Instruction set emulation for guest operating systems

Abstract

The described implementations relate to virtual computing techniques. One implementation provides a technique that can include receiving a request to execute an application. The application can include first application instructions from a guest instruction set architecture. The technique can also include loading an emulator and a guest operating system into an execution context with the application. The emulator can translate the first application instructions into second application instructions from a host instruction set architecture. The technique can also include running the application by executing the second application instructions.

Description

BACKGROUND

Computer applications are often programmed to interface with an operating system that provides various functions to the applications. For example, operating systems can provide lower-level services such as device drivers and memory management routines, or higher-level services such as graphical user interface application programming interfaces (“APIs”). In some cases, a particular application may only be provided for a particular operating system. For example, an application may be implemented on one or more versions of Windows®, but may not have been ported to Unix® or other operating systems.

Computer applications are also often limited to certain instruction set architectures. For example, an application can be compiled into binary code for one instruction set architecture, e.g., x86. This binary code generally will not execute on a different instruction set architecture, e.g., a PowerPC processor. Rather, different compiled binaries normally need to be generated for each instruction set architecture on which the application is expected to run.

Traditional techniques to allow application portability across operating systems and instruction set architectures tend to be relatively inefficient and cumbersome. For example, virtual machine monitors can allow a copy of a guest operating system to execute on a computer system with a different underlying host operating system. However, this approach generally involves executing full copies of both the guest operating system as well as the underlying host operating system. Likewise, central processing unit (“CPU”) emulators can be used to translate code into different instruction sets at runtime instead of using precompiled binaries. However, these emulators generally have a substantial memory and processor footprint that do not lend them to efficient integration with a guest operating system executing on a virtual machine.

SUMMARY

This document relates to virtual computing techniques. One implementation is manifested as a technique that can include receiving a request to execute an application. The application can include first application instructions from a guest instruction set architecture. The technique can also include loading an emulator and a guest operating system into an execution context with the application. The emulator can translate the first application instructions into second application instructions from a host instruction set architecture. The guest operating system can provide at least one guest operating system interface that is called by the application. The technique can also include running the application by executing the second application instructions.

Another implementation is manifested as a system that can include an emulator configured to receive a request from a guest operating system to load first instructions in a guest instruction set architecture. The emulator can also be configured to check a translation table to determine whether corresponding second instructions are available in a host instruction set architecture, and, in a first instance when the second instructions are available, load the second instructions. The emulator can also be configured to, in a second instance when the second instructions are not available, delay translating the first instructions until a request to execute the first instructions is received. The system can also include at least one processing device configured to execute the emulator.

Another implementation is manifested as a computer-readable storage media that can include instructions. The instructions can be executed by one or more processing devices, which can cause the one or more processing devices to perform receiving a request to execute an application on a host device. The host device can have a host instruction set architecture and a host operating system. The application can include one or more application binaries. The acts can also include checking application metadata associated with the application to identify an instruction set architecture and an operating system for the application. When the instruction set architecture for the application corresponds to the host instruction set architecture and the operating system for the application is the host operating system, the application can be loaded in an execution context. When the instruction set architecture for the application corresponds to the host instruction set architecture and the operating system for the application is different than the host operating system, the application can be loaded together with a guest operating system for the application an execution context. When the instruction set architecture for the application is different than the host instruction set architecture and the operating system for the application is different than the host operating system, the application, an emulator, and a guest operating system for the application can be loaded together in an execution context.

The above listed examples are intended to provide a quick reference to aid the reader and are not intended to define the scope of the concepts described herein.

BRIEF DESCRIPTION OF THE DRAWINGS

The accompanying drawings illustrate implementations of the concepts conveyed in the present document. Features of the illustrated implementations can be more readily understood by reference to the following description taken in conjunction with the accompanying drawings. Like reference numbers in the various drawings are used wherever feasible to indicate like elements. Further, the left-most numeral of each reference number conveys the figure and associated discussion where the reference number is first introduced.

FIGS. 1 and 5 show exemplary architectures of a host computing device in accordance with some implementations of the present concepts.

FIG. 2 shows an exemplary logical architecture in accordance with some implementations of the present concepts.

FIG. 3 shows an exemplary architecture of an emulator in accordance with some implementations of the present concepts.

FIG. 4A-4C show an exemplary table in accordance with some implementations of the present concepts.

FIG. 6 shows an exemplary system in accordance with some implementations of the present concepts.

FIGS. 7-9 show flowcharts of exemplary methods that can be accomplished in accordance with some implementations of the present concepts.

DETAILED DESCRIPTION

Overview

This document relates to virtual computing techniques performed on a host device, and more particularly to implementing a guest operating system (“OS”) and/or an application in a different instruction set architecture (“ISA”) than that of the host device. Generally speaking, the host device can have a host OS operating thereon, which can execute in the native ISA of the host device. The host OS can interact directly with the hardware of the host device to provide normal OS functionality. For example, the host OS can provide functionality such as memory management, thread creation, thread synchronization, input/output streams, creating windows and/or dialog boxes, etc.

Generally speaking, the host OS can provide this functionality to various applications via one or more interfaces such as APIs or application binary interfaces (“ABIs”). The host OS interfaces are collectively referred to herein as the “host OS application interface set.” In many cases, applications are configured to make calls to functions provided by a particular OS. If these functions are supported by the host OS application interface set, the application can generally interface directly with the host OS. If not, the application can have limited or no compatibility with the host OS. For example, a Unix® machine may not implement any Windows® application interfaces, and thus a Windows® application typically will not run directly on a Unix® machine. Even different versions of an operating system may deprecate individual functions, implement them in unexpected ways, or otherwise render the operating system incompatible with the application. For example, an application that is configured to run on Windows® 98 may attempt to call various application interfaces that are deprecated in a newer OS such as Windows® 7. This may render the application partially or completely inoperable on a machine that is running Windows® 7.

One way to handle this problem is to implement a guest operating system that runs on the host operating system. The guest OS can also implement a set of application interfaces (API's or ABI's), referred to herein collectively as the “guest OS application interface set.” If the guest operating system implements individual application interface calls as expected by the application, the application can interface with the guest operating system and function properly. This is generally true regardless of whether the host OS application interface set also supports the functions called by the application. However, as mentioned above, running two full copies of different operating systems concurrently can often be quite inefficient. Moreover, this problem can be compounded when the application executes in a different instruction set than the hardware of the host device.

