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 = 1ULL << 31,
755 ARM_HWCAP_A64_GCS = 1ULL << 32,
756 ARM_HWCAP_A64_CMPBR = 1ULL << 33,
757 ARM_HWCAP_A64_FPRCVT = 1ULL << 34,
758 ARM_HWCAP_A64_F8MM8 = 1ULL << 35,
759 ARM_HWCAP_A64_F8MM4 = 1ULL << 36,
760 ARM_HWCAP_A64_SVE_F16MM = 1ULL << 37,
761 ARM_HWCAP_A64_SVE_ELTPERM = 1ULL << 38,
762 ARM_HWCAP_A64_SVE_AES2 = 1ULL << 39,
763 ARM_HWCAP_A64_SVE_BFSCALE = 1ULL << 40,
764 ARM_HWCAP_A64_SVE2P2 = 1ULL << 41,
765 ARM_HWCAP_A64_SME2P2 = 1ULL << 42,
766 ARM_HWCAP_A64_SME_SBITPERM = 1ULL << 43,
767 ARM_HWCAP_A64_SME_AES = 1ULL << 44,
768 ARM_HWCAP_A64_SME_SFEXPA = 1ULL << 45,
769 ARM_HWCAP_A64_SME_STMOP = 1ULL << 46,
770 ARM_HWCAP_A64_SME_SMOP4 = 1ULL << 47,
771
772 ARM_HWCAP2_A64_DCPODP = 1 << 0,
773 ARM_HWCAP2_A64_SVE2 = 1 << 1,
774 ARM_HWCAP2_A64_SVEAES = 1 << 2,
775 ARM_HWCAP2_A64_SVEPMULL = 1 << 3,
776 ARM_HWCAP2_A64_SVEBITPERM = 1 << 4,
777 ARM_HWCAP2_A64_SVESHA3 = 1 << 5,
778 ARM_HWCAP2_A64_SVESM4 = 1 << 6,
779 ARM_HWCAP2_A64_FLAGM2 = 1 << 7,
780 ARM_HWCAP2_A64_FRINT = 1 << 8,
781 ARM_HWCAP2_A64_SVEI8MM = 1 << 9,
782 ARM_HWCAP2_A64_SVEF32MM = 1 << 10,
783 ARM_HWCAP2_A64_SVEF64MM = 1 << 11,
784 ARM_HWCAP2_A64_SVEBF16 = 1 << 12,
785 ARM_HWCAP2_A64_I8MM = 1 << 13,
786 ARM_HWCAP2_A64_BF16 = 1 << 14,
787 ARM_HWCAP2_A64_DGH = 1 << 15,
788 ARM_HWCAP2_A64_RNG = 1 << 16,
789 ARM_HWCAP2_A64_BTI = 1 << 17,
790 ARM_HWCAP2_A64_MTE = 1 << 18,
791 ARM_HWCAP2_A64_ECV = 1 << 19,
792 ARM_HWCAP2_A64_AFP = 1 << 20,
793 ARM_HWCAP2_A64_RPRES = 1 << 21,
794 ARM_HWCAP2_A64_MTE3 = 1 << 22,
795 ARM_HWCAP2_A64_SME = 1 << 23,
796 ARM_HWCAP2_A64_SME_I16I64 = 1 << 24,
797 ARM_HWCAP2_A64_SME_F64F64 = 1 << 25,
798 ARM_HWCAP2_A64_SME_I8I32 = 1 << 26,
799 ARM_HWCAP2_A64_SME_F16F32 = 1 << 27,
800 ARM_HWCAP2_A64_SME_B16F32 = 1 << 28,
801 ARM_HWCAP2_A64_SME_F32F32 = 1 << 29,
802 ARM_HWCAP2_A64_SME_FA64 = 1 << 30,
803 ARM_HWCAP2_A64_WFXT = 1ULL << 31,
804 ARM_HWCAP2_A64_EBF16 = 1ULL << 32,
805 ARM_HWCAP2_A64_SVE_EBF16 = 1ULL << 33,
806 ARM_HWCAP2_A64_CSSC = 1ULL << 34,
807 ARM_HWCAP2_A64_RPRFM = 1ULL << 35,
808 ARM_HWCAP2_A64_SVE2P1 = 1ULL << 36,
809 ARM_HWCAP2_A64_SME2 = 1ULL << 37,
810 ARM_HWCAP2_A64_SME2P1 = 1ULL << 38,
811 ARM_HWCAP2_A64_SME_I16I32 = 1ULL << 39,
812 ARM_HWCAP2_A64_SME_BI32I32 = 1ULL << 40,
813 ARM_HWCAP2_A64_SME_B16B16 = 1ULL << 41,
814 ARM_HWCAP2_A64_SME_F16F16 = 1ULL << 42,
815 ARM_HWCAP2_A64_MOPS = 1ULL << 43,
816 ARM_HWCAP2_A64_HBC = 1ULL << 44,
817 ARM_HWCAP2_A64_SVE_B16B16 = 1ULL << 45,
818 ARM_HWCAP2_A64_LRCPC3 = 1ULL << 46,
819 ARM_HWCAP2_A64_LSE128 = 1ULL << 47,
820 ARM_HWCAP2_A64_FPMR = 1ULL << 48,
821 ARM_HWCAP2_A64_LUT = 1ULL << 49,
822 ARM_HWCAP2_A64_FAMINMAX = 1ULL << 50,
823 ARM_HWCAP2_A64_F8CVT = 1ULL << 51,
824 ARM_HWCAP2_A64_F8FMA = 1ULL << 52,
825 ARM_HWCAP2_A64_F8DP4 = 1ULL << 53,
826 ARM_HWCAP2_A64_F8DP2 = 1ULL << 54,
827 ARM_HWCAP2_A64_F8E4M3 = 1ULL << 55,
828 ARM_HWCAP2_A64_F8E5M2 = 1ULL << 56,
829 ARM_HWCAP2_A64_SME_LUTV2 = 1ULL << 57,
830 ARM_HWCAP2_A64_SME_F8F16 = 1ULL << 58,
831 ARM_HWCAP2_A64_SME_F8F32 = 1ULL << 59,
832 ARM_HWCAP2_A64_SME_SF8FMA = 1ULL << 60,
833 ARM_HWCAP2_A64_SME_SF8DP4 = 1ULL << 61,
834 ARM_HWCAP2_A64_SME_SF8DP2 = 1ULL << 62,
835 ARM_HWCAP2_A64_POE = 1ULL << 63,
836 };
837
838 #define ELF_HWCAP get_elf_hwcap()
839 #define ELF_HWCAP2 get_elf_hwcap2()
840
841 #define GET_FEATURE_ID(feat, hwcap) \
842 do { if (cpu_isar_feature(feat, cpu)) { hwcaps |= hwcap; } } while (0)
843
get_elf_hwcap(void)844 uint32_t get_elf_hwcap(void)
845 {
846 ARMCPU *cpu = ARM_CPU(thread_cpu);
847 uint32_t hwcaps = 0;
848
849 hwcaps |= ARM_HWCAP_A64_FP;
850 hwcaps |= ARM_HWCAP_A64_ASIMD;
851 hwcaps |= ARM_HWCAP_A64_CPUID;
852
853 /* probe for the extra features */
854
855 GET_FEATURE_ID(aa64_aes, ARM_HWCAP_A64_AES);
856 GET_FEATURE_ID(aa64_pmull, ARM_HWCAP_A64_PMULL);
857 GET_FEATURE_ID(aa64_sha1, ARM_HWCAP_A64_SHA1);
858 GET_FEATURE_ID(aa64_sha256, ARM_HWCAP_A64_SHA2);
859 GET_FEATURE_ID(aa64_sha512, ARM_HWCAP_A64_SHA512);
860 GET_FEATURE_ID(aa64_crc32, ARM_HWCAP_A64_CRC32);
861 GET_FEATURE_ID(aa64_sha3, ARM_HWCAP_A64_SHA3);
862 GET_FEATURE_ID(aa64_sm3, ARM_HWCAP_A64_SM3);
863 GET_FEATURE_ID(aa64_sm4, ARM_HWCAP_A64_SM4);
864 GET_FEATURE_ID(aa64_fp16, ARM_HWCAP_A64_FPHP | ARM_HWCAP_A64_ASIMDHP);
865 GET_FEATURE_ID(aa64_atomics, ARM_HWCAP_A64_ATOMICS);
866 GET_FEATURE_ID(aa64_lse2, ARM_HWCAP_A64_USCAT);
867 GET_FEATURE_ID(aa64_rdm, ARM_HWCAP_A64_ASIMDRDM);
868 GET_FEATURE_ID(aa64_dp, ARM_HWCAP_A64_ASIMDDP);
869 GET_FEATURE_ID(aa64_fcma, ARM_HWCAP_A64_FCMA);
870 GET_FEATURE_ID(aa64_sve, ARM_HWCAP_A64_SVE);
871 GET_FEATURE_ID(aa64_pauth, ARM_HWCAP_A64_PACA | ARM_HWCAP_A64_PACG);
872 GET_FEATURE_ID(aa64_fhm, ARM_HWCAP_A64_ASIMDFHM);
873 GET_FEATURE_ID(aa64_dit, ARM_HWCAP_A64_DIT);
874 GET_FEATURE_ID(aa64_jscvt, ARM_HWCAP_A64_JSCVT);
875 GET_FEATURE_ID(aa64_sb, ARM_HWCAP_A64_SB);
876 GET_FEATURE_ID(aa64_condm_4, ARM_HWCAP_A64_FLAGM);
877 GET_FEATURE_ID(aa64_dcpop, ARM_HWCAP_A64_DCPOP);
878 GET_FEATURE_ID(aa64_rcpc_8_3, ARM_HWCAP_A64_LRCPC);
879 GET_FEATURE_ID(aa64_rcpc_8_4, ARM_HWCAP_A64_ILRCPC);
880
881 return hwcaps;
882 }
883
get_elf_hwcap2(void)884 uint64_t get_elf_hwcap2(void)
885 {
886 ARMCPU *cpu = ARM_CPU(thread_cpu);
887 uint64_t hwcaps = 0;
888
889 GET_FEATURE_ID(aa64_dcpodp, ARM_HWCAP2_A64_DCPODP);
890 GET_FEATURE_ID(aa64_sve2, ARM_HWCAP2_A64_SVE2);
891 GET_FEATURE_ID(aa64_sve2_aes, ARM_HWCAP2_A64_SVEAES);
892 GET_FEATURE_ID(aa64_sve2_pmull128, ARM_HWCAP2_A64_SVEPMULL);
893 GET_FEATURE_ID(aa64_sve2_bitperm, ARM_HWCAP2_A64_SVEBITPERM);
894 GET_FEATURE_ID(aa64_sve2_sha3, ARM_HWCAP2_A64_SVESHA3);
895 GET_FEATURE_ID(aa64_sve2_sm4, ARM_HWCAP2_A64_SVESM4);
896 GET_FEATURE_ID(aa64_condm_5, ARM_HWCAP2_A64_FLAGM2);
897 GET_FEATURE_ID(aa64_frint, ARM_HWCAP2_A64_FRINT);
898 GET_FEATURE_ID(aa64_sve_i8mm, ARM_HWCAP2_A64_SVEI8MM);
899 GET_FEATURE_ID(aa64_sve_f32mm, ARM_HWCAP2_A64_SVEF32MM);
900 GET_FEATURE_ID(aa64_sve_f64mm, ARM_HWCAP2_A64_SVEF64MM);
901 GET_FEATURE_ID(aa64_sve_bf16, ARM_HWCAP2_A64_SVEBF16);
902 GET_FEATURE_ID(aa64_i8mm, ARM_HWCAP2_A64_I8MM);
903 GET_FEATURE_ID(aa64_bf16, ARM_HWCAP2_A64_BF16);
904 GET_FEATURE_ID(aa64_rndr, ARM_HWCAP2_A64_RNG);
905 GET_FEATURE_ID(aa64_bti, ARM_HWCAP2_A64_BTI);
906 GET_FEATURE_ID(aa64_mte, ARM_HWCAP2_A64_MTE);
907 GET_FEATURE_ID(aa64_mte3, ARM_HWCAP2_A64_MTE3);
908 GET_FEATURE_ID(aa64_sme, (ARM_HWCAP2_A64_SME |
909 ARM_HWCAP2_A64_SME_F32F32 |
910 ARM_HWCAP2_A64_SME_B16F32 |
911 ARM_HWCAP2_A64_SME_F16F32 |
912 ARM_HWCAP2_A64_SME_I8I32));
913 GET_FEATURE_ID(aa64_sme_f64f64, ARM_HWCAP2_A64_SME_F64F64);
914 GET_FEATURE_ID(aa64_sme_i16i64, ARM_HWCAP2_A64_SME_I16I64);
915 GET_FEATURE_ID(aa64_sme_fa64, ARM_HWCAP2_A64_SME_FA64);
916 GET_FEATURE_ID(aa64_hbc, ARM_HWCAP2_A64_HBC);
917 GET_FEATURE_ID(aa64_mops, ARM_HWCAP2_A64_MOPS);
918
919 return hwcaps;
920 }
921
elf_hwcap_str(uint32_t bit)922 const char *elf_hwcap_str(uint32_t bit)
923 {
924 static const char * const hwcap_str[] = {
925 [__builtin_ctz(ARM_HWCAP_A64_FP )] = "fp",
926 [__builtin_ctz(ARM_HWCAP_A64_ASIMD )] = "asimd",
927 [__builtin_ctz(ARM_HWCAP_A64_EVTSTRM )] = "evtstrm",
928 [__builtin_ctz(ARM_HWCAP_A64_AES )] = "aes",
929 [__builtin_ctz(ARM_HWCAP_A64_PMULL )] = "pmull",
930 [__builtin_ctz(ARM_HWCAP_A64_SHA1 )] = "sha1",
931 [__builtin_ctz(ARM_HWCAP_A64_SHA2 )] = "sha2",
932 [__builtin_ctz(ARM_HWCAP_A64_CRC32 )] = "crc32",
933 [__builtin_ctz(ARM_HWCAP_A64_ATOMICS )] = "atomics",
934 [__builtin_ctz(ARM_HWCAP_A64_FPHP )] = "fphp",
935 [__builtin_ctz(ARM_HWCAP_A64_ASIMDHP )] = "asimdhp",
936 [__builtin_ctz(ARM_HWCAP_A64_CPUID )] = "cpuid",
937 [__builtin_ctz(ARM_HWCAP_A64_ASIMDRDM)] = "asimdrdm",
938 [__builtin_ctz(ARM_HWCAP_A64_JSCVT )] = "jscvt",
939 [__builtin_ctz(ARM_HWCAP_A64_FCMA )] = "fcma",
940 [__builtin_ctz(ARM_HWCAP_A64_LRCPC )] = "lrcpc",
941 [__builtin_ctz(ARM_HWCAP_A64_DCPOP )] = "dcpop",
942 [__builtin_ctz(ARM_HWCAP_A64_SHA3 )] = "sha3",
943 [__builtin_ctz(ARM_HWCAP_A64_SM3 )] = "sm3",
944 [__builtin_ctz(ARM_HWCAP_A64_SM4 )] = "sm4",
945 [__builtin_ctz(ARM_HWCAP_A64_ASIMDDP )] = "asimddp",
946 [__builtin_ctz(ARM_HWCAP_A64_SHA512 )] = "sha512",
947 [__builtin_ctz(ARM_HWCAP_A64_SVE )] = "sve",
948 [__builtin_ctz(ARM_HWCAP_A64_ASIMDFHM)] = "asimdfhm",
949 [__builtin_ctz(ARM_HWCAP_A64_DIT )] = "dit",
950 [__builtin_ctz(ARM_HWCAP_A64_USCAT )] = "uscat",
951 [__builtin_ctz(ARM_HWCAP_A64_ILRCPC )] = "ilrcpc",
952 [__builtin_ctz(ARM_HWCAP_A64_FLAGM )] = "flagm",
953 [__builtin_ctz(ARM_HWCAP_A64_SSBS )] = "ssbs",
954 [__builtin_ctz(ARM_HWCAP_A64_SB )] = "sb",
955 [__builtin_ctz(ARM_HWCAP_A64_PACA )] = "paca",
956 [__builtin_ctz(ARM_HWCAP_A64_PACG )] = "pacg",
957 [__builtin_ctzll(ARM_HWCAP_A64_GCS )] = "gcs",
958 [__builtin_ctzll(ARM_HWCAP_A64_CMPBR )] = "cmpbr",
959 [__builtin_ctzll(ARM_HWCAP_A64_FPRCVT)] = "fprcvt",
960 [__builtin_ctzll(ARM_HWCAP_A64_F8MM8 )] = "f8mm8",
961 [__builtin_ctzll(ARM_HWCAP_A64_F8MM4 )] = "f8mm4",
962 [__builtin_ctzll(ARM_HWCAP_A64_SVE_F16MM)] = "svef16mm",
963 [__builtin_ctzll(ARM_HWCAP_A64_SVE_ELTPERM)] = "sveeltperm",
964 [__builtin_ctzll(ARM_HWCAP_A64_SVE_AES2)] = "sveaes2",
965 [__builtin_ctzll(ARM_HWCAP_A64_SVE_BFSCALE)] = "svebfscale",
966 [__builtin_ctzll(ARM_HWCAP_A64_SVE2P2)] = "sve2p2",
967 [__builtin_ctzll(ARM_HWCAP_A64_SME2P2)] = "sme2p2",
968 [__builtin_ctzll(ARM_HWCAP_A64_SME_SBITPERM)] = "smesbitperm",
969 [__builtin_ctzll(ARM_HWCAP_A64_SME_AES)] = "smeaes",
970 [__builtin_ctzll(ARM_HWCAP_A64_SME_SFEXPA)] = "smesfexpa",
971 [__builtin_ctzll(ARM_HWCAP_A64_SME_STMOP)] = "smestmop",
972 [__builtin_ctzll(ARM_HWCAP_A64_SME_SMOP4)] = "smesmop4",
973 };
974
975 return bit < ARRAY_SIZE(hwcap_str) ? hwcap_str[bit] : NULL;
976 }
977
elf_hwcap2_str(uint32_t bit)978 const char *elf_hwcap2_str(uint32_t bit)
979 {
980 static const char * const hwcap_str[] = {
981 [__builtin_ctz(ARM_HWCAP2_A64_DCPODP )] = "dcpodp",
982 [__builtin_ctz(ARM_HWCAP2_A64_SVE2 )] = "sve2",
983 [__builtin_ctz(ARM_HWCAP2_A64_SVEAES )] = "sveaes",
984 [__builtin_ctz(ARM_HWCAP2_A64_SVEPMULL )] = "svepmull",
985 [__builtin_ctz(ARM_HWCAP2_A64_SVEBITPERM )] = "svebitperm",
986 [__builtin_ctz(ARM_HWCAP2_A64_SVESHA3 )] = "svesha3",
987 [__builtin_ctz(ARM_HWCAP2_A64_SVESM4 )] = "svesm4",
988 [__builtin_ctz(ARM_HWCAP2_A64_FLAGM2 )] = "flagm2",
989 [__builtin_ctz(ARM_HWCAP2_A64_FRINT )] = "frint",
990 [__builtin_ctz(ARM_HWCAP2_A64_SVEI8MM )] = "svei8mm",
991 [__builtin_ctz(ARM_HWCAP2_A64_SVEF32MM )] = "svef32mm",
992 [__builtin_ctz(ARM_HWCAP2_A64_SVEF64MM )] = "svef64mm",
993 [__builtin_ctz(ARM_HWCAP2_A64_SVEBF16 )] = "svebf16",
994 [__builtin_ctz(ARM_HWCAP2_A64_I8MM )] = "i8mm",
995 [__builtin_ctz(ARM_HWCAP2_A64_BF16 )] = "bf16",
996 [__builtin_ctz(ARM_HWCAP2_A64_DGH )] = "dgh",
997 [__builtin_ctz(ARM_HWCAP2_A64_RNG )] = "rng",
