Merge tag 'linux-user-for-7.1-pull-request' of https://gitlab.com/laurent_vivier...
[qemu.git] / linux-user / elfload.c
1 /* This is the Linux kernel elf-loading code, ported into user space */
2 #include "qemu/osdep.h"
3 #include <sys/param.h>
4
5 #include <sys/resource.h>
6 #include <sys/shm.h>
7
8 #include "qemu.h"
9 #include "disas/disas.h"
10 #include "qemu/path.h"
11 #include "qemu/queue.h"
12 #include "qemu/guest-random.h"
13 #include "qemu/units.h"
14
15 #ifdef _ARCH_PPC64
16 #undef ARCH_DLINFO
17 #undef ELF_PLATFORM
18 #undef ELF_HWCAP
19 #undef ELF_HWCAP2
20 #undef ELF_CLASS
21 #undef ELF_DATA
22 #undef ELF_ARCH
23 #endif
24
25 #define ELF_OSABI ELFOSABI_SYSV
26
27 /* from personality.h */
28
29 /*
30 * Flags for bug emulation.
31 *
32 * These occupy the top three bytes.
33 */
34 enum {
35 ADDR_NO_RANDOMIZE = 0x0040000, /* disable randomization of VA space */
36 FDPIC_FUNCPTRS = 0x0080000, /* userspace function ptrs point to
37 descriptors (signal handling) */
38 MMAP_PAGE_ZERO = 0x0100000,
39 ADDR_COMPAT_LAYOUT = 0x0200000,
40 READ_IMPLIES_EXEC = 0x0400000,
41 ADDR_LIMIT_32BIT = 0x0800000,
42 SHORT_INODE = 0x1000000,
43 WHOLE_SECONDS = 0x2000000,
44 STICKY_TIMEOUTS = 0x4000000,
45 ADDR_LIMIT_3GB = 0x8000000,
46 };
47
48 /*
49 * Personality types.
50 *
51 * These go in the low byte. Avoid using the top bit, it will
52 * conflict with error returns.
53 */
54 enum {
55 PER_LINUX = 0x0000,
56 PER_LINUX_32BIT = 0x0000 | ADDR_LIMIT_32BIT,
57 PER_LINUX_FDPIC = 0x0000 | FDPIC_FUNCPTRS,
58 PER_SVR4 = 0x0001 | STICKY_TIMEOUTS | MMAP_PAGE_ZERO,
59 PER_SVR3 = 0x0002 | STICKY_TIMEOUTS | SHORT_INODE,
60 PER_SCOSVR3 = 0x0003 | STICKY_TIMEOUTS | WHOLE_SECONDS | SHORT_INODE,
61 PER_OSR5 = 0x0003 | STICKY_TIMEOUTS | WHOLE_SECONDS,
62 PER_WYSEV386 = 0x0004 | STICKY_TIMEOUTS | SHORT_INODE,
63 PER_ISCR4 = 0x0005 | STICKY_TIMEOUTS,
64 PER_BSD = 0x0006,
65 PER_SUNOS = 0x0006 | STICKY_TIMEOUTS,
66 PER_XENIX = 0x0007 | STICKY_TIMEOUTS | SHORT_INODE,
67 PER_LINUX32 = 0x0008,
68 PER_LINUX32_3GB = 0x0008 | ADDR_LIMIT_3GB,
69 PER_IRIX32 = 0x0009 | STICKY_TIMEOUTS,/* IRIX5 32-bit */
70 PER_IRIXN32 = 0x000a | STICKY_TIMEOUTS,/* IRIX6 new 32-bit */
71 PER_IRIX64 = 0x000b | STICKY_TIMEOUTS,/* IRIX6 64-bit */
72 PER_RISCOS = 0x000c,
73 PER_SOLARIS = 0x000d | STICKY_TIMEOUTS,
74 PER_UW7 = 0x000e | STICKY_TIMEOUTS | MMAP_PAGE_ZERO,
75 PER_OSF4 = 0x000f, /* OSF/1 v4 */
76 PER_HPUX = 0x0010,
77 PER_MASK = 0x00ff,
78 };
79
80 /*
81 * Return the base personality without flags.
82 */
83 #define personality(pers) (pers & PER_MASK)
84
85 int info_is_fdpic(struct image_info *info)
86 {
87 return info->personality == PER_LINUX_FDPIC;
88 }
89
90 /* this flag is uneffective under linux too, should be deleted */
91 #ifndef MAP_DENYWRITE
92 #define MAP_DENYWRITE 0
93 #endif
94
95 /* should probably go in elf.h */
96 #ifndef ELIBBAD
97 #define ELIBBAD 80
98 #endif
99
100 #ifdef TARGET_WORDS_BIGENDIAN
101 #define ELF_DATA ELFDATA2MSB
102 #else
103 #define ELF_DATA ELFDATA2LSB
104 #endif
105
106 #ifdef TARGET_ABI_MIPSN32
107 typedef abi_ullong target_elf_greg_t;
108 #define tswapreg(ptr) tswap64(ptr)
109 #else
110 typedef abi_ulong target_elf_greg_t;
111 #define tswapreg(ptr) tswapal(ptr)
112 #endif
113
114 #ifdef USE_UID16
115 typedef abi_ushort target_uid_t;
116 typedef abi_ushort target_gid_t;
117 #else
118 typedef abi_uint target_uid_t;
119 typedef abi_uint target_gid_t;
120 #endif
121 typedef abi_int target_pid_t;
122
123 #ifdef TARGET_I386
124
125 #define ELF_PLATFORM get_elf_platform()
126
127 static const char *get_elf_platform(void)
128 {
129 static char elf_platform[] = "i386";
130 int family = object_property_get_int(OBJECT(thread_cpu), "family", NULL);
131 if (family > 6)
132 family = 6;
133 if (family >= 3)
134 elf_platform[1] = '0' + family;
135 return elf_platform;
136 }
137
138 #define ELF_HWCAP get_elf_hwcap()
139
140 static uint32_t get_elf_hwcap(void)
141 {
142 X86CPU *cpu = X86_CPU(thread_cpu);
143
144 return cpu->env.features[FEAT_1_EDX];
145 }
146
147 #ifdef TARGET_X86_64
148 #define ELF_START_MMAP 0x2aaaaab000ULL
149
150 #define ELF_CLASS ELFCLASS64
151 #define ELF_ARCH EM_X86_64
152
153 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop)
154 {
155 regs->rax = 0;
156 regs->rsp = infop->start_stack;
157 regs->rip = infop->entry;
158 }
159
160 #define ELF_NREG 27
161 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
162
163 /*
164 * Note that ELF_NREG should be 29 as there should be place for
165 * TRAPNO and ERR "registers" as well but linux doesn't dump
166 * those.
167 *
168 * See linux kernel: arch/x86/include/asm/elf.h
169 */
170 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUX86State *env)
171 {
172 (*regs)[0] = env->regs[15];
173 (*regs)[1] = env->regs[14];
174 (*regs)[2] = env->regs[13];
175 (*regs)[3] = env->regs[12];
176 (*regs)[4] = env->regs[R_EBP];
177 (*regs)[5] = env->regs[R_EBX];
178 (*regs)[6] = env->regs[11];
179 (*regs)[7] = env->regs[10];
180 (*regs)[8] = env->regs[9];
181 (*regs)[9] = env->regs[8];
182 (*regs)[10] = env->regs[R_EAX];
183 (*regs)[11] = env->regs[R_ECX];
184 (*regs)[12] = env->regs[R_EDX];
185 (*regs)[13] = env->regs[R_ESI];
186 (*regs)[14] = env->regs[R_EDI];
187 (*regs)[15] = env->regs[R_EAX]; /* XXX */
188 (*regs)[16] = env->eip;
189 (*regs)[17] = env->segs[R_CS].selector & 0xffff;
190 (*regs)[18] = env->eflags;
191 (*regs)[19] = env->regs[R_ESP];
192 (*regs)[20] = env->segs[R_SS].selector & 0xffff;
193 (*regs)[21] = env->segs[R_FS].selector & 0xffff;
194 (*regs)[22] = env->segs[R_GS].selector & 0xffff;
195 (*regs)[23] = env->segs[R_DS].selector & 0xffff;
196 (*regs)[24] = env->segs[R_ES].selector & 0xffff;
197 (*regs)[25] = env->segs[R_FS].selector & 0xffff;
198 (*regs)[26] = env->segs[R_GS].selector & 0xffff;
199 }
200
201 #else
202
203 #define ELF_START_MMAP 0x80000000
204
205 /*
206 * This is used to ensure we don't load something for the wrong architecture.
207 */
208 #define elf_check_arch(x) ( ((x) == EM_386) || ((x) == EM_486) )
209
210 /*
211 * These are used to set parameters in the core dumps.
212 */
213 #define ELF_CLASS ELFCLASS32
214 #define ELF_ARCH EM_386
215
216 static inline void init_thread(struct target_pt_regs *regs,
217 struct image_info *infop)
218 {
219 regs->esp = infop->start_stack;
220 regs->eip = infop->entry;
221
222 /* SVR4/i386 ABI (pages 3-31, 3-32) says that when the program
223 starts %edx contains a pointer to a function which might be
224 registered using `atexit'. This provides a mean for the
225 dynamic linker to call DT_FINI functions for shared libraries
226 that have been loaded before the code runs.
227
228 A value of 0 tells we have no such handler. */
229 regs->edx = 0;
230 }
231
232 #define ELF_NREG 17
233 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
234
235 /*
236 * Note that ELF_NREG should be 19 as there should be place for
237 * TRAPNO and ERR "registers" as well but linux doesn't dump
238 * those.
239 *
240 * See linux kernel: arch/x86/include/asm/elf.h
241 */
242 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUX86State *env)
243 {
244 (*regs)[0] = env->regs[R_EBX];
245 (*regs)[1] = env->regs[R_ECX];
246 (*regs)[2] = env->regs[R_EDX];
247 (*regs)[3] = env->regs[R_ESI];
248 (*regs)[4] = env->regs[R_EDI];
249 (*regs)[5] = env->regs[R_EBP];
250 (*regs)[6] = env->regs[R_EAX];
251 (*regs)[7] = env->segs[R_DS].selector & 0xffff;
252 (*regs)[8] = env->segs[R_ES].selector & 0xffff;
253 (*regs)[9] = env->segs[R_FS].selector & 0xffff;
254 (*regs)[10] = env->segs[R_GS].selector & 0xffff;
255 (*regs)[11] = env->regs[R_EAX]; /* XXX */
256 (*regs)[12] = env->eip;
257 (*regs)[13] = env->segs[R_CS].selector & 0xffff;
258 (*regs)[14] = env->eflags;
259 (*regs)[15] = env->regs[R_ESP];
260 (*regs)[16] = env->segs[R_SS].selector & 0xffff;
261 }
262 #endif
263
264 #define USE_ELF_CORE_DUMP
265 #define ELF_EXEC_PAGESIZE 4096
266
267 #endif
268
269 #ifdef TARGET_ARM
270
271 #ifndef TARGET_AARCH64
272 /* 32 bit ARM definitions */
273
274 #define ELF_START_MMAP 0x80000000
275
276 #define ELF_ARCH EM_ARM
277 #define ELF_CLASS ELFCLASS32
278
279 static inline void init_thread(struct target_pt_regs *regs,
280 struct image_info *infop)
281 {
282 abi_long stack = infop->start_stack;
283 memset(regs, 0, sizeof(*regs));
284
285 regs->uregs[16] = ARM_CPU_MODE_USR;
286 if (infop->entry & 1) {
287 regs->uregs[16] |= CPSR_T;
288 }
289 regs->uregs[15] = infop->entry & 0xfffffffe;
290 regs->uregs[13] = infop->start_stack;
291 /* FIXME - what to for failure of get_user()? */
292 get_user_ual(regs->uregs[2], stack + 8); /* envp */
293 get_user_ual(regs->uregs[1], stack + 4); /* envp */
294 /* XXX: it seems that r0 is zeroed after ! */
295 regs->uregs[0] = 0;
296 /* For uClinux PIC binaries. */
297 /* XXX: Linux does this only on ARM with no MMU (do we care ?) */
298 regs->uregs[10] = infop->start_data;
299
300 /* Support ARM FDPIC. */
301 if (info_is_fdpic(infop)) {
302 /* As described in the ABI document, r7 points to the loadmap info
303 * prepared by the kernel. If an interpreter is needed, r8 points
304 * to the interpreter loadmap and r9 points to the interpreter
305 * PT_DYNAMIC info. If no interpreter is needed, r8 is zero, and
306 * r9 points to the main program PT_DYNAMIC info.
307 */
308 regs->uregs[7] = infop->loadmap_addr;
309 if (infop->interpreter_loadmap_addr) {
310 /* Executable is dynamically loaded. */
311 regs->uregs[8] = infop->interpreter_loadmap_addr;
312 regs->uregs[9] = infop->interpreter_pt_dynamic_addr;
313 } else {
314 regs->uregs[8] = 0;
315 regs->uregs[9] = infop->pt_dynamic_addr;
316 }
317 }
318 }
319
320 #define ELF_NREG 18
321 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
322
323 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUARMState *env)
324 {
325 (*regs)[0] = tswapreg(env->regs[0]);
326 (*regs)[1] = tswapreg(env->regs[1]);
327 (*regs)[2] = tswapreg(env->regs[2]);
328 (*regs)[3] = tswapreg(env->regs[3]);
329 (*regs)[4] = tswapreg(env->regs[4]);
330 (*regs)[5] = tswapreg(env->regs[5]);
331 (*regs)[6] = tswapreg(env->regs[6]);
332 (*regs)[7] = tswapreg(env->regs[7]);
333 (*regs)[8] = tswapreg(env->regs[8]);
334 (*regs)[9] = tswapreg(env->regs[9]);
335 (*regs)[10] = tswapreg(env->regs[10]);
336 (*regs)[11] = tswapreg(env->regs[11]);
337 (*regs)[12] = tswapreg(env->regs[12]);
338 (*regs)[13] = tswapreg(env->regs[13]);
339 (*regs)[14] = tswapreg(env->regs[14]);
340 (*regs)[15] = tswapreg(env->regs[15]);
341
342 (*regs)[16] = tswapreg(cpsr_read((CPUARMState *)env));
343 (*regs)[17] = tswapreg(env->regs[0]); /* XXX */
344 }
345
346 #define USE_ELF_CORE_DUMP
347 #define ELF_EXEC_PAGESIZE 4096
348
349 enum
350 {
351 ARM_HWCAP_ARM_SWP = 1 << 0,
352 ARM_HWCAP_ARM_HALF = 1 << 1,
353 ARM_HWCAP_ARM_THUMB = 1 << 2,
354 ARM_HWCAP_ARM_26BIT = 1 << 3,
355 ARM_HWCAP_ARM_FAST_MULT = 1 << 4,
356 ARM_HWCAP_ARM_FPA = 1 << 5,
357 ARM_HWCAP_ARM_VFP = 1 << 6,
358 ARM_HWCAP_ARM_EDSP = 1 << 7,
359 ARM_HWCAP_ARM_JAVA = 1 << 8,
360 ARM_HWCAP_ARM_IWMMXT = 1 << 9,
361 ARM_HWCAP_ARM_CRUNCH = 1 << 10,
362 ARM_HWCAP_ARM_THUMBEE = 1 << 11,
363 ARM_HWCAP_ARM_NEON = 1 << 12,
364 ARM_HWCAP_ARM_VFPv3 = 1 << 13,
365 ARM_HWCAP_ARM_VFPv3D16 = 1 << 14,
366 ARM_HWCAP_ARM_TLS = 1 << 15,
367 ARM_HWCAP_ARM_VFPv4 = 1 << 16,
368 ARM_HWCAP_ARM_IDIVA = 1 << 17,
369 ARM_HWCAP_ARM_IDIVT = 1 << 18,
370 ARM_HWCAP_ARM_VFPD32 = 1 << 19,
371 ARM_HWCAP_ARM_LPAE = 1 << 20,
372 ARM_HWCAP_ARM_EVTSTRM = 1 << 21,
373 };
374
375 enum {
376 ARM_HWCAP2_ARM_AES = 1 << 0,
377 ARM_HWCAP2_ARM_PMULL = 1 << 1,
378 ARM_HWCAP2_ARM_SHA1 = 1 << 2,
379 ARM_HWCAP2_ARM_SHA2 = 1 << 3,
380 ARM_HWCAP2_ARM_CRC32 = 1 << 4,
381 };
382
383 /* The commpage only exists for 32 bit kernels */
384
385 /* Return 1 if the proposed guest space is suitable for the guest.
386 * Return 0 if the proposed guest space isn't suitable, but another
387 * address space should be tried.
388 * Return -1 if there is no way the proposed guest space can be
389 * valid regardless of the base.
390 * The guest code may leave a page mapped and populate it if the
391 * address is suitable.
392 */
393 static int init_guest_commpage(unsigned long guest_base,
394 unsigned long guest_size)
395 {
396 unsigned long real_start, test_page_addr;
397
398 /* We need to check that we can force a fault on access to the
399 * commpage at 0xffff0fxx
400 */
401 test_page_addr = guest_base + (0xffff0f00 & qemu_host_page_mask);
402
403 /* If the commpage lies within the already allocated guest space,
404 * then there is no way we can allocate it.
405 *
406 * You may be thinking that that this check is redundant because
407 * we already validated the guest size against MAX_RESERVED_VA;
408 * but if qemu_host_page_mask is unusually large, then
409 * test_page_addr may be lower.
410 */
411 if (test_page_addr >= guest_base
412 && test_page_addr < (guest_base + guest_size)) {
413 return -1;
414 }
415
416 /* Note it needs to be writeable to let us initialise it */
417 real_start = (unsigned long)
418 mmap((void *)test_page_addr, qemu_host_page_size,
419 PROT_READ | PROT_WRITE,
420 MAP_ANONYMOUS | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
421
422 /* If we can't map it then try another address */
423 if (real_start == -1ul) {
424 return 0;
425 }
426
427 if (real_start != test_page_addr) {
428 /* OS didn't put the page where we asked - unmap and reject */
429 munmap((void *)real_start, qemu_host_page_size);
430 return 0;
431 }
432
433 /* Leave the page mapped
434 * Populate it (mmap should have left it all 0'd)
435 */
436
437 /* Kernel helper versions */
438 __put_user(5, (uint32_t *)g2h(0xffff0ffcul));
439
440 /* Now it's populated make it RO */
441 if (mprotect((void *)test_page_addr, qemu_host_page_size, PROT_READ)) {
442 perror("Protecting guest commpage");
443 exit(-1);
444 }
445
446 return 1; /* All good */
447 }
448
449 #define ELF_HWCAP get_elf_hwcap()
450 #define ELF_HWCAP2 get_elf_hwcap2()
451
452 static uint32_t get_elf_hwcap(void)
453 {
454 ARMCPU *cpu = ARM_CPU(thread_cpu);
455 uint32_t hwcaps = 0;
456
457 hwcaps |= ARM_HWCAP_ARM_SWP;
458 hwcaps |= ARM_HWCAP_ARM_HALF;
459 hwcaps |= ARM_HWCAP_ARM_THUMB;
460 hwcaps |= ARM_HWCAP_ARM_FAST_MULT;
461
462 /* probe for the extra features */
463 #define GET_FEATURE(feat, hwcap) \
464 do { if (arm_feature(&cpu->env, feat)) { hwcaps |= hwcap; } } while (0)
465
466 #define GET_FEATURE_ID(feat, hwcap) \
467 do { if (cpu_isar_feature(feat, cpu)) { hwcaps |= hwcap; } } while (0)
468
469 /* EDSP is in v5TE and above, but all our v5 CPUs are v5TE */
470 GET_FEATURE(ARM_FEATURE_V5, ARM_HWCAP_ARM_EDSP);
471 GET_FEATURE(ARM_FEATURE_VFP, ARM_HWCAP_ARM_VFP);
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_VFP3, ARM_HWCAP_ARM_VFPv3);
476 GET_FEATURE(ARM_FEATURE_V6K, ARM_HWCAP_ARM_TLS);
477 GET_FEATURE(ARM_FEATURE_VFP4, ARM_HWCAP_ARM_VFPv4);
478 GET_FEATURE_ID(arm_div, ARM_HWCAP_ARM_IDIVA);
479 GET_FEATURE_ID(thumb_div, ARM_HWCAP_ARM_IDIVT);
480 /* All QEMU's VFPv3 CPUs have 32 registers, see VFP_DREG in translate.c.
