Get rid of the ThunkData stubs, these are not functions.
[wine/multimedia.git] / documentation / winedev-kernel.sgml
blob9b4648ba9306964d525b43f1decc08a0740c2c75
1 <chapter>
2 <title>Kernel modules</title>
3 <para>
4 This section cover the kernel modules. As already stated, Wine
5 implements the NT architecture, hence provides NTDLL for the
6 core kernel functions, and KERNEL32, which is the
7 implementation of the basis of the Win32 subsystem, on top of
8 NTDLL.
9 </para>
10 <sect1 id="ntdll">
11 <title>NTDLL</title>
12 <para>
13 NTDLL provides most of the services you'd expect from a
14 kernel.
15 </para>
16 <para>
17 Process and thread management are part of them (even if
18 process management is still mainly done in KERNEL32, unlike
19 NT). A Windows process runs as a Unix process, and a Windows
20 thread runs as a Unix thread.
21 </para>
22 <para>
23 Wine also provide fibers (which is the Windows name of
24 co-routines).
25 </para>
26 <para>
27 Most of the Windows memory handling (Heap, Global and Local
28 functions, virtual memory...) are easily mapped upon their
29 Unix equivalents. Note the NTDLL doesn't know about 16 bit
30 memory, which is only handled in KERNEL32/KRNL386.EXE (and
31 also the DOS routines).
32 </para>
34 <sect2>
35 <title>File management</title>
36 <para>
37 Wine uses some configuration in order to map Windows
38 filenames (either defined with drive letters, or as UNC
39 names) to the unix filenames. Wine also uses some
40 incantation so that most of file related APIs can also
41 take full unix names. This is handy when passing filenames
42 on the command line.
43 </para>
44 <para>
45 File handles can be waitable objects, as Windows define
46 them.
47 </para>
48 <para>
49 Asynchronous I/O is implemented on file handles by
50 queueing pseudo APC. They are not real APC in the sense
51 that they have the same priority as the threads in the
52 considered process (while APCs on NT have normally a
53 higher priority). These APCs get called when invoking
54 Wine server (which should lead to correct behavior when the
55 program ends up waiting on some object - waiting always
56 implies calling Wine server).
57 </para>
58 <para>
59 FIXME: this should be enhanced and updated to latest work
60 on FS.
61 </para>
62 </sect2>
64 <sect2>
65 <title>Synchronization</title>
66 <para>
67 Most of the synchronization (between threads or processes)
68 is done in Wine server, which handles both the waiting
69 operation (on a single object or a set of objects) and the
70 signaling of objects.
71 </para>
72 </sect2>
74 <sect2>
75 <title>Module (DLL) loading</title>
76 <para>
77 Wine is able to load any NE and PE module. In all cases,
78 the module's binary code is directly executed by the
79 processor.
80 </para>
81 </sect2>
83 <sect2>
84 <title>Device management</title>
85 <para>
86 Wine allows usage a wide variety of devices:
87 <itemizedlist>
88 <listitem>
89 <para>
90 Communication ports are mapped to Unix
91 communication ports (if they have sufficient
92 permissions).
93 </para>
94 </listitem>
95 <listitem>
96 <para>
97 Parallel ports are mapped to Unix parallel ports (if
98 they have sufficient permissions).
99 </para>
100 </listitem>
101 <listitem>
102 <para>
103 CDROM: the Windows device I/O control calls are
104 mapped onto Unix <function>ioctl()</function>.
105 </para>
106 </listitem>
107 <listitem>
108 <para>
109 Some Win9x VxDs are supported, by rewriting some of
110 their internal behavior. But this support is
111 limited. Portable programs to Windows NT shouldn't
112 need them.
113 </para>
114 <para>
115 Wine will not support native VxD.
116 </para>
117 </listitem>
118 </itemizedlist>
119 </para>
120 </sect2>
121 <sect2 id="threading">
122 <title>Multi-threading in Wine</title>
124 <para>
125 This section will assume you understand the basics of
126 multithreading. If not there are plenty of good tutorials
127 available on the net to get you started.
128 </para>
130 <para>
131 Threading in Wine is somewhat complex due to several
132 factors. The first is the advanced level of multithreading
133 support provided by Windows - there are far more threading
134 related constructs available in Win32 than the Linux
135 equivalent (pthreads). The second is the need to be able to
136 map Win32 threads to native Linux threads which provides us
137 with benefits like having the kernel schedule them without
138 our intervention. While it's possible to implement threading
139 entirely without kernel support, doing so is not desirable
140 on most platforms that Wine runs on.
