File:  [Qemu by Fabrice Bellard] / qemu / qemu-tech.texi
Revision (vendor branch): download - view: text, annotated - select for diffs
Tue Apr 24 16:40:49 2018 UTC (2 years, 4 months ago) by root
Branches: qemu, MAIN
CVS tags: qemu0090, qemu0082, qemu0081, HEAD
qemu 0.8.1

\input texinfo @c -*- texinfo -*-
@c %**start of header
@settitle QEMU Internals
@exampleindent 0
@paragraphindent 0
@c %**end of header

@sp 7
@center @titlefont{QEMU Internals}
@sp 3
@end titlepage
@end iftex

@node Top

* Introduction::
* QEMU Internals::
* Regression Tests::
* Index::
@end menu
@end ifnottex


@node Introduction
@chapter Introduction

* intro_features::        Features
* intro_x86_emulation::   x86 emulation
* intro_arm_emulation::   ARM emulation
* intro_ppc_emulation::   PowerPC emulation
* intro_sparc_emulation:: SPARC emulation
@end menu

@node intro_features
@section Features

QEMU is a FAST! processor emulator using a portable dynamic

QEMU has two operating modes:

@itemize @minus

Full system emulation. In this mode, QEMU emulates a full system
(usually a PC), including a processor and various peripherals. It can
be used to launch an different Operating System without rebooting the
PC or to debug system code.

User mode emulation (Linux host only). In this mode, QEMU can launch
Linux processes compiled for one CPU on another CPU. It can be used to
launch the Wine Windows API emulator (@url{}) or
to ease cross-compilation and cross-debugging.

@end itemize

As QEMU requires no host kernel driver to run, it is very safe and
easy to use.

QEMU generic features:


@item User space only or full system emulation.

@item Using dynamic translation to native code for reasonable speed.

@item Working on x86 and PowerPC hosts. Being tested on ARM, Sparc32, Alpha and S390.

@item Self-modifying code support.

@item Precise exceptions support.

@item The virtual CPU is a library (@code{libqemu}) which can be used 
in other projects (look at @file{qemu/tests/qruncom.c} to have an
example of user mode @code{libqemu} usage).

@end itemize

QEMU user mode emulation features:
@item Generic Linux system call converter, including most ioctls.

@item clone() emulation using native CPU clone() to use Linux scheduler for threads.

@item Accurate signal handling by remapping host signals to target signals. 
@end itemize

QEMU full system emulation features:
@item QEMU can either use a full software MMU for maximum portability or use the host system call mmap() to simulate the target MMU.
@end itemize

@node intro_x86_emulation
@section x86 emulation

QEMU x86 target features:


@item The virtual x86 CPU supports 16 bit and 32 bit addressing with segmentation. 
LDT/GDT and IDT are emulated. VM86 mode is also supported to run DOSEMU.

@item Support of host page sizes bigger than 4KB in user mode emulation.

@item QEMU can emulate itself on x86.

@item An extensive Linux x86 CPU test program is included @file{tests/test-i386}. 
It can be used to test other x86 virtual CPUs.

@end itemize

Current QEMU limitations:


@item No SSE/MMX support (yet).

@item No x86-64 support.

@item IPC syscalls are missing.

@item The x86 segment limits and access rights are not tested at every 
memory access (yet). Hopefully, very few OSes seem to rely on that for
normal use.

@item On non x86 host CPUs, @code{double}s are used instead of the non standard 
10 byte @code{long double}s of x86 for floating point emulation to get
maximum performances.

@end itemize

@node intro_arm_emulation
@section ARM emulation


@item Full ARM 7 user emulation.

@item NWFPE FPU support included in user Linux emulation.

@item Can run most ARM Linux binaries.

@end itemize

@node intro_ppc_emulation
@section PowerPC emulation


@item Full PowerPC 32 bit emulation, including privileged instructions, 
FPU and MMU.

@item Can run most PowerPC Linux binaries.

@end itemize

@node intro_sparc_emulation
@section SPARC emulation


@item Somewhat complete SPARC V8 emulation, including privileged
instructions, FPU and MMU. SPARC V9 emulation includes most privileged
instructions, FPU and I/D MMU, but misses VIS instructions.

@item Can run some 32-bit SPARC Linux binaries.

@end itemize

Current QEMU limitations:


@item Tagged add/subtract instructions are not supported, but they are
probably not used.

