xref: /openbmc/qemu/linux-user/elfload.c (revision 8ab5e8a5)
1 /* This is the Linux kernel elf-loading code, ported into user space */
2 #include "qemu/osdep.h"
3 #include <sys/param.h>
4 
5 #include <sys/resource.h>
6 #include <sys/shm.h>
7 
8 #include "qemu.h"
9 #include "user-internals.h"
10 #include "signal-common.h"
11 #include "loader.h"
12 #include "user-mmap.h"
13 #include "disas/disas.h"
14 #include "qemu/bitops.h"
15 #include "qemu/path.h"
16 #include "qemu/queue.h"
17 #include "qemu/guest-random.h"
18 #include "qemu/units.h"
19 #include "qemu/selfmap.h"
20 #include "qapi/error.h"
21 #include "target_signal.h"
22 #include "accel/tcg/debuginfo.h"
23 
24 #ifdef _ARCH_PPC64
25 #undef ARCH_DLINFO
26 #undef ELF_PLATFORM
27 #undef ELF_HWCAP
28 #undef ELF_HWCAP2
29 #undef ELF_CLASS
30 #undef ELF_DATA
31 #undef ELF_ARCH
32 #endif
33 
34 #define ELF_OSABI   ELFOSABI_SYSV
35 
36 /* from personality.h */
37 
38 /*
39  * Flags for bug emulation.
40  *
41  * These occupy the top three bytes.
42  */
43 enum {
44     ADDR_NO_RANDOMIZE = 0x0040000,      /* disable randomization of VA space */
45     FDPIC_FUNCPTRS =    0x0080000,      /* userspace function ptrs point to
46                                            descriptors (signal handling) */
47     MMAP_PAGE_ZERO =    0x0100000,
48     ADDR_COMPAT_LAYOUT = 0x0200000,
49     READ_IMPLIES_EXEC = 0x0400000,
50     ADDR_LIMIT_32BIT =  0x0800000,
51     SHORT_INODE =       0x1000000,
52     WHOLE_SECONDS =     0x2000000,
53     STICKY_TIMEOUTS =   0x4000000,
54     ADDR_LIMIT_3GB =    0x8000000,
55 };
56 
57 /*
58  * Personality types.
59  *
60  * These go in the low byte.  Avoid using the top bit, it will
61  * conflict with error returns.
62  */
63 enum {
64     PER_LINUX =         0x0000,
65     PER_LINUX_32BIT =   0x0000 | ADDR_LIMIT_32BIT,
66     PER_LINUX_FDPIC =   0x0000 | FDPIC_FUNCPTRS,
67     PER_SVR4 =          0x0001 | STICKY_TIMEOUTS | MMAP_PAGE_ZERO,
68     PER_SVR3 =          0x0002 | STICKY_TIMEOUTS | SHORT_INODE,
69     PER_SCOSVR3 =       0x0003 | STICKY_TIMEOUTS | WHOLE_SECONDS | SHORT_INODE,
70     PER_OSR5 =          0x0003 | STICKY_TIMEOUTS | WHOLE_SECONDS,
71     PER_WYSEV386 =      0x0004 | STICKY_TIMEOUTS | SHORT_INODE,
72     PER_ISCR4 =         0x0005 | STICKY_TIMEOUTS,
73     PER_BSD =           0x0006,
74     PER_SUNOS =         0x0006 | STICKY_TIMEOUTS,
75     PER_XENIX =         0x0007 | STICKY_TIMEOUTS | SHORT_INODE,
76     PER_LINUX32 =       0x0008,
77     PER_LINUX32_3GB =   0x0008 | ADDR_LIMIT_3GB,
78     PER_IRIX32 =        0x0009 | STICKY_TIMEOUTS,/* IRIX5 32-bit */
79     PER_IRIXN32 =       0x000a | STICKY_TIMEOUTS,/* IRIX6 new 32-bit */
80     PER_IRIX64 =        0x000b | STICKY_TIMEOUTS,/* IRIX6 64-bit */
81     PER_RISCOS =        0x000c,
82     PER_SOLARIS =       0x000d | STICKY_TIMEOUTS,
83     PER_UW7 =           0x000e | STICKY_TIMEOUTS | MMAP_PAGE_ZERO,
84     PER_OSF4 =          0x000f,                  /* OSF/1 v4 */
85     PER_HPUX =          0x0010,
86     PER_MASK =          0x00ff,
87 };
88 
89 /*
90  * Return the base personality without flags.
91  */
92 #define personality(pers)       (pers & PER_MASK)
93 
94 int info_is_fdpic(struct image_info *info)
95 {
96     return info->personality == PER_LINUX_FDPIC;
97 }
98 
99 /* this flag is uneffective under linux too, should be deleted */
100 #ifndef MAP_DENYWRITE
101 #define MAP_DENYWRITE 0
102 #endif
103 
104 /* should probably go in elf.h */
105 #ifndef ELIBBAD
106 #define ELIBBAD 80
107 #endif
108 
109 #if TARGET_BIG_ENDIAN
110 #define ELF_DATA        ELFDATA2MSB
111 #else
112 #define ELF_DATA        ELFDATA2LSB
113 #endif
114 
115 #ifdef TARGET_ABI_MIPSN32
116 typedef abi_ullong      target_elf_greg_t;
117 #define tswapreg(ptr)   tswap64(ptr)
118 #else
119 typedef abi_ulong       target_elf_greg_t;
120 #define tswapreg(ptr)   tswapal(ptr)
121 #endif
122 
123 #ifdef USE_UID16
124 typedef abi_ushort      target_uid_t;
125 typedef abi_ushort      target_gid_t;
126 #else
127 typedef abi_uint        target_uid_t;
128 typedef abi_uint        target_gid_t;
129 #endif
130 typedef abi_int         target_pid_t;
131 
132 #ifdef TARGET_I386
133 
134 #define ELF_HWCAP get_elf_hwcap()
135 
136 static uint32_t get_elf_hwcap(void)
137 {
138     X86CPU *cpu = X86_CPU(thread_cpu);
139 
140     return cpu->env.features[FEAT_1_EDX];
141 }
142 
143 #ifdef TARGET_X86_64
144 #define ELF_START_MMAP 0x2aaaaab000ULL
145 
146 #define ELF_CLASS      ELFCLASS64
147 #define ELF_ARCH       EM_X86_64
148 
149 #define ELF_PLATFORM   "x86_64"
150 
151 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop)
152 {
153     regs->rax = 0;
154     regs->rsp = infop->start_stack;
155     regs->rip = infop->entry;
156 }
157 
158 #define ELF_NREG    27
159 typedef target_elf_greg_t  target_elf_gregset_t[ELF_NREG];
160 
161 /*
162  * Note that ELF_NREG should be 29 as there should be place for
163  * TRAPNO and ERR "registers" as well but linux doesn't dump
164  * those.
165  *
166  * See linux kernel: arch/x86/include/asm/elf.h
167  */
168 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUX86State *env)
169 {
170     (*regs)[0] = tswapreg(env->regs[15]);
171     (*regs)[1] = tswapreg(env->regs[14]);
172     (*regs)[2] = tswapreg(env->regs[13]);
173     (*regs)[3] = tswapreg(env->regs[12]);
174     (*regs)[4] = tswapreg(env->regs[R_EBP]);
175     (*regs)[5] = tswapreg(env->regs[R_EBX]);
176     (*regs)[6] = tswapreg(env->regs[11]);
177     (*regs)[7] = tswapreg(env->regs[10]);
178     (*regs)[8] = tswapreg(env->regs[9]);
179     (*regs)[9] = tswapreg(env->regs[8]);
180     (*regs)[10] = tswapreg(env->regs[R_EAX]);
181     (*regs)[11] = tswapreg(env->regs[R_ECX]);
182     (*regs)[12] = tswapreg(env->regs[R_EDX]);
183     (*regs)[13] = tswapreg(env->regs[R_ESI]);
184     (*regs)[14] = tswapreg(env->regs[R_EDI]);
185     (*regs)[15] = tswapreg(env->regs[R_EAX]); /* XXX */
186     (*regs)[16] = tswapreg(env->eip);
187     (*regs)[17] = tswapreg(env->segs[R_CS].selector & 0xffff);
188     (*regs)[18] = tswapreg(env->eflags);
189     (*regs)[19] = tswapreg(env->regs[R_ESP]);
190     (*regs)[20] = tswapreg(env->segs[R_SS].selector & 0xffff);
191     (*regs)[21] = tswapreg(env->segs[R_FS].selector & 0xffff);
192     (*regs)[22] = tswapreg(env->segs[R_GS].selector & 0xffff);
193     (*regs)[23] = tswapreg(env->segs[R_DS].selector & 0xffff);
194     (*regs)[24] = tswapreg(env->segs[R_ES].selector & 0xffff);
195     (*regs)[25] = tswapreg(env->segs[R_FS].selector & 0xffff);
196     (*regs)[26] = tswapreg(env->segs[R_GS].selector & 0xffff);
197 }
198 
199 #if ULONG_MAX > UINT32_MAX
200 #define INIT_GUEST_COMMPAGE
201 static bool init_guest_commpage(void)
202 {
203     /*
204      * The vsyscall page is at a high negative address aka kernel space,
205      * which means that we cannot actually allocate it with target_mmap.
206      * We still should be able to use page_set_flags, unless the user
207      * has specified -R reserved_va, which would trigger an assert().
208      */
209     if (reserved_va != 0 &&
210         TARGET_VSYSCALL_PAGE + TARGET_PAGE_SIZE >= reserved_va) {
211         error_report("Cannot allocate vsyscall page");
212         exit(EXIT_FAILURE);
213     }
214     page_set_flags(TARGET_VSYSCALL_PAGE,
215                    TARGET_VSYSCALL_PAGE + TARGET_PAGE_SIZE,
216                    PAGE_EXEC | PAGE_VALID);
217     return true;
218 }
219 #endif
220 #else
221 
222 #define ELF_START_MMAP 0x80000000
223 
224 /*
225  * This is used to ensure we don't load something for the wrong architecture.
226  */
227 #define elf_check_arch(x) ( ((x) == EM_386) || ((x) == EM_486) )
228 
229 /*
230  * These are used to set parameters in the core dumps.
231  */
232 #define ELF_CLASS       ELFCLASS32
233 #define ELF_ARCH        EM_386
234 
235 #define ELF_PLATFORM get_elf_platform()
236 #define EXSTACK_DEFAULT true
237 
238 static const char *get_elf_platform(void)
239 {
240     static char elf_platform[] = "i386";
241     int family = object_property_get_int(OBJECT(thread_cpu), "family", NULL);
242     if (family > 6) {
243         family = 6;
244     }
245     if (family >= 3) {
246         elf_platform[1] = '0' + family;
247     }
248     return elf_platform;
249 }
250 
251 static inline void init_thread(struct target_pt_regs *regs,
252                                struct image_info *infop)
253 {
254     regs->esp = infop->start_stack;
255     regs->eip = infop->entry;
256 
257     /* SVR4/i386 ABI (pages 3-31, 3-32) says that when the program
258        starts %edx contains a pointer to a function which might be
259        registered using `atexit'.  This provides a mean for the
260        dynamic linker to call DT_FINI functions for shared libraries
261        that have been loaded before the code runs.
262 
263        A value of 0 tells we have no such handler.  */
264     regs->edx = 0;
265 }
266 
267 #define ELF_NREG    17
268 typedef target_elf_greg_t  target_elf_gregset_t[ELF_NREG];
269 
270 /*
271  * Note that ELF_NREG should be 19 as there should be place for
272  * TRAPNO and ERR "registers" as well but linux doesn't dump
273  * those.
274  *
275  * See linux kernel: arch/x86/include/asm/elf.h
276  */
277 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUX86State *env)
278 {
279     (*regs)[0] = tswapreg(env->regs[R_EBX]);
280     (*regs)[1] = tswapreg(env->regs[R_ECX]);
281     (*regs)[2] = tswapreg(env->regs[R_EDX]);
282     (*regs)[3] = tswapreg(env->regs[R_ESI]);
283     (*regs)[4] = tswapreg(env->regs[R_EDI]);
284     (*regs)[5] = tswapreg(env->regs[R_EBP]);
285     (*regs)[6] = tswapreg(env->regs[R_EAX]);
286     (*regs)[7] = tswapreg(env->segs[R_DS].selector & 0xffff);
287     (*regs)[8] = tswapreg(env->segs[R_ES].selector & 0xffff);
288     (*regs)[9] = tswapreg(env->segs[R_FS].selector & 0xffff);
289     (*regs)[10] = tswapreg(env->segs[R_GS].selector & 0xffff);
290     (*regs)[11] = tswapreg(env->regs[R_EAX]); /* XXX */
291     (*regs)[12] = tswapreg(env->eip);
292     (*regs)[13] = tswapreg(env->segs[R_CS].selector & 0xffff);
293     (*regs)[14] = tswapreg(env->eflags);
294     (*regs)[15] = tswapreg(env->regs[R_ESP]);
295     (*regs)[16] = tswapreg(env->segs[R_SS].selector & 0xffff);
296 }
297 #endif
298 
299 #define USE_ELF_CORE_DUMP
300 #define ELF_EXEC_PAGESIZE       4096
301 
302 #endif
303 
304 #ifdef TARGET_ARM
305 
306 #ifndef TARGET_AARCH64
307 /* 32 bit ARM definitions */
308 
309 #define ELF_START_MMAP 0x80000000
310 
311 #define ELF_ARCH        EM_ARM
312 #define ELF_CLASS       ELFCLASS32
313 #define EXSTACK_DEFAULT true
314 
315 static inline void init_thread(struct target_pt_regs *regs,
316                                struct image_info *infop)
317 {
318     abi_long stack = infop->start_stack;
319     memset(regs, 0, sizeof(*regs));
320 
321     regs->uregs[16] = ARM_CPU_MODE_USR;
322     if (infop->entry & 1) {
323         regs->uregs[16] |= CPSR_T;
324     }
325     regs->uregs[15] = infop->entry & 0xfffffffe;
326     regs->uregs[13] = infop->start_stack;
327     /* FIXME - what to for failure of get_user()? */
328     get_user_ual(regs->uregs[2], stack + 8); /* envp */
329     get_user_ual(regs->uregs[1], stack + 4); /* envp */
330     /* XXX: it seems that r0 is zeroed after ! */
331     regs->uregs[0] = 0;
332     /* For uClinux PIC binaries.  */
333     /* XXX: Linux does this only on ARM with no MMU (do we care ?) */
334     regs->uregs[10] = infop->start_data;
335 
336     /* Support ARM FDPIC.  */
337     if (info_is_fdpic(infop)) {
338         /* As described in the ABI document, r7 points to the loadmap info
339          * prepared by the kernel. If an interpreter is needed, r8 points
340          * to the interpreter loadmap and r9 points to the interpreter
341          * PT_DYNAMIC info. If no interpreter is needed, r8 is zero, and
342          * r9 points to the main program PT_DYNAMIC info.
343          */
344         regs->uregs[7] = infop->loadmap_addr;
345         if (infop->interpreter_loadmap_addr) {
346             /* Executable is dynamically loaded.  */
347             regs->uregs[8] = infop->interpreter_loadmap_addr;
348             regs->uregs[9] = infop->interpreter_pt_dynamic_addr;
349         } else {
350             regs->uregs[8] = 0;
351             regs->uregs[9] = infop->pt_dynamic_addr;
352         }
353     }
354 }
355 
356 #define ELF_NREG    18
357 typedef target_elf_greg_t  target_elf_gregset_t[ELF_NREG];
358 
359 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUARMState *env)
360 {
361     (*regs)[0] = tswapreg(env->regs[0]);
362     (*regs)[1] = tswapreg(env->regs[1]);
363     (*regs)[2] = tswapreg(env->regs[2]);
364     (*regs)[3] = tswapreg(env->regs[3]);
365     (*regs)[4] = tswapreg(env->regs[4]);
366     (*regs)[5] = tswapreg(env->regs[5]);
367     (*regs)[6] = tswapreg(env->regs[6]);
368     (*regs)[7] = tswapreg(env->regs[7]);
369     (*regs)[8] = tswapreg(env->regs[8]);
370     (*regs)[9] = tswapreg(env->regs[9]);
371     (*regs)[10] = tswapreg(env->regs[10]);
372     (*regs)[11] = tswapreg(env->regs[11]);
373     (*regs)[12] = tswapreg(env->regs[12]);
374     (*regs)[13] = tswapreg(env->regs[13]);
375     (*regs)[14] = tswapreg(env->regs[14]);
376     (*regs)[15] = tswapreg(env->regs[15]);
377 
378     (*regs)[16] = tswapreg(cpsr_read((CPUARMState *)env));
379     (*regs)[17] = tswapreg(env->regs[0]); /* XXX */
380 }
381 
382 #define USE_ELF_CORE_DUMP
383 #define ELF_EXEC_PAGESIZE       4096
384 
385 enum
386 {
387     ARM_HWCAP_ARM_SWP       = 1 << 0,
388     ARM_HWCAP_ARM_HALF      = 1 << 1,
389     ARM_HWCAP_ARM_THUMB     = 1 << 2,
390     ARM_HWCAP_ARM_26BIT     = 1 << 3,
391     ARM_HWCAP_ARM_FAST_MULT = 1 << 4,
392     ARM_HWCAP_ARM_FPA       = 1 << 5,
393     ARM_HWCAP_ARM_VFP       = 1 << 6,
394     ARM_HWCAP_ARM_EDSP      = 1 << 7,
395     ARM_HWCAP_ARM_JAVA      = 1 << 8,
396     ARM_HWCAP_ARM_IWMMXT    = 1 << 9,
397     ARM_HWCAP_ARM_CRUNCH    = 1 << 10,
398     ARM_HWCAP_ARM_THUMBEE   = 1 << 11,
399     ARM_HWCAP_ARM_NEON      = 1 << 12,
400     ARM_HWCAP_ARM_VFPv3     = 1 << 13,
401     ARM_HWCAP_ARM_VFPv3D16  = 1 << 14,
402     ARM_HWCAP_ARM_TLS       = 1 << 15,
403     ARM_HWCAP_ARM_VFPv4     = 1 << 16,
404     ARM_HWCAP_ARM_IDIVA     = 1 << 17,
405     ARM_HWCAP_ARM_IDIVT     = 1 << 18,
406     ARM_HWCAP_ARM_VFPD32    = 1 << 19,
407     ARM_HWCAP_ARM_LPAE      = 1 << 20,
408     ARM_HWCAP_ARM_EVTSTRM   = 1 << 21,
409 };
410 
411 enum {
412     ARM_HWCAP2_ARM_AES      = 1 << 0,
413     ARM_HWCAP2_ARM_PMULL    = 1 << 1,
414     ARM_HWCAP2_ARM_SHA1     = 1 << 2,
415     ARM_HWCAP2_ARM_SHA2     = 1 << 3,
416     ARM_HWCAP2_ARM_CRC32    = 1 << 4,
417 };
418 
419 /* The commpage only exists for 32 bit kernels */
420 
421 #define HI_COMMPAGE (intptr_t)0xffff0f00u
422 
423 static bool init_guest_commpage(void)
424 {
425     abi_ptr commpage = HI_COMMPAGE & -qemu_host_page_size;
426     void *want = g2h_untagged(commpage);
427     void *addr = mmap(want, qemu_host_page_size, PROT_READ | PROT_WRITE,
428                       MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0);
429 
430     if (addr == MAP_FAILED) {
431         perror("Allocating guest commpage");
432         exit(EXIT_FAILURE);
433     }
434     if (addr != want) {
435         return false;
436     }
437 
438     /* Set kernel helper versions; rest of page is 0.  */
439     __put_user(5, (uint32_t *)g2h_untagged(0xffff0ffcu));
440 
441     if (mprotect(addr, qemu_host_page_size, PROT_READ)) {
442         perror("Protecting guest commpage");
443         exit(EXIT_FAILURE);
444     }
445 
446     page_set_flags(commpage, commpage + qemu_host_page_size,
447                    PAGE_READ | PAGE_EXEC | PAGE_VALID);
448     return true;
449 }
450 
451 #define ELF_HWCAP get_elf_hwcap()
452 #define ELF_HWCAP2 get_elf_hwcap2()
453 
454 static uint32_t get_elf_hwcap(void)
455 {
456     ARMCPU *cpu = ARM_CPU(thread_cpu);
457     uint32_t hwcaps = 0;
458 
459     hwcaps |= ARM_HWCAP_ARM_SWP;
460     hwcaps |= ARM_HWCAP_ARM_HALF;
461     hwcaps |= ARM_HWCAP_ARM_THUMB;
462     hwcaps |= ARM_HWCAP_ARM_FAST_MULT;
463 
464     /* probe for the extra features */
465 #define GET_FEATURE(feat, hwcap) \
466     do { if (arm_feature(&cpu->env, feat)) { hwcaps |= hwcap; } } while (0)
467 
468 #define GET_FEATURE_ID(feat, hwcap) \
469     do { if (cpu_isar_feature(feat, cpu)) { hwcaps |= hwcap; } } while (0)
470 
471     /* EDSP is in v5TE and above, but all our v5 CPUs are v5TE */
472     GET_FEATURE(ARM_FEATURE_V5, ARM_HWCAP_ARM_EDSP);
473     GET_FEATURE(ARM_FEATURE_IWMMXT, ARM_HWCAP_ARM_IWMMXT);
474     GET_FEATURE(ARM_FEATURE_THUMB2EE, ARM_HWCAP_ARM_THUMBEE);
475     GET_FEATURE(ARM_FEATURE_NEON, ARM_HWCAP_ARM_NEON);
476     GET_FEATURE(ARM_FEATURE_V6K, ARM_HWCAP_ARM_TLS);
477     GET_FEATURE(ARM_FEATURE_LPAE, ARM_HWCAP_ARM_LPAE);
478     GET_FEATURE_ID(aa32_arm_div, ARM_HWCAP_ARM_IDIVA);
479     GET_FEATURE_ID(aa32_thumb_div, ARM_HWCAP_ARM_IDIVT);
480     GET_FEATURE_ID(aa32_vfp, ARM_HWCAP_ARM_VFP);
