File:  [Qemu by Fabrice Bellard] / qemu / qemu-tech.texi
Revision (vendor branch): download - view: text, annotated - select for diffs
Tue Apr 24 17:34:46 2018 UTC (3 years, 3 months ago) by root
Branches: qemu, MAIN
CVS tags: qemu0125, qemu0124, qemu0123, qemu0122, qemu0121, qemu0120, HEAD
qemu 0.12.0

    1: \input texinfo @c -*- texinfo -*-
    2: @c %**start of header
    3: @setfilename
    4: @settitle QEMU Internals
    5: @exampleindent 0
    6: @paragraphindent 0
    7: @c %**end of header
    9: @iftex
   10: @titlepage
   11: @sp 7
   12: @center @titlefont{QEMU Internals}
   13: @sp 3
   14: @end titlepage
   15: @end iftex
   17: @ifnottex
   18: @node Top
   19: @top
   21: @menu
   22: * Introduction::
   23: * QEMU Internals::
   24: * Regression Tests::
   25: * Index::
   26: @end menu
   27: @end ifnottex
   29: @contents
   31: @node Introduction
   32: @chapter Introduction
   34: @menu
   35: * intro_features::        Features
   36: * intro_x86_emulation::   x86 and x86-64 emulation
   37: * intro_arm_emulation::   ARM emulation
   38: * intro_mips_emulation::  MIPS emulation
   39: * intro_ppc_emulation::   PowerPC emulation
   40: * intro_sparc_emulation:: Sparc32 and Sparc64 emulation
   41: * intro_other_emulation:: Other CPU emulation
   42: @end menu
   44: @node intro_features
   45: @section Features
   47: QEMU is a FAST! processor emulator using a portable dynamic
   48: translator.
   50: QEMU has two operating modes:
   52: @itemize @minus
   54: @item
   55: Full system emulation. In this mode (full platform virtualization),
   56: QEMU emulates a full system (usually a PC), including a processor and
   57: various peripherals. It can be used to launch several different
   58: Operating Systems at once without rebooting the host machine or to
   59: debug system code.
   61: @item
   62: User mode emulation. In this mode (application level virtualization),
   63: QEMU can launch processes compiled for one CPU on another CPU, however
   64: the Operating Systems must match. This can be used for example to ease
   65: cross-compilation and cross-debugging.
   66: @end itemize
   68: As QEMU requires no host kernel driver to run, it is very safe and
   69: easy to use.
   71: QEMU generic features:
   73: @itemize
   75: @item User space only or full system emulation.
   77: @item Using dynamic translation to native code for reasonable speed.
   79: @item
   80: Working on x86, x86_64 and PowerPC32/64 hosts. Being tested on ARM,
   81: HPPA, Sparc32 and Sparc64. Previous versions had some support for
   82: Alpha and S390 hosts, but TCG (see below) doesn't support those yet.
   84: @item Self-modifying code support.
   86: @item Precise exceptions support.
   88: @item The virtual CPU is a library (@code{libqemu}) which can be used
   89: in other projects (look at @file{qemu/tests/qruncom.c} to have an
   90: example of user mode @code{libqemu} usage).
   92: @item
   93: Floating point library supporting both full software emulation and
   94: native host FPU instructions.
   96: @end itemize
   98: QEMU user mode emulation features:
   99: @itemize
  100: @item Generic Linux system call converter, including most ioctls.
  102: @item clone() emulation using native CPU clone() to use Linux scheduler for threads.
  104: @item Accurate signal handling by remapping host signals to target signals.
  105: @end itemize
  107: Linux user emulator (Linux host only) can be used to launch the Wine
  108: Windows API emulator (@url{}). A Darwin user
  109: emulator (Darwin hosts only) exists and a BSD user emulator for BSD
  110: hosts is under development. It would also be possible to develop a
  111: similar user emulator for Solaris.
  113: QEMU full system emulation features:
  114: @itemize
  115: @item
  116: QEMU uses a full software MMU for maximum portability.
