorted on pr_type. */
2586     if (have_prev_type && pr_type <= *prev_type) {
2587         if (pr_type == *prev_type) {
2588             error_setg(errp, "Duplicate property in PT_GNU_PROPERTY");
2589         } else {
2590             error_setg(errp, "Unsorted property in PT_GNU_PROPERTY");
2591         }
2592         return false;
2593     }
2594     *prev_type = pr_type;
2595 
2596     if (!arch_parse_elf_property(pr_type, pr_datasz, data, info, errp)) {
2597         return false;
2598     }
2599 
2600     *off += 2 * sizeof(uint32_t) + step;
2601     return true;
2602 
2603  error_data:
2604     error_setg(errp, "Ill-formed property in PT_GNU_PROPERTY");
2605     return false;
2606 }
2607 
2608 /* Process NT_GNU_PROPERTY_TYPE_0. */
2609 static bool parse_elf_properties(int image_fd,
2610                                  struct image_info *info,
2611                                  const struct elf_phdr *phdr,
2612                                  char bprm_buf[BPRM_BUF_SIZE],
2613                                  Error **errp)
2614 {
2615     union {
2616         struct elf_note nhdr;
2617         uint32_t data[NOTE_DATA_SZ / sizeof(uint32_t)];
2618     } note;
2619 
2620     int n, off, datasz;
2621     bool have_prev_type;
2622     uint32_t prev_type;
2623 
2624     /* Unless the arch requires properties, ignore them. */
2625     if (!ARCH_USE_GNU_PROPERTY) {
2626         return true;
2627     }
2628 
2629     /* If the properties are crazy large, that's too bad. */
2630     n = phdr->p_filesz;
2631     if (n > sizeof(note)) {
2632         error_setg(errp, "PT_GNU_PROPERTY too large");
2633         return false;
2634     }
2635     if (n < sizeof(note.nhdr)) {
2636         error_setg(errp, "PT_GNU_PROPERTY too small");
2637         return false;
2638     }
2639 
2640     if (phdr->p_offset + n <= BPRM_BUF_SIZE) {
2641         memcpy(&note, bprm_buf + phdr->p_offset, n);
2642     } else {
2643         ssize_t len = pread(image_fd, &note, n, phdr->p_offset);
2644         if (len != n) {
2645             error_setg_errno(errp, errno, "Error reading file header");
2646             return false;
2647         }
2648     }
2649 
2650     /*
2651      * The contents of a valid PT_GNU_PROPERTY is a sequence
2652      * of uint32_t -- swap them all now.
2653      */
2654 #ifdef BSWAP_NEEDED
2655     for (int i = 0; i < n / 4; i++) {
2656         bswap32s(note.data + i);
2657     }
2658 #endif
2659 
2660     /*
2661      * Note that nhdr is 3 words, and that the "name" described by namesz
2662      * immediately follows nhdr and is thus at the 4th word.  Further, all
2663      * of the inputs to the kernel's round_up are multiples of 4.
2664      */
2665     if (note.nhdr.n_type != NT_GNU_PROPERTY_TYPE_0 ||
2666         note.nhdr.n_namesz != NOTE_NAME_SZ ||
2667         note.data[3] != GNU0_MAGIC) {
2668         error_setg(errp, "Invalid note in PT_GNU_PROPERTY");
2669         return false;
2670     }
2671     off = sizeof(note.nhdr) + NOTE_NAME_SZ;
2672 
2673     datasz = note.nhdr.n_descsz + off;
2674     if (datasz > n) {
2675         error_setg(errp, "Invalid note size in PT_GNU_PROPERTY");
2676         return false;
2677     }
2678 
2679     have_prev_type = false;
2680     prev_type = 0;
2681     while (1) {
2682         if (off == datasz) {
2683             return true;  /* end, exit ok */
2684         }
2685         if (!parse_elf_property(note.data, &off, datasz, info,
2686                                 have_prev_type, &prev_type, errp)) {
2687             return false;
2688         }
2689         have_prev_type = true;
2690     }
2691 }
2692 
2693 /* Load an ELF image into the address space.
2694 
2695    IMAGE_NAME is the filename of the image, to use in error messages.
2696    IMAGE_FD is the open file descriptor for the image.
2697 
2698    BPRM_BUF is a copy of the beginning of the file; this of course
2699    contains the elf file header at offset 0.  It is assumed that this
2700    buffer is sufficiently aligned to present no problems to the host
2701    in accessing data at aligned offsets within the buffer.
2702 
2703    On return: INFO values will be filled in, as necessary or available.  */
2704 
2705 static void load_elf_image(const char *image_name, int image_fd,
2706                            struct image_info *info, char **pinterp_name,
2707                            char bprm_buf[BPRM_BUF_SIZE])
2708 {
2709     struct elfhdr *ehdr = (struct elfhdr *)bprm_buf;
2710     struct elf_phdr *phdr;
2711     abi_ulong load_addr, load_bias, loaddr, hiaddr, error;
2712     int i, retval, prot_exec;
2713     Error *err = NULL;
2714 
2715     /* First of all, some simple consistency checks */
2716     if (!elf_check_ident(ehdr)) {
2717         error_setg(&err, "Invalid ELF image for this architecture");
2718         goto exit_errmsg;
2719     }
2720     bswap_ehdr(ehdr);
2721     if (!elf_check_ehdr(ehdr)) {
2722         error_setg(&err, "Invalid ELF image for this architecture");
2723         goto exit_errmsg;
2724     }
2725 
2726     i = ehdr->e_phnum * sizeof(struct elf_phdr);
2727     if (ehdr->e_phoff + i <= BPRM_BUF_SIZE) {
2728         phdr = (struct elf_phdr *)(bprm_buf + ehdr->e_phoff);
2729     } else {
2730         phdr = (struct elf_phdr *) alloca(i);
2731         retval = pread(image_fd, phdr, i, ehdr->e_phoff);
2732         if (retval != i) {
2733             goto exit_read;
2734         }
2735     }
2736     bswap_phdr(phdr, ehdr->e_phnum);
2737 
2738     info->nsegs = 0;
2739     info->pt_dynamic_addr = 0;
2740 
2741     mmap_lock();
2742 
2743     /*
2744      * Find the maximum size of the image and allocate an appropriate
2745      * amount of memory to handle that.  Locate the interpreter, if any.
2746      */
2747     loaddr = -1, hiaddr = 0;
2748     info->alignment = 0;
2749     for (i = 0; i < ehdr->e_phnum; ++i) {
2750         struct elf_phdr *eppnt = phdr + i;
2751         if (eppnt->p_type == PT_LOAD) {
2752             abi_ulong a = eppnt->p_vaddr - eppnt->p_offset;
2753             if (a < loaddr) {
2754                 loaddr = a;
2755             }
2756             a = eppnt->p_vaddr + eppnt->p_memsz;
2757             if (a > hiaddr) {
2758                 hiaddr = a;
2759             }
2760             ++info->nsegs;
2761             info->alignment |= eppnt->p_align;
2762         } else if (eppnt->p_type == PT_INTERP && pinterp_name) {
2763             g_autofree char *interp_name = NULL;
2764 
2765             if (*pinterp_name) {
2766                 error_setg(&err, "Multiple PT_INTERP entries");
2767                 goto exit_errmsg;
2768             }
2769 
2770             interp_name = g_malloc(eppnt->p_filesz);
2771 
2772             if (eppnt->p_offset + eppnt->p_filesz <= BPRM_BUF_SIZE) {
2773                 memcpy(interp_name, bprm_buf + eppnt->p_offset,
2774                        eppnt->p_filesz);
2775             } else {
2776                 retval = pread(image_fd, interp_name, eppnt->p_filesz,
2777                                eppnt->p_offset);
2778                 if (retval != eppnt->p_filesz) {
2779                     goto exit_read;
2780                 }
2781             }
2782             if (interp_name[eppnt->p_filesz - 1] != 0) {
2783                 error_setg(&err, "Invalid PT_INTERP entry");
2784                 goto exit_errmsg;
2785             }
2786             *pinterp_name = g_steal_pointer(&interp_name);
2787         } else if (eppnt->p_type == PT_GNU_PROPERTY) {
2788             if (!parse_elf_properties(image_fd, info, eppnt, bprm_buf, &err)) {
2789                 goto exit_errmsg;
2790             }
2791         }
2792     }
2793 
2794     if (pinterp_name != NULL) {
2795         /*
2796          * This is the main executable.
