?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
softfloat: Move div_floats to softfloat-parts.c.inc
[qemu/ar7.git] / linux-user / elfload.c
blobfc9c4f12be92bd4eec03e9e7803f4d28deef12a8
1 /* This is the Linux kernel elf-loading code, ported into user space */
2 #include "qemu/osdep.h"
3 #include <sys/param.h>
4
5 #include <sys/resource.h>
6 #include <sys/shm.h>
7
8 #include "qemu.h"
9 #include "disas/disas.h"
10 #include "qemu/bitops.h"
11 #include "qemu/path.h"
12 #include "qemu/queue.h"
13 #include "qemu/guest-random.h"
14 #include "qemu/units.h"
15 #include "qemu/selfmap.h"
16 #include "qapi/error.h"
17
18 #ifdef _ARCH_PPC64
19 #undef ARCH_DLINFO
20 #undef ELF_PLATFORM
21 #undef ELF_HWCAP
22 #undef ELF_HWCAP2
23 #undef ELF_CLASS
24 #undef ELF_DATA
25 #undef ELF_ARCH
26 #endif
27
28 #define ELF_OSABI ELFOSABI_SYSV
29
30 /* from personality.h */
31
32 /*
33 * Flags for bug emulation.
34 *
35 * These occupy the top three bytes.
36 */
37 enum {
38 ADDR_NO_RANDOMIZE = 0x0040000, /* disable randomization of VA space */
39 FDPIC_FUNCPTRS = 0x0080000, /* userspace function ptrs point to
40 descriptors (signal handling) */
41 MMAP_PAGE_ZERO = 0x0100000,
42 ADDR_COMPAT_LAYOUT = 0x0200000,
43 READ_IMPLIES_EXEC = 0x0400000,
44 ADDR_LIMIT_32BIT = 0x0800000,
45 SHORT_INODE = 0x1000000,
46 WHOLE_SECONDS = 0x2000000,
47 STICKY_TIMEOUTS = 0x4000000,
48 ADDR_LIMIT_3GB = 0x8000000,
49 };
50
51 /*
52 * Personality types.
53 *
54 * These go in the low byte. Avoid using the top bit, it will
55 * conflict with error returns.
56 */
57 enum {
58 PER_LINUX = 0x0000,
59 PER_LINUX_32BIT = 0x0000 | ADDR_LIMIT_32BIT,
60 PER_LINUX_FDPIC = 0x0000 | FDPIC_FUNCPTRS,
61 PER_SVR4 = 0x0001 | STICKY_TIMEOUTS | MMAP_PAGE_ZERO,
62 PER_SVR3 = 0x0002 | STICKY_TIMEOUTS | SHORT_INODE,
63 PER_SCOSVR3 = 0x0003 | STICKY_TIMEOUTS | WHOLE_SECONDS | SHORT_INODE,
64 PER_OSR5 = 0x0003 | STICKY_TIMEOUTS | WHOLE_SECONDS,
65 PER_WYSEV386 = 0x0004 | STICKY_TIMEOUTS | SHORT_INODE,
66 PER_ISCR4 = 0x0005 | STICKY_TIMEOUTS,
67 PER_BSD = 0x0006,
68 PER_SUNOS = 0x0006 | STICKY_TIMEOUTS,
69 PER_XENIX = 0x0007 | STICKY_TIMEOUTS | SHORT_INODE,
70 PER_LINUX32 = 0x0008,
71 PER_LINUX32_3GB = 0x0008 | ADDR_LIMIT_3GB,
72 PER_IRIX32 = 0x0009 | STICKY_TIMEOUTS,/* IRIX5 32-bit */
73 PER_IRIXN32 = 0x000a | STICKY_TIMEOUTS,/* IRIX6 new 32-bit */
74 PER_IRIX64 = 0x000b | STICKY_TIMEOUTS,/* IRIX6 64-bit */
75 PER_RISCOS = 0x000c,
76 PER_SOLARIS = 0x000d | STICKY_TIMEOUTS,
77 PER_UW7 = 0x000e | STICKY_TIMEOUTS | MMAP_PAGE_ZERO,
78 PER_OSF4 = 0x000f, /* OSF/1 v4 */
79 PER_HPUX = 0x0010,
80 PER_MASK = 0x00ff,
81 };
82
83 /*
84 * Return the base personality without flags.
85 */
86 #define personality(pers) (pers & PER_MASK)
87
88 int info_is_fdpic(struct image_info *info)
89 {
90 return info->personality == PER_LINUX_FDPIC;
91 }
92
93 /* this flag is uneffective under linux too, should be deleted */
94 #ifndef MAP_DENYWRITE
95 #define MAP_DENYWRITE 0
96 #endif
97
98 /* should probably go in elf.h */
99 #ifndef ELIBBAD
100 #define ELIBBAD 80
101 #endif
102
103 #ifdef TARGET_WORDS_BIGENDIAN
104 #define ELF_DATA ELFDATA2MSB
105 #else
106 #define ELF_DATA ELFDATA2LSB
107 #endif
108
109 #ifdef TARGET_ABI_MIPSN32
110 typedef abi_ullong target_elf_greg_t;
111 #define tswapreg(ptr) tswap64(ptr)
112 #else
113 typedef abi_ulong target_elf_greg_t;
114 #define tswapreg(ptr) tswapal(ptr)
115 #endif
116
117 #ifdef USE_UID16
118 typedef abi_ushort target_uid_t;
119 typedef abi_ushort target_gid_t;
120 #else
121 typedef abi_uint target_uid_t;
122 typedef abi_uint target_gid_t;
123 #endif
124 typedef abi_int target_pid_t;
125
126 #ifdef TARGET_I386
127
128 #define ELF_PLATFORM get_elf_platform()
129
130 static const char *get_elf_platform(void)
131 {
132 static char elf_platform[] = "i386";
133 int family = object_property_get_int(OBJECT(thread_cpu), "family", NULL);
134 if (family > 6)
135 family = 6;
136 if (family >= 3)
137 elf_platform[1] = '0' + family;
138 return elf_platform;
139 }
140
141 #define ELF_HWCAP get_elf_hwcap()
142
143 static uint32_t get_elf_hwcap(void)
144 {
145 X86CPU *cpu = X86_CPU(thread_cpu);
146
147 return cpu->env.features[FEAT_1_EDX];
148 }
149
150 #ifdef TARGET_X86_64
151 #define ELF_START_MMAP 0x2aaaaab000ULL
152
153 #define ELF_CLASS ELFCLASS64
154 #define ELF_ARCH EM_X86_64
155
156 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop)
157 {
158 regs->rax = 0;
159 regs->rsp = infop->start_stack;
160 regs->rip = infop->entry;
161 }
162
163 #define ELF_NREG 27
164 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
165
166 /*
167 * Note that ELF_NREG should be 29 as there should be place for
168 * TRAPNO and ERR "registers" as well but linux doesn't dump
169 * those.
170 *
171 * See linux kernel: arch/x86/include/asm/elf.h
172 */
173 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUX86State *env)
174 {
175 (*regs)[0] = env->regs[15];
176 (*regs)[1] = env->regs[14];
177 (*regs)[2] = env->regs[13];
178 (*regs)[3] = env->regs[12];
179 (*regs)[4] = env->regs[R_EBP];
180 (*regs)[5] = env->regs[R_EBX];
181 (*regs)[6] = env->regs[11];
182 (*regs)[7] = env->regs[10];
183 (*regs)[8] = env->regs[9];
184 (*regs)[9] = env->regs[8];
185 (*regs)[10] = env->regs[R_EAX];
186 (*regs)[11] = env->regs[R_ECX];
187 (*regs)[12] = env->regs[R_EDX];
188 (*regs)[13] = env->regs[R_ESI];
189 (*regs)[14] = env->regs[R_EDI];
190 (*regs)[15] = env->regs[R_EAX]; /* XXX */
191 (*regs)[16] = env->eip;
192 (*regs)[17] = env->segs[R_CS].selector & 0xffff;
193 (*regs)[18] = env->eflags;
194 (*regs)[19] = env->regs[R_ESP];
195 (*regs)[20] = env->segs[R_SS].selector & 0xffff;
196 (*regs)[21] = env->segs[R_FS].selector & 0xffff;
197 (*regs)[22] = env->segs[R_GS].selector & 0xffff;
198 (*regs)[23] = env->segs[R_DS].selector & 0xffff;
199 (*regs)[24] = env->segs[R_ES].selector & 0xffff;
200 (*regs)[25] = env->segs[R_FS].selector & 0xffff;
201 (*regs)[26] = env->segs[R_GS].selector & 0xffff;
202 }
203
204 #else
205
206 #define ELF_START_MMAP 0x80000000
207
208 /*
209 * This is used to ensure we don't load something for the wrong architecture.
210 */
211 #define elf_check_arch(x) ( ((x) == EM_386) || ((x) == EM_486) )
212
213 /*
214 * These are used to set parameters in the core dumps.
215 */
216 #define ELF_CLASS ELFCLASS32
217 #define ELF_ARCH EM_386
218
219 static inline void init_thread(struct target_pt_regs *regs,
220 struct image_info *infop)
221 {
222 regs->esp = infop->start_stack;
223 regs->eip = infop->entry;
224
225 /* SVR4/i386 ABI (pages 3-31, 3-32) says that when the program
226 starts %edx contains a pointer to a function which might be
227 registered using `atexit'. This provides a mean for the
228 dynamic linker to call DT_FINI functions for shared libraries
229 that have been loaded before the code runs.
230
231 A value of 0 tells we have no such handler. */
232 regs->edx = 0;
233 }
234
235 #define ELF_NREG 17
236 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
237
238 /*
239 * Note that ELF_NREG should be 19 as there should be place for
240 * TRAPNO and ERR "registers" as well but linux doesn't dump
241 * those.
242 *
243 * See linux kernel: arch/x86/include/asm/elf.h
244 */
245 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUX86State *env)
246 {
247 (*regs)[0] = env->regs[R_EBX];
248 (*regs)[1] = env->regs[R_ECX];
249 (*regs)[2] = env->regs[R_EDX];
250 (*regs)[3] = env->regs[R_ESI];
251 (*regs)[4] = env->regs[R_EDI];
252 (*regs)[5] = env->regs[R_EBP];
253 (*regs)[6] = env->regs[R_EAX];
254 (*regs)[7] = env->segs[R_DS].selector & 0xffff;
255 (*regs)[8] = env->segs[R_ES].selector & 0xffff;
256 (*regs)[9] = env->segs[R_FS].selector & 0xffff;
257 (*regs)[10] = env->segs[R_GS].selector & 0xffff;
258 (*regs)[11] = env->regs[R_EAX]; /* XXX */
259 (*regs)[12] = env->eip;
260 (*regs)[13] = env->segs[R_CS].selector & 0xffff;
261 (*regs)[14] = env->eflags;
262 (*regs)[15] = env->regs[R_ESP];
263 (*regs)[16] = env->segs[R_SS].selector & 0xffff;
264 }
265 #endif
266
267 #define USE_ELF_CORE_DUMP
268 #define ELF_EXEC_PAGESIZE 4096
269
270 #endif
271
272 #ifdef TARGET_ARM
273
274 #ifndef TARGET_AARCH64
275 /* 32 bit ARM definitions */
276
277 #define ELF_START_MMAP 0x80000000
278
279 #define ELF_ARCH EM_ARM
280 #define ELF_CLASS ELFCLASS32
281
282 static inline void init_thread(struct target_pt_regs *regs,
283 struct image_info *infop)
284 {
285 abi_long stack = infop->start_stack;
286 memset(regs, 0, sizeof(*regs));
287
288 regs->uregs[16] = ARM_CPU_MODE_USR;
289 if (infop->entry & 1) {
290 regs->uregs[16] |= CPSR_T;
291 }
292 regs->uregs[15] = infop->entry & 0xfffffffe;
293 regs->uregs[13] = infop->start_stack;
294 /* FIXME - what to for failure of get_user()? */
295 get_user_ual(regs->uregs[2], stack + 8); /* envp */
296 get_user_ual(regs->uregs[1], stack + 4); /* envp */
297 /* XXX: it seems that r0 is zeroed after ! */
298 regs->uregs[0] = 0;
299 /* For uClinux PIC binaries. */
300 /* XXX: Linux does this only on ARM with no MMU (do we care ?) */
301 regs->uregs[10] = infop->start_data;
302
303 /* Support ARM FDPIC. */
304 if (info_is_fdpic(infop)) {
305 /* As described in the ABI document, r7 points to the loadmap info
306 * prepared by the kernel. If an interpreter is needed, r8 points
307 * to the interpreter loadmap and r9 points to the interpreter
308 * PT_DYNAMIC info. If no interpreter is needed, r8 is zero, and
309 * r9 points to the main program PT_DYNAMIC info.
310 */
311 regs->uregs[7] = infop->loadmap_addr;
312 if (infop->interpreter_loadmap_addr) {
313 /* Executable is dynamically loaded. */
314 regs->uregs[8] = infop->interpreter_loadmap_addr;
315 regs->uregs[9] = infop->interpreter_pt_dynamic_addr;
316 } else {
317 regs->uregs[8] = 0;
318 regs->uregs[9] = infop->pt_dynamic_addr;
319 }
320 }
321 }
322
323 #define ELF_NREG 18
324 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
325
326 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUARMState *env)
327 {
328 (*regs)[0] = tswapreg(env->regs[0]);
329 (*regs)[1] = tswapreg(env->regs[1]);
330 (*regs)[2] = tswapreg(env->regs[2]);
331 (*regs)[3] = tswapreg(env->regs[3]);
332 (*regs)[4] = tswapreg(env->regs[4]);
333 (*regs)[5] = tswapreg(env->regs[5]);
334 (*regs)[6] = tswapreg(env->regs[6]);
335 (*regs)[7] = tswapreg(env->regs[7]);
336 (*regs)[8] = tswapreg(env->regs[8]);
337 (*regs)[9] = tswapreg(env->regs[9]);
338 (*regs)[10] = tswapreg(env->regs[10]);
339 (*regs)[11] = tswapreg(env->regs[11]);
340 (*regs)[12] = tswapreg(env->regs[12]);
341 (*regs)[13] = tswapreg(env->regs[13]);
342 (*regs)[14] = tswapreg(env->regs[14]);
343 (*regs)[15] = tswapreg(env->regs[15]);
344
345 (*regs)[16] = tswapreg(cpsr_read((CPUARMState *)env));
346 (*regs)[17] = tswapreg(env->regs[0]); /* XXX */
347 }
348
349 #define USE_ELF_CORE_DUMP
350 #define ELF_EXEC_PAGESIZE 4096
351
352 enum
353 {
354 ARM_HWCAP_ARM_SWP = 1 << 0,
355 ARM_HWCAP_ARM_HALF = 1 << 1,
356 ARM_HWCAP_ARM_THUMB = 1 << 2,
357 ARM_HWCAP_ARM_26BIT = 1 << 3,
358 ARM_HWCAP_ARM_FAST_MULT = 1 << 4,
359 ARM_HWCAP_ARM_FPA = 1 << 5,
360 ARM_HWCAP_ARM_VFP = 1 << 6,
361 ARM_HWCAP_ARM_EDSP = 1 << 7,
362 ARM_HWCAP_ARM_JAVA = 1 << 8,
363 ARM_HWCAP_ARM_IWMMXT = 1 << 9,
364 ARM_HWCAP_ARM_CRUNCH = 1 << 10,
365 ARM_HWCAP_ARM_THUMBEE = 1 << 11,
366 ARM_HWCAP_ARM_NEON = 1 << 12,
367 ARM_HWCAP_ARM_VFPv3 = 1 << 13,
368 ARM_HWCAP_ARM_VFPv3D16 = 1 << 14,
369 ARM_HWCAP_ARM_TLS = 1 << 15,
370 ARM_HWCAP_ARM_VFPv4 = 1 << 16,
371 ARM_HWCAP_ARM_IDIVA = 1 << 17,
372 ARM_HWCAP_ARM_IDIVT = 1 << 18,
373 ARM_HWCAP_ARM_VFPD32 = 1 << 19,
374 ARM_HWCAP_ARM_LPAE = 1 << 20,
375 ARM_HWCAP_ARM_EVTSTRM = 1 << 21,
376 };
377
378 enum {
379 ARM_HWCAP2_ARM_AES = 1 << 0,
380 ARM_HWCAP2_ARM_PMULL = 1 << 1,
381 ARM_HWCAP2_ARM_SHA1 = 1 << 2,
382 ARM_HWCAP2_ARM_SHA2 = 1 << 3,
383 ARM_HWCAP2_ARM_CRC32 = 1 << 4,
384 };
385
386 /* The commpage only exists for 32 bit kernels */
387
388 #define ARM_COMMPAGE (intptr_t)0xffff0f00u
389
390 static bool init_guest_commpage(void)
391 {
392 void *want = g2h_untagged(ARM_COMMPAGE & -qemu_host_page_size);
393 void *addr = mmap(want, qemu_host_page_size, PROT_READ | PROT_WRITE,
394 MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0);
395
396 if (addr == MAP_FAILED) {
397 perror("Allocating guest commpage");
398 exit(EXIT_FAILURE);
399 }
400 if (addr != want) {
401 return false;
402 }
403
404 /* Set kernel helper versions; rest of page is 0. */
405 __put_user(5, (uint32_t *)g2h_untagged(0xffff0ffcu));
406
407 if (mprotect(addr, qemu_host_page_size, PROT_READ)) {
408 perror("Protecting guest commpage");
409 exit(EXIT_FAILURE);
410 }
411 return true;
412 }
413
414 #define ELF_HWCAP get_elf_hwcap()
415 #define ELF_HWCAP2 get_elf_hwcap2()
416
417 static uint32_t get_elf_hwcap(void)
418 {
419 ARMCPU *cpu = ARM_CPU(thread_cpu);
420 uint32_t hwcaps = 0;
421
422 hwcaps |= ARM_HWCAP_ARM_SWP;
423 hwcaps |= ARM_HWCAP_ARM_HALF;
424 hwcaps |= ARM_HWCAP_ARM_THUMB;
425 hwcaps |= ARM_HWCAP_ARM_FAST_MULT;
426
427 /* probe for the extra features */
428 #define GET_FEATURE(feat, hwcap) \
429 do { if (arm_feature(&cpu->env, feat)) { hwcaps |= hwcap; } } while (0)
430
431 #define GET_FEATURE_ID(feat, hwcap) \
432 do { if (cpu_isar_feature(feat, cpu)) { hwcaps |= hwcap; } } while (0)
433
434 /* EDSP is in v5TE and above, but all our v5 CPUs are v5TE */
435 GET_FEATURE(ARM_FEATURE_V5, ARM_HWCAP_ARM_EDSP);
436 GET_FEATURE(ARM_FEATURE_IWMMXT, ARM_HWCAP_ARM_IWMMXT);
437 GET_FEATURE(ARM_FEATURE_THUMB2EE, ARM_HWCAP_ARM_THUMBEE);
438 GET_FEATURE(ARM_FEATURE_NEON, ARM_HWCAP_ARM_NEON);
439 GET_FEATURE(ARM_FEATURE_V6K, ARM_HWCAP_ARM_TLS);
440 GET_FEATURE(ARM_FEATURE_LPAE, ARM_HWCAP_ARM_LPAE);
441 GET_FEATURE_ID(aa32_arm_div, ARM_HWCAP_ARM_IDIVA);