141 </para>
143 <sect3>
144 <title> Threading support in Win32 </title>
146 <para>
147 Win32 is an unusually thread friendly API. Not only is it
148 entirely thread safe, but it provides many different
149 facilities for working with threads. These range from the
150 basics such as starting and stopping threads, to the
151 extremely complex such as injecting threads into other
152 processes and COM inter-thread marshalling.
153 </para>
155 <para>
156 One of the primary challenges of writing Wine code
157 therefore is ensuring that all our DLLs are thread safe,
158 free of race conditions and so on. This isn't simple -
159 don't be afraid to ask if you aren't sure whether a piece
160 of code is thread safe or not!
161 </para>
163 <para>
164 Win32 provides many different ways you can make your code
165 thread safe however the most common are <emphasis>critical
166 section</emphasis> and the <emphasis>interlocked
167 functions</emphasis>. Critical sections are a type of
168 mutex designed to protect a geographic area of code. If
169 you don't want multiple threads running in a piece of code
170 at once, you can protect them with calls to
171 <function>EnterCriticalSection</function> and
172 <function>LeaveCriticalSection</function>. The first call
173 to <function>EnterCriticalSection</function> by a thread
174 will lock the section and continue without stopping. If
175 another thread calls it then it will block until the
176 original thread calls
177 <function>LeaveCriticalSection</function> again.
178 </para>
180 <para>
181 It is therefore vitally important that if you use critical
182 sections to make some code thread-safe, that you check
183 every possible codepath out of the code to ensure that any
184 held sections are left. Code like this:
185 </para>
187 <programlisting>
188 if (res != ERROR_SUCCESS) return res;
189 </programlisting>
191 <para>
192 is extremely suspect in a function that also contains a
193 call to <function>EnterCriticalSection</function>. Be
194 careful.
195 </para>
197 <para>
198 If a thread blocks while waiting for another thread to
199 leave a critical section, you will see an error from the
200 <function>RtlpWaitForCriticalSection</function> function,
201 along with a note of which thread is holding the
202 lock. This only appears after a certain timeout, normally
203 a few seconds. It's possible the thread holding the lock
204 is just being really slow which is why Wine won't
205 terminate the app like a non-checked build of Windows
206 would, but the most common cause is that for some reason a
207 thread forgot to call
208 <function>LeaveCriticalSection</function>, or died while
209 holding the lock (perhaps because it was in turn waiting
210 for another lock). This doesn't just happen in Wine code:
211 a deadlock while waiting for a critical section could be
212 due to a bug in the app triggered by a slight difference
213 in the emulation.
214 </para>
216 <para>
217 Another popular mechanism available is the use of
218 functions like <function>InterlockedIncrement</function>
219 and <function>InterlockedExchange</function>. These make
220 use of native CPU abilities to execute a single
221 instruction while ensuring any other processors on the
222 system cannot access memory, and allow you to do common
223 operations like add/remove/check a variable in thread-safe
224 code without holding a mutex. These are useful for
225 reference counting especially in free-threaded (thread
226 safe) COM objects.
227 </para>
229 <para>
230 Finally, the usage of TLS slots are also popular. TLS
231 stands for thread-local storage, and is a set of slots
232 scoped local to a thread which you can store pointers
233 in. Look on MSDN for the <function>TlsAlloc</function>
234 function to learn more about the Win32 implementation of
235 this. Essentially, the contents of a given slot will be
236 different in each thread, so you can use this to store
237 data that is only meaningful in the context of a single
238 thread. On recent versions of Linux the __thread keyword
239 provides a convenient interface to this functionality - a
240 more portable API is exposed in the pthread
241 library. However, these facilities is not used by Wine,
242 rather, we implement Win32 TLS entirely ourselves.
243 </para>
244 </sect3>
246 <sect3>
247 <title> SysLevels </title>
249 <para>
250 SysLevels are an undocumented Windows-internal
251 thread-safety system. They are basically critical sections
252 which must be taken in a particular order. The mechanism
253 is generic but there are always three syslevels: level 1
254 is the Win16 mutex, level 2 is the USER mutex and level 3
255 is the GDI mutex.
