?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
accel/tcg: Assert that tb->size != 0 after translation
[qemu/ar7.git] / linux-user / elfload.c
blob0e832b2649f6bb87c9f620b4f535b3d5b8338cb3
1 /* This is the Linux kernel elf-loading code, ported into user space */
2 #include "qemu/osdep.h"
3 #include <sys/param.h>
4
5 #include <sys/resource.h>
6 #include <sys/shm.h>
7
8 #include "qemu.h"
9 #include "disas/disas.h"
10 #include "qemu/bitops.h"
11 #include "qemu/path.h"
12 #include "qemu/queue.h"
13 #include "qemu/guest-random.h"
14 #include "qemu/units.h"
15 #include "qemu/selfmap.h"
16 #include "qapi/error.h"
17
18 #ifdef _ARCH_PPC64
19 #undef ARCH_DLINFO
20 #undef ELF_PLATFORM
21 #undef ELF_HWCAP
22 #undef ELF_HWCAP2
23 #undef ELF_CLASS
24 #undef ELF_DATA
25 #undef ELF_ARCH
26 #endif
27
28 #define ELF_OSABI ELFOSABI_SYSV
29
30 /* from personality.h */
31
32 /*
33 * Flags for bug emulation.
34 *
35 * These occupy the top three bytes.
36 */
37 enum {
38 ADDR_NO_RANDOMIZE = 0x0040000, /* disable randomization of VA space */
39 FDPIC_FUNCPTRS = 0x0080000, /* userspace function ptrs point to
40 descriptors (signal handling) */
41 MMAP_PAGE_ZERO = 0x0100000,
42 ADDR_COMPAT_LAYOUT = 0x0200000,
43 READ_IMPLIES_EXEC = 0x0400000,
44 ADDR_LIMIT_32BIT = 0x0800000,
45 SHORT_INODE = 0x1000000,
46 WHOLE_SECONDS = 0x2000000,
47 STICKY_TIMEOUTS = 0x4000000,
48 ADDR_LIMIT_3GB = 0x8000000,
49 };
50
51 /*
52 * Personality types.
53 *
54 * These go in the low byte. Avoid using the top bit, it will
55 * conflict with error returns.
56 */
57 enum {
58 PER_LINUX = 0x0000,
59 PER_LINUX_32BIT = 0x0000 | ADDR_LIMIT_32BIT,
60 PER_LINUX_FDPIC = 0x0000 | FDPIC_FUNCPTRS,
61 PER_SVR4 = 0x0001 | STICKY_TIMEOUTS | MMAP_PAGE_ZERO,
62 PER_SVR3 = 0x0002 | STICKY_TIMEOUTS | SHORT_INODE,
63 PER_SCOSVR3 = 0x0003 | STICKY_TIMEOUTS | WHOLE_SECONDS | SHORT_INODE,
64 PER_OSR5 = 0x0003 | STICKY_TIMEOUTS | WHOLE_SECONDS,
65 PER_WYSEV386 = 0x0004 | STICKY_TIMEOUTS | SHORT_INODE,
66 PER_ISCR4 = 0x0005 | STICKY_TIMEOUTS,
67 PER_BSD = 0x0006,
68 PER_SUNOS = 0x0006 | STICKY_TIMEOUTS,
69 PER_XENIX = 0x0007 | STICKY_TIMEOUTS | SHORT_INODE,
70 PER_LINUX32 = 0x0008,
71 PER_LINUX32_3GB = 0x0008 | ADDR_LIMIT_3GB,
72 PER_IRIX32 = 0x0009 | STICKY_TIMEOUTS,/* IRIX5 32-bit */
73 PER_IRIXN32 = 0x000a | STICKY_TIMEOUTS,/* IRIX6 new 32-bit */
74 PER_IRIX64 = 0x000b | STICKY_TIMEOUTS,/* IRIX6 64-bit */
75 PER_RISCOS = 0x000c,
76 PER_SOLARIS = 0x000d | STICKY_TIMEOUTS,
77 PER_UW7 = 0x000e | STICKY_TIMEOUTS | MMAP_PAGE_ZERO,
78 PER_OSF4 = 0x000f, /* OSF/1 v4 */
79 PER_HPUX = 0x0010,
80 PER_MASK = 0x00ff,
81 };
82
83 /*
84 * Return the base personality without flags.
85 */
86 #define personality(pers) (pers & PER_MASK)
87
88 int info_is_fdpic(struct image_info *info)
89 {
90 return info->personality == PER_LINUX_FDPIC;
91 }
92
93 /* this flag is uneffective under linux too, should be deleted */
94 #ifndef MAP_DENYWRITE
95 #define MAP_DENYWRITE 0
96 #endif
97
98 /* should probably go in elf.h */
99 #ifndef ELIBBAD
100 #define ELIBBAD 80
101 #endif
102
103 #ifdef TARGET_WORDS_BIGENDIAN
104 #define ELF_DATA ELFDATA2MSB
105 #else
106 #define ELF_DATA ELFDATA2LSB
107 #endif
108
109 #ifdef TARGET_ABI_MIPSN32
110 typedef abi_ullong target_elf_greg_t;
111 #define tswapreg(ptr) tswap64(ptr)
112 #else
113 typedef abi_ulong target_elf_greg_t;
114 #define tswapreg(ptr) tswapal(ptr)
115 #endif
116
117 #ifdef USE_UID16
118 typedef abi_ushort target_uid_t;
119 typedef abi_ushort target_gid_t;
120 #else
121 typedef abi_uint target_uid_t;
122 typedef abi_uint target_gid_t;
123 #endif
124 typedef abi_int target_pid_t;
125
126 #ifdef TARGET_I386
127
128 #define ELF_PLATFORM get_elf_platform()
129
130 static const char *get_elf_platform(void)
131 {
132 static char elf_platform[] = "i386";
133 int family = object_property_get_int(OBJECT(thread_cpu), "family", NULL);
134 if (family > 6)
135 family = 6;
136 if (family >= 3)
137 elf_platform[1] = '0' + family;
138 return elf_platform;
139 }
140
141 #define ELF_HWCAP get_elf_hwcap()
142
143 static uint32_t get_elf_hwcap(void)
144 {
145 X86CPU *cpu = X86_CPU(thread_cpu);
146
147 return cpu->env.features[FEAT_1_EDX];
148 }
149
150 #ifdef TARGET_X86_64
151 #define ELF_START_MMAP 0x2aaaaab000ULL
152
153 #define ELF_CLASS ELFCLASS64
154 #define ELF_ARCH EM_X86_64
155
156 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop)
157 {
158 regs->rax = 0;
159 regs->rsp = infop->start_stack;
160 regs->rip = infop->entry;
161 }
162
163 #define ELF_NREG 27
164 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
165
166 /*
167 * Note that ELF_NREG should be 29 as there should be place for
168 * TRAPNO and ERR "registers" as well but linux doesn't dump
169 * those.
170 *
171 * See linux kernel: arch/x86/include/asm/elf.h
172 */
173 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUX86State *env)
174 {
175 (*regs)[0] = env->regs[15];
176 (*regs)[1] = env->regs[14];
177 (*regs)[2] = env->regs[13];
178 (*regs)[3] = env->regs[12];
179 (*regs)[4] = env->regs[R_EBP];
180 (*regs)[5] = env->regs[R_EBX];
181 (*regs)[6] = env->regs[11];
182 (*regs)[7] = env->regs[10];
183 (*regs)[8] = env->regs[9];
184 (*regs)[9] = env->regs[8];
185 (*regs)[10] = env->regs[R_EAX];
186 (*regs)[11] = env->regs[R_ECX];
187 (*regs)[12] = env->regs[R_EDX];
188 (*regs)[13] = env->regs[R_ESI];
189 (*regs)[14] = env->regs[R_EDI];
190 (*regs)[15] = env->regs[R_EAX]; /* XXX */
191 (*regs)[16] = env->eip;
192 (*regs)[17] = env->segs[R_CS].selector & 0xffff;
193 (*regs)[18] = env->eflags;
194 (*regs)[19] = env->regs[R_ESP];
195 (*regs)[20] = env->segs[R_SS].selector & 0xffff;
196 (*regs)[21] = env->segs[R_FS].selector & 0xffff;
197 (*regs)[22] = env->segs[R_GS].selector & 0xffff;
198 (*regs)[23] = env->segs[R_DS].selector & 0xffff;
199 (*regs)[24] = env->segs[R_ES].selector & 0xffff;
200 (*regs)[25] = env->segs[R_FS].selector & 0xffff;
201 (*regs)[26] = env->segs[R_GS].selector & 0xffff;
202 }
203
204 #else
205
206 #define ELF_START_MMAP 0x80000000
207
208 /*
209 * This is used to ensure we don't load something for the wrong architecture.
210 */
211 #define elf_check_arch(x) ( ((x) == EM_386) || ((x) == EM_486) )
212
213 /*
214 * These are used to set parameters in the core dumps.
215 */
216 #define ELF_CLASS ELFCLASS32
217 #define ELF_ARCH EM_386
218
219 static inline void init_thread(struct target_pt_regs *regs,
220 struct image_info *infop)
221 {
222 regs->esp = infop->start_stack;
223 regs->eip = infop->entry;
224
225 /* SVR4/i386 ABI (pages 3-31, 3-32) says that when the program
226 starts %edx contains a pointer to a function which might be
227 registered using `atexit'. This provides a mean for the
228 dynamic linker to call DT_FINI functions for shared libraries
229 that have been loaded before the code runs.
230
231 A value of 0 tells we have no such handler. */
232 regs->edx = 0;
233 }
234
235 #define ELF_NREG 17
236 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
237
238 /*
239 * Note that ELF_NREG should be 19 as there should be place for
240 * TRAPNO and ERR "registers" as well but linux doesn't dump
241 * those.
242 *
243 * See linux kernel: arch/x86/include/asm/elf.h
244 */
245 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUX86State *env)
246 {
247 (*regs)[0] = env->regs[R_EBX];
248 (*regs)[1] = env->regs[R_ECX];
249 (*regs)[2] = env->regs[R_EDX];
250 (*regs)[3] = env->regs[R_ESI];
251 (*regs)[4] = env->regs[R_EDI];
252 (*regs)[5] = env->regs[R_EBP];
253 (*regs)[6] = env->regs[R_EAX];
254 (*regs)[7] = env->segs[R_DS].selector & 0xffff;
255 (*regs)[8] = env->segs[R_ES].selector & 0xffff;
256 (*regs)[9] = env->segs[R_FS].selector & 0xffff;
257 (*regs)[10] = env->segs[R_GS].selector & 0xffff;
258 (*regs)[11] = env->regs[R_EAX]; /* XXX */
259 (*regs)[12] = env->eip;
260 (*regs)[13] = env->segs[R_CS].selector & 0xffff;
261 (*regs)[14] = env->eflags;
262 (*regs)[15] = env->regs[R_ESP];
263 (*regs)[16] = env->segs[R_SS].selector & 0xffff;
264 }
265 #endif
266
267 #define USE_ELF_CORE_DUMP
268 #define ELF_EXEC_PAGESIZE 4096
269
270 #endif
271
272 #ifdef TARGET_ARM
273
274 #ifndef TARGET_AARCH64
275 /* 32 bit ARM definitions */
276
277 #define ELF_START_MMAP 0x80000000
278
279 #define ELF_ARCH EM_ARM
280 #define ELF_CLASS ELFCLASS32
281
282 static inline void init_thread(struct target_pt_regs *regs,
283 struct image_info *infop)
284 {
285 abi_long stack = infop->start_stack;
286 memset(regs, 0, sizeof(*regs));
287
288 regs->uregs[16] = ARM_CPU_MODE_USR;
289 if (infop->entry & 1) {
290 regs->uregs[16] |= CPSR_T;
291 }
292 regs->uregs[15] = infop->entry & 0xfffffffe;
293 regs->uregs[13] = infop->start_stack;
294 /* FIXME - what to for failure of get_user()? */
295 get_user_ual(regs->uregs[2], stack + 8); /* envp */
296 get_user_ual(regs->uregs[1], stack + 4); /* envp */
297 /* XXX: it seems that r0 is zeroed after ! */
298 regs->uregs[0] = 0;
299 /* For uClinux PIC binaries. */
300 /* XXX: Linux does this only on ARM with no MMU (do we care ?) */
301 regs->uregs[10] = infop->start_data;
302
303 /* Support ARM FDPIC. */
304 if (info_is_fdpic(infop)) {
305 /* As described in the ABI document, r7 points to the loadmap info
306 * prepared by the kernel. If an interpreter is needed, r8 points
307 * to the interpreter loadmap and r9 points to the interpreter
308 * PT_DYNAMIC info. If no interpreter is needed, r8 is zero, and
309 * r9 points to the main program PT_DYNAMIC info.
310 */
311 regs->uregs[7] = infop->loadmap_addr;
312 if (infop->interpreter_loadmap_addr) {
313 /* Executable is dynamically loaded. */
314 regs->uregs[8] = infop->interpreter_loadmap_addr;
315 regs->uregs[9] = infop->interpreter_pt_dynamic_addr;
316 } else {
317 regs->uregs[8] = 0;
318 regs->uregs[9] = infop->pt_dynamic_addr;
319 }
320 }
321 }
322
323 #define ELF_NREG 18
324 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
325
326 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUARMState *env)
327 {
328 (*regs)[0] = tswapreg(env->regs[0]);
329 (*regs)[1] = tswapreg(env->regs[1]);
330 (*regs)[2] = tswapreg(env->regs[2]);
331 (*regs)[3] = tswapreg(env->regs[3]);
332 (*regs)[4] = tswapreg(env->regs[4]);
333 (*regs)[5] = tswapreg(env->regs[5]);
334 (*regs)[6] = tswapreg(env->regs[6]);
335 (*regs)[7] = tswapreg(env->regs[7]);
336 (*regs)[8] = tswapreg(env->regs[8]);
337 (*regs)[9] = tswapreg(env->regs[9]);
338 (*regs)[10] = tswapreg(env->regs[10]);
339 (*regs)[11] = tswapreg(env->regs[11]);
340 (*regs)[12] = tswapreg(env->regs[12]);
341 (*regs)[13] = tswapreg(env->regs[13]);
342 (*regs)[14] = tswapreg(env->regs[14]);
343 (*regs)[15] = tswapreg(env->regs[15]);
344
345 (*regs)[16] = tswapreg(cpsr_read((CPUARMState *)env));
346 (*regs)[17] = tswapreg(env->regs[0]); /* XXX */
347 }
348
349 #define USE_ELF_CORE_DUMP
350 #define ELF_EXEC_PAGESIZE 4096
351
352 enum
353 {
354 ARM_HWCAP_ARM_SWP = 1 << 0,
355 ARM_HWCAP_ARM_HALF = 1 << 1,
356 ARM_HWCAP_ARM_THUMB = 1 << 2,
357 ARM_HWCAP_ARM_26BIT = 1 << 3,
358 ARM_HWCAP_ARM_FAST_MULT = 1 << 4,
359 ARM_HWCAP_ARM_FPA = 1 << 5,
360 ARM_HWCAP_ARM_VFP = 1 << 6,
361 ARM_HWCAP_ARM_EDSP = 1 << 7,
362 ARM_HWCAP_ARM_JAVA = 1 << 8,
363 ARM_HWCAP_ARM_IWMMXT = 1 << 9,
364 ARM_HWCAP_ARM_CRUNCH = 1 << 10,
365 ARM_HWCAP_ARM_THUMBEE = 1 << 11,
366 ARM_HWCAP_ARM_NEON = 1 << 12,
367 ARM_HWCAP_ARM_VFPv3 = 1 << 13,
368 ARM_HWCAP_ARM_VFPv3D16 = 1 << 14,
369 ARM_HWCAP_ARM_TLS = 1 << 15,
370 ARM_HWCAP_ARM_VFPv4 = 1 << 16,
371 ARM_HWCAP_ARM_IDIVA = 1 << 17,
372 ARM_HWCAP_ARM_IDIVT = 1 << 18,
373 ARM_HWCAP_ARM_VFPD32 = 1 << 19,
374 ARM_HWCAP_ARM_LPAE = 1 << 20,
375 ARM_HWCAP_ARM_EVTSTRM = 1 << 21,
376 };
377
378 enum {
379 ARM_HWCAP2_ARM_AES = 1 << 0,
380 ARM_HWCAP2_ARM_PMULL = 1 << 1,
381 ARM_HWCAP2_ARM_SHA1 = 1 << 2,
382 ARM_HWCAP2_ARM_SHA2 = 1 << 3,
383 ARM_HWCAP2_ARM_CRC32 = 1 << 4,
384 };
385
386 /* The commpage only exists for 32 bit kernels */
387
388 #define ARM_COMMPAGE (intptr_t)0xffff0f00u
389
390 static bool init_guest_commpage(void)
391 {
392 void *want = g2h_untagged(ARM_COMMPAGE & -qemu_host_page_size);
393 void *addr = mmap(want, qemu_host_page_size, PROT_READ | PROT_WRITE,
394 MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0);
395
396 if (addr == MAP_FAILED) {
397 perror("Allocating guest commpage");
398 exit(EXIT_FAILURE);
399 }
400 if (addr != want) {
401 return false;
402 }
403
404 /* Set kernel helper versions; rest of page is 0. */
405 __put_user(5, (uint32_t *)g2h_untagged(0xffff0ffcu));
406
407 if (mprotect(addr, qemu_host_page_size, PROT_READ)) {
408 perror("Protecting guest commpage");
409 exit(EXIT_FAILURE);
410 }
411 return true;
412 }
413
414 #define ELF_HWCAP get_elf_hwcap()
415 #define ELF_HWCAP2 get_elf_hwcap2()
416
417 static uint32_t get_elf_hwcap(void)
418 {
419 ARMCPU *cpu = ARM_CPU(thread_cpu);
420 uint32_t hwcaps = 0;
421
422 hwcaps |= ARM_HWCAP_ARM_SWP;
423 hwcaps |= ARM_HWCAP_ARM_HALF;
424 hwcaps |= ARM_HWCAP_ARM_THUMB;
425 hwcaps |= ARM_HWCAP_ARM_FAST_MULT;
426
427 /* probe for the extra features */
428 #define GET_FEATURE(feat, hwcap) \
429 do { if (arm_feature(&cpu->env, feat)) { hwcaps |= hwcap; } } while (0)
430
431 #define GET_FEATURE_ID(feat, hwcap) \
432 do { if (cpu_isar_feature(feat, cpu)) { hwcaps |= hwcap; } } while (0)
433
434 /* EDSP is in v5TE and above, but all our v5 CPUs are v5TE */
435 GET_FEATURE(ARM_FEATURE_V5, ARM_HWCAP_ARM_EDSP);