442 GET_FEATURE_ID(aa32_thumb_div, ARM_HWCAP_ARM_IDIVT);
443 GET_FEATURE_ID(aa32_vfp, ARM_HWCAP_ARM_VFP);
444
445 if (cpu_isar_feature(aa32_fpsp_v3, cpu) ||
446 cpu_isar_feature(aa32_fpdp_v3, cpu)) {
447 hwcaps |= ARM_HWCAP_ARM_VFPv3;
448 if (cpu_isar_feature(aa32_simd_r32, cpu)) {
449 hwcaps |= ARM_HWCAP_ARM_VFPD32;
450 } else {
451 hwcaps |= ARM_HWCAP_ARM_VFPv3D16;
452 }
453 }
454 GET_FEATURE_ID(aa32_simdfmac, ARM_HWCAP_ARM_VFPv4);
455
456 return hwcaps;
457 }
458
459 static uint32_t get_elf_hwcap2(void)
460 {
461 ARMCPU *cpu = ARM_CPU(thread_cpu);
462 uint32_t hwcaps = 0;
463
464 GET_FEATURE_ID(aa32_aes, ARM_HWCAP2_ARM_AES);
465 GET_FEATURE_ID(aa32_pmull, ARM_HWCAP2_ARM_PMULL);
466 GET_FEATURE_ID(aa32_sha1, ARM_HWCAP2_ARM_SHA1);
467 GET_FEATURE_ID(aa32_sha2, ARM_HWCAP2_ARM_SHA2);
468 GET_FEATURE_ID(aa32_crc32, ARM_HWCAP2_ARM_CRC32);
469 return hwcaps;
470 }
471
472 #undef GET_FEATURE
473 #undef GET_FEATURE_ID
474
475 #define ELF_PLATFORM get_elf_platform()
476
477 static const char *get_elf_platform(void)
478 {
479 CPUARMState *env = thread_cpu->env_ptr;
480
481 #ifdef TARGET_WORDS_BIGENDIAN
482 # define END "b"
483 #else
484 # define END "l"
485 #endif
486
487 if (arm_feature(env, ARM_FEATURE_V8)) {
488 return "v8" END;
489 } else if (arm_feature(env, ARM_FEATURE_V7)) {
490 if (arm_feature(env, ARM_FEATURE_M)) {
491 return "v7m" END;
492 } else {
493 return "v7" END;
494 }
495 } else if (arm_feature(env, ARM_FEATURE_V6)) {
496 return "v6" END;
497 } else if (arm_feature(env, ARM_FEATURE_V5)) {
498 return "v5" END;
499 } else {
500 return "v4" END;
501 }
502
503 #undef END
504 }
505
506 #else
507 /* 64 bit ARM definitions */
508 #define ELF_START_MMAP 0x80000000
509
510 #define ELF_ARCH EM_AARCH64
511 #define ELF_CLASS ELFCLASS64
512 #ifdef TARGET_WORDS_BIGENDIAN
513 # define ELF_PLATFORM "aarch64_be"
514 #else
515 # define ELF_PLATFORM "aarch64"
516 #endif
517
518 static inline void init_thread(struct target_pt_regs *regs,
519 struct image_info *infop)
520 {
521 abi_long stack = infop->start_stack;
522 memset(regs, 0, sizeof(*regs));
523
524 regs->pc = infop->entry & ~0x3ULL;
525 regs->sp = stack;
526 }
527
528 #define ELF_NREG 34
529 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
530
531 static void elf_core_copy_regs(target_elf_gregset_t *regs,
532 const CPUARMState *env)
533 {
534 int i;
535
536 for (i = 0; i < 32; i++) {
537 (*regs)[i] = tswapreg(env->xregs[i]);
538 }
539 (*regs)[32] = tswapreg(env->pc);
540 (*regs)[33] = tswapreg(pstate_read((CPUARMState *)env));
541 }
542
543 #define USE_ELF_CORE_DUMP
544 #define ELF_EXEC_PAGESIZE 4096
545
546 enum {
547 ARM_HWCAP_A64_FP = 1 << 0,
548 ARM_HWCAP_A64_ASIMD = 1 << 1,
549 ARM_HWCAP_A64_EVTSTRM = 1 << 2,
550 ARM_HWCAP_A64_AES = 1 << 3,
551 ARM_HWCAP_A64_PMULL = 1 << 4,
552 ARM_HWCAP_A64_SHA1 = 1 << 5,
553 ARM_HWCAP_A64_SHA2 = 1 << 6,
554 ARM_HWCAP_A64_CRC32 = 1 << 7,
555 ARM_HWCAP_A64_ATOMICS = 1 << 8,
556 ARM_HWCAP_A64_FPHP = 1 << 9,
557 ARM_HWCAP_A64_ASIMDHP = 1 << 10,
558 ARM_HWCAP_A64_CPUID = 1 << 11,
559 ARM_HWCAP_A64_ASIMDRDM = 1 << 12,
560 ARM_HWCAP_A64_JSCVT = 1 << 13,
561 ARM_HWCAP_A64_FCMA = 1 << 14,
562 ARM_HWCAP_A64_LRCPC = 1 << 15,
563 ARM_HWCAP_A64_DCPOP = 1 << 16,
564 ARM_HWCAP_A64_SHA3 = 1 << 17,
565 ARM_HWCAP_A64_SM3 = 1 << 18,
566 ARM_HWCAP_A64_SM4 = 1 << 19,
567 ARM_HWCAP_A64_ASIMDDP = 1 << 20,
568 ARM_HWCAP_A64_SHA512 = 1 << 21,
569 ARM_HWCAP_A64_SVE = 1 << 22,
570 ARM_HWCAP_A64_ASIMDFHM = 1 << 23,
571 ARM_HWCAP_A64_DIT = 1 << 24,
572 ARM_HWCAP_A64_USCAT = 1 << 25,
573 ARM_HWCAP_A64_ILRCPC = 1 << 26,
574 ARM_HWCAP_A64_FLAGM = 1 << 27,
575 ARM_HWCAP_A64_SSBS = 1 << 28,
576 ARM_HWCAP_A64_SB = 1 << 29,
577 ARM_HWCAP_A64_PACA = 1 << 30,
578 ARM_HWCAP_A64_PACG = 1UL << 31,
579
580 ARM_HWCAP2_A64_DCPODP = 1 << 0,
581 ARM_HWCAP2_A64_SVE2 = 1 << 1,
582 ARM_HWCAP2_A64_SVEAES = 1 << 2,
583 ARM_HWCAP2_A64_SVEPMULL = 1 << 3,
584 ARM_HWCAP2_A64_SVEBITPERM = 1 << 4,
585 ARM_HWCAP2_A64_SVESHA3 = 1 << 5,
586 ARM_HWCAP2_A64_SVESM4 = 1 << 6,
587 ARM_HWCAP2_A64_FLAGM2 = 1 << 7,
588 ARM_HWCAP2_A64_FRINT = 1 << 8,
589 ARM_HWCAP2_A64_SVEI8MM = 1 << 9,
590 ARM_HWCAP2_A64_SVEF32MM = 1 << 10,
591 ARM_HWCAP2_A64_SVEF64MM = 1 << 11,
592 ARM_HWCAP2_A64_SVEBF16 = 1 << 12,
593 ARM_HWCAP2_A64_I8MM = 1 << 13,
594 ARM_HWCAP2_A64_BF16 = 1 << 14,
595 ARM_HWCAP2_A64_DGH = 1 << 15,
596 ARM_HWCAP2_A64_RNG = 1 << 16,
597 ARM_HWCAP2_A64_BTI = 1 << 17,
598 ARM_HWCAP2_A64_MTE = 1 << 18,
599 };
600
601 #define ELF_HWCAP get_elf_hwcap()
602 #define ELF_HWCAP2 get_elf_hwcap2()
603
604 #define GET_FEATURE_ID(feat, hwcap) \
605 do { if (cpu_isar_feature(feat, cpu)) { hwcaps |= hwcap; } } while (0)
606
607 static uint32_t get_elf_hwcap(void)
608 {
609 ARMCPU *cpu = ARM_CPU(thread_cpu);
610 uint32_t hwcaps = 0;
611
612 hwcaps |= ARM_HWCAP_A64_FP;
613 hwcaps |= ARM_HWCAP_A64_ASIMD;
614 hwcaps |= ARM_HWCAP_A64_CPUID;
615
616 /* probe for the extra features */
617
618 GET_FEATURE_ID(aa64_aes, ARM_HWCAP_A64_AES);
619 GET_FEATURE_ID(aa64_pmull, ARM_HWCAP_A64_PMULL);
620 GET_FEATURE_ID(aa64_sha1, ARM_HWCAP_A64_SHA1);
621 GET_FEATURE_ID(aa64_sha256, ARM_HWCAP_A64_SHA2);
622 GET_FEATURE_ID(aa64_sha512, ARM_HWCAP_A64_SHA512);
623 GET_FEATURE_ID(aa64_crc32, ARM_HWCAP_A64_CRC32);
624 GET_FEATURE_ID(aa64_sha3, ARM_HWCAP_A64_SHA3);
625 GET_FEATURE_ID(aa64_sm3, ARM_HWCAP_A64_SM3);
626 GET_FEATURE_ID(aa64_sm4, ARM_HWCAP_A64_SM4);
627 GET_FEATURE_ID(aa64_fp16, ARM_HWCAP_A64_FPHP | ARM_HWCAP_A64_ASIMDHP);
628 GET_FEATURE_ID(aa64_atomics, ARM_HWCAP_A64_ATOMICS);
629 GET_FEATURE_ID(aa64_rdm, ARM_HWCAP_A64_ASIMDRDM);
630 GET_FEATURE_ID(aa64_dp, ARM_HWCAP_A64_ASIMDDP);
631 GET_FEATURE_ID(aa64_fcma, ARM_HWCAP_A64_FCMA);
632 GET_FEATURE_ID(aa64_sve, ARM_HWCAP_A64_SVE);
633 GET_FEATURE_ID(aa64_pauth, ARM_HWCAP_A64_PACA | ARM_HWCAP_A64_PACG);
634 GET_FEATURE_ID(aa64_fhm, ARM_HWCAP_A64_ASIMDFHM);
635 GET_FEATURE_ID(aa64_jscvt, ARM_HWCAP_A64_JSCVT);
636 GET_FEATURE_ID(aa64_sb, ARM_HWCAP_A64_SB);
637 GET_FEATURE_ID(aa64_condm_4, ARM_HWCAP_A64_FLAGM);
638 GET_FEATURE_ID(aa64_dcpop, ARM_HWCAP_A64_DCPOP);
639 GET_FEATURE_ID(aa64_rcpc_8_3, ARM_HWCAP_A64_LRCPC);
640 GET_FEATURE_ID(aa64_rcpc_8_4, ARM_HWCAP_A64_ILRCPC);
641
642 return hwcaps;
643 }
644
645 static uint32_t get_elf_hwcap2(void)
646 {
647 ARMCPU *cpu = ARM_CPU(thread_cpu);
648 uint32_t hwcaps = 0;
649
650 GET_FEATURE_ID(aa64_dcpodp, ARM_HWCAP2_A64_DCPODP);
651 GET_FEATURE_ID(aa64_condm_5, ARM_HWCAP2_A64_FLAGM2);
652 GET_FEATURE_ID(aa64_frint, ARM_HWCAP2_A64_FRINT);
653 GET_FEATURE_ID(aa64_rndr, ARM_HWCAP2_A64_RNG);
654 GET_FEATURE_ID(aa64_bti, ARM_HWCAP2_A64_BTI);
655 GET_FEATURE_ID(aa64_mte, ARM_HWCAP2_A64_MTE);
656
657 return hwcaps;
658 }
659
660 #undef GET_FEATURE_ID
661
662 #endif /* not TARGET_AARCH64 */
663 #endif /* TARGET_ARM */
664
665 #ifdef TARGET_SPARC
666 #ifdef TARGET_SPARC64
667
668 #define ELF_START_MMAP 0x80000000
669 #define ELF_HWCAP (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \
670 | HWCAP_SPARC_MULDIV | HWCAP_SPARC_V9)
671 #ifndef TARGET_ABI32
672 #define elf_check_arch(x) ( (x) == EM_SPARCV9 || (x) == EM_SPARC32PLUS )
673 #else
674 #define elf_check_arch(x) ( (x) == EM_SPARC32PLUS || (x) == EM_SPARC )
675 #endif
676
677 #define ELF_CLASS ELFCLASS64
678 #define ELF_ARCH EM_SPARCV9
679
680 #define STACK_BIAS 2047
681
682 static inline void init_thread(struct target_pt_regs *regs,
683 struct image_info *infop)
684 {
685 #ifndef TARGET_ABI32
686 regs->tstate = 0;
687 #endif
688 regs->pc = infop->entry;
689 regs->npc = regs->pc + 4;
690 regs->y = 0;
691 #ifdef TARGET_ABI32
692 regs->u_regs[14] = infop->start_stack - 16 * 4;
693 #else
694 if (personality(infop->personality) == PER_LINUX32)
695 regs->u_regs[14] = infop->start_stack - 16 * 4;
696 else
697 regs->u_regs[14] = infop->start_stack - 16 * 8 - STACK_BIAS;
698 #endif
699 }
700
701 #else
702 #define ELF_START_MMAP 0x80000000
703 #define ELF_HWCAP (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \
704 | HWCAP_SPARC_MULDIV)
705
706 #define ELF_CLASS ELFCLASS32
707 #define ELF_ARCH EM_SPARC
708
709 static inline void init_thread(struct target_pt_regs *regs,
710 struct image_info *infop)
711 {
712 regs->psr = 0;
713 regs->pc = infop->entry;
714 regs->npc = regs->pc + 4;
715 regs->y = 0;
716 regs->u_regs[14] = infop->start_stack - 16 * 4;
717 }
718
719 #endif
720 #endif
721
722 #ifdef TARGET_PPC
723
724 #define ELF_MACHINE PPC_ELF_MACHINE
725 #define ELF_START_MMAP 0x80000000
726
727 #if defined(TARGET_PPC64) && !defined(TARGET_ABI32)
728
729 #define elf_check_arch(x) ( (x) == EM_PPC64 )
730
731 #define ELF_CLASS ELFCLASS64
732
733 #else
734
735 #define ELF_CLASS ELFCLASS32
736
737 #endif
738
739 #define ELF_ARCH EM_PPC
740
741 /* Feature masks for the Aux Vector Hardware Capabilities (AT_HWCAP).
742 See arch/powerpc/include/asm/cputable.h. */
743 enum {
744 QEMU_PPC_FEATURE_32 = 0x80000000,
745 QEMU_PPC_FEATURE_64 = 0x40000000,
746 QEMU_PPC_FEATURE_601_INSTR = 0x20000000,
747 QEMU_PPC_FEATURE_HAS_ALTIVEC = 0x10000000,
748 QEMU_PPC_FEATURE_HAS_FPU = 0x08000000,
749 QEMU_PPC_FEATURE_HAS_MMU = 0x04000000,
750 QEMU_PPC_FEATURE_HAS_4xxMAC = 0x02000000,
751 QEMU_PPC_FEATURE_UNIFIED_CACHE = 0x01000000,
752 QEMU_PPC_FEATURE_HAS_SPE = 0x00800000,
753 QEMU_PPC_FEATURE_HAS_EFP_SINGLE = 0x00400000,
754 QEMU_PPC_FEATURE_HAS_EFP_DOUBLE = 0x00200000,
755 QEMU_PPC_FEATURE_NO_TB = 0x00100000,
756 QEMU_PPC_FEATURE_POWER4 = 0x00080000,
757 QEMU_PPC_FEATURE_POWER5 = 0x00040000,
758 QEMU_PPC_FEATURE_POWER5_PLUS = 0x00020000,
759 QEMU_PPC_FEATURE_CELL = 0x00010000,
760 QEMU_PPC_FEATURE_BOOKE = 0x00008000,
761 QEMU_PPC_FEATURE_SMT = 0x00004000,
762 QEMU_PPC_FEATURE_ICACHE_SNOOP = 0x00002000,
763 QEMU_PPC_FEATURE_ARCH_2_05 = 0x00001000,
764 QEMU_PPC_FEATURE_PA6T = 0x00000800,
765 QEMU_PPC_FEATURE_HAS_DFP = 0x00000400,
766 QEMU_PPC_FEATURE_POWER6_EXT = 0x00000200,
767 QEMU_PPC_FEATURE_ARCH_2_06 = 0x00000100,
768 QEMU_PPC_FEATURE_HAS_VSX = 0x00000080,
769 QEMU_PPC_FEATURE_PSERIES_PERFMON_COMPAT = 0x00000040,
770
771 QEMU_PPC_FEATURE_TRUE_LE = 0x00000002,
772 QEMU_PPC_FEATURE_PPC_LE = 0x00000001,
773
774 /* Feature definitions in AT_HWCAP2. */
775 QEMU_PPC_FEATURE2_ARCH_2_07 = 0x80000000, /* ISA 2.07 */
776 QEMU_PPC_FEATURE2_HAS_HTM = 0x40000000, /* Hardware Transactional Memory */
777 QEMU_PPC_FEATURE2_HAS_DSCR = 0x20000000, /* Data Stream Control Register */
778 QEMU_PPC_FEATURE2_HAS_EBB = 0x10000000, /* Event Base Branching */
779 QEMU_PPC_FEATURE2_HAS_ISEL = 0x08000000, /* Integer Select */
780 QEMU_PPC_FEATURE2_HAS_TAR = 0x04000000, /* Target Address Register */
781 QEMU_PPC_FEATURE2_VEC_CRYPTO = 0x02000000,
782 QEMU_PPC_FEATURE2_HTM_NOSC = 0x01000000,
783 QEMU_PPC_FEATURE2_ARCH_3_00 = 0x00800000, /* ISA 3.00 */
784 QEMU_PPC_FEATURE2_HAS_IEEE128 = 0x00400000, /* VSX IEEE Bin Float 128-bit */
785 QEMU_PPC_FEATURE2_DARN = 0x00200000, /* darn random number insn */
786 QEMU_PPC_FEATURE2_SCV = 0x00100000, /* scv syscall */
787 QEMU_PPC_FEATURE2_HTM_NO_SUSPEND = 0x00080000, /* TM w/o suspended state */
788 };
789
790 #define ELF_HWCAP get_elf_hwcap()
791
792 static uint32_t get_elf_hwcap(void)
793 {
794 PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);
795 uint32_t features = 0;
796
797 /* We don't have to be terribly complete here; the high points are
798 Altivec/FP/SPE support. Anything else is just a bonus. */
799 #define GET_FEATURE(flag, feature) \
800 do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0)
801 #define GET_FEATURE2(flags, feature) \
802 do { \
803 if ((cpu->env.insns_flags2 & flags) == flags) { \
804 features |= feature; \
805 } \
806 } while (0)
807 GET_FEATURE(PPC_64B, QEMU_PPC_FEATURE_64);
808 GET_FEATURE(PPC_FLOAT, QEMU_PPC_FEATURE_HAS_FPU);
809 GET_FEATURE(PPC_ALTIVEC, QEMU_PPC_FEATURE_HAS_ALTIVEC);
810 GET_FEATURE(PPC_SPE, QEMU_PPC_FEATURE_HAS_SPE);
811 GET_FEATURE(PPC_SPE_SINGLE, QEMU_PPC_FEATURE_HAS_EFP_SINGLE);
812 GET_FEATURE(PPC_SPE_DOUBLE, QEMU_PPC_FEATURE_HAS_EFP_DOUBLE);
813 GET_FEATURE(PPC_BOOKE, QEMU_PPC_FEATURE_BOOKE);
814 GET_FEATURE(PPC_405_MAC, QEMU_PPC_FEATURE_HAS_4xxMAC);
815 GET_FEATURE2(PPC2_DFP, QEMU_PPC_FEATURE_HAS_DFP);
816 GET_FEATURE2(PPC2_VSX, QEMU_PPC_FEATURE_HAS_VSX);
817 GET_FEATURE2((PPC2_PERM_ISA206 | PPC2_DIVE_ISA206 | PPC2_ATOMIC_ISA206 |
818 PPC2_FP_CVT_ISA206 | PPC2_FP_TST_ISA206),
819 QEMU_PPC_FEATURE_ARCH_2_06);
820 #undef GET_FEATURE
821 #undef GET_FEATURE2
822
823 return features;
824 }
825
826 #define ELF_HWCAP2 get_elf_hwcap2()
827
828 static uint32_t get_elf_hwcap2(void)
829 {
830 PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);
831 uint32_t features = 0;
832
833 #define GET_FEATURE(flag, feature) \
834 do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0)
835 #define GET_FEATURE2(flag, feature) \
836 do { if (cpu->env.insns_flags2 & flag) { features |= feature; } } while (0)
837
838 GET_FEATURE(PPC_ISEL, QEMU_PPC_FEATURE2_HAS_ISEL);
839 GET_FEATURE2(PPC2_BCTAR_ISA207, QEMU_PPC_FEATURE2_HAS_TAR);
840 GET_FEATURE2((PPC2_BCTAR_ISA207 | PPC2_LSQ_ISA207 | PPC2_ALTIVEC_207 |
841 PPC2_ISA207S), QEMU_PPC_FEATURE2_ARCH_2_07 |
842 QEMU_PPC_FEATURE2_VEC_CRYPTO);
843 GET_FEATURE2(PPC2_ISA300, QEMU_PPC_FEATURE2_ARCH_3_00 |
844 QEMU_PPC_FEATURE2_DARN);
845
846 #undef GET_FEATURE
847 #undef GET_FEATURE2
848
849 return features;
850 }
851
852 /*
853 * The requirements here are:
854 * - keep the final alignment of sp (sp & 0xf)
855 * - make sure the 32-bit value at the first 16 byte aligned position of
856 * AUXV is greater than 16 for glibc compatibility.
857 * AT_IGNOREPPC is used for that.
858 * - for compatibility with glibc ARCH_DLINFO must always be defined on PPC,
859 * even if DLINFO_ARCH_ITEMS goes to zero or is undefined.