256 </para>
258 <para>
259 When entering a syslevel, the code (in
260 <filename>dlls/kernel/syslevel.c</filename>) will check
261 that a higher syslevel is not already held and produce an
262 error if so. This is because it's not legal to enter level
263 2 while holding level 3 - first, you must leave level 3.
264 </para>
266 <para>
267 Throughout the code you may see calls to
268 <function>_ConfirmSysLevel()</function> and
269 <function>_CheckNotSysLevel()</function>. These functions
270 are essentially assertions about the syslevel states and
271 can be used to check that the rules have not been
272 accidentally violated. In particular,
273 <function>_CheckNotSysLevel()</function> will break
274 (probably into the debugger) if the check fails. If this
275 happens the solution is to get a backtrace and find out,
276 by reading the source of the wine functions called along
277 the way, how Wine got into the invalid state.
278 </para>
280 </sect3>
282 <sect3>
283 <title> POSIX threading vs kernel threading </title>
285 <para>
286 Wine runs in one of two modes: either pthreads (posix
287 threading) or kthreads (kernel threading). This section
288 explains the differences between them. The one that is
289 used is automatically selected on startup by a small test
290 program which then execs the correct binary, either
291 wine-kthread or wine-pthread. On NPTL-enabled systems
292 pthreads will be used, and on older non-NPTL systems
293 kthreads is selected.
294 </para>
296 <para>
297 Let's start with a bit of history. Back in the dark ages
298 when Wines threading support was first implemented a
299 problem was faced - Windows had much more capable
300 threading APIs than Linux did. This presented a problem -
301 Wine works either by reimplementing an API entirely or by
302 mapping it onto the underlying systems equivalent. How
303 could Win32 threading be implemented using a library which
304 did not have all the neeed features? The answer, of
305 course, was that it couldn't be.
306 </para>
308 <para>
309 On Linux the pthreads interface is used to start, stop and
310 control threads. The pthreads library in turn is based on
311 top of so-called "kernel threads" which are created using
312 the <function>clone(2)</function> syscall. Pthreads
313 provides a nicer (more portable) interface to this
314 functionality and also provides APIs for controlling
315 mutexes. There is a <ulink
316 url="http://www.llnl.gov/computing/tutorials/pthreads/">
317 good tutorial on pthreads </ulink> available if you want
318 to learn more.
319 </para>
321 <para>
322 As pthreads did not provide the necessary semantics to
323 implement Win32 threading, the decision was made to
324 implement Win32 threading on top of the underlying kernel
325 threads by using syscalls like <function>clone</function>
326 directly. This provided maximum flexibility and allowed a
327 correct implementation but caused some bad side
328 effects. Most notably, all the userland Linux APIs assumed
329 that the user was utilising the pthreads library. Some
330 only enabled thread safety when they detected that
331 pthreads was in use - this is true of glibc, for
332 instance. Worse, pthreads and pure kernel threads had
333 strange interactions when run in the same process yet some
334 libraries used by Wine used pthreads internally. Throw in
335 source code porting using WineLib - where you have both
336 UNIX and Win32 code in the same process - and chaos was
337 the result.
338 </para>
340 <para>
341 The solution was simple yet ingenius: Wine would provide
342 its own implementation of the pthread library
343 <emphasis>inside</emphasis> its own binary. Due to the
344 semantics of ELF symbol scoping, this would cause Wines
345 own implementations to override any implementation loaded
346 later on (like the real libpthread.so). Therefore, any
347 calls to the pthread APIs in external libraries would be
348 linked to Wines instead of the systems pthreads library,
349 and Wine implemented pthreads by using the standard
350 Windows threading APIs it in turn implemented itself.
351 </para>
353 <para>
354 As a result, libraries that only became thread-safe in the
355 presence of a loaded pthreads implementation would now do
356 so, and any external code that used pthreads would
357 actually end up creating Win32 threads that Wine was aware
358 of and controlled. This worked quite nicely for a long
359 time, even though it required doing some extremely
360 un-kosher things like overriding internal libc structures
361 and functions. That is, it worked until NPTL was developed
362 at which point the underlying thread implementation on
363 Linux changed dramatically.
364 </para>
366 <para>
367 The fake pthread implementation can be found in
368 <filename>loader/kthread.c</filename>, which is used to
369 produce to wine-kthread binary. In contrast,
370 loader/pthread.c produces the wine-pthread binary which is
371 used on newer NPTL systems.
