bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 1 | \input texinfo @c -*- texinfo -*- |
bellard | debc706 | 2006-04-30 21:58:41 +0000 | [diff] [blame] | 2 | @c %**start of header |
| 3 | @setfilename qemu-tech.info |
Stefan Weil | e080e78 | 2010-02-05 23:52:00 +0100 | [diff] [blame] | 4 | |
| 5 | @documentlanguage en |
| 6 | @documentencoding UTF-8 |
| 7 | |
bellard | debc706 | 2006-04-30 21:58:41 +0000 | [diff] [blame] | 8 | @settitle QEMU Internals |
| 9 | @exampleindent 0 |
| 10 | @paragraphindent 0 |
| 11 | @c %**end of header |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 12 | |
Stefan Weil | a1a32b0 | 2010-02-05 23:51:59 +0100 | [diff] [blame] | 13 | @ifinfo |
| 14 | @direntry |
| 15 | * QEMU Internals: (qemu-tech). The QEMU Emulator Internals. |
| 16 | @end direntry |
| 17 | @end ifinfo |
| 18 | |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 19 | @iftex |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 20 | @titlepage |
| 21 | @sp 7 |
| 22 | @center @titlefont{QEMU Internals} |
| 23 | @sp 3 |
| 24 | @end titlepage |
| 25 | @end iftex |
| 26 | |
bellard | debc706 | 2006-04-30 21:58:41 +0000 | [diff] [blame] | 27 | @ifnottex |
| 28 | @node Top |
| 29 | @top |
| 30 | |
| 31 | @menu |
| 32 | * Introduction:: |
| 33 | * QEMU Internals:: |
| 34 | * Regression Tests:: |
| 35 | * Index:: |
| 36 | @end menu |
| 37 | @end ifnottex |
| 38 | |
| 39 | @contents |
| 40 | |
| 41 | @node Introduction |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 42 | @chapter Introduction |
| 43 | |
bellard | debc706 | 2006-04-30 21:58:41 +0000 | [diff] [blame] | 44 | @menu |
Max Filippov | 3aeaea6 | 2011-10-10 14:48:23 +0400 | [diff] [blame] | 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 |
bellard | debc706 | 2006-04-30 21:58:41 +0000 | [diff] [blame] | 53 | @end menu |
| 54 | |
| 55 | @node intro_features |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 56 | @section Features |
| 57 | |
| 58 | QEMU is a FAST! processor emulator using a portable dynamic |
| 59 | translator. |
| 60 | |
| 61 | QEMU has two operating modes: |
| 62 | |
| 63 | @itemize @minus |
| 64 | |
ths | 5fafdf2 | 2007-09-16 21:08:06 +0000 | [diff] [blame] | 65 | @item |
blueswir1 | 998a050 | 2008-10-09 18:52:04 +0000 | [diff] [blame] | 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 |
| 70 | debug system code. |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 71 | |
ths | 5fafdf2 | 2007-09-16 21:08:06 +0000 | [diff] [blame] | 72 | @item |
blueswir1 | 998a050 | 2008-10-09 18:52:04 +0000 | [diff] [blame] | 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. |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 77 | @end itemize |
| 78 | |
| 79 | As QEMU requires no host kernel driver to run, it is very safe and |
| 80 | easy to use. |
| 81 | |
| 82 | QEMU generic features: |
| 83 | |
ths | 5fafdf2 | 2007-09-16 21:08:06 +0000 | [diff] [blame] | 84 | @itemize |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 85 | |
| 86 | @item User space only or full system emulation. |
| 87 | |
bellard | debc706 | 2006-04-30 21:58:41 +0000 | [diff] [blame] | 88 | @item Using dynamic translation to native code for reasonable speed. |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 89 | |
blueswir1 | 998a050 | 2008-10-09 18:52:04 +0000 | [diff] [blame] | 90 | @item |
| 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. |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 94 | |
| 95 | @item Self-modifying code support. |
| 96 | |
| 97 | @item Precise exceptions support. |
| 98 | |
blueswir1 | 998a050 | 2008-10-09 18:52:04 +0000 | [diff] [blame] | 99 | @item |
| 100 | Floating point library supporting both full software emulation and |
| 101 | native host FPU instructions. |
| 102 | |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 103 | @end itemize |
| 104 | |
| 105 | QEMU user mode emulation features: |
ths | 5fafdf2 | 2007-09-16 21:08:06 +0000 | [diff] [blame] | 106 | @itemize |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 107 | @item Generic Linux system call converter, including most ioctls. |
| 108 | |
| 109 | @item clone() emulation using native CPU clone() to use Linux scheduler for threads. |
| 110 | |
ths | 5fafdf2 | 2007-09-16 21:08:06 +0000 | [diff] [blame] | 111 | @item Accurate signal handling by remapping host signals to target signals. |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 112 | @end itemize |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 113 | |
blueswir1 | 998a050 | 2008-10-09 18:52:04 +0000 | [diff] [blame] | 114 | Linux user emulator (Linux host only) can be used to launch the Wine |
| 115 | Windows API emulator (@url{http://www.winehq.org}). A Darwin user |
| 116 | emulator (Darwin hosts only) exists and a BSD user emulator for BSD |
| 117 | hosts is under development. It would also be possible to develop a |
| 118 | similar user emulator for Solaris. |
| 119 | |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 120 | QEMU full system emulation features: |
ths | 5fafdf2 | 2007-09-16 21:08:06 +0000 | [diff] [blame] | 121 | @itemize |
blueswir1 | 998a050 | 2008-10-09 18:52:04 +0000 | [diff] [blame] | 122 | @item |
| 123 | QEMU uses a full software MMU for maximum portability. |
| 124 | |
| 125 | @item |
Anthony Liguori | 4a1418e | 2009-08-10 17:07:24 -0500 | [diff] [blame] | 126 | QEMU can optionally use an in-kernel accelerator, like kvm. The accelerators |
| 127 | execute some of the guest code natively, while |
blueswir1 | 998a050 | 2008-10-09 18:52:04 +0000 | [diff] [blame] | 128 | continuing to emulate the rest of the machine. |
| 129 | |
| 130 | @item |
| 131 | Various hardware devices can be emulated and in some cases, host |
| 132 | devices (e.g. serial and parallel ports, USB, drives) can be used |
| 133 | transparently by the guest Operating System. Host device passthrough |
| 134 | can be used for talking to external physical peripherals (e.g. a |
| 135 | webcam, modem or tape drive). |
| 136 | |
| 137 | @item |
| 138 | Symmetric multiprocessing (SMP) even on a host with a single CPU. On a |
| 139 | SMP host system, QEMU can use only one CPU fully due to difficulty in |
| 140 | implementing atomic memory accesses efficiently. |
| 141 | |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 142 | @end itemize |
| 143 | |
bellard | debc706 | 2006-04-30 21:58:41 +0000 | [diff] [blame] | 144 | @node intro_x86_emulation |
blueswir1 | 998a050 | 2008-10-09 18:52:04 +0000 | [diff] [blame] | 145 | @section x86 and x86-64 emulation |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 146 | |
| 147 | QEMU x86 target features: |
| 148 | |
ths | 5fafdf2 | 2007-09-16 21:08:06 +0000 | [diff] [blame] | 149 | @itemize |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 150 | |
ths | 5fafdf2 | 2007-09-16 21:08:06 +0000 | [diff] [blame] | 151 | @item The virtual x86 CPU supports 16 bit and 32 bit addressing with segmentation. |
blueswir1 | 998a050 | 2008-10-09 18:52:04 +0000 | [diff] [blame] | 152 | LDT/GDT and IDT are emulated. VM86 mode is also supported to run |
| 153 | DOSEMU. There is some support for MMX/3DNow!, SSE, SSE2, SSE3, SSSE3, |
| 154 | and SSE4 as well as x86-64 SVM. |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 155 | |
| 156 | @item Support of host page sizes bigger than 4KB in user mode emulation. |
| 157 | |
| 158 | @item QEMU can emulate itself on x86. |
| 159 | |
ths | 5fafdf2 | 2007-09-16 21:08:06 +0000 | [diff] [blame] | 160 | @item An extensive Linux x86 CPU test program is included @file{tests/test-i386}. |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 161 | It can be used to test other x86 virtual CPUs. |
| 162 | |
| 163 | @end itemize |
| 164 | |
| 165 | Current QEMU limitations: |
| 166 | |
ths | 5fafdf2 | 2007-09-16 21:08:06 +0000 | [diff] [blame] | 167 | @itemize |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 168 | |
blueswir1 | 998a050 | 2008-10-09 18:52:04 +0000 | [diff] [blame] | 169 | @item Limited x86-64 support. |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 170 | |
| 171 | @item IPC syscalls are missing. |
| 172 | |
ths | 5fafdf2 | 2007-09-16 21:08:06 +0000 | [diff] [blame] | 173 | @item The x86 segment limits and access rights are not tested at every |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 174 | memory access (yet). Hopefully, very few OSes seem to rely on that for |
| 175 | normal use. |
| 176 | |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 177 | @end itemize |
| 178 | |
bellard | debc706 | 2006-04-30 21:58:41 +0000 | [diff] [blame] | 179 | @node intro_arm_emulation |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 180 | @section ARM emulation |
| 181 | |
| 182 | @itemize |
| 183 | |
| 184 | @item Full ARM 7 user emulation. |
| 185 | |
| 186 | @item NWFPE FPU support included in user Linux emulation. |
| 187 | |
| 188 | @item Can run most ARM Linux binaries. |
| 189 | |
| 190 | @end itemize |
| 191 | |
ths | 24d4de4 | 2007-07-11 10:24:28 +0000 | [diff] [blame] | 192 | @node intro_mips_emulation |
| 193 | @section MIPS emulation |
| 194 | |
| 195 | @itemize |
| 196 | |
| 197 | @item The system emulation allows full MIPS32/MIPS64 Release 2 emulation, |
| 198 | including privileged instructions, FPU and MMU, in both little and big |
| 199 | endian modes. |
| 200 | |
| 201 | @item The Linux userland emulation can run many 32 bit MIPS Linux binaries. |
| 202 | |
| 203 | @end itemize |
| 204 | |
| 205 | Current QEMU limitations: |
| 206 | |
| 207 | @itemize |
| 208 | |
| 209 | @item Self-modifying code is not always handled correctly. |
| 210 | |
| 211 | @item 64 bit userland emulation is not implemented. |
| 212 | |
| 213 | @item The system emulation is not complete enough to run real firmware. |
| 214 | |
ths | b1f4523 | 2007-07-12 09:03:30 +0000 | [diff] [blame] | 215 | @item The watchpoint debug facility is not implemented. |
| 216 | |
ths | 24d4de4 | 2007-07-11 10:24:28 +0000 | [diff] [blame] | 217 | @end itemize |
| 218 | |
bellard | debc706 | 2006-04-30 21:58:41 +0000 | [diff] [blame] | 219 | @node intro_ppc_emulation |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 220 | @section PowerPC emulation |
| 221 | |
| 222 | @itemize |
| 223 | |
ths | 5fafdf2 | 2007-09-16 21:08:06 +0000 | [diff] [blame] | 224 | @item Full PowerPC 32 bit emulation, including privileged instructions, |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 225 | FPU and MMU. |
| 226 | |
| 227 | @item Can run most PowerPC Linux binaries. |
| 228 | |
| 229 | @end itemize |
| 230 | |
bellard | debc706 | 2006-04-30 21:58:41 +0000 | [diff] [blame] | 231 | @node intro_sparc_emulation |
blueswir1 | 998a050 | 2008-10-09 18:52:04 +0000 | [diff] [blame] | 232 | @section Sparc32 and Sparc64 emulation |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 233 | |
| 234 | @itemize |
| 235 | |
blueswir1 | f6b647c | 2007-04-05 18:40:23 +0000 | [diff] [blame] | 236 | @item Full SPARC V8 emulation, including privileged |
bellard | 3475187 | 2005-07-02 14:31:34 +0000 | [diff] [blame] | 237 | instructions, FPU and MMU. SPARC V9 emulation includes most privileged |
blueswir1 | a785e42 | 2007-10-20 08:09:05 +0000 | [diff] [blame] | 238 | and VIS instructions, FPU and I/D MMU. Alignment is fully enforced. |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 239 | |
blueswir1 | a785e42 | 2007-10-20 08:09:05 +0000 | [diff] [blame] | 240 | @item Can run most 32-bit SPARC Linux binaries, SPARC32PLUS Linux binaries and |
| 241 | some 64-bit SPARC Linux binaries. |
bellard | 3475187 | 2005-07-02 14:31:34 +0000 | [diff] [blame] | 242 | |
| 243 | @end itemize |
| 244 | |
| 245 | Current QEMU limitations: |
| 246 | |
ths | 5fafdf2 | 2007-09-16 21:08:06 +0000 | [diff] [blame] | 247 | @itemize |
bellard | 3475187 | 2005-07-02 14:31:34 +0000 | [diff] [blame] | 248 | |
bellard | 3475187 | 2005-07-02 14:31:34 +0000 | [diff] [blame] | 249 | @item IPC syscalls are missing. |
| 250 | |
blueswir1 | 1f58732 | 2007-11-25 18:40:20 +0000 | [diff] [blame] | 251 | @item Floating point exception support is buggy. |
bellard | 3475187 | 2005-07-02 14:31:34 +0000 | [diff] [blame] | 252 | |
| 253 | @item Atomic instructions are not correctly implemented. |
| 254 | |
blueswir1 | 998a050 | 2008-10-09 18:52:04 +0000 | [diff] [blame] | 255 | @item There are still some problems with Sparc64 emulators. |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 256 | |
| 257 | @end itemize |
| 258 | |
Max Filippov | 3aeaea6 | 2011-10-10 14:48:23 +0400 | [diff] [blame] | 259 | @node intro_xtensa_emulation |
| 260 | @section Xtensa emulation |
| 261 | |
| 262 | @itemize |
| 263 | |
| 264 | @item Core Xtensa ISA emulation, including most options: code density, |
| 265 | loop, extended L32R, 16- and 32-bit multiplication, 32-bit division, |
| 266 | MAC16, miscellaneous operations, boolean, multiprocessor synchronization, |
| 267 | conditional store, exceptions, relocatable vectors, unaligned exception, |
| 268 | interrupts (including high priority and timer), hardware alignment, |
| 269 | region protection, region translation, MMU, windowed registers, thread |
| 270 | pointer, processor ID. |
| 271 | |
| 272 | @item Not implemented options: FP coprocessor, coprocessor context, |
| 273 | data/instruction cache (including cache prefetch and locking), XLMI, |
| 274 | processor interface, debug. Also options not covered by the core ISA |
| 275 | (e.g. FLIX, wide branches) are not implemented. |
| 276 | |
| 277 | @item Can run most Xtensa Linux binaries. |
| 278 | |
| 279 | @item New core configuration that requires no additional instructions |
| 280 | may be created from overlay with minimal amount of hand-written code. |
| 281 | |
| 282 | @end itemize |
| 283 | |
blueswir1 | 998a050 | 2008-10-09 18:52:04 +0000 | [diff] [blame] | 284 | @node intro_other_emulation |
| 285 | @section Other CPU emulation |
| 286 | |
| 287 | In addition to the above, QEMU supports emulation of other CPUs with |
| 288 | varying levels of success. These are: |
| 289 | |
| 290 | @itemize |
| 291 | |
| 292 | @item |
| 293 | Alpha |
| 294 | @item |
| 295 | CRIS |
| 296 | @item |
| 297 | M68k |
| 298 | @item |
| 299 | SH4 |
| 300 | @end itemize |
| 301 | |
bellard | debc706 | 2006-04-30 21:58:41 +0000 | [diff] [blame] | 302 | @node QEMU Internals |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 303 | @chapter QEMU Internals |
| 304 | |
bellard | debc706 | 2006-04-30 21:58:41 +0000 | [diff] [blame] | 305 | @menu |
| 306 | * QEMU compared to other emulators:: |
| 307 | * Portable dynamic translation:: |
bellard | debc706 | 2006-04-30 21:58:41 +0000 | [diff] [blame] | 308 | * Condition code optimisations:: |
| 309 | * CPU state optimisations:: |
| 310 | * Translation cache:: |
| 311 | * Direct block chaining:: |
| 312 | * Self-modifying code and translated code invalidation:: |
| 313 | * Exception support:: |
| 314 | * MMU emulation:: |
blueswir1 | 998a050 | 2008-10-09 18:52:04 +0000 | [diff] [blame] | 315 | * Device emulation:: |
bellard | debc706 | 2006-04-30 21:58:41 +0000 | [diff] [blame] | 316 | * Hardware interrupts:: |
| 317 | * User emulation specific details:: |
| 318 | * Bibliography:: |
| 319 | @end menu |
| 320 | |
| 321 | @node QEMU compared to other emulators |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 322 | @section QEMU compared to other emulators |
| 323 | |
| 324 | Like bochs [3], QEMU emulates an x86 CPU. But QEMU is much faster than |
| 325 | bochs as it uses dynamic compilation. Bochs is closely tied to x86 PC |
| 326 | emulation while QEMU can emulate several processors. |
| 327 | |
| 328 | Like Valgrind [2], QEMU does user space emulation and dynamic |
| 329 | translation. Valgrind is mainly a memory debugger while QEMU has no |
| 330 | support for it (QEMU could be used to detect out of bound memory |
| 331 | accesses as Valgrind, but it has no support to track uninitialised data |
| 332 | as Valgrind does). The Valgrind dynamic translator generates better code |
| 333 | than QEMU (in particular it does register allocation) but it is closely |
| 334 | tied to an x86 host and target and has no support for precise exceptions |
| 335 | and system emulation. |
| 336 | |
| 337 | EM86 [4] is the closest project to user space QEMU (and QEMU still uses |
| 338 | some of its code, in particular the ELF file loader). EM86 was limited |
| 339 | to an alpha host and used a proprietary and slow interpreter (the |
| 340 | interpreter part of the FX!32 Digital Win32 code translator [5]). |
| 341 | |
| 342 | TWIN [6] is a Windows API emulator like Wine. It is less accurate than |
| 343 | Wine but includes a protected mode x86 interpreter to launch x86 Windows |
bellard | 36d54d1 | 2004-09-05 16:04:16 +0000 | [diff] [blame] | 344 | executables. Such an approach has greater potential because most of the |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 345 | Windows API is executed natively but it is far more difficult to develop |
| 346 | because all the data structures and function parameters exchanged |
| 347 | between the API and the x86 code must be converted. |
| 348 | |
| 349 | User mode Linux [7] was the only solution before QEMU to launch a |
| 350 | Linux kernel as a process while not needing any host kernel |
| 351 | patches. However, user mode Linux requires heavy kernel patches while |
| 352 | QEMU accepts unpatched Linux kernels. The price to pay is that QEMU is |
| 353 | slower. |
| 354 | |
blueswir1 | 998a050 | 2008-10-09 18:52:04 +0000 | [diff] [blame] | 355 | The Plex86 [8] PC virtualizer is done in the same spirit as the now |
| 356 | obsolete qemu-fast system emulator. It requires a patched Linux kernel |
| 357 | to work (you cannot launch the same kernel on your PC), but the |
| 358 | patches are really small. As it is a PC virtualizer (no emulation is |
| 359 | done except for some privileged instructions), it has the potential of |
| 360 | being faster than QEMU. The downside is that a complicated (and |
| 361 | potentially unsafe) host kernel patch is needed. |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 362 | |
| 363 | The commercial PC Virtualizers (VMWare [9], VirtualPC [10], TwoOStwo |
| 364 | [11]) are faster than QEMU, but they all need specific, proprietary |
| 365 | and potentially unsafe host drivers. Moreover, they are unable to |
| 366 | provide cycle exact simulation as an emulator can. |
| 367 | |
blueswir1 | 998a050 | 2008-10-09 18:52:04 +0000 | [diff] [blame] | 368 | VirtualBox [12], Xen [13] and KVM [14] are based on QEMU. QEMU-SystemC |
| 369 | [15] uses QEMU to simulate a system where some hardware devices are |
| 370 | developed in SystemC. |
| 371 | |
bellard | debc706 | 2006-04-30 21:58:41 +0000 | [diff] [blame] | 372 | @node Portable dynamic translation |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 373 | @section Portable dynamic translation |
| 374 | |
| 375 | QEMU is a dynamic translator. When it first encounters a piece of code, |
| 376 | it converts it to the host instruction set. Usually dynamic translators |
| 377 | are very complicated and highly CPU dependent. QEMU uses some tricks |
| 378 | which make it relatively easily portable and simple while achieving good |
| 379 | performances. |
| 380 | |
blueswir1 | 998a050 | 2008-10-09 18:52:04 +0000 | [diff] [blame] | 381 | After the release of version 0.9.1, QEMU switched to a new method of |
| 382 | generating code, Tiny Code Generator or TCG. TCG relaxes the |
| 383 | dependency on the exact version of the compiler used. The basic idea |
| 384 | is to split every target instruction into a couple of RISC-like TCG |
| 385 | ops (see @code{target-i386/translate.c}). Some optimizations can be |
| 386 | performed at this stage, including liveness analysis and trivial |
| 387 | constant expression evaluation. TCG ops are then implemented in the |
| 388 | host CPU back end, also known as TCG target (see |
| 389 | @code{tcg/i386/tcg-target.c}). For more information, please take a |
| 390 | look at @code{tcg/README}. |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 391 | |
bellard | debc706 | 2006-04-30 21:58:41 +0000 | [diff] [blame] | 392 | @node Condition code optimisations |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 393 | @section Condition code optimisations |
| 394 | |
blueswir1 | 998a050 | 2008-10-09 18:52:04 +0000 | [diff] [blame] | 395 | Lazy evaluation of CPU condition codes (@code{EFLAGS} register on x86) |
| 396 | is important for CPUs where every instruction sets the condition |
| 397 | codes. It tends to be less important on conventional RISC systems |
Blue Swirl | f0f26a0 | 2009-05-10 18:23:46 +0000 | [diff] [blame] | 398 | where condition codes are only updated when explicitly requested. On |
| 399 | Sparc64, costly update of both 32 and 64 bit condition codes can be |
| 400 | avoided with lazy evaluation. |
blueswir1 | 998a050 | 2008-10-09 18:52:04 +0000 | [diff] [blame] | 401 | |
| 402 | Instead of computing the condition codes after each x86 instruction, |
| 403 | QEMU just stores one operand (called @code{CC_SRC}), the result |
| 404 | (called @code{CC_DST}) and the type of operation (called |
| 405 | @code{CC_OP}). When the condition codes are needed, the condition |
| 406 | codes can be calculated using this information. In addition, an |
| 407 | optimized calculation can be performed for some instruction types like |
| 408 | conditional branches. |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 409 | |
ths | 1235fc0 | 2008-06-03 19:51:57 +0000 | [diff] [blame] | 410 | @code{CC_OP} is almost never explicitly set in the generated code |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 411 | because it is known at translation time. |
| 412 | |
Blue Swirl | f0f26a0 | 2009-05-10 18:23:46 +0000 | [diff] [blame] | 413 | The lazy condition code evaluation is used on x86, m68k, cris and |
| 414 | Sparc. ARM uses a simplified variant for the N and Z flags. |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 415 | |
bellard | debc706 | 2006-04-30 21:58:41 +0000 | [diff] [blame] | 416 | @node CPU state optimisations |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 417 | @section CPU state optimisations |
| 418 | |
blueswir1 | 998a050 | 2008-10-09 18:52:04 +0000 | [diff] [blame] | 419 | The target CPUs have many internal states which change the way it |
| 420 | evaluates instructions. In order to achieve a good speed, the |
| 421 | translation phase considers that some state information of the virtual |
| 422 | CPU cannot change in it. The state is recorded in the Translation |
| 423 | Block (TB). If the state changes (e.g. privilege level), a new TB will |
| 424 | be generated and the previous TB won't be used anymore until the state |
| 425 | matches the state recorded in the previous TB. For example, if the SS, |
| 426 | DS and ES segments have a zero base, then the translator does not even |
| 427 | generate an addition for the segment base. |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 428 | |
| 429 | [The FPU stack pointer register is not handled that way yet]. |
| 430 | |
bellard | debc706 | 2006-04-30 21:58:41 +0000 | [diff] [blame] | 431 | @node Translation cache |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 432 | @section Translation cache |
| 433 | |
陳韋任 | 27c8efc | 2011-11-05 01:14:44 +0800 | [diff] [blame] | 434 | A 32 MByte cache holds the most recently used translations. For |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 435 | simplicity, it is completely flushed when it is full. A translation unit |
| 436 | contains just a single basic block (a block of x86 instructions |
| 437 | terminated by a jump or by a virtual CPU state change which the |
| 438 | translator cannot deduce statically). |
| 439 | |
bellard | debc706 | 2006-04-30 21:58:41 +0000 | [diff] [blame] | 440 | @node Direct block chaining |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 441 | @section Direct block chaining |
| 442 | |
| 443 | After each translated basic block is executed, QEMU uses the simulated |
| 444 | Program Counter (PC) and other cpu state informations (such as the CS |
| 445 | segment base value) to find the next basic block. |
| 446 | |
| 447 | In order to accelerate the most common cases where the new simulated PC |
| 448 | is known, QEMU can patch a basic block so that it jumps directly to the |
| 449 | next one. |
| 450 | |
| 451 | The most portable code uses an indirect jump. An indirect jump makes |
| 452 | it easier to make the jump target modification atomic. On some host |
| 453 | architectures (such as x86 or PowerPC), the @code{JUMP} opcode is |
| 454 | directly patched so that the block chaining has no overhead. |
| 455 | |
bellard | debc706 | 2006-04-30 21:58:41 +0000 | [diff] [blame] | 456 | @node Self-modifying code and translated code invalidation |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 457 | @section Self-modifying code and translated code invalidation |
| 458 | |
| 459 | Self-modifying code is a special challenge in x86 emulation because no |
| 460 | instruction cache invalidation is signaled by the application when code |
| 461 | is modified. |
| 462 | |
| 463 | When translated code is generated for a basic block, the corresponding |
blueswir1 | 998a050 | 2008-10-09 18:52:04 +0000 | [diff] [blame] | 464 | host page is write protected if it is not already read-only. Then, if |
| 465 | a write access is done to the page, Linux raises a SEGV signal. QEMU |
| 466 | then invalidates all the translated code in the page and enables write |
| 467 | accesses to the page. |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 468 | |
| 469 | Correct translated code invalidation is done efficiently by maintaining |
| 470 | a linked list of every translated block contained in a given page. Other |
ths | 5fafdf2 | 2007-09-16 21:08:06 +0000 | [diff] [blame] | 471 | linked lists are also maintained to undo direct block chaining. |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 472 | |
blueswir1 | 998a050 | 2008-10-09 18:52:04 +0000 | [diff] [blame] | 473 | On RISC targets, correctly written software uses memory barriers and |
| 474 | cache flushes, so some of the protection above would not be |
| 475 | necessary. However, QEMU still requires that the generated code always |
| 476 | matches the target instructions in memory in order to handle |
| 477 | exceptions correctly. |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 478 | |
bellard | debc706 | 2006-04-30 21:58:41 +0000 | [diff] [blame] | 479 | @node Exception support |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 480 | @section Exception support |
| 481 | |
| 482 | longjmp() is used when an exception such as division by zero is |
ths | 5fafdf2 | 2007-09-16 21:08:06 +0000 | [diff] [blame] | 483 | encountered. |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 484 | |
| 485 | The host SIGSEGV and SIGBUS signal handlers are used to get invalid |
blueswir1 | 998a050 | 2008-10-09 18:52:04 +0000 | [diff] [blame] | 486 | memory accesses. The simulated program counter is found by |
| 487 | retranslating the corresponding basic block and by looking where the |
| 488 | host program counter was at the exception point. |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 489 | |
| 490 | The virtual CPU cannot retrieve the exact @code{EFLAGS} register because |
| 491 | in some cases it is not computed because of condition code |
| 492 | optimisations. It is not a big concern because the emulated code can |
| 493 | still be restarted in any cases. |
| 494 | |
bellard | debc706 | 2006-04-30 21:58:41 +0000 | [diff] [blame] | 495 | @node MMU emulation |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 496 | @section MMU emulation |
| 497 | |
blueswir1 | 998a050 | 2008-10-09 18:52:04 +0000 | [diff] [blame] | 498 | For system emulation QEMU supports a soft MMU. In that mode, the MMU |
| 499 | virtual to physical address translation is done at every memory |
| 500 | access. QEMU uses an address translation cache to speed up the |
| 501 | translation. |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 502 | |
| 503 | In order to avoid flushing the translated code each time the MMU |
| 504 | mappings change, QEMU uses a physically indexed translation cache. It |
ths | 5fafdf2 | 2007-09-16 21:08:06 +0000 | [diff] [blame] | 505 | means that each basic block is indexed with its physical address. |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 506 | |
| 507 | When MMU mappings change, only the chaining of the basic blocks is |
| 508 | reset (i.e. a basic block can no longer jump directly to another one). |
| 509 | |
blueswir1 | 998a050 | 2008-10-09 18:52:04 +0000 | [diff] [blame] | 510 | @node Device emulation |
| 511 | @section Device emulation |
| 512 | |
| 513 | Systems emulated by QEMU are organized by boards. At initialization |
| 514 | phase, each board instantiates a number of CPUs, devices, RAM and |
| 515 | ROM. Each device in turn can assign I/O ports or memory areas (for |
| 516 | MMIO) to its handlers. When the emulation starts, an access to the |
| 517 | ports or MMIO memory areas assigned to the device causes the |
| 518 | corresponding handler to be called. |
| 519 | |
| 520 | RAM and ROM are handled more optimally, only the offset to the host |
| 521 | memory needs to be added to the guest address. |
| 522 | |
| 523 | The video RAM of VGA and other display cards is special: it can be |
| 524 | read or written directly like RAM, but write accesses cause the memory |
| 525 | to be marked with VGA_DIRTY flag as well. |
| 526 | |
| 527 | QEMU supports some device classes like serial and parallel ports, USB, |
| 528 | drives and network devices, by providing APIs for easier connection to |
| 529 | the generic, higher level implementations. The API hides the |
| 530 | implementation details from the devices, like native device use or |
| 531 | advanced block device formats like QCOW. |
| 532 | |
| 533 | Usually the devices implement a reset method and register support for |
| 534 | saving and loading of the device state. The devices can also use |
| 535 | timers, especially together with the use of bottom halves (BHs). |
| 536 | |
bellard | debc706 | 2006-04-30 21:58:41 +0000 | [diff] [blame] | 537 | @node Hardware interrupts |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 538 | @section Hardware interrupts |
| 539 | |
| 540 | In order to be faster, QEMU does not check at every basic block if an |
Stefan Weil | e8dc093 | 2011-01-07 21:31:39 +0100 | [diff] [blame] | 541 | hardware interrupt is pending. Instead, the user must asynchronously |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 542 | call a specific function to tell that an interrupt is pending. This |
| 543 | function resets the chaining of the currently executing basic |
| 544 | block. It ensures that the execution will return soon in the main loop |
| 545 | of the CPU emulator. Then the main loop can test if the interrupt is |
| 546 | pending and handle it. |
| 547 | |
bellard | debc706 | 2006-04-30 21:58:41 +0000 | [diff] [blame] | 548 | @node User emulation specific details |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 549 | @section User emulation specific details |
| 550 | |
| 551 | @subsection Linux system call translation |
| 552 | |
| 553 | QEMU includes a generic system call translator for Linux. It means that |
| 554 | the parameters of the system calls can be converted to fix the |
| 555 | endianness and 32/64 bit issues. The IOCTLs are converted with a generic |
| 556 | type description system (see @file{ioctls.h} and @file{thunk.c}). |
| 557 | |
| 558 | QEMU supports host CPUs which have pages bigger than 4KB. It records all |
| 559 | the mappings the process does and try to emulated the @code{mmap()} |
| 560 | system calls in cases where the host @code{mmap()} call would fail |
| 561 | because of bad page alignment. |
| 562 | |
| 563 | @subsection Linux signals |
| 564 | |
| 565 | Normal and real-time signals are queued along with their information |
| 566 | (@code{siginfo_t}) as it is done in the Linux kernel. Then an interrupt |
| 567 | request is done to the virtual CPU. When it is interrupted, one queued |
| 568 | signal is handled by generating a stack frame in the virtual CPU as the |
| 569 | Linux kernel does. The @code{sigreturn()} system call is emulated to return |
| 570 | from the virtual signal handler. |
| 571 | |
| 572 | Some signals (such as SIGALRM) directly come from the host. Other |
Stefan Weil | e8dc093 | 2011-01-07 21:31:39 +0100 | [diff] [blame] | 573 | signals are synthesized from the virtual CPU exceptions such as SIGFPE |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 574 | when a division by zero is done (see @code{main.c:cpu_loop()}). |
| 575 | |
| 576 | The blocked signal mask is still handled by the host Linux kernel so |
| 577 | that most signal system calls can be redirected directly to the host |
| 578 | Linux kernel. Only the @code{sigaction()} and @code{sigreturn()} system |
| 579 | calls need to be fully emulated (see @file{signal.c}). |
| 580 | |
| 581 | @subsection clone() system call and threads |
| 582 | |
| 583 | The Linux clone() system call is usually used to create a thread. QEMU |
| 584 | uses the host clone() system call so that real host threads are created |
| 585 | for each emulated thread. One virtual CPU instance is created for each |
| 586 | thread. |
| 587 | |
| 588 | The virtual x86 CPU atomic operations are emulated with a global lock so |
| 589 | that their semantic is preserved. |
| 590 | |
| 591 | Note that currently there are still some locking issues in QEMU. In |
| 592 | particular, the translated cache flush is not protected yet against |
| 593 | reentrancy. |
| 594 | |
| 595 | @subsection Self-virtualization |
| 596 | |
| 597 | QEMU was conceived so that ultimately it can emulate itself. Although |
| 598 | it is not very useful, it is an important test to show the power of the |
| 599 | emulator. |
| 600 | |
| 601 | Achieving self-virtualization is not easy because there may be address |
blueswir1 | 998a050 | 2008-10-09 18:52:04 +0000 | [diff] [blame] | 602 | space conflicts. QEMU user emulators solve this problem by being an |
| 603 | executable ELF shared object as the ld-linux.so ELF interpreter. That |
| 604 | way, it can be relocated at load time. |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 605 | |
bellard | debc706 | 2006-04-30 21:58:41 +0000 | [diff] [blame] | 606 | @node Bibliography |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 607 | @section Bibliography |
| 608 | |
| 609 | @table @asis |
| 610 | |
ths | 5fafdf2 | 2007-09-16 21:08:06 +0000 | [diff] [blame] | 611 | @item [1] |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 612 | @url{http://citeseer.nj.nec.com/piumarta98optimizing.html}, Optimizing |
| 613 | direct threaded code by selective inlining (1998) by Ian Piumarta, Fabio |
| 614 | Riccardi. |
| 615 | |
| 616 | @item [2] |
| 617 | @url{http://developer.kde.org/~sewardj/}, Valgrind, an open-source |
| 618 | memory debugger for x86-GNU/Linux, by Julian Seward. |
| 619 | |
| 620 | @item [3] |
| 621 | @url{http://bochs.sourceforge.net/}, the Bochs IA-32 Emulator Project, |
| 622 | by Kevin Lawton et al. |
| 623 | |
| 624 | @item [4] |
| 625 | @url{http://www.cs.rose-hulman.edu/~donaldlf/em86/index.html}, the EM86 |
| 626 | x86 emulator on Alpha-Linux. |
| 627 | |
| 628 | @item [5] |
bellard | debc706 | 2006-04-30 21:58:41 +0000 | [diff] [blame] | 629 | @url{http://www.usenix.org/publications/library/proceedings/usenix-nt97/@/full_papers/chernoff/chernoff.pdf}, |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 630 | DIGITAL FX!32: Running 32-Bit x86 Applications on Alpha NT, by Anton |
| 631 | Chernoff and Ray Hookway. |
| 632 | |
| 633 | @item [6] |
| 634 | @url{http://www.willows.com/}, Windows API library emulation from |
| 635 | Willows Software. |
| 636 | |
| 637 | @item [7] |
ths | 5fafdf2 | 2007-09-16 21:08:06 +0000 | [diff] [blame] | 638 | @url{http://user-mode-linux.sourceforge.net/}, |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 639 | The User-mode Linux Kernel. |
| 640 | |
| 641 | @item [8] |
ths | 5fafdf2 | 2007-09-16 21:08:06 +0000 | [diff] [blame] | 642 | @url{http://www.plex86.org/}, |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 643 | The new Plex86 project. |
| 644 | |
| 645 | @item [9] |
ths | 5fafdf2 | 2007-09-16 21:08:06 +0000 | [diff] [blame] | 646 | @url{http://www.vmware.com/}, |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 647 | The VMWare PC virtualizer. |
| 648 | |
| 649 | @item [10] |
ths | 5fafdf2 | 2007-09-16 21:08:06 +0000 | [diff] [blame] | 650 | @url{http://www.microsoft.com/windowsxp/virtualpc/}, |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 651 | The VirtualPC PC virtualizer. |
| 652 | |
| 653 | @item [11] |
ths | 5fafdf2 | 2007-09-16 21:08:06 +0000 | [diff] [blame] | 654 | @url{http://www.twoostwo.org/}, |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 655 | The TwoOStwo PC virtualizer. |
| 656 | |
blueswir1 | 998a050 | 2008-10-09 18:52:04 +0000 | [diff] [blame] | 657 | @item [12] |
| 658 | @url{http://virtualbox.org/}, |
| 659 | The VirtualBox PC virtualizer. |
| 660 | |
| 661 | @item [13] |
| 662 | @url{http://www.xen.org/}, |
| 663 | The Xen hypervisor. |
| 664 | |
| 665 | @item [14] |
| 666 | @url{http://kvm.qumranet.com/kvmwiki/Front_Page}, |
| 667 | Kernel Based Virtual Machine (KVM). |
| 668 | |
| 669 | @item [15] |
| 670 | @url{http://www.greensocs.com/projects/QEMUSystemC}, |
| 671 | QEMU-SystemC, a hardware co-simulator. |
| 672 | |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 673 | @end table |
| 674 | |
bellard | debc706 | 2006-04-30 21:58:41 +0000 | [diff] [blame] | 675 | @node Regression Tests |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 676 | @chapter Regression Tests |
| 677 | |
| 678 | In the directory @file{tests/}, various interesting testing programs |
ths | b1f4523 | 2007-07-12 09:03:30 +0000 | [diff] [blame] | 679 | are available. They are used for regression testing. |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 680 | |
bellard | debc706 | 2006-04-30 21:58:41 +0000 | [diff] [blame] | 681 | @menu |
| 682 | * test-i386:: |
| 683 | * linux-test:: |
bellard | debc706 | 2006-04-30 21:58:41 +0000 | [diff] [blame] | 684 | @end menu |
| 685 | |
| 686 | @node test-i386 |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 687 | @section @file{test-i386} |
| 688 | |
| 689 | This program executes most of the 16 bit and 32 bit x86 instructions and |
| 690 | generates a text output. It can be compared with the output obtained with |
| 691 | a real CPU or another emulator. The target @code{make test} runs this |
| 692 | program and a @code{diff} on the generated output. |
| 693 | |
| 694 | The Linux system call @code{modify_ldt()} is used to create x86 selectors |
| 695 | to test some 16 bit addressing and 32 bit with segmentation cases. |
| 696 | |
| 697 | The Linux system call @code{vm86()} is used to test vm86 emulation. |
| 698 | |
| 699 | Various exceptions are raised to test most of the x86 user space |
| 700 | exception reporting. |
| 701 | |
bellard | debc706 | 2006-04-30 21:58:41 +0000 | [diff] [blame] | 702 | @node linux-test |
bellard | 1f67313 | 2004-04-04 15:21:17 +0000 | [diff] [blame] | 703 | @section @file{linux-test} |
| 704 | |
| 705 | This program tests various Linux system calls. It is used to verify |
| 706 | that the system call parameters are correctly converted between target |
| 707 | and host CPUs. |
| 708 | |
bellard | debc706 | 2006-04-30 21:58:41 +0000 | [diff] [blame] | 709 | @node Index |
| 710 | @chapter Index |
| 711 | @printindex cp |
| 712 | |
| 713 | @bye |