436 GET_FEATURE(ARM_FEATURE_IWMMXT, ARM_HWCAP_ARM_IWMMXT);
437 GET_FEATURE(ARM_FEATURE_THUMB2EE, ARM_HWCAP_ARM_THUMBEE);
438 GET_FEATURE(ARM_FEATURE_NEON, ARM_HWCAP_ARM_NEON);
439 GET_FEATURE(ARM_FEATURE_V6K, ARM_HWCAP_ARM_TLS);
440 GET_FEATURE(ARM_FEATURE_LPAE, ARM_HWCAP_ARM_LPAE);
441 GET_FEATURE_ID(aa32_arm_div, ARM_HWCAP_ARM_IDIVA);
442 GET_FEATURE_ID(aa32_thumb_div, ARM_HWCAP_ARM_IDIVT);
443 GET_FEATURE_ID(aa32_vfp, ARM_HWCAP_ARM_VFP);
444
445 if (cpu_isar_feature(aa32_fpsp_v3, cpu) ||
446 cpu_isar_feature(aa32_fpdp_v3, cpu)) {
447 hwcaps |= ARM_HWCAP_ARM_VFPv3;
448 if (cpu_isar_feature(aa32_simd_r32, cpu)) {
449 hwcaps |= ARM_HWCAP_ARM_VFPD32;
450 } else {
451 hwcaps |= ARM_HWCAP_ARM_VFPv3D16;
452 }
453 }
454 GET_FEATURE_ID(aa32_simdfmac, ARM_HWCAP_ARM_VFPv4);
455
456 return hwcaps;
457 }
458
459 static uint32_t get_elf_hwcap2(void)
460 {
461 ARMCPU *cpu = ARM_CPU(thread_cpu);
462 uint32_t hwcaps = 0;
463
464 GET_FEATURE_ID(aa32_aes, ARM_HWCAP2_ARM_AES);
465 GET_FEATURE_ID(aa32_pmull, ARM_HWCAP2_ARM_PMULL);
466 GET_FEATURE_ID(aa32_sha1, ARM_HWCAP2_ARM_SHA1);
467 GET_FEATURE_ID(aa32_sha2, ARM_HWCAP2_ARM_SHA2);
468 GET_FEATURE_ID(aa32_crc32, ARM_HWCAP2_ARM_CRC32);
469 return hwcaps;
470 }
471
472 #undef GET_FEATURE
473 #undef GET_FEATURE_ID
474
475 #define ELF_PLATFORM get_elf_platform()
476
477 static const char *get_elf_platform(void)
478 {
479 CPUARMState *env = thread_cpu->env_ptr;
480
481 #ifdef TARGET_WORDS_BIGENDIAN
482 # define END "b"
483 #else
484 # define END "l"
485 #endif
486
487 if (arm_feature(env, ARM_FEATURE_V8)) {
488 return "v8" END;
489 } else if (arm_feature(env, ARM_FEATURE_V7)) {
490 if (arm_feature(env, ARM_FEATURE_M)) {
491 return "v7m" END;
492 } else {
493 return "v7" END;
494 }
495 } else if (arm_feature(env, ARM_FEATURE_V6)) {
496 return "v6" END;
497 } else if (arm_feature(env, ARM_FEATURE_V5)) {
498 return "v5" END;
499 } else {
500 return "v4" END;
501 }
502
503 #undef END
504 }
505
506 #else
507 /* 64 bit ARM definitions */
508 #define ELF_START_MMAP 0x80000000
509
510 #define ELF_ARCH EM_AARCH64
511 #define ELF_CLASS ELFCLASS64
512 #ifdef TARGET_WORDS_BIGENDIAN
513 # define ELF_PLATFORM "aarch64_be"
514 #else
515 # define ELF_PLATFORM "aarch64"
516 #endif
517
518 static inline void init_thread(struct target_pt_regs *regs,
519 struct image_info *infop)
520 {
521 abi_long stack = infop->start_stack;
522 memset(regs, 0, sizeof(*regs));
523
524 regs->pc = infop->entry & ~0x3ULL;
525 regs->sp = stack;
526 }
527
528 #define ELF_NREG 34
529 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
530
531 static void elf_core_copy_regs(target_elf_gregset_t *regs,
532 const CPUARMState *env)
533 {
534 int i;
535
536 for (i = 0; i < 32; i++) {
537 (*regs)[i] = tswapreg(env->xregs[i]);
538 }
539 (*regs)[32] = tswapreg(env->pc);
540 (*regs)[33] = tswapreg(pstate_read((CPUARMState *)env));
541 }
542
543 #define USE_ELF_CORE_DUMP
544 #define ELF_EXEC_PAGESIZE 4096
545
546 enum {
547 ARM_HWCAP_A64_FP = 1 << 0,
548 ARM_HWCAP_A64_ASIMD = 1 << 1,
549 ARM_HWCAP_A64_EVTSTRM = 1 << 2,
550 ARM_HWCAP_A64_AES = 1 << 3,
551 ARM_HWCAP_A64_PMULL = 1 << 4,
552 ARM_HWCAP_A64_SHA1 = 1 << 5,
553 ARM_HWCAP_A64_SHA2 = 1 << 6,
554 ARM_HWCAP_A64_CRC32 = 1 << 7,
555 ARM_HWCAP_A64_ATOMICS = 1 << 8,
556 ARM_HWCAP_A64_FPHP = 1 << 9,
557 ARM_HWCAP_A64_ASIMDHP = 1 << 10,
558 ARM_HWCAP_A64_CPUID = 1 << 11,
559 ARM_HWCAP_A64_ASIMDRDM = 1 << 12,
560 ARM_HWCAP_A64_JSCVT = 1 << 13,
561 ARM_HWCAP_A64_FCMA = 1 << 14,
562 ARM_HWCAP_A64_LRCPC = 1 << 15,
563 ARM_HWCAP_A64_DCPOP = 1 << 16,
564 ARM_HWCAP_A64_SHA3 = 1 << 17,
565 ARM_HWCAP_A64_SM3 = 1 << 18,
566 ARM_HWCAP_A64_SM4 = 1 << 19,
567 ARM_HWCAP_A64_ASIMDDP = 1 << 20,
568 ARM_HWCAP_A64_SHA512 = 1 << 21,
569 ARM_HWCAP_A64_SVE = 1 << 22,
570 ARM_HWCAP_A64_ASIMDFHM = 1 << 23,
571 ARM_HWCAP_A64_DIT = 1 << 24,
572 ARM_HWCAP_A64_USCAT = 1 << 25,
573 ARM_HWCAP_A64_ILRCPC = 1 << 26,
574 ARM_HWCAP_A64_FLAGM = 1 << 27,
575 ARM_HWCAP_A64_SSBS = 1 << 28,
576 ARM_HWCAP_A64_SB = 1 << 29,
577 ARM_HWCAP_A64_PACA = 1 << 30,
578 ARM_HWCAP_A64_PACG = 1UL << 31,
579
580 ARM_HWCAP2_A64_DCPODP = 1 << 0,
581 ARM_HWCAP2_A64_SVE2 = 1 << 1,
582 ARM_HWCAP2_A64_SVEAES = 1 << 2,
583 ARM_HWCAP2_A64_SVEPMULL = 1 << 3,
584 ARM_HWCAP2_A64_SVEBITPERM = 1 << 4,
585 ARM_HWCAP2_A64_SVESHA3 = 1 << 5,
586 ARM_HWCAP2_A64_SVESM4 = 1 << 6,
587 ARM_HWCAP2_A64_FLAGM2 = 1 << 7,
588 ARM_HWCAP2_A64_FRINT = 1 << 8,
589 ARM_HWCAP2_A64_SVEI8MM = 1 << 9,
590 ARM_HWCAP2_A64_SVEF32MM = 1 << 10,
591 ARM_HWCAP2_A64_SVEF64MM = 1 << 11,
592 ARM_HWCAP2_A64_SVEBF16 = 1 << 12,
593 ARM_HWCAP2_A64_I8MM = 1 << 13,
594 ARM_HWCAP2_A64_BF16 = 1 << 14,
595 ARM_HWCAP2_A64_DGH = 1 << 15,
596 ARM_HWCAP2_A64_RNG = 1 << 16,
597 ARM_HWCAP2_A64_BTI = 1 << 17,
598 ARM_HWCAP2_A64_MTE = 1 << 18,
599 };
600
601 #define ELF_HWCAP get_elf_hwcap()
602 #define ELF_HWCAP2 get_elf_hwcap2()
603
604 #define GET_FEATURE_ID(feat, hwcap) \
605 do { if (cpu_isar_feature(feat, cpu)) { hwcaps |= hwcap; } } while (0)
606
607 static uint32_t get_elf_hwcap(void)
608 {
609 ARMCPU *cpu = ARM_CPU(thread_cpu);
610 uint32_t hwcaps = 0;
611
612 hwcaps |= ARM_HWCAP_A64_FP;
613 hwcaps |= ARM_HWCAP_A64_ASIMD;
614 hwcaps |= ARM_HWCAP_A64_CPUID;
615
616 /* probe for the extra features */
617
618 GET_FEATURE_ID(aa64_aes, ARM_HWCAP_A64_AES);
619 GET_FEATURE_ID(aa64_pmull, ARM_HWCAP_A64_PMULL);
620 GET_FEATURE_ID(aa64_sha1, ARM_HWCAP_A64_SHA1);
621 GET_FEATURE_ID(aa64_sha256, ARM_HWCAP_A64_SHA2);
622 GET_FEATURE_ID(aa64_sha512, ARM_HWCAP_A64_SHA512);
623 GET_FEATURE_ID(aa64_crc32, ARM_HWCAP_A64_CRC32);
624 GET_FEATURE_ID(aa64_sha3, ARM_HWCAP_A64_SHA3);
625 GET_FEATURE_ID(aa64_sm3, ARM_HWCAP_A64_SM3);
626 GET_FEATURE_ID(aa64_sm4, ARM_HWCAP_A64_SM4);
627 GET_FEATURE_ID(aa64_fp16, ARM_HWCAP_A64_FPHP | ARM_HWCAP_A64_ASIMDHP);
628 GET_FEATURE_ID(aa64_atomics, ARM_HWCAP_A64_ATOMICS);
629 GET_FEATURE_ID(aa64_rdm, ARM_HWCAP_A64_ASIMDRDM);
630 GET_FEATURE_ID(aa64_dp, ARM_HWCAP_A64_ASIMDDP);
631 GET_FEATURE_ID(aa64_fcma, ARM_HWCAP_A64_FCMA);
632 GET_FEATURE_ID(aa64_sve, ARM_HWCAP_A64_SVE);
633 GET_FEATURE_ID(aa64_pauth, ARM_HWCAP_A64_PACA | ARM_HWCAP_A64_PACG);
634 GET_FEATURE_ID(aa64_fhm, ARM_HWCAP_A64_ASIMDFHM);
635 GET_FEATURE_ID(aa64_jscvt, ARM_HWCAP_A64_JSCVT);
636 GET_FEATURE_ID(aa64_sb, ARM_HWCAP_A64_SB);
637 GET_FEATURE_ID(aa64_condm_4, ARM_HWCAP_A64_FLAGM);
638 GET_FEATURE_ID(aa64_dcpop, ARM_HWCAP_A64_DCPOP);
639 GET_FEATURE_ID(aa64_rcpc_8_3, ARM_HWCAP_A64_LRCPC);
640 GET_FEATURE_ID(aa64_rcpc_8_4, ARM_HWCAP_A64_ILRCPC);
641
642 return hwcaps;
643 }
644
645 static uint32_t get_elf_hwcap2(void)
646 {
647 ARMCPU *cpu = ARM_CPU(thread_cpu);
648 uint32_t hwcaps = 0;
649
650 GET_FEATURE_ID(aa64_dcpodp, ARM_HWCAP2_A64_DCPODP);
651 GET_FEATURE_ID(aa64_condm_5, ARM_HWCAP2_A64_FLAGM2);
652 GET_FEATURE_ID(aa64_frint, ARM_HWCAP2_A64_FRINT);
653 GET_FEATURE_ID(aa64_rndr, ARM_HWCAP2_A64_RNG);
654 GET_FEATURE_ID(aa64_bti, ARM_HWCAP2_A64_BTI);
655 GET_FEATURE_ID(aa64_mte, ARM_HWCAP2_A64_MTE);
656
657 return hwcaps;
658 }
659
660 #undef GET_FEATURE_ID
661
662 #endif /* not TARGET_AARCH64 */
663 #endif /* TARGET_ARM */
664
665 #ifdef TARGET_SPARC
666 #ifdef TARGET_SPARC64
667
668 #define ELF_START_MMAP 0x80000000
669 #define ELF_HWCAP (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \
670 | HWCAP_SPARC_MULDIV | HWCAP_SPARC_V9)
671 #ifndef TARGET_ABI32
672 #define elf_check_arch(x) ( (x) == EM_SPARCV9 || (x) == EM_SPARC32PLUS )
673 #else
674 #define elf_check_arch(x) ( (x) == EM_SPARC32PLUS || (x) == EM_SPARC )
675 #endif
676
677 #define ELF_CLASS ELFCLASS64
678 #define ELF_ARCH EM_SPARCV9
679 #else
680 #define ELF_START_MMAP 0x80000000
681 #define ELF_HWCAP (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \
682 | HWCAP_SPARC_MULDIV)
683 #define ELF_CLASS ELFCLASS32
684 #define ELF_ARCH EM_SPARC
685 #endif /* TARGET_SPARC64 */
686
687 static inline void init_thread(struct target_pt_regs *regs,
688 struct image_info *infop)
689 {
690 /* Note that target_cpu_copy_regs does not read psr/tstate. */
691 regs->pc = infop->entry;
692 regs->npc = regs->pc + 4;
693 regs->y = 0;
694 regs->u_regs[14] = (infop->start_stack - 16 * sizeof(abi_ulong)
695 - TARGET_STACK_BIAS);
696 }
697 #endif /* TARGET_SPARC */
698
699 #ifdef TARGET_PPC
700
701 #define ELF_MACHINE PPC_ELF_MACHINE
702 #define ELF_START_MMAP 0x80000000
703
704 #if defined(TARGET_PPC64) && !defined(TARGET_ABI32)
705
706 #define elf_check_arch(x) ( (x) == EM_PPC64 )
707
708 #define ELF_CLASS ELFCLASS64
709
710 #else
711
712 #define ELF_CLASS ELFCLASS32
713
714 #endif
715
716 #define ELF_ARCH EM_PPC
717
718 /* Feature masks for the Aux Vector Hardware Capabilities (AT_HWCAP).
719 See arch/powerpc/include/asm/cputable.h. */
720 enum {
721 QEMU_PPC_FEATURE_32 = 0x80000000,
722 QEMU_PPC_FEATURE_64 = 0x40000000,
723 QEMU_PPC_FEATURE_601_INSTR = 0x20000000,
724 QEMU_PPC_FEATURE_HAS_ALTIVEC = 0x10000000,
725 QEMU_PPC_FEATURE_HAS_FPU = 0x08000000,
726 QEMU_PPC_FEATURE_HAS_MMU = 0x04000000,
727 QEMU_PPC_FEATURE_HAS_4xxMAC = 0x02000000,
728 QEMU_PPC_FEATURE_UNIFIED_CACHE = 0x01000000,
729 QEMU_PPC_FEATURE_HAS_SPE = 0x00800000,
730 QEMU_PPC_FEATURE_HAS_EFP_SINGLE = 0x00400000,
731 QEMU_PPC_FEATURE_HAS_EFP_DOUBLE = 0x00200000,
732 QEMU_PPC_FEATURE_NO_TB = 0x00100000,
733 QEMU_PPC_FEATURE_POWER4 = 0x00080000,
734 QEMU_PPC_FEATURE_POWER5 = 0x00040000,
735 QEMU_PPC_FEATURE_POWER5_PLUS = 0x00020000,
736 QEMU_PPC_FEATURE_CELL = 0x00010000,
737 QEMU_PPC_FEATURE_BOOKE = 0x00008000,
738 QEMU_PPC_FEATURE_SMT = 0x00004000,
739 QEMU_PPC_FEATURE_ICACHE_SNOOP = 0x00002000,
740 QEMU_PPC_FEATURE_ARCH_2_05 = 0x00001000,
741 QEMU_PPC_FEATURE_PA6T = 0x00000800,
742 QEMU_PPC_FEATURE_HAS_DFP = 0x00000400,
743 QEMU_PPC_FEATURE_POWER6_EXT = 0x00000200,
744 QEMU_PPC_FEATURE_ARCH_2_06 = 0x00000100,
745 QEMU_PPC_FEATURE_HAS_VSX = 0x00000080,
746 QEMU_PPC_FEATURE_PSERIES_PERFMON_COMPAT = 0x00000040,
747
748 QEMU_PPC_FEATURE_TRUE_LE = 0x00000002,
749 QEMU_PPC_FEATURE_PPC_LE = 0x00000001,
750
751 /* Feature definitions in AT_HWCAP2. */
752 QEMU_PPC_FEATURE2_ARCH_2_07 = 0x80000000, /* ISA 2.07 */
753 QEMU_PPC_FEATURE2_HAS_HTM = 0x40000000, /* Hardware Transactional Memory */
754 QEMU_PPC_FEATURE2_HAS_DSCR = 0x20000000, /* Data Stream Control Register */
755 QEMU_PPC_FEATURE2_HAS_EBB = 0x10000000, /* Event Base Branching */
756 QEMU_PPC_FEATURE2_HAS_ISEL = 0x08000000, /* Integer Select */
757 QEMU_PPC_FEATURE2_HAS_TAR = 0x04000000, /* Target Address Register */
758 QEMU_PPC_FEATURE2_VEC_CRYPTO = 0x02000000,
759 QEMU_PPC_FEATURE2_HTM_NOSC = 0x01000000,
760 QEMU_PPC_FEATURE2_ARCH_3_00 = 0x00800000, /* ISA 3.00 */
761 QEMU_PPC_FEATURE2_HAS_IEEE128 = 0x00400000, /* VSX IEEE Bin Float 128-bit */
762 QEMU_PPC_FEATURE2_DARN = 0x00200000, /* darn random number insn */
763 QEMU_PPC_FEATURE2_SCV = 0x00100000, /* scv syscall */
764 QEMU_PPC_FEATURE2_HTM_NO_SUSPEND = 0x00080000, /* TM w/o suspended state */
765 };
766
767 #define ELF_HWCAP get_elf_hwcap()
768
769 static uint32_t get_elf_hwcap(void)
770 {
771 PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);
772 uint32_t features = 0;
773
774 /* We don't have to be terribly complete here; the high points are
775 Altivec/FP/SPE support. Anything else is just a bonus. */
776 #define GET_FEATURE(flag, feature) \
777 do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0)
778 #define GET_FEATURE2(flags, feature) \
779 do { \
780 if ((cpu->env.insns_flags2 & flags) == flags) { \
781 features |= feature; \
782 } \
783 } while (0)
784 GET_FEATURE(PPC_64B, QEMU_PPC_FEATURE_64);
785 GET_FEATURE(PPC_FLOAT, QEMU_PPC_FEATURE_HAS_FPU);
786 GET_FEATURE(PPC_ALTIVEC, QEMU_PPC_FEATURE_HAS_ALTIVEC);
787 GET_FEATURE(PPC_SPE, QEMU_PPC_FEATURE_HAS_SPE);
788 GET_FEATURE(PPC_SPE_SINGLE, QEMU_PPC_FEATURE_HAS_EFP_SINGLE);
789 GET_FEATURE(PPC_SPE_DOUBLE, QEMU_PPC_FEATURE_HAS_EFP_DOUBLE);
790 GET_FEATURE(PPC_BOOKE, QEMU_PPC_FEATURE_BOOKE);
791 GET_FEATURE(PPC_405_MAC, QEMU_PPC_FEATURE_HAS_4xxMAC);
792 GET_FEATURE2(PPC2_DFP, QEMU_PPC_FEATURE_HAS_DFP);
793 GET_FEATURE2(PPC2_VSX, QEMU_PPC_FEATURE_HAS_VSX);
794 GET_FEATURE2((PPC2_PERM_ISA206 | PPC2_DIVE_ISA206 | PPC2_ATOMIC_ISA206 |
795 PPC2_FP_CVT_ISA206 | PPC2_FP_TST_ISA206),
796 QEMU_PPC_FEATURE_ARCH_2_06);
797 #undef GET_FEATURE
798 #undef GET_FEATURE2
799
800 return features;
801 }
802
803 #define ELF_HWCAP2 get_elf_hwcap2()
804
805 static uint32_t get_elf_hwcap2(void)
806 {
807 PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);
808 uint32_t features = 0;
809
810 #define GET_FEATURE(flag, feature) \
811 do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0)
812 #define GET_FEATURE2(flag, feature) \
813 do { if (cpu->env.insns_flags2 & flag) { features |= feature; } } while (0)
814
815 GET_FEATURE(PPC_ISEL, QEMU_PPC_FEATURE2_HAS_ISEL);
816 GET_FEATURE2(PPC2_BCTAR_ISA207, QEMU_PPC_FEATURE2_HAS_TAR);
817 GET_FEATURE2((PPC2_BCTAR_ISA207 | PPC2_LSQ_ISA207 | PPC2_ALTIVEC_207 |
818 PPC2_ISA207S), QEMU_PPC_FEATURE2_ARCH_2_07 |
819 QEMU_PPC_FEATURE2_VEC_CRYPTO);
820 GET_FEATURE2(PPC2_ISA300, QEMU_PPC_FEATURE2_ARCH_3_00 |
821 QEMU_PPC_FEATURE2_DARN);
822
823 #undef GET_FEATURE
824 #undef GET_FEATURE2
825
826 return features;
827 }
828
829 /*
830 * The requirements here are:
831 * - keep the final alignment of sp (sp & 0xf)
832 * - make sure the 32-bit value at the first 16 byte aligned position of
833 * AUXV is greater than 16 for glibc compatibility.
