]> git.kernelconcepts.de Git - karo-tx-linux.git/blob - tools/kvm/kvm-run.c
bd7691fbb8e01779de9a1dda5f91f9bbd9688e09
[karo-tx-linux.git] / tools / kvm / kvm-run.c
1 #include <stdio.h>
2 #include <string.h>
3 #include <signal.h>
4 #include <unistd.h>
5 #include <stdlib.h>
6 #include <termios.h>
7 #include <sys/utsname.h>
8 #include <sys/types.h>
9 #include <sys/stat.h>
10 #include <ctype.h>
11
12 /* user defined header files */
13 #include <linux/types.h>
14 #include <kvm/kvm.h>
15 #include <kvm/kvm-cpu.h>
16 #include <kvm/8250-serial.h>
17 #include <kvm/virtio-blk.h>
18 #include <kvm/virtio-net.h>
19 #include <kvm/virtio-console.h>
20 #include <kvm/virtio-rng.h>
21 #include <kvm/disk-image.h>
22 #include <kvm/util.h>
23 #include <kvm/pci.h>
24 #include <kvm/rtc.h>
25 #include <kvm/term.h>
26 #include <kvm/ioport.h>
27 #include <kvm/threadpool.h>
28 #include <kvm/barrier.h>
29 #include <kvm/symbol.h>
30 #include <kvm/virtio-9p.h>
31
32 /* header files for gitish interface  */
33 #include <kvm/kvm-run.h>
34 #include <kvm/parse-options.h>
35 #include <kvm/mutex.h>
36
37 #define DEFAULT_KVM_DEV         "/dev/kvm"
38 #define DEFAULT_CONSOLE         "serial"
39 #define DEFAULT_NETWORK         "virtio"
40 #define DEFAULT_HOST_ADDR       "192.168.33.2"
41 #define DEFAULT_GUEST_MAC       "00:11:22:33:44:55"
42 #define DEFAULT_SCRIPT          "none"
43
44 #define MB_SHIFT                (20)
45 #define MIN_RAM_SIZE_MB         (64ULL)
46 #define MIN_RAM_SIZE_BYTE       (MIN_RAM_SIZE_MB << MB_SHIFT)
47
48 static struct kvm *kvm;
49 static struct kvm_cpu *kvm_cpus[KVM_NR_CPUS];
50 static __thread struct kvm_cpu *current_kvm_cpu;
51
52 static u64 ram_size;
53 static u8  image_count;
54 static const char *kernel_cmdline;
55 static const char *kernel_filename;
56 static const char *vmlinux_filename;
57 static const char *initrd_filename;
58 static const char *image_filename[MAX_DISK_IMAGES];
59 static const char *console;
60 static const char *kvm_dev;
61 static const char *network;
62 static const char *host_ip_addr;
63 static const char *guest_mac;
64 static const char *script;
65 static const char *virtio_9p_dir;
66 static bool single_step;
67 static bool readonly_image[MAX_DISK_IMAGES];
68 static bool virtio_rng;
69 extern bool ioport_debug;
70 extern int  active_console;
71
72 bool do_debug_print = false;
73
74 static int nrcpus;
75
76 static const char * const run_usage[] = {
77         "kvm run [<options>] [<kernel image>]",
78         NULL
79 };
80
81 static int img_name_parser(const struct option *opt, const char *arg, int unset)
82 {
83         char *sep;
84
85         if (image_count >= MAX_DISK_IMAGES)
86                 die("Currently only 4 images are supported");
87
88         image_filename[image_count] = arg;
89         sep = strstr(arg, ",");
90         if (sep) {
91                 if (strcmp(sep + 1, "ro") == 0)
92                         readonly_image[image_count] = 1;
93                 *sep = 0;
94         }
95
96         image_count++;
97
98         return 0;
99 }
100
101 static const struct option options[] = {
102         OPT_GROUP("Basic options:"),
103         OPT_INTEGER('c', "cpus", &nrcpus, "Number of CPUs"),
104         OPT_U64('m', "mem", &ram_size, "Virtual machine memory size in MiB."),
105         OPT_CALLBACK('d', "disk", NULL, "image", "Disk image", img_name_parser),
106         OPT_STRING('\0', "console", &console, "serial or virtio",
107                         "Console to use"),
108         OPT_BOOLEAN('\0', "rng", &virtio_rng,
109                         "Enable virtio Random Number Generator"),
110         OPT_STRING('\0', "kvm-dev", &kvm_dev, "kvm-dev", "KVM device file"),
111         OPT_STRING('\0', "virtio-9p", &virtio_9p_dir, "root dir",
112                         "Enable 9p over virtio"),
113
114         OPT_GROUP("Kernel options:"),
115         OPT_STRING('k', "kernel", &kernel_filename, "kernel",
116                         "Kernel to boot in virtual machine"),
117         OPT_STRING('i', "initrd", &initrd_filename, "initrd",
118                         "Initial RAM disk image"),
119         OPT_STRING('p', "params", &kernel_cmdline, "params",
