]> git.proxmox.com Git - mirror_qemu.git/blob - linux-user/elfload.c
projects / mirror_qemu.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
linux-user: Rewrite non-fixed probe_guest_base
[mirror_qemu.git] / linux-user / elfload.c
1 /* This is the Linux kernel elf-loading code, ported into user space */
2 #include "qemu/osdep.h"
3 #include <sys/param.h>
4
5 #include <sys/resource.h>
6 #include <sys/shm.h>
7
8 #include "qemu.h"
9 #include "user-internals.h"
10 #include "signal-common.h"
11 #include "loader.h"
12 #include "user-mmap.h"
13 #include "disas/disas.h"
14 #include "qemu/bitops.h"
15 #include "qemu/path.h"
16 #include "qemu/queue.h"
17 #include "qemu/guest-random.h"
18 #include "qemu/units.h"
19 #include "qemu/selfmap.h"
20 #include "qemu/lockable.h"
21 #include "qapi/error.h"
22 #include "qemu/error-report.h"
23 #include "target_signal.h"
24 #include "accel/tcg/debuginfo.h"
25
26 #ifdef _ARCH_PPC64
27 #undef ARCH_DLINFO
28 #undef ELF_PLATFORM
29 #undef ELF_HWCAP
30 #undef ELF_HWCAP2
31 #undef ELF_CLASS
32 #undef ELF_DATA
33 #undef ELF_ARCH
34 #endif
35
36 #define ELF_OSABI ELFOSABI_SYSV
37
38 /* from personality.h */
39
40 /*
41 * Flags for bug emulation.
42 *
43 * These occupy the top three bytes.
44 */
45 enum {
46 ADDR_NO_RANDOMIZE = 0x0040000, /* disable randomization of VA space */
47 FDPIC_FUNCPTRS = 0x0080000, /* userspace function ptrs point to
48 descriptors (signal handling) */
49 MMAP_PAGE_ZERO = 0x0100000,
50 ADDR_COMPAT_LAYOUT = 0x0200000,
51 READ_IMPLIES_EXEC = 0x0400000,
52 ADDR_LIMIT_32BIT = 0x0800000,
53 SHORT_INODE = 0x1000000,
54 WHOLE_SECONDS = 0x2000000,
55 STICKY_TIMEOUTS = 0x4000000,
56 ADDR_LIMIT_3GB = 0x8000000,
57 };
58
59 /*
60 * Personality types.
61 *
62 * These go in the low byte. Avoid using the top bit, it will
63 * conflict with error returns.
64 */
65 enum {
66 PER_LINUX = 0x0000,
67 PER_LINUX_32BIT = 0x0000 | ADDR_LIMIT_32BIT,
68 PER_LINUX_FDPIC = 0x0000 | FDPIC_FUNCPTRS,
69 PER_SVR4 = 0x0001 | STICKY_TIMEOUTS | MMAP_PAGE_ZERO,
70 PER_SVR3 = 0x0002 | STICKY_TIMEOUTS | SHORT_INODE,
71 PER_SCOSVR3 = 0x0003 | STICKY_TIMEOUTS | WHOLE_SECONDS | SHORT_INODE,
72 PER_OSR5 = 0x0003 | STICKY_TIMEOUTS | WHOLE_SECONDS,
73 PER_WYSEV386 = 0x0004 | STICKY_TIMEOUTS | SHORT_INODE,
74 PER_ISCR4 = 0x0005 | STICKY_TIMEOUTS,
75 PER_BSD = 0x0006,
76 PER_SUNOS = 0x0006 | STICKY_TIMEOUTS,
77 PER_XENIX = 0x0007 | STICKY_TIMEOUTS | SHORT_INODE,
78 PER_LINUX32 = 0x0008,
79 PER_LINUX32_3GB = 0x0008 | ADDR_LIMIT_3GB,
80 PER_IRIX32 = 0x0009 | STICKY_TIMEOUTS,/* IRIX5 32-bit */
81 PER_IRIXN32 = 0x000a | STICKY_TIMEOUTS,/* IRIX6 new 32-bit */
82 PER_IRIX64 = 0x000b | STICKY_TIMEOUTS,/* IRIX6 64-bit */
83 PER_RISCOS = 0x000c,
84 PER_SOLARIS = 0x000d | STICKY_TIMEOUTS,
85 PER_UW7 = 0x000e | STICKY_TIMEOUTS | MMAP_PAGE_ZERO,
86 PER_OSF4 = 0x000f, /* OSF/1 v4 */
87 PER_HPUX = 0x0010,
88 PER_MASK = 0x00ff,
89 };
90
91 /*
92 * Return the base personality without flags.
93 */
94 #define personality(pers) (pers & PER_MASK)
95
96 int info_is_fdpic(struct image_info *info)
97 {
98 return info->personality == PER_LINUX_FDPIC;
99 }
100
101 /* this flag is uneffective under linux too, should be deleted */
102 #ifndef MAP_DENYWRITE
103 #define MAP_DENYWRITE 0
104 #endif
105
106 /* should probably go in elf.h */
107 #ifndef ELIBBAD
108 #define ELIBBAD 80
109 #endif
110
111 #if TARGET_BIG_ENDIAN
112 #define ELF_DATA ELFDATA2MSB
113 #else
114 #define ELF_DATA ELFDATA2LSB
115 #endif
116
117 #ifdef TARGET_ABI_MIPSN32
118 typedef abi_ullong target_elf_greg_t;
119 #define tswapreg(ptr) tswap64(ptr)
120 #else
121 typedef abi_ulong target_elf_greg_t;
122 #define tswapreg(ptr) tswapal(ptr)
123 #endif
124
125 #ifdef USE_UID16
126 typedef abi_ushort target_uid_t;
127 typedef abi_ushort target_gid_t;
128 #else
129 typedef abi_uint target_uid_t;
130 typedef abi_uint target_gid_t;
131 #endif
132 typedef abi_int target_pid_t;
133
134 #ifdef TARGET_I386
135
136 #define ELF_HWCAP get_elf_hwcap()
137
138 static uint32_t get_elf_hwcap(void)
139 {
140 X86CPU *cpu = X86_CPU(thread_cpu);
141
142 return cpu->env.features[FEAT_1_EDX];
143 }
144
145 #ifdef TARGET_X86_64
146 #define ELF_START_MMAP 0x2aaaaab000ULL
147
148 #define ELF_CLASS ELFCLASS64
149 #define ELF_ARCH EM_X86_64
150
151 #define ELF_PLATFORM "x86_64"
152
153 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop)
154 {
155 regs->rax = 0;
156 regs->rsp = infop->start_stack;
157 regs->rip = infop->entry;
158 }
159
160 #define ELF_NREG 27
161 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
162
163 /*
164 * Note that ELF_NREG should be 29 as there should be place for
165 * TRAPNO and ERR "registers" as well but linux doesn't dump
166 * those.
167 *
168 * See linux kernel: arch/x86/include/asm/elf.h
169 */
170 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUX86State *env)
171 {
172 (*regs)[0] = tswapreg(env->regs[15]);
173 (*regs)[1] = tswapreg(env->regs[14]);
174 (*regs)[2] = tswapreg(env->regs[13]);
175 (*regs)[3] = tswapreg(env->regs[12]);
176 (*regs)[4] = tswapreg(env->regs[R_EBP]);
177 (*regs)[5] = tswapreg(env->regs[R_EBX]);
178 (*regs)[6] = tswapreg(env->regs[11]);
179 (*regs)[7] = tswapreg(env->regs[10]);
180 (*regs)[8] = tswapreg(env->regs[9]);
181 (*regs)[9] = tswapreg(env->regs[8]);
182 (*regs)[10] = tswapreg(env->regs[R_EAX]);
183 (*regs)[11] = tswapreg(env->regs[R_ECX]);
184 (*regs)[12] = tswapreg(env->regs[R_EDX]);
185 (*regs)[13] = tswapreg(env->regs[R_ESI]);
186 (*regs)[14] = tswapreg(env->regs[R_EDI]);
187 (*regs)[15] = tswapreg(env->regs[R_EAX]); /* XXX */
188 (*regs)[16] = tswapreg(env->eip);
189 (*regs)[17] = tswapreg(env->segs[R_CS].selector & 0xffff);
190 (*regs)[18] = tswapreg(env->eflags);
191 (*regs)[19] = tswapreg(env->regs[R_ESP]);
192 (*regs)[20] = tswapreg(env->segs[R_SS].selector & 0xffff);
193 (*regs)[21] = tswapreg(env->segs[R_FS].selector & 0xffff);
194 (*regs)[22] = tswapreg(env->segs[R_GS].selector & 0xffff);
195 (*regs)[23] = tswapreg(env->segs[R_DS].selector & 0xffff);
196 (*regs)[24] = tswapreg(env->segs[R_ES].selector & 0xffff);
197 (*regs)[25] = tswapreg(env->segs[R_FS].selector & 0xffff);
198 (*regs)[26] = tswapreg(env->segs[R_GS].selector & 0xffff);
199 }
200
201 #if ULONG_MAX > UINT32_MAX
202 #define INIT_GUEST_COMMPAGE
203 static bool init_guest_commpage(void)
204 {
205 /*
206 * The vsyscall page is at a high negative address aka kernel space,
207 * which means that we cannot actually allocate it with target_mmap.
208 * We still should be able to use page_set_flags, unless the user
209 * has specified -R reserved_va, which would trigger an assert().
210 */
211 if (reserved_va != 0 &&
212 TARGET_VSYSCALL_PAGE + TARGET_PAGE_SIZE - 1 > reserved_va) {
213 error_report("Cannot allocate vsyscall page");
214 exit(EXIT_FAILURE);
215 }
216 page_set_flags(TARGET_VSYSCALL_PAGE,
217 TARGET_VSYSCALL_PAGE | ~TARGET_PAGE_MASK,
218 PAGE_EXEC | PAGE_VALID);
219 return true;
220 }
221 #endif
222 #else
223
224 #define ELF_START_MMAP 0x80000000
225
226 /*
227 * This is used to ensure we don't load something for the wrong architecture.
228 */
229 #define elf_check_arch(x) ( ((x) == EM_386) || ((x) == EM_486) )
230
231 /*
232 * These are used to set parameters in the core dumps.
233 */
234 #define ELF_CLASS ELFCLASS32
235 #define ELF_ARCH EM_386
236
237 #define ELF_PLATFORM get_elf_platform()
238 #define EXSTACK_DEFAULT true
239
240 static const char *get_elf_platform(void)
241 {
242 static char elf_platform[] = "i386";
243 int family = object_property_get_int(OBJECT(thread_cpu), "family", NULL);
244 if (family > 6) {
245 family = 6;
246 }
247 if (family >= 3) {
248 elf_platform[1] = '0' + family;
249 }
250 return elf_platform;
251 }
252
253 static inline void init_thread(struct target_pt_regs *regs,
254 struct image_info *infop)
255 {
256 regs->esp = infop->start_stack;
257 regs->eip = infop->entry;
258
259 /* SVR4/i386 ABI (pages 3-31, 3-32) says that when the program
260 starts %edx contains a pointer to a function which might be
261 registered using `atexit'. This provides a mean for the
262 dynamic linker to call DT_FINI functions for shared libraries
263 that have been loaded before the code runs.
264
265 A value of 0 tells we have no such handler. */
266 regs->edx = 0;
267 }
268
269 #define ELF_NREG 17
270 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
271
272 /*
273 * Note that ELF_NREG should be 19 as there should be place for
274 * TRAPNO and ERR "registers" as well but linux doesn't dump
275 * those.
276 *
277 * See linux kernel: arch/x86/include/asm/elf.h
278 */
279 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUX86State *env)
280 {
281 (*regs)[0] = tswapreg(env->regs[R_EBX]);
282 (*regs)[1] = tswapreg(env->regs[R_ECX]);
283 (*regs)[2] = tswapreg(env->regs[R_EDX]);
284 (*regs)[3] = tswapreg(env->regs[R_ESI]);
285 (*regs)[4] = tswapreg(env->regs[R_EDI]);
286 (*regs)[5] = tswapreg(env->regs[R_EBP]);
287 (*regs)[6] = tswapreg(env->regs[R_EAX]);
288 (*regs)[7] = tswapreg(env->segs[R_DS].selector & 0xffff);
289 (*regs)[8] = tswapreg(env->segs[R_ES].selector & 0xffff);
290 (*regs)[9] = tswapreg(env->segs[R_FS].selector & 0xffff);
291 (*regs)[10] = tswapreg(env->segs[R_GS].selector & 0xffff);
292 (*regs)[11] = tswapreg(env->regs[R_EAX]); /* XXX */
293 (*regs)[12] = tswapreg(env->eip);
294 (*regs)[13] = tswapreg(env->segs[R_CS].selector & 0xffff);
295 (*regs)[14] = tswapreg(env->eflags);
296 (*regs)[15] = tswapreg(env->regs[R_ESP]);
297 (*regs)[16] = tswapreg(env->segs[R_SS].selector & 0xffff);
298 }
299 #endif
300
301 #define USE_ELF_CORE_DUMP
302 #define ELF_EXEC_PAGESIZE 4096
303
304 #endif
305
306 #ifdef TARGET_ARM
307
308 #ifndef TARGET_AARCH64
309 /* 32 bit ARM definitions */
310
311 #define ELF_START_MMAP 0x80000000
312
313 #define ELF_ARCH EM_ARM
314 #define ELF_CLASS ELFCLASS32
315 #define EXSTACK_DEFAULT true
316
317 static inline void init_thread(struct target_pt_regs *regs,
318 struct image_info *infop)
319 {
320 abi_long stack = infop->start_stack;
321 memset(regs, 0, sizeof(*regs));
322
323 regs->uregs[16] = ARM_CPU_MODE_USR;
324 if (infop->entry & 1) {
325 regs->uregs[16] |= CPSR_T;
326 }
327 regs->uregs[15] = infop->entry & 0xfffffffe;
328 regs->uregs[13] = infop->start_stack;
329 /* FIXME - what to for failure of get_user()? */
330 get_user_ual(regs->uregs[2], stack + 8); /* envp */
331 get_user_ual(regs->uregs[1], stack + 4); /* envp */
332 /* XXX: it seems that r0 is zeroed after ! */
333 regs->uregs[0] = 0;
334 /* For uClinux PIC binaries. */
335 /* XXX: Linux does this only on ARM with no MMU (do we care ?) */
336 regs->uregs[10] = infop->start_data;
337
338 /* Support ARM FDPIC. */
339 if (info_is_fdpic(infop)) {
340 /* As described in the ABI document, r7 points to the loadmap info
341 * prepared by the kernel. If an interpreter is needed, r8 points
342 * to the interpreter loadmap and r9 points to the interpreter
343 * PT_DYNAMIC info. If no interpreter is needed, r8 is zero, and
344 * r9 points to the main program PT_DYNAMIC info.
345 */
346 regs->uregs[7] = infop->loadmap_addr;
347 if (infop->interpreter_loadmap_addr) {
348 /* Executable is dynamically loaded. */
349 regs->uregs[8] = infop->interpreter_loadmap_addr;
350 regs->uregs[9] = infop->interpreter_pt_dynamic_addr;
351 } else {
352 regs->uregs[8] = 0;
353 regs->uregs[9] = infop->pt_dynamic_addr;
354 }
355 }
356 }
357
358 #define ELF_NREG 18
359 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
360
361 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUARMState *env)
362 {
363 (*regs)[0] = tswapreg(env->regs[0]);
364 (*regs)[1] = tswapreg(env->regs[1]);
365 (*regs)[2] = tswapreg(env->regs[2]);
366 (*regs)[3] = tswapreg(env->regs[3]);
367 (*regs)[4] = tswapreg(env->regs[4]);
368 (*regs)[5] = tswapreg(env->regs[5]);
369 (*regs)[6] = tswapreg(env->regs[6]);
370 (*regs)[7] = tswapreg(env->regs[7]);
371 (*regs)[8] = tswapreg(env->regs[8]);
372 (*regs)[9] = tswapreg(env->regs[9]);
373 (*regs)[10] = tswapreg(env->regs[10]);
374 (*regs)[11] = tswapreg(env->regs[11]);
375 (*regs)[12] = tswapreg(env->regs[12]);
376 (*regs)[13] = tswapreg(env->regs[13]);
377 (*regs)[14] = tswapreg(env->regs[14]);
378 (*regs)[15] = tswapreg(env->regs[15]);
379
380 (*regs)[16] = tswapreg(cpsr_read((CPUARMState *)env));
381 (*regs)[17] = tswapreg(env->regs[0]); /* XXX */
382 }
383
384 #define USE_ELF_CORE_DUMP
385 #define ELF_EXEC_PAGESIZE 4096
386
387 enum
388 {
389 ARM_HWCAP_ARM_SWP = 1 << 0,
390 ARM_HWCAP_ARM_HALF = 1 << 1,
391 ARM_HWCAP_ARM_THUMB = 1 << 2,
392 ARM_HWCAP_ARM_26BIT = 1 << 3,
393 ARM_HWCAP_ARM_FAST_MULT = 1 << 4,
394 ARM_HWCAP_ARM_FPA = 1 << 5,
395 ARM_HWCAP_ARM_VFP = 1 << 6,
396 ARM_HWCAP_ARM_EDSP = 1 << 7,
397 ARM_HWCAP_ARM_JAVA = 1 << 8,
398 ARM_HWCAP_ARM_IWMMXT = 1 << 9,
399 ARM_HWCAP_ARM_CRUNCH = 1 << 10,
400 ARM_HWCAP_ARM_THUMBEE = 1 << 11,
401 ARM_HWCAP_ARM_NEON = 1 << 12,
402 ARM_HWCAP_ARM_VFPv3 = 1 << 13,
403 ARM_HWCAP_ARM_VFPv3D16 = 1 << 14,
404 ARM_HWCAP_ARM_TLS = 1 << 15,
405 ARM_HWCAP_ARM_VFPv4 = 1 << 16,
406 ARM_HWCAP_ARM_IDIVA = 1 << 17,
407 ARM_HWCAP_ARM_IDIVT = 1 << 18,
408 ARM_HWCAP_ARM_VFPD32 = 1 << 19,
409 ARM_HWCAP_ARM_LPAE = 1 << 20,
410 ARM_HWCAP_ARM_EVTSTRM = 1 << 21,
411 };
412
413 enum {
414 ARM_HWCAP2_ARM_AES = 1 << 0,
415 ARM_HWCAP2_ARM_PMULL = 1 << 1,
416 ARM_HWCAP2_ARM_SHA1 = 1 << 2,
417 ARM_HWCAP2_ARM_SHA2 = 1 << 3,
418 ARM_HWCAP2_ARM_CRC32 = 1 << 4,
419 };
420
421 /* The commpage only exists for 32 bit kernels */
422
423 #define HI_COMMPAGE (intptr_t)0xffff0f00u
424
425 static bool init_guest_commpage(void)
426 {
427 ARMCPU *cpu = ARM_CPU(thread_cpu);
428 abi_ptr commpage;
429 void *want;
430 void *addr;
431
432 /*
433 * M-profile allocates maximum of 2GB address space, so can never
434 * allocate the commpage. Skip it.
