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/prctl.h>
6 #include <sys/resource.h>
7 #include <sys/shm.h>
8
9 #include "qemu.h"
10 #include "user/tswap-target.h"
11 #include "user/page-protection.h"
12 #include "exec/page-protection.h"
13 #include "exec/mmap-lock.h"
14 #include "exec/translation-block.h"
15 #include "exec/tswap.h"
16 #include "user/guest-base.h"
17 #include "user-internals.h"
18 #include "signal-common.h"
19 #include "loader.h"
20 #include "user-mmap.h"
21 #include "disas/disas.h"
22 #include "qemu/bitops.h"
23 #include "qemu/path.h"
24 #include "qemu/queue.h"
25 #include "qemu/guest-random.h"
26 #include "qemu/units.h"
27 #include "qemu/selfmap.h"
28 #include "qemu/lockable.h"
29 #include "qapi/error.h"
30 #include "qemu/error-report.h"
31 #include "target_signal.h"
32 #include "tcg/debuginfo.h"
33
34 #ifdef TARGET_ARM
35 #include "target/arm/cpu-features.h"
36 #endif
37
38 #ifdef _ARCH_PPC64
39 #undef ARCH_DLINFO
40 #undef ELF_PLATFORM
41 #undef ELF_HWCAP
42 #undef ELF_HWCAP2
43 #undef ELF_CLASS
44 #undef ELF_DATA
45 #undef ELF_ARCH
46 #endif
47
48 #ifndef TARGET_ARCH_HAS_SIGTRAMP_PAGE
49 #define TARGET_ARCH_HAS_SIGTRAMP_PAGE 0
50 #endif
51
52 typedef struct {
53 const uint8_t *image;
54 const uint32_t *relocs;
55 unsigned image_size;
56 unsigned reloc_count;
57 unsigned sigreturn_ofs;
58 unsigned rt_sigreturn_ofs;
59 } VdsoImageInfo;
60
61 #define ELF_OSABI ELFOSABI_SYSV
62
63 /* from personality.h */
64
65 /*
66 * Flags for bug emulation.
67 *
68 * These occupy the top three bytes.
69 */
70 enum {
71 ADDR_NO_RANDOMIZE = 0x0040000, /* disable randomization of VA space */
72 FDPIC_FUNCPTRS = 0x0080000, /* userspace function ptrs point to
73 descriptors (signal handling) */
74 MMAP_PAGE_ZERO = 0x0100000,
75 ADDR_COMPAT_LAYOUT = 0x0200000,
76 READ_IMPLIES_EXEC = 0x0400000,
77 ADDR_LIMIT_32BIT = 0x0800000,
78 SHORT_INODE = 0x1000000,
79 WHOLE_SECONDS = 0x2000000,
80 STICKY_TIMEOUTS = 0x4000000,
81 ADDR_LIMIT_3GB = 0x8000000,
82 };
83
84 /*
85 * Personality types.
86 *
87 * These go in the low byte. Avoid using the top bit, it will
88 * conflict with error returns.
89 */
90 enum {
91 PER_LINUX = 0x0000,
92 PER_LINUX_32BIT = 0x0000 | ADDR_LIMIT_32BIT,
93 PER_LINUX_FDPIC = 0x0000 | FDPIC_FUNCPTRS,
94 PER_SVR4 = 0x0001 | STICKY_TIMEOUTS | MMAP_PAGE_ZERO,
95 PER_SVR3 = 0x0002 | STICKY_TIMEOUTS | SHORT_INODE,
96 PER_SCOSVR3 = 0x0003 | STICKY_TIMEOUTS | WHOLE_SECONDS | SHORT_INODE,
97 PER_OSR5 = 0x0003 | STICKY_TIMEOUTS | WHOLE_SECONDS,
98 PER_WYSEV386 = 0x0004 | STICKY_TIMEOUTS | SHORT_INODE,
99 PER_ISCR4 = 0x0005 | STICKY_TIMEOUTS,
100 PER_BSD = 0x0006,
101 PER_SUNOS = 0x0006 | STICKY_TIMEOUTS,
102 PER_XENIX = 0x0007 | STICKY_TIMEOUTS | SHORT_INODE,
103 PER_LINUX32 = 0x0008,
104 PER_LINUX32_3GB = 0x0008 | ADDR_LIMIT_3GB,
105 PER_IRIX32 = 0x0009 | STICKY_TIMEOUTS,/* IRIX5 32-bit */
106 PER_IRIXN32 = 0x000a | STICKY_TIMEOUTS,/* IRIX6 new 32-bit */
107 PER_IRIX64 = 0x000b | STICKY_TIMEOUTS,/* IRIX6 64-bit */
108 PER_RISCOS = 0x000c,
109 PER_SOLARIS = 0x000d | STICKY_TIMEOUTS,
110 PER_UW7 = 0x000e | STICKY_TIMEOUTS | MMAP_PAGE_ZERO,
111 PER_OSF4 = 0x000f, /* OSF/1 v4 */
112 PER_HPUX = 0x0010,
113 PER_MASK = 0x00ff,
114 };
115
116 /*
117 * Return the base personality without flags.
118 */
119 #define personality(pers) (pers & PER_MASK)
120
info_is_fdpic(struct image_info * info)121 int info_is_fdpic(struct image_info *info)
122 {
123 return info->personality == PER_LINUX_FDPIC;
124 }
125
126 /* this flag is uneffective under linux too, should be deleted */
127 #ifndef MAP_DENYWRITE
128 #define MAP_DENYWRITE 0
129 #endif
130
131 /* should probably go in elf.h */
132 #ifndef ELIBBAD
133 #define ELIBBAD 80
134 #endif
135
136 #if TARGET_BIG_ENDIAN
137 #define ELF_DATA ELFDATA2MSB
138 #else
139 #define ELF_DATA ELFDATA2LSB
140 #endif
141
142 #ifdef TARGET_ABI_MIPSN32
143 typedef abi_ullong target_elf_greg_t;
144 #define tswapreg(ptr) tswap64(ptr)
145 #else
146 typedef abi_ulong target_elf_greg_t;
147 #define tswapreg(ptr) tswapal(ptr)
148 #endif
149
150 #ifdef USE_UID16
151 typedef abi_ushort target_uid_t;
152 typedef abi_ushort target_gid_t;
153 #else
154 typedef abi_uint target_uid_t;
155 typedef abi_uint target_gid_t;
156 #endif
157 typedef abi_int target_pid_t;
158
159 #ifdef TARGET_I386
160
161 #define ELF_HWCAP get_elf_hwcap()
162
get_elf_hwcap(void)163 static uint32_t get_elf_hwcap(void)
164 {
165 X86CPU *cpu = X86_CPU(thread_cpu);
166
167 return cpu->env.features[FEAT_1_EDX];
168 }
169
170 #ifdef TARGET_X86_64
171 #define ELF_CLASS ELFCLASS64
172 #define ELF_ARCH EM_X86_64
173
174 #define ELF_PLATFORM "x86_64"
175
init_thread(struct target_pt_regs * regs,struct image_info * infop)176 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop)
177 {
178 regs->rax = 0;
179 regs->rsp = infop->start_stack;
180 regs->rip = infop->entry;
181 }
182
183 #define ELF_NREG 27
184 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
185
186 /*
187 * Note that ELF_NREG should be 29 as there should be place for
188 * TRAPNO and ERR "registers" as well but linux doesn't dump
189 * those.
190 *
191 * See linux kernel: arch/x86/include/asm/elf.h
192 */
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUX86State * env)193 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUX86State *env)
194 {
195 (*regs)[0] = tswapreg(env->regs[15]);
196 (*regs)[1] = tswapreg(env->regs[14]);
197 (*regs)[2] = tswapreg(env->regs[13]);
198 (*regs)[3] = tswapreg(env->regs[12]);
199 (*regs)[4] = tswapreg(env->regs[R_EBP]);
200 (*regs)[5] = tswapreg(env->regs[R_EBX]);
201 (*regs)[6] = tswapreg(env->regs[11]);
202 (*regs)[7] = tswapreg(env->regs[10]);
203 (*regs)[8] = tswapreg(env->regs[9]);
204 (*regs)[9] = tswapreg(env->regs[8]);
205 (*regs)[10] = tswapreg(env->regs[R_EAX]);
206 (*regs)[11] = tswapreg(env->regs[R_ECX]);
207 (*regs)[12] = tswapreg(env->regs[R_EDX]);
208 (*regs)[13] = tswapreg(env->regs[R_ESI]);
209 (*regs)[14] = tswapreg(env->regs[R_EDI]);
210 (*regs)[15] = tswapreg(get_task_state(env_cpu_const(env))->orig_ax);
211 (*regs)[16] = tswapreg(env->eip);
212 (*regs)[17] = tswapreg(env->segs[R_CS].selector & 0xffff);
213 (*regs)[18] = tswapreg(env->eflags);
214 (*regs)[19] = tswapreg(env->regs[R_ESP]);
215 (*regs)[20] = tswapreg(env->segs[R_SS].selector & 0xffff);
216 (*regs)[21] = tswapreg(env->segs[R_FS].selector & 0xffff);
217 (*regs)[22] = tswapreg(env->segs[R_GS].selector & 0xffff);
218 (*regs)[23] = tswapreg(env->segs[R_DS].selector & 0xffff);
219 (*regs)[24] = tswapreg(env->segs[R_ES].selector & 0xffff);
220 (*regs)[25] = tswapreg(env->segs[R_FS].selector & 0xffff);
221 (*regs)[26] = tswapreg(env->segs[R_GS].selector & 0xffff);
222 }
223
224 #if ULONG_MAX > UINT32_MAX
225 #define INIT_GUEST_COMMPAGE
init_guest_commpage(void)226 static bool init_guest_commpage(void)
227 {
228 /*
229 * The vsyscall page is at a high negative address aka kernel space,
230 * which means that we cannot actually allocate it with target_mmap.
231 * We still should be able to use page_set_flags, unless the user
232 * has specified -R reserved_va, which would trigger an assert().
233 */
234 if (reserved_va != 0 &&
235 TARGET_VSYSCALL_PAGE + TARGET_PAGE_SIZE - 1 > reserved_va) {
236 error_report("Cannot allocate vsyscall page");
237 exit(EXIT_FAILURE);
238 }
239 page_set_flags(TARGET_VSYSCALL_PAGE,
240 TARGET_VSYSCALL_PAGE | ~TARGET_PAGE_MASK,
241 PAGE_EXEC | PAGE_VALID);
242 return true;
243 }
244 #endif
245 #else
246
247 /*
248 * This is used to ensure we don't load something for the wrong architecture.
249 */
250 #define elf_check_arch(x) ( ((x) == EM_386) || ((x) == EM_486) )
251
252 /*
253 * These are used to set parameters in the core dumps.
254 */
255 #define ELF_CLASS ELFCLASS32
256 #define ELF_ARCH EM_386
257
258 #define ELF_PLATFORM get_elf_platform()
259 #define EXSTACK_DEFAULT true
260
get_elf_platform(void)261 static const char *get_elf_platform(void)
262 {
263 static char elf_platform[] = "i386";
264 int family = object_property_get_int(OBJECT(thread_cpu), "family", NULL);
265 if (family > 6) {
266 family = 6;
267 }
268 if (family >= 3) {
269 elf_platform[1] = '0' + family;
270 }
271 return elf_platform;
272 }
273
init_thread(struct target_pt_regs * regs,struct image_info * infop)274 static inline void init_thread(struct target_pt_regs *regs,
275 struct image_info *infop)
276 {
277 regs->esp = infop->start_stack;
278 regs->eip = infop->entry;
279
280 /* SVR4/i386 ABI (pages 3-31, 3-32) says that when the program
281 starts %edx contains a pointer to a function which might be
282 registered using `atexit'. This provides a mean for the
283 dynamic linker to call DT_FINI functions for shared libraries
284 that have been loaded before the code runs.
285
286 A value of 0 tells we have no such handler. */
287 regs->edx = 0;
288 }
289
290 #define ELF_NREG 17
291 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
292
293 /*
294 * Note that ELF_NREG should be 19 as there should be place for
295 * TRAPNO and ERR "registers" as well but linux doesn't dump
296 * those.
297 *
298 * See linux kernel: arch/x86/include/asm/elf.h
299 */
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUX86State * env)300 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUX86State *env)
301 {
302 (*regs)[0] = tswapreg(env->regs[R_EBX]);
303 (*regs)[1] = tswapreg(env->regs[R_ECX]);
304 (*regs)[2] = tswapreg(env->regs[R_EDX]);
305 (*regs)[3] = tswapreg(env->regs[R_ESI]);
306 (*regs)[4] = tswapreg(env->regs[R_EDI]);
307 (*regs)[5] = tswapreg(env->regs[R_EBP]);
308 (*regs)[6] = tswapreg(env->regs[R_EAX]);
309 (*regs)[7] = tswapreg(env->segs[R_DS].selector & 0xffff);
310 (*regs)[8] = tswapreg(env->segs[R_ES].selector & 0xffff);
311 (*regs)[9] = tswapreg(env->segs[R_FS].selector & 0xffff);
312 (*regs)[10] = tswapreg(env->segs[R_GS].selector & 0xffff);
313 (*regs)[11] = tswapreg(get_task_state(env_cpu_const(env))->orig_ax);
314 (*regs)[12] = tswapreg(env->eip);
315 (*regs)[13] = tswapreg(env->segs[R_CS].selector & 0xffff);
316 (*regs)[14] = tswapreg(env->eflags);
317 (*regs)[15] = tswapreg(env->regs[R_ESP]);
318 (*regs)[16] = tswapreg(env->segs[R_SS].selector & 0xffff);
319 }
320
321 /*
322 * i386 is the only target which supplies AT_SYSINFO for the vdso.
323 * All others only supply AT_SYSINFO_EHDR.
324 */
325 #define DLINFO_ARCH_ITEMS (vdso_info != NULL)
326 #define ARCH_DLINFO \
327 do { \
328 if (vdso_info) { \
329 NEW_AUX_ENT(AT_SYSINFO, vdso_info->entry); \
330 } \
331 } while (0)
332
333 #endif /* TARGET_X86_64 */
334
335 #define VDSO_HEADER "vdso.c.inc"
336
337 #define USE_ELF_CORE_DUMP
338 #define ELF_EXEC_PAGESIZE 4096
339
340 #endif /* TARGET_I386 */
341
342 #ifdef TARGET_ARM
343
344 #ifndef TARGET_AARCH64
345 /* 32 bit ARM definitions */
346
347 #define ELF_ARCH EM_ARM
348 #define ELF_CLASS ELFCLASS32
349 #define EXSTACK_DEFAULT true
350
init_thread(struct target_pt_regs * regs,struct image_info * infop)351 static inline void init_thread(struct target_pt_regs *regs,
352 struct image_info *infop)
353 {
354 abi_long stack = infop->start_stack;
355 memset(regs, 0, sizeof(*regs));
356
357 regs->uregs[16] = ARM_CPU_MODE_USR;
358 if (infop->entry & 1) {
359 regs->uregs[16] |= CPSR_T;
360 }
361 regs->uregs[15] = infop->entry & 0xfffffffe;
362 regs->uregs[13] = infop->start_stack;
363 /* FIXME - what to for failure of get_user()? */
364 get_user_ual(regs->uregs[2], stack + 8); /* envp */
365 get_user_ual(regs->uregs[1], stack + 4); /* envp */
366 /* XXX: it seems that r0 is zeroed after ! */
367 regs->uregs[0] = 0;
368 /* For uClinux PIC binaries. */
369 /* XXX: Linux does this only on ARM with no MMU (do we care ?) */
370 regs->uregs[10] = infop->start_data;
371
372 /* Support ARM FDPIC. */
373 if (info_is_fdpic(infop)) {
374 /* As described in the ABI document, r7 points to the loadmap info
375 * prepared by the kernel. If an interpreter is needed, r8 points
376 * to the interpreter loadmap and r9 points to the interpreter
377 * PT_DYNAMIC info. If no interpreter is needed, r8 is zero, and
378 * r9 points to the main program PT_DYNAMIC info.
379 */
380 regs->uregs[7] = infop->loadmap_addr;
381 if (infop->interpreter_loadmap_addr) {
382 /* Executable is dynamically loaded. */
383 regs->uregs[8] = infop->interpreter_loadmap_addr;
384 regs->uregs[9] = infop->interpreter_pt_dynamic_addr;
385 } else {
386 regs->uregs[8] = 0;
387 regs->uregs[9] = infop->pt_dynamic_addr;
388 }
389 }
390 }
391
392 #define ELF_NREG 18
393 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
394
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUARMState * env)395 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUARMState *env)
396 {
397 (*regs)[0] = tswapreg(env->regs[0]);
398 (*regs)[1] = tswapreg(env->regs[1]);
399 (*regs)[2] = tswapreg(env->regs[2]);
400 (*regs)[3] = tswapreg(env->regs[3]);
401 (*regs)[4] = tswapreg(env->regs[4]);
402 (*regs)[5] = tswapreg(env->regs[5]);
403 (*regs)[6] = tswapreg(env->regs[6]);
404 (*regs)[7] = tswapreg(env->regs[7]);
405 (*regs)[8] = tswapreg(env->regs[8]);
406 (*regs)[9] = tswapreg(env->regs[9]);
407 (*regs)[10] = tswapreg(env->regs[10]);
408 (*regs)[11] = tswapreg(env->regs[11]);
409 (*regs)[12] = tswapreg(env->regs[12]);
410 (*regs)[13] = tswapreg(env->regs[13]);
411 (*regs)[14] = tswapreg(env->regs[14]);
412 (*regs)[15] = tswapreg(env->regs[15]);
413
414 (*regs)[16] = tswapreg(cpsr_read((CPUARMState *)env));
415 (*regs)[17] = tswapreg(env->regs[0]); /* XXX */
416 }
417
418 #define USE_ELF_CORE_DUMP
419 #define ELF_EXEC_PAGESIZE 4096
420
421 enum
422 {
423 ARM_HWCAP_ARM_SWP = 1 << 0,
424 ARM_HWCAP_ARM_HALF = 1 << 1,
425 ARM_HWCAP_ARM_THUMB = 1 << 2,
426 ARM_HWCAP_ARM_26BIT = 1 << 3,
427 ARM_HWCAP_ARM_FAST_MULT = 1 << 4,
428 ARM_HWCAP_ARM_FPA = 1 << 5,
429 ARM_HWCAP_ARM_VFP = 1 << 6,
430 ARM_HWCAP_ARM_EDSP = 1 << 7,
431 ARM_HWCAP_ARM_JAVA = 1 << 8,
432 ARM_HWCAP_ARM_IWMMXT = 1 << 9,
433 ARM_HWCAP_ARM_CRUNCH = 1 << 10,
434 ARM_HWCAP_ARM_THUMBEE = 1 << 11,
435 ARM_HWCAP_ARM_NEON = 1 << 12,
436 ARM_HWCAP_ARM_VFPv3 = 1 << 13,
437 ARM_HWCAP_ARM_VFPv3D16 = 1 << 14,
438 ARM_HWCAP_ARM_TLS = 1 << 15,
439 ARM_HWCAP_ARM_VFPv4 = 1 << 16,
440 ARM_HWCAP_ARM_IDIVA = 1 << 17,
441 ARM_HWCAP_ARM_IDIVT = 1 << 18,
442 ARM_HWCAP_ARM_VFPD32 = 1 << 19,
443 ARM_HWCAP_ARM_LPAE = 1 << 20,
444 ARM_HWCAP_ARM_EVTSTRM = 1 << 21,
445 ARM_HWCAP_ARM_FPHP = 1 << 22,
446 ARM_HWCAP_ARM_ASIMDHP = 1 << 23,
447 ARM_HWCAP_ARM_ASIMDDP = 1 << 24,
448 ARM_HWCAP_ARM_ASIMDFHM = 1 << 25,
449 ARM_HWCAP_ARM_ASIMDBF16 = 1 << 26,
450 ARM_HWCAP_ARM_I8MM = 1 << 27,
451 };
452
453 enum {
454 ARM_HWCAP2_ARM_AES = 1 << 0,
455 ARM_HWCAP2_ARM_PMULL = 1 << 1,
456 ARM_HWCAP2_ARM_SHA1 = 1 << 2,
457 ARM_HWCAP2_ARM_SHA2 = 1 << 3,
458 ARM_HWCAP2_ARM_CRC32 = 1 << 4,
459 ARM_HWCAP2_ARM_SB = 1 << 5,
460 ARM_HWCAP2_ARM_SSBS = 1 << 6,
461 };
462
463 /* The commpage only exists for 32 bit kernels */
464
465 #define HI_COMMPAGE (intptr_t)0xffff0f00u
466
init_guest_commpage(void)467 static bool init_guest_commpage(void)
468 {
469 ARMCPU *cpu = ARM_CPU(thread_cpu);
470 int host_page_size = qemu_real_host_page_size();
471 abi_ptr commpage;
472 void *want;
473 void *addr;
474
475 /*
476 * M-profile allocates maximum of 2GB address space, so can never
477 * allocate the commpage. Skip it.
478 */
479 if (arm_feature(&cpu->env, ARM_FEATURE_M)) {
480 return true;
481 }
482
483 commpage = HI_COMMPAGE & -host_page_size;
484 want = g2h_untagged(commpage);
485 addr = mmap(want, host_page_size, PROT_READ | PROT_WRITE,
486 MAP_ANONYMOUS | MAP_PRIVATE |
487 (commpage < reserved_va ? MAP_FIXED : MAP_FIXED_NOREPLACE),
488 -1, 0);
489
490 if (addr == MAP_FAILED) {
491 perror("Allocating guest commpage");
492 exit(EXIT_FAILURE);
493 }
494 if (addr != want) {
495 return false;
496 }
497
498 /* Set kernel helper versions; rest of page is 0. */
499 __put_user(5, (uint32_t *)g2h_untagged(0xffff0ffcu));
500
501 if (mprotect(addr, host_page_size, PROT_READ)) {
502 perror("Protecting guest commpage");
503 exit(EXIT_FAILURE);
504 }
505
506 page_set_flags(commpage, commpage | (host_page_size - 1),
507 PAGE_READ | PAGE_EXEC | PAGE_VALID);
508 return true;
509 }
510
511 #define ELF_HWCAP get_elf_hwcap()
512 #define ELF_HWCAP2 get_elf_hwcap2()
513
get_elf_hwcap(void)514 uint32_t get_elf_hwcap(void)
515 {
516 ARMCPU *cpu = ARM_CPU(thread_cpu);
517 uint32_t hwcaps = 0;
518
519 hwcaps |= ARM_HWCAP_ARM_SWP;
520 hwcaps |= ARM_HWCAP_ARM_HALF;
521 hwcaps |= ARM_HWCAP_ARM_THUMB;
522 hwcaps |= ARM_HWCAP_ARM_FAST_MULT;
523
524 /* probe for the extra features */
525 #define GET_FEATURE(feat, hwcap) \
526 do { if (arm_feature(&cpu->env, feat)) { hwcaps |= hwcap; } } while (0)
527
528 #define GET_FEATURE_ID(feat, hwcap) \
529 do { if (cpu_isar_feature(feat, cpu)) { hwcaps |= hwcap; } } while (0)
530
531 /* EDSP is in v5TE and above, but all our v5 CPUs are v5TE */
532 GET_FEATURE(ARM_FEATURE_V5, ARM_HWCAP_ARM_EDSP);
533 GET_FEATURE(ARM_FEATURE_IWMMXT, ARM_HWCAP_ARM_IWMMXT);
534 GET_FEATURE(ARM_FEATURE_THUMB2EE, ARM_HWCAP_ARM_THUMBEE);
535 GET_FEATURE(ARM_FEATURE_NEON, ARM_HWCAP_ARM_NEON);
536 GET_FEATURE(ARM_FEATURE_V6K, ARM_HWCAP_ARM_TLS);
537 GET_FEATURE(ARM_FEATURE_LPAE, ARM_HWCAP_ARM_LPAE);
538 GET_FEATURE_ID(aa32_arm_div, ARM_HWCAP_ARM_IDIVA);
539 GET_FEATURE_ID(aa32_thumb_div, ARM_HWCAP_ARM_IDIVT);
540 GET_FEATURE_ID(aa32_vfp, ARM_HWCAP_ARM_VFP);
541
542 if (cpu_isar_feature(aa32_fpsp_v3, cpu) ||
543 cpu_isar_feature(aa32_fpdp_v3, cpu)) {
544 hwcaps |= ARM_HWCAP_ARM_VFPv3;
545 if (cpu_isar_feature(aa32_simd_r32, cpu)) {
546 hwcaps |= ARM_HWCAP_ARM_VFPD32;
547 } else {
548 hwcaps |= ARM_HWCAP_ARM_VFPv3D16;
549 }
550 }
551 GET_FEATURE_ID(aa32_simdfmac, ARM_HWCAP_ARM_VFPv4);
552 /*
553 * MVFR1.FPHP and .SIMDHP must be in sync, and QEMU uses the same
554 * isar_feature function for both. The kernel reports them as two hwcaps.
