xref: /qemu/linux-user/elfload.c (revision 74781c08)
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(env->regs[R_EAX]); /* XXX */
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(env->regs[R_EAX]); /* XXX */
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 #else
663 /* 64 bit ARM definitions */
664 
665 #define ELF_ARCH        EM_AARCH64
666 #define ELF_CLASS       ELFCLASS64
667 #if TARGET_BIG_ENDIAN
668 # define ELF_PLATFORM    "aarch64_be"
669 #else
670 # define ELF_PLATFORM    "aarch64"
671 #endif
672 
init_thread(struct target_pt_regs * regs,struct image_info * infop)673 static inline void init_thread(struct target_pt_regs *regs,
674                                struct image_info *infop)
675 {
676     abi_long stack = infop->start_stack;
677     memset(regs, 0, sizeof(*regs));
678 
679     regs->pc = infop->entry & ~0x3ULL;
680     regs->sp = stack;
681 }
682 
683 #define ELF_NREG    34
684 typedef target_elf_greg_t  target_elf_gregset_t[ELF_NREG];
685 
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUARMState * env)686 static void elf_core_copy_regs(target_elf_gregset_t *regs,
687                                const CPUARMState *env)
688 {
689     int i;
690 
691     for (i = 0; i < 32; i++) {
692         (*regs)[i] = tswapreg(env->xregs[i]);
693     }
694     (*regs)[32] = tswapreg(env->pc);
695     (*regs)[33] = tswapreg(pstate_read((CPUARMState *)env));
696 }
697 
698 #define USE_ELF_CORE_DUMP
699 #define ELF_EXEC_PAGESIZE       4096
700 
701 enum {
702     ARM_HWCAP_A64_FP            = 1 << 0,
703     ARM_HWCAP_A64_ASIMD         = 1 << 1,
704     ARM_HWCAP_A64_EVTSTRM       = 1 << 2,
705     ARM_HWCAP_A64_AES           = 1 << 3,
706     ARM_HWCAP_A64_PMULL         = 1 << 4,
707     ARM_HWCAP_A64_SHA1          = 1 << 5,
708     ARM_HWCAP_A64_SHA2          = 1 << 6,
709     ARM_HWCAP_A64_CRC32         = 1 << 7,
710     ARM_HWCAP_A64_ATOMICS       = 1 << 8,
711     ARM_HWCAP_A64_FPHP          = 1 << 9,
712     ARM_HWCAP_A64_ASIMDHP       = 1 << 10,
713     ARM_HWCAP_A64_CPUID         = 1 << 11,
714     ARM_HWCAP_A64_ASIMDRDM      = 1 << 12,
715     ARM_HWCAP_A64_JSCVT         = 1 << 13,
716     ARM_HWCAP_A64_FCMA          = 1 << 14,
717     ARM_HWCAP_A64_LRCPC         = 1 << 15,
718     ARM_HWCAP_A64_DCPOP         = 1 << 16,
719     ARM_HWCAP_A64_SHA3          = 1 << 17,
720     ARM_HWCAP_A64_SM3           = 1 << 18,
721     ARM_HWCAP_A64_SM4           = 1 << 19,
722     ARM_HWCAP_A64_ASIMDDP       = 1 << 20,
723     ARM_HWCAP_A64_SHA512        = 1 << 21,
724     ARM_HWCAP_A64_SVE           = 1 << 22,
725     ARM_HWCAP_A64_ASIMDFHM      = 1 << 23,
726     ARM_HWCAP_A64_DIT           = 1 << 24,
727     ARM_HWCAP_A64_USCAT         = 1 << 25,
728     ARM_HWCAP_A64_ILRCPC        = 1 << 26,
729     ARM_HWCAP_A64_FLAGM         = 1 << 27,
730     ARM_HWCAP_A64_SSBS          = 1 << 28,
731     ARM_HWCAP_A64_SB            = 1 << 29,
732     ARM_HWCAP_A64_PACA          = 1 << 30,
733     ARM_HWCAP_A64_PACG          = 1UL << 31,
734 
735     ARM_HWCAP2_A64_DCPODP       = 1 << 0,
736     ARM_HWCAP2_A64_SVE2         = 1 << 1,
737     ARM_HWCAP2_A64_SVEAES       = 1 << 2,
738     ARM_HWCAP2_A64_SVEPMULL     = 1 << 3,
739     ARM_HWCAP2_A64_SVEBITPERM   = 1 << 4,
740     ARM_HWCAP2_A64_SVESHA3      = 1 << 5,
741     ARM_HWCAP2_A64_SVESM4       = 1 << 6,
742     ARM_HWCAP2_A64_FLAGM2       = 1 << 7,
743     ARM_HWCAP2_A64_FRINT        = 1 << 8,
744     ARM_HWCAP2_A64_SVEI8MM      = 1 << 9,
745     ARM_HWCAP2_A64_SVEF32MM     = 1 << 10,
746     ARM_HWCAP2_A64_SVEF64MM     = 1 << 11,
747     ARM_HWCAP2_A64_SVEBF16      = 1 << 12,
748     ARM_HWCAP2_A64_I8MM         = 1 << 13,
749     ARM_HWCAP2_A64_BF16         = 1 << 14,
750     ARM_HWCAP2_A64_DGH          = 1 << 15,
751     ARM_HWCAP2_A64_RNG          = 1 << 16,
752     ARM_HWCAP2_A64_BTI          = 1 << 17,
753     ARM_HWCAP2_A64_MTE          = 1 << 18,
754     ARM_HWCAP2_A64_ECV          = 1 << 19,
755     ARM_HWCAP2_A64_AFP          = 1 << 20,
756     ARM_HWCAP2_A64_RPRES        = 1 << 21,
757     ARM_HWCAP2_A64_MTE3         = 1 << 22,
758     ARM_HWCAP2_A64_SME          = 1 << 23,
759     ARM_HWCAP2_A64_SME_I16I64   = 1 << 24,
760     ARM_HWCAP2_A64_SME_F64F64   = 1 << 25,
761     ARM_HWCAP2_A64_SME_I8I32    = 1 << 26,
762     ARM_HWCAP2_A64_SME_F16F32   = 1 << 27,
763     ARM_HWCAP2_A64_SME_B16F32   = 1 << 28,
764     ARM_HWCAP2_A64_SME_F32F32   = 1 << 29,
765     ARM_HWCAP2_A64_SME_FA64     = 1 << 30,
766     ARM_HWCAP2_A64_WFXT         = 1ULL << 31,
767     ARM_HWCAP2_A64_EBF16        = 1ULL << 32,
768     ARM_HWCAP2_A64_SVE_EBF16    = 1ULL << 33,
769     ARM_HWCAP2_A64_CSSC         = 1ULL << 34,
770     ARM_HWCAP2_A64_RPRFM        = 1ULL << 35,
771     ARM_HWCAP2_A64_SVE2P1       = 1ULL << 36,
772     ARM_HWCAP2_A64_SME2         = 1ULL << 37,
773     ARM_HWCAP2_A64_SME2P1       = 1ULL << 38,
774     ARM_HWCAP2_A64_SME_I16I32   = 1ULL << 39,
775     ARM_HWCAP2_A64_SME_BI32I32  = 1ULL << 40,
776     ARM_HWCAP2_A64_SME_B16B16   = 1ULL << 41,
777     ARM_HWCAP2_A64_SME_F16F16   = 1ULL << 42,
778     ARM_HWCAP2_A64_MOPS         = 1ULL << 43,
779     ARM_HWCAP2_A64_HBC          = 1ULL << 44,
780 };
781 
782 #define ELF_HWCAP   get_elf_hwcap()
783 #define ELF_HWCAP2  get_elf_hwcap2()
784 
785 #define GET_FEATURE_ID(feat, hwcap) \
786     do { if (cpu_isar_feature(feat, cpu)) { hwcaps |= hwcap; } } while (0)
787 
get_elf_hwcap(void)788 uint32_t get_elf_hwcap(void)
789 {
790     ARMCPU *cpu = ARM_CPU(thread_cpu);
791     uint32_t hwcaps = 0;
792 
793     hwcaps |= ARM_HWCAP_A64_FP;
794     hwcaps |= ARM_HWCAP_A64_ASIMD;
795     hwcaps |= ARM_HWCAP_A64_CPUID;
796 
797     /* probe for the extra features */
798 
799     GET_FEATURE_ID(aa64_aes, ARM_HWCAP_A64_AES);
800     GET_FEATURE_ID(aa64_pmull, ARM_HWCAP_A64_PMULL);
801     GET_FEATURE_ID(aa64_sha1, ARM_HWCAP_A64_SHA1);
802     GET_FEATURE_ID(aa64_sha256, ARM_HWCAP_A64_SHA2);
803     GET_FEATURE_ID(aa64_sha512, ARM_HWCAP_A64_SHA512);
804     GET_FEATURE_ID(aa64_crc32, ARM_HWCAP_A64_CRC32);
805     GET_FEATURE_ID(aa64_sha3, ARM_HWCAP_A64_SHA3);
806     GET_FEATURE_ID(aa64_sm3, ARM_HWCAP_A64_SM3);
807     GET_FEATURE_ID(aa64_sm4, ARM_HWCAP_A64_SM4);
808     GET_FEATURE_ID(aa64_fp16, ARM_HWCAP_A64_FPHP | ARM_HWCAP_A64_ASIMDHP);
809     GET_FEATURE_ID(aa64_atomics, ARM_HWCAP_A64_ATOMICS);
810     GET_FEATURE_ID(aa64_lse2, ARM_HWCAP_A64_USCAT);
811     GET_FEATURE_ID(aa64_rdm, ARM_HWCAP_A64_ASIMDRDM);
812     GET_FEATURE_ID(aa64_dp, ARM_HWCAP_A64_ASIMDDP);
813     GET_FEATURE_ID(aa64_fcma, ARM_HWCAP_A64_FCMA);
814     GET_FEATURE_ID(aa64_sve, ARM_HWCAP_A64_SVE);
815     GET_FEATURE_ID(aa64_pauth, ARM_HWCAP_A64_PACA | ARM_HWCAP_A64_PACG);
816     GET_FEATURE_ID(aa64_fhm, ARM_HWCAP_A64_ASIMDFHM);
817     GET_FEATURE_ID(aa64_dit, ARM_HWCAP_A64_DIT);
818     GET_FEATURE_ID(aa64_jscvt, ARM_HWCAP_A64_JSCVT);
819     GET_FEATURE_ID(aa64_sb, ARM_HWCAP_A64_SB);
820     GET_FEATURE_ID(aa64_condm_4, ARM_HWCAP_A64_FLAGM);
821     GET_FEATURE_ID(aa64_dcpop, ARM_HWCAP_A64_DCPOP);
822     GET_FEATURE_ID(aa64_rcpc_8_3, ARM_HWCAP_A64_LRCPC);
823     GET_FEATURE_ID(aa64_rcpc_8_4, ARM_HWCAP_A64_ILRCPC);
824 
825     return hwcaps;
826 }
827 
get_elf_hwcap2(void)828 uint64_t get_elf_hwcap2(void)
829 {
830     ARMCPU *cpu = ARM_CPU(thread_cpu);
831     uint64_t hwcaps = 0;
832 
833     GET_FEATURE_ID(aa64_dcpodp, ARM_HWCAP2_A64_DCPODP);
834     GET_FEATURE_ID(aa64_sve2, ARM_HWCAP2_A64_SVE2);
835     GET_FEATURE_ID(aa64_sve2_aes, ARM_HWCAP2_A64_SVEAES);
836     GET_FEATURE_ID(aa64_sve2_pmull128, ARM_HWCAP2_A64_SVEPMULL);
837     GET_FEATURE_ID(aa64_sve2_bitperm, ARM_HWCAP2_A64_SVEBITPERM);
838     GET_FEATURE_ID(aa64_sve2_sha3, ARM_HWCAP2_A64_SVESHA3);
839     GET_FEATURE_ID(aa64_sve2_sm4, ARM_HWCAP2_A64_SVESM4);
840     GET_FEATURE_ID(aa64_condm_5, ARM_HWCAP2_A64_FLAGM2);
841     GET_FEATURE_ID(aa64_frint, ARM_HWCAP2_A64_FRINT);
842     GET_FEATURE_ID(aa64_sve_i8mm, ARM_HWCAP2_A64_SVEI8MM);
843     GET_FEATURE_ID(aa64_sve_f32mm, ARM_HWCAP2_A64_SVEF32MM);
844     GET_FEATURE_ID(aa64_sve_f64mm, ARM_HWCAP2_A64_SVEF64MM);
845     GET_FEATURE_ID(aa64_sve_bf16, ARM_HWCAP2_A64_SVEBF16);
846     GET_FEATURE_ID(aa64_i8mm, ARM_HWCAP2_A64_I8MM);
847     GET_FEATURE_ID(aa64_bf16, ARM_HWCAP2_A64_BF16);
848     GET_FEATURE_ID(aa64_rndr, ARM_HWCAP2_A64_RNG);
849     GET_FEATURE_ID(aa64_bti, ARM_HWCAP2_A64_BTI);
850     GET_FEATURE_ID(aa64_mte, ARM_HWCAP2_A64_MTE);
851     GET_FEATURE_ID(aa64_mte3, ARM_HWCAP2_A64_MTE3);
852     GET_FEATURE_ID(aa64_sme, (ARM_HWCAP2_A64_SME |
853                               ARM_HWCAP2_A64_SME_F32F32 |
854                               ARM_HWCAP2_A64_SME_B16F32 |
855                               ARM_HWCAP2_A64_SME_F16F32 |
856                               ARM_HWCAP2_A64_SME_I8I32));
857     GET_FEATURE_ID(aa64_sme_f64f64, ARM_HWCAP2_A64_SME_F64F64);
858     GET_FEATURE_ID(aa64_sme_i16i64, ARM_HWCAP2_A64_SME_I16I64);
859     GET_FEATURE_ID(aa64_sme_fa64, ARM_HWCAP2_A64_SME_FA64);
860     GET_FEATURE_ID(aa64_hbc, ARM_HWCAP2_A64_HBC);
861     GET_FEATURE_ID(aa64_mops, ARM_HWCAP2_A64_MOPS);
862 
863     return hwcaps;
864 }
865 
elf_hwcap_str(uint32_t bit)866 const char *elf_hwcap_str(uint32_t bit)
867 {
868     static const char *hwcap_str[] = {
869     [__builtin_ctz(ARM_HWCAP_A64_FP      )] = "fp",
870     [__builtin_ctz(ARM_HWCAP_A64_ASIMD   )] = "asimd",
871     [__builtin_ctz(ARM_HWCAP_A64_EVTSTRM )] = "evtstrm",
872     [__builtin_ctz(ARM_HWCAP_A64_AES     )] = "aes",
873     [__builtin_ctz(ARM_HWCAP_A64_PMULL   )] = "pmull",
874     [__builtin_ctz(ARM_HWCAP_A64_SHA1    )] = "sha1",
875     [__builtin_ctz(ARM_HWCAP_A64_SHA2    )] = "sha2",
876     [__builtin_ctz(ARM_HWCAP_A64_CRC32   )] = "crc32",
877     [__builtin_ctz(ARM_HWCAP_A64_ATOMICS )] = "atomics",
878     [__builtin_ctz(ARM_HWCAP_A64_FPHP    )] = "fphp",
879     [__builtin_ctz(ARM_HWCAP_A64_ASIMDHP )] = "asimdhp",
880     [__builtin_ctz(ARM_HWCAP_A64_CPUID   )] = "cpuid",
881     [__builtin_ctz(ARM_HWCAP_A64_ASIMDRDM)] = "asimdrdm",
882     [__builtin_ctz(ARM_HWCAP_A64_JSCVT   )] = "jscvt",
883     [__builtin_ctz(ARM_HWCAP_A64_FCMA    )] = "fcma",
884     [__builtin_ctz(ARM_HWCAP_A64_LRCPC   )] = "lrcpc",
885     [__builtin_ctz(ARM_HWCAP_A64_DCPOP   )] = "dcpop",
886     [__builtin_ctz(ARM_HWCAP_A64_SHA3    )] = "sha3",
887     [__builtin_ctz(ARM_HWCAP_A64_SM3     )] = "sm3",
888     [__builtin_ctz(ARM_HWCAP_A64_SM4     )] = "sm4",
889     [__builtin_ctz(ARM_HWCAP_A64_ASIMDDP )] = "asimddp",
890     [__builtin_ctz(ARM_HWCAP_A64_SHA512  )] = "sha512",
891     [__builtin_ctz(ARM_HWCAP_A64_SVE     )] = "sve",
892     [__builtin_ctz(ARM_HWCAP_A64_ASIMDFHM)] = "asimdfhm",
893     [__builtin_ctz(ARM_HWCAP_A64_DIT     )] = "dit",
894     [__builtin_ctz(ARM_HWCAP_A64_USCAT   )] = "uscat",
895     [__builtin_ctz(ARM_HWCAP_A64_ILRCPC  )] = "ilrcpc",
896     [__builtin_ctz(ARM_HWCAP_A64_FLAGM   )] = "flagm",
897     [__builtin_ctz(ARM_HWCAP_A64_SSBS    )] = "ssbs",
898     [__builtin_ctz(ARM_HWCAP_A64_SB      )] = "sb",
899     [__builtin_ctz(ARM_HWCAP_A64_PACA    )] = "paca",
900     [__builtin_ctz(ARM_HWCAP_A64_PACG    )] = "pacg",
901     };
902 
903     return bit < ARRAY_SIZE(hwcap_str) ? hwcap_str[bit] : NULL;
904 }
905 
elf_hwcap2_str(uint32_t bit)906 const char *elf_hwcap2_str(uint32_t bit)
907 {
908     static const char *hwcap_str[] = {
909     [__builtin_ctz(ARM_HWCAP2_A64_DCPODP       )] = "dcpodp",
910     [__builtin_ctz(ARM_HWCAP2_A64_SVE2         )] = "sve2",
911     [__builtin_ctz(ARM_HWCAP2_A64_SVEAES       )] = "sveaes",
912     [__builtin_ctz(ARM_HWCAP2_A64_SVEPMULL     )] = "svepmull",
913     [__builtin_ctz(ARM_HWCAP2_A64_SVEBITPERM   )] = "svebitperm",
914     [__builtin_ctz(ARM_HWCAP2_A64_SVESHA3      )] = "svesha3",
915     [__builtin_ctz(ARM_HWCAP2_A64_SVESM4       )] = "svesm4",
916     [__builtin_ctz(ARM_HWCAP2_A64_FLAGM2       )] = "flagm2",
917     [__builtin_ctz(ARM_HWCAP2_A64_FRINT        )] = "frint",
918     [__builtin_ctz(ARM_HWCAP2_A64_SVEI8MM      )] = "svei8mm",
919     [__builtin_ctz(ARM_HWCAP2_A64_SVEF32MM     )] = "svef32mm",
920     [__builtin_ctz(ARM_HWCAP2_A64_SVEF64MM     )] = "svef64mm",
921     [__builtin_ctz(ARM_HWCAP2_A64_SVEBF16      )] = "svebf16",
922     [__builtin_ctz(ARM_HWCAP2_A64_I8MM         )] = "i8mm",
923     [__builtin_ctz(ARM_HWCAP2_A64_BF16         )] = "bf16",
924     [__builtin_ctz(ARM_HWCAP2_A64_DGH          )] = "dgh",
925     [__builtin_ctz(ARM_HWCAP2_A64_RNG          )] = "rng",
926     [__builtin_ctz(ARM_HWCAP2_A64_BTI          )] = "bti",
927     [__builtin_ctz(ARM_HWCAP2_A64_MTE          )] = "mte",
928     [__builtin_ctz(ARM_HWCAP2_A64_ECV          )] = "ecv",
929     [__builtin_ctz(ARM_HWCAP2_A64_AFP          )] = "afp",
930     [__builtin_ctz(ARM_HWCAP2_A64_RPRES        )] = "rpres",
931     [__builtin_ctz(ARM_HWCAP2_A64_MTE3         )] = "mte3",
932     [__builtin_ctz(ARM_HWCAP2_A64_SME          )] = "sme",
933     [__builtin_ctz(ARM_HWCAP2_A64_SME_I16I64   )] = "smei16i64",
934     [__builtin_ctz(ARM_HWCAP2_A64_SME_F64F64   )] = "smef64f64",
935     [__builtin_ctz(ARM_HWCAP2_A64_SME_I8I32    )] = "smei8i32",
936     [__builtin_ctz(ARM_HWCAP2_A64_SME_F16F32   )] = "smef16f32",
937     [__builtin_ctz(ARM_HWCAP2_A64_SME_B16F32   )] = "smeb16f32",
938     [__builtin_ctz(ARM_HWCAP2_A64_SME_F32F32   )] = "smef32f32",
939     [__builtin_ctz(ARM_HWCAP2_A64_SME_FA64     )] = "smefa64",
940     [__builtin_ctz(ARM_HWCAP2_A64_WFXT         )] = "wfxt",
941     [__builtin_ctzll(ARM_HWCAP2_A64_EBF16      )] = "ebf16",
942     [__builtin_ctzll(ARM_HWCAP2_A64_SVE_EBF16  )] = "sveebf16",
943     [__builtin_ctzll(ARM_HWCAP2_A64_CSSC       )] = "cssc",
944     [__builtin_ctzll(ARM_HWCAP2_A64_RPRFM      )] = "rprfm",
945     [__builtin_ctzll(ARM_HWCAP2_A64_SVE2P1     )] = "sve2p1",
946     [__builtin_ctzll(ARM_HWCAP2_A64_SME2       )] = "sme2",
947     [__builtin_ctzll(ARM_HWCAP2_A64_SME2P1     )] = "sme2p1",
948     [__builtin_ctzll(ARM_HWCAP2_A64_SME_I16I32 )] = "smei16i32",
949     [__builtin_ctzll(ARM_HWCAP2_A64_SME_BI32I32)] = "smebi32i32",
950     [__builtin_ctzll(ARM_HWCAP2_A64_SME_B16B16 )] = "smeb16b16",
951     [__builtin_ctzll(ARM_HWCAP2_A64_SME_F16F16 )] = "smef16f16",
952     [__builtin_ctzll(ARM_HWCAP2_A64_MOPS       )] = "mops",
953     [__builtin_ctzll(ARM_HWCAP2_A64_HBC        )] = "hbc",
954     };
955 
956     return bit < ARRAY_SIZE(hwcap_str) ? hwcap_str[bit] : NULL;
957 }
958 
959 #undef GET_FEATURE_ID
960 
961 #endif /* not TARGET_AARCH64 */
962 
963 #if TARGET_BIG_ENDIAN
964 # define VDSO_HEADER  "vdso-be.c.inc"
965 #else
966 # define VDSO_HEADER  "vdso-le.c.inc"
967 #endif
968 
969 #endif /* TARGET_ARM */
970 
971 #ifdef TARGET_SPARC
972 
973 #ifndef TARGET_SPARC64
974 # define ELF_CLASS  ELFCLASS32
975 # define ELF_ARCH   EM_SPARC
976 #elif defined(TARGET_ABI32)
977 # define ELF_CLASS  ELFCLASS32
978 # define elf_check_arch(x) ((x) == EM_SPARC32PLUS || (x) == EM_SPARC)
979 #else
980 # define ELF_CLASS  ELFCLASS64
981 # define ELF_ARCH   EM_SPARCV9
982 #endif
983 
984 #include "elf.h"
985 
986 #define ELF_HWCAP get_elf_hwcap()
987 
get_elf_hwcap(void)988 static uint32_t get_elf_hwcap(void)
989 {
990     /* There are not many sparc32 hwcap bits -- we have all of them. */
991     uint32_t r = HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR |
992                  HWCAP_SPARC_SWAP | HWCAP_SPARC_MULDIV;
993 
994 #ifdef TARGET_SPARC64
995     CPUSPARCState *env = cpu_env(thread_cpu);
996     uint32_t features = env->def.features;
997 
998     r |= HWCAP_SPARC_V9 | HWCAP_SPARC_V8PLUS;
999     /* 32x32 multiply and divide are efficient. */
1000     r |= HWCAP_SPARC_MUL32 | HWCAP_SPARC_DIV32;
1001     /* We don't have an internal feature bit for this. */
1002     r |= HWCAP_SPARC_POPC;
1003     r |= features & CPU_FEATURE_FSMULD ? HWCAP_SPARC_FSMULD : 0;
1004     r |= features & CPU_FEATURE_VIS1 ? HWCAP_SPARC_VIS : 0;
1005     r |= features & CPU_FEATURE_VIS2 ? HWCAP_SPARC_VIS2 : 0;
1006 #endif
1007 
1008     return r;
1009 }
1010 
init_thread(struct target_pt_regs * regs,struct image_info * infop)1011 static inline void init_thread(struct target_pt_regs *regs,
1012                                struct image_info *infop)
1013 {
1014     /* Note that target_cpu_copy_regs does not read psr/tstate. */
1015     regs->pc = infop->entry;
1016     regs->npc = regs->pc + 4;
1017     regs->y = 0;
1018     regs->u_regs[14] = (infop->start_stack - 16 * sizeof(abi_ulong)
1019                         - TARGET_STACK_BIAS);
1020 }
1021 #endif /* TARGET_SPARC */
1022 
1023 #ifdef TARGET_PPC
1024 
1025 #define ELF_MACHINE    PPC_ELF_MACHINE
1026 
1027 #if defined(TARGET_PPC64)
1028 
1029 #define elf_check_arch(x) ( (x) == EM_PPC64 )
1030 
1031 #define ELF_CLASS       ELFCLASS64
1032 
1033 #else
1034 
1035 #define ELF_CLASS       ELFCLASS32
1036 #define EXSTACK_DEFAULT true
1037 
1038 #endif
1039 
1040 #define ELF_ARCH        EM_PPC
1041 
1042 /* Feature masks for the Aux Vector Hardware Capabilities (AT_HWCAP).
