xref: /qemu/linux-user/main.c (revision 003f1536)
1 /*
2  *  qemu user main
3  *
4  *  Copyright (c) 2003-2008 Fabrice Bellard
5  *
6  *  This program is free software; you can redistribute it and/or modify
7  *  it under the terms of the GNU General Public License as published by
8  *  the Free Software Foundation; either version 2 of the License, or
9  *  (at your option) any later version.
10  *
11  *  This program is distributed in the hope that it will be useful,
12  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
13  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14  *  GNU General Public License for more details.
15  *
16  *  You should have received a copy of the GNU General Public License
17  *  along with this program; if not, see <http://www.gnu.org/licenses/>.
18  */
19 
20 #include "qemu/osdep.h"
21 #include "qemu/help-texts.h"
22 #include "qemu/units.h"
23 #include "qemu/accel.h"
24 #include "qemu-version.h"
25 #include <sys/syscall.h>
26 #include <sys/resource.h>
27 #include <sys/shm.h>
28 #include <linux/binfmts.h>
29 
30 #include "qapi/error.h"
31 #include "qemu.h"
32 #include "user-internals.h"
33 #include "qemu/path.h"
34 #include "qemu/queue.h"
35 #include "qemu/config-file.h"
36 #include "qemu/cutils.h"
37 #include "qemu/error-report.h"
38 #include "qemu/help_option.h"
39 #include "qemu/module.h"
40 #include "qemu/plugin.h"
41 #include "exec/exec-all.h"
42 #include "exec/gdbstub.h"
43 #include "gdbstub/user.h"
44 #include "tcg/startup.h"
45 #include "qemu/timer.h"
46 #include "qemu/envlist.h"
47 #include "qemu/guest-random.h"
48 #include "elf.h"
49 #include "trace/control.h"
50 #include "target_elf.h"
51 #include "cpu_loop-common.h"
52 #include "crypto/init.h"
53 #include "fd-trans.h"
54 #include "signal-common.h"
55 #include "loader.h"
56 #include "user-mmap.h"
57 #include "tcg/perf.h"
58 
59 #ifdef CONFIG_SEMIHOSTING
60 #include "semihosting/semihost.h"
61 #endif
62 
63 #ifndef AT_FLAGS_PRESERVE_ARGV0
64 #define AT_FLAGS_PRESERVE_ARGV0_BIT 0
65 #define AT_FLAGS_PRESERVE_ARGV0 (1 << AT_FLAGS_PRESERVE_ARGV0_BIT)
66 #endif
67 
68 char *exec_path;
69 char real_exec_path[PATH_MAX];
70 
71 static bool opt_one_insn_per_tb;
72 static const char *argv0;
73 static const char *gdbstub;
74 static envlist_t *envlist;
75 static const char *cpu_model;
76 static const char *cpu_type;
77 static const char *seed_optarg;
78 unsigned long mmap_min_addr;
79 uintptr_t guest_base;
80 bool have_guest_base;
81 
82 /*
83  * Used to implement backwards-compatibility for the `-strace`, and
84  * QEMU_STRACE options. Without this, the QEMU_LOG can be overwritten by
85  * -strace, or vice versa.
86  */
87 static bool enable_strace;
88 
89 /*
90  * The last log mask given by the user in an environment variable or argument.
91  * Used to support command line arguments overriding environment variables.
92  */
93 static int last_log_mask;
94 static const char *last_log_filename;
95 
96 /*
97  * When running 32-on-64 we should make sure we can fit all of the possible
98  * guest address space into a contiguous chunk of virtual host memory.
99  *
100  * This way we will never overlap with our own libraries or binaries or stack
101  * or anything else that QEMU maps.
102  *
103  * Many cpus reserve the high bit (or more than one for some 64-bit cpus)
104  * of the address for the kernel.  Some cpus rely on this and user space
105  * uses the high bit(s) for pointer tagging and the like.  For them, we
106  * must preserve the expected address space.
