CPU虚拟化系列文章1——x86架构CPU虚拟化

{"type":"doc","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"本文摘自于王柏生、谢广军撰写的《深度探索Linux系统虚拟化：原理与实现》一书，介绍了CPU虚拟化的基本概念，探讨了x86架构在虚拟化时面临的障碍，以及为支持CPU虚拟化，Intel在硬件层面实现的扩展VMX。同时，介绍了在VMX扩展支持下，虚拟CPU从Host模式到Guest模式，再回到Host模式的完整生命周期。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"link","attrs":{"href":"https://link.zhihu.com/?target=https%3A//item.jd.com/12742101.html","title":null},"content":[{"type":"text","text":"https://item.jd.com/12742101.html","attrs":{}}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/ca/ca7990e93dacf97a8d62d49216b533a9.jpeg","alt":null,"title":null,"style":null,"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Gerald J. Popek和Robert P. Goldberg在1974年发表的论文“Formal Requirements for Virtualizable Third Generation Architectures”中提出了虚拟化的3个条件：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"1）等价性，","attrs":{}},{"type":"text","text":"即VMM需要在宿主机上为虚拟机模拟出一个本质上与物理机一致的环境。虚拟机在这个环境上运行与其在物理机上运行别无二致，除了可能因为资源竞争或者VMM的干预导致在虚拟环境中表现略有差异，比如虚拟机的I/O、网络等因宿主机的限速或者多个虚拟机共享资源，导致速度可能要比独占物理机时慢一些。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"2）高效性，","attrs":{}},{"type":"text","text":"即虚拟机指令执行的性能与其在物理机上运行相比并无明显损耗。该标准要求虚拟机中的绝大部分指令无须VMM干预而直接运行在物理CPU上，比如我们在x86架构上通过Qemu运行的ARM系统并不是虚拟化，而是模拟。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"3）资源控制，","attrs":{}},{"type":"text","text":"即VMM可以完全控制系统资源。由VMM控制协调宿主机资源给各个虚拟机，而不能由虚拟机控制了宿主机的资源。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"1　陷入和模拟模型","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"为了满足Gerald J. Popek和Robert P. Goldberg提出的虚拟化的3个条件，一个典型的解决方案是陷入和模拟（Trap and Emulate）模型。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"一般来说，处理器分为两种运行模式：系统模式和用户模式。相应地，CPU的指令也分为特权指令和非特权指令。特权指令只能在系统模式运行，如果在用户模式运行就将触发处理器异常。操作系统允许内核运行在系统模式，因为内核需要管理系统资源，需要运行特权指令，而普通的用户程序则运行在用户模式。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在陷入和模拟模型下，虚拟机的用户程序仍然运行在用户模式，但是虚拟机的内核也将运行在用户模式，这种方式称为特权级压缩（Ring Compression）。在这种方式下，虚拟机中的非特权指令直接运行在处理器上，满足了虚拟化标准中高效的要求，即大部分指令无须VMM干预直接在处理器上运行。但是，当虚拟机执行特权指令时，因为是在用户模式下运行，将触发处理器异常，从而陷入VMM中，由VMM代理虚拟机完成系统资源的访问，即所谓的模拟（emulate）。如此，又满足了虚拟化标准中VMM控制系统资源的要求，虚拟机将不会因为可以直接运行特权指令而修改宿主机的资源，从而破坏宿主机的环境。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"2　x86架构虚拟化的障碍","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Gerald J. Popek和Robert P. Goldberg指出，修改系统资源的，或者在不同模式下行为有不同表现的，都属于敏感指令。在虚拟化场景下，VMM需要监测这些敏感指令。一个支持虚拟化的体系架构的敏感指令都属于特权指令，即在非特权级别执行这些敏感指令时CPU会抛出异常，进入VMM的异常处理函数，从而实现了控制VM访问敏感资源的目的。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"但是，x86架构恰恰不能满足这个准则。x86架构并不是所有的敏感指令都是特权指令，有些敏感指令在非特权模式下执行时并不会抛出异常，此时VMM就无法拦截处理VM的行为了。我们以修改FLAGS寄存器中的IF（Interrupt Flag）为例，我们首先使用指令pushf将FLAGS寄存器的内容压到栈中，然后将栈顶的IF清零，最后使用popf指令从栈中恢复FLAGS寄存器。如果虚拟机内核没有运行在ring 0，x86的CPU并不会抛出异常，而只是默默地忽略指令popf，因此虚拟机关闭IF的目的并没有生效。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"有人提出半虚拟化的解决方案，即修改Guest的代码，但是这不符合虚拟化的透明准则。后来，人们提出了二进制翻译的方案，包括静态翻译和动态翻译。静态翻译就是在运行前扫描整个可执行文件，对敏感指令进行翻译，形成一个新的文件。然而，静态翻译必须提前处理，而且对于有些指令只有在运行时才会产生的副作用，无法静态处理。