435 */
436 if (arm_feature(&cpu->env, ARM_FEATURE_M)) {
437 return true;
438 }
439
440 commpage = HI_COMMPAGE & -qemu_host_page_size;
441 want = g2h_untagged(commpage);
442 addr = mmap(want, qemu_host_page_size, PROT_READ | PROT_WRITE,
443 MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0);
444
445 if (addr == MAP_FAILED) {
446 perror("Allocating guest commpage");
447 exit(EXIT_FAILURE);
448 }
449 if (addr != want) {
450 return false;
451 }
452
453 /* Set kernel helper versions; rest of page is 0. */
454 __put_user(5, (uint32_t *)g2h_untagged(0xffff0ffcu));
455
456 if (mprotect(addr, qemu_host_page_size, PROT_READ)) {
457 perror("Protecting guest commpage");
458 exit(EXIT_FAILURE);
459 }
460
461 page_set_flags(commpage, commpage | ~qemu_host_page_mask,
462 PAGE_READ | PAGE_EXEC | PAGE_VALID);
463 return true;
464 }
465
466 #define ELF_HWCAP get_elf_hwcap()
467 #define ELF_HWCAP2 get_elf_hwcap2()
468
469 static uint32_t get_elf_hwcap(void)
470 {
471 ARMCPU *cpu = ARM_CPU(thread_cpu);
472 uint32_t hwcaps = 0;
473
474 hwcaps |= ARM_HWCAP_ARM_SWP;
475 hwcaps |= ARM_HWCAP_ARM_HALF;
476 hwcaps |= ARM_HWCAP_ARM_THUMB;
477 hwcaps |= ARM_HWCAP_ARM_FAST_MULT;
478
479 /* probe for the extra features */
480 #define GET_FEATURE(feat, hwcap) \
481 do { if (arm_feature(&cpu->env, feat)) { hwcaps |= hwcap; } } while (0)
482
483 #define GET_FEATURE_ID(feat, hwcap) \
484 do { if (cpu_isar_feature(feat, cpu)) { hwcaps |= hwcap; } } while (0)
485
486 /* EDSP is in v5TE and above, but all our v5 CPUs are v5TE */
487 GET_FEATURE(ARM_FEATURE_V5, ARM_HWCAP_ARM_EDSP);
488 GET_FEATURE(ARM_FEATURE_IWMMXT, ARM_HWCAP_ARM_IWMMXT);
489 GET_FEATURE(ARM_FEATURE_THUMB2EE, ARM_HWCAP_ARM_THUMBEE);
490 GET_FEATURE(ARM_FEATURE_NEON, ARM_HWCAP_ARM_NEON);
491 GET_FEATURE(ARM_FEATURE_V6K, ARM_HWCAP_ARM_TLS);
492 GET_FEATURE(ARM_FEATURE_LPAE, ARM_HWCAP_ARM_LPAE);
493 GET_FEATURE_ID(aa32_arm_div, ARM_HWCAP_ARM_IDIVA);
494 GET_FEATURE_ID(aa32_thumb_div, ARM_HWCAP_ARM_IDIVT);
495 GET_FEATURE_ID(aa32_vfp, ARM_HWCAP_ARM_VFP);
496
497 if (cpu_isar_feature(aa32_fpsp_v3, cpu) ||
498 cpu_isar_feature(aa32_fpdp_v3, cpu)) {
499 hwcaps |= ARM_HWCAP_ARM_VFPv3;
500 if (cpu_isar_feature(aa32_simd_r32, cpu)) {
501 hwcaps |= ARM_HWCAP_ARM_VFPD32;
502 } else {
503 hwcaps |= ARM_HWCAP_ARM_VFPv3D16;
504 }
505 }
506 GET_FEATURE_ID(aa32_simdfmac, ARM_HWCAP_ARM_VFPv4);
507
508 return hwcaps;
509 }
510
511 static uint32_t get_elf_hwcap2(void)
512 {
513 ARMCPU *cpu = ARM_CPU(thread_cpu);
514 uint32_t hwcaps = 0;
515
516 GET_FEATURE_ID(aa32_aes, ARM_HWCAP2_ARM_AES);
517 GET_FEATURE_ID(aa32_pmull, ARM_HWCAP2_ARM_PMULL);
518 GET_FEATURE_ID(aa32_sha1, ARM_HWCAP2_ARM_SHA1);
519 GET_FEATURE_ID(aa32_sha2, ARM_HWCAP2_ARM_SHA2);
520 GET_FEATURE_ID(aa32_crc32, ARM_HWCAP2_ARM_CRC32);
521 return hwcaps;
522 }
523
524 #undef GET_FEATURE
525 #undef GET_FEATURE_ID
526
527 #define ELF_PLATFORM get_elf_platform()
528
529 static const char *get_elf_platform(void)
530 {
531 CPUARMState *env = thread_cpu->env_ptr;
532
533 #if TARGET_BIG_ENDIAN
534 # define END "b"
535 #else
536 # define END "l"
537 #endif
538
539 if (arm_feature(env, ARM_FEATURE_V8)) {
540 return "v8" END;
541 } else if (arm_feature(env, ARM_FEATURE_V7)) {
542 if (arm_feature(env, ARM_FEATURE_M)) {
543 return "v7m" END;
544 } else {
545 return "v7" END;
546 }
547 } else if (arm_feature(env, ARM_FEATURE_V6)) {
548 return "v6" END;
549 } else if (arm_feature(env, ARM_FEATURE_V5)) {
550 return "v5" END;
551 } else {
552 return "v4" END;
553 }
554
555 #undef END
556 }
557
558 #else
559 /* 64 bit ARM definitions */
560 #define ELF_START_MMAP 0x80000000
561
562 #define ELF_ARCH EM_AARCH64
563 #define ELF_CLASS ELFCLASS64
564 #if TARGET_BIG_ENDIAN
565 # define ELF_PLATFORM "aarch64_be"
566 #else
567 # define ELF_PLATFORM "aarch64"
568 #endif
569
570 static inline void init_thread(struct target_pt_regs *regs,
571 struct image_info *infop)
572 {
573 abi_long stack = infop->start_stack;
574 memset(regs, 0, sizeof(*regs));
575
576 regs->pc = infop->entry & ~0x3ULL;
577 regs->sp = stack;
578 }
579
580 #define ELF_NREG 34
581 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
582
583 static void elf_core_copy_regs(target_elf_gregset_t *regs,
584 const CPUARMState *env)
585 {
586 int i;
587
588 for (i = 0; i < 32; i++) {
589 (*regs)[i] = tswapreg(env->xregs[i]);
590 }
591 (*regs)[32] = tswapreg(env->pc);
592 (*regs)[33] = tswapreg(pstate_read((CPUARMState *)env));
593 }
594
595 #define USE_ELF_CORE_DUMP
596 #define ELF_EXEC_PAGESIZE 4096
597
598 enum {
599 ARM_HWCAP_A64_FP = 1 << 0,
600 ARM_HWCAP_A64_ASIMD = 1 << 1,
601 ARM_HWCAP_A64_EVTSTRM = 1 << 2,
602 ARM_HWCAP_A64_AES = 1 << 3,
603 ARM_HWCAP_A64_PMULL = 1 << 4,
604 ARM_HWCAP_A64_SHA1 = 1 << 5,
605 ARM_HWCAP_A64_SHA2 = 1 << 6,
606 ARM_HWCAP_A64_CRC32 = 1 << 7,
607 ARM_HWCAP_A64_ATOMICS = 1 << 8,
608 ARM_HWCAP_A64_FPHP = 1 << 9,
609 ARM_HWCAP_A64_ASIMDHP = 1 << 10,
610 ARM_HWCAP_A64_CPUID = 1 << 11,
611 ARM_HWCAP_A64_ASIMDRDM = 1 << 12,
612 ARM_HWCAP_A64_JSCVT = 1 << 13,
613 ARM_HWCAP_A64_FCMA = 1 << 14,
614 ARM_HWCAP_A64_LRCPC = 1 << 15,
615 ARM_HWCAP_A64_DCPOP = 1 << 16,
616 ARM_HWCAP_A64_SHA3 = 1 << 17,
617 ARM_HWCAP_A64_SM3 = 1 << 18,
618 ARM_HWCAP_A64_SM4 = 1 << 19,
619 ARM_HWCAP_A64_ASIMDDP = 1 << 20,
620 ARM_HWCAP_A64_SHA512 = 1 << 21,
621 ARM_HWCAP_A64_SVE = 1 << 22,
622 ARM_HWCAP_A64_ASIMDFHM = 1 << 23,
623 ARM_HWCAP_A64_DIT = 1 << 24,
624 ARM_HWCAP_A64_USCAT = 1 << 25,
625 ARM_HWCAP_A64_ILRCPC = 1 << 26,
626 ARM_HWCAP_A64_FLAGM = 1 << 27,
627 ARM_HWCAP_A64_SSBS = 1 << 28,
628 ARM_HWCAP_A64_SB = 1 << 29,
629 ARM_HWCAP_A64_PACA = 1 << 30,
630 ARM_HWCAP_A64_PACG = 1UL << 31,
631
632 ARM_HWCAP2_A64_DCPODP = 1 << 0,
633 ARM_HWCAP2_A64_SVE2 = 1 << 1,
634 ARM_HWCAP2_A64_SVEAES = 1 << 2,
635 ARM_HWCAP2_A64_SVEPMULL = 1 << 3,
636 ARM_HWCAP2_A64_SVEBITPERM = 1 << 4,
637 ARM_HWCAP2_A64_SVESHA3 = 1 << 5,
638 ARM_HWCAP2_A64_SVESM4 = 1 << 6,
639 ARM_HWCAP2_A64_FLAGM2 = 1 << 7,
640 ARM_HWCAP2_A64_FRINT = 1 << 8,
641 ARM_HWCAP2_A64_SVEI8MM = 1 << 9,
642 ARM_HWCAP2_A64_SVEF32MM = 1 << 10,
643 ARM_HWCAP2_A64_SVEF64MM = 1 << 11,
644 ARM_HWCAP2_A64_SVEBF16 = 1 << 12,
645 ARM_HWCAP2_A64_I8MM = 1 << 13,
646 ARM_HWCAP2_A64_BF16 = 1 << 14,
647 ARM_HWCAP2_A64_DGH = 1 << 15,
648 ARM_HWCAP2_A64_RNG = 1 << 16,
649 ARM_HWCAP2_A64_BTI = 1 << 17,
650 ARM_HWCAP2_A64_MTE = 1 << 18,
651 ARM_HWCAP2_A64_ECV = 1 << 19,
652 ARM_HWCAP2_A64_AFP = 1 << 20,
653 ARM_HWCAP2_A64_RPRES = 1 << 21,
654 ARM_HWCAP2_A64_MTE3 = 1 << 22,
655 ARM_HWCAP2_A64_SME = 1 << 23,
656 ARM_HWCAP2_A64_SME_I16I64 = 1 << 24,
657 ARM_HWCAP2_A64_SME_F64F64 = 1 << 25,
658 ARM_HWCAP2_A64_SME_I8I32 = 1 << 26,
659 ARM_HWCAP2_A64_SME_F16F32 = 1 << 27,
660 ARM_HWCAP2_A64_SME_B16F32 = 1 << 28,
661 ARM_HWCAP2_A64_SME_F32F32 = 1 << 29,
662 ARM_HWCAP2_A64_SME_FA64 = 1 << 30,
663 };
664
665 #define ELF_HWCAP get_elf_hwcap()
666 #define ELF_HWCAP2 get_elf_hwcap2()
667
668 #define GET_FEATURE_ID(feat, hwcap) \
669 do { if (cpu_isar_feature(feat, cpu)) { hwcaps |= hwcap; } } while (0)
670
671 static uint32_t get_elf_hwcap(void)
672 {
673 ARMCPU *cpu = ARM_CPU(thread_cpu);
674 uint32_t hwcaps = 0;
675
676 hwcaps |= ARM_HWCAP_A64_FP;
677 hwcaps |= ARM_HWCAP_A64_ASIMD;
678 hwcaps |= ARM_HWCAP_A64_CPUID;
679
680 /* probe for the extra features */
681
682 GET_FEATURE_ID(aa64_aes, ARM_HWCAP_A64_AES);
683 GET_FEATURE_ID(aa64_pmull, ARM_HWCAP_A64_PMULL);
684 GET_FEATURE_ID(aa64_sha1, ARM_HWCAP_A64_SHA1);
685 GET_FEATURE_ID(aa64_sha256, ARM_HWCAP_A64_SHA2);
686 GET_FEATURE_ID(aa64_sha512, ARM_HWCAP_A64_SHA512);
687 GET_FEATURE_ID(aa64_crc32, ARM_HWCAP_A64_CRC32);
688 GET_FEATURE_ID(aa64_sha3, ARM_HWCAP_A64_SHA3);
689 GET_FEATURE_ID(aa64_sm3, ARM_HWCAP_A64_SM3);
690 GET_FEATURE_ID(aa64_sm4, ARM_HWCAP_A64_SM4);
691 GET_FEATURE_ID(aa64_fp16, ARM_HWCAP_A64_FPHP | ARM_HWCAP_A64_ASIMDHP);
692 GET_FEATURE_ID(aa64_atomics, ARM_HWCAP_A64_ATOMICS);
693 GET_FEATURE_ID(aa64_rdm, ARM_HWCAP_A64_ASIMDRDM);
694 GET_FEATURE_ID(aa64_dp, ARM_HWCAP_A64_ASIMDDP);
695 GET_FEATURE_ID(aa64_fcma, ARM_HWCAP_A64_FCMA);
696 GET_FEATURE_ID(aa64_sve, ARM_HWCAP_A64_SVE);
697 GET_FEATURE_ID(aa64_pauth, ARM_HWCAP_A64_PACA | ARM_HWCAP_A64_PACG);
698 GET_FEATURE_ID(aa64_fhm, ARM_HWCAP_A64_ASIMDFHM);
699 GET_FEATURE_ID(aa64_jscvt, ARM_HWCAP_A64_JSCVT);
700 GET_FEATURE_ID(aa64_sb, ARM_HWCAP_A64_SB);
701 GET_FEATURE_ID(aa64_condm_4, ARM_HWCAP_A64_FLAGM);
702 GET_FEATURE_ID(aa64_dcpop, ARM_HWCAP_A64_DCPOP);
703 GET_FEATURE_ID(aa64_rcpc_8_3, ARM_HWCAP_A64_LRCPC);
704 GET_FEATURE_ID(aa64_rcpc_8_4, ARM_HWCAP_A64_ILRCPC);
705
706 return hwcaps;
707 }
708
709 static uint32_t get_elf_hwcap2(void)
710 {
711 ARMCPU *cpu = ARM_CPU(thread_cpu);
712 uint32_t hwcaps = 0;
713
714 GET_FEATURE_ID(aa64_dcpodp, ARM_HWCAP2_A64_DCPODP);
715 GET_FEATURE_ID(aa64_sve2, ARM_HWCAP2_A64_SVE2);
716 GET_FEATURE_ID(aa64_sve2_aes, ARM_HWCAP2_A64_SVEAES);
717 GET_FEATURE_ID(aa64_sve2_pmull128, ARM_HWCAP2_A64_SVEPMULL);
718 GET_FEATURE_ID(aa64_sve2_bitperm, ARM_HWCAP2_A64_SVEBITPERM);
719 GET_FEATURE_ID(aa64_sve2_sha3, ARM_HWCAP2_A64_SVESHA3);
720 GET_FEATURE_ID(aa64_sve2_sm4, ARM_HWCAP2_A64_SVESM4);
721 GET_FEATURE_ID(aa64_condm_5, ARM_HWCAP2_A64_FLAGM2);
722 GET_FEATURE_ID(aa64_frint, ARM_HWCAP2_A64_FRINT);
723 GET_FEATURE_ID(aa64_sve_i8mm, ARM_HWCAP2_A64_SVEI8MM);
724 GET_FEATURE_ID(aa64_sve_f32mm, ARM_HWCAP2_A64_SVEF32MM);
725 GET_FEATURE_ID(aa64_sve_f64mm, ARM_HWCAP2_A64_SVEF64MM);
726 GET_FEATURE_ID(aa64_sve_bf16, ARM_HWCAP2_A64_SVEBF16);
727 GET_FEATURE_ID(aa64_i8mm, ARM_HWCAP2_A64_I8MM);
728 GET_FEATURE_ID(aa64_bf16, ARM_HWCAP2_A64_BF16);
729 GET_FEATURE_ID(aa64_rndr, ARM_HWCAP2_A64_RNG);
730 GET_FEATURE_ID(aa64_bti, ARM_HWCAP2_A64_BTI);
731 GET_FEATURE_ID(aa64_mte, ARM_HWCAP2_A64_MTE);
732 GET_FEATURE_ID(aa64_sme, (ARM_HWCAP2_A64_SME |
733 ARM_HWCAP2_A64_SME_F32F32 |
734 ARM_HWCAP2_A64_SME_B16F32 |
735 ARM_HWCAP2_A64_SME_F16F32 |
736 ARM_HWCAP2_A64_SME_I8I32));
737 GET_FEATURE_ID(aa64_sme_f64f64, ARM_HWCAP2_A64_SME_F64F64);
738 GET_FEATURE_ID(aa64_sme_i16i64, ARM_HWCAP2_A64_SME_I16I64);
739 GET_FEATURE_ID(aa64_sme_fa64, ARM_HWCAP2_A64_SME_FA64);
740
741 return hwcaps;
742 }
743
744 #undef GET_FEATURE_ID
745
746 #endif /* not TARGET_AARCH64 */
747 #endif /* TARGET_ARM */
748
749 #ifdef TARGET_SPARC
750 #ifdef TARGET_SPARC64
751
752 #define ELF_START_MMAP 0x80000000
753 #define ELF_HWCAP (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \
754 | HWCAP_SPARC_MULDIV | HWCAP_SPARC_V9)
755 #ifndef TARGET_ABI32
756 #define elf_check_arch(x) ( (x) == EM_SPARCV9 || (x) == EM_SPARC32PLUS )
757 #else
758 #define elf_check_arch(x) ( (x) == EM_SPARC32PLUS || (x) == EM_SPARC )
759 #endif
760
761 #define ELF_CLASS ELFCLASS64
762 #define ELF_ARCH EM_SPARCV9
763 #else
764 #define ELF_START_MMAP 0x80000000
765 #define ELF_HWCAP (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR | HWCAP_SPARC_SWAP \
766 | HWCAP_SPARC_MULDIV)
767 #define ELF_CLASS ELFCLASS32
768 #define ELF_ARCH EM_SPARC
769 #endif /* TARGET_SPARC64 */
770
771 static inline void init_thread(struct target_pt_regs *regs,
772 struct image_info *infop)
773 {
774 /* Note that target_cpu_copy_regs does not read psr/tstate. */
775 regs->pc = infop->entry;
776 regs->npc = regs->pc + 4;
777 regs->y = 0;
778 regs->u_regs[14] = (infop->start_stack - 16 * sizeof(abi_ulong)
779 - TARGET_STACK_BIAS);
780 }
781 #endif /* TARGET_SPARC */
782
783 #ifdef TARGET_PPC
784
785 #define ELF_MACHINE PPC_ELF_MACHINE
786 #define ELF_START_MMAP 0x80000000
787
788 #if defined(TARGET_PPC64)
789
790 #define elf_check_arch(x) ( (x) == EM_PPC64 )
791
792 #define ELF_CLASS ELFCLASS64
793
794 #else
795
796 #define ELF_CLASS ELFCLASS32
797 #define EXSTACK_DEFAULT true
798
799 #endif
800
801 #define ELF_ARCH EM_PPC
802
803 /* Feature masks for the Aux Vector Hardware Capabilities (AT_HWCAP).
804 See arch/powerpc/include/asm/cputable.h. */
805 enum {
806 QEMU_PPC_FEATURE_32 = 0x80000000,
807 QEMU_PPC_FEATURE_64 = 0x40000000,
808 QEMU_PPC_FEATURE_601_INSTR = 0x20000000,
809 QEMU_PPC_FEATURE_HAS_ALTIVEC = 0x10000000,
810 QEMU_PPC_FEATURE_HAS_FPU = 0x08000000,
811 QEMU_PPC_FEATURE_HAS_MMU = 0x04000000,
812 QEMU_PPC_FEATURE_HAS_4xxMAC = 0x02000000,
813 QEMU_PPC_FEATURE_UNIFIED_CACHE = 0x01000000,
814 QEMU_PPC_FEATURE_HAS_SPE = 0x00800000,
815 QEMU_PPC_FEATURE_HAS_EFP_SINGLE = 0x00400000,
816 QEMU_PPC_FEATURE_HAS_EFP_DOUBLE = 0x00200000,
817 QEMU_PPC_FEATURE_NO_TB = 0x00100000,
818 QEMU_PPC_FEATURE_POWER4 = 0x00080000,
819 QEMU_PPC_FEATURE_POWER5 = 0x00040000,
820 QEMU_PPC_FEATURE_POWER5_PLUS = 0x00020000,
821 QEMU_PPC_FEATURE_CELL = 0x00010000,
822 QEMU_PPC_FEATURE_BOOKE = 0x00008000,
823 QEMU_PPC_FEATURE_SMT = 0x00004000,
824 QEMU_PPC_FEATURE_ICACHE_SNOOP = 0x00002000,
825 QEMU_PPC_FEATURE_ARCH_2_05 = 0x00001000,
826 QEMU_PPC_FEATURE_PA6T = 0x00000800,
827 QEMU_PPC_FEATURE_HAS_DFP = 0x00000400,
828 QEMU_PPC_FEATURE_POWER6_EXT = 0x00000200,
829 QEMU_PPC_FEATURE_ARCH_2_06 = 0x00000100,
830 QEMU_PPC_FEATURE_HAS_VSX = 0x00000080,
831 QEMU_PPC_FEATURE_PSERIES_PERFMON_COMPAT = 0x00000040,
832
833 QEMU_PPC_FEATURE_TRUE_LE = 0x00000002,
834 QEMU_PPC_FEATURE_PPC_LE = 0x00000001,
835
836 /* Feature definitions in AT_HWCAP2. */
837 QEMU_PPC_FEATURE2_ARCH_2_07 = 0x80000000, /* ISA 2.07 */
838 QEMU_PPC_FEATURE2_HAS_HTM = 0x40000000, /* Hardware Transactional Memory */
839 QEMU_PPC_FEATURE2_HAS_DSCR = 0x20000000, /* Data Stream Control Register */
840 QEMU_PPC_FEATURE2_HAS_EBB = 0x10000000, /* Event Base Branching */
841 QEMU_PPC_FEATURE2_HAS_ISEL = 0x08000000, /* Integer Select */
842 QEMU_PPC_FEATURE2_HAS_TAR = 0x04000000, /* Target Address Register */
843 QEMU_PPC_FEATURE2_VEC_CRYPTO = 0x02000000,
844 QEMU_PPC_FEATURE2_HTM_NOSC = 0x01000000,
845 QEMU_PPC_FEATURE2_ARCH_3_00 = 0x00800000, /* ISA 3.00 */
846 QEMU_PPC_FEATURE2_HAS_IEEE128 = 0x00400000, /* VSX IEEE Bin Float 128-bit */
847 QEMU_PPC_FEATURE2_DARN = 0x00200000, /* darn random number insn */
848 QEMU_PPC_FEATURE2_SCV = 0x00100000, /* scv syscall */
849 QEMU_PPC_FEATURE2_HTM_NO_SUSPEND = 0x00080000, /* TM w/o suspended state */
850 QEMU_PPC_FEATURE2_ARCH_3_1 = 0x00040000, /* ISA 3.1 */
851 QEMU_PPC_FEATURE2_MMA = 0x00020000, /* Matrix-Multiply Assist */
852 };
853
854 #define ELF_HWCAP get_elf_hwcap()
855
856 static uint32_t get_elf_hwcap(void)
857 {
858 PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);
859 uint32_t features = 0;
860
861 /* We don't have to be terribly complete here; the high points are
862 Altivec/FP/SPE support. Anything else is just a bonus. */
863 #define GET_FEATURE(flag, feature) \
864 do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0)
865 #define GET_FEATURE2(flags, feature) \
866 do { \
867 if ((cpu->env.insns_flags2 & flags) == flags) { \
868 features |= feature; \
869 } \
870 } while (0)
871 GET_FEATURE(PPC_64B, QEMU_PPC_FEATURE_64);
872 GET_FEATURE(PPC_FLOAT, QEMU_PPC_FEATURE_HAS_FPU);
873 GET_FEATURE(PPC_ALTIVEC, QEMU_PPC_FEATURE_HAS_ALTIVEC);
874 GET_FEATURE(PPC_SPE, QEMU_PPC_FEATURE_HAS_SPE);
875 GET_FEATURE(PPC_SPE_SINGLE, QEMU_PPC_FEATURE_HAS_EFP_SINGLE);
876 GET_FEATURE(PPC_SPE_DOUBLE, QEMU_PPC_FEATURE_HAS_EFP_DOUBLE);
877 GET_FEATURE(PPC_BOOKE, QEMU_PPC_FEATURE_BOOKE);
878 GET_FEATURE(PPC_405_MAC, QEMU_PPC_FEATURE_HAS_4xxMAC);
879 GET_FEATURE2(PPC2_DFP, QEMU_PPC_FEATURE_HAS_DFP);
880 GET_FEATURE2(PPC2_VSX, QEMU_PPC_FEATURE_HAS_VSX);
881 GET_FEATURE2((PPC2_PERM_ISA206 | PPC2_DIVE_ISA206 | PPC2_ATOMIC_ISA206 |
882 PPC2_FP_CVT_ISA206 | PPC2_FP_TST_ISA206),
883 QEMU_PPC_FEATURE_ARCH_2_06);
884 #undef GET_FEATURE
885 #undef GET_FEATURE2
886
887 return features;
888 }
889
890 #define ELF_HWCAP2 get_elf_hwcap2()
891
892 static uint32_t get_elf_hwcap2(void)
893 {
894 PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);
895 uint32_t features = 0;
896
897 #define GET_FEATURE(flag, feature) \
898 do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0)
899 #define GET_FEATURE2(flag, feature) \
900 do { if (cpu->env.insns_flags2 & flag) { features |= feature; } } while (0)
901
902 GET_FEATURE(PPC_ISEL, QEMU_PPC_FEATURE2_HAS_ISEL);
903 GET_FEATURE2(PPC2_BCTAR_ISA207, QEMU_PPC_FEATURE2_HAS_TAR);
904 GET_FEATURE2((PPC2_BCTAR_ISA207 | PPC2_LSQ_ISA207 | PPC2_ALTIVEC_207 |
905 PPC2_ISA207S), QEMU_PPC_FEATURE2_ARCH_2_07 |
906 QEMU_PPC_FEATURE2_VEC_CRYPTO);
907 GET_FEATURE2(PPC2_ISA300, QEMU_PPC_FEATURE2_ARCH_3_00 |
908 QEMU_PPC_FEATURE2_DARN | QEMU_PPC_FEATURE2_HAS_IEEE128);
909 GET_FEATURE2(PPC2_ISA310, QEMU_PPC_FEATURE2_ARCH_3_1 |
910 QEMU_PPC_FEATURE2_MMA);
911
912 #undef GET_FEATURE
913 #undef GET_FEATURE2
914
915 return features;
916 }
917
918 /*
919 * The requirements here are:
920 * - keep the final alignment of sp (sp & 0xf)
921 * - make sure the 32-bit value at the first 16 byte aligned position of
922 * AUXV is greater than 16 for glibc compatibility.
923 * AT_IGNOREPPC is used for that.
924 * - for compatibility with glibc ARCH_DLINFO must always be defined on PPC,
925 * even if DLINFO_ARCH_ITEMS goes to zero or is undefined.