834 * AT_IGNOREPPC is used for that.
835 * - for compatibility with glibc ARCH_DLINFO must always be defined on PPC,
836 * even if DLINFO_ARCH_ITEMS goes to zero or is undefined.
837 */
838 #define DLINFO_ARCH_ITEMS 5
839 #define ARCH_DLINFO \
840 do { \
841 PowerPCCPU *cpu = POWERPC_CPU(thread_cpu); \
842 /* \
843 * Handle glibc compatibility: these magic entries must \
844 * be at the lowest addresses in the final auxv. \
845 */ \
846 NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC); \
847 NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC); \
848 NEW_AUX_ENT(AT_DCACHEBSIZE, cpu->env.dcache_line_size); \
849 NEW_AUX_ENT(AT_ICACHEBSIZE, cpu->env.icache_line_size); \
850 NEW_AUX_ENT(AT_UCACHEBSIZE, 0); \
851 } while (0)
852
853 static inline void init_thread(struct target_pt_regs *_regs, struct image_info *infop)
854 {
855 _regs->gpr[1] = infop->start_stack;
856 #if defined(TARGET_PPC64) && !defined(TARGET_ABI32)
857 if (get_ppc64_abi(infop) < 2) {
858 uint64_t val;
859 get_user_u64(val, infop->entry + 8);
860 _regs->gpr[2] = val + infop->load_bias;
861 get_user_u64(val, infop->entry);
862 infop->entry = val + infop->load_bias;
863 } else {
864 _regs->gpr[12] = infop->entry; /* r12 set to global entry address */
865 }
866 #endif
867 _regs->nip = infop->entry;
868 }
869
870 /* See linux kernel: arch/powerpc/include/asm/elf.h. */
871 #define ELF_NREG 48
872 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
873
874 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUPPCState *env)
875 {
876 int i;
877 target_ulong ccr = 0;
878
879 for (i = 0; i < ARRAY_SIZE(env->gpr); i++) {
880 (*regs)[i] = tswapreg(env->gpr[i]);
881 }
882
883 (*regs)[32] = tswapreg(env->nip);
884 (*regs)[33] = tswapreg(env->msr);
885 (*regs)[35] = tswapreg(env->ctr);
886 (*regs)[36] = tswapreg(env->lr);
887 (*regs)[37] = tswapreg(env->xer);
888
889 for (i = 0; i < ARRAY_SIZE(env->crf); i++) {
890 ccr |= env->crf[i] << (32 - ((i + 1) * 4));
891 }
892 (*regs)[38] = tswapreg(ccr);
893 }
894
895 #define USE_ELF_CORE_DUMP
896 #define ELF_EXEC_PAGESIZE 4096
897
898 #endif
899
900 #ifdef TARGET_MIPS
901
902 #define ELF_START_MMAP 0x80000000
903
904 #ifdef TARGET_MIPS64
905 #define ELF_CLASS ELFCLASS64
906 #else
907 #define ELF_CLASS ELFCLASS32
908 #endif
909 #define ELF_ARCH EM_MIPS
910
911 #define elf_check_arch(x) ((x) == EM_MIPS || (x) == EM_NANOMIPS)
912
913 #ifdef TARGET_ABI_MIPSN32
914 #define elf_check_abi(x) ((x) & EF_MIPS_ABI2)
915 #else
916 #define elf_check_abi(x) (!((x) & EF_MIPS_ABI2))
917 #endif
918
919 static inline void init_thread(struct target_pt_regs *regs,
920 struct image_info *infop)
921 {
922 regs->cp0_status = 2 << CP0St_KSU;
923 regs->cp0_epc = infop->entry;
924 regs->regs[29] = infop->start_stack;
925 }
926
927 /* See linux kernel: arch/mips/include/asm/elf.h. */
928 #define ELF_NREG 45
929 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
930
931 /* See linux kernel: arch/mips/include/asm/reg.h. */
932 enum {
933 #ifdef TARGET_MIPS64
934 TARGET_EF_R0 = 0,
935 #else
936 TARGET_EF_R0 = 6,
937 #endif
938 TARGET_EF_R26 = TARGET_EF_R0 + 26,
939 TARGET_EF_R27 = TARGET_EF_R0 + 27,
940 TARGET_EF_LO = TARGET_EF_R0 + 32,
941 TARGET_EF_HI = TARGET_EF_R0 + 33,
942 TARGET_EF_CP0_EPC = TARGET_EF_R0 + 34,
943 TARGET_EF_CP0_BADVADDR = TARGET_EF_R0 + 35,
944 TARGET_EF_CP0_STATUS = TARGET_EF_R0 + 36,
945 TARGET_EF_CP0_CAUSE = TARGET_EF_R0 + 37
946 };
947
948 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */
949 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMIPSState *env)
950 {
951 int i;
952
953 for (i = 0; i < TARGET_EF_R0; i++) {
954 (*regs)[i] = 0;
955 }
956 (*regs)[TARGET_EF_R0] = 0;
957
958 for (i = 1; i < ARRAY_SIZE(env->active_tc.gpr); i++) {
959 (*regs)[TARGET_EF_R0 + i] = tswapreg(env->active_tc.gpr[i]);
960 }
961
962 (*regs)[TARGET_EF_R26] = 0;
963 (*regs)[TARGET_EF_R27] = 0;
964 (*regs)[TARGET_EF_LO] = tswapreg(env->active_tc.LO[0]);
965 (*regs)[TARGET_EF_HI] = tswapreg(env->active_tc.HI[0]);
966 (*regs)[TARGET_EF_CP0_EPC] = tswapreg(env->active_tc.PC);
967 (*regs)[TARGET_EF_CP0_BADVADDR] = tswapreg(env->CP0_BadVAddr);
968 (*regs)[TARGET_EF_CP0_STATUS] = tswapreg(env->CP0_Status);
969 (*regs)[TARGET_EF_CP0_CAUSE] = tswapreg(env->CP0_Cause);
970 }
971
972 #define USE_ELF_CORE_DUMP
973 #define ELF_EXEC_PAGESIZE 4096
974
975 /* See arch/mips/include/uapi/asm/hwcap.h. */
976 enum {
977 HWCAP_MIPS_R6 = (1 << 0),
978 HWCAP_MIPS_MSA = (1 << 1),
979 HWCAP_MIPS_CRC32 = (1 << 2),
980 HWCAP_MIPS_MIPS16 = (1 << 3),
981 HWCAP_MIPS_MDMX = (1 << 4),
982 HWCAP_MIPS_MIPS3D = (1 << 5),
983 HWCAP_MIPS_SMARTMIPS = (1 << 6),
984 HWCAP_MIPS_DSP = (1 << 7),
985 HWCAP_MIPS_DSP2 = (1 << 8),
986 HWCAP_MIPS_DSP3 = (1 << 9),
987 HWCAP_MIPS_MIPS16E2 = (1 << 10),
988 HWCAP_LOONGSON_MMI = (1 << 11),
989 HWCAP_LOONGSON_EXT = (1 << 12),
990 HWCAP_LOONGSON_EXT2 = (1 << 13),
991 HWCAP_LOONGSON_CPUCFG = (1 << 14),
992 };
993
994 #define ELF_HWCAP get_elf_hwcap()
995
996 #define GET_FEATURE_INSN(_flag, _hwcap) \
997 do { if (cpu->env.insn_flags & (_flag)) { hwcaps |= _hwcap; } } while (0)
998
999 #define GET_FEATURE_REG_SET(_reg, _mask, _hwcap) \
1000 do { if (cpu->env._reg & (_mask)) { hwcaps |= _hwcap; } } while (0)
1001
1002 #define GET_FEATURE_REG_EQU(_reg, _start, _length, _val, _hwcap) \
1003 do { \
1004 if (extract32(cpu->env._reg, (_start), (_length)) == (_val)) { \
1005 hwcaps |= _hwcap; \
1006 } \
1007 } while (0)
1008
1009 static uint32_t get_elf_hwcap(void)
1010 {
1011 MIPSCPU *cpu = MIPS_CPU(thread_cpu);
1012 uint32_t hwcaps = 0;
1013
1014 GET_FEATURE_REG_EQU(CP0_Config0, CP0C0_AR, CP0C0_AR_LENGTH,
1015 2, HWCAP_MIPS_R6);
1016 GET_FEATURE_REG_SET(CP0_Config3, 1 << CP0C3_MSAP, HWCAP_MIPS_MSA);
1017 GET_FEATURE_INSN(ASE_LMMI, HWCAP_LOONGSON_MMI);
1018 GET_FEATURE_INSN(ASE_LEXT, HWCAP_LOONGSON_EXT);
1019
1020 return hwcaps;
1021 }
1022
1023 #undef GET_FEATURE_REG_EQU
1024 #undef GET_FEATURE_REG_SET
1025 #undef GET_FEATURE_INSN
1026
1027 #endif /* TARGET_MIPS */
1028
1029 #ifdef TARGET_MICROBLAZE
1030
1031 #define ELF_START_MMAP 0x80000000
1032
1033 #define elf_check_arch(x) ( (x) == EM_MICROBLAZE || (x) == EM_MICROBLAZE_OLD)
1034
1035 #define ELF_CLASS ELFCLASS32
1036 #define ELF_ARCH EM_MICROBLAZE
1037
1038 static inline void init_thread(struct target_pt_regs *regs,
1039 struct image_info *infop)
1040 {
1041 regs->pc = infop->entry;
1042 regs->r1 = infop->start_stack;
1043
1044 }
1045
1046 #define ELF_EXEC_PAGESIZE 4096
1047
1048 #define USE_ELF_CORE_DUMP
1049 #define ELF_NREG 38
1050 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1051
1052 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */
1053 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMBState *env)
1054 {
1055 int i, pos = 0;
1056
1057 for (i = 0; i < 32; i++) {
1058 (*regs)[pos++] = tswapreg(env->regs[i]);
1059 }
1060
1061 (*regs)[pos++] = tswapreg(env->pc);
1062 (*regs)[pos++] = tswapreg(mb_cpu_read_msr(env));
1063 (*regs)[pos++] = 0;
1064 (*regs)[pos++] = tswapreg(env->ear);
1065 (*regs)[pos++] = 0;
1066 (*regs)[pos++] = tswapreg(env->esr);
1067 }
1068
1069 #endif /* TARGET_MICROBLAZE */
1070
1071 #ifdef TARGET_NIOS2
1072
1073 #define ELF_START_MMAP 0x80000000
1074
1075 #define elf_check_arch(x) ((x) == EM_ALTERA_NIOS2)
1076
1077 #define ELF_CLASS ELFCLASS32
1078 #define ELF_ARCH EM_ALTERA_NIOS2
1079
1080 static void init_thread(struct target_pt_regs *regs, struct image_info *infop)
1081 {
1082 regs->ea = infop->entry;
1083 regs->sp = infop->start_stack;
1084 regs->estatus = 0x3;
1085 }
1086
1087 #define ELF_EXEC_PAGESIZE 4096
1088
1089 #define USE_ELF_CORE_DUMP
1090 #define ELF_NREG 49
1091 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1092
1093 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */
1094 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1095 const CPUNios2State *env)
1096 {
1097 int i;
1098
1099 (*regs)[0] = -1;
1100 for (i = 1; i < 8; i++) /* r0-r7 */
1101 (*regs)[i] = tswapreg(env->regs[i + 7]);
1102
1103 for (i = 8; i < 16; i++) /* r8-r15 */
1104 (*regs)[i] = tswapreg(env->regs[i - 8]);
1105
1106 for (i = 16; i < 24; i++) /* r16-r23 */
1107 (*regs)[i] = tswapreg(env->regs[i + 7]);
1108 (*regs)[24] = -1; /* R_ET */
1109 (*regs)[25] = -1; /* R_BT */
1110 (*regs)[26] = tswapreg(env->regs[R_GP]);
1111 (*regs)[27] = tswapreg(env->regs[R_SP]);
1112 (*regs)[28] = tswapreg(env->regs[R_FP]);
1113 (*regs)[29] = tswapreg(env->regs[R_EA]);
1114 (*regs)[30] = -1; /* R_SSTATUS */
1115 (*regs)[31] = tswapreg(env->regs[R_RA]);
1116
1117 (*regs)[32] = tswapreg(env->regs[R_PC]);
1118
1119 (*regs)[33] = -1; /* R_STATUS */
1120 (*regs)[34] = tswapreg(env->regs[CR_ESTATUS]);
1121
1122 for (i = 35; i < 49; i++) /* ... */
1123 (*regs)[i] = -1;
1124 }
1125
1126 #endif /* TARGET_NIOS2 */
1127
1128 #ifdef TARGET_OPENRISC
1129
1130 #define ELF_START_MMAP 0x08000000
1131
1132 #define ELF_ARCH EM_OPENRISC
1133 #define ELF_CLASS ELFCLASS32
1134 #define ELF_DATA ELFDATA2MSB
1135
1136 static inline void init_thread(struct target_pt_regs *regs,
1137 struct image_info *infop)
1138 {
1139 regs->pc = infop->entry;
1140 regs->gpr[1] = infop->start_stack;
1141 }
1142
1143 #define USE_ELF_CORE_DUMP
1144 #define ELF_EXEC_PAGESIZE 8192
1145
1146 /* See linux kernel arch/openrisc/include/asm/elf.h. */
1147 #define ELF_NREG 34 /* gprs and pc, sr */
1148 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1149
1150 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1151 const CPUOpenRISCState *env)
1152 {
1153 int i;
1154
1155 for (i = 0; i < 32; i++) {
1156 (*regs)[i] = tswapreg(cpu_get_gpr(env, i));
1157 }
1158 (*regs)[32] = tswapreg(env->pc);
1159 (*regs)[33] = tswapreg(cpu_get_sr(env));
1160 }
1161 #define ELF_HWCAP 0
1162 #define ELF_PLATFORM NULL
1163
1164 #endif /* TARGET_OPENRISC */
1165
1166 #ifdef TARGET_SH4
1167
1168 #define ELF_START_MMAP 0x80000000
1169
1170 #define ELF_CLASS ELFCLASS32
1171 #define ELF_ARCH EM_SH
1172
1173 static inline void init_thread(struct target_pt_regs *regs,
1174 struct image_info *infop)
1175 {
1176 /* Check other registers XXXXX */
1177 regs->pc = infop->entry;
1178 regs->regs[15] = infop->start_stack;
1179 }
1180
1181 /* See linux kernel: arch/sh/include/asm/elf.h. */
1182 #define ELF_NREG 23
1183 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1184
1185 /* See linux kernel: arch/sh/include/asm/ptrace.h. */
1186 enum {
1187 TARGET_REG_PC = 16,
1188 TARGET_REG_PR = 17,
1189 TARGET_REG_SR = 18,
1190 TARGET_REG_GBR = 19,
1191 TARGET_REG_MACH = 20,
1192 TARGET_REG_MACL = 21,
1193 TARGET_REG_SYSCALL = 22
1194 };
1195
1196 static inline void elf_core_copy_regs(target_elf_gregset_t *regs,
1197 const CPUSH4State *env)
1198 {
1199 int i;
1200
1201 for (i = 0; i < 16; i++) {
1202 (*regs)[i] = tswapreg(env->gregs[i]);
1203 }
1204
1205 (*regs)[TARGET_REG_PC] = tswapreg(env->pc);
1206 (*regs)[TARGET_REG_PR] = tswapreg(env->pr);
1207 (*regs)[TARGET_REG_SR] = tswapreg(env->sr);
1208 (*regs)[TARGET_REG_GBR] = tswapreg(env->gbr);
1209 (*regs)[TARGET_REG_MACH] = tswapreg(env->mach);
1210 (*regs)[TARGET_REG_MACL] = tswapreg(env->macl);
1211 (*regs)[TARGET_REG_SYSCALL] = 0; /* FIXME */
1212 }
1213
1214 #define USE_ELF_CORE_DUMP
1215 #define ELF_EXEC_PAGESIZE 4096
1216
1217 enum {
1218 SH_CPU_HAS_FPU = 0x0001, /* Hardware FPU support */
1219 SH_CPU_HAS_P2_FLUSH_BUG = 0x0002, /* Need to flush the cache in P2 area */
1220 SH_CPU_HAS_MMU_PAGE_ASSOC = 0x0004, /* SH3: TLB way selection bit support */
1221 SH_CPU_HAS_DSP = 0x0008, /* SH-DSP: DSP support */
1222 SH_CPU_HAS_PERF_COUNTER = 0x0010, /* Hardware performance counters */
1223 SH_CPU_HAS_PTEA = 0x0020, /* PTEA register */
1224 SH_CPU_HAS_LLSC = 0x0040, /* movli.l/movco.l */
1225 SH_CPU_HAS_L2_CACHE = 0x0080, /* Secondary cache / URAM */
1226 SH_CPU_HAS_OP32 = 0x0100, /* 32-bit instruction support */
1227 SH_CPU_HAS_PTEAEX = 0x0200, /* PTE ASID Extension support */
1228 };
1229
1230 #define ELF_HWCAP get_elf_hwcap()
1231
1232 static uint32_t get_elf_hwcap(void)
1233 {
1234 SuperHCPU *cpu = SUPERH_CPU(thread_cpu);
1235 uint32_t hwcap = 0;
1236
1237 hwcap |= SH_CPU_HAS_FPU;
1238
1239 if (cpu->env.features & SH_FEATURE_SH4A) {
1240 hwcap |= SH_CPU_HAS_LLSC;
1241 }
1242
1243 return hwcap;
1244 }
1245
1246 #endif
1247
1248 #ifdef TARGET_CRIS
1249
1250 #define ELF_START_MMAP 0x80000000
1251
1252 #define ELF_CLASS ELFCLASS32
1253 #define ELF_ARCH EM_CRIS
1254
1255 static inline void init_thread(struct target_pt_regs *regs,
1256 struct image_info *infop)
1257 {
1258 regs->erp = infop->entry;
1259 }
1260
1261 #define ELF_EXEC_PAGESIZE 8192
1262
1263 #endif
1264
1265 #ifdef TARGET_M68K
1266
1267 #define ELF_START_MMAP 0x80000000
1268
1269 #define ELF_CLASS ELFCLASS32
1270 #define ELF_ARCH EM_68K
1271
1272 /* ??? Does this need to do anything?
