xref: /openbmc/qemu/linux-user/elfload.c (revision e5d832b6)
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 #if TARGET_BIG_ENDIAN
663 #include "elf.h"
664 #include "vdso-be8.c.inc"
665 #include "vdso-be32.c.inc"
666 
vdso_image_info(uint32_t elf_flags)667 static const VdsoImageInfo *vdso_image_info(uint32_t elf_flags)
668 {
669     return (EF_ARM_EABI_VERSION(elf_flags) >= EF_ARM_EABI_VER4
670             && (elf_flags & EF_ARM_BE8)
671             ? &vdso_be8_image_info
672             : &vdso_be32_image_info);
673 }
674 #define vdso_image_info vdso_image_info
675 #else
676 # define VDSO_HEADER  "vdso-le.c.inc"
677 #endif
678 
679 #else
680 /* 64 bit ARM definitions */
681 
682 #define ELF_ARCH        EM_AARCH64
683 #define ELF_CLASS       ELFCLASS64
684 #if TARGET_BIG_ENDIAN
685 # define ELF_PLATFORM    "aarch64_be"
686 #else
687 # define ELF_PLATFORM    "aarch64"
688 #endif
689 
init_thread(struct target_pt_regs * regs,struct image_info * infop)690 static inline void init_thread(struct target_pt_regs *regs,
691                                struct image_info *infop)
692 {
693     abi_long stack = infop->start_stack;
694     memset(regs, 0, sizeof(*regs));
695 
696     regs->pc = infop->entry & ~0x3ULL;
697     regs->sp = stack;
698 }
699 
700 #define ELF_NREG    34
701 typedef target_elf_greg_t  target_elf_gregset_t[ELF_NREG];
702 
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUARMState * env)703 static void elf_core_copy_regs(target_elf_gregset_t *regs,
704                                const CPUARMState *env)
705 {
706     int i;
707 
708     for (i = 0; i < 32; i++) {
709         (*regs)[i] = tswapreg(env->xregs[i]);
710     }
711     (*regs)[32] = tswapreg(env->pc);
712     (*regs)[33] = tswapreg(pstate_read((CPUARMState *)env));
713 }
714 
715 #define USE_ELF_CORE_DUMP
716 #define ELF_EXEC_PAGESIZE       4096
717 
718 enum {
719     ARM_HWCAP_A64_FP            = 1 << 0,
720     ARM_HWCAP_A64_ASIMD         = 1 << 1,
721     ARM_HWCAP_A64_EVTSTRM       = 1 << 2,
722     ARM_HWCAP_A64_AES           = 1 << 3,
723     ARM_HWCAP_A64_PMULL         = 1 << 4,
724     ARM_HWCAP_A64_SHA1          = 1 << 5,
725     ARM_HWCAP_A64_SHA2          = 1 << 6,
726     ARM_HWCAP_A64_CRC32         = 1 << 7,
727     ARM_HWCAP_A64_ATOMICS       = 1 << 8,
728     ARM_HWCAP_A64_FPHP          = 1 << 9,
729     ARM_HWCAP_A64_ASIMDHP       = 1 << 10,
730     ARM_HWCAP_A64_CPUID         = 1 << 11,
731     ARM_HWCAP_A64_ASIMDRDM      = 1 << 12,
732     ARM_HWCAP_A64_JSCVT         = 1 << 13,
733     ARM_HWCAP_A64_FCMA          = 1 << 14,
734     ARM_HWCAP_A64_LRCPC         = 1 << 15,
735     ARM_HWCAP_A64_DCPOP         = 1 << 16,
736     ARM_HWCAP_A64_SHA3          = 1 << 17,
737     ARM_HWCAP_A64_SM3           = 1 << 18,
738     ARM_HWCAP_A64_SM4           = 1 << 19,
739     ARM_HWCAP_A64_ASIMDDP       = 1 << 20,
740     ARM_HWCAP_A64_SHA512        = 1 << 21,
741     ARM_HWCAP_A64_SVE           = 1 << 22,
742     ARM_HWCAP_A64_ASIMDFHM      = 1 << 23,
743     ARM_HWCAP_A64_DIT           = 1 << 24,
744     ARM_HWCAP_A64_USCAT         = 1 << 25,
745     ARM_HWCAP_A64_ILRCPC        = 1 << 26,
746     ARM_HWCAP_A64_FLAGM         = 1 << 27,
747     ARM_HWCAP_A64_SSBS          = 1 << 28,
748     ARM_HWCAP_A64_SB            = 1 << 29,
749     ARM_HWCAP_A64_PACA          = 1 << 30,
750     ARM_HWCAP_A64_PACG          = 1UL << 31,
751 
752     ARM_HWCAP2_A64_DCPODP       = 1 << 0,
753     ARM_HWCAP2_A64_SVE2         = 1 << 1,
754     ARM_HWCAP2_A64_SVEAES       = 1 << 2,
755     ARM_HWCAP2_A64_SVEPMULL     = 1 << 3,
756     ARM_HWCAP2_A64_SVEBITPERM   = 1 << 4,
757     ARM_HWCAP2_A64_SVESHA3      = 1 << 5,
758     ARM_HWCAP2_A64_SVESM4       = 1 << 6,
759     ARM_HWCAP2_A64_FLAGM2       = 1 << 7,
760     ARM_HWCAP2_A64_FRINT        = 1 << 8,
761     ARM_HWCAP2_A64_SVEI8MM      = 1 << 9,
762     ARM_HWCAP2_A64_SVEF32MM     = 1 << 10,
763     ARM_HWCAP2_A64_SVEF64MM     = 1 << 11,
764     ARM_HWCAP2_A64_SVEBF16      = 1 << 12,
765     ARM_HWCAP2_A64_I8MM         = 1 << 13,
766     ARM_HWCAP2_A64_BF16         = 1 << 14,
767     ARM_HWCAP2_A64_DGH          = 1 << 15,
768     ARM_HWCAP2_A64_RNG          = 1 << 16,
769     ARM_HWCAP2_A64_BTI          = 1 << 17,
770     ARM_HWCAP2_A64_MTE          = 1 << 18,
771     ARM_HWCAP2_A64_ECV          = 1 << 19,
772     ARM_HWCAP2_A64_AFP          = 1 << 20,
773     ARM_HWCAP2_A64_RPRES        = 1 << 21,
774     ARM_HWCAP2_A64_MTE3         = 1 << 22,
775     ARM_HWCAP2_A64_SME          = 1 << 23,
776     ARM_HWCAP2_A64_SME_I16I64   = 1 << 24,
777     ARM_HWCAP2_A64_SME_F64F64   = 1 << 25,
778     ARM_HWCAP2_A64_SME_I8I32    = 1 << 26,
779     ARM_HWCAP2_A64_SME_F16F32   = 1 << 27,
780     ARM_HWCAP2_A64_SME_B16F32   = 1 << 28,
781     ARM_HWCAP2_A64_SME_F32F32   = 1 << 29,
782     ARM_HWCAP2_A64_SME_FA64     = 1 << 30,
783     ARM_HWCAP2_A64_WFXT         = 1ULL << 31,
784     ARM_HWCAP2_A64_EBF16        = 1ULL << 32,
785     ARM_HWCAP2_A64_SVE_EBF16    = 1ULL << 33,
786     ARM_HWCAP2_A64_CSSC         = 1ULL << 34,
787     ARM_HWCAP2_A64_RPRFM        = 1ULL << 35,
788     ARM_HWCAP2_A64_SVE2P1       = 1ULL << 36,
789     ARM_HWCAP2_A64_SME2         = 1ULL << 37,
790     ARM_HWCAP2_A64_SME2P1       = 1ULL << 38,
791     ARM_HWCAP2_A64_SME_I16I32   = 1ULL << 39,
792     ARM_HWCAP2_A64_SME_BI32I32  = 1ULL << 40,
793     ARM_HWCAP2_A64_SME_B16B16   = 1ULL << 41,
794     ARM_HWCAP2_A64_SME_F16F16   = 1ULL << 42,
795     ARM_HWCAP2_A64_MOPS         = 1ULL << 43,
796     ARM_HWCAP2_A64_HBC          = 1ULL << 44,
797 };
798 
799 #define ELF_HWCAP   get_elf_hwcap()
800 #define ELF_HWCAP2  get_elf_hwcap2()
801 
802 #define GET_FEATURE_ID(feat, hwcap) \
803     do { if (cpu_isar_feature(feat, cpu)) { hwcaps |= hwcap; } } while (0)
804 
get_elf_hwcap(void)805 uint32_t get_elf_hwcap(void)
806 {
807     ARMCPU *cpu = ARM_CPU(thread_cpu);
808     uint32_t hwcaps = 0;
809 
810     hwcaps |= ARM_HWCAP_A64_FP;
811     hwcaps |= ARM_HWCAP_A64_ASIMD;
812     hwcaps |= ARM_HWCAP_A64_CPUID;
813 
814     /* probe for the extra features */
815 
816     GET_FEATURE_ID(aa64_aes, ARM_HWCAP_A64_AES);
817     GET_FEATURE_ID(aa64_pmull, ARM_HWCAP_A64_PMULL);
818     GET_FEATURE_ID(aa64_sha1, ARM_HWCAP_A64_SHA1);
819     GET_FEATURE_ID(aa64_sha256, ARM_HWCAP_A64_SHA2);
820     GET_FEATURE_ID(aa64_sha512, ARM_HWCAP_A64_SHA512);
821     GET_FEATURE_ID(aa64_crc32, ARM_HWCAP_A64_CRC32);
822     GET_FEATURE_ID(aa64_sha3, ARM_HWCAP_A64_SHA3);
823     GET_FEATURE_ID(aa64_sm3, ARM_HWCAP_A64_SM3);
824     GET_FEATURE_ID(aa64_sm4, ARM_HWCAP_A64_SM4);
825     GET_FEATURE_ID(aa64_fp16, ARM_HWCAP_A64_FPHP | ARM_HWCAP_A64_ASIMDHP);
826     GET_FEATURE_ID(aa64_atomics, ARM_HWCAP_A64_ATOMICS);
827     GET_FEATURE_ID(aa64_lse2, ARM_HWCAP_A64_USCAT);
828     GET_FEATURE_ID(aa64_rdm, ARM_HWCAP_A64_ASIMDRDM);
829     GET_FEATURE_ID(aa64_dp, ARM_HWCAP_A64_ASIMDDP);
830     GET_FEATURE_ID(aa64_fcma, ARM_HWCAP_A64_FCMA);
831     GET_FEATURE_ID(aa64_sve, ARM_HWCAP_A64_SVE);
832     GET_FEATURE_ID(aa64_pauth, ARM_HWCAP_A64_PACA | ARM_HWCAP_A64_PACG);
833     GET_FEATURE_ID(aa64_fhm, ARM_HWCAP_A64_ASIMDFHM);
834     GET_FEATURE_ID(aa64_dit, ARM_HWCAP_A64_DIT);
835     GET_FEATURE_ID(aa64_jscvt, ARM_HWCAP_A64_JSCVT);
836     GET_FEATURE_ID(aa64_sb, ARM_HWCAP_A64_SB);
837     GET_FEATURE_ID(aa64_condm_4, ARM_HWCAP_A64_FLAGM);
838     GET_FEATURE_ID(aa64_dcpop, ARM_HWCAP_A64_DCPOP);
839     GET_FEATURE_ID(aa64_rcpc_8_3, ARM_HWCAP_A64_LRCPC);
840     GET_FEATURE_ID(aa64_rcpc_8_4, ARM_HWCAP_A64_ILRCPC);
841 
842     return hwcaps;
843 }
844 
get_elf_hwcap2(void)845 uint64_t get_elf_hwcap2(void)
846 {
847     ARMCPU *cpu = ARM_CPU(thread_cpu);
848     uint64_t hwcaps = 0;
849 
850     GET_FEATURE_ID(aa64_dcpodp, ARM_HWCAP2_A64_DCPODP);
851     GET_FEATURE_ID(aa64_sve2, ARM_HWCAP2_A64_SVE2);
852     GET_FEATURE_ID(aa64_sve2_aes, ARM_HWCAP2_A64_SVEAES);
853     GET_FEATURE_ID(aa64_sve2_pmull128, ARM_HWCAP2_A64_SVEPMULL);
854     GET_FEATURE_ID(aa64_sve2_bitperm, ARM_HWCAP2_A64_SVEBITPERM);
855     GET_FEATURE_ID(aa64_sve2_sha3, ARM_HWCAP2_A64_SVESHA3);
856     GET_FEATURE_ID(aa64_sve2_sm4, ARM_HWCAP2_A64_SVESM4);
857     GET_FEATURE_ID(aa64_condm_5, ARM_HWCAP2_A64_FLAGM2);
858     GET_FEATURE_ID(aa64_frint, ARM_HWCAP2_A64_FRINT);
859     GET_FEATURE_ID(aa64_sve_i8mm, ARM_HWCAP2_A64_SVEI8MM);
860     GET_FEATURE_ID(aa64_sve_f32mm, ARM_HWCAP2_A64_SVEF32MM);
861     GET_FEATURE_ID(aa64_sve_f64mm, ARM_HWCAP2_A64_SVEF64MM);
862     GET_FEATURE_ID(aa64_sve_bf16, ARM_HWCAP2_A64_SVEBF16);
863     GET_FEATURE_ID(aa64_i8mm, ARM_HWCAP2_A64_I8MM);
864     GET_FEATURE_ID(aa64_bf16, ARM_HWCAP2_A64_BF16);
865     GET_FEATURE_ID(aa64_rndr, ARM_HWCAP2_A64_RNG);
866     GET_FEATURE_ID(aa64_bti, ARM_HWCAP2_A64_BTI);
867     GET_FEATURE_ID(aa64_mte, ARM_HWCAP2_A64_MTE);
868     GET_FEATURE_ID(aa64_mte3, ARM_HWCAP2_A64_MTE3);
869     GET_FEATURE_ID(aa64_sme, (ARM_HWCAP2_A64_SME |
870                               ARM_HWCAP2_A64_SME_F32F32 |
871                               ARM_HWCAP2_A64_SME_B16F32 |
872                               ARM_HWCAP2_A64_SME_F16F32 |
873                               ARM_HWCAP2_A64_SME_I8I32));
874     GET_FEATURE_ID(aa64_sme_f64f64, ARM_HWCAP2_A64_SME_F64F64);
875     GET_FEATURE_ID(aa64_sme_i16i64, ARM_HWCAP2_A64_SME_I16I64);
876     GET_FEATURE_ID(aa64_sme_fa64, ARM_HWCAP2_A64_SME_FA64);
877     GET_FEATURE_ID(aa64_hbc, ARM_HWCAP2_A64_HBC);
878     GET_FEATURE_ID(aa64_mops, ARM_HWCAP2_A64_MOPS);
879 
880     return hwcaps;
881 }
882 
elf_hwcap_str(uint32_t bit)883 const char *elf_hwcap_str(uint32_t bit)
884 {
885     static const char *hwcap_str[] = {
886     [__builtin_ctz(ARM_HWCAP_A64_FP      )] = "fp",
887     [__builtin_ctz(ARM_HWCAP_A64_ASIMD   )] = "asimd",
888     [__builtin_ctz(ARM_HWCAP_A64_EVTSTRM )] = "evtstrm",
889     [__builtin_ctz(ARM_HWCAP_A64_AES     )] = "aes",
890     [__builtin_ctz(ARM_HWCAP_A64_PMULL   )] = "pmull",
891     [__builtin_ctz(ARM_HWCAP_A64_SHA1    )] = "sha1",
892     [__builtin_ctz(ARM_HWCAP_A64_SHA2    )] = "sha2",
893     [__builtin_ctz(ARM_HWCAP_A64_CRC32   )] = "crc32",
894     [__builtin_ctz(ARM_HWCAP_A64_ATOMICS )] = "atomics",
895     [__builtin_ctz(ARM_HWCAP_A64_FPHP    )] = "fphp",
896     [__builtin_ctz(ARM_HWCAP_A64_ASIMDHP )] = "asimdhp",
897     [__builtin_ctz(ARM_HWCAP_A64_CPUID   )] = "cpuid",
898     [__builtin_ctz(ARM_HWCAP_A64_ASIMDRDM)] = "asimdrdm",
899     [__builtin_ctz(ARM_HWCAP_A64_JSCVT   )] = "jscvt",
900     [__builtin_ctz(ARM_HWCAP_A64_FCMA    )] = "fcma",
901     [__builtin_ctz(ARM_HWCAP_A64_LRCPC   )] = "lrcpc",
902     [__builtin_ctz(ARM_HWCAP_A64_DCPOP   )] = "dcpop",
903     [__builtin_ctz(ARM_HWCAP_A64_SHA3    )] = "sha3",
904     [__builtin_ctz(ARM_HWCAP_A64_SM3     )] = "sm3",
905     [__builtin_ctz(ARM_HWCAP_A64_SM4     )] = "sm4",
906     [__builtin_ctz(ARM_HWCAP_A64_ASIMDDP )] = "asimddp",
907     [__builtin_ctz(ARM_HWCAP_A64_SHA512  )] = "sha512",
908     [__builtin_ctz(ARM_HWCAP_A64_SVE     )] = "sve",
909     [__builtin_ctz(ARM_HWCAP_A64_ASIMDFHM)] = "asimdfhm",
910     [__builtin_ctz(ARM_HWCAP_A64_DIT     )] = "dit",
911     [__builtin_ctz(ARM_HWCAP_A64_USCAT   )] = "uscat",
912     [__builtin_ctz(ARM_HWCAP_A64_ILRCPC  )] = "ilrcpc",
913     [__builtin_ctz(ARM_HWCAP_A64_FLAGM   )] = "flagm",
914     [__builtin_ctz(ARM_HWCAP_A64_SSBS    )] = "ssbs",
915     [__builtin_ctz(ARM_HWCAP_A64_SB      )] = "sb",
916     [__builtin_ctz(ARM_HWCAP_A64_PACA    )] = "paca",
917     [__builtin_ctz(ARM_HWCAP_A64_PACG    )] = "pacg",
918     };
919 
920     return bit < ARRAY_SIZE(hwcap_str) ? hwcap_str[bit] : NULL;
921 }
922 
elf_hwcap2_str(uint32_t bit)923 const char *elf_hwcap2_str(uint32_t bit)
924 {
925     static const char *hwcap_str[] = {
926     [__builtin_ctz(ARM_HWCAP2_A64_DCPODP       )] = "dcpodp",
927     [__builtin_ctz(ARM_HWCAP2_A64_SVE2         )] = "sve2",
928     [__builtin_ctz(ARM_HWCAP2_A64_SVEAES       )] = "sveaes",
929     [__builtin_ctz(ARM_HWCAP2_A64_SVEPMULL     )] = "svepmull",
930     [__builtin_ctz(ARM_HWCAP2_A64_SVEBITPERM   )] = "svebitperm",
931     [__builtin_ctz(ARM_HWCAP2_A64_SVESHA3      )] = "svesha3",
932     [__builtin_ctz(ARM_HWCAP2_A64_SVESM4       )] = "svesm4",
933     [__builtin_ctz(ARM_HWCAP2_A64_FLAGM2       )] = "flagm2",
934     [__builtin_ctz(ARM_HWCAP2_A64_FRINT        )] = "frint",
935     [__builtin_ctz(ARM_HWCAP2_A64_SVEI8MM      )] = "svei8mm",
936     [__builtin_ctz(ARM_HWCAP2_A64_SVEF32MM     )] = "svef32mm",
937     [__builtin_ctz(ARM_HWCAP2_A64_SVEF64MM     )] = "svef64mm",
938     [__builtin_ctz(ARM_HWCAP2_A64_SVEBF16      )] = "svebf16",
939     [__builtin_ctz(ARM_HWCAP2_A64_I8MM         )] = "i8mm",
940     [__builtin_ctz(ARM_HWCAP2_A64_BF16         )] = "bf16",
941     [__builtin_ctz(ARM_HWCAP2_A64_DGH          )] = "dgh",
942     [__builtin_ctz(ARM_HWCAP2_A64_RNG          )] = "rng",
943     [__builtin_ctz(ARM_HWCAP2_A64_BTI          )] = "bti",
944     [__builtin_ctz(ARM_HWCAP2_A64_MTE          )] = "mte",
945     [__builtin_ctz(ARM_HWCAP2_A64_ECV          )] = "ecv",
946     [__builtin_ctz(ARM_HWCAP2_A64_AFP          )] = "afp",
947     [__builtin_ctz(ARM_HWCAP2_A64_RPRES        )] = "rpres",
948     [__builtin_ctz(ARM_HWCAP2_A64_MTE3         )] = "mte3",
949     [__builtin_ctz(ARM_HWCAP2_A64_SME          )] = "sme",
950     [__builtin_ctz(ARM_HWCAP2_A64_SME_I16I64   )] = "smei16i64",
951     [__builtin_ctz(ARM_HWCAP2_A64_SME_F64F64   )] = "smef64f64",
952     [__builtin_ctz(ARM_HWCAP2_A64_SME_I8I32    )] = "smei8i32",
953     [__builtin_ctz(ARM_HWCAP2_A64_SME_F16F32   )] = "smef16f32",
954     [__builtin_ctz(ARM_HWCAP2_A64_SME_B16F32   )] = "smeb16f32",
955     [__builtin_ctz(ARM_HWCAP2_A64_SME_F32F32   )] = "smef32f32",
956     [__builtin_ctz(ARM_HWCAP2_A64_SME_FA64     )] = "smefa64",
957     [__builtin_ctz(ARM_HWCAP2_A64_WFXT         )] = "wfxt",
958     [__builtin_ctzll(ARM_HWCAP2_A64_EBF16      )] = "ebf16",
959     [__builtin_ctzll(ARM_HWCAP2_A64_SVE_EBF16  )] = "sveebf16",
960     [__builtin_ctzll(ARM_HWCAP2_A64_CSSC       )] = "cssc",
961     [__builtin_ctzll(ARM_HWCAP2_A64_RPRFM      )] = "rprfm",
962     [__builtin_ctzll(ARM_HWCAP2_A64_SVE2P1     )] = "sve2p1",
963     [__builtin_ctzll(ARM_HWCAP2_A64_SME2       )] = "sme2",
964     [__builtin_ctzll(ARM_HWCAP2_A64_SME2P1     )] = "sme2p1",
965     [__builtin_ctzll(ARM_HWCAP2_A64_SME_I16I32 )] = "smei16i32",
966     [__builtin_ctzll(ARM_HWCAP2_A64_SME_BI32I32)] = "smebi32i32",
967     [__builtin_ctzll(ARM_HWCAP2_A64_SME_B16B16 )] = "smeb16b16",
968     [__builtin_ctzll(ARM_HWCAP2_A64_SME_F16F16 )] = "smef16f16",
969     [__builtin_ctzll(ARM_HWCAP2_A64_MOPS       )] = "mops",
970     [__builtin_ctzll(ARM_HWCAP2_A64_HBC        )] = "hbc",
971     };
972 
973     return bit < ARRAY_SIZE(hwcap_str) ? hwcap_str[bit] : NULL;
974 }
975 
976 #undef GET_FEATURE_ID
977 
978 #if TARGET_BIG_ENDIAN
979 # define VDSO_HEADER  "vdso-be.c.inc"
980 #else
981 # define VDSO_HEADER  "vdso-le.c.inc"
982 #endif
983 
984 #endif /* not TARGET_AARCH64 */
985 
986 #endif /* TARGET_ARM */
987 
988 #ifdef TARGET_SPARC
989 
990 #ifndef TARGET_SPARC64
991 # define ELF_CLASS  ELFCLASS32
992 # define ELF_ARCH   EM_SPARC
993 #elif defined(TARGET_ABI32)
994 # define ELF_CLASS  ELFCLASS32
995 # define elf_check_arch(x) ((x) == EM_SPARC32PLUS || (x) == EM_SPARC)
996 #else
997 # define ELF_CLASS  ELFCLASS64
998 # define ELF_ARCH   EM_SPARCV9
999 #endif
1000 
1001 #include "elf.h"
1002 
1003 #define ELF_HWCAP get_elf_hwcap()
1004 
get_elf_hwcap(void)1005 static uint32_t get_elf_hwcap(void)
1006 {
1007     /* There are not many sparc32 hwcap bits -- we have all of them. */
1008     uint32_t r = HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR |
1009                  HWCAP_SPARC_SWAP | HWCAP_SPARC_MULDIV;
1010 
1011 #ifdef TARGET_SPARC64
1012     CPUSPARCState *env = cpu_env(thread_cpu);
1013     uint32_t features = env->def.features;
1014 
1015     r |= HWCAP_SPARC_V9 | HWCAP_SPARC_V8PLUS;
1016     /* 32x32 multiply and divide are efficient. */
1017     r |= HWCAP_SPARC_MUL32 | HWCAP_SPARC_DIV32;
1018     /* We don't have an internal feature bit for this. */
1019     r |= HWCAP_SPARC_POPC;
1020     r |= features & CPU_FEATURE_FSMULD ? HWCAP_SPARC_FSMULD : 0;
1021     r |= features & CPU_FEATURE_VIS1 ? HWCAP_SPARC_VIS : 0;
1022     r |= features & CPU_FEATURE_VIS2 ? HWCAP_SPARC_VIS2 : 0;
1023     r |= features & CPU_FEATURE_FMAF ? HWCAP_SPARC_FMAF : 0;
1024     r |= features & CPU_FEATURE_VIS3 ? HWCAP_SPARC_VIS3 : 0;
1025     r |= features & CPU_FEATURE_IMA ? HWCAP_SPARC_IMA : 0;
1026 #endif
1027 
1028     return r;
1029 }
1030 
init_thread(struct target_pt_regs * regs,struct image_info * infop)1031 static inline void init_thread(struct target_pt_regs *regs,
1032                                struct image_info *infop)
1033 {
1034     /* Note that target_cpu_copy_regs does not read psr/tstate. */
1035     regs->pc = infop->entry;
1036     regs->npc = regs->pc + 4;
1037     regs->y = 0;
1038     regs->u_regs[14] = (infop->start_stack - 16 * sizeof(abi_ulong)
1039                         - TARGET_STACK_BIAS);
1040 }
1041 #endif /* TARGET_SPARC */
1042 
1043 #ifdef TARGET_PPC
1044 
1045 #define ELF_MACHINE    PPC_ELF_MACHINE
1046 
1047 #if defined(TARGET_PPC64)
1048 
1049 #define elf_check_arch(x) ( (x) == EM_PPC64 )
1050 
1051 #define ELF_CLASS       ELFCLASS64
1052 
1053 #else
1054 
1055 #define ELF_CLASS       ELFCLASS32
1056 #define EXSTACK_DEFAULT true
1057 
1058 #endif
1059 
1060 #define ELF_ARCH        EM_PPC
1061 
1062 /* Feature masks for the Aux Vector Hardware Capabilities (AT_HWCAP).
