?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
sev/i386: Allow AP booting under SEV-ES
[qemu/ar7.git] / linux-user / elfload.c
blobf5bd4076fcf5043b397456bfbf8c14fe9106b168
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(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(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 };
590
591 #define ELF_HWCAP get_elf_hwcap()
592 #define ELF_HWCAP2 get_elf_hwcap2()
593
594 #define GET_FEATURE_ID(feat, hwcap) \
595 do { if (cpu_isar_feature(feat, cpu)) { hwcaps |= hwcap; } } while (0)
596
597 static uint32_t get_elf_hwcap(void)
598 {
599 ARMCPU *cpu = ARM_CPU(thread_cpu);
600 uint32_t hwcaps = 0;
601
602 hwcaps |= ARM_HWCAP_A64_FP;
603 hwcaps |= ARM_HWCAP_A64_ASIMD;
604 hwcaps |= ARM_HWCAP_A64_CPUID;
605
606 /* probe for the extra features */
607
608 GET_FEATURE_ID(aa64_aes, ARM_HWCAP_A64_AES);
609 GET_FEATURE_ID(aa64_pmull, ARM_HWCAP_A64_PMULL);
610 GET_FEATURE_ID(aa64_sha1, ARM_HWCAP_A64_SHA1);
611 GET_FEATURE_ID(aa64_sha256, ARM_HWCAP_A64_SHA2);
612 GET_FEATURE_ID(aa64_sha512, ARM_HWCAP_A64_SHA512);
613 GET_FEATURE_ID(aa64_crc32, ARM_HWCAP_A64_CRC32);
614 GET_FEATURE_ID(aa64_sha3, ARM_HWCAP_A64_SHA3);
615 GET_FEATURE_ID(aa64_sm3, ARM_HWCAP_A64_SM3);
616 GET_FEATURE_ID(aa64_sm4, ARM_HWCAP_A64_SM4);
617 GET_FEATURE_ID(aa64_fp16, ARM_HWCAP_A64_FPHP | ARM_HWCAP_A64_ASIMDHP);
618 GET_FEATURE_ID(aa64_atomics, ARM_HWCAP_A64_ATOMICS);
619 GET_FEATURE_ID(aa64_rdm, ARM_HWCAP_A64_ASIMDRDM);
620 GET_FEATURE_ID(aa64_dp, ARM_HWCAP_A64_ASIMDDP);
621 GET_FEATURE_ID(aa64_fcma, ARM_HWCAP_A64_FCMA);
622 GET_FEATURE_ID(aa64_sve, ARM_HWCAP_A64_SVE);
623 GET_FEATURE_ID(aa64_pauth, ARM_HWCAP_A64_PACA | ARM_HWCAP_A64_PACG);
624 GET_FEATURE_ID(aa64_fhm, ARM_HWCAP_A64_ASIMDFHM);
625 GET_FEATURE_ID(aa64_jscvt, ARM_HWCAP_A64_JSCVT);
626 GET_FEATURE_ID(aa64_sb, ARM_HWCAP_A64_SB);
627 GET_FEATURE_ID(aa64_condm_4, ARM_HWCAP_A64_FLAGM);
628 GET_FEATURE_ID(aa64_dcpop, ARM_HWCAP_A64_DCPOP);
629 GET_FEATURE_ID(aa64_rcpc_8_3, ARM_HWCAP_A64_LRCPC);
630 GET_FEATURE_ID(aa64_rcpc_8_4, ARM_HWCAP_A64_ILRCPC);
631
632 return hwcaps;
633 }
634
635 static uint32_t get_elf_hwcap2(void)
636 {
637 ARMCPU *cpu = ARM_CPU(thread_cpu);
638 uint32_t hwcaps = 0;
639
640 GET_FEATURE_ID(aa64_dcpodp, ARM_HWCAP2_A64_DCPODP);
641 GET_FEATURE_ID(aa64_condm_5, ARM_HWCAP2_A64_FLAGM2);
642 GET_FEATURE_ID(aa64_frint, ARM_HWCAP2_A64_FRINT);
643
644 return hwcaps;
645 }
646
647 #undef GET_FEATURE_ID
648
649 #endif /* not TARGET_AARCH64 */
650 #endif /* TARGET_ARM */
651
652 #ifdef TARGET_SPARC
653 #ifdef TARGET_SPARC64
654
655 #define ELF_START_MMAP 0x80000000
656 #define ELF_HWCAP (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \
657 | HWCAP_SPARC_MULDIV | HWCAP_SPARC_V9)
658 #ifndef TARGET_ABI32
659 #define elf_check_arch(x) ( (x) == EM_SPARCV9 || (x) == EM_SPARC32PLUS )
660 #else
661 #define elf_check_arch(x) ( (x) == EM_SPARC32PLUS || (x) == EM_SPARC )
662 #endif
663
664 #define ELF_CLASS ELFCLASS64
665 #define ELF_ARCH EM_SPARCV9
666
667 #define STACK_BIAS 2047
668
669 static inline void init_thread(struct target_pt_regs *regs,
670 struct image_info *infop)
671 {
672 #ifndef TARGET_ABI32
673 regs->tstate = 0;
674 #endif
675 regs->pc = infop->entry;
676 regs->npc = regs->pc + 4;
677 regs->y = 0;
678 #ifdef TARGET_ABI32
679 regs->u_regs[14] = infop->start_stack - 16 * 4;
680 #else
681 if (personality(infop->personality) == PER_LINUX32)
682 regs->u_regs[14] = infop->start_stack - 16 * 4;
683 else
684 regs->u_regs[14] = infop->start_stack - 16 * 8 - STACK_BIAS;
685 #endif
686 }
687
688 #else
689 #define ELF_START_MMAP 0x80000000
690 #define ELF_HWCAP (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \
691 | HWCAP_SPARC_MULDIV)
692
693 #define ELF_CLASS ELFCLASS32
694 #define ELF_ARCH EM_SPARC
695
696 static inline void init_thread(struct target_pt_regs *regs,
697 struct image_info *infop)
698 {
699 regs->psr = 0;
700 regs->pc = infop->entry;
701 regs->npc = regs->pc + 4;
702 regs->y = 0;
703 regs->u_regs[14] = infop->start_stack - 16 * 4;
704 }
705
706 #endif
707 #endif
708
709 #ifdef TARGET_PPC
710
711 #define ELF_MACHINE PPC_ELF_MACHINE
712 #define ELF_START_MMAP 0x80000000
713
714 #if defined(TARGET_PPC64) && !defined(TARGET_ABI32)
715
716 #define elf_check_arch(x) ( (x) == EM_PPC64 )
717
718 #define ELF_CLASS ELFCLASS64
719
720 #else
721
722 #define ELF_CLASS ELFCLASS32
723
724 #endif
725
726 #define ELF_ARCH EM_PPC
727
728 /* Feature masks for the Aux Vector Hardware Capabilities (AT_HWCAP).
729 See arch/powerpc/include/asm/cputable.h. */
730 enum {
731 QEMU_PPC_FEATURE_32 = 0x80000000,
732 QEMU_PPC_FEATURE_64 = 0x40000000,
733 QEMU_PPC_FEATURE_601_INSTR = 0x20000000,
734 QEMU_PPC_FEATURE_HAS_ALTIVEC = 0x10000000,
735 QEMU_PPC_FEATURE_HAS_FPU = 0x08000000,
736 QEMU_PPC_FEATURE_HAS_MMU = 0x04000000,
737 QEMU_PPC_FEATURE_HAS_4xxMAC = 0x02000000,
738 QEMU_PPC_FEATURE_UNIFIED_CACHE = 0x01000000,
739 QEMU_PPC_FEATURE_HAS_SPE = 0x00800000,
740 QEMU_PPC_FEATURE_HAS_EFP_SINGLE = 0x00400000,
741 QEMU_PPC_FEATURE_HAS_EFP_DOUBLE = 0x00200000,
742 QEMU_PPC_FEATURE_NO_TB = 0x00100000,
743 QEMU_PPC_FEATURE_POWER4 = 0x00080000,
744 QEMU_PPC_FEATURE_POWER5 = 0x00040000,
745 QEMU_PPC_FEATURE_POWER5_PLUS = 0x00020000,
746 QEMU_PPC_FEATURE_CELL = 0x00010000,
747 QEMU_PPC_FEATURE_BOOKE = 0x00008000,
748 QEMU_PPC_FEATURE_SMT = 0x00004000,
749 QEMU_PPC_FEATURE_ICACHE_SNOOP = 0x00002000,
750 QEMU_PPC_FEATURE_ARCH_2_05 = 0x00001000,
751 QEMU_PPC_FEATURE_PA6T = 0x00000800,
752 QEMU_PPC_FEATURE_HAS_DFP = 0x00000400,
753 QEMU_PPC_FEATURE_POWER6_EXT = 0x00000200,
754 QEMU_PPC_FEATURE_ARCH_2_06 = 0x00000100,
755 QEMU_PPC_FEATURE_HAS_VSX = 0x00000080,
756 QEMU_PPC_FEATURE_PSERIES_PERFMON_COMPAT = 0x00000040,
757
758 QEMU_PPC_FEATURE_TRUE_LE = 0x00000002,
759 QEMU_PPC_FEATURE_PPC_LE = 0x00000001,
760
761 /* Feature definitions in AT_HWCAP2. */
762 QEMU_PPC_FEATURE2_ARCH_2_07 = 0x80000000, /* ISA 2.07 */
763 QEMU_PPC_FEATURE2_HAS_HTM = 0x40000000, /* Hardware Transactional Memory */
764 QEMU_PPC_FEATURE2_HAS_DSCR = 0x20000000, /* Data Stream Control Register */
765 QEMU_PPC_FEATURE2_HAS_EBB = 0x10000000, /* Event Base Branching */
766 QEMU_PPC_FEATURE2_HAS_ISEL = 0x08000000, /* Integer Select */
767 QEMU_PPC_FEATURE2_HAS_TAR = 0x04000000, /* Target Address Register */
768 QEMU_PPC_FEATURE2_VEC_CRYPTO = 0x02000000,
769 QEMU_PPC_FEATURE2_HTM_NOSC = 0x01000000,
770 QEMU_PPC_FEATURE2_ARCH_3_00 = 0x00800000, /* ISA 3.00 */
771 QEMU_PPC_FEATURE2_HAS_IEEE128 = 0x00400000, /* VSX IEEE Bin Float 128-bit */
772 QEMU_PPC_FEATURE2_DARN = 0x00200000, /* darn random number insn */
773 QEMU_PPC_FEATURE2_SCV = 0x00100000, /* scv syscall */
774 QEMU_PPC_FEATURE2_HTM_NO_SUSPEND = 0x00080000, /* TM w/o suspended state */
775 };
776
777 #define ELF_HWCAP get_elf_hwcap()
778
779 static uint32_t get_elf_hwcap(void)
780 {
781 PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);
782 uint32_t features = 0;
783
784 /* We don't have to be terribly complete here; the high points are
785 Altivec/FP/SPE support. Anything else is just a bonus. */
786 #define GET_FEATURE(flag, feature) \
787 do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0)
788 #define GET_FEATURE2(flags, feature) \
789 do { \
790 if ((cpu->env.insns_flags2 & flags) == flags) { \
791 features |= feature; \
792 } \
793 } while (0)
794 GET_FEATURE(PPC_64B, QEMU_PPC_FEATURE_64);
795 GET_FEATURE(PPC_FLOAT, QEMU_PPC_FEATURE_HAS_FPU);
796 GET_FEATURE(PPC_ALTIVEC, QEMU_PPC_FEATURE_HAS_ALTIVEC);
797 GET_FEATURE(PPC_SPE, QEMU_PPC_FEATURE_HAS_SPE);
798 GET_FEATURE(PPC_SPE_SINGLE, QEMU_PPC_FEATURE_HAS_EFP_SINGLE);
799 GET_FEATURE(PPC_SPE_DOUBLE, QEMU_PPC_FEATURE_HAS_EFP_DOUBLE);
800 GET_FEATURE(PPC_BOOKE, QEMU_PPC_FEATURE_BOOKE);
801 GET_FEATURE(PPC_405_MAC, QEMU_PPC_FEATURE_HAS_4xxMAC);
802 GET_FEATURE2(PPC2_DFP, QEMU_PPC_FEATURE_HAS_DFP);
803 GET_FEATURE2(PPC2_VSX, QEMU_PPC_FEATURE_HAS_VSX);
804 GET_FEATURE2((PPC2_PERM_ISA206 | PPC2_DIVE_ISA206 | PPC2_ATOMIC_ISA206 |
805 PPC2_FP_CVT_ISA206 | PPC2_FP_TST_ISA206),
806 QEMU_PPC_FEATURE_ARCH_2_06);
807 #undef GET_FEATURE
808 #undef GET_FEATURE2
809
810 return features;
811 }
812
813 #define ELF_HWCAP2 get_elf_hwcap2()
814
815 static uint32_t get_elf_hwcap2(void)
816 {
817 PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);
818 uint32_t features = 0;
819
820 #define GET_FEATURE(flag, feature) \
821 do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0)
822 #define GET_FEATURE2(flag, feature) \
823 do { if (cpu->env.insns_flags2 & flag) { features |= feature; } } while (0)
824
825 GET_FEATURE(PPC_ISEL, QEMU_PPC_FEATURE2_HAS_ISEL);
826 GET_FEATURE2(PPC2_BCTAR_ISA207, QEMU_PPC_FEATURE2_HAS_TAR);
827 GET_FEATURE2((PPC2_BCTAR_ISA207 | PPC2_LSQ_ISA207 | PPC2_ALTIVEC_207 |
828 PPC2_ISA207S), QEMU_PPC_FEATURE2_ARCH_2_07 |
829 QEMU_PPC_FEATURE2_VEC_CRYPTO);
830 GET_FEATURE2(PPC2_ISA300, QEMU_PPC_FEATURE2_ARCH_3_00 |
831 QEMU_PPC_FEATURE2_DARN);
832
833 #undef GET_FEATURE
834 #undef GET_FEATURE2
835
836 return features;
837 }
838
839 /*
840 * The requirements here are:
841 * - keep the final alignment of sp (sp & 0xf)
842 * - make sure the 32-bit value at the first 16 byte aligned position of
843 * AUXV is greater than 16 for glibc compatibility.
844 * AT_IGNOREPPC is used for that.
845 * - for compatibility with glibc ARCH_DLINFO must always be defined on PPC,
846 * even if DLINFO_ARCH_ITEMS goes to zero or is undefined.