1273 #define ELF_PLAT_INIT(_r) */
1274
1275 static inline void init_thread(struct target_pt_regs *regs,
1276 struct image_info *infop)
1277 {
1278 regs->usp = infop->start_stack;
1279 regs->sr = 0;
1280 regs->pc = infop->entry;
1281 }
1282
1283 /* See linux kernel: arch/m68k/include/asm/elf.h. */
1284 #define ELF_NREG 20
1285 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1286
1287 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUM68KState *env)
1288 {
1289 (*regs)[0] = tswapreg(env->dregs[1]);
1290 (*regs)[1] = tswapreg(env->dregs[2]);
1291 (*regs)[2] = tswapreg(env->dregs[3]);
1292 (*regs)[3] = tswapreg(env->dregs[4]);
1293 (*regs)[4] = tswapreg(env->dregs[5]);
1294 (*regs)[5] = tswapreg(env->dregs[6]);
1295 (*regs)[6] = tswapreg(env->dregs[7]);
1296 (*regs)[7] = tswapreg(env->aregs[0]);
1297 (*regs)[8] = tswapreg(env->aregs[1]);
1298 (*regs)[9] = tswapreg(env->aregs[2]);
1299 (*regs)[10] = tswapreg(env->aregs[3]);
1300 (*regs)[11] = tswapreg(env->aregs[4]);
1301 (*regs)[12] = tswapreg(env->aregs[5]);
1302 (*regs)[13] = tswapreg(env->aregs[6]);
1303 (*regs)[14] = tswapreg(env->dregs[0]);
1304 (*regs)[15] = tswapreg(env->aregs[7]);
1305 (*regs)[16] = tswapreg(env->dregs[0]); /* FIXME: orig_d0 */
1306 (*regs)[17] = tswapreg(env->sr);
1307 (*regs)[18] = tswapreg(env->pc);
1308 (*regs)[19] = 0; /* FIXME: regs->format | regs->vector */
1309 }
1310
1311 #define USE_ELF_CORE_DUMP
1312 #define ELF_EXEC_PAGESIZE 8192
1313
1314 #endif
1315
1316 #ifdef TARGET_ALPHA
1317
1318 #define ELF_START_MMAP (0x30000000000ULL)
1319
1320 #define ELF_CLASS ELFCLASS64
1321 #define ELF_ARCH EM_ALPHA
1322
1323 static inline void init_thread(struct target_pt_regs *regs,
1324 struct image_info *infop)
1325 {
1326 regs->pc = infop->entry;
1327 regs->ps = 8;
1328 regs->usp = infop->start_stack;
1329 }
1330
1331 #define ELF_EXEC_PAGESIZE 8192
1332
1333 #endif /* TARGET_ALPHA */
1334
1335 #ifdef TARGET_S390X
1336
1337 #define ELF_START_MMAP (0x20000000000ULL)
1338
1339 #define ELF_CLASS ELFCLASS64
1340 #define ELF_DATA ELFDATA2MSB
1341 #define ELF_ARCH EM_S390
1342
1343 #include "elf.h"
1344
1345 #define ELF_HWCAP get_elf_hwcap()
1346
1347 #define GET_FEATURE(_feat, _hwcap) \
1348 do { if (s390_has_feat(_feat)) { hwcap |= _hwcap; } } while (0)
1349
1350 static uint32_t get_elf_hwcap(void)
1351 {
1352 /*
1353 * Let's assume we always have esan3 and zarch.
1354 * 31-bit processes can use 64-bit registers (high gprs).
1355 */
1356 uint32_t hwcap = HWCAP_S390_ESAN3 | HWCAP_S390_ZARCH | HWCAP_S390_HIGH_GPRS;
1357
1358 GET_FEATURE(S390_FEAT_STFLE, HWCAP_S390_STFLE);
1359 GET_FEATURE(S390_FEAT_MSA, HWCAP_S390_MSA);
1360 GET_FEATURE(S390_FEAT_LONG_DISPLACEMENT, HWCAP_S390_LDISP);
1361 GET_FEATURE(S390_FEAT_EXTENDED_IMMEDIATE, HWCAP_S390_EIMM);
1362 if (s390_has_feat(S390_FEAT_EXTENDED_TRANSLATION_3) &&
1363 s390_has_feat(S390_FEAT_ETF3_ENH)) {
1364 hwcap |= HWCAP_S390_ETF3EH;
1365 }
1366 GET_FEATURE(S390_FEAT_VECTOR, HWCAP_S390_VXRS);
1367
1368 return hwcap;
1369 }
1370
1371 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop)
1372 {
1373 regs->psw.addr = infop->entry;
1374 regs->psw.mask = PSW_MASK_64 | PSW_MASK_32;
1375 regs->gprs[15] = infop->start_stack;
1376 }
1377
1378 /* See linux kernel: arch/s390/include/uapi/asm/ptrace.h (s390_regs). */
1379 #define ELF_NREG 27
1380 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1381
1382 enum {
1383 TARGET_REG_PSWM = 0,
1384 TARGET_REG_PSWA = 1,
1385 TARGET_REG_GPRS = 2,
1386 TARGET_REG_ARS = 18,
1387 TARGET_REG_ORIG_R2 = 26,
1388 };
1389
1390 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1391 const CPUS390XState *env)
1392 {
1393 int i;
1394 uint32_t *aregs;
1395
1396 (*regs)[TARGET_REG_PSWM] = tswapreg(env->psw.mask);
1397 (*regs)[TARGET_REG_PSWA] = tswapreg(env->psw.addr);
1398 for (i = 0; i < 16; i++) {
1399 (*regs)[TARGET_REG_GPRS + i] = tswapreg(env->regs[i]);
1400 }
1401 aregs = (uint32_t *)&((*regs)[TARGET_REG_ARS]);
1402 for (i = 0; i < 16; i++) {
1403 aregs[i] = tswap32(env->aregs[i]);
1404 }
1405 (*regs)[TARGET_REG_ORIG_R2] = 0;
1406 }
1407
1408 #define USE_ELF_CORE_DUMP
1409 #define ELF_EXEC_PAGESIZE 4096
1410
1411 #endif /* TARGET_S390X */
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 #ifdef TARGET_HEXAGON
1518
1519 #define ELF_START_MMAP 0x20000000
1520
1521 #define ELF_CLASS ELFCLASS32
1522 #define ELF_ARCH EM_HEXAGON
1523
1524 static inline void init_thread(struct target_pt_regs *regs,
1525 struct image_info *infop)
1526 {
1527 regs->sepc = infop->entry;
1528 regs->sp = infop->start_stack;
1529 }
1530
1531 #endif /* TARGET_HEXAGON */
1532
1533 #ifndef ELF_PLATFORM
1534 #define ELF_PLATFORM (NULL)
1535 #endif
1536
1537 #ifndef ELF_MACHINE
1538 #define ELF_MACHINE ELF_ARCH
1539 #endif
1540
1541 #ifndef elf_check_arch
1542 #define elf_check_arch(x) ((x) == ELF_ARCH)
1543 #endif
1544
1545 #ifndef elf_check_abi
1546 #define elf_check_abi(x) (1)
1547 #endif
1548
1549 #ifndef ELF_HWCAP
1550 #define ELF_HWCAP 0
1551 #endif
1552
1553 #ifndef STACK_GROWS_DOWN
1554 #define STACK_GROWS_DOWN 1
1555 #endif
1556
1557 #ifndef STACK_ALIGNMENT
1558 #define STACK_ALIGNMENT 16
1559 #endif
1560
1561 #ifdef TARGET_ABI32
1562 #undef ELF_CLASS
1563 #define ELF_CLASS ELFCLASS32
1564 #undef bswaptls
1565 #define bswaptls(ptr) bswap32s(ptr)
1566 #endif
1567
1568 #include "elf.h"
1569
1570 /* We must delay the following stanzas until after "elf.h". */
1571 #if defined(TARGET_AARCH64)
1572
1573 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz,
1574 const uint32_t *data,
1575 struct image_info *info,
1576 Error **errp)
1577 {
1578 if (pr_type == GNU_PROPERTY_AARCH64_FEATURE_1_AND) {
1579 if (pr_datasz != sizeof(uint32_t)) {
1580 error_setg(errp, "Ill-formed GNU_PROPERTY_AARCH64_FEATURE_1_AND");
1581 return false;
1582 }
1583 /* We will extract GNU_PROPERTY_AARCH64_FEATURE_1_BTI later. */
1584 info->note_flags = *data;
1585 }
1586 return true;
1587 }
1588 #define ARCH_USE_GNU_PROPERTY 1
1589
1590 #else
1591
1592 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz,
1593 const uint32_t *data,
1594 struct image_info *info,
1595 Error **errp)
1596 {
1597 g_assert_not_reached();
1598 }
1599 #define ARCH_USE_GNU_PROPERTY 0
1600
1601 #endif
1602
1603 struct exec
1604 {
1605 unsigned int a_info; /* Use macros N_MAGIC, etc for access */
1606 unsigned int a_text; /* length of text, in bytes */
1607 unsigned int a_data; /* length of data, in bytes */
1608 unsigned int a_bss; /* length of uninitialized data area, in bytes */
1609 unsigned int a_syms; /* length of symbol table data in file, in bytes */
1610 unsigned int a_entry; /* start address */
1611 unsigned int a_trsize; /* length of relocation info for text, in bytes */
1612 unsigned int a_drsize; /* length of relocation info for data, in bytes */
1613 };
1614
1615
1616 #define N_MAGIC(exec) ((exec).a_info & 0xffff)
1617 #define OMAGIC 0407
1618 #define NMAGIC 0410
1619 #define ZMAGIC 0413
1620 #define QMAGIC 0314
1621
1622 /* Necessary parameters */
1623 #define TARGET_ELF_EXEC_PAGESIZE \
1624 (((eppnt->p_align & ~qemu_host_page_mask) != 0) ? \
1625 TARGET_PAGE_SIZE : MAX(qemu_host_page_size, TARGET_PAGE_SIZE))
1626 #define TARGET_ELF_PAGELENGTH(_v) ROUND_UP((_v), TARGET_ELF_EXEC_PAGESIZE)
1627 #define TARGET_ELF_PAGESTART(_v) ((_v) & \
1628 ~(abi_ulong)(TARGET_ELF_EXEC_PAGESIZE-1))
1629 #define TARGET_ELF_PAGEOFFSET(_v) ((_v) & (TARGET_ELF_EXEC_PAGESIZE-1))
1630
1631 #define DLINFO_ITEMS 16
1632
1633 static inline void memcpy_fromfs(void * to, const void * from, unsigned long n)
1634 {
1635 memcpy(to, from, n);
1636 }
1637
1638 #ifdef BSWAP_NEEDED
1639 static void bswap_ehdr(struct elfhdr *ehdr)
1640 {
1641 bswap16s(&ehdr->e_type); /* Object file type */
1642 bswap16s(&ehdr->e_machine); /* Architecture */
1643 bswap32s(&ehdr->e_version); /* Object file version */
1644 bswaptls(&ehdr->e_entry); /* Entry point virtual address */
1645 bswaptls(&ehdr->e_phoff); /* Program header table file offset */
1646 bswaptls(&ehdr->e_shoff); /* Section header table file offset */
1647 bswap32s(&ehdr->e_flags); /* Processor-specific flags */
1648 bswap16s(&ehdr->e_ehsize); /* ELF header size in bytes */
1649 bswap16s(&ehdr->e_phentsize); /* Program header table entry size */
1650 bswap16s(&ehdr->e_phnum); /* Program header table entry count */
1651 bswap16s(&ehdr->e_shentsize); /* Section header table entry size */
1652 bswap16s(&ehdr->e_shnum); /* Section header table entry count */
1653 bswap16s(&ehdr->e_shstrndx); /* Section header string table index */
1654 }
1655
1656 static void bswap_phdr(struct elf_phdr *phdr, int phnum)
1657 {
1658 int i;
1659 for (i = 0; i < phnum; ++i, ++phdr) {
1660 bswap32s(&phdr->p_type); /* Segment type */
1661 bswap32s(&phdr->p_flags); /* Segment flags */
1662 bswaptls(&phdr->p_offset); /* Segment file offset */
1663 bswaptls(&phdr->p_vaddr); /* Segment virtual address */
1664 bswaptls(&phdr->p_paddr); /* Segment physical address */
1665 bswaptls(&phdr->p_filesz); /* Segment size in file */
1666 bswaptls(&phdr->p_memsz); /* Segment size in memory */
1667 bswaptls(&phdr->p_align); /* Segment alignment */
1668 }
1669 }
1670
1671 static void bswap_shdr(struct elf_shdr *shdr, int shnum)
1672 {
1673 int i;
1674 for (i = 0; i < shnum; ++i, ++shdr) {
1675 bswap32s(&shdr->sh_name);
1676 bswap32s(&shdr->sh_type);
1677 bswaptls(&shdr->sh_flags);
1678 bswaptls(&shdr->sh_addr);
1679 bswaptls(&shdr->sh_offset);
1680 bswaptls(&shdr->sh_size);
1681 bswap32s(&shdr->sh_link);
1682 bswap32s(&shdr->sh_info);
1683 bswaptls(&shdr->sh_addralign);
1684 bswaptls(&shdr->sh_entsize);
1685 }
1686 }
1687
1688 static void bswap_sym(struct elf_sym *sym)
1689 {
1690 bswap32s(&sym->st_name);
1691 bswaptls(&sym->st_value);
1692 bswaptls(&sym->st_size);
1693 bswap16s(&sym->st_shndx);
1694 }
1695
1696 #ifdef TARGET_MIPS
1697 static void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags)
1698 {
1699 bswap16s(&abiflags->version);
1700 bswap32s(&abiflags->ases);
1701 bswap32s(&abiflags->isa_ext);
1702 bswap32s(&abiflags->flags1);
1703 bswap32s(&abiflags->flags2);
1704 }
1705 #endif
1706 #else
1707 static inline void bswap_ehdr(struct elfhdr *ehdr) { }
1708 static inline void bswap_phdr(struct elf_phdr *phdr, int phnum) { }
1709 static inline void bswap_shdr(struct elf_shdr *shdr, int shnum) { }
1710 static inline void bswap_sym(struct elf_sym *sym) { }
1711 #ifdef TARGET_MIPS
1712 static inline void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags) { }
1713 #endif
1714 #endif
1715
1716 #ifdef USE_ELF_CORE_DUMP
1717 static int elf_core_dump(int, const CPUArchState *);
1718 #endif /* USE_ELF_CORE_DUMP */
1719 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias);
1720
1721 /* Verify the portions of EHDR within E_IDENT for the target.
1722 This can be performed before bswapping the entire header. */
1723 static bool elf_check_ident(struct elfhdr *ehdr)
1724 {
1725 return (ehdr->e_ident[EI_MAG0] == ELFMAG0
1726 && ehdr->e_ident[EI_MAG1] == ELFMAG1
1727 && ehdr->e_ident[EI_MAG2] == ELFMAG2
1728 && ehdr->e_ident[EI_MAG3] == ELFMAG3
1729 && ehdr->e_ident[EI_CLASS] == ELF_CLASS
1730 && ehdr->e_ident[EI_DATA] == ELF_DATA
1731 && ehdr->e_ident[EI_VERSION] == EV_CURRENT);
1732 }
1733
1734 /* Verify the portions of EHDR outside of E_IDENT for the target.
1735 This has to wait until after bswapping the header. */
1736 static bool elf_check_ehdr(struct elfhdr *ehdr)
1737 {
1738 return (elf_check_arch(ehdr->e_machine)
1739 && elf_check_abi(ehdr->e_flags)
1740 && ehdr->e_ehsize == sizeof(struct elfhdr)
1741 && ehdr->e_phentsize == sizeof(struct elf_phdr)
1742 && (ehdr->e_type == ET_EXEC || ehdr->e_type == ET_DYN));
1743 }
1744
1745 /*
1746 * 'copy_elf_strings()' copies argument/envelope strings from user
1747 * memory to free pages in kernel mem. These are in a format ready
1748 * to be put directly into the top of new user memory.
1749 *
1750 */
1751 static abi_ulong copy_elf_strings(int argc, char **argv, char *scratch,
1752 abi_ulong p, abi_ulong stack_limit)
1753 {
1754 char *tmp;
1755 int len, i;
1756 abi_ulong top = p;
1757
1758 if (!p) {
1759 return 0; /* bullet-proofing */
1760 }
1761
1762 if (STACK_GROWS_DOWN) {
1763 int offset = ((p - 1) % TARGET_PAGE_SIZE) + 1;
1764 for (i = argc - 1; i >= 0; --i) {
1765 tmp = argv[i];
1766 if (!tmp) {
1767 fprintf(stderr, "VFS: argc is wrong");
1768 exit(-1);
1769 }
1770 len = strlen(tmp) + 1;
1771 tmp += len;
1772
1773 if (len > (p - stack_limit)) {
1774 return 0;
1775 }
1776 while (len) {
1777 int bytes_to_copy = (len > offset) ? offset : len;
1778 tmp -= bytes_to_copy;
1779 p -= bytes_to_copy;
1780 offset -= bytes_to_copy;
1781 len -= bytes_to_copy;
1782
1783 memcpy_fromfs(scratch + offset, tmp, bytes_to_copy);
1784
1785 if (offset == 0) {
1786 memcpy_to_target(p, scratch, top - p);
1787 top = p;
1788 offset = TARGET_PAGE_SIZE;
1789 }
1790 }
1791 }
1792 if (p != top) {
1793 memcpy_to_target(p, scratch + offset, top - p);
1794 }
1795 } else {
1796 int remaining = TARGET_PAGE_SIZE - (p % TARGET_PAGE_SIZE);
1797 for (i = 0; i < argc; ++i) {
1798 tmp = argv[i];
1799 if (!tmp) {
1800 fprintf(stderr, "VFS: argc is wrong");
1801 exit(-1);
1802 }
1803 len = strlen(tmp) + 1;
1804 if (len > (stack_limit - p)) {
1805 return 0;
1806 }
1807 while (len) {
1808 int bytes_to_copy = (len > remaining) ? remaining : len;
1809
1810 memcpy_fromfs(scratch + (p - top), tmp, bytes_to_copy);
1811
1812 tmp += bytes_to_copy;
1813 remaining -= bytes_to_copy;
1814 p += bytes_to_copy;
1815 len -= bytes_to_copy;
1816
1817 if (remaining == 0) {
1818 memcpy_to_target(top, scratch, p - top);
1819 top = p;
1820 remaining = TARGET_PAGE_SIZE;
1821 }
1822 }
1823 }
1824 if (p != top) {
1825 memcpy_to_target(top, scratch, p - top);
1826 }
1827 }
1828
1829 return p;
1830 }
1831
1832 /* Older linux kernels provide up to MAX_ARG_PAGES (default: 32) of
1833 * argument/environment space. Newer kernels (>2.6.33) allow more,
1834 * dependent on stack size, but guarantee at least 32 pages for
1835 * backwards compatibility.