The disclosed implementations provide a guest operating system that can support applications that execute in a different ISA than the host device. The guest OS can interface to an OS virtualization monitor (“OS VM”) that provides a relatively small set of abstractions to the guest OS. These abstractions can be built via individual binary calls from the guest OS to the OS virtualization monitor. Collectively, the individual binary calls implemented by the OS virtualization monitor are referred to herein as the “OS VM interface set” or “VM interface set.”

The guest OS can make calls to implement the guest OS interface set in a manner that is consistent with the expectations of applications that are configured to use individual guest OS interfaces. If the application is in a different instruction set than the host device, the application and the guest OS can execute on a CPU emulator. The guest OS can make calls to the VM interface set to implement the guest OS application interface set as expected by the application. From the perspective of the application, the guest OS is supporting the application via memory management, thread creation, thread synchronization, input/output streams, creating windows and/or dialog boxes, etc. In some implementations, the application can execute on the CPU emulator in the same execution context as the guest OS, e.g., a single user-mode process.

Using the above-described architecture, the application can execute using both a different instruction set and a different OS application interface set than is provided by the underlying host device and OS. Moreover, because the application can execute together with the guest OS in a user-mode process, the guest OS can be implemented in user-mode instructions that share security permissions with user-mode instructions of the application. This can also substantially reduce the resource footprint of the guest OS and CPU emulator, because the guest OS and CPU emulator do not need to provide support for kernel mode instructions in the guest ISA. Instead, only the host OS needs to execute kernel mode instructions in the native ISA.

Example System

For purposes of explanation, consider introductory FIG. 1. FIG. 1 shows an exemplary architecture of a host computing device 100 that is configured to accomplish the concepts described above and below. Host computing device 100 can include a central processing unit (“CPU”) 101 that is operably connected to a memory 102. For example, CPU 101 can be a reduced instruction set computing (RISC) or complex instruction set computing (CISC) microprocessor that is connected to memory 102 via a bus. Memory 102 can be a volatile storage device such as a random access memory (RAM), or a non-volatile memory such as FLASH memory. Although not shown in FIG. 1, host computing device 100 can also include various input/output devices, e.g., a keyboard, a mouse, a display, a printer, etc. Furthermore, the host computing device can include one or more non-volatile storage devices, such as a hard disc drive (HDD), optical (compact disc/digital video disc) drive, tape drive, etc. Generally speaking, any data processed by host computing device 100 can be stored in memory 102, and can also be committed to non-volatile storage. As used herein, the term “computer-readable media” can include transitory and non-transitory instructions. In contrast, the term “computer-readable storage media” excludes transitory instances, and includes volatile or non-volatile storage devices such as random access memory, optical disks, hard drives, flash drives, etc.

Memory 102 of host computing device 100 can include various components that implement certain processing described herein. For example, memory 102 can include a host operating system 103. Generally speaking, host operating system 103 can be a fully functional operating system such as found on a modern personal computer, server, tablet, cell phone, or other computing device. For example, host operating system 103 can be a version of Windows®, Unix® (including related operating systems such as OS X®), Linux® (including related operating systems such as Android®), a real-time and/or embedded operating system, etc. Host operating system 103 can include one or more binary files that are in the native or host ISA for CPU 101.

Memory 102 can also include an operating system virtualization monitor 104 and an execution context 105. Execution context 105 can include a process, thread, task, or other execution context that includes an emulator 106, a guest operating system 107, and an application 108. Note that host OS 103 and/or OS virtualization monitor 104 can generally be implemented via one or more other execution contexts. In some implementations, execution context 105 includes a single address space that is shared by emulator 106, guest OS 107, and application 108. Said another way, execution context 105 includes memory that is allocated specifically for use by any of emulator 106, guest OS 107, and application 108, and that is not available for use by other execution contexts on host computing device 100.

Generally speaking, guest operating system 107 can be configured to provide a guest OS application interface set for applications such as application 108. Operating system virtualization monitor 104 can be configured to provide a set of virtualization interfaces to guest operating system 107, e.g., the OS VM interface set. Host OS 103 can be configured to provide a host OS application interface set to OS VM 104. Generally speaking, emulator 106, OS VM 104, and host OS 103 can be implemented in the host ISA. Application 108 and guest operating system 107 can each include one or more binary files that include instructions in the guest ISA which can be different than the host ISA.

Emulator 106 can be configured to provide instructions that are translated from the guest ISA to the host ISA so that the translated instructions can be executed on CPU 101. For example, emulator 106 can be configured to compile a set of instructions from the guest ISA into the host ISA. Emulator 106 can also be configured to obtain a previously-translated set of instructions in the host ISA. For example, emulator 106 can be configured to track instructions from the guest ISA that have already been compiled into the host ISA and use the previously-compiled instructions rather than recompiling them. Emulator 106 can also interpret and/or just-in-time compile guest ISA instructions into the host ISA, as set forth in more detail below.

In some implementations, CPU 101 can have different levels of privileges for individual instructions. For example, an x86 processor can have certain instructions that can only be executed in a “kernel mode” that are not available to “user mode” programs. In such implementations, host operating system 103 can include binaries that execute in kernel mode, but only allow access to CPU 101 via user mode instructions for execution context 105. Said differently, execution context 105 can be a user mode execution context, whereas host operating system 103 can include at least one kernel mode execution context, e.g., a Windows® system process. More generally, host operating system 103 can include at least one privileged execution context that has at least some higher privileges than execution context 105.

In some implementations, host operating system 103 can implement lower-level operating system functionality such as timers, interrupts, input/output, virtual memory (e.g., accessing a page table maintained by CPU 101), pointer validation, and/or context switching by switching to the privileged execution context, e.g., kernel mode. Host operating system 103 can also implement “rich” operating system functionality via one or more interfaces for application configuration, graphical user interfaces, etc., in a non-privileged execution environment such as a user mode process. Host operating system 103 can provide both the lower-level and rich functionality via the host OS application interface set.