847 */
848 #define DLINFO_ARCH_ITEMS 5
849 #define ARCH_DLINFO \
850 do { \
851 PowerPCCPU *cpu = POWERPC_CPU(thread_cpu); \
852 /* \
853 * Handle glibc compatibility: these magic entries must \
854 * be at the lowest addresses in the final auxv. \
855 */ \
856 NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC); \
857 NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC); \
858 NEW_AUX_ENT(AT_DCACHEBSIZE, cpu->env.dcache_line_size); \
859 NEW_AUX_ENT(AT_ICACHEBSIZE, cpu->env.icache_line_size); \
860 NEW_AUX_ENT(AT_UCACHEBSIZE, 0); \
861 } while (0)
862
863 static inline void init_thread(struct target_pt_regs *_regs, struct image_info *infop)
864 {
865 _regs->gpr[1] = infop->start_stack;
866 #if defined(TARGET_PPC64) && !defined(TARGET_ABI32)
867 if (get_ppc64_abi(infop) < 2) {
868 uint64_t val;
869 get_user_u64(val, infop->entry + 8);
870 _regs->gpr[2] = val + infop->load_bias;
871 get_user_u64(val, infop->entry);
872 infop->entry = val + infop->load_bias;
873 } else {
874 _regs->gpr[12] = infop->entry; /* r12 set to global entry address */
875 }
876 #endif
877 _regs->nip = infop->entry;
878 }
879
880 /* See linux kernel: arch/powerpc/include/asm/elf.h. */
881 #define ELF_NREG 48
882 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
883
884 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUPPCState *env)
885 {
886 int i;
887 target_ulong ccr = 0;
888
889 for (i = 0; i < ARRAY_SIZE(env->gpr); i++) {
890 (*regs)[i] = tswapreg(env->gpr[i]);
891 }
892
893 (*regs)[32] = tswapreg(env->nip);
894 (*regs)[33] = tswapreg(env->msr);
895 (*regs)[35] = tswapreg(env->ctr);
896 (*regs)[36] = tswapreg(env->lr);
897 (*regs)[37] = tswapreg(env->xer);
898
899 for (i = 0; i < ARRAY_SIZE(env->crf); i++) {
900 ccr |= env->crf[i] << (32 - ((i + 1) * 4));
901 }
902 (*regs)[38] = tswapreg(ccr);
903 }
904
905 #define USE_ELF_CORE_DUMP
906 #define ELF_EXEC_PAGESIZE 4096
907
908 #endif
909
910 #ifdef TARGET_MIPS
911
912 #define ELF_START_MMAP 0x80000000
913
914 #ifdef TARGET_MIPS64
915 #define ELF_CLASS ELFCLASS64
916 #else
917 #define ELF_CLASS ELFCLASS32
918 #endif
919 #define ELF_ARCH EM_MIPS
920
921 #define elf_check_arch(x) ((x) == EM_MIPS || (x) == EM_NANOMIPS)
922
923 #ifdef TARGET_ABI_MIPSN32
924 #define elf_check_abi(x) ((x) & EF_MIPS_ABI2)
925 #else
926 #define elf_check_abi(x) (!((x) & EF_MIPS_ABI2))
927 #endif
928
929 static inline void init_thread(struct target_pt_regs *regs,
930 struct image_info *infop)
931 {
932 regs->cp0_status = 2 << CP0St_KSU;
933 regs->cp0_epc = infop->entry;
934 regs->regs[29] = infop->start_stack;
935 }
936
937 /* See linux kernel: arch/mips/include/asm/elf.h. */
938 #define ELF_NREG 45
939 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
940
941 /* See linux kernel: arch/mips/include/asm/reg.h. */
942 enum {
943 #ifdef TARGET_MIPS64
944 TARGET_EF_R0 = 0,
945 #else
946 TARGET_EF_R0 = 6,
947 #endif
948 TARGET_EF_R26 = TARGET_EF_R0 + 26,
949 TARGET_EF_R27 = TARGET_EF_R0 + 27,
950 TARGET_EF_LO = TARGET_EF_R0 + 32,
951 TARGET_EF_HI = TARGET_EF_R0 + 33,
952 TARGET_EF_CP0_EPC = TARGET_EF_R0 + 34,
953 TARGET_EF_CP0_BADVADDR = TARGET_EF_R0 + 35,
954 TARGET_EF_CP0_STATUS = TARGET_EF_R0 + 36,
955 TARGET_EF_CP0_CAUSE = TARGET_EF_R0 + 37
956 };
957
958 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */
959 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMIPSState *env)
960 {
961 int i;
962
963 for (i = 0; i < TARGET_EF_R0; i++) {
964 (*regs)[i] = 0;
965 }
966 (*regs)[TARGET_EF_R0] = 0;
967
968 for (i = 1; i < ARRAY_SIZE(env->active_tc.gpr); i++) {
969 (*regs)[TARGET_EF_R0 + i] = tswapreg(env->active_tc.gpr[i]);
970 }
971
972 (*regs)[TARGET_EF_R26] = 0;
973 (*regs)[TARGET_EF_R27] = 0;
974 (*regs)[TARGET_EF_LO] = tswapreg(env->active_tc.LO[0]);
975 (*regs)[TARGET_EF_HI] = tswapreg(env->active_tc.HI[0]);
976 (*regs)[TARGET_EF_CP0_EPC] = tswapreg(env->active_tc.PC);
977 (*regs)[TARGET_EF_CP0_BADVADDR] = tswapreg(env->CP0_BadVAddr);
978 (*regs)[TARGET_EF_CP0_STATUS] = tswapreg(env->CP0_Status);
979 (*regs)[TARGET_EF_CP0_CAUSE] = tswapreg(env->CP0_Cause);
980 }
981
982 #define USE_ELF_CORE_DUMP
983 #define ELF_EXEC_PAGESIZE 4096
984
985 /* See arch/mips/include/uapi/asm/hwcap.h. */
986 enum {
987 HWCAP_MIPS_R6 = (1 << 0),
988 HWCAP_MIPS_MSA = (1 << 1),
989 HWCAP_MIPS_CRC32 = (1 << 2),
990 HWCAP_MIPS_MIPS16 = (1 << 3),
991 HWCAP_MIPS_MDMX = (1 << 4),
992 HWCAP_MIPS_MIPS3D = (1 << 5),
993 HWCAP_MIPS_SMARTMIPS = (1 << 6),
994 HWCAP_MIPS_DSP = (1 << 7),
995 HWCAP_MIPS_DSP2 = (1 << 8),
996 HWCAP_MIPS_DSP3 = (1 << 9),
997 HWCAP_MIPS_MIPS16E2 = (1 << 10),
998 HWCAP_LOONGSON_MMI = (1 << 11),
999 HWCAP_LOONGSON_EXT = (1 << 12),
1000 HWCAP_LOONGSON_EXT2 = (1 << 13),
1001 HWCAP_LOONGSON_CPUCFG = (1 << 14),
1002 };
1003
1004 #define ELF_HWCAP get_elf_hwcap()
1005
1006 #define GET_FEATURE_INSN(_flag, _hwcap) \
1007 do { if (cpu->env.insn_flags & (_flag)) { hwcaps |= _hwcap; } } while (0)
1008
1009 #define GET_FEATURE_REG_SET(_reg, _mask, _hwcap) \
1010 do { if (cpu->env._reg & (_mask)) { hwcaps |= _hwcap; } } while (0)
1011
1012 #define GET_FEATURE_REG_EQU(_reg, _start, _length, _val, _hwcap) \
1013 do { \
1014 if (extract32(cpu->env._reg, (_start), (_length)) == (_val)) { \
1015 hwcaps |= _hwcap; \
1016 } \
1017 } while (0)
1018
1019 static uint32_t get_elf_hwcap(void)
1020 {
1021 MIPSCPU *cpu = MIPS_CPU(thread_cpu);
1022 uint32_t hwcaps = 0;
1023
1024 GET_FEATURE_REG_EQU(CP0_Config0, CP0C0_AR, CP0C0_AR_LENGTH,
1025 2, HWCAP_MIPS_R6);
1026 GET_FEATURE_REG_SET(CP0_Config3, 1 << CP0C3_MSAP, HWCAP_MIPS_MSA);
1027 GET_FEATURE_INSN(ASE_LMMI, HWCAP_LOONGSON_MMI);
1028 GET_FEATURE_INSN(ASE_LEXT, HWCAP_LOONGSON_EXT);
1029
1030 return hwcaps;
1031 }
1032
1033 #undef GET_FEATURE_REG_EQU
1034 #undef GET_FEATURE_REG_SET
1035 #undef GET_FEATURE_INSN
1036
1037 #endif /* TARGET_MIPS */
1038
1039 #ifdef TARGET_MICROBLAZE
1040
1041 #define ELF_START_MMAP 0x80000000
1042
1043 #define elf_check_arch(x) ( (x) == EM_MICROBLAZE || (x) == EM_MICROBLAZE_OLD)
1044
1045 #define ELF_CLASS ELFCLASS32
1046 #define ELF_ARCH EM_MICROBLAZE
1047
1048 static inline void init_thread(struct target_pt_regs *regs,
1049 struct image_info *infop)
1050 {
1051 regs->pc = infop->entry;
1052 regs->r1 = infop->start_stack;
1053
1054 }
1055
1056 #define ELF_EXEC_PAGESIZE 4096
1057
1058 #define USE_ELF_CORE_DUMP
1059 #define ELF_NREG 38
1060 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1061
1062 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */
1063 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMBState *env)
1064 {
1065 int i, pos = 0;
1066
1067 for (i = 0; i < 32; i++) {
1068 (*regs)[pos++] = tswapreg(env->regs[i]);
1069 }
1070
1071 (*regs)[pos++] = tswapreg(env->pc);
1072 (*regs)[pos++] = tswapreg(mb_cpu_read_msr(env));
1073 (*regs)[pos++] = 0;
1074 (*regs)[pos++] = tswapreg(env->ear);
1075 (*regs)[pos++] = 0;
1076 (*regs)[pos++] = tswapreg(env->esr);
1077 }
1078
1079 #endif /* TARGET_MICROBLAZE */
1080
1081 #ifdef TARGET_NIOS2
1082
1083 #define ELF_START_MMAP 0x80000000
1084
1085 #define elf_check_arch(x) ((x) == EM_ALTERA_NIOS2)
1086
1087 #define ELF_CLASS ELFCLASS32
1088 #define ELF_ARCH EM_ALTERA_NIOS2
1089
1090 static void init_thread(struct target_pt_regs *regs, struct image_info *infop)
1091 {
1092 regs->ea = infop->entry;
1093 regs->sp = infop->start_stack;
1094 regs->estatus = 0x3;
1095 }
1096
1097 #define ELF_EXEC_PAGESIZE 4096
1098
1099 #define USE_ELF_CORE_DUMP
1100 #define ELF_NREG 49
1101 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1102
1103 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */
1104 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1105 const CPUNios2State *env)
1106 {
1107 int i;
1108
1109 (*regs)[0] = -1;
1110 for (i = 1; i < 8; i++) /* r0-r7 */
1111 (*regs)[i] = tswapreg(env->regs[i + 7]);
1112
1113 for (i = 8; i < 16; i++) /* r8-r15 */
1114 (*regs)[i] = tswapreg(env->regs[i - 8]);
1115
1116 for (i = 16; i < 24; i++) /* r16-r23 */
1117 (*regs)[i] = tswapreg(env->regs[i + 7]);
1118 (*regs)[24] = -1; /* R_ET */
1119 (*regs)[25] = -1; /* R_BT */
1120 (*regs)[26] = tswapreg(env->regs[R_GP]);
1121 (*regs)[27] = tswapreg(env->regs[R_SP]);
1122 (*regs)[28] = tswapreg(env->regs[R_FP]);
1123 (*regs)[29] = tswapreg(env->regs[R_EA]);
1124 (*regs)[30] = -1; /* R_SSTATUS */
1125 (*regs)[31] = tswapreg(env->regs[R_RA]);
1126
1127 (*regs)[32] = tswapreg(env->regs[R_PC]);
1128
1129 (*regs)[33] = -1; /* R_STATUS */
1130 (*regs)[34] = tswapreg(env->regs[CR_ESTATUS]);
1131
1132 for (i = 35; i < 49; i++) /* ... */
1133 (*regs)[i] = -1;
1134 }
1135
1136 #endif /* TARGET_NIOS2 */
1137
1138 #ifdef TARGET_OPENRISC
1139
1140 #define ELF_START_MMAP 0x08000000
1141
1142 #define ELF_ARCH EM_OPENRISC
1143 #define ELF_CLASS ELFCLASS32
1144 #define ELF_DATA ELFDATA2MSB
1145
1146 static inline void init_thread(struct target_pt_regs *regs,
1147 struct image_info *infop)
1148 {
1149 regs->pc = infop->entry;
1150 regs->gpr[1] = infop->start_stack;
1151 }
1152
1153 #define USE_ELF_CORE_DUMP
1154 #define ELF_EXEC_PAGESIZE 8192
1155
1156 /* See linux kernel arch/openrisc/include/asm/elf.h. */
1157 #define ELF_NREG 34 /* gprs and pc, sr */
1158 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1159
1160 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1161 const CPUOpenRISCState *env)
1162 {
1163 int i;
1164
1165 for (i = 0; i < 32; i++) {
1166 (*regs)[i] = tswapreg(cpu_get_gpr(env, i));
1167 }
1168 (*regs)[32] = tswapreg(env->pc);
1169 (*regs)[33] = tswapreg(cpu_get_sr(env));
1170 }
1171 #define ELF_HWCAP 0
1172 #define ELF_PLATFORM NULL
1173
1174 #endif /* TARGET_OPENRISC */
1175
1176 #ifdef TARGET_SH4
1177
1178 #define ELF_START_MMAP 0x80000000
1179
1180 #define ELF_CLASS ELFCLASS32
1181 #define ELF_ARCH EM_SH
1182
1183 static inline void init_thread(struct target_pt_regs *regs,
1184 struct image_info *infop)
1185 {
1186 /* Check other registers XXXXX */
1187 regs->pc = infop->entry;
1188 regs->regs[15] = infop->start_stack;
1189 }
1190
1191 /* See linux kernel: arch/sh/include/asm/elf.h. */
1192 #define ELF_NREG 23
1193 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1194
1195 /* See linux kernel: arch/sh/include/asm/ptrace.h. */
1196 enum {
1197 TARGET_REG_PC = 16,
1198 TARGET_REG_PR = 17,
1199 TARGET_REG_SR = 18,
1200 TARGET_REG_GBR = 19,
1201 TARGET_REG_MACH = 20,
1202 TARGET_REG_MACL = 21,
1203 TARGET_REG_SYSCALL = 22
1204 };
1205
1206 static inline void elf_core_copy_regs(target_elf_gregset_t *regs,
1207 const CPUSH4State *env)
1208 {
1209 int i;
1210
1211 for (i = 0; i < 16; i++) {
1212 (*regs)[i] = tswapreg(env->gregs[i]);
1213 }
1214
1215 (*regs)[TARGET_REG_PC] = tswapreg(env->pc);
1216 (*regs)[TARGET_REG_PR] = tswapreg(env->pr);
1217 (*regs)[TARGET_REG_SR] = tswapreg(env->sr);
1218 (*regs)[TARGET_REG_GBR] = tswapreg(env->gbr);
1219 (*regs)[TARGET_REG_MACH] = tswapreg(env->mach);
1220 (*regs)[TARGET_REG_MACL] = tswapreg(env->macl);
1221 (*regs)[TARGET_REG_SYSCALL] = 0; /* FIXME */
1222 }
1223
1224 #define USE_ELF_CORE_DUMP
1225 #define ELF_EXEC_PAGESIZE 4096
1226
1227 enum {
1228 SH_CPU_HAS_FPU = 0x0001, /* Hardware FPU support */
1229 SH_CPU_HAS_P2_FLUSH_BUG = 0x0002, /* Need to flush the cache in P2 area */
1230 SH_CPU_HAS_MMU_PAGE_ASSOC = 0x0004, /* SH3: TLB way selection bit support */
1231 SH_CPU_HAS_DSP = 0x0008, /* SH-DSP: DSP support */
1232 SH_CPU_HAS_PERF_COUNTER = 0x0010, /* Hardware performance counters */
1233 SH_CPU_HAS_PTEA = 0x0020, /* PTEA register */
1234 SH_CPU_HAS_LLSC = 0x0040, /* movli.l/movco.l */
1235 SH_CPU_HAS_L2_CACHE = 0x0080, /* Secondary cache / URAM */
1236 SH_CPU_HAS_OP32 = 0x0100, /* 32-bit instruction support */
1237 SH_CPU_HAS_PTEAEX = 0x0200, /* PTE ASID Extension support */
1238 };
1239
1240 #define ELF_HWCAP get_elf_hwcap()
1241
1242 static uint32_t get_elf_hwcap(void)
1243 {
1244 SuperHCPU *cpu = SUPERH_CPU(thread_cpu);
1245 uint32_t hwcap = 0;
1246
1247 hwcap |= SH_CPU_HAS_FPU;
1248
1249 if (cpu->env.features & SH_FEATURE_SH4A) {
1250 hwcap |= SH_CPU_HAS_LLSC;
1251 }
1252
1253 return hwcap;
1254 }
1255
1256 #endif
1257
1258 #ifdef TARGET_CRIS
1259
1260 #define ELF_START_MMAP 0x80000000
1261
1262 #define ELF_CLASS ELFCLASS32
1263 #define ELF_ARCH EM_CRIS
1264
1265 static inline void init_thread(struct target_pt_regs *regs,
1266 struct image_info *infop)
1267 {
1268 regs->erp = infop->entry;
1269 }
1270
1271 #define ELF_EXEC_PAGESIZE 8192
1272
1273 #endif
1274
1275 #ifdef TARGET_M68K
1276
1277 #define ELF_START_MMAP 0x80000000
1278
1279 #define ELF_CLASS ELFCLASS32
1280 #define ELF_ARCH EM_68K
1281
1282 /* ??? Does this need to do anything?
1283 #define ELF_PLAT_INIT(_r) */
1284
1285 static inline void init_thread(struct target_pt_regs *regs,
1286 struct image_info *infop)
1287 {
1288 regs->usp = infop->start_stack;
1289 regs->sr = 0;
1290 regs->pc = infop->entry;
1291 }
1292
1293 /* See linux kernel: arch/m68k/include/asm/elf.h. */
1294 #define ELF_NREG 20
1295 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1296
1297 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUM68KState *env)
1298 {
1299 (*regs)[0] = tswapreg(env->dregs[1]);
1300 (*regs)[1] = tswapreg(env->dregs[2]);
1301 (*regs)[2] = tswapreg(env->dregs[3]);
1302 (*regs)[3] = tswapreg(env->dregs[4]);
1303 (*regs)[4] = tswapreg(env->dregs[5]);
1304 (*regs)[5] = tswapreg(env->dregs[6]);
1305 (*regs)[6] = tswapreg(env->dregs[7]);
1306 (*regs)[7] = tswapreg(env->aregs[0]);
1307 (*regs)[8] = tswapreg(env->aregs[1]);
1308 (*regs)[9] = tswapreg(env->aregs[2]);
1309 (*regs)[10] = tswapreg(env->aregs[3]);
1310 (*regs)[11] = tswapreg(env->aregs[4]);
1311 (*regs)[12] = tswapreg(env->aregs[5]);
1312 (*regs)[13] = tswapreg(env->aregs[6]);
1313 (*regs)[14] = tswapreg(env->dregs[0]);
1314 (*regs)[15] = tswapreg(env->aregs[7]);
1315 (*regs)[16] = tswapreg(env->dregs[0]); /* FIXME: orig_d0 */
1316 (*regs)[17] = tswapreg(env->sr);
1317 (*regs)[18] = tswapreg(env->pc);
1318 (*regs)[19] = 0; /* FIXME: regs->format | regs->vector */
1319 }
1320
1321 #define USE_ELF_CORE_DUMP
1322 #define ELF_EXEC_PAGESIZE 8192
1323
1324 #endif
1325
1326 #ifdef TARGET_ALPHA
1327
1328 #define ELF_START_MMAP (0x30000000000ULL)
1329
1330 #define ELF_CLASS ELFCLASS64
1331 #define ELF_ARCH EM_ALPHA
1332
1333 static inline void init_thread(struct target_pt_regs *regs,
1334 struct image_info *infop)
1335 {
1336 regs->pc = infop->entry;
1337 regs->ps = 8;
1338 regs->usp = infop->start_stack;
1339 }
1340
1341 #define ELF_EXEC_PAGESIZE 8192
1342
1343 #endif /* TARGET_ALPHA */
1344
1345 #ifdef TARGET_S390X
1346
1347 #define ELF_START_MMAP (0x20000000000ULL)
1348
1349 #define ELF_CLASS ELFCLASS64
1350 #define ELF_DATA ELFDATA2MSB
1351 #define ELF_ARCH EM_S390
1352
1353 #include "elf.h"
1354
1355 #define ELF_HWCAP get_elf_hwcap()
1356
1357 #define GET_FEATURE(_feat, _hwcap) \
1358 do { if (s390_has_feat(_feat)) { hwcap |= _hwcap; } } while (0)
1359
1360 static uint32_t get_elf_hwcap(void)
1361 {
1362 /*
1363 * Let's assume we always have esan3 and zarch.
