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