2797          *
2798          * Reserve extra space for brk.
2799          * We hold on to this space while placing the interpreter
2800          * and the stack, lest they be placed immediately after
2801          * the data segment and block allocation from the brk.
2802          *
2803          * 16MB is chosen as "large enough" without being so large as
2804          * to allow the result to not fit with a 32-bit guest on a
2805          * 32-bit host. However some 64 bit guests (e.g. s390x)
2806          * attempt to place their heap further ahead and currently
2807          * nothing stops them smashing into QEMUs address space.
2808          */
2809 #if TARGET_LONG_BITS == 64
2810         info->reserve_brk = 32 * MiB;
2811 #else
2812         info->reserve_brk = 16 * MiB;
2813 #endif
2814         hiaddr += info->reserve_brk;
2815 
2816         if (ehdr->e_type == ET_EXEC) {
2817             /*
2818              * Make sure that the low address does not conflict with
2819              * MMAP_MIN_ADDR or the QEMU application itself.
2820              */
2821             probe_guest_base(image_name, loaddr, hiaddr);
2822         } else {
2823             /*
2824              * The binary is dynamic, but we still need to
2825              * select guest_base.  In this case we pass a size.
2826              */
2827             probe_guest_base(image_name, 0, hiaddr - loaddr);
2828         }
2829     }
2830 
2831     /*
2832      * Reserve address space for all of this.
2833      *
2834      * In the case of ET_EXEC, we supply MAP_FIXED so that we get
2835      * exactly the address range that is required.
2836      *
2837      * Otherwise this is ET_DYN, and we are searching for a location
2838      * that can hold the memory space required.  If the image is
2839      * pre-linked, LOADDR will be non-zero, and the kernel should
2840      * honor that address if it happens to be free.
2841      *
2842      * In both cases, we will overwrite pages in this range with mappings
2843      * from the executable.
2844      */
2845     load_addr = target_mmap(loaddr, hiaddr - loaddr, PROT_NONE,
2846                             MAP_PRIVATE | MAP_ANON | MAP_NORESERVE |
2847                             (ehdr->e_type == ET_EXEC ? MAP_FIXED : 0),
2848                             -1, 0);
2849     if (load_addr == -1) {
2850         goto exit_mmap;
2851     }
2852     load_bias = load_addr - loaddr;
2853 
2854     if (elf_is_fdpic(ehdr)) {
2855         struct elf32_fdpic_loadseg *loadsegs = info->loadsegs =
2856             g_malloc(sizeof(*loadsegs) * info->nsegs);
2857 
2858         for (i = 0; i < ehdr->e_phnum; ++i) {
2859             switch (phdr[i].p_type) {
2860             case PT_DYNAMIC:
2861                 info->pt_dynamic_addr = phdr[i].p_vaddr + load_bias;
2862                 break;
2863             case PT_LOAD:
2864                 loadsegs->addr = phdr[i].p_vaddr + load_bias;
2865                 loadsegs->p_vaddr = phdr[i].p_vaddr;
2866                 loadsegs->p_memsz = phdr[i].p_memsz;
2867                 ++loadsegs;
2868                 break;
2869             }
2870         }
2871     }
2872 
2873     info->load_bias = load_bias;
2874     info->code_offset = load_bias;
2875     info->data_offset = load_bias;
2876     info->load_addr = load_addr;
2877     info->entry = ehdr->e_entry + load_bias;
2878     info->start_code = -1;
2879     info->end_code = 0;
2880     info->start_data = -1;
2881     info->end_data = 0;
2882     info->brk = 0;
2883     info->elf_flags = ehdr->e_flags;
2884 
2885     prot_exec = PROT_EXEC;
2886 #ifdef TARGET_AARCH64
2887     /*
2888      * If the BTI feature is present, this indicates that the executable
2889      * pages of the startup binary should be mapped with PROT_BTI, so that
2890      * branch targets are enforced.
2891      *
2892      * The startup binary is either the interpreter or the static executable.
2893      * The interpreter is responsible for all pages of a dynamic executable.
2894      *
2895      * Elf notes are backward compatible to older cpus.
2896      * Do not enable BTI unless it is supported.
2897      */
2898     if ((info->note_flags & GNU_PROPERTY_AARCH64_FEATURE_1_BTI)
2899         && (pinterp_name == NULL || *pinterp_name == 0)
2900         && cpu_isar_feature(aa64_bti, ARM_CPU(thread_cpu))) {
2901         prot_exec |= TARGET_PROT_BTI;
2902     }
2903 #endif
2904 
2905     for (i = 0; i < ehdr->e_phnum; i++) {
2906         struct elf_phdr *eppnt = phdr + i;
2907         if (eppnt->p_type == PT_LOAD) {
2908             abi_ulong vaddr, vaddr_po, vaddr_ps, vaddr_ef, vaddr_em, vaddr_len;
2909             int elf_prot = 0;
2910 
2911             if (eppnt->p_flags & PF_R) {
2912                 elf_prot |= PROT_READ;
2913             }
2914             if (eppnt->p_flags & PF_W) {
2915                 elf_prot |= PROT_WRITE;
2916             }
2917             if (eppnt->p_flags & PF_X) {
2918                 elf_prot |= prot_exec;
2919             }
2920 
2921             vaddr = load_bias + eppnt->p_vaddr;
2922             vaddr_po = TARGET_ELF_PAGEOFFSET(vaddr);
2923             vaddr_ps = TARGET_ELF_PAGESTART(vaddr);
2924 
2925             vaddr_ef = vaddr + eppnt->p_filesz;
2926             vaddr_em = vaddr + eppnt->p_memsz;
2927 
2928             /*
2929              * Some segments may be completely empty, with a non-zero p_memsz
2930              * but no backing file segment.
2931              */
2932             if (eppnt->p_filesz != 0) {
2933                 vaddr_len = TARGET_ELF_PAGELENGTH(eppnt->p_filesz + vaddr_po);
2934                 error = target_mmap(vaddr_ps, vaddr_len, elf_prot,
2935                                     MAP_PRIVATE | MAP_FIXED,
2936                                     image_fd, eppnt->p_offset - vaddr_po);
2937 
2938                 if (error == -1) {
2939                     goto exit_mmap;
2940                 }
2941 
2942                 /*
2943                  * If the load segment requests extra zeros (e.g. bss), map it.
