xref: /openbmc/qemu/linux-user/elfload.c (revision 954a6c4f)
1 /* This is the Linux kernel elf-loading code, ported into user space */
2 #include "qemu/osdep.h"
3 #include <sys/param.h>
4 
5 #include <sys/resource.h>
6 #include <sys/shm.h>
7 
8 #include "qemu.h"
9 #include "user-internals.h"
10 #include "signal-common.h"
11 #include "loader.h"
12 #include "user-mmap.h"
13 #include "disas/disas.h"
14 #include "qemu/bitops.h"
15 #include "qemu/path.h"
16 #include "qemu/queue.h"
17 #include "qemu/guest-random.h"
18 #include "qemu/units.h"
19 #include "qemu/selfmap.h"
20 #include "qapi/error.h"
21 #include "target_signal.h"
22 
23 #ifdef _ARCH_PPC64
24 #undef ARCH_DLINFO
25 #undef ELF_PLATFORM
26 #undef ELF_HWCAP
27 #undef ELF_HWCAP2
28 #undef ELF_CLASS
29 #undef ELF_DATA
30 #undef ELF_ARCH
31 #endif
32 
33 #define ELF_OSABI   ELFOSABI_SYSV
34 
35 /* from personality.h */
36 
37 /*
38  * Flags for bug emulation.
39  *
40  * These occupy the top three bytes.
41  */
42 enum {
43     ADDR_NO_RANDOMIZE = 0x0040000,      /* disable randomization of VA space */
44     FDPIC_FUNCPTRS =    0x0080000,      /* userspace function ptrs point to
45                                            descriptors (signal handling) */
46     MMAP_PAGE_ZERO =    0x0100000,
47     ADDR_COMPAT_LAYOUT = 0x0200000,
48     READ_IMPLIES_EXEC = 0x0400000,
49     ADDR_LIMIT_32BIT =  0x0800000,
50     SHORT_INODE =       0x1000000,
51     WHOLE_SECONDS =     0x2000000,
52     STICKY_TIMEOUTS =   0x4000000,
53     ADDR_LIMIT_3GB =    0x8000000,
54 };
55 
56 /*
57  * Personality types.
58  *
59  * These go in the low byte.  Avoid using the top bit, it will
60  * conflict with error returns.
61  */
62 enum {
63     PER_LINUX =         0x0000,
64     PER_LINUX_32BIT =   0x0000 | ADDR_LIMIT_32BIT,
65     PER_LINUX_FDPIC =   0x0000 | FDPIC_FUNCPTRS,
66     PER_SVR4 =          0x0001 | STICKY_TIMEOUTS | MMAP_PAGE_ZERO,
67     PER_SVR3 =          0x0002 | STICKY_TIMEOUTS | SHORT_INODE,
68     PER_SCOSVR3 =       0x0003 | STICKY_TIMEOUTS | WHOLE_SECONDS | SHORT_INODE,
69     PER_OSR5 =          0x0003 | STICKY_TIMEOUTS | WHOLE_SECONDS,
70     PER_WYSEV386 =      0x0004 | STICKY_TIMEOUTS | SHORT_INODE,
71     PER_ISCR4 =         0x0005 | STICKY_TIMEOUTS,
72     PER_BSD =           0x0006,
73     PER_SUNOS =         0x0006 | STICKY_TIMEOUTS,
74     PER_XENIX =         0x0007 | STICKY_TIMEOUTS | SHORT_INODE,
75     PER_LINUX32 =       0x0008,
76     PER_LINUX32_3GB =   0x0008 | ADDR_LIMIT_3GB,
77     PER_IRIX32 =        0x0009 | STICKY_TIMEOUTS,/* IRIX5 32-bit */
78     PER_IRIXN32 =       0x000a | STICKY_TIMEOUTS,/* IRIX6 new 32-bit */
79     PER_IRIX64 =        0x000b | STICKY_TIMEOUTS,/* IRIX6 64-bit */
80     PER_RISCOS =        0x000c,
81     PER_SOLARIS =       0x000d | STICKY_TIMEOUTS,
82     PER_UW7 =           0x000e | STICKY_TIMEOUTS | MMAP_PAGE_ZERO,
83     PER_OSF4 =          0x000f,                  /* OSF/1 v4 */
84     PER_HPUX =          0x0010,
85     PER_MASK =          0x00ff,
86 };
87 
88 /*
89  * Return the base personality without flags.
90  */
91 #define personality(pers)       (pers & PER_MASK)
92 
93 int info_is_fdpic(struct image_info *info)
94 {
95     return info->personality == PER_LINUX_FDPIC;
96 }
97 
98 /* this flag is uneffective under linux too, should be deleted */
99 #ifndef MAP_DENYWRITE
100 #define MAP_DENYWRITE 0
101 #endif
102 
103 /* should probably go in elf.h */
104 #ifndef ELIBBAD
105 #define ELIBBAD 80
106 #endif
107 
108 #if TARGET_BIG_ENDIAN
109 #define ELF_DATA        ELFDATA2MSB
110 #else
111 #define ELF_DATA        ELFDATA2LSB
112 #endif
113 
114 #ifdef TARGET_ABI_MIPSN32
115 typedef abi_ullong      target_elf_greg_t;
116 #define tswapreg(ptr)   tswap64(ptr)
117 #else
118 typedef abi_ulong       target_elf_greg_t;
119 #define tswapreg(ptr)   tswapal(ptr)
120 #endif
121 
122 #ifdef USE_UID16
123 typedef abi_ushort      target_uid_t;
124 typedef abi_ushort      target_gid_t;
125 #else
126 typedef abi_uint        target_uid_t;
127 typedef abi_uint        target_gid_t;
128 #endif
129 typedef abi_int         target_pid_t;
130 
131 #ifdef TARGET_I386
132 
133 #define ELF_HWCAP get_elf_hwcap()
134 
135 static uint32_t get_elf_hwcap(void)
136 {
137     X86CPU *cpu = X86_CPU(thread_cpu);
138 
139     return cpu->env.features[FEAT_1_EDX];
140 }
141 
142 #ifdef TARGET_X86_64
143 #define ELF_START_MMAP 0x2aaaaab000ULL
144 
145 #define ELF_CLASS      ELFCLASS64
146 #define ELF_ARCH       EM_X86_64
147 
148 #define ELF_PLATFORM   "x86_64"
149 
150 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop)
151 {
152     regs->rax = 0;
153     regs->rsp = infop->start_stack;
154     regs->rip = infop->entry;
155 }
156 
157 #define ELF_NREG    27
158 typedef target_elf_greg_t  target_elf_gregset_t[ELF_NREG];
159 
160 /*
161  * Note that ELF_NREG should be 29 as there should be place for
162  * TRAPNO and ERR "registers" as well but linux doesn't dump
163  * those.
164  *
165  * See linux kernel: arch/x86/include/asm/elf.h
166  */
167 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUX86State *env)
168 {
169     (*regs)[0] = tswapreg(env->regs[15]);
170     (*regs)[1] = tswapreg(env->regs[14]);
171     (*regs)[2] = tswapreg(env->regs[13]);
172     (*regs)[3] = tswapreg(env->regs[12]);
173     (*regs)[4] = tswapreg(env->regs[R_EBP]);
174     (*regs)[5] = tswapreg(env->regs[R_EBX]);
175     (*regs)[6] = tswapreg(env->regs[11]);
176     (*regs)[7] = tswapreg(env->regs[10]);
177     (*regs)[8] = tswapreg(env->regs[9]);
178     (*regs)[9] = tswapreg(env->regs[8]);
179     (*regs)[10] = tswapreg(env->regs[R_EAX]);
180     (*regs)[11] = tswapreg(env->regs[R_ECX]);
181     (*regs)[12] = tswapreg(env->regs[R_EDX]);
182     (*regs)[13] = tswapreg(env->regs[R_ESI]);
183     (*regs)[14] = tswapreg(env->regs[R_EDI]);
184     (*regs)[15] = tswapreg(env->regs[R_EAX]); /* XXX */
185     (*regs)[16] = tswapreg(env->eip);
186     (*regs)[17] = tswapreg(env->segs[R_CS].selector & 0xffff);
187     (*regs)[18] = tswapreg(env->eflags);
188     (*regs)[19] = tswapreg(env->regs[R_ESP]);
189     (*regs)[20] = tswapreg(env->segs[R_SS].selector & 0xffff);
190     (*regs)[21] = tswapreg(env->segs[R_FS].selector & 0xffff);
191     (*regs)[22] = tswapreg(env->segs[R_GS].selector & 0xffff);
192     (*regs)[23] = tswapreg(env->segs[R_DS].selector & 0xffff);
193     (*regs)[24] = tswapreg(env->segs[R_ES].selector & 0xffff);
194     (*regs)[25] = tswapreg(env->segs[R_FS].selector & 0xffff);
195     (*regs)[26] = tswapreg(env->segs[R_GS].selector & 0xffff);
196 }
197 
198 #if ULONG_MAX > UINT32_MAX
199 #define INIT_GUEST_COMMPAGE
200 static bool init_guest_commpage(void)
201 {
202     /*
203      * The vsyscall page is at a high negative address aka kernel space,
204      * which means that we cannot actually allocate it with target_mmap.
205      * We still should be able to use page_set_flags, unless the user
206      * has specified -R reserved_va, which would trigger an assert().
207      */
208     if (reserved_va != 0 &&
209         TARGET_VSYSCALL_PAGE + TARGET_PAGE_SIZE >= reserved_va) {
210         error_report("Cannot allocate vsyscall page");
211         exit(EXIT_FAILURE);
212     }
213     page_set_flags(TARGET_VSYSCALL_PAGE,
214                    TARGET_VSYSCALL_PAGE + TARGET_PAGE_SIZE,
215                    PAGE_EXEC | PAGE_VALID);
216     return true;
217 }
218 #endif
219 #else
220 
221 #define ELF_START_MMAP 0x80000000
222 
223 /*
224  * This is used to ensure we don't load something for the wrong architecture.
225  */
226 #define elf_check_arch(x) ( ((x) == EM_386) || ((x) == EM_486) )
227 
228 /*
229  * These are used to set parameters in the core dumps.
230  */
231 #define ELF_CLASS       ELFCLASS32
232 #define ELF_ARCH        EM_386
233 
234 #define ELF_PLATFORM get_elf_platform()
235 #define EXSTACK_DEFAULT true
236 
237 static const char *get_elf_platform(void)
238 {
239     static char elf_platform[] = "i386";
240     int family = object_property_get_int(OBJECT(thread_cpu), "family", NULL);
241     if (family > 6) {
242         family = 6;
243     }
244     if (family >= 3) {
245         elf_platform[1] = '0' + family;
246     }
247     return elf_platform;
248 }
249 
250 static inline void init_thread(struct target_pt_regs *regs,
251                                struct image_info *infop)
252 {
253     regs->esp = infop->start_stack;
254     regs->eip = infop->entry;
255 
256     /* SVR4/i386 ABI (pages 3-31, 3-32) says that when the program
257        starts %edx contains a pointer to a function which might be
258        registered using `atexit'.  This provides a mean for the
259        dynamic linker to call DT_FINI functions for shared libraries
260        that have been loaded before the code runs.
261 
262        A value of 0 tells we have no such handler.  */
263     regs->edx = 0;
264 }
265 
266 #define ELF_NREG    17
267 typedef target_elf_greg_t  target_elf_gregset_t[ELF_NREG];
268 
269 /*
270  * Note that ELF_NREG should be 19 as there should be place for
271  * TRAPNO and ERR "registers" as well but linux doesn't dump
272  * those.
273  *
274  * See linux kernel: arch/x86/include/asm/elf.h
275  */
276 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUX86State *env)
277 {
278     (*regs)[0] = tswapreg(env->regs[R_EBX]);
279     (*regs)[1] = tswapreg(env->regs[R_ECX]);
280     (*regs)[2] = tswapreg(env->regs[R_EDX]);
281     (*regs)[3] = tswapreg(env->regs[R_ESI]);
282     (*regs)[4] = tswapreg(env->regs[R_EDI]);
283     (*regs)[5] = tswapreg(env->regs[R_EBP]);
284     (*regs)[6] = tswapreg(env->regs[R_EAX]);
285     (*regs)[7] = tswapreg(env->segs[R_DS].selector & 0xffff);
286     (*regs)[8] = tswapreg(env->segs[R_ES].selector & 0xffff);
287     (*regs)[9] = tswapreg(env->segs[R_FS].selector & 0xffff);
288     (*regs)[10] = tswapreg(env->segs[R_GS].selector & 0xffff);
289     (*regs)[11] = tswapreg(env->regs[R_EAX]); /* XXX */
290     (*regs)[12] = tswapreg(env->eip);
291     (*regs)[13] = tswapreg(env->segs[R_CS].selector & 0xffff);
292     (*regs)[14] = tswapreg(env->eflags);
293     (*regs)[15] = tswapreg(env->regs[R_ESP]);
294     (*regs)[16] = tswapreg(env->segs[R_SS].selector & 0xffff);
295 }
296 #endif
297 
298 #define USE_ELF_CORE_DUMP
299 #define ELF_EXEC_PAGESIZE       4096
300 
301 #endif
302 
303 #ifdef TARGET_ARM
304 
305 #ifndef TARGET_AARCH64
306 /* 32 bit ARM definitions */
307 
308 #define ELF_START_MMAP 0x80000000
309 
310 #define ELF_ARCH        EM_ARM
311 #define ELF_CLASS       ELFCLASS32
312 #define EXSTACK_DEFAULT true
313 
314 static inline void init_thread(struct target_pt_regs *regs,
315                                struct image_info *infop)
316 {
317     abi_long stack = infop->start_stack;
318     memset(regs, 0, sizeof(*regs));
319 
320     regs->uregs[16] = ARM_CPU_MODE_USR;
321     if (infop->entry & 1) {
322         regs->uregs[16] |= CPSR_T;
323     }
324     regs->uregs[15] = infop->entry & 0xfffffffe;
325     regs->uregs[13] = infop->start_stack;
326     /* FIXME - what to for failure of get_user()? */
327     get_user_ual(regs->uregs[2], stack + 8); /* envp */
328     get_user_ual(regs->uregs[1], stack + 4); /* envp */
329     /* XXX: it seems that r0 is zeroed after ! */
330     regs->uregs[0] = 0;
331     /* For uClinux PIC binaries.  */
332     /* XXX: Linux does this only on ARM with no MMU (do we care ?) */
333     regs->uregs[10] = infop->start_data;
334 
335     /* Support ARM FDPIC.  */
336     if (info_is_fdpic(infop)) {
337         /* As described in the ABI document, r7 points to the loadmap info
338          * prepared by the kernel. If an interpreter is needed, r8 points
339          * to the interpreter loadmap and r9 points to the interpreter
340          * PT_DYNAMIC info. If no interpreter is needed, r8 is zero, and
341          * r9 points to the main program PT_DYNAMIC info.
342          */
343         regs->uregs[7] = infop->loadmap_addr;
344         if (infop->interpreter_loadmap_addr) {
345             /* Executable is dynamically loaded.  */
346             regs->uregs[8] = infop->interpreter_loadmap_addr;
347             regs->uregs[9] = infop->interpreter_pt_dynamic_addr;
348         } else {
349             regs->uregs[8] = 0;
350             regs->uregs[9] = infop->pt_dynamic_addr;
351         }
352     }
353 }
354 
355 #define ELF_NREG    18
356 typedef target_elf_greg_t  target_elf_gregset_t[ELF_NREG];
357 
358 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUARMState *env)
359 {
360     (*regs)[0] = tswapreg(env->regs[0]);
361     (*regs)[1] = tswapreg(env->regs[1]);
362     (*regs)[2] = tswapreg(env->regs[2]);
363     (*regs)[3] = tswapreg(env->regs[3]);
364     (*regs)[4] = tswapreg(env->regs[4]);
365     (*regs)[5] = tswapreg(env->regs[5]);
366     (*regs)[6] = tswapreg(env->regs[6]);
367     (*regs)[7] = tswapreg(env->regs[7]);
368     (*regs)[8] = tswapreg(env->regs[8]);
369     (*regs)[9] = tswapreg(env->regs[9]);
370     (*regs)[10] = tswapreg(env->regs[10]);
371     (*regs)[11] = tswapreg(env->regs[11]);
372     (*regs)[12] = tswapreg(env->regs[12]);
373     (*regs)[13] = tswapreg(env->regs[13]);
374     (*regs)[14] = tswapreg(env->regs[14]);
375     (*regs)[15] = tswapreg(env->regs[15]);
376 
377     (*regs)[16] = tswapreg(cpsr_read((CPUARMState *)env));
378     (*regs)[17] = tswapreg(env->regs[0]); /* XXX */
379 }
380 
381 #define USE_ELF_CORE_DUMP
382 #define ELF_EXEC_PAGESIZE       4096
383 
384 enum
385 {
386     ARM_HWCAP_ARM_SWP       = 1 << 0,
387     ARM_HWCAP_ARM_HALF      = 1 << 1,
388     ARM_HWCAP_ARM_THUMB     = 1 << 2,
389     ARM_HWCAP_ARM_26BIT     = 1 << 3,
390     ARM_HWCAP_ARM_FAST_MULT = 1 << 4,
391     ARM_HWCAP_ARM_FPA       = 1 << 5,
392     ARM_HWCAP_ARM_VFP       = 1 << 6,
393     ARM_HWCAP_ARM_EDSP      = 1 << 7,
394     ARM_HWCAP_ARM_JAVA      = 1 << 8,
395     ARM_HWCAP_ARM_IWMMXT    = 1 << 9,
396     ARM_HWCAP_ARM_CRUNCH    = 1 << 10,
397     ARM_HWCAP_ARM_THUMBEE   = 1 << 11,
398     ARM_HWCAP_ARM_NEON      = 1 << 12,
399     ARM_HWCAP_ARM_VFPv3     = 1 << 13,
400     ARM_HWCAP_ARM_VFPv3D16  = 1 << 14,
401     ARM_HWCAP_ARM_TLS       = 1 << 15,
402     ARM_HWCAP_ARM_VFPv4     = 1 << 16,
403     ARM_HWCAP_ARM_IDIVA     = 1 << 17,
404     ARM_HWCAP_ARM_IDIVT     = 1 << 18,
405     ARM_HWCAP_ARM_VFPD32    = 1 << 19,
406     ARM_HWCAP_ARM_LPAE      = 1 << 20,
407     ARM_HWCAP_ARM_EVTSTRM   = 1 << 21,
408 };
409 
410 enum {
411     ARM_HWCAP2_ARM_AES      = 1 << 0,
412     ARM_HWCAP2_ARM_PMULL    = 1 << 1,
413     ARM_HWCAP2_ARM_SHA1     = 1 << 2,
414     ARM_HWCAP2_ARM_SHA2     = 1 << 3,
415     ARM_HWCAP2_ARM_CRC32    = 1 << 4,
416 };
417 
418 /* The commpage only exists for 32 bit kernels */
419 
420 #define HI_COMMPAGE (intptr_t)0xffff0f00u
421 
422 static bool init_guest_commpage(void)
423 {
424     abi_ptr commpage = HI_COMMPAGE & -qemu_host_page_size;
425     void *want = g2h_untagged(commpage);
426     void *addr = mmap(want, qemu_host_page_size, PROT_READ | PROT_WRITE,
427                       MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0);
428 
429     if (addr == MAP_FAILED) {
430         perror("Allocating guest commpage");
431         exit(EXIT_FAILURE);
432     }
433     if (addr != want) {
434         return false;
435     }
436 
437     /* Set kernel helper versions; rest of page is 0.  */
438     __put_user(5, (uint32_t *)g2h_untagged(0xffff0ffcu));
439 
440     if (mprotect(addr, qemu_host_page_size, PROT_READ)) {
441         perror("Protecting guest commpage");
442         exit(EXIT_FAILURE);
443     }
444 
445     page_set_flags(commpage, commpage + qemu_host_page_size,
446                    PAGE_READ | PAGE_EXEC | PAGE_VALID);
447     return true;
448 }
449 
450 #define ELF_HWCAP get_elf_hwcap()
451 #define ELF_HWCAP2 get_elf_hwcap2()
452 
453 static uint32_t get_elf_hwcap(void)
454 {
455     ARMCPU *cpu = ARM_CPU(thread_cpu);
456     uint32_t hwcaps = 0;
457 
458     hwcaps |= ARM_HWCAP_ARM_SWP;
459     hwcaps |= ARM_HWCAP_ARM_HALF;
460     hwcaps |= ARM_HWCAP_ARM_THUMB;
461     hwcaps |= ARM_HWCAP_ARM_FAST_MULT;
462 
463     /* probe for the extra features */
464 #define GET_FEATURE(feat, hwcap) \
465     do { if (arm_feature(&cpu->env, feat)) { hwcaps |= hwcap; } } while (0)
466 
467 #define GET_FEATURE_ID(feat, hwcap) \
468     do { if (cpu_isar_feature(feat, cpu)) { hwcaps |= hwcap; } } while (0)
469 
470     /* EDSP is in v5TE and above, but all our v5 CPUs are v5TE */
471     GET_FEATURE(ARM_FEATURE_V5, ARM_HWCAP_ARM_EDSP);
472     GET_FEATURE(ARM_FEATURE_IWMMXT, ARM_HWCAP_ARM_IWMMXT);
473     GET_FEATURE(ARM_FEATURE_THUMB2EE, ARM_HWCAP_ARM_THUMBEE);
474     GET_FEATURE(ARM_FEATURE_NEON, ARM_HWCAP_ARM_NEON);
475     GET_FEATURE(ARM_FEATURE_V6K, ARM_HWCAP_ARM_TLS);
476     GET_FEATURE(ARM_FEATURE_LPAE, ARM_HWCAP_ARM_LPAE);
477     GET_FEATURE_ID(aa32_arm_div, ARM_HWCAP_ARM_IDIVA);
478     GET_FEATURE_ID(aa32_thumb_div, ARM_HWCAP_ARM_IDIVT);
479     GET_FEATURE_ID(aa32_vfp, ARM_HWCAP_ARM_VFP);
480 