107  */
108 #ifndef MAX_RESERVED_VA
109 # if HOST_LONG_BITS > TARGET_VIRT_ADDR_SPACE_BITS
110 #  if TARGET_VIRT_ADDR_SPACE_BITS == 32 && \
111       (TARGET_LONG_BITS == 32 || defined(TARGET_ABI32))
112 #   define MAX_RESERVED_VA(CPU)  0xfffffffful
113 #  else
114 #   define MAX_RESERVED_VA(CPU)  ((1ul << TARGET_VIRT_ADDR_SPACE_BITS) - 1)
115 #  endif
116 # else
117 #  define MAX_RESERVED_VA(CPU)  0
118 # endif
119 #endif
120 
121 unsigned long reserved_va;
122 
123 static void usage(int exitcode);
124 
125 static const char *interp_prefix = CONFIG_QEMU_INTERP_PREFIX;
126 const char *qemu_uname_release;
127 
128 #if !defined(TARGET_DEFAULT_STACK_SIZE)
129 /* XXX: on x86 MAP_GROWSDOWN only works if ESP <= address + 32, so
130    we allocate a bigger stack. Need a better solution, for example
131    by remapping the process stack directly at the right place */
132 #define TARGET_DEFAULT_STACK_SIZE	8 * 1024 * 1024UL
133 #endif
134 
135 unsigned long guest_stack_size = TARGET_DEFAULT_STACK_SIZE;
136 
137 /***********************************************************/
138 /* Helper routines for implementing atomic operations.  */
139 
140 /* Make sure everything is in a consistent state for calling fork().  */
141 void fork_start(void)
142 {
143     start_exclusive();
144     mmap_fork_start();
145     cpu_list_lock();
146     qemu_plugin_user_prefork_lock();
147 }
148 
149 void fork_end(int child)
150 {
151     qemu_plugin_user_postfork(child);
152     mmap_fork_end(child);
153     if (child) {
154         CPUState *cpu, *next_cpu;
155         /* Child processes created by fork() only have a single thread.
156            Discard information about the parent threads.  */
157         CPU_FOREACH_SAFE(cpu, next_cpu) {
158             if (cpu != thread_cpu) {
159                 QTAILQ_REMOVE_RCU(&cpus_queue, cpu, node);
160             }
161         }
162         qemu_init_cpu_list();
163         gdbserver_fork(thread_cpu);
164     } else {
165         cpu_list_unlock();
166     }
167     /*
168      * qemu_init_cpu_list() reinitialized the child exclusive state, but we
169      * also need to keep current_cpu consistent, so call end_exclusive() for
170      * both child and parent.
171      */
172     end_exclusive();
173 }
174 
175 __thread CPUState *thread_cpu;
176 
177 bool qemu_cpu_is_self(CPUState *cpu)
178 {
179     return thread_cpu == cpu;
180 }
181 
182 void qemu_cpu_kick(CPUState *cpu)
183 {
184     cpu_exit(cpu);
185 }
186 
187 void task_settid(TaskState *ts)
188 {
189     if (ts->ts_tid == 0) {
190         ts->ts_tid = (pid_t)syscall(SYS_gettid);
191     }
192 }
193 
194 void stop_all_tasks(void)
195 {
196     /*
197      * We trust that when using NPTL, start_exclusive()
198      * handles thread stopping correctly.
199      */
200     start_exclusive();
201 }
202 
203 /* Assumes contents are already zeroed.  */
204 void init_task_state(TaskState *ts)
205 {
206     long ticks_per_sec;
207     struct timespec bt;
208 
209     ts->used = 1;
210     ts->sigaltstack_used = (struct target_sigaltstack) {
211         .ss_sp = 0,
212         .ss_size = 0,
213         .ss_flags = TARGET_SS_DISABLE,
214     };
215 
216     /* Capture task start time relative to system boot */
217 
218     ticks_per_sec = sysconf(_SC_CLK_TCK);
219 
220     if ((ticks_per_sec > 0) && !clock_gettime(CLOCK_BOOTTIME, &bt)) {
221         /* start_boottime is expressed in clock ticks */
222         ts->start_boottime = bt.tv_sec * (uint64_t) ticks_per_sec;
223         ts->start_boottime += bt.tv_nsec * (uint64_t) ticks_per_sec /
224                               NANOSECONDS_PER_SECOND;
225     }
226 }
227 
228 CPUArchState *cpu_copy(CPUArchState *env)
229 {
230     CPUState *cpu = env_cpu(env);
231     CPUState *new_cpu = cpu_create(cpu_type);
232     CPUArchState *new_env = cpu_env(new_cpu);
233     CPUBreakpoint *bp;
234 
235     /* Reset non arch specific state */
236     cpu_reset(new_cpu);
237 
238     new_cpu->tcg_cflags = cpu->tcg_cflags;
239     memcpy(new_env, env, sizeof(CPUArchState));
240 #if defined(TARGET_I386) || defined(TARGET_X86_64)
241     new_env->gdt.base = target_mmap(0, sizeof(uint64_t) * TARGET_GDT_ENTRIES,
242                                     PROT_READ | PROT_WRITE,
243                                     MAP_ANONYMOUS | MAP_PRIVATE, -1, 0);
244     memcpy(g2h_untagged(new_env->gdt.base), g2h_untagged(env->gdt.base),
245            sizeof(uint64_t) * TARGET_GDT_ENTRIES);
246     OBJECT(new_cpu)->free = OBJECT(cpu)->free;
247 #endif
248 
249     /* Clone all break/watchpoints.