2944                  */
2945                 if (eppnt->p_filesz < eppnt->p_memsz) {
2946                     zero_bss(vaddr_ef, vaddr_em, elf_prot);
2947                 }
2948             } else if (eppnt->p_memsz != 0) {
2949                 vaddr_len = TARGET_ELF_PAGELENGTH(eppnt->p_memsz + vaddr_po);
2950                 error = target_mmap(vaddr_ps, vaddr_len, elf_prot,
2951                                     MAP_PRIVATE | MAP_FIXED | MAP_ANONYMOUS,
2952                                     -1, 0);
2953 
2954                 if (error == -1) {
2955                     goto exit_mmap;
2956                 }
2957             }
2958 
2959             /* Find the full program boundaries.  */
2960             if (elf_prot & PROT_EXEC) {
2961                 if (vaddr < info->start_code) {
2962                     info->start_code = vaddr;
2963                 }
2964                 if (vaddr_ef > info->end_code) {
2965                     info->end_code = vaddr_ef;
2966                 }
2967             }
2968             if (elf_prot & PROT_WRITE) {
2969                 if (vaddr < info->start_data) {
2970                     info->start_data = vaddr;
2971                 }
2972                 if (vaddr_ef > info->end_data) {
2973                     info->end_data = vaddr_ef;
2974                 }
2975             }
2976             if (vaddr_em > info->brk) {
2977                 info->brk = vaddr_em;
2978             }
2979 #ifdef TARGET_MIPS
2980         } else if (eppnt->p_type == PT_MIPS_ABIFLAGS) {
2981             Mips_elf_abiflags_v0 abiflags;
2982             if (eppnt->p_filesz < sizeof(Mips_elf_abiflags_v0)) {
2983                 error_setg(&err, "Invalid PT_MIPS_ABIFLAGS entry");
2984                 goto exit_errmsg;
2985             }
2986             if (eppnt->p_offset + eppnt->p_filesz <= BPRM_BUF_SIZE) {
2987                 memcpy(&abiflags, bprm_buf + eppnt->p_offset,
2988                        sizeof(Mips_elf_abiflags_v0));
2989             } else {
2990                 retval = pread(image_fd, &abiflags, sizeof(Mips_elf_abiflags_v0),
2991                                eppnt->p_offset);
2992                 if (retval != sizeof(Mips_elf_abiflags_v0)) {
2993                     goto exit_read;
2994                 }
2995             }
2996             bswap_mips_abiflags(&abiflags);
2997             info->fp_abi = abiflags.fp_abi;
2998 #endif
2999         }
3000     }
3001 
3002     if (info->end_data == 0) {
3003         info->start_data = info->end_code;
3004         info->end_data = info->end_code;
3005     }
3006 
3007     if (qemu_log_enabled()) {
3008         load_symbols(ehdr, image_fd, load_bias);
3009     }
3010 
3011     mmap_unlock();
3012 
3013     close(image_fd);
3014     return;
3015 
3016  exit_read:
3017     if (retval >= 0) {
3018         error_setg(&err, "Incomplete read of file header");
3019     } else {
3020         error_setg_errno(&err, errno, "Error reading file header");
3021     }
3022     goto exit_errmsg;
3023  exit_mmap:
3024     error_setg_errno(&err, errno, "Error mapping file");
3025     goto exit_errmsg;
3026  exit_errmsg:
3027     error_reportf_err(err, "%s: ", image_name);
3028     exit(-1);
3029 }
3030 
3031 static void load_elf_interp(const char *filename, struct image_info *info,
3032                             char bprm_buf[BPRM_BUF_SIZE])
3033 {
3034     int fd, retval;
3035     Error *err = NULL;
3036 
3037     fd = open(path(filename), O_RDONLY);
3038     if (fd < 0) {
3039         error_setg_file_open(&err, errno, filename);
3040         error_report_err(err);
3041         exit(-1);
3042     }
3043 
3044     retval = read(fd, bprm_buf, BPRM_BUF_SIZE);
3045     if (retval < 0) {
3046         error_setg_errno(&err, errno, "Error reading file header");
3047         error_reportf_err(err, "%s: ", filename);
3048         exit(-1);
3049     }
3050 
3051     if (retval < BPRM_BUF_SIZE) {
3052         memset(bprm_buf + retval, 0, BPRM_BUF_SIZE - retval);
3053     }
3054 
3055     load_elf_image(filename, fd, info, NULL, bprm_buf);
3056 }
3057 
3058 static int symfind(const void *s0, const void *s1)
3059 {
3060     target_ulong addr = *(target_ulong *)s0;
3061     struct elf_sym *sym = (struct elf_sym *)s1;
3062     int result = 0;
3063     if (addr < sym->st_value) {
3064         result = -1;
3065     } else if (addr >= sym->st_value + sym->st_size) {
3066         result = 1;
3067     }
3068     return result;
3069 }
3070 
3071 static const char *lookup_symbolxx(struct syminfo *s, target_ulong orig_addr)
3072 {
3073 #if ELF_CLASS == ELFCLASS32
3074     struct elf_sym *syms = s->disas_symtab.elf32;
3075 #else
3076     struct elf_sym *syms = s->disas_symtab.elf64;
3077 #endif
3078 
3079     // binary search
3080     struct elf_sym *sym;
3081 
3082     sym = bsearch(&orig_addr, syms, s->disas_num_syms, sizeof(*syms), symfind);
3083     if (sym != NULL) {
3084         return s->disas_strtab + sym->st_name;
3085     }
3086 
3087     return "";
3088 }
3089 
3090 /* FIXME: This should use elf_ops.h  */
3091 static int symcmp(const void *s0, const void *s1)
3092 {
3093     struct elf_sym *sym0 = (struct elf_sym *)s0;
3094     struct elf_sym *sym1 = (struct elf_sym *)s1;
3095     return (sym0->st_value < sym1->st_value)
3096         ? -1
3097         : ((sym0->st_value > sym1->st_value) ? 1 : 0);
3098 }
3099 
3100 /* Best attempt to load symbols from this ELF object. */
3101 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias)
3102 {
3103     int i, shnum, nsyms, sym_idx = 0, str_idx = 0;
3104     uint64_t segsz;
3105     struct elf_shdr *shdr;
3106     char *strings = NULL;
3107     struct syminfo *s = NULL;
3108     struct elf_sym *new_syms, *syms = NULL;
3109 
3110     shnum = hdr->e_shnum;
3111     i = shnum * sizeof(struct elf_shdr);
3112     shdr = (struct elf_shdr *)alloca(i);
3113     if (pread(fd, shdr, i, hdr->e_shoff) != i) {
3114         return;
3115     }
3116 
3117     bswap_shdr(shdr, shnum);
3118     for (i = 0; i < shnum; ++i) {
3119         if (shdr[i].sh_type == SHT_SYMTAB) {
3120             sym_idx = i;
3121             str_idx = shdr[i].sh_link;
3122             goto found;
3123         }
3124     }
3125 
3126     /* There will be no symbol table if the file was stripped.  */
3127     return;
3128 
3129  found:
3130     /* Now know where the strtab and symtab are.  Snarf them.  */
3131     s = g_try_new(struct syminfo, 1);
3132     if (!s) {
3133         goto give_up;
3134     }
3135 
3136     segsz = shdr[str_idx].sh_size;
3137     s->disas_strtab = strings = g_try_malloc(segsz);
3138     if (!strings ||
3139         pread(fd, strings, segsz, shdr[str_idx].sh_offset) != segsz) {
3140         goto give_up;
3141     }
3142 
3143     segsz = shdr[sym_idx].sh_size;
3144     syms = g_try_malloc(segsz);
3145     if (!syms || pread(fd, syms, segsz, shdr[sym_idx].sh_offset) != segsz) {
3146         goto give_up;
3147     }
3148 
3149     if (segsz / sizeof(struct elf_sym) > INT_MAX) {
3150         /* Implausibly large symbol table: give up rather than ploughing
3151          * on with the number of symbols calculation overflowing
3152          */
3153         goto give_up;
3154     }
3155     nsyms = segsz / sizeof(struct elf_sym);
3156     for (i = 0; i < nsyms; ) {
3157         bswap_sym(syms + i);
3158         /* Throw away entries which we do not need.  */
3159         if (syms[i].st_shndx == SHN_UNDEF
3160             || syms[i].st_shndx >= SHN_LORESERVE
3161             || ELF_ST_TYPE(syms[i].st_info) != STT_FUNC) {
3162             if (i < --nsyms) {
3163                 syms[i] = syms[nsyms];
3164             }
3165         } else {
3166 #if defined(TARGET_ARM) || defined (TARGET_MIPS)
3167             /* The bottom address bit marks a Thumb or MIPS16 symbol.  */
3168             syms[i].st_value &= ~(target_ulong)1;
3169 #endif
3170             syms[i].st_value += load_bias;
3171             i++;
3172         }
3173     }
3174 
3175     /* No "useful" symbol.  */
3176     if (nsyms == 0) {
3177         goto give_up;
3178     }
3179 
3180     /* Attempt to free the storage associated with the local symbols
3181        that we threw away.  Whether or not this has any effect on the
3182        memory allocation depends on the malloc implementation and how
3183        many symbols we managed to discard.  */
3184     new_syms = g_try_renew(struct elf_sym, syms, nsyms);
3185     if (new_syms == NULL) {
3186         goto give_up;
3187     }
3188     syms = new_syms;
3189 
3190     qsort(syms, nsyms, sizeof(*syms), symcmp);
3191 
3192     s->disas_num_syms = nsyms;
3193 #if ELF_CLASS == ELFCLASS32
3194     s->disas_symtab.elf32 = syms;
3195 #else
3196     s->disas_symtab.elf64 = syms;
3197 #endif
3198     s->lookup_symbol = lookup_symbolxx;
3199     s->next = syminfos;
3200     syminfos = s;
3201 
3202     return;
3203 
3204 give_up:
3205     g_free(s);
3206     g_free(strings);
3207     g_free(syms);
3208 }
3209 
3210 uint32_t get_elf_eflags(int fd)
3211 {
3212     struct elfhdr ehdr;
3213     off_t offset;
3214     int ret;
3215 
3216     /* Read ELF header */
3217     offset = lseek(fd, 0, SEEK_SET);
3218     if (offset == (off_t) -1) {
3219         return 0;
3220     }
3221     ret = read(fd, &ehdr, sizeof(ehdr));
3222     if (ret < sizeof(ehdr)) {
3223         return 0;
3224     }
3225     offset = lseek(fd, offset, SEEK_SET);
3226     if (offset == (off_t) -1) {
3227         return 0;
3228     }
3229 
3230     /* Check ELF signature */
3231     if (!elf_check_ident(&ehdr)) {
3232         return 0;
3233     }
3234 
3235     /* check header */
3236     bswap_ehdr(&ehdr);
3237     if (!elf_check_ehdr(&ehdr)) {
3238         return 0;
3239     }
3240 
3241     /* return architecture id */
3242     return ehdr.e_flags;
3243 }
3244 
3245 int load_elf_binary(struct linux_binprm *bprm, struct image_info *info)
3246 {
3247     struct image_info interp_info;
3248     struct elfhdr elf_ex;
3249     char *elf_interpreter = NULL;
3250     char *scratch;
3251 
3252     memset(&interp_info, 0, sizeof(interp_info));
3253 #ifdef TARGET_MIPS
3254     interp_info.fp_abi = MIPS_ABI_FP_UNKNOWN;
3255 #endif
3256 
3257     info->start_mmap = (abi_ulong)ELF_START_MMAP;
3258 
3259     load_elf_image(bprm->filename, bprm->fd, info,
3260                    &elf_interpreter, bprm->buf);
3261 
3262     /* ??? We need a copy of the elf header for passing to create_elf_tables.