860 */
861 #define DLINFO_ARCH_ITEMS 5
862 #define ARCH_DLINFO \
863 do { \
864 PowerPCCPU *cpu = POWERPC_CPU(thread_cpu); \
865 /* \
866 * Handle glibc compatibility: these magic entries must \
867 * be at the lowest addresses in the final auxv. \
868 */ \
869 NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC); \
870 NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC); \
871 NEW_AUX_ENT(AT_DCACHEBSIZE, cpu->env.dcache_line_size); \
872 NEW_AUX_ENT(AT_ICACHEBSIZE, cpu->env.icache_line_size); \
873 NEW_AUX_ENT(AT_UCACHEBSIZE, 0); \
874 } while (0)
875
876 static inline void init_thread(struct target_pt_regs *_regs, struct image_info *infop)
877 {
878 _regs->gpr[1] = infop->start_stack;
879 #if defined(TARGET_PPC64) && !defined(TARGET_ABI32)
880 if (get_ppc64_abi(infop) < 2) {
881 uint64_t val;
882 get_user_u64(val, infop->entry + 8);
883 _regs->gpr[2] = val + infop->load_bias;
884 get_user_u64(val, infop->entry);
885 infop->entry = val + infop->load_bias;
886 } else {
887 _regs->gpr[12] = infop->entry; /* r12 set to global entry address */
888 }
889 #endif
890 _regs->nip = infop->entry;
891 }
892
893 /* See linux kernel: arch/powerpc/include/asm/elf.h. */
894 #define ELF_NREG 48
895 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
896
897 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUPPCState *env)
898 {
899 int i;
900 target_ulong ccr = 0;
901
902 for (i = 0; i < ARRAY_SIZE(env->gpr); i++) {
903 (*regs)[i] = tswapreg(env->gpr[i]);
904 }
905
906 (*regs)[32] = tswapreg(env->nip);
907 (*regs)[33] = tswapreg(env->msr);
908 (*regs)[35] = tswapreg(env->ctr);
909 (*regs)[36] = tswapreg(env->lr);
910 (*regs)[37] = tswapreg(env->xer);
911
912 for (i = 0; i < ARRAY_SIZE(env->crf); i++) {
913 ccr |= env->crf[i] << (32 - ((i + 1) * 4));
914 }
915 (*regs)[38] = tswapreg(ccr);
916 }
917
918 #define USE_ELF_CORE_DUMP
919 #define ELF_EXEC_PAGESIZE 4096
920
921 #endif
922
923 #ifdef TARGET_MIPS
924
925 #define ELF_START_MMAP 0x80000000
926
927 #ifdef TARGET_MIPS64
928 #define ELF_CLASS ELFCLASS64
929 #else
930 #define ELF_CLASS ELFCLASS32
931 #endif
932 #define ELF_ARCH EM_MIPS
933
934 #define elf_check_arch(x) ((x) == EM_MIPS || (x) == EM_NANOMIPS)
935
936 #ifdef TARGET_ABI_MIPSN32
937 #define elf_check_abi(x) ((x) & EF_MIPS_ABI2)
938 #else
939 #define elf_check_abi(x) (!((x) & EF_MIPS_ABI2))
940 #endif
941
942 static inline void init_thread(struct target_pt_regs *regs,
943 struct image_info *infop)
944 {
945 regs->cp0_status = 2 << CP0St_KSU;
946 regs->cp0_epc = infop->entry;
947 regs->regs[29] = infop->start_stack;
948 }
949
950 /* See linux kernel: arch/mips/include/asm/elf.h. */
951 #define ELF_NREG 45
952 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
953
954 /* See linux kernel: arch/mips/include/asm/reg.h. */
955 enum {
956 #ifdef TARGET_MIPS64
957 TARGET_EF_R0 = 0,
958 #else
959 TARGET_EF_R0 = 6,
960 #endif
961 TARGET_EF_R26 = TARGET_EF_R0 + 26,
962 TARGET_EF_R27 = TARGET_EF_R0 + 27,
963 TARGET_EF_LO = TARGET_EF_R0 + 32,
964 TARGET_EF_HI = TARGET_EF_R0 + 33,
965 TARGET_EF_CP0_EPC = TARGET_EF_R0 + 34,
966 TARGET_EF_CP0_BADVADDR = TARGET_EF_R0 + 35,
967 TARGET_EF_CP0_STATUS = TARGET_EF_R0 + 36,
968 TARGET_EF_CP0_CAUSE = TARGET_EF_R0 + 37
969 };
970
971 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */
972 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMIPSState *env)
973 {
974 int i;
975
976 for (i = 0; i < TARGET_EF_R0; i++) {
977 (*regs)[i] = 0;
978 }
979 (*regs)[TARGET_EF_R0] = 0;
980
981 for (i = 1; i < ARRAY_SIZE(env->active_tc.gpr); i++) {
982 (*regs)[TARGET_EF_R0 + i] = tswapreg(env->active_tc.gpr[i]);
983 }
984
985 (*regs)[TARGET_EF_R26] = 0;
986 (*regs)[TARGET_EF_R27] = 0;
987 (*regs)[TARGET_EF_LO] = tswapreg(env->active_tc.LO[0]);
988 (*regs)[TARGET_EF_HI] = tswapreg(env->active_tc.HI[0]);
989 (*regs)[TARGET_EF_CP0_EPC] = tswapreg(env->active_tc.PC);
990 (*regs)[TARGET_EF_CP0_BADVADDR] = tswapreg(env->CP0_BadVAddr);
991 (*regs)[TARGET_EF_CP0_STATUS] = tswapreg(env->CP0_Status);
992 (*regs)[TARGET_EF_CP0_CAUSE] = tswapreg(env->CP0_Cause);
993 }
994
995 #define USE_ELF_CORE_DUMP
996 #define ELF_EXEC_PAGESIZE 4096
997
998 /* See arch/mips/include/uapi/asm/hwcap.h. */
999 enum {
1000 HWCAP_MIPS_R6 = (1 << 0),
1001 HWCAP_MIPS_MSA = (1 << 1),
1002 HWCAP_MIPS_CRC32 = (1 << 2),
1003 HWCAP_MIPS_MIPS16 = (1 << 3),
1004 HWCAP_MIPS_MDMX = (1 << 4),
1005 HWCAP_MIPS_MIPS3D = (1 << 5),
1006 HWCAP_MIPS_SMARTMIPS = (1 << 6),
1007 HWCAP_MIPS_DSP = (1 << 7),
1008 HWCAP_MIPS_DSP2 = (1 << 8),
1009 HWCAP_MIPS_DSP3 = (1 << 9),
1010 HWCAP_MIPS_MIPS16E2 = (1 << 10),
1011 HWCAP_LOONGSON_MMI = (1 << 11),
1012 HWCAP_LOONGSON_EXT = (1 << 12),
1013 HWCAP_LOONGSON_EXT2 = (1 << 13),
1014 HWCAP_LOONGSON_CPUCFG = (1 << 14),
1015 };
1016
1017 #define ELF_HWCAP get_elf_hwcap()
1018
1019 #define GET_FEATURE_INSN(_flag, _hwcap) \
1020 do { if (cpu->env.insn_flags & (_flag)) { hwcaps |= _hwcap; } } while (0)
1021
1022 #define GET_FEATURE_REG_SET(_reg, _mask, _hwcap) \
1023 do { if (cpu->env._reg & (_mask)) { hwcaps |= _hwcap; } } while (0)
1024
1025 #define GET_FEATURE_REG_EQU(_reg, _start, _length, _val, _hwcap) \
1026 do { \
1027 if (extract32(cpu->env._reg, (_start), (_length)) == (_val)) { \
1028 hwcaps |= _hwcap; \
1029 } \
1030 } while (0)
1031
1032 static uint32_t get_elf_hwcap(void)
1033 {
1034 MIPSCPU *cpu = MIPS_CPU(thread_cpu);
1035 uint32_t hwcaps = 0;
1036
1037 GET_FEATURE_REG_EQU(CP0_Config0, CP0C0_AR, CP0C0_AR_LENGTH,
1038 2, HWCAP_MIPS_R6);
1039 GET_FEATURE_REG_SET(CP0_Config3, 1 << CP0C3_MSAP, HWCAP_MIPS_MSA);
1040 GET_FEATURE_INSN(ASE_LMMI, HWCAP_LOONGSON_MMI);
1041 GET_FEATURE_INSN(ASE_LEXT, HWCAP_LOONGSON_EXT);
1042
1043 return hwcaps;
1044 }
1045
1046 #undef GET_FEATURE_REG_EQU
1047 #undef GET_FEATURE_REG_SET
1048 #undef GET_FEATURE_INSN
1049
1050 #endif /* TARGET_MIPS */
1051
1052 #ifdef TARGET_MICROBLAZE
1053
1054 #define ELF_START_MMAP 0x80000000
1055
1056 #define elf_check_arch(x) ( (x) == EM_MICROBLAZE || (x) == EM_MICROBLAZE_OLD)
1057
1058 #define ELF_CLASS ELFCLASS32
1059 #define ELF_ARCH EM_MICROBLAZE
1060
1061 static inline void init_thread(struct target_pt_regs *regs,
1062 struct image_info *infop)
1063 {
1064 regs->pc = infop->entry;
1065 regs->r1 = infop->start_stack;
1066
1067 }
1068
1069 #define ELF_EXEC_PAGESIZE 4096
1070
1071 #define USE_ELF_CORE_DUMP
1072 #define ELF_NREG 38
1073 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1074
1075 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */
1076 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMBState *env)
1077 {
1078 int i, pos = 0;
1079
1080 for (i = 0; i < 32; i++) {
1081 (*regs)[pos++] = tswapreg(env->regs[i]);
1082 }
1083
1084 (*regs)[pos++] = tswapreg(env->pc);
1085 (*regs)[pos++] = tswapreg(mb_cpu_read_msr(env));
1086 (*regs)[pos++] = 0;
1087 (*regs)[pos++] = tswapreg(env->ear);
1088 (*regs)[pos++] = 0;
1089 (*regs)[pos++] = tswapreg(env->esr);
1090 }
1091
1092 #endif /* TARGET_MICROBLAZE */
1093
1094 #ifdef TARGET_NIOS2
1095
1096 #define ELF_START_MMAP 0x80000000
1097
1098 #define elf_check_arch(x) ((x) == EM_ALTERA_NIOS2)
1099
1100 #define ELF_CLASS ELFCLASS32
1101 #define ELF_ARCH EM_ALTERA_NIOS2
1102
1103 static void init_thread(struct target_pt_regs *regs, struct image_info *infop)
1104 {
1105 regs->ea = infop->entry;
1106 regs->sp = infop->start_stack;
1107 regs->estatus = 0x3;
1108 }
1109
1110 #define ELF_EXEC_PAGESIZE 4096
1111
1112 #define USE_ELF_CORE_DUMP
1113 #define ELF_NREG 49
1114 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1115
1116 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */
1117 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1118 const CPUNios2State *env)
1119 {
1120 int i;
1121
1122 (*regs)[0] = -1;
1123 for (i = 1; i < 8; i++) /* r0-r7 */
1124 (*regs)[i] = tswapreg(env->regs[i + 7]);
1125
1126 for (i = 8; i < 16; i++) /* r8-r15 */
1127 (*regs)[i] = tswapreg(env->regs[i - 8]);
1128
1129 for (i = 16; i < 24; i++) /* r16-r23 */
1130 (*regs)[i] = tswapreg(env->regs[i + 7]);
1131 (*regs)[24] = -1; /* R_ET */
1132 (*regs)[25] = -1; /* R_BT */
1133 (*regs)[26] = tswapreg(env->regs[R_GP]);
1134 (*regs)[27] = tswapreg(env->regs[R_SP]);
1135 (*regs)[28] = tswapreg(env->regs[R_FP]);
1136 (*regs)[29] = tswapreg(env->regs[R_EA]);
1137 (*regs)[30] = -1; /* R_SSTATUS */
1138 (*regs)[31] = tswapreg(env->regs[R_RA]);
1139
1140 (*regs)[32] = tswapreg(env->regs[R_PC]);
1141
1142 (*regs)[33] = -1; /* R_STATUS */
1143 (*regs)[34] = tswapreg(env->regs[CR_ESTATUS]);
1144
1145 for (i = 35; i < 49; i++) /* ... */
1146 (*regs)[i] = -1;
1147 }
1148
1149 #endif /* TARGET_NIOS2 */
1150
1151 #ifdef TARGET_OPENRISC
1152
1153 #define ELF_START_MMAP 0x08000000
1154
1155 #define ELF_ARCH EM_OPENRISC
1156 #define ELF_CLASS ELFCLASS32
1157 #define ELF_DATA ELFDATA2MSB
1158
1159 static inline void init_thread(struct target_pt_regs *regs,
1160 struct image_info *infop)
1161 {
1162 regs->pc = infop->entry;
1163 regs->gpr[1] = infop->start_stack;
1164 }
1165
1166 #define USE_ELF_CORE_DUMP
1167 #define ELF_EXEC_PAGESIZE 8192
1168
1169 /* See linux kernel arch/openrisc/include/asm/elf.h. */
1170 #define ELF_NREG 34 /* gprs and pc, sr */
1171 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1172
1173 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1174 const CPUOpenRISCState *env)
1175 {
1176 int i;
1177
1178 for (i = 0; i < 32; i++) {
1179 (*regs)[i] = tswapreg(cpu_get_gpr(env, i));
1180 }
1181 (*regs)[32] = tswapreg(env->pc);
1182 (*regs)[33] = tswapreg(cpu_get_sr(env));
1183 }
1184 #define ELF_HWCAP 0
1185 #define ELF_PLATFORM NULL
1186
1187 #endif /* TARGET_OPENRISC */
1188
1189 #ifdef TARGET_SH4
1190
1191 #define ELF_START_MMAP 0x80000000
1192
1193 #define ELF_CLASS ELFCLASS32
1194 #define ELF_ARCH EM_SH
1195
1196 static inline void init_thread(struct target_pt_regs *regs,
1197 struct image_info *infop)
1198 {
1199 /* Check other registers XXXXX */
1200 regs->pc = infop->entry;
1201 regs->regs[15] = infop->start_stack;
1202 }
1203
1204 /* See linux kernel: arch/sh/include/asm/elf.h. */
1205 #define ELF_NREG 23
1206 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1207
1208 /* See linux kernel: arch/sh/include/asm/ptrace.h. */
1209 enum {
1210 TARGET_REG_PC = 16,
1211 TARGET_REG_PR = 17,
1212 TARGET_REG_SR = 18,
1213 TARGET_REG_GBR = 19,
1214 TARGET_REG_MACH = 20,
1215 TARGET_REG_MACL = 21,
1216 TARGET_REG_SYSCALL = 22
1217 };
1218
1219 static inline void elf_core_copy_regs(target_elf_gregset_t *regs,
1220 const CPUSH4State *env)
1221 {
1222 int i;
1223
1224 for (i = 0; i < 16; i++) {
1225 (*regs)[i] = tswapreg(env->gregs[i]);
1226 }
1227
1228 (*regs)[TARGET_REG_PC] = tswapreg(env->pc);
1229 (*regs)[TARGET_REG_PR] = tswapreg(env->pr);
1230 (*regs)[TARGET_REG_SR] = tswapreg(env->sr);
1231 (*regs)[TARGET_REG_GBR] = tswapreg(env->gbr);
1232 (*regs)[TARGET_REG_MACH] = tswapreg(env->mach);
1233 (*regs)[TARGET_REG_MACL] = tswapreg(env->macl);
1234 (*regs)[TARGET_REG_SYSCALL] = 0; /* FIXME */
1235 }
1236
1237 #define USE_ELF_CORE_DUMP
1238 #define ELF_EXEC_PAGESIZE 4096
1239
1240 enum {
1241 SH_CPU_HAS_FPU = 0x0001, /* Hardware FPU support */
1242 SH_CPU_HAS_P2_FLUSH_BUG = 0x0002, /* Need to flush the cache in P2 area */
1243 SH_CPU_HAS_MMU_PAGE_ASSOC = 0x0004, /* SH3: TLB way selection bit support */
1244 SH_CPU_HAS_DSP = 0x0008, /* SH-DSP: DSP support */
1245 SH_CPU_HAS_PERF_COUNTER = 0x0010, /* Hardware performance counters */
1246 SH_CPU_HAS_PTEA = 0x0020, /* PTEA register */
1247 SH_CPU_HAS_LLSC = 0x0040, /* movli.l/movco.l */
1248 SH_CPU_HAS_L2_CACHE = 0x0080, /* Secondary cache / URAM */
1249 SH_CPU_HAS_OP32 = 0x0100, /* 32-bit instruction support */
1250 SH_CPU_HAS_PTEAEX = 0x0200, /* PTE ASID Extension support */
1251 };
1252
1253 #define ELF_HWCAP get_elf_hwcap()
1254
1255 static uint32_t get_elf_hwcap(void)
1256 {
1257 SuperHCPU *cpu = SUPERH_CPU(thread_cpu);
1258 uint32_t hwcap = 0;
1259
1260 hwcap |= SH_CPU_HAS_FPU;
1261
1262 if (cpu->env.features & SH_FEATURE_SH4A) {
1263 hwcap |= SH_CPU_HAS_LLSC;
1264 }
1265
1266 return hwcap;
1267 }
1268
1269 #endif
1270
1271 #ifdef TARGET_CRIS
1272
1273 #define ELF_START_MMAP 0x80000000
1274
1275 #define ELF_CLASS ELFCLASS32
1276 #define ELF_ARCH EM_CRIS
1277
1278 static inline void init_thread(struct target_pt_regs *regs,
1279 struct image_info *infop)
1280 {
1281 regs->erp = infop->entry;
1282 }
1283
1284 #define ELF_EXEC_PAGESIZE 8192
1285
1286 #endif
1287
1288 #ifdef TARGET_M68K
1289
1290 #define ELF_START_MMAP 0x80000000
1291
1292 #define ELF_CLASS ELFCLASS32
1293 #define ELF_ARCH EM_68K
1294
1295 /* ??? Does this need to do anything?
1296 #define ELF_PLAT_INIT(_r) */
1297
1298 static inline void init_thread(struct target_pt_regs *regs,
1299 struct image_info *infop)
1300 {
1301 regs->usp = infop->start_stack;
1302 regs->sr = 0;
1303 regs->pc = infop->entry;
1304 }
1305
1306 /* See linux kernel: arch/m68k/include/asm/elf.h. */
1307 #define ELF_NREG 20
1308 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1309
1310 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUM68KState *env)
1311 {
1312 (*regs)[0] = tswapreg(env->dregs[1]);
1313 (*regs)[1] = tswapreg(env->dregs[2]);
1314 (*regs)[2] = tswapreg(env->dregs[3]);
1315 (*regs)[3] = tswapreg(env->dregs[4]);
1316 (*regs)[4] = tswapreg(env->dregs[5]);
1317 (*regs)[5] = tswapreg(env->dregs[6]);
1318 (*regs)[6] = tswapreg(env->dregs[7]);
1319 (*regs)[7] = tswapreg(env->aregs[0]);
1320 (*regs)[8] = tswapreg(env->aregs[1]);
1321 (*regs)[9] = tswapreg(env->aregs[2]);
1322 (*regs)[10] = tswapreg(env->aregs[3]);
1323 (*regs)[11] = tswapreg(env->aregs[4]);
1324 (*regs)[12] = tswapreg(env->aregs[5]);
1325 (*regs)[13] = tswapreg(env->aregs[6]);
1326 (*regs)[14] = tswapreg(env->dregs[0]);
1327 (*regs)[15] = tswapreg(env->aregs[7]);
1328 (*regs)[16] = tswapreg(env->dregs[0]); /* FIXME: orig_d0 */
1329 (*regs)[17] = tswapreg(env->sr);
1330 (*regs)[18] = tswapreg(env->pc);
1331 (*regs)[19] = 0; /* FIXME: regs->format | regs->vector */
1332 }
1333
1334 #define USE_ELF_CORE_DUMP
1335 #define ELF_EXEC_PAGESIZE 8192
1336
1337 #endif
1338
1339 #ifdef TARGET_ALPHA
1340
1341 #define ELF_START_MMAP (0x30000000000ULL)
1342
1343 #define ELF_CLASS ELFCLASS64
1344 #define ELF_ARCH EM_ALPHA
1345
1346 static inline void init_thread(struct target_pt_regs *regs,
1347 struct image_info *infop)
1348 {
1349 regs->pc = infop->entry;
1350 regs->ps = 8;
1351 regs->usp = infop->start_stack;
1352 }
1353
1354 #define ELF_EXEC_PAGESIZE 8192
1355
1356 #endif /* TARGET_ALPHA */
1357
1358 #ifdef TARGET_S390X
1359
1360 #define ELF_START_MMAP (0x20000000000ULL)
1361
1362 #define ELF_CLASS ELFCLASS64
1363 #define ELF_DATA ELFDATA2MSB
1364 #define ELF_ARCH EM_S390
1365
1366 #include "elf.h"
1367
1368 #define ELF_HWCAP get_elf_hwcap()
1369
1370 #define GET_FEATURE(_feat, _hwcap) \
1371 do { if (s390_has_feat(_feat)) { hwcap |= _hwcap; } } while (0)
1372
1373 static uint32_t get_elf_hwcap(void)
1374 {
1375 /*
1376 * Let's assume we always have esan3 and zarch.
1377 * 31-bit processes can use 64-bit registers (high gprs).