1063    See arch/powerpc/include/asm/cputable.h.  */
1064 enum {
1065     QEMU_PPC_FEATURE_32 = 0x80000000,
1066     QEMU_PPC_FEATURE_64 = 0x40000000,
1067     QEMU_PPC_FEATURE_601_INSTR = 0x20000000,
1068     QEMU_PPC_FEATURE_HAS_ALTIVEC = 0x10000000,
1069     QEMU_PPC_FEATURE_HAS_FPU = 0x08000000,
1070     QEMU_PPC_FEATURE_HAS_MMU = 0x04000000,
1071     QEMU_PPC_FEATURE_HAS_4xxMAC = 0x02000000,
1072     QEMU_PPC_FEATURE_UNIFIED_CACHE = 0x01000000,
1073     QEMU_PPC_FEATURE_HAS_SPE = 0x00800000,
1074     QEMU_PPC_FEATURE_HAS_EFP_SINGLE = 0x00400000,
1075     QEMU_PPC_FEATURE_HAS_EFP_DOUBLE = 0x00200000,
1076     QEMU_PPC_FEATURE_NO_TB = 0x00100000,
1077     QEMU_PPC_FEATURE_POWER4 = 0x00080000,
1078     QEMU_PPC_FEATURE_POWER5 = 0x00040000,
1079     QEMU_PPC_FEATURE_POWER5_PLUS = 0x00020000,
1080     QEMU_PPC_FEATURE_CELL = 0x00010000,
1081     QEMU_PPC_FEATURE_BOOKE = 0x00008000,
1082     QEMU_PPC_FEATURE_SMT = 0x00004000,
1083     QEMU_PPC_FEATURE_ICACHE_SNOOP = 0x00002000,
1084     QEMU_PPC_FEATURE_ARCH_2_05 = 0x00001000,
1085     QEMU_PPC_FEATURE_PA6T = 0x00000800,
1086     QEMU_PPC_FEATURE_HAS_DFP = 0x00000400,
1087     QEMU_PPC_FEATURE_POWER6_EXT = 0x00000200,
1088     QEMU_PPC_FEATURE_ARCH_2_06 = 0x00000100,
1089     QEMU_PPC_FEATURE_HAS_VSX = 0x00000080,
1090     QEMU_PPC_FEATURE_PSERIES_PERFMON_COMPAT = 0x00000040,
1091 
1092     QEMU_PPC_FEATURE_TRUE_LE = 0x00000002,
1093     QEMU_PPC_FEATURE_PPC_LE = 0x00000001,
1094 
1095     /* Feature definitions in AT_HWCAP2.  */
1096     QEMU_PPC_FEATURE2_ARCH_2_07 = 0x80000000, /* ISA 2.07 */
1097     QEMU_PPC_FEATURE2_HAS_HTM = 0x40000000, /* Hardware Transactional Memory */
1098     QEMU_PPC_FEATURE2_HAS_DSCR = 0x20000000, /* Data Stream Control Register */
1099     QEMU_PPC_FEATURE2_HAS_EBB = 0x10000000, /* Event Base Branching */
1100     QEMU_PPC_FEATURE2_HAS_ISEL = 0x08000000, /* Integer Select */
1101     QEMU_PPC_FEATURE2_HAS_TAR = 0x04000000, /* Target Address Register */
1102     QEMU_PPC_FEATURE2_VEC_CRYPTO = 0x02000000,
1103     QEMU_PPC_FEATURE2_HTM_NOSC = 0x01000000,
1104     QEMU_PPC_FEATURE2_ARCH_3_00 = 0x00800000, /* ISA 3.00 */
1105     QEMU_PPC_FEATURE2_HAS_IEEE128 = 0x00400000, /* VSX IEEE Bin Float 128-bit */
1106     QEMU_PPC_FEATURE2_DARN = 0x00200000, /* darn random number insn */
1107     QEMU_PPC_FEATURE2_SCV = 0x00100000, /* scv syscall */
1108     QEMU_PPC_FEATURE2_HTM_NO_SUSPEND = 0x00080000, /* TM w/o suspended state */
1109     QEMU_PPC_FEATURE2_ARCH_3_1 = 0x00040000, /* ISA 3.1 */
1110     QEMU_PPC_FEATURE2_MMA = 0x00020000, /* Matrix-Multiply Assist */
1111 };
1112 
1113 #define ELF_HWCAP get_elf_hwcap()
1114 
get_elf_hwcap(void)1115 static uint32_t get_elf_hwcap(void)
1116 {
1117     PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);
1118     uint32_t features = 0;
1119 
1120     /* We don't have to be terribly complete here; the high points are
1121        Altivec/FP/SPE support.  Anything else is just a bonus.  */
1122 #define GET_FEATURE(flag, feature)                                      \
1123     do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0)
1124 #define GET_FEATURE2(flags, feature) \
1125     do { \
1126         if ((cpu->env.insns_flags2 & flags) == flags) { \
1127             features |= feature; \
1128         } \
1129     } while (0)
1130     GET_FEATURE(PPC_64B, QEMU_PPC_FEATURE_64);
1131     GET_FEATURE(PPC_FLOAT, QEMU_PPC_FEATURE_HAS_FPU);
1132     GET_FEATURE(PPC_ALTIVEC, QEMU_PPC_FEATURE_HAS_ALTIVEC);
1133     GET_FEATURE(PPC_SPE, QEMU_PPC_FEATURE_HAS_SPE);
1134     GET_FEATURE(PPC_SPE_SINGLE, QEMU_PPC_FEATURE_HAS_EFP_SINGLE);
1135     GET_FEATURE(PPC_SPE_DOUBLE, QEMU_PPC_FEATURE_HAS_EFP_DOUBLE);
1136     GET_FEATURE(PPC_BOOKE, QEMU_PPC_FEATURE_BOOKE);
1137     GET_FEATURE(PPC_405_MAC, QEMU_PPC_FEATURE_HAS_4xxMAC);
1138     GET_FEATURE2(PPC2_DFP, QEMU_PPC_FEATURE_HAS_DFP);
1139     GET_FEATURE2(PPC2_VSX, QEMU_PPC_FEATURE_HAS_VSX);
1140     GET_FEATURE2((PPC2_PERM_ISA206 | PPC2_DIVE_ISA206 | PPC2_ATOMIC_ISA206 |
1141                   PPC2_FP_CVT_ISA206 | PPC2_FP_TST_ISA206),
1142                   QEMU_PPC_FEATURE_ARCH_2_06);
1143 #undef GET_FEATURE
1144 #undef GET_FEATURE2
1145 
1146     return features;
1147 }
1148 
1149 #define ELF_HWCAP2 get_elf_hwcap2()
1150 
get_elf_hwcap2(void)1151 static uint32_t get_elf_hwcap2(void)
1152 {
1153     PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);
1154     uint32_t features = 0;
1155 
1156 #define GET_FEATURE(flag, feature)                                      \
1157     do { if (cpu->env.insns_flags & flag) { features |= feature; } } while (0)
1158 #define GET_FEATURE2(flag, feature)                                      \
1159     do { if (cpu->env.insns_flags2 & flag) { features |= feature; } } while (0)
1160 
1161     GET_FEATURE(PPC_ISEL, QEMU_PPC_FEATURE2_HAS_ISEL);
1162     GET_FEATURE2(PPC2_BCTAR_ISA207, QEMU_PPC_FEATURE2_HAS_TAR);
1163     GET_FEATURE2((PPC2_BCTAR_ISA207 | PPC2_LSQ_ISA207 | PPC2_ALTIVEC_207 |
1164                   PPC2_ISA207S), QEMU_PPC_FEATURE2_ARCH_2_07 |
1165                   QEMU_PPC_FEATURE2_VEC_CRYPTO);
1166     GET_FEATURE2(PPC2_ISA300, QEMU_PPC_FEATURE2_ARCH_3_00 |
1167                  QEMU_PPC_FEATURE2_DARN | QEMU_PPC_FEATURE2_HAS_IEEE128);
1168     GET_FEATURE2(PPC2_ISA310, QEMU_PPC_FEATURE2_ARCH_3_1 |
1169                  QEMU_PPC_FEATURE2_MMA);
1170 
1171 #undef GET_FEATURE
1172 #undef GET_FEATURE2
1173 
1174     return features;
1175 }
1176 
1177 /*
1178  * The requirements here are:
1179  * - keep the final alignment of sp (sp & 0xf)
1180  * - make sure the 32-bit value at the first 16 byte aligned position of
1181  *   AUXV is greater than 16 for glibc compatibility.
1182  *   AT_IGNOREPPC is used for that.
1183  * - for compatibility with glibc ARCH_DLINFO must always be defined on PPC,
1184  *   even if DLINFO_ARCH_ITEMS goes to zero or is undefined.
1185  */
1186 #define DLINFO_ARCH_ITEMS       5
1187 #define ARCH_DLINFO                                     \
1188     do {                                                \
1189         PowerPCCPU *cpu = POWERPC_CPU(thread_cpu);              \
1190         /*                                              \
1191          * Handle glibc compatibility: these magic entries must \
1192          * be at the lowest addresses in the final auxv.        \
1193          */                                             \
1194         NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC);        \
1195         NEW_AUX_ENT(AT_IGNOREPPC, AT_IGNOREPPC);        \
1196         NEW_AUX_ENT(AT_DCACHEBSIZE, cpu->env.dcache_line_size); \
1197         NEW_AUX_ENT(AT_ICACHEBSIZE, cpu->env.icache_line_size); \
1198         NEW_AUX_ENT(AT_UCACHEBSIZE, 0);                 \
1199     } while (0)
1200 
init_thread(struct target_pt_regs * _regs,struct image_info * infop)1201 static inline void init_thread(struct target_pt_regs *_regs, struct image_info *infop)
1202 {
1203     _regs->gpr[1] = infop->start_stack;
1204 #if defined(TARGET_PPC64)
1205     if (get_ppc64_abi(infop) < 2) {
1206         uint64_t val;
1207         get_user_u64(val, infop->entry + 8);
1208         _regs->gpr[2] = val + infop->load_bias;
1209         get_user_u64(val, infop->entry);
1210         infop->entry = val + infop->load_bias;
1211     } else {
1212         _regs->gpr[12] = infop->entry;  /* r12 set to global entry address */
1213     }
1214 #endif
1215     _regs->nip = infop->entry;
1216 }
1217 
1218 /* See linux kernel: arch/powerpc/include/asm/elf.h.  */
1219 #define ELF_NREG 48
1220 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1221 
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUPPCState * env)1222 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUPPCState *env)
1223 {
1224     int i;
1225     target_ulong ccr = 0;
1226 
1227     for (i = 0; i < ARRAY_SIZE(env->gpr); i++) {
1228         (*regs)[i] = tswapreg(env->gpr[i]);
1229     }
1230 
1231     (*regs)[32] = tswapreg(env->nip);
1232     (*regs)[33] = tswapreg(env->msr);
1233     (*regs)[35] = tswapreg(env->ctr);
1234     (*regs)[36] = tswapreg(env->lr);
1235     (*regs)[37] = tswapreg(cpu_read_xer(env));
1236 
1237     ccr = ppc_get_cr(env);
1238     (*regs)[38] = tswapreg(ccr);
1239 }
1240 
1241 #define USE_ELF_CORE_DUMP
1242 #define ELF_EXEC_PAGESIZE       4096
1243 
1244 #ifndef TARGET_PPC64
1245 # define VDSO_HEADER  "vdso-32.c.inc"
1246 #elif TARGET_BIG_ENDIAN
1247 # define VDSO_HEADER  "vdso-64.c.inc"
1248 #else
1249 # define VDSO_HEADER  "vdso-64le.c.inc"
1250 #endif
1251 
1252 #endif
1253 
1254 #ifdef TARGET_LOONGARCH64
1255 
1256 #define ELF_CLASS   ELFCLASS64
1257 #define ELF_ARCH    EM_LOONGARCH
1258 #define EXSTACK_DEFAULT true
1259 
1260 #define elf_check_arch(x) ((x) == EM_LOONGARCH)
1261 
1262 #define VDSO_HEADER "vdso.c.inc"
1263 
init_thread(struct target_pt_regs * regs,struct image_info * infop)1264 static inline void init_thread(struct target_pt_regs *regs,
1265                                struct image_info *infop)
1266 {
1267     /*Set crmd PG,DA = 1,0 */
1268     regs->csr.crmd = 2 << 3;
1269     regs->csr.era = infop->entry;
1270     regs->regs[3] = infop->start_stack;
1271 }
1272 
1273 /* See linux kernel: arch/loongarch/include/asm/elf.h */
1274 #define ELF_NREG 45
1275 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1276 
1277 enum {
1278     TARGET_EF_R0 = 0,
1279     TARGET_EF_CSR_ERA = TARGET_EF_R0 + 33,
1280     TARGET_EF_CSR_BADV = TARGET_EF_R0 + 34,
1281 };
1282 
elf_core_copy_regs(target_elf_gregset_t * regs,const CPULoongArchState * env)1283 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1284                                const CPULoongArchState *env)
1285 {
1286     int i;
1287 
1288     (*regs)[TARGET_EF_R0] = 0;
1289 
1290     for (i = 1; i < ARRAY_SIZE(env->gpr); i++) {
1291         (*regs)[TARGET_EF_R0 + i] = tswapreg(env->gpr[i]);
1292     }
1293 
1294     (*regs)[TARGET_EF_CSR_ERA] = tswapreg(env->pc);
1295     (*regs)[TARGET_EF_CSR_BADV] = tswapreg(env->CSR_BADV);
1296 }
1297 
1298 #define USE_ELF_CORE_DUMP
1299 #define ELF_EXEC_PAGESIZE        4096
1300 
1301 #define ELF_HWCAP get_elf_hwcap()
1302 
1303 /* See arch/loongarch/include/uapi/asm/hwcap.h */
1304 enum {
1305     HWCAP_LOONGARCH_CPUCFG   = (1 << 0),
1306     HWCAP_LOONGARCH_LAM      = (1 << 1),
1307     HWCAP_LOONGARCH_UAL      = (1 << 2),
1308     HWCAP_LOONGARCH_FPU      = (1 << 3),
1309     HWCAP_LOONGARCH_LSX      = (1 << 4),
1310     HWCAP_LOONGARCH_LASX     = (1 << 5),
1311     HWCAP_LOONGARCH_CRC32    = (1 << 6),
1312     HWCAP_LOONGARCH_COMPLEX  = (1 << 7),
1313     HWCAP_LOONGARCH_CRYPTO   = (1 << 8),
1314     HWCAP_LOONGARCH_LVZ      = (1 << 9),
1315     HWCAP_LOONGARCH_LBT_X86  = (1 << 10),
1316     HWCAP_LOONGARCH_LBT_ARM  = (1 << 11),
1317     HWCAP_LOONGARCH_LBT_MIPS = (1 << 12),
1318 };
1319 
get_elf_hwcap(void)1320 static uint32_t get_elf_hwcap(void)
1321 {
1322     LoongArchCPU *cpu = LOONGARCH_CPU(thread_cpu);
1323     uint32_t hwcaps = 0;
1324 
1325     hwcaps |= HWCAP_LOONGARCH_CRC32;
1326 
1327     if (FIELD_EX32(cpu->env.cpucfg[1], CPUCFG1, UAL)) {
1328         hwcaps |= HWCAP_LOONGARCH_UAL;
1329     }
1330 
1331     if (FIELD_EX32(cpu->env.cpucfg[2], CPUCFG2, FP)) {
1332         hwcaps |= HWCAP_LOONGARCH_FPU;
1333     }
1334 
1335     if (FIELD_EX32(cpu->env.cpucfg[2], CPUCFG2, LAM)) {
1336         hwcaps |= HWCAP_LOONGARCH_LAM;
1337     }
1338 
1339     if (FIELD_EX32(cpu->env.cpucfg[2], CPUCFG2, LSX)) {
1340         hwcaps |= HWCAP_LOONGARCH_LSX;
1341     }
1342 
1343     if (FIELD_EX32(cpu->env.cpucfg[2], CPUCFG2, LASX)) {
1344         hwcaps |= HWCAP_LOONGARCH_LASX;
1345     }
1346 
1347     return hwcaps;
1348 }
1349 
1350 #define ELF_PLATFORM "loongarch"
1351 
1352 #endif /* TARGET_LOONGARCH64 */
1353 
1354 #ifdef TARGET_MIPS
1355 
1356 #ifdef TARGET_MIPS64
1357 #define ELF_CLASS   ELFCLASS64
1358 #else
1359 #define ELF_CLASS   ELFCLASS32
1360 #endif
1361 #define ELF_ARCH    EM_MIPS
1362 #define EXSTACK_DEFAULT true
1363 
1364 #ifdef TARGET_ABI_MIPSN32
1365 #define elf_check_abi(x) ((x) & EF_MIPS_ABI2)
1366 #else
1367 #define elf_check_abi(x) (!((x) & EF_MIPS_ABI2))
1368 #endif
1369 
1370 #define ELF_BASE_PLATFORM get_elf_base_platform()
1371 
1372 #define MATCH_PLATFORM_INSN(_flags, _base_platform)      \
1373     do { if ((cpu->env.insn_flags & (_flags)) == _flags) \
1374     { return _base_platform; } } while (0)
1375 
get_elf_base_platform(void)1376 static const char *get_elf_base_platform(void)
1377 {
1378     MIPSCPU *cpu = MIPS_CPU(thread_cpu);
1379 
1380     /* 64 bit ISAs goes first */
1381     MATCH_PLATFORM_INSN(CPU_MIPS64R6, "mips64r6");
1382     MATCH_PLATFORM_INSN(CPU_MIPS64R5, "mips64r5");
1383     MATCH_PLATFORM_INSN(CPU_MIPS64R2, "mips64r2");
1384     MATCH_PLATFORM_INSN(CPU_MIPS64R1, "mips64");
1385     MATCH_PLATFORM_INSN(CPU_MIPS5, "mips5");
1386     MATCH_PLATFORM_INSN(CPU_MIPS4, "mips4");
1387     MATCH_PLATFORM_INSN(CPU_MIPS3, "mips3");
1388 
1389     /* 32 bit ISAs */
1390     MATCH_PLATFORM_INSN(CPU_MIPS32R6, "mips32r6");
1391     MATCH_PLATFORM_INSN(CPU_MIPS32R5, "mips32r5");
1392     MATCH_PLATFORM_INSN(CPU_MIPS32R2, "mips32r2");