3263        If we do nothing, we'll have overwritten this when we re-use bprm->buf
3264        when we load the interpreter.  */
3265     elf_ex = *(struct elfhdr *)bprm->buf;
3266 
3267     /* Do this so that we can load the interpreter, if need be.  We will
3268        change some of these later */
3269     bprm->p = setup_arg_pages(bprm, info);
3270 
3271     scratch = g_new0(char, TARGET_PAGE_SIZE);
3272     if (STACK_GROWS_DOWN) {
3273         bprm->p = copy_elf_strings(1, &bprm->filename, scratch,
3274                                    bprm->p, info->stack_limit);
3275         info->file_string = bprm->p;
3276         bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch,
3277                                    bprm->p, info->stack_limit);
3278         info->env_strings = bprm->p;
3279         bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch,
3280                                    bprm->p, info->stack_limit);
3281         info->arg_strings = bprm->p;
3282     } else {
3283         info->arg_strings = bprm->p;
3284         bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch,
3285                                    bprm->p, info->stack_limit);
3286         info->env_strings = bprm->p;
3287         bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch,
3288                                    bprm->p, info->stack_limit);
3289         info->file_string = bprm->p;
3290         bprm->p = copy_elf_strings(1, &bprm->filename, scratch,
3291                                    bprm->p, info->stack_limit);
3292     }
3293 
3294     g_free(scratch);
3295 
3296     if (!bprm->p) {
3297         fprintf(stderr, "%s: %s\n", bprm->filename, strerror(E2BIG));
3298         exit(-1);
3299     }
3300 
3301     if (elf_interpreter) {
3302         load_elf_interp(elf_interpreter, &interp_info, bprm->buf);
3303 
3304         /* If the program interpreter is one of these two, then assume
3305            an iBCS2 image.  Otherwise assume a native linux image.  */
3306 
3307         if (strcmp(elf_interpreter, "/usr/lib/libc.so.1") == 0
3308             || strcmp(elf_interpreter, "/usr/lib/ld.so.1") == 0) {
3309             info->personality = PER_SVR4;
3310 
3311             /* Why this, you ask???  Well SVr4 maps page 0 as read-only,
3312                and some applications "depend" upon this behavior.  Since
3313                we do not have the power to recompile these, we emulate
3314                the SVr4 behavior.  Sigh.  */
3315             target_mmap(0, qemu_host_page_size, PROT_READ | PROT_EXEC,
3316                         MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
3317         }
3318 #ifdef TARGET_MIPS
3319         info->interp_fp_abi = interp_info.fp_abi;
3320 #endif
3321     }
3322 
3323     /*
3324      * TODO: load a vdso, which would also contain the signal trampolines.
3325      * Otherwise, allocate a private page to hold them.
3326      */
3327     if (TARGET_ARCH_HAS_SIGTRAMP_PAGE) {
3328         abi_long tramp_page = target_mmap(0, TARGET_PAGE_SIZE,
3329                                           PROT_READ | PROT_WRITE,
3330                                           MAP_PRIVATE | MAP_ANON, -1, 0);
3331         if (tramp_page == -1) {
3332             return -errno;
3333         }
3334 
3335         setup_sigtramp(tramp_page);
3336         target_mprotect(tramp_page, TARGET_PAGE_SIZE, PROT_READ | PROT_EXEC);
3337     }
3338 
3339     bprm->p = create_elf_tables(bprm->p, bprm->argc, bprm->envc, &elf_ex,
3340                                 info, (elf_interpreter ? &interp_info : NULL));
3341     info->start_stack = bprm->p;
3342 
3343     /* If we have an interpreter, set that as the program's entry point.
3344        Copy the load_bias as well, to help PPC64 interpret the entry
3345        point as a function descriptor.  Do this after creating elf tables
3346        so that we copy the original program entry point into the AUXV.  */
3347     if (elf_interpreter) {
3348         info->load_bias = interp_info.load_bias;
3349         info->entry = interp_info.entry;
3350         g_free(elf_interpreter);
3351     }
3352 
3353 #ifdef USE_ELF_CORE_DUMP
3354     bprm->core_dump = &elf_core_dump;
3355 #endif
3356 
3357     /*
3358      * If we reserved extra space for brk, release it now.
3359      * The implementation of do_brk in syscalls.c expects to be able
3360      * to mmap pages in this space.
3361      */
3362     if (info->reserve_brk) {
3363         abi_ulong start_brk = HOST_PAGE_ALIGN(info->brk);
3364         abi_ulong end_brk = HOST_PAGE_ALIGN(info->brk + info->reserve_brk);
3365         target_munmap(start_brk, end_brk - start_brk);
3366     }
3367 
3368     return 0;
3369 }
3370 
3371 #ifdef USE_ELF_CORE_DUMP
3372 /*
3373  * Definitions to generate Intel SVR4-like core files.
3374  * These mostly have the same names as the SVR4 types with "target_elf_"
3375  * tacked on the front to prevent clashes with linux definitions,
3376  * and the typedef forms have been avoided.  This is mostly like
3377  * the SVR4 structure, but more Linuxy, with things that Linux does
3378  * not support and which gdb doesn't really use excluded.
3379  *
3380  * Fields we don't dump (their contents is zero) in linux-user qemu
3381  * are marked with XXX.
3382  *
3383  * Core dump code is copied from linux kernel (fs/binfmt_elf.c).
3384  *
3385  * Porting ELF coredump for target is (quite) simple process.  First you
3386  * define USE_ELF_CORE_DUMP in target ELF code (where init_thread() for
3387  * the target resides):
3388  *
3389  * #define USE_ELF_CORE_DUMP
3390  *
3391  * Next you define type of register set used for dumping.  ELF specification
3392  * says that it needs to be array of elf_greg_t that has size of ELF_NREG.
