dmusic: Fix IDirectMusic8Impl_CreateMusicBuffer buffer signature.
[wine/multimedia.git] / tools / winebuild / relay.c
blobcc7dd37cb918710adc9481e6cbccb60252fe0165
1 /*
2 * Relay calls helper routines
4 * Copyright 1993 Robert J. Amstadt
5 * Copyright 1995 Martin von Loewis
6 * Copyright 1995, 1996, 1997 Alexandre Julliard
7 * Copyright 1997 Eric Youngdale
8 * Copyright 1999 Ulrich Weigand
10 * This library is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU Lesser General Public
12 * License as published by the Free Software Foundation; either
13 * version 2.1 of the License, or (at your option) any later version.
15 * This library is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * Lesser General Public License for more details.
20 * You should have received a copy of the GNU Lesser General Public
21 * License along with this library; if not, write to the Free Software
22 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
25 #include "config.h"
26 #include "wine/port.h"
28 #include <ctype.h>
29 #include <stdarg.h>
31 #include "build.h"
33 /* offset of the stack pointer relative to %fs:(0) */
34 #define STACKOFFSET 0xc0 /* FIELD_OFFSET(TEB,WOW32Reserved) */
36 /* fix this if the ntdll_thread_regs structure is changed */
37 #define GS_OFFSET 0x1d8 /* FIELD_OFFSET(TEB,SystemReserved2) + FIELD_OFFSET(ntdll_thread_data,gs) */
39 #define DPMI_VIF_OFFSET (0x1fc + 0) /* FIELD_OFFSET(TEB,GdiTebBatch) + FIELD_OFFSET(WINE_VM86_TEB_INFO,dpmi_vif) */
40 #define VM86_PENDING_OFFSET (0x1fc + 4) /* FIELD_OFFSET(TEB,GdiTebBatch) + FIELD_OFFSET(WINE_VM86_TEB_INFO,vm86_pending) */
42 static void function_header( const char *name )
44 output( "\n\t.align %d\n", get_alignment(4) );
45 output( "\t%s\n", func_declaration(name) );
46 output( "%s\n", asm_globl(name) );
50 /*******************************************************************
51 * BuildCallFrom16Core
53 * This routine builds the core routines used in 16->32 thunks:
54 * CallFrom16Word, CallFrom16Long, CallFrom16Register, and CallFrom16Thunk.
56 * These routines are intended to be called via a far call (with 32-bit
57 * operand size) from 16-bit code. The 16-bit code stub must push %bp,
58 * the 32-bit entry point to be called, and the argument conversion
59 * routine to be used (see stack layout below).
61 * The core routine completes the STACK16FRAME on the 16-bit stack and
62 * switches to the 32-bit stack. Then, the argument conversion routine
63 * is called; it gets passed the 32-bit entry point and a pointer to the
64 * 16-bit arguments (on the 16-bit stack) as parameters. (You can either
65 * use conversion routines automatically generated by BuildCallFrom16,
66 * or write your own for special purposes.)
68 * The conversion routine must call the 32-bit entry point, passing it
69 * the converted arguments, and return its return value to the core.
70 * After the conversion routine has returned, the core switches back
71 * to the 16-bit stack, converts the return value to the DX:AX format
72 * (CallFrom16Long), and returns to the 16-bit call stub. All parameters,
73 * including %bp, are popped off the stack.
75 * The 16-bit call stub now returns to the caller, popping the 16-bit
76 * arguments if necessary (pascal calling convention).
78 * In the case of a 'register' function, CallFrom16Register fills a
79 * CONTEXT86 structure with the values all registers had at the point
80 * the first instruction of the 16-bit call stub was about to be
81 * executed. A pointer to this CONTEXT86 is passed as third parameter
82 * to the argument conversion routine, which typically passes it on
83 * to the called 32-bit entry point.
85 * CallFrom16Thunk is a special variant used by the implementation of
86 * the Win95 16->32 thunk functions C16ThkSL and C16ThkSL01 and is
87 * implemented as follows:
88 * On entry, the EBX register is set up to contain a flat pointer to the
89 * 16-bit stack such that EBX+22 points to the first argument.
90 * Then, the entry point is called, while EBP is set up to point
91 * to the return address (on the 32-bit stack).
92 * The called function returns with CX set to the number of bytes
93 * to be popped of the caller's stack.
95 * Stack layout upon entry to the core routine (STACK16FRAME):
96 * ... ...