1043    See arch/powerpc/include/asm/cputable.h.  */
1044 enum {
1045     QEMU_PPC_FEATURE_32 = 0x80000000,
1046     QEMU_PPC_FEATURE_64 = 0x40000000,
1047     QEMU_PPC_FEATURE_601_INSTR = 0x20000000,
1048     QEMU_PPC_FEATURE_HAS_ALTIVEC = 0x10000000,
1049     QEMU_PPC_FEATURE_HAS_FPU = 0x08000000,
1050     QEMU_PPC_FEATURE_HAS_MMU = 0x04000000,
1051     QEMU_PPC_FEATURE_HAS_4xxMAC = 0x02000000,
1052     QEMU_PPC_FEATURE_UNIFIED_CACHE = 0x01000000,
1053     QEMU_PPC_FEATURE_HAS_SPE = 0x00800000,
1054     QEMU_PPC_FEATURE_HAS_EFP_SINGLE = 0x00400000,
1055     QEMU_PPC_FEATURE_HAS_EFP_DOUBLE = 0x00200000,
1056     QEMU_PPC_FEATURE_NO_TB = 0x00100000,
1057     QEMU_PPC_FEATURE_POWER4 = 0x00080000,
1058     QEMU_PPC_FEATURE_POWER5 = 0x00040000,
1059     QEMU_PPC_FEATURE_POWER5_PLUS = 0x00020000,
1060     QEMU_PPC_FEATURE_CELL = 0x00010000,
1061     QEMU_PPC_FEATURE_BOOKE = 0x00008000,
1062     QEMU_PPC_FEATURE_SMT = 0x00004000,
1063     QEMU_PPC_FEATURE_ICACHE_SNOOP = 0x00002000,
1064     QEMU_PPC_FEATURE_ARCH_2_05 = 0x00001000,
1065     QEMU_PPC_FEATURE_PA6T = 0x00000800,
1066     QEMU_PPC_FEATURE_HAS_DFP = 0x00000400,
1067     QEMU_PPC_FEATURE_POWER6_EXT = 0x00000200,
1068     QEMU_PPC_FEATURE_ARCH_2_06 = 0x00000100,
1069     QEMU_PPC_FEATURE_HAS_VSX = 0x00000080,
1070     QEMU_PPC_FEATURE_PSERIES_PERFMON_COMPAT = 0x00000040,
1071 
1072     QEMU_PPC_FEATURE_TRUE_LE = 0x00000002,
1073     QEMU_PPC_FEATURE_PPC_LE = 0x00000001,
1074 
1075     /* Feature definitions in AT_HWCAP2.  */
1076     QEMU_PPC_FEATURE2_ARCH_2_07 = 0x80000000, /* ISA 2.07 */
1077     QEMU_PPC_FEATURE2_HAS_HTM = 0x40000000, /* Hardware Transactional Memory */
1078     QEMU_PPC_FEATURE2_HAS_DSCR = 0x20000000, /* Data Stream Control Register */
1079     QEMU_PPC_FEATURE2_HAS_EBB = 0x10000000, /* Event Base Branching */
1080     QEMU_PPC_FEATURE2_HAS_ISEL = 0x08000000, /* Integer Select */
1081     QEMU_PPC_FEATURE2_HAS_TAR = 0x04000000, /* Target Address Register */
1082     QEMU_PPC_FEATURE2_VEC_CRYPTO = 0x02000000,
1083     QEMU_PPC_FEATURE2_HTM_NOSC = 0x01000000,
1084     QEMU_PPC_FEATURE2_ARCH_3_00 = 0x00800000, /* ISA 3.00 */
1085     QEMU_PPC_FEATURE2_HAS_IEEE128 = 0x00400000, /* VSX IEEE Bin Float 128-bit */
1086     QEMU_PPC_FEATURE2_DARN = 0x00200000, /* darn random number insn */
1087     QEMU_PPC_FEATURE2_SCV = 0x00100000, /* scv syscall */
1088     QEMU_PPC_FEATURE2_HTM_NO_SUSPEND = 0x00080000, /* TM w/o suspended state */
1089     QEMU_PPC_FEATURE2_ARCH_3_1 = 0x00040000, /* ISA 3.1 */
1090     QEMU_PPC_FEATURE2_MMA = 0x00020000, /* Matrix-Multiply Assist */
1091 };
1092 
1093 #define ELF_HWCAP get_elf_hwcap()
1094 
get_elf_hwcap(void)1095 static uint32_t get_elf_hwcap(void)
1096 {
1097     PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);
1098     uint32_t features = 0;
1099 
1100     /* We don't have to be terribly complete here; the high points are
1101        Altivec/FP/SPE support.  Anything else is just a bonus.  */
1102 #define GET_FEATURE(flag, feature)                                      \
1103     do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0)
1104 #define GET_FEATURE2(flags, feature) \
1105     do { \
1106         if ((cpu->env.insns_flags2 & flags) == flags) { \
1107             features |= feature; \
1108         } \
1109     } while (0)
1110     GET_FEATURE(PPC_64B, QEMU_PPC_FEATURE_64);
1111     GET_FEATURE(PPC_FLOAT, QEMU_PPC_FEATURE_HAS_FPU);
1112     GET_FEATURE(PPC_ALTIVEC, QEMU_PPC_FEATURE_HAS_ALTIVEC);
1113     GET_FEATURE(PPC_SPE, QEMU_PPC_FEATURE_HAS_SPE);
1114     GET_FEATURE(PPC_SPE_SINGLE, QEMU_PPC_FEATURE_HAS_EFP_SINGLE);
1115     GET_FEATURE(PPC_SPE_DOUBLE, QEMU_PPC_FEATURE_HAS_EFP_DOUBLE);
1116     GET_FEATURE(PPC_BOOKE, QEMU_PPC_FEATURE_BOOKE);
1117     GET_FEATURE(PPC_405_MAC, QEMU_PPC_FEATURE_HAS_4xxMAC);
1118     GET_FEATURE2(PPC2_DFP, QEMU_PPC_FEATURE_HAS_DFP);
1119     GET_FEATURE2(PPC2_VSX, QEMU_PPC_FEATURE_HAS_VSX);
1120     GET_FEATURE2((PPC2_PERM_ISA206 | PPC2_DIVE_ISA206 | PPC2_ATOMIC_ISA206 |
1121                   PPC2_FP_CVT_ISA206 | PPC2_FP_TST_ISA206),
1122                   QEMU_PPC_FEATURE_ARCH_2_06);
1123 #undef GET_FEATURE
1124 #undef GET_FEATURE2
1125 
1126     return features;
1127 }
1128 
1129 #define ELF_HWCAP2 get_elf_hwcap2()
1130 
get_elf_hwcap2(void)1131 static uint32_t get_elf_hwcap2(void)
1132 {
1133     PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);
1134     uint32_t features = 0;
1135 
1136 #define GET_FEATURE(flag, feature)                                      \
1137     do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0)
1138 #define GET_FEATURE2(flag, feature)                                      \
1139     do { if (cpu->env.insns_flags2 & flag) { features |= feature; } } while (0)
1140 
1141     GET_FEATURE(PPC_ISEL, QEMU_PPC_FEATURE2_HAS_ISEL);
1142     GET_FEATURE2(PPC2_BCTAR_ISA207, QEMU_PPC_FEATURE2_HAS_TAR);
1143     GET_FEATURE2((PPC2_BCTAR_ISA207 | PPC2_LSQ_ISA207 | PPC2_ALTIVEC_207 |
1144                   PPC2_ISA207S), QEMU_PPC_FEATURE2_ARCH_2_07 |
1145                   QEMU_PPC_FEATURE2_VEC_CRYPTO);
1146     GET_FEATURE2(PPC2_ISA300, QEMU_PPC_FEATURE2_ARCH_3_00 |
1147                  QEMU_PPC_FEATURE2_DARN | QEMU_PPC_FEATURE2_HAS_IEEE128);
1148     GET_FEATURE2(PPC2_ISA310, QEMU_PPC_FEATURE2_ARCH_3_1 |
1149                  QEMU_PPC_FEATURE2_MMA);
1150 
1151 #undef GET_FEATURE
1152 #undef GET_FEATURE2
1153 
1154     return features;
1155 }
1156 
1157 /*
1158  * The requirements here are:
1159  * - keep the final alignment of sp (sp & 0xf)
1160  * - make sure the 32-bit value at the first 16 byte aligned position of
1161  *   AUXV is greater than 16 for glibc compatibility.
1162  *   AT_IGNOREPPC is used for that.
1163  * - for compatibility with glibc ARCH_DLINFO must always be defined on PPC,
1164  *   even if DLINFO_ARCH_ITEMS goes to zero or is undefined.
1165  */
1166 #define DLINFO_ARCH_ITEMS       5
1167 #define ARCH_DLINFO                                     \
1168     do {                                                \
1169         PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);              \
1170         /*                                              \
1171          * Handle glibc compatibility: these magic entries must \
1172          * be at the lowest addresses in the final auxv.        \
1173          */                                             \
1174         NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC);        \
1175         NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC);        \
1176         NEW_AUX_ENT(AT_DCACHEBSIZE, cpu->env.dcache_line_size); \
1177         NEW_AUX_ENT(AT_ICACHEBSIZE, cpu->env.icache_line_size); \
1178         NEW_AUX_ENT(AT_UCACHEBSIZE, 0);                 \
1179     } while (0)
1180 
init_thread(struct target_pt_regs * _regs,struct image_info * infop)1181 static inline void init_thread(struct target_pt_regs *_regs, struct image_info *infop)
1182 {
1183     _regs->gpr[1] = infop->start_stack;
1184 #if defined(TARGET_PPC64)
1185     if (get_ppc64_abi(infop) < 2) {
1186         uint64_t val;
1187         get_user_u64(val, infop->entry + 8);
1188         _regs->gpr[2] = val + infop->load_bias;
1189         get_user_u64(val, infop->entry);
1190         infop->entry = val + infop->load_bias;
1191     } else {
1192         _regs->gpr[12] = infop->entry;  /* r12 set to global entry address */
1193     }
1194 #endif
1195     _regs->nip = infop->entry;
1196 }
1197 
1198 /* See linux kernel: arch/powerpc/include/asm/elf.h.  */
1199 #define ELF_NREG 48
1200 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1201 
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUPPCState * env)1202 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUPPCState *env)
1203 {
1204     int i;
1205     target_ulong ccr = 0;
1206 
1207     for (i = 0; i < ARRAY_SIZE(env->gpr); i++) {
1208         (*regs)[i] = tswapreg(env->gpr[i]);
1209     }
1210 
1211     (*regs)[32] = tswapreg(env->nip);
1212     (*regs)[33] = tswapreg(env->msr);
1213     (*regs)[35] = tswapreg(env->ctr);
1214     (*regs)[36] = tswapreg(env->lr);
1215     (*regs)[37] = tswapreg(cpu_read_xer(env));
1216 
1217     ccr = ppc_get_cr(env);
1218     (*regs)[38] = tswapreg(ccr);
1219 }
1220 
1221 #define USE_ELF_CORE_DUMP
1222 #define ELF_EXEC_PAGESIZE       4096
1223 
1224 #ifndef TARGET_PPC64
1225 # define VDSO_HEADER  "vdso-32.c.inc"
1226 #elif TARGET_BIG_ENDIAN
1227 # define VDSO_HEADER  "vdso-64.c.inc"
1228 #else
1229 # define VDSO_HEADER  "vdso-64le.c.inc"
1230 #endif
1231 
1232 #endif
1233 
1234 #ifdef TARGET_LOONGARCH64
1235 
1236 #define ELF_CLASS   ELFCLASS64
1237 #define ELF_ARCH    EM_LOONGARCH
1238 #define EXSTACK_DEFAULT true
1239 
1240 #define elf_check_arch(x) ((x) == EM_LOONGARCH)
1241 
1242 #define VDSO_HEADER "vdso.c.inc"
1243 
init_thread(struct target_pt_regs * regs,struct image_info * infop)1244 static inline void init_thread(struct target_pt_regs *regs,
1245                                struct image_info *infop)
1246 {
1247     /*Set crmd PG,DA = 1,0 */
1248     regs->csr.crmd = 2 << 3;
1249     regs->csr.era = infop->entry;
1250     regs->regs[3] = infop->start_stack;
1251 }
1252 
1253 /* See linux kernel: arch/loongarch/include/asm/elf.h */
1254 #define ELF_NREG 45
1255 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1256 
1257 enum {
1258     TARGET_EF_R0 = 0,
1259     TARGET_EF_CSR_ERA = TARGET_EF_R0 + 33,
1260     TARGET_EF_CSR_BADV = TARGET_EF_R0 + 34,
1261 };
1262 
elf_core_copy_regs(target_elf_gregset_t * regs,const CPULoongArchState * env)1263 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1264                                const CPULoongArchState *env)
1265 {
1266     int i;
1267 
1268     (*regs)[TARGET_EF_R0] = 0;
1269 
1270     for (i = 1; i < ARRAY_SIZE(env->gpr); i++) {
1271         (*regs)[TARGET_EF_R0 + i] = tswapreg(env->gpr[i]);
1272     }
1273 
1274     (*regs)[TARGET_EF_CSR_ERA] = tswapreg(env->pc);
1275     (*regs)[TARGET_EF_CSR_BADV] = tswapreg(env->CSR_BADV);
1276 }
1277 
1278 #define USE_ELF_CORE_DUMP
1279 #define ELF_EXEC_PAGESIZE        4096
1280 
1281 #define ELF_HWCAP get_elf_hwcap()
1282 
1283 /* See arch/loongarch/include/uapi/asm/hwcap.h */
1284 enum {
1285     HWCAP_LOONGARCH_CPUCFG   = (1 << 0),
1286     HWCAP_LOONGARCH_LAM      = (1 << 1),
1287     HWCAP_LOONGARCH_UAL      = (1 << 2),
1288     HWCAP_LOONGARCH_FPU      = (1 << 3),
1289     HWCAP_LOONGARCH_LSX      = (1 << 4),
1290     HWCAP_LOONGARCH_LASX     = (1 << 5),
1291     HWCAP_LOONGARCH_CRC32    = (1 << 6),
1292     HWCAP_LOONGARCH_COMPLEX  = (1 << 7),
1293     HWCAP_LOONGARCH_CRYPTO   = (1 << 8),
1294     HWCAP_LOONGARCH_LVZ      = (1 << 9),
1295     HWCAP_LOONGARCH_LBT_X86  = (1 << 10),
1296     HWCAP_LOONGARCH_LBT_ARM  = (1 << 11),
1297     HWCAP_LOONGARCH_LBT_MIPS = (1 << 12),
1298 };
1299 
get_elf_hwcap(void)1300 static uint32_t get_elf_hwcap(void)
1301 {
1302     LoongArchCPU *cpu = LOONGARCH_CPU(thread_cpu);
1303     uint32_t hwcaps = 0;
1304 
1305     hwcaps |= HWCAP_LOONGARCH_CRC32;
1306 
1307     if (FIELD_EX32(cpu->env.cpucfg[1], CPUCFG1, UAL)) {
1308         hwcaps |= HWCAP_LOONGARCH_UAL;
1309     }
1310 
1311     if (FIELD_EX32(cpu->env.cpucfg[2], CPUCFG2, FP)) {
1312         hwcaps |= HWCAP_LOONGARCH_FPU;
1313     }
1314 
1315     if (FIELD_EX32(cpu->env.cpucfg[2], CPUCFG2, LAM)) {
1316         hwcaps |= HWCAP_LOONGARCH_LAM;
1317     }
1318 
1319     if (FIELD_EX32(cpu->env.cpucfg[2], CPUCFG2, LSX)) {
1320         hwcaps |= HWCAP_LOONGARCH_LSX;
1321     }
1322 
1323     if (FIELD_EX32(cpu->env.cpucfg[2], CPUCFG2, LASX)) {
1324         hwcaps |= HWCAP_LOONGARCH_LASX;
1325     }
1326 
1327     return hwcaps;
1328 }
1329 
1330 #define ELF_PLATFORM "loongarch"
1331 
1332 #endif /* TARGET_LOONGARCH64 */
1333 
1334 #ifdef TARGET_MIPS
1335 
1336 #ifdef TARGET_MIPS64
1337 #define ELF_CLASS   ELFCLASS64
1338 #else
1339 #define ELF_CLASS   ELFCLASS32
1340 #endif
1341 #define ELF_ARCH    EM_MIPS
1342 #define EXSTACK_DEFAULT true
1343 
1344 #ifdef TARGET_ABI_MIPSN32
1345 #define elf_check_abi(x) ((x) & EF_MIPS_ABI2)
1346 #else
1347 #define elf_check_abi(x) (!((x) & EF_MIPS_ABI2))
1348 #endif
1349 
1350 #define ELF_BASE_PLATFORM get_elf_base_platform()
1351 
1352 #define MATCH_PLATFORM_INSN(_flags, _base_platform)      \
1353     do { if ((cpu->env.insn_flags & (_flags)) == _flags) \
1354     { return _base_platform; } } while (0)
1355 
get_elf_base_platform(void)1356 static const char *get_elf_base_platform(void)
1357 {
1358     MIPSCPU *cpu = MIPS_CPU(thread_cpu);
1359 
1360     /* 64 bit ISAs goes first */
1361     MATCH_PLATFORM_INSN(CPU_MIPS64R6, "mips64r6");
1362     MATCH_PLATFORM_INSN(CPU_MIPS64R5, "mips64r5");
1363     MATCH_PLATFORM_INSN(CPU_MIPS64R2, "mips64r2");
1364     MATCH_PLATFORM_INSN(CPU_MIPS64R1, "mips64");
1365     MATCH_PLATFORM_INSN(CPU_MIPS5, "mips5");
1366     MATCH_PLATFORM_INSN(CPU_MIPS4, "mips4");
1367     MATCH_PLATFORM_INSN(CPU_MIPS3, "mips3");
1368 
1369     /* 32 bit ISAs */
1370     MATCH_PLATFORM_INSN(CPU_MIPS32R6, "mips32r6");
1371     MATCH_PLATFORM_INSN(CPU_MIPS32R5, "mips32r5");
1372     MATCH_PLATFORM_INSN(CPU_MIPS32R2, "mips32r2");
1373     MATCH_PLATFORM_INSN(CPU_MIPS32R1, "mips32");
1374     MATCH_PLATFORM_INSN(CPU_MIPS2, "mips2");
1375 
1376     /* Fallback */
1377     return "mips";
1378 }
1379 #undef MATCH_PLATFORM_INSN
1380 
init_thread(struct target_pt_regs * regs,struct image_info * infop)1381 static inline void init_thread(struct target_pt_regs *regs,
1382                                struct image_info *infop)
1383 {
1384     regs->cp0_status = 2 << CP0St_KSU;
1385     regs->cp0_epc = infop->entry;
1386     regs->regs[29] = infop->start_stack;
1387 }
1388 
1389 /* See linux kernel: arch/mips/include/asm/elf.h.  */
1390 #define ELF_NREG 45
1391 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1392 
1393 /* See linux kernel: arch/mips/include/asm/reg.h.  */
1394 enum {
1395 #ifdef TARGET_MIPS64
1396     TARGET_EF_R0 = 0,
1397 #else
1398     TARGET_EF_R0 = 6,
1399 #endif
1400     TARGET_EF_R26 = TARGET_EF_R0 + 26,
1401     TARGET_EF_R27 = TARGET_EF_R0 + 27,
1402     TARGET_EF_LO = TARGET_EF_R0 + 32,
1403     TARGET_EF_HI = TARGET_EF_R0 + 33,
1404     TARGET_EF_CP0_EPC = TARGET_EF_R0 + 34,
1405     TARGET_EF_CP0_BADVADDR = TARGET_EF_R0 + 35,
1406     TARGET_EF_CP0_STATUS = TARGET_EF_R0 + 36,
1407     TARGET_EF_CP0_CAUSE = TARGET_EF_R0 + 37
1408 };
1409 
1410 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs.  */
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUMIPSState * env)1411 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMIPSState *env)
1412 {
1413     int i;
1414 
1415     for (i = 0; i < TARGET_EF_R0; i++) {
1416         (*regs)[i] = 0;
1417     }
1418     (*regs)[TARGET_EF_R0] = 0;
1419 
1420     for (i = 1; i < ARRAY_SIZE(env->active_tc.gpr); i++) {
1421         (*regs)[TARGET_EF_R0 + i] = tswapreg(env->active_tc.gpr[i]);
1422     }
1423 
1424     (*regs)[TARGET_EF_R26] = 0;
1425     (*regs)[TARGET_EF_R27] = 0;
1426     (*regs)[TARGET_EF_LO] = tswapreg(env->active_tc.LO[0]);
1427     (*regs)[TARGET_EF_HI] = tswapreg(env->active_tc.HI[0]);
1428     (*regs)[TARGET_EF_CP0_EPC] = tswapreg(env->active_tc.PC);
1429     (*regs)[TARGET_EF_CP0_BADVADDR] = tswapreg(env->CP0_BadVAddr);
1430     (*regs)[TARGET_EF_CP0_STATUS] = tswapreg(env->CP0_Status);
1431     (*regs)[TARGET_EF_CP0_CAUSE] = tswapreg(env->CP0_Cause);
1432 }
1433 
1434 #define USE_ELF_CORE_DUMP
1435 #define ELF_EXEC_PAGESIZE        4096
1436 
1437 /* See arch/mips/include/uapi/asm/hwcap.h.  */
1438 enum {
1439     HWCAP_MIPS_R6           = (1 << 0),
1440     HWCAP_MIPS_MSA          = (1 << 1),
1441     HWCAP_MIPS_CRC32        = (1 << 2),
1442     HWCAP_MIPS_MIPS16       = (1 << 3),
1443     HWCAP_MIPS_MDMX         = (1 << 4),
1444     HWCAP_MIPS_MIPS3D       = (1 << 5),
1445     HWCAP_MIPS_SMARTMIPS    = (1 << 6),
1446     HWCAP_MIPS_DSP          = (1 << 7),
1447     HWCAP_MIPS_DSP2         = (1 << 8),
1448     HWCAP_MIPS_DSP3         = (1 << 9),
1449     HWCAP_MIPS_MIPS16E2     = (1 << 10),
1450     HWCAP_LOONGSON_MMI      = (1 << 11),
1451     HWCAP_LOONGSON_EXT      = (1 << 12),
1452     HWCAP_LOONGSON_EXT2     = (1 << 13),
1453     HWCAP_LOONGSON_CPUCFG   = (1 << 14),
1454 };
1455 
1456 #define ELF_HWCAP get_elf_hwcap()
1457 