@item IPC syscalls are missing.

@item 128-bit floating point operations are not supported, though none of the
real CPUs implement them either. FCMPE[SD] are not correctly
implemented.  Floating point exception support is untested.

@item Alignment is not enforced at all.

@item Atomic instructions are not correctly implemented.

@item Sparc64 emulators are not usable for anything yet.

@end itemize

@node QEMU Internals
@chapter QEMU Internals

* QEMU compared to other emulators::
* Portable dynamic translation::
* Register allocation::
* Condition code optimisations::
* CPU state optimisations::
* Translation cache::
* Direct block chaining::
* Self-modifying code and translated code invalidation::
* Exception support::
* MMU emulation::
* Hardware interrupts::
* User emulation specific details::
* Bibliography::
@end menu

@node QEMU compared to other emulators
@section QEMU compared to other emulators

Like bochs [3], QEMU emulates an x86 CPU. But QEMU is much faster than
bochs as it uses dynamic compilation. Bochs is closely tied to x86 PC
emulation while QEMU can emulate several processors.

Like Valgrind [2], QEMU does user space emulation and dynamic
translation. Valgrind is mainly a memory debugger while QEMU has no
support for it (QEMU could be used to detect out of bound memory
accesses as Valgrind, but it has no support to track uninitialised data
as Valgrind does). The Valgrind dynamic translator generates better code
than QEMU (in particular it does register allocation) but it is closely
tied to an x86 host and target and has no support for precise exceptions
and system emulation.

EM86 [4] is the closest project to user space QEMU (and QEMU still uses
some of its code, in particular the ELF file loader). EM86 was limited
to an alpha host and used a proprietary and slow interpreter (the
interpreter part of the FX!32 Digital Win32 code translator [5]).

TWIN [6] is a Windows API emulator like Wine. It is less accurate than
Wine but includes a protected mode x86 interpreter to launch x86 Windows
executables. Such an approach has greater potential because most of the
Windows API is executed natively but it is far more difficult to develop
because all the data structures and function parameters exchanged
between the API and the x86 code must be converted.

User mode Linux [7] was the only solution before QEMU to launch a
Linux kernel as a process while not needing any host kernel
patches. However, user mode Linux requires heavy kernel patches while
QEMU accepts unpatched Linux kernels. The price to pay is that QEMU is

The new Plex86 [8] PC virtualizer is done in the same spirit as the
qemu-fast system emulator. It requires a patched Linux kernel to work
(you cannot launch the same kernel on your PC), but the patches are
really small. As it is a PC virtualizer (no emulation is done except
for some priveledged instructions), it has the potential of being
faster than QEMU. The downside is that a complicated (and potentially
unsafe) host kernel patch is needed.

The commercial PC Virtualizers (VMWare [9], VirtualPC [10], TwoOStwo
[11]) are faster than QEMU, but they all need specific, proprietary
and potentially unsafe host drivers. Moreover, they are unable to
provide cycle exact simulation as an emulator can.

@node Portable dynamic translation
@section Portable dynamic translation

QEMU is a dynamic translator. When it first encounters a piece of code,
it converts it to the host instruction set. Usually dynamic translators
are very complicated and highly CPU dependent. QEMU uses some tricks
which make it relatively easily portable and simple while achieving good

The basic idea is to split every x86 instruction into fewer simpler
instructions. Each simple instruction is implemented by a piece of C
code (see @file{target-i386/op.c}). Then a compile time tool
(@file{dyngen}) takes the corresponding object file (@file{op.o})
to generate a dynamic code generator which concatenates the simple
instructions to build a function (see @file{op.h:dyngen_code()}).

In essence, the process is similar to [1], but more work is done at
compile time. 

A key idea to get optimal performances is that constant parameters can
be passed to the simple operations. For that purpose, dummy ELF
relocations are generated with gcc for each constant parameter. Then,
the tool (@file{dyngen}) can locate the relocations and generate the
appriopriate C code to resolve them when building the dynamic code.

That way, QEMU is no more difficult to port than a dynamic linker.

To go even faster, GCC static register variables are used to keep the
state of the virtual CPU.

@node Register allocation
@section Register allocation

Since QEMU uses fixed simple instructions, no efficient register
allocation can be done. However, because RISC CPUs have a lot of
register, most of the virtual CPU state can be put in registers without
doing complicated register allocation.

