CPU虛擬化系列文章1——x86架構CPU虛擬化

{"type":"doc","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"本文摘自於王柏生、謝廣軍撰寫的《深度探索Linux系統虛擬化：原理與實現》一書，介紹了CPU虛擬化的基本概念，探討了x86架構在虛擬化時面臨的障礙，以及爲支持CPU虛擬化，Intel在硬件層面實現的擴展VMX。同時，介紹了在VMX擴展支持下，虛擬CPU從Host模式到Guest模式，再回到Host模式的完整生命週期。","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年發表的論文“Formal Requirements for Virtualizable Third Generation Architectures”中提出了虛擬化的3個條件：","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）等價性，","attrs":{}},{"type":"text","text":"即VMM需要在宿主機上爲虛擬機模擬出一個本質上與物理機一致的環境。虛擬機在這個環境上運行與其在物理機上運行別無二致，除了可能因爲資源競爭或者VMM的干預導致在虛擬環境中表現略有差異，比如虛擬機的I/O、網絡等因宿主機的限速或者多個虛擬機共享資源，導致速度可能要比獨佔物理機時慢一些。","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":"即虛擬機指令執行的性能與其在物理機上運行相比並無明顯損耗。該標準要求虛擬機中的絕大部分指令無須VMM干預而直接運行在物理CPU上，比如我們在x86架構上通過Qemu運行的ARM系統並不是虛擬化，而是模擬。","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）資源控制，","attrs":{}},{"type":"text","text":"即VMM可以完全控制系統資源。由VMM控制協調宿主機資源給各個虛擬機，而不能由虛擬機控制了宿主機的資源。","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　陷入和模擬模型","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":"爲了滿足Gerald J. Popek和Robert P. Goldberg提出的虛擬化的3個條件，一個典型的解決方案是陷入和模擬（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":"一般來說，處理器分爲兩種運行模式：系統模式和用戶模式。相應地，CPU的指令也分爲特權指令和非特權指令。特權指令只能在系統模式運行，如果在用戶模式運行就將觸發處理器異常。操作系統允許內核運行在系統模式，因爲內核需要管理系統資源，需要運行特權指令，而普通的用戶程序則運行在用戶模式。","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":"在陷入和模擬模型下，虛擬機的用戶程序仍然運行在用戶模式，但是虛擬機的內核也將運行在用戶模式，這種方式稱爲特權級壓縮（Ring Compression）。在這種方式下，虛擬機中的非特權指令直接運行在處理器上，滿足了虛擬化標準中高效的要求，即大部分指令無須VMM干預直接在處理器上運行。但是，當虛擬機執行特權指令時，因爲是在用戶模式下運行，將觸發處理器異常，從而陷入VMM中，由VMM代理虛擬機完成系統資源的訪問，即所謂的模擬（emulate）。如此，又滿足了虛擬化標準中VMM控制系統資源的要求，虛擬機將不會因爲可以直接運行特權指令而修改宿主機的資源，從而破壞宿主機的環境。","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架構虛擬化的障礙","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，修改系統資源的，或者在不同模式下行爲有不同表現的，都屬於敏感指令。在虛擬化場景下，VMM需要監測這些敏感指令。一個支持虛擬化的體系架構的敏感指令都屬於特權指令，即在非特權級別執行這些敏感指令時CPU會拋出異常，進入VMM的異常處理函數，從而實現了控制VM訪問敏感資源的目的。","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架構恰恰不能滿足這個準則。x86架構並不是所有的敏感指令都是特權指令，有些敏感指令在非特權模式下執行時並不會拋出異常，此時VMM就無法攔截處理VM的行爲了。我們以修改FLAGS寄存器中的IF（Interrupt Flag）爲例，我們首先使用指令pushf將FLAGS寄存器的內容壓到棧中，然後將棧頂的IF清零，最後使用popf指令從棧中恢復FLAGS寄存器。如果虛擬機內核沒有運行在ring 0，x86的CPU並不會拋出異常，而只是默默地忽略指令popf，因此虛擬機關閉IF的目的並沒有生效。","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":"有人提出半虛擬化的解決方案，即修改Guest的代碼，但是這不符合虛擬化的透明準則。後來，人們提出了二進制翻譯的方案，包括靜態翻譯和動態翻譯。靜態翻譯就是在運行前掃描整個可執行文件，對敏感指令進行翻譯，形成一個新的文件。然而，靜態翻譯必須提前處理，而且對於有些指令只有在運行時纔會產生的副作用，無法靜態處理。於是，動態翻譯應運而生，即在運行時以代碼塊爲單元動態地修改二進制代碼。動態翻譯在很多VMM中得到應用，而且優化的效果非常不錯。","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":"雖然大家從軟件層面採用了多種方案來解決x86架構在虛擬化時遇到的問題，但是這些解決方案除了引入了額外的開銷外，還給VMM的實現帶來了巨大的複雜性。於是，Intel嘗試從硬件層面解決這個問題。Intel並沒有將那些非特權的敏感指令修改爲特權指令，因爲並不是所有的特權指令都需要攔截處理。舉一個典型的例子，每當操作系統內核切換進程時，都會切換cr3寄存器，使其指向當前運行進程的頁表。但是，當使用影子頁表進行GVA到HPA的映射時，VMM模塊需要捕獲Guest每一次設置cr3寄存器的操作，使其指向影子頁表。