97 * (sp+24) word first 16-bit arg
98 * (sp+22) word cs
99 * (sp+20) word ip
100 * (sp+18) word bp
101 * (sp+14) long 32-bit entry point (reused for Win16 mutex recursion count)
102 * (sp+12) word ip of actual entry point (necessary for relay debugging)
103 * (sp+8) long relay (argument conversion) function entry point
104 * (sp+4) long cs of 16-bit entry point
105 * (sp) long ip of 16-bit entry point
107 * Added on the stack:
108 * (sp-2) word saved gs
109 * (sp-4) word saved fs
110 * (sp-6) word saved es
111 * (sp-8) word saved ds
112 * (sp-12) long saved ebp
113 * (sp-16) long saved ecx
114 * (sp-20) long saved edx
115 * (sp-24) long saved previous stack
117 static void BuildCallFrom16Core( int reg_func, int thunk )
119 /* Function header */
120 if (thunk) function_header( "__wine_call_from_16_thunk" );
121 else if (reg_func) function_header( "__wine_call_from_16_regs" );
122 else function_header( "__wine_call_from_16" );
124 /* Create STACK16FRAME (except STACK32FRAME link) */
125 output( "\tpushw %%gs\n" );
126 output( "\tpushw %%fs\n" );
127 output( "\tpushw %%es\n" );
128 output( "\tpushw %%ds\n" );
129 output( "\tpushl %%ebp\n" );
130 output( "\tpushl %%ecx\n" );
131 output( "\tpushl %%edx\n" );
133 /* Save original EFlags register */
134 if (reg_func) output( "\tpushfl\n" );
136 if ( UsePIC )
138 output( "\tcall 1f\n" );
139 output( "1:\tpopl %%ecx\n" );
140 output( "\t.byte 0x2e\n\tmovl %s-1b(%%ecx),%%edx\n", asm_name("CallTo16_DataSelector") );
142 else
143 output( "\t.byte 0x2e\n\tmovl %s,%%edx\n", asm_name("CallTo16_DataSelector") );
145 /* Load 32-bit segment registers */
146 output( "\tmovw %%dx, %%ds\n" );
147 output( "\tmovw %%dx, %%es\n" );
149 if ( UsePIC )
150 output( "\tmovw %s-1b(%%ecx), %%fs\n", asm_name("CallTo16_TebSelector") );
151 else
152 output( "\tmovw %s, %%fs\n", asm_name("CallTo16_TebSelector") );
154 output( "\t.byte 0x64\n\tmov (%d),%%gs\n", GS_OFFSET );
156 /* Translate STACK16FRAME base to flat offset in %edx */
157 output( "\tmovw %%ss, %%dx\n" );
158 output( "\tandl $0xfff8, %%edx\n" );
159 output( "\tshrl $1, %%edx\n" );
160 if (UsePIC)
162 output( "\taddl wine_ldt_copy_ptr-1b(%%ecx),%%edx\n" );
163 output( "\tmovl (%%edx), %%edx\n" );
165 else
166 output( "\tmovl %s(%%edx), %%edx\n", asm_name("wine_ldt_copy") );
167 output( "\tmovzwl %%sp, %%ebp\n" );
168 output( "\tleal %d(%%ebp,%%edx), %%edx\n", reg_func ? 0 : -4 );
170 /* Get saved flags into %ecx */
171 if (reg_func) output( "\tpopl %%ecx\n" );
173 /* Get the 32-bit stack pointer from the TEB and complete STACK16FRAME */
174 output( "\t.byte 0x64\n\tmovl (%d), %%ebp\n", STACKOFFSET );
175 output( "\tpushl %%ebp\n" );
177 /* Switch stacks */
178 output( "\t.byte 0x64\n\tmovw %%ss, (%d)\n", STACKOFFSET + 2 );
179 output( "\t.byte 0x64\n\tmovw %%sp, (%d)\n", STACKOFFSET );
180 output( "\tpushl %%ds\n" );
181 output( "\tpopl %%ss\n" );
182 output( "\tmovl %%ebp, %%esp\n" );
183 output( "\taddl $0x20,%%ebp\n"); /* FIELD_OFFSET(STACK32FRAME,ebp) */
186 /* At this point:
187 STACK16FRAME is completely set up
188 DS, ES, SS: flat data segment
189 FS: current TEB
190 ESP: points to last STACK32FRAME
191 EBP: points to ebp member of last STACK32FRAME
192 EDX: points to current STACK16FRAME
193 ECX: contains saved flags
194 all other registers: unchanged */
196 /* Special case: C16ThkSL stub */
197 if ( thunk )
199 /* Set up registers as expected and call thunk */
200 output( "\tleal 0x1a(%%edx),%%ebx\n" ); /* sizeof(STACK16FRAME)-22 */
