SlideShare a Scribd company logo

Linux KVM のコードを追いかけてみよう

•
•
Tsuyoshi OZAWA
Tsuyoshi OZAWA

The document discusses Linux KVM (Kernel-based Virtual Machine) and how it enables full virtualization on x86 hardware. KVM uses Intel VT-x and AMD-V virtualization extensions to allow a Linux kernel to function as a hypervisor. Guest virtual machines see a bare metal interface while the host kernel manages scheduling and resource allocation. Qemu is used as a processor emulator to add missing guest architectures.

1 of 260
Downloaded 334 times
Linux KVM /VM 1 by Tsuyoshi Ozawa, softlab@univ. of Tsukuba. Twitter: oza_x86 2009 8 13 1
2009 8 13 2
1. VMM 2. Intel-VT 3. Linux KVM KVM 2009 8 13 3
VMM 2009 8 13 4
VMM 2009 8 13 5
Virtual Machine Monitor 2009 8 13 6
Virtual Machine Monitor 1. 2. (OS) 2009 8 13 7
VMM 2009 8 13 8
2009 8 13 9
2009 8 13 10
Qemu 2009 8 13 11
2009 8 13 12
2009 8 13 13
Qemu 2009 8 13 14
2009 8 13 15
CPU 2009 8 13 16
Main Memory ... ALU movl %eax %ebx ... 2009 8 13 17
Main Memory ... ALU movl %eax %ebx ... fetch 2009 8 13 18
Qemu ... 2009 8 13 19
Qemu Main Memory ... ALU movl %eax %ebx ... fetch 2009 8 13 20
opcode = mem[pc]; execute(opcode); 2009 8 13 21
mov opcode = mem[pc]; execute(opcode); jmp 2009 8 13 22
• 2009 8 13 23
2009 8 13 24
opcode[0..N] = mem[pc]; mid_ops=compile(opcode); execute(mid_ops); 2009 8 13 25
opcode[0..N] = mem[pc]; mid_ops=compile(opcode); execute(mid_ops); 2009 8 13 26
opcode[0..N] = mem[pc]; mid_ops=compile(opcode); execute(mid_ops); 2009 8 13 27
2009 8 13 28
2009 8 13 29
• • • 2009 8 13 30
2009 8 13 31
2009 8 13 32
VMM 2009 8 13 33
KVM 2009 8 13 34
• 2009 8 13 35
.section .data message: .ascii "hello,gas!n" .section .text .global _start _start: movl $4,%eax movl $1,%ebx movl $message,%ecx movl $12,%edx int $0x80 2009 8 13 36
./a.out hello,gas! 2009 8 13 37
Hello world 2009 8 13 38
printf(“HelloWorld¥n”) 2009 8 13 39
printf 2009 8 13 40
2009 8 13 41
2009 8 13 42
write 2009 8 13 43
write out 2009 8 13 44
.section .data message: .ascii "hello,gas!n" .section .text .global _start _start: movl $4,%eax movl $1,%ebx movl $message,%ecx out movl $12,%edx int $0x80 2009 8 13 45
... outb %ax ... 2009 8 13 46
Program Display CPU outb 2009 8 13 47
hello,gas! 2009 8 13 48
• 2009 8 13 49
• VM 2009 8 13 50
.section .data message: .ascii "hello,gas!n" .section .text .global _start _start: movl $4,%eax movl $1,%ebx movl $message,%ecx out movl $12,%edx int $0x80 2009 8 13 51
... outb $hogehoge ... 2009 8 13 52
hello,gas! 2009 8 13 53
OS . 2009 8 13 54
VM Virtual Program Display VirtualCPU outb 2009 8 13 55
VM Program Real Display VirtualCPU outb 2009 8 13 56
hello,gas! on host system?! 2009 8 13 57
2009 8 13 58
2009 8 13 59
2009 8 13 60
2009 8 13 61
VM out Virtual Program Display CPU 2009 8 13 62
VM out Virtual Program Display CPU outb 2009 8 13 63
VM out 2009 8 13 64
Emu Code trap CPU 2009 8 13 65
Emu Code CPU handle 2009 8 13 66
VM out Emu ret Code CPU 2009 8 13 67
VM out ack Virtual Program Display CPU 2009 8 13 68
VM out out 2009 8 13 69
2009 8 13 70
2009 8 13 71
x86 2009 8 13 72
? 2009 8 13 73
2009 8 13 74
2009 8 13 75
? 2009 8 13 76
2009 8 13 77
x86 ? 2009 8 13 78
... ? 2009 8 13 79
2009 8 13 80
2009 8 13 81
x86 • • Hypervisor Call • Xen 2009 8 13 82
x86 • • LilyVM 2009 8 13 83
x86 • • (JIT ) • VMWare 2009 8 13 84
2009 8 13 85
Intel ( AMD-V ) 2009 8 13 86
Intel-VT 2009 8 13 87
CPU ? 2009 8 13 88
User Privilege 2009 8 13 89
User Privilege 2009 8 13 90