481 
482     if (cpu_isar_feature(aa32_fpsp_v3, cpu) ||
483         cpu_isar_feature(aa32_fpdp_v3, cpu)) {
484         hwcaps |= ARM_HWCAP_ARM_VFPv3;
485         if (cpu_isar_feature(aa32_simd_r32, cpu)) {
486             hwcaps |= ARM_HWCAP_ARM_VFPD32;
487         } else {
488             hwcaps |= ARM_HWCAP_ARM_VFPv3D16;
489         }
490     }
491     GET_FEATURE_ID(aa32_simdfmac, ARM_HWCAP_ARM_VFPv4);
492 
493     return hwcaps;
494 }
495 
496 static uint32_t get_elf_hwcap2(void)
497 {
498     ARMCPU *cpu = ARM_CPU(thread_cpu);
499     uint32_t hwcaps = 0;
500 
501     GET_FEATURE_ID(aa32_aes, ARM_HWCAP2_ARM_AES);
502     GET_FEATURE_ID(aa32_pmull, ARM_HWCAP2_ARM_PMULL);
503     GET_FEATURE_ID(aa32_sha1, ARM_HWCAP2_ARM_SHA1);
504     GET_FEATURE_ID(aa32_sha2, ARM_HWCAP2_ARM_SHA2);
505     GET_FEATURE_ID(aa32_crc32, ARM_HWCAP2_ARM_CRC32);
506     return hwcaps;
507 }
508 
509 #undef GET_FEATURE
510 #undef GET_FEATURE_ID
511 
512 #define ELF_PLATFORM get_elf_platform()
513 
514 static const char *get_elf_platform(void)
515 {
516     CPUARMState *env = thread_cpu->env_ptr;
517 
518 #if TARGET_BIG_ENDIAN
519 # define END  "b"
520 #else
521 # define END  "l"
522 #endif
523 
524     if (arm_feature(env, ARM_FEATURE_V8)) {
525         return "v8" END;
526     } else if (arm_feature(env, ARM_FEATURE_V7)) {
527         if (arm_feature(env, ARM_FEATURE_M)) {
528             return "v7m" END;
529         } else {
530             return "v7" END;
531         }
532     } else if (arm_feature(env, ARM_FEATURE_V6)) {
533         return "v6" END;
534     } else if (arm_feature(env, ARM_FEATURE_V5)) {
535         return "v5" END;
536     } else {
537         return "v4" END;
538     }
539 
540 #undef END
541 }
542 
543 #else
544 /* 64 bit ARM definitions */
545 #define ELF_START_MMAP 0x80000000
546 
547 #define ELF_ARCH        EM_AARCH64
548 #define ELF_CLASS       ELFCLASS64
549 #if TARGET_BIG_ENDIAN
550 # define ELF_PLATFORM    "aarch64_be"
551 #else
552 # define ELF_PLATFORM    "aarch64"
553 #endif
554 
555 static inline void init_thread(struct target_pt_regs *regs,
556                                struct image_info *infop)
557 {
558     abi_long stack = infop->start_stack;
559     memset(regs, 0, sizeof(*regs));
560 
561     regs->pc = infop->entry & ~0x3ULL;
562     regs->sp = stack;
563 }
564 
565 #define ELF_NREG    34
566 typedef target_elf_greg_t  target_elf_gregset_t[ELF_NREG];
567 
568 static void elf_core_copy_regs(target_elf_gregset_t *regs,
569                                const CPUARMState *env)
570 {
571     int i;
572 
573     for (i = 0; i < 32; i++) {
574         (*regs)[i] = tswapreg(env->xregs[i]);
575     }
576     (*regs)[32] = tswapreg(env->pc);
577     (*regs)[33] = tswapreg(pstate_read((CPUARMState *)env));
578 }
579 
580 #define USE_ELF_CORE_DUMP
581 #define ELF_EXEC_PAGESIZE       4096
582 
583 enum {
584     ARM_HWCAP_A64_FP            = 1 << 0,
585     ARM_HWCAP_A64_ASIMD         = 1 << 1,
586     ARM_HWCAP_A64_EVTSTRM       = 1 << 2,
587     ARM_HWCAP_A64_AES           = 1 << 3,
588     ARM_HWCAP_A64_PMULL         = 1 << 4,
589     ARM_HWCAP_A64_SHA1          = 1 << 5,
590     ARM_HWCAP_A64_SHA2          = 1 << 6,
591     ARM_HWCAP_A64_CRC32         = 1 << 7,
592     ARM_HWCAP_A64_ATOMICS       = 1 << 8,
593     ARM_HWCAP_A64_FPHP          = 1 << 9,
594     ARM_HWCAP_A64_ASIMDHP       = 1 << 10,
595     ARM_HWCAP_A64_CPUID         = 1 << 11,
596     ARM_HWCAP_A64_ASIMDRDM      = 1 << 12,
597     ARM_HWCAP_A64_JSCVT         = 1 << 13,
598     ARM_HWCAP_A64_FCMA          = 1 << 14,
599     ARM_HWCAP_A64_LRCPC         = 1 << 15,
600     ARM_HWCAP_A64_DCPOP         = 1 << 16,
601     ARM_HWCAP_A64_SHA3          = 1 << 17,
602     ARM_HWCAP_A64_SM3           = 1 << 18,
603     ARM_HWCAP_A64_SM4           = 1 << 19,
604     ARM_HWCAP_A64_ASIMDDP       = 1 << 20,
605     ARM_HWCAP_A64_SHA512        = 1 << 21,
606     ARM_HWCAP_A64_SVE           = 1 << 22,
607     ARM_HWCAP_A64_ASIMDFHM      = 1 << 23,
608     ARM_HWCAP_A64_DIT           = 1 << 24,
609     ARM_HWCAP_A64_USCAT         = 1 << 25,
610     ARM_HWCAP_A64_ILRCPC        = 1 << 26,
611     ARM_HWCAP_A64_FLAGM         = 1 << 27,
612     ARM_HWCAP_A64_SSBS          = 1 << 28,
613     ARM_HWCAP_A64_SB            = 1 << 29,
614     ARM_HWCAP_A64_PACA          = 1 << 30,
615     ARM_HWCAP_A64_PACG          = 1UL << 31,
616 
617     ARM_HWCAP2_A64_DCPODP       = 1 << 0,
618     ARM_HWCAP2_A64_SVE2         = 1 << 1,
619     ARM_HWCAP2_A64_SVEAES       = 1 << 2,
620     ARM_HWCAP2_A64_SVEPMULL     = 1 << 3,
621     ARM_HWCAP2_A64_SVEBITPERM   = 1 << 4,
622     ARM_HWCAP2_A64_SVESHA3      = 1 << 5,
623     ARM_HWCAP2_A64_SVESM4       = 1 << 6,
624     ARM_HWCAP2_A64_FLAGM2       = 1 << 7,
625     ARM_HWCAP2_A64_FRINT        = 1 << 8,
626     ARM_HWCAP2_A64_SVEI8MM      = 1 << 9,
627     ARM_HWCAP2_A64_SVEF32MM     = 1 << 10,
628     ARM_HWCAP2_A64_SVEF64MM     = 1 << 11,
629     ARM_HWCAP2_A64_SVEBF16      = 1 << 12,
630     ARM_HWCAP2_A64_I8MM         = 1 << 13,
631     ARM_HWCAP2_A64_BF16         = 1 << 14,
632     ARM_HWCAP2_A64_DGH          = 1 << 15,
633     ARM_HWCAP2_A64_RNG          = 1 << 16,
634     ARM_HWCAP2_A64_BTI          = 1 << 17,
635     ARM_HWCAP2_A64_MTE          = 1 << 18,
636     ARM_HWCAP2_A64_ECV          = 1 << 19,
637     ARM_HWCAP2_A64_AFP          = 1 << 20,
638     ARM_HWCAP2_A64_RPRES        = 1 << 21,
639     ARM_HWCAP2_A64_MTE3         = 1 << 22,
640     ARM_HWCAP2_A64_SME          = 1 << 23,
641     ARM_HWCAP2_A64_SME_I16I64   = 1 << 24,
642     ARM_HWCAP2_A64_SME_F64F64   = 1 << 25,
643     ARM_HWCAP2_A64_SME_I8I32    = 1 << 26,
644     ARM_HWCAP2_A64_SME_F16F32   = 1 << 27,
645     ARM_HWCAP2_A64_SME_B16F32   = 1 << 28,
646     ARM_HWCAP2_A64_SME_F32F32   = 1 << 29,
647     ARM_HWCAP2_A64_SME_FA64     = 1 << 30,
648 };
649 
650 #define ELF_HWCAP   get_elf_hwcap()
651 #define ELF_HWCAP2  get_elf_hwcap2()
652 
653 #define GET_FEATURE_ID(feat, hwcap) \
654     do { if (cpu_isar_feature(feat, cpu)) { hwcaps |= hwcap; } } while (0)
655 
656 static uint32_t get_elf_hwcap(void)
657 {
658     ARMCPU *cpu = ARM_CPU(thread_cpu);
659     uint32_t hwcaps = 0;
660 
661     hwcaps |= ARM_HWCAP_A64_FP;
662     hwcaps |= ARM_HWCAP_A64_ASIMD;
663     hwcaps |= ARM_HWCAP_A64_CPUID;
664 
665     /* probe for the extra features */
666 
667     GET_FEATURE_ID(aa64_aes, ARM_HWCAP_A64_AES);
668     GET_FEATURE_ID(aa64_pmull, ARM_HWCAP_A64_PMULL);
669     GET_FEATURE_ID(aa64_sha1, ARM_HWCAP_A64_SHA1);
670     GET_FEATURE_ID(aa64_sha256, ARM_HWCAP_A64_SHA2);
671     GET_FEATURE_ID(aa64_sha512, ARM_HWCAP_A64_SHA512);
672     GET_FEATURE_ID(aa64_crc32, ARM_HWCAP_A64_CRC32);
673     GET_FEATURE_ID(aa64_sha3, ARM_HWCAP_A64_SHA3);
674     GET_FEATURE_ID(aa64_sm3, ARM_HWCAP_A64_SM3);
675     GET_FEATURE_ID(aa64_sm4, ARM_HWCAP_A64_SM4);
676     GET_FEATURE_ID(aa64_fp16, ARM_HWCAP_A64_FPHP | ARM_HWCAP_A64_ASIMDHP);
677     GET_FEATURE_ID(aa64_atomics, ARM_HWCAP_A64_ATOMICS);
678     GET_FEATURE_ID(aa64_rdm, ARM_HWCAP_A64_ASIMDRDM);
679     GET_FEATURE_ID(aa64_dp, ARM_HWCAP_A64_ASIMDDP);
680     GET_FEATURE_ID(aa64_fcma, ARM_HWCAP_A64_FCMA);
681     GET_FEATURE_ID(aa64_sve, ARM_HWCAP_A64_SVE);
682     GET_FEATURE_ID(aa64_pauth, ARM_HWCAP_A64_PACA | ARM_HWCAP_A64_PACG);
683     GET_FEATURE_ID(aa64_fhm, ARM_HWCAP_A64_ASIMDFHM);
684     GET_FEATURE_ID(aa64_jscvt, ARM_HWCAP_A64_JSCVT);
685     GET_FEATURE_ID(aa64_sb, ARM_HWCAP_A64_SB);
686     GET_FEATURE_ID(aa64_condm_4, ARM_HWCAP_A64_FLAGM);
687     GET_FEATURE_ID(aa64_dcpop, ARM_HWCAP_A64_DCPOP);
688     GET_FEATURE_ID(aa64_rcpc_8_3, ARM_HWCAP_A64_LRCPC);
689     GET_FEATURE_ID(aa64_rcpc_8_4, ARM_HWCAP_A64_ILRCPC);
690 
691     return hwcaps;
692 }
693 
694 static uint32_t get_elf_hwcap2(void)
695 {
696     ARMCPU *cpu = ARM_CPU(thread_cpu);
697     uint32_t hwcaps = 0;
698 
699     GET_FEATURE_ID(aa64_dcpodp, ARM_HWCAP2_A64_DCPODP);
700     GET_FEATURE_ID(aa64_sve2, ARM_HWCAP2_A64_SVE2);
701     GET_FEATURE_ID(aa64_sve2_aes, ARM_HWCAP2_A64_SVEAES);
702     GET_FEATURE_ID(aa64_sve2_pmull128, ARM_HWCAP2_A64_SVEPMULL);
703     GET_FEATURE_ID(aa64_sve2_bitperm, ARM_HWCAP2_A64_SVEBITPERM);
704     GET_FEATURE_ID(aa64_sve2_sha3, ARM_HWCAP2_A64_SVESHA3);
705     GET_FEATURE_ID(aa64_sve2_sm4, ARM_HWCAP2_A64_SVESM4);
706     GET_FEATURE_ID(aa64_condm_5, ARM_HWCAP2_A64_FLAGM2);
707     GET_FEATURE_ID(aa64_frint, ARM_HWCAP2_A64_FRINT);
708     GET_FEATURE_ID(aa64_sve_i8mm, ARM_HWCAP2_A64_SVEI8MM);
709     GET_FEATURE_ID(aa64_sve_f32mm, ARM_HWCAP2_A64_SVEF32MM);
710     GET_FEATURE_ID(aa64_sve_f64mm, ARM_HWCAP2_A64_SVEF64MM);
711     GET_FEATURE_ID(aa64_sve_bf16, ARM_HWCAP2_A64_SVEBF16);
712     GET_FEATURE_ID(aa64_i8mm, ARM_HWCAP2_A64_I8MM);
713     GET_FEATURE_ID(aa64_bf16, ARM_HWCAP2_A64_BF16);
714     GET_FEATURE_ID(aa64_rndr, ARM_HWCAP2_A64_RNG);
715     GET_FEATURE_ID(aa64_bti, ARM_HWCAP2_A64_BTI);
716     GET_FEATURE_ID(aa64_mte, ARM_HWCAP2_A64_MTE);
717     GET_FEATURE_ID(aa64_sme, (ARM_HWCAP2_A64_SME |
718                               ARM_HWCAP2_A64_SME_F32F32 |
719                               ARM_HWCAP2_A64_SME_B16F32 |
720                               ARM_HWCAP2_A64_SME_F16F32 |
721                               ARM_HWCAP2_A64_SME_I8I32));
722     GET_FEATURE_ID(aa64_sme_f64f64, ARM_HWCAP2_A64_SME_F64F64);
723     GET_FEATURE_ID(aa64_sme_i16i64, ARM_HWCAP2_A64_SME_I16I64);
724     GET_FEATURE_ID(aa64_sme_fa64, ARM_HWCAP2_A64_SME_FA64);
725 
726     return hwcaps;
727 }
728 
729 #undef GET_FEATURE_ID
730 
731 #endif /* not TARGET_AARCH64 */
732 #endif /* TARGET_ARM */
733 
734 #ifdef TARGET_SPARC
735 #ifdef TARGET_SPARC64
736 
737 #define ELF_START_MMAP 0x80000000
738 #define ELF_HWCAP  (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \
739                     | HWCAP_SPARC_MULDIV | HWCAP_SPARC_V9)
740 #ifndef TARGET_ABI32
741 #define elf_check_arch(x) ( (x) == EM_SPARCV9 || (x) == EM_SPARC32PLUS )
742 #else
743 #define elf_check_arch(x) ( (x) == EM_SPARC32PLUS || (x) == EM_SPARC )
744 #endif
745 
746 #define ELF_CLASS   ELFCLASS64
747 #define ELF_ARCH    EM_SPARCV9
748 #else
749 #define ELF_START_MMAP 0x80000000
750 #define ELF_HWCAP  (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \
751                     | HWCAP_SPARC_MULDIV)
752 #define ELF_CLASS   ELFCLASS32
753 #define ELF_ARCH    EM_SPARC
754 #endif /* TARGET_SPARC64 */
755 
756 static inline void init_thread(struct target_pt_regs *regs,
757                                struct image_info *infop)
758 {
759     /* Note that target_cpu_copy_regs does not read psr/tstate. */
760     regs->pc = infop->entry;
761     regs->npc = regs->pc + 4;
762     regs->y = 0;
763     regs->u_regs[14] = (infop->start_stack - 16 * sizeof(abi_ulong)
764                         - TARGET_STACK_BIAS);
765 }
766 #endif /* TARGET_SPARC */
767 
768 #ifdef TARGET_PPC
769 
770 #define ELF_MACHINE    PPC_ELF_MACHINE
771 #define ELF_START_MMAP 0x80000000
772 
773 #if defined(TARGET_PPC64)
774 
775 #define elf_check_arch(x) ( (x) == EM_PPC64 )
776 
777 #define ELF_CLASS       ELFCLASS64
778 
779 #else
780 
781 #define ELF_CLASS       ELFCLASS32
782 #define EXSTACK_DEFAULT true
783 
784 #endif
785 
786 #define ELF_ARCH        EM_PPC
787 
788 /* Feature masks for the Aux Vector Hardware Capabilities (AT_HWCAP).
789    See arch/powerpc/include/asm/cputable.h.  */
790 enum {
791     QEMU_PPC_FEATURE_32 = 0x80000000,
792     QEMU_PPC_FEATURE_64 = 0x40000000,
793     QEMU_PPC_FEATURE_601_INSTR = 0x20000000,
794     QEMU_PPC_FEATURE_HAS_ALTIVEC = 0x10000000,
795     QEMU_PPC_FEATURE_HAS_FPU = 0x08000000,
796     QEMU_PPC_FEATURE_HAS_MMU = 0x04000000,
797     QEMU_PPC_FEATURE_HAS_4xxMAC = 0x02000000,
798     QEMU_PPC_FEATURE_UNIFIED_CACHE = 0x01000000,
799     QEMU_PPC_FEATURE_HAS_SPE = 0x00800000,
800     QEMU_PPC_FEATURE_HAS_EFP_SINGLE = 0x00400000,
801     QEMU_PPC_FEATURE_HAS_EFP_DOUBLE = 0x00200000,
802     QEMU_PPC_FEATURE_NO_TB = 0x00100000,
803     QEMU_PPC_FEATURE_POWER4 = 0x00080000,
804     QEMU_PPC_FEATURE_POWER5 = 0x00040000,
805     QEMU_PPC_FEATURE_POWER5_PLUS = 0x00020000,
806     QEMU_PPC_FEATURE_CELL = 0x00010000,
807     QEMU_PPC_FEATURE_BOOKE = 0x00008000,
808     QEMU_PPC_FEATURE_SMT = 0x00004000,
809     QEMU_PPC_FEATURE_ICACHE_SNOOP = 0x00002000,
810     QEMU_PPC_FEATURE_ARCH_2_05 = 0x00001000,
811     QEMU_PPC_FEATURE_PA6T = 0x00000800,
812     QEMU_PPC_FEATURE_HAS_DFP = 0x00000400,
813     QEMU_PPC_FEATURE_POWER6_EXT = 0x00000200,
814     QEMU_PPC_FEATURE_ARCH_2_06 = 0x00000100,
815     QEMU_PPC_FEATURE_HAS_VSX = 0x00000080,
816     QEMU_PPC_FEATURE_PSERIES_PERFMON_COMPAT = 0x00000040,
817 
818     QEMU_PPC_FEATURE_TRUE_LE = 0x00000002,
819     QEMU_PPC_FEATURE_PPC_LE = 0x00000001,
820 
821     /* Feature definitions in AT_HWCAP2.  */
822     QEMU_PPC_FEATURE2_ARCH_2_07 = 0x80000000, /* ISA 2.07 */
823     QEMU_PPC_FEATURE2_HAS_HTM = 0x40000000, /* Hardware Transactional Memory */
824     QEMU_PPC_FEATURE2_HAS_DSCR = 0x20000000, /* Data Stream Control Register */
825     QEMU_PPC_FEATURE2_HAS_EBB = 0x10000000, /* Event Base Branching */
826     QEMU_PPC_FEATURE2_HAS_ISEL = 0x08000000, /* Integer Select */
827     QEMU_PPC_FEATURE2_HAS_TAR = 0x04000000, /* Target Address Register */
828     QEMU_PPC_FEATURE2_VEC_CRYPTO = 0x02000000,
829     QEMU_PPC_FEATURE2_HTM_NOSC = 0x01000000,
830     QEMU_PPC_FEATURE2_ARCH_3_00 = 0x00800000, /* ISA 3.00 */
831     QEMU_PPC_FEATURE2_HAS_IEEE128 = 0x00400000, /* VSX IEEE Bin Float 128-bit */
832     QEMU_PPC_FEATURE2_DARN = 0x00200000, /* darn random number insn */
833     QEMU_PPC_FEATURE2_SCV = 0x00100000, /* scv syscall */
834     QEMU_PPC_FEATURE2_HTM_NO_SUSPEND = 0x00080000, /* TM w/o suspended state */
835     QEMU_PPC_FEATURE2_ARCH_3_1 = 0x00040000, /* ISA 3.1 */
836     QEMU_PPC_FEATURE2_MMA = 0x00020000, /* Matrix-Multiply Assist */
837 };
838 
839 #define ELF_HWCAP get_elf_hwcap()
840 
841 static uint32_t get_elf_hwcap(void)
842 {
843     PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);
844     uint32_t features = 0;
845 
846     /* We don't have to be terribly complete here; the high points are
847        Altivec/FP/SPE support.  Anything else is just a bonus.  */
848 #define GET_FEATURE(flag, feature)                                      \
849     do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0)
850 #define GET_FEATURE2(flags, feature) \
851     do { \
852         if ((cpu->env.insns_flags2 & flags) == flags) { \
853             features |= feature; \
854         } \
855     } while (0)
856     GET_FEATURE(PPC_64B, QEMU_PPC_FEATURE_64);
857     GET_FEATURE(PPC_FLOAT, QEMU_PPC_FEATURE_HAS_FPU);
858     GET_FEATURE(PPC_ALTIVEC, QEMU_PPC_FEATURE_HAS_ALTIVEC);
859     GET_FEATURE(PPC_SPE, QEMU_PPC_FEATURE_HAS_SPE);
860     GET_FEATURE(PPC_SPE_SINGLE, QEMU_PPC_FEATURE_HAS_EFP_SINGLE);
861     GET_FEATURE(PPC_SPE_DOUBLE, QEMU_PPC_FEATURE_HAS_EFP_DOUBLE);
862     GET_FEATURE(PPC_BOOKE, QEMU_PPC_FEATURE_BOOKE);
863     GET_FEATURE(PPC_405_MAC, QEMU_PPC_FEATURE_HAS_4xxMAC);
864     GET_FEATURE2(PPC2_DFP, QEMU_PPC_FEATURE_HAS_DFP);
865     GET_FEATURE2(PPC2_VSX, QEMU_PPC_FEATURE_HAS_VSX);
866     GET_FEATURE2((PPC2_PERM_ISA206 | PPC2_DIVE_ISA206 | PPC2_ATOMIC_ISA206 |
867                   PPC2_FP_CVT_ISA206 | PPC2_FP_TST_ISA206),
868                   QEMU_PPC_FEATURE_ARCH_2_06);
869 #undef GET_FEATURE
870 #undef GET_FEATURE2
871 
872     return features;
873 }
874 
875 #define ELF_HWCAP2 get_elf_hwcap2()
876 
877 static uint32_t get_elf_hwcap2(void)
878 {
879     PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);
880     uint32_t features = 0;
881 
882 #define GET_FEATURE(flag, feature)                                      \
883     do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0)
884 #define GET_FEATURE2(flag, feature)                                      \
885     do { if (cpu->env.insns_flags2 & flag) { features |= feature; } } while (0)
886 
887     GET_FEATURE(PPC_ISEL, QEMU_PPC_FEATURE2_HAS_ISEL);
888     GET_FEATURE2(PPC2_BCTAR_ISA207, QEMU_PPC_FEATURE2_HAS_TAR);
889     GET_FEATURE2((PPC2_BCTAR_ISA207 | PPC2_LSQ_ISA207 | PPC2_ALTIVEC_207 |
890                   PPC2_ISA207S), QEMU_PPC_FEATURE2_ARCH_2_07 |
891                   QEMU_PPC_FEATURE2_VEC_CRYPTO);
892     GET_FEATURE2(PPC2_ISA300, QEMU_PPC_FEATURE2_ARCH_3_00 |
893                  QEMU_PPC_FEATURE2_DARN | QEMU_PPC_FEATURE2_HAS_IEEE128);
894     GET_FEATURE2(PPC2_ISA310, QEMU_PPC_FEATURE2_ARCH_3_1 |
895                  QEMU_PPC_FEATURE2_MMA);
896 
897 #undef GET_FEATURE
898 #undef GET_FEATURE2
899 
900     return features;
901 }
902 
903 /*
904  * The requirements here are:
905  * - keep the final alignment of sp (sp & 0xf)
906  * - make sure the 32-bit value at the first 16 byte aligned position of
907  *   AUXV is greater than 16 for glibc compatibility.
908  *   AT_IGNOREPPC is used for that.
909  * - for compatibility with glibc ARCH_DLINFO must always be defined on PPC,
910  *   even if DLINFO_ARCH_ITEMS goes to zero or is undefined.
