root / qemu-tech.texi @ ed23fbd9
History | View | Annotate | Download (20 kB)
1 | 1f673135 | bellard | \input texinfo @c -*- texinfo -*- |
---|---|---|---|
2 | debc7065 | bellard | @c %**start of header |
3 | debc7065 | bellard | @setfilename qemu-tech.info |
4 | debc7065 | bellard | @settitle QEMU Internals |
5 | debc7065 | bellard | @exampleindent 0 |
6 | debc7065 | bellard | @paragraphindent 0 |
7 | debc7065 | bellard | @c %**end of header |
8 | 1f673135 | bellard | |
9 | 1f673135 | bellard | @iftex |
10 | 1f673135 | bellard | @titlepage |
11 | 1f673135 | bellard | @sp 7 |
12 | 1f673135 | bellard | @center @titlefont{QEMU Internals} |
13 | 1f673135 | bellard | @sp 3 |
14 | 1f673135 | bellard | @end titlepage |
15 | 1f673135 | bellard | @end iftex |
16 | 1f673135 | bellard | |
17 | debc7065 | bellard | @ifnottex |
18 | debc7065 | bellard | @node Top |
19 | debc7065 | bellard | @top |
20 | debc7065 | bellard | |
21 | debc7065 | bellard | @menu |
22 | debc7065 | bellard | * Introduction:: |
23 | debc7065 | bellard | * QEMU Internals:: |
24 | debc7065 | bellard | * Regression Tests:: |
25 | debc7065 | bellard | * Index:: |
26 | debc7065 | bellard | @end menu |
27 | debc7065 | bellard | @end ifnottex |
28 | debc7065 | bellard | |
29 | debc7065 | bellard | @contents |
30 | debc7065 | bellard | |
31 | debc7065 | bellard | @node Introduction |
32 | 1f673135 | bellard | @chapter Introduction |
33 | 1f673135 | bellard | |
34 | debc7065 | bellard | @menu |
35 | debc7065 | bellard | * intro_features:: Features |
36 | debc7065 | bellard | * intro_x86_emulation:: x86 emulation |
37 | debc7065 | bellard | * intro_arm_emulation:: ARM emulation |
38 | 24d4de45 | ths | * intro_mips_emulation:: MIPS emulation |
39 | debc7065 | bellard | * intro_ppc_emulation:: PowerPC emulation |
40 | debc7065 | bellard | * intro_sparc_emulation:: SPARC emulation |
41 | debc7065 | bellard | @end menu |
42 | debc7065 | bellard | |
43 | debc7065 | bellard | @node intro_features |
44 | 1f673135 | bellard | @section Features |
45 | 1f673135 | bellard | |
46 | 1f673135 | bellard | QEMU is a FAST! processor emulator using a portable dynamic |
47 | 1f673135 | bellard | translator. |
48 | 1f673135 | bellard | |
49 | 1f673135 | bellard | QEMU has two operating modes: |
50 | 1f673135 | bellard | |
51 | 1f673135 | bellard | @itemize @minus |
52 | 1f673135 | bellard | |
53 | 5fafdf24 | ths | @item |
54 | 1f673135 | bellard | Full system emulation. In this mode, QEMU emulates a full system |
55 | b671f9ed | bellard | (usually a PC), including a processor and various peripherals. It can |
56 | 1f673135 | bellard | be used to launch an different Operating System without rebooting the |
57 | 1f673135 | bellard | PC or to debug system code. |
58 | 1f673135 | bellard | |
59 | 5fafdf24 | ths | @item |
60 | 1f673135 | bellard | User mode emulation (Linux host only). In this mode, QEMU can launch |
61 | 1f673135 | bellard | Linux processes compiled for one CPU on another CPU. It can be used to |
62 | 1f673135 | bellard | launch the Wine Windows API emulator (@url{http://www.winehq.org}) or |
63 | 1f673135 | bellard | to ease cross-compilation and cross-debugging. |
64 | 1f673135 | bellard | |
65 | 1f673135 | bellard | @end itemize |
66 | 1f673135 | bellard | |
67 | 1f673135 | bellard | As QEMU requires no host kernel driver to run, it is very safe and |
68 | 1f673135 | bellard | easy to use. |
69 | 1f673135 | bellard | |
70 | 1f673135 | bellard | QEMU generic features: |
71 | 1f673135 | bellard | |
72 | 5fafdf24 | ths | @itemize |
73 | 1f673135 | bellard | |
74 | 1f673135 | bellard | @item User space only or full system emulation. |
75 | 1f673135 | bellard | |
76 | debc7065 | bellard | @item Using dynamic translation to native code for reasonable speed. |
77 | 1f673135 | bellard | |
78 | 1f673135 | bellard | @item Working on x86 and PowerPC hosts. Being tested on ARM, Sparc32, Alpha and S390. |
79 | 1f673135 | bellard | |
80 | 1f673135 | bellard | @item Self-modifying code support. |
81 | 1f673135 | bellard | |
82 | 1f673135 | bellard | @item Precise exceptions support. |
83 | 1f673135 | bellard | |
84 | 5fafdf24 | ths | @item The virtual CPU is a library (@code{libqemu}) which can be used |
85 | ad6a4837 | bellard | in other projects (look at @file{qemu/tests/qruncom.c} to have an |
86 | ad6a4837 | bellard | example of user mode @code{libqemu} usage). |
87 | 1f673135 | bellard | |
88 | 1f673135 | bellard | @end itemize |
89 | 1f673135 | bellard | |
90 | 1f673135 | bellard | QEMU user mode emulation features: |
91 | 5fafdf24 | ths | @itemize |
92 | 1f673135 | bellard | @item Generic Linux system call converter, including most ioctls. |
93 | 1f673135 | bellard | |