250        Note: Once we support ptrace with hw-debug register access, make sure
251        BP_CPU break/watchpoints are handled correctly on clone. */
252     QTAILQ_INIT(&new_cpu->breakpoints);
253     QTAILQ_FOREACH(bp, &cpu->breakpoints, entry) {
254         cpu_breakpoint_insert(new_cpu, bp->pc, bp->flags, NULL);
255     }
256 
257     return new_env;
258 }
259 
260 static void handle_arg_help(const char *arg)
261 {
262     usage(EXIT_SUCCESS);
263 }
264 
265 static void handle_arg_log(const char *arg)
266 {
267     last_log_mask = qemu_str_to_log_mask(arg);
268     if (!last_log_mask) {
269         qemu_print_log_usage(stdout);
270         exit(EXIT_FAILURE);
271     }
272 }
273 
274 static void handle_arg_dfilter(const char *arg)
275 {
276     qemu_set_dfilter_ranges(arg, &error_fatal);
277 }
278 
279 static void handle_arg_log_filename(const char *arg)
280 {
281     last_log_filename = arg;
282 }
283 
284 static void handle_arg_set_env(const char *arg)
285 {
286     char *r, *p, *token;
287     r = p = strdup(arg);
288     while ((token = strsep(&p, ",")) != NULL) {
289         if (envlist_setenv(envlist, token) != 0) {
290             usage(EXIT_FAILURE);
291         }
292     }
293     free(r);
294 }
295 
296 static void handle_arg_unset_env(const char *arg)
297 {
298     char *r, *p, *token;
299     r = p = strdup(arg);
300     while ((token = strsep(&p, ",")) != NULL) {
301         if (envlist_unsetenv(envlist, token) != 0) {
302             usage(EXIT_FAILURE);
303         }
304     }
305     free(r);
306 }
307 
308 static void handle_arg_argv0(const char *arg)
309 {
310     argv0 = strdup(arg);
311 }
312 
313 static void handle_arg_stack_size(const char *arg)
314 {
315     char *p;
316     guest_stack_size = strtoul(arg, &p, 0);
317     if (guest_stack_size == 0) {
318         usage(EXIT_FAILURE);
319     }
320 
321     if (*p == 'M') {
322         guest_stack_size *= MiB;
323     } else if (*p == 'k' || *p == 'K') {
324         guest_stack_size *= KiB;
325     }
326 }
327 
328 static void handle_arg_ld_prefix(const char *arg)
329 {
330     interp_prefix = strdup(arg);
331 }
332 
333 static void handle_arg_pagesize(const char *arg)
334 {
335     qemu_host_page_size = atoi(arg);
336     if (qemu_host_page_size == 0 ||
337         (qemu_host_page_size & (qemu_host_page_size - 1)) != 0) {
338         fprintf(stderr, "page size must be a power of two\n");
339         exit(EXIT_FAILURE);
340     }
341 }
342 
343 static void handle_arg_seed(const char *arg)
344 {
345     seed_optarg = arg;
346 }
347 
348 static void handle_arg_gdb(const char *arg)
349 {
350     gdbstub = g_strdup(arg);
351 }
352 
353 static void handle_arg_uname(const char *arg)
354 {
355     qemu_uname_release = strdup(arg);
356 }
357 
358 static void handle_arg_cpu(const char *arg)
359 {
360     cpu_model = strdup(arg);
361     if (cpu_model == NULL || is_help_option(cpu_model)) {
362         list_cpus();
363         exit(EXIT_FAILURE);
364     }
365 }
366 
367 static void handle_arg_guest_base(const char *arg)
368 {
369     guest_base = strtol(arg, NULL, 0);
370     have_guest_base = true;
371 }
372 
373 static void handle_arg_reserved_va(const char *arg)
374 {
375     char *p;
376     int shift = 0;
377     unsigned long val;
378 
379     val = strtoul(arg, &p, 0);
380     switch (*p) {
381     case 'k':
382     case 'K':
383         shift = 10;
384         break;
385     case 'M':
386         shift = 20;
387         break;
388     case 'G':
389         shift = 30;
390         break;
391     }
392     if (shift) {
393         unsigned long unshifted = val;
394         p++;
395         val <<= shift;
396         if (val >> shift != unshifted) {
397             fprintf(stderr, "Reserved virtual address too big\n");
398             exit(EXIT_FAILURE);
399         }
400     }
401     if (*p) {
402         fprintf(stderr, "Unrecognised -R size suffix '%s'\n", p);
403         exit(EXIT_FAILURE);
404     }
405     /* The representation is size - 1, with 0 remaining "default". */
406     reserved_va = val ? val - 1 : 0;
407 }
408 
409 static void handle_arg_one_insn_per_tb(const char *arg)
410 {
411     opt_one_insn_per_tb = true;
412 }
413 
414 static void handle_arg_strace(const char *arg)
415 {
416     enable_strace = true;
417 }
418 
419 static void handle_arg_version(const char *arg)
420 {
421     printf("qemu-" TARGET_NAME " version " QEMU_FULL_VERSION
422            "\n" QEMU_COPYRIGHT "\n");
423     exit(EXIT_SUCCESS);
424 }
425 
426 static void handle_arg_trace(const char *arg)
427 {
428     trace_opt_parse(arg);
429 }
430 
431 #if defined(TARGET_XTENSA)
432 static void handle_arg_abi_call0(const char *arg)
433 {
434     xtensa_set_abi_call0();
435 }
436 #endif
437 
438 static void handle_arg_perfmap(const char *arg)
439 {
440     perf_enable_perfmap();
441 }
442 
443 static void handle_arg_jitdump(const char *arg)
444 {
445     perf_enable_jitdump();
446 }
447 
448 static QemuPluginList plugins = QTAILQ_HEAD_INITIALIZER(plugins);
449 
450 #ifdef CONFIG_PLUGIN
451 static void handle_arg_plugin(const char *arg)
452 {
453     qemu_plugin_opt_parse(arg, &plugins);
454 }
455 #endif
456 
457 struct qemu_argument {
458     const char *argv;
459     const char *env;
460     bool has_arg;
461     void (*handle_opt)(const char *arg);
462     const char *example;
463     const char *help;
464 };
465 
466 static const struct qemu_argument arg_table[] = {
467     {"h",          "",                 false, handle_arg_help,
468      "",           "print this help"},
469     {"help",       "",                 false, handle_arg_help,
470      "",           ""},
471     {"g",          "QEMU_GDB",         true,  handle_arg_gdb,
472      "port",       "wait gdb connection to 'port'"},
473     {"L",          "QEMU_LD_PREFIX",   true,  handle_arg_ld_prefix,
474      "path",       "set the elf interpreter prefix to 'path'"},
475     {"s",          "QEMU_STACK_SIZE",  true,  handle_arg_stack_size,
476      "size",       "set the stack size to 'size' bytes"},
477     {"cpu",        "QEMU_CPU",         true,  handle_arg_cpu,
478      "model",      "select CPU (-cpu help for list)"},
479     {"E",          "QEMU_SET_ENV",     true,  handle_arg_set_env,
480      "var=value",  "sets targets environment variable (see below)"},
481     {"U",          "QEMU_UNSET_ENV",   true,  handle_arg_unset_env,
482      "var",        "unsets targets environment variable (see below)"},
483     {"0",          "QEMU_ARGV0",       true,  handle_arg_argv0,
484      "argv0",      "forces target process argv[0] to be 'argv0'"},
485     {"r",          "QEMU_UNAME",       true,  handle_arg_uname,
486      "uname",      "set qemu uname release string to 'uname'"},
487     {"B",          "QEMU_GUEST_BASE",  true,  handle_arg_guest_base,
488      "address",    "set guest_base address to 'address'"},
489     {"R",          "QEMU_RESERVED_VA", true,  handle_arg_reserved_va,
490      "size",       "reserve 'size' bytes for guest virtual address space"},
491     {"d",          "QEMU_LOG",         true,  handle_arg_log,
492      "item[,...]", "enable logging of specified items "
493      "(use '-d help' for a list of items)"},
494     {"dfilter",    "QEMU_DFILTER",     true,  handle_arg_dfilter,
495      "range[,...]","filter logging based on address range"},
496     {"D",          "QEMU_LOG_FILENAME", true, handle_arg_log_filename,
497      "logfile",     "write logs to 'logfile' (default stderr)"},
498     {"p",          "QEMU_PAGESIZE",    true,  handle_arg_pagesize,
499      "pagesize",   "set the host page size to 'pagesize'"},
500     {"one-insn-per-tb",
501                    "QEMU_ONE_INSN_PER_TB",  false, handle_arg_one_insn_per_tb,
502      "",           "run with one guest instruction per emulated TB"},
503     {"strace",     "QEMU_STRACE",      false, handle_arg_strace,
504      "",           "log system calls"},
505     {"seed",       "QEMU_RAND_SEED",   true,  handle_arg_seed,
506      "",           "Seed for pseudo-random number generator"},
507     {"trace",      "QEMU_TRACE",       true,  handle_arg_trace,
508      "",           "[[enable=]<pattern>][,events=<file>][,file=<file>]"},
509 #ifdef CONFIG_PLUGIN
510     {"plugin",     "QEMU_PLUGIN",      true,  handle_arg_plugin,
511      "",           "[file=]<file>[,<argname>=<argvalue>]"},
512 #endif
513     {"version",    "QEMU_VERSION",     false, handle_arg_version,
514      "",           "display version information and exit"},
515 #if defined(TARGET_XTENSA)
516     {"xtensa-abi-call0", "QEMU_XTENSA_ABI_CALL0", false, handle_arg_abi_call0,
517      "",           "assume CALL0 Xtensa ABI"},
518 #endif
519     {"perfmap",    "QEMU_PERFMAP",     false, handle_arg_perfmap,
520      "",           "Generate a /tmp/perf-${pid}.map file for perf"},
521     {"jitdump",    "QEMU_JITDUMP",     false, handle_arg_jitdump,
522      "",           "Generate a jit-${pid}.dump file for perf"},
523     {NULL, NULL, false, NULL, NULL, NULL}
524 };
525 
526 static void usage(int exitcode)
527 {
528     const struct qemu_argument *arginfo;
529     int maxarglen;
530     int maxenvlen;
531 
532     printf("usage: qemu-" TARGET_NAME " [options] program [arguments...]\n"
533            "Linux CPU emulator (compiled for " TARGET_NAME " emulation)\n"
534            "\n"
535            "Options and associated environment variables:\n"
536            "\n");
537 
538     /* Calculate column widths. We must always have at least enough space
539      * for the column header.
540      */
541     maxarglen = strlen("Argument");
542     maxenvlen = strlen("Env-variable");
543 
544     for (arginfo = arg_table; arginfo->handle_opt != NULL; arginfo++) {
545         int arglen = strlen(arginfo->argv);
546         if (arginfo->has_arg) {
547             arglen += strlen(arginfo->example) + 1;
548         }
549         if (strlen(arginfo->env) > maxenvlen) {
550             maxenvlen = strlen(arginfo->env);
551         }
552         if (arglen > maxarglen) {
553             maxarglen = arglen;
554         }
555     }
556 
557     printf("%-*s %-*s Description\n", maxarglen+1, "Argument",
558             maxenvlen, "Env-variable");
559 
560     for (arginfo = arg_table; arginfo->handle_opt != NULL; arginfo++) {
561         if (arginfo->has_arg) {
562             printf("-%s %-*s %-*s %s\n", arginfo->argv,
563                    (int)(maxarglen - strlen(arginfo->argv) - 1),
564                    arginfo->example, maxenvlen, arginfo->env, arginfo->help);
565         } else {
566             printf("-%-*s %-*s %s\n", maxarglen, arginfo->argv,
567                     maxenvlen, arginfo->env,
568                     arginfo->help);
569         }
570     }
571 
572     printf("\n"
573            "Defaults:\n"
574            "QEMU_LD_PREFIX  = %s\n"
575            "QEMU_STACK_SIZE = %ld byte\n",
576            interp_prefix,
577            guest_stack_size);
578 
579     printf("\n"
580            "You can use -E and -U options or the QEMU_SET_ENV and\n"
581            "QEMU_UNSET_ENV environment variables to set and unset\n"
582            "environment variables for the target process.\n"
583            "It is possible to provide several variables by separating them\n"
584            "by commas in getsubopt(3) style. Additionally it is possible to\n"
585            "provide the -E and -U options multiple times.\n"
586            "The following lines are equivalent:\n"
587            "    -E var1=val2 -E var2=val2 -U LD_PRELOAD -U LD_DEBUG\n"
588            "    -E var1=val2,var2=val2 -U LD_PRELOAD,LD_DEBUG\n"
589            "    QEMU_SET_ENV=var1=val2,var2=val2 QEMU_UNSET_ENV=LD_PRELOAD,LD_DEBUG\n"
590            "Note that if you provide several changes to a single variable\n"
591            "the last change will stay in effect.\n"
592            "\n"
593            QEMU_HELP_BOTTOM "\n");
594 
595     exit(exitcode);
596 }
597 
598 static int parse_args(int argc, char **argv)
599 {
600     const char *r;
601     int optind;
602     const struct qemu_argument *arginfo;
603 
604     for (arginfo = arg_table; arginfo->handle_opt != NULL; arginfo++) {
605         if (arginfo->env == NULL) {
606             continue;
607         }
608 
609         r = getenv(arginfo->env);
610         if (r != NULL) {
611             arginfo->handle_opt(r);
612         }
613     }
614 
615     optind = 1;
616     for (;;) {
617         if (optind >= argc) {
618             break;
619         }
620         r = argv[optind];
621         if (r[0] != '-') {
622             break;
623         }
624         optind++;
625         r++;
626         if (!strcmp(r, "-")) {
627             break;
628         }
629         /* Treat --foo the same as -foo.  */
630         if (r[0] == '-') {
631             r++;
632         }
633 
634         for (arginfo = arg_table; arginfo->handle_opt != NULL; arginfo++) {
635             if (!strcmp(r, arginfo->argv)) {
636                 if (arginfo->has_arg) {
637                     if (optind >= argc) {
638                         (void) fprintf(stderr,
639                             "qemu: missing argument for option '%s'\n", r);
640                         exit(EXIT_FAILURE);
641                     }
642                     arginfo->handle_opt(argv[optind]);
643                     optind++;
644                 } else {
645                     arginfo->handle_opt(NULL);
646                 }
647                 break;
648             }
649         }
650 
651         /* no option matched the current argv */
652         if (arginfo->handle_opt == NULL) {
653             (void) fprintf(stderr, "qemu: unknown option '%s'\n", r);
654             exit(EXIT_FAILURE);
655         }
656     }
657 
658     if (optind >= argc) {
659         (void) fprintf(stderr, "qemu: no user program specified\n");
660         exit(EXIT_FAILURE);
661     }
662 
663     exec_path = argv[optind];
664 
665     return optind;
666 }
667 
668 int main(int argc, char **argv, char **envp)
669 {
670     struct target_pt_regs regs1, *regs = &regs1;
671     struct image_info info1, *info = &info1;
672     struct linux_binprm bprm;
673     TaskState *ts;
674     CPUArchState *env;
675     CPUState *cpu;
676     int optind;
677     char **target_environ, **wrk;
678     char **target_argv;
679     int target_argc;
680     int i;
681     int ret;
682     int execfd;
683     unsigned long max_reserved_va;
684     bool preserve_argv0;
685 
686     error_init(argv[0]);
687     module_call_init(MODULE_INIT_TRACE);
688     qemu_init_cpu_list();
689     module_call_init(MODULE_INIT_QOM);
690 
691     envlist = envlist_create();
692 
693     /*
694      * add current environment into the list
695      * envlist_setenv adds to the front of the list; to preserve environ
696      * order add from back to front
697      */
698     for (wrk = environ; *wrk != NULL; wrk++) {
699         continue;
700     }
701     while (wrk != environ) {
702         wrk--;
703         (void) envlist_setenv(envlist, *wrk);
704     }
705 
706     /* Read the stack limit from the kernel.  If it's "unlimited",
707        then we can do little else besides use the default.  */
708     {
709         struct rlimit lim;
710         if (getrlimit(RLIMIT_STACK, &lim) == 0
711             && lim.rlim_cur != RLIM_INFINITY
712             && lim.rlim_cur == (target_long)lim.rlim_cur
713             && lim.rlim_cur > guest_stack_size) {
714             guest_stack_size = lim.rlim_cur;
715         }
716     }
717 
718     cpu_model = NULL;
719 
720     qemu_add_opts(&qemu_trace_opts);
721     qemu_plugin_add_opts();
722 
723     optind = parse_args(argc, argv);
724 
725     qemu_set_log_filename_flags(last_log_filename,
726                                 last_log_mask | (enable_strace * LOG_STRACE),
727                                 &error_fatal);
728 
729     if (!trace_init_backends()) {
730         exit(1);
731     }
732     trace_init_file();
733     qemu_plugin_load_list(&plugins, &error_fatal);
734 
735     /* Zero out regs */
736     memset(regs, 0, sizeof(struct target_pt_regs));
737 
738     /* Zero out image_info */
739     memset(info, 0, sizeof(struct image_info));
740 
741     memset(&bprm, 0, sizeof (bprm));
742 
743     /* Scan interp_prefix dir for replacement files. */
744     init_paths(interp_prefix);
745 
746     init_qemu_uname_release();
747 
748     /*
749      * Manage binfmt-misc open-binary flag
750      */
751     execfd = qemu_getauxval(AT_EXECFD);
752     if (execfd == 0) {
753         execfd = open(exec_path, O_RDONLY);
754         if (execfd < 0) {
755             printf("Error while loading %s: %s\n", exec_path, strerror(errno));
756             _exit(EXIT_FAILURE);
757         }
758     }
759 
760     /* Resolve executable file name to full path name */
761     if (realpath(exec_path, real_exec_path)) {
762         exec_path = real_exec_path;
763     }
764 
765     /*
766      * get binfmt_misc flags
767      */
768     preserve_argv0 = !!(qemu_getauxval(AT_FLAGS) & AT_FLAGS_PRESERVE_ARGV0);
769 
770     /*
771      * Manage binfmt-misc preserve-arg[0] flag
772      *    argv[optind]     full path to the binary
773      *    argv[optind + 1] original argv[0]
774      */
775     if (optind + 1 < argc && preserve_argv0) {
776         optind++;
777     }
778 
779     if (cpu_model == NULL) {
780         cpu_model = cpu_get_model(get_elf_eflags(execfd));
781     }
782     cpu_type = parse_cpu_option(cpu_model);
783 
784     /* init tcg before creating CPUs and to get qemu_host_page_size */
785     {
786         AccelState *accel = current_accel();
787         AccelClass *ac = ACCEL_GET_CLASS(accel);
788 
789         accel_init_interfaces(ac);
790         object_property_set_bool(OBJECT(accel), "one-insn-per-tb",
791                                  opt_one_insn_per_tb, &error_abort);
792         ac->init_machine(NULL);
793     }
794     cpu = cpu_create(cpu_type);
795     env = cpu_env(cpu);
796     cpu_reset(cpu);
797     thread_cpu = cpu;
798 
799     /*
800      * Reserving too much vm space via mmap can run into problems
801      * with rlimits, oom due to page table creation, etc.  We will
802      * still try it, if directed by the command-line option, but
803      * not by default.