911  */
912 #define DLINFO_ARCH_ITEMS       5
913 #define ARCH_DLINFO                                     \
914     do {                                                \
915         PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);              \
916         /*                                              \
917          * Handle glibc compatibility: these magic entries must \
918          * be at the lowest addresses in the final auxv.        \
919          */                                             \
920         NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC);        \
921         NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC);        \
922         NEW_AUX_ENT(AT_DCACHEBSIZE, cpu->env.dcache_line_size); \
923         NEW_AUX_ENT(AT_ICACHEBSIZE, cpu->env.icache_line_size); \
924         NEW_AUX_ENT(AT_UCACHEBSIZE, 0);                 \
925     } while (0)
926 
927 static inline void init_thread(struct target_pt_regs *_regs, struct image_info *infop)
928 {
929     _regs->gpr[1] = infop->start_stack;
930 #if defined(TARGET_PPC64)
931     if (get_ppc64_abi(infop) < 2) {
932         uint64_t val;
933         get_user_u64(val, infop->entry + 8);
934         _regs->gpr[2] = val + infop->load_bias;
935         get_user_u64(val, infop->entry);
936         infop->entry = val + infop->load_bias;
937     } else {
938         _regs->gpr[12] = infop->entry;  /* r12 set to global entry address */
939     }
940 #endif
941     _regs->nip = infop->entry;
942 }
943 
944 /* See linux kernel: arch/powerpc/include/asm/elf.h.  */
945 #define ELF_NREG 48
946 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
947 
948 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUPPCState *env)
949 {
950     int i;
951     target_ulong ccr = 0;
952 
953     for (i = 0; i < ARRAY_SIZE(env->gpr); i++) {
954         (*regs)[i] = tswapreg(env->gpr[i]);
955     }
956 
957     (*regs)[32] = tswapreg(env->nip);
958     (*regs)[33] = tswapreg(env->msr);
959     (*regs)[35] = tswapreg(env->ctr);
960     (*regs)[36] = tswapreg(env->lr);
961     (*regs)[37] = tswapreg(cpu_read_xer(env));
962 
963     for (i = 0; i < ARRAY_SIZE(env->crf); i++) {
964         ccr |= env->crf[i] << (32 - ((i + 1) * 4));
965     }
966     (*regs)[38] = tswapreg(ccr);
967 }
968 
969 #define USE_ELF_CORE_DUMP
970 #define ELF_EXEC_PAGESIZE       4096
971 
972 #endif
973 
974 #ifdef TARGET_LOONGARCH64
975 
976 #define ELF_START_MMAP 0x80000000
977 
978 #define ELF_CLASS   ELFCLASS64
979 #define ELF_ARCH    EM_LOONGARCH
980 #define EXSTACK_DEFAULT true
981 
982 #define elf_check_arch(x) ((x) == EM_LOONGARCH)
983 
984 static inline void init_thread(struct target_pt_regs *regs,
985                                struct image_info *infop)
986 {
987     /*Set crmd PG,DA = 1,0 */
988     regs->csr.crmd = 2 << 3;
989     regs->csr.era = infop->entry;
990     regs->regs[3] = infop->start_stack;
991 }
992 
993 /* See linux kernel: arch/loongarch/include/asm/elf.h */
994 #define ELF_NREG 45
995 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
996 
997 enum {
998     TARGET_EF_R0 = 0,
999     TARGET_EF_CSR_ERA = TARGET_EF_R0 + 33,
1000     TARGET_EF_CSR_BADV = TARGET_EF_R0 + 34,
1001 };
1002 
1003 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1004                                const CPULoongArchState *env)
1005 {
1006     int i;
1007 
1008     (*regs)[TARGET_EF_R0] = 0;
1009 
1010     for (i = 1; i < ARRAY_SIZE(env->gpr); i++) {
1011         (*regs)[TARGET_EF_R0 + i] = tswapreg(env->gpr[i]);
1012     }
1013 
1014     (*regs)[TARGET_EF_CSR_ERA] = tswapreg(env->pc);
1015     (*regs)[TARGET_EF_CSR_BADV] = tswapreg(env->CSR_BADV);
1016 }
1017 
1018 #define USE_ELF_CORE_DUMP
1019 #define ELF_EXEC_PAGESIZE        4096
1020 
1021 #define ELF_HWCAP get_elf_hwcap()
1022 
1023 /* See arch/loongarch/include/uapi/asm/hwcap.h */
1024 enum {
1025     HWCAP_LOONGARCH_CPUCFG   = (1 << 0),
1026     HWCAP_LOONGARCH_LAM      = (1 << 1),
1027     HWCAP_LOONGARCH_UAL      = (1 << 2),
1028     HWCAP_LOONGARCH_FPU      = (1 << 3),
1029     HWCAP_LOONGARCH_LSX      = (1 << 4),
1030     HWCAP_LOONGARCH_LASX     = (1 << 5),
1031     HWCAP_LOONGARCH_CRC32    = (1 << 6),
1032     HWCAP_LOONGARCH_COMPLEX  = (1 << 7),
1033     HWCAP_LOONGARCH_CRYPTO   = (1 << 8),
1034     HWCAP_LOONGARCH_LVZ      = (1 << 9),
1035     HWCAP_LOONGARCH_LBT_X86  = (1 << 10),
1036     HWCAP_LOONGARCH_LBT_ARM  = (1 << 11),
1037     HWCAP_LOONGARCH_LBT_MIPS = (1 << 12),
1038 };
1039 
1040 static uint32_t get_elf_hwcap(void)
1041 {
1042     LoongArchCPU *cpu = LOONGARCH_CPU(thread_cpu);
1043     uint32_t hwcaps = 0;
1044 
1045     hwcaps |= HWCAP_LOONGARCH_CRC32;
1046 
1047     if (FIELD_EX32(cpu->env.cpucfg[1], CPUCFG1, UAL)) {
1048         hwcaps |= HWCAP_LOONGARCH_UAL;
1049     }
1050 
1051     if (FIELD_EX32(cpu->env.cpucfg[2], CPUCFG2, FP)) {
1052         hwcaps |= HWCAP_LOONGARCH_FPU;
1053     }
1054 
1055     if (FIELD_EX32(cpu->env.cpucfg[2], CPUCFG2, LAM)) {
1056         hwcaps |= HWCAP_LOONGARCH_LAM;
1057     }
1058 
1059     return hwcaps;
1060 }
1061 
1062 #define ELF_PLATFORM "loongarch"
1063 
1064 #endif /* TARGET_LOONGARCH64 */
1065 
1066 #ifdef TARGET_MIPS
1067 
1068 #define ELF_START_MMAP 0x80000000
1069 
1070 #ifdef TARGET_MIPS64
1071 #define ELF_CLASS   ELFCLASS64
1072 #else
1073 #define ELF_CLASS   ELFCLASS32
1074 #endif
1075 #define ELF_ARCH    EM_MIPS
1076 #define EXSTACK_DEFAULT true
1077 
1078 #ifdef TARGET_ABI_MIPSN32
1079 #define elf_check_abi(x) ((x) & EF_MIPS_ABI2)
1080 #else
1081 #define elf_check_abi(x) (!((x) & EF_MIPS_ABI2))
1082 #endif
1083 
1084 #define ELF_BASE_PLATFORM get_elf_base_platform()
1085 
1086 #define MATCH_PLATFORM_INSN(_flags, _base_platform)      \
1087     do { if ((cpu->env.insn_flags & (_flags)) == _flags) \
1088     { return _base_platform; } } while (0)
1089 
1090 static const char *get_elf_base_platform(void)
1091 {
1092     MIPSCPU *cpu = MIPS_CPU(thread_cpu);
1093 
1094     /* 64 bit ISAs goes first */
1095     MATCH_PLATFORM_INSN(CPU_MIPS64R6, "mips64r6");
1096     MATCH_PLATFORM_INSN(CPU_MIPS64R5, "mips64r5");
1097     MATCH_PLATFORM_INSN(CPU_MIPS64R2, "mips64r2");
1098     MATCH_PLATFORM_INSN(CPU_MIPS64R1, "mips64");
1099     MATCH_PLATFORM_INSN(CPU_MIPS5, "mips5");
1100     MATCH_PLATFORM_INSN(CPU_MIPS4, "mips4");
1101     MATCH_PLATFORM_INSN(CPU_MIPS3, "mips3");
1102 
1103     /* 32 bit ISAs */
1104     MATCH_PLATFORM_INSN(CPU_MIPS32R6, "mips32r6");
1105     MATCH_PLATFORM_INSN(CPU_MIPS32R5, "mips32r5");
1106     MATCH_PLATFORM_INSN(CPU_MIPS32R2, "mips32r2");
1107     MATCH_PLATFORM_INSN(CPU_MIPS32R1, "mips32");
1108     MATCH_PLATFORM_INSN(CPU_MIPS2, "mips2");
1109 
1110     /* Fallback */
1111     return "mips";
1112 }
1113 #undef MATCH_PLATFORM_INSN
1114 
1115 static inline void init_thread(struct target_pt_regs *regs,
1116                                struct image_info *infop)
1117 {
1118     regs->cp0_status = 2 << CP0St_KSU;
1119     regs->cp0_epc = infop->entry;
1120     regs->regs[29] = infop->start_stack;
1121 }
1122 
1123 /* See linux kernel: arch/mips/include/asm/elf.h.  */
1124 #define ELF_NREG 45
1125 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1126 
1127 /* See linux kernel: arch/mips/include/asm/reg.h.  */
1128 enum {
1129 #ifdef TARGET_MIPS64
1130     TARGET_EF_R0 = 0,
1131 #else
1132     TARGET_EF_R0 = 6,
1133 #endif
1134     TARGET_EF_R26 = TARGET_EF_R0 + 26,
1135     TARGET_EF_R27 = TARGET_EF_R0 + 27,
1136     TARGET_EF_LO = TARGET_EF_R0 + 32,
1137     TARGET_EF_HI = TARGET_EF_R0 + 33,
1138     TARGET_EF_CP0_EPC = TARGET_EF_R0 + 34,
1139     TARGET_EF_CP0_BADVADDR = TARGET_EF_R0 + 35,
1140     TARGET_EF_CP0_STATUS = TARGET_EF_R0 + 36,
1141     TARGET_EF_CP0_CAUSE = TARGET_EF_R0 + 37
1142 };
1143 
1144 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs.  */
1145 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMIPSState *env)
1146 {
1147     int i;
1148 
1149     for (i = 0; i < TARGET_EF_R0; i++) {
1150         (*regs)[i] = 0;
1151     }
1152     (*regs)[TARGET_EF_R0] = 0;
1153 
1154     for (i = 1; i < ARRAY_SIZE(env->active_tc.gpr); i++) {
1155         (*regs)[TARGET_EF_R0 + i] = tswapreg(env->active_tc.gpr[i]);
1156     }
1157 
1158     (*regs)[TARGET_EF_R26] = 0;
1159     (*regs)[TARGET_EF_R27] = 0;
1160     (*regs)[TARGET_EF_LO] = tswapreg(env->active_tc.LO[0]);
1161     (*regs)[TARGET_EF_HI] = tswapreg(env->active_tc.HI[0]);
1162     (*regs)[TARGET_EF_CP0_EPC] = tswapreg(env->active_tc.PC);
1163     (*regs)[TARGET_EF_CP0_BADVADDR] = tswapreg(env->CP0_BadVAddr);
1164     (*regs)[TARGET_EF_CP0_STATUS] = tswapreg(env->CP0_Status);
1165     (*regs)[TARGET_EF_CP0_CAUSE] = tswapreg(env->CP0_Cause);
1166 }
1167 
1168 #define USE_ELF_CORE_DUMP
1169 #define ELF_EXEC_PAGESIZE        4096
1170 
1171 /* See arch/mips/include/uapi/asm/hwcap.h.  */
1172 enum {
1173     HWCAP_MIPS_R6           = (1 << 0),
1174     HWCAP_MIPS_MSA          = (1 << 1),
1175     HWCAP_MIPS_CRC32        = (1 << 2),
1176     HWCAP_MIPS_MIPS16       = (1 << 3),
1177     HWCAP_MIPS_MDMX         = (1 << 4),
1178     HWCAP_MIPS_MIPS3D       = (1 << 5),
1179     HWCAP_MIPS_SMARTMIPS    = (1 << 6),
1180     HWCAP_MIPS_DSP          = (1 << 7),
1181     HWCAP_MIPS_DSP2         = (1 << 8),
1182     HWCAP_MIPS_DSP3         = (1 << 9),
1183     HWCAP_MIPS_MIPS16E2     = (1 << 10),
1184     HWCAP_LOONGSON_MMI      = (1 << 11),
1185     HWCAP_LOONGSON_EXT      = (1 << 12),
1186     HWCAP_LOONGSON_EXT2     = (1 << 13),
1187     HWCAP_LOONGSON_CPUCFG   = (1 << 14),
1188 };
1189 
1190 #define ELF_HWCAP get_elf_hwcap()
1191 
1192 #define GET_FEATURE_INSN(_flag, _hwcap) \
1193     do { if (cpu->env.insn_flags & (_flag)) { hwcaps |= _hwcap; } } while (0)
1194 
1195 #define GET_FEATURE_REG_SET(_reg, _mask, _hwcap) \
1196     do { if (cpu->env._reg & (_mask)) { hwcaps |= _hwcap; } } while (0)
1197 
1198 #define GET_FEATURE_REG_EQU(_reg, _start, _length, _val, _hwcap) \
1199     do { \
1200         if (extract32(cpu->env._reg, (_start), (_length)) == (_val)) { \
1201             hwcaps |= _hwcap; \
1202         } \
1203     } while (0)
1204 
1205 static uint32_t get_elf_hwcap(void)
1206 {
1207     MIPSCPU *cpu = MIPS_CPU(thread_cpu);
1208     uint32_t hwcaps = 0;
1209 
1210     GET_FEATURE_REG_EQU(CP0_Config0, CP0C0_AR, CP0C0_AR_LENGTH,
1211                         2, HWCAP_MIPS_R6);
1212     GET_FEATURE_REG_SET(CP0_Config3, 1 << CP0C3_MSAP, HWCAP_MIPS_MSA);
1213     GET_FEATURE_INSN(ASE_LMMI, HWCAP_LOONGSON_MMI);
1214     GET_FEATURE_INSN(ASE_LEXT, HWCAP_LOONGSON_EXT);
1215 
1216     return hwcaps;
1217 }
1218 
1219 #undef GET_FEATURE_REG_EQU
1220 #undef GET_FEATURE_REG_SET
1221 #undef GET_FEATURE_INSN
1222 
1223 #endif /* TARGET_MIPS */
1224 
1225 #ifdef TARGET_MICROBLAZE
1226 
1227 #define ELF_START_MMAP 0x80000000
1228 
1229 #define elf_check_arch(x) ( (x) == EM_MICROBLAZE || (x) == EM_MICROBLAZE_OLD)
1230 
1231 #define ELF_CLASS   ELFCLASS32
1232 #define ELF_ARCH    EM_MICROBLAZE
1233 
1234 static inline void init_thread(struct target_pt_regs *regs,
1235                                struct image_info *infop)
1236 {
1237     regs->pc = infop->entry;
1238     regs->r1 = infop->start_stack;
1239 
1240 }
1241 
1242 #define ELF_EXEC_PAGESIZE        4096
1243 
1244 #define USE_ELF_CORE_DUMP
1245 #define ELF_NREG 38
1246 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1247 
1248 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs.  */
1249 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMBState *env)
1250 {
1251     int i, pos = 0;
1252 
1253     for (i = 0; i < 32; i++) {
1254         (*regs)[pos++] = tswapreg(env->regs[i]);
1255     }
1256 
1257     (*regs)[pos++] = tswapreg(env->pc);
1258     (*regs)[pos++] = tswapreg(mb_cpu_read_msr(env));
1259     (*regs)[pos++] = 0;
1260     (*regs)[pos++] = tswapreg(env->ear);
1261     (*regs)[pos++] = 0;
1262     (*regs)[pos++] = tswapreg(env->esr);
1263 }
1264 
1265 #endif /* TARGET_MICROBLAZE */
1266 
1267 #ifdef TARGET_NIOS2
1268 
1269 #define ELF_START_MMAP 0x80000000
1270 
1271 #define elf_check_arch(x) ((x) == EM_ALTERA_NIOS2)
1272 
1273 #define ELF_CLASS   ELFCLASS32
1274 #define ELF_ARCH    EM_ALTERA_NIOS2
1275 
1276 static void init_thread(struct target_pt_regs *regs, struct image_info *infop)
1277 {
1278     regs->ea = infop->entry;
1279     regs->sp = infop->start_stack;
1280 }
1281 
1282 #define LO_COMMPAGE  TARGET_PAGE_SIZE
1283 
1284 static bool init_guest_commpage(void)
1285 {
1286     static const uint8_t kuser_page[4 + 2 * 64] = {
1287         /* __kuser_helper_version */
1288         [0x00] = 0x02, 0x00, 0x00, 0x00,
1289 
1290         /* __kuser_cmpxchg */
1291         [0x04] = 0x3a, 0x6c, 0x3b, 0x00,  /* trap 16 */
1292                  0x3a, 0x28, 0x00, 0xf8,  /* ret */
1293 
1294         /* __kuser_sigtramp */
1295         [0x44] = 0xc4, 0x22, 0x80, 0x00,  /* movi r2, __NR_rt_sigreturn */
1296                  0x3a, 0x68, 0x3b, 0x00,  /* trap 0 */
1297     };
1298 
1299     void *want = g2h_untagged(LO_COMMPAGE & -qemu_host_page_size);
1300     void *addr = mmap(want, qemu_host_page_size, PROT_READ | PROT_WRITE,
1301                       MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0);
1302 
1303     if (addr == MAP_FAILED) {
1304         perror("Allocating guest commpage");
1305         exit(EXIT_FAILURE);
1306     }
1307     if (addr != want) {
1308         return false;
1309     }
1310 
1311     memcpy(addr, kuser_page, sizeof(kuser_page));
1312 
1313     if (mprotect(addr, qemu_host_page_size, PROT_READ)) {
1314         perror("Protecting guest commpage");
1315         exit(EXIT_FAILURE);
1316     }
1317 
1318     page_set_flags(LO_COMMPAGE, LO_COMMPAGE + TARGET_PAGE_SIZE,
1319                    PAGE_READ | PAGE_EXEC | PAGE_VALID);
1320     return true;
1321 }
1322 
1323 #define ELF_EXEC_PAGESIZE        4096
1324 
1325 #define USE_ELF_CORE_DUMP
1326 #define ELF_NREG 49
1327 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1328 
1329 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs.  */
1330 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1331                                const CPUNios2State *env)
1332 {
1333     int i;
1334 
1335     (*regs)[0] = -1;
1336     for (i = 1; i < 8; i++)    /* r0-r7 */
1337         (*regs)[i] = tswapreg(env->regs[i + 7]);
1338 
1339     for (i = 8; i < 16; i++)   /* r8-r15 */
1340         (*regs)[i] = tswapreg(env->regs[i - 8]);
1341 
1342     for (i = 16; i < 24; i++)  /* r16-r23 */
1343         (*regs)[i] = tswapreg(env->regs[i + 7]);
1344     (*regs)[24] = -1;    /* R_ET */
1345     (*regs)[25] = -1;    /* R_BT */
1346     (*regs)[26] = tswapreg(env->regs[R_GP]);
1347     (*regs)[27] = tswapreg(env->regs[R_SP]);
1348     (*regs)[28] = tswapreg(env->regs[R_FP]);
1349     (*regs)[29] = tswapreg(env->regs[R_EA]);
1350     (*regs)[30] = -1;    /* R_SSTATUS */
1351     (*regs)[31] = tswapreg(env->regs[R_RA]);
1352 
1353     (*regs)[32] = tswapreg(env->pc);
1354 
1355     (*regs)[33] = -1; /* R_STATUS */
1356     (*regs)[34] = tswapreg(env->regs[CR_ESTATUS]);
1357 
1358     for (i = 35; i < 49; i++)    /* ... */
1359         (*regs)[i] = -1;
1360 }
1361 
1362 #endif /* TARGET_NIOS2 */
1363 
1364 #ifdef TARGET_OPENRISC
1365 
1366 #define ELF_START_MMAP 0x08000000
1367 
1368 #define ELF_ARCH EM_OPENRISC
1369 #define ELF_CLASS ELFCLASS32
1370 #define ELF_DATA  ELFDATA2MSB
1371 
1372 static inline void init_thread(struct target_pt_regs *regs,
1373                                struct image_info *infop)
1374 {
1375     regs->pc = infop->entry;
1376     regs->gpr[1] = infop->start_stack;
1377 }
1378 
1379 #define USE_ELF_CORE_DUMP
1380 #define ELF_EXEC_PAGESIZE 8192
1381 
1382 /* See linux kernel arch/openrisc/include/asm/elf.h.  */
1383 #define ELF_NREG 34 /* gprs and pc, sr */
1384 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1385 
1386 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1387                                const CPUOpenRISCState *env)
1388 {
1389     int i;
1390 
1391     for (i = 0; i < 32; i++) {
1392         (*regs)[i] = tswapreg(cpu_get_gpr(env, i));
1393     }
1394     (*regs)[32] = tswapreg(env->pc);
1395     (*regs)[33] = tswapreg(cpu_get_sr(env));
1396 }
1397 #define ELF_HWCAP 0
1398 #define ELF_PLATFORM NULL
1399 
1400 #endif /* TARGET_OPENRISC */
1401 
1402 #ifdef TARGET_SH4
1403 
1404 #define ELF_START_MMAP 0x80000000
1405 
1406 #define ELF_CLASS ELFCLASS32
1407 #define ELF_ARCH  EM_SH
1408 
1409 static inline void init_thread(struct target_pt_regs *regs,
1410                                struct image_info *infop)
1411 {
1412     /* Check other registers XXXXX */
1413     regs->pc = infop->entry;
1414     regs->regs[15] = infop->start_stack;
1415 }
1416 
1417 /* See linux kernel: arch/sh/include/asm/elf.h.  */
1418 #define ELF_NREG 23
1419 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1420 
1421 /* See linux kernel: arch/sh/include/asm/ptrace.h.  */
1422 enum {
1423     TARGET_REG_PC = 16,
1424     TARGET_REG_PR = 17,
1425     TARGET_REG_SR = 18,
1426     TARGET_REG_GBR = 19,
1427     TARGET_REG_MACH = 20,
1428     TARGET_REG_MACL = 21,
1429     TARGET_REG_SYSCALL = 22
1430 };
1431 
1432 static inline void elf_core_copy_regs(target_elf_gregset_t *regs,
1433                                       const CPUSH4State *env)
1434 {
1435     int i;
1436 
1437     for (i = 0; i < 16; i++) {
1438         (*regs)[i] = tswapreg(env->gregs[i]);
1439     }
1440 
1441     (*regs)[TARGET_REG_PC] = tswapreg(env->pc);
1442     (*regs)[TARGET_REG_PR] = tswapreg(env->pr);
1443     (*regs)[TARGET_REG_SR] = tswapreg(env->sr);
1444     (*regs)[TARGET_REG_GBR] = tswapreg(env->gbr);
1445     (*regs)[TARGET_REG_MACH] = tswapreg(env->mach);
1446     (*regs)[TARGET_REG_MACL] = tswapreg(env->macl);
1447     (*regs)[TARGET_REG_SYSCALL] = 0; /* FIXME */
1448 }
1449 
1450 #define USE_ELF_CORE_DUMP
1451 #define ELF_EXEC_PAGESIZE        4096
1452 
1453 enum {
1454     SH_CPU_HAS_FPU            = 0x0001, /* Hardware FPU support */
1455     SH_CPU_HAS_P2_FLUSH_BUG   = 0x0002, /* Need to flush the cache in P2 area */
1456     SH_CPU_HAS_MMU_PAGE_ASSOC = 0x0004, /* SH3: TLB way selection bit support */
1457     SH_CPU_HAS_DSP            = 0x0008, /* SH-DSP: DSP support */
1458     SH_CPU_HAS_PERF_COUNTER   = 0x0010, /* Hardware performance counters */
1459     SH_CPU_HAS_PTEA           = 0x0020, /* PTEA register */
1460     SH_CPU_HAS_LLSC           = 0x0040, /* movli.l/movco.l */
1461     SH_CPU_HAS_L2_CACHE       = 0x0080, /* Secondary cache / URAM */
1462     SH_CPU_HAS_OP32           = 0x0100, /* 32-bit instruction support */
1463     SH_CPU_HAS_PTEAEX         = 0x0200, /* PTE ASID Extension support */
1464 };
1465 
1466 #define ELF_HWCAP get_elf_hwcap()
1467 
1468 static uint32_t get_elf_hwcap(void)
1469 {
1470     SuperHCPU *cpu = SUPERH_CPU(thread_cpu);
1471     uint32_t hwcap = 0;
1472 
1473     hwcap |= SH_CPU_HAS_FPU;
1474 
1475     if (cpu->env.features & SH_FEATURE_SH4A) {
1476         hwcap |= SH_CPU_HAS_LLSC;
1477     }
1478 
1479     return hwcap;
1480 }
1481 
1482 #endif
1483 
1484 #ifdef TARGET_CRIS
1485 
1486 #define ELF_START_MMAP 0x80000000
1487 
1488 #define ELF_CLASS ELFCLASS32
1489 #define ELF_ARCH  EM_CRIS
1490 
1491 static inline void init_thread(struct target_pt_regs *regs,
1492                                struct image_info *infop)
1493 {
1494     regs->erp = infop->entry;
1495 }
1496 
1497 #define ELF_EXEC_PAGESIZE        8192
1498 
1499 #endif
1500 
1501 #ifdef TARGET_M68K
1502 
1503 #define ELF_START_MMAP 0x80000000
1504 
1505 #define ELF_CLASS       ELFCLASS32
1506 #define ELF_ARCH        EM_68K
1507 
1508 /* ??? Does this need to do anything?