120                         "Kernel command line arguments"),
121
122         OPT_GROUP("Networking options:"),
123         OPT_STRING('n', "network", &network, "virtio",
124                         "Network to use"),
125         OPT_STRING('\0', "host-ip-addr", &host_ip_addr, "a.b.c.d",
126                         "Assign this address to the host side networking"),
127         OPT_STRING('\0', "guest-mac", &guest_mac, "aa:bb:cc:dd:ee:ff",
128                         "Assign this address to the guest side NIC"),
129         OPT_STRING('\0', "tapscript", &script, "Script path",
130                          "Assign a script to process created tap device"),
131
132         OPT_GROUP("Debug options:"),
133         OPT_BOOLEAN('\0', "debug", &do_debug_print,
134                         "Enable debug messages"),
135         OPT_BOOLEAN('\0', "debug-single-step", &single_step,
136                         "Enable single stepping"),
137         OPT_BOOLEAN('\0', "debug-ioport-debug", &ioport_debug,
138                         "Enable ioport debugging"),
139         OPT_END()
140 };
141
142 /*
143  * Serialize debug printout so that the output of multiple vcpus does not
144  * get mixed up:
145  */
146 static int printout_done;
147
148 static void handle_sigusr1(int sig)
149 {
150         struct kvm_cpu *cpu = current_kvm_cpu;
151
152         if (!cpu)
153                 return;
154
155         printf("\n #\n # vCPU #%ld's dump:\n #\n", cpu->cpu_id);
156         kvm_cpu__show_registers(cpu);
157         kvm_cpu__show_code(cpu);
158         kvm_cpu__show_page_tables(cpu);
159         fflush(stdout);
160         printout_done = 1;
161         mb();
162 }
163
164 static void handle_sigquit(int sig)
165 {
166         int i;
167
168         for (i = 0; i < nrcpus; i++) {
169                 struct kvm_cpu *cpu = kvm_cpus[i];
170
171                 if (!cpu)
172                         continue;
173
174                 printout_done = 0;
175                 pthread_kill(cpu->thread, SIGUSR1);
176                 /*
177                  * Wait for the vCPU to dump state before signalling
178                  * the next thread. Since this is debug code it does
179                  * not matter that we are burning CPU time a bit:
180                  */
181                 while (!printout_done)
182                         mb();
183         }
184
185         serial8250__inject_sysrq(kvm);
186 }
187
188 static void handle_sigalrm(int sig)
189 {
190         serial8250__inject_interrupt(kvm);
191         virtio_console__inject_interrupt(kvm);
192 }
193
194 static void *kvm_cpu_thread(void *arg)
195 {
196         current_kvm_cpu         = arg;
197
198         if (kvm_cpu__start(current_kvm_cpu))
199                 goto panic_kvm;
200
201         kvm_cpu__delete(current_kvm_cpu);
202
203         return (void *) (intptr_t) 0;
204
205 panic_kvm:
206         fprintf(stderr, "KVM exit reason: %u (\"%s\")\n",
207                 current_kvm_cpu->kvm_run->exit_reason,
208                 kvm_exit_reasons[current_kvm_cpu->kvm_run->exit_reason]);
209         if (current_kvm_cpu->kvm_run->exit_reason == KVM_EXIT_UNKNOWN)
210                 fprintf(stderr, "KVM exit code: 0x%Lu\n",
211                         current_kvm_cpu->kvm_run->hw.hardware_exit_reason);
212
213         kvm_cpu__show_registers(current_kvm_cpu);
214         kvm_cpu__show_code(current_kvm_cpu);
215         kvm_cpu__show_page_tables(current_kvm_cpu);
216
217         kvm_cpu__delete(current_kvm_cpu);
218
219         return (void *) (intptr_t) 1;
220 }
221
222 static char kernel[PATH_MAX];
223
224 static const char *host_kernels[] = {
225         "/boot/vmlinuz",
226         "/boot/bzImage",
227         NULL
228 };
229
230 static const char *default_kernels[] = {
231         "./bzImage",
232         "../../arch/x86/boot/bzImage",
233         NULL
234 };
235
236 static const char *default_vmlinux[] = {
237         "../../../vmlinux",
238         "../../vmlinux",
239         NULL
240 };
241
242 static void kernel_usage_with_options(void)
243 {
244         const char **k;
245         struct utsname uts;
246