555 */
556 GET_FEATURE_ID(aa32_fp16_arith, ARM_HWCAP_ARM_FPHP);
557 GET_FEATURE_ID(aa32_fp16_arith, ARM_HWCAP_ARM_ASIMDHP);
558 GET_FEATURE_ID(aa32_dp, ARM_HWCAP_ARM_ASIMDDP);
559 GET_FEATURE_ID(aa32_fhm, ARM_HWCAP_ARM_ASIMDFHM);
560 GET_FEATURE_ID(aa32_bf16, ARM_HWCAP_ARM_ASIMDBF16);
561 GET_FEATURE_ID(aa32_i8mm, ARM_HWCAP_ARM_I8MM);
562
563 return hwcaps;
564 }
565
get_elf_hwcap2(void)566 uint64_t get_elf_hwcap2(void)
567 {
568 ARMCPU *cpu = ARM_CPU(thread_cpu);
569 uint64_t hwcaps = 0;
570
571 GET_FEATURE_ID(aa32_aes, ARM_HWCAP2_ARM_AES);
572 GET_FEATURE_ID(aa32_pmull, ARM_HWCAP2_ARM_PMULL);
573 GET_FEATURE_ID(aa32_sha1, ARM_HWCAP2_ARM_SHA1);
574 GET_FEATURE_ID(aa32_sha2, ARM_HWCAP2_ARM_SHA2);
575 GET_FEATURE_ID(aa32_crc32, ARM_HWCAP2_ARM_CRC32);
576 GET_FEATURE_ID(aa32_sb, ARM_HWCAP2_ARM_SB);
577 GET_FEATURE_ID(aa32_ssbs, ARM_HWCAP2_ARM_SSBS);
578 return hwcaps;
579 }
580
elf_hwcap_str(uint32_t bit)581 const char *elf_hwcap_str(uint32_t bit)
582 {
583 static const char *hwcap_str[] = {
584 [__builtin_ctz(ARM_HWCAP_ARM_SWP )] = "swp",
585 [__builtin_ctz(ARM_HWCAP_ARM_HALF )] = "half",
586 [__builtin_ctz(ARM_HWCAP_ARM_THUMB )] = "thumb",
587 [__builtin_ctz(ARM_HWCAP_ARM_26BIT )] = "26bit",
588 [__builtin_ctz(ARM_HWCAP_ARM_FAST_MULT)] = "fast_mult",
589 [__builtin_ctz(ARM_HWCAP_ARM_FPA )] = "fpa",
590 [__builtin_ctz(ARM_HWCAP_ARM_VFP )] = "vfp",
591 [__builtin_ctz(ARM_HWCAP_ARM_EDSP )] = "edsp",
592 [__builtin_ctz(ARM_HWCAP_ARM_JAVA )] = "java",
593 [__builtin_ctz(ARM_HWCAP_ARM_IWMMXT )] = "iwmmxt",
594 [__builtin_ctz(ARM_HWCAP_ARM_CRUNCH )] = "crunch",
595 [__builtin_ctz(ARM_HWCAP_ARM_THUMBEE )] = "thumbee",
596 [__builtin_ctz(ARM_HWCAP_ARM_NEON )] = "neon",
597 [__builtin_ctz(ARM_HWCAP_ARM_VFPv3 )] = "vfpv3",
598 [__builtin_ctz(ARM_HWCAP_ARM_VFPv3D16 )] = "vfpv3d16",
599 [__builtin_ctz(ARM_HWCAP_ARM_TLS )] = "tls",
600 [__builtin_ctz(ARM_HWCAP_ARM_VFPv4 )] = "vfpv4",
601 [__builtin_ctz(ARM_HWCAP_ARM_IDIVA )] = "idiva",
602 [__builtin_ctz(ARM_HWCAP_ARM_IDIVT )] = "idivt",
603 [__builtin_ctz(ARM_HWCAP_ARM_VFPD32 )] = "vfpd32",
604 [__builtin_ctz(ARM_HWCAP_ARM_LPAE )] = "lpae",
605 [__builtin_ctz(ARM_HWCAP_ARM_EVTSTRM )] = "evtstrm",
606 [__builtin_ctz(ARM_HWCAP_ARM_FPHP )] = "fphp",
607 [__builtin_ctz(ARM_HWCAP_ARM_ASIMDHP )] = "asimdhp",
608 [__builtin_ctz(ARM_HWCAP_ARM_ASIMDDP )] = "asimddp",
609 [__builtin_ctz(ARM_HWCAP_ARM_ASIMDFHM )] = "asimdfhm",
610 [__builtin_ctz(ARM_HWCAP_ARM_ASIMDBF16)] = "asimdbf16",
611 [__builtin_ctz(ARM_HWCAP_ARM_I8MM )] = "i8mm",
612 };
613
614 return bit < ARRAY_SIZE(hwcap_str) ? hwcap_str[bit] : NULL;
615 }
616
elf_hwcap2_str(uint32_t bit)617 const char *elf_hwcap2_str(uint32_t bit)
618 {
619 static const char *hwcap_str[] = {
620 [__builtin_ctz(ARM_HWCAP2_ARM_AES )] = "aes",
621 [__builtin_ctz(ARM_HWCAP2_ARM_PMULL)] = "pmull",
622 [__builtin_ctz(ARM_HWCAP2_ARM_SHA1 )] = "sha1",
623 [__builtin_ctz(ARM_HWCAP2_ARM_SHA2 )] = "sha2",
624 [__builtin_ctz(ARM_HWCAP2_ARM_CRC32)] = "crc32",
625 [__builtin_ctz(ARM_HWCAP2_ARM_SB )] = "sb",
626 [__builtin_ctz(ARM_HWCAP2_ARM_SSBS )] = "ssbs",
627 };
628
629 return bit < ARRAY_SIZE(hwcap_str) ? hwcap_str[bit] : NULL;
630 }
631
632 #undef GET_FEATURE
633 #undef GET_FEATURE_ID
634
635 #define ELF_PLATFORM get_elf_platform()
636
get_elf_platform(void)637 static const char *get_elf_platform(void)
638 {
639 CPUARMState *env = cpu_env(thread_cpu);
640
641 #if TARGET_BIG_ENDIAN
642 # define END "b"
643 #else
644 # define END "l"
645 #endif
646
647 if (arm_feature(env, ARM_FEATURE_V8)) {
648 return "v8" END;
649 } else if (arm_feature(env, ARM_FEATURE_V7)) {
650 if (arm_feature(env, ARM_FEATURE_M)) {
651 return "v7m" END;
652 } else {
653 return "v7" END;
654 }
655 } else if (arm_feature(env, ARM_FEATURE_V6)) {
656 return "v6" END;
657 } else if (arm_feature(env, ARM_FEATURE_V5)) {
658 return "v5" END;
659 } else {
660 return "v4" END;
661 }
662
663 #undef END
664 }
665
666 #if TARGET_BIG_ENDIAN
667 #include "elf.h"
668 #include "vdso-be8.c.inc"
669 #include "vdso-be32.c.inc"
670
vdso_image_info(uint32_t elf_flags)671 static const VdsoImageInfo *vdso_image_info(uint32_t elf_flags)
672 {
673 return (EF_ARM_EABI_VERSION(elf_flags) >= EF_ARM_EABI_VER4
674 && (elf_flags & EF_ARM_BE8)
675 ? &vdso_be8_image_info
676 : &vdso_be32_image_info);
677 }
678 #define vdso_image_info vdso_image_info
679 #else
680 # define VDSO_HEADER "vdso-le.c.inc"
681 #endif
682
683 #else
684 /* 64 bit ARM definitions */
685
686 #define ELF_ARCH EM_AARCH64
687 #define ELF_CLASS ELFCLASS64
688 #if TARGET_BIG_ENDIAN
689 # define ELF_PLATFORM "aarch64_be"
690 #else
691 # define ELF_PLATFORM "aarch64"
692 #endif
693
init_thread(struct target_pt_regs * regs,struct image_info * infop)694 static inline void init_thread(struct target_pt_regs *regs,
695 struct image_info *infop)
696 {
697 abi_long stack = infop->start_stack;
698 memset(regs, 0, sizeof(*regs));
699
700 regs->pc = infop->entry & ~0x3ULL;
701 regs->sp = stack;
702 }
703
704 #define ELF_NREG 34
705 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
706
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUARMState * env)707 static void elf_core_copy_regs(target_elf_gregset_t *regs,
708 const CPUARMState *env)
709 {
710 int i;
711
712 for (i = 0; i < 32; i++) {
713 (*regs)[i] = tswapreg(env->xregs[i]);
714 }
715 (*regs)[32] = tswapreg(env->pc);
716 (*regs)[33] = tswapreg(pstate_read((CPUARMState *)env));
717 }
718
719 #define USE_ELF_CORE_DUMP
720 #define ELF_EXEC_PAGESIZE 4096
721
722 enum {
723 ARM_HWCAP_A64_FP = 1 << 0,
724 ARM_HWCAP_A64_ASIMD = 1 << 1,
725 ARM_HWCAP_A64_EVTSTRM = 1 << 2,
726 ARM_HWCAP_A64_AES = 1 << 3,
727 ARM_HWCAP_A64_PMULL = 1 << 4,
728 ARM_HWCAP_A64_SHA1 = 1 << 5,
729 ARM_HWCAP_A64_SHA2 = 1 << 6,
730 ARM_HWCAP_A64_CRC32 = 1 << 7,
731 ARM_HWCAP_A64_ATOMICS = 1 << 8,
732 ARM_HWCAP_A64_FPHP = 1 << 9,
733 ARM_HWCAP_A64_ASIMDHP = 1 << 10,
734 ARM_HWCAP_A64_CPUID = 1 << 11,
735 ARM_HWCAP_A64_ASIMDRDM = 1 << 12,
736 ARM_HWCAP_A64_JSCVT = 1 << 13,
737 ARM_HWCAP_A64_FCMA = 1 << 14,
738 ARM_HWCAP_A64_LRCPC = 1 << 15,
739 ARM_HWCAP_A64_DCPOP = 1 << 16,
740 ARM_HWCAP_A64_SHA3 = 1 << 17,
741 ARM_HWCAP_A64_SM3 = 1 << 18,
742 ARM_HWCAP_A64_SM4 = 1 << 19,
743 ARM_HWCAP_A64_ASIMDDP = 1 << 20,
744 ARM_HWCAP_A64_SHA512 = 1 << 21,
745 ARM_HWCAP_A64_SVE = 1 << 22,
746 ARM_HWCAP_A64_ASIMDFHM = 1 << 23,
747 ARM_HWCAP_A64_DIT = 1 << 24,
748 ARM_HWCAP_A64_USCAT = 1 << 25,
749 ARM_HWCAP_A64_ILRCPC = 1 << 26,
750 ARM_HWCAP_A64_FLAGM = 1 << 27,
751 ARM_HWCAP_A64_SSBS = 1 << 28,
752 ARM_HWCAP_A64_SB = 1 << 29,
753 ARM_HWCAP_A64_PACA = 1 << 30,
754 ARM_HWCAP_A64_PACG = 1UL << 31,
755
756 ARM_HWCAP2_A64_DCPODP = 1 << 0,
757 ARM_HWCAP2_A64_SVE2 = 1 << 1,
758 ARM_HWCAP2_A64_SVEAES = 1 << 2,
759 ARM_HWCAP2_A64_SVEPMULL = 1 << 3,
760 ARM_HWCAP2_A64_SVEBITPERM = 1 << 4,
761 ARM_HWCAP2_A64_SVESHA3 = 1 << 5,
762 ARM_HWCAP2_A64_SVESM4 = 1 << 6,
763 ARM_HWCAP2_A64_FLAGM2 = 1 << 7,
764 ARM_HWCAP2_A64_FRINT = 1 << 8,
765 ARM_HWCAP2_A64_SVEI8MM = 1 << 9,
766 ARM_HWCAP2_A64_SVEF32MM = 1 << 10,
767 ARM_HWCAP2_A64_SVEF64MM = 1 << 11,
768 ARM_HWCAP2_A64_SVEBF16 = 1 << 12,
769 ARM_HWCAP2_A64_I8MM = 1 << 13,
770 ARM_HWCAP2_A64_BF16 = 1 << 14,
771 ARM_HWCAP2_A64_DGH = 1 << 15,
772 ARM_HWCAP2_A64_RNG = 1 << 16,
773 ARM_HWCAP2_A64_BTI = 1 << 17,
774 ARM_HWCAP2_A64_MTE = 1 << 18,
775 ARM_HWCAP2_A64_ECV = 1 << 19,
776 ARM_HWCAP2_A64_AFP = 1 << 20,
777 ARM_HWCAP2_A64_RPRES = 1 << 21,
778 ARM_HWCAP2_A64_MTE3 = 1 << 22,
779 ARM_HWCAP2_A64_SME = 1 << 23,
780 ARM_HWCAP2_A64_SME_I16I64 = 1 << 24,
781 ARM_HWCAP2_A64_SME_F64F64 = 1 << 25,
782 ARM_HWCAP2_A64_SME_I8I32 = 1 << 26,
783 ARM_HWCAP2_A64_SME_F16F32 = 1 << 27,
784 ARM_HWCAP2_A64_SME_B16F32 = 1 << 28,
785 ARM_HWCAP2_A64_SME_F32F32 = 1 << 29,
786 ARM_HWCAP2_A64_SME_FA64 = 1 << 30,
787 ARM_HWCAP2_A64_WFXT = 1ULL << 31,
788 ARM_HWCAP2_A64_EBF16 = 1ULL << 32,
789 ARM_HWCAP2_A64_SVE_EBF16 = 1ULL << 33,
790 ARM_HWCAP2_A64_CSSC = 1ULL << 34,
791 ARM_HWCAP2_A64_RPRFM = 1ULL << 35,
792 ARM_HWCAP2_A64_SVE2P1 = 1ULL << 36,
793 ARM_HWCAP2_A64_SME2 = 1ULL << 37,
794 ARM_HWCAP2_A64_SME2P1 = 1ULL << 38,
795 ARM_HWCAP2_A64_SME_I16I32 = 1ULL << 39,
796 ARM_HWCAP2_A64_SME_BI32I32 = 1ULL << 40,
797 ARM_HWCAP2_A64_SME_B16B16 = 1ULL << 41,
798 ARM_HWCAP2_A64_SME_F16F16 = 1ULL << 42,
799 ARM_HWCAP2_A64_MOPS = 1ULL << 43,
800 ARM_HWCAP2_A64_HBC = 1ULL << 44,
801 };
802
803 #define ELF_HWCAP get_elf_hwcap()
804 #define ELF_HWCAP2 get_elf_hwcap2()
805
806 #define GET_FEATURE_ID(feat, hwcap) \
807 do { if (cpu_isar_feature(feat, cpu)) { hwcaps |= hwcap; } } while (0)
808
get_elf_hwcap(void)809 uint32_t get_elf_hwcap(void)
810 {
811 ARMCPU *cpu = ARM_CPU(thread_cpu);
812 uint32_t hwcaps = 0;
813
814 hwcaps |= ARM_HWCAP_A64_FP;
815 hwcaps |= ARM_HWCAP_A64_ASIMD;
816 hwcaps |= ARM_HWCAP_A64_CPUID;
817
818 /* probe for the extra features */
819
820 GET_FEATURE_ID(aa64_aes, ARM_HWCAP_A64_AES);
821 GET_FEATURE_ID(aa64_pmull, ARM_HWCAP_A64_PMULL);
822 GET_FEATURE_ID(aa64_sha1, ARM_HWCAP_A64_SHA1);
823 GET_FEATURE_ID(aa64_sha256, ARM_HWCAP_A64_SHA2);
824 GET_FEATURE_ID(aa64_sha512, ARM_HWCAP_A64_SHA512);
825 GET_FEATURE_ID(aa64_crc32, ARM_HWCAP_A64_CRC32);
826 GET_FEATURE_ID(aa64_sha3, ARM_HWCAP_A64_SHA3);
827 GET_FEATURE_ID(aa64_sm3, ARM_HWCAP_A64_SM3);
828 GET_FEATURE_ID(aa64_sm4, ARM_HWCAP_A64_SM4);
829 GET_FEATURE_ID(aa64_fp16, ARM_HWCAP_A64_FPHP | ARM_HWCAP_A64_ASIMDHP);
830 GET_FEATURE_ID(aa64_atomics, ARM_HWCAP_A64_ATOMICS);
831 GET_FEATURE_ID(aa64_lse2, ARM_HWCAP_A64_USCAT);
832 GET_FEATURE_ID(aa64_rdm, ARM_HWCAP_A64_ASIMDRDM);
833 GET_FEATURE_ID(aa64_dp, ARM_HWCAP_A64_ASIMDDP);
834 GET_FEATURE_ID(aa64_fcma, ARM_HWCAP_A64_FCMA);
835 GET_FEATURE_ID(aa64_sve, ARM_HWCAP_A64_SVE);
836 GET_FEATURE_ID(aa64_pauth, ARM_HWCAP_A64_PACA | ARM_HWCAP_A64_PACG);
837 GET_FEATURE_ID(aa64_fhm, ARM_HWCAP_A64_ASIMDFHM);
838 GET_FEATURE_ID(aa64_dit, ARM_HWCAP_A64_DIT);
839 GET_FEATURE_ID(aa64_jscvt, ARM_HWCAP_A64_JSCVT);
840 GET_FEATURE_ID(aa64_sb, ARM_HWCAP_A64_SB);
841 GET_FEATURE_ID(aa64_condm_4, ARM_HWCAP_A64_FLAGM);
842 GET_FEATURE_ID(aa64_dcpop, ARM_HWCAP_A64_DCPOP);
843 GET_FEATURE_ID(aa64_rcpc_8_3, ARM_HWCAP_A64_LRCPC);
844 GET_FEATURE_ID(aa64_rcpc_8_4, ARM_HWCAP_A64_ILRCPC);
845
846 return hwcaps;
847 }
848
get_elf_hwcap2(void)849 uint64_t get_elf_hwcap2(void)
850 {
851 ARMCPU *cpu = ARM_CPU(thread_cpu);
852 uint64_t hwcaps = 0;
853
854 GET_FEATURE_ID(aa64_dcpodp, ARM_HWCAP2_A64_DCPODP);
855 GET_FEATURE_ID(aa64_sve2, ARM_HWCAP2_A64_SVE2);
856 GET_FEATURE_ID(aa64_sve2_aes, ARM_HWCAP2_A64_SVEAES);
857 GET_FEATURE_ID(aa64_sve2_pmull128, ARM_HWCAP2_A64_SVEPMULL);
858 GET_FEATURE_ID(aa64_sve2_bitperm, ARM_HWCAP2_A64_SVEBITPERM);
859 GET_FEATURE_ID(aa64_sve2_sha3, ARM_HWCAP2_A64_SVESHA3);
860 GET_FEATURE_ID(aa64_sve2_sm4, ARM_HWCAP2_A64_SVESM4);
861 GET_FEATURE_ID(aa64_condm_5, ARM_HWCAP2_A64_FLAGM2);
862 GET_FEATURE_ID(aa64_frint, ARM_HWCAP2_A64_FRINT);
863 GET_FEATURE_ID(aa64_sve_i8mm, ARM_HWCAP2_A64_SVEI8MM);
864 GET_FEATURE_ID(aa64_sve_f32mm, ARM_HWCAP2_A64_SVEF32MM);
865 GET_FEATURE_ID(aa64_sve_f64mm, ARM_HWCAP2_A64_SVEF64MM);
866 GET_FEATURE_ID(aa64_sve_bf16, ARM_HWCAP2_A64_SVEBF16);
867 GET_FEATURE_ID(aa64_i8mm, ARM_HWCAP2_A64_I8MM);
868 GET_FEATURE_ID(aa64_bf16, ARM_HWCAP2_A64_BF16);
869 GET_FEATURE_ID(aa64_rndr, ARM_HWCAP2_A64_RNG);
870 GET_FEATURE_ID(aa64_bti, ARM_HWCAP2_A64_BTI);
871 GET_FEATURE_ID(aa64_mte, ARM_HWCAP2_A64_MTE);
872 GET_FEATURE_ID(aa64_mte3, ARM_HWCAP2_A64_MTE3);
873 GET_FEATURE_ID(aa64_sme, (ARM_HWCAP2_A64_SME |
874 ARM_HWCAP2_A64_SME_F32F32 |
875 ARM_HWCAP2_A64_SME_B16F32 |
876 ARM_HWCAP2_A64_SME_F16F32 |
877 ARM_HWCAP2_A64_SME_I8I32));
878 GET_FEATURE_ID(aa64_sme_f64f64, ARM_HWCAP2_A64_SME_F64F64);
879 GET_FEATURE_ID(aa64_sme_i16i64, ARM_HWCAP2_A64_SME_I16I64);
880 GET_FEATURE_ID(aa64_sme_fa64, ARM_HWCAP2_A64_SME_FA64);
881 GET_FEATURE_ID(aa64_hbc, ARM_HWCAP2_A64_HBC);
882 GET_FEATURE_ID(aa64_mops, ARM_HWCAP2_A64_MOPS);
883
884 return hwcaps;
885 }
886
elf_hwcap_str(uint32_t bit)887 const char *elf_hwcap_str(uint32_t bit)
888 {
889 static const char *hwcap_str[] = {
890 [__builtin_ctz(ARM_HWCAP_A64_FP )] = "fp",
891 [__builtin_ctz(ARM_HWCAP_A64_ASIMD )] = "asimd",
892 [__builtin_ctz(ARM_HWCAP_A64_EVTSTRM )] = "evtstrm",
893 [__builtin_ctz(ARM_HWCAP_A64_AES )] = "aes",
894 [__builtin_ctz(ARM_HWCAP_A64_PMULL )] = "pmull",
895 [__builtin_ctz(ARM_HWCAP_A64_SHA1 )] = "sha1",
896 [__builtin_ctz(ARM_HWCAP_A64_SHA2 )] = "sha2",
897 [__builtin_ctz(ARM_HWCAP_A64_CRC32 )] = "crc32",
898 [__builtin_ctz(ARM_HWCAP_A64_ATOMICS )] = "atomics",
899 [__builtin_ctz(ARM_HWCAP_A64_FPHP )] = "fphp",
900 [__builtin_ctz(ARM_HWCAP_A64_ASIMDHP )] = "asimdhp",
901 [__builtin_ctz(ARM_HWCAP_A64_CPUID )] = "cpuid",
902 [__builtin_ctz(ARM_HWCAP_A64_ASIMDRDM)] = "asimdrdm",
903 [__builtin_ctz(ARM_HWCAP_A64_JSCVT )] = "jscvt",
904 [__builtin_ctz(ARM_HWCAP_A64_FCMA )] = "fcma",
905 [__builtin_ctz(ARM_HWCAP_A64_LRCPC )] = "lrcpc",
906 [__builtin_ctz(ARM_HWCAP_A64_DCPOP )] = "dcpop",
907 [__builtin_ctz(ARM_HWCAP_A64_SHA3 )] = "sha3",
908 [__builtin_ctz(ARM_HWCAP_A64_SM3 )] = "sm3",
909 [__builtin_ctz(ARM_HWCAP_A64_SM4 )] = "sm4",
910 [__builtin_ctz(ARM_HWCAP_A64_ASIMDDP )] = "asimddp",
911 [__builtin_ctz(ARM_HWCAP_A64_SHA512 )] = "sha512",
912 [__builtin_ctz(ARM_HWCAP_A64_SVE )] = "sve",
913 [__builtin_ctz(ARM_HWCAP_A64_ASIMDFHM)] = "asimdfhm",
914 [__builtin_ctz(ARM_HWCAP_A64_DIT )] = "dit",
915 [__builtin_ctz(ARM_HWCAP_A64_USCAT )] = "uscat",
916 [__builtin_ctz(ARM_HWCAP_A64_ILRCPC )] = "ilrcpc",
917 [__builtin_ctz(ARM_HWCAP_A64_FLAGM )] = "flagm",
918 [__builtin_ctz(ARM_HWCAP_A64_SSBS )] = "ssbs",
919 [__builtin_ctz(ARM_HWCAP_A64_SB )] = "sb",
920 [__builtin_ctz(ARM_HWCAP_A64_PACA )] = "paca",
921 [__builtin_ctz(ARM_HWCAP_A64_PACG )] = "pacg",
922 };
923
924 return bit < ARRAY_SIZE(hwcap_str) ? hwcap_str[bit] : NULL;
925 }
926
elf_hwcap2_str(uint32_t bit)927 const char *elf_hwcap2_str(uint32_t bit)
928 {
929 static const char *hwcap_str[] = {
930 [__builtin_ctz(ARM_HWCAP2_A64_DCPODP )] = "dcpodp",
931 [__builtin_ctz(ARM_HWCAP2_A64_SVE2 )] = "sve2",
932 [__builtin_ctz(ARM_HWCAP2_A64_SVEAES )] = "sveaes",
933 [__builtin_ctz(ARM_HWCAP2_A64_SVEPMULL )] = "svepmull",
934 [__builtin_ctz(ARM_HWCAP2_A64_SVEBITPERM )] = "svebitperm",
935 [__builtin_ctz(ARM_HWCAP2_A64_SVESHA3 )] = "svesha3",
936 [__builtin_ctz(ARM_HWCAP2_A64_SVESM4 )] = "svesm4",
937 [__builtin_ctz(ARM_HWCAP2_A64_FLAGM2 )] = "flagm2",
938 [__builtin_ctz(ARM_HWCAP2_A64_FRINT )] = "frint",
939 [__builtin_ctz(ARM_HWCAP2_A64_SVEI8MM )] = "svei8mm",
940 [__builtin_ctz(ARM_HWCAP2_A64_SVEF32MM )] = "svef32mm",
941 [__builtin_ctz(ARM_HWCAP2_A64_SVEF64MM )] = "svef64mm",
942 [__builtin_ctz(ARM_HWCAP2_A64_SVEBF16 )] = "svebf16",
943 [__builtin_ctz(ARM_HWCAP2_A64_I8MM )] = "i8mm",
944 [__builtin_ctz(ARM_HWCAP2_A64_BF16 )] = "bf16",
945 [__builtin_ctz(ARM_HWCAP2_A64_DGH )] = "dgh",
946 [__builtin_ctz(ARM_HWCAP2_A64_RNG )] = "rng",
947 [__builtin_ctz(ARM_HWCAP2_A64_BTI )] = "bti",
948 [__builtin_ctz(ARM_HWCAP2_A64_MTE )] = "mte",
949 [__builtin_ctz(ARM_HWCAP2_A64_ECV )] = "ecv",
950 [__builtin_ctz(ARM_HWCAP2_A64_AFP )] = "afp",
951 [__builtin_ctz(ARM_HWCAP2_A64_RPRES )] = "rpres",
952 [__builtin_ctz(ARM_HWCAP2_A64_MTE3 )] = "mte3",
953 [__builtin_ctz(ARM_HWCAP2_A64_SME )] = "sme",
954 [__builtin_ctz(ARM_HWCAP2_A64_SME_I16I64 )] = "smei16i64",
955 [__builtin_ctz(ARM_HWCAP2_A64_SME_F64F64 )] = "smef64f64",
956 [__builtin_ctz(ARM_HWCAP2_A64_SME_I8I32 )] = "smei8i32",
957 [__builtin_ctz(ARM_HWCAP2_A64_SME_F16F32 )] = "smef16f32",
958 [__builtin_ctz(ARM_HWCAP2_A64_SME_B16F32 )] = "smeb16f32",
959 [__builtin_ctz(ARM_HWCAP2_A64_SME_F32F32 )] = "smef32f32",
960 [__builtin_ctz(ARM_HWCAP2_A64_SME_FA64 )] = "smefa64",
961 [__builtin_ctz(ARM_HWCAP2_A64_WFXT )] = "wfxt",
962 [__builtin_ctzll(ARM_HWCAP2_A64_EBF16 )] = "ebf16",
963 [__builtin_ctzll(ARM_HWCAP2_A64_SVE_EBF16 )] = "sveebf16",
964 [__builtin_ctzll(ARM_HWCAP2_A64_CSSC )] = "cssc",
965 [__builtin_ctzll(ARM_HWCAP2_A64_RPRFM )] = "rprfm",
966 [__builtin_ctzll(ARM_HWCAP2_A64_SVE2P1 )] = "sve2p1",
967 [__builtin_ctzll(ARM_HWCAP2_A64_SME2 )] = "sme2",
968 [__builtin_ctzll(ARM_HWCAP2_A64_SME2P1 )] = "sme2p1",
969 [__builtin_ctzll(ARM_HWCAP2_A64_SME_I16I32 )] = "smei16i32",
970 [__builtin_ctzll(ARM_HWCAP2_A64_SME_BI32I32)] = "smebi32i32",
971 [__builtin_ctzll(ARM_HWCAP2_A64_SME_B16B16 )] = "smeb16b16",
972 [__builtin_ctzll(ARM_HWCAP2_A64_SME_F16F16 )] = "smef16f16",
973 [__builtin_ctzll(ARM_HWCAP2_A64_MOPS )] = "mops",
974 [__builtin_ctzll(ARM_HWCAP2_A64_HBC )] = "hbc",
975 };
976
977 return bit < ARRAY_SIZE(hwcap_str) ? hwcap_str[bit] : NULL;
978 }
979
980 #undef GET_FEATURE_ID
981
982 #if TARGET_BIG_ENDIAN
983 # define VDSO_HEADER "vdso-be.c.inc"
984 #else
985 # define VDSO_HEADER "vdso-le.c.inc"
986 #endif
987
988 #endif /* not TARGET_AARCH64 */
989
990 #endif /* TARGET_ARM */
991
992 #ifdef TARGET_SPARC
993
994 #ifndef TARGET_SPARC64
995 # define ELF_CLASS ELFCLASS32
996 # define ELF_ARCH EM_SPARC
997 #elif defined(TARGET_ABI32)
998 # define ELF_CLASS ELFCLASS32
999 # define elf_check_arch(x) ((x) == EM_SPARC32PLUS || (x) == EM_SPARC)
1000 #else
1001 # define ELF_CLASS ELFCLASS64
1002 # define ELF_ARCH EM_SPARCV9
1003 #endif
1004
1005 #include "elf.h"
1006
1007 #define ELF_HWCAP get_elf_hwcap()
1008
get_elf_hwcap(void)1009 static uint32_t get_elf_hwcap(void)
1010 {
1011 /* There are not many sparc32 hwcap bits -- we have all of them. */
1012 uint32_t r = HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR |
1013 HWCAP_SPARC_SWAP | HWCAP_SPARC_MULDIV;
1014
1015 #ifdef TARGET_SPARC64
1016 CPUSPARCState *env = cpu_env(thread_cpu);
1017 uint32_t features = env->def.features;
1018
1019 r |= HWCAP_SPARC_V9 | HWCAP_SPARC_V8PLUS;
1020 /* 32x32 multiply and divide are efficient. */
1021 r |= HWCAP_SPARC_MUL32 | HWCAP_SPARC_DIV32;
1022 /* We don't have an internal feature bit for this. */
1023 r |= HWCAP_SPARC_POPC;
1024 r |= features & CPU_FEATURE_FSMULD ? HWCAP_SPARC_FSMULD : 0;
1025 r |= features & CPU_FEATURE_VIS1 ? HWCAP_SPARC_VIS : 0;
1026 r |= features & CPU_FEATURE_VIS2 ? HWCAP_SPARC_VIS2 : 0;
1027 r |= features & CPU_FEATURE_FMAF ? HWCAP_SPARC_FMAF : 0;
1028 r |= features & CPU_FEATURE_VIS3 ? HWCAP_SPARC_VIS3 : 0;
1029 r |= features & CPU_FEATURE_IMA ? HWCAP_SPARC_IMA : 0;
1030 #endif
1031
1032 return r;
1033 }
1034
init_thread(struct target_pt_regs * regs,struct image_info * infop)1035 static inline void init_thread(struct target_pt_regs *regs,
1036 struct image_info *infop)
1037 {
1038 /* Note that target_cpu_copy_regs does not read psr/tstate. */
1039 regs->pc = infop->entry;
1040 regs->npc = regs->pc + 4;
1041 regs->y = 0;
1042 regs->u_regs[14] = (infop->start_stack - 16 * sizeof(abi_ulong)
1043 - TARGET_STACK_BIAS);
1044 }
1045 #endif /* TARGET_SPARC */
1046
1047 #ifdef TARGET_PPC
1048
1049 #define ELF_MACHINE PPC_ELF_MACHINE
1050
1051 #if defined(TARGET_PPC64)
1052
1053 #define elf_check_arch(x) ( (x) == EM_PPC64 )
1054
1055 #define ELF_CLASS ELFCLASS64
1056
1057 #else
1058
1059 #define ELF_CLASS ELFCLASS32
1060 #define EXSTACK_DEFAULT true
1061
1062 #endif
1063
1064 #define ELF_ARCH EM_PPC
1065
1066 /* Feature masks for the Aux Vector Hardware Capabilities (AT_HWCAP).