1378 */
1379 uint32_t hwcap = HWCAP_S390_ESAN3 | HWCAP_S390_ZARCH | HWCAP_S390_HIGH_GPRS;
1380
1381 GET_FEATURE(S390_FEAT_STFLE, HWCAP_S390_STFLE);
1382 GET_FEATURE(S390_FEAT_MSA, HWCAP_S390_MSA);
1383 GET_FEATURE(S390_FEAT_LONG_DISPLACEMENT, HWCAP_S390_LDISP);
1384 GET_FEATURE(S390_FEAT_EXTENDED_IMMEDIATE, HWCAP_S390_EIMM);
1385 if (s390_has_feat(S390_FEAT_EXTENDED_TRANSLATION_3) &&
1386 s390_has_feat(S390_FEAT_ETF3_ENH)) {
1387 hwcap |= HWCAP_S390_ETF3EH;
1388 }
1389 GET_FEATURE(S390_FEAT_VECTOR, HWCAP_S390_VXRS);
1390
1391 return hwcap;
1392 }
1393
1394 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop)
1395 {
1396 regs->psw.addr = infop->entry;
1397 regs->psw.mask = PSW_MASK_64 | PSW_MASK_32;
1398 regs->gprs[15] = infop->start_stack;
1399 }
1400
1401 #endif /* TARGET_S390X */
1402
1403 #ifdef TARGET_RISCV
1404
1405 #define ELF_START_MMAP 0x80000000
1406 #define ELF_ARCH EM_RISCV
1407
1408 #ifdef TARGET_RISCV32
1409 #define ELF_CLASS ELFCLASS32
1410 #else
1411 #define ELF_CLASS ELFCLASS64
1412 #endif
1413
1414 static inline void init_thread(struct target_pt_regs *regs,
1415 struct image_info *infop)
1416 {
1417 regs->sepc = infop->entry;
1418 regs->sp = infop->start_stack;
1419 }
1420
1421 #define ELF_EXEC_PAGESIZE 4096
1422
1423 #endif /* TARGET_RISCV */
1424
1425 #ifdef TARGET_HPPA
1426
1427 #define ELF_START_MMAP 0x80000000
1428 #define ELF_CLASS ELFCLASS32
1429 #define ELF_ARCH EM_PARISC
1430 #define ELF_PLATFORM "PARISC"
1431 #define STACK_GROWS_DOWN 0
1432 #define STACK_ALIGNMENT 64
1433
1434 static inline void init_thread(struct target_pt_regs *regs,
1435 struct image_info *infop)
1436 {
1437 regs->iaoq[0] = infop->entry;
1438 regs->iaoq[1] = infop->entry + 4;
1439 regs->gr[23] = 0;
1440 regs->gr[24] = infop->arg_start;
1441 regs->gr[25] = (infop->arg_end - infop->arg_start) / sizeof(abi_ulong);
1442 /* The top-of-stack contains a linkage buffer. */
1443 regs->gr[30] = infop->start_stack + 64;
1444 regs->gr[31] = infop->entry;
1445 }
1446
1447 #endif /* TARGET_HPPA */
1448
1449 #ifdef TARGET_XTENSA
1450
1451 #define ELF_START_MMAP 0x20000000
1452
1453 #define ELF_CLASS ELFCLASS32
1454 #define ELF_ARCH EM_XTENSA
1455
1456 static inline void init_thread(struct target_pt_regs *regs,
1457 struct image_info *infop)
1458 {
1459 regs->windowbase = 0;
1460 regs->windowstart = 1;
1461 regs->areg[1] = infop->start_stack;
1462 regs->pc = infop->entry;
1463 }
1464
1465 /* See linux kernel: arch/xtensa/include/asm/elf.h. */
1466 #define ELF_NREG 128
1467 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1468
1469 enum {
1470 TARGET_REG_PC,
1471 TARGET_REG_PS,
1472 TARGET_REG_LBEG,
1473 TARGET_REG_LEND,
1474 TARGET_REG_LCOUNT,
1475 TARGET_REG_SAR,
1476 TARGET_REG_WINDOWSTART,
1477 TARGET_REG_WINDOWBASE,
1478 TARGET_REG_THREADPTR,
1479 TARGET_REG_AR0 = 64,
1480 };
1481
1482 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1483 const CPUXtensaState *env)
1484 {
1485 unsigned i;
1486
1487 (*regs)[TARGET_REG_PC] = tswapreg(env->pc);
1488 (*regs)[TARGET_REG_PS] = tswapreg(env->sregs[PS] & ~PS_EXCM);
1489 (*regs)[TARGET_REG_LBEG] = tswapreg(env->sregs[LBEG]);
1490 (*regs)[TARGET_REG_LEND] = tswapreg(env->sregs[LEND]);
1491 (*regs)[TARGET_REG_LCOUNT] = tswapreg(env->sregs[LCOUNT]);
1492 (*regs)[TARGET_REG_SAR] = tswapreg(env->sregs[SAR]);
1493 (*regs)[TARGET_REG_WINDOWSTART] = tswapreg(env->sregs[WINDOW_START]);
1494 (*regs)[TARGET_REG_WINDOWBASE] = tswapreg(env->sregs[WINDOW_BASE]);
1495 (*regs)[TARGET_REG_THREADPTR] = tswapreg(env->uregs[THREADPTR]);
1496 xtensa_sync_phys_from_window((CPUXtensaState *)env);
1497 for (i = 0; i < env->config->nareg; ++i) {
1498 (*regs)[TARGET_REG_AR0 + i] = tswapreg(env->phys_regs[i]);
1499 }
1500 }
1501
1502 #define USE_ELF_CORE_DUMP
1503 #define ELF_EXEC_PAGESIZE 4096
1504
1505 #endif /* TARGET_XTENSA */
1506
1507 #ifdef TARGET_HEXAGON
1508
1509 #define ELF_START_MMAP 0x20000000
1510
1511 #define ELF_CLASS ELFCLASS32
1512 #define ELF_ARCH EM_HEXAGON
1513
1514 static inline void init_thread(struct target_pt_regs *regs,
1515 struct image_info *infop)
1516 {
1517 regs->sepc = infop->entry;
1518 regs->sp = infop->start_stack;
1519 }
1520
1521 #endif /* TARGET_HEXAGON */
1522
1523 #ifndef ELF_PLATFORM
1524 #define ELF_PLATFORM (NULL)
1525 #endif
1526
1527 #ifndef ELF_MACHINE
1528 #define ELF_MACHINE ELF_ARCH
1529 #endif
1530
1531 #ifndef elf_check_arch
1532 #define elf_check_arch(x) ((x) == ELF_ARCH)
1533 #endif
1534
1535 #ifndef elf_check_abi
1536 #define elf_check_abi(x) (1)
1537 #endif
1538
1539 #ifndef ELF_HWCAP
1540 #define ELF_HWCAP 0
1541 #endif
1542
1543 #ifndef STACK_GROWS_DOWN
1544 #define STACK_GROWS_DOWN 1
1545 #endif
1546
1547 #ifndef STACK_ALIGNMENT
1548 #define STACK_ALIGNMENT 16
1549 #endif
1550
1551 #ifdef TARGET_ABI32
1552 #undef ELF_CLASS
1553 #define ELF_CLASS ELFCLASS32
1554 #undef bswaptls
1555 #define bswaptls(ptr) bswap32s(ptr)
1556 #endif
1557
1558 #include "elf.h"
1559
1560 /* We must delay the following stanzas until after "elf.h". */
1561 #if defined(TARGET_AARCH64)
1562
1563 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz,
1564 const uint32_t *data,
1565 struct image_info *info,
1566 Error **errp)
1567 {
1568 if (pr_type == GNU_PROPERTY_AARCH64_FEATURE_1_AND) {
1569 if (pr_datasz != sizeof(uint32_t)) {
1570 error_setg(errp, "Ill-formed GNU_PROPERTY_AARCH64_FEATURE_1_AND");
1571 return false;
1572 }
1573 /* We will extract GNU_PROPERTY_AARCH64_FEATURE_1_BTI later. */
1574 info->note_flags = *data;
1575 }
1576 return true;
1577 }
1578 #define ARCH_USE_GNU_PROPERTY 1
1579
1580 #else
1581
1582 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz,
1583 const uint32_t *data,
1584 struct image_info *info,
1585 Error **errp)
1586 {
1587 g_assert_not_reached();
1588 }
1589 #define ARCH_USE_GNU_PROPERTY 0
1590
1591 #endif
1592
1593 struct exec
1594 {
1595 unsigned int a_info; /* Use macros N_MAGIC, etc for access */
1596 unsigned int a_text; /* length of text, in bytes */
1597 unsigned int a_data; /* length of data, in bytes */
1598 unsigned int a_bss; /* length of uninitialized data area, in bytes */
1599 unsigned int a_syms; /* length of symbol table data in file, in bytes */
1600 unsigned int a_entry; /* start address */
1601 unsigned int a_trsize; /* length of relocation info for text, in bytes */
1602 unsigned int a_drsize; /* length of relocation info for data, in bytes */
1603 };
1604
1605
1606 #define N_MAGIC(exec) ((exec).a_info & 0xffff)
1607 #define OMAGIC 0407
1608 #define NMAGIC 0410
1609 #define ZMAGIC 0413
1610 #define QMAGIC 0314
1611
1612 /* Necessary parameters */
1613 #define TARGET_ELF_EXEC_PAGESIZE \
1614 (((eppnt->p_align & ~qemu_host_page_mask) != 0) ? \
1615 TARGET_PAGE_SIZE : MAX(qemu_host_page_size, TARGET_PAGE_SIZE))
1616 #define TARGET_ELF_PAGELENGTH(_v) ROUND_UP((_v), TARGET_ELF_EXEC_PAGESIZE)
1617 #define TARGET_ELF_PAGESTART(_v) ((_v) & \
1618 ~(abi_ulong)(TARGET_ELF_EXEC_PAGESIZE-1))
1619 #define TARGET_ELF_PAGEOFFSET(_v) ((_v) & (TARGET_ELF_EXEC_PAGESIZE-1))
1620
1621 #define DLINFO_ITEMS 16
1622
1623 static inline void memcpy_fromfs(void * to, const void * from, unsigned long n)
1624 {
1625 memcpy(to, from, n);
1626 }
1627
1628 #ifdef BSWAP_NEEDED
1629 static void bswap_ehdr(struct elfhdr *ehdr)
1630 {
1631 bswap16s(&ehdr->e_type); /* Object file type */
1632 bswap16s(&ehdr->e_machine); /* Architecture */
1633 bswap32s(&ehdr->e_version); /* Object file version */
1634 bswaptls(&ehdr->e_entry); /* Entry point virtual address */
1635 bswaptls(&ehdr->e_phoff); /* Program header table file offset */
1636 bswaptls(&ehdr->e_shoff); /* Section header table file offset */
1637 bswap32s(&ehdr->e_flags); /* Processor-specific flags */
1638 bswap16s(&ehdr->e_ehsize); /* ELF header size in bytes */
1639 bswap16s(&ehdr->e_phentsize); /* Program header table entry size */
1640 bswap16s(&ehdr->e_phnum); /* Program header table entry count */
1641 bswap16s(&ehdr->e_shentsize); /* Section header table entry size */
1642 bswap16s(&ehdr->e_shnum); /* Section header table entry count */
1643 bswap16s(&ehdr->e_shstrndx); /* Section header string table index */
1644 }
1645
1646 static void bswap_phdr(struct elf_phdr *phdr, int phnum)
1647 {
1648 int i;
1649 for (i = 0; i < phnum; ++i, ++phdr) {
1650 bswap32s(&phdr->p_type); /* Segment type */
1651 bswap32s(&phdr->p_flags); /* Segment flags */
1652 bswaptls(&phdr->p_offset); /* Segment file offset */
1653 bswaptls(&phdr->p_vaddr); /* Segment virtual address */
1654 bswaptls(&phdr->p_paddr); /* Segment physical address */
1655 bswaptls(&phdr->p_filesz); /* Segment size in file */
1656 bswaptls(&phdr->p_memsz); /* Segment size in memory */
1657 bswaptls(&phdr->p_align); /* Segment alignment */
1658 }
1659 }
1660
1661 static void bswap_shdr(struct elf_shdr *shdr, int shnum)
1662 {
1663 int i;
1664 for (i = 0; i < shnum; ++i, ++shdr) {
1665 bswap32s(&shdr->sh_name);
1666 bswap32s(&shdr->sh_type);
1667 bswaptls(&shdr->sh_flags);
1668 bswaptls(&shdr->sh_addr);
1669 bswaptls(&shdr->sh_offset);
1670 bswaptls(&shdr->sh_size);
1671 bswap32s(&shdr->sh_link);
1672 bswap32s(&shdr->sh_info);
1673 bswaptls(&shdr->sh_addralign);
1674 bswaptls(&shdr->sh_entsize);
1675 }
1676 }
1677
1678 static void bswap_sym(struct elf_sym *sym)
1679 {
1680 bswap32s(&sym->st_name);
1681 bswaptls(&sym->st_value);
1682 bswaptls(&sym->st_size);
1683 bswap16s(&sym->st_shndx);
1684 }
1685
1686 #ifdef TARGET_MIPS
1687 static void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags)
1688 {
1689 bswap16s(&abiflags->version);
1690 bswap32s(&abiflags->ases);
1691 bswap32s(&abiflags->isa_ext);
1692 bswap32s(&abiflags->flags1);
1693 bswap32s(&abiflags->flags2);
1694 }
1695 #endif
1696 #else
1697 static inline void bswap_ehdr(struct elfhdr *ehdr) { }
1698 static inline void bswap_phdr(struct elf_phdr *phdr, int phnum) { }
1699 static inline void bswap_shdr(struct elf_shdr *shdr, int shnum) { }
1700 static inline void bswap_sym(struct elf_sym *sym) { }
1701 #ifdef TARGET_MIPS
1702 static inline void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags) { }
1703 #endif
1704 #endif
1705
1706 #ifdef USE_ELF_CORE_DUMP
1707 static int elf_core_dump(int, const CPUArchState *);
1708 #endif /* USE_ELF_CORE_DUMP */
1709 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias);
1710
1711 /* Verify the portions of EHDR within E_IDENT for the target.
1712 This can be performed before bswapping the entire header. */
1713 static bool elf_check_ident(struct elfhdr *ehdr)
1714 {
1715 return (ehdr->e_ident[EI_MAG0] == ELFMAG0
1716 && ehdr->e_ident[EI_MAG1] == ELFMAG1
1717 && ehdr->e_ident[EI_MAG2] == ELFMAG2
1718 && ehdr->e_ident[EI_MAG3] == ELFMAG3
1719 && ehdr->e_ident[EI_CLASS] == ELF_CLASS
1720 && ehdr->e_ident[EI_DATA] == ELF_DATA
1721 && ehdr->e_ident[EI_VERSION] == EV_CURRENT);
1722 }
1723
1724 /* Verify the portions of EHDR outside of E_IDENT for the target.
1725 This has to wait until after bswapping the header. */
1726 static bool elf_check_ehdr(struct elfhdr *ehdr)
1727 {
1728 return (elf_check_arch(ehdr->e_machine)
1729 && elf_check_abi(ehdr->e_flags)
1730 && ehdr->e_ehsize == sizeof(struct elfhdr)
1731 && ehdr->e_phentsize == sizeof(struct elf_phdr)
1732 && (ehdr->e_type == ET_EXEC || ehdr->e_type == ET_DYN));
1733 }
1734
1735 /*
1736 * 'copy_elf_strings()' copies argument/envelope strings from user
1737 * memory to free pages in kernel mem. These are in a format ready
1738 * to be put directly into the top of new user memory.
1739 *
1740 */
1741 static abi_ulong copy_elf_strings(int argc, char **argv, char *scratch,
1742 abi_ulong p, abi_ulong stack_limit)
1743 {
1744 char *tmp;
1745 int len, i;
1746 abi_ulong top = p;
1747
1748 if (!p) {
1749 return 0; /* bullet-proofing */
1750 }
1751
1752 if (STACK_GROWS_DOWN) {
1753 int offset = ((p - 1) % TARGET_PAGE_SIZE) + 1;
1754 for (i = argc - 1; i >= 0; --i) {
1755 tmp = argv[i];
1756 if (!tmp) {
1757 fprintf(stderr, "VFS: argc is wrong");
1758 exit(-1);
1759 }
1760 len = strlen(tmp) + 1;
1761 tmp += len;
1762
1763 if (len > (p - stack_limit)) {
1764 return 0;
1765 }
1766 while (len) {
1767 int bytes_to_copy = (len > offset) ? offset : len;
1768 tmp -= bytes_to_copy;
1769 p -= bytes_to_copy;
1770 offset -= bytes_to_copy;
1771 len -= bytes_to_copy;
1772
1773 memcpy_fromfs(scratch + offset, tmp, bytes_to_copy);
1774
1775 if (offset == 0) {
1776 memcpy_to_target(p, scratch, top - p);
1777 top = p;
1778 offset = TARGET_PAGE_SIZE;
1779 }
1780 }
1781 }
1782 if (p != top) {
1783 memcpy_to_target(p, scratch + offset, top - p);
1784 }
1785 } else {
1786 int remaining = TARGET_PAGE_SIZE - (p % TARGET_PAGE_SIZE);
1787 for (i = 0; i < argc; ++i) {
1788 tmp = argv[i];
1789 if (!tmp) {
1790 fprintf(stderr, "VFS: argc is wrong");
1791 exit(-1);
1792 }
1793 len = strlen(tmp) + 1;
1794 if (len > (stack_limit - p)) {
1795 return 0;
1796 }
1797 while (len) {
1798 int bytes_to_copy = (len > remaining) ? remaining : len;
1799
1800 memcpy_fromfs(scratch + (p - top), tmp, bytes_to_copy);
1801
1802 tmp += bytes_to_copy;
1803 remaining -= bytes_to_copy;
1804 p += bytes_to_copy;
1805 len -= bytes_to_copy;
1806
1807 if (remaining == 0) {
1808 memcpy_to_target(top, scratch, p - top);
1809 top = p;
1810 remaining = TARGET_PAGE_SIZE;
1811 }
1812 }
1813 }
1814 if (p != top) {
1815 memcpy_to_target(top, scratch, p - top);
1816 }
1817 }
1818
1819 return p;
1820 }
1821
1822 /* Older linux kernels provide up to MAX_ARG_PAGES (default: 32) of
1823 * argument/environment space. Newer kernels (>2.6.33) allow more,
1824 * dependent on stack size, but guarantee at least 32 pages for
1825 * backwards compatibility.
1826 */
1827 #define STACK_LOWER_LIMIT (32 * TARGET_PAGE_SIZE)
1828
1829 static abi_ulong setup_arg_pages(struct linux_binprm *bprm,
1830 struct image_info *info)
1831 {
1832 abi_ulong size, error, guard;
1833
1834 size = guest_stack_size;
1835 if (size < STACK_LOWER_LIMIT) {
1836 size = STACK_LOWER_LIMIT;
1837 }
1838 guard = TARGET_PAGE_SIZE;
1839 if (guard < qemu_real_host_page_size) {
1840 guard = qemu_real_host_page_size;
1841 }
1842
1843 error = target_mmap(0, size + guard, PROT_READ | PROT_WRITE,
1844 MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
1845 if (error == -1) {
1846 perror("mmap stack");
1847 exit(-1);
1848 }
1849
1850 /* We reserve one extra page at the top of the stack as guard. */
1851 if (STACK_GROWS_DOWN) {
1852 target_mprotect(error, guard, PROT_NONE);
1853 info->stack_limit = error + guard;
1854 return info->stack_limit + size - sizeof(void *);
1855 } else {
1856 target_mprotect(error + size, guard, PROT_NONE);
1857 info->stack_limit = error + size;
1858 return error;
1859 }
1860 }
1861
1862 /* Map and zero the bss. We need to explicitly zero any fractional pages
1863 after the data section (i.e. bss). */
1864 static void zero_bss(abi_ulong elf_bss, abi_ulong last_bss, int prot)
1865 {
1866 uintptr_t host_start, host_map_start, host_end;
1867
1868 last_bss = TARGET_PAGE_ALIGN(last_bss);
1869
1870 /* ??? There is confusion between qemu_real_host_page_size and
1871 qemu_host_page_size here and elsewhere in target_mmap, which
1872 may lead to the end of the data section mapping from the file
1873 not being mapped. At least there was an explicit test and
1874 comment for that here, suggesting that "the file size must
1875 be known". The comment probably pre-dates the introduction
1876 of the fstat system call in target_mmap which does in fact
1877 find out the size. What isn't clear is if the workaround
1878 here is still actually needed. For now, continue with it,
1879 but merge it with the "normal" mmap that would allocate the bss. */
1880
1881 host_start = (uintptr_t) g2h_untagged(elf_bss);
1882 host_end = (uintptr_t) g2h_untagged(last_bss);
1883 host_map_start = REAL_HOST_PAGE_ALIGN(host_start);
1884
1885 if (host_map_start < host_end) {
1886 void *p = mmap((void *)host_map_start, host_end - host_map_start,
1887 prot, MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
1888 if (p == MAP_FAILED) {
1889 perror("cannot mmap brk");
1890 exit(-1);
1891 }
1892 }
1893
1894 /* Ensure that the bss page(s) are valid */
1895 if ((page_get_flags(last_bss-1) & prot) != prot) {
1896 page_set_flags(elf_bss & TARGET_PAGE_MASK, last_bss, prot | PAGE_VALID);
1897 }
1898
1899 if (host_start < host_map_start) {
1900 memset((void *)host_start, 0, host_map_start - host_start);
1901 }
1902 }
1903
1904 #ifdef TARGET_ARM
1905 static int elf_is_fdpic(struct elfhdr *exec)
1906 {
1907 return exec->e_ident[EI_OSABI] == ELFOSABI_ARM_FDPIC;
1908 }
1909 #else
1910 /* Default implementation, always false. */
1911 static int elf_is_fdpic(struct elfhdr *exec)
1912 {
1913 return 0;
1914 }
1915 #endif
1916
1917 static abi_ulong loader_build_fdpic_loadmap(struct image_info *info, abi_ulong sp)
1918 {
1919 uint16_t n;
1920 struct elf32_fdpic_loadseg *loadsegs = info->loadsegs;
1921
1922 /* elf32_fdpic_loadseg */
1923 n = info->nsegs;
1924 while (n--) {
1925 sp -= 12;
1926 put_user_u32(loadsegs[n].addr, sp+0);
1927 put_user_u32(loadsegs[n].p_vaddr, sp+4);
1928 put_user_u32(loadsegs[n].p_memsz, sp+8);
1929 }
1930
1931 /* elf32_fdpic_loadmap */
1932 sp -= 4;
1933 put_user_u16(0, sp+0); /* version */
1934 put_user_u16(info->nsegs, sp+2); /* nsegs */
1935
1936 info->personality = PER_LINUX_FDPIC;
1937 info->loadmap_addr = sp;
1938
1939 return sp;
1940 }
1941
1942 static abi_ulong create_elf_tables(abi_ulong p, int argc, int envc,
1943 struct elfhdr *exec,
1944 struct image_info *info,
1945 struct image_info *interp_info)
1946 {
1947 abi_ulong sp;
1948 abi_ulong u_argc, u_argv, u_envp, u_auxv;
1949 int size;
1950 int i;
1951 abi_ulong u_rand_bytes;
1952 uint8_t k_rand_bytes[16];
1953 abi_ulong u_platform;
1954 const char *k_platform;
1955 const int n = sizeof(elf_addr_t);
1956
1957 sp = p;
1958
1959 /* Needs to be before we load the env/argc/... */
1960 if (elf_is_fdpic(exec)) {
1961 /* Need 4 byte alignment for these structs */
1962 sp &= ~3;
1963 sp = loader_build_fdpic_loadmap(info, sp);
1964 info->other_info = interp_info;
1965 if (interp_info) {
1966 interp_info->other_info = info;
1967 sp = loader_build_fdpic_loadmap(interp_info, sp);
1968 info->interpreter_loadmap_addr = interp_info->loadmap_addr;
1969 info->interpreter_pt_dynamic_addr = interp_info->pt_dynamic_addr;
1970 } else {
1971 info->interpreter_loadmap_addr = 0;
1972 info->interpreter_pt_dynamic_addr = 0;
1973 }
1974 }
1975
1976 u_platform = 0;
1977 k_platform = ELF_PLATFORM;
1978 if (k_platform) {
1979 size_t len = strlen(k_platform) + 1;
1980 if (STACK_GROWS_DOWN) {
1981 sp -= (len + n - 1) & ~(n - 1);
1982 u_platform = sp;
1983 /* FIXME - check return value of memcpy_to_target() for failure */
1984 memcpy_to_target(sp, k_platform, len);
1985 } else {
1986 memcpy_to_target(sp, k_platform, len);
1987 u_platform = sp;
1988 sp += len + 1;
1989 }
1990 }
1991
1992 /* Provide 16 byte alignment for the PRNG, and basic alignment for
1993 * the argv and envp pointers.
