xref: /openbmc/qemu/linux-user/elfload.c (revision 284d697c)
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 "disas/disas.h"
10 #include "qemu/path.h"
11 #include "qemu/queue.h"
12 #include "qemu/guest-random.h"
13 #include "qemu/units.h"
14 #include "qemu/selfmap.h"
15 #include "qapi/error.h"
16 
17 #ifdef _ARCH_PPC64
18 #undef ARCH_DLINFO
19 #undef ELF_PLATFORM
20 #undef ELF_HWCAP
21 #undef ELF_HWCAP2
22 #undef ELF_CLASS
23 #undef ELF_DATA
24 #undef ELF_ARCH
25 #endif
26 
27 #define ELF_OSABI   ELFOSABI_SYSV
28 
29 /* from personality.h */
30 
31 /*
32  * Flags for bug emulation.
33  *
34  * These occupy the top three bytes.
35  */
36 enum {
37     ADDR_NO_RANDOMIZE = 0x0040000,      /* disable randomization of VA space */
38     FDPIC_FUNCPTRS =    0x0080000,      /* userspace function ptrs point to
39                                            descriptors (signal handling) */
40     MMAP_PAGE_ZERO =    0x0100000,
41     ADDR_COMPAT_LAYOUT = 0x0200000,
42     READ_IMPLIES_EXEC = 0x0400000,
43     ADDR_LIMIT_32BIT =  0x0800000,
44     SHORT_INODE =       0x1000000,
45     WHOLE_SECONDS =     0x2000000,
46     STICKY_TIMEOUTS =   0x4000000,
47     ADDR_LIMIT_3GB =    0x8000000,
48 };
49 
50 /*
51  * Personality types.
52  *
53  * These go in the low byte.  Avoid using the top bit, it will
54  * conflict with error returns.
55  */
56 enum {
57     PER_LINUX =         0x0000,
58     PER_LINUX_32BIT =   0x0000 | ADDR_LIMIT_32BIT,
59     PER_LINUX_FDPIC =   0x0000 | FDPIC_FUNCPTRS,
60     PER_SVR4 =          0x0001 | STICKY_TIMEOUTS | MMAP_PAGE_ZERO,
61     PER_SVR3 =          0x0002 | STICKY_TIMEOUTS | SHORT_INODE,
62     PER_SCOSVR3 =       0x0003 | STICKY_TIMEOUTS | WHOLE_SECONDS | SHORT_INODE,
63     PER_OSR5 =          0x0003 | STICKY_TIMEOUTS | WHOLE_SECONDS,
64     PER_WYSEV386 =      0x0004 | STICKY_TIMEOUTS | SHORT_INODE,
65     PER_ISCR4 =         0x0005 | STICKY_TIMEOUTS,
66     PER_BSD =           0x0006,
67     PER_SUNOS =         0x0006 | STICKY_TIMEOUTS,
68     PER_XENIX =         0x0007 | STICKY_TIMEOUTS | SHORT_INODE,
69     PER_LINUX32 =       0x0008,
70     PER_LINUX32_3GB =   0x0008 | ADDR_LIMIT_3GB,
71     PER_IRIX32 =        0x0009 | STICKY_TIMEOUTS,/* IRIX5 32-bit */
72     PER_IRIXN32 =       0x000a | STICKY_TIMEOUTS,/* IRIX6 new 32-bit */
73     PER_IRIX64 =        0x000b | STICKY_TIMEOUTS,/* IRIX6 64-bit */
74     PER_RISCOS =        0x000c,
75     PER_SOLARIS =       0x000d | STICKY_TIMEOUTS,
76     PER_UW7 =           0x000e | STICKY_TIMEOUTS | MMAP_PAGE_ZERO,
77     PER_OSF4 =          0x000f,                  /* OSF/1 v4 */
78     PER_HPUX =          0x0010,
79     PER_MASK =          0x00ff,
80 };
81 
82 /*
83  * Return the base personality without flags.
84  */
85 #define personality(pers)       (pers & PER_MASK)
86 
87 int info_is_fdpic(struct image_info *info)
88 {
89     return info->personality == PER_LINUX_FDPIC;
90 }
91 
92 /* this flag is uneffective under linux too, should be deleted */
93 #ifndef MAP_DENYWRITE
94 #define MAP_DENYWRITE 0
95 #endif
96 
97 /* should probably go in elf.h */
98 #ifndef ELIBBAD
99 #define ELIBBAD 80
100 #endif
101 
102 #ifdef TARGET_WORDS_BIGENDIAN
103 #define ELF_DATA        ELFDATA2MSB
104 #else
105 #define ELF_DATA        ELFDATA2LSB
106 #endif
107 
108 #ifdef TARGET_ABI_MIPSN32
109 typedef abi_ullong      target_elf_greg_t;
110 #define tswapreg(ptr)   tswap64(ptr)
111 #else
112 typedef abi_ulong       target_elf_greg_t;
113 #define tswapreg(ptr)   tswapal(ptr)
114 #endif
115 
116 #ifdef USE_UID16
117 typedef abi_ushort      target_uid_t;
118 typedef abi_ushort      target_gid_t;
119 #else
120 typedef abi_uint        target_uid_t;
121 typedef abi_uint        target_gid_t;
122 #endif
123 typedef abi_int         target_pid_t;
124 
125 #ifdef TARGET_I386
126 
127 #define ELF_PLATFORM get_elf_platform()
128 
129 static const char *get_elf_platform(void)
130 {
131     static char elf_platform[] = "i386";
132     int family = object_property_get_int(OBJECT(thread_cpu), "family", NULL);
133     if (family > 6)
134         family = 6;
135     if (family >= 3)
136         elf_platform[1] = '0' + family;
137     return elf_platform;
138 }
139 
140 #define ELF_HWCAP get_elf_hwcap()
141 
142 static uint32_t get_elf_hwcap(void)
143 {
144     X86CPU *cpu = X86_CPU(thread_cpu);
145 
146     return cpu->env.features[FEAT_1_EDX];
147 }
148 
149 #ifdef TARGET_X86_64
150 #define ELF_START_MMAP 0x2aaaaab000ULL
151 
152 #define ELF_CLASS      ELFCLASS64
153 #define ELF_ARCH       EM_X86_64
154 
155 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop)
156 {
157     regs->rax = 0;
158     regs->rsp = infop->start_stack;
159     regs->rip = infop->entry;
160 }
161 
162 #define ELF_NREG    27
163 typedef target_elf_greg_t  target_elf_gregset_t[ELF_NREG];
164 
165 /*
166  * Note that ELF_NREG should be 29 as there should be place for
167  * TRAPNO and ERR "registers" as well but linux doesn't dump
168  * those.
169  *
170  * See linux kernel: arch/x86/include/asm/elf.h
171  */
172 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUX86State *env)
173 {
174     (*regs)[0] = env->regs[15];
175     (*regs)[1] = env->regs[14];
176     (*regs)[2] = env->regs[13];
177     (*regs)[3] = env->regs[12];
178     (*regs)[4] = env->regs[R_EBP];
179     (*regs)[5] = env->regs[R_EBX];
180     (*regs)[6] = env->regs[11];
181     (*regs)[7] = env->regs[10];
182     (*regs)[8] = env->regs[9];
183     (*regs)[9] = env->regs[8];
184     (*regs)[10] = env->regs[R_EAX];
185     (*regs)[11] = env->regs[R_ECX];
186     (*regs)[12] = env->regs[R_EDX];
187     (*regs)[13] = env->regs[R_ESI];
188     (*regs)[14] = env->regs[R_EDI];
189     (*regs)[15] = env->regs[R_EAX]; /* XXX */
190     (*regs)[16] = env->eip;
191     (*regs)[17] = env->segs[R_CS].selector & 0xffff;
192     (*regs)[18] = env->eflags;
193     (*regs)[19] = env->regs[R_ESP];
194     (*regs)[20] = env->segs[R_SS].selector & 0xffff;
195     (*regs)[21] = env->segs[R_FS].selector & 0xffff;
196     (*regs)[22] = env->segs[R_GS].selector & 0xffff;
197     (*regs)[23] = env->segs[R_DS].selector & 0xffff;
198     (*regs)[24] = env->segs[R_ES].selector & 0xffff;
199     (*regs)[25] = env->segs[R_FS].selector & 0xffff;
200     (*regs)[26] = env->segs[R_GS].selector & 0xffff;
201 }
202 
203 #else
204 
205 #define ELF_START_MMAP 0x80000000
206 
207 /*
208  * This is used to ensure we don't load something for the wrong architecture.
209  */
210 #define elf_check_arch(x) ( ((x) == EM_386) || ((x) == EM_486) )
211 
212 /*
213  * These are used to set parameters in the core dumps.
214  */
215 #define ELF_CLASS       ELFCLASS32
216 #define ELF_ARCH        EM_386
217 
218 static inline void init_thread(struct target_pt_regs *regs,
219                                struct image_info *infop)
220 {
221     regs->esp = infop->start_stack;
222     regs->eip = infop->entry;
223 
224     /* SVR4/i386 ABI (pages 3-31, 3-32) says that when the program
225        starts %edx contains a pointer to a function which might be
226        registered using `atexit'.  This provides a mean for the
227        dynamic linker to call DT_FINI functions for shared libraries
228        that have been loaded before the code runs.
229 
230        A value of 0 tells we have no such handler.  */
231     regs->edx = 0;
232 }
233 
234 #define ELF_NREG    17
235 typedef target_elf_greg_t  target_elf_gregset_t[ELF_NREG];
236 
237 /*
238  * Note that ELF_NREG should be 19 as there should be place for
239  * TRAPNO and ERR "registers" as well but linux doesn't dump
240  * those.
241  *
242  * See linux kernel: arch/x86/include/asm/elf.h
243  */
244 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUX86State *env)
245 {
246     (*regs)[0] = env->regs[R_EBX];
247     (*regs)[1] = env->regs[R_ECX];
248     (*regs)[2] = env->regs[R_EDX];
249     (*regs)[3] = env->regs[R_ESI];
250     (*regs)[4] = env->regs[R_EDI];
251     (*regs)[5] = env->regs[R_EBP];
252     (*regs)[6] = env->regs[R_EAX];
253     (*regs)[7] = env->segs[R_DS].selector & 0xffff;
254     (*regs)[8] = env->segs[R_ES].selector & 0xffff;
255     (*regs)[9] = env->segs[R_FS].selector & 0xffff;
256     (*regs)[10] = env->segs[R_GS].selector & 0xffff;
257     (*regs)[11] = env->regs[R_EAX]; /* XXX */
258     (*regs)[12] = env->eip;
259     (*regs)[13] = env->segs[R_CS].selector & 0xffff;
260     (*regs)[14] = env->eflags;
261     (*regs)[15] = env->regs[R_ESP];
262     (*regs)[16] = env->segs[R_SS].selector & 0xffff;
263 }
264 #endif
265 
266 #define USE_ELF_CORE_DUMP
267 #define ELF_EXEC_PAGESIZE       4096
268 
269 #endif
270 
271 #ifdef TARGET_ARM
272 
273 #ifndef TARGET_AARCH64
274 /* 32 bit ARM definitions */
275 
276 #define ELF_START_MMAP 0x80000000
277 
278 #define ELF_ARCH        EM_ARM
279 #define ELF_CLASS       ELFCLASS32
280 
281 static inline void init_thread(struct target_pt_regs *regs,
282                                struct image_info *infop)
283 {
284     abi_long stack = infop->start_stack;
285     memset(regs, 0, sizeof(*regs));
286 
287     regs->uregs[16] = ARM_CPU_MODE_USR;
288     if (infop->entry & 1) {
289         regs->uregs[16] |= CPSR_T;
290     }
291     regs->uregs[15] = infop->entry & 0xfffffffe;
292     regs->uregs[13] = infop->start_stack;
293     /* FIXME - what to for failure of get_user()? */
294     get_user_ual(regs->uregs[2], stack + 8); /* envp */
295     get_user_ual(regs->uregs[1], stack + 4); /* envp */
296     /* XXX: it seems that r0 is zeroed after ! */
297     regs->uregs[0] = 0;
298     /* For uClinux PIC binaries.  */
299     /* XXX: Linux does this only on ARM with no MMU (do we care ?) */
300     regs->uregs[10] = infop->start_data;
301 
302     /* Support ARM FDPIC.  */
303     if (info_is_fdpic(infop)) {
304         /* As described in the ABI document, r7 points to the loadmap info
305          * prepared by the kernel. If an interpreter is needed, r8 points
306          * to the interpreter loadmap and r9 points to the interpreter
307          * PT_DYNAMIC info. If no interpreter is needed, r8 is zero, and
308          * r9 points to the main program PT_DYNAMIC info.
309          */
310         regs->uregs[7] = infop->loadmap_addr;
311         if (infop->interpreter_loadmap_addr) {
312             /* Executable is dynamically loaded.  */
313             regs->uregs[8] = infop->interpreter_loadmap_addr;
314             regs->uregs[9] = infop->interpreter_pt_dynamic_addr;
315         } else {
316             regs->uregs[8] = 0;
317             regs->uregs[9] = infop->pt_dynamic_addr;
318         }
319     }
320 }
321 
322 #define ELF_NREG    18
323 typedef target_elf_greg_t  target_elf_gregset_t[ELF_NREG];
324 
325 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUARMState *env)
326 {
327     (*regs)[0] = tswapreg(env->regs[0]);
328     (*regs)[1] = tswapreg(env->regs[1]);
329     (*regs)[2] = tswapreg(env->regs[2]);
330     (*regs)[3] = tswapreg(env->regs[3]);
331     (*regs)[4] = tswapreg(env->regs[4]);
332     (*regs)[5] = tswapreg(env->regs[5]);
333     (*regs)[6] = tswapreg(env->regs[6]);
334     (*regs)[7] = tswapreg(env->regs[7]);
335     (*regs)[8] = tswapreg(env->regs[8]);
336     (*regs)[9] = tswapreg(env->regs[9]);
337     (*regs)[10] = tswapreg(env->regs[10]);
338     (*regs)[11] = tswapreg(env->regs[11]);
339     (*regs)[12] = tswapreg(env->regs[12]);
340     (*regs)[13] = tswapreg(env->regs[13]);
341     (*regs)[14] = tswapreg(env->regs[14]);
342     (*regs)[15] = tswapreg(env->regs[15]);
343 
344     (*regs)[16] = tswapreg(cpsr_read((CPUARMState *)env));
345     (*regs)[17] = tswapreg(env->regs[0]); /* XXX */
346 }
347 
348 #define USE_ELF_CORE_DUMP
349 #define ELF_EXEC_PAGESIZE       4096
350 
351 enum
352 {
353     ARM_HWCAP_ARM_SWP       = 1 << 0,
354     ARM_HWCAP_ARM_HALF      = 1 << 1,
355     ARM_HWCAP_ARM_THUMB     = 1 << 2,
356     ARM_HWCAP_ARM_26BIT     = 1 << 3,
357     ARM_HWCAP_ARM_FAST_MULT = 1 << 4,
358     ARM_HWCAP_ARM_FPA       = 1 << 5,
359     ARM_HWCAP_ARM_VFP       = 1 << 6,
360     ARM_HWCAP_ARM_EDSP      = 1 << 7,
361     ARM_HWCAP_ARM_JAVA      = 1 << 8,
362     ARM_HWCAP_ARM_IWMMXT    = 1 << 9,
363     ARM_HWCAP_ARM_CRUNCH    = 1 << 10,
364     ARM_HWCAP_ARM_THUMBEE   = 1 << 11,
365     ARM_HWCAP_ARM_NEON      = 1 << 12,
366     ARM_HWCAP_ARM_VFPv3     = 1 << 13,
367     ARM_HWCAP_ARM_VFPv3D16  = 1 << 14,
368     ARM_HWCAP_ARM_TLS       = 1 << 15,
369     ARM_HWCAP_ARM_VFPv4     = 1 << 16,
370     ARM_HWCAP_ARM_IDIVA     = 1 << 17,
371     ARM_HWCAP_ARM_IDIVT     = 1 << 18,
372     ARM_HWCAP_ARM_VFPD32    = 1 << 19,
373     ARM_HWCAP_ARM_LPAE      = 1 << 20,
374     ARM_HWCAP_ARM_EVTSTRM   = 1 << 21,
375 };
376 
377 enum {
378     ARM_HWCAP2_ARM_AES      = 1 << 0,
379     ARM_HWCAP2_ARM_PMULL    = 1 << 1,
380     ARM_HWCAP2_ARM_SHA1     = 1 << 2,
381     ARM_HWCAP2_ARM_SHA2     = 1 << 3,
382     ARM_HWCAP2_ARM_CRC32    = 1 << 4,
383 };
384 
385 /* The commpage only exists for 32 bit kernels */
386 
387 #define ARM_COMMPAGE (intptr_t)0xffff0f00u
388 
389 static bool init_guest_commpage(void)
390 {
391     void *want = g2h(ARM_COMMPAGE & -qemu_host_page_size);
392     void *addr = mmap(want, qemu_host_page_size, PROT_READ | PROT_WRITE,
393                       MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0);
394 
395     if (addr == MAP_FAILED) {
396         perror("Allocating guest commpage");
397         exit(EXIT_FAILURE);
398     }
399     if (addr != want) {
400         return false;
401     }
402 
403     /* Set kernel helper versions; rest of page is 0.  */
404     __put_user(5, (uint32_t *)g2h(0xffff0ffcu));
405 
406     if (mprotect(addr, qemu_host_page_size, PROT_READ)) {
407         perror("Protecting guest commpage");
408         exit(EXIT_FAILURE);
409     }
410     return true;
411 }
412 
413 #define ELF_HWCAP get_elf_hwcap()
414 #define ELF_HWCAP2 get_elf_hwcap2()
415 
416 static uint32_t get_elf_hwcap(void)
417 {
418     ARMCPU *cpu = ARM_CPU(thread_cpu);
419     uint32_t hwcaps = 0;
420 
421     hwcaps |= ARM_HWCAP_ARM_SWP;
422     hwcaps |= ARM_HWCAP_ARM_HALF;
423     hwcaps |= ARM_HWCAP_ARM_THUMB;
424     hwcaps |= ARM_HWCAP_ARM_FAST_MULT;
425 
426     /* probe for the extra features */
427 #define GET_FEATURE(feat, hwcap) \
428     do { if (arm_feature(&cpu->env, feat)) { hwcaps |= hwcap; } } while (0)
429 
430 #define GET_FEATURE_ID(feat, hwcap) \
431     do { if (cpu_isar_feature(feat, cpu)) { hwcaps |= hwcap; } } while (0)
432 
433     /* EDSP is in v5TE and above, but all our v5 CPUs are v5TE */
434     GET_FEATURE(ARM_FEATURE_V5, ARM_HWCAP_ARM_EDSP);
435     GET_FEATURE(ARM_FEATURE_IWMMXT, ARM_HWCAP_ARM_IWMMXT);
436     GET_FEATURE(ARM_FEATURE_THUMB2EE, ARM_HWCAP_ARM_THUMBEE);
437     GET_FEATURE(ARM_FEATURE_NEON, ARM_HWCAP_ARM_NEON);
438     GET_FEATURE(ARM_FEATURE_V6K, ARM_HWCAP_ARM_TLS);
439     GET_FEATURE(ARM_FEATURE_LPAE, ARM_HWCAP_ARM_LPAE);
440     GET_FEATURE_ID(aa32_arm_div, ARM_HWCAP_ARM_IDIVA);
