Guest OS, Qemu, KVM工作流程

 

这里主要介绍基于x86平台的Guest Os, Qemu, Kvm工作流程，如图，通过KVM APIs可以将qemu的command传递到kvm：

1.创建VM

system_fd = open("/dev/kvm", xxx);

vm_fd = ioctl(system_fd, KVM_CREATE_VM, xxx);

2.创建VCPU

vcpu_fd = kvm_vm_ioctl(vm_fd, VM_CREATE_VCPU, xxx);

3.运行KVM

status = kvm_vcpu_ioctl(vcpu_fd, KVM_RUN, xxx);

Qemu通过KVM APIs进入KVM后，KVM会切入Guest OS，假如Guest OS运行运行，需要访问IO等，也就是说要访问physical device，那么Qemu与KVM就要进行emulate。 如果是KVM emulate的则由KVM emulate，然后切回Guest OS。如果是Qemu emulate的，则从KVM中进入Qemu，等Qemu中的device model执行完emulate之后，再次在Qemu中调用kvm_vcpu_ioctl(vcpu_fd, KVM_RUN, xxx)进入KVM运行，然后再切回Guest OS.

(图片勘误，如果KVM can emulate那么emulate之后应该层层返回到kvm_x86_ops->run(vcpu)，然后才切入guest os，不是直接切入，图画完了，不好修改)

 

Qemu是一个应用程序，所以入口函数当然是main函数，但是一些被type_init修饰的函数会在main函数之前运行。这里分析的代码是emulate x86 的一款i440板子。main函数中会调用在main函数中会调用kvm_init函数来创建一个VM(virtual machine)，然后调用机器硬件初始化相关的函数，对PCI，memory等进行emulate。然后调用qemu_thread_create创建线程，这个函数会调用pthread_create创建一个线程，每个VCPU依靠一个线程来运行。在线程的处理函数qemu_kvm_cpu_thread_fn中，会调用kvm_init_vcpu来创建一个VCPU(virtual CPU)，然后调用kvm_vcpu_ioctl，参数KVM_RUN，这样就进入KVM中了。进入KVM中第一个执行的函数名字相同，也叫kvm_vcpu_ioctl，最终会调用到kvm_x86_ops->run()进入到Guest OS，如果Guest OS要写某个端口，会产生一条IO instruction，这时会从Guest OS中退出，调用kvm_x86_ops->handle_exit函数，其实这个函数被赋值为vmx_handle_exit，最终会调用到kvm_vmx_exit_handlers[exit_reason](vcpu)，kvm_vmx_exit_handlers是一个函数指针，会根据产生事件的类型来匹配使用那个函数。这里因为是ioport访问产生的退出，所以选择handle_io函数。

 

5549static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
5550        [EXIT_REASON_EXCEPTION_NMI]           = handle_exception,
5551        [EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
5552        [EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
5553        [EXIT_REASON_NMI_WINDOW]              = handle_nmi_window,
5554        [EXIT_REASON_IO_INSTRUCTION]          = handle_io,
5555        [EXIT_REASON_CR_ACCESS]               = handle_cr,
5556        [EXIT_REASON_DR_ACCESS]               = handle_dr,
5557        [EXIT_REASON_CPUID]                   = handle_cpuid,
5558        [EXIT_REASON_MSR_READ]                = handle_rdmsr,
5559        [EXIT_REASON_MSR_WRITE]               = handle_wrmsr,
5560        [EXIT_REASON_PENDING_INTERRUPT]       = handle_interrupt_window,
5561        [EXIT_REASON_HLT]                     = handle_halt,
5562        [EXIT_REASON_INVD]                    = handle_invd,
5563        [EXIT_REASON_INVLPG]                  = handle_invlpg,
5564        [EXIT_REASON_VMCALL]                  = handle_vmcall,
5565        [EXIT_REASON_VMCLEAR]                 = handle_vmclear,
5566        [EXIT_REASON_VMLAUNCH]                = handle_vmlaunch,
5567        [EXIT_REASON_VMPTRLD]                 = handle_vmptrld,
5568        [EXIT_REASON_VMPTRST]                 = handle_vmptrst,
5569        [EXIT_REASON_VMREAD]                  = handle_vmread,
5570        [EXIT_REASON_VMRESUME]                = handle_vmresume,
5571        [EXIT_REASON_VMWRITE]                 = handle_vmwrite,
5572        [EXIT_REASON_VMOFF]                   = handle_vmoff,
5573        [EXIT_REASON_VMON]                    = handle_vmon,
5574        [EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
5575        [EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
5576        [EXIT_REASON_WBINVD]                  = handle_wbinvd,
5577        [EXIT_REASON_XSETBV]                  = handle_xsetbv,
5578        [EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
5579        [EXIT_REASON_MCE_DURING_VMENTRY]      = handle_machine_check,
5580        [EXIT_REASON_EPT_VIOLATION]           = handle_ept_violation,
5581        [EXIT_REASON_EPT_MISCONFIG]           = handle_ept_misconfig,
5582        [EXIT_REASON_PAUSE_INSTRUCTION]       = handle_pause,
5583        [EXIT_REASON_MWAIT_INSTRUCTION]       = handle_invalid_op,
5584        [EXIT_REASON_MONITOR_INSTRUCTION]     = handle_invalid_op,
5585};

如果KVM中的handle_io函数可以处理，那么处理完了再次切入Guest OS。如果是在Qemu中emulate，那么在KVM中的代码执行完后，会再次回到Qemu中，调用Qemu中的kvm_handle_io函数，如果可以处理，那么再次调用kvm_vcpu_ioctl，参数KVM_RUN，进入KVM，否则出错退出。