1364 * 31-bit processes can use 64-bit registers (high gprs).
1365 */
1366 uint32_t hwcap = HWCAP_S390_ESAN3 | HWCAP_S390_ZARCH | HWCAP_S390_HIGH_GPRS;
1367
1368 GET_FEATURE(S390_FEAT_STFLE, HWCAP_S390_STFLE);
1369 GET_FEATURE(S390_FEAT_MSA, HWCAP_S390_MSA);
1370 GET_FEATURE(S390_FEAT_LONG_DISPLACEMENT, HWCAP_S390_LDISP);
1371 GET_FEATURE(S390_FEAT_EXTENDED_IMMEDIATE, HWCAP_S390_EIMM);
1372 if (s390_has_feat(S390_FEAT_EXTENDED_TRANSLATION_3) &&
1373 s390_has_feat(S390_FEAT_ETF3_ENH)) {
1374 hwcap |= HWCAP_S390_ETF3EH;
1375 }
1376 GET_FEATURE(S390_FEAT_VECTOR, HWCAP_S390_VXRS);
1377
1378 return hwcap;
1379 }
1380
1381 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop)
1382 {
1383 regs->psw.addr = infop->entry;
1384 regs->psw.mask = PSW_MASK_64 | PSW_MASK_32;
1385 regs->gprs[15] = infop->start_stack;
1386 }
1387
1388 #endif /* TARGET_S390X */
1389
1390 #ifdef TARGET_TILEGX
1391
1392 /* 42 bits real used address, a half for user mode */
1393 #define ELF_START_MMAP (0x00000020000000000ULL)
1394
1395 #define elf_check_arch(x) ((x) == EM_TILEGX)
1396
1397 #define ELF_CLASS ELFCLASS64
1398 #define ELF_DATA ELFDATA2LSB
1399 #define ELF_ARCH EM_TILEGX
1400
1401 static inline void init_thread(struct target_pt_regs *regs,
1402 struct image_info *infop)
1403 {
1404 regs->pc = infop->entry;
1405 regs->sp = infop->start_stack;
1406
1407 }
1408
1409 #define ELF_EXEC_PAGESIZE 65536 /* TILE-Gx page size is 64KB */
1410
1411 #endif /* TARGET_TILEGX */
1412
1413 #ifdef TARGET_RISCV
1414
1415 #define ELF_START_MMAP 0x80000000
1416 #define ELF_ARCH EM_RISCV
1417
1418 #ifdef TARGET_RISCV32
1419 #define ELF_CLASS ELFCLASS32
1420 #else
1421 #define ELF_CLASS ELFCLASS64
1422 #endif
1423
1424 static inline void init_thread(struct target_pt_regs *regs,
1425 struct image_info *infop)
1426 {
1427 regs->sepc = infop->entry;
1428 regs->sp = infop->start_stack;
1429 }
1430
1431 #define ELF_EXEC_PAGESIZE 4096
1432
1433 #endif /* TARGET_RISCV */
1434
1435 #ifdef TARGET_HPPA
1436
1437 #define ELF_START_MMAP 0x80000000
1438 #define ELF_CLASS ELFCLASS32
1439 #define ELF_ARCH EM_PARISC
1440 #define ELF_PLATFORM "PARISC"
1441 #define STACK_GROWS_DOWN 0
1442 #define STACK_ALIGNMENT 64
1443
1444 static inline void init_thread(struct target_pt_regs *regs,
1445 struct image_info *infop)
1446 {
1447 regs->iaoq[0] = infop->entry;
1448 regs->iaoq[1] = infop->entry + 4;
1449 regs->gr[23] = 0;
1450 regs->gr[24] = infop->arg_start;
1451 regs->gr[25] = (infop->arg_end - infop->arg_start) / sizeof(abi_ulong);
1452 /* The top-of-stack contains a linkage buffer. */
1453 regs->gr[30] = infop->start_stack + 64;
1454 regs->gr[31] = infop->entry;
1455 }
1456
1457 #endif /* TARGET_HPPA */
1458
1459 #ifdef TARGET_XTENSA
1460
1461 #define ELF_START_MMAP 0x20000000
1462
1463 #define ELF_CLASS ELFCLASS32
1464 #define ELF_ARCH EM_XTENSA
1465
1466 static inline void init_thread(struct target_pt_regs *regs,
1467 struct image_info *infop)
1468 {
1469 regs->windowbase = 0;
1470 regs->windowstart = 1;
1471 regs->areg[1] = infop->start_stack;
1472 regs->pc = infop->entry;
1473 }
1474
1475 /* See linux kernel: arch/xtensa/include/asm/elf.h. */
1476 #define ELF_NREG 128
1477 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1478
1479 enum {
1480 TARGET_REG_PC,
1481 TARGET_REG_PS,
1482 TARGET_REG_LBEG,
1483 TARGET_REG_LEND,
1484 TARGET_REG_LCOUNT,
1485 TARGET_REG_SAR,
1486 TARGET_REG_WINDOWSTART,
1487 TARGET_REG_WINDOWBASE,
1488 TARGET_REG_THREADPTR,
1489 TARGET_REG_AR0 = 64,
1490 };
1491
1492 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1493 const CPUXtensaState *env)
1494 {
1495 unsigned i;
1496
1497 (*regs)[TARGET_REG_PC] = tswapreg(env->pc);
1498 (*regs)[TARGET_REG_PS] = tswapreg(env->sregs[PS] & ~PS_EXCM);
1499 (*regs)[TARGET_REG_LBEG] = tswapreg(env->sregs[LBEG]);
1500 (*regs)[TARGET_REG_LEND] = tswapreg(env->sregs[LEND]);
1501 (*regs)[TARGET_REG_LCOUNT] = tswapreg(env->sregs[LCOUNT]);
1502 (*regs)[TARGET_REG_SAR] = tswapreg(env->sregs[SAR]);
1503 (*regs)[TARGET_REG_WINDOWSTART] = tswapreg(env->sregs[WINDOW_START]);
1504 (*regs)[TARGET_REG_WINDOWBASE] = tswapreg(env->sregs[WINDOW_BASE]);
1505 (*regs)[TARGET_REG_THREADPTR] = tswapreg(env->uregs[THREADPTR]);
1506 xtensa_sync_phys_from_window((CPUXtensaState *)env);
1507 for (i = 0; i < env->config->nareg; ++i) {
1508 (*regs)[TARGET_REG_AR0 + i] = tswapreg(env->phys_regs[i]);
1509 }
1510 }
1511
1512 #define USE_ELF_CORE_DUMP
1513 #define ELF_EXEC_PAGESIZE 4096
1514
1515 #endif /* TARGET_XTENSA */
1516
1517 #ifndef ELF_PLATFORM
1518 #define ELF_PLATFORM (NULL)
1519 #endif
1520
1521 #ifndef ELF_MACHINE
1522 #define ELF_MACHINE ELF_ARCH
1523 #endif
1524
1525 #ifndef elf_check_arch
1526 #define elf_check_arch(x) ((x) == ELF_ARCH)
1527 #endif
1528
1529 #ifndef elf_check_abi
1530 #define elf_check_abi(x) (1)
1531 #endif
1532
1533 #ifndef ELF_HWCAP
1534 #define ELF_HWCAP 0
1535 #endif
1536
1537 #ifndef STACK_GROWS_DOWN
1538 #define STACK_GROWS_DOWN 1
1539 #endif
1540
1541 #ifndef STACK_ALIGNMENT
1542 #define STACK_ALIGNMENT 16
1543 #endif
1544
1545 #ifdef TARGET_ABI32
1546 #undef ELF_CLASS
1547 #define ELF_CLASS ELFCLASS32
1548 #undef bswaptls
1549 #define bswaptls(ptr) bswap32s(ptr)
1550 #endif
1551
1552 #include "elf.h"
1553
1554 /* We must delay the following stanzas until after "elf.h". */
1555 #if defined(TARGET_AARCH64)
1556
1557 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz,
1558 const uint32_t *data,
1559 struct image_info *info,
1560 Error **errp)
1561 {
1562 if (pr_type == GNU_PROPERTY_AARCH64_FEATURE_1_AND) {
1563 if (pr_datasz != sizeof(uint32_t)) {
1564 error_setg(errp, "Ill-formed GNU_PROPERTY_AARCH64_FEATURE_1_AND");
1565 return false;
1566 }
1567 /* We will extract GNU_PROPERTY_AARCH64_FEATURE_1_BTI later. */
1568 info->note_flags = *data;
1569 }
1570 return true;
1571 }
1572 #define ARCH_USE_GNU_PROPERTY 1
1573
1574 #else
1575
1576 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz,
1577 const uint32_t *data,
1578 struct image_info *info,
1579 Error **errp)
1580 {
1581 g_assert_not_reached();
1582 }
1583 #define ARCH_USE_GNU_PROPERTY 0
1584
1585 #endif
1586
1587 struct exec
1588 {
1589 unsigned int a_info; /* Use macros N_MAGIC, etc for access */
1590 unsigned int a_text; /* length of text, in bytes */
1591 unsigned int a_data; /* length of data, in bytes */
1592 unsigned int a_bss; /* length of uninitialized data area, in bytes */
1593 unsigned int a_syms; /* length of symbol table data in file, in bytes */
1594 unsigned int a_entry; /* start address */
1595 unsigned int a_trsize; /* length of relocation info for text, in bytes */
1596 unsigned int a_drsize; /* length of relocation info for data, in bytes */
1597 };
1598
1599
1600 #define N_MAGIC(exec) ((exec).a_info & 0xffff)
1601 #define OMAGIC 0407
1602 #define NMAGIC 0410
1603 #define ZMAGIC 0413
1604 #define QMAGIC 0314
1605
1606 /* Necessary parameters */
1607 #define TARGET_ELF_EXEC_PAGESIZE \
1608 (((eppnt->p_align & ~qemu_host_page_mask) != 0) ? \
1609 TARGET_PAGE_SIZE : MAX(qemu_host_page_size, TARGET_PAGE_SIZE))
1610 #define TARGET_ELF_PAGELENGTH(_v) ROUND_UP((_v), TARGET_ELF_EXEC_PAGESIZE)
1611 #define TARGET_ELF_PAGESTART(_v) ((_v) & \
1612 ~(abi_ulong)(TARGET_ELF_EXEC_PAGESIZE-1))
1613 #define TARGET_ELF_PAGEOFFSET(_v) ((_v) & (TARGET_ELF_EXEC_PAGESIZE-1))
1614
1615 #define DLINFO_ITEMS 16
1616
1617 static inline void memcpy_fromfs(void * to, const void * from, unsigned long n)
1618 {
1619 memcpy(to, from, n);
1620 }
1621
1622 #ifdef BSWAP_NEEDED
1623 static void bswap_ehdr(struct elfhdr *ehdr)
1624 {
1625 bswap16s(&ehdr->e_type); /* Object file type */
1626 bswap16s(&ehdr->e_machine); /* Architecture */
1627 bswap32s(&ehdr->e_version); /* Object file version */
1628 bswaptls(&ehdr->e_entry); /* Entry point virtual address */
1629 bswaptls(&ehdr->e_phoff); /* Program header table file offset */
1630 bswaptls(&ehdr->e_shoff); /* Section header table file offset */
1631 bswap32s(&ehdr->e_flags); /* Processor-specific flags */
1632 bswap16s(&ehdr->e_ehsize); /* ELF header size in bytes */
1633 bswap16s(&ehdr->e_phentsize); /* Program header table entry size */
1634 bswap16s(&ehdr->e_phnum); /* Program header table entry count */
1635 bswap16s(&ehdr->e_shentsize); /* Section header table entry size */
1636 bswap16s(&ehdr->e_shnum); /* Section header table entry count */
1637 bswap16s(&ehdr->e_shstrndx); /* Section header string table index */
1638 }
1639
1640 static void bswap_phdr(struct elf_phdr *phdr, int phnum)
1641 {
1642 int i;
1643 for (i = 0; i < phnum; ++i, ++phdr) {
1644 bswap32s(&phdr->p_type); /* Segment type */
1645 bswap32s(&phdr->p_flags); /* Segment flags */
1646 bswaptls(&phdr->p_offset); /* Segment file offset */
1647 bswaptls(&phdr->p_vaddr); /* Segment virtual address */
1648 bswaptls(&phdr->p_paddr); /* Segment physical address */
1649 bswaptls(&phdr->p_filesz); /* Segment size in file */
1650 bswaptls(&phdr->p_memsz); /* Segment size in memory */
1651 bswaptls(&phdr->p_align); /* Segment alignment */
1652 }
1653 }
1654
1655 static void bswap_shdr(struct elf_shdr *shdr, int shnum)
1656 {
1657 int i;
1658 for (i = 0; i < shnum; ++i, ++shdr) {
1659 bswap32s(&shdr->sh_name);
1660 bswap32s(&shdr->sh_type);
1661 bswaptls(&shdr->sh_flags);
1662 bswaptls(&shdr->sh_addr);
1663 bswaptls(&shdr->sh_offset);
1664 bswaptls(&shdr->sh_size);
1665 bswap32s(&shdr->sh_link);
1666 bswap32s(&shdr->sh_info);
1667 bswaptls(&shdr->sh_addralign);
1668 bswaptls(&shdr->sh_entsize);
1669 }
1670 }
1671
1672 static void bswap_sym(struct elf_sym *sym)
1673 {
1674 bswap32s(&sym->st_name);
1675 bswaptls(&sym->st_value);
1676 bswaptls(&sym->st_size);
1677 bswap16s(&sym->st_shndx);
1678 }
1679
1680 #ifdef TARGET_MIPS
1681 static void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags)
1682 {
1683 bswap16s(&abiflags->version);
1684 bswap32s(&abiflags->ases);
1685 bswap32s(&abiflags->isa_ext);
1686 bswap32s(&abiflags->flags1);
1687 bswap32s(&abiflags->flags2);
1688 }
1689 #endif
1690 #else
1691 static inline void bswap_ehdr(struct elfhdr *ehdr) { }
1692 static inline void bswap_phdr(struct elf_phdr *phdr, int phnum) { }
1693 static inline void bswap_shdr(struct elf_shdr *shdr, int shnum) { }
1694 static inline void bswap_sym(struct elf_sym *sym) { }
1695 #ifdef TARGET_MIPS
1696 static inline void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags) { }
1697 #endif
1698 #endif
1699
1700 #ifdef USE_ELF_CORE_DUMP
1701 static int elf_core_dump(int, const CPUArchState *);
1702 #endif /* USE_ELF_CORE_DUMP */
1703 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias);
1704
1705 /* Verify the portions of EHDR within E_IDENT for the target.
1706 This can be performed before bswapping the entire header. */
1707 static bool elf_check_ident(struct elfhdr *ehdr)
1708 {
1709 return (ehdr->e_ident[EI_MAG0] == ELFMAG0
1710 && ehdr->e_ident[EI_MAG1] == ELFMAG1
1711 && ehdr->e_ident[EI_MAG2] == ELFMAG2
1712 && ehdr->e_ident[EI_MAG3] == ELFMAG3
1713 && ehdr->e_ident[EI_CLASS] == ELF_CLASS
1714 && ehdr->e_ident[EI_DATA] == ELF_DATA
1715 && ehdr->e_ident[EI_VERSION] == EV_CURRENT);
1716 }
1717
1718 /* Verify the portions of EHDR outside of E_IDENT for the target.
1719 This has to wait until after bswapping the header. */
1720 static bool elf_check_ehdr(struct elfhdr *ehdr)
1721 {
1722 return (elf_check_arch(ehdr->e_machine)
1723 && elf_check_abi(ehdr->e_flags)
1724 && ehdr->e_ehsize == sizeof(struct elfhdr)
1725 && ehdr->e_phentsize == sizeof(struct elf_phdr)
1726 && (ehdr->e_type == ET_EXEC || ehdr->e_type == ET_DYN));
1727 }
1728
1729 /*
1730 * 'copy_elf_strings()' copies argument/envelope strings from user
1731 * memory to free pages in kernel mem. These are in a format ready
1732 * to be put directly into the top of new user memory.
1733 *
1734 */
1735 static abi_ulong copy_elf_strings(int argc, char **argv, char *scratch,
1736 abi_ulong p, abi_ulong stack_limit)
1737 {
1738 char *tmp;
1739 int len, i;
1740 abi_ulong top = p;
1741
1742 if (!p) {
1743 return 0; /* bullet-proofing */
1744 }
1745
1746 if (STACK_GROWS_DOWN) {
1747 int offset = ((p - 1) % TARGET_PAGE_SIZE) + 1;
1748 for (i = argc - 1; i >= 0; --i) {
1749 tmp = argv[i];
1750 if (!tmp) {
1751 fprintf(stderr, "VFS: argc is wrong");
1752 exit(-1);
1753 }
1754 len = strlen(tmp) + 1;
1755 tmp += len;
1756
1757 if (len > (p - stack_limit)) {
1758 return 0;
1759 }
1760 while (len) {
1761 int bytes_to_copy = (len > offset) ? offset : len;
1762 tmp -= bytes_to_copy;
1763 p -= bytes_to_copy;
1764 offset -= bytes_to_copy;
1765 len -= bytes_to_copy;
1766
1767 memcpy_fromfs(scratch + offset, tmp, bytes_to_copy);
1768
1769 if (offset == 0) {
1770 memcpy_to_target(p, scratch, top - p);
1771 top = p;
1772 offset = TARGET_PAGE_SIZE;
1773 }
1774 }
1775 }
1776 if (p != top) {
1777 memcpy_to_target(p, scratch + offset, top - p);
1778 }
1779 } else {
1780 int remaining = TARGET_PAGE_SIZE - (p % TARGET_PAGE_SIZE);
1781 for (i = 0; i < argc; ++i) {
1782 tmp = argv[i];
1783 if (!tmp) {
1784 fprintf(stderr, "VFS: argc is wrong");
1785 exit(-1);
1786 }
1787 len = strlen(tmp) + 1;
1788 if (len > (stack_limit - p)) {
1789 return 0;
1790 }
1791 while (len) {
1792 int bytes_to_copy = (len > remaining) ? remaining : len;
1793
1794 memcpy_fromfs(scratch + (p - top), tmp, bytes_to_copy);
1795
1796 tmp += bytes_to_copy;
1797 remaining -= bytes_to_copy;
1798 p += bytes_to_copy;
1799 len -= bytes_to_copy;
1800
1801 if (remaining == 0) {
1802 memcpy_to_target(top, scratch, p - top);
1803 top = p;
1804 remaining = TARGET_PAGE_SIZE;
1805 }
1806 }
1807 }
1808 if (p != top) {
1809 memcpy_to_target(top, scratch, p - top);
1810 }
1811 }
1812
1813 return p;
1814 }
1815
1816 /* Older linux kernels provide up to MAX_ARG_PAGES (default: 32) of
1817 * argument/environment space. Newer kernels (>2.6.33) allow more,
1818 * dependent on stack size, but guarantee at least 32 pages for
1819 * backwards compatibility.