804      */
805     max_reserved_va = MAX_RESERVED_VA(cpu);
806     if (reserved_va != 0) {
807         if ((reserved_va + 1) % qemu_host_page_size) {
808             char *s = size_to_str(qemu_host_page_size);
809             fprintf(stderr, "Reserved virtual address not aligned mod %s\n", s);
810             g_free(s);
811             exit(EXIT_FAILURE);
812         }
813         if (max_reserved_va && reserved_va > max_reserved_va) {
814             fprintf(stderr, "Reserved virtual address too big\n");
815             exit(EXIT_FAILURE);
816         }
817     } else if (HOST_LONG_BITS == 64 && TARGET_VIRT_ADDR_SPACE_BITS <= 32) {
818         /* MAX_RESERVED_VA + 1 is a large power of 2, so is aligned. */
819         reserved_va = max_reserved_va;
820     }
821 
822     /*
823      * Temporarily disable
824      *   "comparison is always false due to limited range of data type"
825      * due to comparison between (possible) uint64_t and uintptr_t.
826      */
827 #pragma GCC diagnostic push
828 #pragma GCC diagnostic ignored "-Wtype-limits"
829 
830     /*
831      * Select an initial value for task_unmapped_base that is in range.
832      */
833     if (reserved_va) {
834         if (TASK_UNMAPPED_BASE < reserved_va) {
835             task_unmapped_base = TASK_UNMAPPED_BASE;
836         } else {
837             /* The most common default formula is TASK_SIZE / 3. */
838             task_unmapped_base = TARGET_PAGE_ALIGN(reserved_va / 3);
839         }
840     } else if (TASK_UNMAPPED_BASE < UINTPTR_MAX) {
841         task_unmapped_base = TASK_UNMAPPED_BASE;
842     } else {
843         /* 32-bit host: pick something medium size. */
844         task_unmapped_base = 0x10000000;
845     }
846     mmap_next_start = task_unmapped_base;
847 
848     /* Similarly for elf_et_dyn_base. */
849     if (reserved_va) {
850         if (ELF_ET_DYN_BASE < reserved_va) {
851             elf_et_dyn_base = ELF_ET_DYN_BASE;
852         } else {
853             /* The most common default formula is TASK_SIZE / 3 * 2. */
854             elf_et_dyn_base = TARGET_PAGE_ALIGN(reserved_va / 3) * 2;
855         }
856     } else if (ELF_ET_DYN_BASE < UINTPTR_MAX) {
857         elf_et_dyn_base = ELF_ET_DYN_BASE;
858     } else {
859         /* 32-bit host: pick something medium size. */
860         elf_et_dyn_base = 0x18000000;
861     }
862 
863 #pragma GCC diagnostic pop
864 
865     {
866         Error *err = NULL;
867         if (seed_optarg != NULL) {
868             qemu_guest_random_seed_main(seed_optarg, &err);
869         } else {
870             qcrypto_init(&err);
871         }
872         if (err) {
873             error_reportf_err(err, "cannot initialize crypto: ");
874             exit(1);
875         }
876     }
877 
878     target_environ = envlist_to_environ(envlist, NULL);
879     envlist_free(envlist);
880 
881     /*
882      * Read in mmap_min_addr kernel parameter.  This value is used
883      * When loading the ELF image to determine whether guest_base
884      * is needed.  It is also used in mmap_find_vma.