1393     MATCH_PLATFORM_INSN(CPU_MIPS32R1, "mips32");
1394     MATCH_PLATFORM_INSN(CPU_MIPS2, "mips2");
1395 
1396     /* Fallback */
1397     return "mips";
1398 }
1399 #undef MATCH_PLATFORM_INSN
1400 
init_thread(struct target_pt_regs * regs,struct image_info * infop)1401 static inline void init_thread(struct target_pt_regs *regs,
1402                                struct image_info *infop)
1403 {
1404     regs->cp0_status = 2 << CP0St_KSU;
1405     regs->cp0_epc = infop->entry;
1406     regs->regs[29] = infop->start_stack;
1407 }
1408 
1409 /* See linux kernel: arch/mips/include/asm/elf.h.  */
1410 #define ELF_NREG 45
1411 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1412 
1413 /* See linux kernel: arch/mips/include/asm/reg.h.  */
1414 enum {
1415 #ifdef TARGET_MIPS64
1416     TARGET_EF_R0 = 0,
1417 #else
1418     TARGET_EF_R0 = 6,
1419 #endif
1420     TARGET_EF_R26 = TARGET_EF_R0 + 26,
1421     TARGET_EF_R27 = TARGET_EF_R0 + 27,
1422     TARGET_EF_LO = TARGET_EF_R0 + 32,
1423     TARGET_EF_HI = TARGET_EF_R0 + 33,
1424     TARGET_EF_CP0_EPC = TARGET_EF_R0 + 34,
1425     TARGET_EF_CP0_BADVADDR = TARGET_EF_R0 + 35,
1426     TARGET_EF_CP0_STATUS = TARGET_EF_R0 + 36,
1427     TARGET_EF_CP0_CAUSE = TARGET_EF_R0 + 37
1428 };
1429 
1430 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs.  */
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUMIPSState * env)1431 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMIPSState *env)
1432 {
1433     int i;
1434 
1435     for (i = 0; i < TARGET_EF_R0; i++) {
1436         (*regs)[i] = 0;
1437     }
1438     (*regs)[TARGET_EF_R0] = 0;
1439 
1440     for (i = 1; i < ARRAY_SIZE(env->active_tc.gpr); i++) {
1441         (*regs)[TARGET_EF_R0 + i] = tswapreg(env->active_tc.gpr[i]);
1442     }
1443 
1444     (*regs)[TARGET_EF_R26] = 0;
1445     (*regs)[TARGET_EF_R27] = 0;
1446     (*regs)[TARGET_EF_LO] = tswapreg(env->active_tc.LO[0]);
1447     (*regs)[TARGET_EF_HI] = tswapreg(env->active_tc.HI[0]);
1448     (*regs)[TARGET_EF_CP0_EPC] = tswapreg(env->active_tc.PC);
1449     (*regs)[TARGET_EF_CP0_BADVADDR] = tswapreg(env->CP0_BadVAddr);
1450     (*regs)[TARGET_EF_CP0_STATUS] = tswapreg(env->CP0_Status);
1451     (*regs)[TARGET_EF_CP0_CAUSE] = tswapreg(env->CP0_Cause);
1452 }
1453 
1454 #define USE_ELF_CORE_DUMP
1455 #define ELF_EXEC_PAGESIZE        4096
1456 
1457 /* See arch/mips/include/uapi/asm/hwcap.h.  */
1458 enum {
1459     HWCAP_MIPS_R6           = (1 << 0),
1460     HWCAP_MIPS_MSA          = (1 << 1),
1461     HWCAP_MIPS_CRC32        = (1 << 2),
1462     HWCAP_MIPS_MIPS16       = (1 << 3),
1463     HWCAP_MIPS_MDMX         = (1 << 4),
1464     HWCAP_MIPS_MIPS3D       = (1 << 5),
1465     HWCAP_MIPS_SMARTMIPS    = (1 << 6),
1466     HWCAP_MIPS_DSP          = (1 << 7),
1467     HWCAP_MIPS_DSP2         = (1 << 8),
1468     HWCAP_MIPS_DSP3         = (1 << 9),
1469     HWCAP_MIPS_MIPS16E2     = (1 << 10),
1470     HWCAP_LOONGSON_MMI      = (1 << 11),
1471     HWCAP_LOONGSON_EXT      = (1 << 12),
1472     HWCAP_LOONGSON_EXT2     = (1 << 13),
1473     HWCAP_LOONGSON_CPUCFG   = (1 << 14),
1474 };
1475 
1476 #define ELF_HWCAP get_elf_hwcap()
1477 
1478 #define GET_FEATURE_INSN(_flag, _hwcap) \
1479     do { if (cpu->env.insn_flags & (_flag)) { hwcaps |= _hwcap; } } while (0)
1480 
1481 #define GET_FEATURE_REG_SET(_reg, _mask, _hwcap) \
1482     do { if (cpu->env._reg & (_mask)) { hwcaps |= _hwcap; } } while (0)
1483 
1484 #define GET_FEATURE_REG_EQU(_reg, _start, _length, _val, _hwcap) \
1485     do { \
1486         if (extract32(cpu->env._reg, (_start), (_length)) == (_val)) { \
1487             hwcaps |= _hwcap; \
1488         } \
1489     } while (0)
1490 
get_elf_hwcap(void)1491 static uint32_t get_elf_hwcap(void)
1492 {
1493     MIPSCPU *cpu = MIPS_CPU(thread_cpu);
1494     uint32_t hwcaps = 0;
1495 
1496     GET_FEATURE_REG_EQU(CP0_Config0, CP0C0_AR, CP0C0_AR_LENGTH,
1497                         2, HWCAP_MIPS_R6);
1498     GET_FEATURE_REG_SET(CP0_Config3, 1 << CP0C3_MSAP, HWCAP_MIPS_MSA);
1499     GET_FEATURE_INSN(ASE_LMMI, HWCAP_LOONGSON_MMI);
1500     GET_FEATURE_INSN(ASE_LEXT, HWCAP_LOONGSON_EXT);
1501 
1502     return hwcaps;
1503 }
1504 
1505 #undef GET_FEATURE_REG_EQU
1506 #undef GET_FEATURE_REG_SET
1507 #undef GET_FEATURE_INSN
1508 
1509 #endif /* TARGET_MIPS */
1510 
1511 #ifdef TARGET_MICROBLAZE
1512 
1513 #define elf_check_arch(x) ( (x) == EM_MICROBLAZE || (x) == EM_MICROBLAZE_OLD)
1514 
1515 #define ELF_CLASS   ELFCLASS32
1516 #define ELF_ARCH    EM_MICROBLAZE
1517 
init_thread(struct target_pt_regs * regs,struct image_info * infop)1518 static inline void init_thread(struct target_pt_regs *regs,
1519                                struct image_info *infop)
1520 {
1521     regs->pc = infop->entry;
1522     regs->r1 = infop->start_stack;
1523 
1524 }
1525 
1526 #define ELF_EXEC_PAGESIZE        4096
1527 
1528 #define USE_ELF_CORE_DUMP
1529 #define ELF_NREG 38
1530 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1531 
1532 /* See linux kernel: arch/mips/kernel/process.c:elf_dump_regs.  */
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUMBState * env)1533 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUMBState *env)
1534 {
1535     int i, pos = 0;
1536 
1537     for (i = 0; i < 32; i++) {
1538         (*regs)[pos++] = tswapreg(env->regs[i]);
1539     }
1540 
1541     (*regs)[pos++] = tswapreg(env->pc);
1542     (*regs)[pos++] = tswapreg(mb_cpu_read_msr(env));
1543     (*regs)[pos++] = 0;
1544     (*regs)[pos++] = tswapreg(env->ear);
1545     (*regs)[pos++] = 0;
1546     (*regs)[pos++] = tswapreg(env->esr);
1547 }
1548 
1549 #endif /* TARGET_MICROBLAZE */
1550 
1551 #ifdef TARGET_OPENRISC
1552 
1553 #define ELF_ARCH EM_OPENRISC
1554 #define ELF_CLASS ELFCLASS32
1555 #define ELF_DATA  ELFDATA2MSB
1556 
init_thread(struct target_pt_regs * regs,struct image_info * infop)1557 static inline void init_thread(struct target_pt_regs *regs,
1558                                struct image_info *infop)
1559 {
1560     regs->pc = infop->entry;
1561     regs->gpr[1] = infop->start_stack;
1562 }
1563 
1564 #define USE_ELF_CORE_DUMP
1565 #define ELF_EXEC_PAGESIZE 8192
1566 
1567 /* See linux kernel arch/openrisc/include/asm/elf.h.  */
1568 #define ELF_NREG 34 /* gprs and pc, sr */
1569 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1570 
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUOpenRISCState * env)1571 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1572                                const CPUOpenRISCState *env)
1573 {
1574     int i;
1575 
1576     for (i = 0; i < 32; i++) {
1577         (*regs)[i] = tswapreg(cpu_get_gpr(env, i));
1578     }
1579     (*regs)[32] = tswapreg(env->pc);
1580     (*regs)[33] = tswapreg(cpu_get_sr(env));
1581 }
1582 #define ELF_HWCAP 0
1583 #define ELF_PLATFORM NULL
1584 
1585 #endif /* TARGET_OPENRISC */
1586 
1587 #ifdef TARGET_SH4
1588 
1589 #define ELF_CLASS ELFCLASS32
1590 #define ELF_ARCH  EM_SH
1591 
init_thread(struct target_pt_regs * regs,struct image_info * infop)1592 static inline void init_thread(struct target_pt_regs *regs,
1593                                struct image_info *infop)
1594 {
1595     /* Check other registers XXXXX */
1596     regs->pc = infop->entry;
1597     regs->regs[15] = infop->start_stack;
1598 }
1599 
1600 /* See linux kernel: arch/sh/include/asm/elf.h.  */
1601 #define ELF_NREG 23
1602 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1603 
1604 /* See linux kernel: arch/sh/include/asm/ptrace.h.  */
1605 enum {
1606     TARGET_REG_PC = 16,
1607     TARGET_REG_PR = 17,
1608     TARGET_REG_SR = 18,
1609     TARGET_REG_GBR = 19,
1610     TARGET_REG_MACH = 20,
1611     TARGET_REG_MACL = 21,
1612     TARGET_REG_SYSCALL = 22
1613 };
1614 
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUSH4State * env)1615 static inline void elf_core_copy_regs(target_elf_gregset_t *regs,
1616                                       const CPUSH4State *env)
1617 {
1618     int i;
1619 
1620     for (i = 0; i < 16; i++) {
1621         (*regs)[i] = tswapreg(env->gregs[i]);
1622     }
1623 
1624     (*regs)[TARGET_REG_PC] = tswapreg(env->pc);
1625     (*regs)[TARGET_REG_PR] = tswapreg(env->pr);
1626     (*regs)[TARGET_REG_SR] = tswapreg(env->sr);
1627     (*regs)[TARGET_REG_GBR] = tswapreg(env->gbr);
1628     (*regs)[TARGET_REG_MACH] = tswapreg(env->mach);
1629     (*regs)[TARGET_REG_MACL] = tswapreg(env->macl);
1630     (*regs)[TARGET_REG_SYSCALL] = 0; /* FIXME */
1631 }
1632 
1633 #define USE_ELF_CORE_DUMP
1634 #define ELF_EXEC_PAGESIZE        4096
1635 
1636 enum {
1637     SH_CPU_HAS_FPU            = 0x0001, /* Hardware FPU support */
1638     SH_CPU_HAS_P2_FLUSH_BUG   = 0x0002, /* Need to flush the cache in P2 area */
1639     SH_CPU_HAS_MMU_PAGE_ASSOC = 0x0004, /* SH3: TLB way selection bit support */
1640     SH_CPU_HAS_DSP            = 0x0008, /* SH-DSP: DSP support */
1641     SH_CPU_HAS_PERF_COUNTER   = 0x0010, /* Hardware performance counters */
1642     SH_CPU_HAS_PTEA           = 0x0020, /* PTEA register */
1643     SH_CPU_HAS_LLSC           = 0x0040, /* movli.l/movco.l */
1644     SH_CPU_HAS_L2_CACHE       = 0x0080, /* Secondary cache / URAM */
1645     SH_CPU_HAS_OP32           = 0x0100, /* 32-bit instruction support */
1646     SH_CPU_HAS_PTEAEX         = 0x0200, /* PTE ASID Extension support */
1647 };
1648 
1649 #define ELF_HWCAP get_elf_hwcap()
1650 
get_elf_hwcap(void)1651 static uint32_t get_elf_hwcap(void)
1652 {
1653     SuperHCPU *cpu = SUPERH_CPU(thread_cpu);
1654     uint32_t hwcap = 0;
1655 
1656     hwcap |= SH_CPU_HAS_FPU;
1657 
1658     if (cpu->env.features & SH_FEATURE_SH4A) {
1659         hwcap |= SH_CPU_HAS_LLSC;
1660     }
1661 
1662     return hwcap;
1663 }
1664 
1665 #endif
1666 
1667 #ifdef TARGET_CRIS
1668 
1669 #define ELF_CLASS ELFCLASS32
1670 #define ELF_ARCH  EM_CRIS
1671 
init_thread(struct target_pt_regs * regs,struct image_info * infop)1672 static inline void init_thread(struct target_pt_regs *regs,
1673                                struct image_info *infop)
1674 {
1675     regs->erp = infop->entry;
1676 }
1677 
1678 #define ELF_EXEC_PAGESIZE        8192
1679 
1680 #endif
1681 
1682 #ifdef TARGET_M68K
1683 
1684 #define ELF_CLASS       ELFCLASS32
1685 #define ELF_ARCH        EM_68K
1686 
1687 /* ??? Does this need to do anything?
1688    #define ELF_PLAT_INIT(_r) */
1689 
init_thread(struct target_pt_regs * regs,struct image_info * infop)1690 static inline void init_thread(struct target_pt_regs *regs,
1691                                struct image_info *infop)
1692 {
1693     regs->usp = infop->start_stack;
1694     regs->sr = 0;
1695     regs->pc = infop->entry;
1696 }
1697 
1698 /* See linux kernel: arch/m68k/include/asm/elf.h.  */
1699 #define ELF_NREG 20
1700 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1701 
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUM68KState * env)1702 static void elf_core_copy_regs(target_elf_gregset_t *regs, const CPUM68KState *env)
1703 {
1704     (*regs)[0] = tswapreg(env->dregs[1]);
1705     (*regs)[1] = tswapreg(env->dregs[2]);
1706     (*regs)[2] = tswapreg(env->dregs[3]);
1707     (*regs)[3] = tswapreg(env->dregs[4]);
1708     (*regs)[4] = tswapreg(env->dregs[5]);
1709     (*regs)[5] = tswapreg(env->dregs[6]);
1710     (*regs)[6] = tswapreg(env->dregs[7]);
1711     (*regs)[7] = tswapreg(env->aregs[0]);
1712     (*regs)[8] = tswapreg(env->aregs[1]);
1713     (*regs)[9] = tswapreg(env->aregs[2]);
1714     (*regs)[10] = tswapreg(env->aregs[3]);
1715     (*regs)[11] = tswapreg(env->aregs[4]);
1716     (*regs)[12] = tswapreg(env->aregs[5]);
1717     (*regs)[13] = tswapreg(env->aregs[6]);
1718     (*regs)[14] = tswapreg(env->dregs[0]);
1719     (*regs)[15] = tswapreg(env->aregs[7]);
1720     (*regs)[16] = tswapreg(env->dregs[0]); /* FIXME: orig_d0 */
1721     (*regs)[17] = tswapreg(env->sr);
1722     (*regs)[18] = tswapreg(env->pc);
1723     (*regs)[19] = 0;  /* FIXME: regs->format | regs->vector */
1724 }
1725 
1726 #define USE_ELF_CORE_DUMP
1727 #define ELF_EXEC_PAGESIZE       8192
1728 
1729 #endif
1730 
1731 #ifdef TARGET_ALPHA
1732 
1733 #define ELF_CLASS      ELFCLASS64
1734 #define ELF_ARCH       EM_ALPHA
1735 
init_thread(struct target_pt_regs * regs,struct image_info * infop)1736 static inline void init_thread(struct target_pt_regs *regs,
1737                                struct image_info *infop)
1738 {
1739     regs->pc = infop->entry;
1740     regs->ps = 8;
1741     regs->usp = infop->start_stack;
1742 }
1743 
1744 #define ELF_EXEC_PAGESIZE        8192
1745 
1746 #endif /* TARGET_ALPHA */
1747 
1748 #ifdef TARGET_S390X
1749 
1750 #define ELF_CLASS	ELFCLASS64
1751 #define ELF_DATA	ELFDATA2MSB
1752 #define ELF_ARCH	EM_S390
1753 
1754 #include "elf.h"
1755 
1756 #define ELF_HWCAP get_elf_hwcap()
1757 
1758 #define GET_FEATURE(_feat, _hwcap) \
1759     do { if (s390_has_feat(_feat)) { hwcap |= _hwcap; } } while (0)
1760 
get_elf_hwcap(void)1761 uint32_t get_elf_hwcap(void)
1762 {
1763     /*
1764      * Let's assume we always have esan3 and zarch.
1765      * 31-bit processes can use 64-bit registers (high gprs).