1820 */
1821 #define STACK_LOWER_LIMIT (32 * TARGET_PAGE_SIZE)
1822
1823 static abi_ulong setup_arg_pages(struct linux_binprm *bprm,
1824 struct image_info *info)
1825 {
1826 abi_ulong size, error, guard;
1827
1828 size = guest_stack_size;
1829 if (size < STACK_LOWER_LIMIT) {
1830 size = STACK_LOWER_LIMIT;
1831 }
1832 guard = TARGET_PAGE_SIZE;
1833 if (guard < qemu_real_host_page_size) {
1834 guard = qemu_real_host_page_size;
1835 }
1836
1837 error = target_mmap(0, size + guard, PROT_READ | PROT_WRITE,
1838 MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
1839 if (error == -1) {
1840 perror("mmap stack");
1841 exit(-1);
1842 }
1843
1844 /* We reserve one extra page at the top of the stack as guard. */
1845 if (STACK_GROWS_DOWN) {
1846 target_mprotect(error, guard, PROT_NONE);
1847 info->stack_limit = error + guard;
1848 return info->stack_limit + size - sizeof(void *);
1849 } else {
1850 target_mprotect(error + size, guard, PROT_NONE);
1851 info->stack_limit = error + size;
1852 return error;
1853 }
1854 }
1855
1856 /* Map and zero the bss. We need to explicitly zero any fractional pages
1857 after the data section (i.e. bss). */
1858 static void zero_bss(abi_ulong elf_bss, abi_ulong last_bss, int prot)
1859 {
1860 uintptr_t host_start, host_map_start, host_end;
1861
1862 last_bss = TARGET_PAGE_ALIGN(last_bss);
1863
1864 /* ??? There is confusion between qemu_real_host_page_size and
1865 qemu_host_page_size here and elsewhere in target_mmap, which
1866 may lead to the end of the data section mapping from the file
1867 not being mapped. At least there was an explicit test and
1868 comment for that here, suggesting that "the file size must
1869 be known". The comment probably pre-dates the introduction
1870 of the fstat system call in target_mmap which does in fact
1871 find out the size. What isn't clear is if the workaround
1872 here is still actually needed. For now, continue with it,
1873 but merge it with the "normal" mmap that would allocate the bss. */
1874
1875 host_start = (uintptr_t) g2h(elf_bss);
1876 host_end = (uintptr_t) g2h(last_bss);
1877 host_map_start = REAL_HOST_PAGE_ALIGN(host_start);
1878
1879 if (host_map_start < host_end) {
1880 void *p = mmap((void *)host_map_start, host_end - host_map_start,
1881 prot, MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
1882 if (p == MAP_FAILED) {
1883 perror("cannot mmap brk");
1884 exit(-1);
1885 }
1886 }
1887
1888 /* Ensure that the bss page(s) are valid */
1889 if ((page_get_flags(last_bss-1) & prot) != prot) {
1890 page_set_flags(elf_bss & TARGET_PAGE_MASK, last_bss, prot | PAGE_VALID);
1891 }
1892
1893 if (host_start < host_map_start) {
1894 memset((void *)host_start, 0, host_map_start - host_start);
1895 }
1896 }
1897
1898 #ifdef TARGET_ARM
1899 static int elf_is_fdpic(struct elfhdr *exec)
1900 {
1901 return exec->e_ident[EI_OSABI] == ELFOSABI_ARM_FDPIC;
1902 }
1903 #else
1904 /* Default implementation, always false. */
1905 static int elf_is_fdpic(struct elfhdr *exec)
1906 {
1907 return 0;
1908 }
1909 #endif
1910
1911 static abi_ulong loader_build_fdpic_loadmap(struct image_info *info, abi_ulong sp)
1912 {
1913 uint16_t n;
1914 struct elf32_fdpic_loadseg *loadsegs = info->loadsegs;
1915
1916 /* elf32_fdpic_loadseg */
1917 n = info->nsegs;
1918 while (n--) {
1919 sp -= 12;
1920 put_user_u32(loadsegs[n].addr, sp+0);
1921 put_user_u32(loadsegs[n].p_vaddr, sp+4);
1922 put_user_u32(loadsegs[n].p_memsz, sp+8);
1923 }
1924
1925 /* elf32_fdpic_loadmap */
1926 sp -= 4;
1927 put_user_u16(0, sp+0); /* version */
1928 put_user_u16(info->nsegs, sp+2); /* nsegs */
1929
1930 info->personality = PER_LINUX_FDPIC;
1931 info->loadmap_addr = sp;
1932
1933 return sp;
1934 }
1935
1936 static abi_ulong create_elf_tables(abi_ulong p, int argc, int envc,
1937 struct elfhdr *exec,
1938 struct image_info *info,
1939 struct image_info *interp_info)
1940 {
1941 abi_ulong sp;
1942 abi_ulong u_argc, u_argv, u_envp, u_auxv;
1943 int size;
1944 int i;
1945 abi_ulong u_rand_bytes;
1946 uint8_t k_rand_bytes[16];
1947 abi_ulong u_platform;
1948 const char *k_platform;
1949 const int n = sizeof(elf_addr_t);
1950
1951 sp = p;
1952
1953 /* Needs to be before we load the env/argc/... */
1954 if (elf_is_fdpic(exec)) {
1955 /* Need 4 byte alignment for these structs */
1956 sp &= ~3;
1957 sp = loader_build_fdpic_loadmap(info, sp);
1958 info->other_info = interp_info;
1959 if (interp_info) {
1960 interp_info->other_info = info;
1961 sp = loader_build_fdpic_loadmap(interp_info, sp);
1962 info->interpreter_loadmap_addr = interp_info->loadmap_addr;
1963 info->interpreter_pt_dynamic_addr = interp_info->pt_dynamic_addr;
1964 } else {
1965 info->interpreter_loadmap_addr = 0;
1966 info->interpreter_pt_dynamic_addr = 0;
1967 }
1968 }
1969
1970 u_platform = 0;
1971 k_platform = ELF_PLATFORM;
1972 if (k_platform) {
1973 size_t len = strlen(k_platform) + 1;
1974 if (STACK_GROWS_DOWN) {
1975 sp -= (len + n - 1) & ~(n - 1);
1976 u_platform = sp;
1977 /* FIXME - check return value of memcpy_to_target() for failure */
1978 memcpy_to_target(sp, k_platform, len);
1979 } else {
1980 memcpy_to_target(sp, k_platform, len);
1981 u_platform = sp;
1982 sp += len + 1;
1983 }
1984 }
1985
1986 /* Provide 16 byte alignment for the PRNG, and basic alignment for
1987 * the argv and envp pointers.
1988 */
1989 if (STACK_GROWS_DOWN) {
1990 sp = QEMU_ALIGN_DOWN(sp, 16);
1991 } else {
1992 sp = QEMU_ALIGN_UP(sp, 16);
1993 }
1994
1995 /*
1996 * Generate 16 random bytes for userspace PRNG seeding.
1997 */
1998 qemu_guest_getrandom_nofail(k_rand_bytes, sizeof(k_rand_bytes));
1999 if (STACK_GROWS_DOWN) {
2000 sp -= 16;
2001 u_rand_bytes = sp;
2002 /* FIXME - check return value of memcpy_to_target() for failure */
2003 memcpy_to_target(sp, k_rand_bytes, 16);
2004 } else {
2005 memcpy_to_target(sp, k_rand_bytes, 16);
2006 u_rand_bytes = sp;
2007 sp += 16;
2008 }
2009
2010 size = (DLINFO_ITEMS + 1) * 2;
2011 if (k_platform)
2012 size += 2;
2013 #ifdef DLINFO_ARCH_ITEMS
2014 size += DLINFO_ARCH_ITEMS * 2;
2015 #endif
2016 #ifdef ELF_HWCAP2
2017 size += 2;
2018 #endif
2019 info->auxv_len = size * n;
2020
2021 size += envc + argc + 2;
2022 size += 1; /* argc itself */
2023 size *= n;
2024
2025 /* Allocate space and finalize stack alignment for entry now. */
2026 if (STACK_GROWS_DOWN) {
2027 u_argc = QEMU_ALIGN_DOWN(sp - size, STACK_ALIGNMENT);
2028 sp = u_argc;
2029 } else {
2030 u_argc = sp;
2031 sp = QEMU_ALIGN_UP(sp + size, STACK_ALIGNMENT);
2032 }
2033
2034 u_argv = u_argc + n;
2035 u_envp = u_argv + (argc + 1) * n;
2036 u_auxv = u_envp + (envc + 1) * n;
2037 info->saved_auxv = u_auxv;
2038 info->arg_start = u_argv;
2039 info->arg_end = u_argv + argc * n;
2040
2041 /* This is correct because Linux defines
2042 * elf_addr_t as Elf32_Off / Elf64_Off
2043 */
2044 #define NEW_AUX_ENT(id, val) do { \
2045 put_user_ual(id, u_auxv); u_auxv += n; \
2046 put_user_ual(val, u_auxv); u_auxv += n; \
2047 } while(0)
2048
2049 #ifdef ARCH_DLINFO
2050 /*
2051 * ARCH_DLINFO must come first so platform specific code can enforce
2052 * special alignment requirements on the AUXV if necessary (eg. PPC).
2053 */
2054 ARCH_DLINFO;
2055 #endif
2056 /* There must be exactly DLINFO_ITEMS entries here, or the assert
2057 * on info->auxv_len will trigger.
2058 */
2059 NEW_AUX_ENT(AT_PHDR, (abi_ulong)(info->load_addr + exec->e_phoff));
2060 NEW_AUX_ENT(AT_PHENT, (abi_ulong)(sizeof (struct elf_phdr)));
2061 NEW_AUX_ENT(AT_PHNUM, (abi_ulong)(exec->e_phnum));
2062 if ((info->alignment & ~qemu_host_page_mask) != 0) {
2063 /* Target doesn't support host page size alignment */
2064 NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(TARGET_PAGE_SIZE));
2065 } else {
2066 NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(MAX(TARGET_PAGE_SIZE,
2067 qemu_host_page_size)));
2068 }
2069 NEW_AUX_ENT(AT_BASE, (abi_ulong)(interp_info ? interp_info->load_addr : 0));
2070 NEW_AUX_ENT(AT_FLAGS, (abi_ulong)0);
2071 NEW_AUX_ENT(AT_ENTRY, info->entry);
2072 NEW_AUX_ENT(AT_UID, (abi_ulong) getuid());
2073 NEW_AUX_ENT(AT_EUID, (abi_ulong) geteuid());
2074 NEW_AUX_ENT(AT_GID, (abi_ulong) getgid());
2075 NEW_AUX_ENT(AT_EGID, (abi_ulong) getegid());
2076 NEW_AUX_ENT(AT_HWCAP, (abi_ulong) ELF_HWCAP);
2077 NEW_AUX_ENT(AT_CLKTCK, (abi_ulong) sysconf(_SC_CLK_TCK));
2078 NEW_AUX_ENT(AT_RANDOM, (abi_ulong) u_rand_bytes);
2079 NEW_AUX_ENT(AT_SECURE, (abi_ulong) qemu_getauxval(AT_SECURE));
2080 NEW_AUX_ENT(AT_EXECFN, info->file_string);
2081
2082 #ifdef ELF_HWCAP2
2083 NEW_AUX_ENT(AT_HWCAP2, (abi_ulong) ELF_HWCAP2);
2084 #endif
2085
2086 if (u_platform) {
2087 NEW_AUX_ENT(AT_PLATFORM, u_platform);
2088 }
2089 NEW_AUX_ENT (AT_NULL, 0);
2090 #undef NEW_AUX_ENT
2091
2092 /* Check that our initial calculation of the auxv length matches how much
2093 * we actually put into it.
2094 */
2095 assert(info->auxv_len == u_auxv - info->saved_auxv);
2096
2097 put_user_ual(argc, u_argc);
2098
2099 p = info->arg_strings;
2100 for (i = 0; i < argc; ++i) {
2101 put_user_ual(p, u_argv);
2102 u_argv += n;
2103 p += target_strlen(p) + 1;
2104 }
2105 put_user_ual(0, u_argv);
2106
2107 p = info->env_strings;
2108 for (i = 0; i < envc; ++i) {
2109 put_user_ual(p, u_envp);
2110 u_envp += n;
2111 p += target_strlen(p) + 1;
2112 }
2113 put_user_ual(0, u_envp);
2114
2115 return sp;
2116 }
2117
2118 #ifndef ARM_COMMPAGE
2119 #define ARM_COMMPAGE 0
2120 #define init_guest_commpage() true
2121 #endif
2122
2123 static void pgb_fail_in_use(const char *image_name)
2124 {
2125 error_report("%s: requires virtual address space that is in use "
2126 "(omit the -B option or choose a different value)",
2127 image_name);
2128 exit(EXIT_FAILURE);
2129 }
2130
2131 static void pgb_have_guest_base(const char *image_name, abi_ulong guest_loaddr,
2132 abi_ulong guest_hiaddr, long align)
2133 {
2134 const int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE;
2135 void *addr, *test;
2136
2137 if (!QEMU_IS_ALIGNED(guest_base, align)) {
2138 fprintf(stderr, "Requested guest base 0x%lx does not satisfy "
2139 "host minimum alignment (0x%lx)\n",
2140 guest_base, align);
2141 exit(EXIT_FAILURE);
2142 }
2143
2144 /* Sanity check the guest binary. */
2145 if (reserved_va) {
2146 if (guest_hiaddr > reserved_va) {
2147 error_report("%s: requires more than reserved virtual "
2148 "address space (0x%" PRIx64 " > 0x%lx)",
2149 image_name, (uint64_t)guest_hiaddr, reserved_va);
2150 exit(EXIT_FAILURE);
2151 }
2152 } else {
2153 #if HOST_LONG_BITS < TARGET_ABI_BITS
2154 if ((guest_hiaddr - guest_base) > ~(uintptr_t)0) {
2155 error_report("%s: requires more virtual address space "
2156 "than the host can provide (0x%" PRIx64 ")",
2157 image_name, (uint64_t)guest_hiaddr - guest_base);
2158 exit(EXIT_FAILURE);
2159 }
2160 #endif
2161 }
2162
2163 /*
2164 * Expand the allocation to the entire reserved_va.
2165 * Exclude the mmap_min_addr hole.
2166 */
2167 if (reserved_va) {
2168 guest_loaddr = (guest_base >= mmap_min_addr ? 0
2169 : mmap_min_addr - guest_base);
2170 guest_hiaddr = reserved_va;
2171 }
2172
2173 /* Reserve the address space for the binary, or reserved_va. */
2174 test = g2h(guest_loaddr);
2175 addr = mmap(test, guest_hiaddr - guest_loaddr, PROT_NONE, flags, -1, 0);
2176 if (test != addr) {
2177 pgb_fail_in_use(image_name);
2178 }
2179 }
2180
2181 /**
2182 * pgd_find_hole_fallback: potential mmap address
2183 * @guest_size: size of available space
2184 * @brk: location of break
2185 * @align: memory alignment
2186 *
2187 * This is a fallback method for finding a hole in the host address
2188 * space if we don't have the benefit of being able to access
2189 * /proc/self/map. It can potentially take a very long time as we can
2190 * only dumbly iterate up the host address space seeing if the
2191 * allocation would work.
