BACKGROUND

The present disclosure relates to the resource management of virtual machine(s) using hardware address mapping, and, more specifically, to facilitate direct access to devices from virtual machines, utilizing control of hardware address translation facilities.

The virtualization of machine resources has been of significant interest for some time; however, with processors becoming more diverse and complex, such as processors that are deeply pipelined/super pipelined, hyper-threaded, on-chip multi-processing capable, and processors having Explicitly Parallel Instruction Computing (EPIC) architecture, and with larger instruction and data caches, virtualization of machine resources is becoming an even greater interest.

Many attempts have been made to make virtualization more efficient. For example, some vendors offer software products that have a virtual machine system that permits a machine to be virtualized, such that the underlying hardware resources of the machine appears as one or more independently operating virtual machines (VM). Typically, a Virtual Machine Monitor (VMM) may be a thin layer of software running on a computer and presenting to other software an abstraction of one or more VMs. Each VM, on the other hand, may function as a self-contained platform, running its own operating system (OS), or a copy of the OS, and/or a software application. Software executing within a VM is collectively referred to as "guest software". Some commercial solutions that provide software VMs include VMware, Inc. (VMware) of Palo Alto, Calif. and VirtualPC by Microsoft Corp. of Redmond, Wash.

Typical software based virtualization solutions utilize the privilege level protections in common processor hardware by running the entire guest software (e.g. OS and applications) at a lower privilege level, and virtualizes guest attempts to execute privileged operations by trapping to the VMM and emulating guest execution through the VMM software. However, typical operating systems expect to utilize these privilege levels to protect OS kernel from user-mode applications. For example, the Intel IA-32 processor architecture defines four privilege levels, specified as ring 0 to ring 3 in decreasing order of privilege.

Processor hardware support for virtualization, such as matter disclosed in previously filed patent application Ser. No. 09/752,134, titled "New Processor Mode For Limiting The Operation Of Guest Software Running On A Virtual Machine Supported By A Virtual Machine Monitor," by Neiger, et al., improves upon the software only virtualization solutions, through additional processor operation modes to differentiate between guest and VMM execution. Additionally it may allow flexible control of guest operations than what is offered by the coarse-grained ring based privilege levels. These additional processor operating modes for virtualization may preserve the typical ring-based processor hardware privilege levels for use by both guest and VMM software.

Typically, input/output I/O devices in the system are managed by the VMM in either one of two modes: shared mode or dedicated mode. In this context an I/O device is any device that facilitates the transfer of data to or from the core of a computer system. Examples of I/O devices may include, but are not limited to: keyboards, pointing devices, graphics adapters, network interface controllers, memory devices, hard disks, diskettes, and writable media. However, other devices are within the scope of the disclosed subject matter.

In shared mode, each VM may typically see and interact with a set of virtual devices. The VMM is typically responsible for emulating these virtual devices and mapping the virtual devices to physical devices in the system. The virtual device emulation is typically used to allow a plurality of virtual machines to share a physical device in, for example, a substantially simultaneous or time-multiplexed fashion. For example, each VM may be exposed a virtual network interface controller (NIC) device by the VMM, and the VMM may map virtual NICs from multiple VMs to a physical NIC in the system, thereby allowing these virtual machines to share the same network connection.

In dedicated mode, a VMM may assign certain physical devices in the platform for exclusive use by a specific virtual machine. I/O device assignment to VMs is typically done to improve performance. Device assignment to VMs is typically done indirectly through the VMM, such that, the VM to which a physical device is assigned is exposed to a virtual device, and the virtual device emulation in the VMM may be optimized to utilize the knowledge that the underlying physical device is assigned for exclusive use for the VM. This kind of VM I/O device assignment has inherent limits (such as reduced performance) since the guest software access to the assigned device is indirectly through the VMM. In addition to performance limitations, software emulated virtual devices also negatively impacts functionality and robustness. Since the software emulated virtual devices are often simpler devices with limited capabilities, any additional capabilities provided by the underlying physical device will not be available to the VM. The extra capabilities offered by the physical device may also be lost due to the virtualization and device emulation overheads. Since virtual device emulation requires the device drivers for the physical devices to be running as part of the privileged VMM software, it also reduces overall system robustness, as errors in any of these drivers can crash the entire system including all the VMs.