372 </para>
374 <para>
375 NPTL is a new threading subsystem for Linux that hugely
376 improves its performance and flexibility. By allowing
377 threads to become much more scalable and adding new
378 pthread APIs, NPTL made Linux competitive with Windows in
379 the multi-threaded world. Unfortunately it also broke many
380 assumptions made by Wine (as well as other applications
381 such as the Sun JVM and RealPlayer) in the process.
382 </para>
384 <para>
385 There was, however, some good news. NPTL made Linux
386 threading powerful enough that Win32 threads could now be
387 implemented on top of pthreads like any other normal
388 application. There would no longer be problems with mixing
389 win32-kthreads and pthreads created by external libraries,
390 and no need to override glibc internals. As you can see
391 from the relative sizes of the
392 <filename>loader/kthread.c</filename> and
393 <filename>loader/pthread.c</filename> files, the
394 difference in code complexity is considerable. NPTL also
395 made several other semantic changes to things such as
396 signal delivery so changes were required in many different
397 places in Wine.
398 </para>
400 <para>
401 On non-Linux systems the threading interface is typically
402 not powerful enough to replicate the semantics Win32
403 applications expect and so kthreads with the pthread
404 overrides are used.
405 </para>
406 </sect3>
408 <sect3>
409 <title> The Win32 thread environment </title>
411 <para>
412 All Win32 code, whether from a native EXE/DLL or in Wine
413 itself, expects certain constructs to be present in its
414 environment. This section explores what those constructs
415 are and how Wine sets them up. The lack of this
416 environment is one thing that makes it hard to use Wine
417 code directly from standard Linux applications - in order
418 to interact with Win32 code a thread must first be
419 "adopted" by Wine.
420 </para>
422 <para>
423 The first thing Win32 code requires is the
424 <emphasis>TEB</emphasis> or "Thread Environment
425 Block". This is an internal (undocumented) Windows
426 structure associated with every thread which stores a
427 variety of things such as TLS slots, a pointer to the
428 threads message queue, the last error code and so on. You
429 can see the definition of the TEB in
430 <filename>include/thread.h</filename>, or at least what we
431 know of it so far. Being internal and subject to change,
432 the layout of the TEB has had to be reverse engineered
433 from scratch.
434 </para>
436 <para>
437 A pointer to the TEB is stored in the %fs register and can
438 be accessed using <function>NtCurrentTeb()</function> from
439 within Wine code. %fs actually stores a selector, and
440 setting it therefore requires modifying the processes
441 local descriptor table (LDT) - the code to do this is in
442 <filename>lib/wine/ldt.c</filename>.
443 </para>
445 <para>
446 The TEB is required by nearly all Win32 code run in the
447 Wine environment, as any wineserver RPC will use it, which
448 in turn implies that any code which could possibly block
449 (for instance by using a critical section) needs it. The
450 TEB also holds the SEH exception handler chain as the
451 first element, so if when disassembling you see code like
452 this:
453 </para>
455 <programlisting> movl %esp, %fs:0 </programlisting>
457 <para>
458 ... then you are seeing the program set up an SEH handler
459 frame. All threads must have at least one SEH entry, which
460 normally points to the backstop handler which is
461 ultimately responsible for popping up the all-too-familiar
462 "This program has performed an illegal operation and will
463 be terminated" message. On Wine we just drop straight into
464 the debugger. A full description of SEH is out of the
465 scope of this section, however there are some good
466 articles in MSJ if you are interested.
467 </para>
469 <para>
470 All Win32-aware threads must have a wineserver
471 connection. Many different APIs require the ability to
472 communicate with the wineserver. In turn, the wineserver
473 must be aware of Win32 threads in order to be able to
474 accurately report information to other parts of the program
475 and do things like route inter-thread messages, dispatch
476 APCs (asynchronous procedure calls) and so on. Therefore a
477 part of thread initialization is initializing the thread
478 serverside. The result is not only correct information in
479 the server, but a set of file descriptors the thread can use
480 to communicate with the server - the request fd, reply fd
481 and wait fd (used for blocking).
482 </para>
483 </sect3>
484 </sect2>
485 </sect1>
487 <sect1>
488 <title>KERNEL Module</title>
490 <para>
491 FIXME: Needs some content...
492 </para>
493 <sect2 id="consoles">
494 <title>Consoles in Wine</title>
495 <para>
496 As described in the Wine User Guide's CUI section, Wine
497 manipulates three kinds of "consoles" in order to support
498 properly the Win32 CUI API.