However, because application 108 is generally configured to interface with guest OS 107 instead of host OS 103, application 108 may not be configured to use the host OS application interface set. Instead, guest operating system 107 can provide such operating system functionality via a guest OS application interface set. The guest OS application interface set can include at least some application interfaces that are particular to guest operating system 107 and that may not be provided by host operating system 103. Because these interfaces are provided by guest OS 107, application 108 does not need to be ported, recompiled, or otherwise modified to interface directly with host operating system 103. Moreover, because guest operating system 107 and application 108 are operable on emulator 106, application 108 can also execute binary code that is in a different ISA than the native ISA for CPU 101. Thus, application 108 can run directly within execution context 105 even though application 108 may have originally been built for both a different operating system and a different architecture than host computing device 100.

Note, however, that in some implementations, part or all of guest OS 107 and application 108 can run in a kernel mode process, e.g., execution context 105. OS VM 104 can cause the kernel mode process to appear to application 108 as a user-mode process. For example, OS VM 104 can prevent application 108 from executing any kernel mode instructions, while OS VM 104 can allow guest OS 107 to execute kernel mode instructions.

OS VM 104 can also service certain requests by application 108 with responses that are consistent with application 108 executing in a user-mode process. For example, application 108 can execute an instruction in the guest ISA that reads a register value from emulator 106. The register value can represent the privilege mode in which application 108 executes in the guest ISA. OS VM 104 can respond to application 108 with a value that indicates that application 108 is executing in a non-privileged mode. This can be true even when execution context 105 is actually a privileged execution context provided by host OS 103 and can execute privileged instructions in the host ISA.

As a specific example, consider a scenario where the guest ISA is an x86 instruction set. Application 108 can execute an x86 PUSHFD instruction and receive a response indicating that application 108 is executing in Ring 3 of the x86 ISA. However, execution context 105 (e.g., guest OS 107 and/or emulator 106) may actually have access to Ring 0 x86 functionality. Note that this may also mean that corresponding host ISA instructions can be executed by guest OS 107 and/or emulator 106, but not application 108.

Furthermore, note that, in some implementations, guest operating system 107 does not need to fully implement certain lower-level operating system functionality discussed above with respect to host operating system 103. Rather, guest operating system 107 can be built to use calls to the VM interface set to provide such functionality to application 108. OS VM 104 can translate these calls into corresponding calls from the host OS application interface set and host OS 103 can handle the calls in the host ISA. This implementation can reduce the footprint of guest operating system 107 while still providing the flexibility to support the complete guest OS application interface set.

FIG. 2 shows an exemplary logical architecture of communications between certain components of host computing device 100. Generally speaking, application 108 can include one or more binary files in the guest ISA. These binaries can be translated by emulator 106 into the host ISA and run directly on CPU 101 in the host ISA.

However, some or all of the application binaries may use calls to the guest OS application interface set. These calls are also generally in the guest ISA, but rely on guest OS 107 to perform some functionality for application 108. Guest OS 107 can include one or more binaries in the guest ISA that implement the functions provided via the guest OS application interface set. Some of these binaries can also be translated by emulator 106 into the host ISA and execute directly on CPU 101. For example, in some implementations, guest OS 107 can draw a window for application 108 using only its own translated code via emulator 106. Guest OS 107 may be able to do so because the window can be drawn without using privileged instructions such as kernel mode instructions.

However, in some cases, application 108 may make a guest OS interface call that guest OS 107 cannot satisfy directly. For example, application 108 may request to access a file or create a new thread. Guest OS 107 may need to interface with OS virtualization monitor 104 to implement these operations, because these operations may require privileged (e.g., kernel mode) instructions and/or file system access privileges. Guest OS 107 can access files, create new threads, etc. by making calls to the VM interface set in the guest ISA. Emulator 106 can translate these calls into the host ISA. The translated VM interface calls can be provided to OS VM 104, which can execute them in the host ISA.

In some cases, the VM interface calls are implemented by OS VM 104 via one or more calls to host OS application interfaces. Indeed, some VM interface calls are simply wrappers for the underlying host OS calls, e.g., creating threads or allocating virtual memory. In other cases, the guest OS can add a layer of functionality on top of host OS 103 calls to OS VM 104. To do so, OS VM 104 can make calls to high level services provided by host OS 103, such as file input/output, threading, synchronization, etc.

For example, OS VM 104 can enforce permissions for application 108, guest OS 107, and/or emulator 106. OS VM 104 can do so using a single set of permissions associated with execution context 105 for various operations. Thus, for example, OS VM 104 can prevent application 108 from accessing a particular file by not allowing any instructions from execution context 105 to access the file. This implementation allows guest OS 107 and application 108 to share a common set of permissions.

Guest OS 107 can also provide device support for devices such as displays, printers, keyboards, mice, etc. via a remote desktop protocol (“RDP) instead of using a conventional device driver. Guest OS 107 can do so by receiving calls to the guest OS application interface set from application 108 and communicating with the devices via the RDP protocol. The RDP protocol can be implemented via communication pipes that are routed through OS VM 104 to host OS 103. Host OS 103 can receive the RDP calls and control the devices via device drivers that are part of host OS 103. Note that the communication pipes can cross machine boundaries, e.g., over a network or other type of communication link.

FIG. 3 shows an exemplary architecture of emulator 106. Generally speaking, emulator 106 can use different techniques to obtain translated binaries for guest operating system 107 and/or application 108. For example, CPU emulator 106 can include a precompiler 301, a just-in-time (“JIT”) compiler 302, and/or an interpreter 303.

Using precompiler 301, CPU emulator 106 can translate binary files of guest OS 107 and/or application 108 from the guest ISA to the host ISA. Generally speaking, precompiler 301 can translate the binaries before the binaries are invoked for execution, e.g., precompiler 301 can translate binaries for guest OS 107 offline before guest OS 107 is loaded into execution context 105. Likewise, precompiler 301 can translate the binaries for application 108 before application 108 is loaded into execution context 105. For example, precompiler 301 can load one or more binaries of application 108 and/or guest OS 107 from storage into memory 102, translate them from the guest ISA to the host ISA, and store the corresponding host ISA binaries for later usage.

