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