998 [__builtin_ctz(ARM_HWCAP2_A64_BTI )] = "bti",
999 [__builtin_ctz(ARM_HWCAP2_A64_MTE )] = "mte",
1000 [__builtin_ctz(ARM_HWCAP2_A64_ECV )] = "ecv",
1001 [__builtin_ctz(ARM_HWCAP2_A64_AFP )] = "afp",
1002 [__builtin_ctz(ARM_HWCAP2_A64_RPRES )] = "rpres",
1003 [__builtin_ctz(ARM_HWCAP2_A64_MTE3 )] = "mte3",
1004 [__builtin_ctz(ARM_HWCAP2_A64_SME )] = "sme",
1005 [__builtin_ctz(ARM_HWCAP2_A64_SME_I16I64 )] = "smei16i64",
1006 [__builtin_ctz(ARM_HWCAP2_A64_SME_F64F64 )] = "smef64f64",
1007 [__builtin_ctz(ARM_HWCAP2_A64_SME_I8I32 )] = "smei8i32",
1008 [__builtin_ctz(ARM_HWCAP2_A64_SME_F16F32 )] = "smef16f32",
1009 [__builtin_ctz(ARM_HWCAP2_A64_SME_B16F32 )] = "smeb16f32",
1010 [__builtin_ctz(ARM_HWCAP2_A64_SME_F32F32 )] = "smef32f32",
1011 [__builtin_ctz(ARM_HWCAP2_A64_SME_FA64 )] = "smefa64",
1012 [__builtin_ctz(ARM_HWCAP2_A64_WFXT )] = "wfxt",
1013 [__builtin_ctzll(ARM_HWCAP2_A64_EBF16 )] = "ebf16",
1014 [__builtin_ctzll(ARM_HWCAP2_A64_SVE_EBF16 )] = "sveebf16",
1015 [__builtin_ctzll(ARM_HWCAP2_A64_CSSC )] = "cssc",
1016 [__builtin_ctzll(ARM_HWCAP2_A64_RPRFM )] = "rprfm",
1017 [__builtin_ctzll(ARM_HWCAP2_A64_SVE2P1 )] = "sve2p1",
1018 [__builtin_ctzll(ARM_HWCAP2_A64_SME2 )] = "sme2",
1019 [__builtin_ctzll(ARM_HWCAP2_A64_SME2P1 )] = "sme2p1",
1020 [__builtin_ctzll(ARM_HWCAP2_A64_SME_I16I32 )] = "smei16i32",
1021 [__builtin_ctzll(ARM_HWCAP2_A64_SME_BI32I32)] = "smebi32i32",
1022 [__builtin_ctzll(ARM_HWCAP2_A64_SME_B16B16 )] = "smeb16b16",
1023 [__builtin_ctzll(ARM_HWCAP2_A64_SME_F16F16 )] = "smef16f16",
1024 [__builtin_ctzll(ARM_HWCAP2_A64_MOPS )] = "mops",
1025 [__builtin_ctzll(ARM_HWCAP2_A64_HBC )] = "hbc",
1026 [__builtin_ctzll(ARM_HWCAP2_A64_SVE_B16B16 )] = "sveb16b16",
1027 [__builtin_ctzll(ARM_HWCAP2_A64_LRCPC3 )] = "lrcpc3",
1028 [__builtin_ctzll(ARM_HWCAP2_A64_LSE128 )] = "lse128",
1029 [__builtin_ctzll(ARM_HWCAP2_A64_FPMR )] = "fpmr",
1030 [__builtin_ctzll(ARM_HWCAP2_A64_LUT )] = "lut",
1031 [__builtin_ctzll(ARM_HWCAP2_A64_FAMINMAX )] = "faminmax",
1032 [__builtin_ctzll(ARM_HWCAP2_A64_F8CVT )] = "f8cvt",
1033 [__builtin_ctzll(ARM_HWCAP2_A64_F8FMA )] = "f8fma",
1034 [__builtin_ctzll(ARM_HWCAP2_A64_F8DP4 )] = "f8dp4",
1035 [__builtin_ctzll(ARM_HWCAP2_A64_F8DP2 )] = "f8dp2",
1036 [__builtin_ctzll(ARM_HWCAP2_A64_F8E4M3 )] = "f8e4m3",
1037 [__builtin_ctzll(ARM_HWCAP2_A64_F8E5M2 )] = "f8e5m2",
1038 [__builtin_ctzll(ARM_HWCAP2_A64_SME_LUTV2 )] = "smelutv2",
1039 [__builtin_ctzll(ARM_HWCAP2_A64_SME_F8F16 )] = "smef8f16",
1040 [__builtin_ctzll(ARM_HWCAP2_A64_SME_F8F32 )] = "smef8f32",
1041 [__builtin_ctzll(ARM_HWCAP2_A64_SME_SF8DP4 )] = "smesf8dp4",
1042 [__builtin_ctzll(ARM_HWCAP2_A64_SME_SF8DP2 )] = "smesf8dp2",
1043 [__builtin_ctzll(ARM_HWCAP2_A64_POE )] = "poe",
1044 };
1045
1046 return bit < ARRAY_SIZE(hwcap_str) ? hwcap_str[bit] : NULL;
1047 }
1048
1049 #undef GET_FEATURE_ID
1050
1051 #if TARGET_BIG_ENDIAN
1052 # define VDSO_HEADER "vdso-be.c.inc"
1053 #else
1054 # define VDSO_HEADER "vdso-le.c.inc"
1055 #endif
1056
1057 #endif /* not TARGET_AARCH64 */
1058
1059 #endif /* TARGET_ARM */
1060
1061 #ifdef TARGET_SPARC
1062
1063 #ifndef TARGET_SPARC64
1064 # define ELF_CLASS ELFCLASS32
1065 # define ELF_ARCH EM_SPARC
1066 #elif defined(TARGET_ABI32)
1067 # define ELF_CLASS ELFCLASS32
1068 # define elf_check_arch(x) ((x) == EM_SPARC32PLUS || (x) == EM_SPARC)
1069 #else
1070 # define ELF_CLASS ELFCLASS64
1071 # define ELF_ARCH EM_SPARCV9
1072 #endif
1073
1074 #include "elf.h"
1075
1076 #define ELF_HWCAP get_elf_hwcap()
1077
get_elf_hwcap(void)1078 static uint32_t get_elf_hwcap(void)
1079 {
1080 /* There are not many sparc32 hwcap bits -- we have all of them. */
1081 uint32_t r = HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR |
1082 HWCAP_SPARC_SWAP | HWCAP_SPARC_MULDIV;
1083
1084 #ifdef TARGET_SPARC64
1085 CPUSPARCState *env = cpu_env(thread_cpu);
1086 uint32_t features = env->def.features;
1087
1088 r |= HWCAP_SPARC_V9 | HWCAP_SPARC_V8PLUS;
1089 /* 32x32 multiply and divide are efficient. */
1090 r |= HWCAP_SPARC_MUL32 | HWCAP_SPARC_DIV32;
1091 /* We don't have an internal feature bit for this. */
1092 r |= HWCAP_SPARC_POPC;
1093 r |= features & CPU_FEATURE_FSMULD ? HWCAP_SPARC_FSMULD : 0;
1094 r |= features & CPU_FEATURE_VIS1 ? HWCAP_SPARC_VIS : 0;
1095 r |= features & CPU_FEATURE_VIS2 ? HWCAP_SPARC_VIS2 : 0;
1096 r |= features & CPU_FEATURE_FMAF ? HWCAP_SPARC_FMAF : 0;
1097 r |= features & CPU_FEATURE_VIS3 ? HWCAP_SPARC_VIS3 : 0;
1098 r |= features & CPU_FEATURE_IMA ? HWCAP_SPARC_IMA : 0;
1099 #endif
1100
1101 return r;
1102 }
1103
init_thread(struct target_pt_regs * regs,struct image_info * infop)1104 static inline void init_thread(struct target_pt_regs *regs,
1105 struct image_info *infop)
1106 {
1107 /* Note that target_cpu_copy_regs does not read psr/tstate. */
1108 regs->pc = infop->entry;
1109 regs->npc = regs->pc + 4;
1110 regs->y = 0;
1111 regs->u_regs[14] = (infop->start_stack - 16 * sizeof(abi_ulong)
1112 - TARGET_STACK_BIAS);
1113 }
1114 #endif /* TARGET_SPARC */
1115
1116 #ifdef TARGET_PPC
1117
1118 #define ELF_MACHINE PPC_ELF_MACHINE
1119
1120 #if defined(TARGET_PPC64)
1121
1122 #define elf_check_arch(x) ( (x) == EM_PPC64 )
1123
1124 #define ELF_CLASS ELFCLASS64
1125
1126 #else
1127
1128 #define ELF_CLASS ELFCLASS32
1129 #define EXSTACK_DEFAULT true
1130
1131 #endif
1132
1133 #define ELF_ARCH EM_PPC
1134
1135 /* Feature masks for the Aux Vector Hardware Capabilities (AT_HWCAP).
1136 See arch/powerpc/include/asm/cputable.h. */
1137 enum {
1138 QEMU_PPC_FEATURE_32 = 0x80000000,
1139 QEMU_PPC_FEATURE_64 = 0x40000000,
1140 QEMU_PPC_FEATURE_601_INSTR = 0x20000000,
1141 QEMU_PPC_FEATURE_HAS_ALTIVEC = 0x10000000,
1142 QEMU_PPC_FEATURE_HAS_FPU = 0x08000000,
1143 QEMU_PPC_FEATURE_HAS_MMU = 0x04000000,
1144 QEMU_PPC_FEATURE_HAS_4xxMAC = 0x02000000,
1145 QEMU_PPC_FEATURE_UNIFIED_CACHE = 0x01000000,
1146 QEMU_PPC_FEATURE_HAS_SPE = 0x00800000,
1147 QEMU_PPC_FEATURE_HAS_EFP_SINGLE = 0x00400000,
1148 QEMU_PPC_FEATURE_HAS_EFP_DOUBLE = 0x00200000,
1149 QEMU_PPC_FEATURE_NO_TB = 0x00100000,
1150 QEMU_PPC_FEATURE_POWER4 = 0x00080000,
1151 QEMU_PPC_FEATURE_POWER5 = 0x00040000,
1152 QEMU_PPC_FEATURE_POWER5_PLUS = 0x00020000,
1153 QEMU_PPC_FEATURE_CELL = 0x00010000,
1154 QEMU_PPC_FEATURE_BOOKE = 0x00008000,
1155 QEMU_PPC_FEATURE_SMT = 0x00004000,
1156 QEMU_PPC_FEATURE_ICACHE_SNOOP = 0x00002000,
1157 QEMU_PPC_FEATURE_ARCH_2_05 = 0x00001000,
1158 QEMU_PPC_FEATURE_PA6T = 0x00000800,
1159 QEMU_PPC_FEATURE_HAS_DFP = 0x00000400,
1160 QEMU_PPC_FEATURE_POWER6_EXT = 0x00000200,
1161 QEMU_PPC_FEATURE_ARCH_2_06 = 0x00000100,
1162 QEMU_PPC_FEATURE_HAS_VSX = 0x00000080,
1163 QEMU_PPC_FEATURE_PSERIES_PERFMON_COMPAT = 0x00000040,
1164
1165 QEMU_PPC_FEATURE_TRUE_LE = 0x00000002,
1166 QEMU_PPC_FEATURE_PPC_LE = 0x00000001,
1167
1168 /* Feature definitions in AT_HWCAP2. */
1169 QEMU_PPC_FEATURE2_ARCH_2_07 = 0x80000000, /* ISA 2.07 */
1170 QEMU_PPC_FEATURE2_HAS_HTM = 0x40000000, /* Hardware Transactional Memory */
1171 QEMU_PPC_FEATURE2_HAS_DSCR = 0x20000000, /* Data Stream Control Register */
1172 QEMU_PPC_FEATURE2_HAS_EBB = 0x10000000, /* Event Base Branching */
1173 QEMU_PPC_FEATURE2_HAS_ISEL = 0x08000000, /* Integer Select */
1174 QEMU_PPC_FEATURE2_HAS_TAR = 0x04000000, /* Target Address Register */
1175 QEMU_PPC_FEATURE2_VEC_CRYPTO = 0x02000000,
1176 QEMU_PPC_FEATURE2_HTM_NOSC = 0x01000000,
1177 QEMU_PPC_FEATURE2_ARCH_3_00 = 0x00800000, /* ISA 3.00 */
1178 QEMU_PPC_FEATURE2_HAS_IEEE128 = 0x00400000, /* VSX IEEE Bin Float 128-bit */
1179 QEMU_PPC_FEATURE2_DARN = 0x00200000, /* darn random number insn */
1180 QEMU_PPC_FEATURE2_SCV = 0x00100000, /* scv syscall */
1181 QEMU_PPC_FEATURE2_HTM_NO_SUSPEND = 0x00080000, /* TM w/o suspended state */
1182 QEMU_PPC_FEATURE2_ARCH_3_1 = 0x00040000, /* ISA 3.1 */
1183 QEMU_PPC_FEATURE2_MMA = 0x00020000, /* Matrix-Multiply Assist */
1184 };
1185
1186 #define ELF_HWCAP get_elf_hwcap()
1187
get_elf_hwcap(void)1188 static uint32_t get_elf_hwcap(void)
1189 {
1190 PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);
1191 uint32_t features = 0;
1192
1193 /* We don't have to be terribly complete here; the high points are
1194 Altivec/FP/SPE support. Anything else is just a bonus. */
1195 #define GET_FEATURE(flag, feature) \
1196 do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0)
1197 #define GET_FEATURE2(flags, feature) \
1198 do { \
1199 if ((cpu->env.insns_flags2 & flags) == flags) { \
1200 features |= feature; \
1201 } \
1202 } while (0)
1203 GET_FEATURE(PPC_64B, QEMU_PPC_FEATURE_64);
1204 GET_FEATURE(PPC_FLOAT, QEMU_PPC_FEATURE_HAS_FPU);
1205 GET_FEATURE(PPC_ALTIVEC, QEMU_PPC_FEATURE_HAS_ALTIVEC);
1206 GET_FEATURE(PPC_SPE, QEMU_PPC_FEATURE_HAS_SPE);
1207 GET_FEATURE(PPC_SPE_SINGLE, QEMU_PPC_FEATURE_HAS_EFP_SINGLE);
1208 GET_FEATURE(PPC_SPE_DOUBLE, QEMU_PPC_FEATURE_HAS_EFP_DOUBLE);
1209 GET_FEATURE(PPC_BOOKE, QEMU_PPC_FEATURE_BOOKE);
1210 GET_FEATURE(PPC_405_MAC, QEMU_PPC_FEATURE_HAS_4xxMAC);
1211 GET_FEATURE2(PPC2_DFP, QEMU_PPC_FEATURE_HAS_DFP);
1212 GET_FEATURE2(PPC2_VSX, QEMU_PPC_FEATURE_HAS_VSX);
1213 GET_FEATURE2((PPC2_PERM_ISA206 | PPC2_DIVE_ISA206 | PPC2_ATOMIC_ISA206 |
1214 PPC2_FP_CVT_ISA206 | PPC2_FP_TST_ISA206),
1215 QEMU_PPC_FEATURE_ARCH_2_06);
1216 #undef GET_FEATURE
1217 #undef GET_FEATURE2
1218
1219 return features;
1220 }
1221
1222 #define ELF_HWCAP2 get_elf_hwcap2()
1223
get_elf_hwcap2(void)1224 static uint32_t get_elf_hwcap2(void)
1225 {
1226 PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);
1227 uint32_t features = 0;
1228
1229 #define GET_FEATURE(flag, feature) \
1230 do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0)
1231 #define GET_FEATURE2(flag, feature) \
1232 do { if (cpu->env.insns_flags2 & flag) { features |= feature; } } while (0)
1233
1234 GET_FEATURE(PPC_ISEL, QEMU_PPC_FEATURE2_HAS_ISEL);
1235 GET_FEATURE2(PPC2_BCTAR_ISA207, QEMU_PPC_FEATURE2_HAS_TAR);
1236 GET_FEATURE2((PPC2_BCTAR_ISA207 | PPC2_LSQ_ISA207 | PPC2_ALTIVEC_207 |
1237 PPC2_ISA207S), QEMU_PPC_FEATURE2_ARCH_2_07 |
1238 QEMU_PPC_FEATURE2_VEC_CRYPTO);
1239 GET_FEATURE2(PPC2_ISA300, QEMU_PPC_FEATURE2_ARCH_3_00 |
1240 QEMU_PPC_FEATURE2_DARN | QEMU_PPC_FEATURE2_HAS_IEEE128);
1241 GET_FEATURE2(PPC2_ISA310, QEMU_PPC_FEATURE2_ARCH_3_1 |
1242 QEMU_PPC_FEATURE2_MMA);
1243
1244 #undef GET_FEATURE
1245 #undef GET_FEATURE2
1246
1247 return features;
1248 }
1249
1250 /*
1251 * The requirements here are:
1252 * - keep the final alignment of sp (sp & 0xf)
1253 * - make sure the 32-bit value at the first 16 byte aligned position of
1254 * AUXV is greater than 16 for glibc compatibility.
1255 * AT_IGNOREPPC is used for that.
1256 * - for compatibility with glibc ARCH_DLINFO must always be defined on PPC,
1257 * even if DLINFO_ARCH_ITEMS goes to zero or is undefined.