Under some circumstances, precompilation is not necessarily suitable. For example, a user may request to initiate application 108 and/or guest operating system 107 on host computing device 100 before precompiled binaries in the host ISA are obtained by emulator 106. As another example, application 108 may generate code at runtime in the guest ISA which needs to be translated into the host ISA while application 108 and guest OS 107 are already running. Under such circumstances, the binaries can be translated at runtime by just-in-time compiler 302 and/or interpreter 303.

Generally speaking, just-in-time compiler 302 compiles a group of instructions (binary) in the guest ISA the first time they are executed, caches the compiled instructions as one or more binaries in the corresponding host ISA, then executes the cached host ISA binaries each time the corresponding binaries in the guest ISA are invoked afterwards. In contrast, interpreter 303 translates guest ISA instructions into host ISA sequentially and generally does not compile or cache the translated host ISA instructions. Rather, each time the guest ISA instructions are invoked, interpreter 303 performs the translation and executes the host ISA instructions, but does not retain the translated host ISA instructions for later use.

Note that, in some implementations, the translation techniques mentioned above can be performed using source files instead of binaries. For example, one or more source files of guest OS 107 and/or application 108 can be compiled and/or linked by precompiler 301, just-in-time compiler 302, and/or interpreter 303 instead of performing the translation of binaries. Generally speaking, however, many applications and operating systems are distributed only as binaries and the techniques disclosed herein can be performed without access to source code for either guest OS 107 or application 108.

Note that there may be certain performance trade-offs associated with which of the above-mentioned techniques is used for translation. When possible, precompiling can provide good performance because the host ISA binaries are available when application 108 and/or guest OS 107 begins execution. This is because there is no need to translate the guest ISA instructions at runtime, the precompilation process can be performed offline prior to their execution. Furthermore, the precompiled binaries can be resident in shared memory. This can be particularly useful in circumstances where multiple applications execute the same binary code on emulator 106, because the applications themselves may have different address spaces in their respective execution contexts. By storing the precompiled host ISA binary in shared memory, the applications sharing the host ISA binary do not need to each have a copy of the host ISA binary in the address space for their respective execution contexts. Rather, the host ISA binary can be shared between multiple execution contexts, each of which can include an application and/or guest OS.

Just-in-time compiling, on the other hand, generally occurs while application 108 and/or guest OS 107 is already executing. Moreover, just-in-time compiling generally results in a host ISA binary that is stored in the private address space for the application, e.g., in memory that is allocated to the application's execution context. The first time a given set of instructions is executed by application 108, the instructions are compiled into the host ISA. This can result in a certain amount of latency because the compilation can use a substantial amount of computational resources, e.g., heavy processor loading on CPU 101, substantial usage of memory 102, and/or the time to read the guest ISA binary from storage into memory 102.

However, when using just-in-time compiler 302, once a given set of instructions has been compiled into the host ISA, they do not need to be recompiled each time the corresponding guest ISA instructions are invoked. In contrast, interpreter 303 generally translates guest ISA instructions sequentially and does not cache translated instructions. This can prevent the one-time latency issue mentioned above for just-in-time compilation. However, because the host ISA instructions are not cached for later use by interpreter 303, the host ISA instructions need to be generated each time the corresponding guest ISA instructions are invoked. Also, many interpreters produce code that is not optimized or is otherwise relatively inefficient in comparison to code that is compiled by precompiler 301 and/or just-in-time compiler 302.

Emulator 106 can also maintain a translation table 304. For example, translation table 304 can be used by emulator 106 to check whether a particular set of guest ISA instructions have been translated into host ISA instructions and loaded for execution. By doing this check before translating the guest ISA instructions with precompiler 301, just-in-time compiler 302, or interpreter 303, emulator 106 can avoid unnecessary compilation overhead. Note that the term “table” is used generically to refer to many different data structures and that translation table 304 can be implemented using arrays, structures, classes, linked lists, trees, etc.

FIG. 4A-4C collectively illustrate an exemplary translation table 304 as it is populated by emulator 106 over time. Generally speaking, FIG. 4A illustrates translation table 304 prior to invoking application 108, FIG. 4B illustrates translation table 304 after application 108 loads a first binary file (App_DLL_1), and FIG. 4C illustrates translation table 304 after application 108 loads a second binary file (App_DLL_2). For the purposes of the following example, guest OS 107 has two binary files, GuestOS_DLL_1 and GuestOS_DLL_2, that are already precompiled and loaded in execution context 105 at the beginning of the following example.

As shown in FIG. 4A, translation table 304 can include a file name column 401, a guest ISA address column 402, and a host ISA address column 403. File name column 401 is provided to illustrate logical names for the binaries provided by guest operating system 107 and/or application 108 discussed herein. For example, guest operating system 107 can provide interfaces to functions that create and close a window via GuestOS_DLL_1, and interfaces to functions that create or manage virtual memory via GuestOS_DLL_2. These binaries can have addresses in the guest ISA shown at guest ISA address 402. The binaries can be translated into the host ISA, e.g., by precompiler 301, prior to executing guest OS 107 and/or application 108 on host computing device 100. The corresponding host ISA binaries can be loaded into memory and have host ISA addresses as shown in host ISA address column 403.

Application 108 can make calls to the guest OS application interface set by linking to the guest OS binaries at runtime. From the perspective of application 108, these binaries are located at the guest ISA addresses at column 402. When they are loaded by emulator 106, however, emulator 106 updates translation table 304 with the corresponding host ISA address 403. Note that host ISA address 403 can be a virtual memory address within the address space allocated to execution context 105, and may actually correspond to different physical addresses in memory 102. In such implementations, CPU 101 and/or host OS 103 can translate the virtual addresses to actual physical addresses on memory 102.

As described below, each time a load request for a guest ISA binary is received, emulator 106 can check translation table 304 to determine whether the corresponding host ISA binary has been previously compiled and loaded into memory. If so, emulator 106 does not need to translate the binary into the host ISA. Instead, emulator 106 can simply wait until the binary at the corresponding host ISA address 403 is executed and run the translated code at that time.