1509    #define ELF_PLAT_INIT(_r) */
1510 
1511 static inline void init_thread(struct target_pt_regs *regs,
1512                                struct image_info *infop)
1513 {
1514     regs->usp = infop->start_stack;
1515     regs->sr = 0;
1516     regs->pc = infop->entry;
1517 }
1518 
1519 /* See linux kernel: arch/m68k/include/asm/elf.h.  */
1520 #define ELF_NREG 20
1521 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1522 
1523 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUM68KState *env)
1524 {
1525     (*regs)[0] = tswapreg(env->dregs[1]);
1526     (*regs)[1] = tswapreg(env->dregs[2]);
1527     (*regs)[2] = tswapreg(env->dregs[3]);
1528     (*regs)[3] = tswapreg(env->dregs[4]);
1529     (*regs)[4] = tswapreg(env->dregs[5]);
1530     (*regs)[5] = tswapreg(env->dregs[6]);
1531     (*regs)[6] = tswapreg(env->dregs[7]);
1532     (*regs)[7] = tswapreg(env->aregs[0]);
1533     (*regs)[8] = tswapreg(env->aregs[1]);
1534     (*regs)[9] = tswapreg(env->aregs[2]);
1535     (*regs)[10] = tswapreg(env->aregs[3]);
1536     (*regs)[11] = tswapreg(env->aregs[4]);
1537     (*regs)[12] = tswapreg(env->aregs[5]);
1538     (*regs)[13] = tswapreg(env->aregs[6]);
1539     (*regs)[14] = tswapreg(env->dregs[0]);
1540     (*regs)[15] = tswapreg(env->aregs[7]);
1541     (*regs)[16] = tswapreg(env->dregs[0]); /* FIXME: orig_d0 */
1542     (*regs)[17] = tswapreg(env->sr);
1543     (*regs)[18] = tswapreg(env->pc);
1544     (*regs)[19] = 0;  /* FIXME: regs->format | regs->vector */
1545 }
1546 
1547 #define USE_ELF_CORE_DUMP
1548 #define ELF_EXEC_PAGESIZE       8192
1549 
1550 #endif
1551 
1552 #ifdef TARGET_ALPHA
1553 
1554 #define ELF_START_MMAP (0x30000000000ULL)
1555 
1556 #define ELF_CLASS      ELFCLASS64
1557 #define ELF_ARCH       EM_ALPHA
1558 
1559 static inline void init_thread(struct target_pt_regs *regs,
1560                                struct image_info *infop)
1561 {
1562     regs->pc = infop->entry;
1563     regs->ps = 8;
1564     regs->usp = infop->start_stack;
1565 }
1566 
1567 #define ELF_EXEC_PAGESIZE        8192
1568 
1569 #endif /* TARGET_ALPHA */
1570 
1571 #ifdef TARGET_S390X
1572 
1573 #define ELF_START_MMAP (0x20000000000ULL)
1574 
1575 #define ELF_CLASS	ELFCLASS64
1576 #define ELF_DATA	ELFDATA2MSB
1577 #define ELF_ARCH	EM_S390
1578 
1579 #include "elf.h"
1580 
1581 #define ELF_HWCAP get_elf_hwcap()
1582 
1583 #define GET_FEATURE(_feat, _hwcap) \
1584     do { if (s390_has_feat(_feat)) { hwcap |= _hwcap; } } while (0)
1585 
1586 static uint32_t get_elf_hwcap(void)
1587 {
1588     /*
1589      * Let's assume we always have esan3 and zarch.
1590      * 31-bit processes can use 64-bit registers (high gprs).
1591      */
1592     uint32_t hwcap = HWCAP_S390_ESAN3 | HWCAP_S390_ZARCH | HWCAP_S390_HIGH_GPRS;
1593 
1594     GET_FEATURE(S390_FEAT_STFLE, HWCAP_S390_STFLE);
1595     GET_FEATURE(S390_FEAT_MSA, HWCAP_S390_MSA);
1596     GET_FEATURE(S390_FEAT_LONG_DISPLACEMENT, HWCAP_S390_LDISP);
1597     GET_FEATURE(S390_FEAT_EXTENDED_IMMEDIATE, HWCAP_S390_EIMM);
1598     if (s390_has_feat(S390_FEAT_EXTENDED_TRANSLATION_3) &&
1599         s390_has_feat(S390_FEAT_ETF3_ENH)) {
1600         hwcap |= HWCAP_S390_ETF3EH;
1601     }
1602     GET_FEATURE(S390_FEAT_VECTOR, HWCAP_S390_VXRS);
1603     GET_FEATURE(S390_FEAT_VECTOR_ENH, HWCAP_S390_VXRS_EXT);
1604 
1605     return hwcap;
1606 }
1607 
1608 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop)
1609 {
1610     regs->psw.addr = infop->entry;
1611     regs->psw.mask = PSW_MASK_64 | PSW_MASK_32;
1612     regs->gprs[15] = infop->start_stack;
1613 }
1614 
1615 /* See linux kernel: arch/s390/include/uapi/asm/ptrace.h (s390_regs).  */
1616 #define ELF_NREG 27
1617 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1618 
1619 enum {
1620     TARGET_REG_PSWM = 0,
1621     TARGET_REG_PSWA = 1,
1622     TARGET_REG_GPRS = 2,
1623     TARGET_REG_ARS = 18,
1624     TARGET_REG_ORIG_R2 = 26,
1625 };
1626 
1627 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1628                                const CPUS390XState *env)
1629 {
1630     int i;
1631     uint32_t *aregs;
1632 
1633     (*regs)[TARGET_REG_PSWM] = tswapreg(env->psw.mask);
1634     (*regs)[TARGET_REG_PSWA] = tswapreg(env->psw.addr);
1635     for (i = 0; i < 16; i++) {
1636         (*regs)[TARGET_REG_GPRS + i] = tswapreg(env->regs[i]);
1637     }
1638     aregs = (uint32_t *)&((*regs)[TARGET_REG_ARS]);
1639     for (i = 0; i < 16; i++) {
1640         aregs[i] = tswap32(env->aregs[i]);
1641     }
1642     (*regs)[TARGET_REG_ORIG_R2] = 0;
1643 }
1644 
1645 #define USE_ELF_CORE_DUMP
1646 #define ELF_EXEC_PAGESIZE 4096
1647 
1648 #endif /* TARGET_S390X */
1649 
1650 #ifdef TARGET_RISCV
1651 
1652 #define ELF_START_MMAP 0x80000000
1653 #define ELF_ARCH  EM_RISCV
1654 
1655 #ifdef TARGET_RISCV32
1656 #define ELF_CLASS ELFCLASS32
1657 #else
1658 #define ELF_CLASS ELFCLASS64
1659 #endif
1660 
1661 #define ELF_HWCAP get_elf_hwcap()
1662 
1663 static uint32_t get_elf_hwcap(void)
1664 {
1665 #define MISA_BIT(EXT) (1 << (EXT - 'A'))
1666     RISCVCPU *cpu = RISCV_CPU(thread_cpu);
1667     uint32_t mask = MISA_BIT('I') | MISA_BIT('M') | MISA_BIT('A')
1668                     | MISA_BIT('F') | MISA_BIT('D') | MISA_BIT('C');
1669 
1670     return cpu->env.misa_ext & mask;
1671 #undef MISA_BIT
1672 }
1673 
1674 static inline void init_thread(struct target_pt_regs *regs,
1675                                struct image_info *infop)
1676 {
1677     regs->sepc = infop->entry;
1678     regs->sp = infop->start_stack;
1679 }
1680 
1681 #define ELF_EXEC_PAGESIZE 4096
1682 
1683 #endif /* TARGET_RISCV */
1684 
1685 #ifdef TARGET_HPPA
1686 
1687 #define ELF_START_MMAP  0x80000000
1688 #define ELF_CLASS       ELFCLASS32
1689 #define ELF_ARCH        EM_PARISC
1690 #define ELF_PLATFORM    "PARISC"
1691 #define STACK_GROWS_DOWN 0
1692 #define STACK_ALIGNMENT  64
1693 
1694 static inline void init_thread(struct target_pt_regs *regs,
1695                                struct image_info *infop)
1696 {
1697     regs->iaoq[0] = infop->entry;
1698     regs->iaoq[1] = infop->entry + 4;
1699     regs->gr[23] = 0;
1700     regs->gr[24] = infop->argv;
1701     regs->gr[25] = infop->argc;
1702     /* The top-of-stack contains a linkage buffer.  */
1703     regs->gr[30] = infop->start_stack + 64;
1704     regs->gr[31] = infop->entry;
1705 }
1706 
1707 #define LO_COMMPAGE  0
1708 
1709 static bool init_guest_commpage(void)
1710 {
1711     void *want = g2h_untagged(LO_COMMPAGE);
1712     void *addr = mmap(want, qemu_host_page_size, PROT_NONE,
1713                       MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0);
1714 
1715     if (addr == MAP_FAILED) {
1716         perror("Allocating guest commpage");
1717         exit(EXIT_FAILURE);
1718     }
1719     if (addr != want) {
1720         return false;
1721     }
1722 
1723     /*
1724      * On Linux, page zero is normally marked execute only + gateway.
1725      * Normal read or write is supposed to fail (thus PROT_NONE above),
1726      * but specific offsets have kernel code mapped to raise permissions
1727      * and implement syscalls.  Here, simply mark the page executable.
1728      * Special case the entry points during translation (see do_page_zero).
1729      */
1730     page_set_flags(LO_COMMPAGE, LO_COMMPAGE + TARGET_PAGE_SIZE,
1731                    PAGE_EXEC | PAGE_VALID);
1732     return true;
1733 }
1734 
1735 #endif /* TARGET_HPPA */
1736 
1737 #ifdef TARGET_XTENSA
1738 
1739 #define ELF_START_MMAP 0x20000000
1740 
1741 #define ELF_CLASS       ELFCLASS32
1742 #define ELF_ARCH        EM_XTENSA
1743 
1744 static inline void init_thread(struct target_pt_regs *regs,
1745                                struct image_info *infop)
1746 {
1747     regs->windowbase = 0;
1748     regs->windowstart = 1;
1749     regs->areg[1] = infop->start_stack;
1750     regs->pc = infop->entry;
1751 }
1752 
1753 /* See linux kernel: arch/xtensa/include/asm/elf.h.  */
1754 #define ELF_NREG 128
1755 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1756 
1757 enum {
1758     TARGET_REG_PC,
1759     TARGET_REG_PS,
1760     TARGET_REG_LBEG,
1761     TARGET_REG_LEND,
1762     TARGET_REG_LCOUNT,
1763     TARGET_REG_SAR,
1764     TARGET_REG_WINDOWSTART,
1765     TARGET_REG_WINDOWBASE,
1766     TARGET_REG_THREADPTR,
1767     TARGET_REG_AR0 = 64,
1768 };
1769 
1770 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1771                                const CPUXtensaState *env)
1772 {
1773     unsigned i;
1774 
1775     (*regs)[TARGET_REG_PC] = tswapreg(env->pc);
1776     (*regs)[TARGET_REG_PS] = tswapreg(env->sregs[PS] & ~PS_EXCM);
1777     (*regs)[TARGET_REG_LBEG] = tswapreg(env->sregs[LBEG]);
1778     (*regs)[TARGET_REG_LEND] = tswapreg(env->sregs[LEND]);
1779     (*regs)[TARGET_REG_LCOUNT] = tswapreg(env->sregs[LCOUNT]);
1780     (*regs)[TARGET_REG_SAR] = tswapreg(env->sregs[SAR]);
1781     (*regs)[TARGET_REG_WINDOWSTART] = tswapreg(env->sregs[WINDOW_START]);
1782     (*regs)[TARGET_REG_WINDOWBASE] = tswapreg(env->sregs[WINDOW_BASE]);
1783     (*regs)[TARGET_REG_THREADPTR] = tswapreg(env->uregs[THREADPTR]);
1784     xtensa_sync_phys_from_window((CPUXtensaState *)env);
1785     for (i = 0; i < env->config->nareg; ++i) {
1786         (*regs)[TARGET_REG_AR0 + i] = tswapreg(env->phys_regs[i]);
1787     }
1788 }
1789 
1790 #define USE_ELF_CORE_DUMP
1791 #define ELF_EXEC_PAGESIZE       4096
1792 
1793 #endif /* TARGET_XTENSA */
1794 
1795 #ifdef TARGET_HEXAGON
1796 
1797 #define ELF_START_MMAP 0x20000000
1798 
1799 #define ELF_CLASS       ELFCLASS32
1800 #define ELF_ARCH        EM_HEXAGON
1801 
1802 static inline void init_thread(struct target_pt_regs *regs,
1803                                struct image_info *infop)
1804 {
1805     regs->sepc = infop->entry;
1806     regs->sp = infop->start_stack;
1807 }
1808 
1809 #endif /* TARGET_HEXAGON */
1810 
1811 #ifndef ELF_BASE_PLATFORM
1812 #define ELF_BASE_PLATFORM (NULL)
1813 #endif
1814 
1815 #ifndef ELF_PLATFORM
1816 #define ELF_PLATFORM (NULL)
1817 #endif
1818 
1819 #ifndef ELF_MACHINE
1820 #define ELF_MACHINE ELF_ARCH
1821 #endif
1822 
1823 #ifndef elf_check_arch
1824 #define elf_check_arch(x) ((x) == ELF_ARCH)
1825 #endif
1826 
1827 #ifndef elf_check_abi
1828 #define elf_check_abi(x) (1)
1829 #endif
1830 
1831 #ifndef ELF_HWCAP
1832 #define ELF_HWCAP 0
1833 #endif
1834 
1835 #ifndef STACK_GROWS_DOWN
1836 #define STACK_GROWS_DOWN 1
1837 #endif
1838 
1839 #ifndef STACK_ALIGNMENT
1840 #define STACK_ALIGNMENT 16
1841 #endif
1842 
1843 #ifdef TARGET_ABI32
1844 #undef ELF_CLASS
1845 #define ELF_CLASS ELFCLASS32
1846 #undef bswaptls
1847 #define bswaptls(ptr) bswap32s(ptr)
1848 #endif
1849 
1850 #ifndef EXSTACK_DEFAULT
1851 #define EXSTACK_DEFAULT false
1852 #endif
1853 
1854 #include "elf.h"
1855 
1856 /* We must delay the following stanzas until after "elf.h". */
1857 #if defined(TARGET_AARCH64)
1858 
1859 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz,
1860                                     const uint32_t *data,
1861                                     struct image_info *info,
1862                                     Error **errp)
1863 {
1864     if (pr_type == GNU_PROPERTY_AARCH64_FEATURE_1_AND) {
1865         if (pr_datasz != sizeof(uint32_t)) {
1866             error_setg(errp, "Ill-formed GNU_PROPERTY_AARCH64_FEATURE_1_AND");
1867             return false;
1868         }
1869         /* We will extract GNU_PROPERTY_AARCH64_FEATURE_1_BTI later. */
1870         info->note_flags = *data;
1871     }
1872     return true;
1873 }
1874 #define ARCH_USE_GNU_PROPERTY 1
1875 
1876 #else
1877 
1878 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz,
1879                                     const uint32_t *data,
1880                                     struct image_info *info,
1881                                     Error **errp)
1882 {
1883     g_assert_not_reached();
1884 }
1885 #define ARCH_USE_GNU_PROPERTY 0
1886 
1887 #endif
1888 
1889 struct exec
1890 {
1891     unsigned int a_info;   /* Use macros N_MAGIC, etc for access */
1892     unsigned int a_text;   /* length of text, in bytes */
1893     unsigned int a_data;   /* length of data, in bytes */
1894     unsigned int a_bss;    /* length of uninitialized data area, in bytes */
1895     unsigned int a_syms;   /* length of symbol table data in file, in bytes */
1896     unsigned int a_entry;  /* start address */
1897     unsigned int a_trsize; /* length of relocation info for text, in bytes */
1898     unsigned int a_drsize; /* length of relocation info for data, in bytes */
1899 };
1900 
1901 
1902 #define N_MAGIC(exec) ((exec).a_info & 0xffff)
1903 #define OMAGIC 0407
1904 #define NMAGIC 0410
1905 #define ZMAGIC 0413
1906 #define QMAGIC 0314
1907 
1908 /* Necessary parameters */
1909 #define TARGET_ELF_EXEC_PAGESIZE \
1910         (((eppnt->p_align & ~qemu_host_page_mask) != 0) ? \
1911          TARGET_PAGE_SIZE : MAX(qemu_host_page_size, TARGET_PAGE_SIZE))
1912 #define TARGET_ELF_PAGELENGTH(_v) ROUND_UP((_v), TARGET_ELF_EXEC_PAGESIZE)
1913 #define TARGET_ELF_PAGESTART(_v) ((_v) & \
1914                                  ~(abi_ulong)(TARGET_ELF_EXEC_PAGESIZE-1))
1915 #define TARGET_ELF_PAGEOFFSET(_v) ((_v) & (TARGET_ELF_EXEC_PAGESIZE-1))
1916 
1917 #define DLINFO_ITEMS 16
1918 
1919 static inline void memcpy_fromfs(void * to, const void * from, unsigned long n)
1920 {
1921     memcpy(to, from, n);
1922 }
1923 
1924 #ifdef BSWAP_NEEDED
1925 static void bswap_ehdr(struct elfhdr *ehdr)
1926 {
1927     bswap16s(&ehdr->e_type);            /* Object file type */
1928     bswap16s(&ehdr->e_machine);         /* Architecture */
1929     bswap32s(&ehdr->e_version);         /* Object file version */
1930     bswaptls(&ehdr->e_entry);           /* Entry point virtual address */
1931     bswaptls(&ehdr->e_phoff);           /* Program header table file offset */
1932     bswaptls(&ehdr->e_shoff);           /* Section header table file offset */
1933     bswap32s(&ehdr->e_flags);           /* Processor-specific flags */
1934     bswap16s(&ehdr->e_ehsize);          /* ELF header size in bytes */
1935     bswap16s(&ehdr->e_phentsize);       /* Program header table entry size */
1936     bswap16s(&ehdr->e_phnum);           /* Program header table entry count */
1937     bswap16s(&ehdr->e_shentsize);       /* Section header table entry size */
1938     bswap16s(&ehdr->e_shnum);           /* Section header table entry count */
1939     bswap16s(&ehdr->e_shstrndx);        /* Section header string table index */
1940 }
1941 
1942 static void bswap_phdr(struct elf_phdr *phdr, int phnum)
1943 {
1944     int i;
1945     for (i = 0; i < phnum; ++i, ++phdr) {
1946         bswap32s(&phdr->p_type);        /* Segment type */
1947         bswap32s(&phdr->p_flags);       /* Segment flags */
1948         bswaptls(&phdr->p_offset);      /* Segment file offset */
1949         bswaptls(&phdr->p_vaddr);       /* Segment virtual address */
1950         bswaptls(&phdr->p_paddr);       /* Segment physical address */
1951         bswaptls(&phdr->p_filesz);      /* Segment size in file */
1952         bswaptls(&phdr->p_memsz);       /* Segment size in memory */
1953         bswaptls(&phdr->p_align);       /* Segment alignment */
1954     }
1955 }
1956 
1957 static void bswap_shdr(struct elf_shdr *shdr, int shnum)
1958 {
1959     int i;
1960     for (i = 0; i < shnum; ++i, ++shdr) {
1961         bswap32s(&shdr->sh_name);
1962         bswap32s(&shdr->sh_type);
1963         bswaptls(&shdr->sh_flags);
1964         bswaptls(&shdr->sh_addr);
1965         bswaptls(&shdr->sh_offset);
1966         bswaptls(&shdr->sh_size);
1967         bswap32s(&shdr->sh_link);
1968         bswap32s(&shdr->sh_info);
1969         bswaptls(&shdr->sh_addralign);
1970         bswaptls(&shdr->sh_entsize);
1971     }
1972 }
1973 
1974 static void bswap_sym(struct elf_sym *sym)
1975 {
1976     bswap32s(&sym->st_name);
1977     bswaptls(&sym->st_value);
1978     bswaptls(&sym->st_size);
1979     bswap16s(&sym->st_shndx);
1980 }
1981 
1982 #ifdef TARGET_MIPS
1983 static void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags)
1984 {
1985     bswap16s(&abiflags->version);
1986     bswap32s(&abiflags->ases);
1987     bswap32s(&abiflags->isa_ext);
1988     bswap32s(&abiflags->flags1);
1989     bswap32s(&abiflags->flags2);
1990 }
1991 #endif
1992 #else
1993 static inline void bswap_ehdr(struct elfhdr *ehdr) { }
1994 static inline void bswap_phdr(struct elf_phdr *phdr, int phnum) { }
1995 static inline void bswap_shdr(struct elf_shdr *shdr, int shnum) { }
1996 static inline void bswap_sym(struct elf_sym *sym) { }
1997 #ifdef TARGET_MIPS
1998 static inline void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags) { }
1999 #endif
2000 #endif
2001 
2002 #ifdef USE_ELF_CORE_DUMP
2003 static int elf_core_dump(int, const CPUArchState *);
2004 #endif /* USE_ELF_CORE_DUMP */
2005 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias);
2006 
2007 /* Verify the portions of EHDR within E_IDENT for the target.
2008    This can be performed before bswapping the entire header.  */
2009 static bool elf_check_ident(struct elfhdr *ehdr)
2010 {
2011     return (ehdr->e_ident[EI_MAG0] == ELFMAG0
2012             && ehdr->e_ident[EI_MAG1] == ELFMAG1
2013             && ehdr->e_ident[EI_MAG2] == ELFMAG2
2014             && ehdr->e_ident[EI_MAG3] == ELFMAG3
2015             && ehdr->e_ident[EI_CLASS] == ELF_CLASS
2016             && ehdr->e_ident[EI_DATA] == ELF_DATA
2017             && ehdr->e_ident[EI_VERSION] == EV_CURRENT);
2018 }
2019 
2020 /* Verify the portions of EHDR outside of E_IDENT for the target.
2021    This has to wait until after bswapping the header.  */
2022 static bool elf_check_ehdr(struct elfhdr *ehdr)
2023 {
2024     return (elf_check_arch(ehdr->e_machine)
2025             && elf_check_abi(ehdr->e_flags)
2026             && ehdr->e_ehsize == sizeof(struct elfhdr)
2027             && ehdr->e_phentsize == sizeof(struct elf_phdr)
2028             && (ehdr->e_type == ET_EXEC || ehdr->e_type == ET_DYN));
2029 }
2030 
2031 /*
2032  * 'copy_elf_strings()' copies argument/envelope strings from user
2033  * memory to free pages in kernel mem. These are in a format ready
2034  * to be put directly into the top of new user memory.