1258 */
1259 #define DLINFO_ARCH_ITEMS 5
1260 #define ARCH_DLINFO \
1261 do { \
1262 PowerPCCPU *cpu = POWERPC_CPU(thread_cpu); \
1263 /* \
1264 * Handle glibc compatibility: these magic entries must \
1265 * be at the lowest addresses in the final auxv. \
1266 */ \
1267 NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC); \
1268 NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC); \
1269 NEW_AUX_ENT(AT_DCACHEBSIZE, cpu->env.dcache_line_size); \
1270 NEW_AUX_ENT(AT_ICACHEBSIZE, cpu->env.icache_line_size); \
1271 NEW_AUX_ENT(AT_UCACHEBSIZE, 0); \
1272 } while (0)
1273
init_thread(struct target_pt_regs * _regs,struct image_info * infop)1274 static inline void init_thread(struct target_pt_regs *_regs, struct image_info *infop)
1275 {
1276 _regs->gpr[1] = infop->start_stack;
1277 #if defined(TARGET_PPC64)
1278 if (get_ppc64_abi(infop) < 2) {
1279 uint64_t val;
1280 get_user_u64(val, infop->entry + 8);
1281 _regs->gpr[2] = val + infop->load_bias;
1282 get_user_u64(val, infop->entry);
1283 infop->entry = val + infop->load_bias;
1284 } else {
1285 _regs->gpr[12] = infop->entry; /* r12 set to global entry address */
1286 }
1287 #endif
1288 _regs->nip = infop->entry;
1289 }
1290
1291 /* See linux kernel: arch/powerpc/include/asm/elf.h. */
1292 #define ELF_NREG 48
1293 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1294
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUPPCState * env)1295 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUPPCState *env)
1296 {
1297 int i;
1298 target_ulong ccr = 0;
1299
1300 for (i = 0; i < ARRAY_SIZE(env->gpr); i++) {
1301 (*regs)[i] = tswapreg(env->gpr[i]);
1302 }
1303
1304 (*regs)[32] = tswapreg(env->nip);
1305 (*regs)[33] = tswapreg(env->msr);
1306 (*regs)[35] = tswapreg(env->ctr);
1307 (*regs)[36] = tswapreg(env->lr);
1308 (*regs)[37] = tswapreg(cpu_read_xer(env));
1309
1310 ccr = ppc_get_cr(env);
1311 (*regs)[38] = tswapreg(ccr);
1312 }
1313
1314 #define USE_ELF_CORE_DUMP
1315 #define ELF_EXEC_PAGESIZE 4096
1316
1317 #ifndef TARGET_PPC64
1318 # define VDSO_HEADER "vdso-32.c.inc"
1319 #elif TARGET_BIG_ENDIAN
1320 # define VDSO_HEADER "vdso-64.c.inc"
1321 #else
1322 # define VDSO_HEADER "vdso-64le.c.inc"
1323 #endif
1324
1325 #endif
1326
1327 #ifdef TARGET_LOONGARCH64
1328
1329 #define ELF_CLASS ELFCLASS64
1330 #define ELF_ARCH EM_LOONGARCH
1331 #define EXSTACK_DEFAULT true
1332
1333 #define elf_check_arch(x) ((x) == EM_LOONGARCH)
1334
1335 #define VDSO_HEADER "vdso.c.inc"
1336
init_thread(struct target_pt_regs * regs,struct image_info * infop)1337 static inline void init_thread(struct target_pt_regs *regs,
1338 struct image_info *infop)
1339 {
1340 /*Set crmd PG,DA = 1,0 */
1341 regs->csr.crmd = 2 << 3;
1342 regs->csr.era = infop->entry;
1343 regs->regs[3] = infop->start_stack;
1344 }
1345
1346 /* See linux kernel: arch/loongarch/include/asm/elf.h */
1347 #define ELF_NREG 45
1348 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1349
1350 enum {
1351 TARGET_EF_R0 = 0,
1352 TARGET_EF_CSR_ERA = TARGET_EF_R0 + 33,
1353 TARGET_EF_CSR_BADV = TARGET_EF_R0 + 34,
1354 };
1355
elf_core_copy_regs(target_elf_gregset_t * regs,const CPULoongArchState * env)1356 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1357 const CPULoongArchState *env)
1358 {
1359 int i;
1360
1361 (*regs)[TARGET_EF_R0] = 0;
1362
1363 for (i = 1; i < ARRAY_SIZE(env->gpr); i++) {
1364 (*regs)[TARGET_EF_R0 + i] = tswapreg(env->gpr[i]);
1365 }
1366
1367 (*regs)[TARGET_EF_CSR_ERA] = tswapreg(env->pc);
1368 (*regs)[TARGET_EF_CSR_BADV] = tswapreg(env->CSR_BADV);
1369 }
1370
1371 #define USE_ELF_CORE_DUMP
1372 #define ELF_EXEC_PAGESIZE 4096
1373
1374 #define ELF_HWCAP get_elf_hwcap()
1375
1376 /* See arch/loongarch/include/uapi/asm/hwcap.h */
1377 enum {
1378 HWCAP_LOONGARCH_CPUCFG = (1 << 0),
1379 HWCAP_LOONGARCH_LAM = (1 << 1),
1380 HWCAP_LOONGARCH_UAL = (1 << 2),
1381 HWCAP_LOONGARCH_FPU = (1 << 3),
1382 HWCAP_LOONGARCH_LSX = (1 << 4),
1383 HWCAP_LOONGARCH_LASX = (1 << 5),
1384 HWCAP_LOONGARCH_CRC32 = (1 << 6),
1385 HWCAP_LOONGARCH_COMPLEX = (1 << 7),
1386 HWCAP_LOONGARCH_CRYPTO = (1 << 8),
1387 HWCAP_LOONGARCH_LVZ = (1 << 9),
1388 HWCAP_LOONGARCH_LBT_X86 = (1 << 10),
1389 HWCAP_LOONGARCH_LBT_ARM = (1 << 11),
1390 HWCAP_LOONGARCH_LBT_MIPS = (1 << 12),
1391 };
1392
get_elf_hwcap(void)1393 static uint32_t get_elf_hwcap(void)
1394 {
1395 LoongArchCPU *cpu = LOONGARCH_CPU(thread_cpu);
1396 uint32_t hwcaps = 0;
1397
1398 hwcaps |= HWCAP_LOONGARCH_CRC32;
1399
1400 if (FIELD_EX32(cpu->env.cpucfg[1], CPUCFG1, UAL)) {
1401 hwcaps |= HWCAP_LOONGARCH_UAL;
1402 }
1403
1404 if (FIELD_EX32(cpu->env.cpucfg[2], CPUCFG2, FP)) {
1405 hwcaps |= HWCAP_LOONGARCH_FPU;
1406 }
1407
1408 if (FIELD_EX32(cpu->env.cpucfg[2], CPUCFG2, LAM)) {
1409 hwcaps |= HWCAP_LOONGARCH_LAM;
1410 }
1411
1412 if (FIELD_EX32(cpu->env.cpucfg[2], CPUCFG2, LSX)) {
1413 hwcaps |= HWCAP_LOONGARCH_LSX;
1414 }
1415
1416 if (FIELD_EX32(cpu->env.cpucfg[2], CPUCFG2, LASX)) {
1417 hwcaps |= HWCAP_LOONGARCH_LASX;
1418 }
1419
1420 return hwcaps;
1421 }
1422
1423 #define ELF_PLATFORM "loongarch"
1424
1425 #endif /* TARGET_LOONGARCH64 */
1426
1427 #ifdef TARGET_MIPS
1428
1429 #ifdef TARGET_MIPS64
1430 #define ELF_CLASS ELFCLASS64
1431 #else
1432 #define ELF_CLASS ELFCLASS32
1433 #endif
1434 #define ELF_ARCH EM_MIPS
1435 #define EXSTACK_DEFAULT true
1436
1437 #ifdef TARGET_ABI_MIPSN32
1438 #define elf_check_abi(x) ((x) & EF_MIPS_ABI2)
1439 #else
1440 #define elf_check_abi(x) (!((x) & EF_MIPS_ABI2))
1441 #endif
1442
1443 #define ELF_BASE_PLATFORM get_elf_base_platform()
1444
1445 #define MATCH_PLATFORM_INSN(_flags, _base_platform) \
1446 do { if ((cpu->env.insn_flags & (_flags)) == _flags) \
1447 { return _base_platform; } } while (0)
1448
get_elf_base_platform(void)1449 static const char *get_elf_base_platform(void)
1450 {
1451 MIPSCPU *cpu = MIPS_CPU(thread_cpu);
1452
1453 /* 64 bit ISAs goes first */
1454 MATCH_PLATFORM_INSN(CPU_MIPS64R6, "mips64r6");
1455 MATCH_PLATFORM_INSN(CPU_MIPS64R5, "mips64r5");
1456 MATCH_PLATFORM_INSN(CPU_MIPS64R2, "mips64r2");
1457 MATCH_PLATFORM_INSN(CPU_MIPS64R1, "mips64");
1458 MATCH_PLATFORM_INSN(CPU_MIPS5, "mips5");
1459 MATCH_PLATFORM_INSN(CPU_MIPS4, "mips4");
1460 MATCH_PLATFORM_INSN(CPU_MIPS3, "mips3");
1461
1462 /* 32 bit ISAs */
1463 MATCH_PLATFORM_INSN(CPU_MIPS32R6, "mips32r6");
1464 MATCH_PLATFORM_INSN(CPU_MIPS32R5, "mips32r5");
1465 MATCH_PLATFORM_INSN(CPU_MIPS32R2, "mips32r2");
1466 MATCH_PLATFORM_INSN(CPU_MIPS32R1, "mips32");
1467 MATCH_PLATFORM_INSN(CPU_MIPS2, "mips2");
1468
1469 /* Fallback */
1470 return "mips";
1471 }
1472 #undef MATCH_PLATFORM_INSN
1473
init_thread(struct target_pt_regs * regs,struct image_info * infop)1474 static inline void init_thread(struct target_pt_regs *regs,
1475 struct image_info *infop)
1476 {
1477 regs->cp0_status = 2 << CP0St_KSU;
1478 regs->cp0_epc = infop->entry;
1479 regs->regs[29] = infop->start_stack;
1480 }
1481
1482 /* See linux kernel: arch/mips/include/asm/elf.h. */
1483 #define ELF_NREG 45
1484 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1485
1486 /* See linux kernel: arch/mips/include/asm/reg.h. */
1487 enum {
1488 #ifdef TARGET_MIPS64
1489 TARGET_EF_R0 = 0,
1490 #else
1491 TARGET_EF_R0 = 6,
1492 #endif
1493 TARGET_EF_R26 = TARGET_EF_R0 + 26,
1494 TARGET_EF_R27 = TARGET_EF_R0 + 27,
1495 TARGET_EF_LO = TARGET_EF_R0 + 32,
1496 TARGET_EF_HI = TARGET_EF_R0 + 33,
1497 TARGET_EF_CP0_EPC = TARGET_EF_R0 + 34,
1498 TARGET_EF_CP0_BADVADDR = TARGET_EF_R0 + 35,
1499 TARGET_EF_CP0_STATUS = TARGET_EF_R0 + 36,
1500 TARGET_EF_CP0_CAUSE = TARGET_EF_R0 + 37
1501 };
1502
1503 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUMIPSState * env)1504 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMIPSState *env)
1505 {
1506 int i;
1507
1508 for (i = 0; i < TARGET_EF_R0; i++) {
1509 (*regs)[i] = 0;
1510 }
1511 (*regs)[TARGET_EF_R0] = 0;
1512
1513 for (i = 1; i < ARRAY_SIZE(env->active_tc.gpr); i++) {
1514 (*regs)[TARGET_EF_R0 + i] = tswapreg(env->active_tc.gpr[i]);
1515 }
1516
1517 (*regs)[TARGET_EF_R26] = 0;
1518 (*regs)[TARGET_EF_R27] = 0;
1519 (*regs)[TARGET_EF_LO] = tswapreg(env->active_tc.LO[0]);
1520 (*regs)[TARGET_EF_HI] = tswapreg(env->active_tc.HI[0]);
1521 (*regs)[TARGET_EF_CP0_EPC] = tswapreg(env->active_tc.PC);
1522 (*regs)[TARGET_EF_CP0_BADVADDR] = tswapreg(env->CP0_BadVAddr);
1523 (*regs)[TARGET_EF_CP0_STATUS] = tswapreg(env->CP0_Status);
1524 (*regs)[TARGET_EF_CP0_CAUSE] = tswapreg(env->CP0_Cause);
1525 }
1526
1527 #define USE_ELF_CORE_DUMP
1528 #define ELF_EXEC_PAGESIZE 4096
1529
1530 /* See arch/mips/include/uapi/asm/hwcap.h. */
1531 enum {
1532 HWCAP_MIPS_R6 = (1 << 0),
1533 HWCAP_MIPS_MSA = (1 << 1),
1534 HWCAP_MIPS_CRC32 = (1 << 2),
1535 HWCAP_MIPS_MIPS16 = (1 << 3),
1536 HWCAP_MIPS_MDMX = (1 << 4),
1537 HWCAP_MIPS_MIPS3D = (1 << 5),
1538 HWCAP_MIPS_SMARTMIPS = (1 << 6),
1539 HWCAP_MIPS_DSP = (1 << 7),
1540 HWCAP_MIPS_DSP2 = (1 << 8),
1541 HWCAP_MIPS_DSP3 = (1 << 9),
1542 HWCAP_MIPS_MIPS16E2 = (1 << 10),
1543 HWCAP_LOONGSON_MMI = (1 << 11),
1544 HWCAP_LOONGSON_EXT = (1 << 12),
1545 HWCAP_LOONGSON_EXT2 = (1 << 13),
1546 HWCAP_LOONGSON_CPUCFG = (1 << 14),
1547 };
1548
1549 #define ELF_HWCAP get_elf_hwcap()
1550
1551 #define GET_FEATURE_INSN(_flag, _hwcap) \
1552 do { if (cpu->env.insn_flags & (_flag)) { hwcaps |= _hwcap; } } while (0)
1553
1554 #define GET_FEATURE_REG_SET(_reg, _mask, _hwcap) \
1555 do { if (cpu->env._reg & (_mask)) { hwcaps |= _hwcap; } } while (0)
1556
1557 #define GET_FEATURE_REG_EQU(_reg, _start, _length, _val, _hwcap) \
1558 do { \
1559 if (extract32(cpu->env._reg, (_start), (_length)) == (_val)) { \
1560 hwcaps |= _hwcap; \
1561 } \
1562 } while (0)
1563
get_elf_hwcap(void)1564 static uint32_t get_elf_hwcap(void)
1565 {
1566 MIPSCPU *cpu = MIPS_CPU(thread_cpu);
1567 uint32_t hwcaps = 0;
1568
1569 GET_FEATURE_REG_EQU(CP0_Config0, CP0C0_AR, CP0C0_AR_LENGTH,
1570 2, HWCAP_MIPS_R6);
1571 GET_FEATURE_REG_SET(CP0_Config3, 1 << CP0C3_MSAP, HWCAP_MIPS_MSA);
1572 GET_FEATURE_INSN(ASE_LMMI, HWCAP_LOONGSON_MMI);
1573 GET_FEATURE_INSN(ASE_LEXT, HWCAP_LOONGSON_EXT);
1574
1575 return hwcaps;
1576 }
1577
1578 #undef GET_FEATURE_REG_EQU
1579 #undef GET_FEATURE_REG_SET
1580 #undef GET_FEATURE_INSN
1581
1582 #endif /* TARGET_MIPS */
1583
1584 #ifdef TARGET_MICROBLAZE
1585
1586 #define elf_check_arch(x) ( (x) == EM_MICROBLAZE || (x) == EM_MICROBLAZE_OLD)
1587
1588 #define ELF_CLASS ELFCLASS32
1589 #define ELF_ARCH EM_MICROBLAZE
1590
init_thread(struct target_pt_regs * regs,struct image_info * infop)1591 static inline void init_thread(struct target_pt_regs *regs,
1592 struct image_info *infop)
1593 {
1594 regs->pc = infop->entry;
1595 regs->r1 = infop->start_stack;
1596
1597 }
1598
1599 #define ELF_EXEC_PAGESIZE 4096
1600
1601 #define USE_ELF_CORE_DUMP
1602 #define ELF_NREG 38
1603 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1604
1605 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs. */
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUMBState * env)1606 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMBState *env)
1607 {
1608 int i, pos = 0;
1609
1610 for (i = 0; i < 32; i++) {
1611 (*regs)[pos++] = tswapreg(env->regs[i]);
1612 }
1613
1614 (*regs)[pos++] = tswapreg(env->pc);
1615 (*regs)[pos++] = tswapreg(mb_cpu_read_msr(env));
1616 (*regs)[pos++] = 0;
1617 (*regs)[pos++] = tswapreg(env->ear);
1618 (*regs)[pos++] = 0;
1619 (*regs)[pos++] = tswapreg(env->esr);
1620 }
1621
1622 #endif /* TARGET_MICROBLAZE */
1623
1624 #ifdef TARGET_OPENRISC
1625
1626 #define ELF_ARCH EM_OPENRISC
1627 #define ELF_CLASS ELFCLASS32
1628 #define ELF_DATA ELFDATA2MSB
1629
init_thread(struct target_pt_regs * regs,struct image_info * infop)1630 static inline void init_thread(struct target_pt_regs *regs,
1631 struct image_info *infop)
1632 {
1633 regs->pc = infop->entry;
1634 regs->gpr[1] = infop->start_stack;
1635 }
1636
1637 #define USE_ELF_CORE_DUMP
1638 #define ELF_EXEC_PAGESIZE 8192
1639
1640 /* See linux kernel arch/openrisc/include/asm/elf.h. */
1641 #define ELF_NREG 34 /* gprs and pc, sr */
1642 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1643
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUOpenRISCState * env)1644 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1645 const CPUOpenRISCState *env)
1646 {
1647 int i;
1648
1649 for (i = 0; i < 32; i++) {
1650 (*regs)[i] = tswapreg(cpu_get_gpr(env, i));
1651 }
1652 (*regs)[32] = tswapreg(env->pc);
1653 (*regs)[33] = tswapreg(cpu_get_sr(env));
1654 }
1655 #define ELF_HWCAP 0
1656 #define ELF_PLATFORM NULL
1657
1658 #endif /* TARGET_OPENRISC */
1659
1660 #ifdef TARGET_SH4
1661
1662 #define ELF_CLASS ELFCLASS32
1663 #define ELF_ARCH EM_SH
1664
init_thread(struct target_pt_regs * regs,struct image_info * infop)1665 static inline void init_thread(struct target_pt_regs *regs,
1666 struct image_info *infop)
1667 {
1668 /* Check other registers XXXXX */
1669 regs->pc = infop->entry;
1670 regs->regs[15] = infop->start_stack;
1671 }
1672
1673 /* See linux kernel: arch/sh/include/asm/elf.h. */
1674 #define ELF_NREG 23
1675 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1676
1677 /* See linux kernel: arch/sh/include/asm/ptrace.h. */
1678 enum {
1679 TARGET_REG_PC = 16,
1680 TARGET_REG_PR = 17,
1681 TARGET_REG_SR = 18,
1682 TARGET_REG_GBR = 19,
1683 TARGET_REG_MACH = 20,
1684 TARGET_REG_MACL = 21,
1685 TARGET_REG_SYSCALL = 22
1686 };
1687
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUSH4State * env)1688 static inline void elf_core_copy_regs(target_elf_gregset_t *regs,
1689 const CPUSH4State *env)
1690 {
1691 int i;
1692
1693 for (i = 0; i < 16; i++) {
1694 (*regs)[i] = tswapreg(env->gregs[i]);
1695 }
1696
1697 (*regs)[TARGET_REG_PC] = tswapreg(env->pc);
1698 (*regs)[TARGET_REG_PR] = tswapreg(env->pr);
1699 (*regs)[TARGET_REG_SR] = tswapreg(env->sr);
1700 (*regs)[TARGET_REG_GBR] = tswapreg(env->gbr);
1701 (*regs)[TARGET_REG_MACH] = tswapreg(env->mach);
1702 (*regs)[TARGET_REG_MACL] = tswapreg(env->macl);
1703 (*regs)[TARGET_REG_SYSCALL] = 0; /* FIXME */
1704 }
1705
1706 #define USE_ELF_CORE_DUMP
1707 #define ELF_EXEC_PAGESIZE 4096
1708
1709 enum {
1710 SH_CPU_HAS_FPU = 0x0001, /* Hardware FPU support */
1711 SH_CPU_HAS_P2_FLUSH_BUG = 0x0002, /* Need to flush the cache in P2 area */
1712 SH_CPU_HAS_MMU_PAGE_ASSOC = 0x0004, /* SH3: TLB way selection bit support */
1713 SH_CPU_HAS_DSP = 0x0008, /* SH-DSP: DSP support */
1714 SH_CPU_HAS_PERF_COUNTER = 0x0010, /* Hardware performance counters */
1715 SH_CPU_HAS_PTEA = 0x0020, /* PTEA register */
1716 SH_CPU_HAS_LLSC = 0x0040, /* movli.l/movco.l */
1717 SH_CPU_HAS_L2_CACHE = 0x0080, /* Secondary cache / URAM */
1718 SH_CPU_HAS_OP32 = 0x0100, /* 32-bit instruction support */
1719 SH_CPU_HAS_PTEAEX = 0x0200, /* PTE ASID Extension support */
1720 };
1721
1722 #define ELF_HWCAP get_elf_hwcap()
1723
get_elf_hwcap(void)1724 static uint32_t get_elf_hwcap(void)
1725 {
1726 SuperHCPU *cpu = SUPERH_CPU(thread_cpu);
1727 uint32_t hwcap = 0;
1728
1729 hwcap |= SH_CPU_HAS_FPU;
1730
1731 if (cpu->env.features & SH_FEATURE_SH4A) {
1732 hwcap |= SH_CPU_HAS_LLSC;
1733 }
1734
1735 return hwcap;
1736 }
1737
1738 #endif
1739
1740 #ifdef TARGET_M68K
1741
1742 #define ELF_CLASS ELFCLASS32
1743 #define ELF_ARCH EM_68K
1744
1745 /* ??? Does this need to do anything?
1746 #define ELF_PLAT_INIT(_r) */
1747
init_thread(struct target_pt_regs * regs,struct image_info * infop)1748 static inline void init_thread(struct target_pt_regs *regs,
1749 struct image_info *infop)
1750 {
1751 regs->usp = infop->start_stack;
1752 regs->sr = 0;
1753 regs->pc = infop->entry;
1754 }
1755
1756 /* See linux kernel: arch/m68k/include/asm/elf.h. */
1757 #define ELF_NREG 20
1758 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1759
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUM68KState * env)1760 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUM68KState *env)
1761 {
1762 (*regs)[0] = tswapreg(env->dregs[1]);
1763 (*regs)[1] = tswapreg(env->dregs[2]);
1764 (*regs)[2] = tswapreg(env->dregs[3]);
1765 (*regs)[3] = tswapreg(env->dregs[4]);
1766 (*regs)[4] = tswapreg(env->dregs[5]);
1767 (*regs)[5] = tswapreg(env->dregs[6]);
1768 (*regs)[6] = tswapreg(env->dregs[7]);
1769 (*regs)[7] = tswapreg(env->aregs[0]);
1770 (*regs)[8] = tswapreg(env->aregs[1]);
1771 (*regs)[9] = tswapreg(env->aregs[2]);
1772 (*regs)[10] = tswapreg(env->aregs[3]);
1773 (*regs)[11] = tswapreg(env->aregs[4]);
1774 (*regs)[12] = tswapreg(env->aregs[5]);
1775 (*regs)[13] = tswapreg(env->aregs[6]);
1776 (*regs)[14] = tswapreg(env->dregs[0]);
1777 (*regs)[15] = tswapreg(env->aregs[7]);
1778 (*regs)[16] = tswapreg(env->dregs[0]); /* FIXME: orig_d0 */
1779 (*regs)[17] = tswapreg(env->sr);
1780 (*regs)[18] = tswapreg(env->pc);
1781 (*regs)[19] = 0; /* FIXME: regs->format | regs->vector */
1782 }
1783
1784 #define USE_ELF_CORE_DUMP
1785 #define ELF_EXEC_PAGESIZE 8192
1786
1787 #endif
1788
1789 #ifdef TARGET_ALPHA
1790
1791 #define ELF_CLASS ELFCLASS64
1792 #define ELF_ARCH EM_ALPHA
1793
init_thread(struct target_pt_regs * regs,struct image_info * infop)1794 static inline void init_thread(struct target_pt_regs *regs,
1795 struct image_info *infop)
1796 {
1797 regs->pc = infop->entry;
1798 regs->ps = 8;
1799 regs->usp = infop->start_stack;
1800 }
1801
1802 #define ELF_EXEC_PAGESIZE 8192
1803
1804 #endif /* TARGET_ALPHA */
1805
1806 #ifdef TARGET_S390X
1807
1808 #define ELF_CLASS ELFCLASS64
1809 #define ELF_DATA ELFDATA2MSB
1810 #define ELF_ARCH EM_S390
1811
1812 #include "elf.h"
1813
1814 #define ELF_HWCAP get_elf_hwcap()
1815
1816 #define GET_FEATURE(_feat, _hwcap) \
1817 do { if (s390_has_feat(_feat)) { hwcap |= _hwcap; } } while (0)
1818
get_elf_hwcap(void)1819 uint32_t get_elf_hwcap(void)
1820 {
1821 /*
1822 * Let's assume we always have esan3 and zarch.
1823 * 31-bit processes can use 64-bit registers (high gprs).