1836 */
1837 #define STACK_LOWER_LIMIT (32 * TARGET_PAGE_SIZE)
1838
1839 static abi_ulong setup_arg_pages(struct linux_binprm *bprm,
1840 struct image_info *info)
1841 {
1842 abi_ulong size, error, guard;
1843
1844 size = guest_stack_size;
1845 if (size < STACK_LOWER_LIMIT) {
1846 size = STACK_LOWER_LIMIT;
1847 }
1848 guard = TARGET_PAGE_SIZE;
1849 if (guard < qemu_real_host_page_size) {
1850 guard = qemu_real_host_page_size;
1851 }
1852
1853 error = target_mmap(0, size + guard, PROT_READ | PROT_WRITE,
1854 MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
1855 if (error == -1) {
1856 perror("mmap stack");
1857 exit(-1);
1858 }
1859
1860 /* We reserve one extra page at the top of the stack as guard. */
1861 if (STACK_GROWS_DOWN) {
1862 target_mprotect(error, guard, PROT_NONE);
1863 info->stack_limit = error + guard;
1864 return info->stack_limit + size - sizeof(void *);
1865 } else {
1866 target_mprotect(error + size, guard, PROT_NONE);
1867 info->stack_limit = error + size;
1868 return error;
1869 }
1870 }
1871
1872 /* Map and zero the bss. We need to explicitly zero any fractional pages
1873 after the data section (i.e. bss). */
1874 static void zero_bss(abi_ulong elf_bss, abi_ulong last_bss, int prot)
1875 {
1876 uintptr_t host_start, host_map_start, host_end;
1877
1878 last_bss = TARGET_PAGE_ALIGN(last_bss);
1879
1880 /* ??? There is confusion between qemu_real_host_page_size and
1881 qemu_host_page_size here and elsewhere in target_mmap, which
1882 may lead to the end of the data section mapping from the file
1883 not being mapped. At least there was an explicit test and
1884 comment for that here, suggesting that "the file size must
1885 be known". The comment probably pre-dates the introduction
1886 of the fstat system call in target_mmap which does in fact
1887 find out the size. What isn't clear is if the workaround
1888 here is still actually needed. For now, continue with it,
1889 but merge it with the "normal" mmap that would allocate the bss. */
1890
1891 host_start = (uintptr_t) g2h_untagged(elf_bss);
1892 host_end = (uintptr_t) g2h_untagged(last_bss);
1893 host_map_start = REAL_HOST_PAGE_ALIGN(host_start);
1894
1895 if (host_map_start < host_end) {
1896 void *p = mmap((void *)host_map_start, host_end - host_map_start,
1897 prot, MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
1898 if (p == MAP_FAILED) {
1899 perror("cannot mmap brk");
1900 exit(-1);
1901 }
1902 }
1903
1904 /* Ensure that the bss page(s) are valid */
1905 if ((page_get_flags(last_bss-1) & prot) != prot) {
1906 page_set_flags(elf_bss & TARGET_PAGE_MASK, last_bss, prot | PAGE_VALID);
1907 }
1908
1909 if (host_start < host_map_start) {
1910 memset((void *)host_start, 0, host_map_start - host_start);
1911 }
1912 }
1913
1914 #ifdef TARGET_ARM
1915 static int elf_is_fdpic(struct elfhdr *exec)
1916 {
1917 return exec->e_ident[EI_OSABI] == ELFOSABI_ARM_FDPIC;
1918 }
1919 #else
1920 /* Default implementation, always false. */
1921 static int elf_is_fdpic(struct elfhdr *exec)
1922 {
1923 return 0;
1924 }
1925 #endif
1926
1927 static abi_ulong loader_build_fdpic_loadmap(struct image_info *info, abi_ulong sp)
1928 {
1929 uint16_t n;
1930 struct elf32_fdpic_loadseg *loadsegs = info->loadsegs;
1931
1932 /* elf32_fdpic_loadseg */
1933 n = info->nsegs;
1934 while (n--) {
1935 sp -= 12;
1936 put_user_u32(loadsegs[n].addr, sp+0);
1937 put_user_u32(loadsegs[n].p_vaddr, sp+4);
1938 put_user_u32(loadsegs[n].p_memsz, sp+8);
1939 }
1940
1941 /* elf32_fdpic_loadmap */
1942 sp -= 4;
1943 put_user_u16(0, sp+0); /* version */
1944 put_user_u16(info->nsegs, sp+2); /* nsegs */
1945
1946 info->personality = PER_LINUX_FDPIC;
1947 info->loadmap_addr = sp;
1948
1949 return sp;
1950 }
1951
1952 static abi_ulong create_elf_tables(abi_ulong p, int argc, int envc,
1953 struct elfhdr *exec,
1954 struct image_info *info,
1955 struct image_info *interp_info)
1956 {
1957 abi_ulong sp;
1958 abi_ulong u_argc, u_argv, u_envp, u_auxv;
1959 int size;
1960 int i;
1961 abi_ulong u_rand_bytes;
1962 uint8_t k_rand_bytes[16];
1963 abi_ulong u_platform;
1964 const char *k_platform;
1965 const int n = sizeof(elf_addr_t);
1966
1967 sp = p;
1968
1969 /* Needs to be before we load the env/argc/... */
1970 if (elf_is_fdpic(exec)) {
1971 /* Need 4 byte alignment for these structs */
1972 sp &= ~3;
1973 sp = loader_build_fdpic_loadmap(info, sp);
1974 info->other_info = interp_info;
1975 if (interp_info) {
1976 interp_info->other_info = info;
1977 sp = loader_build_fdpic_loadmap(interp_info, sp);
1978 info->interpreter_loadmap_addr = interp_info->loadmap_addr;
1979 info->interpreter_pt_dynamic_addr = interp_info->pt_dynamic_addr;
1980 } else {
1981 info->interpreter_loadmap_addr = 0;
1982 info->interpreter_pt_dynamic_addr = 0;
1983 }
1984 }
1985
1986 u_platform = 0;
1987 k_platform = ELF_PLATFORM;
1988 if (k_platform) {
1989 size_t len = strlen(k_platform) + 1;
1990 if (STACK_GROWS_DOWN) {
1991 sp -= (len + n - 1) & ~(n - 1);
1992 u_platform = sp;
1993 /* FIXME - check return value of memcpy_to_target() for failure */
1994 memcpy_to_target(sp, k_platform, len);
1995 } else {
1996 memcpy_to_target(sp, k_platform, len);
1997 u_platform = sp;
1998 sp += len + 1;
1999 }
2000 }
2001
2002 /* Provide 16 byte alignment for the PRNG, and basic alignment for
2003 * the argv and envp pointers.
2004 */
2005 if (STACK_GROWS_DOWN) {
2006 sp = QEMU_ALIGN_DOWN(sp, 16);
2007 } else {
2008 sp = QEMU_ALIGN_UP(sp, 16);
2009 }
2010
2011 /*
2012 * Generate 16 random bytes for userspace PRNG seeding.
2013 */
2014 qemu_guest_getrandom_nofail(k_rand_bytes, sizeof(k_rand_bytes));
2015 if (STACK_GROWS_DOWN) {
2016 sp -= 16;
2017 u_rand_bytes = sp;
2018 /* FIXME - check return value of memcpy_to_target() for failure */
2019 memcpy_to_target(sp, k_rand_bytes, 16);
2020 } else {
2021 memcpy_to_target(sp, k_rand_bytes, 16);
2022 u_rand_bytes = sp;
2023 sp += 16;
2024 }
2025
2026 size = (DLINFO_ITEMS + 1) * 2;
2027 if (k_platform)
2028 size += 2;
2029 #ifdef DLINFO_ARCH_ITEMS
2030 size += DLINFO_ARCH_ITEMS * 2;
2031 #endif
2032 #ifdef ELF_HWCAP2
2033 size += 2;
2034 #endif
2035 info->auxv_len = size * n;
2036
2037 size += envc + argc + 2;
2038 size += 1; /* argc itself */
2039 size *= n;
2040
2041 /* Allocate space and finalize stack alignment for entry now. */
2042 if (STACK_GROWS_DOWN) {
2043 u_argc = QEMU_ALIGN_DOWN(sp - size, STACK_ALIGNMENT);
2044 sp = u_argc;
2045 } else {
2046 u_argc = sp;
2047 sp = QEMU_ALIGN_UP(sp + size, STACK_ALIGNMENT);
2048 }
2049
2050 u_argv = u_argc + n;
2051 u_envp = u_argv + (argc + 1) * n;
2052 u_auxv = u_envp + (envc + 1) * n;
2053 info->saved_auxv = u_auxv;
2054 info->arg_start = u_argv;
2055 info->arg_end = u_argv + argc * n;
2056
2057 /* This is correct because Linux defines
2058 * elf_addr_t as Elf32_Off / Elf64_Off
2059 */
2060 #define NEW_AUX_ENT(id, val) do { \
2061 put_user_ual(id, u_auxv); u_auxv += n; \
2062 put_user_ual(val, u_auxv); u_auxv += n; \
2063 } while(0)
2064
2065 #ifdef ARCH_DLINFO
2066 /*
2067 * ARCH_DLINFO must come first so platform specific code can enforce
2068 * special alignment requirements on the AUXV if necessary (eg. PPC).
2069 */
2070 ARCH_DLINFO;
2071 #endif
2072 /* There must be exactly DLINFO_ITEMS entries here, or the assert
2073 * on info->auxv_len will trigger.
2074 */
2075 NEW_AUX_ENT(AT_PHDR, (abi_ulong)(info->load_addr + exec->e_phoff));
2076 NEW_AUX_ENT(AT_PHENT, (abi_ulong)(sizeof (struct elf_phdr)));
2077 NEW_AUX_ENT(AT_PHNUM, (abi_ulong)(exec->e_phnum));
2078 if ((info->alignment & ~qemu_host_page_mask) != 0) {
2079 /* Target doesn't support host page size alignment */
2080 NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(TARGET_PAGE_SIZE));
2081 } else {
2082 NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(MAX(TARGET_PAGE_SIZE,
2083 qemu_host_page_size)));
2084 }
2085 NEW_AUX_ENT(AT_BASE, (abi_ulong)(interp_info ? interp_info->load_addr : 0));
2086 NEW_AUX_ENT(AT_FLAGS, (abi_ulong)0);
2087 NEW_AUX_ENT(AT_ENTRY, info->entry);
2088 NEW_AUX_ENT(AT_UID, (abi_ulong) getuid());
2089 NEW_AUX_ENT(AT_EUID, (abi_ulong) geteuid());
2090 NEW_AUX_ENT(AT_GID, (abi_ulong) getgid());
2091 NEW_AUX_ENT(AT_EGID, (abi_ulong) getegid());
2092 NEW_AUX_ENT(AT_HWCAP, (abi_ulong) ELF_HWCAP);
2093 NEW_AUX_ENT(AT_CLKTCK, (abi_ulong) sysconf(_SC_CLK_TCK));
2094 NEW_AUX_ENT(AT_RANDOM, (abi_ulong) u_rand_bytes);
2095 NEW_AUX_ENT(AT_SECURE, (abi_ulong) qemu_getauxval(AT_SECURE));
2096 NEW_AUX_ENT(AT_EXECFN, info->file_string);
2097
2098 #ifdef ELF_HWCAP2
2099 NEW_AUX_ENT(AT_HWCAP2, (abi_ulong) ELF_HWCAP2);
2100 #endif
2101
2102 if (u_platform) {
2103 NEW_AUX_ENT(AT_PLATFORM, u_platform);
2104 }
2105 NEW_AUX_ENT (AT_NULL, 0);
2106 #undef NEW_AUX_ENT
2107
2108 /* Check that our initial calculation of the auxv length matches how much
2109 * we actually put into it.
2110 */
2111 assert(info->auxv_len == u_auxv - info->saved_auxv);
2112
2113 put_user_ual(argc, u_argc);
2114
2115 p = info->arg_strings;
2116 for (i = 0; i < argc; ++i) {
2117 put_user_ual(p, u_argv);
2118 u_argv += n;
2119 p += target_strlen(p) + 1;
2120 }
2121 put_user_ual(0, u_argv);
2122
2123 p = info->env_strings;
2124 for (i = 0; i < envc; ++i) {
2125 put_user_ual(p, u_envp);
2126 u_envp += n;
2127 p += target_strlen(p) + 1;
2128 }
2129 put_user_ual(0, u_envp);
2130
2131 return sp;
2132 }
2133
2134 #ifndef ARM_COMMPAGE
2135 #define ARM_COMMPAGE 0
2136 #define init_guest_commpage() true
2137 #endif
2138
2139 static void pgb_fail_in_use(const char *image_name)
2140 {
2141 error_report("%s: requires virtual address space that is in use "
2142 "(omit the -B option or choose a different value)",
2143 image_name);
2144 exit(EXIT_FAILURE);
2145 }
2146
2147 static void pgb_have_guest_base(const char *image_name, abi_ulong guest_loaddr,
2148 abi_ulong guest_hiaddr, long align)
2149 {
2150 const int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE;
2151 void *addr, *test;
2152
2153 if (!QEMU_IS_ALIGNED(guest_base, align)) {
2154 fprintf(stderr, "Requested guest base %p does not satisfy "
2155 "host minimum alignment (0x%lx)\n",
2156 (void *)guest_base, align);
2157 exit(EXIT_FAILURE);
2158 }
2159
2160 /* Sanity check the guest binary. */
2161 if (reserved_va) {
2162 if (guest_hiaddr > reserved_va) {
2163 error_report("%s: requires more than reserved virtual "
2164 "address space (0x%" PRIx64 " > 0x%lx)",
2165 image_name, (uint64_t)guest_hiaddr, reserved_va);
2166 exit(EXIT_FAILURE);
2167 }
2168 } else {
2169 #if HOST_LONG_BITS < TARGET_ABI_BITS
2170 if ((guest_hiaddr - guest_base) > ~(uintptr_t)0) {
2171 error_report("%s: requires more virtual address space "
2172 "than the host can provide (0x%" PRIx64 ")",
2173 image_name, (uint64_t)guest_hiaddr - guest_base);
2174 exit(EXIT_FAILURE);
2175 }
2176 #endif
2177 }
2178
2179 /*
2180 * Expand the allocation to the entire reserved_va.
2181 * Exclude the mmap_min_addr hole.
2182 */
2183 if (reserved_va) {
2184 guest_loaddr = (guest_base >= mmap_min_addr ? 0
2185 : mmap_min_addr - guest_base);
2186 guest_hiaddr = reserved_va;
2187 }
2188
2189 /* Reserve the address space for the binary, or reserved_va. */
2190 test = g2h_untagged(guest_loaddr);
2191 addr = mmap(test, guest_hiaddr - guest_loaddr, PROT_NONE, flags, -1, 0);
2192 if (test != addr) {
2193 pgb_fail_in_use(image_name);
2194 }
2195 }
2196
2197 /**
2198 * pgd_find_hole_fallback: potential mmap address
2199 * @guest_size: size of available space
2200 * @brk: location of break
2201 * @align: memory alignment
2202 *
2203 * This is a fallback method for finding a hole in the host address
2204 * space if we don't have the benefit of being able to access
2205 * /proc/self/map. It can potentially take a very long time as we can
2206 * only dumbly iterate up the host address space seeing if the
2207 * allocation would work.
2208 */
2209 static uintptr_t pgd_find_hole_fallback(uintptr_t guest_size, uintptr_t brk,
2210 long align, uintptr_t offset)
2211 {
2212 uintptr_t base;
2213
2214 /* Start (aligned) at the bottom and work our way up */
2215 base = ROUND_UP(mmap_min_addr, align);
2216
2217 while (true) {
2218 uintptr_t align_start, end;
2219 align_start = ROUND_UP(base, align);
2220 end = align_start + guest_size + offset;
2221
2222 /* if brk is anywhere in the range give ourselves some room to grow. */
2223 if (align_start <= brk && brk < end) {
2224 base = brk + (16 * MiB);
2225 continue;
2226 } else if (align_start + guest_size < align_start) {
2227 /* we have run out of space */
2228 return -1;
2229 } else {
2230 int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE |
2231 MAP_FIXED_NOREPLACE;
2232 void * mmap_start = mmap((void *) align_start, guest_size,
2233 PROT_NONE, flags, -1, 0);
2234 if (mmap_start != MAP_FAILED) {
2235 munmap(mmap_start, guest_size);
2236 if (mmap_start == (void *) align_start) {
2237 return (uintptr_t) mmap_start + offset;
2238 }
2239 }
2240 base += qemu_host_page_size;
2241 }
2242 }
2243 }
2244
2245 /* Return value for guest_base, or -1 if no hole found. */
2246 static uintptr_t pgb_find_hole(uintptr_t guest_loaddr, uintptr_t guest_size,
2247 long align, uintptr_t offset)
2248 {
2249 GSList *maps, *iter;
2250 uintptr_t this_start, this_end, next_start, brk;
2251 intptr_t ret = -1;
2252
2253 assert(QEMU_IS_ALIGNED(guest_loaddr, align));
2254
2255 maps = read_self_maps();
2256
2257 /* Read brk after we've read the maps, which will malloc. */
2258 brk = (uintptr_t)sbrk(0);
2259
2260 if (!maps) {
2261 ret = pgd_find_hole_fallback(guest_size, brk, align, offset);
2262 return ret == -1 ? -1 : ret - guest_loaddr;
2263 }
2264
2265 /* The first hole is before the first map entry. */
2266 this_start = mmap_min_addr;
2267
2268 for (iter = maps; iter;
2269 this_start = next_start, iter = g_slist_next(iter)) {
2270 uintptr_t align_start, hole_size;
2271
2272 this_end = ((MapInfo *)iter->data)->start;
2273 next_start = ((MapInfo *)iter->data)->end;
2274 align_start = ROUND_UP(this_start + offset, align);
2275
2276 /* Skip holes that are too small. */
2277 if (align_start >= this_end) {
2278 continue;
2279 }
2280 hole_size = this_end - align_start;
2281 if (hole_size < guest_size) {
2282 continue;
2283 }
2284
2285 /* If this hole contains brk, give ourselves some room to grow. */
2286 if (this_start <= brk && brk < this_end) {
2287 hole_size -= guest_size;
2288 if (sizeof(uintptr_t) == 8 && hole_size >= 1 * GiB) {
2289 align_start += 1 * GiB;
2290 } else if (hole_size >= 16 * MiB) {
2291 align_start += 16 * MiB;
2292 } else {
2293 align_start = (this_end - guest_size) & -align;
2294 if (align_start < this_start) {
2295 continue;
2296 }
2297 }
2298 }
2299
2300 /* Record the lowest successful match. */
2301 if (ret < 0) {
2302 ret = align_start - guest_loaddr;
2303 }
2304 /* If this hole contains the identity map, select it. */
2305 if (align_start <= guest_loaddr &&
2306 guest_loaddr + guest_size <= this_end) {
2307 ret = 0;
2308 }
2309 /* If this hole ends above the identity map, stop looking. */
2310 if (this_end >= guest_loaddr) {
2311 break;
2312 }
2313 }
2314 free_self_maps(maps);
2315
2316 return ret;
2317 }
2318
2319 static void pgb_static(const char *image_name, abi_ulong orig_loaddr,
2320 abi_ulong orig_hiaddr, long align)
2321 {
2322 uintptr_t loaddr = orig_loaddr;
2323 uintptr_t hiaddr = orig_hiaddr;
2324 uintptr_t offset = 0;
2325 uintptr_t addr;
2326
2327 if (hiaddr != orig_hiaddr) {
2328 error_report("%s: requires virtual address space that the "
2329 "host cannot provide (0x%" PRIx64 ")",
2330 image_name, (uint64_t)orig_hiaddr);
2331 exit(EXIT_FAILURE);
2332 }
2333
2334 loaddr &= -align;
2335 if (ARM_COMMPAGE) {
2336 /*
2337 * Extend the allocation to include the commpage.
2338 * For a 64-bit host, this is just 4GiB; for a 32-bit host we
2339 * need to ensure there is space bellow the guest_base so we
2340 * can map the commpage in the place needed when the address
2341 * arithmetic wraps around.