481     if (cpu_isar_feature(aa32_fpsp_v3, cpu) ||
482         cpu_isar_feature(aa32_fpdp_v3, cpu)) {
483         hwcaps |= ARM_HWCAP_ARM_VFPv3;
484         if (cpu_isar_feature(aa32_simd_r32, cpu)) {
485             hwcaps |= ARM_HWCAP_ARM_VFPD32;
486         } else {
487             hwcaps |= ARM_HWCAP_ARM_VFPv3D16;
488         }
489     }
490     GET_FEATURE_ID(aa32_simdfmac, ARM_HWCAP_ARM_VFPv4);
491 
492     return hwcaps;
493 }
494 
495 static uint32_t get_elf_hwcap2(void)
496 {
497     ARMCPU *cpu = ARM_CPU(thread_cpu);
498     uint32_t hwcaps = 0;
499 
500     GET_FEATURE_ID(aa32_aes, ARM_HWCAP2_ARM_AES);
501     GET_FEATURE_ID(aa32_pmull, ARM_HWCAP2_ARM_PMULL);
502     GET_FEATURE_ID(aa32_sha1, ARM_HWCAP2_ARM_SHA1);
503     GET_FEATURE_ID(aa32_sha2, ARM_HWCAP2_ARM_SHA2);
504     GET_FEATURE_ID(aa32_crc32, ARM_HWCAP2_ARM_CRC32);
505     return hwcaps;
506 }
507 
508 #undef GET_FEATURE
509 #undef GET_FEATURE_ID
510 
511 #define ELF_PLATFORM get_elf_platform()
512 
513 static const char *get_elf_platform(void)
514 {
515     CPUARMState *env = thread_cpu->env_ptr;
516 
517 #if TARGET_BIG_ENDIAN
518 # define END  "b"
519 #else
520 # define END  "l"
521 #endif
522 
523     if (arm_feature(env, ARM_FEATURE_V8)) {
524         return "v8" END;
525     } else if (arm_feature(env, ARM_FEATURE_V7)) {
526         if (arm_feature(env, ARM_FEATURE_M)) {
527             return "v7m" END;
528         } else {
529             return "v7" END;
530         }
531     } else if (arm_feature(env, ARM_FEATURE_V6)) {
532         return "v6" END;
533     } else if (arm_feature(env, ARM_FEATURE_V5)) {
534         return "v5" END;
535     } else {
536         return "v4" END;
537     }
538 
539 #undef END
540 }
541 
542 #else
543 /* 64 bit ARM definitions */
544 #define ELF_START_MMAP 0x80000000
545 
546 #define ELF_ARCH        EM_AARCH64
547 #define ELF_CLASS       ELFCLASS64
548 #if TARGET_BIG_ENDIAN
549 # define ELF_PLATFORM    "aarch64_be"
550 #else
551 # define ELF_PLATFORM    "aarch64"
552 #endif
553 
554 static inline void init_thread(struct target_pt_regs *regs,
555                                struct image_info *infop)
556 {
557     abi_long stack = infop->start_stack;
558     memset(regs, 0, sizeof(*regs));
559 
560     regs->pc = infop->entry & ~0x3ULL;
561     regs->sp = stack;
562 }
563 
564 #define ELF_NREG    34
565 typedef target_elf_greg_t  target_elf_gregset_t[ELF_NREG];
566 
567 static void elf_core_copy_regs(target_elf_gregset_t *regs,
568                                const CPUARMState *env)
569 {
570     int i;
571 
572     for (i = 0; i < 32; i++) {
573         (*regs)[i] = tswapreg(env->xregs[i]);
574     }
575     (*regs)[32] = tswapreg(env->pc);
576     (*regs)[33] = tswapreg(pstate_read((CPUARMState *)env));
577 }
578 
579 #define USE_ELF_CORE_DUMP
580 #define ELF_EXEC_PAGESIZE       4096
581 
582 enum {
583     ARM_HWCAP_A64_FP            = 1 << 0,
584     ARM_HWCAP_A64_ASIMD         = 1 << 1,
585     ARM_HWCAP_A64_EVTSTRM       = 1 << 2,
586     ARM_HWCAP_A64_AES           = 1 << 3,
587     ARM_HWCAP_A64_PMULL         = 1 << 4,
588     ARM_HWCAP_A64_SHA1          = 1 << 5,
589     ARM_HWCAP_A64_SHA2          = 1 << 6,
590     ARM_HWCAP_A64_CRC32         = 1 << 7,
591     ARM_HWCAP_A64_ATOMICS       = 1 << 8,
592     ARM_HWCAP_A64_FPHP          = 1 << 9,
593     ARM_HWCAP_A64_ASIMDHP       = 1 << 10,
594     ARM_HWCAP_A64_CPUID         = 1 << 11,
595     ARM_HWCAP_A64_ASIMDRDM      = 1 << 12,
596     ARM_HWCAP_A64_JSCVT         = 1 << 13,
597     ARM_HWCAP_A64_FCMA          = 1 << 14,
598     ARM_HWCAP_A64_LRCPC         = 1 << 15,
599     ARM_HWCAP_A64_DCPOP         = 1 << 16,
600     ARM_HWCAP_A64_SHA3          = 1 << 17,
601     ARM_HWCAP_A64_SM3           = 1 << 18,
602     ARM_HWCAP_A64_SM4           = 1 << 19,
603     ARM_HWCAP_A64_ASIMDDP       = 1 << 20,
604     ARM_HWCAP_A64_SHA512        = 1 << 21,
605     ARM_HWCAP_A64_SVE           = 1 << 22,
606     ARM_HWCAP_A64_ASIMDFHM      = 1 << 23,
607     ARM_HWCAP_A64_DIT           = 1 << 24,
608     ARM_HWCAP_A64_USCAT         = 1 << 25,
609     ARM_HWCAP_A64_ILRCPC        = 1 << 26,
610     ARM_HWCAP_A64_FLAGM         = 1 << 27,
611     ARM_HWCAP_A64_SSBS          = 1 << 28,
612     ARM_HWCAP_A64_SB            = 1 << 29,
613     ARM_HWCAP_A64_PACA          = 1 << 30,
614     ARM_HWCAP_A64_PACG          = 1UL << 31,
615 
616     ARM_HWCAP2_A64_DCPODP       = 1 << 0,
617     ARM_HWCAP2_A64_SVE2         = 1 << 1,
618     ARM_HWCAP2_A64_SVEAES       = 1 << 2,
619     ARM_HWCAP2_A64_SVEPMULL     = 1 << 3,
620     ARM_HWCAP2_A64_SVEBITPERM   = 1 << 4,
621     ARM_HWCAP2_A64_SVESHA3      = 1 << 5,
622     ARM_HWCAP2_A64_SVESM4       = 1 << 6,
623     ARM_HWCAP2_A64_FLAGM2       = 1 << 7,
624     ARM_HWCAP2_A64_FRINT        = 1 << 8,
625     ARM_HWCAP2_A64_SVEI8MM      = 1 << 9,
626     ARM_HWCAP2_A64_SVEF32MM     = 1 << 10,
627     ARM_HWCAP2_A64_SVEF64MM     = 1 << 11,
628     ARM_HWCAP2_A64_SVEBF16      = 1 << 12,
629     ARM_HWCAP2_A64_I8MM         = 1 << 13,
630     ARM_HWCAP2_A64_BF16         = 1 << 14,
631     ARM_HWCAP2_A64_DGH          = 1 << 15,
632     ARM_HWCAP2_A64_RNG          = 1 << 16,
633     ARM_HWCAP2_A64_BTI          = 1 << 17,
634     ARM_HWCAP2_A64_MTE          = 1 << 18,
635     ARM_HWCAP2_A64_ECV          = 1 << 19,
636     ARM_HWCAP2_A64_AFP          = 1 << 20,
637     ARM_HWCAP2_A64_RPRES        = 1 << 21,
638     ARM_HWCAP2_A64_MTE3         = 1 << 22,
639     ARM_HWCAP2_A64_SME          = 1 << 23,
640     ARM_HWCAP2_A64_SME_I16I64   = 1 << 24,
641     ARM_HWCAP2_A64_SME_F64F64   = 1 << 25,
642     ARM_HWCAP2_A64_SME_I8I32    = 1 << 26,
643     ARM_HWCAP2_A64_SME_F16F32   = 1 << 27,
644     ARM_HWCAP2_A64_SME_B16F32   = 1 << 28,
645     ARM_HWCAP2_A64_SME_F32F32   = 1 << 29,
646     ARM_HWCAP2_A64_SME_FA64     = 1 << 30,
647 };
648 
649 #define ELF_HWCAP   get_elf_hwcap()
650 #define ELF_HWCAP2  get_elf_hwcap2()
651 
652 #define GET_FEATURE_ID(feat, hwcap) \
653     do { if (cpu_isar_feature(feat, cpu)) { hwcaps |= hwcap; } } while (0)
654 
655 static uint32_t get_elf_hwcap(void)
656 {
657     ARMCPU *cpu = ARM_CPU(thread_cpu);
658     uint32_t hwcaps = 0;
659 
660     hwcaps |= ARM_HWCAP_A64_FP;
661     hwcaps |= ARM_HWCAP_A64_ASIMD;
662     hwcaps |= ARM_HWCAP_A64_CPUID;
663 
664     /* probe for the extra features */
665 
666     GET_FEATURE_ID(aa64_aes, ARM_HWCAP_A64_AES);
667     GET_FEATURE_ID(aa64_pmull, ARM_HWCAP_A64_PMULL);
668     GET_FEATURE_ID(aa64_sha1, ARM_HWCAP_A64_SHA1);
669     GET_FEATURE_ID(aa64_sha256, ARM_HWCAP_A64_SHA2);
670     GET_FEATURE_ID(aa64_sha512, ARM_HWCAP_A64_SHA512);
671     GET_FEATURE_ID(aa64_crc32, ARM_HWCAP_A64_CRC32);
672     GET_FEATURE_ID(aa64_sha3, ARM_HWCAP_A64_SHA3);
673     GET_FEATURE_ID(aa64_sm3, ARM_HWCAP_A64_SM3);
674     GET_FEATURE_ID(aa64_sm4, ARM_HWCAP_A64_SM4);
675     GET_FEATURE_ID(aa64_fp16, ARM_HWCAP_A64_FPHP | ARM_HWCAP_A64_ASIMDHP);
676     GET_FEATURE_ID(aa64_atomics, ARM_HWCAP_A64_ATOMICS);
677     GET_FEATURE_ID(aa64_rdm, ARM_HWCAP_A64_ASIMDRDM);
678     GET_FEATURE_ID(aa64_dp, ARM_HWCAP_A64_ASIMDDP);
679     GET_FEATURE_ID(aa64_fcma, ARM_HWCAP_A64_FCMA);
680     GET_FEATURE_ID(aa64_sve, ARM_HWCAP_A64_SVE);
681     GET_FEATURE_ID(aa64_pauth, ARM_HWCAP_A64_PACA | ARM_HWCAP_A64_PACG);
682     GET_FEATURE_ID(aa64_fhm, ARM_HWCAP_A64_ASIMDFHM);
683     GET_FEATURE_ID(aa64_jscvt, ARM_HWCAP_A64_JSCVT);
684     GET_FEATURE_ID(aa64_sb, ARM_HWCAP_A64_SB);
685     GET_FEATURE_ID(aa64_condm_4, ARM_HWCAP_A64_FLAGM);
686     GET_FEATURE_ID(aa64_dcpop, ARM_HWCAP_A64_DCPOP);
687     GET_FEATURE_ID(aa64_rcpc_8_3, ARM_HWCAP_A64_LRCPC);
688     GET_FEATURE_ID(aa64_rcpc_8_4, ARM_HWCAP_A64_ILRCPC);
689 
690     return hwcaps;
691 }
692 
693 static uint32_t get_elf_hwcap2(void)
694 {
695     ARMCPU *cpu = ARM_CPU(thread_cpu);
696     uint32_t hwcaps = 0;
697 
698     GET_FEATURE_ID(aa64_dcpodp, ARM_HWCAP2_A64_DCPODP);
699     GET_FEATURE_ID(aa64_sve2, ARM_HWCAP2_A64_SVE2);
700     GET_FEATURE_ID(aa64_sve2_aes, ARM_HWCAP2_A64_SVEAES);
701     GET_FEATURE_ID(aa64_sve2_pmull128, ARM_HWCAP2_A64_SVEPMULL);
702     GET_FEATURE_ID(aa64_sve2_bitperm, ARM_HWCAP2_A64_SVEBITPERM);
703     GET_FEATURE_ID(aa64_sve2_sha3, ARM_HWCAP2_A64_SVESHA3);
704     GET_FEATURE_ID(aa64_sve2_sm4, ARM_HWCAP2_A64_SVESM4);
705     GET_FEATURE_ID(aa64_condm_5, ARM_HWCAP2_A64_FLAGM2);
706     GET_FEATURE_ID(aa64_frint, ARM_HWCAP2_A64_FRINT);
707     GET_FEATURE_ID(aa64_sve_i8mm, ARM_HWCAP2_A64_SVEI8MM);
708     GET_FEATURE_ID(aa64_sve_f32mm, ARM_HWCAP2_A64_SVEF32MM);
709     GET_FEATURE_ID(aa64_sve_f64mm, ARM_HWCAP2_A64_SVEF64MM);
710     GET_FEATURE_ID(aa64_sve_bf16, ARM_HWCAP2_A64_SVEBF16);
711     GET_FEATURE_ID(aa64_i8mm, ARM_HWCAP2_A64_I8MM);
712     GET_FEATURE_ID(aa64_bf16, ARM_HWCAP2_A64_BF16);
713     GET_FEATURE_ID(aa64_rndr, ARM_HWCAP2_A64_RNG);
714     GET_FEATURE_ID(aa64_bti, ARM_HWCAP2_A64_BTI);
715     GET_FEATURE_ID(aa64_mte, ARM_HWCAP2_A64_MTE);
716     GET_FEATURE_ID(aa64_sme, (ARM_HWCAP2_A64_SME |
717                               ARM_HWCAP2_A64_SME_F32F32 |
718                               ARM_HWCAP2_A64_SME_B16F32 |
719                               ARM_HWCAP2_A64_SME_F16F32 |
720                               ARM_HWCAP2_A64_SME_I8I32));
721     GET_FEATURE_ID(aa64_sme_f64f64, ARM_HWCAP2_A64_SME_F64F64);
722     GET_FEATURE_ID(aa64_sme_i16i64, ARM_HWCAP2_A64_SME_I16I64);
723     GET_FEATURE_ID(aa64_sme_fa64, ARM_HWCAP2_A64_SME_FA64);
724 
725     return hwcaps;
726 }
727 
728 #undef GET_FEATURE_ID
729 
730 #endif /* not TARGET_AARCH64 */
731 #endif /* TARGET_ARM */
732 
733 #ifdef TARGET_SPARC
734 #ifdef TARGET_SPARC64
735 
736 #define ELF_START_MMAP 0x80000000
737 #define ELF_HWCAP  (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \
738                     | HWCAP_SPARC_MULDIV | HWCAP_SPARC_V9)
739 #ifndef TARGET_ABI32
740 #define elf_check_arch(x) ( (x) == EM_SPARCV9 || (x) == EM_SPARC32PLUS )
741 #else
742 #define elf_check_arch(x) ( (x) == EM_SPARC32PLUS || (x) == EM_SPARC )
743 #endif
744 
745 #define ELF_CLASS   ELFCLASS64
746 #define ELF_ARCH    EM_SPARCV9
747 #else
748 #define ELF_START_MMAP 0x80000000
749 #define ELF_HWCAP  (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \
750                     | HWCAP_SPARC_MULDIV)
751 #define ELF_CLASS   ELFCLASS32
752 #define ELF_ARCH    EM_SPARC
753 #endif /* TARGET_SPARC64 */
754 
755 static inline void init_thread(struct target_pt_regs *regs,
756                                struct image_info *infop)
757 {
758     /* Note that target_cpu_copy_regs does not read psr/tstate. */
759     regs->pc = infop->entry;
760     regs->npc = regs->pc + 4;
761     regs->y = 0;
762     regs->u_regs[14] = (infop->start_stack - 16 * sizeof(abi_ulong)
763                         - TARGET_STACK_BIAS);
764 }
765 #endif /* TARGET_SPARC */
766 
767 #ifdef TARGET_PPC
768 
769 #define ELF_MACHINE    PPC_ELF_MACHINE
770 #define ELF_START_MMAP 0x80000000
771 
772 #if defined(TARGET_PPC64)
773 
774 #define elf_check_arch(x) ( (x) == EM_PPC64 )
775 
776 #define ELF_CLASS       ELFCLASS64
777 
778 #else
779 
780 #define ELF_CLASS       ELFCLASS32
781 #define EXSTACK_DEFAULT true
782 
783 #endif
784 
785 #define ELF_ARCH        EM_PPC
786 
787 /* Feature masks for the Aux Vector Hardware Capabilities (AT_HWCAP).
788    See arch/powerpc/include/asm/cputable.h.  */
789 enum {
790     QEMU_PPC_FEATURE_32 = 0x80000000,
791     QEMU_PPC_FEATURE_64 = 0x40000000,
792     QEMU_PPC_FEATURE_601_INSTR = 0x20000000,
793     QEMU_PPC_FEATURE_HAS_ALTIVEC = 0x10000000,
794     QEMU_PPC_FEATURE_HAS_FPU = 0x08000000,
795     QEMU_PPC_FEATURE_HAS_MMU = 0x04000000,
796     QEMU_PPC_FEATURE_HAS_4xxMAC = 0x02000000,
797     QEMU_PPC_FEATURE_UNIFIED_CACHE = 0x01000000,
798     QEMU_PPC_FEATURE_HAS_SPE = 0x00800000,
799     QEMU_PPC_FEATURE_HAS_EFP_SINGLE = 0x00400000,
800     QEMU_PPC_FEATURE_HAS_EFP_DOUBLE = 0x00200000,
801     QEMU_PPC_FEATURE_NO_TB = 0x00100000,
802     QEMU_PPC_FEATURE_POWER4 = 0x00080000,
803     QEMU_PPC_FEATURE_POWER5 = 0x00040000,
804     QEMU_PPC_FEATURE_POWER5_PLUS = 0x00020000,
805     QEMU_PPC_FEATURE_CELL = 0x00010000,
806     QEMU_PPC_FEATURE_BOOKE = 0x00008000,
807     QEMU_PPC_FEATURE_SMT = 0x00004000,
808     QEMU_PPC_FEATURE_ICACHE_SNOOP = 0x00002000,
809     QEMU_PPC_FEATURE_ARCH_2_05 = 0x00001000,
810     QEMU_PPC_FEATURE_PA6T = 0x00000800,
811     QEMU_PPC_FEATURE_HAS_DFP = 0x00000400,
812     QEMU_PPC_FEATURE_POWER6_EXT = 0x00000200,
813     QEMU_PPC_FEATURE_ARCH_2_06 = 0x00000100,
814     QEMU_PPC_FEATURE_HAS_VSX = 0x00000080,
815     QEMU_PPC_FEATURE_PSERIES_PERFMON_COMPAT = 0x00000040,
816 
817     QEMU_PPC_FEATURE_TRUE_LE = 0x00000002,
818     QEMU_PPC_FEATURE_PPC_LE = 0x00000001,
819 
820     /* Feature definitions in AT_HWCAP2.  */
821     QEMU_PPC_FEATURE2_ARCH_2_07 = 0x80000000, /* ISA 2.07 */
822     QEMU_PPC_FEATURE2_HAS_HTM = 0x40000000, /* Hardware Transactional Memory */
823     QEMU_PPC_FEATURE2_HAS_DSCR = 0x20000000, /* Data Stream Control Register */
824     QEMU_PPC_FEATURE2_HAS_EBB = 0x10000000, /* Event Base Branching */
825     QEMU_PPC_FEATURE2_HAS_ISEL = 0x08000000, /* Integer Select */
826     QEMU_PPC_FEATURE2_HAS_TAR = 0x04000000, /* Target Address Register */
827     QEMU_PPC_FEATURE2_VEC_CRYPTO = 0x02000000,
828     QEMU_PPC_FEATURE2_HTM_NOSC = 0x01000000,
829     QEMU_PPC_FEATURE2_ARCH_3_00 = 0x00800000, /* ISA 3.00 */
830     QEMU_PPC_FEATURE2_HAS_IEEE128 = 0x00400000, /* VSX IEEE Bin Float 128-bit */
831     QEMU_PPC_FEATURE2_DARN = 0x00200000, /* darn random number insn */
832     QEMU_PPC_FEATURE2_SCV = 0x00100000, /* scv syscall */
833     QEMU_PPC_FEATURE2_HTM_NO_SUSPEND = 0x00080000, /* TM w/o suspended state */
834     QEMU_PPC_FEATURE2_ARCH_3_1 = 0x00040000, /* ISA 3.1 */
835     QEMU_PPC_FEATURE2_MMA = 0x00020000, /* Matrix-Multiply Assist */
836 };
837 
838 #define ELF_HWCAP get_elf_hwcap()
839 
840 static uint32_t get_elf_hwcap(void)
841 {
842     PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);
843     uint32_t features = 0;
844 
845     /* We don't have to be terribly complete here; the high points are
846        Altivec/FP/SPE support.  Anything else is just a bonus.  */
847 #define GET_FEATURE(flag, feature)                                      \
848     do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0)
849 #define GET_FEATURE2(flags, feature) \
850     do { \
851         if ((cpu->env.insns_flags2 & flags) == flags) { \
852             features |= feature; \
853         } \
854     } while (0)
855     GET_FEATURE(PPC_64B, QEMU_PPC_FEATURE_64);
856     GET_FEATURE(PPC_FLOAT, QEMU_PPC_FEATURE_HAS_FPU);
857     GET_FEATURE(PPC_ALTIVEC, QEMU_PPC_FEATURE_HAS_ALTIVEC);
858     GET_FEATURE(PPC_SPE, QEMU_PPC_FEATURE_HAS_SPE);
859     GET_FEATURE(PPC_SPE_SINGLE, QEMU_PPC_FEATURE_HAS_EFP_SINGLE);
860     GET_FEATURE(PPC_SPE_DOUBLE, QEMU_PPC_FEATURE_HAS_EFP_DOUBLE);
861     GET_FEATURE(PPC_BOOKE, QEMU_PPC_FEATURE_BOOKE);
862     GET_FEATURE(PPC_405_MAC, QEMU_PPC_FEATURE_HAS_4xxMAC);
863     GET_FEATURE2(PPC2_DFP, QEMU_PPC_FEATURE_HAS_DFP);
864     GET_FEATURE2(PPC2_VSX, QEMU_PPC_FEATURE_HAS_VSX);
865     GET_FEATURE2((PPC2_PERM_ISA206 | PPC2_DIVE_ISA206 | PPC2_ATOMIC_ISA206 |
866                   PPC2_FP_CVT_ISA206 | PPC2_FP_TST_ISA206),
867                   QEMU_PPC_FEATURE_ARCH_2_06);
868 #undef GET_FEATURE
869 #undef GET_FEATURE2
870 
871     return features;
872 }
873 
874 #define ELF_HWCAP2 get_elf_hwcap2()
875 
876 static uint32_t get_elf_hwcap2(void)
877 {
878     PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);
879     uint32_t features = 0;
880 
881 #define GET_FEATURE(flag, feature)                                      \
882     do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0)
883 #define GET_FEATURE2(flag, feature)                                      \
884     do { if (cpu->env.insns_flags2 & flag) { features |= feature; } } while (0)
885 
886     GET_FEATURE(PPC_ISEL, QEMU_PPC_FEATURE2_HAS_ISEL);
887     GET_FEATURE2(PPC2_BCTAR_ISA207, QEMU_PPC_FEATURE2_HAS_TAR);
888     GET_FEATURE2((PPC2_BCTAR_ISA207 | PPC2_LSQ_ISA207 | PPC2_ALTIVEC_207 |
889                   PPC2_ISA207S), QEMU_PPC_FEATURE2_ARCH_2_07 |
890                   QEMU_PPC_FEATURE2_VEC_CRYPTO);
891     GET_FEATURE2(PPC2_ISA300, QEMU_PPC_FEATURE2_ARCH_3_00 |
892                  QEMU_PPC_FEATURE2_DARN | QEMU_PPC_FEATURE2_HAS_IEEE128);
893     GET_FEATURE2(PPC2_ISA310, QEMU_PPC_FEATURE2_ARCH_3_1 |
894                  QEMU_PPC_FEATURE2_MMA);
895 
896 #undef GET_FEATURE
897 #undef GET_FEATURE2
898 
899     return features;
900 }
901 
902 /*
903  * The requirements here are:
904  * - keep the final alignment of sp (sp & 0xf)
905  * - make sure the 32-bit value at the first 16 byte aligned position of
906  *   AUXV is greater than 16 for glibc compatibility.
907  *   AT_IGNOREPPC is used for that.
908  * - for compatibility with glibc ARCH_DLINFO must always be defined on PPC,
909  *   even if DLINFO_ARCH_ITEMS goes to zero or is undefined.