499 </para>
500 <para>
501 The following table describes the main implementation
502 differences between the three approaches.
503 <table>
504 <title>Function consoles implementation comparison</title>
505 <tgroup cols="4" align="left">
506 <thead>
507 <row>
508 <entry>Function</entry>
509 <entry>Bare streams</entry>
510 <entry>Wineconsole &amp; user backend</entry>
511 <entry>Wineconsole &amp; curses backend</entry>
512 </row>
513 </thead>
514 <tbody>
515 <row>
516 <entry>
517 Console as a Win32 Object (and associated
518 handles)
519 </entry>
520 <entry>
521 No specific Win32 object is used in this
522 case. The handles manipulated for the standard
523 Win32 streams are in fact "bare handles" to
524 their corresponding Unix streams. The mode
525 manipulation functions
526 (<function>GetConsoleMode</function> /
527 <function>SetConsoleMode</function>) are not
528 supported.
529 </entry>
530 <entry>
531 Implemented in server, and a specific Winelib
532 program (wineconsole) is in charge of the
533 rendering and user input. The mode manipulation
534 functions behave as expected.
535 </entry>
536 <entry>
537 Implemented in server, and a specific Winelib
538 program (wineconsole) is in charge of the
539 rendering and user input. The mode manipulation
540 functions behave as expected.
541 </entry>
542 </row>
543 <row>
544 <entry>
545 Inheritance (including handling in
546 <function>CreateProcess</function> of
547 <constant>CREATE_DETACHED</constant>,
548 <constant>CREATE_NEW_CONSOLE</constant> flags).
549 </entry>
550 <entry>
551 Not supported. Every process child of a process
552 will inherit the Unix streams, so will also
553 inherit the Win32 standard streams.
554 </entry>
555 <entry>
556 Fully supported (each new console creation will
557 be handled by the creation of a new USER32
558 window)
559 </entry>
560 <entry>
561 Fully supported, except for the creation of a
562 new console, which will be rendered on the same
563 Unix terminal as the previous one, leading to
564 unpredictable results.
565 </entry>
566 </row>
567 <row>
568 <entry>
569 <function>ReadFile</function> /
570 <function>WriteFile</function>
571 operations
572 </entry>
573 <entry>Fully supported</entry>
574 <entry>Fully supported</entry>
575 <entry>Fully supported</entry>
576 </row>
577 <row>
578 <entry>
579 Screen-buffer manipulation (creation, deletion,
580 resizing...)
581 </entry>
582 <entry>Not supported</entry>
583 <entry>Fully supported</entry>
584 <entry>
585 Partly supported (this won't work too well as we
586 don't control (so far) the size of underlying
587 Unix terminal
588 </entry>
589 </row>
590 <row>
591 <entry>
592 APIs for reading/writing screen-buffer content,
593 cursor position
594 </entry>
595 <entry>Not supported</entry>
596 <entry>Fully supported</entry>
597 <entry>Fully supported</entry>
598 </row>
599 <row>
600 <entry>
601 APIs for manipulating the rendering window size
602 </entry>
603 <entry>Not supported</entry>
604 <entry>Fully supported</entry>
605 <entry>
606 Partly supported (this won't work too well as we
607 don't control (so far) the size of underlying
608 Unix terminal
609 </entry>
610 </row>
611 <row>
612 <entry>
613 Signaling (in particular, Ctrl-C handling)
614 </entry>
615 <entry>
616 Nothing is done, which means that Ctrl-C will
617 generate (as usual) a
618 <constant>SIGINT</constant> which will terminate
619 the program.
620 </entry>
621 <entry>
622 Partly supported (Ctrl-C behaves as expected,
623 however the other Win32 CUI signaling isn't
624 properly implemented).
625 </entry>
626 <entry>
627 Partly supported (Ctrl-C behaves as expected,
628 however the other Win32 CUI signaling isn't
629 properly implemented).
630 </entry>
631 </row>
632 </tbody>
633 </tgroup>
634 </table>
635 </para>
637 <para>
638 The Win32 objects behind a console can be created in
639 several occasions:
640 <itemizedlist>
641 <listitem>
642 <para>
643 When the program is started from wineconsole, a new
644 console object is created and will be used
645 (inherited) by the process launched from
646 wineconsole.
647 </para>
648 </listitem>
649 <listitem>
650 <para>
651 When a program, which isn't attached to a console,
652 calls <function>AllocConsole</function>, Wine then
653 launches wineconsole, and attaches the current
654 program to this console. In this mode, the USER32
655 mode is always selected as Wine cannot tell the
656 current state of the Unix console.
657 </para>
658 </listitem>
659 </itemizedlist>
660 </para>
661 <para>
662 Please also note, that starting a child process with the
663 <constant>CREATE_NEW_CONSOLE</constant> flag, will end-up
664 calling <function>AllocConsole</function> in the child
665 process, hence creating a wineconsole with the USER32
666 backend.
667 </para>
668 </sect2>
669 </sect1>
671 <sect1 id="initialization">
673 <title> The Wine initialization process </title>
675 <para>
676 Wine has a rather complex startup procedure, so unlike many
677 programs the best place to begin exploring the code-base is
678 <emphasis>not</emphasis> in fact at the
679 <function>main()</function> function but instead at some of the
680 more straightforward DLLs that exist on the periphery such as
681 MSI, the widget library (in USER and COMCTL32) etc. The purpose
682 of this section is to document and explain how Wine starts up
683 from the moment the user runs "wine myprogram.exe" to the point
684 at which myprogram gets control.
685 </para>
687 <sect2>
688 <title> First Steps </title>
690 <para>
691 The actual wine binary that the user runs does not do very much, in fact it is only
692 responsible for checking the threading model in use (NPTL vs LinuxThreads) and then invoking
693 a new binary which performs the next stage in the startup sequence. See the threading chapter
694 for more information on this check and why it's necessary. You can find this code in
695 <filename>loader/glibc.c</filename>. The result of this check is an exec of either
696 wine-pthread or wine-kthread, potentially (on Linux) via
697 the <emphasis>preloader</emphasis>. We need to use separate binaries here because overriding
698 the native pthreads library requires us to exploit a property of ELF symbol fixup semantics:
699 it's not possible to do this without starting a new process.
700 </para>
702 <para>
703 The Wine preloader is found in <filename>loader/preloader.c</filename>, and is required in
704 order to impose a Win32 style address space layout upon the newly created Win32 process. The
705 details of what this does is covered in the address space layout chapter. The preloader is a
706 statically linked ELF binary which is passed the name of the actual Wine binary to run (either
707 wine-kthread or wine-pthread) along with the arguments the user passed in from the command
708 line. The preloader is an unusual program: it does not have a main() function. In standard ELF
709 applications, the entry point is actually at a symbol named _start: this is provided by the
710 standard gcc infrastructure and normally jumps to <function>__libc_start_main</function> which
711 initializes glibc before passing control to the main function as defined by the programmer.
712 </para>
714 <para>
715 The preloader takes control direct from the entry point for a few reasons. Firstly, it is
716 required that glibc is not initialized twice: the result of such behaviour is undefined and
717 subject to change without notice. Secondly, it's possible that as part of initializing glibc,
718 the address space layout could be changed - for instance, any call to malloc will initialize a
719 heap arena which modifies the VM mappings. Finally, glibc does not return to _start at any
720 point, so by reusing it we avoid the need to recreate the ELF bootstrap stack (env, argv,
721 auxiliary array etc).
722 </para>
724 <para>
725 The preloader is responsible for two things: protecting important regions of the address
726 space so the dynamic linker does not map shared libraries into them, and once that is done
727 loading the real Wine binary off disk, linking it and starting it up. Normally all this is
728 done automatically by glibc and the kernel but as we intercepted this process by using a
729 static binary it's up to us to restart the process. The bulk of the code in the preloader is
730 about loading wine-[pk]thread and ld-linux.so.2 off disk, linking them together, then
731 starting the dynamic linking process.
732 </para>
734 <para>
735 One of the last things the preloader does before jumping into the dynamic linker is scan the
736 symbol table of the loaded Wine binary and set the value of a global variable directly: this
737 is a more efficient way of passing information to the main Wine program than flattening the
738 data structures into an environment variable or command line parameter then unpacking it on
739 the other side, but it achieves pretty much the same thing. The global variable set points to
740 the preload descriptor table, which contains the VMA regions protected by the preloader. This
741 allows Wine to unmap them once the dynamic linker has been run, so leaving gaps we can
742 initialize properly later on.