2035  *
2036  */
2037 static abi_ulong copy_elf_strings(int argc, char **argv, char *scratch,
2038                                   abi_ulong p, abi_ulong stack_limit)
2039 {
2040     char *tmp;
2041     int len, i;
2042     abi_ulong top = p;
2043 
2044     if (!p) {
2045         return 0;       /* bullet-proofing */
2046     }
2047 
2048     if (STACK_GROWS_DOWN) {
2049         int offset = ((p - 1) % TARGET_PAGE_SIZE) + 1;
2050         for (i = argc - 1; i >= 0; --i) {
2051             tmp = argv[i];
2052             if (!tmp) {
2053                 fprintf(stderr, "VFS: argc is wrong");
2054                 exit(-1);
2055             }
2056             len = strlen(tmp) + 1;
2057             tmp += len;
2058 
2059             if (len > (p - stack_limit)) {
2060                 return 0;
2061             }
2062             while (len) {
2063                 int bytes_to_copy = (len > offset) ? offset : len;
2064                 tmp -= bytes_to_copy;
2065                 p -= bytes_to_copy;
2066                 offset -= bytes_to_copy;
2067                 len -= bytes_to_copy;
2068 
2069                 memcpy_fromfs(scratch + offset, tmp, bytes_to_copy);
2070 
2071                 if (offset == 0) {
2072                     memcpy_to_target(p, scratch, top - p);
2073                     top = p;
2074                     offset = TARGET_PAGE_SIZE;
2075                 }
2076             }
2077         }
2078         if (p != top) {
2079             memcpy_to_target(p, scratch + offset, top - p);
2080         }
2081     } else {
2082         int remaining = TARGET_PAGE_SIZE - (p % TARGET_PAGE_SIZE);
2083         for (i = 0; i < argc; ++i) {
2084             tmp = argv[i];
2085             if (!tmp) {
2086                 fprintf(stderr, "VFS: argc is wrong");
2087                 exit(-1);
2088             }
2089             len = strlen(tmp) + 1;
2090             if (len > (stack_limit - p)) {
2091                 return 0;
2092             }
2093             while (len) {
2094                 int bytes_to_copy = (len > remaining) ? remaining : len;
2095 
2096                 memcpy_fromfs(scratch + (p - top), tmp, bytes_to_copy);
2097 
2098                 tmp += bytes_to_copy;
2099                 remaining -= bytes_to_copy;
2100                 p += bytes_to_copy;
2101                 len -= bytes_to_copy;
2102 
2103                 if (remaining == 0) {
2104                     memcpy_to_target(top, scratch, p - top);
2105                     top = p;
2106                     remaining = TARGET_PAGE_SIZE;
2107                 }
2108             }
2109         }
2110         if (p != top) {
2111             memcpy_to_target(top, scratch, p - top);
2112         }
2113     }
2114 
2115     return p;
2116 }
2117 
2118 /* Older linux kernels provide up to MAX_ARG_PAGES (default: 32) of
2119  * argument/environment space. Newer kernels (>2.6.33) allow more,
2120  * dependent on stack size, but guarantee at least 32 pages for
2121  * backwards compatibility.
2122  */
2123 #define STACK_LOWER_LIMIT (32 * TARGET_PAGE_SIZE)
2124 
2125 static abi_ulong setup_arg_pages(struct linux_binprm *bprm,
2126                                  struct image_info *info)
2127 {
2128     abi_ulong size, error, guard;
2129     int prot;
2130 
2131     size = guest_stack_size;
2132     if (size < STACK_LOWER_LIMIT) {
2133         size = STACK_LOWER_LIMIT;
2134     }
2135 
2136     if (STACK_GROWS_DOWN) {
2137         guard = TARGET_PAGE_SIZE;
2138         if (guard < qemu_real_host_page_size()) {
2139             guard = qemu_real_host_page_size();
2140         }
2141     } else {
2142         /* no guard page for hppa target where stack grows upwards. */
2143         guard = 0;
2144     }
2145 
2146     prot = PROT_READ | PROT_WRITE;
2147     if (info->exec_stack) {
2148         prot |= PROT_EXEC;
2149     }
2150     error = target_mmap(0, size + guard, prot,
2151                         MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
2152     if (error == -1) {
2153         perror("mmap stack");
2154         exit(-1);
2155     }
2156 
2157     /* We reserve one extra page at the top of the stack as guard.  */
2158     if (STACK_GROWS_DOWN) {
2159         target_mprotect(error, guard, PROT_NONE);
2160         info->stack_limit = error + guard;
2161         return info->stack_limit + size - sizeof(void *);
2162     } else {
2163         info->stack_limit = error + size;
2164         return error;
2165     }
2166 }
2167 
2168 /* Map and zero the bss.  We need to explicitly zero any fractional pages
2169    after the data section (i.e. bss).  */
2170 static void zero_bss(abi_ulong elf_bss, abi_ulong last_bss, int prot)
2171 {
2172     uintptr_t host_start, host_map_start, host_end;
2173 
2174     last_bss = TARGET_PAGE_ALIGN(last_bss);
2175 
2176     /* ??? There is confusion between qemu_real_host_page_size and
2177        qemu_host_page_size here and elsewhere in target_mmap, which
2178        may lead to the end of the data section mapping from the file
2179        not being mapped.  At least there was an explicit test and
2180        comment for that here, suggesting that "the file size must
2181        be known".  The comment probably pre-dates the introduction
2182        of the fstat system call in target_mmap which does in fact
2183        find out the size.  What isn't clear is if the workaround
2184        here is still actually needed.  For now, continue with it,
2185        but merge it with the "normal" mmap that would allocate the bss.  */
2186 
2187     host_start = (uintptr_t) g2h_untagged(elf_bss);
2188     host_end = (uintptr_t) g2h_untagged(last_bss);
2189     host_map_start = REAL_HOST_PAGE_ALIGN(host_start);
2190 
2191     if (host_map_start < host_end) {
2192         void *p = mmap((void *)host_map_start, host_end - host_map_start,
2193                        prot, MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
2194         if (p == MAP_FAILED) {
2195             perror("cannot mmap brk");
2196             exit(-1);
2197         }
2198     }
2199 
2200     /* Ensure that the bss page(s) are valid */
2201     if ((page_get_flags(last_bss-1) & prot) != prot) {
2202         page_set_flags(elf_bss & TARGET_PAGE_MASK, last_bss, prot | PAGE_VALID);
2203     }
2204 
2205     if (host_start < host_map_start) {
2206         memset((void *)host_start, 0, host_map_start - host_start);
2207     }
2208 }
2209 
2210 #ifdef TARGET_ARM
2211 static int elf_is_fdpic(struct elfhdr *exec)
2212 {
2213     return exec->e_ident[EI_OSABI] == ELFOSABI_ARM_FDPIC;
2214 }
2215 #else
2216 /* Default implementation, always false.  */
2217 static int elf_is_fdpic(struct elfhdr *exec)
2218 {
2219     return 0;
2220 }
2221 #endif
2222 
2223 static abi_ulong loader_build_fdpic_loadmap(struct image_info *info, abi_ulong sp)
2224 {
2225     uint16_t n;
2226     struct elf32_fdpic_loadseg *loadsegs = info->loadsegs;
2227 
2228     /* elf32_fdpic_loadseg */
2229     n = info->nsegs;
2230     while (n--) {
2231         sp -= 12;
2232         put_user_u32(loadsegs[n].addr, sp+0);
2233         put_user_u32(loadsegs[n].p_vaddr, sp+4);
2234         put_user_u32(loadsegs[n].p_memsz, sp+8);
2235     }
2236 
2237     /* elf32_fdpic_loadmap */
2238     sp -= 4;
2239     put_user_u16(0, sp+0); /* version */
2240     put_user_u16(info->nsegs, sp+2); /* nsegs */
2241 
2242     info->personality = PER_LINUX_FDPIC;
2243     info->loadmap_addr = sp;
2244 
2245     return sp;
2246 }
2247 
2248 static abi_ulong create_elf_tables(abi_ulong p, int argc, int envc,
2249                                    struct elfhdr *exec,
2250                                    struct image_info *info,
2251                                    struct image_info *interp_info)
2252 {
2253     abi_ulong sp;
2254     abi_ulong u_argc, u_argv, u_envp, u_auxv;
2255     int size;
2256     int i;
2257     abi_ulong u_rand_bytes;
2258     uint8_t k_rand_bytes[16];
2259     abi_ulong u_platform, u_base_platform;
2260     const char *k_platform, *k_base_platform;
2261     const int n = sizeof(elf_addr_t);
2262 
2263     sp = p;
2264 
2265     /* Needs to be before we load the env/argc/... */
2266     if (elf_is_fdpic(exec)) {
2267         /* Need 4 byte alignment for these structs */
2268         sp &= ~3;
2269         sp = loader_build_fdpic_loadmap(info, sp);
2270         info->other_info = interp_info;
2271         if (interp_info) {
2272             interp_info->other_info = info;
2273             sp = loader_build_fdpic_loadmap(interp_info, sp);
2274             info->interpreter_loadmap_addr = interp_info->loadmap_addr;
2275             info->interpreter_pt_dynamic_addr = interp_info->pt_dynamic_addr;
2276         } else {
2277             info->interpreter_loadmap_addr = 0;
2278             info->interpreter_pt_dynamic_addr = 0;
2279         }
2280     }
2281 
2282     u_base_platform = 0;
2283     k_base_platform = ELF_BASE_PLATFORM;
2284     if (k_base_platform) {
2285         size_t len = strlen(k_base_platform) + 1;
2286         if (STACK_GROWS_DOWN) {
2287             sp -= (len + n - 1) & ~(n - 1);
2288             u_base_platform = sp;
2289             /* FIXME - check return value of memcpy_to_target() for failure */
2290             memcpy_to_target(sp, k_base_platform, len);
2291         } else {
2292             memcpy_to_target(sp, k_base_platform, len);
2293             u_base_platform = sp;
2294             sp += len + 1;
2295         }
2296     }
2297 
2298     u_platform = 0;
2299     k_platform = ELF_PLATFORM;
2300     if (k_platform) {
2301         size_t len = strlen(k_platform) + 1;
2302         if (STACK_GROWS_DOWN) {
2303             sp -= (len + n - 1) & ~(n - 1);
2304             u_platform = sp;
2305             /* FIXME - check return value of memcpy_to_target() for failure */
2306             memcpy_to_target(sp, k_platform, len);
2307         } else {
2308             memcpy_to_target(sp, k_platform, len);
2309             u_platform = sp;
2310             sp += len + 1;
2311         }
2312     }
2313 
2314     /* Provide 16 byte alignment for the PRNG, and basic alignment for
2315      * the argv and envp pointers.
2316      */
2317     if (STACK_GROWS_DOWN) {
2318         sp = QEMU_ALIGN_DOWN(sp, 16);
2319     } else {
2320         sp = QEMU_ALIGN_UP(sp, 16);
2321     }
2322 
2323     /*
2324      * Generate 16 random bytes for userspace PRNG seeding.
2325      */
2326     qemu_guest_getrandom_nofail(k_rand_bytes, sizeof(k_rand_bytes));
2327     if (STACK_GROWS_DOWN) {
2328         sp -= 16;
2329         u_rand_bytes = sp;
2330         /* FIXME - check return value of memcpy_to_target() for failure */
2331         memcpy_to_target(sp, k_rand_bytes, 16);
2332     } else {
2333         memcpy_to_target(sp, k_rand_bytes, 16);
2334         u_rand_bytes = sp;
2335         sp += 16;
2336     }
2337 
2338     size = (DLINFO_ITEMS + 1) * 2;
2339     if (k_base_platform)
2340         size += 2;
2341     if (k_platform)
2342         size += 2;
2343 #ifdef DLINFO_ARCH_ITEMS
2344     size += DLINFO_ARCH_ITEMS * 2;
2345 #endif
2346 #ifdef ELF_HWCAP2
2347     size += 2;
2348 #endif
2349     info->auxv_len = size * n;
2350 
2351     size += envc + argc + 2;
2352     size += 1;  /* argc itself */
2353     size *= n;
2354 
2355     /* Allocate space and finalize stack alignment for entry now.  */
2356     if (STACK_GROWS_DOWN) {
2357         u_argc = QEMU_ALIGN_DOWN(sp - size, STACK_ALIGNMENT);
2358         sp = u_argc;
2359     } else {
2360         u_argc = sp;
2361         sp = QEMU_ALIGN_UP(sp + size, STACK_ALIGNMENT);
2362     }
2363 
2364     u_argv = u_argc + n;
2365     u_envp = u_argv + (argc + 1) * n;
2366     u_auxv = u_envp + (envc + 1) * n;
2367     info->saved_auxv = u_auxv;
2368     info->argc = argc;
2369     info->envc = envc;
2370     info->argv = u_argv;
2371     info->envp = u_envp;
2372 
2373     /* This is correct because Linux defines
2374      * elf_addr_t as Elf32_Off / Elf64_Off
2375      */
2376 #define NEW_AUX_ENT(id, val) do {               \
2377         put_user_ual(id, u_auxv);  u_auxv += n; \
2378         put_user_ual(val, u_auxv); u_auxv += n; \
2379     } while(0)
2380 
2381 #ifdef ARCH_DLINFO
2382     /*
2383      * ARCH_DLINFO must come first so platform specific code can enforce
2384      * special alignment requirements on the AUXV if necessary (eg. PPC).
2385      */
2386     ARCH_DLINFO;
2387 #endif
2388     /* There must be exactly DLINFO_ITEMS entries here, or the assert
2389      * on info->auxv_len will trigger.
2390      */
2391     NEW_AUX_ENT(AT_PHDR, (abi_ulong)(info->load_addr + exec->e_phoff));
2392     NEW_AUX_ENT(AT_PHENT, (abi_ulong)(sizeof (struct elf_phdr)));
2393     NEW_AUX_ENT(AT_PHNUM, (abi_ulong)(exec->e_phnum));
2394     if ((info->alignment & ~qemu_host_page_mask) != 0) {
2395         /* Target doesn't support host page size alignment */
2396         NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(TARGET_PAGE_SIZE));
2397     } else {
2398         NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(MAX(TARGET_PAGE_SIZE,
2399                                                qemu_host_page_size)));
2400     }
2401     NEW_AUX_ENT(AT_BASE, (abi_ulong)(interp_info ? interp_info->load_addr : 0));
2402     NEW_AUX_ENT(AT_FLAGS, (abi_ulong)0);
2403     NEW_AUX_ENT(AT_ENTRY, info->entry);
2404     NEW_AUX_ENT(AT_UID, (abi_ulong) getuid());
2405     NEW_AUX_ENT(AT_EUID, (abi_ulong) geteuid());
2406     NEW_AUX_ENT(AT_GID, (abi_ulong) getgid());
2407     NEW_AUX_ENT(AT_EGID, (abi_ulong) getegid());
2408     NEW_AUX_ENT(AT_HWCAP, (abi_ulong) ELF_HWCAP);
2409     NEW_AUX_ENT(AT_CLKTCK, (abi_ulong) sysconf(_SC_CLK_TCK));
2410     NEW_AUX_ENT(AT_RANDOM, (abi_ulong) u_rand_bytes);
2411     NEW_AUX_ENT(AT_SECURE, (abi_ulong) qemu_getauxval(AT_SECURE));
2412     NEW_AUX_ENT(AT_EXECFN, info->file_string);
2413 
2414 #ifdef ELF_HWCAP2
2415     NEW_AUX_ENT(AT_HWCAP2, (abi_ulong) ELF_HWCAP2);
2416 #endif
2417 
2418     if (u_base_platform) {
2419         NEW_AUX_ENT(AT_BASE_PLATFORM, u_base_platform);
2420     }
2421     if (u_platform) {
2422         NEW_AUX_ENT(AT_PLATFORM, u_platform);
2423     }
2424     NEW_AUX_ENT (AT_NULL, 0);
2425 #undef NEW_AUX_ENT
2426 
2427     /* Check that our initial calculation of the auxv length matches how much
2428      * we actually put into it.
2429      */
2430     assert(info->auxv_len == u_auxv - info->saved_auxv);
2431 
2432     put_user_ual(argc, u_argc);
2433 
2434     p = info->arg_strings;
2435     for (i = 0; i < argc; ++i) {
2436         put_user_ual(p, u_argv);
2437         u_argv += n;
2438         p += target_strlen(p) + 1;
2439     }
2440     put_user_ual(0, u_argv);
2441 
2442     p = info->env_strings;
2443     for (i = 0; i < envc; ++i) {
2444         put_user_ual(p, u_envp);
2445         u_envp += n;
2446         p += target_strlen(p) + 1;
2447     }
2448     put_user_ual(0, u_envp);
2449 
2450     return sp;
2451 }
2452 
2453 #if defined(HI_COMMPAGE)
2454 #define LO_COMMPAGE -1
2455 #elif defined(LO_COMMPAGE)
2456 #define HI_COMMPAGE 0
2457 #else
2458 #define HI_COMMPAGE 0
2459 #define LO_COMMPAGE -1
2460 #ifndef INIT_GUEST_COMMPAGE
2461 #define init_guest_commpage() true
2462 #endif
2463 #endif
2464 
2465 static void pgb_fail_in_use(const char *image_name)
2466 {
2467     error_report("%s: requires virtual address space that is in use "
2468                  "(omit the -B option or choose a different value)",
2469                  image_name);
2470     exit(EXIT_FAILURE);
2471 }
2472 
2473 static void pgb_have_guest_base(const char *image_name, abi_ulong guest_loaddr,
2474                                 abi_ulong guest_hiaddr, long align)
2475 {
2476     const int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE;
2477     void *addr, *test;
2478 
2479     if (!QEMU_IS_ALIGNED(guest_base, align)) {
2480         fprintf(stderr, "Requested guest base %p does not satisfy "
2481                 "host minimum alignment (0x%lx)\n",
2482                 (void *)guest_base, align);
2483         exit(EXIT_FAILURE);
2484     }
2485 
2486     /* Sanity check the guest binary. */
2487     if (reserved_va) {
2488         if (guest_hiaddr > reserved_va) {
2489             error_report("%s: requires more than reserved virtual "
2490                          "address space (0x%" PRIx64 " > 0x%lx)",
2491                          image_name, (uint64_t)guest_hiaddr, reserved_va);
2492             exit(EXIT_FAILURE);
2493         }
2494     } else {
2495 #if HOST_LONG_BITS < TARGET_ABI_BITS
2496         if ((guest_hiaddr - guest_base) > ~(uintptr_t)0) {
2497             error_report("%s: requires more virtual address space "
2498                          "than the host can provide (0x%" PRIx64 ")",
2499                          image_name, (uint64_t)guest_hiaddr - guest_base);
2500             exit(EXIT_FAILURE);
2501         }
2502 #endif
2503     }
2504 
2505     /*
2506      * Expand the allocation to the entire reserved_va.
2507      * Exclude the mmap_min_addr hole.
2508      */
2509     if (reserved_va) {
2510         guest_loaddr = (guest_base >= mmap_min_addr ? 0
2511                         : mmap_min_addr - guest_base);
2512         guest_hiaddr = reserved_va;
2513     }
2514 
2515     /* Reserve the address space for the binary, or reserved_va. */
2516     test = g2h_untagged(guest_loaddr);
2517     addr = mmap(test, guest_hiaddr - guest_loaddr, PROT_NONE, flags, -1, 0);
2518     if (test != addr) {
2519         pgb_fail_in_use(image_name);
2520     }
2521     qemu_log_mask(CPU_LOG_PAGE,
2522                   "%s: base @ %p for " TARGET_ABI_FMT_ld " bytes\n",
2523                   __func__, addr, guest_hiaddr - guest_loaddr);
2524 }
2525 
2526 /**
2527  * pgd_find_hole_fallback: potential mmap address
2528  * @guest_size: size of available space
2529  * @brk: location of break
2530  * @align: memory alignment
2531  *
2532  * This is a fallback method for finding a hole in the host address
2533  * space if we don't have the benefit of being able to access
2534  * /proc/self/map. It can potentially take a very long time as we can
2535  * only dumbly iterate up the host address space seeing if the
2536  * allocation would work.
2537  */
2538 static uintptr_t pgd_find_hole_fallback(uintptr_t guest_size, uintptr_t brk,
2539                                         long align, uintptr_t offset)
2540 {
2541     uintptr_t base;
2542 
2543     /* Start (aligned) at the bottom and work our way up */
2544     base = ROUND_UP(mmap_min_addr, align);
2545 
2546     while (true) {
2547         uintptr_t align_start, end;
2548         align_start = ROUND_UP(base, align);
2549         end = align_start + guest_size + offset;
2550 
2551         /* if brk is anywhere in the range give ourselves some room to grow. */
2552         if (align_start <= brk && brk < end) {
2553             base = brk + (16 * MiB);
2554             continue;
2555         } else if (align_start + guest_size < align_start) {
2556             /* we have run out of space */
2557             return -1;
2558         } else {
2559             int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE |
2560                 MAP_FIXED_NOREPLACE;
2561             void * mmap_start = mmap((void *) align_start, guest_size,
2562                                      PROT_NONE, flags, -1, 0);
2563             if (mmap_start != MAP_FAILED) {
2564                 munmap(mmap_start, guest_size);
2565                 if (mmap_start == (void *) align_start) {
2566                     qemu_log_mask(CPU_LOG_PAGE,
2567                                   "%s: base @ %p for %" PRIdPTR" bytes\n",
2568                                   __func__, mmap_start + offset, guest_size);
2569                     return (uintptr_t) mmap_start + offset;
2570                 }
2571             }
2572             base += qemu_host_page_size;
2573         }
2574     }
2575 }
2576 
2577 /* Return value for guest_base, or -1 if no hole found. */
2578 static uintptr_t pgb_find_hole(uintptr_t guest_loaddr, uintptr_t guest_size,
2579                                long align, uintptr_t offset)
2580 {
2581     GSList *maps, *iter;
2582     uintptr_t this_start, this_end, next_start, brk;
2583     intptr_t ret = -1;
2584 
2585     assert(QEMU_IS_ALIGNED(guest_loaddr, align));
2586 
2587     maps = read_self_maps();
2588 
2589     /* Read brk after we've read the maps, which will malloc. */
2590     brk = (uintptr_t)sbrk(0);
2591 
2592     if (!maps) {
2593         return pgd_find_hole_fallback(guest_size, brk, align, offset);
2594     }
2595 
2596     /* The first hole is before the first map entry. */
2597     this_start = mmap_min_addr;
2598 
2599     for (iter = maps; iter;
2600          this_start = next_start, iter = g_slist_next(iter)) {
2601         uintptr_t align_start, hole_size;
2602 
2603         this_end = ((MapInfo *)iter->data)->start;
2604         next_start = ((MapInfo *)iter->data)->end;
2605         align_start = ROUND_UP(this_start + offset, align);
2606 
2607         /* Skip holes that are too small. */
2608         if (align_start >= this_end) {
2609             continue;
2610         }
2611         hole_size = this_end - align_start;
2612         if (hole_size < guest_size) {
2613             continue;
2614         }
2615 
2616         /* If this hole contains brk, give ourselves some room to grow. */
2617         if (this_start <= brk && brk < this_end) {
2618             hole_size -= guest_size;
2619             if (sizeof(uintptr_t) == 8 && hole_size >= 1 * GiB) {
2620                 align_start += 1 * GiB;
2621             } else if (hole_size >= 16 * MiB) {
2622                 align_start += 16 * MiB;
2623             } else {
2624                 align_start = (this_end - guest_size) & -align;
2625                 if (align_start < this_start) {
2626                     continue;
2627                 }
2628             }
2629         }
2630 
2631         /* Record the lowest successful match. */
2632         if (ret < 0) {
2633             ret = align_start;
2634         }
2635         /* If this hole contains the identity map, select it. */
2636         if (align_start <= guest_loaddr &&
2637             guest_loaddr + guest_size <= this_end) {
2638             ret = 0;
2639         }
2640         /* If this hole ends above the identity map, stop looking. */
2641         if (this_end >= guest_loaddr) {
2642             break;
2643         }
2644     }
2645     free_self_maps(maps);
2646 
2647     if (ret != -1) {
2648         qemu_log_mask(CPU_LOG_PAGE, "%s: base @ %" PRIxPTR
2649                       " for %" PRIuPTR " bytes\n",
2650                       __func__, ret, guest_size);
2651     }
2652 
2653     return ret;
2654 }
2655 
2656 static void pgb_static(const char *image_name, abi_ulong orig_loaddr,
2657                        abi_ulong orig_hiaddr, long align)
2658 {
2659     uintptr_t loaddr = orig_loaddr;
2660     uintptr_t hiaddr = orig_hiaddr;
2661     uintptr_t offset = 0;
2662     uintptr_t addr;
2663 
2664     if (hiaddr != orig_hiaddr) {
2665         error_report("%s: requires virtual address space that the "
2666                      "host cannot provide (0x%" PRIx64 ")",
2667                      image_name, (uint64_t)orig_hiaddr);
2668         exit(EXIT_FAILURE);
2669     }
2670 
2671     loaddr &= -align;
2672     if (HI_COMMPAGE) {
2673         /*
2674          * Extend the allocation to include the commpage.