910  */
911 #define DLINFO_ARCH_ITEMS       5
912 #define ARCH_DLINFO                                     \
913     do {                                                \
914         PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);              \
915         /*                                              \
916          * Handle glibc compatibility: these magic entries must \
917          * be at the lowest addresses in the final auxv.        \
918          */                                             \
919         NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC);        \
920         NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC);        \
921         NEW_AUX_ENT(AT_DCACHEBSIZE, cpu->env.dcache_line_size); \
922         NEW_AUX_ENT(AT_ICACHEBSIZE, cpu->env.icache_line_size); \
923         NEW_AUX_ENT(AT_UCACHEBSIZE, 0);                 \
924     } while (0)
925 
926 static inline void init_thread(struct target_pt_regs *_regs, struct image_info *infop)
927 {
928     _regs->gpr[1] = infop->start_stack;
929 #if defined(TARGET_PPC64)
930     if (get_ppc64_abi(infop) < 2) {
931         uint64_t val;
932         get_user_u64(val, infop->entry + 8);
933         _regs->gpr[2] = val + infop->load_bias;
934         get_user_u64(val, infop->entry);
935         infop->entry = val + infop->load_bias;
936     } else {
937         _regs->gpr[12] = infop->entry;  /* r12 set to global entry address */
938     }
939 #endif
940     _regs->nip = infop->entry;
941 }
942 
943 /* See linux kernel: arch/powerpc/include/asm/elf.h.  */
944 #define ELF_NREG 48
945 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
946 
947 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUPPCState *env)
948 {
949     int i;
950     target_ulong ccr = 0;
951 
952     for (i = 0; i < ARRAY_SIZE(env->gpr); i++) {
953         (*regs)[i] = tswapreg(env->gpr[i]);
954     }
955 
956     (*regs)[32] = tswapreg(env->nip);
957     (*regs)[33] = tswapreg(env->msr);
958     (*regs)[35] = tswapreg(env->ctr);
959     (*regs)[36] = tswapreg(env->lr);
960     (*regs)[37] = tswapreg(cpu_read_xer(env));
961 
962     for (i = 0; i < ARRAY_SIZE(env->crf); i++) {
963         ccr |= env->crf[i] << (32 - ((i + 1) * 4));
964     }
965     (*regs)[38] = tswapreg(ccr);
966 }
967 
968 #define USE_ELF_CORE_DUMP
969 #define ELF_EXEC_PAGESIZE       4096
970 
971 #endif
972 
973 #ifdef TARGET_LOONGARCH64
974 
975 #define ELF_START_MMAP 0x80000000
976 
977 #define ELF_CLASS   ELFCLASS64
978 #define ELF_ARCH    EM_LOONGARCH
979 #define EXSTACK_DEFAULT true
980 
981 #define elf_check_arch(x) ((x) == EM_LOONGARCH)
982 
983 static inline void init_thread(struct target_pt_regs *regs,
984                                struct image_info *infop)
985 {
986     /*Set crmd PG,DA = 1,0 */
987     regs->csr.crmd = 2 << 3;
988     regs->csr.era = infop->entry;
989     regs->regs[3] = infop->start_stack;
990 }
991 
992 /* See linux kernel: arch/loongarch/include/asm/elf.h */
993 #define ELF_NREG 45
994 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
995 
996 enum {
997     TARGET_EF_R0 = 0,
998     TARGET_EF_CSR_ERA = TARGET_EF_R0 + 33,
999     TARGET_EF_CSR_BADV = TARGET_EF_R0 + 34,
1000 };
1001 
1002 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1003                                const CPULoongArchState *env)
1004 {
1005     int i;
1006 
1007     (*regs)[TARGET_EF_R0] = 0;
1008 
1009     for (i = 1; i < ARRAY_SIZE(env->gpr); i++) {
1010         (*regs)[TARGET_EF_R0 + i] = tswapreg(env->gpr[i]);
1011     }
1012 
1013     (*regs)[TARGET_EF_CSR_ERA] = tswapreg(env->pc);
1014     (*regs)[TARGET_EF_CSR_BADV] = tswapreg(env->CSR_BADV);
1015 }
1016 
1017 #define USE_ELF_CORE_DUMP
1018 #define ELF_EXEC_PAGESIZE        4096
1019 
1020 #define ELF_HWCAP get_elf_hwcap()
1021 
1022 /* See arch/loongarch/include/uapi/asm/hwcap.h */
1023 enum {
1024     HWCAP_LOONGARCH_CPUCFG   = (1 << 0),
1025     HWCAP_LOONGARCH_LAM      = (1 << 1),
1026     HWCAP_LOONGARCH_UAL      = (1 << 2),
1027     HWCAP_LOONGARCH_FPU      = (1 << 3),
1028     HWCAP_LOONGARCH_LSX      = (1 << 4),
1029     HWCAP_LOONGARCH_LASX     = (1 << 5),
1030     HWCAP_LOONGARCH_CRC32    = (1 << 6),
1031     HWCAP_LOONGARCH_COMPLEX  = (1 << 7),
1032     HWCAP_LOONGARCH_CRYPTO   = (1 << 8),
1033     HWCAP_LOONGARCH_LVZ      = (1 << 9),
1034     HWCAP_LOONGARCH_LBT_X86  = (1 << 10),
1035     HWCAP_LOONGARCH_LBT_ARM  = (1 << 11),
1036     HWCAP_LOONGARCH_LBT_MIPS = (1 << 12),
1037 };
1038 
1039 static uint32_t get_elf_hwcap(void)
1040 {
1041     LoongArchCPU *cpu = LOONGARCH_CPU(thread_cpu);
1042     uint32_t hwcaps = 0;
1043 
1044     hwcaps |= HWCAP_LOONGARCH_CRC32;
1045 
1046     if (FIELD_EX32(cpu->env.cpucfg[1], CPUCFG1, UAL)) {
1047         hwcaps |= HWCAP_LOONGARCH_UAL;
1048     }
1049 
1050     if (FIELD_EX32(cpu->env.cpucfg[2], CPUCFG2, FP)) {
1051         hwcaps |= HWCAP_LOONGARCH_FPU;
1052     }
1053 
1054     if (FIELD_EX32(cpu->env.cpucfg[2], CPUCFG2, LAM)) {
1055         hwcaps |= HWCAP_LOONGARCH_LAM;
1056     }
1057 
1058     return hwcaps;
1059 }
1060 
1061 #define ELF_PLATFORM "loongarch"
1062 
1063 #endif /* TARGET_LOONGARCH64 */
1064 
1065 #ifdef TARGET_MIPS
1066 
1067 #define ELF_START_MMAP 0x80000000
1068 
1069 #ifdef TARGET_MIPS64
1070 #define ELF_CLASS   ELFCLASS64
1071 #else
1072 #define ELF_CLASS   ELFCLASS32
1073 #endif
1074 #define ELF_ARCH    EM_MIPS
1075 #define EXSTACK_DEFAULT true
1076 
1077 #ifdef TARGET_ABI_MIPSN32
1078 #define elf_check_abi(x) ((x) & EF_MIPS_ABI2)
1079 #else
1080 #define elf_check_abi(x) (!((x) & EF_MIPS_ABI2))
1081 #endif
1082 
1083 #define ELF_BASE_PLATFORM get_elf_base_platform()
1084 
1085 #define MATCH_PLATFORM_INSN(_flags, _base_platform)      \
1086     do { if ((cpu->env.insn_flags & (_flags)) == _flags) \
1087     { return _base_platform; } } while (0)
1088 
1089 static const char *get_elf_base_platform(void)
1090 {
1091     MIPSCPU *cpu = MIPS_CPU(thread_cpu);
1092 
1093     /* 64 bit ISAs goes first */
1094     MATCH_PLATFORM_INSN(CPU_MIPS64R6, "mips64r6");
1095     MATCH_PLATFORM_INSN(CPU_MIPS64R5, "mips64r5");
1096     MATCH_PLATFORM_INSN(CPU_MIPS64R2, "mips64r2");
1097     MATCH_PLATFORM_INSN(CPU_MIPS64R1, "mips64");
1098     MATCH_PLATFORM_INSN(CPU_MIPS5, "mips5");
1099     MATCH_PLATFORM_INSN(CPU_MIPS4, "mips4");
1100     MATCH_PLATFORM_INSN(CPU_MIPS3, "mips3");
1101 
1102     /* 32 bit ISAs */
1103     MATCH_PLATFORM_INSN(CPU_MIPS32R6, "mips32r6");
1104     MATCH_PLATFORM_INSN(CPU_MIPS32R5, "mips32r5");
1105     MATCH_PLATFORM_INSN(CPU_MIPS32R2, "mips32r2");
1106     MATCH_PLATFORM_INSN(CPU_MIPS32R1, "mips32");
1107     MATCH_PLATFORM_INSN(CPU_MIPS2, "mips2");
1108 
1109     /* Fallback */
1110     return "mips";
1111 }
1112 #undef MATCH_PLATFORM_INSN
1113 
1114 static inline void init_thread(struct target_pt_regs *regs,
1115                                struct image_info *infop)
1116 {
1117     regs->cp0_status = 2 << CP0St_KSU;
1118     regs->cp0_epc = infop->entry;
1119     regs->regs[29] = infop->start_stack;
1120 }
1121 
1122 /* See linux kernel: arch/mips/include/asm/elf.h.  */
1123 #define ELF_NREG 45
1124 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1125 
1126 /* See linux kernel: arch/mips/include/asm/reg.h.  */
1127 enum {
1128 #ifdef TARGET_MIPS64
1129     TARGET_EF_R0 = 0,
1130 #else
1131     TARGET_EF_R0 = 6,
1132 #endif
1133     TARGET_EF_R26 = TARGET_EF_R0 + 26,
1134     TARGET_EF_R27 = TARGET_EF_R0 + 27,
1135     TARGET_EF_LO = TARGET_EF_R0 + 32,
1136     TARGET_EF_HI = TARGET_EF_R0 + 33,
1137     TARGET_EF_CP0_EPC = TARGET_EF_R0 + 34,
1138     TARGET_EF_CP0_BADVADDR = TARGET_EF_R0 + 35,
1139     TARGET_EF_CP0_STATUS = TARGET_EF_R0 + 36,
1140     TARGET_EF_CP0_CAUSE = TARGET_EF_R0 + 37
1141 };
1142 
1143 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs.  */
1144 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMIPSState *env)
1145 {
1146     int i;
1147 
1148     for (i = 0; i < TARGET_EF_R0; i++) {
1149         (*regs)[i] = 0;
1150     }
1151     (*regs)[TARGET_EF_R0] = 0;
1152 
1153     for (i = 1; i < ARRAY_SIZE(env->active_tc.gpr); i++) {
1154         (*regs)[TARGET_EF_R0 + i] = tswapreg(env->active_tc.gpr[i]);
1155     }
1156 
1157     (*regs)[TARGET_EF_R26] = 0;
1158     (*regs)[TARGET_EF_R27] = 0;
1159     (*regs)[TARGET_EF_LO] = tswapreg(env->active_tc.LO[0]);
1160     (*regs)[TARGET_EF_HI] = tswapreg(env->active_tc.HI[0]);
1161     (*regs)[TARGET_EF_CP0_EPC] = tswapreg(env->active_tc.PC);
1162     (*regs)[TARGET_EF_CP0_BADVADDR] = tswapreg(env->CP0_BadVAddr);
1163     (*regs)[TARGET_EF_CP0_STATUS] = tswapreg(env->CP0_Status);
1164     (*regs)[TARGET_EF_CP0_CAUSE] = tswapreg(env->CP0_Cause);
1165 }
1166 
1167 #define USE_ELF_CORE_DUMP
1168 #define ELF_EXEC_PAGESIZE        4096
1169 
1170 /* See arch/mips/include/uapi/asm/hwcap.h.  */
1171 enum {
1172     HWCAP_MIPS_R6           = (1 << 0),
1173     HWCAP_MIPS_MSA          = (1 << 1),
1174     HWCAP_MIPS_CRC32        = (1 << 2),
1175     HWCAP_MIPS_MIPS16       = (1 << 3),
1176     HWCAP_MIPS_MDMX         = (1 << 4),
1177     HWCAP_MIPS_MIPS3D       = (1 << 5),
1178     HWCAP_MIPS_SMARTMIPS    = (1 << 6),
1179     HWCAP_MIPS_DSP          = (1 << 7),
1180     HWCAP_MIPS_DSP2         = (1 << 8),
1181     HWCAP_MIPS_DSP3         = (1 << 9),
1182     HWCAP_MIPS_MIPS16E2     = (1 << 10),
1183     HWCAP_LOONGSON_MMI      = (1 << 11),
1184     HWCAP_LOONGSON_EXT      = (1 << 12),
1185     HWCAP_LOONGSON_EXT2     = (1 << 13),
1186     HWCAP_LOONGSON_CPUCFG   = (1 << 14),
1187 };
1188 
1189 #define ELF_HWCAP get_elf_hwcap()
1190 
1191 #define GET_FEATURE_INSN(_flag, _hwcap) \
1192     do { if (cpu->env.insn_flags & (_flag)) { hwcaps |= _hwcap; } } while (0)
1193 
1194 #define GET_FEATURE_REG_SET(_reg, _mask, _hwcap) \
1195     do { if (cpu->env._reg & (_mask)) { hwcaps |= _hwcap; } } while (0)
1196 
1197 #define GET_FEATURE_REG_EQU(_reg, _start, _length, _val, _hwcap) \
1198     do { \
1199         if (extract32(cpu->env._reg, (_start), (_length)) == (_val)) { \
1200             hwcaps |= _hwcap; \
1201         } \
1202     } while (0)
1203 
1204 static uint32_t get_elf_hwcap(void)
1205 {
1206     MIPSCPU *cpu = MIPS_CPU(thread_cpu);
1207     uint32_t hwcaps = 0;
1208 
1209     GET_FEATURE_REG_EQU(CP0_Config0, CP0C0_AR, CP0C0_AR_LENGTH,
1210                         2, HWCAP_MIPS_R6);
1211     GET_FEATURE_REG_SET(CP0_Config3, 1 << CP0C3_MSAP, HWCAP_MIPS_MSA);
1212     GET_FEATURE_INSN(ASE_LMMI, HWCAP_LOONGSON_MMI);
1213     GET_FEATURE_INSN(ASE_LEXT, HWCAP_LOONGSON_EXT);
1214 
1215     return hwcaps;
1216 }
1217 
1218 #undef GET_FEATURE_REG_EQU
1219 #undef GET_FEATURE_REG_SET
1220 #undef GET_FEATURE_INSN
1221 
1222 #endif /* TARGET_MIPS */
1223 
1224 #ifdef TARGET_MICROBLAZE
1225 
1226 #define ELF_START_MMAP 0x80000000
1227 
1228 #define elf_check_arch(x) ( (x) == EM_MICROBLAZE || (x) == EM_MICROBLAZE_OLD)
1229 
1230 #define ELF_CLASS   ELFCLASS32
1231 #define ELF_ARCH    EM_MICROBLAZE
1232 
1233 static inline void init_thread(struct target_pt_regs *regs,
1234                                struct image_info *infop)
1235 {
1236     regs->pc = infop->entry;
1237     regs->r1 = infop->start_stack;
1238 
1239 }
1240 
1241 #define ELF_EXEC_PAGESIZE        4096
1242 
1243 #define USE_ELF_CORE_DUMP
1244 #define ELF_NREG 38
1245 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1246 
1247 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs.  */
1248 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMBState *env)
1249 {
1250     int i, pos = 0;
1251 
1252     for (i = 0; i < 32; i++) {
1253         (*regs)[pos++] = tswapreg(env->regs[i]);
1254     }
1255 
1256     (*regs)[pos++] = tswapreg(env->pc);
1257     (*regs)[pos++] = tswapreg(mb_cpu_read_msr(env));
1258     (*regs)[pos++] = 0;
1259     (*regs)[pos++] = tswapreg(env->ear);
1260     (*regs)[pos++] = 0;
1261     (*regs)[pos++] = tswapreg(env->esr);
1262 }
1263 
1264 #endif /* TARGET_MICROBLAZE */
1265 
1266 #ifdef TARGET_NIOS2
1267 
1268 #define ELF_START_MMAP 0x80000000
1269 
1270 #define elf_check_arch(x) ((x) == EM_ALTERA_NIOS2)
1271 
1272 #define ELF_CLASS   ELFCLASS32
1273 #define ELF_ARCH    EM_ALTERA_NIOS2
1274 
1275 static void init_thread(struct target_pt_regs *regs, struct image_info *infop)
1276 {
1277     regs->ea = infop->entry;
1278     regs->sp = infop->start_stack;
1279 }
1280 
1281 #define LO_COMMPAGE  TARGET_PAGE_SIZE
1282 
1283 static bool init_guest_commpage(void)
1284 {
1285     static const uint8_t kuser_page[4 + 2 * 64] = {
1286         /* __kuser_helper_version */
1287         [0x00] = 0x02, 0x00, 0x00, 0x00,
1288 
1289         /* __kuser_cmpxchg */
1290         [0x04] = 0x3a, 0x6c, 0x3b, 0x00,  /* trap 16 */
1291                  0x3a, 0x28, 0x00, 0xf8,  /* ret */
1292 
1293         /* __kuser_sigtramp */
1294         [0x44] = 0xc4, 0x22, 0x80, 0x00,  /* movi r2, __NR_rt_sigreturn */
1295                  0x3a, 0x68, 0x3b, 0x00,  /* trap 0 */
1296     };
1297 
1298     void *want = g2h_untagged(LO_COMMPAGE & -qemu_host_page_size);
1299     void *addr = mmap(want, qemu_host_page_size, PROT_READ | PROT_WRITE,
1300                       MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0);
1301 
1302     if (addr == MAP_FAILED) {
1303         perror("Allocating guest commpage");
1304         exit(EXIT_FAILURE);
1305     }
1306     if (addr != want) {
1307         return false;
1308     }
1309 
1310     memcpy(addr, kuser_page, sizeof(kuser_page));
1311 
1312     if (mprotect(addr, qemu_host_page_size, PROT_READ)) {
1313         perror("Protecting guest commpage");
1314         exit(EXIT_FAILURE);
1315     }
1316 
1317     page_set_flags(LO_COMMPAGE, LO_COMMPAGE + TARGET_PAGE_SIZE,
1318                    PAGE_READ | PAGE_EXEC | PAGE_VALID);
1319     return true;
1320 }
1321 
1322 #define ELF_EXEC_PAGESIZE        4096
1323 
1324 #define USE_ELF_CORE_DUMP
1325 #define ELF_NREG 49
1326 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1327 
1328 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs.  */
1329 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1330                                const CPUNios2State *env)
1331 {
1332     int i;
1333 
1334     (*regs)[0] = -1;
1335     for (i = 1; i < 8; i++)    /* r0-r7 */
1336         (*regs)[i] = tswapreg(env->regs[i + 7]);
1337 
1338     for (i = 8; i < 16; i++)   /* r8-r15 */
1339         (*regs)[i] = tswapreg(env->regs[i - 8]);
1340 
1341     for (i = 16; i < 24; i++)  /* r16-r23 */
1342         (*regs)[i] = tswapreg(env->regs[i + 7]);
1343     (*regs)[24] = -1;    /* R_ET */
1344     (*regs)[25] = -1;    /* R_BT */
1345     (*regs)[26] = tswapreg(env->regs[R_GP]);
1346     (*regs)[27] = tswapreg(env->regs[R_SP]);
1347     (*regs)[28] = tswapreg(env->regs[R_FP]);
1348     (*regs)[29] = tswapreg(env->regs[R_EA]);
1349     (*regs)[30] = -1;    /* R_SSTATUS */
1350     (*regs)[31] = tswapreg(env->regs[R_RA]);
1351 
1352     (*regs)[32] = tswapreg(env->pc);
1353 
1354     (*regs)[33] = -1; /* R_STATUS */
1355     (*regs)[34] = tswapreg(env->regs[CR_ESTATUS]);
1356 
1357     for (i = 35; i < 49; i++)    /* ... */
1358         (*regs)[i] = -1;
1359 }
1360 
1361 #endif /* TARGET_NIOS2 */
1362 
1363 #ifdef TARGET_OPENRISC
1364 
1365 #define ELF_START_MMAP 0x08000000
1366 
1367 #define ELF_ARCH EM_OPENRISC
1368 #define ELF_CLASS ELFCLASS32
1369 #define ELF_DATA  ELFDATA2MSB
1370 
1371 static inline void init_thread(struct target_pt_regs *regs,
1372                                struct image_info *infop)
1373 {
1374     regs->pc = infop->entry;
1375     regs->gpr[1] = infop->start_stack;
1376 }
1377 
1378 #define USE_ELF_CORE_DUMP
1379 #define ELF_EXEC_PAGESIZE 8192
1380 
1381 /* See linux kernel arch/openrisc/include/asm/elf.h.  */
1382 #define ELF_NREG 34 /* gprs and pc, sr */
1383 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1384 
1385 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1386                                const CPUOpenRISCState *env)
1387 {
1388     int i;
1389 
1390     for (i = 0; i < 32; i++) {
1391         (*regs)[i] = tswapreg(cpu_get_gpr(env, i));
1392     }
1393     (*regs)[32] = tswapreg(env->pc);
1394     (*regs)[33] = tswapreg(cpu_get_sr(env));
1395 }
1396 #define ELF_HWCAP 0
1397 #define ELF_PLATFORM NULL
1398 
1399 #endif /* TARGET_OPENRISC */
1400 
1401 #ifdef TARGET_SH4
1402 
1403 #define ELF_START_MMAP 0x80000000
1404 
1405 #define ELF_CLASS ELFCLASS32
1406 #define ELF_ARCH  EM_SH
1407 
1408 static inline void init_thread(struct target_pt_regs *regs,
1409                                struct image_info *infop)
1410 {
1411     /* Check other registers XXXXX */
1412     regs->pc = infop->entry;
1413     regs->regs[15] = infop->start_stack;
1414 }
1415 
1416 /* See linux kernel: arch/sh/include/asm/elf.h.  */
1417 #define ELF_NREG 23
1418 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1419 
1420 /* See linux kernel: arch/sh/include/asm/ptrace.h.  */
1421 enum {
1422     TARGET_REG_PC = 16,
1423     TARGET_REG_PR = 17,
1424     TARGET_REG_SR = 18,
1425     TARGET_REG_GBR = 19,
1426     TARGET_REG_MACH = 20,
1427     TARGET_REG_MACL = 21,
1428     TARGET_REG_SYSCALL = 22
1429 };
1430 
1431 static inline void elf_core_copy_regs(target_elf_gregset_t *regs,
1432                                       const CPUSH4State *env)
1433 {
1434     int i;
1435 
1436     for (i = 0; i < 16; i++) {
1437         (*regs)[i] = tswapreg(env->gregs[i]);
1438     }
1439 
1440     (*regs)[TARGET_REG_PC] = tswapreg(env->pc);
1441     (*regs)[TARGET_REG_PR] = tswapreg(env->pr);
1442     (*regs)[TARGET_REG_SR] = tswapreg(env->sr);
1443     (*regs)[TARGET_REG_GBR] = tswapreg(env->gbr);
1444     (*regs)[TARGET_REG_MACH] = tswapreg(env->mach);
1445     (*regs)[TARGET_REG_MACL] = tswapreg(env->macl);
1446     (*regs)[TARGET_REG_SYSCALL] = 0; /* FIXME */
1447 }
1448 
1449 #define USE_ELF_CORE_DUMP
1450 #define ELF_EXEC_PAGESIZE        4096
1451 
1452 enum {
1453     SH_CPU_HAS_FPU            = 0x0001, /* Hardware FPU support */
1454     SH_CPU_HAS_P2_FLUSH_BUG   = 0x0002, /* Need to flush the cache in P2 area */
1455     SH_CPU_HAS_MMU_PAGE_ASSOC = 0x0004, /* SH3: TLB way selection bit support */
1456     SH_CPU_HAS_DSP            = 0x0008, /* SH-DSP: DSP support */
1457     SH_CPU_HAS_PERF_COUNTER   = 0x0010, /* Hardware performance counters */
1458     SH_CPU_HAS_PTEA           = 0x0020, /* PTEA register */
1459     SH_CPU_HAS_LLSC           = 0x0040, /* movli.l/movco.l */
1460     SH_CPU_HAS_L2_CACHE       = 0x0080, /* Secondary cache / URAM */
1461     SH_CPU_HAS_OP32           = 0x0100, /* 32-bit instruction support */
1462     SH_CPU_HAS_PTEAEX         = 0x0200, /* PTE ASID Extension support */
1463 };
1464 
1465 #define ELF_HWCAP get_elf_hwcap()
1466 
1467 static uint32_t get_elf_hwcap(void)
1468 {
1469     SuperHCPU *cpu = SUPERH_CPU(thread_cpu);
1470     uint32_t hwcap = 0;
1471 
1472     hwcap |= SH_CPU_HAS_FPU;
1473 
1474     if (cpu->env.features & SH_FEATURE_SH4A) {
1475         hwcap |= SH_CPU_HAS_LLSC;
1476     }
1477 
1478     return hwcap;
1479 }
1480 
1481 #endif
1482 
1483 #ifdef TARGET_CRIS
1484 
1485 #define ELF_START_MMAP 0x80000000
1486 
1487 #define ELF_CLASS ELFCLASS32
1488 #define ELF_ARCH  EM_CRIS
1489 
1490 static inline void init_thread(struct target_pt_regs *regs,
1491                                struct image_info *infop)
1492 {
1493     regs->erp = infop->entry;
1494 }
1495 
1496 #define ELF_EXEC_PAGESIZE        8192
1497 
1498 #endif
1499 
1500 #ifdef TARGET_M68K
1501 
1502 #define ELF_START_MMAP 0x80000000
1503 
1504 #define ELF_CLASS       ELFCLASS32
1505 #define ELF_ARCH        EM_68K
1506 
1507 /* ??? Does this need to do anything?
1508    #define ELF_PLAT_INIT(_r) */
1509 
1510 static inline void init_thread(struct target_pt_regs *regs,
1511                                struct image_info *infop)
1512 {
1513     regs->usp = infop->start_stack;
1514     regs->sr = 0;
1515     regs->pc = infop->entry;
1516 }
1517 
1518 /* See linux kernel: arch/m68k/include/asm/elf.h.  */
1519 #define ELF_NREG 20
1520 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1521 
1522 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUM68KState *env)
1523 {
1524     (*regs)[0] = tswapreg(env->dregs[1]);
1525     (*regs)[1] = tswapreg(env->dregs[2]);
1526     (*regs)[2] = tswapreg(env->dregs[3]);
1527     (*regs)[3] = tswapreg(env->dregs[4]);
1528     (*regs)[4] = tswapreg(env->dregs[5]);
1529     (*regs)[5] = tswapreg(env->dregs[6]);
1530     (*regs)[6] = tswapreg(env->dregs[7]);
1531     (*regs)[7] = tswapreg(env->aregs[0]);
1532     (*regs)[8] = tswapreg(env->aregs[1]);
1533     (*regs)[9] = tswapreg(env->aregs[2]);
1534     (*regs)[10] = tswapreg(env->aregs[3]);
1535     (*regs)[11] = tswapreg(env->aregs[4]);
1536     (*regs)[12] = tswapreg(env->aregs[5]);
1537     (*regs)[13] = tswapreg(env->aregs[6]);
1538     (*regs)[14] = tswapreg(env->dregs[0]);
1539     (*regs)[15] = tswapreg(env->aregs[7]);
1540     (*regs)[16] = tswapreg(env->dregs[0]); /* FIXME: orig_d0 */
1541     (*regs)[17] = tswapreg(env->sr);
1542     (*regs)[18] = tswapreg(env->pc);
1543     (*regs)[19] = 0;  /* FIXME: regs->format | regs->vector */
1544 }
1545 
1546 #define USE_ELF_CORE_DUMP
1547 #define ELF_EXEC_PAGESIZE       8192
1548 
1549 #endif
1550 
1551 #ifdef TARGET_ALPHA
1552 
1553 #define ELF_START_MMAP (0x30000000000ULL)
1554 
1555 #define ELF_CLASS      ELFCLASS64
1556 #define ELF_ARCH       EM_ALPHA
1557 
1558 static inline void init_thread(struct target_pt_regs *regs,
1559                                struct image_info *infop)
1560 {
1561     regs->pc = infop->entry;
1562     regs->ps = 8;
1563     regs->usp = infop->start_stack;
1564 }
1565 
1566 #define ELF_EXEC_PAGESIZE        8192
1567 
1568 #endif /* TARGET_ALPHA */
1569 
1570 #ifdef TARGET_S390X
1571 
1572 #define ELF_START_MMAP (0x20000000000ULL)
1573 
1574 #define ELF_CLASS	ELFCLASS64
1575 #define ELF_DATA	ELFDATA2MSB
1576 #define ELF_ARCH	EM_S390
1577 
1578 #include "elf.h"
1579 
1580 #define ELF_HWCAP get_elf_hwcap()
1581 
1582 #define GET_FEATURE(_feat, _hwcap) \
1583     do { if (s390_has_feat(_feat)) { hwcap |= _hwcap; } } while (0)
1584 
1585 static uint32_t get_elf_hwcap(void)
1586 {
1587     /*
1588      * Let's assume we always have esan3 and zarch.
1589      * 31-bit processes can use 64-bit registers (high gprs).
1590      */
1591     uint32_t hwcap = HWCAP_S390_ESAN3 | HWCAP_S390_ZARCH | HWCAP_S390_HIGH_GPRS;
1592 
1593     GET_FEATURE(S390_FEAT_STFLE, HWCAP_S390_STFLE);
1594     GET_FEATURE(S390_FEAT_MSA, HWCAP_S390_MSA);
1595     GET_FEATURE(S390_FEAT_LONG_DISPLACEMENT, HWCAP_S390_LDISP);
1596     GET_FEATURE(S390_FEAT_EXTENDED_IMMEDIATE, HWCAP_S390_EIMM);
1597     if (s390_has_feat(S390_FEAT_EXTENDED_TRANSLATION_3) &&
1598         s390_has_feat(S390_FEAT_ETF3_ENH)) {
1599         hwcap |= HWCAP_S390_ETF3EH;
1600     }
1601     GET_FEATURE(S390_FEAT_VECTOR, HWCAP_S390_VXRS);
1602     GET_FEATURE(S390_FEAT_VECTOR_ENH, HWCAP_S390_VXRS_EXT);
1603 
1604     return hwcap;
1605 }
1606 
1607 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop)
1608 {
1609     regs->psw.addr = infop->entry;
1610     regs->psw.mask = PSW_MASK_64 | PSW_MASK_32;
1611     regs->gprs[15] = infop->start_stack;
1612 }
1613 
1614 /* See linux kernel: arch/s390/include/uapi/asm/ptrace.h (s390_regs).  */
1615 #define ELF_NREG 27
1616 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1617 
1618 enum {
1619     TARGET_REG_PSWM = 0,
1620     TARGET_REG_PSWA = 1,
1621     TARGET_REG_GPRS = 2,
1622     TARGET_REG_ARS = 18,
1623     TARGET_REG_ORIG_R2 = 26,
1624 };
1625 
1626 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1627                                const CPUS390XState *env)
1628 {
1629     int i;
1630     uint32_t *aregs;
1631 
1632     (*regs)[TARGET_REG_PSWM] = tswapreg(env->psw.mask);
1633     (*regs)[TARGET_REG_PSWA] = tswapreg(env->psw.addr);
1634     for (i = 0; i < 16; i++) {
1635         (*regs)[TARGET_REG_GPRS + i] = tswapreg(env->regs[i]);
1636     }
1637     aregs = (uint32_t *)&((*regs)[TARGET_REG_ARS]);
1638     for (i = 0; i < 16; i++) {
1639         aregs[i] = tswap32(env->aregs[i]);
1640     }
1641     (*regs)[TARGET_REG_ORIG_R2] = 0;
1642 }
1643 
1644 #define USE_ELF_CORE_DUMP
1645 #define ELF_EXEC_PAGESIZE 4096
1646 
1647 #endif /* TARGET_S390X */
1648 
1649 #ifdef TARGET_RISCV
1650 
1651 #define ELF_START_MMAP 0x80000000
1652 #define ELF_ARCH  EM_RISCV
1653 
1654 #ifdef TARGET_RISCV32
1655 #define ELF_CLASS ELFCLASS32
1656 #else
1657 #define ELF_CLASS ELFCLASS64
1658 #endif
1659 
1660 #define ELF_HWCAP get_elf_hwcap()
1661 
1662 static uint32_t get_elf_hwcap(void)
1663 {
1664 #define MISA_BIT(EXT) (1 << (EXT - 'A'))
1665     RISCVCPU *cpu = RISCV_CPU(thread_cpu);
1666     uint32_t mask = MISA_BIT('I') | MISA_BIT('M') | MISA_BIT('A')
1667                     | MISA_BIT('F') | MISA_BIT('D') | MISA_BIT('C');
1668 
1669     return cpu->env.misa_ext & mask;
1670 #undef MISA_BIT
1671 }
1672 
1673 static inline void init_thread(struct target_pt_regs *regs,
1674                                struct image_info *infop)
1675 {
1676     regs->sepc = infop->entry;
1677     regs->sp = infop->start_stack;
1678 }
1679 
1680 #define ELF_EXEC_PAGESIZE 4096
1681 
1682 #endif /* TARGET_RISCV */
1683 
1684 #ifdef TARGET_HPPA
1685 
1686 #define ELF_START_MMAP  0x80000000
1687 #define ELF_CLASS       ELFCLASS32
1688 #define ELF_ARCH        EM_PARISC
1689 #define ELF_PLATFORM    "PARISC"
1690 #define STACK_GROWS_DOWN 0
1691 #define STACK_ALIGNMENT  64
1692 
1693 static inline void init_thread(struct target_pt_regs *regs,
1694                                struct image_info *infop)
1695 {
1696     regs->iaoq[0] = infop->entry;
1697     regs->iaoq[1] = infop->entry + 4;
1698     regs->gr[23] = 0;
1699     regs->gr[24] = infop->argv;
1700     regs->gr[25] = infop->argc;
1701     /* The top-of-stack contains a linkage buffer.  */
1702     regs->gr[30] = infop->start_stack + 64;
1703     regs->gr[31] = infop->entry;
1704 }
1705 
1706 #define LO_COMMPAGE  0
1707 
1708 static bool init_guest_commpage(void)
1709 {
1710     void *want = g2h_untagged(LO_COMMPAGE);
1711     void *addr = mmap(want, qemu_host_page_size, PROT_NONE,
1712                       MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0);
1713 
1714     if (addr == MAP_FAILED) {
1715         perror("Allocating guest commpage");
1716         exit(EXIT_FAILURE);
1717     }
1718     if (addr != want) {
1719         return false;
1720     }
1721 
1722     /*
1723      * On Linux, page zero is normally marked execute only + gateway.
1724      * Normal read or write is supposed to fail (thus PROT_NONE above),
1725      * but specific offsets have kernel code mapped to raise permissions
1726      * and implement syscalls.  Here, simply mark the page executable.
1727      * Special case the entry points during translation (see do_page_zero).
1728      */
1729     page_set_flags(LO_COMMPAGE, LO_COMMPAGE + TARGET_PAGE_SIZE,
1730                    PAGE_EXEC | PAGE_VALID);
1731     return true;
1732 }
1733 
1734 #endif /* TARGET_HPPA */
1735 
1736 #ifdef TARGET_XTENSA
1737 
1738 #define ELF_START_MMAP 0x20000000
1739 
1740 #define ELF_CLASS       ELFCLASS32
1741 #define ELF_ARCH        EM_XTENSA
1742 
1743 static inline void init_thread(struct target_pt_regs *regs,
1744                                struct image_info *infop)
1745 {
1746     regs->windowbase = 0;
1747     regs->windowstart = 1;
1748     regs->areg[1] = infop->start_stack;
1749     regs->pc = infop->entry;
1750 }
1751 
1752 /* See linux kernel: arch/xtensa/include/asm/elf.h.  */
1753 #define ELF_NREG 128
1754 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1755 
1756 enum {
1757     TARGET_REG_PC,
1758     TARGET_REG_PS,
1759     TARGET_REG_LBEG,
1760     TARGET_REG_LEND,
1761     TARGET_REG_LCOUNT,
1762     TARGET_REG_SAR,
1763     TARGET_REG_WINDOWSTART,
1764     TARGET_REG_WINDOWBASE,
1765     TARGET_REG_THREADPTR,
1766     TARGET_REG_AR0 = 64,
1767 };
1768 
1769 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1770                                const CPUXtensaState *env)
1771 {
1772     unsigned i;
1773 
1774     (*regs)[TARGET_REG_PC] = tswapreg(env->pc);
1775     (*regs)[TARGET_REG_PS] = tswapreg(env->sregs[PS] & ~PS_EXCM);
1776     (*regs)[TARGET_REG_LBEG] = tswapreg(env->sregs[LBEG]);
1777     (*regs)[TARGET_REG_LEND] = tswapreg(env->sregs[LEND]);
1778     (*regs)[TARGET_REG_LCOUNT] = tswapreg(env->sregs[LCOUNT]);
1779     (*regs)[TARGET_REG_SAR] = tswapreg(env->sregs[SAR]);
1780     (*regs)[TARGET_REG_WINDOWSTART] = tswapreg(env->sregs[WINDOW_START]);
1781     (*regs)[TARGET_REG_WINDOWBASE] = tswapreg(env->sregs[WINDOW_BASE]);
1782     (*regs)[TARGET_REG_THREADPTR] = tswapreg(env->uregs[THREADPTR]);
1783     xtensa_sync_phys_from_window((CPUXtensaState *)env);
1784     for (i = 0; i < env->config->nareg; ++i) {
1785         (*regs)[TARGET_REG_AR0 + i] = tswapreg(env->phys_regs[i]);
1786     }
1787 }
1788 
1789 #define USE_ELF_CORE_DUMP
1790 #define ELF_EXEC_PAGESIZE       4096
1791 
1792 #endif /* TARGET_XTENSA */
1793 
1794 #ifdef TARGET_HEXAGON
1795 
1796 #define ELF_START_MMAP 0x20000000
1797 
1798 #define ELF_CLASS       ELFCLASS32
1799 #define ELF_ARCH        EM_HEXAGON
1800 
1801 static inline void init_thread(struct target_pt_regs *regs,
1802                                struct image_info *infop)
1803 {
1804     regs->sepc = infop->entry;
1805     regs->sp = infop->start_stack;
1806 }
1807 
1808 #endif /* TARGET_HEXAGON */
1809 
1810 #ifndef ELF_BASE_PLATFORM
1811 #define ELF_BASE_PLATFORM (NULL)
1812 #endif
1813 
1814 #ifndef ELF_PLATFORM
1815 #define ELF_PLATFORM (NULL)
1816 #endif
1817 
1818 #ifndef ELF_MACHINE
1819 #define ELF_MACHINE ELF_ARCH
1820 #endif
1821 
1822 #ifndef elf_check_arch
1823 #define elf_check_arch(x) ((x) == ELF_ARCH)
1824 #endif
1825 
1826 #ifndef elf_check_abi
1827 #define elf_check_abi(x) (1)
1828 #endif
1829 
1830 #ifndef ELF_HWCAP
1831 #define ELF_HWCAP 0
1832 #endif
1833 
1834 #ifndef STACK_GROWS_DOWN
1835 #define STACK_GROWS_DOWN 1
1836 #endif
1837 
1838 #ifndef STACK_ALIGNMENT
1839 #define STACK_ALIGNMENT 16
1840 #endif
1841 
1842 #ifdef TARGET_ABI32
1843 #undef ELF_CLASS
1844 #define ELF_CLASS ELFCLASS32
1845 #undef bswaptls
1846 #define bswaptls(ptr) bswap32s(ptr)
1847 #endif
1848 
1849 #ifndef EXSTACK_DEFAULT
1850 #define EXSTACK_DEFAULT false
1851 #endif
1852 
1853 #include "elf.h"
1854 
1855 /* We must delay the following stanzas until after "elf.h". */
1856 #if defined(TARGET_AARCH64)
1857 
1858 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz,
1859                                     const uint32_t *data,
1860                                     struct image_info *info,
1861                                     Error **errp)
1862 {
1863     if (pr_type == GNU_PROPERTY_AARCH64_FEATURE_1_AND) {
1864         if (pr_datasz != sizeof(uint32_t)) {
1865             error_setg(errp, "Ill-formed GNU_PROPERTY_AARCH64_FEATURE_1_AND");
1866             return false;
1867         }
1868         /* We will extract GNU_PROPERTY_AARCH64_FEATURE_1_BTI later. */
1869         info->note_flags = *data;
1870     }
1871     return true;
1872 }
1873 #define ARCH_USE_GNU_PROPERTY 1
1874 
1875 #else
1876 
1877 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz,
1878                                     const uint32_t *data,
1879                                     struct image_info *info,
1880                                     Error **errp)
1881 {
1882     g_assert_not_reached();
1883 }
1884 #define ARCH_USE_GNU_PROPERTY 0
1885 
1886 #endif
1887 
1888 struct exec
1889 {
1890     unsigned int a_info;   /* Use macros N_MAGIC, etc for access */
1891     unsigned int a_text;   /* length of text, in bytes */
1892     unsigned int a_data;   /* length of data, in bytes */
1893     unsigned int a_bss;    /* length of uninitialized data area, in bytes */
1894     unsigned int a_syms;   /* length of symbol table data in file, in bytes */
1895     unsigned int a_entry;  /* start address */
1896     unsigned int a_trsize; /* length of relocation info for text, in bytes */
1897     unsigned int a_drsize; /* length of relocation info for data, in bytes */
1898 };
1899 
1900 
1901 #define N_MAGIC(exec) ((exec).a_info & 0xffff)
1902 #define OMAGIC 0407
1903 #define NMAGIC 0410
1904 #define ZMAGIC 0413
1905 #define QMAGIC 0314
1906 
1907 /* Necessary parameters */
1908 #define TARGET_ELF_EXEC_PAGESIZE \
1909         (((eppnt->p_align & ~qemu_host_page_mask) != 0) ? \
1910          TARGET_PAGE_SIZE : MAX(qemu_host_page_size, TARGET_PAGE_SIZE))
1911 #define TARGET_ELF_PAGELENGTH(_v) ROUND_UP((_v), TARGET_ELF_EXEC_PAGESIZE)
1912 #define TARGET_ELF_PAGESTART(_v) ((_v) & \
1913                                  ~(abi_ulong)(TARGET_ELF_EXEC_PAGESIZE-1))
1914 #define TARGET_ELF_PAGEOFFSET(_v) ((_v) & (TARGET_ELF_EXEC_PAGESIZE-1))
1915 
1916 #define DLINFO_ITEMS 16
1917 
1918 static inline void memcpy_fromfs(void * to, const void * from, unsigned long n)
1919 {
1920     memcpy(to, from, n);
1921 }
1922 
1923 #ifdef BSWAP_NEEDED
1924 static void bswap_ehdr(struct elfhdr *ehdr)
1925 {
1926     bswap16s(&ehdr->e_type);            /* Object file type */
1927     bswap16s(&ehdr->e_machine);         /* Architecture */
1928     bswap32s(&ehdr->e_version);         /* Object file version */
1929     bswaptls(&ehdr->e_entry);           /* Entry point virtual address */
1930     bswaptls(&ehdr->e_phoff);           /* Program header table file offset */
1931     bswaptls(&ehdr->e_shoff);           /* Section header table file offset */
1932     bswap32s(&ehdr->e_flags);           /* Processor-specific flags */
1933     bswap16s(&ehdr->e_ehsize);          /* ELF header size in bytes */
1934     bswap16s(&ehdr->e_phentsize);       /* Program header table entry size */
1935     bswap16s(&ehdr->e_phnum);           /* Program header table entry count */
1936     bswap16s(&ehdr->e_shentsize);       /* Section header table entry size */
1937     bswap16s(&ehdr->e_shnum);           /* Section header table entry count */
1938     bswap16s(&ehdr->e_shstrndx);        /* Section header string table index */
1939 }
1940 
1941 static void bswap_phdr(struct elf_phdr *phdr, int phnum)
1942 {
1943     int i;
1944     for (i = 0; i < phnum; ++i, ++phdr) {
1945         bswap32s(&phdr->p_type);        /* Segment type */
1946         bswap32s(&phdr->p_flags);       /* Segment flags */
1947         bswaptls(&phdr->p_offset);      /* Segment file offset */
1948         bswaptls(&phdr->p_vaddr);       /* Segment virtual address */
1949         bswaptls(&phdr->p_paddr);       /* Segment physical address */
1950         bswaptls(&phdr->p_filesz);      /* Segment size in file */
1951         bswaptls(&phdr->p_memsz);       /* Segment size in memory */
1952         bswaptls(&phdr->p_align);       /* Segment alignment */
1953     }
1954 }
1955 
1956 static void bswap_shdr(struct elf_shdr *shdr, int shnum)
1957 {
1958     int i;
1959     for (i = 0; i < shnum; ++i, ++shdr) {
1960         bswap32s(&shdr->sh_name);
1961         bswap32s(&shdr->sh_type);
1962         bswaptls(&shdr->sh_flags);
1963         bswaptls(&shdr->sh_addr);
1964         bswaptls(&shdr->sh_offset);
1965         bswaptls(&shdr->sh_size);
1966         bswap32s(&shdr->sh_link);
1967         bswap32s(&shdr->sh_info);
1968         bswaptls(&shdr->sh_addralign);
1969         bswaptls(&shdr->sh_entsize);
1970     }
1971 }
1972 
1973 static void bswap_sym(struct elf_sym *sym)
1974 {
1975     bswap32s(&sym->st_name);
1976     bswaptls(&sym->st_value);
1977     bswaptls(&sym->st_size);
1978     bswap16s(&sym->st_shndx);
1979 }
1980 
1981 #ifdef TARGET_MIPS
1982 static void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags)
1983 {
1984     bswap16s(&abiflags->version);
1985     bswap32s(&abiflags->ases);
1986     bswap32s(&abiflags->isa_ext);
1987     bswap32s(&abiflags->flags1);
1988     bswap32s(&abiflags->flags2);
1989 }
1990 #endif
1991 #else
1992 static inline void bswap_ehdr(struct elfhdr *ehdr) { }
1993 static inline void bswap_phdr(struct elf_phdr *phdr, int phnum) { }
1994 static inline void bswap_shdr(struct elf_shdr *shdr, int shnum) { }
1995 static inline void bswap_sym(struct elf_sym *sym) { }
1996 #ifdef TARGET_MIPS
1997 static inline void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags) { }
1998 #endif
1999 #endif
2000 
2001 #ifdef USE_ELF_CORE_DUMP
2002 static int elf_core_dump(int, const CPUArchState *);
2003 #endif /* USE_ELF_CORE_DUMP */
2004 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias);
2005 
2006 /* Verify the portions of EHDR within E_IDENT for the target.
2007    This can be performed before bswapping the entire header.  */
2008 static bool elf_check_ident(struct elfhdr *ehdr)
2009 {
2010     return (ehdr->e_ident[EI_MAG0] == ELFMAG0
2011             && ehdr->e_ident[EI_MAG1] == ELFMAG1
2012             && ehdr->e_ident[EI_MAG2] == ELFMAG2
2013             && ehdr->e_ident[EI_MAG3] == ELFMAG3
2014             && ehdr->e_ident[EI_CLASS] == ELF_CLASS
2015             && ehdr->e_ident[EI_DATA] == ELF_DATA
2016             && ehdr->e_ident[EI_VERSION] == EV_CURRENT);
2017 }
2018 
2019 /* Verify the portions of EHDR outside of E_IDENT for the target.
2020    This has to wait until after bswapping the header.  */
2021 static bool elf_check_ehdr(struct elfhdr *ehdr)
2022 {
2023     return (elf_check_arch(ehdr->e_machine)
2024             && elf_check_abi(ehdr->e_flags)
2025             && ehdr->e_ehsize == sizeof(struct elfhdr)
2026             && ehdr->e_phentsize == sizeof(struct elf_phdr)
2027             && (ehdr->e_type == ET_EXEC || ehdr->e_type == ET_DYN));
2028 }
2029 
2030 /*
2031  * 'copy_elf_strings()' copies argument/envelope strings from user
2032  * memory to free pages in kernel mem. These are in a format ready
2033  * to be put directly into the top of new user memory.
2034  *
2035  */
2036 static abi_ulong copy_elf_strings(int argc, char **argv, char *scratch,
2037                                   abi_ulong p, abi_ulong stack_limit)
2038 {
2039     char *tmp;
2040     int len, i;
2041     abi_ulong top = p;
2042 
2043     if (!p) {
2044         return 0;       /* bullet-proofing */
2045     }
2046 
2047     if (STACK_GROWS_DOWN) {
2048         int offset = ((p - 1) % TARGET_PAGE_SIZE) + 1;
2049         for (i = argc - 1; i >= 0; --i) {
2050             tmp = argv[i];
2051             if (!tmp) {
2052                 fprintf(stderr, "VFS: argc is wrong");
2053                 exit(-1);
2054             }
2055             len = strlen(tmp) + 1;
2056             tmp += len;
2057 
2058             if (len > (p - stack_limit)) {
2059                 return 0;
2060             }
2061             while (len) {
2062                 int bytes_to_copy = (len > offset) ? offset : len;
2063                 tmp -= bytes_to_copy;
2064                 p -= bytes_to_copy;
2065                 offset -= bytes_to_copy;
2066                 len -= bytes_to_copy;
2067 
2068                 memcpy_fromfs(scratch + offset, tmp, bytes_to_copy);
2069 
2070                 if (offset == 0) {
2071                     memcpy_to_target(p, scratch, top - p);
2072                     top = p;
2073                     offset = TARGET_PAGE_SIZE;
2074                 }
2075             }
2076         }
2077         if (p != top) {
2078             memcpy_to_target(p, scratch + offset, top - p);
2079         }
2080     } else {
2081         int remaining = TARGET_PAGE_SIZE - (p % TARGET_PAGE_SIZE);
2082         for (i = 0; i < argc; ++i) {
2083             tmp = argv[i];
2084             if (!tmp) {
2085                 fprintf(stderr, "VFS: argc is wrong");
2086                 exit(-1);
2087             }
2088             len = strlen(tmp) + 1;
2089             if (len > (stack_limit - p)) {
2090                 return 0;
2091             }
2092             while (len) {
2093                 int bytes_to_copy = (len > remaining) ? remaining : len;
2094 
2095                 memcpy_fromfs(scratch + (p - top), tmp, bytes_to_copy);
2096 
2097                 tmp += bytes_to_copy;
2098                 remaining -= bytes_to_copy;
2099                 p += bytes_to_copy;
2100                 len -= bytes_to_copy;
2101 
2102                 if (remaining == 0) {
2103                     memcpy_to_target(top, scratch, p - top);
2104                     top = p;
2105                     remaining = TARGET_PAGE_SIZE;
2106                 }
2107             }
2108         }
2109         if (p != top) {
2110             memcpy_to_target(top, scratch, p - top);
2111         }
2112     }
2113 
2114     return p;
2115 }
2116 
2117 /* Older linux kernels provide up to MAX_ARG_PAGES (default: 32) of
2118  * argument/environment space. Newer kernels (>2.6.33) allow more,
2119  * dependent on stack size, but guarantee at least 32 pages for
2120  * backwards compatibility.
2121  */
2122 #define STACK_LOWER_LIMIT (32 * TARGET_PAGE_SIZE)
2123 
2124 static abi_ulong setup_arg_pages(struct linux_binprm *bprm,
2125                                  struct image_info *info)
2126 {
2127     abi_ulong size, error, guard;
2128     int prot;
2129 
2130     size = guest_stack_size;
2131     if (size < STACK_LOWER_LIMIT) {
2132         size = STACK_LOWER_LIMIT;
2133     }
2134 
2135     if (STACK_GROWS_DOWN) {
2136         guard = TARGET_PAGE_SIZE;
2137         if (guard < qemu_real_host_page_size()) {
2138             guard = qemu_real_host_page_size();
2139         }
2140     } else {
2141         /* no guard page for hppa target where stack grows upwards. */
2142         guard = 0;
2143     }
2144 
2145     prot = PROT_READ | PROT_WRITE;
2146     if (info->exec_stack) {
2147         prot |= PROT_EXEC;
2148     }
2149     error = target_mmap(0, size + guard, prot,
2150                         MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
2151     if (error == -1) {
2152         perror("mmap stack");
2153         exit(-1);
2154     }
2155 
2156     /* We reserve one extra page at the top of the stack as guard.  */
2157     if (STACK_GROWS_DOWN) {
2158         target_mprotect(error, guard, PROT_NONE);
2159         info->stack_limit = error + guard;
2160         return info->stack_limit + size - sizeof(void *);
2161     } else {
2162         info->stack_limit = error + size;
2163         return error;
2164     }
2165 }
2166 
2167 /* Map and zero the bss.  We need to explicitly zero any fractional pages
2168    after the data section (i.e. bss).  */
2169 static void zero_bss(abi_ulong elf_bss, abi_ulong last_bss, int prot)
2170 {
2171     uintptr_t host_start, host_map_start, host_end;
2172 
2173     last_bss = TARGET_PAGE_ALIGN(last_bss);
2174 
2175     /* ??? There is confusion between qemu_real_host_page_size and
2176        qemu_host_page_size here and elsewhere in target_mmap, which
2177        may lead to the end of the data section mapping from the file
2178        not being mapped.  At least there was an explicit test and
2179        comment for that here, suggesting that "the file size must
2180        be known".  The comment probably pre-dates the introduction
2181        of the fstat system call in target_mmap which does in fact
2182        find out the size.  What isn't clear is if the workaround
2183        here is still actually needed.  For now, continue with it,
2184        but merge it with the "normal" mmap that would allocate the bss.  */
2185 
2186     host_start = (uintptr_t) g2h_untagged(elf_bss);
2187     host_end = (uintptr_t) g2h_untagged(last_bss);
2188     host_map_start = REAL_HOST_PAGE_ALIGN(host_start);
2189 
2190     if (host_map_start < host_end) {
2191         void *p = mmap((void *)host_map_start, host_end - host_map_start,
2192                        prot, MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
2193         if (p == MAP_FAILED) {
2194             perror("cannot mmap brk");
2195             exit(-1);
2196         }
2197     }
2198 
2199     /* Ensure that the bss page(s) are valid */
2200     if ((page_get_flags(last_bss-1) & prot) != prot) {
2201         page_set_flags(elf_bss & TARGET_PAGE_MASK, last_bss, prot | PAGE_VALID);
2202     }
2203 
2204     if (host_start < host_map_start) {
2205         memset((void *)host_start, 0, host_map_start - host_start);
2206     }
2207 }
2208 
2209 #ifdef TARGET_ARM
2210 static int elf_is_fdpic(struct elfhdr *exec)
2211 {
2212     return exec->e_ident[EI_OSABI] == ELFOSABI_ARM_FDPIC;
2213 }
2214 #else
2215 /* Default implementation, always false.  */
2216 static int elf_is_fdpic(struct elfhdr *exec)
2217 {
2218     return 0;
2219 }
2220 #endif
2221 
2222 static abi_ulong loader_build_fdpic_loadmap(struct image_info *info, abi_ulong sp)
2223 {
2224     uint16_t n;
2225     struct elf32_fdpic_loadseg *loadsegs = info->loadsegs;
2226 
2227     /* elf32_fdpic_loadseg */
2228     n = info->nsegs;
2229     while (n--) {
2230         sp -= 12;
2231         put_user_u32(loadsegs[n].addr, sp+0);
2232         put_user_u32(loadsegs[n].p_vaddr, sp+4);
2233         put_user_u32(loadsegs[n].p_memsz, sp+8);
2234     }
2235 
2236     /* elf32_fdpic_loadmap */
2237     sp -= 4;
2238     put_user_u16(0, sp+0); /* version */
2239     put_user_u16(info->nsegs, sp+2); /* nsegs */
2240 
2241     info->personality = PER_LINUX_FDPIC;
2242     info->loadmap_addr = sp;
2243 
2244     return sp;
2245 }
2246 
2247 static abi_ulong create_elf_tables(abi_ulong p, int argc, int envc,
2248                                    struct elfhdr *exec,
2249                                    struct image_info *info,
2250                                    struct image_info *interp_info)
2251 {
2252     abi_ulong sp;
2253     abi_ulong u_argc, u_argv, u_envp, u_auxv;
2254     int size;
2255     int i;
2256     abi_ulong u_rand_bytes;
2257     uint8_t k_rand_bytes[16];
2258     abi_ulong u_platform, u_base_platform;
2259     const char *k_platform, *k_base_platform;
2260     const int n = sizeof(elf_addr_t);
2261 
2262     sp = p;
2263 
2264     /* Needs to be before we load the env/argc/... */
2265     if (elf_is_fdpic(exec)) {
2266         /* Need 4 byte alignment for these structs */
2267         sp &= ~3;
2268         sp = loader_build_fdpic_loadmap(info, sp);
2269         info->other_info = interp_info;
2270         if (interp_info) {
2271             interp_info->other_info = info;
2272             sp = loader_build_fdpic_loadmap(interp_info, sp);
2273             info->interpreter_loadmap_addr = interp_info->loadmap_addr;
2274             info->interpreter_pt_dynamic_addr = interp_info->pt_dynamic_addr;
2275         } else {
2276             info->interpreter_loadmap_addr = 0;
2277             info->interpreter_pt_dynamic_addr = 0;
2278         }
2279     }
2280 
2281     u_base_platform = 0;
2282     k_base_platform = ELF_BASE_PLATFORM;
2283     if (k_base_platform) {
2284         size_t len = strlen(k_base_platform) + 1;
2285         if (STACK_GROWS_DOWN) {
2286             sp -= (len + n - 1) & ~(n - 1);
2287             u_base_platform = sp;
2288             /* FIXME - check return value of memcpy_to_target() for failure */
2289             memcpy_to_target(sp, k_base_platform, len);
2290         } else {
2291             memcpy_to_target(sp, k_base_platform, len);
2292             u_base_platform = sp;
2293             sp += len + 1;
2294         }
2295     }
2296 
2297     u_platform = 0;
2298     k_platform = ELF_PLATFORM;
2299     if (k_platform) {
2300         size_t len = strlen(k_platform) + 1;
2301         if (STACK_GROWS_DOWN) {
2302             sp -= (len + n - 1) & ~(n - 1);
2303             u_platform = sp;
2304             /* FIXME - check return value of memcpy_to_target() for failure */
2305             memcpy_to_target(sp, k_platform, len);
2306         } else {
2307             memcpy_to_target(sp, k_platform, len);
2308             u_platform = sp;
2309             sp += len + 1;
2310         }
2311     }
2312 
2313     /* Provide 16 byte alignment for the PRNG, and basic alignment for
2314      * the argv and envp pointers.
2315      */
2316     if (STACK_GROWS_DOWN) {
2317         sp = QEMU_ALIGN_DOWN(sp, 16);
2318     } else {
2319         sp = QEMU_ALIGN_UP(sp, 16);
2320     }
2321 
2322     /*
2323      * Generate 16 random bytes for userspace PRNG seeding.
2324      */
2325     qemu_guest_getrandom_nofail(k_rand_bytes, sizeof(k_rand_bytes));
2326     if (STACK_GROWS_DOWN) {
2327         sp -= 16;
2328         u_rand_bytes = sp;
2329         /* FIXME - check return value of memcpy_to_target() for failure */
2330         memcpy_to_target(sp, k_rand_bytes, 16);
2331     } else {
2332         memcpy_to_target(sp, k_rand_bytes, 16);
2333         u_rand_bytes = sp;
2334         sp += 16;
2335     }
2336 
2337     size = (DLINFO_ITEMS + 1) * 2;
2338     if (k_base_platform)
2339         size += 2;
2340     if (k_platform)
2341         size += 2;
2342 #ifdef DLINFO_ARCH_ITEMS
2343     size += DLINFO_ARCH_ITEMS * 2;
2344 #endif
2345 #ifdef ELF_HWCAP2
2346     size += 2;
2347 #endif
2348     info->auxv_len = size * n;
2349 
2350     size += envc + argc + 2;
2351     size += 1;  /* argc itself */
2352     size *= n;
2353 
2354     /* Allocate space and finalize stack alignment for entry now.  */
2355     if (STACK_GROWS_DOWN) {
2356         u_argc = QEMU_ALIGN_DOWN(sp - size, STACK_ALIGNMENT);
2357         sp = u_argc;
2358     } else {
2359         u_argc = sp;
2360         sp = QEMU_ALIGN_UP(sp + size, STACK_ALIGNMENT);
2361     }
2362 
2363     u_argv = u_argc + n;
2364     u_envp = u_argv + (argc + 1) * n;
2365     u_auxv = u_envp + (envc + 1) * n;
2366     info->saved_auxv = u_auxv;
2367     info->argc = argc;
2368     info->envc = envc;
2369     info->argv = u_argv;
2370     info->envp = u_envp;
2371 
2372     /* This is correct because Linux defines
2373      * elf_addr_t as Elf32_Off / Elf64_Off
2374      */
2375 #define NEW_AUX_ENT(id, val) do {               \
2376         put_user_ual(id, u_auxv);  u_auxv += n; \
2377         put_user_ual(val, u_auxv); u_auxv += n; \
2378     } while(0)
2379 
2380 #ifdef ARCH_DLINFO
2381     /*
2382      * ARCH_DLINFO must come first so platform specific code can enforce
2383      * special alignment requirements on the AUXV if necessary (eg. PPC).
2384      */
2385     ARCH_DLINFO;
2386 #endif
2387     /* There must be exactly DLINFO_ITEMS entries here, or the assert
2388      * on info->auxv_len will trigger.
2389      */
2390     NEW_AUX_ENT(AT_PHDR, (abi_ulong)(info->load_addr + exec->e_phoff));
2391     NEW_AUX_ENT(AT_PHENT, (abi_ulong)(sizeof (struct elf_phdr)));
2392     NEW_AUX_ENT(AT_PHNUM, (abi_ulong)(exec->e_phnum));
2393     if ((info->alignment & ~qemu_host_page_mask) != 0) {
2394         /* Target doesn't support host page size alignment */
2395         NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(TARGET_PAGE_SIZE));
2396     } else {
2397         NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(MAX(TARGET_PAGE_SIZE,
2398                                                qemu_host_page_size)));
2399     }
2400     NEW_AUX_ENT(AT_BASE, (abi_ulong)(interp_info ? interp_info->load_addr : 0));
2401     NEW_AUX_ENT(AT_FLAGS, (abi_ulong)0);
2402     NEW_AUX_ENT(AT_ENTRY, info->entry);
2403     NEW_AUX_ENT(AT_UID, (abi_ulong) getuid());
2404     NEW_AUX_ENT(AT_EUID, (abi_ulong) geteuid());
2405     NEW_AUX_ENT(AT_GID, (abi_ulong) getgid());
2406     NEW_AUX_ENT(AT_EGID, (abi_ulong) getegid());
2407     NEW_AUX_ENT(AT_HWCAP, (abi_ulong) ELF_HWCAP);
2408     NEW_AUX_ENT(AT_CLKTCK, (abi_ulong) sysconf(_SC_CLK_TCK));
2409     NEW_AUX_ENT(AT_RANDOM, (abi_ulong) u_rand_bytes);
2410     NEW_AUX_ENT(AT_SECURE, (abi_ulong) qemu_getauxval(AT_SECURE));
2411     NEW_AUX_ENT(AT_EXECFN, info->file_string);
2412 
2413 #ifdef ELF_HWCAP2
2414     NEW_AUX_ENT(AT_HWCAP2, (abi_ulong) ELF_HWCAP2);
2415 #endif
2416 
2417     if (u_base_platform) {
2418         NEW_AUX_ENT(AT_BASE_PLATFORM, u_base_platform);
2419     }
2420     if (u_platform) {
2421         NEW_AUX_ENT(AT_PLATFORM, u_platform);
2422     }
2423     NEW_AUX_ENT (AT_NULL, 0);
2424 #undef NEW_AUX_ENT
2425 
2426     /* Check that our initial calculation of the auxv length matches how much
2427      * we actually put into it.
2428      */
2429     assert(info->auxv_len == u_auxv - info->saved_auxv);
2430 
2431     put_user_ual(argc, u_argc);
2432 
2433     p = info->arg_strings;
2434     for (i = 0; i < argc; ++i) {
2435         put_user_ual(p, u_argv);
2436         u_argv += n;
2437         p += target_strlen(p) + 1;
2438     }
2439     put_user_ual(0, u_argv);
2440 
2441     p = info->env_strings;
2442     for (i = 0; i < envc; ++i) {
2443         put_user_ual(p, u_envp);
2444         u_envp += n;
2445         p += target_strlen(p) + 1;
2446     }
2447     put_user_ual(0, u_envp);
2448 
2449     return sp;
2450 }
2451 
2452 #if defined(HI_COMMPAGE)
2453 #define LO_COMMPAGE -1
2454 #elif defined(LO_COMMPAGE)
2455 #define HI_COMMPAGE 0
2456 #else
2457 #define HI_COMMPAGE 0
2458 #define LO_COMMPAGE -1
2459 #ifndef INIT_GUEST_COMMPAGE
2460 #define init_guest_commpage() true
2461 #endif
2462 #endif
2463 
2464 static void pgb_fail_in_use(const char *image_name)
2465 {
2466     error_report("%s: requires virtual address space that is in use "
2467                  "(omit the -B option or choose a different value)",
2468                  image_name);
2469     exit(EXIT_FAILURE);
2470 }
2471 
2472 static void pgb_have_guest_base(const char *image_name, abi_ulong guest_loaddr,
2473                                 abi_ulong guest_hiaddr, long align)
2474 {
2475     const int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE;
2476     void *addr, *test;
2477 
2478     if (!QEMU_IS_ALIGNED(guest_base, align)) {
2479         fprintf(stderr, "Requested guest base %p does not satisfy "
2480                 "host minimum alignment (0x%lx)\n",
2481                 (void *)guest_base, align);
2482         exit(EXIT_FAILURE);
2483     }
2484 
2485     /* Sanity check the guest binary. */
2486     if (reserved_va) {
2487         if (guest_hiaddr > reserved_va) {
2488             error_report("%s: requires more than reserved virtual "
2489                          "address space (0x%" PRIx64 " > 0x%lx)",
2490                          image_name, (uint64_t)guest_hiaddr, reserved_va);
2491             exit(EXIT_FAILURE);
2492         }
2493     } else {
2494 #if HOST_LONG_BITS < TARGET_ABI_BITS
2495         if ((guest_hiaddr - guest_base) > ~(uintptr_t)0) {
2496             error_report("%s: requires more virtual address space "
2497                          "than the host can provide (0x%" PRIx64 ")",
2498                          image_name, (uint64_t)guest_hiaddr - guest_base);
2499             exit(EXIT_FAILURE);
2500         }
2501 #endif
2502     }
2503 
2504     /*
2505      * Expand the allocation to the entire reserved_va.
2506      * Exclude the mmap_min_addr hole.
2507      */
2508     if (reserved_va) {
2509         guest_loaddr = (guest_base >= mmap_min_addr ? 0
2510                         : mmap_min_addr - guest_base);
2511         guest_hiaddr = reserved_va;
2512     }
2513 
2514     /* Reserve the address space for the binary, or reserved_va. */
2515     test = g2h_untagged(guest_loaddr);
2516     addr = mmap(test, guest_hiaddr - guest_loaddr, PROT_NONE, flags, -1, 0);
2517     if (test != addr) {
2518         pgb_fail_in_use(image_name);
2519     }
2520     qemu_log_mask(CPU_LOG_PAGE,
2521                   "%s: base @ %p for " TARGET_ABI_FMT_ld " bytes\n",
2522                   __func__, addr, guest_hiaddr - guest_loaddr);
2523 }
2524 
2525 /**
2526  * pgd_find_hole_fallback: potential mmap address
2527  * @guest_size: size of available space
2528  * @brk: location of break
2529  * @align: memory alignment
2530  *
2531  * This is a fallback method for finding a hole in the host address
2532  * space if we don't have the benefit of being able to access
2533  * /proc/self/map. It can potentially take a very long time as we can
2534  * only dumbly iterate up the host address space seeing if the
2535  * allocation would work.
2536  */
2537 static uintptr_t pgd_find_hole_fallback(uintptr_t guest_size, uintptr_t brk,
2538                                         long align, uintptr_t offset)
2539 {
2540     uintptr_t base;
2541 
2542     /* Start (aligned) at the bottom and work our way up */
2543     base = ROUND_UP(mmap_min_addr, align);
2544 
2545     while (true) {
2546         uintptr_t align_start, end;
2547         align_start = ROUND_UP(base, align);
2548         end = align_start + guest_size + offset;
2549 
2550         /* if brk is anywhere in the range give ourselves some room to grow. */
2551         if (align_start <= brk && brk < end) {
2552             base = brk + (16 * MiB);
2553             continue;
2554         } else if (align_start + guest_size < align_start) {
2555             /* we have run out of space */
2556             return -1;
2557         } else {
2558             int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE |
2559                 MAP_FIXED_NOREPLACE;
2560             void * mmap_start = mmap((void *) align_start, guest_size,
2561                                      PROT_NONE, flags, -1, 0);
2562             if (mmap_start != MAP_FAILED) {
2563                 munmap(mmap_start, guest_size);
2564                 if (mmap_start == (void *) align_start) {
2565                     qemu_log_mask(CPU_LOG_PAGE,
2566                                   "%s: base @ %p for %" PRIdPTR" bytes\n",
2567                                   __func__, mmap_start + offset, guest_size);
2568                     return (uintptr_t) mmap_start + offset;
2569                 }
2570             }
2571             base += qemu_host_page_size;
2572         }
2573     }
2574 }
2575 
2576 /* Return value for guest_base, or -1 if no hole found. */
2577 static uintptr_t pgb_find_hole(uintptr_t guest_loaddr, uintptr_t guest_size,
2578                                long align, uintptr_t offset)
2579 {
2580     GSList *maps, *iter;
2581     uintptr_t this_start, this_end, next_start, brk;
2582     intptr_t ret = -1;
2583 
2584     assert(QEMU_IS_ALIGNED(guest_loaddr, align));
2585 
2586     maps = read_self_maps();
2587 
2588     /* Read brk after we've read the maps, which will malloc. */
2589     brk = (uintptr_t)sbrk(0);
2590 
2591     if (!maps) {
2592         return pgd_find_hole_fallback(guest_size, brk, align, offset);
2593     }
2594 
2595     /* The first hole is before the first map entry. */
2596     this_start = mmap_min_addr;
2597 
2598     for (iter = maps; iter;
2599          this_start = next_start, iter = g_slist_next(iter)) {
2600         uintptr_t align_start, hole_size;
2601 
2602         this_end = ((MapInfo *)iter->data)->start;
2603         next_start = ((MapInfo *)iter->data)->end;
2604         align_start = ROUND_UP(this_start + offset, align);
2605 
2606         /* Skip holes that are too small. */
2607         if (align_start >= this_end) {
2608             continue;
2609         }
2610         hole_size = this_end - align_start;
2611         if (hole_size < guest_size) {
2612             continue;
2613         }
2614 
2615         /* If this hole contains brk, give ourselves some room to grow. */
2616         if (this_start <= brk && brk < this_end) {
2617             hole_size -= guest_size;
2618             if (sizeof(uintptr_t) == 8 && hole_size >= 1 * GiB) {
2619                 align_start += 1 * GiB;
2620             } else if (hole_size >= 16 * MiB) {
2621                 align_start += 16 * MiB;
2622             } else {
2623                 align_start = (this_end - guest_size) & -align;
2624                 if (align_start < this_start) {
2625                     continue;
2626                 }
2627             }
2628         }
2629 
2630         /* Record the lowest successful match. */
2631         if (ret < 0) {
2632             ret = align_start;
2633         }
2634         /* If this hole contains the identity map, select it. */
2635         if (align_start <= guest_loaddr &&
2636             guest_loaddr + guest_size <= this_end) {
2637             ret = 0;
2638         }
2639         /* If this hole ends above the identity map, stop looking. */
2640         if (this_end >= guest_loaddr) {
2641             break;
2642         }
2643     }
2644     free_self_maps(maps);
2645 
2646     if (ret != -1) {
2647         qemu_log_mask(CPU_LOG_PAGE, "%s: base @ %" PRIxPTR
2648                       " for %" PRIuPTR " bytes\n",
2649                       __func__, ret, guest_size);
2650     }
2651 
2652     return ret;
2653 }
2654 
2655 static void pgb_static(const char *image_name, abi_ulong orig_loaddr,
2656                        abi_ulong orig_hiaddr, long align)
2657 {
2658     uintptr_t loaddr = orig_loaddr;
2659     uintptr_t hiaddr = orig_hiaddr;
2660     uintptr_t offset = 0;
2661     uintptr_t addr;
2662 
2663     if (hiaddr != orig_hiaddr) {
2664         error_report("%s: requires virtual address space that the "
2665                      "host cannot provide (0x%" PRIx64 ")",
2666                      image_name, (uint64_t)orig_hiaddr);
2667         exit(EXIT_FAILURE);
2668     }
2669 
2670     loaddr &= -align;
2671     if (HI_COMMPAGE) {
2672         /*
2673          * Extend the allocation to include the commpage.