1766      */
1767     uint32_t hwcap = HWCAP_S390_ESAN3 | HWCAP_S390_ZARCH | HWCAP_S390_HIGH_GPRS;
1768 
1769     GET_FEATURE(S390_FEAT_STFLE, HWCAP_S390_STFLE);
1770     GET_FEATURE(S390_FEAT_MSA, HWCAP_S390_MSA);
1771     GET_FEATURE(S390_FEAT_LONG_DISPLACEMENT, HWCAP_S390_LDISP);
1772     GET_FEATURE(S390_FEAT_EXTENDED_IMMEDIATE, HWCAP_S390_EIMM);
1773     if (s390_has_feat(S390_FEAT_EXTENDED_TRANSLATION_3) &&
1774         s390_has_feat(S390_FEAT_ETF3_ENH)) {
1775         hwcap |= HWCAP_S390_ETF3EH;
1776     }
1777     GET_FEATURE(S390_FEAT_VECTOR, HWCAP_S390_VXRS);
1778     GET_FEATURE(S390_FEAT_VECTOR_ENH, HWCAP_S390_VXRS_EXT);
1779     GET_FEATURE(S390_FEAT_VECTOR_ENH2, HWCAP_S390_VXRS_EXT2);
1780 
1781     return hwcap;
1782 }
1783 
elf_hwcap_str(uint32_t bit)1784 const char *elf_hwcap_str(uint32_t bit)
1785 {
1786     static const char *hwcap_str[] = {
1787         [HWCAP_S390_NR_ESAN3]     = "esan3",
1788         [HWCAP_S390_NR_ZARCH]     = "zarch",
1789         [HWCAP_S390_NR_STFLE]     = "stfle",
1790         [HWCAP_S390_NR_MSA]       = "msa",
1791         [HWCAP_S390_NR_LDISP]     = "ldisp",
1792         [HWCAP_S390_NR_EIMM]      = "eimm",
1793         [HWCAP_S390_NR_DFP]       = "dfp",
1794         [HWCAP_S390_NR_HPAGE]     = "edat",
1795         [HWCAP_S390_NR_ETF3EH]    = "etf3eh",
1796         [HWCAP_S390_NR_HIGH_GPRS] = "highgprs",
1797         [HWCAP_S390_NR_TE]        = "te",
1798         [HWCAP_S390_NR_VXRS]      = "vx",
1799         [HWCAP_S390_NR_VXRS_BCD]  = "vxd",
1800         [HWCAP_S390_NR_VXRS_EXT]  = "vxe",
1801         [HWCAP_S390_NR_GS]        = "gs",
1802         [HWCAP_S390_NR_VXRS_EXT2] = "vxe2",
1803         [HWCAP_S390_NR_VXRS_PDE]  = "vxp",
1804         [HWCAP_S390_NR_SORT]      = "sort",
1805         [HWCAP_S390_NR_DFLT]      = "dflt",
1806         [HWCAP_S390_NR_NNPA]      = "nnpa",
1807         [HWCAP_S390_NR_PCI_MIO]   = "pcimio",
1808         [HWCAP_S390_NR_SIE]       = "sie",
1809     };
1810 
1811     return bit < ARRAY_SIZE(hwcap_str) ? hwcap_str[bit] : NULL;
1812 }
1813 
init_thread(struct target_pt_regs * regs,struct image_info * infop)1814 static inline void init_thread(struct target_pt_regs *regs, struct image_info *infop)
1815 {
1816     regs->psw.addr = infop->entry;
1817     regs->psw.mask = PSW_MASK_DAT | PSW_MASK_IO | PSW_MASK_EXT | \
1818                      PSW_MASK_MCHECK | PSW_MASK_PSTATE | PSW_MASK_64 | \
1819                      PSW_MASK_32;
1820     regs->gprs[15] = infop->start_stack;
1821 }
1822 
1823 /* See linux kernel: arch/s390/include/uapi/asm/ptrace.h (s390_regs).  */
1824 #define ELF_NREG 27
1825 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1826 
1827 enum {
1828     TARGET_REG_PSWM = 0,
1829     TARGET_REG_PSWA = 1,
1830     TARGET_REG_GPRS = 2,
1831     TARGET_REG_ARS = 18,
1832     TARGET_REG_ORIG_R2 = 26,
1833 };
1834 
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUS390XState * env)1835 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1836                                const CPUS390XState *env)
1837 {
1838     int i;
1839     uint32_t *aregs;
1840 
1841     (*regs)[TARGET_REG_PSWM] = tswapreg(env->psw.mask);
1842     (*regs)[TARGET_REG_PSWA] = tswapreg(env->psw.addr);
1843     for (i = 0; i < 16; i++) {
1844         (*regs)[TARGET_REG_GPRS + i] = tswapreg(env->regs[i]);
1845     }
1846     aregs = (uint32_t *)&((*regs)[TARGET_REG_ARS]);
1847     for (i = 0; i < 16; i++) {
1848         aregs[i] = tswap32(env->aregs[i]);
1849     }
1850     (*regs)[TARGET_REG_ORIG_R2] = 0;
1851 }
1852 
1853 #define USE_ELF_CORE_DUMP
1854 #define ELF_EXEC_PAGESIZE 4096
1855 
1856 #define VDSO_HEADER "vdso.c.inc"
1857 
1858 #endif /* TARGET_S390X */
1859 
1860 #ifdef TARGET_RISCV
1861 
1862 #define ELF_ARCH  EM_RISCV
1863 
1864 #ifdef TARGET_RISCV32
1865 #define ELF_CLASS ELFCLASS32
1866 #define VDSO_HEADER "vdso-32.c.inc"
1867 #else
1868 #define ELF_CLASS ELFCLASS64
1869 #define VDSO_HEADER "vdso-64.c.inc"
1870 #endif
1871 
1872 #define ELF_HWCAP get_elf_hwcap()
1873 
get_elf_hwcap(void)1874 static uint32_t get_elf_hwcap(void)
1875 {
1876 #define MISA_BIT(EXT) (1 << (EXT - 'A'))
1877     RISCVCPU *cpu = RISCV_CPU(thread_cpu);
1878     uint32_t mask = MISA_BIT('I') | MISA_BIT('M') | MISA_BIT('A')
1879                     | MISA_BIT('F') | MISA_BIT('D') | MISA_BIT('C')
1880                     | MISA_BIT('V');
1881 
1882     return cpu->env.misa_ext & mask;
1883 #undef MISA_BIT
1884 }
1885 
init_thread(struct target_pt_regs * regs,struct image_info * infop)1886 static inline void init_thread(struct target_pt_regs *regs,
1887                                struct image_info *infop)
1888 {
1889     regs->sepc = infop->entry;
1890     regs->sp = infop->start_stack;
1891 }
1892 
1893 #define ELF_EXEC_PAGESIZE 4096
1894 
1895 #endif /* TARGET_RISCV */
1896 
1897 #ifdef TARGET_HPPA
1898 
1899 #define ELF_CLASS       ELFCLASS32
1900 #define ELF_ARCH        EM_PARISC
1901 #define ELF_PLATFORM    "PARISC"
1902 #define STACK_GROWS_DOWN 0
1903 #define STACK_ALIGNMENT  64
1904 
1905 #define VDSO_HEADER "vdso.c.inc"
1906 
init_thread(struct target_pt_regs * regs,struct image_info * infop)1907 static inline void init_thread(struct target_pt_regs *regs,
1908                                struct image_info *infop)
1909 {
1910     regs->iaoq[0] = infop->entry | PRIV_USER;
1911     regs->iaoq[1] = regs->iaoq[0] + 4;
1912     regs->gr[23] = 0;
1913     regs->gr[24] = infop->argv;
1914     regs->gr[25] = infop->argc;
1915     /* The top-of-stack contains a linkage buffer.  */
1916     regs->gr[30] = infop->start_stack + 64;
1917     regs->gr[31] = infop->entry;
1918 }
1919 
1920 #define LO_COMMPAGE  0
1921 
init_guest_commpage(void)1922 static bool init_guest_commpage(void)
1923 {
1924     /* If reserved_va, then we have already mapped 0 page on the host. */
1925     if (!reserved_va) {
1926         void *want, *addr;
1927 
1928         want = g2h_untagged(LO_COMMPAGE);
1929         addr = mmap(want, TARGET_PAGE_SIZE, PROT_NONE,
1930                     MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED_NOREPLACE, -1, 0);
1931         if (addr == MAP_FAILED) {
1932             perror("Allocating guest commpage");
1933             exit(EXIT_FAILURE);
1934         }
1935         if (addr != want) {
1936             return false;
1937         }
1938     }
1939 
1940     /*
1941      * On Linux, page zero is normally marked execute only + gateway.
1942      * Normal read or write is supposed to fail (thus PROT_NONE above),
1943      * but specific offsets have kernel code mapped to raise permissions
1944      * and implement syscalls.  Here, simply mark the page executable.
1945      * Special case the entry points during translation (see do_page_zero).
1946      */
1947     page_set_flags(LO_COMMPAGE, LO_COMMPAGE | ~TARGET_PAGE_MASK,
1948                    PAGE_EXEC | PAGE_VALID);
1949     return true;
1950 }
1951 
1952 #endif /* TARGET_HPPA */
1953 
1954 #ifdef TARGET_XTENSA
1955 
1956 #define ELF_CLASS       ELFCLASS32
1957 #define ELF_ARCH        EM_XTENSA
1958 
init_thread(struct target_pt_regs * regs,struct image_info * infop)1959 static inline void init_thread(struct target_pt_regs *regs,
1960                                struct image_info *infop)
1961 {
1962     regs->windowbase = 0;
1963     regs->windowstart = 1;
1964     regs->areg[1] = infop->start_stack;
1965     regs->pc = infop->entry;
1966     if (info_is_fdpic(infop)) {
1967         regs->areg[4] = infop->loadmap_addr;
1968         regs->areg[5] = infop->interpreter_loadmap_addr;
1969         if (infop->interpreter_loadmap_addr) {
1970             regs->areg[6] = infop->interpreter_pt_dynamic_addr;
1971         } else {
1972             regs->areg[6] = infop->pt_dynamic_addr;
1973         }
1974     }
1975 }
1976 
1977 /* See linux kernel: arch/xtensa/include/asm/elf.h.  */
1978 #define ELF_NREG 128
1979 typedef target_elf_greg_t target_elf_gregset_t[ELF_NREG];
1980 
1981 enum {
1982     TARGET_REG_PC,
1983     TARGET_REG_PS,
1984     TARGET_REG_LBEG,
1985     TARGET_REG_LEND,
1986     TARGET_REG_LCOUNT,
1987     TARGET_REG_SAR,
1988     TARGET_REG_WINDOWSTART,
1989     TARGET_REG_WINDOWBASE,
1990     TARGET_REG_THREADPTR,
1991     TARGET_REG_AR0 = 64,
1992 };
1993 
elf_core_copy_regs(target_elf_gregset_t * regs,const CPUXtensaState * env)1994 static void elf_core_copy_regs(target_elf_gregset_t *regs,
1995                                const CPUXtensaState *env)
1996 {
1997     unsigned i;
1998 
1999     (*regs)[TARGET_REG_PC] = tswapreg(env->pc);
2000     (*regs)[TARGET_REG_PS] = tswapreg(env->sregs[PS] & ~PS_EXCM);
2001     (*regs)[TARGET_REG_LBEG] = tswapreg(env->sregs[LBEG]);
2002     (*regs)[TARGET_REG_LEND] = tswapreg(env->sregs[LEND]);
2003     (*regs)[TARGET_REG_LCOUNT] = tswapreg(env->sregs[LCOUNT]);
2004     (*regs)[TARGET_REG_SAR] = tswapreg(env->sregs[SAR]);
2005     (*regs)[TARGET_REG_WINDOWSTART] = tswapreg(env->sregs[WINDOW_START]);
2006     (*regs)[TARGET_REG_WINDOWBASE] = tswapreg(env->sregs[WINDOW_BASE]);
2007     (*regs)[TARGET_REG_THREADPTR] = tswapreg(env->uregs[THREADPTR]);
2008     xtensa_sync_phys_from_window((CPUXtensaState *)env);
2009     for (i = 0; i < env->config->nareg; ++i) {
2010         (*regs)[TARGET_REG_AR0 + i] = tswapreg(env->phys_regs[i]);
2011     }
2012 }
2013 
2014 #define USE_ELF_CORE_DUMP
2015 #define ELF_EXEC_PAGESIZE       4096
2016 
2017 #endif /* TARGET_XTENSA */
2018 
2019 #ifdef TARGET_HEXAGON
2020 
2021 #define ELF_CLASS       ELFCLASS32
2022 #define ELF_ARCH        EM_HEXAGON
2023 
init_thread(struct target_pt_regs * regs,struct image_info * infop)2024 static inline void init_thread(struct target_pt_regs *regs,
2025                                struct image_info *infop)
2026 {
2027     regs->sepc = infop->entry;
2028     regs->sp = infop->start_stack;
2029 }
2030 
2031 #endif /* TARGET_HEXAGON */
2032 
2033 #ifndef ELF_BASE_PLATFORM
2034 #define ELF_BASE_PLATFORM (NULL)
2035 #endif
2036 
2037 #ifndef ELF_PLATFORM
2038 #define ELF_PLATFORM (NULL)
2039 #endif
2040 
2041 #ifndef ELF_MACHINE
2042 #define ELF_MACHINE ELF_ARCH
2043 #endif
2044 
2045 #ifndef elf_check_arch
2046 #define elf_check_arch(x) ((x) == ELF_ARCH)
2047 #endif
2048 
2049 #ifndef elf_check_abi
2050 #define elf_check_abi(x) (1)
2051 #endif
2052 
2053 #ifndef ELF_HWCAP
2054 #define ELF_HWCAP 0
2055 #endif
2056 
2057 #ifndef STACK_GROWS_DOWN
2058 #define STACK_GROWS_DOWN 1
2059 #endif
2060 
2061 #ifndef STACK_ALIGNMENT
2062 #define STACK_ALIGNMENT 16
2063 #endif
2064 
2065 #ifdef TARGET_ABI32
2066 #undef ELF_CLASS
2067 #define ELF_CLASS ELFCLASS32
2068 #undef bswaptls
2069 #define bswaptls(ptr) bswap32s(ptr)
2070 #endif
2071 
2072 #ifndef EXSTACK_DEFAULT
2073 #define EXSTACK_DEFAULT false
2074 #endif
2075 
2076 #include "elf.h"
2077 
2078 /* We must delay the following stanzas until after "elf.h". */
2079 #if defined(TARGET_AARCH64)
2080 
arch_parse_elf_property(uint32_t pr_type,uint32_t pr_datasz,const uint32_t * data,struct image_info * info,Error ** errp)2081 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz,
2082                                     const uint32_t *data,
2083                                     struct image_info *info,
2084                                     Error **errp)
2085 {
2086     if (pr_type == GNU_PROPERTY_AARCH64_FEATURE_1_AND) {
2087         if (pr_datasz != sizeof(uint32_t)) {
2088             error_setg(errp, "Ill-formed GNU_PROPERTY_AARCH64_FEATURE_1_AND");
2089             return false;
2090         }
2091         /* We will extract GNU_PROPERTY_AARCH64_FEATURE_1_BTI later. */
2092         info->note_flags = *data;
2093     }
2094     return true;
2095 }
2096 #define ARCH_USE_GNU_PROPERTY 1
2097 
2098 #else
2099 
arch_parse_elf_property(uint32_t pr_type,uint32_t pr_datasz,const uint32_t * data,struct image_info * info,Error ** errp)2100 static bool arch_parse_elf_property(uint32_t pr_type, uint32_t pr_datasz,
2101                                     const uint32_t *data,
2102                                     struct image_info *info,
2103                                     Error **errp)
2104 {
2105     g_assert_not_reached();
2106 }
2107 #define ARCH_USE_GNU_PROPERTY 0
2108 
2109 #endif
2110 
2111 struct exec
2112 {
2113     unsigned int a_info;   /* Use macros N_MAGIC, etc for access */
2114     unsigned int a_text;   /* length of text, in bytes */
2115     unsigned int a_data;   /* length of data, in bytes */
2116     unsigned int a_bss;    /* length of uninitialized data area, in bytes */
2117     unsigned int a_syms;   /* length of symbol table data in file, in bytes */
2118     unsigned int a_entry;  /* start address */
2119     unsigned int a_trsize; /* length of relocation info for text, in bytes */
2120     unsigned int a_drsize; /* length of relocation info for data, in bytes */
2121 };
2122 
2123 
2124 #define N_MAGIC(exec) ((exec).a_info & 0xffff)
2125 #define OMAGIC 0407
2126 #define NMAGIC 0410
2127 #define ZMAGIC 0413
2128 #define QMAGIC 0314
2129 
2130 #define DLINFO_ITEMS 16
2131 
memcpy_fromfs(void * to,const void * from,unsigned long n)2132 static inline void memcpy_fromfs(void * to, const void * from, unsigned long n)
2133 {
2134     memcpy(to, from, n);
2135 }
2136 
2137 #ifdef BSWAP_NEEDED
bswap_ehdr(struct elfhdr * ehdr)2138 static void bswap_ehdr(struct elfhdr *ehdr)
2139 {
2140     bswap16s(&ehdr->e_type);            /* Object file type */
2141     bswap16s(&ehdr->e_machine);         /* Architecture */
2142     bswap32s(&ehdr->e_version);         /* Object file version */
2143     bswaptls(&ehdr->e_entry);           /* Entry point virtual address */
2144     bswaptls(&ehdr->e_phoff);           /* Program header table file offset */
2145     bswaptls(&ehdr->e_shoff);           /* Section header table file offset */
2146     bswap32s(&ehdr->e_flags);           /* Processor-specific flags */
2147     bswap16s(&ehdr->e_ehsize);          /* ELF header size in bytes */
2148     bswap16s(&ehdr->e_phentsize);       /* Program header table entry size */
2149     bswap16s(&ehdr->e_phnum);           /* Program header table entry count */
2150     bswap16s(&ehdr->e_shentsize);       /* Section header table entry size */
2151     bswap16s(&ehdr->e_shnum);           /* Section header table entry count */
2152     bswap16s(&ehdr->e_shstrndx);        /* Section header string table index */
2153 }
2154 
bswap_phdr(struct elf_phdr * phdr,int phnum)2155 static void bswap_phdr(struct elf_phdr *phdr, int phnum)
2156 {
2157     int i;
2158     for (i = 0; i < phnum; ++i, ++phdr) {
2159         bswap32s(&phdr->p_type);        /* Segment type */
2160         bswap32s(&phdr->p_flags);       /* Segment flags */
2161         bswaptls(&phdr->p_offset);      /* Segment file offset */
2162         bswaptls(&phdr->p_vaddr);       /* Segment virtual address */
2163         bswaptls(&phdr->p_paddr);       /* Segment physical address */
2164         bswaptls(&phdr->p_filesz);      /* Segment size in file */
2165         bswaptls(&phdr->p_memsz);       /* Segment size in memory */
2166         bswaptls(&phdr->p_align);       /* Segment alignment */
2167     }
2168 }
2169 
bswap_shdr(struct elf_shdr * shdr,int shnum)2170 static void bswap_shdr(struct elf_shdr *shdr, int shnum)
2171 {
2172     int i;
2173     for (i = 0; i < shnum; ++i, ++shdr) {
2174         bswap32s(&shdr->sh_name);
2175         bswap32s(&shdr->sh_type);
2176         bswaptls(&shdr->sh_flags);
2177         bswaptls(&shdr->sh_addr);
2178         bswaptls(&shdr->sh_offset);
2179         bswaptls(&shdr->sh_size);
2180         bswap32s(&shdr->sh_link);
2181         bswap32s(&shdr->sh_info);
2182         bswaptls(&shdr->sh_addralign);
2183         bswaptls(&shdr->sh_entsize);
2184     }
2185 }
2186 
bswap_sym(struct elf_sym * sym)2187 static void bswap_sym(struct elf_sym *sym)
2188 {
2189     bswap32s(&sym->st_name);
2190     bswaptls(&sym->st_value);
2191     bswaptls(&sym->st_size);
2192     bswap16s(&sym->st_shndx);
2193 }
2194 
2195 #ifdef TARGET_MIPS
bswap_mips_abiflags(Mips_elf_abiflags_v0 * abiflags)2196 static void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags)
2197 {
2198     bswap16s(&abiflags->version);
2199     bswap32s(&abiflags->ases);
2200     bswap32s(&abiflags->isa_ext);
2201     bswap32s(&abiflags->flags1);
2202     bswap32s(&abiflags->flags2);
2203 }
2204 #endif
2205 #else
bswap_ehdr(struct elfhdr * ehdr)2206 static inline void bswap_ehdr(struct elfhdr *ehdr) { }
bswap_phdr(struct elf_phdr * phdr,int phnum)2207 static inline void bswap_phdr(struct elf_phdr *phdr, int phnum) { }
bswap_shdr(struct elf_shdr * shdr,int shnum)2208 static inline void bswap_shdr(struct elf_shdr *shdr, int shnum) { }
bswap_sym(struct elf_sym * sym)2209 static inline void bswap_sym(struct elf_sym *sym) { }
2210 #ifdef TARGET_MIPS
bswap_mips_abiflags(Mips_elf_abiflags_v0 * abiflags)2211 static inline void bswap_mips_abiflags(Mips_elf_abiflags_v0 *abiflags) { }
2212 #endif
2213 #endif
2214 
2215 #ifdef USE_ELF_CORE_DUMP
2216 static int elf_core_dump(int, const CPUArchState *);
2217 #endif /* USE_ELF_CORE_DUMP */
2218 static void load_symbols(struct elfhdr *hdr, const ImageSource *src,
2219                          abi_ulong load_bias);
2220 
2221 /* Verify the portions of EHDR within E_IDENT for the target.
2222    This can be performed before bswapping the entire header.  */
elf_check_ident(struct elfhdr * ehdr)2223 static bool elf_check_ident(struct elfhdr *ehdr)
2224 {
2225     return (ehdr->e_ident[EI_MAG0] == ELFMAG0
2226             && ehdr->e_ident[EI_MAG1] == ELFMAG1
2227             && ehdr->e_ident[EI_MAG2] == ELFMAG2
2228             && ehdr->e_ident[EI_MAG3] == ELFMAG3
2229             && ehdr->e_ident[EI_CLASS] == ELF_CLASS
2230             && ehdr->e_ident[EI_DATA] == ELF_DATA
2231             && ehdr->e_ident[EI_VERSION] == EV_CURRENT);
2232 }
2233 
2234 /* Verify the portions of EHDR outside of E_IDENT for the target.
2235    This has to wait until after bswapping the header.  */
elf_check_ehdr(struct elfhdr * ehdr)2236 static bool elf_check_ehdr(struct elfhdr *ehdr)
2237 {
2238     return (elf_check_arch(ehdr->e_machine)
2239             && elf_check_abi(ehdr->e_flags)
2240             && ehdr->e_ehsize == sizeof(struct elfhdr)
2241             && ehdr->e_phentsize == sizeof(struct elf_phdr)
2242             && (ehdr->e_type == ET_EXEC || ehdr->e_type == ET_DYN));
2243 }
2244 
2245 /*
2246  * 'copy_elf_strings()' copies argument/envelope strings from user
2247  * memory to free pages in kernel mem. These are in a format ready
2248  * to be put directly into the top of new user memory.
2249  *
2250  */
copy_elf_strings(int argc,char ** argv,char * scratch,abi_ulong p,abi_ulong stack_limit)2251 static abi_ulong copy_elf_strings(int argc, char **argv, char *scratch,
2252                                   abi_ulong p, abi_ulong stack_limit)
2253 {
2254     char *tmp;
2255     int len, i;
2256     abi_ulong top = p;
2257 
2258     if (!p) {
2259         return 0;       /* bullet-proofing */
2260     }
2261 
2262     if (STACK_GROWS_DOWN) {
2263         int offset = ((p - 1) % TARGET_PAGE_SIZE) + 1;
2264         for (i = argc - 1; i >= 0; --i) {
2265             tmp = argv[i];
2266             if (!tmp) {
2267                 fprintf(stderr, "VFS: argc is wrong");
2268                 exit(-1);
2269             }
2270             len = strlen(tmp) + 1;
2271             tmp += len;
2272 
2273             if (len > (p - stack_limit)) {
2274                 return 0;
2275             }
2276             while (len) {
2277                 int bytes_to_copy = (len > offset) ? offset : len;
2278                 tmp -= bytes_to_copy;
2279                 p -= bytes_to_copy;
2280                 offset -= bytes_to_copy;
2281                 len -= bytes_to_copy;
2282 
2283                 memcpy_fromfs(scratch + offset, tmp, bytes_to_copy);
2284 
2285                 if (offset == 0) {
2286                     memcpy_to_target(p, scratch, top - p);
2287                     top = p;
2288                     offset = TARGET_PAGE_SIZE;
2289                 }
2290             }
2291         }
2292         if (p != top) {
2293             memcpy_to_target(p, scratch + offset, top - p);
2294         }
2295     } else {
2296         int remaining = TARGET_PAGE_SIZE - (p % TARGET_PAGE_SIZE);
2297         for (i = 0; i < argc; ++i) {
2298             tmp = argv[i];
2299             if (!tmp) {
2300                 fprintf(stderr, "VFS: argc is wrong");
2301                 exit(-1);
2302             }
2303             len = strlen(tmp) + 1;
2304             if (len > (stack_limit - p)) {
2305                 return 0;
2306             }
2307             while (len) {
2308                 int bytes_to_copy = (len > remaining) ? remaining : len;
2309 
2310                 memcpy_fromfs(scratch + (p - top), tmp, bytes_to_copy);
2311 
2312                 tmp += bytes_to_copy;
2313                 remaining -= bytes_to_copy;
2314                 p += bytes_to_copy;
2315                 len -= bytes_to_copy;
2316 
2317                 if (remaining == 0) {
2318                     memcpy_to_target(top, scratch, p - top);
2319                     top = p;
2320                     remaining = TARGET_PAGE_SIZE;
2321                 }
2322             }
2323         }
2324         if (p != top) {
2325             memcpy_to_target(top, scratch, p - top);
2326         }
2327     }
2328 
2329     return p;
2330 }
2331 
2332 /* Older linux kernels provide up to MAX_ARG_PAGES (default: 32) of
2333  * argument/environment space. Newer kernels (>2.6.33) allow more,
2334  * dependent on stack size, but guarantee at least 32 pages for
2335  * backwards compatibility.