743 </para>
745 </sect2>
747 <sect2>
748 <title> Starting the emulator </title>
750 <para>
751 The process of starting up the emulator itself is mostly one of chaining through various
752 initializer functions defined in the core libraries and DLLs: libwine, then NTDLL, then kernel32.
753 </para>
755 <para>
756 Both the wine-pthread and wine-kthread binaries share a common <function>main</function>
757 function, defined in <filename>loader/main.c</filename>, so no matter which binary is selected
758 after the preloader has run we start here. This passes the information provided by the
759 preloader into libwine and then calls wine_init, defined
760 in <filename>libs/wine/loader.c</filename>. This is where the emulation really starts:
761 <function>wine_init</function> can, with the correct preparation,
762 be called from programs other than the wine loader itself.
763 </para>
765 <para>
766 <function>wine_init</function> does some very basic setup tasks such as initializing the
767 debugging infrastructure, yet more address space manipulation (see the information on the
768 4G/4G VM split in the address space chapter), before loading NTDLL - the core of both Wine and
769 the Windows NT series - and jumping to the <function>__wine_process_init</function> function defined
770 in <filename>dlls/ntdll/loader.c</filename>
771 </para>
773 <para>
774 This function is responsible for initializing the primary Win32 environment. In thread_init(),
775 it sets up the TEB, the wineserver connection for the main thread and the process heap. See
776 the threading chapter for more information on this.
777 </para>
779 <para>
780 Finally, it loads and jumps to <function>__wine_kernel_init</function> in kernel32.dll: this
781 is defined in <filename>dlls/kernel32/process.c</filename>. This is where the bulk of the work
782 is done. The kernel32 initialization code retrieves the startup info for the process from the
783 server, initializes the registry, sets up the drive mapping system and locale data, then
784 begins loading the requested application itself. Each process has a STARTUPINFO block that can
785 be passed into <function>CreateProcess</function> specifying various things like how the first
786 window should be displayed: this is sent to the new process via the wineserver.
787 </para>
789 <para>
790 After determining the type of file given to Wine by the user (a Win32 EXE file, a Win16 EXE, a
791 Winelib app etc), the program is loaded into memory (which may involve loading and
792 initializing other DLLs, the bulk of Wines startup code), before control reaches the end of
793 <function>__wine_kernel_init</function>. This function ends with the new process stack being
794 initialized, and start_process being called on the new stack. Nearly there!
795 </para>
797 <para>
798 The final element of initializing Wine is starting the newly loaded program
799 itself. <function>start_process</function> sets up the SEH backstop handler, calls
800 <function>LdrInitializeThunk</function> which performs the last part of the process
801 initialization (such as performing relocations and calling the DllMains with PROCESS_ATTACH),
802 grabs the entry point of the executable and then on this line:
803 </para>
805 <programlisting>
806 ExitProcess( entry( peb ) );
807 </programlisting>
809 <para>
810 ... jumps to the entry point of the program. At this point the users program is running and
811 the API provided by Wine is ready to be used. When entry returns,
812 the <function>ExitProcess</function> API will be used to initialize a graceful shutdown.
813 </para>
814 </sect2>
815 </sect1>
817 <sect1 id="seh">
818 <title>Structured Exception Handling</title>
820 <para>
821 Structured Exception Handling (or SEH) is an implementation of
822 exceptions inside the Windows core. It allows code written in
823 different languages to throw exceptions across DLL boundaries, and
824 Windows reports various errors like access violations by throwing
825 them. This section looks at how it works, and how it's implemented
826 in Wine.
827 </para>
829 <sect2>
830 <title> How SEH works </title>
832 <para>
833 SEH is based on embedding EXCEPTION_REGISTRATION_RECORD
834 structures in the stack. Together they form a linked list rooted
835 at offset zero in the TEB (see the threading section if you
836 don't know what this is). A registration record points to a
837 handler function, and when an exception is thrown the handlers
838 are executed in turn. Each handler returns a code, and they can
839 elect to either continue through the handler chain or it can
840 handle the exception and then restart the program. This is
841 referred to as unwinding the stack. After each handler is called
842 it's popped off the chain.
843 </para>
845 <para>
846 Before the system begins unwinding the stack, it runs vectored
847 handlers. This is an extension to SEH available in Windows XP,
848 and allows registered functions to get a first chance to watch
849 or deal with any exceptions thrown in the entire program, from
850 any thread.