2674          * For a 64-bit host, this is just 4GiB; for a 32-bit host we
2675          * need to ensure there is space bellow the guest_base so we
2676          * can map the commpage in the place needed when the address
2677          * arithmetic wraps around.
2678          */
2679         if (sizeof(uintptr_t) == 8 || loaddr >= 0x80000000u) {
2680             hiaddr = (uintptr_t) 4 << 30;
2681         } else {
2682             offset = -(HI_COMMPAGE & -align);
2683         }
2684     } else if (LO_COMMPAGE != -1) {
2685         loaddr = MIN(loaddr, LO_COMMPAGE & -align);
2686     }
2687 
2688     addr = pgb_find_hole(loaddr, hiaddr - loaddr, align, offset);
2689     if (addr == -1) {
2690         /*
2691          * If HI_COMMPAGE, there *might* be a non-consecutive allocation
2692          * that can satisfy both.  But as the normal arm32 link base address
2693          * is ~32k, and we extend down to include the commpage, making the
2694          * overhead only ~96k, this is unlikely.
2695          */
2696         error_report("%s: Unable to allocate %#zx bytes of "
2697                      "virtual address space", image_name,
2698                      (size_t)(hiaddr - loaddr));
2699         exit(EXIT_FAILURE);
2700     }
2701 
2702     guest_base = addr;
2703 
2704     qemu_log_mask(CPU_LOG_PAGE, "%s: base @ %"PRIxPTR" for %" PRIuPTR" bytes\n",
2705                   __func__, addr, hiaddr - loaddr);
2706 }
2707 
2708 static void pgb_dynamic(const char *image_name, long align)
2709 {
2710     /*
2711      * The executable is dynamic and does not require a fixed address.
2712      * All we need is a commpage that satisfies align.
2713      * If we do not need a commpage, leave guest_base == 0.
2714      */
2715     if (HI_COMMPAGE) {
2716         uintptr_t addr, commpage;
2717 
2718         /* 64-bit hosts should have used reserved_va. */
2719         assert(sizeof(uintptr_t) == 4);
2720 
2721         /*
2722          * By putting the commpage at the first hole, that puts guest_base
2723          * just above that, and maximises the positive guest addresses.
2724          */
2725         commpage = HI_COMMPAGE & -align;
2726         addr = pgb_find_hole(commpage, -commpage, align, 0);
2727         assert(addr != -1);
2728         guest_base = addr;
2729     }
2730 }
2731 
2732 static void pgb_reserved_va(const char *image_name, abi_ulong guest_loaddr,
2733                             abi_ulong guest_hiaddr, long align)
2734 {
2735     int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE;
2736     void *addr, *test;
2737 
2738     if (guest_hiaddr > reserved_va) {
2739         error_report("%s: requires more than reserved virtual "
2740                      "address space (0x%" PRIx64 " > 0x%lx)",
2741                      image_name, (uint64_t)guest_hiaddr, reserved_va);
2742         exit(EXIT_FAILURE);
2743     }
2744 
2745     /* Widen the "image" to the entire reserved address space. */
2746     pgb_static(image_name, 0, reserved_va, align);
2747 
2748     /* osdep.h defines this as 0 if it's missing */
2749     flags |= MAP_FIXED_NOREPLACE;
2750 
2751     /* Reserve the memory on the host. */
2752     assert(guest_base != 0);
2753     test = g2h_untagged(0);
2754     addr = mmap(test, reserved_va, PROT_NONE, flags, -1, 0);
2755     if (addr == MAP_FAILED || addr != test) {
2756         error_report("Unable to reserve 0x%lx bytes of virtual address "
2757                      "space at %p (%s) for use as guest address space (check your "
2758                      "virtual memory ulimit setting, min_mmap_addr or reserve less "
2759                      "using -R option)", reserved_va, test, strerror(errno));
2760         exit(EXIT_FAILURE);
2761     }
2762 
2763     qemu_log_mask(CPU_LOG_PAGE, "%s: base @ %p for %lu bytes\n",
2764                   __func__, addr, reserved_va);
2765 }
2766 
2767 void probe_guest_base(const char *image_name, abi_ulong guest_loaddr,
2768                       abi_ulong guest_hiaddr)
2769 {
2770     /* In order to use host shmat, we must be able to honor SHMLBA.  */
2771     uintptr_t align = MAX(SHMLBA, qemu_host_page_size);
2772 
2773     if (have_guest_base) {
2774         pgb_have_guest_base(image_name, guest_loaddr, guest_hiaddr, align);
2775     } else if (reserved_va) {
2776         pgb_reserved_va(image_name, guest_loaddr, guest_hiaddr, align);
2777     } else if (guest_loaddr) {
2778         pgb_static(image_name, guest_loaddr, guest_hiaddr, align);
2779     } else {
2780         pgb_dynamic(image_name, align);
2781     }
2782 
2783     /* Reserve and initialize the commpage. */
2784     if (!init_guest_commpage()) {
2785         /*
2786          * With have_guest_base, the user has selected the address and
2787          * we are trying to work with that.  Otherwise, we have selected
2788          * free space and init_guest_commpage must succeeded.
2789          */
2790         assert(have_guest_base);
2791         pgb_fail_in_use(image_name);
2792     }
2793 
2794     assert(QEMU_IS_ALIGNED(guest_base, align));
2795     qemu_log_mask(CPU_LOG_PAGE, "Locating guest address space "
2796                   "@ 0x%" PRIx64 "\n", (uint64_t)guest_base);
2797 }
2798 
2799 enum {
2800     /* The string "GNU\0" as a magic number. */
2801     GNU0_MAGIC = const_le32('G' | 'N' << 8 | 'U' << 16),
2802     NOTE_DATA_SZ = 1 * KiB,
2803     NOTE_NAME_SZ = 4,
2804     ELF_GNU_PROPERTY_ALIGN = ELF_CLASS == ELFCLASS32 ? 4 : 8,
2805 };
2806 
2807 /*
2808  * Process a single gnu_property entry.
2809  * Return false for error.
2810  */
2811 static bool parse_elf_property(const uint32_t *data, int *off, int datasz,
2812                                struct image_info *info, bool have_prev_type,
2813                                uint32_t *prev_type, Error **errp)
2814 {
2815     uint32_t pr_type, pr_datasz, step;
2816 
2817     if (*off > datasz || !QEMU_IS_ALIGNED(*off, ELF_GNU_PROPERTY_ALIGN)) {
2818         goto error_data;
2819     }
2820     datasz -= *off;
2821     data += *off / sizeof(uint32_t);
2822 
2823     if (datasz < 2 * sizeof(uint32_t)) {
2824         goto error_data;
2825     }
2826     pr_type = data[0];
2827     pr_datasz = data[1];
2828     data += 2;
2829     datasz -= 2 * sizeof(uint32_t);
2830     step = ROUND_UP(pr_datasz, ELF_GNU_PROPERTY_ALIGN);
2831     if (step > datasz) {
2832         goto error_data;
2833     }
2834 
2835     /* Properties are supposed to be unique and sorted on pr_type. */
2836     if (have_prev_type && pr_type <= *prev_type) {
2837         if (pr_type == *prev_type) {
2838             error_setg(errp, "Duplicate property in PT_GNU_PROPERTY");
2839         } else {
2840             error_setg(errp, "Unsorted property in PT_GNU_PROPERTY");
2841         }
2842         return false;
2843     }
2844     *prev_type = pr_type;
2845 
2846     if (!arch_parse_elf_property(pr_type, pr_datasz, data, info, errp)) {
2847         return false;
2848     }
2849 
2850     *off += 2 * sizeof(uint32_t) + step;
2851     return true;
2852 
2853  error_data:
2854     error_setg(errp, "Ill-formed property in PT_GNU_PROPERTY");
2855     return false;
2856 }
2857 
2858 /* Process NT_GNU_PROPERTY_TYPE_0. */
2859 static bool parse_elf_properties(int image_fd,
2860                                  struct image_info *info,
2861                                  const struct elf_phdr *phdr,
2862                                  char bprm_buf[BPRM_BUF_SIZE],
2863                                  Error **errp)
2864 {
2865     union {
2866         struct elf_note nhdr;
2867         uint32_t data[NOTE_DATA_SZ / sizeof(uint32_t)];
2868     } note;
2869 
2870     int n, off, datasz;
2871     bool have_prev_type;
2872     uint32_t prev_type;
2873 
2874     /* Unless the arch requires properties, ignore them. */
2875     if (!ARCH_USE_GNU_PROPERTY) {
2876         return true;
2877     }
2878 
2879     /* If the properties are crazy large, that's too bad. */
2880     n = phdr->p_filesz;
2881     if (n > sizeof(note)) {
2882         error_setg(errp, "PT_GNU_PROPERTY too large");
2883         return false;
2884     }
2885     if (n < sizeof(note.nhdr)) {
2886         error_setg(errp, "PT_GNU_PROPERTY too small");
2887         return false;
2888     }
2889 
2890     if (phdr->p_offset + n <= BPRM_BUF_SIZE) {
2891         memcpy(&note, bprm_buf + phdr->p_offset, n);
2892     } else {
2893         ssize_t len = pread(image_fd, &note, n, phdr->p_offset);
2894         if (len != n) {
2895             error_setg_errno(errp, errno, "Error reading file header");
2896             return false;
2897         }
2898     }
2899 
2900     /*
2901      * The contents of a valid PT_GNU_PROPERTY is a sequence
2902      * of uint32_t -- swap them all now.
2903      */
2904 #ifdef BSWAP_NEEDED
2905     for (int i = 0; i < n / 4; i++) {
2906         bswap32s(note.data + i);
2907     }
2908 #endif
2909 
2910     /*
2911      * Note that nhdr is 3 words, and that the "name" described by namesz
2912      * immediately follows nhdr and is thus at the 4th word.  Further, all
2913      * of the inputs to the kernel's round_up are multiples of 4.
2914      */
2915     if (note.nhdr.n_type != NT_GNU_PROPERTY_TYPE_0 ||
2916         note.nhdr.n_namesz != NOTE_NAME_SZ ||
2917         note.data[3] != GNU0_MAGIC) {
2918         error_setg(errp, "Invalid note in PT_GNU_PROPERTY");
2919         return false;
2920     }
2921     off = sizeof(note.nhdr) + NOTE_NAME_SZ;
2922 
2923     datasz = note.nhdr.n_descsz + off;
2924     if (datasz > n) {
2925         error_setg(errp, "Invalid note size in PT_GNU_PROPERTY");
2926         return false;
2927     }
2928 
2929     have_prev_type = false;
2930     prev_type = 0;
2931     while (1) {
2932         if (off == datasz) {
2933             return true;  /* end, exit ok */
2934         }
2935         if (!parse_elf_property(note.data, &off, datasz, info,
2936                                 have_prev_type, &prev_type, errp)) {
2937             return false;
2938         }
2939         have_prev_type = true;
2940     }
2941 }
2942 
2943 /* Load an ELF image into the address space.
2944 
2945    IMAGE_NAME is the filename of the image, to use in error messages.
2946    IMAGE_FD is the open file descriptor for the image.
2947 
2948    BPRM_BUF is a copy of the beginning of the file; this of course
2949    contains the elf file header at offset 0.  It is assumed that this
2950    buffer is sufficiently aligned to present no problems to the host
2951    in accessing data at aligned offsets within the buffer.
2952 
2953    On return: INFO values will be filled in, as necessary or available.  */
2954 
2955 static void load_elf_image(const char *image_name, int image_fd,
2956                            struct image_info *info, char **pinterp_name,
2957                            char bprm_buf[BPRM_BUF_SIZE])
2958 {
2959     struct elfhdr *ehdr = (struct elfhdr *)bprm_buf;
2960     struct elf_phdr *phdr;
2961     abi_ulong load_addr, load_bias, loaddr, hiaddr, error;
2962     int i, retval, prot_exec;
2963     Error *err = NULL;
2964 
2965     /* First of all, some simple consistency checks */
2966     if (!elf_check_ident(ehdr)) {
2967         error_setg(&err, "Invalid ELF image for this architecture");
2968         goto exit_errmsg;
2969     }
2970     bswap_ehdr(ehdr);
2971     if (!elf_check_ehdr(ehdr)) {
2972         error_setg(&err, "Invalid ELF image for this architecture");
2973         goto exit_errmsg;
2974     }
2975 
2976     i = ehdr->e_phnum * sizeof(struct elf_phdr);
2977     if (ehdr->e_phoff + i <= BPRM_BUF_SIZE) {
2978         phdr = (struct elf_phdr *)(bprm_buf + ehdr->e_phoff);
2979     } else {
2980         phdr = (struct elf_phdr *) alloca(i);
2981         retval = pread(image_fd, phdr, i, ehdr->e_phoff);
2982         if (retval != i) {
2983             goto exit_read;
2984         }
2985     }
2986     bswap_phdr(phdr, ehdr->e_phnum);
2987 
2988     info->nsegs = 0;
2989     info->pt_dynamic_addr = 0;
2990 
2991     mmap_lock();
2992 
2993     /*
2994      * Find the maximum size of the image and allocate an appropriate
2995      * amount of memory to handle that.  Locate the interpreter, if any.
2996      */
2997     loaddr = -1, hiaddr = 0;
2998     info->alignment = 0;
2999     info->exec_stack = EXSTACK_DEFAULT;
3000     for (i = 0; i < ehdr->e_phnum; ++i) {
3001         struct elf_phdr *eppnt = phdr + i;
3002         if (eppnt->p_type == PT_LOAD) {
3003             abi_ulong a = eppnt->p_vaddr - eppnt->p_offset;
3004             if (a < loaddr) {
3005                 loaddr = a;
3006             }
3007             a = eppnt->p_vaddr + eppnt->p_memsz;
3008             if (a > hiaddr) {
3009                 hiaddr = a;
3010             }
3011             ++info->nsegs;
3012             info->alignment |= eppnt->p_align;
3013         } else if (eppnt->p_type == PT_INTERP && pinterp_name) {
3014             g_autofree char *interp_name = NULL;
3015 
3016             if (*pinterp_name) {
3017                 error_setg(&err, "Multiple PT_INTERP entries");
3018                 goto exit_errmsg;
3019             }
3020 
3021             interp_name = g_malloc(eppnt->p_filesz);
3022 
3023             if (eppnt->p_offset + eppnt->p_filesz <= BPRM_BUF_SIZE) {
3024                 memcpy(interp_name, bprm_buf + eppnt->p_offset,
3025                        eppnt->p_filesz);
3026             } else {
3027                 retval = pread(image_fd, interp_name, eppnt->p_filesz,
3028                                eppnt->p_offset);
3029                 if (retval != eppnt->p_filesz) {
3030                     goto exit_read;
3031                 }
3032             }
3033             if (interp_name[eppnt->p_filesz - 1] != 0) {
3034                 error_setg(&err, "Invalid PT_INTERP entry");
3035                 goto exit_errmsg;
3036             }
3037             *pinterp_name = g_steal_pointer(&interp_name);
3038         } else if (eppnt->p_type == PT_GNU_PROPERTY) {
3039             if (!parse_elf_properties(image_fd, info, eppnt, bprm_buf, &err)) {
3040                 goto exit_errmsg;
3041             }
3042         } else if (eppnt->p_type == PT_GNU_STACK) {
3043             info->exec_stack = eppnt->p_flags & PF_X;
3044         }
3045     }
3046 
3047     if (pinterp_name != NULL) {
3048         /*
3049          * This is the main executable.
3050          *
3051          * Reserve extra space for brk.
3052          * We hold on to this space while placing the interpreter
3053          * and the stack, lest they be placed immediately after
3054          * the data segment and block allocation from the brk.
3055          *
3056          * 16MB is chosen as "large enough" without being so large as
3057          * to allow the result to not fit with a 32-bit guest on a
3058          * 32-bit host. However some 64 bit guests (e.g. s390x)
3059          * attempt to place their heap further ahead and currently
3060          * nothing stops them smashing into QEMUs address space.
3061          */
3062 #if TARGET_LONG_BITS == 64
3063         info->reserve_brk = 32 * MiB;
3064 #else
3065         info->reserve_brk = 16 * MiB;
3066 #endif
3067         hiaddr += info->reserve_brk;
3068 
3069         if (ehdr->e_type == ET_EXEC) {
3070             /*
3071              * Make sure that the low address does not conflict with
3072              * MMAP_MIN_ADDR or the QEMU application itself.
3073              */
3074             probe_guest_base(image_name, loaddr, hiaddr);
3075         } else {
3076             /*
3077              * The binary is dynamic, but we still need to
3078              * select guest_base.  In this case we pass a size.
3079              */
3080             probe_guest_base(image_name, 0, hiaddr - loaddr);
3081         }
3082     }
3083 
3084     /*
3085      * Reserve address space for all of this.
3086      *
3087      * In the case of ET_EXEC, we supply MAP_FIXED so that we get
3088      * exactly the address range that is required.
3089      *
3090      * Otherwise this is ET_DYN, and we are searching for a location
3091      * that can hold the memory space required.  If the image is
3092      * pre-linked, LOADDR will be non-zero, and the kernel should
3093      * honor that address if it happens to be free.
3094      *
3095      * In both cases, we will overwrite pages in this range with mappings
3096      * from the executable.
3097      */
3098     load_addr = target_mmap(loaddr, hiaddr - loaddr, PROT_NONE,
3099                             MAP_PRIVATE | MAP_ANON | MAP_NORESERVE |
3100                             (ehdr->e_type == ET_EXEC ? MAP_FIXED : 0),
3101                             -1, 0);
3102     if (load_addr == -1) {
3103         goto exit_mmap;
3104     }
3105     load_bias = load_addr - loaddr;
3106 
3107     if (elf_is_fdpic(ehdr)) {
3108         struct elf32_fdpic_loadseg *loadsegs = info->loadsegs =
3109             g_malloc(sizeof(*loadsegs) * info->nsegs);
3110 
3111         for (i = 0; i < ehdr->e_phnum; ++i) {
3112             switch (phdr[i].p_type) {
3113             case PT_DYNAMIC:
3114                 info->pt_dynamic_addr = phdr[i].p_vaddr + load_bias;
3115                 break;
3116             case PT_LOAD:
3117                 loadsegs->addr = phdr[i].p_vaddr + load_bias;
3118                 loadsegs->p_vaddr = phdr[i].p_vaddr;
3119                 loadsegs->p_memsz = phdr[i].p_memsz;
3120                 ++loadsegs;
3121                 break;
3122             }
3123         }
3124     }
3125 
3126     info->load_bias = load_bias;
3127     info->code_offset = load_bias;
3128     info->data_offset = load_bias;
3129     info->load_addr = load_addr;
3130     info->entry = ehdr->e_entry + load_bias;
3131     info->start_code = -1;
3132     info->end_code = 0;
3133     info->start_data = -1;
3134     info->end_data = 0;
3135     info->brk = 0;
3136     info->elf_flags = ehdr->e_flags;
3137 
3138     prot_exec = PROT_EXEC;
3139 #ifdef TARGET_AARCH64
3140     /*
3141      * If the BTI feature is present, this indicates that the executable
3142      * pages of the startup binary should be mapped with PROT_BTI, so that
3143      * branch targets are enforced.
3144      *
3145      * The startup binary is either the interpreter or the static executable.
3146      * The interpreter is responsible for all pages of a dynamic executable.
3147      *
3148      * Elf notes are backward compatible to older cpus.
3149      * Do not enable BTI unless it is supported.
3150      */
3151     if ((info->note_flags & GNU_PROPERTY_AARCH64_FEATURE_1_BTI)
3152         && (pinterp_name == NULL || *pinterp_name == 0)
3153         && cpu_isar_feature(aa64_bti, ARM_CPU(thread_cpu))) {
3154         prot_exec |= TARGET_PROT_BTI;
3155     }
3156 #endif
3157 
3158     for (i = 0; i < ehdr->e_phnum; i++) {
3159         struct elf_phdr *eppnt = phdr + i;
3160         if (eppnt->p_type == PT_LOAD) {
3161             abi_ulong vaddr, vaddr_po, vaddr_ps, vaddr_ef, vaddr_em, vaddr_len;
3162             int elf_prot = 0;
3163 
3164             if (eppnt->p_flags & PF_R) {
3165                 elf_prot |= PROT_READ;
3166             }
3167             if (eppnt->p_flags & PF_W) {
3168                 elf_prot |= PROT_WRITE;
3169             }
3170             if (eppnt->p_flags & PF_X) {
3171                 elf_prot |= prot_exec;
3172             }
3173 
3174             vaddr = load_bias + eppnt->p_vaddr;
3175             vaddr_po = TARGET_ELF_PAGEOFFSET(vaddr);
3176             vaddr_ps = TARGET_ELF_PAGESTART(vaddr);
3177 
3178             vaddr_ef = vaddr + eppnt->p_filesz;
3179             vaddr_em = vaddr + eppnt->p_memsz;
3180 
3181             /*
3182              * Some segments may be completely empty, with a non-zero p_memsz
3183              * but no backing file segment.
3184              */
3185             if (eppnt->p_filesz != 0) {
3186                 vaddr_len = TARGET_ELF_PAGELENGTH(eppnt->p_filesz + vaddr_po);
3187                 error = target_mmap(vaddr_ps, vaddr_len, elf_prot,
3188                                     MAP_PRIVATE | MAP_FIXED,
3189                                     image_fd, eppnt->p_offset - vaddr_po);
3190 
3191                 if (error == -1) {
3192                     goto exit_mmap;
3193                 }
3194 
3195                 /*
3196                  * If the load segment requests extra zeros (e.g. bss), map it.