481 * Note that the ARM_HWCAP_ARM_VFPv3D16 bit is always the inverse of
482 * ARM_HWCAP_ARM_VFPD32 (and so always clear for QEMU); it is unrelated
483 * to our VFP_FP16 feature bit.
484 */
485 GET_FEATURE(ARM_FEATURE_VFP3, ARM_HWCAP_ARM_VFPD32);
486 GET_FEATURE(ARM_FEATURE_LPAE, ARM_HWCAP_ARM_LPAE);
487
488 return hwcaps;
489 }
490
491 static uint32_t get_elf_hwcap2(void)
492 {
493 ARMCPU *cpu = ARM_CPU(thread_cpu);
494 uint32_t hwcaps = 0;
495
496 GET_FEATURE_ID(aa32_aes, ARM_HWCAP2_ARM_AES);
497 GET_FEATURE_ID(aa32_pmull, ARM_HWCAP2_ARM_PMULL);
498 GET_FEATURE_ID(aa32_sha1, ARM_HWCAP2_ARM_SHA1);
499 GET_FEATURE_ID(aa32_sha2, ARM_HWCAP2_ARM_SHA2);
500 GET_FEATURE_ID(aa32_crc32, ARM_HWCAP2_ARM_CRC32);
501 return hwcaps;
502 }
503
504 #undef GET_FEATURE
505 #undef GET_FEATURE_ID
506
507 #define ELF_PLATFORM get_elf_platform()
508
509 static const char *get_elf_platform(void)
510 {
511 CPUARMState *env = thread_cpu->env_ptr;
512
513 #ifdef TARGET_WORDS_BIGENDIAN
514 # define END "b"
515 #else
516 # define END "l"
517 #endif
518
519 if (arm_feature(env, ARM_FEATURE_V8)) {
520 return "v8" END;
521 } else if (arm_feature(env, ARM_FEATURE_V7)) {
522 if (arm_feature(env, ARM_FEATURE_M)) {
523 return "v7m" END;
524 } else {
525 return "v7" END;
526 }
527 } else if (arm_feature(env, ARM_FEATURE_V6)) {
528 return "v6" END;
529 } else if (arm_feature(env, ARM_FEATURE_V5)) {
530 return "v5" END;
531 } else {
532 return "v4" END;
533 }
534
535 #undef END
536 }
537
538 #else
539 /* 64 bit ARM definitions */
540 #define ELF_START_MMAP 0x80000000
541
542 #define ELF_ARCH EM_AARCH64
543 #define ELF_CLASS ELFCLASS64
544 #ifdef TARGET_WORDS_BIGENDIAN
545 # define ELF_PLATFORM "aarch64_be"
546 #else
547 # define ELF_PLATFORM "aarch64"
548 #endif
549
550 static inline void init_thread(struct target_pt_regs *regs,
551 struct image_info *infop)
552 {
553 abi_long stack = infop->start_stack;
554 memset(regs, 0, sizeof(*regs));
555
556 regs->pc = infop->entry & ~0x3ULL;
557 regs->sp = stack;
558 }
559
560 #define ELF_NREG 34
561 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
562
563 static void elf_core_copy_regs(target_elf_gregset_t *regs,
564 const CPUARMState *env)
565 {
566 int i;
567
568 for (i = 0; i < 32; i++) {
569 (*regs)[i] = tswapreg(env->xregs[i]);
570 }
571 (*regs)[32] = tswapreg(env->pc);
572 (*regs)[33] = tswapreg(pstate_read((CPUARMState *)env));
573 }
574
575 #define USE_ELF_CORE_DUMP
576 #define ELF_EXEC_PAGESIZE 4096
577
578 enum {
579 ARM_HWCAP_A64_FP = 1 << 0,
580 ARM_HWCAP_A64_ASIMD = 1 << 1,
581 ARM_HWCAP_A64_EVTSTRM = 1 << 2,
582 ARM_HWCAP_A64_AES = 1 << 3,
583 ARM_HWCAP_A64_PMULL = 1 << 4,
584 ARM_HWCAP_A64_SHA1 = 1 << 5,
585 ARM_HWCAP_A64_SHA2 = 1 << 6,
586 ARM_HWCAP_A64_CRC32 = 1 << 7,
587 ARM_HWCAP_A64_ATOMICS = 1 << 8,
588 ARM_HWCAP_A64_FPHP = 1 << 9,
589 ARM_HWCAP_A64_ASIMDHP = 1 << 10,
590 ARM_HWCAP_A64_CPUID = 1 << 11,
591 ARM_HWCAP_A64_ASIMDRDM = 1 << 12,
592 ARM_HWCAP_A64_JSCVT = 1 << 13,
593 ARM_HWCAP_A64_FCMA = 1 << 14,
594 ARM_HWCAP_A64_LRCPC = 1 << 15,
595 ARM_HWCAP_A64_DCPOP = 1 << 16,
596 ARM_HWCAP_A64_SHA3 = 1 << 17,
597 ARM_HWCAP_A64_SM3 = 1 << 18,
598 ARM_HWCAP_A64_SM4 = 1 << 19,
599 ARM_HWCAP_A64_ASIMDDP = 1 << 20,
600 ARM_HWCAP_A64_SHA512 = 1 << 21,
601 ARM_HWCAP_A64_SVE = 1 << 22,
602 ARM_HWCAP_A64_ASIMDFHM = 1 << 23,
603 ARM_HWCAP_A64_DIT = 1 << 24,
604 ARM_HWCAP_A64_USCAT = 1 << 25,
605 ARM_HWCAP_A64_ILRCPC = 1 << 26,
606 ARM_HWCAP_A64_FLAGM = 1 << 27,
607 ARM_HWCAP_A64_SSBS = 1 << 28,
608 ARM_HWCAP_A64_SB = 1 << 29,
609 ARM_HWCAP_A64_PACA = 1 << 30,
610 ARM_HWCAP_A64_PACG = 1UL << 31,
611
612 ARM_HWCAP2_A64_DCPODP = 1 << 0,
613 ARM_HWCAP2_A64_SVE2 = 1 << 1,
614 ARM_HWCAP2_A64_SVEAES = 1 << 2,
615 ARM_HWCAP2_A64_SVEPMULL = 1 << 3,
616 ARM_HWCAP2_A64_SVEBITPERM = 1 << 4,
617 ARM_HWCAP2_A64_SVESHA3 = 1 << 5,
618 ARM_HWCAP2_A64_SVESM4 = 1 << 6,
619 ARM_HWCAP2_A64_FLAGM2 = 1 << 7,
620 ARM_HWCAP2_A64_FRINT = 1 << 8,
621 };
622
623 #define ELF_HWCAP get_elf_hwcap()
624 #define ELF_HWCAP2 get_elf_hwcap2()
625
626 #define GET_FEATURE_ID(feat, hwcap) \
627 do { if (cpu_isar_feature(feat, cpu)) { hwcaps |= hwcap; } } while (0)
628
629 static uint32_t get_elf_hwcap(void)
630 {
631 ARMCPU *cpu = ARM_CPU(thread_cpu);
632 uint32_t hwcaps = 0;
633
634 hwcaps |= ARM_HWCAP_A64_FP;
635 hwcaps |= ARM_HWCAP_A64_ASIMD;
636 hwcaps |= ARM_HWCAP_A64_CPUID;
637
638 /* probe for the extra features */
639
640 GET_FEATURE_ID(aa64_aes, ARM_HWCAP_A64_AES);
641 GET_FEATURE_ID(aa64_pmull, ARM_HWCAP_A64_PMULL);
642 GET_FEATURE_ID(aa64_sha1, ARM_HWCAP_A64_SHA1);
643 GET_FEATURE_ID(aa64_sha256, ARM_HWCAP_A64_SHA2);
644 GET_FEATURE_ID(aa64_sha512, ARM_HWCAP_A64_SHA512);
645 GET_FEATURE_ID(aa64_crc32, ARM_HWCAP_A64_CRC32);
646 GET_FEATURE_ID(aa64_sha3, ARM_HWCAP_A64_SHA3);
647 GET_FEATURE_ID(aa64_sm3, ARM_HWCAP_A64_SM3);
648 GET_FEATURE_ID(aa64_sm4, ARM_HWCAP_A64_SM4);
649 GET_FEATURE_ID(aa64_fp16, ARM_HWCAP_A64_FPHP | ARM_HWCAP_A64_ASIMDHP);
650 GET_FEATURE_ID(aa64_atomics, ARM_HWCAP_A64_ATOMICS);
651 GET_FEATURE_ID(aa64_rdm, ARM_HWCAP_A64_ASIMDRDM);
652 GET_FEATURE_ID(aa64_dp, ARM_HWCAP_A64_ASIMDDP);
653 GET_FEATURE_ID(aa64_fcma, ARM_HWCAP_A64_FCMA);
654 GET_FEATURE_ID(aa64_sve, ARM_HWCAP_A64_SVE);
655 GET_FEATURE_ID(aa64_pauth, ARM_HWCAP_A64_PACA | ARM_HWCAP_A64_PACG);
656 GET_FEATURE_ID(aa64_fhm, ARM_HWCAP_A64_ASIMDFHM);
657 GET_FEATURE_ID(aa64_jscvt, ARM_HWCAP_A64_JSCVT);
658 GET_FEATURE_ID(aa64_sb, ARM_HWCAP_A64_SB);
659 GET_FEATURE_ID(aa64_condm_4, ARM_HWCAP_A64_FLAGM);
660 GET_FEATURE_ID(aa64_dcpop, ARM_HWCAP_A64_DCPOP);
661
662 return hwcaps;
663 }
664
665 static uint32_t get_elf_hwcap2(void)
666 {
667 ARMCPU *cpu = ARM_CPU(thread_cpu);
668 uint32_t hwcaps = 0;
669
670 GET_FEATURE_ID(aa64_dcpodp, ARM_HWCAP2_A64_DCPODP);
671 GET_FEATURE_ID(aa64_condm_5, ARM_HWCAP2_A64_FLAGM2);
672 GET_FEATURE_ID(aa64_frint, ARM_HWCAP2_A64_FRINT);
673
674 return hwcaps;
675 }
676
677 #undef GET_FEATURE_ID
678
679 #endif /* not TARGET_AARCH64 */
680 #endif /* TARGET_ARM */
681
682 #ifdef TARGET_SPARC
683 #ifdef TARGET_SPARC64
684
685 #define ELF_START_MMAP 0x80000000
686 #define ELF_HWCAP (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \
687 | HWCAP_SPARC_MULDIV | HWCAP_SPARC_V9)
688 #ifndef TARGET_ABI32
689 #define elf_check_arch(x) ( (x) == EM_SPARCV9 || (x) == EM_SPARC32PLUS )
690 #else
691 #define elf_check_arch(x) ( (x) == EM_SPARC32PLUS || (x) == EM_SPARC )
692 #endif
693
694 #define ELF_CLASS ELFCLASS64
695 #define ELF_ARCH EM_SPARCV9
696
697 #define STACK_BIAS 2047
698
699 static inline void init_thread(struct target_pt_regs *regs,
700 struct image_info *infop)
701 {
702 #ifndef TARGET_ABI32
703 regs->tstate = 0;
704 #endif
705 regs->pc = infop->entry;
706 regs->npc = regs->pc + 4;
707 regs->y = 0;
708 #ifdef TARGET_ABI32
709 regs->u_regs[14] = infop->start_stack - 16 * 4;
710 #else
711 if (personality(infop->personality) == PER_LINUX32)
712 regs->u_regs[14] = infop->start_stack - 16 * 4;
713 else
714 regs->u_regs[14] = infop->start_stack - 16 * 8 - STACK_BIAS;
715 #endif
716 }
717
718 #else
719 #define ELF_START_MMAP 0x80000000
720 #define ELF_HWCAP (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \
721 | HWCAP_SPARC_MULDIV)
722
723 #define ELF_CLASS ELFCLASS32
724 #define ELF_ARCH EM_SPARC
725
726 static inline void init_thread(struct target_pt_regs *regs,
727 struct image_info *infop)
728 {
729 regs->psr = 0;
730 regs->pc = infop->entry;
731 regs->npc = regs->pc + 4;
732 regs->y = 0;
733 regs->u_regs[14] = infop->start_stack - 16 * 4;
734 }
735
736 #endif
737 #endif
738
739 #ifdef TARGET_PPC
740
741 #define ELF_MACHINE PPC_ELF_MACHINE
742 #define ELF_START_MMAP 0x80000000
743
744 #if defined(TARGET_PPC64) && !defined(TARGET_ABI32)
745
746 #define elf_check_arch(x) ( (x) == EM_PPC64 )
747
748 #define ELF_CLASS ELFCLASS64
749
750 #else
751
752 #define ELF_CLASS ELFCLASS32
753
754 #endif
755
756 #define ELF_ARCH EM_PPC
757
758 /* Feature masks for the Aux Vector Hardware Capabilities (AT_HWCAP).
759 See arch/powerpc/include/asm/cputable.h. */
760 enum {
761 QEMU_PPC_FEATURE_32 = 0x80000000,
762 QEMU_PPC_FEATURE_64 = 0x40000000,
763 QEMU_PPC_FEATURE_601_INSTR = 0x20000000,
764 QEMU_PPC_FEATURE_HAS_ALTIVEC = 0x10000000,
765 QEMU_PPC_FEATURE_HAS_FPU = 0x08000000,
766 QEMU_PPC_FEATURE_HAS_MMU = 0x04000000,
767 QEMU_PPC_FEATURE_HAS_4xxMAC = 0x02000000,
768 QEMU_PPC_FEATURE_UNIFIED_CACHE = 0x01000000,
769 QEMU_PPC_FEATURE_HAS_SPE = 0x00800000,
770 QEMU_PPC_FEATURE_HAS_EFP_SINGLE = 0x00400000,
771 QEMU_PPC_FEATURE_HAS_EFP_DOUBLE = 0x00200000,
772 QEMU_PPC_FEATURE_NO_TB = 0x00100000,
773 QEMU_PPC_FEATURE_POWER4 = 0x00080000,
774 QEMU_PPC_FEATURE_POWER5 = 0x00040000,
775 QEMU_PPC_FEATURE_POWER5_PLUS = 0x00020000,
776 QEMU_PPC_FEATURE_CELL = 0x00010000,
777 QEMU_PPC_FEATURE_BOOKE = 0x00008000,
778 QEMU_PPC_FEATURE_SMT = 0x00004000,
779 QEMU_PPC_FEATURE_ICACHE_SNOOP = 0x00002000,
780 QEMU_PPC_FEATURE_ARCH_2_05 = 0x00001000,
781 QEMU_PPC_FEATURE_PA6T = 0x00000800,
782 QEMU_PPC_FEATURE_HAS_DFP = 0x00000400,
783 QEMU_PPC_FEATURE_POWER6_EXT = 0x00000200,
784 QEMU_PPC_FEATURE_ARCH_2_06 = 0x00000100,
785 QEMU_PPC_FEATURE_HAS_VSX = 0x00000080,
786 QEMU_PPC_FEATURE_PSERIES_PERFMON_COMPAT = 0x00000040,
787
788 QEMU_PPC_FEATURE_TRUE_LE = 0x00000002,
789 QEMU_PPC_FEATURE_PPC_LE = 0x00000001,
790
791 /* Feature definitions in AT_HWCAP2. */
792 QEMU_PPC_FEATURE2_ARCH_2_07 = 0x80000000, /* ISA 2.07 */
793 QEMU_PPC_FEATURE2_HAS_HTM = 0x40000000, /* Hardware Transactional Memory */
794 QEMU_PPC_FEATURE2_HAS_DSCR = 0x20000000, /* Data Stream Control Register */
795 QEMU_PPC_FEATURE2_HAS_EBB = 0x10000000, /* Event Base Branching */
796 QEMU_PPC_FEATURE2_HAS_ISEL = 0x08000000, /* Integer Select */
797 QEMU_PPC_FEATURE2_HAS_TAR = 0x04000000, /* Target Address Register */
798 QEMU_PPC_FEATURE2_VEC_CRYPTO = 0x02000000,
799 QEMU_PPC_FEATURE2_HTM_NOSC = 0x01000000,
800 QEMU_PPC_FEATURE2_ARCH_3_00 = 0x00800000, /* ISA 3.00 */
801 QEMU_PPC_FEATURE2_HAS_IEEE128 = 0x00400000, /* VSX IEEE Bin Float 128-bit */
802 QEMU_PPC_FEATURE2_DARN = 0x00200000, /* darn random number insn */
803 QEMU_PPC_FEATURE2_SCV = 0x00100000, /* scv syscall */
804 QEMU_PPC_FEATURE2_HTM_NO_SUSPEND = 0x00080000, /* TM w/o suspended state */
805 };
806
807 #define ELF_HWCAP get_elf_hwcap()
808
809 static uint32_t get_elf_hwcap(void)
810 {
811 PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);
812 uint32_t features = 0;
813
814 /* We don't have to be terribly complete here; the high points are
815 Altivec/FP/SPE support. Anything else is just a bonus. */
816 #define GET_FEATURE(flag, feature) \
817 do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0)
818 #define GET_FEATURE2(flags, feature) \
819 do { \
820 if ((cpu->env.insns_flags2 & flags) == flags) { \
821 features |= feature; \
822 } \
823 } while (0)
824 GET_FEATURE(PPC_64B, QEMU_PPC_FEATURE_64);
825 GET_FEATURE(PPC_FLOAT, QEMU_PPC_FEATURE_HAS_FPU);
826 GET_FEATURE(PPC_ALTIVEC, QEMU_PPC_FEATURE_HAS_ALTIVEC);
827 GET_FEATURE(PPC_SPE, QEMU_PPC_FEATURE_HAS_SPE);
828 GET_FEATURE(PPC_SPE_SINGLE, QEMU_PPC_FEATURE_HAS_EFP_SINGLE);
829 GET_FEATURE(PPC_SPE_DOUBLE, QEMU_PPC_FEATURE_HAS_EFP_DOUBLE);
830 GET_FEATURE(PPC_BOOKE, QEMU_PPC_FEATURE_BOOKE);
831 GET_FEATURE(PPC_405_MAC, QEMU_PPC_FEATURE_HAS_4xxMAC);
832 GET_FEATURE2(PPC2_DFP, QEMU_PPC_FEATURE_HAS_DFP);
833 GET_FEATURE2(PPC2_VSX, QEMU_PPC_FEATURE_HAS_VSX);
834 GET_FEATURE2((PPC2_PERM_ISA206 | PPC2_DIVE_ISA206 | PPC2_ATOMIC_ISA206 |
835 PPC2_FP_CVT_ISA206 | PPC2_FP_TST_ISA206),
836 QEMU_PPC_FEATURE_ARCH_2_06);
837 #undef GET_FEATURE
838 #undef GET_FEATURE2
839
840 return features;
841 }
842
843 #define ELF_HWCAP2 get_elf_hwcap2()
844
845 static uint32_t get_elf_hwcap2(void)
846 {
847 PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);
848 uint32_t features = 0;
849
850 #define GET_FEATURE(flag, feature) \
851 do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0)
852 #define GET_FEATURE2(flag, feature) \
853 do { if (cpu->env.insns_flags2 & flag) { features |= feature; } } while (0)
854
855 GET_FEATURE(PPC_ISEL, QEMU_PPC_FEATURE2_HAS_ISEL);
856 GET_FEATURE2(PPC2_BCTAR_ISA207, QEMU_PPC_FEATURE2_HAS_TAR);
857 GET_FEATURE2((PPC2_BCTAR_ISA207 | PPC2_LSQ_ISA207 | PPC2_ALTIVEC_207 |
858 PPC2_ISA207S), QEMU_PPC_FEATURE2_ARCH_2_07 |
859 QEMU_PPC_FEATURE2_VEC_CRYPTO);
860 GET_FEATURE2(PPC2_ISA300, QEMU_PPC_FEATURE2_ARCH_3_00 |
861 QEMU_PPC_FEATURE2_DARN);
862
863 #undef GET_FEATURE
864 #undef GET_FEATURE2
865
866 return features;
867 }
868
869 /*
870 * The requirements here are:
871 * - keep the final alignment of sp (sp & 0xf)
872 * - make sure the 32-bit value at the first 16 byte aligned position of
873 * AUXV is greater than 16 for glibc compatibility.
