xref: /openbmc/qemu/linux-user/elfload.c (revision a495eba0)
1 /* This is the Linux kernel elf-loading code, ported into user space */
2 #include "qemu/osdep.h"
3 #include <sys/param.h>
4 
5 #include <sys/resource.h>
6 #include <sys/shm.h>
7 
8 #include "qemu.h"
9 #include "user-internals.h"
10 #include "signal-common.h"
11 #include "loader.h"
12 #include "user-mmap.h"
13 #include "disas/disas.h"
14 #include "qemu/bitops.h"
15 #include "qemu/path.h"
16 #include "qemu/queue.h"
17 #include "qemu/guest-random.h"
18 #include "qemu/units.h"
19 #include "qemu/selfmap.h"
20 #include "qapi/error.h"
21 #include "target_signal.h"
22 
23 #ifdef _ARCH_PPC64
24 #undef ARCH_DLINFO
25 #undef ELF_PLATFORM
26 #undef ELF_HWCAP
27 #undef ELF_HWCAP2
28 #undef ELF_CLASS
29 #undef ELF_DATA
30 #undef ELF_ARCH
31 #endif
32 
33 #define ELF_OSABI   ELFOSABI_SYSV
34 
35 /* from personality.h */
36 
37 /*
38  * Flags for bug emulation.
39  *
40  * These occupy the top three bytes.
41  */
42 enum {
43     ADDR_NO_RANDOMIZE = 0x0040000,      /* disable randomization of VA space */
44     FDPIC_FUNCPTRS =    0x0080000,      /* userspace function ptrs point to
45                                            descriptors (signal handling) */
46     MMAP_PAGE_ZERO =    0x0100000,
47     ADDR_COMPAT_LAYOUT = 0x0200000,
48     READ_IMPLIES_EXEC = 0x0400000,
49     ADDR_LIMIT_32BIT =  0x0800000,
50     SHORT_INODE =       0x1000000,
51     WHOLE_SECONDS =     0x2000000,
52     STICKY_TIMEOUTS =   0x4000000,
53     ADDR_LIMIT_3GB =    0x8000000,
54 };
55 
56 /*
57  * Personality types.
58  *
59  * These go in the low byte.  Avoid using the top bit, it will
60  * conflict with error returns.
61  */
62 enum {
63     PER_LINUX =         0x0000,
64     PER_LINUX_32BIT =   0x0000 | ADDR_LIMIT_32BIT,
65     PER_LINUX_FDPIC =   0x0000 | FDPIC_FUNCPTRS,
66     PER_SVR4 =          0x0001 | STICKY_TIMEOUTS | MMAP_PAGE_ZERO,
67     PER_SVR3 =          0x0002 | STICKY_TIMEOUTS | SHORT_INODE,
68     PER_SCOSVR3 =       0x0003 | STICKY_TIMEOUTS | WHOLE_SECONDS | SHORT_INODE,
69     PER_OSR5 =          0x0003 | STICKY_TIMEOUTS | WHOLE_SECONDS,
70     PER_WYSEV386 =      0x0004 | STICKY_TIMEOUTS | SHORT_INODE,
71     PER_ISCR4 =         0x0005 | STICKY_TIMEOUTS,
72     PER_BSD =           0x0006,
73     PER_SUNOS =         0x0006 | STICKY_TIMEOUTS,
74     PER_XENIX =         0x0007 | STICKY_TIMEOUTS | SHORT_INODE,
75     PER_LINUX32 =       0x0008,
76     PER_LINUX32_3GB =   0x0008 | ADDR_LIMIT_3GB,
77     PER_IRIX32 =        0x0009 | STICKY_TIMEOUTS,/* IRIX5 32-bit */
78     PER_IRIXN32 =       0x000a | STICKY_TIMEOUTS,/* IRIX6 new 32-bit */
79     PER_IRIX64 =        0x000b | STICKY_TIMEOUTS,/* IRIX6 64-bit */
80     PER_RISCOS =        0x000c,
81     PER_SOLARIS =       0x000d | STICKY_TIMEOUTS,
82     PER_UW7 =           0x000e | STICKY_TIMEOUTS | MMAP_PAGE_ZERO,
83     PER_OSF4 =          0x000f,                  /* OSF/1 v4 */
84     PER_HPUX =          0x0010,
85     PER_MASK =          0x00ff,
86 };
87 
88 /*
89  * Return the base personality without flags.
90  */
91 #define personality(pers)       (pers & PER_MASK)
92 
93 int info_is_fdpic(struct image_info *info)
94 {
95     return info->personality == PER_LINUX_FDPIC;
96 }
97 
98 /* this flag is uneffective under linux too, should be deleted */
99 #ifndef MAP_DENYWRITE
100 #define MAP_DENYWRITE 0
101 #endif
102 
103 /* should probably go in elf.h */
104 #ifndef ELIBBAD
105 #define ELIBBAD 80
106 #endif
107 
108 #if TARGET_BIG_ENDIAN
109 #define ELF_DATA        ELFDATA2MSB
110 #else
111 #define ELF_DATA        ELFDATA2LSB
112 #endif
113 
114 #ifdef TARGET_ABI_MIPSN32
115 typedef abi_ullong      target_elf_greg_t;
116 #define tswapreg(ptr)   tswap64(ptr)
117 #else
118 typedef abi_ulong       target_elf_greg_t;
119 #define tswapreg(ptr)   tswapal(ptr)
120 #endif
121 
122 #ifdef USE_UID16
123 typedef abi_ushort      target_uid_t;
124 typedef abi_ushort      target_gid_t;
125 #else
126 typedef abi_uint        target_uid_t;
127 typedef abi_uint        target_gid_t;
128 #endif
129 typedef abi_int         target_pid_t;
130 
131 #ifdef TARGET_I386
132 
133 #define ELF_HWCAP get_elf_hwcap()
134 
135 static uint32_t get_elf_hwcap(void)
136 {
137     X86CPU *cpu = X86_CPU(thread_cpu);
138 
139     return cpu->env.features[FEAT_1_EDX];
140 }
141 
142 #ifdef TARGET_X86_64
143 #define ELF_START_MMAP 0x2aaaaab000ULL
144 
145 #define ELF_CLASS      ELFCLASS64
146 #define ELF_ARCH       EM_X86_64
147 
148 #define ELF_PLATFORM   "x86_64"
149 
150 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop)
151 {
152     regs->rax = 0;
153     regs->rsp = infop->start_stack;
154     regs->rip = infop->entry;
155 }
156 
157 #define ELF_NREG    27
158 typedef target_elf_greg_t  target_elf_gregset_t[ELF_NREG];
159 
160 /*
161  * Note that ELF_NREG should be 29 as there should be place for
162  * TRAPNO and ERR "registers" as well but linux doesn't dump
163  * those.
164  *
165  * See linux kernel: arch/x86/include/asm/elf.h
166  */
167 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUX86State *env)
168 {
169     (*regs)[0] = tswapreg(env->regs[15]);
170     (*regs)[1] = tswapreg(env->regs[14]);
171     (*regs)[2] = tswapreg(env->regs[13]);
172     (*regs)[3] = tswapreg(env->regs[12]);
173     (*regs)[4] = tswapreg(env->regs[R_EBP]);
174     (*regs)[5] = tswapreg(env->regs[R_EBX]);
175     (*regs)[6] = tswapreg(env->regs[11]);
176     (*regs)[7] = tswapreg(env->regs[10]);
177     (*regs)[8] = tswapreg(env->regs[9]);
178     (*regs)[9] = tswapreg(env->regs[8]);
179     (*regs)[10] = tswapreg(env->regs[R_EAX]);
180     (*regs)[11] = tswapreg(env->regs[R_ECX]);
181     (*regs)[12] = tswapreg(env->regs[R_EDX]);
182     (*regs)[13] = tswapreg(env->regs[R_ESI]);
183     (*regs)[14] = tswapreg(env->regs[R_EDI]);
184     (*regs)[15] = tswapreg(env->regs[R_EAX]); /* XXX */
185     (*regs)[16] = tswapreg(env->eip);
186     (*regs)[17] = tswapreg(env->segs[R_CS].selector & 0xffff);
187     (*regs)[18] = tswapreg(env->eflags);
188     (*regs)[19] = tswapreg(env->regs[R_ESP]);
189     (*regs)[20] = tswapreg(env->segs[R_SS].selector & 0xffff);
190     (*regs)[21] = tswapreg(env->segs[R_FS].selector & 0xffff);
191     (*regs)[22] = tswapreg(env->segs[R_GS].selector & 0xffff);
192     (*regs)[23] = tswapreg(env->segs[R_DS].selector & 0xffff);
193     (*regs)[24] = tswapreg(env->segs[R_ES].selector & 0xffff);
194     (*regs)[25] = tswapreg(env->segs[R_FS].selector & 0xffff);
195     (*regs)[26] = tswapreg(env->segs[R_GS].selector & 0xffff);
196 }
197 
198 #else
199 
200 #define ELF_START_MMAP 0x80000000
201 
202 /*
203  * This is used to ensure we don't load something for the wrong architecture.
204  */
205 #define elf_check_arch(x) ( ((x) == EM_386) || ((x) == EM_486) )
206 
207 /*
208  * These are used to set parameters in the core dumps.
209  */
210 #define ELF_CLASS       ELFCLASS32
211 #define ELF_ARCH        EM_386
212 
213 #define ELF_PLATFORM get_elf_platform()
214 
215 static const char *get_elf_platform(void)
216 {
217     static char elf_platform[] = "i386";
218     int family = object_property_get_int(OBJECT(thread_cpu), "family", NULL);
219     if (family > 6) {
220         family = 6;
221     }
222     if (family >= 3) {
223         elf_platform[1] = '0' + family;
224     }
225     return elf_platform;
226 }
227 
228 static inline void init_thread(struct target_pt_regs *regs,
229                                struct image_info *infop)
230 {
231     regs->esp = infop->start_stack;
232     regs->eip = infop->entry;
233 
234     /* SVR4/i386 ABI (pages 3-31, 3-32) says that when the program
235        starts %edx contains a pointer to a function which might be
236        registered using `atexit'.  This provides a mean for the
237        dynamic linker to call DT_FINI functions for shared libraries
238        that have been loaded before the code runs.
239 
240        A value of 0 tells we have no such handler.  */
241     regs->edx = 0;
242 }
243 
244 #define ELF_NREG    17
245 typedef target_elf_greg_t  target_elf_gregset_t[ELF_NREG];
246 
247 /*
248  * Note that ELF_NREG should be 19 as there should be place for
249  * TRAPNO and ERR "registers" as well but linux doesn't dump
250  * those.
251  *
252  * See linux kernel: arch/x86/include/asm/elf.h
253  */
254 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUX86State *env)
255 {
256     (*regs)[0] = tswapreg(env->regs[R_EBX]);
257     (*regs)[1] = tswapreg(env->regs[R_ECX]);
258     (*regs)[2] = tswapreg(env->regs[R_EDX]);
259     (*regs)[3] = tswapreg(env->regs[R_ESI]);
260     (*regs)[4] = tswapreg(env->regs[R_EDI]);
261     (*regs)[5] = tswapreg(env->regs[R_EBP]);
262     (*regs)[6] = tswapreg(env->regs[R_EAX]);
263     (*regs)[7] = tswapreg(env->segs[R_DS].selector & 0xffff);
264     (*regs)[8] = tswapreg(env->segs[R_ES].selector & 0xffff);
265     (*regs)[9] = tswapreg(env->segs[R_FS].selector & 0xffff);
266     (*regs)[10] = tswapreg(env->segs[R_GS].selector & 0xffff);
267     (*regs)[11] = tswapreg(env->regs[R_EAX]); /* XXX */
268     (*regs)[12] = tswapreg(env->eip);
269     (*regs)[13] = tswapreg(env->segs[R_CS].selector & 0xffff);
270     (*regs)[14] = tswapreg(env->eflags);
271     (*regs)[15] = tswapreg(env->regs[R_ESP]);
272     (*regs)[16] = tswapreg(env->segs[R_SS].selector & 0xffff);
273 }
274 #endif
275 
276 #define USE_ELF_CORE_DUMP
277 #define ELF_EXEC_PAGESIZE       4096
278 
279 #endif
280 
281 #ifdef TARGET_ARM
282 
283 #ifndef TARGET_AARCH64
284 /* 32 bit ARM definitions */
285 
286 #define ELF_START_MMAP 0x80000000
287 
288 #define ELF_ARCH        EM_ARM
289 #define ELF_CLASS       ELFCLASS32
290 
291 static inline void init_thread(struct target_pt_regs *regs,
292                                struct image_info *infop)
293 {
294     abi_long stack = infop->start_stack;
295     memset(regs, 0, sizeof(*regs));
296 
297     regs->uregs[16] = ARM_CPU_MODE_USR;
298     if (infop->entry & 1) {
299         regs->uregs[16] |= CPSR_T;
300     }
301     regs->uregs[15] = infop->entry & 0xfffffffe;
302     regs->uregs[13] = infop->start_stack;
303     /* FIXME - what to for failure of get_user()? */
304     get_user_ual(regs->uregs[2], stack + 8); /* envp */
305     get_user_ual(regs->uregs[1], stack + 4); /* envp */
306     /* XXX: it seems that r0 is zeroed after ! */
307     regs->uregs[0] = 0;
308     /* For uClinux PIC binaries.  */
309     /* XXX: Linux does this only on ARM with no MMU (do we care ?) */
310     regs->uregs[10] = infop->start_data;
311 
312     /* Support ARM FDPIC.  */
313     if (info_is_fdpic(infop)) {
314         /* As described in the ABI document, r7 points to the loadmap info
315          * prepared by the kernel. If an interpreter is needed, r8 points
316          * to the interpreter loadmap and r9 points to the interpreter
317          * PT_DYNAMIC info. If no interpreter is needed, r8 is zero, and
318          * r9 points to the main program PT_DYNAMIC info.
319          */
320         regs->uregs[7] = infop->loadmap_addr;
321         if (infop->interpreter_loadmap_addr) {
322             /* Executable is dynamically loaded.  */
323             regs->uregs[8] = infop->interpreter_loadmap_addr;
324             regs->uregs[9] = infop->interpreter_pt_dynamic_addr;
325         } else {
326             regs->uregs[8] = 0;
327             regs->uregs[9] = infop->pt_dynamic_addr;
328         }
329     }
330 }
331 
332 #define ELF_NREG    18
333 typedef target_elf_greg_t  target_elf_gregset_t[ELF_NREG];
334 
335 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUARMState *env)
336 {
337     (*regs)[0] = tswapreg(env->regs[0]);
338     (*regs)[1] = tswapreg(env->regs[1]);
339     (*regs)[2] = tswapreg(env->regs[2]);
340     (*regs)[3] = tswapreg(env->regs[3]);
341     (*regs)[4] = tswapreg(env->regs[4]);
342     (*regs)[5] = tswapreg(env->regs[5]);
343     (*regs)[6] = tswapreg(env->regs[6]);
344     (*regs)[7] = tswapreg(env->regs[7]);
345     (*regs)[8] = tswapreg(env->regs[8]);
346     (*regs)[9] = tswapreg(env->regs[9]);
347     (*regs)[10] = tswapreg(env->regs[10]);
348     (*regs)[11] = tswapreg(env->regs[11]);
349     (*regs)[12] = tswapreg(env->regs[12]);
350     (*regs)[13] = tswapreg(env->regs[13]);
351     (*regs)[14] = tswapreg(env->regs[14]);
352     (*regs)[15] = tswapreg(env->regs[15]);
353 
354     (*regs)[16] = tswapreg(cpsr_read((CPUARMState *)env));
355     (*regs)[17] = tswapreg(env->regs[0]); /* XXX */
356 }
357 
358 #define USE_ELF_CORE_DUMP
359 #define ELF_EXEC_PAGESIZE       4096
360 
361 enum
362 {
363     ARM_HWCAP_ARM_SWP       = 1 << 0,
364     ARM_HWCAP_ARM_HALF      = 1 << 1,
365     ARM_HWCAP_ARM_THUMB     = 1 << 2,
366     ARM_HWCAP_ARM_26BIT     = 1 << 3,
367     ARM_HWCAP_ARM_FAST_MULT = 1 << 4,
368     ARM_HWCAP_ARM_FPA       = 1 << 5,
369     ARM_HWCAP_ARM_VFP       = 1 << 6,
370     ARM_HWCAP_ARM_EDSP      = 1 << 7,
371     ARM_HWCAP_ARM_JAVA      = 1 << 8,
372     ARM_HWCAP_ARM_IWMMXT    = 1 << 9,
373     ARM_HWCAP_ARM_CRUNCH    = 1 << 10,
374     ARM_HWCAP_ARM_THUMBEE   = 1 << 11,
375     ARM_HWCAP_ARM_NEON      = 1 << 12,
376     ARM_HWCAP_ARM_VFPv3     = 1 << 13,
377     ARM_HWCAP_ARM_VFPv3D16  = 1 << 14,
378     ARM_HWCAP_ARM_TLS       = 1 << 15,
379     ARM_HWCAP_ARM_VFPv4     = 1 << 16,
380     ARM_HWCAP_ARM_IDIVA     = 1 << 17,
381     ARM_HWCAP_ARM_IDIVT     = 1 << 18,
382     ARM_HWCAP_ARM_VFPD32    = 1 << 19,
383     ARM_HWCAP_ARM_LPAE      = 1 << 20,
384     ARM_HWCAP_ARM_EVTSTRM   = 1 << 21,
385 };
386 
387 enum {
388     ARM_HWCAP2_ARM_AES      = 1 << 0,
389     ARM_HWCAP2_ARM_PMULL    = 1 << 1,
390     ARM_HWCAP2_ARM_SHA1     = 1 << 2,
391     ARM_HWCAP2_ARM_SHA2     = 1 << 3,
392     ARM_HWCAP2_ARM_CRC32    = 1 << 4,
393 };
394 
395 /* The commpage only exists for 32 bit kernels */
396 
397 #define HI_COMMPAGE (intptr_t)0xffff0f00u
398 
399 static bool init_guest_commpage(void)
400 {
401     void *want = g2h_untagged(HI_COMMPAGE & -qemu_host_page_size);
402     void *addr = mmap(want, qemu_host_page_size, PROT_READ | PROT_WRITE,
403                       MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0);
404 
405     if (addr == MAP_FAILED) {
406         perror("Allocating guest commpage");
407         exit(EXIT_FAILURE);
408     }
409     if (addr != want) {
410         return false;
411     }
412 
413     /* Set kernel helper versions; rest of page is 0.  */
414     __put_user(5, (uint32_t *)g2h_untagged(0xffff0ffcu));
415 
416     if (mprotect(addr, qemu_host_page_size, PROT_READ)) {
417         perror("Protecting guest commpage");
418         exit(EXIT_FAILURE);
419     }
420     return true;
421 }
422 
423 #define ELF_HWCAP get_elf_hwcap()
424 #define ELF_HWCAP2 get_elf_hwcap2()
425 
426 static uint32_t get_elf_hwcap(void)
427 {
428     ARMCPU *cpu = ARM_CPU(thread_cpu);
429     uint32_t hwcaps = 0;
430 
431     hwcaps |= ARM_HWCAP_ARM_SWP;
432     hwcaps |= ARM_HWCAP_ARM_HALF;
433     hwcaps |= ARM_HWCAP_ARM_THUMB;
434     hwcaps |= ARM_HWCAP_ARM_FAST_MULT;
435 
436     /* probe for the extra features */
437 #define GET_FEATURE(feat, hwcap) \
438     do { if (arm_feature(&cpu->env, feat)) { hwcaps |= hwcap; } } while (0)
439 
440 #define GET_FEATURE_ID(feat, hwcap) \
441     do { if (cpu_isar_feature(feat, cpu)) { hwcaps |= hwcap; } } while (0)
442 
443     /* EDSP is in v5TE and above, but all our v5 CPUs are v5TE */
444     GET_FEATURE(ARM_FEATURE_V5, ARM_HWCAP_ARM_EDSP);
445     GET_FEATURE(ARM_FEATURE_IWMMXT, ARM_HWCAP_ARM_IWMMXT);
446     GET_FEATURE(ARM_FEATURE_THUMB2EE, ARM_HWCAP_ARM_THUMBEE);
447     GET_FEATURE(ARM_FEATURE_NEON, ARM_HWCAP_ARM_NEON);
448     GET_FEATURE(ARM_FEATURE_V6K, ARM_HWCAP_ARM_TLS);
449     GET_FEATURE(ARM_FEATURE_LPAE, ARM_HWCAP_ARM_LPAE);
450     GET_FEATURE_ID(aa32_arm_div, ARM_HWCAP_ARM_IDIVA);
451     GET_FEATURE_ID(aa32_thumb_div, ARM_HWCAP_ARM_IDIVT);
452     GET_FEATURE_ID(aa32_vfp, ARM_HWCAP_ARM_VFP);