3393  *
3394  * typedef <target_regtype> target_elf_greg_t;
3395  * #define ELF_NREG <number of registers>
3396  * typedef taret_elf_greg_t target_elf_gregset_t[ELF_NREG];
3397  *
3398  * Last step is to implement target specific function that copies registers
3399  * from given cpu into just specified register set.  Prototype is:
3400  *
3401  * static void elf_core_copy_regs(taret_elf_gregset_t *regs,
3402  *                                const CPUArchState *env);
3403  *
3404  * Parameters:
3405  *     regs - copy register values into here (allocated and zeroed by caller)
3406  *     env - copy registers from here
3407  *
3408  * Example for ARM target is provided in this file.
3409  */
3410 
3411 /* An ELF note in memory */
3412 struct memelfnote {
3413     const char *name;
3414     size_t     namesz;
3415     size_t     namesz_rounded;
3416     int        type;
3417     size_t     datasz;
3418     size_t     datasz_rounded;
3419     void       *data;
3420     size_t     notesz;
3421 };
3422 
3423 struct target_elf_siginfo {
3424     abi_int    si_signo; /* signal number */
3425     abi_int    si_code;  /* extra code */
3426     abi_int    si_errno; /* errno */
3427 };
3428 
3429 struct target_elf_prstatus {
3430     struct target_elf_siginfo pr_info;      /* Info associated with signal */
3431     abi_short          pr_cursig;    /* Current signal */
3432     abi_ulong          pr_sigpend;   /* XXX */
3433     abi_ulong          pr_sighold;   /* XXX */
3434     target_pid_t       pr_pid;
3435     target_pid_t       pr_ppid;
3436     target_pid_t       pr_pgrp;
3437     target_pid_t       pr_sid;
3438     struct target_timeval pr_utime;  /* XXX User time */
3439     struct target_timeval pr_stime;  /* XXX System time */
3440     struct target_timeval pr_cutime; /* XXX Cumulative user time */
3441     struct target_timeval pr_cstime; /* XXX Cumulative system time */
3442     target_elf_gregset_t      pr_reg;       /* GP registers */
3443     abi_int            pr_fpvalid;   /* XXX */
3444 };
3445 
3446 #define ELF_PRARGSZ     (80) /* Number of chars for args */
3447 
3448 struct target_elf_prpsinfo {
3449     char         pr_state;       /* numeric process state */
3450     char         pr_sname;       /* char for pr_state */
3451     char         pr_zomb;        /* zombie */
3452     char         pr_nice;        /* nice val */
3453     abi_ulong    pr_flag;        /* flags */
3454     target_uid_t pr_uid;
3455     target_gid_t pr_gid;
3456     target_pid_t pr_pid, pr_ppid, pr_pgrp, pr_sid;
3457     /* Lots missing */
3458     char    pr_fname[16] QEMU_NONSTRING; /* filename of executable */
3459     char    pr_psargs[ELF_PRARGSZ]; /* initial part of arg list */
3460 };
3461 
3462 /* Here is the structure in which status of each thread is captured. */
3463 struct elf_thread_status {
3464     QTAILQ_ENTRY(elf_thread_status)  ets_link;
3465     struct target_elf_prstatus prstatus;   /* NT_PRSTATUS */
3466 #if 0
3467     elf_fpregset_t fpu;             /* NT_PRFPREG */
3468     struct task_struct *thread;
3469     elf_fpxregset_t xfpu;           /* ELF_CORE_XFPREG_TYPE */
3470 #endif
3471     struct memelfnote notes[1];
3472     int num_notes;
3473 };
3474 
3475 struct elf_note_info {
3476     struct memelfnote   *notes;
3477     struct target_elf_prstatus *prstatus;  /* NT_PRSTATUS */
3478     struct target_elf_prpsinfo *psinfo;    /* NT_PRPSINFO */
3479 
3480     QTAILQ_HEAD(, elf_thread_status) thread_list;
3481 #if 0
3482     /*
3483      * Current version of ELF coredump doesn't support
3484      * dumping fp regs etc.
3485      */
3486     elf_fpregset_t *fpu;
3487     elf_fpxregset_t *xfpu;
3488     int thread_status_size;
3489 #endif
3490     int notes_size;
3491     int numnote;
3492 };
3493 
3494 struct vm_area_struct {
3495     target_ulong   vma_start;  /* start vaddr of memory region */
3496     target_ulong   vma_end;    /* end vaddr of memory region */
3497     abi_ulong      vma_flags;  /* protection etc. flags for the region */
3498     QTAILQ_ENTRY(vm_area_struct) vma_link;
3499 };
3500 
3501 struct mm_struct {
3502     QTAILQ_HEAD(, vm_area_struct) mm_mmap;
3503     int mm_count;           /* number of mappings */
3504 };
3505 
3506 static struct mm_struct *vma_init(void);
3507 static void vma_delete(struct mm_struct *);
3508 static int vma_add_mapping(struct mm_struct *, target_ulong,
3509                            target_ulong, abi_ulong);
3510 static int vma_get_mapping_count(const struct mm_struct *);
3511 static struct vm_area_struct *vma_first(const struct mm_struct *);
3512 static struct vm_area_struct *vma_next(struct vm_area_struct *);
3513 static abi_ulong vma_dump_size(const struct vm_area_struct *);
3514 static int vma_walker(void *priv, target_ulong start, target_ulong end,
3515                       unsigned long flags);
3516 
3517 static void fill_elf_header(struct elfhdr *, int, uint16_t, uint32_t);
3518 static void fill_note(struct memelfnote *, const char *, int,
3519                       unsigned int, void *);
3520 static void fill_prstatus(struct target_elf_prstatus *, const TaskState *, int);
3521 static int fill_psinfo(struct target_elf_prpsinfo *, const TaskState *);
3522 static void fill_auxv_note(struct memelfnote *, const TaskState *);
3523 static void fill_elf_note_phdr(struct elf_phdr *, int, off_t);
3524 static size_t note_size(const struct memelfnote *);
3525 static void free_note_info(struct elf_note_info *);
3526 static int fill_note_info(struct elf_note_info *, long, const CPUArchState *);
3527 static void fill_thread_info(struct elf_note_info *, const CPUArchState *);
3528 
3529 static int dump_write(int, const void *, size_t);
3530 static int write_note(struct memelfnote *, int);
3531 static int write_note_info(struct elf_note_info *, int);
3532 
3533 #ifdef BSWAP_NEEDED
3534 static void bswap_prstatus(struct target_elf_prstatus *prstatus)
3535 {
3536     prstatus->pr_info.si_signo = tswap32(prstatus->pr_info.si_signo);
3537     prstatus->pr_info.si_code = tswap32(prstatus->pr_info.si_code);
3538     prstatus->pr_info.si_errno = tswap32(prstatus->pr_info.si_errno);
3539     prstatus->pr_cursig = tswap16(prstatus->pr_cursig);
3540     prstatus->pr_sigpend = tswapal(prstatus->pr_sigpend);
3541     prstatus->pr_sighold = tswapal(prstatus->pr_sighold);
3542     prstatus->pr_pid = tswap32(prstatus->pr_pid);
3543     prstatus->pr_ppid = tswap32(prstatus->pr_ppid);
3544     prstatus->pr_pgrp = tswap32(prstatus->pr_pgrp);
3545     prstatus->pr_sid = tswap32(prstatus->pr_sid);
3546     /* cpu times are not filled, so we skip them */
3547     /* regs should be in correct format already */
3548     prstatus->pr_fpvalid = tswap32(prstatus->pr_fpvalid);
3549 }
3550 
3551 static void bswap_psinfo(struct target_elf_prpsinfo *psinfo)
3552 {
3553     psinfo->pr_flag = tswapal(psinfo->pr_flag);
3554     psinfo->pr_uid = tswap16(psinfo->pr_uid);
3555     psinfo->pr_gid = tswap16(psinfo->pr_gid);
3556     psinfo->pr_pid = tswap32(psinfo->pr_pid);
3557     psinfo->pr_ppid = tswap32(psinfo->pr_ppid);
3558     psinfo->pr_pgrp = tswap32(psinfo->pr_pgrp);
3559     psinfo->pr_sid = tswap32(psinfo->pr_sid);
3560 }
3561 
3562 static void bswap_note(struct elf_note *en)
3563 {
3564     bswap32s(&en->n_namesz);
3565     bswap32s(&en->n_descsz);
3566     bswap32s(&en->n_type);
3567 }
3568 #else
3569 static inline void bswap_prstatus(struct target_elf_prstatus *p) { }
3570 static inline void bswap_psinfo(struct target_elf_prpsinfo *p) {}
3571 static inline void bswap_note(struct elf_note *en) { }
3572 #endif /* BSWAP_NEEDED */
3573 
3574 /*
3575  * Minimal support for linux memory regions.  These are needed
3576  * when we are finding out what memory exactly belongs to
3577  * emulated process.  No locks needed here, as long as
3578  * thread that received the signal is stopped.