201 output( "\tleal -4(%%esp), %%ebp\n" );
203 output( "\tcall *0x26(%%edx)\n"); /* FIELD_OFFSET(STACK16FRAME,entry_point) */
205 /* Switch stack back */
206 output( "\t.byte 0x64\n\tmovw (%d), %%ss\n", STACKOFFSET+2 );
207 output( "\t.byte 0x64\n\tmovzwl (%d), %%esp\n", STACKOFFSET );
208 output( "\t.byte 0x64\n\tpopl (%d)\n", STACKOFFSET );
210 /* Restore registers and return directly to caller */
211 output( "\taddl $8, %%esp\n" );
212 output( "\tpopl %%ebp\n" );
213 output( "\tpopw %%ds\n" );
214 output( "\tpopw %%es\n" );
215 output( "\tpopw %%fs\n" );
216 output( "\tpopw %%gs\n" );
217 output( "\taddl $20, %%esp\n" );
219 output( "\txorb %%ch, %%ch\n" );
220 output( "\tpopl %%ebx\n" );
221 output( "\taddw %%cx, %%sp\n" );
222 output( "\tpush %%ebx\n" );
224 output( "\t.byte 0x66\n" );
225 output( "\tlret\n" );
227 output_function_size( "__wine_call_from_16_thunk" );
228 return;
232 /* Build register CONTEXT */
233 if ( reg_func )
235 output( "\tsubl $0x2cc,%%esp\n" ); /* sizeof(CONTEXT86) */
237 output( "\tmovl %%ecx,0xc0(%%esp)\n" ); /* EFlags */
239 output( "\tmovl %%eax,0xb0(%%esp)\n" ); /* Eax */
240 output( "\tmovl %%ebx,0xa4(%%esp)\n" ); /* Ebx */
241 output( "\tmovl %%esi,0xa0(%%esp)\n" ); /* Esi */
242 output( "\tmovl %%edi,0x9c(%%esp)\n" ); /* Edi */
244 output( "\tmovl 0x0c(%%edx),%%eax\n"); /* FIELD_OFFSET(STACK16FRAME,ebp) */
245 output( "\tmovl %%eax,0xb4(%%esp)\n" ); /* Ebp */
246 output( "\tmovl 0x08(%%edx),%%eax\n"); /* FIELD_OFFSET(STACK16FRAME,ecx) */
247 output( "\tmovl %%eax,0xac(%%esp)\n" ); /* Ecx */
248 output( "\tmovl 0x04(%%edx),%%eax\n"); /* FIELD_OFFSET(STACK16FRAME,edx) */
249 output( "\tmovl %%eax,0xa8(%%esp)\n" ); /* Edx */
251 output( "\tmovzwl 0x10(%%edx),%%eax\n"); /* FIELD_OFFSET(STACK16FRAME,ds) */
252 output( "\tmovl %%eax,0x98(%%esp)\n" ); /* SegDs */
253 output( "\tmovzwl 0x12(%%edx),%%eax\n"); /* FIELD_OFFSET(STACK16FRAME,es) */
254 output( "\tmovl %%eax,0x94(%%esp)\n" ); /* SegEs */
255 output( "\tmovzwl 0x14(%%edx),%%eax\n"); /* FIELD_OFFSET(STACK16FRAME,fs) */
256 output( "\tmovl %%eax,0x90(%%esp)\n" ); /* SegFs */
257 output( "\tmovzwl 0x16(%%edx),%%eax\n"); /* FIELD_OFFSET(STACK16FRAME,gs) */
258 output( "\tmovl %%eax,0x8c(%%esp)\n" ); /* SegGs */
260 output( "\tmovzwl 0x2e(%%edx),%%eax\n"); /* FIELD_OFFSET(STACK16FRAME,cs) */
261 output( "\tmovl %%eax,0xbc(%%esp)\n" ); /* SegCs */
262 output( "\tmovzwl 0x2c(%%edx),%%eax\n"); /* FIELD_OFFSET(STACK16FRAME,ip) */
263 output( "\tmovl %%eax,0xb8(%%esp)\n" ); /* Eip */
265 output( "\t.byte 0x64\n\tmovzwl (%d), %%eax\n", STACKOFFSET+2 );
266 output( "\tmovl %%eax,0xc8(%%esp)\n" ); /* SegSs */
267 output( "\t.byte 0x64\n\tmovzwl (%d), %%eax\n", STACKOFFSET );
268 output( "\taddl $0x2c,%%eax\n"); /* FIELD_OFFSET(STACK16FRAME,ip) */
269 output( "\tmovl %%eax,0xc4(%%esp)\n" ); /* Esp */
270 #if 0
271 output( "\tfsave 0x1c(%%esp)\n" ); /* FloatSave */
272 #endif
274 /* Push address of CONTEXT86 structure -- popped by the relay routine */
275 output( "\tmovl %%esp,%%eax\n" );
276 output( "\tandl $~15,%%esp\n" );
277 output( "\tsubl $4,%%esp\n" );
278 output( "\tpushl %%eax\n" );
280 else
282 output( "\tsubl $8,%%esp\n" );
283 output( "\tandl $~15,%%esp\n" );
284 output( "\taddl $8,%%esp\n" );
287 /* Call relay routine (which will call the API entry point) */
288 output( "\tleal 0x30(%%edx),%%eax\n" ); /* sizeof(STACK16FRAME) */
289 output( "\tpushl %%eax\n" );
