1 \input texinfo @c -*- texinfo -*-
3 @settitle QEMU CPU Emulator Reference Documentation
6 @center @titlefont{QEMU CPU Emulator Reference Documentation}
14 QEMU is a FAST! processor emulator. By using dynamic translation it
15 achieves a reasonnable speed while being easy to port on new host
18 QEMU has two operating modes:
20 @item User mode emulation. In this mode, QEMU can launch Linux processes
21 compiled for one CPU on another CPU. Linux system calls are converted
22 because of endianness and 32/64 bit mismatches. The Wine Windows API
23 emulator (@url{http://www.winehq.org}) and the DOSEMU DOS emulator
24 (@url{www.dosemu.org}) are the main targets for QEMU.
26 @item Full system emulation. In this mode, QEMU emulates a full
27 system, including a processor and various peripherials. Currently, it
28 is only used to launch an x86 Linux kernel on an x86 Linux system. It
29 enables easier testing and debugging of system code. It can also be
30 used to provide virtual hosting of several virtual PCs on a single
35 As QEMU requires no host kernel patches to run, it is very safe and
38 QEMU generic features:
42 @item User space only or full system emulation.
44 @item Using dynamic translation to native code for reasonnable speed.
46 @item Working on x86 and PowerPC hosts. Being tested on ARM, Sparc32, Alpha and S390.
48 @item Self-modifying code support.
50 @item Precise exceptions support.
52 @item The virtual CPU is a library (@code{libqemu}) which can be used
57 QEMU user mode emulation features:
59 @item Generic Linux system call converter, including most ioctls.
61 @item clone() emulation using native CPU clone() to use Linux scheduler for threads.
63 @item Accurate signal handling by remapping host signals to target signals.
67 QEMU full system emulation features:
69 @item Using mmap() system calls to simulate the MMU
72 @section x86 emulation
74 QEMU x86 target features:
78 @item The virtual x86 CPU supports 16 bit and 32 bit addressing with segmentation.
79 LDT/GDT and IDT are emulated. VM86 mode is also supported to run DOSEMU.
81 @item Support of host page sizes bigger than 4KB in user mode emulation.
83 @item QEMU can emulate itself on x86.
85 @item An extensive Linux x86 CPU test program is included @file{tests/test-i386}.
86 It can be used to test other x86 virtual CPUs.
90 Current QEMU limitations:
94 @item No SSE/MMX support (yet).
96 @item No x86-64 support.
98 @item IPC syscalls are missing.
100 @item The x86 segment limits and access rights are not tested at every
103 @item On non x86 host CPUs, @code{double}s are used instead of the non standard
104 10 byte @code{long double}s of x86 for floating point emulation to get
105 maximum performances.
107 @item Full system emulation only works if no data are mapped above the virtual address
110 @item Some priviledged instructions or behaviors are missing. Only the ones
111 needed for proper Linux kernel operation are emulated.
113 @item No memory separation between the kernel and the user processes is done.
114 It will be implemented very soon.
118 @section ARM emulation
122 @item ARM emulation can currently launch small programs while using the
123 generic dynamic code generation architecture of QEMU.
125 @item No FPU support (yet).
127 @item No automatic regression testing (yet).
131 @chapter QEMU User space emulator invocation
135 If you need to compile QEMU, please read the @file{README} which gives
136 the related information.
138 In order to launch a Linux process, QEMU needs the process executable
139 itself and all the target (x86) dynamic libraries used by it.