926 */
927 #define DLINFO_ARCH_ITEMS 5
928 #define ARCH_DLINFO \
929 do { \
930 PowerPCCPU *cpu = POWERPC_CPU(thread_cpu); \
931 /* \
932 * Handle glibc compatibility: these magic entries must \
933 * be at the lowest addresses in the final auxv. \
934 */ \
935 NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC); \
936 NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC); \
937 NEW_AUX_ENT(AT_DCACHEBSIZE, cpu->env.dcache_line_size); \
938 NEW_AUX_ENT(AT_ICACHEBSIZE, cpu->env.icache_line_size); \
939 NEW_AUX_ENT(AT_UCACHEBSIZE, 0); \
940 } while (0)
941
942 static inline void init_thread(struct target_pt_regs *_regs, struct image_info *infop)
943 {
944 _regs->gpr[1] = infop->start_stack;
945 #if defined(TARGET_PPC64)
946 if (get_ppc64_abi(infop) < 2) {
947 uint64_t val;
948 get_user_u64(val, infop->entry + 8);
949 _regs->gpr[2] = val + infop->load_bias;
950 get_user_u64(val, infop->entry);
951 infop->entry = val + infop->load_bias;
952 } else {
953 _regs->gpr[12] = infop->entry; /* r12 set to global entry address */
954 }
955 #endif
956 _regs->nip = infop->entry;
957 }
958
959 /* See linux kernel: arch/powerpc/include/asm/elf.h. */
960 #define ELF_NREG 48
961 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
962
963 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUPPCState *env)
964 {
965 int i;
966 target_ulong ccr = 0;
967
968 for (i = 0; i < ARRAY_SIZE(env->gpr); i++) {
969 (*regs)[i] = tswapreg(env->gpr[i]);
970 }
971
972 (*regs)[32] = tswapreg(env->nip);
973 (*regs)[33] = tswapreg(env->msr);
974 (*regs)[35] = tswapreg(env->ctr);
975 (*regs)[36] = tswapreg(env->lr);
976 (*regs)[37] = tswapreg(cpu_read_xer(env));
977
978 ccr = ppc_get_cr(env);
979 (*regs)[38] = tswapreg(ccr);
980 }
981
982 #define USE_ELF_CORE_DUMP
983 #define ELF_EXEC_PAGESIZE 4096
984
985 #endif
986
987 #ifdef TARGET_LOONGARCH64
988
989 #define ELF_START_MMAP 0x80000000
990
991 #define ELF_CLASS ELFCLASS64
992 #define ELF_ARCH EM_LOONGARCH
993 #define EXSTACK_DEFAULT true
994
995 #define elf_check_arch(x) ((x) == EM_LOONGARCH)
996
997 static inline void init_thread(struct target_pt_regs *regs,
998 struct image_info *infop)
999 {
1000 /*Set crmd PG,DA = 1,0 */
1001 regs->csr.crmd = 2 << 3;
1002 regs->csr.era = infop->entry;
1003 regs->regs[3] = infop->start_stack;
1004 }
1005
1006 /* See linux kernel: arch/loongarch/include/asm/elf.h */
1007 #define ELF_NREG 45
1008 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1009
1010 enum {
1011 TARGET_EF_R0 = 0,
1012 TARGET_EF_CSR_ERA = TARGET_EF_R0 + 33,
1013 TARGET_EF_CSR_BADV = TARGET_EF_R0 + 34,
1014 };
1015
1016 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1017 const CPULoongArchState *env)
1018 {
1019 int i;
1020
1021 (*regs)[TARGET_EF_R0] = 0;
1022
1023 for (i = 1; i < ARRAY_SIZE(env->gpr); i++) {
1024 (*regs)[TARGET_EF_R0 + i] = tswapreg(env->gpr[i]);
1025 }
1026
1027 (*regs)[TARGET_EF_CSR_ERA] = tswapreg(env->pc);
1028 (*regs)[TARGET_EF_CSR_BADV] = tswapreg(env->CSR_BADV);
1029 }
1030
1031 #define USE_ELF_CORE_DUMP
1032 #define ELF_EXEC_PAGESIZE 4096
1033
1034 #define ELF_HWCAP get_elf_hwcap()
1035
1036 /* See arch/loongarch/include/uapi/asm/hwcap.h */
1037 enum {
1038 HWCAP_LOONGARCH_CPUCFG = (1 << 0),
1039 HWCAP_LOONGARCH_LAM = (1 << 1),
1040 HWCAP_LOONGARCH_UAL = (1 << 2),
1041 HWCAP_LOONGARCH_FPU = (1 << 3),
1042 HWCAP_LOONGARCH_LSX = (1 << 4),
1043 HWCAP_LOONGARCH_LASX = (1 << 5),
1044 HWCAP_LOONGARCH_CRC32 = (1 << 6),
1045 HWCAP_LOONGARCH_COMPLEX = (1 << 7),
1046 HWCAP_LOONGARCH_CRYPTO = (1 << 8),
1047 HWCAP_LOONGARCH_LVZ = (1 << 9),
1048 HWCAP_LOONGARCH_LBT_X86 = (1 << 10),
1049 HWCAP_LOONGARCH_LBT_ARM = (1 << 11),
1050 HWCAP_LOONGARCH_LBT_MIPS = (1 << 12),
1051 };
1052
1053 static uint32_t get_elf_hwcap(void)
1054 {
1055 LoongArchCPU *cpu = LOONGARCH_CPU(thread_cpu);
1056 uint32_t hwcaps = 0;
1057
1058 hwcaps |= HWCAP_LOONGARCH_CRC32;
1059
1060 if (FIELD_EX32(cpu->env.cpucfg[1], CPUCFG1, UAL)) {
1061 hwcaps |= HWCAP_LOONGARCH_UAL;
1062 }
1063
1064 if (FIELD_EX32(cpu->env.cpucfg[2], CPUCFG2, FP)) {
1065 hwcaps |= HWCAP_LOONGARCH_FPU;
1066 }
1067
1068 if (FIELD_EX32(cpu->env.cpucfg[2], CPUCFG2, LAM)) {
1069 hwcaps |= HWCAP_LOONGARCH_LAM;
1070 }
1071
1072 return hwcaps;
1073 }
1074
1075 #define ELF_PLATFORM "loongarch"
1076
1077 #endif /* TARGET_LOONGARCH64 */
1078
1079 #ifdef TARGET_MIPS
1080
1081 #define ELF_START_MMAP 0x80000000
1082
1083 #ifdef TARGET_MIPS64
1084 #define ELF_CLASS ELFCLASS64
1085 #else
1086 #define ELF_CLASS ELFCLASS32
1087 #endif
1088 #define ELF_ARCH EM_MIPS
1089 #define EXSTACK_DEFAULT true
1090
1091 #ifdef TARGET_ABI_MIPSN32
1092 #define elf_check_abi(x) ((x) & EF_MIPS_ABI2)
1093 #else
1094 #define elf_check_abi(x) (!((x) & EF_MIPS_ABI2))
1095 #endif
1096
1097 #define ELF_BASE_PLATFORM get_elf_base_platform()
1098
1099 #define MATCH_PLATFORM_INSN(_flags, _base_platform) \
1100 do { if ((cpu->env.insn_flags & (_flags)) == _flags) \
1101 { return _base_platform; } } while (0)
1102
1103 static const char *get_elf_base_platform(void)
1104 {
1105 MIPSCPU *cpu = MIPS_CPU(thread_cpu);
1106
1107 /* 64 bit ISAs goes first */
1108 MATCH_PLATFORM_INSN(CPU_MIPS64R6, "mips64r6");
1109 MATCH_PLATFORM_INSN(CPU_MIPS64R5, "mips64r5");
1110 MATCH_PLATFORM_INSN(CPU_MIPS64R2, "mips64r2");
1111 MATCH_PLATFORM_INSN(CPU_MIPS64R1, "mips64");
1112 MATCH_PLATFORM_INSN(CPU_MIPS5, "mips5");
1113 MATCH_PLATFORM_INSN(CPU_MIPS4, "mips4");
1114 MATCH_PLATFORM_INSN(CPU_MIPS3, "mips3");
1115
1116 /* 32 bit ISAs */
1117 MATCH_PLATFORM_INSN(CPU_MIPS32R6, "mips32r6");
1118 MATCH_PLATFORM_INSN(CPU_MIPS32R5, "mips32r5");
1119 MATCH_PLATFORM_INSN(CPU_MIPS32R2, "mips32r2");
1120 MATCH_PLATFORM_INSN(CPU_MIPS32R1, "mips32");
1121 MATCH_PLATFORM_INSN(CPU_MIPS2, "mips2");
1122
1123 /* Fallback */
1124 return "mips";
1125 }
1126 #undef MATCH_PLATFORM_INSN
1127
1128 static inline void init_thread(struct target_pt_regs *regs,
1129 struct image_info *infop)
1130 {
1131 regs->cp0_status = 2 << CP0St_KSU;
1132 regs->cp0_epc = infop->entry;
1133 regs->regs[29] = infop->start_stack;
1134 }
1135
1136 /* See linux kernel: arch/mips/include/asm/elf.h. */
1137 #define ELF_NREG 45
1138 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1139
1140 /* See linux kernel: arch/mips/include/asm/reg.h. */
1141 enum {
1142 #ifdef TARGET_MIPS64
1143 TARGET_EF_R0 = 0,
1144 #else
1145 TARGET_EF_R0 = 6,
1146 #endif
1147 TARGET_EF_R26 = TARGET_EF_R0 + 26,
1148 TARGET_EF_R27 = TARGET_EF_R0 + 27,
1149 TARGET_EF_LO = TARGET_EF_R0 + 32,
1150 TARGET_EF_HI = TARGET_EF_R0 + 33,
1151 TARGET_EF_CP0_EPC = TARGET_EF_R0 + 34,
1152 TARGET_EF_CP0_BADVADDR = TARGET_EF_R0 + 35,
1153 TARGET_EF_CP0_STATUS = TARGET_EF_R0 + 36,
1154 TARGET_EF_CP0_CAUSE = TARGET_EF_R0 + 37
1155 };
1156
1157 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */
1158 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMIPSState *env)
1159 {
1160 int i;
1161
1162 for (i = 0; i < TARGET_EF_R0; i++) {
1163 (*regs)[i] = 0;
1164 }
1165 (*regs)[TARGET_EF_R0] = 0;
1166
1167 for (i = 1; i < ARRAY_SIZE(env->active_tc.gpr); i++) {
1168 (*regs)[TARGET_EF_R0 + i] = tswapreg(env->active_tc.gpr[i]);
1169 }
1170
1171 (*regs)[TARGET_EF_R26] = 0;
1172 (*regs)[TARGET_EF_R27] = 0;
1173 (*regs)[TARGET_EF_LO] = tswapreg(env->active_tc.LO[0]);
1174 (*regs)[TARGET_EF_HI] = tswapreg(env->active_tc.HI[0]);
1175 (*regs)[TARGET_EF_CP0_EPC] = tswapreg(env->active_tc.PC);
1176 (*regs)[TARGET_EF_CP0_BADVADDR] = tswapreg(env->CP0_BadVAddr);
1177 (*regs)[TARGET_EF_CP0_STATUS] = tswapreg(env->CP0_Status);
1178 (*regs)[TARGET_EF_CP0_CAUSE] = tswapreg(env->CP0_Cause);
1179 }
1180
1181 #define USE_ELF_CORE_DUMP
1182 #define ELF_EXEC_PAGESIZE 4096
1183
1184 /* See arch/mips/include/uapi/asm/hwcap.h. */
1185 enum {
1186 HWCAP_MIPS_R6 = (1 << 0),
1187 HWCAP_MIPS_MSA = (1 << 1),
1188 HWCAP_MIPS_CRC32 = (1 << 2),
1189 HWCAP_MIPS_MIPS16 = (1 << 3),
1190 HWCAP_MIPS_MDMX = (1 << 4),
1191 HWCAP_MIPS_MIPS3D = (1 << 5),
1192 HWCAP_MIPS_SMARTMIPS = (1 << 6),
1193 HWCAP_MIPS_DSP = (1 << 7),
1194 HWCAP_MIPS_DSP2 = (1 << 8),
1195 HWCAP_MIPS_DSP3 = (1 << 9),
1196 HWCAP_MIPS_MIPS16E2 = (1 << 10),
1197 HWCAP_LOONGSON_MMI = (1 << 11),
1198 HWCAP_LOONGSON_EXT = (1 << 12),
1199 HWCAP_LOONGSON_EXT2 = (1 << 13),
1200 HWCAP_LOONGSON_CPUCFG = (1 << 14),
1201 };
1202
1203 #define ELF_HWCAP get_elf_hwcap()
1204
1205 #define GET_FEATURE_INSN(_flag, _hwcap) \
1206 do { if (cpu->env.insn_flags & (_flag)) { hwcaps |= _hwcap; } } while (0)
1207
1208 #define GET_FEATURE_REG_SET(_reg, _mask, _hwcap) \
1209 do { if (cpu->env._reg & (_mask)) { hwcaps |= _hwcap; } } while (0)
1210
1211 #define GET_FEATURE_REG_EQU(_reg, _start, _length, _val, _hwcap) \
1212 do { \
1213 if (extract32(cpu->env._reg, (_start), (_length)) == (_val)) { \
1214 hwcaps |= _hwcap; \
1215 } \
1216 } while (0)
1217
1218 static uint32_t get_elf_hwcap(void)
1219 {
1220 MIPSCPU *cpu = MIPS_CPU(thread_cpu);
1221 uint32_t hwcaps = 0;
1222
1223 GET_FEATURE_REG_EQU(CP0_Config0, CP0C0_AR, CP0C0_AR_LENGTH,
1224 2, HWCAP_MIPS_R6);
1225 GET_FEATURE_REG_SET(CP0_Config3, 1 << CP0C3_MSAP, HWCAP_MIPS_MSA);
1226 GET_FEATURE_INSN(ASE_LMMI, HWCAP_LOONGSON_MMI);
1227 GET_FEATURE_INSN(ASE_LEXT, HWCAP_LOONGSON_EXT);
1228
1229 return hwcaps;
1230 }
1231
1232 #undef GET_FEATURE_REG_EQU
1233 #undef GET_FEATURE_REG_SET
1234 #undef GET_FEATURE_INSN
1235
1236 #endif /* TARGET_MIPS */
1237
1238 #ifdef TARGET_MICROBLAZE
1239
1240 #define ELF_START_MMAP 0x80000000
1241
1242 #define elf_check_arch(x) ( (x) == EM_MICROBLAZE || (x) == EM_MICROBLAZE_OLD)
1243
1244 #define ELF_CLASS ELFCLASS32
1245 #define ELF_ARCH EM_MICROBLAZE
1246
1247 static inline void init_thread(struct target_pt_regs *regs,
1248 struct image_info *infop)
1249 {
1250 regs->pc = infop->entry;
1251 regs->r1 = infop->start_stack;
1252
1253 }
1254
1255 #define ELF_EXEC_PAGESIZE 4096
1256
1257 #define USE_ELF_CORE_DUMP
1258 #define ELF_NREG 38
1259 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1260
1261 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */
1262 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMBState *env)
1263 {
1264 int i, pos = 0;
1265
1266 for (i = 0; i < 32; i++) {
1267 (*regs)[pos++] = tswapreg(env->regs[i]);
1268 }
1269
1270 (*regs)[pos++] = tswapreg(env->pc);
1271 (*regs)[pos++] = tswapreg(mb_cpu_read_msr(env));
1272 (*regs)[pos++] = 0;
1273 (*regs)[pos++] = tswapreg(env->ear);
1274 (*regs)[pos++] = 0;
1275 (*regs)[pos++] = tswapreg(env->esr);
1276 }
1277
1278 #endif /* TARGET_MICROBLAZE */
1279
1280 #ifdef TARGET_NIOS2
1281
1282 #define ELF_START_MMAP 0x80000000
1283
1284 #define elf_check_arch(x) ((x) == EM_ALTERA_NIOS2)
1285
1286 #define ELF_CLASS ELFCLASS32
1287 #define ELF_ARCH EM_ALTERA_NIOS2
1288
1289 static void init_thread(struct target_pt_regs *regs, struct image_info *infop)
1290 {
1291 regs->ea = infop->entry;
1292 regs->sp = infop->start_stack;
1293 }
1294
1295 #define LO_COMMPAGE TARGET_PAGE_SIZE
1296
1297 static bool init_guest_commpage(void)
1298 {
1299 static const uint8_t kuser_page[4 + 2 * 64] = {
1300 /* __kuser_helper_version */
1301 [0x00] = 0x02, 0x00, 0x00, 0x00,
1302
1303 /* __kuser_cmpxchg */
1304 [0x04] = 0x3a, 0x6c, 0x3b, 0x00, /* trap 16 */
1305 0x3a, 0x28, 0x00, 0xf8, /* ret */
1306
1307 /* __kuser_sigtramp */
1308 [0x44] = 0xc4, 0x22, 0x80, 0x00, /* movi r2, __NR_rt_sigreturn */
1309 0x3a, 0x68, 0x3b, 0x00, /* trap 0 */
1310 };
1311
1312 void *want = g2h_untagged(LO_COMMPAGE & -qemu_host_page_size);
1313 void *addr = mmap(want, qemu_host_page_size, PROT_READ | PROT_WRITE,
1314 MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0);
1315
1316 if (addr == MAP_FAILED) {
1317 perror("Allocating guest commpage");
1318 exit(EXIT_FAILURE);
1319 }
1320 if (addr != want) {
1321 return false;
1322 }
1323
1324 memcpy(addr, kuser_page, sizeof(kuser_page));
1325
1326 if (mprotect(addr, qemu_host_page_size, PROT_READ)) {
1327 perror("Protecting guest commpage");
1328 exit(EXIT_FAILURE);
1329 }
1330
1331 page_set_flags(LO_COMMPAGE, LO_COMMPAGE | ~TARGET_PAGE_MASK,
1332 PAGE_READ | PAGE_EXEC | PAGE_VALID);
1333 return true;
1334 }
1335
1336 #define ELF_EXEC_PAGESIZE 4096
1337
1338 #define USE_ELF_CORE_DUMP
1339 #define ELF_NREG 49
1340 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1341
1342 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */
1343 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1344 const CPUNios2State *env)
1345 {
1346 int i;
1347
1348 (*regs)[0] = -1;
1349 for (i = 1; i < 8; i++) /* r0-r7 */
1350 (*regs)[i] = tswapreg(env->regs[i + 7]);
1351
1352 for (i = 8; i < 16; i++) /* r8-r15 */
1353 (*regs)[i] = tswapreg(env->regs[i - 8]);
1354
1355 for (i = 16; i < 24; i++) /* r16-r23 */
1356 (*regs)[i] = tswapreg(env->regs[i + 7]);
1357 (*regs)[24] = -1; /* R_ET */
1358 (*regs)[25] = -1; /* R_BT */
1359 (*regs)[26] = tswapreg(env->regs[R_GP]);
1360 (*regs)[27] = tswapreg(env->regs[R_SP]);
1361 (*regs)[28] = tswapreg(env->regs[R_FP]);
1362 (*regs)[29] = tswapreg(env->regs[R_EA]);
1363 (*regs)[30] = -1; /* R_SSTATUS */
1364 (*regs)[31] = tswapreg(env->regs[R_RA]);
1365
1366 (*regs)[32] = tswapreg(env->pc);
1367
1368 (*regs)[33] = -1; /* R_STATUS */
1369 (*regs)[34] = tswapreg(env->regs[CR_ESTATUS]);
1370
1371 for (i = 35; i < 49; i++) /* ... */
1372 (*regs)[i] = -1;
1373 }
1374
1375 #endif /* TARGET_NIOS2 */
1376
1377 #ifdef TARGET_OPENRISC
1378
1379 #define ELF_START_MMAP 0x08000000
1380
1381 #define ELF_ARCH EM_OPENRISC
1382 #define ELF_CLASS ELFCLASS32
1383 #define ELF_DATA ELFDATA2MSB
1384
1385 static inline void init_thread(struct target_pt_regs *regs,
1386 struct image_info *infop)
1387 {
1388 regs->pc = infop->entry;
1389 regs->gpr[1] = infop->start_stack;
1390 }
1391
1392 #define USE_ELF_CORE_DUMP
1393 #define ELF_EXEC_PAGESIZE 8192
1394
1395 /* See linux kernel arch/openrisc/include/asm/elf.h. */
1396 #define ELF_NREG 34 /* gprs and pc, sr */
1397 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1398
1399 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1400 const CPUOpenRISCState *env)
1401 {
1402 int i;
1403
1404 for (i = 0; i < 32; i++) {
1405 (*regs)[i] = tswapreg(cpu_get_gpr(env, i));
1406 }
1407 (*regs)[32] = tswapreg(env->pc);
1408 (*regs)[33] = tswapreg(cpu_get_sr(env));
1409 }
1410 #define ELF_HWCAP 0
1411 #define ELF_PLATFORM NULL
1412
1413 #endif /* TARGET_OPENRISC */
1414
1415 #ifdef TARGET_SH4
1416
1417 #define ELF_START_MMAP 0x80000000
1418
1419 #define ELF_CLASS ELFCLASS32
1420 #define ELF_ARCH EM_SH
1421
1422 static inline void init_thread(struct target_pt_regs *regs,
1423 struct image_info *infop)
1424 {
1425 /* Check other registers XXXXX */
1426 regs->pc = infop->entry;
1427 regs->regs[15] = infop->start_stack;
1428 }
1429
1430 /* See linux kernel: arch/sh/include/asm/elf.h. */
1431 #define ELF_NREG 23
1432 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1433
1434 /* See linux kernel: arch/sh/include/asm/ptrace.h. */
1435 enum {
1436 TARGET_REG_PC = 16,
1437 TARGET_REG_PR = 17,
1438 TARGET_REG_SR = 18,
1439 TARGET_REG_GBR = 19,
1440 TARGET_REG_MACH = 20,
1441 TARGET_REG_MACL = 21,
1442 TARGET_REG_SYSCALL = 22
1443 };
1444
1445 static inline void elf_core_copy_regs(target_elf_gregset_t *regs,
1446 const CPUSH4State *env)
1447 {
1448 int i;
1449
1450 for (i = 0; i < 16; i++) {
1451 (*regs)[i] = tswapreg(env->gregs[i]);
1452 }
1453
1454 (*regs)[TARGET_REG_PC] = tswapreg(env->pc);
1455 (*regs)[TARGET_REG_PR] = tswapreg(env->pr);
1456 (*regs)[TARGET_REG_SR] = tswapreg(env->sr);
1457 (*regs)[TARGET_REG_GBR] = tswapreg(env->gbr);
1458 (*regs)[TARGET_REG_MACH] = tswapreg(env->mach);
1459 (*regs)[TARGET_REG_MACL] = tswapreg(env->macl);
1460 (*regs)[TARGET_REG_SYSCALL] = 0; /* FIXME */
1461 }
1462
1463 #define USE_ELF_CORE_DUMP
1464 #define ELF_EXEC_PAGESIZE 4096
1465
1466 enum {
1467 SH_CPU_HAS_FPU = 0x0001, /* Hardware FPU support */
1468 SH_CPU_HAS_P2_FLUSH_BUG = 0x0002, /* Need to flush the cache in P2 area */
1469 SH_CPU_HAS_MMU_PAGE_ASSOC = 0x0004, /* SH3: TLB way selection bit support */
1470 SH_CPU_HAS_DSP = 0x0008, /* SH-DSP: DSP support */
1471 SH_CPU_HAS_PERF_COUNTER = 0x0010, /* Hardware performance counters */
1472 SH_CPU_HAS_PTEA = 0x0020, /* PTEA register */
1473 SH_CPU_HAS_LLSC = 0x0040, /* movli.l/movco.l */
1474 SH_CPU_HAS_L2_CACHE = 0x0080, /* Secondary cache / URAM */
1475 SH_CPU_HAS_OP32 = 0x0100, /* 32-bit instruction support */
1476 SH_CPU_HAS_PTEAEX = 0x0200, /* PTE ASID Extension support */
1477 };
1478
1479 #define ELF_HWCAP get_elf_hwcap()
1480
1481 static uint32_t get_elf_hwcap(void)
1482 {
1483 SuperHCPU *cpu = SUPERH_CPU(thread_cpu);
1484 uint32_t hwcap = 0;
1485
1486 hwcap |= SH_CPU_HAS_FPU;
1487
1488 if (cpu->env.features & SH_FEATURE_SH4A) {
1489 hwcap |= SH_CPU_HAS_LLSC;
1490 }
1491
1492 return hwcap;
1493 }
1494
1495 #endif
1496
1497 #ifdef TARGET_CRIS
1498
1499 #define ELF_START_MMAP 0x80000000
1500
1501 #define ELF_CLASS ELFCLASS32
1502 #define ELF_ARCH EM_CRIS
1503
1504 static inline void init_thread(struct target_pt_regs *regs,
1505 struct image_info *infop)
1506 {
1507 regs->erp = infop->entry;
1508 }
1509
1510 #define ELF_EXEC_PAGESIZE 8192
1511
1512 #endif
1513
1514 #ifdef TARGET_M68K
1515
1516 #define ELF_START_MMAP 0x80000000
1517
1518 #define ELF_CLASS ELFCLASS32
1519 #define ELF_ARCH EM_68K
1520
1521 /* ??? Does this need to do anything?