2192 */
2193 static uintptr_t pgd_find_hole_fallback(uintptr_t guest_size, uintptr_t brk,
2194 long align, uintptr_t offset)
2195 {
2196 uintptr_t base;
2197
2198 /* Start (aligned) at the bottom and work our way up */
2199 base = ROUND_UP(mmap_min_addr, align);
2200
2201 while (true) {
2202 uintptr_t align_start, end;
2203 align_start = ROUND_UP(base, align);
2204 end = align_start + guest_size + offset;
2205
2206 /* if brk is anywhere in the range give ourselves some room to grow. */
2207 if (align_start <= brk && brk < end) {
2208 base = brk + (16 * MiB);
2209 continue;
2210 } else if (align_start + guest_size < align_start) {
2211 /* we have run out of space */
2212 return -1;
2213 } else {
2214 int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE |
2215 MAP_FIXED_NOREPLACE;
2216 void * mmap_start = mmap((void *) align_start, guest_size,
2217 PROT_NONE, flags, -1, 0);
2218 if (mmap_start != MAP_FAILED) {
2219 munmap((void *) align_start, guest_size);
2220 if (MAP_FIXED_NOREPLACE != 0 ||
2221 mmap_start == (void *) align_start) {
2222 return (uintptr_t) mmap_start + offset;
2223 }
2224 }
2225 base += qemu_host_page_size;
2226 }
2227 }
2228 }
2229
2230 /* Return value for guest_base, or -1 if no hole found. */
2231 static uintptr_t pgb_find_hole(uintptr_t guest_loaddr, uintptr_t guest_size,
2232 long align, uintptr_t offset)
2233 {
2234 GSList *maps, *iter;
2235 uintptr_t this_start, this_end, next_start, brk;
2236 intptr_t ret = -1;
2237
2238 assert(QEMU_IS_ALIGNED(guest_loaddr, align));
2239
2240 maps = read_self_maps();
2241
2242 /* Read brk after we've read the maps, which will malloc. */
2243 brk = (uintptr_t)sbrk(0);
2244
2245 if (!maps) {
2246 return pgd_find_hole_fallback(guest_size, brk, align, offset);
2247 }
2248
2249 /* The first hole is before the first map entry. */
2250 this_start = mmap_min_addr;
2251
2252 for (iter = maps; iter;
2253 this_start = next_start, iter = g_slist_next(iter)) {
2254 uintptr_t align_start, hole_size;
2255
2256 this_end = ((MapInfo *)iter->data)->start;
2257 next_start = ((MapInfo *)iter->data)->end;
2258 align_start = ROUND_UP(this_start + offset, align);
2259
2260 /* Skip holes that are too small. */
2261 if (align_start >= this_end) {
2262 continue;
2263 }
2264 hole_size = this_end - align_start;
2265 if (hole_size < guest_size) {
2266 continue;
2267 }
2268
2269 /* If this hole contains brk, give ourselves some room to grow. */
2270 if (this_start <= brk && brk < this_end) {
2271 hole_size -= guest_size;
2272 if (sizeof(uintptr_t) == 8 && hole_size >= 1 * GiB) {
2273 align_start += 1 * GiB;
2274 } else if (hole_size >= 16 * MiB) {
2275 align_start += 16 * MiB;
2276 } else {
2277 align_start = (this_end - guest_size) & -align;
2278 if (align_start < this_start) {
2279 continue;
2280 }
2281 }
2282 }
2283
2284 /* Record the lowest successful match. */
2285 if (ret < 0) {
2286 ret = align_start - guest_loaddr;
2287 }
2288 /* If this hole contains the identity map, select it. */
2289 if (align_start <= guest_loaddr &&
2290 guest_loaddr + guest_size <= this_end) {
2291 ret = 0;
2292 }
2293 /* If this hole ends above the identity map, stop looking. */
2294 if (this_end >= guest_loaddr) {
2295 break;
2296 }
2297 }
2298 free_self_maps(maps);
2299
2300 return ret;
2301 }
2302
2303 static void pgb_static(const char *image_name, abi_ulong orig_loaddr,
2304 abi_ulong orig_hiaddr, long align)
2305 {
2306 uintptr_t loaddr = orig_loaddr;
2307 uintptr_t hiaddr = orig_hiaddr;
2308 uintptr_t offset = 0;
2309 uintptr_t addr;
2310
2311 if (hiaddr != orig_hiaddr) {
2312 error_report("%s: requires virtual address space that the "
2313 "host cannot provide (0x%" PRIx64 ")",
2314 image_name, (uint64_t)orig_hiaddr);
2315 exit(EXIT_FAILURE);
2316 }
2317
2318 loaddr &= -align;
2319 if (ARM_COMMPAGE) {
2320 /*
2321 * Extend the allocation to include the commpage.
2322 * For a 64-bit host, this is just 4GiB; for a 32-bit host we
2323 * need to ensure there is space bellow the guest_base so we
2324 * can map the commpage in the place needed when the address
2325 * arithmetic wraps around.
2326 */
2327 if (sizeof(uintptr_t) == 8 || loaddr >= 0x80000000u) {
2328 hiaddr = (uintptr_t) 4 << 30;
2329 } else {
2330 offset = -(ARM_COMMPAGE & -align);
2331 }
2332 }
2333
2334 addr = pgb_find_hole(loaddr, hiaddr - loaddr, align, offset);
2335 if (addr == -1) {
2336 /*
2337 * If ARM_COMMPAGE, there *might* be a non-consecutive allocation
2338 * that can satisfy both. But as the normal arm32 link base address
2339 * is ~32k, and we extend down to include the commpage, making the
2340 * overhead only ~96k, this is unlikely.
2341 */
2342 error_report("%s: Unable to allocate %#zx bytes of "
2343 "virtual address space", image_name,
2344 (size_t)(hiaddr - loaddr));
2345 exit(EXIT_FAILURE);
2346 }
2347
2348 guest_base = addr;
2349 }
2350
2351 static void pgb_dynamic(const char *image_name, long align)
2352 {
2353 /*
2354 * The executable is dynamic and does not require a fixed address.
2355 * All we need is a commpage that satisfies align.
2356 * If we do not need a commpage, leave guest_base == 0.
2357 */
2358 if (ARM_COMMPAGE) {
2359 uintptr_t addr, commpage;
2360
2361 /* 64-bit hosts should have used reserved_va. */
2362 assert(sizeof(uintptr_t) == 4);
2363
2364 /*
2365 * By putting the commpage at the first hole, that puts guest_base
2366 * just above that, and maximises the positive guest addresses.
2367 */
2368 commpage = ARM_COMMPAGE & -align;
2369 addr = pgb_find_hole(commpage, -commpage, align, 0);
2370 assert(addr != -1);
2371 guest_base = addr;
2372 }
2373 }
2374
2375 static void pgb_reserved_va(const char *image_name, abi_ulong guest_loaddr,
2376 abi_ulong guest_hiaddr, long align)
2377 {
2378 int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE;
2379 void *addr, *test;
2380
2381 if (guest_hiaddr > reserved_va) {
2382 error_report("%s: requires more than reserved virtual "
2383 "address space (0x%" PRIx64 " > 0x%lx)",
2384 image_name, (uint64_t)guest_hiaddr, reserved_va);
2385 exit(EXIT_FAILURE);
2386 }
2387
2388 /* Widen the "image" to the entire reserved address space. */
2389 pgb_static(image_name, 0, reserved_va, align);
2390
2391 /* osdep.h defines this as 0 if it's missing */
2392 flags |= MAP_FIXED_NOREPLACE;
2393
2394 /* Reserve the memory on the host. */
2395 assert(guest_base != 0);
2396 test = g2h(0);
2397 addr = mmap(test, reserved_va, PROT_NONE, flags, -1, 0);
2398 if (addr == MAP_FAILED || addr != test) {
2399 error_report("Unable to reserve 0x%lx bytes of virtual address "
2400 "space at %p (%s) for use as guest address space (check your"
2401 "virtual memory ulimit setting, min_mmap_addr or reserve less "
2402 "using -R option)", reserved_va, test, strerror(errno));
2403 exit(EXIT_FAILURE);
2404 }
2405 }
2406
2407 void probe_guest_base(const char *image_name, abi_ulong guest_loaddr,
2408 abi_ulong guest_hiaddr)
2409 {
2410 /* In order to use host shmat, we must be able to honor SHMLBA. */
2411 uintptr_t align = MAX(SHMLBA, qemu_host_page_size);
2412
2413 if (have_guest_base) {
2414 pgb_have_guest_base(image_name, guest_loaddr, guest_hiaddr, align);
2415 } else if (reserved_va) {
2416 pgb_reserved_va(image_name, guest_loaddr, guest_hiaddr, align);
2417 } else if (guest_loaddr) {
2418 pgb_static(image_name, guest_loaddr, guest_hiaddr, align);
2419 } else {
2420 pgb_dynamic(image_name, align);
2421 }
2422
2423 /* Reserve and initialize the commpage. */
2424 if (!init_guest_commpage()) {
2425 /*
2426 * With have_guest_base, the user has selected the address and
2427 * we are trying to work with that. Otherwise, we have selected
2428 * free space and init_guest_commpage must succeeded.
2429 */
2430 assert(have_guest_base);
2431 pgb_fail_in_use(image_name);
2432 }
2433
2434 assert(QEMU_IS_ALIGNED(guest_base, align));
2435 qemu_log_mask(CPU_LOG_PAGE, "Locating guest address space "
2436 "@ 0x%" PRIx64 "\n", (uint64_t)guest_base);
2437 }
2438
2439 enum {
2440 /* The string "GNU\0" as a magic number. */
2441 GNU0_MAGIC = const_le32('G' | 'N' << 8 | 'U' << 16),
2442 NOTE_DATA_SZ = 1 * KiB,
2443 NOTE_NAME_SZ = 4,
2444 ELF_GNU_PROPERTY_ALIGN = ELF_CLASS == ELFCLASS32 ? 4 : 8,
2445 };
2446
2447 /*
2448 * Process a single gnu_property entry.
2449 * Return false for error.
2450 */
2451 static bool parse_elf_property(const uint32_t *data, int *off, int datasz,
2452 struct image_info *info, bool have_prev_type,
2453 uint32_t *prev_type, Error **errp)
2454 {
2455 uint32_t pr_type, pr_datasz, step;
2456
2457 if (*off > datasz || !QEMU_IS_ALIGNED(*off, ELF_GNU_PROPERTY_ALIGN)) {
2458 goto error_data;
2459 }
2460 datasz -= *off;
2461 data += *off / sizeof(uint32_t);
2462
2463 if (datasz < 2 * sizeof(uint32_t)) {
2464 goto error_data;
2465 }
2466 pr_type = data[0];
2467 pr_datasz = data[1];
2468 data += 2;
2469 datasz -= 2 * sizeof(uint32_t);
2470 step = ROUND_UP(pr_datasz, ELF_GNU_PROPERTY_ALIGN);
2471 if (step > datasz) {
2472 goto error_data;
2473 }
2474
2475 /* Properties are supposed to be unique and sorted on pr_type. */
2476 if (have_prev_type && pr_type <= *prev_type) {
2477 if (pr_type == *prev_type) {
2478 error_setg(errp, "Duplicate property in PT_GNU_PROPERTY");
2479 } else {
2480 error_setg(errp, "Unsorted property in PT_GNU_PROPERTY");
2481 }
2482 return false;
2483 }
2484 *prev_type = pr_type;
2485
2486 if (!arch_parse_elf_property(pr_type, pr_datasz, data, info, errp)) {
2487 return false;
2488 }
2489
2490 *off += 2 * sizeof(uint32_t) + step;
2491 return true;
2492
2493 error_data:
2494 error_setg(errp, "Ill-formed property in PT_GNU_PROPERTY");
2495 return false;
2496 }
2497
2498 /* Process NT_GNU_PROPERTY_TYPE_0. */
2499 static bool parse_elf_properties(int image_fd,
2500 struct image_info *info,
2501 const struct elf_phdr *phdr,
2502 char bprm_buf[BPRM_BUF_SIZE],
2503 Error **errp)
2504 {
2505 union {
2506 struct elf_note nhdr;
2507 uint32_t data[NOTE_DATA_SZ / sizeof(uint32_t)];
2508 } note;
2509
2510 int n, off, datasz;
2511 bool have_prev_type;
2512 uint32_t prev_type;
2513
2514 /* Unless the arch requires properties, ignore them. */
2515 if (!ARCH_USE_GNU_PROPERTY) {
2516 return true;
2517 }
2518
2519 /* If the properties are crazy large, that's too bad. */
2520 n = phdr->p_filesz;
2521 if (n > sizeof(note)) {
2522 error_setg(errp, "PT_GNU_PROPERTY too large");
2523 return false;
2524 }
2525 if (n < sizeof(note.nhdr)) {
2526 error_setg(errp, "PT_GNU_PROPERTY too small");
2527 return false;
2528 }
2529
2530 if (phdr->p_offset + n <= BPRM_BUF_SIZE) {
2531 memcpy(&note, bprm_buf + phdr->p_offset, n);
2532 } else {
2533 ssize_t len = pread(image_fd, &note, n, phdr->p_offset);
2534 if (len != n) {
2535 error_setg_errno(errp, errno, "Error reading file header");
2536 return false;
2537 }
2538 }
2539
2540 /*
2541 * The contents of a valid PT_GNU_PROPERTY is a sequence
2542 * of uint32_t -- swap them all now.
2543 */
2544 #ifdef BSWAP_NEEDED
2545 for (int i = 0; i < n / 4; i++) {
2546 bswap32s(note.data + i);
2547 }
2548 #endif
2549
2550 /*
2551 * Note that nhdr is 3 words, and that the "name" described by namesz
2552 * immediately follows nhdr and is thus at the 4th word. Further, all
2553 * of the inputs to the kernel's round_up are multiples of 4.
2554 */
2555 if (note.nhdr.n_type != NT_GNU_PROPERTY_TYPE_0 ||
2556 note.nhdr.n_namesz != NOTE_NAME_SZ ||
2557 note.data[3] != GNU0_MAGIC) {
2558 error_setg(errp, "Invalid note in PT_GNU_PROPERTY");
2559 return false;
2560 }
2561 off = sizeof(note.nhdr) + NOTE_NAME_SZ;
2562
2563 datasz = note.nhdr.n_descsz + off;
2564 if (datasz > n) {
2565 error_setg(errp, "Invalid note size in PT_GNU_PROPERTY");
2566 return false;
2567 }
2568
2569 have_prev_type = false;
2570 prev_type = 0;
2571 while (1) {
2572 if (off == datasz) {
2573 return true; /* end, exit ok */
2574 }
2575 if (!parse_elf_property(note.data, &off, datasz, info,
2576 have_prev_type, &prev_type, errp)) {
2577 return false;
2578 }
2579 have_prev_type = true;
2580 }
2581 }
2582
2583 /* Load an ELF image into the address space.
2584
2585 IMAGE_NAME is the filename of the image, to use in error messages.
2586 IMAGE_FD is the open file descriptor for the image.
2587
2588 BPRM_BUF is a copy of the beginning of the file; this of course
2589 contains the elf file header at offset 0. It is assumed that this
2590 buffer is sufficiently aligned to present no problems to the host
2591 in accessing data at aligned offsets within the buffer.
2592
2593 On return: INFO values will be filled in, as necessary or available. */
2594
2595 static void load_elf_image(const char *image_name, int image_fd,
2596 struct image_info *info, char **pinterp_name,
2597 char bprm_buf[BPRM_BUF_SIZE])
2598 {
2599 struct elfhdr *ehdr = (struct elfhdr *)bprm_buf;
2600 struct elf_phdr *phdr;
2601 abi_ulong load_addr, load_bias, loaddr, hiaddr, error;
2602 int i, retval, prot_exec;
2603 Error *err = NULL;
2604
2605 /* First of all, some simple consistency checks */
2606 if (!elf_check_ident(ehdr)) {
2607 error_setg(&err, "Invalid ELF image for this architecture");
2608 goto exit_errmsg;
2609 }
2610 bswap_ehdr(ehdr);
2611 if (!elf_check_ehdr(ehdr)) {
2612 error_setg(&err, "Invalid ELF image for this architecture");
2613 goto exit_errmsg;
2614 }
2615
2616 i = ehdr->e_phnum * sizeof(struct elf_phdr);
2617 if (ehdr->e_phoff + i <= BPRM_BUF_SIZE) {
2618 phdr = (struct elf_phdr *)(bprm_buf + ehdr->e_phoff);
2619 } else {
2620 phdr = (struct elf_phdr *) alloca(i);
2621 retval = pread(image_fd, phdr, i, ehdr->e_phoff);
2622 if (retval != i) {
2623 goto exit_read;
2624 }
2625 }
2626 bswap_phdr(phdr, ehdr->e_phnum);
2627
2628 info->nsegs = 0;
2629 info->pt_dynamic_addr = 0;
2630
2631 mmap_lock();
2632
2633 /*
2634 * Find the maximum size of the image and allocate an appropriate
2635 * amount of memory to handle that. Locate the interpreter, if any.
2636 */
2637 loaddr = -1, hiaddr = 0;
2638 info->alignment = 0;
2639 for (i = 0; i < ehdr->e_phnum; ++i) {
2640 struct elf_phdr *eppnt = phdr + i;
2641 if (eppnt->p_type == PT_LOAD) {
2642 abi_ulong a = eppnt->p_vaddr - eppnt->p_offset;
2643 if (a < loaddr) {
2644 loaddr = a;
2645 }
2646 a = eppnt->p_vaddr + eppnt->p_memsz;
2647 if (a > hiaddr) {
2648 hiaddr = a;
2649 }
2650 ++info->nsegs;
2651 info->alignment |= eppnt->p_align;
2652 } else if (eppnt->p_type == PT_INTERP && pinterp_name) {
2653 g_autofree char *interp_name = NULL;
2654
2655 if (*pinterp_name) {
2656 error_setg(&err, "Multiple PT_INTERP entries");
2657 goto exit_errmsg;
2658 }
2659
2660 interp_name = g_malloc(eppnt->p_filesz);
2661
2662 if (eppnt->p_offset + eppnt->p_filesz <= BPRM_BUF_SIZE) {
2663 memcpy(interp_name, bprm_buf + eppnt->p_offset,
2664 eppnt->p_filesz);
2665 } else {
2666 retval = pread(image_fd, interp_name, eppnt->p_filesz,
2667 eppnt->p_offset);
2668 if (retval != eppnt->p_filesz) {
2669 goto exit_read;
2670 }
2671 }
2672 if (interp_name[eppnt->p_filesz - 1] != 0) {
2673 error_setg(&err, "Invalid PT_INTERP entry");
2674 goto exit_errmsg;
2675 }
2676 *pinterp_name = g_steal_pointer(&interp_name);
2677 } else if (eppnt->p_type == PT_GNU_PROPERTY) {
2678 if (!parse_elf_properties(image_fd, info, eppnt, bprm_buf, &err)) {
2679 goto exit_errmsg;
2680 }
2681 }
2682 }
2683
2684 if (pinterp_name != NULL) {
2685 /*
2686 * This is the main executable.
2687 *
2688 * Reserve extra space for brk.
2689 * We hold on to this space while placing the interpreter
2690 * and the stack, lest they be placed immediately after
2691 * the data segment and block allocation from the brk.
2692 *
2693 * 16MB is chosen as "large enough" without being so large
2694 * as to allow the result to not fit with a 32-bit guest on
2695 * a 32-bit host.