1458 #define GET_FEATURE_INSN(_flag, _hwcap) \
1459     do { if (cpu->env.insn_flags & (_flag)) { hwcaps |= _hwcap; } } while (0)
1460 
1461 #define GET_FEATURE_REG_SET(_reg, _mask, _hwcap) \
1462     do { if (cpu->env._reg & (_mask)) { hwcaps |= _hwcap; } } while (0)
1463 
1464 #define GET_FEATURE_REG_EQU(_reg, _start, _length, _val, _hwcap) \
1465     do { \
1466         if (extract32(cpu->env._reg, (_start), (_length)) == (_val)) { \
1467             hwcaps |= _hwcap; \
1468         } \
1469     } while (0)
1470 
get_elf_hwcap(void)1471 static uint32_t get_elf_hwcap(void)
1472 {
1473     MIPSCPU *cpu = MIPS_CPU(thread_cpu);
1474     uint32_t hwcaps = 0;
1475 
1476     GET_FEATURE_REG_EQU(CP0_Config0, CP0C0_AR, CP0C0_AR_LENGTH,
1477                         2, HWCAP_MIPS_R6);
1478     GET_FEATURE_REG_SET(CP0_Config3, 1 << CP0C3_MSAP, HWCAP_MIPS_MSA);
1479     GET_FEATURE_INSN(ASE_LMMI, HWCAP_LOONGSON_MMI);
1480     GET_FEATURE_INSN(ASE_LEXT, HWCAP_LOONGSON_EXT);
1481 
1482     return hwcaps;
1483 }
1484 
1485 #undef GET_FEATURE_REG_EQU
1486 #undef GET_FEATURE_REG_SET
1487 #undef GET_FEATURE_INSN
1488 
1489 #endif /* TARGET_MIPS */
1490 
1491 #ifdef TARGET_MICROBLAZE
1492 
1493 #define elf_check_arch(x) ( (x) == EM_MICROBLAZE || (x) == EM_MICROBLAZE_OLD)
1494 
1495 #define ELF_CLASS   ELFCLASS32
1496 #define ELF_ARCH    EM_MICROBLAZE
1497 
init_thread(struct target_pt_regs * regs,struct image_info * infop)1498 static inline void init_thread(struct target_pt_regs *regs,
1499                                struct image_info *infop)
1500 {
1501     regs->pc = infop->entry;
1502     regs->r1 = infop->start_stack;
1503 
1504 }
1505 
1506 #define ELF_EXEC_PAGESIZE        4096
1507 
1508 #define USE_ELF_CORE_DUMP
1509 #define ELF_NREG 38
1510 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1511 
1512 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs.  */
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUMBState * env)1513 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMBState *env)
1514 {
1515     int i, pos = 0;
1516 
1517     for (i = 0; i < 32; i++) {
1518         (*regs)[pos++] = tswapreg(env->regs[i]);
1519     }
1520 
1521     (*regs)[pos++] = tswapreg(env->pc);
1522     (*regs)[pos++] = tswapreg(mb_cpu_read_msr(env));
1523     (*regs)[pos++] = 0;
1524     (*regs)[pos++] = tswapreg(env->ear);
1525     (*regs)[pos++] = 0;
1526     (*regs)[pos++] = tswapreg(env->esr);
1527 }
1528 
1529 #endif /* TARGET_MICROBLAZE */
1530 
1531 #ifdef TARGET_OPENRISC
1532 
1533 #define ELF_ARCH EM_OPENRISC
1534 #define ELF_CLASS ELFCLASS32
1535 #define ELF_DATA  ELFDATA2MSB
1536 
init_thread(struct target_pt_regs * regs,struct image_info * infop)1537 static inline void init_thread(struct target_pt_regs *regs,
1538                                struct image_info *infop)
1539 {
1540     regs->pc = infop->entry;
1541     regs->gpr[1] = infop->start_stack;
1542 }
1543 
1544 #define USE_ELF_CORE_DUMP
1545 #define ELF_EXEC_PAGESIZE 8192
1546 
1547 /* See linux kernel arch/openrisc/include/asm/elf.h.  */
1548 #define ELF_NREG 34 /* gprs and pc, sr */
1549 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1550 
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUOpenRISCState * env)1551 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1552                                const CPUOpenRISCState *env)
1553 {
1554     int i;
1555 
1556     for (i = 0; i < 32; i++) {
1557         (*regs)[i] = tswapreg(cpu_get_gpr(env, i));
1558     }
1559     (*regs)[32] = tswapreg(env->pc);
1560     (*regs)[33] = tswapreg(cpu_get_sr(env));
1561 }
1562 #define ELF_HWCAP 0
1563 #define ELF_PLATFORM NULL
1564 
1565 #endif /* TARGET_OPENRISC */
1566 
1567 #ifdef TARGET_SH4
1568 
1569 #define ELF_CLASS ELFCLASS32
1570 #define ELF_ARCH  EM_SH
1571 
init_thread(struct target_pt_regs * regs,struct image_info * infop)1572 static inline void init_thread(struct target_pt_regs *regs,
1573                                struct image_info *infop)
1574 {
1575     /* Check other registers XXXXX */
1576     regs->pc = infop->entry;
1577     regs->regs[15] = infop->start_stack;
1578 }
1579 
1580 /* See linux kernel: arch/sh/include/asm/elf.h.  */
1581 #define ELF_NREG 23
1582 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1583 
1584 /* See linux kernel: arch/sh/include/asm/ptrace.h.  */
1585 enum {
1586     TARGET_REG_PC = 16,
1587     TARGET_REG_PR = 17,
1588     TARGET_REG_SR = 18,
1589     TARGET_REG_GBR = 19,
1590     TARGET_REG_MACH = 20,
1591     TARGET_REG_MACL = 21,
1592     TARGET_REG_SYSCALL = 22
1593 };
1594 
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUSH4State * env)1595 static inline void elf_core_copy_regs(target_elf_gregset_t *regs,
1596                                       const CPUSH4State *env)
1597 {
1598     int i;
1599 
1600     for (i = 0; i < 16; i++) {
1601         (*regs)[i] = tswapreg(env->gregs[i]);
1602     }
1603 
1604     (*regs)[TARGET_REG_PC] = tswapreg(env->pc);
1605     (*regs)[TARGET_REG_PR] = tswapreg(env->pr);
1606     (*regs)[TARGET_REG_SR] = tswapreg(env->sr);
1607     (*regs)[TARGET_REG_GBR] = tswapreg(env->gbr);
1608     (*regs)[TARGET_REG_MACH] = tswapreg(env->mach);
1609     (*regs)[TARGET_REG_MACL] = tswapreg(env->macl);
1610     (*regs)[TARGET_REG_SYSCALL] = 0; /* FIXME */
1611 }
1612 
1613 #define USE_ELF_CORE_DUMP
1614 #define ELF_EXEC_PAGESIZE        4096
1615 
1616 enum {
1617     SH_CPU_HAS_FPU            = 0x0001, /* Hardware FPU support */
1618     SH_CPU_HAS_P2_FLUSH_BUG   = 0x0002, /* Need to flush the cache in P2 area */
1619     SH_CPU_HAS_MMU_PAGE_ASSOC = 0x0004, /* SH3: TLB way selection bit support */
1620     SH_CPU_HAS_DSP            = 0x0008, /* SH-DSP: DSP support */
1621     SH_CPU_HAS_PERF_COUNTER   = 0x0010, /* Hardware performance counters */
1622     SH_CPU_HAS_PTEA           = 0x0020, /* PTEA register */
1623     SH_CPU_HAS_LLSC           = 0x0040, /* movli.l/movco.l */
1624     SH_CPU_HAS_L2_CACHE       = 0x0080, /* Secondary cache / URAM */
1625     SH_CPU_HAS_OP32           = 0x0100, /* 32-bit instruction support */
1626     SH_CPU_HAS_PTEAEX         = 0x0200, /* PTE ASID Extension support */
1627 };
1628 
1629 #define ELF_HWCAP get_elf_hwcap()
1630 
get_elf_hwcap(void)1631 static uint32_t get_elf_hwcap(void)
1632 {
1633     SuperHCPU *cpu = SUPERH_CPU(thread_cpu);
1634     uint32_t hwcap = 0;
1635 
1636     hwcap |= SH_CPU_HAS_FPU;
1637 
1638     if (cpu->env.features & SH_FEATURE_SH4A) {
1639         hwcap |= SH_CPU_HAS_LLSC;
1640     }
1641 
1642     return hwcap;
1643 }
1644 
1645 #endif
1646 
1647 #ifdef TARGET_CRIS
1648 
1649 #define ELF_CLASS ELFCLASS32
1650 #define ELF_ARCH  EM_CRIS
1651 
init_thread(struct target_pt_regs * regs,struct image_info * infop)1652 static inline void init_thread(struct target_pt_regs *regs,
1653                                struct image_info *infop)
1654 {
1655     regs->erp = infop->entry;
1656 }
1657 
1658 #define ELF_EXEC_PAGESIZE        8192
1659 
1660 #endif
1661 
1662 #ifdef TARGET_M68K
1663 
1664 #define ELF_CLASS       ELFCLASS32
1665 #define ELF_ARCH        EM_68K
1666 
1667 /* ??? Does this need to do anything?
1668    #define ELF_PLAT_INIT(_r) */
1669 
init_thread(struct target_pt_regs * regs,struct image_info * infop)1670 static inline void init_thread(struct target_pt_regs *regs,
1671                                struct image_info *infop)
1672 {
1673     regs->usp = infop->start_stack;
1674     regs->sr = 0;
1675     regs->pc = infop->entry;
1676 }
1677 
1678 /* See linux kernel: arch/m68k/include/asm/elf.h.  */
1679 #define ELF_NREG 20
1680 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1681 
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUM68KState * env)1682 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUM68KState *env)
1683 {
1684     (*regs)[0] = tswapreg(env->dregs[1]);
1685     (*regs)[1] = tswapreg(env->dregs[2]);
1686     (*regs)[2] = tswapreg(env->dregs[3]);
1687     (*regs)[3] = tswapreg(env->dregs[4]);
1688     (*regs)[4] = tswapreg(env->dregs[5]);
1689     (*regs)[5] = tswapreg(env->dregs[6]);
1690     (*regs)[6] = tswapreg(env->dregs[7]);
1691     (*regs)[7] = tswapreg(env->aregs[0]);
1692     (*regs)[8] = tswapreg(env->aregs[1]);
1693     (*regs)[9] = tswapreg(env->aregs[2]);
1694     (*regs)[10] = tswapreg(env->aregs[3]);
1695     (*regs)[11] = tswapreg(env->aregs[4]);
1696     (*regs)[12] = tswapreg(env->aregs[5]);
1697     (*regs)[13] = tswapreg(env->aregs[6]);
1698     (*regs)[14] = tswapreg(env->dregs[0]);
1699     (*regs)[15] = tswapreg(env->aregs[7]);
1700     (*regs)[16] = tswapreg(env->dregs[0]); /* FIXME: orig_d0 */
1701     (*regs)[17] = tswapreg(env->sr);
1702     (*regs)[18] = tswapreg(env->pc);
1703     (*regs)[19] = 0;  /* FIXME: regs->format | regs->vector */
1704 }
1705 
1706 #define USE_ELF_CORE_DUMP
1707 #define ELF_EXEC_PAGESIZE       8192
1708 
1709 #endif
1710 
1711 #ifdef TARGET_ALPHA
1712 
1713 #define ELF_CLASS      ELFCLASS64
1714 #define ELF_ARCH       EM_ALPHA
1715 
init_thread(struct target_pt_regs * regs,struct image_info * infop)1716 static inline void init_thread(struct target_pt_regs *regs,
1717                                struct image_info *infop)
1718 {
1719     regs->pc = infop->entry;
1720     regs->ps = 8;
1721     regs->usp = infop->start_stack;
1722 }
1723 
1724 #define ELF_EXEC_PAGESIZE        8192
1725 
1726 #endif /* TARGET_ALPHA */
1727 
1728 #ifdef TARGET_S390X
1729 
1730 #define ELF_CLASS	ELFCLASS64
1731 #define ELF_DATA	ELFDATA2MSB
1732 #define ELF_ARCH	EM_S390
1733 
1734 #include "elf.h"
1735 
1736 #define ELF_HWCAP get_elf_hwcap()
1737 
1738 #define GET_FEATURE(_feat, _hwcap) \
1739     do { if (s390_has_feat(_feat)) { hwcap |= _hwcap; } } while (0)
1740 
get_elf_hwcap(void)1741 uint32_t get_elf_hwcap(void)
1742 {
1743     /*
1744      * Let's assume we always have esan3 and zarch.
1745      * 31-bit processes can use 64-bit registers (high gprs).
1746      */
1747     uint32_t hwcap = HWCAP_S390_ESAN3 | HWCAP_S390_ZARCH | HWCAP_S390_HIGH_GPRS;
1748 
1749     GET_FEATURE(S390_FEAT_STFLE, HWCAP_S390_STFLE);
1750     GET_FEATURE(S390_FEAT_MSA, HWCAP_S390_MSA);
1751     GET_FEATURE(S390_FEAT_LONG_DISPLACEMENT, HWCAP_S390_LDISP);
1752     GET_FEATURE(S390_FEAT_EXTENDED_IMMEDIATE, HWCAP_S390_EIMM);
1753     if (s390_has_feat(S390_FEAT_EXTENDED_TRANSLATION_3) &&
1754         s390_has_feat(S390_FEAT_ETF3_ENH)) {
1755         hwcap |= HWCAP_S390_ETF3EH;
1756     }
1757     GET_FEATURE(S390_FEAT_VECTOR, HWCAP_S390_VXRS);
1758     GET_FEATURE(S390_FEAT_VECTOR_ENH, HWCAP_S390_VXRS_EXT);
1759     GET_FEATURE(S390_FEAT_VECTOR_ENH2, HWCAP_S390_VXRS_EXT2);
1760 
1761     return hwcap;
1762 }
1763 
elf_hwcap_str(uint32_t bit)1764 const char *elf_hwcap_str(uint32_t bit)
1765 {
1766     static const char *hwcap_str[] = {
1767         [HWCAP_S390_NR_ESAN3]     = "esan3",
1768         [HWCAP_S390_NR_ZARCH]     = "zarch",
1769         [HWCAP_S390_NR_STFLE]     = "stfle",
1770         [HWCAP_S390_NR_MSA]       = "msa",
1771         [HWCAP_S390_NR_LDISP]     = "ldisp",
1772         [HWCAP_S390_NR_EIMM]      = "eimm",
1773         [HWCAP_S390_NR_DFP]       = "dfp",
1774         [HWCAP_S390_NR_HPAGE]     = "edat",
1775         [HWCAP_S390_NR_ETF3EH]    = "etf3eh",
1776         [HWCAP_S390_NR_HIGH_GPRS] = "highgprs",
1777         [HWCAP_S390_NR_TE]        = "te",
1778         [HWCAP_S390_NR_VXRS]      = "vx",
1779         [HWCAP_S390_NR_VXRS_BCD]  = "vxd",
1780         [HWCAP_S390_NR_VXRS_EXT]  = "vxe",
1781         [HWCAP_S390_NR_GS]        = "gs",
1782         [HWCAP_S390_NR_VXRS_EXT2] = "vxe2",
1783         [HWCAP_S390_NR_VXRS_PDE]  = "vxp",
1784         [HWCAP_S390_NR_SORT]      = "sort",
1785         [HWCAP_S390_NR_DFLT]      = "dflt",
1786         [HWCAP_S390_NR_NNPA]      = "nnpa",
1787         [HWCAP_S390_NR_PCI_MIO]   = "pcimio",
1788         [HWCAP_S390_NR_SIE]       = "sie",
1789     };
1790 
1791     return bit < ARRAY_SIZE(hwcap_str) ? hwcap_str[bit] : NULL;
1792 }
1793 
init_thread(struct target_pt_regs * regs,struct image_info * infop)1794 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop)
1795 {
1796     regs->psw.addr = infop->entry;
1797     regs->psw.mask = PSW_MASK_DAT | PSW_MASK_IO | PSW_MASK_EXT | \
1798                      PSW_MASK_MCHECK | PSW_MASK_PSTATE | PSW_MASK_64 | \
1799                      PSW_MASK_32;
1800     regs->gprs[15] = infop->start_stack;
1801 }
1802 
1803 /* See linux kernel: arch/s390/include/uapi/asm/ptrace.h (s390_regs).  */
1804 #define ELF_NREG 27
1805 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1806 
1807 enum {
1808     TARGET_REG_PSWM = 0,
1809     TARGET_REG_PSWA = 1,
1810     TARGET_REG_GPRS = 2,
1811     TARGET_REG_ARS = 18,
1812     TARGET_REG_ORIG_R2 = 26,
1813 };
1814 
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUS390XState * env)1815 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1816                                const CPUS390XState *env)
1817 {
1818     int i;
1819     uint32_t *aregs;
1820 
1821     (*regs)[TARGET_REG_PSWM] = tswapreg(env->psw.mask);
1822     (*regs)[TARGET_REG_PSWA] = tswapreg(env->psw.addr);
1823     for (i = 0; i < 16; i++) {
1824         (*regs)[TARGET_REG_GPRS + i] = tswapreg(env->regs[i]);
1825     }
1826     aregs = (uint32_t *)&((*regs)[TARGET_REG_ARS]);
1827     for (i = 0; i < 16; i++) {
1828         aregs[i] = tswap32(env->aregs[i]);
1829     }
1830     (*regs)[TARGET_REG_ORIG_R2] = 0;
1831 }
1832 
1833 #define USE_ELF_CORE_DUMP
1834 #define ELF_EXEC_PAGESIZE 4096
1835 
1836 #define VDSO_HEADER "vdso.c.inc"
1837 
1838 #endif /* TARGET_S390X */
1839 
1840 #ifdef TARGET_RISCV
1841 
1842 #define ELF_ARCH  EM_RISCV
1843 
1844 #ifdef TARGET_RISCV32
1845 #define ELF_CLASS ELFCLASS32
1846 #define VDSO_HEADER "vdso-32.c.inc"
1847 #else
1848 #define ELF_CLASS ELFCLASS64
1849 #define VDSO_HEADER "vdso-64.c.inc"
1850 #endif
1851 
1852 #define ELF_HWCAP get_elf_hwcap()
1853 
get_elf_hwcap(void)1854 static uint32_t get_elf_hwcap(void)
1855 {
1856 #define MISA_BIT(EXT) (1 << (EXT - 'A'))
1857     RISCVCPU *cpu = RISCV_CPU(thread_cpu);
1858     uint32_t mask = MISA_BIT('I') | MISA_BIT('M') | MISA_BIT('A')
1859                     | MISA_BIT('F') | MISA_BIT('D') | MISA_BIT('C')
1860                     | MISA_BIT('V');
1861 
1862     return cpu->env.misa_ext & mask;
1863 #undef MISA_BIT
1864 }
1865 
init_thread(struct target_pt_regs * regs,struct image_info * infop)1866 static inline void init_thread(struct target_pt_regs *regs,
1867                                struct image_info *infop)
1868 {
1869     regs->sepc = infop->entry;
1870     regs->sp = infop->start_stack;
1871 }
1872 
1873 #define ELF_EXEC_PAGESIZE 4096
1874 
1875 #endif /* TARGET_RISCV */
1876 
1877 #ifdef TARGET_HPPA
1878 
1879 #define ELF_CLASS       ELFCLASS32
1880 #define ELF_ARCH        EM_PARISC
1881 #define ELF_PLATFORM    "PARISC"
1882 #define STACK_GROWS_DOWN 0
1883 #define STACK_ALIGNMENT  64
1884 
1885 #define VDSO_HEADER "vdso.c.inc"
1886 
init_thread(struct target_pt_regs * regs,struct image_info * infop)1887 static inline void init_thread(struct target_pt_regs *regs,
1888                                struct image_info *infop)
1889 {
1890     regs->iaoq[0] = infop->entry;
1891     regs->iaoq[1] = infop->entry + 4;
1892     regs->gr[23] = 0;
1893     regs->gr[24] = infop->argv;
1894     regs->gr[25] = infop->argc;
1895     /* The top-of-stack contains a linkage buffer.  */
1896     regs->gr[30] = infop->start_stack + 64;
1897     regs->gr[31] = infop->entry;
1898 }
1899 
1900 #define LO_COMMPAGE  0
1901 
init_guest_commpage(void)1902 static bool init_guest_commpage(void)
1903 {
1904     /* If reserved_va, then we have already mapped 0 page on the host. */
1905     if (!reserved_va) {
1906         void *want, *addr;
1907 
1908         want = g2h_untagged(LO_COMMPAGE);
1909         addr = mmap(want, TARGET_PAGE_SIZE, PROT_NONE,
1910                     MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED_NOREPLACE, -1, 0);
1911         if (addr == MAP_FAILED) {
1912             perror("Allocating guest commpage");
1913             exit(EXIT_FAILURE);
1914         }
1915         if (addr != want) {
1916             return false;
1917         }
1918     }
1919 
1920     /*
1921      * On Linux, page zero is normally marked execute only + gateway.