2675          * For a 64-bit host, this is just 4GiB; for a 32-bit host we
2676          * need to ensure there is space bellow the guest_base so we
2677          * can map the commpage in the place needed when the address
2678          * arithmetic wraps around.
2679          */
2680         if (sizeof(uintptr_t) == 8 || loaddr >= 0x80000000u) {
2681             hiaddr = (uintptr_t) 4 << 30;
2682         } else {
2683             offset = -(HI_COMMPAGE & -align);
2684         }
2685     } else if (LO_COMMPAGE != -1) {
2686         loaddr = MIN(loaddr, LO_COMMPAGE & -align);
2687     }
2688 
2689     addr = pgb_find_hole(loaddr, hiaddr - loaddr, align, offset);
2690     if (addr == -1) {
2691         /*
2692          * If HI_COMMPAGE, there *might* be a non-consecutive allocation
2693          * that can satisfy both.  But as the normal arm32 link base address
2694          * is ~32k, and we extend down to include the commpage, making the
2695          * overhead only ~96k, this is unlikely.
2696          */
2697         error_report("%s: Unable to allocate %#zx bytes of "
2698                      "virtual address space", image_name,
2699                      (size_t)(hiaddr - loaddr));
2700         exit(EXIT_FAILURE);
2701     }
2702 
2703     guest_base = addr;
2704 
2705     qemu_log_mask(CPU_LOG_PAGE, "%s: base @ %"PRIxPTR" for %" PRIuPTR" bytes\n",
2706                   __func__, addr, hiaddr - loaddr);
2707 }
2708 
2709 static void pgb_dynamic(const char *image_name, long align)
2710 {
2711     /*
2712      * The executable is dynamic and does not require a fixed address.
2713      * All we need is a commpage that satisfies align.
2714      * If we do not need a commpage, leave guest_base == 0.
2715      */
2716     if (HI_COMMPAGE) {
2717         uintptr_t addr, commpage;
2718 
2719         /* 64-bit hosts should have used reserved_va. */
2720         assert(sizeof(uintptr_t) == 4);
2721 
2722         /*
2723          * By putting the commpage at the first hole, that puts guest_base
2724          * just above that, and maximises the positive guest addresses.
2725          */
2726         commpage = HI_COMMPAGE & -align;
2727         addr = pgb_find_hole(commpage, -commpage, align, 0);
2728         assert(addr != -1);
2729         guest_base = addr;
2730     }
2731 }
2732 
2733 static void pgb_reserved_va(const char *image_name, abi_ulong guest_loaddr,
2734                             abi_ulong guest_hiaddr, long align)
2735 {
2736     int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE;
2737     void *addr, *test;
2738 
2739     if (guest_hiaddr > reserved_va) {
2740         error_report("%s: requires more than reserved virtual "
2741                      "address space (0x%" PRIx64 " > 0x%lx)",
2742                      image_name, (uint64_t)guest_hiaddr, reserved_va);
2743         exit(EXIT_FAILURE);
2744     }
2745 
2746     /* Widen the "image" to the entire reserved address space. */
2747     pgb_static(image_name, 0, reserved_va, align);
2748 
2749     /* osdep.h defines this as 0 if it's missing */
2750     flags |= MAP_FIXED_NOREPLACE;
2751 
2752     /* Reserve the memory on the host. */
2753     assert(guest_base != 0);
2754     test = g2h_untagged(0);
2755     addr = mmap(test, reserved_va, PROT_NONE, flags, -1, 0);
2756     if (addr == MAP_FAILED || addr != test) {
2757         error_report("Unable to reserve 0x%lx bytes of virtual address "
2758                      "space at %p (%s) for use as guest address space (check your "
2759                      "virtual memory ulimit setting, min_mmap_addr or reserve less "
2760                      "using -R option)", reserved_va, test, strerror(errno));
2761         exit(EXIT_FAILURE);
2762     }
2763 
2764     qemu_log_mask(CPU_LOG_PAGE, "%s: base @ %p for %lu bytes\n",
2765                   __func__, addr, reserved_va);
2766 }
2767 
2768 void probe_guest_base(const char *image_name, abi_ulong guest_loaddr,
2769                       abi_ulong guest_hiaddr)
2770 {
2771     /* In order to use host shmat, we must be able to honor SHMLBA.  */
2772     uintptr_t align = MAX(SHMLBA, qemu_host_page_size);
2773 
2774     if (have_guest_base) {
2775         pgb_have_guest_base(image_name, guest_loaddr, guest_hiaddr, align);
2776     } else if (reserved_va) {
2777         pgb_reserved_va(image_name, guest_loaddr, guest_hiaddr, align);
2778     } else if (guest_loaddr) {
2779         pgb_static(image_name, guest_loaddr, guest_hiaddr, align);
2780     } else {
2781         pgb_dynamic(image_name, align);
2782     }
2783 
2784     /* Reserve and initialize the commpage. */
2785     if (!init_guest_commpage()) {
2786         /*
2787          * With have_guest_base, the user has selected the address and
2788          * we are trying to work with that.  Otherwise, we have selected
2789          * free space and init_guest_commpage must succeeded.
2790          */
2791         assert(have_guest_base);
2792         pgb_fail_in_use(image_name);
2793     }
2794 
2795     assert(QEMU_IS_ALIGNED(guest_base, align));
2796     qemu_log_mask(CPU_LOG_PAGE, "Locating guest address space "
2797                   "@ 0x%" PRIx64 "\n", (uint64_t)guest_base);
2798 }
2799 
2800 enum {
2801     /* The string "GNU\0" as a magic number. */
2802     GNU0_MAGIC = const_le32('G' | 'N' << 8 | 'U' << 16),
2803     NOTE_DATA_SZ = 1 * KiB,
2804     NOTE_NAME_SZ = 4,
2805     ELF_GNU_PROPERTY_ALIGN = ELF_CLASS == ELFCLASS32 ? 4 : 8,
2806 };
2807 
2808 /*
2809  * Process a single gnu_property entry.
2810  * Return false for error.
2811  */
2812 static bool parse_elf_property(const uint32_t *data, int *off, int datasz,
2813                                struct image_info *info, bool have_prev_type,
2814                                uint32_t *prev_type, Error **errp)
2815 {
2816     uint32_t pr_type, pr_datasz, step;
2817 
2818     if (*off > datasz || !QEMU_IS_ALIGNED(*off, ELF_GNU_PROPERTY_ALIGN)) {
2819         goto error_data;
2820     }
2821     datasz -= *off;
2822     data += *off / sizeof(uint32_t);
2823 
2824     if (datasz < 2 * sizeof(uint32_t)) {
2825         goto error_data;
2826     }
2827     pr_type = data[0];
2828     pr_datasz = data[1];
2829     data += 2;
2830     datasz -= 2 * sizeof(uint32_t);
2831     step = ROUND_UP(pr_datasz, ELF_GNU_PROPERTY_ALIGN);
2832     if (step > datasz) {
2833         goto error_data;
2834     }
2835 
2836     /* Properties are supposed to be unique and sorted on pr_type. */
2837     if (have_prev_type && pr_type <= *prev_type) {
2838         if (pr_type == *prev_type) {
2839             error_setg(errp, "Duplicate property in PT_GNU_PROPERTY");
2840         } else {
2841             error_setg(errp, "Unsorted property in PT_GNU_PROPERTY");
2842         }
2843         return false;
2844     }
2845     *prev_type = pr_type;
2846 
2847     if (!arch_parse_elf_property(pr_type, pr_datasz, data, info, errp)) {
2848         return false;
2849     }
2850 
2851     *off += 2 * sizeof(uint32_t) + step;
2852     return true;
2853 
2854  error_data:
2855     error_setg(errp, "Ill-formed property in PT_GNU_PROPERTY");
2856     return false;
2857 }
2858 
2859 /* Process NT_GNU_PROPERTY_TYPE_0. */
2860 static bool parse_elf_properties(int image_fd,
2861                                  struct image_info *info,
2862                                  const struct elf_phdr *phdr,
2863                                  char bprm_buf[BPRM_BUF_SIZE],
2864                                  Error **errp)
2865 {
2866     union {
2867         struct elf_note nhdr;
2868         uint32_t data[NOTE_DATA_SZ / sizeof(uint32_t)];
2869     } note;
2870 
2871     int n, off, datasz;
2872     bool have_prev_type;
2873     uint32_t prev_type;
2874 
2875     /* Unless the arch requires properties, ignore them. */
2876     if (!ARCH_USE_GNU_PROPERTY) {
2877         return true;
2878     }
2879 
2880     /* If the properties are crazy large, that's too bad. */
2881     n = phdr->p_filesz;
2882     if (n > sizeof(note)) {
2883         error_setg(errp, "PT_GNU_PROPERTY too large");
2884         return false;
2885     }
2886     if (n < sizeof(note.nhdr)) {
2887         error_setg(errp, "PT_GNU_PROPERTY too small");
2888         return false;
2889     }
2890 
2891     if (phdr->p_offset + n <= BPRM_BUF_SIZE) {
2892         memcpy(&note, bprm_buf + phdr->p_offset, n);
2893     } else {
2894         ssize_t len = pread(image_fd, &note, n, phdr->p_offset);
2895         if (len != n) {
2896             error_setg_errno(errp, errno, "Error reading file header");
2897             return false;
2898         }
2899     }
2900 
2901     /*
2902      * The contents of a valid PT_GNU_PROPERTY is a sequence
2903      * of uint32_t -- swap them all now.
2904      */
2905 #ifdef BSWAP_NEEDED
2906     for (int i = 0; i < n / 4; i++) {
2907         bswap32s(note.data + i);
2908     }
2909 #endif
2910 
2911     /*
2912      * Note that nhdr is 3 words, and that the "name" described by namesz
2913      * immediately follows nhdr and is thus at the 4th word.  Further, all
2914      * of the inputs to the kernel's round_up are multiples of 4.
2915      */
2916     if (note.nhdr.n_type != NT_GNU_PROPERTY_TYPE_0 ||
2917         note.nhdr.n_namesz != NOTE_NAME_SZ ||
2918         note.data[3] != GNU0_MAGIC) {
2919         error_setg(errp, "Invalid note in PT_GNU_PROPERTY");
2920         return false;
2921     }
2922     off = sizeof(note.nhdr) + NOTE_NAME_SZ;
2923 
2924     datasz = note.nhdr.n_descsz + off;
2925     if (datasz > n) {
2926         error_setg(errp, "Invalid note size in PT_GNU_PROPERTY");
2927         return false;
2928     }
2929 
2930     have_prev_type = false;
2931     prev_type = 0;
2932     while (1) {
2933         if (off == datasz) {
2934             return true;  /* end, exit ok */
2935         }
2936         if (!parse_elf_property(note.data, &off, datasz, info,
2937                                 have_prev_type, &prev_type, errp)) {
2938             return false;
2939         }
2940         have_prev_type = true;
2941     }
2942 }
2943 
2944 /* Load an ELF image into the address space.
2945 
2946    IMAGE_NAME is the filename of the image, to use in error messages.
2947    IMAGE_FD is the open file descriptor for the image.
2948 
2949    BPRM_BUF is a copy of the beginning of the file; this of course
2950    contains the elf file header at offset 0.  It is assumed that this
2951    buffer is sufficiently aligned to present no problems to the host
2952    in accessing data at aligned offsets within the buffer.
2953 
2954    On return: INFO values will be filled in, as necessary or available.  */
2955 
2956 static void load_elf_image(const char *image_name, int image_fd,
2957                            struct image_info *info, char **pinterp_name,
2958                            char bprm_buf[BPRM_BUF_SIZE])
2959 {
2960     struct elfhdr *ehdr = (struct elfhdr *)bprm_buf;
2961     struct elf_phdr *phdr;
2962     abi_ulong load_addr, load_bias, loaddr, hiaddr, error;
2963     int i, retval, prot_exec;
2964     Error *err = NULL;
2965 
2966     /* First of all, some simple consistency checks */
2967     if (!elf_check_ident(ehdr)) {
2968         error_setg(&err, "Invalid ELF image for this architecture");
2969         goto exit_errmsg;
2970     }
2971     bswap_ehdr(ehdr);
2972     if (!elf_check_ehdr(ehdr)) {
2973         error_setg(&err, "Invalid ELF image for this architecture");
2974         goto exit_errmsg;
2975     }
2976 
2977     i = ehdr->e_phnum * sizeof(struct elf_phdr);
2978     if (ehdr->e_phoff + i <= BPRM_BUF_SIZE) {
2979         phdr = (struct elf_phdr *)(bprm_buf + ehdr->e_phoff);
2980     } else {
2981         phdr = (struct elf_phdr *) alloca(i);
2982         retval = pread(image_fd, phdr, i, ehdr->e_phoff);
2983         if (retval != i) {
2984             goto exit_read;
2985         }
2986     }
2987     bswap_phdr(phdr, ehdr->e_phnum);
2988 
2989     info->nsegs = 0;
2990     info->pt_dynamic_addr = 0;
2991 
2992     mmap_lock();
2993 
2994     /*
2995      * Find the maximum size of the image and allocate an appropriate
2996      * amount of memory to handle that.  Locate the interpreter, if any.
2997      */
2998     loaddr = -1, hiaddr = 0;
2999     info->alignment = 0;
3000     info->exec_stack = EXSTACK_DEFAULT;
3001     for (i = 0; i < ehdr->e_phnum; ++i) {
3002         struct elf_phdr *eppnt = phdr + i;
3003         if (eppnt->p_type == PT_LOAD) {
3004             abi_ulong a = eppnt->p_vaddr - eppnt->p_offset;
3005             if (a < loaddr) {
3006                 loaddr = a;
3007             }
3008             a = eppnt->p_vaddr + eppnt->p_memsz;
3009             if (a > hiaddr) {
3010                 hiaddr = a;
3011             }
3012             ++info->nsegs;
3013             info->alignment |= eppnt->p_align;
3014         } else if (eppnt->p_type == PT_INTERP && pinterp_name) {
3015             g_autofree char *interp_name = NULL;
3016 
3017             if (*pinterp_name) {
3018                 error_setg(&err, "Multiple PT_INTERP entries");
3019                 goto exit_errmsg;
3020             }
3021 
3022             interp_name = g_malloc(eppnt->p_filesz);
3023 
3024             if (eppnt->p_offset + eppnt->p_filesz <= BPRM_BUF_SIZE) {
3025                 memcpy(interp_name, bprm_buf + eppnt->p_offset,
3026                        eppnt->p_filesz);
3027             } else {
3028                 retval = pread(image_fd, interp_name, eppnt->p_filesz,
3029                                eppnt->p_offset);
3030                 if (retval != eppnt->p_filesz) {
3031                     goto exit_read;
3032                 }
3033             }
3034             if (interp_name[eppnt->p_filesz - 1] != 0) {
3035                 error_setg(&err, "Invalid PT_INTERP entry");
3036                 goto exit_errmsg;
3037             }
3038             *pinterp_name = g_steal_pointer(&interp_name);
3039         } else if (eppnt->p_type == PT_GNU_PROPERTY) {
3040             if (!parse_elf_properties(image_fd, info, eppnt, bprm_buf, &err)) {
3041                 goto exit_errmsg;
3042             }
3043         } else if (eppnt->p_type == PT_GNU_STACK) {
3044             info->exec_stack = eppnt->p_flags & PF_X;
3045         }
3046     }
3047 
3048     if (pinterp_name != NULL) {
3049         /*
3050          * This is the main executable.
3051          *
3052          * Reserve extra space for brk.
3053          * We hold on to this space while placing the interpreter
3054          * and the stack, lest they be placed immediately after
3055          * the data segment and block allocation from the brk.
3056          *
3057          * 16MB is chosen as "large enough" without being so large as
3058          * to allow the result to not fit with a 32-bit guest on a
3059          * 32-bit host. However some 64 bit guests (e.g. s390x)
3060          * attempt to place their heap further ahead and currently
3061          * nothing stops them smashing into QEMUs address space.
3062          */
3063 #if TARGET_LONG_BITS == 64
3064         info->reserve_brk = 32 * MiB;
3065 #else
3066         info->reserve_brk = 16 * MiB;
3067 #endif
3068         hiaddr += info->reserve_brk;
3069 
3070         if (ehdr->e_type == ET_EXEC) {
3071             /*
3072              * Make sure that the low address does not conflict with
3073              * MMAP_MIN_ADDR or the QEMU application itself.
3074              */
3075             probe_guest_base(image_name, loaddr, hiaddr);
3076         } else {
3077             /*
3078              * The binary is dynamic, but we still need to
3079              * select guest_base.  In this case we pass a size.
3080              */
3081             probe_guest_base(image_name, 0, hiaddr - loaddr);
3082         }
3083     }
3084 
3085     /*
3086      * Reserve address space for all of this.
3087      *
3088      * In the case of ET_EXEC, we supply MAP_FIXED so that we get
3089      * exactly the address range that is required.
3090      *
3091      * Otherwise this is ET_DYN, and we are searching for a location
3092      * that can hold the memory space required.  If the image is
3093      * pre-linked, LOADDR will be non-zero, and the kernel should
3094      * honor that address if it happens to be free.
3095      *
3096      * In both cases, we will overwrite pages in this range with mappings
3097      * from the executable.
3098      */
3099     load_addr = target_mmap(loaddr, hiaddr - loaddr, PROT_NONE,
3100                             MAP_PRIVATE | MAP_ANON | MAP_NORESERVE |
3101                             (ehdr->e_type == ET_EXEC ? MAP_FIXED : 0),
3102                             -1, 0);
3103     if (load_addr == -1) {
3104         goto exit_mmap;
3105     }
3106     load_bias = load_addr - loaddr;
3107 
3108     if (elf_is_fdpic(ehdr)) {
3109         struct elf32_fdpic_loadseg *loadsegs = info->loadsegs =
3110             g_malloc(sizeof(*loadsegs) * info->nsegs);
3111 
3112         for (i = 0; i < ehdr->e_phnum; ++i) {
3113             switch (phdr[i].p_type) {
3114             case PT_DYNAMIC:
3115                 info->pt_dynamic_addr = phdr[i].p_vaddr + load_bias;
3116                 break;
3117             case PT_LOAD:
3118                 loadsegs->addr = phdr[i].p_vaddr + load_bias;
3119                 loadsegs->p_vaddr = phdr[i].p_vaddr;
3120                 loadsegs->p_memsz = phdr[i].p_memsz;
3121                 ++loadsegs;
3122                 break;
3123             }
3124         }
3125     }
3126 
3127     info->load_bias = load_bias;
3128     info->code_offset = load_bias;
3129     info->data_offset = load_bias;
3130     info->load_addr = load_addr;
3131     info->entry = ehdr->e_entry + load_bias;
3132     info->start_code = -1;
3133     info->end_code = 0;
3134     info->start_data = -1;
3135     info->end_data = 0;
3136     info->brk = 0;
3137     info->elf_flags = ehdr->e_flags;
3138 
3139     prot_exec = PROT_EXEC;
3140 #ifdef TARGET_AARCH64
3141     /*
3142      * If the BTI feature is present, this indicates that the executable
3143      * pages of the startup binary should be mapped with PROT_BTI, so that
3144      * branch targets are enforced.
3145      *
3146      * The startup binary is either the interpreter or the static executable.
3147      * The interpreter is responsible for all pages of a dynamic executable.
3148      *
3149      * Elf notes are backward compatible to older cpus.
3150      * Do not enable BTI unless it is supported.
3151      */
3152     if ((info->note_flags & GNU_PROPERTY_AARCH64_FEATURE_1_BTI)
3153         && (pinterp_name == NULL || *pinterp_name == 0)
3154         && cpu_isar_feature(aa64_bti, ARM_CPU(thread_cpu))) {
3155         prot_exec |= TARGET_PROT_BTI;
3156     }
3157 #endif
3158 
3159     for (i = 0; i < ehdr->e_phnum; i++) {
3160         struct elf_phdr *eppnt = phdr + i;
3161         if (eppnt->p_type == PT_LOAD) {
3162             abi_ulong vaddr, vaddr_po, vaddr_ps, vaddr_ef, vaddr_em, vaddr_len;
3163             int elf_prot = 0;
3164 
3165             if (eppnt->p_flags & PF_R) {
3166                 elf_prot |= PROT_READ;
3167             }
3168             if (eppnt->p_flags & PF_W) {
3169                 elf_prot |= PROT_WRITE;
3170             }
3171             if (eppnt->p_flags & PF_X) {
3172                 elf_prot |= prot_exec;
3173             }
3174 
3175             vaddr = load_bias + eppnt->p_vaddr;
3176             vaddr_po = TARGET_ELF_PAGEOFFSET(vaddr);
3177             vaddr_ps = TARGET_ELF_PAGESTART(vaddr);
3178 
3179             vaddr_ef = vaddr + eppnt->p_filesz;
3180             vaddr_em = vaddr + eppnt->p_memsz;
3181 
3182             /*
3183              * Some segments may be completely empty, with a non-zero p_memsz
3184              * but no backing file segment.
3185              */
3186             if (eppnt->p_filesz != 0) {
3187                 vaddr_len = TARGET_ELF_PAGELENGTH(eppnt->p_filesz + vaddr_po);
3188                 error = target_mmap(vaddr_ps, vaddr_len, elf_prot,
3189                                     MAP_PRIVATE | MAP_FIXED,
3190                                     image_fd, eppnt->p_offset - vaddr_po);
3191 
3192                 if (error == -1) {
3193                     goto exit_mmap;
3194                 }
3195 
3196                 /*
3197                  * If the load segment requests extra zeros (e.g. bss), map it.