874 * AT_IGNOREPPC is used for that.
875 * - for compatibility with glibc ARCH_DLINFO must always be defined on PPC,
876 * even if DLINFO_ARCH_ITEMS goes to zero or is undefined.
877 */
878 #define DLINFO_ARCH_ITEMS 5
879 #define ARCH_DLINFO \
880 do { \
881 PowerPCCPU *cpu = POWERPC_CPU(thread_cpu); \
882 /* \
883 * Handle glibc compatibility: these magic entries must \
884 * be at the lowest addresses in the final auxv. \
885 */ \
886 NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC); \
887 NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC); \
888 NEW_AUX_ENT(AT_DCACHEBSIZE, cpu->env.dcache_line_size); \
889 NEW_AUX_ENT(AT_ICACHEBSIZE, cpu->env.icache_line_size); \
890 NEW_AUX_ENT(AT_UCACHEBSIZE, 0); \
891 } while (0)
892
893 static inline void init_thread(struct target_pt_regs *_regs, struct image_info *infop)
894 {
895 _regs->gpr[1] = infop->start_stack;
896 #if defined(TARGET_PPC64) && !defined(TARGET_ABI32)
897 if (get_ppc64_abi(infop) < 2) {
898 uint64_t val;
899 get_user_u64(val, infop->entry + 8);
900 _regs->gpr[2] = val + infop->load_bias;
901 get_user_u64(val, infop->entry);
902 infop->entry = val + infop->load_bias;
903 } else {
904 _regs->gpr[12] = infop->entry; /* r12 set to global entry address */
905 }
906 #endif
907 _regs->nip = infop->entry;
908 }
909
910 /* See linux kernel: arch/powerpc/include/asm/elf.h. */
911 #define ELF_NREG 48
912 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
913
914 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUPPCState *env)
915 {
916 int i;
917 target_ulong ccr = 0;
918
919 for (i = 0; i < ARRAY_SIZE(env->gpr); i++) {
920 (*regs)[i] = tswapreg(env->gpr[i]);
921 }
922
923 (*regs)[32] = tswapreg(env->nip);
924 (*regs)[33] = tswapreg(env->msr);
925 (*regs)[35] = tswapreg(env->ctr);
926 (*regs)[36] = tswapreg(env->lr);
927 (*regs)[37] = tswapreg(env->xer);
928
929 for (i = 0; i < ARRAY_SIZE(env->crf); i++) {
930 ccr |= env->crf[i] << (32 - ((i + 1) * 4));
931 }
932 (*regs)[38] = tswapreg(ccr);
933 }
934
935 #define USE_ELF_CORE_DUMP
936 #define ELF_EXEC_PAGESIZE 4096
937
938 #endif
939
940 #ifdef TARGET_MIPS
941
942 #define ELF_START_MMAP 0x80000000
943
944 #ifdef TARGET_MIPS64
945 #define ELF_CLASS ELFCLASS64
946 #else
947 #define ELF_CLASS ELFCLASS32
948 #endif
949 #define ELF_ARCH EM_MIPS
950
951 #define elf_check_arch(x) ((x) == EM_MIPS || (x) == EM_NANOMIPS)
952
953 static inline void init_thread(struct target_pt_regs *regs,
954 struct image_info *infop)
955 {
956 regs->cp0_status = 2 << CP0St_KSU;
957 regs->cp0_epc = infop->entry;
958 regs->regs[29] = infop->start_stack;
959 }
960
961 /* See linux kernel: arch/mips/include/asm/elf.h. */
962 #define ELF_NREG 45
963 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
964
965 /* See linux kernel: arch/mips/include/asm/reg.h. */
966 enum {
967 #ifdef TARGET_MIPS64
968 TARGET_EF_R0 = 0,
969 #else
970 TARGET_EF_R0 = 6,
971 #endif
972 TARGET_EF_R26 = TARGET_EF_R0 + 26,
973 TARGET_EF_R27 = TARGET_EF_R0 + 27,
974 TARGET_EF_LO = TARGET_EF_R0 + 32,
975 TARGET_EF_HI = TARGET_EF_R0 + 33,
976 TARGET_EF_CP0_EPC = TARGET_EF_R0 + 34,
977 TARGET_EF_CP0_BADVADDR = TARGET_EF_R0 + 35,
978 TARGET_EF_CP0_STATUS = TARGET_EF_R0 + 36,
979 TARGET_EF_CP0_CAUSE = TARGET_EF_R0 + 37
980 };
981
982 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */
983 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMIPSState *env)
984 {
985 int i;
986
987 for (i = 0; i < TARGET_EF_R0; i++) {
988 (*regs)[i] = 0;
989 }
990 (*regs)[TARGET_EF_R0] = 0;
991
992 for (i = 1; i < ARRAY_SIZE(env->active_tc.gpr); i++) {
993 (*regs)[TARGET_EF_R0 + i] = tswapreg(env->active_tc.gpr[i]);
994 }
995
996 (*regs)[TARGET_EF_R26] = 0;
997 (*regs)[TARGET_EF_R27] = 0;
998 (*regs)[TARGET_EF_LO] = tswapreg(env->active_tc.LO[0]);
999 (*regs)[TARGET_EF_HI] = tswapreg(env->active_tc.HI[0]);
1000 (*regs)[TARGET_EF_CP0_EPC] = tswapreg(env->active_tc.PC);
1001 (*regs)[TARGET_EF_CP0_BADVADDR] = tswapreg(env->CP0_BadVAddr);
1002 (*regs)[TARGET_EF_CP0_STATUS] = tswapreg(env->CP0_Status);
1003 (*regs)[TARGET_EF_CP0_CAUSE] = tswapreg(env->CP0_Cause);
1004 }
1005
1006 #define USE_ELF_CORE_DUMP
1007 #define ELF_EXEC_PAGESIZE 4096
1008
1009 /* See arch/mips/include/uapi/asm/hwcap.h. */
1010 enum {
1011 HWCAP_MIPS_R6 = (1 << 0),
1012 HWCAP_MIPS_MSA = (1 << 1),
1013 };
1014
1015 #define ELF_HWCAP get_elf_hwcap()
1016
1017 static uint32_t get_elf_hwcap(void)
1018 {
1019 MIPSCPU *cpu = MIPS_CPU(thread_cpu);
1020 uint32_t hwcaps = 0;
1021
1022 #define GET_FEATURE(flag, hwcap) \
1023 do { if (cpu->env.insn_flags & (flag)) { hwcaps |= hwcap; } } while (0)
1024
1025 GET_FEATURE(ISA_MIPS32R6 | ISA_MIPS64R6, HWCAP_MIPS_R6);
1026 GET_FEATURE(ASE_MSA, HWCAP_MIPS_MSA);
1027
1028 #undef GET_FEATURE
1029
1030 return hwcaps;
1031 }
1032
1033 #endif /* TARGET_MIPS */
1034
1035 #ifdef TARGET_MICROBLAZE
1036
1037 #define ELF_START_MMAP 0x80000000
1038
1039 #define elf_check_arch(x) ( (x) == EM_MICROBLAZE || (x) == EM_MICROBLAZE_OLD)
1040
1041 #define ELF_CLASS ELFCLASS32
1042 #define ELF_ARCH EM_MICROBLAZE
1043
1044 static inline void init_thread(struct target_pt_regs *regs,
1045 struct image_info *infop)
1046 {
1047 regs->pc = infop->entry;
1048 regs->r1 = infop->start_stack;
1049
1050 }
1051
1052 #define ELF_EXEC_PAGESIZE 4096
1053
1054 #define USE_ELF_CORE_DUMP
1055 #define ELF_NREG 38
1056 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1057
1058 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */
1059 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMBState *env)
1060 {
1061 int i, pos = 0;
1062
1063 for (i = 0; i < 32; i++) {
1064 (*regs)[pos++] = tswapreg(env->regs[i]);
1065 }
1066
1067 for (i = 0; i < 6; i++) {
1068 (*regs)[pos++] = tswapreg(env->sregs[i]);
1069 }
1070 }
1071
1072 #endif /* TARGET_MICROBLAZE */
1073
1074 #ifdef TARGET_NIOS2
1075
1076 #define ELF_START_MMAP 0x80000000
1077
1078 #define elf_check_arch(x) ((x) == EM_ALTERA_NIOS2)
1079
1080 #define ELF_CLASS ELFCLASS32
1081 #define ELF_ARCH EM_ALTERA_NIOS2
1082
1083 static void init_thread(struct target_pt_regs *regs, struct image_info *infop)
1084 {
1085 regs->ea = infop->entry;
1086 regs->sp = infop->start_stack;
1087 regs->estatus = 0x3;
1088 }
1089
1090 #define ELF_EXEC_PAGESIZE 4096
1091
1092 #define USE_ELF_CORE_DUMP
1093 #define ELF_NREG 49
1094 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1095
1096 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */
1097 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1098 const CPUNios2State *env)
1099 {
1100 int i;
1101
1102 (*regs)[0] = -1;
1103 for (i = 1; i < 8; i++) /* r0-r7 */
1104 (*regs)[i] = tswapreg(env->regs[i + 7]);
1105
1106 for (i = 8; i < 16; i++) /* r8-r15 */
1107 (*regs)[i] = tswapreg(env->regs[i - 8]);
1108
1109 for (i = 16; i < 24; i++) /* r16-r23 */
1110 (*regs)[i] = tswapreg(env->regs[i + 7]);
1111 (*regs)[24] = -1; /* R_ET */
1112 (*regs)[25] = -1; /* R_BT */
1113 (*regs)[26] = tswapreg(env->regs[R_GP]);
1114 (*regs)[27] = tswapreg(env->regs[R_SP]);
1115 (*regs)[28] = tswapreg(env->regs[R_FP]);
1116 (*regs)[29] = tswapreg(env->regs[R_EA]);
1117 (*regs)[30] = -1; /* R_SSTATUS */
1118 (*regs)[31] = tswapreg(env->regs[R_RA]);
1119
1120 (*regs)[32] = tswapreg(env->regs[R_PC]);
1121
1122 (*regs)[33] = -1; /* R_STATUS */
1123 (*regs)[34] = tswapreg(env->regs[CR_ESTATUS]);
1124
1125 for (i = 35; i < 49; i++) /* ... */
1126 (*regs)[i] = -1;
1127 }
1128
1129 #endif /* TARGET_NIOS2 */
1130
1131 #ifdef TARGET_OPENRISC
1132
1133 #define ELF_START_MMAP 0x08000000
1134
1135 #define ELF_ARCH EM_OPENRISC
1136 #define ELF_CLASS ELFCLASS32
1137 #define ELF_DATA ELFDATA2MSB
1138
1139 static inline void init_thread(struct target_pt_regs *regs,
1140 struct image_info *infop)
1141 {
1142 regs->pc = infop->entry;
1143 regs->gpr[1] = infop->start_stack;
1144 }
1145
1146 #define USE_ELF_CORE_DUMP
1147 #define ELF_EXEC_PAGESIZE 8192
1148
1149 /* See linux kernel arch/openrisc/include/asm/elf.h. */
1150 #define ELF_NREG 34 /* gprs and pc, sr */
1151 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1152
1153 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1154 const CPUOpenRISCState *env)
1155 {
1156 int i;
1157
1158 for (i = 0; i < 32; i++) {
1159 (*regs)[i] = tswapreg(cpu_get_gpr(env, i));
1160 }
1161 (*regs)[32] = tswapreg(env->pc);
1162 (*regs)[33] = tswapreg(cpu_get_sr(env));
1163 }
1164 #define ELF_HWCAP 0
1165 #define ELF_PLATFORM NULL
1166
1167 #endif /* TARGET_OPENRISC */
1168
1169 #ifdef TARGET_SH4
1170
1171 #define ELF_START_MMAP 0x80000000
1172
1173 #define ELF_CLASS ELFCLASS32
1174 #define ELF_ARCH EM_SH
1175
1176 static inline void init_thread(struct target_pt_regs *regs,
1177 struct image_info *infop)
1178 {
1179 /* Check other registers XXXXX */
1180 regs->pc = infop->entry;
1181 regs->regs[15] = infop->start_stack;
1182 }
1183
1184 /* See linux kernel: arch/sh/include/asm/elf.h. */
1185 #define ELF_NREG 23
1186 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1187
1188 /* See linux kernel: arch/sh/include/asm/ptrace.h. */
1189 enum {
1190 TARGET_REG_PC = 16,
1191 TARGET_REG_PR = 17,
1192 TARGET_REG_SR = 18,
1193 TARGET_REG_GBR = 19,
1194 TARGET_REG_MACH = 20,
1195 TARGET_REG_MACL = 21,
1196 TARGET_REG_SYSCALL = 22
1197 };
1198
1199 static inline void elf_core_copy_regs(target_elf_gregset_t *regs,
1200 const CPUSH4State *env)
1201 {
1202 int i;
1203
1204 for (i = 0; i < 16; i++) {
1205 (*regs)[i] = tswapreg(env->gregs[i]);
1206 }
1207
1208 (*regs)[TARGET_REG_PC] = tswapreg(env->pc);
1209 (*regs)[TARGET_REG_PR] = tswapreg(env->pr);
1210 (*regs)[TARGET_REG_SR] = tswapreg(env->sr);
1211 (*regs)[TARGET_REG_GBR] = tswapreg(env->gbr);
1212 (*regs)[TARGET_REG_MACH] = tswapreg(env->mach);
1213 (*regs)[TARGET_REG_MACL] = tswapreg(env->macl);
1214 (*regs)[TARGET_REG_SYSCALL] = 0; /* FIXME */
1215 }
1216
1217 #define USE_ELF_CORE_DUMP
1218 #define ELF_EXEC_PAGESIZE 4096
1219
1220 enum {
1221 SH_CPU_HAS_FPU = 0x0001, /* Hardware FPU support */
1222 SH_CPU_HAS_P2_FLUSH_BUG = 0x0002, /* Need to flush the cache in P2 area */
1223 SH_CPU_HAS_MMU_PAGE_ASSOC = 0x0004, /* SH3: TLB way selection bit support */
1224 SH_CPU_HAS_DSP = 0x0008, /* SH-DSP: DSP support */
1225 SH_CPU_HAS_PERF_COUNTER = 0x0010, /* Hardware performance counters */
1226 SH_CPU_HAS_PTEA = 0x0020, /* PTEA register */
1227 SH_CPU_HAS_LLSC = 0x0040, /* movli.l/movco.l */
1228 SH_CPU_HAS_L2_CACHE = 0x0080, /* Secondary cache / URAM */
1229 SH_CPU_HAS_OP32 = 0x0100, /* 32-bit instruction support */
1230 SH_CPU_HAS_PTEAEX = 0x0200, /* PTE ASID Extension support */
1231 };
1232
1233 #define ELF_HWCAP get_elf_hwcap()
1234
1235 static uint32_t get_elf_hwcap(void)
1236 {
1237 SuperHCPU *cpu = SUPERH_CPU(thread_cpu);
1238 uint32_t hwcap = 0;
1239
1240 hwcap |= SH_CPU_HAS_FPU;
1241
1242 if (cpu->env.features & SH_FEATURE_SH4A) {
1243 hwcap |= SH_CPU_HAS_LLSC;
1244 }
1245
1246 return hwcap;
1247 }
1248
1249 #endif
1250
1251 #ifdef TARGET_CRIS
1252
1253 #define ELF_START_MMAP 0x80000000
1254
1255 #define ELF_CLASS ELFCLASS32
1256 #define ELF_ARCH EM_CRIS
1257
1258 static inline void init_thread(struct target_pt_regs *regs,
1259 struct image_info *infop)
1260 {
1261 regs->erp = infop->entry;
1262 }
1263
1264 #define ELF_EXEC_PAGESIZE 8192
1265
1266 #endif
1267
1268 #ifdef TARGET_M68K
1269
1270 #define ELF_START_MMAP 0x80000000
1271
1272 #define ELF_CLASS ELFCLASS32
1273 #define ELF_ARCH EM_68K
1274
1275 /* ??? Does this need to do anything?