1067 See arch/powerpc/include/asm/cputable.h. */
1068 enum {
1069 QEMU_PPC_FEATURE_32 = 0x80000000,
1070 QEMU_PPC_FEATURE_64 = 0x40000000,
1071 QEMU_PPC_FEATURE_601_INSTR = 0x20000000,
1072 QEMU_PPC_FEATURE_HAS_ALTIVEC = 0x10000000,
1073 QEMU_PPC_FEATURE_HAS_FPU = 0x08000000,
1074 QEMU_PPC_FEATURE_HAS_MMU = 0x04000000,
1075 QEMU_PPC_FEATURE_HAS_4xxMAC = 0x02000000,
1076 QEMU_PPC_FEATURE_UNIFIED_CACHE = 0x01000000,
1077 QEMU_PPC_FEATURE_HAS_SPE = 0x00800000,
1078 QEMU_PPC_FEATURE_HAS_EFP_SINGLE = 0x00400000,
1079 QEMU_PPC_FEATURE_HAS_EFP_DOUBLE = 0x00200000,
1080 QEMU_PPC_FEATURE_NO_TB = 0x00100000,
1081 QEMU_PPC_FEATURE_POWER4 = 0x00080000,
1082 QEMU_PPC_FEATURE_POWER5 = 0x00040000,
1083 QEMU_PPC_FEATURE_POWER5_PLUS = 0x00020000,
1084 QEMU_PPC_FEATURE_CELL = 0x00010000,
1085 QEMU_PPC_FEATURE_BOOKE = 0x00008000,
1086 QEMU_PPC_FEATURE_SMT = 0x00004000,
1087 QEMU_PPC_FEATURE_ICACHE_SNOOP = 0x00002000,
1088 QEMU_PPC_FEATURE_ARCH_2_05 = 0x00001000,
1089 QEMU_PPC_FEATURE_PA6T = 0x00000800,
1090 QEMU_PPC_FEATURE_HAS_DFP = 0x00000400,
1091 QEMU_PPC_FEATURE_POWER6_EXT = 0x00000200,
1092 QEMU_PPC_FEATURE_ARCH_2_06 = 0x00000100,
1093 QEMU_PPC_FEATURE_HAS_VSX = 0x00000080,
1094 QEMU_PPC_FEATURE_PSERIES_PERFMON_COMPAT = 0x00000040,
1095
1096 QEMU_PPC_FEATURE_TRUE_LE = 0x00000002,
1097 QEMU_PPC_FEATURE_PPC_LE = 0x00000001,
1098
1099 /* Feature definitions in AT_HWCAP2. */
1100 QEMU_PPC_FEATURE2_ARCH_2_07 = 0x80000000, /* ISA 2.07 */
1101 QEMU_PPC_FEATURE2_HAS_HTM = 0x40000000, /* Hardware Transactional Memory */
1102 QEMU_PPC_FEATURE2_HAS_DSCR = 0x20000000, /* Data Stream Control Register */
1103 QEMU_PPC_FEATURE2_HAS_EBB = 0x10000000, /* Event Base Branching */
1104 QEMU_PPC_FEATURE2_HAS_ISEL = 0x08000000, /* Integer Select */
1105 QEMU_PPC_FEATURE2_HAS_TAR = 0x04000000, /* Target Address Register */
1106 QEMU_PPC_FEATURE2_VEC_CRYPTO = 0x02000000,
1107 QEMU_PPC_FEATURE2_HTM_NOSC = 0x01000000,
1108 QEMU_PPC_FEATURE2_ARCH_3_00 = 0x00800000, /* ISA 3.00 */
1109 QEMU_PPC_FEATURE2_HAS_IEEE128 = 0x00400000, /* VSX IEEE Bin Float 128-bit */
1110 QEMU_PPC_FEATURE2_DARN = 0x00200000, /* darn random number insn */
1111 QEMU_PPC_FEATURE2_SCV = 0x00100000, /* scv syscall */
1112 QEMU_PPC_FEATURE2_HTM_NO_SUSPEND = 0x00080000, /* TM w/o suspended state */
1113 QEMU_PPC_FEATURE2_ARCH_3_1 = 0x00040000, /* ISA 3.1 */
1114 QEMU_PPC_FEATURE2_MMA = 0x00020000, /* Matrix-Multiply Assist */
1115 };
1116
1117 #define ELF_HWCAP get_elf_hwcap()
1118
get_elf_hwcap(void)1119 static uint32_t get_elf_hwcap(void)
1120 {
1121 PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);
1122 uint32_t features = 0;
1123
1124 /* We don't have to be terribly complete here; the high points are
1125 Altivec/FP/SPE support. Anything else is just a bonus. */
1126 #define GET_FEATURE(flag, feature) \
1127 do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0)
1128 #define GET_FEATURE2(flags, feature) \
1129 do { \
1130 if ((cpu->env.insns_flags2 & flags) == flags) { \
1131 features |= feature; \
1132 } \
1133 } while (0)
1134 GET_FEATURE(PPC_64B, QEMU_PPC_FEATURE_64);
1135 GET_FEATURE(PPC_FLOAT, QEMU_PPC_FEATURE_HAS_FPU);
1136 GET_FEATURE(PPC_ALTIVEC, QEMU_PPC_FEATURE_HAS_ALTIVEC);
1137 GET_FEATURE(PPC_SPE, QEMU_PPC_FEATURE_HAS_SPE);
1138 GET_FEATURE(PPC_SPE_SINGLE, QEMU_PPC_FEATURE_HAS_EFP_SINGLE);
1139 GET_FEATURE(PPC_SPE_DOUBLE, QEMU_PPC_FEATURE_HAS_EFP_DOUBLE);
1140 GET_FEATURE(PPC_BOOKE, QEMU_PPC_FEATURE_BOOKE);
1141 GET_FEATURE(PPC_405_MAC, QEMU_PPC_FEATURE_HAS_4xxMAC);
1142 GET_FEATURE2(PPC2_DFP, QEMU_PPC_FEATURE_HAS_DFP);
1143 GET_FEATURE2(PPC2_VSX, QEMU_PPC_FEATURE_HAS_VSX);
1144 GET_FEATURE2((PPC2_PERM_ISA206 | PPC2_DIVE_ISA206 | PPC2_ATOMIC_ISA206 |
1145 PPC2_FP_CVT_ISA206 | PPC2_FP_TST_ISA206),
1146 QEMU_PPC_FEATURE_ARCH_2_06);
1147 #undef GET_FEATURE
1148 #undef GET_FEATURE2
1149
1150 return features;
1151 }
1152
1153 #define ELF_HWCAP2 get_elf_hwcap2()
1154
get_elf_hwcap2(void)1155 static uint32_t get_elf_hwcap2(void)
1156 {
1157 PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);
1158 uint32_t features = 0;
1159
1160 #define GET_FEATURE(flag, feature) \
1161 do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0)
1162 #define GET_FEATURE2(flag, feature) \
1163 do { if (cpu->env.insns_flags2 & flag) { features |= feature; } } while (0)
1164
1165 GET_FEATURE(PPC_ISEL, QEMU_PPC_FEATURE2_HAS_ISEL);
1166 GET_FEATURE2(PPC2_BCTAR_ISA207, QEMU_PPC_FEATURE2_HAS_TAR);
1167 GET_FEATURE2((PPC2_BCTAR_ISA207 | PPC2_LSQ_ISA207 | PPC2_ALTIVEC_207 |
1168 PPC2_ISA207S), QEMU_PPC_FEATURE2_ARCH_2_07 |
1169 QEMU_PPC_FEATURE2_VEC_CRYPTO);
1170 GET_FEATURE2(PPC2_ISA300, QEMU_PPC_FEATURE2_ARCH_3_00 |
1171 QEMU_PPC_FEATURE2_DARN | QEMU_PPC_FEATURE2_HAS_IEEE128);
1172 GET_FEATURE2(PPC2_ISA310, QEMU_PPC_FEATURE2_ARCH_3_1 |
1173 QEMU_PPC_FEATURE2_MMA);
1174
1175 #undef GET_FEATURE
1176 #undef GET_FEATURE2
1177
1178 return features;
1179 }
1180
1181 /*
1182 * The requirements here are:
1183 * - keep the final alignment of sp (sp & 0xf)
1184 * - make sure the 32-bit value at the first 16 byte aligned position of
1185 * AUXV is greater than 16 for glibc compatibility.
1186 * AT_IGNOREPPC is used for that.
1187 * - for compatibility with glibc ARCH_DLINFO must always be defined on PPC,
1188 * even if DLINFO_ARCH_ITEMS goes to zero or is undefined.
1189 */
1190 #define DLINFO_ARCH_ITEMS 5
1191 #define ARCH_DLINFO \
1192 do { \
1193 PowerPCCPU *cpu = POWERPC_CPU(thread_cpu); \
1194 /* \
1195 * Handle glibc compatibility: these magic entries must \
1196 * be at the lowest addresses in the final auxv. \
1197 */ \
1198 NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC); \
1199 NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC); \
1200 NEW_AUX_ENT(AT_DCACHEBSIZE, cpu->env.dcache_line_size); \
1201 NEW_AUX_ENT(AT_ICACHEBSIZE, cpu->env.icache_line_size); \
1202 NEW_AUX_ENT(AT_UCACHEBSIZE, 0); \
1203 } while (0)
1204
init_thread(struct target_pt_regs * _regs,struct image_info * infop)1205 static inline void init_thread(struct target_pt_regs *_regs, struct image_info *infop)
1206 {
1207 _regs->gpr[1] = infop->start_stack;
1208 #if defined(TARGET_PPC64)
1209 if (get_ppc64_abi(infop) < 2) {
1210 uint64_t val;
1211 get_user_u64(val, infop->entry + 8);
1212 _regs->gpr[2] = val + infop->load_bias;
1213 get_user_u64(val, infop->entry);
1214 infop->entry = val + infop->load_bias;
1215 } else {
1216 _regs->gpr[12] = infop->entry; /* r12 set to global entry address */
1217 }
1218 #endif
1219 _regs->nip = infop->entry;
1220 }
1221
1222 /* See linux kernel: arch/powerpc/include/asm/elf.h. */
1223 #define ELF_NREG 48
1224 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1225
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUPPCState * env)1226 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUPPCState *env)
1227 {
1228 int i;
1229 target_ulong ccr = 0;
1230
1231 for (i = 0; i < ARRAY_SIZE(env->gpr); i++) {
1232 (*regs)[i] = tswapreg(env->gpr[i]);
1233 }
1234
1235 (*regs)[32] = tswapreg(env->nip);
1236 (*regs)[33] = tswapreg(env->msr);
1237 (*regs)[35] = tswapreg(env->ctr);
1238 (*regs)[36] = tswapreg(env->lr);
1239 (*regs)[37] = tswapreg(cpu_read_xer(env));
1240
1241 ccr = ppc_get_cr(env);
1242 (*regs)[38] = tswapreg(ccr);
1243 }
1244
1245 #define USE_ELF_CORE_DUMP
1246 #define ELF_EXEC_PAGESIZE 4096
1247
1248 #ifndef TARGET_PPC64
1249 # define VDSO_HEADER "vdso-32.c.inc"
1250 #elif TARGET_BIG_ENDIAN
1251 # define VDSO_HEADER "vdso-64.c.inc"
1252 #else
1253 # define VDSO_HEADER "vdso-64le.c.inc"
1254 #endif
1255
1256 #endif
1257
1258 #ifdef TARGET_LOONGARCH64
1259
1260 #define ELF_CLASS ELFCLASS64
1261 #define ELF_ARCH EM_LOONGARCH
1262 #define EXSTACK_DEFAULT true
1263
1264 #define elf_check_arch(x) ((x) == EM_LOONGARCH)
1265
1266 #define VDSO_HEADER "vdso.c.inc"
1267
init_thread(struct target_pt_regs * regs,struct image_info * infop)1268 static inline void init_thread(struct target_pt_regs *regs,
1269 struct image_info *infop)
1270 {
1271 /*Set crmd PG,DA = 1,0 */
1272 regs->csr.crmd = 2 << 3;
1273 regs->csr.era = infop->entry;
1274 regs->regs[3] = infop->start_stack;
1275 }
1276
1277 /* See linux kernel: arch/loongarch/include/asm/elf.h */
1278 #define ELF_NREG 45
1279 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1280
1281 enum {
1282 TARGET_EF_R0 = 0,
1283 TARGET_EF_CSR_ERA = TARGET_EF_R0 + 33,
1284 TARGET_EF_CSR_BADV = TARGET_EF_R0 + 34,
1285 };
1286
elf_core_copy_regs(target_elf_gregset_t * regs,const CPULoongArchState * env)1287 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1288 const CPULoongArchState *env)
1289 {
1290 int i;
1291
1292 (*regs)[TARGET_EF_R0] = 0;
1293
1294 for (i = 1; i < ARRAY_SIZE(env->gpr); i++) {
1295 (*regs)[TARGET_EF_R0 + i] = tswapreg(env->gpr[i]);
1296 }
1297
1298 (*regs)[TARGET_EF_CSR_ERA] = tswapreg(env->pc);
1299 (*regs)[TARGET_EF_CSR_BADV] = tswapreg(env->CSR_BADV);
1300 }
1301
1302 #define USE_ELF_CORE_DUMP
1303 #define ELF_EXEC_PAGESIZE 4096
1304
1305 #define ELF_HWCAP get_elf_hwcap()
1306
1307 /* See arch/loongarch/include/uapi/asm/hwcap.h */
1308 enum {
1309 HWCAP_LOONGARCH_CPUCFG = (1 << 0),
1310 HWCAP_LOONGARCH_LAM = (1 << 1),
1311 HWCAP_LOONGARCH_UAL = (1 << 2),
1312 HWCAP_LOONGARCH_FPU = (1 << 3),
1313 HWCAP_LOONGARCH_LSX = (1 << 4),
1314 HWCAP_LOONGARCH_LASX = (1 << 5),
1315 HWCAP_LOONGARCH_CRC32 = (1 << 6),
1316 HWCAP_LOONGARCH_COMPLEX = (1 << 7),
1317 HWCAP_LOONGARCH_CRYPTO = (1 << 8),
1318 HWCAP_LOONGARCH_LVZ = (1 << 9),
1319 HWCAP_LOONGARCH_LBT_X86 = (1 << 10),
1320 HWCAP_LOONGARCH_LBT_ARM = (1 << 11),
1321 HWCAP_LOONGARCH_LBT_MIPS = (1 << 12),
1322 };
1323
get_elf_hwcap(void)1324 static uint32_t get_elf_hwcap(void)
1325 {
1326 LoongArchCPU *cpu = LOONGARCH_CPU(thread_cpu);
1327 uint32_t hwcaps = 0;
1328
1329 hwcaps |= HWCAP_LOONGARCH_CRC32;
1330
1331 if (FIELD_EX32(cpu->env.cpucfg[1], CPUCFG1, UAL)) {
1332 hwcaps |= HWCAP_LOONGARCH_UAL;
1333 }
1334
1335 if (FIELD_EX32(cpu->env.cpucfg[2], CPUCFG2, FP)) {
1336 hwcaps |= HWCAP_LOONGARCH_FPU;
1337 }
1338
1339 if (FIELD_EX32(cpu->env.cpucfg[2], CPUCFG2, LAM)) {
1340 hwcaps |= HWCAP_LOONGARCH_LAM;
1341 }
1342
1343 if (FIELD_EX32(cpu->env.cpucfg[2], CPUCFG2, LSX)) {
1344 hwcaps |= HWCAP_LOONGARCH_LSX;
1345 }
1346
1347 if (FIELD_EX32(cpu->env.cpucfg[2], CPUCFG2, LASX)) {
1348 hwcaps |= HWCAP_LOONGARCH_LASX;
1349 }
1350
1351 return hwcaps;
1352 }
1353
1354 #define ELF_PLATFORM "loongarch"
1355
1356 #endif /* TARGET_LOONGARCH64 */
1357
1358 #ifdef TARGET_MIPS
1359
1360 #ifdef TARGET_MIPS64
1361 #define ELF_CLASS ELFCLASS64
1362 #else
1363 #define ELF_CLASS ELFCLASS32
1364 #endif
1365 #define ELF_ARCH EM_MIPS
1366 #define EXSTACK_DEFAULT true
1367
1368 #ifdef TARGET_ABI_MIPSN32
1369 #define elf_check_abi(x) ((x) & EF_MIPS_ABI2)
1370 #else
1371 #define elf_check_abi(x) (!((x) & EF_MIPS_ABI2))
1372 #endif
1373
1374 #define ELF_BASE_PLATFORM get_elf_base_platform()
1375
1376 #define MATCH_PLATFORM_INSN(_flags, _base_platform) \
1377 do { if ((cpu->env.insn_flags & (_flags)) == _flags) \
1378 { return _base_platform; } } while (0)
1379
get_elf_base_platform(void)1380 static const char *get_elf_base_platform(void)
1381 {
1382 MIPSCPU *cpu = MIPS_CPU(thread_cpu);
1383
1384 /* 64 bit ISAs goes first */
1385 MATCH_PLATFORM_INSN(CPU_MIPS64R6, "mips64r6");
1386 MATCH_PLATFORM_INSN(CPU_MIPS64R5, "mips64r5");
1387 MATCH_PLATFORM_INSN(CPU_MIPS64R2, "mips64r2");
1388 MATCH_PLATFORM_INSN(CPU_MIPS64R1, "mips64");
1389 MATCH_PLATFORM_INSN(CPU_MIPS5, "mips5");
1390 MATCH_PLATFORM_INSN(CPU_MIPS4, "mips4");
1391 MATCH_PLATFORM_INSN(CPU_MIPS3, "mips3");
1392
1393 /* 32 bit ISAs */
1394 MATCH_PLATFORM_INSN(CPU_MIPS32R6, "mips32r6");
1395 MATCH_PLATFORM_INSN(CPU_MIPS32R5, "mips32r5");
1396 MATCH_PLATFORM_INSN(CPU_MIPS32R2, "mips32r2");
1397 MATCH_PLATFORM_INSN(CPU_MIPS32R1, "mips32");
1398 MATCH_PLATFORM_INSN(CPU_MIPS2, "mips2");
1399
1400 /* Fallback */
1401 return "mips";
1402 }
1403 #undef MATCH_PLATFORM_INSN
1404
init_thread(struct target_pt_regs * regs,struct image_info * infop)1405 static inline void init_thread(struct target_pt_regs *regs,
1406 struct image_info *infop)
1407 {
1408 regs->cp0_status = 2 << CP0St_KSU;
1409 regs->cp0_epc = infop->entry;
1410 regs->regs[29] = infop->start_stack;
1411 }
1412
1413 /* See linux kernel: arch/mips/include/asm/elf.h. */
1414 #define ELF_NREG 45
1415 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1416
1417 /* See linux kernel: arch/mips/include/asm/reg.h. */
1418 enum {
1419 #ifdef TARGET_MIPS64
1420 TARGET_EF_R0 = 0,
1421 #else
1422 TARGET_EF_R0 = 6,
1423 #endif
1424 TARGET_EF_R26 = TARGET_EF_R0 + 26,
1425 TARGET_EF_R27 = TARGET_EF_R0 + 27,
1426 TARGET_EF_LO = TARGET_EF_R0 + 32,
1427 TARGET_EF_HI = TARGET_EF_R0 + 33,
1428 TARGET_EF_CP0_EPC = TARGET_EF_R0 + 34,
1429 TARGET_EF_CP0_BADVADDR = TARGET_EF_R0 + 35,
1430 TARGET_EF_CP0_STATUS = TARGET_EF_R0 + 36,
1431 TARGET_EF_CP0_CAUSE = TARGET_EF_R0 + 37
1432 };
1433
1434 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUMIPSState * env)1435 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMIPSState *env)
1436 {
1437 int i;
1438
1439 for (i = 0; i < TARGET_EF_R0; i++) {
1440 (*regs)[i] = 0;
1441 }
1442 (*regs)[TARGET_EF_R0] = 0;
1443
1444 for (i = 1; i < ARRAY_SIZE(env->active_tc.gpr); i++) {
1445 (*regs)[TARGET_EF_R0 + i] = tswapreg(env->active_tc.gpr[i]);
1446 }
1447
1448 (*regs)[TARGET_EF_R26] = 0;
1449 (*regs)[TARGET_EF_R27] = 0;
1450 (*regs)[TARGET_EF_LO] = tswapreg(env->active_tc.LO[0]);
1451 (*regs)[TARGET_EF_HI] = tswapreg(env->active_tc.HI[0]);
1452 (*regs)[TARGET_EF_CP0_EPC] = tswapreg(env->active_tc.PC);
1453 (*regs)[TARGET_EF_CP0_BADVADDR] = tswapreg(env->CP0_BadVAddr);
1454 (*regs)[TARGET_EF_CP0_STATUS] = tswapreg(env->CP0_Status);
1455 (*regs)[TARGET_EF_CP0_CAUSE] = tswapreg(env->CP0_Cause);
1456 }
1457
1458 #define USE_ELF_CORE_DUMP
1459 #define ELF_EXEC_PAGESIZE 4096
1460
1461 /* See arch/mips/include/uapi/asm/hwcap.h. */
1462 enum {
1463 HWCAP_MIPS_R6 = (1 << 0),
1464 HWCAP_MIPS_MSA = (1 << 1),
1465 HWCAP_MIPS_CRC32 = (1 << 2),
1466 HWCAP_MIPS_MIPS16 = (1 << 3),
1467 HWCAP_MIPS_MDMX = (1 << 4),
1468 HWCAP_MIPS_MIPS3D = (1 << 5),
1469 HWCAP_MIPS_SMARTMIPS = (1 << 6),
1470 HWCAP_MIPS_DSP = (1 << 7),
1471 HWCAP_MIPS_DSP2 = (1 << 8),
1472 HWCAP_MIPS_DSP3 = (1 << 9),
1473 HWCAP_MIPS_MIPS16E2 = (1 << 10),
1474 HWCAP_LOONGSON_MMI = (1 << 11),
1475 HWCAP_LOONGSON_EXT = (1 << 12),
1476 HWCAP_LOONGSON_EXT2 = (1 << 13),
1477 HWCAP_LOONGSON_CPUCFG = (1 << 14),
1478 };
1479
1480 #define ELF_HWCAP get_elf_hwcap()