而當啓用了硬件層面的EPT支持後，cr3寄存器不再需要指向影子頁表，其仍然指向Guest的進程的頁表。因此，VMM無須再捕捉Guest設置cr3寄存器的操作，也就是說，雖然寫cr3寄存器是一個特權操作，但這個操作不需要陷入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開發了VT技術以支持虛擬化，爲CPU增加了Virtual-Machine Extensions，簡稱VMX。一旦啓動了CPU的VMX支持，CPU將提供兩種運行模式：VMX Root Mode和VMX non-Root Mode，每一種模式都支持ring 0 ～ ring 3。VMM運行在VMX Root Mode，除了支持VMX外，VMX Root Mode和普通的模式並無本質區別。VM運行在VMX non-Root Mode，Guest無須再採用特權級壓縮方式，Guest kernel可以直接運行在VMX non-Root Mode的ring 0中，如圖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":"圖1　VMX運行模式","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":"處於VMX Root Mode的VMM可以通過執行CPU提供的虛擬化指令VMLaunch切換到VMX non-Root Mode，因爲這個過程相當於進入Guest，所以通常也被稱爲VM entry。當Guest內部執行了敏感指令，比如某些I/O操作後，將觸發CPU發生陷入的動作，從VMX non-Root Mode切換回VMX Root Mode，這個過程相當於退出VM，所以也稱爲VM exit。然後VMM將對Guest 的操作進行模擬。相比於將Guest的內核也運行在用戶模式（ring 1 ～ ring 3）的方式，支持VMX的CPU有以下3點不同：","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）運行於Guest模式時，Guest用戶空間的系統調用直接陷入Guest模式的內核空間，而不再是陷入Host模式的內核空間。","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）對於外部中斷，因爲需要由VMM控制系統的資源，所以處於Guest模式的CPU收到外部中斷後，則觸發CPU從Guest模式退出到Host模式，由Host內核處理外部中斷。處理完中斷後，再重新切入Guest模式。爲了提高I/O效率，Intel支持外設透傳模式，在這種模式下，Guest不必產生VM exit，“設備虛擬化”一章將討論這種特殊方式。","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）不再是所有的特權指令都會導致處於Guest模式的CPU發生VM exit，僅當運行敏感指令時纔會導致CPU從Guest模式陷入Host模式，因爲有的特權指令並不需要由VMM介入處理。","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":"如同一個CPU可以分時運行多個任務一樣，每個任務有自己的上下文，由調度器在調度時切換上下文，從而實現同一個CPU同時運行多個任務。在虛擬化場景下，同一個物理CPU“一人分飾多角”，分時運行着Host及Guest，在不同模式間按需切換，因此，不同模式也需要保存自己的上下文。爲此，VMX設計了一個保存上下文的數據結構：VMCS。每一個Guest都有一個VMCS實例，當物理CPU加載了不同的VMCS時，將運行不同的Guest如圖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":"圖2　多個Guest切換","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中主要保存着兩大類數據，一類是狀態，包括Host的狀態和Guest的狀態，另外一類是控制Guest運行時的行爲。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，保存虛擬機狀態的區域。當發生VM exit時，Guest的狀態將保存在這個區域；當VM entry時，這些狀態將被裝載到CPU中。這些都是硬件層面的自動行爲，無須VMM編碼干預。","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，保存宿主機狀態的區域。當發生VM entry時，CPU自動將宿主機狀態保存到這個區域；當發生VM exit時，CPU自動從VMCS恢復宿主機狀態到物理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。當虛擬機發生VM exit時，VMM需要知道導致VM exit的原因，然後才能“對症下藥”，進行相應的模擬操作。爲此，CPU會自動將Guest退出的原因保存在這個區域，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。這個區域中的各種字段控制着虛擬機運行時的一些行爲，比如設置Guest運行時訪問cr3寄存器時是否觸發VM exit；控制VM entry與VM exit時行爲的VM-entry control fields和VM-exit control fields。此外還有很多不同功能的區域，我們不再一一列舉，讀者如有需要可以查閱Intel手冊。","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":"在創建VCPU時，KVM模塊將爲每個VCPU申請一個VMCS，每次CPU準備切入Guest模式時，將設置其VMCS指針指向即將切入的Guest對應的VMCS實例：","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":"並不是所有的狀態都由CPU自動保存與恢復，我們還需要考慮效率。以cr2寄存器爲例，大多數時候，從Guest退出Host到再次進入Guest期間，Host並不會改變cr2寄存器的值，而且寫cr2的開銷很大，如果每次VM entry時都更新一次cr2，除了浪費CPU的算力毫無意義。