885      */
886     {
887         FILE *fp;
888 
889         if ((fp = fopen("/proc/sys/vm/mmap_min_addr", "r")) != NULL) {
890             unsigned long tmp;
891             if (fscanf(fp, "%lu", &tmp) == 1 && tmp != 0) {
892                 mmap_min_addr = tmp;
893                 qemu_log_mask(CPU_LOG_PAGE, "host mmap_min_addr=0x%lx\n",
894                               mmap_min_addr);
895             }
896             fclose(fp);
897         }
898     }
899 
900     /*
901      * We prefer to not make NULL pointers accessible to QEMU.
902      * If we're in a chroot with no /proc, fall back to 1 page.
903      */
904     if (mmap_min_addr == 0) {
905         mmap_min_addr = qemu_host_page_size;
906         qemu_log_mask(CPU_LOG_PAGE,
907                       "host mmap_min_addr=0x%lx (fallback)\n",
908                       mmap_min_addr);
909     }
910 
911     /*
912      * Prepare copy of argv vector for target.
913      */
914     target_argc = argc - optind;
915     target_argv = calloc(target_argc + 1, sizeof (char *));
916     if (target_argv == NULL) {
917         (void) fprintf(stderr, "Unable to allocate memory for target_argv\n");
918         exit(EXIT_FAILURE);
919     }
920 
921     /*
922      * If argv0 is specified (using '-0' switch) we replace
923      * argv[0] pointer with the given one.
924      */
925     i = 0;
926     if (argv0 != NULL) {
927         target_argv[i++] = strdup(argv0);
928     }
929     for (; i < target_argc; i++) {
930         target_argv[i] = strdup(argv[optind + i]);
931     }
932     target_argv[target_argc] = NULL;
933 
934     ts = g_new0(TaskState, 1);
935     init_task_state(ts);
936     /* build Task State */
937     ts->info = info;
938     ts->bprm = &bprm;
939     cpu->opaque = ts;
940     task_settid(ts);
941 
942     fd_trans_init();
943 
944     ret = loader_exec(execfd, exec_path, target_argv, target_environ, regs,
945         info, &bprm);
946     if (ret != 0) {
947         printf("Error while loading %s: %s\n", exec_path, strerror(-ret));
948         _exit(EXIT_FAILURE);
949     }
950 
951     for (wrk = target_environ; *wrk; wrk++) {
952         g_free(*wrk);
953     }
954 
955     g_free(target_environ);
956 
957     if (qemu_loglevel_mask(CPU_LOG_PAGE)) {
958         FILE *f = qemu_log_trylock();
959         if (f) {
960             fprintf(f, "guest_base  %p\n", (void *)guest_base);
961             fprintf(f, "page layout changed following binary load\n");
962             page_dump(f);
963 
964             fprintf(f, "end_code    0x" TARGET_ABI_FMT_lx "\n",
965                     info->end_code);
966             fprintf(f, "start_code  0x" TARGET_ABI_FMT_lx "\n",
967                     info->start_code);
968             fprintf(f, "start_data  0x" TARGET_ABI_FMT_lx "\n",
969                     info->start_data);
970             fprintf(f, "end_data    0x" TARGET_ABI_FMT_lx "\n",
971                     info->end_data);
972             fprintf(f, "start_stack 0x" TARGET_ABI_FMT_lx "\n",
973                     info->start_stack);
974             fprintf(f, "brk         0x" TARGET_ABI_FMT_lx "\n",
975                     info->brk);
976             fprintf(f, "entry       0x" TARGET_ABI_FMT_lx "\n",
977                     info->entry);
978             fprintf(f, "argv_start  0x" TARGET_ABI_FMT_lx "\n",
979                     info->argv);
980             fprintf(f, "env_start   0x" TARGET_ABI_FMT_lx "\n",
981                     info->envp);
982             fprintf(f, "auxv_start  0x" TARGET_ABI_FMT_lx "\n",
983                     info->saved_auxv);
984             qemu_log_unlock(f);
985         }
986     }
987 
988     target_set_brk(info->brk);
989     syscall_init();
990     signal_init();
991 
992     /* Now that we've loaded the binary, GUEST_BASE is fixed.  Delay
993        generating the prologue until now so that the prologue can take
994        the real value of GUEST_BASE into account.  */
995     tcg_prologue_init();
996 
997     target_cpu_copy_regs(env, regs);
998 
999     if (gdbstub) {
1000         if (gdbserver_start(gdbstub) < 0) {
1001             fprintf(stderr, "qemu: could not open gdbserver on %s\n",
1002                     gdbstub);
1003             exit(EXIT_FAILURE);
1004         }
1005         gdb_handlesig(cpu, 0);
1006     }
1007 
1008 #ifdef CONFIG_SEMIHOSTING
1009     qemu_semihosting_guestfd_init();
1010 #endif
1011 
1012     cpu_loop(env);
1013     /* never exits */
1014     return 0;
1015 }
1016