441     GET_FEATURE_ID(aa32_thumb_div, ARM_HWCAP_ARM_IDIVT);
442     GET_FEATURE_ID(aa32_vfp, ARM_HWCAP_ARM_VFP);
443 
444     if (cpu_isar_feature(aa32_fpsp_v3, cpu) ||
445         cpu_isar_feature(aa32_fpdp_v3, cpu)) {
446         hwcaps |= ARM_HWCAP_ARM_VFPv3;
447         if (cpu_isar_feature(aa32_simd_r32, cpu)) {
448             hwcaps |= ARM_HWCAP_ARM_VFPD32;
449         } else {
450             hwcaps |= ARM_HWCAP_ARM_VFPv3D16;
451         }
452     }
453     GET_FEATURE_ID(aa32_simdfmac, ARM_HWCAP_ARM_VFPv4);
454 
455     return hwcaps;
456 }
457 
458 static uint32_t get_elf_hwcap2(void)
459 {
460     ARMCPU *cpu = ARM_CPU(thread_cpu);
461     uint32_t hwcaps = 0;
462 
463     GET_FEATURE_ID(aa32_aes, ARM_HWCAP2_ARM_AES);
464     GET_FEATURE_ID(aa32_pmull, ARM_HWCAP2_ARM_PMULL);
465     GET_FEATURE_ID(aa32_sha1, ARM_HWCAP2_ARM_SHA1);
466     GET_FEATURE_ID(aa32_sha2, ARM_HWCAP2_ARM_SHA2);
467     GET_FEATURE_ID(aa32_crc32, ARM_HWCAP2_ARM_CRC32);
468     return hwcaps;
469 }
470 
471 #undef GET_FEATURE
472 #undef GET_FEATURE_ID
473 
474 #define ELF_PLATFORM get_elf_platform()
475 
476 static const char *get_elf_platform(void)
477 {
478     CPUARMState *env = thread_cpu->env_ptr;
479 
480 #ifdef TARGET_WORDS_BIGENDIAN
481 # define END  "b"
482 #else
483 # define END  "l"
484 #endif
485 
486     if (arm_feature(env, ARM_FEATURE_V8)) {
487         return "v8" END;
488     } else if (arm_feature(env, ARM_FEATURE_V7)) {
489         if (arm_feature(env, ARM_FEATURE_M)) {
490             return "v7m" END;
491         } else {
492             return "v7" END;
493         }
494     } else if (arm_feature(env, ARM_FEATURE_V6)) {
495         return "v6" END;
496     } else if (arm_feature(env, ARM_FEATURE_V5)) {
497         return "v5" END;
498     } else {
499         return "v4" END;
500     }
501 
502 #undef END
503 }
504 
505 #else
506 /* 64 bit ARM definitions */
507 #define ELF_START_MMAP 0x80000000
508 
509 #define ELF_ARCH        EM_AARCH64
510 #define ELF_CLASS       ELFCLASS64
511 #ifdef TARGET_WORDS_BIGENDIAN
512 # define ELF_PLATFORM    "aarch64_be"
513 #else
514 # define ELF_PLATFORM    "aarch64"
515 #endif
516 
517 static inline void init_thread(struct target_pt_regs *regs,
518                                struct image_info *infop)
519 {
520     abi_long stack = infop->start_stack;
521     memset(regs, 0, sizeof(*regs));
522 
523     regs->pc = infop->entry & ~0x3ULL;
524     regs->sp = stack;
525 }
526 
527 #define ELF_NREG    34
528 typedef target_elf_greg_t  target_elf_gregset_t[ELF_NREG];
529 
530 static void elf_core_copy_regs(target_elf_gregset_t *regs,
531                                const CPUARMState *env)
532 {
533     int i;
534 
535     for (i = 0; i < 32; i++) {
536         (*regs)[i] = tswapreg(env->xregs[i]);
537     }
538     (*regs)[32] = tswapreg(env->pc);
539     (*regs)[33] = tswapreg(pstate_read((CPUARMState *)env));
540 }
541 
542 #define USE_ELF_CORE_DUMP
543 #define ELF_EXEC_PAGESIZE       4096
544 
545 enum {
546     ARM_HWCAP_A64_FP            = 1 << 0,
547     ARM_HWCAP_A64_ASIMD         = 1 << 1,
548     ARM_HWCAP_A64_EVTSTRM       = 1 << 2,
549     ARM_HWCAP_A64_AES           = 1 << 3,
550     ARM_HWCAP_A64_PMULL         = 1 << 4,
551     ARM_HWCAP_A64_SHA1          = 1 << 5,
552     ARM_HWCAP_A64_SHA2          = 1 << 6,
553     ARM_HWCAP_A64_CRC32         = 1 << 7,
554     ARM_HWCAP_A64_ATOMICS       = 1 << 8,
555     ARM_HWCAP_A64_FPHP          = 1 << 9,
556     ARM_HWCAP_A64_ASIMDHP       = 1 << 10,
557     ARM_HWCAP_A64_CPUID         = 1 << 11,
558     ARM_HWCAP_A64_ASIMDRDM      = 1 << 12,
559     ARM_HWCAP_A64_JSCVT         = 1 << 13,
560     ARM_HWCAP_A64_FCMA          = 1 << 14,
561     ARM_HWCAP_A64_LRCPC         = 1 << 15,
562     ARM_HWCAP_A64_DCPOP         = 1 << 16,
563     ARM_HWCAP_A64_SHA3          = 1 << 17,
564     ARM_HWCAP_A64_SM3           = 1 << 18,
565     ARM_HWCAP_A64_SM4           = 1 << 19,
566     ARM_HWCAP_A64_ASIMDDP       = 1 << 20,
567     ARM_HWCAP_A64_SHA512        = 1 << 21,
568     ARM_HWCAP_A64_SVE           = 1 << 22,
569     ARM_HWCAP_A64_ASIMDFHM      = 1 << 23,
570     ARM_HWCAP_A64_DIT           = 1 << 24,
571     ARM_HWCAP_A64_USCAT         = 1 << 25,
572     ARM_HWCAP_A64_ILRCPC        = 1 << 26,
573     ARM_HWCAP_A64_FLAGM         = 1 << 27,
574     ARM_HWCAP_A64_SSBS          = 1 << 28,
575     ARM_HWCAP_A64_SB            = 1 << 29,
576     ARM_HWCAP_A64_PACA          = 1 << 30,
577     ARM_HWCAP_A64_PACG          = 1UL << 31,
578 
579     ARM_HWCAP2_A64_DCPODP       = 1 << 0,
580     ARM_HWCAP2_A64_SVE2         = 1 << 1,
581     ARM_HWCAP2_A64_SVEAES       = 1 << 2,
582     ARM_HWCAP2_A64_SVEPMULL     = 1 << 3,
583     ARM_HWCAP2_A64_SVEBITPERM   = 1 << 4,
584     ARM_HWCAP2_A64_SVESHA3      = 1 << 5,
585     ARM_HWCAP2_A64_SVESM4       = 1 << 6,
586     ARM_HWCAP2_A64_FLAGM2       = 1 << 7,
587     ARM_HWCAP2_A64_FRINT        = 1 << 8,
588 };
589 
590 #define ELF_HWCAP   get_elf_hwcap()
591 #define ELF_HWCAP2  get_elf_hwcap2()
592 
593 #define GET_FEATURE_ID(feat, hwcap) \
594     do { if (cpu_isar_feature(feat, cpu)) { hwcaps |= hwcap; } } while (0)
595 
596 static uint32_t get_elf_hwcap(void)
597 {
598     ARMCPU *cpu = ARM_CPU(thread_cpu);
599     uint32_t hwcaps = 0;
600 
601     hwcaps |= ARM_HWCAP_A64_FP;
602     hwcaps |= ARM_HWCAP_A64_ASIMD;
603     hwcaps |= ARM_HWCAP_A64_CPUID;
604 
605     /* probe for the extra features */
606 
607     GET_FEATURE_ID(aa64_aes, ARM_HWCAP_A64_AES);
608     GET_FEATURE_ID(aa64_pmull, ARM_HWCAP_A64_PMULL);
609     GET_FEATURE_ID(aa64_sha1, ARM_HWCAP_A64_SHA1);
610     GET_FEATURE_ID(aa64_sha256, ARM_HWCAP_A64_SHA2);
611     GET_FEATURE_ID(aa64_sha512, ARM_HWCAP_A64_SHA512);
612     GET_FEATURE_ID(aa64_crc32, ARM_HWCAP_A64_CRC32);
613     GET_FEATURE_ID(aa64_sha3, ARM_HWCAP_A64_SHA3);
614     GET_FEATURE_ID(aa64_sm3, ARM_HWCAP_A64_SM3);
615     GET_FEATURE_ID(aa64_sm4, ARM_HWCAP_A64_SM4);
616     GET_FEATURE_ID(aa64_fp16, ARM_HWCAP_A64_FPHP | ARM_HWCAP_A64_ASIMDHP);
617     GET_FEATURE_ID(aa64_atomics, ARM_HWCAP_A64_ATOMICS);
618     GET_FEATURE_ID(aa64_rdm, ARM_HWCAP_A64_ASIMDRDM);
619     GET_FEATURE_ID(aa64_dp, ARM_HWCAP_A64_ASIMDDP);
620     GET_FEATURE_ID(aa64_fcma, ARM_HWCAP_A64_FCMA);
621     GET_FEATURE_ID(aa64_sve, ARM_HWCAP_A64_SVE);
622     GET_FEATURE_ID(aa64_pauth, ARM_HWCAP_A64_PACA | ARM_HWCAP_A64_PACG);
623     GET_FEATURE_ID(aa64_fhm, ARM_HWCAP_A64_ASIMDFHM);
624     GET_FEATURE_ID(aa64_jscvt, ARM_HWCAP_A64_JSCVT);
625     GET_FEATURE_ID(aa64_sb, ARM_HWCAP_A64_SB);
626     GET_FEATURE_ID(aa64_condm_4, ARM_HWCAP_A64_FLAGM);
627     GET_FEATURE_ID(aa64_dcpop, ARM_HWCAP_A64_DCPOP);
628     GET_FEATURE_ID(aa64_rcpc_8_3, ARM_HWCAP_A64_LRCPC);
629     GET_FEATURE_ID(aa64_rcpc_8_4, ARM_HWCAP_A64_ILRCPC);
630 
631     return hwcaps;
632 }
633 
634 static uint32_t get_elf_hwcap2(void)
635 {
636     ARMCPU *cpu = ARM_CPU(thread_cpu);
637     uint32_t hwcaps = 0;
638 
639     GET_FEATURE_ID(aa64_dcpodp, ARM_HWCAP2_A64_DCPODP);
640     GET_FEATURE_ID(aa64_condm_5, ARM_HWCAP2_A64_FLAGM2);
641     GET_FEATURE_ID(aa64_frint, ARM_HWCAP2_A64_FRINT);
642 
643     return hwcaps;
644 }
645 
646 #undef GET_FEATURE_ID
647 
648 #endif /* not TARGET_AARCH64 */
649 #endif /* TARGET_ARM */
650 
651 #ifdef TARGET_SPARC
652 #ifdef TARGET_SPARC64
653 
654 #define ELF_START_MMAP 0x80000000
655 #define ELF_HWCAP  (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \
656                     | HWCAP_SPARC_MULDIV | HWCAP_SPARC_V9)
657 #ifndef TARGET_ABI32
658 #define elf_check_arch(x) ( (x) == EM_SPARCV9 || (x) == EM_SPARC32PLUS )
659 #else
660 #define elf_check_arch(x) ( (x) == EM_SPARC32PLUS || (x) == EM_SPARC )
661 #endif
662 
663 #define ELF_CLASS   ELFCLASS64
664 #define ELF_ARCH    EM_SPARCV9
665 
666 #define STACK_BIAS              2047
667 
668 static inline void init_thread(struct target_pt_regs *regs,
669                                struct image_info *infop)
670 {
671 #ifndef TARGET_ABI32
672     regs->tstate = 0;
673 #endif
674     regs->pc = infop->entry;
675     regs->npc = regs->pc + 4;
676     regs->y = 0;
677 #ifdef TARGET_ABI32
678     regs->u_regs[14] = infop->start_stack - 16 * 4;
679 #else
680     if (personality(infop->personality) == PER_LINUX32)
681         regs->u_regs[14] = infop->start_stack - 16 * 4;
682     else
683         regs->u_regs[14] = infop->start_stack - 16 * 8 - STACK_BIAS;
684 #endif
685 }
686 
687 #else
688 #define ELF_START_MMAP 0x80000000
689 #define ELF_HWCAP  (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \
690                     | HWCAP_SPARC_MULDIV)
691 
692 #define ELF_CLASS   ELFCLASS32
693 #define ELF_ARCH    EM_SPARC
694 
695 static inline void init_thread(struct target_pt_regs *regs,
696                                struct image_info *infop)
697 {
698     regs->psr = 0;
699     regs->pc = infop->entry;
700     regs->npc = regs->pc + 4;
701     regs->y = 0;
702     regs->u_regs[14] = infop->start_stack - 16 * 4;
703 }
704 
705 #endif
706 #endif
707 
708 #ifdef TARGET_PPC
709 
710 #define ELF_MACHINE    PPC_ELF_MACHINE
711 #define ELF_START_MMAP 0x80000000
712 
713 #if defined(TARGET_PPC64) && !defined(TARGET_ABI32)
714 
715 #define elf_check_arch(x) ( (x) == EM_PPC64 )
716 
717 #define ELF_CLASS       ELFCLASS64
718 
719 #else
720 
721 #define ELF_CLASS       ELFCLASS32
722 
723 #endif
724 
725 #define ELF_ARCH        EM_PPC
726 
727 /* Feature masks for the Aux Vector Hardware Capabilities (AT_HWCAP).
728    See arch/powerpc/include/asm/cputable.h.  */
729 enum {
730     QEMU_PPC_FEATURE_32 = 0x80000000,
731     QEMU_PPC_FEATURE_64 = 0x40000000,
732     QEMU_PPC_FEATURE_601_INSTR = 0x20000000,
733     QEMU_PPC_FEATURE_HAS_ALTIVEC = 0x10000000,
734     QEMU_PPC_FEATURE_HAS_FPU = 0x08000000,
735     QEMU_PPC_FEATURE_HAS_MMU = 0x04000000,
736     QEMU_PPC_FEATURE_HAS_4xxMAC = 0x02000000,
737     QEMU_PPC_FEATURE_UNIFIED_CACHE = 0x01000000,
738     QEMU_PPC_FEATURE_HAS_SPE = 0x00800000,
739     QEMU_PPC_FEATURE_HAS_EFP_SINGLE = 0x00400000,
740     QEMU_PPC_FEATURE_HAS_EFP_DOUBLE = 0x00200000,
741     QEMU_PPC_FEATURE_NO_TB = 0x00100000,
742     QEMU_PPC_FEATURE_POWER4 = 0x00080000,
743     QEMU_PPC_FEATURE_POWER5 = 0x00040000,
744     QEMU_PPC_FEATURE_POWER5_PLUS = 0x00020000,
745     QEMU_PPC_FEATURE_CELL = 0x00010000,
746     QEMU_PPC_FEATURE_BOOKE = 0x00008000,
747     QEMU_PPC_FEATURE_SMT = 0x00004000,
748     QEMU_PPC_FEATURE_ICACHE_SNOOP = 0x00002000,
749     QEMU_PPC_FEATURE_ARCH_2_05 = 0x00001000,
750     QEMU_PPC_FEATURE_PA6T = 0x00000800,
751     QEMU_PPC_FEATURE_HAS_DFP = 0x00000400,
752     QEMU_PPC_FEATURE_POWER6_EXT = 0x00000200,
753     QEMU_PPC_FEATURE_ARCH_2_06 = 0x00000100,
754     QEMU_PPC_FEATURE_HAS_VSX = 0x00000080,
755     QEMU_PPC_FEATURE_PSERIES_PERFMON_COMPAT = 0x00000040,
756 
757     QEMU_PPC_FEATURE_TRUE_LE = 0x00000002,
758     QEMU_PPC_FEATURE_PPC_LE = 0x00000001,
759 
760     /* Feature definitions in AT_HWCAP2.  */
761     QEMU_PPC_FEATURE2_ARCH_2_07 = 0x80000000, /* ISA 2.07 */
762     QEMU_PPC_FEATURE2_HAS_HTM = 0x40000000, /* Hardware Transactional Memory */
763     QEMU_PPC_FEATURE2_HAS_DSCR = 0x20000000, /* Data Stream Control Register */
764     QEMU_PPC_FEATURE2_HAS_EBB = 0x10000000, /* Event Base Branching */
765     QEMU_PPC_FEATURE2_HAS_ISEL = 0x08000000, /* Integer Select */
766     QEMU_PPC_FEATURE2_HAS_TAR = 0x04000000, /* Target Address Register */
767     QEMU_PPC_FEATURE2_VEC_CRYPTO = 0x02000000,
768     QEMU_PPC_FEATURE2_HTM_NOSC = 0x01000000,
769     QEMU_PPC_FEATURE2_ARCH_3_00 = 0x00800000, /* ISA 3.00 */
770     QEMU_PPC_FEATURE2_HAS_IEEE128 = 0x00400000, /* VSX IEEE Bin Float 128-bit */
771     QEMU_PPC_FEATURE2_DARN = 0x00200000, /* darn random number insn */
772     QEMU_PPC_FEATURE2_SCV = 0x00100000, /* scv syscall */
773     QEMU_PPC_FEATURE2_HTM_NO_SUSPEND = 0x00080000, /* TM w/o suspended state */
774 };
775 
776 #define ELF_HWCAP get_elf_hwcap()
777 
778 static uint32_t get_elf_hwcap(void)
779 {
780     PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);
781     uint32_t features = 0;
782 
783     /* We don't have to be terribly complete here; the high points are
784        Altivec/FP/SPE support.  Anything else is just a bonus.  */
785 #define GET_FEATURE(flag, feature)                                      \
786     do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0)
787 #define GET_FEATURE2(flags, feature) \
788     do { \
789         if ((cpu->env.insns_flags2 & flags) == flags) { \
790             features |= feature; \
791         } \
792     } while (0)
793     GET_FEATURE(PPC_64B, QEMU_PPC_FEATURE_64);
794     GET_FEATURE(PPC_FLOAT, QEMU_PPC_FEATURE_HAS_FPU);
795     GET_FEATURE(PPC_ALTIVEC, QEMU_PPC_FEATURE_HAS_ALTIVEC);
796     GET_FEATURE(PPC_SPE, QEMU_PPC_FEATURE_HAS_SPE);
797     GET_FEATURE(PPC_SPE_SINGLE, QEMU_PPC_FEATURE_HAS_EFP_SINGLE);
798     GET_FEATURE(PPC_SPE_DOUBLE, QEMU_PPC_FEATURE_HAS_EFP_DOUBLE);
799     GET_FEATURE(PPC_BOOKE, QEMU_PPC_FEATURE_BOOKE);
800     GET_FEATURE(PPC_405_MAC, QEMU_PPC_FEATURE_HAS_4xxMAC);
801     GET_FEATURE2(PPC2_DFP, QEMU_PPC_FEATURE_HAS_DFP);
802     GET_FEATURE2(PPC2_VSX, QEMU_PPC_FEATURE_HAS_VSX);
803     GET_FEATURE2((PPC2_PERM_ISA206 | PPC2_DIVE_ISA206 | PPC2_ATOMIC_ISA206 |
804                   PPC2_FP_CVT_ISA206 | PPC2_FP_TST_ISA206),
805                   QEMU_PPC_FEATURE_ARCH_2_06);
806 #undef GET_FEATURE
807 #undef GET_FEATURE2
808 
809     return features;
810 }
811 
812 #define ELF_HWCAP2 get_elf_hwcap2()
813 
814 static uint32_t get_elf_hwcap2(void)
815 {
816     PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);
817     uint32_t features = 0;
818 
819 #define GET_FEATURE(flag, feature)                                      \
820     do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0)
821 #define GET_FEATURE2(flag, feature)                                      \
822     do { if (cpu->env.insns_flags2 & flag) { features |= feature; } } while (0)
823 
824     GET_FEATURE(PPC_ISEL, QEMU_PPC_FEATURE2_HAS_ISEL);
825     GET_FEATURE2(PPC2_BCTAR_ISA207, QEMU_PPC_FEATURE2_HAS_TAR);
826     GET_FEATURE2((PPC2_BCTAR_ISA207 | PPC2_LSQ_ISA207 | PPC2_ALTIVEC_207 |
827                   PPC2_ISA207S), QEMU_PPC_FEATURE2_ARCH_2_07 |
828                   QEMU_PPC_FEATURE2_VEC_CRYPTO);
829     GET_FEATURE2(PPC2_ISA300, QEMU_PPC_FEATURE2_ARCH_3_00 |
830                  QEMU_PPC_FEATURE2_DARN);
831 
832 #undef GET_FEATURE
833 #undef GET_FEATURE2
834 
835     return features;
836 }
837 
838 /*
839  * The requirements here are:
840  * - keep the final alignment of sp (sp & 0xf)
841  * - make sure the 32-bit value at the first 16 byte aligned position of
842  *   AUXV is greater than 16 for glibc compatibility.
843  *   AT_IGNOREPPC is used for that.
844  * - for compatibility with glibc ARCH_DLINFO must always be defined on PPC,
845  *   even if DLINFO_ARCH_ITEMS goes to zero or is undefined.