于是，动态翻译应运而生，即在运行时以代码块为单元动态地修改二进制代码。动态翻译在很多VMM中得到应用，而且优化的效果非常不错。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"3　VMX","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"虽然大家从软件层面采用了多种方案来解决x86架构在虚拟化时遇到的问题，但是这些解决方案除了引入了额外的开销外，还给VMM的实现带来了巨大的复杂性。于是，Intel尝试从硬件层面解决这个问题。Intel并没有将那些非特权的敏感指令修改为特权指令，因为并不是所有的特权指令都需要拦截处理。举一个典型的例子，每当操作系统内核切换进程时，都会切换cr3寄存器，使其指向当前运行进程的页表。但是，当使用影子页表进行GVA到HPA的映射时，VMM模块需要捕获Guest每一次设置cr3寄存器的操作，使其指向影子页表。而当启用了硬件层面的EPT支持后，cr3寄存器不再需要指向影子页表，其仍然指向Guest的进程的页表。因此，VMM无须再捕捉Guest设置cr3寄存器的操作，也就是说，虽然写cr3寄存器是一个特权操作，但这个操作不需要陷入VMM。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Intel开发了VT技术以支持虚拟化，为CPU增加了Virtual-Machine Extensions，简称VMX。一旦启动了CPU的VMX支持，CPU将提供两种运行模式：VMX Root Mode和VMX non-Root Mode，每一种模式都支持ring 0 ～ ring 3。VMM运行在VMX Root Mode，除了支持VMX外，VMX Root Mode和普通的模式并无本质区别。VM运行在VMX non-Root Mode，Guest无须再采用特权级压缩方式，Guest kernel可以直接运行在VMX non-Root Mode的ring 0中，如图1所示。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/25/25dcc620fb86bcd94f80be620f1a7eef.jpeg","alt":null,"title":null,"style":null,"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":"center","origin":null},"content":[{"type":"text","text":"图1　VMX运行模式","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"处于VMX Root Mode的VMM可以通过执行CPU提供的虚拟化指令VMLaunch切换到VMX non-Root Mode，因为这个过程相当于进入Guest，所以通常也被称为VM entry。当Guest内部执行了敏感指令，比如某些I/O操作后，将触发CPU发生陷入的动作，从VMX non-Root Mode切换回VMX Root Mode，这个过程相当于退出VM，所以也称为VM exit。然后VMM将对Guest 的操作进行模拟。相比于将Guest的内核也运行在用户模式（ring 1 ～ ring 3）的方式，支持VMX的CPU有以下3点不同：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"1）运行于Guest模式时，Guest用户空间的系统调用直接陷入Guest模式的内核空间，而不再是陷入Host模式的内核空间。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"2）对于外部中断，因为需要由VMM控制系统的资源，所以处于Guest模式的CPU收到外部中断后，则触发CPU从Guest模式退出到Host模式，由Host内核处理外部中断。处理完中断后，再重新切入Guest模式。为了提高I/O效率，Intel支持外设透传模式，在这种模式下，Guest不必产生VM exit，“设备虚拟化”一章将讨论这种特殊方式。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"3）不再是所有的特权指令都会导致处于Guest模式的CPU发生VM exit，仅当运行敏感指令时才会导致CPU从Guest模式陷入Host模式，因为有的特权指令并不需要由VMM介入处理。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如同一个CPU可以分时运行多个任务一样，每个任务有自己的上下文，由调度器在调度时切换上下文，从而实现同一个CPU同时运行多个任务。在虚拟化场景下，同一个物理CPU“一人分饰多角”，分时运行着Host及Guest，在不同模式间按需切换，因此，不同模式也需要保存自己的上下文。为此，VMX设计了一个保存上下文的数据结构：VMCS。每一个Guest都有一个VMCS实例，当物理CPU加载了不同的VMCS时，将运行不同的Guest如图2所示。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/aa/aa901a938124a7477c572ca5e1159257.jpeg","alt":null,"title":null,"style":null,"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":"center","origin":null},"content":[{"type":"text","text":"图2　多个Guest切换","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"VMCS中主要保存着两大类数据，一类是状态，包括Host的状态和Guest的状态，另外一类是控制Guest运行时的行为。其中：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"1）Guest-state area，保存虚拟机状态的区域。当发生VM exit时，Guest的状态将保存在这个区域；当VM entry时，这些状态将被装载到CPU中。这些都是硬件层面的自动行为，无须VMM编码干预。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"2）Host-state area，保存宿主机状态的区域。当发生VM entry时，CPU自动将宿主机状态保存到这个区域；当发生VM exit时，CPU自动从VMCS恢复宿主机状态到物理CPU。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"3）VM-exit information fields。当虚拟机发生VM exit时，VMM需要知道导致VM exit的原因，然后才能“对症下药”，进行相应的模拟操作。为此，CPU会自动将Guest退出的原因保存在这个区域，供VMM使用。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"4）VM-execution control fields。这个区域中的各种字段控制着虚拟机运行时的一些行为，比如设置Guest运行时访问cr3寄存器时是否触发VM exit；控制VM entry与VM exit时行为的VM-entry control fields和VM-exit control fields。此外还有很多不同功能的区域，我们不再一一列举，读者如有需要可以查阅Intel手册。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在创建VCPU时，KVM模块将为每个VCPU申请一个VMCS，每次CPU准备切入Guest模式时，将设置其VMCS指针指向即将切入的Guest对应的VMCS实例：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":null},"content":[{"type":"text","text":"commit 6aa8b732ca01c3d7a54e93f4d701b8aabbe60fb7\n[PATCH] kvm: userspace interface\nlinux.git/drivers/kvm/vmx.c\n\nstatic struct kvm_vcpu *vmx_vcpu_load(struct kvm_vcpu *vcpu)\n{\n u64 phys_addr = __pa(vcpu->vmcs);\n int cpu;\n\n cpu = get_cpu();\n …\n if (per_cpu(current_vmcs, cpu) != vcpu->vmcs) {\n …\n per_cpu(current_vmcs, cpu) = vcpu->vmcs;\n asm volatile (ASM_VMX_VMPTRLD_RAX \"; setna %0\"\n : \"=g\"(error) : \"a\"(&phys_addr), \"m\"(phys_addr)\n : \"cc\");\n …\n }\n …\n}","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"并不是所有的状态都由CPU自动保存与恢复，我们还需要考虑效率。以cr2寄存器为例，大多数时候，从Guest退出Host到再次进入Guest期间，Host并不会改变cr2寄存器的值，而且写cr2的开销很大，如果每次VM entry时都更新一次cr2，除了浪费CPU的算力毫无意义。因此，将这些状态交给VMM，由软件自行控制更为合理。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"4　VCPU生命周期","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"对于每个虚拟处理器（VCPU），VMM使用一个线程来代表VCPU这个实体。在Guest运转过程中，每个VCPU基本都在如图3所示的状态中不断地转换。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/8e/8e3df2ab4e9df32dbb59c72d8c8eaa05.jpeg","alt":null,"title":null,"style":null,"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":"center","origin":null},"content":[{"type":"text","text":"图3　VCPU生命周期","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":null,"normalizeStart":1},"content":[{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"在用户空间准备好后，VCPU所在线程向内核中KVM模块发起一个ioctl请求KVM_RUN，告知内核中的KVM模块，用户空间的操作已经完成，可以切入Guest模式运行Guest了。","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"在进入内核态后，KVM模块将调用CPU提供的虚拟化指令切入Guest模式。如果是首次运行Guest，则使用VMLaunch指令，否则使用VMResume指令。在这个切换过程中，首先，CPU的状态（也就是Host的状态）将会被保存到VMCS中存储Host状态的区域，非CPU自动保存的状态由KVM负责保存。然后，加载存储在VMCS中的Guest的状态到物理CPU，非CPU自动恢复的状态则由KVM负责恢复。","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":3,"align":null,"origin":null},"content":[{"type":"text","text":"物理CPU切入Guest模式，运行Guest指令。当执行Guest指令遇到敏感指令时，CPU将从Guest模式切回到Host模式的ring 0，进入Host内核的KVM模块。在这个切换过程中，首先，CPU的状态（也就是Guest的状态）将会被保存到VMCS中存储Guest状态的区域，然后，加载存储在VMCS中的Host的状态到物理CPU。同样的，非CPU自动保存的状态由KVM模块负责保存。","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":4,"align":null,"origin":null},"content":[{"type":"text","text":"处于内核态的KVM模块从VMCS中读取虚拟机退出原因，尝试在内核中处理。如果内核中可以处理，那么虚拟机就不必再切换到Host模式的用户态了，处理完后，直接快速切回Guest。