2342 */
2343 if (sizeof(uintptr_t) == 8 || loaddr >= 0x80000000u) {
2344 hiaddr = (uintptr_t) 4 << 30;
2345 } else {
2346 offset = -(ARM_COMMPAGE & -align);
2347 }
2348 }
2349
2350 addr = pgb_find_hole(loaddr, hiaddr - loaddr, align, offset);
2351 if (addr == -1) {
2352 /*
2353 * If ARM_COMMPAGE, there *might* be a non-consecutive allocation
2354 * that can satisfy both. But as the normal arm32 link base address
2355 * is ~32k, and we extend down to include the commpage, making the
2356 * overhead only ~96k, this is unlikely.
2357 */
2358 error_report("%s: Unable to allocate %#zx bytes of "
2359 "virtual address space", image_name,
2360 (size_t)(hiaddr - loaddr));
2361 exit(EXIT_FAILURE);
2362 }
2363
2364 guest_base = addr;
2365 }
2366
2367 static void pgb_dynamic(const char *image_name, long align)
2368 {
2369 /*
2370 * The executable is dynamic and does not require a fixed address.
2371 * All we need is a commpage that satisfies align.
2372 * If we do not need a commpage, leave guest_base == 0.
2373 */
2374 if (ARM_COMMPAGE) {
2375 uintptr_t addr, commpage;
2376
2377 /* 64-bit hosts should have used reserved_va. */
2378 assert(sizeof(uintptr_t) == 4);
2379
2380 /*
2381 * By putting the commpage at the first hole, that puts guest_base
2382 * just above that, and maximises the positive guest addresses.
2383 */
2384 commpage = ARM_COMMPAGE & -align;
2385 addr = pgb_find_hole(commpage, -commpage, align, 0);
2386 assert(addr != -1);
2387 guest_base = addr;
2388 }
2389 }
2390
2391 static void pgb_reserved_va(const char *image_name, abi_ulong guest_loaddr,
2392 abi_ulong guest_hiaddr, long align)
2393 {
2394 int flags = MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE;
2395 void *addr, *test;
2396
2397 if (guest_hiaddr > reserved_va) {
2398 error_report("%s: requires more than reserved virtual "
2399 "address space (0x%" PRIx64 " > 0x%lx)",
2400 image_name, (uint64_t)guest_hiaddr, reserved_va);
2401 exit(EXIT_FAILURE);
2402 }
2403
2404 /* Widen the "image" to the entire reserved address space. */
2405 pgb_static(image_name, 0, reserved_va, align);
2406
2407 /* osdep.h defines this as 0 if it's missing */
2408 flags |= MAP_FIXED_NOREPLACE;
2409
2410 /* Reserve the memory on the host. */
2411 assert(guest_base != 0);
2412 test = g2h_untagged(0);
2413 addr = mmap(test, reserved_va, PROT_NONE, flags, -1, 0);
2414 if (addr == MAP_FAILED || addr != test) {
2415 error_report("Unable to reserve 0x%lx bytes of virtual address "
2416 "space at %p (%s) for use as guest address space (check your"
2417 "virtual memory ulimit setting, min_mmap_addr or reserve less "
2418 "using -R option)", reserved_va, test, strerror(errno));
2419 exit(EXIT_FAILURE);
2420 }
2421 }
2422
2423 void probe_guest_base(const char *image_name, abi_ulong guest_loaddr,
2424 abi_ulong guest_hiaddr)
2425 {
2426 /* In order to use host shmat, we must be able to honor SHMLBA. */
2427 uintptr_t align = MAX(SHMLBA, qemu_host_page_size);
2428
2429 if (have_guest_base) {
2430 pgb_have_guest_base(image_name, guest_loaddr, guest_hiaddr, align);
2431 } else if (reserved_va) {
2432 pgb_reserved_va(image_name, guest_loaddr, guest_hiaddr, align);
2433 } else if (guest_loaddr) {
2434 pgb_static(image_name, guest_loaddr, guest_hiaddr, align);
2435 } else {
2436 pgb_dynamic(image_name, align);
2437 }
2438
2439 /* Reserve and initialize the commpage. */
2440 if (!init_guest_commpage()) {
2441 /*
2442 * With have_guest_base, the user has selected the address and
2443 * we are trying to work with that. Otherwise, we have selected
2444 * free space and init_guest_commpage must succeeded.
2445 */
2446 assert(have_guest_base);
2447 pgb_fail_in_use(image_name);
2448 }
2449
2450 assert(QEMU_IS_ALIGNED(guest_base, align));
2451 qemu_log_mask(CPU_LOG_PAGE, "Locating guest address space "
2452 "@ 0x%" PRIx64 "\n", (uint64_t)guest_base);
2453 }
2454
2455 enum {
2456 /* The string "GNU\0" as a magic number. */
2457 GNU0_MAGIC = const_le32('G' | 'N' << 8 | 'U' << 16),
2458 NOTE_DATA_SZ = 1 * KiB,
2459 NOTE_NAME_SZ = 4,
2460 ELF_GNU_PROPERTY_ALIGN = ELF_CLASS == ELFCLASS32 ? 4 : 8,
2461 };
2462
2463 /*
2464 * Process a single gnu_property entry.
2465 * Return false for error.
2466 */
2467 static bool parse_elf_property(const uint32_t *data, int *off, int datasz,
2468 struct image_info *info, bool have_prev_type,
2469 uint32_t *prev_type, Error **errp)
2470 {
2471 uint32_t pr_type, pr_datasz, step;
2472
2473 if (*off > datasz || !QEMU_IS_ALIGNED(*off, ELF_GNU_PROPERTY_ALIGN)) {
2474 goto error_data;
2475 }
2476 datasz -= *off;
2477 data += *off / sizeof(uint32_t);
2478
2479 if (datasz < 2 * sizeof(uint32_t)) {
2480 goto error_data;
2481 }
2482 pr_type = data[0];
2483 pr_datasz = data[1];
2484 data += 2;
2485 datasz -= 2 * sizeof(uint32_t);
2486 step = ROUND_UP(pr_datasz, ELF_GNU_PROPERTY_ALIGN);
2487 if (step > datasz) {
2488 goto error_data;
2489 }
2490
2491 /* Properties are supposed to be unique and sorted on pr_type. */
2492 if (have_prev_type && pr_type <= *prev_type) {
2493 if (pr_type == *prev_type) {
2494 error_setg(errp, "Duplicate property in PT_GNU_PROPERTY");
2495 } else {
2496 error_setg(errp, "Unsorted property in PT_GNU_PROPERTY");
2497 }
2498 return false;
2499 }
2500 *prev_type = pr_type;
2501
2502 if (!arch_parse_elf_property(pr_type, pr_datasz, data, info, errp)) {
2503 return false;
2504 }
2505
2506 *off += 2 * sizeof(uint32_t) + step;
2507 return true;
2508
2509 error_data:
2510 error_setg(errp, "Ill-formed property in PT_GNU_PROPERTY");
2511 return false;
2512 }
2513
2514 /* Process NT_GNU_PROPERTY_TYPE_0. */
2515 static bool parse_elf_properties(int image_fd,
2516 struct image_info *info,
2517 const struct elf_phdr *phdr,
2518 char bprm_buf[BPRM_BUF_SIZE],
2519 Error **errp)
2520 {
2521 union {
2522 struct elf_note nhdr;
2523 uint32_t data[NOTE_DATA_SZ / sizeof(uint32_t)];
2524 } note;
2525
2526 int n, off, datasz;
2527 bool have_prev_type;
2528 uint32_t prev_type;
2529
2530 /* Unless the arch requires properties, ignore them. */
2531 if (!ARCH_USE_GNU_PROPERTY) {
2532 return true;
2533 }
2534
2535 /* If the properties are crazy large, that's too bad. */
2536 n = phdr->p_filesz;
2537 if (n > sizeof(note)) {
2538 error_setg(errp, "PT_GNU_PROPERTY too large");
2539 return false;
2540 }
2541 if (n < sizeof(note.nhdr)) {
2542 error_setg(errp, "PT_GNU_PROPERTY too small");
2543 return false;
2544 }
2545
2546 if (phdr->p_offset + n <= BPRM_BUF_SIZE) {
2547 memcpy(&note, bprm_buf + phdr->p_offset, n);
2548 } else {
2549 ssize_t len = pread(image_fd, &note, n, phdr->p_offset);
2550 if (len != n) {
2551 error_setg_errno(errp, errno, "Error reading file header");
2552 return false;
2553 }
2554 }
2555
2556 /*
2557 * The contents of a valid PT_GNU_PROPERTY is a sequence
2558 * of uint32_t -- swap them all now.
2559 */
2560 #ifdef BSWAP_NEEDED
2561 for (int i = 0; i < n / 4; i++) {
2562 bswap32s(note.data + i);
2563 }
2564 #endif
2565
2566 /*
2567 * Note that nhdr is 3 words, and that the "name" described by namesz
2568 * immediately follows nhdr and is thus at the 4th word. Further, all
2569 * of the inputs to the kernel's round_up are multiples of 4.
2570 */
2571 if (note.nhdr.n_type != NT_GNU_PROPERTY_TYPE_0 ||
2572 note.nhdr.n_namesz != NOTE_NAME_SZ ||
2573 note.data[3] != GNU0_MAGIC) {
2574 error_setg(errp, "Invalid note in PT_GNU_PROPERTY");
2575 return false;
2576 }
2577 off = sizeof(note.nhdr) + NOTE_NAME_SZ;
2578
2579 datasz = note.nhdr.n_descsz + off;
2580 if (datasz > n) {
2581 error_setg(errp, "Invalid note size in PT_GNU_PROPERTY");
2582 return false;
2583 }
2584
2585 have_prev_type = false;
2586 prev_type = 0;
2587 while (1) {
2588 if (off == datasz) {
2589 return true; /* end, exit ok */
2590 }
2591 if (!parse_elf_property(note.data, &off, datasz, info,
2592 have_prev_type, &prev_type, errp)) {
2593 return false;
2594 }
2595 have_prev_type = true;
2596 }
2597 }
2598
2599 /* Load an ELF image into the address space.
2600
2601 IMAGE_NAME is the filename of the image, to use in error messages.
2602 IMAGE_FD is the open file descriptor for the image.
2603
2604 BPRM_BUF is a copy of the beginning of the file; this of course
2605 contains the elf file header at offset 0. It is assumed that this
2606 buffer is sufficiently aligned to present no problems to the host
2607 in accessing data at aligned offsets within the buffer.
2608
2609 On return: INFO values will be filled in, as necessary or available. */
2610
2611 static void load_elf_image(const char *image_name, int image_fd,
2612 struct image_info *info, char **pinterp_name,
2613 char bprm_buf[BPRM_BUF_SIZE])
2614 {
2615 struct elfhdr *ehdr = (struct elfhdr *)bprm_buf;
2616 struct elf_phdr *phdr;
2617 abi_ulong load_addr, load_bias, loaddr, hiaddr, error;
2618 int i, retval, prot_exec;
2619 Error *err = NULL;
2620
2621 /* First of all, some simple consistency checks */
2622 if (!elf_check_ident(ehdr)) {
2623 error_setg(&err, "Invalid ELF image for this architecture");
2624 goto exit_errmsg;
2625 }
2626 bswap_ehdr(ehdr);
2627 if (!elf_check_ehdr(ehdr)) {
2628 error_setg(&err, "Invalid ELF image for this architecture");
2629 goto exit_errmsg;
2630 }
2631
2632 i = ehdr->e_phnum * sizeof(struct elf_phdr);
2633 if (ehdr->e_phoff + i <= BPRM_BUF_SIZE) {
2634 phdr = (struct elf_phdr *)(bprm_buf + ehdr->e_phoff);
2635 } else {
2636 phdr = (struct elf_phdr *) alloca(i);
2637 retval = pread(image_fd, phdr, i, ehdr->e_phoff);
2638 if (retval != i) {
2639 goto exit_read;
2640 }
2641 }
2642 bswap_phdr(phdr, ehdr->e_phnum);
2643
2644 info->nsegs = 0;
2645 info->pt_dynamic_addr = 0;
2646
2647 mmap_lock();
2648
2649 /*
2650 * Find the maximum size of the image and allocate an appropriate
2651 * amount of memory to handle that. Locate the interpreter, if any.
2652 */
2653 loaddr = -1, hiaddr = 0;
2654 info->alignment = 0;
2655 for (i = 0; i < ehdr->e_phnum; ++i) {
2656 struct elf_phdr *eppnt = phdr + i;
2657 if (eppnt->p_type == PT_LOAD) {
2658 abi_ulong a = eppnt->p_vaddr - eppnt->p_offset;
2659 if (a < loaddr) {
2660 loaddr = a;
2661 }
2662 a = eppnt->p_vaddr + eppnt->p_memsz;
2663 if (a > hiaddr) {
2664 hiaddr = a;
2665 }
2666 ++info->nsegs;
2667 info->alignment |= eppnt->p_align;
2668 } else if (eppnt->p_type == PT_INTERP && pinterp_name) {
2669 g_autofree char *interp_name = NULL;
2670
2671 if (*pinterp_name) {
2672 error_setg(&err, "Multiple PT_INTERP entries");
2673 goto exit_errmsg;
2674 }
2675
2676 interp_name = g_malloc(eppnt->p_filesz);
2677
2678 if (eppnt->p_offset + eppnt->p_filesz <= BPRM_BUF_SIZE) {
2679 memcpy(interp_name, bprm_buf + eppnt->p_offset,
2680 eppnt->p_filesz);
2681 } else {
2682 retval = pread(image_fd, interp_name, eppnt->p_filesz,
2683 eppnt->p_offset);
2684 if (retval != eppnt->p_filesz) {
2685 goto exit_read;
2686 }
2687 }
2688 if (interp_name[eppnt->p_filesz - 1] != 0) {
2689 error_setg(&err, "Invalid PT_INTERP entry");
2690 goto exit_errmsg;
2691 }
2692 *pinterp_name = g_steal_pointer(&interp_name);
2693 } else if (eppnt->p_type == PT_GNU_PROPERTY) {
2694 if (!parse_elf_properties(image_fd, info, eppnt, bprm_buf, &err)) {
2695 goto exit_errmsg;
2696 }
2697 }
2698 }
2699
2700 if (pinterp_name != NULL) {
2701 /*
2702 * This is the main executable.
2703 *
2704 * Reserve extra space for brk.
2705 * We hold on to this space while placing the interpreter
2706 * and the stack, lest they be placed immediately after
2707 * the data segment and block allocation from the brk.
2708 *
2709 * 16MB is chosen as "large enough" without being so large
2710 * as to allow the result to not fit with a 32-bit guest on
2711 * a 32-bit host.
2712 */
2713 info->reserve_brk = 16 * MiB;
2714 hiaddr += info->reserve_brk;
2715
2716 if (ehdr->e_type == ET_EXEC) {
2717 /*
2718 * Make sure that the low address does not conflict with
2719 * MMAP_MIN_ADDR or the QEMU application itself.
2720 */
2721 probe_guest_base(image_name, loaddr, hiaddr);
2722 } else {
2723 /*
2724 * The binary is dynamic, but we still need to
2725 * select guest_base. In this case we pass a size.
2726 */
2727 probe_guest_base(image_name, 0, hiaddr - loaddr);
2728 }
2729 }
2730
2731 /*
2732 * Reserve address space for all of this.
2733 *
2734 * In the case of ET_EXEC, we supply MAP_FIXED so that we get
2735 * exactly the address range that is required.
2736 *
2737 * Otherwise this is ET_DYN, and we are searching for a location
2738 * that can hold the memory space required. If the image is
2739 * pre-linked, LOADDR will be non-zero, and the kernel should
2740 * honor that address if it happens to be free.
2741 *
2742 * In both cases, we will overwrite pages in this range with mappings
2743 * from the executable.
2744 */
2745 load_addr = target_mmap(loaddr, hiaddr - loaddr, PROT_NONE,
2746 MAP_PRIVATE | MAP_ANON | MAP_NORESERVE |
2747 (ehdr->e_type == ET_EXEC ? MAP_FIXED : 0),
2748 -1, 0);
2749 if (load_addr == -1) {
2750 goto exit_mmap;
2751 }
2752 load_bias = load_addr - loaddr;
2753
2754 if (elf_is_fdpic(ehdr)) {
2755 struct elf32_fdpic_loadseg *loadsegs = info->loadsegs =
2756 g_malloc(sizeof(*loadsegs) * info->nsegs);
2757
2758 for (i = 0; i < ehdr->e_phnum; ++i) {
2759 switch (phdr[i].p_type) {
2760 case PT_DYNAMIC:
2761 info->pt_dynamic_addr = phdr[i].p_vaddr + load_bias;
2762 break;
2763 case PT_LOAD:
2764 loadsegs->addr = phdr[i].p_vaddr + load_bias;
2765 loadsegs->p_vaddr = phdr[i].p_vaddr;
2766 loadsegs->p_memsz = phdr[i].p_memsz;
2767 ++loadsegs;
2768 break;
2769 }
2770 }
2771 }
2772
2773 info->load_bias = load_bias;
2774 info->code_offset = load_bias;
2775 info->data_offset = load_bias;
2776 info->load_addr = load_addr;
2777 info->entry = ehdr->e_entry + load_bias;
2778 info->start_code = -1;
2779 info->end_code = 0;
2780 info->start_data = -1;
2781 info->end_data = 0;
2782 info->brk = 0;
2783 info->elf_flags = ehdr->e_flags;
2784
2785 prot_exec = PROT_EXEC;
2786 #ifdef TARGET_AARCH64
2787 /*
2788 * If the BTI feature is present, this indicates that the executable
2789 * pages of the startup binary should be mapped with PROT_BTI, so that
2790 * branch targets are enforced.
2791 *
2792 * The startup binary is either the interpreter or the static executable.
2793 * The interpreter is responsible for all pages of a dynamic executable.
2794 *
2795 * Elf notes are backward compatible to older cpus.
2796 * Do not enable BTI unless it is supported.
2797 */
2798 if ((info->note_flags & GNU_PROPERTY_AARCH64_FEATURE_1_BTI)
2799 && (pinterp_name == NULL || *pinterp_name == 0)
2800 && cpu_isar_feature(aa64_bti, ARM_CPU(thread_cpu))) {
2801 prot_exec |= TARGET_PROT_BTI;
2802 }
2803 #endif
2804
2805 for (i = 0; i < ehdr->e_phnum; i++) {
2806 struct elf_phdr *eppnt = phdr + i;
2807 if (eppnt->p_type == PT_LOAD) {
2808 abi_ulong vaddr, vaddr_po, vaddr_ps, vaddr_ef, vaddr_em, vaddr_len;
2809 int elf_prot = 0;
2810
2811 if (eppnt->p_flags & PF_R) {
2812 elf_prot |= PROT_READ;
2813 }
2814 if (eppnt->p_flags & PF_W) {
2815 elf_prot |= PROT_WRITE;
2816 }
2817 if (eppnt->p_flags & PF_X) {
2818 elf_prot |= prot_exec;
2819 }
2820
2821 vaddr = load_bias + eppnt->p_vaddr;
2822 vaddr_po = TARGET_ELF_PAGEOFFSET(vaddr);
2823 vaddr_ps = TARGET_ELF_PAGESTART(vaddr);
2824
2825 vaddr_ef = vaddr + eppnt->p_filesz;
2826 vaddr_em = vaddr + eppnt->p_memsz;
2827
2828 /*
2829 * Some segments may be completely empty, with a non-zero p_memsz
2830 * but no backing file segment.