FIG. 4B illustrates translation table 304 in a configuration after application 108 is initiated and requests to load App_DLL_1 into execution context 105. Application 108 can load this binary via guest ISA instructions that reference the guest ISA address for App_DLL_1, e.g., 0x0F001100. Emulator 106 receives the file load operation from application 108 (e.g., via a call from application 108 to guest OS 107) and can check translation table 304 to determine whether a corresponding host ISA binary has already been loaded.

If the host ISA binary has already been loaded, the binary is ready to execute. If the host ISA binary has already been compiled but not loaded, e.g. precompiled and available in storage, them emulator 106 can load the precompiled binary and update translation table 304. If the host ISA binary is not yet available on storage, emulator 106 can wait until execution time and use just-in-time compiler 302 to translate App_DLL_1 into the host ISA. At that time, emulator 106 can make a new entry in translation table 304 that includes the guest ISA address for App_DLL_1 as well as the corresponding host ISA address where the translated binary is loaded. Then, emulator 106 will not need to recompile App_DLL_1 if application 108 attempts to load this file again. Rather, emulator 106 can receive the guest ISA address for this file from application 108, determine that there is a corresponding host ISA address that is already loaded with a host ISA binary by checking translation table 304, and later execute the corresponding host ISA binary without needing to translate the file into the host ISA again.

FIG. 4C illustrates translation table 304 in a configuration after application 108 loads a binary file entitled App_DLL_2. As discussed above, the first time App_DLL_2 is loaded, an entry is made into translation table 304. The binary can be translated at execution time by just-in-time compiler 302 or can be precompiled. Upon loading the translated binary, emulator 106 makes another new entry in translation table 304 that includes the guest ISA address for App_DLL_2, as well as the corresponding host ISA address for the corresponding translated binary.

To the extent feasible, it can be useful to precompile as many binaries as possible for guest OS 107 and/or application 108. For example, if host computing device 100 is intended to permanently support the guest OS 107, it can be desirable to precompile all of the binaries for the guest OS, or at least those binaries that are most likely to be used at runtime by applications. Then, the binaries are available to be loaded and updated in translation table 304 responsive to receiving the corresponding guest ISA load instruction. Likewise, if application 108 is expected to remain on host computing device 100 for any sustained period of time, it can also be desirable to precompile all binaries associated with application 108. In some implementations, emulator 106 and/or OS VM monitor 104 can track usage statistics associated with individual guest OS binaries and/or application binaries and precompile more frequently-used binaries while leaving other binaries to be compiled at runtime by just-in-time compiler 302 and/or interpreter 303. This can reduce the amount of memory required for execution context 105 except in instances where the infrequently-used binaries are actually loaded by application 108 and/or guest OS 107.

However, in some implementations, it may be more feasible or desirable to not precompile all, or even any, of the binaries associated with guest OS 107 and/or application 108. For example, if the execution of guest OS 107 and/or application 108 is intended to be temporary, e.g., a single use, it can be more efficient to interpret and/or just-in-time compile all of the binaries into the host ISA.

Furthermore, certain binaries may be particular to certain hardware. For example, sometimes two different processors can implement the same basic host ISA. One of the processors may use an arithmetic logic unit (“ALU”) to emulate floating point operations with integer operations, while the other may have an additional piece of hardware, e.g., a floating-point unit, that accepts certain floating-point instructions that are not in the basic host ISA. Under such circumstances, it is generally not desirable to compile binaries with the floating point instructions for the processor that does not have a floating point unit. Instead, these binaries would generally be compiled with the corresponding integer operations to implement the emulated floating point operations.

To accommodate these distinctions, emulator 106 can be configured to precompile all of the binaries for application 108 and/or guest OS 107 without the floating point instructions. At runtime, emulator 106 can perform a check to see whether CPU 101 includes the floating point unit. If so, binaries that include floating point operations can be just-in-time compiled or interpreted at runtime instead of executing the precompiled binaries. Otherwise, the precompiled floating point emulation integer operations can be executed instead. Alternatively, the binaries can be precompiled both with and without the floating point instructions ahead of time, and emulator 106 can load the appropriate binary at runtime after determining whether CPU 101 supports the floating point instructions.

Furthermore, note that guest operating system 107 may be substantially reduced in size relative to a “typical” full operating system. In some implementations, guest operating system 107 may even have a smaller memory, disk, and/or processor footprint than application 108. Accordingly, in some implementations, guest operating system 107 and/or application 108 can be configured together in an installer that installs both guest OS 107 and application 108 together on host computing device 100. In such implementations, the installer may check to ensure that host computing device 100 has an appropriate emulator 106 for the guest ISA before going ahead with the installation. The installer can also check CPU 101 to see whether CPU 101 supports certain instructions, such as the aforementioned floating point instructions, and install different binaries for application 108 and/or guest OS 107 depending on whether the instructions are supported.

Using the techniques introduced above, it is possible to support applications configured for various instruction sets and operating systems concurrently on host computing device 100. FIG. 5 illustrates memory 102 in such a configuration. As shown in FIG. 5, another execution context 501 can reside in memory 102, and include another guest operating system 503 as well as another emulator 504. Guest operating system 503 can be the same operating system as guest OS 107, e.g., can provide the same application interface set, or can be a different operating system. Likewise, emulator 504 can support the same ISA as emulator 106 or a different ISA.

Memory 102 can also include an execution context 505 with an application 506 and a guest OS 507. Note, however, that execution context 505 is shown without an emulator. Execution context 505 may be appropriate when guest OS 507 and application 506 are implemented in the host ISA, but application 506 is configured for guest OS 507 instead of host OS 103. Each of execution context 105, 501, and 505 can be configured to communicate to host OS 103 via OS virtualization monitor 104.

Execution context 508 can include an application 509. Generally speaking, execution context 508 may be appropriate for applications that are in the host ISA and are implemented in the host ISA. Note that, because application 509 is configured for host OS 103, execution context 508 may communicate directly with host OS 103 instead of with OS virtualization monitor 104.

FIG. 5 illustrates several aspects of the disclosed implementations. First, note that each application can run in a different execution context, e.g., with a different address space. Moreover, when a guest operating system and/or emulator is needed to support a given application, the guest OS and/or emulator can be loaded into the execution context with the application. This allows the application, guest OS, and/or emulator within a given execution context to share a common set of permissions. For example, execution context 105 may have read-only access to a group of files. OS VM 104 can receive host ISA instructions from emulator 106 and enforce the permissions so that none of application 108, guest OS 107, and/or emulator 106 improperly writes to one of these files. From the perspective of OS VM 104 and/or host OS 103, it does not matter whether the application, guest OS, or CPU emulator is attempting an impermissible action.