1994 */
1995 if (STACK_GROWS_DOWN) {
1996 sp = QEMU_ALIGN_DOWN(sp, 16);
1997 } else {
1998 sp = QEMU_ALIGN_UP(sp, 16);
1999 }
2000
2001 /*
2002 * Generate 16 random bytes for userspace PRNG seeding.
2003 */
2004 qemu_guest_getrandom_nofail(k_rand_bytes, sizeof(k_rand_bytes));
2005 if (STACK_GROWS_DOWN) {
2006 sp -= 16;
2007 u_rand_bytes = sp;
2008 /* FIXME - check return value of memcpy_to_target() for failure */
2009 memcpy_to_target(sp, k_rand_bytes, 16);
2010 } else {
2011 memcpy_to_target(sp, k_rand_bytes, 16);
2012 u_rand_bytes = sp;
2013 sp += 16;
2014 }
2015
2016 size = (DLINFO_ITEMS + 1) * 2;
2017 if (k_platform)
2018 size += 2;
2019 #ifdef DLINFO_ARCH_ITEMS
2020 size += DLINFO_ARCH_ITEMS * 2;
2021 #endif
2022 #ifdef ELF_HWCAP2
2023 size += 2;
2024 #endif
2025 info->auxv_len = size * n;
2026
2027 size += envc + argc + 2;
2028 size += 1; /* argc itself */
2029 size *= n;
2030
2031 /* Allocate space and finalize stack alignment for entry now. */
2032 if (STACK_GROWS_DOWN) {
2033 u_argc = QEMU_ALIGN_DOWN(sp - size, STACK_ALIGNMENT);
2034 sp = u_argc;
2035 } else {
2036 u_argc = sp;
2037 sp = QEMU_ALIGN_UP(sp + size, STACK_ALIGNMENT);
2038 }
2039
2040 u_argv = u_argc + n;
2041 u_envp = u_argv + (argc + 1) * n;
2042 u_auxv = u_envp + (envc + 1) * n;
2043 info->saved_auxv = u_auxv;
2044 info->arg_start = u_argv;
2045 info->arg_end = u_argv + argc * n;
2046
2047 /* This is correct because Linux defines
2048 * elf_addr_t as Elf32_Off / Elf64_Off
2049 */
2050 #define NEW_AUX_ENT(id, val) do { \
2051 put_user_ual(id, u_auxv); u_auxv += n; \
2052 put_user_ual(val, u_auxv); u_auxv += n; \
2053 } while(0)
2054
2055 #ifdef ARCH_DLINFO
2056 /*
2057 * ARCH_DLINFO must come first so platform specific code can enforce
2058 * special alignment requirements on the AUXV if necessary (eg. PPC).
2059 */
2060 ARCH_DLINFO;
2061 #endif
2062 /* There must be exactly DLINFO_ITEMS entries here, or the assert
2063 * on info->auxv_len will trigger.
2064 */
2065 NEW_AUX_ENT(AT_PHDR, (abi_ulong)(info->load_addr + exec->e_phoff));
2066 NEW_AUX_ENT(AT_PHENT, (abi_ulong)(sizeof (struct elf_phdr)));
2067 NEW_AUX_ENT(AT_PHNUM, (abi_ulong)(exec->e_phnum));
2068 if ((info->alignment & ~qemu_host_page_mask) != 0) {
2069 /* Target doesn't support host page size alignment */
2070 NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(TARGET_PAGE_SIZE));
2071 } else {
2072 NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(MAX(TARGET_PAGE_SIZE,
2073 qemu_host_page_size)));
2074 }
2075 NEW_AUX_ENT(AT_BASE, (abi_ulong)(interp_info ? interp_info->load_addr : 0));
2076 NEW_AUX_ENT(AT_FLAGS, (abi_ulong)0);
2077 NEW_AUX_ENT(AT_ENTRY, info->entry);
2078 NEW_AUX_ENT(AT_UID, (abi_ulong) getuid());
2079 NEW_AUX_ENT(AT_EUID, (abi_ulong) geteuid());
2080 NEW_AUX_ENT(AT_GID, (abi_ulong) getgid());
2081 NEW_AUX_ENT(AT_EGID, (abi_ulong) getegid());
2082 NEW_AUX_ENT(AT_HWCAP, (abi_ulong) ELF_HWCAP);
2083 NEW_AUX_ENT(AT_CLKTCK, (abi_ulong) sysconf(_SC_CLK_TCK));
2084 NEW_AUX_ENT(AT_RANDOM, (abi_ulong) u_rand_bytes);
2085 NEW_AUX_ENT(AT_SECURE, (abi_ulong) qemu_getauxval(AT_SECURE));
2086 NEW_AUX_ENT(AT_EXECFN, info->file_string);
2087
2088 #ifdef ELF_HWCAP2
2089 NEW_AUX_ENT(AT_HWCAP2, (abi_ulong) ELF_HWCAP2);
2090 #endif
2091
2092 if (u_platform) {
2093 NEW_AUX_ENT(AT_PLATFORM, u_platform);
2094 }
2095 NEW_AUX_ENT (AT_NULL, 0);
2096 #undef NEW_AUX_ENT
2097
2098 /* Check that our initial calculation of the auxv length matches how much
2099 * we actually put into it.
2100 */
2101 assert(info->auxv_len == u_auxv - info->saved_auxv);
2102
2103 put_user_ual(argc, u_argc);
2104
2105 p = info->arg_strings;
2106 for (i = 0; i < argc; ++i) {
2107 put_user_ual(p, u_argv);
2108 u_argv += n;
2109 p += target_strlen(p) + 1;
2110 }
2111 put_user_ual(0, u_argv);
2112
2113 p = info->env_strings;
2114 for (i = 0; i < envc; ++i) {
2115 put_user_ual(p, u_envp);
2116 u_envp += n;
2117 p += target_strlen(p) + 1;
2118 }
2119 put_user_ual(0, u_envp);
2120
2121 return sp;
2122 }
2123
2124 #ifndef ARM_COMMPAGE
2125 #define ARM_COMMPAGE 0
2126 #define init_guest_commpage() true
2127 #endif
2128
2129 static void pgb_fail_in_use(const char *image_name)
2130 {
2131 error_report("%s: requires virtual address space that is in use "
2132 "(omit the -B option or choose a different value)",
2133 image_name);
2134 exit(EXIT_FAILURE);
2135 }
2136
2137 static void pgb_have_guest_base(const char *image_name, abi_ulong guest_loaddr,
2138 abi_ulong guest_hiaddr, long align)
2139 {
2140 const int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE;
2141 void *addr, *test;
2142
2143 if (!QEMU_IS_ALIGNED(guest_base, align)) {
2144 fprintf(stderr, "Requested guest base %p does not satisfy "
2145 "host minimum alignment (0x%lx)\n",
2146 (void *)guest_base, align);
2147 exit(EXIT_FAILURE);
2148 }
2149
2150 /* Sanity check the guest binary. */
2151 if (reserved_va) {
2152 if (guest_hiaddr > reserved_va) {
2153 error_report("%s: requires more than reserved virtual "
2154 "address space (0x%" PRIx64 " > 0x%lx)",
2155 image_name, (uint64_t)guest_hiaddr, reserved_va);
2156 exit(EXIT_FAILURE);
2157 }
2158 } else {
2159 #if HOST_LONG_BITS < TARGET_ABI_BITS
2160 if ((guest_hiaddr - guest_base) > ~(uintptr_t)0) {
2161 error_report("%s: requires more virtual address space "
2162 "than the host can provide (0x%" PRIx64 ")",
2163 image_name, (uint64_t)guest_hiaddr - guest_base);
2164 exit(EXIT_FAILURE);
2165 }
2166 #endif
2167 }
2168
2169 /*
2170 * Expand the allocation to the entire reserved_va.
2171 * Exclude the mmap_min_addr hole.
2172 */
2173 if (reserved_va) {
2174 guest_loaddr = (guest_base >= mmap_min_addr ? 0
2175 : mmap_min_addr - guest_base);
2176 guest_hiaddr = reserved_va;
2177 }
2178
2179 /* Reserve the address space for the binary, or reserved_va. */
2180 test = g2h_untagged(guest_loaddr);
2181 addr = mmap(test, guest_hiaddr - guest_loaddr, PROT_NONE, flags, -1, 0);
2182 if (test != addr) {
2183 pgb_fail_in_use(image_name);
2184 }
2185 }
2186
2187 /**
2188 * pgd_find_hole_fallback: potential mmap address
2189 * @guest_size: size of available space
2190 * @brk: location of break
2191 * @align: memory alignment
2192 *
2193 * This is a fallback method for finding a hole in the host address
2194 * space if we don't have the benefit of being able to access
2195 * /proc/self/map. It can potentially take a very long time as we can
2196 * only dumbly iterate up the host address space seeing if the
2197 * allocation would work.
2198 */
2199 static uintptr_t pgd_find_hole_fallback(uintptr_t guest_size, uintptr_t brk,
2200 long align, uintptr_t offset)
2201 {
2202 uintptr_t base;
2203
2204 /* Start (aligned) at the bottom and work our way up */
2205 base = ROUND_UP(mmap_min_addr, align);
2206
2207 while (true) {
2208 uintptr_t align_start, end;
2209 align_start = ROUND_UP(base, align);
2210 end = align_start + guest_size + offset;
2211
2212 /* if brk is anywhere in the range give ourselves some room to grow. */
2213 if (align_start <= brk && brk < end) {
2214 base = brk + (16 * MiB);
2215 continue;
2216 } else if (align_start + guest_size < align_start) {
2217 /* we have run out of space */
2218 return -1;
2219 } else {
2220 int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE |
2221 MAP_FIXED_NOREPLACE;
2222 void * mmap_start = mmap((void *) align_start, guest_size,
2223 PROT_NONE, flags, -1, 0);
2224 if (mmap_start != MAP_FAILED) {
2225 munmap(mmap_start, guest_size);
2226 if (mmap_start == (void *) align_start) {
2227 return (uintptr_t) mmap_start + offset;
2228 }
2229 }
2230 base += qemu_host_page_size;
2231 }
2232 }
2233 }
2234
2235 /* Return value for guest_base, or -1 if no hole found. */
2236 static uintptr_t pgb_find_hole(uintptr_t guest_loaddr, uintptr_t guest_size,
2237 long align, uintptr_t offset)
2238 {
2239 GSList *maps, *iter;
2240 uintptr_t this_start, this_end, next_start, brk;
2241 intptr_t ret = -1;
2242
2243 assert(QEMU_IS_ALIGNED(guest_loaddr, align));
2244
2245 maps = read_self_maps();
2246
2247 /* Read brk after we've read the maps, which will malloc. */
2248 brk = (uintptr_t)sbrk(0);
2249
2250 if (!maps) {
2251 ret = pgd_find_hole_fallback(guest_size, brk, align, offset);
2252 return ret == -1 ? -1 : ret - guest_loaddr;
2253 }
2254
2255 /* The first hole is before the first map entry. */
2256 this_start = mmap_min_addr;
2257
2258 for (iter = maps; iter;
2259 this_start = next_start, iter = g_slist_next(iter)) {
2260 uintptr_t align_start, hole_size;
2261
2262 this_end = ((MapInfo *)iter->data)->start;
2263 next_start = ((MapInfo *)iter->data)->end;
2264 align_start = ROUND_UP(this_start + offset, align);
2265
2266 /* Skip holes that are too small. */
2267 if (align_start >= this_end) {
2268 continue;
2269 }
2270 hole_size = this_end - align_start;
2271 if (hole_size < guest_size) {
2272 continue;
2273 }
2274
2275 /* If this hole contains brk, give ourselves some room to grow. */
2276 if (this_start <= brk && brk < this_end) {
2277 hole_size -= guest_size;
2278 if (sizeof(uintptr_t) == 8 && hole_size >= 1 * GiB) {
2279 align_start += 1 * GiB;
2280 } else if (hole_size >= 16 * MiB) {
2281 align_start += 16 * MiB;
2282 } else {
2283 align_start = (this_end - guest_size) & -align;
2284 if (align_start < this_start) {
2285 continue;
2286 }
2287 }
2288 }
2289
2290 /* Record the lowest successful match. */
2291 if (ret < 0) {
2292 ret = align_start - guest_loaddr;
2293 }
2294 /* If this hole contains the identity map, select it. */
2295 if (align_start <= guest_loaddr &&
2296 guest_loaddr + guest_size <= this_end) {
2297 ret = 0;
2298 }
2299 /* If this hole ends above the identity map, stop looking. */
2300 if (this_end >= guest_loaddr) {
2301 break;
2302 }
2303 }
2304 free_self_maps(maps);
2305
2306 return ret;
2307 }
2308
2309 static void pgb_static(const char *image_name, abi_ulong orig_loaddr,
2310 abi_ulong orig_hiaddr, long align)
2311 {
2312 uintptr_t loaddr = orig_loaddr;
2313 uintptr_t hiaddr = orig_hiaddr;
2314 uintptr_t offset = 0;
2315 uintptr_t addr;
2316
2317 if (hiaddr != orig_hiaddr) {
2318 error_report("%s: requires virtual address space that the "
2319 "host cannot provide (0x%" PRIx64 ")",
2320 image_name, (uint64_t)orig_hiaddr);
2321 exit(EXIT_FAILURE);
2322 }
2323
2324 loaddr &= -align;
2325 if (ARM_COMMPAGE) {
2326 /*
2327 * Extend the allocation to include the commpage.
2328 * For a 64-bit host, this is just 4GiB; for a 32-bit host we
2329 * need to ensure there is space bellow the guest_base so we
2330 * can map the commpage in the place needed when the address
2331 * arithmetic wraps around.
2332 */
2333 if (sizeof(uintptr_t) == 8 || loaddr >= 0x80000000u) {
2334 hiaddr = (uintptr_t) 4 << 30;
2335 } else {
2336 offset = -(ARM_COMMPAGE & -align);
2337 }
2338 }
2339
2340 addr = pgb_find_hole(loaddr, hiaddr - loaddr, align, offset);
2341 if (addr == -1) {
2342 /*
2343 * If ARM_COMMPAGE, there *might* be a non-consecutive allocation
2344 * that can satisfy both. But as the normal arm32 link base address
2345 * is ~32k, and we extend down to include the commpage, making the
2346 * overhead only ~96k, this is unlikely.
2347 */
2348 error_report("%s: Unable to allocate %#zx bytes of "
2349 "virtual address space", image_name,
2350 (size_t)(hiaddr - loaddr));
2351 exit(EXIT_FAILURE);
2352 }
2353
2354 guest_base = addr;
2355 }
2356
2357 static void pgb_dynamic(const char *image_name, long align)
2358 {
2359 /*
2360 * The executable is dynamic and does not require a fixed address.
2361 * All we need is a commpage that satisfies align.
2362 * If we do not need a commpage, leave guest_base == 0.
2363 */
2364 if (ARM_COMMPAGE) {
2365 uintptr_t addr, commpage;
2366
2367 /* 64-bit hosts should have used reserved_va. */
2368 assert(sizeof(uintptr_t) == 4);
2369
2370 /*
2371 * By putting the commpage at the first hole, that puts guest_base
2372 * just above that, and maximises the positive guest addresses.
2373 */
2374 commpage = ARM_COMMPAGE & -align;
2375 addr = pgb_find_hole(commpage, -commpage, align, 0);
2376 assert(addr != -1);
2377 guest_base = addr;
2378 }
2379 }
2380
2381 static void pgb_reserved_va(const char *image_name, abi_ulong guest_loaddr,
2382 abi_ulong guest_hiaddr, long align)
2383 {
2384 int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE;
2385 void *addr, *test;
2386
2387 if (guest_hiaddr > reserved_va) {
2388 error_report("%s: requires more than reserved virtual "
2389 "address space (0x%" PRIx64 " > 0x%lx)",
2390 image_name, (uint64_t)guest_hiaddr, reserved_va);
2391 exit(EXIT_FAILURE);
2392 }
2393
2394 /* Widen the "image" to the entire reserved address space. */
2395 pgb_static(image_name, 0, reserved_va, align);
2396
2397 /* osdep.h defines this as 0 if it's missing */
2398 flags |= MAP_FIXED_NOREPLACE;
2399
2400 /* Reserve the memory on the host. */
2401 assert(guest_base != 0);
2402 test = g2h_untagged(0);
2403 addr = mmap(test, reserved_va, PROT_NONE, flags, -1, 0);
2404 if (addr == MAP_FAILED || addr != test) {
2405 error_report("Unable to reserve 0x%lx bytes of virtual address "
2406 "space at %p (%s) for use as guest address space (check your"
2407 "virtual memory ulimit setting, min_mmap_addr or reserve less "
2408 "using -R option)", reserved_va, test, strerror(errno));
2409 exit(EXIT_FAILURE);
2410 }
2411 }
2412
2413 void probe_guest_base(const char *image_name, abi_ulong guest_loaddr,
2414 abi_ulong guest_hiaddr)
2415 {
2416 /* In order to use host shmat, we must be able to honor SHMLBA. */
2417 uintptr_t align = MAX(SHMLBA, qemu_host_page_size);
2418
2419 if (have_guest_base) {
2420 pgb_have_guest_base(image_name, guest_loaddr, guest_hiaddr, align);
2421 } else if (reserved_va) {
2422 pgb_reserved_va(image_name, guest_loaddr, guest_hiaddr, align);
2423 } else if (guest_loaddr) {
2424 pgb_static(image_name, guest_loaddr, guest_hiaddr, align);
2425 } else {
2426 pgb_dynamic(image_name, align);
2427 }
2428
2429 /* Reserve and initialize the commpage. */
2430 if (!init_guest_commpage()) {
2431 /*
2432 * With have_guest_base, the user has selected the address and
2433 * we are trying to work with that. Otherwise, we have selected
2434 * free space and init_guest_commpage must succeeded.