1522 #define ELF_PLAT_INIT(_r) */
1523
1524 static inline void init_thread(struct target_pt_regs *regs,
1525 struct image_info *infop)
1526 {
1527 regs->usp = infop->start_stack;
1528 regs->sr = 0;
1529 regs->pc = infop->entry;
1530 }
1531
1532 /* See linux kernel: arch/m68k/include/asm/elf.h. */
1533 #define ELF_NREG 20
1534 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1535
1536 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUM68KState *env)
1537 {
1538 (*regs)[0] = tswapreg(env->dregs[1]);
1539 (*regs)[1] = tswapreg(env->dregs[2]);
1540 (*regs)[2] = tswapreg(env->dregs[3]);
1541 (*regs)[3] = tswapreg(env->dregs[4]);
1542 (*regs)[4] = tswapreg(env->dregs[5]);
1543 (*regs)[5] = tswapreg(env->dregs[6]);
1544 (*regs)[6] = tswapreg(env->dregs[7]);
1545 (*regs)[7] = tswapreg(env->aregs[0]);
1546 (*regs)[8] = tswapreg(env->aregs[1]);
1547 (*regs)[9] = tswapreg(env->aregs[2]);
1548 (*regs)[10] = tswapreg(env->aregs[3]);
1549 (*regs)[11] = tswapreg(env->aregs[4]);
1550 (*regs)[12] = tswapreg(env->aregs[5]);
1551 (*regs)[13] = tswapreg(env->aregs[6]);
1552 (*regs)[14] = tswapreg(env->dregs[0]);
1553 (*regs)[15] = tswapreg(env->aregs[7]);
1554 (*regs)[16] = tswapreg(env->dregs[0]); /* FIXME: orig_d0 */
1555 (*regs)[17] = tswapreg(env->sr);
1556 (*regs)[18] = tswapreg(env->pc);
1557 (*regs)[19] = 0; /* FIXME: regs->format | regs->vector */
1558 }
1559
1560 #define USE_ELF_CORE_DUMP
1561 #define ELF_EXEC_PAGESIZE 8192
1562
1563 #endif
1564
1565 #ifdef TARGET_ALPHA
1566
1567 #define ELF_START_MMAP (0x30000000000ULL)
1568
1569 #define ELF_CLASS ELFCLASS64
1570 #define ELF_ARCH EM_ALPHA
1571
1572 static inline void init_thread(struct target_pt_regs *regs,
1573 struct image_info *infop)
1574 {
1575 regs->pc = infop->entry;
1576 regs->ps = 8;
1577 regs->usp = infop->start_stack;
1578 }
1579
1580 #define ELF_EXEC_PAGESIZE 8192
1581
1582 #endif /* TARGET_ALPHA */
1583
1584 #ifdef TARGET_S390X
1585
1586 #define ELF_START_MMAP (0x20000000000ULL)
1587
1588 #define ELF_CLASS ELFCLASS64
1589 #define ELF_DATA ELFDATA2MSB
1590 #define ELF_ARCH EM_S390
1591
1592 #include "elf.h"
1593
1594 #define ELF_HWCAP get_elf_hwcap()
1595
1596 #define GET_FEATURE(_feat, _hwcap) \
1597 do { if (s390_has_feat(_feat)) { hwcap |= _hwcap; } } while (0)
1598
1599 uint32_t get_elf_hwcap(void)
1600 {
1601 /*
1602 * Let's assume we always have esan3 and zarch.
1603 * 31-bit processes can use 64-bit registers (high gprs).
1604 */
1605 uint32_t hwcap = HWCAP_S390_ESAN3 | HWCAP_S390_ZARCH | HWCAP_S390_HIGH_GPRS;
1606
1607 GET_FEATURE(S390_FEAT_STFLE, HWCAP_S390_STFLE);
1608 GET_FEATURE(S390_FEAT_MSA, HWCAP_S390_MSA);
1609 GET_FEATURE(S390_FEAT_LONG_DISPLACEMENT, HWCAP_S390_LDISP);
1610 GET_FEATURE(S390_FEAT_EXTENDED_IMMEDIATE, HWCAP_S390_EIMM);
1611 if (s390_has_feat(S390_FEAT_EXTENDED_TRANSLATION_3) &&
1612 s390_has_feat(S390_FEAT_ETF3_ENH)) {
1613 hwcap |= HWCAP_S390_ETF3EH;
1614 }
1615 GET_FEATURE(S390_FEAT_VECTOR, HWCAP_S390_VXRS);
1616 GET_FEATURE(S390_FEAT_VECTOR_ENH, HWCAP_S390_VXRS_EXT);
1617
1618 return hwcap;
1619 }
1620
1621 const char *elf_hwcap_str(uint32_t bit)
1622 {
1623 static const char *hwcap_str[] = {
1624 [HWCAP_S390_NR_ESAN3] = "esan3",
1625 [HWCAP_S390_NR_ZARCH] = "zarch",
1626 [HWCAP_S390_NR_STFLE] = "stfle",
1627 [HWCAP_S390_NR_MSA] = "msa",
1628 [HWCAP_S390_NR_LDISP] = "ldisp",
1629 [HWCAP_S390_NR_EIMM] = "eimm",
1630 [HWCAP_S390_NR_DFP] = "dfp",
1631 [HWCAP_S390_NR_HPAGE] = "edat",
1632 [HWCAP_S390_NR_ETF3EH] = "etf3eh",
1633 [HWCAP_S390_NR_HIGH_GPRS] = "highgprs",
1634 [HWCAP_S390_NR_TE] = "te",
1635 [HWCAP_S390_NR_VXRS] = "vx",
1636 [HWCAP_S390_NR_VXRS_BCD] = "vxd",
1637 [HWCAP_S390_NR_VXRS_EXT] = "vxe",
1638 [HWCAP_S390_NR_GS] = "gs",
1639 [HWCAP_S390_NR_VXRS_EXT2] = "vxe2",
1640 [HWCAP_S390_NR_VXRS_PDE] = "vxp",
1641 [HWCAP_S390_NR_SORT] = "sort",
1642 [HWCAP_S390_NR_DFLT] = "dflt",
1643 [HWCAP_S390_NR_NNPA] = "nnpa",
1644 [HWCAP_S390_NR_PCI_MIO] = "pcimio",
1645 [HWCAP_S390_NR_SIE] = "sie",
1646 };
1647
1648 return bit < ARRAY_SIZE(hwcap_str) ? hwcap_str[bit] : NULL;
1649 }
1650
1651 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop)
1652 {
1653 regs->psw.addr = infop->entry;
1654 regs->psw.mask = PSW_MASK_DAT | PSW_MASK_IO | PSW_MASK_EXT | \
1655 PSW_MASK_MCHECK | PSW_MASK_PSTATE | PSW_MASK_64 | \
1656 PSW_MASK_32;
1657 regs->gprs[15] = infop->start_stack;
1658 }
1659
1660 /* See linux kernel: arch/s390/include/uapi/asm/ptrace.h (s390_regs). */
1661 #define ELF_NREG 27
1662 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1663
1664 enum {
1665 TARGET_REG_PSWM = 0,
1666 TARGET_REG_PSWA = 1,
1667 TARGET_REG_GPRS = 2,
1668 TARGET_REG_ARS = 18,
1669 TARGET_REG_ORIG_R2 = 26,
1670 };
1671
1672 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1673 const CPUS390XState *env)
1674 {
1675 int i;
1676 uint32_t *aregs;
1677
1678 (*regs)[TARGET_REG_PSWM] = tswapreg(env->psw.mask);
1679 (*regs)[TARGET_REG_PSWA] = tswapreg(env->psw.addr);
1680 for (i = 0; i < 16; i++) {
1681 (*regs)[TARGET_REG_GPRS + i] = tswapreg(env->regs[i]);
1682 }
1683 aregs = (uint32_t *)&((*regs)[TARGET_REG_ARS]);
1684 for (i = 0; i < 16; i++) {
1685 aregs[i] = tswap32(env->aregs[i]);
1686 }
1687 (*regs)[TARGET_REG_ORIG_R2] = 0;
1688 }
1689
1690 #define USE_ELF_CORE_DUMP
1691 #define ELF_EXEC_PAGESIZE 4096
1692
1693 #endif /* TARGET_S390X */
1694
1695 #ifdef TARGET_RISCV
1696
1697 #define ELF_START_MMAP 0x80000000
1698 #define ELF_ARCH EM_RISCV
1699
1700 #ifdef TARGET_RISCV32
1701 #define ELF_CLASS ELFCLASS32
1702 #else
1703 #define ELF_CLASS ELFCLASS64
1704 #endif
1705
1706 #define ELF_HWCAP get_elf_hwcap()
1707
1708 static uint32_t get_elf_hwcap(void)
1709 {
1710 #define MISA_BIT(EXT) (1 << (EXT - 'A'))
1711 RISCVCPU *cpu = RISCV_CPU(thread_cpu);
1712 uint32_t mask = MISA_BIT('I') | MISA_BIT('M') | MISA_BIT('A')
1713 | MISA_BIT('F') | MISA_BIT('D') | MISA_BIT('C')
1714 | MISA_BIT('V');
1715
1716 return cpu->env.misa_ext & mask;
1717 #undef MISA_BIT
1718 }
1719
1720 static inline void init_thread(struct target_pt_regs *regs,
1721 struct image_info *infop)
1722 {
1723 regs->sepc = infop->entry;
1724 regs->sp = infop->start_stack;
1725 }
1726
1727 #define ELF_EXEC_PAGESIZE 4096
1728
1729 #endif /* TARGET_RISCV */
1730
1731 #ifdef TARGET_HPPA
1732
1733 #define ELF_START_MMAP 0x80000000
1734 #define ELF_CLASS ELFCLASS32
1735 #define ELF_ARCH EM_PARISC
1736 #define ELF_PLATFORM "PARISC"
1737 #define STACK_GROWS_DOWN 0
1738 #define STACK_ALIGNMENT 64
1739
1740 static inline void init_thread(struct target_pt_regs *regs,
1741 struct image_info *infop)
1742 {
1743 regs->iaoq[0] = infop->entry;
1744 regs->iaoq[1] = infop->entry + 4;
1745 regs->gr[23] = 0;
1746 regs->gr[24] = infop->argv;
1747 regs->gr[25] = infop->argc;
1748 /* The top-of-stack contains a linkage buffer. */
1749 regs->gr[30] = infop->start_stack + 64;
1750 regs->gr[31] = infop->entry;
1751 }
1752
1753 #define LO_COMMPAGE 0
1754
1755 static bool init_guest_commpage(void)
1756 {
1757 void *want = g2h_untagged(LO_COMMPAGE);
1758 void *addr = mmap(want, qemu_host_page_size, PROT_NONE,
1759 MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0);
1760
1761 if (addr == MAP_FAILED) {
1762 perror("Allocating guest commpage");
1763 exit(EXIT_FAILURE);
1764 }
1765 if (addr != want) {
1766 return false;
1767 }
1768
1769 /*
1770 * On Linux, page zero is normally marked execute only + gateway.
1771 * Normal read or write is supposed to fail (thus PROT_NONE above),
1772 * but specific offsets have kernel code mapped to raise permissions
1773 * and implement syscalls. Here, simply mark the page executable.
1774 * Special case the entry points during translation (see do_page_zero).
1775 */
1776 page_set_flags(LO_COMMPAGE, LO_COMMPAGE | ~TARGET_PAGE_MASK,
1777 PAGE_EXEC | PAGE_VALID);
1778 return true;
1779 }
1780
1781 #endif /* TARGET_HPPA */
1782
1783 #ifdef TARGET_XTENSA
1784
1785 #define ELF_START_MMAP 0x20000000
1786
1787 #define ELF_CLASS ELFCLASS32
1788 #define ELF_ARCH EM_XTENSA
1789
1790 static inline void init_thread(struct target_pt_regs *regs,
1791 struct image_info *infop)
1792 {
1793 regs->windowbase = 0;
1794 regs->windowstart = 1;
1795 regs->areg[1] = infop->start_stack;
1796 regs->pc = infop->entry;
1797 if (info_is_fdpic(infop)) {
1798 regs->areg[4] = infop->loadmap_addr;
1799 regs->areg[5] = infop->interpreter_loadmap_addr;
1800 if (infop->interpreter_loadmap_addr) {
1801 regs->areg[6] = infop->interpreter_pt_dynamic_addr;
1802 } else {
1803 regs->areg[6] = infop->pt_dynamic_addr;
1804 }
1805 }
1806 }
1807
1808 /* See linux kernel: arch/xtensa/include/asm/elf.h. */
1809 #define ELF_NREG 128
1810 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1811
1812 enum {
1813 TARGET_REG_PC,
1814 TARGET_REG_PS,
1815 TARGET_REG_LBEG,
1816 TARGET_REG_LEND,
1817 TARGET_REG_LCOUNT,
1818 TARGET_REG_SAR,
1819 TARGET_REG_WINDOWSTART,
1820 TARGET_REG_WINDOWBASE,
1821 TARGET_REG_THREADPTR,
1822 TARGET_REG_AR0 = 64,
1823 };
1824
1825 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1826 const CPUXtensaState *env)
1827 {
1828 unsigned i;
1829
1830 (*regs)[TARGET_REG_PC] = tswapreg(env->pc);
1831 (*regs)[TARGET_REG_PS] = tswapreg(env->sregs[PS] & ~PS_EXCM);
1832 (*regs)[TARGET_REG_LBEG] = tswapreg(env->sregs[LBEG]);
1833 (*regs)[TARGET_REG_LEND] = tswapreg(env->sregs[LEND]);
1834 (*regs)[TARGET_REG_LCOUNT] = tswapreg(env->sregs[LCOUNT]);
1835 (*regs)[TARGET_REG_SAR] = tswapreg(env->sregs[SAR]);
1836 (*regs)[TARGET_REG_WINDOWSTART] = tswapreg(env->sregs[WINDOW_START]);
1837 (*regs)[TARGET_REG_WINDOWBASE] = tswapreg(env->sregs[WINDOW_BASE]);
1838 (*regs)[TARGET_REG_THREADPTR] = tswapreg(env->uregs[THREADPTR]);
1839 xtensa_sync_phys_from_window((CPUXtensaState *)env);
1840 for (i = 0; i < env->config->nareg; ++i) {
1841 (*regs)[TARGET_REG_AR0 + i] = tswapreg(env->phys_regs[i]);
1842 }
1843 }
1844
1845 #define USE_ELF_CORE_DUMP
1846 #define ELF_EXEC_PAGESIZE 4096
1847
1848 #endif /* TARGET_XTENSA */
1849
1850 #ifdef TARGET_HEXAGON
1851
1852 #define ELF_START_MMAP 0x20000000
1853
1854 #define ELF_CLASS ELFCLASS32
1855 #define ELF_ARCH EM_HEXAGON
1856
1857 static inline void init_thread(struct target_pt_regs *regs,
1858 struct image_info *infop)
1859 {
1860 regs->sepc = infop->entry;
1861 regs->sp = infop->start_stack;
1862 }
1863
1864 #endif /* TARGET_HEXAGON */
1865
1866 #ifndef ELF_BASE_PLATFORM
1867 #define ELF_BASE_PLATFORM (NULL)
1868 #endif
1869
1870 #ifndef ELF_PLATFORM
1871 #define ELF_PLATFORM (NULL)
1872 #endif
1873
1874 #ifndef ELF_MACHINE
1875 #define ELF_MACHINE ELF_ARCH
1876 #endif
1877
1878 #ifndef elf_check_arch
1879 #define elf_check_arch(x) ((x) == ELF_ARCH)
1880 #endif
1881
1882 #ifndef elf_check_abi
1883 #define elf_check_abi(x) (1)
1884 #endif
1885
1886 #ifndef ELF_HWCAP
1887 #define ELF_HWCAP 0
1888 #endif
1889
1890 #ifndef STACK_GROWS_DOWN
1891 #define STACK_GROWS_DOWN 1
1892 #endif
1893
1894 #ifndef STACK_ALIGNMENT
1895 #define STACK_ALIGNMENT 16
1896 #endif
1897
1898 #ifdef TARGET_ABI32
1899 #undef ELF_CLASS
1900 #define ELF_CLASS ELFCLASS32
1901 #undef bswaptls
1902 #define bswaptls(ptr) bswap32s(ptr)
1903 #endif
1904
1905 #ifndef EXSTACK_DEFAULT
1906 #define EXSTACK_DEFAULT false
1907 #endif
1908
1909 #include "elf.h"
1910
1911 /* We must delay the following stanzas until after "elf.h". */
1912 #if defined(TARGET_AARCH64)
1913
1914 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz,
1915 const uint32_t *data,
1916 struct image_info *info,
1917 Error **errp)
1918 {
1919 if (pr_type == GNU_PROPERTY_AARCH64_FEATURE_1_AND) {
1920 if (pr_datasz != sizeof(uint32_t)) {
1921 error_setg(errp, "Ill-formed GNU_PROPERTY_AARCH64_FEATURE_1_AND");
1922 return false;
1923 }
1924 /* We will extract GNU_PROPERTY_AARCH64_FEATURE_1_BTI later. */
1925 info->note_flags = *data;
1926 }
1927 return true;
1928 }
1929 #define ARCH_USE_GNU_PROPERTY 1
1930
1931 #else
1932
1933 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz,
1934 const uint32_t *data,
1935 struct image_info *info,
1936 Error **errp)
1937 {
1938 g_assert_not_reached();
1939 }
1940 #define ARCH_USE_GNU_PROPERTY 0
1941
1942 #endif
1943
1944 struct exec
1945 {
1946 unsigned int a_info; /* Use macros N_MAGIC, etc for access */
1947 unsigned int a_text; /* length of text, in bytes */
1948 unsigned int a_data; /* length of data, in bytes */
1949 unsigned int a_bss; /* length of uninitialized data area, in bytes */
1950 unsigned int a_syms; /* length of symbol table data in file, in bytes */
1951 unsigned int a_entry; /* start address */
1952 unsigned int a_trsize; /* length of relocation info for text, in bytes */
1953 unsigned int a_drsize; /* length of relocation info for data, in bytes */
1954 };
1955
1956
1957 #define N_MAGIC(exec) ((exec).a_info & 0xffff)
1958 #define OMAGIC 0407
1959 #define NMAGIC 0410
1960 #define ZMAGIC 0413
1961 #define QMAGIC 0314
1962
1963 #define DLINFO_ITEMS 16
1964
1965 static inline void memcpy_fromfs(void * to, const void * from, unsigned long n)
1966 {
1967 memcpy(to, from, n);
1968 }
1969
1970 #ifdef BSWAP_NEEDED
1971 static void bswap_ehdr(struct elfhdr *ehdr)
1972 {
1973 bswap16s(&ehdr->e_type); /* Object file type */
1974 bswap16s(&ehdr->e_machine); /* Architecture */
1975 bswap32s(&ehdr->e_version); /* Object file version */
1976 bswaptls(&ehdr->e_entry); /* Entry point virtual address */
1977 bswaptls(&ehdr->e_phoff); /* Program header table file offset */
1978 bswaptls(&ehdr->e_shoff); /* Section header table file offset */
1979 bswap32s(&ehdr->e_flags); /* Processor-specific flags */
1980 bswap16s(&ehdr->e_ehsize); /* ELF header size in bytes */
1981 bswap16s(&ehdr->e_phentsize); /* Program header table entry size */
1982 bswap16s(&ehdr->e_phnum); /* Program header table entry count */
1983 bswap16s(&ehdr->e_shentsize); /* Section header table entry size */
1984 bswap16s(&ehdr->e_shnum); /* Section header table entry count */
1985 bswap16s(&ehdr->e_shstrndx); /* Section header string table index */
1986 }
1987
1988 static void bswap_phdr(struct elf_phdr *phdr, int phnum)
1989 {
1990 int i;
1991 for (i = 0; i < phnum; ++i, ++phdr) {
1992 bswap32s(&phdr->p_type); /* Segment type */
1993 bswap32s(&phdr->p_flags); /* Segment flags */
1994 bswaptls(&phdr->p_offset); /* Segment file offset */
1995 bswaptls(&phdr->p_vaddr); /* Segment virtual address */
1996 bswaptls(&phdr->p_paddr); /* Segment physical address */
1997 bswaptls(&phdr->p_filesz); /* Segment size in file */
1998 bswaptls(&phdr->p_memsz); /* Segment size in memory */
1999 bswaptls(&phdr->p_align); /* Segment alignment */
2000 }
2001 }
2002
2003 static void bswap_shdr(struct elf_shdr *shdr, int shnum)
2004 {
2005 int i;
2006 for (i = 0; i < shnum; ++i, ++shdr) {
2007 bswap32s(&shdr->sh_name);
2008 bswap32s(&shdr->sh_type);
2009 bswaptls(&shdr->sh_flags);
2010 bswaptls(&shdr->sh_addr);
2011 bswaptls(&shdr->sh_offset);
2012 bswaptls(&shdr->sh_size);
2013 bswap32s(&shdr->sh_link);
2014 bswap32s(&shdr->sh_info);
2015 bswaptls(&shdr->sh_addralign);
2016 bswaptls(&shdr->sh_entsize);
2017 }
2018 }
2019
2020 static void bswap_sym(struct elf_sym *sym)
2021 {
2022 bswap32s(&sym->st_name);
2023 bswaptls(&sym->st_value);
2024 bswaptls(&sym->st_size);
2025 bswap16s(&sym->st_shndx);
2026 }
2027
2028 #ifdef TARGET_MIPS
2029 static void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags)
2030 {
2031 bswap16s(&abiflags->version);
2032 bswap32s(&abiflags->ases);
2033 bswap32s(&abiflags->isa_ext);
2034 bswap32s(&abiflags->flags1);
2035 bswap32s(&abiflags->flags2);
2036 }
2037 #endif
2038 #else
2039 static inline void bswap_ehdr(struct elfhdr *ehdr) { }
2040 static inline void bswap_phdr(struct elf_phdr *phdr, int phnum) { }
2041 static inline void bswap_shdr(struct elf_shdr *shdr, int shnum) { }
2042 static inline void bswap_sym(struct elf_sym *sym) { }
2043 #ifdef TARGET_MIPS
2044 static inline void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags) { }
2045 #endif
2046 #endif
2047
2048 #ifdef USE_ELF_CORE_DUMP
2049 static int elf_core_dump(int, const CPUArchState *);
2050 #endif /* USE_ELF_CORE_DUMP */
2051 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias);
2052
2053 /* Verify the portions of EHDR within E_IDENT for the target.