Furthermore, using OS VM 104 can allow different emulators and/or guest operating systems the flexibility to run on other host devices with different underlying host operating systems and/or instruction sets. This is because OS VM 104 provides a layer of abstraction between the emulators/guest OS's and the underlying host device. Generally speaking, the guest OS and/or emulators use calls to the OS VM interface set instead of directly using the host OS application interface set. Thus, all that is needed for a given emulator and/or guest OS to run on a different host OS is an OS VM monitor that properly implements the OS VM interface set. Even if the other host device uses a different instruction set, the emulator and/or OS can simply be recompiled to the ISA for the host device where they will run.

The following is an exemplary interface set that can be provided by OS VM 104:

// Virtual Memory
DKSTATUS
DkVirtualMemoryAllocate(
inout PVOID *BaseAddress,
inout PSIZE_T  RegionSize,
in ULONG  AllocationType,
in ULONG  Protect);
DKSTATUS
DkVirtualMemoryFree(
in PVOID BaseAddress,
in SIZE_T RegionSize,
in ULONG FreeType);
DKSTATUS
DkVirtualMemoryProtect(
inout PVOID BaseAddress,
inout SIZE_T RegionSize,
in ULONG NewProtect,
out PULONG Old Protect);
// IPC
BOOL
DkPipeFork(
in HANDLE Handle,
out PULONG64 Token,
out PHANDLE NewHandle);
BOOL
DkSelfPipeCreate(
out PHANDLE Handle1,
out PHANDLE Handle2,
out PULONG64 Token);
ULONG
DkPipeRead(
in HANDLE Handle,
in BOOL Async,
in PVOID AsyncToken,
inout PVOID *Buffer,
in ULONG Length,
in_opt PLONG64 Timeout);
ULONG
DkPipeWrite(
in HANDLE Handle,
in BOOL Async,
in PVOID AsyncToken,
in PVOID Buffer,
in ULONG Length);
ULONG
DkPipeSelect(
in ULONG Count,
in const HANDLE *Handles,
in_opt PLONG64 Timeout);
ULONG
DkPipePeek(
in HANDLE Handle);
// Isolated File Access
PVOID
DkFileOpen(
in PUNICODE_STRING pUri,
in_opt PVOID DesiredAddress,
in ACCESS_MASK DesiredAccess,
in ULONG ShareMode,
in ULONG CreateDisposition,
in ULONG CreateOptions,
in SIZE_T Offset,
inout_opt PSIZE_T ViewSize);
BOOL
DkFileTruncate(
in PUNICODE_STRING Uri,
in SIZE_T Length);
DKSTATUS
DkFileUnmap(
in PVOID addr);
BOOL
DkFileSync(
in PVOID addr);
BOOL
DkFileUnlink(
in PUNICODE_STRING Uri);
DKSTATUS
DkFileAttributesQuery(
in PUNICODE_STRING Uri,
out PDK_FILE_ATTRIBUTES Attrs);
// Threading
BOOL
DkThreadCreate(
in SIZE_T StackSize,
in PDK_THREAD_START Address,
in_opt PVOID Parameter,
in ULONG CreationFlags,
out_opt PHANDLE Pipe,
out_opt PULONG64 PipeToken);
VOID
DkThreadExit( );
BOOL
DkProcessCreate(
in_opt PUNICODE_STRING Appl,
in_opt PUNICODE_STRING CmdLin,
out_opt PHANDLE Pipe,
out_opt PULONG64 PipeToken);
VOID
DkProcessExit( );
// Other
BOOL
DkSystemTimeQuery(
out PLONG64 SystemTime);
BOOL
DkRandomBitsRead(
in out PVOID Buf,
in SIZE_T BufSize);
BOOL
DkDebugOutput(
in PUNICODE_STRING Message);

As mentioned above, each guest OS can use calls to the VM interface set instead of implementing the guest OS functionality directly. In some cases, a guest OS application interface is a simple wrapper for an underlying VM interface call. This can be the case for virtual memory allocation, thread creation, etc. For example, guest OS 107 and guest OS 503 may use different calling conventions for their respective memory allocation interfaces, e.g., different number, order, or names of parameters, etc. However, both guest OS 107 and guest OS 503 can implement virtual memory allocation with calls to DkVirtualMemoryAllocate.

Other guest OS functionality may be built using non-trivial binaries that still can utilize the underlying VM interfaces. For example, files, locks, timers, etc., can all be built inside of each individual guest OS. Furthermore, recall that each guest OS can also have one or more binaries that execute directly and do not necessarily use calls to the VM interface set. For example, guest OS 107 and guest OS 503 can support different interface calls for creating graphical windows. These interface calls may be implemented directly by executing the corresponding guest OS binaries which may not involve directly calling the VM interface set. In other words, unmodified guest OS binaries may be executed directly instead of modifying the guest OS to use calls to the VM interface set.

In some implementations, this is accomplished by providing modified versions of some binaries that would normally be associated with the guest OS. For example, a Windows® guest OS may include a binary called comctl32.dll that provides various graphical features such as toolbars and tabs. Comctl32.dll may, in turn, rely on calls to another binary, kernelbase.dll. Typically, guest OS 107 is distributed with both of these.dll files. In some disclosed implementations, comctl32.dll runs in execution context 105, and calls from comctl32.dll are passed to a customized version of kernelbase.dll that runs in OS VM 104 to support guest OS 107. This version of kernelbase.dll may rely on calls to the VM interface set. This is a particular example of how user mode operating system functionality can be moved into an execution context along with a supported application and an appropriate emulator.

In some implementations, applications can migrate across device boundaries to execute on a different host device. If the other host device implements the VM interface set and the host OS thereon supports the RDP protocol, this can be accomplished by copying the address space and execution context 105 over to the other host device. For example, one or more pages of memory used by guest OS 107 and/or application 108 as well as register values or other state information can be provided to the other host device.

