xref: /kvmtool/builtin-run.c (revision 1dc24dcc723991c63fb7747ff829b1e1eb38975f)
1 #include "kvm/builtin-run.h"
2 
3 #include "kvm/builtin-setup.h"
4 #include "kvm/virtio-balloon.h"
5 #include "kvm/virtio-console.h"
6 #include "kvm/parse-options.h"
7 #include "kvm/8250-serial.h"
8 #include "kvm/framebuffer.h"
9 #include "kvm/disk-image.h"
10 #include "kvm/threadpool.h"
11 #include "kvm/virtio-scsi.h"
12 #include "kvm/virtio-blk.h"
13 #include "kvm/virtio-net.h"
14 #include "kvm/virtio-rng.h"
15 #include "kvm/ioeventfd.h"
16 #include "kvm/virtio-9p.h"
17 #include "kvm/barrier.h"
18 #include "kvm/kvm-cpu.h"
19 #include "kvm/ioport.h"
20 #include "kvm/symbol.h"
21 #include "kvm/i8042.h"
22 #include "kvm/mutex.h"
23 #include "kvm/term.h"
24 #include "kvm/util.h"
25 #include "kvm/strbuf.h"
26 #include "kvm/vesa.h"
27 #include "kvm/irq.h"
28 #include "kvm/kvm.h"
29 #include "kvm/pci.h"
30 #include "kvm/rtc.h"
31 #include "kvm/sdl.h"
32 #include "kvm/vnc.h"
33 #include "kvm/guest_compat.h"
34 #include "kvm/pci-shmem.h"
35 #include "kvm/kvm-ipc.h"
36 #include "kvm/builtin-debug.h"
37 
38 #include <linux/types.h>
39 #include <linux/err.h>
40 
41 #include <sys/utsname.h>
42 #include <sys/types.h>
43 #include <sys/stat.h>
44 #include <termios.h>
45 #include <signal.h>
46 #include <stdlib.h>
47 #include <string.h>
48 #include <unistd.h>
49 #include <ctype.h>
50 #include <stdio.h>
51 
52 #define MB_SHIFT		(20)
53 #define KB_SHIFT		(10)
54 #define GB_SHIFT		(30)
55 
56 struct kvm *kvm;
57 struct kvm_cpu **kvm_cpus;
58 __thread struct kvm_cpu *current_kvm_cpu;
59 
60 static int  kvm_run_wrapper;
61 extern int  debug_iodelay;
62 
63 bool do_debug_print = false;
64 
65 static int nrcpus;
66 static int vidmode = -1;
67 
68 extern char _binary_guest_init_start;
69 extern char _binary_guest_init_size;
70 
71 static const char * const run_usage[] = {
72 	"lkvm run [<options>] [<kernel image>]",
73 	NULL
74 };
75 
76 enum {
77 	KVM_RUN_DEFAULT,
78 	KVM_RUN_SANDBOX,
79 };
80 
81 void kvm_run_set_wrapper_sandbox(void)
82 {
83 	kvm_run_wrapper = KVM_RUN_SANDBOX;
84 }
85 
86 static int img_name_parser(const struct option *opt, const char *arg, int unset)
87 {
88 	char path[PATH_MAX];
89 	const char *cur;
90 	struct stat st;
91 	char *sep;
92 
93 	if (stat(arg, &st) == 0 &&
94 	    S_ISDIR(st.st_mode)) {
95 		char tmp[PATH_MAX];
96 
97 		if (kvm->cfg.using_rootfs)
98 			die("Please use only one rootfs directory atmost");
99 
100 		if (realpath(arg, tmp) == 0 ||
101 		    virtio_9p__register(kvm, tmp, "/dev/root") < 0)
102 			die("Unable to initialize virtio 9p");
103 		kvm->cfg.using_rootfs = 1;
104 		return 0;
105 	}
106 
107 	snprintf(path, PATH_MAX, "%s%s", kvm__get_dir(), arg);
108 
109 	if (stat(path, &st) == 0 &&
110 	    S_ISDIR(st.st_mode)) {
111 		char tmp[PATH_MAX];
112 
113 		if (kvm->cfg.using_rootfs)
114 			die("Please use only one rootfs directory atmost");
115 
116 		if (realpath(path, tmp) == 0 ||
117 		    virtio_9p__register(kvm, tmp, "/dev/root") < 0)
118 			die("Unable to initialize virtio 9p");
119 		if (virtio_9p__register(kvm, "/", "hostfs") < 0)
120 			die("Unable to initialize virtio 9p");
121 		kvm_setup_resolv(arg);
122 		kvm->cfg.using_rootfs = kvm->cfg.custom_rootfs = 1;
123 		kvm->cfg.custom_rootfs_name = arg;
124 		return 0;
125 	}
126 
127 	if (kvm->cfg.image_count >= MAX_DISK_IMAGES)
128 		die("Currently only 4 images are supported");
129 
130 	kvm->cfg.disk_image[kvm->cfg.image_count].filename = arg;
131 	cur = arg;
132 
133 	if (strncmp(arg, "scsi:", 5) == 0) {
134 		sep = strstr(arg, ":");
135 		if (sep)
136 			kvm->cfg.disk_image[kvm->cfg.image_count].wwpn = sep + 1;
137 		sep = strstr(sep + 1, ":");
138 		if (sep) {
139 			*sep = 0;
140 			kvm->cfg.disk_image[kvm->cfg.image_count].tpgt = sep + 1;
141 		}
142 		cur = sep + 1;
143 	}
144 
145 	do {
146 		sep = strstr(cur, ",");
147 		if (sep) {
148 			if (strncmp(sep + 1, "ro", 2) == 0)
149 				kvm->cfg.disk_image[kvm->cfg.image_count].readonly = true;
150 			else if (strncmp(sep + 1, "direct", 6) == 0)
151 				kvm->cfg.disk_image[kvm->cfg.image_count].direct = true;
152 			*sep = 0;
153 			cur = sep + 1;
154 		}
155 	} while (sep);
156 
157 	kvm->cfg.image_count++;
158 
159 	return 0;
160 }
161 
162 static int virtio_9p_rootdir_parser(const struct option *opt, const char *arg, int unset)
163 {
164 	char *tag_name;
165 	char tmp[PATH_MAX];
166 
167 	/*
168 	 * 9p dir can be of the form dirname,tag_name or
169 	 * just dirname. In the later case we use the
170 	 * default tag name
171 	 */
172 	tag_name = strstr(arg, ",");
173 	if (tag_name) {
174 		*tag_name = '\0';
175 		tag_name++;
176 	}
177 	if (realpath(arg, tmp)) {
178 		if (virtio_9p__register(kvm, tmp, tag_name) < 0)
179 			die("Unable to initialize virtio 9p");
180 	} else
181 		die("Failed resolving 9p path");
182 	return 0;
183 }
184 
185 static int tty_parser(const struct option *opt, const char *arg, int unset)
186 {
187 	int tty = atoi(arg);
188 
189 	term_set_tty(tty);
190 
191 	return 0;
192 }
193 
194 static inline void str_to_mac(const char *str, char *mac)
195 {
196 	sscanf(str, "%hhx:%hhx:%hhx:%hhx:%hhx:%hhx",
197 		mac, mac+1, mac+2, mac+3, mac+4, mac+5);
198 }
199 static int set_net_param(struct virtio_net_params *p, const char *param,
200 				const char *val)
201 {
202 	if (strcmp(param, "guest_mac") == 0) {
203 		str_to_mac(val, p->guest_mac);
204 	} else if (strcmp(param, "mode") == 0) {
205 		if (!strncmp(val, "user", 4)) {
206 			int i;
207 
208 			for (i = 0; i < kvm->cfg.num_net_devices; i++)
209 				if (kvm->cfg.net_params[i].mode == NET_MODE_USER)
210 					die("Only one usermode network device allowed at a time");
211 			p->mode = NET_MODE_USER;
212 		} else if (!strncmp(val, "tap", 3)) {
213 			p->mode = NET_MODE_TAP;
214 		} else if (!strncmp(val, "none", 4)) {
215 			kvm->cfg.no_net = 1;
216 			return -1;
217 		} else
218 			die("Unknown network mode %s, please use user, tap or none", kvm->cfg.network);
219 	} else if (strcmp(param, "script") == 0) {
220 		p->script = strdup(val);
221 	} else if (strcmp(param, "guest_ip") == 0) {
222 		p->guest_ip = strdup(val);
223 	} else if (strcmp(param, "host_ip") == 0) {
224 		p->host_ip = strdup(val);
225 	} else if (strcmp(param, "trans") == 0) {
226 		p->trans = strdup(val);
227 	} else if (strcmp(param, "vhost") == 0) {
228 		p->vhost = atoi(val);
229 	} else if (strcmp(param, "fd") == 0) {
230 		p->fd = atoi(val);
231 	} else
232 		die("Unknown network parameter %s", param);
233 
234 	return 0;
235 }
236 
237 static int netdev_parser(const struct option *opt, const char *arg, int unset)
238 {
239 	struct virtio_net_params p;
240 	char *buf = NULL, *cmd = NULL, *cur = NULL;
241 	bool on_cmd = true;
242 
243 	if (arg) {
244 		buf = strdup(arg);
245 		if (buf == NULL)
246 			die("Failed allocating new net buffer");
247 		cur = strtok(buf, ",=");
248 	}
249 
250 	p = (struct virtio_net_params) {
251 		.guest_ip	= DEFAULT_GUEST_ADDR,
252 		.host_ip	= DEFAULT_HOST_ADDR,
253 		.script		= DEFAULT_SCRIPT,
254 		.mode		= NET_MODE_TAP,
255 	};
256 
257 	str_to_mac(DEFAULT_GUEST_MAC, p.guest_mac);
258 	p.guest_mac[5] += kvm->cfg.num_net_devices;
259 
260 	while (cur) {
261 		if (on_cmd) {
262 			cmd = cur;
263 		} else {
264 			if (set_net_param(&p, cmd, cur) < 0)
265 				goto done;
266 		}
267 		on_cmd = !on_cmd;
268 
269 		cur = strtok(NULL, ",=");
270 	};
271 
272 	kvm->cfg.num_net_devices++;
273 
274 	kvm->cfg.net_params = realloc(kvm->cfg.net_params, kvm->cfg.num_net_devices * sizeof(*kvm->cfg.net_params));
275 	if (kvm->cfg.net_params == NULL)
276 		die("Failed adding new network device");
277 
278 	kvm->cfg.net_params[kvm->cfg.num_net_devices - 1] = p;
279 
280 done:
281 	free(buf);
282 	return 0;
283 }
284 
285 static int shmem_parser(const struct option *opt, const char *arg, int unset)
286 {
287 	const u64 default_size = SHMEM_DEFAULT_SIZE;
288 	const u64 default_phys_addr = SHMEM_DEFAULT_ADDR;
289 	const char *default_handle = SHMEM_DEFAULT_HANDLE;
290 	struct shmem_info *si = malloc(sizeof(struct shmem_info));
291 	u64 phys_addr;
292 	u64 size;
293 	char *handle = NULL;
294 	int create = 0;
295 	const char *p = arg;
296 	char *next;
297 	int base = 10;
298 	int verbose = 0;
299 
300 	const int skip_pci = strlen("pci:");
301 	if (verbose)
302 		pr_info("shmem_parser(%p,%s,%d)", opt, arg, unset);
303 	/* parse out optional addr family */
304 	if (strcasestr(p, "pci:")) {
305 		p += skip_pci;
306 	} else if (strcasestr(p, "mem:")) {
307 		die("I can't add to E820 map yet.\n");
308 	}
309 	/* parse out physical addr */
310 	base = 10;
311 	if (strcasestr(p, "0x"))
312 		base = 16;
313 	phys_addr = strtoll(p, &next, base);
314 	if (next == p && phys_addr == 0) {
315 		pr_info("shmem: no physical addr specified, using default.");
316 		phys_addr = default_phys_addr;
317 	}
318 	if (*next != ':' && *next != '\0')
319 		die("shmem: unexpected chars after phys addr.\n");
320 	if (*next == '\0')
321 		p = next;
322 	else
323 		p = next + 1;
324 	/* parse out size */
325 	base = 10;
326 	if (strcasestr(p, "0x"))
327 		base = 16;
328 	size = strtoll(p, &next, base);
329 	if (next == p && size == 0) {
330 		pr_info("shmem: no size specified, using default.");
331 		size = default_size;
332 	}
333 	/* look for [KMGkmg][Bb]*  uses base 2. */
334 	int skip_B = 0;
335 	if (strspn(next, "KMGkmg")) {	/* might have a prefix */
336 		if (*(next + 1) == 'B' || *(next + 1) == 'b')
337 			skip_B = 1;
338 		switch (*next) {
339 		case 'K':
340 		case 'k':
341 			size = size << KB_SHIFT;
342 			break;
343 		case 'M':
344 		case 'm':
345 			size = size << MB_SHIFT;
346 			break;
347 		case 'G':
348 		case 'g':
349 			size = size << GB_SHIFT;
350 			break;
351 		default:
352 			die("shmem: bug in detecting size prefix.");
353 			break;
354 		}
355 		next += 1 + skip_B;
356 	}
357 	if (*next != ':' && *next != '\0') {
358 		die("shmem: unexpected chars after phys size. <%c><%c>\n",
359 		    *next, *p);
360 	}
361 	if (*next == '\0')
362 		p = next;
363 	else
364 		p = next + 1;
365 	/* parse out optional shmem handle */
366 	const int skip_handle = strlen("handle=");
367 	next = strcasestr(p, "handle=");
368 	if (*p && next) {
369 		if (p != next)
370 			die("unexpected chars before handle\n");
371 		p += skip_handle;
372 		next = strchrnul(p, ':');
373 		if (next - p) {
374 			handle = malloc(next - p + 1);
375 			strncpy(handle, p, next - p);
376 			handle[next - p] = '\0';	/* just in case. */
377 		}
378 		if (*next == '\0')
379 			p = next;
380 		else
381 			p = next + 1;
382 	}
383 	/* parse optional create flag to see if we should create shm seg. */
384 	if (*p && strcasestr(p, "create")) {
385 		create = 1;
386 		p += strlen("create");
387 	}
388 	if (*p != '\0')
389 		die("shmem: unexpected trailing chars\n");
390 	if (handle == NULL) {
391 		handle = malloc(strlen(default_handle) + 1);
392 		strcpy(handle, default_handle);
393 	}
394 	if (verbose) {
395 		pr_info("shmem: phys_addr = %llx", phys_addr);
396 		pr_info("shmem: size      = %llx", size);
397 		pr_info("shmem: handle    = %s", handle);
398 		pr_info("shmem: create    = %d", create);
399 	}
400 
401 	si->phys_addr = phys_addr;
402 	si->size = size;
403 	si->handle = handle;
404 	si->create = create;
405 	pci_shmem__register_mem(si);	/* ownership of si, etc. passed on. */
406 	return 0;
407 }
408 
409 #define BUILD_OPTIONS(name, cfg)					\
410 	struct option name[] = {					\
411 	OPT_GROUP("Basic options:"),					\
412 	OPT_STRING('\0', "name", &(cfg)->guest_name, "guest name",	\
413 			"A name for the guest"),			\
414 	OPT_INTEGER('c', "cpus", &nrcpus, "Number of CPUs"),		\
415 	OPT_U64('m', "mem", &(cfg)->ram_size, "Virtual machine memory size\
416 		in MiB."),						\
417 	OPT_CALLBACK('\0', "shmem", NULL,				\
418 		     "[pci:]<addr>:<size>[:handle=<handle>][:create]",	\
419 		     "Share host shmem with guest via pci device",	\
420 		     shmem_parser, NULL),				\
421 	OPT_CALLBACK('d', "disk", kvm, "image or rootfs_dir", "Disk 	\
422 			image or rootfs directory", img_name_parser,	\
423 			NULL),						\
424 	OPT_BOOLEAN('\0', "balloon", &(cfg)->balloon, "Enable virtio	\
425 			balloon"),					\
426 	OPT_BOOLEAN('\0', "vnc", &(cfg)->vnc, "Enable VNC framebuffer"),\
427 	OPT_BOOLEAN('\0', "sdl", &(cfg)->sdl, "Enable SDL framebuffer"),\
428 	OPT_BOOLEAN('\0', "rng", &(cfg)->virtio_rng, "Enable virtio Random\
429 			Number Generator"),				\
430 	OPT_CALLBACK('\0', "9p", NULL, "dir_to_share,tag_name",		\
431 		     "Enable virtio 9p to share files between host and	\
432 		     guest", virtio_9p_rootdir_parser, NULL),		\
433 	OPT_STRING('\0', "console", &(cfg)->console, "serial, virtio or	\
434 			hv", "Console to use"),				\
435 	OPT_STRING('\0', "dev", &(cfg)->dev, "device_file",		\
436 			"KVM device file"),				\
437 	OPT_CALLBACK('\0', "tty", NULL, "tty id",			\
438 		     "Remap guest TTY into a pty on the host",		\
439 		     tty_parser, NULL),					\
440 	OPT_STRING('\0', "sandbox", &(cfg)->sandbox, "script",		\
441 			"Run this script when booting into custom	\
442 			rootfs"),					\
443 	OPT_STRING('\0', "hugetlbfs", &(cfg)->hugetlbfs_path, "path",	\
444 			"Hugetlbfs path"),				\
445 									\
446 	OPT_GROUP("Kernel options:"),					\
447 	OPT_STRING('k', "kernel", &(cfg)->kernel_filename, "kernel",	\
448 			"Kernel to boot in virtual machine"),		\
449 	OPT_STRING('i', "initrd", &(cfg)->initrd_filename, "initrd",	\
450 			"Initial RAM disk image"),			\
451 	OPT_STRING('p', "params", &(cfg)->kernel_cmdline, "params",	\
452 			"Kernel command line arguments"),		\
453 	OPT_STRING('f', "firmware", &(cfg)->firmware_filename, "firmware",\
454 			"Firmware image to boot in virtual machine"),	\
455 									\
456 	OPT_GROUP("Networking options:"),				\
457 	OPT_CALLBACK_DEFAULT('n', "network", NULL, "network params",	\
458 		     "Create a new guest NIC",				\
459 		     netdev_parser, NULL, NULL),			\
460 	OPT_BOOLEAN('\0', "no-dhcp", &(cfg)->no_dhcp, "Disable kernel DHCP\
461 			in rootfs mode"),				\
462 									\
463 	OPT_GROUP("BIOS options:"),					\
464 	OPT_INTEGER('\0', "vidmode", &vidmode,				\
465 		    "Video mode"),					\
466 									\
467 	OPT_GROUP("Debug options:"),					\
468 	OPT_BOOLEAN('\0', "debug", &do_debug_print,			\
469 			"Enable debug messages"),			\
470 	OPT_BOOLEAN('\0', "debug-single-step", &(cfg)->single_step,	\
471 			"Enable single stepping"),			\
472 	OPT_BOOLEAN('\0', "debug-ioport", &(cfg)->ioport_debug,		\
473 			"Enable ioport debugging"),			\
474 	OPT_BOOLEAN('\0', "debug-mmio", &(cfg)->mmio_debug,		\
475 			"Enable MMIO debugging"),			\
476 	OPT_INTEGER('\0', "debug-iodelay", &debug_iodelay,		\
477 			"Delay IO by millisecond"),			\
478 	OPT_END()							\
479 	};
480 
481 /*
482  * Serialize debug printout so that the output of multiple vcpus does not
483  * get mixed up:
484  */
485 static int printout_done;
486 
487 static void handle_sigusr1(int sig)
488 {
489 	struct kvm_cpu *cpu = current_kvm_cpu;
490 	int fd = kvm_cpu__get_debug_fd();
491 
492 	if (!cpu || cpu->needs_nmi)
493 		return;
494 
495 	dprintf(fd, "\n #\n # vCPU #%ld's dump:\n #\n", cpu->cpu_id);
496 	kvm_cpu__show_registers(cpu);
497 	kvm_cpu__show_code(cpu);
498 	kvm_cpu__show_page_tables(cpu);
499 	fflush(stdout);
500 	printout_done = 1;
501 	mb();
502 }
503 
504 /* Pause/resume the guest using SIGUSR2 */
505 static int is_paused;
506 
507 static void handle_pause(int fd, u32 type, u32 len, u8 *msg)
508 {
509 	if (WARN_ON(len))
510 		return;
511 
512 	if (type == KVM_IPC_RESUME && is_paused) {
513 		kvm->vm_state = KVM_VMSTATE_RUNNING;
514 		kvm__continue();
515 	} else if (type == KVM_IPC_PAUSE && !is_paused) {
516 		kvm->vm_state = KVM_VMSTATE_PAUSED;
517 		ioctl(kvm->vm_fd, KVM_KVMCLOCK_CTRL);
518 		kvm__pause();
519 	} else {
520 		return;
521 	}
522 
523 	is_paused = !is_paused;
524 }
525 
526 static void handle_vmstate(int fd, u32 type, u32 len, u8 *msg)
527 {
528 	int r = 0;
529 
530 	if (type == KVM_IPC_VMSTATE)
531 		r = write(fd, &kvm->vm_state, sizeof(kvm->vm_state));
532 
533 	if (r < 0)
534 		pr_warning("Failed sending VMSTATE");
535 }
536 
537 static void handle_debug(int fd, u32 type, u32 len, u8 *msg)
538 {
539 	int i;
540 	struct debug_cmd_params *params;
541 	u32 dbg_type;
542 	u32 vcpu;
543 
544 	if (WARN_ON(type != KVM_IPC_DEBUG || len != sizeof(*params)))
545 		return;
546 
547 	params = (void *)msg;
548 	dbg_type = params->dbg_type;
549 	vcpu = params->cpu;
550 
551 	if (dbg_type & KVM_DEBUG_CMD_TYPE_SYSRQ)
552 		serial8250__inject_sysrq(kvm, params->sysrq);
553 
554 	if (dbg_type & KVM_DEBUG_CMD_TYPE_NMI) {
555 		if ((int)vcpu >= kvm->nrcpus)
556 			return;
557 
558 		kvm_cpus[vcpu]->needs_nmi = 1;
559 		pthread_kill(kvm_cpus[vcpu]->thread, SIGUSR1);
560 	}
561 
562 	if (!(dbg_type & KVM_DEBUG_CMD_TYPE_DUMP))
563 		return;
564 
565 	for (i = 0; i < nrcpus; i++) {
566 		struct kvm_cpu *cpu = kvm_cpus[i];
567 
568 		if (!cpu)
569 			continue;
570 
571 		printout_done = 0;
572 
573 		kvm_cpu__set_debug_fd(fd);
574 		pthread_kill(cpu->thread, SIGUSR1);
575 		/*
576 		 * Wait for the vCPU to dump state before signalling
577 		 * the next thread. Since this is debug code it does
578 		 * not matter that we are burning CPU time a bit:
579 		 */
580 		while (!printout_done)
581 			mb();
582 	}
583 
584 	close(fd);
585 
586 	serial8250__inject_sysrq(kvm, 'p');
587 }
588 
589 static void handle_sigalrm(int sig)
590 {
591 	kvm__arch_periodic_poll(kvm);
592 }
593 
594 static void handle_stop(int fd, u32 type, u32 len, u8 *msg)
595 {
596 	if (WARN_ON(type != KVM_IPC_STOP || len))
597 		return;
598 
599 	kvm_cpu__reboot();
600 }
601 
602 static void *kvm_cpu_thread(void *arg)
603 {
604 	current_kvm_cpu		= arg;
605 
606 	if (kvm_cpu__start(current_kvm_cpu))
607 		goto panic_kvm;
608 
609 	return (void *) (intptr_t) 0;
610 
611 panic_kvm:
612 	fprintf(stderr, "KVM exit reason: %u (\"%s\")\n",
613 		current_kvm_cpu->kvm_run->exit_reason,
614 		kvm_exit_reasons[current_kvm_cpu->kvm_run->exit_reason]);
615 	if (current_kvm_cpu->kvm_run->exit_reason == KVM_EXIT_UNKNOWN)
616 		fprintf(stderr, "KVM exit code: 0x%Lu\n",
617 			current_kvm_cpu->kvm_run->hw.hardware_exit_reason);
618 
619 	kvm_cpu__set_debug_fd(STDOUT_FILENO);
620 	kvm_cpu__show_registers(current_kvm_cpu);
621 	kvm_cpu__show_code(current_kvm_cpu);
622 	kvm_cpu__show_page_tables(current_kvm_cpu);
623 
624 	return (void *) (intptr_t) 1;
625 }
626 
627 static char kernel[PATH_MAX];
628 
629 static const char *host_kernels[] = {
630 	"/boot/vmlinuz",
631 	"/boot/bzImage",
632 	NULL
633 };
634 
635 static const char *default_kernels[] = {
636 	"./bzImage",
637 	"arch/" BUILD_ARCH "/boot/bzImage",
638 	"../../arch/" BUILD_ARCH "/boot/bzImage",
639 	NULL
640 };
641 
642 static const char *default_vmlinux[] = {
643 	"vmlinux",
644 	"../../../vmlinux",
645 	"../../vmlinux",
646 	NULL
647 };
648 
649 static void kernel_usage_with_options(void)
650 {
651 	const char **k;
652 	struct utsname uts;
653 
654 	fprintf(stderr, "Fatal: could not find default kernel image in:\n");
655 	k = &default_kernels[0];
656 	while (*k) {
657 		fprintf(stderr, "\t%s\n", *k);
658 		k++;
659 	}
660 
661 	if (uname(&uts) < 0)
662 		return;
663 
664 	k = &host_kernels[0];
665 	while (*k) {
666 		if (snprintf(kernel, PATH_MAX, "%s-%s", *k, uts.release) < 0)
667 			return;
668 		fprintf(stderr, "\t%s\n", kernel);
669 		k++;
670 	}
671 	fprintf(stderr, "\nPlease see '%s run --help' for more options.\n\n",
672 		KVM_BINARY_NAME);
673 }
674 
675 static u64 host_ram_size(void)
676 {
677 	long page_size;
678 	long nr_pages;
679 
680 	nr_pages	= sysconf(_SC_PHYS_PAGES);
681 	if (nr_pages < 0) {
682 		pr_warning("sysconf(_SC_PHYS_PAGES) failed");
683 		return 0;
684 	}
685 
686 	page_size	= sysconf(_SC_PAGE_SIZE);
687 	if (page_size < 0) {
688 		pr_warning("sysconf(_SC_PAGE_SIZE) failed");
689 		return 0;
690 	}
691 
692 	return (nr_pages * page_size) >> MB_SHIFT;
693 }
694 
695 /*
696  * If user didn't specify how much memory it wants to allocate for the guest,
697  * avoid filling the whole host RAM.
698  */
699 #define RAM_SIZE_RATIO		0.8
700 
701 static u64 get_ram_size(int nr_cpus)
702 {
703 	u64 available;
704 	u64 ram_size;
705 
706 	ram_size	= 64 * (nr_cpus + 3);
707 
708 	available	= host_ram_size() * RAM_SIZE_RATIO;
709 	if (!available)
710 		available = MIN_RAM_SIZE_MB;
711 
712 	if (ram_size > available)
713 		ram_size	= available;
714 
715 	return ram_size;
716 }
717 
718 static const char *find_kernel(void)
719 {
720 	const char **k;
721 	struct stat st;
722 	struct utsname uts;
723 
724 	k = &default_kernels[0];
725 	while (*k) {
726 		if (stat(*k, &st) < 0 || !S_ISREG(st.st_mode)) {
727 			k++;
728 			continue;
729 		}
730 		strncpy(kernel, *k, PATH_MAX);
731 		return kernel;
732 	}
733 
734 	if (uname(&uts) < 0)
735 		return NULL;
736 
737 	k = &host_kernels[0];
738 	while (*k) {
739 		if (snprintf(kernel, PATH_MAX, "%s-%s", *k, uts.release) < 0)
740 			return NULL;
741 
742 		if (stat(kernel, &st) < 0 || !S_ISREG(st.st_mode)) {
743 			k++;
744 			continue;
745 		}
746 		return kernel;
747 
748 	}
749 	return NULL;
750 }
751 
752 static const char *find_vmlinux(void)
753 {
754 	const char **vmlinux;
755 
756 	vmlinux = &default_vmlinux[0];
757 	while (*vmlinux) {
758 		struct stat st;
759 
760 		if (stat(*vmlinux, &st) < 0 || !S_ISREG(st.st_mode)) {
761 			vmlinux++;
762 			continue;
763 		}
764 		return *vmlinux;
765 	}
766 	return NULL;
767 }
768 
769 void kvm_run_help(void)
770 {
771 	BUILD_OPTIONS(options, &kvm->cfg);
772 	usage_with_options(run_usage, options);
773 }
774 
775 static int kvm_setup_guest_init(void)
776 {
777 	const char *rootfs = kvm->cfg.custom_rootfs_name;
778 	char tmp[PATH_MAX];
779 	size_t size;
780 	int fd, ret;
781 	char *data;
782 
783 	/* Setup /virt/init */
784 	size = (size_t)&_binary_guest_init_size;
785 	data = (char *)&_binary_guest_init_start;
786 	snprintf(tmp, PATH_MAX, "%s%s/virt/init", kvm__get_dir(), rootfs);
787 	remove(tmp);
788 	fd = open(tmp, O_CREAT | O_WRONLY, 0755);
789 	if (fd < 0)
790 		die("Fail to setup %s", tmp);
791 	ret = xwrite(fd, data, size);
792 	if (ret < 0)
793 		die("Fail to setup %s", tmp);
794 	close(fd);
795 
796 	return 0;
797 }
798 
799 static int kvm_run_set_sandbox(void)
800 {
801 	const char *guestfs_name = kvm->cfg.custom_rootfs_name;
802 	char path[PATH_MAX], script[PATH_MAX], *tmp;
803 
804 	snprintf(path, PATH_MAX, "%s%s/virt/sandbox.sh", kvm__get_dir(), guestfs_name);
805 
806 	remove(path);
807 
808 	if (kvm->cfg.sandbox == NULL)
809 		return 0;
810 
811 	tmp = realpath(kvm->cfg.sandbox, NULL);
812 	if (tmp == NULL)
813 		return -ENOMEM;
814 
815 	snprintf(script, PATH_MAX, "/host/%s", tmp);
816 	free(tmp);
817 
818 	return symlink(script, path);
819 }
820 
821 static void kvm_write_sandbox_cmd_exactly(int fd, const char *arg)
822 {
823 	const char *single_quote;
824 
825 	if (!*arg) { /* zero length string */
826 		if (write(fd, "''", 2) <= 0)
827 			die("Failed writing sandbox script");
828 		return;
829 	}
830 
831 	while (*arg) {
832 		single_quote = strchrnul(arg, '\'');
833 
834 		/* write non-single-quote string as #('string') */
835 		if (arg != single_quote) {
836 			if (write(fd, "'", 1) <= 0 ||
837 			    write(fd, arg, single_quote - arg) <= 0 ||
838 			    write(fd, "'", 1) <= 0)
839 				die("Failed writing sandbox script");
840 		}
841 
842 		/* write single quote as #("'") */
843 		if (*single_quote) {
844 			if (write(fd, "\"'\"", 3) <= 0)
845 				die("Failed writing sandbox script");
846 		} else
847 			break;
848 
849 		arg = single_quote + 1;
850 	}
851 }
852 
853 static void resolve_program(const char *src, char *dst, size_t len)
854 {
855 	struct stat st;
856 	int err;
857 
858 	err = stat(src, &st);
859 
860 	if (!err && S_ISREG(st.st_mode)) {
861 		char resolved_path[PATH_MAX];
862 
863 		if (!realpath(src, resolved_path))
864 			die("Unable to resolve program %s: %s\n", src, strerror(errno));
865 
866 		snprintf(dst, len, "/host%s", resolved_path);
867 	} else
868 		strncpy(dst, src, len);
869 }
870 
871 static void kvm_run_write_sandbox_cmd(const char **argv, int argc)
872 {
873 	const char script_hdr[] = "#! /bin/bash\n\n";
874 	char program[PATH_MAX];
875 	int fd;
876 
877 	remove(kvm->cfg.sandbox);
878 
879 	fd = open(kvm->cfg.sandbox, O_RDWR | O_CREAT, 0777);
880 	if (fd < 0)
881 		die("Failed creating sandbox script");
882 
883 	if (write(fd, script_hdr, sizeof(script_hdr) - 1) <= 0)
884 		die("Failed writing sandbox script");
885 
886 	resolve_program(argv[0], program, PATH_MAX);
887 	kvm_write_sandbox_cmd_exactly(fd, program);
888 
889 	argv++;
890 	argc--;
891 
892 	while (argc) {
893 		if (write(fd, " ", 1) <= 0)
894 			die("Failed writing sandbox script");
895 
896 		kvm_write_sandbox_cmd_exactly(fd, argv[0]);
897 		argv++;
898 		argc--;
899 	}
900 	if (write(fd, "\n", 1) <= 0)
901 		die("Failed writing sandbox script");
902 
903 	close(fd);
904 }
905 
906 static int kvm_cmd_run_init(int argc, const char **argv)
907 {
908 	static char real_cmdline[2048], default_name[20];
909 	struct framebuffer *fb = NULL;
910 	unsigned int nr_online_cpus;
911 	int max_cpus, recommended_cpus;
912 	int i, r;
913 
914 	kvm = kvm__new();
915 	if (IS_ERR(kvm))
916 		return PTR_ERR(kvm);
917 
918 	signal(SIGALRM, handle_sigalrm);
919 	kvm_ipc__register_handler(KVM_IPC_DEBUG, handle_debug);
920 	signal(SIGUSR1, handle_sigusr1);
921 	kvm_ipc__register_handler(KVM_IPC_PAUSE, handle_pause);
922 	kvm_ipc__register_handler(KVM_IPC_RESUME, handle_pause);
923 	kvm_ipc__register_handler(KVM_IPC_STOP, handle_stop);
924 	kvm_ipc__register_handler(KVM_IPC_VMSTATE, handle_vmstate);
925 
926 	nr_online_cpus = sysconf(_SC_NPROCESSORS_ONLN);
927 	kvm->cfg.custom_rootfs_name = "default";
928 
929 	while (argc != 0) {
930 		BUILD_OPTIONS(options, &kvm->cfg);
931 		argc = parse_options(argc, argv, options, run_usage,
932 				PARSE_OPT_STOP_AT_NON_OPTION |
933 				PARSE_OPT_KEEP_DASHDASH);
934 		if (argc != 0) {
935 			/* Cusrom options, should have been handled elsewhere */
936 			if (strcmp(argv[0], "--") == 0) {
937 				if (kvm_run_wrapper == KVM_RUN_SANDBOX) {
938 					kvm->cfg.sandbox = DEFAULT_SANDBOX_FILENAME;
939 					kvm_run_write_sandbox_cmd(argv+1, argc-1);
940 					break;
941 				}
942 			}
943 
944 			if ((kvm_run_wrapper == KVM_RUN_DEFAULT && kvm->cfg.kernel_filename) ||
945 				(kvm_run_wrapper == KVM_RUN_SANDBOX && kvm->cfg.sandbox)) {
946 				fprintf(stderr, "Cannot handle parameter: "
947 						"%s\n", argv[0]);
948 				usage_with_options(run_usage, options);
949 				free(kvm);
950 				return -EINVAL;
951 			}
952 			if (kvm_run_wrapper == KVM_RUN_SANDBOX) {
953 				/*
954 				 * first unhandled parameter is treated as
955 				 * sandbox command
956 				 */
957 				kvm->cfg.sandbox = DEFAULT_SANDBOX_FILENAME;
958 				kvm_run_write_sandbox_cmd(argv, argc);
959 			} else {
960 				/*
961 				 * first unhandled parameter is treated as a kernel
962 				 * image
963 				 */
964 				kvm->cfg.kernel_filename = argv[0];
965 			}
966 			argv++;
967 			argc--;
968 		}
969 
970 	}
971 
972 	if (!kvm->cfg.kernel_filename)
973 		kvm->cfg.kernel_filename = find_kernel();
974 
975 	if (!kvm->cfg.kernel_filename) {
976 		kernel_usage_with_options();
977 		return -EINVAL;
978 	}
979 
980 	kvm->cfg.vmlinux_filename = find_vmlinux();
981 
982 	if (nrcpus == 0)
983 		nrcpus = nr_online_cpus;
984 
985 	if (!kvm->cfg.ram_size)
986 		kvm->cfg.ram_size = get_ram_size(nrcpus);
987 
988 	if (kvm->cfg.ram_size < MIN_RAM_SIZE_MB)
989 		die("Not enough memory specified: %lluMB (min %lluMB)", kvm->cfg.ram_size, MIN_RAM_SIZE_MB);
990 
991 	if (kvm->cfg.ram_size > host_ram_size())
992 		pr_warning("Guest memory size %lluMB exceeds host physical RAM size %lluMB", kvm->cfg.ram_size, host_ram_size());
993 
994 	kvm->cfg.ram_size <<= MB_SHIFT;
995 
996 	if (!kvm->cfg.dev)
997 		kvm->cfg.dev = DEFAULT_KVM_DEV;
998 
999 	if (!kvm->cfg.console)
1000 		kvm->cfg.console = DEFAULT_CONSOLE;
1001 
1002 	if (!strncmp(kvm->cfg.console, "virtio", 6))
1003 		kvm->cfg.active_console  = CONSOLE_VIRTIO;
1004 	else if (!strncmp(kvm->cfg.console, "serial", 6))
1005 		kvm->cfg.active_console  = CONSOLE_8250;
1006 	else if (!strncmp(kvm->cfg.console, "hv", 2))
1007 		kvm->cfg.active_console = CONSOLE_HV;
1008 	else
1009 		pr_warning("No console!");
1010 
1011 	if (!kvm->cfg.host_ip)
1012 		kvm->cfg.host_ip = DEFAULT_HOST_ADDR;
1013 
1014 	if (!kvm->cfg.guest_ip)
1015 		kvm->cfg.guest_ip = DEFAULT_GUEST_ADDR;
1016 
1017 	if (!kvm->cfg.guest_mac)
1018 		kvm->cfg.guest_mac = DEFAULT_GUEST_MAC;
1019 
1020 	if (!kvm->cfg.host_mac)
1021 		kvm->cfg.host_mac = DEFAULT_HOST_MAC;
1022 
1023 	if (!kvm->cfg.script)
1024 		kvm->cfg.script = DEFAULT_SCRIPT;
1025 
1026 	term_init();
1027 
1028 	if (!kvm->cfg.guest_name) {
1029 		if (kvm->cfg.custom_rootfs) {
1030 			kvm->cfg.guest_name = kvm->cfg.custom_rootfs_name;
1031 		} else {
1032 			sprintf(default_name, "guest-%u", getpid());
1033 			kvm->cfg.guest_name = default_name;
1034 		}
1035 	}
1036 
1037 	r = kvm__init(kvm);
1038 	if (r)
1039 		goto fail;
1040 
1041 	r = ioeventfd__init(kvm);
1042 	if (r < 0) {
1043 		pr_err("ioeventfd__init() failed with error %d\n", r);
1044 		goto fail;
1045 	}
1046 
1047 	max_cpus = kvm__max_cpus(kvm);
1048 	recommended_cpus = kvm__recommended_cpus(kvm);
1049 
1050 	if (nrcpus > max_cpus) {
1051 		printf("  # Limit the number of CPUs to %d\n", max_cpus);
1052 		nrcpus = max_cpus;
1053 	} else if (nrcpus > recommended_cpus) {
1054 		printf("  # Warning: The maximum recommended amount of VCPUs"
1055 			" is %d\n", recommended_cpus);
1056 	}
1057 
1058 	kvm->nrcpus = nrcpus;
1059 
1060 	/* Alloc one pointer too many, so array ends up 0-terminated */
1061 	kvm_cpus = calloc(nrcpus + 1, sizeof(void *));
1062 	if (!kvm_cpus)
1063 		die("Couldn't allocate array for %d CPUs", nrcpus);
1064 
1065 	r = irq__init(kvm);
1066 	if (r < 0) {
1067 		pr_err("irq__init() failed with error %d\n", r);
1068 		goto fail;
1069 	}
1070 
1071 	r = pci__init(kvm);
1072 	if (r < 0) {
1073 		pr_err("pci__init() failed with error %d\n", r);
1074 		goto fail;
1075 	}
1076 
1077 	r = ioport__init(kvm);
1078 	if (r < 0) {
1079 		pr_err("ioport__init() failed with error %d\n", r);
1080 		goto fail;
1081 	}
1082 
1083 	/*
1084 	 * vidmode should be either specified
1085 	 * either set by default
1086 	 */
1087 	if (kvm->cfg.vnc || kvm->cfg.sdl) {
1088 		if (vidmode == -1)
1089 			vidmode = 0x312;
1090 	} else {
1091 		vidmode = 0;
1092 	}
1093 
1094 	memset(real_cmdline, 0, sizeof(real_cmdline));
1095 	kvm__arch_set_cmdline(real_cmdline, kvm->cfg.vnc || kvm->cfg.sdl);
1096 
1097 	if (strlen(real_cmdline) > 0)
1098 		strcat(real_cmdline, " ");
1099 
1100 	if (kvm->cfg.kernel_cmdline)
1101 		strlcat(real_cmdline, kvm->cfg.kernel_cmdline, sizeof(real_cmdline));
1102 
1103 	if (!kvm->cfg.using_rootfs && !kvm->cfg.disk_image[0].filename && !kvm->cfg.initrd_filename) {
1104 		char tmp[PATH_MAX];
1105 
1106 		kvm_setup_create_new(kvm->cfg.custom_rootfs_name);
1107 		kvm_setup_resolv(kvm->cfg.custom_rootfs_name);
1108 
1109 		snprintf(tmp, PATH_MAX, "%s%s", kvm__get_dir(), "default");
1110 		if (virtio_9p__register(kvm, tmp, "/dev/root") < 0)
1111 			die("Unable to initialize virtio 9p");
1112 		if (virtio_9p__register(kvm, "/", "hostfs") < 0)
1113 			die("Unable to initialize virtio 9p");
1114 		kvm->cfg.using_rootfs = kvm->cfg.custom_rootfs = 1;
1115 	}
1116 
1117 	if (kvm->cfg.using_rootfs) {
1118 		strcat(real_cmdline, " root=/dev/root rw rootflags=rw,trans=virtio,version=9p2000.L rootfstype=9p");
1119 		if (kvm->cfg.custom_rootfs) {
1120 			kvm_run_set_sandbox();
1121 
1122 			strcat(real_cmdline, " init=/virt/init");
1123 
1124 			if (!kvm->cfg.no_dhcp)
1125 				strcat(real_cmdline, "  ip=dhcp");
1126 			if (kvm_setup_guest_init())
1127 				die("Failed to setup init for guest.");
1128 		}
1129 	} else if (!strstr(real_cmdline, "root=")) {
1130 		strlcat(real_cmdline, " root=/dev/vda rw ", sizeof(real_cmdline));
1131 	}
1132 
1133 	if (kvm->cfg.image_count) {
1134 		kvm->nr_disks = kvm->cfg.image_count;
1135 		kvm->disks = disk_image__open_all((struct disk_image_params *)&kvm->cfg.disk_image, kvm->cfg.image_count);
1136 		if (IS_ERR(kvm->disks)) {
1137 			r = PTR_ERR(kvm->disks);
1138 			pr_err("disk_image__open_all() failed with error %ld\n",
1139 					PTR_ERR(kvm->disks));
1140 			goto fail;
1141 		}
1142 	}
1143 
1144 	printf("  # %s run -k %s -m %Lu -c %d --name %s\n", KVM_BINARY_NAME,
1145 		kvm->cfg.kernel_filename, kvm->cfg.ram_size / 1024 / 1024, nrcpus, kvm->cfg.guest_name);
1146 
1147 	if (!kvm->cfg.firmware_filename) {
1148 		if (!kvm__load_kernel(kvm, kvm->cfg.kernel_filename,
1149 				kvm->cfg.initrd_filename, real_cmdline, vidmode))
1150 			die("unable to load kernel %s", kvm->cfg.kernel_filename);
1151 
1152 		kvm->vmlinux = kvm->cfg.vmlinux_filename;
1153 		r = symbol_init(kvm);
1154 		if (r < 0)
1155 			pr_debug("symbol_init() failed with error %d\n", r);
1156 	}
1157 
1158 	ioport__setup_arch();
1159 
1160 	r = rtc__init(kvm);
1161 	if (r < 0) {
1162 		pr_err("rtc__init() failed with error %d\n", r);
1163 		goto fail;
1164 	}
1165 
1166 	r = serial8250__init(kvm);
1167 	if (r < 0) {
1168 		pr_err("serial__init() failed with error %d\n", r);
1169 		goto fail;
1170 	}
1171 
1172 	r = virtio_blk__init(kvm);
1173 	if (r < 0) {
1174 		pr_err("virtio_blk__init() failed with error %d\n", r);
1175 		goto fail;
1176 	}
1177 
1178 	r = virtio_scsi_init(kvm);
1179 	if (r < 0) {
1180 		pr_err("virtio_scsi_init() failed with error %d\n", r);
1181 		goto fail;
1182 	}
1183 
1184 
1185 	if (kvm->cfg.active_console == CONSOLE_VIRTIO)
1186 		virtio_console__init(kvm);
1187 
1188 	if (kvm->cfg.virtio_rng)
1189 		virtio_rng__init(kvm);
1190 
1191 	if (kvm->cfg.balloon)
1192 		virtio_bln__init(kvm);
1193 
1194 	if (!kvm->cfg.network)
1195 		kvm->cfg.network = DEFAULT_NETWORK;
1196 
1197 	virtio_9p__init(kvm);
1198 
1199 	for (i = 0; i < kvm->cfg.num_net_devices; i++) {
1200 		kvm->cfg.net_params[i].kvm = kvm;
1201 		virtio_net__init(&kvm->cfg.net_params[i]);
1202 	}
1203 
1204 	if (kvm->cfg.num_net_devices == 0 && kvm->cfg.no_net == 0) {
1205 		struct virtio_net_params net_params;
1206 
1207 		net_params = (struct virtio_net_params) {
1208 			.guest_ip	= kvm->cfg.guest_ip,
1209 			.host_ip	= kvm->cfg.host_ip,
1210 			.kvm		= kvm,
1211 			.script		= kvm->cfg.script,
1212 			.mode		= NET_MODE_USER,
1213 		};
1214 		str_to_mac(kvm->cfg.guest_mac, net_params.guest_mac);
1215 		str_to_mac(kvm->cfg.host_mac, net_params.host_mac);
1216 
1217 		virtio_net__init(&net_params);
1218 	}
1219 
1220 	kvm__init_ram(kvm);
1221 
1222 #ifdef CONFIG_X86
1223 	kbd__init(kvm);
1224 #endif
1225 
1226 	pci_shmem__init(kvm);
1227 
1228 	if (kvm->cfg.vnc || kvm->cfg.sdl) {
1229 		fb = vesa__init(kvm);
1230 		if (IS_ERR(fb)) {
1231 			pr_err("vesa__init() failed with error %ld\n", PTR_ERR(fb));
1232 			goto fail;
1233 		}
1234 	}
1235 
1236 	if (kvm->cfg.vnc && fb) {
1237 		r = vnc__init(fb);
1238 		if (r < 0) {
1239 			pr_err("vnc__init() failed with error %d\n", r);
1240 			goto fail;
1241 		}
1242 	}
1243 
1244 	if (kvm->cfg.sdl && fb) {
1245 		sdl__init(fb);
1246 		if (r < 0) {
1247 			pr_err("sdl__init() failed with error %d\n", r);
1248 			goto fail;
1249 		}
1250 	}
1251 
1252 	r = fb__start();
1253 	if (r < 0) {
1254 		pr_err("fb__init() failed with error %d\n", r);
1255 		goto fail;
1256 	}
1257 
1258 	/* Device init all done; firmware init must
1259 	 * come after this (it may set up device trees etc.)
1260 	 */
1261 
1262 	kvm__start_timer(kvm);
1263 
1264 	if (kvm->cfg.firmware_filename) {
1265 		if (!kvm__load_firmware(kvm, kvm->cfg.firmware_filename))
1266 			die("unable to load firmware image %s: %s", kvm->cfg.firmware_filename, strerror(errno));
1267 	} else {
1268 		kvm__arch_setup_firmware(kvm);
1269 		if (r < 0) {
1270 			pr_err("kvm__arch_setup_firmware() failed with error %d\n", r);
1271 			goto fail;
1272 		}
1273 	}
1274 
1275 	for (i = 0; i < nrcpus; i++) {
1276 		kvm_cpus[i] = kvm_cpu__init(kvm, i);
1277 		if (!kvm_cpus[i])
1278 			die("unable to initialize KVM VCPU");
1279 	}
1280 
1281 	thread_pool__init(nr_online_cpus);
1282 fail:
1283 	return r;
1284 }
1285 
1286 static int kvm_cmd_run_work(void)
1287 {
1288 	int i, r = -1;
1289 	void *ret = NULL;
1290 
1291 	for (i = 0; i < nrcpus; i++) {
1292 		if (pthread_create(&kvm_cpus[i]->thread, NULL, kvm_cpu_thread, kvm_cpus[i]) != 0)
1293 			die("unable to create KVM VCPU thread");
1294 	}
1295 
1296 	/* Only VCPU #0 is going to exit by itself when shutting down */
1297 	if (pthread_join(kvm_cpus[0]->thread, &ret) != 0)
1298 		r = 0;
1299 
1300 	kvm_cpu__delete(kvm_cpus[0]);
1301 	kvm_cpus[0] = NULL;
1302 
1303 	for (i = 1; i < nrcpus; i++) {
1304 		if (kvm_cpus[i]->is_running) {
1305 			pthread_kill(kvm_cpus[i]->thread, SIGKVMEXIT);
1306 			if (pthread_join(kvm_cpus[i]->thread, &ret) != 0)
1307 				die("pthread_join");
1308 			kvm_cpu__delete(kvm_cpus[i]);
1309 		}
1310 		if (ret == NULL)
1311 			r = 0;
1312 	}
1313 
1314 	return r;
1315 }
1316 
1317 static void kvm_cmd_run_exit(int guest_ret)
1318 {
1319 	int r = 0;
1320 
1321 	compat__print_all_messages();
1322 
1323 	r = symbol_exit(kvm);
1324 	if (r < 0)
1325 		pr_warning("symbol_exit() failed with error %d\n", r);
1326 
1327 	r = irq__exit(kvm);
1328 	if (r < 0)
1329 		pr_warning("irq__exit() failed with error %d\n", r);
1330 
1331 	fb__stop();
1332 
1333 	r = virtio_scsi_exit(kvm);
1334 	if (r < 0)
1335 		pr_warning("virtio_scsi_exit() failed with error %d\n", r);
1336 
1337 	r = virtio_blk__exit(kvm);
1338 	if (r < 0)
1339 		pr_warning("virtio_blk__exit() failed with error %d\n", r);
1340 
1341 	r = virtio_rng__exit(kvm);
1342 	if (r < 0)
1343 		pr_warning("virtio_rng__exit() failed with error %d\n", r);
1344 
1345 	r = disk_image__close_all(kvm->disks, kvm->cfg.image_count);
1346 	if (r < 0)
1347 		pr_warning("disk_image__close_all() failed with error %d\n", r);
1348 
1349 	r = serial8250__exit(kvm);
1350 	if (r < 0)
1351 		pr_warning("serial8250__exit() failed with error %d\n", r);
1352 
1353 	r = rtc__exit(kvm);
1354 	if (r < 0)
1355 		pr_warning("rtc__exit() failed with error %d\n", r);
1356 
1357 	r = kvm__arch_free_firmware(kvm);
1358 	if (r < 0)
1359 		pr_warning("kvm__arch_free_firmware() failed with error %d\n", r);
1360 
1361 	r = ioport__exit(kvm);
1362 	if (r < 0)
1363 		pr_warning("ioport__exit() failed with error %d\n", r);
1364 
1365 	r = ioeventfd__exit(kvm);
1366 	if (r < 0)
1367 		pr_warning("ioeventfd__exit() failed with error %d\n", r);
1368 
1369 	r = pci__exit(kvm);
1370 	if (r < 0)
1371 		pr_warning("pci__exit() failed with error %d\n", r);
1372 
1373 	r = kvm__exit(kvm);
1374 	if (r < 0)
1375 		pr_warning("pci__exit() failed with error %d\n", r);
1376 
1377 	free(kvm_cpus);
1378 
1379 	if (guest_ret == 0)
1380 		printf("\n  # KVM session ended normally.\n");
1381 }
1382 
1383 int kvm_cmd_run(int argc, const char **argv, const char *prefix)
1384 {
1385 	int r, ret = -EFAULT;
1386 
1387 	r = kvm_cmd_run_init(argc, argv);
1388 	if (r < 0)
1389 		return r;
1390 
1391 	ret = kvm_cmd_run_work();
1392 	kvm_cmd_run_exit(ret);
1393 
1394 	return ret;
1395 }
1396