453 
454     if (cpu_isar_feature(aa32_fpsp_v3, cpu) ||
455         cpu_isar_feature(aa32_fpdp_v3, cpu)) {
456         hwcaps |= ARM_HWCAP_ARM_VFPv3;
457         if (cpu_isar_feature(aa32_simd_r32, cpu)) {
458             hwcaps |= ARM_HWCAP_ARM_VFPD32;
459         } else {
460             hwcaps |= ARM_HWCAP_ARM_VFPv3D16;
461         }
462     }
463     GET_FEATURE_ID(aa32_simdfmac, ARM_HWCAP_ARM_VFPv4);
464 
465     return hwcaps;
466 }
467 
468 static uint32_t get_elf_hwcap2(void)
469 {
470     ARMCPU *cpu = ARM_CPU(thread_cpu);
471     uint32_t hwcaps = 0;
472 
473     GET_FEATURE_ID(aa32_aes, ARM_HWCAP2_ARM_AES);
474     GET_FEATURE_ID(aa32_pmull, ARM_HWCAP2_ARM_PMULL);
475     GET_FEATURE_ID(aa32_sha1, ARM_HWCAP2_ARM_SHA1);
476     GET_FEATURE_ID(aa32_sha2, ARM_HWCAP2_ARM_SHA2);
477     GET_FEATURE_ID(aa32_crc32, ARM_HWCAP2_ARM_CRC32);
478     return hwcaps;
479 }
480 
481 #undef GET_FEATURE
482 #undef GET_FEATURE_ID
483 
484 #define ELF_PLATFORM get_elf_platform()
485 
486 static const char *get_elf_platform(void)
487 {
488     CPUARMState *env = thread_cpu->env_ptr;
489 
490 #if TARGET_BIG_ENDIAN
491 # define END  "b"
492 #else
493 # define END  "l"
494 #endif
495 
496     if (arm_feature(env, ARM_FEATURE_V8)) {
497         return "v8" END;
498     } else if (arm_feature(env, ARM_FEATURE_V7)) {
499         if (arm_feature(env, ARM_FEATURE_M)) {
500             return "v7m" END;
501         } else {
502             return "v7" END;
503         }
504     } else if (arm_feature(env, ARM_FEATURE_V6)) {
505         return "v6" END;
506     } else if (arm_feature(env, ARM_FEATURE_V5)) {
507         return "v5" END;
508     } else {
509         return "v4" END;
510     }
511 
512 #undef END
513 }
514 
515 #else
516 /* 64 bit ARM definitions */
517 #define ELF_START_MMAP 0x80000000
518 
519 #define ELF_ARCH        EM_AARCH64
520 #define ELF_CLASS       ELFCLASS64
521 #if TARGET_BIG_ENDIAN
522 # define ELF_PLATFORM    "aarch64_be"
523 #else
524 # define ELF_PLATFORM    "aarch64"
525 #endif
526 
527 static inline void init_thread(struct target_pt_regs *regs,
528                                struct image_info *infop)
529 {
530     abi_long stack = infop->start_stack;
531     memset(regs, 0, sizeof(*regs));
532 
533     regs->pc = infop->entry & ~0x3ULL;
534     regs->sp = stack;
535 }
536 
537 #define ELF_NREG    34
538 typedef target_elf_greg_t  target_elf_gregset_t[ELF_NREG];
539 
540 static void elf_core_copy_regs(target_elf_gregset_t *regs,
541                                const CPUARMState *env)
542 {
543     int i;
544 
545     for (i = 0; i < 32; i++) {
546         (*regs)[i] = tswapreg(env->xregs[i]);
547     }
548     (*regs)[32] = tswapreg(env->pc);
549     (*regs)[33] = tswapreg(pstate_read((CPUARMState *)env));
550 }
551 
552 #define USE_ELF_CORE_DUMP
553 #define ELF_EXEC_PAGESIZE       4096
554 
555 enum {
556     ARM_HWCAP_A64_FP            = 1 << 0,
557     ARM_HWCAP_A64_ASIMD         = 1 << 1,
558     ARM_HWCAP_A64_EVTSTRM       = 1 << 2,
559     ARM_HWCAP_A64_AES           = 1 << 3,
560     ARM_HWCAP_A64_PMULL         = 1 << 4,
561     ARM_HWCAP_A64_SHA1          = 1 << 5,
562     ARM_HWCAP_A64_SHA2          = 1 << 6,
563     ARM_HWCAP_A64_CRC32         = 1 << 7,
564     ARM_HWCAP_A64_ATOMICS       = 1 << 8,
565     ARM_HWCAP_A64_FPHP          = 1 << 9,
566     ARM_HWCAP_A64_ASIMDHP       = 1 << 10,
567     ARM_HWCAP_A64_CPUID         = 1 << 11,
568     ARM_HWCAP_A64_ASIMDRDM      = 1 << 12,
569     ARM_HWCAP_A64_JSCVT         = 1 << 13,
570     ARM_HWCAP_A64_FCMA          = 1 << 14,
571     ARM_HWCAP_A64_LRCPC         = 1 << 15,
572     ARM_HWCAP_A64_DCPOP         = 1 << 16,
573     ARM_HWCAP_A64_SHA3          = 1 << 17,
574     ARM_HWCAP_A64_SM3           = 1 << 18,
575     ARM_HWCAP_A64_SM4           = 1 << 19,
576     ARM_HWCAP_A64_ASIMDDP       = 1 << 20,
577     ARM_HWCAP_A64_SHA512        = 1 << 21,
578     ARM_HWCAP_A64_SVE           = 1 << 22,
579     ARM_HWCAP_A64_ASIMDFHM      = 1 << 23,
580     ARM_HWCAP_A64_DIT           = 1 << 24,
581     ARM_HWCAP_A64_USCAT         = 1 << 25,
582     ARM_HWCAP_A64_ILRCPC        = 1 << 26,
583     ARM_HWCAP_A64_FLAGM         = 1 << 27,
584     ARM_HWCAP_A64_SSBS          = 1 << 28,
585     ARM_HWCAP_A64_SB            = 1 << 29,
586     ARM_HWCAP_A64_PACA          = 1 << 30,
587     ARM_HWCAP_A64_PACG          = 1UL << 31,
588 
589     ARM_HWCAP2_A64_DCPODP       = 1 << 0,
590     ARM_HWCAP2_A64_SVE2         = 1 << 1,
591     ARM_HWCAP2_A64_SVEAES       = 1 << 2,
592     ARM_HWCAP2_A64_SVEPMULL     = 1 << 3,
593     ARM_HWCAP2_A64_SVEBITPERM   = 1 << 4,
594     ARM_HWCAP2_A64_SVESHA3      = 1 << 5,
595     ARM_HWCAP2_A64_SVESM4       = 1 << 6,
596     ARM_HWCAP2_A64_FLAGM2       = 1 << 7,
597     ARM_HWCAP2_A64_FRINT        = 1 << 8,
598     ARM_HWCAP2_A64_SVEI8MM      = 1 << 9,
599     ARM_HWCAP2_A64_SVEF32MM     = 1 << 10,
600     ARM_HWCAP2_A64_SVEF64MM     = 1 << 11,
601     ARM_HWCAP2_A64_SVEBF16      = 1 << 12,
602     ARM_HWCAP2_A64_I8MM         = 1 << 13,
603     ARM_HWCAP2_A64_BF16         = 1 << 14,
604     ARM_HWCAP2_A64_DGH          = 1 << 15,
605     ARM_HWCAP2_A64_RNG          = 1 << 16,
606     ARM_HWCAP2_A64_BTI          = 1 << 17,
607     ARM_HWCAP2_A64_MTE          = 1 << 18,
608 };
609 
610 #define ELF_HWCAP   get_elf_hwcap()
611 #define ELF_HWCAP2  get_elf_hwcap2()
612 
613 #define GET_FEATURE_ID(feat, hwcap) \
614     do { if (cpu_isar_feature(feat, cpu)) { hwcaps |= hwcap; } } while (0)
615 
616 static uint32_t get_elf_hwcap(void)
617 {
618     ARMCPU *cpu = ARM_CPU(thread_cpu);
619     uint32_t hwcaps = 0;
620 
621     hwcaps |= ARM_HWCAP_A64_FP;
622     hwcaps |= ARM_HWCAP_A64_ASIMD;
623     hwcaps |= ARM_HWCAP_A64_CPUID;
624 
625     /* probe for the extra features */
626 
627     GET_FEATURE_ID(aa64_aes, ARM_HWCAP_A64_AES);
628     GET_FEATURE_ID(aa64_pmull, ARM_HWCAP_A64_PMULL);
629     GET_FEATURE_ID(aa64_sha1, ARM_HWCAP_A64_SHA1);
630     GET_FEATURE_ID(aa64_sha256, ARM_HWCAP_A64_SHA2);
631     GET_FEATURE_ID(aa64_sha512, ARM_HWCAP_A64_SHA512);
632     GET_FEATURE_ID(aa64_crc32, ARM_HWCAP_A64_CRC32);
633     GET_FEATURE_ID(aa64_sha3, ARM_HWCAP_A64_SHA3);
634     GET_FEATURE_ID(aa64_sm3, ARM_HWCAP_A64_SM3);
635     GET_FEATURE_ID(aa64_sm4, ARM_HWCAP_A64_SM4);
636     GET_FEATURE_ID(aa64_fp16, ARM_HWCAP_A64_FPHP | ARM_HWCAP_A64_ASIMDHP);
637     GET_FEATURE_ID(aa64_atomics, ARM_HWCAP_A64_ATOMICS);
638     GET_FEATURE_ID(aa64_rdm, ARM_HWCAP_A64_ASIMDRDM);
639     GET_FEATURE_ID(aa64_dp, ARM_HWCAP_A64_ASIMDDP);
640     GET_FEATURE_ID(aa64_fcma, ARM_HWCAP_A64_FCMA);
641     GET_FEATURE_ID(aa64_sve, ARM_HWCAP_A64_SVE);
642     GET_FEATURE_ID(aa64_pauth, ARM_HWCAP_A64_PACA | ARM_HWCAP_A64_PACG);
643     GET_FEATURE_ID(aa64_fhm, ARM_HWCAP_A64_ASIMDFHM);
644     GET_FEATURE_ID(aa64_jscvt, ARM_HWCAP_A64_JSCVT);
645     GET_FEATURE_ID(aa64_sb, ARM_HWCAP_A64_SB);
646     GET_FEATURE_ID(aa64_condm_4, ARM_HWCAP_A64_FLAGM);
647     GET_FEATURE_ID(aa64_dcpop, ARM_HWCAP_A64_DCPOP);
648     GET_FEATURE_ID(aa64_rcpc_8_3, ARM_HWCAP_A64_LRCPC);
649     GET_FEATURE_ID(aa64_rcpc_8_4, ARM_HWCAP_A64_ILRCPC);
650 
651     return hwcaps;
652 }
653 
654 static uint32_t get_elf_hwcap2(void)
655 {
656     ARMCPU *cpu = ARM_CPU(thread_cpu);
657     uint32_t hwcaps = 0;
658 
659     GET_FEATURE_ID(aa64_dcpodp, ARM_HWCAP2_A64_DCPODP);
660     GET_FEATURE_ID(aa64_sve2, ARM_HWCAP2_A64_SVE2);
661     GET_FEATURE_ID(aa64_sve2_aes, ARM_HWCAP2_A64_SVEAES);
662     GET_FEATURE_ID(aa64_sve2_pmull128, ARM_HWCAP2_A64_SVEPMULL);
663     GET_FEATURE_ID(aa64_sve2_bitperm, ARM_HWCAP2_A64_SVEBITPERM);
664     GET_FEATURE_ID(aa64_sve2_sha3, ARM_HWCAP2_A64_SVESHA3);
665     GET_FEATURE_ID(aa64_sve2_sm4, ARM_HWCAP2_A64_SVESM4);
666     GET_FEATURE_ID(aa64_condm_5, ARM_HWCAP2_A64_FLAGM2);
667     GET_FEATURE_ID(aa64_frint, ARM_HWCAP2_A64_FRINT);
668     GET_FEATURE_ID(aa64_sve_i8mm, ARM_HWCAP2_A64_SVEI8MM);
669     GET_FEATURE_ID(aa64_sve_f32mm, ARM_HWCAP2_A64_SVEF32MM);
670     GET_FEATURE_ID(aa64_sve_f64mm, ARM_HWCAP2_A64_SVEF64MM);
671     GET_FEATURE_ID(aa64_sve_bf16, ARM_HWCAP2_A64_SVEBF16);
672     GET_FEATURE_ID(aa64_i8mm, ARM_HWCAP2_A64_I8MM);
673     GET_FEATURE_ID(aa64_bf16, ARM_HWCAP2_A64_BF16);
674     GET_FEATURE_ID(aa64_rndr, ARM_HWCAP2_A64_RNG);
675     GET_FEATURE_ID(aa64_bti, ARM_HWCAP2_A64_BTI);
676     GET_FEATURE_ID(aa64_mte, ARM_HWCAP2_A64_MTE);
677 
678     return hwcaps;
679 }
680 
681 #undef GET_FEATURE_ID
682 
683 #endif /* not TARGET_AARCH64 */
684 #endif /* TARGET_ARM */
685 
686 #ifdef TARGET_SPARC
687 #ifdef TARGET_SPARC64
688 
689 #define ELF_START_MMAP 0x80000000
690 #define ELF_HWCAP  (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \
691                     | HWCAP_SPARC_MULDIV | HWCAP_SPARC_V9)
692 #ifndef TARGET_ABI32
693 #define elf_check_arch(x) ( (x) == EM_SPARCV9 || (x) == EM_SPARC32PLUS )
694 #else
695 #define elf_check_arch(x) ( (x) == EM_SPARC32PLUS || (x) == EM_SPARC )
696 #endif
697 
698 #define ELF_CLASS   ELFCLASS64
699 #define ELF_ARCH    EM_SPARCV9
700 #else
701 #define ELF_START_MMAP 0x80000000
702 #define ELF_HWCAP  (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \
703                     | HWCAP_SPARC_MULDIV)
704 #define ELF_CLASS   ELFCLASS32
705 #define ELF_ARCH    EM_SPARC
706 #endif /* TARGET_SPARC64 */
707 
708 static inline void init_thread(struct target_pt_regs *regs,
709                                struct image_info *infop)
710 {
711     /* Note that target_cpu_copy_regs does not read psr/tstate. */
712     regs->pc = infop->entry;
713     regs->npc = regs->pc + 4;
714     regs->y = 0;
715     regs->u_regs[14] = (infop->start_stack - 16 * sizeof(abi_ulong)
716                         - TARGET_STACK_BIAS);
717 }
718 #endif /* TARGET_SPARC */
719 
720 #ifdef TARGET_PPC
721 
722 #define ELF_MACHINE    PPC_ELF_MACHINE
723 #define ELF_START_MMAP 0x80000000
724 
725 #if defined(TARGET_PPC64)
726 
727 #define elf_check_arch(x) ( (x) == EM_PPC64 )
728 
729 #define ELF_CLASS       ELFCLASS64
730 
731 #else
732 
733 #define ELF_CLASS       ELFCLASS32
734 
735 #endif
736 
737 #define ELF_ARCH        EM_PPC
738 
739 /* Feature masks for the Aux Vector Hardware Capabilities (AT_HWCAP).
740    See arch/powerpc/include/asm/cputable.h.  */
741 enum {
742     QEMU_PPC_FEATURE_32 = 0x80000000,
743     QEMU_PPC_FEATURE_64 = 0x40000000,
744     QEMU_PPC_FEATURE_601_INSTR = 0x20000000,
745     QEMU_PPC_FEATURE_HAS_ALTIVEC = 0x10000000,
746     QEMU_PPC_FEATURE_HAS_FPU = 0x08000000,
747     QEMU_PPC_FEATURE_HAS_MMU = 0x04000000,
748     QEMU_PPC_FEATURE_HAS_4xxMAC = 0x02000000,
749     QEMU_PPC_FEATURE_UNIFIED_CACHE = 0x01000000,
750     QEMU_PPC_FEATURE_HAS_SPE = 0x00800000,
751     QEMU_PPC_FEATURE_HAS_EFP_SINGLE = 0x00400000,
752     QEMU_PPC_FEATURE_HAS_EFP_DOUBLE = 0x00200000,
753     QEMU_PPC_FEATURE_NO_TB = 0x00100000,
754     QEMU_PPC_FEATURE_POWER4 = 0x00080000,
755     QEMU_PPC_FEATURE_POWER5 = 0x00040000,
756     QEMU_PPC_FEATURE_POWER5_PLUS = 0x00020000,
757     QEMU_PPC_FEATURE_CELL = 0x00010000,
758     QEMU_PPC_FEATURE_BOOKE = 0x00008000,
759     QEMU_PPC_FEATURE_SMT = 0x00004000,
760     QEMU_PPC_FEATURE_ICACHE_SNOOP = 0x00002000,
761     QEMU_PPC_FEATURE_ARCH_2_05 = 0x00001000,
762     QEMU_PPC_FEATURE_PA6T = 0x00000800,
763     QEMU_PPC_FEATURE_HAS_DFP = 0x00000400,
764     QEMU_PPC_FEATURE_POWER6_EXT = 0x00000200,
765     QEMU_PPC_FEATURE_ARCH_2_06 = 0x00000100,
766     QEMU_PPC_FEATURE_HAS_VSX = 0x00000080,
767     QEMU_PPC_FEATURE_PSERIES_PERFMON_COMPAT = 0x00000040,
768 
769     QEMU_PPC_FEATURE_TRUE_LE = 0x00000002,
770     QEMU_PPC_FEATURE_PPC_LE = 0x00000001,
771 
772     /* Feature definitions in AT_HWCAP2.  */
773     QEMU_PPC_FEATURE2_ARCH_2_07 = 0x80000000, /* ISA 2.07 */
774     QEMU_PPC_FEATURE2_HAS_HTM = 0x40000000, /* Hardware Transactional Memory */
775     QEMU_PPC_FEATURE2_HAS_DSCR = 0x20000000, /* Data Stream Control Register */
776     QEMU_PPC_FEATURE2_HAS_EBB = 0x10000000, /* Event Base Branching */
777     QEMU_PPC_FEATURE2_HAS_ISEL = 0x08000000, /* Integer Select */
778     QEMU_PPC_FEATURE2_HAS_TAR = 0x04000000, /* Target Address Register */
779     QEMU_PPC_FEATURE2_VEC_CRYPTO = 0x02000000,
780     QEMU_PPC_FEATURE2_HTM_NOSC = 0x01000000,
781     QEMU_PPC_FEATURE2_ARCH_3_00 = 0x00800000, /* ISA 3.00 */
782     QEMU_PPC_FEATURE2_HAS_IEEE128 = 0x00400000, /* VSX IEEE Bin Float 128-bit */
783     QEMU_PPC_FEATURE2_DARN = 0x00200000, /* darn random number insn */
784     QEMU_PPC_FEATURE2_SCV = 0x00100000, /* scv syscall */
785     QEMU_PPC_FEATURE2_HTM_NO_SUSPEND = 0x00080000, /* TM w/o suspended state */
786     QEMU_PPC_FEATURE2_ARCH_3_1 = 0x00040000, /* ISA 3.1 */
787     QEMU_PPC_FEATURE2_MMA = 0x00020000, /* Matrix-Multiply Assist */
788 };
789 
790 #define ELF_HWCAP get_elf_hwcap()
791 
792 static uint32_t get_elf_hwcap(void)
793 {
794     PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);
795     uint32_t features = 0;
796 
797     /* We don't have to be terribly complete here; the high points are
798        Altivec/FP/SPE support.  Anything else is just a bonus.  */
799 #define GET_FEATURE(flag, feature)                                      \
800     do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0)
801 #define GET_FEATURE2(flags, feature) \
802     do { \
803         if ((cpu->env.insns_flags2 & flags) == flags) { \
804             features |= feature; \
805         } \
806     } while (0)
807     GET_FEATURE(PPC_64B, QEMU_PPC_FEATURE_64);
808     GET_FEATURE(PPC_FLOAT, QEMU_PPC_FEATURE_HAS_FPU);
809     GET_FEATURE(PPC_ALTIVEC, QEMU_PPC_FEATURE_HAS_ALTIVEC);
810     GET_FEATURE(PPC_SPE, QEMU_PPC_FEATURE_HAS_SPE);
811     GET_FEATURE(PPC_SPE_SINGLE, QEMU_PPC_FEATURE_HAS_EFP_SINGLE);
812     GET_FEATURE(PPC_SPE_DOUBLE, QEMU_PPC_FEATURE_HAS_EFP_DOUBLE);
813     GET_FEATURE(PPC_BOOKE, QEMU_PPC_FEATURE_BOOKE);
814     GET_FEATURE(PPC_405_MAC, QEMU_PPC_FEATURE_HAS_4xxMAC);
815     GET_FEATURE2(PPC2_DFP, QEMU_PPC_FEATURE_HAS_DFP);
816     GET_FEATURE2(PPC2_VSX, QEMU_PPC_FEATURE_HAS_VSX);
817     GET_FEATURE2((PPC2_PERM_ISA206 | PPC2_DIVE_ISA206 | PPC2_ATOMIC_ISA206 |
818                   PPC2_FP_CVT_ISA206 | PPC2_FP_TST_ISA206),
819                   QEMU_PPC_FEATURE_ARCH_2_06);
820 #undef GET_FEATURE
821 #undef GET_FEATURE2
822 
823     return features;
824 }
825 
826 #define ELF_HWCAP2 get_elf_hwcap2()
827 
828 static uint32_t get_elf_hwcap2(void)
829 {
830     PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);
831     uint32_t features = 0;
832 
833 #define GET_FEATURE(flag, feature)                                      \
834     do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0)
835 #define GET_FEATURE2(flag, feature)                                      \
836     do { if (cpu->env.insns_flags2 & flag) { features |= feature; } } while (0)
837 
838     GET_FEATURE(PPC_ISEL, QEMU_PPC_FEATURE2_HAS_ISEL);
839     GET_FEATURE2(PPC2_BCTAR_ISA207, QEMU_PPC_FEATURE2_HAS_TAR);
840     GET_FEATURE2((PPC2_BCTAR_ISA207 | PPC2_LSQ_ISA207 | PPC2_ALTIVEC_207 |
841                   PPC2_ISA207S), QEMU_PPC_FEATURE2_ARCH_2_07 |
842                   QEMU_PPC_FEATURE2_VEC_CRYPTO);
843     GET_FEATURE2(PPC2_ISA300, QEMU_PPC_FEATURE2_ARCH_3_00 |
844                  QEMU_PPC_FEATURE2_DARN | QEMU_PPC_FEATURE2_HAS_IEEE128);
845     GET_FEATURE2(PPC2_ISA310, QEMU_PPC_FEATURE2_ARCH_3_1 |
846                  QEMU_PPC_FEATURE2_MMA);
847 
848 #undef GET_FEATURE
849 #undef GET_FEATURE2
850 
851     return features;
852 }
853 
854 /*
855  * The requirements here are:
856  * - keep the final alignment of sp (sp & 0xf)
857  * - make sure the 32-bit value at the first 16 byte aligned position of
858  *   AUXV is greater than 16 for glibc compatibility.
859  *   AT_IGNOREPPC is used for that.
860  * - for compatibility with glibc ARCH_DLINFO must always be defined on PPC,
861  *   even if DLINFO_ARCH_ITEMS goes to zero or is undefined.