2435 */
2436 assert(have_guest_base);
2437 pgb_fail_in_use(image_name);
2438 }
2439
2440 assert(QEMU_IS_ALIGNED(guest_base, align));
2441 qemu_log_mask(CPU_LOG_PAGE, "Locating guest address space "
2442 "@ 0x%" PRIx64 "\n", (uint64_t)guest_base);
2443 }
2444
2445 enum {
2446 /* The string "GNU\0" as a magic number. */
2447 GNU0_MAGIC = const_le32('G' | 'N' << 8 | 'U' << 16),
2448 NOTE_DATA_SZ = 1 * KiB,
2449 NOTE_NAME_SZ = 4,
2450 ELF_GNU_PROPERTY_ALIGN = ELF_CLASS == ELFCLASS32 ? 4 : 8,
2451 };
2452
2453 /*
2454 * Process a single gnu_property entry.
2455 * Return false for error.
2456 */
2457 static bool parse_elf_property(const uint32_t *data, int *off, int datasz,
2458 struct image_info *info, bool have_prev_type,
2459 uint32_t *prev_type, Error **errp)
2460 {
2461 uint32_t pr_type, pr_datasz, step;
2462
2463 if (*off > datasz || !QEMU_IS_ALIGNED(*off, ELF_GNU_PROPERTY_ALIGN)) {
2464 goto error_data;
2465 }
2466 datasz -= *off;
2467 data += *off / sizeof(uint32_t);
2468
2469 if (datasz < 2 * sizeof(uint32_t)) {
2470 goto error_data;
2471 }
2472 pr_type = data[0];
2473 pr_datasz = data[1];
2474 data += 2;
2475 datasz -= 2 * sizeof(uint32_t);
2476 step = ROUND_UP(pr_datasz, ELF_GNU_PROPERTY_ALIGN);
2477 if (step > datasz) {
2478 goto error_data;
2479 }
2480
2481 /* Properties are supposed to be unique and sorted on pr_type. */
2482 if (have_prev_type && pr_type <= *prev_type) {
2483 if (pr_type == *prev_type) {
2484 error_setg(errp, "Duplicate property in PT_GNU_PROPERTY");
2485 } else {
2486 error_setg(errp, "Unsorted property in PT_GNU_PROPERTY");
2487 }
2488 return false;
2489 }
2490 *prev_type = pr_type;
2491
2492 if (!arch_parse_elf_property(pr_type, pr_datasz, data, info, errp)) {
2493 return false;
2494 }
2495
2496 *off += 2 * sizeof(uint32_t) + step;
2497 return true;
2498
2499 error_data:
2500 error_setg(errp, "Ill-formed property in PT_GNU_PROPERTY");
2501 return false;
2502 }
2503
2504 /* Process NT_GNU_PROPERTY_TYPE_0. */
2505 static bool parse_elf_properties(int image_fd,
2506 struct image_info *info,
2507 const struct elf_phdr *phdr,
2508 char bprm_buf[BPRM_BUF_SIZE],
2509 Error **errp)
2510 {
2511 union {
2512 struct elf_note nhdr;
2513 uint32_t data[NOTE_DATA_SZ / sizeof(uint32_t)];
2514 } note;
2515
2516 int n, off, datasz;
2517 bool have_prev_type;
2518 uint32_t prev_type;
2519
2520 /* Unless the arch requires properties, ignore them. */
2521 if (!ARCH_USE_GNU_PROPERTY) {
2522 return true;
2523 }
2524
2525 /* If the properties are crazy large, that's too bad. */
2526 n = phdr->p_filesz;
2527 if (n > sizeof(note)) {
2528 error_setg(errp, "PT_GNU_PROPERTY too large");
2529 return false;
2530 }
2531 if (n < sizeof(note.nhdr)) {
2532 error_setg(errp, "PT_GNU_PROPERTY too small");
2533 return false;
2534 }
2535
2536 if (phdr->p_offset + n <= BPRM_BUF_SIZE) {
2537 memcpy(&note, bprm_buf + phdr->p_offset, n);
2538 } else {
2539 ssize_t len = pread(image_fd, &note, n, phdr->p_offset);
2540 if (len != n) {
2541 error_setg_errno(errp, errno, "Error reading file header");
2542 return false;
2543 }
2544 }
2545
2546 /*
2547 * The contents of a valid PT_GNU_PROPERTY is a sequence
2548 * of uint32_t -- swap them all now.
2549 */
2550 #ifdef BSWAP_NEEDED
2551 for (int i = 0; i < n / 4; i++) {
2552 bswap32s(note.data + i);
2553 }
2554 #endif
2555
2556 /*
2557 * Note that nhdr is 3 words, and that the "name" described by namesz
2558 * immediately follows nhdr and is thus at the 4th word. Further, all
2559 * of the inputs to the kernel's round_up are multiples of 4.
2560 */
2561 if (note.nhdr.n_type != NT_GNU_PROPERTY_TYPE_0 ||
2562 note.nhdr.n_namesz != NOTE_NAME_SZ ||
2563 note.data[3] != GNU0_MAGIC) {
2564 error_setg(errp, "Invalid note in PT_GNU_PROPERTY");
2565 return false;
2566 }
2567 off = sizeof(note.nhdr) + NOTE_NAME_SZ;
2568
2569 datasz = note.nhdr.n_descsz + off;
2570 if (datasz > n) {
2571 error_setg(errp, "Invalid note size in PT_GNU_PROPERTY");
2572 return false;
2573 }
2574
2575 have_prev_type = false;
2576 prev_type = 0;
2577 while (1) {
2578 if (off == datasz) {
2579 return true; /* end, exit ok */
2580 }
2581 if (!parse_elf_property(note.data, &off, datasz, info,
2582 have_prev_type, &prev_type, errp)) {
2583 return false;
2584 }
2585 have_prev_type = true;
2586 }
2587 }
2588
2589 /* Load an ELF image into the address space.
2590
2591 IMAGE_NAME is the filename of the image, to use in error messages.
2592 IMAGE_FD is the open file descriptor for the image.
2593
2594 BPRM_BUF is a copy of the beginning of the file; this of course
2595 contains the elf file header at offset 0. It is assumed that this
2596 buffer is sufficiently aligned to present no problems to the host
2597 in accessing data at aligned offsets within the buffer.
2598
2599 On return: INFO values will be filled in, as necessary or available. */
2600
2601 static void load_elf_image(const char *image_name, int image_fd,
2602 struct image_info *info, char **pinterp_name,
2603 char bprm_buf[BPRM_BUF_SIZE])
2604 {
2605 struct elfhdr *ehdr = (struct elfhdr *)bprm_buf;
2606 struct elf_phdr *phdr;
2607 abi_ulong load_addr, load_bias, loaddr, hiaddr, error;
2608 int i, retval, prot_exec;
2609 Error *err = NULL;
2610
2611 /* First of all, some simple consistency checks */
2612 if (!elf_check_ident(ehdr)) {
2613 error_setg(&err, "Invalid ELF image for this architecture");
2614 goto exit_errmsg;
2615 }
2616 bswap_ehdr(ehdr);
2617 if (!elf_check_ehdr(ehdr)) {
2618 error_setg(&err, "Invalid ELF image for this architecture");
2619 goto exit_errmsg;
2620 }
2621
2622 i = ehdr->e_phnum * sizeof(struct elf_phdr);
2623 if (ehdr->e_phoff + i <= BPRM_BUF_SIZE) {
2624 phdr = (struct elf_phdr *)(bprm_buf + ehdr->e_phoff);
2625 } else {
2626 phdr = (struct elf_phdr *) alloca(i);
2627 retval = pread(image_fd, phdr, i, ehdr->e_phoff);
2628 if (retval != i) {
2629 goto exit_read;
2630 }
2631 }
2632 bswap_phdr(phdr, ehdr->e_phnum);
2633
2634 info->nsegs = 0;
2635 info->pt_dynamic_addr = 0;
2636
2637 mmap_lock();
2638
2639 /*
2640 * Find the maximum size of the image and allocate an appropriate
2641 * amount of memory to handle that. Locate the interpreter, if any.
2642 */
2643 loaddr = -1, hiaddr = 0;
2644 info->alignment = 0;
2645 for (i = 0; i < ehdr->e_phnum; ++i) {
2646 struct elf_phdr *eppnt = phdr + i;
2647 if (eppnt->p_type == PT_LOAD) {
2648 abi_ulong a = eppnt->p_vaddr - eppnt->p_offset;
2649 if (a < loaddr) {
2650 loaddr = a;
2651 }
2652 a = eppnt->p_vaddr + eppnt->p_memsz;
2653 if (a > hiaddr) {
2654 hiaddr = a;
2655 }
2656 ++info->nsegs;
2657 info->alignment |= eppnt->p_align;
2658 } else if (eppnt->p_type == PT_INTERP && pinterp_name) {
2659 g_autofree char *interp_name = NULL;
2660
2661 if (*pinterp_name) {
2662 error_setg(&err, "Multiple PT_INTERP entries");
2663 goto exit_errmsg;
2664 }
2665
2666 interp_name = g_malloc(eppnt->p_filesz);
2667
2668 if (eppnt->p_offset + eppnt->p_filesz <= BPRM_BUF_SIZE) {
2669 memcpy(interp_name, bprm_buf + eppnt->p_offset,
2670 eppnt->p_filesz);
2671 } else {
2672 retval = pread(image_fd, interp_name, eppnt->p_filesz,
2673 eppnt->p_offset);
2674 if (retval != eppnt->p_filesz) {
2675 goto exit_read;
2676 }
2677 }
2678 if (interp_name[eppnt->p_filesz - 1] != 0) {
2679 error_setg(&err, "Invalid PT_INTERP entry");
2680 goto exit_errmsg;
2681 }
2682 *pinterp_name = g_steal_pointer(&interp_name);
2683 } else if (eppnt->p_type == PT_GNU_PROPERTY) {
2684 if (!parse_elf_properties(image_fd, info, eppnt, bprm_buf, &err)) {
2685 goto exit_errmsg;
2686 }
2687 }
2688 }
2689
2690 if (pinterp_name != NULL) {
2691 /*
2692 * This is the main executable.
2693 *
2694 * Reserve extra space for brk.
2695 * We hold on to this space while placing the interpreter
2696 * and the stack, lest they be placed immediately after
2697 * the data segment and block allocation from the brk.
2698 *
2699 * 16MB is chosen as "large enough" without being so large
2700 * as to allow the result to not fit with a 32-bit guest on
2701 * a 32-bit host.
2702 */
2703 info->reserve_brk = 16 * MiB;
2704 hiaddr += info->reserve_brk;
2705
2706 if (ehdr->e_type == ET_EXEC) {
2707 /*
2708 * Make sure that the low address does not conflict with
2709 * MMAP_MIN_ADDR or the QEMU application itself.
2710 */
2711 probe_guest_base(image_name, loaddr, hiaddr);
2712 } else {
2713 /*
2714 * The binary is dynamic, but we still need to
2715 * select guest_base. In this case we pass a size.
2716 */
2717 probe_guest_base(image_name, 0, hiaddr - loaddr);
2718 }
2719 }
2720
2721 /*
2722 * Reserve address space for all of this.
2723 *
2724 * In the case of ET_EXEC, we supply MAP_FIXED so that we get
2725 * exactly the address range that is required.
2726 *
2727 * Otherwise this is ET_DYN, and we are searching for a location
2728 * that can hold the memory space required. If the image is
2729 * pre-linked, LOADDR will be non-zero, and the kernel should
2730 * honor that address if it happens to be free.
2731 *
2732 * In both cases, we will overwrite pages in this range with mappings
2733 * from the executable.
2734 */
2735 load_addr = target_mmap(loaddr, hiaddr - loaddr, PROT_NONE,
2736 MAP_PRIVATE | MAP_ANON | MAP_NORESERVE |
2737 (ehdr->e_type == ET_EXEC ? MAP_FIXED : 0),
2738 -1, 0);
2739 if (load_addr == -1) {
2740 goto exit_mmap;
2741 }
2742 load_bias = load_addr - loaddr;
2743
2744 if (elf_is_fdpic(ehdr)) {
2745 struct elf32_fdpic_loadseg *loadsegs = info->loadsegs =
2746 g_malloc(sizeof(*loadsegs) * info->nsegs);
2747
2748 for (i = 0; i < ehdr->e_phnum; ++i) {
2749 switch (phdr[i].p_type) {
2750 case PT_DYNAMIC:
2751 info->pt_dynamic_addr = phdr[i].p_vaddr + load_bias;
2752 break;
2753 case PT_LOAD:
2754 loadsegs->addr = phdr[i].p_vaddr + load_bias;
2755 loadsegs->p_vaddr = phdr[i].p_vaddr;
2756 loadsegs->p_memsz = phdr[i].p_memsz;
2757 ++loadsegs;
2758 break;
2759 }
2760 }
2761 }
2762
2763 info->load_bias = load_bias;
2764 info->code_offset = load_bias;
2765 info->data_offset = load_bias;
2766 info->load_addr = load_addr;
2767 info->entry = ehdr->e_entry + load_bias;
2768 info->start_code = -1;
2769 info->end_code = 0;
2770 info->start_data = -1;
2771 info->end_data = 0;
2772 info->brk = 0;
2773 info->elf_flags = ehdr->e_flags;
2774
2775 prot_exec = PROT_EXEC;
2776 #ifdef TARGET_AARCH64
2777 /*
2778 * If the BTI feature is present, this indicates that the executable
2779 * pages of the startup binary should be mapped with PROT_BTI, so that
2780 * branch targets are enforced.
2781 *
2782 * The startup binary is either the interpreter or the static executable.
2783 * The interpreter is responsible for all pages of a dynamic executable.
2784 *
2785 * Elf notes are backward compatible to older cpus.
2786 * Do not enable BTI unless it is supported.
2787 */
2788 if ((info->note_flags & GNU_PROPERTY_AARCH64_FEATURE_1_BTI)
2789 && (pinterp_name == NULL || *pinterp_name == 0)
2790 && cpu_isar_feature(aa64_bti, ARM_CPU(thread_cpu))) {
2791 prot_exec |= TARGET_PROT_BTI;
2792 }
2793 #endif
2794
2795 for (i = 0; i < ehdr->e_phnum; i++) {
2796 struct elf_phdr *eppnt = phdr + i;
2797 if (eppnt->p_type == PT_LOAD) {
2798 abi_ulong vaddr, vaddr_po, vaddr_ps, vaddr_ef, vaddr_em, vaddr_len;
2799 int elf_prot = 0;
2800
2801 if (eppnt->p_flags & PF_R) {
2802 elf_prot |= PROT_READ;
2803 }
2804 if (eppnt->p_flags & PF_W) {
2805 elf_prot |= PROT_WRITE;
2806 }
2807 if (eppnt->p_flags & PF_X) {
2808 elf_prot |= prot_exec;
2809 }
2810
2811 vaddr = load_bias + eppnt->p_vaddr;
2812 vaddr_po = TARGET_ELF_PAGEOFFSET(vaddr);
2813 vaddr_ps = TARGET_ELF_PAGESTART(vaddr);
2814
2815 vaddr_ef = vaddr + eppnt->p_filesz;
2816 vaddr_em = vaddr + eppnt->p_memsz;
2817
2818 /*
2819 * Some segments may be completely empty, with a non-zero p_memsz
2820 * but no backing file segment.
2821 */
2822 if (eppnt->p_filesz != 0) {
2823 vaddr_len = TARGET_ELF_PAGELENGTH(eppnt->p_filesz + vaddr_po);
2824 error = target_mmap(vaddr_ps, vaddr_len, elf_prot,
2825 MAP_PRIVATE | MAP_FIXED,
2826 image_fd, eppnt->p_offset - vaddr_po);
2827
2828 if (error == -1) {
2829 goto exit_mmap;
2830 }
2831
2832 /*
2833 * If the load segment requests extra zeros (e.g. bss), map it.
2834 */
2835 if (eppnt->p_filesz < eppnt->p_memsz) {
2836 zero_bss(vaddr_ef, vaddr_em, elf_prot);
2837 }
2838 } else if (eppnt->p_memsz != 0) {
2839 vaddr_len = TARGET_ELF_PAGELENGTH(eppnt->p_memsz + vaddr_po);
2840 error = target_mmap(vaddr_ps, vaddr_len, elf_prot,
2841 MAP_PRIVATE | MAP_FIXED | MAP_ANONYMOUS,
2842 -1, 0);
2843
2844 if (error == -1) {
2845 goto exit_mmap;
2846 }
2847 }
2848
2849 /* Find the full program boundaries. */
2850 if (elf_prot & PROT_EXEC) {
2851 if (vaddr < info->start_code) {
2852 info->start_code = vaddr;
2853 }
2854 if (vaddr_ef > info->end_code) {
2855 info->end_code = vaddr_ef;
2856 }
2857 }
2858 if (elf_prot & PROT_WRITE) {
2859 if (vaddr < info->start_data) {
2860 info->start_data = vaddr;
2861 }
2862 if (vaddr_ef > info->end_data) {
2863 info->end_data = vaddr_ef;
2864 }
2865 }
2866 if (vaddr_em > info->brk) {
2867 info->brk = vaddr_em;
2868 }
2869 #ifdef TARGET_MIPS
2870 } else if (eppnt->p_type == PT_MIPS_ABIFLAGS) {
2871 Mips_elf_abiflags_v0 abiflags;
2872 if (eppnt->p_filesz < sizeof(Mips_elf_abiflags_v0)) {
2873 error_setg(&err, "Invalid PT_MIPS_ABIFLAGS entry");
2874 goto exit_errmsg;
2875 }
2876 if (eppnt->p_offset + eppnt->p_filesz <= BPRM_BUF_SIZE) {
2877 memcpy(&abiflags, bprm_buf + eppnt->p_offset,
2878 sizeof(Mips_elf_abiflags_v0));
2879 } else {
2880 retval = pread(image_fd, &abiflags, sizeof(Mips_elf_abiflags_v0),
2881 eppnt->p_offset);
2882 if (retval != sizeof(Mips_elf_abiflags_v0)) {
2883 goto exit_read;
2884 }
2885 }
2886 bswap_mips_abiflags(&abiflags);
2887 info->fp_abi = abiflags.fp_abi;
2888 #endif
2889 }
2890 }
2891
2892 if (info->end_data == 0) {
2893 info->start_data = info->end_code;
2894 info->end_data = info->end_code;
2895 }
2896
2897 if (qemu_log_enabled()) {
2898 load_symbols(ehdr, image_fd, load_bias);
2899 }
2900
2901 mmap_unlock();
2902
2903 close(image_fd);
2904 return;
2905
2906 exit_read:
2907 if (retval >= 0) {
2908 error_setg(&err, "Incomplete read of file header");
2909 } else {
2910 error_setg_errno(&err, errno, "Error reading file header");
2911 }
2912 goto exit_errmsg;
2913 exit_mmap:
2914 error_setg_errno(&err, errno, "Error mapping file");
2915 goto exit_errmsg;
2916 exit_errmsg:
2917 error_reportf_err(err, "%s: ", image_name);
2918 exit(-1);
2919 }
2920
2921 static void load_elf_interp(const char *filename, struct image_info *info,
2922 char bprm_buf[BPRM_BUF_SIZE])
2923 {
2924 int fd, retval;
2925 Error *err = NULL;
2926
2927 fd = open(path(filename), O_RDONLY);
2928 if (fd < 0) {
2929 error_setg_file_open(&err, errno, filename);
2930 error_report_err(err);
2931 exit(-1);
2932 }
2933
2934 retval = read(fd, bprm_buf, BPRM_BUF_SIZE);
2935 if (retval < 0) {
2936 error_setg_errno(&err, errno, "Error reading file header");
2937 error_reportf_err(err, "%s: ", filename);
2938 exit(-1);
2939 }
2940
2941 if (retval < BPRM_BUF_SIZE) {
2942 memset(bprm_buf + retval, 0, BPRM_BUF_SIZE - retval);
2943 }
2944
2945 load_elf_image(filename, fd, info, NULL, bprm_buf);
2946 }
2947
2948 static int symfind(const void *s0, const void *s1)
2949 {
2950 target_ulong addr = *(target_ulong *)s0;
2951 struct elf_sym *sym = (struct elf_sym *)s1;
2952 int result = 0;
2953 if (addr < sym->st_value) {
2954 result = -1;
2955 } else if (addr >= sym->st_value + sym->st_size) {
2956 result = 1;
2957 }
2958 return result;
2959 }
2960
2961 static const char *lookup_symbolxx(struct syminfo *s, target_ulong orig_addr)
2962 {
2963 #if ELF_CLASS == ELFCLASS32
2964 struct elf_sym *syms = s->disas_symtab.elf32;
2965 #else
2966 struct elf_sym *syms = s->disas_symtab.elf64;
2967 #endif
2968
2969 // binary search
2970 struct elf_sym *sym;
2971
2972 sym = bsearch(&orig_addr, syms, s->disas_num_syms, sizeof(*syms), symfind);
2973 if (sym != NULL) {
2974 return s->disas_strtab + sym->st_name;
2975 }
2976
2977 return "";
2978 }
2979
2980 /* FIXME: This should use elf_ops.h */
2981 static int symcmp(const void *s0, const void *s1)
2982 {
2983 struct elf_sym *sym0 = (struct elf_sym *)s0;
2984 struct elf_sym *sym1 = (struct elf_sym *)s1;
2985 return (sym0->st_value < sym1->st_value)
2986 ? -1
2987 : ((sym0->st_value > sym1->st_value) ? 1 : 0);
2988 }
2989
2990 /* Best attempt to load symbols from this ELF object. */
2991 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias)
2992 {
2993 int i, shnum, nsyms, sym_idx = 0, str_idx = 0;
2994 uint64_t segsz;
2995 struct elf_shdr *shdr;
2996 char *strings = NULL;
2997 struct syminfo *s = NULL;
2998 struct elf_sym *new_syms, *syms = NULL;
2999
3000 shnum = hdr->e_shnum;
3001 i = shnum * sizeof(struct elf_shdr);
3002 shdr = (struct elf_shdr *)alloca(i);
3003 if (pread(fd, shdr, i, hdr->e_shoff) != i) {
3004 return;
3005 }
3006
3007 bswap_shdr(shdr, shnum);
3008 for (i = 0; i < shnum; ++i) {
3009 if (shdr[i].sh_type == SHT_SYMTAB) {
3010 sym_idx = i;
3011 str_idx = shdr[i].sh_link;
3012 goto found;
3013 }
3014 }
3015
3016 /* There will be no symbol table if the file was stripped. */
3017 return;
3018
3019 found:
3020 /* Now know where the strtab and symtab are. Snarf them. */
3021 s = g_try_new(struct syminfo, 1);
3022 if (!s) {
3023 goto give_up;
3024 }
3025
3026 segsz = shdr[str_idx].sh_size;
3027 s->disas_strtab = strings = g_try_malloc(segsz);
3028 if (!strings ||
3029 pread(fd, strings, segsz, shdr[str_idx].sh_offset) != segsz) {
3030 goto give_up;
3031 }
3032
3033 segsz = shdr[sym_idx].sh_size;
3034 syms = g_try_malloc(segsz);
3035 if (!syms || pread(fd, syms, segsz, shdr[sym_idx].sh_offset) != segsz) {
3036 goto give_up;
3037 }
3038
3039 if (segsz / sizeof(struct elf_sym) > INT_MAX) {
3040 /* Implausibly large symbol table: give up rather than ploughing
3041 * on with the number of symbols calculation overflowing
3042 */
3043 goto give_up;
3044 }
3045 nsyms = segsz / sizeof(struct elf_sym);
3046 for (i = 0; i < nsyms; ) {
3047 bswap_sym(syms + i);
3048 /* Throw away entries which we do not need. */
3049 if (syms[i].st_shndx == SHN_UNDEF
3050 || syms[i].st_shndx >= SHN_LORESERVE
3051 || ELF_ST_TYPE(syms[i].st_info) != STT_FUNC) {
3052 if (i < --nsyms) {
3053 syms[i] = syms[nsyms];
3054 }
3055 } else {
3056 #if defined(TARGET_ARM) || defined (TARGET_MIPS)
3057 /* The bottom address bit marks a Thumb or MIPS16 symbol. */
3058 syms[i].st_value &= ~(target_ulong)1;
3059 #endif
3060 syms[i].st_value += load_bias;
3061 i++;
3062 }
3063 }
3064
3065 /* No "useful" symbol. */
3066 if (nsyms == 0) {
3067 goto give_up;
3068 }
3069
3070 /* Attempt to free the storage associated with the local symbols
3071 that we threw away. Whether or not this has any effect on the
3072 memory allocation depends on the malloc implementation and how
3073 many symbols we managed to discard. */
3074 new_syms = g_try_renew(struct elf_sym, syms, nsyms);
3075 if (new_syms == NULL) {
3076 goto give_up;
3077 }
3078 syms = new_syms;
3079
3080 qsort(syms, nsyms, sizeof(*syms), symcmp);
3081
3082 s->disas_num_syms = nsyms;
3083 #if ELF_CLASS == ELFCLASS32
3084 s->disas_symtab.elf32 = syms;
3085 #else
3086 s->disas_symtab.elf64 = syms;
3087 #endif
3088 s->lookup_symbol = lookup_symbolxx;
3089 s->next = syminfos;
3090 syminfos = s;
3091
3092 return;
3093
3094 give_up:
3095 g_free(s);
3096 g_free(strings);
3097 g_free(syms);
3098 }
3099
3100 uint32_t get_elf_eflags(int fd)
3101 {
3102 struct elfhdr ehdr;
3103 off_t offset;
3104 int ret;
3105
3106 /* Read ELF header */
3107 offset = lseek(fd, 0, SEEK_SET);
3108 if (offset == (off_t) -1) {
3109 return 0;
3110 }
3111 ret = read(fd, &ehdr, sizeof(ehdr));
3112 if (ret < sizeof(ehdr)) {
3113 return 0;
3114 }
3115 offset = lseek(fd, offset, SEEK_SET);
3116 if (offset == (off_t) -1) {
3117 return 0;
3118 }
3119
3120 /* Check ELF signature */
3121 if (!elf_check_ident(&ehdr)) {
3122 return 0;
3123 }
3124
3125 /* check header */
3126 bswap_ehdr(&ehdr);
3127 if (!elf_check_ehdr(&ehdr)) {
3128 return 0;
3129 }
3130
3131 /* return architecture id */
3132 return ehdr.e_flags;
3133 }
3134
3135 int load_elf_binary(struct linux_binprm *bprm, struct image_info *info)
3136 {
3137 struct image_info interp_info;
3138 struct elfhdr elf_ex;
3139 char *elf_interpreter = NULL;
3140 char *scratch;
3141
3142 memset(&interp_info, 0, sizeof(interp_info));
3143 #ifdef TARGET_MIPS
3144 interp_info.fp_abi = MIPS_ABI_FP_UNKNOWN;
3145 #endif
3146
3147 info->start_mmap = (abi_ulong)ELF_START_MMAP;
3148
3149 load_elf_image(bprm->filename, bprm->fd, info,
3150 &elf_interpreter, bprm->buf);
3151
3152 /* ??? We need a copy of the elf header for passing to create_elf_tables.
3153 If we do nothing, we'll have overwritten this when we re-use bprm->buf
3154 when we load the interpreter. */
3155 elf_ex = *(struct elfhdr *)bprm->buf;
3156
3157 /* Do this so that we can load the interpreter, if need be. We will
3158 change some of these later */
3159 bprm->p = setup_arg_pages(bprm, info);
3160
3161 scratch = g_new0(char, TARGET_PAGE_SIZE);
3162 if (STACK_GROWS_DOWN) {
3163 bprm->p = copy_elf_strings(1, &bprm->filename, scratch,
3164 bprm->p, info->stack_limit);
3165 info->file_string = bprm->p;
3166 bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch,
3167 bprm->p, info->stack_limit);
3168 info->env_strings = bprm->p;
3169 bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch,
3170 bprm->p, info->stack_limit);
3171 info->arg_strings = bprm->p;
3172 } else {
3173 info->arg_strings = bprm->p;
3174 bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch,
3175 bprm->p, info->stack_limit);
3176 info->env_strings = bprm->p;
3177 bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch,
3178 bprm->p, info->stack_limit);
3179 info->file_string = bprm->p;
3180 bprm->p = copy_elf_strings(1, &bprm->filename, scratch,
3181 bprm->p, info->stack_limit);
3182 }
3183
3184 g_free(scratch);
3185
3186 if (!bprm->p) {
3187 fprintf(stderr, "%s: %s\n", bprm->filename, strerror(E2BIG));
3188 exit(-1);
3189 }
3190
3191 if (elf_interpreter) {
3192 load_elf_interp(elf_interpreter, &interp_info, bprm->buf);
3193
3194 /* If the program interpreter is one of these two, then assume
3195 an iBCS2 image. Otherwise assume a native linux image. */
3196
3197 if (strcmp(elf_interpreter, "/usr/lib/libc.so.1") == 0
3198 || strcmp(elf_interpreter, "/usr/lib/ld.so.1") == 0) {
3199 info->personality = PER_SVR4;
3200
3201 /* Why this, you ask??? Well SVr4 maps page 0 as read-only,
3202 and some applications "depend" upon this behavior. Since
3203 we do not have the power to recompile these, we emulate
3204 the SVr4 behavior. Sigh. */
3205 target_mmap(0, qemu_host_page_size, PROT_READ | PROT_EXEC,
3206 MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
3207 }
3208 #ifdef TARGET_MIPS
3209 info->interp_fp_abi = interp_info.fp_abi;
3210 #endif
3211 }
3212
3213 bprm->p = create_elf_tables(bprm->p, bprm->argc, bprm->envc, &elf_ex,
3214 info, (elf_interpreter ? &interp_info : NULL));
3215 info->start_stack = bprm->p;
3216
3217 /* If we have an interpreter, set that as the program's entry point.
3218 Copy the load_bias as well, to help PPC64 interpret the entry
3219 point as a function descriptor. Do this after creating elf tables
3220 so that we copy the original program entry point into the AUXV. */
3221 if (elf_interpreter) {
3222 info->load_bias = interp_info.load_bias;
3223 info->entry = interp_info.entry;
3224 g_free(elf_interpreter);
3225 }
3226
3227 #ifdef USE_ELF_CORE_DUMP
3228 bprm->core_dump = &elf_core_dump;
3229 #endif
3230
3231 /*
3232 * If we reserved extra space for brk, release it now.
3233 * The implementation of do_brk in syscalls.c expects to be able
3234 * to mmap pages in this space.
3235 */
3236 if (info->reserve_brk) {
3237 abi_ulong start_brk = HOST_PAGE_ALIGN(info->brk);
3238 abi_ulong end_brk = HOST_PAGE_ALIGN(info->brk + info->reserve_brk);
3239 target_munmap(start_brk, end_brk - start_brk);
3240 }
3241
3242 return 0;
3243 }
3244
3245 #ifdef USE_ELF_CORE_DUMP
3246 /*
3247 * Definitions to generate Intel SVR4-like core files.
3248 * These mostly have the same names as the SVR4 types with "target_elf_"
3249 * tacked on the front to prevent clashes with linux definitions,
3250 * and the typedef forms have been avoided. This is mostly like
3251 * the SVR4 structure, but more Linuxy, with things that Linux does
3252 * not support and which gdb doesn't really use excluded.
3253 *
3254 * Fields we don't dump (their contents is zero) in linux-user qemu
3255 * are marked with XXX.
3256 *
3257 * Core dump code is copied from linux kernel (fs/binfmt_elf.c).
3258 *
3259 * Porting ELF coredump for target is (quite) simple process. First you
3260 * define USE_ELF_CORE_DUMP in target ELF code (where init_thread() for
3261 * the target resides):
3262 *
3263 * #define USE_ELF_CORE_DUMP
3264 *
3265 * Next you define type of register set used for dumping. ELF specification
3266 * says that it needs to be array of elf_greg_t that has size of ELF_NREG.
3267 *
3268 * typedef <target_regtype> target_elf_greg_t;
3269 * #define ELF_NREG <number of registers>
3270 * typedef taret_elf_greg_t target_elf_gregset_t[ELF_NREG];
3271 *
3272 * Last step is to implement target specific function that copies registers
3273 * from given cpu into just specified register set. Prototype is:
3274 *
3275 * static void elf_core_copy_regs(taret_elf_gregset_t *regs,
3276 * const CPUArchState *env);
3277 *
3278 * Parameters:
3279 * regs - copy register values into here (allocated and zeroed by caller)
3280 * env - copy registers from here
3281 *
3282 * Example for ARM target is provided in this file.
3283 */
3284
3285 /* An ELF note in memory */
3286 struct memelfnote {
3287 const char *name;
3288 size_t namesz;
3289 size_t namesz_rounded;
3290 int type;
3291 size_t datasz;
3292 size_t datasz_rounded;
3293 void *data;
3294 size_t notesz;
3295 };
3296
3297 struct target_elf_siginfo {
3298 abi_int si_signo; /* signal number */
3299 abi_int si_code; /* extra code */
3300 abi_int si_errno; /* errno */
3301 };
3302
3303 struct target_elf_prstatus {
3304 struct target_elf_siginfo pr_info; /* Info associated with signal */
3305 abi_short pr_cursig; /* Current signal */
3306 abi_ulong pr_sigpend; /* XXX */
3307 abi_ulong pr_sighold; /* XXX */
3308 target_pid_t pr_pid;
3309 target_pid_t pr_ppid;
3310 target_pid_t pr_pgrp;
3311 target_pid_t pr_sid;
3312 struct target_timeval pr_utime; /* XXX User time */
3313 struct target_timeval pr_stime; /* XXX System time */
3314 struct target_timeval pr_cutime; /* XXX Cumulative user time */
3315 struct target_timeval pr_cstime; /* XXX Cumulative system time */
3316 target_elf_gregset_t pr_reg; /* GP registers */
3317 abi_int pr_fpvalid; /* XXX */
3318 };
3319
3320 #define ELF_PRARGSZ (80) /* Number of chars for args */
3321
3322 struct target_elf_prpsinfo {
3323 char pr_state; /* numeric process state */
3324 char pr_sname; /* char for pr_state */
3325 char pr_zomb; /* zombie */
3326 char pr_nice; /* nice val */
3327 abi_ulong pr_flag; /* flags */
3328 target_uid_t pr_uid;
3329 target_gid_t pr_gid;
3330 target_pid_t pr_pid, pr_ppid, pr_pgrp, pr_sid;
3331 /* Lots missing */
3332 char pr_fname[16] QEMU_NONSTRING; /* filename of executable */
3333 char pr_psargs[ELF_PRARGSZ]; /* initial part of arg list */
3334 };
3335
3336 /* Here is the structure in which status of each thread is captured. */
3337 struct elf_thread_status {
3338 QTAILQ_ENTRY(elf_thread_status) ets_link;
3339 struct target_elf_prstatus prstatus; /* NT_PRSTATUS */
3340 #if 0
3341 elf_fpregset_t fpu; /* NT_PRFPREG */
3342 struct task_struct *thread;
3343 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */
3344 #endif
3345 struct memelfnote notes[1];
3346 int num_notes;
3347 };
3348
3349 struct elf_note_info {
3350 struct memelfnote *notes;
3351 struct target_elf_prstatus *prstatus; /* NT_PRSTATUS */
3352 struct target_elf_prpsinfo *psinfo; /* NT_PRPSINFO */
3353
3354 QTAILQ_HEAD(, elf_thread_status) thread_list;
3355 #if 0
3356 /*
3357 * Current version of ELF coredump doesn't support
3358 * dumping fp regs etc.
3359 */
3360 elf_fpregset_t *fpu;
3361 elf_fpxregset_t *xfpu;
3362 int thread_status_size;
3363 #endif
3364 int notes_size;
3365 int numnote;
3366 };
3367
3368 struct vm_area_struct {
3369 target_ulong vma_start; /* start vaddr of memory region */
3370 target_ulong vma_end; /* end vaddr of memory region */
3371 abi_ulong vma_flags; /* protection etc. flags for the region */
3372 QTAILQ_ENTRY(vm_area_struct) vma_link;
3373 };
3374
3375 struct mm_struct {
3376 QTAILQ_HEAD(, vm_area_struct) mm_mmap;
3377 int mm_count; /* number of mappings */
3378 };
3379
3380 static struct mm_struct *vma_init(void);
3381 static void vma_delete(struct mm_struct *);
3382 static int vma_add_mapping(struct mm_struct *, target_ulong,
3383 target_ulong, abi_ulong);
3384 static int vma_get_mapping_count(const struct mm_struct *);
3385 static struct vm_area_struct *vma_first(const struct mm_struct *);
3386 static struct vm_area_struct *vma_next(struct vm_area_struct *);
3387 static abi_ulong vma_dump_size(const struct vm_area_struct *);
3388 static int vma_walker(void *priv, target_ulong start, target_ulong end,
3389 unsigned long flags);
3390
3391 static void fill_elf_header(struct elfhdr *, int, uint16_t, uint32_t);
3392 static void fill_note(struct memelfnote *, const char *, int,
3393 unsigned int, void *);
3394 static void fill_prstatus(struct target_elf_prstatus *, const TaskState *, int);
3395 static int fill_psinfo(struct target_elf_prpsinfo *, const TaskState *);
3396 static void fill_auxv_note(struct memelfnote *, const TaskState *);
3397 static void fill_elf_note_phdr(struct elf_phdr *, int, off_t);
3398 static size_t note_size(const struct memelfnote *);
3399 static void free_note_info(struct elf_note_info *);
3400 static int fill_note_info(struct elf_note_info *, long, const CPUArchState *);
3401 static void fill_thread_info(struct elf_note_info *, const CPUArchState *);
3402 static int core_dump_filename(const TaskState *, char *, size_t);
3403
3404 static int dump_write(int, const void *, size_t);
3405 static int write_note(struct memelfnote *, int);
3406 static int write_note_info(struct elf_note_info *, int);
3407
3408 #ifdef BSWAP_NEEDED
3409 static void bswap_prstatus(struct target_elf_prstatus *prstatus)
3410 {
3411 prstatus->pr_info.si_signo = tswap32(prstatus->pr_info.si_signo);
3412 prstatus->pr_info.si_code = tswap32(prstatus->pr_info.si_code);
3413 prstatus->pr_info.si_errno = tswap32(prstatus->pr_info.si_errno);
3414 prstatus->pr_cursig = tswap16(prstatus->pr_cursig);
3415 prstatus->pr_sigpend = tswapal(prstatus->pr_sigpend);
3416 prstatus->pr_sighold = tswapal(prstatus->pr_sighold);
3417 prstatus->pr_pid = tswap32(prstatus->pr_pid);
3418 prstatus->pr_ppid = tswap32(prstatus->pr_ppid);
3419 prstatus->pr_pgrp = tswap32(prstatus->pr_pgrp);
3420 prstatus->pr_sid = tswap32(prstatus->pr_sid);
3421 /* cpu times are not filled, so we skip them */
3422 /* regs should be in correct format already */
3423 prstatus->pr_fpvalid = tswap32(prstatus->pr_fpvalid);
3424 }
3425
3426 static void bswap_psinfo(struct target_elf_prpsinfo *psinfo)
3427 {
3428 psinfo->pr_flag = tswapal(psinfo->pr_flag);
3429 psinfo->pr_uid = tswap16(psinfo->pr_uid);
3430 psinfo->pr_gid = tswap16(psinfo->pr_gid);
3431 psinfo->pr_pid = tswap32(psinfo->pr_pid);
3432 psinfo->pr_ppid = tswap32(psinfo->pr_ppid);
3433 psinfo->pr_pgrp = tswap32(psinfo->pr_pgrp);
3434 psinfo->pr_sid = tswap32(psinfo->pr_sid);
3435 }
3436
3437 static void bswap_note(struct elf_note *en)
3438 {
3439 bswap32s(&en->n_namesz);
3440 bswap32s(&en->n_descsz);
3441 bswap32s(&en->n_type);
3442 }
3443 #else
3444 static inline void bswap_prstatus(struct target_elf_prstatus *p) { }
3445 static inline void bswap_psinfo(struct target_elf_prpsinfo *p) {}
3446 static inline void bswap_note(struct elf_note *en) { }
3447 #endif /* BSWAP_NEEDED */
3448
3449 /*
3450 * Minimal support for linux memory regions. These are needed
3451 * when we are finding out what memory exactly belongs to
3452 * emulated process. No locks needed here, as long as
3453 * thread that received the signal is stopped.
3454 */
3455
3456 static struct mm_struct *vma_init(void)
3457 {
3458 struct mm_struct *mm;
3459
3460 if ((mm = g_malloc(sizeof (*mm))) == NULL)
3461 return (NULL);
3462
3463 mm->mm_count = 0;
3464 QTAILQ_INIT(&mm->mm_mmap);
3465
3466 return (mm);
3467 }
3468
3469 static void vma_delete(struct mm_struct *mm)
3470 {
3471 struct vm_area_struct *vma;
3472
3473 while ((vma = vma_first(mm)) != NULL) {
3474 QTAILQ_REMOVE(&mm->mm_mmap, vma, vma_link);
3475 g_free(vma);
3476 }
3477 g_free(mm);
3478 }
3479
3480 static int vma_add_mapping(struct mm_struct *mm, target_ulong start,
3481 target_ulong end, abi_ulong flags)
3482 {
3483 struct vm_area_struct *vma;
3484
3485 if ((vma = g_malloc0(sizeof (*vma))) == NULL)
3486 return (-1);
3487
3488 vma->vma_start = start;
3489 vma->vma_end = end;
3490 vma->vma_flags = flags;
3491
3492 QTAILQ_INSERT_TAIL(&mm->mm_mmap, vma, vma_link);
3493 mm->mm_count++;
3494
3495 return (0);
3496 }
3497
3498 static struct vm_area_struct *vma_first(const struct mm_struct *mm)
3499 {
3500 return (QTAILQ_FIRST(&mm->mm_mmap));
3501 }
3502
3503 static struct vm_area_struct *vma_next(struct vm_area_struct *vma)
3504 {
3505 return (QTAILQ_NEXT(vma, vma_link));
3506 }
3507
3508 static int vma_get_mapping_count(const struct mm_struct *mm)
3509 {
3510 return (mm->mm_count);
3511 }
3512
3513 /*
3514 * Calculate file (dump) size of given memory region.