3579  */
3580 
3581 static struct mm_struct *vma_init(void)
3582 {
3583     struct mm_struct *mm;
3584 
3585     if ((mm = g_malloc(sizeof (*mm))) == NULL)
3586         return (NULL);
3587 
3588     mm->mm_count = 0;
3589     QTAILQ_INIT(&mm->mm_mmap);
3590 
3591     return (mm);
3592 }
3593 
3594 static void vma_delete(struct mm_struct *mm)
3595 {
3596     struct vm_area_struct *vma;
3597 
3598     while ((vma = vma_first(mm)) != NULL) {
3599         QTAILQ_REMOVE(&mm->mm_mmap, vma, vma_link);
3600         g_free(vma);
3601     }
3602     g_free(mm);
3603 }
3604 
3605 static int vma_add_mapping(struct mm_struct *mm, target_ulong start,
3606                            target_ulong end, abi_ulong flags)
3607 {
3608     struct vm_area_struct *vma;
3609 
3610     if ((vma = g_malloc0(sizeof (*vma))) == NULL)
3611         return (-1);
3612 
3613     vma->vma_start = start;
3614     vma->vma_end = end;
3615     vma->vma_flags = flags;
3616 
3617     QTAILQ_INSERT_TAIL(&mm->mm_mmap, vma, vma_link);
3618     mm->mm_count++;
3619 
3620     return (0);
3621 }
3622 
3623 static struct vm_area_struct *vma_first(const struct mm_struct *mm)
3624 {
3625     return (QTAILQ_FIRST(&mm->mm_mmap));
3626 }
3627 
3628 static struct vm_area_struct *vma_next(struct vm_area_struct *vma)
3629 {
3630     return (QTAILQ_NEXT(vma, vma_link));
3631 }
3632 
3633 static int vma_get_mapping_count(const struct mm_struct *mm)
3634 {
3635     return (mm->mm_count);
3636 }
3637 
3638 /*
3639  * Calculate file (dump) size of given memory region.
3640  */
3641 static abi_ulong vma_dump_size(const struct vm_area_struct *vma)
3642 {
3643     /* if we cannot even read the first page, skip it */
3644     if (!access_ok_untagged(VERIFY_READ, vma->vma_start, TARGET_PAGE_SIZE))
3645         return (0);
3646 
3647     /*
3648      * Usually we don't dump executable pages as they contain
3649      * non-writable code that debugger can read directly from
3650      * target library etc.  However, thread stacks are marked
3651      * also executable so we read in first page of given region
3652      * and check whether it contains elf header.  If there is
3653      * no elf header, we dump it.
3654      */
3655     if (vma->vma_flags & PROT_EXEC) {
3656         char page[TARGET_PAGE_SIZE];
3657 
3658         if (copy_from_user(page, vma->vma_start, sizeof (page))) {
3659             return 0;
3660         }
3661         if ((page[EI_MAG0] == ELFMAG0) &&
3662             (page[EI_MAG1] == ELFMAG1) &&
3663             (page[EI_MAG2] == ELFMAG2) &&
3664             (page[EI_MAG3] == ELFMAG3)) {
3665             /*
3666              * Mappings are possibly from ELF binary.  Don't dump
3667              * them.
3668              */
3669             return (0);
3670         }
3671     }
3672 
3673     return (vma->vma_end - vma->vma_start);
3674 }
3675 
3676 static int vma_walker(void *priv, target_ulong start, target_ulong end,
3677                       unsigned long flags)
3678 {
3679     struct mm_struct *mm = (struct mm_struct *)priv;
3680 
3681     vma_add_mapping(mm, start, end, flags);
3682     return (0);
3683 }
3684 
3685 static void fill_note(struct memelfnote *note, const char *name, int type,
3686                       unsigned int sz, void *data)
3687 {
3688     unsigned int namesz;
3689 
3690     namesz = strlen(name) + 1;
3691     note->name = name;
3692     note->namesz = namesz;
3693     note->namesz_rounded = roundup(namesz, sizeof (int32_t));
3694     note->type = type;
3695     note->datasz = sz;
3696     note->datasz_rounded = roundup(sz, sizeof (int32_t));
3697 
3698     note->data = data;
3699 
3700     /*
3701      * We calculate rounded up note size here as specified by
3702      * ELF document.
3703      */
3704     note->notesz = sizeof (struct elf_note) +
3705         note->namesz_rounded + note->datasz_rounded;
3706 }
3707 
3708 static void fill_elf_header(struct elfhdr *elf, int segs, uint16_t machine,
3709                             uint32_t flags)
3710 {
3711     (void) memset(elf, 0, sizeof(*elf));
3712 
3713     (void) memcpy(elf->e_ident, ELFMAG, SELFMAG);
3714     elf->e_ident[EI_CLASS] = ELF_CLASS;
3715     elf->e_ident[EI_DATA] = ELF_DATA;
3716     elf->e_ident[EI_VERSION] = EV_CURRENT;
3717     elf->e_ident[EI_OSABI] = ELF_OSABI;
3718 
3719     elf->e_type = ET_CORE;
3720     elf->e_machine = machine;
3721     elf->e_version = EV_CURRENT;
3722     elf->e_phoff = sizeof(struct elfhdr);
3723     elf->e_flags = flags;
3724     elf->e_ehsize = sizeof(struct elfhdr);
3725     elf->e_phentsize = sizeof(struct elf_phdr);
3726     elf->e_phnum = segs;
3727 
3728     bswap_ehdr(elf);
3729 }
3730 
3731 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, off_t offset)
3732 {
3733     phdr->p_type = PT_NOTE;
3734     phdr->p_offset = offset;
3735     phdr->p_vaddr = 0;
3736     phdr->p_paddr = 0;
3737     phdr->p_filesz = sz;
3738     phdr->p_memsz = 0;
3739     phdr->p_flags = 0;
3740     phdr->p_align = 0;
3741 
3742     bswap_phdr(phdr, 1);
3743 }
3744 
3745 static size_t note_size(const struct memelfnote *note)
3746 {
3747     return (note->notesz);
3748 }
3749 
3750 static void fill_prstatus(struct target_elf_prstatus *prstatus,
3751                           const TaskState *ts, int signr)
3752 {
3753     (void) memset(prstatus, 0, sizeof (*prstatus));
3754     prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
3755     prstatus->pr_pid = ts->ts_tid;
3756     prstatus->pr_ppid = getppid();
3757     prstatus->pr_pgrp = getpgrp();
3758     prstatus->pr_sid = getsid(0);
3759 
3760     bswap_prstatus(prstatus);
3761 }
3762 
3763 static int fill_psinfo(struct target_elf_prpsinfo *psinfo, const TaskState *ts)
3764 {
3765     char *base_filename;
3766     unsigned int i, len;
3767 
3768     (void) memset(psinfo, 0, sizeof (*psinfo));
3769 
3770     len = ts->info->env_strings - ts->info->arg_strings;
3771     if (len >= ELF_PRARGSZ)
3772         len = ELF_PRARGSZ - 1;
3773     if (copy_from_user(&psinfo->pr_psargs, ts->info->arg_strings, len)) {
3774         return -EFAULT;
3775     }
3776     for (i = 0; i < len; i++)
3777         if (psinfo->pr_psargs[i] == 0)
3778             psinfo->pr_psargs[i] = ' ';
3779     psinfo->pr_psargs[len] = 0;
3780 
3781     psinfo->pr_pid = getpid();
3782     psinfo->pr_ppid = getppid();
3783     psinfo->pr_pgrp = getpgrp();
3784     psinfo->pr_sid = getsid(0);
3785     psinfo->pr_uid = getuid();
3786     psinfo->pr_gid = getgid();
3787 
3788     base_filename = g_path_get_basename(ts->bprm->filename);
3789     /*
3790      * Using strncpy here is fine: at max-length,
3791      * this field is not NUL-terminated.