94 | 1f673135 | bellard | @item clone() emulation using native CPU clone() to use Linux scheduler for threads. |
95 | 1f673135 | bellard | |
96 | 5fafdf24 | ths | @item Accurate signal handling by remapping host signals to target signals. |
97 | 1f673135 | bellard | @end itemize |
98 | 1f673135 | bellard | |
99 | 1f673135 | bellard | QEMU full system emulation features: |
100 | 5fafdf24 | ths | @itemize |
101 | 1f673135 | bellard | @item QEMU can either use a full software MMU for maximum portability or use the host system call mmap() to simulate the target MMU. |
102 | 1f673135 | bellard | @end itemize |
103 | 1f673135 | bellard | |
104 | debc7065 | bellard | @node intro_x86_emulation |
105 | 1f673135 | bellard | @section x86 emulation |
106 | 1f673135 | bellard | |
107 | 1f673135 | bellard | QEMU x86 target features: |
108 | 1f673135 | bellard | |
109 | 5fafdf24 | ths | @itemize |
110 | 1f673135 | bellard | |
111 | 5fafdf24 | ths | @item The virtual x86 CPU supports 16 bit and 32 bit addressing with segmentation. |
112 | 1f673135 | bellard | LDT/GDT and IDT are emulated. VM86 mode is also supported to run DOSEMU. |
113 | 1f673135 | bellard | |
114 | 1f673135 | bellard | @item Support of host page sizes bigger than 4KB in user mode emulation. |
115 | 1f673135 | bellard | |
116 | 1f673135 | bellard | @item QEMU can emulate itself on x86. |
117 | 1f673135 | bellard | |
118 | 5fafdf24 | ths | @item An extensive Linux x86 CPU test program is included @file{tests/test-i386}. |
119 | 1f673135 | bellard | It can be used to test other x86 virtual CPUs. |
120 | 1f673135 | bellard | |
121 | 1f673135 | bellard | @end itemize |
122 | 1f673135 | bellard | |
123 | 1f673135 | bellard | Current QEMU limitations: |
124 | 1f673135 | bellard | |
125 | 5fafdf24 | ths | @itemize |
126 | 1f673135 | bellard | |
127 | 1f673135 | bellard | @item No SSE/MMX support (yet). |
128 | 1f673135 | bellard | |
129 | 1f673135 | bellard | @item No x86-64 support. |
130 | 1f673135 | bellard | |
131 | 1f673135 | bellard | @item IPC syscalls are missing. |
132 | 1f673135 | bellard | |
133 | 5fafdf24 | ths | @item The x86 segment limits and access rights are not tested at every |
134 | 1f673135 | bellard | memory access (yet). Hopefully, very few OSes seem to rely on that for |
135 | 1f673135 | bellard | normal use. |
136 | 1f673135 | bellard | |
137 | 5fafdf24 | ths | @item On non x86 host CPUs, @code{double}s are used instead of the non standard |
138 | 1f673135 | bellard | 10 byte @code{long double}s of x86 for floating point emulation to get |
139 | 1f673135 | bellard | maximum performances. |
140 | 1f673135 | bellard | |
141 | 1f673135 | bellard | @end itemize |
142 | 1f673135 | bellard | |
143 | debc7065 | bellard | @node intro_arm_emulation |
144 | 1f673135 | bellard | @section ARM emulation |
145 | 1f673135 | bellard | |
146 | 1f673135 | bellard | @itemize |
147 | 1f673135 | bellard | |
148 | 1f673135 | bellard | @item Full ARM 7 user emulation. |
149 | 1f673135 | bellard | |
150 | 1f673135 | bellard | @item NWFPE FPU support included in user Linux emulation. |
151 | 1f673135 | bellard | |
152 | 1f673135 | bellard | @item Can run most ARM Linux binaries. |
153 | 1f673135 | bellard | |
154 | 1f673135 | bellard | @end itemize |
155 | 1f673135 | bellard | |
156 | 24d4de45 | ths | @node intro_mips_emulation |
157 | 24d4de45 | ths | @section MIPS emulation |
158 | 24d4de45 | ths | |
159 | 24d4de45 | ths | @itemize |
160 | 24d4de45 | ths | |
161 | 24d4de45 | ths | @item The system emulation allows full MIPS32/MIPS64 Release 2 emulation, |
162 | 24d4de45 | ths | including privileged instructions, FPU and MMU, in both little and big |
163 | 24d4de45 | ths | endian modes. |
164 | 24d4de45 | ths | |
165 | 24d4de45 | ths | @item The Linux userland emulation can run many 32 bit MIPS Linux binaries. |
166 | 24d4de45 | ths | |
167 | 24d4de45 | ths | @end itemize |
168 | 24d4de45 | ths | |
169 | 24d4de45 | ths | Current QEMU limitations: |
170 | 24d4de45 | ths | |
171 | 24d4de45 | ths | @itemize |
172 | 24d4de45 | ths | |
173 | 24d4de45 | ths | @item Self-modifying code is not always handled correctly. |
174 | 24d4de45 | ths | |
175 | 24d4de45 | ths | @item 64 bit userland emulation is not implemented. |
176 | 24d4de45 | ths | |
177 | 24d4de45 | ths | @item The system emulation is not complete enough to run real firmware. |
178 | 24d4de45 | ths | |
179 | b1f45238 | ths | @item The watchpoint debug facility is not implemented. |
180 | b1f45238 | ths | |
181 | 24d4de45 | ths | @end itemize |
182 | 24d4de45 | ths | |
183 | debc7065 | bellard | @node intro_ppc_emulation |
184 | 1f673135 | bellard | @section PowerPC emulation |
185 | 1f673135 | bellard | |
186 | 1f673135 | bellard | @itemize |
187 | 1f673135 | bellard | |
188 | 5fafdf24 | ths | @item Full PowerPC 32 bit emulation, including privileged instructions, |
189 | 1f673135 | bellard | FPU and MMU. |
190 | 1f673135 | bellard | |
191 | 1f673135 | bellard | @item Can run most PowerPC Linux binaries. |
192 | 1f673135 | bellard | |
193 | 1f673135 | bellard | @end itemize |
194 | 1f673135 | bellard | |
195 | debc7065 | bellard | @node intro_sparc_emulation |
196 | 1f673135 | bellard | @section SPARC emulation |
197 | 1f673135 | bellard | |
198 | 1f673135 | bellard | @itemize |
199 | 1f673135 | bellard | |
200 | f6b647cd | blueswir1 | @item Full SPARC V8 emulation, including privileged |
201 | 3475187d | bellard | instructions, FPU and MMU. SPARC V9 emulation includes most privileged |
202 | a785e42e | blueswir1 | and VIS instructions, FPU and I/D MMU. Alignment is fully enforced. |
203 | 1f673135 | bellard | |
204 | a785e42e | blueswir1 | @item Can run most 32-bit SPARC Linux binaries, SPARC32PLUS Linux binaries and |
205 | a785e42e | blueswir1 | some 64-bit SPARC Linux binaries. |
206 | 3475187d | bellard | |
207 | 3475187d | bellard | @end itemize |
208 | 3475187d | bellard | |
209 | 3475187d | bellard | Current QEMU limitations: |
210 | 3475187d | bellard | |
211 | 5fafdf24 | ths | @itemize |
212 | 3475187d | bellard | |
213 | 3475187d | bellard | @item IPC syscalls are missing. |
214 | 3475187d | bellard | |
215 | 1f587329 | blueswir1 | @item Floating point exception support is buggy. |
216 | 3475187d | bellard | |
217 | 3475187d | bellard | @item Atomic instructions are not correctly implemented. |
218 | 3475187d | bellard | |
219 | 3475187d | bellard | @item Sparc64 emulators are not usable for anything yet. |
220 | 1f673135 | bellard | |
221 | 1f673135 | bellard | @end itemize |
222 | 1f673135 | bellard | |
223 | debc7065 | bellard | @node QEMU Internals |
224 | 1f673135 | bellard | @chapter QEMU Internals |
225 | 1f673135 | bellard | |
226 | debc7065 | bellard | @menu |
227 | debc7065 | bellard | * QEMU compared to other emulators:: |
228 | debc7065 | bellard | * Portable dynamic translation:: |
229 | debc7065 | bellard | * Register allocation:: |
230 | debc7065 | bellard | * Condition code optimisations:: |
231 | debc7065 | bellard | * CPU state optimisations:: |
232 | debc7065 | bellard | * Translation cache:: |
233 | debc7065 | bellard | * Direct block chaining:: |
234 | debc7065 | bellard | * Self-modifying code and translated code invalidation:: |
235 | debc7065 | bellard | * Exception support:: |
236 | debc7065 | bellard | * MMU emulation:: |
237 | debc7065 | bellard | * Hardware interrupts:: |
238 | debc7065 | bellard | * User emulation specific details:: |
239 | debc7065 | bellard | * Bibliography:: |
240 | debc7065 | bellard | @end menu |
241 | debc7065 | bellard | |
242 | debc7065 | bellard | @node QEMU compared to other emulators |
243 | 1f673135 | bellard | @section QEMU compared to other emulators |
244 | 1f673135 | bellard | |
245 | 1f673135 | bellard | Like bochs [3], QEMU emulates an x86 CPU. But QEMU is much faster than |
246 | 1f673135 | bellard | bochs as it uses dynamic compilation. Bochs is closely tied to x86 PC |
247 | 1f673135 | bellard | emulation while QEMU can emulate several processors. |
248 | 1f673135 | bellard | |
249 | 1f673135 | bellard | Like Valgrind [2], QEMU does user space emulation and dynamic |
250 | 1f673135 | bellard | translation. Valgrind is mainly a memory debugger while QEMU has no |
251 | 1f673135 | bellard | support for it (QEMU could be used to detect out of bound memory |
252 | 1f673135 | bellard | accesses as Valgrind, but it has no support to track uninitialised data |
253 | 1f673135 | bellard | as Valgrind does). The Valgrind dynamic translator generates better code |
254 | 1f673135 | bellard | than QEMU (in particular it does register allocation) but it is closely |
255 | 1f673135 | bellard | tied to an x86 host and target and has no support for precise exceptions |
256 | 1f673135 | bellard | and system emulation. |
257 | 1f673135 | bellard | |
258 | 1f673135 | bellard | EM86 [4] is the closest project to user space QEMU (and QEMU still uses |
259 | 1f673135 | bellard | some of its code, in particular the ELF file loader). EM86 was limited |
260 | 1f673135 | bellard | to an alpha host and used a proprietary and slow interpreter (the |
261 | 1f673135 | bellard | interpreter part of the FX!32 Digital Win32 code translator [5]). |
262 | 1f673135 | bellard | |
263 | 1f673135 | bellard | TWIN [6] is a Windows API emulator like Wine. It is less accurate than |
264 | 1f673135 | bellard | Wine but includes a protected mode x86 interpreter to launch x86 Windows |
265 | 36d54d15 | bellard | executables. Such an approach has greater potential because most of the |
266 | 1f673135 | bellard | Windows API is executed natively but it is far more difficult to develop |
267 | 1f673135 | bellard | because all the data structures and function parameters exchanged |
268 | 1f673135 | bellard | between the API and the x86 code must be converted. |
269 | 1f673135 | bellard | |
270 | 1f673135 | bellard | User mode Linux [7] was the only solution before QEMU to launch a |
271 | 1f673135 | bellard | Linux kernel as a process while not needing any host kernel |
272 | 1f673135 | bellard | patches. However, user mode Linux requires heavy kernel patches while |
273 | 1f673135 | bellard | QEMU accepts unpatched Linux kernels. The price to pay is that QEMU is |
274 | 1f673135 | bellard | slower. |
275 | 1f673135 | bellard | |
276 | 1f673135 | bellard | The new Plex86 [8] PC virtualizer is done in the same spirit as the |
277 | 1f673135 | bellard | qemu-fast system emulator. It requires a patched Linux kernel to work |
278 | 1f673135 | bellard | (you cannot launch the same kernel on your PC), but the patches are |
279 | 1f673135 | bellard | really small. As it is a PC virtualizer (no emulation is done except |
280 | 1235fc06 | ths | for some privileged instructions), it has the potential of being |
281 | 1f673135 | bellard | faster than QEMU. The downside is that a complicated (and potentially |
282 | 1f673135 | bellard | unsafe) host kernel patch is needed. |
283 | 1f673135 | bellard | |
284 | 1f673135 | bellard | The commercial PC Virtualizers (VMWare [9], VirtualPC [10], TwoOStwo |
285 | 1f673135 | bellard | [11]) are faster than QEMU, but they all need specific, proprietary |
286 | 1f673135 | bellard | and potentially unsafe host drivers. Moreover, they are unable to |
287 | 1f673135 | bellard | provide cycle exact simulation as an emulator can. |
288 | 1f673135 | bellard | |
289 | debc7065 | bellard | @node Portable dynamic translation |
290 | 1f673135 | bellard | @section Portable dynamic translation |
291 | 1f673135 | bellard | |
292 | 1f673135 | bellard | QEMU is a dynamic translator. When it first encounters a piece of code, |
293 | 1f673135 | bellard | it converts it to the host instruction set. Usually dynamic translators |
294 | 1f673135 | bellard | are very complicated and highly CPU dependent. QEMU uses some tricks |
295 | 1f673135 | bellard | which make it relatively easily portable and simple while achieving good |
296 | 1f673135 | bellard | performances. |
297 | 1f673135 | bellard | |
298 | 1f673135 | bellard | The basic idea is to split every x86 instruction into fewer simpler |
299 | 1f673135 | bellard | instructions. Each simple instruction is implemented by a piece of C |
300 | 1f673135 | bellard | code (see @file{target-i386/op.c}). Then a compile time tool |
301 | 1f673135 | bellard | (@file{dyngen}) takes the corresponding object file (@file{op.o}) |
302 | 1f673135 | bellard | to generate a dynamic code generator which concatenates the simple |
303 | 1f673135 | bellard | instructions to build a function (see @file{op.h:dyngen_code()}). |
304 | 1f673135 | bellard | |
305 | 1f673135 | bellard | In essence, the process is similar to [1], but more work is done at |
306 | 5fafdf24 | ths | compile time. |
307 | 1f673135 | bellard | |
308 | 1f673135 | bellard | A key idea to get optimal performances is that constant parameters can |
309 | 1f673135 | bellard | be passed to the simple operations. For that purpose, dummy ELF |
310 | 1f673135 | bellard | relocations are generated with gcc for each constant parameter. Then, |
311 | 1f673135 | bellard | the tool (@file{dyngen}) can locate the relocations and generate the |
312 | 1f673135 | bellard | appriopriate C code to resolve them when building the dynamic code. |
313 | 1f673135 | bellard | |
314 | 1f673135 | bellard | That way, QEMU is no more difficult to port than a dynamic linker. |
315 | 1f673135 | bellard | |
316 | 1f673135 | bellard | To go even faster, GCC static register variables are used to keep the |
317 | 1f673135 | bellard | state of the virtual CPU. |
318 | 1f673135 | bellard | |
319 | debc7065 | bellard | @node Register allocation |
320 | 1f673135 | bellard | @section Register allocation |
321 | 1f673135 | bellard | |
322 | 1f673135 | bellard | Since QEMU uses fixed simple instructions, no efficient register |
323 | 1f673135 | bellard | allocation can be done. However, because RISC CPUs have a lot of |
324 | 1f673135 | bellard | register, most of the virtual CPU state can be put in registers without |
325 | 1f673135 | bellard | doing complicated register allocation. |
326 | 1f673135 | bellard | |
327 | debc7065 | bellard | @node Condition code optimisations |
328 | 1f673135 | bellard | @section Condition code optimisations |
329 | 1f673135 | bellard | |
330 | 1f673135 | bellard | Good CPU condition codes emulation (@code{EFLAGS} register on x86) is a |
331 | 1f673135 | bellard | critical point to get good performances. QEMU uses lazy condition code |
332 | 1f673135 | bellard | evaluation: instead of computing the condition codes after each x86 |
333 | 1f673135 | bellard | instruction, it just stores one operand (called @code{CC_SRC}), the |
334 | 1f673135 | bellard | result (called @code{CC_DST}) and the type of operation (called |
335 | 1f673135 | bellard | @code{CC_OP}). |
336 | 1f673135 | bellard | |
337 | 1235fc06 | ths | @code{CC_OP} is almost never explicitly set in the generated code |
338 | 1f673135 | bellard | because it is known at translation time. |
339 | 1f673135 | bellard | |
340 | 1f673135 | bellard | In order to increase performances, a backward pass is performed on the |
341 | 1f673135 | bellard | generated simple instructions (see |
342 | 1f673135 | bellard | @code{target-i386/translate.c:optimize_flags()}). When it can be proved that |
343 | 1f673135 | bellard | the condition codes are not needed by the next instructions, no |
344 | 1f673135 | bellard | condition codes are computed at all. |
345 | 1f673135 | bellard | |
346 | debc7065 | bellard | @node CPU state optimisations |
347 | 1f673135 | bellard | @section CPU state optimisations |
348 | 1f673135 | bellard | |
349 | 1f673135 | bellard | The x86 CPU has many internal states which change the way it evaluates |
350 | 1f673135 | bellard | instructions. In order to achieve a good speed, the translation phase |
351 | 1f673135 | bellard | considers that some state information of the virtual x86 CPU cannot |
352 | 1f673135 | bellard | change in it. For example, if the SS, DS and ES segments have a zero |
353 | 1f673135 | bellard | base, then the translator does not even generate an addition for the |
354 | 1f673135 | bellard | segment base. |
355 | 1f673135 | bellard | |
356 | 1f673135 | bellard | [The FPU stack pointer register is not handled that way yet]. |
357 | 1f673135 | bellard | |
358 | debc7065 | bellard | @node Translation cache |
359 | 1f673135 | bellard | @section Translation cache |
360 | 1f673135 | bellard | |
361 | 15a34c63 | bellard | A 16 MByte cache holds the most recently used translations. For |
362 | 1f673135 | bellard | simplicity, it is completely flushed when it is full. A translation unit |
363 | 1f673135 | bellard | contains just a single basic block (a block of x86 instructions |
364 | 1f673135 | bellard | terminated by a jump or by a virtual CPU state change which the |
365 | 1f673135 | bellard | translator cannot deduce statically). |
366 | 1f673135 | bellard | |
367 | debc7065 | bellard | @node Direct block chaining |
368 | 1f673135 | bellard | @section Direct block chaining |
369 | 1f673135 | bellard | |
370 | 1f673135 | bellard | After each translated basic block is executed, QEMU uses the simulated |
371 | 1f673135 | bellard | Program Counter (PC) and other cpu state informations (such as the CS |
372 | 1f673135 | bellard | segment base value) to find the next basic block. |
373 | 1f673135 | bellard | |
374 | 1f673135 | bellard | In order to accelerate the most common cases where the new simulated PC |
375 | 1f673135 | bellard | is known, QEMU can patch a basic block so that it jumps directly to the |
376 | 1f673135 | bellard | next one. |
377 | 1f673135 | bellard | |
378 | 1f673135 | bellard | The most portable code uses an indirect jump. An indirect jump makes |
379 | 1f673135 | bellard | it easier to make the jump target modification atomic. On some host |
380 | 1f673135 | bellard | architectures (such as x86 or PowerPC), the @code{JUMP} opcode is |
381 | 1f673135 | bellard | directly patched so that the block chaining has no overhead. |
382 | 1f673135 | bellard | |
383 | debc7065 | bellard | @node Self-modifying code and translated code invalidation |
384 | 1f673135 | bellard | @section Self-modifying code and translated code invalidation |
385 | 1f673135 | bellard | |
386 | 1f673135 | bellard | Self-modifying code is a special challenge in x86 emulation because no |
387 | 1f673135 | bellard | instruction cache invalidation is signaled by the application when code |
388 | 1f673135 | bellard | is modified. |
389 | 1f673135 | bellard | |
390 | 1f673135 | bellard | When translated code is generated for a basic block, the corresponding |
391 | 1f673135 | bellard | host page is write protected if it is not already read-only (with the |
392 | 1f673135 | bellard | system call @code{mprotect()}). Then, if a write access is done to the |
393 | 1f673135 | bellard | page, Linux raises a SEGV signal. QEMU then invalidates all the |
394 | 1f673135 | bellard | translated code in the page and enables write accesses to the page. |
395 | 1f673135 | bellard | |
396 | 1f673135 | bellard | Correct translated code invalidation is done efficiently by maintaining |
397 | 1f673135 | bellard | a linked list of every translated block contained in a given page. Other |
398 | 5fafdf24 | ths | linked lists are also maintained to undo direct block chaining. |
399 | 1f673135 | bellard | |
400 | 1f673135 | bellard | Although the overhead of doing @code{mprotect()} calls is important, |
401 | 1f673135 | bellard | most MSDOS programs can be emulated at reasonnable speed with QEMU and |
402 | 1f673135 | bellard | DOSEMU. |
403 | 1f673135 | bellard | |
404 | 1f673135 | bellard | Note that QEMU also invalidates pages of translated code when it detects |
405 | 1f673135 | bellard | that memory mappings are modified with @code{mmap()} or @code{munmap()}. |
406 | 1f673135 | bellard | |
407 | 1f673135 | bellard | When using a software MMU, the code invalidation is more efficient: if |
408 | 1f673135 | bellard | a given code page is invalidated too often because of write accesses, |
409 | 1f673135 | bellard | then a bitmap representing all the code inside the page is |
410 | 1f673135 | bellard | built. Every store into that page checks the bitmap to see if the code |
411 | 1f673135 | bellard | really needs to be invalidated. It avoids invalidating the code when |
412 | 1f673135 | bellard | only data is modified in the page. |
413 | 1f673135 | bellard | |
414 | debc7065 | bellard | @node Exception support |
415 | 1f673135 | bellard | @section Exception support |
416 | 1f673135 | bellard | |
417 | 1f673135 | bellard | longjmp() is used when an exception such as division by zero is |
418 | 5fafdf24 | ths | encountered. |
419 | 1f673135 | bellard | |
420 | 1f673135 | bellard | The host SIGSEGV and SIGBUS signal handlers are used to get invalid |
421 | 1f673135 | bellard | memory accesses. The exact CPU state can be retrieved because all the |
422 | 1f673135 | bellard | x86 registers are stored in fixed host registers. The simulated program |
423 | 1f673135 | bellard | counter is found by retranslating the corresponding basic block and by |
424 | 1f673135 | bellard | looking where the host program counter was at the exception point. |
425 | 1f673135 | bellard | |
426 | 1f673135 | bellard | The virtual CPU cannot retrieve the exact @code{EFLAGS} register because |
427 | 1f673135 | bellard | in some cases it is not computed because of condition code |
428 | 1f673135 | bellard | optimisations. It is not a big concern because the emulated code can |
429 | 1f673135 | bellard | still be restarted in any cases. |
430 | 1f673135 | bellard | |
431 | debc7065 | bellard | @node MMU emulation |
432 | 1f673135 | bellard | @section MMU emulation |
433 | 1f673135 | bellard | |
434 | 1f673135 | bellard | For system emulation, QEMU uses the mmap() system call to emulate the |
435 | 1f673135 | bellard | target CPU MMU. It works as long the emulated OS does not use an area |
436 | 1f673135 | bellard | reserved by the host OS (such as the area above 0xc0000000 on x86 |
437 | 1f673135 | bellard | Linux). |
438 | 1f673135 | bellard | |
439 | 1f673135 | bellard | In order to be able to launch any OS, QEMU also supports a soft |
440 | 1f673135 | bellard | MMU. In that mode, the MMU virtual to physical address translation is |
441 | 1f673135 | bellard | done at every memory access. QEMU uses an address translation cache to |
442 | 1f673135 | bellard | speed up the translation. |
443 | 1f673135 | bellard | |
444 | 1f673135 | bellard | In order to avoid flushing the translated code each time the MMU |
445 | 1f673135 | bellard | mappings change, QEMU uses a physically indexed translation cache. It |
446 | 5fafdf24 | ths | means that each basic block is indexed with its physical address. |
447 | 1f673135 | bellard | |
448 | 1f673135 | bellard | When MMU mappings change, only the chaining of the basic blocks is |
449 | 1f673135 | bellard | reset (i.e. a basic block can no longer jump directly to another one). |
450 | 1f673135 | bellard | |
451 | debc7065 | bellard | @node Hardware interrupts |
452 | 1f673135 | bellard | @section Hardware interrupts |
453 | 1f673135 | bellard | |
454 | 1f673135 | bellard | In order to be faster, QEMU does not check at every basic block if an |
455 | 1f673135 | bellard | hardware interrupt is pending. Instead, the user must asynchrously |
456 | 1f673135 | bellard | call a specific function to tell that an interrupt is pending. This |
457 | 1f673135 | bellard | function resets the chaining of the currently executing basic |
458 | 1f673135 | bellard | block. It ensures that the execution will return soon in the main loop |
459 | 1f673135 | bellard | of the CPU emulator. Then the main loop can test if the interrupt is |
460 | 1f673135 | bellard | pending and handle it. |
461 | 1f673135 | bellard | |
462 | debc7065 | bellard | @node User emulation specific details |
463 | 1f673135 | bellard | @section User emulation specific details |
464 | 1f673135 | bellard | |
465 | 1f673135 | bellard | @subsection Linux system call translation |
466 | 1f673135 | bellard | |
467 | 1f673135 | bellard | QEMU includes a generic system call translator for Linux. It means that |
468 | 1f673135 | bellard | the parameters of the system calls can be converted to fix the |
469 | 1f673135 | bellard | endianness and 32/64 bit issues. The IOCTLs are converted with a generic |
470 | 1f673135 | bellard | type description system (see @file{ioctls.h} and @file{thunk.c}). |
471 | 1f673135 | bellard | |
472 | 1f673135 | bellard | QEMU supports host CPUs which have pages bigger than 4KB. It records all |
473 | 1f673135 | bellard | the mappings the process does and try to emulated the @code{mmap()} |
474 | 1f673135 | bellard | system calls in cases where the host @code{mmap()} call would fail |
475 | 1f673135 | bellard | because of bad page alignment. |
476 | 1f673135 | bellard | |
477 | 1f673135 | bellard | @subsection Linux signals |
478 | 1f673135 | bellard | |
479 | 1f673135 | bellard | Normal and real-time signals are queued along with their information |
480 | 1f673135 | bellard | (@code{siginfo_t}) as it is done in the Linux kernel. Then an interrupt |
481 | 1f673135 | bellard | request is done to the virtual CPU. When it is interrupted, one queued |
482 | 1f673135 | bellard | signal is handled by generating a stack frame in the virtual CPU as the |
483 | 1f673135 | bellard | Linux kernel does. The @code{sigreturn()} system call is emulated to return |
484 | 1f673135 | bellard | from the virtual signal handler. |
485 | 1f673135 | bellard | |
486 | 1f673135 | bellard | Some signals (such as SIGALRM) directly come from the host. Other |
487 | 1f673135 | bellard | signals are synthetized from the virtual CPU exceptions such as SIGFPE |
488 | 1f673135 | bellard | when a division by zero is done (see @code{main.c:cpu_loop()}). |
489 | 1f673135 | bellard | |
490 | 1f673135 | bellard | The blocked signal mask is still handled by the host Linux kernel so |
491 | 1f673135 | bellard | that most signal system calls can be redirected directly to the host |
492 | 1f673135 | bellard | Linux kernel. Only the @code{sigaction()} and @code{sigreturn()} system |
493 | 1f673135 | bellard | calls need to be fully emulated (see @file{signal.c}). |
494 | 1f673135 | bellard | |
495 | 1f673135 | bellard | @subsection clone() system call and threads |
496 | 1f673135 | bellard | |
497 | 1f673135 | bellard | The Linux clone() system call is usually used to create a thread. QEMU |
498 | 1f673135 | bellard | uses the host clone() system call so that real host threads are created |
499 | 1f673135 | bellard | for each emulated thread. One virtual CPU instance is created for each |
500 | 1f673135 | bellard | thread. |
501 | 1f673135 | bellard | |
502 | 1f673135 | bellard | The virtual x86 CPU atomic operations are emulated with a global lock so |
503 | 1f673135 | bellard | that their semantic is preserved. |
504 | 1f673135 | bellard | |
505 | 1f673135 | bellard | Note that currently there are still some locking issues in QEMU. In |
506 | 1f673135 | bellard | particular, the translated cache flush is not protected yet against |
507 | 1f673135 | bellard | reentrancy. |
508 | 1f673135 | bellard | |
509 | 1f673135 | bellard | @subsection Self-virtualization |
510 | 1f673135 | bellard | |
511 | 1f673135 | bellard | QEMU was conceived so that ultimately it can emulate itself. Although |
512 | 1f673135 | bellard | it is not very useful, it is an important test to show the power of the |
513 | 1f673135 | bellard | emulator. |
514 | 1f673135 | bellard | |
515 | 1f673135 | bellard | Achieving self-virtualization is not easy because there may be address |
516 | 1f673135 | bellard | space conflicts. QEMU solves this problem by being an executable ELF |
517 | 1f673135 | bellard | shared object as the ld-linux.so ELF interpreter. That way, it can be |
518 | 1f673135 | bellard | relocated at load time. |
519 | 1f673135 | bellard | |
520 | debc7065 | bellard | @node Bibliography |
521 | 1f673135 | bellard | @section Bibliography |
522 | 1f673135 | bellard | |
523 | 1f673135 | bellard | @table @asis |
524 | 1f673135 | bellard | |
525 | 5fafdf24 | ths | @item [1] |
526 | 1f673135 | bellard | @url{http://citeseer.nj.nec.com/piumarta98optimizing.html}, Optimizing |
527 | 1f673135 | bellard | direct threaded code by selective inlining (1998) by Ian Piumarta, Fabio |
528 | 1f673135 | bellard | Riccardi. |
529 | 1f673135 | bellard | |
530 | 1f673135 | bellard | @item [2] |
531 | 1f673135 | bellard | @url{http://developer.kde.org/~sewardj/}, Valgrind, an open-source |
532 | 1f673135 | bellard | memory debugger for x86-GNU/Linux, by Julian Seward. |
533 | 1f673135 | bellard | |
534 | 1f673135 | bellard | @item [3] |
535 | 1f673135 | bellard | @url{http://bochs.sourceforge.net/}, the Bochs IA-32 Emulator Project, |
536 | 1f673135 | bellard | by Kevin Lawton et al. |
537 | 1f673135 | bellard | |
538 | 1f673135 | bellard | @item [4] |
539 | 1f673135 | bellard | @url{http://www.cs.rose-hulman.edu/~donaldlf/em86/index.html}, the EM86 |
540 | 1f673135 | bellard | x86 emulator on Alpha-Linux. |
541 | 1f673135 | bellard | |
542 | 1f673135 | bellard | @item [5] |
543 | debc7065 | bellard | @url{http://www.usenix.org/publications/library/proceedings/usenix-nt97/@/full_papers/chernoff/chernoff.pdf}, |
544 | 1f673135 | bellard | DIGITAL FX!32: Running 32-Bit x86 Applications on Alpha NT, by Anton |
545 | 1f673135 | bellard | Chernoff and Ray Hookway. |
546 | 1f673135 | bellard | |
547 | 1f673135 | bellard | @item [6] |
548 | 1f673135 | bellard | @url{http://www.willows.com/}, Windows API library emulation from |
549 | 1f673135 | bellard | Willows Software. |
550 | 1f673135 | bellard | |
551 | 1f673135 | bellard | @item [7] |
552 | 5fafdf24 | ths | @url{http://user-mode-linux.sourceforge.net/}, |
553 | 1f673135 | bellard | The User-mode Linux Kernel. |
554 | 1f673135 | bellard | |
555 | 1f673135 | bellard | @item [8] |
556 | 5fafdf24 | ths | @url{http://www.plex86.org/}, |
557 | 1f673135 | bellard | The new Plex86 project. |
558 | 1f673135 | bellard | |
559 | 1f673135 | bellard | @item [9] |
560 | 5fafdf24 | ths | @url{http://www.vmware.com/}, |
561 | 1f673135 | bellard | The VMWare PC virtualizer. |
562 | 1f673135 | bellard | |
563 | 1f673135 | bellard | @item [10] |
564 | 5fafdf24 | ths | @url{http://www.microsoft.com/windowsxp/virtualpc/}, |
565 | 1f673135 | bellard | The VirtualPC PC virtualizer. |
566 | 1f673135 | bellard | |
567 | 1f673135 | bellard | @item [11] |
568 | 5fafdf24 | ths | @url{http://www.twoostwo.org/}, |
569 | 1f673135 | bellard | The TwoOStwo PC virtualizer. |
570 | 1f673135 | bellard | |
571 | 1f673135 | bellard | @end table |
572 | 1f673135 | bellard | |
573 | debc7065 | bellard | @node Regression Tests |
574 | 1f673135 | bellard | @chapter Regression Tests |
575 | 1f673135 | bellard | |
576 | 1f673135 | bellard | In the directory @file{tests/}, various interesting testing programs |
577 | b1f45238 | ths | are available. They are used for regression testing. |
578 | 1f673135 | bellard | |
579 | debc7065 | bellard | @menu |
580 | debc7065 | bellard | * test-i386:: |
581 | debc7065 | bellard | * linux-test:: |
582 | debc7065 | bellard | * qruncom.c:: |
583 | debc7065 | bellard | @end menu |
584 | debc7065 | bellard | |
585 | debc7065 | bellard | @node test-i386 |
586 | 1f673135 | bellard | @section @file{test-i386} |
587 | 1f673135 | bellard | |
588 | 1f673135 | bellard | This program executes most of the 16 bit and 32 bit x86 instructions and |
589 | 1f673135 | bellard | generates a text output. It can be compared with the output obtained with |
590 | 1f673135 | bellard | a real CPU or another emulator. The target @code{make test} runs this |
591 | 1f673135 | bellard | program and a @code{diff} on the generated output. |
592 | 1f673135 | bellard | |
593 | 1f673135 | bellard | The Linux system call @code{modify_ldt()} is used to create x86 selectors |
594 | 1f673135 | bellard | to test some 16 bit addressing and 32 bit with segmentation cases. |
595 | 1f673135 | bellard | |
596 | 1f673135 | bellard | The Linux system call @code{vm86()} is used to test vm86 emulation. |
597 | 1f673135 | bellard | |
598 | 1f673135 | bellard | Various exceptions are raised to test most of the x86 user space |
599 | 1f673135 | bellard | exception reporting. |
600 | 1f673135 | bellard | |
601 | debc7065 | bellard | @node linux-test |
602 | 1f673135 | bellard | @section @file{linux-test} |
603 | 1f673135 | bellard | |
604 | 1f673135 | bellard | This program tests various Linux system calls. It is used to verify |
605 | 1f673135 | bellard | that the system call parameters are correctly converted between target |
606 | 1f673135 | bellard | and host CPUs. |
607 | 1f673135 | bellard | |
608 | debc7065 | bellard | @node qruncom.c |
609 | 15a34c63 | bellard | @section @file{qruncom.c} |
610 | 1f673135 | bellard | |
611 | 15a34c63 | bellard | Example of usage of @code{libqemu} to emulate a user mode i386 CPU. |
612 | debc7065 | bellard | |
613 | debc7065 | bellard | @node Index |
614 | debc7065 | bellard | @chapter Index |
615 | debc7065 | bellard | @printindex cp |
616 | debc7065 | bellard | |
617 | debc7065 | bellard | @bye |