3198                  */
3199                 if (eppnt->p_filesz < eppnt->p_memsz) {
3200                     zero_bss(vaddr_ef, vaddr_em, elf_prot);
3201                 }
3202             } else if (eppnt->p_memsz != 0) {
3203                 vaddr_len = TARGET_ELF_PAGELENGTH(eppnt->p_memsz + vaddr_po);
3204                 error = target_mmap(vaddr_ps, vaddr_len, elf_prot,
3205                                     MAP_PRIVATE | MAP_FIXED | MAP_ANONYMOUS,
3206                                     -1, 0);
3207 
3208                 if (error == -1) {
3209                     goto exit_mmap;
3210                 }
3211             }
3212 
3213             /* Find the full program boundaries.  */
3214             if (elf_prot & PROT_EXEC) {
3215                 if (vaddr < info->start_code) {
3216                     info->start_code = vaddr;
3217                 }
3218                 if (vaddr_ef > info->end_code) {
3219                     info->end_code = vaddr_ef;
3220                 }
3221             }
3222             if (elf_prot & PROT_WRITE) {
3223                 if (vaddr < info->start_data) {
3224                     info->start_data = vaddr;
3225                 }
3226                 if (vaddr_ef > info->end_data) {
3227                     info->end_data = vaddr_ef;
3228                 }
3229             }
3230             if (vaddr_em > info->brk) {
3231                 info->brk = vaddr_em;
3232             }
3233 #ifdef TARGET_MIPS
3234         } else if (eppnt->p_type == PT_MIPS_ABIFLAGS) {
3235             Mips_elf_abiflags_v0 abiflags;
3236             if (eppnt->p_filesz < sizeof(Mips_elf_abiflags_v0)) {
3237                 error_setg(&err, "Invalid PT_MIPS_ABIFLAGS entry");
3238                 goto exit_errmsg;
3239             }
3240             if (eppnt->p_offset + eppnt->p_filesz <= BPRM_BUF_SIZE) {
3241                 memcpy(&abiflags, bprm_buf + eppnt->p_offset,
3242                        sizeof(Mips_elf_abiflags_v0));
3243             } else {
3244                 retval = pread(image_fd, &abiflags, sizeof(Mips_elf_abiflags_v0),
3245                                eppnt->p_offset);
3246                 if (retval != sizeof(Mips_elf_abiflags_v0)) {
3247                     goto exit_read;
3248                 }
3249             }
3250             bswap_mips_abiflags(&abiflags);
3251             info->fp_abi = abiflags.fp_abi;
3252 #endif
3253         }
3254     }
3255 
3256     if (info->end_data == 0) {
3257         info->start_data = info->end_code;
3258         info->end_data = info->end_code;
3259     }
3260 
3261     if (qemu_log_enabled()) {
3262         load_symbols(ehdr, image_fd, load_bias);
3263     }
3264 
3265     debuginfo_report_elf(image_name, image_fd, load_bias);
3266 
3267     mmap_unlock();
3268 
3269     close(image_fd);
3270     return;
3271 
3272  exit_read:
3273     if (retval >= 0) {
3274         error_setg(&err, "Incomplete read of file header");
3275     } else {
3276         error_setg_errno(&err, errno, "Error reading file header");
3277     }
3278     goto exit_errmsg;
3279  exit_mmap:
3280     error_setg_errno(&err, errno, "Error mapping file");
3281     goto exit_errmsg;
3282  exit_errmsg:
3283     error_reportf_err(err, "%s: ", image_name);
3284     exit(-1);
3285 }
3286 
3287 static void load_elf_interp(const char *filename, struct image_info *info,
3288                             char bprm_buf[BPRM_BUF_SIZE])
3289 {
3290     int fd, retval;
3291     Error *err = NULL;
3292 
3293     fd = open(path(filename), O_RDONLY);
3294     if (fd < 0) {
3295         error_setg_file_open(&err, errno, filename);
3296         error_report_err(err);
3297         exit(-1);
3298     }
3299 
3300     retval = read(fd, bprm_buf, BPRM_BUF_SIZE);
3301     if (retval < 0) {
3302         error_setg_errno(&err, errno, "Error reading file header");
3303         error_reportf_err(err, "%s: ", filename);
3304         exit(-1);
3305     }
3306 
3307     if (retval < BPRM_BUF_SIZE) {
3308         memset(bprm_buf + retval, 0, BPRM_BUF_SIZE - retval);
3309     }
3310 
3311     load_elf_image(filename, fd, info, NULL, bprm_buf);
3312 }
3313 
3314 static int symfind(const void *s0, const void *s1)
3315 {
3316     target_ulong addr = *(target_ulong *)s0;
3317     struct elf_sym *sym = (struct elf_sym *)s1;
3318     int result = 0;
3319     if (addr < sym->st_value) {
3320         result = -1;
3321     } else if (addr >= sym->st_value + sym->st_size) {
3322         result = 1;
3323     }
3324     return result;
3325 }
3326 
3327 static const char *lookup_symbolxx(struct syminfo *s, target_ulong orig_addr)
3328 {
3329 #if ELF_CLASS == ELFCLASS32
3330     struct elf_sym *syms = s->disas_symtab.elf32;
3331 #else
3332     struct elf_sym *syms = s->disas_symtab.elf64;
3333 #endif
3334 
3335     // binary search
3336     struct elf_sym *sym;
3337 
3338     sym = bsearch(&orig_addr, syms, s->disas_num_syms, sizeof(*syms), symfind);
3339     if (sym != NULL) {
3340         return s->disas_strtab + sym->st_name;
3341     }
3342 
3343     return "";
3344 }
3345 
3346 /* FIXME: This should use elf_ops.h  */
3347 static int symcmp(const void *s0, const void *s1)
3348 {
3349     struct elf_sym *sym0 = (struct elf_sym *)s0;
3350     struct elf_sym *sym1 = (struct elf_sym *)s1;
3351     return (sym0->st_value < sym1->st_value)
3352         ? -1
3353         : ((sym0->st_value > sym1->st_value) ? 1 : 0);
3354 }
3355 
3356 /* Best attempt to load symbols from this ELF object. */
3357 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias)
3358 {
3359     int i, shnum, nsyms, sym_idx = 0, str_idx = 0;
3360     uint64_t segsz;
3361     struct elf_shdr *shdr;
3362     char *strings = NULL;
3363     struct syminfo *s = NULL;
3364     struct elf_sym *new_syms, *syms = NULL;
3365 
3366     shnum = hdr->e_shnum;
3367     i = shnum * sizeof(struct elf_shdr);
3368     shdr = (struct elf_shdr *)alloca(i);
3369     if (pread(fd, shdr, i, hdr->e_shoff) != i) {
3370         return;
3371     }
3372 
3373     bswap_shdr(shdr, shnum);
3374     for (i = 0; i < shnum; ++i) {
3375         if (shdr[i].sh_type == SHT_SYMTAB) {
3376             sym_idx = i;
3377             str_idx = shdr[i].sh_link;
3378             goto found;
3379         }
3380     }
3381 
3382     /* There will be no symbol table if the file was stripped.  */
3383     return;
3384 
3385  found:
3386     /* Now know where the strtab and symtab are.  Snarf them.  */
3387     s = g_try_new(struct syminfo, 1);
3388     if (!s) {
3389         goto give_up;
3390     }
3391 
3392     segsz = shdr[str_idx].sh_size;
3393     s->disas_strtab = strings = g_try_malloc(segsz);
3394     if (!strings ||
3395         pread(fd, strings, segsz, shdr[str_idx].sh_offset) != segsz) {
3396         goto give_up;
3397     }
3398 
3399     segsz = shdr[sym_idx].sh_size;
3400     syms = g_try_malloc(segsz);
3401     if (!syms || pread(fd, syms, segsz, shdr[sym_idx].sh_offset) != segsz) {
3402         goto give_up;
3403     }
3404 
3405     if (segsz / sizeof(struct elf_sym) > INT_MAX) {
3406         /* Implausibly large symbol table: give up rather than ploughing
3407          * on with the number of symbols calculation overflowing
3408          */
3409         goto give_up;
3410     }
3411     nsyms = segsz / sizeof(struct elf_sym);
3412     for (i = 0; i < nsyms; ) {
3413         bswap_sym(syms + i);
3414         /* Throw away entries which we do not need.  */
3415         if (syms[i].st_shndx == SHN_UNDEF
3416             || syms[i].st_shndx >= SHN_LORESERVE
3417             || ELF_ST_TYPE(syms[i].st_info) != STT_FUNC) {
3418             if (i < --nsyms) {
3419                 syms[i] = syms[nsyms];
3420             }
3421         } else {
3422 #if defined(TARGET_ARM) || defined (TARGET_MIPS)
3423             /* The bottom address bit marks a Thumb or MIPS16 symbol.  */
3424             syms[i].st_value &= ~(target_ulong)1;
3425 #endif
3426             syms[i].st_value += load_bias;
3427             i++;
3428         }
3429     }
3430 
3431     /* No "useful" symbol.  */
3432     if (nsyms == 0) {
3433         goto give_up;
3434     }
3435 
3436     /* Attempt to free the storage associated with the local symbols
3437        that we threw away.  Whether or not this has any effect on the
3438        memory allocation depends on the malloc implementation and how
3439        many symbols we managed to discard.  */
3440     new_syms = g_try_renew(struct elf_sym, syms, nsyms);
3441     if (new_syms == NULL) {
3442         goto give_up;
3443     }
3444     syms = new_syms;
3445 
3446     qsort(syms, nsyms, sizeof(*syms), symcmp);
3447 
3448     s->disas_num_syms = nsyms;
3449 #if ELF_CLASS == ELFCLASS32
3450     s->disas_symtab.elf32 = syms;
3451 #else
3452     s->disas_symtab.elf64 = syms;
3453 #endif
3454     s->lookup_symbol = lookup_symbolxx;
3455     s->next = syminfos;
3456     syminfos = s;
3457 
3458     return;
3459 
3460 give_up:
3461     g_free(s);
3462     g_free(strings);
3463     g_free(syms);
3464 }
3465 
3466 uint32_t get_elf_eflags(int fd)
3467 {
3468     struct elfhdr ehdr;
3469     off_t offset;
3470     int ret;
3471 
3472     /* Read ELF header */
3473     offset = lseek(fd, 0, SEEK_SET);
3474     if (offset == (off_t) -1) {
3475         return 0;
3476     }
3477     ret = read(fd, &ehdr, sizeof(ehdr));
3478     if (ret < sizeof(ehdr)) {
3479         return 0;
3480     }
3481     offset = lseek(fd, offset, SEEK_SET);
3482     if (offset == (off_t) -1) {
3483         return 0;
3484     }
3485 
3486     /* Check ELF signature */
3487     if (!elf_check_ident(&ehdr)) {
3488         return 0;
3489     }
3490 
3491     /* check header */
3492     bswap_ehdr(&ehdr);
3493     if (!elf_check_ehdr(&ehdr)) {
3494         return 0;
3495     }
3496 
3497     /* return architecture id */
3498     return ehdr.e_flags;
3499 }
3500 
3501 int load_elf_binary(struct linux_binprm *bprm, struct image_info *info)
3502 {
3503     struct image_info interp_info;
3504     struct elfhdr elf_ex;
3505     char *elf_interpreter = NULL;
3506     char *scratch;
3507 
3508     memset(&interp_info, 0, sizeof(interp_info));
3509 #ifdef TARGET_MIPS
3510     interp_info.fp_abi = MIPS_ABI_FP_UNKNOWN;
3511 #endif
3512 
3513     info->start_mmap = (abi_ulong)ELF_START_MMAP;
3514 
3515     load_elf_image(bprm->filename, bprm->fd, info,
3516                    &elf_interpreter, bprm->buf);
3517 
3518     /* ??? We need a copy of the elf header for passing to create_elf_tables.
3519        If we do nothing, we'll have overwritten this when we re-use bprm->buf
3520        when we load the interpreter.  */
3521     elf_ex = *(struct elfhdr *)bprm->buf;
3522 
3523     /* Do this so that we can load the interpreter, if need be.  We will
3524        change some of these later */
3525     bprm->p = setup_arg_pages(bprm, info);
3526 
3527     scratch = g_new0(char, TARGET_PAGE_SIZE);
3528     if (STACK_GROWS_DOWN) {
3529         bprm->p = copy_elf_strings(1, &bprm->filename, scratch,
3530                                    bprm->p, info->stack_limit);
3531         info->file_string = bprm->p;
3532         bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch,
3533                                    bprm->p, info->stack_limit);
3534         info->env_strings = bprm->p;
3535         bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch,
3536                                    bprm->p, info->stack_limit);
3537         info->arg_strings = bprm->p;
3538     } else {
3539         info->arg_strings = bprm->p;
3540         bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch,
3541                                    bprm->p, info->stack_limit);
3542         info->env_strings = bprm->p;
3543         bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch,
3544                                    bprm->p, info->stack_limit);
3545         info->file_string = bprm->p;
3546         bprm->p = copy_elf_strings(1, &bprm->filename, scratch,
3547                                    bprm->p, info->stack_limit);
3548     }
3549 
3550     g_free(scratch);
3551 
3552     if (!bprm->p) {
3553         fprintf(stderr, "%s: %s\n", bprm->filename, strerror(E2BIG));
3554         exit(-1);
3555     }
3556 
3557     if (elf_interpreter) {
3558         load_elf_interp(elf_interpreter, &interp_info, bprm->buf);
3559 
3560         /* If the program interpreter is one of these two, then assume
3561            an iBCS2 image.  Otherwise assume a native linux image.  */
3562 
3563         if (strcmp(elf_interpreter, "/usr/lib/libc.so.1") == 0
3564             || strcmp(elf_interpreter, "/usr/lib/ld.so.1") == 0) {
3565             info->personality = PER_SVR4;
3566 
3567             /* Why this, you ask???  Well SVr4 maps page 0 as read-only,
3568                and some applications "depend" upon this behavior.  Since
3569                we do not have the power to recompile these, we emulate
3570                the SVr4 behavior.  Sigh.  */
3571             target_mmap(0, qemu_host_page_size, PROT_READ | PROT_EXEC,
3572                         MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
3573         }
3574 #ifdef TARGET_MIPS
3575         info->interp_fp_abi = interp_info.fp_abi;
3576 #endif
3577     }
3578 
3579     /*
3580      * TODO: load a vdso, which would also contain the signal trampolines.
3581      * Otherwise, allocate a private page to hold them.
3582      */
3583     if (TARGET_ARCH_HAS_SIGTRAMP_PAGE) {
3584         abi_long tramp_page = target_mmap(0, TARGET_PAGE_SIZE,
3585                                           PROT_READ | PROT_WRITE,
3586                                           MAP_PRIVATE | MAP_ANON, -1, 0);
3587         if (tramp_page == -1) {
3588             return -errno;
3589         }
3590 
3591         setup_sigtramp(tramp_page);
3592         target_mprotect(tramp_page, TARGET_PAGE_SIZE, PROT_READ | PROT_EXEC);
3593     }
3594 
3595     bprm->p = create_elf_tables(bprm->p, bprm->argc, bprm->envc, &elf_ex,
3596                                 info, (elf_interpreter ? &interp_info : NULL));
3597     info->start_stack = bprm->p;
3598 
3599     /* If we have an interpreter, set that as the program's entry point.
3600        Copy the load_bias as well, to help PPC64 interpret the entry
3601        point as a function descriptor.  Do this after creating elf tables
3602        so that we copy the original program entry point into the AUXV.  */
3603     if (elf_interpreter) {
3604         info->load_bias = interp_info.load_bias;
3605         info->entry = interp_info.entry;
3606         g_free(elf_interpreter);
3607     }
3608 
3609 #ifdef USE_ELF_CORE_DUMP
3610     bprm->core_dump = &elf_core_dump;
3611 #endif
3612 
3613     /*
3614      * If we reserved extra space for brk, release it now.
3615      * The implementation of do_brk in syscalls.c expects to be able
3616      * to mmap pages in this space.
3617      */
3618     if (info->reserve_brk) {
3619         abi_ulong start_brk = HOST_PAGE_ALIGN(info->brk);
3620         abi_ulong end_brk = HOST_PAGE_ALIGN(info->brk + info->reserve_brk);
3621         target_munmap(start_brk, end_brk - start_brk);
3622     }
3623 
3624     return 0;
3625 }
3626 
3627 #ifdef USE_ELF_CORE_DUMP
3628 /*
3629  * Definitions to generate Intel SVR4-like core files.
3630  * These mostly have the same names as the SVR4 types with "target_elf_"
3631  * tacked on the front to prevent clashes with linux definitions,
3632  * and the typedef forms have been avoided.  This is mostly like
3633  * the SVR4 structure, but more Linuxy, with things that Linux does
3634  * not support and which gdb doesn't really use excluded.
3635  *
3636  * Fields we don't dump (their contents is zero) in linux-user qemu
3637  * are marked with XXX.
3638  *
3639  * Core dump code is copied from linux kernel (fs/binfmt_elf.c).
3640  *
3641  * Porting ELF coredump for target is (quite) simple process.  First you
3642  * define USE_ELF_CORE_DUMP in target ELF code (where init_thread() for
3643  * the target resides):
3644  *
3645  * #define USE_ELF_CORE_DUMP
3646  *
3647  * Next you define type of register set used for dumping.  ELF specification
3648  * says that it needs to be array of elf_greg_t that has size of ELF_NREG.
3649  *
3650  * typedef <target_regtype> target_elf_greg_t;
3651  * #define ELF_NREG <number of registers>
3652  * typedef taret_elf_greg_t target_elf_gregset_t[ELF_NREG];
3653  *
3654  * Last step is to implement target specific function that copies registers
3655  * from given cpu into just specified register set.  Prototype is:
3656  *
3657  * static void elf_core_copy_regs(taret_elf_gregset_t *regs,
3658  *                                const CPUArchState *env);
3659  *
3660  * Parameters:
3661  *     regs - copy register values into here (allocated and zeroed by caller)
3662  *     env - copy registers from here
3663  *
3664  * Example for ARM target is provided in this file.
3665  */
3666 
3667 /* An ELF note in memory */
3668 struct memelfnote {
3669     const char *name;
3670     size_t     namesz;
3671     size_t     namesz_rounded;
3672     int        type;
3673     size_t     datasz;
3674     size_t     datasz_rounded;
3675     void       *data;
3676     size_t     notesz;
3677 };
3678 
3679 struct target_elf_siginfo {
3680     abi_int    si_signo; /* signal number */
3681     abi_int    si_code;  /* extra code */
3682     abi_int    si_errno; /* errno */
3683 };
3684 
3685 struct target_elf_prstatus {
3686     struct target_elf_siginfo pr_info;      /* Info associated with signal */
3687     abi_short          pr_cursig;    /* Current signal */
3688     abi_ulong          pr_sigpend;   /* XXX */
3689     abi_ulong          pr_sighold;   /* XXX */
3690     target_pid_t       pr_pid;
3691     target_pid_t       pr_ppid;
3692     target_pid_t       pr_pgrp;
3693     target_pid_t       pr_sid;
3694     struct target_timeval pr_utime;  /* XXX User time */
3695     struct target_timeval pr_stime;  /* XXX System time */
3696     struct target_timeval pr_cutime; /* XXX Cumulative user time */
3697     struct target_timeval pr_cstime; /* XXX Cumulative system time */
3698     target_elf_gregset_t      pr_reg;       /* GP registers */
3699     abi_int            pr_fpvalid;   /* XXX */
3700 };
3701 
3702 #define ELF_PRARGSZ     (80) /* Number of chars for args */
3703 
3704 struct target_elf_prpsinfo {
3705     char         pr_state;       /* numeric process state */
3706     char         pr_sname;       /* char for pr_state */
3707     char         pr_zomb;        /* zombie */
3708     char         pr_nice;        /* nice val */
3709     abi_ulong    pr_flag;        /* flags */
3710     target_uid_t pr_uid;
3711     target_gid_t pr_gid;
3712     target_pid_t pr_pid, pr_ppid, pr_pgrp, pr_sid;
3713     /* Lots missing */
3714     char    pr_fname[16] QEMU_NONSTRING; /* filename of executable */
3715     char    pr_psargs[ELF_PRARGSZ]; /* initial part of arg list */
3716 };
3717 
3718 /* Here is the structure in which status of each thread is captured. */
3719 struct elf_thread_status {
3720     QTAILQ_ENTRY(elf_thread_status)  ets_link;
3721     struct target_elf_prstatus prstatus;   /* NT_PRSTATUS */
3722 #if 0
3723     elf_fpregset_t fpu;             /* NT_PRFPREG */
3724     struct task_struct *thread;
3725     elf_fpxregset_t xfpu;           /* ELF_CORE_XFPREG_TYPE */
3726 #endif
3727     struct memelfnote notes[1];
3728     int num_notes;
3729 };
3730 
3731 struct elf_note_info {
3732     struct memelfnote   *notes;
3733     struct target_elf_prstatus *prstatus;  /* NT_PRSTATUS */
3734     struct target_elf_prpsinfo *psinfo;    /* NT_PRPSINFO */
3735 
3736     QTAILQ_HEAD(, elf_thread_status) thread_list;
3737 #if 0
3738     /*
3739      * Current version of ELF coredump doesn't support
3740      * dumping fp regs etc.
3741      */
3742     elf_fpregset_t *fpu;
3743     elf_fpxregset_t *xfpu;
3744     int thread_status_size;
3745 #endif
3746     int notes_size;
3747     int numnote;
3748 };
3749 
3750 struct vm_area_struct {
3751     target_ulong   vma_start;  /* start vaddr of memory region */
3752     target_ulong   vma_end;    /* end vaddr of memory region */
3753     abi_ulong      vma_flags;  /* protection etc. flags for the region */
3754     QTAILQ_ENTRY(vm_area_struct) vma_link;
3755 };
3756 
3757 struct mm_struct {
3758     QTAILQ_HEAD(, vm_area_struct) mm_mmap;
3759     int mm_count;           /* number of mappings */
3760 };
3761 
3762 static struct mm_struct *vma_init(void);
3763 static void vma_delete(struct mm_struct *);
3764 static int vma_add_mapping(struct mm_struct *, target_ulong,
3765                            target_ulong, abi_ulong);
3766 static int vma_get_mapping_count(const struct mm_struct *);
3767 static struct vm_area_struct *vma_first(const struct mm_struct *);
3768 static struct vm_area_struct *vma_next(struct vm_area_struct *);
3769 static abi_ulong vma_dump_size(const struct vm_area_struct *);
3770 static int vma_walker(void *priv, target_ulong start, target_ulong end,
3771                       unsigned long flags);
3772 
3773 static void fill_elf_header(struct elfhdr *, int, uint16_t, uint32_t);
3774 static void fill_note(struct memelfnote *, const char *, int,
3775                       unsigned int, void *);
3776 static void fill_prstatus(struct target_elf_prstatus *, const TaskState *, int);
3777 static int fill_psinfo(struct target_elf_prpsinfo *, const TaskState *);
3778 static void fill_auxv_note(struct memelfnote *, const TaskState *);
3779 static void fill_elf_note_phdr(struct elf_phdr *, int, off_t);
3780 static size_t note_size(const struct memelfnote *);
3781 static void free_note_info(struct elf_note_info *);
3782 static int fill_note_info(struct elf_note_info *, long, const CPUArchState *);
3783 static void fill_thread_info(struct elf_note_info *, const CPUArchState *);
3784 
3785 static int dump_write(int, const void *, size_t);
3786 static int write_note(struct memelfnote *, int);
3787 static int write_note_info(struct elf_note_info *, int);
3788 
3789 #ifdef BSWAP_NEEDED
3790 static void bswap_prstatus(struct target_elf_prstatus *prstatus)
3791 {
3792     prstatus->pr_info.si_signo = tswap32(prstatus->pr_info.si_signo);
3793     prstatus->pr_info.si_code = tswap32(prstatus->pr_info.si_code);
3794     prstatus->pr_info.si_errno = tswap32(prstatus->pr_info.si_errno);
3795     prstatus->pr_cursig = tswap16(prstatus->pr_cursig);
3796     prstatus->pr_sigpend = tswapal(prstatus->pr_sigpend);
3797     prstatus->pr_sighold = tswapal(prstatus->pr_sighold);
3798     prstatus->pr_pid = tswap32(prstatus->pr_pid);
3799     prstatus->pr_ppid = tswap32(prstatus->pr_ppid);
3800     prstatus->pr_pgrp = tswap32(prstatus->pr_pgrp);
3801     prstatus->pr_sid = tswap32(prstatus->pr_sid);
3802     /* cpu times are not filled, so we skip them */
3803     /* regs should be in correct format already */
3804     prstatus->pr_fpvalid = tswap32(prstatus->pr_fpvalid);
3805 }
3806 
3807 static void bswap_psinfo(struct target_elf_prpsinfo *psinfo)
3808 {
3809     psinfo->pr_flag = tswapal(psinfo->pr_flag);
3810     psinfo->pr_uid = tswap16(psinfo->pr_uid);
3811     psinfo->pr_gid = tswap16(psinfo->pr_gid);
3812     psinfo->pr_pid = tswap32(psinfo->pr_pid);
3813     psinfo->pr_ppid = tswap32(psinfo->pr_ppid);
3814     psinfo->pr_pgrp = tswap32(psinfo->pr_pgrp);
3815     psinfo->pr_sid = tswap32(psinfo->pr_sid);
3816 }
3817 
3818 static void bswap_note(struct elf_note *en)
3819 {
3820     bswap32s(&en->n_namesz);
3821     bswap32s(&en->n_descsz);
3822     bswap32s(&en->n_type);
3823 }
3824 #else
3825 static inline void bswap_prstatus(struct target_elf_prstatus *p) { }
3826 static inline void bswap_psinfo(struct target_elf_prpsinfo *p) {}
3827 static inline void bswap_note(struct elf_note *en) { }
3828 #endif /* BSWAP_NEEDED */
3829 
3830 /*
3831  * Minimal support for linux memory regions.  These are needed
3832  * when we are finding out what memory exactly belongs to
3833  * emulated process.  No locks needed here, as long as
3834  * thread that received the signal is stopped.