1824 */
1825 uint32_t hwcap = HWCAP_S390_ESAN3 | HWCAP_S390_ZARCH | HWCAP_S390_HIGH_GPRS;
1826
1827 GET_FEATURE(S390_FEAT_STFLE, HWCAP_S390_STFLE);
1828 GET_FEATURE(S390_FEAT_MSA, HWCAP_S390_MSA);
1829 GET_FEATURE(S390_FEAT_LONG_DISPLACEMENT, HWCAP_S390_LDISP);
1830 GET_FEATURE(S390_FEAT_EXTENDED_IMMEDIATE, HWCAP_S390_EIMM);
1831 if (s390_has_feat(S390_FEAT_EXTENDED_TRANSLATION_3) &&
1832 s390_has_feat(S390_FEAT_ETF3_ENH)) {
1833 hwcap |= HWCAP_S390_ETF3EH;
1834 }
1835 GET_FEATURE(S390_FEAT_VECTOR, HWCAP_S390_VXRS);
1836 GET_FEATURE(S390_FEAT_VECTOR_ENH, HWCAP_S390_VXRS_EXT);
1837 GET_FEATURE(S390_FEAT_VECTOR_ENH2, HWCAP_S390_VXRS_EXT2);
1838
1839 return hwcap;
1840 }
1841
elf_hwcap_str(uint32_t bit)1842 const char *elf_hwcap_str(uint32_t bit)
1843 {
1844 static const char *hwcap_str[] = {
1845 [HWCAP_S390_NR_ESAN3] = "esan3",
1846 [HWCAP_S390_NR_ZARCH] = "zarch",
1847 [HWCAP_S390_NR_STFLE] = "stfle",
1848 [HWCAP_S390_NR_MSA] = "msa",
1849 [HWCAP_S390_NR_LDISP] = "ldisp",
1850 [HWCAP_S390_NR_EIMM] = "eimm",
1851 [HWCAP_S390_NR_DFP] = "dfp",
1852 [HWCAP_S390_NR_HPAGE] = "edat",
1853 [HWCAP_S390_NR_ETF3EH] = "etf3eh",
1854 [HWCAP_S390_NR_HIGH_GPRS] = "highgprs",
1855 [HWCAP_S390_NR_TE] = "te",
1856 [HWCAP_S390_NR_VXRS] = "vx",
1857 [HWCAP_S390_NR_VXRS_BCD] = "vxd",
1858 [HWCAP_S390_NR_VXRS_EXT] = "vxe",
1859 [HWCAP_S390_NR_GS] = "gs",
1860 [HWCAP_S390_NR_VXRS_EXT2] = "vxe2",
1861 [HWCAP_S390_NR_VXRS_PDE] = "vxp",
1862 [HWCAP_S390_NR_SORT] = "sort",
1863 [HWCAP_S390_NR_DFLT] = "dflt",
1864 [HWCAP_S390_NR_NNPA] = "nnpa",
1865 [HWCAP_S390_NR_PCI_MIO] = "pcimio",
1866 [HWCAP_S390_NR_SIE] = "sie",
1867 };
1868
1869 return bit < ARRAY_SIZE(hwcap_str) ? hwcap_str[bit] : NULL;
1870 }
1871
init_thread(struct target_pt_regs * regs,struct image_info * infop)1872 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop)
1873 {
1874 regs->psw.addr = infop->entry;
1875 regs->psw.mask = PSW_MASK_DAT | PSW_MASK_IO | PSW_MASK_EXT | \
1876 PSW_MASK_MCHECK | PSW_MASK_PSTATE | PSW_MASK_64 | \
1877 PSW_MASK_32;
1878 regs->gprs[15] = infop->start_stack;
1879 }
1880
1881 /* See linux kernel: arch/s390/include/uapi/asm/ptrace.h (s390_regs). */
1882 #define ELF_NREG 27
1883 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1884
1885 enum {
1886 TARGET_REG_PSWM = 0,
1887 TARGET_REG_PSWA = 1,
1888 TARGET_REG_GPRS = 2,
1889 TARGET_REG_ARS = 18,
1890 TARGET_REG_ORIG_R2 = 26,
1891 };
1892
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUS390XState * env)1893 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1894 const CPUS390XState *env)
1895 {
1896 int i;
1897 uint32_t *aregs;
1898
1899 (*regs)[TARGET_REG_PSWM] = tswapreg(env->psw.mask);
1900 (*regs)[TARGET_REG_PSWA] = tswapreg(env->psw.addr);
1901 for (i = 0; i < 16; i++) {
1902 (*regs)[TARGET_REG_GPRS + i] = tswapreg(env->regs[i]);
1903 }
1904 aregs = (uint32_t *)&((*regs)[TARGET_REG_ARS]);
1905 for (i = 0; i < 16; i++) {
1906 aregs[i] = tswap32(env->aregs[i]);
1907 }
1908 (*regs)[TARGET_REG_ORIG_R2] = 0;
1909 }
1910
1911 #define USE_ELF_CORE_DUMP
1912 #define ELF_EXEC_PAGESIZE 4096
1913
1914 #define VDSO_HEADER "vdso.c.inc"
1915
1916 #endif /* TARGET_S390X */
1917
1918 #ifdef TARGET_RISCV
1919
1920 #define ELF_ARCH EM_RISCV
1921
1922 #ifdef TARGET_RISCV32
1923 #define ELF_CLASS ELFCLASS32
1924 #define VDSO_HEADER "vdso-32.c.inc"
1925 #else
1926 #define ELF_CLASS ELFCLASS64
1927 #define VDSO_HEADER "vdso-64.c.inc"
1928 #endif
1929
1930 #define ELF_HWCAP get_elf_hwcap()
1931
get_elf_hwcap(void)1932 static uint32_t get_elf_hwcap(void)
1933 {
1934 #define MISA_BIT(EXT) (1 << (EXT - 'A'))
1935 RISCVCPU *cpu = RISCV_CPU(thread_cpu);
1936 uint32_t mask = MISA_BIT('I') | MISA_BIT('M') | MISA_BIT('A')
1937 | MISA_BIT('F') | MISA_BIT('D') | MISA_BIT('C')
1938 | MISA_BIT('V');
1939
1940 return cpu->env.misa_ext & mask;
1941 #undef MISA_BIT
1942 }
1943
init_thread(struct target_pt_regs * regs,struct image_info * infop)1944 static inline void init_thread(struct target_pt_regs *regs,
1945 struct image_info *infop)
1946 {
1947 regs->sepc = infop->entry;
1948 regs->sp = infop->start_stack;
1949 }
1950
1951 #define ELF_EXEC_PAGESIZE 4096
1952
1953 #endif /* TARGET_RISCV */
1954
1955 #ifdef TARGET_HPPA
1956
1957 #define ELF_CLASS ELFCLASS32
1958 #define ELF_ARCH EM_PARISC
1959 #define ELF_PLATFORM "PARISC"
1960 #define STACK_GROWS_DOWN 0
1961 #define STACK_ALIGNMENT 64
1962
1963 #define VDSO_HEADER "vdso.c.inc"
1964
init_thread(struct target_pt_regs * regs,struct image_info * infop)1965 static inline void init_thread(struct target_pt_regs *regs,
1966 struct image_info *infop)
1967 {
1968 regs->iaoq[0] = infop->entry | PRIV_USER;
1969 regs->iaoq[1] = regs->iaoq[0] + 4;
1970 regs->gr[23] = 0;
1971 regs->gr[24] = infop->argv;
1972 regs->gr[25] = infop->argc;
1973 /* The top-of-stack contains a linkage buffer. */
1974 regs->gr[30] = infop->start_stack + 64;
1975 regs->gr[31] = infop->entry;
1976 }
1977
1978 #define LO_COMMPAGE 0
1979
init_guest_commpage(void)1980 static bool init_guest_commpage(void)
1981 {
1982 /* If reserved_va, then we have already mapped 0 page on the host. */
1983 if (!reserved_va) {
1984 void *want, *addr;
1985
1986 want = g2h_untagged(LO_COMMPAGE);
1987 addr = mmap(want, TARGET_PAGE_SIZE, PROT_NONE,
1988 MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED_NOREPLACE, -1, 0);
1989 if (addr == MAP_FAILED) {
1990 perror("Allocating guest commpage");
1991 exit(EXIT_FAILURE);
1992 }
1993 if (addr != want) {
1994 return false;
1995 }
1996 }
1997
1998 /*
1999 * On Linux, page zero is normally marked execute only + gateway.
2000 * Normal read or write is supposed to fail (thus PROT_NONE above),
2001 * but specific offsets have kernel code mapped to raise permissions
2002 * and implement syscalls. Here, simply mark the page executable.
2003 * Special case the entry points during translation (see do_page_zero).
2004 */
2005 page_set_flags(LO_COMMPAGE, LO_COMMPAGE | ~TARGET_PAGE_MASK,
2006 PAGE_EXEC | PAGE_VALID);
2007 return true;
2008 }
2009
2010 #endif /* TARGET_HPPA */
2011
2012 #ifdef TARGET_XTENSA
2013
2014 #define ELF_CLASS ELFCLASS32
2015 #define ELF_ARCH EM_XTENSA
2016
init_thread(struct target_pt_regs * regs,struct image_info * infop)2017 static inline void init_thread(struct target_pt_regs *regs,
2018 struct image_info *infop)
2019 {
2020 regs->windowbase = 0;
2021 regs->windowstart = 1;
2022 regs->areg[1] = infop->start_stack;
2023 regs->pc = infop->entry;
2024 if (info_is_fdpic(infop)) {
2025 regs->areg[4] = infop->loadmap_addr;
2026 regs->areg[5] = infop->interpreter_loadmap_addr;
2027 if (infop->interpreter_loadmap_addr) {
2028 regs->areg[6] = infop->interpreter_pt_dynamic_addr;
2029 } else {
2030 regs->areg[6] = infop->pt_dynamic_addr;
2031 }
2032 }
2033 }
2034
2035 /* See linux kernel: arch/xtensa/include/asm/elf.h. */
2036 #define ELF_NREG 128
2037 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
2038
2039 enum {
2040 TARGET_REG_PC,
2041 TARGET_REG_PS,
2042 TARGET_REG_LBEG,
2043 TARGET_REG_LEND,
2044 TARGET_REG_LCOUNT,
2045 TARGET_REG_SAR,
2046 TARGET_REG_WINDOWSTART,
2047 TARGET_REG_WINDOWBASE,
2048 TARGET_REG_THREADPTR,
2049 TARGET_REG_AR0 = 64,
2050 };
2051
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUXtensaState * env)2052 static void elf_core_copy_regs(target_elf_gregset_t *regs,
2053 const CPUXtensaState *env)
2054 {
2055 unsigned i;
2056
2057 (*regs)[TARGET_REG_PC] = tswapreg(env->pc);
2058 (*regs)[TARGET_REG_PS] = tswapreg(env->sregs[PS] & ~PS_EXCM);
2059 (*regs)[TARGET_REG_LBEG] = tswapreg(env->sregs[LBEG]);
2060 (*regs)[TARGET_REG_LEND] = tswapreg(env->sregs[LEND]);
2061 (*regs)[TARGET_REG_LCOUNT] = tswapreg(env->sregs[LCOUNT]);
2062 (*regs)[TARGET_REG_SAR] = tswapreg(env->sregs[SAR]);
2063 (*regs)[TARGET_REG_WINDOWSTART] = tswapreg(env->sregs[WINDOW_START]);
2064 (*regs)[TARGET_REG_WINDOWBASE] = tswapreg(env->sregs[WINDOW_BASE]);
2065 (*regs)[TARGET_REG_THREADPTR] = tswapreg(env->uregs[THREADPTR]);
2066 xtensa_sync_phys_from_window((CPUXtensaState *)env);
2067 for (i = 0; i < env->config->nareg; ++i) {
2068 (*regs)[TARGET_REG_AR0 + i] = tswapreg(env->phys_regs[i]);
2069 }
2070 }
2071
2072 #define USE_ELF_CORE_DUMP
2073 #define ELF_EXEC_PAGESIZE 4096
2074
2075 #endif /* TARGET_XTENSA */
2076
2077 #ifdef TARGET_HEXAGON
2078
2079 #define ELF_CLASS ELFCLASS32
2080 #define ELF_ARCH EM_HEXAGON
2081
init_thread(struct target_pt_regs * regs,struct image_info * infop)2082 static inline void init_thread(struct target_pt_regs *regs,
2083 struct image_info *infop)
2084 {
2085 regs->sepc = infop->entry;
2086 regs->sp = infop->start_stack;
2087 }
2088
2089 #endif /* TARGET_HEXAGON */
2090
2091 #ifndef ELF_BASE_PLATFORM
2092 #define ELF_BASE_PLATFORM (NULL)
2093 #endif
2094
2095 #ifndef ELF_PLATFORM
2096 #define ELF_PLATFORM (NULL)
2097 #endif
2098
2099 #ifndef ELF_MACHINE
2100 #define ELF_MACHINE ELF_ARCH
2101 #endif
2102
2103 #ifndef elf_check_arch
2104 #define elf_check_arch(x) ((x) == ELF_ARCH)
2105 #endif
2106
2107 #ifndef elf_check_abi
2108 #define elf_check_abi(x) (1)
2109 #endif
2110
2111 #ifndef ELF_HWCAP
2112 #define ELF_HWCAP 0
2113 #endif
2114
2115 #ifndef STACK_GROWS_DOWN
2116 #define STACK_GROWS_DOWN 1
2117 #endif
2118
2119 #ifndef STACK_ALIGNMENT
2120 #define STACK_ALIGNMENT 16
2121 #endif
2122
2123 #ifdef TARGET_ABI32
2124 #undef ELF_CLASS
2125 #define ELF_CLASS ELFCLASS32
2126 #undef bswaptls
2127 #define bswaptls(ptr) bswap32s(ptr)
2128 #endif
2129
2130 #ifndef EXSTACK_DEFAULT
2131 #define EXSTACK_DEFAULT false
2132 #endif
2133
2134 #include "elf.h"
2135
2136 /* We must delay the following stanzas until after "elf.h". */
2137 #if defined(TARGET_AARCH64)
2138
arch_parse_elf_property(uint32_t pr_type,uint32_t pr_datasz,const uint32_t * data,struct image_info * info,Error ** errp)2139 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz,
2140 const uint32_t *data,
2141 struct image_info *info,
2142 Error **errp)
2143 {
2144 if (pr_type == GNU_PROPERTY_AARCH64_FEATURE_1_AND) {
2145 if (pr_datasz != sizeof(uint32_t)) {
2146 error_setg(errp, "Ill-formed GNU_PROPERTY_AARCH64_FEATURE_1_AND");
2147 return false;
2148 }
2149 /* We will extract GNU_PROPERTY_AARCH64_FEATURE_1_BTI later. */
2150 info->note_flags = *data;
2151 }
2152 return true;
2153 }
2154 #define ARCH_USE_GNU_PROPERTY 1
2155
2156 #else
2157
arch_parse_elf_property(uint32_t pr_type,uint32_t pr_datasz,const uint32_t * data,struct image_info * info,Error ** errp)2158 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz,
2159 const uint32_t *data,
2160 struct image_info *info,
2161 Error **errp)
2162 {
2163 g_assert_not_reached();
2164 }
2165 #define ARCH_USE_GNU_PROPERTY 0
2166
2167 #endif
2168
2169 struct exec
2170 {
2171 unsigned int a_info; /* Use macros N_MAGIC, etc for access */
2172 unsigned int a_text; /* length of text, in bytes */
2173 unsigned int a_data; /* length of data, in bytes */
2174 unsigned int a_bss; /* length of uninitialized data area, in bytes */
2175 unsigned int a_syms; /* length of symbol table data in file, in bytes */
2176 unsigned int a_entry; /* start address */
2177 unsigned int a_trsize; /* length of relocation info for text, in bytes */
2178 unsigned int a_drsize; /* length of relocation info for data, in bytes */
2179 };
2180
2181
2182 #define N_MAGIC(exec) ((exec).a_info & 0xffff)
2183 #define OMAGIC 0407
2184 #define NMAGIC 0410
2185 #define ZMAGIC 0413
2186 #define QMAGIC 0314
2187
2188 #define DLINFO_ITEMS 16
2189
memcpy_fromfs(void * to,const void * from,unsigned long n)2190 static inline void memcpy_fromfs(void * to, const void * from, unsigned long n)
2191 {
2192 memcpy(to, from, n);
2193 }
2194
bswap_ehdr(struct elfhdr * ehdr)2195 static void bswap_ehdr(struct elfhdr *ehdr)
2196 {
2197 if (!target_needs_bswap()) {
2198 return;
2199 }
2200
2201 bswap16s(&ehdr->e_type); /* Object file type */
2202 bswap16s(&ehdr->e_machine); /* Architecture */
2203 bswap32s(&ehdr->e_version); /* Object file version */
2204 bswaptls(&ehdr->e_entry); /* Entry point virtual address */
2205 bswaptls(&ehdr->e_phoff); /* Program header table file offset */
2206 bswaptls(&ehdr->e_shoff); /* Section header table file offset */
2207 bswap32s(&ehdr->e_flags); /* Processor-specific flags */
2208 bswap16s(&ehdr->e_ehsize); /* ELF header size in bytes */
2209 bswap16s(&ehdr->e_phentsize); /* Program header table entry size */
2210 bswap16s(&ehdr->e_phnum); /* Program header table entry count */
2211 bswap16s(&ehdr->e_shentsize); /* Section header table entry size */
2212 bswap16s(&ehdr->e_shnum); /* Section header table entry count */
2213 bswap16s(&ehdr->e_shstrndx); /* Section header string table index */
2214 }
2215
bswap_phdr(struct elf_phdr * phdr,int phnum)2216 static void bswap_phdr(struct elf_phdr *phdr, int phnum)
2217 {
2218 if (!target_needs_bswap()) {
2219 return;
2220 }
2221
2222 for (int i = 0; i < phnum; ++i, ++phdr) {
2223 bswap32s(&phdr->p_type); /* Segment type */
2224 bswap32s(&phdr->p_flags); /* Segment flags */
2225 bswaptls(&phdr->p_offset); /* Segment file offset */
2226 bswaptls(&phdr->p_vaddr); /* Segment virtual address */
2227 bswaptls(&phdr->p_paddr); /* Segment physical address */
2228 bswaptls(&phdr->p_filesz); /* Segment size in file */
2229 bswaptls(&phdr->p_memsz); /* Segment size in memory */
2230 bswaptls(&phdr->p_align); /* Segment alignment */
2231 }
2232 }
2233
bswap_shdr(struct elf_shdr * shdr,int shnum)2234 static void bswap_shdr(struct elf_shdr *shdr, int shnum)
2235 {
2236 if (!target_needs_bswap()) {
2237 return;
2238 }
2239
2240 for (int i = 0; i < shnum; ++i, ++shdr) {
2241 bswap32s(&shdr->sh_name);
2242 bswap32s(&shdr->sh_type);
2243 bswaptls(&shdr->sh_flags);
2244 bswaptls(&shdr->sh_addr);
2245 bswaptls(&shdr->sh_offset);
2246 bswaptls(&shdr->sh_size);
2247 bswap32s(&shdr->sh_link);
2248 bswap32s(&shdr->sh_info);
2249 bswaptls(&shdr->sh_addralign);
2250 bswaptls(&shdr->sh_entsize);
2251 }
2252 }
2253
bswap_sym(struct elf_sym * sym)2254 static void bswap_sym(struct elf_sym *sym)
2255 {
2256 if (!target_needs_bswap()) {
2257 return;
2258 }
2259
2260 bswap32s(&sym->st_name);
2261 bswaptls(&sym->st_value);
2262 bswaptls(&sym->st_size);
2263 bswap16s(&sym->st_shndx);
2264 }
2265
2266 #ifdef TARGET_MIPS
bswap_mips_abiflags(Mips_elf_abiflags_v0 * abiflags)2267 static void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags)
2268 {
2269 if (!target_needs_bswap()) {
2270 return;
2271 }
2272
2273 bswap16s(&abiflags->version);
2274 bswap32s(&abiflags->ases);
2275 bswap32s(&abiflags->isa_ext);
2276 bswap32s(&abiflags->flags1);
2277 bswap32s(&abiflags->flags2);
2278 }
2279 #endif
2280
2281 #ifdef USE_ELF_CORE_DUMP
2282 static int elf_core_dump(int, const CPUArchState *);
2283 #endif /* USE_ELF_CORE_DUMP */
2284 static void load_symbols(struct elfhdr *hdr, const ImageSource *src,
2285 abi_ulong load_bias);
2286
2287 /* Verify the portions of EHDR within E_IDENT for the target.
2288 This can be performed before bswapping the entire header. */
elf_check_ident(struct elfhdr * ehdr)2289 static bool elf_check_ident(struct elfhdr *ehdr)
2290 {
2291 return (ehdr->e_ident[EI_MAG0] == ELFMAG0
2292 && ehdr->e_ident[EI_MAG1] == ELFMAG1
2293 && ehdr->e_ident[EI_MAG2] == ELFMAG2
2294 && ehdr->e_ident[EI_MAG3] == ELFMAG3
2295 && ehdr->e_ident[EI_CLASS] == ELF_CLASS
2296 && ehdr->e_ident[EI_DATA] == ELF_DATA
2297 && ehdr->e_ident[EI_VERSION] == EV_CURRENT);
2298 }
2299
2300 /* Verify the portions of EHDR outside of E_IDENT for the target.
2301 This has to wait until after bswapping the header. */
elf_check_ehdr(struct elfhdr * ehdr)2302 static bool elf_check_ehdr(struct elfhdr *ehdr)
2303 {
2304 return (elf_check_arch(ehdr->e_machine)
2305 && elf_check_abi(ehdr->e_flags)
2306 && ehdr->e_ehsize == sizeof(struct elfhdr)
2307 && ehdr->e_phentsize == sizeof(struct elf_phdr)
2308 && (ehdr->e_type == ET_EXEC || ehdr->e_type == ET_DYN));
2309 }
2310
2311 /*
2312 * 'copy_elf_strings()' copies argument/envelope strings from user
2313 * memory to free pages in kernel mem. These are in a format ready
2314 * to be put directly into the top of new user memory.
2315 *
2316 */
copy_elf_strings(int argc,char ** argv,char * scratch,abi_ulong p,abi_ulong stack_limit)2317 static abi_ulong copy_elf_strings(int argc, char **argv, char *scratch,
2318 abi_ulong p, abi_ulong stack_limit)
2319 {
2320 char *tmp;
2321 int len, i;
2322 abi_ulong top = p;
2323
2324 if (!p) {
2325 return 0; /* bullet-proofing */
2326 }
2327
2328 if (STACK_GROWS_DOWN) {
2329 int offset = ((p - 1) % TARGET_PAGE_SIZE) + 1;
2330 for (i = argc - 1; i >= 0; --i) {
2331 tmp = argv[i];
2332 if (!tmp) {
2333 fprintf(stderr, "VFS: argc is wrong");
2334 exit(-1);
2335 }
2336 len = strlen(tmp) + 1;
2337 tmp += len;
2338
2339 if (len > (p - stack_limit)) {
2340 return 0;
2341 }
2342 while (len) {
2343 int bytes_to_copy = (len > offset) ? offset : len;
2344 tmp -= bytes_to_copy;
2345 p -= bytes_to_copy;
2346 offset -= bytes_to_copy;
2347 len -= bytes_to_copy;
2348
2349 memcpy_fromfs(scratch + offset, tmp, bytes_to_copy);
2350
2351 if (offset == 0) {
2352 memcpy_to_target(p, scratch, top - p);
2353 top = p;
2354 offset = TARGET_PAGE_SIZE;
2355 }
2356 }
2357 }
2358 if (p != top) {
2359 memcpy_to_target(p, scratch + offset, top - p);
2360 }
2361 } else {
2362 int remaining = TARGET_PAGE_SIZE - (p % TARGET_PAGE_SIZE);
2363 for (i = 0; i < argc; ++i) {
2364 tmp = argv[i];
2365 if (!tmp) {
2366 fprintf(stderr, "VFS: argc is wrong");
2367 exit(-1);
2368 }
2369 len = strlen(tmp) + 1;
2370 if (len > (stack_limit - p)) {
2371 return 0;
2372 }
2373 while (len) {
2374 int bytes_to_copy = (len > remaining) ? remaining : len;
2375
2376 memcpy_fromfs(scratch + (p - top), tmp, bytes_to_copy);
2377
2378 tmp += bytes_to_copy;
2379 remaining -= bytes_to_copy;
2380 p += bytes_to_copy;
2381 len -= bytes_to_copy;
2382
2383 if (remaining == 0) {
2384 memcpy_to_target(top, scratch, p - top);
2385 top = p;
2386 remaining = TARGET_PAGE_SIZE;
2387 }
2388 }
2389 }
2390 if (p != top) {
2391 memcpy_to_target(top, scratch, p - top);
2392 }
2393 }
2394
2395 return p;
2396 }
2397
2398 /* Older linux kernels provide up to MAX_ARG_PAGES (default: 32) of
2399 * argument/environment space. Newer kernels (>2.6.33) allow more,
2400 * dependent on stack size, but guarantee at least 32 pages for
2401 * backwards compatibility.
