1
\input texinfo @c -*- texinfo -*-
3
@setfilename qemu-tech.info
6
@documentencoding UTF-8
8
@settitle QEMU Internals
15
* QEMU Internals: (qemu-tech). The QEMU Emulator Internals.
22
@center @titlefont{QEMU Internals}
45
* intro_features:: Features
46
* intro_x86_emulation:: x86 and x86-64 emulation
47
* intro_arm_emulation:: ARM emulation
48
* intro_mips_emulation:: MIPS emulation
49
* intro_ppc_emulation:: PowerPC emulation
50
* intro_sparc_emulation:: Sparc32 and Sparc64 emulation
51
* intro_xtensa_emulation:: Xtensa emulation
52
* intro_other_emulation:: Other CPU emulation
58
QEMU is a FAST! processor emulator using a portable dynamic
61
QEMU has two operating modes:
66
Full system emulation. In this mode (full platform virtualization),
67
QEMU emulates a full system (usually a PC), including a processor and
68
various peripherals. It can be used to launch several different
69
Operating Systems at once without rebooting the host machine or to
73
User mode emulation. In this mode (application level virtualization),
74
QEMU can launch processes compiled for one CPU on another CPU, however
75
the Operating Systems must match. This can be used for example to ease
76
cross-compilation and cross-debugging.
79
As QEMU requires no host kernel driver to run, it is very safe and
82
QEMU generic features:
86
@item User space only or full system emulation.
88
@item Using dynamic translation to native code for reasonable speed.
91
Working on x86, x86_64 and PowerPC32/64 hosts. Being tested on ARM,
92
HPPA, Sparc32 and Sparc64. Previous versions had some support for
93
Alpha and S390 hosts, but TCG (see below) doesn't support those yet.
95
@item Self-modifying code support.
97
@item Precise exceptions support.
99
@item The virtual CPU is a library (@code{libqemu}) which can be used
100
in other projects (look at @file{qemu/tests/qruncom.c} to have an
101
example of user mode @code{libqemu} usage).
104
Floating point library supporting both full software emulation and
105
native host FPU instructions.
109
QEMU user mode emulation features:
111
@item Generic Linux system call converter, including most ioctls.
113
@item clone() emulation using native CPU clone() to use Linux scheduler for threads.
115
@item Accurate signal handling by remapping host signals to target signals.
118
Linux user emulator (Linux host only) can be used to launch the Wine
119
Windows API emulator (@url{http://www.winehq.org}). A Darwin user
120
emulator (Darwin hosts only) exists and a BSD user emulator for BSD
121
hosts is under development. It would also be possible to develop a
122
similar user emulator for Solaris.
124
QEMU full system emulation features:
127
QEMU uses a full software MMU for maximum portability.
130
QEMU can optionally use an in-kernel accelerator, like kvm. The accelerators
131
execute some of the guest code natively, while
132
continuing to emulate the rest of the machine.
135
Various hardware devices can be emulated and in some cases, host
136
devices (e.g. serial and parallel ports, USB, drives) can be used
137
transparently by the guest Operating System. Host device passthrough
138
can be used for talking to external physical peripherals (e.g. a
139
webcam, modem or tape drive).
142
Symmetric multiprocessing (SMP) even on a host with a single CPU. On a
143
SMP host system, QEMU can use only one CPU fully due to difficulty in
144
implementing atomic memory accesses efficiently.
148
@node intro_x86_emulation
149
@section x86 and x86-64 emulation
151
QEMU x86 target features:
155
@item The virtual x86 CPU supports 16 bit and 32 bit addressing with segmentation.
156
LDT/GDT and IDT are emulated. VM86 mode is also supported to run
157
DOSEMU. There is some support for MMX/3DNow!, SSE, SSE2, SSE3, SSSE3,
158
and SSE4 as well as x86-64 SVM.
160
@item Support of host page sizes bigger than 4KB in user mode emulation.
162
@item QEMU can emulate itself on x86.
164
@item An extensive Linux x86 CPU test program is included @file{tests/test-i386}.
165
It can be used to test other x86 virtual CPUs.
169
Current QEMU limitations:
173
@item Limited x86-64 support.
175
@item IPC syscalls are missing.
177
@item The x86 segment limits and access rights are not tested at every
178
memory access (yet). Hopefully, very few OSes seem to rely on that for
183
@node intro_arm_emulation
184
@section ARM emulation
188
@item Full ARM 7 user emulation.