1481
1482 #define GET_FEATURE_INSN(_flag, _hwcap) \
1483 do { if (cpu->env.insn_flags & (_flag)) { hwcaps |= _hwcap; } } while (0)
1484
1485 #define GET_FEATURE_REG_SET(_reg, _mask, _hwcap) \
1486 do { if (cpu->env._reg & (_mask)) { hwcaps |= _hwcap; } } while (0)
1487
1488 #define GET_FEATURE_REG_EQU(_reg, _start, _length, _val, _hwcap) \
1489 do { \
1490 if (extract32(cpu->env._reg, (_start), (_length)) == (_val)) { \
1491 hwcaps |= _hwcap; \
1492 } \
1493 } while (0)
1494
get_elf_hwcap(void)1495 static uint32_t get_elf_hwcap(void)
1496 {
1497 MIPSCPU *cpu = MIPS_CPU(thread_cpu);
1498 uint32_t hwcaps = 0;
1499
1500 GET_FEATURE_REG_EQU(CP0_Config0, CP0C0_AR, CP0C0_AR_LENGTH,
1501 2, HWCAP_MIPS_R6);
1502 GET_FEATURE_REG_SET(CP0_Config3, 1 << CP0C3_MSAP, HWCAP_MIPS_MSA);
1503 GET_FEATURE_INSN(ASE_LMMI, HWCAP_LOONGSON_MMI);
1504 GET_FEATURE_INSN(ASE_LEXT, HWCAP_LOONGSON_EXT);
1505
1506 return hwcaps;
1507 }
1508
1509 #undef GET_FEATURE_REG_EQU
1510 #undef GET_FEATURE_REG_SET
1511 #undef GET_FEATURE_INSN
1512
1513 #endif /* TARGET_MIPS */
1514
1515 #ifdef TARGET_MICROBLAZE
1516
1517 #define elf_check_arch(x) ( (x) == EM_MICROBLAZE || (x) == EM_MICROBLAZE_OLD)
1518
1519 #define ELF_CLASS ELFCLASS32
1520 #define ELF_ARCH EM_MICROBLAZE
1521
init_thread(struct target_pt_regs * regs,struct image_info * infop)1522 static inline void init_thread(struct target_pt_regs *regs,
1523 struct image_info *infop)
1524 {
1525 regs->pc = infop->entry;
1526 regs->r1 = infop->start_stack;
1527
1528 }
1529
1530 #define ELF_EXEC_PAGESIZE 4096
1531
1532 #define USE_ELF_CORE_DUMP
1533 #define ELF_NREG 38
1534 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1535
1536 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUMBState * env)1537 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMBState *env)
1538 {
1539 int i, pos = 0;
1540
1541 for (i = 0; i < 32; i++) {
1542 (*regs)[pos++] = tswapreg(env->regs[i]);
1543 }
1544
1545 (*regs)[pos++] = tswapreg(env->pc);
1546 (*regs)[pos++] = tswapreg(mb_cpu_read_msr(env));
1547 (*regs)[pos++] = 0;
1548 (*regs)[pos++] = tswapreg(env->ear);
1549 (*regs)[pos++] = 0;
1550 (*regs)[pos++] = tswapreg(env->esr);
1551 }
1552
1553 #endif /* TARGET_MICROBLAZE */
1554
1555 #ifdef TARGET_OPENRISC
1556
1557 #define ELF_ARCH EM_OPENRISC
1558 #define ELF_CLASS ELFCLASS32
1559 #define ELF_DATA ELFDATA2MSB
1560
init_thread(struct target_pt_regs * regs,struct image_info * infop)1561 static inline void init_thread(struct target_pt_regs *regs,
1562 struct image_info *infop)
1563 {
1564 regs->pc = infop->entry;
1565 regs->gpr[1] = infop->start_stack;
1566 }
1567
1568 #define USE_ELF_CORE_DUMP
1569 #define ELF_EXEC_PAGESIZE 8192
1570
1571 /* See linux kernel arch/openrisc/include/asm/elf.h. */
1572 #define ELF_NREG 34 /* gprs and pc, sr */
1573 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1574
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUOpenRISCState * env)1575 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1576 const CPUOpenRISCState *env)
1577 {
1578 int i;
1579
1580 for (i = 0; i < 32; i++) {
1581 (*regs)[i] = tswapreg(cpu_get_gpr(env, i));
1582 }
1583 (*regs)[32] = tswapreg(env->pc);
1584 (*regs)[33] = tswapreg(cpu_get_sr(env));
1585 }
1586 #define ELF_HWCAP 0
1587 #define ELF_PLATFORM NULL
1588
1589 #endif /* TARGET_OPENRISC */
1590
1591 #ifdef TARGET_SH4
1592
1593 #define ELF_CLASS ELFCLASS32
1594 #define ELF_ARCH EM_SH
1595
init_thread(struct target_pt_regs * regs,struct image_info * infop)1596 static inline void init_thread(struct target_pt_regs *regs,
1597 struct image_info *infop)
1598 {
1599 /* Check other registers XXXXX */
1600 regs->pc = infop->entry;
1601 regs->regs[15] = infop->start_stack;
1602 }
1603
1604 /* See linux kernel: arch/sh/include/asm/elf.h. */
1605 #define ELF_NREG 23
1606 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1607
1608 /* See linux kernel: arch/sh/include/asm/ptrace.h. */
1609 enum {
1610 TARGET_REG_PC = 16,
1611 TARGET_REG_PR = 17,
1612 TARGET_REG_SR = 18,
1613 TARGET_REG_GBR = 19,
1614 TARGET_REG_MACH = 20,
1615 TARGET_REG_MACL = 21,
1616 TARGET_REG_SYSCALL = 22
1617 };
1618
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUSH4State * env)1619 static inline void elf_core_copy_regs(target_elf_gregset_t *regs,
1620 const CPUSH4State *env)
1621 {
1622 int i;
1623
1624 for (i = 0; i < 16; i++) {
1625 (*regs)[i] = tswapreg(env->gregs[i]);
1626 }
1627
1628 (*regs)[TARGET_REG_PC] = tswapreg(env->pc);
1629 (*regs)[TARGET_REG_PR] = tswapreg(env->pr);
1630 (*regs)[TARGET_REG_SR] = tswapreg(env->sr);
1631 (*regs)[TARGET_REG_GBR] = tswapreg(env->gbr);
1632 (*regs)[TARGET_REG_MACH] = tswapreg(env->mach);
1633 (*regs)[TARGET_REG_MACL] = tswapreg(env->macl);
1634 (*regs)[TARGET_REG_SYSCALL] = 0; /* FIXME */
1635 }
1636
1637 #define USE_ELF_CORE_DUMP
1638 #define ELF_EXEC_PAGESIZE 4096
1639
1640 enum {
1641 SH_CPU_HAS_FPU = 0x0001, /* Hardware FPU support */
1642 SH_CPU_HAS_P2_FLUSH_BUG = 0x0002, /* Need to flush the cache in P2 area */
1643 SH_CPU_HAS_MMU_PAGE_ASSOC = 0x0004, /* SH3: TLB way selection bit support */
1644 SH_CPU_HAS_DSP = 0x0008, /* SH-DSP: DSP support */
1645 SH_CPU_HAS_PERF_COUNTER = 0x0010, /* Hardware performance counters */
1646 SH_CPU_HAS_PTEA = 0x0020, /* PTEA register */
1647 SH_CPU_HAS_LLSC = 0x0040, /* movli.l/movco.l */
1648 SH_CPU_HAS_L2_CACHE = 0x0080, /* Secondary cache / URAM */
1649 SH_CPU_HAS_OP32 = 0x0100, /* 32-bit instruction support */
1650 SH_CPU_HAS_PTEAEX = 0x0200, /* PTE ASID Extension support */
1651 };
1652
1653 #define ELF_HWCAP get_elf_hwcap()
1654
get_elf_hwcap(void)1655 static uint32_t get_elf_hwcap(void)
1656 {
1657 SuperHCPU *cpu = SUPERH_CPU(thread_cpu);
1658 uint32_t hwcap = 0;
1659
1660 hwcap |= SH_CPU_HAS_FPU;
1661
1662 if (cpu->env.features & SH_FEATURE_SH4A) {
1663 hwcap |= SH_CPU_HAS_LLSC;
1664 }
1665
1666 return hwcap;
1667 }
1668
1669 #endif
1670
1671 #ifdef TARGET_M68K
1672
1673 #define ELF_CLASS ELFCLASS32
1674 #define ELF_ARCH EM_68K
1675
1676 /* ??? Does this need to do anything?
1677 #define ELF_PLAT_INIT(_r) */
1678
init_thread(struct target_pt_regs * regs,struct image_info * infop)1679 static inline void init_thread(struct target_pt_regs *regs,
1680 struct image_info *infop)
1681 {
1682 regs->usp = infop->start_stack;
1683 regs->sr = 0;
1684 regs->pc = infop->entry;
1685 }
1686
1687 /* See linux kernel: arch/m68k/include/asm/elf.h. */
1688 #define ELF_NREG 20
1689 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1690
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUM68KState * env)1691 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUM68KState *env)
1692 {
1693 (*regs)[0] = tswapreg(env->dregs[1]);
1694 (*regs)[1] = tswapreg(env->dregs[2]);
1695 (*regs)[2] = tswapreg(env->dregs[3]);
1696 (*regs)[3] = tswapreg(env->dregs[4]);
1697 (*regs)[4] = tswapreg(env->dregs[5]);
1698 (*regs)[5] = tswapreg(env->dregs[6]);
1699 (*regs)[6] = tswapreg(env->dregs[7]);
1700 (*regs)[7] = tswapreg(env->aregs[0]);
1701 (*regs)[8] = tswapreg(env->aregs[1]);
1702 (*regs)[9] = tswapreg(env->aregs[2]);
1703 (*regs)[10] = tswapreg(env->aregs[3]);
1704 (*regs)[11] = tswapreg(env->aregs[4]);
1705 (*regs)[12] = tswapreg(env->aregs[5]);
1706 (*regs)[13] = tswapreg(env->aregs[6]);
1707 (*regs)[14] = tswapreg(env->dregs[0]);
1708 (*regs)[15] = tswapreg(env->aregs[7]);
1709 (*regs)[16] = tswapreg(env->dregs[0]); /* FIXME: orig_d0 */
1710 (*regs)[17] = tswapreg(env->sr);
1711 (*regs)[18] = tswapreg(env->pc);
1712 (*regs)[19] = 0; /* FIXME: regs->format | regs->vector */
1713 }
1714
1715 #define USE_ELF_CORE_DUMP
1716 #define ELF_EXEC_PAGESIZE 8192
1717
1718 #endif
1719
1720 #ifdef TARGET_ALPHA
1721
1722 #define ELF_CLASS ELFCLASS64
1723 #define ELF_ARCH EM_ALPHA
1724
init_thread(struct target_pt_regs * regs,struct image_info * infop)1725 static inline void init_thread(struct target_pt_regs *regs,
1726 struct image_info *infop)
1727 {
1728 regs->pc = infop->entry;
1729 regs->ps = 8;
1730 regs->usp = infop->start_stack;
1731 }
1732
1733 #define ELF_EXEC_PAGESIZE 8192
1734
1735 #endif /* TARGET_ALPHA */
1736
1737 #ifdef TARGET_S390X
1738
1739 #define ELF_CLASS ELFCLASS64
1740 #define ELF_DATA ELFDATA2MSB
1741 #define ELF_ARCH EM_S390
1742
1743 #include "elf.h"
1744
1745 #define ELF_HWCAP get_elf_hwcap()
1746
1747 #define GET_FEATURE(_feat, _hwcap) \
1748 do { if (s390_has_feat(_feat)) { hwcap |= _hwcap; } } while (0)
1749
get_elf_hwcap(void)1750 uint32_t get_elf_hwcap(void)
1751 {
1752 /*
1753 * Let's assume we always have esan3 and zarch.
1754 * 31-bit processes can use 64-bit registers (high gprs).
1755 */
1756 uint32_t hwcap = HWCAP_S390_ESAN3 | HWCAP_S390_ZARCH | HWCAP_S390_HIGH_GPRS;
1757
1758 GET_FEATURE(S390_FEAT_STFLE, HWCAP_S390_STFLE);
1759 GET_FEATURE(S390_FEAT_MSA, HWCAP_S390_MSA);
1760 GET_FEATURE(S390_FEAT_LONG_DISPLACEMENT, HWCAP_S390_LDISP);
1761 GET_FEATURE(S390_FEAT_EXTENDED_IMMEDIATE, HWCAP_S390_EIMM);
1762 if (s390_has_feat(S390_FEAT_EXTENDED_TRANSLATION_3) &&
1763 s390_has_feat(S390_FEAT_ETF3_ENH)) {
1764 hwcap |= HWCAP_S390_ETF3EH;
1765 }
1766 GET_FEATURE(S390_FEAT_VECTOR, HWCAP_S390_VXRS);
1767 GET_FEATURE(S390_FEAT_VECTOR_ENH, HWCAP_S390_VXRS_EXT);
1768 GET_FEATURE(S390_FEAT_VECTOR_ENH2, HWCAP_S390_VXRS_EXT2);
1769
1770 return hwcap;
1771 }
1772
elf_hwcap_str(uint32_t bit)1773 const char *elf_hwcap_str(uint32_t bit)
1774 {
1775 static const char *hwcap_str[] = {
1776 [HWCAP_S390_NR_ESAN3] = "esan3",
1777 [HWCAP_S390_NR_ZARCH] = "zarch",
1778 [HWCAP_S390_NR_STFLE] = "stfle",
1779 [HWCAP_S390_NR_MSA] = "msa",
1780 [HWCAP_S390_NR_LDISP] = "ldisp",
1781 [HWCAP_S390_NR_EIMM] = "eimm",
1782 [HWCAP_S390_NR_DFP] = "dfp",
1783 [HWCAP_S390_NR_HPAGE] = "edat",
1784 [HWCAP_S390_NR_ETF3EH] = "etf3eh",
1785 [HWCAP_S390_NR_HIGH_GPRS] = "highgprs",
1786 [HWCAP_S390_NR_TE] = "te",
1787 [HWCAP_S390_NR_VXRS] = "vx",
1788 [HWCAP_S390_NR_VXRS_BCD] = "vxd",
1789 [HWCAP_S390_NR_VXRS_EXT] = "vxe",
1790 [HWCAP_S390_NR_GS] = "gs",
1791 [HWCAP_S390_NR_VXRS_EXT2] = "vxe2",
1792 [HWCAP_S390_NR_VXRS_PDE] = "vxp",
1793 [HWCAP_S390_NR_SORT] = "sort",
1794 [HWCAP_S390_NR_DFLT] = "dflt",
1795 [HWCAP_S390_NR_NNPA] = "nnpa",
1796 [HWCAP_S390_NR_PCI_MIO] = "pcimio",
1797 [HWCAP_S390_NR_SIE] = "sie",
1798 };
1799
1800 return bit < ARRAY_SIZE(hwcap_str) ? hwcap_str[bit] : NULL;
1801 }
1802
init_thread(struct target_pt_regs * regs,struct image_info * infop)1803 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop)
1804 {
1805 regs->psw.addr = infop->entry;
1806 regs->psw.mask = PSW_MASK_DAT | PSW_MASK_IO | PSW_MASK_EXT | \
1807 PSW_MASK_MCHECK | PSW_MASK_PSTATE | PSW_MASK_64 | \
1808 PSW_MASK_32;
1809 regs->gprs[15] = infop->start_stack;
1810 }
1811
1812 /* See linux kernel: arch/s390/include/uapi/asm/ptrace.h (s390_regs). */
1813 #define ELF_NREG 27
1814 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1815
1816 enum {
1817 TARGET_REG_PSWM = 0,
1818 TARGET_REG_PSWA = 1,
1819 TARGET_REG_GPRS = 2,
1820 TARGET_REG_ARS = 18,
1821 TARGET_REG_ORIG_R2 = 26,
1822 };
1823
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUS390XState * env)1824 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1825 const CPUS390XState *env)
1826 {
1827 int i;
1828 uint32_t *aregs;
1829
1830 (*regs)[TARGET_REG_PSWM] = tswapreg(env->psw.mask);
1831 (*regs)[TARGET_REG_PSWA] = tswapreg(env->psw.addr);
1832 for (i = 0; i < 16; i++) {
1833 (*regs)[TARGET_REG_GPRS + i] = tswapreg(env->regs[i]);
1834 }
1835 aregs = (uint32_t *)&((*regs)[TARGET_REG_ARS]);
1836 for (i = 0; i < 16; i++) {
1837 aregs[i] = tswap32(env->aregs[i]);
1838 }
1839 (*regs)[TARGET_REG_ORIG_R2] = 0;
1840 }
1841
1842 #define USE_ELF_CORE_DUMP
1843 #define ELF_EXEC_PAGESIZE 4096
1844
1845 #define VDSO_HEADER "vdso.c.inc"
1846
1847 #endif /* TARGET_S390X */
1848
1849 #ifdef TARGET_RISCV
1850
1851 #define ELF_ARCH EM_RISCV
1852
1853 #ifdef TARGET_RISCV32
1854 #define ELF_CLASS ELFCLASS32
1855 #define VDSO_HEADER "vdso-32.c.inc"
1856 #else
1857 #define ELF_CLASS ELFCLASS64
1858 #define VDSO_HEADER "vdso-64.c.inc"
1859 #endif
1860
1861 #define ELF_HWCAP get_elf_hwcap()
1862
get_elf_hwcap(void)1863 static uint32_t get_elf_hwcap(void)
1864 {
1865 #define MISA_BIT(EXT) (1 << (EXT - 'A'))
1866 RISCVCPU *cpu = RISCV_CPU(thread_cpu);
1867 uint32_t mask = MISA_BIT('I') | MISA_BIT('M') | MISA_BIT('A')
1868 | MISA_BIT('F') | MISA_BIT('D') | MISA_BIT('C')
1869 | MISA_BIT('V');
1870
1871 return cpu->env.misa_ext & mask;
1872 #undef MISA_BIT
1873 }
1874
init_thread(struct target_pt_regs * regs,struct image_info * infop)1875 static inline void init_thread(struct target_pt_regs *regs,
1876 struct image_info *infop)
1877 {
1878 regs->sepc = infop->entry;
1879 regs->sp = infop->start_stack;
1880 }
1881
1882 #define ELF_EXEC_PAGESIZE 4096
1883
1884 #endif /* TARGET_RISCV */
1885
1886 #ifdef TARGET_HPPA
1887
1888 #define ELF_CLASS ELFCLASS32
1889 #define ELF_ARCH EM_PARISC
1890 #define ELF_PLATFORM "PARISC"
1891 #define STACK_GROWS_DOWN 0
1892 #define STACK_ALIGNMENT 64
1893
1894 #define VDSO_HEADER "vdso.c.inc"
1895
init_thread(struct target_pt_regs * regs,struct image_info * infop)1896 static inline void init_thread(struct target_pt_regs *regs,
1897 struct image_info *infop)
1898 {
1899 regs->iaoq[0] = infop->entry | PRIV_USER;
1900 regs->iaoq[1] = regs->iaoq[0] + 4;
1901 regs->gr[23] = 0;
1902 regs->gr[24] = infop->argv;
1903 regs->gr[25] = infop->argc;
1904 /* The top-of-stack contains a linkage buffer. */
1905 regs->gr[30] = infop->start_stack + 64;
1906 regs->gr[31] = infop->entry;
1907 }
1908
1909 #define LO_COMMPAGE 0
1910
init_guest_commpage(void)1911 static bool init_guest_commpage(void)
1912 {
1913 /* If reserved_va, then we have already mapped 0 page on the host. */
1914 if (!reserved_va) {
1915 void *want, *addr;
1916
1917 want = g2h_untagged(LO_COMMPAGE);
1918 addr = mmap(want, TARGET_PAGE_SIZE, PROT_NONE,
1919 MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED_NOREPLACE, -1, 0);
1920 if (addr == MAP_FAILED) {
1921 perror("Allocating guest commpage");
1922 exit(EXIT_FAILURE);
1923 }
1924 if (addr != want) {
1925 return false;
1926 }
1927 }
1928
1929 /*
1930 * On Linux, page zero is normally marked execute only + gateway.
1931 * Normal read or write is supposed to fail (thus PROT_NONE above),
1932 * but specific offsets have kernel code mapped to raise permissions
1933 * and implement syscalls. Here, simply mark the page executable.
1934 * Special case the entry points during translation (see do_page_zero).