2336  */
2337 #define STACK_LOWER_LIMIT (32 * TARGET_PAGE_SIZE)
2338 
setup_arg_pages(struct linux_binprm * bprm,struct image_info * info)2339 static abi_ulong setup_arg_pages(struct linux_binprm *bprm,
2340                                  struct image_info *info)
2341 {
2342     abi_ulong size, error, guard;
2343     int prot;
2344 
2345     size = guest_stack_size;
2346     if (size < STACK_LOWER_LIMIT) {
2347         size = STACK_LOWER_LIMIT;
2348     }
2349 
2350     if (STACK_GROWS_DOWN) {
2351         guard = TARGET_PAGE_SIZE;
2352         if (guard < qemu_real_host_page_size()) {
2353             guard = qemu_real_host_page_size();
2354         }
2355     } else {
2356         /* no guard page for hppa target where stack grows upwards. */
2357         guard = 0;
2358     }
2359 
2360     prot = PROT_READ | PROT_WRITE;
2361     if (info->exec_stack) {
2362         prot |= PROT_EXEC;
2363     }
2364     error = target_mmap(0, size + guard, prot,
2365                         MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
2366     if (error == -1) {
2367         perror("mmap stack");
2368         exit(-1);
2369     }
2370 
2371     /* We reserve one extra page at the top of the stack as guard.  */
2372     if (STACK_GROWS_DOWN) {
2373         target_mprotect(error, guard, PROT_NONE);
2374         info->stack_limit = error + guard;
2375         return info->stack_limit + size - sizeof(void *);
2376     } else {
2377         info->stack_limit = error + size;
2378         return error;
2379     }
2380 }
2381 
2382 /**
2383  * zero_bss:
2384  *
2385  * Map and zero the bss.  We need to explicitly zero any fractional pages
2386  * after the data section (i.e. bss).  Return false on mapping failure.
2387  */
zero_bss(abi_ulong start_bss,abi_ulong end_bss,int prot,Error ** errp)2388 static bool zero_bss(abi_ulong start_bss, abi_ulong end_bss,
2389                      int prot, Error **errp)
2390 {
2391     abi_ulong align_bss;
2392 
2393     /* We only expect writable bss; the code segment shouldn't need this. */
2394     if (!(prot & PROT_WRITE)) {
2395         error_setg(errp, "PT_LOAD with non-writable bss");
2396         return false;
2397     }
2398 
2399     align_bss = TARGET_PAGE_ALIGN(start_bss);
2400     end_bss = TARGET_PAGE_ALIGN(end_bss);
2401 
2402     if (start_bss < align_bss) {
2403         int flags = page_get_flags(start_bss);
2404 
2405         if (!(flags & PAGE_RWX)) {
2406             /*
2407              * The whole address space of the executable was reserved
2408              * at the start, therefore all pages will be VALID.
2409              * But assuming there are no PROT_NONE PT_LOAD segments,
2410              * a PROT_NONE page means no data all bss, and we can
2411              * simply extend the new anon mapping back to the start
2412              * of the page of bss.
2413              */
2414             align_bss -= TARGET_PAGE_SIZE;
2415         } else {
2416             /*
2417              * The start of the bss shares a page with something.
2418              * The only thing that we expect is the data section,
2419              * which would already be marked writable.
2420              * Overlapping the RX code segment seems malformed.
2421              */
2422             if (!(flags & PAGE_WRITE)) {
2423                 error_setg(errp, "PT_LOAD with bss overlapping "
2424                            "non-writable page");
2425                 return false;
2426             }
2427 
2428             /* The page is already mapped and writable. */
2429             memset(g2h_untagged(start_bss), 0, align_bss - start_bss);
2430         }
2431     }
2432 
2433     if (align_bss < end_bss &&
2434         target_mmap(align_bss, end_bss - align_bss, prot,
2435                     MAP_FIXED | MAP_PRIVATE | MAP_ANON, -1, 0) == -1) {
2436         error_setg_errno(errp, errno, "Error mapping bss");
2437         return false;
2438     }
2439     return true;
2440 }
2441 
2442 #if defined(TARGET_ARM)
elf_is_fdpic(struct elfhdr * exec)2443 static int elf_is_fdpic(struct elfhdr *exec)
2444 {
2445     return exec->e_ident[EI_OSABI] == ELFOSABI_ARM_FDPIC;
2446 }
2447 #elif defined(TARGET_XTENSA)
elf_is_fdpic(struct elfhdr * exec)2448 static int elf_is_fdpic(struct elfhdr *exec)
2449 {
2450     return exec->e_ident[EI_OSABI] == ELFOSABI_XTENSA_FDPIC;
2451 }
2452 #else
2453 /* Default implementation, always false.  */
elf_is_fdpic(struct elfhdr * exec)2454 static int elf_is_fdpic(struct elfhdr *exec)
2455 {
2456     return 0;
2457 }
2458 #endif
2459 
loader_build_fdpic_loadmap(struct image_info * info,abi_ulong sp)2460 static abi_ulong loader_build_fdpic_loadmap(struct image_info *info, abi_ulong sp)
2461 {
2462     uint16_t n;
2463     struct elf32_fdpic_loadseg *loadsegs = info->loadsegs;
2464 
2465     /* elf32_fdpic_loadseg */
2466     n = info->nsegs;
2467     while (n--) {
2468         sp -= 12;
2469         put_user_u32(loadsegs[n].addr, sp+0);
2470         put_user_u32(loadsegs[n].p_vaddr, sp+4);
2471         put_user_u32(loadsegs[n].p_memsz, sp+8);
2472     }
2473 
2474     /* elf32_fdpic_loadmap */
2475     sp -= 4;
2476     put_user_u16(0, sp+0); /* version */
2477     put_user_u16(info->nsegs, sp+2); /* nsegs */
2478 
2479     info->personality = PER_LINUX_FDPIC;
2480     info->loadmap_addr = sp;
2481 
2482     return sp;
2483 }
2484 
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)2485 static abi_ulong create_elf_tables(abi_ulong p, int argc, int envc,
2486                                    struct elfhdr *exec,
2487                                    struct image_info *info,
2488                                    struct image_info *interp_info,
2489                                    struct image_info *vdso_info)
2490 {
2491     abi_ulong sp;
2492     abi_ulong u_argc, u_argv, u_envp, u_auxv;
2493     int size;
2494     int i;
2495     abi_ulong u_rand_bytes;
2496     uint8_t k_rand_bytes[16];
2497     abi_ulong u_platform, u_base_platform;
2498     const char *k_platform, *k_base_platform;
2499     const int n = sizeof(elf_addr_t);
2500 
2501     sp = p;
2502 
2503     /* Needs to be before we load the env/argc/... */
2504     if (elf_is_fdpic(exec)) {
2505         /* Need 4 byte alignment for these structs */
2506         sp &= ~3;
2507         sp = loader_build_fdpic_loadmap(info, sp);
2508         info->other_info = interp_info;
2509         if (interp_info) {
2510             interp_info->other_info = info;
2511             sp = loader_build_fdpic_loadmap(interp_info, sp);
2512             info->interpreter_loadmap_addr = interp_info->loadmap_addr;
2513             info->interpreter_pt_dynamic_addr = interp_info->pt_dynamic_addr;
2514         } else {
2515             info->interpreter_loadmap_addr = 0;
2516             info->interpreter_pt_dynamic_addr = 0;
2517         }
2518     }
2519 
2520     u_base_platform = 0;
2521     k_base_platform = ELF_BASE_PLATFORM;
2522     if (k_base_platform) {
2523         size_t len = strlen(k_base_platform) + 1;
2524         if (STACK_GROWS_DOWN) {
2525             sp -= (len + n - 1) & ~(n - 1);
2526             u_base_platform = sp;
2527             /* FIXME - check return value of memcpy_to_target() for failure */
2528             memcpy_to_target(sp, k_base_platform, len);
2529         } else {
2530             memcpy_to_target(sp, k_base_platform, len);
2531             u_base_platform = sp;
2532             sp += len + 1;
2533         }
2534     }
2535 
2536     u_platform = 0;
2537     k_platform = ELF_PLATFORM;
2538     if (k_platform) {
2539         size_t len = strlen(k_platform) + 1;
2540         if (STACK_GROWS_DOWN) {
2541             sp -= (len + n - 1) & ~(n - 1);
2542             u_platform = sp;
2543             /* FIXME - check return value of memcpy_to_target() for failure */
2544             memcpy_to_target(sp, k_platform, len);
2545         } else {
2546             memcpy_to_target(sp, k_platform, len);
2547             u_platform = sp;
2548             sp += len + 1;
2549         }
2550     }
2551 
2552     /* Provide 16 byte alignment for the PRNG, and basic alignment for
2553      * the argv and envp pointers.
2554      */
2555     if (STACK_GROWS_DOWN) {
2556         sp = QEMU_ALIGN_DOWN(sp, 16);
2557     } else {
2558         sp = QEMU_ALIGN_UP(sp, 16);
2559     }
2560 
2561     /*
2562      * Generate 16 random bytes for userspace PRNG seeding.
2563      */
2564     qemu_guest_getrandom_nofail(k_rand_bytes, sizeof(k_rand_bytes));
2565     if (STACK_GROWS_DOWN) {
2566         sp -= 16;
2567         u_rand_bytes = sp;
2568         /* FIXME - check return value of memcpy_to_target() for failure */
2569         memcpy_to_target(sp, k_rand_bytes, 16);
2570     } else {
2571         memcpy_to_target(sp, k_rand_bytes, 16);
2572         u_rand_bytes = sp;
2573         sp += 16;
2574     }
2575 
2576     size = (DLINFO_ITEMS + 1) * 2;
2577     if (k_base_platform) {
2578         size += 2;
2579     }
2580     if (k_platform) {
2581         size += 2;
2582     }
2583     if (vdso_info) {
2584         size += 2;
2585     }
2586 #ifdef DLINFO_ARCH_ITEMS
2587     size += DLINFO_ARCH_ITEMS * 2;
2588 #endif
2589 #ifdef ELF_HWCAP2
2590     size += 2;
2591 #endif
2592     info->auxv_len = size * n;
2593 
2594     size += envc + argc + 2;
2595     size += 1;  /* argc itself */
2596     size *= n;
2597 
2598     /* Allocate space and finalize stack alignment for entry now.  */
2599     if (STACK_GROWS_DOWN) {
2600         u_argc = QEMU_ALIGN_DOWN(sp - size, STACK_ALIGNMENT);
2601         sp = u_argc;
2602     } else {
2603         u_argc = sp;
2604         sp = QEMU_ALIGN_UP(sp + size, STACK_ALIGNMENT);
2605     }
2606 
2607     u_argv = u_argc + n;
2608     u_envp = u_argv + (argc + 1) * n;
2609     u_auxv = u_envp + (envc + 1) * n;
2610     info->saved_auxv = u_auxv;
2611     info->argc = argc;
2612     info->envc = envc;
2613     info->argv = u_argv;
2614     info->envp = u_envp;
2615 
2616     /* This is correct because Linux defines
2617      * elf_addr_t as Elf32_Off / Elf64_Off
2618      */
2619 #define NEW_AUX_ENT(id, val) do {               \
2620         put_user_ual(id, u_auxv);  u_auxv += n; \
2621         put_user_ual(val, u_auxv); u_auxv += n; \
2622     } while(0)
2623 
2624 #ifdef ARCH_DLINFO
2625     /*
2626      * ARCH_DLINFO must come first so platform specific code can enforce
2627      * special alignment requirements on the AUXV if necessary (eg. PPC).
2628      */
2629     ARCH_DLINFO;
2630 #endif
2631     /* There must be exactly DLINFO_ITEMS entries here, or the assert
2632      * on info->auxv_len will trigger.
2633      */
2634     NEW_AUX_ENT(AT_PHDR, (abi_ulong)(info->load_addr + exec->e_phoff));
2635     NEW_AUX_ENT(AT_PHENT, (abi_ulong)(sizeof (struct elf_phdr)));
2636     NEW_AUX_ENT(AT_PHNUM, (abi_ulong)(exec->e_phnum));
2637     NEW_AUX_ENT(AT_PAGESZ, (abi_ulong)(TARGET_PAGE_SIZE));
2638     NEW_AUX_ENT(AT_BASE, (abi_ulong)(interp_info ? interp_info->load_addr : 0));
2639     NEW_AUX_ENT(AT_FLAGS, (abi_ulong)0);
2640     NEW_AUX_ENT(AT_ENTRY, info->entry);
2641     NEW_AUX_ENT(AT_UID, (abi_ulong) getuid());
2642     NEW_AUX_ENT(AT_EUID, (abi_ulong) geteuid());
2643     NEW_AUX_ENT(AT_GID, (abi_ulong) getgid());
2644     NEW_AUX_ENT(AT_EGID, (abi_ulong) getegid());
2645     NEW_AUX_ENT(AT_HWCAP, (abi_ulong) ELF_HWCAP);
2646     NEW_AUX_ENT(AT_CLKTCK, (abi_ulong) sysconf(_SC_CLK_TCK));
2647     NEW_AUX_ENT(AT_RANDOM, (abi_ulong) u_rand_bytes);
2648     NEW_AUX_ENT(AT_SECURE, (abi_ulong) qemu_getauxval(AT_SECURE));
2649     NEW_AUX_ENT(AT_EXECFN, info->file_string);
2650 
2651 #ifdef ELF_HWCAP2
2652     NEW_AUX_ENT(AT_HWCAP2, (abi_ulong) ELF_HWCAP2);
2653 #endif
2654 
2655     if (u_base_platform) {
2656         NEW_AUX_ENT(AT_BASE_PLATFORM, u_base_platform);
2657     }
2658     if (u_platform) {
2659         NEW_AUX_ENT(AT_PLATFORM, u_platform);
2660     }
2661     if (vdso_info) {
2662         NEW_AUX_ENT(AT_SYSINFO_EHDR, vdso_info->load_addr);
2663     }
2664     NEW_AUX_ENT (AT_NULL, 0);
2665 #undef NEW_AUX_ENT
2666 
2667     /* Check that our initial calculation of the auxv length matches how much
2668      * we actually put into it.
2669      */
2670     assert(info->auxv_len == u_auxv - info->saved_auxv);
2671 
2672     put_user_ual(argc, u_argc);
2673 
2674     p = info->arg_strings;
2675     for (i = 0; i < argc; ++i) {
2676         put_user_ual(p, u_argv);
2677         u_argv += n;
2678         p += target_strlen(p) + 1;
2679     }
2680     put_user_ual(0, u_argv);
2681 
2682     p = info->env_strings;
2683     for (i = 0; i < envc; ++i) {
2684         put_user_ual(p, u_envp);
2685         u_envp += n;
2686         p += target_strlen(p) + 1;
2687     }
2688     put_user_ual(0, u_envp);
2689 
2690     return sp;
2691 }
2692 
2693 #if defined(HI_COMMPAGE)
2694 #define LO_COMMPAGE -1
2695 #elif defined(LO_COMMPAGE)
2696 #define HI_COMMPAGE 0
2697 #else
2698 #define HI_COMMPAGE 0
2699 #define LO_COMMPAGE -1
2700 #ifndef INIT_GUEST_COMMPAGE
2701 #define init_guest_commpage() true
2702 #endif
2703 #endif
2704 
2705 /**
2706  * pgb_try_mmap:
2707  * @addr: host start address
2708  * @addr_last: host last address
2709  * @keep: do not unmap the probe region
2710  *
2711  * Return 1 if [@addr, @addr_last] is not mapped in the host,
2712  * return 0 if it is not available to map, and -1 on mmap error.
2713  * If @keep, the region is left mapped on success, otherwise unmapped.
2714  */
pgb_try_mmap(uintptr_t addr,uintptr_t addr_last,bool keep)2715 static int pgb_try_mmap(uintptr_t addr, uintptr_t addr_last, bool keep)
2716 {
2717     size_t size = addr_last - addr + 1;
2718     void *p = mmap((void *)addr, size, PROT_NONE,
2719                    MAP_ANONYMOUS | MAP_PRIVATE |
2720                    MAP_NORESERVE | MAP_FIXED_NOREPLACE, -1, 0);
2721     int ret;
2722 
2723     if (p == MAP_FAILED) {
2724         return errno == EEXIST ? 0 : -1;
2725     }
2726     ret = p == (void *)addr;
2727     if (!keep || !ret) {
2728         munmap(p, size);
2729     }
2730     return ret;
2731 }
2732 
2733 /**
2734  * pgb_try_mmap_skip_brk(uintptr_t addr, uintptr_t size, uintptr_t brk)
2735  * @addr: host address
2736  * @addr_last: host last address
2737  * @brk: host brk
2738  *
2739  * Like pgb_try_mmap, but additionally reserve some memory following brk.
2740  */
pgb_try_mmap_skip_brk(uintptr_t addr,uintptr_t addr_last,uintptr_t brk,bool keep)2741 static int pgb_try_mmap_skip_brk(uintptr_t addr, uintptr_t addr_last,
2742                                  uintptr_t brk, bool keep)
2743 {
2744     uintptr_t brk_last = brk + 16 * MiB - 1;
2745 
2746     /* Do not map anything close to the host brk. */
2747     if (addr <= brk_last && brk <= addr_last) {
2748         return 0;
2749     }
2750     return pgb_try_mmap(addr, addr_last, keep);
2751 }
2752 
2753 /**
2754  * pgb_try_mmap_set:
2755  * @ga: set of guest addrs
2756  * @base: guest_base
2757  * @brk: host brk
2758  *
2759  * Return true if all @ga can be mapped by the host at @base.
2760  * On success, retain the mapping at index 0 for reserved_va.
2761  */
2762 
2763 typedef struct PGBAddrs {
2764     uintptr_t bounds[3][2]; /* start/last pairs */
2765     int nbounds;
2766 } PGBAddrs;
2767 
pgb_try_mmap_set(const PGBAddrs * ga,uintptr_t base,uintptr_t brk)2768 static bool pgb_try_mmap_set(const PGBAddrs *ga, uintptr_t base, uintptr_t brk)
2769 {
2770     for (int i = ga->nbounds - 1; i >= 0; --i) {
2771         if (pgb_try_mmap_skip_brk(ga->bounds[i][0] + base,
2772                                   ga->bounds[i][1] + base,
2773                                   brk, i == 0 && reserved_va) <= 0) {
2774             return false;
2775         }
2776     }
2777     return true;
2778 }
2779 
2780 /**
2781  * pgb_addr_set:
2782  * @ga: output set of guest addrs
2783  * @guest_loaddr: guest image low address
2784  * @guest_loaddr: guest image high address
2785  * @identity: create for identity mapping
2786  *
2787  * Fill in @ga with the image, COMMPAGE and NULL page.
2788  */
pgb_addr_set(PGBAddrs * ga,abi_ulong guest_loaddr,abi_ulong guest_hiaddr,bool try_identity)2789 static bool pgb_addr_set(PGBAddrs *ga, abi_ulong guest_loaddr,
2790                          abi_ulong guest_hiaddr, bool try_identity)
2791 {
2792     int n;
2793 
2794     /*
2795      * With a low commpage, or a guest mapped very low,
2796      * we may not be able to use the identity map.
2797      */
2798     if (try_identity) {
2799         if (LO_COMMPAGE != -1 && LO_COMMPAGE < mmap_min_addr) {
2800             return false;
2801         }
2802         if (guest_loaddr != 0 && guest_loaddr < mmap_min_addr) {
2803             return false;
2804         }
2805     }
2806 
2807     memset(ga, 0, sizeof(*ga));
2808     n = 0;
2809 
2810     if (reserved_va) {
2811         ga->bounds[n][0] = try_identity ? mmap_min_addr : 0;
2812         ga->bounds[n][1] = reserved_va;
2813         n++;
2814         /* LO_COMMPAGE and NULL handled by reserving from 0. */
2815     } else {
2816         /* Add any LO_COMMPAGE or NULL page. */
2817         if (LO_COMMPAGE != -1) {
2818             ga->bounds[n][0] = 0;
2819             ga->bounds[n][1] = LO_COMMPAGE + TARGET_PAGE_SIZE - 1;
2820             n++;
2821         } else if (!try_identity) {
2822             ga->bounds[n][0] = 0;
2823             ga->bounds[n][1] = TARGET_PAGE_SIZE - 1;
2824             n++;
2825         }
2826 
2827         /* Add the guest image for ET_EXEC. */
2828         if (guest_loaddr) {
2829             ga->bounds[n][0] = guest_loaddr;
2830             ga->bounds[n][1] = guest_hiaddr;
2831             n++;
2832         }
2833     }
2834 
2835     /*
2836      * Temporarily disable
2837      *   "comparison is always false due to limited range of data type"
2838      * due to comparison between unsigned and (possible) 0.