@node Condition code optimisations
@section Condition code optimisations

Good CPU condition codes emulation (@code{EFLAGS} register on x86) is a
critical point to get good performances. QEMU uses lazy condition code
evaluation: instead of computing the condition codes after each x86
instruction, it just stores one operand (called @code{CC_SRC}), the
result (called @code{CC_DST}) and the type of operation (called

@code{CC_OP} is almost never explicitely set in the generated code
because it is known at translation time.

In order to increase performances, a backward pass is performed on the
generated simple instructions (see
@code{target-i386/translate.c:optimize_flags()}). When it can be proved that
the condition codes are not needed by the next instructions, no
condition codes are computed at all.

@node CPU state optimisations
@section CPU state optimisations

The x86 CPU has many internal states which change the way it evaluates
instructions. In order to achieve a good speed, the translation phase
considers that some state information of the virtual x86 CPU cannot
change in it. For example, if the SS, DS and ES segments have a zero
base, then the translator does not even generate an addition for the
segment base.

[The FPU stack pointer register is not handled that way yet].

@node Translation cache
@section Translation cache

A 16 MByte cache holds the most recently used translations. For
simplicity, it is completely flushed when it is full. A translation unit
contains just a single basic block (a block of x86 instructions
terminated by a jump or by a virtual CPU state change which the
translator cannot deduce statically).

@node Direct block chaining
@section Direct block chaining

After each translated basic block is executed, QEMU uses the simulated
Program Counter (PC) and other cpu state informations (such as the CS
segment base value) to find the next basic block.

In order to accelerate the most common cases where the new simulated PC
is known, QEMU can patch a basic block so that it jumps directly to the
next one.

The most portable code uses an indirect jump. An indirect jump makes
it easier to make the jump target modification atomic. On some host
architectures (such as x86 or PowerPC), the @code{JUMP} opcode is
directly patched so that the block chaining has no overhead.

@node Self-modifying code and translated code invalidation
@section Self-modifying code and translated code invalidation

Self-modifying code is a special challenge in x86 emulation because no
instruction cache invalidation is signaled by the application when code
is modified.

When translated code is generated for a basic block, the corresponding
host page is write protected if it is not already read-only (with the
system call @code{mprotect()}). Then, if a write access is done to the
page, Linux raises a SEGV signal. QEMU then invalidates all the
translated code in the page and enables write accesses to the page.

Correct translated code invalidation is done efficiently by maintaining
a linked list of every translated block contained in a given page. Other
linked lists are also maintained to undo direct block chaining. 

Although the overhead of doing @code{mprotect()} calls is important,
most MSDOS programs can be emulated at reasonnable speed with QEMU and

Note that QEMU also invalidates pages of translated code when it detects
that memory mappings are modified with @code{mmap()} or @code{munmap()}.

When using a software MMU, the code invalidation is more efficient: if
a given code page is invalidated too often because of write accesses,
then a bitmap representing all the code inside the page is
built. Every store into that page checks the bitmap to see if the code
really needs to be invalidated. It avoids invalidating the code when
only data is modified in the page.

@node Exception support
@section Exception support

longjmp() is used when an exception such as division by zero is

The host SIGSEGV and SIGBUS signal handlers are used to get invalid
memory accesses. The exact CPU state can be retrieved because all the
x86 registers are stored in fixed host registers. The simulated program
counter is found by retranslating the corresponding basic block and by
looking where the host program counter was at the exception point.

The virtual CPU cannot retrieve the exact @code{EFLAGS} register because
in some cases it is not computed because of condition code
optimisations. It is not a big concern because the emulated code can
still be restarted in any cases.

@node MMU emulation
@section MMU emulation

For system emulation, QEMU uses the mmap() system call to emulate the
target CPU MMU. It works as long the emulated OS does not use an area
reserved by the host OS (such as the area above 0xc0000000 on x86

In order to be able to launch any OS, QEMU also supports a soft
MMU. In that mode, the MMU virtual to physical address translation is
done at every memory access. QEMU uses an address translation cache to
speed up the translation.

In order to avoid flushing the translated code each time the MMU
mappings change, QEMU uses a physically indexed translation cache. It
means that each basic block is indexed with its physical address. 

When MMU mappings change, only the chaining of the basic blocks is
reset (i.e. a basic block can no longer jump directly to another one).

@node Hardware interrupts
@section Hardware interrupts

In order to be faster, QEMU does not check at every basic block if an
hardware interrupt is pending. Instead, the user must asynchrously
call a specific function to tell that an interrupt is pending. This
function resets the chaining of the currently executing basic
block. It ensures that the execution will return soon in the main loop
of the CPU emulator. Then the main loop can test if the interrupt is
pending and handle it.

@node User emulation specific details
@section User emulation specific details

@subsection Linux system call translation

QEMU includes a generic system call translator for Linux. It means that
the parameters of the system calls can be converted to fix the
endianness and 32/64 bit issues. The IOCTLs are converted with a generic
type description system (see @file{ioctls.h} and @file{thunk.c}).

QEMU supports host CPUs which have pages bigger than 4KB. It records all
the mappings the process does and try to emulated the @code{mmap()}
system calls in cases where the host @code{mmap()} call would fail
because of bad page alignment.

@subsection Linux signals

Normal and real-time signals are queued along with their information
(@code{siginfo_t}) as it is done in the Linux kernel. Then an interrupt
request is done to the virtual CPU. When it is interrupted, one queued
signal is handled by generating a stack frame in the virtual CPU as the
Linux kernel does. The @code{sigreturn()} system call is emulated to return
from the virtual signal handler.

Some signals (such as SIGALRM) directly come from the host. Other
signals are synthetized from the virtual CPU exceptions such as SIGFPE
when a division by zero is done (see @code{main.c:cpu_loop()}).

The blocked signal mask is still handled by the host Linux kernel so
that most signal system calls can be redirected directly to the host
Linux kernel. Only the @code{sigaction()} and @code{sigreturn()} system
calls need to be fully emulated (see @file{signal.c}).

@subsection clone() system call and threads

The Linux clone() system call is usually used to create a thread. QEMU
uses the host clone() system call so that real host threads are created
for each emulated thread. One virtual CPU instance is created for each

The virtual x86 CPU atomic operations are emulated with a global lock so
that their semantic is preserved.

Note that currently there are still some locking issues in QEMU. In
particular, the translated cache flush is not protected yet against

@subsection Self-virtualization

QEMU was conceived so that ultimately it can emulate itself. Although
it is not very useful, it is an important test to show the power of the

Achieving self-virtualization is not easy because there may be address
space conflicts. QEMU solves this problem by being an executable ELF
shared object as the ELF interpreter. That way, it can be
relocated at load time.

@node Bibliography
@section Bibliography

@table @asis

@item [1] 
@url{}, Optimizing
direct threaded code by selective inlining (1998) by Ian Piumarta, Fabio

@item [2]
@url{}, Valgrind, an open-source
memory debugger for x86-GNU/Linux, by Julian Seward.

@item [3]
@url{}, the Bochs IA-32 Emulator Project,
by Kevin Lawton et al.

@item [4]
@url{}, the EM86
x86 emulator on Alpha-Linux.

@item [5]
DIGITAL FX!32: Running 32-Bit x86 Applications on Alpha NT, by Anton
Chernoff and Ray Hookway.

@item [6]
@url{}, Windows API library emulation from
Willows Software.

@item [7]
The User-mode Linux Kernel.

@item [8]
The new Plex86 project.

@item [9]
The VMWare PC virtualizer.

@item [10]
The VirtualPC PC virtualizer.

@item [11]
The TwoOStwo PC virtualizer.

@end table

@node Regression Tests
@chapter Regression Tests

In the directory @file{tests/}, various interesting testing programs
are available. There are used for regression testing.

* test-i386::
* linux-test::
* qruncom.c::
@end menu

@node test-i386
@section @file{test-i386}

This program executes most of the 16 bit and 32 bit x86 instructions and
generates a text output. It can be compared with the output obtained with
a real CPU or another emulator. The target @code{make test} runs this
program and a @code{diff} on the generated output.

The Linux system call @code{modify_ldt()} is used to create x86 selectors
to test some 16 bit addressing and 32 bit with segmentation cases.

The Linux system call @code{vm86()} is used to test vm86 emulation.

Various exceptions are raised to test most of the x86 user space
exception reporting.

@node linux-test
@section @file{linux-test}

This program tests various Linux system calls. It is used to verify
that the system call parameters are correctly converted between target
and host CPUs.

@node qruncom.c
@section @file{qruncom.c}

Example of usage of @code{libqemu} to emulate a user mode i386 CPU.

@node Index
@chapter Index
@printindex cp