1276 #define ELF_PLAT_INIT(_r) */
1277
1278 static inline void init_thread(struct target_pt_regs *regs,
1279 struct image_info *infop)
1280 {
1281 regs->usp = infop->start_stack;
1282 regs->sr = 0;
1283 regs->pc = infop->entry;
1284 }
1285
1286 /* See linux kernel: arch/m68k/include/asm/elf.h. */
1287 #define ELF_NREG 20
1288 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1289
1290 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUM68KState *env)
1291 {
1292 (*regs)[0] = tswapreg(env->dregs[1]);
1293 (*regs)[1] = tswapreg(env->dregs[2]);
1294 (*regs)[2] = tswapreg(env->dregs[3]);
1295 (*regs)[3] = tswapreg(env->dregs[4]);
1296 (*regs)[4] = tswapreg(env->dregs[5]);
1297 (*regs)[5] = tswapreg(env->dregs[6]);
1298 (*regs)[6] = tswapreg(env->dregs[7]);
1299 (*regs)[7] = tswapreg(env->aregs[0]);
1300 (*regs)[8] = tswapreg(env->aregs[1]);
1301 (*regs)[9] = tswapreg(env->aregs[2]);
1302 (*regs)[10] = tswapreg(env->aregs[3]);
1303 (*regs)[11] = tswapreg(env->aregs[4]);
1304 (*regs)[12] = tswapreg(env->aregs[5]);
1305 (*regs)[13] = tswapreg(env->aregs[6]);
1306 (*regs)[14] = tswapreg(env->dregs[0]);
1307 (*regs)[15] = tswapreg(env->aregs[7]);
1308 (*regs)[16] = tswapreg(env->dregs[0]); /* FIXME: orig_d0 */
1309 (*regs)[17] = tswapreg(env->sr);
1310 (*regs)[18] = tswapreg(env->pc);
1311 (*regs)[19] = 0; /* FIXME: regs->format | regs->vector */
1312 }
1313
1314 #define USE_ELF_CORE_DUMP
1315 #define ELF_EXEC_PAGESIZE 8192
1316
1317 #endif
1318
1319 #ifdef TARGET_ALPHA
1320
1321 #define ELF_START_MMAP (0x30000000000ULL)
1322
1323 #define ELF_CLASS ELFCLASS64
1324 #define ELF_ARCH EM_ALPHA
1325
1326 static inline void init_thread(struct target_pt_regs *regs,
1327 struct image_info *infop)
1328 {
1329 regs->pc = infop->entry;
1330 regs->ps = 8;
1331 regs->usp = infop->start_stack;
1332 }
1333
1334 #define ELF_EXEC_PAGESIZE 8192
1335
1336 #endif /* TARGET_ALPHA */
1337
1338 #ifdef TARGET_S390X
1339
1340 #define ELF_START_MMAP (0x20000000000ULL)
1341
1342 #define ELF_CLASS ELFCLASS64
1343 #define ELF_DATA ELFDATA2MSB
1344 #define ELF_ARCH EM_S390
1345
1346 #include "elf.h"
1347
1348 #define ELF_HWCAP get_elf_hwcap()
1349
1350 #define GET_FEATURE(_feat, _hwcap) \
1351 do { if (s390_has_feat(_feat)) { hwcap |= _hwcap; } } while (0)
1352
1353 static uint32_t get_elf_hwcap(void)
1354 {
1355 /*
1356 * Let's assume we always have esan3 and zarch.
1357 * 31-bit processes can use 64-bit registers (high gprs).
1358 */
1359 uint32_t hwcap = HWCAP_S390_ESAN3 | HWCAP_S390_ZARCH | HWCAP_S390_HIGH_GPRS;
1360
1361 GET_FEATURE(S390_FEAT_STFLE, HWCAP_S390_STFLE);
1362 GET_FEATURE(S390_FEAT_MSA, HWCAP_S390_MSA);
1363 GET_FEATURE(S390_FEAT_LONG_DISPLACEMENT, HWCAP_S390_LDISP);
1364 GET_FEATURE(S390_FEAT_EXTENDED_IMMEDIATE, HWCAP_S390_EIMM);
1365 if (s390_has_feat(S390_FEAT_EXTENDED_TRANSLATION_3) &&
1366 s390_has_feat(S390_FEAT_ETF3_ENH)) {
1367 hwcap |= HWCAP_S390_ETF3EH;
1368 }
1369 GET_FEATURE(S390_FEAT_VECTOR, HWCAP_S390_VXRS);
1370
1371 return hwcap;
1372 }
1373
1374 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop)
1375 {
1376 regs->psw.addr = infop->entry;
1377 regs->psw.mask = PSW_MASK_64 | PSW_MASK_32;
1378 regs->gprs[15] = infop->start_stack;
1379 }
1380
1381 #endif /* TARGET_S390X */
1382
1383 #ifdef TARGET_TILEGX
1384
1385 /* 42 bits real used address, a half for user mode */
1386 #define ELF_START_MMAP (0x00000020000000000ULL)
1387
1388 #define elf_check_arch(x) ((x) == EM_TILEGX)
1389
1390 #define ELF_CLASS ELFCLASS64
1391 #define ELF_DATA ELFDATA2LSB
1392 #define ELF_ARCH EM_TILEGX
1393
1394 static inline void init_thread(struct target_pt_regs *regs,
1395 struct image_info *infop)
1396 {
1397 regs->pc = infop->entry;
1398 regs->sp = infop->start_stack;
1399
1400 }
1401
1402 #define ELF_EXEC_PAGESIZE 65536 /* TILE-Gx page size is 64KB */
1403
1404 #endif /* TARGET_TILEGX */
1405
1406 #ifdef TARGET_RISCV
1407
1408 #define ELF_START_MMAP 0x80000000
1409 #define ELF_ARCH EM_RISCV
1410
1411 #ifdef TARGET_RISCV32
1412 #define ELF_CLASS ELFCLASS32
1413 #else
1414 #define ELF_CLASS ELFCLASS64
1415 #endif
1416
1417 static inline void init_thread(struct target_pt_regs *regs,
1418 struct image_info *infop)
1419 {
1420 regs->sepc = infop->entry;
1421 regs->sp = infop->start_stack;
1422 }
1423
1424 #define ELF_EXEC_PAGESIZE 4096
1425
1426 #endif /* TARGET_RISCV */
1427
1428 #ifdef TARGET_HPPA
1429
1430 #define ELF_START_MMAP 0x80000000
1431 #define ELF_CLASS ELFCLASS32
1432 #define ELF_ARCH EM_PARISC
1433 #define ELF_PLATFORM "PARISC"
1434 #define STACK_GROWS_DOWN 0
1435 #define STACK_ALIGNMENT 64
1436
1437 static inline void init_thread(struct target_pt_regs *regs,
1438 struct image_info *infop)
1439 {
1440 regs->iaoq[0] = infop->entry;
1441 regs->iaoq[1] = infop->entry + 4;
1442 regs->gr[23] = 0;
1443 regs->gr[24] = infop->arg_start;
1444 regs->gr[25] = (infop->arg_end - infop->arg_start) / sizeof(abi_ulong);
1445 /* The top-of-stack contains a linkage buffer. */
1446 regs->gr[30] = infop->start_stack + 64;
1447 regs->gr[31] = infop->entry;
1448 }
1449
1450 #endif /* TARGET_HPPA */
1451
1452 #ifdef TARGET_XTENSA
1453
1454 #define ELF_START_MMAP 0x20000000
1455
1456 #define ELF_CLASS ELFCLASS32
1457 #define ELF_ARCH EM_XTENSA
1458
1459 static inline void init_thread(struct target_pt_regs *regs,
1460 struct image_info *infop)
1461 {
1462 regs->windowbase = 0;
1463 regs->windowstart = 1;
1464 regs->areg[1] = infop->start_stack;
1465 regs->pc = infop->entry;
1466 }
1467
1468 /* See linux kernel: arch/xtensa/include/asm/elf.h. */
1469 #define ELF_NREG 128
1470 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1471
1472 enum {
1473 TARGET_REG_PC,
1474 TARGET_REG_PS,
1475 TARGET_REG_LBEG,
1476 TARGET_REG_LEND,
1477 TARGET_REG_LCOUNT,
1478 TARGET_REG_SAR,
1479 TARGET_REG_WINDOWSTART,
1480 TARGET_REG_WINDOWBASE,
1481 TARGET_REG_THREADPTR,
1482 TARGET_REG_AR0 = 64,
1483 };
1484
1485 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1486 const CPUXtensaState *env)
1487 {
1488 unsigned i;
1489
1490 (*regs)[TARGET_REG_PC] = tswapreg(env->pc);
1491 (*regs)[TARGET_REG_PS] = tswapreg(env->sregs[PS] & ~PS_EXCM);
1492 (*regs)[TARGET_REG_LBEG] = tswapreg(env->sregs[LBEG]);
1493 (*regs)[TARGET_REG_LEND] = tswapreg(env->sregs[LEND]);
1494 (*regs)[TARGET_REG_LCOUNT] = tswapreg(env->sregs[LCOUNT]);
1495 (*regs)[TARGET_REG_SAR] = tswapreg(env->sregs[SAR]);
1496 (*regs)[TARGET_REG_WINDOWSTART] = tswapreg(env->sregs[WINDOW_START]);
1497 (*regs)[TARGET_REG_WINDOWBASE] = tswapreg(env->sregs[WINDOW_BASE]);
1498 (*regs)[TARGET_REG_THREADPTR] = tswapreg(env->uregs[THREADPTR]);
1499 xtensa_sync_phys_from_window((CPUXtensaState *)env);
1500 for (i = 0; i < env->config->nareg; ++i) {
1501 (*regs)[TARGET_REG_AR0 + i] = tswapreg(env->phys_regs[i]);
1502 }
1503 }
1504
1505 #define USE_ELF_CORE_DUMP
1506 #define ELF_EXEC_PAGESIZE 4096
1507
1508 #endif /* TARGET_XTENSA */
1509
1510 #ifndef ELF_PLATFORM
1511 #define ELF_PLATFORM (NULL)
1512 #endif
1513
1514 #ifndef ELF_MACHINE
1515 #define ELF_MACHINE ELF_ARCH
1516 #endif
1517
1518 #ifndef elf_check_arch
1519 #define elf_check_arch(x) ((x) == ELF_ARCH)
1520 #endif
1521
1522 #ifndef ELF_HWCAP
1523 #define ELF_HWCAP 0
1524 #endif
1525
1526 #ifndef STACK_GROWS_DOWN
1527 #define STACK_GROWS_DOWN 1
1528 #endif
1529
1530 #ifndef STACK_ALIGNMENT
1531 #define STACK_ALIGNMENT 16
1532 #endif
1533
1534 #ifdef TARGET_ABI32
1535 #undef ELF_CLASS
1536 #define ELF_CLASS ELFCLASS32
1537 #undef bswaptls
1538 #define bswaptls(ptr) bswap32s(ptr)
1539 #endif
1540
1541 #include "elf.h"
1542
1543 struct exec
1544 {
1545 unsigned int a_info; /* Use macros N_MAGIC, etc for access */
1546 unsigned int a_text; /* length of text, in bytes */
1547 unsigned int a_data; /* length of data, in bytes */
1548 unsigned int a_bss; /* length of uninitialized data area, in bytes */
1549 unsigned int a_syms; /* length of symbol table data in file, in bytes */
1550 unsigned int a_entry; /* start address */
1551 unsigned int a_trsize; /* length of relocation info for text, in bytes */
1552 unsigned int a_drsize; /* length of relocation info for data, in bytes */
1553 };
1554
1555
1556 #define N_MAGIC(exec) ((exec).a_info & 0xffff)
1557 #define OMAGIC 0407
1558 #define NMAGIC 0410
1559 #define ZMAGIC 0413
1560 #define QMAGIC 0314
1561
1562 /* Necessary parameters */
1563 #define TARGET_ELF_EXEC_PAGESIZE \
1564 (((eppnt->p_align & ~qemu_host_page_mask) != 0) ? \
1565 TARGET_PAGE_SIZE : MAX(qemu_host_page_size, TARGET_PAGE_SIZE))
1566 #define TARGET_ELF_PAGELENGTH(_v) ROUND_UP((_v), TARGET_ELF_EXEC_PAGESIZE)
1567 #define TARGET_ELF_PAGESTART(_v) ((_v) & \
1568 ~(abi_ulong)(TARGET_ELF_EXEC_PAGESIZE-1))
1569 #define TARGET_ELF_PAGEOFFSET(_v) ((_v) & (TARGET_ELF_EXEC_PAGESIZE-1))
1570
1571 #define DLINFO_ITEMS 15
1572
1573 static inline void memcpy_fromfs(void * to, const void * from, unsigned long n)
1574 {
1575 memcpy(to, from, n);
1576 }
1577
1578 #ifdef BSWAP_NEEDED
1579 static void bswap_ehdr(struct elfhdr *ehdr)
1580 {
1581 bswap16s(&ehdr->e_type); /* Object file type */
1582 bswap16s(&ehdr->e_machine); /* Architecture */
1583 bswap32s(&ehdr->e_version); /* Object file version */
1584 bswaptls(&ehdr->e_entry); /* Entry point virtual address */
1585 bswaptls(&ehdr->e_phoff); /* Program header table file offset */
1586 bswaptls(&ehdr->e_shoff); /* Section header table file offset */
1587 bswap32s(&ehdr->e_flags); /* Processor-specific flags */
1588 bswap16s(&ehdr->e_ehsize); /* ELF header size in bytes */
1589 bswap16s(&ehdr->e_phentsize); /* Program header table entry size */
1590 bswap16s(&ehdr->e_phnum); /* Program header table entry count */
1591 bswap16s(&ehdr->e_shentsize); /* Section header table entry size */
1592 bswap16s(&ehdr->e_shnum); /* Section header table entry count */
1593 bswap16s(&ehdr->e_shstrndx); /* Section header string table index */
1594 }
1595
1596 static void bswap_phdr(struct elf_phdr *phdr, int phnum)
1597 {
1598 int i;
1599 for (i = 0; i < phnum; ++i, ++phdr) {
1600 bswap32s(&phdr->p_type); /* Segment type */
1601 bswap32s(&phdr->p_flags); /* Segment flags */
1602 bswaptls(&phdr->p_offset); /* Segment file offset */
1603 bswaptls(&phdr->p_vaddr); /* Segment virtual address */
1604 bswaptls(&phdr->p_paddr); /* Segment physical address */
1605 bswaptls(&phdr->p_filesz); /* Segment size in file */
1606 bswaptls(&phdr->p_memsz); /* Segment size in memory */
1607 bswaptls(&phdr->p_align); /* Segment alignment */
1608 }
1609 }
1610
1611 static void bswap_shdr(struct elf_shdr *shdr, int shnum)
1612 {
1613 int i;
1614 for (i = 0; i < shnum; ++i, ++shdr) {
1615 bswap32s(&shdr->sh_name);
1616 bswap32s(&shdr->sh_type);
1617 bswaptls(&shdr->sh_flags);
1618 bswaptls(&shdr->sh_addr);
1619 bswaptls(&shdr->sh_offset);
1620 bswaptls(&shdr->sh_size);
1621 bswap32s(&shdr->sh_link);
1622 bswap32s(&shdr->sh_info);
1623 bswaptls(&shdr->sh_addralign);
1624 bswaptls(&shdr->sh_entsize);
1625 }
1626 }
1627
1628 static void bswap_sym(struct elf_sym *sym)
1629 {
1630 bswap32s(&sym->st_name);
1631 bswaptls(&sym->st_value);
1632 bswaptls(&sym->st_size);
1633 bswap16s(&sym->st_shndx);
1634 }
1635
1636 #ifdef TARGET_MIPS
1637 static void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags)
1638 {
1639 bswap16s(&abiflags->version);
1640 bswap32s(&abiflags->ases);
1641 bswap32s(&abiflags->isa_ext);
1642 bswap32s(&abiflags->flags1);
1643 bswap32s(&abiflags->flags2);
1644 }
1645 #endif
1646 #else
1647 static inline void bswap_ehdr(struct elfhdr *ehdr) { }
1648 static inline void bswap_phdr(struct elf_phdr *phdr, int phnum) { }
1649 static inline void bswap_shdr(struct elf_shdr *shdr, int shnum) { }
1650 static inline void bswap_sym(struct elf_sym *sym) { }
1651 #ifdef TARGET_MIPS
1652 static inline void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags) { }
1653 #endif
1654 #endif
1655
1656 #ifdef USE_ELF_CORE_DUMP
1657 static int elf_core_dump(int, const CPUArchState *);
1658 #endif /* USE_ELF_CORE_DUMP */
1659 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias);
1660
1661 /* Verify the portions of EHDR within E_IDENT for the target.
1662 This can be performed before bswapping the entire header. */
1663 static bool elf_check_ident(struct elfhdr *ehdr)
1664 {
1665 return (ehdr->e_ident[EI_MAG0] == ELFMAG0
1666 && ehdr->e_ident[EI_MAG1] == ELFMAG1
1667 && ehdr->e_ident[EI_MAG2] == ELFMAG2
1668 && ehdr->e_ident[EI_MAG3] == ELFMAG3
1669 && ehdr->e_ident[EI_CLASS] == ELF_CLASS
1670 && ehdr->e_ident[EI_DATA] == ELF_DATA
1671 && ehdr->e_ident[EI_VERSION] == EV_CURRENT);
1672 }
1673
1674 /* Verify the portions of EHDR outside of E_IDENT for the target.
1675 This has to wait until after bswapping the header. */
1676 static bool elf_check_ehdr(struct elfhdr *ehdr)
1677 {
1678 return (elf_check_arch(ehdr->e_machine)
1679 && ehdr->e_ehsize == sizeof(struct elfhdr)
1680 && ehdr->e_phentsize == sizeof(struct elf_phdr)
1681 && (ehdr->e_type == ET_EXEC || ehdr->e_type == ET_DYN));
1682 }
1683
1684 /*
1685 * 'copy_elf_strings()' copies argument/envelope strings from user
1686 * memory to free pages in kernel mem. These are in a format ready
1687 * to be put directly into the top of new user memory.
1688 *
1689 */
1690 static abi_ulong copy_elf_strings(int argc, char **argv, char *scratch,
1691 abi_ulong p, abi_ulong stack_limit)
1692 {
1693 char *tmp;
1694 int len, i;
1695 abi_ulong top = p;
1696
1697 if (!p) {
1698 return 0; /* bullet-proofing */
1699 }
1700
1701 if (STACK_GROWS_DOWN) {
1702 int offset = ((p - 1) % TARGET_PAGE_SIZE) + 1;
1703 for (i = argc - 1; i >= 0; --i) {
1704 tmp = argv[i];
1705 if (!tmp) {
1706 fprintf(stderr, "VFS: argc is wrong");
1707 exit(-1);
1708 }
1709 len = strlen(tmp) + 1;
1710 tmp += len;
1711
1712 if (len > (p - stack_limit)) {
1713 return 0;
1714 }
1715 while (len) {
1716 int bytes_to_copy = (len > offset) ? offset : len;
1717 tmp -= bytes_to_copy;
1718 p -= bytes_to_copy;
1719 offset -= bytes_to_copy;
1720 len -= bytes_to_copy;
1721
1722 memcpy_fromfs(scratch + offset, tmp, bytes_to_copy);
1723
1724 if (offset == 0) {
1725 memcpy_to_target(p, scratch, top - p);
1726 top = p;
1727 offset = TARGET_PAGE_SIZE;
1728 }
1729 }
1730 }
1731 if (p != top) {
1732 memcpy_to_target(p, scratch + offset, top - p);
1733 }
1734 } else {
1735 int remaining = TARGET_PAGE_SIZE - (p % TARGET_PAGE_SIZE);
1736 for (i = 0; i < argc; ++i) {
1737 tmp = argv[i];
1738 if (!tmp) {
1739 fprintf(stderr, "VFS: argc is wrong");
1740 exit(-1);
1741 }
1742 len = strlen(tmp) + 1;
1743 if (len > (stack_limit - p)) {
1744 return 0;
1745 }
1746 while (len) {
1747 int bytes_to_copy = (len > remaining) ? remaining : len;
1748
1749 memcpy_fromfs(scratch + (p - top), tmp, bytes_to_copy);
1750
1751 tmp += bytes_to_copy;
1752 remaining -= bytes_to_copy;
1753 p += bytes_to_copy;
1754 len -= bytes_to_copy;
1755
1756 if (remaining == 0) {
1757 memcpy_to_target(top, scratch, p - top);
1758 top = p;
1759 remaining = TARGET_PAGE_SIZE;
1760 }
1761 }
1762 }
1763 if (p != top) {
1764 memcpy_to_target(top, scratch, p - top);
1765 }
1766 }
1767
1768 return p;
1769 }
1770
1771 /* Older linux kernels provide up to MAX_ARG_PAGES (default: 32) of
1772 * argument/environment space. Newer kernels (>2.6.33) allow more,
1773 * dependent on stack size, but guarantee at least 32 pages for
1774 * backwards compatibility.