1935 */
1936 page_set_flags(LO_COMMPAGE, LO_COMMPAGE | ~TARGET_PAGE_MASK,
1937 PAGE_EXEC | PAGE_VALID);
1938 return true;
1939 }
1940
1941 #endif /* TARGET_HPPA */
1942
1943 #ifdef TARGET_XTENSA
1944
1945 #define ELF_CLASS ELFCLASS32
1946 #define ELF_ARCH EM_XTENSA
1947
init_thread(struct target_pt_regs * regs,struct image_info * infop)1948 static inline void init_thread(struct target_pt_regs *regs,
1949 struct image_info *infop)
1950 {
1951 regs->windowbase = 0;
1952 regs->windowstart = 1;
1953 regs->areg[1] = infop->start_stack;
1954 regs->pc = infop->entry;
1955 if (info_is_fdpic(infop)) {
1956 regs->areg[4] = infop->loadmap_addr;
1957 regs->areg[5] = infop->interpreter_loadmap_addr;
1958 if (infop->interpreter_loadmap_addr) {
1959 regs->areg[6] = infop->interpreter_pt_dynamic_addr;
1960 } else {
1961 regs->areg[6] = infop->pt_dynamic_addr;
1962 }
1963 }
1964 }
1965
1966 /* See linux kernel: arch/xtensa/include/asm/elf.h. */
1967 #define ELF_NREG 128
1968 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1969
1970 enum {
1971 TARGET_REG_PC,
1972 TARGET_REG_PS,
1973 TARGET_REG_LBEG,
1974 TARGET_REG_LEND,
1975 TARGET_REG_LCOUNT,
1976 TARGET_REG_SAR,
1977 TARGET_REG_WINDOWSTART,
1978 TARGET_REG_WINDOWBASE,
1979 TARGET_REG_THREADPTR,
1980 TARGET_REG_AR0 = 64,
1981 };
1982
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUXtensaState * env)1983 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1984 const CPUXtensaState *env)
1985 {
1986 unsigned i;
1987
1988 (*regs)[TARGET_REG_PC] = tswapreg(env->pc);
1989 (*regs)[TARGET_REG_PS] = tswapreg(env->sregs[PS] & ~PS_EXCM);
1990 (*regs)[TARGET_REG_LBEG] = tswapreg(env->sregs[LBEG]);
1991 (*regs)[TARGET_REG_LEND] = tswapreg(env->sregs[LEND]);
1992 (*regs)[TARGET_REG_LCOUNT] = tswapreg(env->sregs[LCOUNT]);
1993 (*regs)[TARGET_REG_SAR] = tswapreg(env->sregs[SAR]);
1994 (*regs)[TARGET_REG_WINDOWSTART] = tswapreg(env->sregs[WINDOW_START]);
1995 (*regs)[TARGET_REG_WINDOWBASE] = tswapreg(env->sregs[WINDOW_BASE]);
1996 (*regs)[TARGET_REG_THREADPTR] = tswapreg(env->uregs[THREADPTR]);
1997 xtensa_sync_phys_from_window((CPUXtensaState *)env);
1998 for (i = 0; i < env->config->nareg; ++i) {
1999 (*regs)[TARGET_REG_AR0 + i] = tswapreg(env->phys_regs[i]);
2000 }
2001 }
2002
2003 #define USE_ELF_CORE_DUMP
2004 #define ELF_EXEC_PAGESIZE 4096
2005
2006 #endif /* TARGET_XTENSA */
2007
2008 #ifdef TARGET_HEXAGON
2009
2010 #define ELF_CLASS ELFCLASS32
2011 #define ELF_ARCH EM_HEXAGON
2012
init_thread(struct target_pt_regs * regs,struct image_info * infop)2013 static inline void init_thread(struct target_pt_regs *regs,
2014 struct image_info *infop)
2015 {
2016 regs->sepc = infop->entry;
2017 regs->sp = infop->start_stack;
2018 }
2019
2020 #endif /* TARGET_HEXAGON */
2021
2022 #ifndef ELF_BASE_PLATFORM
2023 #define ELF_BASE_PLATFORM (NULL)
2024 #endif
2025
2026 #ifndef ELF_PLATFORM
2027 #define ELF_PLATFORM (NULL)
2028 #endif
2029
2030 #ifndef ELF_MACHINE
2031 #define ELF_MACHINE ELF_ARCH
2032 #endif
2033
2034 #ifndef elf_check_arch
2035 #define elf_check_arch(x) ((x) == ELF_ARCH)
2036 #endif
2037
2038 #ifndef elf_check_abi
2039 #define elf_check_abi(x) (1)
2040 #endif
2041
2042 #ifndef ELF_HWCAP
2043 #define ELF_HWCAP 0
2044 #endif
2045
2046 #ifndef STACK_GROWS_DOWN
2047 #define STACK_GROWS_DOWN 1
2048 #endif
2049
2050 #ifndef STACK_ALIGNMENT
2051 #define STACK_ALIGNMENT 16
2052 #endif
2053
2054 #ifdef TARGET_ABI32
2055 #undef ELF_CLASS
2056 #define ELF_CLASS ELFCLASS32
2057 #undef bswaptls
2058 #define bswaptls(ptr) bswap32s(ptr)
2059 #endif
2060
2061 #ifndef EXSTACK_DEFAULT
2062 #define EXSTACK_DEFAULT false
2063 #endif
2064
2065 #include "elf.h"
2066
2067 /* We must delay the following stanzas until after "elf.h". */
2068 #if defined(TARGET_AARCH64)
2069
arch_parse_elf_property(uint32_t pr_type,uint32_t pr_datasz,const uint32_t * data,struct image_info * info,Error ** errp)2070 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz,
2071 const uint32_t *data,
2072 struct image_info *info,
2073 Error **errp)
2074 {
2075 if (pr_type == GNU_PROPERTY_AARCH64_FEATURE_1_AND) {
2076 if (pr_datasz != sizeof(uint32_t)) {
2077 error_setg(errp, "Ill-formed GNU_PROPERTY_AARCH64_FEATURE_1_AND");
2078 return false;
2079 }
2080 /* We will extract GNU_PROPERTY_AARCH64_FEATURE_1_BTI later. */
2081 info->note_flags = *data;
2082 }
2083 return true;
2084 }
2085 #define ARCH_USE_GNU_PROPERTY 1
2086
2087 #else
2088
arch_parse_elf_property(uint32_t pr_type,uint32_t pr_datasz,const uint32_t * data,struct image_info * info,Error ** errp)2089 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz,
2090 const uint32_t *data,
2091 struct image_info *info,
2092 Error **errp)
2093 {
2094 g_assert_not_reached();
2095 }
2096 #define ARCH_USE_GNU_PROPERTY 0
2097
2098 #endif
2099
2100 struct exec
2101 {
2102 unsigned int a_info; /* Use macros N_MAGIC, etc for access */
2103 unsigned int a_text; /* length of text, in bytes */
2104 unsigned int a_data; /* length of data, in bytes */
2105 unsigned int a_bss; /* length of uninitialized data area, in bytes */
2106 unsigned int a_syms; /* length of symbol table data in file, in bytes */
2107 unsigned int a_entry; /* start address */
2108 unsigned int a_trsize; /* length of relocation info for text, in bytes */
2109 unsigned int a_drsize; /* length of relocation info for data, in bytes */
2110 };
2111
2112
2113 #define N_MAGIC(exec) ((exec).a_info & 0xffff)
2114 #define OMAGIC 0407
2115 #define NMAGIC 0410
2116 #define ZMAGIC 0413
2117 #define QMAGIC 0314
2118
2119 #define DLINFO_ITEMS 16
2120
memcpy_fromfs(void * to,const void * from,unsigned long n)2121 static inline void memcpy_fromfs(void * to, const void * from, unsigned long n)
2122 {
2123 memcpy(to, from, n);
2124 }
2125
bswap_ehdr(struct elfhdr * ehdr)2126 static void bswap_ehdr(struct elfhdr *ehdr)
2127 {
2128 if (!target_needs_bswap()) {
2129 return;
2130 }
2131
2132 bswap16s(&ehdr->e_type); /* Object file type */
2133 bswap16s(&ehdr->e_machine); /* Architecture */
2134 bswap32s(&ehdr->e_version); /* Object file version */
2135 bswaptls(&ehdr->e_entry); /* Entry point virtual address */
2136 bswaptls(&ehdr->e_phoff); /* Program header table file offset */
2137 bswaptls(&ehdr->e_shoff); /* Section header table file offset */
2138 bswap32s(&ehdr->e_flags); /* Processor-specific flags */
2139 bswap16s(&ehdr->e_ehsize); /* ELF header size in bytes */
2140 bswap16s(&ehdr->e_phentsize); /* Program header table entry size */
2141 bswap16s(&ehdr->e_phnum); /* Program header table entry count */
2142 bswap16s(&ehdr->e_shentsize); /* Section header table entry size */
2143 bswap16s(&ehdr->e_shnum); /* Section header table entry count */
2144 bswap16s(&ehdr->e_shstrndx); /* Section header string table index */
2145 }
2146
bswap_phdr(struct elf_phdr * phdr,int phnum)2147 static void bswap_phdr(struct elf_phdr *phdr, int phnum)
2148 {
2149 if (!target_needs_bswap()) {
2150 return;
2151 }
2152
2153 for (int i = 0; i < phnum; ++i, ++phdr) {
2154 bswap32s(&phdr->p_type); /* Segment type */
2155 bswap32s(&phdr->p_flags); /* Segment flags */
2156 bswaptls(&phdr->p_offset); /* Segment file offset */
2157 bswaptls(&phdr->p_vaddr); /* Segment virtual address */
2158 bswaptls(&phdr->p_paddr); /* Segment physical address */
2159 bswaptls(&phdr->p_filesz); /* Segment size in file */
2160 bswaptls(&phdr->p_memsz); /* Segment size in memory */
2161 bswaptls(&phdr->p_align); /* Segment alignment */
2162 }
2163 }
2164
bswap_shdr(struct elf_shdr * shdr,int shnum)2165 static void bswap_shdr(struct elf_shdr *shdr, int shnum)
2166 {
2167 if (!target_needs_bswap()) {
2168 return;
2169 }
2170
2171 for (int i = 0; i < shnum; ++i, ++shdr) {
2172 bswap32s(&shdr->sh_name);
2173 bswap32s(&shdr->sh_type);
2174 bswaptls(&shdr->sh_flags);
2175 bswaptls(&shdr->sh_addr);
2176 bswaptls(&shdr->sh_offset);
2177 bswaptls(&shdr->sh_size);
2178 bswap32s(&shdr->sh_link);
2179 bswap32s(&shdr->sh_info);
2180 bswaptls(&shdr->sh_addralign);
2181 bswaptls(&shdr->sh_entsize);
2182 }
2183 }
2184
bswap_sym(struct elf_sym * sym)2185 static void bswap_sym(struct elf_sym *sym)
2186 {
2187 if (!target_needs_bswap()) {
2188 return;
2189 }
2190
2191 bswap32s(&sym->st_name);
2192 bswaptls(&sym->st_value);
2193 bswaptls(&sym->st_size);
2194 bswap16s(&sym->st_shndx);
2195 }
2196
2197 #ifdef TARGET_MIPS
bswap_mips_abiflags(Mips_elf_abiflags_v0 * abiflags)2198 static void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags)
2199 {
2200 if (!target_needs_bswap()) {
2201 return;
2202 }
2203
2204 bswap16s(&abiflags->version);
2205 bswap32s(&abiflags->ases);
2206 bswap32s(&abiflags->isa_ext);
2207 bswap32s(&abiflags->flags1);
2208 bswap32s(&abiflags->flags2);
2209 }
2210 #endif
2211
2212 #ifdef USE_ELF_CORE_DUMP
2213 static int elf_core_dump(int, const CPUArchState *);
2214 #endif /* USE_ELF_CORE_DUMP */
2215 static void load_symbols(struct elfhdr *hdr, const ImageSource *src,
2216 abi_ulong load_bias);
2217
2218 /* Verify the portions of EHDR within E_IDENT for the target.
2219 This can be performed before bswapping the entire header. */
elf_check_ident(struct elfhdr * ehdr)2220 static bool elf_check_ident(struct elfhdr *ehdr)
2221 {
2222 return (ehdr->e_ident[EI_MAG0] == ELFMAG0
2223 && ehdr->e_ident[EI_MAG1] == ELFMAG1
2224 && ehdr->e_ident[EI_MAG2] == ELFMAG2
2225 && ehdr->e_ident[EI_MAG3] == ELFMAG3
2226 && ehdr->e_ident[EI_CLASS] == ELF_CLASS
2227 && ehdr->e_ident[EI_DATA] == ELF_DATA
2228 && ehdr->e_ident[EI_VERSION] == EV_CURRENT);
2229 }
2230
2231 /* Verify the portions of EHDR outside of E_IDENT for the target.
2232 This has to wait until after bswapping the header. */
elf_check_ehdr(struct elfhdr * ehdr)2233 static bool elf_check_ehdr(struct elfhdr *ehdr)
2234 {
2235 return (elf_check_arch(ehdr->e_machine)
2236 && elf_check_abi(ehdr->e_flags)
2237 && ehdr->e_ehsize == sizeof(struct elfhdr)
2238 && ehdr->e_phentsize == sizeof(struct elf_phdr)
2239 && (ehdr->e_type == ET_EXEC || ehdr->e_type == ET_DYN));
2240 }
2241
2242 /*
2243 * 'copy_elf_strings()' copies argument/envelope strings from user
2244 * memory to free pages in kernel mem. These are in a format ready
2245 * to be put directly into the top of new user memory.
2246 *
2247 */
copy_elf_strings(int argc,char ** argv,char * scratch,abi_ulong p,abi_ulong stack_limit)2248 static abi_ulong copy_elf_strings(int argc, char **argv, char *scratch,
2249 abi_ulong p, abi_ulong stack_limit)
2250 {
2251 char *tmp;
2252 int len, i;
2253 abi_ulong top = p;
2254
2255 if (!p) {
2256 return 0; /* bullet-proofing */
2257 }
2258
2259 if (STACK_GROWS_DOWN) {
2260 int offset = ((p - 1) % TARGET_PAGE_SIZE) + 1;
2261 for (i = argc - 1; i >= 0; --i) {
2262 tmp = argv[i];
2263 if (!tmp) {
2264 fprintf(stderr, "VFS: argc is wrong");
2265 exit(-1);
2266 }
2267 len = strlen(tmp) + 1;
2268 tmp += len;
2269
2270 if (len > (p - stack_limit)) {
2271 return 0;
2272 }
2273 while (len) {
2274 int bytes_to_copy = (len > offset) ? offset : len;
2275 tmp -= bytes_to_copy;
2276 p -= bytes_to_copy;
2277 offset -= bytes_to_copy;
2278 len -= bytes_to_copy;
2279
2280 memcpy_fromfs(scratch + offset, tmp, bytes_to_copy);
2281
2282 if (offset == 0) {
2283 memcpy_to_target(p, scratch, top - p);
2284 top = p;
2285 offset = TARGET_PAGE_SIZE;
2286 }
2287 }
2288 }
2289 if (p != top) {
2290 memcpy_to_target(p, scratch + offset, top - p);
2291 }
2292 } else {
2293 int remaining = TARGET_PAGE_SIZE - (p % TARGET_PAGE_SIZE);
2294 for (i = 0; i < argc; ++i) {
2295 tmp = argv[i];
2296 if (!tmp) {
2297 fprintf(stderr, "VFS: argc is wrong");
2298 exit(-1);
2299 }
2300 len = strlen(tmp) + 1;
2301 if (len > (stack_limit - p)) {
2302 return 0;
2303 }
2304 while (len) {
2305 int bytes_to_copy = (len > remaining) ? remaining : len;
2306
2307 memcpy_fromfs(scratch + (p - top), tmp, bytes_to_copy);
2308
2309 tmp += bytes_to_copy;
2310 remaining -= bytes_to_copy;
2311 p += bytes_to_copy;
2312 len -= bytes_to_copy;
2313
2314 if (remaining == 0) {
2315 memcpy_to_target(top, scratch, p - top);
2316 top = p;
2317 remaining = TARGET_PAGE_SIZE;
2318 }
2319 }
2320 }
2321 if (p != top) {
2322 memcpy_to_target(top, scratch, p - top);
2323 }
2324 }
2325
2326 return p;
2327 }
2328
2329 /* Older linux kernels provide up to MAX_ARG_PAGES (default: 32) of
2330 * argument/environment space. Newer kernels (>2.6.33) allow more,
2331 * dependent on stack size, but guarantee at least 32 pages for
2332 * backwards compatibility.
2333 */
2334 #define STACK_LOWER_LIMIT (32 * TARGET_PAGE_SIZE)
2335
setup_arg_pages(struct linux_binprm * bprm,struct image_info * info)2336 static abi_ulong setup_arg_pages(struct linux_binprm *bprm,
2337 struct image_info *info)
2338 {
2339 abi_ulong size, error, guard;
2340 int prot;
2341
2342 size = guest_stack_size;
2343 if (size < STACK_LOWER_LIMIT) {
2344 size = STACK_LOWER_LIMIT;
2345 }
2346
2347 if (STACK_GROWS_DOWN) {
2348 guard = TARGET_PAGE_SIZE;
2349 if (guard < qemu_real_host_page_size()) {
2350 guard = qemu_real_host_page_size();
2351 }
2352 } else {
2353 /* no guard page for hppa target where stack grows upwards. */
2354 guard = 0;
2355 }
2356
2357 prot = PROT_READ | PROT_WRITE;
2358 if (info->exec_stack) {
2359 prot |= PROT_EXEC;
2360 }
2361 error = target_mmap(0, size + guard, prot,
2362 MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
2363 if (error == -1) {
2364 perror("mmap stack");
2365 exit(-1);
2366 }
2367
2368 /* We reserve one extra page at the top of the stack as guard. */
2369 if (STACK_GROWS_DOWN) {
2370 target_mprotect(error, guard, PROT_NONE);
2371 info->stack_limit = error + guard;
2372 return info->stack_limit + size - sizeof(void *);
2373 } else {
2374 info->stack_limit = error + size;
2375 return error;
2376 }
2377 }
2378
2379 /**
2380 * zero_bss:
2381 *
2382 * Map and zero the bss. We need to explicitly zero any fractional pages
2383 * after the data section (i.e. bss). Return false on mapping failure.
2384 */
zero_bss(abi_ulong start_bss,abi_ulong end_bss,int prot,Error ** errp)2385 static bool zero_bss(abi_ulong start_bss, abi_ulong end_bss,
2386 int prot, Error **errp)
2387 {
2388 abi_ulong align_bss;
2389
2390 /* We only expect writable bss; the code segment shouldn't need this. */
2391 if (!(prot & PROT_WRITE)) {
2392 error_setg(errp, "PT_LOAD with non-writable bss");
2393 return false;
2394 }
2395
2396 align_bss = TARGET_PAGE_ALIGN(start_bss);
2397 end_bss = TARGET_PAGE_ALIGN(end_bss);
2398
2399 if (start_bss < align_bss) {
2400 int flags = page_get_flags(start_bss);
2401
2402 if (!(flags & PAGE_RWX)) {
2403 /*
2404 * The whole address space of the executable was reserved
2405 * at the start, therefore all pages will be VALID.
2406 * But assuming there are no PROT_NONE PT_LOAD segments,
2407 * a PROT_NONE page means no data all bss, and we can
2408 * simply extend the new anon mapping back to the start
2409 * of the page of bss.
2410 */
2411 align_bss -= TARGET_PAGE_SIZE;
2412 } else {
2413 /*
2414 * The start of the bss shares a page with something.
2415 * The only thing that we expect is the data section,
2416 * which would already be marked writable.
2417 * Overlapping the RX code segment seems malformed.
2418 */
2419 if (!(flags & PAGE_WRITE)) {
2420 error_setg(errp, "PT_LOAD with bss overlapping "
2421 "non-writable page");
2422 return false;
2423 }
2424
2425 /* The page is already mapped and writable. */
2426 memset(g2h_untagged(start_bss), 0, align_bss - start_bss);
2427 }
2428 }
2429
2430 if (align_bss < end_bss &&
2431 target_mmap(align_bss, end_bss - align_bss, prot,
2432 MAP_FIXED | MAP_PRIVATE | MAP_ANON, -1, 0) == -1) {
2433 error_setg_errno(errp, errno, "Error mapping bss");
2434 return false;
2435 }
2436 return true;
2437 }
2438
2439 #if defined(TARGET_ARM)
elf_is_fdpic(struct elfhdr * exec)2440 static int elf_is_fdpic(struct elfhdr *exec)
2441 {
2442 return exec->e_ident[EI_OSABI] == ELFOSABI_ARM_FDPIC;
2443 }
2444 #elif defined(TARGET_XTENSA)
elf_is_fdpic(struct elfhdr * exec)2445 static int elf_is_fdpic(struct elfhdr *exec)
2446 {
2447 return exec->e_ident[EI_OSABI] == ELFOSABI_XTENSA_FDPIC;
2448 }
2449 #else
2450 /* Default implementation, always false. */
elf_is_fdpic(struct elfhdr * exec)2451 static int elf_is_fdpic(struct elfhdr *exec)
2452 {
2453 return 0;
2454 }
2455 #endif
2456
loader_build_fdpic_loadmap(struct image_info * info,abi_ulong sp)2457 static abi_ulong loader_build_fdpic_loadmap(struct image_info *info, abi_ulong sp)
2458 {
2459 uint16_t n;
2460 struct elf32_fdpic_loadseg *loadsegs = info->loadsegs;
2461
2462 /* elf32_fdpic_loadseg */
2463 n = info->nsegs;
2464 while (n--) {
2465 sp -= 12;
2466 put_user_u32(loadsegs[n].addr, sp+0);
2467 put_user_u32(loadsegs[n].p_vaddr, sp+4);
2468 put_user_u32(loadsegs[n].p_memsz, sp+8);
2469 }
2470
2471 /* elf32_fdpic_loadmap */
2472 sp -= 4;
2473 put_user_u16(0, sp+0); /* version */
2474 put_user_u16(info->nsegs, sp+2); /* nsegs */
2475
2476 info->personality = PER_LINUX_FDPIC;
2477 info->loadmap_addr = sp;
2478
2479 return sp;
2480 }
2481
create_elf_tables(abi_ulong p,int argc,int envc,struct elfhdr * exec,struct image_info * info,struct image_info * interp_info,struct image_info * vdso_info)2482 static abi_ulong create_elf_tables(abi_ulong p, int argc, int envc,
2483 struct elfhdr *exec,
2484 struct image_info *info,
2485 struct image_info *interp_info,
2486 struct image_info *vdso_info)
2487 {
2488 abi_ulong sp;
2489 abi_ulong u_argc, u_argv, u_envp, u_auxv;
2490 int size;
2491 int i;
2492 abi_ulong u_rand_bytes;
2493 uint8_t k_rand_bytes[16];
2494 abi_ulong u_platform, u_base_platform;
2495 const char *k_platform, *k_base_platform;
2496 const int n = sizeof(elf_addr_t);
2497
2498 sp = p;
2499
2500 /* Needs to be before we load the env/argc/... */
2501 if (elf_is_fdpic(exec)) {
2502 /* Need 4 byte alignment for these structs */
2503 sp &= ~3;
2504 sp = loader_build_fdpic_loadmap(info, sp);
2505 info->other_info = interp_info;
2506 if (interp_info) {
2507 interp_info->other_info = info;
2508 sp = loader_build_fdpic_loadmap(interp_info, sp);
2509 info->interpreter_loadmap_addr = interp_info->loadmap_addr;
2510 info->interpreter_pt_dynamic_addr = interp_info->pt_dynamic_addr;
2511 } else {
2512 info->interpreter_loadmap_addr = 0;
2513 info->interpreter_pt_dynamic_addr = 0;
2514 }
2515 }
2516
2517 u_base_platform = 0;
2518 k_base_platform = ELF_BASE_PLATFORM;
2519 if (k_base_platform) {
2520 size_t len = strlen(k_base_platform) + 1;
2521 if (STACK_GROWS_DOWN) {
2522 sp -= (len + n - 1) & ~(n - 1);
2523 u_base_platform = sp;
2524 /* FIXME - check return value of memcpy_to_target() for failure */
2525 memcpy_to_target(sp, k_base_platform, len);
2526 } else {
2527 memcpy_to_target(sp, k_base_platform, len);
2528 u_base_platform = sp;
2529 sp += len + 1;
2530 }
2531 }
2532
2533 u_platform = 0;
2534 k_platform = ELF_PLATFORM;
2535 if (k_platform) {
2536 size_t len = strlen(k_platform) + 1;
2537 if (STACK_GROWS_DOWN) {
2538 sp -= (len + n - 1) & ~(n - 1);
2539 u_platform = sp;
2540 /* FIXME - check return value of memcpy_to_target() for failure */
2541 memcpy_to_target(sp, k_platform, len);
2542 } else {
2543 memcpy_to_target(sp, k_platform, len);
2544 u_platform = sp;
2545 sp += len + 1;
2546 }
2547 }
2548
2549 /* Provide 16 byte alignment for the PRNG, and basic alignment for
2550 * the argv and envp pointers.
2551 */
2552 if (STACK_GROWS_DOWN) {
2553 sp = QEMU_ALIGN_DOWN(sp, 16);
2554 } else {
2555 sp = QEMU_ALIGN_UP(sp, 16);
2556 }
2557
2558 /*
2559 * Generate 16 random bytes for userspace PRNG seeding.
2560 */
2561 qemu_guest_getrandom_nofail(k_rand_bytes, sizeof(k_rand_bytes));
2562 if (STACK_GROWS_DOWN) {
2563 sp -= 16;
2564 u_rand_bytes = sp;
2565 /* FIXME - check return value of memcpy_to_target() for failure */
2566 memcpy_to_target(sp, k_rand_bytes, 16);
2567 } else {
2568 memcpy_to_target(sp, k_rand_bytes, 16);
2569 u_rand_bytes = sp;
2570 sp += 16;
2571 }
2572
2573 size = (DLINFO_ITEMS + 1) * 2;
2574 if (k_base_platform) {
2575 size += 2;
2576 }
2577 if (k_platform) {
2578 size += 2;
2579 }
2580 if (vdso_info) {
2581 size += 2;
2582 }
2583 #ifdef DLINFO_ARCH_ITEMS
2584 size += DLINFO_ARCH_ITEMS * 2;
2585 #endif
2586 #ifdef ELF_HWCAP2
2587 size += 2;
2588 #endif
2589 info->auxv_len = size * n;
2590
2591 size += envc + argc + 2;
2592 size += 1; /* argc itself */
2593 size *= n;
2594
2595 /* Allocate space and finalize stack alignment for entry now. */
2596 if (STACK_GROWS_DOWN) {
2597 u_argc = QEMU_ALIGN_DOWN(sp - size, STACK_ALIGNMENT);
2598 sp = u_argc;
2599 } else {
2600 u_argc = sp;
2601 sp = QEMU_ALIGN_UP(sp + size, STACK_ALIGNMENT);
2602 }
2603
2604 u_argv = u_argc + n;
2605 u_envp = u_argv + (argc + 1) * n;
2606 u_auxv = u_envp + (envc + 1) * n;
2607 info->saved_auxv = u_auxv;
2608 info->argc = argc;
2609 info->envc = envc;
2610 info->argv = u_argv;
2611 info->envp = u_envp;
2612
2613 /* This is correct because Linux defines
2614 * elf_addr_t as Elf32_Off / Elf64_Off
2615 */
2616 #define NEW_AUX_ENT(id, val) do { \
2617 put_user_ual(id, u_auxv); u_auxv += n; \
2618 put_user_ual(val, u_auxv); u_auxv += n; \
2619 } while(0)
2620
2621 #ifdef ARCH_DLINFO
2622 /*
2623 * ARCH_DLINFO must come first so platform specific code can enforce
2624 * special alignment requirements on the AUXV if necessary (eg. PPC).