846  */
847 #define DLINFO_ARCH_ITEMS       5
848 #define ARCH_DLINFO                                     \
849     do {                                                \
850         PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);              \
851         /*                                              \
852          * Handle glibc compatibility: these magic entries must \
853          * be at the lowest addresses in the final auxv.        \
854          */                                             \
855         NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC);        \
856         NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC);        \
857         NEW_AUX_ENT(AT_DCACHEBSIZE, cpu->env.dcache_line_size); \
858         NEW_AUX_ENT(AT_ICACHEBSIZE, cpu->env.icache_line_size); \
859         NEW_AUX_ENT(AT_UCACHEBSIZE, 0);                 \
860     } while (0)
861 
862 static inline void init_thread(struct target_pt_regs *_regs, struct image_info *infop)
863 {
864     _regs->gpr[1] = infop->start_stack;
865 #if defined(TARGET_PPC64) && !defined(TARGET_ABI32)
866     if (get_ppc64_abi(infop) < 2) {
867         uint64_t val;
868         get_user_u64(val, infop->entry + 8);
869         _regs->gpr[2] = val + infop->load_bias;
870         get_user_u64(val, infop->entry);
871         infop->entry = val + infop->load_bias;
872     } else {
873         _regs->gpr[12] = infop->entry;  /* r12 set to global entry address */
874     }
875 #endif
876     _regs->nip = infop->entry;
877 }
878 
879 /* See linux kernel: arch/powerpc/include/asm/elf.h.  */
880 #define ELF_NREG 48
881 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
882 
883 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUPPCState *env)
884 {
885     int i;
886     target_ulong ccr = 0;
887 
888     for (i = 0; i < ARRAY_SIZE(env->gpr); i++) {
889         (*regs)[i] = tswapreg(env->gpr[i]);
890     }
891 
892     (*regs)[32] = tswapreg(env->nip);
893     (*regs)[33] = tswapreg(env->msr);
894     (*regs)[35] = tswapreg(env->ctr);
895     (*regs)[36] = tswapreg(env->lr);
896     (*regs)[37] = tswapreg(env->xer);
897 
898     for (i = 0; i < ARRAY_SIZE(env->crf); i++) {
899         ccr |= env->crf[i] << (32 - ((i + 1) * 4));
900     }
901     (*regs)[38] = tswapreg(ccr);
902 }
903 
904 #define USE_ELF_CORE_DUMP
905 #define ELF_EXEC_PAGESIZE       4096
906 
907 #endif
908 
909 #ifdef TARGET_MIPS
910 
911 #define ELF_START_MMAP 0x80000000
912 
913 #ifdef TARGET_MIPS64
914 #define ELF_CLASS   ELFCLASS64
915 #else
916 #define ELF_CLASS   ELFCLASS32
917 #endif
918 #define ELF_ARCH    EM_MIPS
919 
920 #define elf_check_arch(x) ((x) == EM_MIPS || (x) == EM_NANOMIPS)
921 
922 #ifdef TARGET_ABI_MIPSN32
923 #define elf_check_abi(x) ((x) & EF_MIPS_ABI2)
924 #else
925 #define elf_check_abi(x) (!((x) & EF_MIPS_ABI2))
926 #endif
927 
928 static inline void init_thread(struct target_pt_regs *regs,
929                                struct image_info *infop)
930 {
931     regs->cp0_status = 2 << CP0St_KSU;
932     regs->cp0_epc = infop->entry;
933     regs->regs[29] = infop->start_stack;
934 }
935 
936 /* See linux kernel: arch/mips/include/asm/elf.h.  */
937 #define ELF_NREG 45
938 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
939 
940 /* See linux kernel: arch/mips/include/asm/reg.h.  */
941 enum {
942 #ifdef TARGET_MIPS64
943     TARGET_EF_R0 = 0,
944 #else
945     TARGET_EF_R0 = 6,
946 #endif
947     TARGET_EF_R26 = TARGET_EF_R0 + 26,
948     TARGET_EF_R27 = TARGET_EF_R0 + 27,
949     TARGET_EF_LO = TARGET_EF_R0 + 32,
950     TARGET_EF_HI = TARGET_EF_R0 + 33,
951     TARGET_EF_CP0_EPC = TARGET_EF_R0 + 34,
952     TARGET_EF_CP0_BADVADDR = TARGET_EF_R0 + 35,
953     TARGET_EF_CP0_STATUS = TARGET_EF_R0 + 36,
954     TARGET_EF_CP0_CAUSE = TARGET_EF_R0 + 37
955 };
956 
957 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs.  */
958 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMIPSState *env)
959 {
960     int i;
961 
962     for (i = 0; i < TARGET_EF_R0; i++) {
963         (*regs)[i] = 0;
964     }
965     (*regs)[TARGET_EF_R0] = 0;
966 
967     for (i = 1; i < ARRAY_SIZE(env->active_tc.gpr); i++) {
968         (*regs)[TARGET_EF_R0 + i] = tswapreg(env->active_tc.gpr[i]);
969     }
970 
971     (*regs)[TARGET_EF_R26] = 0;
972     (*regs)[TARGET_EF_R27] = 0;
973     (*regs)[TARGET_EF_LO] = tswapreg(env->active_tc.LO[0]);
974     (*regs)[TARGET_EF_HI] = tswapreg(env->active_tc.HI[0]);
975     (*regs)[TARGET_EF_CP0_EPC] = tswapreg(env->active_tc.PC);
976     (*regs)[TARGET_EF_CP0_BADVADDR] = tswapreg(env->CP0_BadVAddr);
977     (*regs)[TARGET_EF_CP0_STATUS] = tswapreg(env->CP0_Status);
978     (*regs)[TARGET_EF_CP0_CAUSE] = tswapreg(env->CP0_Cause);
979 }
980 
981 #define USE_ELF_CORE_DUMP
982 #define ELF_EXEC_PAGESIZE        4096
983 
984 /* See arch/mips/include/uapi/asm/hwcap.h.  */
985 enum {
986     HWCAP_MIPS_R6           = (1 << 0),
987     HWCAP_MIPS_MSA          = (1 << 1),
988 };
989 
990 #define ELF_HWCAP get_elf_hwcap()
991 
992 static uint32_t get_elf_hwcap(void)
993 {
994     MIPSCPU *cpu = MIPS_CPU(thread_cpu);
995     uint32_t hwcaps = 0;
996 
997 #define GET_FEATURE(flag, hwcap) \
998     do { if (cpu->env.insn_flags & (flag)) { hwcaps |= hwcap; } } while (0)
999 
1000     GET_FEATURE(ISA_MIPS32R6 | ISA_MIPS64R6, HWCAP_MIPS_R6);
1001     GET_FEATURE(ASE_MSA, HWCAP_MIPS_MSA);
1002 
1003 #undef GET_FEATURE
1004 
1005     return hwcaps;
1006 }
1007 
1008 #endif /* TARGET_MIPS */
1009 
1010 #ifdef TARGET_MICROBLAZE
1011 
1012 #define ELF_START_MMAP 0x80000000
1013 
1014 #define elf_check_arch(x) ( (x) == EM_MICROBLAZE || (x) == EM_MICROBLAZE_OLD)
1015 
1016 #define ELF_CLASS   ELFCLASS32
1017 #define ELF_ARCH    EM_MICROBLAZE
1018 
1019 static inline void init_thread(struct target_pt_regs *regs,
1020                                struct image_info *infop)
1021 {
1022     regs->pc = infop->entry;
1023     regs->r1 = infop->start_stack;
1024 
1025 }
1026 
1027 #define ELF_EXEC_PAGESIZE        4096
1028 
1029 #define USE_ELF_CORE_DUMP
1030 #define ELF_NREG 38
1031 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1032 
1033 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs.  */
1034 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMBState *env)
1035 {
1036     int i, pos = 0;
1037 
1038     for (i = 0; i < 32; i++) {
1039         (*regs)[pos++] = tswapreg(env->regs[i]);
1040     }
1041 
1042     (*regs)[pos++] = tswapreg(env->pc);
1043     (*regs)[pos++] = tswapreg(mb_cpu_read_msr(env));
1044     (*regs)[pos++] = 0;
1045     (*regs)[pos++] = tswapreg(env->ear);
1046     (*regs)[pos++] = 0;
1047     (*regs)[pos++] = tswapreg(env->esr);
1048 }
1049 
1050 #endif /* TARGET_MICROBLAZE */
1051 
1052 #ifdef TARGET_NIOS2
1053 
1054 #define ELF_START_MMAP 0x80000000
1055 
1056 #define elf_check_arch(x) ((x) == EM_ALTERA_NIOS2)
1057 
1058 #define ELF_CLASS   ELFCLASS32
1059 #define ELF_ARCH    EM_ALTERA_NIOS2
1060 
1061 static void init_thread(struct target_pt_regs *regs, struct image_info *infop)
1062 {
1063     regs->ea = infop->entry;
1064     regs->sp = infop->start_stack;
1065     regs->estatus = 0x3;
1066 }
1067 
1068 #define ELF_EXEC_PAGESIZE        4096
1069 
1070 #define USE_ELF_CORE_DUMP
1071 #define ELF_NREG 49
1072 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1073 
1074 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs.  */
1075 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1076                                const CPUNios2State *env)
1077 {
1078     int i;
1079 
1080     (*regs)[0] = -1;
1081     for (i = 1; i < 8; i++)    /* r0-r7 */
1082         (*regs)[i] = tswapreg(env->regs[i + 7]);
1083 
1084     for (i = 8; i < 16; i++)   /* r8-r15 */
1085         (*regs)[i] = tswapreg(env->regs[i - 8]);
1086 
1087     for (i = 16; i < 24; i++)  /* r16-r23 */
1088         (*regs)[i] = tswapreg(env->regs[i + 7]);
1089     (*regs)[24] = -1;    /* R_ET */
1090     (*regs)[25] = -1;    /* R_BT */
1091     (*regs)[26] = tswapreg(env->regs[R_GP]);
1092     (*regs)[27] = tswapreg(env->regs[R_SP]);
1093     (*regs)[28] = tswapreg(env->regs[R_FP]);
1094     (*regs)[29] = tswapreg(env->regs[R_EA]);
1095     (*regs)[30] = -1;    /* R_SSTATUS */
1096     (*regs)[31] = tswapreg(env->regs[R_RA]);
1097 
1098     (*regs)[32] = tswapreg(env->regs[R_PC]);
1099 
1100     (*regs)[33] = -1; /* R_STATUS */
1101     (*regs)[34] = tswapreg(env->regs[CR_ESTATUS]);
1102 
1103     for (i = 35; i < 49; i++)    /* ... */
1104         (*regs)[i] = -1;
1105 }
1106 
1107 #endif /* TARGET_NIOS2 */
1108 
1109 #ifdef TARGET_OPENRISC
1110 
1111 #define ELF_START_MMAP 0x08000000
1112 
1113 #define ELF_ARCH EM_OPENRISC
1114 #define ELF_CLASS ELFCLASS32
1115 #define ELF_DATA  ELFDATA2MSB
1116 
1117 static inline void init_thread(struct target_pt_regs *regs,
1118                                struct image_info *infop)
1119 {
1120     regs->pc = infop->entry;
1121     regs->gpr[1] = infop->start_stack;
1122 }
1123 
1124 #define USE_ELF_CORE_DUMP
1125 #define ELF_EXEC_PAGESIZE 8192
1126 
1127 /* See linux kernel arch/openrisc/include/asm/elf.h.  */
1128 #define ELF_NREG 34 /* gprs and pc, sr */
1129 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1130 
1131 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1132                                const CPUOpenRISCState *env)
1133 {
1134     int i;
1135 
1136     for (i = 0; i < 32; i++) {
1137         (*regs)[i] = tswapreg(cpu_get_gpr(env, i));
1138     }
1139     (*regs)[32] = tswapreg(env->pc);
1140     (*regs)[33] = tswapreg(cpu_get_sr(env));
1141 }
1142 #define ELF_HWCAP 0
1143 #define ELF_PLATFORM NULL
1144 
1145 #endif /* TARGET_OPENRISC */
1146 
1147 #ifdef TARGET_SH4
1148 
1149 #define ELF_START_MMAP 0x80000000
1150 
1151 #define ELF_CLASS ELFCLASS32
1152 #define ELF_ARCH  EM_SH
1153 
1154 static inline void init_thread(struct target_pt_regs *regs,
1155                                struct image_info *infop)
1156 {
1157     /* Check other registers XXXXX */
1158     regs->pc = infop->entry;
1159     regs->regs[15] = infop->start_stack;
1160 }
1161 
1162 /* See linux kernel: arch/sh/include/asm/elf.h.  */
1163 #define ELF_NREG 23
1164 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1165 
1166 /* See linux kernel: arch/sh/include/asm/ptrace.h.  */
1167 enum {
1168     TARGET_REG_PC = 16,
1169     TARGET_REG_PR = 17,
1170     TARGET_REG_SR = 18,
1171     TARGET_REG_GBR = 19,
1172     TARGET_REG_MACH = 20,
1173     TARGET_REG_MACL = 21,
1174     TARGET_REG_SYSCALL = 22
1175 };
1176 
1177 static inline void elf_core_copy_regs(target_elf_gregset_t *regs,
1178                                       const CPUSH4State *env)
1179 {
1180     int i;
1181 
1182     for (i = 0; i < 16; i++) {
1183         (*regs)[i] = tswapreg(env->gregs[i]);
1184     }
1185 
1186     (*regs)[TARGET_REG_PC] = tswapreg(env->pc);
1187     (*regs)[TARGET_REG_PR] = tswapreg(env->pr);
1188     (*regs)[TARGET_REG_SR] = tswapreg(env->sr);
1189     (*regs)[TARGET_REG_GBR] = tswapreg(env->gbr);
1190     (*regs)[TARGET_REG_MACH] = tswapreg(env->mach);
1191     (*regs)[TARGET_REG_MACL] = tswapreg(env->macl);
1192     (*regs)[TARGET_REG_SYSCALL] = 0; /* FIXME */
1193 }
1194 
1195 #define USE_ELF_CORE_DUMP
1196 #define ELF_EXEC_PAGESIZE        4096
1197 
1198 enum {
1199     SH_CPU_HAS_FPU            = 0x0001, /* Hardware FPU support */
1200     SH_CPU_HAS_P2_FLUSH_BUG   = 0x0002, /* Need to flush the cache in P2 area */
1201     SH_CPU_HAS_MMU_PAGE_ASSOC = 0x0004, /* SH3: TLB way selection bit support */
1202     SH_CPU_HAS_DSP            = 0x0008, /* SH-DSP: DSP support */
1203     SH_CPU_HAS_PERF_COUNTER   = 0x0010, /* Hardware performance counters */
1204     SH_CPU_HAS_PTEA           = 0x0020, /* PTEA register */
1205     SH_CPU_HAS_LLSC           = 0x0040, /* movli.l/movco.l */
1206     SH_CPU_HAS_L2_CACHE       = 0x0080, /* Secondary cache / URAM */
1207     SH_CPU_HAS_OP32           = 0x0100, /* 32-bit instruction support */
1208     SH_CPU_HAS_PTEAEX         = 0x0200, /* PTE ASID Extension support */
1209 };
1210 
1211 #define ELF_HWCAP get_elf_hwcap()
1212 
1213 static uint32_t get_elf_hwcap(void)
1214 {
1215     SuperHCPU *cpu = SUPERH_CPU(thread_cpu);
1216     uint32_t hwcap = 0;
1217 
1218     hwcap |= SH_CPU_HAS_FPU;
1219 
1220     if (cpu->env.features & SH_FEATURE_SH4A) {
1221         hwcap |= SH_CPU_HAS_LLSC;
1222     }
1223 
1224     return hwcap;
1225 }
1226 
1227 #endif
1228 
1229 #ifdef TARGET_CRIS
1230 
1231 #define ELF_START_MMAP 0x80000000
1232 
1233 #define ELF_CLASS ELFCLASS32
1234 #define ELF_ARCH  EM_CRIS
1235 
1236 static inline void init_thread(struct target_pt_regs *regs,
1237                                struct image_info *infop)
1238 {
1239     regs->erp = infop->entry;
1240 }
1241 
1242 #define ELF_EXEC_PAGESIZE        8192
1243 
1244 #endif
1245 
1246 #ifdef TARGET_M68K
1247 
1248 #define ELF_START_MMAP 0x80000000
1249 
1250 #define ELF_CLASS       ELFCLASS32
1251 #define ELF_ARCH        EM_68K
1252 
1253 /* ??? Does this need to do anything?
1254    #define ELF_PLAT_INIT(_r) */
1255 
1256 static inline void init_thread(struct target_pt_regs *regs,
1257                                struct image_info *infop)
1258 {
1259     regs->usp = infop->start_stack;
1260     regs->sr = 0;
1261     regs->pc = infop->entry;
1262 }
1263 
1264 /* See linux kernel: arch/m68k/include/asm/elf.h.  */
1265 #define ELF_NREG 20
1266 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1267 
1268 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUM68KState *env)
1269 {
1270     (*regs)[0] = tswapreg(env->dregs[1]);
1271     (*regs)[1] = tswapreg(env->dregs[2]);
1272     (*regs)[2] = tswapreg(env->dregs[3]);
1273     (*regs)[3] = tswapreg(env->dregs[4]);
1274     (*regs)[4] = tswapreg(env->dregs[5]);
1275     (*regs)[5] = tswapreg(env->dregs[6]);
1276     (*regs)[6] = tswapreg(env->dregs[7]);
1277     (*regs)[7] = tswapreg(env->aregs[0]);
1278     (*regs)[8] = tswapreg(env->aregs[1]);
1279     (*regs)[9] = tswapreg(env->aregs[2]);
1280     (*regs)[10] = tswapreg(env->aregs[3]);
1281     (*regs)[11] = tswapreg(env->aregs[4]);
1282     (*regs)[12] = tswapreg(env->aregs[5]);
1283     (*regs)[13] = tswapreg(env->aregs[6]);
1284     (*regs)[14] = tswapreg(env->dregs[0]);
1285     (*regs)[15] = tswapreg(env->aregs[7]);
1286     (*regs)[16] = tswapreg(env->dregs[0]); /* FIXME: orig_d0 */
1287     (*regs)[17] = tswapreg(env->sr);
1288     (*regs)[18] = tswapreg(env->pc);
1289     (*regs)[19] = 0;  /* FIXME: regs->format | regs->vector */
1290 }
1291 
1292 #define USE_ELF_CORE_DUMP
1293 #define ELF_EXEC_PAGESIZE       8192
1294 
1295 #endif
1296 
1297 #ifdef TARGET_ALPHA
1298 
1299 #define ELF_START_MMAP (0x30000000000ULL)
1300 
1301 #define ELF_CLASS      ELFCLASS64
1302 #define ELF_ARCH       EM_ALPHA
1303 
1304 static inline void init_thread(struct target_pt_regs *regs,
1305                                struct image_info *infop)
1306 {
1307     regs->pc = infop->entry;
1308     regs->ps = 8;
1309     regs->usp = infop->start_stack;
1310 }
1311 
1312 #define ELF_EXEC_PAGESIZE        8192
1313 
1314 #endif /* TARGET_ALPHA */
1315 
1316 #ifdef TARGET_S390X
1317 
1318 #define ELF_START_MMAP (0x20000000000ULL)
1319 
1320 #define ELF_CLASS	ELFCLASS64
1321 #define ELF_DATA	ELFDATA2MSB
1322 #define ELF_ARCH	EM_S390
1323 
1324 #include "elf.h"
1325 
1326 #define ELF_HWCAP get_elf_hwcap()
1327 
1328 #define GET_FEATURE(_feat, _hwcap) \
1329     do { if (s390_has_feat(_feat)) { hwcap |= _hwcap; } } while (0)
1330 
1331 static uint32_t get_elf_hwcap(void)
1332 {
1333     /*
1334      * Let's assume we always have esan3 and zarch.
1335      * 31-bit processes can use 64-bit registers (high gprs).