1775 */
1776 #define STACK_LOWER_LIMIT (32 * TARGET_PAGE_SIZE)
1777
1778 static abi_ulong setup_arg_pages(struct linux_binprm *bprm,
1779 struct image_info *info)
1780 {
1781 abi_ulong size, error, guard;
1782
1783 size = guest_stack_size;
1784 if (size < STACK_LOWER_LIMIT) {
1785 size = STACK_LOWER_LIMIT;
1786 }
1787 guard = TARGET_PAGE_SIZE;
1788 if (guard < qemu_real_host_page_size) {
1789 guard = qemu_real_host_page_size;
1790 }
1791
1792 error = target_mmap(0, size + guard, PROT_READ | PROT_WRITE,
1793 MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
1794 if (error == -1) {
1795 perror("mmap stack");
1796 exit(-1);
1797 }
1798
1799 /* We reserve one extra page at the top of the stack as guard. */
1800 if (STACK_GROWS_DOWN) {
1801 target_mprotect(error, guard, PROT_NONE);
1802 info->stack_limit = error + guard;
1803 return info->stack_limit + size - sizeof(void *);
1804 } else {
1805 target_mprotect(error + size, guard, PROT_NONE);
1806 info->stack_limit = error + size;
1807 return error;
1808 }
1809 }
1810
1811 /* Map and zero the bss. We need to explicitly zero any fractional pages
1812 after the data section (i.e. bss). */
1813 static void zero_bss(abi_ulong elf_bss, abi_ulong last_bss, int prot)
1814 {
1815 uintptr_t host_start, host_map_start, host_end;
1816
1817 last_bss = TARGET_PAGE_ALIGN(last_bss);
1818
1819 /* ??? There is confusion between qemu_real_host_page_size and
1820 qemu_host_page_size here and elsewhere in target_mmap, which
1821 may lead to the end of the data section mapping from the file
1822 not being mapped. At least there was an explicit test and
1823 comment for that here, suggesting that "the file size must
1824 be known". The comment probably pre-dates the introduction
1825 of the fstat system call in target_mmap which does in fact
1826 find out the size. What isn't clear is if the workaround
1827 here is still actually needed. For now, continue with it,
1828 but merge it with the "normal" mmap that would allocate the bss. */
1829
1830 host_start = (uintptr_t) g2h(elf_bss);
1831 host_end = (uintptr_t) g2h(last_bss);
1832 host_map_start = REAL_HOST_PAGE_ALIGN(host_start);
1833
1834 if (host_map_start < host_end) {
1835 void *p = mmap((void *)host_map_start, host_end - host_map_start,
1836 prot, MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
1837 if (p == MAP_FAILED) {
1838 perror("cannot mmap brk");
1839 exit(-1);
1840 }
1841 }
1842
1843 /* Ensure that the bss page(s) are valid */
1844 if ((page_get_flags(last_bss-1) & prot) != prot) {
1845 page_set_flags(elf_bss & TARGET_PAGE_MASK, last_bss, prot | PAGE_VALID);
1846 }
1847
1848 if (host_start < host_map_start) {
1849 memset((void *)host_start, 0, host_map_start - host_start);
1850 }
1851 }
1852
1853 #ifdef TARGET_ARM
1854 static int elf_is_fdpic(struct elfhdr *exec)
1855 {
1856 return exec->e_ident[EI_OSABI] == ELFOSABI_ARM_FDPIC;
1857 }
1858 #else
1859 /* Default implementation, always false. */
1860 static int elf_is_fdpic(struct elfhdr *exec)
1861 {
1862 return 0;
1863 }
1864 #endif
1865
1866 static abi_ulong loader_build_fdpic_loadmap(struct image_info *info, abi_ulong sp)
1867 {
1868 uint16_t n;
1869 struct elf32_fdpic_loadseg *loadsegs = info->loadsegs;
1870
1871 /* elf32_fdpic_loadseg */
1872 n = info->nsegs;
1873 while (n--) {
1874 sp -= 12;
1875 put_user_u32(loadsegs[n].addr, sp+0);
1876 put_user_u32(loadsegs[n].p_vaddr, sp+4);
1877 put_user_u32(loadsegs[n].p_memsz, sp+8);
1878 }
1879
1880 /* elf32_fdpic_loadmap */
1881 sp -= 4;
1882 put_user_u16(0, sp+0); /* version */
1883 put_user_u16(info->nsegs, sp+2); /* nsegs */
1884
1885 info->personality = PER_LINUX_FDPIC;
1886 info->loadmap_addr = sp;
1887
1888 return sp;
1889 }
1890
1891 static abi_ulong create_elf_tables(abi_ulong p, int argc, int envc,
1892 struct elfhdr *exec,
1893 struct image_info *info,
1894 struct image_info *interp_info)
1895 {
1896 abi_ulong sp;
1897 abi_ulong u_argc, u_argv, u_envp, u_auxv;
1898 int size;
1899 int i;
1900 abi_ulong u_rand_bytes;
1901 uint8_t k_rand_bytes[16];
1902 abi_ulong u_platform;
1903 const char *k_platform;
1904 const int n = sizeof(elf_addr_t);
1905
1906 sp = p;
1907
1908 /* Needs to be before we load the env/argc/... */
1909 if (elf_is_fdpic(exec)) {
1910 /* Need 4 byte alignment for these structs */
1911 sp &= ~3;
1912 sp = loader_build_fdpic_loadmap(info, sp);
1913 info->other_info = interp_info;
1914 if (interp_info) {
1915 interp_info->other_info = info;
1916 sp = loader_build_fdpic_loadmap(interp_info, sp);
1917 info->interpreter_loadmap_addr = interp_info->loadmap_addr;
1918 info->interpreter_pt_dynamic_addr = interp_info->pt_dynamic_addr;
1919 } else {
1920 info->interpreter_loadmap_addr = 0;
1921 info->interpreter_pt_dynamic_addr = 0;
1922 }
1923 }
1924
1925 u_platform = 0;
1926 k_platform = ELF_PLATFORM;
1927 if (k_platform) {
1928 size_t len = strlen(k_platform) + 1;
1929 if (STACK_GROWS_DOWN) {
1930 sp -= (len + n - 1) & ~(n - 1);
1931 u_platform = sp;
1932 /* FIXME - check return value of memcpy_to_target() for failure */
1933 memcpy_to_target(sp, k_platform, len);
1934 } else {
1935 memcpy_to_target(sp, k_platform, len);
1936 u_platform = sp;
1937 sp += len + 1;
1938 }
1939 }
1940
1941 /* Provide 16 byte alignment for the PRNG, and basic alignment for
1942 * the argv and envp pointers.
1943 */
1944 if (STACK_GROWS_DOWN) {
1945 sp = QEMU_ALIGN_DOWN(sp, 16);
1946 } else {
1947 sp = QEMU_ALIGN_UP(sp, 16);
1948 }
1949
1950 /*
1951 * Generate 16 random bytes for userspace PRNG seeding.
1952 */
1953 qemu_guest_getrandom_nofail(k_rand_bytes, sizeof(k_rand_bytes));
1954 if (STACK_GROWS_DOWN) {
1955 sp -= 16;
1956 u_rand_bytes = sp;
1957 /* FIXME - check return value of memcpy_to_target() for failure */
1958 memcpy_to_target(sp, k_rand_bytes, 16);
1959 } else {
1960 memcpy_to_target(sp, k_rand_bytes, 16);
1961 u_rand_bytes = sp;
1962 sp += 16;
1963 }
1964
1965 size = (DLINFO_ITEMS + 1) * 2;
1966 if (k_platform)
1967 size += 2;
1968 #ifdef DLINFO_ARCH_ITEMS
1969 size += DLINFO_ARCH_ITEMS * 2;
1970 #endif
1971 #ifdef ELF_HWCAP2
1972 size += 2;
1973 #endif
1974 info->auxv_len = size * n;
1975
1976 size += envc + argc + 2;
1977 size += 1; /* argc itself */
1978 size *= n;
1979
1980 /* Allocate space and finalize stack alignment for entry now. */
1981 if (STACK_GROWS_DOWN) {
1982 u_argc = QEMU_ALIGN_DOWN(sp - size, STACK_ALIGNMENT);
1983 sp = u_argc;
1984 } else {
1985 u_argc = sp;
1986 sp = QEMU_ALIGN_UP(sp + size, STACK_ALIGNMENT);
1987 }
1988
1989 u_argv = u_argc + n;
1990 u_envp = u_argv + (argc + 1) * n;
1991 u_auxv = u_envp + (envc + 1) * n;
1992 info->saved_auxv = u_auxv;
1993 info->arg_start = u_argv;
1994 info->arg_end = u_argv + argc * n;
1995
1996 /* This is correct because Linux defines
1997 * elf_addr_t as Elf32_Off / Elf64_Off
1998 */
1999 #define NEW_AUX_ENT(id, val) do { \
2000 put_user_ual(id, u_auxv); u_auxv += n; \
2001 put_user_ual(val, u_auxv); u_auxv += n; \
2002 } while(0)
2003
2004 #ifdef ARCH_DLINFO
2005 /*
2006 * ARCH_DLINFO must come first so platform specific code can enforce
2007 * special alignment requirements on the AUXV if necessary (eg. PPC).
2008 */
2009 ARCH_DLINFO;
2010 #endif
2011 /* There must be exactly DLINFO_ITEMS entries here, or the assert
2012 * on info->auxv_len will trigger.
2013 */
2014 NEW_AUX_ENT(AT_PHDR, (abi_ulong)(info->load_addr + exec->e_phoff));
2015 NEW_AUX_ENT(AT_PHENT, (abi_ulong)(sizeof (struct elf_phdr)));
2016 NEW_AUX_ENT(AT_PHNUM, (abi_ulong)(exec->e_phnum));
2017 if ((info->alignment & ~qemu_host_page_mask) != 0) {
2018 /* Target doesn't support host page size alignment */
2019 NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(TARGET_PAGE_SIZE));
2020 } else {
2021 NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(MAX(TARGET_PAGE_SIZE,
2022 qemu_host_page_size)));
2023 }
2024 NEW_AUX_ENT(AT_BASE, (abi_ulong)(interp_info ? interp_info->load_addr : 0));
2025 NEW_AUX_ENT(AT_FLAGS, (abi_ulong)0);
2026 NEW_AUX_ENT(AT_ENTRY, info->entry);
2027 NEW_AUX_ENT(AT_UID, (abi_ulong) getuid());
2028 NEW_AUX_ENT(AT_EUID, (abi_ulong) geteuid());
2029 NEW_AUX_ENT(AT_GID, (abi_ulong) getgid());
2030 NEW_AUX_ENT(AT_EGID, (abi_ulong) getegid());
2031 NEW_AUX_ENT(AT_HWCAP, (abi_ulong) ELF_HWCAP);
2032 NEW_AUX_ENT(AT_CLKTCK, (abi_ulong) sysconf(_SC_CLK_TCK));
2033 NEW_AUX_ENT(AT_RANDOM, (abi_ulong) u_rand_bytes);
2034 NEW_AUX_ENT(AT_SECURE, (abi_ulong) qemu_getauxval(AT_SECURE));
2035
2036 #ifdef ELF_HWCAP2
2037 NEW_AUX_ENT(AT_HWCAP2, (abi_ulong) ELF_HWCAP2);
2038 #endif
2039
2040 if (u_platform) {
2041 NEW_AUX_ENT(AT_PLATFORM, u_platform);
2042 }
2043 NEW_AUX_ENT (AT_NULL, 0);
2044 #undef NEW_AUX_ENT
2045
2046 /* Check that our initial calculation of the auxv length matches how much
2047 * we actually put into it.
2048 */
2049 assert(info->auxv_len == u_auxv - info->saved_auxv);
2050
2051 put_user_ual(argc, u_argc);
2052
2053 p = info->arg_strings;
2054 for (i = 0; i < argc; ++i) {
2055 put_user_ual(p, u_argv);
2056 u_argv += n;
2057 p += target_strlen(p) + 1;
2058 }
2059 put_user_ual(0, u_argv);
2060
2061 p = info->env_strings;
2062 for (i = 0; i < envc; ++i) {
2063 put_user_ual(p, u_envp);
2064 u_envp += n;
2065 p += target_strlen(p) + 1;
2066 }
2067 put_user_ual(0, u_envp);
2068
2069 return sp;
2070 }
2071
2072 unsigned long init_guest_space(unsigned long host_start,
2073 unsigned long host_size,
2074 unsigned long guest_start,
2075 bool fixed)
2076 {
2077 /* In order to use host shmat, we must be able to honor SHMLBA. */
2078 unsigned long align = MAX(SHMLBA, qemu_host_page_size);
2079 unsigned long current_start, aligned_start;
2080 int flags;
2081
2082 assert(host_start || host_size);
2083
2084 /* If just a starting address is given, then just verify that
2085 * address. */
2086 if (host_start && !host_size) {
2087 #if defined(TARGET_ARM) && !defined(TARGET_AARCH64)
2088 if (init_guest_commpage(host_start, host_size) != 1) {
2089 return (unsigned long)-1;
2090 }
2091 #endif
2092 return host_start;
2093 }
2094
2095 /* Setup the initial flags and start address. */
2096 current_start = host_start & -align;
2097 flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE;
2098 if (fixed) {
2099 flags |= MAP_FIXED;
2100 }
2101
2102 /* Otherwise, a non-zero size region of memory needs to be mapped
2103 * and validated. */
2104
2105 #if defined(TARGET_ARM) && !defined(TARGET_AARCH64)
2106 /* On 32-bit ARM, we need to map not just the usable memory, but
2107 * also the commpage. Try to find a suitable place by allocating
2108 * a big chunk for all of it. If host_start, then the naive
2109 * strategy probably does good enough.
2110 */
2111 if (!host_start) {
2112 unsigned long guest_full_size, host_full_size, real_start;
2113
2114 guest_full_size =
2115 (0xffff0f00 & qemu_host_page_mask) + qemu_host_page_size;
2116 host_full_size = guest_full_size - guest_start;
2117 real_start = (unsigned long)
2118 mmap(NULL, host_full_size, PROT_NONE, flags, -1, 0);
2119 if (real_start == (unsigned long)-1) {
2120 if (host_size < host_full_size - qemu_host_page_size) {
2121 /* We failed to map a continous segment, but we're
2122 * allowed to have a gap between the usable memory and
2123 * the commpage where other things can be mapped.
2124 * This sparseness gives us more flexibility to find
2125 * an address range.
2126 */
2127 goto naive;
2128 }
2129 return (unsigned long)-1;
2130 }
2131 munmap((void *)real_start, host_full_size);
2132 if (real_start & (align - 1)) {
2133 /* The same thing again, but with extra
2134 * so that we can shift around alignment.
2135 */
2136 unsigned long real_size = host_full_size + qemu_host_page_size;
2137 real_start = (unsigned long)
2138 mmap(NULL, real_size, PROT_NONE, flags, -1, 0);
2139 if (real_start == (unsigned long)-1) {
2140 if (host_size < host_full_size - qemu_host_page_size) {
2141 goto naive;
2142 }
2143 return (unsigned long)-1;
2144 }
2145 munmap((void *)real_start, real_size);
2146 real_start = ROUND_UP(real_start, align);
2147 }
2148 current_start = real_start;
2149 }
2150 naive:
2151 #endif
2152
2153 while (1) {
2154 unsigned long real_start, real_size, aligned_size;
2155 aligned_size = real_size = host_size;
2156
2157 /* Do not use mmap_find_vma here because that is limited to the
2158 * guest address space. We are going to make the
2159 * guest address space fit whatever we're given.
2160 */
2161 real_start = (unsigned long)
2162 mmap((void *)current_start, host_size, PROT_NONE, flags, -1, 0);
2163 if (real_start == (unsigned long)-1) {
2164 return (unsigned long)-1;
2165 }
2166
2167 /* Check to see if the address is valid. */
2168 if (host_start && real_start != current_start) {
2169 goto try_again;
2170 }
2171
2172 /* Ensure the address is properly aligned. */
2173 if (real_start & (align - 1)) {
2174 /* Ideally, we adjust like
2175 *
2176 * pages: [ ][ ][ ][ ][ ]
2177 * old: [ real ]
2178 * [ aligned ]
2179 * new: [ real ]
2180 * [ aligned ]
2181 *
2182 * But if there is something else mapped right after it,
2183 * then obviously it won't have room to grow, and the
2184 * kernel will put the new larger real someplace else with
2185 * unknown alignment (if we made it to here, then
2186 * fixed=false). Which is why we grow real by a full page
2187 * size, instead of by part of one; so that even if we get
2188 * moved, we can still guarantee alignment. But this does
2189 * mean that there is a padding of < 1 page both before
2190 * and after the aligned range; the "after" could could
2191 * cause problems for ARM emulation where it could butt in
2192 * to where we need to put the commpage.
2193 */
2194 munmap((void *)real_start, host_size);
2195 real_size = aligned_size + align;
2196 real_start = (unsigned long)
2197 mmap((void *)real_start, real_size, PROT_NONE, flags, -1, 0);
2198 if (real_start == (unsigned long)-1) {
2199 return (unsigned long)-1;
2200 }
2201 aligned_start = ROUND_UP(real_start, align);
2202 } else {
2203 aligned_start = real_start;
2204 }
2205
2206 #if defined(TARGET_ARM) && !defined(TARGET_AARCH64)
2207 /* On 32-bit ARM, we need to also be able to map the commpage. */
2208 int valid = init_guest_commpage(aligned_start - guest_start,
2209 aligned_size + guest_start);
2210 if (valid == -1) {
2211 munmap((void *)real_start, real_size);
2212 return (unsigned long)-1;
2213 } else if (valid == 0) {
2214 goto try_again;
2215 }
2216 #endif
2217
2218 /* If nothing has said `return -1` or `goto try_again` yet,
2219 * then the address we have is good.
2220 */
2221 break;
2222
2223 try_again:
2224 /* That address didn't work. Unmap and try a different one.
2225 * The address the host picked because is typically right at
2226 * the top of the host address space and leaves the guest with
2227 * no usable address space. Resort to a linear search. We
2228 * already compensated for mmap_min_addr, so this should not
2229 * happen often. Probably means we got unlucky and host
2230 * address space randomization put a shared library somewhere
2231 * inconvenient.
2232 *
2233 * This is probably a good strategy if host_start, but is
2234 * probably a bad strategy if not, which means we got here
2235 * because of trouble with ARM commpage setup.