2054 This can be performed before bswapping the entire header. */
2055 static bool elf_check_ident(struct elfhdr *ehdr)
2056 {
2057 return (ehdr->e_ident[EI_MAG0] == ELFMAG0
2058 && ehdr->e_ident[EI_MAG1] == ELFMAG1
2059 && ehdr->e_ident[EI_MAG2] == ELFMAG2
2060 && ehdr->e_ident[EI_MAG3] == ELFMAG3
2061 && ehdr->e_ident[EI_CLASS] == ELF_CLASS
2062 && ehdr->e_ident[EI_DATA] == ELF_DATA
2063 && ehdr->e_ident[EI_VERSION] == EV_CURRENT);
2064 }
2065
2066 /* Verify the portions of EHDR outside of E_IDENT for the target.
2067 This has to wait until after bswapping the header. */
2068 static bool elf_check_ehdr(struct elfhdr *ehdr)
2069 {
2070 return (elf_check_arch(ehdr->e_machine)
2071 && elf_check_abi(ehdr->e_flags)
2072 && ehdr->e_ehsize == sizeof(struct elfhdr)
2073 && ehdr->e_phentsize == sizeof(struct elf_phdr)
2074 && (ehdr->e_type == ET_EXEC || ehdr->e_type == ET_DYN));
2075 }
2076
2077 /*
2078 * 'copy_elf_strings()' copies argument/envelope strings from user
2079 * memory to free pages in kernel mem. These are in a format ready
2080 * to be put directly into the top of new user memory.
2081 *
2082 */
2083 static abi_ulong copy_elf_strings(int argc, char **argv, char *scratch,
2084 abi_ulong p, abi_ulong stack_limit)
2085 {
2086 char *tmp;
2087 int len, i;
2088 abi_ulong top = p;
2089
2090 if (!p) {
2091 return 0; /* bullet-proofing */
2092 }
2093
2094 if (STACK_GROWS_DOWN) {
2095 int offset = ((p - 1) % TARGET_PAGE_SIZE) + 1;
2096 for (i = argc - 1; i >= 0; --i) {
2097 tmp = argv[i];
2098 if (!tmp) {
2099 fprintf(stderr, "VFS: argc is wrong");
2100 exit(-1);
2101 }
2102 len = strlen(tmp) + 1;
2103 tmp += len;
2104
2105 if (len > (p - stack_limit)) {
2106 return 0;
2107 }
2108 while (len) {
2109 int bytes_to_copy = (len > offset) ? offset : len;
2110 tmp -= bytes_to_copy;
2111 p -= bytes_to_copy;
2112 offset -= bytes_to_copy;
2113 len -= bytes_to_copy;
2114
2115 memcpy_fromfs(scratch + offset, tmp, bytes_to_copy);
2116
2117 if (offset == 0) {
2118 memcpy_to_target(p, scratch, top - p);
2119 top = p;
2120 offset = TARGET_PAGE_SIZE;
2121 }
2122 }
2123 }
2124 if (p != top) {
2125 memcpy_to_target(p, scratch + offset, top - p);
2126 }
2127 } else {
2128 int remaining = TARGET_PAGE_SIZE - (p % TARGET_PAGE_SIZE);
2129 for (i = 0; i < argc; ++i) {
2130 tmp = argv[i];
2131 if (!tmp) {
2132 fprintf(stderr, "VFS: argc is wrong");
2133 exit(-1);
2134 }
2135 len = strlen(tmp) + 1;
2136 if (len > (stack_limit - p)) {
2137 return 0;
2138 }
2139 while (len) {
2140 int bytes_to_copy = (len > remaining) ? remaining : len;
2141
2142 memcpy_fromfs(scratch + (p - top), tmp, bytes_to_copy);
2143
2144 tmp += bytes_to_copy;
2145 remaining -= bytes_to_copy;
2146 p += bytes_to_copy;
2147 len -= bytes_to_copy;
2148
2149 if (remaining == 0) {
2150 memcpy_to_target(top, scratch, p - top);
2151 top = p;
2152 remaining = TARGET_PAGE_SIZE;
2153 }
2154 }
2155 }
2156 if (p != top) {
2157 memcpy_to_target(top, scratch, p - top);
2158 }
2159 }
2160
2161 return p;
2162 }
2163
2164 /* Older linux kernels provide up to MAX_ARG_PAGES (default: 32) of
2165 * argument/environment space. Newer kernels (>2.6.33) allow more,
2166 * dependent on stack size, but guarantee at least 32 pages for
2167 * backwards compatibility.
2168 */
2169 #define STACK_LOWER_LIMIT (32 * TARGET_PAGE_SIZE)
2170
2171 static abi_ulong setup_arg_pages(struct linux_binprm *bprm,
2172 struct image_info *info)
2173 {
2174 abi_ulong size, error, guard;
2175 int prot;
2176
2177 size = guest_stack_size;
2178 if (size < STACK_LOWER_LIMIT) {
2179 size = STACK_LOWER_LIMIT;
2180 }
2181
2182 if (STACK_GROWS_DOWN) {
2183 guard = TARGET_PAGE_SIZE;
2184 if (guard < qemu_real_host_page_size()) {
2185 guard = qemu_real_host_page_size();
2186 }
2187 } else {
2188 /* no guard page for hppa target where stack grows upwards. */
2189 guard = 0;
2190 }
2191
2192 prot = PROT_READ | PROT_WRITE;
2193 if (info->exec_stack) {
2194 prot |= PROT_EXEC;
2195 }
2196 error = target_mmap(0, size + guard, prot,
2197 MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
2198 if (error == -1) {
2199 perror("mmap stack");
2200 exit(-1);
2201 }
2202
2203 /* We reserve one extra page at the top of the stack as guard. */
2204 if (STACK_GROWS_DOWN) {
2205 target_mprotect(error, guard, PROT_NONE);
2206 info->stack_limit = error + guard;
2207 return info->stack_limit + size - sizeof(void *);
2208 } else {
2209 info->stack_limit = error + size;
2210 return error;
2211 }
2212 }
2213
2214 /**
2215 * zero_bss:
2216 *
2217 * Map and zero the bss. We need to explicitly zero any fractional pages
2218 * after the data section (i.e. bss). Return false on mapping failure.
2219 */
2220 static bool zero_bss(abi_ulong start_bss, abi_ulong end_bss, int prot)
2221 {
2222 abi_ulong align_bss;
2223
2224 align_bss = TARGET_PAGE_ALIGN(start_bss);
2225 end_bss = TARGET_PAGE_ALIGN(end_bss);
2226
2227 if (start_bss < align_bss) {
2228 int flags = page_get_flags(start_bss);
2229
2230 if (!(flags & PAGE_VALID)) {
2231 /* Map the start of the bss. */
2232 align_bss -= TARGET_PAGE_SIZE;
2233 } else if (flags & PAGE_WRITE) {
2234 /* The page is already mapped writable. */
2235 memset(g2h_untagged(start_bss), 0, align_bss - start_bss);
2236 } else {
2237 /* Read-only zeros? */
2238 g_assert_not_reached();
2239 }
2240 }
2241
2242 return align_bss >= end_bss ||
2243 target_mmap(align_bss, end_bss - align_bss, prot,
2244 MAP_FIXED | MAP_PRIVATE | MAP_ANON, -1, 0) != -1;
2245 }
2246
2247 #if defined(TARGET_ARM)
2248 static int elf_is_fdpic(struct elfhdr *exec)
2249 {
2250 return exec->e_ident[EI_OSABI] == ELFOSABI_ARM_FDPIC;
2251 }
2252 #elif defined(TARGET_XTENSA)
2253 static int elf_is_fdpic(struct elfhdr *exec)
2254 {
2255 return exec->e_ident[EI_OSABI] == ELFOSABI_XTENSA_FDPIC;
2256 }
2257 #else
2258 /* Default implementation, always false. */
2259 static int elf_is_fdpic(struct elfhdr *exec)
2260 {
2261 return 0;
2262 }
2263 #endif
2264
2265 static abi_ulong loader_build_fdpic_loadmap(struct image_info *info, abi_ulong sp)
2266 {
2267 uint16_t n;
2268 struct elf32_fdpic_loadseg *loadsegs = info->loadsegs;
2269
2270 /* elf32_fdpic_loadseg */
2271 n = info->nsegs;
2272 while (n--) {
2273 sp -= 12;
2274 put_user_u32(loadsegs[n].addr, sp+0);
2275 put_user_u32(loadsegs[n].p_vaddr, sp+4);
2276 put_user_u32(loadsegs[n].p_memsz, sp+8);
2277 }
2278
2279 /* elf32_fdpic_loadmap */
2280 sp -= 4;
2281 put_user_u16(0, sp+0); /* version */
2282 put_user_u16(info->nsegs, sp+2); /* nsegs */
2283
2284 info->personality = PER_LINUX_FDPIC;
2285 info->loadmap_addr = sp;
2286
2287 return sp;
2288 }
2289
2290 static abi_ulong create_elf_tables(abi_ulong p, int argc, int envc,
2291 struct elfhdr *exec,
2292 struct image_info *info,
2293 struct image_info *interp_info)
2294 {
2295 abi_ulong sp;
2296 abi_ulong u_argc, u_argv, u_envp, u_auxv;
2297 int size;
2298 int i;
2299 abi_ulong u_rand_bytes;
2300 uint8_t k_rand_bytes[16];
2301 abi_ulong u_platform, u_base_platform;
2302 const char *k_platform, *k_base_platform;
2303 const int n = sizeof(elf_addr_t);
2304
2305 sp = p;
2306
2307 /* Needs to be before we load the env/argc/... */
2308 if (elf_is_fdpic(exec)) {
2309 /* Need 4 byte alignment for these structs */
2310 sp &= ~3;
2311 sp = loader_build_fdpic_loadmap(info, sp);
2312 info->other_info = interp_info;
2313 if (interp_info) {
2314 interp_info->other_info = info;
2315 sp = loader_build_fdpic_loadmap(interp_info, sp);
2316 info->interpreter_loadmap_addr = interp_info->loadmap_addr;
2317 info->interpreter_pt_dynamic_addr = interp_info->pt_dynamic_addr;
2318 } else {
2319 info->interpreter_loadmap_addr = 0;
2320 info->interpreter_pt_dynamic_addr = 0;
2321 }
2322 }
2323
2324 u_base_platform = 0;
2325 k_base_platform = ELF_BASE_PLATFORM;
2326 if (k_base_platform) {
2327 size_t len = strlen(k_base_platform) + 1;
2328 if (STACK_GROWS_DOWN) {
2329 sp -= (len + n - 1) & ~(n - 1);
2330 u_base_platform = sp;
2331 /* FIXME - check return value of memcpy_to_target() for failure */
2332 memcpy_to_target(sp, k_base_platform, len);
2333 } else {
2334 memcpy_to_target(sp, k_base_platform, len);
2335 u_base_platform = sp;
2336 sp += len + 1;
2337 }
2338 }
2339
2340 u_platform = 0;
2341 k_platform = ELF_PLATFORM;
2342 if (k_platform) {
2343 size_t len = strlen(k_platform) + 1;
2344 if (STACK_GROWS_DOWN) {
2345 sp -= (len + n - 1) & ~(n - 1);
2346 u_platform = sp;
2347 /* FIXME - check return value of memcpy_to_target() for failure */
2348 memcpy_to_target(sp, k_platform, len);
2349 } else {
2350 memcpy_to_target(sp, k_platform, len);
2351 u_platform = sp;
2352 sp += len + 1;
2353 }
2354 }
2355
2356 /* Provide 16 byte alignment for the PRNG, and basic alignment for
2357 * the argv and envp pointers.
2358 */
2359 if (STACK_GROWS_DOWN) {
2360 sp = QEMU_ALIGN_DOWN(sp, 16);
2361 } else {
2362 sp = QEMU_ALIGN_UP(sp, 16);
2363 }
2364
2365 /*
2366 * Generate 16 random bytes for userspace PRNG seeding.
2367 */
2368 qemu_guest_getrandom_nofail(k_rand_bytes, sizeof(k_rand_bytes));
2369 if (STACK_GROWS_DOWN) {
2370 sp -= 16;
2371 u_rand_bytes = sp;
2372 /* FIXME - check return value of memcpy_to_target() for failure */
2373 memcpy_to_target(sp, k_rand_bytes, 16);
2374 } else {
2375 memcpy_to_target(sp, k_rand_bytes, 16);
2376 u_rand_bytes = sp;
2377 sp += 16;
2378 }
2379
2380 size = (DLINFO_ITEMS + 1) * 2;
2381 if (k_base_platform)
2382 size += 2;
2383 if (k_platform)
2384 size += 2;
2385 #ifdef DLINFO_ARCH_ITEMS
2386 size += DLINFO_ARCH_ITEMS * 2;
2387 #endif
2388 #ifdef ELF_HWCAP2
2389 size += 2;
2390 #endif
2391 info->auxv_len = size * n;
2392
2393 size += envc + argc + 2;
2394 size += 1; /* argc itself */
2395 size *= n;
2396
2397 /* Allocate space and finalize stack alignment for entry now. */
2398 if (STACK_GROWS_DOWN) {
2399 u_argc = QEMU_ALIGN_DOWN(sp - size, STACK_ALIGNMENT);
2400 sp = u_argc;
2401 } else {
2402 u_argc = sp;
2403 sp = QEMU_ALIGN_UP(sp + size, STACK_ALIGNMENT);
2404 }
2405
2406 u_argv = u_argc + n;
2407 u_envp = u_argv + (argc + 1) * n;
2408 u_auxv = u_envp + (envc + 1) * n;
2409 info->saved_auxv = u_auxv;
2410 info->argc = argc;
2411 info->envc = envc;
2412 info->argv = u_argv;
2413 info->envp = u_envp;
2414
2415 /* This is correct because Linux defines
2416 * elf_addr_t as Elf32_Off / Elf64_Off
2417 */
2418 #define NEW_AUX_ENT(id, val) do { \
2419 put_user_ual(id, u_auxv); u_auxv += n; \
2420 put_user_ual(val, u_auxv); u_auxv += n; \
2421 } while(0)
2422
2423 #ifdef ARCH_DLINFO
2424 /*
2425 * ARCH_DLINFO must come first so platform specific code can enforce
2426 * special alignment requirements on the AUXV if necessary (eg. PPC).
2427 */
2428 ARCH_DLINFO;
2429 #endif
2430 /* There must be exactly DLINFO_ITEMS entries here, or the assert
2431 * on info->auxv_len will trigger.
2432 */
2433 NEW_AUX_ENT(AT_PHDR, (abi_ulong)(info->load_addr + exec->e_phoff));
2434 NEW_AUX_ENT(AT_PHENT, (abi_ulong)(sizeof (struct elf_phdr)));
2435 NEW_AUX_ENT(AT_PHNUM, (abi_ulong)(exec->e_phnum));
2436 if ((info->alignment & ~qemu_host_page_mask) != 0) {
2437 /* Target doesn't support host page size alignment */
2438 NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(TARGET_PAGE_SIZE));
2439 } else {
2440 NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(MAX(TARGET_PAGE_SIZE,
2441 qemu_host_page_size)));
2442 }
2443 NEW_AUX_ENT(AT_BASE, (abi_ulong)(interp_info ? interp_info->load_addr : 0));
2444 NEW_AUX_ENT(AT_FLAGS, (abi_ulong)0);
2445 NEW_AUX_ENT(AT_ENTRY, info->entry);
2446 NEW_AUX_ENT(AT_UID, (abi_ulong) getuid());
2447 NEW_AUX_ENT(AT_EUID, (abi_ulong) geteuid());
2448 NEW_AUX_ENT(AT_GID, (abi_ulong) getgid());
2449 NEW_AUX_ENT(AT_EGID, (abi_ulong) getegid());
2450 NEW_AUX_ENT(AT_HWCAP, (abi_ulong) ELF_HWCAP);
2451 NEW_AUX_ENT(AT_CLKTCK, (abi_ulong) sysconf(_SC_CLK_TCK));
2452 NEW_AUX_ENT(AT_RANDOM, (abi_ulong) u_rand_bytes);
2453 NEW_AUX_ENT(AT_SECURE, (abi_ulong) qemu_getauxval(AT_SECURE));
2454 NEW_AUX_ENT(AT_EXECFN, info->file_string);
2455
2456 #ifdef ELF_HWCAP2
2457 NEW_AUX_ENT(AT_HWCAP2, (abi_ulong) ELF_HWCAP2);
2458 #endif
2459
2460 if (u_base_platform) {
2461 NEW_AUX_ENT(AT_BASE_PLATFORM, u_base_platform);
2462 }
2463 if (u_platform) {
2464 NEW_AUX_ENT(AT_PLATFORM, u_platform);
2465 }
2466 NEW_AUX_ENT (AT_NULL, 0);
2467 #undef NEW_AUX_ENT
2468
2469 /* Check that our initial calculation of the auxv length matches how much
2470 * we actually put into it.
2471 */
2472 assert(info->auxv_len == u_auxv - info->saved_auxv);
2473
2474 put_user_ual(argc, u_argc);
2475
2476 p = info->arg_strings;
2477 for (i = 0; i < argc; ++i) {
2478 put_user_ual(p, u_argv);
2479 u_argv += n;
2480 p += target_strlen(p) + 1;
2481 }
2482 put_user_ual(0, u_argv);
2483
2484 p = info->env_strings;
2485 for (i = 0; i < envc; ++i) {
2486 put_user_ual(p, u_envp);
2487 u_envp += n;
2488 p += target_strlen(p) + 1;
2489 }
2490 put_user_ual(0, u_envp);
2491
2492 return sp;
2493 }
2494
2495 #if defined(HI_COMMPAGE)
2496 #define LO_COMMPAGE -1
2497 #elif defined(LO_COMMPAGE)
2498 #define HI_COMMPAGE 0
2499 #else
2500 #define HI_COMMPAGE 0
2501 #define LO_COMMPAGE -1
2502 #ifndef INIT_GUEST_COMMPAGE
2503 #define init_guest_commpage() true
2504 #endif
2505 #endif
2506
2507 /**
2508 * pgb_try_mmap:
2509 * @addr: host start address
2510 * @addr_last: host last address
2511 * @keep: do not unmap the probe region
2512 *
2513 * Return 1 if [@addr, @addr_last] is not mapped in the host,
2514 * return 0 if it is not available to map, and -1 on mmap error.
2515 * If @keep, the region is left mapped on success, otherwise unmapped.
2516 */
2517 static int pgb_try_mmap(uintptr_t addr, uintptr_t addr_last, bool keep)
2518 {
2519 size_t size = addr_last - addr + 1;
2520 void *p = mmap((void *)addr, size, PROT_NONE,
2521 MAP_ANONYMOUS | MAP_PRIVATE |
2522 MAP_NORESERVE | MAP_FIXED_NOREPLACE, -1, 0);
2523 int ret;
2524
2525 if (p == MAP_FAILED) {
2526 return errno == EEXIST ? 0 : -1;
2527 }
2528 ret = p == (void *)addr;
2529 if (!keep || !ret) {
2530 munmap(p, size);
2531 }
2532 return ret;
2533 }
2534
2535 /**
2536 * pgb_try_mmap_skip_brk(uintptr_t addr, uintptr_t size, uintptr_t brk)
2537 * @addr: host address
2538 * @addr_last: host last address
2539 * @brk: host brk
2540 *
2541 * Like pgb_try_mmap, but additionally reserve some memory following brk.
2542 */
2543 static int pgb_try_mmap_skip_brk(uintptr_t addr, uintptr_t addr_last,
2544 uintptr_t brk, bool keep)
2545 {
2546 uintptr_t brk_last = brk + 16 * MiB - 1;
2547
2548 /* Do not map anything close to the host brk. */
2549 if (addr <= brk_last && brk <= addr_last) {
2550 return 0;
2551 }
2552 return pgb_try_mmap(addr, addr_last, keep);
2553 }
2554
2555 /**
2556 * pgb_try_mmap_set:
2557 * @ga: set of guest addrs
2558 * @base: guest_base
2559 * @brk: host brk
2560 *
2561 * Return true if all @ga can be mapped by the host at @base.
2562 * On success, retain the mapping at index 0 for reserved_va.
2563 */
2564
2565 typedef struct PGBAddrs {
2566 uintptr_t bounds[3][2]; /* start/last pairs */
2567 int nbounds;
2568 } PGBAddrs;
2569
2570 static bool pgb_try_mmap_set(const PGBAddrs *ga, uintptr_t base, uintptr_t brk)
2571 {
2572 for (int i = ga->nbounds - 1; i >= 0; --i) {
2573 if (pgb_try_mmap_skip_brk(ga->bounds[i][0] + base,
2574 ga->bounds[i][1] + base,
2575 brk, i == 0 && reserved_va) <= 0) {
2576 return false;
2577 }
2578 }
2579 return true;
2580 }
2581
2582 /**
2583 * pgb_addr_set:
2584 * @ga: output set of guest addrs
2585 * @guest_loaddr: guest image low address
2586 * @guest_loaddr: guest image high address
2587 * @identity: create for identity mapping
2588 *
2589 * Fill in @ga with the image, COMMPAGE and NULL page.
2590 */
2591 static bool pgb_addr_set(PGBAddrs *ga, abi_ulong guest_loaddr,
2592 abi_ulong guest_hiaddr, bool try_identity)
2593 {
2594 int n;
2595
2596 /*
2597 * With a low commpage, or a guest mapped very low,
2598 * we may not be able to use the identity map.
2599 */
2600 if (try_identity) {
2601 if (LO_COMMPAGE != -1 && LO_COMMPAGE < mmap_min_addr) {
2602 return false;
2603 }
2604 if (guest_loaddr != 0 && guest_loaddr < mmap_min_addr) {
2605 return false;
2606 }
2607 }
2608
2609 memset(ga, 0, sizeof(*ga));
2610 n = 0;
2611
2612 if (reserved_va) {
2613 ga->bounds[n][0] = try_identity ? mmap_min_addr : 0;
2614 ga->bounds[n][1] = reserved_va;
2615 n++;
2616 /* LO_COMMPAGE and NULL handled by reserving from 0. */
2617 } else {
2618 /* Add any LO_COMMPAGE or NULL page. */
2619 if (LO_COMMPAGE != -1) {
2620 ga->bounds[n][0] = 0;
2621 ga->bounds[n][1] = LO_COMMPAGE + TARGET_PAGE_SIZE - 1;
2622 n++;
2623 } else if (!try_identity) {
2624 ga->bounds[n][0] = 0;
2625 ga->bounds[n][1] = TARGET_PAGE_SIZE - 1;
2626 n++;
2627 }
2628
2629 /* Add the guest image for ET_EXEC. */
2630 if (guest_loaddr) {
2631 ga->bounds[n][0] = guest_loaddr;
2632 ga->bounds[n][1] = guest_hiaddr;
2633 n++;
2634 }
2635 }
2636
2637 /*
2638 * Temporarily disable
2639 * "comparison is always false due to limited range of data type"
2640 * due to comparison between unsigned and (possible) 0.
2641 */
2642 #pragma GCC diagnostic push
2643 #pragma GCC diagnostic ignored "-Wtype-limits"
2644
2645 /* Add any HI_COMMPAGE not covered by reserved_va. */
2646 if (reserved_va < HI_COMMPAGE) {
2647 ga->bounds[n][0] = HI_COMMPAGE & qemu_host_page_mask;
2648 ga->bounds[n][1] = HI_COMMPAGE + TARGET_PAGE_SIZE - 1;
2649 n++;
2650 }
2651
2652 #pragma GCC diagnostic pop
2653
2654 ga->nbounds = n;
2655 return true;
2656 }
2657
2658 static void pgb_fail_in_use(const char *image_name)
2659 {
2660 error_report("%s: requires virtual address space that is in use "
2661 "(omit the -B option or choose a different value)",
2662 image_name);
2663 exit(EXIT_FAILURE);
2664 }
2665
2666 static void pgb_fixed(const char *image_name, uintptr_t guest_loaddr,
2667 uintptr_t guest_hiaddr, uintptr_t align)
2668 {
2669 PGBAddrs ga;
2670 uintptr_t brk = (uintptr_t)sbrk(0);
2671
2672 if (!QEMU_IS_ALIGNED(guest_base, align)) {
2673 fprintf(stderr, "Requested guest base %p does not satisfy "
2674 "host minimum alignment (0x%" PRIxPTR ")\n",
2675 (void *)guest_base, align);
2676 exit(EXIT_FAILURE);
2677 }
2678
2679 if (!pgb_addr_set(&ga, guest_loaddr, guest_hiaddr, !guest_base)
2680 || !pgb_try_mmap_set(&ga, guest_base, brk)) {
2681 pgb_fail_in_use(image_name);
2682 }
2683 }
2684
2685 /**
2686 * pgb_find_fallback:
2687 *
2688 * This is a fallback method for finding holes in the host address space
2689 * if we don't have the benefit of being able to access /proc/self/map.