3197                  */
3198                 if (eppnt->p_filesz < eppnt->p_memsz) {
3199                     zero_bss(vaddr_ef, vaddr_em, elf_prot);
3200                 }
3201             } else if (eppnt->p_memsz != 0) {
3202                 vaddr_len = TARGET_ELF_PAGELENGTH(eppnt->p_memsz + vaddr_po);
3203                 error = target_mmap(vaddr_ps, vaddr_len, elf_prot,
3204                                     MAP_PRIVATE | MAP_FIXED | MAP_ANONYMOUS,
3205                                     -1, 0);
3206 
3207                 if (error == -1) {
3208                     goto exit_mmap;
3209                 }
3210             }
3211 
3212             /* Find the full program boundaries.  */
3213             if (elf_prot & PROT_EXEC) {
3214                 if (vaddr < info->start_code) {
3215                     info->start_code = vaddr;
3216                 }
3217                 if (vaddr_ef > info->end_code) {
3218                     info->end_code = vaddr_ef;
3219                 }
3220             }
3221             if (elf_prot & PROT_WRITE) {
3222                 if (vaddr < info->start_data) {
3223                     info->start_data = vaddr;
3224                 }
3225                 if (vaddr_ef > info->end_data) {
3226                     info->end_data = vaddr_ef;
3227                 }
3228             }
3229             if (vaddr_em > info->brk) {
3230                 info->brk = vaddr_em;
3231             }
3232 #ifdef TARGET_MIPS
3233         } else if (eppnt->p_type == PT_MIPS_ABIFLAGS) {
3234             Mips_elf_abiflags_v0 abiflags;
3235             if (eppnt->p_filesz < sizeof(Mips_elf_abiflags_v0)) {
3236                 error_setg(&err, "Invalid PT_MIPS_ABIFLAGS entry");
3237                 goto exit_errmsg;
3238             }
3239             if (eppnt->p_offset + eppnt->p_filesz <= BPRM_BUF_SIZE) {
3240                 memcpy(&abiflags, bprm_buf + eppnt->p_offset,
3241                        sizeof(Mips_elf_abiflags_v0));
3242             } else {
3243                 retval = pread(image_fd, &abiflags, sizeof(Mips_elf_abiflags_v0),
3244                                eppnt->p_offset);
3245                 if (retval != sizeof(Mips_elf_abiflags_v0)) {
3246                     goto exit_read;
3247                 }
3248             }
3249             bswap_mips_abiflags(&abiflags);
3250             info->fp_abi = abiflags.fp_abi;
3251 #endif
3252         }
3253     }
3254 
3255     if (info->end_data == 0) {
3256         info->start_data = info->end_code;
3257         info->end_data = info->end_code;
3258     }
3259 
3260     if (qemu_log_enabled()) {
3261         load_symbols(ehdr, image_fd, load_bias);
3262     }
3263 
3264     mmap_unlock();
3265 
3266     close(image_fd);
3267     return;
3268 
3269  exit_read:
3270     if (retval >= 0) {
3271         error_setg(&err, "Incomplete read of file header");
3272     } else {
3273         error_setg_errno(&err, errno, "Error reading file header");
3274     }
3275     goto exit_errmsg;
3276  exit_mmap:
3277     error_setg_errno(&err, errno, "Error mapping file");
3278     goto exit_errmsg;
3279  exit_errmsg:
3280     error_reportf_err(err, "%s: ", image_name);
3281     exit(-1);
3282 }
3283 
3284 static void load_elf_interp(const char *filename, struct image_info *info,
3285                             char bprm_buf[BPRM_BUF_SIZE])
3286 {
3287     int fd, retval;
3288     Error *err = NULL;
3289 
3290     fd = open(path(filename), O_RDONLY);
3291     if (fd < 0) {
3292         error_setg_file_open(&err, errno, filename);
3293         error_report_err(err);
3294         exit(-1);
3295     }
3296 
3297     retval = read(fd, bprm_buf, BPRM_BUF_SIZE);
3298     if (retval < 0) {
3299         error_setg_errno(&err, errno, "Error reading file header");
3300         error_reportf_err(err, "%s: ", filename);
3301         exit(-1);
3302     }
3303 
3304     if (retval < BPRM_BUF_SIZE) {
3305         memset(bprm_buf + retval, 0, BPRM_BUF_SIZE - retval);
3306     }
3307 
3308     load_elf_image(filename, fd, info, NULL, bprm_buf);
3309 }
3310 
3311 static int symfind(const void *s0, const void *s1)
3312 {
3313     target_ulong addr = *(target_ulong *)s0;
3314     struct elf_sym *sym = (struct elf_sym *)s1;
3315     int result = 0;
3316     if (addr < sym->st_value) {
3317         result = -1;
3318     } else if (addr >= sym->st_value + sym->st_size) {
3319         result = 1;
3320     }
3321     return result;
3322 }
3323 
3324 static const char *lookup_symbolxx(struct syminfo *s, target_ulong orig_addr)
3325 {
3326 #if ELF_CLASS == ELFCLASS32
3327     struct elf_sym *syms = s->disas_symtab.elf32;
3328 #else
3329     struct elf_sym *syms = s->disas_symtab.elf64;
3330 #endif
3331 
3332     // binary search
3333     struct elf_sym *sym;
3334 
3335     sym = bsearch(&orig_addr, syms, s->disas_num_syms, sizeof(*syms), symfind);
3336     if (sym != NULL) {
3337         return s->disas_strtab + sym->st_name;
3338     }
3339 
3340     return "";
3341 }
3342 
3343 /* FIXME: This should use elf_ops.h  */
3344 static int symcmp(const void *s0, const void *s1)
3345 {
3346     struct elf_sym *sym0 = (struct elf_sym *)s0;
3347     struct elf_sym *sym1 = (struct elf_sym *)s1;
3348     return (sym0->st_value < sym1->st_value)
3349         ? -1
3350         : ((sym0->st_value > sym1->st_value) ? 1 : 0);
3351 }
3352 
3353 /* Best attempt to load symbols from this ELF object. */
3354 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias)
3355 {
3356     int i, shnum, nsyms, sym_idx = 0, str_idx = 0;
3357     uint64_t segsz;
3358     struct elf_shdr *shdr;
3359     char *strings = NULL;
3360     struct syminfo *s = NULL;
3361     struct elf_sym *new_syms, *syms = NULL;
3362 
3363     shnum = hdr->e_shnum;
3364     i = shnum * sizeof(struct elf_shdr);
3365     shdr = (struct elf_shdr *)alloca(i);
3366     if (pread(fd, shdr, i, hdr->e_shoff) != i) {
3367         return;
3368     }
3369 
3370     bswap_shdr(shdr, shnum);
3371     for (i = 0; i < shnum; ++i) {
3372         if (shdr[i].sh_type == SHT_SYMTAB) {
3373             sym_idx = i;
3374             str_idx = shdr[i].sh_link;
3375             goto found;
3376         }
3377     }
3378 
3379     /* There will be no symbol table if the file was stripped.  */
3380     return;
3381 
3382  found:
3383     /* Now know where the strtab and symtab are.  Snarf them.  */
3384     s = g_try_new(struct syminfo, 1);
3385     if (!s) {
3386         goto give_up;
3387     }
3388 
3389     segsz = shdr[str_idx].sh_size;
3390     s->disas_strtab = strings = g_try_malloc(segsz);
3391     if (!strings ||
3392         pread(fd, strings, segsz, shdr[str_idx].sh_offset) != segsz) {
3393         goto give_up;
3394     }
3395 
3396     segsz = shdr[sym_idx].sh_size;
3397     syms = g_try_malloc(segsz);
3398     if (!syms || pread(fd, syms, segsz, shdr[sym_idx].sh_offset) != segsz) {
3399         goto give_up;
3400     }
3401 
3402     if (segsz / sizeof(struct elf_sym) > INT_MAX) {
3403         /* Implausibly large symbol table: give up rather than ploughing
3404          * on with the number of symbols calculation overflowing
3405          */
3406         goto give_up;
3407     }
3408     nsyms = segsz / sizeof(struct elf_sym);
3409     for (i = 0; i < nsyms; ) {
3410         bswap_sym(syms + i);
3411         /* Throw away entries which we do not need.  */
3412         if (syms[i].st_shndx == SHN_UNDEF
3413             || syms[i].st_shndx >= SHN_LORESERVE
3414             || ELF_ST_TYPE(syms[i].st_info) != STT_FUNC) {
3415             if (i < --nsyms) {
3416                 syms[i] = syms[nsyms];
3417             }
3418         } else {
3419 #if defined(TARGET_ARM) || defined (TARGET_MIPS)
3420             /* The bottom address bit marks a Thumb or MIPS16 symbol.  */
3421             syms[i].st_value &= ~(target_ulong)1;
3422 #endif
3423             syms[i].st_value += load_bias;
3424             i++;
3425         }
3426     }
3427 
3428     /* No "useful" symbol.  */
3429     if (nsyms == 0) {
3430         goto give_up;
3431     }
3432 
3433     /* Attempt to free the storage associated with the local symbols
3434        that we threw away.  Whether or not this has any effect on the
3435        memory allocation depends on the malloc implementation and how
3436        many symbols we managed to discard.  */
3437     new_syms = g_try_renew(struct elf_sym, syms, nsyms);
3438     if (new_syms == NULL) {
3439         goto give_up;
3440     }
3441     syms = new_syms;
3442 
3443     qsort(syms, nsyms, sizeof(*syms), symcmp);
3444 
3445     s->disas_num_syms = nsyms;
3446 #if ELF_CLASS == ELFCLASS32
3447     s->disas_symtab.elf32 = syms;
3448 #else
3449     s->disas_symtab.elf64 = syms;
3450 #endif
3451     s->lookup_symbol = lookup_symbolxx;
3452     s->next = syminfos;
3453     syminfos = s;
3454 
3455     return;
3456 
3457 give_up:
3458     g_free(s);
3459     g_free(strings);
3460     g_free(syms);
3461 }
3462 
3463 uint32_t get_elf_eflags(int fd)
3464 {
3465     struct elfhdr ehdr;
3466     off_t offset;
3467     int ret;
3468 
3469     /* Read ELF header */
3470     offset = lseek(fd, 0, SEEK_SET);
3471     if (offset == (off_t) -1) {
3472         return 0;
3473     }
3474     ret = read(fd, &ehdr, sizeof(ehdr));
3475     if (ret < sizeof(ehdr)) {
3476         return 0;
3477     }
3478     offset = lseek(fd, offset, SEEK_SET);
3479     if (offset == (off_t) -1) {
3480         return 0;
3481     }
3482 
3483     /* Check ELF signature */
3484     if (!elf_check_ident(&ehdr)) {
3485         return 0;
3486     }
3487 
3488     /* check header */
3489     bswap_ehdr(&ehdr);
3490     if (!elf_check_ehdr(&ehdr)) {
3491         return 0;
3492     }
3493 
3494     /* return architecture id */
3495     return ehdr.e_flags;
3496 }
3497 
3498 int load_elf_binary(struct linux_binprm *bprm, struct image_info *info)
3499 {
3500     struct image_info interp_info;
3501     struct elfhdr elf_ex;
3502     char *elf_interpreter = NULL;
3503     char *scratch;
3504 
3505     memset(&interp_info, 0, sizeof(interp_info));
3506 #ifdef TARGET_MIPS
3507     interp_info.fp_abi = MIPS_ABI_FP_UNKNOWN;
3508 #endif
3509 
3510     info->start_mmap = (abi_ulong)ELF_START_MMAP;
3511 
3512     load_elf_image(bprm->filename, bprm->fd, info,
3513                    &elf_interpreter, bprm->buf);
3514 
3515     /* ??? We need a copy of the elf header for passing to create_elf_tables.
3516        If we do nothing, we'll have overwritten this when we re-use bprm->buf
3517        when we load the interpreter.  */
3518     elf_ex = *(struct elfhdr *)bprm->buf;
3519 
3520     /* Do this so that we can load the interpreter, if need be.  We will
3521        change some of these later */
3522     bprm->p = setup_arg_pages(bprm, info);
3523 
3524     scratch = g_new0(char, TARGET_PAGE_SIZE);
3525     if (STACK_GROWS_DOWN) {
3526         bprm->p = copy_elf_strings(1, &bprm->filename, scratch,
3527                                    bprm->p, info->stack_limit);
3528         info->file_string = bprm->p;
3529         bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch,
3530                                    bprm->p, info->stack_limit);
3531         info->env_strings = bprm->p;
3532         bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch,
3533                                    bprm->p, info->stack_limit);
3534         info->arg_strings = bprm->p;
3535     } else {
3536         info->arg_strings = bprm->p;
3537         bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch,
3538                                    bprm->p, info->stack_limit);
3539         info->env_strings = bprm->p;
3540         bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch,
3541                                    bprm->p, info->stack_limit);
3542         info->file_string = bprm->p;
3543         bprm->p = copy_elf_strings(1, &bprm->filename, scratch,
3544                                    bprm->p, info->stack_limit);
3545     }
3546 
3547     g_free(scratch);
3548 
3549     if (!bprm->p) {
3550         fprintf(stderr, "%s: %s\n", bprm->filename, strerror(E2BIG));
3551         exit(-1);
3552     }
3553 
3554     if (elf_interpreter) {
3555         load_elf_interp(elf_interpreter, &interp_info, bprm->buf);
3556 
3557         /* If the program interpreter is one of these two, then assume
3558            an iBCS2 image.  Otherwise assume a native linux image.  */
3559 
3560         if (strcmp(elf_interpreter, "/usr/lib/libc.so.1") == 0
3561             || strcmp(elf_interpreter, "/usr/lib/ld.so.1") == 0) {
3562             info->personality = PER_SVR4;
3563 
3564             /* Why this, you ask???  Well SVr4 maps page 0 as read-only,
3565                and some applications "depend" upon this behavior.  Since
3566                we do not have the power to recompile these, we emulate
3567                the SVr4 behavior.  Sigh.  */
3568             target_mmap(0, qemu_host_page_size, PROT_READ | PROT_EXEC,
3569                         MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
3570         }
3571 #ifdef TARGET_MIPS
3572         info->interp_fp_abi = interp_info.fp_abi;
3573 #endif
3574     }
3575 
3576     /*
3577      * TODO: load a vdso, which would also contain the signal trampolines.
3578      * Otherwise, allocate a private page to hold them.
3579      */
3580     if (TARGET_ARCH_HAS_SIGTRAMP_PAGE) {
3581         abi_long tramp_page = target_mmap(0, TARGET_PAGE_SIZE,
3582                                           PROT_READ | PROT_WRITE,
3583                                           MAP_PRIVATE | MAP_ANON, -1, 0);
3584         if (tramp_page == -1) {
3585             return -errno;
3586         }
3587 
3588         setup_sigtramp(tramp_page);
3589         target_mprotect(tramp_page, TARGET_PAGE_SIZE, PROT_READ | PROT_EXEC);
3590     }
3591 
3592     bprm->p = create_elf_tables(bprm->p, bprm->argc, bprm->envc, &elf_ex,
3593                                 info, (elf_interpreter ? &interp_info : NULL));
3594     info->start_stack = bprm->p;
3595 
3596     /* If we have an interpreter, set that as the program's entry point.
3597        Copy the load_bias as well, to help PPC64 interpret the entry
3598        point as a function descriptor.  Do this after creating elf tables
3599        so that we copy the original program entry point into the AUXV.  */
3600     if (elf_interpreter) {
3601         info->load_bias = interp_info.load_bias;
3602         info->entry = interp_info.entry;
3603         g_free(elf_interpreter);
3604     }
3605 
3606 #ifdef USE_ELF_CORE_DUMP
3607     bprm->core_dump = &elf_core_dump;
3608 #endif
3609 
3610     /*
3611      * If we reserved extra space for brk, release it now.
3612      * The implementation of do_brk in syscalls.c expects to be able
3613      * to mmap pages in this space.
3614      */
3615     if (info->reserve_brk) {
3616         abi_ulong start_brk = HOST_PAGE_ALIGN(info->brk);
3617         abi_ulong end_brk = HOST_PAGE_ALIGN(info->brk + info->reserve_brk);
3618         target_munmap(start_brk, end_brk - start_brk);
3619     }
3620 
3621     return 0;
3622 }
3623 
3624 #ifdef USE_ELF_CORE_DUMP
3625 /*
3626  * Definitions to generate Intel SVR4-like core files.
3627  * These mostly have the same names as the SVR4 types with "target_elf_"
3628  * tacked on the front to prevent clashes with linux definitions,
3629  * and the typedef forms have been avoided.  This is mostly like
3630  * the SVR4 structure, but more Linuxy, with things that Linux does
3631  * not support and which gdb doesn't really use excluded.
3632  *
3633  * Fields we don't dump (their contents is zero) in linux-user qemu
3634  * are marked with XXX.
3635  *
3636  * Core dump code is copied from linux kernel (fs/binfmt_elf.c).
3637  *
3638  * Porting ELF coredump for target is (quite) simple process.  First you
3639  * define USE_ELF_CORE_DUMP in target ELF code (where init_thread() for
3640  * the target resides):
3641  *
3642  * #define USE_ELF_CORE_DUMP
3643  *
3644  * Next you define type of register set used for dumping.  ELF specification
3645  * says that it needs to be array of elf_greg_t that has size of ELF_NREG.
3646  *
3647  * typedef <target_regtype> target_elf_greg_t;
3648  * #define ELF_NREG <number of registers>
3649  * typedef taret_elf_greg_t target_elf_gregset_t[ELF_NREG];
3650  *
3651  * Last step is to implement target specific function that copies registers
3652  * from given cpu into just specified register set.  Prototype is:
3653  *
3654  * static void elf_core_copy_regs(taret_elf_gregset_t *regs,
3655  *                                const CPUArchState *env);
3656  *
3657  * Parameters:
3658  *     regs - copy register values into here (allocated and zeroed by caller)
3659  *     env - copy registers from here
3660  *
3661  * Example for ARM target is provided in this file.
3662  */
3663 
3664 /* An ELF note in memory */
3665 struct memelfnote {
3666     const char *name;
3667     size_t     namesz;
3668     size_t     namesz_rounded;
3669     int        type;
3670     size_t     datasz;
3671     size_t     datasz_rounded;
3672     void       *data;
3673     size_t     notesz;
3674 };
3675 
3676 struct target_elf_siginfo {
3677     abi_int    si_signo; /* signal number */
3678     abi_int    si_code;  /* extra code */
3679     abi_int    si_errno; /* errno */
3680 };
3681 
3682 struct target_elf_prstatus {
3683     struct target_elf_siginfo pr_info;      /* Info associated with signal */
3684     abi_short          pr_cursig;    /* Current signal */
3685     abi_ulong          pr_sigpend;   /* XXX */
3686     abi_ulong          pr_sighold;   /* XXX */
3687     target_pid_t       pr_pid;
3688     target_pid_t       pr_ppid;
3689     target_pid_t       pr_pgrp;
3690     target_pid_t       pr_sid;
3691     struct target_timeval pr_utime;  /* XXX User time */
3692     struct target_timeval pr_stime;  /* XXX System time */
3693     struct target_timeval pr_cutime; /* XXX Cumulative user time */
3694     struct target_timeval pr_cstime; /* XXX Cumulative system time */
3695     target_elf_gregset_t      pr_reg;       /* GP registers */
3696     abi_int            pr_fpvalid;   /* XXX */
3697 };
3698 
3699 #define ELF_PRARGSZ     (80) /* Number of chars for args */
3700 
3701 struct target_elf_prpsinfo {
3702     char         pr_state;       /* numeric process state */
3703     char         pr_sname;       /* char for pr_state */
3704     char         pr_zomb;        /* zombie */
3705     char         pr_nice;        /* nice val */
3706     abi_ulong    pr_flag;        /* flags */
3707     target_uid_t pr_uid;
3708     target_gid_t pr_gid;
3709     target_pid_t pr_pid, pr_ppid, pr_pgrp, pr_sid;
3710     /* Lots missing */
3711     char    pr_fname[16] QEMU_NONSTRING; /* filename of executable */
3712     char    pr_psargs[ELF_PRARGSZ]; /* initial part of arg list */
3713 };
3714 
3715 /* Here is the structure in which status of each thread is captured. */
3716 struct elf_thread_status {
3717     QTAILQ_ENTRY(elf_thread_status)  ets_link;
3718     struct target_elf_prstatus prstatus;   /* NT_PRSTATUS */
3719 #if 0
3720     elf_fpregset_t fpu;             /* NT_PRFPREG */
3721     struct task_struct *thread;
3722     elf_fpxregset_t xfpu;           /* ELF_CORE_XFPREG_TYPE */
3723 #endif
3724     struct memelfnote notes[1];
3725     int num_notes;
3726 };
3727 
3728 struct elf_note_info {
3729     struct memelfnote   *notes;
3730     struct target_elf_prstatus *prstatus;  /* NT_PRSTATUS */
3731     struct target_elf_prpsinfo *psinfo;    /* NT_PRPSINFO */
3732 
3733     QTAILQ_HEAD(, elf_thread_status) thread_list;
3734 #if 0
3735     /*
3736      * Current version of ELF coredump doesn't support
3737      * dumping fp regs etc.
3738      */
3739     elf_fpregset_t *fpu;
3740     elf_fpxregset_t *xfpu;
3741     int thread_status_size;
3742 #endif
3743     int notes_size;
3744     int numnote;
3745 };
3746 
3747 struct vm_area_struct {
3748     target_ulong   vma_start;  /* start vaddr of memory region */
3749     target_ulong   vma_end;    /* end vaddr of memory region */
3750     abi_ulong      vma_flags;  /* protection etc. flags for the region */
3751     QTAILQ_ENTRY(vm_area_struct) vma_link;
3752 };
3753 
3754 struct mm_struct {
3755     QTAILQ_HEAD(, vm_area_struct) mm_mmap;
3756     int mm_count;           /* number of mappings */
3757 };
3758 
3759 static struct mm_struct *vma_init(void);
3760 static void vma_delete(struct mm_struct *);
3761 static int vma_add_mapping(struct mm_struct *, target_ulong,
3762                            target_ulong, abi_ulong);
3763 static int vma_get_mapping_count(const struct mm_struct *);
3764 static struct vm_area_struct *vma_first(const struct mm_struct *);
3765 static struct vm_area_struct *vma_next(struct vm_area_struct *);
3766 static abi_ulong vma_dump_size(const struct vm_area_struct *);
3767 static int vma_walker(void *priv, target_ulong start, target_ulong end,
3768                       unsigned long flags);
3769 
3770 static void fill_elf_header(struct elfhdr *, int, uint16_t, uint32_t);
3771 static void fill_note(struct memelfnote *, const char *, int,
3772                       unsigned int, void *);
3773 static void fill_prstatus(struct target_elf_prstatus *, const TaskState *, int);
3774 static int fill_psinfo(struct target_elf_prpsinfo *, const TaskState *);
3775 static void fill_auxv_note(struct memelfnote *, const TaskState *);
3776 static void fill_elf_note_phdr(struct elf_phdr *, int, off_t);
3777 static size_t note_size(const struct memelfnote *);
3778 static void free_note_info(struct elf_note_info *);
3779 static int fill_note_info(struct elf_note_info *, long, const CPUArchState *);
3780 static void fill_thread_info(struct elf_note_info *, const CPUArchState *);
3781 
3782 static int dump_write(int, const void *, size_t);
3783 static int write_note(struct memelfnote *, int);
3784 static int write_note_info(struct elf_note_info *, int);
3785 
3786 #ifdef BSWAP_NEEDED
3787 static void bswap_prstatus(struct target_elf_prstatus *prstatus)
3788 {
3789     prstatus->pr_info.si_signo = tswap32(prstatus->pr_info.si_signo);
3790     prstatus->pr_info.si_code = tswap32(prstatus->pr_info.si_code);
3791     prstatus->pr_info.si_errno = tswap32(prstatus->pr_info.si_errno);
3792     prstatus->pr_cursig = tswap16(prstatus->pr_cursig);
3793     prstatus->pr_sigpend = tswapal(prstatus->pr_sigpend);
3794     prstatus->pr_sighold = tswapal(prstatus->pr_sighold);
3795     prstatus->pr_pid = tswap32(prstatus->pr_pid);
3796     prstatus->pr_ppid = tswap32(prstatus->pr_ppid);
3797     prstatus->pr_pgrp = tswap32(prstatus->pr_pgrp);
3798     prstatus->pr_sid = tswap32(prstatus->pr_sid);
3799     /* cpu times are not filled, so we skip them */
3800     /* regs should be in correct format already */
3801     prstatus->pr_fpvalid = tswap32(prstatus->pr_fpvalid);
3802 }
3803 
3804 static void bswap_psinfo(struct target_elf_prpsinfo *psinfo)
3805 {
3806     psinfo->pr_flag = tswapal(psinfo->pr_flag);
3807     psinfo->pr_uid = tswap16(psinfo->pr_uid);
3808     psinfo->pr_gid = tswap16(psinfo->pr_gid);
3809     psinfo->pr_pid = tswap32(psinfo->pr_pid);
3810     psinfo->pr_ppid = tswap32(psinfo->pr_ppid);
3811     psinfo->pr_pgrp = tswap32(psinfo->pr_pgrp);
3812     psinfo->pr_sid = tswap32(psinfo->pr_sid);
3813 }
3814 
3815 static void bswap_note(struct elf_note *en)
3816 {
3817     bswap32s(&en->n_namesz);
3818     bswap32s(&en->n_descsz);
3819     bswap32s(&en->n_type);
3820 }
3821 #else
3822 static inline void bswap_prstatus(struct target_elf_prstatus *p) { }
3823 static inline void bswap_psinfo(struct target_elf_prpsinfo *p) {}
3824 static inline void bswap_note(struct elf_note *en) { }
3825 #endif /* BSWAP_NEEDED */
3826 
3827 /*
3828  * Minimal support for linux memory regions.  These are needed
3829  * when we are finding out what memory exactly belongs to
3830  * emulated process.  No locks needed here, as long as
3831  * thread that received the signal is stopped.