2839      */
2840 #pragma GCC diagnostic push
2841 #pragma GCC diagnostic ignored "-Wtype-limits"
2842 
2843     /* Add any HI_COMMPAGE not covered by reserved_va. */
2844     if (reserved_va < HI_COMMPAGE) {
2845         ga->bounds[n][0] = HI_COMMPAGE & qemu_real_host_page_mask();
2846         ga->bounds[n][1] = HI_COMMPAGE + TARGET_PAGE_SIZE - 1;
2847         n++;
2848     }
2849 
2850 #pragma GCC diagnostic pop
2851 
2852     ga->nbounds = n;
2853     return true;
2854 }
2855 
pgb_fail_in_use(const char * image_name)2856 static void pgb_fail_in_use(const char *image_name)
2857 {
2858     error_report("%s: requires virtual address space that is in use "
2859                  "(omit the -B option or choose a different value)",
2860                  image_name);
2861     exit(EXIT_FAILURE);
2862 }
2863 
pgb_fixed(const char * image_name,uintptr_t guest_loaddr,uintptr_t guest_hiaddr,uintptr_t align)2864 static void pgb_fixed(const char *image_name, uintptr_t guest_loaddr,
2865                       uintptr_t guest_hiaddr, uintptr_t align)
2866 {
2867     PGBAddrs ga;
2868     uintptr_t brk = (uintptr_t)sbrk(0);
2869 
2870     if (!QEMU_IS_ALIGNED(guest_base, align)) {
2871         fprintf(stderr, "Requested guest base %p does not satisfy "
2872                 "host minimum alignment (0x%" PRIxPTR ")\n",
2873                 (void *)guest_base, align);
2874         exit(EXIT_FAILURE);
2875     }
2876 
2877     if (!pgb_addr_set(&ga, guest_loaddr, guest_hiaddr, !guest_base)
2878         || !pgb_try_mmap_set(&ga, guest_base, brk)) {
2879         pgb_fail_in_use(image_name);
2880     }
2881 }
2882 
2883 /**
2884  * pgb_find_fallback:
2885  *
2886  * This is a fallback method for finding holes in the host address space
2887  * if we don't have the benefit of being able to access /proc/self/map.
2888  * It can potentially take a very long time as we can only dumbly iterate
2889  * up the host address space seeing if the allocation would work.
2890  */
pgb_find_fallback(const PGBAddrs * ga,uintptr_t align,uintptr_t brk)2891 static uintptr_t pgb_find_fallback(const PGBAddrs *ga, uintptr_t align,
2892                                    uintptr_t brk)
2893 {
2894     /* TODO: come up with a better estimate of how much to skip. */
2895     uintptr_t skip = sizeof(uintptr_t) == 4 ? MiB : GiB;
2896 
2897     for (uintptr_t base = skip; ; base += skip) {
2898         base = ROUND_UP(base, align);
2899         if (pgb_try_mmap_set(ga, base, brk)) {
2900             return base;
2901         }
2902         if (base >= -skip) {
2903             return -1;
2904         }
2905     }
2906 }
2907 
pgb_try_itree(const PGBAddrs * ga,uintptr_t base,IntervalTreeRoot * root)2908 static uintptr_t pgb_try_itree(const PGBAddrs *ga, uintptr_t base,
2909                                IntervalTreeRoot *root)
2910 {
2911     for (int i = ga->nbounds - 1; i >= 0; --i) {
2912         uintptr_t s = base + ga->bounds[i][0];
2913         uintptr_t l = base + ga->bounds[i][1];
2914         IntervalTreeNode *n;
2915 
2916         if (l < s) {
2917             /* Wraparound. Skip to advance S to mmap_min_addr. */
2918             return mmap_min_addr - s;
2919         }
2920 
2921         n = interval_tree_iter_first(root, s, l);
2922         if (n != NULL) {
2923             /* Conflict.  Skip to advance S to LAST + 1. */
2924             return n->last - s + 1;
2925         }
2926     }
2927     return 0;  /* success */
2928 }
2929 
pgb_find_itree(const PGBAddrs * ga,IntervalTreeRoot * root,uintptr_t align,uintptr_t brk)2930 static uintptr_t pgb_find_itree(const PGBAddrs *ga, IntervalTreeRoot *root,
2931                                 uintptr_t align, uintptr_t brk)
2932 {
2933     uintptr_t last = sizeof(uintptr_t) == 4 ? MiB : GiB;
2934     uintptr_t base, skip;
2935 
2936     while (true) {
2937         base = ROUND_UP(last, align);
2938         if (base < last) {
2939             return -1;
2940         }
2941 
2942         skip = pgb_try_itree(ga, base, root);
2943         if (skip == 0) {
2944             break;
2945         }
2946 
2947         last = base + skip;
2948         if (last < base) {
2949             return -1;
2950         }
2951     }
2952 
2953     /*
2954      * We've chosen 'base' based on holes in the interval tree,
2955      * but we don't yet know if it is a valid host address.
2956      * Because it is the first matching hole, if the host addresses
2957      * are invalid we know there are no further matches.
2958      */
2959     return pgb_try_mmap_set(ga, base, brk) ? base : -1;
2960 }
2961 
pgb_dynamic(const char * image_name,uintptr_t guest_loaddr,uintptr_t guest_hiaddr,uintptr_t align)2962 static void pgb_dynamic(const char *image_name, uintptr_t guest_loaddr,
2963                         uintptr_t guest_hiaddr, uintptr_t align)
2964 {
2965     IntervalTreeRoot *root;
2966     uintptr_t brk, ret;
2967     PGBAddrs ga;
2968 
2969     /* Try the identity map first. */
2970     if (pgb_addr_set(&ga, guest_loaddr, guest_hiaddr, true)) {
2971         brk = (uintptr_t)sbrk(0);
2972         if (pgb_try_mmap_set(&ga, 0, brk)) {
2973             guest_base = 0;
2974             return;
2975         }
2976     }
2977 
2978     /*
2979      * Rebuild the address set for non-identity map.
2980      * This differs in the mapping of the guest NULL page.
2981      */
2982     pgb_addr_set(&ga, guest_loaddr, guest_hiaddr, false);
2983 
2984     root = read_self_maps();
2985 
2986     /* Read brk after we've read the maps, which will malloc. */
2987     brk = (uintptr_t)sbrk(0);
2988 
2989     if (!root) {
2990         ret = pgb_find_fallback(&ga, align, brk);
2991     } else {
2992         /*
2993          * Reserve the area close to the host brk.
2994          * This will be freed with the rest of the tree.
2995          */
2996         IntervalTreeNode *b = g_new0(IntervalTreeNode, 1);
2997         b->start = brk;
2998         b->last = brk + 16 * MiB - 1;
2999         interval_tree_insert(b, root);
3000 
3001         ret = pgb_find_itree(&ga, root, align, brk);
3002         free_self_maps(root);
3003     }
3004 
3005     if (ret == -1) {
3006         int w = TARGET_LONG_BITS / 4;
3007 
3008         error_report("%s: Unable to find a guest_base to satisfy all "
3009                      "guest address mapping requirements", image_name);
3010 
3011         for (int i = 0; i < ga.nbounds; ++i) {
3012             error_printf("  %0*" PRIx64 "-%0*" PRIx64 "\n",
3013                          w, (uint64_t)ga.bounds[i][0],
3014                          w, (uint64_t)ga.bounds[i][1]);
3015         }
3016         exit(EXIT_FAILURE);
3017     }
3018     guest_base = ret;
3019 }
3020 
probe_guest_base(const char * image_name,abi_ulong guest_loaddr,abi_ulong guest_hiaddr)3021 void probe_guest_base(const char *image_name, abi_ulong guest_loaddr,
3022                       abi_ulong guest_hiaddr)
3023 {
3024     /* In order to use host shmat, we must be able to honor SHMLBA.  */
3025     uintptr_t align = MAX(SHMLBA, TARGET_PAGE_SIZE);
3026 
3027     /* Sanity check the guest binary. */
3028     if (reserved_va) {
3029         if (guest_hiaddr > reserved_va) {
3030             error_report("%s: requires more than reserved virtual "
3031                          "address space (0x%" PRIx64 " > 0x%lx)",
3032                          image_name, (uint64_t)guest_hiaddr, reserved_va);
3033             exit(EXIT_FAILURE);
3034         }
3035     } else {
3036         if (guest_hiaddr != (uintptr_t)guest_hiaddr) {
3037             error_report("%s: requires more virtual address space "
3038                          "than the host can provide (0x%" PRIx64 ")",
3039                          image_name, (uint64_t)guest_hiaddr + 1);
3040             exit(EXIT_FAILURE);
3041         }
3042     }
3043 
3044     if (have_guest_base) {
3045         pgb_fixed(image_name, guest_loaddr, guest_hiaddr, align);
3046     } else {
3047         pgb_dynamic(image_name, guest_loaddr, guest_hiaddr, align);
3048     }
3049 
3050     /* Reserve and initialize the commpage. */
3051     if (!init_guest_commpage()) {
3052         /* We have already probed for the commpage being free. */
3053         g_assert_not_reached();
3054     }
3055 
3056     assert(QEMU_IS_ALIGNED(guest_base, align));
3057     qemu_log_mask(CPU_LOG_PAGE, "Locating guest address space "
3058                   "@ 0x%" PRIx64 "\n", (uint64_t)guest_base);
3059 }
3060 
3061 enum {
3062     /* The string "GNU\0" as a magic number. */
3063     GNU0_MAGIC = const_le32('G' | 'N' << 8 | 'U' << 16),
3064     NOTE_DATA_SZ = 1 * KiB,
3065     NOTE_NAME_SZ = 4,
3066     ELF_GNU_PROPERTY_ALIGN = ELF_CLASS == ELFCLASS32 ? 4 : 8,
3067 };
3068 
3069 /*
3070  * Process a single gnu_property entry.
3071  * Return false for error.
3072  */
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)3073 static bool parse_elf_property(const uint32_t *data, int *off, int datasz,
3074                                struct image_info *info, bool have_prev_type,
3075                                uint32_t *prev_type, Error **errp)
3076 {
3077     uint32_t pr_type, pr_datasz, step;
3078 
3079     if (*off > datasz || !QEMU_IS_ALIGNED(*off, ELF_GNU_PROPERTY_ALIGN)) {
3080         goto error_data;
3081     }
3082     datasz -= *off;
3083     data += *off / sizeof(uint32_t);
3084 
3085     if (datasz < 2 * sizeof(uint32_t)) {
3086         goto error_data;
3087     }
3088     pr_type = data[0];
3089     pr_datasz = data[1];
3090     data += 2;
3091     datasz -= 2 * sizeof(uint32_t);
3092     step = ROUND_UP(pr_datasz, ELF_GNU_PROPERTY_ALIGN);
3093     if (step > datasz) {
3094         goto error_data;
3095     }
3096 
3097     /* Properties are supposed to be unique and sorted on pr_type. */
3098     if (have_prev_type && pr_type <= *prev_type) {
3099         if (pr_type == *prev_type) {
3100             error_setg(errp, "Duplicate property in PT_GNU_PROPERTY");
3101         } else {
3102             error_setg(errp, "Unsorted property in PT_GNU_PROPERTY");
3103         }
3104         return false;
3105     }
3106     *prev_type = pr_type;
3107 
3108     if (!arch_parse_elf_property(pr_type, pr_datasz, data, info, errp)) {
3109         return false;
3110     }
3111 
3112     *off += 2 * sizeof(uint32_t) + step;
3113     return true;
3114 
3115  error_data:
3116     error_setg(errp, "Ill-formed property in PT_GNU_PROPERTY");
3117     return false;
3118 }
3119 
3120 /* Process NT_GNU_PROPERTY_TYPE_0. */
parse_elf_properties(const ImageSource * src,struct image_info * info,const struct elf_phdr * phdr,Error ** errp)3121 static bool parse_elf_properties(const ImageSource *src,
3122                                  struct image_info *info,
3123                                  const struct elf_phdr *phdr,
3124                                  Error **errp)
3125 {
3126     union {
3127         struct elf_note nhdr;
3128         uint32_t data[NOTE_DATA_SZ / sizeof(uint32_t)];
3129     } note;
3130 
3131     int n, off, datasz;
3132     bool have_prev_type;
3133     uint32_t prev_type;
3134 
3135     /* Unless the arch requires properties, ignore them. */
3136     if (!ARCH_USE_GNU_PROPERTY) {
3137         return true;
3138     }
3139 
3140     /* If the properties are crazy large, that's too bad. */
3141     n = phdr->p_filesz;
3142     if (n > sizeof(note)) {
3143         error_setg(errp, "PT_GNU_PROPERTY too large");
3144         return false;
3145     }
3146     if (n < sizeof(note.nhdr)) {
3147         error_setg(errp, "PT_GNU_PROPERTY too small");
3148         return false;
3149     }
3150 
3151     if (!imgsrc_read(&note, phdr->p_offset, n, src, errp)) {
3152         return false;
3153     }
3154 
3155     /*
3156      * The contents of a valid PT_GNU_PROPERTY is a sequence of uint32_t.
3157      * Swap most of them now, beyond the header and namesz.
3158      */
3159 #ifdef BSWAP_NEEDED
3160     for (int i = 4; i < n / 4; i++) {
3161         bswap32s(note.data + i);
3162     }
3163 #endif
3164 
3165     /*
3166      * Note that nhdr is 3 words, and that the "name" described by namesz
3167      * immediately follows nhdr and is thus at the 4th word.  Further, all
3168      * of the inputs to the kernel's round_up are multiples of 4.
3169      */
3170     if (tswap32(note.nhdr.n_type) != NT_GNU_PROPERTY_TYPE_0 ||
3171         tswap32(note.nhdr.n_namesz) != NOTE_NAME_SZ ||
3172         note.data[3] != GNU0_MAGIC) {
3173         error_setg(errp, "Invalid note in PT_GNU_PROPERTY");
3174         return false;
3175     }
3176     off = sizeof(note.nhdr) + NOTE_NAME_SZ;
3177 
3178     datasz = tswap32(note.nhdr.n_descsz) + off;
3179     if (datasz > n) {
3180         error_setg(errp, "Invalid note size in PT_GNU_PROPERTY");
3181         return false;
3182     }
3183 
3184     have_prev_type = false;
3185     prev_type = 0;
3186     while (1) {
3187         if (off == datasz) {
3188             return true;  /* end, exit ok */
3189         }
3190         if (!parse_elf_property(note.data, &off, datasz, info,
3191                                 have_prev_type, &prev_type, errp)) {
3192             return false;
3193         }
3194         have_prev_type = true;
3195     }
3196 }
3197 
3198 /**
3199  * load_elf_image: Load an ELF image into the address space.
3200  * @image_name: the filename of the image, to use in error messages.
3201  * @src: the ImageSource from which to read.
3202  * @info: info collected from the loaded image.
3203  * @ehdr: the ELF header, not yet bswapped.
3204  * @pinterp_name: record any PT_INTERP string found.
3205  *
3206  * On return: @info values will be filled in, as necessary or available.
3207  */
3208 
load_elf_image(const char * image_name,const ImageSource * src,struct image_info * info,struct elfhdr * ehdr,char ** pinterp_name)3209 static void load_elf_image(const char *image_name, const ImageSource *src,
3210                            struct image_info *info, struct elfhdr *ehdr,
3211                            char **pinterp_name)
3212 {
3213     g_autofree struct elf_phdr *phdr = NULL;
3214     abi_ulong load_addr, load_bias, loaddr, hiaddr, error;
3215     int i, prot_exec;
3216     Error *err = NULL;
3217 
3218     /*
3219      * First of all, some simple consistency checks.
3220      * Note that we rely on the bswapped ehdr staying in bprm_buf,
3221      * for later use by load_elf_binary and create_elf_tables.
3222      */
3223     if (!imgsrc_read(ehdr, 0, sizeof(*ehdr), src, &err)) {
3224         goto exit_errmsg;
3225     }
3226     if (!elf_check_ident(ehdr)) {
3227         error_setg(&err, "Invalid ELF image for this architecture");
3228         goto exit_errmsg;
3229     }
3230     bswap_ehdr(ehdr);
3231     if (!elf_check_ehdr(ehdr)) {
3232         error_setg(&err, "Invalid ELF image for this architecture");
3233         goto exit_errmsg;
3234     }
3235 
3236     phdr = imgsrc_read_alloc(ehdr->e_phoff,
3237                              ehdr->e_phnum * sizeof(struct elf_phdr),
3238                              src, &err);
3239     if (phdr == NULL) {
3240         goto exit_errmsg;
3241     }
3242     bswap_phdr(phdr, ehdr->e_phnum);
3243 
3244     info->nsegs = 0;
3245     info->pt_dynamic_addr = 0;
3246 
3247     mmap_lock();
3248 
3249     /*
3250      * Find the maximum size of the image and allocate an appropriate
3251      * amount of memory to handle that.  Locate the interpreter, if any.
3252      */
3253     loaddr = -1, hiaddr = 0;
3254     info->alignment = 0;
3255     info->exec_stack = EXSTACK_DEFAULT;
3256     for (i = 0; i < ehdr->e_phnum; ++i) {
3257         struct elf_phdr *eppnt = phdr + i;
3258         if (eppnt->p_type == PT_LOAD) {
3259             abi_ulong a = eppnt->p_vaddr & TARGET_PAGE_MASK;
3260             if (a < loaddr) {
3261                 loaddr = a;
3262             }
3263             a = eppnt->p_vaddr + eppnt->p_memsz - 1;
3264             if (a > hiaddr) {
3265                 hiaddr = a;
3266             }
3267             ++info->nsegs;
3268             info->alignment |= eppnt->p_align;
3269         } else if (eppnt->p_type == PT_INTERP && pinterp_name) {
3270             g_autofree char *interp_name = NULL;
3271 
3272             if (*pinterp_name) {
3273                 error_setg(&err, "Multiple PT_INTERP entries");
3274                 goto exit_errmsg;
3275             }
3276 
3277             interp_name = imgsrc_read_alloc(eppnt->p_offset, eppnt->p_filesz,
3278                                             src, &err);
3279             if (interp_name == NULL) {
3280                 goto exit_errmsg;
3281             }
3282             if (interp_name[eppnt->p_filesz - 1] != 0) {
3283                 error_setg(&err, "Invalid PT_INTERP entry");
3284                 goto exit_errmsg;
3285             }
3286             *pinterp_name = g_steal_pointer(&interp_name);
3287         } else if (eppnt->p_type == PT_GNU_PROPERTY) {
3288             if (!parse_elf_properties(src, info, eppnt, &err)) {
3289                 goto exit_errmsg;
3290             }
3291         } else if (eppnt->p_type == PT_GNU_STACK) {
3292             info->exec_stack = eppnt->p_flags & PF_X;
3293         }
3294     }
3295 
3296     load_addr = loaddr;
3297 
3298     if (pinterp_name != NULL) {
3299         if (ehdr->e_type == ET_EXEC) {
3300             /*
3301              * Make sure that the low address does not conflict with
3302              * MMAP_MIN_ADDR or the QEMU application itself.
3303              */
3304             probe_guest_base(image_name, loaddr, hiaddr);
3305         } else {
3306             abi_ulong align;
3307 
3308             /*
3309              * The binary is dynamic, but we still need to
3310              * select guest_base.  In this case we pass a size.
3311              */
3312             probe_guest_base(image_name, 0, hiaddr - loaddr);
3313 
3314             /*
3315              * Avoid collision with the loader by providing a different
3316              * default load address.
3317              */
3318             load_addr += elf_et_dyn_base;
3319 
3320             /*
3321              * TODO: Better support for mmap alignment is desirable.
3322              * Since we do not have complete control over the guest
3323              * address space, we prefer the kernel to choose some address
3324              * rather than force the use of LOAD_ADDR via MAP_FIXED.
3325              * But without MAP_FIXED we cannot guarantee alignment,
3326              * only suggest it.
3327              */
3328             align = pow2ceil(info->alignment);
3329             if (align) {
3330                 load_addr &= -align;
3331             }
3332         }
3333     }
3334 
3335     /*
3336      * Reserve address space for all of this.
3337      *
3338      * In the case of ET_EXEC, we supply MAP_FIXED_NOREPLACE so that we get
3339      * exactly the address range that is required.  Without reserved_va,
3340      * the guest address space is not isolated.  We have attempted to avoid
3341      * conflict with the host program itself via probe_guest_base, but using
3342      * MAP_FIXED_NOREPLACE instead of MAP_FIXED provides an extra check.
3343      *
3344      * Otherwise this is ET_DYN, and we are searching for a location
3345      * that can hold the memory space required.  If the image is
3346      * pre-linked, LOAD_ADDR will be non-zero, and the kernel should
3347      * honor that address if it happens to be free.
3348      *
3349      * In both cases, we will overwrite pages in this range with mappings
3350      * from the executable.
3351      */
3352     load_addr = target_mmap(load_addr, (size_t)hiaddr - loaddr + 1, PROT_NONE,
3353                             MAP_PRIVATE | MAP_ANON | MAP_NORESERVE |
3354                             (ehdr->e_type == ET_EXEC ? MAP_FIXED_NOREPLACE : 0),
3355                             -1, 0);
3356     if (load_addr == -1) {
3357         goto exit_mmap;
3358     }
3359     load_bias = load_addr - loaddr;
3360 
3361     if (elf_is_fdpic(ehdr)) {
3362         struct elf32_fdpic_loadseg *loadsegs = info->loadsegs =
3363             g_malloc(sizeof(*loadsegs) * info->nsegs);
3364 
3365         for (i = 0; i < ehdr->e_phnum; ++i) {
3366             switch (phdr[i].p_type) {
3367             case PT_DYNAMIC:
3368                 info->pt_dynamic_addr = phdr[i].p_vaddr + load_bias;
3369                 break;
3370             case PT_LOAD:
3371                 loadsegs->addr = phdr[i].p_vaddr + load_bias;
3372                 loadsegs->p_vaddr = phdr[i].p_vaddr;
3373                 loadsegs->p_memsz = phdr[i].p_memsz;
3374                 ++loadsegs;
3375                 break;
3376             }
3377         }
3378     }
3379 
3380     info->load_bias = load_bias;
3381     info->code_offset = load_bias;
3382     info->data_offset = load_bias;
3383     info->load_addr = load_addr;
3384     info->entry = ehdr->e_entry + load_bias;
3385     info->start_code = -1;
3386     info->end_code = 0;
3387     info->start_data = -1;
3388     info->end_data = 0;
3389     /* Usual start for brk is after all sections of the main executable. */
3390     info->brk = TARGET_PAGE_ALIGN(hiaddr + load_bias);
3391     info->elf_flags = ehdr->e_flags;
3392 
3393     prot_exec = PROT_EXEC;
3394 #ifdef TARGET_AARCH64
3395     /*
3396      * If the BTI feature is present, this indicates that the executable
3397      * pages of the startup binary should be mapped with PROT_BTI, so that
3398      * branch targets are enforced.
3399      *
3400      * The startup binary is either the interpreter or the static executable.
3401      * The interpreter is responsible for all pages of a dynamic executable.