1922      * Normal read or write is supposed to fail (thus PROT_NONE above),
1923      * but specific offsets have kernel code mapped to raise permissions
1924      * and implement syscalls.  Here, simply mark the page executable.
1925      * Special case the entry points during translation (see do_page_zero).
1926      */
1927     page_set_flags(LO_COMMPAGE, LO_COMMPAGE | ~TARGET_PAGE_MASK,
1928                    PAGE_EXEC | PAGE_VALID);
1929     return true;
1930 }
1931 
1932 #endif /* TARGET_HPPA */
1933 
1934 #ifdef TARGET_XTENSA
1935 
1936 #define ELF_CLASS       ELFCLASS32
1937 #define ELF_ARCH        EM_XTENSA
1938 
init_thread(struct target_pt_regs * regs,struct image_info * infop)1939 static inline void init_thread(struct target_pt_regs *regs,
1940                                struct image_info *infop)
1941 {
1942     regs->windowbase = 0;
1943     regs->windowstart = 1;
1944     regs->areg[1] = infop->start_stack;
1945     regs->pc = infop->entry;
1946     if (info_is_fdpic(infop)) {
1947         regs->areg[4] = infop->loadmap_addr;
1948         regs->areg[5] = infop->interpreter_loadmap_addr;
1949         if (infop->interpreter_loadmap_addr) {
1950             regs->areg[6] = infop->interpreter_pt_dynamic_addr;
1951         } else {
1952             regs->areg[6] = infop->pt_dynamic_addr;
1953         }
1954     }
1955 }
1956 
1957 /* See linux kernel: arch/xtensa/include/asm/elf.h.  */
1958 #define ELF_NREG 128
1959 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1960 
1961 enum {
1962     TARGET_REG_PC,
1963     TARGET_REG_PS,
1964     TARGET_REG_LBEG,
1965     TARGET_REG_LEND,
1966     TARGET_REG_LCOUNT,
1967     TARGET_REG_SAR,
1968     TARGET_REG_WINDOWSTART,
1969     TARGET_REG_WINDOWBASE,
1970     TARGET_REG_THREADPTR,
1971     TARGET_REG_AR0 = 64,
1972 };
1973 
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUXtensaState * env)1974 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1975                                const CPUXtensaState *env)
1976 {
1977     unsigned i;
1978 
1979     (*regs)[TARGET_REG_PC] = tswapreg(env->pc);
1980     (*regs)[TARGET_REG_PS] = tswapreg(env->sregs[PS] & ~PS_EXCM);
1981     (*regs)[TARGET_REG_LBEG] = tswapreg(env->sregs[LBEG]);
1982     (*regs)[TARGET_REG_LEND] = tswapreg(env->sregs[LEND]);
1983     (*regs)[TARGET_REG_LCOUNT] = tswapreg(env->sregs[LCOUNT]);
1984     (*regs)[TARGET_REG_SAR] = tswapreg(env->sregs[SAR]);
1985     (*regs)[TARGET_REG_WINDOWSTART] = tswapreg(env->sregs[WINDOW_START]);
1986     (*regs)[TARGET_REG_WINDOWBASE] = tswapreg(env->sregs[WINDOW_BASE]);
1987     (*regs)[TARGET_REG_THREADPTR] = tswapreg(env->uregs[THREADPTR]);
1988     xtensa_sync_phys_from_window((CPUXtensaState *)env);
1989     for (i = 0; i < env->config->nareg; ++i) {
1990         (*regs)[TARGET_REG_AR0 + i] = tswapreg(env->phys_regs[i]);
1991     }
1992 }
1993 
1994 #define USE_ELF_CORE_DUMP
1995 #define ELF_EXEC_PAGESIZE       4096
1996 
1997 #endif /* TARGET_XTENSA */
1998 
1999 #ifdef TARGET_HEXAGON
2000 
2001 #define ELF_CLASS       ELFCLASS32
2002 #define ELF_ARCH        EM_HEXAGON
2003 
init_thread(struct target_pt_regs * regs,struct image_info * infop)2004 static inline void init_thread(struct target_pt_regs *regs,
2005                                struct image_info *infop)
2006 {
2007     regs->sepc = infop->entry;
2008     regs->sp = infop->start_stack;
2009 }
2010 
2011 #endif /* TARGET_HEXAGON */
2012 
2013 #ifndef ELF_BASE_PLATFORM
2014 #define ELF_BASE_PLATFORM (NULL)
2015 #endif
2016 
2017 #ifndef ELF_PLATFORM
2018 #define ELF_PLATFORM (NULL)
2019 #endif
2020 
2021 #ifndef ELF_MACHINE
2022 #define ELF_MACHINE ELF_ARCH
2023 #endif
2024 
2025 #ifndef elf_check_arch
2026 #define elf_check_arch(x) ((x) == ELF_ARCH)
2027 #endif
2028 
2029 #ifndef elf_check_abi
2030 #define elf_check_abi(x) (1)
2031 #endif
2032 
2033 #ifndef ELF_HWCAP
2034 #define ELF_HWCAP 0
2035 #endif
2036 
2037 #ifndef STACK_GROWS_DOWN
2038 #define STACK_GROWS_DOWN 1
2039 #endif
2040 
2041 #ifndef STACK_ALIGNMENT
2042 #define STACK_ALIGNMENT 16
2043 #endif
2044 
2045 #ifdef TARGET_ABI32
2046 #undef ELF_CLASS
2047 #define ELF_CLASS ELFCLASS32
2048 #undef bswaptls
2049 #define bswaptls(ptr) bswap32s(ptr)
2050 #endif
2051 
2052 #ifndef EXSTACK_DEFAULT
2053 #define EXSTACK_DEFAULT false
2054 #endif
2055 
2056 #include "elf.h"
2057 
2058 /* We must delay the following stanzas until after "elf.h". */
2059 #if defined(TARGET_AARCH64)
2060 
arch_parse_elf_property(uint32_t pr_type,uint32_t pr_datasz,const uint32_t * data,struct image_info * info,Error ** errp)2061 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz,
2062                                     const uint32_t *data,
2063                                     struct image_info *info,
2064                                     Error **errp)
2065 {
2066     if (pr_type == GNU_PROPERTY_AARCH64_FEATURE_1_AND) {
2067         if (pr_datasz != sizeof(uint32_t)) {
2068             error_setg(errp, "Ill-formed GNU_PROPERTY_AARCH64_FEATURE_1_AND");
2069             return false;
2070         }
2071         /* We will extract GNU_PROPERTY_AARCH64_FEATURE_1_BTI later. */
2072         info->note_flags = *data;
2073     }
2074     return true;
2075 }
2076 #define ARCH_USE_GNU_PROPERTY 1
2077 
2078 #else
2079 
arch_parse_elf_property(uint32_t pr_type,uint32_t pr_datasz,const uint32_t * data,struct image_info * info,Error ** errp)2080 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz,
2081                                     const uint32_t *data,
2082                                     struct image_info *info,
2083                                     Error **errp)
2084 {
2085     g_assert_not_reached();
2086 }
2087 #define ARCH_USE_GNU_PROPERTY 0
2088 
2089 #endif
2090 
2091 struct exec
2092 {
2093     unsigned int a_info;   /* Use macros N_MAGIC, etc for access */
2094     unsigned int a_text;   /* length of text, in bytes */
2095     unsigned int a_data;   /* length of data, in bytes */
2096     unsigned int a_bss;    /* length of uninitialized data area, in bytes */
2097     unsigned int a_syms;   /* length of symbol table data in file, in bytes */
2098     unsigned int a_entry;  /* start address */
2099     unsigned int a_trsize; /* length of relocation info for text, in bytes */
2100     unsigned int a_drsize; /* length of relocation info for data, in bytes */
2101 };
2102 
2103 
2104 #define N_MAGIC(exec) ((exec).a_info & 0xffff)
2105 #define OMAGIC 0407
2106 #define NMAGIC 0410
2107 #define ZMAGIC 0413
2108 #define QMAGIC 0314
2109 
2110 #define DLINFO_ITEMS 16
2111 
memcpy_fromfs(void * to,const void * from,unsigned long n)2112 static inline void memcpy_fromfs(void * to, const void * from, unsigned long n)
2113 {
2114     memcpy(to, from, n);
2115 }
2116 
2117 #ifdef BSWAP_NEEDED
bswap_ehdr(struct elfhdr * ehdr)2118 static void bswap_ehdr(struct elfhdr *ehdr)
2119 {
2120     bswap16s(&ehdr->e_type);            /* Object file type */
2121     bswap16s(&ehdr->e_machine);         /* Architecture */
2122     bswap32s(&ehdr->e_version);         /* Object file version */
2123     bswaptls(&ehdr->e_entry);           /* Entry point virtual address */
2124     bswaptls(&ehdr->e_phoff);           /* Program header table file offset */
2125     bswaptls(&ehdr->e_shoff);           /* Section header table file offset */
2126     bswap32s(&ehdr->e_flags);           /* Processor-specific flags */
2127     bswap16s(&ehdr->e_ehsize);          /* ELF header size in bytes */
2128     bswap16s(&ehdr->e_phentsize);       /* Program header table entry size */
2129     bswap16s(&ehdr->e_phnum);           /* Program header table entry count */
2130     bswap16s(&ehdr->e_shentsize);       /* Section header table entry size */
2131     bswap16s(&ehdr->e_shnum);           /* Section header table entry count */
2132     bswap16s(&ehdr->e_shstrndx);        /* Section header string table index */
2133 }
2134 
bswap_phdr(struct elf_phdr * phdr,int phnum)2135 static void bswap_phdr(struct elf_phdr *phdr, int phnum)
2136 {
2137     int i;
2138     for (i = 0; i < phnum; ++i, ++phdr) {
2139         bswap32s(&phdr->p_type);        /* Segment type */
2140         bswap32s(&phdr->p_flags);       /* Segment flags */
2141         bswaptls(&phdr->p_offset);      /* Segment file offset */
2142         bswaptls(&phdr->p_vaddr);       /* Segment virtual address */
2143         bswaptls(&phdr->p_paddr);       /* Segment physical address */
2144         bswaptls(&phdr->p_filesz);      /* Segment size in file */
2145         bswaptls(&phdr->p_memsz);       /* Segment size in memory */
2146         bswaptls(&phdr->p_align);       /* Segment alignment */
2147     }
2148 }
2149 
bswap_shdr(struct elf_shdr * shdr,int shnum)2150 static void bswap_shdr(struct elf_shdr *shdr, int shnum)
2151 {
2152     int i;
2153     for (i = 0; i < shnum; ++i, ++shdr) {
2154         bswap32s(&shdr->sh_name);
2155         bswap32s(&shdr->sh_type);
2156         bswaptls(&shdr->sh_flags);
2157         bswaptls(&shdr->sh_addr);
2158         bswaptls(&shdr->sh_offset);
2159         bswaptls(&shdr->sh_size);
2160         bswap32s(&shdr->sh_link);
2161         bswap32s(&shdr->sh_info);
2162         bswaptls(&shdr->sh_addralign);
2163         bswaptls(&shdr->sh_entsize);
2164     }
2165 }
2166 
bswap_sym(struct elf_sym * sym)2167 static void bswap_sym(struct elf_sym *sym)
2168 {
2169     bswap32s(&sym->st_name);
2170     bswaptls(&sym->st_value);
2171     bswaptls(&sym->st_size);
2172     bswap16s(&sym->st_shndx);
2173 }
2174 
2175 #ifdef TARGET_MIPS
bswap_mips_abiflags(Mips_elf_abiflags_v0 * abiflags)2176 static void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags)
2177 {
2178     bswap16s(&abiflags->version);
2179     bswap32s(&abiflags->ases);
2180     bswap32s(&abiflags->isa_ext);
2181     bswap32s(&abiflags->flags1);
2182     bswap32s(&abiflags->flags2);
2183 }
2184 #endif
2185 #else
bswap_ehdr(struct elfhdr * ehdr)2186 static inline void bswap_ehdr(struct elfhdr *ehdr) { }
bswap_phdr(struct elf_phdr * phdr,int phnum)2187 static inline void bswap_phdr(struct elf_phdr *phdr, int phnum) { }
bswap_shdr(struct elf_shdr * shdr,int shnum)2188 static inline void bswap_shdr(struct elf_shdr *shdr, int shnum) { }
bswap_sym(struct elf_sym * sym)2189 static inline void bswap_sym(struct elf_sym *sym) { }
2190 #ifdef TARGET_MIPS
bswap_mips_abiflags(Mips_elf_abiflags_v0 * abiflags)2191 static inline void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags) { }
2192 #endif
2193 #endif
2194 
2195 #ifdef USE_ELF_CORE_DUMP
2196 static int elf_core_dump(int, const CPUArchState *);
2197 #endif /* USE_ELF_CORE_DUMP */
2198 static void load_symbols(struct elfhdr *hdr, const ImageSource *src,
2199                          abi_ulong load_bias);
2200 
2201 /* Verify the portions of EHDR within E_IDENT for the target.
2202    This can be performed before bswapping the entire header.  */
elf_check_ident(struct elfhdr * ehdr)2203 static bool elf_check_ident(struct elfhdr *ehdr)
2204 {
2205     return (ehdr->e_ident[EI_MAG0] == ELFMAG0
2206             && ehdr->e_ident[EI_MAG1] == ELFMAG1
2207             && ehdr->e_ident[EI_MAG2] == ELFMAG2
2208             && ehdr->e_ident[EI_MAG3] == ELFMAG3
2209             && ehdr->e_ident[EI_CLASS] == ELF_CLASS
2210             && ehdr->e_ident[EI_DATA] == ELF_DATA
2211             && ehdr->e_ident[EI_VERSION] == EV_CURRENT);
2212 }
2213 
2214 /* Verify the portions of EHDR outside of E_IDENT for the target.
2215    This has to wait until after bswapping the header.  */
elf_check_ehdr(struct elfhdr * ehdr)2216 static bool elf_check_ehdr(struct elfhdr *ehdr)
2217 {
2218     return (elf_check_arch(ehdr->e_machine)
2219             && elf_check_abi(ehdr->e_flags)
2220             && ehdr->e_ehsize == sizeof(struct elfhdr)
2221             && ehdr->e_phentsize == sizeof(struct elf_phdr)
2222             && (ehdr->e_type == ET_EXEC || ehdr->e_type == ET_DYN));
2223 }
2224 
2225 /*
2226  * 'copy_elf_strings()' copies argument/envelope strings from user
2227  * memory to free pages in kernel mem. These are in a format ready
2228  * to be put directly into the top of new user memory.
2229  *
2230  */
copy_elf_strings(int argc,char ** argv,char * scratch,abi_ulong p,abi_ulong stack_limit)2231 static abi_ulong copy_elf_strings(int argc, char **argv, char *scratch,
2232                                   abi_ulong p, abi_ulong stack_limit)
2233 {
2234     char *tmp;
2235     int len, i;
2236     abi_ulong top = p;
2237 
2238     if (!p) {
2239         return 0;       /* bullet-proofing */
2240     }
2241 
2242     if (STACK_GROWS_DOWN) {
2243         int offset = ((p - 1) % TARGET_PAGE_SIZE) + 1;
2244         for (i = argc - 1; i >= 0; --i) {
2245             tmp = argv[i];
2246             if (!tmp) {
2247                 fprintf(stderr, "VFS: argc is wrong");
2248                 exit(-1);
2249             }
2250             len = strlen(tmp) + 1;
2251             tmp += len;
2252 
2253             if (len > (p - stack_limit)) {
2254                 return 0;
2255             }
2256             while (len) {
2257                 int bytes_to_copy = (len > offset) ? offset : len;
2258                 tmp -= bytes_to_copy;
2259                 p -= bytes_to_copy;
2260                 offset -= bytes_to_copy;
2261                 len -= bytes_to_copy;
2262 
2263                 memcpy_fromfs(scratch + offset, tmp, bytes_to_copy);
2264 
2265                 if (offset == 0) {
2266                     memcpy_to_target(p, scratch, top - p);
2267                     top = p;
2268                     offset = TARGET_PAGE_SIZE;
2269                 }
2270             }
2271         }
2272         if (p != top) {
2273             memcpy_to_target(p, scratch + offset, top - p);
2274         }
2275     } else {
2276         int remaining = TARGET_PAGE_SIZE - (p % TARGET_PAGE_SIZE);
2277         for (i = 0; i < argc; ++i) {
2278             tmp = argv[i];
2279             if (!tmp) {
2280                 fprintf(stderr, "VFS: argc is wrong");
2281                 exit(-1);
2282             }
2283             len = strlen(tmp) + 1;
2284             if (len > (stack_limit - p)) {
2285                 return 0;
2286             }
2287             while (len) {
2288                 int bytes_to_copy = (len > remaining) ? remaining : len;
2289 
2290                 memcpy_fromfs(scratch + (p - top), tmp, bytes_to_copy);
2291 
2292                 tmp += bytes_to_copy;
2293                 remaining -= bytes_to_copy;
2294                 p += bytes_to_copy;
2295                 len -= bytes_to_copy;
2296 
2297                 if (remaining == 0) {
2298                     memcpy_to_target(top, scratch, p - top);
2299                     top = p;
2300                     remaining = TARGET_PAGE_SIZE;
2301                 }
2302             }
2303         }
2304         if (p != top) {
2305             memcpy_to_target(top, scratch, p - top);
2306         }
2307     }
2308 
2309     return p;
2310 }
2311 
2312 /* Older linux kernels provide up to MAX_ARG_PAGES (default: 32) of
2313  * argument/environment space. Newer kernels (>2.6.33) allow more,
2314  * dependent on stack size, but guarantee at least 32 pages for
2315  * backwards compatibility.
