xref: /openbmc/linux/arch/um/os-Linux/main.c (revision d3878bb8)
152c653b3SJeff Dike /*
22eb5f31bSAnton Ivanov  * Copyright (C) 2015 Thomas Meyer (thomas@m3y3r.de)
3ba180fd4SJeff Dike  * Copyright (C) 2000 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
452c653b3SJeff Dike  * Licensed under the GPL
552c653b3SJeff Dike  */
652c653b3SJeff Dike 
752c653b3SJeff Dike #include <stdio.h>
852c653b3SJeff Dike #include <stdlib.h>
9ba180fd4SJeff Dike #include <unistd.h>
1052c653b3SJeff Dike #include <errno.h>
11ba180fd4SJeff Dike #include <signal.h>
12ba180fd4SJeff Dike #include <string.h>
1352c653b3SJeff Dike #include <sys/resource.h>
1437185b33SAl Viro #include <as-layout.h>
1537185b33SAl Viro #include <init.h>
1637185b33SAl Viro #include <kern_util.h>
1737185b33SAl Viro #include <os.h>
1837185b33SAl Viro #include <um_malloc.h>
1952c653b3SJeff Dike 
2052c653b3SJeff Dike #define PGD_BOUND (4 * 1024 * 1024)
2152c653b3SJeff Dike #define STACKSIZE (8 * 1024 * 1024)
2252c653b3SJeff Dike #define THREAD_NAME_LEN (256)
2352c653b3SJeff Dike 
24b743ac54SRichard Weinberger long elf_aux_hwcap;
25b743ac54SRichard Weinberger 
2652c653b3SJeff Dike static void set_stklim(void)
2752c653b3SJeff Dike {
2852c653b3SJeff Dike 	struct rlimit lim;
2952c653b3SJeff Dike 
3052c653b3SJeff Dike 	if (getrlimit(RLIMIT_STACK, &lim) < 0) {
3152c653b3SJeff Dike 		perror("getrlimit");
3252c653b3SJeff Dike 		exit(1);
3352c653b3SJeff Dike 	}
3452c653b3SJeff Dike 	if ((lim.rlim_cur == RLIM_INFINITY) || (lim.rlim_cur > STACKSIZE)) {
3552c653b3SJeff Dike 		lim.rlim_cur = STACKSIZE;
3652c653b3SJeff Dike 		if (setrlimit(RLIMIT_STACK, &lim) < 0) {
3752c653b3SJeff Dike 			perror("setrlimit");
3852c653b3SJeff Dike 			exit(1);
3952c653b3SJeff Dike 		}
4052c653b3SJeff Dike 	}
4152c653b3SJeff Dike }
4252c653b3SJeff Dike 
4352c653b3SJeff Dike static __init void do_uml_initcalls(void)
4452c653b3SJeff Dike {
4552c653b3SJeff Dike 	initcall_t *call;
4652c653b3SJeff Dike 
4752c653b3SJeff Dike 	call = &__uml_initcall_start;
48f2183125SJeff Dike 	while (call < &__uml_initcall_end) {
4952c653b3SJeff Dike 		(*call)();
5052c653b3SJeff Dike 		call++;
5152c653b3SJeff Dike 	}
5252c653b3SJeff Dike }
5352c653b3SJeff Dike 
5452c653b3SJeff Dike static void last_ditch_exit(int sig)
5552c653b3SJeff Dike {
5652c653b3SJeff Dike 	uml_cleanup();
5752c653b3SJeff Dike 	exit(1);
5852c653b3SJeff Dike }
5952c653b3SJeff Dike 
604b84c69bSJeff Dike static void install_fatal_handler(int sig)
614b84c69bSJeff Dike {
624b84c69bSJeff Dike 	struct sigaction action;
634b84c69bSJeff Dike 
644b84c69bSJeff Dike 	/* All signals are enabled in this handler ... */
654b84c69bSJeff Dike 	sigemptyset(&action.sa_mask);
664b84c69bSJeff Dike 
67ba180fd4SJeff Dike 	/*
68ba180fd4SJeff Dike 	 * ... including the signal being handled, plus we want the
694b84c69bSJeff Dike 	 * handler reset to the default behavior, so that if an exit
704b84c69bSJeff Dike 	 * handler is hanging for some reason, the UML will just die
714b84c69bSJeff Dike 	 * after this signal is sent a second time.
724b84c69bSJeff Dike 	 */
734b84c69bSJeff Dike 	action.sa_flags = SA_RESETHAND | SA_NODEFER;
744b84c69bSJeff Dike 	action.sa_restorer = NULL;
754b84c69bSJeff Dike 	action.sa_handler = last_ditch_exit;
764b84c69bSJeff Dike 	if (sigaction(sig, &action, NULL) < 0) {
774b84c69bSJeff Dike 		printf("failed to install handler for signal %d - errno = %d\n",
78c9a3072dSWANG Cong 		       sig, errno);
794b84c69bSJeff Dike 		exit(1);
804b84c69bSJeff Dike 	}
814b84c69bSJeff Dike }
824b84c69bSJeff Dike 
830ce451acSRichard Weinberger #define UML_LIB_PATH	":" OS_LIB_PATH "/uml"
84cb98cdcdSMattia Dongili 
85cb98cdcdSMattia Dongili static void setup_env_path(void)
86cb98cdcdSMattia Dongili {
87cb98cdcdSMattia Dongili 	char *new_path = NULL;
88cb98cdcdSMattia Dongili 	char *old_path = NULL;
89cb98cdcdSMattia Dongili 	int path_len = 0;
90cb98cdcdSMattia Dongili 
91cb98cdcdSMattia Dongili 	old_path = getenv("PATH");
92ba180fd4SJeff Dike 	/*
93ba180fd4SJeff Dike 	 * if no PATH variable is set or it has an empty value
94cb98cdcdSMattia Dongili 	 * just use the default + /usr/lib/uml
95cb98cdcdSMattia Dongili 	 */
96cb98cdcdSMattia Dongili 	if (!old_path || (path_len = strlen(old_path)) == 0) {
97c9a3072dSWANG Cong 		if (putenv("PATH=:/bin:/usr/bin/" UML_LIB_PATH))
98c9a3072dSWANG Cong 			perror("couldn't putenv");
99cb98cdcdSMattia Dongili 		return;
100cb98cdcdSMattia Dongili 	}
101cb98cdcdSMattia Dongili 
102cb98cdcdSMattia Dongili 	/* append /usr/lib/uml to the existing path */
103cb98cdcdSMattia Dongili 	path_len += strlen("PATH=" UML_LIB_PATH) + 1;
104cb98cdcdSMattia Dongili 	new_path = malloc(path_len);
105cb98cdcdSMattia Dongili 	if (!new_path) {
106c9a3072dSWANG Cong 		perror("couldn't malloc to set a new PATH");
107cb98cdcdSMattia Dongili 		return;
108cb98cdcdSMattia Dongili 	}
109cb98cdcdSMattia Dongili 	snprintf(new_path, path_len, "PATH=%s" UML_LIB_PATH, old_path);
110c9a3072dSWANG Cong 	if (putenv(new_path)) {
111c9a3072dSWANG Cong 		perror("couldn't putenv to set a new PATH");
112c9a3072dSWANG Cong 		free(new_path);
113c9a3072dSWANG Cong 	}
114cb98cdcdSMattia Dongili }
115cb98cdcdSMattia Dongili 
11652c653b3SJeff Dike extern void scan_elf_aux( char **envp);
11752c653b3SJeff Dike 
11836e45463SJeff Dike int __init main(int argc, char **argv, char **envp)
11952c653b3SJeff Dike {
12052c653b3SJeff Dike 	char **new_argv;
12152c653b3SJeff Dike 	int ret, i, err;
12252c653b3SJeff Dike 
12352c653b3SJeff Dike 	set_stklim();
12452c653b3SJeff Dike 
125cb98cdcdSMattia Dongili 	setup_env_path();
126cb98cdcdSMattia Dongili 
12725012721SRichard Weinberger 	setsid();
12825012721SRichard Weinberger 
12952c653b3SJeff Dike 	new_argv = malloc((argc + 1) * sizeof(char *));
13052c653b3SJeff Dike 	if (new_argv == NULL) {
13152c653b3SJeff Dike 		perror("Mallocing argv");
13252c653b3SJeff Dike 		exit(1);
13352c653b3SJeff Dike 	}
13452c653b3SJeff Dike 	for (i = 0; i < argc; i++) {
13552c653b3SJeff Dike 		new_argv[i] = strdup(argv[i]);
13652c653b3SJeff Dike 		if (new_argv[i] == NULL) {
13752c653b3SJeff Dike 			perror("Mallocing an arg");
13852c653b3SJeff Dike 			exit(1);
13952c653b3SJeff Dike 		}
14052c653b3SJeff Dike 	}
14152c653b3SJeff Dike 	new_argv[argc] = NULL;
14252c653b3SJeff Dike 
143ba180fd4SJeff Dike 	/*
144ba180fd4SJeff Dike 	 * Allow these signals to bring down a UML if all other
1454b84c69bSJeff Dike 	 * methods of control fail.
1464b84c69bSJeff Dike 	 */
1474b84c69bSJeff Dike 	install_fatal_handler(SIGINT);
1484b84c69bSJeff Dike 	install_fatal_handler(SIGTERM);
14952c653b3SJeff Dike 
150b743ac54SRichard Weinberger #ifdef CONFIG_ARCH_REUSE_HOST_VSYSCALL_AREA
15152c653b3SJeff Dike 	scan_elf_aux(envp);
152b743ac54SRichard Weinberger #endif
15352c653b3SJeff Dike 
15452c653b3SJeff Dike 	do_uml_initcalls();
1559fcb663bSAnton Ivanov 	change_sig(SIGPIPE, 0);
15652c653b3SJeff Dike 	ret = linux_main(argc, argv);
15752c653b3SJeff Dike 
158ba180fd4SJeff Dike 	/*
159ba180fd4SJeff Dike 	 * Disable SIGPROF - I have no idea why libc doesn't do this or turn
16052c653b3SJeff Dike 	 * off the profiling time, but UML dies with a SIGPROF just before
16152c653b3SJeff Dike 	 * exiting when profiling is active.
16252c653b3SJeff Dike 	 */
16352c653b3SJeff Dike 	change_sig(SIGPROF, 0);
16452c653b3SJeff Dike 
165ba180fd4SJeff Dike 	/*
166ba180fd4SJeff Dike 	 * This signal stuff used to be in the reboot case.  However,
1672eb5f31bSAnton Ivanov 	 * sometimes a timer signal can come in when we're halting (reproducably
16852c653b3SJeff Dike 	 * when writing out gcov information, presumably because that takes
16952c653b3SJeff Dike 	 * some time) and cause a segfault.
17052c653b3SJeff Dike 	 */
17152c653b3SJeff Dike 
1722eb5f31bSAnton Ivanov 	/* stop timers and set timer signal to be ignored */
1732eb5f31bSAnton Ivanov 	os_timer_disable();
17452c653b3SJeff Dike 
17552c653b3SJeff Dike 	/* disable SIGIO for the fds and set SIGIO to be ignored */
17652c653b3SJeff Dike 	err = deactivate_all_fds();
17752c653b3SJeff Dike 	if (err)
17852c653b3SJeff Dike 		printf("deactivate_all_fds failed, errno = %d\n", -err);
17952c653b3SJeff Dike 
180ba180fd4SJeff Dike 	/*
181ba180fd4SJeff Dike 	 * Let any pending signals fire now.  This ensures
18252c653b3SJeff Dike 	 * that they won't be delivered after the exec, when
18352c653b3SJeff Dike 	 * they are definitely not expected.
18452c653b3SJeff Dike 	 */
18552c653b3SJeff Dike 	unblock_signals();
18652c653b3SJeff Dike 
187d3878bb8SMasami Hiramatsu 	os_info("\n");
18852c653b3SJeff Dike 	/* Reboot */
18952c653b3SJeff Dike 	if (ret) {
19052c653b3SJeff Dike 		execvp(new_argv[0], new_argv);
19152c653b3SJeff Dike 		perror("Failed to exec kernel");
19252c653b3SJeff Dike 		ret = 1;
19352c653b3SJeff Dike 	}
194a5ed1ffaSJeff Dike 	return uml_exitcode;
19552c653b3SJeff Dike }
19652c653b3SJeff Dike 
19752c653b3SJeff Dike extern void *__real_malloc(int);
19852c653b3SJeff Dike 
19952c653b3SJeff Dike void *__wrap_malloc(int size)
20052c653b3SJeff Dike {
20152c653b3SJeff Dike 	void *ret;
20252c653b3SJeff Dike 
2036aa802ceSJeff Dike 	if (!kmalloc_ok)
204a5ed1ffaSJeff Dike 		return __real_malloc(size);
205c539ab73SJeff Dike 	else if (size <= UM_KERN_PAGE_SIZE)
206c539ab73SJeff Dike 		/* finding contiguous pages can be hard*/
20743f5b308SJeff Dike 		ret = uml_kmalloc(size, UM_GFP_KERNEL);
208e4c4bf99SJeff Dike 	else ret = vmalloc(size);
20952c653b3SJeff Dike 
210ba180fd4SJeff Dike 	/*
211ba180fd4SJeff Dike 	 * glibc people insist that if malloc fails, errno should be
21252c653b3SJeff Dike 	 * set by malloc as well. So we do.