3402      *
3403      * Elf notes are backward compatible to older cpus.
3404      * Do not enable BTI unless it is supported.
3405      */
3406     if ((info->note_flags & GNU_PROPERTY_AARCH64_FEATURE_1_BTI)
3407         && (pinterp_name == NULL || *pinterp_name == 0)
3408         && cpu_isar_feature(aa64_bti, ARM_CPU(thread_cpu))) {
3409         prot_exec |= TARGET_PROT_BTI;
3410     }
3411 #endif
3412 
3413     for (i = 0; i < ehdr->e_phnum; i++) {
3414         struct elf_phdr *eppnt = phdr + i;
3415         if (eppnt->p_type == PT_LOAD) {
3416             abi_ulong vaddr, vaddr_po, vaddr_ps, vaddr_ef, vaddr_em;
3417             int elf_prot = 0;
3418 
3419             if (eppnt->p_flags & PF_R) {
3420                 elf_prot |= PROT_READ;
3421             }
3422             if (eppnt->p_flags & PF_W) {
3423                 elf_prot |= PROT_WRITE;
3424             }
3425             if (eppnt->p_flags & PF_X) {
3426                 elf_prot |= prot_exec;
3427             }
3428 
3429             vaddr = load_bias + eppnt->p_vaddr;
3430             vaddr_po = vaddr & ~TARGET_PAGE_MASK;
3431             vaddr_ps = vaddr & TARGET_PAGE_MASK;
3432 
3433             vaddr_ef = vaddr + eppnt->p_filesz;
3434             vaddr_em = vaddr + eppnt->p_memsz;
3435 
3436             /*
3437              * Some segments may be completely empty, with a non-zero p_memsz
3438              * but no backing file segment.
3439              */
3440             if (eppnt->p_filesz != 0) {
3441                 error = imgsrc_mmap(vaddr_ps, eppnt->p_filesz + vaddr_po,
3442                                     elf_prot, MAP_PRIVATE | MAP_FIXED,
3443                                     src, eppnt->p_offset - vaddr_po);
3444                 if (error == -1) {
3445                     goto exit_mmap;
3446                 }
3447             }
3448 
3449             /* If the load segment requests extra zeros (e.g. bss), map it. */
3450             if (vaddr_ef < vaddr_em &&
3451                 !zero_bss(vaddr_ef, vaddr_em, elf_prot, &err)) {
3452                 goto exit_errmsg;
3453             }
3454 
3455             /* Find the full program boundaries.  */
3456             if (elf_prot & PROT_EXEC) {
3457                 if (vaddr < info->start_code) {
3458                     info->start_code = vaddr;
3459                 }
3460                 if (vaddr_ef > info->end_code) {
3461                     info->end_code = vaddr_ef;
3462                 }
3463             }
3464             if (elf_prot & PROT_WRITE) {
3465                 if (vaddr < info->start_data) {
3466                     info->start_data = vaddr;
3467                 }
3468                 if (vaddr_ef > info->end_data) {
3469                     info->end_data = vaddr_ef;
3470                 }
3471             }
3472 #ifdef TARGET_MIPS
3473         } else if (eppnt->p_type == PT_MIPS_ABIFLAGS) {
3474             Mips_elf_abiflags_v0 abiflags;
3475 
3476             if (!imgsrc_read(&abiflags, eppnt->p_offset, sizeof(abiflags),
3477                              src, &err)) {
3478                 goto exit_errmsg;
3479             }
3480             bswap_mips_abiflags(&abiflags);
3481             info->fp_abi = abiflags.fp_abi;
3482 #endif
3483         }
3484     }
3485 
3486     if (info->end_data == 0) {
3487         info->start_data = info->end_code;
3488         info->end_data = info->end_code;
3489     }
3490 
3491     if (qemu_log_enabled()) {
3492         load_symbols(ehdr, src, load_bias);
3493     }
3494 
3495     debuginfo_report_elf(image_name, src->fd, load_bias);
3496 
3497     mmap_unlock();
3498 
3499     close(src->fd);
3500     return;
3501 
3502  exit_mmap:
3503     error_setg_errno(&err, errno, "Error mapping file");
3504     goto exit_errmsg;
3505  exit_errmsg:
3506     error_reportf_err(err, "%s: ", image_name);
3507     exit(-1);
3508 }
3509 
load_elf_interp(const char * filename,struct image_info * info,char bprm_buf[BPRM_BUF_SIZE])3510 static void load_elf_interp(const char *filename, struct image_info *info,
3511                             char bprm_buf[BPRM_BUF_SIZE])
3512 {
3513     struct elfhdr ehdr;
3514     ImageSource src;
3515     int fd, retval;
3516     Error *err = NULL;
3517 
3518     fd = open(path(filename), O_RDONLY);
3519     if (fd < 0) {
3520         error_setg_file_open(&err, errno, filename);
3521         error_report_err(err);
3522         exit(-1);
3523     }
3524 
3525     retval = read(fd, bprm_buf, BPRM_BUF_SIZE);
3526     if (retval < 0) {
3527         error_setg_errno(&err, errno, "Error reading file header");
3528         error_reportf_err(err, "%s: ", filename);
3529         exit(-1);
3530     }
3531 
3532     src.fd = fd;
3533     src.cache = bprm_buf;
3534     src.cache_size = retval;
3535 
3536     load_elf_image(filename, &src, info, &ehdr, NULL);
3537 }
3538 
3539 #ifndef vdso_image_info
3540 #ifdef VDSO_HEADER
3541 #include VDSO_HEADER
3542 #define  vdso_image_info(flags)  &vdso_image_info
3543 #else
3544 #define  vdso_image_info(flags)  NULL
3545 #endif /* VDSO_HEADER */
3546 #endif /* vdso_image_info */
3547 
load_elf_vdso(struct image_info * info,const VdsoImageInfo * vdso)3548 static void load_elf_vdso(struct image_info *info, const VdsoImageInfo *vdso)
3549 {
3550     ImageSource src;
3551     struct elfhdr ehdr;
3552     abi_ulong load_bias, load_addr;
3553 
3554     src.fd = -1;
3555     src.cache = vdso->image;
3556     src.cache_size = vdso->image_size;
3557 
3558     load_elf_image("<internal-vdso>", &src, info, &ehdr, NULL);
3559     load_addr = info->load_addr;
3560     load_bias = info->load_bias;
3561 
3562     /*
3563      * We need to relocate the VDSO image.  The one built into the kernel
3564      * is built for a fixed address.  The one built for QEMU is not, since
3565      * that requires close control of the guest address space.
3566      * We pre-processed the image to locate all of the addresses that need
3567      * to be updated.
3568      */
3569     for (unsigned i = 0, n = vdso->reloc_count; i < n; i++) {
3570         abi_ulong *addr = g2h_untagged(load_addr + vdso->relocs[i]);
3571         *addr = tswapal(tswapal(*addr) + load_bias);
3572     }
3573 
3574     /* Install signal trampolines, if present. */
3575     if (vdso->sigreturn_ofs) {
3576         default_sigreturn = load_addr + vdso->sigreturn_ofs;
3577     }
3578     if (vdso->rt_sigreturn_ofs) {
3579         default_rt_sigreturn = load_addr + vdso->rt_sigreturn_ofs;
3580     }
3581 
3582     /* Remove write from VDSO segment. */
3583     target_mprotect(info->start_data, info->end_data - info->start_data,
3584                     PROT_READ | PROT_EXEC);
3585 }
3586 
symfind(const void * s0,const void * s1)3587 static int symfind(const void *s0, const void *s1)
3588 {
3589     struct elf_sym *sym = (struct elf_sym *)s1;
3590     __typeof(sym->st_value) addr = *(uint64_t *)s0;
3591     int result = 0;
3592 
3593     if (addr < sym->st_value) {
3594         result = -1;
3595     } else if (addr >= sym->st_value + sym->st_size) {
3596         result = 1;
3597     }
3598     return result;
3599 }
3600 
lookup_symbolxx(struct syminfo * s,uint64_t orig_addr)3601 static const char *lookup_symbolxx(struct syminfo *s, uint64_t orig_addr)
3602 {
3603 #if ELF_CLASS == ELFCLASS32
3604     struct elf_sym *syms = s->disas_symtab.elf32;
3605 #else
3606     struct elf_sym *syms = s->disas_symtab.elf64;
3607 #endif
3608 
3609     // binary search
3610     struct elf_sym *sym;
3611 
3612     sym = bsearch(&orig_addr, syms, s->disas_num_syms, sizeof(*syms), symfind);
3613     if (sym != NULL) {
3614         return s->disas_strtab + sym->st_name;
3615     }
3616 
3617     return "";
3618 }
3619 
3620 /* FIXME: This should use elf_ops.h.inc  */
symcmp(const void * s0,const void * s1)3621 static int symcmp(const void *s0, const void *s1)
3622 {
3623     struct elf_sym *sym0 = (struct elf_sym *)s0;
3624     struct elf_sym *sym1 = (struct elf_sym *)s1;
3625     return (sym0->st_value < sym1->st_value)
3626         ? -1
3627         : ((sym0->st_value > sym1->st_value) ? 1 : 0);
3628 }
3629 
3630 /* Best attempt to load symbols from this ELF object. */
load_symbols(struct elfhdr * hdr,const ImageSource * src,abi_ulong load_bias)3631 static void load_symbols(struct elfhdr *hdr, const ImageSource *src,
3632                          abi_ulong load_bias)
3633 {
3634     int i, shnum, nsyms, sym_idx = 0, str_idx = 0;
3635     g_autofree struct elf_shdr *shdr = NULL;
3636     char *strings = NULL;
3637     struct elf_sym *syms = NULL;
3638     struct elf_sym *new_syms;
3639     uint64_t segsz;
3640 
3641     shnum = hdr->e_shnum;
3642     shdr = imgsrc_read_alloc(hdr->e_shoff, shnum * sizeof(struct elf_shdr),
3643                              src, NULL);
3644     if (shdr == NULL) {
3645         return;
3646     }
3647 
3648     bswap_shdr(shdr, shnum);
3649     for (i = 0; i < shnum; ++i) {
3650         if (shdr[i].sh_type == SHT_SYMTAB) {
3651             sym_idx = i;
3652             str_idx = shdr[i].sh_link;
3653             goto found;
3654         }
3655     }
3656 
3657     /* There will be no symbol table if the file was stripped.  */
3658     return;
3659 
3660  found:
3661     /* Now know where the strtab and symtab are.  Snarf them.  */
3662 
3663     segsz = shdr[str_idx].sh_size;
3664     strings = g_try_malloc(segsz);
3665     if (!strings) {
3666         goto give_up;
3667     }
3668     if (!imgsrc_read(strings, shdr[str_idx].sh_offset, segsz, src, NULL)) {
3669         goto give_up;
3670     }
3671 
3672     segsz = shdr[sym_idx].sh_size;
3673     if (segsz / sizeof(struct elf_sym) > INT_MAX) {
3674         /*
3675          * Implausibly large symbol table: give up rather than ploughing
3676          * on with the number of symbols calculation overflowing.
3677          */
3678         goto give_up;
3679     }
3680     nsyms = segsz / sizeof(struct elf_sym);
3681     syms = g_try_malloc(segsz);
3682     if (!syms) {
3683         goto give_up;
3684     }
3685     if (!imgsrc_read(syms, shdr[sym_idx].sh_offset, segsz, src, NULL)) {
3686         goto give_up;
3687     }
3688 
3689     for (i = 0; i < nsyms; ) {
3690         bswap_sym(syms + i);
3691         /* Throw away entries which we do not need.  */
3692         if (syms[i].st_shndx == SHN_UNDEF
3693             || syms[i].st_shndx >= SHN_LORESERVE
3694             || ELF_ST_TYPE(syms[i].st_info) != STT_FUNC) {
3695             if (i < --nsyms) {
3696                 syms[i] = syms[nsyms];
3697             }
3698         } else {
3699 #if defined(TARGET_ARM) || defined (TARGET_MIPS)
3700             /* The bottom address bit marks a Thumb or MIPS16 symbol.  */
3701             syms[i].st_value &= ~(target_ulong)1;
3702 #endif
3703             syms[i].st_value += load_bias;
3704             i++;
3705         }
3706     }
3707 
3708     /* No "useful" symbol.  */
3709     if (nsyms == 0) {
3710         goto give_up;
3711     }
3712 
3713     /*
3714      * Attempt to free the storage associated with the local symbols
3715      * that we threw away.  Whether or not this has any effect on the
3716      * memory allocation depends on the malloc implementation and how
3717      * many symbols we managed to discard.
3718      */
3719     new_syms = g_try_renew(struct elf_sym, syms, nsyms);
3720     if (new_syms == NULL) {
3721         goto give_up;
3722     }
3723     syms = new_syms;
3724 
3725     qsort(syms, nsyms, sizeof(*syms), symcmp);
3726 
3727     {
3728         struct syminfo *s = g_new(struct syminfo, 1);
3729 
3730         s->disas_strtab = strings;
3731         s->disas_num_syms = nsyms;
3732 #if ELF_CLASS == ELFCLASS32
3733         s->disas_symtab.elf32 = syms;
3734 #else
3735         s->disas_symtab.elf64 = syms;
3736 #endif
3737         s->lookup_symbol = lookup_symbolxx;
3738         s->next = syminfos;
3739         syminfos = s;
3740     }
3741     return;
3742 
3743  give_up:
3744     g_free(strings);
3745     g_free(syms);
3746 }
3747 
get_elf_eflags(int fd)3748 uint32_t get_elf_eflags(int fd)
3749 {
3750     struct elfhdr ehdr;
3751     off_t offset;
3752     int ret;
3753 
3754     /* Read ELF header */
3755     offset = lseek(fd, 0, SEEK_SET);
3756     if (offset == (off_t) -1) {
3757         return 0;
3758     }
3759     ret = read(fd, &ehdr, sizeof(ehdr));
3760     if (ret < sizeof(ehdr)) {
3761         return 0;
3762     }
3763     offset = lseek(fd, offset, SEEK_SET);
3764     if (offset == (off_t) -1) {
3765         return 0;
3766     }
3767 
3768     /* Check ELF signature */
3769     if (!elf_check_ident(&ehdr)) {
3770         return 0;
3771     }
3772 
3773     /* check header */
3774     bswap_ehdr(&ehdr);
3775     if (!elf_check_ehdr(&ehdr)) {
3776         return 0;
3777     }
3778 
3779     /* return architecture id */
3780     return ehdr.e_flags;
3781 }
3782 
load_elf_binary(struct linux_binprm * bprm,struct image_info * info)3783 int load_elf_binary(struct linux_binprm *bprm, struct image_info *info)
3784 {
3785     /*
3786      * We need a copy of the elf header for passing to create_elf_tables.
3787      * We will have overwritten the original when we re-use bprm->buf
3788      * while loading the interpreter.  Allocate the storage for this now
3789      * and let elf_load_image do any swapping that may be required.
3790      */
3791     struct elfhdr ehdr;
3792     struct image_info interp_info, vdso_info;
3793     char *elf_interpreter = NULL;
3794     char *scratch;
3795 
3796     memset(&interp_info, 0, sizeof(interp_info));
3797 #ifdef TARGET_MIPS
3798     interp_info.fp_abi = MIPS_ABI_FP_UNKNOWN;
3799 #endif
3800 
3801     load_elf_image(bprm->filename, &bprm->src, info, &ehdr, &elf_interpreter);
3802 
3803     /* Do this so that we can load the interpreter, if need be.  We will
3804        change some of these later */
3805     bprm->p = setup_arg_pages(bprm, info);
3806 
3807     scratch = g_new0(char, TARGET_PAGE_SIZE);
3808     if (STACK_GROWS_DOWN) {
3809         bprm->p = copy_elf_strings(1, &bprm->filename, scratch,
3810                                    bprm->p, info->stack_limit);
3811         info->file_string = bprm->p;
3812         bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch,
3813                                    bprm->p, info->stack_limit);
3814         info->env_strings = bprm->p;
3815         bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch,
3816                                    bprm->p, info->stack_limit);
3817         info->arg_strings = bprm->p;
3818     } else {
3819         info->arg_strings = bprm->p;
3820         bprm->p = copy_elf_strings(bprm->argc, bprm->argv, scratch,
3821                                    bprm->p, info->stack_limit);
3822         info->env_strings = bprm->p;
3823         bprm->p = copy_elf_strings(bprm->envc, bprm->envp, scratch,
3824                                    bprm->p, info->stack_limit);
3825         info->file_string = bprm->p;
3826         bprm->p = copy_elf_strings(1, &bprm->filename, scratch,
3827                                    bprm->p, info->stack_limit);
3828     }
3829 
3830     g_free(scratch);
3831 
3832     if (!bprm->p) {
3833         fprintf(stderr, "%s: %s\n", bprm->filename, strerror(E2BIG));
3834         exit(-1);
3835     }
3836 
3837     if (elf_interpreter) {
3838         load_elf_interp(elf_interpreter, &interp_info, bprm->buf);
3839 
3840         /*
3841          * While unusual because of ELF_ET_DYN_BASE, if we are unlucky
3842          * with the mappings the interpreter can be loaded above but
3843          * near the main executable, which can leave very little room
3844          * for the heap.
3845          * If the current brk has less than 16MB, use the end of the
3846          * interpreter.
3847          */
3848         if (interp_info.brk > info->brk &&
3849             interp_info.load_bias - info->brk < 16 * MiB)  {
3850             info->brk = interp_info.brk;
3851         }
3852 
3853         /* If the program interpreter is one of these two, then assume
3854            an iBCS2 image.  Otherwise assume a native linux image.  */
3855 
3856         if (strcmp(elf_interpreter, "/usr/lib/libc.so.1") == 0
3857             || strcmp(elf_interpreter, "/usr/lib/ld.so.1") == 0) {
3858             info->personality = PER_SVR4;
3859 
3860             /* Why this, you ask???  Well SVr4 maps page 0 as read-only,
3861                and some applications "depend" upon this behavior.  Since
3862                we do not have the power to recompile these, we emulate
3863                the SVr4 behavior.  Sigh.  */
3864             target_mmap(0, TARGET_PAGE_SIZE, PROT_READ | PROT_EXEC,
3865                         MAP_FIXED_NOREPLACE | MAP_PRIVATE | MAP_ANONYMOUS,
3866                         -1, 0);
3867         }
3868 #ifdef TARGET_MIPS
3869         info->interp_fp_abi = interp_info.fp_abi;
3870 #endif
3871     }
3872 
3873     /*
3874      * Load a vdso if available, which will amongst other things contain the
3875      * signal trampolines.  Otherwise, allocate a separate page for them.
3876      */
3877     const VdsoImageInfo *vdso = vdso_image_info(info->elf_flags);
3878     if (vdso) {
3879         load_elf_vdso(&vdso_info, vdso);
3880         info->vdso = vdso_info.load_bias;
3881     } else if (TARGET_ARCH_HAS_SIGTRAMP_PAGE) {
3882         abi_long tramp_page = target_mmap(0, TARGET_PAGE_SIZE,
3883                                           PROT_READ | PROT_WRITE,
3884                                           MAP_PRIVATE | MAP_ANON, -1, 0);
3885         if (tramp_page == -1) {
3886             return -errno;
3887         }
3888 
3889         setup_sigtramp(tramp_page);
3890         target_mprotect(tramp_page, TARGET_PAGE_SIZE, PROT_READ | PROT_EXEC);
3891     }
3892 
3893     bprm->p = create_elf_tables(bprm->p, bprm->argc, bprm->envc, &ehdr, info,
3894                                 elf_interpreter ? &interp_info : NULL,
3895                                 vdso ? &vdso_info : NULL);
3896     info->start_stack = bprm->p;
3897 
3898     /* If we have an interpreter, set that as the program's entry point.
3899        Copy the load_bias as well, to help PPC64 interpret the entry
3900        point as a function descriptor.  Do this after creating elf tables
3901        so that we copy the original program entry point into the AUXV.  */
3902     if (elf_interpreter) {
3903         info->load_bias = interp_info.load_bias;
3904         info->entry = interp_info.entry;
3905         g_free(elf_interpreter);
3906     }
3907 
3908 #ifdef USE_ELF_CORE_DUMP
3909     bprm->core_dump = &elf_core_dump;
3910 #endif
3911 
3912     return 0;
3913 }
3914 
3915 #ifdef USE_ELF_CORE_DUMP
3916 #include "exec/translate-all.h"
3917 
3918 /*
3919  * Definitions to generate Intel SVR4-like core files.
3920  * These mostly have the same names as the SVR4 types with "target_elf_"
3921  * tacked on the front to prevent clashes with linux definitions,
3922  * and the typedef forms have been avoided.  This is mostly like
3923  * the SVR4 structure, but more Linuxy, with things that Linux does
3924  * not support and which gdb doesn't really use excluded.
3925  *
3926  * Fields we don't dump (their contents is zero) in linux-user qemu
3927  * are marked with XXX.
3928  *
3929  * Core dump code is copied from linux kernel (fs/binfmt_elf.c).
3930  *
3931  * Porting ELF coredump for target is (quite) simple process.  First you
3932  * define USE_ELF_CORE_DUMP in target ELF code (where init_thread() for
3933  * the target resides):
3934  *
3935  * #define USE_ELF_CORE_DUMP
3936  *
3937  * Next you define type of register set used for dumping.  ELF specification
3938  * says that it needs to be array of elf_greg_t that has size of ELF_NREG.
3939  *
3940  * typedef <target_regtype> target_elf_greg_t;
3941  * #define ELF_NREG <number of registers>
3942  * typedef taret_elf_greg_t target_elf_gregset_t[ELF_NREG];
3943  *
3944  * Last step is to implement target specific function that copies registers
3945  * from given cpu into just specified register set.  Prototype is:
3946  *
3947  * static void elf_core_copy_regs(taret_elf_gregset_t *regs,
3948  *                                const CPUArchState *env);
3949  *
3950  * Parameters:
3951  *     regs - copy register values into here (allocated and zeroed by caller)
3952  *     env - copy registers from here
3953  *
3954  * Example for ARM target is provided in this file.