3792      */
3793     (void) strncpy(psinfo->pr_fname, base_filename,
3794                    sizeof(psinfo->pr_fname));
3795 
3796     g_free(base_filename);
3797     bswap_psinfo(psinfo);
3798     return (0);
3799 }
3800 
3801 static void fill_auxv_note(struct memelfnote *note, const TaskState *ts)
3802 {
3803     elf_addr_t auxv = (elf_addr_t)ts->info->saved_auxv;
3804     elf_addr_t orig_auxv = auxv;
3805     void *ptr;
3806     int len = ts->info->auxv_len;
3807 
3808     /*
3809      * Auxiliary vector is stored in target process stack.  It contains
3810      * {type, value} pairs that we need to dump into note.  This is not
3811      * strictly necessary but we do it here for sake of completeness.
3812      */
3813 
3814     /* read in whole auxv vector and copy it to memelfnote */
3815     ptr = lock_user(VERIFY_READ, orig_auxv, len, 0);
3816     if (ptr != NULL) {
3817         fill_note(note, "CORE", NT_AUXV, len, ptr);
3818         unlock_user(ptr, auxv, len);
3819     }
3820 }
3821 
3822 /*
3823  * Constructs name of coredump file.  We have following convention
3824  * for the name:
3825  *     qemu_<basename-of-target-binary>_<date>-<time>_<pid>.core
3826  *
3827  * Returns the filename
3828  */
3829 static char *core_dump_filename(const TaskState *ts)
3830 {
3831     g_autoptr(GDateTime) now = g_date_time_new_now_local();
3832     g_autofree char *nowstr = g_date_time_format(now, "%Y%m%d-%H%M%S");
3833     g_autofree char *base_filename = g_path_get_basename(ts->bprm->filename);
3834 
3835     return g_strdup_printf("qemu_%s_%s_%d.core",
3836                            base_filename, nowstr, (int)getpid());
3837 }
3838 
3839 static int dump_write(int fd, const void *ptr, size_t size)
3840 {
3841     const char *bufp = (const char *)ptr;
3842     ssize_t bytes_written, bytes_left;
3843     struct rlimit dumpsize;
3844     off_t pos;
3845 
3846     bytes_written = 0;
3847     getrlimit(RLIMIT_CORE, &dumpsize);
3848     if ((pos = lseek(fd, 0, SEEK_CUR))==-1) {
3849         if (errno == ESPIPE) { /* not a seekable stream */
3850             bytes_left = size;
3851         } else {
3852             return pos;
3853         }
3854     } else {
3855         if (dumpsize.rlim_cur <= pos) {
3856             return -1;
3857         } else if (dumpsize.rlim_cur == RLIM_INFINITY) {
3858             bytes_left = size;
3859         } else {
3860             size_t limit_left=dumpsize.rlim_cur - pos;
3861             bytes_left = limit_left >= size ? size : limit_left ;
3862         }
3863     }
3864 
3865     /*
3866      * In normal conditions, single write(2) should do but
3867      * in case of socket etc. this mechanism is more portable.
3868      */
3869     do {
3870         bytes_written = write(fd, bufp, bytes_left);
3871         if (bytes_written < 0) {
3872             if (errno == EINTR)
3873                 continue;
3874             return (-1);
3875         } else if (bytes_written == 0) { /* eof */
3876             return (-1);
3877         }
3878         bufp += bytes_written;
3879         bytes_left -= bytes_written;
3880     } while (bytes_left > 0);
3881 
3882     return (0);
3883 }
3884 
3885 static int write_note(struct memelfnote *men, int fd)
3886 {
3887     struct elf_note en;
3888 
3889     en.n_namesz = men->namesz;
3890     en.n_type = men->type;
3891     en.n_descsz = men->datasz;
3892 
3893     bswap_note(&en);
3894 
3895     if (dump_write(fd, &en, sizeof(en)) != 0)
3896         return (-1);
3897     if (dump_write(fd, men->name, men->namesz_rounded) != 0)
3898         return (-1);
3899     if (dump_write(fd, men->data, men->datasz_rounded) != 0)
3900         return (-1);
3901 
3902     return (0);
3903 }
3904 
3905 static void fill_thread_info(struct elf_note_info *info, const CPUArchState *env)
3906 {
3907     CPUState *cpu = env_cpu((CPUArchState *)env);
3908     TaskState *ts = (TaskState *)cpu->opaque;
3909     struct elf_thread_status *ets;
3910 
3911     ets = g_malloc0(sizeof (*ets));
3912     ets->num_notes = 1; /* only prstatus is dumped */
3913     fill_prstatus(&ets->prstatus, ts, 0);
3914     elf_core_copy_regs(&ets->prstatus.pr_reg, env);
3915     fill_note(&ets->notes[0], "CORE", NT_PRSTATUS, sizeof (ets->prstatus),
3916               &ets->prstatus);
3917 
3918     QTAILQ_INSERT_TAIL(&info->thread_list, ets, ets_link);
3919 
3920     info->notes_size += note_size(&ets->notes[0]);
3921 }
3922 
3923 static void init_note_info(struct elf_note_info *info)
3924 {
3925     /* Initialize the elf_note_info structure so that it is at
3926      * least safe to call free_note_info() on it. Must be
3927      * called before calling fill_note_info().
3928      */
3929     memset(info, 0, sizeof (*info));
3930     QTAILQ_INIT(&info->thread_list);
3931 }
3932 
3933 static int fill_note_info(struct elf_note_info *info,
3934                           long signr, const CPUArchState *env)
3935 {
3936 #define NUMNOTES 3
3937     CPUState *cpu = env_cpu((CPUArchState *)env);
3938     TaskState *ts = (TaskState *)cpu->opaque;
3939     int i;
3940 
3941     info->notes = g_new0(struct memelfnote, NUMNOTES);
3942     if (info->notes == NULL)
3943         return (-ENOMEM);
3944     info->prstatus = g_malloc0(sizeof (*info->prstatus));
3945     if (info->prstatus == NULL)
3946         return (-ENOMEM);
3947     info->psinfo = g_malloc0(sizeof (*info->psinfo));
3948     if (info->prstatus == NULL)
3949         return (-ENOMEM);
3950 
3951     /*
3952      * First fill in status (and registers) of current thread
3953      * including process info & aux vector.
3954      */
3955     fill_prstatus(info->prstatus, ts, signr);
3956     elf_core_copy_regs(&info->prstatus->pr_reg, env);
3957     fill_note(&info->notes[0], "CORE", NT_PRSTATUS,
3958               sizeof (*info->prstatus), info->prstatus);
3959     fill_psinfo(info->psinfo, ts);
3960     fill_note(&info->notes[1], "CORE", NT_PRPSINFO,
3961               sizeof (*info->psinfo), info->psinfo);
3962     fill_auxv_note(&info->notes[2], ts);
3963     info->numnote = 3;
3964 
3965     info->notes_size = 0;
3966     for (i = 0; i < info->numnote; i++)
3967         info->notes_size += note_size(&info->notes[i]);
3968 
3969     /* read and fill status of all threads */
3970     cpu_list_lock();
3971     CPU_FOREACH(cpu) {
3972         if (cpu == thread_cpu) {
3973             continue;
3974         }
3975         fill_thread_info(info, (CPUArchState *)cpu->env_ptr);
3976     }
3977     cpu_list_unlock();
3978 
3979     return (0);
3980 }
3981 
3982 static void free_note_info(struct elf_note_info *info)
3983 {
3984     struct elf_thread_status *ets;
3985 
3986     while (!QTAILQ_EMPTY(&info->thread_list)) {
3987         ets = QTAILQ_FIRST(&info->thread_list);
3988         QTAILQ_REMOVE(&info->thread_list, ets, ets_link);
3989         g_free(ets);
3990     }
3991 
3992     g_free(info->prstatus);
3993     g_free(info->psinfo);
3994     g_free(info->notes);
3995 }
3996 
3997 static int write_note_info(struct elf_note_info *info, int fd)
3998 {
3999     struct elf_thread_status *ets;
4000     int i, error = 0;
4001 
4002     /* write prstatus, psinfo and auxv for current thread */
4003     for (i = 0; i < info->numnote; i++)
4004         if ((error = write_note(&info->notes[i], fd)) != 0)
4005             return (error);
4006 
4007     /* write prstatus for each thread */
4008     QTAILQ_FOREACH(ets, &info->thread_list, ets_link) {
4009         if ((error = write_note(&ets->notes[0], fd)) != 0)
4010             return (error);
4011     }
4012 
4013     return (0);
4014 }
4015 
4016 /*
4017  * Write out ELF coredump.
