xref: /qemu/linux-user/elfload.c (revision 4d7dd4ed)
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 "qemu/lockable.h"
21 #include "qapi/error.h"
22 #include "qemu/error-report.h"
23 #include "target_signal.h"
24 #include "accel/tcg/debuginfo.h"
25 
26 #ifdef TARGET_ARM
27 #include "target/arm/cpu-features.h"
28 #endif
29 
30 #ifdef _ARCH_PPC64
31 #undef ARCH_DLINFO
32 #undef ELF_PLATFORM
33 #undef ELF_HWCAP
34 #undef ELF_HWCAP2
35 #undef ELF_CLASS
36 #undef ELF_DATA
37 #undef ELF_ARCH
38 #endif
39 
40 #ifndef TARGET_ARCH_HAS_SIGTRAMP_PAGE
41 #define TARGET_ARCH_HAS_SIGTRAMP_PAGE 0
42 #endif
43 
44 typedef struct {
45     const uint8_t *image;
46     const uint32_t *relocs;
47     unsigned image_size;
48     unsigned reloc_count;
49     unsigned sigreturn_ofs;
50     unsigned rt_sigreturn_ofs;
51 } VdsoImageInfo;
52 
53 #define ELF_OSABI   ELFOSABI_SYSV
54 
55 /* from personality.h */
56 
57 /*
58  * Flags for bug emulation.
59  *
60  * These occupy the top three bytes.
61  */
62 enum {
63     ADDR_NO_RANDOMIZE = 0x0040000,      /* disable randomization of VA space */
64     FDPIC_FUNCPTRS =    0x0080000,      /* userspace function ptrs point to
65                                            descriptors (signal handling) */
66     MMAP_PAGE_ZERO =    0x0100000,
67     ADDR_COMPAT_LAYOUT = 0x0200000,
68     READ_IMPLIES_EXEC = 0x0400000,
69     ADDR_LIMIT_32BIT =  0x0800000,
70     SHORT_INODE =       0x1000000,
71     WHOLE_SECONDS =     0x2000000,
72     STICKY_TIMEOUTS =   0x4000000,
73     ADDR_LIMIT_3GB =    0x8000000,
74 };
75 
76 /*
77  * Personality types.
78  *
79  * These go in the low byte.  Avoid using the top bit, it will
80  * conflict with error returns.
81  */
82 enum {
83     PER_LINUX =         0x0000,
84     PER_LINUX_32BIT =   0x0000 | ADDR_LIMIT_32BIT,
85     PER_LINUX_FDPIC =   0x0000 | FDPIC_FUNCPTRS,
86     PER_SVR4 =          0x0001 | STICKY_TIMEOUTS | MMAP_PAGE_ZERO,
87     PER_SVR3 =          0x0002 | STICKY_TIMEOUTS | SHORT_INODE,
88     PER_SCOSVR3 =       0x0003 | STICKY_TIMEOUTS | WHOLE_SECONDS | SHORT_INODE,
89     PER_OSR5 =          0x0003 | STICKY_TIMEOUTS | WHOLE_SECONDS,
90     PER_WYSEV386 =      0x0004 | STICKY_TIMEOUTS | SHORT_INODE,
91     PER_ISCR4 =         0x0005 | STICKY_TIMEOUTS,
92     PER_BSD =           0x0006,
93     PER_SUNOS =         0x0006 | STICKY_TIMEOUTS,
94     PER_XENIX =         0x0007 | STICKY_TIMEOUTS | SHORT_INODE,
95     PER_LINUX32 =       0x0008,
96     PER_LINUX32_3GB =   0x0008 | ADDR_LIMIT_3GB,
97     PER_IRIX32 =        0x0009 | STICKY_TIMEOUTS,/* IRIX5 32-bit */
98     PER_IRIXN32 =       0x000a | STICKY_TIMEOUTS,/* IRIX6 new 32-bit */
99     PER_IRIX64 =        0x000b | STICKY_TIMEOUTS,/* IRIX6 64-bit */
100     PER_RISCOS =        0x000c,
101     PER_SOLARIS =       0x000d | STICKY_TIMEOUTS,
102     PER_UW7 =           0x000e | STICKY_TIMEOUTS | MMAP_PAGE_ZERO,
103     PER_OSF4 =          0x000f,                  /* OSF/1 v4 */
104     PER_HPUX =          0x0010,
105     PER_MASK =          0x00ff,
106 };
107 
108 /*
109  * Return the base personality without flags.
110  */
111 #define personality(pers)       (pers & PER_MASK)
112 
113 int info_is_fdpic(struct image_info *info)
114 {
115     return info->personality == PER_LINUX_FDPIC;
116 }
117 
118 /* this flag is uneffective under linux too, should be deleted */
119 #ifndef MAP_DENYWRITE
120 #define MAP_DENYWRITE 0
121 #endif
122 
123 /* should probably go in elf.h */
124 #ifndef ELIBBAD
125 #define ELIBBAD 80
126 #endif
127 
128 #if TARGET_BIG_ENDIAN
129 #define ELF_DATA        ELFDATA2MSB
130 #else
131 #define ELF_DATA        ELFDATA2LSB
132 #endif
133 
134 #ifdef TARGET_ABI_MIPSN32
135 typedef abi_ullong      target_elf_greg_t;
136 #define tswapreg(ptr)   tswap64(ptr)
137 #else
138 typedef abi_ulong       target_elf_greg_t;
139 #define tswapreg(ptr)   tswapal(ptr)
140 #endif
141 
142 #ifdef USE_UID16
143 typedef abi_ushort      target_uid_t;
144 typedef abi_ushort      target_gid_t;
145 #else
146 typedef abi_uint        target_uid_t;
147 typedef abi_uint        target_gid_t;
148 #endif
149 typedef abi_int         target_pid_t;
150 
151 #ifdef TARGET_I386
152 
153 #define ELF_HWCAP get_elf_hwcap()
154 
155 static uint32_t get_elf_hwcap(void)
156 {
157     X86CPU *cpu = X86_CPU(thread_cpu);
158 
159     return cpu->env.features[FEAT_1_EDX];
160 }
161 
162 #ifdef TARGET_X86_64
163 #define ELF_CLASS      ELFCLASS64
164 #define ELF_ARCH       EM_X86_64
165 
166 #define ELF_PLATFORM   "x86_64"
167 
168 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop)
169 {
170     regs->rax = 0;
171     regs->rsp = infop->start_stack;
172     regs->rip = infop->entry;
173 }
174 
175 #define ELF_NREG    27
176 typedef target_elf_greg_t  target_elf_gregset_t[ELF_NREG];
177 
178 /*
179  * Note that ELF_NREG should be 29 as there should be place for
180  * TRAPNO and ERR "registers" as well but linux doesn't dump
181  * those.
182  *
183  * See linux kernel: arch/x86/include/asm/elf.h
184  */
185 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUX86State *env)
186 {
187     (*regs)[0] = tswapreg(env->regs[15]);
188     (*regs)[1] = tswapreg(env->regs[14]);
189     (*regs)[2] = tswapreg(env->regs[13]);
190     (*regs)[3] = tswapreg(env->regs[12]);
191     (*regs)[4] = tswapreg(env->regs[R_EBP]);
192     (*regs)[5] = tswapreg(env->regs[R_EBX]);
193     (*regs)[6] = tswapreg(env->regs[11]);
194     (*regs)[7] = tswapreg(env->regs[10]);
195     (*regs)[8] = tswapreg(env->regs[9]);
196     (*regs)[9] = tswapreg(env->regs[8]);
197     (*regs)[10] = tswapreg(env->regs[R_EAX]);
198     (*regs)[11] = tswapreg(env->regs[R_ECX]);
199     (*regs)[12] = tswapreg(env->regs[R_EDX]);
200     (*regs)[13] = tswapreg(env->regs[R_ESI]);
201     (*regs)[14] = tswapreg(env->regs[R_EDI]);
202     (*regs)[15] = tswapreg(env->regs[R_EAX]); /* XXX */
203     (*regs)[16] = tswapreg(env->eip);
204     (*regs)[17] = tswapreg(env->segs[R_CS].selector & 0xffff);
205     (*regs)[18] = tswapreg(env->eflags);
206     (*regs)[19] = tswapreg(env->regs[R_ESP]);
207     (*regs)[20] = tswapreg(env->segs[R_SS].selector & 0xffff);
208     (*regs)[21] = tswapreg(env->segs[R_FS].selector & 0xffff);
209     (*regs)[22] = tswapreg(env->segs[R_GS].selector & 0xffff);
210     (*regs)[23] = tswapreg(env->segs[R_DS].selector & 0xffff);
211     (*regs)[24] = tswapreg(env->segs[R_ES].selector & 0xffff);
212     (*regs)[25] = tswapreg(env->segs[R_FS].selector & 0xffff);
213     (*regs)[26] = tswapreg(env->segs[R_GS].selector & 0xffff);
214 }
215 
216 #if ULONG_MAX > UINT32_MAX
217 #define INIT_GUEST_COMMPAGE
218 static bool init_guest_commpage(void)
219 {
220     /*
221      * The vsyscall page is at a high negative address aka kernel space,
222      * which means that we cannot actually allocate it with target_mmap.
223      * We still should be able to use page_set_flags, unless the user
224      * has specified -R reserved_va, which would trigger an assert().
225      */
226     if (reserved_va != 0 &&
227         TARGET_VSYSCALL_PAGE + TARGET_PAGE_SIZE - 1 > reserved_va) {
228         error_report("Cannot allocate vsyscall page");
229         exit(EXIT_FAILURE);
230     }
231     page_set_flags(TARGET_VSYSCALL_PAGE,
232                    TARGET_VSYSCALL_PAGE | ~TARGET_PAGE_MASK,
233                    PAGE_EXEC | PAGE_VALID);
234     return true;
235 }
236 #endif
237 #else
238 
239 /*
240  * This is used to ensure we don't load something for the wrong architecture.
241  */
242 #define elf_check_arch(x) ( ((x) == EM_386) || ((x) == EM_486) )
243 
244 /*
245  * These are used to set parameters in the core dumps.
246  */
247 #define ELF_CLASS       ELFCLASS32
248 #define ELF_ARCH        EM_386
249 
250 #define ELF_PLATFORM get_elf_platform()
251 #define EXSTACK_DEFAULT true
252 
253 static const char *get_elf_platform(void)
254 {
255     static char elf_platform[] = "i386";
256     int family = object_property_get_int(OBJECT(thread_cpu), "family", NULL);
257     if (family > 6) {
258         family = 6;
259     }
260     if (family >= 3) {
261         elf_platform[1] = '0' + family;
262     }
263     return elf_platform;
264 }
265 
266 static inline void init_thread(struct target_pt_regs *regs,
267                                struct image_info *infop)
268 {
269     regs->esp = infop->start_stack;
270     regs->eip = infop->entry;
271 
272     /* SVR4/i386 ABI (pages 3-31, 3-32) says that when the program
273        starts %edx contains a pointer to a function which might be
274        registered using `atexit'.  This provides a mean for the
275        dynamic linker to call DT_FINI functions for shared libraries
276        that have been loaded before the code runs.
277 
278        A value of 0 tells we have no such handler.  */
279     regs->edx = 0;
280 }
281 
282 #define ELF_NREG    17
283 typedef target_elf_greg_t  target_elf_gregset_t[ELF_NREG];
284 
285 /*
286  * Note that ELF_NREG should be 19 as there should be place for
287  * TRAPNO and ERR "registers" as well but linux doesn't dump
288  * those.
289  *
290  * See linux kernel: arch/x86/include/asm/elf.h
291  */
292 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUX86State *env)
293 {
294     (*regs)[0] = tswapreg(env->regs[R_EBX]);
295     (*regs)[1] = tswapreg(env->regs[R_ECX]);
296     (*regs)[2] = tswapreg(env->regs[R_EDX]);
297     (*regs)[3] = tswapreg(env->regs[R_ESI]);
298     (*regs)[4] = tswapreg(env->regs[R_EDI]);
299     (*regs)[5] = tswapreg(env->regs[R_EBP]);
300     (*regs)[6] = tswapreg(env->regs[R_EAX]);
301     (*regs)[7] = tswapreg(env->segs[R_DS].selector & 0xffff);
302     (*regs)[8] = tswapreg(env->segs[R_ES].selector & 0xffff);
303     (*regs)[9] = tswapreg(env->segs[R_FS].selector & 0xffff);
304     (*regs)[10] = tswapreg(env->segs[R_GS].selector & 0xffff);
305     (*regs)[11] = tswapreg(env->regs[R_EAX]); /* XXX */
306     (*regs)[12] = tswapreg(env->eip);
307     (*regs)[13] = tswapreg(env->segs[R_CS].selector & 0xffff);
308     (*regs)[14] = tswapreg(env->eflags);
309     (*regs)[15] = tswapreg(env->regs[R_ESP]);
310     (*regs)[16] = tswapreg(env->segs[R_SS].selector & 0xffff);
311 }
312 
313 /*
314  * i386 is the only target which supplies AT_SYSINFO for the vdso.
315  * All others only supply AT_SYSINFO_EHDR.
316  */
317 #define DLINFO_ARCH_ITEMS (vdso_info != NULL)
318 #define ARCH_DLINFO                                     \
319     do {                                                \
320         if (vdso_info) {                                \
321             NEW_AUX_ENT(AT_SYSINFO, vdso_info->entry);  \
322         }                                               \
323     } while (0)
324 
325 #endif /* TARGET_X86_64 */
326 
327 #define VDSO_HEADER "vdso.c.inc"
328 
329 #define USE_ELF_CORE_DUMP
330 #define ELF_EXEC_PAGESIZE       4096
331 
332 #endif /* TARGET_I386 */
333 
334 #ifdef TARGET_ARM
335 
336 #ifndef TARGET_AARCH64
337 /* 32 bit ARM definitions */
338 
339 #define ELF_ARCH        EM_ARM
340 #define ELF_CLASS       ELFCLASS32
341 #define EXSTACK_DEFAULT true
342 
343 static inline void init_thread(struct target_pt_regs *regs,
344                                struct image_info *infop)
345 {
346     abi_long stack = infop->start_stack;
347     memset(regs, 0, sizeof(*regs));
348 
349     regs->uregs[16] = ARM_CPU_MODE_USR;
350     if (infop->entry & 1) {
351         regs->uregs[16] |= CPSR_T;
352     }
353     regs->uregs[15] = infop->entry & 0xfffffffe;
354     regs->uregs[13] = infop->start_stack;
355     /* FIXME - what to for failure of get_user()? */
356     get_user_ual(regs->uregs[2], stack + 8); /* envp */
357     get_user_ual(regs->uregs[1], stack + 4); /* envp */
358     /* XXX: it seems that r0 is zeroed after ! */
359     regs->uregs[0] = 0;
360     /* For uClinux PIC binaries.  */
361     /* XXX: Linux does this only on ARM with no MMU (do we care ?) */
362     regs->uregs[10] = infop->start_data;
363 
364     /* Support ARM FDPIC.  */
365     if (info_is_fdpic(infop)) {
366         /* As described in the ABI document, r7 points to the loadmap info
367          * prepared by the kernel. If an interpreter is needed, r8 points
368          * to the interpreter loadmap and r9 points to the interpreter
369          * PT_DYNAMIC info. If no interpreter is needed, r8 is zero, and
370          * r9 points to the main program PT_DYNAMIC info.
371          */
372         regs->uregs[7] = infop->loadmap_addr;
373         if (infop->interpreter_loadmap_addr) {
374             /* Executable is dynamically loaded.  */
375             regs->uregs[8] = infop->interpreter_loadmap_addr;
376             regs->uregs[9] = infop->interpreter_pt_dynamic_addr;
377         } else {
378             regs->uregs[8] = 0;
379             regs->uregs[9] = infop->pt_dynamic_addr;
380         }
381     }
382 }
383 
384 #define ELF_NREG    18
385 typedef target_elf_greg_t  target_elf_gregset_t[ELF_NREG];
386 
387 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUARMState *env)
388 {
389     (*regs)[0] = tswapreg(env->regs[0]);
390     (*regs)[1] = tswapreg(env->regs[1]);
391     (*regs)[2] = tswapreg(env->regs[2]);
392     (*regs)[3] = tswapreg(env->regs[3]);
393     (*regs)[4] = tswapreg(env->regs[4]);
394     (*regs)[5] = tswapreg(env->regs[5]);
395     (*regs)[6] = tswapreg(env->regs[6]);
396     (*regs)[7] = tswapreg(env->regs[7]);
397     (*regs)[8] = tswapreg(env->regs[8]);
398     (*regs)[9] = tswapreg(env->regs[9]);
399     (*regs)[10] = tswapreg(env->regs[10]);
400     (*regs)[11] = tswapreg(env->regs[11]);
401     (*regs)[12] = tswapreg(env->regs[12]);
402     (*regs)[13] = tswapreg(env->regs[13]);
403     (*regs)[14] = tswapreg(env->regs[14]);
404     (*regs)[15] = tswapreg(env->regs[15]);
405 
406     (*regs)[16] = tswapreg(cpsr_read((CPUARMState *)env));
407     (*regs)[17] = tswapreg(env->regs[0]); /* XXX */
408 }
409 
410 #define USE_ELF_CORE_DUMP
411 #define ELF_EXEC_PAGESIZE       4096
412 
413 enum
414 {
415     ARM_HWCAP_ARM_SWP       = 1 << 0,
416     ARM_HWCAP_ARM_HALF      = 1 << 1,
417     ARM_HWCAP_ARM_THUMB     = 1 << 2,
418     ARM_HWCAP_ARM_26BIT     = 1 << 3,
419     ARM_HWCAP_ARM_FAST_MULT = 1 << 4,
420     ARM_HWCAP_ARM_FPA       = 1 << 5,
421     ARM_HWCAP_ARM_VFP       = 1 << 6,
422     ARM_HWCAP_ARM_EDSP      = 1 << 7,
423     ARM_HWCAP_ARM_JAVA      = 1 << 8,
424     ARM_HWCAP_ARM_IWMMXT    = 1 << 9,
425     ARM_HWCAP_ARM_CRUNCH    = 1 << 10,
426     ARM_HWCAP_ARM_THUMBEE   = 1 << 11,
427     ARM_HWCAP_ARM_NEON      = 1 << 12,
428     ARM_HWCAP_ARM_VFPv3     = 1 << 13,
429     ARM_HWCAP_ARM_VFPv3D16  = 1 << 14,
430     ARM_HWCAP_ARM_TLS       = 1 << 15,
431     ARM_HWCAP_ARM_VFPv4     = 1 << 16,
432     ARM_HWCAP_ARM_IDIVA     = 1 << 17,
433     ARM_HWCAP_ARM_IDIVT     = 1 << 18,
434     ARM_HWCAP_ARM_VFPD32    = 1 << 19,
435     ARM_HWCAP_ARM_LPAE      = 1 << 20,
436     ARM_HWCAP_ARM_EVTSTRM   = 1 << 21,
437     ARM_HWCAP_ARM_FPHP      = 1 << 22,
438     ARM_HWCAP_ARM_ASIMDHP   = 1 << 23,
439     ARM_HWCAP_ARM_ASIMDDP   = 1 << 24,
440     ARM_HWCAP_ARM_ASIMDFHM  = 1 << 25,
441     ARM_HWCAP_ARM_ASIMDBF16 = 1 << 26,
442     ARM_HWCAP_ARM_I8MM      = 1 << 27,
443 };
444 
445 enum {
446     ARM_HWCAP2_ARM_AES      = 1 << 0,
447     ARM_HWCAP2_ARM_PMULL    = 1 << 1,
448     ARM_HWCAP2_ARM_SHA1     = 1 << 2,
449     ARM_HWCAP2_ARM_SHA2     = 1 << 3,
450     ARM_HWCAP2_ARM_CRC32    = 1 << 4,
451     ARM_HWCAP2_ARM_SB       = 1 << 5,
452     ARM_HWCAP2_ARM_SSBS     = 1 << 6,
453 };
454 
455 /* The commpage only exists for 32 bit kernels */
456 
457 #define HI_COMMPAGE (intptr_t)0xffff0f00u
458 
459 static bool init_guest_commpage(void)
460 {
461     ARMCPU *cpu = ARM_CPU(thread_cpu);
462     abi_ptr commpage;
463     void *want;
464     void *addr;
465 
466     /*
467      * M-profile allocates maximum of 2GB address space, so can never
468      * allocate the commpage.  Skip it.
469      */
470     if (arm_feature(&cpu->env, ARM_FEATURE_M)) {
471         return true;
472     }
473 
474     commpage = HI_COMMPAGE & -qemu_host_page_size;
475     want = g2h_untagged(commpage);
476     addr = mmap(want, qemu_host_page_size, PROT_READ | PROT_WRITE,
477                 MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0);
478 
479     if (addr == MAP_FAILED) {
480         perror("Allocating guest commpage");
481         exit(EXIT_FAILURE);
482     }
483     if (addr != want) {
484         return false;
485     }
486 
487     /* Set kernel helper versions; rest of page is 0.  */
488     __put_user(5, (uint32_t *)g2h_untagged(0xffff0ffcu));
489 
490     if (mprotect(addr, qemu_host_page_size, PROT_READ)) {
491         perror("Protecting guest commpage");
492         exit(EXIT_FAILURE);
493     }
494 
495     page_set_flags(commpage, commpage | ~qemu_host_page_mask,
496                    PAGE_READ | PAGE_EXEC | PAGE_VALID);
497     return true;
498 }
499 
500 #define ELF_HWCAP get_elf_hwcap()
501 #define ELF_HWCAP2 get_elf_hwcap2()
502 
503 uint32_t get_elf_hwcap(void)
504 {
505     ARMCPU *cpu = ARM_CPU(thread_cpu);
506     uint32_t hwcaps = 0;
507 
508     hwcaps |= ARM_HWCAP_ARM_SWP;
509     hwcaps |= ARM_HWCAP_ARM_HALF;
510     hwcaps |= ARM_HWCAP_ARM_THUMB;
511     hwcaps |= ARM_HWCAP_ARM_FAST_MULT;
512 
513     /* probe for the extra features */
514 #define GET_FEATURE(feat, hwcap) \
515     do { if (arm_feature(&cpu->env, feat)) { hwcaps |= hwcap; } } while (0)
516 
517 #define GET_FEATURE_ID(feat, hwcap) \
518     do { if (cpu_isar_feature(feat, cpu)) { hwcaps |= hwcap; } } while (0)
519 
520     /* EDSP is in v5TE and above, but all our v5 CPUs are v5TE */
521     GET_FEATURE(ARM_FEATURE_V5, ARM_HWCAP_ARM_EDSP);
522     GET_FEATURE(ARM_FEATURE_IWMMXT, ARM_HWCAP_ARM_IWMMXT);
523     GET_FEATURE(ARM_FEATURE_THUMB2EE, ARM_HWCAP_ARM_THUMBEE);
524     GET_FEATURE(ARM_FEATURE_NEON, ARM_HWCAP_ARM_NEON);
525     GET_FEATURE(ARM_FEATURE_V6K, ARM_HWCAP_ARM_TLS);
526     GET_FEATURE(ARM_FEATURE_LPAE, ARM_HWCAP_ARM_LPAE);
527     GET_FEATURE_ID(aa32_arm_div, ARM_HWCAP_ARM_IDIVA);
528     GET_FEATURE_ID(aa32_thumb_div, ARM_HWCAP_ARM_IDIVT);
529     GET_FEATURE_ID(aa32_vfp, ARM_HWCAP_ARM_VFP);
530 
531     if (cpu_isar_feature(aa32_fpsp_v3, cpu) ||
532         cpu_isar_feature(aa32_fpdp_v3, cpu)) {
533         hwcaps |= ARM_HWCAP_ARM_VFPv3;
534         if (cpu_isar_feature(aa32_simd_r32, cpu)) {
535             hwcaps |= ARM_HWCAP_ARM_VFPD32;
536         } else {
537             hwcaps |= ARM_HWCAP_ARM_VFPv3D16;
538         }
539     }
540     GET_FEATURE_ID(aa32_simdfmac, ARM_HWCAP_ARM_VFPv4);
541     /*
542      * MVFR1.FPHP and .SIMDHP must be in sync, and QEMU uses the same
543      * isar_feature function for both. The kernel reports them as two hwcaps.