3955  */
3956 
3957 struct target_elf_siginfo {
3958     abi_int    si_signo; /* signal number */
3959     abi_int    si_code;  /* extra code */
3960     abi_int    si_errno; /* errno */
3961 };
3962 
3963 struct target_elf_prstatus {
3964     struct target_elf_siginfo pr_info;      /* Info associated with signal */
3965     abi_short          pr_cursig;    /* Current signal */
3966     abi_ulong          pr_sigpend;   /* XXX */
3967     abi_ulong          pr_sighold;   /* XXX */
3968     target_pid_t       pr_pid;
3969     target_pid_t       pr_ppid;
3970     target_pid_t       pr_pgrp;
3971     target_pid_t       pr_sid;
3972     struct target_timeval pr_utime;  /* XXX User time */
3973     struct target_timeval pr_stime;  /* XXX System time */
3974     struct target_timeval pr_cutime; /* XXX Cumulative user time */
3975     struct target_timeval pr_cstime; /* XXX Cumulative system time */
3976     target_elf_gregset_t      pr_reg;       /* GP registers */
3977     abi_int            pr_fpvalid;   /* XXX */
3978 };
3979 
3980 #define ELF_PRARGSZ     (80) /* Number of chars for args */
3981 
3982 struct target_elf_prpsinfo {
3983     char         pr_state;       /* numeric process state */
3984     char         pr_sname;       /* char for pr_state */
3985     char         pr_zomb;        /* zombie */
3986     char         pr_nice;        /* nice val */
3987     abi_ulong    pr_flag;        /* flags */
3988     target_uid_t pr_uid;
3989     target_gid_t pr_gid;
3990     target_pid_t pr_pid, pr_ppid, pr_pgrp, pr_sid;
3991     /* Lots missing */
3992     char    pr_fname[16] QEMU_NONSTRING; /* filename of executable */
3993     char    pr_psargs[ELF_PRARGSZ]; /* initial part of arg list */
3994 };
3995 
3996 #ifdef BSWAP_NEEDED
bswap_prstatus(struct target_elf_prstatus * prstatus)3997 static void bswap_prstatus(struct target_elf_prstatus *prstatus)
3998 {
3999     prstatus->pr_info.si_signo = tswap32(prstatus->pr_info.si_signo);
4000     prstatus->pr_info.si_code = tswap32(prstatus->pr_info.si_code);
4001     prstatus->pr_info.si_errno = tswap32(prstatus->pr_info.si_errno);
4002     prstatus->pr_cursig = tswap16(prstatus->pr_cursig);
4003     prstatus->pr_sigpend = tswapal(prstatus->pr_sigpend);
4004     prstatus->pr_sighold = tswapal(prstatus->pr_sighold);
4005     prstatus->pr_pid = tswap32(prstatus->pr_pid);
4006     prstatus->pr_ppid = tswap32(prstatus->pr_ppid);
4007     prstatus->pr_pgrp = tswap32(prstatus->pr_pgrp);
4008     prstatus->pr_sid = tswap32(prstatus->pr_sid);
4009     /* cpu times are not filled, so we skip them */
4010     /* regs should be in correct format already */
4011     prstatus->pr_fpvalid = tswap32(prstatus->pr_fpvalid);
4012 }
4013 
bswap_psinfo(struct target_elf_prpsinfo * psinfo)4014 static void bswap_psinfo(struct target_elf_prpsinfo *psinfo)
4015 {
4016     psinfo->pr_flag = tswapal(psinfo->pr_flag);
4017     psinfo->pr_uid = tswap16(psinfo->pr_uid);
4018     psinfo->pr_gid = tswap16(psinfo->pr_gid);
4019     psinfo->pr_pid = tswap32(psinfo->pr_pid);
4020     psinfo->pr_ppid = tswap32(psinfo->pr_ppid);
4021     psinfo->pr_pgrp = tswap32(psinfo->pr_pgrp);
4022     psinfo->pr_sid = tswap32(psinfo->pr_sid);
4023 }
4024 
bswap_note(struct elf_note * en)4025 static void bswap_note(struct elf_note *en)
4026 {
4027     bswap32s(&en->n_namesz);
4028     bswap32s(&en->n_descsz);
4029     bswap32s(&en->n_type);
4030 }
4031 #else
bswap_prstatus(struct target_elf_prstatus * p)4032 static inline void bswap_prstatus(struct target_elf_prstatus *p) { }
bswap_psinfo(struct target_elf_prpsinfo * p)4033 static inline void bswap_psinfo(struct target_elf_prpsinfo *p) {}
bswap_note(struct elf_note * en)4034 static inline void bswap_note(struct elf_note *en) { }
4035 #endif /* BSWAP_NEEDED */
4036 
4037 /*
4038  * Calculate file (dump) size of given memory region.
4039  */
vma_dump_size(target_ulong start,target_ulong end,unsigned long flags)4040 static size_t vma_dump_size(target_ulong start, target_ulong end,
4041                             unsigned long flags)
4042 {
4043     /* The area must be readable. */
4044     if (!(flags & PAGE_READ)) {
4045         return 0;
4046     }
4047 
4048     /*
4049      * Usually we don't dump executable pages as they contain
4050      * non-writable code that debugger can read directly from
4051      * target library etc. If there is no elf header, we dump it.
4052      */
4053     if (!(flags & PAGE_WRITE_ORG) &&
4054         (flags & PAGE_EXEC) &&
4055         memcmp(g2h_untagged(start), ELFMAG, SELFMAG) == 0) {
4056         return 0;
4057     }
4058 
4059     return end - start;
4060 }
4061 
size_note(const char * name,size_t datasz)4062 static size_t size_note(const char *name, size_t datasz)
4063 {
4064     size_t namesz = strlen(name) + 1;
4065 
4066     namesz = ROUND_UP(namesz, 4);
4067     datasz = ROUND_UP(datasz, 4);
4068 
4069     return sizeof(struct elf_note) + namesz + datasz;
4070 }
4071 
fill_note(void ** pptr,int type,const char * name,size_t datasz)4072 static void *fill_note(void **pptr, int type, const char *name, size_t datasz)
4073 {
4074     void *ptr = *pptr;
4075     struct elf_note *n = ptr;
4076     size_t namesz = strlen(name) + 1;
4077 
4078     n->n_namesz = namesz;
4079     n->n_descsz = datasz;
4080     n->n_type = type;
4081     bswap_note(n);
4082 
4083     ptr += sizeof(*n);
4084     memcpy(ptr, name, namesz);
4085 
4086     namesz = ROUND_UP(namesz, 4);
4087     datasz = ROUND_UP(datasz, 4);
4088 
4089     *pptr = ptr + namesz + datasz;
4090     return ptr + namesz;
4091 }
4092 
fill_elf_header(struct elfhdr * elf,int segs,uint16_t machine,uint32_t flags)4093 static void fill_elf_header(struct elfhdr *elf, int segs, uint16_t machine,
4094                             uint32_t flags)
4095 {
4096     memcpy(elf->e_ident, ELFMAG, SELFMAG);
4097 
4098     elf->e_ident[EI_CLASS] = ELF_CLASS;
4099     elf->e_ident[EI_DATA] = ELF_DATA;
4100     elf->e_ident[EI_VERSION] = EV_CURRENT;
4101     elf->e_ident[EI_OSABI] = ELF_OSABI;
4102 
4103     elf->e_type = ET_CORE;
4104     elf->e_machine = machine;
4105     elf->e_version = EV_CURRENT;
4106     elf->e_phoff = sizeof(struct elfhdr);
4107     elf->e_flags = flags;
4108     elf->e_ehsize = sizeof(struct elfhdr);
4109     elf->e_phentsize = sizeof(struct elf_phdr);
4110     elf->e_phnum = segs;
4111 
4112     bswap_ehdr(elf);
4113 }
4114 
fill_elf_note_phdr(struct elf_phdr * phdr,size_t sz,off_t offset)4115 static void fill_elf_note_phdr(struct elf_phdr *phdr, size_t sz, off_t offset)
4116 {
4117     phdr->p_type = PT_NOTE;
4118     phdr->p_offset = offset;
4119     phdr->p_filesz = sz;
4120 
4121     bswap_phdr(phdr, 1);
4122 }
4123 
fill_prstatus_note(void * data,CPUState * cpu,int signr)4124 static void fill_prstatus_note(void *data, CPUState *cpu, int signr)
4125 {
4126     /*
4127      * Because note memory is only aligned to 4, and target_elf_prstatus
4128      * may well have higher alignment requirements, fill locally and
4129      * memcpy to the destination afterward.
4130      */
4131     struct target_elf_prstatus prstatus = {
4132         .pr_info.si_signo = signr,
4133         .pr_cursig = signr,
4134         .pr_pid = get_task_state(cpu)->ts_tid,
4135         .pr_ppid = getppid(),
4136         .pr_pgrp = getpgrp(),
4137         .pr_sid = getsid(0),
4138     };
4139 
4140     elf_core_copy_regs(&prstatus.pr_reg, cpu_env(cpu));
4141     bswap_prstatus(&prstatus);
4142     memcpy(data, &prstatus, sizeof(prstatus));
4143 }
4144 
fill_prpsinfo_note(void * data,const TaskState * ts)4145 static void fill_prpsinfo_note(void *data, const TaskState *ts)
4146 {
4147     /*
4148      * Because note memory is only aligned to 4, and target_elf_prpsinfo
4149      * may well have higher alignment requirements, fill locally and
4150      * memcpy to the destination afterward.
4151      */
4152     struct target_elf_prpsinfo psinfo = {
4153         .pr_pid = getpid(),
4154         .pr_ppid = getppid(),
4155         .pr_pgrp = getpgrp(),
4156         .pr_sid = getsid(0),
4157         .pr_uid = getuid(),
4158         .pr_gid = getgid(),
4159     };
4160     char *base_filename;
4161     size_t len;
4162 
4163     len = ts->info->env_strings - ts->info->arg_strings;
4164     len = MIN(len, ELF_PRARGSZ);
4165     memcpy(&psinfo.pr_psargs, g2h_untagged(ts->info->arg_strings), len);
4166     for (size_t i = 0; i < len; i++) {
4167         if (psinfo.pr_psargs[i] == 0) {
4168             psinfo.pr_psargs[i] = ' ';
4169         }
4170     }
4171 
4172     base_filename = g_path_get_basename(ts->bprm->filename);
4173     /*
4174      * Using strncpy here is fine: at max-length,
4175      * this field is not NUL-terminated.
4176      */
4177     strncpy(psinfo.pr_fname, base_filename, sizeof(psinfo.pr_fname));
4178     g_free(base_filename);
4179 
4180     bswap_psinfo(&psinfo);
4181     memcpy(data, &psinfo, sizeof(psinfo));
4182 }
4183 
fill_auxv_note(void * data,const TaskState * ts)4184 static void fill_auxv_note(void *data, const TaskState *ts)
4185 {
4186     memcpy(data, g2h_untagged(ts->info->saved_auxv), ts->info->auxv_len);
4187 }
4188 
4189 /*
4190  * Constructs name of coredump file.  We have following convention
4191  * for the name:
4192  *     qemu_<basename-of-target-binary>_<date>-<time>_<pid>.core
4193  *
4194  * Returns the filename
4195  */
core_dump_filename(const TaskState * ts)4196 static char *core_dump_filename(const TaskState *ts)
4197 {
4198     g_autoptr(GDateTime) now = g_date_time_new_now_local();
4199     g_autofree char *nowstr = g_date_time_format(now, "%Y%m%d-%H%M%S");
4200     g_autofree char *base_filename = g_path_get_basename(ts->bprm->filename);
4201 
4202     return g_strdup_printf("qemu_%s_%s_%d.core",
4203                            base_filename, nowstr, (int)getpid());
4204 }
4205 
dump_write(int fd,const void * ptr,size_t size)4206 static int dump_write(int fd, const void *ptr, size_t size)
4207 {
4208     const char *bufp = (const char *)ptr;
4209     ssize_t bytes_written, bytes_left;
4210 
4211     bytes_written = 0;
4212     bytes_left = size;
4213 
4214     /*
4215      * In normal conditions, single write(2) should do but
4216      * in case of socket etc. this mechanism is more portable.
4217      */
4218     do {
4219         bytes_written = write(fd, bufp, bytes_left);
4220         if (bytes_written < 0) {
4221             if (errno == EINTR)
4222                 continue;
4223             return (-1);
4224         } else if (bytes_written == 0) { /* eof */
4225             return (-1);
4226         }
4227         bufp += bytes_written;
4228         bytes_left -= bytes_written;
4229     } while (bytes_left > 0);
4230 
4231     return (0);
4232 }
4233 
wmr_page_unprotect_regions(void * opaque,target_ulong start,target_ulong end,unsigned long flags)4234 static int wmr_page_unprotect_regions(void *opaque, target_ulong start,
4235                                       target_ulong end, unsigned long flags)
4236 {
4237     if ((flags & (PAGE_WRITE | PAGE_WRITE_ORG)) == PAGE_WRITE_ORG) {
4238         size_t step = MAX(TARGET_PAGE_SIZE, qemu_real_host_page_size());
4239 
4240         while (1) {
4241             page_unprotect(start, 0);
4242             if (end - start <= step) {
4243                 break;
4244             }
4245             start += step;
4246         }
4247     }
4248     return 0;
4249 }
4250 
4251 typedef struct {
4252     unsigned count;
4253     size_t size;
4254 } CountAndSizeRegions;
4255 
wmr_count_and_size_regions(void * opaque,target_ulong start,target_ulong end,unsigned long flags)4256 static int wmr_count_and_size_regions(void *opaque, target_ulong start,
4257                                       target_ulong end, unsigned long flags)
4258 {
4259     CountAndSizeRegions *css = opaque;
4260 
4261     css->count++;
4262     css->size += vma_dump_size(start, end, flags);
4263     return 0;
4264 }
4265 
4266 typedef struct {
4267     struct elf_phdr *phdr;
4268     off_t offset;
4269 } FillRegionPhdr;
4270 
wmr_fill_region_phdr(void * opaque,target_ulong start,target_ulong end,unsigned long flags)4271 static int wmr_fill_region_phdr(void *opaque, target_ulong start,
4272                                 target_ulong end, unsigned long flags)
4273 {
4274     FillRegionPhdr *d = opaque;
4275     struct elf_phdr *phdr = d->phdr;
4276 
4277     phdr->p_type = PT_LOAD;
4278     phdr->p_vaddr = start;
4279     phdr->p_paddr = 0;
4280     phdr->p_filesz = vma_dump_size(start, end, flags);
4281     phdr->p_offset = d->offset;
4282     d->offset += phdr->p_filesz;
4283     phdr->p_memsz = end - start;
4284     phdr->p_flags = (flags & PAGE_READ ? PF_R : 0)
4285                   | (flags & PAGE_WRITE_ORG ? PF_W : 0)
4286                   | (flags & PAGE_EXEC ? PF_X : 0);
4287     phdr->p_align = ELF_EXEC_PAGESIZE;
4288 
4289     bswap_phdr(phdr, 1);
4290     d->phdr = phdr + 1;
4291     return 0;
4292 }
4293 
wmr_write_region(void * opaque,target_ulong start,target_ulong end,unsigned long flags)4294 static int wmr_write_region(void *opaque, target_ulong start,
4295                             target_ulong end, unsigned long flags)
4296 {
4297     int fd = *(int *)opaque;
4298     size_t size = vma_dump_size(start, end, flags);
4299 
4300     if (!size) {
4301         return 0;
4302     }
4303     return dump_write(fd, g2h_untagged(start), size);
4304 }
4305 
4306 /*
4307  * Write out ELF coredump.
4308  *
4309  * See documentation of ELF object file format in:
4310  * http://www.caldera.com/developers/devspecs/gabi41.pdf
4311  *
4312  * Coredump format in linux is following:
4313  *
4314  * 0   +----------------------+         \
4315  *     | ELF header           | ET_CORE  |
4316  *     +----------------------+          |
4317  *     | ELF program headers  |          |--- headers
4318  *     | - NOTE section       |          |
4319  *     | - PT_LOAD sections   |          |
4320  *     +----------------------+         /
4321  *     | NOTEs:               |
4322  *     | - NT_PRSTATUS        |
4323  *     | - NT_PRSINFO         |
4324  *     | - NT_AUXV            |
4325  *     +----------------------+ <-- aligned to target page
4326  *     | Process memory dump  |
4327  *     :                      :
4328  *     .                      .
4329  *     :                      :
4330  *     |                      |
4331  *     +----------------------+
4332  *
4333  * NT_PRSTATUS -> struct elf_prstatus (per thread)
4334  * NT_PRSINFO  -> struct elf_prpsinfo
4335  * NT_AUXV is array of { type, value } pairs (see fill_auxv_note()).
4336  *
4337  * Format follows System V format as close as possible.  Current
4338  * version limitations are as follows:
4339  *     - no floating point registers are dumped
4340  *
4341  * Function returns 0 in case of success, negative errno otherwise.
4342  *
4343  * TODO: make this work also during runtime: it should be
4344  * possible to force coredump from running process and then
4345  * continue processing.  For example qemu could set up SIGUSR2
4346  * handler (provided that target process haven't registered
4347  * handler for that) that does the dump when signal is received.
4348  */
elf_core_dump(int signr,const CPUArchState * env)4349 static int elf_core_dump(int signr, const CPUArchState *env)
4350 {
4351     const CPUState *cpu = env_cpu((CPUArchState *)env);
4352     const TaskState *ts = (const TaskState *)get_task_state((CPUState *)cpu);
4353     struct rlimit dumpsize;
4354     CountAndSizeRegions css;
4355     off_t offset, note_offset, data_offset;
4356     size_t note_size;
4357     int cpus, ret;
4358     int fd = -1;
4359     CPUState *cpu_iter;
4360 
4361     if (prctl(PR_GET_DUMPABLE) == 0) {
4362         return 0;
4363     }
4364 
4365     if (getrlimit(RLIMIT_CORE, &dumpsize) < 0 || dumpsize.rlim_cur == 0) {
4366         return 0;
4367     }
4368 
4369     cpu_list_lock();
4370     mmap_lock();
4371 
4372     /* By unprotecting, we merge vmas that might be split. */
4373     walk_memory_regions(NULL, wmr_page_unprotect_regions);
4374 
4375     /*
4376      * Walk through target process memory mappings and
4377      * set up structure containing this information.
4378      */
4379     memset(&css, 0, sizeof(css));
4380     walk_memory_regions(&css, wmr_count_and_size_regions);
4381 
4382     cpus = 0;
4383     CPU_FOREACH(cpu_iter) {
4384         cpus++;
4385     }
4386 
4387     offset = sizeof(struct elfhdr);
4388     offset += (css.count + 1) * sizeof(struct elf_phdr);
4389     note_offset = offset;
4390 
4391     offset += size_note("CORE", ts->info->auxv_len);
4392     offset += size_note("CORE", sizeof(struct target_elf_prpsinfo));
4393     offset += size_note("CORE", sizeof(struct target_elf_prstatus)) * cpus;
4394     note_size = offset - note_offset;
4395     data_offset = ROUND_UP(offset, ELF_EXEC_PAGESIZE);
4396 
4397     /* Do not dump if the corefile size exceeds the limit. */
4398     if (dumpsize.rlim_cur != RLIM_INFINITY
4399         && dumpsize.rlim_cur < data_offset + css.size) {
4400         errno = 0;
4401         goto out;
4402     }
4403 
4404     {
4405         g_autofree char *corefile = core_dump_filename(ts);
4406         fd = open(corefile, O_WRONLY | O_CREAT | O_TRUNC,
4407                   S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH);
4408     }
4409     if (fd < 0) {
4410         goto out;
4411     }
4412 
4413     /*
4414      * There is a fair amount of alignment padding within the notes
4415      * as well as preceeding the process memory.  Allocate a zeroed
4416      * block to hold it all.  Write all of the headers directly into
4417      * this buffer and then write it out as a block.
4418      */
4419     {
4420         g_autofree void *header = g_malloc0(data_offset);
4421         FillRegionPhdr frp;
4422         void *hptr, *dptr;
4423 
4424         /* Create elf file header. */
4425         hptr = header;
4426         fill_elf_header(hptr, css.count + 1, ELF_MACHINE, 0);
4427         hptr += sizeof(struct elfhdr);
4428 
4429         /* Create elf program headers. */
4430         fill_elf_note_phdr(hptr, note_size, note_offset);
4431         hptr += sizeof(struct elf_phdr);
4432 
4433         frp.phdr = hptr;
4434         frp.offset = data_offset;
4435         walk_memory_regions(&frp, wmr_fill_region_phdr);
4436         hptr = frp.phdr;
4437 
4438         /* Create the notes. */
4439         dptr = fill_note(&hptr, NT_AUXV, "CORE", ts->info->auxv_len);
4440         fill_auxv_note(dptr, ts);
4441 
4442         dptr = fill_note(&hptr, NT_PRPSINFO, "CORE",
4443                          sizeof(struct target_elf_prpsinfo));
4444         fill_prpsinfo_note(dptr, ts);
4445 
4446         CPU_FOREACH(cpu_iter) {
4447             dptr = fill_note(&hptr, NT_PRSTATUS, "CORE",
4448                              sizeof(struct target_elf_prstatus));
4449             fill_prstatus_note(dptr, cpu_iter, cpu_iter == cpu ? signr : 0);
4450         }
4451 
4452         if (dump_write(fd, header, data_offset) < 0) {
4453             goto out;
4454         }
4455     }
4456 
4457     /*
4458      * Finally write process memory into the corefile as well.
4459      */
4460     if (walk_memory_regions(&fd, wmr_write_region) < 0) {
4461         goto out;
4462     }
4463     errno = 0;
4464 
4465  out:
4466     ret = -errno;
4467     mmap_unlock();
4468     cpu_list_unlock();
4469     if (fd >= 0) {
4470         close(fd);
4471     }
4472     return ret;
4473 }
4474 #endif /* USE_ELF_CORE_DUMP */
4475 
do_init_thread(struct target_pt_regs * regs,struct image_info * infop)4476 void do_init_thread(struct target_pt_regs *regs, struct image_info *infop)
4477 {
4478     init_thread(regs, infop);
4479 }
4480