]>
Commit | Line | Data |
---|---|---|
1 | \input texinfo @c -*- texinfo -*- | |
2 | ||
3 | @settitle QEMU x86 Emulator Reference Documentation | |
4 | @titlepage | |
5 | @sp 7 | |
6 | @center @titlefont{QEMU x86 Emulator Reference Documentation} | |
7 | @sp 3 | |
8 | @end titlepage | |
9 | ||
10 | @chapter Introduction | |
11 | ||
12 | QEMU is an x86 processor emulator. Its purpose is to run x86 Linux | |
13 | processes on non-x86 Linux architectures such as PowerPC or ARM. By | |
14 | using dynamic translation it achieves a reasonnable speed while being | |
15 | easy to port on new host CPUs. Its main goal is to be able to launch the | |
16 | @code{Wine} Windows API emulator (@url{http://www.winehq.org}) on | |
17 | non-x86 CPUs. | |
18 | ||
19 | QEMU features: | |
20 | ||
21 | @itemize | |
22 | ||
23 | @item User space only x86 emulator. | |
24 | ||
25 | @item Currently ported on i386, PowerPC and S390. | |
26 | ||
27 | @item Using dynamic translation to native code for reasonnable speed. | |
28 | ||
29 | @item The virtual x86 CPU supports 16 bit and 32 bit addressing with segmentation. | |
30 | User space LDT and GDT are emulated. VM86 mode is also supported | |
31 | (experimental). | |
32 | ||
33 | @item Generic Linux system call converter, including most ioctls. | |
34 | ||
35 | @item clone() emulation using native CPU clone() to use Linux scheduler for threads. | |
36 | ||
37 | @item Accurate signal handling by remapping host signals to virtual x86 signals. | |
38 | ||
39 | @item QEMU can emulate itself on x86 (experimental). | |
40 | ||
41 | @item The virtual x86 CPU is a library (@code{libqemu}) which can be used | |
42 | in other projects. | |
43 | ||
44 | @item An extensive Linux x86 CPU test program is included @file{tests/test-i386}. | |
45 | It can be used to test other x86 virtual CPUs. | |
46 | ||
47 | @end itemize | |
48 | ||
49 | Current QEMU Limitations: | |
50 | ||
51 | @itemize | |
52 | ||
53 | @item Not all x86 exceptions are precise (yet). [Very few programs need that]. | |
54 | ||
55 | @item No support for self-modifying code (yet). [Very few programs need that, a notable exception is QEMU itself !]. | |
56 | ||
57 | @item No SSE/MMX support (yet). | |
58 | ||
59 | @item No x86-64 support. | |
60 | ||
61 | @item Some Linux syscalls are missing. | |
62 | ||
63 | @item The x86 segment limits and access rights are not tested at every | |
64 | memory access (and will never be to have good performances). | |
65 | ||
66 | @item On non x86 host CPUs, @code{double}s are used instead of the non standard | |
67 | 10 byte @code{long double}s of x86 for floating point emulation to get | |
68 | maximum performances. | |
69 | ||
70 | @end itemize | |
71 | ||
72 | @chapter Invocation | |
73 | ||
74 | @section Quick Start | |
75 | ||
76 | In order to launch a Linux process, QEMU needs the process executable | |
77 | itself and all the target (x86) dynamic libraries used by it. | |
78 | ||
79 | @itemize | |
80 | ||
81 | @item On x86, you can just try to launch any process by using the native | |
82 | libraries: | |
83 | ||
84 | @example | |
85 | qemu -L / /bin/ls | |
86 | @end example | |
87 | ||
88 | @code{-L /} tells that the x86 dynamic linker must be searched with a | |
89 | @file{/} prefix. | |
90 | ||
91 | @item Since QEMU is also a linux process, you can launch qemu with qemu: | |
92 | ||
93 | @example | |
94 | qemu -L / qemu -L / /bin/ls | |
95 | @end example | |
96 | ||
97 | @item On non x86 CPUs, you need first to download at least an x86 glibc | |
98 | (@file{qemu-XXX-i386-glibc21.tar.gz} on the QEMU web page). Ensure that | |
99 | @code{LD_LIBRARY_PATH} is not set: | |
100 | ||
101 | @example | |
102 | unset LD_LIBRARY_PATH | |
103 | @end example | |
104 | ||
105 | Then you can launch the precompiled @file{ls} x86 executable: | |
106 | ||
107 | @example | |
108 | qemu /usr/local/qemu-i386/bin/ls-i386 | |
109 | @end example | |
110 | You can look at @file{/usr/local/qemu-i386/bin/qemu-conf.sh} so that | |
111 | QEMU is automatically launched by the Linux kernel when you try to | |
112 | launch x86 executables. It requires the @code{binfmt_misc} module in the | |
113 | Linux kernel. | |
114 | ||
115 | @item The x86 version of QEMU is also included. You can try weird things such as: | |
116 | @example | |
117 | qemu /usr/local/qemu-i386/bin/qemu-i386 /usr/local/qemu-i386/bin/ls-i386 | |
118 | @end example | |
119 | ||
120 | @end itemize | |
121 | ||
122 | @section Wine launch (Currently only tested when emulating x86 on x86) | |
123 | ||
124 | @itemize | |
125 | ||
126 | @item Ensure that you have a working QEMU with the x86 glibc | |
127 | distribution (see previous section). In order to verify it, you must be | |
128 | able to do: | |
129 | ||
130 | @example | |
131 | qemu /usr/local/qemu-i386/bin/ls-i386 | |
132 | @end example | |
133 | ||
134 | @item Download the binary x86 Wine install | |
135 | (@file{qemu-XXX-i386-wine.tar.gz} on the QEMU web page). | |
136 | ||
137 | @item Configure Wine on your account. Look at the provided script | |
138 | @file{/usr/local/qemu-i386/bin/wine-conf.sh}. Your previous | |
139 | @code{$@{HOME@}/.wine} directory is saved to @code{$@{HOME@}/.wine.org}. | |
140 | ||
141 | @item Then you can try the example @file{putty.exe}: | |
142 | ||
143 | @example | |
144 | qemu /usr/local/qemu-i386/wine/bin/wine /usr/local/qemu-i386/wine/c/Program\ Files/putty.exe | |
145 | @end example | |
146 | ||
147 | @end itemize | |
148 | ||
149 | @section Command line options | |
150 | ||
151 | @example | |
152 | usage: qemu [-h] [-d] [-L path] [-s size] program [arguments...] | |
153 | @end example | |
154 | ||
155 | @table @samp | |
156 | @item -h | |
157 | Print the help | |
158 | @item -d | |
159 | Activate log (logfile=/tmp/qemu.log) | |
160 | @item -L path | |
161 | Set the x86 elf interpreter prefix (default=/usr/local/qemu-i386) | |
162 | @item -s size | |
163 | Set the x86 stack size in bytes (default=524288) | |
164 | @end table | |
165 | ||
166 | @chapter QEMU Internals | |
167 | ||
168 | @section QEMU compared to other emulators | |
169 | ||
170 | Unlike bochs [3], QEMU emulates only a user space x86 CPU. It means that | |
171 | you cannot launch an operating system with it. The benefit is that it is | |
172 | simpler and faster due to the fact that some of the low level CPU state | |
173 | can be ignored (in particular, no virtual memory needs to be emulated). | |
174 | ||
175 | Like Valgrind [2], QEMU does user space emulation and dynamic | |
176 | translation. Valgrind is mainly a memory debugger while QEMU has no | |
177 | support for it (QEMU could be used to detect out of bound memory accesses | |
178 | as Valgrind, but it has no support to track uninitialised data as | |
179 | Valgrind does). Valgrind dynamic translator generates better code than | |
180 | QEMU (in particular it does register allocation) but it is closely tied | |
181 | to an x86 host. | |
182 | ||
183 | EM86 [4] is the closest project to QEMU (and QEMU still uses some of its | |
184 | code, in particular the ELF file loader). EM86 was limited to an alpha | |
185 | host and used a proprietary and slow interpreter (the interpreter part | |
186 | of the FX!32 Digital Win32 code translator [5]). | |
187 | ||
188 | TWIN [6] is a Windows API emulator like Wine. It is less accurate than | |
189 | Wine but includes a protected mode x86 interpreter to launch x86 Windows | |
190 | executables. Such an approach as greater potential because most of the | |
191 | Windows API is executed natively but it is far more difficult to develop | |
192 | because all the data structures and function parameters exchanged | |
193 | between the API and the x86 code must be converted. | |
194 | ||
195 | @section Portable dynamic translation | |
196 | ||
197 | QEMU is a dynamic translator. When it first encounters a piece of code, | |
198 | it converts it to the host instruction set. Usually dynamic translators | |
199 | are very complicated and highly CPU dependant. QEMU uses some tricks | |
200 | which make it relatively easily portable and simple while achieving good | |
201 | performances. | |
202 | ||
203 | The basic idea is to split every x86 instruction into fewer simpler | |
204 | instructions. Each simple instruction is implemented by a piece of C | |
205 | code (see @file{op-i386.c}). Then a compile time tool (@file{dyngen}) | |
206 | takes the corresponding object file (@file{op-i386.o}) to generate a | |
207 | dynamic code generator which concatenates the simple instructions to | |
208 | build a function (see @file{op-i386.h:dyngen_code()}). | |
209 | ||
210 | In essence, the process is similar to [1], but more work is done at | |
211 | compile time. | |
212 | ||
213 | A key idea to get optimal performances is that constant parameters can | |
214 | be passed to the simple operations. For that purpose, dummy ELF | |
215 | relocations are generated with gcc for each constant parameter. Then, | |
216 | the tool (@file{dyngen}) can locate the relocations and generate the | |
217 | appriopriate C code to resolve them when building the dynamic code. | |
218 | ||
219 | That way, QEMU is no more difficult to port than a dynamic linker. | |
220 | ||
221 | To go even faster, GCC static register variables are used to keep the | |
222 | state of the virtual CPU. | |
223 | ||
224 | @section Register allocation | |
225 | ||
226 | Since QEMU uses fixed simple instructions, no efficient register | |
227 | allocation can be done. However, because RISC CPUs have a lot of | |
228 | register, most of the virtual CPU state can be put in registers without | |
229 | doing complicated register allocation. | |
230 | ||
231 | @section Condition code optimisations | |
232 | ||
233 | Good CPU condition codes emulation (@code{EFLAGS} register on x86) is a | |
234 | critical point to get good performances. QEMU uses lazy condition code | |
235 | evaluation: instead of computing the condition codes after each x86 | |
236 | instruction, it just stores one operand (called @code{CC_SRC}), the | |
237 | result (called @code{CC_DST}) and the type of operation (called | |
238 | @code{CC_OP}). | |
239 | ||
240 | @code{CC_OP} is almost never explicitely set in the generated code | |
241 | because it is known at translation time. | |
242 | ||
243 | In order to increase performances, a backward pass is performed on the | |
244 | generated simple instructions (see | |
245 | @code{translate-i386.c:optimize_flags()}). When it can be proved that | |
246 | the condition codes are not needed by the next instructions, no | |
247 | condition codes are computed at all. | |
248 | ||
249 | @section CPU state optimisations | |
250 | ||
251 | The x86 CPU has many internal states which change the way it evaluates | |
252 | instructions. In order to achieve a good speed, the translation phase | |
253 | considers that some state information of the virtual x86 CPU cannot | |
254 | change in it. For example, if the SS, DS and ES segments have a zero | |
255 | base, then the translator does not even generate an addition for the | |
256 | segment base. | |
257 | ||
258 | [The FPU stack pointer register is not handled that way yet]. | |
259 | ||
260 | @section Translation cache | |
261 | ||
262 | A 2MByte cache holds the most recently used translations. For | |
263 | simplicity, it is completely flushed when it is full. A translation unit | |
264 | contains just a single basic block (a block of x86 instructions | |
265 | terminated by a jump or by a virtual CPU state change which the | |
266 | translator cannot deduce statically). | |
267 | ||
268 | [Currently, the translated code is not patched if it jumps to another | |
269 | translated code]. | |
270 | ||
271 | @section Exception support | |
272 | ||
273 | longjmp() is used when an exception such as division by zero is | |
274 | encountered. The host SIGSEGV and SIGBUS signal handlers are used to get | |
275 | invalid memory accesses. | |
276 | ||
277 | [Currently, the virtual CPU cannot retrieve the exact CPU state in some | |
278 | exceptions, although it could except for the @code{EFLAGS} register]. | |
279 | ||
280 | @section Linux system call translation | |
281 | ||
282 | QEMU includes a generic system call translator for Linux. It means that | |
283 | the parameters of the system calls can be converted to fix the | |
284 | endianness and 32/64 bit issues. The IOCTLs are converted with a generic | |
285 | type description system (see @file{ioctls.h} and @file{thunk.c}). | |
286 | ||
287 | @section Linux signals | |
288 | ||
289 | Normal and real-time signals are queued along with their information | |
290 | (@code{siginfo_t}) as it is done in the Linux kernel. Then an interrupt | |
291 | request is done to the virtual CPU. When it is interrupted, one queued | |
292 | signal is handled by generating a stack frame in the virtual CPU as the | |
293 | Linux kernel does. The @code{sigreturn()} system call is emulated to return | |
294 | from the virtual signal handler. | |
295 | ||
296 | Some signals (such as SIGALRM) directly come from the host. Other | |
297 | signals are synthetized from the virtual CPU exceptions such as SIGFPE | |
298 | when a division by zero is done (see @code{main.c:cpu_loop()}). | |
299 | ||
300 | The blocked signal mask is still handled by the host Linux kernel so | |
301 | that most signal system calls can be redirected directly to the host | |
302 | Linux kernel. Only the @code{sigaction()} and @code{sigreturn()} system | |
303 | calls need to be fully emulated (see @file{signal.c}). | |
304 | ||
305 | @section clone() system call and threads | |
306 | ||
307 | The Linux clone() system call is usually used to create a thread. QEMU | |
308 | uses the host clone() system call so that real host threads are created | |
309 | for each emulated thread. One virtual CPU instance is created for each | |
310 | thread. | |
311 | ||
312 | The virtual x86 CPU atomic operations are emulated with a global lock so | |
313 | that their semantic is preserved. | |
314 | ||
315 | @section Self-virtualization | |
316 | ||
317 | QEMU was conceived so that ultimately it can emulate itself. Althought | |
318 | it is not very useful, it is an important test to show the power of the | |
319 | emulator. | |
320 | ||
321 | Achieving self-virtualization is not easy because there may be address | |
322 | space conflicts. QEMU solves this problem by being an executable ELF | |
323 | shared object as the ld-linux.so ELF interpreter. That way, it can be | |
324 | relocated at load time. | |
325 | ||
326 | Since self-modifying code is not supported yet, QEMU cannot emulate | |
327 | itself in case of translation cache flush. This limitation will be | |
328 | suppressed soon. | |
329 | ||
330 | @section Bibliography | |
331 | ||
332 | @table @asis | |
333 | ||
334 | @item [1] | |
335 | @url{http://citeseer.nj.nec.com/piumarta98optimizing.html}, Optimizing | |
336 | direct threaded code by selective inlining (1998) by Ian Piumarta, Fabio | |
337 | Riccardi. | |
338 | ||
339 | @item [2] | |
340 | @url{http://developer.kde.org/~sewardj/}, Valgrind, an open-source | |
341 | memory debugger for x86-GNU/Linux, by Julian Seward. | |
342 | ||
343 | @item [3] | |
344 | @url{http://bochs.sourceforge.net/}, the Bochs IA-32 Emulator Project, | |
345 | by Kevin Lawton et al. | |
346 | ||
347 | @item [4] | |
348 | @url{http://www.cs.rose-hulman.edu/~donaldlf/em86/index.html}, the EM86 | |
349 | x86 emulator on Alpha-Linux. | |
350 | ||
351 | @item [5] | |
352 | @url{http://www.usenix.org/publications/library/proceedings/usenix-nt97/full_papers/chernoff/chernoff.pdf}, | |
353 | DIGITAL FX!32: Running 32-Bit x86 Applications on Alpha NT, by Anton | |
354 | Chernoff and Ray Hookway. | |
355 | ||
356 | @item [6] | |
357 | @url{http://www.willows.com/}, Windows API library emulation from | |
358 | Willows Software. | |
359 | ||
360 | @end table | |
361 | ||
362 | @chapter Regression Tests | |
363 | ||
364 | In the directory @file{tests/}, various interesting x86 testing programs | |
365 | are available. There are used for regression testing. | |
366 | ||
367 | @section @file{hello} | |
368 | ||
369 | Very simple statically linked x86 program, just to test QEMU during a | |
370 | port to a new host CPU. | |
371 | ||
372 | @section @file{test-i386} | |
373 | ||
374 | This program executes most of the 16 bit and 32 bit x86 instructions and | |
375 | generates a text output. It can be compared with the output obtained with | |
376 | a real CPU or another emulator. The target @code{make test} runs this | |
377 | program and a @code{diff} on the generated output. | |
378 | ||
379 | The Linux system call @code{modify_ldt()} is used to create x86 selectors | |
380 | to test some 16 bit addressing and 32 bit with segmentation cases. | |
381 | ||
382 | @section @file{testsig} | |
383 | ||
384 | This program tests various signal cases, including SIGFPE, SIGSEGV and | |
385 | SIGILL. | |
386 | ||
387 | @section @file{testclone} | |
388 | ||
389 | Tests the @code{clone()} system call (basic test). | |
390 | ||
391 | @section @file{testthread} | |
392 | ||
393 | Tests the glibc threads (more complicated than @code{clone()} because signals | |
394 | are also used). | |
395 | ||
396 | @section @file{sha1} | |
397 | ||
398 | It is a simple benchmark. Care must be taken to interpret the results | |
399 | because it mostly tests the ability of the virtual CPU to optimize the | |
400 | @code{rol} x86 instruction and the condition code computations. | |
401 | ||
402 | @section @file{runcom} | |
403 | ||
404 | A very simple MSDOS emulator to test the Linux vm86() system call | |
405 | emulation. The excellent 54 byte @file{pi_10.com} PI number calculator | |
406 | can be launched with it. @file{pi_10.com} was written by Bertram | |
407 | Felgenhauer (more information at @url{http://www.boo.net/~jasonp/pipage.html}). |