2316  */
2317 #define STACK_LOWER_LIMIT (32 * TARGET_PAGE_SIZE)
2318 
setup_arg_pages(struct linux_binprm * bprm,struct image_info * info)2319 static abi_ulong setup_arg_pages(struct linux_binprm *bprm,
2320                                  struct image_info *info)
2321 {
2322     abi_ulong size, error, guard;
2323     int prot;
2324 
2325     size = guest_stack_size;
2326     if (size < STACK_LOWER_LIMIT) {
2327         size = STACK_LOWER_LIMIT;
2328     }
2329 
2330     if (STACK_GROWS_DOWN) {
2331         guard = TARGET_PAGE_SIZE;
2332         if (guard < qemu_real_host_page_size()) {
2333             guard = qemu_real_host_page_size();
2334         }
2335     } else {
2336         /* no guard page for hppa target where stack grows upwards. */
2337         guard = 0;
2338     }
2339 
2340     prot = PROT_READ | PROT_WRITE;
2341     if (info->exec_stack) {
2342         prot |= PROT_EXEC;
2343     }
2344     error = target_mmap(0, size + guard, prot,
2345                         MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
2346     if (error == -1) {
2347         perror("mmap stack");
2348         exit(-1);
2349     }
2350 
2351     /* We reserve one extra page at the top of the stack as guard.  */
2352     if (STACK_GROWS_DOWN) {
2353         target_mprotect(error, guard, PROT_NONE);
2354         info->stack_limit = error + guard;
2355         return info->stack_limit + size - sizeof(void *);
2356     } else {
2357         info->stack_limit = error + size;
2358         return error;
2359     }
2360 }
2361 
2362 /**
2363  * zero_bss:
2364  *
2365  * Map and zero the bss.  We need to explicitly zero any fractional pages
2366  * after the data section (i.e. bss).  Return false on mapping failure.
2367  */
zero_bss(abi_ulong start_bss,abi_ulong end_bss,int prot,Error ** errp)2368 static bool zero_bss(abi_ulong start_bss, abi_ulong end_bss,
2369                      int prot, Error **errp)
2370 {
2371     abi_ulong align_bss;
2372 
2373     /* We only expect writable bss; the code segment shouldn't need this. */
2374     if (!(prot & PROT_WRITE)) {
2375         error_setg(errp, "PT_LOAD with non-writable bss");
2376         return false;
2377     }
2378 
2379     align_bss = TARGET_PAGE_ALIGN(start_bss);
2380     end_bss = TARGET_PAGE_ALIGN(end_bss);
2381 
2382     if (start_bss < align_bss) {
2383         int flags = page_get_flags(start_bss);
2384 
2385         if (!(flags & PAGE_RWX)) {
2386             /*
2387              * The whole address space of the executable was reserved
2388              * at the start, therefore all pages will be VALID.
2389              * But assuming there are no PROT_NONE PT_LOAD segments,
2390              * a PROT_NONE page means no data all bss, and we can
2391              * simply extend the new anon mapping back to the start
2392              * of the page of bss.
2393              */
2394             align_bss -= TARGET_PAGE_SIZE;
2395         } else {
2396             /*
2397              * The start of the bss shares a page with something.
2398              * The only thing that we expect is the data section,
2399              * which would already be marked writable.
2400              * Overlapping the RX code segment seems malformed.
2401              */
2402             if (!(flags & PAGE_WRITE)) {
2403                 error_setg(errp, "PT_LOAD with bss overlapping "
2404                            "non-writable page");
2405                 return false;
2406             }
2407 
2408             /* The page is already mapped and writable. */
2409             memset(g2h_untagged(start_bss), 0, align_bss - start_bss);
2410         }
2411     }
2412 
2413     if (align_bss < end_bss &&
2414         target_mmap(align_bss, end_bss - align_bss, prot,
2415                     MAP_FIXED | MAP_PRIVATE | MAP_ANON, -1, 0) == -1) {
2416         error_setg_errno(errp, errno, "Error mapping bss");
2417         return false;
2418     }
2419     return true;
2420 }
2421 
2422 #if defined(TARGET_ARM)
elf_is_fdpic(struct elfhdr * exec)2423 static int elf_is_fdpic(struct elfhdr *exec)
2424 {
2425     return exec->e_ident[EI_OSABI] == ELFOSABI_ARM_FDPIC;
2426 }
2427 #elif defined(TARGET_XTENSA)
elf_is_fdpic(struct elfhdr * exec)2428 static int elf_is_fdpic(struct elfhdr *exec)
2429 {
2430     return exec->e_ident[EI_OSABI] == ELFOSABI_XTENSA_FDPIC;
2431 }
2432 #else
2433 /* Default implementation, always false.  */
elf_is_fdpic(struct elfhdr * exec)2434 static int elf_is_fdpic(struct elfhdr *exec)
2435 {
2436     return 0;
2437 }
2438 #endif
2439 
loader_build_fdpic_loadmap(struct image_info * info,abi_ulong sp)2440 static abi_ulong loader_build_fdpic_loadmap(struct image_info *info, abi_ulong sp)
2441 {
2442     uint16_t n;
2443     struct elf32_fdpic_loadseg *loadsegs = info->loadsegs;
2444 
2445     /* elf32_fdpic_loadseg */
2446     n = info->nsegs;
2447     while (n--) {
2448         sp -= 12;
2449         put_user_u32(loadsegs[n].addr, sp+0);
2450         put_user_u32(loadsegs[n].p_vaddr, sp+4);
2451         put_user_u32(loadsegs[n].p_memsz, sp+8);
2452     }
2453 
2454     /* elf32_fdpic_loadmap */
2455     sp -= 4;
2456     put_user_u16(0, sp+0); /* version */
2457     put_user_u16(info->nsegs, sp+2); /* nsegs */
2458 
2459     info->personality = PER_LINUX_FDPIC;
2460     info->loadmap_addr = sp;
2461 
2462     return sp;
2463 }
2464 
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)2465 static abi_ulong create_elf_tables(abi_ulong p, int argc, int envc,
2466                                    struct elfhdr *exec,
2467                                    struct image_info *info,
2468                                    struct image_info *interp_info,
2469                                    struct image_info *vdso_info)
2470 {
2471     abi_ulong sp;
2472     abi_ulong u_argc, u_argv, u_envp, u_auxv;
2473     int size;
2474     int i;
2475     abi_ulong u_rand_bytes;
2476     uint8_t k_rand_bytes[16];
2477     abi_ulong u_platform, u_base_platform;
2478     const char *k_platform, *k_base_platform;
2479     const int n = sizeof(elf_addr_t);
2480 
2481     sp = p;
2482 
2483     /* Needs to be before we load the env/argc/... */
2484     if (elf_is_fdpic(exec)) {
2485         /* Need 4 byte alignment for these structs */
2486         sp &= ~3;
2487         sp = loader_build_fdpic_loadmap(info, sp);
2488         info->other_info = interp_info;
2489         if (interp_info) {
2490             interp_info->other_info = info;
2491             sp = loader_build_fdpic_loadmap(interp_info, sp);
2492             info->interpreter_loadmap_addr = interp_info->loadmap_addr;
2493             info->interpreter_pt_dynamic_addr = interp_info->pt_dynamic_addr;
2494         } else {
2495             info->interpreter_loadmap_addr = 0;
2496             info->interpreter_pt_dynamic_addr = 0;
2497         }
2498     }
2499 
2500     u_base_platform = 0;
2501     k_base_platform = ELF_BASE_PLATFORM;
2502     if (k_base_platform) {
2503         size_t len = strlen(k_base_platform) + 1;
2504         if (STACK_GROWS_DOWN) {
2505             sp -= (len + n - 1) & ~(n - 1);
2506             u_base_platform = sp;
2507             /* FIXME - check return value of memcpy_to_target() for failure */
2508             memcpy_to_target(sp, k_base_platform, len);
2509         } else {
2510             memcpy_to_target(sp, k_base_platform, len);
2511             u_base_platform = sp;
2512             sp += len + 1;
2513         }
2514     }
2515 
2516     u_platform = 0;
2517     k_platform = ELF_PLATFORM;
2518     if (k_platform) {
2519         size_t len = strlen(k_platform) + 1;
2520         if (STACK_GROWS_DOWN) {
2521             sp -= (len + n - 1) & ~(n - 1);
2522             u_platform = sp;
2523             /* FIXME - check return value of memcpy_to_target() for failure */
2524             memcpy_to_target(sp, k_platform, len);
2525         } else {
2526             memcpy_to_target(sp, k_platform, len);
2527             u_platform = sp;
2528             sp += len + 1;
2529         }
2530     }
2531 
2532     /* Provide 16 byte alignment for the PRNG, and basic alignment for
2533      * the argv and envp pointers.
2534      */
2535     if (STACK_GROWS_DOWN) {
2536         sp = QEMU_ALIGN_DOWN(sp, 16);
2537     } else {
2538         sp = QEMU_ALIGN_UP(sp, 16);
2539     }
2540 
2541     /*
2542      * Generate 16 random bytes for userspace PRNG seeding.
2543      */
2544     qemu_guest_getrandom_nofail(k_rand_bytes, sizeof(k_rand_bytes));
2545     if (STACK_GROWS_DOWN) {
2546         sp -= 16;
2547         u_rand_bytes = sp;
2548         /* FIXME - check return value of memcpy_to_target() for failure */
2549         memcpy_to_target(sp, k_rand_bytes, 16);
2550     } else {
2551         memcpy_to_target(sp, k_rand_bytes, 16);
2552         u_rand_bytes = sp;
2553         sp += 16;
2554     }
2555 
2556     size = (DLINFO_ITEMS + 1) * 2;
2557     if (k_base_platform) {
2558         size += 2;
2559     }
2560     if (k_platform) {
2561         size += 2;
2562     }
2563     if (vdso_info) {
2564         size += 2;
2565     }
2566 #ifdef DLINFO_ARCH_ITEMS
2567     size += DLINFO_ARCH_ITEMS * 2;
2568 #endif
2569 #ifdef ELF_HWCAP2
2570     size += 2;
2571 #endif
2572     info->auxv_len = size * n;
2573 
2574     size += envc + argc + 2;
2575     size += 1;  /* argc itself */
2576     size *= n;
2577 
2578     /* Allocate space and finalize stack alignment for entry now.  */
2579     if (STACK_GROWS_DOWN) {
2580         u_argc = QEMU_ALIGN_DOWN(sp - size, STACK_ALIGNMENT);
2581         sp = u_argc;
2582     } else {
2583         u_argc = sp;
2584         sp = QEMU_ALIGN_UP(sp + size, STACK_ALIGNMENT);
2585     }
2586 
2587     u_argv = u_argc + n;
2588     u_envp = u_argv + (argc + 1) * n;
2589     u_auxv = u_envp + (envc + 1) * n;
2590     info->saved_auxv = u_auxv;
2591     info->argc = argc;
2592     info->envc = envc;
2593     info->argv = u_argv;
2594     info->envp = u_envp;
2595 
2596     /* This is correct because Linux defines
2597      * elf_addr_t as Elf32_Off / Elf64_Off
2598      */
2599 #define NEW_AUX_ENT(id, val) do {               \
2600         put_user_ual(id, u_auxv);  u_auxv += n; \
2601         put_user_ual(val, u_auxv); u_auxv += n; \
2602     } while(0)
2603 
2604 #ifdef ARCH_DLINFO
2605     /*
2606      * ARCH_DLINFO must come first so platform specific code can enforce
2607      * special alignment requirements on the AUXV if necessary (eg. PPC).
2608      */
2609     ARCH_DLINFO;
2610 #endif
2611     /* There must be exactly DLINFO_ITEMS entries here, or the assert
2612      * on info->auxv_len will trigger.
2613      */
2614     NEW_AUX_ENT(AT_PHDR, (abi_ulong)(info->load_addr + exec->e_phoff));
2615     NEW_AUX_ENT(AT_PHENT, (abi_ulong)(sizeof (struct elf_phdr)));
2616     NEW_AUX_ENT(AT_PHNUM, (abi_ulong)(exec->e_phnum));
2617     NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(TARGET_PAGE_SIZE));
2618     NEW_AUX_ENT(AT_BASE, (abi_ulong)(interp_info ? interp_info->load_addr : 0));
2619     NEW_AUX_ENT(AT_FLAGS, (abi_ulong)0);
2620     NEW_AUX_ENT(AT_ENTRY, info->entry);
2621     NEW_AUX_ENT(AT_UID, (abi_ulong) getuid());
2622     NEW_AUX_ENT(AT_EUID, (abi_ulong) geteuid());
2623     NEW_AUX_ENT(AT_GID, (abi_ulong) getgid());
2624     NEW_AUX_ENT(AT_EGID, (abi_ulong) getegid());
2625     NEW_AUX_ENT(AT_HWCAP, (abi_ulong) ELF_HWCAP);
2626     NEW_AUX_ENT(AT_CLKTCK, (abi_ulong) sysconf(_SC_CLK_TCK));
2627     NEW_AUX_ENT(AT_RANDOM, (abi_ulong) u_rand_bytes);
2628     NEW_AUX_ENT(AT_SECURE, (abi_ulong) qemu_getauxval(AT_SECURE));
2629     NEW_AUX_ENT(AT_EXECFN, info->file_string);
2630 
2631 #ifdef ELF_HWCAP2
2632     NEW_AUX_ENT(AT_HWCAP2, (abi_ulong) ELF_HWCAP2);
2633 #endif
2634 
2635     if (u_base_platform) {
2636         NEW_AUX_ENT(AT_BASE_PLATFORM, u_base_platform);
2637     }
2638     if (u_platform) {
2639         NEW_AUX_ENT(AT_PLATFORM, u_platform);
2640     }
2641     if (vdso_info) {
2642         NEW_AUX_ENT(AT_SYSINFO_EHDR, vdso_info->load_addr);
2643     }
2644     NEW_AUX_ENT (AT_NULL, 0);
2645 #undef NEW_AUX_ENT
2646 
2647     /* Check that our initial calculation of the auxv length matches how much
2648      * we actually put into it.
2649      */
2650     assert(info->auxv_len == u_auxv - info->saved_auxv);
2651 
2652     put_user_ual(argc, u_argc);
2653 
2654     p = info->arg_strings;
2655     for (i = 0; i < argc; ++i) {
2656         put_user_ual(p, u_argv);
2657         u_argv += n;
2658         p += target_strlen(p) + 1;
2659     }
2660     put_user_ual(0, u_argv);
2661 
2662     p = info->env_strings;
2663     for (i = 0; i < envc; ++i) {
2664         put_user_ual(p, u_envp);
2665         u_envp += n;
2666         p += target_strlen(p) + 1;
2667     }
2668     put_user_ual(0, u_envp);
2669 
2670     return sp;
2671 }
2672 
2673 #if defined(HI_COMMPAGE)
2674 #define LO_COMMPAGE -1
2675 #elif defined(LO_COMMPAGE)
2676 #define HI_COMMPAGE 0
2677 #else
2678 #define HI_COMMPAGE 0
2679 #define LO_COMMPAGE -1
2680 #ifndef INIT_GUEST_COMMPAGE
2681 #define init_guest_commpage() true
2682 #endif
2683 #endif
2684 
2685 /**
2686  * pgb_try_mmap:
2687  * @addr: host start address
2688  * @addr_last: host last address
2689  * @keep: do not unmap the probe region
2690  *
2691  * Return 1 if [@addr, @addr_last] is not mapped in the host,
2692  * return 0 if it is not available to map, and -1 on mmap error.
2693  * If @keep, the region is left mapped on success, otherwise unmapped.
2694  */
pgb_try_mmap(uintptr_t addr,uintptr_t addr_last,bool keep)2695 static int pgb_try_mmap(uintptr_t addr, uintptr_t addr_last, bool keep)
2696 {
2697     size_t size = addr_last - addr + 1;
2698     void *p = mmap((void *)addr, size, PROT_NONE,
2699                    MAP_ANONYMOUS | MAP_PRIVATE |
2700                    MAP_NORESERVE | MAP_FIXED_NOREPLACE, -1, 0);
2701     int ret;
2702 
2703     if (p == MAP_FAILED) {
2704         return errno == EEXIST ? 0 : -1;
2705     }
2706     ret = p == (void *)addr;
2707     if (!keep || !ret) {
2708         munmap(p, size);
2709     }
2710     return ret;
2711 }
2712 
2713 /**
2714  * pgb_try_mmap_skip_brk(uintptr_t addr, uintptr_t size, uintptr_t brk)
2715  * @addr: host address
2716  * @addr_last: host last address
2717  * @brk: host brk
2718  *
2719  * Like pgb_try_mmap, but additionally reserve some memory following brk.
2720  */
pgb_try_mmap_skip_brk(uintptr_t addr,uintptr_t addr_last,uintptr_t brk,bool keep)2721 static int pgb_try_mmap_skip_brk(uintptr_t addr, uintptr_t addr_last,
2722                                  uintptr_t brk, bool keep)
2723 {
2724     uintptr_t brk_last = brk + 16 * MiB - 1;
2725 
2726     /* Do not map anything close to the host brk. */
2727     if (addr <= brk_last && brk <= addr_last) {
2728         return 0;
2729     }
2730     return pgb_try_mmap(addr, addr_last, keep);
2731 }
2732 
2733 /**
2734  * pgb_try_mmap_set:
2735  * @ga: set of guest addrs
2736  * @base: guest_base
2737  * @brk: host brk
2738  *
2739  * Return true if all @ga can be mapped by the host at @base.
2740  * On success, retain the mapping at index 0 for reserved_va.
2741  */
2742 
2743 typedef struct PGBAddrs {
2744     uintptr_t bounds[3][2]; /* start/last pairs */
2745     int nbounds;
2746 } PGBAddrs;
2747 
pgb_try_mmap_set(const PGBAddrs * ga,uintptr_t base,uintptr_t brk)2748 static bool pgb_try_mmap_set(const PGBAddrs *ga, uintptr_t base, uintptr_t brk)
2749 {
2750     for (int i = ga->nbounds - 1; i >= 0; --i) {
2751         if (pgb_try_mmap_skip_brk(ga->bounds[i][0] + base,
2752                                   ga->bounds[i][1] + base,
2753                                   brk, i == 0 && reserved_va) <= 0) {
2754             return false;
2755         }
2756     }
2757     return true;
2758 }
2759 
2760 /**
2761  * pgb_addr_set:
2762  * @ga: output set of guest addrs
2763  * @guest_loaddr: guest image low address
2764  * @guest_loaddr: guest image high address
2765  * @identity: create for identity mapping
2766  *
2767  * Fill in @ga with the image, COMMPAGE and NULL page.
2768  */
pgb_addr_set(PGBAddrs * ga,abi_ulong guest_loaddr,abi_ulong guest_hiaddr,bool try_identity)2769 static bool pgb_addr_set(PGBAddrs *ga, abi_ulong guest_loaddr,
2770                          abi_ulong guest_hiaddr, bool try_identity)
2771 {
2772     int n;
2773 
2774     /*
2775      * With a low commpage, or a guest mapped very low,
2776      * we may not be able to use the identity map.
2777      */
2778     if (try_identity) {
2779         if (LO_COMMPAGE != -1 && LO_COMMPAGE < mmap_min_addr) {
2780             return false;
2781         }
2782         if (guest_loaddr != 0 && guest_loaddr < mmap_min_addr) {
2783             return false;
2784         }
2785     }
2786 
2787     memset(ga, 0, sizeof(*ga));
2788     n = 0;
2789 
2790     if (reserved_va) {
2791         ga->bounds[n][0] = try_identity ? mmap_min_addr : 0;
2792         ga->bounds[n][1] = reserved_va;
2793         n++;
2794         /* LO_COMMPAGE and NULL handled by reserving from 0. */
2795     } else {
2796         /* Add any LO_COMMPAGE or NULL page. */
2797         if (LO_COMMPAGE != -1) {
2798             ga->bounds[n][0] = 0;
2799             ga->bounds[n][1] = LO_COMMPAGE + TARGET_PAGE_SIZE - 1;
2800             n++;
2801         } else if (!try_identity) {
2802             ga->bounds[n][0] = 0;
2803             ga->bounds[n][1] = TARGET_PAGE_SIZE - 1;
2804             n++;
2805         }
2806 
2807         /* Add the guest image for ET_EXEC. */
2808         if (guest_loaddr) {
2809             ga->bounds[n][0] = guest_loaddr;
2810             ga->bounds[n][1] = guest_hiaddr;
2811             n++;
2812         }
2813     }
2814 
2815     /*
2816      * Temporarily disable
2817      *   "comparison is always false due to limited range of data type"
2818      * due to comparison between unsigned and (possible) 0.
2819      */
2820 #pragma GCC diagnostic push
2821 #pragma GCC diagnostic ignored "-Wtype-limits"
2822 
2823     /* Add any HI_COMMPAGE not covered by reserved_va. */
2824     if (reserved_va < HI_COMMPAGE) {
2825         ga->bounds[n][0] = HI_COMMPAGE & qemu_real_host_page_mask();
2826         ga->bounds[n][1] = HI_COMMPAGE + TARGET_PAGE_SIZE - 1;
2827         n++;
2828     }
2829 
2830 #pragma GCC diagnostic pop
2831 
2832     ga->nbounds = n;
2833     return true;
2834 }
2835 
pgb_fail_in_use(const char * image_name)2836 static void pgb_fail_in_use(const char *image_name)
2837 {
2838     error_report("%s: requires virtual address space that is in use "
2839                  "(omit the -B option or choose a different value)",
2840                  image_name);
2841     exit(EXIT_FAILURE);
2842 }
2843 
pgb_fixed(const char * image_name,uintptr_t guest_loaddr,uintptr_t guest_hiaddr,uintptr_t align)2844 static void pgb_fixed(const char *image_name, uintptr_t guest_loaddr,
2845                       uintptr_t guest_hiaddr, uintptr_t align)
2846 {
2847     PGBAddrs ga;
2848     uintptr_t brk = (uintptr_t)sbrk(0);
2849 
2850     if (!QEMU_IS_ALIGNED(guest_base, align)) {
2851         fprintf(stderr, "Requested guest base %p does not satisfy "
2852                 "host minimum alignment (0x%" PRIxPTR ")\n",
2853                 (void *)guest_base, align);
2854         exit(EXIT_FAILURE);
2855     }
2856 
2857     if (!pgb_addr_set(&ga, guest_loaddr, guest_hiaddr, !guest_base)
2858         || !pgb_try_mmap_set(&ga, guest_base, brk)) {
2859         pgb_fail_in_use(image_name);
2860     }
2861 }
2862 
2863 /**
2864  * pgb_find_fallback:
2865  *
2866  * This is a fallback method for finding holes in the host address space
2867  * if we don't have the benefit of being able to access /proc/self/map.