2696 */
2697 info->reserve_brk = 16 * MiB;
2698 hiaddr += info->reserve_brk;
2699
2700 if (ehdr->e_type == ET_EXEC) {
2701 /*
2702 * Make sure that the low address does not conflict with
2703 * MMAP_MIN_ADDR or the QEMU application itself.
2704 */
2705 probe_guest_base(image_name, loaddr, hiaddr);
2706 } else {
2707 /*
2708 * The binary is dynamic, but we still need to
2709 * select guest_base. In this case we pass a size.
2710 */
2711 probe_guest_base(image_name, 0, hiaddr - loaddr);
2712 }
2713 }
2714
2715 /*
2716 * Reserve address space for all of this.
2717 *
2718 * In the case of ET_EXEC, we supply MAP_FIXED so that we get
2719 * exactly the address range that is required.
2720 *
2721 * Otherwise this is ET_DYN, and we are searching for a location
2722 * that can hold the memory space required. If the image is
2723 * pre-linked, LOADDR will be non-zero, and the kernel should
2724 * honor that address if it happens to be free.
2725 *
2726 * In both cases, we will overwrite pages in this range with mappings
2727 * from the executable.
2728 */
2729 load_addr = target_mmap(loaddr, hiaddr - loaddr, PROT_NONE,
2730 MAP_PRIVATE | MAP_ANON | MAP_NORESERVE |
2731 (ehdr->e_type == ET_EXEC ? MAP_FIXED : 0),
2732 -1, 0);
2733 if (load_addr == -1) {
2734 goto exit_mmap;
2735 }
2736 load_bias = load_addr - loaddr;
2737
2738 if (elf_is_fdpic(ehdr)) {
2739 struct elf32_fdpic_loadseg *loadsegs = info->loadsegs =
2740 g_malloc(sizeof(*loadsegs) * info->nsegs);
2741
2742 for (i = 0; i < ehdr->e_phnum; ++i) {
2743 switch (phdr[i].p_type) {
2744 case PT_DYNAMIC:
2745 info->pt_dynamic_addr = phdr[i].p_vaddr + load_bias;
2746 break;
2747 case PT_LOAD:
2748 loadsegs->addr = phdr[i].p_vaddr + load_bias;
2749 loadsegs->p_vaddr = phdr[i].p_vaddr;
2750 loadsegs->p_memsz = phdr[i].p_memsz;
2751 ++loadsegs;
2752 break;
2753 }
2754 }
2755 }
2756
2757 info->load_bias = load_bias;
2758 info->code_offset = load_bias;
2759 info->data_offset = load_bias;
2760 info->load_addr = load_addr;
2761 info->entry = ehdr->e_entry + load_bias;
2762 info->start_code = -1;
2763 info->end_code = 0;
2764 info->start_data = -1;
2765 info->end_data = 0;
2766 info->brk = 0;
2767 info->elf_flags = ehdr->e_flags;
2768
2769 prot_exec = PROT_EXEC;
2770 #ifdef TARGET_AARCH64
2771 /*
2772 * If the BTI feature is present, this indicates that the executable
2773 * pages of the startup binary should be mapped with PROT_BTI, so that
2774 * branch targets are enforced.
2775 *
2776 * The startup binary is either the interpreter or the static executable.
2777 * The interpreter is responsible for all pages of a dynamic executable.
2778 *
2779 * Elf notes are backward compatible to older cpus.
2780 * Do not enable BTI unless it is supported.
2781 */
2782 if ((info->note_flags & GNU_PROPERTY_AARCH64_FEATURE_1_BTI)
2783 && (pinterp_name == NULL || *pinterp_name == 0)
2784 && cpu_isar_feature(aa64_bti, ARM_CPU(thread_cpu))) {
2785 prot_exec |= TARGET_PROT_BTI;
2786 }
2787 #endif
2788
2789 for (i = 0; i < ehdr->e_phnum; i++) {
2790 struct elf_phdr *eppnt = phdr + i;
2791 if (eppnt->p_type == PT_LOAD) {
2792 abi_ulong vaddr, vaddr_po, vaddr_ps, vaddr_ef, vaddr_em, vaddr_len;
2793 int elf_prot = 0;
2794
2795 if (eppnt->p_flags & PF_R) {
2796 elf_prot |= PROT_READ;
2797 }
2798 if (eppnt->p_flags & PF_W) {
2799 elf_prot |= PROT_WRITE;
2800 }
2801 if (eppnt->p_flags & PF_X) {
2802 elf_prot |= prot_exec;
2803 }
2804
2805 vaddr = load_bias + eppnt->p_vaddr;
2806 vaddr_po = TARGET_ELF_PAGEOFFSET(vaddr);
2807 vaddr_ps = TARGET_ELF_PAGESTART(vaddr);
2808
2809 vaddr_ef = vaddr + eppnt->p_filesz;
2810 vaddr_em = vaddr + eppnt->p_memsz;
2811
2812 /*
2813 * Some segments may be completely empty, with a non-zero p_memsz
2814 * but no backing file segment.
2815 */
2816 if (eppnt->p_filesz != 0) {
2817 vaddr_len = TARGET_ELF_PAGELENGTH(eppnt->p_filesz + vaddr_po);
2818 error = target_mmap(vaddr_ps, vaddr_len, elf_prot,
2819 MAP_PRIVATE | MAP_FIXED,
2820 image_fd, eppnt->p_offset - vaddr_po);
2821
2822 if (error == -1) {
2823 goto exit_mmap;
2824 }
2825
2826 /*
2827 * If the load segment requests extra zeros (e.g. bss), map it.
2828 */
2829 if (eppnt->p_filesz < eppnt->p_memsz) {
2830 zero_bss(vaddr_ef, vaddr_em, elf_prot);
2831 }
2832 } else if (eppnt->p_memsz != 0) {
2833 vaddr_len = TARGET_ELF_PAGELENGTH(eppnt->p_memsz + vaddr_po);
2834 error = target_mmap(vaddr_ps, vaddr_len, elf_prot,
2835 MAP_PRIVATE | MAP_FIXED | MAP_ANONYMOUS,
2836 -1, 0);
2837
2838 if (error == -1) {
2839 goto exit_mmap;
2840 }
2841 }
2842
2843 /* Find the full program boundaries. */
2844 if (elf_prot & PROT_EXEC) {
2845 if (vaddr < info->start_code) {
2846 info->start_code = vaddr;
2847 }
2848 if (vaddr_ef > info->end_code) {
2849 info->end_code = vaddr_ef;
2850 }
2851 }
2852 if (elf_prot & PROT_WRITE) {
2853 if (vaddr < info->start_data) {
2854 info->start_data = vaddr;
2855 }
2856 if (vaddr_ef > info->end_data) {
2857 info->end_data = vaddr_ef;
2858 }
2859 }
2860 if (vaddr_em > info->brk) {
2861 info->brk = vaddr_em;
2862 }
2863 #ifdef TARGET_MIPS
2864 } else if (eppnt->p_type == PT_MIPS_ABIFLAGS) {
2865 Mips_elf_abiflags_v0 abiflags;
2866 if (eppnt->p_filesz < sizeof(Mips_elf_abiflags_v0)) {
2867 error_setg(&err, "Invalid PT_MIPS_ABIFLAGS entry");
2868 goto exit_errmsg;
2869 }
2870 if (eppnt->p_offset + eppnt->p_filesz <= BPRM_BUF_SIZE) {
2871 memcpy(&abiflags, bprm_buf + eppnt->p_offset,
2872 sizeof(Mips_elf_abiflags_v0));
2873 } else {
2874 retval = pread(image_fd, &abiflags, sizeof(Mips_elf_abiflags_v0),
2875 eppnt->p_offset);
2876 if (retval != sizeof(Mips_elf_abiflags_v0)) {
2877 goto exit_read;
2878 }
2879 }
2880 bswap_mips_abiflags(&abiflags);
2881 info->fp_abi = abiflags.fp_abi;
2882 #endif
2883 }
2884 }
2885
2886 if (info->end_data == 0) {
2887 info->start_data = info->end_code;
2888 info->end_data = info->end_code;
2889 }
2890
2891 if (qemu_log_enabled()) {
2892 load_symbols(ehdr, image_fd, load_bias);
2893 }
2894
2895 mmap_unlock();
2896
2897 close(image_fd);
2898 return;
2899
2900 exit_read:
2901 if (retval >= 0) {
2902 error_setg(&err, "Incomplete read of file header");
2903 } else {
2904 error_setg_errno(&err, errno, "Error reading file header");
2905 }
2906 goto exit_errmsg;
2907 exit_mmap:
2908 error_setg_errno(&err, errno, "Error mapping file");
2909 goto exit_errmsg;
2910 exit_errmsg:
2911 error_reportf_err(err, "%s: ", image_name);
2912 exit(-1);
2913 }
2914
2915 static void load_elf_interp(const char *filename, struct image_info *info,
2916 char bprm_buf[BPRM_BUF_SIZE])
2917 {
2918 int fd, retval;
2919 Error *err = NULL;
2920
2921 fd = open(path(filename), O_RDONLY);
2922 if (fd < 0) {
2923 error_setg_file_open(&err, errno, filename);
2924 error_report_err(err);
2925 exit(-1);
2926 }
2927
2928 retval = read(fd, bprm_buf, BPRM_BUF_SIZE);
2929 if (retval < 0) {
2930 error_setg_errno(&err, errno, "Error reading file header");
2931 error_reportf_err(err, "%s: ", filename);
2932 exit(-1);
2933 }
2934
2935 if (retval < BPRM_BUF_SIZE) {
2936 memset(bprm_buf + retval, 0, BPRM_BUF_SIZE - retval);
2937 }
2938
2939 load_elf_image(filename, fd, info, NULL, bprm_buf);
2940 }
2941
2942 static int symfind(const void *s0, const void *s1)
2943 {
2944 target_ulong addr = *(target_ulong *)s0;
2945 struct elf_sym *sym = (struct elf_sym *)s1;
2946 int result = 0;
2947 if (addr < sym->st_value) {
2948 result = -1;
2949 } else if (addr >= sym->st_value + sym->st_size) {
2950 result = 1;
2951 }
2952 return result;
2953 }
2954
2955 static const char *lookup_symbolxx(struct syminfo *s, target_ulong orig_addr)
2956 {
2957 #if ELF_CLASS == ELFCLASS32
2958 struct elf_sym *syms = s->disas_symtab.elf32;
2959 #else
2960 struct elf_sym *syms = s->disas_symtab.elf64;
2961 #endif
2962
2963 // binary search
2964 struct elf_sym *sym;
2965
2966 sym = bsearch(&orig_addr, syms, s->disas_num_syms, sizeof(*syms), symfind);
2967 if (sym != NULL) {
2968 return s->disas_strtab + sym->st_name;
2969 }
2970
2971 return "";
2972 }
2973
2974 /* FIXME: This should use elf_ops.h */
2975 static int symcmp(const void *s0, const void *s1)
2976 {
2977 struct elf_sym *sym0 = (struct elf_sym *)s0;
2978 struct elf_sym *sym1 = (struct elf_sym *)s1;
2979 return (sym0->st_value < sym1->st_value)
2980 ? -1
2981 : ((sym0->st_value > sym1->st_value) ? 1 : 0);
2982 }
2983
2984 /* Best attempt to load symbols from this ELF object. */
2985 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias)
2986 {
2987 int i, shnum, nsyms, sym_idx = 0, str_idx = 0;
2988 uint64_t segsz;
2989 struct elf_shdr *shdr;
2990 char *strings = NULL;
2991 struct syminfo *s = NULL;
2992 struct elf_sym *new_syms, *syms = NULL;
2993
2994 shnum = hdr->e_shnum;
2995 i = shnum * sizeof(struct elf_shdr);
2996 shdr = (struct elf_shdr *)alloca(i);
2997 if (pread(fd, shdr, i, hdr->e_shoff) != i) {
2998 return;
2999 }
3000
3001 bswap_shdr(shdr, shnum);
3002 for (i = 0; i < shnum; ++i) {
3003 if (shdr[i].sh_type == SHT_SYMTAB) {
3004 sym_idx = i;
3005 str_idx = shdr[i].sh_link;
3006 goto found;
3007 }
3008 }
3009
3010 /* There will be no symbol table if the file was stripped. */
3011 return;
3012
3013 found:
3014 /* Now know where the strtab and symtab are. Snarf them. */
3015 s = g_try_new(struct syminfo, 1);
3016 if (!s) {
3017 goto give_up;
3018 }
3019
3020 segsz = shdr[str_idx].sh_size;
3021 s->disas_strtab = strings = g_try_malloc(segsz);
3022 if (!strings ||
3023 pread(fd, strings, segsz, shdr[str_idx].sh_offset) != segsz) {
3024 goto give_up;
3025 }
3026
3027 segsz = shdr[sym_idx].sh_size;
3028 syms = g_try_malloc(segsz);
3029 if (!syms || pread(fd, syms, segsz, shdr[sym_idx].sh_offset) != segsz) {
3030 goto give_up;
3031 }
3032
3033 if (segsz / sizeof(struct elf_sym) > INT_MAX) {
3034 /* Implausibly large symbol table: give up rather than ploughing
3035 * on with the number of symbols calculation overflowing
3036 */
3037 goto give_up;
3038 }
3039 nsyms = segsz / sizeof(struct elf_sym);
3040 for (i = 0; i < nsyms; ) {
3041 bswap_sym(syms + i);
3042 /* Throw away entries which we do not need. */
3043 if (syms[i].st_shndx == SHN_UNDEF
3044 || syms[i].st_shndx >= SHN_LORESERVE
3045 || ELF_ST_TYPE(syms[i].st_info) != STT_FUNC) {
3046 if (i < --nsyms) {
3047 syms[i] = syms[nsyms];
3048 }
3049 } else {
3050 #if defined(TARGET_ARM) || defined (TARGET_MIPS)
3051 /* The bottom address bit marks a Thumb or MIPS16 symbol. */
3052 syms[i].st_value &= ~(target_ulong)1;
3053 #endif
3054 syms[i].st_value += load_bias;
3055 i++;
3056 }
3057 }
3058
3059 /* No "useful" symbol. */
3060 if (nsyms == 0) {
3061 goto give_up;
3062 }
3063
3064 /* Attempt to free the storage associated with the local symbols
3065 that we threw away. Whether or not this has any effect on the
3066 memory allocation depends on the malloc implementation and how
3067 many symbols we managed to discard. */
3068 new_syms = g_try_renew(struct elf_sym, syms, nsyms);
3069 if (new_syms == NULL) {
3070 goto give_up;
3071 }
3072 syms = new_syms;
3073
3074 qsort(syms, nsyms, sizeof(*syms), symcmp);
3075
3076 s->disas_num_syms = nsyms;
3077 #if ELF_CLASS == ELFCLASS32
3078 s->disas_symtab.elf32 = syms;
3079 #else
3080 s->disas_symtab.elf64 = syms;
3081 #endif
3082 s->lookup_symbol = lookup_symbolxx;
3083 s->next = syminfos;
3084 syminfos = s;
3085
3086 return;
3087
3088 give_up:
3089 g_free(s);
3090 g_free(strings);
3091 g_free(syms);
3092 }
3093
3094 uint32_t get_elf_eflags(int fd)
3095 {
3096 struct elfhdr ehdr;
3097 off_t offset;
3098 int ret;
3099
3100 /* Read ELF header */
3101 offset = lseek(fd, 0, SEEK_SET);
3102 if (offset == (off_t) -1) {
3103 return 0;
3104 }
3105 ret = read(fd, &ehdr, sizeof(ehdr));
3106 if (ret < sizeof(ehdr)) {
3107 return 0;
3108 }
3109 offset = lseek(fd, offset, SEEK_SET);
3110 if (offset == (off_t) -1) {
3111 return 0;
3112 }
3113
3114 /* Check ELF signature */
3115 if (!elf_check_ident(&ehdr)) {
3116 return 0;
3117 }
3118
3119 /* check header */
3120 bswap_ehdr(&ehdr);
3121 if (!elf_check_ehdr(&ehdr)) {
3122 return 0;
3123 }
3124
3125 /* return architecture id */
3126 return ehdr.e_flags;
3127 }
3128
3129 int load_elf_binary(struct linux_binprm *bprm, struct image_info *info)
3130 {
3131 struct image_info interp_info;
3132 struct elfhdr elf_ex;
3133 char *elf_interpreter = NULL;
3134 char *scratch;
3135
3136 memset(&interp_info, 0, sizeof(interp_info));
3137 #ifdef TARGET_MIPS
3138 interp_info.fp_abi = MIPS_ABI_FP_UNKNOWN;
3139 #endif
3140
3141 info->start_mmap = (abi_ulong)ELF_START_MMAP;
3142
3143 load_elf_image(bprm->filename, bprm->fd, info,
3144 &elf_interpreter, bprm->buf);
3145
3146 /* ??? We need a copy of the elf header for passing to create_elf_tables.