2831 */
2832 if (eppnt->p_filesz != 0) {
2833 vaddr_len = TARGET_ELF_PAGELENGTH(eppnt->p_filesz + vaddr_po);
2834 error = target_mmap(vaddr_ps, vaddr_len, elf_prot,
2835 MAP_PRIVATE | MAP_FIXED,
2836 image_fd, eppnt->p_offset - vaddr_po);
2837
2838 if (error == -1) {
2839 goto exit_mmap;
2840 }
2841
2842 /*
2843 * If the load segment requests extra zeros (e.g. bss), map it.
2844 */
2845 if (eppnt->p_filesz < eppnt->p_memsz) {
2846 zero_bss(vaddr_ef, vaddr_em, elf_prot);
2847 }
2848 } else if (eppnt->p_memsz != 0) {
2849 vaddr_len = TARGET_ELF_PAGELENGTH(eppnt->p_memsz + vaddr_po);
2850 error = target_mmap(vaddr_ps, vaddr_len, elf_prot,
2851 MAP_PRIVATE | MAP_FIXED | MAP_ANONYMOUS,
2852 -1, 0);
2853
2854 if (error == -1) {
2855 goto exit_mmap;
2856 }
2857 }
2858
2859 /* Find the full program boundaries. */
2860 if (elf_prot & PROT_EXEC) {
2861 if (vaddr < info->start_code) {
2862 info->start_code = vaddr;
2863 }
2864 if (vaddr_ef > info->end_code) {
2865 info->end_code = vaddr_ef;
2866 }
2867 }
2868 if (elf_prot & PROT_WRITE) {
2869 if (vaddr < info->start_data) {
2870 info->start_data = vaddr;
2871 }
2872 if (vaddr_ef > info->end_data) {
2873 info->end_data = vaddr_ef;
2874 }
2875 }
2876 if (vaddr_em > info->brk) {
2877 info->brk = vaddr_em;
2878 }
2879 #ifdef TARGET_MIPS
2880 } else if (eppnt->p_type == PT_MIPS_ABIFLAGS) {
2881 Mips_elf_abiflags_v0 abiflags;
2882 if (eppnt->p_filesz < sizeof(Mips_elf_abiflags_v0)) {
2883 error_setg(&err, "Invalid PT_MIPS_ABIFLAGS entry");
2884 goto exit_errmsg;
2885 }
2886 if (eppnt->p_offset + eppnt->p_filesz <= BPRM_BUF_SIZE) {
2887 memcpy(&abiflags, bprm_buf + eppnt->p_offset,
2888 sizeof(Mips_elf_abiflags_v0));
2889 } else {
2890 retval = pread(image_fd, &abiflags, sizeof(Mips_elf_abiflags_v0),
2891 eppnt->p_offset);
2892 if (retval != sizeof(Mips_elf_abiflags_v0)) {
2893 goto exit_read;
2894 }
2895 }
2896 bswap_mips_abiflags(&abiflags);
2897 info->fp_abi = abiflags.fp_abi;
2898 #endif
2899 }
2900 }
2901
2902 if (info->end_data == 0) {
2903 info->start_data = info->end_code;
2904 info->end_data = info->end_code;
2905 }
2906
2907 if (qemu_log_enabled()) {
2908 load_symbols(ehdr, image_fd, load_bias);
2909 }
2910
2911 mmap_unlock();
2912
2913 close(image_fd);
2914 return;
2915
2916 exit_read:
2917 if (retval >= 0) {
2918 error_setg(&err, "Incomplete read of file header");
2919 } else {
2920 error_setg_errno(&err, errno, "Error reading file header");
2921 }
2922 goto exit_errmsg;
2923 exit_mmap:
2924 error_setg_errno(&err, errno, "Error mapping file");
2925 goto exit_errmsg;
2926 exit_errmsg:
2927 error_reportf_err(err, "%s: ", image_name);
2928 exit(-1);
2929 }
2930
2931 static void load_elf_interp(const char *filename, struct image_info *info,
2932 char bprm_buf[BPRM_BUF_SIZE])
2933 {
2934 int fd, retval;
2935 Error *err = NULL;
2936
2937 fd = open(path(filename), O_RDONLY);
2938 if (fd < 0) {
2939 error_setg_file_open(&err, errno, filename);
2940 error_report_err(err);
2941 exit(-1);
2942 }
2943
2944 retval = read(fd, bprm_buf, BPRM_BUF_SIZE);
2945 if (retval < 0) {
2946 error_setg_errno(&err, errno, "Error reading file header");
2947 error_reportf_err(err, "%s: ", filename);
2948 exit(-1);
2949 }
2950
2951 if (retval < BPRM_BUF_SIZE) {
2952 memset(bprm_buf + retval, 0, BPRM_BUF_SIZE - retval);
2953 }
2954
2955 load_elf_image(filename, fd, info, NULL, bprm_buf);
2956 }
2957
2958 static int symfind(const void *s0, const void *s1)
2959 {
2960 target_ulong addr = *(target_ulong *)s0;
2961 struct elf_sym *sym = (struct elf_sym *)s1;
2962 int result = 0;
2963 if (addr < sym->st_value) {
2964 result = -1;
2965 } else if (addr >= sym->st_value + sym->st_size) {
2966 result = 1;
2967 }
2968 return result;
2969 }
2970
2971 static const char *lookup_symbolxx(struct syminfo *s, target_ulong orig_addr)
2972 {
2973 #if ELF_CLASS == ELFCLASS32
2974 struct elf_sym *syms = s->disas_symtab.elf32;
2975 #else
2976 struct elf_sym *syms = s->disas_symtab.elf64;
2977 #endif
2978
2979 // binary search
2980 struct elf_sym *sym;
2981
2982 sym = bsearch(&orig_addr, syms, s->disas_num_syms, sizeof(*syms), symfind);
2983 if (sym != NULL) {
2984 return s->disas_strtab + sym->st_name;
2985 }
2986
2987 return "";
2988 }
2989
2990 /* FIXME: This should use elf_ops.h */
2991 static int symcmp(const void *s0, const void *s1)
2992 {
2993 struct elf_sym *sym0 = (struct elf_sym *)s0;
2994 struct elf_sym *sym1 = (struct elf_sym *)s1;
2995 return (sym0->st_value < sym1->st_value)
2996 ? -1
2997 : ((sym0->st_value > sym1->st_value) ? 1 : 0);
2998 }
2999
3000 /* Best attempt to load symbols from this ELF object. */
3001 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias)
3002 {
3003 int i, shnum, nsyms, sym_idx = 0, str_idx = 0;
3004 uint64_t segsz;
3005 struct elf_shdr *shdr;
3006 char *strings = NULL;
3007 struct syminfo *s = NULL;
3008 struct elf_sym *new_syms, *syms = NULL;
3009
3010 shnum = hdr->e_shnum;
3011 i = shnum * sizeof(struct elf_shdr);
3012 shdr = (struct elf_shdr *)alloca(i);
3013 if (pread(fd, shdr, i, hdr->e_shoff) != i) {
3014 return;
3015 }
3016
3017 bswap_shdr(shdr, shnum);
3018 for (i = 0; i < shnum; ++i) {
3019 if (shdr[i].sh_type == SHT_SYMTAB) {
3020 sym_idx = i;
3021 str_idx = shdr[i].sh_link;
3022 goto found;
3023 }
3024 }
3025
3026 /* There will be no symbol table if the file was stripped. */
3027 return;
3028
3029 found:
3030 /* Now know where the strtab and symtab are. Snarf them. */
3031 s = g_try_new(struct syminfo, 1);
3032 if (!s) {
3033 goto give_up;
3034 }
3035
3036 segsz = shdr[str_idx].sh_size;
3037 s->disas_strtab = strings = g_try_malloc(segsz);
3038 if (!strings ||
3039 pread(fd, strings, segsz, shdr[str_idx].sh_offset) != segsz) {
3040 goto give_up;
3041 }
3042
3043 segsz = shdr[sym_idx].sh_size;
3044 syms = g_try_malloc(segsz);
3045 if (!syms || pread(fd, syms, segsz, shdr[sym_idx].sh_offset) != segsz) {
3046 goto give_up;
3047 }
3048
3049 if (segsz / sizeof(struct elf_sym) > INT_MAX) {
3050 /* Implausibly large symbol table: give up rather than ploughing
3051 * on with the number of symbols calculation overflowing
3052 */
3053 goto give_up;
3054 }
3055 nsyms = segsz / sizeof(struct elf_sym);
3056 for (i = 0; i < nsyms; ) {
3057 bswap_sym(syms + i);
3058 /* Throw away entries which we do not need. */
3059 if (syms[i].st_shndx == SHN_UNDEF
3060 || syms[i].st_shndx >= SHN_LORESERVE
3061 || ELF_ST_TYPE(syms[i].st_info) != STT_FUNC) {
3062 if (i < --nsyms) {
3063 syms[i] = syms[nsyms];
3064 }
3065 } else {
3066 #if defined(TARGET_ARM) || defined (TARGET_MIPS)
3067 /* The bottom address bit marks a Thumb or MIPS16 symbol. */
3068 syms[i].st_value &= ~(target_ulong)1;
3069 #endif
3070 syms[i].st_value += load_bias;
3071 i++;
3072 }
3073 }
3074
3075 /* No "useful" symbol. */
3076 if (nsyms == 0) {
3077 goto give_up;
3078 }
3079
3080 /* Attempt to free the storage associated with the local symbols
3081 that we threw away. Whether or not this has any effect on the
3082 memory allocation depends on the malloc implementation and how
3083 many symbols we managed to discard. */
3084 new_syms = g_try_renew(struct elf_sym, syms, nsyms);
3085 if (new_syms == NULL) {
3086 goto give_up;
3087 }
3088 syms = new_syms;
3089
3090 qsort(syms, nsyms, sizeof(*syms), symcmp);
3091
3092 s->disas_num_syms = nsyms;
3093 #if ELF_CLASS == ELFCLASS32
3094 s->disas_symtab.elf32 = syms;
3095 #else
3096 s->disas_symtab.elf64 = syms;
3097 #endif
3098 s->lookup_symbol = lookup_symbolxx;
3099 s->next = syminfos;
3100 syminfos = s;
3101
3102 return;
3103
3104 give_up:
3105 g_free(s);
3106 g_free(strings);
3107 g_free(syms);
3108 }
3109
3110 uint32_t get_elf_eflags(int fd)
3111 {
3112 struct elfhdr ehdr;
3113 off_t offset;
3114 int ret;
3115
3116 /* Read ELF header */
3117 offset = lseek(fd, 0, SEEK_SET);
3118 if (offset == (off_t) -1) {
3119 return 0;
3120 }
3121 ret = read(fd, &ehdr, sizeof(ehdr));
3122 if (ret < sizeof(ehdr)) {
3123 return 0;
3124 }
3125 offset = lseek(fd, offset, SEEK_SET);
3126 if (offset == (off_t) -1) {
3127 return 0;
3128 }
3129
3130 /* Check ELF signature */
3131 if (!elf_check_ident(&ehdr)) {
3132 return 0;
3133 }
3134
3135 /* check header */
3136 bswap_ehdr(&ehdr);
3137 if (!elf_check_ehdr(&ehdr)) {
3138 return 0;
3139 }
3140
3141 /* return architecture id */
3142 return ehdr.e_flags;
3143 }
3144
3145 int load_elf_binary(struct linux_binprm *bprm, struct image_info *info)
3146 {
3147 struct image_info interp_info;
3148 struct elfhdr elf_ex;
3149 char *elf_interpreter = NULL;
3150 char *scratch;
3151
3152 memset(&interp_info, 0, sizeof(interp_info));
3153 #ifdef TARGET_MIPS
3154 interp_info.fp_abi = MIPS_ABI_FP_UNKNOWN;
3155 #endif
3156
3157 info->start_mmap = (abi_ulong)ELF_START_MMAP;
3158
3159 load_elf_image(bprm->filename, bprm->fd, info,
3160 &elf_interpreter, bprm->buf);
3161
3162 /* ??? We need a copy of the elf header for passing to create_elf_tables.
3163 If we do nothing, we'll have overwritten this when we re-use bprm->buf
3164 when we load the interpreter. */
3165 elf_ex = *(struct elfhdr *)bprm->buf;
3166
3167 /* Do this so that we can load the interpreter, if need be. We will
3168 change some of these later */
3169 bprm->p = setup_arg_pages(bprm, info);
3170
3171 scratch = g_new0(char, TARGET_PAGE_SIZE);
3172 if (STACK_GROWS_DOWN) {
3173 bprm->p = copy_elf_strings(1, &bprm->filename, scratch,
3174 bprm->p, info->stack_limit);
3175 info->file_string = bprm->p;
3176 bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch,
3177 bprm->p, info->stack_limit);
3178 info->env_strings = bprm->p;
3179 bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch,
3180 bprm->p, info->stack_limit);
3181 info->arg_strings = bprm->p;
3182 } else {
3183 info->arg_strings = bprm->p;
3184 bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch,
3185 bprm->p, info->stack_limit);
3186 info->env_strings = bprm->p;
3187 bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch,
3188 bprm->p, info->stack_limit);
3189 info->file_string = bprm->p;
3190 bprm->p = copy_elf_strings(1, &bprm->filename, scratch,
3191 bprm->p, info->stack_limit);
3192 }
3193
3194 g_free(scratch);
3195
3196 if (!bprm->p) {
3197 fprintf(stderr, "%s: %s\n", bprm->filename, strerror(E2BIG));
3198 exit(-1);
3199 }
3200
3201 if (elf_interpreter) {
3202 load_elf_interp(elf_interpreter, &interp_info, bprm->buf);
3203
3204 /* If the program interpreter is one of these two, then assume
3205 an iBCS2 image. Otherwise assume a native linux image. */
3206
3207 if (strcmp(elf_interpreter, "/usr/lib/libc.so.1") == 0
3208 || strcmp(elf_interpreter, "/usr/lib/ld.so.1") == 0) {
3209 info->personality = PER_SVR4;
3210
3211 /* Why this, you ask??? Well SVr4 maps page 0 as read-only,
3212 and some applications "depend" upon this behavior. Since
3213 we do not have the power to recompile these, we emulate
3214 the SVr4 behavior. Sigh. */
3215 target_mmap(0, qemu_host_page_size, PROT_READ | PROT_EXEC,
3216 MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
3217 }
3218 #ifdef TARGET_MIPS
3219 info->interp_fp_abi = interp_info.fp_abi;
3220 #endif
3221 }
3222
3223 bprm->p = create_elf_tables(bprm->p, bprm->argc, bprm->envc, &elf_ex,
3224 info, (elf_interpreter ? &interp_info : NULL));
3225 info->start_stack = bprm->p;
3226
3227 /* If we have an interpreter, set that as the program's entry point.
3228 Copy the load_bias as well, to help PPC64 interpret the entry
3229 point as a function descriptor. Do this after creating elf tables
3230 so that we copy the original program entry point into the AUXV. */
3231 if (elf_interpreter) {
3232 info->load_bias = interp_info.load_bias;
3233 info->entry = interp_info.entry;
3234 g_free(elf_interpreter);
3235 }
3236
3237 #ifdef USE_ELF_CORE_DUMP
3238 bprm->core_dump = &elf_core_dump;
3239 #endif
3240
3241 /*
3242 * If we reserved extra space for brk, release it now.
3243 * The implementation of do_brk in syscalls.c expects to be able
3244 * to mmap pages in this space.
3245 */
3246 if (info->reserve_brk) {
3247 abi_ulong start_brk = HOST_PAGE_ALIGN(info->brk);
3248 abi_ulong end_brk = HOST_PAGE_ALIGN(info->brk + info->reserve_brk);
3249 target_munmap(start_brk, end_brk - start_brk);
3250 }
3251
3252 return 0;
3253 }
3254
3255 #ifdef USE_ELF_CORE_DUMP
3256 /*
3257 * Definitions to generate Intel SVR4-like core files.
3258 * These mostly have the same names as the SVR4 types with "target_elf_"
3259 * tacked on the front to prevent clashes with linux definitions,
3260 * and the typedef forms have been avoided. This is mostly like
3261 * the SVR4 structure, but more Linuxy, with things that Linux does
3262 * not support and which gdb doesn't really use excluded.
3263 *
3264 * Fields we don't dump (their contents is zero) in linux-user qemu
3265 * are marked with XXX.
3266 *
3267 * Core dump code is copied from linux kernel (fs/binfmt_elf.c).
3268 *
3269 * Porting ELF coredump for target is (quite) simple process. First you
3270 * define USE_ELF_CORE_DUMP in target ELF code (where init_thread() for
3271 * the target resides):
3272 *
3273 * #define USE_ELF_CORE_DUMP
3274 *
3275 * Next you define type of register set used for dumping. ELF specification
3276 * says that it needs to be array of elf_greg_t that has size of ELF_NREG.
3277 *
3278 * typedef <target_regtype> target_elf_greg_t;
3279 * #define ELF_NREG <number of registers>
3280 * typedef taret_elf_greg_t target_elf_gregset_t[ELF_NREG];
3281 *
3282 * Last step is to implement target specific function that copies registers
3283 * from given cpu into just specified register set. Prototype is:
3284 *
3285 * static void elf_core_copy_regs(taret_elf_gregset_t *regs,
3286 * const CPUArchState *env);
3287 *
3288 * Parameters:
3289 * regs - copy register values into here (allocated and zeroed by caller)
3290 * env - copy registers from here
3291 *
3292 * Example for ARM target is provided in this file.
3293 */
3294
3295 /* An ELF note in memory */
3296 struct memelfnote {
3297 const char *name;
3298 size_t namesz;
3299 size_t namesz_rounded;
3300 int type;
3301 size_t datasz;
3302 size_t datasz_rounded;
3303 void *data;
3304 size_t notesz;
3305 };
3306
3307 struct target_elf_siginfo {
3308 abi_int si_signo; /* signal number */
3309 abi_int si_code; /* extra code */
3310 abi_int si_errno; /* errno */
3311 };
3312
3313 struct target_elf_prstatus {
3314 struct target_elf_siginfo pr_info; /* Info associated with signal */
3315 abi_short pr_cursig; /* Current signal */
3316 abi_ulong pr_sigpend; /* XXX */
3317 abi_ulong pr_sighold; /* XXX */
3318 target_pid_t pr_pid;
3319 target_pid_t pr_ppid;
3320 target_pid_t pr_pgrp;
3321 target_pid_t pr_sid;
3322 struct target_timeval pr_utime; /* XXX User time */
3323 struct target_timeval pr_stime; /* XXX System time */
3324 struct target_timeval pr_cutime; /* XXX Cumulative user time */
3325 struct target_timeval pr_cstime; /* XXX Cumulative system time */
3326 target_elf_gregset_t pr_reg; /* GP registers */
3327 abi_int pr_fpvalid; /* XXX */
3328 };
3329
3330 #define ELF_PRARGSZ (80) /* Number of chars for args */
3331
3332 struct target_elf_prpsinfo {
3333 char pr_state; /* numeric process state */
3334 char pr_sname; /* char for pr_state */
3335 char pr_zomb; /* zombie */
3336 char pr_nice; /* nice val */
3337 abi_ulong pr_flag; /* flags */
3338 target_uid_t pr_uid;
3339 target_gid_t pr_gid;
3340 target_pid_t pr_pid, pr_ppid, pr_pgrp, pr_sid;
3341 /* Lots missing */
3342 char pr_fname[16] QEMU_NONSTRING; /* filename of executable */
3343 char pr_psargs[ELF_PRARGSZ]; /* initial part of arg list */
3344 };
3345
3346 /* Here is the structure in which status of each thread is captured. */
3347 struct elf_thread_status {
3348 QTAILQ_ENTRY(elf_thread_status) ets_link;
3349 struct target_elf_prstatus prstatus; /* NT_PRSTATUS */
3350 #if 0
3351 elf_fpregset_t fpu; /* NT_PRFPREG */
3352 struct task_struct *thread;
3353 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */
3354 #endif
3355 struct memelfnote notes[1];
3356 int num_notes;
3357 };
3358
3359 struct elf_note_info {
3360 struct memelfnote *notes;
3361 struct target_elf_prstatus *prstatus; /* NT_PRSTATUS */
3362 struct target_elf_prpsinfo *psinfo; /* NT_PRPSINFO */
3363
3364 QTAILQ_HEAD(, elf_thread_status) thread_list;
3365 #if 0
3366 /*
3367 * Current version of ELF coredump doesn't support
3368 * dumping fp regs etc.