290 output( "\tpushl 0x26(%%edx)\n"); /* FIELD_OFFSET(STACK16FRAME,entry_point) */
291 output( "\tcall *0x20(%%edx)\n"); /* FIELD_OFFSET(STACK16FRAME,relay) */
293 if ( reg_func )
295 output( "\tleal -748(%%ebp),%%ebx\n" ); /* sizeof(CONTEXT) + FIELD_OFFSET(STACK32FRAME,ebp) */
297 /* Switch stack back */
298 output( "\t.byte 0x64\n\tmovw (%d), %%ss\n", STACKOFFSET+2 );
299 output( "\t.byte 0x64\n\tmovzwl (%d), %%esp\n", STACKOFFSET );
300 output( "\t.byte 0x64\n\tpopl (%d)\n", STACKOFFSET );
302 /* Get return address to CallFrom16 stub */
303 output( "\taddw $0x14,%%sp\n" ); /* FIELD_OFFSET(STACK16FRAME,callfrom_ip)-4 */
304 output( "\tpopl %%eax\n" );
305 output( "\tpopl %%edx\n" );
307 /* Restore all registers from CONTEXT */
308 output( "\tmovw 0xc8(%%ebx),%%ss\n"); /* SegSs */
309 output( "\tmovl 0xc4(%%ebx),%%esp\n"); /* Esp */
310 output( "\taddl $4, %%esp\n" ); /* room for final return address */
312 output( "\tpushw 0xbc(%%ebx)\n"); /* SegCs */
313 output( "\tpushw 0xb8(%%ebx)\n"); /* Eip */
314 output( "\tpushl %%edx\n" );
315 output( "\tpushl %%eax\n" );
316 output( "\tpushl 0xc0(%%ebx)\n"); /* EFlags */
317 output( "\tpushl 0x98(%%ebx)\n"); /* SegDs */
319 output( "\tpushl 0x94(%%ebx)\n"); /* SegEs */
320 output( "\tpopl %%es\n" );
321 output( "\tpushl 0x90(%%ebx)\n"); /* SegFs */
322 output( "\tpopl %%fs\n" );
323 output( "\tpushl 0x8c(%%ebx)\n"); /* SegGs */
324 output( "\tpopl %%gs\n" );
326 output( "\tmovl 0xb4(%%ebx),%%ebp\n"); /* Ebp */
327 output( "\tmovl 0xa0(%%ebx),%%esi\n"); /* Esi */
328 output( "\tmovl 0x9c(%%ebx),%%edi\n"); /* Edi */
329 output( "\tmovl 0xb0(%%ebx),%%eax\n"); /* Eax */
330 output( "\tmovl 0xa8(%%ebx),%%edx\n"); /* Edx */
331 output( "\tmovl 0xac(%%ebx),%%ecx\n"); /* Ecx */
332 output( "\tmovl 0xa4(%%ebx),%%ebx\n"); /* Ebx */
334 output( "\tpopl %%ds\n" );
335 output( "\tpopfl\n" );
336 output( "\tlret\n" );
338 output_function_size( "__wine_call_from_16_regs" );
340 else
342 /* Switch stack back */
343 output( "\t.byte 0x64\n\tmovw (%d), %%ss\n", STACKOFFSET+2 );
344 output( "\t.byte 0x64\n\tmovzwl (%d), %%esp\n", STACKOFFSET );
345 output( "\t.byte 0x64\n\tpopl (%d)\n", STACKOFFSET );
347 /* Restore registers */
348 output( "\tpopl %%edx\n" );
349 output( "\tpopl %%ecx\n" );
350 output( "\tpopl %%ebp\n" );
351 output( "\tpopw %%ds\n" );
352 output( "\tpopw %%es\n" );
353 output( "\tpopw %%fs\n" );
354 output( "\tpopw %%gs\n" );
356 /* Return to return stub which will return to caller */
357 output( "\tlret $12\n" );
359 output_function_size( "__wine_call_from_16" );
364 /*******************************************************************
365 * BuildCallTo16Core
367 * This routine builds the core routines used in 32->16 thunks:
369 * extern DWORD WINAPI wine_call_to_16( FARPROC16 target, DWORD cbArgs, PEXCEPTION_HANDLER handler );
370 * extern void WINAPI wine_call_to_16_regs( CONTEXT86 *context, DWORD cbArgs, PEXCEPTION_HANDLER handler );
372 * These routines can be called directly from 32-bit code.
374 * All routines expect that the 16-bit stack contents (arguments) and the
375 * return address (segptr to CallTo16_Ret) were already set up by the
376 * caller; nb_args must contain the number of bytes to be conserved. The
377 * 16-bit SS:SP will be set accordingly.
379 * All other registers are either taken from the CONTEXT86 structure
380 * or else set to default values. The target routine address is either
381 * given directly or taken from the CONTEXT86.