143 @item On x86, you can just try to launch any process by using the native
150 @code{-L /} tells that the x86 dynamic linker must be searched with a
153 @item Since QEMU is also a linux process, you can launch qemu with qemu:
156 qemu -L / qemu -L / /bin/ls
159 @item On non x86 CPUs, you need first to download at least an x86 glibc
160 (@file{qemu-XXX-i386-glibc21.tar.gz} on the QEMU web page). Ensure that
161 @code{LD_LIBRARY_PATH} is not set:
164 unset LD_LIBRARY_PATH
167 Then you can launch the precompiled @file{ls} x86 executable:
170 qemu /usr/local/qemu-i386/bin/ls-i386
172 You can look at @file{/usr/local/qemu-i386/bin/qemu-conf.sh} so that
173 QEMU is automatically launched by the Linux kernel when you try to
174 launch x86 executables. It requires the @code{binfmt_misc} module in the
177 @item The x86 version of QEMU is also included. You can try weird things such as:
179 qemu /usr/local/qemu-i386/bin/qemu-i386 /usr/local/qemu-i386/bin/ls-i386
188 @item Ensure that you have a working QEMU with the x86 glibc
189 distribution (see previous section). In order to verify it, you must be
193 qemu /usr/local/qemu-i386/bin/ls-i386
196 @item Download the binary x86 Wine install
197 (@file{qemu-XXX-i386-wine.tar.gz} on the QEMU web page).
199 @item Configure Wine on your account. Look at the provided script
200 @file{/usr/local/qemu-i386/bin/wine-conf.sh}. Your previous
201 @code{$@{HOME@}/.wine} directory is saved to @code{$@{HOME@}/.wine.org}.
203 @item Then you can try the example @file{putty.exe}:
206 qemu /usr/local/qemu-i386/wine/bin/wine /usr/local/qemu-i386/wine/c/Program\ Files/putty.exe
211 @section Command line options
214 usage: qemu [-h] [-d] [-L path] [-s size] program [arguments...]
221 Set the x86 elf interpreter prefix (default=/usr/local/qemu-i386)
223 Set the x86 stack size in bytes (default=524288)
230 Activate log (logfile=/tmp/qemu.log)
232 Act as if the host page size was 'pagesize' bytes
235 @chapter QEMU System emulator invocation
239 This section explains how to launch a Linux kernel inside QEMU.
243 Download the archive @file{vl-test-xxx.tar.gz} containing a Linux kernel
244 and an initrd (initial Ram Disk). The archive also contains a
245 precompiled version of @file{vl}, the QEMU System emulator.
247 @item Optional: If you want network support (for example to launch X11 examples), you
248 must copy the script @file{vl-ifup} in @file{/etc} and configure
249 properly @code{sudo} so that the command @code{ifconfig} contained in
250 @file{vl-ifup} can be executed as root. You must verify that your host
251 kernel supports the TUN/TAP network interfaces: the device
252 @file{/dev/net/tun} must be present.
254 When network is enabled, there is a virtual network connection between
255 the host kernel and the emulated kernel. The emulated kernel is seen
256 from the host kernel at IP address 172.20.0.2 and the host kernel is
257 seen from the emulated kernel at IP address 172.20.0.1.
259 @item Launch @code{vl.sh}. You should have the following output:
263 connected to host network interface: tun0
264 Uncompressing Linux... Ok, booting the kernel.
265 Linux version 2.4.20 (bellard@voyager) (gcc version 2.95.2 20000220 (Debian GNU/Linux)) #42 Wed Jun 25 14:16:12 CEST 2003
266 BIOS-provided physical RAM map:
267 BIOS-88: 0000000000000000 - 000000000009f000 (usable)
268 BIOS-88: 0000000000100000 - 0000000002000000 (usable)
269 32MB LOWMEM available.
270 On node 0 totalpages: 8192
274 Kernel command line: root=/dev/ram ramdisk_size=6144
276 Detected 501.785 MHz processor.
277 Calibrating delay loop... 973.20 BogoMIPS
278 Memory: 24776k/32768k available (725k kernel code, 7604k reserved, 151k data, 48k init, 0k highmem)
279 Dentry cache hash table entries: 4096 (order: 3, 32768 bytes)
280 Inode cache hash table entries: 2048 (order: 2, 16384 bytes)
281 Mount-cache hash table entries: 512 (order: 0, 4096 bytes)
282 Buffer-cache hash table entries: 1024 (order: 0, 4096 bytes)
283 Page-cache hash table entries: 8192 (order: 3, 32768 bytes)
284 CPU: Intel Pentium Pro stepping 03
285 Checking 'hlt' instruction... OK.
286 POSIX conformance testing by UNIFIX
287 Linux NET4.0 for Linux 2.4
288 Based upon Swansea University Computer Society NET3.039
289 Initializing RT netlink socket
292 pty: 256 Unix98 ptys configured
293 Serial driver version 5.05c (2001-07-08) with no serial options enabled
294 ttyS00 at 0x03f8 (irq = 4) is a 16450
295 ne.c:v1.10 9/23/94 Donald Becker (becker@scyld.com)
296 Last modified Nov 1, 2000 by Paul Gortmaker
297 NE*000 ethercard probe at 0x300: 52 54 00 12 34 56
298 eth0: NE2000 found at 0x300, using IRQ 9.
299 RAMDISK driver initialized: 16 RAM disks of 6144K size 1024 blocksize
300 NET4: Linux TCP/IP 1.0 for NET4.0
301 IP Protocols: ICMP, UDP, TCP, IGMP
302 IP: routing cache hash table of 512 buckets, 4Kbytes
303 TCP: Hash tables configured (established 2048 bind 2048)
304 NET4: Unix domain sockets 1.0/SMP for Linux NET4.0.
305 RAMDISK: ext2 filesystem found at block 0
306 RAMDISK: Loading 6144 blocks [1 disk] into ram disk... done.
307 Freeing initrd memory: 6144k freed
308 VFS: Mounted root (ext2 filesystem).
309 Freeing unused kernel memory: 48k freed
310 sh: can't access tty; job control turned off
315 Then you can play with the kernel inside the virtual serial console. You
316 can launch @code{ls} for example. Type @key{Ctrl-a h} to have an help
317 about the keys you can type inside the virtual serial console. In
318 particular, use @key{Ctrl-a x} to exit QEMU and use @key{Ctrl-a b} as
322 If the network is enabled, launch the script @file{/etc/linuxrc} in the
323 emulator (don't forget the leading dot):
328 Then enable X11 connections on your PC from the emulated Linux:
333 You can now launch @file{xterm} or @file{xlogo} and verify that you have
334 a real Virtual Linux system !
341 A 2.5.66 kernel is also included in the vl-test archive. Just
342 replace the bzImage in vl.sh to try it.
345 vl creates a temporary file in @var{$VLTMPDIR} (@file{/tmp} is the
346 default) containing all the simulated PC memory. If possible, try to use
347 a temporary directory using the tmpfs filesystem to avoid too many
348 unnecessary disk accesses.
351 The example initrd is a modified version of the one made by Kevin
352 Lawton for the plex86 Project (@url{www.plex86.org}).
356 @section Kernel Compilation
358 You can use any Linux kernel within QEMU provided it is mapped at
359 address 0x90000000 (the default is 0xc0000000). You must modify only two
360 lines in the kernel source:
362 In asm/page.h, replace
364 #define __PAGE_OFFSET (0xc0000000)
368 #define __PAGE_OFFSET (0x90000000)
371 And in arch/i386/vmlinux.lds, replace
373 . = 0xc0000000 + 0x100000;
377 . = 0x90000000 + 0x100000;
380 The file config-2.4.20 gives the configuration of the example kernel.
387 As you would do to make a real kernel. Then you can use with QEMU
388 exactly the same kernel as you would boot on your PC (in
389 @file{arch/i386/boot/bzImage}).
391 If you are not using a 2.5 kernel as host kernel but if you use a target
392 2.5 kernel, you must also ensure that the 'HZ' define is set to 100
393 (1000 is the default) as QEMU cannot currently emulate timers at
394 frequencies greater than 100 Hz on host Linux systems < 2.5. In
395 asm/param.h, replace:
398 # define HZ 1000 /* Internal kernel timer frequency */
402 # define HZ 100 /* Internal kernel timer frequency */
405 If you have problems running your kernel, verify that neither the SMP nor
406 HIGHMEM configuration options are activated.
408 @section PC Emulation
410 QEMU emulates the following PC peripherials:
414 PIC (interrupt controler)
420 Serial port (port=0x3f8, irq=4)
422 NE2000 network adapter (port=0x300, irq=9)
424 Dumb VGA (to print the @code{Uncompressing Linux} message)
429 QEMU has a primitive support to work with gdb, so that you can do
430 'Ctrl-C' while the kernel is running and inspect its state.
432 In order to use gdb, launch vl with the '-s' option. It will wait for a
435 > vl -s arch/i386/boot/bzImage initrd-2.4.20.img root=/dev/ram0 ramdisk_size=6144
436 Connected to host network interface: tun0
437 Waiting gdb connection on port 1234
440 Then launch gdb on the 'vmlinux' executable:
445 In gdb, connect to QEMU:
447 (gdb) target remote locahost:1234
450 Then you can use gdb normally. For example, type 'c' to launch the kernel:
455 WARNING: breakpoints and single stepping are not yet supported.
457 @chapter QEMU Internals
459 @section QEMU compared to other emulators
461 Like bochs [3], QEMU emulates an x86 CPU. But QEMU is much faster than
462 bochs as it uses dynamic compilation and because it uses the host MMU to
463 simulate the x86 MMU. The downside is that currently the emulation is
464 not as accurate as bochs (for example, you cannot currently run Windows
467 Like Valgrind [2], QEMU does user space emulation and dynamic
468 translation. Valgrind is mainly a memory debugger while QEMU has no
469 support for it (QEMU could be used to detect out of bound memory
470 accesses as Valgrind, but it has no support to track uninitialised data
471 as Valgrind does). The Valgrind dynamic translator generates better code
472 than QEMU (in particular it does register allocation) but it is closely
473 tied to an x86 host and target and has no support for precise exceptions
474 and system emulation.
476 EM86 [4] is the closest project to user space QEMU (and QEMU still uses
477 some of its code, in particular the ELF file loader). EM86 was limited
478 to an alpha host and used a proprietary and slow interpreter (the
479 interpreter part of the FX!32 Digital Win32 code translator [5]).
481 TWIN [6] is a Windows API emulator like Wine. It is less accurate than
482 Wine but includes a protected mode x86 interpreter to launch x86 Windows
483 executables. Such an approach as greater potential because most of the
484 Windows API is executed natively but it is far more difficult to develop
485 because all the data structures and function parameters exchanged
486 between the API and the x86 code must be converted.
488 User mode Linux [7] was the only solution before QEMU to launch a Linux
489 kernel as a process while not needing any host kernel patches. However,
490 user mode Linux requires heavy kernel patches while QEMU accepts
491 unpatched Linux kernels. It would be interesting to compare the
492 performance of the two approaches.
494 The new Plex86 [8] PC virtualizer is done in the same spirit as the QEMU
495 system emulator. It requires a patched Linux kernel to work (you cannot
496 launch the same kernel on your PC), but the patches are really small. As
497 it is a PC virtualizer (no emulation is done except for some priveledged
498 instructions), it has the potential of being faster than QEMU. The
499 downside is that a complicated (and potentially unsafe) host kernel
502 @section Portable dynamic translation
504 QEMU is a dynamic translator. When it first encounters a piece of code,
505 it converts it to the host instruction set. Usually dynamic translators
506 are very complicated and highly CPU dependent. QEMU uses some tricks
507 which make it relatively easily portable and simple while achieving good
510 The basic idea is to split every x86 instruction into fewer simpler
511 instructions. Each simple instruction is implemented by a piece of C
512 code (see @file{op-i386.c}). Then a compile time tool (@file{dyngen})
513 takes the corresponding object file (@file{op-i386.o}) to generate a
514 dynamic code generator which concatenates the simple instructions to
515 build a function (see @file{op-i386.h:dyngen_code()}).
517 In essence, the process is similar to [1], but more work is done at
520 A key idea to get optimal performances is that constant parameters can
521 be passed to the simple operations. For that purpose, dummy ELF
522 relocations are generated with gcc for each constant parameter. Then,
523 the tool (@file{dyngen}) can locate the relocations and generate the
524 appriopriate C code to resolve them when building the dynamic code.
526 That way, QEMU is no more difficult to port than a dynamic linker.
528 To go even faster, GCC static register variables are used to keep the
529 state of the virtual CPU.
531 @section Register allocation
533 Since QEMU uses fixed simple instructions, no efficient register
534 allocation can be done. However, because RISC CPUs have a lot of
535 register, most of the virtual CPU state can be put in registers without
536 doing complicated register allocation.
538 @section Condition code optimisations
540 Good CPU condition codes emulation (@code{EFLAGS} register on x86) is a
541 critical point to get good performances. QEMU uses lazy condition code
542 evaluation: instead of computing the condition codes after each x86
543 instruction, it just stores one operand (called @code{CC_SRC}), the
544 result (called @code{CC_DST}) and the type of operation (called
547 @code{CC_OP} is almost never explicitely set in the generated code
548 because it is known at translation time.
550 In order to increase performances, a backward pass is performed on the
551 generated simple instructions (see
552 @code{translate-i386.c:optimize_flags()}). When it can be proved that
553 the condition codes are not needed by the next instructions, no
554 condition codes are computed at all.
556 @section CPU state optimisations
558 The x86 CPU has many internal states which change the way it evaluates
559 instructions. In order to achieve a good speed, the translation phase
560 considers that some state information of the virtual x86 CPU cannot
561 change in it. For example, if the SS, DS and ES segments have a zero
562 base, then the translator does not even generate an addition for the
565 [The FPU stack pointer register is not handled that way yet].
567 @section Translation cache
569 A 2MByte cache holds the most recently used translations. For
570 simplicity, it is completely flushed when it is full. A translation unit
571 contains just a single basic block (a block of x86 instructions
572 terminated by a jump or by a virtual CPU state change which the
573 translator cannot deduce statically).
575 @section Direct block chaining
577 After each translated basic block is executed, QEMU uses the simulated
578 Program Counter (PC) and other cpu state informations (such as the CS
579 segment base value) to find the next basic block.
581 In order to accelerate the most common cases where the new simulated PC
582 is known, QEMU can patch a basic block so that it jumps directly to the
585 The most portable code uses an indirect jump. An indirect jump makes it
586 easier to make the jump target modification atomic. On some
587 architectures (such as PowerPC), the @code{JUMP} opcode is directly
588 patched so that the block chaining has no overhead.
590 @section Self-modifying code and translated code invalidation
592 Self-modifying code is a special challenge in x86 emulation because no
593 instruction cache invalidation is signaled by the application when code
596 When translated code is generated for a basic block, the corresponding
597 host page is write protected if it is not already read-only (with the
598 system call @code{mprotect()}). Then, if a write access is done to the
599 page, Linux raises a SEGV signal. QEMU then invalidates all the
600 translated code in the page and enables write accesses to the page.
602 Correct translated code invalidation is done efficiently by maintaining
603 a linked list of every translated block contained in a given page. Other
604 linked lists are also maintained to undo direct block chaining.
606 Althought the overhead of doing @code{mprotect()} calls is important,
607 most MSDOS programs can be emulated at reasonnable speed with QEMU and
610 Note that QEMU also invalidates pages of translated code when it detects
611 that memory mappings are modified with @code{mmap()} or @code{munmap()}.
613 @section Exception support
615 longjmp() is used when an exception such as division by zero is
618 The host SIGSEGV and SIGBUS signal handlers are used to get invalid
619 memory accesses. The exact CPU state can be retrieved because all the
620 x86 registers are stored in fixed host registers. The simulated program
621 counter is found by retranslating the corresponding basic block and by
622 looking where the host program counter was at the exception point.
624 The virtual CPU cannot retrieve the exact @code{EFLAGS} register because
625 in some cases it is not computed because of condition code
626 optimisations. It is not a big concern because the emulated code can
627 still be restarted in any cases.
629 @section Linux system call translation
631 QEMU includes a generic system call translator for Linux. It means that
632 the parameters of the system calls can be converted to fix the
633 endianness and 32/64 bit issues. The IOCTLs are converted with a generic
634 type description system (see @file{ioctls.h} and @file{thunk.c}).
636 QEMU supports host CPUs which have pages bigger than 4KB. It records all
637 the mappings the process does and try to emulated the @code{mmap()}
638 system calls in cases where the host @code{mmap()} call would fail
639 because of bad page alignment.
641 @section Linux signals
643 Normal and real-time signals are queued along with their information
644 (@code{siginfo_t}) as it is done in the Linux kernel. Then an interrupt
645 request is done to the virtual CPU. When it is interrupted, one queued
646 signal is handled by generating a stack frame in the virtual CPU as the
647 Linux kernel does. The @code{sigreturn()} system call is emulated to return
648 from the virtual signal handler.
650 Some signals (such as SIGALRM) directly come from the host. Other
651 signals are synthetized from the virtual CPU exceptions such as SIGFPE
652 when a division by zero is done (see @code{main.c:cpu_loop()}).
654 The blocked signal mask is still handled by the host Linux kernel so
655 that most signal system calls can be redirected directly to the host
656 Linux kernel. Only the @code{sigaction()} and @code{sigreturn()} system
657 calls need to be fully emulated (see @file{signal.c}).
659 @section clone() system call and threads
661 The Linux clone() system call is usually used to create a thread. QEMU
662 uses the host clone() system call so that real host threads are created
663 for each emulated thread. One virtual CPU instance is created for each
666 The virtual x86 CPU atomic operations are emulated with a global lock so
667 that their semantic is preserved.
669 Note that currently there are still some locking issues in QEMU. In
670 particular, the translated cache flush is not protected yet against
673 @section Self-virtualization
675 QEMU was conceived so that ultimately it can emulate itself. Althought
676 it is not very useful, it is an important test to show the power of the
679 Achieving self-virtualization is not easy because there may be address
680 space conflicts. QEMU solves this problem by being an executable ELF
681 shared object as the ld-linux.so ELF interpreter. That way, it can be
682 relocated at load time.
684 @section MMU emulation
686 For system emulation, QEMU uses the mmap() system call to emulate the
687 target CPU MMU. It works as long the emulated OS does not use an area
688 reserved by the host OS (such as the area above 0xc0000000 on x86
691 It is planned to add a slower but more precise MMU emulation
694 @section Bibliography
699 @url{http://citeseer.nj.nec.com/piumarta98optimizing.html}, Optimizing
700 direct threaded code by selective inlining (1998) by Ian Piumarta, Fabio
704 @url{http://developer.kde.org/~sewardj/}, Valgrind, an open-source
705 memory debugger for x86-GNU/Linux, by Julian Seward.
708 @url{http://bochs.sourceforge.net/}, the Bochs IA-32 Emulator Project,
709 by Kevin Lawton et al.
712 @url{http://www.cs.rose-hulman.edu/~donaldlf/em86/index.html}, the EM86
713 x86 emulator on Alpha-Linux.
716 @url{http://www.usenix.org/publications/library/proceedings/usenix-nt97/full_papers/chernoff/chernoff.pdf},
717 DIGITAL FX!32: Running 32-Bit x86 Applications on Alpha NT, by Anton
718 Chernoff and Ray Hookway.
721 @url{http://www.willows.com/}, Windows API library emulation from
725 @url{http://user-mode-linux.sourceforge.net/},
726 The User-mode Linux Kernel.
729 @url{http://www.plex86.org/},
730 The new Plex86 project.
734 @chapter Regression Tests
736 In the directory @file{tests/}, various interesting testing programs
737 are available. There are used for regression testing.
739 @section @file{hello-i386}
741 Very simple statically linked x86 program, just to test QEMU during a
742 port to a new host CPU.
744 @section @file{hello-arm}
746 Very simple statically linked ARM program, just to test QEMU during a
747 port to a new host CPU.
749 @section @file{test-i386}
751 This program executes most of the 16 bit and 32 bit x86 instructions and
752 generates a text output. It can be compared with the output obtained with
753 a real CPU or another emulator. The target @code{make test} runs this
754 program and a @code{diff} on the generated output.
756 The Linux system call @code{modify_ldt()} is used to create x86 selectors
757 to test some 16 bit addressing and 32 bit with segmentation cases.
759 The Linux system call @code{vm86()} is used to test vm86 emulation.
761 Various exceptions are raised to test most of the x86 user space
766 It is a simple benchmark. Care must be taken to interpret the results
767 because it mostly tests the ability of the virtual CPU to optimize the
768 @code{rol} x86 instruction and the condition code computations.