2402 */
2403 #define STACK_LOWER_LIMIT (32 * TARGET_PAGE_SIZE)
2404
setup_arg_pages(struct linux_binprm * bprm,struct image_info * info)2405 static abi_ulong setup_arg_pages(struct linux_binprm *bprm,
2406 struct image_info *info)
2407 {
2408 abi_ulong size, error, guard;
2409 int prot;
2410
2411 size = guest_stack_size;
2412 if (size < STACK_LOWER_LIMIT) {
2413 size = STACK_LOWER_LIMIT;
2414 }
2415
2416 if (STACK_GROWS_DOWN) {
2417 guard = TARGET_PAGE_SIZE;
2418 if (guard < qemu_real_host_page_size()) {
2419 guard = qemu_real_host_page_size();
2420 }
2421 } else {
2422 /* no guard page for hppa target where stack grows upwards. */
2423 guard = 0;
2424 }
2425
2426 prot = PROT_READ | PROT_WRITE;
2427 if (info->exec_stack) {
2428 prot |= PROT_EXEC;
2429 }
2430 error = target_mmap(0, size + guard, prot,
2431 MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
2432 if (error == -1) {
2433 perror("mmap stack");
2434 exit(-1);
2435 }
2436
2437 /* We reserve one extra page at the top of the stack as guard. */
2438 if (STACK_GROWS_DOWN) {
2439 target_mprotect(error, guard, PROT_NONE);
2440 info->stack_limit = error + guard;
2441 return info->stack_limit + size - sizeof(void *);
2442 } else {
2443 info->stack_limit = error + size;
2444 return error;
2445 }
2446 }
2447
2448 /**
2449 * zero_bss:
2450 *
2451 * Map and zero the bss. We need to explicitly zero any fractional pages
2452 * after the data section (i.e. bss). Return false on mapping failure.
2453 */
zero_bss(abi_ulong start_bss,abi_ulong end_bss,int prot,Error ** errp)2454 static bool zero_bss(abi_ulong start_bss, abi_ulong end_bss,
2455 int prot, Error **errp)
2456 {
2457 abi_ulong align_bss;
2458
2459 /* We only expect writable bss; the code segment shouldn't need this. */
2460 if (!(prot & PROT_WRITE)) {
2461 error_setg(errp, "PT_LOAD with non-writable bss");
2462 return false;
2463 }
2464
2465 align_bss = TARGET_PAGE_ALIGN(start_bss);
2466 end_bss = TARGET_PAGE_ALIGN(end_bss);
2467
2468 if (start_bss < align_bss) {
2469 int flags = page_get_flags(start_bss);
2470
2471 if (!(flags & PAGE_RWX)) {
2472 /*
2473 * The whole address space of the executable was reserved
2474 * at the start, therefore all pages will be VALID.
2475 * But assuming there are no PROT_NONE PT_LOAD segments,
2476 * a PROT_NONE page means no data all bss, and we can
2477 * simply extend the new anon mapping back to the start
2478 * of the page of bss.
2479 */
2480 align_bss -= TARGET_PAGE_SIZE;
2481 } else {
2482 /*
2483 * The start of the bss shares a page with something.
2484 * The only thing that we expect is the data section,
2485 * which would already be marked writable.
2486 * Overlapping the RX code segment seems malformed.
2487 */
2488 if (!(flags & PAGE_WRITE)) {
2489 error_setg(errp, "PT_LOAD with bss overlapping "
2490 "non-writable page");
2491 return false;
2492 }
2493
2494 /* The page is already mapped and writable. */
2495 memset(g2h_untagged(start_bss), 0, align_bss - start_bss);
2496 }
2497 }
2498
2499 if (align_bss < end_bss &&
2500 target_mmap(align_bss, end_bss - align_bss, prot,
2501 MAP_FIXED | MAP_PRIVATE | MAP_ANON, -1, 0) == -1) {
2502 error_setg_errno(errp, errno, "Error mapping bss");
2503 return false;
2504 }
2505 return true;
2506 }
2507
2508 #if defined(TARGET_ARM)
elf_is_fdpic(struct elfhdr * exec)2509 static int elf_is_fdpic(struct elfhdr *exec)
2510 {
2511 return exec->e_ident[EI_OSABI] == ELFOSABI_ARM_FDPIC;
2512 }
2513 #elif defined(TARGET_XTENSA)
elf_is_fdpic(struct elfhdr * exec)2514 static int elf_is_fdpic(struct elfhdr *exec)
2515 {
2516 return exec->e_ident[EI_OSABI] == ELFOSABI_XTENSA_FDPIC;
2517 }
2518 #else
2519 /* Default implementation, always false. */
elf_is_fdpic(struct elfhdr * exec)2520 static int elf_is_fdpic(struct elfhdr *exec)
2521 {
2522 return 0;
2523 }
2524 #endif
2525
loader_build_fdpic_loadmap(struct image_info * info,abi_ulong sp)2526 static abi_ulong loader_build_fdpic_loadmap(struct image_info *info, abi_ulong sp)
2527 {
2528 uint16_t n;
2529 struct elf32_fdpic_loadseg *loadsegs = info->loadsegs;
2530
2531 /* elf32_fdpic_loadseg */
2532 n = info->nsegs;
2533 while (n--) {
2534 sp -= 12;
2535 put_user_u32(loadsegs[n].addr, sp+0);
2536 put_user_u32(loadsegs[n].p_vaddr, sp+4);
2537 put_user_u32(loadsegs[n].p_memsz, sp+8);
2538 }
2539
2540 /* elf32_fdpic_loadmap */
2541 sp -= 4;
2542 put_user_u16(0, sp+0); /* version */
2543 put_user_u16(info->nsegs, sp+2); /* nsegs */
2544
2545 info->personality = PER_LINUX_FDPIC;
2546 info->loadmap_addr = sp;
2547
2548 return sp;
2549 }
2550
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)2551 static abi_ulong create_elf_tables(abi_ulong p, int argc, int envc,
2552 struct elfhdr *exec,
2553 struct image_info *info,
2554 struct image_info *interp_info,
2555 struct image_info *vdso_info)
2556 {
2557 abi_ulong sp;
2558 abi_ulong u_argc, u_argv, u_envp, u_auxv;
2559 int size;
2560 int i;
2561 abi_ulong u_rand_bytes;
2562 uint8_t k_rand_bytes[16];
2563 abi_ulong u_platform, u_base_platform;
2564 const char *k_platform, *k_base_platform;
2565 const int n = sizeof(elf_addr_t);
2566
2567 sp = p;
2568
2569 /* Needs to be before we load the env/argc/... */
2570 if (elf_is_fdpic(exec)) {
2571 /* Need 4 byte alignment for these structs */
2572 sp &= ~3;
2573 sp = loader_build_fdpic_loadmap(info, sp);
2574 info->other_info = interp_info;
2575 if (interp_info) {
2576 interp_info->other_info = info;
2577 sp = loader_build_fdpic_loadmap(interp_info, sp);
2578 info->interpreter_loadmap_addr = interp_info->loadmap_addr;
2579 info->interpreter_pt_dynamic_addr = interp_info->pt_dynamic_addr;
2580 } else {
2581 info->interpreter_loadmap_addr = 0;
2582 info->interpreter_pt_dynamic_addr = 0;
2583 }
2584 }
2585
2586 u_base_platform = 0;
2587 k_base_platform = ELF_BASE_PLATFORM;
2588 if (k_base_platform) {
2589 size_t len = strlen(k_base_platform) + 1;
2590 if (STACK_GROWS_DOWN) {
2591 sp -= (len + n - 1) & ~(n - 1);
2592 u_base_platform = sp;
2593 /* FIXME - check return value of memcpy_to_target() for failure */
2594 memcpy_to_target(sp, k_base_platform, len);
2595 } else {
2596 memcpy_to_target(sp, k_base_platform, len);
2597 u_base_platform = sp;
2598 sp += len + 1;
2599 }
2600 }
2601
2602 u_platform = 0;
2603 k_platform = ELF_PLATFORM;
2604 if (k_platform) {
2605 size_t len = strlen(k_platform) + 1;
2606 if (STACK_GROWS_DOWN) {
2607 sp -= (len + n - 1) & ~(n - 1);
2608 u_platform = sp;
2609 /* FIXME - check return value of memcpy_to_target() for failure */
2610 memcpy_to_target(sp, k_platform, len);
2611 } else {
2612 memcpy_to_target(sp, k_platform, len);
2613 u_platform = sp;
2614 sp += len + 1;
2615 }
2616 }
2617
2618 /* Provide 16 byte alignment for the PRNG, and basic alignment for
2619 * the argv and envp pointers.
2620 */
2621 if (STACK_GROWS_DOWN) {
2622 sp = QEMU_ALIGN_DOWN(sp, 16);
2623 } else {
2624 sp = QEMU_ALIGN_UP(sp, 16);
2625 }
2626
2627 /*
2628 * Generate 16 random bytes for userspace PRNG seeding.
2629 */
2630 qemu_guest_getrandom_nofail(k_rand_bytes, sizeof(k_rand_bytes));
2631 if (STACK_GROWS_DOWN) {
2632 sp -= 16;
2633 u_rand_bytes = sp;
2634 /* FIXME - check return value of memcpy_to_target() for failure */
2635 memcpy_to_target(sp, k_rand_bytes, 16);
2636 } else {
2637 memcpy_to_target(sp, k_rand_bytes, 16);
2638 u_rand_bytes = sp;
2639 sp += 16;
2640 }
2641
2642 size = (DLINFO_ITEMS + 1) * 2;
2643 if (k_base_platform) {
2644 size += 2;
2645 }
2646 if (k_platform) {
2647 size += 2;
2648 }
2649 if (vdso_info) {
2650 size += 2;
2651 }
2652 #ifdef DLINFO_ARCH_ITEMS
2653 size += DLINFO_ARCH_ITEMS * 2;
2654 #endif
2655 #ifdef ELF_HWCAP2
2656 size += 2;
2657 #endif
2658 info->auxv_len = size * n;
2659
2660 size += envc + argc + 2;
2661 size += 1; /* argc itself */
2662 size *= n;
2663
2664 /* Allocate space and finalize stack alignment for entry now. */
2665 if (STACK_GROWS_DOWN) {
2666 u_argc = QEMU_ALIGN_DOWN(sp - size, STACK_ALIGNMENT);
2667 sp = u_argc;
2668 } else {
2669 u_argc = sp;
2670 sp = QEMU_ALIGN_UP(sp + size, STACK_ALIGNMENT);
2671 }
2672
2673 u_argv = u_argc + n;
2674 u_envp = u_argv + (argc + 1) * n;
2675 u_auxv = u_envp + (envc + 1) * n;
2676 info->saved_auxv = u_auxv;
2677 info->argc = argc;
2678 info->envc = envc;
2679 info->argv = u_argv;
2680 info->envp = u_envp;
2681
2682 /* This is correct because Linux defines
2683 * elf_addr_t as Elf32_Off / Elf64_Off
2684 */
2685 #define NEW_AUX_ENT(id, val) do { \
2686 put_user_ual(id, u_auxv); u_auxv += n; \
2687 put_user_ual(val, u_auxv); u_auxv += n; \
2688 } while(0)
2689
2690 #ifdef ARCH_DLINFO
2691 /*
2692 * ARCH_DLINFO must come first so platform specific code can enforce
2693 * special alignment requirements on the AUXV if necessary (eg. PPC).
2694 */
2695 ARCH_DLINFO;
2696 #endif
2697 /* There must be exactly DLINFO_ITEMS entries here, or the assert
2698 * on info->auxv_len will trigger.
2699 */
2700 NEW_AUX_ENT(AT_PHDR, (abi_ulong)(info->load_addr + exec->e_phoff));
2701 NEW_AUX_ENT(AT_PHENT, (abi_ulong)(sizeof (struct elf_phdr)));
2702 NEW_AUX_ENT(AT_PHNUM, (abi_ulong)(exec->e_phnum));
2703 NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(TARGET_PAGE_SIZE));
2704 NEW_AUX_ENT(AT_BASE, (abi_ulong)(interp_info ? interp_info->load_addr : 0));
2705 NEW_AUX_ENT(AT_FLAGS, (abi_ulong)0);
2706 NEW_AUX_ENT(AT_ENTRY, info->entry);
2707 NEW_AUX_ENT(AT_UID, (abi_ulong) getuid());
2708 NEW_AUX_ENT(AT_EUID, (abi_ulong) geteuid());
2709 NEW_AUX_ENT(AT_GID, (abi_ulong) getgid());
2710 NEW_AUX_ENT(AT_EGID, (abi_ulong) getegid());
2711 NEW_AUX_ENT(AT_HWCAP, (abi_ulong) ELF_HWCAP);
2712 NEW_AUX_ENT(AT_CLKTCK, (abi_ulong) sysconf(_SC_CLK_TCK));
2713 NEW_AUX_ENT(AT_RANDOM, (abi_ulong) u_rand_bytes);
2714 NEW_AUX_ENT(AT_SECURE, (abi_ulong) qemu_getauxval(AT_SECURE));
2715 NEW_AUX_ENT(AT_EXECFN, info->file_string);
2716
2717 #ifdef ELF_HWCAP2
2718 NEW_AUX_ENT(AT_HWCAP2, (abi_ulong) ELF_HWCAP2);
2719 #endif
2720
2721 if (u_base_platform) {
2722 NEW_AUX_ENT(AT_BASE_PLATFORM, u_base_platform);
2723 }
2724 if (u_platform) {
2725 NEW_AUX_ENT(AT_PLATFORM, u_platform);
2726 }
2727 if (vdso_info) {
2728 NEW_AUX_ENT(AT_SYSINFO_EHDR, vdso_info->load_addr);
2729 }
2730 NEW_AUX_ENT (AT_NULL, 0);
2731 #undef NEW_AUX_ENT
2732
2733 /* Check that our initial calculation of the auxv length matches how much
2734 * we actually put into it.
2735 */
2736 assert(info->auxv_len == u_auxv - info->saved_auxv);
2737
2738 put_user_ual(argc, u_argc);
2739
2740 p = info->arg_strings;
2741 for (i = 0; i < argc; ++i) {
2742 put_user_ual(p, u_argv);
2743 u_argv += n;
2744 p += target_strlen(p) + 1;
2745 }
2746 put_user_ual(0, u_argv);
2747
2748 p = info->env_strings;
2749 for (i = 0; i < envc; ++i) {
2750 put_user_ual(p, u_envp);
2751 u_envp += n;
2752 p += target_strlen(p) + 1;
2753 }
2754 put_user_ual(0, u_envp);
2755
2756 return sp;
2757 }
2758
2759 #if defined(HI_COMMPAGE)
2760 #define LO_COMMPAGE -1
2761 #elif defined(LO_COMMPAGE)
2762 #define HI_COMMPAGE 0
2763 #else
2764 #define HI_COMMPAGE 0
2765 #define LO_COMMPAGE -1
2766 #ifndef INIT_GUEST_COMMPAGE
2767 #define init_guest_commpage() true
2768 #endif
2769 #endif
2770
2771 /**
2772 * pgb_try_mmap:
2773 * @addr: host start address
2774 * @addr_last: host last address
2775 * @keep: do not unmap the probe region
2776 *
2777 * Return 1 if [@addr, @addr_last] is not mapped in the host,
2778 * return 0 if it is not available to map, and -1 on mmap error.
2779 * If @keep, the region is left mapped on success, otherwise unmapped.
2780 */
pgb_try_mmap(uintptr_t addr,uintptr_t addr_last,bool keep)2781 static int pgb_try_mmap(uintptr_t addr, uintptr_t addr_last, bool keep)
2782 {
2783 size_t size = addr_last - addr + 1;
2784 void *p = mmap((void *)addr, size, PROT_NONE,
2785 MAP_ANONYMOUS | MAP_PRIVATE |
2786 MAP_NORESERVE | MAP_FIXED_NOREPLACE, -1, 0);
2787 int ret;
2788
2789 if (p == MAP_FAILED) {
2790 return errno == EEXIST ? 0 : -1;
2791 }
2792 ret = p == (void *)addr;
2793 if (!keep || !ret) {
2794 munmap(p, size);
2795 }
2796 return ret;
2797 }
2798
2799 /**
2800 * pgb_try_mmap_skip_brk(uintptr_t addr, uintptr_t size, uintptr_t brk)
2801 * @addr: host address
2802 * @addr_last: host last address
2803 * @brk: host brk
2804 *
2805 * Like pgb_try_mmap, but additionally reserve some memory following brk.
2806 */
pgb_try_mmap_skip_brk(uintptr_t addr,uintptr_t addr_last,uintptr_t brk,bool keep)2807 static int pgb_try_mmap_skip_brk(uintptr_t addr, uintptr_t addr_last,
2808 uintptr_t brk, bool keep)
2809 {
2810 uintptr_t brk_last = brk + 16 * MiB - 1;
2811
2812 /* Do not map anything close to the host brk. */
2813 if (addr <= brk_last && brk <= addr_last) {
2814 return 0;
2815 }
2816 return pgb_try_mmap(addr, addr_last, keep);
2817 }
2818
2819 /**
2820 * pgb_try_mmap_set:
2821 * @ga: set of guest addrs
2822 * @base: guest_base
2823 * @brk: host brk
2824 *
2825 * Return true if all @ga can be mapped by the host at @base.
2826 * On success, retain the mapping at index 0 for reserved_va.
2827 */
2828
2829 typedef struct PGBAddrs {
2830 uintptr_t bounds[3][2]; /* start/last pairs */
2831 int nbounds;
2832 } PGBAddrs;
2833
pgb_try_mmap_set(const PGBAddrs * ga,uintptr_t base,uintptr_t brk)2834 static bool pgb_try_mmap_set(const PGBAddrs *ga, uintptr_t base, uintptr_t brk)
2835 {
2836 for (int i = ga->nbounds - 1; i >= 0; --i) {
2837 if (pgb_try_mmap_skip_brk(ga->bounds[i][0] + base,
2838 ga->bounds[i][1] + base,
2839 brk, i == 0 && reserved_va) <= 0) {
2840 return false;
2841 }
2842 }
2843 return true;
2844 }
2845
2846 /**
2847 * pgb_addr_set:
2848 * @ga: output set of guest addrs
2849 * @guest_loaddr: guest image low address
2850 * @guest_loaddr: guest image high address
2851 * @identity: create for identity mapping
2852 *
2853 * Fill in @ga with the image, COMMPAGE and NULL page.
2854 */
pgb_addr_set(PGBAddrs * ga,abi_ulong guest_loaddr,abi_ulong guest_hiaddr,bool try_identity)2855 static bool pgb_addr_set(PGBAddrs *ga, abi_ulong guest_loaddr,
2856 abi_ulong guest_hiaddr, bool try_identity)
2857 {
2858 int n;
2859
2860 /*
2861 * With a low commpage, or a guest mapped very low,
2862 * we may not be able to use the identity map.
2863 */
2864 if (try_identity) {
2865 if (LO_COMMPAGE != -1 && LO_COMMPAGE < mmap_min_addr) {
2866 return false;
2867 }
2868 if (guest_loaddr != 0 && guest_loaddr < mmap_min_addr) {
2869 return false;
2870 }
2871 }
2872
2873 memset(ga, 0, sizeof(*ga));
2874 n = 0;
2875
2876 if (reserved_va) {
2877 ga->bounds[n][0] = try_identity ? mmap_min_addr : 0;
2878 ga->bounds[n][1] = reserved_va;
2879 n++;
2880 /* LO_COMMPAGE and NULL handled by reserving from 0. */
2881 } else {
2882 /* Add any LO_COMMPAGE or NULL page. */
2883 if (LO_COMMPAGE != -1) {
2884 ga->bounds[n][0] = 0;
2885 ga->bounds[n][1] = LO_COMMPAGE + TARGET_PAGE_SIZE - 1;
2886 n++;
2887 } else if (!try_identity) {
2888 ga->bounds[n][0] = 0;
2889 ga->bounds[n][1] = TARGET_PAGE_SIZE - 1;
2890 n++;
2891 }
2892
2893 /* Add the guest image for ET_EXEC. */
2894 if (guest_loaddr) {
2895 ga->bounds[n][0] = guest_loaddr;
2896 ga->bounds[n][1] = guest_hiaddr;
2897 n++;
2898 }
2899 }
2900
2901 /*
2902 * Temporarily disable
2903 * "comparison is always false due to limited range of data type"
2904 * due to comparison between unsigned and (possible) 0.