3832  */
3833 
3834 static struct mm_struct *vma_init(void)
3835 {
3836     struct mm_struct *mm;
3837 
3838     if ((mm = g_malloc(sizeof (*mm))) == NULL)
3839         return (NULL);
3840 
3841     mm->mm_count = 0;
3842     QTAILQ_INIT(&mm->mm_mmap);
3843 
3844     return (mm);
3845 }
3846 
3847 static void vma_delete(struct mm_struct *mm)
3848 {
3849     struct vm_area_struct *vma;
3850 
3851     while ((vma = vma_first(mm)) != NULL) {
3852         QTAILQ_REMOVE(&mm->mm_mmap, vma, vma_link);
3853         g_free(vma);
3854     }
3855     g_free(mm);
3856 }
3857 
3858 static int vma_add_mapping(struct mm_struct *mm, target_ulong start,
3859                            target_ulong end, abi_ulong flags)
3860 {
3861     struct vm_area_struct *vma;
3862 
3863     if ((vma = g_malloc0(sizeof (*vma))) == NULL)
3864         return (-1);
3865 
3866     vma->vma_start = start;
3867     vma->vma_end = end;
3868     vma->vma_flags = flags;
3869 
3870     QTAILQ_INSERT_TAIL(&mm->mm_mmap, vma, vma_link);
3871     mm->mm_count++;
3872 
3873     return (0);
3874 }
3875 
3876 static struct vm_area_struct *vma_first(const struct mm_struct *mm)
3877 {
3878     return (QTAILQ_FIRST(&mm->mm_mmap));
3879 }
3880 
3881 static struct vm_area_struct *vma_next(struct vm_area_struct *vma)
3882 {
3883     return (QTAILQ_NEXT(vma, vma_link));
3884 }
3885 
3886 static int vma_get_mapping_count(const struct mm_struct *mm)
3887 {
3888     return (mm->mm_count);
3889 }
3890 
3891 /*
3892  * Calculate file (dump) size of given memory region.
3893  */
3894 static abi_ulong vma_dump_size(const struct vm_area_struct *vma)
3895 {
3896     /* if we cannot even read the first page, skip it */
3897     if (!access_ok_untagged(VERIFY_READ, vma->vma_start, TARGET_PAGE_SIZE))
3898         return (0);
3899 
3900     /*
3901      * Usually we don't dump executable pages as they contain
3902      * non-writable code that debugger can read directly from
3903      * target library etc.  However, thread stacks are marked
3904      * also executable so we read in first page of given region
3905      * and check whether it contains elf header.  If there is
3906      * no elf header, we dump it.
3907      */
3908     if (vma->vma_flags & PROT_EXEC) {
3909         char page[TARGET_PAGE_SIZE];
3910 
3911         if (copy_from_user(page, vma->vma_start, sizeof (page))) {
3912             return 0;
3913         }
3914         if ((page[EI_MAG0] == ELFMAG0) &&
3915             (page[EI_MAG1] == ELFMAG1) &&
3916             (page[EI_MAG2] == ELFMAG2) &&
3917             (page[EI_MAG3] == ELFMAG3)) {
3918             /*
3919              * Mappings are possibly from ELF binary.  Don't dump
3920              * them.
3921              */
3922             return (0);
3923         }
3924     }
3925 
3926     return (vma->vma_end - vma->vma_start);
3927 }
3928 
3929 static int vma_walker(void *priv, target_ulong start, target_ulong end,
3930                       unsigned long flags)
3931 {
3932     struct mm_struct *mm = (struct mm_struct *)priv;
3933 
3934     vma_add_mapping(mm, start, end, flags);
3935     return (0);
3936 }
3937 
3938 static void fill_note(struct memelfnote *note, const char *name, int type,
3939                       unsigned int sz, void *data)
3940 {
3941     unsigned int namesz;
3942 
3943     namesz = strlen(name) + 1;
3944     note->name = name;
3945     note->namesz = namesz;
3946     note->namesz_rounded = roundup(namesz, sizeof (int32_t));
3947     note->type = type;
3948     note->datasz = sz;
3949     note->datasz_rounded = roundup(sz, sizeof (int32_t));
3950 
3951     note->data = data;
3952 
3953     /*
3954      * We calculate rounded up note size here as specified by
3955      * ELF document.
3956      */
3957     note->notesz = sizeof (struct elf_note) +
3958         note->namesz_rounded + note->datasz_rounded;
3959 }
3960 
3961 static void fill_elf_header(struct elfhdr *elf, int segs, uint16_t machine,
3962                             uint32_t flags)
3963 {
3964     (void) memset(elf, 0, sizeof(*elf));
3965 
3966     (void) memcpy(elf->e_ident, ELFMAG, SELFMAG);
3967     elf->e_ident[EI_CLASS] = ELF_CLASS;
3968     elf->e_ident[EI_DATA] = ELF_DATA;
3969     elf->e_ident[EI_VERSION] = EV_CURRENT;
3970     elf->e_ident[EI_OSABI] = ELF_OSABI;
3971 
3972     elf->e_type = ET_CORE;
3973     elf->e_machine = machine;
3974     elf->e_version = EV_CURRENT;
3975     elf->e_phoff = sizeof(struct elfhdr);
3976     elf->e_flags = flags;
3977     elf->e_ehsize = sizeof(struct elfhdr);
3978     elf->e_phentsize = sizeof(struct elf_phdr);
3979     elf->e_phnum = segs;
3980 
3981     bswap_ehdr(elf);
3982 }
3983 
3984 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, off_t offset)
3985 {
3986     phdr->p_type = PT_NOTE;
3987     phdr->p_offset = offset;
3988     phdr->p_vaddr = 0;
3989     phdr->p_paddr = 0;
3990     phdr->p_filesz = sz;
3991     phdr->p_memsz = 0;
3992     phdr->p_flags = 0;
3993     phdr->p_align = 0;
3994 
3995     bswap_phdr(phdr, 1);
3996 }
3997 
3998 static size_t note_size(const struct memelfnote *note)
3999 {
4000     return (note->notesz);
4001 }
4002 
4003 static void fill_prstatus(struct target_elf_prstatus *prstatus,
4004                           const TaskState *ts, int signr)
4005 {
4006     (void) memset(prstatus, 0, sizeof (*prstatus));
4007     prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
4008     prstatus->pr_pid = ts->ts_tid;
4009     prstatus->pr_ppid = getppid();
4010     prstatus->pr_pgrp = getpgrp();
4011     prstatus->pr_sid = getsid(0);
4012 
4013     bswap_prstatus(prstatus);
4014 }
4015 
4016 static int fill_psinfo(struct target_elf_prpsinfo *psinfo, const TaskState *ts)
4017 {
4018     char *base_filename;
4019     unsigned int i, len;
4020 
4021     (void) memset(psinfo, 0, sizeof (*psinfo));
4022 
4023     len = ts->info->env_strings - ts->info->arg_strings;
4024     if (len >= ELF_PRARGSZ)
4025         len = ELF_PRARGSZ - 1;
4026     if (copy_from_user(&psinfo->pr_psargs, ts->info->arg_strings, len)) {
4027         return -EFAULT;
4028     }
4029     for (i = 0; i < len; i++)
4030         if (psinfo->pr_psargs[i] == 0)
4031             psinfo->pr_psargs[i] = ' ';
4032     psinfo->pr_psargs[len] = 0;
4033 
4034     psinfo->pr_pid = getpid();
4035     psinfo->pr_ppid = getppid();
4036     psinfo->pr_pgrp = getpgrp();
4037     psinfo->pr_sid = getsid(0);
4038     psinfo->pr_uid = getuid();
4039     psinfo->pr_gid = getgid();
4040 
4041     base_filename = g_path_get_basename(ts->bprm->filename);
4042     /*
4043      * Using strncpy here is fine: at max-length,
4044      * this field is not NUL-terminated.
4045      */
4046     (void) strncpy(psinfo->pr_fname, base_filename,
4047                    sizeof(psinfo->pr_fname));
4048 
4049     g_free(base_filename);
4050     bswap_psinfo(psinfo);
4051     return (0);
4052 }
4053 
4054 static void fill_auxv_note(struct memelfnote *note, const TaskState *ts)
4055 {
4056     elf_addr_t auxv = (elf_addr_t)ts->info->saved_auxv;
4057     elf_addr_t orig_auxv = auxv;
4058     void *ptr;
4059     int len = ts->info->auxv_len;
4060 
4061     /*
4062      * Auxiliary vector is stored in target process stack.  It contains
4063      * {type, value} pairs that we need to dump into note.  This is not
4064      * strictly necessary but we do it here for sake of completeness.
4065      */
4066 
4067     /* read in whole auxv vector and copy it to memelfnote */
4068     ptr = lock_user(VERIFY_READ, orig_auxv, len, 0);
4069     if (ptr != NULL) {
4070         fill_note(note, "CORE", NT_AUXV, len, ptr);
4071         unlock_user(ptr, auxv, len);
4072     }
4073 }
4074 
4075 /*
4076  * Constructs name of coredump file.  We have following convention
4077  * for the name:
4078  *     qemu_<basename-of-target-binary>_<date>-<time>_<pid>.core
4079  *
4080  * Returns the filename
4081  */
4082 static char *core_dump_filename(const TaskState *ts)
4083 {
4084     g_autoptr(GDateTime) now = g_date_time_new_now_local();
4085     g_autofree char *nowstr = g_date_time_format(now, "%Y%m%d-%H%M%S");
4086     g_autofree char *base_filename = g_path_get_basename(ts->bprm->filename);
4087 
4088     return g_strdup_printf("qemu_%s_%s_%d.core",
4089                            base_filename, nowstr, (int)getpid());
4090 }
4091 
4092 static int dump_write(int fd, const void *ptr, size_t size)
4093 {
4094     const char *bufp = (const char *)ptr;
4095     ssize_t bytes_written, bytes_left;
4096     struct rlimit dumpsize;
4097     off_t pos;
4098 
4099     bytes_written = 0;
4100     getrlimit(RLIMIT_CORE, &dumpsize);
4101     if ((pos = lseek(fd, 0, SEEK_CUR))==-1) {
4102         if (errno == ESPIPE) { /* not a seekable stream */
4103             bytes_left = size;
4104         } else {
4105             return pos;
4106         }
4107     } else {
4108         if (dumpsize.rlim_cur <= pos) {
4109             return -1;
4110         } else if (dumpsize.rlim_cur == RLIM_INFINITY) {
4111             bytes_left = size;
4112         } else {
4113             size_t limit_left=dumpsize.rlim_cur - pos;
4114             bytes_left = limit_left >= size ? size : limit_left ;
4115         }
4116     }
4117 
4118     /*
4119      * In normal conditions, single write(2) should do but
4120      * in case of socket etc. this mechanism is more portable.
4121      */
4122     do {
4123         bytes_written = write(fd, bufp, bytes_left);
4124         if (bytes_written < 0) {
4125             if (errno == EINTR)
4126                 continue;
4127             return (-1);
4128         } else if (bytes_written == 0) { /* eof */
4129             return (-1);
4130         }
4131         bufp += bytes_written;
4132         bytes_left -= bytes_written;
4133     } while (bytes_left > 0);
4134 
4135     return (0);
4136 }
4137 
4138 static int write_note(struct memelfnote *men, int fd)
4139 {
4140     struct elf_note en;
4141 
4142     en.n_namesz = men->namesz;
4143     en.n_type = men->type;
4144     en.n_descsz = men->datasz;
4145 
4146     bswap_note(&en);
4147 
4148     if (dump_write(fd, &en, sizeof(en)) != 0)
4149         return (-1);
4150     if (dump_write(fd, men->name, men->namesz_rounded) != 0)
4151         return (-1);
4152     if (dump_write(fd, men->data, men->datasz_rounded) != 0)
4153         return (-1);
4154 
4155     return (0);
4156 }
4157 
4158 static void fill_thread_info(struct elf_note_info *info, const CPUArchState *env)
4159 {
4160     CPUState *cpu = env_cpu((CPUArchState *)env);
4161     TaskState *ts = (TaskState *)cpu->opaque;
4162     struct elf_thread_status *ets;
4163 
4164     ets = g_malloc0(sizeof (*ets));
4165     ets->num_notes = 1; /* only prstatus is dumped */
4166     fill_prstatus(&ets->prstatus, ts, 0);
4167     elf_core_copy_regs(&ets->prstatus.pr_reg, env);
4168     fill_note(&ets->notes[0], "CORE", NT_PRSTATUS, sizeof (ets->prstatus),
4169               &ets->prstatus);
4170 
4171     QTAILQ_INSERT_TAIL(&info->thread_list, ets, ets_link);
4172 
4173     info->notes_size += note_size(&ets->notes[0]);
4174 }
4175 
4176 static void init_note_info(struct elf_note_info *info)
4177 {
4178     /* Initialize the elf_note_info structure so that it is at
4179      * least safe to call free_note_info() on it. Must be
4180      * called before calling fill_note_info().
4181      */
4182     memset(info, 0, sizeof (*info));
4183     QTAILQ_INIT(&info->thread_list);
4184 }
4185 
4186 static int fill_note_info(struct elf_note_info *info,
4187                           long signr, const CPUArchState *env)
4188 {
4189 #define NUMNOTES 3
4190     CPUState *cpu = env_cpu((CPUArchState *)env);
4191     TaskState *ts = (TaskState *)cpu->opaque;
4192     int i;
4193 
4194     info->notes = g_new0(struct memelfnote, NUMNOTES);
4195     if (info->notes == NULL)
4196         return (-ENOMEM);
4197     info->prstatus = g_malloc0(sizeof (*info->prstatus));
4198     if (info->prstatus == NULL)
4199         return (-ENOMEM);
4200     info->psinfo = g_malloc0(sizeof (*info->psinfo));
4201     if (info->prstatus == NULL)
4202         return (-ENOMEM);
4203 
4204     /*
4205      * First fill in status (and registers) of current thread
4206      * including process info & aux vector.
4207      */
4208     fill_prstatus(info->prstatus, ts, signr);
4209     elf_core_copy_regs(&info->prstatus->pr_reg, env);
4210     fill_note(&info->notes[0], "CORE", NT_PRSTATUS,
4211               sizeof (*info->prstatus), info->prstatus);
4212     fill_psinfo(info->psinfo, ts);
4213     fill_note(&info->notes[1], "CORE", NT_PRPSINFO,
4214               sizeof (*info->psinfo), info->psinfo);
4215     fill_auxv_note(&info->notes[2], ts);
4216     info->numnote = 3;
4217 
4218     info->notes_size = 0;
4219     for (i = 0; i < info->numnote; i++)
4220         info->notes_size += note_size(&info->notes[i]);
4221 
4222     /* read and fill status of all threads */
4223     cpu_list_lock();
4224     CPU_FOREACH(cpu) {
4225         if (cpu == thread_cpu) {
4226             continue;
4227         }
4228         fill_thread_info(info, cpu->env_ptr);
4229     }
4230     cpu_list_unlock();
4231 
4232     return (0);
4233 }
4234 
4235 static void free_note_info(struct elf_note_info *info)
4236 {
4237     struct elf_thread_status *ets;
4238 
4239     while (!QTAILQ_EMPTY(&info->thread_list)) {
4240         ets = QTAILQ_FIRST(&info->thread_list);
4241         QTAILQ_REMOVE(&info->thread_list, ets, ets_link);
4242         g_free(ets);
4243     }
4244 
4245     g_free(info->prstatus);
4246     g_free(info->psinfo);
4247     g_free(info->notes);
4248 }
4249 
4250 static int write_note_info(struct elf_note_info *info, int fd)
4251 {
4252     struct elf_thread_status *ets;
4253     int i, error = 0;
4254 
4255     /* write prstatus, psinfo and auxv for current thread */
4256     for (i = 0; i < info->numnote; i++)
4257         if ((error = write_note(&info->notes[i], fd)) != 0)
4258             return (error);
4259 
4260     /* write prstatus for each thread */
4261     QTAILQ_FOREACH(ets, &info->thread_list, ets_link) {
4262         if ((error = write_note(&ets->notes[0], fd)) != 0)
4263             return (error);
4264     }
4265 
4266     return (0);
4267 }
4268 
4269 /*
4270  * Write out ELF coredump.
4271  *
4272  * See documentation of ELF object file format in:
4273  * http://www.caldera.com/developers/devspecs/gabi41.pdf
4274  *
4275  * Coredump format in linux is following:
4276  *
4277  * 0   +----------------------+         \
4278  *     | ELF header           | ET_CORE  |
4279  *     +----------------------+          |
4280  *     | ELF program headers  |          |--- headers
4281  *     | - NOTE section       |          |
4282  *     | - PT_LOAD sections   |          |
4283  *     +----------------------+         /
4284  *     | NOTEs:               |
4285  *     | - NT_PRSTATUS        |
4286  *     | - NT_PRSINFO         |
4287  *     | - NT_AUXV            |
4288  *     +----------------------+ <-- aligned to target page
4289  *     | Process memory dump  |
4290  *     :                      :
4291  *     .                      .
4292  *     :                      :
4293  *     |                      |
4294  *     +----------------------+
4295  *
4296  * NT_PRSTATUS -> struct elf_prstatus (per thread)
4297  * NT_PRSINFO  -> struct elf_prpsinfo
4298  * NT_AUXV is array of { type, value } pairs (see fill_auxv_note()).
4299  *
4300  * Format follows System V format as close as possible.  Current
4301  * version limitations are as follows:
4302  *     - no floating point registers are dumped
4303  *
4304  * Function returns 0 in case of success, negative errno otherwise.
4305  *
4306  * TODO: make this work also during runtime: it should be
4307  * possible to force coredump from running process and then
4308  * continue processing.  For example qemu could set up SIGUSR2
4309  * handler (provided that target process haven't registered
4310  * handler for that) that does the dump when signal is received.
4311  */
4312 static int elf_core_dump(int signr, const CPUArchState *env)
4313 {
4314     const CPUState *cpu = env_cpu((CPUArchState *)env);
4315     const TaskState *ts = (const TaskState *)cpu->opaque;
4316     struct vm_area_struct *vma = NULL;
4317     g_autofree char *corefile = NULL;
4318     struct elf_note_info info;
4319     struct elfhdr elf;
4320     struct elf_phdr phdr;
4321     struct rlimit dumpsize;
4322     struct mm_struct *mm = NULL;
4323     off_t offset = 0, data_offset = 0;
4324     int segs = 0;
4325     int fd = -1;
4326 
4327     init_note_info(&info);
4328 
4329     errno = 0;
4330     getrlimit(RLIMIT_CORE, &dumpsize);
4331     if (dumpsize.rlim_cur == 0)
4332         return 0;
4333 
4334     corefile = core_dump_filename(ts);
4335 
4336     if ((fd = open(corefile, O_WRONLY | O_CREAT,
4337                    S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH)) < 0)
4338         return (-errno);
4339 
4340     /*
4341      * Walk through target process memory mappings and
4342      * set up structure containing this information.  After
4343      * this point vma_xxx functions can be used.
4344      */
4345     if ((mm = vma_init()) == NULL)
4346         goto out;
4347 
4348     walk_memory_regions(mm, vma_walker);
4349     segs = vma_get_mapping_count(mm);
4350 
4351     /*
4352      * Construct valid coredump ELF header.  We also
4353      * add one more segment for notes.
4354      */
4355     fill_elf_header(&elf, segs + 1, ELF_MACHINE, 0);
4356     if (dump_write(fd, &elf, sizeof (elf)) != 0)
4357         goto out;
4358 
4359     /* fill in the in-memory version of notes */
4360     if (fill_note_info(&info, signr, env) < 0)
4361         goto out;
4362 
4363     offset += sizeof (elf);                             /* elf header */
4364     offset += (segs + 1) * sizeof (struct elf_phdr);    /* program headers */
4365 
4366     /* write out notes program header */
4367     fill_elf_note_phdr(&phdr, info.notes_size, offset);
4368 
4369     offset += info.notes_size;
4370     if (dump_write(fd, &phdr, sizeof (phdr)) != 0)
4371         goto out;
4372 
4373     /*
4374      * ELF specification wants data to start at page boundary so
4375      * we align it here.
4376      */
4377     data_offset = offset = roundup(offset, ELF_EXEC_PAGESIZE);
4378 
4379     /*
4380      * Write program headers for memory regions mapped in
4381      * the target process.
4382      */
4383     for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) {
4384         (void) memset(&phdr, 0, sizeof (phdr));
4385 
4386         phdr.p_type = PT_LOAD;
4387         phdr.p_offset = offset;
4388         phdr.p_vaddr = vma->vma_start;
4389         phdr.p_paddr = 0;
4390         phdr.p_filesz = vma_dump_size(vma);
4391         offset += phdr.p_filesz;
4392         phdr.p_memsz = vma->vma_end - vma->vma_start;
4393         phdr.p_flags = vma->vma_flags & PROT_READ ? PF_R : 0;
4394         if (vma->vma_flags & PROT_WRITE)
4395             phdr.p_flags |= PF_W;
4396         if (vma->vma_flags & PROT_EXEC)
4397             phdr.p_flags |= PF_X;
4398         phdr.p_align = ELF_EXEC_PAGESIZE;
4399 
4400         bswap_phdr(&phdr, 1);
4401         if (dump_write(fd, &phdr, sizeof(phdr)) != 0) {
4402             goto out;
4403         }
4404     }
4405 
4406     /*
4407      * Next we write notes just after program headers.  No
4408      * alignment needed here.
4409      */
4410     if (write_note_info(&info, fd) < 0)
4411         goto out;
4412 
4413     /* align data to page boundary */
4414     if (lseek(fd, data_offset, SEEK_SET) != data_offset)
4415         goto out;
4416 
4417     /*
4418      * Finally we can dump process memory into corefile as well.
4419      */
4420     for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) {
4421         abi_ulong addr;
4422         abi_ulong end;
4423 
4424         end = vma->vma_start + vma_dump_size(vma);
4425 
4426         for (addr = vma->vma_start; addr < end;
4427              addr += TARGET_PAGE_SIZE) {
4428             char page[TARGET_PAGE_SIZE];
4429             int error;
4430 
4431             /*
4432              *  Read in page from target process memory and
4433              *  write it to coredump file.
4434              */
4435             error = copy_from_user(page, addr, sizeof (page));
4436             if (error != 0) {
4437                 (void) fprintf(stderr, "unable to dump " TARGET_ABI_FMT_lx "\n",
4438                                addr);
4439                 errno = -error;
4440                 goto out;
4441             }
4442             if (dump_write(fd, page, TARGET_PAGE_SIZE) < 0)
4443                 goto out;
4444         }
4445     }
4446 
4447  out:
4448     free_note_info(&info);
4449     if (mm != NULL)
4450         vma_delete(mm);
4451     (void) close(fd);
4452 
4453     if (errno != 0)
4454         return (-errno);
4455     return (0);
4456 }
4457 #endif /* USE_ELF_CORE_DUMP */
4458 
4459 void do_init_thread(struct target_pt_regs *regs, struct image_info *infop)
4460 {
4461     init_thread(regs, infop);
4462 }
4463