544      */
545     GET_FEATURE_ID(aa32_fp16_arith, ARM_HWCAP_ARM_FPHP);
546     GET_FEATURE_ID(aa32_fp16_arith, ARM_HWCAP_ARM_ASIMDHP);
547     GET_FEATURE_ID(aa32_dp, ARM_HWCAP_ARM_ASIMDDP);
548     GET_FEATURE_ID(aa32_fhm, ARM_HWCAP_ARM_ASIMDFHM);
549     GET_FEATURE_ID(aa32_bf16, ARM_HWCAP_ARM_ASIMDBF16);
550     GET_FEATURE_ID(aa32_i8mm, ARM_HWCAP_ARM_I8MM);
551 
552     return hwcaps;
553 }
554 
555 uint32_t get_elf_hwcap2(void)
556 {
557     ARMCPU *cpu = ARM_CPU(thread_cpu);
558     uint32_t hwcaps = 0;
559 
560     GET_FEATURE_ID(aa32_aes, ARM_HWCAP2_ARM_AES);
561     GET_FEATURE_ID(aa32_pmull, ARM_HWCAP2_ARM_PMULL);
562     GET_FEATURE_ID(aa32_sha1, ARM_HWCAP2_ARM_SHA1);
563     GET_FEATURE_ID(aa32_sha2, ARM_HWCAP2_ARM_SHA2);
564     GET_FEATURE_ID(aa32_crc32, ARM_HWCAP2_ARM_CRC32);
565     GET_FEATURE_ID(aa32_sb, ARM_HWCAP2_ARM_SB);
566     GET_FEATURE_ID(aa32_ssbs, ARM_HWCAP2_ARM_SSBS);
567     return hwcaps;
568 }
569 
570 const char *elf_hwcap_str(uint32_t bit)
571 {
572     static const char *hwcap_str[] = {
573     [__builtin_ctz(ARM_HWCAP_ARM_SWP      )] = "swp",
574     [__builtin_ctz(ARM_HWCAP_ARM_HALF     )] = "half",
575     [__builtin_ctz(ARM_HWCAP_ARM_THUMB    )] = "thumb",
576     [__builtin_ctz(ARM_HWCAP_ARM_26BIT    )] = "26bit",
577     [__builtin_ctz(ARM_HWCAP_ARM_FAST_MULT)] = "fast_mult",
578     [__builtin_ctz(ARM_HWCAP_ARM_FPA      )] = "fpa",
579     [__builtin_ctz(ARM_HWCAP_ARM_VFP      )] = "vfp",
580     [__builtin_ctz(ARM_HWCAP_ARM_EDSP     )] = "edsp",
581     [__builtin_ctz(ARM_HWCAP_ARM_JAVA     )] = "java",
582     [__builtin_ctz(ARM_HWCAP_ARM_IWMMXT   )] = "iwmmxt",
583     [__builtin_ctz(ARM_HWCAP_ARM_CRUNCH   )] = "crunch",
584     [__builtin_ctz(ARM_HWCAP_ARM_THUMBEE  )] = "thumbee",
585     [__builtin_ctz(ARM_HWCAP_ARM_NEON     )] = "neon",
586     [__builtin_ctz(ARM_HWCAP_ARM_VFPv3    )] = "vfpv3",
587     [__builtin_ctz(ARM_HWCAP_ARM_VFPv3D16 )] = "vfpv3d16",
588     [__builtin_ctz(ARM_HWCAP_ARM_TLS      )] = "tls",
589     [__builtin_ctz(ARM_HWCAP_ARM_VFPv4    )] = "vfpv4",
590     [__builtin_ctz(ARM_HWCAP_ARM_IDIVA    )] = "idiva",
591     [__builtin_ctz(ARM_HWCAP_ARM_IDIVT    )] = "idivt",
592     [__builtin_ctz(ARM_HWCAP_ARM_VFPD32   )] = "vfpd32",
593     [__builtin_ctz(ARM_HWCAP_ARM_LPAE     )] = "lpae",
594     [__builtin_ctz(ARM_HWCAP_ARM_EVTSTRM  )] = "evtstrm",
595     [__builtin_ctz(ARM_HWCAP_ARM_FPHP     )] = "fphp",
596     [__builtin_ctz(ARM_HWCAP_ARM_ASIMDHP  )] = "asimdhp",
597     [__builtin_ctz(ARM_HWCAP_ARM_ASIMDDP  )] = "asimddp",
598     [__builtin_ctz(ARM_HWCAP_ARM_ASIMDFHM )] = "asimdfhm",
599     [__builtin_ctz(ARM_HWCAP_ARM_ASIMDBF16)] = "asimdbf16",
600     [__builtin_ctz(ARM_HWCAP_ARM_I8MM     )] = "i8mm",
601     };
602 
603     return bit < ARRAY_SIZE(hwcap_str) ? hwcap_str[bit] : NULL;
604 }
605 
606 const char *elf_hwcap2_str(uint32_t bit)
607 {
608     static const char *hwcap_str[] = {
609     [__builtin_ctz(ARM_HWCAP2_ARM_AES  )] = "aes",
610     [__builtin_ctz(ARM_HWCAP2_ARM_PMULL)] = "pmull",
611     [__builtin_ctz(ARM_HWCAP2_ARM_SHA1 )] = "sha1",
612     [__builtin_ctz(ARM_HWCAP2_ARM_SHA2 )] = "sha2",
613     [__builtin_ctz(ARM_HWCAP2_ARM_CRC32)] = "crc32",
614     [__builtin_ctz(ARM_HWCAP2_ARM_SB   )] = "sb",
615     [__builtin_ctz(ARM_HWCAP2_ARM_SSBS )] = "ssbs",
616     };
617 
618     return bit < ARRAY_SIZE(hwcap_str) ? hwcap_str[bit] : NULL;
619 }
620 
621 #undef GET_FEATURE
622 #undef GET_FEATURE_ID
623 
624 #define ELF_PLATFORM get_elf_platform()
625 
626 static const char *get_elf_platform(void)
627 {
628     CPUARMState *env = cpu_env(thread_cpu);
629 
630 #if TARGET_BIG_ENDIAN
631 # define END  "b"
632 #else
633 # define END  "l"
634 #endif
635 
636     if (arm_feature(env, ARM_FEATURE_V8)) {
637         return "v8" END;
638     } else if (arm_feature(env, ARM_FEATURE_V7)) {
639         if (arm_feature(env, ARM_FEATURE_M)) {
640             return "v7m" END;
641         } else {
642             return "v7" END;
643         }
644     } else if (arm_feature(env, ARM_FEATURE_V6)) {
645         return "v6" END;
646     } else if (arm_feature(env, ARM_FEATURE_V5)) {
647         return "v5" END;
648     } else {
649         return "v4" END;
650     }
651 
652 #undef END
653 }
654 
655 #else
656 /* 64 bit ARM definitions */
657 
658 #define ELF_ARCH        EM_AARCH64
659 #define ELF_CLASS       ELFCLASS64
660 #if TARGET_BIG_ENDIAN
661 # define ELF_PLATFORM    "aarch64_be"
662 #else
663 # define ELF_PLATFORM    "aarch64"
664 #endif
665 
666 static inline void init_thread(struct target_pt_regs *regs,
667                                struct image_info *infop)
668 {
669     abi_long stack = infop->start_stack;
670     memset(regs, 0, sizeof(*regs));
671 
672     regs->pc = infop->entry & ~0x3ULL;
673     regs->sp = stack;
674 }
675 
676 #define ELF_NREG    34
677 typedef target_elf_greg_t  target_elf_gregset_t[ELF_NREG];
678 
679 static void elf_core_copy_regs(target_elf_gregset_t *regs,
680                                const CPUARMState *env)
681 {
682     int i;
683 
684     for (i = 0; i < 32; i++) {
685         (*regs)[i] = tswapreg(env->xregs[i]);
686     }
687     (*regs)[32] = tswapreg(env->pc);
688     (*regs)[33] = tswapreg(pstate_read((CPUARMState *)env));
689 }
690 
691 #define USE_ELF_CORE_DUMP
692 #define ELF_EXEC_PAGESIZE       4096
693 
694 enum {
695     ARM_HWCAP_A64_FP            = 1 << 0,
696     ARM_HWCAP_A64_ASIMD         = 1 << 1,
697     ARM_HWCAP_A64_EVTSTRM       = 1 << 2,
698     ARM_HWCAP_A64_AES           = 1 << 3,
699     ARM_HWCAP_A64_PMULL         = 1 << 4,
700     ARM_HWCAP_A64_SHA1          = 1 << 5,
701     ARM_HWCAP_A64_SHA2          = 1 << 6,
702     ARM_HWCAP_A64_CRC32         = 1 << 7,
703     ARM_HWCAP_A64_ATOMICS       = 1 << 8,
704     ARM_HWCAP_A64_FPHP          = 1 << 9,
705     ARM_HWCAP_A64_ASIMDHP       = 1 << 10,
706     ARM_HWCAP_A64_CPUID         = 1 << 11,
707     ARM_HWCAP_A64_ASIMDRDM      = 1 << 12,
708     ARM_HWCAP_A64_JSCVT         = 1 << 13,
709     ARM_HWCAP_A64_FCMA          = 1 << 14,
710     ARM_HWCAP_A64_LRCPC         = 1 << 15,
711     ARM_HWCAP_A64_DCPOP         = 1 << 16,
712     ARM_HWCAP_A64_SHA3          = 1 << 17,
713     ARM_HWCAP_A64_SM3           = 1 << 18,
714     ARM_HWCAP_A64_SM4           = 1 << 19,
715     ARM_HWCAP_A64_ASIMDDP       = 1 << 20,
716     ARM_HWCAP_A64_SHA512        = 1 << 21,
717     ARM_HWCAP_A64_SVE           = 1 << 22,
718     ARM_HWCAP_A64_ASIMDFHM      = 1 << 23,
719     ARM_HWCAP_A64_DIT           = 1 << 24,
720     ARM_HWCAP_A64_USCAT         = 1 << 25,
721     ARM_HWCAP_A64_ILRCPC        = 1 << 26,
722     ARM_HWCAP_A64_FLAGM         = 1 << 27,
723     ARM_HWCAP_A64_SSBS          = 1 << 28,
724     ARM_HWCAP_A64_SB            = 1 << 29,
725     ARM_HWCAP_A64_PACA          = 1 << 30,
726     ARM_HWCAP_A64_PACG          = 1UL << 31,
727 
728     ARM_HWCAP2_A64_DCPODP       = 1 << 0,
729     ARM_HWCAP2_A64_SVE2         = 1 << 1,
730     ARM_HWCAP2_A64_SVEAES       = 1 << 2,
731     ARM_HWCAP2_A64_SVEPMULL     = 1 << 3,
732     ARM_HWCAP2_A64_SVEBITPERM   = 1 << 4,
733     ARM_HWCAP2_A64_SVESHA3      = 1 << 5,
734     ARM_HWCAP2_A64_SVESM4       = 1 << 6,
735     ARM_HWCAP2_A64_FLAGM2       = 1 << 7,
736     ARM_HWCAP2_A64_FRINT        = 1 << 8,
737     ARM_HWCAP2_A64_SVEI8MM      = 1 << 9,
738     ARM_HWCAP2_A64_SVEF32MM     = 1 << 10,
739     ARM_HWCAP2_A64_SVEF64MM     = 1 << 11,
740     ARM_HWCAP2_A64_SVEBF16      = 1 << 12,
741     ARM_HWCAP2_A64_I8MM         = 1 << 13,
742     ARM_HWCAP2_A64_BF16         = 1 << 14,
743     ARM_HWCAP2_A64_DGH          = 1 << 15,
744     ARM_HWCAP2_A64_RNG          = 1 << 16,
745     ARM_HWCAP2_A64_BTI          = 1 << 17,
746     ARM_HWCAP2_A64_MTE          = 1 << 18,
747     ARM_HWCAP2_A64_ECV          = 1 << 19,
748     ARM_HWCAP2_A64_AFP          = 1 << 20,
749     ARM_HWCAP2_A64_RPRES        = 1 << 21,
750     ARM_HWCAP2_A64_MTE3         = 1 << 22,
751     ARM_HWCAP2_A64_SME          = 1 << 23,
752     ARM_HWCAP2_A64_SME_I16I64   = 1 << 24,
753     ARM_HWCAP2_A64_SME_F64F64   = 1 << 25,
754     ARM_HWCAP2_A64_SME_I8I32    = 1 << 26,
755     ARM_HWCAP2_A64_SME_F16F32   = 1 << 27,
756     ARM_HWCAP2_A64_SME_B16F32   = 1 << 28,
757     ARM_HWCAP2_A64_SME_F32F32   = 1 << 29,
758     ARM_HWCAP2_A64_SME_FA64     = 1 << 30,
759     ARM_HWCAP2_A64_WFXT         = 1ULL << 31,
760     ARM_HWCAP2_A64_EBF16        = 1ULL << 32,
761     ARM_HWCAP2_A64_SVE_EBF16    = 1ULL << 33,
762     ARM_HWCAP2_A64_CSSC         = 1ULL << 34,
763     ARM_HWCAP2_A64_RPRFM        = 1ULL << 35,
764     ARM_HWCAP2_A64_SVE2P1       = 1ULL << 36,
765     ARM_HWCAP2_A64_SME2         = 1ULL << 37,
766     ARM_HWCAP2_A64_SME2P1       = 1ULL << 38,
767     ARM_HWCAP2_A64_SME_I16I32   = 1ULL << 39,
768     ARM_HWCAP2_A64_SME_BI32I32  = 1ULL << 40,
769     ARM_HWCAP2_A64_SME_B16B16   = 1ULL << 41,
770     ARM_HWCAP2_A64_SME_F16F16   = 1ULL << 42,
771     ARM_HWCAP2_A64_MOPS         = 1ULL << 43,
772     ARM_HWCAP2_A64_HBC          = 1ULL << 44,
773 };
774 
775 #define ELF_HWCAP   get_elf_hwcap()
776 #define ELF_HWCAP2  get_elf_hwcap2()
777 
778 #define GET_FEATURE_ID(feat, hwcap) \
779     do { if (cpu_isar_feature(feat, cpu)) { hwcaps |= hwcap; } } while (0)
780 
781 uint32_t get_elf_hwcap(void)
782 {
783     ARMCPU *cpu = ARM_CPU(thread_cpu);
784     uint32_t hwcaps = 0;
785 
786     hwcaps |= ARM_HWCAP_A64_FP;
787     hwcaps |= ARM_HWCAP_A64_ASIMD;
788     hwcaps |= ARM_HWCAP_A64_CPUID;
789 
790     /* probe for the extra features */
791 
792     GET_FEATURE_ID(aa64_aes, ARM_HWCAP_A64_AES);
793     GET_FEATURE_ID(aa64_pmull, ARM_HWCAP_A64_PMULL);
794     GET_FEATURE_ID(aa64_sha1, ARM_HWCAP_A64_SHA1);
795     GET_FEATURE_ID(aa64_sha256, ARM_HWCAP_A64_SHA2);
796     GET_FEATURE_ID(aa64_sha512, ARM_HWCAP_A64_SHA512);
797     GET_FEATURE_ID(aa64_crc32, ARM_HWCAP_A64_CRC32);
798     GET_FEATURE_ID(aa64_sha3, ARM_HWCAP_A64_SHA3);
799     GET_FEATURE_ID(aa64_sm3, ARM_HWCAP_A64_SM3);
800     GET_FEATURE_ID(aa64_sm4, ARM_HWCAP_A64_SM4);
801     GET_FEATURE_ID(aa64_fp16, ARM_HWCAP_A64_FPHP | ARM_HWCAP_A64_ASIMDHP);
802     GET_FEATURE_ID(aa64_atomics, ARM_HWCAP_A64_ATOMICS);
803     GET_FEATURE_ID(aa64_rdm, ARM_HWCAP_A64_ASIMDRDM);
804     GET_FEATURE_ID(aa64_dp, ARM_HWCAP_A64_ASIMDDP);
805     GET_FEATURE_ID(aa64_fcma, ARM_HWCAP_A64_FCMA);
806     GET_FEATURE_ID(aa64_sve, ARM_HWCAP_A64_SVE);
807     GET_FEATURE_ID(aa64_pauth, ARM_HWCAP_A64_PACA | ARM_HWCAP_A64_PACG);
808     GET_FEATURE_ID(aa64_fhm, ARM_HWCAP_A64_ASIMDFHM);
809     GET_FEATURE_ID(aa64_jscvt, ARM_HWCAP_A64_JSCVT);
810     GET_FEATURE_ID(aa64_sb, ARM_HWCAP_A64_SB);
811     GET_FEATURE_ID(aa64_condm_4, ARM_HWCAP_A64_FLAGM);
812     GET_FEATURE_ID(aa64_dcpop, ARM_HWCAP_A64_DCPOP);
813     GET_FEATURE_ID(aa64_rcpc_8_3, ARM_HWCAP_A64_LRCPC);
814     GET_FEATURE_ID(aa64_rcpc_8_4, ARM_HWCAP_A64_ILRCPC);
815 
816     return hwcaps;
817 }
818 
819 uint32_t get_elf_hwcap2(void)
820 {
821     ARMCPU *cpu = ARM_CPU(thread_cpu);
822     uint32_t hwcaps = 0;
823 
824     GET_FEATURE_ID(aa64_dcpodp, ARM_HWCAP2_A64_DCPODP);
825     GET_FEATURE_ID(aa64_sve2, ARM_HWCAP2_A64_SVE2);
826     GET_FEATURE_ID(aa64_sve2_aes, ARM_HWCAP2_A64_SVEAES);
827     GET_FEATURE_ID(aa64_sve2_pmull128, ARM_HWCAP2_A64_SVEPMULL);
828     GET_FEATURE_ID(aa64_sve2_bitperm, ARM_HWCAP2_A64_SVEBITPERM);
829     GET_FEATURE_ID(aa64_sve2_sha3, ARM_HWCAP2_A64_SVESHA3);
830     GET_FEATURE_ID(aa64_sve2_sm4, ARM_HWCAP2_A64_SVESM4);
831     GET_FEATURE_ID(aa64_condm_5, ARM_HWCAP2_A64_FLAGM2);
832     GET_FEATURE_ID(aa64_frint, ARM_HWCAP2_A64_FRINT);
833     GET_FEATURE_ID(aa64_sve_i8mm, ARM_HWCAP2_A64_SVEI8MM);
834     GET_FEATURE_ID(aa64_sve_f32mm, ARM_HWCAP2_A64_SVEF32MM);
835     GET_FEATURE_ID(aa64_sve_f64mm, ARM_HWCAP2_A64_SVEF64MM);
836     GET_FEATURE_ID(aa64_sve_bf16, ARM_HWCAP2_A64_SVEBF16);
837     GET_FEATURE_ID(aa64_i8mm, ARM_HWCAP2_A64_I8MM);
838     GET_FEATURE_ID(aa64_bf16, ARM_HWCAP2_A64_BF16);
839     GET_FEATURE_ID(aa64_rndr, ARM_HWCAP2_A64_RNG);
840     GET_FEATURE_ID(aa64_bti, ARM_HWCAP2_A64_BTI);
841     GET_FEATURE_ID(aa64_mte, ARM_HWCAP2_A64_MTE);
842     GET_FEATURE_ID(aa64_sme, (ARM_HWCAP2_A64_SME |
843                               ARM_HWCAP2_A64_SME_F32F32 |
844                               ARM_HWCAP2_A64_SME_B16F32 |
845                               ARM_HWCAP2_A64_SME_F16F32 |
846                               ARM_HWCAP2_A64_SME_I8I32));
847     GET_FEATURE_ID(aa64_sme_f64f64, ARM_HWCAP2_A64_SME_F64F64);
848     GET_FEATURE_ID(aa64_sme_i16i64, ARM_HWCAP2_A64_SME_I16I64);
849     GET_FEATURE_ID(aa64_sme_fa64, ARM_HWCAP2_A64_SME_FA64);
850     GET_FEATURE_ID(aa64_hbc, ARM_HWCAP2_A64_HBC);
851     GET_FEATURE_ID(aa64_mops, ARM_HWCAP2_A64_MOPS);
852 
853     return hwcaps;
854 }
855 
856 const char *elf_hwcap_str(uint32_t bit)
857 {
858     static const char *hwcap_str[] = {
859     [__builtin_ctz(ARM_HWCAP_A64_FP      )] = "fp",
860     [__builtin_ctz(ARM_HWCAP_A64_ASIMD   )] = "asimd",
861     [__builtin_ctz(ARM_HWCAP_A64_EVTSTRM )] = "evtstrm",
862     [__builtin_ctz(ARM_HWCAP_A64_AES     )] = "aes",
863     [__builtin_ctz(ARM_HWCAP_A64_PMULL   )] = "pmull",
864     [__builtin_ctz(ARM_HWCAP_A64_SHA1    )] = "sha1",
865     [__builtin_ctz(ARM_HWCAP_A64_SHA2    )] = "sha2",
866     [__builtin_ctz(ARM_HWCAP_A64_CRC32   )] = "crc32",
867     [__builtin_ctz(ARM_HWCAP_A64_ATOMICS )] = "atomics",
868     [__builtin_ctz(ARM_HWCAP_A64_FPHP    )] = "fphp",
869     [__builtin_ctz(ARM_HWCAP_A64_ASIMDHP )] = "asimdhp",
870     [__builtin_ctz(ARM_HWCAP_A64_CPUID   )] = "cpuid",
871     [__builtin_ctz(ARM_HWCAP_A64_ASIMDRDM)] = "asimdrdm",
872     [__builtin_ctz(ARM_HWCAP_A64_JSCVT   )] = "jscvt",
873     [__builtin_ctz(ARM_HWCAP_A64_FCMA    )] = "fcma",
874     [__builtin_ctz(ARM_HWCAP_A64_LRCPC   )] = "lrcpc",
875     [__builtin_ctz(ARM_HWCAP_A64_DCPOP   )] = "dcpop",
876     [__builtin_ctz(ARM_HWCAP_A64_SHA3    )] = "sha3",
877     [__builtin_ctz(ARM_HWCAP_A64_SM3     )] = "sm3",
878     [__builtin_ctz(ARM_HWCAP_A64_SM4     )] = "sm4",
879     [__builtin_ctz(ARM_HWCAP_A64_ASIMDDP )] = "asimddp",
880     [__builtin_ctz(ARM_HWCAP_A64_SHA512  )] = "sha512",
881     [__builtin_ctz(ARM_HWCAP_A64_SVE     )] = "sve",
882     [__builtin_ctz(ARM_HWCAP_A64_ASIMDFHM)] = "asimdfhm",
883     [__builtin_ctz(ARM_HWCAP_A64_DIT     )] = "dit",
884     [__builtin_ctz(ARM_HWCAP_A64_USCAT   )] = "uscat",
885     [__builtin_ctz(ARM_HWCAP_A64_ILRCPC  )] = "ilrcpc",
886     [__builtin_ctz(ARM_HWCAP_A64_FLAGM   )] = "flagm",
887     [__builtin_ctz(ARM_HWCAP_A64_SSBS    )] = "ssbs",
888     [__builtin_ctz(ARM_HWCAP_A64_SB      )] = "sb",
889     [__builtin_ctz(ARM_HWCAP_A64_PACA    )] = "paca",
890     [__builtin_ctz(ARM_HWCAP_A64_PACG    )] = "pacg",
891     };
892 
893     return bit < ARRAY_SIZE(hwcap_str) ? hwcap_str[bit] : NULL;
894 }
895 
896 const char *elf_hwcap2_str(uint32_t bit)
897 {
898     static const char *hwcap_str[] = {
899     [__builtin_ctz(ARM_HWCAP2_A64_DCPODP       )] = "dcpodp",
900     [__builtin_ctz(ARM_HWCAP2_A64_SVE2         )] = "sve2",
901     [__builtin_ctz(ARM_HWCAP2_A64_SVEAES       )] = "sveaes",
902     [__builtin_ctz(ARM_HWCAP2_A64_SVEPMULL     )] = "svepmull",
903     [__builtin_ctz(ARM_HWCAP2_A64_SVEBITPERM   )] = "svebitperm",
904     [__builtin_ctz(ARM_HWCAP2_A64_SVESHA3      )] = "svesha3",
905     [__builtin_ctz(ARM_HWCAP2_A64_SVESM4       )] = "svesm4",
906     [__builtin_ctz(ARM_HWCAP2_A64_FLAGM2       )] = "flagm2",
907     [__builtin_ctz(ARM_HWCAP2_A64_FRINT        )] = "frint",
908     [__builtin_ctz(ARM_HWCAP2_A64_SVEI8MM      )] = "svei8mm",
909     [__builtin_ctz(ARM_HWCAP2_A64_SVEF32MM     )] = "svef32mm",
910     [__builtin_ctz(ARM_HWCAP2_A64_SVEF64MM     )] = "svef64mm",
911     [__builtin_ctz(ARM_HWCAP2_A64_SVEBF16      )] = "svebf16",
912     [__builtin_ctz(ARM_HWCAP2_A64_I8MM         )] = "i8mm",
913     [__builtin_ctz(ARM_HWCAP2_A64_BF16         )] = "bf16",
914     [__builtin_ctz(ARM_HWCAP2_A64_DGH          )] = "dgh",
915     [__builtin_ctz(ARM_HWCAP2_A64_RNG          )] = "rng",
916     [__builtin_ctz(ARM_HWCAP2_A64_BTI          )] = "bti",
917     [__builtin_ctz(ARM_HWCAP2_A64_MTE          )] = "mte",
918     [__builtin_ctz(ARM_HWCAP2_A64_ECV          )] = "ecv",
919     [__builtin_ctz(ARM_HWCAP2_A64_AFP          )] = "afp",
920     [__builtin_ctz(ARM_HWCAP2_A64_RPRES        )] = "rpres",
921     [__builtin_ctz(ARM_HWCAP2_A64_MTE3         )] = "mte3",
922     [__builtin_ctz(ARM_HWCAP2_A64_SME          )] = "sme",
923     [__builtin_ctz(ARM_HWCAP2_A64_SME_I16I64   )] = "smei16i64",
924     [__builtin_ctz(ARM_HWCAP2_A64_SME_F64F64   )] = "smef64f64",
925     [__builtin_ctz(ARM_HWCAP2_A64_SME_I8I32    )] = "smei8i32",
926     [__builtin_ctz(ARM_HWCAP2_A64_SME_F16F32   )] = "smef16f32",
927     [__builtin_ctz(ARM_HWCAP2_A64_SME_B16F32   )] = "smeb16f32",
928     [__builtin_ctz(ARM_HWCAP2_A64_SME_F32F32   )] = "smef32f32",
929     [__builtin_ctz(ARM_HWCAP2_A64_SME_FA64     )] = "smefa64",
930     [__builtin_ctz(ARM_HWCAP2_A64_WFXT         )] = "wfxt",
931     [__builtin_ctzll(ARM_HWCAP2_A64_EBF16      )] = "ebf16",
932     [__builtin_ctzll(ARM_HWCAP2_A64_SVE_EBF16  )] = "sveebf16",
933     [__builtin_ctzll(ARM_HWCAP2_A64_CSSC       )] = "cssc",
934     [__builtin_ctzll(ARM_HWCAP2_A64_RPRFM      )] = "rprfm",
935     [__builtin_ctzll(ARM_HWCAP2_A64_SVE2P1     )] = "sve2p1",
936     [__builtin_ctzll(ARM_HWCAP2_A64_SME2       )] = "sme2",
937     [__builtin_ctzll(ARM_HWCAP2_A64_SME2P1     )] = "sme2p1",
938     [__builtin_ctzll(ARM_HWCAP2_A64_SME_I16I32 )] = "smei16i32",
939     [__builtin_ctzll(ARM_HWCAP2_A64_SME_BI32I32)] = "smebi32i32",
940     [__builtin_ctzll(ARM_HWCAP2_A64_SME_B16B16 )] = "smeb16b16",
941     [__builtin_ctzll(ARM_HWCAP2_A64_SME_F16F16 )] = "smef16f16",
942     [__builtin_ctzll(ARM_HWCAP2_A64_MOPS       )] = "mops",
943     [__builtin_ctzll(ARM_HWCAP2_A64_HBC        )] = "hbc",
944     };
945 
946     return bit < ARRAY_SIZE(hwcap_str) ? hwcap_str[bit] : NULL;
947 }
948 
949 #undef GET_FEATURE_ID
950 
951 #endif /* not TARGET_AARCH64 */
952 
953 #if TARGET_BIG_ENDIAN
954 # define VDSO_HEADER  "vdso-be.c.inc"
955 #else
956 # define VDSO_HEADER  "vdso-le.c.inc"
957 #endif
958 
959 #endif /* TARGET_ARM */
960 
961 #ifdef TARGET_SPARC
962 #ifdef TARGET_SPARC64
963 
964 #define ELF_HWCAP  (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \
965                     | HWCAP_SPARC_MULDIV | HWCAP_SPARC_V9)
966 #ifndef TARGET_ABI32
967 #define elf_check_arch(x) ( (x) == EM_SPARCV9 || (x) == EM_SPARC32PLUS )
968 #else
969 #define elf_check_arch(x) ( (x) == EM_SPARC32PLUS || (x) == EM_SPARC )
970 #endif
971 
972 #define ELF_CLASS   ELFCLASS64
973 #define ELF_ARCH    EM_SPARCV9
974 #else
975 #define ELF_HWCAP  (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \
976                     | HWCAP_SPARC_MULDIV)
977 #define ELF_CLASS   ELFCLASS32
978 #define ELF_ARCH    EM_SPARC
979 #endif /* TARGET_SPARC64 */
980 
981 static inline void init_thread(struct target_pt_regs *regs,
982                                struct image_info *infop)
983 {
984     /* Note that target_cpu_copy_regs does not read psr/tstate. */
985     regs->pc = infop->entry;
986     regs->npc = regs->pc + 4;
987     regs->y = 0;
988     regs->u_regs[14] = (infop->start_stack - 16 * sizeof(abi_ulong)
989                         - TARGET_STACK_BIAS);
990 }
991 #endif /* TARGET_SPARC */
992 
993 #ifdef TARGET_PPC
994 
995 #define ELF_MACHINE    PPC_ELF_MACHINE
996 
997 #if defined(TARGET_PPC64)
998 
999 #define elf_check_arch(x) ( (x) == EM_PPC64 )
1000 
1001 #define ELF_CLASS       ELFCLASS64
1002 
1003 #else
1004 
1005 #define ELF_CLASS       ELFCLASS32
1006 #define EXSTACK_DEFAULT true
1007 
1008 #endif
1009 
1010 #define ELF_ARCH        EM_PPC
1011 
1012 /* Feature masks for the Aux Vector Hardware Capabilities (AT_HWCAP).