2009 8 13 91
User User warp Privilege Privilege 2009 8 13 92
2009 8 13 93
User User Privilege Privilege 2009 8 13 94
User User Privilege Privilege 2009 8 13 95
User User Privilege Privilege trap 2009 8 13 96
User User Privilege Privilege Intel-VT 2009 8 13 97
User Privilege 2009 8 13 98
VMX-Root Mode (VM ) 2009 8 13 99
User Privilege 2009 8 13 100
VMX-non root Mode ( ) 2009 8 13 101
VMX non Root Mode 2009 8 13 102
User User Privilege Privilege 2009 8 13 103
vmlaunch vmresume 2009 8 13 104
KVM 2009 8 13 105
KVM 2009 8 13 106
Kernel-based Virtul Machine 2009 8 13 107
KVM 2009 8 13 108
KVM • •Intel-VT • •Qemu • •Linux • •Linux 2009 8 13 109
KVM • •Intel-VT • •Qemu • •Linux • •Linux KVM 2009 8 13 110
KVM • •VM • • • • OS 2009 8 13 111
2009 8 13 112
2009 8 13 113
Main loop 2009 8 13 114
KVM 2009 8 13 115
User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 116
2009 8 13 117
VMX VMX non Root Root Mode Mode 2009 8 13 118
2009 8 13 119
User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 120
VMX non Root Mode 2009 8 13 121
User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 122
User Qemu KVM Program Qemu Linux Guest KVM Kernel Kernel 2009 8 13 123
./qemu -hda disk1.img (-hda ) 2009 8 13 124
User Qemu Program $ ./qemu -hda disk1 Linux Guest KVM Kernel Kernel 2009 8 13 125
User Qemu Program Qemu Linux Guest KVM Kernel Kernel 2009 8 13 126
Qemu 2009 8 13 127
User Qemu Program IOCTL Linux Guest KVM Kernel Kernel 2009 8 13 128
User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 129
2009 8 13 130
2009 8 13 131
vmlaunch vmresume 2009 8 13 132
VMX non Root Mode 2009 8 13 133
User Qemu Program Linux Guest KVM VT Kernel Kernel 2009 8 13 134
2009 8 13 135
vmlaunch vmresume 2009 8 13 136
User Qemu Program vmlaunch / vmresume Linux Guest KVM Kernel Kernel 2009 8 13 137
User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 138
VMX non root Mode 2009 8 13 139
User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 140
2009 8 13 141
VMX non Root Mode IO 2009 8 13 142
2009 8 13 143
outb %ax 2009 8 13 144
User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 145
User Qemu Program Linux Guest KVM Kernel Kernel outb %ax 2009 8 13 146
VMX Root Mode 2009 8 13 147
User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 148
User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 149
2009 8 13 150
Qemu 2009 8 13 151
User Qemu Program IOCTL return Linux Guest KVM Kernel Kernel 2009 8 13 152
User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 153
User Qemu Program io Linux Guest KVM Kernel Kernel 2009 8 13 154
User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 155
User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 156
2009 8 13 157
IOCTL CPU 2009 8 13 158
User Qemu Program IOCTL Linux Guest KVM Kernel Kernel 2009 8 13 159
2009 8 13 160
... 2009 8 13 161
User Qemu Program IOCTL Linux Guest KVM Kernel Kernel vmlaunch vmresume 2009 8 13 162
2009 8 13 163
2009 8 13 164
KVM kvm-85 2009 8 13 165
2009 8 13 166
User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 167
Qemu 2009 8 13 168
$./qemu -hda disk1 ... 2009 8 13 169
qemu/vl.c main.c 2009 8 13 170
4439 int 4440 { main (int argc, char **argv, char **envp) // . Shadow page table . 5425 kvm_qemu_create_context() // 5493 kvm_init(smp_cpus); ... // qemu/hw/pc.c 5554 machine->init(ram_size, vga_ram_size, boot_devices, 5555 kernel_filename, kernel_cmdline, 5556 initrd_filename, cpu_model); ... ... 5742 main_loop(); ... 5747 } 2009 8 13 171