2690 * It can potentially take a very long time as we can only dumbly iterate
2691 * up the host address space seeing if the allocation would work.
2692 */
2693 static uintptr_t pgb_find_fallback(const PGBAddrs *ga, uintptr_t align,
2694 uintptr_t brk)
2695 {
2696 /* TODO: come up with a better estimate of how much to skip. */
2697 uintptr_t skip = sizeof(uintptr_t) == 4 ? MiB : GiB;
2698
2699 for (uintptr_t base = skip; ; base += skip) {
2700 base = ROUND_UP(base, align);
2701 if (pgb_try_mmap_set(ga, base, brk)) {
2702 return base;
2703 }
2704 if (base >= -skip) {
2705 return -1;
2706 }
2707 }
2708 }
2709
2710 static uintptr_t pgb_try_itree(const PGBAddrs *ga, uintptr_t base,
2711 IntervalTreeRoot *root)
2712 {
2713 for (int i = ga->nbounds - 1; i >= 0; --i) {
2714 uintptr_t s = base + ga->bounds[i][0];
2715 uintptr_t l = base + ga->bounds[i][1];
2716 IntervalTreeNode *n;
2717
2718 if (l < s) {
2719 /* Wraparound. Skip to advance S to mmap_min_addr. */
2720 return mmap_min_addr - s;
2721 }
2722
2723 n = interval_tree_iter_first(root, s, l);
2724 if (n != NULL) {
2725 /* Conflict. Skip to advance S to LAST + 1. */
2726 return n->last - s + 1;
2727 }
2728 }
2729 return 0; /* success */
2730 }
2731
2732 static uintptr_t pgb_find_itree(const PGBAddrs *ga, IntervalTreeRoot *root,
2733 uintptr_t align, uintptr_t brk)
2734 {
2735 uintptr_t last = mmap_min_addr;
2736 uintptr_t base, skip;
2737
2738 while (true) {
2739 base = ROUND_UP(last, align);
2740 if (base < last) {
2741 return -1;
2742 }
2743
2744 skip = pgb_try_itree(ga, base, root);
2745 if (skip == 0) {
2746 break;
2747 }
2748
2749 last = base + skip;
2750 if (last < base) {
2751 return -1;
2752 }
2753 }
2754
2755 /*
2756 * We've chosen 'base' based on holes in the interval tree,
2757 * but we don't yet know if it is a valid host address.
2758 * Because it is the first matching hole, if the host addresses
2759 * are invalid we know there are no further matches.
2760 */
2761 return pgb_try_mmap_set(ga, base, brk) ? base : -1;
2762 }
2763
2764 static void pgb_dynamic(const char *image_name, uintptr_t guest_loaddr,
2765 uintptr_t guest_hiaddr, uintptr_t align)
2766 {
2767 IntervalTreeRoot *root;
2768 uintptr_t brk, ret;
2769 PGBAddrs ga;
2770
2771 assert(QEMU_IS_ALIGNED(guest_loaddr, align));
2772
2773 /* Try the identity map first. */
2774 if (pgb_addr_set(&ga, guest_loaddr, guest_hiaddr, true)) {
2775 brk = (uintptr_t)sbrk(0);
2776 if (pgb_try_mmap_set(&ga, 0, brk)) {
2777 guest_base = 0;
2778 return;
2779 }
2780 }
2781
2782 /*
2783 * Rebuild the address set for non-identity map.
2784 * This differs in the mapping of the guest NULL page.
2785 */
2786 pgb_addr_set(&ga, guest_loaddr, guest_hiaddr, false);
2787
2788 root = read_self_maps();
2789
2790 /* Read brk after we've read the maps, which will malloc. */
2791 brk = (uintptr_t)sbrk(0);
2792
2793 if (!root) {
2794 ret = pgb_find_fallback(&ga, align, brk);
2795 } else {
2796 /*
2797 * Reserve the area close to the host brk.
2798 * This will be freed with the rest of the tree.
2799 */
2800 IntervalTreeNode *b = g_new0(IntervalTreeNode, 1);
2801 b->start = brk;
2802 b->last = brk + 16 * MiB - 1;
2803 interval_tree_insert(b, root);
2804
2805 ret = pgb_find_itree(&ga, root, align, brk);
2806 free_self_maps(root);
2807 }
2808
2809 if (ret == -1) {
2810 int w = TARGET_LONG_BITS / 4;
2811
2812 error_report("%s: Unable to find a guest_base to satisfy all "
2813 "guest address mapping requirements", image_name);
2814
2815 for (int i = 0; i < ga.nbounds; ++i) {
2816 error_printf(" %0*" PRIx64 "-%0*" PRIx64 "\n",
2817 w, (uint64_t)ga.bounds[i][0],
2818 w, (uint64_t)ga.bounds[i][1]);
2819 }
2820 exit(EXIT_FAILURE);
2821 }
2822 guest_base = ret;
2823 }
2824
2825 void probe_guest_base(const char *image_name, abi_ulong guest_loaddr,
2826 abi_ulong guest_hiaddr)
2827 {
2828 /* In order to use host shmat, we must be able to honor SHMLBA. */
2829 uintptr_t align = MAX(SHMLBA, qemu_host_page_size);
2830
2831 /* Sanity check the guest binary. */
2832 if (reserved_va) {
2833 if (guest_hiaddr > reserved_va) {
2834 error_report("%s: requires more than reserved virtual "
2835 "address space (0x%" PRIx64 " > 0x%lx)",
2836 image_name, (uint64_t)guest_hiaddr, reserved_va);
2837 exit(EXIT_FAILURE);
2838 }
2839 } else {
2840 if (guest_hiaddr != (uintptr_t)guest_hiaddr) {
2841 error_report("%s: requires more virtual address space "
2842 "than the host can provide (0x%" PRIx64 ")",
2843 image_name, (uint64_t)guest_hiaddr + 1);
2844 exit(EXIT_FAILURE);
2845 }
2846 }
2847
2848 if (have_guest_base) {
2849 pgb_fixed(image_name, guest_loaddr, guest_hiaddr, align);
2850 } else {
2851 pgb_dynamic(image_name, guest_loaddr, guest_hiaddr, align);
2852 }
2853
2854 /* Reserve and initialize the commpage. */
2855 if (!init_guest_commpage()) {
2856 /* We have already probed for the commpage being free. */
2857 g_assert_not_reached();
2858 }
2859
2860 assert(QEMU_IS_ALIGNED(guest_base, align));
2861 qemu_log_mask(CPU_LOG_PAGE, "Locating guest address space "
2862 "@ 0x%" PRIx64 "\n", (uint64_t)guest_base);
2863 }
2864
2865 enum {
2866 /* The string "GNU\0" as a magic number. */
2867 GNU0_MAGIC = const_le32('G' | 'N' << 8 | 'U' << 16),
2868 NOTE_DATA_SZ = 1 * KiB,
2869 NOTE_NAME_SZ = 4,
2870 ELF_GNU_PROPERTY_ALIGN = ELF_CLASS == ELFCLASS32 ? 4 : 8,
2871 };
2872
2873 /*
2874 * Process a single gnu_property entry.
2875 * Return false for error.
2876 */
2877 static bool parse_elf_property(const uint32_t *data, int *off, int datasz,
2878 struct image_info *info, bool have_prev_type,
2879 uint32_t *prev_type, Error **errp)
2880 {
2881 uint32_t pr_type, pr_datasz, step;
2882
2883 if (*off > datasz || !QEMU_IS_ALIGNED(*off, ELF_GNU_PROPERTY_ALIGN)) {
2884 goto error_data;
2885 }
2886 datasz -= *off;
2887 data += *off / sizeof(uint32_t);
2888
2889 if (datasz < 2 * sizeof(uint32_t)) {
2890 goto error_data;
2891 }
2892 pr_type = data[0];
2893 pr_datasz = data[1];
2894 data += 2;
2895 datasz -= 2 * sizeof(uint32_t);
2896 step = ROUND_UP(pr_datasz, ELF_GNU_PROPERTY_ALIGN);
2897 if (step > datasz) {
2898 goto error_data;
2899 }
2900
2901 /* Properties are supposed to be unique and sorted on pr_type. */
2902 if (have_prev_type && pr_type <= *prev_type) {
2903 if (pr_type == *prev_type) {
2904 error_setg(errp, "Duplicate property in PT_GNU_PROPERTY");
2905 } else {
2906 error_setg(errp, "Unsorted property in PT_GNU_PROPERTY");
2907 }
2908 return false;
2909 }
2910 *prev_type = pr_type;
2911
2912 if (!arch_parse_elf_property(pr_type, pr_datasz, data, info, errp)) {
2913 return false;
2914 }
2915
2916 *off += 2 * sizeof(uint32_t) + step;
2917 return true;
2918
2919 error_data:
2920 error_setg(errp, "Ill-formed property in PT_GNU_PROPERTY");
2921 return false;
2922 }
2923
2924 /* Process NT_GNU_PROPERTY_TYPE_0. */
2925 static bool parse_elf_properties(int image_fd,
2926 struct image_info *info,
2927 const struct elf_phdr *phdr,
2928 char bprm_buf[BPRM_BUF_SIZE],
2929 Error **errp)
2930 {
2931 union {
2932 struct elf_note nhdr;
2933 uint32_t data[NOTE_DATA_SZ / sizeof(uint32_t)];
2934 } note;
2935
2936 int n, off, datasz;
2937 bool have_prev_type;
2938 uint32_t prev_type;
2939
2940 /* Unless the arch requires properties, ignore them. */
2941 if (!ARCH_USE_GNU_PROPERTY) {
2942 return true;
2943 }
2944
2945 /* If the properties are crazy large, that's too bad. */
2946 n = phdr->p_filesz;
2947 if (n > sizeof(note)) {
2948 error_setg(errp, "PT_GNU_PROPERTY too large");
2949 return false;
2950 }
2951 if (n < sizeof(note.nhdr)) {
2952 error_setg(errp, "PT_GNU_PROPERTY too small");
2953 return false;
2954 }
2955
2956 if (phdr->p_offset + n <= BPRM_BUF_SIZE) {
2957 memcpy(&note, bprm_buf + phdr->p_offset, n);
2958 } else {
2959 ssize_t len = pread(image_fd, &note, n, phdr->p_offset);
2960 if (len != n) {
2961 error_setg_errno(errp, errno, "Error reading file header");
2962 return false;
2963 }
2964 }
2965
2966 /*
2967 * The contents of a valid PT_GNU_PROPERTY is a sequence
2968 * of uint32_t -- swap them all now.
2969 */
2970 #ifdef BSWAP_NEEDED
2971 for (int i = 0; i < n / 4; i++) {
2972 bswap32s(note.data + i);
2973 }
2974 #endif
2975
2976 /*
2977 * Note that nhdr is 3 words, and that the "name" described by namesz
2978 * immediately follows nhdr and is thus at the 4th word. Further, all
2979 * of the inputs to the kernel's round_up are multiples of 4.
2980 */
2981 if (note.nhdr.n_type != NT_GNU_PROPERTY_TYPE_0 ||
2982 note.nhdr.n_namesz != NOTE_NAME_SZ ||
2983 note.data[3] != GNU0_MAGIC) {
2984 error_setg(errp, "Invalid note in PT_GNU_PROPERTY");
2985 return false;
2986 }
2987 off = sizeof(note.nhdr) + NOTE_NAME_SZ;
2988
2989 datasz = note.nhdr.n_descsz + off;
2990 if (datasz > n) {
2991 error_setg(errp, "Invalid note size in PT_GNU_PROPERTY");
2992 return false;
2993 }
2994
2995 have_prev_type = false;
2996 prev_type = 0;
2997 while (1) {
2998 if (off == datasz) {
2999 return true; /* end, exit ok */
3000 }
3001 if (!parse_elf_property(note.data, &off, datasz, info,
3002 have_prev_type, &prev_type, errp)) {
3003 return false;
3004 }
3005 have_prev_type = true;
3006 }
3007 }
3008
3009 /* Load an ELF image into the address space.
3010
3011 IMAGE_NAME is the filename of the image, to use in error messages.
3012 IMAGE_FD is the open file descriptor for the image.
3013
3014 BPRM_BUF is a copy of the beginning of the file; this of course
3015 contains the elf file header at offset 0. It is assumed that this
3016 buffer is sufficiently aligned to present no problems to the host
3017 in accessing data at aligned offsets within the buffer.
3018
3019 On return: INFO values will be filled in, as necessary or available. */
3020
3021 static void load_elf_image(const char *image_name, int image_fd,
3022 struct image_info *info, char **pinterp_name,
3023 char bprm_buf[BPRM_BUF_SIZE])
3024 {
3025 struct elfhdr *ehdr = (struct elfhdr *)bprm_buf;
3026 struct elf_phdr *phdr;
3027 abi_ulong load_addr, load_bias, loaddr, hiaddr, error;
3028 int i, retval, prot_exec;
3029 Error *err = NULL;
3030
3031 /* First of all, some simple consistency checks */
3032 if (!elf_check_ident(ehdr)) {
3033 error_setg(&err, "Invalid ELF image for this architecture");
3034 goto exit_errmsg;
3035 }
3036 bswap_ehdr(ehdr);
3037 if (!elf_check_ehdr(ehdr)) {
3038 error_setg(&err, "Invalid ELF image for this architecture");
3039 goto exit_errmsg;
3040 }
3041
3042 i = ehdr->e_phnum * sizeof(struct elf_phdr);
3043 if (ehdr->e_phoff + i <= BPRM_BUF_SIZE) {
3044 phdr = (struct elf_phdr *)(bprm_buf + ehdr->e_phoff);
3045 } else {
3046 phdr = (struct elf_phdr *) alloca(i);
3047 retval = pread(image_fd, phdr, i, ehdr->e_phoff);
3048 if (retval != i) {
3049 goto exit_read;
3050 }
3051 }
3052 bswap_phdr(phdr, ehdr->e_phnum);
3053
3054 info->nsegs = 0;
3055 info->pt_dynamic_addr = 0;
3056
3057 mmap_lock();
3058
3059 /*
3060 * Find the maximum size of the image and allocate an appropriate
3061 * amount of memory to handle that. Locate the interpreter, if any.
3062 */
3063 loaddr = -1, hiaddr = 0;
3064 info->alignment = 0;
3065 info->exec_stack = EXSTACK_DEFAULT;
3066 for (i = 0; i < ehdr->e_phnum; ++i) {
3067 struct elf_phdr *eppnt = phdr + i;
3068 if (eppnt->p_type == PT_LOAD) {
3069 abi_ulong a = eppnt->p_vaddr - eppnt->p_offset;
3070 if (a < loaddr) {
3071 loaddr = a;
3072 }
3073 a = eppnt->p_vaddr + eppnt->p_memsz - 1;
3074 if (a > hiaddr) {
3075 hiaddr = a;
3076 }
3077 ++info->nsegs;
3078 info->alignment |= eppnt->p_align;
3079 } else if (eppnt->p_type == PT_INTERP && pinterp_name) {
3080 g_autofree char *interp_name = NULL;
3081
3082 if (*pinterp_name) {
3083 error_setg(&err, "Multiple PT_INTERP entries");
3084 goto exit_errmsg;
3085 }
3086
3087 interp_name = g_malloc(eppnt->p_filesz);
3088
3089 if (eppnt->p_offset + eppnt->p_filesz <= BPRM_BUF_SIZE) {
3090 memcpy(interp_name, bprm_buf + eppnt->p_offset,
3091 eppnt->p_filesz);
3092 } else {
3093 retval = pread(image_fd, interp_name, eppnt->p_filesz,
3094 eppnt->p_offset);
3095 if (retval != eppnt->p_filesz) {
3096 goto exit_read;
3097 }
3098 }
3099 if (interp_name[eppnt->p_filesz - 1] != 0) {
3100 error_setg(&err, "Invalid PT_INTERP entry");
3101 goto exit_errmsg;
3102 }
3103 *pinterp_name = g_steal_pointer(&interp_name);
3104 } else if (eppnt->p_type == PT_GNU_PROPERTY) {
3105 if (!parse_elf_properties(image_fd, info, eppnt, bprm_buf, &err)) {
3106 goto exit_errmsg;
3107 }
3108 } else if (eppnt->p_type == PT_GNU_STACK) {
3109 info->exec_stack = eppnt->p_flags & PF_X;
3110 }
3111 }
3112
3113 load_addr = loaddr;
3114
3115 if (pinterp_name != NULL) {
3116 if (ehdr->e_type == ET_EXEC) {
3117 /*
3118 * Make sure that the low address does not conflict with
3119 * MMAP_MIN_ADDR or the QEMU application itself.
3120 */
3121 probe_guest_base(image_name, loaddr, hiaddr);
3122 } else {
3123 abi_ulong align;
3124
3125 /*
3126 * The binary is dynamic, but we still need to
3127 * select guest_base. In this case we pass a size.
3128 */
3129 probe_guest_base(image_name, 0, hiaddr - loaddr);
3130
3131 /*
3132 * Avoid collision with the loader by providing a different
3133 * default load address.
3134 */
3135 load_addr += elf_et_dyn_base;
3136
3137 /*
3138 * TODO: Better support for mmap alignment is desirable.
3139 * Since we do not have complete control over the guest
3140 * address space, we prefer the kernel to choose some address
3141 * rather than force the use of LOAD_ADDR via MAP_FIXED.
3142 * But without MAP_FIXED we cannot guarantee alignment,
3143 * only suggest it.
3144 */
3145 align = pow2ceil(info->alignment);
3146 if (align) {
3147 load_addr &= -align;
3148 }
3149 }
3150 }
3151
3152 /*
3153 * Reserve address space for all of this.
3154 *
3155 * In the case of ET_EXEC, we supply MAP_FIXED_NOREPLACE so that we get
3156 * exactly the address range that is required. Without reserved_va,
3157 * the guest address space is not isolated. We have attempted to avoid
3158 * conflict with the host program itself via probe_guest_base, but using
3159 * MAP_FIXED_NOREPLACE instead of MAP_FIXED provides an extra check.
3160 *
3161 * Otherwise this is ET_DYN, and we are searching for a location
3162 * that can hold the memory space required. If the image is
3163 * pre-linked, LOAD_ADDR will be non-zero, and the kernel should
3164 * honor that address if it happens to be free.
3165 *
3166 * In both cases, we will overwrite pages in this range with mappings
3167 * from the executable.
3168 */
3169 load_addr = target_mmap(load_addr, (size_t)hiaddr - loaddr + 1, PROT_NONE,
3170 MAP_PRIVATE | MAP_ANON | MAP_NORESERVE |
3171 (ehdr->e_type == ET_EXEC ? MAP_FIXED_NOREPLACE : 0),
3172 -1, 0);
3173 if (load_addr == -1) {
3174 goto exit_mmap;
3175 }
3176 load_bias = load_addr - loaddr;
3177
3178 if (elf_is_fdpic(ehdr)) {
3179 struct elf32_fdpic_loadseg *loadsegs = info->loadsegs =
3180 g_malloc(sizeof(*loadsegs) * info->nsegs);
3181
3182 for (i = 0; i < ehdr->e_phnum; ++i) {
3183 switch (phdr[i].p_type) {
3184 case PT_DYNAMIC:
3185 info->pt_dynamic_addr = phdr[i].p_vaddr + load_bias;
3186 break;
3187 case PT_LOAD:
3188 loadsegs->addr = phdr[i].p_vaddr + load_bias;
3189 loadsegs->p_vaddr = phdr[i].p_vaddr;
3190 loadsegs->p_memsz = phdr[i].p_memsz;
3191 ++loadsegs;
3192 break;
3193 }
3194 }
3195 }
3196
3197 info->load_bias = load_bias;
3198 info->code_offset = load_bias;
3199 info->data_offset = load_bias;
3200 info->load_addr = load_addr;
3201 info->entry = ehdr->e_entry + load_bias;
3202 info->start_code = -1;
3203 info->end_code = 0;
3204 info->start_data = -1;
3205 info->end_data = 0;
3206 /* Usual start for brk is after all sections of the main executable. */
3207 info->brk = TARGET_PAGE_ALIGN(hiaddr);
3208 info->elf_flags = ehdr->e_flags;
3209
3210 prot_exec = PROT_EXEC;
3211 #ifdef TARGET_AARCH64
3212 /*
3213 * If the BTI feature is present, this indicates that the executable
3214 * pages of the startup binary should be mapped with PROT_BTI, so that
3215 * branch targets are enforced.
3216 *
3217 * The startup binary is either the interpreter or the static executable.
3218 * The interpreter is responsible for all pages of a dynamic executable.
3219 *
3220 * Elf notes are backward compatible to older cpus.
3221 * Do not enable BTI unless it is supported.
3222 */
3223 if ((info->note_flags & GNU_PROPERTY_AARCH64_FEATURE_1_BTI)
3224 && (pinterp_name == NULL || *pinterp_name == 0)
3225 && cpu_isar_feature(aa64_bti, ARM_CPU(thread_cpu))) {
3226 prot_exec |= TARGET_PROT_BTI;
3227 }
3228 #endif
3229
3230 for (i = 0; i < ehdr->e_phnum; i++) {
3231 struct elf_phdr *eppnt = phdr + i;
3232 if (eppnt->p_type == PT_LOAD) {
3233 abi_ulong vaddr, vaddr_po, vaddr_ps, vaddr_ef, vaddr_em;
3234 int elf_prot = 0;
3235
3236 if (eppnt->p_flags & PF_R) {
3237 elf_prot |= PROT_READ;
3238 }
3239 if (eppnt->p_flags & PF_W) {
3240 elf_prot |= PROT_WRITE;
3241 }
3242 if (eppnt->p_flags & PF_X) {
3243 elf_prot |= prot_exec;
3244 }
3245
3246 vaddr = load_bias + eppnt->p_vaddr;
3247 vaddr_po = vaddr & ~TARGET_PAGE_MASK;
3248 vaddr_ps = vaddr & TARGET_PAGE_MASK;
3249
3250 vaddr_ef = vaddr + eppnt->p_filesz;
3251 vaddr_em = vaddr + eppnt->p_memsz;
3252
3253 /*
3254 * Some segments may be completely empty, with a non-zero p_memsz
3255 * but no backing file segment.