3369 */
3370 elf_fpregset_t *fpu;
3371 elf_fpxregset_t *xfpu;
3372 int thread_status_size;
3373 #endif
3374 int notes_size;
3375 int numnote;
3376 };
3377
3378 struct vm_area_struct {
3379 target_ulong vma_start; /* start vaddr of memory region */
3380 target_ulong vma_end; /* end vaddr of memory region */
3381 abi_ulong vma_flags; /* protection etc. flags for the region */
3382 QTAILQ_ENTRY(vm_area_struct) vma_link;
3383 };
3384
3385 struct mm_struct {
3386 QTAILQ_HEAD(, vm_area_struct) mm_mmap;
3387 int mm_count; /* number of mappings */
3388 };
3389
3390 static struct mm_struct *vma_init(void);
3391 static void vma_delete(struct mm_struct *);
3392 static int vma_add_mapping(struct mm_struct *, target_ulong,
3393 target_ulong, abi_ulong);
3394 static int vma_get_mapping_count(const struct mm_struct *);
3395 static struct vm_area_struct *vma_first(const struct mm_struct *);
3396 static struct vm_area_struct *vma_next(struct vm_area_struct *);
3397 static abi_ulong vma_dump_size(const struct vm_area_struct *);
3398 static int vma_walker(void *priv, target_ulong start, target_ulong end,
3399 unsigned long flags);
3400
3401 static void fill_elf_header(struct elfhdr *, int, uint16_t, uint32_t);
3402 static void fill_note(struct memelfnote *, const char *, int,
3403 unsigned int, void *);
3404 static void fill_prstatus(struct target_elf_prstatus *, const TaskState *, int);
3405 static int fill_psinfo(struct target_elf_prpsinfo *, const TaskState *);
3406 static void fill_auxv_note(struct memelfnote *, const TaskState *);
3407 static void fill_elf_note_phdr(struct elf_phdr *, int, off_t);
3408 static size_t note_size(const struct memelfnote *);
3409 static void free_note_info(struct elf_note_info *);
3410 static int fill_note_info(struct elf_note_info *, long, const CPUArchState *);
3411 static void fill_thread_info(struct elf_note_info *, const CPUArchState *);
3412
3413 static int dump_write(int, const void *, size_t);
3414 static int write_note(struct memelfnote *, int);
3415 static int write_note_info(struct elf_note_info *, int);
3416
3417 #ifdef BSWAP_NEEDED
3418 static void bswap_prstatus(struct target_elf_prstatus *prstatus)
3419 {
3420 prstatus->pr_info.si_signo = tswap32(prstatus->pr_info.si_signo);
3421 prstatus->pr_info.si_code = tswap32(prstatus->pr_info.si_code);
3422 prstatus->pr_info.si_errno = tswap32(prstatus->pr_info.si_errno);
3423 prstatus->pr_cursig = tswap16(prstatus->pr_cursig);
3424 prstatus->pr_sigpend = tswapal(prstatus->pr_sigpend);
3425 prstatus->pr_sighold = tswapal(prstatus->pr_sighold);
3426 prstatus->pr_pid = tswap32(prstatus->pr_pid);
3427 prstatus->pr_ppid = tswap32(prstatus->pr_ppid);
3428 prstatus->pr_pgrp = tswap32(prstatus->pr_pgrp);
3429 prstatus->pr_sid = tswap32(prstatus->pr_sid);
3430 /* cpu times are not filled, so we skip them */
3431 /* regs should be in correct format already */
3432 prstatus->pr_fpvalid = tswap32(prstatus->pr_fpvalid);
3433 }
3434
3435 static void bswap_psinfo(struct target_elf_prpsinfo *psinfo)
3436 {
3437 psinfo->pr_flag = tswapal(psinfo->pr_flag);
3438 psinfo->pr_uid = tswap16(psinfo->pr_uid);
3439 psinfo->pr_gid = tswap16(psinfo->pr_gid);
3440 psinfo->pr_pid = tswap32(psinfo->pr_pid);
3441 psinfo->pr_ppid = tswap32(psinfo->pr_ppid);
3442 psinfo->pr_pgrp = tswap32(psinfo->pr_pgrp);
3443 psinfo->pr_sid = tswap32(psinfo->pr_sid);
3444 }
3445
3446 static void bswap_note(struct elf_note *en)
3447 {
3448 bswap32s(&en->n_namesz);
3449 bswap32s(&en->n_descsz);
3450 bswap32s(&en->n_type);
3451 }
3452 #else
3453 static inline void bswap_prstatus(struct target_elf_prstatus *p) { }
3454 static inline void bswap_psinfo(struct target_elf_prpsinfo *p) {}
3455 static inline void bswap_note(struct elf_note *en) { }
3456 #endif /* BSWAP_NEEDED */
3457
3458 /*
3459 * Minimal support for linux memory regions. These are needed
3460 * when we are finding out what memory exactly belongs to
3461 * emulated process. No locks needed here, as long as
3462 * thread that received the signal is stopped.
3463 */
3464
3465 static struct mm_struct *vma_init(void)
3466 {
3467 struct mm_struct *mm;
3468
3469 if ((mm = g_malloc(sizeof (*mm))) == NULL)
3470 return (NULL);
3471
3472 mm->mm_count = 0;
3473 QTAILQ_INIT(&mm->mm_mmap);
3474
3475 return (mm);
3476 }
3477
3478 static void vma_delete(struct mm_struct *mm)
3479 {
3480 struct vm_area_struct *vma;
3481
3482 while ((vma = vma_first(mm)) != NULL) {
3483 QTAILQ_REMOVE(&mm->mm_mmap, vma, vma_link);
3484 g_free(vma);
3485 }
3486 g_free(mm);
3487 }
3488
3489 static int vma_add_mapping(struct mm_struct *mm, target_ulong start,
3490 target_ulong end, abi_ulong flags)
3491 {
3492 struct vm_area_struct *vma;
3493
3494 if ((vma = g_malloc0(sizeof (*vma))) == NULL)
3495 return (-1);
3496
3497 vma->vma_start = start;
3498 vma->vma_end = end;
3499 vma->vma_flags = flags;
3500
3501 QTAILQ_INSERT_TAIL(&mm->mm_mmap, vma, vma_link);
3502 mm->mm_count++;
3503
3504 return (0);
3505 }
3506
3507 static struct vm_area_struct *vma_first(const struct mm_struct *mm)
3508 {
3509 return (QTAILQ_FIRST(&mm->mm_mmap));
3510 }
3511
3512 static struct vm_area_struct *vma_next(struct vm_area_struct *vma)
3513 {
3514 return (QTAILQ_NEXT(vma, vma_link));
3515 }
3516
3517 static int vma_get_mapping_count(const struct mm_struct *mm)
3518 {
3519 return (mm->mm_count);
3520 }
3521
3522 /*
3523 * Calculate file (dump) size of given memory region.
3524 */
3525 static abi_ulong vma_dump_size(const struct vm_area_struct *vma)
3526 {
3527 /* if we cannot even read the first page, skip it */
3528 if (!access_ok_untagged(VERIFY_READ, vma->vma_start, TARGET_PAGE_SIZE))
3529 return (0);
3530
3531 /*
3532 * Usually we don't dump executable pages as they contain
3533 * non-writable code that debugger can read directly from
3534 * target library etc. However, thread stacks are marked
3535 * also executable so we read in first page of given region
3536 * and check whether it contains elf header. If there is
3537 * no elf header, we dump it.
3538 */
3539 if (vma->vma_flags & PROT_EXEC) {
3540 char page[TARGET_PAGE_SIZE];
3541
3542 if (copy_from_user(page, vma->vma_start, sizeof (page))) {
3543 return 0;
3544 }
3545 if ((page[EI_MAG0] == ELFMAG0) &&
3546 (page[EI_MAG1] == ELFMAG1) &&
3547 (page[EI_MAG2] == ELFMAG2) &&
3548 (page[EI_MAG3] == ELFMAG3)) {
3549 /*
3550 * Mappings are possibly from ELF binary. Don't dump
3551 * them.
3552 */
3553 return (0);
3554 }
3555 }
3556
3557 return (vma->vma_end - vma->vma_start);
3558 }
3559
3560 static int vma_walker(void *priv, target_ulong start, target_ulong end,
3561 unsigned long flags)
3562 {
3563 struct mm_struct *mm = (struct mm_struct *)priv;
3564
3565 vma_add_mapping(mm, start, end, flags);
3566 return (0);
3567 }
3568
3569 static void fill_note(struct memelfnote *note, const char *name, int type,
3570 unsigned int sz, void *data)
3571 {
3572 unsigned int namesz;
3573
3574 namesz = strlen(name) + 1;
3575 note->name = name;
3576 note->namesz = namesz;
3577 note->namesz_rounded = roundup(namesz, sizeof (int32_t));
3578 note->type = type;
3579 note->datasz = sz;
3580 note->datasz_rounded = roundup(sz, sizeof (int32_t));
3581
3582 note->data = data;
3583
3584 /*
3585 * We calculate rounded up note size here as specified by
3586 * ELF document.
3587 */
3588 note->notesz = sizeof (struct elf_note) +
3589 note->namesz_rounded + note->datasz_rounded;
3590 }
3591
3592 static void fill_elf_header(struct elfhdr *elf, int segs, uint16_t machine,
3593 uint32_t flags)
3594 {
3595 (void) memset(elf, 0, sizeof(*elf));
3596
3597 (void) memcpy(elf->e_ident, ELFMAG, SELFMAG);
3598 elf->e_ident[EI_CLASS] = ELF_CLASS;
3599 elf->e_ident[EI_DATA] = ELF_DATA;
3600 elf->e_ident[EI_VERSION] = EV_CURRENT;
3601 elf->e_ident[EI_OSABI] = ELF_OSABI;
3602
3603 elf->e_type = ET_CORE;
3604 elf->e_machine = machine;
3605 elf->e_version = EV_CURRENT;
3606 elf->e_phoff = sizeof(struct elfhdr);
3607 elf->e_flags = flags;
3608 elf->e_ehsize = sizeof(struct elfhdr);
3609 elf->e_phentsize = sizeof(struct elf_phdr);
3610 elf->e_phnum = segs;
3611
3612 bswap_ehdr(elf);
3613 }
3614
3615 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, off_t offset)
3616 {
3617 phdr->p_type = PT_NOTE;
3618 phdr->p_offset = offset;
3619 phdr->p_vaddr = 0;
3620 phdr->p_paddr = 0;
3621 phdr->p_filesz = sz;
3622 phdr->p_memsz = 0;
3623 phdr->p_flags = 0;
3624 phdr->p_align = 0;
3625
3626 bswap_phdr(phdr, 1);
3627 }
3628
3629 static size_t note_size(const struct memelfnote *note)
3630 {
3631 return (note->notesz);
3632 }
3633
3634 static void fill_prstatus(struct target_elf_prstatus *prstatus,
3635 const TaskState *ts, int signr)
3636 {
3637 (void) memset(prstatus, 0, sizeof (*prstatus));
3638 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
3639 prstatus->pr_pid = ts->ts_tid;
3640 prstatus->pr_ppid = getppid();
3641 prstatus->pr_pgrp = getpgrp();
3642 prstatus->pr_sid = getsid(0);
3643
3644 bswap_prstatus(prstatus);
3645 }
3646
3647 static int fill_psinfo(struct target_elf_prpsinfo *psinfo, const TaskState *ts)
3648 {
3649 char *base_filename;
3650 unsigned int i, len;
3651
3652 (void) memset(psinfo, 0, sizeof (*psinfo));
3653
3654 len = ts->info->env_strings - ts->info->arg_strings;
3655 if (len >= ELF_PRARGSZ)
3656 len = ELF_PRARGSZ - 1;
3657 if (copy_from_user(&psinfo->pr_psargs, ts->info->arg_strings, len)) {
3658 return -EFAULT;
3659 }
3660 for (i = 0; i < len; i++)
3661 if (psinfo->pr_psargs[i] == 0)
3662 psinfo->pr_psargs[i] = ' ';
3663 psinfo->pr_psargs[len] = 0;
3664
3665 psinfo->pr_pid = getpid();
3666 psinfo->pr_ppid = getppid();
3667 psinfo->pr_pgrp = getpgrp();
3668 psinfo->pr_sid = getsid(0);
3669 psinfo->pr_uid = getuid();
3670 psinfo->pr_gid = getgid();
3671
3672 base_filename = g_path_get_basename(ts->bprm->filename);
3673 /*
3674 * Using strncpy here is fine: at max-length,
3675 * this field is not NUL-terminated.
3676 */
3677 (void) strncpy(psinfo->pr_fname, base_filename,
3678 sizeof(psinfo->pr_fname));
3679
3680 g_free(base_filename);
3681 bswap_psinfo(psinfo);
3682 return (0);
3683 }
3684
3685 static void fill_auxv_note(struct memelfnote *note, const TaskState *ts)
3686 {
3687 elf_addr_t auxv = (elf_addr_t)ts->info->saved_auxv;
3688 elf_addr_t orig_auxv = auxv;
3689 void *ptr;
3690 int len = ts->info->auxv_len;
3691
3692 /*
3693 * Auxiliary vector is stored in target process stack. It contains
3694 * {type, value} pairs that we need to dump into note. This is not
3695 * strictly necessary but we do it here for sake of completeness.
3696 */
3697
3698 /* read in whole auxv vector and copy it to memelfnote */
3699 ptr = lock_user(VERIFY_READ, orig_auxv, len, 0);
3700 if (ptr != NULL) {
3701 fill_note(note, "CORE", NT_AUXV, len, ptr);
3702 unlock_user(ptr, auxv, len);
3703 }
3704 }
3705
3706 /*
3707 * Constructs name of coredump file. We have following convention
3708 * for the name:
3709 * qemu_<basename-of-target-binary>_<date>-<time>_<pid>.core
3710 *
3711 * Returns the filename
3712 */
3713 static char *core_dump_filename(const TaskState *ts)
3714 {
3715 g_autoptr(GDateTime) now = g_date_time_new_now_local();
3716 g_autofree char *nowstr = g_date_time_format(now, "%Y%m%d-%H%M%S");
3717 g_autofree char *base_filename = g_path_get_basename(ts->bprm->filename);
3718
3719 return g_strdup_printf("qemu_%s_%s_%d.core",
3720 base_filename, nowstr, (int)getpid());
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 g_autofree char *corefile = NULL;
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 corefile = core_dump_filename(ts);
3966
3967 if ((fd = open(corefile, O_WRONLY | O_CREAT,
3968 S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH)) < 0)
3969 return (-errno);
3970
3971 /*
3972 * Walk through target process memory mappings and
3973 * set up structure containing this information. After
3974 * this point vma_xxx functions can be used.
3975 */
3976 if ((mm = vma_init()) == NULL)
3977 goto out;
3978
3979 walk_memory_regions(mm, vma_walker);
3980 segs = vma_get_mapping_count(mm);
3981
3982 /*
3983 * Construct valid coredump ELF header. We also
3984 * add one more segment for notes.
3985 */
3986 fill_elf_header(&elf, segs + 1, ELF_MACHINE, 0);
3987 if (dump_write(fd, &elf, sizeof (elf)) != 0)
3988 goto out;
3989
3990 /* fill in the in-memory version of notes */
3991 if (fill_note_info(&info, signr, env) < 0)
3992 goto out;
3993
3994 offset += sizeof (elf); /* elf header */
3995 offset += (segs + 1) * sizeof (struct elf_phdr); /* program headers */
3996
3997 /* write out notes program header */
3998 fill_elf_note_phdr(&phdr, info.notes_size, offset);
3999
4000 offset += info.notes_size;
4001 if (dump_write(fd, &phdr, sizeof (phdr)) != 0)
4002 goto out;
4003
4004 /*
4005 * ELF specification wants data to start at page boundary so
4006 * we align it here.
4007 */
4008 data_offset = offset = roundup(offset, ELF_EXEC_PAGESIZE);
4009
4010 /*
4011 * Write program headers for memory regions mapped in
4012 * the target process.
4013 */
4014 for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) {
4015 (void) memset(&phdr, 0, sizeof (phdr));
4016
4017 phdr.p_type = PT_LOAD;
4018 phdr.p_offset = offset;
4019 phdr.p_vaddr = vma->vma_start;
4020 phdr.p_paddr = 0;
4021 phdr.p_filesz = vma_dump_size(vma);
4022 offset += phdr.p_filesz;
4023 phdr.p_memsz = vma->vma_end - vma->vma_start;
4024 phdr.p_flags = vma->vma_flags & PROT_READ ? PF_R : 0;
4025 if (vma->vma_flags & PROT_WRITE)
4026 phdr.p_flags |= PF_W;
4027 if (vma->vma_flags & PROT_EXEC)
4028 phdr.p_flags |= PF_X;
4029 phdr.p_align = ELF_EXEC_PAGESIZE;
4030
4031 bswap_phdr(&phdr, 1);
4032 if (dump_write(fd, &phdr, sizeof(phdr)) != 0) {
4033 goto out;
4034 }
4035 }
4036
4037 /*
4038 * Next we write notes just after program headers. No
4039 * alignment needed here.
4040 */
4041 if (write_note_info(&info, fd) < 0)
4042 goto out;
4043
4044 /* align data to page boundary */
4045 if (lseek(fd, data_offset, SEEK_SET) != data_offset)
4046 goto out;
4047
4048 /*
4049 * Finally we can dump process memory into corefile as well.
4050 */
4051 for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) {
4052 abi_ulong addr;
4053 abi_ulong end;
4054
4055 end = vma->vma_start + vma_dump_size(vma);
4056
4057 for (addr = vma->vma_start; addr < end;
4058 addr += TARGET_PAGE_SIZE) {
4059 char page[TARGET_PAGE_SIZE];
4060 int error;
4061
4062 /*
4063 * Read in page from target process memory and
4064 * write it to coredump file.
4065 */
4066 error = copy_from_user(page, addr, sizeof (page));
4067 if (error != 0) {
4068 (void) fprintf(stderr, "unable to dump " TARGET_ABI_FMT_lx "\n",
4069 addr);
4070 errno = -error;
4071 goto out;
4072 }
4073 if (dump_write(fd, page, TARGET_PAGE_SIZE) < 0)
4074 goto out;
4075 }
4076 }
4077
4078 out:
4079 free_note_info(&info);
4080 if (mm != NULL)
4081 vma_delete(mm);
4082 (void) close(fd);
4083
4084 if (errno != 0)
4085 return (-errno);
4086 return (0);
4087 }
4088 #endif /* USE_ELF_CORE_DUMP */
4089
4090 void do_init_thread(struct target_pt_regs *regs, struct image_info *infop)
4091 {
4092 init_thread(regs, infop);
4093 }
AR7 emulation for QEMU