247         fprintf(stderr, "Fatal: could not find default kernel image in:\n");
248         k = &default_kernels[0];
249         while (*k) {
250                 fprintf(stderr, "\t%s\n", *k);
251                 k++;
252         }
253
254         if (uname(&uts) < 0)
255                 return;
256
257         k = &host_kernels[0];
258         while (*k) {
259                 if (snprintf(kernel, PATH_MAX, "%s-%s", *k, uts.release) < 0)
260                         return;
261                 fprintf(stderr, "\t%s\n", kernel);
262                 k++;
263         }
264         fprintf(stderr, "\nPlease see 'kvm run --help' for more options.\n\n");
265 }
266
267 static u64 host_ram_size(void)
268 {
269         long page_size;
270         long nr_pages;
271
272         nr_pages        = sysconf(_SC_PHYS_PAGES);
273         if (nr_pages < 0) {
274                 pr_warning("sysconf(_SC_PHYS_PAGES) failed");
275                 return 0;
276         }
277
278         page_size       = sysconf(_SC_PAGE_SIZE);
279         if (page_size < 0) {
280                 pr_warning("sysconf(_SC_PAGE_SIZE) failed");
281                 return 0;
282         }
283
284         return (nr_pages * page_size) >> MB_SHIFT;
285 }
286
287 /*
288  * If user didn't specify how much memory it wants to allocate for the guest,
289  * avoid filling the whole host RAM.
290  */
291 #define RAM_SIZE_RATIO          0.8
292
293 static u64 get_ram_size(int nr_cpus)
294 {
295         long available;
296         long ram_size;
297
298         ram_size        = 64 * (nr_cpus + 3);
299
300         available       = host_ram_size() * RAM_SIZE_RATIO;
301         if (!available)
302                 available = MIN_RAM_SIZE_MB;
303
304         if (ram_size > available)
305                 ram_size        = available;
306
307         return ram_size;
308 }
309
310 static const char *find_kernel(void)
311 {
312         const char **k;
313         struct stat st;
314         struct utsname uts;
315
316         k = &default_kernels[0];
317         while (*k) {
318                 if (stat(*k, &st) < 0 || !S_ISREG(st.st_mode)) {
319                         k++;
320                         continue;
321                 }
322                 strncpy(kernel, *k, PATH_MAX);
323                 return kernel;
324         }
325
326         if (uname(&uts) < 0)
327                 return NULL;
328
329         k = &host_kernels[0];
330         while (*k) {
331                 if (snprintf(kernel, PATH_MAX, "%s-%s", *k, uts.release) < 0)
332                         return NULL;
333
334                 if (stat(kernel, &st) < 0 || !S_ISREG(st.st_mode)) {
335                         k++;
336                         continue;
337                 }
338                 return kernel;
339
340         }
341         return NULL;
342 }
343
344 static const char *find_vmlinux(void)
345 {
346         const char **vmlinux;
347
348         vmlinux = &default_vmlinux[0];
349         while (*vmlinux) {
350                 struct stat st;
351
352                 if (stat(*vmlinux, &st) < 0 || !S_ISREG(st.st_mode)) {
353                         vmlinux++;
354                         continue;
355                 }
356                 return *vmlinux;
357         }
358         return NULL;
359 }
360
361 static int root_device(char *dev, long *part)
362 {
363         struct stat st;
364
365         if (stat("/", &st) < 0)
366                 return -1;
367
368         *part = minor(st.st_dev);
369
370         sprintf(dev, "/dev/block/%u:0", major(st.st_dev));
371         if (access(dev, R_OK) < 0)
372                 return -1;
373
374         return 0;
375 }
376
377 static char *host_image(char *cmd_line, size_t size)
378 {
379         char *t;
380         char device[PATH_MAX];
381         long part = 0;
382
383         t = malloc(PATH_MAX);
384         if (!t)
385                 return NULL;
386
387         /* check for the root file system */
388         if (root_device(device, &part) < 0) {
389                 free(t);
390                 return NULL;
391         }
392         strncpy(t, device, PATH_MAX);
393         if (!strstr(cmd_line, "root=")) {