  118: @item
  119: QEMU can optionally use an in-kernel accelerator, like kvm. The accelerators 
  120: execute some of the guest code natively, while
  121: continuing to emulate the rest of the machine.
  123: @item
  124: Various hardware devices can be emulated and in some cases, host
  125: devices (e.g. serial and parallel ports, USB, drives) can be used
  126: transparently by the guest Operating System. Host device passthrough
  127: can be used for talking to external physical peripherals (e.g. a
  128: webcam, modem or tape drive).
  130: @item
  131: Symmetric multiprocessing (SMP) even on a host with a single CPU. On a
  132: SMP host system, QEMU can use only one CPU fully due to difficulty in
  133: implementing atomic memory accesses efficiently.
  135: @end itemize
  137: @node intro_x86_emulation
  138: @section x86 and x86-64 emulation
  140: QEMU x86 target features:
  142: @itemize
  144: @item The virtual x86 CPU supports 16 bit and 32 bit addressing with segmentation.
  145: LDT/GDT and IDT are emulated. VM86 mode is also supported to run
  146: DOSEMU. There is some support for MMX/3DNow!, SSE, SSE2, SSE3, SSSE3,
  147: and SSE4 as well as x86-64 SVM.
  149: @item Support of host page sizes bigger than 4KB in user mode emulation.
  151: @item QEMU can emulate itself on x86.
  153: @item An extensive Linux x86 CPU test program is included @file{tests/test-i386}.
  154: It can be used to test other x86 virtual CPUs.
  156: @end itemize
  158: Current QEMU limitations:
  160: @itemize
  162: @item Limited x86-64 support.
  164: @item IPC syscalls are missing.
  166: @item The x86 segment limits and access rights are not tested at every
  167: memory access (yet). Hopefully, very few OSes seem to rely on that for
  168: normal use.
  170: @end itemize
  172: @node intro_arm_emulation
  173: @section ARM emulation
  175: @itemize
  177: @item Full ARM 7 user emulation.
  179: @item NWFPE FPU support included in user Linux emulation.
  181: @item Can run most ARM Linux binaries.
  183: @end itemize
  185: @node intro_mips_emulation
  186: @section MIPS emulation
  188: @itemize
  190: @item The system emulation allows full MIPS32/MIPS64 Release 2 emulation,
  191: including privileged instructions, FPU and MMU, in both little and big
  192: endian modes.
  194: @item The Linux userland emulation can run many 32 bit MIPS Linux binaries.
  196: @end itemize
  198: Current QEMU limitations:
  200: @itemize
  202: @item Self-modifying code is not always handled correctly.
  204: @item 64 bit userland emulation is not implemented.
  206: @item The system emulation is not complete enough to run real firmware.
  208: @item The watchpoint debug facility is not implemented.
  210: @end itemize
  212: @node intro_ppc_emulation
  213: @section PowerPC emulation
  215: @itemize
  217: @item Full PowerPC 32 bit emulation, including privileged instructions,
  218: FPU and MMU.
  220: @item Can run most PowerPC Linux binaries.
  222: @end itemize
  224: @node intro_sparc_emulation
  225: @section Sparc32 and Sparc64 emulation
  227: @itemize
  229: @item Full SPARC V8 emulation, including privileged
  230: instructions, FPU and MMU. SPARC V9 emulation includes most privileged
  231: and VIS instructions, FPU and I/D MMU. Alignment is fully enforced.
  233: @item Can run most 32-bit SPARC Linux binaries, SPARC32PLUS Linux binaries and
  234: some 64-bit SPARC Linux binaries.
  236: @end itemize
  238: Current QEMU limitations:
  240: @itemize
  242: @item IPC syscalls are missing.
  244: @item Floating point exception support is buggy.
  246: @item Atomic instructions are not correctly implemented.
  248: @item There are still some problems with Sparc64 emulators.
  250: @end itemize
  252: @node intro_other_emulation
  253: @section Other CPU emulation
  255: In addition to the above, QEMU supports emulation of other CPUs with
  256: varying levels of success. These are:
  258: @itemize
  260: @item
  261: Alpha
  262: @item
  263: CRIS
  264: @item
  265: M68k
  266: @item
  267: SH4
  268: @end itemize
  270: @node QEMU Internals
  271: @chapter QEMU Internals
  273: @menu
  274: * QEMU compared to other emulators::
  275: * Portable dynamic translation::
  276: * Condition code optimisations::
  277: * CPU state optimisations::
  278: * Translation cache::
  279: * Direct block chaining::
  280: * Self-modifying code and translated code invalidation::
  281: * Exception support::
  282: * MMU emulation::
  283: * Device emulation::
  284: * Hardware interrupts::
  285: * User emulation specific details::
  286: * Bibliography::
  287: @end menu
  289: @node QEMU compared to other emulators
  290: @section QEMU compared to other emulators
  292: Like bochs [3], QEMU emulates an x86 CPU. But QEMU is much faster than
  293: bochs as it uses dynamic compilation. Bochs is closely tied to x86 PC
  294: emulation while QEMU can emulate several processors.
  296: Like Valgrind [2], QEMU does user space emulation and dynamic
  297: translation. Valgrind is mainly a memory debugger while QEMU has no
  298: support for it (QEMU could be used to detect out of bound memory
  299: accesses as Valgrind, but it has no support to track uninitialised data
  300: as Valgrind does). The Valgrind dynamic translator generates better code
  301: than QEMU (in particular it does register allocation) but it is closely
  302: tied to an x86 host and target and has no support for precise exceptions
  303: and system emulation.
  305: EM86 [4] is the closest project to user space QEMU (and QEMU still uses
  306: some of its code, in particular the ELF file loader). EM86 was limited
  307: to an alpha host and used a proprietary and slow interpreter (the
  308: interpreter part of the FX!32 Digital Win32 code translator [5]).
  310: TWIN [6] is a Windows API emulator like Wine. It is less accurate than
  311: Wine but includes a protected mode x86 interpreter to launch x86 Windows
  312: executables. Such an approach has greater potential because most of the
  313: Windows API is executed natively but it is far more difficult to develop
  314: because all the data structures and function parameters exchanged
  315: between the API and the x86 code must be converted.
  317: User mode Linux [7] was the only solution before QEMU to launch a
  318: Linux kernel as a process while not needing any host kernel
  319: patches. However, user mode Linux requires heavy kernel patches while
  320: QEMU accepts unpatched Linux kernels. The price to pay is that QEMU is
  321: slower.
  323: The Plex86 [8] PC virtualizer is done in the same spirit as the now
  324: obsolete qemu-fast system emulator. It requires a patched Linux kernel
  325: to work (you cannot launch the same kernel on your PC), but the
  326: patches are really small. As it is a PC virtualizer (no emulation is
  327: done except for some privileged instructions), it has the potential of
  328: being faster than QEMU. The downside is that a complicated (and
  329: potentially unsafe) host kernel patch is needed.
  331: The commercial PC Virtualizers (VMWare [9], VirtualPC [10], TwoOStwo
  332: [11]) are faster than QEMU, but they all need specific, proprietary
  333: and potentially unsafe host drivers. Moreover, they are unable to
  334: provide cycle exact simulation as an emulator can.
  336: VirtualBox [12], Xen [13] and KVM [14] are based on QEMU. QEMU-SystemC
  337: [15] uses QEMU to simulate a system where some hardware devices are
  338: developed in SystemC.
  340: @node Portable dynamic translation
  341: @section Portable dynamic translation
  343: QEMU is a dynamic translator. When it first encounters a piece of code,
  344: it converts it to the host instruction set. Usually dynamic translators
  345: are very complicated and highly CPU dependent. QEMU uses some tricks
  346: which make it relatively easily portable and simple while achieving good
  347: performances.
  349: After the release of version 0.9.1, QEMU switched to a new method of
  350: generating code, Tiny Code Generator or TCG. TCG relaxes the
  351: dependency on the exact version of the compiler used. The basic idea
  352: is to split every target instruction into a couple of RISC-like TCG
  353: ops (see @code{target-i386/translate.c}). Some optimizations can be
  354: performed at this stage, including liveness analysis and trivial
  355: constant expression evaluation. TCG ops are then implemented in the
  356: host CPU back end, also known as TCG target (see
  357: @code{tcg/i386/tcg-target.c}). For more information, please take a
  358: look at @code{tcg/README}.
  360: @node Condition code optimisations
  361: @section Condition code optimisations
  363: Lazy evaluation of CPU condition codes (@code{EFLAGS} register on x86)
  364: is important for CPUs where every instruction sets the condition
  365: codes. It tends to be less important on conventional RISC systems
  366: where condition codes are only updated when explicitly requested. On
  367: Sparc64, costly update of both 32 and 64 bit condition codes can be
  368: avoided with lazy evaluation.
  370: Instead of computing the condition codes after each x86 instruction,
  371: QEMU just stores one operand (called @code{CC_SRC}), the result
  372: (called @code{CC_DST}) and the type of operation (called
  373: @code{CC_OP}). When the condition codes are needed, the condition
  374: codes can be calculated using this information. In addition, an
  375: optimized calculation can be performed for some instruction types like
  376: conditional branches.
  378: @code{CC_OP} is almost never explicitly set in the generated code
  379: because it is known at translation time.
  381: The lazy condition code evaluation is used on x86, m68k, cris and
  382: Sparc. ARM uses a simplified variant for the N and Z flags.
  384: @node CPU state optimisations
  385: @section CPU state optimisations
  387: The target CPUs have many internal states which change the way it
  388: evaluates instructions. In order to achieve a good speed, the
  389: translation phase considers that some state information of the virtual
  390: CPU cannot change in it. The state is recorded in the Translation
  391: Block (TB). If the state changes (e.g. privilege level), a new TB will
  392: be generated and the previous TB won't be used anymore until the state
  393: matches the state recorded in the previous TB. For example, if the SS,
  394: DS and ES segments have a zero base, then the translator does not even
  395: generate an addition for the segment base.
  397: [The FPU stack pointer register is not handled that way yet].
  399: @node Translation cache
  400: @section Translation cache
  402: A 16 MByte cache holds the most recently used translations. For
  403: simplicity, it is completely flushed when it is full. A translation unit
  404: contains just a single basic block (a block of x86 instructions
  405: terminated by a jump or by a virtual CPU state change which the
  406: translator cannot deduce statically).
  408: @node Direct block chaining
  409: @section Direct block chaining
  411: After each translated basic block is executed, QEMU uses the simulated
  412: Program Counter (PC) and other cpu state informations (such as the CS
  413: segment base value) to find the next basic block.
  415: In order to accelerate the most common cases where the new simulated PC
  416: is known, QEMU can patch a basic block so that it jumps directly to the
  417: next one.
  419: The most portable code uses an indirect jump. An indirect jump makes
  420: it easier to make the jump target modification atomic. On some host
  421: architectures (such as x86 or PowerPC), the @code{JUMP} opcode is
  422: directly patched so that the block chaining has no overhead.
  424: @node Self-modifying code and translated code invalidation
  425: @section Self-modifying code and translated code invalidation
  427: Self-modifying code is a special challenge in x86 emulation because no
  428: instruction cache invalidation is signaled by the application when code
  429: is modified.
  431: When translated code is generated for a basic block, the corresponding
  432: host page is write protected if it is not already read-only. Then, if
  433: a write access is done to the page, Linux raises a SEGV signal. QEMU
  434: then invalidates all the translated code in the page and enables write
  435: accesses to the page.
  437: Correct translated code invalidation is done efficiently by maintaining
  438: a linked list of every translated block contained in a given page. Other
  439: linked lists are also maintained to undo direct block chaining.
  441: On RISC targets, correctly written software uses memory barriers and
  442: cache flushes, so some of the protection above would not be
  443: necessary. However, QEMU still requires that the generated code always
  444: matches the target instructions in memory in order to handle
  445: exceptions correctly.
  447: @node Exception support
  448: @section Exception support
  450: longjmp() is used when an exception such as division by zero is
  451: encountered.
  453: The host SIGSEGV and SIGBUS signal handlers are used to get invalid
  454: memory accesses. The simulated program counter is found by
  455: retranslating the corresponding basic block and by looking where the
  456: host program counter was at the exception point.
  458: The virtual CPU cannot retrieve the exact @code{EFLAGS} register because
  459: in some cases it is not computed because of condition code
  460: optimisations. It is not a big concern because the emulated code can
  461: still be restarted in any cases.
  463: @node MMU emulation
  464: @section MMU emulation
  466: For system emulation QEMU supports a soft MMU. In that mode, the MMU
  467: virtual to physical address translation is done at every memory
  468: access. QEMU uses an address translation cache to speed up the
  469: translation.
  471: In order to avoid flushing the translated code each time the MMU
  472: mappings change, QEMU uses a physically indexed translation cache. It
  473: means that each basic block is indexed with its physical address.
  475: When MMU mappings change, only the chaining of the basic blocks is
  476: reset (i.e. a basic block can no longer jump directly to another one).
  478: @node Device emulation
  479: @section Device emulation
  481: Systems emulated by QEMU are organized by boards. At initialization
  482: phase, each board instantiates a number of CPUs, devices, RAM and
  483: ROM. Each device in turn can assign I/O ports or memory areas (for
  484: MMIO) to its handlers. When the emulation starts, an access to the
  485: ports or MMIO memory areas assigned to the device causes the
  486: corresponding handler to be called.
  488: RAM and ROM are handled more optimally, only the offset to the host
  489: memory needs to be added to the guest address.
  491: The video RAM of VGA and other display cards is special: it can be
  492: read or written directly like RAM, but write accesses cause the memory
  493: to be marked with VGA_DIRTY flag as well.
  495: QEMU supports some device classes like serial and parallel ports, USB,
  496: drives and network devices, by providing APIs for easier connection to
  497: the generic, higher level implementations. The API hides the
  498: implementation details from the devices, like native device use or
  499: advanced block device formats like QCOW.
  501: Usually the devices implement a reset method and register support for
  502: saving and loading of the device state. The devices can also use
  503: timers, especially together with the use of bottom halves (BHs).
  505: @node Hardware interrupts
  506: @section Hardware interrupts
  508: In order to be faster, QEMU does not check at every basic block if an
  509: hardware interrupt is pending. Instead, the user must asynchrously
  510: call a specific function to tell that an interrupt is pending. This
  511: function resets the chaining of the currently executing basic
  512: block. It ensures that the execution will return soon in the main loop
  513: of the CPU emulator. Then the main loop can test if the interrupt is
  514: pending and handle it.
  516: @node User emulation specific details
  517: @section User emulation specific details
  519: @subsection Linux system call translation
  521: QEMU includes a generic system call translator for Linux. It means that
  522: the parameters of the system calls can be converted to fix the
  523: endianness and 32/64 bit issues. The IOCTLs are converted with a generic
  524: type description system (see @file{ioctls.h} and @file{thunk.c}).
  526: QEMU supports host CPUs which have pages bigger than 4KB. It records all
  527: the mappings the process does and try to emulated the @code{mmap()}
  528: system calls in cases where the host @code{mmap()} call would fail
  529: because of bad page alignment.
  531: @subsection Linux signals
  533: Normal and real-time signals are queued along with their information
  534: (@code{siginfo_t}) as it is done in the Linux kernel. Then an interrupt
  535: request is done to the virtual CPU. When it is interrupted, one queued
  536: signal is handled by generating a stack frame in the virtual CPU as the
  537: Linux kernel does. The @code{sigreturn()} system call is emulated to return
  538: from the virtual signal handler.
  540: Some signals (such as SIGALRM) directly come from the host. Other
  541: signals are synthetized from the virtual CPU exceptions such as SIGFPE
  542: when a division by zero is done (see @code{main.c:cpu_loop()}).
  544: The blocked signal mask is still handled by the host Linux kernel so
  545: that most signal system calls can be redirected directly to the host
  546: Linux kernel. Only the @code{sigaction()} and @code{sigreturn()} system
  547: calls need to be fully emulated (see @file{signal.c}).
  549: @subsection clone() system call and threads
  551: The Linux clone() system call is usually used to create a thread. QEMU
  552: uses the host clone() system call so that real host threads are created
  553: for each emulated thread. One virtual CPU instance is created for each
  554: thread.
  556: The virtual x86 CPU atomic operations are emulated with a global lock so
  557: that their semantic is preserved.
  559: Note that currently there are still some locking issues in QEMU. In
  560: particular, the translated cache flush is not protected yet against
  561: reentrancy.
  563: @subsection Self-virtualization
  565: QEMU was conceived so that ultimately it can emulate itself. Although
  566: it is not very useful, it is an important test to show the power of the
  567: emulator.
  569: Achieving self-virtualization is not easy because there may be address
  570: space conflicts. QEMU user emulators solve this problem by being an
  571: executable ELF shared object as the ELF interpreter. That
  572: way, it can be relocated at load time.
  574: @node Bibliography
  575: @section Bibliography
  577: @table @asis
  579: @item [1]
  580: @url{}, Optimizing
  581: direct threaded code by selective inlining (1998) by Ian Piumarta, Fabio
  582: Riccardi.
  584: @item [2]
  585: @url{}, Valgrind, an open-source
  586: memory debugger for x86-GNU/Linux, by Julian Seward.
  588: @item [3]
  589: @url{}, the Bochs IA-32 Emulator Project,
  590: by Kevin Lawton et al.
  592: @item [4]
  593: @url{}, the EM86
  594: x86 emulator on Alpha-Linux.
  596: @item [5]
  597: @url{},
  598: DIGITAL FX!32: Running 32-Bit x86 Applications on Alpha NT, by Anton
  599: Chernoff and Ray Hookway.
  601: @item [6]
  602: @url{}, Windows API library emulation from
  603: Willows Software.
  605: @item [7]
  606: @url{},
  607: The User-mode Linux Kernel.
  609: @item [8]
  610: @url{},
  611: The new Plex86 project.
  613: @item [9]
  614: @url{},
  615: The VMWare PC virtualizer.
  617: @item [10]
  618: @url{},
  619: The VirtualPC PC virtualizer.
  621: @item [11]
  622: @url{},
  623: The TwoOStwo PC virtualizer.
  625: @item [12]
  626: @url{},
  627: The VirtualBox PC virtualizer.
  629: @item [13]
  630: @url{},
  631: The Xen hypervisor.
  633: @item [14]
  634: @url{},
  635: Kernel Based Virtual Machine (KVM).
  637: @item [15]
  638: @url{},
  639: QEMU-SystemC, a hardware co-simulator.
  641: @end table
  643: @node Regression Tests
  644: @chapter Regression Tests
  646: In the directory @file{tests/}, various interesting testing programs
  647: are available. They are used for regression testing.
  649: @menu
  650: * test-i386::
  651: * linux-test::
  652: * qruncom.c::
  653: @end menu
  655: @node test-i386
  656: @section @file{test-i386}
  658: This program executes most of the 16 bit and 32 bit x86 instructions and
  659: generates a text output. It can be compared with the output obtained with
  660: a real CPU or another emulator. The target @code{make test} runs this
  661: program and a @code{diff} on the generated output.
  663: The Linux system call @code{modify_ldt()} is used to create x86 selectors
  664: to test some 16 bit addressing and 32 bit with segmentation cases.
  666: The Linux system call @code{vm86()} is used to test vm86 emulation.
  668: Various exceptions are raised to test most of the x86 user space
  669: exception reporting.
  671: @node linux-test
  672: @section @file{linux-test}
  674: This program tests various Linux system calls. It is used to verify
  675: that the system call parameters are correctly converted between target
  676: and host CPUs.
  678: @node qruncom.c
  679: @section @file{qruncom.c}
  681: Example of usage of @code{libqemu} to emulate a user mode i386 CPU.
  683: @node Index
  684: @chapter Index
  685: @printindex cp
  687: @bye