190
@item NWFPE FPU support included in user Linux emulation.
192
@item Can run most ARM Linux binaries.
196
@node intro_mips_emulation
197
@section MIPS emulation
201
@item The system emulation allows full MIPS32/MIPS64 Release 2 emulation,
202
including privileged instructions, FPU and MMU, in both little and big
205
@item The Linux userland emulation can run many 32 bit MIPS Linux binaries.
209
Current QEMU limitations:
213
@item Self-modifying code is not always handled correctly.
215
@item 64 bit userland emulation is not implemented.
217
@item The system emulation is not complete enough to run real firmware.
219
@item The watchpoint debug facility is not implemented.
223
@node intro_ppc_emulation
224
@section PowerPC emulation
228
@item Full PowerPC 32 bit emulation, including privileged instructions,
231
@item Can run most PowerPC Linux binaries.
235
@node intro_sparc_emulation
236
@section Sparc32 and Sparc64 emulation
240
@item Full SPARC V8 emulation, including privileged
241
instructions, FPU and MMU. SPARC V9 emulation includes most privileged
242
and VIS instructions, FPU and I/D MMU. Alignment is fully enforced.
244
@item Can run most 32-bit SPARC Linux binaries, SPARC32PLUS Linux binaries and
245
some 64-bit SPARC Linux binaries.
249
Current QEMU limitations:
253
@item IPC syscalls are missing.
255
@item Floating point exception support is buggy.
257
@item Atomic instructions are not correctly implemented.
259
@item There are still some problems with Sparc64 emulators.
263
@node intro_xtensa_emulation
264
@section Xtensa emulation
268
@item Core Xtensa ISA emulation, including most options: code density,
269
loop, extended L32R, 16- and 32-bit multiplication, 32-bit division,
270
MAC16, miscellaneous operations, boolean, multiprocessor synchronization,
271
conditional store, exceptions, relocatable vectors, unaligned exception,
272
interrupts (including high priority and timer), hardware alignment,
273
region protection, region translation, MMU, windowed registers, thread
274
pointer, processor ID.
276
@item Not implemented options: FP coprocessor, coprocessor context,
277
data/instruction cache (including cache prefetch and locking), XLMI,
278
processor interface, debug. Also options not covered by the core ISA
279
(e.g. FLIX, wide branches) are not implemented.
281
@item Can run most Xtensa Linux binaries.
283
@item New core configuration that requires no additional instructions
284
may be created from overlay with minimal amount of hand-written code.
288
@node intro_other_emulation
289
@section Other CPU emulation
291
In addition to the above, QEMU supports emulation of other CPUs with
292
varying levels of success. These are:
307
@chapter QEMU Internals
310
* QEMU compared to other emulators::
311
* Portable dynamic translation::
312
* Condition code optimisations::
313
* CPU state optimisations::
314
* Translation cache::
315
* Direct block chaining::
316
* Self-modifying code and translated code invalidation::
317
* Exception support::
320
* Hardware interrupts::
321
* User emulation specific details::
325
@node QEMU compared to other emulators
326
@section QEMU compared to other emulators
328
Like bochs [3], QEMU emulates an x86 CPU. But QEMU is much faster than
329
bochs as it uses dynamic compilation. Bochs is closely tied to x86 PC
330
emulation while QEMU can emulate several processors.
332
Like Valgrind [2], QEMU does user space emulation and dynamic
333
translation. Valgrind is mainly a memory debugger while QEMU has no
334
support for it (QEMU could be used to detect out of bound memory
335
accesses as Valgrind, but it has no support to track uninitialised data
336
as Valgrind does). The Valgrind dynamic translator generates better code
337
than QEMU (in particular it does register allocation) but it is closely
338
tied to an x86 host and target and has no support for precise exceptions
339
and system emulation.
341
EM86 [4] is the closest project to user space QEMU (and QEMU still uses
342
some of its code, in particular the ELF file loader). EM86 was limited
343
to an alpha host and used a proprietary and slow interpreter (the
344
interpreter part of the FX!32 Digital Win32 code translator [5]).
346
TWIN [6] is a Windows API emulator like Wine. It is less accurate than
347
Wine but includes a protected mode x86 interpreter to launch x86 Windows
348
executables. Such an approach has greater potential because most of the
349
Windows API is executed natively but it is far more difficult to develop
350
because all the data structures and function parameters exchanged
351
between the API and the x86 code must be converted.
353
User mode Linux [7] was the only solution before QEMU to launch a
354
Linux kernel as a process while not needing any host kernel
355
patches. However, user mode Linux requires heavy kernel patches while
356
QEMU accepts unpatched Linux kernels. The price to pay is that QEMU is
359
The Plex86 [8] PC virtualizer is done in the same spirit as the now
360
obsolete qemu-fast system emulator. It requires a patched Linux kernel
361
to work (you cannot launch the same kernel on your PC), but the
362
patches are really small. As it is a PC virtualizer (no emulation is
363
done except for some privileged instructions), it has the potential of
364
being faster than QEMU. The downside is that a complicated (and
365
potentially unsafe) host kernel patch is needed.
367
The commercial PC Virtualizers (VMWare [9], VirtualPC [10], TwoOStwo
368
[11]) are faster than QEMU, but they all need specific, proprietary
369
and potentially unsafe host drivers. Moreover, they are unable to
370
provide cycle exact simulation as an emulator can.
372
VirtualBox [12], Xen [13] and KVM [14] are based on QEMU. QEMU-SystemC
373
[15] uses QEMU to simulate a system where some hardware devices are
374
developed in SystemC.
376
@node Portable dynamic translation
377
@section Portable dynamic translation
379
QEMU is a dynamic translator. When it first encounters a piece of code,
380
it converts it to the host instruction set. Usually dynamic translators
381
are very complicated and highly CPU dependent. QEMU uses some tricks
382
which make it relatively easily portable and simple while achieving good
385
After the release of version 0.9.1, QEMU switched to a new method of
386
generating code, Tiny Code Generator or TCG. TCG relaxes the
387
dependency on the exact version of the compiler used. The basic idea
388
is to split every target instruction into a couple of RISC-like TCG
389
ops (see @code{target-i386/translate.c}). Some optimizations can be
390
performed at this stage, including liveness analysis and trivial
391
constant expression evaluation. TCG ops are then implemented in the
392
host CPU back end, also known as TCG target (see
393
@code{tcg/i386/tcg-target.c}). For more information, please take a
394
look at @code{tcg/README}.
396
@node Condition code optimisations
397
@section Condition code optimisations
399
Lazy evaluation of CPU condition codes (@code{EFLAGS} register on x86)
400
is important for CPUs where every instruction sets the condition
401
codes. It tends to be less important on conventional RISC systems
402
where condition codes are only updated when explicitly requested. On
403
Sparc64, costly update of both 32 and 64 bit condition codes can be
404
avoided with lazy evaluation.
406
Instead of computing the condition codes after each x86 instruction,
407
QEMU just stores one operand (called @code{CC_SRC}), the result
408
(called @code{CC_DST}) and the type of operation (called
409
@code{CC_OP}). When the condition codes are needed, the condition
410
codes can be calculated using this information. In addition, an
411
optimized calculation can be performed for some instruction types like
412
conditional branches.
414
@code{CC_OP} is almost never explicitly set in the generated code
415
because it is known at translation time.
417
The lazy condition code evaluation is used on x86, m68k, cris and
418
Sparc. ARM uses a simplified variant for the N and Z flags.
420
@node CPU state optimisations
421
@section CPU state optimisations
423
The target CPUs have many internal states which change the way it
424
evaluates instructions. In order to achieve a good speed, the
425
translation phase considers that some state information of the virtual
426
CPU cannot change in it. The state is recorded in the Translation
427
Block (TB). If the state changes (e.g. privilege level), a new TB will
428
be generated and the previous TB won't be used anymore until the state
429
matches the state recorded in the previous TB. For example, if the SS,
430
DS and ES segments have a zero base, then the translator does not even
431
generate an addition for the segment base.
433
[The FPU stack pointer register is not handled that way yet].
435
@node Translation cache
436
@section Translation cache
438
A 32 MByte cache holds the most recently used translations. For
439
simplicity, it is completely flushed when it is full. A translation unit
440
contains just a single basic block (a block of x86 instructions
441
terminated by a jump or by a virtual CPU state change which the
442
translator cannot deduce statically).
444
@node Direct block chaining
445
@section Direct block chaining
447
After each translated basic block is executed, QEMU uses the simulated
448
Program Counter (PC) and other cpu state informations (such as the CS
449
segment base value) to find the next basic block.
451
In order to accelerate the most common cases where the new simulated PC
452
is known, QEMU can patch a basic block so that it jumps directly to the
455
The most portable code uses an indirect jump. An indirect jump makes
456
it easier to make the jump target modification atomic. On some host
457
architectures (such as x86 or PowerPC), the @code{JUMP} opcode is
458
directly patched so that the block chaining has no overhead.
460
@node Self-modifying code and translated code invalidation
461
@section Self-modifying code and translated code invalidation
463
Self-modifying code is a special challenge in x86 emulation because no
464
instruction cache invalidation is signaled by the application when code
467
When translated code is generated for a basic block, the corresponding
468
host page is write protected if it is not already read-only. Then, if
469
a write access is done to the page, Linux raises a SEGV signal. QEMU
470
then invalidates all the translated code in the page and enables write
471
accesses to the page.
473
Correct translated code invalidation is done efficiently by maintaining
474
a linked list of every translated block contained in a given page. Other
475
linked lists are also maintained to undo direct block chaining.
477
On RISC targets, correctly written software uses memory barriers and
478
cache flushes, so some of the protection above would not be
479
necessary. However, QEMU still requires that the generated code always
480
matches the target instructions in memory in order to handle
481
exceptions correctly.
483
@node Exception support
484
@section Exception support
486
longjmp() is used when an exception such as division by zero is
489
The host SIGSEGV and SIGBUS signal handlers are used to get invalid
490
memory accesses. The simulated program counter is found by
491
retranslating the corresponding basic block and by looking where the
492
host program counter was at the exception point.
494
The virtual CPU cannot retrieve the exact @code{EFLAGS} register because
495
in some cases it is not computed because of condition code
496
optimisations. It is not a big concern because the emulated code can
497
still be restarted in any cases.
500
@section MMU emulation
502
For system emulation QEMU supports a soft MMU. In that mode, the MMU
503
virtual to physical address translation is done at every memory
504
access. QEMU uses an address translation cache to speed up the
507
In order to avoid flushing the translated code each time the MMU
508
mappings change, QEMU uses a physically indexed translation cache. It
509
means that each basic block is indexed with its physical address.
511
When MMU mappings change, only the chaining of the basic blocks is
512
reset (i.e. a basic block can no longer jump directly to another one).
514
@node Device emulation
515
@section Device emulation
517
Systems emulated by QEMU are organized by boards. At initialization
518
phase, each board instantiates a number of CPUs, devices, RAM and
519
ROM. Each device in turn can assign I/O ports or memory areas (for
520
MMIO) to its handlers. When the emulation starts, an access to the
521
ports or MMIO memory areas assigned to the device causes the
522
corresponding handler to be called.
524
RAM and ROM are handled more optimally, only the offset to the host
525
memory needs to be added to the guest address.
527
The video RAM of VGA and other display cards is special: it can be
528
read or written directly like RAM, but write accesses cause the memory
529
to be marked with VGA_DIRTY flag as well.
531
QEMU supports some device classes like serial and parallel ports, USB,
532
drives and network devices, by providing APIs for easier connection to
533
the generic, higher level implementations. The API hides the
534
implementation details from the devices, like native device use or
535
advanced block device formats like QCOW.
537
Usually the devices implement a reset method and register support for
538
saving and loading of the device state. The devices can also use
539
timers, especially together with the use of bottom halves (BHs).
541
@node Hardware interrupts
542
@section Hardware interrupts
544
In order to be faster, QEMU does not check at every basic block if an
545
hardware interrupt is pending. Instead, the user must asynchronously
546
call a specific function to tell that an interrupt is pending. This
547
function resets the chaining of the currently executing basic
548
block. It ensures that the execution will return soon in the main loop
549
of the CPU emulator. Then the main loop can test if the interrupt is
550
pending and handle it.
552
@node User emulation specific details
553
@section User emulation specific details
555
@subsection Linux system call translation
557
QEMU includes a generic system call translator for Linux. It means that
558
the parameters of the system calls can be converted to fix the
559
endianness and 32/64 bit issues. The IOCTLs are converted with a generic
560
type description system (see @file{ioctls.h} and @file{thunk.c}).
562
QEMU supports host CPUs which have pages bigger than 4KB. It records all
563
the mappings the process does and try to emulated the @code{mmap()}
564
system calls in cases where the host @code{mmap()} call would fail
565
because of bad page alignment.
567
@subsection Linux signals
569
Normal and real-time signals are queued along with their information
570
(@code{siginfo_t}) as it is done in the Linux kernel. Then an interrupt
571
request is done to the virtual CPU. When it is interrupted, one queued
572
signal is handled by generating a stack frame in the virtual CPU as the
573
Linux kernel does. The @code{sigreturn()} system call is emulated to return
574
from the virtual signal handler.
576
Some signals (such as SIGALRM) directly come from the host. Other
577
signals are synthesized from the virtual CPU exceptions such as SIGFPE
578
when a division by zero is done (see @code{main.c:cpu_loop()}).
580
The blocked signal mask is still handled by the host Linux kernel so
581
that most signal system calls can be redirected directly to the host
582
Linux kernel. Only the @code{sigaction()} and @code{sigreturn()} system
583
calls need to be fully emulated (see @file{signal.c}).
585
@subsection clone() system call and threads
587
The Linux clone() system call is usually used to create a thread. QEMU
588
uses the host clone() system call so that real host threads are created
589
for each emulated thread. One virtual CPU instance is created for each
592
The virtual x86 CPU atomic operations are emulated with a global lock so
593
that their semantic is preserved.
595
Note that currently there are still some locking issues in QEMU. In
596
particular, the translated cache flush is not protected yet against
599
@subsection Self-virtualization
601
QEMU was conceived so that ultimately it can emulate itself. Although
602
it is not very useful, it is an important test to show the power of the
605
Achieving self-virtualization is not easy because there may be address
606
space conflicts. QEMU user emulators solve this problem by being an
607
executable ELF shared object as the ld-linux.so ELF interpreter. That
608
way, it can be relocated at load time.
611
@section Bibliography
616
@url{http://citeseer.nj.nec.com/piumarta98optimizing.html}, Optimizing
617
direct threaded code by selective inlining (1998) by Ian Piumarta, Fabio
621
@url{http://developer.kde.org/~sewardj/}, Valgrind, an open-source
622
memory debugger for x86-GNU/Linux, by Julian Seward.
625
@url{http://bochs.sourceforge.net/}, the Bochs IA-32 Emulator Project,
626
by Kevin Lawton et al.
629
@url{http://www.cs.rose-hulman.edu/~donaldlf/em86/index.html}, the EM86
630
x86 emulator on Alpha-Linux.
633
@url{http://www.usenix.org/publications/library/proceedings/usenix-nt97/@/full_papers/chernoff/chernoff.pdf},
634
DIGITAL FX!32: Running 32-Bit x86 Applications on Alpha NT, by Anton
635
Chernoff and Ray Hookway.
638
@url{http://www.willows.com/}, Windows API library emulation from
642
@url{http://user-mode-linux.sourceforge.net/},
643
The User-mode Linux Kernel.
646
@url{http://www.plex86.org/},
647
The new Plex86 project.
650
@url{http://www.vmware.com/},
651
The VMWare PC virtualizer.
654
@url{http://www.microsoft.com/windowsxp/virtualpc/},
655
The VirtualPC PC virtualizer.
658
@url{http://www.twoostwo.org/},
659
The TwoOStwo PC virtualizer.
662
@url{http://virtualbox.org/},
663
The VirtualBox PC virtualizer.
666
@url{http://www.xen.org/},
670
@url{http://kvm.qumranet.com/kvmwiki/Front_Page},
671
Kernel Based Virtual Machine (KVM).
674
@url{http://www.greensocs.com/projects/QEMUSystemC},
675
QEMU-SystemC, a hardware co-simulator.
679
@node Regression Tests
680
@chapter Regression Tests
682
In the directory @file{tests/}, various interesting testing programs
683
are available. They are used for regression testing.
692
@section @file{test-i386}
694
This program executes most of the 16 bit and 32 bit x86 instructions and
695
generates a text output. It can be compared with the output obtained with
696
a real CPU or another emulator. The target @code{make test} runs this
697
program and a @code{diff} on the generated output.
699
The Linux system call @code{modify_ldt()} is used to create x86 selectors
700
to test some 16 bit addressing and 32 bit with segmentation cases.
702
The Linux system call @code{vm86()} is used to test vm86 emulation.
704
Various exceptions are raised to test most of the x86 user space
708
@section @file{linux-test}
710
This program tests various Linux system calls. It is used to verify
711
that the system call parameters are correctly converted between target
715
@section @file{qruncom.c}
717
Example of usage of @code{libqemu} to emulate a user mode i386 CPU.