383 static void BuildCallTo16Core( int reg_func )
385 const char *name = reg_func ? "wine_call_to_16_regs" : "wine_call_to_16";
387 /* Function header */
388 function_header( name );
390 /* Function entry sequence */
391 output_cfi( ".cfi_startproc" );
392 output( "\tpushl %%ebp\n" );
393 output_cfi( ".cfi_adjust_cfa_offset 4" );
394 output_cfi( ".cfi_rel_offset %%ebp,0" );
395 output( "\tmovl %%esp, %%ebp\n" );
396 output_cfi( ".cfi_def_cfa_register %%ebp" );
398 /* Save the 32-bit registers */
399 output( "\tpushl %%ebx\n" );
400 output_cfi( ".cfi_rel_offset %%ebx,-4" );
401 output( "\tpushl %%esi\n" );
402 output_cfi( ".cfi_rel_offset %%esi,-8" );
403 output( "\tpushl %%edi\n" );
404 output_cfi( ".cfi_rel_offset %%edi,-12" );
405 output( "\t.byte 0x64\n\tmov %%gs,(%d)\n", GS_OFFSET );
407 /* Setup exception frame */
408 output( "\t.byte 0x64\n\tpushl (%d)\n", STACKOFFSET );
409 output( "\tpushl 16(%%ebp)\n" ); /* handler */
410 output( "\t.byte 0x64\n\tpushl (0)\n" );
411 output( "\t.byte 0x64\n\tmovl %%esp,(0)\n" );
413 /* Call the actual CallTo16 routine (simulate a lcall) */
414 output( "\tpushl %%cs\n" );
415 output( "\tcall .L%s\n", name );
417 /* Remove exception frame */
418 output( "\t.byte 0x64\n\tpopl (0)\n" );
419 output( "\taddl $4, %%esp\n" );
420 output( "\t.byte 0x64\n\tpopl (%d)\n", STACKOFFSET );
422 if ( !reg_func )
424 /* Convert return value */
425 output( "\tandl $0xffff,%%eax\n" );
426 output( "\tshll $16,%%edx\n" );
427 output( "\torl %%edx,%%eax\n" );
429 else
432 * Modify CONTEXT86 structure to contain new values
434 * NOTE: We restore only EAX, EBX, EDX, EDX, EBP, and ESP.
435 * The segment registers as well as ESI and EDI should
436 * not be modified by a well-behaved 16-bit routine in
437 * any case. [If necessary, we could restore them as well,
438 * at the cost of a somewhat less efficient return path.]
441 output( "\tmovl 0x14(%%esp),%%edi\n" ); /* FIELD_OFFSET(STACK32FRAME,target) - FIELD_OFFSET(STACK32FRAME,edi) */
442 /* everything above edi has been popped already */
444 output( "\tmovl %%eax,0xb0(%%edi)\n"); /* Eax */
445 output( "\tmovl %%ebx,0xa4(%%edi)\n"); /* Ebx */
446 output( "\tmovl %%ecx,0xac(%%edi)\n"); /* Ecx */
447 output( "\tmovl %%edx,0xa8(%%edi)\n"); /* Edx */
448 output( "\tmovl %%ebp,0xb4(%%edi)\n"); /* Ebp */
449 output( "\tmovl %%esi,0xc4(%%edi)\n"); /* Esp */
450 /* The return glue code saved %esp into %esi */
453 /* Restore the 32-bit registers */
454 output( "\tpopl %%edi\n" );
455 output_cfi( ".cfi_same_value %%edi" );
456 output( "\tpopl %%esi\n" );
457 output_cfi( ".cfi_same_value %%esi" );
458 output( "\tpopl %%ebx\n" );
459 output_cfi( ".cfi_same_value %%ebx" );
461 /* Function exit sequence */
462 output( "\tpopl %%ebp\n" );
463 output_cfi( ".cfi_def_cfa %%esp,4" );
464 output_cfi( ".cfi_same_value %%ebp" );
465 output( "\tret $12\n" );
466 output_cfi( ".cfi_endproc" );
469 /* Start of the actual CallTo16 routine */
471 output( ".L%s:\n", name );
473 /* Switch to the 16-bit stack */
474 output( "\tmovl %%esp,%%edx\n" );
475 output( "\t.byte 0x64\n\tmovw (%d),%%ss\n", STACKOFFSET + 2);
476 output( "\t.byte 0x64\n\tmovw (%d),%%sp\n", STACKOFFSET );
477 output( "\t.byte 0x64\n\tmovl %%edx,(%d)\n", STACKOFFSET );
479 /* Make %bp point to the previous stackframe (built by CallFrom16) */
480 output( "\tmovzwl %%sp,%%ebp\n" );
481 output( "\tleal 0x2a(%%ebp),%%ebp\n"); /* FIELD_OFFSET(STACK16FRAME,bp) */
483 /* Add the specified offset to the new sp */
484 output( "\tsubw 0x2c(%%edx), %%sp\n"); /* FIELD_OFFSET(STACK32FRAME,nb_args) */
486 if (reg_func)
488 /* Push the called routine address */
489 output( "\tmovl 0x28(%%edx),%%edx\n"); /* FIELD_OFFSET(STACK32FRAME,target) */
490 output( "\tpushw 0xbc(%%edx)\n"); /* SegCs */
491 output( "\tpushw 0xb8(%%edx)\n"); /* Eip */
493 /* Get the registers */
494 output( "\tpushw 0x98(%%edx)\n"); /* SegDs */
495 output( "\tpushl 0x94(%%edx)\n"); /* SegEs */
496 output( "\tpopl %%es\n" );
497 output( "\tpushl 0x90(%%edx)\n"); /* SegFs */
498 output( "\tpopl %%fs\n" );