3147 If we do nothing, we'll have overwritten this when we re-use bprm->buf
3148 when we load the interpreter. */
3149 elf_ex = *(struct elfhdr *)bprm->buf;
3150
3151 /* Do this so that we can load the interpreter, if need be. We will
3152 change some of these later */
3153 bprm->p = setup_arg_pages(bprm, info);
3154
3155 scratch = g_new0(char, TARGET_PAGE_SIZE);
3156 if (STACK_GROWS_DOWN) {
3157 bprm->p = copy_elf_strings(1, &bprm->filename, scratch,
3158 bprm->p, info->stack_limit);
3159 info->file_string = bprm->p;
3160 bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch,
3161 bprm->p, info->stack_limit);
3162 info->env_strings = bprm->p;
3163 bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch,
3164 bprm->p, info->stack_limit);
3165 info->arg_strings = bprm->p;
3166 } else {
3167 info->arg_strings = bprm->p;
3168 bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch,
3169 bprm->p, info->stack_limit);
3170 info->env_strings = bprm->p;
3171 bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch,
3172 bprm->p, info->stack_limit);
3173 info->file_string = bprm->p;
3174 bprm->p = copy_elf_strings(1, &bprm->filename, scratch,
3175 bprm->p, info->stack_limit);
3176 }
3177
3178 g_free(scratch);
3179
3180 if (!bprm->p) {
3181 fprintf(stderr, "%s: %s\n", bprm->filename, strerror(E2BIG));
3182 exit(-1);
3183 }
3184
3185 if (elf_interpreter) {
3186 load_elf_interp(elf_interpreter, &interp_info, bprm->buf);
3187
3188 /* If the program interpreter is one of these two, then assume
3189 an iBCS2 image. Otherwise assume a native linux image. */
3190
3191 if (strcmp(elf_interpreter, "/usr/lib/libc.so.1") == 0
3192 || strcmp(elf_interpreter, "/usr/lib/ld.so.1") == 0) {
3193 info->personality = PER_SVR4;
3194
3195 /* Why this, you ask??? Well SVr4 maps page 0 as read-only,
3196 and some applications "depend" upon this behavior. Since
3197 we do not have the power to recompile these, we emulate
3198 the SVr4 behavior. Sigh. */
3199 target_mmap(0, qemu_host_page_size, PROT_READ | PROT_EXEC,
3200 MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
3201 }
3202 #ifdef TARGET_MIPS
3203 info->interp_fp_abi = interp_info.fp_abi;
3204 #endif
3205 }
3206
3207 bprm->p = create_elf_tables(bprm->p, bprm->argc, bprm->envc, &elf_ex,
3208 info, (elf_interpreter ? &interp_info : NULL));
3209 info->start_stack = bprm->p;
3210
3211 /* If we have an interpreter, set that as the program's entry point.
3212 Copy the load_bias as well, to help PPC64 interpret the entry
3213 point as a function descriptor. Do this after creating elf tables
3214 so that we copy the original program entry point into the AUXV. */
3215 if (elf_interpreter) {
3216 info->load_bias = interp_info.load_bias;
3217 info->entry = interp_info.entry;
3218 g_free(elf_interpreter);
3219 }
3220
3221 #ifdef USE_ELF_CORE_DUMP
3222 bprm->core_dump = &elf_core_dump;
3223 #endif
3224
3225 /*
3226 * If we reserved extra space for brk, release it now.
3227 * The implementation of do_brk in syscalls.c expects to be able
3228 * to mmap pages in this space.
3229 */
3230 if (info->reserve_brk) {
3231 abi_ulong start_brk = HOST_PAGE_ALIGN(info->brk);
3232 abi_ulong end_brk = HOST_PAGE_ALIGN(info->brk + info->reserve_brk);
3233 target_munmap(start_brk, end_brk - start_brk);
3234 }
3235
3236 return 0;
3237 }
3238
3239 #ifdef USE_ELF_CORE_DUMP
3240 /*
3241 * Definitions to generate Intel SVR4-like core files.
3242 * These mostly have the same names as the SVR4 types with "target_elf_"
3243 * tacked on the front to prevent clashes with linux definitions,
3244 * and the typedef forms have been avoided. This is mostly like
3245 * the SVR4 structure, but more Linuxy, with things that Linux does
3246 * not support and which gdb doesn't really use excluded.
3247 *
3248 * Fields we don't dump (their contents is zero) in linux-user qemu
3249 * are marked with XXX.
3250 *
3251 * Core dump code is copied from linux kernel (fs/binfmt_elf.c).
3252 *
3253 * Porting ELF coredump for target is (quite) simple process. First you
3254 * define USE_ELF_CORE_DUMP in target ELF code (where init_thread() for
3255 * the target resides):
3256 *
3257 * #define USE_ELF_CORE_DUMP
3258 *
3259 * Next you define type of register set used for dumping. ELF specification
3260 * says that it needs to be array of elf_greg_t that has size of ELF_NREG.
3261 *
3262 * typedef <target_regtype> target_elf_greg_t;
3263 * #define ELF_NREG <number of registers>
3264 * typedef taret_elf_greg_t target_elf_gregset_t[ELF_NREG];
3265 *
3266 * Last step is to implement target specific function that copies registers
3267 * from given cpu into just specified register set. Prototype is:
3268 *
3269 * static void elf_core_copy_regs(taret_elf_gregset_t *regs,
3270 * const CPUArchState *env);
3271 *
3272 * Parameters:
3273 * regs - copy register values into here (allocated and zeroed by caller)
3274 * env - copy registers from here
3275 *
3276 * Example for ARM target is provided in this file.
3277 */
3278
3279 /* An ELF note in memory */
3280 struct memelfnote {
3281 const char *name;
3282 size_t namesz;
3283 size_t namesz_rounded;
3284 int type;
3285 size_t datasz;
3286 size_t datasz_rounded;
3287 void *data;
3288 size_t notesz;
3289 };
3290
3291 struct target_elf_siginfo {
3292 abi_int si_signo; /* signal number */
3293 abi_int si_code; /* extra code */
3294 abi_int si_errno; /* errno */
3295 };
3296
3297 struct target_elf_prstatus {
3298 struct target_elf_siginfo pr_info; /* Info associated with signal */
3299 abi_short pr_cursig; /* Current signal */
3300 abi_ulong pr_sigpend; /* XXX */
3301 abi_ulong pr_sighold; /* XXX */
3302 target_pid_t pr_pid;
3303 target_pid_t pr_ppid;
3304 target_pid_t pr_pgrp;
3305 target_pid_t pr_sid;
3306 struct target_timeval pr_utime; /* XXX User time */
3307 struct target_timeval pr_stime; /* XXX System time */
3308 struct target_timeval pr_cutime; /* XXX Cumulative user time */
3309 struct target_timeval pr_cstime; /* XXX Cumulative system time */
3310 target_elf_gregset_t pr_reg; /* GP registers */
3311 abi_int pr_fpvalid; /* XXX */
3312 };
3313
3314 #define ELF_PRARGSZ (80) /* Number of chars for args */
3315
3316 struct target_elf_prpsinfo {
3317 char pr_state; /* numeric process state */
3318 char pr_sname; /* char for pr_state */
3319 char pr_zomb; /* zombie */
3320 char pr_nice; /* nice val */
3321 abi_ulong pr_flag; /* flags */
3322 target_uid_t pr_uid;
3323 target_gid_t pr_gid;
3324 target_pid_t pr_pid, pr_ppid, pr_pgrp, pr_sid;
3325 /* Lots missing */
3326 char pr_fname[16] QEMU_NONSTRING; /* filename of executable */
3327 char pr_psargs[ELF_PRARGSZ]; /* initial part of arg list */
3328 };
3329
3330 /* Here is the structure in which status of each thread is captured. */
3331 struct elf_thread_status {
3332 QTAILQ_ENTRY(elf_thread_status) ets_link;
3333 struct target_elf_prstatus prstatus; /* NT_PRSTATUS */
3334 #if 0
3335 elf_fpregset_t fpu; /* NT_PRFPREG */
3336 struct task_struct *thread;
3337 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */
3338 #endif
3339 struct memelfnote notes[1];
3340 int num_notes;
3341 };
3342
3343 struct elf_note_info {
3344 struct memelfnote *notes;
3345 struct target_elf_prstatus *prstatus; /* NT_PRSTATUS */
3346 struct target_elf_prpsinfo *psinfo; /* NT_PRPSINFO */
3347
3348 QTAILQ_HEAD(, elf_thread_status) thread_list;
3349 #if 0
3350 /*
3351 * Current version of ELF coredump doesn't support
3352 * dumping fp regs etc.
3353 */
3354 elf_fpregset_t *fpu;
3355 elf_fpxregset_t *xfpu;
3356 int thread_status_size;
3357 #endif
3358 int notes_size;
3359 int numnote;
3360 };
3361
3362 struct vm_area_struct {
3363 target_ulong vma_start; /* start vaddr of memory region */
3364 target_ulong vma_end; /* end vaddr of memory region */
3365 abi_ulong vma_flags; /* protection etc. flags for the region */
3366 QTAILQ_ENTRY(vm_area_struct) vma_link;
3367 };
3368
3369 struct mm_struct {
3370 QTAILQ_HEAD(, vm_area_struct) mm_mmap;
3371 int mm_count; /* number of mappings */
3372 };
3373
3374 static struct mm_struct *vma_init(void);
3375 static void vma_delete(struct mm_struct *);
3376 static int vma_add_mapping(struct mm_struct *, target_ulong,
3377 target_ulong, abi_ulong);
3378 static int vma_get_mapping_count(const struct mm_struct *);
3379 static struct vm_area_struct *vma_first(const struct mm_struct *);
3380 static struct vm_area_struct *vma_next(struct vm_area_struct *);
3381 static abi_ulong vma_dump_size(const struct vm_area_struct *);
3382 static int vma_walker(void *priv, target_ulong start, target_ulong end,
3383 unsigned long flags);
3384
3385 static void fill_elf_header(struct elfhdr *, int, uint16_t, uint32_t);
3386 static void fill_note(struct memelfnote *, const char *, int,
3387 unsigned int, void *);
3388 static void fill_prstatus(struct target_elf_prstatus *, const TaskState *, int);
3389 static int fill_psinfo(struct target_elf_prpsinfo *, const TaskState *);
3390 static void fill_auxv_note(struct memelfnote *, const TaskState *);
3391 static void fill_elf_note_phdr(struct elf_phdr *, int, off_t);
3392 static size_t note_size(const struct memelfnote *);
3393 static void free_note_info(struct elf_note_info *);
3394 static int fill_note_info(struct elf_note_info *, long, const CPUArchState *);
3395 static void fill_thread_info(struct elf_note_info *, const CPUArchState *);
3396 static int core_dump_filename(const TaskState *, char *, size_t);
3397
3398 static int dump_write(int, const void *, size_t);
3399 static int write_note(struct memelfnote *, int);
3400 static int write_note_info(struct elf_note_info *, int);
3401
3402 #ifdef BSWAP_NEEDED
3403 static void bswap_prstatus(struct target_elf_prstatus *prstatus)
3404 {
3405 prstatus->pr_info.si_signo = tswap32(prstatus->pr_info.si_signo);
3406 prstatus->pr_info.si_code = tswap32(prstatus->pr_info.si_code);
3407 prstatus->pr_info.si_errno = tswap32(prstatus->pr_info.si_errno);
3408 prstatus->pr_cursig = tswap16(prstatus->pr_cursig);
3409 prstatus->pr_sigpend = tswapal(prstatus->pr_sigpend);
3410 prstatus->pr_sighold = tswapal(prstatus->pr_sighold);
3411 prstatus->pr_pid = tswap32(prstatus->pr_pid);
3412 prstatus->pr_ppid = tswap32(prstatus->pr_ppid);
3413 prstatus->pr_pgrp = tswap32(prstatus->pr_pgrp);
3414 prstatus->pr_sid = tswap32(prstatus->pr_sid);
3415 /* cpu times are not filled, so we skip them */
3416 /* regs should be in correct format already */
3417 prstatus->pr_fpvalid = tswap32(prstatus->pr_fpvalid);
3418 }
3419
3420 static void bswap_psinfo(struct target_elf_prpsinfo *psinfo)
3421 {
3422 psinfo->pr_flag = tswapal(psinfo->pr_flag);
3423 psinfo->pr_uid = tswap16(psinfo->pr_uid);
3424 psinfo->pr_gid = tswap16(psinfo->pr_gid);
3425 psinfo->pr_pid = tswap32(psinfo->pr_pid);
3426 psinfo->pr_ppid = tswap32(psinfo->pr_ppid);
3427 psinfo->pr_pgrp = tswap32(psinfo->pr_pgrp);
3428 psinfo->pr_sid = tswap32(psinfo->pr_sid);
3429 }
3430
3431 static void bswap_note(struct elf_note *en)
3432 {
3433 bswap32s(&en->n_namesz);
3434 bswap32s(&en->n_descsz);
3435 bswap32s(&en->n_type);
3436 }
3437 #else
3438 static inline void bswap_prstatus(struct target_elf_prstatus *p) { }
3439 static inline void bswap_psinfo(struct target_elf_prpsinfo *p) {}
3440 static inline void bswap_note(struct elf_note *en) { }
3441 #endif /* BSWAP_NEEDED */
3442
3443 /*
3444 * Minimal support for linux memory regions. These are needed
3445 * when we are finding out what memory exactly belongs to
3446 * emulated process. No locks needed here, as long as
3447 * thread that received the signal is stopped.
3448 */
3449
3450 static struct mm_struct *vma_init(void)
3451 {
3452 struct mm_struct *mm;
3453
3454 if ((mm = g_malloc(sizeof (*mm))) == NULL)
3455 return (NULL);
3456
3457 mm->mm_count = 0;
3458 QTAILQ_INIT(&mm->mm_mmap);
3459
3460 return (mm);
3461 }
3462
3463 static void vma_delete(struct mm_struct *mm)
3464 {
3465 struct vm_area_struct *vma;
3466
3467 while ((vma = vma_first(mm)) != NULL) {
3468 QTAILQ_REMOVE(&mm->mm_mmap, vma, vma_link);
3469 g_free(vma);
3470 }
3471 g_free(mm);
3472 }
3473
3474 static int vma_add_mapping(struct mm_struct *mm, target_ulong start,
3475 target_ulong end, abi_ulong flags)
3476 {
3477 struct vm_area_struct *vma;
3478
3479 if ((vma = g_malloc0(sizeof (*vma))) == NULL)
3480 return (-1);
3481
3482 vma->vma_start = start;
3483 vma->vma_end = end;
3484 vma->vma_flags = flags;
3485
3486 QTAILQ_INSERT_TAIL(&mm->mm_mmap, vma, vma_link);
3487 mm->mm_count++;
3488
3489 return (0);
3490 }
3491
3492 static struct vm_area_struct *vma_first(const struct mm_struct *mm)
3493 {
3494 return (QTAILQ_FIRST(&mm->mm_mmap));
3495 }
3496
3497 static struct vm_area_struct *vma_next(struct vm_area_struct *vma)
3498 {
3499 return (QTAILQ_NEXT(vma, vma_link));
3500 }
3501
3502 static int vma_get_mapping_count(const struct mm_struct *mm)
3503 {
3504 return (mm->mm_count);
3505 }
3506
3507 /*
3508 * Calculate file (dump) size of given memory region.
3509 */
3510 static abi_ulong vma_dump_size(const struct vm_area_struct *vma)
3511 {
3512 /* if we cannot even read the first page, skip it */
3513 if (!access_ok(VERIFY_READ, vma->vma_start, TARGET_PAGE_SIZE))
3514 return (0);
3515
3516 /*
3517 * Usually we don't dump executable pages as they contain
3518 * non-writable code that debugger can read directly from
3519 * target library etc. However, thread stacks are marked
3520 * also executable so we read in first page of given region
3521 * and check whether it contains elf header. If there is
3522 * no elf header, we dump it.
3523 */
3524 if (vma->vma_flags & PROT_EXEC) {
3525 char page[TARGET_PAGE_SIZE];
3526
3527 if (copy_from_user(page, vma->vma_start, sizeof (page))) {
3528 return 0;
3529 }
3530 if ((page[EI_MAG0] == ELFMAG0) &&
3531 (page[EI_MAG1] == ELFMAG1) &&
3532 (page[EI_MAG2] == ELFMAG2) &&
3533 (page[EI_MAG3] == ELFMAG3)) {
3534 /*
3535 * Mappings are possibly from ELF binary. Don't dump
3536 * them.
3537 */
3538 return (0);
3539 }
3540 }
3541
3542 return (vma->vma_end - vma->vma_start);
3543 }
3544
3545 static int vma_walker(void *priv, target_ulong start, target_ulong end,
3546 unsigned long flags)
3547 {
3548 struct mm_struct *mm = (struct mm_struct *)priv;
3549
3550 vma_add_mapping(mm, start, end, flags);
3551 return (0);
3552 }
3553
3554 static void fill_note(struct memelfnote *note, const char *name, int type,
3555 unsigned int sz, void *data)
3556 {
3557 unsigned int namesz;
3558
3559 namesz = strlen(name) + 1;
3560 note->name = name;
3561 note->namesz = namesz;
3562 note->namesz_rounded = roundup(namesz, sizeof (int32_t));
3563 note->type = type;
3564 note->datasz = sz;
3565 note->datasz_rounded = roundup(sz, sizeof (int32_t));
3566
3567 note->data = data;
3568
3569 /*
3570 * We calculate rounded up note size here as specified by
3571 * ELF document.