2905 */
2906 #pragma GCC diagnostic push
2907 #pragma GCC diagnostic ignored "-Wtype-limits"
2908
2909 /* Add any HI_COMMPAGE not covered by reserved_va. */
2910 if (reserved_va < HI_COMMPAGE) {
2911 ga->bounds[n][0] = HI_COMMPAGE & qemu_real_host_page_mask();
2912 ga->bounds[n][1] = HI_COMMPAGE + TARGET_PAGE_SIZE - 1;
2913 n++;
2914 }
2915
2916 #pragma GCC diagnostic pop
2917
2918 ga->nbounds = n;
2919 return true;
2920 }
2921
pgb_fail_in_use(const char * image_name)2922 static void pgb_fail_in_use(const char *image_name)
2923 {
2924 error_report("%s: requires virtual address space that is in use "
2925 "(omit the -B option or choose a different value)",
2926 image_name);
2927 exit(EXIT_FAILURE);
2928 }
2929
pgb_fixed(const char * image_name,uintptr_t guest_loaddr,uintptr_t guest_hiaddr,uintptr_t align)2930 static void pgb_fixed(const char *image_name, uintptr_t guest_loaddr,
2931 uintptr_t guest_hiaddr, uintptr_t align)
2932 {
2933 PGBAddrs ga;
2934 uintptr_t brk = (uintptr_t)sbrk(0);
2935
2936 if (!QEMU_IS_ALIGNED(guest_base, align)) {
2937 fprintf(stderr, "Requested guest base %p does not satisfy "
2938 "host minimum alignment (0x%" PRIxPTR ")\n",
2939 (void *)guest_base, align);
2940 exit(EXIT_FAILURE);
2941 }
2942
2943 if (!pgb_addr_set(&ga, guest_loaddr, guest_hiaddr, !guest_base)
2944 || !pgb_try_mmap_set(&ga, guest_base, brk)) {
2945 pgb_fail_in_use(image_name);
2946 }
2947 }
2948
2949 /**
2950 * pgb_find_fallback:
2951 *
2952 * This is a fallback method for finding holes in the host address space
2953 * if we don't have the benefit of being able to access /proc/self/map.
2954 * It can potentially take a very long time as we can only dumbly iterate
2955 * up the host address space seeing if the allocation would work.
2956 */
pgb_find_fallback(const PGBAddrs * ga,uintptr_t align,uintptr_t brk)2957 static uintptr_t pgb_find_fallback(const PGBAddrs *ga, uintptr_t align,
2958 uintptr_t brk)
2959 {
2960 /* TODO: come up with a better estimate of how much to skip. */
2961 uintptr_t skip = sizeof(uintptr_t) == 4 ? MiB : GiB;
2962
2963 for (uintptr_t base = skip; ; base += skip) {
2964 base = ROUND_UP(base, align);
2965 if (pgb_try_mmap_set(ga, base, brk)) {
2966 return base;
2967 }
2968 if (base >= -skip) {
2969 return -1;
2970 }
2971 }
2972 }
2973
pgb_try_itree(const PGBAddrs * ga,uintptr_t base,IntervalTreeRoot * root)2974 static uintptr_t pgb_try_itree(const PGBAddrs *ga, uintptr_t base,
2975 IntervalTreeRoot *root)
2976 {
2977 for (int i = ga->nbounds - 1; i >= 0; --i) {
2978 uintptr_t s = base + ga->bounds[i][0];
2979 uintptr_t l = base + ga->bounds[i][1];
2980 IntervalTreeNode *n;
2981
2982 if (l < s) {
2983 /* Wraparound. Skip to advance S to mmap_min_addr. */
2984 return mmap_min_addr - s;
2985 }
2986
2987 n = interval_tree_iter_first(root, s, l);
2988 if (n != NULL) {
2989 /* Conflict. Skip to advance S to LAST + 1. */
2990 return n->last - s + 1;
2991 }
2992 }
2993 return 0; /* success */
2994 }
2995
pgb_find_itree(const PGBAddrs * ga,IntervalTreeRoot * root,uintptr_t align,uintptr_t brk)2996 static uintptr_t pgb_find_itree(const PGBAddrs *ga, IntervalTreeRoot *root,
2997 uintptr_t align, uintptr_t brk)
2998 {
2999 uintptr_t last = sizeof(uintptr_t) == 4 ? MiB : GiB;
3000 uintptr_t base, skip;
3001
3002 while (true) {
3003 base = ROUND_UP(last, align);
3004 if (base < last) {
3005 return -1;
3006 }
3007
3008 skip = pgb_try_itree(ga, base, root);
3009 if (skip == 0) {
3010 break;
3011 }
3012
3013 last = base + skip;
3014 if (last < base) {
3015 return -1;
3016 }
3017 }
3018
3019 /*
3020 * We've chosen 'base' based on holes in the interval tree,
3021 * but we don't yet know if it is a valid host address.
3022 * Because it is the first matching hole, if the host addresses
3023 * are invalid we know there are no further matches.
3024 */
3025 return pgb_try_mmap_set(ga, base, brk) ? base : -1;
3026 }
3027
pgb_dynamic(const char * image_name,uintptr_t guest_loaddr,uintptr_t guest_hiaddr,uintptr_t align)3028 static void pgb_dynamic(const char *image_name, uintptr_t guest_loaddr,
3029 uintptr_t guest_hiaddr, uintptr_t align)
3030 {
3031 IntervalTreeRoot *root;
3032 uintptr_t brk, ret;
3033 PGBAddrs ga;
3034
3035 /* Try the identity map first. */
3036 if (pgb_addr_set(&ga, guest_loaddr, guest_hiaddr, true)) {
3037 brk = (uintptr_t)sbrk(0);
3038 if (pgb_try_mmap_set(&ga, 0, brk)) {
3039 guest_base = 0;
3040 return;
3041 }
3042 }
3043
3044 /*
3045 * Rebuild the address set for non-identity map.
3046 * This differs in the mapping of the guest NULL page.
3047 */
3048 pgb_addr_set(&ga, guest_loaddr, guest_hiaddr, false);
3049
3050 root = read_self_maps();
3051
3052 /* Read brk after we've read the maps, which will malloc. */
3053 brk = (uintptr_t)sbrk(0);
3054
3055 if (!root) {
3056 ret = pgb_find_fallback(&ga, align, brk);
3057 } else {
3058 /*
3059 * Reserve the area close to the host brk.
3060 * This will be freed with the rest of the tree.
3061 */
3062 IntervalTreeNode *b = g_new0(IntervalTreeNode, 1);
3063 b->start = brk;
3064 b->last = brk + 16 * MiB - 1;
3065 interval_tree_insert(b, root);
3066
3067 ret = pgb_find_itree(&ga, root, align, brk);
3068 free_self_maps(root);
3069 }
3070
3071 if (ret == -1) {
3072 int w = TARGET_LONG_BITS / 4;
3073
3074 error_report("%s: Unable to find a guest_base to satisfy all "
3075 "guest address mapping requirements", image_name);
3076
3077 for (int i = 0; i < ga.nbounds; ++i) {
3078 error_printf(" %0*" PRIx64 "-%0*" PRIx64 "\n",
3079 w, (uint64_t)ga.bounds[i][0],
3080 w, (uint64_t)ga.bounds[i][1]);
3081 }
3082 exit(EXIT_FAILURE);
3083 }
3084 guest_base = ret;
3085 }
3086
probe_guest_base(const char * image_name,abi_ulong guest_loaddr,abi_ulong guest_hiaddr)3087 void probe_guest_base(const char *image_name, abi_ulong guest_loaddr,
3088 abi_ulong guest_hiaddr)
3089 {
3090 /* In order to use host shmat, we must be able to honor SHMLBA. */
3091 uintptr_t align = MAX(SHMLBA, TARGET_PAGE_SIZE);
3092
3093 /* Sanity check the guest binary. */
3094 if (reserved_va) {
3095 if (guest_hiaddr > reserved_va) {
3096 error_report("%s: requires more than reserved virtual "
3097 "address space (0x%" PRIx64 " > 0x%lx)",
3098 image_name, (uint64_t)guest_hiaddr, reserved_va);
3099 exit(EXIT_FAILURE);
3100 }
3101 } else {
3102 if (guest_hiaddr != (uintptr_t)guest_hiaddr) {
3103 error_report("%s: requires more virtual address space "
3104 "than the host can provide (0x%" PRIx64 ")",
3105 image_name, (uint64_t)guest_hiaddr + 1);
3106 exit(EXIT_FAILURE);
3107 }
3108 }
3109
3110 if (have_guest_base) {
3111 pgb_fixed(image_name, guest_loaddr, guest_hiaddr, align);
3112 } else {
3113 pgb_dynamic(image_name, guest_loaddr, guest_hiaddr, align);
3114 }
3115
3116 /* Reserve and initialize the commpage. */
3117 if (!init_guest_commpage()) {
3118 /* We have already probed for the commpage being free. */
3119 g_assert_not_reached();
3120 }
3121
3122 assert(QEMU_IS_ALIGNED(guest_base, align));
3123 qemu_log_mask(CPU_LOG_PAGE, "Locating guest address space "
3124 "@ 0x%" PRIx64 "\n", (uint64_t)guest_base);
3125 }
3126
3127 enum {
3128 /* The string "GNU\0" as a magic number. */
3129 GNU0_MAGIC = const_le32('G' | 'N' << 8 | 'U' << 16),
3130 NOTE_DATA_SZ = 1 * KiB,
3131 NOTE_NAME_SZ = 4,
3132 ELF_GNU_PROPERTY_ALIGN = ELF_CLASS == ELFCLASS32 ? 4 : 8,
3133 };
3134
3135 /*
3136 * Process a single gnu_property entry.
3137 * Return false for error.
3138 */
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)3139 static bool parse_elf_property(const uint32_t *data, int *off, int datasz,
3140 struct image_info *info, bool have_prev_type,
3141 uint32_t *prev_type, Error **errp)
3142 {
3143 uint32_t pr_type, pr_datasz, step;
3144
3145 if (*off > datasz || !QEMU_IS_ALIGNED(*off, ELF_GNU_PROPERTY_ALIGN)) {
3146 goto error_data;
3147 }
3148 datasz -= *off;
3149 data += *off / sizeof(uint32_t);
3150
3151 if (datasz < 2 * sizeof(uint32_t)) {
3152 goto error_data;
3153 }
3154 pr_type = data[0];
3155 pr_datasz = data[1];
3156 data += 2;
3157 datasz -= 2 * sizeof(uint32_t);
3158 step = ROUND_UP(pr_datasz, ELF_GNU_PROPERTY_ALIGN);
3159 if (step > datasz) {
3160 goto error_data;
3161 }
3162
3163 /* Properties are supposed to be unique and sorted on pr_type. */
3164 if (have_prev_type && pr_type <= *prev_type) {
3165 if (pr_type == *prev_type) {
3166 error_setg(errp, "Duplicate property in PT_GNU_PROPERTY");
3167 } else {
3168 error_setg(errp, "Unsorted property in PT_GNU_PROPERTY");
3169 }
3170 return false;
3171 }
3172 *prev_type = pr_type;
3173
3174 if (!arch_parse_elf_property(pr_type, pr_datasz, data, info, errp)) {
3175 return false;
3176 }
3177
3178 *off += 2 * sizeof(uint32_t) + step;
3179 return true;
3180
3181 error_data:
3182 error_setg(errp, "Ill-formed property in PT_GNU_PROPERTY");
3183 return false;
3184 }
3185
3186 /* Process NT_GNU_PROPERTY_TYPE_0. */
parse_elf_properties(const ImageSource * src,struct image_info * info,const struct elf_phdr * phdr,Error ** errp)3187 static bool parse_elf_properties(const ImageSource *src,
3188 struct image_info *info,
3189 const struct elf_phdr *phdr,
3190 Error **errp)
3191 {
3192 union {
3193 struct elf_note nhdr;
3194 uint32_t data[NOTE_DATA_SZ / sizeof(uint32_t)];
3195 } note;
3196
3197 int n, off, datasz;
3198 bool have_prev_type;
3199 uint32_t prev_type;
3200
3201 /* Unless the arch requires properties, ignore them. */
3202 if (!ARCH_USE_GNU_PROPERTY) {
3203 return true;
3204 }
3205
3206 /* If the properties are crazy large, that's too bad. */
3207 n = phdr->p_filesz;
3208 if (n > sizeof(note)) {
3209 error_setg(errp, "PT_GNU_PROPERTY too large");
3210 return false;
3211 }
3212 if (n < sizeof(note.nhdr)) {
3213 error_setg(errp, "PT_GNU_PROPERTY too small");
3214 return false;
3215 }
3216
3217 if (!imgsrc_read(¬e, phdr->p_offset, n, src, errp)) {
3218 return false;
3219 }
3220
3221 /*
3222 * The contents of a valid PT_GNU_PROPERTY is a sequence of uint32_t.
3223 * Swap most of them now, beyond the header and namesz.
3224 */
3225 if (target_needs_bswap()) {
3226 for (int i = 4; i < n / 4; i++) {
3227 bswap32s(note.data + i);
3228 }
3229 }
3230
3231 /*
3232 * Note that nhdr is 3 words, and that the "name" described by namesz
3233 * immediately follows nhdr and is thus at the 4th word. Further, all
3234 * of the inputs to the kernel's round_up are multiples of 4.
3235 */
3236 if (tswap32(note.nhdr.n_type) != NT_GNU_PROPERTY_TYPE_0 ||
3237 tswap32(note.nhdr.n_namesz) != NOTE_NAME_SZ ||
3238 note.data[3] != GNU0_MAGIC) {
3239 error_setg(errp, "Invalid note in PT_GNU_PROPERTY");
3240 return false;
3241 }
3242 off = sizeof(note.nhdr) + NOTE_NAME_SZ;
3243
3244 datasz = tswap32(note.nhdr.n_descsz) + off;
3245 if (datasz > n) {
3246 error_setg(errp, "Invalid note size in PT_GNU_PROPERTY");
3247 return false;
3248 }
3249
3250 have_prev_type = false;
3251 prev_type = 0;
3252 while (1) {
3253 if (off == datasz) {
3254 return true; /* end, exit ok */
3255 }
3256 if (!parse_elf_property(note.data, &off, datasz, info,
3257 have_prev_type, &prev_type, errp)) {
3258 return false;
3259 }
3260 have_prev_type = true;
3261 }
3262 }
3263
3264 /**
3265 * load_elf_image: Load an ELF image into the address space.
3266 * @image_name: the filename of the image, to use in error messages.
3267 * @src: the ImageSource from which to read.
3268 * @info: info collected from the loaded image.
3269 * @ehdr: the ELF header, not yet bswapped.
3270 * @pinterp_name: record any PT_INTERP string found.
3271 *
3272 * On return: @info values will be filled in, as necessary or available.
3273 */
3274
load_elf_image(const char * image_name,const ImageSource * src,struct image_info * info,struct elfhdr * ehdr,char ** pinterp_name)3275 static void load_elf_image(const char *image_name, const ImageSource *src,
3276 struct image_info *info, struct elfhdr *ehdr,
3277 char **pinterp_name)
3278 {
3279 g_autofree struct elf_phdr *phdr = NULL;
3280 abi_ulong load_addr, load_bias, loaddr, hiaddr, error, align;
3281 size_t reserve_size, align_size;
3282 int i, prot_exec;
3283 Error *err = NULL;
3284
3285 /*
3286 * First of all, some simple consistency checks.
3287 * Note that we rely on the bswapped ehdr staying in bprm_buf,
3288 * for later use by load_elf_binary and create_elf_tables.
3289 */
3290 if (!imgsrc_read(ehdr, 0, sizeof(*ehdr), src, &err)) {
3291 goto exit_errmsg;
3292 }
3293 if (!elf_check_ident(ehdr)) {
3294 error_setg(&err, "Invalid ELF image for this architecture");
3295 goto exit_errmsg;
3296 }
3297 bswap_ehdr(ehdr);
3298 if (!elf_check_ehdr(ehdr)) {
3299 error_setg(&err, "Invalid ELF image for this architecture");
3300 goto exit_errmsg;
3301 }
3302
3303 phdr = imgsrc_read_alloc(ehdr->e_phoff,
3304 ehdr->e_phnum * sizeof(struct elf_phdr),
3305 src, &err);
3306 if (phdr == NULL) {
3307 goto exit_errmsg;
3308 }
3309 bswap_phdr(phdr, ehdr->e_phnum);
3310
3311 info->nsegs = 0;
3312 info->pt_dynamic_addr = 0;
3313
3314 mmap_lock();
3315
3316 /*
3317 * Find the maximum size of the image and allocate an appropriate
3318 * amount of memory to handle that. Locate the interpreter, if any.
3319 */
3320 loaddr = -1, hiaddr = 0;
3321 align = 0;
3322 info->exec_stack = EXSTACK_DEFAULT;
3323 for (i = 0; i < ehdr->e_phnum; ++i) {
3324 struct elf_phdr *eppnt = phdr + i;
3325 if (eppnt->p_type == PT_LOAD) {
3326 abi_ulong a = eppnt->p_vaddr & TARGET_PAGE_MASK;
3327 if (a < loaddr) {
3328 loaddr = a;
3329 }
3330 a = eppnt->p_vaddr + eppnt->p_memsz - 1;
3331 if (a > hiaddr) {
3332 hiaddr = a;
3333 }
3334 ++info->nsegs;
3335 align |= eppnt->p_align;
3336 } else if (eppnt->p_type == PT_INTERP && pinterp_name) {
3337 g_autofree char *interp_name = NULL;
3338
3339 if (*pinterp_name) {
3340 error_setg(&err, "Multiple PT_INTERP entries");
3341 goto exit_errmsg;
3342 }
3343
3344 interp_name = imgsrc_read_alloc(eppnt->p_offset, eppnt->p_filesz,
3345 src, &err);
3346 if (interp_name == NULL) {
3347 goto exit_errmsg;
3348 }
3349 if (interp_name[eppnt->p_filesz - 1] != 0) {
3350 error_setg(&err, "Invalid PT_INTERP entry");
3351 goto exit_errmsg;
3352 }
3353 *pinterp_name = g_steal_pointer(&interp_name);
3354 } else if (eppnt->p_type == PT_GNU_PROPERTY) {
3355 if (!parse_elf_properties(src, info, eppnt, &err)) {
3356 goto exit_errmsg;
3357 }
3358 } else if (eppnt->p_type == PT_GNU_STACK) {
3359 info->exec_stack = eppnt->p_flags & PF_X;
3360 }
3361 }
3362
3363 load_addr = loaddr;
3364
3365 align = pow2ceil(align);
3366
3367 if (pinterp_name != NULL) {
3368 if (ehdr->e_type == ET_EXEC) {
3369 /*
3370 * Make sure that the low address does not conflict with
3371 * MMAP_MIN_ADDR or the QEMU application itself.
3372 */
3373 probe_guest_base(image_name, loaddr, hiaddr);
3374 } else {
3375 /*
3376 * The binary is dynamic, but we still need to
3377 * select guest_base. In this case we pass a size.
3378 */
3379 probe_guest_base(image_name, 0, hiaddr - loaddr);
3380
3381 /*
3382 * Avoid collision with the loader by providing a different
3383 * default load address.
3384 */
3385 load_addr += elf_et_dyn_base;
3386
3387 /*
3388 * TODO: Better support for mmap alignment is desirable.
3389 * Since we do not have complete control over the guest
3390 * address space, we prefer the kernel to choose some address
3391 * rather than force the use of LOAD_ADDR via MAP_FIXED.
3392 */
3393 if (align) {
3394 load_addr &= -align;
3395 }
3396 }
3397 }
3398
3399 /*
3400 * Reserve address space for all of this.
3401 *
3402 * In the case of ET_EXEC, we supply MAP_FIXED_NOREPLACE so that we get
3403 * exactly the address range that is required. Without reserved_va,
3404 * the guest address space is not isolated. We have attempted to avoid
3405 * conflict with the host program itself via probe_guest_base, but using
3406 * MAP_FIXED_NOREPLACE instead of MAP_FIXED provides an extra check.
3407 *
3408 * Otherwise this is ET_DYN, and we are searching for a location
3409 * that can hold the memory space required. If the image is
3410 * pre-linked, LOAD_ADDR will be non-zero, and the kernel should
3411 * honor that address if it happens to be free.
3412 *
3413 * In both cases, we will overwrite pages in this range with mappings
3414 * from the executable.
3415 */
3416 reserve_size = (size_t)hiaddr - loaddr + 1;
3417 align_size = reserve_size;
3418
3419 if (ehdr->e_type != ET_EXEC && align > qemu_real_host_page_size()) {
3420 align_size += align - 1;
3421 }
3422
3423 load_addr = target_mmap(load_addr, align_size, PROT_NONE,
3424 MAP_PRIVATE | MAP_ANON | MAP_NORESERVE |
3425 (ehdr->e_type == ET_EXEC ? MAP_FIXED_NOREPLACE : 0),
3426 -1, 0);
3427 if (load_addr == -1) {
3428 goto exit_mmap;
3429 }
3430
3431 if (align_size != reserve_size) {
3432 abi_ulong align_addr = ROUND_UP(load_addr, align);
3433 abi_ulong align_end = TARGET_PAGE_ALIGN(align_addr + reserve_size);
3434 abi_ulong load_end = TARGET_PAGE_ALIGN(load_addr + align_size);
3435
3436 if (align_addr != load_addr) {
3437 target_munmap(load_addr, align_addr - load_addr);
3438 }
3439 if (align_end != load_end) {
3440 target_munmap(align_end, load_end - align_end);
3441 }
3442 load_addr = align_addr;
3443 }
3444
3445 load_bias = load_addr - loaddr;
3446
3447 if (elf_is_fdpic(ehdr)) {
3448 struct elf32_fdpic_loadseg *loadsegs = info->loadsegs =
3449 g_malloc(sizeof(*loadsegs) * info->nsegs);
3450
3451 for (i = 0; i < ehdr->e_phnum; ++i) {
3452 switch (phdr[i].p_type) {
3453 case PT_DYNAMIC:
3454 info->pt_dynamic_addr = phdr[i].p_vaddr + load_bias;
3455 break;
3456 case PT_LOAD:
3457 loadsegs->addr = phdr[i].p_vaddr + load_bias;
3458 loadsegs->p_vaddr = phdr[i].p_vaddr;
3459 loadsegs->p_memsz = phdr[i].p_memsz;
3460 ++loadsegs;
3461 break;
3462 }
3463 }
3464 }
3465
3466 info->load_bias = load_bias;
3467 info->code_offset = load_bias;
3468 info->data_offset = load_bias;
3469 info->load_addr = load_addr;
3470 info->entry = ehdr->e_entry + load_bias;
3471 info->start_code = -1;
3472 info->end_code = 0;
3473 info->start_data = -1;
3474 info->end_data = 0;
3475 /* Usual start for brk is after all sections of the main executable. */
3476 info->brk = TARGET_PAGE_ALIGN(hiaddr + load_bias);
3477 info->elf_flags = ehdr->e_flags;
3478
3479 prot_exec = PROT_EXEC;
3480 #ifdef TARGET_AARCH64
3481 /*
3482 * If the BTI feature is present, this indicates that the executable
3483 * pages of the startup binary should be mapped with PROT_BTI, so that
3484 * branch targets are enforced.