1336      */
1337     uint32_t hwcap = HWCAP_S390_ESAN3 | HWCAP_S390_ZARCH | HWCAP_S390_HIGH_GPRS;
1338 
1339     GET_FEATURE(S390_FEAT_STFLE, HWCAP_S390_STFLE);
1340     GET_FEATURE(S390_FEAT_MSA, HWCAP_S390_MSA);
1341     GET_FEATURE(S390_FEAT_LONG_DISPLACEMENT, HWCAP_S390_LDISP);
1342     GET_FEATURE(S390_FEAT_EXTENDED_IMMEDIATE, HWCAP_S390_EIMM);
1343     if (s390_has_feat(S390_FEAT_EXTENDED_TRANSLATION_3) &&
1344         s390_has_feat(S390_FEAT_ETF3_ENH)) {
1345         hwcap |= HWCAP_S390_ETF3EH;
1346     }
1347     GET_FEATURE(S390_FEAT_VECTOR, HWCAP_S390_VXRS);
1348 
1349     return hwcap;
1350 }
1351 
1352 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop)
1353 {
1354     regs->psw.addr = infop->entry;
1355     regs->psw.mask = PSW_MASK_64 | PSW_MASK_32;
1356     regs->gprs[15] = infop->start_stack;
1357 }
1358 
1359 #endif /* TARGET_S390X */
1360 
1361 #ifdef TARGET_TILEGX
1362 
1363 /* 42 bits real used address, a half for user mode */
1364 #define ELF_START_MMAP (0x00000020000000000ULL)
1365 
1366 #define elf_check_arch(x) ((x) == EM_TILEGX)
1367 
1368 #define ELF_CLASS   ELFCLASS64
1369 #define ELF_DATA    ELFDATA2LSB
1370 #define ELF_ARCH    EM_TILEGX
1371 
1372 static inline void init_thread(struct target_pt_regs *regs,
1373                                struct image_info *infop)
1374 {
1375     regs->pc = infop->entry;
1376     regs->sp = infop->start_stack;
1377 
1378 }
1379 
1380 #define ELF_EXEC_PAGESIZE        65536 /* TILE-Gx page size is 64KB */
1381 
1382 #endif /* TARGET_TILEGX */
1383 
1384 #ifdef TARGET_RISCV
1385 
1386 #define ELF_START_MMAP 0x80000000
1387 #define ELF_ARCH  EM_RISCV
1388 
1389 #ifdef TARGET_RISCV32
1390 #define ELF_CLASS ELFCLASS32
1391 #else
1392 #define ELF_CLASS ELFCLASS64
1393 #endif
1394 
1395 static inline void init_thread(struct target_pt_regs *regs,
1396                                struct image_info *infop)
1397 {
1398     regs->sepc = infop->entry;
1399     regs->sp = infop->start_stack;
1400 }
1401 
1402 #define ELF_EXEC_PAGESIZE 4096
1403 
1404 #endif /* TARGET_RISCV */
1405 
1406 #ifdef TARGET_HPPA
1407 
1408 #define ELF_START_MMAP  0x80000000
1409 #define ELF_CLASS       ELFCLASS32
1410 #define ELF_ARCH        EM_PARISC
1411 #define ELF_PLATFORM    "PARISC"
1412 #define STACK_GROWS_DOWN 0
1413 #define STACK_ALIGNMENT  64
1414 
1415 static inline void init_thread(struct target_pt_regs *regs,
1416                                struct image_info *infop)
1417 {
1418     regs->iaoq[0] = infop->entry;
1419     regs->iaoq[1] = infop->entry + 4;
1420     regs->gr[23] = 0;
1421     regs->gr[24] = infop->arg_start;
1422     regs->gr[25] = (infop->arg_end - infop->arg_start) / sizeof(abi_ulong);
1423     /* The top-of-stack contains a linkage buffer.  */
1424     regs->gr[30] = infop->start_stack + 64;
1425     regs->gr[31] = infop->entry;
1426 }
1427 
1428 #endif /* TARGET_HPPA */
1429 
1430 #ifdef TARGET_XTENSA
1431 
1432 #define ELF_START_MMAP 0x20000000
1433 
1434 #define ELF_CLASS       ELFCLASS32
1435 #define ELF_ARCH        EM_XTENSA
1436 
1437 static inline void init_thread(struct target_pt_regs *regs,
1438                                struct image_info *infop)
1439 {
1440     regs->windowbase = 0;
1441     regs->windowstart = 1;
1442     regs->areg[1] = infop->start_stack;
1443     regs->pc = infop->entry;
1444 }
1445 
1446 /* See linux kernel: arch/xtensa/include/asm/elf.h.  */
1447 #define ELF_NREG 128
1448 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1449 
1450 enum {
1451     TARGET_REG_PC,
1452     TARGET_REG_PS,
1453     TARGET_REG_LBEG,
1454     TARGET_REG_LEND,
1455     TARGET_REG_LCOUNT,
1456     TARGET_REG_SAR,
1457     TARGET_REG_WINDOWSTART,
1458     TARGET_REG_WINDOWBASE,
1459     TARGET_REG_THREADPTR,
1460     TARGET_REG_AR0 = 64,
1461 };
1462 
1463 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1464                                const CPUXtensaState *env)
1465 {
1466     unsigned i;
1467 
1468     (*regs)[TARGET_REG_PC] = tswapreg(env->pc);
1469     (*regs)[TARGET_REG_PS] = tswapreg(env->sregs[PS] & ~PS_EXCM);
1470     (*regs)[TARGET_REG_LBEG] = tswapreg(env->sregs[LBEG]);
1471     (*regs)[TARGET_REG_LEND] = tswapreg(env->sregs[LEND]);
1472     (*regs)[TARGET_REG_LCOUNT] = tswapreg(env->sregs[LCOUNT]);
1473     (*regs)[TARGET_REG_SAR] = tswapreg(env->sregs[SAR]);
1474     (*regs)[TARGET_REG_WINDOWSTART] = tswapreg(env->sregs[WINDOW_START]);
1475     (*regs)[TARGET_REG_WINDOWBASE] = tswapreg(env->sregs[WINDOW_BASE]);
1476     (*regs)[TARGET_REG_THREADPTR] = tswapreg(env->uregs[THREADPTR]);
1477     xtensa_sync_phys_from_window((CPUXtensaState *)env);
1478     for (i = 0; i < env->config->nareg; ++i) {
1479         (*regs)[TARGET_REG_AR0 + i] = tswapreg(env->phys_regs[i]);
1480     }
1481 }
1482 
1483 #define USE_ELF_CORE_DUMP
1484 #define ELF_EXEC_PAGESIZE       4096
1485 
1486 #endif /* TARGET_XTENSA */
1487 
1488 #ifndef ELF_PLATFORM
1489 #define ELF_PLATFORM (NULL)
1490 #endif
1491 
1492 #ifndef ELF_MACHINE
1493 #define ELF_MACHINE ELF_ARCH
1494 #endif
1495 
1496 #ifndef elf_check_arch
1497 #define elf_check_arch(x) ((x) == ELF_ARCH)
1498 #endif
1499 
1500 #ifndef elf_check_abi
1501 #define elf_check_abi(x) (1)
1502 #endif
1503 
1504 #ifndef ELF_HWCAP
1505 #define ELF_HWCAP 0
1506 #endif
1507 
1508 #ifndef STACK_GROWS_DOWN
1509 #define STACK_GROWS_DOWN 1
1510 #endif
1511 
1512 #ifndef STACK_ALIGNMENT
1513 #define STACK_ALIGNMENT 16
1514 #endif
1515 
1516 #ifdef TARGET_ABI32
1517 #undef ELF_CLASS
1518 #define ELF_CLASS ELFCLASS32
1519 #undef bswaptls
1520 #define bswaptls(ptr) bswap32s(ptr)
1521 #endif
1522 
1523 #include "elf.h"
1524 
1525 /* We must delay the following stanzas until after "elf.h". */
1526 #if defined(TARGET_AARCH64)
1527 
1528 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz,
1529                                     const uint32_t *data,
1530                                     struct image_info *info,
1531                                     Error **errp)
1532 {
1533     if (pr_type == GNU_PROPERTY_AARCH64_FEATURE_1_AND) {
1534         if (pr_datasz != sizeof(uint32_t)) {
1535             error_setg(errp, "Ill-formed GNU_PROPERTY_AARCH64_FEATURE_1_AND");
1536             return false;
1537         }
1538         /* We will extract GNU_PROPERTY_AARCH64_FEATURE_1_BTI later. */
1539         info->note_flags = *data;
1540     }
1541     return true;
1542 }
1543 #define ARCH_USE_GNU_PROPERTY 1
1544 
1545 #else
1546 
1547 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz,
1548                                     const uint32_t *data,
1549                                     struct image_info *info,
1550                                     Error **errp)
1551 {
1552     g_assert_not_reached();
1553 }
1554 #define ARCH_USE_GNU_PROPERTY 0
1555 
1556 #endif
1557 
1558 struct exec
1559 {
1560     unsigned int a_info;   /* Use macros N_MAGIC, etc for access */
1561     unsigned int a_text;   /* length of text, in bytes */
1562     unsigned int a_data;   /* length of data, in bytes */
1563     unsigned int a_bss;    /* length of uninitialized data area, in bytes */
1564     unsigned int a_syms;   /* length of symbol table data in file, in bytes */
1565     unsigned int a_entry;  /* start address */
1566     unsigned int a_trsize; /* length of relocation info for text, in bytes */
1567     unsigned int a_drsize; /* length of relocation info for data, in bytes */
1568 };
1569 
1570 
1571 #define N_MAGIC(exec) ((exec).a_info & 0xffff)
1572 #define OMAGIC 0407
1573 #define NMAGIC 0410
1574 #define ZMAGIC 0413
1575 #define QMAGIC 0314
1576 
1577 /* Necessary parameters */
1578 #define TARGET_ELF_EXEC_PAGESIZE \
1579         (((eppnt->p_align & ~qemu_host_page_mask) != 0) ? \
1580          TARGET_PAGE_SIZE : MAX(qemu_host_page_size, TARGET_PAGE_SIZE))
1581 #define TARGET_ELF_PAGELENGTH(_v) ROUND_UP((_v), TARGET_ELF_EXEC_PAGESIZE)
1582 #define TARGET_ELF_PAGESTART(_v) ((_v) & \
1583                                  ~(abi_ulong)(TARGET_ELF_EXEC_PAGESIZE-1))
1584 #define TARGET_ELF_PAGEOFFSET(_v) ((_v) & (TARGET_ELF_EXEC_PAGESIZE-1))
1585 
1586 #define DLINFO_ITEMS 16
1587 
1588 static inline void memcpy_fromfs(void * to, const void * from, unsigned long n)
1589 {
1590     memcpy(to, from, n);
1591 }
1592 
1593 #ifdef BSWAP_NEEDED
1594 static void bswap_ehdr(struct elfhdr *ehdr)
1595 {
1596     bswap16s(&ehdr->e_type);            /* Object file type */
1597     bswap16s(&ehdr->e_machine);         /* Architecture */
1598     bswap32s(&ehdr->e_version);         /* Object file version */
1599     bswaptls(&ehdr->e_entry);           /* Entry point virtual address */
1600     bswaptls(&ehdr->e_phoff);           /* Program header table file offset */
1601     bswaptls(&ehdr->e_shoff);           /* Section header table file offset */
1602     bswap32s(&ehdr->e_flags);           /* Processor-specific flags */
1603     bswap16s(&ehdr->e_ehsize);          /* ELF header size in bytes */
1604     bswap16s(&ehdr->e_phentsize);       /* Program header table entry size */
1605     bswap16s(&ehdr->e_phnum);           /* Program header table entry count */
1606     bswap16s(&ehdr->e_shentsize);       /* Section header table entry size */
1607     bswap16s(&ehdr->e_shnum);           /* Section header table entry count */
1608     bswap16s(&ehdr->e_shstrndx);        /* Section header string table index */
1609 }
1610 
1611 static void bswap_phdr(struct elf_phdr *phdr, int phnum)
1612 {
1613     int i;
1614     for (i = 0; i < phnum; ++i, ++phdr) {
1615         bswap32s(&phdr->p_type);        /* Segment type */
1616         bswap32s(&phdr->p_flags);       /* Segment flags */
1617         bswaptls(&phdr->p_offset);      /* Segment file offset */
1618         bswaptls(&phdr->p_vaddr);       /* Segment virtual address */
1619         bswaptls(&phdr->p_paddr);       /* Segment physical address */
1620         bswaptls(&phdr->p_filesz);      /* Segment size in file */
1621         bswaptls(&phdr->p_memsz);       /* Segment size in memory */
1622         bswaptls(&phdr->p_align);       /* Segment alignment */
1623     }
1624 }
1625 
1626 static void bswap_shdr(struct elf_shdr *shdr, int shnum)
1627 {
1628     int i;
1629     for (i = 0; i < shnum; ++i, ++shdr) {
1630         bswap32s(&shdr->sh_name);
1631         bswap32s(&shdr->sh_type);
1632         bswaptls(&shdr->sh_flags);
1633         bswaptls(&shdr->sh_addr);
1634         bswaptls(&shdr->sh_offset);
1635         bswaptls(&shdr->sh_size);
1636         bswap32s(&shdr->sh_link);
1637         bswap32s(&shdr->sh_info);
1638         bswaptls(&shdr->sh_addralign);
1639         bswaptls(&shdr->sh_entsize);
1640     }
1641 }
1642 
1643 static void bswap_sym(struct elf_sym *sym)
1644 {
1645     bswap32s(&sym->st_name);
1646     bswaptls(&sym->st_value);
1647     bswaptls(&sym->st_size);
1648     bswap16s(&sym->st_shndx);
1649 }
1650 
1651 #ifdef TARGET_MIPS
1652 static void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags)
1653 {
1654     bswap16s(&abiflags->version);
1655     bswap32s(&abiflags->ases);
1656     bswap32s(&abiflags->isa_ext);
1657     bswap32s(&abiflags->flags1);
1658     bswap32s(&abiflags->flags2);
1659 }
1660 #endif
1661 #else
1662 static inline void bswap_ehdr(struct elfhdr *ehdr) { }
1663 static inline void bswap_phdr(struct elf_phdr *phdr, int phnum) { }
1664 static inline void bswap_shdr(struct elf_shdr *shdr, int shnum) { }
1665 static inline void bswap_sym(struct elf_sym *sym) { }
1666 #ifdef TARGET_MIPS
1667 static inline void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags) { }
1668 #endif
1669 #endif
1670 
1671 #ifdef USE_ELF_CORE_DUMP
1672 static int elf_core_dump(int, const CPUArchState *);
1673 #endif /* USE_ELF_CORE_DUMP */
1674 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias);
1675 
1676 /* Verify the portions of EHDR within E_IDENT for the target.
1677    This can be performed before bswapping the entire header.  */
1678 static bool elf_check_ident(struct elfhdr *ehdr)
1679 {
1680     return (ehdr->e_ident[EI_MAG0] == ELFMAG0
1681             && ehdr->e_ident[EI_MAG1] == ELFMAG1
1682             && ehdr->e_ident[EI_MAG2] == ELFMAG2
1683             && ehdr->e_ident[EI_MAG3] == ELFMAG3
1684             && ehdr->e_ident[EI_CLASS] == ELF_CLASS
1685             && ehdr->e_ident[EI_DATA] == ELF_DATA
1686             && ehdr->e_ident[EI_VERSION] == EV_CURRENT);
1687 }
1688 
1689 /* Verify the portions of EHDR outside of E_IDENT for the target.
1690    This has to wait until after bswapping the header.  */
1691 static bool elf_check_ehdr(struct elfhdr *ehdr)
1692 {
1693     return (elf_check_arch(ehdr->e_machine)
1694             && elf_check_abi(ehdr->e_flags)
1695             && ehdr->e_ehsize == sizeof(struct elfhdr)
1696             && ehdr->e_phentsize == sizeof(struct elf_phdr)
1697             && (ehdr->e_type == ET_EXEC || ehdr->e_type == ET_DYN));
1698 }
1699 
1700 /*
1701  * 'copy_elf_strings()' copies argument/envelope strings from user
1702  * memory to free pages in kernel mem. These are in a format ready
1703  * to be put directly into the top of new user memory.
1704  *
1705  */
1706 static abi_ulong copy_elf_strings(int argc, char **argv, char *scratch,
1707                                   abi_ulong p, abi_ulong stack_limit)
1708 {
1709     char *tmp;
1710     int len, i;
1711     abi_ulong top = p;
1712 
1713     if (!p) {
1714         return 0;       /* bullet-proofing */
1715     }
1716 
1717     if (STACK_GROWS_DOWN) {
1718         int offset = ((p - 1) % TARGET_PAGE_SIZE) + 1;
1719         for (i = argc - 1; i >= 0; --i) {
1720             tmp = argv[i];
1721             if (!tmp) {
1722                 fprintf(stderr, "VFS: argc is wrong");
1723                 exit(-1);
1724             }
1725             len = strlen(tmp) + 1;
1726             tmp += len;
1727 
1728             if (len > (p - stack_limit)) {
1729                 return 0;
1730             }
1731             while (len) {
1732                 int bytes_to_copy = (len > offset) ? offset : len;
1733                 tmp -= bytes_to_copy;
1734                 p -= bytes_to_copy;
1735                 offset -= bytes_to_copy;
1736                 len -= bytes_to_copy;
1737 
1738                 memcpy_fromfs(scratch + offset, tmp, bytes_to_copy);
1739 
1740                 if (offset == 0) {
1741                     memcpy_to_target(p, scratch, top - p);
1742                     top = p;
1743                     offset = TARGET_PAGE_SIZE;
1744                 }
1745             }
1746         }
1747         if (p != top) {
1748             memcpy_to_target(p, scratch + offset, top - p);
1749         }
1750     } else {
1751         int remaining = TARGET_PAGE_SIZE - (p % TARGET_PAGE_SIZE);
1752         for (i = 0; i < argc; ++i) {
1753             tmp = argv[i];
1754             if (!tmp) {
1755                 fprintf(stderr, "VFS: argc is wrong");
1756                 exit(-1);
1757             }
1758             len = strlen(tmp) + 1;
1759             if (len > (stack_limit - p)) {
1760                 return 0;
1761             }
1762             while (len) {
1763                 int bytes_to_copy = (len > remaining) ? remaining : len;
1764 
1765                 memcpy_fromfs(scratch + (p - top), tmp, bytes_to_copy);
1766 
1767                 tmp += bytes_to_copy;
1768                 remaining -= bytes_to_copy;
1769                 p += bytes_to_copy;
1770                 len -= bytes_to_copy;
1771 
1772                 if (remaining == 0) {
1773                     memcpy_to_target(top, scratch, p - top);
1774                     top = p;
1775                     remaining = TARGET_PAGE_SIZE;
1776                 }
1777             }
1778         }
1779         if (p != top) {
1780             memcpy_to_target(top, scratch, p - top);
1781         }
1782     }
1783 
1784     return p;
1785 }
1786 
1787 /* Older linux kernels provide up to MAX_ARG_PAGES (default: 32) of
1788  * argument/environment space. Newer kernels (>2.6.33) allow more,
1789  * dependent on stack size, but guarantee at least 32 pages for
1790  * backwards compatibility.
1791  */
1792 #define STACK_LOWER_LIMIT (32 * TARGET_PAGE_SIZE)
1793 
1794 static abi_ulong setup_arg_pages(struct linux_binprm *bprm,
1795                                  struct image_info *info)
1796 {
1797     abi_ulong size, error, guard;
1798 
1799     size = guest_stack_size;
1800     if (size < STACK_LOWER_LIMIT) {
1801         size = STACK_LOWER_LIMIT;
1802     }
1803     guard = TARGET_PAGE_SIZE;
1804     if (guard < qemu_real_host_page_size) {
1805         guard = qemu_real_host_page_size;
1806     }
1807 
1808     error = target_mmap(0, size + guard, PROT_READ | PROT_WRITE,
1809                         MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
1810     if (error == -1) {
1811         perror("mmap stack");
1812         exit(-1);
1813     }
1814 
1815     /* We reserve one extra page at the top of the stack as guard.  */
1816     if (STACK_GROWS_DOWN) {
1817         target_mprotect(error, guard, PROT_NONE);
1818         info->stack_limit = error + guard;
1819         return info->stack_limit + size - sizeof(void *);
1820     } else {
1821         target_mprotect(error + size, guard, PROT_NONE);
1822         info->stack_limit = error + size;
1823         return error;
1824     }
1825 }
1826 
1827 /* Map and zero the bss.  We need to explicitly zero any fractional pages
1828    after the data section (i.e. bss).  */
1829 static void zero_bss(abi_ulong elf_bss, abi_ulong last_bss, int prot)
1830 {
1831     uintptr_t host_start, host_map_start, host_end;
1832 
1833     last_bss = TARGET_PAGE_ALIGN(last_bss);
1834 
1835     /* ??? There is confusion between qemu_real_host_page_size and
1836        qemu_host_page_size here and elsewhere in target_mmap, which
1837        may lead to the end of the data section mapping from the file
1838        not being mapped.  At least there was an explicit test and
1839        comment for that here, suggesting that "the file size must
1840        be known".  The comment probably pre-dates the introduction
1841        of the fstat system call in target_mmap which does in fact
1842        find out the size.  What isn't clear is if the workaround
1843        here is still actually needed.  For now, continue with it,
1844        but merge it with the "normal" mmap that would allocate the bss.  */
1845 
1846     host_start = (uintptr_t) g2h(elf_bss);
1847     host_end = (uintptr_t) g2h(last_bss);
1848     host_map_start = REAL_HOST_PAGE_ALIGN(host_start);
1849 
1850     if (host_map_start < host_end) {
1851         void *p = mmap((void *)host_map_start, host_end - host_map_start,
1852                        prot, MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
1853         if (p == MAP_FAILED) {
1854             perror("cannot mmap brk");
1855             exit(-1);
1856         }
1857     }
1858 
1859     /* Ensure that the bss page(s) are valid */
1860     if ((page_get_flags(last_bss-1) & prot) != prot) {
1861         page_set_flags(elf_bss & TARGET_PAGE_MASK, last_bss, prot | PAGE_VALID);
1862     }
1863 
1864     if (host_start < host_map_start) {
1865         memset((void *)host_start, 0, host_map_start - host_start);
1866     }
1867 }
1868 
1869 #ifdef TARGET_ARM
1870 static int elf_is_fdpic(struct elfhdr *exec)
1871 {
1872     return exec->e_ident[EI_OSABI] == ELFOSABI_ARM_FDPIC;
1873 }
1874 #else
1875 /* Default implementation, always false.  */
1876 static int elf_is_fdpic(struct elfhdr *exec)
1877 {
1878     return 0;
1879 }
1880 #endif
1881 
1882 static abi_ulong loader_build_fdpic_loadmap(struct image_info *info, abi_ulong sp)
1883 {
1884     uint16_t n;
1885     struct elf32_fdpic_loadseg *loadsegs = info->loadsegs;
1886 
1887     /* elf32_fdpic_loadseg */
1888     n = info->nsegs;
1889     while (n--) {
1890         sp -= 12;
1891         put_user_u32(loadsegs[n].addr, sp+0);
1892         put_user_u32(loadsegs[n].p_vaddr, sp+4);
1893         put_user_u32(loadsegs[n].p_memsz, sp+8);
1894     }
1895 
1896     /* elf32_fdpic_loadmap */
1897     sp -= 4;
1898     put_user_u16(0, sp+0); /* version */
1899     put_user_u16(info->nsegs, sp+2); /* nsegs */
1900 
1901     info->personality = PER_LINUX_FDPIC;
1902     info->loadmap_addr = sp;
1903 
1904     return sp;
1905 }
1906 
1907 static abi_ulong create_elf_tables(abi_ulong p, int argc, int envc,
1908                                    struct elfhdr *exec,
1909                                    struct image_info *info,
1910                                    struct image_info *interp_info)
1911 {
1912     abi_ulong sp;
1913     abi_ulong u_argc, u_argv, u_envp, u_auxv;
1914     int size;
1915     int i;
1916     abi_ulong u_rand_bytes;
1917     uint8_t k_rand_bytes[16];
1918     abi_ulong u_platform;
1919     const char *k_platform;
1920     const int n = sizeof(elf_addr_t);
1921 
1922     sp = p;
1923 
1924     /* Needs to be before we load the env/argc/... */
1925     if (elf_is_fdpic(exec)) {
1926         /* Need 4 byte alignment for these structs */
1927         sp &= ~3;
1928         sp = loader_build_fdpic_loadmap(info, sp);
1929         info->other_info = interp_info;
1930         if (interp_info) {
1931             interp_info->other_info = info;
1932             sp = loader_build_fdpic_loadmap(interp_info, sp);
1933             info->interpreter_loadmap_addr = interp_info->loadmap_addr;
1934             info->interpreter_pt_dynamic_addr = interp_info->pt_dynamic_addr;
1935         } else {
1936             info->interpreter_loadmap_addr = 0;
1937             info->interpreter_pt_dynamic_addr = 0;
1938         }
1939     }
1940 
1941     u_platform = 0;
1942     k_platform = ELF_PLATFORM;
1943     if (k_platform) {
1944         size_t len = strlen(k_platform) + 1;
1945         if (STACK_GROWS_DOWN) {
1946             sp -= (len + n - 1) & ~(n - 1);
1947             u_platform = sp;
1948             /* FIXME - check return value of memcpy_to_target() for failure */
1949             memcpy_to_target(sp, k_platform, len);
1950         } else {
1951             memcpy_to_target(sp, k_platform, len);
1952             u_platform = sp;
1953             sp += len + 1;
1954         }
1955     }
1956 
1957     /* Provide 16 byte alignment for the PRNG, and basic alignment for
1958      * the argv and envp pointers.