394                 char tmp[PATH_MAX];
395                 snprintf(tmp, sizeof(tmp), "root=/dev/vda%ld rw ", part);
396                 strlcat(cmd_line, tmp, size);
397         }
398         return t;
399 }
400
401 void kvm_run_help(void)
402 {
403         usage_with_options(run_usage, options);
404 }
405
406 int kvm_cmd_run(int argc, const char **argv, const char *prefix)
407 {
408         struct virtio_net_parameters net_params;
409         static char real_cmdline[2048];
410         unsigned int nr_online_cpus;
411         int exit_code = 0;
412         int max_cpus;
413         char *hi;
414         int i;
415         void *ret;
416
417         signal(SIGALRM, handle_sigalrm);
418         signal(SIGQUIT, handle_sigquit);
419         signal(SIGUSR1, handle_sigusr1);
420
421         nr_online_cpus = sysconf(_SC_NPROCESSORS_ONLN);
422
423         while (argc != 0) {
424                 argc = parse_options(argc, argv, options, run_usage,
425                                 PARSE_OPT_STOP_AT_NON_OPTION);
426                 if (argc != 0) {
427                         if (kernel_filename) {
428                                 fprintf(stderr, "Cannot handle parameter: "
429                                                 "%s\n", argv[0]);
430                                 usage_with_options(run_usage, options);
431                                 return EINVAL;
432                         }
433                         /* first unhandled parameter is treated as a kernel
434                            image
435                          */
436                         kernel_filename = argv[0];
437                         argv++;
438                         argc--;
439                 }
440
441         }
442
443         if (!kernel_filename)
444                 kernel_filename = find_kernel();
445
446         if (!kernel_filename) {
447                 kernel_usage_with_options();
448                 return EINVAL;
449         }
450
451         vmlinux_filename = find_vmlinux();
452
453         if (nrcpus == 0)
454                 nrcpus = nr_online_cpus;
455         else if (nrcpus < 1 || nrcpus > KVM_NR_CPUS)
456                 die("Number of CPUs %d is out of [1;%d] range", nrcpus, KVM_NR_CPUS);
457
458         if (!ram_size)
459                 ram_size        = get_ram_size(nrcpus);
460
461         if (ram_size < MIN_RAM_SIZE_MB)
462                 die("Not enough memory specified: %lluMB (min %lluMB)", ram_size, MIN_RAM_SIZE_MB);
463
464         if (ram_size > host_ram_size())
465                 pr_warning("Guest memory size %lluMB exceeds host physical RAM size %lluMB", ram_size, host_ram_size());
466
467         ram_size <<= MB_SHIFT;
468
469         if (!kvm_dev)
470                 kvm_dev = DEFAULT_KVM_DEV;
471
472         if (!console)
473                 console = DEFAULT_CONSOLE;
474
475         if (!strncmp(console, "virtio", 6))
476                 active_console  = CONSOLE_VIRTIO;
477         else
478                 active_console  = CONSOLE_8250;
479
480         if (!host_ip_addr)
481                 host_ip_addr = DEFAULT_HOST_ADDR;
482
483         if (!guest_mac)
484                 guest_mac = DEFAULT_GUEST_MAC;
485
486         if (!script)
487                 script = DEFAULT_SCRIPT;
488
489         if (virtio_9p_dir) {
490                 char tmp[PATH_MAX];
491
492                 if (realpath(virtio_9p_dir, tmp))
493                         virtio_9p__init(kvm, tmp);
494                 else
495                         die("Failed resolving 9p path");
496         }
497
498         symbol__init(vmlinux_filename);
499
500         term_init();
501
502         kvm = kvm__init(kvm_dev, ram_size);
503
504         max_cpus = kvm__max_cpus(kvm);
505
506         if (nrcpus > max_cpus) {
507                 printf("  # Limit the number of CPUs to %d\n", max_cpus);
508                 kvm->nrcpus     = max_cpus;
509         }
510
511         kvm->nrcpus = nrcpus;
512
513         memset(real_cmdline, 0, sizeof(real_cmdline));
514         strcpy(real_cmdline, "notsc noapic noacpi pci=conf1 console=ttyS0 earlyprintk=serial");
515         strcat(real_cmdline, " ");
516         if (kernel_cmdline)
517                 strlcat(real_cmdline, kernel_cmdline, sizeof(real_cmdline));
518
519         hi = NULL;
520         if (!image_filename[0]) {
521                 hi = host_image(real_cmdline, sizeof(real_cmdline));
522                 if (hi) {
523                         image_filename[0] = hi;
524                         readonly_image[0] = true;
525                         image_count++;
526                 }
527         }
528
529         if (!strstr(real_cmdline, "root="))
530                 strlcat(real_cmdline, " root=/dev/vda rw ", sizeof(real_cmdline));
531
532         if (image_count) {
533                 kvm->disks = disk_image__open_all(image_filename, readonly_image, image_count);
534                 if (!kvm->disks)
535                         die("Unable to load all disk images.");
536
537                 for (i = 0; i < image_count; i++)
538                         virtio_blk__init(kvm, kvm->disks[i]);
539         }
540
541         free(hi);
542
543         printf("  # kvm run -k %s -m %Lu -c %d\n", kernel_filename, ram_size / 1024 / 1024, nrcpus);
544
545         if (!kvm__load_kernel(kvm, kernel_filename, initrd_filename,
546                                 real_cmdline))
547                 die("unable to load kernel %s", kernel_filename);
548
549         kvm->vmlinux            = vmlinux_filename;
550
551         ioport__setup_legacy();
552
553         rtc__init();
554
555         serial8250__init(kvm);
556
557         pci__init();
558
559         if (active_console == CONSOLE_VIRTIO)
560                 virtio_console__init(kvm);
561
562         if (virtio_rng)
563                 virtio_rng__init(kvm);
564
565         if (!network)
566                 network = DEFAULT_NETWORK;
567
568         if (!strncmp(network, "virtio", 6)) {
569                 net_params = (struct virtio_net_parameters) {
570                         .host_ip = host_ip_addr,
571                         .kvm = kvm,
572                         .script = script
573                 };
574                 sscanf(guest_mac,       "%hhx:%hhx:%hhx:%hhx:%hhx:%hhx",
575                                                         net_params.guest_mac,
576                                                         net_params.guest_mac+1,
577                                                         net_params.guest_mac+2,
578                                                         net_params.guest_mac+3,
579                                                         net_params.guest_mac+4,
580                                                         net_params.guest_mac+5);
581
582                 virtio_net__init(&net_params);
583         }
584
585         kvm__start_timer(kvm);
586
587         kvm__setup_bios(kvm);
588
589         for (i = 0; i < nrcpus; i++) {
590                 kvm_cpus[i] = kvm_cpu__init(kvm, i);
591                 if (!kvm_cpus[i])
592                         die("unable to initialize KVM VCPU");
593
594                 if (single_step)
595                         kvm_cpu__enable_singlestep(kvm_cpus[i]);
596         }
597
598         kvm__init_ram(kvm);
599
600         thread_pool__init(nr_online_cpus);
601
602         for (i = 0; i < nrcpus; i++) {
603                 if (pthread_create(&kvm_cpus[i]->thread, NULL, kvm_cpu_thread, kvm_cpus[i]) != 0)
604                         die("unable to create KVM VCPU thread");
605         }
606
607         /* Only VCPU #0 is going to exit by itself when shutting down */
608         if (pthread_join(kvm_cpus[0]->thread, &ret) != 0)
609                 exit_code = 1;
610
611         for (i = 1; i < nrcpus; i++) {
612                 pthread_kill(kvm_cpus[i]->thread, SIGKVMEXIT);
613                 if (pthread_join(kvm_cpus[i]->thread, &ret) != 0)
614                         die("pthread_join");
615
616                 if (ret != NULL)
617                         exit_code = 1;
618         }
619
620         kvm__delete(kvm);
621
622         if (!exit_code)
623                 printf("\n  # KVM session ended normally.\n");
624
625         return exit_code;
626 }