2868  * It can potentially take a very long time as we can only dumbly iterate
2869  * up the host address space seeing if the allocation would work.
2870  */
pgb_find_fallback(const PGBAddrs * ga,uintptr_t align,uintptr_t brk)2871 static uintptr_t pgb_find_fallback(const PGBAddrs *ga, uintptr_t align,
2872                                    uintptr_t brk)
2873 {
2874     /* TODO: come up with a better estimate of how much to skip. */
2875     uintptr_t skip = sizeof(uintptr_t) == 4 ? MiB : GiB;
2876 
2877     for (uintptr_t base = skip; ; base += skip) {
2878         base = ROUND_UP(base, align);
2879         if (pgb_try_mmap_set(ga, base, brk)) {
2880             return base;
2881         }
2882         if (base >= -skip) {
2883             return -1;
2884         }
2885     }
2886 }
2887 
pgb_try_itree(const PGBAddrs * ga,uintptr_t base,IntervalTreeRoot * root)2888 static uintptr_t pgb_try_itree(const PGBAddrs *ga, uintptr_t base,
2889                                IntervalTreeRoot *root)
2890 {
2891     for (int i = ga->nbounds - 1; i >= 0; --i) {
2892         uintptr_t s = base + ga->bounds[i][0];
2893         uintptr_t l = base + ga->bounds[i][1];
2894         IntervalTreeNode *n;
2895 
2896         if (l < s) {
2897             /* Wraparound. Skip to advance S to mmap_min_addr. */
2898             return mmap_min_addr - s;
2899         }
2900 
2901         n = interval_tree_iter_first(root, s, l);
2902         if (n != NULL) {
2903             /* Conflict.  Skip to advance S to LAST + 1. */
2904             return n->last - s + 1;
2905         }
2906     }
2907     return 0;  /* success */
2908 }
2909 
pgb_find_itree(const PGBAddrs * ga,IntervalTreeRoot * root,uintptr_t align,uintptr_t brk)2910 static uintptr_t pgb_find_itree(const PGBAddrs *ga, IntervalTreeRoot *root,
2911                                 uintptr_t align, uintptr_t brk)
2912 {
2913     uintptr_t last = mmap_min_addr;
2914     uintptr_t base, skip;
2915 
2916     while (true) {
2917         base = ROUND_UP(last, align);
2918         if (base < last) {
2919             return -1;
2920         }
2921 
2922         skip = pgb_try_itree(ga, base, root);
2923         if (skip == 0) {
2924             break;
2925         }
2926 
2927         last = base + skip;
2928         if (last < base) {
2929             return -1;
2930         }
2931     }
2932 
2933     /*
2934      * We've chosen 'base' based on holes in the interval tree,
2935      * but we don't yet know if it is a valid host address.
2936      * Because it is the first matching hole, if the host addresses
2937      * are invalid we know there are no further matches.
2938      */
2939     return pgb_try_mmap_set(ga, base, brk) ? base : -1;
2940 }
2941 
pgb_dynamic(const char * image_name,uintptr_t guest_loaddr,uintptr_t guest_hiaddr,uintptr_t align)2942 static void pgb_dynamic(const char *image_name, uintptr_t guest_loaddr,
2943                         uintptr_t guest_hiaddr, uintptr_t align)
2944 {
2945     IntervalTreeRoot *root;
2946     uintptr_t brk, ret;
2947     PGBAddrs ga;
2948 
2949     /* Try the identity map first. */
2950     if (pgb_addr_set(&ga, guest_loaddr, guest_hiaddr, true)) {
2951         brk = (uintptr_t)sbrk(0);
2952         if (pgb_try_mmap_set(&ga, 0, brk)) {
2953             guest_base = 0;
2954             return;
2955         }
2956     }
2957 
2958     /*
2959      * Rebuild the address set for non-identity map.
2960      * This differs in the mapping of the guest NULL page.
2961      */
2962     pgb_addr_set(&ga, guest_loaddr, guest_hiaddr, false);
2963 
2964     root = read_self_maps();
2965 
2966     /* Read brk after we've read the maps, which will malloc. */
2967     brk = (uintptr_t)sbrk(0);
2968 
2969     if (!root) {
2970         ret = pgb_find_fallback(&ga, align, brk);
2971     } else {
2972         /*
2973          * Reserve the area close to the host brk.
2974          * This will be freed with the rest of the tree.
2975          */
2976         IntervalTreeNode *b = g_new0(IntervalTreeNode, 1);
2977         b->start = brk;
2978         b->last = brk + 16 * MiB - 1;
2979         interval_tree_insert(b, root);
2980 
2981         ret = pgb_find_itree(&ga, root, align, brk);
2982         free_self_maps(root);
2983     }
2984 
2985     if (ret == -1) {
2986         int w = TARGET_LONG_BITS / 4;
2987 
2988         error_report("%s: Unable to find a guest_base to satisfy all "
2989                      "guest address mapping requirements", image_name);
2990 
2991         for (int i = 0; i < ga.nbounds; ++i) {
2992             error_printf("  %0*" PRIx64 "-%0*" PRIx64 "\n",
2993                          w, (uint64_t)ga.bounds[i][0],
2994                          w, (uint64_t)ga.bounds[i][1]);
2995         }
2996         exit(EXIT_FAILURE);
2997     }
2998     guest_base = ret;
2999 }
3000 
probe_guest_base(const char * image_name,abi_ulong guest_loaddr,abi_ulong guest_hiaddr)3001 void probe_guest_base(const char *image_name, abi_ulong guest_loaddr,
3002                       abi_ulong guest_hiaddr)
3003 {
3004     /* In order to use host shmat, we must be able to honor SHMLBA.  */
3005     uintptr_t align = MAX(SHMLBA, TARGET_PAGE_SIZE);
3006 
3007     /* Sanity check the guest binary. */
3008     if (reserved_va) {
3009         if (guest_hiaddr > reserved_va) {
3010             error_report("%s: requires more than reserved virtual "
3011                          "address space (0x%" PRIx64 " > 0x%lx)",
3012                          image_name, (uint64_t)guest_hiaddr, reserved_va);
3013             exit(EXIT_FAILURE);
3014         }
3015     } else {
3016         if (guest_hiaddr != (uintptr_t)guest_hiaddr) {
3017             error_report("%s: requires more virtual address space "
3018                          "than the host can provide (0x%" PRIx64 ")",
3019                          image_name, (uint64_t)guest_hiaddr + 1);
3020             exit(EXIT_FAILURE);
3021         }
3022     }
3023 
3024     if (have_guest_base) {
3025         pgb_fixed(image_name, guest_loaddr, guest_hiaddr, align);
3026     } else {
3027         pgb_dynamic(image_name, guest_loaddr, guest_hiaddr, align);
3028     }
3029 
3030     /* Reserve and initialize the commpage. */
3031     if (!init_guest_commpage()) {
3032         /* We have already probed for the commpage being free. */
3033         g_assert_not_reached();
3034     }
3035 
3036     assert(QEMU_IS_ALIGNED(guest_base, align));
3037     qemu_log_mask(CPU_LOG_PAGE, "Locating guest address space "
3038                   "@ 0x%" PRIx64 "\n", (uint64_t)guest_base);
3039 }
3040 
3041 enum {
3042     /* The string "GNU\0" as a magic number. */
3043     GNU0_MAGIC = const_le32('G' | 'N' << 8 | 'U' << 16),
3044     NOTE_DATA_SZ = 1 * KiB,
3045     NOTE_NAME_SZ = 4,
3046     ELF_GNU_PROPERTY_ALIGN = ELF_CLASS == ELFCLASS32 ? 4 : 8,
3047 };
3048 
3049 /*
3050  * Process a single gnu_property entry.
3051  * Return false for error.
3052  */
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)3053 static bool parse_elf_property(const uint32_t *data, int *off, int datasz,
3054                                struct image_info *info, bool have_prev_type,
3055                                uint32_t *prev_type, Error **errp)
3056 {
3057     uint32_t pr_type, pr_datasz, step;
3058 
3059     if (*off > datasz || !QEMU_IS_ALIGNED(*off, ELF_GNU_PROPERTY_ALIGN)) {
3060         goto error_data;
3061     }
3062     datasz -= *off;
3063     data += *off / sizeof(uint32_t);
3064 
3065     if (datasz < 2 * sizeof(uint32_t)) {
3066         goto error_data;
3067     }
3068     pr_type = data[0];
3069     pr_datasz = data[1];
3070     data += 2;
3071     datasz -= 2 * sizeof(uint32_t);
3072     step = ROUND_UP(pr_datasz, ELF_GNU_PROPERTY_ALIGN);
3073     if (step > datasz) {
3074         goto error_data;
3075     }
3076 
3077     /* Properties are supposed to be unique and sorted on pr_type. */
3078     if (have_prev_type && pr_type <= *prev_type) {
3079         if (pr_type == *prev_type) {
3080             error_setg(errp, "Duplicate property in PT_GNU_PROPERTY");
3081         } else {
3082             error_setg(errp, "Unsorted property in PT_GNU_PROPERTY");
3083         }
3084         return false;
3085     }
3086     *prev_type = pr_type;
3087 
3088     if (!arch_parse_elf_property(pr_type, pr_datasz, data, info, errp)) {
3089         return false;
3090     }
3091 
3092     *off += 2 * sizeof(uint32_t) + step;
3093     return true;
3094 
3095  error_data:
3096     error_setg(errp, "Ill-formed property in PT_GNU_PROPERTY");
3097     return false;
3098 }
3099 
3100 /* Process NT_GNU_PROPERTY_TYPE_0. */
parse_elf_properties(const ImageSource * src,struct image_info * info,const struct elf_phdr * phdr,Error ** errp)3101 static bool parse_elf_properties(const ImageSource *src,
3102                                  struct image_info *info,
3103                                  const struct elf_phdr *phdr,
3104                                  Error **errp)
3105 {
3106     union {
3107         struct elf_note nhdr;
3108         uint32_t data[NOTE_DATA_SZ / sizeof(uint32_t)];
3109     } note;
3110 
3111     int n, off, datasz;
3112     bool have_prev_type;
3113     uint32_t prev_type;
3114 
3115     /* Unless the arch requires properties, ignore them. */
3116     if (!ARCH_USE_GNU_PROPERTY) {
3117         return true;
3118     }
3119 
3120     /* If the properties are crazy large, that's too bad. */
3121     n = phdr->p_filesz;
3122     if (n > sizeof(note)) {
3123         error_setg(errp, "PT_GNU_PROPERTY too large");
3124         return false;
3125     }
3126     if (n < sizeof(note.nhdr)) {
3127         error_setg(errp, "PT_GNU_PROPERTY too small");
3128         return false;
3129     }
3130 
3131     if (!imgsrc_read(&note, phdr->p_offset, n, src, errp)) {
3132         return false;
3133     }
3134 
3135     /*
3136      * The contents of a valid PT_GNU_PROPERTY is a sequence
3137      * of uint32_t -- swap them all now.
3138      */
3139 #ifdef BSWAP_NEEDED
3140     for (int i = 0; i < n / 4; i++) {
3141         bswap32s(note.data + i);
3142     }
3143 #endif
3144 
3145     /*
3146      * Note that nhdr is 3 words, and that the "name" described by namesz
3147      * immediately follows nhdr and is thus at the 4th word.  Further, all
3148      * of the inputs to the kernel's round_up are multiples of 4.
3149      */
3150     if (note.nhdr.n_type != NT_GNU_PROPERTY_TYPE_0 ||
3151         note.nhdr.n_namesz != NOTE_NAME_SZ ||
3152         note.data[3] != GNU0_MAGIC) {
3153         error_setg(errp, "Invalid note in PT_GNU_PROPERTY");
3154         return false;
3155     }
3156     off = sizeof(note.nhdr) + NOTE_NAME_SZ;
3157 
3158     datasz = note.nhdr.n_descsz + off;
3159     if (datasz > n) {
3160         error_setg(errp, "Invalid note size in PT_GNU_PROPERTY");
3161         return false;
3162     }
3163 
3164     have_prev_type = false;
3165     prev_type = 0;
3166     while (1) {
3167         if (off == datasz) {
3168             return true;  /* end, exit ok */
3169         }
3170         if (!parse_elf_property(note.data, &off, datasz, info,
3171                                 have_prev_type, &prev_type, errp)) {
3172             return false;
3173         }
3174         have_prev_type = true;
3175     }
3176 }
3177 
3178 /**
3179  * load_elf_image: Load an ELF image into the address space.
3180  * @image_name: the filename of the image, to use in error messages.
3181  * @src: the ImageSource from which to read.
3182  * @info: info collected from the loaded image.
3183  * @ehdr: the ELF header, not yet bswapped.
3184  * @pinterp_name: record any PT_INTERP string found.
3185  *
3186  * On return: @info values will be filled in, as necessary or available.
3187  */
3188 
load_elf_image(const char * image_name,const ImageSource * src,struct image_info * info,struct elfhdr * ehdr,char ** pinterp_name)3189 static void load_elf_image(const char *image_name, const ImageSource *src,
3190                            struct image_info *info, struct elfhdr *ehdr,
3191                            char **pinterp_name)
3192 {
3193     g_autofree struct elf_phdr *phdr = NULL;
3194     abi_ulong load_addr, load_bias, loaddr, hiaddr, error;
3195     int i, prot_exec;
3196     Error *err = NULL;
3197 
3198     /*
3199      * First of all, some simple consistency checks.
3200      * Note that we rely on the bswapped ehdr staying in bprm_buf,
3201      * for later use by load_elf_binary and create_elf_tables.
3202      */
3203     if (!imgsrc_read(ehdr, 0, sizeof(*ehdr), src, &err)) {
3204         goto exit_errmsg;
3205     }
3206     if (!elf_check_ident(ehdr)) {
3207         error_setg(&err, "Invalid ELF image for this architecture");
3208         goto exit_errmsg;
3209     }
3210     bswap_ehdr(ehdr);
3211     if (!elf_check_ehdr(ehdr)) {
3212         error_setg(&err, "Invalid ELF image for this architecture");
3213         goto exit_errmsg;
3214     }
3215 
3216     phdr = imgsrc_read_alloc(ehdr->e_phoff,
3217                              ehdr->e_phnum * sizeof(struct elf_phdr),
3218                              src, &err);
3219     if (phdr == NULL) {
3220         goto exit_errmsg;
3221     }
3222     bswap_phdr(phdr, ehdr->e_phnum);
3223 
3224     info->nsegs = 0;
3225     info->pt_dynamic_addr = 0;
3226 
3227     mmap_lock();
3228 
3229     /*
3230      * Find the maximum size of the image and allocate an appropriate
3231      * amount of memory to handle that.  Locate the interpreter, if any.
3232      */
3233     loaddr = -1, hiaddr = 0;
3234     info->alignment = 0;
3235     info->exec_stack = EXSTACK_DEFAULT;
3236     for (i = 0; i < ehdr->e_phnum; ++i) {
3237         struct elf_phdr *eppnt = phdr + i;
3238         if (eppnt->p_type == PT_LOAD) {
3239             abi_ulong a = eppnt->p_vaddr & TARGET_PAGE_MASK;
3240             if (a < loaddr) {
3241                 loaddr = a;
3242             }
3243             a = eppnt->p_vaddr + eppnt->p_memsz - 1;
3244             if (a > hiaddr) {
3245                 hiaddr = a;
3246             }
3247             ++info->nsegs;
3248             info->alignment |= eppnt->p_align;
3249         } else if (eppnt->p_type == PT_INTERP && pinterp_name) {
3250             g_autofree char *interp_name = NULL;
3251 
3252             if (*pinterp_name) {
3253                 error_setg(&err, "Multiple PT_INTERP entries");
3254                 goto exit_errmsg;
3255             }
3256 
3257             interp_name = imgsrc_read_alloc(eppnt->p_offset, eppnt->p_filesz,
3258                                             src, &err);
3259             if (interp_name == NULL) {
3260                 goto exit_errmsg;
3261             }
3262             if (interp_name[eppnt->p_filesz - 1] != 0) {
3263                 error_setg(&err, "Invalid PT_INTERP entry");
3264                 goto exit_errmsg;
3265             }
3266             *pinterp_name = g_steal_pointer(&interp_name);
3267         } else if (eppnt->p_type == PT_GNU_PROPERTY) {
3268             if (!parse_elf_properties(src, info, eppnt, &err)) {
3269                 goto exit_errmsg;
3270             }
3271         } else if (eppnt->p_type == PT_GNU_STACK) {
3272             info->exec_stack = eppnt->p_flags & PF_X;
3273         }
3274     }
3275 
3276     load_addr = loaddr;
3277 
3278     if (pinterp_name != NULL) {
3279         if (ehdr->e_type == ET_EXEC) {
3280             /*
3281              * Make sure that the low address does not conflict with
3282              * MMAP_MIN_ADDR or the QEMU application itself.
3283              */
3284             probe_guest_base(image_name, loaddr, hiaddr);
3285         } else {
3286             abi_ulong align;
3287 
3288             /*
3289              * The binary is dynamic, but we still need to
3290              * select guest_base.  In this case we pass a size.
3291              */
3292             probe_guest_base(image_name, 0, hiaddr - loaddr);
3293 
3294             /*
3295              * Avoid collision with the loader by providing a different
3296              * default load address.
3297              */
3298             load_addr += elf_et_dyn_base;
3299 
3300             /*
3301              * TODO: Better support for mmap alignment is desirable.
3302              * Since we do not have complete control over the guest
3303              * address space, we prefer the kernel to choose some address
3304              * rather than force the use of LOAD_ADDR via MAP_FIXED.
3305              * But without MAP_FIXED we cannot guarantee alignment,
3306              * only suggest it.
3307              */
3308             align = pow2ceil(info->alignment);
3309             if (align) {
3310                 load_addr &= -align;
3311             }
3312         }
3313     }
3314 
3315     /*
3316      * Reserve address space for all of this.
3317      *
3318      * In the case of ET_EXEC, we supply MAP_FIXED_NOREPLACE so that we get
3319      * exactly the address range that is required.  Without reserved_va,
3320      * the guest address space is not isolated.  We have attempted to avoid
3321      * conflict with the host program itself via probe_guest_base, but using
3322      * MAP_FIXED_NOREPLACE instead of MAP_FIXED provides an extra check.
3323      *
3324      * Otherwise this is ET_DYN, and we are searching for a location
3325      * that can hold the memory space required.  If the image is
3326      * pre-linked, LOAD_ADDR will be non-zero, and the kernel should
3327      * honor that address if it happens to be free.
3328      *
3329      * In both cases, we will overwrite pages in this range with mappings
3330      * from the executable.
3331      */
3332     load_addr = target_mmap(load_addr, (size_t)hiaddr - loaddr + 1, PROT_NONE,
3333                             MAP_PRIVATE | MAP_ANON | MAP_NORESERVE |
3334                             (ehdr->e_type == ET_EXEC ? MAP_FIXED_NOREPLACE : 0),
3335                             -1, 0);
3336     if (load_addr == -1) {
3337         goto exit_mmap;
3338     }
3339     load_bias = load_addr - loaddr;
3340 
3341     if (elf_is_fdpic(ehdr)) {
3342         struct elf32_fdpic_loadseg *loadsegs = info->loadsegs =
3343             g_malloc(sizeof(*loadsegs) * info->nsegs);
3344 
3345         for (i = 0; i < ehdr->e_phnum; ++i) {
3346             switch (phdr[i].p_type) {
3347             case PT_DYNAMIC:
3348                 info->pt_dynamic_addr = phdr[i].p_vaddr + load_bias;
3349                 break;
3350             case PT_LOAD:
3351                 loadsegs->addr = phdr[i].p_vaddr + load_bias;
3352                 loadsegs->p_vaddr = phdr[i].p_vaddr;
3353                 loadsegs->p_memsz = phdr[i].p_memsz;
3354                 ++loadsegs;
3355                 break;
3356             }
3357         }
3358     }
3359 
3360     info->load_bias = load_bias;
3361     info->code_offset = load_bias;
3362     info->data_offset = load_bias;
3363     info->load_addr = load_addr;
3364     info->entry = ehdr->e_entry + load_bias;
3365     info->start_code = -1;
3366     info->end_code = 0;
3367     info->start_data = -1;
3368     info->end_data = 0;
3369     /* Usual start for brk is after all sections of the main executable. */
3370     info->brk = TARGET_PAGE_ALIGN(hiaddr + load_bias);
3371     info->elf_flags = ehdr->e_flags;
3372 
3373     prot_exec = PROT_EXEC;
3374 #ifdef TARGET_AARCH64
3375     /*
3376      * If the BTI feature is present, this indicates that the executable
3377      * pages of the startup binary should be mapped with PROT_BTI, so that
3378      * branch targets are enforced.
3379      *
3380      * The startup binary is either the interpreter or the static executable.
3381      * The interpreter is responsible for all pages of a dynamic executable.
3382      *
3383      * Elf notes are backward compatible to older cpus.
3384      * Do not enable BTI unless it is supported.