User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 172
User Qemu Program Qemu Linux Guest KVM Kernel Kernel 2009 8 13 173
- •machine->init • • qemu/hw/pc.c pc_init1 2009 8 13 174
2009 8 13 175
2009 8 13 176
pc_init1 pc_new_cpu cpu_init = cpu_x86_init kvm_init_cpu pthread_create ap_main_loop kvm_main_loop_cpu 2009 8 13 177
2009 8 13 178
2009 8 13 179
User Qemu Program IOCTL Linux Guest KVM Kernel Kernel 2009 8 13 180
IOCTL 2009 8 13 181
... qemu/kvm-all.c kvm_cpu_exec 2009 8 13 182
378 static int kvm_main_loop_cpu (CPUState *env) 379 { 394 while (1) { 395 while (!has_work(env)) 396 kvm_main_loop_wait(env, 1000); 397 if (env->interrupt_request & (CPU_INTERRUPT_HARD | CPU_INTERRUPT_NMI)) 398 env->halted = 0; // . .... 404 } 405 if (!env->halted && !env->kvm_cpu_state.init) { 406 kvm_cpu_exec(env); // } } 413 } 2009 8 13 183
int kvm_cpu_exec(){ ... 461 ret = kvm_vcpu_ioctl(env, KVM_RUN, 0); 478 switch (run->exit_reason) { 479 case KVM_EXIT_IO: 481 ret = kvm_handle_io(env, run->io.port, 482 (uint8_t *)run + run>io.data_offset, 483 run->io.direction, 484 run->io.size, 485 run->io.count); 486 break; case KVM_EXIT_MMIO: .... ... 2009 8 13 184
648 int kvm_vcpu_ioctl(CPUState *env, int type, ...) 649 { 658 ret = ioctl(env->kvm_fd, type, arg); 659 if (ret == -1) 660 ret = -errno; 661 662 return ret; 663 } 2009 8 13 185
2009 8 13 186
IOCTL 2009 8 13 187
User Qemu Program IOCTL Linux Guest KVM Kernel Kernel 2009 8 13 188
User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 189
2009 8 13 190
2009 8 13 191
IOCTL 2009 8 13 192
IOCTL 2009 8 13 193
2009 8 13 194
static struct file_operations kvm_vcpu_fops = { .release = kvm_vcpu_release, .unlocked_ioctl = kvm_vcpu_ioctl, .compat_ioctl = kvm_vcpu_ioctl, .mmap = kvm_vcpu_mmap, }; 2009 8 13 195
// file->f_op = kvm_vcpu_fops; 2009 8 13 196
kvm_vcpu_ioctl 2009 8 13 197
static long kvm_vcpu_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg) { switch (ioctl) { case KVM_RUN: r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run); break; .... } 2009 8 13 198
VMX non Root Mode 2009 8 13 199
User Qemu Program Linux Guest KVM VT Kernel Kernel 2009 8 13 200
... 2009 8 13 201
kvm_arch_vcpu_ioctl_run __vcpu_run vcpu_enter_guest vmx_vcpu_run 2009 8 13 202
vmx_vcpu_run kernel/x86/vmx.c 2009 8 13 203
User Qemu Program vmlaunch / vmresume Linux Guest KVM Kernel Kernel 2009 8 13 204
2009 8 13 205
2009 8 13 206
include/asm-x86/vmx.h #define ASM_VMX_VMLAUNCH ".byte 0x0f, 0x01, 0xc2" #define ASM_VMX_VMRESUME ".byte 0x0f, 0x01, 0xc3" 2009 8 13 207
2009 8 13 208
vmlaunch/vmresume 2009 8 13 209
User Qemu Program vmlaunch / vmresume Linux Guest KVM Kernel Kernel 2009 8 13 210
User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 211
VMX non-root mode 2009 8 13 212
2009 8 13 213
VMX non Root Mode IO 2009 8 13 214
2009 8 13 215
2009 8 13 216
outb %ax 2009 8 13 217
User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 218
User Qemu Program Linux Guest KVM Kernel Kernel outb %ax 2009 8 13 219
2009 8 13 220
"jne .Llaunched nt" __ex(ASM_VMX_VMLAUNCH) "nt" "jmp .Lkvm_vmx_return nt" ".Llaunched: " __ex(ASM_VMX_VMRESUME) "nt" ".Lkvm_vmx_return: " vmlaunch /* */ vmresume ... ... 2009 8 13 221
"jne .Llaunched nt" __ex(ASM_VMX_VMLAUNCH) "nt" "jmp .Lkvm_vmx_return nt" ".Llaunched: " __ex(ASM_VMX_VMRESUME) "nt" ".Lkvm_vmx_return: " /* */ ... 2009 8 13 222
2009 8 13 223
2009 8 13 224
out 2009 8 13 225
out in . 2009 8 13 226
Intel-VT 2009 8 13 227