21352c653b3SJeff Dike 	 */
21452c653b3SJeff Dike 	if (ret == NULL)
21552c653b3SJeff Dike 		errno = ENOMEM;
21652c653b3SJeff Dike 
217a5ed1ffaSJeff Dike 	return ret;
21852c653b3SJeff Dike }
21952c653b3SJeff Dike 
22052c653b3SJeff Dike void *__wrap_calloc(int n, int size)
22152c653b3SJeff Dike {
22252c653b3SJeff Dike 	void *ptr = __wrap_malloc(n * size);
22352c653b3SJeff Dike 
224a5ed1ffaSJeff Dike 	if (ptr == NULL)
225a5ed1ffaSJeff Dike 		return NULL;
22652c653b3SJeff Dike 	memset(ptr, 0, n * size);
227a5ed1ffaSJeff Dike 	return ptr;
22852c653b3SJeff Dike }
22952c653b3SJeff Dike 
23052c653b3SJeff Dike extern void __real_free(void *);
23152c653b3SJeff Dike 
23252c653b3SJeff Dike extern unsigned long high_physmem;
23352c653b3SJeff Dike 
23452c653b3SJeff Dike void __wrap_free(void *ptr)
23552c653b3SJeff Dike {
23652c653b3SJeff Dike 	unsigned long addr = (unsigned long) ptr;
23752c653b3SJeff Dike 
238ba180fd4SJeff Dike 	/*
239ba180fd4SJeff Dike 	 * We need to know how the allocation happened, so it can be correctly
24052c653b3SJeff Dike 	 * freed.  This is done by seeing what region of memory the pointer is
24152c653b3SJeff Dike 	 * in -
24252c653b3SJeff Dike 	 * 	physical memory - kmalloc/kfree
24352c653b3SJeff Dike 	 *	kernel virtual memory - vmalloc/vfree
24452c653b3SJeff Dike 	 * 	anywhere else - malloc/free
24552c653b3SJeff Dike 	 * If kmalloc is not yet possible, then either high_physmem and/or
24652c653b3SJeff Dike 	 * end_vm are still 0 (as at startup), in which case we call free, or
24752c653b3SJeff Dike 	 * we have set them, but anyway addr has not been allocated from those
24852c653b3SJeff Dike 	 * areas. So, in both cases __real_free is called.
24952c653b3SJeff Dike 	 *
25052c653b3SJeff Dike 	 * CAN_KMALLOC is checked because it would be bad to free a buffer
25152c653b3SJeff Dike 	 * with kmalloc/vmalloc after they have been turned off during
25252c653b3SJeff Dike 	 * shutdown.
25352c653b3SJeff Dike 	 * XXX: However, we sometimes shutdown CAN_KMALLOC temporarily, so
25452c653b3SJeff Dike 	 * there is a possibility for memory leaks.
25552c653b3SJeff Dike 	 */
25652c653b3SJeff Dike 
25752c653b3SJeff Dike 	if ((addr >= uml_physmem) && (addr < high_physmem)) {
2586aa802ceSJeff Dike 		if (kmalloc_ok)
25952c653b3SJeff Dike 			kfree(ptr);
26052c653b3SJeff Dike 	}
26152c653b3SJeff Dike 	else if ((addr >= start_vm) && (addr < end_vm)) {
2626aa802ceSJeff Dike 		if (kmalloc_ok)
26352c653b3SJeff Dike 			vfree(ptr);
26452c653b3SJeff Dike 	}
26552c653b3SJeff Dike 	else __real_free(ptr);
26652c653b3SJeff Dike }
267