3385      */
3386     if ((info->note_flags & GNU_PROPERTY_AARCH64_FEATURE_1_BTI)
3387         && (pinterp_name == NULL || *pinterp_name == 0)
3388         && cpu_isar_feature(aa64_bti, ARM_CPU(thread_cpu))) {
3389         prot_exec |= TARGET_PROT_BTI;
3390     }
3391 #endif
3392 
3393     for (i = 0; i < ehdr->e_phnum; i++) {
3394         struct elf_phdr *eppnt = phdr + i;
3395         if (eppnt->p_type == PT_LOAD) {
3396             abi_ulong vaddr, vaddr_po, vaddr_ps, vaddr_ef, vaddr_em;
3397             int elf_prot = 0;
3398 
3399             if (eppnt->p_flags & PF_R) {
3400                 elf_prot |= PROT_READ;
3401             }
3402             if (eppnt->p_flags & PF_W) {
3403                 elf_prot |= PROT_WRITE;
3404             }
3405             if (eppnt->p_flags & PF_X) {
3406                 elf_prot |= prot_exec;
3407             }
3408 
3409             vaddr = load_bias + eppnt->p_vaddr;
3410             vaddr_po = vaddr & ~TARGET_PAGE_MASK;
3411             vaddr_ps = vaddr & TARGET_PAGE_MASK;
3412 
3413             vaddr_ef = vaddr + eppnt->p_filesz;
3414             vaddr_em = vaddr + eppnt->p_memsz;
3415 
3416             /*
3417              * Some segments may be completely empty, with a non-zero p_memsz
3418              * but no backing file segment.
3419              */
3420             if (eppnt->p_filesz != 0) {
3421                 error = imgsrc_mmap(vaddr_ps, eppnt->p_filesz + vaddr_po,
3422                                     elf_prot, MAP_PRIVATE | MAP_FIXED,
3423                                     src, eppnt->p_offset - vaddr_po);
3424                 if (error == -1) {
3425                     goto exit_mmap;
3426                 }
3427             }
3428 
3429             /* If the load segment requests extra zeros (e.g. bss), map it. */
3430             if (vaddr_ef < vaddr_em &&
3431                 !zero_bss(vaddr_ef, vaddr_em, elf_prot, &err)) {
3432                 goto exit_errmsg;
3433             }
3434 
3435             /* Find the full program boundaries.  */
3436             if (elf_prot & PROT_EXEC) {
3437                 if (vaddr < info->start_code) {
3438                     info->start_code = vaddr;
3439                 }
3440                 if (vaddr_ef > info->end_code) {
3441                     info->end_code = vaddr_ef;
3442                 }
3443             }
3444             if (elf_prot & PROT_WRITE) {
3445                 if (vaddr < info->start_data) {
3446                     info->start_data = vaddr;
3447                 }
3448                 if (vaddr_ef > info->end_data) {
3449                     info->end_data = vaddr_ef;
3450                 }
3451             }
3452 #ifdef TARGET_MIPS
3453         } else if (eppnt->p_type == PT_MIPS_ABIFLAGS) {
3454             Mips_elf_abiflags_v0 abiflags;
3455 
3456             if (!imgsrc_read(&abiflags, eppnt->p_offset, sizeof(abiflags),
3457                              src, &err)) {
3458                 goto exit_errmsg;
3459             }
3460             bswap_mips_abiflags(&abiflags);
3461             info->fp_abi = abiflags.fp_abi;
3462 #endif
3463         }
3464     }
3465 
3466     if (info->end_data == 0) {
3467         info->start_data = info->end_code;
3468         info->end_data = info->end_code;
3469     }
3470 
3471     if (qemu_log_enabled()) {
3472         load_symbols(ehdr, src, load_bias);
3473     }
3474 
3475     debuginfo_report_elf(image_name, src->fd, load_bias);
3476 
3477     mmap_unlock();
3478 
3479     close(src->fd);
3480     return;
3481 
3482  exit_mmap:
3483     error_setg_errno(&err, errno, "Error mapping file");
3484     goto exit_errmsg;
3485  exit_errmsg:
3486     error_reportf_err(err, "%s: ", image_name);
3487     exit(-1);
3488 }
3489 
load_elf_interp(const char * filename,struct image_info * info,char bprm_buf[BPRM_BUF_SIZE])3490 static void load_elf_interp(const char *filename, struct image_info *info,
3491                             char bprm_buf[BPRM_BUF_SIZE])
3492 {
3493     struct elfhdr ehdr;
3494     ImageSource src;
3495     int fd, retval;
3496     Error *err = NULL;
3497 
3498     fd = open(path(filename), O_RDONLY);
3499     if (fd < 0) {
3500         error_setg_file_open(&err, errno, filename);
3501         error_report_err(err);
3502         exit(-1);
3503     }
3504 
3505     retval = read(fd, bprm_buf, BPRM_BUF_SIZE);
3506     if (retval < 0) {
3507         error_setg_errno(&err, errno, "Error reading file header");
3508         error_reportf_err(err, "%s: ", filename);
3509         exit(-1);
3510     }
3511 
3512     src.fd = fd;
3513     src.cache = bprm_buf;
3514     src.cache_size = retval;
3515 
3516     load_elf_image(filename, &src, info, &ehdr, NULL);
3517 }
3518 
3519 #ifdef VDSO_HEADER
3520 #include VDSO_HEADER
3521 #define  vdso_image_info()  &vdso_image_info
3522 #else
3523 #define  vdso_image_info()  NULL
3524 #endif
3525 
load_elf_vdso(struct image_info * info,const VdsoImageInfo * vdso)3526 static void load_elf_vdso(struct image_info *info, const VdsoImageInfo *vdso)
3527 {
3528     ImageSource src;
3529     struct elfhdr ehdr;
3530     abi_ulong load_bias, load_addr;
3531 
3532     src.fd = -1;
3533     src.cache = vdso->image;
3534     src.cache_size = vdso->image_size;
3535 
3536     load_elf_image("<internal-vdso>", &src, info, &ehdr, NULL);
3537     load_addr = info->load_addr;
3538     load_bias = info->load_bias;
3539 
3540     /*
3541      * We need to relocate the VDSO image.  The one built into the kernel
3542      * is built for a fixed address.  The one built for QEMU is not, since
3543      * that requires close control of the guest address space.
3544      * We pre-processed the image to locate all of the addresses that need
3545      * to be updated.
3546      */
3547     for (unsigned i = 0, n = vdso->reloc_count; i < n; i++) {
3548         abi_ulong *addr = g2h_untagged(load_addr + vdso->relocs[i]);
3549         *addr = tswapal(tswapal(*addr) + load_bias);
3550     }
3551 
3552     /* Install signal trampolines, if present. */
3553     if (vdso->sigreturn_ofs) {
3554         default_sigreturn = load_addr + vdso->sigreturn_ofs;
3555     }
3556     if (vdso->rt_sigreturn_ofs) {
3557         default_rt_sigreturn = load_addr + vdso->rt_sigreturn_ofs;
3558     }
3559 
3560     /* Remove write from VDSO segment. */
3561     target_mprotect(info->start_data, info->end_data - info->start_data,
3562                     PROT_READ | PROT_EXEC);
3563 }
3564 
symfind(const void * s0,const void * s1)3565 static int symfind(const void *s0, const void *s1)
3566 {
3567     struct elf_sym *sym = (struct elf_sym *)s1;
3568     __typeof(sym->st_value) addr = *(uint64_t *)s0;
3569     int result = 0;
3570 
3571     if (addr < sym->st_value) {
3572         result = -1;
3573     } else if (addr >= sym->st_value + sym->st_size) {
3574         result = 1;
3575     }
3576     return result;
3577 }
3578 
lookup_symbolxx(struct syminfo * s,uint64_t orig_addr)3579 static const char *lookup_symbolxx(struct syminfo *s, uint64_t orig_addr)
3580 {
3581 #if ELF_CLASS == ELFCLASS32
3582     struct elf_sym *syms = s->disas_symtab.elf32;
3583 #else
3584     struct elf_sym *syms = s->disas_symtab.elf64;
3585 #endif
3586 
3587     // binary search
3588     struct elf_sym *sym;
3589 
3590     sym = bsearch(&orig_addr, syms, s->disas_num_syms, sizeof(*syms), symfind);
3591     if (sym != NULL) {
3592         return s->disas_strtab + sym->st_name;
3593     }
3594 
3595     return "";
3596 }
3597 
3598 /* FIXME: This should use elf_ops.h.inc  */
symcmp(const void * s0,const void * s1)3599 static int symcmp(const void *s0, const void *s1)
3600 {
3601     struct elf_sym *sym0 = (struct elf_sym *)s0;
3602     struct elf_sym *sym1 = (struct elf_sym *)s1;
3603     return (sym0->st_value < sym1->st_value)
3604         ? -1
3605         : ((sym0->st_value > sym1->st_value) ? 1 : 0);
3606 }
3607 
3608 /* Best attempt to load symbols from this ELF object. */
load_symbols(struct elfhdr * hdr,const ImageSource * src,abi_ulong load_bias)3609 static void load_symbols(struct elfhdr *hdr, const ImageSource *src,
3610                          abi_ulong load_bias)
3611 {
3612     int i, shnum, nsyms, sym_idx = 0, str_idx = 0;
3613     g_autofree struct elf_shdr *shdr = NULL;
3614     char *strings = NULL;
3615     struct elf_sym *syms = NULL;
3616     struct elf_sym *new_syms;
3617     uint64_t segsz;
3618 
3619     shnum = hdr->e_shnum;
3620     shdr = imgsrc_read_alloc(hdr->e_shoff, shnum * sizeof(struct elf_shdr),
3621                              src, NULL);
3622     if (shdr == NULL) {
3623         return;
3624     }
3625 
3626     bswap_shdr(shdr, shnum);
3627     for (i = 0; i < shnum; ++i) {
3628         if (shdr[i].sh_type == SHT_SYMTAB) {
3629             sym_idx = i;
3630             str_idx = shdr[i].sh_link;
3631             goto found;
3632         }
3633     }
3634 
3635     /* There will be no symbol table if the file was stripped.  */
3636     return;
3637 
3638  found:
3639     /* Now know where the strtab and symtab are.  Snarf them.  */
3640 
3641     segsz = shdr[str_idx].sh_size;
3642     strings = g_try_malloc(segsz);
3643     if (!strings) {
3644         goto give_up;
3645     }
3646     if (!imgsrc_read(strings, shdr[str_idx].sh_offset, segsz, src, NULL)) {
3647         goto give_up;
3648     }
3649 
3650     segsz = shdr[sym_idx].sh_size;
3651     if (segsz / sizeof(struct elf_sym) > INT_MAX) {
3652         /*
3653          * Implausibly large symbol table: give up rather than ploughing
3654          * on with the number of symbols calculation overflowing.
3655          */
3656         goto give_up;
3657     }
3658     nsyms = segsz / sizeof(struct elf_sym);
3659     syms = g_try_malloc(segsz);
3660     if (!syms) {
3661         goto give_up;
3662     }
3663     if (!imgsrc_read(syms, shdr[sym_idx].sh_offset, segsz, src, NULL)) {
3664         goto give_up;
3665     }
3666 
3667     for (i = 0; i < nsyms; ) {
3668         bswap_sym(syms + i);
3669         /* Throw away entries which we do not need.  */
3670         if (syms[i].st_shndx == SHN_UNDEF
3671             || syms[i].st_shndx >= SHN_LORESERVE
3672             || ELF_ST_TYPE(syms[i].st_info) != STT_FUNC) {
3673             if (i < --nsyms) {
3674                 syms[i] = syms[nsyms];
3675             }
3676         } else {
3677 #if defined(TARGET_ARM) || defined (TARGET_MIPS)
3678             /* The bottom address bit marks a Thumb or MIPS16 symbol.  */
3679             syms[i].st_value &= ~(target_ulong)1;
3680 #endif
3681             syms[i].st_value += load_bias;
3682             i++;
3683         }
3684     }
3685 
3686     /* No "useful" symbol.  */
3687     if (nsyms == 0) {
3688         goto give_up;
3689     }
3690 
3691     /*
3692      * Attempt to free the storage associated with the local symbols
3693      * that we threw away.  Whether or not this has any effect on the
3694      * memory allocation depends on the malloc implementation and how
3695      * many symbols we managed to discard.
3696      */
3697     new_syms = g_try_renew(struct elf_sym, syms, nsyms);
3698     if (new_syms == NULL) {
3699         goto give_up;
3700     }
3701     syms = new_syms;
3702 
3703     qsort(syms, nsyms, sizeof(*syms), symcmp);
3704 
3705     {
3706         struct syminfo *s = g_new(struct syminfo, 1);
3707 
3708         s->disas_strtab = strings;
3709         s->disas_num_syms = nsyms;
3710 #if ELF_CLASS == ELFCLASS32
3711         s->disas_symtab.elf32 = syms;
3712 #else
3713         s->disas_symtab.elf64 = syms;
3714 #endif
3715         s->lookup_symbol = lookup_symbolxx;
3716         s->next = syminfos;
3717         syminfos = s;
3718     }
3719     return;
3720 
3721  give_up:
3722     g_free(strings);
3723     g_free(syms);
3724 }
3725 
get_elf_eflags(int fd)3726 uint32_t get_elf_eflags(int fd)
3727 {
3728     struct elfhdr ehdr;
3729     off_t offset;
3730     int ret;
3731 
3732     /* Read ELF header */
3733     offset = lseek(fd, 0, SEEK_SET);
3734     if (offset == (off_t) -1) {
3735         return 0;
3736     }
3737     ret = read(fd, &ehdr, sizeof(ehdr));
3738     if (ret < sizeof(ehdr)) {
3739         return 0;
3740     }
3741     offset = lseek(fd, offset, SEEK_SET);
3742     if (offset == (off_t) -1) {
3743         return 0;
3744     }
3745 
3746     /* Check ELF signature */
3747     if (!elf_check_ident(&ehdr)) {
3748         return 0;
3749     }
3750 
3751     /* check header */
3752     bswap_ehdr(&ehdr);
3753     if (!elf_check_ehdr(&ehdr)) {
3754         return 0;
3755     }
3756 
3757     /* return architecture id */
3758     return ehdr.e_flags;
3759 }
3760 
load_elf_binary(struct linux_binprm * bprm,struct image_info * info)3761 int load_elf_binary(struct linux_binprm *bprm, struct image_info *info)
3762 {
3763     /*
3764      * We need a copy of the elf header for passing to create_elf_tables.
3765      * We will have overwritten the original when we re-use bprm->buf
3766      * while loading the interpreter.  Allocate the storage for this now
3767      * and let elf_load_image do any swapping that may be required.
3768      */
3769     struct elfhdr ehdr;
3770     struct image_info interp_info, vdso_info;
3771     char *elf_interpreter = NULL;
3772     char *scratch;
3773 
3774     memset(&interp_info, 0, sizeof(interp_info));
3775 #ifdef TARGET_MIPS
3776     interp_info.fp_abi = MIPS_ABI_FP_UNKNOWN;
3777 #endif
3778 
3779     load_elf_image(bprm->filename, &bprm->src, info, &ehdr, &elf_interpreter);
3780 
3781     /* Do this so that we can load the interpreter, if need be.  We will
3782        change some of these later */
3783     bprm->p = setup_arg_pages(bprm, info);
3784 
3785     scratch = g_new0(char, TARGET_PAGE_SIZE);
3786     if (STACK_GROWS_DOWN) {
3787         bprm->p = copy_elf_strings(1, &bprm->filename, scratch,
3788                                    bprm->p, info->stack_limit);
3789         info->file_string = bprm->p;
3790         bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch,
3791                                    bprm->p, info->stack_limit);
3792         info->env_strings = bprm->p;
3793         bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch,
3794                                    bprm->p, info->stack_limit);
3795         info->arg_strings = bprm->p;
3796     } else {
3797         info->arg_strings = bprm->p;
3798         bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch,
3799                                    bprm->p, info->stack_limit);
3800         info->env_strings = bprm->p;
3801         bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch,
3802                                    bprm->p, info->stack_limit);
3803         info->file_string = bprm->p;
3804         bprm->p = copy_elf_strings(1, &bprm->filename, scratch,
3805                                    bprm->p, info->stack_limit);
3806     }
3807 
3808     g_free(scratch);
3809 
3810     if (!bprm->p) {
3811         fprintf(stderr, "%s: %s\n", bprm->filename, strerror(E2BIG));
3812         exit(-1);
3813     }
3814 
3815     if (elf_interpreter) {
3816         load_elf_interp(elf_interpreter, &interp_info, bprm->buf);
3817 
3818         /*
3819          * While unusual because of ELF_ET_DYN_BASE, if we are unlucky
3820          * with the mappings the interpreter can be loaded above but
3821          * near the main executable, which can leave very little room
3822          * for the heap.
3823          * If the current brk has less than 16MB, use the end of the
3824          * interpreter.
3825          */
3826         if (interp_info.brk > info->brk &&
3827             interp_info.load_bias - info->brk < 16 * MiB)  {
3828             info->brk = interp_info.brk;
3829         }
3830 
3831         /* If the program interpreter is one of these two, then assume
3832            an iBCS2 image.  Otherwise assume a native linux image.  */
3833 
3834         if (strcmp(elf_interpreter, "/usr/lib/libc.so.1") == 0
3835             || strcmp(elf_interpreter, "/usr/lib/ld.so.1") == 0) {
3836             info->personality = PER_SVR4;
3837 
3838             /* Why this, you ask???  Well SVr4 maps page 0 as read-only,
3839                and some applications "depend" upon this behavior.  Since
3840                we do not have the power to recompile these, we emulate
3841                the SVr4 behavior.  Sigh.  */
3842             target_mmap(0, TARGET_PAGE_SIZE, PROT_READ | PROT_EXEC,
3843                         MAP_FIXED_NOREPLACE | MAP_PRIVATE | MAP_ANONYMOUS,
3844                         -1, 0);
3845         }
3846 #ifdef TARGET_MIPS
3847         info->interp_fp_abi = interp_info.fp_abi;
3848 #endif
3849     }
3850 
3851     /*
3852      * Load a vdso if available, which will amongst other things contain the
3853      * signal trampolines.  Otherwise, allocate a separate page for them.
3854      */
3855     const VdsoImageInfo *vdso = vdso_image_info();
3856     if (vdso) {
3857         load_elf_vdso(&vdso_info, vdso);
3858         info->vdso = vdso_info.load_bias;
3859     } else if (TARGET_ARCH_HAS_SIGTRAMP_PAGE) {
3860         abi_long tramp_page = target_mmap(0, TARGET_PAGE_SIZE,
3861                                           PROT_READ | PROT_WRITE,
3862                                           MAP_PRIVATE | MAP_ANON, -1, 0);
3863         if (tramp_page == -1) {
3864             return -errno;
3865         }
3866 
3867         setup_sigtramp(tramp_page);
3868         target_mprotect(tramp_page, TARGET_PAGE_SIZE, PROT_READ | PROT_EXEC);
3869     }
3870 
3871     bprm->p = create_elf_tables(bprm->p, bprm->argc, bprm->envc, &ehdr, info,
3872                                 elf_interpreter ? &interp_info : NULL,
3873                                 vdso ? &vdso_info : NULL);
3874     info->start_stack = bprm->p;
3875 
3876     /* If we have an interpreter, set that as the program's entry point.
3877        Copy the load_bias as well, to help PPC64 interpret the entry
3878        point as a function descriptor.  Do this after creating elf tables
3879        so that we copy the original program entry point into the AUXV.  */
3880     if (elf_interpreter) {
3881         info->load_bias = interp_info.load_bias;
3882         info->entry = interp_info.entry;
3883         g_free(elf_interpreter);
3884     }
3885 
3886 #ifdef USE_ELF_CORE_DUMP
3887     bprm->core_dump = &elf_core_dump;
3888 #endif
3889 
3890     return 0;
3891 }
3892 
3893 #ifdef USE_ELF_CORE_DUMP
3894 #include "exec/translate-all.h"
3895 
3896 /*
3897  * Definitions to generate Intel SVR4-like core files.
3898  * These mostly have the same names as the SVR4 types with "target_elf_"
3899  * tacked on the front to prevent clashes with linux definitions,
3900  * and the typedef forms have been avoided.  This is mostly like
3901  * the SVR4 structure, but more Linuxy, with things that Linux does
3902  * not support and which gdb doesn't really use excluded.
3903  *
3904  * Fields we don't dump (their contents is zero) in linux-user qemu
3905  * are marked with XXX.
3906  *
3907  * Core dump code is copied from linux kernel (fs/binfmt_elf.c).
3908  *
3909  * Porting ELF coredump for target is (quite) simple process.  First you
3910  * define USE_ELF_CORE_DUMP in target ELF code (where init_thread() for
3911  * the target resides):
3912  *
3913  * #define USE_ELF_CORE_DUMP
3914  *
3915  * Next you define type of register set used for dumping.  ELF specification
3916  * says that it needs to be array of elf_greg_t that has size of ELF_NREG.
3917  *
3918  * typedef <target_regtype> target_elf_greg_t;
3919  * #define ELF_NREG <number of registers>
3920  * typedef taret_elf_greg_t target_elf_gregset_t[ELF_NREG];
3921  *
3922  * Last step is to implement target specific function that copies registers
3923  * from given cpu into just specified register set.  Prototype is:
3924  *
3925  * static void elf_core_copy_regs(taret_elf_gregset_t *regs,
3926  *                                const CPUArchState *env);
3927  *
3928  * Parameters:
3929  *     regs - copy register values into here (allocated and zeroed by caller)
3930  *     env - copy registers from here
3931  *
3932  * Example for ARM target is provided in this file.