3835  */
3836 
3837 static struct mm_struct *vma_init(void)
3838 {
3839     struct mm_struct *mm;
3840 
3841     if ((mm = g_malloc(sizeof (*mm))) == NULL)
3842         return (NULL);
3843 
3844     mm->mm_count = 0;
3845     QTAILQ_INIT(&mm->mm_mmap);
3846 
3847     return (mm);
3848 }
3849 
3850 static void vma_delete(struct mm_struct *mm)
3851 {
3852     struct vm_area_struct *vma;
3853 
3854     while ((vma = vma_first(mm)) != NULL) {
3855         QTAILQ_REMOVE(&mm->mm_mmap, vma, vma_link);
3856         g_free(vma);
3857     }
3858     g_free(mm);
3859 }
3860 
3861 static int vma_add_mapping(struct mm_struct *mm, target_ulong start,
3862                            target_ulong end, abi_ulong flags)
3863 {
3864     struct vm_area_struct *vma;
3865 
3866     if ((vma = g_malloc0(sizeof (*vma))) == NULL)
3867         return (-1);
3868 
3869     vma->vma_start = start;
3870     vma->vma_end = end;
3871     vma->vma_flags = flags;
3872 
3873     QTAILQ_INSERT_TAIL(&mm->mm_mmap, vma, vma_link);
3874     mm->mm_count++;
3875 
3876     return (0);
3877 }
3878 
3879 static struct vm_area_struct *vma_first(const struct mm_struct *mm)
3880 {
3881     return (QTAILQ_FIRST(&mm->mm_mmap));
3882 }
3883 
3884 static struct vm_area_struct *vma_next(struct vm_area_struct *vma)
3885 {
3886     return (QTAILQ_NEXT(vma, vma_link));
3887 }
3888 
3889 static int vma_get_mapping_count(const struct mm_struct *mm)
3890 {
3891     return (mm->mm_count);
3892 }
3893 
3894 /*
3895  * Calculate file (dump) size of given memory region.
3896  */
3897 static abi_ulong vma_dump_size(const struct vm_area_struct *vma)
3898 {
3899     /* if we cannot even read the first page, skip it */
3900     if (!access_ok_untagged(VERIFY_READ, vma->vma_start, TARGET_PAGE_SIZE))
3901         return (0);
3902 
3903     /*
3904      * Usually we don't dump executable pages as they contain
3905      * non-writable code that debugger can read directly from
3906      * target library etc.  However, thread stacks are marked
3907      * also executable so we read in first page of given region
3908      * and check whether it contains elf header.  If there is
3909      * no elf header, we dump it.
3910      */
3911     if (vma->vma_flags & PROT_EXEC) {
3912         char page[TARGET_PAGE_SIZE];
3913 
3914         if (copy_from_user(page, vma->vma_start, sizeof (page))) {
3915             return 0;
3916         }
3917         if ((page[EI_MAG0] == ELFMAG0) &&
3918             (page[EI_MAG1] == ELFMAG1) &&
3919             (page[EI_MAG2] == ELFMAG2) &&
3920             (page[EI_MAG3] == ELFMAG3)) {
3921             /*
3922              * Mappings are possibly from ELF binary.  Don't dump
3923              * them.
3924              */
3925             return (0);
3926         }
3927     }
3928 
3929     return (vma->vma_end - vma->vma_start);
3930 }
3931 
3932 static int vma_walker(void *priv, target_ulong start, target_ulong end,
3933                       unsigned long flags)
3934 {
3935     struct mm_struct *mm = (struct mm_struct *)priv;
3936 
3937     vma_add_mapping(mm, start, end, flags);
3938     return (0);
3939 }
3940 
3941 static void fill_note(struct memelfnote *note, const char *name, int type,
3942                       unsigned int sz, void *data)
3943 {
3944     unsigned int namesz;
3945 
3946     namesz = strlen(name) + 1;
3947     note->name = name;
3948     note->namesz = namesz;
3949     note->namesz_rounded = roundup(namesz, sizeof (int32_t));
3950     note->type = type;
3951     note->datasz = sz;
3952     note->datasz_rounded = roundup(sz, sizeof (int32_t));
3953 
3954     note->data = data;
3955 
3956     /*
3957      * We calculate rounded up note size here as specified by
3958      * ELF document.
3959      */
3960     note->notesz = sizeof (struct elf_note) +
3961         note->namesz_rounded + note->datasz_rounded;
3962 }
3963 
3964 static void fill_elf_header(struct elfhdr *elf, int segs, uint16_t machine,
3965                             uint32_t flags)
3966 {
3967     (void) memset(elf, 0, sizeof(*elf));
3968 
3969     (void) memcpy(elf->e_ident, ELFMAG, SELFMAG);
3970     elf->e_ident[EI_CLASS] = ELF_CLASS;
3971     elf->e_ident[EI_DATA] = ELF_DATA;
3972     elf->e_ident[EI_VERSION] = EV_CURRENT;
3973     elf->e_ident[EI_OSABI] = ELF_OSABI;
3974 
3975     elf->e_type = ET_CORE;
3976     elf->e_machine = machine;
3977     elf->e_version = EV_CURRENT;
3978     elf->e_phoff = sizeof(struct elfhdr);
3979     elf->e_flags = flags;
3980     elf->e_ehsize = sizeof(struct elfhdr);
3981     elf->e_phentsize = sizeof(struct elf_phdr);
3982     elf->e_phnum = segs;
3983 
3984     bswap_ehdr(elf);
3985 }
3986 
3987 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, off_t offset)
3988 {
3989     phdr->p_type = PT_NOTE;
3990     phdr->p_offset = offset;
3991     phdr->p_vaddr = 0;
3992     phdr->p_paddr = 0;
3993     phdr->p_filesz = sz;
3994     phdr->p_memsz = 0;
3995     phdr->p_flags = 0;
3996     phdr->p_align = 0;
3997 
3998     bswap_phdr(phdr, 1);
3999 }
4000 
4001 static size_t note_size(const struct memelfnote *note)
4002 {
4003     return (note->notesz);
4004 }
4005 
4006 static void fill_prstatus(struct target_elf_prstatus *prstatus,
4007                           const TaskState *ts, int signr)
4008 {
4009     (void) memset(prstatus, 0, sizeof (*prstatus));
4010     prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
4011     prstatus->pr_pid = ts->ts_tid;
4012     prstatus->pr_ppid = getppid();
4013     prstatus->pr_pgrp = getpgrp();
4014     prstatus->pr_sid = getsid(0);
4015 
4016     bswap_prstatus(prstatus);
4017 }
4018 
4019 static int fill_psinfo(struct target_elf_prpsinfo *psinfo, const TaskState *ts)
4020 {
4021     char *base_filename;
4022     unsigned int i, len;
4023 
4024     (void) memset(psinfo, 0, sizeof (*psinfo));
4025 
4026     len = ts->info->env_strings - ts->info->arg_strings;
4027     if (len >= ELF_PRARGSZ)
4028         len = ELF_PRARGSZ - 1;
4029     if (copy_from_user(&psinfo->pr_psargs, ts->info->arg_strings, len)) {
4030         return -EFAULT;
4031     }
4032     for (i = 0; i < len; i++)
4033         if (psinfo->pr_psargs[i] == 0)
4034             psinfo->pr_psargs[i] = ' ';
4035     psinfo->pr_psargs[len] = 0;
4036 
4037     psinfo->pr_pid = getpid();
4038     psinfo->pr_ppid = getppid();
4039     psinfo->pr_pgrp = getpgrp();
4040     psinfo->pr_sid = getsid(0);
4041     psinfo->pr_uid = getuid();
4042     psinfo->pr_gid = getgid();
4043 
4044     base_filename = g_path_get_basename(ts->bprm->filename);
4045     /*
4046      * Using strncpy here is fine: at max-length,
4047      * this field is not NUL-terminated.
4048      */
4049     (void) strncpy(psinfo->pr_fname, base_filename,
4050                    sizeof(psinfo->pr_fname));
4051 
4052     g_free(base_filename);
4053     bswap_psinfo(psinfo);
4054     return (0);
4055 }
4056 
4057 static void fill_auxv_note(struct memelfnote *note, const TaskState *ts)
4058 {
4059     elf_addr_t auxv = (elf_addr_t)ts->info->saved_auxv;
4060     elf_addr_t orig_auxv = auxv;
4061     void *ptr;
4062     int len = ts->info->auxv_len;
4063 
4064     /*
4065      * Auxiliary vector is stored in target process stack.  It contains
4066      * {type, value} pairs that we need to dump into note.  This is not
4067      * strictly necessary but we do it here for sake of completeness.
4068      */
4069 
4070     /* read in whole auxv vector and copy it to memelfnote */
4071     ptr = lock_user(VERIFY_READ, orig_auxv, len, 0);
4072     if (ptr != NULL) {
4073         fill_note(note, "CORE", NT_AUXV, len, ptr);
4074         unlock_user(ptr, auxv, len);
4075     }
4076 }
4077 
4078 /*
4079  * Constructs name of coredump file.  We have following convention
4080  * for the name:
4081  *     qemu_<basename-of-target-binary>_<date>-<time>_<pid>.core
4082  *
4083  * Returns the filename
4084  */
4085 static char *core_dump_filename(const TaskState *ts)
4086 {
4087     g_autoptr(GDateTime) now = g_date_time_new_now_local();
4088     g_autofree char *nowstr = g_date_time_format(now, "%Y%m%d-%H%M%S");
4089     g_autofree char *base_filename = g_path_get_basename(ts->bprm->filename);
4090 
4091     return g_strdup_printf("qemu_%s_%s_%d.core",
4092                            base_filename, nowstr, (int)getpid());
4093 }
4094 
4095 static int dump_write(int fd, const void *ptr, size_t size)
4096 {
4097     const char *bufp = (const char *)ptr;
4098     ssize_t bytes_written, bytes_left;
4099     struct rlimit dumpsize;
4100     off_t pos;
4101 
4102     bytes_written = 0;
4103     getrlimit(RLIMIT_CORE, &dumpsize);
4104     if ((pos = lseek(fd, 0, SEEK_CUR))==-1) {
4105         if (errno == ESPIPE) { /* not a seekable stream */
4106             bytes_left = size;
4107         } else {
4108             return pos;
4109         }
4110     } else {
4111         if (dumpsize.rlim_cur <= pos) {
4112             return -1;
4113         } else if (dumpsize.rlim_cur == RLIM_INFINITY) {
4114             bytes_left = size;
4115         } else {
4116             size_t limit_left=dumpsize.rlim_cur - pos;
4117             bytes_left = limit_left >= size ? size : limit_left ;
4118         }
4119     }
4120 
4121     /*
4122      * In normal conditions, single write(2) should do but
4123      * in case of socket etc. this mechanism is more portable.
4124      */
4125     do {
4126         bytes_written = write(fd, bufp, bytes_left);
4127         if (bytes_written < 0) {
4128             if (errno == EINTR)
4129                 continue;
4130             return (-1);
4131         } else if (bytes_written == 0) { /* eof */
4132             return (-1);
4133         }
4134         bufp += bytes_written;
4135         bytes_left -= bytes_written;
4136     } while (bytes_left > 0);
4137 
4138     return (0);
4139 }
4140 
4141 static int write_note(struct memelfnote *men, int fd)
4142 {
4143     struct elf_note en;
4144 
4145     en.n_namesz = men->namesz;
4146     en.n_type = men->type;
4147     en.n_descsz = men->datasz;
4148 
4149     bswap_note(&en);
4150 
4151     if (dump_write(fd, &en, sizeof(en)) != 0)
4152         return (-1);
4153     if (dump_write(fd, men->name, men->namesz_rounded) != 0)
4154         return (-1);
4155     if (dump_write(fd, men->data, men->datasz_rounded) != 0)
4156         return (-1);
4157 
4158     return (0);
4159 }
4160 
4161 static void fill_thread_info(struct elf_note_info *info, const CPUArchState *env)
4162 {
4163     CPUState *cpu = env_cpu((CPUArchState *)env);
4164     TaskState *ts = (TaskState *)cpu->opaque;
4165     struct elf_thread_status *ets;
4166 
4167     ets = g_malloc0(sizeof (*ets));
4168     ets->num_notes = 1; /* only prstatus is dumped */
4169     fill_prstatus(&ets->prstatus, ts, 0);
4170     elf_core_copy_regs(&ets->prstatus.pr_reg, env);
4171     fill_note(&ets->notes[0], "CORE", NT_PRSTATUS, sizeof (ets->prstatus),
4172               &ets->prstatus);
4173 
4174     QTAILQ_INSERT_TAIL(&info->thread_list, ets, ets_link);
4175 
4176     info->notes_size += note_size(&ets->notes[0]);
4177 }
4178 
4179 static void init_note_info(struct elf_note_info *info)
4180 {
4181     /* Initialize the elf_note_info structure so that it is at
4182      * least safe to call free_note_info() on it. Must be
4183      * called before calling fill_note_info().
4184      */
4185     memset(info, 0, sizeof (*info));
4186     QTAILQ_INIT(&info->thread_list);
4187 }
4188 
4189 static int fill_note_info(struct elf_note_info *info,
4190                           long signr, const CPUArchState *env)
4191 {
4192 #define NUMNOTES 3
4193     CPUState *cpu = env_cpu((CPUArchState *)env);
4194     TaskState *ts = (TaskState *)cpu->opaque;
4195     int i;
4196 
4197     info->notes = g_new0(struct memelfnote, NUMNOTES);
4198     if (info->notes == NULL)
4199         return (-ENOMEM);
4200     info->prstatus = g_malloc0(sizeof (*info->prstatus));
4201     if (info->prstatus == NULL)
4202         return (-ENOMEM);
4203     info->psinfo = g_malloc0(sizeof (*info->psinfo));
4204     if (info->prstatus == NULL)
4205         return (-ENOMEM);
4206 
4207     /*
4208      * First fill in status (and registers) of current thread
4209      * including process info & aux vector.
4210      */
4211     fill_prstatus(info->prstatus, ts, signr);
4212     elf_core_copy_regs(&info->prstatus->pr_reg, env);
4213     fill_note(&info->notes[0], "CORE", NT_PRSTATUS,
4214               sizeof (*info->prstatus), info->prstatus);
4215     fill_psinfo(info->psinfo, ts);
4216     fill_note(&info->notes[1], "CORE", NT_PRPSINFO,
4217               sizeof (*info->psinfo), info->psinfo);
4218     fill_auxv_note(&info->notes[2], ts);
4219     info->numnote = 3;
4220 
4221     info->notes_size = 0;
4222     for (i = 0; i < info->numnote; i++)
4223         info->notes_size += note_size(&info->notes[i]);
4224 
4225     /* read and fill status of all threads */
4226     cpu_list_lock();
4227     CPU_FOREACH(cpu) {
4228         if (cpu == thread_cpu) {
4229             continue;
4230         }
4231         fill_thread_info(info, cpu->env_ptr);
4232     }
4233     cpu_list_unlock();
4234 
4235     return (0);
4236 }
4237 
4238 static void free_note_info(struct elf_note_info *info)
4239 {
4240     struct elf_thread_status *ets;
4241 
4242     while (!QTAILQ_EMPTY(&info->thread_list)) {
4243         ets = QTAILQ_FIRST(&info->thread_list);
4244         QTAILQ_REMOVE(&info->thread_list, ets, ets_link);
4245         g_free(ets);
4246     }
4247 
4248     g_free(info->prstatus);
4249     g_free(info->psinfo);
4250     g_free(info->notes);
4251 }
4252 
4253 static int write_note_info(struct elf_note_info *info, int fd)
4254 {
4255     struct elf_thread_status *ets;
4256     int i, error = 0;
4257 
4258     /* write prstatus, psinfo and auxv for current thread */
4259     for (i = 0; i < info->numnote; i++)
4260         if ((error = write_note(&info->notes[i], fd)) != 0)
4261             return (error);
4262 
4263     /* write prstatus for each thread */
4264     QTAILQ_FOREACH(ets, &info->thread_list, ets_link) {
4265         if ((error = write_note(&ets->notes[0], fd)) != 0)
4266             return (error);
4267     }
4268 
4269     return (0);
4270 }
4271 
4272 /*
4273  * Write out ELF coredump.
4274  *
4275  * See documentation of ELF object file format in:
4276  * http://www.caldera.com/developers/devspecs/gabi41.pdf
4277  *
4278  * Coredump format in linux is following:
4279  *
4280  * 0   +----------------------+         \
4281  *     | ELF header           | ET_CORE  |
4282  *     +----------------------+          |
4283  *     | ELF program headers  |          |--- headers
4284  *     | - NOTE section       |          |
4285  *     | - PT_LOAD sections   |          |
4286  *     +----------------------+         /
4287  *     | NOTEs:               |
4288  *     | - NT_PRSTATUS        |
4289  *     | - NT_PRSINFO         |
4290  *     | - NT_AUXV            |
4291  *     +----------------------+ <-- aligned to target page
4292  *     | Process memory dump  |
4293  *     :                      :
4294  *     .                      .
4295  *     :                      :
4296  *     |                      |
4297  *     +----------------------+
4298  *
4299  * NT_PRSTATUS -> struct elf_prstatus (per thread)
4300  * NT_PRSINFO  -> struct elf_prpsinfo
4301  * NT_AUXV is array of { type, value } pairs (see fill_auxv_note()).
4302  *
4303  * Format follows System V format as close as possible.  Current
4304  * version limitations are as follows:
4305  *     - no floating point registers are dumped
4306  *
4307  * Function returns 0 in case of success, negative errno otherwise.
4308  *
4309  * TODO: make this work also during runtime: it should be
4310  * possible to force coredump from running process and then
4311  * continue processing.  For example qemu could set up SIGUSR2
4312  * handler (provided that target process haven't registered
4313  * handler for that) that does the dump when signal is received.
4314  */
4315 static int elf_core_dump(int signr, const CPUArchState *env)
4316 {
4317     const CPUState *cpu = env_cpu((CPUArchState *)env);
4318     const TaskState *ts = (const TaskState *)cpu->opaque;
4319     struct vm_area_struct *vma = NULL;
4320     g_autofree char *corefile = NULL;
4321     struct elf_note_info info;
4322     struct elfhdr elf;
4323     struct elf_phdr phdr;
4324     struct rlimit dumpsize;
4325     struct mm_struct *mm = NULL;
4326     off_t offset = 0, data_offset = 0;
4327     int segs = 0;
4328     int fd = -1;
4329 
4330     init_note_info(&info);
4331 
4332     errno = 0;
4333     getrlimit(RLIMIT_CORE, &dumpsize);
4334     if (dumpsize.rlim_cur == 0)
4335         return 0;
4336 
4337     corefile = core_dump_filename(ts);
4338 
4339     if ((fd = open(corefile, O_WRONLY | O_CREAT,
4340                    S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH)) < 0)
4341         return (-errno);
4342 
4343     /*
4344      * Walk through target process memory mappings and
4345      * set up structure containing this information.  After
4346      * this point vma_xxx functions can be used.
4347      */
4348     if ((mm = vma_init()) == NULL)
4349         goto out;
4350 
4351     walk_memory_regions(mm, vma_walker);
4352     segs = vma_get_mapping_count(mm);
4353 
4354     /*
4355      * Construct valid coredump ELF header.  We also
4356      * add one more segment for notes.
4357      */
4358     fill_elf_header(&elf, segs + 1, ELF_MACHINE, 0);
4359     if (dump_write(fd, &elf, sizeof (elf)) != 0)
4360         goto out;
4361 
4362     /* fill in the in-memory version of notes */
4363     if (fill_note_info(&info, signr, env) < 0)
4364         goto out;
4365 
4366     offset += sizeof (elf);                             /* elf header */
4367     offset += (segs + 1) * sizeof (struct elf_phdr);    /* program headers */
4368 
4369     /* write out notes program header */
4370     fill_elf_note_phdr(&phdr, info.notes_size, offset);
4371 
4372     offset += info.notes_size;
4373     if (dump_write(fd, &phdr, sizeof (phdr)) != 0)
4374         goto out;
4375 
4376     /*
4377      * ELF specification wants data to start at page boundary so
4378      * we align it here.
4379      */
4380     data_offset = offset = roundup(offset, ELF_EXEC_PAGESIZE);
4381 
4382     /*
4383      * Write program headers for memory regions mapped in
4384      * the target process.
4385      */
4386     for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) {
4387         (void) memset(&phdr, 0, sizeof (phdr));
4388 
4389         phdr.p_type = PT_LOAD;
4390         phdr.p_offset = offset;
4391         phdr.p_vaddr = vma->vma_start;
4392         phdr.p_paddr = 0;
4393         phdr.p_filesz = vma_dump_size(vma);
4394         offset += phdr.p_filesz;
4395         phdr.p_memsz = vma->vma_end - vma->vma_start;
4396         phdr.p_flags = vma->vma_flags & PROT_READ ? PF_R : 0;
4397         if (vma->vma_flags & PROT_WRITE)
4398             phdr.p_flags |= PF_W;
4399         if (vma->vma_flags & PROT_EXEC)
4400             phdr.p_flags |= PF_X;
4401         phdr.p_align = ELF_EXEC_PAGESIZE;
4402 
4403         bswap_phdr(&phdr, 1);
4404         if (dump_write(fd, &phdr, sizeof(phdr)) != 0) {
4405             goto out;
4406         }
4407     }
4408 
4409     /*
4410      * Next we write notes just after program headers.  No
4411      * alignment needed here.
4412      */
4413     if (write_note_info(&info, fd) < 0)
4414         goto out;
4415 
4416     /* align data to page boundary */
4417     if (lseek(fd, data_offset, SEEK_SET) != data_offset)
4418         goto out;
4419 
4420     /*
4421      * Finally we can dump process memory into corefile as well.
4422      */
4423     for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) {
4424         abi_ulong addr;
4425         abi_ulong end;
4426 
4427         end = vma->vma_start + vma_dump_size(vma);
4428 
4429         for (addr = vma->vma_start; addr < end;
4430              addr += TARGET_PAGE_SIZE) {
4431             char page[TARGET_PAGE_SIZE];
4432             int error;
4433 
4434             /*
4435              *  Read in page from target process memory and
4436              *  write it to coredump file.
4437              */
4438             error = copy_from_user(page, addr, sizeof (page));
4439             if (error != 0) {
4440                 (void) fprintf(stderr, "unable to dump " TARGET_ABI_FMT_lx "\n",
4441                                addr);
4442                 errno = -error;
4443                 goto out;
4444             }
4445             if (dump_write(fd, page, TARGET_PAGE_SIZE) < 0)
4446                 goto out;
4447         }
4448     }
4449 
4450  out:
4451     free_note_info(&info);
4452     if (mm != NULL)
4453         vma_delete(mm);
4454     (void) close(fd);
4455 
4456     if (errno != 0)
4457         return (-errno);
4458     return (0);
4459 }
4460 #endif /* USE_ELF_CORE_DUMP */
4461 
4462 void do_init_thread(struct target_pt_regs *regs, struct image_info *infop)
4463 {
4464     init_thread(regs, infop);
4465 }
4466