2625 */
2626 ARCH_DLINFO;
2627 #endif
2628 /* There must be exactly DLINFO_ITEMS entries here, or the assert
2629 * on info->auxv_len will trigger.
2630 */
2631 NEW_AUX_ENT(AT_PHDR, (abi_ulong)(info->load_addr + exec->e_phoff));
2632 NEW_AUX_ENT(AT_PHENT, (abi_ulong)(sizeof (struct elf_phdr)));
2633 NEW_AUX_ENT(AT_PHNUM, (abi_ulong)(exec->e_phnum));
2634 NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(TARGET_PAGE_SIZE));
2635 NEW_AUX_ENT(AT_BASE, (abi_ulong)(interp_info ? interp_info->load_addr : 0));
2636 NEW_AUX_ENT(AT_FLAGS, (abi_ulong)0);
2637 NEW_AUX_ENT(AT_ENTRY, info->entry);
2638 NEW_AUX_ENT(AT_UID, (abi_ulong) getuid());
2639 NEW_AUX_ENT(AT_EUID, (abi_ulong) geteuid());
2640 NEW_AUX_ENT(AT_GID, (abi_ulong) getgid());
2641 NEW_AUX_ENT(AT_EGID, (abi_ulong) getegid());
2642 NEW_AUX_ENT(AT_HWCAP, (abi_ulong) ELF_HWCAP);
2643 NEW_AUX_ENT(AT_CLKTCK, (abi_ulong) sysconf(_SC_CLK_TCK));
2644 NEW_AUX_ENT(AT_RANDOM, (abi_ulong) u_rand_bytes);
2645 NEW_AUX_ENT(AT_SECURE, (abi_ulong) qemu_getauxval(AT_SECURE));
2646 NEW_AUX_ENT(AT_EXECFN, info->file_string);
2647
2648 #ifdef ELF_HWCAP2
2649 NEW_AUX_ENT(AT_HWCAP2, (abi_ulong) ELF_HWCAP2);
2650 #endif
2651
2652 if (u_base_platform) {
2653 NEW_AUX_ENT(AT_BASE_PLATFORM, u_base_platform);
2654 }
2655 if (u_platform) {
2656 NEW_AUX_ENT(AT_PLATFORM, u_platform);
2657 }
2658 if (vdso_info) {
2659 NEW_AUX_ENT(AT_SYSINFO_EHDR, vdso_info->load_addr);
2660 }
2661 NEW_AUX_ENT (AT_NULL, 0);
2662 #undef NEW_AUX_ENT
2663
2664 /* Check that our initial calculation of the auxv length matches how much
2665 * we actually put into it.
2666 */
2667 assert(info->auxv_len == u_auxv - info->saved_auxv);
2668
2669 put_user_ual(argc, u_argc);
2670
2671 p = info->arg_strings;
2672 for (i = 0; i < argc; ++i) {
2673 put_user_ual(p, u_argv);
2674 u_argv += n;
2675 p += target_strlen(p) + 1;
2676 }
2677 put_user_ual(0, u_argv);
2678
2679 p = info->env_strings;
2680 for (i = 0; i < envc; ++i) {
2681 put_user_ual(p, u_envp);
2682 u_envp += n;
2683 p += target_strlen(p) + 1;
2684 }
2685 put_user_ual(0, u_envp);
2686
2687 return sp;
2688 }
2689
2690 #if defined(HI_COMMPAGE)
2691 #define LO_COMMPAGE -1
2692 #elif defined(LO_COMMPAGE)
2693 #define HI_COMMPAGE 0
2694 #else
2695 #define HI_COMMPAGE 0
2696 #define LO_COMMPAGE -1
2697 #ifndef INIT_GUEST_COMMPAGE
2698 #define init_guest_commpage() true
2699 #endif
2700 #endif
2701
2702 /**
2703 * pgb_try_mmap:
2704 * @addr: host start address
2705 * @addr_last: host last address
2706 * @keep: do not unmap the probe region
2707 *
2708 * Return 1 if [@addr, @addr_last] is not mapped in the host,
2709 * return 0 if it is not available to map, and -1 on mmap error.
2710 * If @keep, the region is left mapped on success, otherwise unmapped.
2711 */
pgb_try_mmap(uintptr_t addr,uintptr_t addr_last,bool keep)2712 static int pgb_try_mmap(uintptr_t addr, uintptr_t addr_last, bool keep)
2713 {
2714 size_t size = addr_last - addr + 1;
2715 void *p = mmap((void *)addr, size, PROT_NONE,
2716 MAP_ANONYMOUS | MAP_PRIVATE |
2717 MAP_NORESERVE | MAP_FIXED_NOREPLACE, -1, 0);
2718 int ret;
2719
2720 if (p == MAP_FAILED) {
2721 return errno == EEXIST ? 0 : -1;
2722 }
2723 ret = p == (void *)addr;
2724 if (!keep || !ret) {
2725 munmap(p, size);
2726 }
2727 return ret;
2728 }
2729
2730 /**
2731 * pgb_try_mmap_skip_brk(uintptr_t addr, uintptr_t size, uintptr_t brk)
2732 * @addr: host address
2733 * @addr_last: host last address
2734 * @brk: host brk
2735 *
2736 * Like pgb_try_mmap, but additionally reserve some memory following brk.
2737 */
pgb_try_mmap_skip_brk(uintptr_t addr,uintptr_t addr_last,uintptr_t brk,bool keep)2738 static int pgb_try_mmap_skip_brk(uintptr_t addr, uintptr_t addr_last,
2739 uintptr_t brk, bool keep)
2740 {
2741 uintptr_t brk_last = brk + 16 * MiB - 1;
2742
2743 /* Do not map anything close to the host brk. */
2744 if (addr <= brk_last && brk <= addr_last) {
2745 return 0;
2746 }
2747 return pgb_try_mmap(addr, addr_last, keep);
2748 }
2749
2750 /**
2751 * pgb_try_mmap_set:
2752 * @ga: set of guest addrs
2753 * @base: guest_base
2754 * @brk: host brk
2755 *
2756 * Return true if all @ga can be mapped by the host at @base.
2757 * On success, retain the mapping at index 0 for reserved_va.
2758 */
2759
2760 typedef struct PGBAddrs {
2761 uintptr_t bounds[3][2]; /* start/last pairs */
2762 int nbounds;
2763 } PGBAddrs;
2764
pgb_try_mmap_set(const PGBAddrs * ga,uintptr_t base,uintptr_t brk)2765 static bool pgb_try_mmap_set(const PGBAddrs *ga, uintptr_t base, uintptr_t brk)
2766 {
2767 for (int i = ga->nbounds - 1; i >= 0; --i) {
2768 if (pgb_try_mmap_skip_brk(ga->bounds[i][0] + base,
2769 ga->bounds[i][1] + base,
2770 brk, i == 0 && reserved_va) <= 0) {
2771 return false;
2772 }
2773 }
2774 return true;
2775 }
2776
2777 /**
2778 * pgb_addr_set:
2779 * @ga: output set of guest addrs
2780 * @guest_loaddr: guest image low address
2781 * @guest_loaddr: guest image high address
2782 * @identity: create for identity mapping
2783 *
2784 * Fill in @ga with the image, COMMPAGE and NULL page.
2785 */
pgb_addr_set(PGBAddrs * ga,abi_ulong guest_loaddr,abi_ulong guest_hiaddr,bool try_identity)2786 static bool pgb_addr_set(PGBAddrs *ga, abi_ulong guest_loaddr,
2787 abi_ulong guest_hiaddr, bool try_identity)
2788 {
2789 int n;
2790
2791 /*
2792 * With a low commpage, or a guest mapped very low,
2793 * we may not be able to use the identity map.
2794 */
2795 if (try_identity) {
2796 if (LO_COMMPAGE != -1 && LO_COMMPAGE < mmap_min_addr) {
2797 return false;
2798 }
2799 if (guest_loaddr != 0 && guest_loaddr < mmap_min_addr) {
2800 return false;
2801 }
2802 }
2803
2804 memset(ga, 0, sizeof(*ga));
2805 n = 0;
2806
2807 if (reserved_va) {
2808 ga->bounds[n][0] = try_identity ? mmap_min_addr : 0;
2809 ga->bounds[n][1] = reserved_va;
2810 n++;
2811 /* LO_COMMPAGE and NULL handled by reserving from 0. */
2812 } else {
2813 /* Add any LO_COMMPAGE or NULL page. */
2814 if (LO_COMMPAGE != -1) {
2815 ga->bounds[n][0] = 0;
2816 ga->bounds[n][1] = LO_COMMPAGE + TARGET_PAGE_SIZE - 1;
2817 n++;
2818 } else if (!try_identity) {
2819 ga->bounds[n][0] = 0;
2820 ga->bounds[n][1] = TARGET_PAGE_SIZE - 1;
2821 n++;
2822 }
2823
2824 /* Add the guest image for ET_EXEC. */
2825 if (guest_loaddr) {
2826 ga->bounds[n][0] = guest_loaddr;
2827 ga->bounds[n][1] = guest_hiaddr;
2828 n++;
2829 }
2830 }
2831
2832 /*
2833 * Temporarily disable
2834 * "comparison is always false due to limited range of data type"
2835 * due to comparison between unsigned and (possible) 0.
2836 */
2837 #pragma GCC diagnostic push
2838 #pragma GCC diagnostic ignored "-Wtype-limits"
2839
2840 /* Add any HI_COMMPAGE not covered by reserved_va. */
2841 if (reserved_va < HI_COMMPAGE) {
2842 ga->bounds[n][0] = HI_COMMPAGE & qemu_real_host_page_mask();
2843 ga->bounds[n][1] = HI_COMMPAGE + TARGET_PAGE_SIZE - 1;
2844 n++;
2845 }
2846
2847 #pragma GCC diagnostic pop
2848
2849 ga->nbounds = n;
2850 return true;
2851 }
2852
pgb_fail_in_use(const char * image_name)2853 static void pgb_fail_in_use(const char *image_name)
2854 {
2855 error_report("%s: requires virtual address space that is in use "
2856 "(omit the -B option or choose a different value)",
2857 image_name);
2858 exit(EXIT_FAILURE);
2859 }
2860
pgb_fixed(const char * image_name,uintptr_t guest_loaddr,uintptr_t guest_hiaddr,uintptr_t align)2861 static void pgb_fixed(const char *image_name, uintptr_t guest_loaddr,
2862 uintptr_t guest_hiaddr, uintptr_t align)
2863 {
2864 PGBAddrs ga;
2865 uintptr_t brk = (uintptr_t)sbrk(0);
2866
2867 if (!QEMU_IS_ALIGNED(guest_base, align)) {
2868 fprintf(stderr, "Requested guest base %p does not satisfy "
2869 "host minimum alignment (0x%" PRIxPTR ")\n",
2870 (void *)guest_base, align);
2871 exit(EXIT_FAILURE);
2872 }
2873
2874 if (!pgb_addr_set(&ga, guest_loaddr, guest_hiaddr, !guest_base)
2875 || !pgb_try_mmap_set(&ga, guest_base, brk)) {
2876 pgb_fail_in_use(image_name);
2877 }
2878 }
2879
2880 /**
2881 * pgb_find_fallback:
2882 *
2883 * This is a fallback method for finding holes in the host address space
2884 * if we don't have the benefit of being able to access /proc/self/map.
2885 * It can potentially take a very long time as we can only dumbly iterate
2886 * up the host address space seeing if the allocation would work.
2887 */
pgb_find_fallback(const PGBAddrs * ga,uintptr_t align,uintptr_t brk)2888 static uintptr_t pgb_find_fallback(const PGBAddrs *ga, uintptr_t align,
2889 uintptr_t brk)
2890 {
2891 /* TODO: come up with a better estimate of how much to skip. */
2892 uintptr_t skip = sizeof(uintptr_t) == 4 ? MiB : GiB;
2893
2894 for (uintptr_t base = skip; ; base += skip) {
2895 base = ROUND_UP(base, align);
2896 if (pgb_try_mmap_set(ga, base, brk)) {
2897 return base;
2898 }
2899 if (base >= -skip) {
2900 return -1;
2901 }
2902 }
2903 }
2904
pgb_try_itree(const PGBAddrs * ga,uintptr_t base,IntervalTreeRoot * root)2905 static uintptr_t pgb_try_itree(const PGBAddrs *ga, uintptr_t base,
2906 IntervalTreeRoot *root)
2907 {
2908 for (int i = ga->nbounds - 1; i >= 0; --i) {
2909 uintptr_t s = base + ga->bounds[i][0];
2910 uintptr_t l = base + ga->bounds[i][1];
2911 IntervalTreeNode *n;
2912
2913 if (l < s) {
2914 /* Wraparound. Skip to advance S to mmap_min_addr. */
2915 return mmap_min_addr - s;
2916 }
2917
2918 n = interval_tree_iter_first(root, s, l);
2919 if (n != NULL) {
2920 /* Conflict. Skip to advance S to LAST + 1. */
2921 return n->last - s + 1;
2922 }
2923 }
2924 return 0; /* success */
2925 }
2926
pgb_find_itree(const PGBAddrs * ga,IntervalTreeRoot * root,uintptr_t align,uintptr_t brk)2927 static uintptr_t pgb_find_itree(const PGBAddrs *ga, IntervalTreeRoot *root,
2928 uintptr_t align, uintptr_t brk)
2929 {
2930 uintptr_t last = sizeof(uintptr_t) == 4 ? MiB : GiB;
2931 uintptr_t base, skip;
2932
2933 while (true) {
2934 base = ROUND_UP(last, align);
2935 if (base < last) {
2936 return -1;
2937 }
2938
2939 skip = pgb_try_itree(ga, base, root);
2940 if (skip == 0) {
2941 break;
2942 }
2943
2944 last = base + skip;
2945 if (last < base) {
2946 return -1;
2947 }
2948 }
2949
2950 /*
2951 * We've chosen 'base' based on holes in the interval tree,
2952 * but we don't yet know if it is a valid host address.
2953 * Because it is the first matching hole, if the host addresses
2954 * are invalid we know there are no further matches.
2955 */
2956 return pgb_try_mmap_set(ga, base, brk) ? base : -1;
2957 }
2958
pgb_dynamic(const char * image_name,uintptr_t guest_loaddr,uintptr_t guest_hiaddr,uintptr_t align)2959 static void pgb_dynamic(const char *image_name, uintptr_t guest_loaddr,
2960 uintptr_t guest_hiaddr, uintptr_t align)
2961 {
2962 IntervalTreeRoot *root;
2963 uintptr_t brk, ret;
2964 PGBAddrs ga;
2965
2966 /* Try the identity map first. */
2967 if (pgb_addr_set(&ga, guest_loaddr, guest_hiaddr, true)) {
2968 brk = (uintptr_t)sbrk(0);
2969 if (pgb_try_mmap_set(&ga, 0, brk)) {
2970 guest_base = 0;
2971 return;
2972 }
2973 }
2974
2975 /*
2976 * Rebuild the address set for non-identity map.
2977 * This differs in the mapping of the guest NULL page.
2978 */
2979 pgb_addr_set(&ga, guest_loaddr, guest_hiaddr, false);
2980
2981 root = read_self_maps();
2982
2983 /* Read brk after we've read the maps, which will malloc. */
2984 brk = (uintptr_t)sbrk(0);
2985
2986 if (!root) {
2987 ret = pgb_find_fallback(&ga, align, brk);
2988 } else {
2989 /*
2990 * Reserve the area close to the host brk.
2991 * This will be freed with the rest of the tree.
2992 */
2993 IntervalTreeNode *b = g_new0(IntervalTreeNode, 1);
2994 b->start = brk;
2995 b->last = brk + 16 * MiB - 1;
2996 interval_tree_insert(b, root);
2997
2998 ret = pgb_find_itree(&ga, root, align, brk);
2999 free_self_maps(root);
3000 }
3001
3002 if (ret == -1) {
3003 int w = TARGET_LONG_BITS / 4;
3004
3005 error_report("%s: Unable to find a guest_base to satisfy all "
3006 "guest address mapping requirements", image_name);
3007
3008 for (int i = 0; i < ga.nbounds; ++i) {
3009 error_printf(" %0*" PRIx64 "-%0*" PRIx64 "\n",
3010 w, (uint64_t)ga.bounds[i][0],
3011 w, (uint64_t)ga.bounds[i][1]);
3012 }
3013 exit(EXIT_FAILURE);
3014 }
3015 guest_base = ret;
3016 }
3017
probe_guest_base(const char * image_name,abi_ulong guest_loaddr,abi_ulong guest_hiaddr)3018 void probe_guest_base(const char *image_name, abi_ulong guest_loaddr,
3019 abi_ulong guest_hiaddr)
3020 {
3021 /* In order to use host shmat, we must be able to honor SHMLBA. */
3022 uintptr_t align = MAX(SHMLBA, TARGET_PAGE_SIZE);
3023
3024 /* Sanity check the guest binary. */
3025 if (reserved_va) {
3026 if (guest_hiaddr > reserved_va) {
3027 error_report("%s: requires more than reserved virtual "
3028 "address space (0x%" PRIx64 " > 0x%lx)",
3029 image_name, (uint64_t)guest_hiaddr, reserved_va);
3030 exit(EXIT_FAILURE);
3031 }
3032 } else {
3033 if (guest_hiaddr != (uintptr_t)guest_hiaddr) {
3034 error_report("%s: requires more virtual address space "
3035 "than the host can provide (0x%" PRIx64 ")",
3036 image_name, (uint64_t)guest_hiaddr + 1);
3037 exit(EXIT_FAILURE);
3038 }
3039 }
3040
3041 if (have_guest_base) {
3042 pgb_fixed(image_name, guest_loaddr, guest_hiaddr, align);
3043 } else {
3044 pgb_dynamic(image_name, guest_loaddr, guest_hiaddr, align);
3045 }
3046
3047 /* Reserve and initialize the commpage. */
3048 if (!init_guest_commpage()) {
3049 /* We have already probed for the commpage being free. */
3050 g_assert_not_reached();
3051 }
3052
3053 assert(QEMU_IS_ALIGNED(guest_base, align));
3054 qemu_log_mask(CPU_LOG_PAGE, "Locating guest address space "
3055 "@ 0x%" PRIx64 "\n", (uint64_t)guest_base);
3056 }
3057
3058 enum {
3059 /* The string "GNU\0" as a magic number. */
3060 GNU0_MAGIC = const_le32('G' | 'N' << 8 | 'U' << 16),
3061 NOTE_DATA_SZ = 1 * KiB,
3062 NOTE_NAME_SZ = 4,
3063 ELF_GNU_PROPERTY_ALIGN = ELF_CLASS == ELFCLASS32 ? 4 : 8,
3064 };
3065
3066 /*
3067 * Process a single gnu_property entry.
3068 * Return false for error.
3069 */
parse_elf_property(const uint32_t * data,int * off,int datasz,struct image_info * info,bool have_prev_type,uint32_t * prev_type,Error ** errp)3070 static bool parse_elf_property(const uint32_t *data, int *off, int datasz,
3071 struct image_info *info, bool have_prev_type,
3072 uint32_t *prev_type, Error **errp)
3073 {
3074 uint32_t pr_type, pr_datasz, step;
3075
3076 if (*off > datasz || !QEMU_IS_ALIGNED(*off, ELF_GNU_PROPERTY_ALIGN)) {
3077 goto error_data;
3078 }
3079 datasz -= *off;
3080 data += *off / sizeof(uint32_t);
3081
3082 if (datasz < 2 * sizeof(uint32_t)) {
3083 goto error_data;
3084 }
3085 pr_type = data[0];
3086 pr_datasz = data[1];
3087 data += 2;
3088 datasz -= 2 * sizeof(uint32_t);
3089 step = ROUND_UP(pr_datasz, ELF_GNU_PROPERTY_ALIGN);
3090 if (step > datasz) {
3091 goto error_data;
3092 }
3093
3094 /* Properties are supposed to be unique and sorted on pr_type. */
3095 if (have_prev_type && pr_type <= *prev_type) {
3096 if (pr_type == *prev_type) {
3097 error_setg(errp, "Duplicate property in PT_GNU_PROPERTY");
3098 } else {
3099 error_setg(errp, "Unsorted property in PT_GNU_PROPERTY");
3100 }
3101 return false;
3102 }
3103 *prev_type = pr_type;
3104
3105 if (!arch_parse_elf_property(pr_type, pr_datasz, data, info, errp)) {
3106 return false;
3107 }
3108
3109 *off += 2 * sizeof(uint32_t) + step;
3110 return true;
3111
3112 error_data:
3113 error_setg(errp, "Ill-formed property in PT_GNU_PROPERTY");
3114 return false;
3115 }
3116
3117 /* Process NT_GNU_PROPERTY_TYPE_0. */
parse_elf_properties(const ImageSource * src,struct image_info * info,const struct elf_phdr * phdr,Error ** errp)3118 static bool parse_elf_properties(const ImageSource *src,
3119 struct image_info *info,
3120 const struct elf_phdr *phdr,
3121 Error **errp)
3122 {
3123 union {
3124 struct elf_note nhdr;
3125 uint32_t data[NOTE_DATA_SZ / sizeof(uint32_t)];
3126 } note;
3127
3128 int n, off, datasz;
3129 bool have_prev_type;
3130 uint32_t prev_type;
3131
3132 /* Unless the arch requires properties, ignore them. */
3133 if (!ARCH_USE_GNU_PROPERTY) {
3134 return true;
3135 }
3136
3137 /* If the properties are crazy large, that's too bad. */
3138 n = phdr->p_filesz;
3139 if (n > sizeof(note)) {
3140 error_setg(errp, "PT_GNU_PROPERTY too large");
3141 return false;
3142 }
3143 if (n < sizeof(note.nhdr)) {
3144 error_setg(errp, "PT_GNU_PROPERTY too small");
3145 return false;
3146 }
3147
3148 if (!imgsrc_read(¬e, phdr->p_offset, n, src, errp)) {
3149 return false;
3150 }
3151
3152 /*
3153 * The contents of a valid PT_GNU_PROPERTY is a sequence of uint32_t.
3154 * Swap most of them now, beyond the header and namesz.
3155 */
3156 if (target_needs_bswap()) {
3157 for (int i = 4; i < n / 4; i++) {
3158 bswap32s(note.data + i);
3159 }
3160 }
3161
3162 /*
3163 * Note that nhdr is 3 words, and that the "name" described by namesz
3164 * immediately follows nhdr and is thus at the 4th word. Further, all
3165 * of the inputs to the kernel's round_up are multiples of 4.
3166 */
3167 if (tswap32(note.nhdr.n_type) != NT_GNU_PROPERTY_TYPE_0 ||
3168 tswap32(note.nhdr.n_namesz) != NOTE_NAME_SZ ||
3169 note.data[3] != GNU0_MAGIC) {
3170 error_setg(errp, "Invalid note in PT_GNU_PROPERTY");
3171 return false;
3172 }
3173 off = sizeof(note.nhdr) + NOTE_NAME_SZ;
3174
3175 datasz = tswap32(note.nhdr.n_descsz) + off;
3176 if (datasz > n) {
3177 error_setg(errp, "Invalid note size in PT_GNU_PROPERTY");
3178 return false;
3179 }
3180
3181 have_prev_type = false;
3182 prev_type = 0;
3183 while (1) {
3184 if (off == datasz) {
3185 return true; /* end, exit ok */
3186 }
3187 if (!parse_elf_property(note.data, &off, datasz, info,
3188 have_prev_type, &prev_type, errp)) {
3189 return false;
3190 }
3191 have_prev_type = true;
3192 }
3193 }
3194
3195 /**
3196 * load_elf_image: Load an ELF image into the address space.
3197 * @image_name: the filename of the image, to use in error messages.
3198 * @src: the ImageSource from which to read.
3199 * @info: info collected from the loaded image.
3200 * @ehdr: the ELF header, not yet bswapped.
3201 * @pinterp_name: record any PT_INTERP string found.
3202 *
3203 * On return: @info values will be filled in, as necessary or available.
3204 */
3205
load_elf_image(const char * image_name,const ImageSource * src,struct image_info * info,struct elfhdr * ehdr,char ** pinterp_name)3206 static void load_elf_image(const char *image_name, const ImageSource *src,
3207 struct image_info *info, struct elfhdr *ehdr,
3208 char **pinterp_name)
3209 {
3210 g_autofree struct elf_phdr *phdr = NULL;
3211 abi_ulong load_addr, load_bias, loaddr, hiaddr, error, align;
3212 size_t reserve_size, align_size;
3213 int i, prot_exec;
3214 Error *err = NULL;
3215
3216 /*
3217 * First of all, some simple consistency checks.