User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 228
2009 8 13 229
Qemu 2009 8 13 230
User Qemu Program IOCTL return Linux Guest KVM Kernel Kernel 2009 8 13 231
User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 232
int kvm_cpu_exec(){ ... 461 ret = kvm_vcpu_ioctl(env, KVM_RUN, 0); 478 switch (run->exit_reason) { 479 case KVM_EXIT_IO: 481 ret = kvm_handle_io(env, run->io.port, 482 (uint8_t *)run + run>io.data_offset, 483 run->io.direction, 484 run->io.size, 485 run->io.count); 486 break; case KVM_EXIT_MMIO: .... ... 2009 8 13 233
outb 2009 8 13 234
int kvm_cpu_exec(){ ... // 461 ret = kvm_vcpu_ioctl(env, KVM_RUN, 0); 478 switch (run->exit_reason) { 479 case KVM_EXIT_IO: 481 ret = kvm_handle_io(env, run->io.port, 482 (uint8_t *)run + run>io.data_offset, 483 run->io.direction, 484 run->io.size, 485 run->io.count); 486 break; case KVM_EXIT_MMIO: outb .... ... 2009 8 13 235
2009 8 13 236
IO 2009 8 13 237
User Qemu Program io Linux Guest KVM Kernel Kernel 2009 8 13 238
2009 8 13 239
out 2009 8 13 240
outb %al 2009 8 13 241
outb %al 2009 8 13 242
( ) 2009 8 13 243
int kvm_cpu_exec(){ ... // 461 ret = kvm_vcpu_ioctl(env, KVM_RUN, 0); 478 switch (run->exit_reason) { 479 case KVM_EXIT_IO: 481 ret = kvm_handle_io(env, run->io.port, 482 (uint8_t *)run + run>io.data_offset, 483 run->io.direction, 484 run->io.size, 485 run->io.count); 486 break; case KVM_EXIT_MMIO: .... ... 2009 8 13 244
kvm_handle_io cpu_outb ioport_write 2009 8 13 245
static void ioport_write(int index, uint32_t address, uint32_t data) { static IOPortWriteFunc *default_func[3] = { default_ioport_writeb, default_ioport_writew, default_ioport_writel }; IOPortWriteFunc *func = ioport_write_table[index][address]; if (!func) func = default_func[index]; func(ioport_opaque[address], address, data); } ioport_write_table 2009 8 13 246
( ) 2009 8 13 247
User Qemu Program io Linux Guest KVM Kernel Kernel 2009 8 13 248
User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 249
User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 250
2009 8 13 251
IOCTL CPU 2009 8 13 252
User Qemu Program IOCTL Linux Guest KVM Kernel Kernel 2009 8 13 253
2009 8 13 254
... 2009 8 13 255
User Qemu Program IOCTL Linux Guest KVM Kernel Kernel vmlaunch 2009 8 13 vmresume 256
2009 8 13 257
378 static int kvm_main_loop_cpu(CPUState *env) 379 { 394 while (1) { 395 while (!has_work(env)) 396 kvm_main_loop_wait(env, 1000); 397 if (env->interrupt_request & (CPU_INTERRUPT_HARD | CPU_INTERRUPT_NMI)) 398 env->halted = 0; // . .... 404 } 405 if (!env->halted && !env->kvm_cpu_state.init) { 406 kvm_cpu_exec(env); // } } 413 } 2009 8 13 258
2009 8 13 259
Twitter 2009 8 13 260

More Related Content

Linux KVM のコードを追いかけてみよう

  • 1. Linux KVM /VM 1 by Tsuyoshi Ozawa, softlab@univ. of Tsukuba. Twitter: oza_x86 2009 8 13 1
  • 2. 2009 8 13 2
  • 3. 1. VMM 2. Intel-VT 3. Linux KVM KVM 2009 8 13 3
  • 4. VMM 2009 8 13 4
  • 5. VMM 2009 8 13 5
  • 6. Virtual Machine Monitor 2009 8 13 6
  • 7. Virtual Machine Monitor 1. 2. (OS) 2009 8 13 7
  • 8. VMM 2009 8 13 8
  • 9. 2009 8 13 9
  • 10. 2009 8 13 10
  • 11. Qemu 2009 8 13 11
  • 12. 2009 8 13 12
  • 13. 2009 8 13 13
  • 14. Qemu 2009 8 13 14
  • 15. 2009 8 13 15
  • 16. CPU 2009 8 13 16
  • 17. Main Memory ... ALU movl %eax %ebx ... 2009 8 13 17
  • 18. Main Memory ... ALU movl %eax %ebx ... fetch 2009 8 13 18
  • 19. Qemu ... 2009 8 13 19
  • 20. Qemu Main Memory ... ALU movl %eax %ebx ... fetch 2009 8 13 20