3515 */
3516 static abi_ulong vma_dump_size(const struct vm_area_struct *vma)
3517 {
3518 /* if we cannot even read the first page, skip it */
3519 if (!access_ok_untagged(VERIFY_READ, vma->vma_start, TARGET_PAGE_SIZE))
3520 return (0);
3521
3522 /*
3523 * Usually we don't dump executable pages as they contain
3524 * non-writable code that debugger can read directly from
3525 * target library etc. However, thread stacks are marked
3526 * also executable so we read in first page of given region
3527 * and check whether it contains elf header. If there is
3528 * no elf header, we dump it.
3529 */
3530 if (vma->vma_flags & PROT_EXEC) {
3531 char page[TARGET_PAGE_SIZE];
3532
3533 if (copy_from_user(page, vma->vma_start, sizeof (page))) {
3534 return 0;
3535 }
3536 if ((page[EI_MAG0] == ELFMAG0) &&
3537 (page[EI_MAG1] == ELFMAG1) &&
3538 (page[EI_MAG2] == ELFMAG2) &&
3539 (page[EI_MAG3] == ELFMAG3)) {
3540 /*
3541 * Mappings are possibly from ELF binary. Don't dump
3542 * them.
3543 */
3544 return (0);
3545 }
3546 }
3547
3548 return (vma->vma_end - vma->vma_start);
3549 }
3550
3551 static int vma_walker(void *priv, target_ulong start, target_ulong end,
3552 unsigned long flags)
3553 {
3554 struct mm_struct *mm = (struct mm_struct *)priv;
3555
3556 vma_add_mapping(mm, start, end, flags);
3557 return (0);
3558 }
3559
3560 static void fill_note(struct memelfnote *note, const char *name, int type,
3561 unsigned int sz, void *data)
3562 {
3563 unsigned int namesz;
3564
3565 namesz = strlen(name) + 1;
3566 note->name = name;
3567 note->namesz = namesz;
3568 note->namesz_rounded = roundup(namesz, sizeof (int32_t));
3569 note->type = type;
3570 note->datasz = sz;
3571 note->datasz_rounded = roundup(sz, sizeof (int32_t));
3572
3573 note->data = data;
3574
3575 /*
3576 * We calculate rounded up note size here as specified by
3577 * ELF document.
3578 */
3579 note->notesz = sizeof (struct elf_note) +
3580 note->namesz_rounded + note->datasz_rounded;
3581 }
3582
3583 static void fill_elf_header(struct elfhdr *elf, int segs, uint16_t machine,
3584 uint32_t flags)
3585 {
3586 (void) memset(elf, 0, sizeof(*elf));
3587
3588 (void) memcpy(elf->e_ident, ELFMAG, SELFMAG);
3589 elf->e_ident[EI_CLASS] = ELF_CLASS;
3590 elf->e_ident[EI_DATA] = ELF_DATA;
3591 elf->e_ident[EI_VERSION] = EV_CURRENT;
3592 elf->e_ident[EI_OSABI] = ELF_OSABI;
3593
3594 elf->e_type = ET_CORE;
3595 elf->e_machine = machine;
3596 elf->e_version = EV_CURRENT;
3597 elf->e_phoff = sizeof(struct elfhdr);
3598 elf->e_flags = flags;
3599 elf->e_ehsize = sizeof(struct elfhdr);
3600 elf->e_phentsize = sizeof(struct elf_phdr);
3601 elf->e_phnum = segs;
3602
3603 bswap_ehdr(elf);
3604 }
3605
3606 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, off_t offset)
3607 {
3608 phdr->p_type = PT_NOTE;
3609 phdr->p_offset = offset;
3610 phdr->p_vaddr = 0;
3611 phdr->p_paddr = 0;
3612 phdr->p_filesz = sz;
3613 phdr->p_memsz = 0;
3614 phdr->p_flags = 0;
3615 phdr->p_align = 0;
3616
3617 bswap_phdr(phdr, 1);
3618 }
3619
3620 static size_t note_size(const struct memelfnote *note)
3621 {
3622 return (note->notesz);
3623 }
3624
3625 static void fill_prstatus(struct target_elf_prstatus *prstatus,
3626 const TaskState *ts, int signr)
3627 {
3628 (void) memset(prstatus, 0, sizeof (*prstatus));
3629 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
3630 prstatus->pr_pid = ts->ts_tid;
3631 prstatus->pr_ppid = getppid();
3632 prstatus->pr_pgrp = getpgrp();
3633 prstatus->pr_sid = getsid(0);
3634
3635 bswap_prstatus(prstatus);
3636 }
3637
3638 static int fill_psinfo(struct target_elf_prpsinfo *psinfo, const TaskState *ts)
3639 {
3640 char *base_filename;
3641 unsigned int i, len;
3642
3643 (void) memset(psinfo, 0, sizeof (*psinfo));
3644
3645 len = ts->info->arg_end - ts->info->arg_start;
3646 if (len >= ELF_PRARGSZ)
3647 len = ELF_PRARGSZ - 1;
3648 if (copy_from_user(&psinfo->pr_psargs, ts->info->arg_start, len))
3649 return -EFAULT;
3650 for (i = 0; i < len; i++)
3651 if (psinfo->pr_psargs[i] == 0)
3652 psinfo->pr_psargs[i] = ' ';
3653 psinfo->pr_psargs[len] = 0;
3654
3655 psinfo->pr_pid = getpid();
3656 psinfo->pr_ppid = getppid();
3657 psinfo->pr_pgrp = getpgrp();
3658 psinfo->pr_sid = getsid(0);
3659 psinfo->pr_uid = getuid();
3660 psinfo->pr_gid = getgid();
3661
3662 base_filename = g_path_get_basename(ts->bprm->filename);
3663 /*
3664 * Using strncpy here is fine: at max-length,
3665 * this field is not NUL-terminated.
3666 */
3667 (void) strncpy(psinfo->pr_fname, base_filename,
3668 sizeof(psinfo->pr_fname));
3669
3670 g_free(base_filename);
3671 bswap_psinfo(psinfo);
3672 return (0);
3673 }
3674
3675 static void fill_auxv_note(struct memelfnote *note, const TaskState *ts)
3676 {
3677 elf_addr_t auxv = (elf_addr_t)ts->info->saved_auxv;
3678 elf_addr_t orig_auxv = auxv;
3679 void *ptr;
3680 int len = ts->info->auxv_len;
3681
3682 /*
3683 * Auxiliary vector is stored in target process stack. It contains
3684 * {type, value} pairs that we need to dump into note. This is not
3685 * strictly necessary but we do it here for sake of completeness.
3686 */
3687
3688 /* read in whole auxv vector and copy it to memelfnote */
3689 ptr = lock_user(VERIFY_READ, orig_auxv, len, 0);
3690 if (ptr != NULL) {
3691 fill_note(note, "CORE", NT_AUXV, len, ptr);
3692 unlock_user(ptr, auxv, len);
3693 }
3694 }
3695
3696 /*
3697 * Constructs name of coredump file. We have following convention
3698 * for the name:
3699 * qemu_<basename-of-target-binary>_<date>-<time>_<pid>.core
3700 *
3701 * Returns 0 in case of success, -1 otherwise (errno is set).
3702 */
3703 static int core_dump_filename(const TaskState *ts, char *buf,
3704 size_t bufsize)
3705 {
3706 char timestamp[64];
3707 char *base_filename = NULL;
3708 struct timeval tv;
3709 struct tm tm;
3710
3711 assert(bufsize >= PATH_MAX);
3712
3713 if (gettimeofday(&tv, NULL) < 0) {
3714 (void) fprintf(stderr, "unable to get current timestamp: %s",
3715 strerror(errno));
3716 return (-1);
3717 }
3718
3719 base_filename = g_path_get_basename(ts->bprm->filename);
3720 (void) strftime(timestamp, sizeof (timestamp), "%Y%m%d-%H%M%S",
3721 localtime_r(&tv.tv_sec, &tm));
3722 (void) snprintf(buf, bufsize, "qemu_%s_%s_%d.core",
3723 base_filename, timestamp, (int)getpid());
3724 g_free(base_filename);
3725
3726 return (0);
3727 }
3728
3729 static int dump_write(int fd, const void *ptr, size_t size)
3730 {
3731 const char *bufp = (const char *)ptr;
3732 ssize_t bytes_written, bytes_left;
3733 struct rlimit dumpsize;
3734 off_t pos;
3735
3736 bytes_written = 0;
3737 getrlimit(RLIMIT_CORE, &dumpsize);
3738 if ((pos = lseek(fd, 0, SEEK_CUR))==-1) {
3739 if (errno == ESPIPE) { /* not a seekable stream */
3740 bytes_left = size;
3741 } else {
3742 return pos;
3743 }
3744 } else {
3745 if (dumpsize.rlim_cur <= pos) {
3746 return -1;
3747 } else if (dumpsize.rlim_cur == RLIM_INFINITY) {
3748 bytes_left = size;
3749 } else {
3750 size_t limit_left=dumpsize.rlim_cur - pos;
3751 bytes_left = limit_left >= size ? size : limit_left ;
3752 }
3753 }
3754
3755 /*
3756 * In normal conditions, single write(2) should do but
3757 * in case of socket etc. this mechanism is more portable.
3758 */
3759 do {
3760 bytes_written = write(fd, bufp, bytes_left);
3761 if (bytes_written < 0) {
3762 if (errno == EINTR)
3763 continue;
3764 return (-1);
3765 } else if (bytes_written == 0) { /* eof */
3766 return (-1);
3767 }
3768 bufp += bytes_written;
3769 bytes_left -= bytes_written;
3770 } while (bytes_left > 0);
3771
3772 return (0);
3773 }
3774
3775 static int write_note(struct memelfnote *men, int fd)
3776 {
3777 struct elf_note en;
3778
3779 en.n_namesz = men->namesz;
3780 en.n_type = men->type;
3781 en.n_descsz = men->datasz;
3782
3783 bswap_note(&en);
3784
3785 if (dump_write(fd, &en, sizeof(en)) != 0)
3786 return (-1);
3787 if (dump_write(fd, men->name, men->namesz_rounded) != 0)
3788 return (-1);
3789 if (dump_write(fd, men->data, men->datasz_rounded) != 0)
3790 return (-1);
3791
3792 return (0);
3793 }
3794
3795 static void fill_thread_info(struct elf_note_info *info, const CPUArchState *env)
3796 {
3797 CPUState *cpu = env_cpu((CPUArchState *)env);
3798 TaskState *ts = (TaskState *)cpu->opaque;
3799 struct elf_thread_status *ets;
3800
3801 ets = g_malloc0(sizeof (*ets));
3802 ets->num_notes = 1; /* only prstatus is dumped */
3803 fill_prstatus(&ets->prstatus, ts, 0);
3804 elf_core_copy_regs(&ets->prstatus.pr_reg, env);
3805 fill_note(&ets->notes[0], "CORE", NT_PRSTATUS, sizeof (ets->prstatus),
3806 &ets->prstatus);
3807
3808 QTAILQ_INSERT_TAIL(&info->thread_list, ets, ets_link);
3809
3810 info->notes_size += note_size(&ets->notes[0]);
3811 }
3812
3813 static void init_note_info(struct elf_note_info *info)
3814 {
3815 /* Initialize the elf_note_info structure so that it is at
3816 * least safe to call free_note_info() on it. Must be
3817 * called before calling fill_note_info().
3818 */
3819 memset(info, 0, sizeof (*info));
3820 QTAILQ_INIT(&info->thread_list);
3821 }
3822
3823 static int fill_note_info(struct elf_note_info *info,
3824 long signr, const CPUArchState *env)
3825 {
3826 #define NUMNOTES 3
3827 CPUState *cpu = env_cpu((CPUArchState *)env);
3828 TaskState *ts = (TaskState *)cpu->opaque;
3829 int i;
3830
3831 info->notes = g_new0(struct memelfnote, NUMNOTES);
3832 if (info->notes == NULL)
3833 return (-ENOMEM);
3834 info->prstatus = g_malloc0(sizeof (*info->prstatus));
3835 if (info->prstatus == NULL)
3836 return (-ENOMEM);
3837 info->psinfo = g_malloc0(sizeof (*info->psinfo));
3838 if (info->prstatus == NULL)
3839 return (-ENOMEM);
3840
3841 /*
3842 * First fill in status (and registers) of current thread
3843 * including process info & aux vector.
3844 */
3845 fill_prstatus(info->prstatus, ts, signr);
3846 elf_core_copy_regs(&info->prstatus->pr_reg, env);
3847 fill_note(&info->notes[0], "CORE", NT_PRSTATUS,
3848 sizeof (*info->prstatus), info->prstatus);
3849 fill_psinfo(info->psinfo, ts);
3850 fill_note(&info->notes[1], "CORE", NT_PRPSINFO,
3851 sizeof (*info->psinfo), info->psinfo);
3852 fill_auxv_note(&info->notes[2], ts);
3853 info->numnote = 3;
3854
3855 info->notes_size = 0;
3856 for (i = 0; i < info->numnote; i++)
3857 info->notes_size += note_size(&info->notes[i]);
3858
3859 /* read and fill status of all threads */
3860 cpu_list_lock();
3861 CPU_FOREACH(cpu) {
3862 if (cpu == thread_cpu) {
3863 continue;
3864 }
3865 fill_thread_info(info, (CPUArchState *)cpu->env_ptr);
3866 }
3867 cpu_list_unlock();
3868
3869 return (0);
3870 }
3871
3872 static void free_note_info(struct elf_note_info *info)
3873 {
3874 struct elf_thread_status *ets;
3875
3876 while (!QTAILQ_EMPTY(&info->thread_list)) {
3877 ets = QTAILQ_FIRST(&info->thread_list);
3878 QTAILQ_REMOVE(&info->thread_list, ets, ets_link);
3879 g_free(ets);
3880 }
3881
3882 g_free(info->prstatus);
3883 g_free(info->psinfo);
3884 g_free(info->notes);
3885 }
3886
3887 static int write_note_info(struct elf_note_info *info, int fd)
3888 {
3889 struct elf_thread_status *ets;
3890 int i, error = 0;
3891
3892 /* write prstatus, psinfo and auxv for current thread */
3893 for (i = 0; i < info->numnote; i++)
3894 if ((error = write_note(&info->notes[i], fd)) != 0)
3895 return (error);
3896
3897 /* write prstatus for each thread */
3898 QTAILQ_FOREACH(ets, &info->thread_list, ets_link) {
3899 if ((error = write_note(&ets->notes[0], fd)) != 0)
3900 return (error);
3901 }
3902
3903 return (0);
3904 }
3905
3906 /*
3907 * Write out ELF coredump.
3908 *
3909 * See documentation of ELF object file format in:
3910 * http://www.caldera.com/developers/devspecs/gabi41.pdf
3911 *
3912 * Coredump format in linux is following:
3913 *
3914 * 0 +----------------------+ \
3915 * | ELF header | ET_CORE |
3916 * +----------------------+ |
3917 * | ELF program headers | |--- headers
3918 * | - NOTE section | |
3919 * | - PT_LOAD sections | |
3920 * +----------------------+ /
3921 * | NOTEs: |
3922 * | - NT_PRSTATUS |
3923 * | - NT_PRSINFO |
3924 * | - NT_AUXV |
3925 * +----------------------+ <-- aligned to target page
3926 * | Process memory dump |
3927 * : :
3928 * . .
3929 * : :
3930 * | |
3931 * +----------------------+
3932 *
3933 * NT_PRSTATUS -> struct elf_prstatus (per thread)
3934 * NT_PRSINFO -> struct elf_prpsinfo
3935 * NT_AUXV is array of { type, value } pairs (see fill_auxv_note()).
3936 *
3937 * Format follows System V format as close as possible. Current
3938 * version limitations are as follows:
3939 * - no floating point registers are dumped
3940 *
3941 * Function returns 0 in case of success, negative errno otherwise.
3942 *
3943 * TODO: make this work also during runtime: it should be
3944 * possible to force coredump from running process and then
3945 * continue processing. For example qemu could set up SIGUSR2
3946 * handler (provided that target process haven't registered
3947 * handler for that) that does the dump when signal is received.
3948 */
3949 static int elf_core_dump(int signr, const CPUArchState *env)
3950 {
3951 const CPUState *cpu = env_cpu((CPUArchState *)env);
3952 const TaskState *ts = (const TaskState *)cpu->opaque;
3953 struct vm_area_struct *vma = NULL;
3954 char corefile[PATH_MAX];
3955 struct elf_note_info info;
3956 struct elfhdr elf;
3957 struct elf_phdr phdr;
3958 struct rlimit dumpsize;
3959 struct mm_struct *mm = NULL;
3960 off_t offset = 0, data_offset = 0;
3961 int segs = 0;
3962 int fd = -1;
3963
3964 init_note_info(&info);
3965
3966 errno = 0;
3967 getrlimit(RLIMIT_CORE, &dumpsize);
3968 if (dumpsize.rlim_cur == 0)
3969 return 0;
3970
3971 if (core_dump_filename(ts, corefile, sizeof (corefile)) < 0)
3972 return (-errno);
3973
3974 if ((fd = open(corefile, O_WRONLY | O_CREAT,
3975 S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH)) < 0)
3976 return (-errno);
3977
3978 /*
3979 * Walk through target process memory mappings and
3980 * set up structure containing this information. After
3981 * this point vma_xxx functions can be used.
3982 */
3983 if ((mm = vma_init()) == NULL)
3984 goto out;
3985
3986 walk_memory_regions(mm, vma_walker);
3987 segs = vma_get_mapping_count(mm);
3988
3989 /*
3990 * Construct valid coredump ELF header. We also
3991 * add one more segment for notes.
3992 */
3993 fill_elf_header(&elf, segs + 1, ELF_MACHINE, 0);
3994 if (dump_write(fd, &elf, sizeof (elf)) != 0)
3995 goto out;
3996
3997 /* fill in the in-memory version of notes */
3998 if (fill_note_info(&info, signr, env) < 0)
3999 goto out;
4000
4001 offset += sizeof (elf); /* elf header */
4002 offset += (segs + 1) * sizeof (struct elf_phdr); /* program headers */
4003
4004 /* write out notes program header */
4005 fill_elf_note_phdr(&phdr, info.notes_size, offset);
4006
4007 offset += info.notes_size;
4008 if (dump_write(fd, &phdr, sizeof (phdr)) != 0)
4009 goto out;
4010
4011 /*
4012 * ELF specification wants data to start at page boundary so
4013 * we align it here.
4014 */
4015 data_offset = offset = roundup(offset, ELF_EXEC_PAGESIZE);
4016
4017 /*
4018 * Write program headers for memory regions mapped in
4019 * the target process.
4020 */
4021 for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) {
4022 (void) memset(&phdr, 0, sizeof (phdr));
4023
4024 phdr.p_type = PT_LOAD;
4025 phdr.p_offset = offset;
4026 phdr.p_vaddr = vma->vma_start;
4027 phdr.p_paddr = 0;
4028 phdr.p_filesz = vma_dump_size(vma);
4029 offset += phdr.p_filesz;
4030 phdr.p_memsz = vma->vma_end - vma->vma_start;
4031 phdr.p_flags = vma->vma_flags & PROT_READ ? PF_R : 0;
4032 if (vma->vma_flags & PROT_WRITE)
4033 phdr.p_flags |= PF_W;
4034 if (vma->vma_flags & PROT_EXEC)
4035 phdr.p_flags |= PF_X;
4036 phdr.p_align = ELF_EXEC_PAGESIZE;
4037
4038 bswap_phdr(&phdr, 1);
4039 if (dump_write(fd, &phdr, sizeof(phdr)) != 0) {
4040 goto out;
4041 }
4042 }
4043
4044 /*
4045 * Next we write notes just after program headers. No
4046 * alignment needed here.
4047 */
4048 if (write_note_info(&info, fd) < 0)
4049 goto out;
4050
4051 /* align data to page boundary */
4052 if (lseek(fd, data_offset, SEEK_SET) != data_offset)
4053 goto out;
4054
4055 /*
4056 * Finally we can dump process memory into corefile as well.
4057 */
4058 for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) {
4059 abi_ulong addr;
4060 abi_ulong end;
4061
4062 end = vma->vma_start + vma_dump_size(vma);
4063
4064 for (addr = vma->vma_start; addr < end;
4065 addr += TARGET_PAGE_SIZE) {
4066 char page[TARGET_PAGE_SIZE];
4067 int error;
4068
4069 /*
4070 * Read in page from target process memory and
4071 * write it to coredump file.
4072 */
4073 error = copy_from_user(page, addr, sizeof (page));
4074 if (error != 0) {
4075 (void) fprintf(stderr, "unable to dump " TARGET_ABI_FMT_lx "\n",
4076 addr);
4077 errno = -error;
4078 goto out;
4079 }
4080 if (dump_write(fd, page, TARGET_PAGE_SIZE) < 0)
4081 goto out;
4082 }
4083 }
4084
4085 out:
4086 free_note_info(&info);
4087 if (mm != NULL)
4088 vma_delete(mm);
4089 (void) close(fd);
4090
4091 if (errno != 0)
4092 return (-errno);
4093 return (0);
4094 }
4095 #endif /* USE_ELF_CORE_DUMP */
4096
4097 void do_init_thread(struct target_pt_regs *regs, struct image_info *infop)
4098 {
4099 init_thread(regs, infop);
4100 }
AR7 emulation for QEMU