862  */
863 #define DLINFO_ARCH_ITEMS       5
864 #define ARCH_DLINFO                                     \
865     do {                                                \
866         PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);              \
867         /*                                              \
868          * Handle glibc compatibility: these magic entries must \
869          * be at the lowest addresses in the final auxv.        \
870          */                                             \
871         NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC);        \
872         NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC);        \
873         NEW_AUX_ENT(AT_DCACHEBSIZE, cpu->env.dcache_line_size); \
874         NEW_AUX_ENT(AT_ICACHEBSIZE, cpu->env.icache_line_size); \
875         NEW_AUX_ENT(AT_UCACHEBSIZE, 0);                 \
876     } while (0)
877 
878 static inline void init_thread(struct target_pt_regs *_regs, struct image_info *infop)
879 {
880     _regs->gpr[1] = infop->start_stack;
881 #if defined(TARGET_PPC64)
882     if (get_ppc64_abi(infop) < 2) {
883         uint64_t val;
884         get_user_u64(val, infop->entry + 8);
885         _regs->gpr[2] = val + infop->load_bias;
886         get_user_u64(val, infop->entry);
887         infop->entry = val + infop->load_bias;
888     } else {
889         _regs->gpr[12] = infop->entry;  /* r12 set to global entry address */
890     }
891 #endif
892     _regs->nip = infop->entry;
893 }
894 
895 /* See linux kernel: arch/powerpc/include/asm/elf.h.  */
896 #define ELF_NREG 48
897 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
898 
899 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUPPCState *env)
900 {
901     int i;
902     target_ulong ccr = 0;
903 
904     for (i = 0; i < ARRAY_SIZE(env->gpr); i++) {
905         (*regs)[i] = tswapreg(env->gpr[i]);
906     }
907 
908     (*regs)[32] = tswapreg(env->nip);
909     (*regs)[33] = tswapreg(env->msr);
910     (*regs)[35] = tswapreg(env->ctr);
911     (*regs)[36] = tswapreg(env->lr);
912     (*regs)[37] = tswapreg(cpu_read_xer(env));
913 
914     for (i = 0; i < ARRAY_SIZE(env->crf); i++) {
915         ccr |= env->crf[i] << (32 - ((i + 1) * 4));
916     }
917     (*regs)[38] = tswapreg(ccr);
918 }
919 
920 #define USE_ELF_CORE_DUMP
921 #define ELF_EXEC_PAGESIZE       4096
922 
923 #endif
924 
925 #ifdef TARGET_LOONGARCH64
926 
927 #define ELF_START_MMAP 0x80000000
928 
929 #define ELF_CLASS   ELFCLASS64
930 #define ELF_ARCH    EM_LOONGARCH
931 
932 #define elf_check_arch(x) ((x) == EM_LOONGARCH)
933 
934 static inline void init_thread(struct target_pt_regs *regs,
935                                struct image_info *infop)
936 {
937     /*Set crmd PG,DA = 1,0 */
938     regs->csr.crmd = 2 << 3;
939     regs->csr.era = infop->entry;
940     regs->regs[3] = infop->start_stack;
941 }
942 
943 /* See linux kernel: arch/loongarch/include/asm/elf.h */
944 #define ELF_NREG 45
945 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
946 
947 enum {
948     TARGET_EF_R0 = 0,
949     TARGET_EF_CSR_ERA = TARGET_EF_R0 + 33,
950     TARGET_EF_CSR_BADV = TARGET_EF_R0 + 34,
951 };
952 
953 static void elf_core_copy_regs(target_elf_gregset_t *regs,
954                                const CPULoongArchState *env)
955 {
956     int i;
957 
958     (*regs)[TARGET_EF_R0] = 0;
959 
960     for (i = 1; i < ARRAY_SIZE(env->gpr); i++) {
961         (*regs)[TARGET_EF_R0 + i] = tswapreg(env->gpr[i]);
962     }
963 
964     (*regs)[TARGET_EF_CSR_ERA] = tswapreg(env->pc);
965     (*regs)[TARGET_EF_CSR_BADV] = tswapreg(env->CSR_BADV);
966 }
967 
968 #define USE_ELF_CORE_DUMP
969 #define ELF_EXEC_PAGESIZE        4096
970 
971 #define ELF_HWCAP get_elf_hwcap()
972 
973 /* See arch/loongarch/include/uapi/asm/hwcap.h */
974 enum {
975     HWCAP_LOONGARCH_CPUCFG   = (1 << 0),
976     HWCAP_LOONGARCH_LAM      = (1 << 1),
977     HWCAP_LOONGARCH_UAL      = (1 << 2),
978     HWCAP_LOONGARCH_FPU      = (1 << 3),
979     HWCAP_LOONGARCH_LSX      = (1 << 4),
980     HWCAP_LOONGARCH_LASX     = (1 << 5),
981     HWCAP_LOONGARCH_CRC32    = (1 << 6),
982     HWCAP_LOONGARCH_COMPLEX  = (1 << 7),
983     HWCAP_LOONGARCH_CRYPTO   = (1 << 8),
984     HWCAP_LOONGARCH_LVZ      = (1 << 9),
985     HWCAP_LOONGARCH_LBT_X86  = (1 << 10),
986     HWCAP_LOONGARCH_LBT_ARM  = (1 << 11),
987     HWCAP_LOONGARCH_LBT_MIPS = (1 << 12),
988 };
989 
990 static uint32_t get_elf_hwcap(void)
991 {
992     LoongArchCPU *cpu = LOONGARCH_CPU(thread_cpu);
993     uint32_t hwcaps = 0;
994 
995     hwcaps |= HWCAP_LOONGARCH_CRC32;
996 
997     if (FIELD_EX32(cpu->env.cpucfg[1], CPUCFG1, UAL)) {
998         hwcaps |= HWCAP_LOONGARCH_UAL;
999     }
1000 
1001     if (FIELD_EX32(cpu->env.cpucfg[2], CPUCFG2, FP)) {
1002         hwcaps |= HWCAP_LOONGARCH_FPU;
1003     }
1004 
1005     if (FIELD_EX32(cpu->env.cpucfg[2], CPUCFG2, LAM)) {
1006         hwcaps |= HWCAP_LOONGARCH_LAM;
1007     }
1008 
1009     return hwcaps;
1010 }
1011 
1012 #define ELF_PLATFORM "loongarch"
1013 
1014 #endif /* TARGET_LOONGARCH64 */
1015 
1016 #ifdef TARGET_MIPS
1017 
1018 #define ELF_START_MMAP 0x80000000
1019 
1020 #ifdef TARGET_MIPS64
1021 #define ELF_CLASS   ELFCLASS64
1022 #else
1023 #define ELF_CLASS   ELFCLASS32
1024 #endif
1025 #define ELF_ARCH    EM_MIPS
1026 
1027 #ifdef TARGET_ABI_MIPSN32
1028 #define elf_check_abi(x) ((x) & EF_MIPS_ABI2)
1029 #else
1030 #define elf_check_abi(x) (!((x) & EF_MIPS_ABI2))
1031 #endif
1032 
1033 static inline void init_thread(struct target_pt_regs *regs,
1034                                struct image_info *infop)
1035 {
1036     regs->cp0_status = 2 << CP0St_KSU;
1037     regs->cp0_epc = infop->entry;
1038     regs->regs[29] = infop->start_stack;
1039 }
1040 
1041 /* See linux kernel: arch/mips/include/asm/elf.h.  */
1042 #define ELF_NREG 45
1043 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1044 
1045 /* See linux kernel: arch/mips/include/asm/reg.h.  */
1046 enum {
1047 #ifdef TARGET_MIPS64
1048     TARGET_EF_R0 = 0,
1049 #else
1050     TARGET_EF_R0 = 6,
1051 #endif
1052     TARGET_EF_R26 = TARGET_EF_R0 + 26,
1053     TARGET_EF_R27 = TARGET_EF_R0 + 27,
1054     TARGET_EF_LO = TARGET_EF_R0 + 32,
1055     TARGET_EF_HI = TARGET_EF_R0 + 33,
1056     TARGET_EF_CP0_EPC = TARGET_EF_R0 + 34,
1057     TARGET_EF_CP0_BADVADDR = TARGET_EF_R0 + 35,
1058     TARGET_EF_CP0_STATUS = TARGET_EF_R0 + 36,
1059     TARGET_EF_CP0_CAUSE = TARGET_EF_R0 + 37
1060 };
1061 
1062 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs.  */
1063 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMIPSState *env)
1064 {
1065     int i;
1066 
1067     for (i = 0; i < TARGET_EF_R0; i++) {
1068         (*regs)[i] = 0;
1069     }
1070     (*regs)[TARGET_EF_R0] = 0;
1071 
1072     for (i = 1; i < ARRAY_SIZE(env->active_tc.gpr); i++) {
1073         (*regs)[TARGET_EF_R0 + i] = tswapreg(env->active_tc.gpr[i]);
1074     }
1075 
1076     (*regs)[TARGET_EF_R26] = 0;
1077     (*regs)[TARGET_EF_R27] = 0;
1078     (*regs)[TARGET_EF_LO] = tswapreg(env->active_tc.LO[0]);
1079     (*regs)[TARGET_EF_HI] = tswapreg(env->active_tc.HI[0]);
1080     (*regs)[TARGET_EF_CP0_EPC] = tswapreg(env->active_tc.PC);
1081     (*regs)[TARGET_EF_CP0_BADVADDR] = tswapreg(env->CP0_BadVAddr);
1082     (*regs)[TARGET_EF_CP0_STATUS] = tswapreg(env->CP0_Status);
1083     (*regs)[TARGET_EF_CP0_CAUSE] = tswapreg(env->CP0_Cause);
1084 }
1085 
1086 #define USE_ELF_CORE_DUMP
1087 #define ELF_EXEC_PAGESIZE        4096
1088 
1089 /* See arch/mips/include/uapi/asm/hwcap.h.  */
1090 enum {
1091     HWCAP_MIPS_R6           = (1 << 0),
1092     HWCAP_MIPS_MSA          = (1 << 1),
1093     HWCAP_MIPS_CRC32        = (1 << 2),
1094     HWCAP_MIPS_MIPS16       = (1 << 3),
1095     HWCAP_MIPS_MDMX         = (1 << 4),
1096     HWCAP_MIPS_MIPS3D       = (1 << 5),
1097     HWCAP_MIPS_SMARTMIPS    = (1 << 6),
1098     HWCAP_MIPS_DSP          = (1 << 7),
1099     HWCAP_MIPS_DSP2         = (1 << 8),
1100     HWCAP_MIPS_DSP3         = (1 << 9),
1101     HWCAP_MIPS_MIPS16E2     = (1 << 10),
1102     HWCAP_LOONGSON_MMI      = (1 << 11),
1103     HWCAP_LOONGSON_EXT      = (1 << 12),
1104     HWCAP_LOONGSON_EXT2     = (1 << 13),
1105     HWCAP_LOONGSON_CPUCFG   = (1 << 14),
1106 };
1107 
1108 #define ELF_HWCAP get_elf_hwcap()
1109 
1110 #define GET_FEATURE_INSN(_flag, _hwcap) \
1111     do { if (cpu->env.insn_flags & (_flag)) { hwcaps |= _hwcap; } } while (0)
1112 
1113 #define GET_FEATURE_REG_SET(_reg, _mask, _hwcap) \
1114     do { if (cpu->env._reg & (_mask)) { hwcaps |= _hwcap; } } while (0)
1115 
1116 #define GET_FEATURE_REG_EQU(_reg, _start, _length, _val, _hwcap) \
1117     do { \
1118         if (extract32(cpu->env._reg, (_start), (_length)) == (_val)) { \
1119             hwcaps |= _hwcap; \
1120         } \
1121     } while (0)
1122 
1123 static uint32_t get_elf_hwcap(void)
1124 {
1125     MIPSCPU *cpu = MIPS_CPU(thread_cpu);
1126     uint32_t hwcaps = 0;
1127 
1128     GET_FEATURE_REG_EQU(CP0_Config0, CP0C0_AR, CP0C0_AR_LENGTH,
1129                         2, HWCAP_MIPS_R6);
1130     GET_FEATURE_REG_SET(CP0_Config3, 1 << CP0C3_MSAP, HWCAP_MIPS_MSA);
1131     GET_FEATURE_INSN(ASE_LMMI, HWCAP_LOONGSON_MMI);
1132     GET_FEATURE_INSN(ASE_LEXT, HWCAP_LOONGSON_EXT);
1133 
1134     return hwcaps;
1135 }
1136 
1137 #undef GET_FEATURE_REG_EQU
1138 #undef GET_FEATURE_REG_SET
1139 #undef GET_FEATURE_INSN
1140 
1141 #endif /* TARGET_MIPS */
1142 
1143 #ifdef TARGET_MICROBLAZE
1144 
1145 #define ELF_START_MMAP 0x80000000
1146 
1147 #define elf_check_arch(x) ( (x) == EM_MICROBLAZE || (x) == EM_MICROBLAZE_OLD)
1148 
1149 #define ELF_CLASS   ELFCLASS32
1150 #define ELF_ARCH    EM_MICROBLAZE
1151 
1152 static inline void init_thread(struct target_pt_regs *regs,
1153                                struct image_info *infop)
1154 {
1155     regs->pc = infop->entry;
1156     regs->r1 = infop->start_stack;
1157 
1158 }
1159 
1160 #define ELF_EXEC_PAGESIZE        4096
1161 
1162 #define USE_ELF_CORE_DUMP
1163 #define ELF_NREG 38
1164 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1165 
1166 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs.  */
1167 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMBState *env)
1168 {
1169     int i, pos = 0;
1170 
1171     for (i = 0; i < 32; i++) {
1172         (*regs)[pos++] = tswapreg(env->regs[i]);
1173     }
1174 
1175     (*regs)[pos++] = tswapreg(env->pc);
1176     (*regs)[pos++] = tswapreg(mb_cpu_read_msr(env));
1177     (*regs)[pos++] = 0;
1178     (*regs)[pos++] = tswapreg(env->ear);
1179     (*regs)[pos++] = 0;
1180     (*regs)[pos++] = tswapreg(env->esr);
1181 }
1182 
1183 #endif /* TARGET_MICROBLAZE */
1184 
1185 #ifdef TARGET_NIOS2
1186 
1187 #define ELF_START_MMAP 0x80000000
1188 
1189 #define elf_check_arch(x) ((x) == EM_ALTERA_NIOS2)
1190 
1191 #define ELF_CLASS   ELFCLASS32
1192 #define ELF_ARCH    EM_ALTERA_NIOS2
1193 
1194 static void init_thread(struct target_pt_regs *regs, struct image_info *infop)
1195 {
1196     regs->ea = infop->entry;
1197     regs->sp = infop->start_stack;
1198 }
1199 
1200 #define LO_COMMPAGE  TARGET_PAGE_SIZE
1201 
1202 static bool init_guest_commpage(void)
1203 {
1204     static const uint8_t kuser_page[4 + 2 * 64] = {
1205         /* __kuser_helper_version */
1206         [0x00] = 0x02, 0x00, 0x00, 0x00,
1207 
1208         /* __kuser_cmpxchg */
1209         [0x04] = 0x3a, 0x6c, 0x3b, 0x00,  /* trap 16 */
1210                  0x3a, 0x28, 0x00, 0xf8,  /* ret */
1211 
1212         /* __kuser_sigtramp */
1213         [0x44] = 0xc4, 0x22, 0x80, 0x00,  /* movi r2, __NR_rt_sigreturn */
1214                  0x3a, 0x68, 0x3b, 0x00,  /* trap 0 */
1215     };
1216 
1217     void *want = g2h_untagged(LO_COMMPAGE & -qemu_host_page_size);
1218     void *addr = mmap(want, qemu_host_page_size, PROT_READ | PROT_WRITE,
1219                       MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0);
1220 
1221     if (addr == MAP_FAILED) {
1222         perror("Allocating guest commpage");
1223         exit(EXIT_FAILURE);
1224     }
1225     if (addr != want) {
1226         return false;
1227     }
1228 
1229     memcpy(addr, kuser_page, sizeof(kuser_page));
1230 
1231     if (mprotect(addr, qemu_host_page_size, PROT_READ)) {
1232         perror("Protecting guest commpage");
1233         exit(EXIT_FAILURE);
1234     }
1235 
1236     page_set_flags(LO_COMMPAGE, LO_COMMPAGE + TARGET_PAGE_SIZE,
1237                    PAGE_READ | PAGE_EXEC | PAGE_VALID);
1238     return true;
1239 }
1240 
1241 #define ELF_EXEC_PAGESIZE        4096
1242 
1243 #define USE_ELF_CORE_DUMP
1244 #define ELF_NREG 49
1245 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1246 
1247 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs.  */
1248 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1249                                const CPUNios2State *env)
1250 {
1251     int i;
1252 
1253     (*regs)[0] = -1;
1254     for (i = 1; i < 8; i++)    /* r0-r7 */
1255         (*regs)[i] = tswapreg(env->regs[i + 7]);
1256 
1257     for (i = 8; i < 16; i++)   /* r8-r15 */
1258         (*regs)[i] = tswapreg(env->regs[i - 8]);
1259 
1260     for (i = 16; i < 24; i++)  /* r16-r23 */
1261         (*regs)[i] = tswapreg(env->regs[i + 7]);
1262     (*regs)[24] = -1;    /* R_ET */
1263     (*regs)[25] = -1;    /* R_BT */
1264     (*regs)[26] = tswapreg(env->regs[R_GP]);
1265     (*regs)[27] = tswapreg(env->regs[R_SP]);
1266     (*regs)[28] = tswapreg(env->regs[R_FP]);
1267     (*regs)[29] = tswapreg(env->regs[R_EA]);
1268     (*regs)[30] = -1;    /* R_SSTATUS */
1269     (*regs)[31] = tswapreg(env->regs[R_RA]);
1270 
1271     (*regs)[32] = tswapreg(env->pc);
1272 
1273     (*regs)[33] = -1; /* R_STATUS */
1274     (*regs)[34] = tswapreg(env->regs[CR_ESTATUS]);
1275 
1276     for (i = 35; i < 49; i++)    /* ... */
1277         (*regs)[i] = -1;
1278 }
1279 
1280 #endif /* TARGET_NIOS2 */
1281 
1282 #ifdef TARGET_OPENRISC
1283 
1284 #define ELF_START_MMAP 0x08000000
1285 
1286 #define ELF_ARCH EM_OPENRISC
1287 #define ELF_CLASS ELFCLASS32
1288 #define ELF_DATA  ELFDATA2MSB
1289 
1290 static inline void init_thread(struct target_pt_regs *regs,
1291                                struct image_info *infop)
1292 {
1293     regs->pc = infop->entry;
1294     regs->gpr[1] = infop->start_stack;
1295 }
1296 
1297 #define USE_ELF_CORE_DUMP
1298 #define ELF_EXEC_PAGESIZE 8192
1299 
1300 /* See linux kernel arch/openrisc/include/asm/elf.h.  */
1301 #define ELF_NREG 34 /* gprs and pc, sr */
1302 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1303 
1304 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1305                                const CPUOpenRISCState *env)
1306 {
1307     int i;
1308 
1309     for (i = 0; i < 32; i++) {
1310         (*regs)[i] = tswapreg(cpu_get_gpr(env, i));
1311     }
1312     (*regs)[32] = tswapreg(env->pc);
1313     (*regs)[33] = tswapreg(cpu_get_sr(env));
1314 }
1315 #define ELF_HWCAP 0
1316 #define ELF_PLATFORM NULL
1317 
1318 #endif /* TARGET_OPENRISC */
1319 
1320 #ifdef TARGET_SH4
1321 
1322 #define ELF_START_MMAP 0x80000000
1323 
1324 #define ELF_CLASS ELFCLASS32
1325 #define ELF_ARCH  EM_SH
1326 
1327 static inline void init_thread(struct target_pt_regs *regs,
1328                                struct image_info *infop)
1329 {
1330     /* Check other registers XXXXX */
1331     regs->pc = infop->entry;
1332     regs->regs[15] = infop->start_stack;
1333 }
1334 
1335 /* See linux kernel: arch/sh/include/asm/elf.h.  */
1336 #define ELF_NREG 23
1337 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1338 
1339 /* See linux kernel: arch/sh/include/asm/ptrace.h.  */
1340 enum {
1341     TARGET_REG_PC = 16,
1342     TARGET_REG_PR = 17,
1343     TARGET_REG_SR = 18,
1344     TARGET_REG_GBR = 19,
1345     TARGET_REG_MACH = 20,
1346     TARGET_REG_MACL = 21,
1347     TARGET_REG_SYSCALL = 22
1348 };
1349 
1350 static inline void elf_core_copy_regs(target_elf_gregset_t *regs,
1351                                       const CPUSH4State *env)
1352 {
1353     int i;
1354 
1355     for (i = 0; i < 16; i++) {
1356         (*regs)[i] = tswapreg(env->gregs[i]);
1357     }
1358 
1359     (*regs)[TARGET_REG_PC] = tswapreg(env->pc);
1360     (*regs)[TARGET_REG_PR] = tswapreg(env->pr);
1361     (*regs)[TARGET_REG_SR] = tswapreg(env->sr);
1362     (*regs)[TARGET_REG_GBR] = tswapreg(env->gbr);
1363     (*regs)[TARGET_REG_MACH] = tswapreg(env->mach);
1364     (*regs)[TARGET_REG_MACL] = tswapreg(env->macl);
1365     (*regs)[TARGET_REG_SYSCALL] = 0; /* FIXME */
1366 }
1367 
1368 #define USE_ELF_CORE_DUMP
1369 #define ELF_EXEC_PAGESIZE        4096
1370 
1371 enum {
1372     SH_CPU_HAS_FPU            = 0x0001, /* Hardware FPU support */
1373     SH_CPU_HAS_P2_FLUSH_BUG   = 0x0002, /* Need to flush the cache in P2 area */
1374     SH_CPU_HAS_MMU_PAGE_ASSOC = 0x0004, /* SH3: TLB way selection bit support */
1375     SH_CPU_HAS_DSP            = 0x0008, /* SH-DSP: DSP support */
1376     SH_CPU_HAS_PERF_COUNTER   = 0x0010, /* Hardware performance counters */
1377     SH_CPU_HAS_PTEA           = 0x0020, /* PTEA register */
1378     SH_CPU_HAS_LLSC           = 0x0040, /* movli.l/movco.l */
1379     SH_CPU_HAS_L2_CACHE       = 0x0080, /* Secondary cache / URAM */
1380     SH_CPU_HAS_OP32           = 0x0100, /* 32-bit instruction support */
1381     SH_CPU_HAS_PTEAEX         = 0x0200, /* PTE ASID Extension support */
1382 };
1383 
1384 #define ELF_HWCAP get_elf_hwcap()
1385 
1386 static uint32_t get_elf_hwcap(void)
1387 {
1388     SuperHCPU *cpu = SUPERH_CPU(thread_cpu);
1389     uint32_t hwcap = 0;
1390 
1391     hwcap |= SH_CPU_HAS_FPU;
1392 
1393     if (cpu->env.features & SH_FEATURE_SH4A) {
1394         hwcap |= SH_CPU_HAS_LLSC;
1395     }
1396 
1397     return hwcap;
1398 }
1399 
1400 #endif
1401 
1402 #ifdef TARGET_CRIS
1403 
1404 #define ELF_START_MMAP 0x80000000
1405 
1406 #define ELF_CLASS ELFCLASS32
1407 #define ELF_ARCH  EM_CRIS
1408 
1409 static inline void init_thread(struct target_pt_regs *regs,
1410                                struct image_info *infop)
1411 {
1412     regs->erp = infop->entry;
1413 }
1414 
1415 #define ELF_EXEC_PAGESIZE        8192
1416 
1417 #endif
1418 
1419 #ifdef TARGET_M68K
1420 
1421 #define ELF_START_MMAP 0x80000000
1422 
1423 #define ELF_CLASS       ELFCLASS32
1424 #define ELF_ARCH        EM_68K
1425 
1426 /* ??? Does this need to do anything?