3485 *
3486 * The startup binary is either the interpreter or the static executable.
3487 * The interpreter is responsible for all pages of a dynamic executable.
3488 *
3489 * Elf notes are backward compatible to older cpus.
3490 * Do not enable BTI unless it is supported.
3491 */
3492 if ((info->note_flags & GNU_PROPERTY_AARCH64_FEATURE_1_BTI)
3493 && (pinterp_name == NULL || *pinterp_name == 0)
3494 && cpu_isar_feature(aa64_bti, ARM_CPU(thread_cpu))) {
3495 prot_exec |= TARGET_PROT_BTI;
3496 }
3497 #endif
3498
3499 for (i = 0; i < ehdr->e_phnum; i++) {
3500 struct elf_phdr *eppnt = phdr + i;
3501 if (eppnt->p_type == PT_LOAD) {
3502 abi_ulong vaddr, vaddr_po, vaddr_ps, vaddr_ef, vaddr_em;
3503 int elf_prot = 0;
3504
3505 if (eppnt->p_flags & PF_R) {
3506 elf_prot |= PROT_READ;
3507 }
3508 if (eppnt->p_flags & PF_W) {
3509 elf_prot |= PROT_WRITE;
3510 }
3511 if (eppnt->p_flags & PF_X) {
3512 elf_prot |= prot_exec;
3513 }
3514
3515 vaddr = load_bias + eppnt->p_vaddr;
3516 vaddr_po = vaddr & ~TARGET_PAGE_MASK;
3517 vaddr_ps = vaddr & TARGET_PAGE_MASK;
3518
3519 vaddr_ef = vaddr + eppnt->p_filesz;
3520 vaddr_em = vaddr + eppnt->p_memsz;
3521
3522 /*
3523 * Some segments may be completely empty, with a non-zero p_memsz
3524 * but no backing file segment.
3525 */
3526 if (eppnt->p_filesz != 0) {
3527 error = imgsrc_mmap(vaddr_ps, eppnt->p_filesz + vaddr_po,
3528 elf_prot, MAP_PRIVATE | MAP_FIXED,
3529 src, eppnt->p_offset - vaddr_po);
3530 if (error == -1) {
3531 goto exit_mmap;
3532 }
3533 }
3534
3535 /* If the load segment requests extra zeros (e.g. bss), map it. */
3536 if (vaddr_ef < vaddr_em &&
3537 !zero_bss(vaddr_ef, vaddr_em, elf_prot, &err)) {
3538 goto exit_errmsg;
3539 }
3540
3541 /* Find the full program boundaries. */
3542 if (elf_prot & PROT_EXEC) {
3543 if (vaddr < info->start_code) {
3544 info->start_code = vaddr;
3545 }
3546 if (vaddr_ef > info->end_code) {
3547 info->end_code = vaddr_ef;
3548 }
3549 }
3550 if (elf_prot & PROT_WRITE) {
3551 if (vaddr < info->start_data) {
3552 info->start_data = vaddr;
3553 }
3554 if (vaddr_ef > info->end_data) {
3555 info->end_data = vaddr_ef;
3556 }
3557 }
3558 #ifdef TARGET_MIPS
3559 } else if (eppnt->p_type == PT_MIPS_ABIFLAGS) {
3560 Mips_elf_abiflags_v0 abiflags;
3561
3562 if (!imgsrc_read(&abiflags, eppnt->p_offset, sizeof(abiflags),
3563 src, &err)) {
3564 goto exit_errmsg;
3565 }
3566 bswap_mips_abiflags(&abiflags);
3567 info->fp_abi = abiflags.fp_abi;
3568 #endif
3569 }
3570 }
3571
3572 if (info->end_data == 0) {
3573 info->start_data = info->end_code;
3574 info->end_data = info->end_code;
3575 }
3576
3577 if (qemu_log_enabled()) {
3578 load_symbols(ehdr, src, load_bias);
3579 }
3580
3581 debuginfo_report_elf(image_name, src->fd, load_bias);
3582
3583 mmap_unlock();
3584
3585 close(src->fd);
3586 return;
3587
3588 exit_mmap:
3589 error_setg_errno(&err, errno, "Error mapping file");
3590 goto exit_errmsg;
3591 exit_errmsg:
3592 error_reportf_err(err, "%s: ", image_name);
3593 exit(-1);
3594 }
3595
load_elf_interp(const char * filename,struct image_info * info,char bprm_buf[BPRM_BUF_SIZE])3596 static void load_elf_interp(const char *filename, struct image_info *info,
3597 char bprm_buf[BPRM_BUF_SIZE])
3598 {
3599 struct elfhdr ehdr;
3600 ImageSource src;
3601 int fd, retval;
3602 Error *err = NULL;
3603
3604 fd = open(path(filename), O_RDONLY);
3605 if (fd < 0) {
3606 error_setg_file_open(&err, errno, filename);
3607 error_report_err(err);
3608 exit(-1);
3609 }
3610
3611 retval = read(fd, bprm_buf, BPRM_BUF_SIZE);
3612 if (retval < 0) {
3613 error_setg_errno(&err, errno, "Error reading file header");
3614 error_reportf_err(err, "%s: ", filename);
3615 exit(-1);
3616 }
3617
3618 src.fd = fd;
3619 src.cache = bprm_buf;
3620 src.cache_size = retval;
3621
3622 load_elf_image(filename, &src, info, &ehdr, NULL);
3623 }
3624
3625 #ifndef vdso_image_info
3626 #ifdef VDSO_HEADER
3627 #include VDSO_HEADER
3628 #define vdso_image_info(flags) &vdso_image_info
3629 #else
3630 #define vdso_image_info(flags) NULL
3631 #endif /* VDSO_HEADER */
3632 #endif /* vdso_image_info */
3633
load_elf_vdso(struct image_info * info,const VdsoImageInfo * vdso)3634 static void load_elf_vdso(struct image_info *info, const VdsoImageInfo *vdso)
3635 {
3636 ImageSource src;
3637 struct elfhdr ehdr;
3638 abi_ulong load_bias, load_addr;
3639
3640 src.fd = -1;
3641 src.cache = vdso->image;
3642 src.cache_size = vdso->image_size;
3643
3644 load_elf_image("<internal-vdso>", &src, info, &ehdr, NULL);
3645 load_addr = info->load_addr;
3646 load_bias = info->load_bias;
3647
3648 /*
3649 * We need to relocate the VDSO image. The one built into the kernel
3650 * is built for a fixed address. The one built for QEMU is not, since
3651 * that requires close control of the guest address space.
3652 * We pre-processed the image to locate all of the addresses that need
3653 * to be updated.
3654 */
3655 for (unsigned i = 0, n = vdso->reloc_count; i < n; i++) {
3656 abi_ulong *addr = g2h_untagged(load_addr + vdso->relocs[i]);
3657 *addr = tswapal(tswapal(*addr) + load_bias);
3658 }
3659
3660 /* Install signal trampolines, if present. */
3661 if (vdso->sigreturn_ofs) {
3662 default_sigreturn = load_addr + vdso->sigreturn_ofs;
3663 }
3664 if (vdso->rt_sigreturn_ofs) {
3665 default_rt_sigreturn = load_addr + vdso->rt_sigreturn_ofs;
3666 }
3667
3668 /* Remove write from VDSO segment. */
3669 target_mprotect(info->start_data, info->end_data - info->start_data,
3670 PROT_READ | PROT_EXEC);
3671 }
3672
symfind(const void * s0,const void * s1)3673 static int symfind(const void *s0, const void *s1)
3674 {
3675 struct elf_sym *sym = (struct elf_sym *)s1;
3676 __typeof(sym->st_value) addr = *(uint64_t *)s0;
3677 int result = 0;
3678
3679 if (addr < sym->st_value) {
3680 result = -1;
3681 } else if (addr >= sym->st_value + sym->st_size) {
3682 result = 1;
3683 }
3684 return result;
3685 }
3686
lookup_symbolxx(struct syminfo * s,uint64_t orig_addr)3687 static const char *lookup_symbolxx(struct syminfo *s, uint64_t orig_addr)
3688 {
3689 #if ELF_CLASS == ELFCLASS32
3690 struct elf_sym *syms = s->disas_symtab.elf32;
3691 #else
3692 struct elf_sym *syms = s->disas_symtab.elf64;
3693 #endif
3694
3695 // binary search
3696 struct elf_sym *sym;
3697
3698 sym = bsearch(&orig_addr, syms, s->disas_num_syms, sizeof(*syms), symfind);
3699 if (sym != NULL) {
3700 return s->disas_strtab + sym->st_name;
3701 }
3702
3703 return "";
3704 }
3705
3706 /* FIXME: This should use elf_ops.h.inc */
symcmp(const void * s0,const void * s1)3707 static int symcmp(const void *s0, const void *s1)
3708 {
3709 struct elf_sym *sym0 = (struct elf_sym *)s0;
3710 struct elf_sym *sym1 = (struct elf_sym *)s1;
3711 return (sym0->st_value < sym1->st_value)
3712 ? -1
3713 : ((sym0->st_value > sym1->st_value) ? 1 : 0);
3714 }
3715
3716 /* Best attempt to load symbols from this ELF object. */
load_symbols(struct elfhdr * hdr,const ImageSource * src,abi_ulong load_bias)3717 static void load_symbols(struct elfhdr *hdr, const ImageSource *src,
3718 abi_ulong load_bias)
3719 {
3720 int i, shnum, nsyms, sym_idx = 0, str_idx = 0;
3721 g_autofree struct elf_shdr *shdr = NULL;
3722 char *strings = NULL;
3723 struct elf_sym *syms = NULL;
3724 struct elf_sym *new_syms;
3725 uint64_t segsz;
3726
3727 shnum = hdr->e_shnum;
3728 shdr = imgsrc_read_alloc(hdr->e_shoff, shnum * sizeof(struct elf_shdr),
3729 src, NULL);
3730 if (shdr == NULL) {
3731 return;
3732 }
3733
3734 bswap_shdr(shdr, shnum);
3735 for (i = 0; i < shnum; ++i) {
3736 if (shdr[i].sh_type == SHT_SYMTAB) {
3737 sym_idx = i;
3738 str_idx = shdr[i].sh_link;
3739 goto found;
3740 }
3741 }
3742
3743 /* There will be no symbol table if the file was stripped. */
3744 return;
3745
3746 found:
3747 /* Now know where the strtab and symtab are. Snarf them. */
3748
3749 segsz = shdr[str_idx].sh_size;
3750 strings = g_try_malloc(segsz);
3751 if (!strings) {
3752 goto give_up;
3753 }
3754 if (!imgsrc_read(strings, shdr[str_idx].sh_offset, segsz, src, NULL)) {
3755 goto give_up;
3756 }
3757
3758 segsz = shdr[sym_idx].sh_size;
3759 if (segsz / sizeof(struct elf_sym) > INT_MAX) {
3760 /*
3761 * Implausibly large symbol table: give up rather than ploughing
3762 * on with the number of symbols calculation overflowing.
3763 */
3764 goto give_up;
3765 }
3766 nsyms = segsz / sizeof(struct elf_sym);
3767 syms = g_try_malloc(segsz);
3768 if (!syms) {
3769 goto give_up;
3770 }
3771 if (!imgsrc_read(syms, shdr[sym_idx].sh_offset, segsz, src, NULL)) {
3772 goto give_up;
3773 }
3774
3775 for (i = 0; i < nsyms; ) {
3776 bswap_sym(syms + i);
3777 /* Throw away entries which we do not need. */
3778 if (syms[i].st_shndx == SHN_UNDEF
3779 || syms[i].st_shndx >= SHN_LORESERVE
3780 || ELF_ST_TYPE(syms[i].st_info) != STT_FUNC) {
3781 if (i < --nsyms) {
3782 syms[i] = syms[nsyms];
3783 }
3784 } else {
3785 #if defined(TARGET_ARM) || defined (TARGET_MIPS)
3786 /* The bottom address bit marks a Thumb or MIPS16 symbol. */
3787 syms[i].st_value &= ~(target_ulong)1;
3788 #endif
3789 syms[i].st_value += load_bias;
3790 i++;
3791 }
3792 }
3793
3794 /* No "useful" symbol. */
3795 if (nsyms == 0) {
3796 goto give_up;
3797 }
3798
3799 /*
3800 * Attempt to free the storage associated with the local symbols
3801 * that we threw away. Whether or not this has any effect on the
3802 * memory allocation depends on the malloc implementation and how
3803 * many symbols we managed to discard.
3804 */
3805 new_syms = g_try_renew(struct elf_sym, syms, nsyms);
3806 if (new_syms == NULL) {
3807 goto give_up;
3808 }
3809 syms = new_syms;
3810
3811 qsort(syms, nsyms, sizeof(*syms), symcmp);
3812
3813 {
3814 struct syminfo *s = g_new(struct syminfo, 1);
3815
3816 s->disas_strtab = strings;
3817 s->disas_num_syms = nsyms;
3818 #if ELF_CLASS == ELFCLASS32
3819 s->disas_symtab.elf32 = syms;
3820 #else
3821 s->disas_symtab.elf64 = syms;
3822 #endif
3823 s->lookup_symbol = lookup_symbolxx;
3824 s->next = syminfos;
3825 syminfos = s;
3826 }
3827 return;
3828
3829 give_up:
3830 g_free(strings);
3831 g_free(syms);
3832 }
3833
get_elf_eflags(int fd)3834 uint32_t get_elf_eflags(int fd)
3835 {
3836 struct elfhdr ehdr;
3837 off_t offset;
3838 int ret;
3839
3840 /* Read ELF header */
3841 offset = lseek(fd, 0, SEEK_SET);
3842 if (offset == (off_t) -1) {
3843 return 0;
3844 }
3845 ret = read(fd, &ehdr, sizeof(ehdr));
3846 if (ret < sizeof(ehdr)) {
3847 return 0;
3848 }
3849 offset = lseek(fd, offset, SEEK_SET);
3850 if (offset == (off_t) -1) {
3851 return 0;
3852 }
3853
3854 /* Check ELF signature */
3855 if (!elf_check_ident(&ehdr)) {
3856 return 0;
3857 }
3858
3859 /* check header */
3860 bswap_ehdr(&ehdr);
3861 if (!elf_check_ehdr(&ehdr)) {
3862 return 0;
3863 }
3864
3865 /* return architecture id */
3866 return ehdr.e_flags;
3867 }
3868
load_elf_binary(struct linux_binprm * bprm,struct image_info * info)3869 int load_elf_binary(struct linux_binprm *bprm, struct image_info *info)
3870 {
3871 /*
3872 * We need a copy of the elf header for passing to create_elf_tables.
3873 * We will have overwritten the original when we re-use bprm->buf
3874 * while loading the interpreter. Allocate the storage for this now
3875 * and let elf_load_image do any swapping that may be required.
3876 */
3877 struct elfhdr ehdr;
3878 struct image_info interp_info, vdso_info;
3879 char *elf_interpreter = NULL;
3880 char *scratch;
3881
3882 memset(&interp_info, 0, sizeof(interp_info));
3883 #ifdef TARGET_MIPS
3884 interp_info.fp_abi = MIPS_ABI_FP_UNKNOWN;
3885 #endif
3886
3887 load_elf_image(bprm->filename, &bprm->src, info, &ehdr, &elf_interpreter);
3888
3889 /* Do this so that we can load the interpreter, if need be. We will
3890 change some of these later */
3891 bprm->p = setup_arg_pages(bprm, info);
3892
3893 scratch = g_new0(char, TARGET_PAGE_SIZE);
3894 if (STACK_GROWS_DOWN) {
3895 bprm->p = copy_elf_strings(1, &bprm->filename, scratch,
3896 bprm->p, info->stack_limit);
3897 info->file_string = bprm->p;
3898 bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch,
3899 bprm->p, info->stack_limit);
3900 info->env_strings = bprm->p;
3901 bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch,
3902 bprm->p, info->stack_limit);
3903 info->arg_strings = bprm->p;
3904 } else {
3905 info->arg_strings = bprm->p;
3906 bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch,
3907 bprm->p, info->stack_limit);
3908 info->env_strings = bprm->p;
3909 bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch,
3910 bprm->p, info->stack_limit);
3911 info->file_string = bprm->p;
3912 bprm->p = copy_elf_strings(1, &bprm->filename, scratch,
3913 bprm->p, info->stack_limit);
3914 }
3915
3916 g_free(scratch);
3917
3918 if (!bprm->p) {
3919 fprintf(stderr, "%s: %s\n", bprm->filename, strerror(E2BIG));
3920 exit(-1);
3921 }
3922
3923 if (elf_interpreter) {
3924 load_elf_interp(elf_interpreter, &interp_info, bprm->buf);
3925
3926 /*
3927 * While unusual because of ELF_ET_DYN_BASE, if we are unlucky
3928 * with the mappings the interpreter can be loaded above but
3929 * near the main executable, which can leave very little room
3930 * for the heap.
3931 * If the current brk has less than 16MB, use the end of the
3932 * interpreter.
3933 */
3934 if (interp_info.brk > info->brk &&
3935 interp_info.load_bias - info->brk < 16 * MiB) {
3936 info->brk = interp_info.brk;
3937 }
3938
3939 /* If the program interpreter is one of these two, then assume
3940 an iBCS2 image. Otherwise assume a native linux image. */
3941
3942 if (strcmp(elf_interpreter, "/usr/lib/libc.so.1") == 0
3943 || strcmp(elf_interpreter, "/usr/lib/ld.so.1") == 0) {
3944 info->personality = PER_SVR4;
3945
3946 /* Why this, you ask??? Well SVr4 maps page 0 as read-only,
3947 and some applications "depend" upon this behavior. Since
3948 we do not have the power to recompile these, we emulate
3949 the SVr4 behavior. Sigh. */
3950 target_mmap(0, TARGET_PAGE_SIZE, PROT_READ | PROT_EXEC,
3951 MAP_FIXED_NOREPLACE | MAP_PRIVATE | MAP_ANONYMOUS,
3952 -1, 0);
3953 }
3954 #ifdef TARGET_MIPS
3955 info->interp_fp_abi = interp_info.fp_abi;
3956 #endif
3957 }
3958
3959 /*
3960 * Load a vdso if available, which will amongst other things contain the
3961 * signal trampolines. Otherwise, allocate a separate page for them.
3962 */
3963 const VdsoImageInfo *vdso = vdso_image_info(info->elf_flags);
3964 if (vdso) {
3965 load_elf_vdso(&vdso_info, vdso);
3966 info->vdso = vdso_info.load_bias;
3967 } else if (TARGET_ARCH_HAS_SIGTRAMP_PAGE) {
3968 abi_long tramp_page = target_mmap(0, TARGET_PAGE_SIZE,
3969 PROT_READ | PROT_WRITE,
3970 MAP_PRIVATE | MAP_ANON, -1, 0);
3971 if (tramp_page == -1) {
3972 return -errno;
3973 }
3974
3975 setup_sigtramp(tramp_page);
3976 target_mprotect(tramp_page, TARGET_PAGE_SIZE, PROT_READ | PROT_EXEC);
3977 }
3978
3979 bprm->p = create_elf_tables(bprm->p, bprm->argc, bprm->envc, &ehdr, info,
3980 elf_interpreter ? &interp_info : NULL,
3981 vdso ? &vdso_info : NULL);
3982 info->start_stack = bprm->p;
3983
3984 /* If we have an interpreter, set that as the program's entry point.
3985 Copy the load_bias as well, to help PPC64 interpret the entry
3986 point as a function descriptor. Do this after creating elf tables
3987 so that we copy the original program entry point into the AUXV. */
3988 if (elf_interpreter) {
3989 info->load_bias = interp_info.load_bias;
3990 info->entry = interp_info.entry;
3991 g_free(elf_interpreter);
3992 }
3993
3994 #ifdef USE_ELF_CORE_DUMP
3995 bprm->core_dump = &elf_core_dump;
3996 #endif
3997
3998 return 0;
3999 }
4000
4001 #ifdef USE_ELF_CORE_DUMP
4002
4003 /*
4004 * Definitions to generate Intel SVR4-like core files.
4005 * These mostly have the same names as the SVR4 types with "target_elf_"
4006 * tacked on the front to prevent clashes with linux definitions,
4007 * and the typedef forms have been avoided. This is mostly like
4008 * the SVR4 structure, but more Linuxy, with things that Linux does
4009 * not support and which gdb doesn't really use excluded.
4010 *
4011 * Fields we don't dump (their contents is zero) in linux-user qemu
4012 * are marked with XXX.
4013 *
4014 * Core dump code is copied from linux kernel (fs/binfmt_elf.c).
4015 *
4016 * Porting ELF coredump for target is (quite) simple process. First you
4017 * define USE_ELF_CORE_DUMP in target ELF code (where init_thread() for
4018 * the target resides):
4019 *
4020 * #define USE_ELF_CORE_DUMP
4021 *
4022 * Next you define type of register set used for dumping. ELF specification
4023 * says that it needs to be array of elf_greg_t that has size of ELF_NREG.