499 output( "\tpushl 0x8c(%%edx)\n"); /* SegGs */
500 output( "\tpopl %%gs\n" );
501 output( "\tmovl 0xb4(%%edx),%%ebp\n"); /* Ebp */
502 output( "\tmovl 0xa0(%%edx),%%esi\n"); /* Esi */
503 output( "\tmovl 0x9c(%%edx),%%edi\n"); /* Edi */
504 output( "\tmovl 0xb0(%%edx),%%eax\n"); /* Eax */
505 output( "\tmovl 0xa4(%%edx),%%ebx\n"); /* Ebx */
506 output( "\tmovl 0xac(%%edx),%%ecx\n"); /* Ecx */
507 output( "\tmovl 0xa8(%%edx),%%edx\n"); /* Edx */
509 /* Get the 16-bit ds */
510 output( "\tpopw %%ds\n" );
512 else /* not a register function */
514 /* Push the called routine address */
515 output( "\tpushl 0x28(%%edx)\n"); /* FIELD_OFFSET(STACK32FRAME,target) */
517 /* Set %fs and %gs to the value saved by the last CallFrom16 */
518 output( "\tpushw -22(%%ebp)\n" ); /* FIELD_OFFSET(STACK16FRAME,fs)-FIELD_OFFSET(STACK16FRAME,bp) */
519 output( "\tpopw %%fs\n" );
520 output( "\tpushw -20(%%ebp)\n" ); /* FIELD_OFFSET(STACK16FRAME,gs)-FIELD_OFFSET(STACK16FRAME,bp) */
521 output( "\tpopw %%gs\n" );
523 /* Set %ds and %es (and %ax just in case) equal to %ss */
524 output( "\tmovw %%ss,%%ax\n" );
525 output( "\tmovw %%ax,%%ds\n" );
526 output( "\tmovw %%ax,%%es\n" );
529 /* Jump to the called routine */
530 output( "\t.byte 0x66\n" );
531 output( "\tlret\n" );
533 /* Function footer */
534 output_function_size( name );
538 /*******************************************************************
539 * BuildRet16Func
541 * Build the return code for 16-bit callbacks
543 static void BuildRet16Func(void)
545 function_header( "__wine_call_to_16_ret" );
547 /* Save %esp into %esi */
548 output( "\tmovl %%esp,%%esi\n" );
550 /* Restore 32-bit segment registers */
552 output( "\t.byte 0x2e\n\tmovl %s", asm_name("CallTo16_DataSelector") );
553 output( "-%s,%%edi\n", asm_name("__wine_call16_start") );
554 output( "\tmovw %%di,%%ds\n" );
555 output( "\tmovw %%di,%%es\n" );
557 output( "\t.byte 0x2e\n\tmov %s", asm_name("CallTo16_TebSelector") );
558 output( "-%s,%%fs\n", asm_name("__wine_call16_start") );
560 output( "\t.byte 0x64\n\tmov (%d),%%gs\n", GS_OFFSET );
562 /* Restore the 32-bit stack */
564 output( "\tmovw %%di,%%ss\n" );
565 output( "\t.byte 0x64\n\tmovl (%d),%%esp\n", STACKOFFSET );
567 /* Return to caller */
569 output( "\tlret\n" );
570 output_function_size( "__wine_call_to_16_ret" );
574 /*******************************************************************
575 * BuildCallTo32CBClient
577 * Call a CBClient relay stub from 32-bit code (KERNEL.620).
579 * Since the relay stub is itself 32-bit, this should not be a problem;
580 * unfortunately, the relay stubs are expected to switch back to a
581 * 16-bit stack (and 16-bit code) after completion :-(
583 * This would conflict with our 16- vs. 32-bit stack handling, so
584 * we simply switch *back* to our 32-bit stack before returning to
585 * the caller ...
587 * The CBClient relay stub expects to be called with the following
588 * 16-bit stack layout, and with ebp and ebx pointing into the 16-bit
589 * stack at the designated places:
591 * ...
592 * (ebp+14) original arguments to the callback routine
593 * (ebp+10) far return address to original caller
594 * (ebp+6) Thunklet target address
595 * (ebp+2) Thunklet relay ID code
596 * (ebp) BP (saved by CBClientGlueSL)
597 * (ebp-2) SI (saved by CBClientGlueSL)
598 * (ebp-4) DI (saved by CBClientGlueSL)
599 * (ebp-6) DS (saved by CBClientGlueSL)
601 * ... buffer space used by the 16-bit side glue for temp copies
603 * (ebx+4) far return address to 16-bit side glue code
604 * (ebx) saved 16-bit ss:sp (pointing to ebx+4)
606 * The 32-bit side glue code accesses both the original arguments (via ebp)
607 * and the temporary copies prepared by the 16-bit side glue (via ebx).
608 * After completion, the stub will load ss:sp from the buffer at ebx
609 * and perform a far return to 16-bit code.
611 * To trick the relay stub into returning to us, we replace the 16-bit
612 * return address to the glue code by a cs:ip pair pointing to our
613 * return entry point (the original return address is saved first).