3256 */
3257 if (eppnt->p_filesz != 0) {
3258 error = target_mmap(vaddr_ps, eppnt->p_filesz + vaddr_po,
3259 elf_prot, MAP_PRIVATE | MAP_FIXED,
3260 image_fd, eppnt->p_offset - vaddr_po);
3261 if (error == -1) {
3262 goto exit_mmap;
3263 }
3264 }
3265
3266 /* If the load segment requests extra zeros (e.g. bss), map it. */
3267 if (vaddr_ef < vaddr_em &&
3268 !zero_bss(vaddr_ef, vaddr_em, elf_prot)) {
3269 goto exit_mmap;
3270 }
3271
3272 /* Find the full program boundaries. */
3273 if (elf_prot & PROT_EXEC) {
3274 if (vaddr < info->start_code) {
3275 info->start_code = vaddr;
3276 }
3277 if (vaddr_ef > info->end_code) {
3278 info->end_code = vaddr_ef;
3279 }
3280 }
3281 if (elf_prot & PROT_WRITE) {
3282 if (vaddr < info->start_data) {
3283 info->start_data = vaddr;
3284 }
3285 if (vaddr_ef > info->end_data) {
3286 info->end_data = vaddr_ef;
3287 }
3288 }
3289 #ifdef TARGET_MIPS
3290 } else if (eppnt->p_type == PT_MIPS_ABIFLAGS) {
3291 Mips_elf_abiflags_v0 abiflags;
3292 if (eppnt->p_filesz < sizeof(Mips_elf_abiflags_v0)) {
3293 error_setg(&err, "Invalid PT_MIPS_ABIFLAGS entry");
3294 goto exit_errmsg;
3295 }
3296 if (eppnt->p_offset + eppnt->p_filesz <= BPRM_BUF_SIZE) {
3297 memcpy(&abiflags, bprm_buf + eppnt->p_offset,
3298 sizeof(Mips_elf_abiflags_v0));
3299 } else {
3300 retval = pread(image_fd, &abiflags, sizeof(Mips_elf_abiflags_v0),
3301 eppnt->p_offset);
3302 if (retval != sizeof(Mips_elf_abiflags_v0)) {
3303 goto exit_read;
3304 }
3305 }
3306 bswap_mips_abiflags(&abiflags);
3307 info->fp_abi = abiflags.fp_abi;
3308 #endif
3309 }
3310 }
3311
3312 if (info->end_data == 0) {
3313 info->start_data = info->end_code;
3314 info->end_data = info->end_code;
3315 }
3316
3317 if (qemu_log_enabled()) {
3318 load_symbols(ehdr, image_fd, load_bias);
3319 }
3320
3321 debuginfo_report_elf(image_name, image_fd, load_bias);
3322
3323 mmap_unlock();
3324
3325 close(image_fd);
3326 return;
3327
3328 exit_read:
3329 if (retval >= 0) {
3330 error_setg(&err, "Incomplete read of file header");
3331 } else {
3332 error_setg_errno(&err, errno, "Error reading file header");
3333 }
3334 goto exit_errmsg;
3335 exit_mmap:
3336 error_setg_errno(&err, errno, "Error mapping file");
3337 goto exit_errmsg;
3338 exit_errmsg:
3339 error_reportf_err(err, "%s: ", image_name);
3340 exit(-1);
3341 }
3342
3343 static void load_elf_interp(const char *filename, struct image_info *info,
3344 char bprm_buf[BPRM_BUF_SIZE])
3345 {
3346 int fd, retval;
3347 Error *err = NULL;
3348
3349 fd = open(path(filename), O_RDONLY);
3350 if (fd < 0) {
3351 error_setg_file_open(&err, errno, filename);
3352 error_report_err(err);
3353 exit(-1);
3354 }
3355
3356 retval = read(fd, bprm_buf, BPRM_BUF_SIZE);
3357 if (retval < 0) {
3358 error_setg_errno(&err, errno, "Error reading file header");
3359 error_reportf_err(err, "%s: ", filename);
3360 exit(-1);
3361 }
3362
3363 if (retval < BPRM_BUF_SIZE) {
3364 memset(bprm_buf + retval, 0, BPRM_BUF_SIZE - retval);
3365 }
3366
3367 load_elf_image(filename, fd, info, NULL, bprm_buf);
3368 }
3369
3370 static int symfind(const void *s0, const void *s1)
3371 {
3372 struct elf_sym *sym = (struct elf_sym *)s1;
3373 __typeof(sym->st_value) addr = *(uint64_t *)s0;
3374 int result = 0;
3375
3376 if (addr < sym->st_value) {
3377 result = -1;
3378 } else if (addr >= sym->st_value + sym->st_size) {
3379 result = 1;
3380 }
3381 return result;
3382 }
3383
3384 static const char *lookup_symbolxx(struct syminfo *s, uint64_t orig_addr)
3385 {
3386 #if ELF_CLASS == ELFCLASS32
3387 struct elf_sym *syms = s->disas_symtab.elf32;
3388 #else
3389 struct elf_sym *syms = s->disas_symtab.elf64;
3390 #endif
3391
3392 // binary search
3393 struct elf_sym *sym;
3394
3395 sym = bsearch(&orig_addr, syms, s->disas_num_syms, sizeof(*syms), symfind);
3396 if (sym != NULL) {
3397 return s->disas_strtab + sym->st_name;
3398 }
3399
3400 return "";
3401 }
3402
3403 /* FIXME: This should use elf_ops.h */
3404 static int symcmp(const void *s0, const void *s1)
3405 {
3406 struct elf_sym *sym0 = (struct elf_sym *)s0;
3407 struct elf_sym *sym1 = (struct elf_sym *)s1;
3408 return (sym0->st_value < sym1->st_value)
3409 ? -1
3410 : ((sym0->st_value > sym1->st_value) ? 1 : 0);
3411 }
3412
3413 /* Best attempt to load symbols from this ELF object. */
3414 static void load_symbols(struct elfhdr *hdr, int fd, abi_ulong load_bias)
3415 {
3416 int i, shnum, nsyms, sym_idx = 0, str_idx = 0;
3417 uint64_t segsz;
3418 struct elf_shdr *shdr;
3419 char *strings = NULL;
3420 struct syminfo *s = NULL;
3421 struct elf_sym *new_syms, *syms = NULL;
3422
3423 shnum = hdr->e_shnum;
3424 i = shnum * sizeof(struct elf_shdr);
3425 shdr = (struct elf_shdr *)alloca(i);
3426 if (pread(fd, shdr, i, hdr->e_shoff) != i) {
3427 return;
3428 }
3429
3430 bswap_shdr(shdr, shnum);
3431 for (i = 0; i < shnum; ++i) {
3432 if (shdr[i].sh_type == SHT_SYMTAB) {
3433 sym_idx = i;
3434 str_idx = shdr[i].sh_link;
3435 goto found;
3436 }
3437 }
3438
3439 /* There will be no symbol table if the file was stripped. */
3440 return;
3441
3442 found:
3443 /* Now know where the strtab and symtab are. Snarf them. */
3444 s = g_try_new(struct syminfo, 1);
3445 if (!s) {
3446 goto give_up;
3447 }
3448
3449 segsz = shdr[str_idx].sh_size;
3450 s->disas_strtab = strings = g_try_malloc(segsz);
3451 if (!strings ||
3452 pread(fd, strings, segsz, shdr[str_idx].sh_offset) != segsz) {
3453 goto give_up;
3454 }
3455
3456 segsz = shdr[sym_idx].sh_size;
3457 syms = g_try_malloc(segsz);
3458 if (!syms || pread(fd, syms, segsz, shdr[sym_idx].sh_offset) != segsz) {
3459 goto give_up;
3460 }
3461
3462 if (segsz / sizeof(struct elf_sym) > INT_MAX) {
3463 /* Implausibly large symbol table: give up rather than ploughing
3464 * on with the number of symbols calculation overflowing
3465 */
3466 goto give_up;
3467 }
3468 nsyms = segsz / sizeof(struct elf_sym);
3469 for (i = 0; i < nsyms; ) {
3470 bswap_sym(syms + i);
3471 /* Throw away entries which we do not need. */
3472 if (syms[i].st_shndx == SHN_UNDEF
3473 || syms[i].st_shndx >= SHN_LORESERVE
3474 || ELF_ST_TYPE(syms[i].st_info) != STT_FUNC) {
3475 if (i < --nsyms) {
3476 syms[i] = syms[nsyms];
3477 }
3478 } else {
3479 #if defined(TARGET_ARM) || defined (TARGET_MIPS)
3480 /* The bottom address bit marks a Thumb or MIPS16 symbol. */
3481 syms[i].st_value &= ~(target_ulong)1;
3482 #endif
3483 syms[i].st_value += load_bias;
3484 i++;
3485 }
3486 }
3487
3488 /* No "useful" symbol. */
3489 if (nsyms == 0) {
3490 goto give_up;
3491 }
3492
3493 /* Attempt to free the storage associated with the local symbols
3494 that we threw away. Whether or not this has any effect on the
3495 memory allocation depends on the malloc implementation and how
3496 many symbols we managed to discard. */
3497 new_syms = g_try_renew(struct elf_sym, syms, nsyms);
3498 if (new_syms == NULL) {
3499 goto give_up;
3500 }
3501 syms = new_syms;
3502
3503 qsort(syms, nsyms, sizeof(*syms), symcmp);
3504
3505 s->disas_num_syms = nsyms;
3506 #if ELF_CLASS == ELFCLASS32
3507 s->disas_symtab.elf32 = syms;
3508 #else
3509 s->disas_symtab.elf64 = syms;
3510 #endif
3511 s->lookup_symbol = lookup_symbolxx;
3512 s->next = syminfos;
3513 syminfos = s;
3514
3515 return;
3516
3517 give_up:
3518 g_free(s);
3519 g_free(strings);
3520 g_free(syms);
3521 }
3522
3523 uint32_t get_elf_eflags(int fd)
3524 {
3525 struct elfhdr ehdr;
3526 off_t offset;
3527 int ret;
3528
3529 /* Read ELF header */
3530 offset = lseek(fd, 0, SEEK_SET);
3531 if (offset == (off_t) -1) {
3532 return 0;
3533 }
3534 ret = read(fd, &ehdr, sizeof(ehdr));
3535 if (ret < sizeof(ehdr)) {
3536 return 0;
3537 }
3538 offset = lseek(fd, offset, SEEK_SET);
3539 if (offset == (off_t) -1) {
3540 return 0;
3541 }
3542
3543 /* Check ELF signature */
3544 if (!elf_check_ident(&ehdr)) {
3545 return 0;
3546 }
3547
3548 /* check header */
3549 bswap_ehdr(&ehdr);
3550 if (!elf_check_ehdr(&ehdr)) {
3551 return 0;
3552 }
3553
3554 /* return architecture id */
3555 return ehdr.e_flags;
3556 }
3557
3558 int load_elf_binary(struct linux_binprm *bprm, struct image_info *info)
3559 {
3560 struct image_info interp_info;
3561 struct elfhdr elf_ex;
3562 char *elf_interpreter = NULL;
3563 char *scratch;
3564
3565 memset(&interp_info, 0, sizeof(interp_info));
3566 #ifdef TARGET_MIPS
3567 interp_info.fp_abi = MIPS_ABI_FP_UNKNOWN;
3568 #endif
3569
3570 info->start_mmap = (abi_ulong)ELF_START_MMAP;
3571
3572 load_elf_image(bprm->filename, bprm->fd, info,
3573 &elf_interpreter, bprm->buf);
3574
3575 /* ??? We need a copy of the elf header for passing to create_elf_tables.
3576 If we do nothing, we'll have overwritten this when we re-use bprm->buf
3577 when we load the interpreter. */
3578 elf_ex = *(struct elfhdr *)bprm->buf;
3579
3580 /* Do this so that we can load the interpreter, if need be. We will
3581 change some of these later */
3582 bprm->p = setup_arg_pages(bprm, info);
3583
3584 scratch = g_new0(char, TARGET_PAGE_SIZE);
3585 if (STACK_GROWS_DOWN) {
3586 bprm->p = copy_elf_strings(1, &bprm->filename, scratch,
3587 bprm->p, info->stack_limit);
3588 info->file_string = bprm->p;
3589 bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch,
3590 bprm->p, info->stack_limit);
3591 info->env_strings = bprm->p;
3592 bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch,
3593 bprm->p, info->stack_limit);
3594 info->arg_strings = bprm->p;
3595 } else {
3596 info->arg_strings = bprm->p;
3597 bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch,
3598 bprm->p, info->stack_limit);
3599 info->env_strings = bprm->p;
3600 bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch,
3601 bprm->p, info->stack_limit);
3602 info->file_string = bprm->p;
3603 bprm->p = copy_elf_strings(1, &bprm->filename, scratch,
3604 bprm->p, info->stack_limit);
3605 }
3606
3607 g_free(scratch);
3608
3609 if (!bprm->p) {
3610 fprintf(stderr, "%s: %s\n", bprm->filename, strerror(E2BIG));
3611 exit(-1);
3612 }
3613
3614 if (elf_interpreter) {
3615 load_elf_interp(elf_interpreter, &interp_info, bprm->buf);
3616
3617 /*
3618 * While unusual because of ELF_ET_DYN_BASE, if we are unlucky
3619 * with the mappings the interpreter can be loaded above but
3620 * near the main executable, which can leave very little room
3621 * for the heap.
3622 * If the current brk has less than 16MB, use the end of the
3623 * interpreter.
3624 */
3625 if (interp_info.brk > info->brk &&
3626 interp_info.load_bias - info->brk < 16 * MiB) {
3627 info->brk = interp_info.brk;
3628 }
3629
3630 /* If the program interpreter is one of these two, then assume
3631 an iBCS2 image. Otherwise assume a native linux image. */
3632
3633 if (strcmp(elf_interpreter, "/usr/lib/libc.so.1") == 0
3634 || strcmp(elf_interpreter, "/usr/lib/ld.so.1") == 0) {
3635 info->personality = PER_SVR4;
3636
3637 /* Why this, you ask??? Well SVr4 maps page 0 as read-only,
3638 and some applications "depend" upon this behavior. Since
3639 we do not have the power to recompile these, we emulate
3640 the SVr4 behavior. Sigh. */
3641 target_mmap(0, qemu_host_page_size, PROT_READ | PROT_EXEC,
3642 MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
3643 }
3644 #ifdef TARGET_MIPS
3645 info->interp_fp_abi = interp_info.fp_abi;
3646 #endif
3647 }
3648
3649 /*
3650 * TODO: load a vdso, which would also contain the signal trampolines.
3651 * Otherwise, allocate a private page to hold them.
3652 */
3653 if (TARGET_ARCH_HAS_SIGTRAMP_PAGE) {
3654 abi_long tramp_page = target_mmap(0, TARGET_PAGE_SIZE,
3655 PROT_READ | PROT_WRITE,
3656 MAP_PRIVATE | MAP_ANON, -1, 0);
3657 if (tramp_page == -1) {
3658 return -errno;
3659 }
3660
3661 setup_sigtramp(tramp_page);
3662 target_mprotect(tramp_page, TARGET_PAGE_SIZE, PROT_READ | PROT_EXEC);
3663 }
3664
3665 bprm->p = create_elf_tables(bprm->p, bprm->argc, bprm->envc, &elf_ex,
3666 info, (elf_interpreter ? &interp_info : NULL));
3667 info->start_stack = bprm->p;
3668
3669 /* If we have an interpreter, set that as the program's entry point.
3670 Copy the load_bias as well, to help PPC64 interpret the entry
3671 point as a function descriptor. Do this after creating elf tables
3672 so that we copy the original program entry point into the AUXV. */
3673 if (elf_interpreter) {
3674 info->load_bias = interp_info.load_bias;
3675 info->entry = interp_info.entry;
3676 g_free(elf_interpreter);
3677 }
3678
3679 #ifdef USE_ELF_CORE_DUMP
3680 bprm->core_dump = &elf_core_dump;
3681 #endif
3682
3683 return 0;
3684 }
3685
3686 #ifdef USE_ELF_CORE_DUMP
3687 /*
3688 * Definitions to generate Intel SVR4-like core files.
3689 * These mostly have the same names as the SVR4 types with "target_elf_"
3690 * tacked on the front to prevent clashes with linux definitions,
3691 * and the typedef forms have been avoided. This is mostly like
3692 * the SVR4 structure, but more Linuxy, with things that Linux does
3693 * not support and which gdb doesn't really use excluded.
3694 *
3695 * Fields we don't dump (their contents is zero) in linux-user qemu
3696 * are marked with XXX.
3697 *
3698 * Core dump code is copied from linux kernel (fs/binfmt_elf.c).
3699 *
3700 * Porting ELF coredump for target is (quite) simple process. First you
3701 * define USE_ELF_CORE_DUMP in target ELF code (where init_thread() for
3702 * the target resides):
3703 *
3704 * #define USE_ELF_CORE_DUMP
3705 *
3706 * Next you define type of register set used for dumping. ELF specification
3707 * says that it needs to be array of elf_greg_t that has size of ELF_NREG.
3708 *
3709 * typedef <target_regtype> target_elf_greg_t;
3710 * #define ELF_NREG <number of registers>
3711 * typedef taret_elf_greg_t target_elf_gregset_t[ELF_NREG];
3712 *
3713 * Last step is to implement target specific function that copies registers
3714 * from given cpu into just specified register set. Prototype is:
3715 *
3716 * static void elf_core_copy_regs(taret_elf_gregset_t *regs,
3717 * const CPUArchState *env);
3718 *
3719 * Parameters:
3720 * regs - copy register values into here (allocated and zeroed by caller)
3721 * env - copy registers from here
3722 *
3723 * Example for ARM target is provided in this file.
3724 */
3725
3726 /* An ELF note in memory */
3727 struct memelfnote {
3728 const char *name;
3729 size_t namesz;
3730 size_t namesz_rounded;
3731 int type;
3732 size_t datasz;
3733 size_t datasz_rounded;
3734 void *data;
3735 size_t notesz;
3736 };
3737
3738 struct target_elf_siginfo {
3739 abi_int si_signo; /* signal number */
3740 abi_int si_code; /* extra code */
3741 abi_int si_errno; /* errno */
3742 };
3743
3744 struct target_elf_prstatus {
3745 struct target_elf_siginfo pr_info; /* Info associated with signal */
3746 abi_short pr_cursig; /* Current signal */
3747 abi_ulong pr_sigpend; /* XXX */
3748 abi_ulong pr_sighold; /* XXX */
3749 target_pid_t pr_pid;
3750 target_pid_t pr_ppid;
3751 target_pid_t pr_pgrp;
3752 target_pid_t pr_sid;
3753 struct target_timeval pr_utime; /* XXX User time */
3754 struct target_timeval pr_stime; /* XXX System time */
3755 struct target_timeval pr_cutime; /* XXX Cumulative user time */
3756 struct target_timeval pr_cstime; /* XXX Cumulative system time */
3757 target_elf_gregset_t pr_reg; /* GP registers */
3758 abi_int pr_fpvalid; /* XXX */
3759 };
3760
3761 #define ELF_PRARGSZ (80) /* Number of chars for args */
3762
3763 struct target_elf_prpsinfo {
3764 char pr_state; /* numeric process state */
3765 char pr_sname; /* char for pr_state */
3766 char pr_zomb; /* zombie */
3767 char pr_nice; /* nice val */
3768 abi_ulong pr_flag; /* flags */
3769 target_uid_t pr_uid;
3770 target_gid_t pr_gid;
3771 target_pid_t pr_pid, pr_ppid, pr_pgrp, pr_sid;
3772 /* Lots missing */
3773 char pr_fname[16] QEMU_NONSTRING; /* filename of executable */
3774 char pr_psargs[ELF_PRARGSZ]; /* initial part of arg list */
3775 };
3776
3777 /* Here is the structure in which status of each thread is captured. */
3778 struct elf_thread_status {
3779 QTAILQ_ENTRY(elf_thread_status) ets_link;
3780 struct target_elf_prstatus prstatus; /* NT_PRSTATUS */
3781 #if 0
3782 elf_fpregset_t fpu; /* NT_PRFPREG */
3783 struct task_struct *thread;
3784 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */
3785 #endif
3786 struct memelfnote notes[1];
3787 int num_notes;
3788 };
3789
3790 struct elf_note_info {
3791 struct memelfnote *notes;
3792 struct target_elf_prstatus *prstatus; /* NT_PRSTATUS */
3793 struct target_elf_prpsinfo *psinfo; /* NT_PRPSINFO */
3794
3795 QTAILQ_HEAD(, elf_thread_status) thread_list;
3796 #if 0
3797 /*
3798 * Current version of ELF coredump doesn't support
3799 * dumping fp regs etc.
3800 */
3801 elf_fpregset_t *fpu;
3802 elf_fpxregset_t *xfpu;
3803 int thread_status_size;
3804 #endif
3805 int notes_size;
3806 int numnote;
3807 };
3808
3809 struct vm_area_struct {
3810 target_ulong vma_start; /* start vaddr of memory region */
3811 target_ulong vma_end; /* end vaddr of memory region */
3812 abi_ulong vma_flags; /* protection etc. flags for the region */
3813 QTAILQ_ENTRY(vm_area_struct) vma_link;
3814 };
3815
3816 struct mm_struct {
3817 QTAILQ_HEAD(, vm_area_struct) mm_mmap;
3818 int mm_count; /* number of mappings */
3819 };
3820
3821 static struct mm_struct *vma_init(void);
3822 static void vma_delete(struct mm_struct *);
3823 static int vma_add_mapping(struct mm_struct *, target_ulong,
3824 target_ulong, abi_ulong);
3825 static int vma_get_mapping_count(const struct mm_struct *);
3826 static struct vm_area_struct *vma_first(const struct mm_struct *);
3827 static struct vm_area_struct *vma_next(struct vm_area_struct *);
3828 static abi_ulong vma_dump_size(const struct vm_area_struct *);
3829 static int vma_walker(void *priv, target_ulong start, target_ulong end,
3830 unsigned long flags);
3831
3832 static void fill_elf_header(struct elfhdr *, int, uint16_t, uint32_t);
3833 static void fill_note(struct memelfnote *, const char *, int,
3834 unsigned int, void *);
3835 static void fill_prstatus(struct target_elf_prstatus *, const TaskState *, int);
3836 static int fill_psinfo(struct target_elf_prpsinfo *, const TaskState *);
3837 static void fill_auxv_note(struct memelfnote *, const TaskState *);
3838 static void fill_elf_note_phdr(struct elf_phdr *, int, off_t);
3839 static size_t note_size(const struct memelfnote *);
3840 static void free_note_info(struct elf_note_info *);
3841 static int fill_note_info(struct elf_note_info *, long, const CPUArchState *);
3842 static void fill_thread_info(struct elf_note_info *, const CPUArchState *);
3843
3844 static int dump_write(int, const void *, size_t);
3845 static int write_note(struct memelfnote *, int);
3846 static int write_note_info(struct elf_note_info *, int);
3847
3848 #ifdef BSWAP_NEEDED
3849 static void bswap_prstatus(struct target_elf_prstatus *prstatus)
3850 {
3851 prstatus->pr_info.si_signo = tswap32(prstatus->pr_info.si_signo);
3852 prstatus->pr_info.si_code = tswap32(prstatus->pr_info.si_code);
3853 prstatus->pr_info.si_errno = tswap32(prstatus->pr_info.si_errno);
3854 prstatus->pr_cursig = tswap16(prstatus->pr_cursig);
3855 prstatus->pr_sigpend = tswapal(prstatus->pr_sigpend);
3856 prstatus->pr_sighold = tswapal(prstatus->pr_sighold);
3857 prstatus->pr_pid = tswap32(prstatus->pr_pid);
3858 prstatus->pr_ppid = tswap32(prstatus->pr_ppid);
3859 prstatus->pr_pgrp = tswap32(prstatus->pr_pgrp);
3860 prstatus->pr_sid = tswap32(prstatus->pr_sid);
3861 /* cpu times are not filled, so we skip them */
3862 /* regs should be in correct format already */
3863 prstatus->pr_fpvalid = tswap32(prstatus->pr_fpvalid);
3864 }
3865
3866 static void bswap_psinfo(struct target_elf_prpsinfo *psinfo)
3867 {
3868 psinfo->pr_flag = tswapal(psinfo->pr_flag);
3869 psinfo->pr_uid = tswap16(psinfo->pr_uid);
3870 psinfo->pr_gid = tswap16(psinfo->pr_gid);
3871 psinfo->pr_pid = tswap32(psinfo->pr_pid);
3872 psinfo->pr_ppid = tswap32(psinfo->pr_ppid);
3873 psinfo->pr_pgrp = tswap32(psinfo->pr_pgrp);
3874 psinfo->pr_sid = tswap32(psinfo->pr_sid);
3875 }
3876
3877 static void bswap_note(struct elf_note *en)
3878 {
3879 bswap32s(&en->n_namesz);
3880 bswap32s(&en->n_descsz);
3881 bswap32s(&en->n_type);
3882 }
3883 #else
3884 static inline void bswap_prstatus(struct target_elf_prstatus *p) { }
3885 static inline void bswap_psinfo(struct target_elf_prpsinfo *p) {}
3886 static inline void bswap_note(struct elf_note *en) { }
3887 #endif /* BSWAP_NEEDED */
3888
3889 /*
3890 * Minimal support for linux memory regions. These are needed
3891 * when we are finding out what memory exactly belongs to
3892 * emulated process. No locks needed here, as long as
3893 * thread that received the signal is stopped.
3894 */
3895
3896 static struct mm_struct *vma_init(void)
3897 {
3898 struct mm_struct *mm;
3899
3900 if ((mm = g_malloc(sizeof (*mm))) == NULL)
3901 return (NULL);
3902
3903 mm->mm_count = 0;
3904 QTAILQ_INIT(&mm->mm_mmap);
3905
3906 return (mm);
3907 }
3908
3909 static void vma_delete(struct mm_struct *mm)
3910 {
3911 struct vm_area_struct *vma;
3912
3913 while ((vma = vma_first(mm)) != NULL) {
3914 QTAILQ_REMOVE(&mm->mm_mmap, vma, vma_link);
3915 g_free(vma);
3916 }
3917 g_free(mm);
3918 }
3919
3920 static int vma_add_mapping(struct mm_struct *mm, target_ulong start,
3921 target_ulong end, abi_ulong flags)
3922 {
3923 struct vm_area_struct *vma;
3924
3925 if ((vma = g_malloc0(sizeof (*vma))) == NULL)
3926 return (-1);
3927
3928 vma->vma_start = start;
3929 vma->vma_end = end;
3930 vma->vma_flags = flags;
3931
3932 QTAILQ_INSERT_TAIL(&mm->mm_mmap, vma, vma_link);
3933 mm->mm_count++;
3934
3935 return (0);
3936 }
3937
3938 static struct vm_area_struct *vma_first(const struct mm_struct *mm)
3939 {
3940 return (QTAILQ_FIRST(&mm->mm_mmap));
3941 }
3942
3943 static struct vm_area_struct *vma_next(struct vm_area_struct *vma)
3944 {
3945 return (QTAILQ_NEXT(vma, vma_link));
3946 }
3947
3948 static int vma_get_mapping_count(const struct mm_struct *mm)
3949 {
3950 return (mm->mm_count);
3951 }
3952
3953 /*
3954 * Calculate file (dump) size of given memory region.
3955 */
3956 static abi_ulong vma_dump_size(const struct vm_area_struct *vma)
3957 {
3958 /* if we cannot even read the first page, skip it */
3959 if (!access_ok_untagged(VERIFY_READ, vma->vma_start, TARGET_PAGE_SIZE))
3960 return (0);
3961
3962 /*
3963 * Usually we don't dump executable pages as they contain
3964 * non-writable code that debugger can read directly from
3965 * target library etc. However, thread stacks are marked
3966 * also executable so we read in first page of given region
3967 * and check whether it contains elf header. If there is
3968 * no elf header, we dump it.
3969 */
3970 if (vma->vma_flags & PROT_EXEC) {
3971 char page[TARGET_PAGE_SIZE];
3972
3973 if (copy_from_user(page, vma->vma_start, sizeof (page))) {
3974 return 0;
3975 }
3976 if ((page[EI_MAG0] == ELFMAG0) &&
3977 (page[EI_MAG1] == ELFMAG1) &&
3978 (page[EI_MAG2] == ELFMAG2) &&
3979 (page[EI_MAG3] == ELFMAG3)) {
3980 /*
3981 * Mappings are possibly from ELF binary. Don't dump
3982 * them.
3983 */
3984 return (0);
3985 }
3986 }
3987
3988 return (vma->vma_end - vma->vma_start);
3989 }
3990
3991 static int vma_walker(void *priv, target_ulong start, target_ulong end,
3992 unsigned long flags)
3993 {
3994 struct mm_struct *mm = (struct mm_struct *)priv;
3995
3996 vma_add_mapping(mm, start, end, flags);
3997 return (0);
3998 }
3999
4000 static void fill_note(struct memelfnote *note, const char *name, int type,
4001 unsigned int sz, void *data)
4002 {
4003 unsigned int namesz;
4004
4005 namesz = strlen(name) + 1;
4006 note->name = name;
4007 note->namesz = namesz;
4008 note->namesz_rounded = roundup(namesz, sizeof (int32_t));
4009 note->type = type;
4010 note->datasz = sz;
4011 note->datasz_rounded = roundup(sz, sizeof (int32_t));
4012
4013 note->data = data;
4014
4015 /*
4016 * We calculate rounded up note size here as specified by
4017 * ELF document.
4018 */
4019 note->notesz = sizeof (struct elf_note) +
4020 note->namesz_rounded + note->datasz_rounded;
4021 }
4022
4023 static void fill_elf_header(struct elfhdr *elf, int segs, uint16_t machine,
4024 uint32_t flags)
4025 {
4026 (void) memset(elf, 0, sizeof(*elf));
4027
4028 (void) memcpy(elf->e_ident, ELFMAG, SELFMAG);
4029 elf->e_ident[EI_CLASS] = ELF_CLASS;
4030 elf->e_ident[EI_DATA] = ELF_DATA;
4031 elf->e_ident[EI_VERSION] = EV_CURRENT;
4032 elf->e_ident[EI_OSABI] = ELF_OSABI;
4033
4034 elf->e_type = ET_CORE;
4035 elf->e_machine = machine;
4036 elf->e_version = EV_CURRENT;
4037 elf->e_phoff = sizeof(struct elfhdr);
4038 elf->e_flags = flags;
4039 elf->e_ehsize = sizeof(struct elfhdr);
4040 elf->e_phentsize = sizeof(struct elf_phdr);
4041 elf->e_phnum = segs;
4042
4043 bswap_ehdr(elf);
4044 }
4045
4046 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, off_t offset)
4047 {
4048 phdr->p_type = PT_NOTE;
4049 phdr->p_offset = offset;
4050 phdr->p_vaddr = 0;
4051 phdr->p_paddr = 0;
4052 phdr->p_filesz = sz;
4053 phdr->p_memsz = 0;
4054 phdr->p_flags = 0;
4055 phdr->p_align = 0;
4056
4057 bswap_phdr(phdr, 1);
4058 }
4059
4060 static size_t note_size(const struct memelfnote *note)
4061 {
4062 return (note->notesz);
4063 }
4064
4065 static void fill_prstatus(struct target_elf_prstatus *prstatus,
4066 const TaskState *ts, int signr)
4067 {
4068 (void) memset(prstatus, 0, sizeof (*prstatus));
4069 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
4070 prstatus->pr_pid = ts->ts_tid;
4071 prstatus->pr_ppid = getppid();
4072 prstatus->pr_pgrp = getpgrp();
4073 prstatus->pr_sid = getsid(0);
4074
4075 bswap_prstatus(prstatus);
4076 }
4077
4078 static int fill_psinfo(struct target_elf_prpsinfo *psinfo, const TaskState *ts)
4079 {
4080 char *base_filename;
4081 unsigned int i, len;
4082
4083 (void) memset(psinfo, 0, sizeof (*psinfo));
4084
4085 len = ts->info->env_strings - ts->info->arg_strings;
4086 if (len >= ELF_PRARGSZ)
4087 len = ELF_PRARGSZ - 1;
4088 if (copy_from_user(&psinfo->pr_psargs, ts->info->arg_strings, len)) {
4089 return -EFAULT;
4090 }
4091 for (i = 0; i < len; i++)
4092 if (psinfo->pr_psargs[i] == 0)
4093 psinfo->pr_psargs[i] = ' ';
4094 psinfo->pr_psargs[len] = 0;
4095
4096 psinfo->pr_pid = getpid();
4097 psinfo->pr_ppid = getppid();
4098 psinfo->pr_pgrp = getpgrp();
4099 psinfo->pr_sid = getsid(0);
4100 psinfo->pr_uid = getuid();
4101 psinfo->pr_gid = getgid();
4102
4103 base_filename = g_path_get_basename(ts->bprm->filename);
4104 /*
4105 * Using strncpy here is fine: at max-length,
4106 * this field is not NUL-terminated.
4107 */
4108 (void) strncpy(psinfo->pr_fname, base_filename,
4109 sizeof(psinfo->pr_fname));
4110
4111 g_free(base_filename);
4112 bswap_psinfo(psinfo);
4113 return (0);
4114 }
4115
4116 static void fill_auxv_note(struct memelfnote *note, const TaskState *ts)
4117 {
4118 elf_addr_t auxv = (elf_addr_t)ts->info->saved_auxv;
4119 elf_addr_t orig_auxv = auxv;
4120 void *ptr;
4121 int len = ts->info->auxv_len;
4122
4123 /*
4124 * Auxiliary vector is stored in target process stack. It contains
4125 * {type, value} pairs that we need to dump into note. This is not
4126 * strictly necessary but we do it here for sake of completeness.
4127 */
4128
4129 /* read in whole auxv vector and copy it to memelfnote */
4130 ptr = lock_user(VERIFY_READ, orig_auxv, len, 0);
4131 if (ptr != NULL) {
4132 fill_note(note, "CORE", NT_AUXV, len, ptr);
4133 unlock_user(ptr, auxv, len);
4134 }
4135 }
4136
4137 /*
4138 * Constructs name of coredump file. We have following convention
4139 * for the name:
4140 * qemu_<basename-of-target-binary>_<date>-<time>_<pid>.core
4141 *
4142 * Returns the filename
4143 */
4144 static char *core_dump_filename(const TaskState *ts)
4145 {
4146 g_autoptr(GDateTime) now = g_date_time_new_now_local();
4147 g_autofree char *nowstr = g_date_time_format(now, "%Y%m%d-%H%M%S");
4148 g_autofree char *base_filename = g_path_get_basename(ts->bprm->filename);
4149
4150 return g_strdup_printf("qemu_%s_%s_%d.core",
4151 base_filename, nowstr, (int)getpid());
4152 }
4153
4154 static int dump_write(int fd, const void *ptr, size_t size)
4155 {
4156 const char *bufp = (const char *)ptr;
4157 ssize_t bytes_written, bytes_left;
4158 struct rlimit dumpsize;
4159 off_t pos;
4160
4161 bytes_written = 0;
4162 getrlimit(RLIMIT_CORE, &dumpsize);
4163 if ((pos = lseek(fd, 0, SEEK_CUR))==-1) {
4164 if (errno == ESPIPE) { /* not a seekable stream */
4165 bytes_left = size;
4166 } else {
4167 return pos;
4168 }
4169 } else {
4170 if (dumpsize.rlim_cur <= pos) {
4171 return -1;
4172 } else if (dumpsize.rlim_cur == RLIM_INFINITY) {
4173 bytes_left = size;
4174 } else {
4175 size_t limit_left=dumpsize.rlim_cur - pos;
4176 bytes_left = limit_left >= size ? size : limit_left ;
4177 }
4178 }
4179
4180 /*
4181 * In normal conditions, single write(2) should do but
4182 * in case of socket etc. this mechanism is more portable.
4183 */
4184 do {
4185 bytes_written = write(fd, bufp, bytes_left);
4186 if (bytes_written < 0) {
4187 if (errno == EINTR)
4188 continue;
4189 return (-1);
4190 } else if (bytes_written == 0) { /* eof */
4191 return (-1);
4192 }
4193 bufp += bytes_written;
4194 bytes_left -= bytes_written;
4195 } while (bytes_left > 0);
4196
4197 return (0);
4198 }
4199
4200 static int write_note(struct memelfnote *men, int fd)
4201 {
4202 struct elf_note en;
4203
4204 en.n_namesz = men->namesz;
4205 en.n_type = men->type;
4206 en.n_descsz = men->datasz;
4207
4208 bswap_note(&en);
4209
4210 if (dump_write(fd, &en, sizeof(en)) != 0)
4211 return (-1);
4212 if (dump_write(fd, men->name, men->namesz_rounded) != 0)
4213 return (-1);
4214 if (dump_write(fd, men->data, men->datasz_rounded) != 0)
4215 return (-1);
4216
4217 return (0);
4218 }
4219
4220 static void fill_thread_info(struct elf_note_info *info, const CPUArchState *env)
4221 {
4222 CPUState *cpu = env_cpu((CPUArchState *)env);
4223 TaskState *ts = (TaskState *)cpu->opaque;
4224 struct elf_thread_status *ets;
4225
4226 ets = g_malloc0(sizeof (*ets));
4227 ets->num_notes = 1; /* only prstatus is dumped */
4228 fill_prstatus(&ets->prstatus, ts, 0);
4229 elf_core_copy_regs(&ets->prstatus.pr_reg, env);
4230 fill_note(&ets->notes[0], "CORE", NT_PRSTATUS, sizeof (ets->prstatus),
4231 &ets->prstatus);
4232
4233 QTAILQ_INSERT_TAIL(&info->thread_list, ets, ets_link);
4234
4235 info->notes_size += note_size(&ets->notes[0]);
4236 }
4237
4238 static void init_note_info(struct elf_note_info *info)
4239 {
4240 /* Initialize the elf_note_info structure so that it is at
4241 * least safe to call free_note_info() on it. Must be
4242 * called before calling fill_note_info().
4243 */
4244 memset(info, 0, sizeof (*info));
4245 QTAILQ_INIT(&info->thread_list);
4246 }
4247
4248 static int fill_note_info(struct elf_note_info *info,
4249 long signr, const CPUArchState *env)
4250 {
4251 #define NUMNOTES 3
4252 CPUState *cpu = env_cpu((CPUArchState *)env);
4253 TaskState *ts = (TaskState *)cpu->opaque;
4254 int i;
4255
4256 info->notes = g_new0(struct memelfnote, NUMNOTES);
4257 if (info->notes == NULL)
4258 return (-ENOMEM);
4259 info->prstatus = g_malloc0(sizeof (*info->prstatus));
4260 if (info->prstatus == NULL)
4261 return (-ENOMEM);
4262 info->psinfo = g_malloc0(sizeof (*info->psinfo));
4263 if (info->prstatus == NULL)
4264 return (-ENOMEM);
4265
4266 /*
4267 * First fill in status (and registers) of current thread
4268 * including process info & aux vector.
4269 */
4270 fill_prstatus(info->prstatus, ts, signr);
4271 elf_core_copy_regs(&info->prstatus->pr_reg, env);
4272 fill_note(&info->notes[0], "CORE", NT_PRSTATUS,
4273 sizeof (*info->prstatus), info->prstatus);
4274 fill_psinfo(info->psinfo, ts);
4275 fill_note(&info->notes[1], "CORE", NT_PRPSINFO,
4276 sizeof (*info->psinfo), info->psinfo);
4277 fill_auxv_note(&info->notes[2], ts);
4278 info->numnote = 3;
4279
4280 info->notes_size = 0;
4281 for (i = 0; i < info->numnote; i++)
4282 info->notes_size += note_size(&info->notes[i]);
4283
4284 /* read and fill status of all threads */
4285 WITH_QEMU_LOCK_GUARD(&qemu_cpu_list_lock) {
4286 CPU_FOREACH(cpu) {
4287 if (cpu == thread_cpu) {
4288 continue;
4289 }
4290 fill_thread_info(info, cpu->env_ptr);
4291 }
4292 }
4293
4294 return (0);
4295 }
4296
4297 static void free_note_info(struct elf_note_info *info)
4298 {
4299 struct elf_thread_status *ets;
4300
4301 while (!QTAILQ_EMPTY(&info->thread_list)) {
4302 ets = QTAILQ_FIRST(&info->thread_list);
4303 QTAILQ_REMOVE(&info->thread_list, ets, ets_link);
4304 g_free(ets);
4305 }
4306
4307 g_free(info->prstatus);
4308 g_free(info->psinfo);
4309 g_free(info->notes);
4310 }
4311
4312 static int write_note_info(struct elf_note_info *info, int fd)
4313 {
4314 struct elf_thread_status *ets;
4315 int i, error = 0;
4316
4317 /* write prstatus, psinfo and auxv for current thread */
4318 for (i = 0; i < info->numnote; i++)
4319 if ((error = write_note(&info->notes[i], fd)) != 0)
4320 return (error);
4321
4322 /* write prstatus for each thread */
4323 QTAILQ_FOREACH(ets, &info->thread_list, ets_link) {
4324 if ((error = write_note(&ets->notes[0], fd)) != 0)
4325 return (error);
4326 }
4327
4328 return (0);
4329 }
4330
4331 /*
4332 * Write out ELF coredump.
4333 *
4334 * See documentation of ELF object file format in:
4335 * http://www.caldera.com/developers/devspecs/gabi41.pdf
4336 *
4337 * Coredump format in linux is following:
4338 *
4339 * 0 +----------------------+ \
4340 * | ELF header | ET_CORE |
4341 * +----------------------+ |
4342 * | ELF program headers | |--- headers
4343 * | - NOTE section | |
4344 * | - PT_LOAD sections | |
4345 * +----------------------+ /
4346 * | NOTEs: |
4347 * | - NT_PRSTATUS |
4348 * | - NT_PRSINFO |
4349 * | - NT_AUXV |
4350 * +----------------------+ <-- aligned to target page
4351 * | Process memory dump |
4352 * : :
4353 * . .
4354 * : :
4355 * | |
4356 * +----------------------+
4357 *
4358 * NT_PRSTATUS -> struct elf_prstatus (per thread)
4359 * NT_PRSINFO -> struct elf_prpsinfo
4360 * NT_AUXV is array of { type, value } pairs (see fill_auxv_note()).
4361 *
4362 * Format follows System V format as close as possible. Current
4363 * version limitations are as follows:
4364 * - no floating point registers are dumped
4365 *
4366 * Function returns 0 in case of success, negative errno otherwise.
4367 *
4368 * TODO: make this work also during runtime: it should be
4369 * possible to force coredump from running process and then
4370 * continue processing. For example qemu could set up SIGUSR2
4371 * handler (provided that target process haven't registered
4372 * handler for that) that does the dump when signal is received.
4373 */
4374 static int elf_core_dump(int signr, const CPUArchState *env)
4375 {
4376 const CPUState *cpu = env_cpu((CPUArchState *)env);
4377 const TaskState *ts = (const TaskState *)cpu->opaque;
4378 struct vm_area_struct *vma = NULL;
4379 g_autofree char *corefile = NULL;
4380 struct elf_note_info info;
4381 struct elfhdr elf;
4382 struct elf_phdr phdr;
4383 struct rlimit dumpsize;
4384 struct mm_struct *mm = NULL;
4385 off_t offset = 0, data_offset = 0;
4386 int segs = 0;
4387 int fd = -1;
4388
4389 init_note_info(&info);
4390
4391 errno = 0;
4392 getrlimit(RLIMIT_CORE, &dumpsize);
4393 if (dumpsize.rlim_cur == 0)
4394 return 0;
4395
4396 corefile = core_dump_filename(ts);
4397
4398 if ((fd = open(corefile, O_WRONLY | O_CREAT,
4399 S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH)) < 0)
4400 return (-errno);
4401
4402 /*
4403 * Walk through target process memory mappings and
4404 * set up structure containing this information. After
4405 * this point vma_xxx functions can be used.
4406 */
4407 if ((mm = vma_init()) == NULL)
4408 goto out;
4409
4410 walk_memory_regions(mm, vma_walker);
4411 segs = vma_get_mapping_count(mm);
4412
4413 /*
4414 * Construct valid coredump ELF header. We also
4415 * add one more segment for notes.
4416 */
4417 fill_elf_header(&elf, segs + 1, ELF_MACHINE, 0);
4418 if (dump_write(fd, &elf, sizeof (elf)) != 0)
4419 goto out;
4420
4421 /* fill in the in-memory version of notes */
4422 if (fill_note_info(&info, signr, env) < 0)
4423 goto out;
4424
4425 offset += sizeof (elf); /* elf header */
4426 offset += (segs + 1) * sizeof (struct elf_phdr); /* program headers */
4427
4428 /* write out notes program header */
4429 fill_elf_note_phdr(&phdr, info.notes_size, offset);
4430
4431 offset += info.notes_size;
4432 if (dump_write(fd, &phdr, sizeof (phdr)) != 0)
4433 goto out;
4434
4435 /*
4436 * ELF specification wants data to start at page boundary so
4437 * we align it here.
4438 */
4439 data_offset = offset = roundup(offset, ELF_EXEC_PAGESIZE);
4440
4441 /*
4442 * Write program headers for memory regions mapped in
4443 * the target process.
4444 */
4445 for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) {
4446 (void) memset(&phdr, 0, sizeof (phdr));
4447
4448 phdr.p_type = PT_LOAD;
4449 phdr.p_offset = offset;
4450 phdr.p_vaddr = vma->vma_start;
4451 phdr.p_paddr = 0;
4452 phdr.p_filesz = vma_dump_size(vma);
4453 offset += phdr.p_filesz;
4454 phdr.p_memsz = vma->vma_end - vma->vma_start;
4455 phdr.p_flags = vma->vma_flags & PROT_READ ? PF_R : 0;
4456 if (vma->vma_flags & PROT_WRITE)
4457 phdr.p_flags |= PF_W;
4458 if (vma->vma_flags & PROT_EXEC)
4459 phdr.p_flags |= PF_X;
4460 phdr.p_align = ELF_EXEC_PAGESIZE;
4461
4462 bswap_phdr(&phdr, 1);
4463 if (dump_write(fd, &phdr, sizeof(phdr)) != 0) {
4464 goto out;
4465 }
4466 }
4467
4468 /*
4469 * Next we write notes just after program headers. No
4470 * alignment needed here.
4471 */
4472 if (write_note_info(&info, fd) < 0)
4473 goto out;
4474
4475 /* align data to page boundary */
4476 if (lseek(fd, data_offset, SEEK_SET) != data_offset)
4477 goto out;
4478
4479 /*
4480 * Finally we can dump process memory into corefile as well.
4481 */
4482 for (vma = vma_first(mm); vma != NULL; vma = vma_next(vma)) {
4483 abi_ulong addr;
4484 abi_ulong end;
4485
4486 end = vma->vma_start + vma_dump_size(vma);
4487
4488 for (addr = vma->vma_start; addr < end;
4489 addr += TARGET_PAGE_SIZE) {
4490 char page[TARGET_PAGE_SIZE];
4491 int error;
4492
4493 /*
4494 * Read in page from target process memory and
4495 * write it to coredump file.
4496 */
4497 error = copy_from_user(page, addr, sizeof (page));
4498 if (error != 0) {
4499 (void) fprintf(stderr, "unable to dump " TARGET_ABI_FMT_lx "\n",
4500 addr);
4501 errno = -error;
4502 goto out;
4503 }
4504 if (dump_write(fd, page, TARGET_PAGE_SIZE) < 0)
4505 goto out;
4506 }
4507 }
4508
4509 out:
4510 free_note_info(&info);
4511 if (mm != NULL)
4512 vma_delete(mm);
4513 (void) close(fd);
4514
4515 if (errno != 0)
4516 return (-errno);
4517 return (0);
4518 }
4519 #endif /* USE_ELF_CORE_DUMP */
4520
4521 void do_init_thread(struct target_pt_regs *regs, struct image_info *infop)
4522 {
4523 init_thread(regs, infop);
4524 }
QEMU mirror