1959      */
1960     if (STACK_GROWS_DOWN) {
1961         sp = QEMU_ALIGN_DOWN(sp, 16);
1962     } else {
1963         sp = QEMU_ALIGN_UP(sp, 16);
1964     }
1965 
1966     /*
1967      * Generate 16 random bytes for userspace PRNG seeding.
1968      */
1969     qemu_guest_getrandom_nofail(k_rand_bytes, sizeof(k_rand_bytes));
1970     if (STACK_GROWS_DOWN) {
1971         sp -= 16;
1972         u_rand_bytes = sp;
1973         /* FIXME - check return value of memcpy_to_target() for failure */
1974         memcpy_to_target(sp, k_rand_bytes, 16);
1975     } else {
1976         memcpy_to_target(sp, k_rand_bytes, 16);
1977         u_rand_bytes = sp;
1978         sp += 16;
1979     }
1980 
1981     size = (DLINFO_ITEMS + 1) * 2;
1982     if (k_platform)
1983         size += 2;
1984 #ifdef DLINFO_ARCH_ITEMS
1985     size += DLINFO_ARCH_ITEMS * 2;
1986 #endif
1987 #ifdef ELF_HWCAP2
1988     size += 2;
1989 #endif
1990     info->auxv_len = size * n;
1991 
1992     size += envc + argc + 2;
1993     size += 1;  /* argc itself */
1994     size *= n;
1995 
1996     /* Allocate space and finalize stack alignment for entry now.  */
1997     if (STACK_GROWS_DOWN) {
1998         u_argc = QEMU_ALIGN_DOWN(sp - size, STACK_ALIGNMENT);
1999         sp = u_argc;
2000     } else {
2001         u_argc = sp;
2002         sp = QEMU_ALIGN_UP(sp + size, STACK_ALIGNMENT);
2003     }
2004 
2005     u_argv = u_argc + n;
2006     u_envp = u_argv + (argc + 1) * n;
2007     u_auxv = u_envp + (envc + 1) * n;
2008     info->saved_auxv = u_auxv;
2009     info->arg_start = u_argv;
2010     info->arg_end = u_argv + argc * n;
2011 
2012     /* This is correct because Linux defines
2013      * elf_addr_t as Elf32_Off / Elf64_Off
2014      */
2015 #define NEW_AUX_ENT(id, val) do {               \
2016         put_user_ual(id, u_auxv);  u_auxv += n; \
2017         put_user_ual(val, u_auxv); u_auxv += n; \
2018     } while(0)
2019 
2020 #ifdef ARCH_DLINFO
2021     /*
2022      * ARCH_DLINFO must come first so platform specific code can enforce
2023      * special alignment requirements on the AUXV if necessary (eg. PPC).
2024      */
2025     ARCH_DLINFO;
2026 #endif
2027     /* There must be exactly DLINFO_ITEMS entries here, or the assert
2028      * on info->auxv_len will trigger.
2029      */
2030     NEW_AUX_ENT(AT_PHDR, (abi_ulong)(info->load_addr + exec->e_phoff));
2031     NEW_AUX_ENT(AT_PHENT, (abi_ulong)(sizeof (struct elf_phdr)));
2032     NEW_AUX_ENT(AT_PHNUM, (abi_ulong)(exec->e_phnum));
2033     if ((info->alignment & ~qemu_host_page_mask) != 0) {
2034         /* Target doesn't support host page size alignment */
2035         NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(TARGET_PAGE_SIZE));
2036     } else {
2037         NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(MAX(TARGET_PAGE_SIZE,
2038                                                qemu_host_page_size)));
2039     }
2040     NEW_AUX_ENT(AT_BASE, (abi_ulong)(interp_info ? interp_info->load_addr : 0));
2041     NEW_AUX_ENT(AT_FLAGS, (abi_ulong)0);
2042     NEW_AUX_ENT(AT_ENTRY, info->entry);
2043     NEW_AUX_ENT(AT_UID, (abi_ulong) getuid());
2044     NEW_AUX_ENT(AT_EUID, (abi_ulong) geteuid());
2045     NEW_AUX_ENT(AT_GID, (abi_ulong) getgid());
2046     NEW_AUX_ENT(AT_EGID, (abi_ulong) getegid());
2047     NEW_AUX_ENT(AT_HWCAP, (abi_ulong) ELF_HWCAP);
2048     NEW_AUX_ENT(AT_CLKTCK, (abi_ulong) sysconf(_SC_CLK_TCK));
2049     NEW_AUX_ENT(AT_RANDOM, (abi_ulong) u_rand_bytes);
2050     NEW_AUX_ENT(AT_SECURE, (abi_ulong) qemu_getauxval(AT_SECURE));
2051     NEW_AUX_ENT(AT_EXECFN, info->file_string);
2052 
2053 #ifdef ELF_HWCAP2
2054     NEW_AUX_ENT(AT_HWCAP2, (abi_ulong) ELF_HWCAP2);
2055 #endif
2056 
2057     if (u_platform) {
2058         NEW_AUX_ENT(AT_PLATFORM, u_platform);
2059     }
2060     NEW_AUX_ENT (AT_NULL, 0);
2061 #undef NEW_AUX_ENT
2062 
2063     /* Check that our initial calculation of the auxv length matches how much
2064      * we actually put into it.
2065      */
2066     assert(info->auxv_len == u_auxv - info->saved_auxv);
2067 
2068     put_user_ual(argc, u_argc);
2069 
2070     p = info->arg_strings;
2071     for (i = 0; i < argc; ++i) {
2072         put_user_ual(p, u_argv);
2073         u_argv += n;
2074         p += target_strlen(p) + 1;
2075     }
2076     put_user_ual(0, u_argv);
2077 
2078     p = info->env_strings;
2079     for (i = 0; i < envc; ++i) {
2080         put_user_ual(p, u_envp);
2081         u_envp += n;
2082         p += target_strlen(p) + 1;
2083     }
2084     put_user_ual(0, u_envp);
2085 
2086     return sp;
2087 }
2088 
2089 #ifndef ARM_COMMPAGE
2090 #define ARM_COMMPAGE 0
2091 #define init_guest_commpage() true
2092 #endif
2093 
2094 static void pgb_fail_in_use(const char *image_name)
2095 {
2096     error_report("%s: requires virtual address space that is in use "
2097                  "(omit the -B option or choose a different value)",
2098                  image_name);
2099     exit(EXIT_FAILURE);
2100 }
2101 
2102 static void pgb_have_guest_base(const char *image_name, abi_ulong guest_loaddr,
2103                                 abi_ulong guest_hiaddr, long align)
2104 {
2105     const int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE;
2106     void *addr, *test;
2107 
2108     if (!QEMU_IS_ALIGNED(guest_base, align)) {
2109         fprintf(stderr, "Requested guest base 0x%lx does not satisfy "
2110                 "host minimum alignment (0x%lx)\n",
2111                 guest_base, align);
2112         exit(EXIT_FAILURE);
2113     }
2114 
2115     /* Sanity check the guest binary. */
2116     if (reserved_va) {
2117         if (guest_hiaddr > reserved_va) {
2118             error_report("%s: requires more than reserved virtual "
2119                          "address space (0x%" PRIx64 " > 0x%lx)",
2120                          image_name, (uint64_t)guest_hiaddr, reserved_va);
2121             exit(EXIT_FAILURE);
2122         }
2123     } else {
2124 #if HOST_LONG_BITS < TARGET_ABI_BITS
2125         if ((guest_hiaddr - guest_base) > ~(uintptr_t)0) {
2126             error_report("%s: requires more virtual address space "
2127                          "than the host can provide (0x%" PRIx64 ")",
2128                          image_name, (uint64_t)guest_hiaddr - guest_base);
2129             exit(EXIT_FAILURE);
2130         }
2131 #endif
2132     }
2133 
2134     /*
2135      * Expand the allocation to the entire reserved_va.
2136      * Exclude the mmap_min_addr hole.
2137      */
2138     if (reserved_va) {
2139         guest_loaddr = (guest_base >= mmap_min_addr ? 0
2140                         : mmap_min_addr - guest_base);
2141         guest_hiaddr = reserved_va;
2142     }
2143 
2144     /* Reserve the address space for the binary, or reserved_va. */
2145     test = g2h(guest_loaddr);
2146     addr = mmap(test, guest_hiaddr - guest_loaddr, PROT_NONE, flags, -1, 0);
2147     if (test != addr) {
2148         pgb_fail_in_use(image_name);
2149     }
2150 }
2151 
2152 /**
2153  * pgd_find_hole_fallback: potential mmap address
2154  * @guest_size: size of available space
2155  * @brk: location of break
2156  * @align: memory alignment
2157  *
2158  * This is a fallback method for finding a hole in the host address
2159  * space if we don't have the benefit of being able to access
2160  * /proc/self/map. It can potentially take a very long time as we can
2161  * only dumbly iterate up the host address space seeing if the
2162  * allocation would work.
2163  */
2164 static uintptr_t pgd_find_hole_fallback(uintptr_t guest_size, uintptr_t brk,
2165                                         long align, uintptr_t offset)
2166 {
2167     uintptr_t base;
2168 
2169     /* Start (aligned) at the bottom and work our way up */
2170     base = ROUND_UP(mmap_min_addr, align);
2171 
2172     while (true) {
2173         uintptr_t align_start, end;
2174         align_start = ROUND_UP(base, align);
2175         end = align_start + guest_size + offset;
2176 
2177         /* if brk is anywhere in the range give ourselves some room to grow. */
2178         if (align_start <= brk && brk < end) {
2179             base = brk + (16 * MiB);
2180             continue;
2181         } else if (align_start + guest_size < align_start) {
2182             /* we have run out of space */
2183             return -1;
2184         } else {
2185             int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE |
2186                 MAP_FIXED_NOREPLACE;
2187             void * mmap_start = mmap((void *) align_start, guest_size,
2188                                      PROT_NONE, flags, -1, 0);
2189             if (mmap_start != MAP_FAILED) {
2190                 munmap((void *) align_start, guest_size);
2191                 if (MAP_FIXED_NOREPLACE || mmap_start == (void *) align_start) {
2192                     return (uintptr_t) mmap_start + offset;
2193                 }
2194             }
2195             base += qemu_host_page_size;
2196         }
2197     }
2198 }
2199 
2200 /* Return value for guest_base, or -1 if no hole found. */
2201 static uintptr_t pgb_find_hole(uintptr_t guest_loaddr, uintptr_t guest_size,
2202                                long align, uintptr_t offset)
2203 {
2204     GSList *maps, *iter;
2205     uintptr_t this_start, this_end, next_start, brk;
2206     intptr_t ret = -1;
2207 
2208     assert(QEMU_IS_ALIGNED(guest_loaddr, align));
2209 
2210     maps = read_self_maps();
2211 
2212     /* Read brk after we've read the maps, which will malloc. */
2213     brk = (uintptr_t)sbrk(0);
2214 
2215     if (!maps) {
2216         return pgd_find_hole_fallback(guest_size, brk, align, offset);
2217     }
2218 
2219     /* The first hole is before the first map entry. */
2220     this_start = mmap_min_addr;
2221 
2222     for (iter = maps; iter;
2223          this_start = next_start, iter = g_slist_next(iter)) {
2224         uintptr_t align_start, hole_size;
2225 
2226         this_end = ((MapInfo *)iter->data)->start;
2227         next_start = ((MapInfo *)iter->data)->end;
2228         align_start = ROUND_UP(this_start + offset, align);
2229 
2230         /* Skip holes that are too small. */
2231         if (align_start >= this_end) {
2232             continue;
2233         }
2234         hole_size = this_end - align_start;
2235         if (hole_size < guest_size) {
2236             continue;
2237         }
2238 
2239         /* If this hole contains brk, give ourselves some room to grow. */
2240         if (this_start <= brk && brk < this_end) {
2241             hole_size -= guest_size;
2242             if (sizeof(uintptr_t) == 8 && hole_size >= 1 * GiB) {
2243                 align_start += 1 * GiB;
2244             } else if (hole_size >= 16 * MiB) {
2245                 align_start += 16 * MiB;
2246             } else {
2247                 align_start = (this_end - guest_size) & -align;
2248                 if (align_start < this_start) {
2249                     continue;
2250                 }
2251             }
2252         }
2253 
2254         /* Record the lowest successful match. */
2255         if (ret < 0) {
2256             ret = align_start - guest_loaddr;
2257         }
2258         /* If this hole contains the identity map, select it. */
2259         if (align_start <= guest_loaddr &&
2260             guest_loaddr + guest_size <= this_end) {
2261             ret = 0;
2262         }
2263         /* If this hole ends above the identity map, stop looking. */
2264         if (this_end >= guest_loaddr) {
2265             break;
2266         }
2267     }
2268     free_self_maps(maps);
2269 
2270     return ret;
2271 }
2272 
2273 static void pgb_static(const char *image_name, abi_ulong orig_loaddr,
2274                        abi_ulong orig_hiaddr, long align)
2275 {
2276     uintptr_t loaddr = orig_loaddr;
2277     uintptr_t hiaddr = orig_hiaddr;
2278     uintptr_t offset = 0;
2279     uintptr_t addr;
2280 
2281     if (hiaddr != orig_hiaddr) {
2282         error_report("%s: requires virtual address space that the "
2283                      "host cannot provide (0x%" PRIx64 ")",
2284                      image_name, (uint64_t)orig_hiaddr);
2285         exit(EXIT_FAILURE);
2286     }
2287 
2288     loaddr &= -align;
2289     if (ARM_COMMPAGE) {
2290         /*
2291          * Extend the allocation to include the commpage.
2292          * For a 64-bit host, this is just 4GiB; for a 32-bit host we
2293          * need to ensure there is space bellow the guest_base so we
2294          * can map the commpage in the place needed when the address
2295          * arithmetic wraps around.
2296          */
2297         if (sizeof(uintptr_t) == 8 || loaddr >= 0x80000000u) {
2298             hiaddr = (uintptr_t) 4 << 30;
2299         } else {
2300             offset = -(ARM_COMMPAGE & -align);
2301         }
2302     }
2303 
2304     addr = pgb_find_hole(loaddr, hiaddr - loaddr, align, offset);
2305     if (addr == -1) {
2306         /*
2307          * If ARM_COMMPAGE, there *might* be a non-consecutive allocation
2308          * that can satisfy both.  But as the normal arm32 link base address
2309          * is ~32k, and we extend down to include the commpage, making the
2310          * overhead only ~96k, this is unlikely.
2311          */
2312         error_report("%s: Unable to allocate %#zx bytes of "
2313                      "virtual address space", image_name,
2314                      (size_t)(hiaddr - loaddr));
2315         exit(EXIT_FAILURE);
2316     }
2317 
2318     guest_base = addr;
2319 }
2320 
2321 static void pgb_dynamic(const char *image_name, long align)
2322 {
2323     /*
2324      * The executable is dynamic and does not require a fixed address.
2325      * All we need is a commpage that satisfies align.
2326      * If we do not need a commpage, leave guest_base == 0.
2327      */
2328     if (ARM_COMMPAGE) {
2329         uintptr_t addr, commpage;
2330 
2331         /* 64-bit hosts should have used reserved_va. */
2332         assert(sizeof(uintptr_t) == 4);
2333 
2334         /*
2335          * By putting the commpage at the first hole, that puts guest_base
2336          * just above that, and maximises the positive guest addresses.
2337          */
2338         commpage = ARM_COMMPAGE & -align;
2339         addr = pgb_find_hole(commpage, -commpage, align, 0);
2340         assert(addr != -1);
2341         guest_base = addr;
2342     }
2343 }
2344 
2345 static void pgb_reserved_va(const char *image_name, abi_ulong guest_loaddr,
2346                             abi_ulong guest_hiaddr, long align)
2347 {
2348     int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE;
2349     void *addr, *test;
2350 
2351     if (guest_hiaddr > reserved_va) {
2352         error_report("%s: requires more than reserved virtual "
2353                      "address space (0x%" PRIx64 " > 0x%lx)",
2354                      image_name, (uint64_t)guest_hiaddr, reserved_va);
2355         exit(EXIT_FAILURE);
2356     }
2357 
2358     /* Widen the "image" to the entire reserved address space. */
2359     pgb_static(image_name, 0, reserved_va, align);
2360 
2361     /* osdep.h defines this as 0 if it's missing */
2362     flags |= MAP_FIXED_NOREPLACE;
2363 
2364     /* Reserve the memory on the host. */
2365     assert(guest_base != 0);
2366     test = g2h(0);
2367     addr = mmap(test, reserved_va, PROT_NONE, flags, -1, 0);
2368     if (addr == MAP_FAILED || addr != test) {
2369         error_report("Unable to reserve 0x%lx bytes of virtual address "
2370                      "space at %p (%s) for use as guest address space (check your"
2371                      "virtual memory ulimit setting, min_mmap_addr or reserve less "
2372                      "using -R option)", reserved_va, test, strerror(errno));
2373         exit(EXIT_FAILURE);
2374     }
2375 }
2376 
2377 void probe_guest_base(const char *image_name, abi_ulong guest_loaddr,
2378                       abi_ulong guest_hiaddr)
2379 {
2380     /* In order to use host shmat, we must be able to honor SHMLBA.  */
2381     uintptr_t align = MAX(SHMLBA, qemu_host_page_size);
2382 
2383     if (have_guest_base) {
2384         pgb_have_guest_base(image_name, guest_loaddr, guest_hiaddr, align);
2385     } else if (reserved_va) {
2386         pgb_reserved_va(image_name, guest_loaddr, guest_hiaddr, align);
2387     } else if (guest_loaddr) {
2388         pgb_static(image_name, guest_loaddr, guest_hiaddr, align);
2389     } else {
2390         pgb_dynamic(image_name, align);
2391     }
2392 
2393     /* Reserve and initialize the commpage. */
2394     if (!init_guest_commpage()) {
2395         /*
2396          * With have_guest_base, the user has selected the address and
2397          * we are trying to work with that.  Otherwise, we have selected
2398          * free space and init_guest_commpage must succeeded.
2399          */
2400         assert(have_guest_base);
2401         pgb_fail_in_use(image_name);
2402     }
2403 
2404     assert(QEMU_IS_ALIGNED(guest_base, align));
2405     qemu_log_mask(CPU_LOG_PAGE, "Locating guest address space "
2406                   "@ 0x%" PRIx64 "\n", (uint64_t)guest_base);
2407 }
2408 
2409 enum {
2410     /* The string "GNU\0" as a magic number. */
2411     GNU0_MAGIC = const_le32('G' | 'N' << 8 | 'U' << 16),
2412     NOTE_DATA_SZ = 1 * KiB,
2413     NOTE_NAME_SZ = 4,
2414     ELF_GNU_PROPERTY_ALIGN = ELF_CLASS == ELFCLASS32 ? 4 : 8,
2415 };
2416 
2417 /*
2418  * Process a single gnu_property entry.
2419  * Return false for error.