614 * Our return stub thus called will then reload the 32-bit ss:esp and
615 * return to 32-bit code (by using and ss:esp value that we have also
616 * pushed onto the 16-bit stack before and a cs:eip values found at
617 * that position on the 32-bit stack). The ss:esp to be restored is
618 * found relative to the 16-bit stack pointer at:
620 * (ebx-4) ss (flat)
621 * (ebx-8) sp (32-bit stack pointer)
623 * The second variant of this routine, CALL32_CBClientEx, which is used
624 * to implement KERNEL.621, has to cope with yet another problem: Here,
625 * the 32-bit side directly returns to the caller of the CBClient thunklet,
626 * restoring registers saved by CBClientGlueSL and cleaning up the stack.
627 * As we have to return to our 32-bit code first, we have to adapt the
628 * layout of our temporary area so as to include values for the registers
629 * that are to be restored, and later (in the implementation of KERNEL.621)
630 * we *really* restore them. The return stub restores DS, DI, SI, and BP
631 * from the stack, skips the next 8 bytes (CBClient relay code / target),
632 * and then performs a lret NN, where NN is the number of arguments to be
633 * removed. Thus, we prepare our temporary area as follows:
635 * (ebx+22) 16-bit cs (this segment)
636 * (ebx+20) 16-bit ip ('16-bit' return entry point)
637 * (ebx+16) 32-bit ss (flat)
638 * (ebx+12) 32-bit sp (32-bit stack pointer)
639 * (ebx+10) 16-bit bp (points to ebx+24)
640 * (ebx+8) 16-bit si (ignored)
641 * (ebx+6) 16-bit di (ignored)
642 * (ebx+4) 16-bit ds (we actually use the flat DS here)
643 * (ebx+2) 16-bit ss (16-bit stack segment)
644 * (ebx+0) 16-bit sp (points to ebx+4)
646 * Note that we ensure that DS is not changed and remains the flat segment,
647 * and the 32-bit stack pointer our own return stub needs fits just
648 * perfectly into the 8 bytes that are skipped by the Windows stub.
649 * One problem is that we have to determine the number of removed arguments,
650 * as these have to be really removed in KERNEL.621. Thus, the BP value
651 * that we place in the temporary area to be restored, contains the value
652 * that SP would have if no arguments were removed. By comparing the actual
653 * value of SP with this value in our return stub we can compute the number
654 * of removed arguments. This is then returned to KERNEL.621.
656 * The stack layout of this function:
657 * (ebp+20) nArgs pointer to variable receiving nr. of args (Ex only)
658 * (ebp+16) esi pointer to caller's esi value
659 * (ebp+12) arg ebp value to be set for relay stub
660 * (ebp+8) func CBClient relay stub address
661 * (ebp+4) ret addr
662 * (ebp) ebp
664 static void BuildCallTo32CBClient( int isEx )
666 function_header( isEx ? "CALL32_CBClientEx" : "CALL32_CBClient" );
668 /* Entry code */
670 output_cfi( ".cfi_startproc" );
671 output( "\tpushl %%ebp\n" );
672 output_cfi( ".cfi_adjust_cfa_offset 4" );
673 output_cfi( ".cfi_rel_offset %%ebp,0" );
674 output( "\tmovl %%esp,%%ebp\n" );
675 output_cfi( ".cfi_def_cfa_register %%ebp" );
676 output( "\tpushl %%edi\n" );
677 output_cfi( ".cfi_rel_offset %%edi,-4" );
678 output( "\tpushl %%esi\n" );
679 output_cfi( ".cfi_rel_offset %%esi,-8" );
680 output( "\tpushl %%ebx\n" );
681 output_cfi( ".cfi_rel_offset %%ebx,-12" );
683 /* Get pointer to temporary area and save the 32-bit stack pointer */
685 output( "\tmovl 16(%%ebp), %%ebx\n" );
686 output( "\tleal -8(%%esp), %%eax\n" );
688 if ( !isEx )
689 output( "\tmovl %%eax, -8(%%ebx)\n" );
690 else
691 output( "\tmovl %%eax, 12(%%ebx)\n" );
693 /* Set up registers and call CBClient relay stub (simulating a far call) */
695 output( "\tmovl 20(%%ebp), %%esi\n" );
696 output( "\tmovl (%%esi), %%esi\n" );
698 output( "\tmovl 8(%%ebp), %%eax\n" );
699 output( "\tmovl 12(%%ebp), %%ebp\n" );
701 output( "\tpushl %%cs\n" );
702 output( "\tcall *%%eax\n" );
704 /* Return new esi value to caller */
706 output( "\tmovl 32(%%esp), %%edi\n" );
707 output( "\tmovl %%esi, (%%edi)\n" );
709 /* Return argument size to caller */
710 if ( isEx )
712 output( "\tmovl 36(%%esp), %%ebx\n" );
713 output( "\tmovl %%ebp, (%%ebx)\n" );
716 /* Restore registers and return */
718 output( "\tpopl %%ebx\n" );
719 output_cfi( ".cfi_same_value %%ebx" );
720 output( "\tpopl %%esi\n" );
721 output_cfi( ".cfi_same_value %%esi" );
722 output( "\tpopl %%edi\n" );
723 output_cfi( ".cfi_same_value %%edi" );
724 output( "\tpopl %%ebp\n" );
725 output_cfi( ".cfi_def_cfa %%esp,4" );
726 output_cfi( ".cfi_same_value %%ebp" );
727 output( "\tret\n" );
728 output_cfi( ".cfi_endproc" );
729 output_function_size( isEx ? "CALL32_CBClientEx" : "CALL32_CBClient" );
731 /* '16-bit' return stub */
733 function_header( isEx ? "CALL32_CBClientEx_Ret" : "CALL32_CBClient_Ret" );
734 if ( !isEx )
736 output( "\tmovzwl %%sp, %%ebx\n" );
737 output( "\tlssl %%ss:-16(%%ebx), %%esp\n" );
739 else
741 output( "\tmovzwl %%bp, %%ebx\n" );
742 output( "\tsubw %%bp, %%sp\n" );
743 output( "\tmovzwl %%sp, %%ebp\n" );
744 output( "\tlssl %%ss:-12(%%ebx), %%esp\n" );
746 output( "\tlret\n" );
747 output_function_size( isEx ? "CALL32_CBClientEx_Ret" : "CALL32_CBClient_Ret" );
751 /*******************************************************************
752 * build_call_from_regs_x86
754 * Build a 32-bit-to-Wine call-back function for a 'register' function.