2236 */
2237 munmap((void *)real_start, real_size);
2238 current_start += align;
2239 if (host_start == current_start) {
2240 /* Theoretically possible if host doesn't have any suitably
2241 * aligned areas. Normally the first mmap will fail.
2242 */
2243 return (unsigned long)-1;
2244 }
2245 }
2246
2247 qemu_log_mask(CPU_LOG_PAGE, "Reserved 0x%lx bytes of guest address space\n", host_size);
2248
2249 return aligned_start;
2250 }
2251
2252 static void probe_guest_base(const char *image_name,
2253 abi_ulong loaddr, abi_ulong hiaddr)
2254 {
2255 /* Probe for a suitable guest base address, if the user has not set
2256 * it explicitly, and set guest_base appropriately.
2257 * In case of error we will print a suitable message and exit.
2258 */
2259 const char *errmsg;
2260 if (!have_guest_base && !reserved_va) {
2261 unsigned long host_start, real_start, host_size;
2262
2263 /* Round addresses to page boundaries. */
2264 loaddr &= qemu_host_page_mask;
2265 hiaddr = HOST_PAGE_ALIGN(hiaddr);
2266
2267 if (loaddr < mmap_min_addr) {
2268 host_start = HOST_PAGE_ALIGN(mmap_min_addr);
2269 } else {
2270 host_start = loaddr;
2271 if (host_start != loaddr) {
2272 errmsg = "Address overflow loading ELF binary";
2273 goto exit_errmsg;
2274 }
2275 }
2276 host_size = hiaddr - loaddr;
2277
2278 /* Setup the initial guest memory space with ranges gleaned from
2279 * the ELF image that is being loaded.
2280 */
2281 real_start = init_guest_space(host_start, host_size, loaddr, false);
2282 if (real_start == (unsigned long)-1) {
2283 errmsg = "Unable to find space for application";
2284 goto exit_errmsg;
2285 }
2286 guest_base = real_start - loaddr;
2287
2288 qemu_log_mask(CPU_LOG_PAGE, "Relocating guest address space from 0x"
2289 TARGET_ABI_FMT_lx " to 0x%lx\n",
2290 loaddr, real_start);
2291 }
2292 return;
2293
2294 exit_errmsg:
2295 fprintf(stderr, "%s: %s\n", image_name, errmsg);
2296 exit(-1);
2297 }
2298
2299
2300 /* Load an ELF image into the address space.
2301
2302 IMAGE_NAME is the filename of the image, to use in error messages.
2303 IMAGE_FD is the open file descriptor for the image.
2304
2305 BPRM_BUF is a copy of the beginning of the file; this of course
2306 contains the elf file header at offset 0. It is assumed that this
2307 buffer is sufficiently aligned to present no problems to the host
2308 in accessing data at aligned offsets within the buffer.
2309
2310 On return: INFO values will be filled in, as necessary or available. */
2311
2312 static void load_elf_image(const char *image_name, int image_fd,
2313 struct image_info *info, char **pinterp_name,
2314 char bprm_buf[BPRM_BUF_SIZE])
2315 {
2316 struct elfhdr *ehdr = (struct elfhdr *)bprm_buf;
2317 struct elf_phdr *phdr;
2318 abi_ulong load_addr, load_bias, loaddr, hiaddr, error;
2319 int i, retval;
2320 const char *errmsg;
2321
2322 /* First of all, some simple consistency checks */
2323 errmsg = "Invalid ELF image for this architecture";
2324 if (!elf_check_ident(ehdr)) {
2325 goto exit_errmsg;
2326 }
2327 bswap_ehdr(ehdr);
2328 if (!elf_check_ehdr(ehdr)) {
2329 goto exit_errmsg;
2330 }
2331
2332 i = ehdr->e_phnum * sizeof(struct elf_phdr);
2333 if (ehdr->e_phoff + i <= BPRM_BUF_SIZE) {
2334 phdr = (struct elf_phdr *)(bprm_buf + ehdr->e_phoff);
2335 } else {
2336 phdr = (struct elf_phdr *) alloca(i);
2337 retval = pread(image_fd, phdr, i, ehdr->e_phoff);
2338 if (retval != i) {
2339 goto exit_read;
2340 }
2341 }
2342 bswap_phdr(phdr, ehdr->e_phnum);
2343
2344 info->nsegs = 0;
2345 info->pt_dynamic_addr = 0;
2346
2347 mmap_lock();
2348
2349 /* Find the maximum size of the image and allocate an appropriate
2350 amount of memory to handle that. */
2351 loaddr = -1, hiaddr = 0;
2352 info->alignment = 0;
2353 for (i = 0; i < ehdr->e_phnum; ++i) {
2354 if (phdr[i].p_type == PT_LOAD) {
2355 abi_ulong a = phdr[i].p_vaddr - phdr[i].p_offset;
2356 if (a < loaddr) {
2357 loaddr = a;
2358 }
2359 a = phdr[i].p_vaddr + phdr[i].p_memsz;
2360 if (a > hiaddr) {
2361 hiaddr = a;
2362 }
2363 ++info->nsegs;
2364 info->alignment |= phdr[i].p_align;
2365 }
2366 }
2367
2368 if (pinterp_name != NULL) {
2369 /*
2370 * This is the main executable.
2371 *
2372 * Reserve extra space for brk.
2373 * We hold on to this space while placing the interpreter
2374 * and the stack, lest they be placed immediately after
2375 * the data segment and block allocation from the brk.
2376 *
2377 * 16MB is chosen as "large enough" without being so large
2378 * as to allow the result to not fit with a 32-bit guest on
2379 * a 32-bit host.
2380 */
2381 info->reserve_brk = 16 * MiB;
2382 hiaddr += info->reserve_brk;
2383
2384 if (ehdr->e_type == ET_EXEC) {
2385 /*
2386 * Make sure that the low address does not conflict with
2387 * MMAP_MIN_ADDR or the QEMU application itself.
2388 */
2389 probe_guest_base(image_name, loaddr, hiaddr);
2390 }
2391 }
2392
2393 /*
2394 * Reserve address space for all of this.
2395 *
2396 * In the case of ET_EXEC, we supply MAP_FIXED so that we get
2397 * exactly the address range that is required.
2398 *
2399 * Otherwise this is ET_DYN, and we are searching for a location
2400 * that can hold the memory space required. If the image is
2401 * pre-linked, LOADDR will be non-zero, and the kernel should
2402 * honor that address if it happens to be free.
2403 *
2404 * In both cases, we will overwrite pages in this range with mappings
2405 * from the executable.
2406 */
2407 load_addr = target_mmap(loaddr, hiaddr - loaddr, PROT_NONE,
2408 MAP_PRIVATE | MAP_ANON | MAP_NORESERVE |
2409 (ehdr->e_type == ET_EXEC ? MAP_FIXED : 0),
2410 -1, 0);
2411 if (load_addr == -1) {
2412 goto exit_perror;
2413 }
2414 load_bias = load_addr - loaddr;
2415
2416 if (elf_is_fdpic(ehdr)) {
2417 struct elf32_fdpic_loadseg *loadsegs = info->loadsegs =
2418 g_malloc(sizeof(*loadsegs) * info->nsegs);
2419
2420 for (i = 0; i < ehdr->e_phnum; ++i) {
2421 switch (phdr[i].p_type) {
2422 case PT_DYNAMIC:
2423 info->pt_dynamic_addr = phdr[i].p_vaddr + load_bias;
2424 break;
2425 case PT_LOAD:
2426 loadsegs->addr = phdr[i].p_vaddr + load_bias;
2427 loadsegs->p_vaddr = phdr[i].p_vaddr;
2428 loadsegs->p_memsz = phdr[i].p_memsz;
2429 ++loadsegs;
2430 break;
2431 }
2432 }
2433 }
2434
2435 info->load_bias = load_bias;
2436 info->code_offset = load_bias;
2437 info->data_offset = load_bias;
2438 info->load_addr = load_addr;
2439 info->entry = ehdr->e_entry + load_bias;
2440 info->start_code = -1;
2441 info->end_code = 0;
2442 info->start_data = -1;
2443 info->end_data = 0;
2444 info->brk = 0;
2445 info->elf_flags = ehdr->e_flags;
2446
2447 for (i = 0; i < ehdr->e_phnum; i++) {
2448 struct elf_phdr *eppnt = phdr + i;
2449 if (eppnt->p_type == PT_LOAD) {
2450 abi_ulong vaddr, vaddr_po, vaddr_ps, vaddr_ef, vaddr_em, vaddr_len;
2451 int elf_prot = 0;
2452
2453 if (eppnt->p_flags & PF_R) elf_prot = PROT_READ;
2454 if (eppnt->p_flags & PF_W) elf_prot |= PROT_WRITE;
2455 if (eppnt->p_flags & PF_X) elf_prot |= PROT_EXEC;
2456
2457 vaddr = load_bias + eppnt->p_vaddr;
2458 vaddr_po = TARGET_ELF_PAGEOFFSET(vaddr);
2459 vaddr_ps = TARGET_ELF_PAGESTART(vaddr);
2460 vaddr_len = TARGET_ELF_PAGELENGTH(eppnt->p_filesz + vaddr_po);
2461
2462 /*
2463 * Some segments may be completely empty without any backing file
2464 * segment, in that case just let zero_bss allocate an empty buffer
2465 * for it.
2466 */
2467 if (eppnt->p_filesz != 0) {
2468 error = target_mmap(vaddr_ps, vaddr_len, elf_prot,
2469 MAP_PRIVATE | MAP_FIXED,
2470 image_fd, eppnt->p_offset - vaddr_po);
2471
2472 if (error == -1) {
2473 goto exit_perror;
2474 }
2475 }
2476
2477 vaddr_ef = vaddr + eppnt->p_filesz;
2478 vaddr_em = vaddr + eppnt->p_memsz;
2479
2480 /* If the load segment requests extra zeros (e.g. bss), map it. */
2481 if (vaddr_ef < vaddr_em) {
2482 zero_bss(vaddr_ef, vaddr_em, elf_prot);
2483 }
2484
2485 /* Find the full program boundaries. */
2486 if (elf_prot & PROT_EXEC) {
2487 if (vaddr < info->start_code) {
2488 info->start_code = vaddr;
2489 }
2490 if (vaddr_ef > info->end_code) {
2491 info->end_code = vaddr_ef;
2492 }
2493 }
2494 if (elf_prot & PROT_WRITE) {
2495 if (vaddr < info->start_data) {
2496 info->start_data = vaddr;
2497 }
2498 if (vaddr_ef > info->end_data) {
2499 info->end_data = vaddr_ef;
2500 }
2501 if (vaddr_em > info->brk) {
2502 info->brk = vaddr_em;
2503 }
2504 }
2505 } else if (eppnt->p_type == PT_INTERP && pinterp_name) {
2506 char *interp_name;
2507
2508 if (*pinterp_name) {
2509 errmsg = "Multiple PT_INTERP entries";
2510 goto exit_errmsg;
2511 }
2512 interp_name = malloc(eppnt->p_filesz);
2513 if (!interp_name) {
2514 goto exit_perror;
2515 }
2516
2517 if (eppnt->p_offset + eppnt->p_filesz <= BPRM_BUF_SIZE) {
2518 memcpy(interp_name, bprm_buf + eppnt->p_offset,
2519 eppnt->p_filesz);
2520 } else {
2521 retval = pread(image_fd, interp_name, eppnt->p_filesz,
2522 eppnt->p_offset);
2523 if (retval != eppnt->p_filesz) {
2524 goto exit_perror;
2525 }
2526 }
2527 if (interp_name[eppnt->p_filesz - 1] != 0) {
2528 errmsg = "Invalid PT_INTERP entry";
2529 goto exit_errmsg;
2530 }
2531 *pinterp_name = interp_name;
2532 #ifdef TARGET_MIPS
2533 } else if (eppnt->p_type == PT_MIPS_ABIFLAGS) {
2534 Mips_elf_abiflags_v0 abiflags;
2535 if (eppnt->p_filesz < sizeof(Mips_elf_abiflags_v0)) {
2536 errmsg = "Invalid PT_MIPS_ABIFLAGS entry";
2537 goto exit_errmsg;
2538 }
2539 if (eppnt->p_offset + eppnt->p_filesz <= BPRM_BUF_SIZE) {
2540 memcpy(&abiflags, bprm_buf + eppnt->p_offset,
2541 sizeof(Mips_elf_abiflags_v0));
2542 } else {
2543 retval = pread(image_fd, &abiflags, sizeof(Mips_elf_abiflags_v0),
2544 eppnt->p_offset);
2545 if (retval != sizeof(Mips_elf_abiflags_v0)) {
2546 goto exit_perror;
2547 }
2548 }
2549 bswap_mips_abiflags(&abiflags);
2550 info->fp_abi = abiflags.fp_abi;
2551 #endif
2552 }
2553 }
2554
2555 if (info->end_data == 0) {
2556 info->start_data = info->end_code;
2557 info->end_data = info->end_code;
2558 info->brk = info->end_code;
2559 }
2560
2561 if (qemu_log_enabled()) {
2562 load_symbols(ehdr, image_fd, load_bias);
2563 }
2564
2565 mmap_unlock();
2566
2567 close(image_fd);
2568 return;
2569
2570 exit_read:
2571 if (retval >= 0) {
2572 errmsg = "Incomplete read of file header";
2573 goto exit_errmsg;
2574 }
2575 exit_perror:
2576 errmsg = strerror(errno);
2577 exit_errmsg:
2578 fprintf(stderr, "%s: %s\n", image_name, errmsg);
2579 exit(-1);
2580 }
2581
2582 static void load_elf_interp(const char *filename, struct image_info *info,
2583 char bprm_buf[BPRM_BUF_SIZE])
2584 {
2585 int fd, retval;
2586
2587 fd = open(path(filename), O_RDONLY);
2588 if (fd < 0) {
2589 goto exit_perror;
2590 }
2591
2592 retval = read(fd, bprm_buf, BPRM_BUF_SIZE);
2593 if (retval < 0) {
2594 goto exit_perror;
2595 }
2596 if (retval < BPRM_BUF_SIZE) {
2597 memset(bprm_buf + retval, 0, BPRM_BUF_SIZE - retval);
2598 }
2599
2600 load_elf_image(filename, fd, info, NULL, bprm_buf);
2601 return;
2602
2603 exit_perror:
2604 fprintf(stderr, "%s: %s\n", filename, strerror(errno));
2605 exit(-1);
2606 }
2607
2608 static int symfind(const void *s0, const void *s1)
2609 {
2610 target_ulong addr = *(target_ulong *)s0;
2611 struct elf_sym *sym = (struct elf_sym *)s1;
2612 int result = 0;
2613 if (addr < sym->st_value) {
2614 result = -1;
2615 } else if (addr >= sym->st_value + sym->st_size) {
2616 result = 1;
2617 }
2618 return result;
2619 }
2620
2621 static const char *lookup_symbolxx(struct syminfo *s, target_ulong orig_addr)
2622 {
2623 #if ELF_CLASS == ELFCLASS32
2624 struct elf_sym *syms = s->disas_symtab.elf32;
2625 #else
2626 struct elf_sym *syms = s->disas_symtab.elf64;
2627 #endif
2628
2629 // binary search
2630 struct elf_sym *sym;
2631
2632 sym = bsearch(&orig_addr, syms, s->disas_num_syms, sizeof(*syms), symfind);
2633 if (sym != NULL) {
2634 return s->disas_strtab + sym->st_name;
2635 }
2636
2637 return "";
2638 }
2639
2640 /* FIXME: This should use elf_ops.h */
2641 static int symcmp(const void *s0, const void *s1)
2642 {
2643 struct elf_sym *sym0 = (struct elf_sym *)s0;
2644 struct elf_sym *sym1 = (struct elf_sym *)s1;
2645 return (sym0->st_value < sym1->st_value)
2646 ? -1
2647 : ((sym0->st_value > sym1->st_value) ? 1 : 0);
2648 }
2649
2650 /* Best attempt to load symbols from this ELF object. */
2651 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias)
2652 {
2653 int i, shnum, nsyms, sym_idx = 0, str_idx = 0;
2654 uint64_t segsz;
2655 struct elf_shdr *shdr;
2656 char *strings = NULL;
2657 struct syminfo *s = NULL;
2658 struct elf_sym *new_syms, *syms = NULL;
2659
2660 shnum = hdr->e_shnum;
2661 i = shnum * sizeof(struct elf_shdr);
2662 shdr = (struct elf_shdr *)alloca(i);
2663 if (pread(fd, shdr, i, hdr->e_shoff) != i) {
2664 return;
2665 }
2666
2667 bswap_shdr(shdr, shnum);
2668 for (i = 0; i < shnum; ++i) {
2669 if (shdr[i].sh_type == SHT_SYMTAB) {
2670 sym_idx = i;
2671 str_idx = shdr[i].sh_link;
2672 goto found;
2673 }
2674 }
2675
2676 /* There will be no symbol table if the file was stripped. */
2677 return;
2678
2679 found:
2680 /* Now know where the strtab and symtab are. Snarf them. */
2681 s = g_try_new(struct syminfo, 1);
2682 if (!s) {
2683 goto give_up;
2684 }
2685
2686 segsz = shdr[str_idx].sh_size;
2687 s->disas_strtab = strings = g_try_malloc(segsz);
2688 if (!strings ||
2689 pread(fd, strings, segsz, shdr[str_idx].sh_offset) != segsz) {
2690 goto give_up;
2691 }
2692
2693 segsz = shdr[sym_idx].sh_size;
2694 syms = g_try_malloc(segsz);
2695 if (!syms || pread(fd, syms, segsz, shdr[sym_idx].sh_offset) != segsz) {
2696 goto give_up;
2697 }
2698
2699 if (segsz / sizeof(struct elf_sym) > INT_MAX) {
2700 /* Implausibly large symbol table: give up rather than ploughing
2701 * on with the number of symbols calculation overflowing
2702 */
2703 goto give_up;
2704 }
2705 nsyms = segsz / sizeof(struct elf_sym);
2706 for (i = 0; i < nsyms; ) {
2707 bswap_sym(syms + i);
2708 /* Throw away entries which we do not need. */
2709 if (syms[i].st_shndx == SHN_UNDEF
2710 || syms[i].st_shndx >= SHN_LORESERVE
2711 || ELF_ST_TYPE(syms[i].st_info) != STT_FUNC) {
2712 if (i < --nsyms) {
2713 syms[i] = syms[nsyms];
2714 }
2715 } else {
2716 #if defined(TARGET_ARM) || defined (TARGET_MIPS)
2717 /* The bottom address bit marks a Thumb or MIPS16 symbol. */
2718 syms[i].st_value &= ~(target_ulong)1;
2719 #endif
2720 syms[i].st_value += load_bias;
2721 i++;
2722 }
2723 }
2724
2725 /* No "useful" symbol. */
2726 if (nsyms == 0) {
2727 goto give_up;
2728 }
2729
2730 /* Attempt to free the storage associated with the local symbols
2731 that we threw away. Whether or not this has any effect on the
2732 memory allocation depends on the malloc implementation and how
2733 many symbols we managed to discard. */
2734 new_syms = g_try_renew(struct elf_sym, syms, nsyms);
2735 if (new_syms == NULL) {
2736 goto give_up;
2737 }
2738 syms = new_syms;
2739
2740 qsort(syms, nsyms, sizeof(*syms), symcmp);
2741
2742 s->disas_num_syms = nsyms;
2743 #if ELF_CLASS == ELFCLASS32
2744 s->disas_symtab.elf32 = syms;
2745 #else
2746 s->disas_symtab.elf64 = syms;
2747 #endif
2748 s->lookup_symbol = lookup_symbolxx;
2749 s->next = syminfos;
2750 syminfos = s;
2751
2752 return;
2753
2754 give_up:
2755 g_free(s);
2756 g_free(strings);
2757 g_free(syms);
2758 }
2759
2760 uint32_t get_elf_eflags(int fd)
2761 {
2762 struct elfhdr ehdr;
2763 off_t offset;
2764 int ret;
2765
2766 /* Read ELF header */
2767 offset = lseek(fd, 0, SEEK_SET);
2768 if (offset == (off_t) -1) {
2769 return 0;
2770 }
2771 ret = read(fd, &ehdr, sizeof(ehdr));
2772 if (ret < sizeof(ehdr)) {
2773 return 0;
2774 }
2775 offset = lseek(fd, offset, SEEK_SET);
2776 if (offset == (off_t) -1) {
2777 return 0;
2778 }
2779
2780 /* Check ELF signature */
2781 if (!elf_check_ident(&ehdr)) {
2782 return 0;
2783 }
2784
2785 /* check header */
2786 bswap_ehdr(&ehdr);
2787 if (!elf_check_ehdr(&ehdr)) {
2788 return 0;
2789 }
2790
2791 /* return architecture id */
2792 return ehdr.e_flags;
2793 }
2794
2795 int load_elf_binary(struct linux_binprm *bprm, struct image_info *info)
2796 {
2797 struct image_info interp_info;
2798 struct elfhdr elf_ex;
2799 char *elf_interpreter = NULL;
2800 char *scratch;
2801
2802 memset(&interp_info, 0, sizeof(interp_info));
2803 #ifdef TARGET_MIPS
2804 interp_info.fp_abi = MIPS_ABI_FP_UNKNOWN;
2805 #endif
2806
2807 info->start_mmap = (abi_ulong)ELF_START_MMAP;
2808
2809 load_elf_image(bprm->filename, bprm->fd, info,
2810 &elf_interpreter, bprm->buf);
2811
2812 /* ??? We need a copy of the elf header for passing to create_elf_tables.