1013    See arch/powerpc/include/asm/cputable.h.  */
1014 enum {
1015     QEMU_PPC_FEATURE_32 = 0x80000000,
1016     QEMU_PPC_FEATURE_64 = 0x40000000,
1017     QEMU_PPC_FEATURE_601_INSTR = 0x20000000,
1018     QEMU_PPC_FEATURE_HAS_ALTIVEC = 0x10000000,
1019     QEMU_PPC_FEATURE_HAS_FPU = 0x08000000,
1020     QEMU_PPC_FEATURE_HAS_MMU = 0x04000000,
1021     QEMU_PPC_FEATURE_HAS_4xxMAC = 0x02000000,
1022     QEMU_PPC_FEATURE_UNIFIED_CACHE = 0x01000000,
1023     QEMU_PPC_FEATURE_HAS_SPE = 0x00800000,
1024     QEMU_PPC_FEATURE_HAS_EFP_SINGLE = 0x00400000,
1025     QEMU_PPC_FEATURE_HAS_EFP_DOUBLE = 0x00200000,
1026     QEMU_PPC_FEATURE_NO_TB = 0x00100000,
1027     QEMU_PPC_FEATURE_POWER4 = 0x00080000,
1028     QEMU_PPC_FEATURE_POWER5 = 0x00040000,
1029     QEMU_PPC_FEATURE_POWER5_PLUS = 0x00020000,
1030     QEMU_PPC_FEATURE_CELL = 0x00010000,
1031     QEMU_PPC_FEATURE_BOOKE = 0x00008000,
1032     QEMU_PPC_FEATURE_SMT = 0x00004000,
1033     QEMU_PPC_FEATURE_ICACHE_SNOOP = 0x00002000,
1034     QEMU_PPC_FEATURE_ARCH_2_05 = 0x00001000,
1035     QEMU_PPC_FEATURE_PA6T = 0x00000800,
1036     QEMU_PPC_FEATURE_HAS_DFP = 0x00000400,
1037     QEMU_PPC_FEATURE_POWER6_EXT = 0x00000200,
1038     QEMU_PPC_FEATURE_ARCH_2_06 = 0x00000100,
1039     QEMU_PPC_FEATURE_HAS_VSX = 0x00000080,
1040     QEMU_PPC_FEATURE_PSERIES_PERFMON_COMPAT = 0x00000040,
1041 
1042     QEMU_PPC_FEATURE_TRUE_LE = 0x00000002,
1043     QEMU_PPC_FEATURE_PPC_LE = 0x00000001,
1044 
1045     /* Feature definitions in AT_HWCAP2.  */
1046     QEMU_PPC_FEATURE2_ARCH_2_07 = 0x80000000, /* ISA 2.07 */
1047     QEMU_PPC_FEATURE2_HAS_HTM = 0x40000000, /* Hardware Transactional Memory */
1048     QEMU_PPC_FEATURE2_HAS_DSCR = 0x20000000, /* Data Stream Control Register */
1049     QEMU_PPC_FEATURE2_HAS_EBB = 0x10000000, /* Event Base Branching */
1050     QEMU_PPC_FEATURE2_HAS_ISEL = 0x08000000, /* Integer Select */
1051     QEMU_PPC_FEATURE2_HAS_TAR = 0x04000000, /* Target Address Register */
1052     QEMU_PPC_FEATURE2_VEC_CRYPTO = 0x02000000,
1053     QEMU_PPC_FEATURE2_HTM_NOSC = 0x01000000,
1054     QEMU_PPC_FEATURE2_ARCH_3_00 = 0x00800000, /* ISA 3.00 */
1055     QEMU_PPC_FEATURE2_HAS_IEEE128 = 0x00400000, /* VSX IEEE Bin Float 128-bit */
1056     QEMU_PPC_FEATURE2_DARN = 0x00200000, /* darn random number insn */
1057     QEMU_PPC_FEATURE2_SCV = 0x00100000, /* scv syscall */
1058     QEMU_PPC_FEATURE2_HTM_NO_SUSPEND = 0x00080000, /* TM w/o suspended state */
1059     QEMU_PPC_FEATURE2_ARCH_3_1 = 0x00040000, /* ISA 3.1 */
1060     QEMU_PPC_FEATURE2_MMA = 0x00020000, /* Matrix-Multiply Assist */
1061 };
1062 
1063 #define ELF_HWCAP get_elf_hwcap()
1064 
1065 static uint32_t get_elf_hwcap(void)
1066 {
1067     PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);
1068     uint32_t features = 0;
1069 
1070     /* We don't have to be terribly complete here; the high points are
1071        Altivec/FP/SPE support.  Anything else is just a bonus.  */
1072 #define GET_FEATURE(flag, feature)                                      \
1073     do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0)
1074 #define GET_FEATURE2(flags, feature) \
1075     do { \
1076         if ((cpu->env.insns_flags2 & flags) == flags) { \
1077             features |= feature; \
1078         } \
1079     } while (0)
1080     GET_FEATURE(PPC_64B, QEMU_PPC_FEATURE_64);
1081     GET_FEATURE(PPC_FLOAT, QEMU_PPC_FEATURE_HAS_FPU);
1082     GET_FEATURE(PPC_ALTIVEC, QEMU_PPC_FEATURE_HAS_ALTIVEC);
1083     GET_FEATURE(PPC_SPE, QEMU_PPC_FEATURE_HAS_SPE);
1084     GET_FEATURE(PPC_SPE_SINGLE, QEMU_PPC_FEATURE_HAS_EFP_SINGLE);
1085     GET_FEATURE(PPC_SPE_DOUBLE, QEMU_PPC_FEATURE_HAS_EFP_DOUBLE);
1086     GET_FEATURE(PPC_BOOKE, QEMU_PPC_FEATURE_BOOKE);
1087     GET_FEATURE(PPC_405_MAC, QEMU_PPC_FEATURE_HAS_4xxMAC);
1088     GET_FEATURE2(PPC2_DFP, QEMU_PPC_FEATURE_HAS_DFP);
1089     GET_FEATURE2(PPC2_VSX, QEMU_PPC_FEATURE_HAS_VSX);
1090     GET_FEATURE2((PPC2_PERM_ISA206 | PPC2_DIVE_ISA206 | PPC2_ATOMIC_ISA206 |
1091                   PPC2_FP_CVT_ISA206 | PPC2_FP_TST_ISA206),
1092                   QEMU_PPC_FEATURE_ARCH_2_06);
1093 #undef GET_FEATURE
1094 #undef GET_FEATURE2
1095 
1096     return features;
1097 }
1098 
1099 #define ELF_HWCAP2 get_elf_hwcap2()
1100 
1101 static uint32_t get_elf_hwcap2(void)
1102 {
1103     PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);
1104     uint32_t features = 0;
1105 
1106 #define GET_FEATURE(flag, feature)                                      \
1107     do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0)
1108 #define GET_FEATURE2(flag, feature)                                      \
1109     do { if (cpu->env.insns_flags2 & flag) { features |= feature; } } while (0)
1110 
1111     GET_FEATURE(PPC_ISEL, QEMU_PPC_FEATURE2_HAS_ISEL);
1112     GET_FEATURE2(PPC2_BCTAR_ISA207, QEMU_PPC_FEATURE2_HAS_TAR);
1113     GET_FEATURE2((PPC2_BCTAR_ISA207 | PPC2_LSQ_ISA207 | PPC2_ALTIVEC_207 |
1114                   PPC2_ISA207S), QEMU_PPC_FEATURE2_ARCH_2_07 |
1115                   QEMU_PPC_FEATURE2_VEC_CRYPTO);
1116     GET_FEATURE2(PPC2_ISA300, QEMU_PPC_FEATURE2_ARCH_3_00 |
1117                  QEMU_PPC_FEATURE2_DARN | QEMU_PPC_FEATURE2_HAS_IEEE128);
1118     GET_FEATURE2(PPC2_ISA310, QEMU_PPC_FEATURE2_ARCH_3_1 |
1119                  QEMU_PPC_FEATURE2_MMA);
1120 
1121 #undef GET_FEATURE
1122 #undef GET_FEATURE2
1123 
1124     return features;
1125 }
1126 
1127 /*
1128  * The requirements here are:
1129  * - keep the final alignment of sp (sp & 0xf)
1130  * - make sure the 32-bit value at the first 16 byte aligned position of
1131  *   AUXV is greater than 16 for glibc compatibility.
1132  *   AT_IGNOREPPC is used for that.
1133  * - for compatibility with glibc ARCH_DLINFO must always be defined on PPC,
1134  *   even if DLINFO_ARCH_ITEMS goes to zero or is undefined.
1135  */
1136 #define DLINFO_ARCH_ITEMS       5
1137 #define ARCH_DLINFO                                     \
1138     do {                                                \
1139         PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);              \
1140         /*                                              \
1141          * Handle glibc compatibility: these magic entries must \
1142          * be at the lowest addresses in the final auxv.        \
1143          */                                             \
1144         NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC);        \
1145         NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC);        \
1146         NEW_AUX_ENT(AT_DCACHEBSIZE, cpu->env.dcache_line_size); \
1147         NEW_AUX_ENT(AT_ICACHEBSIZE, cpu->env.icache_line_size); \
1148         NEW_AUX_ENT(AT_UCACHEBSIZE, 0);                 \
1149     } while (0)
1150 
1151 static inline void init_thread(struct target_pt_regs *_regs, struct image_info *infop)
1152 {
1153     _regs->gpr[1] = infop->start_stack;
1154 #if defined(TARGET_PPC64)
1155     if (get_ppc64_abi(infop) < 2) {
1156         uint64_t val;
1157         get_user_u64(val, infop->entry + 8);
1158         _regs->gpr[2] = val + infop->load_bias;
1159         get_user_u64(val, infop->entry);
1160         infop->entry = val + infop->load_bias;
1161     } else {
1162         _regs->gpr[12] = infop->entry;  /* r12 set to global entry address */
1163     }
1164 #endif
1165     _regs->nip = infop->entry;
1166 }
1167 
1168 /* See linux kernel: arch/powerpc/include/asm/elf.h.  */
1169 #define ELF_NREG 48
1170 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1171 
1172 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUPPCState *env)
1173 {
1174     int i;
1175     target_ulong ccr = 0;
1176 
1177     for (i = 0; i < ARRAY_SIZE(env->gpr); i++) {
1178         (*regs)[i] = tswapreg(env->gpr[i]);
1179     }
1180 
1181     (*regs)[32] = tswapreg(env->nip);
1182     (*regs)[33] = tswapreg(env->msr);
1183     (*regs)[35] = tswapreg(env->ctr);
1184     (*regs)[36] = tswapreg(env->lr);
1185     (*regs)[37] = tswapreg(cpu_read_xer(env));
1186 
1187     ccr = ppc_get_cr(env);
1188     (*regs)[38] = tswapreg(ccr);
1189 }
1190 
1191 #define USE_ELF_CORE_DUMP
1192 #define ELF_EXEC_PAGESIZE       4096
1193 
1194 #ifndef TARGET_PPC64
1195 # define VDSO_HEADER  "vdso-32.c.inc"
1196 #elif TARGET_BIG_ENDIAN
1197 # define VDSO_HEADER  "vdso-64.c.inc"
1198 #else
1199 # define VDSO_HEADER  "vdso-64le.c.inc"
1200 #endif
1201 
1202 #endif
1203 
1204 #ifdef TARGET_LOONGARCH64
1205 
1206 #define ELF_CLASS   ELFCLASS64
1207 #define ELF_ARCH    EM_LOONGARCH
1208 #define EXSTACK_DEFAULT true
1209 
1210 #define elf_check_arch(x) ((x) == EM_LOONGARCH)
1211 
1212 #define VDSO_HEADER "vdso.c.inc"
1213 
1214 static inline void init_thread(struct target_pt_regs *regs,
1215                                struct image_info *infop)
1216 {
1217     /*Set crmd PG,DA = 1,0 */
1218     regs->csr.crmd = 2 << 3;
1219     regs->csr.era = infop->entry;
1220     regs->regs[3] = infop->start_stack;
1221 }
1222 
1223 /* See linux kernel: arch/loongarch/include/asm/elf.h */
1224 #define ELF_NREG 45
1225 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1226 
1227 enum {
1228     TARGET_EF_R0 = 0,
1229     TARGET_EF_CSR_ERA = TARGET_EF_R0 + 33,
1230     TARGET_EF_CSR_BADV = TARGET_EF_R0 + 34,
1231 };
1232 
1233 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1234                                const CPULoongArchState *env)
1235 {
1236     int i;
1237 
1238     (*regs)[TARGET_EF_R0] = 0;
1239 
1240     for (i = 1; i < ARRAY_SIZE(env->gpr); i++) {
1241         (*regs)[TARGET_EF_R0 + i] = tswapreg(env->gpr[i]);
1242     }
1243 
1244     (*regs)[TARGET_EF_CSR_ERA] = tswapreg(env->pc);
1245     (*regs)[TARGET_EF_CSR_BADV] = tswapreg(env->CSR_BADV);
1246 }
1247 
1248 #define USE_ELF_CORE_DUMP
1249 #define ELF_EXEC_PAGESIZE        4096
1250 
1251 #define ELF_HWCAP get_elf_hwcap()
1252 
1253 /* See arch/loongarch/include/uapi/asm/hwcap.h */
1254 enum {
1255     HWCAP_LOONGARCH_CPUCFG   = (1 << 0),
1256     HWCAP_LOONGARCH_LAM      = (1 << 1),
1257     HWCAP_LOONGARCH_UAL      = (1 << 2),
1258     HWCAP_LOONGARCH_FPU      = (1 << 3),
1259     HWCAP_LOONGARCH_LSX      = (1 << 4),
1260     HWCAP_LOONGARCH_LASX     = (1 << 5),
1261     HWCAP_LOONGARCH_CRC32    = (1 << 6),
1262     HWCAP_LOONGARCH_COMPLEX  = (1 << 7),
1263     HWCAP_LOONGARCH_CRYPTO   = (1 << 8),
1264     HWCAP_LOONGARCH_LVZ      = (1 << 9),
1265     HWCAP_LOONGARCH_LBT_X86  = (1 << 10),
1266     HWCAP_LOONGARCH_LBT_ARM  = (1 << 11),
1267     HWCAP_LOONGARCH_LBT_MIPS = (1 << 12),
1268 };
1269 
1270 static uint32_t get_elf_hwcap(void)
1271 {
1272     LoongArchCPU *cpu = LOONGARCH_CPU(thread_cpu);
1273     uint32_t hwcaps = 0;
1274 
1275     hwcaps |= HWCAP_LOONGARCH_CRC32;
1276 
1277     if (FIELD_EX32(cpu->env.cpucfg[1], CPUCFG1, UAL)) {
1278         hwcaps |= HWCAP_LOONGARCH_UAL;
1279     }
1280 
1281     if (FIELD_EX32(cpu->env.cpucfg[2], CPUCFG2, FP)) {
1282         hwcaps |= HWCAP_LOONGARCH_FPU;
1283     }
1284 
1285     if (FIELD_EX32(cpu->env.cpucfg[2], CPUCFG2, LAM)) {
1286         hwcaps |= HWCAP_LOONGARCH_LAM;
1287     }
1288 
1289     if (FIELD_EX32(cpu->env.cpucfg[2], CPUCFG2, LSX)) {
1290         hwcaps |= HWCAP_LOONGARCH_LSX;
1291     }
1292 
1293     if (FIELD_EX32(cpu->env.cpucfg[2], CPUCFG2, LASX)) {
1294         hwcaps |= HWCAP_LOONGARCH_LASX;
1295     }
1296 
1297     return hwcaps;
1298 }
1299 
1300 #define ELF_PLATFORM "loongarch"
1301 
1302 #endif /* TARGET_LOONGARCH64 */
1303 
1304 #ifdef TARGET_MIPS
1305 
1306 #ifdef TARGET_MIPS64
1307 #define ELF_CLASS   ELFCLASS64
1308 #else
1309 #define ELF_CLASS   ELFCLASS32
1310 #endif
1311 #define ELF_ARCH    EM_MIPS
1312 #define EXSTACK_DEFAULT true
1313 
1314 #ifdef TARGET_ABI_MIPSN32
1315 #define elf_check_abi(x) ((x) & EF_MIPS_ABI2)
1316 #else
1317 #define elf_check_abi(x) (!((x) & EF_MIPS_ABI2))
1318 #endif
1319 
1320 #define ELF_BASE_PLATFORM get_elf_base_platform()
1321 
1322 #define MATCH_PLATFORM_INSN(_flags, _base_platform)      \
1323     do { if ((cpu->env.insn_flags & (_flags)) == _flags) \
1324     { return _base_platform; } } while (0)
1325 
1326 static const char *get_elf_base_platform(void)
1327 {
1328     MIPSCPU *cpu = MIPS_CPU(thread_cpu);
1329 
1330     /* 64 bit ISAs goes first */
1331     MATCH_PLATFORM_INSN(CPU_MIPS64R6, "mips64r6");
1332     MATCH_PLATFORM_INSN(CPU_MIPS64R5, "mips64r5");
1333     MATCH_PLATFORM_INSN(CPU_MIPS64R2, "mips64r2");
1334     MATCH_PLATFORM_INSN(CPU_MIPS64R1, "mips64");
1335     MATCH_PLATFORM_INSN(CPU_MIPS5, "mips5");
1336     MATCH_PLATFORM_INSN(CPU_MIPS4, "mips4");
1337     MATCH_PLATFORM_INSN(CPU_MIPS3, "mips3");
1338 
1339     /* 32 bit ISAs */
1340     MATCH_PLATFORM_INSN(CPU_MIPS32R6, "mips32r6");
1341     MATCH_PLATFORM_INSN(CPU_MIPS32R5, "mips32r5");
1342     MATCH_PLATFORM_INSN(CPU_MIPS32R2, "mips32r2");
1343     MATCH_PLATFORM_INSN(CPU_MIPS32R1, "mips32");
1344     MATCH_PLATFORM_INSN(CPU_MIPS2, "mips2");
1345 
1346     /* Fallback */
1347     return "mips";
1348 }
1349 #undef MATCH_PLATFORM_INSN
1350 
1351 static inline void init_thread(struct target_pt_regs *regs,
1352                                struct image_info *infop)
1353 {
1354     regs->cp0_status = 2 << CP0St_KSU;
1355     regs->cp0_epc = infop->entry;
1356     regs->regs[29] = infop->start_stack;
1357 }
1358 
1359 /* See linux kernel: arch/mips/include/asm/elf.h.  */
1360 #define ELF_NREG 45
1361 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1362 
1363 /* See linux kernel: arch/mips/include/asm/reg.h.  */
1364 enum {
1365 #ifdef TARGET_MIPS64
1366     TARGET_EF_R0 = 0,
1367 #else
1368     TARGET_EF_R0 = 6,
1369 #endif
1370     TARGET_EF_R26 = TARGET_EF_R0 + 26,
1371     TARGET_EF_R27 = TARGET_EF_R0 + 27,
1372     TARGET_EF_LO = TARGET_EF_R0 + 32,
1373     TARGET_EF_HI = TARGET_EF_R0 + 33,
1374     TARGET_EF_CP0_EPC = TARGET_EF_R0 + 34,
1375     TARGET_EF_CP0_BADVADDR = TARGET_EF_R0 + 35,
1376     TARGET_EF_CP0_STATUS = TARGET_EF_R0 + 36,
1377     TARGET_EF_CP0_CAUSE = TARGET_EF_R0 + 37
1378 };
1379 
1380 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs.  */
1381 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMIPSState *env)
1382 {
1383     int i;
1384 
1385     for (i = 0; i < TARGET_EF_R0; i++) {
1386         (*regs)[i] = 0;
1387     }
1388     (*regs)[TARGET_EF_R0] = 0;
1389 
1390     for (i = 1; i < ARRAY_SIZE(env->active_tc.gpr); i++) {
1391         (*regs)[TARGET_EF_R0 + i] = tswapreg(env->active_tc.gpr[i]);
1392     }
1393 
1394     (*regs)[TARGET_EF_R26] = 0;
1395     (*regs)[TARGET_EF_R27] = 0;
1396     (*regs)[TARGET_EF_LO] = tswapreg(env->active_tc.LO[0]);
1397     (*regs)[TARGET_EF_HI] = tswapreg(env->active_tc.HI[0]);
1398     (*regs)[TARGET_EF_CP0_EPC] = tswapreg(env->active_tc.PC);
1399     (*regs)[TARGET_EF_CP0_BADVADDR] = tswapreg(env->CP0_BadVAddr);
1400     (*regs)[TARGET_EF_CP0_STATUS] = tswapreg(env->CP0_Status);
1401     (*regs)[TARGET_EF_CP0_CAUSE] = tswapreg(env->CP0_Cause);
1402 }
1403 
1404 #define USE_ELF_CORE_DUMP
1405 #define ELF_EXEC_PAGESIZE        4096
1406 
1407 /* See arch/mips/include/uapi/asm/hwcap.h.  */
1408 enum {
1409     HWCAP_MIPS_R6           = (1 << 0),
1410     HWCAP_MIPS_MSA          = (1 << 1),
1411     HWCAP_MIPS_CRC32        = (1 << 2),
1412     HWCAP_MIPS_MIPS16       = (1 << 3),
1413     HWCAP_MIPS_MDMX         = (1 << 4),
1414     HWCAP_MIPS_MIPS3D       = (1 << 5),
1415     HWCAP_MIPS_SMARTMIPS    = (1 << 6),
1416     HWCAP_MIPS_DSP          = (1 << 7),
1417     HWCAP_MIPS_DSP2         = (1 << 8),
1418     HWCAP_MIPS_DSP3         = (1 << 9),
1419     HWCAP_MIPS_MIPS16E2     = (1 << 10),
1420     HWCAP_LOONGSON_MMI      = (1 << 11),
1421     HWCAP_LOONGSON_EXT      = (1 << 12),
1422     HWCAP_LOONGSON_EXT2     = (1 << 13),
1423     HWCAP_LOONGSON_CPUCFG   = (1 << 14),
1424 };
1425 
1426 #define ELF_HWCAP get_elf_hwcap()
1427 
1428 #define GET_FEATURE_INSN(_flag, _hwcap) \
1429     do { if (cpu->env.insn_flags & (_flag)) { hwcaps |= _hwcap; } } while (0)
1430 
1431 #define GET_FEATURE_REG_SET(_reg, _mask, _hwcap) \
1432     do { if (cpu->env._reg & (_mask)) { hwcaps |= _hwcap; } } while (0)
1433 
1434 #define GET_FEATURE_REG_EQU(_reg, _start, _length, _val, _hwcap) \
1435     do { \
1436         if (extract32(cpu->env._reg, (_start), (_length)) == (_val)) { \
1437             hwcaps |= _hwcap; \
1438         } \
1439     } while (0)
1440 
1441 static uint32_t get_elf_hwcap(void)
1442 {
1443     MIPSCPU *cpu = MIPS_CPU(thread_cpu);
1444     uint32_t hwcaps = 0;
1445 
1446     GET_FEATURE_REG_EQU(CP0_Config0, CP0C0_AR, CP0C0_AR_LENGTH,
1447                         2, HWCAP_MIPS_R6);
1448     GET_FEATURE_REG_SET(CP0_Config3, 1 << CP0C3_MSAP, HWCAP_MIPS_MSA);
1449     GET_FEATURE_INSN(ASE_LMMI, HWCAP_LOONGSON_MMI);
1450     GET_FEATURE_INSN(ASE_LEXT, HWCAP_LOONGSON_EXT);
1451 
1452     return hwcaps;
1453 }
1454 
1455 #undef GET_FEATURE_REG_EQU
1456 #undef GET_FEATURE_REG_SET
1457 #undef GET_FEATURE_INSN
1458 
1459 #endif /* TARGET_MIPS */
1460 
1461 #ifdef TARGET_MICROBLAZE
1462 
1463 #define elf_check_arch(x) ( (x) == EM_MICROBLAZE || (x) == EM_MICROBLAZE_OLD)
1464 
1465 #define ELF_CLASS   ELFCLASS32
1466 #define ELF_ARCH    EM_MICROBLAZE
1467 