DETAILED DESCRIPTION

A physical device in the system may be assigned to a VM such that the physical device resources are mapped for direct access from a virtual machine (VM). As opposed to the typical I/O virtualization solution, described above, that creates a virtual device emulated in software, direct mapping enables a VM to access the resources of the assigned physical device without entering the virtual machine monitor (VMM), thereby avoiding the associated performance overheads. In this context a VMM may include a dedicated software application, a host virtualization driver, or a host firmware. Hereafter, the above will simply be referred to as a "VMM." Direct mapped access may also allow the VM to access and utilize all the capabilities of the physical device without at least some of the limitations typically imposed by virtual device emulation. In addition, direct access to physical devices from VMs may move the device drivers for these physical devices out of the most privileged VMM software into lower privileged guest software, and thus, may improve the robustness of the overall system.

Direct access to physical I/O devices from virtual machines may involve at least two distinct components: (a) partitioning the physical I/O devices across VMs such that partitioned devices are visible only to the VM to which they are assigned; (b) enabling direct access to a partitioned device from the VM through translation mechanisms for different types of interactions between a VM and its assigned device resources, and explicitly limiting translations for other VMs to restrict accessibility to this device. The disclosed subject matter pertains to specific techniques and example embodiments for these components.

Namespaces may be applied to limit partition visibility, where a primary name space may be created to uniquely identify each physical object, secondary name spaces may be created for each virtual machine domain and applying a domain specific mapping function to map objects from primary namespace to each secondary namespace. Operating systems typically discover and configure devices by enumerating buses and devices in the I/O complex, where the enumeration utilizes well-defined naming schemes to identify and communicate with specific I/O agents. For example, Peripheral Component Interconnect (PCI) based I/O defines a depth first enumeration scheme, where each PCI bus in the system is uniquely through a bus number, each device is identified through a device number unique on its local bus and each of the device's functions through a function number unique for the local device.

FIG. 1 is a flowchart illustrating an embodiment of a technique for exposing partitioned physical devices in the system to virtual machine(s) using namespace mapping in accordance with the disclosed subject matter. Block 110 illustrates that the VMM may enumerate and configure the physical devices in the primary namespace by controlling the mechanisms for device discovery, configuration and hot plug operations. In some embodiments, this may be similar to how typical firmware, such as a basic input/output system (BIOS) or operating systems discover I/O devices in a system.

Block 120 illustrates that the VMM may create a namespace mapping function for each VM to map specific devices from its primary namespace to the VMs namespace. In some embodiments, these namespace mapping functions may be generated statically based on user configuration or policies for VM device assignment. In other embodiments, the mapping function may be dynamically generated and applied to respond to system conditions (such as system thermal or power status), user actions (such as device hot plugging) or device in-activity.

Block 130 illustrates that the VMM may trap attempts by guest software in a VM to enumerate and configure devices in its virtual I/O fabric (secondary namespace). One embodiment may utilize virtualization support in processor hardware to explicitly trap and handle device enumeration and configuration operations from VMs with direct hardware access. In one specific embodiment, I/O fabric, specifically PCI, enumeration may be done through specific I/O ports. Hardware support for virtualization may provide I/O port bitmaps to explicitly trap access to specific I/O ports while allowing direct access to others. This is contrasted with existing software only solutions that have no choice but to trap all the I/O ports, which hinders direct access to the I/O ports of a specific device.

Block 140 illustrates that the namespace mapping function may be applied to manage devices that are exposed on a VMs virtual I/O fabric. In one embodiment, the mapping function may enumerate only physical devices assigned exclusively to a VM on its I/O fabric. Another embodiment may expose both assigned physical devices and emulated virtual devices (such as when the physical device is being shared by multiple VMs through the VMM) on a VM's virtual I/O fabric. In yet another embodiment, the VMM may expose a virtual I/O fabric structure to a VM that allows optimization or simplification of management operations on its partitioned devices.

Block 150 illustrates that the VMM may virtualize and expose to a VM, mechanisms for dynamic insertion or removal of assigned physical devices. One embodiment may expose these mechanisms as virtual hot plug controllers or other mechanisms compatible to the emulated I/O fabric and map it over equivalent mechanisms on the physical system. Another embodiment may utilize the dynamic removal and insertion of devices to a VM based on user actions or operating conditions. Given the generality of the VMM based namespace mapping method many other embodiments are possible that can specialize for specific requirements.