2813 If we do nothing, we'll have overwritten this when we re-use bprm->buf
2814 when we load the interpreter. */
2815 elf_ex = *(struct elfhdr *)bprm->buf;
2816
2817 /* Do this so that we can load the interpreter, if need be. We will
2818 change some of these later */
2819 bprm->p = setup_arg_pages(bprm, info);
2820
2821 scratch = g_new0(char, TARGET_PAGE_SIZE);
2822 if (STACK_GROWS_DOWN) {
2823 bprm->p = copy_elf_strings(1, &bprm->filename, scratch,
2824 bprm->p, info->stack_limit);
2825 info->file_string = bprm->p;
2826 bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch,
2827 bprm->p, info->stack_limit);
2828 info->env_strings = bprm->p;
2829 bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch,
2830 bprm->p, info->stack_limit);
2831 info->arg_strings = bprm->p;
2832 } else {
2833 info->arg_strings = bprm->p;
2834 bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch,
2835 bprm->p, info->stack_limit);
2836 info->env_strings = bprm->p;
2837 bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch,
2838 bprm->p, info->stack_limit);
2839 info->file_string = bprm->p;
2840 bprm->p = copy_elf_strings(1, &bprm->filename, scratch,
2841 bprm->p, info->stack_limit);
2842 }
2843
2844 g_free(scratch);
2845
2846 if (!bprm->p) {
2847 fprintf(stderr, "%s: %s\n", bprm->filename, strerror(E2BIG));
2848 exit(-1);
2849 }
2850
2851 if (elf_interpreter) {
2852 load_elf_interp(elf_interpreter, &interp_info, bprm->buf);
2853
2854 /* If the program interpreter is one of these two, then assume
2855 an iBCS2 image. Otherwise assume a native linux image. */
2856
2857 if (strcmp(elf_interpreter, "/usr/lib/libc.so.1") == 0
2858 || strcmp(elf_interpreter, "/usr/lib/ld.so.1") == 0) {
2859 info->personality = PER_SVR4;
2860
2861 /* Why this, you ask??? Well SVr4 maps page 0 as read-only,
2862 and some applications "depend" upon this behavior. Since
2863 we do not have the power to recompile these, we emulate
2864 the SVr4 behavior. Sigh. */
2865 target_mmap(0, qemu_host_page_size, PROT_READ | PROT_EXEC,
2866 MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
2867 }
2868 #ifdef TARGET_MIPS
2869 info->interp_fp_abi = interp_info.fp_abi;
2870 #endif
2871 }
2872
2873 bprm->p = create_elf_tables(bprm->p, bprm->argc, bprm->envc, &elf_ex,
2874 info, (elf_interpreter ? &interp_info : NULL));
2875 info->start_stack = bprm->p;
2876
2877 /* If we have an interpreter, set that as the program's entry point.
2878 Copy the load_bias as well, to help PPC64 interpret the entry
2879 point as a function descriptor. Do this after creating elf tables
2880 so that we copy the original program entry point into the AUXV. */
2881 if (elf_interpreter) {
2882 info->load_bias = interp_info.load_bias;
2883 info->entry = interp_info.entry;
2884 free(elf_interpreter);
2885 }
2886
2887 #ifdef USE_ELF_CORE_DUMP
2888 bprm->core_dump = &elf_core_dump;
2889 #endif
2890
2891 /*
2892 * If we reserved extra space for brk, release it now.
2893 * The implementation of do_brk in syscalls.c expects to be able
2894 * to mmap pages in this space.
2895 */
2896 if (info->reserve_brk) {
2897 abi_ulong start_brk = HOST_PAGE_ALIGN(info->brk);
2898 abi_ulong end_brk = HOST_PAGE_ALIGN(info->brk + info->reserve_brk);
2899 target_munmap(start_brk, end_brk - start_brk);
2900 }
2901
2902 return 0;
2903 }
2904
2905 #ifdef USE_ELF_CORE_DUMP
2906 /*
2907 * Definitions to generate Intel SVR4-like core files.
2908 * These mostly have the same names as the SVR4 types with "target_elf_"
2909 * tacked on the front to prevent clashes with linux definitions,
2910 * and the typedef forms have been avoided. This is mostly like
2911 * the SVR4 structure, but more Linuxy, with things that Linux does
2912 * not support and which gdb doesn't really use excluded.
2913 *
2914 * Fields we don't dump (their contents is zero) in linux-user qemu
2915 * are marked with XXX.
2916 *
2917 * Core dump code is copied from linux kernel (fs/binfmt_elf.c).
2918 *
2919 * Porting ELF coredump for target is (quite) simple process. First you
2920 * define USE_ELF_CORE_DUMP in target ELF code (where init_thread() for
2921 * the target resides):
2922 *
2923 * #define USE_ELF_CORE_DUMP
2924 *
2925 * Next you define type of register set used for dumping. ELF specification
2926 * says that it needs to be array of elf_greg_t that has size of ELF_NREG.
2927 *
2928 * typedef <target_regtype> target_elf_greg_t;
2929 * #define ELF_NREG <number of registers>
2930 * typedef taret_elf_greg_t target_elf_gregset_t[ELF_NREG];
2931 *
2932 * Last step is to implement target specific function that copies registers
2933 * from given cpu into just specified register set. Prototype is:
2934 *
2935 * static void elf_core_copy_regs(taret_elf_gregset_t *regs,
2936 * const CPUArchState *env);
2937 *
2938 * Parameters:
2939 * regs - copy register values into here (allocated and zeroed by caller)
2940 * env - copy registers from here
2941 *
2942 * Example for ARM target is provided in this file.
2943 */
2944
2945 /* An ELF note in memory */
2946 struct memelfnote {
2947 const char *name;
2948 size_t namesz;
2949 size_t namesz_rounded;
2950 int type;
2951 size_t datasz;
2952 size_t datasz_rounded;
2953 void *data;
2954 size_t notesz;
2955 };
2956
2957 struct target_elf_siginfo {
2958 abi_int si_signo; /* signal number */
2959 abi_int si_code; /* extra code */
2960 abi_int si_errno; /* errno */
2961 };
2962
2963 struct target_elf_prstatus {
2964 struct target_elf_siginfo pr_info; /* Info associated with signal */
2965 abi_short pr_cursig; /* Current signal */
2966 abi_ulong pr_sigpend; /* XXX */
2967 abi_ulong pr_sighold; /* XXX */
2968 target_pid_t pr_pid;
2969 target_pid_t pr_ppid;
2970 target_pid_t pr_pgrp;
2971 target_pid_t pr_sid;
2972 struct target_timeval pr_utime; /* XXX User time */
2973 struct target_timeval pr_stime; /* XXX System time */
2974 struct target_timeval pr_cutime; /* XXX Cumulative user time */
2975 struct target_timeval pr_cstime; /* XXX Cumulative system time */
2976 target_elf_gregset_t pr_reg; /* GP registers */
2977 abi_int pr_fpvalid; /* XXX */
2978 };
2979
2980 #define ELF_PRARGSZ (80) /* Number of chars for args */
2981
2982 struct target_elf_prpsinfo {
2983 char pr_state; /* numeric process state */
2984 char pr_sname; /* char for pr_state */
2985 char pr_zomb; /* zombie */
2986 char pr_nice; /* nice val */
2987 abi_ulong pr_flag; /* flags */
2988 target_uid_t pr_uid;
2989 target_gid_t pr_gid;
2990 target_pid_t pr_pid, pr_ppid, pr_pgrp, pr_sid;
2991 /* Lots missing */
2992 char pr_fname[16] QEMU_NONSTRING; /* filename of executable */
2993 char pr_psargs[ELF_PRARGSZ]; /* initial part of arg list */
2994 };
2995
2996 /* Here is the structure in which status of each thread is captured. */
2997 struct elf_thread_status {
2998 QTAILQ_ENTRY(elf_thread_status) ets_link;
2999 struct target_elf_prstatus prstatus; /* NT_PRSTATUS */
3000 #if 0
3001 elf_fpregset_t fpu; /* NT_PRFPREG */
3002 struct task_struct *thread;
3003 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */
3004 #endif
3005 struct memelfnote notes[1];
3006 int num_notes;
3007 };
3008
3009 struct elf_note_info {
3010 struct memelfnote *notes;
3011 struct target_elf_prstatus *prstatus; /* NT_PRSTATUS */
3012 struct target_elf_prpsinfo *psinfo; /* NT_PRPSINFO */
3013
3014 QTAILQ_HEAD(, elf_thread_status) thread_list;
3015 #if 0
3016 /*
3017 * Current version of ELF coredump doesn't support
3018 * dumping fp regs etc.
3019 */
3020 elf_fpregset_t *fpu;
3021 elf_fpxregset_t *xfpu;
3022 int thread_status_size;
3023 #endif
3024 int notes_size;
3025 int numnote;
3026 };
3027
3028 struct vm_area_struct {
3029 target_ulong vma_start; /* start vaddr of memory region */
3030 target_ulong vma_end; /* end vaddr of memory region */
3031 abi_ulong vma_flags; /* protection etc. flags for the region */
3032 QTAILQ_ENTRY(vm_area_struct) vma_link;
3033 };
3034
3035 struct mm_struct {
3036 QTAILQ_HEAD(, vm_area_struct) mm_mmap;
3037 int mm_count; /* number of mappings */
3038 };
3039
3040 static struct mm_struct *vma_init(void);
3041 static void vma_delete(struct mm_struct *);
3042 static int vma_add_mapping(struct mm_struct *, target_ulong,
3043 target_ulong, abi_ulong);
3044 static int vma_get_mapping_count(const struct mm_struct *);
3045 static struct vm_area_struct *vma_first(const struct mm_struct *);
3046 static struct vm_area_struct *vma_next(struct vm_area_struct *);
3047 static abi_ulong vma_dump_size(const struct vm_area_struct *);
3048 static int vma_walker(void *priv, target_ulong start, target_ulong end,
3049 unsigned long flags);
3050
3051 static void fill_elf_header(struct elfhdr *, int, uint16_t, uint32_t);
3052 static void fill_note(struct memelfnote *, const char *, int,
3053 unsigned int, void *);
3054 static void fill_prstatus(struct target_elf_prstatus *, const TaskState *, int);
3055 static int fill_psinfo(struct target_elf_prpsinfo *, const TaskState *);
3056 static void fill_auxv_note(struct memelfnote *, const TaskState *);
3057 static void fill_elf_note_phdr(struct elf_phdr *, int, off_t);
3058 static size_t note_size(const struct memelfnote *);
3059 static void free_note_info(struct elf_note_info *);
3060 static int fill_note_info(struct elf_note_info *, long, const CPUArchState *);
3061 static void fill_thread_info(struct elf_note_info *, const CPUArchState *);
3062 static int core_dump_filename(const TaskState *, char *, size_t);
3063
3064 static int dump_write(int, const void *, size_t);
3065 static int write_note(struct memelfnote *, int);
3066 static int write_note_info(struct elf_note_info *, int);
3067
3068 #ifdef BSWAP_NEEDED
3069 static void bswap_prstatus(struct target_elf_prstatus *prstatus)
3070 {
3071 prstatus->pr_info.si_signo = tswap32(prstatus->pr_info.si_signo);
3072 prstatus->pr_info.si_code = tswap32(prstatus->pr_info.si_code);
3073 prstatus->pr_info.si_errno = tswap32(prstatus->pr_info.si_errno);
3074 prstatus->pr_cursig = tswap16(prstatus->pr_cursig);
3075 prstatus->pr_sigpend = tswapal(prstatus->pr_sigpend);
3076 prstatus->pr_sighold = tswapal(prstatus->pr_sighold);
3077 prstatus->pr_pid = tswap32(prstatus->pr_pid);
3078 prstatus->pr_ppid = tswap32(prstatus->pr_ppid);
3079 prstatus->pr_pgrp = tswap32(prstatus->pr_pgrp);
3080 prstatus->pr_sid = tswap32(prstatus->pr_sid);
3081 /* cpu times are not filled, so we skip them */
3082 /* regs should be in correct format already */
3083 prstatus->pr_fpvalid = tswap32(prstatus->pr_fpvalid);
3084 }
3085
3086 static void bswap_psinfo(struct target_elf_prpsinfo *psinfo)
3087 {
3088 psinfo->pr_flag = tswapal(psinfo->pr_flag);
3089 psinfo->pr_uid = tswap16(psinfo->pr_uid);
3090 psinfo->pr_gid = tswap16(psinfo->pr_gid);
3091 psinfo->pr_pid = tswap32(psinfo->pr_pid);
3092 psinfo->pr_ppid = tswap32(psinfo->pr_ppid);
3093 psinfo->pr_pgrp = tswap32(psinfo->pr_pgrp);
3094 psinfo->pr_sid = tswap32(psinfo->pr_sid);
3095 }
3096
3097 static void bswap_note(struct elf_note *en)
3098 {
3099 bswap32s(&en->n_namesz);
3100 bswap32s(&en->n_descsz);
3101 bswap32s(&en->n_type);
3102 }
3103 #else
3104 static inline void bswap_prstatus(struct target_elf_prstatus *p) { }
3105 static inline void bswap_psinfo(struct target_elf_prpsinfo *p) {}
3106 static inline void bswap_note(struct elf_note *en) { }
3107 #endif /* BSWAP_NEEDED */
3108
3109 /*
3110 * Minimal support for linux memory regions. These are needed
3111 * when we are finding out what memory exactly belongs to
3112 * emulated process. No locks needed here, as long as
3113 * thread that received the signal is stopped.
3114 */
3115
3116 static struct mm_struct *vma_init(void)
3117 {
3118 struct mm_struct *mm;
3119
3120 if ((mm = g_malloc(sizeof (*mm))) == NULL)
3121 return (NULL);
3122
3123 mm->mm_count = 0;
3124 QTAILQ_INIT(&mm->mm_mmap);
3125
3126 return (mm);
3127 }
3128
3129 static void vma_delete(struct mm_struct *mm)
3130 {
3131 struct vm_area_struct *vma;
3132
3133 while ((vma = vma_first(mm)) != NULL) {
3134 QTAILQ_REMOVE(&mm->mm_mmap, vma, vma_link);
3135 g_free(vma);
3136 }
3137 g_free(mm);
3138 }
3139
3140 static int vma_add_mapping(struct mm_struct *mm, target_ulong start,
3141 target_ulong end, abi_ulong flags)
3142 {
3143 struct vm_area_struct *vma;
3144
3145 if ((vma = g_malloc0(sizeof (*vma))) == NULL)
3146 return (-1);
3147
3148 vma->vma_start = start;
3149 vma->vma_end = end;
3150 vma->vma_flags = flags;
3151
3152 QTAILQ_INSERT_TAIL(&mm->mm_mmap, vma, vma_link);
3153 mm->mm_count++;
3154
3155 return (0);
3156 }
3157
3158 static struct vm_area_struct *vma_first(const struct mm_struct *mm)
3159 {
3160 return (QTAILQ_FIRST(&mm->mm_mmap));
3161 }
3162
3163 static struct vm_area_struct *vma_next(struct vm_area_struct *vma)
3164 {
3165 return (QTAILQ_NEXT(vma, vma_link));
3166 }
3167
3168 static int vma_get_mapping_count(const struct mm_struct *mm)
3169 {
3170 return (mm->mm_count);
3171 }
3172
3173 /*
3174 * Calculate file (dump) size of given memory region.
3175 */
3176 static abi_ulong vma_dump_size(const struct vm_area_struct *vma)
3177 {
3178 /* if we cannot even read the first page, skip it */
3179 if (!access_ok(VERIFY_READ, vma->vma_start, TARGET_PAGE_SIZE))
3180 return (0);
3181
3182 /*
3183 * Usually we don't dump executable pages as they contain
3184 * non-writable code that debugger can read directly from
3185 * target library etc. However, thread stacks are marked
3186 * also executable so we read in first page of given region
3187 * and check whether it contains elf header. If there is
3188 * no elf header, we dump it.
3189 */
3190 if (vma->vma_flags & PROT_EXEC) {
3191 char page[TARGET_PAGE_SIZE];
3192
3193 copy_from_user(page, vma->vma_start, sizeof (page));
3194 if ((page[EI_MAG0] == ELFMAG0) &&
3195 (page[EI_MAG1] == ELFMAG1) &&
3196 (page[EI_MAG2] == ELFMAG2) &&
3197 (page[EI_MAG3] == ELFMAG3)) {
3198 /*
3199 * Mappings are possibly from ELF binary. Don't dump
3200 * them.
3201 */
3202 return (0);
3203 }
3204 }
3205
3206 return (vma->vma_end - vma->vma_start);
3207 }
3208
3209 static int vma_walker(void *priv, target_ulong start, target_ulong end,
3210 unsigned long flags)
3211 {
3212 struct mm_struct *mm = (struct mm_struct *)priv;
3213
3214 vma_add_mapping(mm, start, end, flags);
3215 return (0);
3216 }
3217
3218 static void fill_note(struct memelfnote *note, const char *name, int type,
3219 unsigned int sz, void *data)
3220 {
3221 unsigned int namesz;
3222
3223 namesz = strlen(name) + 1;
3224 note->name = name;
3225 note->namesz = namesz;
3226 note->namesz_rounded = roundup(namesz, sizeof (int32_t));
3227 note->type = type;
3228 note->datasz = sz;
3229 note->datasz_rounded = roundup(sz, sizeof (int32_t));
3230
3231 note->data = data;
3232
3233 /*
3234 * We calculate rounded up note size here as specified by
3235 * ELF document.