这种退出也称为轻量级虚拟机退出。","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":5,"align":null,"origin":null},"content":[{"type":"text","text":"如果内核态的KVM模块不能处理虚拟机退出，那么VCPU将再进行一次上下文切换，从Host的内核态切换到Host的用户态，由VMM的用户空间部分进行处理。VMM用户空间处理完毕，再次发起切入Guest模式的指令。在整个虚拟机运行过程中，步骤1～5循环往复。","attrs":{}}]}],"attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"下面是KVM切入、切出Guest的代码：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":null},"content":[{"type":"text","text":"commit 6aa8b732ca01c3d7a54e93f4d701b8aabbe60fb7\n[PATCH] kvm: userspace interface\nlinux.git/drivers/kvm/vmx.c\n\nstatic int vmx_vcpu_run(struct kvm_vcpu *vcpu, …)\n{\n u8 fail;\n u16 fs_sel, gs_sel, ldt_sel;\n int fs_gs_ldt_reload_needed;\n\nagain:\n …\n /* Enter guest mode */\n \"jne launched \\n\\t\"\n ASM_VMX_VMLAUNCH \"\\n\\t\"\n \"jmp kvm_vmx_return \\n\\t\"\n \"launched: \" ASM_VMX_VMRESUME \"\\n\\t\"\n \".globl kvm_vmx_return \\n\\t\"\n \"kvm_vmx_return: \"\n /* Save guest registers, load host registers, keep flags */\n …\n if (kvm_handle_exit(kvm_run, vcpu)) {\n …\n goto again;\n }\n }\n return 0;\n}","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在从Guest退出时，KVM模块首先调用函数kvm_handle_exit尝试在内核空间处理Guest退出。函数kvm_handle_exit有个约定，如果在内核空间可以成功处理虚拟机退出，或者是因为其他干扰比如外部中断导致虚拟机退出等无须切换到Host的用户空间，则返回1；否则返回0，表示需要求助KVM的用户空间处理虚拟机退出，比如需要KVM用户空间的模拟设备处理外设请求。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果内核空间成功处理了虚拟机的退出，则函数kvm_handle_exit返回1，在上述代码中即直接跳转到标签again处，然后程序流程会再次切入Guest。如果函数kvm_handle_exit返回0，则函数vmx_vcpu_run结束执行，CPU从内核空间返回到用户空间，以kvmtool为例，其相关代码片段如下：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":null},"content":[{"type":"text","text":"commit 8d20223edc81c6b199842b36fcd5b0aa1b8d3456\nDump KVM_EXIT_IO details\nkvmtool.git/kvm.c\n\nint main(int argc, char *argv[])\n{\n …\n for (;;) {\n kvm__run(kvm);\n\n switch (kvm->kvm_run->exit_reason) {\n case KVM_EXIT_IO:\n …\n }\n …\n}","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"根据代码可见，kvmtool发起进入Guest的代码处于一个for的无限循环中。当从KVM内核空间返回用户空间后，kvmtool在用户空间处理Guest的请求，比如调用模拟设备处理I/O请求。在处理完Guest的请求后，重新进入下一轮for循环，kvmtool再次请求KVM模块切入Guest。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/ec/ec266841fac439ef402ca60c7e6666d9.jpeg","alt":null,"title":null,"style":null,"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"作者简介：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"王柏生","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"资深技术专家，先后就职于中科院软件所、红旗Linux和百度，现任百度主任架构师。在操作系统、虚拟化技术、分布式系统、云计算、自动驾驶等相关领域耕耘多年，有着丰富的实践经验。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"著有畅销书《深度探索Linux操作系统》（2013年出版）。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"谢广军","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"计算机专业博士，毕业于南开大学计算机系。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"资深技术专家，有多年的IT行业工作经验。现担任百度智能云副总经理，负责云计算相关产品的研发。多年来一直从事操作系统、虚拟化技术、分布式系统、大数据、云计算等相关领域的研发工作，实践经验丰富。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"*本文经出版社授权发布，更多关于虚拟化技术的内容推荐阅读《深度探索Linux系统虚拟化：原理与实现》。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}}]}