FIG. 6 illustrates a system 600 that is configured to accomplish migrating an application. System 600 can include host computing device 100 connected to a network 610. Host computing device can communicate with another host computing device 620 via network 610.

Although not shown in FIG. 6, host computing device 620 can have similar components as those set forth above with respect to host computing device 100. However, host computing device 620 can have a different native ISA than host computing device 100, and can also have a different host OS than host OS 103. If host computing device 620 correctly implements the VM interface and the RDP protocol, application 108 can be copied directly to host device 620 as shown in FIG. 6. For example, at a first time 601, execution context can be executing on host computing device 100 as discussed above. At any time, the execution state of application 108 and/or guest OS 107 can be shared by transmitting execution context 105 over network 610, as shown at a second, subsequent time 602. Host computing device 620 can receive execution context 105, instantiate a new execution context thereon, and copy execution context 105 into the new execution context. Host computing device 620 can have an emulator 621 that can function as discussed above and below with respect to emulator 106. In some implementations, state information (e.g., register values, program counter, etc.) for emulator 106 can be copied to host device 620 and used to initialize emulator 621. This may be particularly useful when host computing device 100 and host computing device 620 share a common ISA that is different than the ISA for application 108.

In the manner described above, application 108 can seamlessly transition across different ISA's and host operating systems in mid-execution. Note that, in some implementations, host computing device 620 can be preconfigured with precompiled binaries for guest OS 107. This can reduce the amount of time it takes for application 108 to be migrated.

Method Implementations

FIG. 7 illustrates a method 700 that is suitable for implementation on host computing device 100 or by other devices or systems. Generally speaking, method 700 relates to loading guest operating systems and/or emulators to support a request to run an application.

A request to start an application can be received at block 701. For example, OS virtualization monitor 104 can receive a request to run an application such as application 108, 502, 506, and/or 509.

Application metadata can be checked at block 702. For example, OS VM 104 can check a manifest associated with application 108 to determine what ISA and/or operating system application 108 is configured to run. In some implementations, the application manifest can be written in a markup language such as extensible markup language (“XML”).

If the application manifest indicates that the application is in the host ISA and is configured to run on the host operating system, method 700 can move to block 703. At block 703, the application can run directly on the host OS without an emulator. For example, OS VM 104 can pass the application to host OS 103. Host OS 103 can create a new execution context and run the application directly within the context. This configuration is generally illustrated by execution context 508 in FIG. 5.

If the application manifest indicates that the application is in the host ISA but is configured to run on a different operating system than the host OS, method 700 can move to block 704. At block 704, the appropriate guest OS is loaded into an execution context with the application. Next, at block 705, the application is executed in the execution context with the guest OS. This configuration is generally illustrated by execution context 505 in FIG. 5.

If the application manifest indicates that the application is in a different ISA than the host ISA and is also configured to run on a different operating system than the host OS, method 700 can move to block 706. At block 706, the appropriate emulator is loaded into an execution context. Next, at block 707, the appropriate guest OS is loaded into the execution context with the emulator. Next, at block 708, the application can run on the guest OS in the guest ISA in the execution context with the emulator and the guest OS. This configuration is generally illustrated by execution contexts 105 and 501 in FIG. 5.

FIG. 8 illustrates a method 800 that is suitable for implementation on host computing device 100 or by other devices or systems. Generally speaking, method 800 relates to runtime support of an application by an emulator that can load precompiled code when the precompiled code is available.

A request to load a file is received at block 801. For example, emulator 106 can receive a request from application 108 and/or guest OS 107 to load a DLL or other binary file at runtime. The request can include one or more instructions in the guest ISA.

A translation table is checked at decision block 802. For example, emulator 106 can check translation table 304 using the guest ISA address of the file that application 108 and/or guest OS 107 is requesting to load. This enables emulator 106 to see if the file has already been loaded into memory, e.g., execution context 105.

If the file has already been compiled into the host ISA and loaded into memory, method 800 moves to block 803. At block 803, the file is in memory and ready to execute.

If the file has not yet been loaded (e.g., has no corresponding host ISA address in translation table 304) and has yet to be compiled (e.g., a compiled version is not available in storage), method 800 moves to block 804. At block 804, the method waits to compile the file until the file is executed. For example, in the case of a .dll file, the method can wait until at least one function from the .dll is called and then compile the file with just-in-time compiler 302 and/or interpreter 303.

If the file has already been compiled but has not yet been loaded (e.g., a compiled binary is available in storage but the file has no corresponding host ISA address in translation table 304), method 800 moves to block 805. At block 805, the file is loaded into memory. Next, at block 806, the table is updated with the address where the compiled host ISA version of the file is loaded. Next, at block 807, the file is ready to execute.

FIG. 9 illustrates a method 900 that is suitable for implementation on host computing device 100 or by other devices or systems. Generally speaking, method 900 relates to runtime support of an application by an emulator that can use precompiled code, interpreted code, and/or just-in-time compiled code.

A request to execute code is received at block 901. For example, emulator 106 can receive a request from application 108 and/or guest OS 107 to execute code at runtime. The request can include one or more instructions in the guest ISA.

A translation table is checked at decision block 902. For example, emulator 106 can check translation table 304 using the guest ISA address of the code that application 108 and/or guest OS 107 is executing. Emulator 106 can thus determine whether the code that will execute is included in a file that has already been loaded into memory.

If the code has already been loaded into memory, method 900 moves to block 903. At block 903, the code is executed.

If the code has not been loaded and just-in-time compiling is not supported or otherwise not being used, method 900 moves to block 904. At block 904, the code is interpreted into the host ISA. At block 905, the interpreted code is executed.

If the code has not yet been loaded into memory and just-in-time compiling is supported, method 900 moves to block 906. At block 906, the code that will be executed is compiled into the host ISA and loaded into memory. Next, at block 907, the translation table is updated to reflect the loaded binary is loaded in memory. Next, at block 908, the code is executed.

Note that the above discussion is largely illustrated at a file level of granularity. In other words, translation table 304 includes host ISA addresses for individual .dll files. However, note that each file can include multiple functions as well as data. In some implementations, more refined granularities can be used. For example, addresses of each function in a file can be tracked individually rather than tracking the file as a whole.