1427    #define ELF_PLAT_INIT(_r) */
1428 
1429 static inline void init_thread(struct target_pt_regs *regs,
1430                                struct image_info *infop)
1431 {
1432     regs->usp = infop->start_stack;
1433     regs->sr = 0;
1434     regs->pc = infop->entry;
1435 }
1436 
1437 /* See linux kernel: arch/m68k/include/asm/elf.h.  */
1438 #define ELF_NREG 20
1439 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1440 
1441 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUM68KState *env)
1442 {
1443     (*regs)[0] = tswapreg(env->dregs[1]);
1444     (*regs)[1] = tswapreg(env->dregs[2]);
1445     (*regs)[2] = tswapreg(env->dregs[3]);
1446     (*regs)[3] = tswapreg(env->dregs[4]);
1447     (*regs)[4] = tswapreg(env->dregs[5]);
1448     (*regs)[5] = tswapreg(env->dregs[6]);
1449     (*regs)[6] = tswapreg(env->dregs[7]);
1450     (*regs)[7] = tswapreg(env->aregs[0]);
1451     (*regs)[8] = tswapreg(env->aregs[1]);
1452     (*regs)[9] = tswapreg(env->aregs[2]);
1453     (*regs)[10] = tswapreg(env->aregs[3]);
1454     (*regs)[11] = tswapreg(env->aregs[4]);
1455     (*regs)[12] = tswapreg(env->aregs[5]);
1456     (*regs)[13] = tswapreg(env->aregs[6]);
1457     (*regs)[14] = tswapreg(env->dregs[0]);
1458     (*regs)[15] = tswapreg(env->aregs[7]);
1459     (*regs)[16] = tswapreg(env->dregs[0]); /* FIXME: orig_d0 */
1460     (*regs)[17] = tswapreg(env->sr);
1461     (*regs)[18] = tswapreg(env->pc);
1462     (*regs)[19] = 0;  /* FIXME: regs->format | regs->vector */
1463 }
1464 
1465 #define USE_ELF_CORE_DUMP
1466 #define ELF_EXEC_PAGESIZE       8192
1467 
1468 #endif
1469 
1470 #ifdef TARGET_ALPHA
1471 
1472 #define ELF_START_MMAP (0x30000000000ULL)
1473 
1474 #define ELF_CLASS      ELFCLASS64
1475 #define ELF_ARCH       EM_ALPHA
1476 
1477 static inline void init_thread(struct target_pt_regs *regs,
1478                                struct image_info *infop)
1479 {
1480     regs->pc = infop->entry;
1481     regs->ps = 8;
1482     regs->usp = infop->start_stack;
1483 }
1484 
1485 #define ELF_EXEC_PAGESIZE        8192
1486 
1487 #endif /* TARGET_ALPHA */
1488 
1489 #ifdef TARGET_S390X
1490 
1491 #define ELF_START_MMAP (0x20000000000ULL)
1492 
1493 #define ELF_CLASS	ELFCLASS64
1494 #define ELF_DATA	ELFDATA2MSB
1495 #define ELF_ARCH	EM_S390
1496 
1497 #include "elf.h"
1498 
1499 #define ELF_HWCAP get_elf_hwcap()
1500 
1501 #define GET_FEATURE(_feat, _hwcap) \
1502     do { if (s390_has_feat(_feat)) { hwcap |= _hwcap; } } while (0)
1503 
1504 static uint32_t get_elf_hwcap(void)
1505 {
1506     /*
1507      * Let's assume we always have esan3 and zarch.
1508      * 31-bit processes can use 64-bit registers (high gprs).
1509      */
1510     uint32_t hwcap = HWCAP_S390_ESAN3 | HWCAP_S390_ZARCH | HWCAP_S390_HIGH_GPRS;
1511 
1512     GET_FEATURE(S390_FEAT_STFLE, HWCAP_S390_STFLE);
1513     GET_FEATURE(S390_FEAT_MSA, HWCAP_S390_MSA);
1514     GET_FEATURE(S390_FEAT_LONG_DISPLACEMENT, HWCAP_S390_LDISP);
1515     GET_FEATURE(S390_FEAT_EXTENDED_IMMEDIATE, HWCAP_S390_EIMM);
1516     if (s390_has_feat(S390_FEAT_EXTENDED_TRANSLATION_3) &&
1517         s390_has_feat(S390_FEAT_ETF3_ENH)) {
1518         hwcap |= HWCAP_S390_ETF3EH;
1519     }
1520     GET_FEATURE(S390_FEAT_VECTOR, HWCAP_S390_VXRS);
1521     GET_FEATURE(S390_FEAT_VECTOR_ENH, HWCAP_S390_VXRS_EXT);
1522 
1523     return hwcap;
1524 }
1525 
1526 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop)
1527 {
1528     regs->psw.addr = infop->entry;
1529     regs->psw.mask = PSW_MASK_64 | PSW_MASK_32;
1530     regs->gprs[15] = infop->start_stack;
1531 }
1532 
1533 /* See linux kernel: arch/s390/include/uapi/asm/ptrace.h (s390_regs).  */
1534 #define ELF_NREG 27
1535 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1536 
1537 enum {
1538     TARGET_REG_PSWM = 0,
1539     TARGET_REG_PSWA = 1,
1540     TARGET_REG_GPRS = 2,
1541     TARGET_REG_ARS = 18,
1542     TARGET_REG_ORIG_R2 = 26,
1543 };
1544 
1545 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1546                                const CPUS390XState *env)
1547 {
1548     int i;
1549     uint32_t *aregs;
1550 
1551     (*regs)[TARGET_REG_PSWM] = tswapreg(env->psw.mask);
1552     (*regs)[TARGET_REG_PSWA] = tswapreg(env->psw.addr);
1553     for (i = 0; i < 16; i++) {
1554         (*regs)[TARGET_REG_GPRS + i] = tswapreg(env->regs[i]);
1555     }
1556     aregs = (uint32_t *)&((*regs)[TARGET_REG_ARS]);
1557     for (i = 0; i < 16; i++) {
1558         aregs[i] = tswap32(env->aregs[i]);
1559     }
1560     (*regs)[TARGET_REG_ORIG_R2] = 0;
1561 }
1562 
1563 #define USE_ELF_CORE_DUMP
1564 #define ELF_EXEC_PAGESIZE 4096
1565 
1566 #endif /* TARGET_S390X */
1567 
1568 #ifdef TARGET_RISCV
1569 
1570 #define ELF_START_MMAP 0x80000000
1571 #define ELF_ARCH  EM_RISCV
1572 
1573 #ifdef TARGET_RISCV32
1574 #define ELF_CLASS ELFCLASS32
1575 #else
1576 #define ELF_CLASS ELFCLASS64
1577 #endif
1578 
1579 #define ELF_HWCAP get_elf_hwcap()
1580 
1581 static uint32_t get_elf_hwcap(void)
1582 {
1583 #define MISA_BIT(EXT) (1 << (EXT - 'A'))
1584     RISCVCPU *cpu = RISCV_CPU(thread_cpu);
1585     uint32_t mask = MISA_BIT('I') | MISA_BIT('M') | MISA_BIT('A')
1586                     | MISA_BIT('F') | MISA_BIT('D') | MISA_BIT('C');
1587 
1588     return cpu->env.misa_ext & mask;
1589 #undef MISA_BIT
1590 }
1591 
1592 static inline void init_thread(struct target_pt_regs *regs,
1593                                struct image_info *infop)
1594 {
1595     regs->sepc = infop->entry;
1596     regs->sp = infop->start_stack;
1597 }
1598 
1599 #define ELF_EXEC_PAGESIZE 4096
1600 
1601 #endif /* TARGET_RISCV */
1602 
1603 #ifdef TARGET_HPPA
1604 
1605 #define ELF_START_MMAP  0x80000000
1606 #define ELF_CLASS       ELFCLASS32
1607 #define ELF_ARCH        EM_PARISC
1608 #define ELF_PLATFORM    "PARISC"
1609 #define STACK_GROWS_DOWN 0
1610 #define STACK_ALIGNMENT  64
1611 
1612 static inline void init_thread(struct target_pt_regs *regs,
1613                                struct image_info *infop)
1614 {
1615     regs->iaoq[0] = infop->entry;
1616     regs->iaoq[1] = infop->entry + 4;
1617     regs->gr[23] = 0;
1618     regs->gr[24] = infop->argv;
1619     regs->gr[25] = infop->argc;
1620     /* The top-of-stack contains a linkage buffer.  */
1621     regs->gr[30] = infop->start_stack + 64;
1622     regs->gr[31] = infop->entry;
1623 }
1624 
1625 #endif /* TARGET_HPPA */
1626 
1627 #ifdef TARGET_XTENSA
1628 
1629 #define ELF_START_MMAP 0x20000000
1630 
1631 #define ELF_CLASS       ELFCLASS32
1632 #define ELF_ARCH        EM_XTENSA
1633 
1634 static inline void init_thread(struct target_pt_regs *regs,
1635                                struct image_info *infop)
1636 {
1637     regs->windowbase = 0;
1638     regs->windowstart = 1;
1639     regs->areg[1] = infop->start_stack;
1640     regs->pc = infop->entry;
1641 }
1642 
1643 /* See linux kernel: arch/xtensa/include/asm/elf.h.  */
1644 #define ELF_NREG 128
1645 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1646 
1647 enum {
1648     TARGET_REG_PC,
1649     TARGET_REG_PS,
1650     TARGET_REG_LBEG,
1651     TARGET_REG_LEND,
1652     TARGET_REG_LCOUNT,
1653     TARGET_REG_SAR,
1654     TARGET_REG_WINDOWSTART,
1655     TARGET_REG_WINDOWBASE,
1656     TARGET_REG_THREADPTR,
1657     TARGET_REG_AR0 = 64,
1658 };
1659 
1660 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1661                                const CPUXtensaState *env)
1662 {
1663     unsigned i;
1664 
1665     (*regs)[TARGET_REG_PC] = tswapreg(env->pc);
1666     (*regs)[TARGET_REG_PS] = tswapreg(env->sregs[PS] & ~PS_EXCM);
1667     (*regs)[TARGET_REG_LBEG] = tswapreg(env->sregs[LBEG]);
1668     (*regs)[TARGET_REG_LEND] = tswapreg(env->sregs[LEND]);
1669     (*regs)[TARGET_REG_LCOUNT] = tswapreg(env->sregs[LCOUNT]);
1670     (*regs)[TARGET_REG_SAR] = tswapreg(env->sregs[SAR]);
1671     (*regs)[TARGET_REG_WINDOWSTART] = tswapreg(env->sregs[WINDOW_START]);
1672     (*regs)[TARGET_REG_WINDOWBASE] = tswapreg(env->sregs[WINDOW_BASE]);
1673     (*regs)[TARGET_REG_THREADPTR] = tswapreg(env->uregs[THREADPTR]);
1674     xtensa_sync_phys_from_window((CPUXtensaState *)env);
1675     for (i = 0; i < env->config->nareg; ++i) {
1676         (*regs)[TARGET_REG_AR0 + i] = tswapreg(env->phys_regs[i]);
1677     }
1678 }
1679 
1680 #define USE_ELF_CORE_DUMP
1681 #define ELF_EXEC_PAGESIZE       4096
1682 
1683 #endif /* TARGET_XTENSA */
1684 
1685 #ifdef TARGET_HEXAGON
1686 
1687 #define ELF_START_MMAP 0x20000000
1688 
1689 #define ELF_CLASS       ELFCLASS32
1690 #define ELF_ARCH        EM_HEXAGON
1691 
1692 static inline void init_thread(struct target_pt_regs *regs,
1693                                struct image_info *infop)
1694 {
1695     regs->sepc = infop->entry;
1696     regs->sp = infop->start_stack;
1697 }
1698 
1699 #endif /* TARGET_HEXAGON */
1700 
1701 #ifndef ELF_PLATFORM
1702 #define ELF_PLATFORM (NULL)
1703 #endif
1704 
1705 #ifndef ELF_MACHINE
1706 #define ELF_MACHINE ELF_ARCH
1707 #endif
1708 
1709 #ifndef elf_check_arch
1710 #define elf_check_arch(x) ((x) == ELF_ARCH)
1711 #endif
1712 
1713 #ifndef elf_check_abi
1714 #define elf_check_abi(x) (1)
1715 #endif
1716 
1717 #ifndef ELF_HWCAP
1718 #define ELF_HWCAP 0
1719 #endif
1720 
1721 #ifndef STACK_GROWS_DOWN
1722 #define STACK_GROWS_DOWN 1
1723 #endif
1724 
1725 #ifndef STACK_ALIGNMENT
1726 #define STACK_ALIGNMENT 16
1727 #endif
1728 
1729 #ifdef TARGET_ABI32
1730 #undef ELF_CLASS
1731 #define ELF_CLASS ELFCLASS32
1732 #undef bswaptls
1733 #define bswaptls(ptr) bswap32s(ptr)
1734 #endif
1735 
1736 #include "elf.h"
1737 
1738 /* We must delay the following stanzas until after "elf.h". */
1739 #if defined(TARGET_AARCH64)
1740 
1741 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz,
1742                                     const uint32_t *data,
1743                                     struct image_info *info,
1744                                     Error **errp)
1745 {
1746     if (pr_type == GNU_PROPERTY_AARCH64_FEATURE_1_AND) {
1747         if (pr_datasz != sizeof(uint32_t)) {
1748             error_setg(errp, "Ill-formed GNU_PROPERTY_AARCH64_FEATURE_1_AND");
1749             return false;
1750         }
1751         /* We will extract GNU_PROPERTY_AARCH64_FEATURE_1_BTI later. */
1752         info->note_flags = *data;
1753     }
1754     return true;
1755 }
1756 #define ARCH_USE_GNU_PROPERTY 1
1757 
1758 #else
1759 
1760 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz,
1761                                     const uint32_t *data,
1762                                     struct image_info *info,
1763                                     Error **errp)
1764 {
1765     g_assert_not_reached();
1766 }
1767 #define ARCH_USE_GNU_PROPERTY 0
1768 
1769 #endif
1770 
1771 struct exec
1772 {
1773     unsigned int a_info;   /* Use macros N_MAGIC, etc for access */
1774     unsigned int a_text;   /* length of text, in bytes */
1775     unsigned int a_data;   /* length of data, in bytes */
1776     unsigned int a_bss;    /* length of uninitialized data area, in bytes */
1777     unsigned int a_syms;   /* length of symbol table data in file, in bytes */
1778     unsigned int a_entry;  /* start address */
1779     unsigned int a_trsize; /* length of relocation info for text, in bytes */
1780     unsigned int a_drsize; /* length of relocation info for data, in bytes */
1781 };
1782 
1783 
1784 #define N_MAGIC(exec) ((exec).a_info & 0xffff)
1785 #define OMAGIC 0407
1786 #define NMAGIC 0410
1787 #define ZMAGIC 0413
1788 #define QMAGIC 0314
1789 
1790 /* Necessary parameters */
1791 #define TARGET_ELF_EXEC_PAGESIZE \
1792         (((eppnt->p_align & ~qemu_host_page_mask) != 0) ? \
1793          TARGET_PAGE_SIZE : MAX(qemu_host_page_size, TARGET_PAGE_SIZE))
1794 #define TARGET_ELF_PAGELENGTH(_v) ROUND_UP((_v), TARGET_ELF_EXEC_PAGESIZE)
1795 #define TARGET_ELF_PAGESTART(_v) ((_v) & \
1796                                  ~(abi_ulong)(TARGET_ELF_EXEC_PAGESIZE-1))
1797 #define TARGET_ELF_PAGEOFFSET(_v) ((_v) & (TARGET_ELF_EXEC_PAGESIZE-1))
1798 
1799 #define DLINFO_ITEMS 16
1800 
1801 static inline void memcpy_fromfs(void * to, const void * from, unsigned long n)
1802 {
1803     memcpy(to, from, n);
1804 }
1805 
1806 #ifdef BSWAP_NEEDED
1807 static void bswap_ehdr(struct elfhdr *ehdr)
1808 {
1809     bswap16s(&ehdr->e_type);            /* Object file type */
1810     bswap16s(&ehdr->e_machine);         /* Architecture */
1811     bswap32s(&ehdr->e_version);         /* Object file version */
1812     bswaptls(&ehdr->e_entry);           /* Entry point virtual address */
1813     bswaptls(&ehdr->e_phoff);           /* Program header table file offset */
1814     bswaptls(&ehdr->e_shoff);           /* Section header table file offset */
1815     bswap32s(&ehdr->e_flags);           /* Processor-specific flags */
1816     bswap16s(&ehdr->e_ehsize);          /* ELF header size in bytes */
1817     bswap16s(&ehdr->e_phentsize);       /* Program header table entry size */
1818     bswap16s(&ehdr->e_phnum);           /* Program header table entry count */
1819     bswap16s(&ehdr->e_shentsize);       /* Section header table entry size */
1820     bswap16s(&ehdr->e_shnum);           /* Section header table entry count */
1821     bswap16s(&ehdr->e_shstrndx);        /* Section header string table index */
1822 }
1823 
1824 static void bswap_phdr(struct elf_phdr *phdr, int phnum)
1825 {
1826     int i;
1827     for (i = 0; i < phnum; ++i, ++phdr) {
1828         bswap32s(&phdr->p_type);        /* Segment type */
1829         bswap32s(&phdr->p_flags);       /* Segment flags */
1830         bswaptls(&phdr->p_offset);      /* Segment file offset */
1831         bswaptls(&phdr->p_vaddr);       /* Segment virtual address */
1832         bswaptls(&phdr->p_paddr);       /* Segment physical address */
1833         bswaptls(&phdr->p_filesz);      /* Segment size in file */
1834         bswaptls(&phdr->p_memsz);       /* Segment size in memory */
1835         bswaptls(&phdr->p_align);       /* Segment alignment */
1836     }
1837 }
1838 
1839 static void bswap_shdr(struct elf_shdr *shdr, int shnum)
1840 {
1841     int i;
1842     for (i = 0; i < shnum; ++i, ++shdr) {
1843         bswap32s(&shdr->sh_name);
1844         bswap32s(&shdr->sh_type);
1845         bswaptls(&shdr->sh_flags);
1846         bswaptls(&shdr->sh_addr);
1847         bswaptls(&shdr->sh_offset);
1848         bswaptls(&shdr->sh_size);
1849         bswap32s(&shdr->sh_link);
1850         bswap32s(&shdr->sh_info);
1851         bswaptls(&shdr->sh_addralign);
1852         bswaptls(&shdr->sh_entsize);
1853     }
1854 }
1855 
1856 static void bswap_sym(struct elf_sym *sym)
1857 {
1858     bswap32s(&sym->st_name);
1859     bswaptls(&sym->st_value);
1860     bswaptls(&sym->st_size);
1861     bswap16s(&sym->st_shndx);
1862 }
1863 
1864 #ifdef TARGET_MIPS
1865 static void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags)
1866 {
1867     bswap16s(&abiflags->version);
1868     bswap32s(&abiflags->ases);
1869     bswap32s(&abiflags->isa_ext);
1870     bswap32s(&abiflags->flags1);
1871     bswap32s(&abiflags->flags2);
1872 }
1873 #endif
1874 #else
1875 static inline void bswap_ehdr(struct elfhdr *ehdr) { }
1876 static inline void bswap_phdr(struct elf_phdr *phdr, int phnum) { }
1877 static inline void bswap_shdr(struct elf_shdr *shdr, int shnum) { }
1878 static inline void bswap_sym(struct elf_sym *sym) { }
1879 #ifdef TARGET_MIPS
1880 static inline void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags) { }
1881 #endif
1882 #endif
1883 
1884 #ifdef USE_ELF_CORE_DUMP
1885 static int elf_core_dump(int, const CPUArchState *);
1886 #endif /* USE_ELF_CORE_DUMP */
1887 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias);
1888 
1889 /* Verify the portions of EHDR within E_IDENT for the target.
1890    This can be performed before bswapping the entire header.  */
1891 static bool elf_check_ident(struct elfhdr *ehdr)
1892 {
1893     return (ehdr->e_ident[EI_MAG0] == ELFMAG0
1894             && ehdr->e_ident[EI_MAG1] == ELFMAG1
1895             && ehdr->e_ident[EI_MAG2] == ELFMAG2
1896             && ehdr->e_ident[EI_MAG3] == ELFMAG3
1897             && ehdr->e_ident[EI_CLASS] == ELF_CLASS
1898             && ehdr->e_ident[EI_DATA] == ELF_DATA
1899             && ehdr->e_ident[EI_VERSION] == EV_CURRENT);
1900 }
1901 
1902 /* Verify the portions of EHDR outside of E_IDENT for the target.
1903    This has to wait until after bswapping the header.  */
1904 static bool elf_check_ehdr(struct elfhdr *ehdr)
1905 {
1906     return (elf_check_arch(ehdr->e_machine)
1907             && elf_check_abi(ehdr->e_flags)
1908             && ehdr->e_ehsize == sizeof(struct elfhdr)
1909             && ehdr->e_phentsize == sizeof(struct elf_phdr)
1910             && (ehdr->e_type == ET_EXEC || ehdr->e_type == ET_DYN));
1911 }
1912 
1913 /*
1914  * 'copy_elf_strings()' copies argument/envelope strings from user
1915  * memory to free pages in kernel mem. These are in a format ready
1916  * to be put directly into the top of new user memory.
1917  *
1918  */
1919 static abi_ulong copy_elf_strings(int argc, char **argv, char *scratch,
1920                                   abi_ulong p, abi_ulong stack_limit)
1921 {
1922     char *tmp;
1923     int len, i;
1924     abi_ulong top = p;
1925 
1926     if (!p) {
1927         return 0;       /* bullet-proofing */
1928     }
1929 
1930     if (STACK_GROWS_DOWN) {
1931         int offset = ((p - 1) % TARGET_PAGE_SIZE) + 1;
1932         for (i = argc - 1; i >= 0; --i) {
1933             tmp = argv[i];
1934             if (!tmp) {
1935                 fprintf(stderr, "VFS: argc is wrong");
1936                 exit(-1);
1937             }
1938             len = strlen(tmp) + 1;
1939             tmp += len;
1940 
1941             if (len > (p - stack_limit)) {
1942                 return 0;
1943             }
1944             while (len) {
1945                 int bytes_to_copy = (len > offset) ? offset : len;
1946                 tmp -= bytes_to_copy;
1947                 p -= bytes_to_copy;
1948                 offset -= bytes_to_copy;
1949                 len -= bytes_to_copy;
1950 
1951                 memcpy_fromfs(scratch + offset, tmp, bytes_to_copy);
1952 
1953                 if (offset == 0) {
1954                     memcpy_to_target(p, scratch, top - p);
1955                     top = p;
1956                     offset = TARGET_PAGE_SIZE;
1957                 }
1958             }
1959         }
1960         if (p != top) {
1961             memcpy_to_target(p, scratch + offset, top - p);
1962         }
1963     } else {
1964         int remaining = TARGET_PAGE_SIZE - (p % TARGET_PAGE_SIZE);
1965         for (i = 0; i < argc; ++i) {
1966             tmp = argv[i];
1967             if (!tmp) {
1968                 fprintf(stderr, "VFS: argc is wrong");
1969                 exit(-1);
1970             }
1971             len = strlen(tmp) + 1;
1972             if (len > (stack_limit - p)) {
1973                 return 0;
1974             }
1975             while (len) {
1976                 int bytes_to_copy = (len > remaining) ? remaining : len;
1977 
1978                 memcpy_fromfs(scratch + (p - top), tmp, bytes_to_copy);
1979 
1980                 tmp += bytes_to_copy;
1981                 remaining -= bytes_to_copy;
1982                 p += bytes_to_copy;
1983                 len -= bytes_to_copy;
1984 
1985                 if (remaining == 0) {
1986                     memcpy_to_target(top, scratch, p - top);
1987                     top = p;
1988                     remaining = TARGET_PAGE_SIZE;
1989                 }
1990             }
1991         }
1992         if (p != top) {
1993             memcpy_to_target(top, scratch, p - top);
1994         }
1995     }
1996 
1997     return p;
1998 }
1999 
2000 /* Older linux kernels provide up to MAX_ARG_PAGES (default: 32) of
2001  * argument/environment space. Newer kernels (>2.6.33) allow more,
2002  * dependent on stack size, but guarantee at least 32 pages for
2003  * backwards compatibility.
2004  */
2005 #define STACK_LOWER_LIMIT (32 * TARGET_PAGE_SIZE)
2006 
2007 static abi_ulong setup_arg_pages(struct linux_binprm *bprm,
2008                                  struct image_info *info)
2009 {
2010     abi_ulong size, error, guard;
2011 
2012     size = guest_stack_size;
2013     if (size < STACK_LOWER_LIMIT) {
2014         size = STACK_LOWER_LIMIT;
2015     }
2016     guard = TARGET_PAGE_SIZE;
2017     if (guard < qemu_real_host_page_size()) {
2018         guard = qemu_real_host_page_size();
2019     }
2020 
2021     error = target_mmap(0, size + guard, PROT_READ | PROT_WRITE,
2022                         MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
2023     if (error == -1) {
2024         perror("mmap stack");
2025         exit(-1);
2026     }
2027 
2028     /* We reserve one extra page at the top of the stack as guard.  */
2029     if (STACK_GROWS_DOWN) {
2030         target_mprotect(error, guard, PROT_NONE);
2031         info->stack_limit = error + guard;
2032         return info->stack_limit + size - sizeof(void *);
2033     } else {
2034         target_mprotect(error + size, guard, PROT_NONE);
2035         info->stack_limit = error + size;
2036         return error;
2037     }
2038 }
2039 
2040 /* Map and zero the bss.  We need to explicitly zero any fractional pages
2041    after the data section (i.e. bss).  */
2042 static void zero_bss(abi_ulong elf_bss, abi_ulong last_bss, int prot)
2043 {
2044     uintptr_t host_start, host_map_start, host_end;
2045 
2046     last_bss = TARGET_PAGE_ALIGN(last_bss);
2047 
2048     /* ??? There is confusion between qemu_real_host_page_size and
2049        qemu_host_page_size here and elsewhere in target_mmap, which
2050        may lead to the end of the data section mapping from the file
2051        not being mapped.  At least there was an explicit test and
2052        comment for that here, suggesting that "the file size must
2053        be known".  The comment probably pre-dates the introduction
2054        of the fstat system call in target_mmap which does in fact
2055        find out the size.  What isn't clear is if the workaround
2056        here is still actually needed.  For now, continue with it,
2057        but merge it with the "normal" mmap that would allocate the bss.  */
2058 
2059     host_start = (uintptr_t) g2h_untagged(elf_bss);
2060     host_end = (uintptr_t) g2h_untagged(last_bss);
2061     host_map_start = REAL_HOST_PAGE_ALIGN(host_start);
2062 
2063     if (host_map_start < host_end) {
2064         void *p = mmap((void *)host_map_start, host_end - host_map_start,
2065                        prot, MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
2066         if (p == MAP_FAILED) {
2067             perror("cannot mmap brk");
2068             exit(-1);
2069         }
2070     }
2071 
2072     /* Ensure that the bss page(s) are valid */
2073     if ((page_get_flags(last_bss-1) & prot) != prot) {
2074         page_set_flags(elf_bss & TARGET_PAGE_MASK, last_bss, prot | PAGE_VALID);
2075     }
2076 
2077     if (host_start < host_map_start) {
2078         memset((void *)host_start, 0, host_map_start - host_start);
2079     }
2080 }
2081 
2082 #ifdef TARGET_ARM
2083 static int elf_is_fdpic(struct elfhdr *exec)
2084 {
2085     return exec->e_ident[EI_OSABI] == ELFOSABI_ARM_FDPIC;
2086 }
2087 #else
2088 /* Default implementation, always false.  */
2089 static int elf_is_fdpic(struct elfhdr *exec)
2090 {
2091     return 0;
2092 }
2093 #endif
2094 
2095 static abi_ulong loader_build_fdpic_loadmap(struct image_info *info, abi_ulong sp)
2096 {
2097     uint16_t n;
2098     struct elf32_fdpic_loadseg *loadsegs = info->loadsegs;
2099 
2100     /* elf32_fdpic_loadseg */
2101     n = info->nsegs;
2102     while (n--) {
2103         sp -= 12;
2104         put_user_u32(loadsegs[n].addr, sp+0);
2105         put_user_u32(loadsegs[n].p_vaddr, sp+4);
2106         put_user_u32(loadsegs[n].p_memsz, sp+8);
2107     }
2108 
2109     /* elf32_fdpic_loadmap */
2110     sp -= 4;
2111     put_user_u16(0, sp+0); /* version */
2112     put_user_u16(info->nsegs, sp+2); /* nsegs */
2113 
2114     info->personality = PER_LINUX_FDPIC;
2115     info->loadmap_addr = sp;
2116 
2117     return sp;
2118 }
2119 
2120 static abi_ulong create_elf_tables(abi_ulong p, int argc, int envc,
2121                                    struct elfhdr *exec,
2122                                    struct image_info *info,
2123                                    struct image_info *interp_info)
2124 {
2125     abi_ulong sp;
2126     abi_ulong u_argc, u_argv, u_envp, u_auxv;
2127     int size;
2128     int i;
2129     abi_ulong u_rand_bytes;
2130     uint8_t k_rand_bytes[16];
2131     abi_ulong u_platform;
2132     const char *k_platform;
2133     const int n = sizeof(elf_addr_t);
2134 
2135     sp = p;
2136 
2137     /* Needs to be before we load the env/argc/... */
2138     if (elf_is_fdpic(exec)) {
2139         /* Need 4 byte alignment for these structs */
2140         sp &= ~3;
2141         sp = loader_build_fdpic_loadmap(info, sp);
2142         info->other_info = interp_info;
2143         if (interp_info) {
2144             interp_info->other_info = info;
2145             sp = loader_build_fdpic_loadmap(interp_info, sp);
2146             info->interpreter_loadmap_addr = interp_info->loadmap_addr;
2147             info->interpreter_pt_dynamic_addr = interp_info->pt_dynamic_addr;
2148         } else {
2149             info->interpreter_loadmap_addr = 0;
2150             info->interpreter_pt_dynamic_addr = 0;
2151         }
2152     }
2153 
2154     u_platform = 0;
2155     k_platform = ELF_PLATFORM;
2156     if (k_platform) {
2157         size_t len = strlen(k_platform) + 1;
2158         if (STACK_GROWS_DOWN) {
2159             sp -= (len + n - 1) & ~(n - 1);
2160             u_platform = sp;
2161             /* FIXME - check return value of memcpy_to_target() for failure */
2162             memcpy_to_target(sp, k_platform, len);
2163         } else {
2164             memcpy_to_target(sp, k_platform, len);
2165             u_platform = sp;
2166             sp += len + 1;
2167         }
2168     }
2169 
2170     /* Provide 16 byte alignment for the PRNG, and basic alignment for
2171      * the argv and envp pointers.
2172      */
2173     if (STACK_GROWS_DOWN) {
2174         sp = QEMU_ALIGN_DOWN(sp, 16);
2175     } else {
2176         sp = QEMU_ALIGN_UP(sp, 16);
2177     }
2178 
2179     /*
2180      * Generate 16 random bytes for userspace PRNG seeding.
2181      */
2182     qemu_guest_getrandom_nofail(k_rand_bytes, sizeof(k_rand_bytes));
2183     if (STACK_GROWS_DOWN) {
2184         sp -= 16;
2185         u_rand_bytes = sp;
2186         /* FIXME - check return value of memcpy_to_target() for failure */
2187         memcpy_to_target(sp, k_rand_bytes, 16);
2188     } else {
2189         memcpy_to_target(sp, k_rand_bytes, 16);
2190         u_rand_bytes = sp;
2191         sp += 16;
2192     }
2193 
2194     size = (DLINFO_ITEMS + 1) * 2;
2195     if (k_platform)
2196         size += 2;
2197 #ifdef DLINFO_ARCH_ITEMS
2198     size += DLINFO_ARCH_ITEMS * 2;
2199 #endif
2200 #ifdef ELF_HWCAP2
2201     size += 2;
2202 #endif
2203     info->auxv_len = size * n;
2204 
2205     size += envc + argc + 2;
2206     size += 1;  /* argc itself */
2207     size *= n;
2208 
2209     /* Allocate space and finalize stack alignment for entry now.  */
2210     if (STACK_GROWS_DOWN) {
2211         u_argc = QEMU_ALIGN_DOWN(sp - size, STACK_ALIGNMENT);
2212         sp = u_argc;
2213     } else {
2214         u_argc = sp;
2215         sp = QEMU_ALIGN_UP(sp + size, STACK_ALIGNMENT);
2216     }
2217 
2218     u_argv = u_argc + n;
2219     u_envp = u_argv + (argc + 1) * n;
2220     u_auxv = u_envp + (envc + 1) * n;
2221     info->saved_auxv = u_auxv;
2222     info->argc = argc;
2223     info->envc = envc;
2224     info->argv = u_argv;
2225     info->envp = u_envp;
2226 
2227     /* This is correct because Linux defines
2228      * elf_addr_t as Elf32_Off / Elf64_Off
2229      */
2230 #define NEW_AUX_ENT(id, val) do {               \
2231         put_user_ual(id, u_auxv);  u_auxv += n; \
2232         put_user_ual(val, u_auxv); u_auxv += n; \
2233     } while(0)
2234 
2235 #ifdef ARCH_DLINFO
2236     /*
2237      * ARCH_DLINFO must come first so platform specific code can enforce
2238      * special alignment requirements on the AUXV if necessary (eg. PPC).
2239      */
2240     ARCH_DLINFO;
2241 #endif
2242     /* There must be exactly DLINFO_ITEMS entries here, or the assert
2243      * on info->auxv_len will trigger.
2244      */
2245     NEW_AUX_ENT(AT_PHDR, (abi_ulong)(info->load_addr + exec->e_phoff));
2246     NEW_AUX_ENT(AT_PHENT, (abi_ulong)(sizeof (struct elf_phdr)));
2247     NEW_AUX_ENT(AT_PHNUM, (abi_ulong)(exec->e_phnum));
2248     if ((info->alignment & ~qemu_host_page_mask) != 0) {
2249         /* Target doesn't support host page size alignment */
2250         NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(TARGET_PAGE_SIZE));
2251     } else {
2252         NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(MAX(TARGET_PAGE_SIZE,
2253                                                qemu_host_page_size)));
2254     }
2255     NEW_AUX_ENT(AT_BASE, (abi_ulong)(interp_info ? interp_info->load_addr : 0));
2256     NEW_AUX_ENT(AT_FLAGS, (abi_ulong)0);
2257     NEW_AUX_ENT(AT_ENTRY, info->entry);
2258     NEW_AUX_ENT(AT_UID, (abi_ulong) getuid());
2259     NEW_AUX_ENT(AT_EUID, (abi_ulong) geteuid());
2260     NEW_AUX_ENT(AT_GID, (abi_ulong) getgid());
2261     NEW_AUX_ENT(AT_EGID, (abi_ulong) getegid());
2262     NEW_AUX_ENT(AT_HWCAP, (abi_ulong) ELF_HWCAP);
2263     NEW_AUX_ENT(AT_CLKTCK, (abi_ulong) sysconf(_SC_CLK_TCK));
2264     NEW_AUX_ENT(AT_RANDOM, (abi_ulong) u_rand_bytes);
2265     NEW_AUX_ENT(AT_SECURE, (abi_ulong) qemu_getauxval(AT_SECURE));
2266     NEW_AUX_ENT(AT_EXECFN, info->file_string);
2267 
2268 #ifdef ELF_HWCAP2
2269     NEW_AUX_ENT(AT_HWCAP2, (abi_ulong) ELF_HWCAP2);
2270 #endif
2271 
2272     if (u_platform) {
2273         NEW_AUX_ENT(AT_PLATFORM, u_platform);
2274     }
2275     NEW_AUX_ENT (AT_NULL, 0);
2276 #undef NEW_AUX_ENT
2277 
2278     /* Check that our initial calculation of the auxv length matches how much
2279      * we actually put into it.
2280      */
2281     assert(info->auxv_len == u_auxv - info->saved_auxv);
2282 
2283     put_user_ual(argc, u_argc);
2284 
2285     p = info->arg_strings;
2286     for (i = 0; i < argc; ++i) {
2287         put_user_ual(p, u_argv);
2288         u_argv += n;
2289         p += target_strlen(p) + 1;
2290     }
2291     put_user_ual(0, u_argv);
2292 
2293     p = info->env_strings;
2294     for (i = 0; i < envc; ++i) {
2295         put_user_ual(p, u_envp);
2296         u_envp += n;
2297         p += target_strlen(p) + 1;
2298     }
2299     put_user_ual(0, u_envp);
2300 
2301     return sp;
2302 }
2303 
2304 #if defined(HI_COMMPAGE)
2305 #define LO_COMMPAGE 0
2306 #elif defined(LO_COMMPAGE)
2307 #define HI_COMMPAGE 0
2308 #else
2309 #define HI_COMMPAGE 0
2310 #define LO_COMMPAGE 0
2311 #define init_guest_commpage() true
2312 #endif
2313 
2314 static void pgb_fail_in_use(const char *image_name)
2315 {
2316     error_report("%s: requires virtual address space that is in use "
2317                  "(omit the -B option or choose a different value)",
2318                  image_name);
2319     exit(EXIT_FAILURE);
2320 }
2321 
2322 static void pgb_have_guest_base(const char *image_name, abi_ulong guest_loaddr,
2323                                 abi_ulong guest_hiaddr, long align)
2324 {
2325     const int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE;
2326     void *addr, *test;
2327 
2328     if (!QEMU_IS_ALIGNED(guest_base, align)) {
2329         fprintf(stderr, "Requested guest base %p does not satisfy "
2330                 "host minimum alignment (0x%lx)\n",
2331                 (void *)guest_base, align);
2332         exit(EXIT_FAILURE);
2333     }
2334 
2335     /* Sanity check the guest binary. */
2336     if (reserved_va) {
2337         if (guest_hiaddr > reserved_va) {
2338             error_report("%s: requires more than reserved virtual "
2339                          "address space (0x%" PRIx64 " > 0x%lx)",
2340                          image_name, (uint64_t)guest_hiaddr, reserved_va);
2341             exit(EXIT_FAILURE);
2342         }
2343     } else {
2344 #if HOST_LONG_BITS < TARGET_ABI_BITS
2345         if ((guest_hiaddr - guest_base) > ~(uintptr_t)0) {
2346             error_report("%s: requires more virtual address space "
2347                          "than the host can provide (0x%" PRIx64 ")",
2348                          image_name, (uint64_t)guest_hiaddr - guest_base);
2349             exit(EXIT_FAILURE);
2350         }
2351 #endif
2352     }
2353 
2354     /*
2355      * Expand the allocation to the entire reserved_va.
2356      * Exclude the mmap_min_addr hole.
2357      */
2358     if (reserved_va) {
2359         guest_loaddr = (guest_base >= mmap_min_addr ? 0
2360                         : mmap_min_addr - guest_base);
2361         guest_hiaddr = reserved_va;
2362     }
2363 
2364     /* Reserve the address space for the binary, or reserved_va. */
2365     test = g2h_untagged(guest_loaddr);
2366     addr = mmap(test, guest_hiaddr - guest_loaddr, PROT_NONE, flags, -1, 0);
2367     if (test != addr) {
2368         pgb_fail_in_use(image_name);
2369     }
2370     qemu_log_mask(CPU_LOG_PAGE,
2371                   "%s: base @ %p for " TARGET_ABI_FMT_ld " bytes\n",
2372                   __func__, addr, guest_hiaddr - guest_loaddr);
2373 }
2374 
2375 /**
2376  * pgd_find_hole_fallback: potential mmap address
2377  * @guest_size: size of available space
2378  * @brk: location of break
2379  * @align: memory alignment
2380  *
2381  * This is a fallback method for finding a hole in the host address
2382  * space if we don't have the benefit of being able to access
2383  * /proc/self/map. It can potentially take a very long time as we can
2384  * only dumbly iterate up the host address space seeing if the
2385  * allocation would work.
2386  */
2387 static uintptr_t pgd_find_hole_fallback(uintptr_t guest_size, uintptr_t brk,
2388                                         long align, uintptr_t offset)
2389 {
2390     uintptr_t base;
2391 
2392     /* Start (aligned) at the bottom and work our way up */
2393     base = ROUND_UP(mmap_min_addr, align);
2394 
2395     while (true) {
2396         uintptr_t align_start, end;
2397         align_start = ROUND_UP(base, align);
2398         end = align_start + guest_size + offset;
2399 
2400         /* if brk is anywhere in the range give ourselves some room to grow. */
2401         if (align_start <= brk && brk < end) {
2402             base = brk + (16 * MiB);
2403             continue;
2404         } else if (align_start + guest_size < align_start) {
2405             /* we have run out of space */
2406             return -1;
2407         } else {
2408             int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE |
2409                 MAP_FIXED_NOREPLACE;
2410             void * mmap_start = mmap((void *) align_start, guest_size,
2411                                      PROT_NONE, flags, -1, 0);
2412             if (mmap_start != MAP_FAILED) {
2413                 munmap(mmap_start, guest_size);
2414                 if (mmap_start == (void *) align_start) {
2415                     qemu_log_mask(CPU_LOG_PAGE,
2416                                   "%s: base @ %p for %" PRIdPTR" bytes\n",
2417                                   __func__, mmap_start + offset, guest_size);
2418                     return (uintptr_t) mmap_start + offset;
2419                 }
2420             }
2421             base += qemu_host_page_size;
2422         }
2423     }
2424 }
2425 
2426 /* Return value for guest_base, or -1 if no hole found. */
2427 static uintptr_t pgb_find_hole(uintptr_t guest_loaddr, uintptr_t guest_size,
2428                                long align, uintptr_t offset)
2429 {
2430     GSList *maps, *iter;
2431     uintptr_t this_start, this_end, next_start, brk;
2432     intptr_t ret = -1;
2433 
2434     assert(QEMU_IS_ALIGNED(guest_loaddr, align));
2435 
2436     maps = read_self_maps();
2437 
2438     /* Read brk after we've read the maps, which will malloc. */
2439     brk = (uintptr_t)sbrk(0);
2440 
2441     if (!maps) {
2442         return pgd_find_hole_fallback(guest_size, brk, align, offset);
2443     }
2444 
2445     /* The first hole is before the first map entry. */
2446     this_start = mmap_min_addr;
2447 
2448     for (iter = maps; iter;
2449          this_start = next_start, iter = g_slist_next(iter)) {
2450         uintptr_t align_start, hole_size;
2451 
2452         this_end = ((MapInfo *)iter->data)->start;
2453         next_start = ((MapInfo *)iter->data)->end;
2454         align_start = ROUND_UP(this_start + offset, align);
2455 
2456         /* Skip holes that are too small. */
2457         if (align_start >= this_end) {
2458             continue;
2459         }
2460         hole_size = this_end - align_start;
2461         if (hole_size < guest_size) {
2462             continue;
2463         }
2464 
2465         /* If this hole contains brk, give ourselves some room to grow. */
2466         if (this_start <= brk && brk < this_end) {
2467             hole_size -= guest_size;
2468             if (sizeof(uintptr_t) == 8 && hole_size >= 1 * GiB) {
2469                 align_start += 1 * GiB;
2470             } else if (hole_size >= 16 * MiB) {
2471                 align_start += 16 * MiB;
2472             } else {
2473                 align_start = (this_end - guest_size) & -align;
2474                 if (align_start < this_start) {
2475                     continue;
2476                 }
2477             }
2478         }
2479 
2480         /* Record the lowest successful match. */
2481         if (ret < 0) {
2482             ret = align_start;
2483         }
2484         /* If this hole contains the identity map, select it. */
2485         if (align_start <= guest_loaddr &&
2486             guest_loaddr + guest_size <= this_end) {
2487             ret = 0;
2488         }
2489         /* If this hole ends above the identity map, stop looking. */
2490         if (this_end >= guest_loaddr) {
2491             break;
2492         }
2493     }
2494     free_self_maps(maps);
2495 
2496     if (ret != -1) {
2497         qemu_log_mask(CPU_LOG_PAGE, "%s: base @ %" PRIxPTR
2498                       " for %" PRIuPTR " bytes\n",
2499                       __func__, ret, guest_size);
2500     }
2501 
2502     return ret;
2503 }
2504 
2505 static void pgb_static(const char *image_name, abi_ulong orig_loaddr,
2506                        abi_ulong orig_hiaddr, long align)
2507 {
2508     uintptr_t loaddr = orig_loaddr;
2509     uintptr_t hiaddr = orig_hiaddr;
2510     uintptr_t offset = 0;
2511     uintptr_t addr;
2512 
2513     if (hiaddr != orig_hiaddr) {
2514         error_report("%s: requires virtual address space that the "
2515                      "host cannot provide (0x%" PRIx64 ")",
2516                      image_name, (uint64_t)orig_hiaddr);
2517         exit(EXIT_FAILURE);
2518     }
2519 
2520     loaddr &= -align;
2521     if (HI_COMMPAGE) {
2522         /*
2523          * Extend the allocation to include the commpage.
2524          * For a 64-bit host, this is just 4GiB; for a 32-bit host we
2525          * need to ensure there is space bellow the guest_base so we
2526          * can map the commpage in the place needed when the address
2527          * arithmetic wraps around.
2528          */
2529         if (sizeof(uintptr_t) == 8 || loaddr >= 0x80000000u) {
2530             hiaddr = (uintptr_t) 4 << 30;
2531         } else {
2532             offset = -(HI_COMMPAGE & -align);
2533         }
2534     } else if (LO_COMMPAGE != 0) {
2535         loaddr = MIN(loaddr, LO_COMMPAGE & -align);
2536     }
2537 
2538     addr = pgb_find_hole(loaddr, hiaddr - loaddr, align, offset);
2539     if (addr == -1) {
2540         /*
2541          * If HI_COMMPAGE, there *might* be a non-consecutive allocation
2542          * that can satisfy both.  But as the normal arm32 link base address
2543          * is ~32k, and we extend down to include the commpage, making the
2544          * overhead only ~96k, this is unlikely.
2545          */
2546         error_report("%s: Unable to allocate %#zx bytes of "
2547                      "virtual address space", image_name,
2548                      (size_t)(hiaddr - loaddr));
2549         exit(EXIT_FAILURE);
2550     }
2551 
2552     guest_base = addr;
2553 
2554     qemu_log_mask(CPU_LOG_PAGE, "%s: base @ %"PRIxPTR" for %" PRIuPTR" bytes\n",
2555                   __func__, addr, hiaddr - loaddr);
2556 }
2557 
2558 static void pgb_dynamic(const char *image_name, long align)
2559 {
2560     /*
2561      * The executable is dynamic and does not require a fixed address.
2562      * All we need is a commpage that satisfies align.
2563      * If we do not need a commpage, leave guest_base == 0.
2564      */
2565     if (HI_COMMPAGE) {
2566         uintptr_t addr, commpage;
2567 
2568         /* 64-bit hosts should have used reserved_va. */
2569         assert(sizeof(uintptr_t) == 4);
2570 
2571         /*
2572          * By putting the commpage at the first hole, that puts guest_base
2573          * just above that, and maximises the positive guest addresses.
2574          */
2575         commpage = HI_COMMPAGE & -align;
2576         addr = pgb_find_hole(commpage, -commpage, align, 0);
2577         assert(addr != -1);
2578         guest_base = addr;
2579     }
2580 }
2581 
2582 static void pgb_reserved_va(const char *image_name, abi_ulong guest_loaddr,
2583                             abi_ulong guest_hiaddr, long align)
2584 {
2585     int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE;
2586     void *addr, *test;
2587 
2588     if (guest_hiaddr > reserved_va) {
2589         error_report("%s: requires more than reserved virtual "
2590                      "address space (0x%" PRIx64 " > 0x%lx)",
2591                      image_name, (uint64_t)guest_hiaddr, reserved_va);
2592         exit(EXIT_FAILURE);
2593     }
2594 
2595     /* Widen the "image" to the entire reserved address space. */
2596     pgb_static(image_name, 0, reserved_va, align);
2597 
2598     /* osdep.h defines this as 0 if it's missing */
2599     flags |= MAP_FIXED_NOREPLACE;
2600 
2601     /* Reserve the memory on the host. */
2602     assert(guest_base != 0);
2603     test = g2h_untagged(0);
2604     addr = mmap(test, reserved_va, PROT_NONE, flags, -1, 0);
2605     if (addr == MAP_FAILED || addr != test) {
2606         error_report("Unable to reserve 0x%lx bytes of virtual address "
2607                      "space at %p (%s) for use as guest address space (check your "
2608                      "virtual memory ulimit setting, min_mmap_addr or reserve less "
2609                      "using -R option)", reserved_va, test, strerror(errno));
2610         exit(EXIT_FAILURE);
2611     }
2612 
2613     qemu_log_mask(CPU_LOG_PAGE, "%s: base @ %p for %lu bytes\n",
2614                   __func__, addr, reserved_va);
2615 }
2616 
2617 void probe_guest_base(const char *image_name, abi_ulong guest_loaddr,
2618                       abi_ulong guest_hiaddr)
2619 {
2620     /* In order to use host shmat, we must be able to honor SHMLBA.  */
2621     uintptr_t align = MAX(SHMLBA, qemu_host_page_size);
2622 
2623     if (have_guest_base) {
2624         pgb_have_guest_base(image_name, guest_loaddr, guest_hiaddr, align);
2625     } else if (reserved_va) {
2626         pgb_reserved_va(image_name, guest_loaddr, guest_hiaddr, align);
2627     } else if (guest_loaddr) {
2628         pgb_static(image_name, guest_loaddr, guest_hiaddr, align);
2629     } else {
2630         pgb_dynamic(image_name, align);
2631     }
2632 
2633     /* Reserve and initialize the commpage. */
2634     if (!init_guest_commpage()) {
2635         /*
2636          * With have_guest_base, the user has selected the address and
2637          * we are trying to work with that.  Otherwise, we have selected
2638          * free space and init_guest_commpage must succeeded.
2639          */
2640         assert(have_guest_base);
2641         pgb_fail_in_use(image_name);
2642     }
2643 
2644     assert(QEMU_IS_ALIGNED(guest_base, align));
2645     qemu_log_mask(CPU_LOG_PAGE, "Locating guest address space "
2646                   "@ 0x%" PRIx64 "\n", (uint64_t)guest_base);
2647 }
2648 
2649 enum {
2650     /* The string "GNU\0" as a magic number. */
2651     GNU0_MAGIC = const_le32('G' | 'N' << 8 | 'U' << 16),
2652     NOTE_DATA_SZ = 1 * KiB,
2653     NOTE_NAME_SZ = 4,
2654     ELF_GNU_PROPERTY_ALIGN = ELF_CLASS == ELFCLASS32 ? 4 : 8,
2655 };
2656 
2657 /*
2658  * Process a single gnu_property entry.
2659  * Return false for error.
2660  */
2661 static bool parse_elf_property(const uint32_t *data, int *off, int datasz,
2662                                struct image_info *info, bool have_prev_type,
2663                                uint32_t *prev_type, Error **errp)
2664 {
2665     uint32_t pr_type, pr_datasz, step;
2666 
2667     if (*off > datasz || !QEMU_IS_ALIGNED(*off, ELF_GNU_PROPERTY_ALIGN)) {
2668         goto error_data;
2669     }
2670     datasz -= *off;
2671     data += *off / sizeof(uint32_t);
2672 
2673     if (datasz < 2 * sizeof(uint32_t)) {
2674         goto error_data;
2675     }
2676     pr_type = data[0];
2677     pr_datasz = data[1];
2678     data += 2;
2679     datasz -= 2 * sizeof(uint32_t);
2680     step = ROUND_UP(pr_datasz, ELF_GNU_PROPERTY_ALIGN);
2681     if (step > datasz) {
2682         goto error_data;
2683     }
2684 
2685     /* Properties are supposed to be unique and sorted on pr_type. */
2686     if (have_prev_type && pr_type <= *prev_type) {
2687         if (pr_type == *prev_type) {
2688             error_setg(errp, "Duplicate property in PT_GNU_PROPERTY");
2689         } else {
2690             error_setg(errp, "Unsorted property in PT_GNU_PROPERTY");
2691         }
2692         return false;
2693     }
2694     *prev_type = pr_type;
2695 
2696     if (!arch_parse_elf_property(pr_type, pr_datasz, data, info, errp)) {
2697         return false;
2698     }
2699 
2700     *off += 2 * sizeof(uint32_t) + step;
2701     return true;
2702 
2703  error_data:
2704     error_setg(errp, "Ill-formed property in PT_GNU_PROPERTY");
2705     return false;
2706 }
2707 
2708 /* Process NT_GNU_PROPERTY_TYPE_0. */
2709 static bool parse_elf_properties(int image_fd,
2710                                  struct image_info *info,
2711                                  const struct elf_phdr *phdr,
2712                                  char bprm_buf[BPRM_BUF_SIZE],
2713                                  Error **errp)
2714 {
2715     union {
2716         struct elf_note nhdr;
2717         uint32_t data[NOTE_DATA_SZ / sizeof(uint32_t)];
2718     } note;
2719 
2720     int n, off, datasz;
2721     bool have_prev_type;
2722     uint32_t prev_type;
2723 
2724     /* Unless the arch requires properties, ignore them. */
2725     if (!ARCH_USE_GNU_PROPERTY) {
2726         return true;
2727     }
2728 
2729     /* If the properties are crazy large, that's too bad. */
2730     n = phdr->p_filesz;
2731     if (n > sizeof(note)) {
2732         error_setg(errp, "PT_GNU_PROPERTY too large");
2733         return false;
2734     }
2735     if (n < sizeof(note.nhdr)) {
2736         error_setg(errp, "PT_GNU_PROPERTY too small");
2737         return false;
2738     }
2739 
2740     if (phdr->p_offset + n <= BPRM_BUF_SIZE) {
2741         memcpy(&note, bprm_buf + phdr->p_offset, n);
2742     } else {
2743         ssize_t len = pread(image_fd, &note, n, phdr->p_offset);
2744         if (len != n) {
2745             error_setg_errno(errp, errno, "Error reading file header");
2746             return false;
2747         }
2748     }
2749 
2750     /*
2751      * The contents of a valid PT_GNU_PROPERTY is a sequence
2752      * of uint32_t -- swap them all now.
2753      */
2754 #ifdef BSWAP_NEEDED
2755     for (int i = 0; i < n / 4; i++) {
2756         bswap32s(note.data + i);
2757     }
2758 #endif
2759 
2760     /*
2761      * Note that nhdr is 3 words, and that the "name" described by namesz
2762      * immediately follows nhdr and is thus at the 4th word.  Further, all
2763      * of the inputs to the kernel's round_up are multiples of 4.
2764      */
2765     if (note.nhdr.n_type != NT_GNU_PROPERTY_TYPE_0 ||
2766         note.nhdr.n_namesz != NOTE_NAME_SZ ||
2767         note.data[3] != GNU0_MAGIC) {
2768         error_setg(errp, "Invalid note in PT_GNU_PROPERTY");
2769         return false;
2770     }
2771     off = sizeof(note.nhdr) + NOTE_NAME_SZ;
2772 
2773     datasz = note.nhdr.n_descsz + off;
2774     if (datasz > n) {
2775         error_setg(errp, "Invalid note size in PT_GNU_PROPERTY");
2776         return false;
2777     }
2778 
2779     have_prev_type = false;
2780     prev_type = 0;
2781     while (1) {
2782         if (off == datasz) {
2783             return true;  /* end, exit ok */
2784         }
2785         if (!parse_elf_property(note.data, &off, datasz, info,
2786                                 have_prev_type, &prev_type, errp)) {
2787             return false;
2788         }
2789         have_prev_type = true;
2790     }
2791 }
2792 
2793 /* Load an ELF image into the address space.
2794 
2795    IMAGE_NAME is the filename of the image, to use in error messages.
2796    IMAGE_FD is the open file descriptor for the image.
2797 
2798    BPRM_BUF is a copy of the beginning of the file; this of course
2799    contains the elf file header at offset 0.  It is assumed that this
2800    buffer is sufficiently aligned to present no problems to the host
2801    in accessing data at aligned offsets within the buffer.
2802 
2803    On return: INFO values will be filled in, as necessary or available.  */
2804 
2805 static void load_elf_image(const char *image_name, int image_fd,
2806                            struct image_info *info, char **pinterp_name,
2807                            char bprm_buf[BPRM_BUF_SIZE])
2808 {
2809     struct elfhdr *ehdr = (struct elfhdr *)bprm_buf;
2810     struct elf_phdr *phdr;
2811     abi_ulong load_addr, load_bias, loaddr, hiaddr, error;
2812     int i, retval, prot_exec;
2813     Error *err = NULL;
2814 
2815     /* First of all, some simple consistency checks */
2816     if (!elf_check_ident(ehdr)) {
2817         error_setg(&err, "Invalid ELF image for this architecture");
2818         goto exit_errmsg;
2819     }
2820     bswap_ehdr(ehdr);
2821     if (!elf_check_ehdr(ehdr)) {
2822         error_setg(&err, "Invalid ELF image for this architecture");
2823         goto exit_errmsg;
2824     }
2825 
2826     i = ehdr->e_phnum * sizeof(struct elf_phdr);
2827     if (ehdr->e_phoff + i <= BPRM_BUF_SIZE) {
2828         phdr = (struct elf_phdr *)(bprm_buf + ehdr->e_phoff);
2829     } else {
2830         phdr = (struct elf_phdr *) alloca(i);
2831         retval = pread(image_fd, phdr, i, ehdr->e_phoff);
2832         if (retval != i) {
2833             goto exit_read;
2834         }
2835     }
2836     bswap_phdr(phdr, ehdr->e_phnum);
2837 
2838     info->nsegs = 0;
2839     info->pt_dynamic_addr = 0;
2840 
2841     mmap_lock();
2842 
2843     /*
2844      * Find the maximum size of the image and allocate an appropriate
2845      * amount of memory to handle that.  Locate the interpreter, if any.
2846      */
2847     loaddr = -1, hiaddr = 0;
2848     info->alignment = 0;
2849     for (i = 0; i < ehdr->e_phnum; ++i) {
2850         struct elf_phdr *eppnt = phdr + i;
2851         if (eppnt->p_type == PT_LOAD) {
2852             abi_ulong a = eppnt->p_vaddr - eppnt->p_offset;
2853             if (a < loaddr) {
2854                 loaddr = a;
2855             }
2856             a = eppnt->p_vaddr + eppnt->p_memsz;
2857             if (a > hiaddr) {
2858                 hiaddr = a;
2859             }
2860             ++info->nsegs;
2861             info->alignment |= eppnt->p_align;
2862         } else if (eppnt->p_type == PT_INTERP && pinterp_name) {
2863             g_autofree char *interp_name = NULL;
2864 
2865             if (*pinterp_name) {
2866                 error_setg(&err, "Multiple PT_INTERP entries");
2867                 goto exit_errmsg;
2868             }
2869 
2870             interp_name = g_malloc(eppnt->p_filesz);
2871 
2872             if (eppnt->p_offset + eppnt->p_filesz <= BPRM_BUF_SIZE) {
2873                 memcpy(interp_name, bprm_buf + eppnt->p_offset,
2874                        eppnt->p_filesz);
2875             } else {
2876                 retval = pread(image_fd, interp_name, eppnt->p_filesz,
2877                                eppnt->p_offset);
2878                 if (retval != eppnt->p_filesz) {
2879                     goto exit_read;
2880                 }
2881             }
2882             if (interp_name[eppnt->p_filesz - 1] != 0) {
2883                 error_setg(&err, "Invalid PT_INTERP entry");
2884                 goto exit_errmsg;
2885             }
2886             *pinterp_name = g_steal_pointer(&interp_name);
2887         } else if (eppnt->p_type == PT_GNU_PROPERTY) {
2888             if (!parse_elf_properties(image_fd, info, eppnt, bprm_buf, &err)) {
2889                 goto exit_errmsg;
2890             }
2891         }
2892     }
2893 
2894     if (pinterp_name != NULL) {
2895         /*
2896          * This is the main executable.
2897          *
2898          * Reserve extra space for brk.
2899          * We hold on to this space while placing the interpreter
2900          * and the stack, lest they be placed immediately after
2901          * the data segment and block allocation from the brk.
2902          *
2903          * 16MB is chosen as "large enough" without being so large as
2904          * to allow the result to not fit with a 32-bit guest on a
2905          * 32-bit host. However some 64 bit guests (e.g. s390x)
2906          * attempt to place their heap further ahead and currently
2907          * nothing stops them smashing into QEMUs address space.
2908          */
2909 #if TARGET_LONG_BITS == 64
2910         info->reserve_brk = 32 * MiB;
2911 #else
2912         info->reserve_brk = 16 * MiB;
2913 #endif
2914         hiaddr += info->reserve_brk;
2915 
2916         if (ehdr->e_type == ET_EXEC) {
2917             /*
2918              * Make sure that the low address does not conflict with
2919              * MMAP_MIN_ADDR or the QEMU application itself.
2920              */
2921             probe_guest_base(image_name, loaddr, hiaddr);
2922         } else {
2923             /*
2924              * The binary is dynamic, but we still need to
2925              * select guest_base.  In this case we pass a size.
2926              */
2927             probe_guest_base(image_name, 0, hiaddr - loaddr);
2928         }
2929     }
2930 
2931     /*
2932      * Reserve address space for all of this.
2933      *
2934      * In the case of ET_EXEC, we supply MAP_FIXED so that we get
2935      * exactly the address range that is required.
2936      *
2937      * Otherwise this is ET_DYN, and we are searching for a location
2938      * that can hold the memory space required.  If the image is
2939      * pre-linked, LOADDR will be non-zero, and the kernel should
2940      * honor that address if it happens to be free.
2941      *
2942      * In both cases, we will overwrite pages in this range with mappings
2943      * from the executable.
2944      */
2945     load_addr = target_mmap(loaddr, hiaddr - loaddr, PROT_NONE,
2946                             MAP_PRIVATE | MAP_ANON | MAP_NORESERVE |
2947                             (ehdr->e_type == ET_EXEC ? MAP_FIXED : 0),
2948                             -1, 0);
2949     if (load_addr == -1) {
2950         goto exit_mmap;
2951     }
2952     load_bias = load_addr - loaddr;
2953 
2954     if (elf_is_fdpic(ehdr)) {
2955         struct elf32_fdpic_loadseg *loadsegs = info->loadsegs =
2956             g_malloc(sizeof(*loadsegs) * info->nsegs);
2957 
2958         for (i = 0; i < ehdr->e_phnum; ++i) {
2959             switch (phdr[i].p_type) {
2960             case PT_DYNAMIC:
2961                 info->pt_dynamic_addr = phdr[i].p_vaddr + load_bias;
2962                 break;
2963             case PT_LOAD:
2964                 loadsegs->addr = phdr[i].p_vaddr + load_bias;
2965                 loadsegs->p_vaddr = phdr[i].p_vaddr;
2966                 loadsegs->p_memsz = phdr[i].p_memsz;
2967                 ++loadsegs;
2968                 break;
2969             }
2970         }
2971     }
2972 
2973     info->load_bias = load_bias;
2974     info->code_offset = load_bias;
2975     info->data_offset = load_bias;
2976     info->load_addr = load_addr;
2977     info->entry = ehdr->e_entry + load_bias;
2978     info->start_code = -1;
2979     info->end_code = 0;
2980     info->start_data = -1;
2981     info->end_data = 0;
2982     info->brk = 0;
2983     info->elf_flags = ehdr->e_flags;
2984 
2985     prot_exec = PROT_EXEC;
2986 #ifdef TARGET_AARCH64
2987     /*
2988      * If the BTI feature is present, this indicates that the executable
2989      * pages of the startup binary should be mapped with PROT_BTI, so that
2990      * branch targets are enforced.
2991      *
2992      * The startup binary is either the interpreter or the static executable.
2993      * The interpreter is responsible for all pages of a dynamic executable.
2994      *
2995      * Elf notes are backward compatible to older cpus.
2996      * Do not enable BTI unless it is supported.
2997      */
2998     if ((info->note_flags & GNU_PROPERTY_AARCH64_FEATURE_1_BTI)
2999         && (pinterp_name == NULL || *pinterp_name == 0)
3000         && cpu_isar_feature(aa64_bti, ARM_CPU(thread_cpu))) {
3001         prot_exec |= TARGET_PROT_BTI;
3002     }
3003 #endif
3004 
3005     for (i = 0; i < ehdr->e_phnum; i++) {
3006         struct elf_phdr *eppnt = phdr + i;
3007         if (eppnt->p_type == PT_LOAD) {
3008             abi_ulong vaddr, vaddr_po, vaddr_ps, vaddr_ef, vaddr_em, vaddr_len;
3009             int elf_prot = 0;
3010 
3011             if (eppnt->p_flags & PF_R) {
3012                 elf_prot |= PROT_READ;
3013             }
3014             if (eppnt->p_flags & PF_W) {
3015                 elf_prot |= PROT_WRITE;
3016             }
3017             if (eppnt->p_flags & PF_X) {
3018                 elf_prot |= prot_exec;
3019             }
3020 
3021             vaddr = load_bias + eppnt->p_vaddr;
3022             vaddr_po = TARGET_ELF_PAGEOFFSET(vaddr);
3023             vaddr_ps = TARGET_ELF_PAGESTART(vaddr);
3024 
3025             vaddr_ef = vaddr + eppnt->p_filesz;
3026             vaddr_em = vaddr + eppnt->p_memsz;
3027 
3028             /*
3029              * Some segments may be completely empty, with a non-zero p_memsz
3030              * but no backing file segment.
3031              */
3032             if (eppnt->p_filesz != 0) {
3033                 vaddr_len = TARGET_ELF_PAGELENGTH(eppnt->p_filesz + vaddr_po);
3034                 error = target_mmap(vaddr_ps, vaddr_len, elf_prot,
3035                                     MAP_PRIVATE | MAP_FIXED,
3036                                     image_fd, eppnt->p_offset - vaddr_po);
3037 
3038                 if (error == -1) {
3039                     goto exit_mmap;
3040                 }
3041 
3042                 /*
3043                  * If the load segment requests extra zeros (e.g. bss), map it.
3044                  */
3045                 if (eppnt->p_filesz < eppnt->p_memsz) {
3046                     zero_bss(vaddr_ef, vaddr_em, elf_prot);
3047                 }
3048             } else if (eppnt->p_memsz != 0) {
3049                 vaddr_len = TARGET_ELF_PAGELENGTH(eppnt->p_memsz + vaddr_po);
3050                 error = target_mmap(vaddr_ps, vaddr_len, elf_prot,
3051                                     MAP_PRIVATE | MAP_FIXED | MAP_ANONYMOUS,
3052                                     -1, 0);
3053 
3054                 if (error == -1) {
3055                     goto exit_mmap;
3056                 }
3057             }
3058 
3059             /* Find the full program boundaries.  */
3060             if (elf_prot & PROT_EXEC) {
3061                 if (vaddr < info->start_code) {
3062                     info->start_code = vaddr;
3063                 }
3064                 if (vaddr_ef > info->end_code) {
3065                     info->end_code = vaddr_ef;
3066                 }
3067             }
3068             if (elf_prot & PROT_WRITE) {
3069                 if (vaddr < info->start_data) {
3070                     info->start_data = vaddr;
3071                 }
3072                 if (vaddr_ef > info->end_data) {
3073                     info->end_data = vaddr_ef;
3074                 }
3075             }
3076             if (vaddr_em > info->brk) {
3077                 info->brk = vaddr_em;
3078             }
3079 #ifdef TARGET_MIPS
3080         } else if (eppnt->p_type == PT_MIPS_ABIFLAGS) {
3081             Mips_elf_abiflags_v0 abiflags;
3082             if (eppnt->p_filesz < sizeof(Mips_elf_abiflags_v0)) {
3083                 error_setg(&err, "Invalid PT_MIPS_ABIFLAGS entry");
3084                 goto exit_errmsg;
3085             }
3086             if (eppnt->p_offset + eppnt->p_filesz <= BPRM_BUF_SIZE) {
3087                 memcpy(&abiflags, bprm_buf + eppnt->p_offset,
3088                        sizeof(Mips_elf_abiflags_v0));
3089             } else {
3090                 retval = pread(image_fd, &abiflags, sizeof(Mips_elf_abiflags_v0),
3091                                eppnt->p_offset);
3092                 if (retval != sizeof(Mips_elf_abiflags_v0)) {
3093                     goto exit_read;
3094                 }
3095             }
3096             bswap_mips_abiflags(&abiflags);
3097             info->fp_abi = abiflags.fp_abi;
3098 #endif
3099         }
3100     }
3101 
3102     if (info->end_data == 0) {
3103         info->start_data = info->end_code;
3104         info->end_data = info->end_code;
3105     }
3106 
3107     if (qemu_log_enabled()) {
3108         load_symbols(ehdr, image_fd, load_bias);
3109     }
3110 
3111     mmap_unlock();
3112 
3113     close(image_fd);
3114     return;
3115 
3116  exit_read:
3117     if (retval >= 0) {
3118         error_setg(&err, "Incomplete read of file header");
3119     } else {
3120         error_setg_errno(&err, errno, "Error reading file header");
3121     }
3122     goto exit_errmsg;
3123  exit_mmap:
3124     error_setg_errno(&err, errno, "Error mapping file");
3125     goto exit_errmsg;
3126  exit_errmsg:
3127     error_reportf_err(err, "%s: ", image_name);
3128     exit(-1);
3129 }
3130 
3131 static void load_elf_interp(const char *filename, struct image_info *info,
3132                             char bprm_buf[BPRM_BUF_SIZE])
3133 {
3134     int fd, retval;
3135     Error *err = NULL;
3136 
3137     fd = open(path(filename), O_RDONLY);
3138     if (fd < 0) {
3139         error_setg_file_open(&err, errno, filename);
3140         error_report_err(err);
3141         exit(-1);
3142     }
3143 
3144     retval = read(fd, bprm_buf, BPRM_BUF_SIZE);
3145     if (retval < 0) {
3146         error_setg_errno(&err, errno, "Error reading file header");
3147         error_reportf_err(err, "%s: ", filename);
3148         exit(-1);
3149     }
3150 
3151     if (retval < BPRM_BUF_SIZE) {
3152         memset(bprm_buf + retval, 0, BPRM_BUF_SIZE - retval);
3153     }
3154 
3155     load_elf_image(filename, fd, info, NULL, bprm_buf);
3156 }
3157 
3158 static int symfind(const void *s0, const void *s1)
3159 {
3160     target_ulong addr = *(target_ulong *)s0;
3161     struct elf_sym *sym = (struct elf_sym *)s1;
3162     int result = 0;
3163     if (addr < sym->st_value) {
3164         result = -1;
3165     } else if (addr >= sym->st_value + sym->st_size) {
3166         result = 1;
3167     }
3168     return result;
3169 }
3170 
3171 static const char *lookup_symbolxx(struct syminfo *s, target_ulong orig_addr)
3172 {
3173 #if ELF_CLASS == ELFCLASS32
3174     struct elf_sym *syms = s->disas_symtab.elf32;
3175 #else
3176     struct elf_sym *syms = s->disas_symtab.elf64;
3177 #endif
3178 
3179     // binary search
3180     struct elf_sym *sym;
3181 
3182     sym = bsearch(&orig_addr, syms, s->disas_num_syms, sizeof(*syms), symfind);
3183     if (sym != NULL) {
3184         return s->disas_strtab + sym->st_name;
3185     }
3186 
3187     return "";
3188 }
3189 
3190 /* FIXME: This should use elf_ops.h  */
3191 static int symcmp(const void *s0, const void *s1)
3192 {
3193     struct elf_sym *sym0 = (struct elf_sym *)s0;
3194     struct elf_sym *sym1 = (struct elf_sym *)s1;
3195     return (sym0->st_value < sym1->st_value)
3196         ? -1
3197         : ((sym0->st_value > sym1->st_value) ? 1 : 0);
3198 }
3199 
3200 /* Best attempt to load symbols from this ELF object. */
3201 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias)
3202 {
3203     int i, shnum, nsyms, sym_idx = 0, str_idx = 0;
3204     uint64_t segsz;
3205     struct elf_shdr *shdr;
3206     char *strings = NULL;
3207     struct syminfo *s = NULL;
3208     struct elf_sym *new_syms, *syms = NULL;
3209 
3210     shnum = hdr->e_shnum;
3211     i = shnum * sizeof(struct elf_shdr);
3212     shdr = (struct elf_shdr *)alloca(i);
3213     if (pread(fd, shdr, i, hdr->e_shoff) != i) {
3214         return;
3215     }
3216 
3217     bswap_shdr(shdr, shnum);
3218     for (i = 0; i < shnum; ++i) {
3219         if (shdr[i].sh_type == SHT_SYMTAB) {
3220             sym_idx = i;
3221             str_idx = shdr[i].sh_link;
3222             goto found;
3223         }
3224     }
3225 
3226     /* There will be no symbol table if the file was stripped.  */
3227     return;
3228 
3229  found:
3230     /* Now know where the strtab and symtab are.  Snarf them.  */
3231     s = g_try_new(struct syminfo, 1);
3232     if (!s) {
3233         goto give_up;
3234     }
3235 
3236     segsz = shdr[str_idx].sh_size;
3237     s->disas_strtab = strings = g_try_malloc(segsz);
3238     if (!strings ||
3239         pread(fd, strings, segsz, shdr[str_idx].sh_offset) != segsz) {
3240         goto give_up;
3241     }
3242 
3243     segsz = shdr[sym_idx].sh_size;
3244     syms = g_try_malloc(segsz);
3245     if (!syms || pread(fd, syms, segsz, shdr[sym_idx].sh_offset) != segsz) {
3246         goto give_up;
3247     }
3248 
3249     if (segsz / sizeof(struct elf_sym) > INT_MAX) {
3250         /* Implausibly large symbol table: give up rather than ploughing
3251          * on with the number of symbols calculation overflowing
3252          */
3253         goto give_up;
3254     }
3255     nsyms = segsz / sizeof(struct elf_sym);
3256     for (i = 0; i < nsyms; ) {
3257         bswap_sym(syms + i);
3258         /* Throw away entries which we do not need.  */
3259         if (syms[i].st_shndx == SHN_UNDEF
3260             || syms[i].st_shndx >= SHN_LORESERVE
3261             || ELF_ST_TYPE(syms[i].st_info) != STT_FUNC) {
3262             if (i < --nsyms) {
3263                 syms[i] = syms[nsyms];
3264             }
3265         } else {
3266 #if defined(TARGET_ARM) || defined (TARGET_MIPS)
3267             /* The bottom address bit marks a Thumb or MIPS16 symbol.  */
3268             syms[i].st_value &= ~(target_ulong)1;
3269 #endif
3270             syms[i].st_value += load_bias;
3271             i++;
3272         }
3273     }
3274 
3275     /* No "useful" symbol.  */
3276     if (nsyms == 0) {
3277         goto give_up;
3278     }
3279 
3280     /* Attempt to free the storage associated with the local symbols
3281        that we threw away.  Whether or not this has any effect on the
3282        memory allocation depends on the malloc implementation and how
3283        many symbols we managed to discard.  */
3284     new_syms = g_try_renew(struct elf_sym, syms, nsyms);
3285     if (new_syms == NULL) {
3286         goto give_up;
3287     }
3288     syms = new_syms;
3289 
3290     qsort(syms, nsyms, sizeof(*syms), symcmp);
3291 
3292     s->disas_num_syms = nsyms;
3293 #if ELF_CLASS == ELFCLASS32
3294     s->disas_symtab.elf32 = syms;
3295 #else
3296     s->disas_symtab.elf64 = syms;
3297 #endif
3298     s->lookup_symbol = lookup_symbolxx;
3299     s->next = syminfos;
3300     syminfos = s;
3301 
3302     return;
3303 
3304 give_up:
3305     g_free(s);
3306     g_free(strings);
3307     g_free(syms);
3308 }
3309 
3310 uint32_t get_elf_eflags(int fd)
3311 {
3312     struct elfhdr ehdr;
3313     off_t offset;
3314     int ret;
3315 
3316     /* Read ELF header */
3317     offset = lseek(fd, 0, SEEK_SET);
3318     if (offset == (off_t) -1) {
3319         return 0;
3320     }
3321     ret = read(fd, &ehdr, sizeof(ehdr));
3322     if (ret < sizeof(ehdr)) {
3323         return 0;
3324     }
3325     offset = lseek(fd, offset, SEEK_SET);
3326     if (offset == (off_t) -1) {
3327         return 0;
3328     }
3329 
3330     /* Check ELF signature */
3331     if (!elf_check_ident(&ehdr)) {
3332         return 0;
3333     }
3334 
3335     /* check header */
3336     bswap_ehdr(&ehdr);
3337     if (!elf_check_ehdr(&ehdr)) {
3338         return 0;
3339     }
3340 
3341     /* return architecture id */
3342     return ehdr.e_flags;
3343 }
3344 
3345 int load_elf_binary(struct linux_binprm *bprm, struct image_info *info)
3346 {
3347     struct image_info interp_info;
3348     struct elfhdr elf_ex;
3349     char *elf_interpreter = NULL;
3350     char *scratch;
3351 
3352     memset(&interp_info, 0, sizeof(interp_info));
3353 #ifdef TARGET_MIPS
3354     interp_info.fp_abi = MIPS_ABI_FP_UNKNOWN;
3355 #endif
3356 
3357     info->start_mmap = (abi_ulong)ELF_START_MMAP;
3358 
3359     load_elf_image(bprm->filename, bprm->fd, info,
3360                    &elf_interpreter, bprm->buf);
3361 
3362     /* ??? We need a copy of the elf header for passing to create_elf_tables.
3363        If we do nothing, we'll have overwritten this when we re-use bprm->buf
3364        when we load the interpreter.  */
3365     elf_ex = *(struct elfhdr *)bprm->buf;
3366 
3367     /* Do this so that we can load the interpreter, if need be.  We will
3368        change some of these later */
3369     bprm->p = setup_arg_pages(bprm, info);
3370 
3371     scratch = g_new0(char, TARGET_PAGE_SIZE);
3372     if (STACK_GROWS_DOWN) {
3373         bprm->p = copy_elf_strings(1, &bprm->filename, scratch,
3374                                    bprm->p, info->stack_limit);
3375         info->file_string = bprm->p;
3376         bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch,
3377                                    bprm->p, info->stack_limit);
3378         info->env_strings = bprm->p;
3379         bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch,
3380                                    bprm->p, info->stack_limit);
3381         info->arg_strings = bprm->p;
3382     } else {
3383         info->arg_strings = bprm->p;
3384         bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch,
3385                                    bprm->p, info->stack_limit);
3386         info->env_strings = bprm->p;
3387         bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch,
3388                                    bprm->p, info->stack_limit);
3389         info->file_string = bprm->p;
3390         bprm->p = copy_elf_strings(1, &bprm->filename, scratch,
3391                                    bprm->p, info->stack_limit);
3392     }
3393 
3394     g_free(scratch);
3395 
3396     if (!bprm->p) {
3397         fprintf(stderr, "%s: %s\n", bprm->filename, strerror(E2BIG));
3398         exit(-1);
3399     }
3400 
3401     if (elf_interpreter) {
3402         load_elf_interp(elf_interpreter, &interp_info, bprm->buf);
3403 
3404         /* If the program interpreter is one of these two, then assume
3405            an iBCS2 image.  Otherwise assume a native linux image.  */
3406 
3407         if (strcmp(elf_interpreter, "/usr/lib/libc.so.1") == 0
3408             || strcmp(elf_interpreter, "/usr/lib/ld.so.1") == 0) {
3409             info->personality = PER_SVR4;
3410 
3411             /* Why this, you ask???  Well SVr4 maps page 0 as read-only,
3412                and some applications "depend" upon this behavior.  Since
3413                we do not have the power to recompile these, we emulate
3414                the SVr4 behavior.  Sigh.  */
3415             target_mmap(0, qemu_host_page_size, PROT_READ | PROT_EXEC,
3416                         MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
3417         }
3418 #ifdef TARGET_MIPS
3419         info->interp_fp_abi = interp_info.fp_abi;
3420 #endif
3421     }
3422 
3423     /*
3424      * TODO: load a vdso, which would also contain the signal trampolines.
3425      * Otherwise, allocate a private page to hold them.
3426      */
3427     if (TARGET_ARCH_HAS_SIGTRAMP_PAGE) {
3428         abi_long tramp_page = target_mmap(0, TARGET_PAGE_SIZE,
3429                                           PROT_READ | PROT_WRITE,
3430                                           MAP_PRIVATE | MAP_ANON, -1, 0);
3431         if (tramp_page == -1) {
3432             return -errno;
3433         }
3434 
3435         setup_sigtramp(tramp_page);
3436         target_mprotect(tramp_page, TARGET_PAGE_SIZE, PROT_READ | PROT_EXEC);
3437     }
3438 
3439     bprm->p = create_elf_tables(bprm->p, bprm->argc, bprm->envc, &elf_ex,
3440                                 info, (elf_interpreter ? &interp_info : NULL));
3441     info->start_stack = bprm->p;
3442 
3443     /* If we have an interpreter, set that as the program's entry point.
3444        Copy the load_bias as well, to help PPC64 interpret the entry
3445        point as a function descriptor.  Do this after creating elf tables
3446        so that we copy the original program entry point into the AUXV.  */
3447     if (elf_interpreter) {
3448         info->load_bias = interp_info.load_bias;
3449         info->entry = interp_info.entry;
3450         g_free(elf_interpreter);
3451     }
3452 
3453 #ifdef USE_ELF_CORE_DUMP
3454     bprm->core_dump = &elf_core_dump;
3455 #endif
3456 
3457     /*
3458      * If we reserved extra space for brk, release it now.
3459      * The implementation of do_brk in syscalls.c expects to be able
3460      * to mmap pages in this space.
3461      */
3462     if (info->reserve_brk) {
3463         abi_ulong start_brk = HOST_PAGE_ALIGN(info->brk);
3464         abi_ulong end_brk = HOST_PAGE_ALIGN(info->brk + info->reserve_brk);
3465         target_munmap(start_brk, end_brk - start_brk);
3466     }
3467 
3468     return 0;
3469 }
3470 
3471 #ifdef USE_ELF_CORE_DUMP
3472 /*
3473  * Definitions to generate Intel SVR4-like core files.
3474  * These mostly have the same names as the SVR4 types with "target_elf_"
3475  * tacked on the front to prevent clashes with linux definitions,
3476  * and the typedef forms have been avoided.  This is mostly like
3477  * the SVR4 structure, but more Linuxy, with things that Linux does
3478  * not support and which gdb doesn't really use excluded.
3479  *
3480  * Fields we don't dump (their contents is zero) in linux-user qemu
3481  * are marked with XXX.
3482  *
3483  * Core dump code is copied from linux kernel (fs/binfmt_elf.c).
3484  *
3485  * Porting ELF coredump for target is (quite) simple process.  First you
3486  * define USE_ELF_CORE_DUMP in target ELF code (where init_thread() for
3487  * the target resides):
3488  *
3489  * #define USE_ELF_CORE_DUMP
3490  *
3491  * Next you define type of register set used for dumping.  ELF specification
3492  * says that it needs to be array of elf_greg_t that has size of ELF_NREG.
3493  *
3494  * typedef <target_regtype> target_elf_greg_t;
3495  * #define ELF_NREG <number of registers>
3496  * typedef taret_elf_greg_t target_elf_gregset_t[ELF_NREG];
3497  *
3498  * Last step is to implement target specific function that copies registers
3499  * from given cpu into just specified register set.  Prototype is:
3500  *
3501  * static void elf_core_copy_regs(taret_elf_gregset_t *regs,
3502  *                                const CPUArchState *env);
3503  *
3504  * Parameters:
3505  *     regs - copy register values into here (allocated and zeroed by caller)
3506  *     env - copy registers from here
3507  *
3508  * Example for ARM target is provided in this file.
3509  */
3510 
3511 /* An ELF note in memory */
3512 struct memelfnote {
3513     const char *name;
3514     size_t     namesz;
3515     size_t     namesz_rounded;
3516     int        type;
3517     size_t     datasz;
3518     size_t     datasz_rounded;
3519     void       *data;
3520     size_t     notesz;
3521 };
3522 
3523 struct target_elf_siginfo {
3524     abi_int    si_signo; /* signal number */
3525     abi_int    si_code;  /* extra code */
3526     abi_int    si_errno; /* errno */
3527 };
3528 
3529 struct target_elf_prstatus {
3530     struct target_elf_siginfo pr_info;      /* Info associated with signal */
3531     abi_short          pr_cursig;    /* Current signal */
3532     abi_ulong          pr_sigpend;   /* XXX */
3533     abi_ulong          pr_sighold;   /* XXX */
3534     target_pid_t       pr_pid;
3535     target_pid_t       pr_ppid;
3536     target_pid_t       pr_pgrp;
3537     target_pid_t       pr_sid;
3538     struct target_timeval pr_utime;  /* XXX User time */
3539     struct target_timeval pr_stime;  /* XXX System time */
3540     struct target_timeval pr_cutime; /* XXX Cumulative user time */
3541     struct target_timeval pr_cstime; /* XXX Cumulative system time */
3542     target_elf_gregset_t      pr_reg;       /* GP registers */
3543     abi_int            pr_fpvalid;   /* XXX */
3544 };
3545 
3546 #define ELF_PRARGSZ     (80) /* Number of chars for args */
3547 
3548 struct target_elf_prpsinfo {
3549     char         pr_state;       /* numeric process state */
3550     char         pr_sname;       /* char for pr_state */
3551     char         pr_zomb;        /* zombie */
3552     char         pr_nice;        /* nice val */
3553     abi_ulong    pr_flag;        /* flags */
3554     target_uid_t pr_uid;
3555     target_gid_t pr_gid;
3556     target_pid_t pr_pid, pr_ppid, pr_pgrp, pr_sid;
3557     /* Lots missing */
3558     char    pr_fname[16] QEMU_NONSTRING; /* filename of executable */
3559     char    pr_psargs[ELF_PRARGSZ]; /* initial part of arg list */
3560 };
3561 
3562 /* Here is the structure in which status of each thread is captured. */
3563 struct elf_thread_status {
3564     QTAILQ_ENTRY(elf_thread_status)  ets_link;
3565     struct target_elf_prstatus prstatus;   /* NT_PRSTATUS */
3566 #if 0
3567     elf_fpregset_t fpu;             /* NT_PRFPREG */
3568     struct task_struct *thread;
3569     elf_fpxregset_t xfpu;           /* ELF_CORE_XFPREG_TYPE */
3570 #endif
3571     struct memelfnote notes[1];
3572     int num_notes;
3573 };
3574 
3575 struct elf_note_info {
3576     struct memelfnote   *notes;
3577     struct target_elf_prstatus *prstatus;  /* NT_PRSTATUS */
3578     struct target_elf_prpsinfo *psinfo;    /* NT_PRPSINFO */
3579 
3580     QTAILQ_HEAD(, elf_thread_status) thread_list;
3581 #if 0
3582     /*
3583      * Current version of ELF coredump doesn't support
3584      * dumping fp regs etc.
3585      */
3586     elf_fpregset_t *fpu;
3587     elf_fpxregset_t *xfpu;
3588     int thread_status_size;
3589 #endif
3590     int notes_size;
3591     int numnote;
3592 };
3593 
3594 struct vm_area_struct {
3595     target_ulong   vma_start;  /* start vaddr of memory region */
3596     target_ulong   vma_end;    /* end vaddr of memory region */
3597     abi_ulong      vma_flags;  /* protection etc. flags for the region */
3598     QTAILQ_ENTRY(vm_area_struct) vma_link;
3599 };
3600 
3601 struct mm_struct {
3602     QTAILQ_HEAD(, vm_area_struct) mm_mmap;
3603     int mm_count;           /* number of mappings */
3604 };
3605 
3606 static struct mm_struct *vma_init(void);
3607 static void vma_delete(struct mm_struct *);
3608 static int vma_add_mapping(struct mm_struct *, target_ulong,
3609                            target_ulong, abi_ulong);
3610 static int vma_get_mapping_count(const struct mm_struct *);
3611 static struct vm_area_struct *vma_first(const struct mm_struct *);
3612 static struct vm_area_struct *vma_next(struct vm_area_struct *);
3613 static abi_ulong vma_dump_size(const struct vm_area_struct *);
3614 static int vma_walker(void *priv, target_ulong start, target_ulong end,
3615                       unsigned long flags);
3616 
3617 static void fill_elf_header(struct elfhdr *, int, uint16_t, uint32_t);
3618 static void fill_note(struct memelfnote *, const char *, int,
3619                       unsigned int, void *);
3620 static void fill_prstatus(struct target_elf_prstatus *, const TaskState *, int);
3621 static int fill_psinfo(struct target_elf_prpsinfo *, const TaskState *);
3622 static void fill_auxv_note(struct memelfnote *, const TaskState *);
3623 static void fill_elf_note_phdr(struct elf_phdr *, int, off_t);
3624 static size_t note_size(const struct memelfnote *);
3625 static void free_note_info(struct elf_note_info *);
3626 static int fill_note_info(struct elf_note_info *, long, const CPUArchState *);
3627 static void fill_thread_info(struct elf_note_info *, const CPUArchState *);
3628 
3629 static int dump_write(int, const void *, size_t);
3630 static int write_note(struct memelfnote *, int);
3631 static int write_note_info(struct elf_note_info *, int);
3632 
3633 #ifdef BSWAP_NEEDED
3634 static void bswap_prstatus(struct target_elf_prstatus *prstatus)
3635 {
3636     prstatus->pr_info.si_signo = tswap32(prstatus->pr_info.si_signo);
3637     prstatus->pr_info.si_code = tswap32(prstatus->pr_info.si_code);
3638     prstatus->pr_info.si_errno = tswap32(prstatus->pr_info.si_errno);
3639     prstatus->pr_cursig = tswap16(prstatus->pr_cursig);
3640     prstatus->pr_sigpend = tswapal(prstatus->pr_sigpend);
3641     prstatus->pr_sighold = tswapal(prstatus->pr_sighold);
3642     prstatus->pr_pid = tswap32(prstatus->pr_pid);
3643     prstatus->pr_ppid = tswap32(prstatus->pr_ppid);
3644     prstatus->pr_pgrp = tswap32(prstatus->pr_pgrp);
3645     prstatus->pr_sid = tswap32(prstatus->pr_sid);
3646     /* cpu times are not filled, so we skip them */
3647     /* regs should be in correct format already */
3648     prstatus->pr_fpvalid = tswap32(prstatus->pr_fpvalid);
3649 }
3650 
3651 static void bswap_psinfo(struct target_elf_prpsinfo *psinfo)
3652 {
3653     psinfo->pr_flag = tswapal(psinfo->pr_flag);
3654     psinfo->pr_uid = tswap16(psinfo->pr_uid);
3655     psinfo->pr_gid = tswap16(psinfo->pr_gid);
3656     psinfo->pr_pid = tswap32(psinfo->pr_pid);
3657     psinfo->pr_ppid = tswap32(psinfo->pr_ppid);
3658     psinfo->pr_pgrp = tswap32(psinfo->pr_pgrp);
3659     psinfo->pr_sid = tswap32(psinfo->pr_sid);
3660 }
3661 
3662 static void bswap_note(struct elf_note *en)
3663 {
3664     bswap32s(&en->n_namesz);
3665     bswap32s(&en->n_descsz);
3666     bswap32s(&en->n_type);
3667 }
3668 #else
3669 static inline void bswap_prstatus(struct target_elf_prstatus *p) { }
3670 static inline void bswap_psinfo(struct target_elf_prpsinfo *p) {}
3671 static inline void bswap_note(struct elf_note *en) { }
3672 #endif /* BSWAP_NEEDED */
3673 
3674 /*
3675  * Minimal support for linux memory regions.  These are needed
3676  * when we are finding out what memory exactly belongs to
3677  * emulated process.  No locks needed here, as long as
3678  * thread that received the signal is stopped.
3679  */
3680 
3681 static struct mm_struct *vma_init(void)
3682 {
3683     struct mm_struct *mm;
3684 
3685     if ((mm = g_malloc(sizeof (*mm))) == NULL)
3686         return (NULL);
3687 
3688     mm->mm_count = 0;
3689     QTAILQ_INIT(&mm->mm_mmap);
3690 
3691     return (mm);
3692 }
3693 
3694 static void vma_delete(struct mm_struct *mm)
3695 {
3696     struct vm_area_struct *vma;
3697 
3698     while ((vma = vma_first(mm)) != NULL) {
3699         QTAILQ_REMOVE(&mm->mm_mmap, vma, vma_link);
3700         g_free(vma);
3701     }
3702     g_free(mm);
3703 }
3704 
3705 static int vma_add_mapping(struct mm_struct *mm, target_ulong start,
3706                            target_ulong end, abi_ulong flags)
3707 {
3708     struct vm_area_struct *vma;
3709 
3710     if ((vma = g_malloc0(sizeof (*vma))) == NULL)
3711         return (-1);
3712 
3713     vma->vma_start = start;
3714     vma->vma_end = end;
3715     vma->vma_flags = flags;
3716 
3717     QTAILQ_INSERT_TAIL(&mm->mm_mmap, vma, vma_link);
3718     mm->mm_count++;
3719 
3720     return (0);
3721 }
3722 
3723 static struct vm_area_struct *vma_first(const struct mm_struct *mm)
3724 {
3725     return (QTAILQ_FIRST(&mm->mm_mmap));
3726 }
3727 
3728 static struct vm_area_struct *vma_next(struct vm_area_struct *vma)
3729 {
3730     return (QTAILQ_NEXT(vma, vma_link));
3731 }
3732 
3733 static int vma_get_mapping_count(const struct mm_struct *mm)
3734 {
3735     return (mm->mm_count);
3736 }
3737 
3738 /*
3739  * Calculate file (dump) size of given memory region.
3740  */
3741 static abi_ulong vma_dump_size(const struct vm_area_struct *vma)
3742 {
3743     /* if we cannot even read the first page, skip it */
3744     if (!access_ok_untagged(VERIFY_READ, vma->vma_start, TARGET_PAGE_SIZE))
3745         return (0);
3746 
3747     /*
3748      * Usually we don't dump executable pages as they contain
3749      * non-writable code that debugger can read directly from
3750      * target library etc.  However, thread stacks are marked
3751      * also executable so we read in first page of given region
3752      * and check whether it contains elf header.  If there is
3753      * no elf header, we dump it.
3754      */
3755     if (vma->vma_flags & PROT_EXEC) {
3756         char page[TARGET_PAGE_SIZE];
3757 
3758         if (copy_from_user(page, vma->vma_start, sizeof (page))) {
3759             return 0;
3760         }
3761         if ((page[EI_MAG0] == ELFMAG0) &&
3762             (page[EI_MAG1] == ELFMAG1) &&
3763             (page[EI_MAG2] == ELFMAG2) &&
3764             (page[EI_MAG3] == ELFMAG3)) {
3765             /*
3766              * Mappings are possibly from ELF binary.  Don't dump
3767              * them.
3768              */
3769             return (0);
3770         }
3771     }
3772 
3773     return (vma->vma_end - vma->vma_start);
3774 }
3775 
3776 static int vma_walker(void *priv, target_ulong start, target_ulong end,
3777                       unsigned long flags)
3778 {
3779     struct mm_struct *mm = (struct mm_struct *)priv;
3780 
3781     vma_add_mapping(mm, start, end, flags);
3782     return (0);
3783 }
3784 
3785 static void fill_note(struct memelfnote *note, const char *name, int type,
3786                       unsigned int sz, void *data)
3787 {
3788     unsigned int namesz;
3789 
3790     namesz = strlen(name) + 1;
3791     note->name = name;
3792     note->namesz = namesz;
3793     note->namesz_rounded = roundup(namesz, sizeof (int32_t));
3794     note->type = type;
3795     note->datasz = sz;
3796     note->datasz_rounded = roundup(sz, sizeof (int32_t));
3797 
3798     note->data = data;
3799 
3800     /*
3801      * We calculate rounded up note size here as specified by
3802      * ELF document.
3803      */
3804     note->notesz = sizeof (struct elf_note) +
3805         note->namesz_rounded + note->datasz_rounded;
3806 }
3807 
3808 static void fill_elf_header(struct elfhdr *elf, int segs, uint16_t machine,
3809                             uint32_t flags)
3810 {
3811     (void) memset(elf, 0, sizeof(*elf));
3812 
3813     (void) memcpy(elf->e_ident, ELFMAG, SELFMAG);
3814     elf->e_ident[EI_CLASS] = ELF_CLASS;
3815     elf->e_ident[EI_DATA] = ELF_DATA;
3816     elf->e_ident[EI_VERSION] = EV_CURRENT;
3817     elf->e_ident[EI_OSABI] = ELF_OSABI;
3818 
3819     elf->e_type = ET_CORE;
3820     elf->e_machine = machine;
3821     elf->e_version = EV_CURRENT;
3822     elf->e_phoff = sizeof(struct elfhdr);
3823     elf->e_flags = flags;
3824     elf->e_ehsize = sizeof(struct elfhdr);
3825     elf->e_phentsize = sizeof(struct elf_phdr);
3826     elf->e_phnum = segs;
3827 
3828     bswap_ehdr(elf);
3829 }
3830 
3831 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, off_t offset)
3832 {
3833     phdr->p_type = PT_NOTE;
3834     phdr->p_offset = offset;
3835     phdr->p_vaddr = 0;
3836     phdr->p_paddr = 0;
3837     phdr->p_filesz = sz;
3838     phdr->p_memsz = 0;
3839     phdr->p_flags = 0;
3840     phdr->p_align = 0;
3841 
3842     bswap_phdr(phdr, 1);
3843 }
3844 
3845 static size_t note_size(const struct memelfnote *note)
3846 {
3847     return (note->notesz);
3848 }
3849 
3850 static void fill_prstatus(struct target_elf_prstatus *prstatus,
3851                           const TaskState *ts, int signr)
3852 {
3853     (void) memset(prstatus, 0, sizeof (*prstatus));
3854     prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
3855     prstatus->pr_pid = ts->ts_tid;
3856     prstatus->pr_ppid = getppid();
3857     prstatus->pr_pgrp = getpgrp();
3858     prstatus->pr_sid = getsid(0);
3859 
3860     bswap_prstatus(prstatus);
3861 }
3862 
3863 static int fill_psinfo(struct target_elf_prpsinfo *psinfo, const TaskState *ts)
3864 {
3865     char *base_filename;
3866     unsigned int i, len;
3867 
3868     (void) memset(psinfo, 0, sizeof (*psinfo));
3869 
3870     len = ts->info->env_strings - ts->info->arg_strings;
3871     if (len >= ELF_PRARGSZ)
3872         len = ELF_PRARGSZ - 1;
3873     if (copy_from_user(&psinfo->pr_psargs, ts->info->arg_strings, len)) {
3874         return -EFAULT;
3875     }
3876     for (i = 0; i < len; i++)
3877         if (psinfo->pr_psargs[i] == 0)
3878             psinfo->pr_psargs[i] = ' ';
3879     psinfo->pr_psargs[len] = 0;
3880 
3881     psinfo->pr_pid = getpid();
3882     psinfo->pr_ppid = getppid();
3883     psinfo->pr_pgrp = getpgrp();
3884     psinfo->pr_sid = getsid(0);
3885     psinfo->pr_uid = getuid();
3886     psinfo->pr_gid = getgid();
3887 
3888     base_filename = g_path_get_basename(ts->bprm->filename);
3889     /*
3890      * Using strncpy here is fine: at max-length,
3891      * this field is not NUL-terminated.
3892      */
3893     (void) strncpy(psinfo->pr_fname, base_filename,
3894                    sizeof(psinfo->pr_fname));
3895 
3896     g_free(base_filename);
3897     bswap_psinfo(psinfo);
3898     return (0);
3899 }
3900 
3901 static void fill_auxv_note(struct memelfnote *note, const TaskState *ts)
3902 {
3903     elf_addr_t auxv = (elf_addr_t)ts->info->saved_auxv;
3904     elf_addr_t orig_auxv = auxv;
3905     void *ptr;
3906     int len = ts->info->auxv_len;
3907 
3908     /*
3909      * Auxiliary vector is stored in target process stack.  It contains
3910      * {type, value} pairs that we need to dump into note.  This is not
3911      * strictly necessary but we do it here for sake of completeness.
3912      */
3913 
3914     /* read in whole auxv vector and copy it to memelfnote */
3915     ptr = lock_user(VERIFY_READ, orig_auxv, len, 0);
3916     if (ptr != NULL) {
3917         fill_note(note, "CORE", NT_AUXV, len, ptr);
3918         unlock_user(ptr, auxv, len);
3919     }
3920 }
3921 
3922 /*
3923  * Constructs name of coredump file.  We have following convention
3924  * for the name:
3925  *     qemu_<basename-of-target-binary>_<date>-<time>_<pid>.core
3926  *
3927  * Returns the filename
3928  */
3929 static char *core_dump_filename(const TaskState *ts)
3930 {
3931     g_autoptr(GDateTime) now = g_date_time_new_now_local();
3932     g_autofree char *nowstr = g_date_time_format(now, "%Y%m%d-%H%M%S");
3933     g_autofree char *base_filename = g_path_get_basename(ts->bprm->filename);
3934 
3935     return g_strdup_printf("qemu_%s_%s_%d.core",
3936                            base_filename, nowstr, (int)getpid());
3937 }
3938 
3939 static int dump_write(int fd, const void *ptr, size_t size)
3940 {
3941     const char *bufp = (const char *)ptr;
3942     ssize_t bytes_written, bytes_left;
3943     struct rlimit dumpsize;
3944     off_t pos;
3945 
3946     bytes_written = 0;
3947     getrlimit(RLIMIT_CORE, &dumpsize);
3948     if ((pos = lseek(fd, 0, SEEK_CUR))==-1) {
3949         if (errno == ESPIPE) { /* not a seekable stream */
3950             bytes_left = size;
3951         } else {
3952             return pos;
3953         }
3954     } else {
3955         if (dumpsize.rlim_cur <= pos) {
3956             return -1;
3957         } else if (dumpsize.rlim_cur == RLIM_INFINITY) {
3958             bytes_left = size;
3959         } else {
3960             size_t limit_left=dumpsize.rlim_cur - pos;
3961             bytes_left = limit_left >= size ? size : limit_left ;
3962         }
3963     }
3964 
3965     /*
3966      * In normal conditions, single write(2) should do but
3967      * in case of socket etc. this mechanism is more portable.
3968      */
3969     do {
3970         bytes_written = write(fd, bufp, bytes_left);
3971         if (bytes_written < 0) {
3972             if (errno == EINTR)
3973                 continue;
3974             return (-1);
3975         } else if (bytes_written == 0) { /* eof */
3976             return (-1);
3977         }
3978         bufp += bytes_written;
3979         bytes_left -= bytes_written;
3980     } while (bytes_left > 0);
3981 
3982     return (0);
3983 }
3984 
3985 static int write_note(struct memelfnote *men, int fd)
3986 {
3987     struct elf_note en;
3988 
3989     en.n_namesz = men->namesz;
3990     en.n_type = men->type;
3991     en.n_descsz = men->datasz;
3992 
3993     bswap_note(&en);
3994 
3995     if (dump_write(fd, &en, sizeof(en)) != 0)
3996         return (-1);
3997     if (dump_write(fd, men->name, men->namesz_rounded) != 0)
3998         return (-1);
3999     if (dump_write(fd, men->data, men->datasz_rounded) != 0)
4000         return (-1);
4001 
4002     return (0);
4003 }
4004 
4005 static void fill_thread_info(struct elf_note_info *info, const CPUArchState *env)
4006 {
4007     CPUState *cpu = env_cpu((CPUArchState *)env);
4008     TaskState *ts = (TaskState *)cpu->opaque;
4009     struct elf_thread_status *ets;
4010 
4011     ets = g_malloc0(sizeof (*ets));
4012     ets->num_notes = 1; /* only prstatus is dumped */
4013     fill_prstatus(&ets->prstatus, ts, 0);
4014     elf_core_copy_regs(&ets->prstatus.pr_reg, env);
4015     fill_note(&ets->notes[0], "CORE", NT_PRSTATUS, sizeof (ets->prstatus),
4016               &ets->prstatus);
4017 
4018     QTAILQ_INSERT_TAIL(&info->thread_list, ets, ets_link);
4019 
4020     info->notes_size += note_size(&ets->notes[0]);
4021 }
4022 
4023 static void init_note_info(struct elf_note_info *info)
4024 {
4025     /* Initialize the elf_note_info structure so that it is at
4026      * least safe to call free_note_info() on it. Must be
4027      * called before calling fill_note_info().
4028      */
4029     memset(info, 0, sizeof (*info));
4030     QTAILQ_INIT(&info->thread_list);
4031 }
4032 
4033 static int fill_note_info(struct elf_note_info *info,
4034                           long signr, const CPUArchState *env)
4035 {
4036 #define NUMNOTES 3
4037     CPUState *cpu = env_cpu((CPUArchState *)env);
4038     TaskState *ts = (TaskState *)cpu->opaque;
4039     int i;
4040 
4041     info->notes = g_new0(struct memelfnote, NUMNOTES);
4042     if (info->notes == NULL)
4043         return (-ENOMEM);
4044     info->prstatus = g_malloc0(sizeof (*info->prstatus));
4045     if (info->prstatus == NULL)
4046         return (-ENOMEM);
4047     info->psinfo = g_malloc0(sizeof (*info->psinfo));
4048     if (info->prstatus == NULL)
4049         return (-ENOMEM);
4050 
4051     /*
4052      * First fill in status (and registers) of current thread
4053      * including process info & aux vector.
4054      */
4055     fill_prstatus(info->prstatus, ts, signr);
4056     elf_core_copy_regs(&info->prstatus->pr_reg, env);
4057     fill_note(&info->notes[0], "CORE", NT_PRSTATUS,
4058               sizeof (*info->prstatus), info->prstatus);
4059     fill_psinfo(info->psinfo, ts);
4060     fill_note(&info->notes[1], "CORE", NT_PRPSINFO,
4061               sizeof (*info->psinfo), info->psinfo);
4062     fill_auxv_note(&info->notes[2], ts);
4063     info->numnote = 3;
4064 
4065     info->notes_size = 0;
4066     for (i = 0; i < info->numnote; i++)
4067         info->notes_size += note_size(&info->notes[i]);
4068 
4069     /* read and fill status of all threads */
4070     cpu_list_lock();
4071     CPU_FOREACH(cpu) {
4072         if (cpu == thread_cpu) {
4073             continue;
4074         }
4075         fill_thread_info(info, cpu->env_ptr);
4076     }
4077     cpu_list_unlock();
4078 
4079     return (0);
4080 }
4081 
4082 static void free_note_info(struct elf_note_info *info)
4083 {
4084     struct elf_thread_status *ets;
4085 
4086     while (!QTAILQ_EMPTY(&info->thread_list)) {
4087         ets = QTAILQ_FIRST(&info->thread_list);
4088         QTAILQ_REMOVE(&info->thread_list, ets, ets_link);
4089         g_free(ets);
4090     }
4091 
4092     g_free(info->prstatus);
4093     g_free(info->psinfo);
4094     g_free(info->notes);
4095 }
4096 
4097 static int write_note_info(struct elf_note_info *info, int fd)
4098 {
4099     struct elf_thread_status *ets;
4100     int i, error = 0;
4101 
4102     /* write prstatus, psinfo and auxv for current thread */
4103     for (i = 0; i < info->numnote; i++)
4104         if ((error = write_note(&info->notes[i], fd)) != 0)
4105             return (error);
4106 
4107     /* write prstatus for each thread */
4108     QTAILQ_FOREACH(ets, &info->thread_list, ets_link) {
4109         if ((error = write_note(&ets->notes[0], fd)) != 0)
4110             return (error);
4111     }
4112 
4113     return (0);
4114 }
4115 
4116 /*
4117  * Write out ELF coredump.
4118  *
4119  * See documentation of ELF object file format in:
4120  * http://www.caldera.com/developers/devspecs/gabi41.pdf
4121  *
4122  * Coredump format in linux is following:
4123  *
4124  * 0   +----------------------+         \
4125  *     | ELF header           | ET_CORE  |
4126  *     +----------------------+          |
4127  *     | ELF program headers  |          |--- headers
4128  *     | - NOTE section       |          |
4129  *     | - PT_LOAD sections   |          |
4130  *     +----------------------+         /
4131  *     | NOTEs:               |
4132  *     | - NT_PRSTATUS        |
4133  *     | - NT_PRSINFO         |
4134  *     | - NT_AUXV            |
4135  *     +----------------------+ <-- aligned to target page
4136  *     | Process memory dump  |
4137  *     :                      :
4138  *     .                      .
4139  *     :                      :
4140  *     |                      |
4141  *     +----------------------+
4142  *
4143  * NT_PRSTATUS -> struct elf_prstatus (per thread)
4144  * NT_PRSINFO  -> struct elf_prpsinfo
4145  * NT_AUXV is array of { type, value } pairs (see fill_auxv_note()).
4146  *
4147  * Format follows System V format as close as possible.  Current
4148  * version limitations are as follows:
4149  *     - no floating point registers are dumped
4150  *
4151  * Function returns 0 in case of success, negative errno otherwise.
4152  *
4153  * TODO: make this work also during runtime: it should be
4154  * possible to force coredump from running process and then
4155  * continue processing.  For example qemu could set up SIGUSR2
4156  * handler (provided that target process haven't registered
4157  * handler for that) that does the dump when signal is received.
4158  */
4159 static int elf_core_dump(int signr, const CPUArchState *env)
4160 {
4161     const CPUState *cpu = env_cpu((CPUArchState *)env);
4162     const TaskState *ts = (const TaskState *)cpu->opaque;
4163     struct vm_area_struct *vma = NULL;
4164     g_autofree char *corefile = NULL;
4165     struct elf_note_info info;
4166     struct elfhdr elf;
4167     struct elf_phdr phdr;
4168     struct rlimit dumpsize;
4169     struct mm_struct *mm = NULL;
4170     off_t offset = 0, data_offset = 0;
4171     int segs = 0;
4172     int fd = -1;
4173 
4174     init_note_info(&info);
4175 
4176     errno = 0;
4177     getrlimit(RLIMIT_CORE, &dumpsize);
4178     if (dumpsize.rlim_cur == 0)
4179         return 0;
4180 
4181     corefile = core_dump_filename(ts);
4182 
4183     if ((fd = open(corefile, O_WRONLY | O_CREAT,
4184                    S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH)) < 0)
4185         return (-errno);
4186 
4187     /*
4188      * Walk through target process memory mappings and
4189      * set up structure containing this information.  After
4190      * this point vma_xxx functions can be used.
4191      */
4192     if ((mm = vma_init()) == NULL)
4193         goto out;
4194 
4195     walk_memory_regions(mm, vma_walker);
4196     segs = vma_get_mapping_count(mm);
4197 
4198     /*
4199      * Construct valid coredump ELF header.  We also
4200      * add one more segment for notes.
4201      */
4202     fill_elf_header(&elf, segs + 1, ELF_MACHINE, 0);
4203     if (dump_write(fd, &elf, sizeof (elf)) != 0)
4204         goto out;
4205 
4206     /* fill in the in-memory version of notes */
4207     if (fill_note_info(&info, signr, env) < 0)
4208         goto out;
4209 
4210     offset += sizeof (elf);                             /* elf header */
4211     offset += (segs + 1) * sizeof (struct elf_phdr);    /* program headers */
4212 
4213     /* write out notes program header */
4214     fill_elf_note_phdr(&phdr, info.notes_size, offset);
4215 
4216     offset += info.notes_size;
4217     if (dump_write(fd, &phdr, sizeof (phdr)) != 0)
4218         goto out;
4219 
4220     /*
4221      * ELF specification wants data to start at page boundary so
4222      * we align it here.
4223      */
4224     data_offset = offset = roundup(offset, ELF_EXEC_PAGESIZE);
4225 
4226     /*
4227      * Write program headers for memory regions mapped in
4228      * the target process.
4229      */
4230     for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) {
4231         (void) memset(&phdr, 0, sizeof (phdr));
4232 
4233         phdr.p_type = PT_LOAD;
4234         phdr.p_offset = offset;
4235         phdr.p_vaddr = vma->vma_start;
4236         phdr.p_paddr = 0;
4237         phdr.p_filesz = vma_dump_size(vma);
4238         offset += phdr.p_filesz;
4239         phdr.p_memsz = vma->vma_end - vma->vma_start;
4240         phdr.p_flags = vma->vma_flags & PROT_READ ? PF_R : 0;
4241         if (vma->vma_flags & PROT_WRITE)
4242             phdr.p_flags |= PF_W;
4243         if (vma->vma_flags & PROT_EXEC)
4244             phdr.p_flags |= PF_X;
4245         phdr.p_align = ELF_EXEC_PAGESIZE;
4246 
4247         bswap_phdr(&phdr, 1);
4248         if (dump_write(fd, &phdr, sizeof(phdr)) != 0) {
4249             goto out;
4250         }
4251     }
4252 
4253     /*
4254      * Next we write notes just after program headers.  No
4255      * alignment needed here.
4256      */
4257     if (write_note_info(&info, fd) < 0)
4258         goto out;
4259 
4260     /* align data to page boundary */
4261     if (lseek(fd, data_offset, SEEK_SET) != data_offset)
4262         goto out;
4263 
4264     /*
4265      * Finally we can dump process memory into corefile as well.
4266      */
4267     for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) {
4268         abi_ulong addr;
4269         abi_ulong end;
4270 
4271         end = vma->vma_start + vma_dump_size(vma);
4272 
4273         for (addr = vma->vma_start; addr < end;
4274              addr += TARGET_PAGE_SIZE) {
4275             char page[TARGET_PAGE_SIZE];
4276             int error;
4277 
4278             /*
4279              *  Read in page from target process memory and
4280              *  write it to coredump file.
4281              */
4282             error = copy_from_user(page, addr, sizeof (page));
4283             if (error != 0) {
4284                 (void) fprintf(stderr, "unable to dump " TARGET_ABI_FMT_lx "\n",
4285                                addr);
4286                 errno = -error;
4287                 goto out;
4288             }
4289             if (dump_write(fd, page, TARGET_PAGE_SIZE) < 0)
4290                 goto out;
4291         }
4292     }
4293 
4294  out:
4295     free_note_info(&info);
4296     if (mm != NULL)
4297         vma_delete(mm);
4298     (void) close(fd);
4299 
4300     if (errno != 0)
4301         return (-errno);
4302     return (0);
4303 }
4304 #endif /* USE_ELF_CORE_DUMP */
4305 
4306 void do_init_thread(struct target_pt_regs *regs, struct image_info *infop)
4307 {
4308     init_thread(regs, infop);
4309 }
4310