4024 *
4025 * typedef <target_regtype> target_elf_greg_t;
4026 * #define ELF_NREG <number of registers>
4027 * typedef taret_elf_greg_t target_elf_gregset_t[ELF_NREG];
4028 *
4029 * Last step is to implement target specific function that copies registers
4030 * from given cpu into just specified register set. Prototype is:
4031 *
4032 * static void elf_core_copy_regs(taret_elf_gregset_t *regs,
4033 * const CPUArchState *env);
4034 *
4035 * Parameters:
4036 * regs - copy register values into here (allocated and zeroed by caller)
4037 * env - copy registers from here
4038 *
4039 * Example for ARM target is provided in this file.
4040 */
4041
4042 struct target_elf_siginfo {
4043 abi_int si_signo; /* signal number */
4044 abi_int si_code; /* extra code */
4045 abi_int si_errno; /* errno */
4046 };
4047
4048 struct target_elf_prstatus {
4049 struct target_elf_siginfo pr_info; /* Info associated with signal */
4050 abi_short pr_cursig; /* Current signal */
4051 abi_ulong pr_sigpend; /* XXX */
4052 abi_ulong pr_sighold; /* XXX */
4053 target_pid_t pr_pid;
4054 target_pid_t pr_ppid;
4055 target_pid_t pr_pgrp;
4056 target_pid_t pr_sid;
4057 struct target_timeval pr_utime; /* XXX User time */
4058 struct target_timeval pr_stime; /* XXX System time */
4059 struct target_timeval pr_cutime; /* XXX Cumulative user time */
4060 struct target_timeval pr_cstime; /* XXX Cumulative system time */
4061 target_elf_gregset_t pr_reg; /* GP registers */
4062 abi_int pr_fpvalid; /* XXX */
4063 };
4064
4065 #define ELF_PRARGSZ (80) /* Number of chars for args */
4066
4067 struct target_elf_prpsinfo {
4068 char pr_state; /* numeric process state */
4069 char pr_sname; /* char for pr_state */
4070 char pr_zomb; /* zombie */
4071 char pr_nice; /* nice val */
4072 abi_ulong pr_flag; /* flags */
4073 target_uid_t pr_uid;
4074 target_gid_t pr_gid;
4075 target_pid_t pr_pid, pr_ppid, pr_pgrp, pr_sid;
4076 /* Lots missing */
4077 char pr_fname[16] QEMU_NONSTRING; /* filename of executable */
4078 char pr_psargs[ELF_PRARGSZ]; /* initial part of arg list */
4079 };
4080
bswap_prstatus(struct target_elf_prstatus * prstatus)4081 static void bswap_prstatus(struct target_elf_prstatus *prstatus)
4082 {
4083 if (!target_needs_bswap()) {
4084 return;
4085 }
4086
4087 prstatus->pr_info.si_signo = tswap32(prstatus->pr_info.si_signo);
4088 prstatus->pr_info.si_code = tswap32(prstatus->pr_info.si_code);
4089 prstatus->pr_info.si_errno = tswap32(prstatus->pr_info.si_errno);
4090 prstatus->pr_cursig = tswap16(prstatus->pr_cursig);
4091 prstatus->pr_sigpend = tswapal(prstatus->pr_sigpend);
4092 prstatus->pr_sighold = tswapal(prstatus->pr_sighold);
4093 prstatus->pr_pid = tswap32(prstatus->pr_pid);
4094 prstatus->pr_ppid = tswap32(prstatus->pr_ppid);
4095 prstatus->pr_pgrp = tswap32(prstatus->pr_pgrp);
4096 prstatus->pr_sid = tswap32(prstatus->pr_sid);
4097 /* cpu times are not filled, so we skip them */
4098 /* regs should be in correct format already */
4099 prstatus->pr_fpvalid = tswap32(prstatus->pr_fpvalid);
4100 }
4101
bswap_psinfo(struct target_elf_prpsinfo * psinfo)4102 static void bswap_psinfo(struct target_elf_prpsinfo *psinfo)
4103 {
4104 if (!target_needs_bswap()) {
4105 return;
4106 }
4107
4108 psinfo->pr_flag = tswapal(psinfo->pr_flag);
4109 psinfo->pr_uid = tswap16(psinfo->pr_uid);
4110 psinfo->pr_gid = tswap16(psinfo->pr_gid);
4111 psinfo->pr_pid = tswap32(psinfo->pr_pid);
4112 psinfo->pr_ppid = tswap32(psinfo->pr_ppid);
4113 psinfo->pr_pgrp = tswap32(psinfo->pr_pgrp);
4114 psinfo->pr_sid = tswap32(psinfo->pr_sid);
4115 }
4116
bswap_note(struct elf_note * en)4117 static void bswap_note(struct elf_note *en)
4118 {
4119 if (!target_needs_bswap()) {
4120 return;
4121 }
4122
4123 bswap32s(&en->n_namesz);
4124 bswap32s(&en->n_descsz);
4125 bswap32s(&en->n_type);
4126 }
4127
4128 /*
4129 * Calculate file (dump) size of given memory region.
4130 */
vma_dump_size(vaddr start,vaddr end,int flags)4131 static size_t vma_dump_size(vaddr start, vaddr end, int flags)
4132 {
4133 /* The area must be readable. */
4134 if (!(flags & PAGE_READ)) {
4135 return 0;
4136 }
4137
4138 /*
4139 * Usually we don't dump executable pages as they contain
4140 * non-writable code that debugger can read directly from
4141 * target library etc. If there is no elf header, we dump it.
4142 */
4143 if (!(flags & PAGE_WRITE_ORG) &&
4144 (flags & PAGE_EXEC) &&
4145 memcmp(g2h_untagged(start), ELFMAG, SELFMAG) == 0) {
4146 return 0;
4147 }
4148
4149 return end - start;
4150 }
4151
size_note(const char * name,size_t datasz)4152 static size_t size_note(const char *name, size_t datasz)
4153 {
4154 size_t namesz = strlen(name) + 1;
4155
4156 namesz = ROUND_UP(namesz, 4);
4157 datasz = ROUND_UP(datasz, 4);
4158
4159 return sizeof(struct elf_note) + namesz + datasz;
4160 }
4161
fill_note(void ** pptr,int type,const char * name,size_t datasz)4162 static void *fill_note(void **pptr, int type, const char *name, size_t datasz)
4163 {
4164 void *ptr = *pptr;
4165 struct elf_note *n = ptr;
4166 size_t namesz = strlen(name) + 1;
4167
4168 n->n_namesz = namesz;
4169 n->n_descsz = datasz;
4170 n->n_type = type;
4171 bswap_note(n);
4172
4173 ptr += sizeof(*n);
4174 memcpy(ptr, name, namesz);
4175
4176 namesz = ROUND_UP(namesz, 4);
4177 datasz = ROUND_UP(datasz, 4);
4178
4179 *pptr = ptr + namesz + datasz;
4180 return ptr + namesz;
4181 }
4182
fill_elf_header(struct elfhdr * elf,int segs,uint16_t machine,uint32_t flags)4183 static void fill_elf_header(struct elfhdr *elf, int segs, uint16_t machine,
4184 uint32_t flags)
4185 {
4186 memcpy(elf->e_ident, ELFMAG, SELFMAG);
4187
4188 elf->e_ident[EI_CLASS] = ELF_CLASS;
4189 elf->e_ident[EI_DATA] = ELF_DATA;
4190 elf->e_ident[EI_VERSION] = EV_CURRENT;
4191 elf->e_ident[EI_OSABI] = ELF_OSABI;
4192
4193 elf->e_type = ET_CORE;
4194 elf->e_machine = machine;
4195 elf->e_version = EV_CURRENT;
4196 elf->e_phoff = sizeof(struct elfhdr);
4197 elf->e_flags = flags;
4198 elf->e_ehsize = sizeof(struct elfhdr);
4199 elf->e_phentsize = sizeof(struct elf_phdr);
4200 elf->e_phnum = segs;
4201
4202 bswap_ehdr(elf);
4203 }
4204
fill_elf_note_phdr(struct elf_phdr * phdr,size_t sz,off_t offset)4205 static void fill_elf_note_phdr(struct elf_phdr *phdr, size_t sz, off_t offset)
4206 {
4207 phdr->p_type = PT_NOTE;
4208 phdr->p_offset = offset;
4209 phdr->p_filesz = sz;
4210
4211 bswap_phdr(phdr, 1);
4212 }
4213
fill_prstatus_note(void * data,CPUState * cpu,int signr)4214 static void fill_prstatus_note(void *data, CPUState *cpu, int signr)
4215 {
4216 /*
4217 * Because note memory is only aligned to 4, and target_elf_prstatus
4218 * may well have higher alignment requirements, fill locally and
4219 * memcpy to the destination afterward.
4220 */
4221 struct target_elf_prstatus prstatus = {
4222 .pr_info.si_signo = signr,
4223 .pr_cursig = signr,
4224 .pr_pid = get_task_state(cpu)->ts_tid,
4225 .pr_ppid = getppid(),
4226 .pr_pgrp = getpgrp(),
4227 .pr_sid = getsid(0),
4228 };
4229
4230 elf_core_copy_regs(&prstatus.pr_reg, cpu_env(cpu));
4231 bswap_prstatus(&prstatus);
4232 memcpy(data, &prstatus, sizeof(prstatus));
4233 }
4234
fill_prpsinfo_note(void * data,const TaskState * ts)4235 static void fill_prpsinfo_note(void *data, const TaskState *ts)
4236 {
4237 /*
4238 * Because note memory is only aligned to 4, and target_elf_prpsinfo
4239 * may well have higher alignment requirements, fill locally and
4240 * memcpy to the destination afterward.
4241 */
4242 struct target_elf_prpsinfo psinfo = {
4243 .pr_pid = getpid(),
4244 .pr_ppid = getppid(),
4245 .pr_pgrp = getpgrp(),
4246 .pr_sid = getsid(0),
4247 .pr_uid = getuid(),
4248 .pr_gid = getgid(),
4249 };
4250 char *base_filename;
4251 size_t len;
4252
4253 len = ts->info->env_strings - ts->info->arg_strings;
4254 len = MIN(len, ELF_PRARGSZ);
4255 memcpy(&psinfo.pr_psargs, g2h_untagged(ts->info->arg_strings), len);
4256 for (size_t i = 0; i < len; i++) {
4257 if (psinfo.pr_psargs[i] == 0) {
4258 psinfo.pr_psargs[i] = ' ';
4259 }
4260 }
4261
4262 base_filename = g_path_get_basename(ts->bprm->filename);
4263 /*
4264 * Using strncpy here is fine: at max-length,
4265 * this field is not NUL-terminated.
4266 */
4267 strncpy(psinfo.pr_fname, base_filename, sizeof(psinfo.pr_fname));
4268 g_free(base_filename);
4269
4270 bswap_psinfo(&psinfo);
4271 memcpy(data, &psinfo, sizeof(psinfo));
4272 }
4273
fill_auxv_note(void * data,const TaskState * ts)4274 static void fill_auxv_note(void *data, const TaskState *ts)
4275 {
4276 memcpy(data, g2h_untagged(ts->info->saved_auxv), ts->info->auxv_len);
4277 }
4278
4279 /*
4280 * Constructs name of coredump file. We have following convention
4281 * for the name:
4282 * qemu_<basename-of-target-binary>_<date>-<time>_<pid>.core
4283 *
4284 * Returns the filename
4285 */
core_dump_filename(const TaskState * ts)4286 static char *core_dump_filename(const TaskState *ts)
4287 {
4288 g_autoptr(GDateTime) now = g_date_time_new_now_local();
4289 g_autofree char *nowstr = g_date_time_format(now, "%Y%m%d-%H%M%S");
4290 g_autofree char *base_filename = g_path_get_basename(ts->bprm->filename);
4291
4292 return g_strdup_printf("qemu_%s_%s_%d.core",
4293 base_filename, nowstr, (int)getpid());
4294 }
4295
dump_write(int fd,const void * ptr,size_t size)4296 static int dump_write(int fd, const void *ptr, size_t size)
4297 {
4298 const char *bufp = (const char *)ptr;
4299 ssize_t bytes_written, bytes_left;
4300
4301 bytes_written = 0;
4302 bytes_left = size;
4303
4304 /*
4305 * In normal conditions, single write(2) should do but
4306 * in case of socket etc. this mechanism is more portable.
4307 */
4308 do {
4309 bytes_written = write(fd, bufp, bytes_left);
4310 if (bytes_written < 0) {
4311 if (errno == EINTR)
4312 continue;
4313 return (-1);
4314 } else if (bytes_written == 0) { /* eof */
4315 return (-1);
4316 }
4317 bufp += bytes_written;
4318 bytes_left -= bytes_written;
4319 } while (bytes_left > 0);
4320
4321 return (0);
4322 }
4323
wmr_page_unprotect_regions(void * opaque,vaddr start,vaddr end,int flags)4324 static int wmr_page_unprotect_regions(void *opaque, vaddr start,
4325 vaddr end, int flags)
4326 {
4327 if ((flags & (PAGE_WRITE | PAGE_WRITE_ORG)) == PAGE_WRITE_ORG) {
4328 size_t step = MAX(TARGET_PAGE_SIZE, qemu_real_host_page_size());
4329
4330 while (1) {
4331 page_unprotect(NULL, start, 0);
4332 if (end - start <= step) {
4333 break;
4334 }
4335 start += step;
4336 }
4337 }
4338 return 0;
4339 }
4340
4341 typedef struct {
4342 unsigned count;
4343 size_t size;
4344 } CountAndSizeRegions;
4345
wmr_count_and_size_regions(void * opaque,vaddr start,vaddr end,int flags)4346 static int wmr_count_and_size_regions(void *opaque, vaddr start,
4347 vaddr end, int flags)
4348 {
4349 CountAndSizeRegions *css = opaque;
4350
4351 css->count++;
4352 css->size += vma_dump_size(start, end, flags);
4353 return 0;
4354 }
4355
4356 typedef struct {
4357 struct elf_phdr *phdr;
4358 off_t offset;
4359 } FillRegionPhdr;
4360
wmr_fill_region_phdr(void * opaque,vaddr start,vaddr end,int flags)4361 static int wmr_fill_region_phdr(void *opaque, vaddr start,
4362 vaddr end, int flags)
4363 {
4364 FillRegionPhdr *d = opaque;
4365 struct elf_phdr *phdr = d->phdr;
4366
4367 phdr->p_type = PT_LOAD;
4368 phdr->p_vaddr = start;
4369 phdr->p_paddr = 0;
4370 phdr->p_filesz = vma_dump_size(start, end, flags);
4371 phdr->p_offset = d->offset;
4372 d->offset += phdr->p_filesz;
4373 phdr->p_memsz = end - start;
4374 phdr->p_flags = (flags & PAGE_READ ? PF_R : 0)
4375 | (flags & PAGE_WRITE_ORG ? PF_W : 0)
4376 | (flags & PAGE_EXEC ? PF_X : 0);
4377 phdr->p_align = ELF_EXEC_PAGESIZE;
4378
4379 bswap_phdr(phdr, 1);
4380 d->phdr = phdr + 1;
4381 return 0;
4382 }
4383
wmr_write_region(void * opaque,vaddr start,vaddr end,int flags)4384 static int wmr_write_region(void *opaque, vaddr start,
4385 vaddr end, int flags)
4386 {
4387 int fd = *(int *)opaque;
4388 size_t size = vma_dump_size(start, end, flags);
4389
4390 if (!size) {
4391 return 0;
4392 }
4393 return dump_write(fd, g2h_untagged(start), size);
4394 }
4395
4396 /*
4397 * Write out ELF coredump.
4398 *
4399 * See documentation of ELF object file format in:
4400 * http://www.caldera.com/developers/devspecs/gabi41.pdf
4401 *
4402 * Coredump format in linux is following:
4403 *
4404 * 0 +----------------------+ \
4405 * | ELF header | ET_CORE |
4406 * +----------------------+ |
4407 * | ELF program headers | |--- headers
4408 * | - NOTE section | |
4409 * | - PT_LOAD sections | |
4410 * +----------------------+ /
4411 * | NOTEs: |
4412 * | - NT_PRSTATUS |
4413 * | - NT_PRSINFO |
4414 * | - NT_AUXV |
4415 * +----------------------+ <-- aligned to target page
4416 * | Process memory dump |
4417 * : :
4418 * . .
4419 * : :
4420 * | |
4421 * +----------------------+
4422 *
4423 * NT_PRSTATUS -> struct elf_prstatus (per thread)
4424 * NT_PRSINFO -> struct elf_prpsinfo
4425 * NT_AUXV is array of { type, value } pairs (see fill_auxv_note()).
4426 *
4427 * Format follows System V format as close as possible. Current
4428 * version limitations are as follows:
4429 * - no floating point registers are dumped
4430 *
4431 * Function returns 0 in case of success, negative errno otherwise.
4432 *
4433 * TODO: make this work also during runtime: it should be
4434 * possible to force coredump from running process and then
4435 * continue processing. For example qemu could set up SIGUSR2
4436 * handler (provided that target process haven't registered
4437 * handler for that) that does the dump when signal is received.
4438 */
elf_core_dump(int signr,const CPUArchState * env)4439 static int elf_core_dump(int signr, const CPUArchState *env)
4440 {
4441 const CPUState *cpu = env_cpu_const(env);
4442 const TaskState *ts = (const TaskState *)get_task_state((CPUState *)cpu);
4443 struct rlimit dumpsize;
4444 CountAndSizeRegions css;
4445 off_t offset, note_offset, data_offset;
4446 size_t note_size;
4447 int cpus, ret;
4448 int fd = -1;
4449 CPUState *cpu_iter;
4450
4451 if (prctl(PR_GET_DUMPABLE) == 0) {
4452 return 0;
4453 }
4454
4455 if (getrlimit(RLIMIT_CORE, &dumpsize) < 0 || dumpsize.rlim_cur == 0) {
4456 return 0;
4457 }
4458
4459 cpu_list_lock();
4460 mmap_lock();
4461
4462 /* By unprotecting, we merge vmas that might be split. */
4463 walk_memory_regions(NULL, wmr_page_unprotect_regions);
4464
4465 /*
4466 * Walk through target process memory mappings and
4467 * set up structure containing this information.
4468 */
4469 memset(&css, 0, sizeof(css));
4470 walk_memory_regions(&css, wmr_count_and_size_regions);
4471
4472 cpus = 0;
4473 CPU_FOREACH(cpu_iter) {
4474 cpus++;
4475 }
4476
4477 offset = sizeof(struct elfhdr);
4478 offset += (css.count + 1) * sizeof(struct elf_phdr);
4479 note_offset = offset;
4480
4481 offset += size_note("CORE", ts->info->auxv_len);
4482 offset += size_note("CORE", sizeof(struct target_elf_prpsinfo));
4483 offset += size_note("CORE", sizeof(struct target_elf_prstatus)) * cpus;
4484 note_size = offset - note_offset;
4485 data_offset = ROUND_UP(offset, ELF_EXEC_PAGESIZE);
4486
4487 /* Do not dump if the corefile size exceeds the limit. */
4488 if (dumpsize.rlim_cur != RLIM_INFINITY
4489 && dumpsize.rlim_cur < data_offset + css.size) {
4490 errno = 0;
4491 goto out;
4492 }
4493
4494 {
4495 g_autofree char *corefile = core_dump_filename(ts);
4496 fd = open(corefile, O_WRONLY | O_CREAT | O_TRUNC,
4497 S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH);
4498 }
4499 if (fd < 0) {
4500 goto out;
4501 }
4502
4503 /*
4504 * There is a fair amount of alignment padding within the notes
4505 * as well as preceeding the process memory. Allocate a zeroed
4506 * block to hold it all. Write all of the headers directly into
4507 * this buffer and then write it out as a block.
4508 */
4509 {
4510 g_autofree void *header = g_malloc0(data_offset);
4511 FillRegionPhdr frp;
4512 void *hptr, *dptr;
4513
4514 /* Create elf file header. */
4515 hptr = header;
4516 fill_elf_header(hptr, css.count + 1, ELF_MACHINE, 0);
4517 hptr += sizeof(struct elfhdr);
4518
4519 /* Create elf program headers. */
4520 fill_elf_note_phdr(hptr, note_size, note_offset);
4521 hptr += sizeof(struct elf_phdr);
4522
4523 frp.phdr = hptr;
4524 frp.offset = data_offset;
4525 walk_memory_regions(&frp, wmr_fill_region_phdr);
4526 hptr = frp.phdr;
4527
4528 /* Create the notes. */
4529 dptr = fill_note(&hptr, NT_AUXV, "CORE", ts->info->auxv_len);
4530 fill_auxv_note(dptr, ts);
4531
4532 dptr = fill_note(&hptr, NT_PRPSINFO, "CORE",
4533 sizeof(struct target_elf_prpsinfo));
4534 fill_prpsinfo_note(dptr, ts);
4535
4536 CPU_FOREACH(cpu_iter) {
4537 dptr = fill_note(&hptr, NT_PRSTATUS, "CORE",
4538 sizeof(struct target_elf_prstatus));
4539 fill_prstatus_note(dptr, cpu_iter, cpu_iter == cpu ? signr : 0);
4540 }
4541
4542 if (dump_write(fd, header, data_offset) < 0) {
4543 goto out;
4544 }
4545 }
4546
4547 /*
4548 * Finally write process memory into the corefile as well.
4549 */
4550 if (walk_memory_regions(&fd, wmr_write_region) < 0) {
4551 goto out;
4552 }
4553 errno = 0;
4554
4555 out:
4556 ret = -errno;
4557 mmap_unlock();
4558 cpu_list_unlock();
4559 if (fd >= 0) {
4560 close(fd);
4561 }
4562 return ret;
4563 }
4564 #endif /* USE_ELF_CORE_DUMP */
4565
do_init_thread(struct target_pt_regs * regs,struct image_info * infop)4566 void do_init_thread(struct target_pt_regs *regs, struct image_info *infop)
4567 {
4568 init_thread(regs, infop);
4569 }
4570