CONCLUSION

Although techniques, methods, devices, systems, etc., pertaining to the above implementations are described in language specific to structural features and/or methodological acts, it is to be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or acts described. Rather, the specific features and acts are disclosed as exemplary forms of implementing the claimed methods, devices, systems, etc.

Claims

1. A host computing device comprising: at least one processing device configured to implement a host instruction set architecture (ISA) that is native to the host computing device; andat least one computer-readable storage medium storing host ISA instructions,wherein the at least one computer-readable storage medium includes a memory, andwherein the host ISA instructions, when executed by the at least one processing device, cause the at least one processing device to:maintain a translation data structure indicating whether guest ISA binaries have been translated into host ISA binaries and loaded into the memory, the translation data structure mapping guest ISA addresses used by a guest operating system or a guest application to reference the guest ISA binaries to host ISA addresses of the host ISA binaries in the memory, the guest ISA binaries being in a guest ISA that is not native to the host computing device;receive a request from the guest operating system or the guest application to load a particular guest ISA binary into the memory;check the translation data structure to determine whether a particular host ISA binary corresponding to the particular guest ISA binary has already been loaded into the memory;in a first instance when the particular host ISA binary corresponding to the particular guest ISA binary has already been loaded into the memory, execute the particular host ISA binary upon request; andin a second instance when the particular guest ISA binary has already been translated into the particular host ISA binary in the host ISA and the particular host ISA binary is not yet loaded into the memory: load the particular host ISA binary into the memory;update the translation data structure with a record indicating that the particular host ISA binary has been translated and is located in the memory at a particular host ISA address, the record having a particular guest ISA address used by the guest operating system or the guest application to reference the particular guest ISA binary; andexecute the particular host ISA binary upon request.

2. The host computing device of claim 1, wherein the host ISA instructions comprise an emulator configured to translate the guest ISA binaries from the guest ISA to the host ISA to obtain the host ISA binaries.

3. The host computing device of claim 2, wherein the emulator comprises a just-in-time compiler configured to: in a third instance when the particular guest ISA binary is not loaded into the memory and has not already been translated, defer translation of the particular guest ISA binary until a subsequent request to execute the particular guest ISA binary is received.

4. The host computing device of claim 2, the emulator further comprising a precompiler configured to translate an individual guest ISA binary in advance of receiving another request to load the individual guest ISA binary.

5. The host computing device according to claim 2, the particular guest ISA binary being part of the guest application.

6. The host computing device according to claim 5, the particular guest ISA binary comprising a shared library that is linked at runtime.

7. The host computing device according to claim 2, the particular guest ISA binary being part of the guest operating system, the host computing device having a host operating system other than the guest operating system.

8. The host computing device of claim 7, wherein another guest ISA binary is part of the guest application, the another guest ISA binary having a call to an interface provided by the guest operating system.

9. The host computing device of claim 8, wherein the particular guest ISA binary that is part of the guest operating system implements the interface.

10. A method performed by a computing device having a native instruction set architecture (ISA), the method comprising: maintaining a translation data structure indicating whether guest ISA binaries have been translated into native ISA binaries and loaded into memory, the translation data structure mapping guest locations of the guest ISA binaries to corresponding host locations of the native ISA binaries, wherein the guest ISA binaries are provided in a guest ISA that is different than the native ISA and the guest locations are used by guest code in the guest ISA to reference the guest ISA binaries;receiving a request from the guest code to execute an individual guest ISA binary;checking the translation data structure to determine whether an individual native ISA binary corresponding to the individual guest ISA binary has already been loaded into memory; andin a first instance when the individual native ISA binary has not already been loaded into memory: interpreting or compiling the individual guest ISA binary to obtain the individual native ISA binary;loading the individual native ISA binary into memory at an individual host location;updating the translation data structure with a record mapping an individual quest location of the individual quest ISA binary to the individual host location of the individual native ISA binary in memory; andexecuting the individual native ISA binary in response to the request; andin a second instance when the individual native ISA binary has already been loaded into memory, executing the individual native ISA binary in response to the request.

11. The method of claim 10, further comprising: compiling the individual guest ISA binary with a just-in-time compiler.

12. The method of claim 10, wherein the native ISA is a reduced instruction set architecture and the guest ISA is a complex instruction set architecture.

13. The method of claim 10, wherein the native ISA is a complex instruction set architecture and the guest ISA is a reduced instruction set architecture.

14. The method of claim 10, wherein the translation data structure comprises a translation data table and the updating comprises: after interpreting or compiling the individual guest ISA binary, adding a new row to the translation data table, the new row comprising the record mapping the individual guest location of the individual guest ISA binary to the individual host location of the individual native ISA binary.

15. The method of claim 14, wherein the translation data table comprises a column that includes identifiers of the guest ISA binaries.

16. The method of claim 15, wherein individual rows of the translation data table map different guest locations of different guest ISA binaries to different host locations.

17. The method of claim 16, wherein the translation data table includes separate columns identifying the different guest locations and the different host locations.

18. A method performed by a computing device, the method comprising: receiving multiple requests to execute guest files;in response to the multiple requests, translating the guest files into native binaries and loading the native binaries into memory;updating a translation data structure with records indicating that the guest files have been translated, the translation data structure having mappings of guest locations of the guest files to corresponding host locations of the native binaries in memory, the guest locations being used by guest code to reference the guest files;executing the native binaries;receiving subsequent requests to execute individual guest files; andin response to the subsequent requests: in first instances when the translation data structure includes records indicating that first native binaries for first guest files remain loaded in memory, executing the first native binaries without recompilation and without reloading into memory; andin second instances when the translation data structure lacks records indicating that second native binaries for second guest files are compiled and loaded in memory and the second native binaries are not compiled and available in storage: compiling the second guest files to obtain the second native binaries;loading the second native binaries into memory;updating the translation data structure with records indicating that the second native binaries have been compiled and loaded into memory; andexecuting the second native binaries.

19. The method of claim 18, wherein the guest files are guest binaries in a guest instruction set architecture and the native binaries are in a native instruction set architecture of the computing device.

20. The method of claim 19, wherein the guest files include both guest application files that are part of a guest application and guest operating system files that are part of a guest operating system, the computing device having a host operating system other than the guest operating system.