3236 */
3237 note->notesz = sizeof (struct elf_note) +
3238 note->namesz_rounded + note->datasz_rounded;
3239 }
3240
3241 static void fill_elf_header(struct elfhdr *elf, int segs, uint16_t machine,
3242 uint32_t flags)
3243 {
3244 (void) memset(elf, 0, sizeof(*elf));
3245
3246 (void) memcpy(elf->e_ident, ELFMAG, SELFMAG);
3247 elf->e_ident[EI_CLASS] = ELF_CLASS;
3248 elf->e_ident[EI_DATA] = ELF_DATA;
3249 elf->e_ident[EI_VERSION] = EV_CURRENT;
3250 elf->e_ident[EI_OSABI] = ELF_OSABI;
3251
3252 elf->e_type = ET_CORE;
3253 elf->e_machine = machine;
3254 elf->e_version = EV_CURRENT;
3255 elf->e_phoff = sizeof(struct elfhdr);
3256 elf->e_flags = flags;
3257 elf->e_ehsize = sizeof(struct elfhdr);
3258 elf->e_phentsize = sizeof(struct elf_phdr);
3259 elf->e_phnum = segs;
3260
3261 bswap_ehdr(elf);
3262 }
3263
3264 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, off_t offset)
3265 {
3266 phdr->p_type = PT_NOTE;
3267 phdr->p_offset = offset;
3268 phdr->p_vaddr = 0;
3269 phdr->p_paddr = 0;
3270 phdr->p_filesz = sz;
3271 phdr->p_memsz = 0;
3272 phdr->p_flags = 0;
3273 phdr->p_align = 0;
3274
3275 bswap_phdr(phdr, 1);
3276 }
3277
3278 static size_t note_size(const struct memelfnote *note)
3279 {
3280 return (note->notesz);
3281 }
3282
3283 static void fill_prstatus(struct target_elf_prstatus *prstatus,
3284 const TaskState *ts, int signr)
3285 {
3286 (void) memset(prstatus, 0, sizeof (*prstatus));
3287 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
3288 prstatus->pr_pid = ts->ts_tid;
3289 prstatus->pr_ppid = getppid();
3290 prstatus->pr_pgrp = getpgrp();
3291 prstatus->pr_sid = getsid(0);
3292
3293 bswap_prstatus(prstatus);
3294 }
3295
3296 static int fill_psinfo(struct target_elf_prpsinfo *psinfo, const TaskState *ts)
3297 {
3298 char *base_filename;
3299 unsigned int i, len;
3300
3301 (void) memset(psinfo, 0, sizeof (*psinfo));
3302
3303 len = ts->info->arg_end - ts->info->arg_start;
3304 if (len >= ELF_PRARGSZ)
3305 len = ELF_PRARGSZ - 1;
3306 if (copy_from_user(&psinfo->pr_psargs, ts->info->arg_start, len))
3307 return -EFAULT;
3308 for (i = 0; i < len; i++)
3309 if (psinfo->pr_psargs[i] == 0)
3310 psinfo->pr_psargs[i] = ' ';
3311 psinfo->pr_psargs[len] = 0;
3312
3313 psinfo->pr_pid = getpid();
3314 psinfo->pr_ppid = getppid();
3315 psinfo->pr_pgrp = getpgrp();
3316 psinfo->pr_sid = getsid(0);
3317 psinfo->pr_uid = getuid();
3318 psinfo->pr_gid = getgid();
3319
3320 base_filename = g_path_get_basename(ts->bprm->filename);
3321 /*
3322 * Using strncpy here is fine: at max-length,
3323 * this field is not NUL-terminated.
3324 */
3325 (void) strncpy(psinfo->pr_fname, base_filename,
3326 sizeof(psinfo->pr_fname));
3327
3328 g_free(base_filename);
3329 bswap_psinfo(psinfo);
3330 return (0);
3331 }
3332
3333 static void fill_auxv_note(struct memelfnote *note, const TaskState *ts)
3334 {
3335 elf_addr_t auxv = (elf_addr_t)ts->info->saved_auxv;
3336 elf_addr_t orig_auxv = auxv;
3337 void *ptr;
3338 int len = ts->info->auxv_len;
3339
3340 /*
3341 * Auxiliary vector is stored in target process stack. It contains
3342 * {type, value} pairs that we need to dump into note. This is not
3343 * strictly necessary but we do it here for sake of completeness.
3344 */
3345
3346 /* read in whole auxv vector and copy it to memelfnote */
3347 ptr = lock_user(VERIFY_READ, orig_auxv, len, 0);
3348 if (ptr != NULL) {
3349 fill_note(note, "CORE", NT_AUXV, len, ptr);
3350 unlock_user(ptr, auxv, len);
3351 }
3352 }
3353
3354 /*
3355 * Constructs name of coredump file. We have following convention
3356 * for the name:
3357 * qemu_<basename-of-target-binary>_<date>-<time>_<pid>.core
3358 *
3359 * Returns 0 in case of success, -1 otherwise (errno is set).
3360 */
3361 static int core_dump_filename(const TaskState *ts, char *buf,
3362 size_t bufsize)
3363 {
3364 char timestamp[64];
3365 char *base_filename = NULL;
3366 struct timeval tv;
3367 struct tm tm;
3368
3369 assert(bufsize >= PATH_MAX);
3370
3371 if (gettimeofday(&tv, NULL) < 0) {
3372 (void) fprintf(stderr, "unable to get current timestamp: %s",
3373 strerror(errno));
3374 return (-1);
3375 }
3376
3377 base_filename = g_path_get_basename(ts->bprm->filename);
3378 (void) strftime(timestamp, sizeof (timestamp), "%Y%m%d-%H%M%S",
3379 localtime_r(&tv.tv_sec, &tm));
3380 (void) snprintf(buf, bufsize, "qemu_%s_%s_%d.core",
3381 base_filename, timestamp, (int)getpid());
3382 g_free(base_filename);
3383
3384 return (0);
3385 }
3386
3387 static int dump_write(int fd, const void *ptr, size_t size)
3388 {
3389 const char *bufp = (const char *)ptr;
3390 ssize_t bytes_written, bytes_left;
3391 struct rlimit dumpsize;
3392 off_t pos;
3393
3394 bytes_written = 0;
3395 getrlimit(RLIMIT_CORE, &dumpsize);
3396 if ((pos = lseek(fd, 0, SEEK_CUR))==-1) {
3397 if (errno == ESPIPE) { /* not a seekable stream */
3398 bytes_left = size;
3399 } else {
3400 return pos;
3401 }
3402 } else {
3403 if (dumpsize.rlim_cur <= pos) {
3404 return -1;
3405 } else if (dumpsize.rlim_cur == RLIM_INFINITY) {
3406 bytes_left = size;
3407 } else {
3408 size_t limit_left=dumpsize.rlim_cur - pos;
3409 bytes_left = limit_left >= size ? size : limit_left ;
3410 }
3411 }
3412
3413 /*
3414 * In normal conditions, single write(2) should do but
3415 * in case of socket etc. this mechanism is more portable.
3416 */
3417 do {
3418 bytes_written = write(fd, bufp, bytes_left);
3419 if (bytes_written < 0) {
3420 if (errno == EINTR)
3421 continue;
3422 return (-1);
3423 } else if (bytes_written == 0) { /* eof */
3424 return (-1);
3425 }
3426 bufp += bytes_written;
3427 bytes_left -= bytes_written;
3428 } while (bytes_left > 0);
3429
3430 return (0);
3431 }
3432
3433 static int write_note(struct memelfnote *men, int fd)
3434 {
3435 struct elf_note en;
3436
3437 en.n_namesz = men->namesz;
3438 en.n_type = men->type;
3439 en.n_descsz = men->datasz;
3440
3441 bswap_note(&en);
3442
3443 if (dump_write(fd, &en, sizeof(en)) != 0)
3444 return (-1);
3445 if (dump_write(fd, men->name, men->namesz_rounded) != 0)
3446 return (-1);
3447 if (dump_write(fd, men->data, men->datasz_rounded) != 0)
3448 return (-1);
3449
3450 return (0);
3451 }
3452
3453 static void fill_thread_info(struct elf_note_info *info, const CPUArchState *env)
3454 {
3455 CPUState *cpu = env_cpu((CPUArchState *)env);
3456 TaskState *ts = (TaskState *)cpu->opaque;
3457 struct elf_thread_status *ets;
3458
3459 ets = g_malloc0(sizeof (*ets));
3460 ets->num_notes = 1; /* only prstatus is dumped */
3461 fill_prstatus(&ets->prstatus, ts, 0);
3462 elf_core_copy_regs(&ets->prstatus.pr_reg, env);
3463 fill_note(&ets->notes[0], "CORE", NT_PRSTATUS, sizeof (ets->prstatus),
3464 &ets->prstatus);
3465
3466 QTAILQ_INSERT_TAIL(&info->thread_list, ets, ets_link);
3467
3468 info->notes_size += note_size(&ets->notes[0]);
3469 }
3470
3471 static void init_note_info(struct elf_note_info *info)
3472 {
3473 /* Initialize the elf_note_info structure so that it is at
3474 * least safe to call free_note_info() on it. Must be
3475 * called before calling fill_note_info().
3476 */
3477 memset(info, 0, sizeof (*info));
3478 QTAILQ_INIT(&info->thread_list);
3479 }
3480
3481 static int fill_note_info(struct elf_note_info *info,
3482 long signr, const CPUArchState *env)
3483 {
3484 #define NUMNOTES 3
3485 CPUState *cpu = env_cpu((CPUArchState *)env);
3486 TaskState *ts = (TaskState *)cpu->opaque;
3487 int i;
3488
3489 info->notes = g_new0(struct memelfnote, NUMNOTES);
3490 if (info->notes == NULL)
3491 return (-ENOMEM);
3492 info->prstatus = g_malloc0(sizeof (*info->prstatus));
3493 if (info->prstatus == NULL)
3494 return (-ENOMEM);
3495 info->psinfo = g_malloc0(sizeof (*info->psinfo));
3496 if (info->prstatus == NULL)
3497 return (-ENOMEM);
3498
3499 /*
3500 * First fill in status (and registers) of current thread
3501 * including process info & aux vector.
3502 */
3503 fill_prstatus(info->prstatus, ts, signr);
3504 elf_core_copy_regs(&info->prstatus->pr_reg, env);
3505 fill_note(&info->notes[0], "CORE", NT_PRSTATUS,
3506 sizeof (*info->prstatus), info->prstatus);
3507 fill_psinfo(info->psinfo, ts);
3508 fill_note(&info->notes[1], "CORE", NT_PRPSINFO,
3509 sizeof (*info->psinfo), info->psinfo);
3510 fill_auxv_note(&info->notes[2], ts);
3511 info->numnote = 3;
3512
3513 info->notes_size = 0;
3514 for (i = 0; i < info->numnote; i++)
3515 info->notes_size += note_size(&info->notes[i]);
3516
3517 /* read and fill status of all threads */
3518 cpu_list_lock();
3519 CPU_FOREACH(cpu) {
3520 if (cpu == thread_cpu) {
3521 continue;
3522 }
3523 fill_thread_info(info, (CPUArchState *)cpu->env_ptr);
3524 }
3525 cpu_list_unlock();
3526
3527 return (0);
3528 }
3529
3530 static void free_note_info(struct elf_note_info *info)
3531 {
3532 struct elf_thread_status *ets;
3533
3534 while (!QTAILQ_EMPTY(&info->thread_list)) {
3535 ets = QTAILQ_FIRST(&info->thread_list);
3536 QTAILQ_REMOVE(&info->thread_list, ets, ets_link);
3537 g_free(ets);
3538 }
3539
3540 g_free(info->prstatus);
3541 g_free(info->psinfo);
3542 g_free(info->notes);
3543 }
3544
3545 static int write_note_info(struct elf_note_info *info, int fd)
3546 {
3547 struct elf_thread_status *ets;
3548 int i, error = 0;
3549
3550 /* write prstatus, psinfo and auxv for current thread */
3551 for (i = 0; i < info->numnote; i++)
3552 if ((error = write_note(&info->notes[i], fd)) != 0)
3553 return (error);
3554
3555 /* write prstatus for each thread */
3556 QTAILQ_FOREACH(ets, &info->thread_list, ets_link) {
3557 if ((error = write_note(&ets->notes[0], fd)) != 0)
3558 return (error);
3559 }
3560
3561 return (0);
3562 }
3563
3564 /*
3565 * Write out ELF coredump.
3566 *
3567 * See documentation of ELF object file format in:
3568 * http://www.caldera.com/developers/devspecs/gabi41.pdf
3569 *
3570 * Coredump format in linux is following:
3571 *
3572 * 0 +----------------------+ \
3573 * | ELF header | ET_CORE |
3574 * +----------------------+ |
3575 * | ELF program headers | |--- headers
3576 * | - NOTE section | |
3577 * | - PT_LOAD sections | |
3578 * +----------------------+ /
3579 * | NOTEs: |
3580 * | - NT_PRSTATUS |
3581 * | - NT_PRSINFO |
3582 * | - NT_AUXV |
3583 * +----------------------+ <-- aligned to target page
3584 * | Process memory dump |
3585 * : :
3586 * . .
3587 * : :
3588 * | |
3589 * +----------------------+
3590 *
3591 * NT_PRSTATUS -> struct elf_prstatus (per thread)
3592 * NT_PRSINFO -> struct elf_prpsinfo
3593 * NT_AUXV is array of { type, value } pairs (see fill_auxv_note()).
3594 *
3595 * Format follows System V format as close as possible. Current
3596 * version limitations are as follows:
3597 * - no floating point registers are dumped
3598 *
3599 * Function returns 0 in case of success, negative errno otherwise.
3600 *
3601 * TODO: make this work also during runtime: it should be
3602 * possible to force coredump from running process and then
3603 * continue processing. For example qemu could set up SIGUSR2
3604 * handler (provided that target process haven't registered
3605 * handler for that) that does the dump when signal is received.
3606 */
3607 static int elf_core_dump(int signr, const CPUArchState *env)
3608 {
3609 const CPUState *cpu = env_cpu((CPUArchState *)env);
3610 const TaskState *ts = (const TaskState *)cpu->opaque;
3611 struct vm_area_struct *vma = NULL;
3612 char corefile[PATH_MAX];
3613 struct elf_note_info info;
3614 struct elfhdr elf;
3615 struct elf_phdr phdr;
3616 struct rlimit dumpsize;
3617 struct mm_struct *mm = NULL;
3618 off_t offset = 0, data_offset = 0;
3619 int segs = 0;
3620 int fd = -1;
3621
3622 init_note_info(&info);
3623
3624 errno = 0;
3625 getrlimit(RLIMIT_CORE, &dumpsize);
3626 if (dumpsize.rlim_cur == 0)
3627 return 0;
3628
3629 if (core_dump_filename(ts, corefile, sizeof (corefile)) < 0)
3630 return (-errno);
3631
3632 if ((fd = open(corefile, O_WRONLY | O_CREAT,
3633 S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH)) < 0)
3634 return (-errno);
3635
3636 /*
3637 * Walk through target process memory mappings and
3638 * set up structure containing this information. After
3639 * this point vma_xxx functions can be used.
3640 */
3641 if ((mm = vma_init()) == NULL)
3642 goto out;
3643
3644 walk_memory_regions(mm, vma_walker);
3645 segs = vma_get_mapping_count(mm);
3646
3647 /*
3648 * Construct valid coredump ELF header. We also
3649 * add one more segment for notes.
3650 */
3651 fill_elf_header(&elf, segs + 1, ELF_MACHINE, 0);
3652 if (dump_write(fd, &elf, sizeof (elf)) != 0)
3653 goto out;
3654
3655 /* fill in the in-memory version of notes */
3656 if (fill_note_info(&info, signr, env) < 0)
3657 goto out;
3658
3659 offset += sizeof (elf); /* elf header */
3660 offset += (segs + 1) * sizeof (struct elf_phdr); /* program headers */
3661
3662 /* write out notes program header */
3663 fill_elf_note_phdr(&phdr, info.notes_size, offset);
3664
3665 offset += info.notes_size;
3666 if (dump_write(fd, &phdr, sizeof (phdr)) != 0)
3667 goto out;
3668
3669 /*
3670 * ELF specification wants data to start at page boundary so
3671 * we align it here.
3672 */
3673 data_offset = offset = roundup(offset, ELF_EXEC_PAGESIZE);
3674
3675 /*
3676 * Write program headers for memory regions mapped in
3677 * the target process.
3678 */
3679 for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) {
3680 (void) memset(&phdr, 0, sizeof (phdr));
3681
3682 phdr.p_type = PT_LOAD;
3683 phdr.p_offset = offset;
3684 phdr.p_vaddr = vma->vma_start;
3685 phdr.p_paddr = 0;
3686 phdr.p_filesz = vma_dump_size(vma);
3687 offset += phdr.p_filesz;
3688 phdr.p_memsz = vma->vma_end - vma->vma_start;
3689 phdr.p_flags = vma->vma_flags & PROT_READ ? PF_R : 0;
3690 if (vma->vma_flags & PROT_WRITE)
3691 phdr.p_flags |= PF_W;
3692 if (vma->vma_flags & PROT_EXEC)
3693 phdr.p_flags |= PF_X;
3694 phdr.p_align = ELF_EXEC_PAGESIZE;
3695
3696 bswap_phdr(&phdr, 1);
3697 if (dump_write(fd, &phdr, sizeof(phdr)) != 0) {
3698 goto out;
3699 }
3700 }
3701
3702 /*
3703 * Next we write notes just after program headers. No
3704 * alignment needed here.
3705 */
3706 if (write_note_info(&info, fd) < 0)
3707 goto out;
3708
3709 /* align data to page boundary */
3710 if (lseek(fd, data_offset, SEEK_SET) != data_offset)
3711 goto out;
3712
3713 /*
3714 * Finally we can dump process memory into corefile as well.
3715 */
3716 for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) {
3717 abi_ulong addr;
3718 abi_ulong end;
3719
3720 end = vma->vma_start + vma_dump_size(vma);
3721
3722 for (addr = vma->vma_start; addr < end;
3723 addr += TARGET_PAGE_SIZE) {
3724 char page[TARGET_PAGE_SIZE];
3725 int error;
3726
3727 /*
3728 * Read in page from target process memory and
3729 * write it to coredump file.
3730 */
3731 error = copy_from_user(page, addr, sizeof (page));
3732 if (error != 0) {
3733 (void) fprintf(stderr, "unable to dump " TARGET_ABI_FMT_lx "\n",
3734 addr);
3735 errno = -error;
3736 goto out;
3737 }
3738 if (dump_write(fd, page, TARGET_PAGE_SIZE) < 0)
3739 goto out;
3740 }
3741 }
3742
3743 out:
3744 free_note_info(&info);
3745 if (mm