  • 21. opcode = mem[pc]; execute(opcode); 2009 8 13 21
  • 22. mov opcode = mem[pc]; execute(opcode); jmp 2009 8 13 22
  • 23. • 2009 8 13 23
  • 24. 2009 8 13 24
  • 25. opcode[0..N] = mem[pc]; mid_ops=compile(opcode); execute(mid_ops); 2009 8 13 25
  • 26. opcode[0..N] = mem[pc]; mid_ops=compile(opcode); execute(mid_ops); 2009 8 13 26
  • 27. opcode[0..N] = mem[pc]; mid_ops=compile(opcode); execute(mid_ops); 2009 8 13 27
  • 28. 2009 8 13 28
  • 29. 2009 8 13 29
  • 30. • • • 2009 8 13 30
  • 31. 2009 8 13 31
  • 32. 2009 8 13 32
  • 33. VMM 2009 8 13 33
  • 34. KVM 2009 8 13 34
  • 35. • 2009 8 13 35
  • 36. .section .data message: .ascii "hello,gas!n" .section .text .global _start _start: movl $4,%eax movl $1,%ebx movl $message,%ecx movl $12,%edx int $0x80 2009 8 13 36
  • 37. ./a.out hello,gas! 2009 8 13 37
  • 38. Hello world 2009 8 13 38
  • 40. printf 2009 8 13 40
  • 41. 2009 8 13 41
  • 42. 2009 8 13 42
  • 43. write 2009 8 13 43
  • 44. write out 2009 8 13 44
  • 45. .section .data message: .ascii "hello,gas!n" .section .text .global _start _start: movl $4,%eax movl $1,%ebx movl $message,%ecx out movl $12,%edx int $0x80 2009 8 13 45
  • 46. ... outb %ax ... 2009 8 13 46
  • 47. Program Display CPU outb 2009 8 13 47
  • 48. hello,gas! 2009 8 13 48
  • 49. • 2009 8 13 49
  • 50. • VM 2009 8 13 50
  • 51. .section .data message: .ascii "hello,gas!n" .section .text .global _start _start: movl $4,%eax movl $1,%ebx movl $message,%ecx out movl $12,%edx int $0x80 2009 8 13 51
  • 52. ... outb $hogehoge ... 2009 8 13 52
  • 53. hello,gas! 2009 8 13 53
  • 54. OS . 2009 8 13 54
  • 55. VM Virtual Program Display VirtualCPU outb 2009 8 13 55
  • 56. VM Program Real Display VirtualCPU outb 2009 8 13 56
  • 57. hello,gas! on host system?! 2009 8 13 57
  • 58. 2009 8 13 58
  • 59. 2009 8 13 59
  • 60. 2009 8 13 60
  • 61. 2009 8 13 61
  • 62. VM out Virtual Program Display CPU 2009 8 13 62
  • 63. VM out Virtual Program Display CPU outb 2009 8 13 63
  • 64. VM out 2009 8 13 64
  • 65. Emu Code trap CPU 2009 8 13 65
  • 66. Emu Code CPU handle 2009 8 13 66
  • 67. VM out Emu ret Code CPU 2009 8 13 67
  • 68. VM out ack Virtual Program Display CPU 2009 8 13 68
  • 69. VM out out 2009 8 13 69
  • 70. 2009 8 13 70
  • 71. 2009 8 13 71
  • 72. x86 2009 8 13 72
  • 73. ? 2009 8 13 73
  • 74. 2009 8 13 74
  • 75. 2009 8 13 75
  • 76. ? 2009 8 13 76
  • 77. 2009 8 13 77
  • 78. x86 ? 2009 8 13 78
  • 79. ... ? 2009 8 13 79
  • 80. 2009 8 13 80
  • 81. 2009 8 13 81
  • 82. x86 • • Hypervisor Call • Xen 2009 8 13 82
  • 83. x86 • • LilyVM 2009 8 13 83
  • 84. x86 • • (JIT ) • VMWare 2009 8 13 84
  • 85. 2009 8 13 85
  • 86. Intel ( AMD-V ) 2009 8 13 86
  • 87. Intel-VT 2009 8 13 87
  • 88. CPU ? 2009 8 13 88
  • 89. User Privilege 2009 8 13 89
  • 90. User Privilege 2009 8 13 90
  • 91. 2009 8 13 91
  • 92. User User warp Privilege Privilege 2009 8 13 92
  • 93. 2009 8 13 93
  • 94. User User Privilege Privilege 2009 8 13 94
  • 95. User User Privilege Privilege 2009 8 13 95
  • 96. User User Privilege Privilege trap 2009 8 13 96
  • 97. User User Privilege Privilege Intel-VT 2009 8 13 97
  • 98. User Privilege 2009 8 13 98
  • 99. VMX-Root Mode (VM ) 2009 8 13 99
  • 100. User Privilege 2009 8 13 100
  • 101. VMX-non root Mode ( ) 2009 8 13 101
  • 102. VMX non Root Mode 2009 8 13 102
  • 103. User User Privilege Privilege 2009 8 13 103
  • 104. vmlaunch vmresume 2009 8 13 104