3218 * Note that we rely on the bswapped ehdr staying in bprm_buf,
3219 * for later use by load_elf_binary and create_elf_tables.
3220 */
3221 if (!imgsrc_read(ehdr, 0, sizeof(*ehdr), src, &err)) {
3222 goto exit_errmsg;
3223 }
3224 if (!elf_check_ident(ehdr)) {
3225 error_setg(&err, "Invalid ELF image for this architecture");
3226 goto exit_errmsg;
3227 }
3228 bswap_ehdr(ehdr);
3229 if (!elf_check_ehdr(ehdr)) {
3230 error_setg(&err, "Invalid ELF image for this architecture");
3231 goto exit_errmsg;
3232 }
3233
3234 phdr = imgsrc_read_alloc(ehdr->e_phoff,
3235 ehdr->e_phnum * sizeof(struct elf_phdr),
3236 src, &err);
3237 if (phdr == NULL) {
3238 goto exit_errmsg;
3239 }
3240 bswap_phdr(phdr, ehdr->e_phnum);
3241
3242 info->nsegs = 0;
3243 info->pt_dynamic_addr = 0;
3244
3245 mmap_lock();
3246
3247 /*
3248 * Find the maximum size of the image and allocate an appropriate
3249 * amount of memory to handle that. Locate the interpreter, if any.
3250 */
3251 loaddr = -1, hiaddr = 0;
3252 align = 0;
3253 info->exec_stack = EXSTACK_DEFAULT;
3254 for (i = 0; i < ehdr->e_phnum; ++i) {
3255 struct elf_phdr *eppnt = phdr + i;
3256 if (eppnt->p_type == PT_LOAD) {
3257 abi_ulong a = eppnt->p_vaddr & TARGET_PAGE_MASK;
3258 if (a < loaddr) {
3259 loaddr = a;
3260 }
3261 a = eppnt->p_vaddr + eppnt->p_memsz - 1;
3262 if (a > hiaddr) {
3263 hiaddr = a;
3264 }
3265 ++info->nsegs;
3266 align |= eppnt->p_align;
3267 } else if (eppnt->p_type == PT_INTERP && pinterp_name) {
3268 g_autofree char *interp_name = NULL;
3269
3270 if (*pinterp_name) {
3271 error_setg(&err, "Multiple PT_INTERP entries");
3272 goto exit_errmsg;
3273 }
3274
3275 interp_name = imgsrc_read_alloc(eppnt->p_offset, eppnt->p_filesz,
3276 src, &err);
3277 if (interp_name == NULL) {
3278 goto exit_errmsg;
3279 }
3280 if (interp_name[eppnt->p_filesz - 1] != 0) {
3281 error_setg(&err, "Invalid PT_INTERP entry");
3282 goto exit_errmsg;
3283 }
3284 *pinterp_name = g_steal_pointer(&interp_name);
3285 } else if (eppnt->p_type == PT_GNU_PROPERTY) {
3286 if (!parse_elf_properties(src, info, eppnt, &err)) {
3287 goto exit_errmsg;
3288 }
3289 } else if (eppnt->p_type == PT_GNU_STACK) {
3290 info->exec_stack = eppnt->p_flags & PF_X;
3291 }
3292 }
3293
3294 load_addr = loaddr;
3295
3296 align = pow2ceil(align);
3297
3298 if (pinterp_name != NULL) {
3299 if (ehdr->e_type == ET_EXEC) {
3300 /*
3301 * Make sure that the low address does not conflict with
3302 * MMAP_MIN_ADDR or the QEMU application itself.
3303 */
3304 probe_guest_base(image_name, loaddr, hiaddr);
3305 } else {
3306 /*
3307 * The binary is dynamic, but we still need to
3308 * select guest_base. In this case we pass a size.
3309 */
3310 probe_guest_base(image_name, 0, hiaddr - loaddr);
3311
3312 /*
3313 * Avoid collision with the loader by providing a different
3314 * default load address.
3315 */
3316 load_addr += elf_et_dyn_base;
3317
3318 /*
3319 * TODO: Better support for mmap alignment is desirable.
3320 * Since we do not have complete control over the guest
3321 * address space, we prefer the kernel to choose some address
3322 * rather than force the use of LOAD_ADDR via MAP_FIXED.
3323 */
3324 if (align) {
3325 load_addr &= -align;
3326 }
3327 }
3328 }
3329
3330 /*
3331 * Reserve address space for all of this.
3332 *
3333 * In the case of ET_EXEC, we supply MAP_FIXED_NOREPLACE so that we get
3334 * exactly the address range that is required. Without reserved_va,
3335 * the guest address space is not isolated. We have attempted to avoid
3336 * conflict with the host program itself via probe_guest_base, but using
3337 * MAP_FIXED_NOREPLACE instead of MAP_FIXED provides an extra check.
3338 *
3339 * Otherwise this is ET_DYN, and we are searching for a location
3340 * that can hold the memory space required. If the image is
3341 * pre-linked, LOAD_ADDR will be non-zero, and the kernel should
3342 * honor that address if it happens to be free.
3343 *
3344 * In both cases, we will overwrite pages in this range with mappings
3345 * from the executable.
3346 */
3347 reserve_size = (size_t)hiaddr - loaddr + 1;
3348 align_size = reserve_size;
3349
3350 if (ehdr->e_type != ET_EXEC && align > qemu_real_host_page_size()) {
3351 align_size += align - 1;
3352 }
3353
3354 load_addr = target_mmap(load_addr, align_size, PROT_NONE,
3355 MAP_PRIVATE | MAP_ANON | MAP_NORESERVE |
3356 (ehdr->e_type == ET_EXEC ? MAP_FIXED_NOREPLACE : 0),
3357 -1, 0);
3358 if (load_addr == -1) {
3359 goto exit_mmap;
3360 }
3361
3362 if (align_size != reserve_size) {
3363 abi_ulong align_addr = ROUND_UP(load_addr, align);
3364 abi_ulong align_end = TARGET_PAGE_ALIGN(align_addr + reserve_size);
3365 abi_ulong load_end = TARGET_PAGE_ALIGN(load_addr + align_size);
3366
3367 if (align_addr != load_addr) {
3368 target_munmap(load_addr, align_addr - load_addr);
3369 }
3370 if (align_end != load_end) {
3371 target_munmap(align_end, load_end - align_end);
3372 }
3373 load_addr = align_addr;
3374 }
3375
3376 load_bias = load_addr - loaddr;
3377
3378 if (elf_is_fdpic(ehdr)) {
3379 struct elf32_fdpic_loadseg *loadsegs = info->loadsegs =
3380 g_malloc(sizeof(*loadsegs) * info->nsegs);
3381
3382 for (i = 0; i < ehdr->e_phnum; ++i) {
3383 switch (phdr[i].p_type) {
3384 case PT_DYNAMIC:
3385 info->pt_dynamic_addr = phdr[i].p_vaddr + load_bias;
3386 break;
3387 case PT_LOAD:
3388 loadsegs->addr = phdr[i].p_vaddr + load_bias;
3389 loadsegs->p_vaddr = phdr[i].p_vaddr;
3390 loadsegs->p_memsz = phdr[i].p_memsz;
3391 ++loadsegs;
3392 break;
3393 }
3394 }
3395 }
3396
3397 info->load_bias = load_bias;
3398 info->code_offset = load_bias;
3399 info->data_offset = load_bias;
3400 info->load_addr = load_addr;
3401 info->entry = ehdr->e_entry + load_bias;
3402 info->start_code = -1;
3403 info->end_code = 0;
3404 info->start_data = -1;
3405 info->end_data = 0;
3406 /* Usual start for brk is after all sections of the main executable. */
3407 info->brk = TARGET_PAGE_ALIGN(hiaddr + load_bias);
3408 info->elf_flags = ehdr->e_flags;
3409
3410 prot_exec = PROT_EXEC;
3411 #ifdef TARGET_AARCH64
3412 /*
3413 * If the BTI feature is present, this indicates that the executable
3414 * pages of the startup binary should be mapped with PROT_BTI, so that
3415 * branch targets are enforced.
3416 *
3417 * The startup binary is either the interpreter or the static executable.
3418 * The interpreter is responsible for all pages of a dynamic executable.
3419 *
3420 * Elf notes are backward compatible to older cpus.
3421 * Do not enable BTI unless it is supported.
3422 */
3423 if ((info->note_flags & GNU_PROPERTY_AARCH64_FEATURE_1_BTI)
3424 && (pinterp_name == NULL || *pinterp_name == 0)
3425 && cpu_isar_feature(aa64_bti, ARM_CPU(thread_cpu))) {
3426 prot_exec |= TARGET_PROT_BTI;
3427 }
3428 #endif
3429
3430 for (i = 0; i < ehdr->e_phnum; i++) {
3431 struct elf_phdr *eppnt = phdr + i;
3432 if (eppnt->p_type == PT_LOAD) {
3433 abi_ulong vaddr, vaddr_po, vaddr_ps, vaddr_ef, vaddr_em;
3434 int elf_prot = 0;
3435
3436 if (eppnt->p_flags & PF_R) {
3437 elf_prot |= PROT_READ;
3438 }
3439 if (eppnt->p_flags & PF_W) {
3440 elf_prot |= PROT_WRITE;
3441 }
3442 if (eppnt->p_flags & PF_X) {
3443 elf_prot |= prot_exec;
3444 }
3445
3446 vaddr = load_bias + eppnt->p_vaddr;
3447 vaddr_po = vaddr & ~TARGET_PAGE_MASK;
3448 vaddr_ps = vaddr & TARGET_PAGE_MASK;
3449
3450 vaddr_ef = vaddr + eppnt->p_filesz;
3451 vaddr_em = vaddr + eppnt->p_memsz;
3452
3453 /*
3454 * Some segments may be completely empty, with a non-zero p_memsz
3455 * but no backing file segment.
3456 */
3457 if (eppnt->p_filesz != 0) {
3458 error = imgsrc_mmap(vaddr_ps, eppnt->p_filesz + vaddr_po,
3459 elf_prot, MAP_PRIVATE | MAP_FIXED,
3460 src, eppnt->p_offset - vaddr_po);
3461 if (error == -1) {
3462 goto exit_mmap;
3463 }
3464 }
3465
3466 /* If the load segment requests extra zeros (e.g. bss), map it. */
3467 if (vaddr_ef < vaddr_em &&
3468 !zero_bss(vaddr_ef, vaddr_em, elf_prot, &err)) {
3469 goto exit_errmsg;
3470 }
3471
3472 /* Find the full program boundaries. */
3473 if (elf_prot & PROT_EXEC) {
3474 if (vaddr < info->start_code) {
3475 info->start_code = vaddr;
3476 }
3477 if (vaddr_ef > info->end_code) {
3478 info->end_code = vaddr_ef;
3479 }
3480 }
3481 if (elf_prot & PROT_WRITE) {
3482 if (vaddr < info->start_data) {
3483 info->start_data = vaddr;
3484 }
3485 if (vaddr_ef > info->end_data) {
3486 info->end_data = vaddr_ef;
3487 }
3488 }
3489 #ifdef TARGET_MIPS
3490 } else if (eppnt->p_type == PT_MIPS_ABIFLAGS) {
3491 Mips_elf_abiflags_v0 abiflags;
3492
3493 if (!imgsrc_read(&abiflags, eppnt->p_offset, sizeof(abiflags),
3494 src, &err)) {
3495 goto exit_errmsg;
3496 }
3497 bswap_mips_abiflags(&abiflags);
3498 info->fp_abi = abiflags.fp_abi;
3499 #endif
3500 }
3501 }
3502
3503 if (info->end_data == 0) {
3504 info->start_data = info->end_code;
3505 info->end_data = info->end_code;
3506 }
3507
3508 if (qemu_log_enabled()) {
3509 load_symbols(ehdr, src, load_bias);
3510 }
3511
3512 debuginfo_report_elf(image_name, src->fd, load_bias);
3513
3514 mmap_unlock();
3515
3516 close(src->fd);
3517 return;
3518
3519 exit_mmap:
3520 error_setg_errno(&err, errno, "Error mapping file");
3521 goto exit_errmsg;
3522 exit_errmsg:
3523 error_reportf_err(err, "%s: ", image_name);
3524 exit(-1);
3525 }
3526
load_elf_interp(const char * filename,struct image_info * info,char bprm_buf[BPRM_BUF_SIZE])3527 static void load_elf_interp(const char *filename, struct image_info *info,
3528 char bprm_buf[BPRM_BUF_SIZE])
3529 {
3530 struct elfhdr ehdr;
3531 ImageSource src;
3532 int fd, retval;
3533 Error *err = NULL;
3534
3535 fd = open(path(filename), O_RDONLY);
3536 if (fd < 0) {
3537 error_setg_file_open(&err, errno, filename);
3538 error_report_err(err);
3539 exit(-1);
3540 }
3541
3542 retval = read(fd, bprm_buf, BPRM_BUF_SIZE);
3543 if (retval < 0) {
3544 error_setg_errno(&err, errno, "Error reading file header");
3545 error_reportf_err(err, "%s: ", filename);
3546 exit(-1);
3547 }
3548
3549 src.fd = fd;
3550 src.cache = bprm_buf;
3551 src.cache_size = retval;
3552
3553 load_elf_image(filename, &src, info, &ehdr, NULL);
3554 }
3555
3556 #ifndef vdso_image_info
3557 #ifdef VDSO_HEADER
3558 #include VDSO_HEADER
3559 #define vdso_image_info(flags) &vdso_image_info
3560 #else
3561 #define vdso_image_info(flags) NULL
3562 #endif /* VDSO_HEADER */
3563 #endif /* vdso_image_info */
3564
load_elf_vdso(struct image_info * info,const VdsoImageInfo * vdso)3565 static void load_elf_vdso(struct image_info *info, const VdsoImageInfo *vdso)
3566 {
3567 ImageSource src;
3568 struct elfhdr ehdr;
3569 abi_ulong load_bias, load_addr;
3570
3571 src.fd = -1;
3572 src.cache = vdso->image;
3573 src.cache_size = vdso->image_size;
3574
3575 load_elf_image("<internal-vdso>", &src, info, &ehdr, NULL);
3576 load_addr = info->load_addr;
3577 load_bias = info->load_bias;
3578
3579 /*
3580 * We need to relocate the VDSO image. The one built into the kernel
3581 * is built for a fixed address. The one built for QEMU is not, since
3582 * that requires close control of the guest address space.
3583 * We pre-processed the image to locate all of the addresses that need
3584 * to be updated.
3585 */
3586 for (unsigned i = 0, n = vdso->reloc_count; i < n; i++) {
3587 abi_ulong *addr = g2h_untagged(load_addr + vdso->relocs[i]);
3588 *addr = tswapal(tswapal(*addr) + load_bias);
3589 }
3590
3591 /* Install signal trampolines, if present. */
3592 if (vdso->sigreturn_ofs) {
3593 default_sigreturn = load_addr + vdso->sigreturn_ofs;
3594 }
3595 if (vdso->rt_sigreturn_ofs) {
3596 default_rt_sigreturn = load_addr + vdso->rt_sigreturn_ofs;
3597 }
3598
3599 /* Remove write from VDSO segment. */
3600 target_mprotect(info->start_data, info->end_data - info->start_data,
3601 PROT_READ | PROT_EXEC);
3602 }
3603
symfind(const void * s0,const void * s1)3604 static int symfind(const void *s0, const void *s1)
3605 {
3606 struct elf_sym *sym = (struct elf_sym *)s1;
3607 __typeof(sym->st_value) addr = *(uint64_t *)s0;
3608 int result = 0;
3609
3610 if (addr < sym->st_value) {
3611 result = -1;
3612 } else if (addr >= sym->st_value + sym->st_size) {
3613 result = 1;
3614 }
3615 return result;
3616 }
3617
lookup_symbolxx(struct syminfo * s,uint64_t orig_addr)3618 static const char *lookup_symbolxx(struct syminfo *s, uint64_t orig_addr)
3619 {
3620 #if ELF_CLASS == ELFCLASS32
3621 struct elf_sym *syms = s->disas_symtab.elf32;
3622 #else
3623 struct elf_sym *syms = s->disas_symtab.elf64;
3624 #endif
3625
3626 // binary search
3627 struct elf_sym *sym;
3628
3629 sym = bsearch(&orig_addr, syms, s->disas_num_syms, sizeof(*syms), symfind);
3630 if (sym != NULL) {
3631 return s->disas_strtab + sym->st_name;
3632 }
3633
3634 return "";
3635 }
3636
3637 /* FIXME: This should use elf_ops.h.inc */
symcmp(const void * s0,const void * s1)3638 static int symcmp(const void *s0, const void *s1)
3639 {
3640 struct elf_sym *sym0 = (struct elf_sym *)s0;
3641 struct elf_sym *sym1 = (struct elf_sym *)s1;
3642 return (sym0->st_value < sym1->st_value)
3643 ? -1
3644 : ((sym0->st_value > sym1->st_value) ? 1 : 0);
3645 }
3646
3647 /* Best attempt to load symbols from this ELF object. */
load_symbols(struct elfhdr * hdr,const ImageSource * src,abi_ulong load_bias)3648 static void load_symbols(struct elfhdr *hdr, const ImageSource *src,
3649 abi_ulong load_bias)
3650 {
3651 int i, shnum, nsyms, sym_idx = 0, str_idx = 0;
3652 g_autofree struct elf_shdr *shdr = NULL;
3653 char *strings = NULL;
3654 struct elf_sym *syms = NULL;
3655 struct elf_sym *new_syms;
3656 uint64_t segsz;
3657
3658 shnum = hdr->e_shnum;
3659 shdr = imgsrc_read_alloc(hdr->e_shoff, shnum * sizeof(struct elf_shdr),
3660 src, NULL);
3661 if (shdr == NULL) {
3662 return;
3663 }
3664
3665 bswap_shdr(shdr, shnum);
3666 for (i = 0; i < shnum; ++i) {
3667 if (shdr[i].sh_type == SHT_SYMTAB) {
3668 sym_idx = i;
3669 str_idx = shdr[i].sh_link;
3670 goto found;
3671 }
3672 }
3673
3674 /* There will be no symbol table if the file was stripped. */
3675 return;
3676
3677 found:
3678 /* Now know where the strtab and symtab are. Snarf them. */
3679
3680 segsz = shdr[str_idx].sh_size;
3681 strings = g_try_malloc(segsz);
3682 if (!strings) {
3683 goto give_up;
3684 }
3685 if (!imgsrc_read(strings, shdr[str_idx].sh_offset, segsz, src, NULL)) {
3686 goto give_up;
3687 }
3688
3689 segsz = shdr[sym_idx].sh_size;
3690 if (segsz / sizeof(struct elf_sym) > INT_MAX) {
3691 /*
3692 * Implausibly large symbol table: give up rather than ploughing
3693 * on with the number of symbols calculation overflowing.
3694 */
3695 goto give_up;
3696 }
3697 nsyms = segsz / sizeof(struct elf_sym);
3698 syms = g_try_malloc(segsz);
3699 if (!syms) {
3700 goto give_up;
3701 }
3702 if (!imgsrc_read(syms, shdr[sym_idx].sh_offset, segsz, src, NULL)) {
3703 goto give_up;
3704 }
3705
3706 for (i = 0; i < nsyms; ) {
3707 bswap_sym(syms + i);
3708 /* Throw away entries which we do not need. */
3709 if (syms[i].st_shndx == SHN_UNDEF
3710 || syms[i].st_shndx >= SHN_LORESERVE
3711 || ELF_ST_TYPE(syms[i].st_info) != STT_FUNC) {
3712 if (i < --nsyms) {
3713 syms[i] = syms[nsyms];
3714 }
3715 } else {
3716 #if defined(TARGET_ARM) || defined (TARGET_MIPS)
3717 /* The bottom address bit marks a Thumb or MIPS16 symbol. */
3718 syms[i].st_value &= ~(target_ulong)1;
3719 #endif
3720 syms[i].st_value += load_bias;
3721 i++;
3722 }
3723 }
3724
3725 /* No "useful" symbol. */
3726 if (nsyms == 0) {
3727 goto give_up;
3728 }
3729
3730 /*
3731 * Attempt to free the storage associated with the local symbols
3732 * that we threw away. Whether or not this has any effect on the
3733 * memory allocation depends on the malloc implementation and how
3734 * many symbols we managed to discard.
3735 */
3736 new_syms = g_try_renew(struct elf_sym, syms, nsyms);
3737 if (new_syms == NULL) {
3738 goto give_up;
3739 }
3740 syms = new_syms;
3741
3742 qsort(syms, nsyms, sizeof(*syms), symcmp);
3743
3744 {
3745 struct syminfo *s = g_new(struct syminfo, 1);
3746
3747 s->disas_strtab = strings;
3748 s->disas_num_syms = nsyms;
3749 #if ELF_CLASS == ELFCLASS32
3750 s->disas_symtab.elf32 = syms;
3751 #else
3752 s->disas_symtab.elf64 = syms;
3753 #endif
3754 s->lookup_symbol = lookup_symbolxx;
3755 s->next = syminfos;
3756 syminfos = s;
3757 }
3758 return;
3759
3760 give_up:
3761 g_free(strings);
3762 g_free(syms);
3763 }
3764
get_elf_eflags(int fd)3765 uint32_t get_elf_eflags(int fd)
3766 {
3767 struct elfhdr ehdr;
3768 off_t offset;
3769 int ret;
3770
3771 /* Read ELF header */
3772 offset = lseek(fd, 0, SEEK_SET);
3773 if (offset == (off_t) -1) {
3774 return 0;
3775 }
3776 ret = read(fd, &ehdr, sizeof(ehdr));
3777 if (ret < sizeof(ehdr)) {
3778 return 0;
3779 }
3780 offset = lseek(fd, offset, SEEK_SET);
3781 if (offset == (off_t) -1) {
3782 return 0;
3783 }
3784
3785 /* Check ELF signature */
3786 if (!elf_check_ident(&ehdr)) {
3787 return 0;
3788 }
3789
3790 /* check header */
3791 bswap_ehdr(&ehdr);
3792 if (!elf_check_ehdr(&ehdr)) {
3793 return 0;
3794 }
3795
3796 /* return architecture id */
3797 return ehdr.e_flags;
3798 }
3799
load_elf_binary(struct linux_binprm * bprm,struct image_info * info)3800 int load_elf_binary(struct linux_binprm *bprm, struct image_info *info)
3801 {
3802 /*
3803 * We need a copy of the elf header for passing to create_elf_tables.
3804 * We will have overwritten the original when we re-use bprm->buf
3805 * while loading the interpreter. Allocate the storage for this now
3806 * and let elf_load_image do any swapping that may be required.
3807 */
3808 struct elfhdr ehdr;
3809 struct image_info interp_info, vdso_info;
3810 char *elf_interpreter = NULL;
3811 char *scratch;
3812
3813 memset(&interp_info, 0, sizeof(interp_info));
3814 #ifdef TARGET_MIPS
3815 interp_info.fp_abi = MIPS_ABI_FP_UNKNOWN;
3816 #endif
3817
3818 load_elf_image(bprm->filename, &bprm->src, info, &ehdr, &elf_interpreter);
3819
3820 /* Do this so that we can load the interpreter, if need be. We will
3821 change some of these later */
3822 bprm->p = setup_arg_pages(bprm, info);
3823
3824 scratch = g_new0(char, TARGET_PAGE_SIZE);
3825 if (STACK_GROWS_DOWN) {
3826 bprm->p = copy_elf_strings(1, &bprm->filename, scratch,
3827 bprm->p, info->stack_limit);
3828 info->file_string = bprm->p;
3829 bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch,
3830 bprm->p, info->stack_limit);
3831 info->env_strings = bprm->p;
3832 bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch,
3833 bprm->p, info->stack_limit);
3834 info->arg_strings = bprm->p;
3835 } else {
3836 info->arg_strings = bprm->p;
3837 bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch,
3838 bprm->p, info->stack_limit);
3839 info->env_strings = bprm->p;
3840 bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch,
3841 bprm->p, info->stack_limit);
3842 info->file_string = bprm->p;
3843 bprm->p = copy_elf_strings(1, &bprm->filename, scratch,
3844 bprm->p, info->stack_limit);
3845 }
3846
3847 g_free(scratch);
3848
3849 if (!bprm->p) {
3850 fprintf(stderr, "%s: %s\n", bprm->filename, strerror(E2BIG));
3851 exit(-1);
3852 }
3853
3854 if (elf_interpreter) {
3855 load_elf_interp(elf_interpreter, &interp_info, bprm->buf);
3856
3857 /*
3858 * While unusual because of ELF_ET_DYN_BASE, if we are unlucky
3859 * with the mappings the interpreter can be loaded above but
3860 * near the main executable, which can leave very little room
3861 * for the heap.
3862 * If the current brk has less than 16MB, use the end of the
3863 * interpreter.
3864 */
3865 if (interp_info.brk > info->brk &&
3866 interp_info.load_bias - info->brk < 16 * MiB) {
3867 info->brk = interp_info.brk;
3868 }
3869
3870 /* If the program interpreter is one of these two, then assume
3871 an iBCS2 image. Otherwise assume a native linux image. */
3872
3873 if (strcmp(elf_interpreter, "/usr/lib/libc.so.1") == 0
3874 || strcmp(elf_interpreter, "/usr/lib/ld.so.1") == 0) {
3875 info->personality = PER_SVR4;
3876
3877 /* Why this, you ask??? Well SVr4 maps page 0 as read-only,
3878 and some applications "depend" upon this behavior. Since
3879 we do not have the power to recompile these, we emulate
3880 the SVr4 behavior. Sigh. */
3881 target_mmap(0, TARGET_PAGE_SIZE, PROT_READ | PROT_EXEC,
3882 MAP_FIXED_NOREPLACE | MAP_PRIVATE | MAP_ANONYMOUS,
3883 -1, 0);
3884 }
3885 #ifdef TARGET_MIPS
3886 info->interp_fp_abi = interp_info.fp_abi;
3887 #endif
3888 }
3889
3890 /*
3891 * Load a vdso if available, which will amongst other things contain the
3892 * signal trampolines. Otherwise, allocate a separate page for them.
3893 */
3894 const VdsoImageInfo *vdso = vdso_image_info(info->elf_flags);
3895 if (vdso) {
3896 load_elf_vdso(&vdso_info, vdso);
3897 info->vdso = vdso_info.load_bias;
3898 } else if (TARGET_ARCH_HAS_SIGTRAMP_PAGE) {
3899 abi_long tramp_page = target_mmap(0, TARGET_PAGE_SIZE,
3900 PROT_READ | PROT_WRITE,
3901 MAP_PRIVATE | MAP_ANON, -1, 0);
3902 if (tramp_page == -1) {
3903 return -errno;
3904 }
3905
3906 setup_sigtramp(tramp_page);
3907 target_mprotect(tramp_page, TARGET_PAGE_SIZE, PROT_READ | PROT_EXEC);
3908 }
3909
3910 bprm->p = create_elf_tables(bprm->p, bprm->argc, bprm->envc, &ehdr, info,
3911 elf_interpreter ? &interp_info : NULL,
3912 vdso ? &vdso_info : NULL);
3913 info->start_stack = bprm->p;
3914
3915 /* If we have an interpreter, set that as the program's entry point.
3916 Copy the load_bias as well, to help PPC64 interpret the entry
3917 point as a function descriptor. Do this after creating elf tables
3918 so that we copy the original program entry point into the AUXV. */
3919 if (elf_interpreter) {
3920 info->load_bias = interp_info.load_bias;
3921 info->entry = interp_info.entry;
3922 g_free(elf_interpreter);
3923 }
3924
3925 #ifdef USE_ELF_CORE_DUMP
3926 bprm->core_dump = &elf_core_dump;
3927 #endif
3928
3929 return 0;
3930 }
3931
3932 #ifdef USE_ELF_CORE_DUMP
3933
3934 /*
3935 * Definitions to generate Intel SVR4-like core files.
3936 * These mostly have the same names as the SVR4 types with "target_elf_"
3937 * tacked on the front to prevent clashes with linux definitions,
3938 * and the typedef forms have been avoided. This is mostly like
3939 * the SVR4 structure, but more Linuxy, with things that Linux does
3940 * not support and which gdb doesn't really use excluded.
3941 *
3942 * Fields we don't dump (their contents is zero) in linux-user qemu
3943 * are marked with XXX.
3944 *
3945 * Core dump code is copied from linux kernel (fs/binfmt_elf.c).
3946 *
3947 * Porting ELF coredump for target is (quite) simple process. First you
3948 * define USE_ELF_CORE_DUMP in target ELF code (where init_thread() for
3949 * the target resides):
3950 *
3951 * #define USE_ELF_CORE_DUMP
3952 *
3953 * Next you define type of register set used for dumping. ELF specification
3954 * says that it needs to be array of elf_greg_t that has size of ELF_NREG.
3955 *
3956 * typedef <target_regtype> target_elf_greg_t;
3957 * #define ELF_NREG <number of registers>
3958 * typedef taret_elf_greg_t target_elf_gregset_t[ELF_NREG];
3959 *
3960 * Last step is to implement target specific function that copies registers
3961 * from given cpu into just specified register set. Prototype is:
3962 *
3963 * static void elf_core_copy_regs(taret_elf_gregset_t *regs,
3964 * const CPUArchState *env);
3965 *
3966 * Parameters:
3967 * regs - copy register values into here (allocated and zeroed by caller)
3968 * env - copy registers from here
3969 *
3970 * Example for ARM target is provided in this file.
3971 */
3972
3973 struct target_elf_siginfo {
3974 abi_int si_signo; /* signal number */
3975 abi_int si_code; /* extra code */
3976 abi_int si_errno; /* errno */
3977 };
3978
3979 struct target_elf_prstatus {
3980 struct target_elf_siginfo pr_info; /* Info associated with signal */
3981 abi_short pr_cursig; /* Current signal */
3982 abi_ulong pr_sigpend; /* XXX */
3983 abi_ulong pr_sighold; /* XXX */
3984 target_pid_t pr_pid;
3985 target_pid_t pr_ppid;
3986 target_pid_t pr_pgrp;
3987 target_pid_t pr_sid;
3988 struct target_timeval pr_utime; /* XXX User time */
3989 struct target_timeval pr_stime; /* XXX System time */
3990 struct target_timeval pr_cutime; /* XXX Cumulative user time */
3991 struct target_timeval pr_cstime; /* XXX Cumulative system time */
3992 target_elf_gregset_t pr_reg; /* GP registers */
3993 abi_int pr_fpvalid; /* XXX */
3994 };
3995
3996 #define ELF_PRARGSZ (80) /* Number of chars for args */
3997
3998 struct target_elf_prpsinfo {
3999 char pr_state; /* numeric process state */
4000 char pr_sname; /* char for pr_state */
4001 char pr_zomb; /* zombie */
4002 char pr_nice; /* nice val */
4003 abi_ulong pr_flag; /* flags */
4004 target_uid_t pr_uid;
4005 target_gid_t pr_gid;
4006 target_pid_t pr_pid, pr_ppid, pr_pgrp, pr_sid;
4007 /* Lots missing */
4008 char pr_fname[16] QEMU_NONSTRING; /* filename of executable */
4009 char pr_psargs[ELF_PRARGSZ]; /* initial part of arg list */
4010 };
4011
bswap_prstatus(struct target_elf_prstatus * prstatus)4012 static void bswap_prstatus(struct target_elf_prstatus *prstatus)
4013 {
4014 if (!target_needs_bswap()) {
4015 return;
4016 }
4017
4018 prstatus->pr_info.si_signo = tswap32(prstatus->pr_info.si_signo);
4019 prstatus->pr_info.si_code = tswap32(prstatus->pr_info.si_code);
4020 prstatus->pr_info.si_errno = tswap32(prstatus->pr_info.si_errno);
4021 prstatus->pr_cursig = tswap16(prstatus->pr_cursig);
4022 prstatus->pr_sigpend = tswapal(prstatus->pr_sigpend);
4023 prstatus->pr_sighold = tswapal(prstatus->pr_sighold);
4024 prstatus->pr_pid = tswap32(prstatus->pr_pid);
4025 prstatus->pr_ppid = tswap32(prstatus->pr_ppid);
4026 prstatus->pr_pgrp = tswap32(prstatus->pr_pgrp);
4027 prstatus->pr_sid = tswap32(prstatus->pr_sid);
4028 /* cpu times are not filled, so we skip them */
4029 /* regs should be in correct format already */
4030 prstatus->pr_fpvalid = tswap32(prstatus->pr_fpvalid);
4031 }
4032
bswap_psinfo(struct target_elf_prpsinfo * psinfo)4033 static void bswap_psinfo(struct target_elf_prpsinfo *psinfo)
4034 {
4035 if (!target_needs_bswap()) {
4036 return;
4037 }
4038
4039 psinfo->pr_flag = tswapal(psinfo->pr_flag);
4040 psinfo->pr_uid = tswap16(psinfo->pr_uid);
4041 psinfo->pr_gid = tswap16(psinfo->pr_gid);
4042 psinfo->pr_pid = tswap32(psinfo->pr_pid);
4043 psinfo->pr_ppid = tswap32(psinfo->pr_ppid);
4044 psinfo->pr_pgrp = tswap32(psinfo->pr_pgrp);
4045 psinfo->pr_sid = tswap32(psinfo->pr_sid);
4046 }
4047
bswap_note(struct elf_note * en)4048 static void bswap_note(struct elf_note *en)
4049 {
4050 if (!target_needs_bswap()) {
4051 return;
4052 }
4053
4054 bswap32s(&en->n_namesz);
4055 bswap32s(&en->n_descsz);
4056 bswap32s(&en->n_type);
4057 }
4058
4059 /*
4060 * Calculate file (dump) size of given memory region.
4061 */
vma_dump_size(vaddr start,vaddr end,int flags)4062 static size_t vma_dump_size(vaddr start, vaddr end, int flags)
4063 {
4064 /* The area must be readable. */
4065 if (!(flags & PAGE_READ)) {
4066 return 0;
4067 }
4068
4069 /*
4070 * Usually we don't dump executable pages as they contain
4071 * non-writable code that debugger can read directly from
4072 * target library etc. If there is no elf header, we dump it.
4073 */
4074 if (!(flags & PAGE_WRITE_ORG) &&
4075 (flags & PAGE_EXEC) &&
4076 memcmp(g2h_untagged(start), ELFMAG, SELFMAG) == 0) {
4077 return 0;
4078 }
4079
4080 return end - start;
4081 }
4082
size_note(const char * name,size_t datasz)4083 static size_t size_note(const char *name, size_t datasz)
4084 {
4085 size_t namesz = strlen(name) + 1;
4086
4087 namesz = ROUND_UP(namesz, 4);
4088 datasz = ROUND_UP(datasz, 4);
4089
4090 return sizeof(struct elf_note) + namesz + datasz;
4091 }
4092
fill_note(void ** pptr,int type,const char * name,size_t datasz)4093 static void *fill_note(void **pptr, int type, const char *name, size_t datasz)
4094 {
4095 void *ptr = *pptr;
4096 struct elf_note *n = ptr;
4097 size_t namesz = strlen(name) + 1;
4098
4099 n->n_namesz = namesz;
4100 n->n_descsz = datasz;
4101 n->n_type = type;
4102 bswap_note(n);
4103
4104 ptr += sizeof(*n);
4105 memcpy(ptr, name, namesz);
4106
4107 namesz = ROUND_UP(namesz, 4);
4108 datasz = ROUND_UP(datasz, 4);
4109
4110 *pptr = ptr + namesz + datasz;
4111 return ptr + namesz;
4112 }
4113
fill_elf_header(struct elfhdr * elf,int segs,uint16_t machine,uint32_t flags)4114 static void fill_elf_header(struct elfhdr *elf, int segs, uint16_t machine,
4115 uint32_t flags)
4116 {
4117 memcpy(elf->e_ident, ELFMAG, SELFMAG);
4118
4119 elf->e_ident[EI_CLASS] = ELF_CLASS;
4120 elf->e_ident[EI_DATA] = ELF_DATA;
4121 elf->e_ident[EI_VERSION] = EV_CURRENT;
4122 elf->e_ident[EI_OSABI] = ELF_OSABI;
4123
4124 elf->e_type = ET_CORE;
4125 elf->e_machine = machine;
4126 elf->e_version = EV_CURRENT;
4127 elf->e_phoff = sizeof(struct elfhdr);
4128 elf->e_flags = flags;
4129 elf->e_ehsize = sizeof(struct elfhdr);
4130 elf->e_phentsize = sizeof(struct elf_phdr);
4131 elf->e_phnum = segs;
4132
4133 bswap_ehdr(elf);
4134 }
4135
fill_elf_note_phdr(struct elf_phdr * phdr,size_t sz,off_t offset)4136 static void fill_elf_note_phdr(struct elf_phdr *phdr, size_t sz, off_t offset)
4137 {
4138 phdr->p_type = PT_NOTE;
4139 phdr->p_offset = offset;
4140 phdr->p_filesz = sz;
4141
4142 bswap_phdr(phdr, 1);
4143 }
4144
fill_prstatus_note(void * data,CPUState * cpu,int signr)4145 static void fill_prstatus_note(void *data, CPUState *cpu, int signr)
4146 {
4147 /*
4148 * Because note memory is only aligned to 4, and target_elf_prstatus
4149 * may well have higher alignment requirements, fill locally and
4150 * memcpy to the destination afterward.
4151 */
4152 struct target_elf_prstatus prstatus = {
4153 .pr_info.si_signo = signr,
4154 .pr_cursig = signr,
4155 .pr_pid = get_task_state(cpu)->ts_tid,
4156 .pr_ppid = getppid(),
4157 .pr_pgrp = getpgrp(),
4158 .pr_sid = getsid(0),
4159 };
4160
4161 elf_core_copy_regs(&prstatus.pr_reg, cpu_env(cpu));
4162 bswap_prstatus(&prstatus);
4163 memcpy(data, &prstatus, sizeof(prstatus));
4164 }
4165
fill_prpsinfo_note(void * data,const TaskState * ts)4166 static void fill_prpsinfo_note(void *data, const TaskState *ts)
4167 {
4168 /*
4169 * Because note memory is only aligned to 4, and target_elf_prpsinfo
4170 * may well have higher alignment requirements, fill locally and
4171 * memcpy to the destination afterward.
4172 */
4173 struct target_elf_prpsinfo psinfo = {
4174 .pr_pid = getpid(),
4175 .pr_ppid = getppid(),
4176 .pr_pgrp = getpgrp(),
4177 .pr_sid = getsid(0),
4178 .pr_uid = getuid(),
4179 .pr_gid = getgid(),
4180 };
4181 char *base_filename;
4182 size_t len;
4183
4184 len = ts->info->env_strings - ts->info->arg_strings;
4185 len = MIN(len, ELF_PRARGSZ);
4186 memcpy(&psinfo.pr_psargs, g2h_untagged(ts->info->arg_strings), len);
4187 for (size_t i = 0; i < len; i++) {
4188 if (psinfo.pr_psargs[i] == 0) {
4189 psinfo.pr_psargs[i] = ' ';
4190 }
4191 }
4192
4193 base_filename = g_path_get_basename(ts->bprm->filename);
4194 /*
4195 * Using strncpy here is fine: at max-length,
4196 * this field is not NUL-terminated.
4197 */
4198 strncpy(psinfo.pr_fname, base_filename, sizeof(psinfo.pr_fname));
4199 g_free(base_filename);
4200
4201 bswap_psinfo(&psinfo);
4202 memcpy(data, &psinfo, sizeof(psinfo));
4203 }
4204
fill_auxv_note(void * data,const TaskState * ts)4205 static void fill_auxv_note(void *data, const TaskState *ts)
4206 {
4207 memcpy(data, g2h_untagged(ts->info->saved_auxv), ts->info->auxv_len);
4208 }
4209
4210 /*
4211 * Constructs name of coredump file. We have following convention
4212 * for the name:
4213 * qemu_<basename-of-target-binary>_<date>-<time>_<pid>.core
4214 *
4215 * Returns the filename
4216 */
core_dump_filename(const TaskState * ts)4217 static char *core_dump_filename(const TaskState *ts)
4218 {
4219 g_autoptr(GDateTime) now = g_date_time_new_now_local();
4220 g_autofree char *nowstr = g_date_time_format(now, "%Y%m%d-%H%M%S");
4221 g_autofree char *base_filename = g_path_get_basename(ts->bprm->filename);
4222
4223 return g_strdup_printf("qemu_%s_%s_%d.core",
4224 base_filename, nowstr, (int)getpid());
4225 }
4226
dump_write(int fd,const void * ptr,size_t size)4227 static int dump_write(int fd, const void *ptr, size_t size)
4228 {
4229 const char *bufp = (const char *)ptr;
4230 ssize_t bytes_written, bytes_left;
4231
4232 bytes_written = 0;
4233 bytes_left = size;
4234
4235 /*
4236 * In normal conditions, single write(2) should do but
4237 * in case of socket etc. this mechanism is more portable.
4238 */
4239 do {
4240 bytes_written = write(fd, bufp, bytes_left);
4241 if (bytes_written < 0) {
4242 if (errno == EINTR)
4243 continue;
4244 return (-1);
4245 } else if (bytes_written == 0) { /* eof */
4246 return (-1);
4247 }
4248 bufp += bytes_written;
4249 bytes_left -= bytes_written;
4250 } while (bytes_left > 0);
4251
4252 return (0);
4253 }
4254
wmr_page_unprotect_regions(void * opaque,vaddr start,vaddr end,int flags)4255 static int wmr_page_unprotect_regions(void *opaque, vaddr start,
4256 vaddr end, int flags)
4257 {
4258 if ((flags & (PAGE_WRITE | PAGE_WRITE_ORG)) == PAGE_WRITE_ORG) {
4259 size_t step = MAX(TARGET_PAGE_SIZE, qemu_real_host_page_size());
4260
4261 while (1) {
4262 page_unprotect(NULL, start, 0);
4263 if (end - start <= step) {
4264 break;
4265 }
4266 start += step;
4267 }
4268 }
4269 return 0;
4270 }
4271
4272 typedef struct {
4273 unsigned count;
4274 size_t size;
4275 } CountAndSizeRegions;
4276
wmr_count_and_size_regions(void * opaque,vaddr start,vaddr end,int flags)4277 static int wmr_count_and_size_regions(void *opaque, vaddr start,
4278 vaddr end, int flags)
4279 {
4280 CountAndSizeRegions *css = opaque;
4281
4282 css->count++;
4283 css->size += vma_dump_size(start, end, flags);
4284 return 0;
4285 }
4286
4287 typedef struct {
4288 struct elf_phdr *phdr;
4289 off_t offset;
4290 } FillRegionPhdr;
4291
wmr_fill_region_phdr(void * opaque,vaddr start,vaddr end,int flags)4292 static int wmr_fill_region_phdr(void *opaque, vaddr start,
4293 vaddr end, int flags)
4294 {
4295 FillRegionPhdr *d = opaque;
4296 struct elf_phdr *phdr = d->phdr;
4297
4298 phdr->p_type = PT_LOAD;
4299 phdr->p_vaddr = start;
4300 phdr->p_paddr = 0;
4301 phdr->p_filesz = vma_dump_size(start, end, flags);
4302 phdr->p_offset = d->offset;
4303 d->offset += phdr->p_filesz;
4304 phdr->p_memsz = end - start;
4305 phdr->p_flags = (flags & PAGE_READ ? PF_R : 0)
4306 | (flags & PAGE_WRITE_ORG ? PF_W : 0)
4307 | (flags & PAGE_EXEC ? PF_X : 0);
4308 phdr->p_align = ELF_EXEC_PAGESIZE;
4309
4310 bswap_phdr(phdr, 1);
4311 d->phdr = phdr + 1;
4312 return 0;
4313 }
4314
wmr_write_region(void * opaque,vaddr start,vaddr end,int flags)4315 static int wmr_write_region(void *opaque, vaddr start,
4316 vaddr end, int flags)
4317 {
4318 int fd = *(int *)opaque;
4319 size_t size = vma_dump_size(start, end, flags);
4320
4321 if (!size) {
4322 return 0;
4323 }
4324 return dump_write(fd, g2h_untagged(start), size);
4325 }
4326
4327 /*
4328 * Write out ELF coredump.
4329 *
4330 * See documentation of ELF object file format in:
4331 * http://www.caldera.com/developers/devspecs/gabi41.pdf
4332 *
4333 * Coredump format in linux is following:
4334 *
4335 * 0 +----------------------+ \
4336 * | ELF header | ET_CORE |
4337 * +----------------------+ |
4338 * | ELF program headers | |--- headers
4339 * | - NOTE section | |
4340 * | - PT_LOAD sections | |
4341 * +----------------------+ /
4342 * | NOTEs: |
4343 * | - NT_PRSTATUS |
4344 * | - NT_PRSINFO |
4345 * | - NT_AUXV |
4346 * +----------------------+ <-- aligned to target page
4347 * | Process memory dump |
4348 * : :
4349 * . .
4350 * : :
4351 * | |
4352 * +----------------------+
4353 *
4354 * NT_PRSTATUS -> struct elf_prstatus (per thread)
4355 * NT_PRSINFO -> struct elf_prpsinfo
4356 * NT_AUXV is array of { type, value } pairs (see fill_auxv_note()).
4357 *
4358 * Format follows System V format as close as possible. Current
4359 * version limitations are as follows:
4360 * - no floating point registers are dumped
4361 *
4362 * Function returns 0 in case of success, negative errno otherwise.
4363 *
4364 * TODO: make this work also during runtime: it should be
4365 * possible to force coredump from running process and then
4366 * continue processing. For example qemu could set up SIGUSR2
4367 * handler (provided that target process haven't registered
4368 * handler for that) that does the dump when signal is received.
4369 */
elf_core_dump(int signr,const CPUArchState * env)4370 static int elf_core_dump(int signr, const CPUArchState *env)
4371 {
4372 const CPUState *cpu = env_cpu_const(env);
4373 const TaskState *ts = (const TaskState *)get_task_state((CPUState *)cpu);
4374 struct rlimit dumpsize;
4375 CountAndSizeRegions css;
4376 off_t offset, note_offset, data_offset;
4377 size_t note_size;
4378 int cpus, ret;
4379 int fd = -1;
4380 CPUState *cpu_iter;
4381
4382 if (prctl(PR_GET_DUMPABLE) == 0) {
4383 return 0;
4384 }
4385
4386 if (getrlimit(RLIMIT_CORE, &dumpsize) < 0 || dumpsize.rlim_cur == 0) {
4387 return 0;
4388 }
4389
4390 cpu_list_lock();
4391 mmap_lock();
4392
4393 /* By unprotecting, we merge vmas that might be split. */
4394 walk_memory_regions(NULL, wmr_page_unprotect_regions);
4395
4396 /*
4397 * Walk through target process memory mappings and
4398 * set up structure containing this information.
4399 */
4400 memset(&css, 0, sizeof(css));
4401 walk_memory_regions(&css, wmr_count_and_size_regions);
4402
4403 cpus = 0;
4404 CPU_FOREACH(cpu_iter) {
4405 cpus++;
4406 }
4407
4408 offset = sizeof(struct elfhdr);
4409 offset += (css.count + 1) * sizeof(struct elf_phdr);
4410 note_offset = offset;
4411
4412 offset += size_note("CORE", ts->info->auxv_len);
4413 offset += size_note("CORE", sizeof(struct target_elf_prpsinfo));
4414 offset += size_note("CORE", sizeof(struct target_elf_prstatus)) * cpus;
4415 note_size = offset - note_offset;
4416 data_offset = ROUND_UP(offset, ELF_EXEC_PAGESIZE);
4417
4418 /* Do not dump if the corefile size exceeds the limit. */
4419 if (dumpsize.rlim_cur != RLIM_INFINITY
4420 && dumpsize.rlim_cur < data_offset + css.size) {
4421 errno = 0;
4422 goto out;
4423 }
4424
4425 {
4426 g_autofree char *corefile = core_dump_filename(ts);
4427 fd = open(corefile, O_WRONLY | O_CREAT | O_TRUNC,
4428 S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH);
4429 }
4430 if (fd < 0) {
4431 goto out;
4432 }
4433
4434 /*
4435 * There is a fair amount of alignment padding within the notes
4436 * as well as preceeding the process memory. Allocate a zeroed
4437 * block to hold it all. Write all of the headers directly into
4438 * this buffer and then write it out as a block.
4439 */
4440 {
4441 g_autofree void *header = g_malloc0(data_offset);
4442 FillRegionPhdr frp;
4443 void *hptr, *dptr;
4444
4445 /* Create elf file header. */
4446 hptr = header;
4447 fill_elf_header(hptr, css.count + 1, ELF_MACHINE, 0);
4448 hptr += sizeof(struct elfhdr);
4449
4450 /* Create elf program headers. */
4451 fill_elf_note_phdr(hptr, note_size, note_offset);
4452 hptr += sizeof(struct elf_phdr);
4453
4454 frp.phdr = hptr;
4455 frp.offset = data_offset;
4456 walk_memory_regions(&frp, wmr_fill_region_phdr);
4457 hptr = frp.phdr;
4458
4459 /* Create the notes. */
4460 dptr = fill_note(&hptr, NT_AUXV, "CORE", ts->info->auxv_len);
4461 fill_auxv_note(dptr, ts);
4462
4463 dptr = fill_note(&hptr, NT_PRPSINFO, "CORE",
4464 sizeof(struct target_elf_prpsinfo));
4465 fill_prpsinfo_note(dptr, ts);
4466
4467 CPU_FOREACH(cpu_iter) {
4468 dptr = fill_note(&hptr, NT_PRSTATUS, "CORE",
4469 sizeof(struct target_elf_prstatus));
4470 fill_prstatus_note(dptr, cpu_iter, cpu_iter == cpu ? signr : 0);
4471 }
4472
4473 if (dump_write(fd, header, data_offset) < 0) {
4474 goto out;
4475 }
4476 }
4477
4478 /*
4479 * Finally write process memory into the corefile as well.
4480 */
4481 if (walk_memory_regions(&fd, wmr_write_region) < 0) {
4482 goto out;
4483 }
4484 errno = 0;
4485
4486 out:
4487 ret = -errno;
4488 mmap_unlock();
4489 cpu_list_unlock();
4490 if (fd >= 0) {
4491 close(fd);
4492 }
4493 return ret;
4494 }
4495 #endif /* USE_ELF_CORE_DUMP */
4496
do_init_thread(struct target_pt_regs * regs,struct image_info * infop)4497 void do_init_thread(struct target_pt_regs *regs, struct image_info *infop)
4498 {
4499 init_thread(regs, infop);
4500 }
4501