]> git.kernelconcepts.de Git - karo-tx-linux.git/blob - tools/kvm/kvm.c
kvm: Document the 'reset vector' hack
[karo-tx-linux.git] / tools / kvm / kvm.c
1 #include "kvm/kvm.h"
2
3 #include <linux/kvm.h>
4
5 #include <asm/bootparam.h>
6
7 #include <sys/ioctl.h>
8 #include <inttypes.h>
9 #include <sys/mman.h>
10 #include <stdbool.h>
11 #include <limits.h>
12 #include <stdarg.h>
13 #include <stdlib.h>
14 #include <string.h>
15 #include <unistd.h>
16 #include <stdio.h>
17 #include <fcntl.h>
18
19 #include "util.h"
20
21 /*
22  * Compatibility code. Remove this when we move to tools/kvm.
23  */
24 #ifndef KVM_EXIT_INTERNAL_ERROR
25 # define KVM_EXIT_INTERNAL_ERROR                17
26 #endif
27
28 #define DEFINE_KVM_EXIT_REASON(reason) [reason] = #reason
29
30 const char *kvm_exit_reasons[] = {
31         DEFINE_KVM_EXIT_REASON(KVM_EXIT_UNKNOWN),
32         DEFINE_KVM_EXIT_REASON(KVM_EXIT_EXCEPTION),
33         DEFINE_KVM_EXIT_REASON(KVM_EXIT_IO),
34         DEFINE_KVM_EXIT_REASON(KVM_EXIT_HYPERCALL),
35         DEFINE_KVM_EXIT_REASON(KVM_EXIT_DEBUG),
36         DEFINE_KVM_EXIT_REASON(KVM_EXIT_HLT),
37         DEFINE_KVM_EXIT_REASON(KVM_EXIT_MMIO),
38         DEFINE_KVM_EXIT_REASON(KVM_EXIT_IRQ_WINDOW_OPEN),
39         DEFINE_KVM_EXIT_REASON(KVM_EXIT_SHUTDOWN),
40         DEFINE_KVM_EXIT_REASON(KVM_EXIT_FAIL_ENTRY),
41         DEFINE_KVM_EXIT_REASON(KVM_EXIT_INTR),
42         DEFINE_KVM_EXIT_REASON(KVM_EXIT_SET_TPR),
43         DEFINE_KVM_EXIT_REASON(KVM_EXIT_TPR_ACCESS),
44         DEFINE_KVM_EXIT_REASON(KVM_EXIT_S390_SIEIC),
45         DEFINE_KVM_EXIT_REASON(KVM_EXIT_S390_RESET),
46         DEFINE_KVM_EXIT_REASON(KVM_EXIT_DCR),
47         DEFINE_KVM_EXIT_REASON(KVM_EXIT_NMI),
48         DEFINE_KVM_EXIT_REASON(KVM_EXIT_INTERNAL_ERROR),
49 };
50
51 static inline void *guest_addr_to_host(struct kvm *self, unsigned long offset)
52 {
53         return self->ram_start + offset;
54 }
55
56 static bool kvm__supports_extension(struct kvm *self, unsigned int extension)
57 {
58         int ret;
59
60         ret = ioctl(self->sys_fd, KVM_CHECK_EXTENSION, extension);
61         if (ret < 0)
62                 return false;
63
64         return ret;
65 }
66
67 static struct kvm *kvm__new(void)
68 {
69         struct kvm *self = calloc(1, sizeof *self);
70
71         if (!self)
72                 die("out of memory");
73
74         return self;
75 }
76
77 struct kvm *kvm__init(void)
78 {
79         struct kvm_userspace_memory_region mem;
80         struct kvm *self;
81         long page_size;
82         int mmap_size;
83         int ret;
84
85         self = kvm__new();
86
87         self->sys_fd = open("/dev/kvm", O_RDWR);
88         if (self->sys_fd < 0)
89                 die_perror("open");
90
91         ret = ioctl(self->sys_fd, KVM_GET_API_VERSION, 0);
92         if (ret != KVM_API_VERSION)
93                 die_perror("KVM_API_VERSION ioctl");
94
95         self->vm_fd = ioctl(self->sys_fd, KVM_CREATE_VM, 0);
96         if (self->vm_fd < 0)
97                 die_perror("KVM_CREATE_VM ioctl");
98
99         if (!kvm__supports_extension(self, KVM_CAP_USER_MEMORY))
100                 die("KVM_CAP_USER_MEMORY is not supported");
101
102         self->ram_size          = 64UL * 1024UL * 1024UL;
103
104         page_size       = sysconf(_SC_PAGESIZE);
105         if (posix_memalign(&self->ram_start, page_size, self->ram_size) != 0)
106                 die("out of memory");
107
108         mem = (struct kvm_userspace_memory_region) {
109                 .slot                   = 0,
110                 .guest_phys_addr        = 0x0UL,
111                 .memory_size            = self->ram_size,
112                 .userspace_addr         = (unsigned long) self->ram_start,
113         };
114
115         ret = ioctl(self->vm_fd, KVM_SET_USER_MEMORY_REGION, &mem, 1);
116         if (ret < 0)
117                 die_perror("KVM_SET_USER_MEMORY_REGION ioctl");
118
119         if (!kvm__supports_extension(self, KVM_CAP_SET_TSS_ADDR))
120                 die("KVM_CAP_SET_TSS_ADDR is not supported");
121
122         ret = ioctl(self->vm_fd, KVM_SET_TSS_ADDR, 0xfffbd000);
123         if (ret < 0)
124                 die_perror("KVM_SET_TSS_ADDR ioctl");
125
126         self->vcpu_fd = ioctl(self->vm_fd, KVM_CREATE_VCPU, 0);
127         if (self->vcpu_fd < 0)
128                 die_perror("KVM_CREATE_VCPU ioctl");
129
130         mmap_size = ioctl(self->sys_fd, KVM_GET_VCPU_MMAP_SIZE, 0);
131         if (mmap_size < 0)
132                 die_perror("KVM_GET_VCPU_MMAP_SIZE ioctl");
133
134         self->kvm_run = mmap(NULL, mmap_size, PROT_READ|PROT_WRITE, MAP_SHARED, self->vcpu_fd, 0);
135         if (self->kvm_run == MAP_FAILED)
136                 die("unable to mmap vcpu fd");
137
138         return self;
139 }
140
141 void kvm__enable_singlestep(struct kvm *self)
142 {
143         struct kvm_guest_debug debug = {
144                 .control        = KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP,
145         };
146
147         if (ioctl(self->vcpu_fd, KVM_SET_GUEST_DEBUG, &debug) < 0)
148                 warning("KVM_SET_GUEST_DEBUG failed");
149 }
150
151 static inline uint32_t segment_to_flat(uint16_t selector, uint16_t offset)
152 {
153         return ((uint32_t)selector << 4) + (uint32_t) offset;
154 }
155
156 /*
157  * HACK ALERT! KVM seems to be unable to run 16-bit real mode if 'cs' selector
158  * does not equal to 0xf000 at the beginning. To work around that, we need a
159  * 'reset vector' at 0xf000:0xfff0 that has an hard-coded jump to 0x000:0x7c000
160  * where we also load the Linux kernel boot sector and setup code.
161  */
162
163 #define RESET_VECTOR_CS         0xf000
164 #define RESET_VECTOR_IP         0xfff0
165
166 static unsigned char reset_vector_code[] = { 0xea, 0x00, 0x7c, 0x00, 0x00 };
167
168 static void load_reset_vector(struct kvm *self)
169 {
170         void *p;
171
172         p       = guest_addr_to_host(self, segment_to_flat(RESET_VECTOR_CS, RESET_VECTOR_IP));
173
174         memcpy(p, reset_vector_code, ARRAY_SIZE(reset_vector_code));
175 }
176
177 #define BOOT_LOADER_CS          0x0000
178 #define BOOT_LOADER_IP          0x7c00
179
180 static int load_flat_binary(struct kvm *kvm, int fd)
181 {
182         void *p;
183         int nr;
184
185         if (lseek(fd, 0, SEEK_SET) < 0)
186                 die_perror("lseek");
187
188         p = guest_addr_to_host(kvm, segment_to_flat(BOOT_LOADER_CS, BOOT_LOADER_IP));
189
190         while ((nr = read(fd, p, 65536)) > 0)
191                 p += nr;
192
193         return true;
194 }
195
196 /*
197  * The protected mode kernel part of a modern bzImage is loaded at 1 MB by
198  * default.
199  */
200 #define BZ_KERNEL_START                 0x100000UL
201
202 static const char *BZIMAGE_MAGIC        = "HdrS";
203
204 #define BZ_DEFAULT_SETUP_SECTS          4
205
206 static bool load_bzimage(struct kvm *kvm, int fd)
207 {
208         unsigned long setup_sects;
209         struct boot_params boot;
210         ssize_t setup_size;
211         void *p;
212         int nr;
213
214         /*
215          * See Documentation/x86/boot.txt for details no bzImage on-disk and
216          * memory layout.
217          */
218
219         if (lseek(fd, 0, SEEK_SET) < 0)
220                 die_perror("lseek");
221
222         read(fd, &boot, sizeof(boot));
223
224         if (memcmp(&boot.hdr.header, BZIMAGE_MAGIC, strlen(BZIMAGE_MAGIC)) != 0)
225                 return false;
226
227         if (lseek(fd, 0, SEEK_SET) < 0)
228                 die_perror("lseek");
229
230         setup_sects = boot.hdr.setup_sects + 1;
231         if (setup_sects == 0)
232                 setup_sects      = BZ_DEFAULT_SETUP_SECTS;
233
234         setup_size = setup_sects << 4;
235         p = guest_addr_to_host(kvm, segment_to_flat(BOOT_LOADER_CS, BOOT_LOADER_IP));
236
237         if (read(fd, p, setup_size) != setup_size)
238                 die_perror("read");
239
240         p = guest_addr_to_host(kvm, BZ_KERNEL_START);
241
242         while ((nr = read(fd, p, 65536)) > 0)
243                 p += nr;
244
245         return true;
246 }
247
248 bool kvm__load_kernel(struct kvm *kvm, const char *kernel_filename)
249 {
250         bool ret;
251         int fd;
252
253         fd = open(kernel_filename, O_RDONLY);
254         if (fd < 0)
255                 die("unable to open kernel");
256
257         ret = load_bzimage(kvm, fd);
258         if (ret)
259                 goto found_kernel;
260
261         ret = load_flat_binary(kvm, fd);
262         if (ret)
263                 goto found_kernel;
264
265         die("%s is not a valid bzImage or flat binary", kernel_filename);
266
267 found_kernel:
268         return ret;
269 }
270
271 static inline uint64_t ip_flat_to_real(struct kvm *self, uint64_t ip)
272 {
273         uint64_t cs = self->sregs.cs.selector;
274
275         return ip - (cs << 4);
276 }
277
278 static inline uint64_t ip_real_to_flat(struct kvm *self, uint64_t ip)
279 {
280         uint64_t cs = self->sregs.cs.selector;
281
282         return ip + (cs << 4);
283 }
284
285 void kvm__reset_vcpu(struct kvm *self)
286 {
287         load_reset_vector(self);
288
289         self->sregs = (struct kvm_sregs) {
290                 .cr0            = 0x60000010ULL,
291                 .cs             = (struct kvm_segment) {
292                         .selector       = RESET_VECTOR_CS,
293                         .base           = 0xffff0000UL,
294                         .limit          = 0xffffU,
295                         .type           = 0x0bU,
296                         .present        = 1,
297                         .dpl            = 0x03,
298                         .s              = 1,
299                 },
300                 .ss             = (struct kvm_segment) {
301                         .limit          = 0xffffU,
302                         .type           = 0x03U,
303                         .present        = 1,
304                         .dpl            = 0x03,
305                         .s              = 1,
306                 },
307                 .ds             = (struct kvm_segment) {
308                         .limit          = 0xffffU,
309                         .type           = 0x03U,
310                         .present        = 1,
311                         .dpl            = 0x03,
312                         .s              = 1,
313                 },
314                 .es             = (struct kvm_segment) {
315                         .limit          = 0xffffU,
316                         .type           = 0x03U,
317                         .present        = 1,
318                         .dpl            = 0x03,
319                         .s              = 1,
320                 },
321                 .fs             = (struct kvm_segment) {
322                         .limit          = 0xffffU,
323                         .type           = 0x03U,
324                         .present        = 1,
325                         .dpl            = 0x03,
326                         .s              = 1,
327                 },
328                 .gs             = (struct kvm_segment) {
329                         .limit          = 0xffffU,
330                         .type           = 0x03U,
331                         .present        = 1,
332                         .dpl            = 0x03,
333                         .s              = 1,
334                 },
335                 .tr             = (struct kvm_segment) {
336                         .limit          = 0xffffU,
337                         .present        = 1,
338                         .type           = 0x03U,
339                 },
340                 .ldt            = (struct kvm_segment) {
341                         .limit          = 0xffffU,
342                         .present        = 1,
343                         .type           = 0x02U,
344                 },
345                 .gdt            = (struct kvm_dtable) {
346                         .limit          = 0xffffU,
347                 },
348                 .idt            = (struct kvm_dtable) {
349                         .limit          = 0xffffU,
350                 },
351         };
352
353         if (ioctl(self->vcpu_fd, KVM_SET_SREGS, &self->sregs) < 0)
354                 die_perror("KVM_SET_SREGS failed");
355
356         self->regs = (struct kvm_regs) {
357                 .rip            = RESET_VECTOR_IP,
358                 /* We start the guest in 16-bit real mode  */
359                 .rflags         = 0x0000000000000002ULL,
360         };
361
362         if (self->regs.rip > USHRT_MAX)
363                 die("ip 0x%" PRIx64 " is too high for real mode", (uint64_t) self->regs.rip);
364
365         if (ioctl(self->vcpu_fd, KVM_SET_REGS, &self->regs) < 0)
366                 die_perror("KVM_SET_REGS failed");
367
368 }
369
370 void kvm__run(struct kvm *self)
371 {
372         if (ioctl(self->vcpu_fd, KVM_RUN, 0) < 0)
373                 die_perror("KVM_RUN failed");
374 }
375
376 static void kvm__emulate_io_out(struct kvm *self, uint16_t port, void *data, int size, uint32_t count)
377 {
378         fprintf(stderr, "%s port=%x, size=%d, count=%" PRIu32 "\n", __func__, port, size, count);
379 }
380
381 static void kvm__emulate_io_in(struct kvm *self, uint16_t port, void *data, int size, uint32_t count)
382 {
383         fprintf(stderr, "%s port=%x, size=%d, count=%" PRIu32 "\n", __func__, port, size, count);
384 }
385
386 void kvm__emulate_io(struct kvm *self, uint16_t port, void *data, int direction, int size, uint32_t count)
387 {
388         if (direction == KVM_EXIT_IO_IN)
389                 kvm__emulate_io_in(self, port, data, size, count);
390         else
391                 kvm__emulate_io_out(self, port, data, size, count);
392 }
393
394 static void print_segment(const char *name, struct kvm_segment *seg)
395 {
396         printf(" %s       %04" PRIx16 "      %016" PRIx64 "  %08" PRIx32 "  %02" PRIx8 "    %x %x   %x  %x %x %x %x\n",
397                 name, (uint16_t) seg->selector, (uint64_t) seg->base, (uint32_t) seg->limit,
398                 (uint8_t) seg->type, seg->present, seg->dpl, seg->db, seg->s, seg->l, seg->g, seg->avl);
399 }
400
401 void kvm__show_registers(struct kvm *self)
402 {
403         unsigned long cr0, cr2, cr3;
404         unsigned long cr4, cr8;
405         unsigned long rax, rbx, rcx;
406         unsigned long rdx, rsi, rdi;
407         unsigned long rbp,  r8,  r9;
408         unsigned long r10, r11, r12;
409         unsigned long r13, r14, r15;
410         unsigned long rip, rsp;
411         struct kvm_sregs sregs;
412         unsigned long rflags;
413         struct kvm_regs regs;
414         int i;
415
416         if (ioctl(self->vcpu_fd, KVM_GET_REGS, &regs) < 0)
417                 die("KVM_GET_REGS failed");
418
419         rflags = regs.rflags;
420
421         rip = regs.rip; rsp = regs.rsp;
422         rax = regs.rax; rbx = regs.rbx; rcx = regs.rcx;
423         rdx = regs.rdx; rsi = regs.rsi; rdi = regs.rdi;
424         rbp = regs.rbp; r8  = regs.r8;  r9  = regs.r9;
425         r10 = regs.r10; r11 = regs.r11; r12 = regs.r12;
426         r13 = regs.r13; r14 = regs.r14; r15 = regs.r15;
427
428         printf("Registers:\n");
429         printf(" rip: %016lx   rsp: %016lx flags: %016lx\n", rip, rsp, rflags);
430         printf(" rax: %016lx   ebx: %016lx   ecx: %016lx\n", rax, rbx, rcx);
431         printf(" rdx: %016lx   rsi: %016lx   rdi: %016lx\n", rdx, rsi, rdi);
432         printf(" rbp: %016lx   r8:  %016lx   r9:  %016lx\n", rbp, r8,  r9);
433         printf(" r10: %016lx   r11: %016lx   r12: %016lx\n", r10, r11, r12);
434         printf(" r13: %016lx   r14: %016lx   r15: %016lx\n", r13, r14, r15);
435
436         if (ioctl(self->vcpu_fd, KVM_GET_SREGS, &sregs) < 0)
437                 die("KVM_GET_REGS failed");
438
439         cr0 = sregs.cr0; cr2 = sregs.cr2; cr3 = sregs.cr3;
440         cr4 = sregs.cr4; cr8 = sregs.cr8;
441
442         printf(" cr0: %016lx   cr2: %016lx   cr3: %016lx\n", cr0, cr2, cr3);
443         printf(" cr4: %016lx   cr8: %016lx\n", cr4, cr8);
444         printf("Segment registers:\n");
445         printf(" register  selector  base              limit     type  p dpl db s l g avl\n");
446         print_segment("cs ", &sregs.cs);
447         print_segment("ss ", &sregs.ss);
448         print_segment("ds ", &sregs.ds);
449         print_segment("es ", &sregs.es);
450         print_segment("fs ", &sregs.fs);
451         print_segment("gs ", &sregs.gs);
452         print_segment("tr ", &sregs.tr);
453         print_segment("ldt", &sregs.ldt);
454         printf(" [ efer: %016lx  apic base: %016lx ]\n", (uint64_t) sregs.efer, (uint64_t) sregs.apic_base);
455         printf("Interrupt bitmap:\n");
456         printf(" ");
457         for (i = 0; i < (KVM_NR_INTERRUPTS + 63) / 64; i++)
458                 printf("%016lx ", (uint64_t) sregs.interrupt_bitmap[i]);
459         printf("\n");
460 }
461
462 void kvm__show_code(struct kvm *self)
463 {
464         unsigned int code_bytes = 64;
465         unsigned int code_prologue = code_bytes * 43 / 64;
466         unsigned int code_len = code_bytes;
467         unsigned char c;
468         unsigned int i;
469         uint8_t *ip;
470
471         if (ioctl(self->vcpu_fd, KVM_GET_REGS, &self->regs) < 0)
472                 die("KVM_GET_REGS failed");
473
474         if (ioctl(self->vcpu_fd, KVM_GET_SREGS, &self->sregs) < 0)
475                 die("KVM_GET_SREGS failed");
476
477         ip = guest_addr_to_host(self, ip_real_to_flat(self, self->regs.rip) - code_prologue);
478
479         printf("Code: ");
480
481         for (i = 0; i < code_len; i++, ip++) {
482                 c = *ip;
483
484                 if (ip == guest_addr_to_host(self, ip_real_to_flat(self, self->regs.rip)))
485                         printf("<%02x> ", c);
486                 else
487                         printf("%02x ", c);
488         }
489
490         printf("\n");
491 }