因此，將這些狀態交給VMM，由軟件自行控制更爲合理。","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生命週期","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":"對於每個虛擬處理器（VCPU），VMM使用一個線程來代表VCPU這個實體。在Guest運轉過程中，每個VCPU基本都在如圖3所示的狀態中不斷地轉換。","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":"圖3　VCPU生命週期","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":"在用戶空間準備好後，VCPU所在線程向內核中KVM模塊發起一個ioctl請求KVM_RUN，告知內核中的KVM模塊，用戶空間的操作已經完成，可以切入Guest模式運行Guest了。","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"在進入內核態後，KVM模塊將調用CPU提供的虛擬化指令切入Guest模式。如果是首次運行Guest，則使用VMLaunch指令，否則使用VMResume指令。在這個切換過程中，首先，CPU的狀態（也就是Host的狀態）將會被保存到VMCS中存儲Host狀態的區域，非CPU自動保存的狀態由KVM負責保存。然後，加載存儲在VMCS中的Guest的狀態到物理CPU，非CPU自動恢復的狀態則由KVM負責恢復。","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，運行Guest指令。當執行Guest指令遇到敏感指令時，CPU將從Guest模式切回到Host模式的ring 0，進入Host內核的KVM模塊。在這個切換過程中，首先，CPU的狀態（也就是Guest的狀態）將會被保存到VMCS中存儲Guest狀態的區域，然後，加載存儲在VMCS中的Host的狀態到物理CPU。同樣的，非CPU自動保存的狀態由KVM模塊負責保存。","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":4,"align":null,"origin":null},"content":[{"type":"text","text":"處於內核態的KVM模塊從VMCS中讀取虛擬機退出原因，嘗試在內核中處理。如果內核中可以處理，那麼虛擬機就不必再切換到Host模式的用戶態了，處理完後，直接快速切回Guest。這種退出也稱爲輕量級虛擬機退出。","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":5,"align":null,"origin":null},"content":[{"type":"text","text":"如果內核態的KVM模塊不能處理虛擬機退出，那麼VCPU將再進行一次上下文切換，從Host的內核態切換到Host的用戶態，由VMM的用戶空間部分進行處理。VMM用戶空間處理完畢，再次發起切入Guest模式的指令。在整個虛擬機運行過程中，步驟1～5循環往復。","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的代碼：","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":"在從Guest退出時，KVM模塊首先調用函數kvm_handle_exit嘗試在內核空間處理Guest退出。函數kvm_handle_exit有個約定，如果在內核空間可以成功處理虛擬機退出，或者是因爲其他干擾比如外部中斷導致虛擬機退出等無須切換到Host的用戶空間，則返回1；否則返回0，表示需要求助KVM的用戶空間處理虛擬機退出，比如需要KVM用戶空間的模擬設備處理外設請求。","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":"如果內核空間成功處理了虛擬機的退出，則函數kvm_handle_exit返回1，在上述代碼中即直接跳轉到標籤again處，然後程序流程會再次切入Guest。如果函數kvm_handle_exit返回0，則函數vmx_vcpu_run結束執行，CPU從內核空間返回到用戶空間，以kvmtool爲例，其相關代碼片段如下：","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":"根據代碼可見，kvmtool發起進入Guest的代碼處於一個for的無限循環中。當從KVM內核空間返回用戶空間後，kvmtool在用戶空間處理Guest的請求，比如調用模擬設備處理I/O請求。在處理完Guest的請求後，重新進入下一輪for循環，kvmtool再次請求KVM模塊切入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":"作者簡介：","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":"資深技術專家，先後就職於中科院軟件所、紅旗Linux和百度，現任百度主任架構師。在操作系統、虛擬化技術、分佈式系統、雲計算、自動駕駛等相關領域耕耘多年，有着豐富的實踐經驗。","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":"著有暢銷書《深度探索Linux操作系統》（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":"謝廣軍","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":"計算機專業博士，畢業於南開大學計算機系。","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":"資深技術專家，有多年的IT行業工作經驗。現擔任百度智能雲副總經理，負責雲計算相關產品的研發。多年來一直從事操作系統、虛擬化技術、分佈式系統、大數據、雲計算等相關領域的研發工作，實踐經驗豐富。","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":"*本文經出版社授權發佈，更多關於虛擬化技術的內容推薦閱讀《深度探索Linux系統虛擬化：原理與實現》。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}}]}