851 </para>
853 <para>
854 A thrown exception is represented by an EXCEPTION_RECORD
855 structure. It consists of a code, flags, an address and an
856 arbitrary number of DWORD parameters. Language runtimes can use
857 these parameters to associate language-specific information with
858 the exception.
859 </para>
861 <para>
862 Exceptions can be triggered by many things. They can be thrown
863 explicitly by using the RaiseException API, or they can be
864 triggered by a crash (ie, translated from a signal). They may be
865 used internally by a language runtime to implement
866 language-specific exceptions. They can also be thrown across
867 DCOM connections.
868 </para>
870 <para>
871 Visual C++ has various extensions to SEH which it uses to
872 implement, eg, object destruction on stack unwind as well as the
873 ability to throw arbitrary types. The code for this is in dlls/msvcrt/except.c
874 </para>
876 </sect2>
878 <sect2>
879 <title> Translating signals to exceptions </title>
881 <para>
882 In Windows, compilers are expected to use the system exception
883 interface, and the kernel itself uses the same interface to
884 dynamically insert exceptions into a running program. By contrast
885 on Linux the exception ABI is implemented at the compiler level
886 (inside GCC and the linker) and the kernel tells a thread of
887 exceptional events by sending <emphasis>signals</emphasis>. The
888 language runtime may or may not translate these signals into
889 native exceptions, but whatever happens the kernel does not care.
890 </para>
892 <para>
893 You may think that if an app crashes, it's game over and it
894 really shouldn't matter how Wine handles this. It's what you might
895 intuitively guess, but you'd be wrong. In fact some Windows
896 programs expect to be able to crash themselves and recover later
897 without the user noticing, some contain buggy binary-only
898 components from third parties and use SEH to swallow crashes, and
899 still others execute priviledged (kernel-level) instructions and
900 expect it to work. In fact, at least one set of APIs (the
901 IsBad*Ptr series) can only be implemented by performing an
902 operation that may crash and returning TRUE if it does, and FALSE
903 if it doesn't! So, Wine needs to not only implement the SEH
904 infrastructure but also translate Unix signals into SEH
905 exceptions.
906 </para>
908 <para>
909 The code to translate signals into exceptions is a part of NTDLL,
910 and can be found in dlls/ntdll/signal_i386.c. This file sets up
911 handlers for various signals during Wine startup, and for the ones
912 that indicate exceptional conditions translates them into
913 exceptions. Some signals are used by Wine internally and have
914 nothing to do with SEH.
915 </para>
917 <para>
918 Signal handlers in Wine run on their own stack. Each thread has
919 its own signal stack which resides 4k after the TEB. This is
920 important for a couple of reasons. Firstly, because there's no
921 guarantee that the app thread which triggered the signal has
922 enough stack space for the Wine signal handling code. In
923 Windows, if a thread hits the limits of its stack it triggers a
924 fault on the stack guard page. The language runtime can use this
925 to grow the stack if it wants to.
927 <!-- fixme: is it really the language runtime that does this? i
928 can't find any code in Wine to reallocate the stack on
929 STATUS_GUARD_PAGE_VIOLATION -->
931 However, because a guard page violation is just a regular
932 segfault to the kernel, that would lead to a nested signal
933 handler and that gets messy really quick so we disallow that in
934 Wine. Secondly, setting up the exception to throw requires
935 modifying the stack of the thread which triggered it, which is
936 quite hard to do when you're still running on it.
937 </para>
939 <para>
940 Windows exceptions typically contain more information than the
941 Unix standard APIs provide. For instance, a
942 STATUS_ACCESS_VIOLATION exception (0xC0000005) structure
943 contains the faulting address, whereas a standard Unix SIGSEGV
944 just tells the app that it crashed. Usually this information is
945 passed as an extra parameter to the signal handler, however its
946 location and contents vary between kernels (BSD, Solaris,
947 etc). This data is provided in a SIGCONTEXT structure, and on
948 entry to the signal handler it contains the register state of
949 the CPU before the signal was sent. Modifying it will cause the
950 kernel to adjust the context before restarting the thread.
951 </para>
953 </sect2>
955 </sect1>
957 </chapter>
959 <!-- Keep this comment at the end of the file
960 Local variables:
961 mode: sgml
962 sgml-parent-document:("wine-devel.sgml" "set" "book" "part" "chapter" "")
963 End: