Get rid of the ThunkData stubs, these are not functions.
[wine/multimedia.git] / documentation / architecture.sgml
blob62f9f89426a10261fe257652e0e3249dbbf91aab
1 <chapter id="architecture">
2 <title>Overview</title>
3 <para>Brief overview of Wine's architecture...</para>
5 <sect1 id="basic-overview">
6 <title>Wine Overview</title>
8 <para>
9 With the fundamental architecture of Wine stabilizing, and
10 people starting to think that we might soon be ready to
11 actually release this thing, it may be time to take a look at
12 how Wine actually works and operates.
13 </para>
15 <sect2>
16 <title>Foreword</title>
17 <para>
18 Wine is often used as a recursive acronym, standing for
19 "Wine Is Not an Emulator". Sometimes it is also known to be
20 used for "Windows Emulator". In a way, both meanings are
21 correct, only seen from different perspectives. The first
22 meaning says that Wine is not a virtual machine, it does not
23 emulate a CPU, and you are not supposed to install
24 Windows nor any Windows device drivers on top of it; rather,
25 Wine is an implementation of the Windows API, and can be
26 used as a library to port Windows applications to Unix. The
27 second meaning, obviously, is that to Windows binaries
28 (<filename>.exe</filename> files), Wine does look like
29 Windows, and emulates its behaviour and quirks rather
30 closely.
31 </para>
32 <note>
33 <title>"Emulator"</title>
34 <para>
35 The "Emulator" perspective should not be thought of as if
36 Wine is a typical inefficient emulation layer that means
37 Wine can't be anything but slow - the faithfulness to the
38 badly designed Windows API may of course impose a minor
39 overhead in some cases, but this is both balanced out by
40 the higher efficiency of the Unix platforms Wine runs on,
41 and that other possible abstraction libraries (like Motif,
42 GTK+, CORBA, etc) has a runtime overhead typically
43 comparable to Wine's.
44 </para>
45 </note>
46 </sect2>
48 <sect2>
49 <title>Executables</title>
50 <para>
51 Wine's main task is to run Windows executables under non
52 Windows operating systems. It supports different types of
53 executables:
54 <itemizedlist>
55 <listitem>
56 <para>
57 DOS executable. Those are even older programs, using
58 the DOS format (either <filename>.com</filename> or
59 <filename>.exe</filename> (the later being also called
60 MZ)).
61 </para>
62 </listitem>
63 <listitem>
64 <para>
65 Windows NE executable, also called 16 bit. They were
66 the native processes run by Windows 2.x and 3.x. NE
67 stands for New Executable &lt;g&gt;.
68 </para>
69 </listitem>
70 <listitem>
71 <para>
72 Windows PE executable. These are programs were
73 introduced in Windows 95 (and became the native
74 formats for all later Windows version), even if 16 bit
75 applications were still supported. PE stands for
76 Portable Executable, in a sense where the format of
77 the executable (as a file) is independent of the CPU
78 (even if the content of the file - the code - is CPU
79 dependent).
80 </para>
81 </listitem>
82 <listitem>
83 <para>
84 WineLib executable. These are applications, written
85 using the Windows API, but compiled as a Unix
86 executable. Wine provides the tools to create such
87 executables.
88 </para>
89 </listitem>
90 </itemizedlist>
91 </para>
92 <para>
93 Let's quickly review the main differences for the supported
94 executables:
95 <table>
96 <title>Wine executables</title>
97 <tgroup cols="5" align="left" colsep="1" rowsep="1">
98 <thead>
99 <row>
100 <entry></entry>
101 <entry>DOS (.COM or .EXE)</entry>
102 <entry>Win16 (NE)</entry>
103 <entry>Win32 (PE)</entry>
104 <entry>WineLib</entry>
105 </row>
106 </thead>
107 <tbody>
108 <row>
109 <entry>Multitasking</entry>
110 <entry>Only one application at a time (except for TSR)</entry>
111 <entry>Cooperative</entry>
112 <entry>Preemptive</entry>
113 <entry>Preemptive</entry>
114 </row>
115 <row>
116 <entry>Address space</entry>
117 <entry>
118 One MB of memory, where each application is loaded
119 and unloaded.
120 </entry>
121 <entry>
122 All 16 bit applications share a single address
123 space, protected mode.
124 </entry>
125 <entry>
126 Each application has it's own address
127 space. Requires MMU support from CPU.
128 </entry>
129 <entry>
130 Each application has it's own address
131 space. Requires MMU support from CPU.
132 </entry>
133 </row>
134 <row>
135 <entry>Windows API</entry>
136 <entry>
137 No Windows API but the DOS API (like <function>Int
138 21h</function> traps).
139 </entry>
140 <entry>
141 Will call the 16 bit Windows API.
142 </entry>
143 <entry>
144 Will call the 32 bit Windows API.
145 </entry>
146 <entry>
147 Will call the 32 bit Windows API, and possibly
148 also the Unix APIs.
149 </entry>
150 </row>
151 <row>
152 <entry>Code (CPU level)</entry>
153 <entry>
154 Only available on x86 in real mode. Code and data
155 are in segmented forms, with 16 bit
156 offsets. Processor is in real mode.
157 </entry>
158 <entry>
159 Only available on IA-32 architectures, code and
160 data are in segmented forms, with 16 bit offsets
161 (hence the 16 bit name). Processor is in protected
162 mode.
163 </entry>
164 <entry>
165 Available (with NT) on several CPUs, including
166 IA-32. On this CPU, uses a flat memory model with
167 32 bit offsets (hence the 32 bit name).
168 </entry>
169 <entry>
170 Flat model, with 32 bit addresses.
171 </entry>
172 </row>
173 <row>
174 <entry>Multi-threading</entry>
175 <entry>Not available.</entry>
176 <entry>Not available.</entry>
177 <entry>
178 Available.
179 </entry>
180 <entry>
181 Available, but must use the Win32 APIs for
182 threading and synchronization, not the Unix ones.
183 </entry>
184 </row>
185 </tbody>
186 </tgroup>
187 </table>
188 </para>
190 <para>
191 Wine deals with this issue by launching a separate Wine
192 process (which is in fact a Unix process) for each Win32
193 process, but not for Win16 tasks. Win16 tasks (as well as
194 DOS programs) are run as different intersynchronized
195 Unix-threads in the same dedicated Wine process; this Wine
196 process is commonly known as a <firstterm>WOW</firstterm>
197 process (Windows on Windows), referring to a similar
198 mechanism used by Windows NT.
199 </para>
200 <para>
201 Synchronization between the Win16 tasks running in the WOW
202 process is normally done through the Win16 mutex - whenever
203 one of them is running, it holds the Win16 mutex, keeping
204 the others from running. When the task wishes to let the
205 other tasks run, the thread releases the Win16 mutex, and
206 one of the waiting threads will then acquire it and let its
207 task run.
208 </para>
209 </sect2>
210 </sect1>
212 <sect1>
213 <title>Standard Windows Architectures</title>
215 <sect2>
216 <title>Windows 9x architecture</title>
218 <para>
219 The windows architecture (Win 9x way) looks like this:
220 <screen>
221 +---------------------+ \
222 | Windows EXE | } application
223 +---------------------+ /
225 +---------+ +---------+ \
226 | Windows | | Windows | \ application & system DLLs
227 | DLL | | DLL | /
228 +---------+ +---------+ /
230 +---------+ +---------+ \
231 | GDI32 | | USER32 | \
232 | DLL | | DLL | \
233 +---------+ +---------+ } core system DLLs
234 +---------------------+ /
235 | Kernel32 DLL | /
236 +---------------------+ /
238 +---------------------+ \
239 | Win9x kernel | } kernel space
240 +---------------------+ /
242 +---------------------+ \
243 | Windows low-level | \ drivers (kernel space)
244 | drivers | /
245 +---------------------+ /
246 </screen>
247 </para>
248 </sect2>
249 <sect2>
250 <title>Windows NT architecture</title>
252 <para>
253 The windows architecture (Windows NT way) looks like the
254 following drawing. Note the new DLL (NTDLL) which allows
255 implementing different subsystems (as win32); kernel32 in NT
256 architecture implements the Win32 subsystem on top of NTDLL.
257 <screen>
258 +---------------------+ \
259 | Windows EXE | } application
260 +---------------------+ /
262 +---------+ +---------+ \
263 | Windows | | Windows | \ application & system DLLs
264 | DLL | | DLL | /
265 +---------+ +---------+ /
267 +---------+ +---------+ +-----------+ \
268 | GDI32 | | USER32 | | | \
269 | DLL | | DLL | | | \
270 +---------+ +---------+ | | \ core system DLLs
271 +---------------------+ | | / (on the left side)
272 | Kernel32 DLL | | Subsystem | /
273 | (Win32 subsystem) | |Posix, OS/2| /
274 +---------------------+ +-----------+ /
276 +---------------------------------------+
277 | NTDLL.DLL |
278 +---------------------------------------+
280 +---------------------------------------+ \
281 | NT kernel | } NT kernel (kernel space)
282 +---------------------------------------+ /
283 +---------------------------------------+ \
284 | Windows low-level drivers | } drivers (kernel space)
285 +---------------------------------------+ /
286 </screen>
287 </para>
288 <para>
289 Note also (not depicted in schema above) that the 16 bit
290 applications are supported in a specific subsystem.
291 Some basic differences between the Win9x and the NT
292 architectures include:
293 <itemizedlist>
294 <listitem>
295 <para>
296 Several subsystems (Win32, Posix...) can be run on NT,
297 while not on Win 9x
298 </para>
299 </listitem>
300 <listitem>
301 <para>
302 Win 9x roots its architecture in 16 bit systems, while
303 NT is truely a 32 bit system.
304 </para>
305 </listitem>
306 <listitem>
307 <para>
308 The drivers model and interfaces in Win 9x and NT are
309 different (even if Microsoft tried to bridge the gap
310 with some support of WDM drivers in Win 98 and above).
311 </para>
312 </listitem>
313 </itemizedlist>
314 </para>
315 </sect2>
316 </sect1>
318 <sect1>
319 <title>Wine architecture</title>
321 <sect2>
322 <title>Global picture</title>
324 <para>
325 Wine implementation is closer to the Windows NT
326 architecture, even if several subsystems are not implemented
327 yet (remind also that 16bit support is implemented in a 32-bit
328 Windows EXE, not as a subsystem). Here's the overall picture:
329 <screen>
330 +---------------------+ \
331 | Windows EXE | } application
332 +---------------------+ /
334 +---------+ +---------+ \
335 | Windows | | Windows | \ application & system DLLs
336 | DLL | | DLL | /
337 +---------+ +---------+ /
339 +---------+ +---------+ +-----------+ +--------+ \
340 | GDI32 | | USER32 | | | | | \
341 | DLL | | DLL | | | | Wine | \
342 +---------+ +---------+ | | | Server | \ core system DLLs
343 +---------------------+ | | | | / (on the left side)
344 | Kernel32 DLL | | Subsystem | | NT-like| /
345 | (Win32 subsystem) | |Posix, OS/2| | Kernel | /
346 +---------------------+ +-----------+ | | /
348 +---------------------------------------+ | |
349 | NTDLL | | |
350 +---------------------------------------+ +--------+
352 +---------------------------------------+ \
353 | Wine executable (wine-?thread) | } unix executable
354 +---------------------------------------+ /
355 +---------------------------------------------------+ \
356 | Wine drivers | } Wine specific DLLs
357 +---------------------------------------------------+ /
359 +------------+ +------------+ +--------------+ \
360 | libc | | libX11 | | other libs | } unix shared libraries
361 +------------+ +------------+ +--------------+ / (user space)
363 +---------------------------------------------------+ \
364 | Unix kernel (Linux,*BSD,Solaris,OS/X) | } (Unix) kernel space
365 +---------------------------------------------------+ /
366 +---------------------------------------------------+ \
367 | Unix device drivers | } Unix drivers (kernel space)
368 +---------------------------------------------------+ /
369 </screen>
370 </para>
372 <para>
373 Wine must at least completely replace the "Big Three" DLLs
374 (KERNEL/KERNEL32, GDI/GDI32, and USER/USER32), which all
375 other DLLs are layered on top of. But since Wine is (for
376 various reasons) leaning towards the NT way of implementing
377 things, the NTDLL is another core DLL to be implemented in
378 Wine, and many KERNEL32 and ADVAPI32 features will be
379 implemented through the NTDLL.
380 </para>
381 <para>
382 As of today, no real subsystem (apart the Win32 one) has
383 been implemented in Wine.
384 </para>
385 <para>
386 The Wine server provides the backbone for the implementation
387 of the core DLLs. It mainly implementents inter-process
388 synchronization and object sharing. It can be seen, from a
389 functional point of view, as a NT kernel (even if the APIs
390 and protocols used between Wine's DLL and the Wine server
391 are Wine specific).
392 </para>
393 <para>
394 Wine uses the Unix drivers to access the various hardware
395 pieces on the box. However, in some cases, Wine will
396 provide a driver (in Windows sense) to a physical hardware
397 device. This driver will be a proxy to the Unix driver
398 (this is the case, for example, for the graphical part
399 with X11 or SDL drivers, audio with OSS or ALSA drivers...).
400 </para>
401 <para>
402 All DLLs provided by Wine try to stick as much as possible
403 to the exported APIs from the Windows platforms. There are
404 rare cases where this is not the case, and have been
405 propertly documented (Wine DLLs export some Wine specific
406 APIs). Usually, those are prefixed with
407 <function>__wine</function>.
408 </para>
409 <para>
410 Let's now review in greater details all of those components.
411 </para>
412 </sect2>
414 <sect2>
415 <title>The Wine server</title>
416 <para>
417 The Wine server is among the most confusing concepts in
418 Wine. What is its function in Wine? Well, to be brief, it
419 provides Inter-Process Communication (IPC),
420 synchronization, and process/thread management. When the
421 Wine server launches, it creates a Unix socket for the
422 current host based on (see below) your home directory's
423 <filename>.wine</filename> subdirectory (or wherever the
424 <constant>WINEPREFIX</constant> environment variable
425 points to) - all Wine processes launched later connects to
426 the Wine server using this socket. (If a Wine server was
427 not already running, the first Wine process will start up
428 the Wine server in auto-terminate mode (i.e. the Wine
429 server will then terminate itself once the last Wine
430 process has terminated).)
431 </para>
432 <para>
433 In earlier versions of Wine the master socket mentioned
434 above was actually created in the configuration directory;
435 either your home directory's <filename>/wine</filename>
436 subdirectory or wherever the
437 <constant>WINEPREFIX</constant> environment variable
438 points>. Since that might not be possible the socket is
439 actually created within the <filename>/tmp</filename>
440 directory with a name that reflects the configuration
441 directory. This means that there can actually be several
442 separate copies of the Wine server running; one per
443 combination of user and configuration directory. Note that
444 you should not have several users using the same
445 configuration directory at the same time; they will have
446 different copies of the Wine server running and this could
447 well lead to problems with the registry information that
448 they are sharing.
449 </para>
450 <para>
451 Every thread in each Wine process has its own request
452 buffer, which is shared with the Wine server. When a
453 thread needs to synchronize or communicate with any other
454 thread or process, it fills out its request buffer, then
455 writes a command code through the socket. The Wine server
456 handles the command as appropriate, while the client
457 thread waits for a reply. In some cases, like with the
458 various <function>WaitFor???</function> synchronization
459 primitives, the server handles it by marking the client
460 thread as waiting and does not send it a reply before the
461 wait condition has been satisfied.
462 </para>
463 <para>
464 The Wine server itself is a single and separate Unix
465 process and does not have its own threading - instead, it
466 is built on top of a large <function>poll()</function>
467 loop that alerts the Wine server whenever anything
468 happens, such as a client having sent a command, or a wait
469 condition having been satisfied. There is thus no danger
470 of race conditions inside the Wine server itself - it is
471 often called upon to do operations that look completely
472 atomic to its clients.
473 </para>
474 <para>
475 Because the Wine server needs to manage processes,
476 threads, shared handles, synchronization, and any related
477 issues, all the clients' Win32 objects are also managed by
478 the Wine server, and the clients must send requests to the
479 Wine server whenever they need to know any Win32 object
480 handle's associated Unix file descriptor (in which case
481 the Wine server duplicates the file descriptor, transmits
482 it back to the client, and leaves it to the client to
483 close the duplicate when the client has finished with
484 it).
485 </para>
486 </sect2>
488 <sect2>
489 <title>
490 Wine builtin DLLs: about Relays, Thunks, and DLL
491 descriptors
492 </title>
493 <para>
494 This section mainly applies to builtin DLLs (DLLs provided
495 by Wine). See section <xref linkend="arch-dlls"> for the
496 details on native vs. builtin DLL handling.
497 </para>
498 <para>
499 Loading a Windows binary into memory isn't that hard by
500 itself, the hard part is all those various DLLs and entry
501 points it imports and expects to be there and function as
502 expected; this is, obviously, what the entire Wine
503 implementation is all about. Wine contains a range of DLL
504 implementations. You can find the DLLs implementation in the
505 <filename>dlls/</filename> directory.
506 </para>
507 <para>
508 Each DLL (at least, the 32 bit version, see below) is
509 implemented in a Unix shared library. The file name of this
510 shared library is the module name of the DLL with a
511 <filename>.dll.so</filename> suffix (or
512 <filename>.drv.so</filename> or any other relevant extension
513 depending on the DLL type). This shared library contains the
514 code itself for the DLL, as well as some more information,
515 as the DLL resources and a Wine specific DLL descriptor.
516 </para>
517 <para>
518 The DLL descriptor, when the DLL is instanciated, is used to
519 create an in-memory PE header, which will provide access to
520 various information about the DLL, including but not limited
521 to its entry point, its resources, its sections, its debug
522 information...
523 </para>
524 <para>
525 The DLL descriptor and entry point table is generated by
526 the <command>winebuild</command> tool (previously just
527 named <command>build</command>), taking DLL specification
528 files with the extension <filename>.spec</filename> as
529 input. Resources (after compilation by
530 <command>wrc</command>) or message tables (after
531 compilation by <command>wmc</command>) are also added to
532 the descriptor by <command>winebuild</command>.
533 </para>
534 <para>
535 Once an application module wants to import a DLL, Wine
536 will look at:
537 <itemizedlist>
538 <listitem>
539 <para>
540 through its list of registered DLLs (in fact, both
541 the already loaded DLLs, and the already loaded
542 shared libraries which has registered a DLL
543 descriptor). Since, the DLL descriptor is
544 automatically registered when the shared library is
545 loaded - remember, registration call is put inside a
546 shared library constructor - using the
547 <constant>PRELOAD</constant> environment variable
548 when running a Wine process can force the
549 registration of some DLL descriptors.
550 </para>
551 </listitem>
552 <listitem>
553 <para>
554 If it's not registered, Wine will look for it on
555 disk, building the shared library name from the DLL
556 module name. Directory searched for are specified by
557 the <constant>WINEDLLPATH</constant> environment
558 variable.
559 </para>
560 </listitem>
561 <listitem>
562 <para>
563 Failing that, it will look for a real Windows
564 <filename>.DLL</filename> file to use, and look
565 through its imports, etc) and use the loading of
566 native DLLs.
567 </para>
568 </listitem>
569 </itemizedlist>
570 </para>
571 <para>
572 After the DLL has been identified (assuming it's still a
573 native one), it's mapped into memory using a
574 <function>dlopen()</function> call. Note, that Wine doesn't
575 use the shared library mechanisms for resolving and/or
576 importing functions between two shared libraries (for two
577 DLLs). The shared library is only used for providing a way
578 to load a piece of code on demand. This piece of code,
579 thanks the DLL descriptor, will provide the same type of
580 information a native DLL would. Wine can then use the same
581 code for native and builtin DLL to handle imports/exports.
582 </para>
583 <para>
584 Wine also relies on the dynamic loading features of the Unix
585 shared libraries to relocate the DLLs if needed (the same
586 DLL can be loaded at different address in two different
587 processes, and even in two consecutive run of the same
588 executable if the order of loading the DLLs differ).
589 </para>
590 <para>
591 The DLL descriptor is registered in the Wine realm using
592 some tricks. The <command>winebuild</command> tool, while
593 creating the code for DLL descriptor, also creates a
594 constructor, that will be called when the shared library is
595 loaded into memory. This constructor will actually register
596 the descriptor to the Wine DLL loader. Hence, before the
597 <function>dlopen</function> call returns, the DLL descriptor
598 will be known and registered. This also helps to deal with
599 the cases where there's still dependencies (at the ELF
600 shared lib level, not at the embedded DLL level) between
601 different shared libraries: the embedded DLLs will be
602 properly registered, and even loaded (from a Windows point
603 of view).
604 </para>
605 <para>
606 Since Wine is 32-bit code itself, and if the compiler
607 supports Windows' calling convention, <type>stdcall</type>
608 (<command>gcc</command> does), Wine can resolve imports
609 into Win32 code by substituting the addresses of the Wine
610 handlers directly without any thunking layer in
611 between. This eliminates the overhead most people
612 associate with "emulation", and is what the applications
613 expect anyway.
614 </para>
615 <para>
616 However, if the user specified <parameter>WINEDEBUG=+relay
617 </parameter>, a thunk layer is inserted between the
618 application imports and the Wine handlers (actually the
619 export table of the DLL is modified, and a thunk is
620 inserted in the table); this layer is known as "relay"
621 because all it does is print out the arguments/return
622 values (by using the argument lists in the DLL
623 descriptor's entry point table), then pass the call on,
624 but it's invaluable for debugging misbehaving calls into
625 Wine code. A similar mechanism also exists between Windows
626 DLLs - Wine can optionally insert thunk layers between
627 them, by using <parameter>WINEDEBUG=+snoop</parameter>,
628 but since no DLL descriptor information exists for
629 non-Wine DLLs, this is less reliable and may lead to
630 crashes.
631 </para>
632 <para>
633 For Win16 code, there is no way around thunking - Wine
634 needs to relay between 16-bit and 32-bit code. These
635 thunks switch between the app's 16-bit stack and Wine's
636 32-bit stack, copies and converts arguments as appropriate
637 (an int is 16 bit 16-bit and 32 bits in 32-bit, pointers
638 are segmented in 16 bit (and also near or far) but are 32
639 bit linear values in 32 bit), and handles the Win16
640 mutex. Some finer control can be obtained on the
641 conversion, see <command>winebuild</command> reference
642 manual for the details. Suffice to say that the kind of
643 intricate stack content juggling this results in, is not
644 exactly suitable study material for beginners.
645 </para>
646 <para>
647 A DLL descriptor is also created for every 16 bit
648 DLL. However, this DLL normally paired with a 32 bit
649 DLL. Either, it's the 16 bit counterpart of the 16 bit DLL
650 (KRNL386.EXE for KERNEL32, USER for USER32...), or a 16
651 bit DLL directly linked to a 32 bit DLL (like SYSTEM for
652 KERNEL32, or DDEML for USER32). In those cases, the 16 bit
653 descriptor(s) is (are) inserted in the same shared library
654 as the the corresponding 32 bit DLL. Wine will also create
655 symbolic links between kernel32.dll.so and system.dll.so
656 so that loading of either
657 <filename>kernel32.dll</filename> or
658 <filename>system.dll</filename> will end up on the same
659 shared library.
660 </para>
661 </sect2>
662 <sect2 id="arch-dlls">
663 <title>Wine/Windows DLLs</title>
665 <para>
666 This document mainly deals with the status of current DLL
667 support by Wine. The Wine ini file currently supports
668 settings to change the load order of DLLs. The load order
669 depends on several issues, which results in different settings
670 for various DLLs.
671 </para>
673 <sect3>
674 <title>Pros of Native DLLs</title>
676 <para>
677 Native DLLs of course guarantee 100% compatibility for
678 routines they implement. For example, using the native USER
679 DLL would maintain a virtually perfect and Windows 95-like
680 look for window borders, dialog controls, and so on. Using
681 the built-in Wine version of this library, on the other
682 hand, would produce a display that does not precisely mimic
683 that of Windows 95. Such subtle differences can be
684 engendered in other important DLLs, such as the common
685 controls library COMMCTRL or the common dialogs library
686 COMMDLG, when built-in Wine DLLs outrank other types in load
687 order.
688 </para>
689 <para>
690 More significant, less aesthetically-oriented problems can
691 result if the built-in Wine version of the SHELL DLL is
692 loaded before the native version of this library. SHELL
693 contains routines such as those used by installer utilities
694 to create desktop shortcuts. Some installers might fail when
695 using Wine's built-in SHELL.
696 </para>
697 </sect3>
699 <sect3>
700 <title>Cons of Native DLLs</title>
702 <para>
703 Not every application performs better under native DLLs. If
704 a library tries to access features of the rest of the system
705 that are not fully implemented in Wine, the native DLL might
706 work much worse than the corresponding built-in one, if at
707 all. For example, the native Windows GDI library must be
708 paired with a Windows display driver, which of course is not
709 present under Intel Unix and Wine.
710 </para>
711 <para>
712 Finally, occasionally built-in Wine DLLs implement more
713 features than the corresponding native Windows DLLs.
714 Probably the most important example of such behavior is the
715 integration of Wine with X provided by Wine's built-in USER
716 DLL. Should the native Windows USER library take load-order
717 precedence, such features as the ability to use the
718 clipboard or drag-and-drop between Wine windows and X
719 windows will be lost.
720 </para>
721 </sect3>
723 <sect3>
724 <title>Deciding Between Native and Built-In DLLs</title>
726 <para>
727 Clearly, there is no one rule-of-thumb regarding which
728 load-order to use. So, you must become familiar with
729 what specific DLLs do and which other DLLs or features
730 a given library interacts with, and use this information
731 to make a case-by-case decision.
732 </para>
733 </sect3>
735 <sect3>
736 <title>Load Order for DLLs</title>
738 <para>
739 Using the DLL sections from the wine configuration file, the
740 load order can be tweaked to a high degree. In general it is
741 advised not to change the settings of the configuration
742 file. The default configuration specifies the right load
743 order for the most important DLLs.
744 </para>
745 <para>
746 The default load order follows this algorithm: for all DLLs
747 which have a fully-functional Wine implementation, or where
748 the native DLL is known not to work, the built-in library
749 will be loaded first. In all other cases, the native DLL
750 takes load-order precedence.
751 </para>
752 <para>
753 The <varname>DefaultLoadOrder</varname> from the
754 [DllDefaults] section specifies for all DLLs which version
755 to try first. See manpage for explanation of the arguments.
756 </para>
757 <para>
758 The [DllOverrides] section deals with DLLs, which need a
759 different-from-default treatment.
760 </para>
761 <para>
762 The [DllPairs] section is for DLLs, which must be loaded in
763 pairs. In general, these are DLLs for either 16-bit or
764 32-bit applications. In most cases in Windows, the 32-bit
765 version cannot be used without its 16-bit counterpart. For
766 Wine, it is customary that the 16-bit implementations rely
767 on the 32-bit implementations and cast the results back to
768 16-bit arguments. Changing anything in this section is bound
769 to result in errors.
770 </para>
771 <para>
772 For the future, the Wine implementation of Windows DLL seems
773 to head towards unifying the 16 and 32 bit DLLs wherever
774 possible, resulting in larger DLLs. They are stored in the
775 <filename>dlls/</filename> subdirectory using the 32-bit
776 name.
777 </para>
778 </sect3>
779 </sect2>
781 <sect2 id="arch-mem">
782 <title>Memory management</title>
783 <para>
784 Every Win32 process in Wine has its own dedicated native
785 process on the host system, and therefore its own address
786 space. This section explores the layout of the Windows
787 address space and how it is emulated.
788 </para>
790 <para>
791 Firstly, a quick recap of how virtual memory works. Physical
792 memory in RAM chips is split into
793 <emphasis>frames</emphasis>, and the memory that each
794 process sees is split into <emphasis>pages</emphasis>. Each
795 process has its own 4 gigabytes of address space (4gig being
796 the maximum space addressable with a 32 bit pointer). Pages
797 can be mapped or unmapped: attempts to access an unmapped
798 page cause an
799 <constant>EXCEPTION_ACCESS_VIOLATION</constant> which has
800 the easily recognizable code of
801 <constant>0xC0000005</constant>. Any page can be mapped to
802 any frame, therefore you can have multiple addresses which
803 actually "contain" the same memory. Pages can also be mapped
804 to things like files or swap space, in which case accessing
805 that page will cause a disk access to read the contents into
806 a free frame.
807 </para>
809 <sect3>
810 <title>Initial layout (in Windows)</title>
811 <para>
812 When a Win32 process starts, it does not have a clear
813 address space to use as it pleases. Many pages are already
814 mapped by the operating system. In particular, the EXE
815 file itself and any DLLs it needs are mapped into memory,
816 and space has been reserved for the stack and a couple of
817 heaps (zones used to allocate memory to the app
818 from). Some of these things need to be at a fixed address,
819 and others can be placed anywhere.
820 </para>
822 <para>
823 The EXE file itself is usually mapped at address 0x400000
824 and up: indeed, most EXEs have their relocation records
825 stripped which means they must be loaded at their base
826 address and cannot be loaded at any other address.
827 </para>
829 <para>
830 DLLs are internally much the same as EXE files but they
831 have relocation records, which means that they can be
832 mapped at any address in the address space. Remember we
833 are not dealing with physical memory here, but rather
834 virtual memory which is different for each
835 process. Therefore <filename>OLEAUT32.DLL</filename> may
836 be loaded at one address in one process, and a totally
837 different one in another. Ensuring all the functions
838 loaded into memory can find each other is the job of the
839 Windows dynamic linker, which is a part of NTDLL.
840 </para>
841 <para>
842 So, we have the EXE and its DLLs mapped into memory. Two
843 other very important regions also exist: the stack and the
844 process heap. The process heap is simply the equivalent of
845 the libc <function>malloc</function> arena on UNIX: it's a
846 region of memory managed by the OS which
847 <function>malloc</function>/<function>HeapAlloc</function>
848 partitions and hands out to the application. Windows
849 applications can create several heaps but the process heap
850 always exists.
851 </para>
852 <para>
853 Windows 9x also implements another kind of heap: the
854 shared heap. The shared heap is unusual in that
855 anything allocated from it will be visible in every other
856 process.
857 </para>
858 </sect3>
860 <sect3>
861 <title>Comparison</title>
862 <para>
863 So far we've assumed the entire 4 gigs of address space is
864 available for the application. In fact that's not so: only
865 the lower 2 gigs are available, the upper 2 gigs are on
866 Windows NT used by the operating system and hold the
867 kernel (from 0x80000000). Why is the kernel mapped into
868 every address space? Mostly for performance: while it's
869 possible to give the kernel its own address space too -
870 this is what Ingo Molnars 4G/4G VM split patch does for
871 Linux - it requires that every system call into the kernel
872 switches address space. As that is a fairly expensive
873 operation (requires flushing the translation lookaside
874 buffers etc) and syscalls are made frequently it's best
875 avoided by keeping the kernel mapped at a constant
876 position in every processes address space.
877 </para>
879 <para>
880 Basically, the comparison of memory mappings looks as
881 follows:
882 <table>
883 <title>Memory layout (Windows and Wine)</title>
884 <tgroup cols="4" align="left" colsep="1" rowsep="1">
885 <thead>
886 <row>
887 <entry>Address</entry>
888 <entry>Windows 9x</entry>
889 <entry>Windows NT</entry>
890 <entry>Linux</entry>
891 </row>
892 </thead>
893 <tbody>
894 <row>
895 <entry>00000000-7fffffff</entry>
896 <entry>User</entry>
897 <entry>User</entry>
898 <entry>User</entry>
899 </row>
900 <row>
901 <entry>80000000-bfffffff</entry>
902 <entry>Shared</entry>
903 <entry>User</entry>
904 <entry>User</entry>
905 </row>
906 <row>
907 <entry>c0000000-ffffffff</entry>
908 <entry>Kernel</entry>
909 <entry>Kernel</entry>
910 <entry>Kernel</entry>
911 </row>
912 </tbody>
913 </tgroup>
914 </table>
915 </para>
917 <para>
918 On Windows 9x, in fact only the upper gigabyte
919 (<constant>0xC0000000</constant> and up) is used by the
920 kernel, the region from 2 to 3 gigs is a shared area used
921 for loading system DLLs and for file mappings. The bottom
922 2 gigs on both NT and 9x are available for the programs
923 memory allocation and stack.
924 </para>
925 </sect3>
927 <sect3>
928 <title>Implementation</title>
929 <para>
930 Wine (with a bit of black magic) is able to map all items
931 at the correct locations as depicted above.
932 </para>
933 <para>
934 Wine also implements the shared heap so native win9x DLLs
935 can be used. This heap is always created at the
936 <constant>SYSTEM_HEAP_BASE</constant> address or
937 <constant>0x80000000</constant> and defaults to 16
938 megabytes in size.
939 </para>
940 <para>
941 There are a few other magic locations. The bottom 64k of
942 memory is deliberately left unmapped to catch null pointer
943 dereferences. The region from 64k to 1mb+64k are reserved
944 for DOS compatibility and contain various DOS data
945 structures. Finally, the address space also contains
946 mappings for the Wine binary itself, any native libaries
947 Wine is using, the glibc malloc arena and so on.
948 </para>
949 </sect3>
951 <sect3 id="address-space">
952 <title>Laying out the address space</title>
954 <para>
955 Up until about the start of 2004, the Linux address space
956 very much resembled the Windows 9x layout: the kernel sat
957 in the top gigabyte, the bottom pages were unmapped to
958 catch null pointer dereferences, and the rest was
959 free. The kernels mmap algorithm was predictable: it would
960 start by mapping files at low addresses and work up from
961 there.
962 </para>
964 <para>
965 The development of a series of new low level patches
966 violated many of these assumptions, and resulted in Wine
967 needing to force the Win32 address space layout upon the
968 system. This section looks at why and how this is done.
969 </para>
971 <para>
972 The exec-shield patch increases security by randomizing
973 the kernels mmap algorithms. Rather than consistently
974 choosing the same addresses given the same sequence of
975 requests, the kernel will now choose randomized
976 addresses. Because the Linux dynamic linker
977 (ld-linux.so.2) loads DSOs into memory by using mmap, this
978 means that DSOs are no longer loaded at predictable
979 addresses, so making it harder to attack software by using
980 buffer overflows. It also attempts to relocate certain
981 binaries into a special low area of memory known as the
982 ASCII armor so making it harder to jump into them when
983 using string based attacks.
984 </para>
986 <para>
987 Prelink is a technology that enhances startup times by
988 precalculating ELF global offset tables then saving the
989 results inside the native binaries themselves. By grid
990 fitting each DSO into the address space, the dynamic
991 linker does not have to perform as many relocations so
992 allowing applications that heavily rely on dynamic linkage
993 to be loaded into memory much quicker. Complex C++
994 applications such as Mozilla, OpenOffice and KDE can
995 especially benefit from this technique.
996 </para>
998 <para>
999 The 4G VM split patch was developed by Ingo Molnar. It
1000 gives the Linux kernel its own address space, thereby
1001 allowing processes to access the maximum addressable
1002 amount of memory on a 32-bit machine: 4 gigabytes. It
1003 allows people with lots of RAM to fully utilise that in
1004 any given process at the cost of performance: the reason
1005 behind giving the kernel a part of each processes address
1006 space was to avoid the overhead of switching on each
1007 syscall.
1008 </para>
1010 <para>
1011 Each of these changes alter the address space in a way
1012 incompatible with Windows. Prelink and exec-shield mean
1013 that the libraries Wine uses can be placed at any point in
1014 the address space: typically this meant that a library was
1015 sitting in the region that the EXE you wanted to run had
1016 to be loaded (remember that unlike DLLs, EXE files cannot
1017 be moved around in memory). The 4G VM split means that
1018 programs could receive pointers to the top gigabyte of
1019 address space which some are not prepared for (they may
1020 store extra information in the high bits of a pointer, for
1021 instance). In particular, in combination with exec-shield
1022 this one is especially deadly as it's possible the process
1023 heap could be allocated beyond ADDRESS_SPACE_LIMIT which
1024 causes Wine initialization to fail.
1025 </para>
1027 <para>
1028 The solution to these problems is for Wine to reserve
1029 particular parts of the address space so that areas that
1030 we don't want the system to use will be avoided. We later
1031 on (re/de)allocate those areas as needed. One problem is
1032 that some of these mappings are put in place automatically
1033 by the dynamic linker: for instance any libraries that
1034 Wine is linked to (like libc, libwine, libpthread etc)
1035 will be mapped into memory before Wine even gets
1036 control. In order to solve that, Wine overrides the
1037 default ELF initialization sequence at a low level and
1038 reserves the needed areas by using direct syscalls into
1039 the kernel (ie without linking against any other code to
1040 do it) before restarting the standard initialization and
1041 letting the dynamic linker continue. This is referred to
1042 as the preloader and is found in loader/preloader.c.
1043 </para>
1045 <para>
1046 Once the usual ELF boot sequence has been completed, some
1047 native libraries may well have been mapped above the 3gig
1048 limit: however, this doesn't matter as 3G is a Windows
1049 limit, not a Linux limit. We still have to prevent the
1050 system from allocating anything else above there (like the
1051 heap or other DLLs) though so Wine performs a binary
1052 search over the upper gig of address space in order to
1053 iteratively fill in the holes with MAP_NORESERVE mappings
1054 so the address space is allocated but the memory to
1055 actually back it is not. This code can be found in libs/wine/mmap.c:reserve_area.
1056 </para>
1057 </sect3>
1058 </sect2>
1060 <sect2>
1061 <title>Processes</title>
1062 <para>
1063 Let's take a closer look at the way Wine loads and run
1064 processes in memory.
1065 </para>
1066 <sect3>
1067 <title>Starting a process from command line</title>
1068 <para>
1069 When starting a Wine process from command line (we'll get
1070 later on to the differences between NE, PE and Winelib
1071 executables), there are a couple of things Wine need to do
1072 first. A first executable is run to check the threading
1073 model of the underlying OS (see <xref linkend="threading">
1074 for the details) and will start the real Wine loader
1075 corresponding to the choosen threading model.
1076 </para>
1077 <para>
1078 Then Wine graps a few elements from the Unix world: the
1079 environment, the program arguments. Then the
1080 <filename>ntdll.dll.so</filename> is loaded into memory
1081 using the standard shared library dynamic loader. When
1082 loaded, NTDLL will mainly first create a decent Windows
1083 environment:
1084 <itemizedlist>
1085 <listitem>
1086 <para>create a PEB and a TEB</para>
1087 </listitem>
1088 <listitem>
1089 <para>
1090 set up the connection to the Wine server - and
1091 eventually launching the Wine server if none runs
1092 </para>
1093 </listitem>
1094 <listitem>
1095 <para>create the process heap</para>
1096 </listitem>
1097 </itemizedlist>
1098 </para>
1099 <para>
1100 Then <filename>Kernel32</filename> is loaded (but now
1101 using the Windows dynamic loading capabilities) and a Wine
1102 specific entry point is called
1103 <function>__wine_kernel_init</function>. This function
1104 will actually handle all the logic of the process loading
1105 and execution, and will never return from it's call.
1106 </para>
1107 <para>
1108 <function>__wine_kernel_init</function> will undergo the
1109 following tasks:
1110 <itemizedlist>
1111 <listitem>
1112 <para>
1113 initialization of program arguments from Unix
1114 program arguments
1115 </para>
1116 </listitem>
1117 <listitem>
1118 <para>
1119 lookup of executable in the file system
1120 </para>
1121 </listitem>
1122 <listitem>
1123 <para>
1124 If the file is not found, then an error is printed
1125 and the Wine loader stops.
1126 </para>
1127 </listitem>
1128 <listitem>
1129 <para>
1130 We'll cover the non-PE file type later on, so assume
1131 for now it's a PE file. The PE module is loaded in
1132 memory using the Windows shared library
1133 mechanism. Note that the dependencies on the module
1134 are not resolved at this point.
1135 </para>
1136 </listitem>
1137 <listitem>
1138 <para>
1139 A new stack is created, which size is given in the
1140 PE header, and this stack is made the one of the
1141 running thread (which is still the only one in the
1142 process). The stack used at startup will no longer
1143 be used.
1144 </para>
1145 </listitem>
1146 <listitem>
1147 <para>
1148 Which this new stack,
1149 <function>ntdll.LdrInitializeThunk</function> is
1150 called which performs the remaining initialization
1151 parts, including resolving all the DLL imports on
1152 the PE module, and doing the init of the TLS slots.
1153 </para>
1154 </listitem>
1155 <listitem>
1156 <para>
1157 Control can now be passed to the
1158 <function>EntryPoint</function> of the PE module,
1159 which will let the executable run.
1160 </para>
1161 </listitem>
1162 </itemizedlist>
1163 </para>
1164 </sect3>
1165 <sect3>
1166 <title>Creating a child process from a running process</title>
1167 <para>
1168 The steps used are closely link to what is done in the
1169 previous case.
1170 </para>
1171 <para>
1172 There are however a few points to look at a bit more
1173 closely. The inner implementation creates the child
1174 process using the <function>fork()</function> and
1175 <function>exec()</function> calls. This means that we
1176 don't need to check again for the threading model, we can
1177 use what the parent (or the grand-parent process...)
1178 started from command line has found.
1179 </para>
1180 <para>
1181 The Win32 process creation allows to pass a lot of
1182 information between the parent and the child. This
1183 includes object handles, windows title, console
1184 parameters, environment strings... Wine makes use of both
1185 the standard Unix inheritance mechanisms (for environment
1186 for example) and the Wine server (to pass from parent to
1187 child a chunk of data containing the relevant information).
1188 </para>
1189 <para>
1190 The previously described loading mechanism will check in
1191 the Wine server if such a chunk exists, and, if so, will
1192 perform the relevant initialization.
1193 </para>
1194 <para>
1195 Some further synchronization is also put in place: a
1196 parent will wait until the child has started, or has
1197 failed. The Wine server is also used to perform those
1198 tasks.
1199 </para>
1200 </sect3>
1201 <sect3>
1202 <title>Starting a Winelib process</title>
1203 <para>
1204 Before going into the gory details, let's first go back to
1205 what a Winelib application is. It can be either a regular
1206 Unix executable, or a more specific Wine beast. This later
1207 form in fact creates two files for a given executable (say
1208 <filename>foo.exe</filename>). The first one, named
1209 <filename>foo</filename> will be a symbolic link to the
1210 Wine loader (<filename>wine</filename>). The second one,
1211 named <filename>foo.exe.so</filename>, is the equivalent
1212 of the <filename>.dll.so</filename> files we've already
1213 described for DLLs. As in Windows, an executable is, among
1214 other things, a module with its import and export
1215 information, as any DLL, it makes sense Wine uses the same
1216 mechanisms for loading native executables and DLLs.
1217 </para>
1218 <para>
1219 When starting a Winelib application from the command line
1220 (say with <command>foo arg1 arg2</command>), the Unix
1221 shell will execute <command>foo</command> as a Unix
1222 executable. Since this is in fact the Wine loader, Wine
1223 will fire up. However, will notice that it hasn't been
1224 started as <command>wine</command> but as
1225 <command>foo</command>, and hence, will try to load (using
1226 Unix shared library mechanism) the second file
1227 <filename>foo.exe.so</filename>. Wine will recognize a 32
1228 bit module (with its descriptor) embedded in the shared
1229 library, and once the shared library loaded, it will
1230 proceed the same path as when loading a standard native PE
1231 executable.
1232 </para>
1233 <para>
1234 Wine needs to implement this second form of executable in
1235 order to maintain the order of initialization of some
1236 elements in the executable. One particular issue is when
1237 dealing with global C++ objects. In standard Unix
1238 executable, the call of the constructor to such objects is
1239 stored in the specific section of the executable
1240 (<function>.init</function> not to name it). All
1241 constructors in this section are called before the
1242 <function>main</function> function is called. Creating a
1243 Wine executable using the first form mentionned above will
1244 let those constructors being called before Wine gets a
1245 chance to initialize itself. So, any constructor using a
1246 Windows API will fail, because Wine infrastructure isn't
1247 in place. The use of the second form for Winelib
1248 executables ensures that we do the initialization using
1249 the following steps:
1250 <itemizedlist>
1251 <listitem>
1252 <para>
1253 initialize the Wine infrastructure
1254 </para>
1255 </listitem>
1256 <listitem>
1257 <para>
1258 load the executable into memory
1259 </para>
1260 </listitem>
1261 <listitem>
1262 <para>
1263 handle the import sections for the executable
1264 </para>
1265 </listitem>
1266 <listitem>
1267 <para>
1268 call the global object constructors (if any). They
1269 now can properly call the Windows APIs
1270 </para>
1271 </listitem>
1272 <listitem>
1273 <para>
1274 call the executable entry point
1275 </para>
1276 </listitem>
1277 </itemizedlist>
1278 </para>
1279 <para>
1280 The attentive reader would have noted that the resolution
1281 of imports for the executable is done, as for a DLL, when
1282 the executable/DLL descriptor is registered. However, this
1283 is done also by adding a specific constructor in the
1284 <function>.init</function> section. For the above describe
1285 scheme to function properly, this constructor must be the
1286 first constructor to be called, before all the other
1287 constructors, generated by the executable itself. The Wine
1288 build chain takes care of that, and also generating the
1289 executable/DLL descriptor for the Winelib executable.
1290 </para>
1291 </sect3>
1292 <sect3>
1293 <title>Starting a NE (Win16) process</title>
1294 <para>
1295 When Wine is requested to run a NE (Win 16 process), it
1296 will in fact hand over the execution of it to a specific
1297 executable <filename>winevdm</filename>. VDM stands for
1298 Virtual DOS Machine. This <filename>winevdm</filename>
1299 will in fact set up the correct 16 bit environment to run
1300 the executable. Any new 16 bit process created by this
1301 executable (or its children) will run into the same
1302 <filename>winevdm</filename> instance. Among one instance,
1303 several functionalities will be provided to those 16 bit
1304 processes, including the cooperative multitasking, sharing
1305 the same address space, managing the selectors for the 16
1306 bit segments needed for code, data and stack.
1307 </para>
1308 <para>
1309 Note that several <filename>winevdm</filename> instances
1310 can run in the same Wine session, but the functionalities
1311 described above are only shared among a given instance,
1312 not among all the instances. <filename>winevdm</filename>
1313 is built as Winelib application, and hence has access to
1314 any facility a 32 bit application has.
1315 </para>
1316 <para>
1317 The behaviour we just described also applies to DOS
1318 executables, which are handled the same way by
1319 <filename>winevdm</filename>.
1320 </para>
1321 </sect3>
1322 </sect2>
1323 <sect2>
1324 <title>Wine drivers</title>
1325 <para>
1326 Wine will not allow running native Windows drivers under
1327 Unix. This comes mainly because (look at the generic
1328 architecture schemas) Wine doesn't implement the kernel
1329 features of Windows (kernel here really means the kernel,
1330 not the KERNEL32 DLL), but rather sets up a proxy layer on
1331 top of the Unix kernel to provide the NTDLL and KERNEL32
1332 features. This means that Wine doesn't provide the inner
1333 infrastructure to run native drivers, either from the Win9x
1334 family or from the NT family.
1335 </para>
1336 <para>
1337 In other words, Wine will only be able to provide access to
1338 a specific device, if and only if, 1/ this device is
1339 supported in Unix (there is Unix-driver to talk to it), 2/
1340 Wine has implemented the proxy code to make the glue between
1341 the API of a Windows driver, and the Unix interface of the
1342 Unix driver.
1343 </para>
1344 <para>
1345 Wine, however, tries to implement in the various DLLs
1346 needing to access devices to do it through the standard
1347 Windows APIs for device drivers in user space. This is for
1348 example the case for the multimedia drivers, where Wine
1349 loads Wine builtin DLLs to talk to the OSS interface, or the
1350 ALSA interface. Those DLLs implement the same interface as
1351 any user space audio driver in Windows.
1352 </para>
1353 </sect2>
1354 </sect1>
1355 </chapter>
1358 <!-- Keep this comment at the end of the file
1359 Local variables:
1360 mode: sgml
1361 sgml-parent-document:("wine-devel.sgml" "set" "book" "part" "chapter" "")
1362 End: