view tools/ioemu/qemu-tech.texi @ 0:7d21f7218375

Exact replica of unstable on 051908 + README-this
author Mukesh Rathor
date Mon May 19 15:34:57 2008 -0700 (2008-05-19)
line source
1 \input texinfo @c -*- texinfo -*-
2 @c %**start of header
3 @setfilename
4 @settitle QEMU Internals
5 @exampleindent 0
6 @paragraphindent 0
7 @c %**end of header
9 @iftex
10 @titlepage
11 @sp 7
12 @center @titlefont{QEMU Internals}
13 @sp 3
14 @end titlepage
15 @end iftex
17 @ifnottex
18 @node Top
19 @top
21 @menu
22 * Introduction::
23 * QEMU Internals::
24 * Regression Tests::
25 * Index::
26 @end menu
27 @end ifnottex
29 @contents
31 @node Introduction
32 @chapter Introduction
34 @menu
35 * intro_features:: Features
36 * intro_x86_emulation:: x86 emulation
37 * intro_arm_emulation:: ARM emulation
38 * intro_ppc_emulation:: PowerPC emulation
39 * intro_sparc_emulation:: SPARC emulation
40 @end menu
42 @node intro_features
43 @section Features
45 QEMU is a FAST! processor emulator using a portable dynamic
46 translator.
48 QEMU has two operating modes:
50 @itemize @minus
52 @item
53 Full system emulation. In this mode, QEMU emulates a full system
54 (usually a PC), including a processor and various peripherals. It can
55 be used to launch an different Operating System without rebooting the
56 PC or to debug system code.
58 @item
59 User mode emulation (Linux host only). In this mode, QEMU can launch
60 Linux processes compiled for one CPU on another CPU. It can be used to
61 launch the Wine Windows API emulator (@url{}) or
62 to ease cross-compilation and cross-debugging.
64 @end itemize
66 As QEMU requires no host kernel driver to run, it is very safe and
67 easy to use.
69 QEMU generic features:
71 @itemize
73 @item User space only or full system emulation.
75 @item Using dynamic translation to native code for reasonable speed.
77 @item Working on x86 and PowerPC hosts. Being tested on ARM, Sparc32, Alpha and S390.
79 @item Self-modifying code support.
81 @item Precise exceptions support.
83 @item The virtual CPU is a library (@code{libqemu}) which can be used
84 in other projects (look at @file{qemu/tests/qruncom.c} to have an
85 example of user mode @code{libqemu} usage).
87 @end itemize
89 QEMU user mode emulation features:
90 @itemize
91 @item Generic Linux system call converter, including most ioctls.
93 @item clone() emulation using native CPU clone() to use Linux scheduler for threads.
95 @item Accurate signal handling by remapping host signals to target signals.
96 @end itemize
98 QEMU full system emulation features:
99 @itemize
100 @item QEMU can either use a full software MMU for maximum portability or use the host system call mmap() to simulate the target MMU.
101 @end itemize
103 @node intro_x86_emulation
104 @section x86 emulation
106 QEMU x86 target features:
108 @itemize
110 @item The virtual x86 CPU supports 16 bit and 32 bit addressing with segmentation.
111 LDT/GDT and IDT are emulated. VM86 mode is also supported to run DOSEMU.
113 @item Support of host page sizes bigger than 4KB in user mode emulation.
115 @item QEMU can emulate itself on x86.
117 @item An extensive Linux x86 CPU test program is included @file{tests/test-i386}.
118 It can be used to test other x86 virtual CPUs.
120 @end itemize
122 Current QEMU limitations:
124 @itemize
126 @item No SSE/MMX support (yet).
128 @item No x86-64 support.
130 @item IPC syscalls are missing.
132 @item The x86 segment limits and access rights are not tested at every
133 memory access (yet). Hopefully, very few OSes seem to rely on that for
134 normal use.
136 @item On non x86 host CPUs, @code{double}s are used instead of the non standard
137 10 byte @code{long double}s of x86 for floating point emulation to get
138 maximum performances.
140 @end itemize
142 @node intro_arm_emulation
143 @section ARM emulation
145 @itemize
147 @item Full ARM 7 user emulation.
149 @item NWFPE FPU support included in user Linux emulation.
151 @item Can run most ARM Linux binaries.
153 @end itemize
155 @node intro_ppc_emulation
156 @section PowerPC emulation
158 @itemize
160 @item Full PowerPC 32 bit emulation, including privileged instructions,
161 FPU and MMU.
163 @item Can run most PowerPC Linux binaries.
165 @end itemize
167 @node intro_sparc_emulation
168 @section SPARC emulation
170 @itemize
172 @item Somewhat complete SPARC V8 emulation, including privileged
173 instructions, FPU and MMU. SPARC V9 emulation includes most privileged
174 instructions, FPU and I/D MMU, but misses VIS instructions.
176 @item Can run some 32-bit SPARC Linux binaries.
178 @end itemize
180 Current QEMU limitations:
182 @itemize
184 @item Tagged add/subtract instructions are not supported, but they are
185 probably not used.
187 @item IPC syscalls are missing.
189 @item 128-bit floating point operations are not supported, though none of the
190 real CPUs implement them either. FCMPE[SD] are not correctly
191 implemented. Floating point exception support is untested.
193 @item Alignment is not enforced at all.
195 @item Atomic instructions are not correctly implemented.
197 @item Sparc64 emulators are not usable for anything yet.
199 @end itemize
201 @node QEMU Internals
202 @chapter QEMU Internals
204 @menu
205 * QEMU compared to other emulators::
206 * Portable dynamic translation::
207 * Register allocation::
208 * Condition code optimisations::
209 * CPU state optimisations::
210 * Translation cache::
211 * Direct block chaining::
212 * Self-modifying code and translated code invalidation::
213 * Exception support::
214 * MMU emulation::
215 * Hardware interrupts::
216 * User emulation specific details::
217 * Bibliography::
218 @end menu
220 @node QEMU compared to other emulators
221 @section QEMU compared to other emulators
223 Like bochs [3], QEMU emulates an x86 CPU. But QEMU is much faster than
224 bochs as it uses dynamic compilation. Bochs is closely tied to x86 PC
225 emulation while QEMU can emulate several processors.
227 Like Valgrind [2], QEMU does user space emulation and dynamic
228 translation. Valgrind is mainly a memory debugger while QEMU has no
229 support for it (QEMU could be used to detect out of bound memory
230 accesses as Valgrind, but it has no support to track uninitialised data
231 as Valgrind does). The Valgrind dynamic translator generates better code
232 than QEMU (in particular it does register allocation) but it is closely
233 tied to an x86 host and target and has no support for precise exceptions
234 and system emulation.
236 EM86 [4] is the closest project to user space QEMU (and QEMU still uses
237 some of its code, in particular the ELF file loader). EM86 was limited
238 to an alpha host and used a proprietary and slow interpreter (the
239 interpreter part of the FX!32 Digital Win32 code translator [5]).
241 TWIN [6] is a Windows API emulator like Wine. It is less accurate than
242 Wine but includes a protected mode x86 interpreter to launch x86 Windows
243 executables. Such an approach has greater potential because most of the
244 Windows API is executed natively but it is far more difficult to develop
245 because all the data structures and function parameters exchanged
246 between the API and the x86 code must be converted.
248 User mode Linux [7] was the only solution before QEMU to launch a
249 Linux kernel as a process while not needing any host kernel
250 patches. However, user mode Linux requires heavy kernel patches while
251 QEMU accepts unpatched Linux kernels. The price to pay is that QEMU is
252 slower.
254 The new Plex86 [8] PC virtualizer is done in the same spirit as the
255 qemu-fast system emulator. It requires a patched Linux kernel to work
256 (you cannot launch the same kernel on your PC), but the patches are
257 really small. As it is a PC virtualizer (no emulation is done except
258 for some priveledged instructions), it has the potential of being
259 faster than QEMU. The downside is that a complicated (and potentially
260 unsafe) host kernel patch is needed.
262 The commercial PC Virtualizers (VMWare [9], VirtualPC [10], TwoOStwo
263 [11]) are faster than QEMU, but they all need specific, proprietary
264 and potentially unsafe host drivers. Moreover, they are unable to
265 provide cycle exact simulation as an emulator can.
267 @node Portable dynamic translation
268 @section Portable dynamic translation
270 QEMU is a dynamic translator. When it first encounters a piece of code,
271 it converts it to the host instruction set. Usually dynamic translators
272 are very complicated and highly CPU dependent. QEMU uses some tricks
273 which make it relatively easily portable and simple while achieving good
274 performances.
276 The basic idea is to split every x86 instruction into fewer simpler
277 instructions. Each simple instruction is implemented by a piece of C
278 code (see @file{target-i386/op.c}). Then a compile time tool
279 (@file{dyngen}) takes the corresponding object file (@file{op.o})
280 to generate a dynamic code generator which concatenates the simple
281 instructions to build a function (see @file{op.h:dyngen_code()}).
283 In essence, the process is similar to [1], but more work is done at
284 compile time.
286 A key idea to get optimal performances is that constant parameters can
287 be passed to the simple operations. For that purpose, dummy ELF
288 relocations are generated with gcc for each constant parameter. Then,
289 the tool (@file{dyngen}) can locate the relocations and generate the
290 appriopriate C code to resolve them when building the dynamic code.
292 That way, QEMU is no more difficult to port than a dynamic linker.
294 To go even faster, GCC static register variables are used to keep the
295 state of the virtual CPU.
297 @node Register allocation
298 @section Register allocation
300 Since QEMU uses fixed simple instructions, no efficient register
301 allocation can be done. However, because RISC CPUs have a lot of
302 register, most of the virtual CPU state can be put in registers without
303 doing complicated register allocation.
305 @node Condition code optimisations
306 @section Condition code optimisations
308 Good CPU condition codes emulation (@code{EFLAGS} register on x86) is a
309 critical point to get good performances. QEMU uses lazy condition code
310 evaluation: instead of computing the condition codes after each x86
311 instruction, it just stores one operand (called @code{CC_SRC}), the
312 result (called @code{CC_DST}) and the type of operation (called
313 @code{CC_OP}).
315 @code{CC_OP} is almost never explicitely set in the generated code
316 because it is known at translation time.
318 In order to increase performances, a backward pass is performed on the
319 generated simple instructions (see
320 @code{target-i386/translate.c:optimize_flags()}). When it can be proved that
321 the condition codes are not needed by the next instructions, no
322 condition codes are computed at all.
324 @node CPU state optimisations
325 @section CPU state optimisations
327 The x86 CPU has many internal states which change the way it evaluates
328 instructions. In order to achieve a good speed, the translation phase
329 considers that some state information of the virtual x86 CPU cannot
330 change in it. For example, if the SS, DS and ES segments have a zero
331 base, then the translator does not even generate an addition for the
332 segment base.
334 [The FPU stack pointer register is not handled that way yet].
336 @node Translation cache
337 @section Translation cache
339 A 16 MByte cache holds the most recently used translations. For
340 simplicity, it is completely flushed when it is full. A translation unit
341 contains just a single basic block (a block of x86 instructions
342 terminated by a jump or by a virtual CPU state change which the
343 translator cannot deduce statically).
345 @node Direct block chaining
346 @section Direct block chaining
348 After each translated basic block is executed, QEMU uses the simulated
349 Program Counter (PC) and other cpu state informations (such as the CS
350 segment base value) to find the next basic block.
352 In order to accelerate the most common cases where the new simulated PC
353 is known, QEMU can patch a basic block so that it jumps directly to the
354 next one.
356 The most portable code uses an indirect jump. An indirect jump makes
357 it easier to make the jump target modification atomic. On some host
358 architectures (such as x86 or PowerPC), the @code{JUMP} opcode is
359 directly patched so that the block chaining has no overhead.
361 @node Self-modifying code and translated code invalidation
362 @section Self-modifying code and translated code invalidation
364 Self-modifying code is a special challenge in x86 emulation because no
365 instruction cache invalidation is signaled by the application when code
366 is modified.
368 When translated code is generated for a basic block, the corresponding
369 host page is write protected if it is not already read-only (with the
370 system call @code{mprotect()}). Then, if a write access is done to the
371 page, Linux raises a SEGV signal. QEMU then invalidates all the
372 translated code in the page and enables write accesses to the page.
374 Correct translated code invalidation is done efficiently by maintaining
375 a linked list of every translated block contained in a given page. Other
376 linked lists are also maintained to undo direct block chaining.
378 Although the overhead of doing @code{mprotect()} calls is important,
379 most MSDOS programs can be emulated at reasonnable speed with QEMU and
382 Note that QEMU also invalidates pages of translated code when it detects
383 that memory mappings are modified with @code{mmap()} or @code{munmap()}.
385 When using a software MMU, the code invalidation is more efficient: if
386 a given code page is invalidated too often because of write accesses,
387 then a bitmap representing all the code inside the page is
388 built. Every store into that page checks the bitmap to see if the code
389 really needs to be invalidated. It avoids invalidating the code when
390 only data is modified in the page.
392 @node Exception support
393 @section Exception support
395 longjmp() is used when an exception such as division by zero is
396 encountered.
398 The host SIGSEGV and SIGBUS signal handlers are used to get invalid
399 memory accesses. The exact CPU state can be retrieved because all the
400 x86 registers are stored in fixed host registers. The simulated program
401 counter is found by retranslating the corresponding basic block and by
402 looking where the host program counter was at the exception point.
404 The virtual CPU cannot retrieve the exact @code{EFLAGS} register because
405 in some cases it is not computed because of condition code
406 optimisations. It is not a big concern because the emulated code can
407 still be restarted in any cases.
409 @node MMU emulation
410 @section MMU emulation
412 For system emulation, QEMU uses the mmap() system call to emulate the
413 target CPU MMU. It works as long the emulated OS does not use an area
414 reserved by the host OS (such as the area above 0xc0000000 on x86
415 Linux).
417 In order to be able to launch any OS, QEMU also supports a soft
418 MMU. In that mode, the MMU virtual to physical address translation is
419 done at every memory access. QEMU uses an address translation cache to
420 speed up the translation.
422 In order to avoid flushing the translated code each time the MMU
423 mappings change, QEMU uses a physically indexed translation cache. It
424 means that each basic block is indexed with its physical address.
426 When MMU mappings change, only the chaining of the basic blocks is
427 reset (i.e. a basic block can no longer jump directly to another one).
429 @node Hardware interrupts
430 @section Hardware interrupts
432 In order to be faster, QEMU does not check at every basic block if an
433 hardware interrupt is pending. Instead, the user must asynchrously
434 call a specific function to tell that an interrupt is pending. This
435 function resets the chaining of the currently executing basic
436 block. It ensures that the execution will return soon in the main loop
437 of the CPU emulator. Then the main loop can test if the interrupt is
438 pending and handle it.
440 @node User emulation specific details
441 @section User emulation specific details
443 @subsection Linux system call translation
445 QEMU includes a generic system call translator for Linux. It means that
446 the parameters of the system calls can be converted to fix the
447 endianness and 32/64 bit issues. The IOCTLs are converted with a generic
448 type description system (see @file{ioctls.h} and @file{thunk.c}).
450 QEMU supports host CPUs which have pages bigger than 4KB. It records all
451 the mappings the process does and try to emulated the @code{mmap()}
452 system calls in cases where the host @code{mmap()} call would fail
453 because of bad page alignment.
455 @subsection Linux signals
457 Normal and real-time signals are queued along with their information
458 (@code{siginfo_t}) as it is done in the Linux kernel. Then an interrupt
459 request is done to the virtual CPU. When it is interrupted, one queued
460 signal is handled by generating a stack frame in the virtual CPU as the
461 Linux kernel does. The @code{sigreturn()} system call is emulated to return
462 from the virtual signal handler.
464 Some signals (such as SIGALRM) directly come from the host. Other
465 signals are synthetized from the virtual CPU exceptions such as SIGFPE
466 when a division by zero is done (see @code{main.c:cpu_loop()}).
468 The blocked signal mask is still handled by the host Linux kernel so
469 that most signal system calls can be redirected directly to the host
470 Linux kernel. Only the @code{sigaction()} and @code{sigreturn()} system
471 calls need to be fully emulated (see @file{signal.c}).
473 @subsection clone() system call and threads
475 The Linux clone() system call is usually used to create a thread. QEMU
476 uses the host clone() system call so that real host threads are created
477 for each emulated thread. One virtual CPU instance is created for each
478 thread.
480 The virtual x86 CPU atomic operations are emulated with a global lock so
481 that their semantic is preserved.
483 Note that currently there are still some locking issues in QEMU. In
484 particular, the translated cache flush is not protected yet against
485 reentrancy.
487 @subsection Self-virtualization
489 QEMU was conceived so that ultimately it can emulate itself. Although
490 it is not very useful, it is an important test to show the power of the
491 emulator.
493 Achieving self-virtualization is not easy because there may be address
494 space conflicts. QEMU solves this problem by being an executable ELF
495 shared object as the ELF interpreter. That way, it can be
496 relocated at load time.
498 @node Bibliography
499 @section Bibliography
501 @table @asis
503 @item [1]
504 @url{}, Optimizing
505 direct threaded code by selective inlining (1998) by Ian Piumarta, Fabio
506 Riccardi.
508 @item [2]
509 @url{}, Valgrind, an open-source
510 memory debugger for x86-GNU/Linux, by Julian Seward.
512 @item [3]
513 @url{}, the Bochs IA-32 Emulator Project,
514 by Kevin Lawton et al.
516 @item [4]
517 @url{}, the EM86
518 x86 emulator on Alpha-Linux.
520 @item [5]
521 @url{},
522 DIGITAL FX!32: Running 32-Bit x86 Applications on Alpha NT, by Anton
523 Chernoff and Ray Hookway.
525 @item [6]
526 @url{}, Windows API library emulation from
527 Willows Software.
529 @item [7]
530 @url{},
531 The User-mode Linux Kernel.
533 @item [8]
534 @url{},
535 The new Plex86 project.
537 @item [9]
538 @url{},
539 The VMWare PC virtualizer.
541 @item [10]
542 @url{},
543 The VirtualPC PC virtualizer.
545 @item [11]
546 @url{},
547 The TwoOStwo PC virtualizer.
549 @end table
551 @node Regression Tests
552 @chapter Regression Tests
554 In the directory @file{tests/}, various interesting testing programs
555 are available. There are used for regression testing.
557 @menu
558 * test-i386::
559 * linux-test::
560 * qruncom.c::
561 @end menu
563 @node test-i386
564 @section @file{test-i386}
566 This program executes most of the 16 bit and 32 bit x86 instructions and
567 generates a text output. It can be compared with the output obtained with
568 a real CPU or another emulator. The target @code{make test} runs this
569 program and a @code{diff} on the generated output.
571 The Linux system call @code{modify_ldt()} is used to create x86 selectors
572 to test some 16 bit addressing and 32 bit with segmentation cases.
574 The Linux system call @code{vm86()} is used to test vm86 emulation.
576 Various exceptions are raised to test most of the x86 user space
577 exception reporting.
579 @node linux-test
580 @section @file{linux-test}
582 This program tests various Linux system calls. It is used to verify
583 that the system call parameters are correctly converted between target
584 and host CPUs.
586 @node qruncom.c
587 @section @file{qruncom.c}
589 Example of usage of @code{libqemu} to emulate a user mode i386 CPU.
591 @node Index
592 @chapter Index
593 @printindex cp
595 @bye