  • 105. KVM 2009 8 13 105
  • 106. KVM 2009 8 13 106
  • 107. Kernel-based Virtul Machine 2009 8 13 107
  • 108. KVM 2009 8 13 108
  • 109. KVM • •Intel-VT • •Qemu • •Linux • •Linux 2009 8 13 109
  • 110. KVM • •Intel-VT • •Qemu • •Linux • •Linux KVM 2009 8 13 110
  • 111. KVM • •VM • • • • OS 2009 8 13 111
  • 112. 2009 8 13 112
  • 113. 2009 8 13 113
  • 114. Main loop 2009 8 13 114
  • 115. KVM 2009 8 13 115
  • 116. User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 116
  • 117. 2009 8 13 117
  • 118. VMX VMX non Root Root Mode Mode 2009 8 13 118
  • 119. 2009 8 13 119
  • 120. User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 120
  • 121. VMX non Root Mode 2009 8 13 121
  • 122. User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 122
  • 123. User Qemu KVM Program Qemu Linux Guest KVM Kernel Kernel 2009 8 13 123
  • 124. ./qemu -hda disk1.img (-hda ) 2009 8 13 124
  • 125. User Qemu Program $ ./qemu -hda disk1 Linux Guest KVM Kernel Kernel 2009 8 13 125
  • 126. User Qemu Program Qemu Linux Guest KVM Kernel Kernel 2009 8 13 126
  • 127. Qemu 2009 8 13 127
  • 128. User Qemu Program IOCTL Linux Guest KVM Kernel Kernel 2009 8 13 128
  • 129. User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 129
  • 130. 2009 8 13 130
  • 131. 2009 8 13 131
  • 132. vmlaunch vmresume 2009 8 13 132
  • 133. VMX non Root Mode 2009 8 13 133
  • 134. User Qemu Program Linux Guest KVM VT Kernel Kernel 2009 8 13 134
  • 135. 2009 8 13 135
  • 136. vmlaunch vmresume 2009 8 13 136
  • 137. User Qemu Program vmlaunch / vmresume Linux Guest KVM Kernel Kernel 2009 8 13 137
  • 138. User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 138
  • 139. VMX non root Mode 2009 8 13 139
  • 140. User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 140
  • 141. 2009 8 13 141
  • 142. VMX non Root Mode IO 2009 8 13 142
  • 143. 2009 8 13 143
  • 144. outb %ax 2009 8 13 144
  • 145. User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 145
  • 146. User Qemu Program Linux Guest KVM Kernel Kernel outb %ax 2009 8 13 146
  • 147. VMX Root Mode 2009 8 13 147
  • 148. User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 148
  • 149. User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 149
  • 150. 2009 8 13 150
  • 151. Qemu 2009 8 13 151
  • 152. User Qemu Program IOCTL return Linux Guest KVM Kernel Kernel 2009 8 13 152
  • 153. User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 153
  • 154. User Qemu Program io Linux Guest KVM Kernel Kernel 2009 8 13 154
  • 155. User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 155
  • 156. User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 156
  • 157. 2009 8 13 157
  • 158. IOCTL CPU 2009 8 13 158
  • 159. User Qemu Program IOCTL Linux Guest KVM Kernel Kernel 2009 8 13 159
  • 160. 2009 8 13 160
  • 161. ... 2009 8 13 161
  • 162. User Qemu Program IOCTL Linux Guest KVM Kernel Kernel vmlaunch vmresume 2009 8 13 162
  • 163. 2009 8 13 163
  • 164. 2009 8 13 164
  • 165. KVM kvm-85 2009 8 13 165
  • 166. 2009 8 13 166
  • 167. User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 167
  • 168. Qemu 2009 8 13 168
  • 169. $./qemu -hda disk1 ... 2009 8 13 169
  • 171. 4439 int 4440 { main (int argc, char **argv, char **envp) // . Shadow page table . 5425 kvm_qemu_create_context() // 5493 kvm_init(smp_cpus); ... // qemu/hw/pc.c 5554 machine->init(ram_size, vga_ram_size, boot_devices, 5555 kernel_filename, kernel_cmdline, 5556 initrd_filename, cpu_model); ... ... 5742 main_loop(); ... 5747 } 2009 8 13 171
  • 172. User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 172
  • 173. User Qemu Program Qemu Linux Guest KVM Kernel Kernel 2009 8 13 173
  • 174. - •machine->init • • qemu/hw/pc.c pc_init1 2009 8 13 174
  • 175. 2009 8 13 175
  • 176. 2009 8 13 176
  • 177. pc_init1 pc_new_cpu cpu_init = cpu_x86_init kvm_init_cpu pthread_create ap_main_loop kvm_main_loop_cpu 2009 8 13 177
  • 178. 2009 8 13 178
  • 179. 2009 8 13 179
  • 180. User Qemu Program IOCTL Linux Guest KVM Kernel Kernel 2009 8 13 180
  • 181. IOCTL 2009 8 13 181
  • 182. ... qemu/kvm-all.c kvm_cpu_exec 2009 8 13 182
  • 183. 378 static int kvm_main_loop_cpu (CPUState *env) 379 { 394 while (1) { 395 while (!has_work(env)) 396 kvm_main_loop_wait(env, 1000); 397 if (env->interrupt_request & (CPU_INTERRUPT_HARD | CPU_INTERRUPT_NMI)) 398 env->halted = 0; // . .... 404 } 405 if (!env->halted && !env->kvm_cpu_state.init) { 406 kvm_cpu_exec(env); // } } 413 } 2009 8 13 183
  • 184. int kvm_cpu_exec(){ ... 461 ret = kvm_vcpu_ioctl(env, KVM_RUN, 0); 478 switch (run->exit_reason) { 479 case KVM_EXIT_IO: 481 ret = kvm_handle_io(env, run->io.port, 482 (uint8_t *)run + run>io.data_offset, 483 run->io.direction, 484 run->io.size, 485 run->io.count); 486 break; case KVM_EXIT_MMIO: .... ... 2009 8 13 184
  • 185. 648 int kvm_vcpu_ioctl(CPUState *env, int type, ...) 649 { 658 ret = ioctl(env->kvm_fd, type, arg); 659 if (ret == -1) 660 ret = -errno; 661 662 return ret; 663 } 2009 8 13 185
  • 186. 2009 8 13 186
  • 187. IOCTL 2009 8 13 187
  • 188. User Qemu Program IOCTL Linux Guest KVM Kernel Kernel 2009 8 13 188
  • 189. User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 189
  • 190. 2009 8 13 190
  • 191. 2009 8 13 191
  • 192. IOCTL 2009 8 13 192
  • 193. IOCTL 2009 8 13 193
  • 194. 2009 8 13 194
  • 195. static struct file_operations kvm_vcpu_fops = { .release = kvm_vcpu_release, .unlocked_ioctl = kvm_vcpu_ioctl, .compat_ioctl = kvm_vcpu_ioctl, .mmap = kvm_vcpu_mmap, }; 2009 8 13 195
  • 196. // file->f_op = kvm_vcpu_fops; 2009 8 13 196
  • 197. kvm_vcpu_ioctl 2009 8 13 197
  • 198. static long kvm_vcpu_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg) { switch (ioctl) { case KVM_RUN: r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run); break; .... } 2009 8 13 198
  • 199. VMX non Root Mode 2009 8 13 199
  • 200. User Qemu Program Linux Guest KVM VT Kernel Kernel 2009 8 13 200
  • 201. ... 2009 8 13 201
  • 202. kvm_arch_vcpu_ioctl_run __vcpu_run vcpu_enter_guest vmx_vcpu_run 2009 8 13 202
  • 203. vmx_vcpu_run kernel/x86/vmx.c 2009 8 13 203
  • 204. User Qemu Program vmlaunch / vmresume Linux Guest KVM Kernel Kernel 2009 8 13 204
  • 205. 2009 8 13 205
  • 206. 2009 8 13 206
  • 207. include/asm-x86/vmx.h #define ASM_VMX_VMLAUNCH ".byte 0x0f, 0x01, 0xc2" #define ASM_VMX_VMRESUME ".byte 0x0f, 0x01, 0xc3" 2009 8 13 207
  • 208. 2009 8 13 208
  • 210. User Qemu Program vmlaunch / vmresume Linux Guest KVM Kernel Kernel 2009 8 13 210
  • 211. User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 211
  • 213. 2009 8 13 213
  • 214. VMX non Root Mode IO 2009 8 13 214
  • 215. 2009 8 13 215
  • 216. 2009 8 13 216
  • 217. outb %ax 2009 8 13 217
  • 218. User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 218
  • 219. User Qemu Program Linux Guest KVM Kernel Kernel outb %ax 2009 8 13 219
  • 220. 2009 8 13 220
  • 221. "jne .Llaunched nt" __ex(ASM_VMX_VMLAUNCH) "nt" "jmp .Lkvm_vmx_return nt" ".Llaunched: " __ex(ASM_VMX_VMRESUME) "nt" ".Lkvm_vmx_return: " vmlaunch /* */ vmresume ... ... 2009 8 13 221
  • 222. "jne .Llaunched nt" __ex(ASM_VMX_VMLAUNCH) "nt" "jmp .Lkvm_vmx_return nt" ".Llaunched: " __ex(ASM_VMX_VMRESUME) "nt" ".Lkvm_vmx_return: " /* */ ... 2009 8 13 222
  • 223. 2009 8 13 223
  • 224. 2009 8 13 224
  • 225. out 2009 8 13 225
  • 226. out in . 2009 8 13 226
  • 227. Intel-VT 2009 8 13 227
  • 228. User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 228
  • 229. 2009 8 13 229
  • 230. Qemu 2009 8 13 230
  • 231. User Qemu Program IOCTL return Linux Guest KVM Kernel Kernel 2009 8 13 231
  • 232. User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 232
  • 233. int kvm_cpu_exec(){ ... 461 ret = kvm_vcpu_ioctl(env, KVM_RUN, 0); 478 switch (run->exit_reason) { 479 case KVM_EXIT_IO: 481 ret = kvm_handle_io(env, run->io.port, 482 (uint8_t *)run + run>io.data_offset, 483 run->io.direction, 484 run->io.size, 485 run->io.count); 486 break; case KVM_EXIT_MMIO: .... ... 2009 8 13 233
  • 234. outb 2009 8 13 234
  • 235. int kvm_cpu_exec(){ ... // 461 ret = kvm_vcpu_ioctl(env, KVM_RUN, 0); 478 switch (run->exit_reason) { 479 case KVM_EXIT_IO: 481 ret = kvm_handle_io(env, run->io.port, 482 (uint8_t *)run + run>io.data_offset, 483 run->io.direction, 484 run->io.size, 485 run->io.count); 486 break; case KVM_EXIT_MMIO: outb .... ... 2009 8 13 235
  • 236. 2009 8 13 236
  • 237. IO 2009 8 13 237
  • 238. User Qemu Program io Linux Guest KVM Kernel Kernel 2009 8 13 238
  • 239. 2009 8 13 239
  • 240. out 2009 8 13 240
  • 241. outb %al 2009 8 13 241
  • 242. outb %al 2009 8 13 242
  • 243. ( ) 2009 8 13 243
  • 244. int kvm_cpu_exec(){ ... // 461 ret = kvm_vcpu_ioctl(env, KVM_RUN, 0); 478 switch (run->exit_reason) { 479 case KVM_EXIT_IO: 481 ret = kvm_handle_io(env, run->io.port, 482 (uint8_t *)run + run>io.data_offset, 483 run->io.direction, 484 run->io.size, 485 run->io.count); 486 break; case KVM_EXIT_MMIO: .... ... 2009 8 13 244
  • 245. kvm_handle_io cpu_outb ioport_write 2009 8 13 245
  • 246. static void ioport_write(int index, uint32_t address, uint32_t data) { static IOPortWriteFunc *default_func[3] = { default_ioport_writeb, default_ioport_writew, default_ioport_writel }; IOPortWriteFunc *func = ioport_write_table[index][address]; if (!func) func = default_func[index]; func(ioport_opaque[address], address, data); } ioport_write_table 2009 8 13 246
  • 247. ( ) 2009 8 13 247
  • 248. User Qemu Program io Linux Guest KVM Kernel Kernel 2009 8 13 248
  • 249. User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 249
  • 250. User Qemu Program Linux Guest KVM Kernel Kernel 2009 8 13 250
  • 251. 2009 8 13 251
  • 252. IOCTL CPU 2009 8 13 252
  • 253. User Qemu Program IOCTL Linux Guest KVM Kernel Kernel 2009 8 13 253
  • 254. 2009 8 13 254
  • 255. ... 2009 8 13 255
  • 256. User Qemu Program IOCTL Linux Guest KVM Kernel Kernel vmlaunch 2009 8 13 vmresume 256
  • 257. 2009 8 13 257
  • 258. 378 static int kvm_main_loop_cpu(CPUState *env) 379 { 394 while (1) { 395 while (!has_work(env)) 396 kvm_main_loop_wait(env, 1000); 397 if (env->interrupt_request & (CPU_INTERRUPT_HARD | CPU_INTERRUPT_NMI)) 398 env->halted = 0; // . .... 404 } 405 if (!env->halted && !env->kvm_cpu_state.init) { 406 kvm_cpu_exec(env); // } } 413 } 2009 8 13 258
  • 259. 2009 8 13 259
  • 260. Twitter 2009 8 13 260