3572 */
3573 note->notesz = sizeof (struct elf_note) +
3574 note->namesz_rounded + note->datasz_rounded;
3575 }
3576
3577 static void fill_elf_header(struct elfhdr *elf, int segs, uint16_t machine,
3578 uint32_t flags)
3579 {
3580 (void) memset(elf, 0, sizeof(*elf));
3581
3582 (void) memcpy(elf->e_ident, ELFMAG, SELFMAG);
3583 elf->e_ident[EI_CLASS] = ELF_CLASS;
3584 elf->e_ident[EI_DATA] = ELF_DATA;
3585 elf->e_ident[EI_VERSION] = EV_CURRENT;
3586 elf->e_ident[EI_OSABI] = ELF_OSABI;
3587
3588 elf->e_type = ET_CORE;
3589 elf->e_machine = machine;
3590 elf->e_version = EV_CURRENT;
3591 elf->e_phoff = sizeof(struct elfhdr);
3592 elf->e_flags = flags;
3593 elf->e_ehsize = sizeof(struct elfhdr);
3594 elf->e_phentsize = sizeof(struct elf_phdr);
3595 elf->e_phnum = segs;
3596
3597 bswap_ehdr(elf);
3598 }
3599
3600 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, off_t offset)
3601 {
3602 phdr->p_type = PT_NOTE;
3603 phdr->p_offset = offset;
3604 phdr->p_vaddr = 0;
3605 phdr->p_paddr = 0;
3606 phdr->p_filesz = sz;
3607 phdr->p_memsz = 0;
3608 phdr->p_flags = 0;
3609 phdr->p_align = 0;
3610
3611 bswap_phdr(phdr, 1);
3612 }
3613
3614 static size_t note_size(const struct memelfnote *note)
3615 {
3616 return (note->notesz);
3617 }
3618
3619 static void fill_prstatus(struct target_elf_prstatus *prstatus,
3620 const TaskState *ts, int signr)
3621 {
3622 (void) memset(prstatus, 0, sizeof (*prstatus));
3623 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
3624 prstatus->pr_pid = ts->ts_tid;
3625 prstatus->pr_ppid = getppid();
3626 prstatus->pr_pgrp = getpgrp();
3627 prstatus->pr_sid = getsid(0);
3628
3629 bswap_prstatus(prstatus);
3630 }
3631
3632 static int fill_psinfo(struct target_elf_prpsinfo *psinfo, const TaskState *ts)
3633 {
3634 char *base_filename;
3635 unsigned int i, len;
3636
3637 (void) memset(psinfo, 0, sizeof (*psinfo));
3638
3639 len = ts->info->arg_end - ts->info->arg_start;
3640 if (len >= ELF_PRARGSZ)
3641 len = ELF_PRARGSZ - 1;
3642 if (copy_from_user(&psinfo->pr_psargs, ts->info->arg_start, len))
3643 return -EFAULT;
3644 for (i = 0; i < len; i++)
3645 if (psinfo->pr_psargs[i] == 0)
3646 psinfo->pr_psargs[i] = ' ';
3647 psinfo->pr_psargs[len] = 0;
3648
3649 psinfo->pr_pid = getpid();
3650 psinfo->pr_ppid = getppid();
3651 psinfo->pr_pgrp = getpgrp();
3652 psinfo->pr_sid = getsid(0);
3653 psinfo->pr_uid = getuid();
3654 psinfo->pr_gid = getgid();
3655
3656 base_filename = g_path_get_basename(ts->bprm->filename);
3657 /*
3658 * Using strncpy here is fine: at max-length,
3659 * this field is not NUL-terminated.
3660 */
3661 (void) strncpy(psinfo->pr_fname, base_filename,
3662 sizeof(psinfo->pr_fname));
3663
3664 g_free(base_filename);
3665 bswap_psinfo(psinfo);
3666 return (0);
3667 }
3668
3669 static void fill_auxv_note(struct memelfnote *note, const TaskState *ts)
3670 {
3671 elf_addr_t auxv = (elf_addr_t)ts->info->saved_auxv;
3672 elf_addr_t orig_auxv = auxv;
3673 void *ptr;
3674 int len = ts->info->auxv_len;
3675
3676 /*
3677 * Auxiliary vector is stored in target process stack. It contains
3678 * {type, value} pairs that we need to dump into note. This is not
3679 * strictly necessary but we do it here for sake of completeness.
3680 */
3681
3682 /* read in whole auxv vector and copy it to memelfnote */
3683 ptr = lock_user(VERIFY_READ, orig_auxv, len, 0);
3684 if (ptr != NULL) {
3685 fill_note(note, "CORE", NT_AUXV, len, ptr);
3686 unlock_user(ptr, auxv, len);
3687 }
3688 }
3689
3690 /*
3691 * Constructs name of coredump file. We have following convention
3692 * for the name:
3693 * qemu_<basename-of-target-binary>_<date>-<time>_<pid>.core
3694 *
3695 * Returns 0 in case of success, -1 otherwise (errno is set).
3696 */
3697 static int core_dump_filename(const TaskState *ts, char *buf,
3698 size_t bufsize)
3699 {
3700 char timestamp[64];
3701 char *base_filename = NULL;
3702 struct timeval tv;
3703 struct tm tm;
3704
3705 assert(bufsize >= PATH_MAX);
3706
3707 if (gettimeofday(&tv, NULL) < 0) {
3708 (void) fprintf(stderr, "unable to get current timestamp: %s",
3709 strerror(errno));
3710 return (-1);
3711 }
3712
3713 base_filename = g_path_get_basename(ts->bprm->filename);
3714 (void) strftime(timestamp, sizeof (timestamp), "%Y%m%d-%H%M%S",
3715 localtime_r(&tv.tv_sec, &tm));
3716 (void) snprintf(buf, bufsize, "qemu_%s_%s_%d.core",
3717 base_filename, timestamp, (int)getpid());
3718 g_free(base_filename);
3719
3720 return (0);
3721 }
3722
3723 static int dump_write(int fd, const void *ptr, size_t size)
3724 {
3725 const char *bufp = (const char *)ptr;
3726 ssize_t bytes_written, bytes_left;
3727 struct rlimit dumpsize;
3728 off_t pos;
3729
3730 bytes_written = 0;
3731 getrlimit(RLIMIT_CORE, &dumpsize);
3732 if ((pos = lseek(fd, 0, SEEK_CUR))==-1) {
3733 if (errno == ESPIPE) { /* not a seekable stream */
3734 bytes_left = size;
3735 } else {
3736 return pos;
3737 }
3738 } else {
3739 if (dumpsize.rlim_cur <= pos) {
3740 return -1;
3741 } else if (dumpsize.rlim_cur == RLIM_INFINITY) {
3742 bytes_left = size;
3743 } else {
3744 size_t limit_left=dumpsize.rlim_cur - pos;
3745 bytes_left = limit_left >= size ? size : limit_left ;
3746 }
3747 }
3748
3749 /*
3750 * In normal conditions, single write(2) should do but
3751 * in case of socket etc. this mechanism is more portable.
3752 */
3753 do {
3754 bytes_written = write(fd, bufp, bytes_left);
3755 if (bytes_written < 0) {
3756 if (errno == EINTR)
3757 continue;
3758 return (-1);
3759 } else if (bytes_written == 0) { /* eof */
3760 return (-1);
3761 }
3762 bufp += bytes_written;
3763 bytes_left -= bytes_written;
3764 } while (bytes_left > 0);
3765
3766 return (0);
3767 }
3768
3769 static int write_note(struct memelfnote *men, int fd)
3770 {
3771 struct elf_note en;
3772
3773 en.n_namesz = men->namesz;
3774 en.n_type = men->type;
3775 en.n_descsz = men->datasz;
3776
3777 bswap_note(&en);
3778
3779 if (dump_write(fd, &en, sizeof(en)) != 0)
3780 return (-1);
3781 if (dump_write(fd, men->name, men->namesz_rounded) != 0)
3782 return (-1);
3783 if (dump_write(fd, men->data, men->datasz_rounded) != 0)
3784 return (-1);
3785
3786 return (0);
3787 }
3788
3789 static void fill_thread_info(struct elf_note_info *info, const CPUArchState *env)
3790 {
3791 CPUState *cpu = env_cpu((CPUArchState *)env);
3792 TaskState *ts = (TaskState *)cpu->opaque;
3793 struct elf_thread_status *ets;
3794
3795 ets = g_malloc0(sizeof (*ets));
3796 ets->num_notes = 1; /* only prstatus is dumped */
3797 fill_prstatus(&ets->prstatus, ts, 0);
3798 elf_core_copy_regs(&ets->prstatus.pr_reg, env);
3799 fill_note(&ets->notes[0], "CORE", NT_PRSTATUS, sizeof (ets->prstatus),
3800 &ets->prstatus);
3801
3802 QTAILQ_INSERT_TAIL(&info->thread_list, ets, ets_link);
3803
3804 info->notes_size += note_size(&ets->notes[0]);
3805 }
3806
3807 static void init_note_info(struct elf_note_info *info)
3808 {
3809 /* Initialize the elf_note_info structure so that it is at
3810 * least safe to call free_note_info() on it. Must be
3811 * called before calling fill_note_info().
3812 */
3813 memset(info, 0, sizeof (*info));
3814 QTAILQ_INIT(&info->thread_list);
3815 }
3816
3817 static int fill_note_info(struct elf_note_info *info,
3818 long signr, const CPUArchState *env)
3819 {
3820 #define NUMNOTES 3
3821 CPUState *cpu = env_cpu((CPUArchState *)env);
3822 TaskState *ts = (TaskState *)cpu->opaque;
3823 int i;
3824
3825 info->notes = g_new0(struct memelfnote, NUMNOTES);
3826 if (info->notes == NULL)
3827 return (-ENOMEM);
3828 info->prstatus = g_malloc0(sizeof (*info->prstatus));
3829 if (info->prstatus == NULL)
3830 return (-ENOMEM);
3831 info->psinfo = g_malloc0(sizeof (*info->psinfo));
3832 if (info->prstatus == NULL)
3833 return (-ENOMEM);
3834
3835 /*
3836 * First fill in status (and registers) of current thread
3837 * including process info & aux vector.
3838 */
3839 fill_prstatus(info->prstatus, ts, signr);
3840 elf_core_copy_regs(&info->prstatus->pr_reg, env);
3841 fill_note(&info->notes[0], "CORE", NT_PRSTATUS,
3842 sizeof (*info->prstatus), info->prstatus);
3843 fill_psinfo(info->psinfo, ts);
3844 fill_note(&info->notes[1], "CORE", NT_PRPSINFO,
3845 sizeof (*info->psinfo), info->psinfo);
3846 fill_auxv_note(&info->notes[2], ts);
3847 info->numnote = 3;
3848
3849 info->notes_size = 0;
3850 for (i = 0; i < info->numnote; i++)
3851 info->notes_size += note_size(&info->notes[i]);
3852
3853 /* read and fill status of all threads */
3854 cpu_list_lock();
3855 CPU_FOREACH(cpu) {
3856 if (cpu == thread_cpu) {
3857 continue;
3858 }
3859 fill_thread_info(info, (CPUArchState *)cpu->env_ptr);
3860 }
3861 cpu_list_unlock();
3862
3863 return (0);
3864 }
3865
3866 static void free_note_info(struct elf_note_info *info)
3867 {
3868 struct elf_thread_status *ets;
3869
3870 while (!QTAILQ_EMPTY(&info->thread_list)) {
3871 ets = QTAILQ_FIRST(&info->thread_list);
3872 QTAILQ_REMOVE(&info->thread_list, ets, ets_link);
3873 g_free(ets);
3874 }
3875
3876 g_free(info->prstatus);
3877 g_free(info->psinfo);
3878 g_free(info->notes);
3879 }
3880
3881 static int write_note_info(struct elf_note_info *info, int fd)
3882 {
3883 struct elf_thread_status *ets;
3884 int i, error = 0;
3885
3886 /* write prstatus, psinfo and auxv for current thread */
3887 for (i = 0; i < info->numnote; i++)
3888 if ((error = write_note(&info->notes[i], fd)) != 0)
3889 return (error);
3890
3891 /* write prstatus for each thread */
3892 QTAILQ_FOREACH(ets, &info->thread_list, ets_link) {
3893 if ((error = write_note(&ets->notes[0], fd)) != 0)
3894 return (error);
3895 }
3896
3897 return (0);
3898 }
3899
3900 /*
3901 * Write out ELF coredump.
3902 *
3903 * See documentation of ELF object file format in:
3904 * http://www.caldera.com/developers/devspecs/gabi41.pdf
3905 *
3906 * Coredump format in linux is following:
3907 *
3908 * 0 +----------------------+ \
3909 * | ELF header | ET_CORE |
3910 * +----------------------+ |
3911 * | ELF program headers | |--- headers
3912 * | - NOTE section | |
3913 * | - PT_LOAD sections | |
3914 * +----------------------+ /
3915 * | NOTEs: |
3916 * | - NT_PRSTATUS |
3917 * | - NT_PRSINFO |
3918 * | - NT_AUXV |
3919 * +----------------------+ <-- aligned to target page
3920 * | Process memory dump |
3921 * : :
3922 * . .
3923 * : :
3924 * | |
3925 * +----------------------+
3926 *
3927 * NT_PRSTATUS -> struct elf_prstatus (per thread)
3928 * NT_PRSINFO -> struct elf_prpsinfo
3929 * NT_AUXV is array of { type, value } pairs (see fill_auxv_note()).
3930 *
3931 * Format follows System V format as close as possible. Current
3932 * version limitations are as follows:
3933 * - no floating point registers are dumped
3934 *
3935 * Function returns 0 in case of success, negative errno otherwise.
3936 *
3937 * TODO: make this work also during runtime: it should be
3938 * possible to force coredump from running process and then
3939 * continue processing. For example qemu could set up SIGUSR2
3940 * handler (provided that target process haven't registered
3941 * handler for that) that does the dump when signal is received.
3942 */
3943 static int elf_core_dump(int signr, const CPUArchState *env)
3944 {
3945 const CPUState *cpu = env_cpu((CPUArchState *)env);
3946 const TaskState *ts = (const TaskState *)cpu->opaque;
3947 struct vm_area_struct *vma = NULL;
3948 char corefile[PATH_MAX];
3949 struct elf_note_info info;
3950 struct elfhdr elf;
3951 struct elf_phdr phdr;
3952 struct rlimit dumpsize;
3953 struct mm_struct *mm = NULL;
3954 off_t offset = 0, data_offset = 0;
3955 int segs = 0;
3956 int fd = -1;
3957
3958 init_note_info(&info);
3959
3960 errno = 0;
3961 getrlimit(RLIMIT_CORE, &dumpsize);
3962 if (dumpsize.rlim_cur == 0)
3963 return 0;
3964
3965 if (core_dump_filename(ts, corefile, sizeof (corefile)) < 0)
3966 return (-errno);
3967
3968 if ((fd = open(corefile, O_WRONLY | O_CREAT,
3969 S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH)) < 0)
3970 return (-errno);
3971
3972 /*
3973 * Walk through target process memory mappings and
3974 * set up structure containing this information. After
3975 * this point vma_xxx functions can be used.
3976 */
3977 if ((mm = vma_init()) == NULL)
3978 goto out;
3979
3980 walk_memory_regions(mm, vma_walker);
3981 segs = vma_get_mapping_count(mm);
3982
3983 /*
3984 * Construct valid coredump ELF header. We also
3985 * add one more segment for notes.
3986 */
3987 fill_elf_header(&elf, segs + 1, ELF_MACHINE, 0);
3988 if (dump_write(fd, &elf, sizeof (elf)) != 0)
3989 goto out;
3990
3991 /* fill in the in-memory version of notes */
3992 if (fill_note_info(&info, signr, env) < 0)
3993 goto out;
3994
3995 offset += sizeof (elf); /* elf header */
3996 offset += (segs + 1) * sizeof (struct elf_phdr); /* program headers */
3997
3998 /* write out notes program header */
3999 fill_elf_note_phdr(&phdr, info.notes_size, offset);
4000
4001 offset += info.notes_size;
4002 if (dump_write(fd, &phdr, sizeof (phdr)) != 0)
4003 goto out;
4004
4005 /*
4006 * ELF specification wants data to start at page boundary so
4007 * we align it here.
4008 */
4009 data_offset = offset = roundup(offset, ELF_EXEC_PAGESIZE);
4010
4011 /*
4012 * Write program headers for memory regions mapped in
4013 * the target process.
4014 */
4015 for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) {
4016 (void) memset(&phdr, 0, sizeof (phdr));
4017
4018 phdr.p_type = PT_LOAD;
4019 phdr.p_offset = offset;
4020 phdr.p_vaddr = vma->vma_start;
4021 phdr.p_paddr = 0;
4022 phdr.p_filesz = vma_dump_size(vma);
4023 offset += phdr.p_filesz;
4024 phdr.p_memsz = vma->vma_end - vma->vma_start;
4025 phdr.p_flags = vma->vma_flags & PROT_READ ? PF_R : 0;
4026 if (vma->vma_flags & PROT_WRITE)
4027 phdr.p_flags |= PF_W;
4028 if (vma->vma_flags & PROT_EXEC)
4029 phdr.p_flags |= PF_X;
4030 phdr.p_align = ELF_EXEC_PAGESIZE;
4031
4032 bswap_phdr(&phdr, 1);
4033 if (dump_write(fd, &phdr, sizeof(phdr)) != 0) {
4034 goto out;
4035 }
4036 }
4037
4038 /*
4039 * Next we write notes just after program headers. No
4040 * alignment needed here.
4041 */
4042 if (write_note_info(&info, fd) < 0)
4043 goto out;
4044
4045 /* align data to page boundary */
4046 if (lseek(fd, data_offset, SEEK_SET) != data_offset)
4047 goto out;
4048
4049 /*
4050 * Finally we can dump process memory into corefile as well.
4051 */
4052 for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) {
4053 abi_ulong addr;
4054 abi_ulong end;
4055
4056 end = vma->vma_start + vma_dump_size(vma);
4057
4058 for (addr = vma->vma_start; addr < end;
4059 addr += TARGET_PAGE_SIZE) {
4060 char page[TARGET_PAGE_SIZE];
4061 int error;
4062
4063 /*
4064 * Read in page from target process memory and
4065 * write it to coredump file.
4066 */
4067 error = copy_from_user(page, addr, sizeof (page));
4068 if (error != 0) {
4069 (void) fprintf(stderr, "unable to dump " TARGET_ABI_FMT_lx "\n",
4070 addr);
4071 errno = -error;
4072 goto out;
4073 }
4074 if (dump_write(fd, page, TARGET_PAGE_SIZE) < 0)
4075 goto out;
4076 }
4077 }
4078
4079 out:
4080 free_note_info(&info);
4081 if (mm != NULL)
4082 vma_delete(mm);
4083 (void) close(fd);
4084
4085 if (errno != 0)
4086 return (-errno);
4087 return (0);
4088 }
4089 #endif /* USE_ELF_CORE_DUMP */
4090
4091 void do_init_thread(struct target_pt_regs *regs, struct image_info *infop)
4092 {
4093 init_thread(regs, infop);
4094 }
AR7 emulation for QEMU