4018  *
4019  * See documentation of ELF object file format in:
4020  * http://www.caldera.com/developers/devspecs/gabi41.pdf
4021  *
4022  * Coredump format in linux is following:
4023  *
4024  * 0   +----------------------+         \
4025  *     | ELF header           | ET_CORE  |
4026  *     +----------------------+          |
4027  *     | ELF program headers  |          |--- headers
4028  *     | - NOTE section       |          |
4029  *     | - PT_LOAD sections   |          |
4030  *     +----------------------+         /
4031  *     | NOTEs:               |
4032  *     | - NT_PRSTATUS        |
4033  *     | - NT_PRSINFO         |
4034  *     | - NT_AUXV            |
4035  *     +----------------------+ <-- aligned to target page
4036  *     | Process memory dump  |
4037  *     :                      :
4038  *     .                      .
4039  *     :                      :
4040  *     |                      |
4041  *     +----------------------+
4042  *
4043  * NT_PRSTATUS -> struct elf_prstatus (per thread)
4044  * NT_PRSINFO  -> struct elf_prpsinfo
4045  * NT_AUXV is array of { type, value } pairs (see fill_auxv_note()).
4046  *
4047  * Format follows System V format as close as possible.  Current
4048  * version limitations are as follows:
4049  *     - no floating point registers are dumped
4050  *
4051  * Function returns 0 in case of success, negative errno otherwise.
4052  *
4053  * TODO: make this work also during runtime: it should be
4054  * possible to force coredump from running process and then
4055  * continue processing.  For example qemu could set up SIGUSR2
4056  * handler (provided that target process haven't registered
4057  * handler for that) that does the dump when signal is received.
4058  */
4059 static int elf_core_dump(int signr, const CPUArchState *env)
4060 {
4061     const CPUState *cpu = env_cpu((CPUArchState *)env);
4062     const TaskState *ts = (const TaskState *)cpu->opaque;
4063     struct vm_area_struct *vma = NULL;
4064     g_autofree char *corefile = NULL;
4065     struct elf_note_info info;
4066     struct elfhdr elf;
4067     struct elf_phdr phdr;
4068     struct rlimit dumpsize;
4069     struct mm_struct *mm = NULL;
4070     off_t offset = 0, data_offset = 0;
4071     int segs = 0;
4072     int fd = -1;
4073 
4074     init_note_info(&info);
4075 
4076     errno = 0;
4077     getrlimit(RLIMIT_CORE, &dumpsize);
4078     if (dumpsize.rlim_cur == 0)
4079         return 0;
4080 
4081     corefile = core_dump_filename(ts);
4082 
4083     if ((fd = open(corefile, O_WRONLY | O_CREAT,
4084                    S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH)) < 0)
4085         return (-errno);
4086 
4087     /*
4088      * Walk through target process memory mappings and
4089      * set up structure containing this information.  After
4090      * this point vma_xxx functions can be used.
4091      */
4092     if ((mm = vma_init()) == NULL)
4093         goto out;
4094 
4095     walk_memory_regions(mm, vma_walker);
4096     segs = vma_get_mapping_count(mm);
4097 
4098     /*
4099      * Construct valid coredump ELF header.  We also
4100      * add one more segment for notes.
4101      */
4102     fill_elf_header(&elf, segs + 1, ELF_MACHINE, 0);
4103     if (dump_write(fd, &elf, sizeof (elf)) != 0)
4104         goto out;
4105 
4106     /* fill in the in-memory version of notes */
4107     if (fill_note_info(&info, signr, env) < 0)
4108         goto out;
4109 
4110     offset += sizeof (elf);                             /* elf header */
4111     offset += (segs + 1) * sizeof (struct elf_phdr);    /* program headers */
4112 
4113     /* write out notes program header */
4114     fill_elf_note_phdr(&phdr, info.notes_size, offset);
4115 
4116     offset += info.notes_size;
4117     if (dump_write(fd, &phdr, sizeof (phdr)) != 0)
4118         goto out;
4119 
4120     /*
4121      * ELF specification wants data to start at page boundary so
4122      * we align it here.
4123      */
4124     data_offset = offset = roundup(offset, ELF_EXEC_PAGESIZE);
4125 
4126     /*
4127      * Write program headers for memory regions mapped in
4128      * the target process.
4129      */
4130     for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) {
4131         (void) memset(&phdr, 0, sizeof (phdr));
4132 
4133         phdr.p_type = PT_LOAD;
4134         phdr.p_offset = offset;
4135         phdr.p_vaddr = vma->vma_start;
4136         phdr.p_paddr = 0;
4137         phdr.p_filesz = vma_dump_size(vma);
4138         offset += phdr.p_filesz;
4139         phdr.p_memsz = vma->vma_end - vma->vma_start;
4140         phdr.p_flags = vma->vma_flags & PROT_READ ? PF_R : 0;
4141         if (vma->vma_flags & PROT_WRITE)
4142             phdr.p_flags |= PF_W;
4143         if (vma->vma_flags & PROT_EXEC)
4144             phdr.p_flags |= PF_X;
4145         phdr.p_align = ELF_EXEC_PAGESIZE;
4146 
4147         bswap_phdr(&phdr, 1);
4148         if (dump_write(fd, &phdr, sizeof(phdr)) != 0) {
4149             goto out;
4150         }
4151     }
4152 
4153     /*
4154      * Next we write notes just after program headers.  No
4155      * alignment needed here.
4156      */
4157     if (write_note_info(&info, fd) < 0)
4158         goto out;
4159 
4160     /* align data to page boundary */
4161     if (lseek(fd, data_offset, SEEK_SET) != data_offset)
4162         goto out;
4163 
4164     /*
4165      * Finally we can dump process memory into corefile as well.
4166      */
4167     for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) {
4168         abi_ulong addr;
4169         abi_ulong end;
4170 
4171         end = vma->vma_start + vma_dump_size(vma);
4172 
4173         for (addr = vma->vma_start; addr < end;
4174              addr += TARGET_PAGE_SIZE) {
4175             char page[TARGET_PAGE_SIZE];
4176             int error;
4177 
4178             /*
4179              *  Read in page from target process memory and
4180              *  write it to coredump file.
4181              */
4182             error = copy_from_user(page, addr, sizeof (page));
4183             if (error != 0) {
4184                 (void) fprintf(stderr, "unable to dump " TARGET_ABI_FMT_lx "\n",
4185                                addr);
4186                 errno = -error;
4187                 goto out;
4188             }
4189             if (dump_write(fd, page, TARGET_PAGE_SIZE) < 0)
4190                 goto out;
4191         }
4192     }
4193 
4194  out:
4195     free_note_info(&info);
4196     if (mm != NULL)
4197         vma_delete(mm);
4198     (void) close(fd);
4199 
4200     if (errno != 0)
4201         return (-errno);
4202     return (0);
4203 }
4204 #endif /* USE_ELF_CORE_DUMP */
4205 
4206 void do_init_thread(struct target_pt_regs *regs, struct image_info *infop)
4207 {
4208     init_thread(regs, infop);
4209 }
4210