3933  */
3934 
3935 struct target_elf_siginfo {
3936     abi_int    si_signo; /* signal number */
3937     abi_int    si_code;  /* extra code */
3938     abi_int    si_errno; /* errno */
3939 };
3940 
3941 struct target_elf_prstatus {
3942     struct target_elf_siginfo pr_info;      /* Info associated with signal */
3943     abi_short          pr_cursig;    /* Current signal */
3944     abi_ulong          pr_sigpend;   /* XXX */
3945     abi_ulong          pr_sighold;   /* XXX */
3946     target_pid_t       pr_pid;
3947     target_pid_t       pr_ppid;
3948     target_pid_t       pr_pgrp;
3949     target_pid_t       pr_sid;
3950     struct target_timeval pr_utime;  /* XXX User time */
3951     struct target_timeval pr_stime;  /* XXX System time */
3952     struct target_timeval pr_cutime; /* XXX Cumulative user time */
3953     struct target_timeval pr_cstime; /* XXX Cumulative system time */
3954     target_elf_gregset_t      pr_reg;       /* GP registers */
3955     abi_int            pr_fpvalid;   /* XXX */
3956 };
3957 
3958 #define ELF_PRARGSZ     (80) /* Number of chars for args */
3959 
3960 struct target_elf_prpsinfo {
3961     char         pr_state;       /* numeric process state */
3962     char         pr_sname;       /* char for pr_state */
3963     char         pr_zomb;        /* zombie */
3964     char         pr_nice;        /* nice val */
3965     abi_ulong    pr_flag;        /* flags */
3966     target_uid_t pr_uid;
3967     target_gid_t pr_gid;
3968     target_pid_t pr_pid, pr_ppid, pr_pgrp, pr_sid;
3969     /* Lots missing */
3970     char    pr_fname[16] QEMU_NONSTRING; /* filename of executable */
3971     char    pr_psargs[ELF_PRARGSZ]; /* initial part of arg list */
3972 };
3973 
3974 #ifdef BSWAP_NEEDED
bswap_prstatus(struct target_elf_prstatus * prstatus)3975 static void bswap_prstatus(struct target_elf_prstatus *prstatus)
3976 {
3977     prstatus->pr_info.si_signo = tswap32(prstatus->pr_info.si_signo);
3978     prstatus->pr_info.si_code = tswap32(prstatus->pr_info.si_code);
3979     prstatus->pr_info.si_errno = tswap32(prstatus->pr_info.si_errno);
3980     prstatus->pr_cursig = tswap16(prstatus->pr_cursig);
3981     prstatus->pr_sigpend = tswapal(prstatus->pr_sigpend);
3982     prstatus->pr_sighold = tswapal(prstatus->pr_sighold);
3983     prstatus->pr_pid = tswap32(prstatus->pr_pid);
3984     prstatus->pr_ppid = tswap32(prstatus->pr_ppid);
3985     prstatus->pr_pgrp = tswap32(prstatus->pr_pgrp);
3986     prstatus->pr_sid = tswap32(prstatus->pr_sid);
3987     /* cpu times are not filled, so we skip them */
3988     /* regs should be in correct format already */
3989     prstatus->pr_fpvalid = tswap32(prstatus->pr_fpvalid);
3990 }
3991 
bswap_psinfo(struct target_elf_prpsinfo * psinfo)3992 static void bswap_psinfo(struct target_elf_prpsinfo *psinfo)
3993 {
3994     psinfo->pr_flag = tswapal(psinfo->pr_flag);
3995     psinfo->pr_uid = tswap16(psinfo->pr_uid);
3996     psinfo->pr_gid = tswap16(psinfo->pr_gid);
3997     psinfo->pr_pid = tswap32(psinfo->pr_pid);
3998     psinfo->pr_ppid = tswap32(psinfo->pr_ppid);
3999     psinfo->pr_pgrp = tswap32(psinfo->pr_pgrp);
4000     psinfo->pr_sid = tswap32(psinfo->pr_sid);
4001 }
4002 
bswap_note(struct elf_note * en)4003 static void bswap_note(struct elf_note *en)
4004 {
4005     bswap32s(&en->n_namesz);
4006     bswap32s(&en->n_descsz);
4007     bswap32s(&en->n_type);
4008 }
4009 #else
bswap_prstatus(struct target_elf_prstatus * p)4010 static inline void bswap_prstatus(struct target_elf_prstatus *p) { }
bswap_psinfo(struct target_elf_prpsinfo * p)4011 static inline void bswap_psinfo(struct target_elf_prpsinfo *p) {}
bswap_note(struct elf_note * en)4012 static inline void bswap_note(struct elf_note *en) { }
4013 #endif /* BSWAP_NEEDED */
4014 
4015 /*
4016  * Calculate file (dump) size of given memory region.
4017  */
vma_dump_size(target_ulong start,target_ulong end,unsigned long flags)4018 static size_t vma_dump_size(target_ulong start, target_ulong end,
4019                             unsigned long flags)
4020 {
4021     /* The area must be readable. */
4022     if (!(flags & PAGE_READ)) {
4023         return 0;
4024     }
4025 
4026     /*
4027      * Usually we don't dump executable pages as they contain
4028      * non-writable code that debugger can read directly from
4029      * target library etc. If there is no elf header, we dump it.
4030      */
4031     if (!(flags & PAGE_WRITE_ORG) &&
4032         (flags & PAGE_EXEC) &&
4033         memcmp(g2h_untagged(start), ELFMAG, SELFMAG) == 0) {
4034         return 0;
4035     }
4036 
4037     return end - start;
4038 }
4039 
size_note(const char * name,size_t datasz)4040 static size_t size_note(const char *name, size_t datasz)
4041 {
4042     size_t namesz = strlen(name) + 1;
4043 
4044     namesz = ROUND_UP(namesz, 4);
4045     datasz = ROUND_UP(datasz, 4);
4046 
4047     return sizeof(struct elf_note) + namesz + datasz;
4048 }
4049 
fill_note(void ** pptr,int type,const char * name,size_t datasz)4050 static void *fill_note(void **pptr, int type, const char *name, size_t datasz)
4051 {
4052     void *ptr = *pptr;
4053     struct elf_note *n = ptr;
4054     size_t namesz = strlen(name) + 1;
4055 
4056     n->n_namesz = namesz;
4057     n->n_descsz = datasz;
4058     n->n_type = type;
4059     bswap_note(n);
4060 
4061     ptr += sizeof(*n);
4062     memcpy(ptr, name, namesz);
4063 
4064     namesz = ROUND_UP(namesz, 4);
4065     datasz = ROUND_UP(datasz, 4);
4066 
4067     *pptr = ptr + namesz + datasz;
4068     return ptr + namesz;
4069 }
4070 
fill_elf_header(struct elfhdr * elf,int segs,uint16_t machine,uint32_t flags)4071 static void fill_elf_header(struct elfhdr *elf, int segs, uint16_t machine,
4072                             uint32_t flags)
4073 {
4074     memcpy(elf->e_ident, ELFMAG, SELFMAG);
4075 
4076     elf->e_ident[EI_CLASS] = ELF_CLASS;
4077     elf->e_ident[EI_DATA] = ELF_DATA;
4078     elf->e_ident[EI_VERSION] = EV_CURRENT;
4079     elf->e_ident[EI_OSABI] = ELF_OSABI;
4080 
4081     elf->e_type = ET_CORE;
4082     elf->e_machine = machine;
4083     elf->e_version = EV_CURRENT;
4084     elf->e_phoff = sizeof(struct elfhdr);
4085     elf->e_flags = flags;
4086     elf->e_ehsize = sizeof(struct elfhdr);
4087     elf->e_phentsize = sizeof(struct elf_phdr);
4088     elf->e_phnum = segs;
4089 
4090     bswap_ehdr(elf);
4091 }
4092 
fill_elf_note_phdr(struct elf_phdr * phdr,size_t sz,off_t offset)4093 static void fill_elf_note_phdr(struct elf_phdr *phdr, size_t sz, off_t offset)
4094 {
4095     phdr->p_type = PT_NOTE;
4096     phdr->p_offset = offset;
4097     phdr->p_filesz = sz;
4098 
4099     bswap_phdr(phdr, 1);
4100 }
4101 
fill_prstatus_note(void * data,const TaskState * ts,CPUState * cpu,int signr)4102 static void fill_prstatus_note(void *data, const TaskState *ts,
4103                                CPUState *cpu, int signr)
4104 {
4105     /*
4106      * Because note memory is only aligned to 4, and target_elf_prstatus
4107      * may well have higher alignment requirements, fill locally and
4108      * memcpy to the destination afterward.
4109      */
4110     struct target_elf_prstatus prstatus = {
4111         .pr_info.si_signo = signr,
4112         .pr_cursig = signr,
4113         .pr_pid = ts->ts_tid,
4114         .pr_ppid = getppid(),
4115         .pr_pgrp = getpgrp(),
4116         .pr_sid = getsid(0),
4117     };
4118 
4119     elf_core_copy_regs(&prstatus.pr_reg, cpu_env(cpu));
4120     bswap_prstatus(&prstatus);
4121     memcpy(data, &prstatus, sizeof(prstatus));
4122 }
4123 
fill_prpsinfo_note(void * data,const TaskState * ts)4124 static void fill_prpsinfo_note(void *data, const TaskState *ts)
4125 {
4126     /*
4127      * Because note memory is only aligned to 4, and target_elf_prpsinfo
4128      * may well have higher alignment requirements, fill locally and
4129      * memcpy to the destination afterward.
4130      */
4131     struct target_elf_prpsinfo psinfo = {
4132         .pr_pid = getpid(),
4133         .pr_ppid = getppid(),
4134         .pr_pgrp = getpgrp(),
4135         .pr_sid = getsid(0),
4136         .pr_uid = getuid(),
4137         .pr_gid = getgid(),
4138     };
4139     char *base_filename;
4140     size_t len;
4141 
4142     len = ts->info->env_strings - ts->info->arg_strings;
4143     len = MIN(len, ELF_PRARGSZ);
4144     memcpy(&psinfo.pr_psargs, g2h_untagged(ts->info->arg_strings), len);
4145     for (size_t i = 0; i < len; i++) {
4146         if (psinfo.pr_psargs[i] == 0) {
4147             psinfo.pr_psargs[i] = ' ';
4148         }
4149     }
4150 
4151     base_filename = g_path_get_basename(ts->bprm->filename);
4152     /*
4153      * Using strncpy here is fine: at max-length,
4154      * this field is not NUL-terminated.
4155      */
4156     strncpy(psinfo.pr_fname, base_filename, sizeof(psinfo.pr_fname));
4157     g_free(base_filename);
4158 
4159     bswap_psinfo(&psinfo);
4160     memcpy(data, &psinfo, sizeof(psinfo));
4161 }
4162 
fill_auxv_note(void * data,const TaskState * ts)4163 static void fill_auxv_note(void *data, const TaskState *ts)
4164 {
4165     memcpy(data, g2h_untagged(ts->info->saved_auxv), ts->info->auxv_len);
4166 }
4167 
4168 /*
4169  * Constructs name of coredump file.  We have following convention
4170  * for the name:
4171  *     qemu_<basename-of-target-binary>_<date>-<time>_<pid>.core
4172  *
4173  * Returns the filename
4174  */
core_dump_filename(const TaskState * ts)4175 static char *core_dump_filename(const TaskState *ts)
4176 {
4177     g_autoptr(GDateTime) now = g_date_time_new_now_local();
4178     g_autofree char *nowstr = g_date_time_format(now, "%Y%m%d-%H%M%S");
4179     g_autofree char *base_filename = g_path_get_basename(ts->bprm->filename);
4180 
4181     return g_strdup_printf("qemu_%s_%s_%d.core",
4182                            base_filename, nowstr, (int)getpid());
4183 }
4184 
dump_write(int fd,const void * ptr,size_t size)4185 static int dump_write(int fd, const void *ptr, size_t size)
4186 {
4187     const char *bufp = (const char *)ptr;
4188     ssize_t bytes_written, bytes_left;
4189 
4190     bytes_written = 0;
4191     bytes_left = size;
4192 
4193     /*
4194      * In normal conditions, single write(2) should do but
4195      * in case of socket etc. this mechanism is more portable.
4196      */
4197     do {
4198         bytes_written = write(fd, bufp, bytes_left);
4199         if (bytes_written < 0) {
4200             if (errno == EINTR)
4201                 continue;
4202             return (-1);
4203         } else if (bytes_written == 0) { /* eof */
4204             return (-1);
4205         }
4206         bufp += bytes_written;
4207         bytes_left -= bytes_written;
4208     } while (bytes_left > 0);
4209 
4210     return (0);
4211 }
4212 
wmr_page_unprotect_regions(void * opaque,target_ulong start,target_ulong end,unsigned long flags)4213 static int wmr_page_unprotect_regions(void *opaque, target_ulong start,
4214                                       target_ulong end, unsigned long flags)
4215 {
4216     if ((flags & (PAGE_WRITE | PAGE_WRITE_ORG)) == PAGE_WRITE_ORG) {
4217         size_t step = MAX(TARGET_PAGE_SIZE, qemu_real_host_page_size());
4218 
4219         while (1) {
4220             page_unprotect(start, 0);
4221             if (end - start <= step) {
4222                 break;
4223             }
4224             start += step;
4225         }
4226     }
4227     return 0;
4228 }
4229 
4230 typedef struct {
4231     unsigned count;
4232     size_t size;
4233 } CountAndSizeRegions;
4234 
wmr_count_and_size_regions(void * opaque,target_ulong start,target_ulong end,unsigned long flags)4235 static int wmr_count_and_size_regions(void *opaque, target_ulong start,
4236                                       target_ulong end, unsigned long flags)
4237 {
4238     CountAndSizeRegions *css = opaque;
4239 
4240     css->count++;
4241     css->size += vma_dump_size(start, end, flags);
4242     return 0;
4243 }
4244 
4245 typedef struct {
4246     struct elf_phdr *phdr;
4247     off_t offset;
4248 } FillRegionPhdr;
4249 
wmr_fill_region_phdr(void * opaque,target_ulong start,target_ulong end,unsigned long flags)4250 static int wmr_fill_region_phdr(void *opaque, target_ulong start,
4251                                 target_ulong end, unsigned long flags)
4252 {
4253     FillRegionPhdr *d = opaque;
4254     struct elf_phdr *phdr = d->phdr;
4255 
4256     phdr->p_type = PT_LOAD;
4257     phdr->p_vaddr = start;
4258     phdr->p_paddr = 0;
4259     phdr->p_filesz = vma_dump_size(start, end, flags);
4260     phdr->p_offset = d->offset;
4261     d->offset += phdr->p_filesz;
4262     phdr->p_memsz = end - start;
4263     phdr->p_flags = (flags & PAGE_READ ? PF_R : 0)
4264                   | (flags & PAGE_WRITE_ORG ? PF_W : 0)
4265                   | (flags & PAGE_EXEC ? PF_X : 0);
4266     phdr->p_align = ELF_EXEC_PAGESIZE;
4267 
4268     bswap_phdr(phdr, 1);
4269     d->phdr = phdr + 1;
4270     return 0;
4271 }
4272 
wmr_write_region(void * opaque,target_ulong start,target_ulong end,unsigned long flags)4273 static int wmr_write_region(void *opaque, target_ulong start,
4274                             target_ulong end, unsigned long flags)
4275 {
4276     int fd = *(int *)opaque;
4277     size_t size = vma_dump_size(start, end, flags);
4278 
4279     if (!size) {
4280         return 0;
4281     }
4282     return dump_write(fd, g2h_untagged(start), size);
4283 }
4284 
4285 /*
4286  * Write out ELF coredump.
4287  *
4288  * See documentation of ELF object file format in:
4289  * http://www.caldera.com/developers/devspecs/gabi41.pdf
4290  *
4291  * Coredump format in linux is following:
4292  *
4293  * 0   +----------------------+         \
4294  *     | ELF header           | ET_CORE  |
4295  *     +----------------------+          |
4296  *     | ELF program headers  |          |--- headers
4297  *     | - NOTE section       |          |
4298  *     | - PT_LOAD sections   |          |
4299  *     +----------------------+         /
4300  *     | NOTEs:               |
4301  *     | - NT_PRSTATUS        |
4302  *     | - NT_PRSINFO         |
4303  *     | - NT_AUXV            |
4304  *     +----------------------+ <-- aligned to target page
4305  *     | Process memory dump  |
4306  *     :                      :
4307  *     .                      .
4308  *     :                      :
4309  *     |                      |
4310  *     +----------------------+
4311  *
4312  * NT_PRSTATUS -> struct elf_prstatus (per thread)
4313  * NT_PRSINFO  -> struct elf_prpsinfo
4314  * NT_AUXV is array of { type, value } pairs (see fill_auxv_note()).
4315  *
4316  * Format follows System V format as close as possible.  Current
4317  * version limitations are as follows:
4318  *     - no floating point registers are dumped
4319  *
4320  * Function returns 0 in case of success, negative errno otherwise.
4321  *
4322  * TODO: make this work also during runtime: it should be
4323  * possible to force coredump from running process and then
4324  * continue processing.  For example qemu could set up SIGUSR2
4325  * handler (provided that target process haven't registered
4326  * handler for that) that does the dump when signal is received.
4327  */
elf_core_dump(int signr,const CPUArchState * env)4328 static int elf_core_dump(int signr, const CPUArchState *env)
4329 {
4330     const CPUState *cpu = env_cpu((CPUArchState *)env);
4331     const TaskState *ts = (const TaskState *)get_task_state((CPUState *)cpu);
4332     struct rlimit dumpsize;
4333     CountAndSizeRegions css;
4334     off_t offset, note_offset, data_offset;
4335     size_t note_size;
4336     int cpus, ret;
4337     int fd = -1;
4338     CPUState *cpu_iter;
4339 
4340     if (prctl(PR_GET_DUMPABLE) == 0) {
4341         return 0;
4342     }
4343 
4344     if (getrlimit(RLIMIT_CORE, &dumpsize) < 0 || dumpsize.rlim_cur == 0) {
4345         return 0;
4346     }
4347 
4348     cpu_list_lock();
4349     mmap_lock();
4350 
4351     /* By unprotecting, we merge vmas that might be split. */
4352     walk_memory_regions(NULL, wmr_page_unprotect_regions);
4353 
4354     /*
4355      * Walk through target process memory mappings and
4356      * set up structure containing this information.
4357      */
4358     memset(&css, 0, sizeof(css));
4359     walk_memory_regions(&css, wmr_count_and_size_regions);
4360 
4361     cpus = 0;
4362     CPU_FOREACH(cpu_iter) {
4363         cpus++;
4364     }
4365 
4366     offset = sizeof(struct elfhdr);
4367     offset += (css.count + 1) * sizeof(struct elf_phdr);
4368     note_offset = offset;
4369 
4370     offset += size_note("CORE", ts->info->auxv_len);
4371     offset += size_note("CORE", sizeof(struct target_elf_prpsinfo));
4372     offset += size_note("CORE", sizeof(struct target_elf_prstatus)) * cpus;
4373     note_size = offset - note_offset;
4374     data_offset = ROUND_UP(offset, ELF_EXEC_PAGESIZE);
4375 
4376     /* Do not dump if the corefile size exceeds the limit. */
4377     if (dumpsize.rlim_cur != RLIM_INFINITY
4378         && dumpsize.rlim_cur < data_offset + css.size) {
4379         errno = 0;
4380         goto out;
4381     }
4382 
4383     {
4384         g_autofree char *corefile = core_dump_filename(ts);
4385         fd = open(corefile, O_WRONLY | O_CREAT | O_TRUNC,
4386                   S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH);
4387     }
4388     if (fd < 0) {
4389         goto out;
4390     }
4391 
4392     /*
4393      * There is a fair amount of alignment padding within the notes
4394      * as well as preceeding the process memory.  Allocate a zeroed
4395      * block to hold it all.  Write all of the headers directly into
4396      * this buffer and then write it out as a block.
4397      */
4398     {
4399         g_autofree void *header = g_malloc0(data_offset);
4400         FillRegionPhdr frp;
4401         void *hptr, *dptr;
4402 
4403         /* Create elf file header. */
4404         hptr = header;
4405         fill_elf_header(hptr, css.count + 1, ELF_MACHINE, 0);
4406         hptr += sizeof(struct elfhdr);
4407 
4408         /* Create elf program headers. */
4409         fill_elf_note_phdr(hptr, note_size, note_offset);
4410         hptr += sizeof(struct elf_phdr);
4411 
4412         frp.phdr = hptr;
4413         frp.offset = data_offset;
4414         walk_memory_regions(&frp, wmr_fill_region_phdr);
4415         hptr = frp.phdr;
4416 
4417         /* Create the notes. */
4418         dptr = fill_note(&hptr, NT_AUXV, "CORE", ts->info->auxv_len);
4419         fill_auxv_note(dptr, ts);
4420 
4421         dptr = fill_note(&hptr, NT_PRPSINFO, "CORE",
4422                          sizeof(struct target_elf_prpsinfo));
4423         fill_prpsinfo_note(dptr, ts);
4424 
4425         CPU_FOREACH(cpu_iter) {
4426             dptr = fill_note(&hptr, NT_PRSTATUS, "CORE",
4427                              sizeof(struct target_elf_prstatus));
4428             fill_prstatus_note(dptr, ts, cpu_iter,
4429                                cpu_iter == cpu ? signr : 0);
4430         }
4431 
4432         if (dump_write(fd, header, data_offset) < 0) {
4433             goto out;
4434         }
4435     }
4436 
4437     /*
4438      * Finally write process memory into the corefile as well.
4439      */
4440     if (walk_memory_regions(&fd, wmr_write_region) < 0) {
4441         goto out;
4442     }
4443     errno = 0;
4444 
4445  out:
4446     ret = -errno;
4447     mmap_unlock();
4448     cpu_list_unlock();
4449     if (fd >= 0) {
4450         close(fd);
4451     }
4452     return ret;
4453 }
4454 #endif /* USE_ELF_CORE_DUMP */
4455 
do_init_thread(struct target_pt_regs * regs,struct image_info * infop)4456 void do_init_thread(struct target_pt_regs *regs, struct image_info *infop)
4457 {
4458     init_thread(regs, infop);
4459 }
4460