1468 static inline void init_thread(struct target_pt_regs *regs,
1469                                struct image_info *infop)
1470 {
1471     regs->pc = infop->entry;
1472     regs->r1 = infop->start_stack;
1473 
1474 }
1475 
1476 #define ELF_EXEC_PAGESIZE        4096
1477 
1478 #define USE_ELF_CORE_DUMP
1479 #define ELF_NREG 38
1480 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1481 
1482 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs.  */
1483 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMBState *env)
1484 {
1485     int i, pos = 0;
1486 
1487     for (i = 0; i < 32; i++) {
1488         (*regs)[pos++] = tswapreg(env->regs[i]);
1489     }
1490 
1491     (*regs)[pos++] = tswapreg(env->pc);
1492     (*regs)[pos++] = tswapreg(mb_cpu_read_msr(env));
1493     (*regs)[pos++] = 0;
1494     (*regs)[pos++] = tswapreg(env->ear);
1495     (*regs)[pos++] = 0;
1496     (*regs)[pos++] = tswapreg(env->esr);
1497 }
1498 
1499 #endif /* TARGET_MICROBLAZE */
1500 
1501 #ifdef TARGET_NIOS2
1502 
1503 #define elf_check_arch(x) ((x) == EM_ALTERA_NIOS2)
1504 
1505 #define ELF_CLASS   ELFCLASS32
1506 #define ELF_ARCH    EM_ALTERA_NIOS2
1507 
1508 static void init_thread(struct target_pt_regs *regs, struct image_info *infop)
1509 {
1510     regs->ea = infop->entry;
1511     regs->sp = infop->start_stack;
1512 }
1513 
1514 #define LO_COMMPAGE  TARGET_PAGE_SIZE
1515 
1516 static bool init_guest_commpage(void)
1517 {
1518     static const uint8_t kuser_page[4 + 2 * 64] = {
1519         /* __kuser_helper_version */
1520         [0x00] = 0x02, 0x00, 0x00, 0x00,
1521 
1522         /* __kuser_cmpxchg */
1523         [0x04] = 0x3a, 0x6c, 0x3b, 0x00,  /* trap 16 */
1524                  0x3a, 0x28, 0x00, 0xf8,  /* ret */
1525 
1526         /* __kuser_sigtramp */
1527         [0x44] = 0xc4, 0x22, 0x80, 0x00,  /* movi r2, __NR_rt_sigreturn */
1528                  0x3a, 0x68, 0x3b, 0x00,  /* trap 0 */
1529     };
1530 
1531     void *want = g2h_untagged(LO_COMMPAGE & -qemu_host_page_size);
1532     void *addr = mmap(want, qemu_host_page_size, PROT_READ | PROT_WRITE,
1533                       MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0);
1534 
1535     if (addr == MAP_FAILED) {
1536         perror("Allocating guest commpage");
1537         exit(EXIT_FAILURE);
1538     }
1539     if (addr != want) {
1540         return false;
1541     }
1542 
1543     memcpy(addr, kuser_page, sizeof(kuser_page));
1544 
1545     if (mprotect(addr, qemu_host_page_size, PROT_READ)) {
1546         perror("Protecting guest commpage");
1547         exit(EXIT_FAILURE);
1548     }
1549 
1550     page_set_flags(LO_COMMPAGE, LO_COMMPAGE | ~TARGET_PAGE_MASK,
1551                    PAGE_READ | PAGE_EXEC | PAGE_VALID);
1552     return true;
1553 }
1554 
1555 #define ELF_EXEC_PAGESIZE        4096
1556 
1557 #define USE_ELF_CORE_DUMP
1558 #define ELF_NREG 49
1559 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1560 
1561 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs.  */
1562 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1563                                const CPUNios2State *env)
1564 {
1565     int i;
1566 
1567     (*regs)[0] = -1;
1568     for (i = 1; i < 8; i++)    /* r0-r7 */
1569         (*regs)[i] = tswapreg(env->regs[i + 7]);
1570 
1571     for (i = 8; i < 16; i++)   /* r8-r15 */
1572         (*regs)[i] = tswapreg(env->regs[i - 8]);
1573 
1574     for (i = 16; i < 24; i++)  /* r16-r23 */
1575         (*regs)[i] = tswapreg(env->regs[i + 7]);
1576     (*regs)[24] = -1;    /* R_ET */
1577     (*regs)[25] = -1;    /* R_BT */
1578     (*regs)[26] = tswapreg(env->regs[R_GP]);
1579     (*regs)[27] = tswapreg(env->regs[R_SP]);
1580     (*regs)[28] = tswapreg(env->regs[R_FP]);
1581     (*regs)[29] = tswapreg(env->regs[R_EA]);
1582     (*regs)[30] = -1;    /* R_SSTATUS */
1583     (*regs)[31] = tswapreg(env->regs[R_RA]);
1584 
1585     (*regs)[32] = tswapreg(env->pc);
1586 
1587     (*regs)[33] = -1; /* R_STATUS */
1588     (*regs)[34] = tswapreg(env->regs[CR_ESTATUS]);
1589 
1590     for (i = 35; i < 49; i++)    /* ... */
1591         (*regs)[i] = -1;
1592 }
1593 
1594 #endif /* TARGET_NIOS2 */
1595 
1596 #ifdef TARGET_OPENRISC
1597 
1598 #define ELF_ARCH EM_OPENRISC
1599 #define ELF_CLASS ELFCLASS32
1600 #define ELF_DATA  ELFDATA2MSB
1601 
1602 static inline void init_thread(struct target_pt_regs *regs,
1603                                struct image_info *infop)
1604 {
1605     regs->pc = infop->entry;
1606     regs->gpr[1] = infop->start_stack;
1607 }
1608 
1609 #define USE_ELF_CORE_DUMP
1610 #define ELF_EXEC_PAGESIZE 8192
1611 
1612 /* See linux kernel arch/openrisc/include/asm/elf.h.  */
1613 #define ELF_NREG 34 /* gprs and pc, sr */
1614 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1615 
1616 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1617                                const CPUOpenRISCState *env)
1618 {
1619     int i;
1620 
1621     for (i = 0; i < 32; i++) {
1622         (*regs)[i] = tswapreg(cpu_get_gpr(env, i));
1623     }
1624     (*regs)[32] = tswapreg(env->pc);
1625     (*regs)[33] = tswapreg(cpu_get_sr(env));
1626 }
1627 #define ELF_HWCAP 0
1628 #define ELF_PLATFORM NULL
1629 
1630 #endif /* TARGET_OPENRISC */
1631 
1632 #ifdef TARGET_SH4
1633 
1634 #define ELF_CLASS ELFCLASS32
1635 #define ELF_ARCH  EM_SH
1636 
1637 static inline void init_thread(struct target_pt_regs *regs,
1638                                struct image_info *infop)
1639 {
1640     /* Check other registers XXXXX */
1641     regs->pc = infop->entry;
1642     regs->regs[15] = infop->start_stack;
1643 }
1644 
1645 /* See linux kernel: arch/sh/include/asm/elf.h.  */
1646 #define ELF_NREG 23
1647 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1648 
1649 /* See linux kernel: arch/sh/include/asm/ptrace.h.  */
1650 enum {
1651     TARGET_REG_PC = 16,
1652     TARGET_REG_PR = 17,
1653     TARGET_REG_SR = 18,
1654     TARGET_REG_GBR = 19,
1655     TARGET_REG_MACH = 20,
1656     TARGET_REG_MACL = 21,
1657     TARGET_REG_SYSCALL = 22
1658 };
1659 
1660 static inline void elf_core_copy_regs(target_elf_gregset_t *regs,
1661                                       const CPUSH4State *env)
1662 {
1663     int i;
1664 
1665     for (i = 0; i < 16; i++) {
1666         (*regs)[i] = tswapreg(env->gregs[i]);
1667     }
1668 
1669     (*regs)[TARGET_REG_PC] = tswapreg(env->pc);
1670     (*regs)[TARGET_REG_PR] = tswapreg(env->pr);
1671     (*regs)[TARGET_REG_SR] = tswapreg(env->sr);
1672     (*regs)[TARGET_REG_GBR] = tswapreg(env->gbr);
1673     (*regs)[TARGET_REG_MACH] = tswapreg(env->mach);
1674     (*regs)[TARGET_REG_MACL] = tswapreg(env->macl);
1675     (*regs)[TARGET_REG_SYSCALL] = 0; /* FIXME */
1676 }
1677 
1678 #define USE_ELF_CORE_DUMP
1679 #define ELF_EXEC_PAGESIZE        4096
1680 
1681 enum {
1682     SH_CPU_HAS_FPU            = 0x0001, /* Hardware FPU support */
1683     SH_CPU_HAS_P2_FLUSH_BUG   = 0x0002, /* Need to flush the cache in P2 area */
1684     SH_CPU_HAS_MMU_PAGE_ASSOC = 0x0004, /* SH3: TLB way selection bit support */
1685     SH_CPU_HAS_DSP            = 0x0008, /* SH-DSP: DSP support */
1686     SH_CPU_HAS_PERF_COUNTER   = 0x0010, /* Hardware performance counters */
1687     SH_CPU_HAS_PTEA           = 0x0020, /* PTEA register */
1688     SH_CPU_HAS_LLSC           = 0x0040, /* movli.l/movco.l */
1689     SH_CPU_HAS_L2_CACHE       = 0x0080, /* Secondary cache / URAM */
1690     SH_CPU_HAS_OP32           = 0x0100, /* 32-bit instruction support */
1691     SH_CPU_HAS_PTEAEX         = 0x0200, /* PTE ASID Extension support */
1692 };
1693 
1694 #define ELF_HWCAP get_elf_hwcap()
1695 
1696 static uint32_t get_elf_hwcap(void)
1697 {
1698     SuperHCPU *cpu = SUPERH_CPU(thread_cpu);
1699     uint32_t hwcap = 0;
1700 
1701     hwcap |= SH_CPU_HAS_FPU;
1702 
1703     if (cpu->env.features & SH_FEATURE_SH4A) {
1704         hwcap |= SH_CPU_HAS_LLSC;
1705     }
1706 
1707     return hwcap;
1708 }
1709 
1710 #endif
1711 
1712 #ifdef TARGET_CRIS
1713 
1714 #define ELF_CLASS ELFCLASS32
1715 #define ELF_ARCH  EM_CRIS
1716 
1717 static inline void init_thread(struct target_pt_regs *regs,
1718                                struct image_info *infop)
1719 {
1720     regs->erp = infop->entry;
1721 }
1722 
1723 #define ELF_EXEC_PAGESIZE        8192
1724 
1725 #endif
1726 
1727 #ifdef TARGET_M68K
1728 
1729 #define ELF_CLASS       ELFCLASS32
1730 #define ELF_ARCH        EM_68K
1731 
1732 /* ??? Does this need to do anything?
1733    #define ELF_PLAT_INIT(_r) */
1734 
1735 static inline void init_thread(struct target_pt_regs *regs,
1736                                struct image_info *infop)
1737 {
1738     regs->usp = infop->start_stack;
1739     regs->sr = 0;
1740     regs->pc = infop->entry;
1741 }
1742 
1743 /* See linux kernel: arch/m68k/include/asm/elf.h.  */
1744 #define ELF_NREG 20
1745 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1746 
1747 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUM68KState *env)
1748 {
1749     (*regs)[0] = tswapreg(env->dregs[1]);
1750     (*regs)[1] = tswapreg(env->dregs[2]);
1751     (*regs)[2] = tswapreg(env->dregs[3]);
1752     (*regs)[3] = tswapreg(env->dregs[4]);
1753     (*regs)[4] = tswapreg(env->dregs[5]);
1754     (*regs)[5] = tswapreg(env->dregs[6]);
1755     (*regs)[6] = tswapreg(env->dregs[7]);
1756     (*regs)[7] = tswapreg(env->aregs[0]);
1757     (*regs)[8] = tswapreg(env->aregs[1]);
1758     (*regs)[9] = tswapreg(env->aregs[2]);
1759     (*regs)[10] = tswapreg(env->aregs[3]);
1760     (*regs)[11] = tswapreg(env->aregs[4]);
1761     (*regs)[12] = tswapreg(env->aregs[5]);
1762     (*regs)[13] = tswapreg(env->aregs[6]);
1763     (*regs)[14] = tswapreg(env->dregs[0]);
1764     (*regs)[15] = tswapreg(env->aregs[7]);
1765     (*regs)[16] = tswapreg(env->dregs[0]); /* FIXME: orig_d0 */
1766     (*regs)[17] = tswapreg(env->sr);
1767     (*regs)[18] = tswapreg(env->pc);
1768     (*regs)[19] = 0;  /* FIXME: regs->format | regs->vector */
1769 }
1770 
1771 #define USE_ELF_CORE_DUMP
1772 #define ELF_EXEC_PAGESIZE       8192
1773 
1774 #endif
1775 
1776 #ifdef TARGET_ALPHA
1777 
1778 #define ELF_CLASS      ELFCLASS64
1779 #define ELF_ARCH       EM_ALPHA
1780 
1781 static inline void init_thread(struct target_pt_regs *regs,
1782                                struct image_info *infop)
1783 {
1784     regs->pc = infop->entry;
1785     regs->ps = 8;
1786     regs->usp = infop->start_stack;
1787 }
1788 
1789 #define ELF_EXEC_PAGESIZE        8192
1790 
1791 #endif /* TARGET_ALPHA */
1792 
1793 #ifdef TARGET_S390X
1794 
1795 #define ELF_CLASS	ELFCLASS64
1796 #define ELF_DATA	ELFDATA2MSB
1797 #define ELF_ARCH	EM_S390
1798 
1799 #include "elf.h"
1800 
1801 #define ELF_HWCAP get_elf_hwcap()
1802 
1803 #define GET_FEATURE(_feat, _hwcap) \
1804     do { if (s390_has_feat(_feat)) { hwcap |= _hwcap; } } while (0)
1805 
1806 uint32_t get_elf_hwcap(void)
1807 {
1808     /*
1809      * Let's assume we always have esan3 and zarch.
1810      * 31-bit processes can use 64-bit registers (high gprs).
1811      */
1812     uint32_t hwcap = HWCAP_S390_ESAN3 | HWCAP_S390_ZARCH | HWCAP_S390_HIGH_GPRS;
1813 
1814     GET_FEATURE(S390_FEAT_STFLE, HWCAP_S390_STFLE);
1815     GET_FEATURE(S390_FEAT_MSA, HWCAP_S390_MSA);
1816     GET_FEATURE(S390_FEAT_LONG_DISPLACEMENT, HWCAP_S390_LDISP);
1817     GET_FEATURE(S390_FEAT_EXTENDED_IMMEDIATE, HWCAP_S390_EIMM);
1818     if (s390_has_feat(S390_FEAT_EXTENDED_TRANSLATION_3) &&
1819         s390_has_feat(S390_FEAT_ETF3_ENH)) {
1820         hwcap |= HWCAP_S390_ETF3EH;
1821     }
1822     GET_FEATURE(S390_FEAT_VECTOR, HWCAP_S390_VXRS);
1823     GET_FEATURE(S390_FEAT_VECTOR_ENH, HWCAP_S390_VXRS_EXT);
1824     GET_FEATURE(S390_FEAT_VECTOR_ENH2, HWCAP_S390_VXRS_EXT2);
1825 
1826     return hwcap;
1827 }
1828 
1829 const char *elf_hwcap_str(uint32_t bit)
1830 {
1831     static const char *hwcap_str[] = {
1832         [HWCAP_S390_NR_ESAN3]     = "esan3",
1833         [HWCAP_S390_NR_ZARCH]     = "zarch",
1834         [HWCAP_S390_NR_STFLE]     = "stfle",
1835         [HWCAP_S390_NR_MSA]       = "msa",
1836         [HWCAP_S390_NR_LDISP]     = "ldisp",
1837         [HWCAP_S390_NR_EIMM]      = "eimm",
1838         [HWCAP_S390_NR_DFP]       = "dfp",
1839         [HWCAP_S390_NR_HPAGE]     = "edat",
1840         [HWCAP_S390_NR_ETF3EH]    = "etf3eh",
1841         [HWCAP_S390_NR_HIGH_GPRS] = "highgprs",
1842         [HWCAP_S390_NR_TE]        = "te",
1843         [HWCAP_S390_NR_VXRS]      = "vx",
1844         [HWCAP_S390_NR_VXRS_BCD]  = "vxd",
1845         [HWCAP_S390_NR_VXRS_EXT]  = "vxe",
1846         [HWCAP_S390_NR_GS]        = "gs",
1847         [HWCAP_S390_NR_VXRS_EXT2] = "vxe2",
1848         [HWCAP_S390_NR_VXRS_PDE]  = "vxp",
1849         [HWCAP_S390_NR_SORT]      = "sort",
1850         [HWCAP_S390_NR_DFLT]      = "dflt",
1851         [HWCAP_S390_NR_NNPA]      = "nnpa",
1852         [HWCAP_S390_NR_PCI_MIO]   = "pcimio",
1853         [HWCAP_S390_NR_SIE]       = "sie",
1854     };
1855 
1856     return bit < ARRAY_SIZE(hwcap_str) ? hwcap_str[bit] : NULL;
1857 }
1858 
1859 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop)
1860 {
1861     regs->psw.addr = infop->entry;
1862     regs->psw.mask = PSW_MASK_DAT | PSW_MASK_IO | PSW_MASK_EXT | \
1863                      PSW_MASK_MCHECK | PSW_MASK_PSTATE | PSW_MASK_64 | \
1864                      PSW_MASK_32;
1865     regs->gprs[15] = infop->start_stack;
1866 }
1867 
1868 /* See linux kernel: arch/s390/include/uapi/asm/ptrace.h (s390_regs).  */
1869 #define ELF_NREG 27
1870 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1871 
1872 enum {
1873     TARGET_REG_PSWM = 0,
1874     TARGET_REG_PSWA = 1,
1875     TARGET_REG_GPRS = 2,
1876     TARGET_REG_ARS = 18,
1877     TARGET_REG_ORIG_R2 = 26,
1878 };
1879 
1880 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1881                                const CPUS390XState *env)
1882 {
1883     int i;
1884     uint32_t *aregs;
1885 
1886     (*regs)[TARGET_REG_PSWM] = tswapreg(env->psw.mask);
1887     (*regs)[TARGET_REG_PSWA] = tswapreg(env->psw.addr);
1888     for (i = 0; i < 16; i++) {
1889         (*regs)[TARGET_REG_GPRS + i] = tswapreg(env->regs[i]);
1890     }
1891     aregs = (uint32_t *)&((*regs)[TARGET_REG_ARS]);
1892     for (i = 0; i < 16; i++) {
1893         aregs[i] = tswap32(env->aregs[i]);
1894     }
1895     (*regs)[TARGET_REG_ORIG_R2] = 0;
1896 }
1897 
1898 #define USE_ELF_CORE_DUMP
1899 #define ELF_EXEC_PAGESIZE 4096
1900 
1901 #define VDSO_HEADER "vdso.c.inc"
1902 
1903 #endif /* TARGET_S390X */
1904 
1905 #ifdef TARGET_RISCV
1906 
1907 #define ELF_ARCH  EM_RISCV
1908 
1909 #ifdef TARGET_RISCV32
1910 #define ELF_CLASS ELFCLASS32
1911 #define VDSO_HEADER "vdso-32.c.inc"
1912 #else
1913 #define ELF_CLASS ELFCLASS64
1914 #define VDSO_HEADER "vdso-64.c.inc"
1915 #endif
1916 
1917 #define ELF_HWCAP get_elf_hwcap()
1918 
1919 static uint32_t get_elf_hwcap(void)
1920 {
1921 #define MISA_BIT(EXT) (1 << (EXT - 'A'))
1922     RISCVCPU *cpu = RISCV_CPU(thread_cpu);
1923     uint32_t mask = MISA_BIT('I') | MISA_BIT('M') | MISA_BIT('A')
1924                     | MISA_BIT('F') | MISA_BIT('D') | MISA_BIT('C')
1925                     | MISA_BIT('V');
1926 
1927     return cpu->env.misa_ext & mask;
1928 #undef MISA_BIT
1929 }
1930 
1931 static inline void init_thread(struct target_pt_regs *regs,
1932                                struct image_info *infop)
1933 {
1934     regs->sepc = infop->entry;
1935     regs->sp = infop->start_stack;
1936 }
1937 
1938 #define ELF_EXEC_PAGESIZE 4096
1939 
1940 #endif /* TARGET_RISCV */
1941 
1942 #ifdef TARGET_HPPA
1943 
1944 #define ELF_CLASS       ELFCLASS32
1945 #define ELF_ARCH        EM_PARISC
1946 #define ELF_PLATFORM    "PARISC"
1947 #define STACK_GROWS_DOWN 0
1948 #define STACK_ALIGNMENT  64
1949 
1950 #define VDSO_HEADER "vdso.c.inc"
1951 
1952 static inline void init_thread(struct target_pt_regs *regs,
1953                                struct image_info *infop)
1954 {
1955     regs->iaoq[0] = infop->entry;
1956     regs->iaoq[1] = infop->entry + 4;
1957     regs->gr[23] = 0;
1958     regs->gr[24] = infop->argv;
1959     regs->gr[25] = infop->argc;
1960     /* The top-of-stack contains a linkage buffer.  */
1961     regs->gr[30] = infop->start_stack + 64;
1962     regs->gr[31] = infop->entry;
1963 }
1964 
1965 #define LO_COMMPAGE  0
1966 
1967 static bool init_guest_commpage(void)
1968 {
1969     void *want = g2h_untagged(LO_COMMPAGE);
1970     void *addr = mmap(want, qemu_host_page_size, PROT_NONE,
1971                       MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0);
1972 
1973     if (addr == MAP_FAILED) {
1974         perror("Allocating guest commpage");
1975         exit(EXIT_FAILURE);
1976     }
1977     if (addr != want) {
1978         return false;
1979     }
1980 
1981     /*
1982      * On Linux, page zero is normally marked execute only + gateway.
1983      * Normal read or write is supposed to fail (thus PROT_NONE above),
1984      * but specific offsets have kernel code mapped to raise permissions
1985      * and implement syscalls.  Here, simply mark the page executable.
1986      * Special case the entry points during translation (see do_page_zero).