A technique to enable the direct mapping of physical device resources to a VM may need to address at least three types of interactions typical between guest software and I/O: (a) processor access to device memory-mapped and I/O mapped resources; (b) device access to system memory; and (c) events generated from device and delivered to processor. Each of these interactions may require some sort of translation function between guest and host attributes that are, ideally, transparent to the guest software and the physical device. Attributes used in the translation function may include: guest and host physical addresses for memory-mapped resources, guest and host I/O port addresses for I/O-mapped resources, guest and host interrupt vector addresses and any other relevant attributes that needs translation. For frequently utilized and performance sensitive interactions, the translation function may be accelerated through hardware support, while in-frequent, less performance sensitive interactions may continued to be proxied and translated indirectly through VMM software. Additionally, to setup and apply the translation function correctly, the guest and host address attributes to assigned device resources need to be tracked by the VMM at all times.

FIG. 2 is a flowchart illustrating an embodiment of a technique for direct mapping resources of partitioned I/O devices to virtual machine(s) utilizing hardware support for virtualization in accordance with the disclosed subject matter. The host attribute settings are typically initialized by system initialization firmware, a host operating system, a VMM or a combination thereof. Block 210 illustrates that the VMM may initialize the guest attribute settings as part of virtual I/O fabric configuration. Specifically, in one embodiment, the VMM or the guest VM firmware may create an initial guest physical address mapping for an exclusive use, as opposed to shared, physical devices. In one embodiment, the VMM may group the addresses of the exclusive use devices in such a way that the addresses of the various devices are contiguous, or otherwise convenient for the VMM to manage. In some embodiments this may be done by the VMM itself, the VM's virtual firmware or a combination of these.

Block 220 illustrates that the VMM may setup the initial translation structures by associating corresponding host and guest attributes or I/O addresses. In one embodiment, an addresses mapping may be created between guest physical and host physical addresses for exclusive use physical devices. In some embodiments, depending on the translation facility, the associations may include additional information (for example the guest kernel virtual address) which may become available only after the guest OS has loaded.

Block 230 illustrates that the VMM may trap any attempts by guest software or host software if VMM is part of a host OS) to re-program the guest (or host) attributes. In one embodiment, a guest (or host) may attempt to re-program the interrupt vectors or address of memory mapped registers of the assigned device as part of processes of re-balancing the VM's I/O hierarchy. The VMM may then trap the attempt to remap the address assignments, and, in one embodiment, re-program the guest to host address translation tables. Block 240 illustrates that the VMM may track the new guest (or host) attributes and updates the associations used for translating.

FIG. 3 is a flowchart illustrating an embodiment of a technique for translating between guest and host physical addresses at a page granularity to enable direct-mapping of memory-mapped device resources without trapping by the VMM. The granularity for translations between guest and host attributes may vary depending on the translation mechanism and the attribute type being translated; therefore, the disclosed matter is not limited to any one level of granularity, page level or otherwise.

Block 310 illustrates a VM may attempt to access a memory-mapped resource of the assigned direct mapped device. In one embodiment, the VM may attempt to access the device utilizing a guest attribute. In one embodiment, the guest OS may use a virtual address for the access attempt which has a valid guest physical address translation in the guest page tables.

Block 320 illustrates that a translation logic may check if an association exists between the guest attribute to a host physical address. Depending on the translation method, the guest attribute checked could be guest physical address, guest virtual address or a combination of these. One embodiment may utilize a virtual memory virtualization technique including a virtual translation lookaside buffer (VTLB) such as the matter disclosed in the previously filed patent application Ser. No. 09/752,133, titled "A Virtual Translation Look aside Buffer," by Neiger, et al. Another embodiment may perform the guest to host translation by utilizing or adding to the address translation facilities in processor hardware.

Block 360 illustrates that, if a valid association check exists, the translation mechanism may use the association to complete the guest access with the resulting host physical address without involving the VMM.

Blocks 330340, &350 illustrate the case when an association is not found in the translation logic. In one embodiment of such a case the VMM may trap the attempt to access the device, thus allowing it to setup the proper associations for possible future translations. Unlike general purpose memory which can be paged by the guest or VMM (host OS) independently, pages containing memory-mapped device addresses are often non-pageable in both guest and VMM. This, along with the technique illustrated in FIG. 2, make the path from Block 320 directly to Block 360 the common case.

Block 330 illustrates that, if no association is found, the translation logic may generate a specific event to transfer execution control to the VMM, or other device. Block 340 illustrates that the VMM or host operating system virtualization driver may associate the virtual address with the physical address without the aid of the translation logic. In one embodiment, page faults may be utilized as the execution control transfer event and the VMM may detect that the trapped page fault is due to access of a direct memory mapped device. In one embodiment, the VMM may to do this by looking at the virtual address in the page table that maps to the guest virtual address and detecting that the virtual address is associated with the direct mapped device by comparing the virtual address to a guest-to-host attribute translation table maintained by the VMM.

Block 350 illustrates that once the proper translation has been determined, the address mapping may be stored within the translation logic for future use. It is contemplated that the translation logic may have a limited number of entries. In one embodiment, a most recently used algorithm may be used to storing and deleting entries from the translation logic; however, other techniques of dealing with a limited translation logic storage capability are within the scope of the disclosed matter. In one embodiment, the VMM may resume execution from the instruction that caused the page fault. Block 360illustrates that the translation mechanism or the VMM may use the association to complete the guest access with the resulting host physical address.

In another embodiment, I/O mapped resource translations may be achieved more simply by opportunistically partitioning the host I/O port space across VMs without any translation and enable direct access from each VM to its partitioned I/O ports. In cases when an I/O port address use conflict is detected between multiple VMs or with the host, the translation may be done by trapping to the VMM as before. Direct access of devices from guest software causes the devices to be programmed by the guest with guest physical addresses for device bus master transactions such as direct memory access DMA. This requires the guest to host translation function to be applied for transactions originating from the device destined to system memory. This translation need to be consistent with any processor translation schemes. In some embodiments these translations may reside on the I/O device itself, on some intermediate chipset logic, or a combination thereof.

FIG. 4 is a block diagram illustrating an embodiment of an apparatus 401 and a system 400 that allows for I/O resource partitioning and direct mapping to virtual machine(s) using hardware support for virtualization in accordance with the disclosed subject matter. In one embodiment, the apparatus may include a VMM 410, having a namespace & mapping controller 415, and an integrated circuit 420, having a translation logic 425. In one embodiment, the integrated circuit may be a processor or a portion of a chipset or other device. In another embodiment, the translation logic may be distributed amongst various integrated circuits. In one embodiment, the system may include the apparatus, a first virtual machine 431, having a virtual device 441, a second virtual machine 432, having a virtual device 442, a shared physical device 495, and an exclusive use direct mapped physical device 490.

The namespace & mapping controller 415 may be capable of managing the visibility of partitioned I/O devices to virtual machines, specifically the first and second virtual machines 431 &432. In one embodiment, the namespace & mapping controller may be capable of performing the technique illustrated by FIG. 1 and described above.

The VMM 410 may, in one embodiment, initialize and track the guest-to-host address and attribute translations. In one embodiment, the VMM may configure and maintain the guest-to-host translations 427. In one embodiment, the guest-to-host translations may be stored within a table in a memory element. In one embodiment, the VMM may be capable of performing the technique illustrated by FIG. 2 and described above.

The translation logic 425 may be capable of providing a guest-to-host attribute translation without involving the VMM 410. In one embodiment, the first VM 431 may use the translation logic to access the direct mapped physical device 490. In another embodiment, the first VM may also use the VMM to access the physical device if the guest-to-host mapping is not accessible to the translation logic. In one embodiment, the translation logic may use the guest-to-host translation table 427 configured by the VMM to facilitate the access of the physical device. In one embodiment, the VMM and translation logic may be capable of performing the technique illustrated by FIG. 3 and described above.

SRC=http://www.freepatentsonline.com/7467381.html

PatentTips - Resource partitioning and direct access utilizing hardware support for virtualization的更多相关文章

  1. Direct Access to Video Encoding and Decoding

    来源:http://asciiwwdc.com/2014/sessions/513   Direct Access to Video Encoding and Decoding  Session 5 ...

  2. 【转】VirtualBox direct access to SD Card in Windows--不错

    原文网址:http://www.sandyscott.net/2013/08/14/virtualbox-direct-drive-access/ I’ve trying to get my Rasp ...

  3. PatentTips - System and method to deprivilege components of a virtual machine monitor

    BACKGROUND INFORMATION An embodiment of the present invention relates generally to virtualization pl ...

  4. 转载:Why using Single Root I/O Virtualization (SR-IOV) can help improve I/O performance and Reduce Costs

    Introduction While server virtualization is being widely deployed in an effort to reduce costs and o ...

  5. (英文版)使用Visual Studio 2015 编写 MASM 汇编程序!

    原文地址:http://kipirvine.com/asm/gettingStartedVS2015/index.htm#CreatingProject Getting Started with MA ...

  6. Virtualization Essentials---Understanding hypervisor

    Original link from : http://searchservervirtualization.techtarget.com/tip/Understanding-hosted-and-b ...

  7. System and method to prioritize large memory page allocation in virtualized systems

    The prioritization of large memory page mapping is a function of the access bits in the L1 page tabl ...

  8. 浅析 Linux 初始化 init 系统,第 1 部分: sysvinit 第 2 部分: UpStart 第 3 部分: Systemd

    浅析 Linux 初始化 init 系统,第 1 部分: sysvinit  第 2 部分: UpStart 第 3 部分: Systemd http://www.ibm.com/developerw ...

  9. EyeQ进展The Evolution of EyeQ

    EyeQ进展The Evolution of EyeQ Mobileye's proven leadership in ADAS technologies is based in our EyeQ f ...

随机推荐

  1. Python正则表达式初识(六)

    继续分享Python正则表达式基础,今天给大家分享的正则表达式特殊符号是“[]”.中括号十分实用,其有特殊含义,其代表的意思是中括号中的字符只要满足其中任意一个就可以.其用法一共有三种,分别对其进行具 ...

  2. Linux下解压缩文件命令总结

    ---------------------------------------------tar命令: -v 可视化-c 新建包-f 指定文件名(除非你用默认用户名)-x 解压target.tar-r ...

  3. angular-cli6使用ng serve --o 控制台报错Invalid Host Header

    angular-cli6使用ng serve --o 运行创建的项目,能正常访问,但是控制台一直在报错,点击报错能看到是因为webpack的webpack-dev-server里报的错, 看到网上介绍 ...

  4. 【Linux下自定义Shell终端提示符】

    目录 基本转义符 字体颜色 背景颜色 移动光标 @ Linux系统终端提示符的特征由系统环境变量 PS1(Prompt String One)定义. 我们可以通过命令echo $PS1来查看当前设置, ...

  5. Linux系统中命令提示符解释-Linux中命令格式、分类-bash中的快捷键

    1.命令提示符 [root@mysqlserver01 opt]# [root@mysqlserver01 ~]# root 当前用户名 @ 固定的(格式) mysqlserver01 主机名 opt ...

  6. BZOJ——T 3732: Network

    http://www.lydsy.com/JudgeOnline/problem.php?id=3732 Time Limit: 10 Sec  Memory Limit: 128 MBSubmit: ...

  7. 《从0到1》读书笔记第2章"像1999 年那样狂欢"第1记:小结及词汇解析

    小结 本章的目的应该是通过90年代末的互联网泡沫的背景,成因.影响,以及教训来教诫人们,在全部人都疯狂的抛身于洪流热潮之中时,我们要冷静的思考辨识出那些不切实际的大众观点,找到隐藏在这些观点后面的反主 ...

  8. Android自己定义视图(一):带下划线的TextView

    package com.francis.underlinetextviewtest; import android.content.Context; import android.content.re ...

  9. 操作excel脚本练习

    # -*- coding: utf-8 -*- import xlrd import xlwt import sys from xlwt import * from xlrd import open_ ...

  10. Fragment-Transaction 源码分析

    概述 这篇文章的简要分析了Activity中的Transaction和add,replace等操作以及backstack的工作原理. 分析transaction源码的原因是因为我在写一个测试代码的时候 ...