2420  */
2421 static bool parse_elf_property(const uint32_t *data, int *off, int datasz,
2422                                struct image_info *info, bool have_prev_type,
2423                                uint32_t *prev_type, Error **errp)
2424 {
2425     uint32_t pr_type, pr_datasz, step;
2426 
2427     if (*off > datasz || !QEMU_IS_ALIGNED(*off, ELF_GNU_PROPERTY_ALIGN)) {
2428         goto error_data;
2429     }
2430     datasz -= *off;
2431     data += *off / sizeof(uint32_t);
2432 
2433     if (datasz < 2 * sizeof(uint32_t)) {
2434         goto error_data;
2435     }
2436     pr_type = data[0];
2437     pr_datasz = data[1];
2438     data += 2;
2439     datasz -= 2 * sizeof(uint32_t);
2440     step = ROUND_UP(pr_datasz, ELF_GNU_PROPERTY_ALIGN);
2441     if (step > datasz) {
2442         goto error_data;
2443     }
2444 
2445     /* Properties are supposed to be unique and sorted on pr_type. */
2446     if (have_prev_type && pr_type <= *prev_type) {
2447         if (pr_type == *prev_type) {
2448             error_setg(errp, "Duplicate property in PT_GNU_PROPERTY");
2449         } else {
2450             error_setg(errp, "Unsorted property in PT_GNU_PROPERTY");
2451         }
2452         return false;
2453     }
2454     *prev_type = pr_type;
2455 
2456     if (!arch_parse_elf_property(pr_type, pr_datasz, data, info, errp)) {
2457         return false;
2458     }
2459 
2460     *off += 2 * sizeof(uint32_t) + step;
2461     return true;
2462 
2463  error_data:
2464     error_setg(errp, "Ill-formed property in PT_GNU_PROPERTY");
2465     return false;
2466 }
2467 
2468 /* Process NT_GNU_PROPERTY_TYPE_0. */
2469 static bool parse_elf_properties(int image_fd,
2470                                  struct image_info *info,
2471                                  const struct elf_phdr *phdr,
2472                                  char bprm_buf[BPRM_BUF_SIZE],
2473                                  Error **errp)
2474 {
2475     union {
2476         struct elf_note nhdr;
2477         uint32_t data[NOTE_DATA_SZ / sizeof(uint32_t)];
2478     } note;
2479 
2480     int n, off, datasz;
2481     bool have_prev_type;
2482     uint32_t prev_type;
2483 
2484     /* Unless the arch requires properties, ignore them. */
2485     if (!ARCH_USE_GNU_PROPERTY) {
2486         return true;
2487     }
2488 
2489     /* If the properties are crazy large, that's too bad. */
2490     n = phdr->p_filesz;
2491     if (n > sizeof(note)) {
2492         error_setg(errp, "PT_GNU_PROPERTY too large");
2493         return false;
2494     }
2495     if (n < sizeof(note.nhdr)) {
2496         error_setg(errp, "PT_GNU_PROPERTY too small");
2497         return false;
2498     }
2499 
2500     if (phdr->p_offset + n <= BPRM_BUF_SIZE) {
2501         memcpy(&note, bprm_buf + phdr->p_offset, n);
2502     } else {
2503         ssize_t len = pread(image_fd, &note, n, phdr->p_offset);
2504         if (len != n) {
2505             error_setg_errno(errp, errno, "Error reading file header");
2506             return false;
2507         }
2508     }
2509 
2510     /*
2511      * The contents of a valid PT_GNU_PROPERTY is a sequence
2512      * of uint32_t -- swap them all now.
2513      */
2514 #ifdef BSWAP_NEEDED
2515     for (int i = 0; i < n / 4; i++) {
2516         bswap32s(note.data + i);
2517     }
2518 #endif
2519 
2520     /*
2521      * Note that nhdr is 3 words, and that the "name" described by namesz
2522      * immediately follows nhdr and is thus at the 4th word.  Further, all
2523      * of the inputs to the kernel's round_up are multiples of 4.
2524      */
2525     if (note.nhdr.n_type != NT_GNU_PROPERTY_TYPE_0 ||
2526         note.nhdr.n_namesz != NOTE_NAME_SZ ||
2527         note.data[3] != GNU0_MAGIC) {
2528         error_setg(errp, "Invalid note in PT_GNU_PROPERTY");
2529         return false;
2530     }
2531     off = sizeof(note.nhdr) + NOTE_NAME_SZ;
2532 
2533     datasz = note.nhdr.n_descsz + off;
2534     if (datasz > n) {
2535         error_setg(errp, "Invalid note size in PT_GNU_PROPERTY");
2536         return false;
2537     }
2538 
2539     have_prev_type = false;
2540     prev_type = 0;
2541     while (1) {
2542         if (off == datasz) {
2543             return true;  /* end, exit ok */
2544         }
2545         if (!parse_elf_property(note.data, &off, datasz, info,
2546                                 have_prev_type, &prev_type, errp)) {
2547             return false;
2548         }
2549         have_prev_type = true;
2550     }
2551 }
2552 
2553 /* Load an ELF image into the address space.
2554 
2555    IMAGE_NAME is the filename of the image, to use in error messages.
2556    IMAGE_FD is the open file descriptor for the image.
2557 
2558    BPRM_BUF is a copy of the beginning of the file; this of course
2559    contains the elf file header at offset 0.  It is assumed that this
2560    buffer is sufficiently aligned to present no problems to the host
2561    in accessing data at aligned offsets within the buffer.
2562 
2563    On return: INFO values will be filled in, as necessary or available.  */
2564 
2565 static void load_elf_image(const char *image_name, int image_fd,
2566                            struct image_info *info, char **pinterp_name,
2567                            char bprm_buf[BPRM_BUF_SIZE])
2568 {
2569     struct elfhdr *ehdr = (struct elfhdr *)bprm_buf;
2570     struct elf_phdr *phdr;
2571     abi_ulong load_addr, load_bias, loaddr, hiaddr, error;
2572     int i, retval, prot_exec;
2573     Error *err = NULL;
2574 
2575     /* First of all, some simple consistency checks */
2576     if (!elf_check_ident(ehdr)) {
2577         error_setg(&err, "Invalid ELF image for this architecture");
2578         goto exit_errmsg;
2579     }
2580     bswap_ehdr(ehdr);
2581     if (!elf_check_ehdr(ehdr)) {
2582         error_setg(&err, "Invalid ELF image for this architecture");
2583         goto exit_errmsg;
2584     }
2585 
2586     i = ehdr->e_phnum * sizeof(struct elf_phdr);
2587     if (ehdr->e_phoff + i <= BPRM_BUF_SIZE) {
2588         phdr = (struct elf_phdr *)(bprm_buf + ehdr->e_phoff);
2589     } else {
2590         phdr = (struct elf_phdr *) alloca(i);
2591         retval = pread(image_fd, phdr, i, ehdr->e_phoff);
2592         if (retval != i) {
2593             goto exit_read;
2594         }
2595     }
2596     bswap_phdr(phdr, ehdr->e_phnum);
2597 
2598     info->nsegs = 0;
2599     info->pt_dynamic_addr = 0;
2600 
2601     mmap_lock();
2602 
2603     /*
2604      * Find the maximum size of the image and allocate an appropriate
2605      * amount of memory to handle that.  Locate the interpreter, if any.
2606      */
2607     loaddr = -1, hiaddr = 0;
2608     info->alignment = 0;
2609     for (i = 0; i < ehdr->e_phnum; ++i) {
2610         struct elf_phdr *eppnt = phdr + i;
2611         if (eppnt->p_type == PT_LOAD) {
2612             abi_ulong a = eppnt->p_vaddr - eppnt->p_offset;
2613             if (a < loaddr) {
2614                 loaddr = a;
2615             }
2616             a = eppnt->p_vaddr + eppnt->p_memsz;
2617             if (a > hiaddr) {
2618                 hiaddr = a;
2619             }
2620             ++info->nsegs;
2621             info->alignment |= eppnt->p_align;
2622         } else if (eppnt->p_type == PT_INTERP && pinterp_name) {
2623             g_autofree char *interp_name = NULL;
2624 
2625             if (*pinterp_name) {
2626                 error_setg(&err, "Multiple PT_INTERP entries");
2627                 goto exit_errmsg;
2628             }
2629 
2630             interp_name = g_malloc(eppnt->p_filesz);
2631 
2632             if (eppnt->p_offset + eppnt->p_filesz <= BPRM_BUF_SIZE) {
2633                 memcpy(interp_name, bprm_buf + eppnt->p_offset,
2634                        eppnt->p_filesz);
2635             } else {
2636                 retval = pread(image_fd, interp_name, eppnt->p_filesz,
2637                                eppnt->p_offset);
2638                 if (retval != eppnt->p_filesz) {
2639                     goto exit_read;
2640                 }
2641             }
2642             if (interp_name[eppnt->p_filesz - 1] != 0) {
2643                 error_setg(&err, "Invalid PT_INTERP entry");
2644                 goto exit_errmsg;
2645             }
2646             *pinterp_name = g_steal_pointer(&interp_name);
2647         } else if (eppnt->p_type == PT_GNU_PROPERTY) {
2648             if (!parse_elf_properties(image_fd, info, eppnt, bprm_buf, &err)) {
2649                 goto exit_errmsg;
2650             }
2651         }
2652     }
2653 
2654     if (pinterp_name != NULL) {
2655         /*
2656          * This is the main executable.
2657          *
2658          * Reserve extra space for brk.
2659          * We hold on to this space while placing the interpreter
2660          * and the stack, lest they be placed immediately after
2661          * the data segment and block allocation from the brk.
2662          *
2663          * 16MB is chosen as "large enough" without being so large
2664          * as to allow the result to not fit with a 32-bit guest on
2665          * a 32-bit host.
2666          */
2667         info->reserve_brk = 16 * MiB;
2668         hiaddr += info->reserve_brk;
2669 
2670         if (ehdr->e_type == ET_EXEC) {
2671             /*
2672              * Make sure that the low address does not conflict with
2673              * MMAP_MIN_ADDR or the QEMU application itself.
2674              */
2675             probe_guest_base(image_name, loaddr, hiaddr);
2676         } else {
2677             /*
2678              * The binary is dynamic, but we still need to
2679              * select guest_base.  In this case we pass a size.
2680              */
2681             probe_guest_base(image_name, 0, hiaddr - loaddr);
2682         }
2683     }
2684 
2685     /*
2686      * Reserve address space for all of this.
2687      *
2688      * In the case of ET_EXEC, we supply MAP_FIXED so that we get
2689      * exactly the address range that is required.
2690      *
2691      * Otherwise this is ET_DYN, and we are searching for a location
2692      * that can hold the memory space required.  If the image is
2693      * pre-linked, LOADDR will be non-zero, and the kernel should
2694      * honor that address if it happens to be free.
2695      *
2696      * In both cases, we will overwrite pages in this range with mappings
2697      * from the executable.
2698      */
2699     load_addr = target_mmap(loaddr, hiaddr - loaddr, PROT_NONE,
2700                             MAP_PRIVATE | MAP_ANON | MAP_NORESERVE |
2701                             (ehdr->e_type == ET_EXEC ? MAP_FIXED : 0),
2702                             -1, 0);
2703     if (load_addr == -1) {
2704         goto exit_mmap;
2705     }
2706     load_bias = load_addr - loaddr;
2707 
2708     if (elf_is_fdpic(ehdr)) {
2709         struct elf32_fdpic_loadseg *loadsegs = info->loadsegs =
2710             g_malloc(sizeof(*loadsegs) * info->nsegs);
2711 
2712         for (i = 0; i < ehdr->e_phnum; ++i) {
2713             switch (phdr[i].p_type) {
2714             case PT_DYNAMIC:
2715                 info->pt_dynamic_addr = phdr[i].p_vaddr + load_bias;
2716                 break;
2717             case PT_LOAD:
2718                 loadsegs->addr = phdr[i].p_vaddr + load_bias;
2719                 loadsegs->p_vaddr = phdr[i].p_vaddr;
2720                 loadsegs->p_memsz = phdr[i].p_memsz;
2721                 ++loadsegs;
2722                 break;
2723             }
2724         }
2725     }
2726 
2727     info->load_bias = load_bias;
2728     info->code_offset = load_bias;
2729     info->data_offset = load_bias;
2730     info->load_addr = load_addr;
2731     info->entry = ehdr->e_entry + load_bias;
2732     info->start_code = -1;
2733     info->end_code = 0;
2734     info->start_data = -1;
2735     info->end_data = 0;
2736     info->brk = 0;
2737     info->elf_flags = ehdr->e_flags;
2738 
2739     prot_exec = PROT_EXEC;
2740 #ifdef TARGET_AARCH64
2741     /*
2742      * If the BTI feature is present, this indicates that the executable
2743      * pages of the startup binary should be mapped with PROT_BTI, so that
2744      * branch targets are enforced.
2745      *
2746      * The startup binary is either the interpreter or the static executable.
2747      * The interpreter is responsible for all pages of a dynamic executable.
2748      *
2749      * Elf notes are backward compatible to older cpus.
2750      * Do not enable BTI unless it is supported.
2751      */
2752     if ((info->note_flags & GNU_PROPERTY_AARCH64_FEATURE_1_BTI)
2753         && (pinterp_name == NULL || *pinterp_name == 0)
2754         && cpu_isar_feature(aa64_bti, ARM_CPU(thread_cpu))) {
2755         prot_exec |= TARGET_PROT_BTI;
2756     }
2757 #endif
2758 
2759     for (i = 0; i < ehdr->e_phnum; i++) {
2760         struct elf_phdr *eppnt = phdr + i;
2761         if (eppnt->p_type == PT_LOAD) {
2762             abi_ulong vaddr, vaddr_po, vaddr_ps, vaddr_ef, vaddr_em, vaddr_len;
2763             int elf_prot = 0;
2764 
2765             if (eppnt->p_flags & PF_R) {
2766                 elf_prot |= PROT_READ;
2767             }
2768             if (eppnt->p_flags & PF_W) {
2769                 elf_prot |= PROT_WRITE;
2770             }
2771             if (eppnt->p_flags & PF_X) {
2772                 elf_prot |= prot_exec;
2773             }
2774 
2775             vaddr = load_bias + eppnt->p_vaddr;
2776             vaddr_po = TARGET_ELF_PAGEOFFSET(vaddr);
2777             vaddr_ps = TARGET_ELF_PAGESTART(vaddr);
2778             vaddr_len = TARGET_ELF_PAGELENGTH(eppnt->p_filesz + vaddr_po);
2779 
2780             /*
2781              * Some segments may be completely empty without any backing file
2782              * segment, in that case just let zero_bss allocate an empty buffer
2783              * for it.
2784              */
2785             if (eppnt->p_filesz != 0) {
2786                 error = target_mmap(vaddr_ps, vaddr_len, elf_prot,
2787                                     MAP_PRIVATE | MAP_FIXED,
2788                                     image_fd, eppnt->p_offset - vaddr_po);
2789 
2790                 if (error == -1) {
2791                     goto exit_mmap;
2792                 }
2793             }
2794 
2795             vaddr_ef = vaddr + eppnt->p_filesz;
2796             vaddr_em = vaddr + eppnt->p_memsz;
2797 
2798             /* If the load segment requests extra zeros (e.g. bss), map it.  */
2799             if (vaddr_ef < vaddr_em) {
2800                 zero_bss(vaddr_ef, vaddr_em, elf_prot);
2801             }
2802 
2803             /* Find the full program boundaries.  */
2804             if (elf_prot & PROT_EXEC) {
2805                 if (vaddr < info->start_code) {
2806                     info->start_code = vaddr;
2807                 }
2808                 if (vaddr_ef > info->end_code) {
2809                     info->end_code = vaddr_ef;
2810                 }
2811             }
2812             if (elf_prot & PROT_WRITE) {
2813                 if (vaddr < info->start_data) {
2814                     info->start_data = vaddr;
2815                 }
2816                 if (vaddr_ef > info->end_data) {
2817                     info->end_data = vaddr_ef;
2818                 }
2819             }
2820             if (vaddr_em > info->brk) {
2821                 info->brk = vaddr_em;
2822             }
2823 #ifdef TARGET_MIPS
2824         } else if (eppnt->p_type == PT_MIPS_ABIFLAGS) {
2825             Mips_elf_abiflags_v0 abiflags;
2826             if (eppnt->p_filesz < sizeof(Mips_elf_abiflags_v0)) {
2827                 error_setg(&err, "Invalid PT_MIPS_ABIFLAGS entry");
2828                 goto exit_errmsg;
2829             }
2830             if (eppnt->p_offset + eppnt->p_filesz <= BPRM_BUF_SIZE) {
2831                 memcpy(&abiflags, bprm_buf + eppnt->p_offset,
2832                        sizeof(Mips_elf_abiflags_v0));
2833             } else {
2834                 retval = pread(image_fd, &abiflags, sizeof(Mips_elf_abiflags_v0),
2835                                eppnt->p_offset);
2836                 if (retval != sizeof(Mips_elf_abiflags_v0)) {
2837                     goto exit_read;
2838                 }
2839             }
2840             bswap_mips_abiflags(&abiflags);
2841             info->fp_abi = abiflags.fp_abi;
2842 #endif
2843         }
2844     }
2845 
2846     if (info->end_data == 0) {
2847         info->start_data = info->end_code;
2848         info->end_data = info->end_code;
2849     }
2850 
2851     if (qemu_log_enabled()) {
2852         load_symbols(ehdr, image_fd, load_bias);
2853     }
2854 
2855     mmap_unlock();
2856 
2857     close(image_fd);
2858     return;
2859 
2860  exit_read:
2861     if (retval >= 0) {
2862         error_setg(&err, "Incomplete read of file header");
2863     } else {
2864         error_setg_errno(&err, errno, "Error reading file header");
2865     }
2866     goto exit_errmsg;
2867  exit_mmap:
2868     error_setg_errno(&err, errno, "Error mapping file");
2869     goto exit_errmsg;
2870  exit_errmsg:
2871     error_reportf_err(err, "%s: ", image_name);
2872     exit(-1);
2873 }
2874 
2875 static void load_elf_interp(const char *filename, struct image_info *info,
2876                             char bprm_buf[BPRM_BUF_SIZE])
2877 {
2878     int fd, retval;
2879     Error *err = NULL;
2880 
2881     fd = open(path(filename), O_RDONLY);
2882     if (fd < 0) {
2883         error_setg_file_open(&err, errno, filename);
2884         error_report_err(err);
2885         exit(-1);
2886     }
2887 
2888     retval = read(fd, bprm_buf, BPRM_BUF_SIZE);
2889     if (retval < 0) {
2890         error_setg_errno(&err, errno, "Error reading file header");
2891         error_reportf_err(err, "%s: ", filename);
2892         exit(-1);
2893     }
2894 
2895     if (retval < BPRM_BUF_SIZE) {
2896         memset(bprm_buf + retval, 0, BPRM_BUF_SIZE - retval);
2897     }
2898 
2899     load_elf_image(filename, fd, info, NULL, bprm_buf);
2900 }
2901 
2902 static int symfind(const void *s0, const void *s1)
2903 {
2904     target_ulong addr = *(target_ulong *)s0;
2905     struct elf_sym *sym = (struct elf_sym *)s1;
2906     int result = 0;
2907     if (addr < sym->st_value) {
2908         result = -1;
2909     } else if (addr >= sym->st_value + sym->st_size) {
2910         result = 1;
2911     }
2912     return result;
2913 }
2914 
2915 static const char *lookup_symbolxx(struct syminfo *s, target_ulong orig_addr)
2916 {
2917 #if ELF_CLASS == ELFCLASS32
2918     struct elf_sym *syms = s->disas_symtab.elf32;
2919 #else
2920     struct elf_sym *syms = s->disas_symtab.elf64;
2921 #endif
2922 
2923     // binary search
2924     struct elf_sym *sym;
2925 
2926     sym = bsearch(&orig_addr, syms, s->disas_num_syms, sizeof(*syms), symfind);
2927     if (sym != NULL) {
2928         return s->disas_strtab + sym->st_name;
2929     }
2930 
2931     return "";
2932 }
2933 
2934 /* FIXME: This should use elf_ops.h  */
2935 static int symcmp(const void *s0, const void *s1)
2936 {
2937     struct elf_sym *sym0 = (struct elf_sym *)s0;
2938     struct elf_sym *sym1 = (struct elf_sym *)s1;
2939     return (sym0->st_value < sym1->st_value)
2940         ? -1
2941         : ((sym0->st_value > sym1->st_value) ? 1 : 0);
2942 }
2943 
2944 /* Best attempt to load symbols from this ELF object. */
2945 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias)
2946 {
2947     int i, shnum, nsyms, sym_idx = 0, str_idx = 0;
2948     uint64_t segsz;
2949     struct elf_shdr *shdr;
2950     char *strings = NULL;
2951     struct syminfo *s = NULL;
2952     struct elf_sym *new_syms, *syms = NULL;
2953 
2954     shnum = hdr->e_shnum;
2955     i = shnum * sizeof(struct elf_shdr);
2956     shdr = (struct elf_shdr *)alloca(i);
2957     if (pread(fd, shdr, i, hdr->e_shoff) != i) {
2958         return;
2959     }
2960 
2961     bswap_shdr(shdr, shnum);
2962     for (i = 0; i < shnum; ++i) {
2963         if (shdr[i].sh_type == SHT_SYMTAB) {
2964             sym_idx = i;
2965             str_idx = shdr[i].sh_link;
2966             goto found;
2967         }
2968     }
2969 
2970     /* There will be no symbol table if the file was stripped.  */
2971     return;
2972 
2973  found:
2974     /* Now know where the strtab and symtab are.  Snarf them.  */
2975     s = g_try_new(struct syminfo, 1);
2976     if (!s) {
2977         goto give_up;
2978     }
2979 
2980     segsz = shdr[str_idx].sh_size;
2981     s->disas_strtab = strings = g_try_malloc(segsz);
2982     if (!strings ||
2983         pread(fd, strings, segsz, shdr[str_idx].sh_offset) != segsz) {
2984         goto give_up;
2985     }
2986 
2987     segsz = shdr[sym_idx].sh_size;
2988     syms = g_try_malloc(segsz);
2989     if (!syms || pread(fd, syms, segsz, shdr[sym_idx].sh_offset) != segsz) {
2990         goto give_up;
2991     }
2992 
2993     if (segsz / sizeof(struct elf_sym) > INT_MAX) {
2994         /* Implausibly large symbol table: give up rather than ploughing
2995          * on with the number of symbols calculation overflowing
2996          */
2997         goto give_up;
2998     }
2999     nsyms = segsz / sizeof(struct elf_sym);
3000     for (i = 0; i < nsyms; ) {
3001         bswap_sym(syms + i);
3002         /* Throw away entries which we do not need.  */
3003         if (syms[i].st_shndx == SHN_UNDEF
3004             || syms[i].st_shndx >= SHN_LORESERVE
3005             || ELF_ST_TYPE(syms[i].st_info) != STT_FUNC) {
3006             if (i < --nsyms) {
3007                 syms[i] = syms[nsyms];
3008             }
3009         } else {
3010 #if defined(TARGET_ARM) || defined (TARGET_MIPS)
3011             /* The bottom address bit marks a Thumb or MIPS16 symbol.  */
3012             syms[i].st_value &= ~(target_ulong)1;
3013 #endif
3014             syms[i].st_value += load_bias;
3015             i++;
3016         }
3017     }
3018 
3019     /* No "useful" symbol.  */
3020     if (nsyms == 0) {
3021         goto give_up;
3022     }
3023 
3024     /* Attempt to free the storage associated with the local symbols
3025        that we threw away.  Whether or not this has any effect on the
3026        memory allocation depends on the malloc implementation and how
3027        many symbols we managed to discard.  */
3028     new_syms = g_try_renew(struct elf_sym, syms, nsyms);
3029     if (new_syms == NULL) {
3030         goto give_up;
3031     }
3032     syms = new_syms;
3033 
3034     qsort(syms, nsyms, sizeof(*syms), symcmp);
3035 
3036     s->disas_num_syms = nsyms;
3037 #if ELF_CLASS == ELFCLASS32
3038     s->disas_symtab.elf32 = syms;
3039 #else
3040     s->disas_symtab.elf64 = syms;
3041 #endif
3042     s->lookup_symbol = lookup_symbolxx;
3043     s->next = syminfos;
3044     syminfos = s;
3045 
3046     return;
3047 
3048 give_up:
3049     g_free(s);
3050     g_free(strings);
3051     g_free(syms);
3052 }
3053 
3054 uint32_t get_elf_eflags(int fd)
3055 {
3056     struct elfhdr ehdr;
3057     off_t offset;
3058     int ret;
3059 
3060     /* Read ELF header */
3061     offset = lseek(fd, 0, SEEK_SET);
3062     if (offset == (off_t) -1) {
3063         return 0;
3064     }
3065     ret = read(fd, &ehdr, sizeof(ehdr));
3066     if (ret < sizeof(ehdr)) {
3067         return 0;
3068     }
3069     offset = lseek(fd, offset, SEEK_SET);
3070     if (offset == (off_t) -1) {
3071         return 0;
3072     }
3073 
3074     /* Check ELF signature */
3075     if (!elf_check_ident(&ehdr)) {
3076         return 0;
3077     }
3078 
3079     /* check header */
3080     bswap_ehdr(&ehdr);
3081     if (!elf_check_ehdr(&ehdr)) {
3082         return 0;
3083     }
3084 
3085     /* return architecture id */
3086     return ehdr.e_flags;
3087 }
3088 
3089 int load_elf_binary(struct linux_binprm *bprm, struct image_info *info)
3090 {
3091     struct image_info interp_info;
3092     struct elfhdr elf_ex;
3093     char *elf_interpreter = NULL;
3094     char *scratch;
3095 
3096     memset(&interp_info, 0, sizeof(interp_info));
3097 #ifdef TARGET_MIPS
3098     interp_info.fp_abi = MIPS_ABI_FP_UNKNOWN;
3099 #endif
3100 
3101     info->start_mmap = (abi_ulong)ELF_START_MMAP;
3102 
3103     load_elf_image(bprm->filename, bprm->fd, info,
3104                    &elf_interpreter, bprm->buf);
3105 
3106     /* ??? We need a copy of the elf header for passing to create_elf_tables.
3107        If we do nothing, we'll have overwritten this when we re-use bprm->buf
3108        when we load the interpreter.  */
3109     elf_ex = *(struct elfhdr *)bprm->buf;
3110 
3111     /* Do this so that we can load the interpreter, if need be.  We will
3112        change some of these later */
3113     bprm->p = setup_arg_pages(bprm, info);
3114 
3115     scratch = g_new0(char, TARGET_PAGE_SIZE);
3116     if (STACK_GROWS_DOWN) {
3117         bprm->p = copy_elf_strings(1, &bprm->filename, scratch,
3118                                    bprm->p, info->stack_limit);
3119         info->file_string = bprm->p;
3120         bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch,
3121                                    bprm->p, info->stack_limit);
3122         info->env_strings = bprm->p;
3123         bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch,
3124                                    bprm->p, info->stack_limit);
3125         info->arg_strings = bprm->p;
3126     } else {
3127         info->arg_strings = bprm->p;
3128         bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch,
3129                                    bprm->p, info->stack_limit);
3130         info->env_strings = bprm->p;
3131         bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch,
3132                                    bprm->p, info->stack_limit);
3133         info->file_string = bprm->p;
3134         bprm->p = copy_elf_strings(1, &bprm->filename, scratch,
3135                                    bprm->p, info->stack_limit);
3136     }
3137 
3138     g_free(scratch);
3139 
3140     if (!bprm->p) {
3141         fprintf(stderr, "%s: %s\n", bprm->filename, strerror(E2BIG));
3142         exit(-1);
3143     }
3144 
3145     if (elf_interpreter) {
3146         load_elf_interp(elf_interpreter, &interp_info, bprm->buf);
3147 
3148         /* If the program interpreter is one of these two, then assume
3149            an iBCS2 image.  Otherwise assume a native linux image.  */
3150 
3151         if (strcmp(elf_interpreter, "/usr/lib/libc.so.1") == 0
3152             || strcmp(elf_interpreter, "/usr/lib/ld.so.1") == 0) {
3153             info->personality = PER_SVR4;
3154 
3155             /* Why this, you ask???  Well SVr4 maps page 0 as read-only,
3156                and some applications "depend" upon this behavior.  Since
3157                we do not have the power to recompile these, we emulate
3158                the SVr4 behavior.  Sigh.  */
3159             target_mmap(0, qemu_host_page_size, PROT_READ | PROT_EXEC,
3160                         MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
3161         }
3162 #ifdef TARGET_MIPS
3163         info->interp_fp_abi = interp_info.fp_abi;
3164 #endif
3165     }
3166 
3167     bprm->p = create_elf_tables(bprm->p, bprm->argc, bprm->envc, &elf_ex,
3168                                 info, (elf_interpreter ? &interp_info : NULL));
3169     info->start_stack = bprm->p;
3170 
3171     /* If we have an interpreter, set that as the program's entry point.
3172        Copy the load_bias as well, to help PPC64 interpret the entry
3173        point as a function descriptor.  Do this after creating elf tables
3174        so that we copy the original program entry point into the AUXV.  */
3175     if (elf_interpreter) {
3176         info->load_bias = interp_info.load_bias;
3177         info->entry = interp_info.entry;
3178         g_free(elf_interpreter);
3179     }
3180 
3181 #ifdef USE_ELF_CORE_DUMP
3182     bprm->core_dump = &elf_core_dump;
3183 #endif
3184 
3185     /*
3186      * If we reserved extra space for brk, release it now.
3187      * The implementation of do_brk in syscalls.c expects to be able
3188      * to mmap pages in this space.
3189      */
3190     if (info->reserve_brk) {
3191         abi_ulong start_brk = HOST_PAGE_ALIGN(info->brk);
3192         abi_ulong end_brk = HOST_PAGE_ALIGN(info->brk + info->reserve_brk);
3193         target_munmap(start_brk, end_brk - start_brk);
3194     }
3195 
3196     return 0;
3197 }
3198 
3199 #ifdef USE_ELF_CORE_DUMP
3200 /*
3201  * Definitions to generate Intel SVR4-like core files.
3202  * These mostly have the same names as the SVR4 types with "target_elf_"
3203  * tacked on the front to prevent clashes with linux definitions,
3204  * and the typedef forms have been avoided.  This is mostly like
3205  * the SVR4 structure, but more Linuxy, with things that Linux does
3206  * not support and which gdb doesn't really use excluded.
3207  *
3208  * Fields we don't dump (their contents is zero) in linux-user qemu
3209  * are marked with XXX.
3210  *
3211  * Core dump code is copied from linux kernel (fs/binfmt_elf.c).
3212  *
3213  * Porting ELF coredump for target is (quite) simple process.  First you
3214  * define USE_ELF_CORE_DUMP in target ELF code (where init_thread() for
3215  * the target resides):
3216  *
3217  * #define USE_ELF_CORE_DUMP
3218  *
3219  * Next you define type of register set used for dumping.  ELF specification
3220  * says that it needs to be array of elf_greg_t that has size of ELF_NREG.
3221  *
3222  * typedef <target_regtype> target_elf_greg_t;
3223  * #define ELF_NREG <number of registers>
3224  * typedef taret_elf_greg_t target_elf_gregset_t[ELF_NREG];
3225  *
3226  * Last step is to implement target specific function that copies registers
3227  * from given cpu into just specified register set.  Prototype is:
3228  *
3229  * static void elf_core_copy_regs(taret_elf_gregset_t *regs,
3230  *                                const CPUArchState *env);
3231  *
3232  * Parameters:
3233  *     regs - copy register values into here (allocated and zeroed by caller)
3234  *     env - copy registers from here
3235  *
3236  * Example for ARM target is provided in this file.
3237  */
3238 
3239 /* An ELF note in memory */
3240 struct memelfnote {
3241     const char *name;
3242     size_t     namesz;
3243     size_t     namesz_rounded;
3244     int        type;
3245     size_t     datasz;
3246     size_t     datasz_rounded;
3247     void       *data;
3248     size_t     notesz;
3249 };
3250 
3251 struct target_elf_siginfo {
3252     abi_int    si_signo; /* signal number */
3253     abi_int    si_code;  /* extra code */
3254     abi_int    si_errno; /* errno */
3255 };
3256 
3257 struct target_elf_prstatus {
3258     struct target_elf_siginfo pr_info;      /* Info associated with signal */
3259     abi_short          pr_cursig;    /* Current signal */
3260     abi_ulong          pr_sigpend;   /* XXX */
3261     abi_ulong          pr_sighold;   /* XXX */
3262     target_pid_t       pr_pid;
3263     target_pid_t       pr_ppid;
3264     target_pid_t       pr_pgrp;
3265     target_pid_t       pr_sid;
3266     struct target_timeval pr_utime;  /* XXX User time */
3267     struct target_timeval pr_stime;  /* XXX System time */
3268     struct target_timeval pr_cutime; /* XXX Cumulative user time */
3269     struct target_timeval pr_cstime; /* XXX Cumulative system time */
3270     target_elf_gregset_t      pr_reg;       /* GP registers */
3271     abi_int            pr_fpvalid;   /* XXX */
3272 };
3273 
3274 #define ELF_PRARGSZ     (80) /* Number of chars for args */
3275 
3276 struct target_elf_prpsinfo {
3277     char         pr_state;       /* numeric process state */
3278     char         pr_sname;       /* char for pr_state */
3279     char         pr_zomb;        /* zombie */
3280     char         pr_nice;        /* nice val */
3281     abi_ulong    pr_flag;        /* flags */
3282     target_uid_t pr_uid;
3283     target_gid_t pr_gid;
3284     target_pid_t pr_pid, pr_ppid, pr_pgrp, pr_sid;
3285     /* Lots missing */
3286     char    pr_fname[16] QEMU_NONSTRING; /* filename of executable */
3287     char    pr_psargs[ELF_PRARGSZ]; /* initial part of arg list */
3288 };
3289 
3290 /* Here is the structure in which status of each thread is captured. */
3291 struct elf_thread_status {
3292     QTAILQ_ENTRY(elf_thread_status)  ets_link;
3293     struct target_elf_prstatus prstatus;   /* NT_PRSTATUS */
3294 #if 0
3295     elf_fpregset_t fpu;             /* NT_PRFPREG */
3296     struct task_struct *thread;
3297     elf_fpxregset_t xfpu;           /* ELF_CORE_XFPREG_TYPE */
3298 #endif
3299     struct memelfnote notes[1];
3300     int num_notes;
3301 };
3302 
3303 struct elf_note_info {
3304     struct memelfnote   *notes;
3305     struct target_elf_prstatus *prstatus;  /* NT_PRSTATUS */
3306     struct target_elf_prpsinfo *psinfo;    /* NT_PRPSINFO */
3307 
3308     QTAILQ_HEAD(, elf_thread_status) thread_list;
3309 #if 0
3310     /*
3311      * Current version of ELF coredump doesn't support
3312      * dumping fp regs etc.
3313      */
3314     elf_fpregset_t *fpu;
3315     elf_fpxregset_t *xfpu;
3316     int thread_status_size;
3317 #endif
3318     int notes_size;
3319     int numnote;
3320 };
3321 
3322 struct vm_area_struct {
3323     target_ulong   vma_start;  /* start vaddr of memory region */
3324     target_ulong   vma_end;    /* end vaddr of memory region */
3325     abi_ulong      vma_flags;  /* protection etc. flags for the region */
3326     QTAILQ_ENTRY(vm_area_struct) vma_link;
3327 };
3328 
3329 struct mm_struct {
3330     QTAILQ_HEAD(, vm_area_struct) mm_mmap;
3331     int mm_count;           /* number of mappings */
3332 };
3333 
3334 static struct mm_struct *vma_init(void);
3335 static void vma_delete(struct mm_struct *);
3336 static int vma_add_mapping(struct mm_struct *, target_ulong,
3337                            target_ulong, abi_ulong);
3338 static int vma_get_mapping_count(const struct mm_struct *);
3339 static struct vm_area_struct *vma_first(const struct mm_struct *);
3340 static struct vm_area_struct *vma_next(struct vm_area_struct *);
3341 static abi_ulong vma_dump_size(const struct vm_area_struct *);
3342 static int vma_walker(void *priv, target_ulong start, target_ulong end,
3343                       unsigned long flags);
3344 
3345 static void fill_elf_header(struct elfhdr *, int, uint16_t, uint32_t);
3346 static void fill_note(struct memelfnote *, const char *, int,
3347                       unsigned int, void *);
3348 static void fill_prstatus(struct target_elf_prstatus *, const TaskState *, int);
3349 static int fill_psinfo(struct target_elf_prpsinfo *, const TaskState *);
3350 static void fill_auxv_note(struct memelfnote *, const TaskState *);
3351 static void fill_elf_note_phdr(struct elf_phdr *, int, off_t);
3352 static size_t note_size(const struct memelfnote *);
3353 static void free_note_info(struct elf_note_info *);
3354 static int fill_note_info(struct elf_note_info *, long, const CPUArchState *);
3355 static void fill_thread_info(struct elf_note_info *, const CPUArchState *);
3356 static int core_dump_filename(const TaskState *, char *, size_t);
3357 
3358 static int dump_write(int, const void *, size_t);
3359 static int write_note(struct memelfnote *, int);
3360 static int write_note_info(struct elf_note_info *, int);
3361 
3362 #ifdef BSWAP_NEEDED
3363 static void bswap_prstatus(struct target_elf_prstatus *prstatus)
3364 {
3365     prstatus->pr_info.si_signo = tswap32(prstatus->pr_info.si_signo);
3366     prstatus->pr_info.si_code = tswap32(prstatus->pr_info.si_code);
3367     prstatus->pr_info.si_errno = tswap32(prstatus->pr_info.si_errno);
3368     prstatus->pr_cursig = tswap16(prstatus->pr_cursig);
3369     prstatus->pr_sigpend = tswapal(prstatus->pr_sigpend);
3370     prstatus->pr_sighold = tswapal(prstatus->pr_sighold);
3371     prstatus->pr_pid = tswap32(prstatus->pr_pid);
3372     prstatus->pr_ppid = tswap32(prstatus->pr_ppid);
3373     prstatus->pr_pgrp = tswap32(prstatus->pr_pgrp);
3374     prstatus->pr_sid = tswap32(prstatus->pr_sid);
3375     /* cpu times are not filled, so we skip them */
3376     /* regs should be in correct format already */
3377     prstatus->pr_fpvalid = tswap32(prstatus->pr_fpvalid);
3378 }
3379 
3380 static void bswap_psinfo(struct target_elf_prpsinfo *psinfo)
3381 {
3382     psinfo->pr_flag = tswapal(psinfo->pr_flag);
3383     psinfo->pr_uid = tswap16(psinfo->pr_uid);
3384     psinfo->pr_gid = tswap16(psinfo->pr_gid);
3385     psinfo->pr_pid = tswap32(psinfo->pr_pid);
3386     psinfo->pr_ppid = tswap32(psinfo->pr_ppid);
3387     psinfo->pr_pgrp = tswap32(psinfo->pr_pgrp);
3388     psinfo->pr_sid = tswap32(psinfo->pr_sid);
3389 }
3390 
3391 static void bswap_note(struct elf_note *en)
3392 {
3393     bswap32s(&en->n_namesz);
3394     bswap32s(&en->n_descsz);
3395     bswap32s(&en->n_type);
3396 }
3397 #else
3398 static inline void bswap_prstatus(struct target_elf_prstatus *p) { }
3399 static inline void bswap_psinfo(struct target_elf_prpsinfo *p) {}
3400 static inline void bswap_note(struct elf_note *en) { }
3401 #endif /* BSWAP_NEEDED */
3402 
3403 /*
3404  * Minimal support for linux memory regions.  These are needed
3405  * when we are finding out what memory exactly belongs to
3406  * emulated process.  No locks needed here, as long as
3407  * thread that received the signal is stopped.
3408  */
3409 
3410 static struct mm_struct *vma_init(void)
3411 {
3412     struct mm_struct *mm;
3413 
3414     if ((mm = g_malloc(sizeof (*mm))) == NULL)
3415         return (NULL);
3416 
3417     mm->mm_count = 0;
3418     QTAILQ_INIT(&mm->mm_mmap);
3419 
3420     return (mm);
3421 }
3422 
3423 static void vma_delete(struct mm_struct *mm)
3424 {
3425     struct vm_area_struct *vma;
3426 
3427     while ((vma = vma_first(mm)) != NULL) {
3428         QTAILQ_REMOVE(&mm->mm_mmap, vma, vma_link);
3429         g_free(vma);
3430     }
3431     g_free(mm);
3432 }
3433 
3434 static int vma_add_mapping(struct mm_struct *mm, target_ulong start,
3435                            target_ulong end, abi_ulong flags)
3436 {
3437     struct vm_area_struct *vma;
3438 
3439     if ((vma = g_malloc0(sizeof (*vma))) == NULL)
3440         return (-1);
3441 
3442     vma->vma_start = start;
3443     vma->vma_end = end;
3444     vma->vma_flags = flags;
3445 
3446     QTAILQ_INSERT_TAIL(&mm->mm_mmap, vma, vma_link);
3447     mm->mm_count++;
3448 
3449     return (0);
3450 }
3451 
3452 static struct vm_area_struct *vma_first(const struct mm_struct *mm)
3453 {
3454     return (QTAILQ_FIRST(&mm->mm_mmap));
3455 }
3456 
3457 static struct vm_area_struct *vma_next(struct vm_area_struct *vma)
3458 {
3459     return (QTAILQ_NEXT(vma, vma_link));
3460 }
3461 
3462 static int vma_get_mapping_count(const struct mm_struct *mm)
3463 {
3464     return (mm->mm_count);
3465 }
3466 
3467 /*
3468  * Calculate file (dump) size of given memory region.
3469  */
3470 static abi_ulong vma_dump_size(const struct vm_area_struct *vma)
3471 {
3472     /* if we cannot even read the first page, skip it */
3473     if (!access_ok(VERIFY_READ, vma->vma_start, TARGET_PAGE_SIZE))
3474         return (0);
3475 
3476     /*
3477      * Usually we don't dump executable pages as they contain
3478      * non-writable code that debugger can read directly from
3479      * target library etc.  However, thread stacks are marked
3480      * also executable so we read in first page of given region
3481      * and check whether it contains elf header.  If there is
3482      * no elf header, we dump it.
3483      */
3484     if (vma->vma_flags & PROT_EXEC) {
3485         char page[TARGET_PAGE_SIZE];
3486 
3487         copy_from_user(page, vma->vma_start, sizeof (page));
3488         if ((page[EI_MAG0] == ELFMAG0) &&
3489             (page[EI_MAG1] == ELFMAG1) &&
3490             (page[EI_MAG2] == ELFMAG2) &&
3491             (page[EI_MAG3] == ELFMAG3)) {
3492             /*
3493              * Mappings are possibly from ELF binary.  Don't dump
3494              * them.
3495              */
3496             return (0);
3497         }
3498     }
3499 
3500     return (vma->vma_end - vma->vma_start);
3501 }
3502 
3503 static int vma_walker(void *priv, target_ulong start, target_ulong end,
3504                       unsigned long flags)
3505 {
3506     struct mm_struct *mm = (struct mm_struct *)priv;
3507 
3508     vma_add_mapping(mm, start, end, flags);
3509     return (0);
3510 }
3511 
3512 static void fill_note(struct memelfnote *note, const char *name, int type,
3513                       unsigned int sz, void *data)
3514 {
3515     unsigned int namesz;
3516 
3517     namesz = strlen(name) + 1;
3518     note->name = name;
3519     note->namesz = namesz;
3520     note->namesz_rounded = roundup(namesz, sizeof (int32_t));
3521     note->type = type;
3522     note->datasz = sz;
3523     note->datasz_rounded = roundup(sz, sizeof (int32_t));
3524 
3525     note->data = data;
3526 
3527     /*
3528      * We calculate rounded up note size here as specified by
3529      * ELF document.
3530      */
3531     note->notesz = sizeof (struct elf_note) +
3532         note->namesz_rounded + note->datasz_rounded;
3533 }
3534 
3535 static void fill_elf_header(struct elfhdr *elf, int segs, uint16_t machine,
3536                             uint32_t flags)
3537 {
3538     (void) memset(elf, 0, sizeof(*elf));
3539 
3540     (void) memcpy(elf->e_ident, ELFMAG, SELFMAG);
3541     elf->e_ident[EI_CLASS] = ELF_CLASS;
3542     elf->e_ident[EI_DATA] = ELF_DATA;
3543     elf->e_ident[EI_VERSION] = EV_CURRENT;
3544     elf->e_ident[EI_OSABI] = ELF_OSABI;
3545 
3546     elf->e_type = ET_CORE;
3547     elf->e_machine = machine;
3548     elf->e_version = EV_CURRENT;
3549     elf->e_phoff = sizeof(struct elfhdr);
3550     elf->e_flags = flags;
3551     elf->e_ehsize = sizeof(struct elfhdr);
3552     elf->e_phentsize = sizeof(struct elf_phdr);
3553     elf->e_phnum = segs;
3554 
3555     bswap_ehdr(elf);
3556 }
3557 
3558 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, off_t offset)
3559 {
3560     phdr->p_type = PT_NOTE;
3561     phdr->p_offset = offset;
3562     phdr->p_vaddr = 0;
3563     phdr->p_paddr = 0;
3564     phdr->p_filesz = sz;
3565     phdr->p_memsz = 0;
3566     phdr->p_flags = 0;
3567     phdr->p_align = 0;
3568 
3569     bswap_phdr(phdr, 1);
3570 }
3571 
3572 static size_t note_size(const struct memelfnote *note)
3573 {
3574     return (note->notesz);
3575 }
3576 
3577 static void fill_prstatus(struct target_elf_prstatus *prstatus,
3578                           const TaskState *ts, int signr)
3579 {
3580     (void) memset(prstatus, 0, sizeof (*prstatus));
3581     prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
3582     prstatus->pr_pid = ts->ts_tid;
3583     prstatus->pr_ppid = getppid();
3584     prstatus->pr_pgrp = getpgrp();
3585     prstatus->pr_sid = getsid(0);
3586 
3587     bswap_prstatus(prstatus);
3588 }
3589 
3590 static int fill_psinfo(struct target_elf_prpsinfo *psinfo, const TaskState *ts)
3591 {
3592     char *base_filename;
3593     unsigned int i, len;
3594 
3595     (void) memset(psinfo, 0, sizeof (*psinfo));
3596 
3597     len = ts->info->arg_end - ts->info->arg_start;
3598     if (len >= ELF_PRARGSZ)
3599         len = ELF_PRARGSZ - 1;
3600     if (copy_from_user(&psinfo->pr_psargs, ts->info->arg_start, len))
3601         return -EFAULT;
3602     for (i = 0; i < len; i++)
3603         if (psinfo->pr_psargs[i] == 0)
3604             psinfo->pr_psargs[i] = ' ';
3605     psinfo->pr_psargs[len] = 0;
3606 
3607     psinfo->pr_pid = getpid();
3608     psinfo->pr_ppid = getppid();
3609     psinfo->pr_pgrp = getpgrp();
3610     psinfo->pr_sid = getsid(0);
3611     psinfo->pr_uid = getuid();
3612     psinfo->pr_gid = getgid();
3613 
3614     base_filename = g_path_get_basename(ts->bprm->filename);
3615     /*
3616      * Using strncpy here is fine: at max-length,
3617      * this field is not NUL-terminated.
3618      */
3619     (void) strncpy(psinfo->pr_fname, base_filename,
3620                    sizeof(psinfo->pr_fname));
3621 
3622     g_free(base_filename);
3623     bswap_psinfo(psinfo);
3624     return (0);
3625 }
3626 
3627 static void fill_auxv_note(struct memelfnote *note, const TaskState *ts)
3628 {
3629     elf_addr_t auxv = (elf_addr_t)ts->info->saved_auxv;
3630     elf_addr_t orig_auxv = auxv;
3631     void *ptr;
3632     int len = ts->info->auxv_len;
3633 
3634     /*
3635      * Auxiliary vector is stored in target process stack.  It contains
3636      * {type, value} pairs that we need to dump into note.  This is not
3637      * strictly necessary but we do it here for sake of completeness.
3638      */
3639 
3640     /* read in whole auxv vector and copy it to memelfnote */
3641     ptr = lock_user(VERIFY_READ, orig_auxv, len, 0);
3642     if (ptr != NULL) {
3643         fill_note(note, "CORE", NT_AUXV, len, ptr);
3644         unlock_user(ptr, auxv, len);
3645     }
3646 }
3647 
3648 /*
3649  * Constructs name of coredump file.  We have following convention
3650  * for the name:
3651  *     qemu_<basename-of-target-binary>_<date>-<time>_<pid>.core
3652  *
3653  * Returns 0 in case of success, -1 otherwise (errno is set).
3654  */
3655 static int core_dump_filename(const TaskState *ts, char *buf,
3656                               size_t bufsize)
3657 {
3658     char timestamp[64];
3659     char *base_filename = NULL;
3660     struct timeval tv;
3661     struct tm tm;
3662 
3663     assert(bufsize >= PATH_MAX);
3664 
3665     if (gettimeofday(&tv, NULL) < 0) {
3666         (void) fprintf(stderr, "unable to get current timestamp: %s",
3667                        strerror(errno));
3668         return (-1);
3669     }
3670 
3671     base_filename = g_path_get_basename(ts->bprm->filename);
3672     (void) strftime(timestamp, sizeof (timestamp), "%Y%m%d-%H%M%S",
3673                     localtime_r(&tv.tv_sec, &tm));
3674     (void) snprintf(buf, bufsize, "qemu_%s_%s_%d.core",
3675                     base_filename, timestamp, (int)getpid());
3676     g_free(base_filename);
3677 
3678     return (0);
3679 }
3680 
3681 static int dump_write(int fd, const void *ptr, size_t size)
3682 {
3683     const char *bufp = (const char *)ptr;
3684     ssize_t bytes_written, bytes_left;
3685     struct rlimit dumpsize;
3686     off_t pos;
3687 
3688     bytes_written = 0;
3689     getrlimit(RLIMIT_CORE, &dumpsize);
3690     if ((pos = lseek(fd, 0, SEEK_CUR))==-1) {
3691         if (errno == ESPIPE) { /* not a seekable stream */
3692             bytes_left = size;
3693         } else {
3694             return pos;
3695         }
3696     } else {
3697         if (dumpsize.rlim_cur <= pos) {
3698             return -1;
3699         } else if (dumpsize.rlim_cur == RLIM_INFINITY) {
3700             bytes_left = size;
3701         } else {
3702             size_t limit_left=dumpsize.rlim_cur - pos;
3703             bytes_left = limit_left >= size ? size : limit_left ;
3704         }
3705     }
3706 
3707     /*
3708      * In normal conditions, single write(2) should do but
3709      * in case of socket etc. this mechanism is more portable.
3710      */
3711     do {
3712         bytes_written = write(fd, bufp, bytes_left);
3713         if (bytes_written < 0) {
3714             if (errno == EINTR)
3715                 continue;
3716             return (-1);
3717         } else if (bytes_written == 0) { /* eof */
3718             return (-1);
3719         }
3720         bufp += bytes_written;
3721         bytes_left -= bytes_written;
3722     } while (bytes_left > 0);
3723 
3724     return (0);
3725 }
3726 
3727 static int write_note(struct memelfnote *men, int fd)
3728 {
3729     struct elf_note en;
3730 
3731     en.n_namesz = men->namesz;
3732     en.n_type = men->type;
3733     en.n_descsz = men->datasz;
3734 
3735     bswap_note(&en);
3736 
3737     if (dump_write(fd, &en, sizeof(en)) != 0)
3738         return (-1);
3739     if (dump_write(fd, men->name, men->namesz_rounded) != 0)
3740         return (-1);
3741     if (dump_write(fd, men->data, men->datasz_rounded) != 0)
3742         return (-1);
3743 
3744     return (0);
3745 }
3746 
3747 static void fill_thread_info(struct elf_note_info *info, const CPUArchState *env)
3748 {
3749     CPUState *cpu = env_cpu((CPUArchState *)env);
3750     TaskState *ts = (TaskState *)cpu->opaque;
3751     struct elf_thread_status *ets;
3752 
3753     ets = g_malloc0(sizeof (*ets));
3754     ets->num_notes = 1; /* only prstatus is dumped */
3755     fill_prstatus(&ets->prstatus, ts, 0);
3756     elf_core_copy_regs(&ets->prstatus.pr_reg, env);
3757     fill_note(&ets->notes[0], "CORE", NT_PRSTATUS, sizeof (ets->prstatus),
3758               &ets->prstatus);
3759 
3760     QTAILQ_INSERT_TAIL(&info->thread_list, ets, ets_link);
3761 
3762     info->notes_size += note_size(&ets->notes[0]);
3763 }
3764 
3765 static void init_note_info(struct elf_note_info *info)
3766 {
3767     /* Initialize the elf_note_info structure so that it is at
3768      * least safe to call free_note_info() on it. Must be
3769      * called before calling fill_note_info().
3770      */
3771     memset(info, 0, sizeof (*info));
3772     QTAILQ_INIT(&info->thread_list);
3773 }
3774 
3775 static int fill_note_info(struct elf_note_info *info,
3776                           long signr, const CPUArchState *env)
3777 {
3778 #define NUMNOTES 3
3779     CPUState *cpu = env_cpu((CPUArchState *)env);
3780     TaskState *ts = (TaskState *)cpu->opaque;
3781     int i;
3782 
3783     info->notes = g_new0(struct memelfnote, NUMNOTES);
3784     if (info->notes == NULL)
3785         return (-ENOMEM);
3786     info->prstatus = g_malloc0(sizeof (*info->prstatus));
3787     if (info->prstatus == NULL)
3788         return (-ENOMEM);
3789     info->psinfo = g_malloc0(sizeof (*info->psinfo));
3790     if (info->prstatus == NULL)
3791         return (-ENOMEM);
3792 
3793     /*
3794      * First fill in status (and registers) of current thread
3795      * including process info & aux vector.
3796      */
3797     fill_prstatus(info->prstatus, ts, signr);
3798     elf_core_copy_regs(&info->prstatus->pr_reg, env);
3799     fill_note(&info->notes[0], "CORE", NT_PRSTATUS,
3800               sizeof (*info->prstatus), info->prstatus);
3801     fill_psinfo(info->psinfo, ts);
3802     fill_note(&info->notes[1], "CORE", NT_PRPSINFO,
3803               sizeof (*info->psinfo), info->psinfo);
3804     fill_auxv_note(&info->notes[2], ts);
3805     info->numnote = 3;
3806 
3807     info->notes_size = 0;
3808     for (i = 0; i < info->numnote; i++)
3809         info->notes_size += note_size(&info->notes[i]);
3810 
3811     /* read and fill status of all threads */
3812     cpu_list_lock();
3813     CPU_FOREACH(cpu) {
3814         if (cpu == thread_cpu) {
3815             continue;
3816         }
3817         fill_thread_info(info, (CPUArchState *)cpu->env_ptr);
3818     }
3819     cpu_list_unlock();
3820 
3821     return (0);
3822 }
3823 
3824 static void free_note_info(struct elf_note_info *info)
3825 {
3826     struct elf_thread_status *ets;
3827 
3828     while (!QTAILQ_EMPTY(&info->thread_list)) {
3829         ets = QTAILQ_FIRST(&info->thread_list);
3830         QTAILQ_REMOVE(&info->thread_list, ets, ets_link);
3831         g_free(ets);
3832     }
3833 
3834     g_free(info->prstatus);
3835     g_free(info->psinfo);
3836     g_free(info->notes);
3837 }
3838 
3839 static int write_note_info(struct elf_note_info *info, int fd)
3840 {
3841     struct elf_thread_status *ets;
3842     int i, error = 0;
3843 
3844     /* write prstatus, psinfo and auxv for current thread */
3845     for (i = 0; i < info->numnote; i++)
3846         if ((error = write_note(&info->notes[i], fd)) != 0)
3847             return (error);
3848 
3849     /* write prstatus for each thread */
3850     QTAILQ_FOREACH(ets, &info->thread_list, ets_link) {
3851         if ((error = write_note(&ets->notes[0], fd)) != 0)
3852             return (error);
3853     }
3854 
3855     return (0);
3856 }
3857 
3858 /*
3859  * Write out ELF coredump.
3860  *
3861  * See documentation of ELF object file format in:
3862  * http://www.caldera.com/developers/devspecs/gabi41.pdf
3863  *
3864  * Coredump format in linux is following:
3865  *
3866  * 0   +----------------------+         \
3867  *     | ELF header           | ET_CORE  |
3868  *     +----------------------+          |
3869  *     | ELF program headers  |          |--- headers
3870  *     | - NOTE section       |          |
3871  *     | - PT_LOAD sections   |          |
3872  *     +----------------------+         /
3873  *     | NOTEs:               |
3874  *     | - NT_PRSTATUS        |
3875  *     | - NT_PRSINFO         |
3876  *     | - NT_AUXV            |
3877  *     +----------------------+ <-- aligned to target page
3878  *     | Process memory dump  |
3879  *     :                      :
3880  *     .                      .
3881  *     :                      :
3882  *     |                      |
3883  *     +----------------------+
3884  *
3885  * NT_PRSTATUS -> struct elf_prstatus (per thread)
3886  * NT_PRSINFO  -> struct elf_prpsinfo
3887  * NT_AUXV is array of { type, value } pairs (see fill_auxv_note()).
3888  *
3889  * Format follows System V format as close as possible.  Current
3890  * version limitations are as follows:
3891  *     - no floating point registers are dumped
3892  *
3893  * Function returns 0 in case of success, negative errno otherwise.
3894  *
3895  * TODO: make this work also during runtime: it should be
3896  * possible to force coredump from running process and then
3897  * continue processing.  For example qemu could set up SIGUSR2
3898  * handler (provided that target process haven't registered
3899  * handler for that) that does the dump when signal is received.
3900  */
3901 static int elf_core_dump(int signr, const CPUArchState *env)
3902 {
3903     const CPUState *cpu = env_cpu((CPUArchState *)env);
3904     const TaskState *ts = (const TaskState *)cpu->opaque;
3905     struct vm_area_struct *vma = NULL;
3906     char corefile[PATH_MAX];
3907     struct elf_note_info info;
3908     struct elfhdr elf;
3909     struct elf_phdr phdr;
3910     struct rlimit dumpsize;
3911     struct mm_struct *mm = NULL;
3912     off_t offset = 0, data_offset = 0;
3913     int segs = 0;
3914     int fd = -1;
3915 
3916     init_note_info(&info);
3917 
3918     errno = 0;
3919     getrlimit(RLIMIT_CORE, &dumpsize);
3920     if (dumpsize.rlim_cur == 0)
3921         return 0;
3922 
3923     if (core_dump_filename(ts, corefile, sizeof (corefile)) < 0)
3924         return (-errno);
3925 
3926     if ((fd = open(corefile, O_WRONLY | O_CREAT,
3927                    S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH)) < 0)
3928         return (-errno);
3929 
3930     /*
3931      * Walk through target process memory mappings and
3932      * set up structure containing this information.  After
3933      * this point vma_xxx functions can be used.
3934      */
3935     if ((mm = vma_init()) == NULL)
3936         goto out;
3937 
3938     walk_memory_regions(mm, vma_walker);
3939     segs = vma_get_mapping_count(mm);
3940 
3941     /*
3942      * Construct valid coredump ELF header.  We also
3943      * add one more segment for notes.
3944      */
3945     fill_elf_header(&elf, segs + 1, ELF_MACHINE, 0);
3946     if (dump_write(fd, &elf, sizeof (elf)) != 0)
3947         goto out;
3948 
3949     /* fill in the in-memory version of notes */
3950     if (fill_note_info(&info, signr, env) < 0)
3951         goto out;
3952 
3953     offset += sizeof (elf);                             /* elf header */
3954     offset += (segs + 1) * sizeof (struct elf_phdr);    /* program headers */
3955 
3956     /* write out notes program header */
3957     fill_elf_note_phdr(&phdr, info.notes_size, offset);
3958 
3959     offset += info.notes_size;
3960     if (dump_write(fd, &phdr, sizeof (phdr)) != 0)
3961         goto out;
3962 
3963     /*
3964      * ELF specification wants data to start at page boundary so
3965      * we align it here.
3966      */
3967     data_offset = offset = roundup(offset, ELF_EXEC_PAGESIZE);
3968 
3969     /*
3970      * Write program headers for memory regions mapped in
3971      * the target process.
3972      */
3973     for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) {
3974         (void) memset(&phdr, 0, sizeof (phdr));
3975 
3976         phdr.p_type = PT_LOAD;
3977         phdr.p_offset = offset;
3978         phdr.p_vaddr = vma->vma_start;
3979         phdr.p_paddr = 0;
3980         phdr.p_filesz = vma_dump_size(vma);
3981         offset += phdr.p_filesz;
3982         phdr.p_memsz = vma->vma_end - vma->vma_start;
3983         phdr.p_flags = vma->vma_flags & PROT_READ ? PF_R : 0;
3984         if (vma->vma_flags & PROT_WRITE)
3985             phdr.p_flags |= PF_W;
3986         if (vma->vma_flags & PROT_EXEC)
3987             phdr.p_flags |= PF_X;
3988         phdr.p_align = ELF_EXEC_PAGESIZE;
3989 
3990         bswap_phdr(&phdr, 1);
3991         if (dump_write(fd, &phdr, sizeof(phdr)) != 0) {
3992             goto out;
3993         }
3994     }
3995 
3996     /*
3997      * Next we write notes just after program headers.  No
3998      * alignment needed here.
3999      */
4000     if (write_note_info(&info, fd) < 0)
4001         goto out;
4002 
4003     /* align data to page boundary */
4004     if (lseek(fd, data_offset, SEEK_SET) != data_offset)
4005         goto out;
4006 
4007     /*
4008      * Finally we can dump process memory into corefile as well.
4009      */
4010     for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) {
4011         abi_ulong addr;
4012         abi_ulong end;
4013 
4014         end = vma->vma_start + vma_dump_size(vma);
4015 
4016         for (addr = vma->vma_start; addr < end;
4017              addr += TARGET_PAGE_SIZE) {
4018             char page[TARGET_PAGE_SIZE];
4019             int error;
4020 
4021             /*
4022              *  Read in page from target process memory and
4023              *  write it to coredump file.
4024              */
4025             error = copy_from_user(page, addr, sizeof (page));
4026             if (error != 0) {
4027                 (void) fprintf(stderr, "unable to dump " TARGET_ABI_FMT_lx "\n",
4028                                addr);
4029                 errno = -error;
4030                 goto out;
4031             }
4032             if (dump_write(fd, page, TARGET_PAGE_SIZE) < 0)
4033                 goto out;
4034         }
4035     }
4036 
4037  out:
4038     free_note_info(&info);
4039     if (mm != NULL)
4040         vma_delete(mm);
4041     (void) close(fd);
4042 
4043     if (errno != 0)
4044         return (-errno);
4045     return (0);
4046 }
4047 #endif /* USE_ELF_CORE_DUMP */
4048 
4049 void do_init_thread(struct target_pt_regs *regs, struct image_info *infop)
4050 {
4051     init_thread(regs, infop);
4052 }
4053