1987      */
1988     page_set_flags(LO_COMMPAGE, LO_COMMPAGE | ~TARGET_PAGE_MASK,
1989                    PAGE_EXEC | PAGE_VALID);
1990     return true;
1991 }
1992 
1993 #endif /* TARGET_HPPA */
1994 
1995 #ifdef TARGET_XTENSA
1996 
1997 #define ELF_CLASS       ELFCLASS32
1998 #define ELF_ARCH        EM_XTENSA
1999 
2000 static inline void init_thread(struct target_pt_regs *regs,
2001                                struct image_info *infop)
2002 {
2003     regs->windowbase = 0;
2004     regs->windowstart = 1;
2005     regs->areg[1] = infop->start_stack;
2006     regs->pc = infop->entry;
2007     if (info_is_fdpic(infop)) {
2008         regs->areg[4] = infop->loadmap_addr;
2009         regs->areg[5] = infop->interpreter_loadmap_addr;
2010         if (infop->interpreter_loadmap_addr) {
2011             regs->areg[6] = infop->interpreter_pt_dynamic_addr;
2012         } else {
2013             regs->areg[6] = infop->pt_dynamic_addr;
2014         }
2015     }
2016 }
2017 
2018 /* See linux kernel: arch/xtensa/include/asm/elf.h.  */
2019 #define ELF_NREG 128
2020 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
2021 
2022 enum {
2023     TARGET_REG_PC,
2024     TARGET_REG_PS,
2025     TARGET_REG_LBEG,
2026     TARGET_REG_LEND,
2027     TARGET_REG_LCOUNT,
2028     TARGET_REG_SAR,
2029     TARGET_REG_WINDOWSTART,
2030     TARGET_REG_WINDOWBASE,
2031     TARGET_REG_THREADPTR,
2032     TARGET_REG_AR0 = 64,
2033 };
2034 
2035 static void elf_core_copy_regs(target_elf_gregset_t *regs,
2036                                const CPUXtensaState *env)
2037 {
2038     unsigned i;
2039 
2040     (*regs)[TARGET_REG_PC] = tswapreg(env->pc);
2041     (*regs)[TARGET_REG_PS] = tswapreg(env->sregs[PS] & ~PS_EXCM);
2042     (*regs)[TARGET_REG_LBEG] = tswapreg(env->sregs[LBEG]);
2043     (*regs)[TARGET_REG_LEND] = tswapreg(env->sregs[LEND]);
2044     (*regs)[TARGET_REG_LCOUNT] = tswapreg(env->sregs[LCOUNT]);
2045     (*regs)[TARGET_REG_SAR] = tswapreg(env->sregs[SAR]);
2046     (*regs)[TARGET_REG_WINDOWSTART] = tswapreg(env->sregs[WINDOW_START]);
2047     (*regs)[TARGET_REG_WINDOWBASE] = tswapreg(env->sregs[WINDOW_BASE]);
2048     (*regs)[TARGET_REG_THREADPTR] = tswapreg(env->uregs[THREADPTR]);
2049     xtensa_sync_phys_from_window((CPUXtensaState *)env);
2050     for (i = 0; i < env->config->nareg; ++i) {
2051         (*regs)[TARGET_REG_AR0 + i] = tswapreg(env->phys_regs[i]);
2052     }
2053 }
2054 
2055 #define USE_ELF_CORE_DUMP
2056 #define ELF_EXEC_PAGESIZE       4096
2057 
2058 #endif /* TARGET_XTENSA */
2059 
2060 #ifdef TARGET_HEXAGON
2061 
2062 #define ELF_CLASS       ELFCLASS32
2063 #define ELF_ARCH        EM_HEXAGON
2064 
2065 static inline void init_thread(struct target_pt_regs *regs,
2066                                struct image_info *infop)
2067 {
2068     regs->sepc = infop->entry;
2069     regs->sp = infop->start_stack;
2070 }
2071 
2072 #endif /* TARGET_HEXAGON */
2073 
2074 #ifndef ELF_BASE_PLATFORM
2075 #define ELF_BASE_PLATFORM (NULL)
2076 #endif
2077 
2078 #ifndef ELF_PLATFORM
2079 #define ELF_PLATFORM (NULL)
2080 #endif
2081 
2082 #ifndef ELF_MACHINE
2083 #define ELF_MACHINE ELF_ARCH
2084 #endif
2085 
2086 #ifndef elf_check_arch
2087 #define elf_check_arch(x) ((x) == ELF_ARCH)
2088 #endif
2089 
2090 #ifndef elf_check_abi
2091 #define elf_check_abi(x) (1)
2092 #endif
2093 
2094 #ifndef ELF_HWCAP
2095 #define ELF_HWCAP 0
2096 #endif
2097 
2098 #ifndef STACK_GROWS_DOWN
2099 #define STACK_GROWS_DOWN 1
2100 #endif
2101 
2102 #ifndef STACK_ALIGNMENT
2103 #define STACK_ALIGNMENT 16
2104 #endif
2105 
2106 #ifdef TARGET_ABI32
2107 #undef ELF_CLASS
2108 #define ELF_CLASS ELFCLASS32
2109 #undef bswaptls
2110 #define bswaptls(ptr) bswap32s(ptr)
2111 #endif
2112 
2113 #ifndef EXSTACK_DEFAULT
2114 #define EXSTACK_DEFAULT false
2115 #endif
2116 
2117 #include "elf.h"
2118 
2119 /* We must delay the following stanzas until after "elf.h". */
2120 #if defined(TARGET_AARCH64)
2121 
2122 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz,
2123                                     const uint32_t *data,
2124                                     struct image_info *info,
2125                                     Error **errp)
2126 {
2127     if (pr_type == GNU_PROPERTY_AARCH64_FEATURE_1_AND) {
2128         if (pr_datasz != sizeof(uint32_t)) {
2129             error_setg(errp, "Ill-formed GNU_PROPERTY_AARCH64_FEATURE_1_AND");
2130             return false;
2131         }
2132         /* We will extract GNU_PROPERTY_AARCH64_FEATURE_1_BTI later. */
2133         info->note_flags = *data;
2134     }
2135     return true;
2136 }
2137 #define ARCH_USE_GNU_PROPERTY 1
2138 
2139 #else
2140 
2141 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz,
2142                                     const uint32_t *data,
2143                                     struct image_info *info,
2144                                     Error **errp)
2145 {
2146     g_assert_not_reached();
2147 }
2148 #define ARCH_USE_GNU_PROPERTY 0
2149 
2150 #endif
2151 
2152 struct exec
2153 {
2154     unsigned int a_info;   /* Use macros N_MAGIC, etc for access */
2155     unsigned int a_text;   /* length of text, in bytes */
2156     unsigned int a_data;   /* length of data, in bytes */
2157     unsigned int a_bss;    /* length of uninitialized data area, in bytes */
2158     unsigned int a_syms;   /* length of symbol table data in file, in bytes */
2159     unsigned int a_entry;  /* start address */
2160     unsigned int a_trsize; /* length of relocation info for text, in bytes */
2161     unsigned int a_drsize; /* length of relocation info for data, in bytes */
2162 };
2163 
2164 
2165 #define N_MAGIC(exec) ((exec).a_info & 0xffff)
2166 #define OMAGIC 0407
2167 #define NMAGIC 0410
2168 #define ZMAGIC 0413
2169 #define QMAGIC 0314
2170 
2171 #define DLINFO_ITEMS 16
2172 
2173 static inline void memcpy_fromfs(void * to, const void * from, unsigned long n)
2174 {
2175     memcpy(to, from, n);
2176 }
2177 
2178 #ifdef BSWAP_NEEDED
2179 static void bswap_ehdr(struct elfhdr *ehdr)
2180 {
2181     bswap16s(&ehdr->e_type);            /* Object file type */
2182     bswap16s(&ehdr->e_machine);         /* Architecture */
2183     bswap32s(&ehdr->e_version);         /* Object file version */
2184     bswaptls(&ehdr->e_entry);           /* Entry point virtual address */
2185     bswaptls(&ehdr->e_phoff);           /* Program header table file offset */
2186     bswaptls(&ehdr->e_shoff);           /* Section header table file offset */
2187     bswap32s(&ehdr->e_flags);           /* Processor-specific flags */
2188     bswap16s(&ehdr->e_ehsize);          /* ELF header size in bytes */
2189     bswap16s(&ehdr->e_phentsize);       /* Program header table entry size */
2190     bswap16s(&ehdr->e_phnum);           /* Program header table entry count */
2191     bswap16s(&ehdr->e_shentsize);       /* Section header table entry size */
2192     bswap16s(&ehdr->e_shnum);           /* Section header table entry count */
2193     bswap16s(&ehdr->e_shstrndx);        /* Section header string table index */
2194 }
2195 
2196 static void bswap_phdr(struct elf_phdr *phdr, int phnum)
2197 {
2198     int i;
2199     for (i = 0; i < phnum; ++i, ++phdr) {
2200         bswap32s(&phdr->p_type);        /* Segment type */
2201         bswap32s(&phdr->p_flags);       /* Segment flags */
2202         bswaptls(&phdr->p_offset);      /* Segment file offset */
2203         bswaptls(&phdr->p_vaddr);       /* Segment virtual address */
2204         bswaptls(&phdr->p_paddr);       /* Segment physical address */
2205         bswaptls(&phdr->p_filesz);      /* Segment size in file */
2206         bswaptls(&phdr->p_memsz);       /* Segment size in memory */
2207         bswaptls(&phdr->p_align);       /* Segment alignment */
2208     }
2209 }
2210 
2211 static void bswap_shdr(struct elf_shdr *shdr, int shnum)
2212 {
2213     int i;
2214     for (i = 0; i < shnum; ++i, ++shdr) {
2215         bswap32s(&shdr->sh_name);
2216         bswap32s(&shdr->sh_type);
2217         bswaptls(&shdr->sh_flags);
2218         bswaptls(&shdr->sh_addr);
2219         bswaptls(&shdr->sh_offset);
2220         bswaptls(&shdr->sh_size);
2221         bswap32s(&shdr->sh_link);
2222         bswap32s(&shdr->sh_info);
2223         bswaptls(&shdr->sh_addralign);
2224         bswaptls(&shdr->sh_entsize);
2225     }
2226 }
2227 
2228 static void bswap_sym(struct elf_sym *sym)
2229 {
2230     bswap32s(&sym->st_name);
2231     bswaptls(&sym->st_value);
2232     bswaptls(&sym->st_size);
2233     bswap16s(&sym->st_shndx);
2234 }
2235 
2236 #ifdef TARGET_MIPS
2237 static void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags)
2238 {
2239     bswap16s(&abiflags->version);
2240     bswap32s(&abiflags->ases);
2241     bswap32s(&abiflags->isa_ext);
2242     bswap32s(&abiflags->flags1);
2243     bswap32s(&abiflags->flags2);
2244 }
2245 #endif
2246 #else
2247 static inline void bswap_ehdr(struct elfhdr *ehdr) { }
2248 static inline void bswap_phdr(struct elf_phdr *phdr, int phnum) { }
2249 static inline void bswap_shdr(struct elf_shdr *shdr, int shnum) { }
2250 static inline void bswap_sym(struct elf_sym *sym) { }
2251 #ifdef TARGET_MIPS
2252 static inline void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags) { }
2253 #endif
2254 #endif
2255 
2256 #ifdef USE_ELF_CORE_DUMP
2257 static int elf_core_dump(int, const CPUArchState *);
2258 #endif /* USE_ELF_CORE_DUMP */
2259 static void load_symbols(struct elfhdr *hdr, const ImageSource *src,
2260                          abi_ulong load_bias);
2261 
2262 /* Verify the portions of EHDR within E_IDENT for the target.
2263    This can be performed before bswapping the entire header.  */
2264 static bool elf_check_ident(struct elfhdr *ehdr)
2265 {
2266     return (ehdr->e_ident[EI_MAG0] == ELFMAG0
2267             && ehdr->e_ident[EI_MAG1] == ELFMAG1
2268             && ehdr->e_ident[EI_MAG2] == ELFMAG2
2269             && ehdr->e_ident[EI_MAG3] == ELFMAG3
2270             && ehdr->e_ident[EI_CLASS] == ELF_CLASS
2271             && ehdr->e_ident[EI_DATA] == ELF_DATA
2272             && ehdr->e_ident[EI_VERSION] == EV_CURRENT);
2273 }
2274 
2275 /* Verify the portions of EHDR outside of E_IDENT for the target.
2276    This has to wait until after bswapping the header.  */
2277 static bool elf_check_ehdr(struct elfhdr *ehdr)
2278 {
2279     return (elf_check_arch(ehdr->e_machine)
2280             && elf_check_abi(ehdr->e_flags)
2281             && ehdr->e_ehsize == sizeof(struct elfhdr)
2282             && ehdr->e_phentsize == sizeof(struct elf_phdr)
2283             && (ehdr->e_type == ET_EXEC || ehdr->e_type == ET_DYN));
2284 }
2285 
2286 /*
2287  * 'copy_elf_strings()' copies argument/envelope strings from user
2288  * memory to free pages in kernel mem. These are in a format ready
2289  * to be put directly into the top of new user memory.
2290  *
2291  */
2292 static abi_ulong copy_elf_strings(int argc, char **argv, char *scratch,
2293                                   abi_ulong p, abi_ulong stack_limit)
2294 {
2295     char *tmp;
2296     int len, i;
2297     abi_ulong top = p;
2298 
2299     if (!p) {
2300         return 0;       /* bullet-proofing */
2301     }
2302 
2303     if (STACK_GROWS_DOWN) {
2304         int offset = ((p - 1) % TARGET_PAGE_SIZE) + 1;
2305         for (i = argc - 1; i >= 0; --i) {
2306             tmp = argv[i];
2307             if (!tmp) {
2308                 fprintf(stderr, "VFS: argc is wrong");
2309                 exit(-1);
2310             }
2311             len = strlen(tmp) + 1;
2312             tmp += len;
2313 
2314             if (len > (p - stack_limit)) {
2315                 return 0;
2316             }
2317             while (len) {
2318                 int bytes_to_copy = (len > offset) ? offset : len;
2319                 tmp -= bytes_to_copy;
2320                 p -= bytes_to_copy;
2321                 offset -= bytes_to_copy;
2322                 len -= bytes_to_copy;
2323 
2324                 memcpy_fromfs(scratch + offset, tmp, bytes_to_copy);
2325 
2326                 if (offset == 0) {
2327                     memcpy_to_target(p, scratch, top - p);
2328                     top = p;
2329                     offset = TARGET_PAGE_SIZE;
2330                 }
2331             }
2332         }
2333         if (p != top) {
2334             memcpy_to_target(p, scratch + offset, top - p);
2335         }
2336     } else {
2337         int remaining = TARGET_PAGE_SIZE - (p % TARGET_PAGE_SIZE);
2338         for (i = 0; i < argc; ++i) {
2339             tmp = argv[i];
2340             if (!tmp) {
2341                 fprintf(stderr, "VFS: argc is wrong");
2342                 exit(-1);
2343             }
2344             len = strlen(tmp) + 1;
2345             if (len > (stack_limit - p)) {
2346                 return 0;
2347             }
2348             while (len) {
2349                 int bytes_to_copy = (len > remaining) ? remaining : len;
2350 
2351                 memcpy_fromfs(scratch + (p - top), tmp, bytes_to_copy);
2352 
2353                 tmp += bytes_to_copy;
2354                 remaining -= bytes_to_copy;
2355                 p += bytes_to_copy;
2356                 len -= bytes_to_copy;
2357 
2358                 if (remaining == 0) {
2359                     memcpy_to_target(top, scratch, p - top);
2360                     top = p;
2361                     remaining = TARGET_PAGE_SIZE;
2362                 }
2363             }
2364         }
2365         if (p != top) {
2366             memcpy_to_target(top, scratch, p - top);
2367         }
2368     }
2369 
2370     return p;
2371 }
2372 
2373 /* Older linux kernels provide up to MAX_ARG_PAGES (default: 32) of
2374  * argument/environment space. Newer kernels (>2.6.33) allow more,
2375  * dependent on stack size, but guarantee at least 32 pages for
2376  * backwards compatibility.
2377  */
2378 #define STACK_LOWER_LIMIT (32 * TARGET_PAGE_SIZE)
2379 
2380 static abi_ulong setup_arg_pages(struct linux_binprm *bprm,
2381                                  struct image_info *info)
2382 {
2383     abi_ulong size, error, guard;
2384     int prot;
2385 
2386     size = guest_stack_size;
2387     if (size < STACK_LOWER_LIMIT) {
2388         size = STACK_LOWER_LIMIT;
2389     }
2390 
2391     if (STACK_GROWS_DOWN) {
2392         guard = TARGET_PAGE_SIZE;
2393         if (guard < qemu_real_host_page_size()) {
2394             guard = qemu_real_host_page_size();
2395         }
2396     } else {
2397         /* no guard page for hppa target where stack grows upwards. */
2398         guard = 0;
2399     }
2400 
2401     prot = PROT_READ | PROT_WRITE;
2402     if (info->exec_stack) {
2403         prot |= PROT_EXEC;
2404     }
2405     error = target_mmap(0, size + guard, prot,
2406                         MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
2407     if (error == -1) {
2408         perror("mmap stack");
2409         exit(-1);
2410     }
2411 
2412     /* We reserve one extra page at the top of the stack as guard.  */
2413     if (STACK_GROWS_DOWN) {
2414         target_mprotect(error, guard, PROT_NONE);
2415         info->stack_limit = error + guard;
2416         return info->stack_limit + size - sizeof(void *);
2417     } else {
2418         info->stack_limit = error + size;
2419         return error;
2420     }
2421 }
2422 
2423 /**
2424  * zero_bss:
2425  *
2426  * Map and zero the bss.  We need to explicitly zero any fractional pages
2427  * after the data section (i.e. bss).  Return false on mapping failure.
2428  */
2429 static bool zero_bss(abi_ulong start_bss, abi_ulong end_bss,
2430                      int prot, Error **errp)
2431 {
2432     abi_ulong align_bss;
2433 
2434     /* We only expect writable bss; the code segment shouldn't need this. */
2435     if (!(prot & PROT_WRITE)) {
2436         error_setg(errp, "PT_LOAD with non-writable bss");
2437         return false;
2438     }
2439 
2440     align_bss = TARGET_PAGE_ALIGN(start_bss);
2441     end_bss = TARGET_PAGE_ALIGN(end_bss);
2442 
2443     if (start_bss < align_bss) {
2444         int flags = page_get_flags(start_bss);
2445 
2446         if (!(flags & PAGE_BITS)) {
2447             /*
2448              * The whole address space of the executable was reserved
2449              * at the start, therefore all pages will be VALID.
2450              * But assuming there are no PROT_NONE PT_LOAD segments,
2451              * a PROT_NONE page means no data all bss, and we can
2452              * simply extend the new anon mapping back to the start
2453              * of the page of bss.
2454              */
2455             align_bss -= TARGET_PAGE_SIZE;
2456         } else {
2457             /*
2458              * The start of the bss shares a page with something.
2459              * The only thing that we expect is the data section,
2460              * which would already be marked writable.
2461              * Overlapping the RX code segment seems malformed.
2462              */
2463             if (!(flags & PAGE_WRITE)) {
2464                 error_setg(errp, "PT_LOAD with bss overlapping "
2465                            "non-writable page");
2466                 return false;
2467             }
2468 
2469             /* The page is already mapped and writable. */
2470             memset(g2h_untagged(start_bss), 0, align_bss - start_bss);
2471         }
2472     }
2473 
2474     if (align_bss < end_bss &&
2475         target_mmap(align_bss, end_bss - align_bss, prot,
2476                     MAP_FIXED | MAP_PRIVATE | MAP_ANON, -1, 0) == -1) {
2477         error_setg_errno(errp, errno, "Error mapping bss");
2478         return false;
2479     }
2480     return true;
2481 }
2482 
2483 #if defined(TARGET_ARM)
2484 static int elf_is_fdpic(struct elfhdr *exec)
2485 {
2486     return exec->e_ident[EI_OSABI] == ELFOSABI_ARM_FDPIC;
2487 }
2488 #elif defined(TARGET_XTENSA)
2489 static int elf_is_fdpic(struct elfhdr *exec)
2490 {
2491     return exec->e_ident[EI_OSABI] == ELFOSABI_XTENSA_FDPIC;
2492 }
2493 #else
2494 /* Default implementation, always false.  */
2495 static int elf_is_fdpic(struct elfhdr *exec)
2496 {
2497     return 0;
2498 }
2499 #endif
2500 
2501 static abi_ulong loader_build_fdpic_loadmap(struct image_info *info, abi_ulong sp)
2502 {
2503     uint16_t n;
2504     struct elf32_fdpic_loadseg *loadsegs = info->loadsegs;
2505 
2506     /* elf32_fdpic_loadseg */
2507     n = info->nsegs;
2508     while (n--) {
2509         sp -= 12;
2510         put_user_u32(loadsegs[n].addr, sp+0);
2511         put_user_u32(loadsegs[n].p_vaddr, sp+4);
2512         put_user_u32(loadsegs[n].p_memsz, sp+8);
2513     }
2514 
2515     /* elf32_fdpic_loadmap */
2516     sp -= 4;
2517     put_user_u16(0, sp+0); /* version */
2518     put_user_u16(info->nsegs, sp+2); /* nsegs */
2519 
2520     info->personality = PER_LINUX_FDPIC;
2521     info->loadmap_addr = sp;
2522 
2523     return sp;
2524 }
2525 
2526 static abi_ulong create_elf_tables(abi_ulong p, int argc, int envc,
2527                                    struct elfhdr *exec,
2528                                    struct image_info *info,
2529                                    struct image_info *interp_info,
2530                                    struct image_info *vdso_info)
2531 {
2532     abi_ulong sp;
2533     abi_ulong u_argc, u_argv, u_envp, u_auxv;
2534     int size;
2535     int i;
2536     abi_ulong u_rand_bytes;
2537     uint8_t k_rand_bytes[16];
2538     abi_ulong u_platform, u_base_platform;
2539     const char *k_platform, *k_base_platform;
2540     const int n = sizeof(elf_addr_t);
2541 
2542     sp = p;
2543 
2544     /* Needs to be before we load the env/argc/... */
2545     if (elf_is_fdpic(exec)) {
2546         /* Need 4 byte alignment for these structs */
2547         sp &= ~3;
2548         sp = loader_build_fdpic_loadmap(info, sp);
2549         info->other_info = interp_info;
2550         if (interp_info) {
2551             interp_info->other_info = info;
2552             sp = loader_build_fdpic_loadmap(interp_info, sp);
2553             info->interpreter_loadmap_addr = interp_info->loadmap_addr;
2554             info->interpreter_pt_dynamic_addr = interp_info->pt_dynamic_addr;
2555         } else {
2556             info->interpreter_loadmap_addr = 0;
2557             info->interpreter_pt_dynamic_addr = 0;
2558         }
2559     }
2560 
2561     u_base_platform = 0;
2562     k_base_platform = ELF_BASE_PLATFORM;
2563     if (k_base_platform) {
2564         size_t len = strlen(k_base_platform) + 1;
2565         if (STACK_GROWS_DOWN) {
2566             sp -= (len + n - 1) & ~(n - 1);
2567             u_base_platform = sp;
2568             /* FIXME - check return value of memcpy_to_target() for failure */
2569             memcpy_to_target(sp, k_base_platform, len);
2570         } else {
2571             memcpy_to_target(sp, k_base_platform, len);
2572             u_base_platform = sp;
2573             sp += len + 1;
2574         }
2575     }
2576 
2577     u_platform = 0;
2578     k_platform = ELF_PLATFORM;
2579     if (k_platform) {
2580         size_t len = strlen(k_platform) + 1;
2581         if (STACK_GROWS_DOWN) {
2582             sp -= (len + n - 1) & ~(n - 1);
2583             u_platform = sp;
2584             /* FIXME - check return value of memcpy_to_target() for failure */
2585             memcpy_to_target(sp, k_platform, len);
2586         } else {
2587             memcpy_to_target(sp, k_platform, len);
2588             u_platform = sp;
2589             sp += len + 1;
2590         }
2591     }
2592 
2593     /* Provide 16 byte alignment for the PRNG, and basic alignment for
2594      * the argv and envp pointers.
2595      */
2596     if (STACK_GROWS_DOWN) {
2597         sp = QEMU_ALIGN_DOWN(sp, 16);
2598     } else {
2599         sp = QEMU_ALIGN_UP(sp, 16);
2600     }
2601 
2602     /*
2603      * Generate 16 random bytes for userspace PRNG seeding.
2604      */
2605     qemu_guest_getrandom_nofail(k_rand_bytes, sizeof(k_rand_bytes));
2606     if (STACK_GROWS_DOWN) {
2607         sp -= 16;
2608         u_rand_bytes = sp;
2609         /* FIXME - check return value of memcpy_to_target() for failure */
2610         memcpy_to_target(sp, k_rand_bytes, 16);
2611     } else {
2612         memcpy_to_target(sp, k_rand_bytes, 16);
2613         u_rand_bytes = sp;
2614         sp += 16;
2615     }
2616 
2617     size = (DLINFO_ITEMS + 1) * 2;
2618     if (k_base_platform) {
2619         size += 2;
2620     }
2621     if (k_platform) {
2622         size += 2;
2623     }
2624     if (vdso_info) {
2625         size += 2;
2626     }
2627 #ifdef DLINFO_ARCH_ITEMS
2628     size += DLINFO_ARCH_ITEMS * 2;
2629 #endif
2630 #ifdef ELF_HWCAP2
2631     size += 2;
2632 #endif
2633     info->auxv_len = size * n;
2634 
2635     size += envc + argc + 2;
2636     size += 1;  /* argc itself */
2637     size *= n;
2638 
2639     /* Allocate space and finalize stack alignment for entry now.  */
2640     if (STACK_GROWS_DOWN) {
2641         u_argc = QEMU_ALIGN_DOWN(sp - size, STACK_ALIGNMENT);
2642         sp = u_argc;
2643     } else {
2644         u_argc = sp;
2645         sp = QEMU_ALIGN_UP(sp + size, STACK_ALIGNMENT);
2646     }
2647 
2648     u_argv = u_argc + n;
2649     u_envp = u_argv + (argc + 1) * n;
2650     u_auxv = u_envp + (envc + 1) * n;
2651     info->saved_auxv = u_auxv;
2652     info->argc = argc;
2653     info->envc = envc;
2654     info->argv = u_argv;
2655     info->envp = u_envp;
2656 
2657     /* This is correct because Linux defines
2658      * elf_addr_t as Elf32_Off / Elf64_Off
2659      */
2660 #define NEW_AUX_ENT(id, val) do {               \
2661         put_user_ual(id, u_auxv);  u_auxv += n; \
2662         put_user_ual(val, u_auxv); u_auxv += n; \
2663     } while(0)
2664 
2665 #ifdef ARCH_DLINFO
2666     /*
2667      * ARCH_DLINFO must come first so platform specific code can enforce
2668      * special alignment requirements on the AUXV if necessary (eg. PPC).
2669      */
2670     ARCH_DLINFO;
2671 #endif
2672     /* There must be exactly DLINFO_ITEMS entries here, or the assert
2673      * on info->auxv_len will trigger.
2674      */
2675     NEW_AUX_ENT(AT_PHDR, (abi_ulong)(info->load_addr + exec->e_phoff));
2676     NEW_AUX_ENT(AT_PHENT, (abi_ulong)(sizeof (struct elf_phdr)));
2677     NEW_AUX_ENT(AT_PHNUM, (abi_ulong)(exec->e_phnum));
2678     if ((info->alignment & ~qemu_host_page_mask) != 0) {
2679         /* Target doesn't support host page size alignment */
2680         NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(TARGET_PAGE_SIZE));
2681     } else {
2682         NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(MAX(TARGET_PAGE_SIZE,
2683                                                qemu_host_page_size)));
2684     }
2685     NEW_AUX_ENT(AT_BASE, (abi_ulong)(interp_info ? interp_info->load_addr : 0));
2686     NEW_AUX_ENT(AT_FLAGS, (abi_ulong)0);
2687     NEW_AUX_ENT(AT_ENTRY, info->entry);
2688     NEW_AUX_ENT(AT_UID, (abi_ulong) getuid());
2689     NEW_AUX_ENT(AT_EUID, (abi_ulong) geteuid());
2690     NEW_AUX_ENT(AT_GID, (abi_ulong) getgid());
2691     NEW_AUX_ENT(AT_EGID, (abi_ulong) getegid());
2692     NEW_AUX_ENT(AT_HWCAP, (abi_ulong) ELF_HWCAP);
2693     NEW_AUX_ENT(AT_CLKTCK, (abi_ulong) sysconf(_SC_CLK_TCK));
2694     NEW_AUX_ENT(AT_RANDOM, (abi_ulong) u_rand_bytes);
2695     NEW_AUX_ENT(AT_SECURE, (abi_ulong) qemu_getauxval(AT_SECURE));
2696     NEW_AUX_ENT(AT_EXECFN, info->file_string);
2697 
2698 #ifdef ELF_HWCAP2
2699     NEW_AUX_ENT(AT_HWCAP2, (abi_ulong) ELF_HWCAP2);
2700 #endif
2701 
2702     if (u_base_platform) {
2703         NEW_AUX_ENT(AT_BASE_PLATFORM, u_base_platform);
2704     }
2705     if (u_platform) {
2706         NEW_AUX_ENT(AT_PLATFORM, u_platform);
2707     }
2708     if (vdso_info) {
2709         NEW_AUX_ENT(AT_SYSINFO_EHDR, vdso_info->load_addr);
2710     }
2711     NEW_AUX_ENT (AT_NULL, 0);
2712 #undef NEW_AUX_ENT
2713 
2714     /* Check that our initial calculation of the auxv length matches how much
2715      * we actually put into it.
2716      */
2717     assert(info->auxv_len == u_auxv - info->saved_auxv);
2718 
2719     put_user_ual(argc, u_argc);
2720 
2721     p = info->arg_strings;
2722     for (i = 0; i < argc; ++i) {
2723         put_user_ual(p, u_argv);
2724         u_argv += n;
2725         p += target_strlen(p) + 1;
2726     }
2727     put_user_ual(0, u_argv);
2728 
2729     p = info->env_strings;
2730     for (i = 0; i < envc; ++i) {
2731         put_user_ual(p, u_envp);
2732         u_envp += n;
2733         p += target_strlen(p) + 1;
2734     }
2735     put_user_ual(0, u_envp);
2736 
2737     return sp;
2738 }
2739 
2740 #if defined(HI_COMMPAGE)
2741 #define LO_COMMPAGE -1
2742 #elif defined(LO_COMMPAGE)
2743 #define HI_COMMPAGE 0
2744 #else
2745 #define HI_COMMPAGE 0
2746 #define LO_COMMPAGE -1
2747 #ifndef INIT_GUEST_COMMPAGE
2748 #define init_guest_commpage() true
2749 #endif
2750 #endif
2751 
2752 /**
2753  * pgb_try_mmap:
2754  * @addr: host start address
2755  * @addr_last: host last address
2756  * @keep: do not unmap the probe region
2757  *
2758  * Return 1 if [@addr, @addr_last] is not mapped in the host,
2759  * return 0 if it is not available to map, and -1 on mmap error.
2760  * If @keep, the region is left mapped on success, otherwise unmapped.
2761  */
2762 static int pgb_try_mmap(uintptr_t addr, uintptr_t addr_last, bool keep)
2763 {
2764     size_t size = addr_last - addr + 1;
2765     void *p = mmap((void *)addr, size, PROT_NONE,
2766                    MAP_ANONYMOUS | MAP_PRIVATE |
2767                    MAP_NORESERVE | MAP_FIXED_NOREPLACE, -1, 0);
2768     int ret;
2769 
2770     if (p == MAP_FAILED) {
2771         return errno == EEXIST ? 0 : -1;
2772     }
2773     ret = p == (void *)addr;
2774     if (!keep || !ret) {
2775         munmap(p, size);
2776     }
2777     return ret;
2778 }
2779 
2780 /**
2781  * pgb_try_mmap_skip_brk(uintptr_t addr, uintptr_t size, uintptr_t brk)
2782  * @addr: host address
2783  * @addr_last: host last address
2784  * @brk: host brk
2785  *
2786  * Like pgb_try_mmap, but additionally reserve some memory following brk.
2787  */
2788 static int pgb_try_mmap_skip_brk(uintptr_t addr, uintptr_t addr_last,
2789                                  uintptr_t brk, bool keep)
2790 {
2791     uintptr_t brk_last = brk + 16 * MiB - 1;
2792 
2793     /* Do not map anything close to the host brk. */
2794     if (addr <= brk_last && brk <= addr_last) {
2795         return 0;
2796     }
2797     return pgb_try_mmap(addr, addr_last, keep);
2798 }
2799 
2800 /**
2801  * pgb_try_mmap_set:
2802  * @ga: set of guest addrs
2803  * @base: guest_base
2804  * @brk: host brk
2805  *
2806  * Return true if all @ga can be mapped by the host at @base.
2807  * On success, retain the mapping at index 0 for reserved_va.
2808  */
2809 
2810 typedef struct PGBAddrs {
2811     uintptr_t bounds[3][2]; /* start/last pairs */
2812     int nbounds;
2813 } PGBAddrs;
2814 
2815 static bool pgb_try_mmap_set(const PGBAddrs *ga, uintptr_t base, uintptr_t brk)
2816 {
2817     for (int i = ga->nbounds - 1; i >= 0; --i) {
2818         if (pgb_try_mmap_skip_brk(ga->bounds[i][0] + base,
2819                                   ga->bounds[i][1] + base,
2820                                   brk, i == 0 && reserved_va) <= 0) {
2821             return false;
2822         }
2823     }
2824     return true;
2825 }
2826 
2827 /**
2828  * pgb_addr_set:
2829  * @ga: output set of guest addrs
2830  * @guest_loaddr: guest image low address
2831  * @guest_loaddr: guest image high address
2832  * @identity: create for identity mapping
2833  *
2834  * Fill in @ga with the image, COMMPAGE and NULL page.
2835  */
2836 static bool pgb_addr_set(PGBAddrs *ga, abi_ulong guest_loaddr,
2837                          abi_ulong guest_hiaddr, bool try_identity)
2838 {
2839     int n;
2840 
2841     /*
2842      * With a low commpage, or a guest mapped very low,
2843      * we may not be able to use the identity map.
2844      */
2845     if (try_identity) {
2846         if (LO_COMMPAGE != -1 && LO_COMMPAGE < mmap_min_addr) {
2847             return false;
2848         }
2849         if (guest_loaddr != 0 && guest_loaddr < mmap_min_addr) {
2850             return false;
2851         }
2852     }
2853 
2854     memset(ga, 0, sizeof(*ga));
2855     n = 0;
2856 
2857     if (reserved_va) {
2858         ga->bounds[n][0] = try_identity ? mmap_min_addr : 0;
2859         ga->bounds[n][1] = reserved_va;
2860         n++;
2861         /* LO_COMMPAGE and NULL handled by reserving from 0. */
2862     } else {
2863         /* Add any LO_COMMPAGE or NULL page. */
2864         if (LO_COMMPAGE != -1) {
2865             ga->bounds[n][0] = 0;
2866             ga->bounds[n][1] = LO_COMMPAGE + TARGET_PAGE_SIZE - 1;
2867             n++;
2868         } else if (!try_identity) {
2869             ga->bounds[n][0] = 0;
2870             ga->bounds[n][1] = TARGET_PAGE_SIZE - 1;
2871             n++;
2872         }
2873 
2874         /* Add the guest image for ET_EXEC. */
2875         if (guest_loaddr) {
2876             ga->bounds[n][0] = guest_loaddr;
2877             ga->bounds[n][1] = guest_hiaddr;
2878             n++;
2879         }
2880     }
2881 
2882     /*
2883      * Temporarily disable
2884      *   "comparison is always false due to limited range of data type"
2885      * due to comparison between unsigned and (possible) 0.
2886      */
2887 #pragma GCC diagnostic push
2888 #pragma GCC diagnostic ignored "-Wtype-limits"
2889 
2890     /* Add any HI_COMMPAGE not covered by reserved_va. */
2891     if (reserved_va < HI_COMMPAGE) {
2892         ga->bounds[n][0] = HI_COMMPAGE & qemu_host_page_mask;
2893         ga->bounds[n][1] = HI_COMMPAGE + TARGET_PAGE_SIZE - 1;
2894         n++;
2895     }
2896 
2897 #pragma GCC diagnostic pop
2898 
2899     ga->nbounds = n;
2900     return true;
2901 }
2902 
2903 static void pgb_fail_in_use(const char *image_name)
2904 {
2905     error_report("%s: requires virtual address space that is in use "
2906                  "(omit the -B option or choose a different value)",
2907                  image_name);
2908     exit(EXIT_FAILURE);
2909 }
2910 
2911 static void pgb_fixed(const char *image_name, uintptr_t guest_loaddr,
2912                       uintptr_t guest_hiaddr, uintptr_t align)
2913 {
2914     PGBAddrs ga;
2915     uintptr_t brk = (uintptr_t)sbrk(0);
2916 
2917     if (!QEMU_IS_ALIGNED(guest_base, align)) {
2918         fprintf(stderr, "Requested guest base %p does not satisfy "
2919                 "host minimum alignment (0x%" PRIxPTR ")\n",
2920                 (void *)guest_base, align);
2921         exit(EXIT_FAILURE);
2922     }
2923 
2924     if (!pgb_addr_set(&ga, guest_loaddr, guest_hiaddr, !guest_base)
2925         || !pgb_try_mmap_set(&ga, guest_base, brk)) {
2926         pgb_fail_in_use(image_name);
2927     }
2928 }
2929 
2930 /**
2931  * pgb_find_fallback:
2932  *
2933  * This is a fallback method for finding holes in the host address space
2934  * if we don't have the benefit of being able to access /proc/self/map.
2935  * It can potentially take a very long time as we can only dumbly iterate
2936  * up the host address space seeing if the allocation would work.
2937  */
2938 static uintptr_t pgb_find_fallback(const PGBAddrs *ga, uintptr_t align,
2939                                    uintptr_t brk)
2940 {
2941     /* TODO: come up with a better estimate of how much to skip. */
2942     uintptr_t skip = sizeof(uintptr_t) == 4 ? MiB : GiB;
2943 
2944     for (uintptr_t base = skip; ; base += skip) {
2945         base = ROUND_UP(base, align);
2946         if (pgb_try_mmap_set(ga, base, brk)) {
2947             return base;
2948         }
2949         if (base >= -skip) {
2950             return -1;
2951         }
2952     }
2953 }
2954 
2955 static uintptr_t pgb_try_itree(const PGBAddrs *ga, uintptr_t base,
2956                                IntervalTreeRoot *root)
2957 {
2958     for (int i = ga->nbounds - 1; i >= 0; --i) {
2959         uintptr_t s = base + ga->bounds[i][0];
2960         uintptr_t l = base + ga->bounds[i][1];
2961         IntervalTreeNode *n;
2962 
2963         if (l < s) {
2964             /* Wraparound. Skip to advance S to mmap_min_addr. */
2965             return mmap_min_addr - s;
2966         }
2967 
2968         n = interval_tree_iter_first(root, s, l);
2969         if (n != NULL) {
2970             /* Conflict.  Skip to advance S to LAST + 1. */
2971             return n->last - s + 1;
2972         }
2973     }
2974     return 0;  /* success */
2975 }
2976 
2977 static uintptr_t pgb_find_itree(const PGBAddrs *ga, IntervalTreeRoot *root,
2978                                 uintptr_t align, uintptr_t brk)
2979 {
2980     uintptr_t last = mmap_min_addr;
2981     uintptr_t base, skip;
2982 
2983     while (true) {
2984         base = ROUND_UP(last, align);
2985         if (base < last) {
2986             return -1;
2987         }
2988 
2989         skip = pgb_try_itree(ga, base, root);
2990         if (skip == 0) {
2991             break;
2992         }
2993 
2994         last = base + skip;
2995         if (last < base) {
2996             return -1;
2997         }
2998     }
2999 
3000     /*
3001      * We've chosen 'base' based on holes in the interval tree,
3002      * but we don't yet know if it is a valid host address.
3003      * Because it is the first matching hole, if the host addresses
3004      * are invalid we know there are no further matches.
3005      */
3006     return pgb_try_mmap_set(ga, base, brk) ? base : -1;
3007 }
3008 
3009 static void pgb_dynamic(const char *image_name, uintptr_t guest_loaddr,
3010                         uintptr_t guest_hiaddr, uintptr_t align)
3011 {
3012     IntervalTreeRoot *root;
3013     uintptr_t brk, ret;
3014     PGBAddrs ga;
3015 
3016     assert(QEMU_IS_ALIGNED(guest_loaddr, align));
3017 
3018     /* Try the identity map first. */
3019     if (pgb_addr_set(&ga, guest_loaddr, guest_hiaddr, true)) {
3020         brk = (uintptr_t)sbrk(0);
3021         if (pgb_try_mmap_set(&ga, 0, brk)) {
3022             guest_base = 0;
3023             return;
3024         }
3025     }
3026 
3027     /*
3028      * Rebuild the address set for non-identity map.
3029      * This differs in the mapping of the guest NULL page.
3030      */
3031     pgb_addr_set(&ga, guest_loaddr, guest_hiaddr, false);
3032 
3033     root = read_self_maps();
3034 
3035     /* Read brk after we've read the maps, which will malloc. */
3036     brk = (uintptr_t)sbrk(0);
3037 
3038     if (!root) {
3039         ret = pgb_find_fallback(&ga, align, brk);
3040     } else {
3041         /*
3042          * Reserve the area close to the host brk.
3043          * This will be freed with the rest of the tree.
3044          */
3045         IntervalTreeNode *b = g_new0(IntervalTreeNode, 1);
3046         b->start = brk;
3047         b->last = brk + 16 * MiB - 1;
3048         interval_tree_insert(b, root);
3049 
3050         ret = pgb_find_itree(&ga, root, align, brk);
3051         free_self_maps(root);
3052     }
3053 
3054     if (ret == -1) {
3055         int w = TARGET_LONG_BITS / 4;
3056 
3057         error_report("%s: Unable to find a guest_base to satisfy all "
3058                      "guest address mapping requirements", image_name);
3059 
3060         for (int i = 0; i < ga.nbounds; ++i) {
3061             error_printf("  %0*" PRIx64 "-%0*" PRIx64 "\n",
3062                          w, (uint64_t)ga.bounds[i][0],
3063                          w, (uint64_t)ga.bounds[i][1]);
3064         }
3065         exit(EXIT_FAILURE);
3066     }
3067     guest_base = ret;
3068 }
3069 
3070 void probe_guest_base(const char *image_name, abi_ulong guest_loaddr,
3071                       abi_ulong guest_hiaddr)
3072 {
3073     /* In order to use host shmat, we must be able to honor SHMLBA.  */
3074     uintptr_t align = MAX(SHMLBA, qemu_host_page_size);
3075 
3076     /* Sanity check the guest binary. */
3077     if (reserved_va) {
3078         if (guest_hiaddr > reserved_va) {
3079             error_report("%s: requires more than reserved virtual "
3080                          "address space (0x%" PRIx64 " > 0x%lx)",
3081                          image_name, (uint64_t)guest_hiaddr, reserved_va);
3082             exit(EXIT_FAILURE);
3083         }
3084     } else {
3085         if (guest_hiaddr != (uintptr_t)guest_hiaddr) {
3086             error_report("%s: requires more virtual address space "
3087                          "than the host can provide (0x%" PRIx64 ")",
3088                          image_name, (uint64_t)guest_hiaddr + 1);
3089             exit(EXIT_FAILURE);
3090         }
3091     }
3092 
3093     if (have_guest_base) {
3094         pgb_fixed(image_name, guest_loaddr, guest_hiaddr, align);
3095     } else {
3096         pgb_dynamic(image_name, guest_loaddr, guest_hiaddr, align);
3097     }
3098 
3099     /* Reserve and initialize the commpage. */
3100     if (!init_guest_commpage()) {
3101         /* We have already probed for the commpage being free. */
3102         g_assert_not_reached();
3103     }
3104 
3105     assert(QEMU_IS_ALIGNED(guest_base, align));
3106     qemu_log_mask(CPU_LOG_PAGE, "Locating guest address space "
3107                   "@ 0x%" PRIx64 "\n", (uint64_t)guest_base);
3108 }
3109 
3110 enum {
3111     /* The string "GNU\0" as a magic number. */
3112     GNU0_MAGIC = const_le32('G' | 'N' << 8 | 'U' << 16),
3113     NOTE_DATA_SZ = 1 * KiB,
3114     NOTE_NAME_SZ = 4,
3115     ELF_GNU_PROPERTY_ALIGN = ELF_CLASS == ELFCLASS32 ? 4 : 8,
3116 };
3117 
3118 /*
3119  * Process a single gnu_property entry.
3120  * Return false for error.
3121  */
3122 static bool parse_elf_property(const uint32_t *data, int *off, int datasz,
3123                                struct image_info *info, bool have_prev_type,
3124                                uint32_t *prev_type, Error **errp)
3125 {
3126     uint32_t pr_type, pr_datasz, step;
3127 
3128     if (*off > datasz || !QEMU_IS_ALIGNED(*off, ELF_GNU_PROPERTY_ALIGN)) {
3129         goto error_data;
3130     }
3131     datasz -= *off;
3132     data += *off / sizeof(uint32_t);
3133 
3134     if (datasz < 2 * sizeof(uint32_t)) {
3135         goto error_data;
3136     }
3137     pr_type = data[0];
3138     pr_datasz = data[1];
3139     data += 2;
3140     datasz -= 2 * sizeof(uint32_t);
3141     step = ROUND_UP(pr_datasz, ELF_GNU_PROPERTY_ALIGN);
3142     if (step > datasz) {
3143         goto error_data;
3144     }
3145 
3146     /* Properties are supposed to be unique and sorted on pr_type. */
3147     if (have_prev_type && pr_type <= *prev_type) {
3148         if (pr_type == *prev_type) {
3149             error_setg(errp, "Duplicate property in PT_GNU_PROPERTY");
3150         } else {
3151             error_setg(errp, "Unsorted property in PT_GNU_PROPERTY");
3152         }
3153         return false;
3154     }
3155     *prev_type = pr_type;
3156 
3157     if (!arch_parse_elf_property(pr_type, pr_datasz, data, info, errp)) {
3158         return false;
3159     }
3160 
3161     *off += 2 * sizeof(uint32_t) + step;
3162     return true;
3163 
3164  error_data:
3165     error_setg(errp, "Ill-formed property in PT_GNU_PROPERTY");
3166     return false;
3167 }
3168 
3169 /* Process NT_GNU_PROPERTY_TYPE_0. */
3170 static bool parse_elf_properties(const ImageSource *src,
3171                                  struct image_info *info,
3172                                  const struct elf_phdr *phdr,
3173                                  Error **errp)
3174 {
3175     union {
3176         struct elf_note nhdr;
3177         uint32_t data[NOTE_DATA_SZ / sizeof(uint32_t)];
3178     } note;
3179 
3180     int n, off, datasz;
3181     bool have_prev_type;
3182     uint32_t prev_type;
3183 
3184     /* Unless the arch requires properties, ignore them. */
3185     if (!ARCH_USE_GNU_PROPERTY) {
3186         return true;
3187     }
3188 
3189     /* If the properties are crazy large, that's too bad. */
3190     n = phdr->p_filesz;
3191     if (n > sizeof(note)) {
3192         error_setg(errp, "PT_GNU_PROPERTY too large");
3193         return false;
3194     }
3195     if (n < sizeof(note.nhdr)) {
3196         error_setg(errp, "PT_GNU_PROPERTY too small");
3197         return false;
3198     }
3199 
3200     if (!imgsrc_read(&note, phdr->p_offset, n, src, errp)) {
3201         return false;
3202     }
3203 
3204     /*
3205      * The contents of a valid PT_GNU_PROPERTY is a sequence
3206      * of uint32_t -- swap them all now.
3207      */
3208 #ifdef BSWAP_NEEDED
3209     for (int i = 0; i < n / 4; i++) {
3210         bswap32s(note.data + i);
3211     }
3212 #endif
3213 
3214     /*
3215      * Note that nhdr is 3 words, and that the "name" described by namesz
3216      * immediately follows nhdr and is thus at the 4th word.  Further, all
3217      * of the inputs to the kernel's round_up are multiples of 4.
3218      */
3219     if (note.nhdr.n_type != NT_GNU_PROPERTY_TYPE_0 ||
3220         note.nhdr.n_namesz != NOTE_NAME_SZ ||
3221         note.data[3] != GNU0_MAGIC) {
3222         error_setg(errp, "Invalid note in PT_GNU_PROPERTY");
3223         return false;
3224     }
3225     off = sizeof(note.nhdr) + NOTE_NAME_SZ;
3226 
3227     datasz = note.nhdr.n_descsz + off;
3228     if (datasz > n) {
3229         error_setg(errp, "Invalid note size in PT_GNU_PROPERTY");
3230         return false;
3231     }
3232 
3233     have_prev_type = false;
3234     prev_type = 0;
3235     while (1) {
3236         if (off == datasz) {
3237             return true;  /* end, exit ok */
3238         }
3239         if (!parse_elf_property(note.data, &off, datasz, info,
3240                                 have_prev_type, &prev_type, errp)) {
3241             return false;
3242         }
3243         have_prev_type = true;
3244     }
3245 }
3246 
3247 /**
3248  * load_elf_image: Load an ELF image into the address space.
3249  * @image_name: the filename of the image, to use in error messages.
3250  * @src: the ImageSource from which to read.
3251  * @info: info collected from the loaded image.
3252  * @ehdr: the ELF header, not yet bswapped.
3253  * @pinterp_name: record any PT_INTERP string found.
3254  *
3255  * On return: @info values will be filled in, as necessary or available.
3256  */
3257 
3258 static void load_elf_image(const char *image_name, const ImageSource *src,
3259                            struct image_info *info, struct elfhdr *ehdr,
3260                            char **pinterp_name)
3261 {
3262     g_autofree struct elf_phdr *phdr = NULL;
3263     abi_ulong load_addr, load_bias, loaddr, hiaddr, error;
3264     int i, prot_exec;
3265     Error *err = NULL;
3266 
3267     /*
3268      * First of all, some simple consistency checks.
3269      * Note that we rely on the bswapped ehdr staying in bprm_buf,
3270      * for later use by load_elf_binary and create_elf_tables.
3271      */
3272     if (!imgsrc_read(ehdr, 0, sizeof(*ehdr), src, &err)) {
3273         goto exit_errmsg;
3274     }
3275     if (!elf_check_ident(ehdr)) {
3276         error_setg(&err, "Invalid ELF image for this architecture");
3277         goto exit_errmsg;
3278     }
3279     bswap_ehdr(ehdr);
3280     if (!elf_check_ehdr(ehdr)) {
3281         error_setg(&err, "Invalid ELF image for this architecture");
3282         goto exit_errmsg;
3283     }
3284 
3285     phdr = imgsrc_read_alloc(ehdr->e_phoff,
3286                              ehdr->e_phnum * sizeof(struct elf_phdr),
3287                              src, &err);
3288     if (phdr == NULL) {
3289         goto exit_errmsg;
3290     }
3291     bswap_phdr(phdr, ehdr->e_phnum);
3292 
3293     info->nsegs = 0;
3294     info->pt_dynamic_addr = 0;
3295 
3296     mmap_lock();
3297 
3298     /*
3299      * Find the maximum size of the image and allocate an appropriate
3300      * amount of memory to handle that.  Locate the interpreter, if any.
3301      */
3302     loaddr = -1, hiaddr = 0;
3303     info->alignment = 0;
3304     info->exec_stack = EXSTACK_DEFAULT;
3305     for (i = 0; i < ehdr->e_phnum; ++i) {
3306         struct elf_phdr *eppnt = phdr + i;
3307         if (eppnt->p_type == PT_LOAD) {
3308             abi_ulong a = eppnt->p_vaddr - eppnt->p_offset;
3309             if (a < loaddr) {
3310                 loaddr = a;
3311             }
3312             a = eppnt->p_vaddr + eppnt->p_memsz - 1;
3313             if (a > hiaddr) {
3314                 hiaddr = a;
3315             }
3316             ++info->nsegs;
3317             info->alignment |= eppnt->p_align;
3318         } else if (eppnt->p_type == PT_INTERP && pinterp_name) {
3319             g_autofree char *interp_name = NULL;
3320 
3321             if (*pinterp_name) {
3322                 error_setg(&err, "Multiple PT_INTERP entries");
3323                 goto exit_errmsg;
3324             }
3325 
3326             interp_name = imgsrc_read_alloc(eppnt->p_offset, eppnt->p_filesz,
3327                                             src, &err);
3328             if (interp_name == NULL) {
3329                 goto exit_errmsg;
3330             }
3331             if (interp_name[eppnt->p_filesz - 1] != 0) {
3332                 error_setg(&err, "Invalid PT_INTERP entry");
3333                 goto exit_errmsg;
3334             }
3335             *pinterp_name = g_steal_pointer(&interp_name);
3336         } else if (eppnt->p_type == PT_GNU_PROPERTY) {
3337             if (!parse_elf_properties(src, info, eppnt, &err)) {
3338                 goto exit_errmsg;
3339             }
3340         } else if (eppnt->p_type == PT_GNU_STACK) {
3341             info->exec_stack = eppnt->p_flags & PF_X;
3342         }
3343     }
3344 
3345     load_addr = loaddr;
3346 
3347     if (pinterp_name != NULL) {
3348         if (ehdr->e_type == ET_EXEC) {
3349             /*
3350              * Make sure that the low address does not conflict with
3351              * MMAP_MIN_ADDR or the QEMU application itself.
3352              */
3353             probe_guest_base(image_name, loaddr, hiaddr);
3354         } else {
3355             abi_ulong align;
3356 
3357             /*
3358              * The binary is dynamic, but we still need to
3359              * select guest_base.  In this case we pass a size.
3360              */
3361             probe_guest_base(image_name, 0, hiaddr - loaddr);
3362 
3363             /*
3364              * Avoid collision with the loader by providing a different
3365              * default load address.
3366              */
3367             load_addr += elf_et_dyn_base;
3368 
3369             /*
3370              * TODO: Better support for mmap alignment is desirable.
3371              * Since we do not have complete control over the guest
3372              * address space, we prefer the kernel to choose some address
3373              * rather than force the use of LOAD_ADDR via MAP_FIXED.
3374              * But without MAP_FIXED we cannot guarantee alignment,
3375              * only suggest it.
3376              */
3377             align = pow2ceil(info->alignment);
3378             if (align) {
3379                 load_addr &= -align;
3380             }
3381         }
3382     }
3383 
3384     /*
3385      * Reserve address space for all of this.
3386      *
3387      * In the case of ET_EXEC, we supply MAP_FIXED_NOREPLACE so that we get
3388      * exactly the address range that is required.  Without reserved_va,
3389      * the guest address space is not isolated.  We have attempted to avoid
3390      * conflict with the host program itself via probe_guest_base, but using
3391      * MAP_FIXED_NOREPLACE instead of MAP_FIXED provides an extra check.
3392      *
3393      * Otherwise this is ET_DYN, and we are searching for a location
3394      * that can hold the memory space required.  If the image is
3395      * pre-linked, LOAD_ADDR will be non-zero, and the kernel should
3396      * honor that address if it happens to be free.
3397      *
3398      * In both cases, we will overwrite pages in this range with mappings
3399      * from the executable.
3400      */
3401     load_addr = target_mmap(load_addr, (size_t)hiaddr - loaddr + 1, PROT_NONE,
3402                             MAP_PRIVATE | MAP_ANON | MAP_NORESERVE |
3403                             (ehdr->e_type == ET_EXEC ? MAP_FIXED_NOREPLACE : 0),
3404                             -1, 0);
3405     if (load_addr == -1) {
3406         goto exit_mmap;
3407     }
3408     load_bias = load_addr - loaddr;
3409 
3410     if (elf_is_fdpic(ehdr)) {
3411         struct elf32_fdpic_loadseg *loadsegs = info->loadsegs =
3412             g_malloc(sizeof(*loadsegs) * info->nsegs);
3413 
3414         for (i = 0; i < ehdr->e_phnum; ++i) {
3415             switch (phdr[i].p_type) {
3416             case PT_DYNAMIC:
3417                 info->pt_dynamic_addr = phdr[i].p_vaddr + load_bias;
3418                 break;
3419             case PT_LOAD:
3420                 loadsegs->addr = phdr[i].p_vaddr + load_bias;
3421                 loadsegs->p_vaddr = phdr[i].p_vaddr;
3422                 loadsegs->p_memsz = phdr[i].p_memsz;
3423                 ++loadsegs;
3424                 break;
3425             }
3426         }
3427     }
3428 
3429     info->load_bias = load_bias;
3430     info->code_offset = load_bias;
3431     info->data_offset = load_bias;
3432     info->load_addr = load_addr;
3433     info->entry = ehdr->e_entry + load_bias;
3434     info->start_code = -1;
3435     info->end_code = 0;
3436     info->start_data = -1;
3437     info->end_data = 0;
3438     /* Usual start for brk is after all sections of the main executable. */
3439     info->brk = TARGET_PAGE_ALIGN(hiaddr + load_bias);
3440     info->elf_flags = ehdr->e_flags;
3441 
3442     prot_exec = PROT_EXEC;
3443 #ifdef TARGET_AARCH64
3444     /*
3445      * If the BTI feature is present, this indicates that the executable
3446      * pages of the startup binary should be mapped with PROT_BTI, so that
3447      * branch targets are enforced.
3448      *
3449      * The startup binary is either the interpreter or the static executable.
3450      * The interpreter is responsible for all pages of a dynamic executable.
3451      *
3452      * Elf notes are backward compatible to older cpus.
3453      * Do not enable BTI unless it is supported.
3454      */
3455     if ((info->note_flags & GNU_PROPERTY_AARCH64_FEATURE_1_BTI)
3456         && (pinterp_name == NULL || *pinterp_name == 0)
3457         && cpu_isar_feature(aa64_bti, ARM_CPU(thread_cpu))) {
3458         prot_exec |= TARGET_PROT_BTI;
3459     }
3460 #endif
3461 
3462     for (i = 0; i < ehdr->e_phnum; i++) {
3463         struct elf_phdr *eppnt = phdr + i;
3464         if (eppnt->p_type == PT_LOAD) {
3465             abi_ulong vaddr, vaddr_po, vaddr_ps, vaddr_ef, vaddr_em;
3466             int elf_prot = 0;
3467 
3468             if (eppnt->p_flags & PF_R) {
3469                 elf_prot |= PROT_READ;
3470             }
3471             if (eppnt->p_flags & PF_W) {
3472                 elf_prot |= PROT_WRITE;
3473             }
3474             if (eppnt->p_flags & PF_X) {
3475                 elf_prot |= prot_exec;
3476             }
3477 
3478             vaddr = load_bias + eppnt->p_vaddr;
3479             vaddr_po = vaddr & ~TARGET_PAGE_MASK;
3480             vaddr_ps = vaddr & TARGET_PAGE_MASK;
3481 
3482             vaddr_ef = vaddr + eppnt->p_filesz;
3483             vaddr_em = vaddr + eppnt->p_memsz;
3484 
3485             /*
3486              * Some segments may be completely empty, with a non-zero p_memsz
3487              * but no backing file segment.
3488              */
3489             if (eppnt->p_filesz != 0) {
3490                 error = imgsrc_mmap(vaddr_ps, eppnt->p_filesz + vaddr_po,
3491                                     elf_prot, MAP_PRIVATE | MAP_FIXED,
3492                                     src, eppnt->p_offset - vaddr_po);
3493                 if (error == -1) {
3494                     goto exit_mmap;
3495                 }
3496             }
3497 
3498             /* If the load segment requests extra zeros (e.g. bss), map it. */
3499             if (vaddr_ef < vaddr_em &&
3500                 !zero_bss(vaddr_ef, vaddr_em, elf_prot, &err)) {
3501                 goto exit_errmsg;
3502             }
3503 
3504             /* Find the full program boundaries.  */
3505             if (elf_prot & PROT_EXEC) {
3506                 if (vaddr < info->start_code) {
3507                     info->start_code = vaddr;
3508                 }
3509                 if (vaddr_ef > info->end_code) {
3510                     info->end_code = vaddr_ef;
3511                 }
3512             }
3513             if (elf_prot & PROT_WRITE) {
3514                 if (vaddr < info->start_data) {
3515                     info->start_data = vaddr;
3516                 }
3517                 if (vaddr_ef > info->end_data) {
3518                     info->end_data = vaddr_ef;
3519                 }
3520             }
3521 #ifdef TARGET_MIPS
3522         } else if (eppnt->p_type == PT_MIPS_ABIFLAGS) {
3523             Mips_elf_abiflags_v0 abiflags;
3524 
3525             if (!imgsrc_read(&abiflags, eppnt->p_offset, sizeof(abiflags),
3526                              src, &err)) {
3527                 goto exit_errmsg;
3528             }
3529             bswap_mips_abiflags(&abiflags);
3530             info->fp_abi = abiflags.fp_abi;
3531 #endif
3532         }
3533     }
3534 
3535     if (info->end_data == 0) {
3536         info->start_data = info->end_code;
3537         info->end_data = info->end_code;
3538     }
3539 
3540     if (qemu_log_enabled()) {
3541         load_symbols(ehdr, src, load_bias);
3542     }
3543 
3544     debuginfo_report_elf(image_name, src->fd, load_bias);
3545 
3546     mmap_unlock();
3547 
3548     close(src->fd);
3549     return;
3550 
3551  exit_mmap:
3552     error_setg_errno(&err, errno, "Error mapping file");
3553     goto exit_errmsg;
3554  exit_errmsg:
3555     error_reportf_err(err, "%s: ", image_name);
3556     exit(-1);
3557 }
3558 
3559 static void load_elf_interp(const char *filename, struct image_info *info,
3560                             char bprm_buf[BPRM_BUF_SIZE])
3561 {
3562     struct elfhdr ehdr;
3563     ImageSource src;
3564     int fd, retval;
3565     Error *err = NULL;
3566 
3567     fd = open(path(filename), O_RDONLY);
3568     if (fd < 0) {
3569         error_setg_file_open(&err, errno, filename);
3570         error_report_err(err);
3571         exit(-1);
3572     }
3573 
3574     retval = read(fd, bprm_buf, BPRM_BUF_SIZE);
3575     if (retval < 0) {
3576         error_setg_errno(&err, errno, "Error reading file header");
3577         error_reportf_err(err, "%s: ", filename);
3578         exit(-1);
3579     }
3580 
3581     src.fd = fd;
3582     src.cache = bprm_buf;
3583     src.cache_size = retval;
3584 
3585     load_elf_image(filename, &src, info, &ehdr, NULL);
3586 }
3587 
3588 #ifdef VDSO_HEADER
3589 #include VDSO_HEADER
3590 #define  vdso_image_info()  &vdso_image_info
3591 #else
3592 #define  vdso_image_info()  NULL
3593 #endif
3594 
3595 static void load_elf_vdso(struct image_info *info, const VdsoImageInfo *vdso)
3596 {
3597     ImageSource src;
3598     struct elfhdr ehdr;
3599     abi_ulong load_bias, load_addr;
3600 
3601     src.fd = -1;
3602     src.cache = vdso->image;
3603     src.cache_size = vdso->image_size;
3604 
3605     load_elf_image("<internal-vdso>", &src, info, &ehdr, NULL);
3606     load_addr = info->load_addr;
3607     load_bias = info->load_bias;
3608 
3609     /*
3610      * We need to relocate the VDSO image.  The one built into the kernel
3611      * is built for a fixed address.  The one built for QEMU is not, since
3612      * that requires close control of the guest address space.
3613      * We pre-processed the image to locate all of the addresses that need
3614      * to be updated.
3615      */
3616     for (unsigned i = 0, n = vdso->reloc_count; i < n; i++) {
3617         abi_ulong *addr = g2h_untagged(load_addr + vdso->relocs[i]);
3618         *addr = tswapal(tswapal(*addr) + load_bias);
3619     }
3620 
3621     /* Install signal trampolines, if present. */
3622     if (vdso->sigreturn_ofs) {
3623         default_sigreturn = load_addr + vdso->sigreturn_ofs;
3624     }
3625     if (vdso->rt_sigreturn_ofs) {
3626         default_rt_sigreturn = load_addr + vdso->rt_sigreturn_ofs;
3627     }
3628 
3629     /* Remove write from VDSO segment. */
3630     target_mprotect(info->start_data, info->end_data - info->start_data,
3631                     PROT_READ | PROT_EXEC);
3632 }
3633 
3634 static int symfind(const void *s0, const void *s1)
3635 {
3636     struct elf_sym *sym = (struct elf_sym *)s1;
3637     __typeof(sym->st_value) addr = *(uint64_t *)s0;
3638     int result = 0;
3639 
3640     if (addr < sym->st_value) {
3641         result = -1;
3642     } else if (addr >= sym->st_value + sym->st_size) {
3643         result = 1;
3644     }
3645     return result;
3646 }
3647 
3648 static const char *lookup_symbolxx(struct syminfo *s, uint64_t orig_addr)
3649 {
3650 #if ELF_CLASS == ELFCLASS32
3651     struct elf_sym *syms = s->disas_symtab.elf32;
3652 #else
3653     struct elf_sym *syms = s->disas_symtab.elf64;
3654 #endif
3655 
3656     // binary search
3657     struct elf_sym *sym;
3658 
3659     sym = bsearch(&orig_addr, syms, s->disas_num_syms, sizeof(*syms), symfind);
3660     if (sym != NULL) {
3661         return s->disas_strtab + sym->st_name;
3662     }
3663 
3664     return "";
3665 }
3666 
3667 /* FIXME: This should use elf_ops.h  */
3668 static int symcmp(const void *s0, const void *s1)
3669 {
3670     struct elf_sym *sym0 = (struct elf_sym *)s0;
3671     struct elf_sym *sym1 = (struct elf_sym *)s1;
3672     return (sym0->st_value < sym1->st_value)
3673         ? -1
3674         : ((sym0->st_value > sym1->st_value) ? 1 : 0);
3675 }
3676 
3677 /* Best attempt to load symbols from this ELF object. */
3678 static void load_symbols(struct elfhdr *hdr, const ImageSource *src,
3679                          abi_ulong load_bias)
3680 {
3681     int i, shnum, nsyms, sym_idx = 0, str_idx = 0;
3682     g_autofree struct elf_shdr *shdr = NULL;
3683     char *strings = NULL;
3684     struct elf_sym *syms = NULL;
3685     struct elf_sym *new_syms;
3686     uint64_t segsz;
3687 
3688     shnum = hdr->e_shnum;
3689     shdr = imgsrc_read_alloc(hdr->e_shoff, shnum * sizeof(struct elf_shdr),
3690                              src, NULL);
3691     if (shdr == NULL) {
3692         return;
3693     }
3694 
3695     bswap_shdr(shdr, shnum);
3696     for (i = 0; i < shnum; ++i) {
3697         if (shdr[i].sh_type == SHT_SYMTAB) {
3698             sym_idx = i;
3699             str_idx = shdr[i].sh_link;
3700             goto found;
3701         }
3702     }
3703 
3704     /* There will be no symbol table if the file was stripped.  */
3705     return;
3706 
3707  found:
3708     /* Now know where the strtab and symtab are.  Snarf them.  */
3709 
3710     segsz = shdr[str_idx].sh_size;
3711     strings = g_try_malloc(segsz);
3712     if (!strings) {
3713         goto give_up;
3714     }
3715     if (!imgsrc_read(strings, shdr[str_idx].sh_offset, segsz, src, NULL)) {
3716         goto give_up;
3717     }
3718 
3719     segsz = shdr[sym_idx].sh_size;
3720     if (segsz / sizeof(struct elf_sym) > INT_MAX) {
3721         /*
3722          * Implausibly large symbol table: give up rather than ploughing
3723          * on with the number of symbols calculation overflowing.
3724          */
3725         goto give_up;
3726     }
3727     nsyms = segsz / sizeof(struct elf_sym);
3728     syms = g_try_malloc(segsz);
3729     if (!syms) {
3730         goto give_up;
3731     }
3732     if (!imgsrc_read(syms, shdr[sym_idx].sh_offset, segsz, src, NULL)) {
3733         goto give_up;
3734     }
3735 
3736     for (i = 0; i < nsyms; ) {
3737         bswap_sym(syms + i);
3738         /* Throw away entries which we do not need.  */
3739         if (syms[i].st_shndx == SHN_UNDEF
3740             || syms[i].st_shndx >= SHN_LORESERVE
3741             || ELF_ST_TYPE(syms[i].st_info) != STT_FUNC) {
3742             if (i < --nsyms) {
3743                 syms[i] = syms[nsyms];
3744             }
3745         } else {
3746 #if defined(TARGET_ARM) || defined (TARGET_MIPS)
3747             /* The bottom address bit marks a Thumb or MIPS16 symbol.  */
3748             syms[i].st_value &= ~(target_ulong)1;
3749 #endif
3750             syms[i].st_value += load_bias;
3751             i++;
3752         }
3753     }
3754 
3755     /* No "useful" symbol.  */
3756     if (nsyms == 0) {
3757         goto give_up;
3758     }
3759 
3760     /*
3761      * Attempt to free the storage associated with the local symbols
3762      * that we threw away.  Whether or not this has any effect on the
3763      * memory allocation depends on the malloc implementation and how
3764      * many symbols we managed to discard.
3765      */
3766     new_syms = g_try_renew(struct elf_sym, syms, nsyms);
3767     if (new_syms == NULL) {
3768         goto give_up;
3769     }
3770     syms = new_syms;
3771 
3772     qsort(syms, nsyms, sizeof(*syms), symcmp);
3773 
3774     {
3775         struct syminfo *s = g_new(struct syminfo, 1);
3776 
3777         s->disas_strtab = strings;
3778         s->disas_num_syms = nsyms;
3779 #if ELF_CLASS == ELFCLASS32
3780         s->disas_symtab.elf32 = syms;
3781 #else
3782         s->disas_symtab.elf64 = syms;
3783 #endif
3784         s->lookup_symbol = lookup_symbolxx;
3785         s->next = syminfos;
3786         syminfos = s;
3787     }
3788     return;
3789 
3790  give_up:
3791     g_free(strings);
3792     g_free(syms);
3793 }
3794 
3795 uint32_t get_elf_eflags(int fd)
3796 {
3797     struct elfhdr ehdr;
3798     off_t offset;
3799     int ret;
3800 
3801     /* Read ELF header */
3802     offset = lseek(fd, 0, SEEK_SET);
3803     if (offset == (off_t) -1) {
3804         return 0;
3805     }
3806     ret = read(fd, &ehdr, sizeof(ehdr));
3807     if (ret < sizeof(ehdr)) {
3808         return 0;
3809     }
3810     offset = lseek(fd, offset, SEEK_SET);
3811     if (offset == (off_t) -1) {
3812         return 0;
3813     }
3814 
3815     /* Check ELF signature */
3816     if (!elf_check_ident(&ehdr)) {
3817         return 0;
3818     }
3819 
3820     /* check header */
3821     bswap_ehdr(&ehdr);
3822     if (!elf_check_ehdr(&ehdr)) {
3823         return 0;
3824     }
3825 
3826     /* return architecture id */
3827     return ehdr.e_flags;
3828 }
3829 
3830 int load_elf_binary(struct linux_binprm *bprm, struct image_info *info)
3831 {
3832     /*
3833      * We need a copy of the elf header for passing to create_elf_tables.
3834      * We will have overwritten the original when we re-use bprm->buf
3835      * while loading the interpreter.  Allocate the storage for this now
3836      * and let elf_load_image do any swapping that may be required.
3837      */
3838     struct elfhdr ehdr;
3839     struct image_info interp_info, vdso_info;
3840     char *elf_interpreter = NULL;
3841     char *scratch;
3842 
3843     memset(&interp_info, 0, sizeof(interp_info));
3844 #ifdef TARGET_MIPS
3845     interp_info.fp_abi = MIPS_ABI_FP_UNKNOWN;
3846 #endif
3847 
3848     load_elf_image(bprm->filename, &bprm->src, info, &ehdr, &elf_interpreter);
3849 
3850     /* Do this so that we can load the interpreter, if need be.  We will
3851        change some of these later */
3852     bprm->p = setup_arg_pages(bprm, info);
3853 
3854     scratch = g_new0(char, TARGET_PAGE_SIZE);
3855     if (STACK_GROWS_DOWN) {
3856         bprm->p = copy_elf_strings(1, &bprm->filename, scratch,
3857                                    bprm->p, info->stack_limit);
3858         info->file_string = bprm->p;
3859         bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch,
3860                                    bprm->p, info->stack_limit);
3861         info->env_strings = bprm->p;
3862         bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch,
3863                                    bprm->p, info->stack_limit);
3864         info->arg_strings = bprm->p;
3865     } else {
3866         info->arg_strings = bprm->p;
3867         bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch,
3868                                    bprm->p, info->stack_limit);
3869         info->env_strings = bprm->p;
3870         bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch,
3871                                    bprm->p, info->stack_limit);
3872         info->file_string = bprm->p;
3873         bprm->p = copy_elf_strings(1, &bprm->filename, scratch,
3874                                    bprm->p, info->stack_limit);
3875     }
3876 
3877     g_free(scratch);
3878 
3879     if (!bprm->p) {
3880         fprintf(stderr, "%s: %s\n", bprm->filename, strerror(E2BIG));
3881         exit(-1);
3882     }
3883 
3884     if (elf_interpreter) {
3885         load_elf_interp(elf_interpreter, &interp_info, bprm->buf);
3886 
3887         /*
3888          * While unusual because of ELF_ET_DYN_BASE, if we are unlucky
3889          * with the mappings the interpreter can be loaded above but
3890          * near the main executable, which can leave very little room
3891          * for the heap.
3892          * If the current brk has less than 16MB, use the end of the
3893          * interpreter.
3894          */
3895         if (interp_info.brk > info->brk &&
3896             interp_info.load_bias - info->brk < 16 * MiB)  {
3897             info->brk = interp_info.brk;
3898         }
3899 
3900         /* If the program interpreter is one of these two, then assume
3901            an iBCS2 image.  Otherwise assume a native linux image.  */
3902 
3903         if (strcmp(elf_interpreter, "/usr/lib/libc.so.1") == 0
3904             || strcmp(elf_interpreter, "/usr/lib/ld.so.1") == 0) {
3905             info->personality = PER_SVR4;
3906 
3907             /* Why this, you ask???  Well SVr4 maps page 0 as read-only,
3908                and some applications "depend" upon this behavior.  Since
3909                we do not have the power to recompile these, we emulate
3910                the SVr4 behavior.  Sigh.  */
3911             target_mmap(0, qemu_host_page_size, PROT_READ | PROT_EXEC,
3912                         MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
3913         }
3914 #ifdef TARGET_MIPS
3915         info->interp_fp_abi = interp_info.fp_abi;
3916 #endif
3917     }
3918 
3919     /*
3920      * Load a vdso if available, which will amongst other things contain the
3921      * signal trampolines.  Otherwise, allocate a separate page for them.
3922      */
3923     const VdsoImageInfo *vdso = vdso_image_info();
3924     if (vdso) {
3925         load_elf_vdso(&vdso_info, vdso);
3926         info->vdso = vdso_info.load_bias;
3927     } else if (TARGET_ARCH_HAS_SIGTRAMP_PAGE) {
3928         abi_long tramp_page = target_mmap(0, TARGET_PAGE_SIZE,
3929                                           PROT_READ | PROT_WRITE,
3930                                           MAP_PRIVATE | MAP_ANON, -1, 0);
3931         if (tramp_page == -1) {
3932             return -errno;
3933         }
3934 
3935         setup_sigtramp(tramp_page);
3936         target_mprotect(tramp_page, TARGET_PAGE_SIZE, PROT_READ | PROT_EXEC);
3937     }
3938 
3939     bprm->p = create_elf_tables(bprm->p, bprm->argc, bprm->envc, &ehdr, info,
3940                                 elf_interpreter ? &interp_info : NULL,
3941                                 vdso ? &vdso_info : NULL);
3942     info->start_stack = bprm->p;
3943 
3944     /* If we have an interpreter, set that as the program's entry point.
3945        Copy the load_bias as well, to help PPC64 interpret the entry
3946        point as a function descriptor.  Do this after creating elf tables
3947        so that we copy the original program entry point into the AUXV.  */
3948     if (elf_interpreter) {
3949         info->load_bias = interp_info.load_bias;
3950         info->entry = interp_info.entry;
3951         g_free(elf_interpreter);
3952     }
3953 
3954 #ifdef USE_ELF_CORE_DUMP
3955     bprm->core_dump = &elf_core_dump;
3956 #endif
3957 
3958     return 0;
3959 }
3960 
3961 #ifdef USE_ELF_CORE_DUMP
3962 /*
3963  * Definitions to generate Intel SVR4-like core files.
3964  * These mostly have the same names as the SVR4 types with "target_elf_"
3965  * tacked on the front to prevent clashes with linux definitions,
3966  * and the typedef forms have been avoided.  This is mostly like
3967  * the SVR4 structure, but more Linuxy, with things that Linux does
3968  * not support and which gdb doesn't really use excluded.
3969  *
3970  * Fields we don't dump (their contents is zero) in linux-user qemu
3971  * are marked with XXX.
3972  *
3973  * Core dump code is copied from linux kernel (fs/binfmt_elf.c).
3974  *
3975  * Porting ELF coredump for target is (quite) simple process.  First you
3976  * define USE_ELF_CORE_DUMP in target ELF code (where init_thread() for
3977  * the target resides):
3978  *
3979  * #define USE_ELF_CORE_DUMP
3980  *
3981  * Next you define type of register set used for dumping.  ELF specification
3982  * says that it needs to be array of elf_greg_t that has size of ELF_NREG.
3983  *
3984  * typedef <target_regtype> target_elf_greg_t;
3985  * #define ELF_NREG <number of registers>
3986  * typedef taret_elf_greg_t target_elf_gregset_t[ELF_NREG];
3987  *
3988  * Last step is to implement target specific function that copies registers
3989  * from given cpu into just specified register set.  Prototype is:
3990  *
3991  * static void elf_core_copy_regs(taret_elf_gregset_t *regs,
3992  *                                const CPUArchState *env);
3993  *
3994  * Parameters:
3995  *     regs - copy register values into here (allocated and zeroed by caller)
3996  *     env - copy registers from here
3997  *
3998  * Example for ARM target is provided in this file.
3999  */
4000 
4001 /* An ELF note in memory */
4002 struct memelfnote {
4003     const char *name;
4004     size_t     namesz;
4005     size_t     namesz_rounded;
4006     int        type;
4007     size_t     datasz;
4008     size_t     datasz_rounded;
4009     void       *data;
4010     size_t     notesz;
4011 };
4012 
4013 struct target_elf_siginfo {
4014     abi_int    si_signo; /* signal number */
4015     abi_int    si_code;  /* extra code */
4016     abi_int    si_errno; /* errno */
4017 };
4018 
4019 struct target_elf_prstatus {
4020     struct target_elf_siginfo pr_info;      /* Info associated with signal */
4021     abi_short          pr_cursig;    /* Current signal */
4022     abi_ulong          pr_sigpend;   /* XXX */
4023     abi_ulong          pr_sighold;   /* XXX */
4024     target_pid_t       pr_pid;
4025     target_pid_t       pr_ppid;
4026     target_pid_t       pr_pgrp;
4027     target_pid_t       pr_sid;
4028     struct target_timeval pr_utime;  /* XXX User time */
4029     struct target_timeval pr_stime;  /* XXX System time */
4030     struct target_timeval pr_cutime; /* XXX Cumulative user time */
4031     struct target_timeval pr_cstime; /* XXX Cumulative system time */
4032     target_elf_gregset_t      pr_reg;       /* GP registers */
4033     abi_int            pr_fpvalid;   /* XXX */
4034 };
4035 
4036 #define ELF_PRARGSZ     (80) /* Number of chars for args */
4037 
4038 struct target_elf_prpsinfo {
4039     char         pr_state;       /* numeric process state */
4040     char         pr_sname;       /* char for pr_state */
4041     char         pr_zomb;        /* zombie */
4042     char         pr_nice;        /* nice val */
4043     abi_ulong    pr_flag;        /* flags */
4044     target_uid_t pr_uid;
4045     target_gid_t pr_gid;
4046     target_pid_t pr_pid, pr_ppid, pr_pgrp, pr_sid;
4047     /* Lots missing */
4048     char    pr_fname[16] QEMU_NONSTRING; /* filename of executable */
4049     char    pr_psargs[ELF_PRARGSZ]; /* initial part of arg list */
4050 };
4051 
4052 /* Here is the structure in which status of each thread is captured. */
4053 struct elf_thread_status {
4054     QTAILQ_ENTRY(elf_thread_status)  ets_link;
4055     struct target_elf_prstatus prstatus;   /* NT_PRSTATUS */
4056 #if 0
4057     elf_fpregset_t fpu;             /* NT_PRFPREG */
4058     struct task_struct *thread;
4059     elf_fpxregset_t xfpu;           /* ELF_CORE_XFPREG_TYPE */
4060 #endif
4061     struct memelfnote notes[1];
4062     int num_notes;
4063 };
4064 
4065 struct elf_note_info {
4066     struct memelfnote   *notes;
4067     struct target_elf_prstatus *prstatus;  /* NT_PRSTATUS */
4068     struct target_elf_prpsinfo *psinfo;    /* NT_PRPSINFO */
4069 
4070     QTAILQ_HEAD(, elf_thread_status) thread_list;
4071 #if 0
4072     /*
4073      * Current version of ELF coredump doesn't support
4074      * dumping fp regs etc.
4075      */
4076     elf_fpregset_t *fpu;
4077     elf_fpxregset_t *xfpu;
4078     int thread_status_size;
4079 #endif
4080     int notes_size;
4081     int numnote;
4082 };
4083 
4084 struct vm_area_struct {
4085     target_ulong   vma_start;  /* start vaddr of memory region */
4086     target_ulong   vma_end;    /* end vaddr of memory region */
4087     abi_ulong      vma_flags;  /* protection etc. flags for the region */
4088     QTAILQ_ENTRY(vm_area_struct) vma_link;
4089 };
4090 
4091 struct mm_struct {
4092     QTAILQ_HEAD(, vm_area_struct) mm_mmap;
4093     int mm_count;           /* number of mappings */
4094 };
4095 
4096 static struct mm_struct *vma_init(void);
4097 static void vma_delete(struct mm_struct *);
4098 static int vma_add_mapping(struct mm_struct *, target_ulong,
4099                            target_ulong, abi_ulong);
4100 static int vma_get_mapping_count(const struct mm_struct *);
4101 static struct vm_area_struct *vma_first(const struct mm_struct *);
4102 static struct vm_area_struct *vma_next(struct vm_area_struct *);
4103 static abi_ulong vma_dump_size(const struct vm_area_struct *);
4104 static int vma_walker(void *priv, target_ulong start, target_ulong end,
4105                       unsigned long flags);
4106 
4107 static void fill_elf_header(struct elfhdr *, int, uint16_t, uint32_t);
4108 static void fill_note(struct memelfnote *, const char *, int,
4109                       unsigned int, void *);
4110 static void fill_prstatus(struct target_elf_prstatus *, const TaskState *, int);
4111 static int fill_psinfo(struct target_elf_prpsinfo *, const TaskState *);
4112 static void fill_auxv_note(struct memelfnote *, const TaskState *);
4113 static void fill_elf_note_phdr(struct elf_phdr *, int, off_t);
4114 static size_t note_size(const struct memelfnote *);
4115 static void free_note_info(struct elf_note_info *);
4116 static int fill_note_info(struct elf_note_info *, long, const CPUArchState *);
4117 static void fill_thread_info(struct elf_note_info *, const CPUArchState *);
4118 
4119 static int dump_write(int, const void *, size_t);
4120 static int write_note(struct memelfnote *, int);
4121 static int write_note_info(struct elf_note_info *, int);
4122 
4123 #ifdef BSWAP_NEEDED
4124 static void bswap_prstatus(struct target_elf_prstatus *prstatus)
4125 {
4126     prstatus->pr_info.si_signo = tswap32(prstatus->pr_info.si_signo);
4127     prstatus->pr_info.si_code = tswap32(prstatus->pr_info.si_code);
4128     prstatus->pr_info.si_errno = tswap32(prstatus->pr_info.si_errno);
4129     prstatus->pr_cursig = tswap16(prstatus->pr_cursig);
4130     prstatus->pr_sigpend = tswapal(prstatus->pr_sigpend);
4131     prstatus->pr_sighold = tswapal(prstatus->pr_sighold);
4132     prstatus->pr_pid = tswap32(prstatus->pr_pid);
4133     prstatus->pr_ppid = tswap32(prstatus->pr_ppid);
4134     prstatus->pr_pgrp = tswap32(prstatus->pr_pgrp);
4135     prstatus->pr_sid = tswap32(prstatus->pr_sid);
4136     /* cpu times are not filled, so we skip them */
4137     /* regs should be in correct format already */
4138     prstatus->pr_fpvalid = tswap32(prstatus->pr_fpvalid);
4139 }
4140 
4141 static void bswap_psinfo(struct target_elf_prpsinfo *psinfo)
4142 {
4143     psinfo->pr_flag = tswapal(psinfo->pr_flag);
4144     psinfo->pr_uid = tswap16(psinfo->pr_uid);
4145     psinfo->pr_gid = tswap16(psinfo->pr_gid);
4146     psinfo->pr_pid = tswap32(psinfo->pr_pid);
4147     psinfo->pr_ppid = tswap32(psinfo->pr_ppid);
4148     psinfo->pr_pgrp = tswap32(psinfo->pr_pgrp);
4149     psinfo->pr_sid = tswap32(psinfo->pr_sid);
4150 }
4151 
4152 static void bswap_note(struct elf_note *en)
4153 {
4154     bswap32s(&en->n_namesz);
4155     bswap32s(&en->n_descsz);
4156     bswap32s(&en->n_type);
4157 }
4158 #else
4159 static inline void bswap_prstatus(struct target_elf_prstatus *p) { }
4160 static inline void bswap_psinfo(struct target_elf_prpsinfo *p) {}
4161 static inline void bswap_note(struct elf_note *en) { }
4162 #endif /* BSWAP_NEEDED */
4163 
4164 /*
4165  * Minimal support for linux memory regions.  These are needed
4166  * when we are finding out what memory exactly belongs to
4167  * emulated process.  No locks needed here, as long as
4168  * thread that received the signal is stopped.
4169  */
4170 
4171 static struct mm_struct *vma_init(void)
4172 {
4173     struct mm_struct *mm;
4174 
4175     if ((mm = g_malloc(sizeof (*mm))) == NULL)
4176         return (NULL);
4177 
4178     mm->mm_count = 0;
4179     QTAILQ_INIT(&mm->mm_mmap);
4180 
4181     return (mm);
4182 }
4183 
4184 static void vma_delete(struct mm_struct *mm)
4185 {
4186     struct vm_area_struct *vma;
4187 
4188     while ((vma = vma_first(mm)) != NULL) {
4189         QTAILQ_REMOVE(&mm->mm_mmap, vma, vma_link);
4190         g_free(vma);
4191     }
4192     g_free(mm);
4193 }
4194 
4195 static int vma_add_mapping(struct mm_struct *mm, target_ulong start,
4196                            target_ulong end, abi_ulong flags)
4197 {
4198     struct vm_area_struct *vma;
4199 
4200     if ((vma = g_malloc0(sizeof (*vma))) == NULL)
4201         return (-1);
4202 
4203     vma->vma_start = start;
4204     vma->vma_end = end;
4205     vma->vma_flags = flags;
4206 
4207     QTAILQ_INSERT_TAIL(&mm->mm_mmap, vma, vma_link);
4208     mm->mm_count++;
4209 
4210     return (0);
4211 }
4212 
4213 static struct vm_area_struct *vma_first(const struct mm_struct *mm)
4214 {
4215     return (QTAILQ_FIRST(&mm->mm_mmap));
4216 }
4217 
4218 static struct vm_area_struct *vma_next(struct vm_area_struct *vma)
4219 {
4220     return (QTAILQ_NEXT(vma, vma_link));
4221 }
4222 
4223 static int vma_get_mapping_count(const struct mm_struct *mm)
4224 {
4225     return (mm->mm_count);
4226 }
4227 
4228 /*
4229  * Calculate file (dump) size of given memory region.
4230  */
4231 static abi_ulong vma_dump_size(const struct vm_area_struct *vma)
4232 {
4233     /* if we cannot even read the first page, skip it */
4234     if (!access_ok_untagged(VERIFY_READ, vma->vma_start, TARGET_PAGE_SIZE))
4235         return (0);
4236 
4237     /*
4238      * Usually we don't dump executable pages as they contain
4239      * non-writable code that debugger can read directly from
4240      * target library etc.  However, thread stacks are marked
4241      * also executable so we read in first page of given region
4242      * and check whether it contains elf header.  If there is
4243      * no elf header, we dump it.
4244      */
4245     if (vma->vma_flags & PROT_EXEC) {
4246         char page[TARGET_PAGE_SIZE];
4247 
4248         if (copy_from_user(page, vma->vma_start, sizeof (page))) {
4249             return 0;
4250         }
4251         if ((page[EI_MAG0] == ELFMAG0) &&
4252             (page[EI_MAG1] == ELFMAG1) &&
4253             (page[EI_MAG2] == ELFMAG2) &&
4254             (page[EI_MAG3] == ELFMAG3)) {
4255             /*
4256              * Mappings are possibly from ELF binary.  Don't dump
4257              * them.
4258              */
4259             return (0);
4260         }
4261     }
4262 
4263     return (vma->vma_end - vma->vma_start);
4264 }
4265 
4266 static int vma_walker(void *priv, target_ulong start, target_ulong end,
4267                       unsigned long flags)
4268 {
4269     struct mm_struct *mm = (struct mm_struct *)priv;
4270 
4271     vma_add_mapping(mm, start, end, flags);
4272     return (0);
4273 }
4274 
4275 static void fill_note(struct memelfnote *note, const char *name, int type,
4276                       unsigned int sz, void *data)
4277 {
4278     unsigned int namesz;
4279 
4280     namesz = strlen(name) + 1;
4281     note->name = name;
4282     note->namesz = namesz;
4283     note->namesz_rounded = roundup(namesz, sizeof (int32_t));
4284     note->type = type;
4285     note->datasz = sz;
4286     note->datasz_rounded = roundup(sz, sizeof (int32_t));
4287 
4288     note->data = data;
4289 
4290     /*
4291      * We calculate rounded up note size here as specified by
4292      * ELF document.
4293      */
4294     note->notesz = sizeof (struct elf_note) +
4295         note->namesz_rounded + note->datasz_rounded;
4296 }
4297 
4298 static void fill_elf_header(struct elfhdr *elf, int segs, uint16_t machine,
4299                             uint32_t flags)
4300 {
4301     (void) memset(elf, 0, sizeof(*elf));
4302 
4303     (void) memcpy(elf->e_ident, ELFMAG, SELFMAG);
4304     elf->e_ident[EI_CLASS] = ELF_CLASS;
4305     elf->e_ident[EI_DATA] = ELF_DATA;
4306     elf->e_ident[EI_VERSION] = EV_CURRENT;
4307     elf->e_ident[EI_OSABI] = ELF_OSABI;
4308 
4309     elf->e_type = ET_CORE;
4310     elf->e_machine = machine;
4311     elf->e_version = EV_CURRENT;
4312     elf->e_phoff = sizeof(struct elfhdr);
4313     elf->e_flags = flags;
4314     elf->e_ehsize = sizeof(struct elfhdr);
4315     elf->e_phentsize = sizeof(struct elf_phdr);
4316     elf->e_phnum = segs;
4317 
4318     bswap_ehdr(elf);
4319 }
4320 
4321 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, off_t offset)
4322 {
4323     phdr->p_type = PT_NOTE;
4324     phdr->p_offset = offset;
4325     phdr->p_vaddr = 0;
4326     phdr->p_paddr = 0;
4327     phdr->p_filesz = sz;
4328     phdr->p_memsz = 0;
4329     phdr->p_flags = 0;
4330     phdr->p_align = 0;
4331 
4332     bswap_phdr(phdr, 1);
4333 }
4334 
4335 static size_t note_size(const struct memelfnote *note)
4336 {
4337     return (note->notesz);
4338 }
4339 
4340 static void fill_prstatus(struct target_elf_prstatus *prstatus,
4341                           const TaskState *ts, int signr)
4342 {
4343     (void) memset(prstatus, 0, sizeof (*prstatus));
4344     prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
4345     prstatus->pr_pid = ts->ts_tid;
4346     prstatus->pr_ppid = getppid();
4347     prstatus->pr_pgrp = getpgrp();
4348     prstatus->pr_sid = getsid(0);
4349 
4350     bswap_prstatus(prstatus);
4351 }
4352 
4353 static int fill_psinfo(struct target_elf_prpsinfo *psinfo, const TaskState *ts)
4354 {
4355     char *base_filename;
4356     unsigned int i, len;
4357 
4358     (void) memset(psinfo, 0, sizeof (*psinfo));
4359 
4360     len = ts->info->env_strings - ts->info->arg_strings;
4361     if (len >= ELF_PRARGSZ)
4362         len = ELF_PRARGSZ - 1;
4363     if (copy_from_user(&psinfo->pr_psargs, ts->info->arg_strings, len)) {
4364         return -EFAULT;
4365     }
4366     for (i = 0; i < len; i++)
4367         if (psinfo->pr_psargs[i] == 0)
4368             psinfo->pr_psargs[i] = ' ';
4369     psinfo->pr_psargs[len] = 0;
4370 
4371     psinfo->pr_pid = getpid();
4372     psinfo->pr_ppid = getppid();
4373     psinfo->pr_pgrp = getpgrp();
4374     psinfo->pr_sid = getsid(0);
4375     psinfo->pr_uid = getuid();
4376     psinfo->pr_gid = getgid();
4377 
4378     base_filename = g_path_get_basename(ts->bprm->filename);
4379     /*
4380      * Using strncpy here is fine: at max-length,
4381      * this field is not NUL-terminated.
4382      */
4383     (void) strncpy(psinfo->pr_fname, base_filename,
4384                    sizeof(psinfo->pr_fname));
4385 
4386     g_free(base_filename);
4387     bswap_psinfo(psinfo);
4388     return (0);
4389 }
4390 
4391 static void fill_auxv_note(struct memelfnote *note, const TaskState *ts)
4392 {
4393     elf_addr_t auxv = (elf_addr_t)ts->info->saved_auxv;
4394     elf_addr_t orig_auxv = auxv;
4395     void *ptr;
4396     int len = ts->info->auxv_len;
4397 
4398     /*
4399      * Auxiliary vector is stored in target process stack.  It contains
4400      * {type, value} pairs that we need to dump into note.  This is not
4401      * strictly necessary but we do it here for sake of completeness.
4402      */
4403 
4404     /* read in whole auxv vector and copy it to memelfnote */
4405     ptr = lock_user(VERIFY_READ, orig_auxv, len, 0);
4406     if (ptr != NULL) {
4407         fill_note(note, "CORE", NT_AUXV, len, ptr);
4408         unlock_user(ptr, auxv, len);
4409     }
4410 }
4411 
4412 /*
4413  * Constructs name of coredump file.  We have following convention
4414  * for the name:
4415  *     qemu_<basename-of-target-binary>_<date>-<time>_<pid>.core
4416  *
4417  * Returns the filename
4418  */
4419 static char *core_dump_filename(const TaskState *ts)
4420 {
4421     g_autoptr(GDateTime) now = g_date_time_new_now_local();
4422     g_autofree char *nowstr = g_date_time_format(now, "%Y%m%d-%H%M%S");
4423     g_autofree char *base_filename = g_path_get_basename(ts->bprm->filename);
4424 
4425     return g_strdup_printf("qemu_%s_%s_%d.core",
4426                            base_filename, nowstr, (int)getpid());
4427 }
4428 
4429 static int dump_write(int fd, const void *ptr, size_t size)
4430 {
4431     const char *bufp = (const char *)ptr;
4432     ssize_t bytes_written, bytes_left;
4433     struct rlimit dumpsize;
4434     off_t pos;
4435 
4436     bytes_written = 0;
4437     getrlimit(RLIMIT_CORE, &dumpsize);
4438     if ((pos = lseek(fd, 0, SEEK_CUR))==-1) {
4439         if (errno == ESPIPE) { /* not a seekable stream */
4440             bytes_left = size;
4441         } else {
4442             return pos;
4443         }
4444     } else {
4445         if (dumpsize.rlim_cur <= pos) {
4446             return -1;
4447         } else if (dumpsize.rlim_cur == RLIM_INFINITY) {
4448             bytes_left = size;
4449         } else {
4450             size_t limit_left=dumpsize.rlim_cur - pos;
4451             bytes_left = limit_left >= size ? size : limit_left ;
4452         }
4453     }
4454 
4455     /*
4456      * In normal conditions, single write(2) should do but
4457      * in case of socket etc. this mechanism is more portable.
4458      */
4459     do {
4460         bytes_written = write(fd, bufp, bytes_left);
4461         if (bytes_written < 0) {
4462             if (errno == EINTR)
4463                 continue;
4464             return (-1);
4465         } else if (bytes_written == 0) { /* eof */
4466             return (-1);
4467         }
4468         bufp += bytes_written;
4469         bytes_left -= bytes_written;
4470     } while (bytes_left > 0);
4471 
4472     return (0);
4473 }
4474 
4475 static int write_note(struct memelfnote *men, int fd)
4476 {
4477     struct elf_note en;
4478 
4479     en.n_namesz = men->namesz;
4480     en.n_type = men->type;
4481     en.n_descsz = men->datasz;
4482 
4483     bswap_note(&en);
4484 
4485     if (dump_write(fd, &en, sizeof(en)) != 0)
4486         return (-1);
4487     if (dump_write(fd, men->name, men->namesz_rounded) != 0)
4488         return (-1);
4489     if (dump_write(fd, men->data, men->datasz_rounded) != 0)
4490         return (-1);
4491 
4492     return (0);
4493 }
4494 
4495 static void fill_thread_info(struct elf_note_info *info, const CPUArchState *env)
4496 {
4497     CPUState *cpu = env_cpu((CPUArchState *)env);
4498     TaskState *ts = (TaskState *)cpu->opaque;
4499     struct elf_thread_status *ets;
4500 
4501     ets = g_malloc0(sizeof (*ets));
4502     ets->num_notes = 1; /* only prstatus is dumped */
4503     fill_prstatus(&ets->prstatus, ts, 0);
4504     elf_core_copy_regs(&ets->prstatus.pr_reg, env);
4505     fill_note(&ets->notes[0], "CORE", NT_PRSTATUS, sizeof (ets->prstatus),
4506               &ets->prstatus);
4507 
4508     QTAILQ_INSERT_TAIL(&info->thread_list, ets, ets_link);
4509 
4510     info->notes_size += note_size(&ets->notes[0]);
4511 }
4512 
4513 static void init_note_info(struct elf_note_info *info)
4514 {
4515     /* Initialize the elf_note_info structure so that it is at
4516      * least safe to call free_note_info() on it. Must be
4517      * called before calling fill_note_info().
4518      */
4519     memset(info, 0, sizeof (*info));
4520     QTAILQ_INIT(&info->thread_list);
4521 }
4522 
4523 static int fill_note_info(struct elf_note_info *info,
4524                           long signr, const CPUArchState *env)
4525 {
4526 #define NUMNOTES 3
4527     CPUState *cpu = env_cpu((CPUArchState *)env);
4528     TaskState *ts = (TaskState *)cpu->opaque;
4529     int i;
4530 
4531     info->notes = g_new0(struct memelfnote, NUMNOTES);
4532     if (info->notes == NULL)
4533         return (-ENOMEM);
4534     info->prstatus = g_malloc0(sizeof (*info->prstatus));
4535     if (info->prstatus == NULL)
4536         return (-ENOMEM);
4537     info->psinfo = g_malloc0(sizeof (*info->psinfo));
4538     if (info->prstatus == NULL)
4539         return (-ENOMEM);
4540 
4541     /*
4542      * First fill in status (and registers) of current thread
4543      * including process info & aux vector.
4544      */
4545     fill_prstatus(info->prstatus, ts, signr);
4546     elf_core_copy_regs(&info->prstatus->pr_reg, env);
4547     fill_note(&info->notes[0], "CORE", NT_PRSTATUS,
4548               sizeof (*info->prstatus), info->prstatus);
4549     fill_psinfo(info->psinfo, ts);
4550     fill_note(&info->notes[1], "CORE", NT_PRPSINFO,
4551               sizeof (*info->psinfo), info->psinfo);
4552     fill_auxv_note(&info->notes[2], ts);
4553     info->numnote = 3;
4554 
4555     info->notes_size = 0;
4556     for (i = 0; i < info->numnote; i++)
4557         info->notes_size += note_size(&info->notes[i]);
4558 
4559     /* read and fill status of all threads */
4560     WITH_QEMU_LOCK_GUARD(&qemu_cpu_list_lock) {
4561         CPU_FOREACH(cpu) {
4562             if (cpu == thread_cpu) {
4563                 continue;
4564             }
4565             fill_thread_info(info, cpu_env(cpu));
4566         }
4567     }
4568 
4569     return (0);
4570 }
4571 
4572 static void free_note_info(struct elf_note_info *info)
4573 {
4574     struct elf_thread_status *ets;
4575 
4576     while (!QTAILQ_EMPTY(&info->thread_list)) {
4577         ets = QTAILQ_FIRST(&info->thread_list);
4578         QTAILQ_REMOVE(&info->thread_list, ets, ets_link);
4579         g_free(ets);
4580     }
4581 
4582     g_free(info->prstatus);
4583     g_free(info->psinfo);
4584     g_free(info->notes);
4585 }
4586 
4587 static int write_note_info(struct elf_note_info *info, int fd)
4588 {
4589     struct elf_thread_status *ets;
4590     int i, error = 0;
4591 
4592     /* write prstatus, psinfo and auxv for current thread */
4593     for (i = 0; i < info->numnote; i++)
4594         if ((error = write_note(&info->notes[i], fd)) != 0)
4595             return (error);
4596 
4597     /* write prstatus for each thread */
4598     QTAILQ_FOREACH(ets, &info->thread_list, ets_link) {
4599         if ((error = write_note(&ets->notes[0], fd)) != 0)
4600             return (error);
4601     }
4602 
4603     return (0);
4604 }
4605 
4606 /*
4607  * Write out ELF coredump.
4608  *
4609  * See documentation of ELF object file format in:
4610  * http://www.caldera.com/developers/devspecs/gabi41.pdf
4611  *
4612  * Coredump format in linux is following:
4613  *
4614  * 0   +----------------------+         \
4615  *     | ELF header           | ET_CORE  |
4616  *     +----------------------+          |
4617  *     | ELF program headers  |          |--- headers
4618  *     | - NOTE section       |          |
4619  *     | - PT_LOAD sections   |          |
4620  *     +----------------------+         /
4621  *     | NOTEs:               |
4622  *     | - NT_PRSTATUS        |
4623  *     | - NT_PRSINFO         |
4624  *     | - NT_AUXV            |
4625  *     +----------------------+ <-- aligned to target page
4626  *     | Process memory dump  |
4627  *     :                      :
4628  *     .                      .
4629  *     :                      :
4630  *     |                      |
4631  *     +----------------------+
4632  *
4633  * NT_PRSTATUS -> struct elf_prstatus (per thread)
4634  * NT_PRSINFO  -> struct elf_prpsinfo
4635  * NT_AUXV is array of { type, value } pairs (see fill_auxv_note()).
4636  *
4637  * Format follows System V format as close as possible.  Current
4638  * version limitations are as follows:
4639  *     - no floating point registers are dumped
4640  *
4641  * Function returns 0 in case of success, negative errno otherwise.
4642  *
4643  * TODO: make this work also during runtime: it should be
4644  * possible to force coredump from running process and then
4645  * continue processing.  For example qemu could set up SIGUSR2
4646  * handler (provided that target process haven't registered
4647  * handler for that) that does the dump when signal is received.
4648  */
4649 static int elf_core_dump(int signr, const CPUArchState *env)
4650 {
4651     const CPUState *cpu = env_cpu((CPUArchState *)env);
4652     const TaskState *ts = (const TaskState *)cpu->opaque;
4653     struct vm_area_struct *vma = NULL;
4654     g_autofree char *corefile = NULL;
4655     struct elf_note_info info;
4656     struct elfhdr elf;
4657     struct elf_phdr phdr;
4658     struct rlimit dumpsize;
4659     struct mm_struct *mm = NULL;
4660     off_t offset = 0, data_offset = 0;
4661     int segs = 0;
4662     int fd = -1;
4663 
4664     init_note_info(&info);
4665 
4666     errno = 0;
4667     getrlimit(RLIMIT_CORE, &dumpsize);
4668     if (dumpsize.rlim_cur == 0)
4669         return 0;
4670 
4671     corefile = core_dump_filename(ts);
4672 
4673     if ((fd = open(corefile, O_WRONLY | O_CREAT,
4674                    S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH)) < 0)
4675         return (-errno);
4676 
4677     /*
4678      * Walk through target process memory mappings and
4679      * set up structure containing this information.  After
4680      * this point vma_xxx functions can be used.
4681      */
4682     if ((mm = vma_init()) == NULL)
4683         goto out;
4684 
4685     walk_memory_regions(mm, vma_walker);
4686     segs = vma_get_mapping_count(mm);
4687 
4688     /*
4689      * Construct valid coredump ELF header.  We also
4690      * add one more segment for notes.
4691      */
4692     fill_elf_header(&elf, segs + 1, ELF_MACHINE, 0);
4693     if (dump_write(fd, &elf, sizeof (elf)) != 0)
4694         goto out;
4695 
4696     /* fill in the in-memory version of notes */
4697     if (fill_note_info(&info, signr, env) < 0)
4698         goto out;
4699 
4700     offset += sizeof (elf);                             /* elf header */
4701     offset += (segs + 1) * sizeof (struct elf_phdr);    /* program headers */
4702 
4703     /* write out notes program header */
4704     fill_elf_note_phdr(&phdr, info.notes_size, offset);
4705 
4706     offset += info.notes_size;
4707     if (dump_write(fd, &phdr, sizeof (phdr)) != 0)
4708         goto out;
4709 
4710     /*
4711      * ELF specification wants data to start at page boundary so
4712      * we align it here.
4713      */
4714     data_offset = offset = roundup(offset, ELF_EXEC_PAGESIZE);
4715 
4716     /*
4717      * Write program headers for memory regions mapped in
4718      * the target process.
4719      */
4720     for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) {
4721         (void) memset(&phdr, 0, sizeof (phdr));
4722 
4723         phdr.p_type = PT_LOAD;
4724         phdr.p_offset = offset;
4725         phdr.p_vaddr = vma->vma_start;
4726         phdr.p_paddr = 0;
4727         phdr.p_filesz = vma_dump_size(vma);
4728         offset += phdr.p_filesz;
4729         phdr.p_memsz = vma->vma_end - vma->vma_start;
4730         phdr.p_flags = vma->vma_flags & PROT_READ ? PF_R : 0;
4731         if (vma->vma_flags & PROT_WRITE)
4732             phdr.p_flags |= PF_W;
4733         if (vma->vma_flags & PROT_EXEC)
4734             phdr.p_flags |= PF_X;
4735         phdr.p_align = ELF_EXEC_PAGESIZE;
4736 
4737         bswap_phdr(&phdr, 1);
4738         if (dump_write(fd, &phdr, sizeof(phdr)) != 0) {
4739             goto out;
4740         }
4741     }
4742 
4743     /*
4744      * Next we write notes just after program headers.  No
4745      * alignment needed here.
4746      */
4747     if (write_note_info(&info, fd) < 0)
4748         goto out;
4749 
4750     /* align data to page boundary */
4751     if (lseek(fd, data_offset, SEEK_SET) != data_offset)
4752         goto out;
4753 
4754     /*
4755      * Finally we can dump process memory into corefile as well.
4756      */
4757     for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) {
4758         abi_ulong addr;
4759         abi_ulong end;
4760 
4761         end = vma->vma_start + vma_dump_size(vma);
4762 
4763         for (addr = vma->vma_start; addr < end;
4764              addr += TARGET_PAGE_SIZE) {
4765             char page[TARGET_PAGE_SIZE];
4766             int error;
4767 
4768             /*
4769              *  Read in page from target process memory and
4770              *  write it to coredump file.
4771              */
4772             error = copy_from_user(page, addr, sizeof (page));
4773             if (error != 0) {
4774                 (void) fprintf(stderr, "unable to dump " TARGET_ABI_FMT_lx "\n",
4775                                addr);
4776                 errno = -error;
4777                 goto out;
4778             }
4779             if (dump_write(fd, page, TARGET_PAGE_SIZE) < 0)
4780                 goto out;
4781         }
4782     }
4783 
4784  out:
4785     free_note_info(&info);
4786     if (mm != NULL)
4787         vma_delete(mm);
4788     (void) close(fd);
4789 
4790     if (errno != 0)
4791         return (-errno);
4792     return (0);
4793 }
4794 #endif /* USE_ELF_CORE_DUMP */
4795 
4796 void do_init_thread(struct target_pt_regs *regs, struct image_info *infop)
4797 {
4798     init_thread(regs, infop);
4799 }
4800