755 * 'args' is the number of dword arguments.
757 * Stack layout:
758 * ...
759 * (ebp+20) first arg
760 * (ebp+16) ret addr to user code
761 * (ebp+12) func to call (relative to relay code ret addr)
762 * (ebp+8) number of args
763 * (ebp+4) ret addr to relay code
764 * (ebp+0) saved ebp
765 * (ebp-128) buffer area to allow stack frame manipulation
766 * (ebp-332) CONTEXT86 struct
767 * (ebp-336) padding for stack alignment
768 * (ebp-336-n) CONTEXT86 *argument
769 * .... other arguments copied from (ebp+12)
771 * The entry point routine is called with a CONTEXT* extra argument,
772 * following the normal args. In this context structure, EIP_reg
773 * contains the return address to user code, and ESP_reg the stack
774 * pointer on return (with the return address and arguments already
775 * removed).
777 static void build_call_from_regs_x86(void)
779 static const int STACK_SPACE = 128 + 0x2cc /* sizeof(CONTEXT86) */;
781 /* Function header */
783 output( "\t.text\n" );
784 function_header( "__wine_call_from_regs" );
786 /* Allocate some buffer space on the stack */
788 output_cfi( ".cfi_startproc" );
789 output( "\tpushl %%ebp\n" );
790 output_cfi( ".cfi_adjust_cfa_offset 4" );
791 output_cfi( ".cfi_rel_offset %%ebp,0" );
792 output( "\tmovl %%esp,%%ebp\n" );
793 output_cfi( ".cfi_def_cfa_register %%ebp" );
794 output( "\tleal -%d(%%esp),%%esp\n", STACK_SPACE );
796 /* Build the context structure */
798 output( "\tmovl %%eax,0xb0(%%esp)\n" ); /* Eax */
799 output( "\tpushfl\n" );
800 output( "\tpopl %%eax\n" );
801 output( "\tmovl %%eax,0xc0(%%esp)\n"); /* EFlags */
802 output( "\tmovl 0(%%ebp),%%eax\n" );
803 output( "\tmovl %%eax,0xb4(%%esp)\n"); /* Ebp */
804 output( "\tmovl %%ebx,0xa4(%%esp)\n"); /* Ebx */
805 output( "\tmovl %%ecx,0xac(%%esp)\n"); /* Ecx */
806 output( "\tmovl %%edx,0xa8(%%esp)\n"); /* Edx */
807 output( "\tmovl %%esi,0xa0(%%esp)\n"); /* Esi */
808 output( "\tmovl %%edi,0x9c(%%esp)\n"); /* Edi */
810 output( "\txorl %%eax,%%eax\n" );
811 output( "\tmovw %%cs,%%ax\n" );
812 output( "\tmovl %%eax,0xbc(%%esp)\n"); /* SegCs */
813 output( "\tmovw %%es,%%ax\n" );
814 output( "\tmovl %%eax,0x94(%%esp)\n"); /* SegEs */
815 output( "\tmovw %%fs,%%ax\n" );
816 output( "\tmovl %%eax,0x90(%%esp)\n"); /* SegFs */
817 output( "\tmovw %%gs,%%ax\n" );
818 output( "\tmovl %%eax,0x8c(%%esp)\n"); /* SegGs */
819 output( "\tmovw %%ss,%%ax\n" );
820 output( "\tmovl %%eax,0xc8(%%esp)\n"); /* SegSs */
821 output( "\tmovw %%ds,%%ax\n" );
822 output( "\tmovl %%eax,0x98(%%esp)\n"); /* SegDs */
823 output( "\tmovw %%ax,%%es\n" ); /* set %es equal to %ds just in case */
825 output( "\tmovl $0x10007,0(%%esp)\n"); /* ContextFlags */
827 output( "\tmovl 16(%%ebp),%%eax\n" ); /* Get %eip at time of call */
828 output( "\tmovl %%eax,0xb8(%%esp)\n"); /* Eip */
830 /* Transfer the arguments */
832 output( "\tmovl 8(%%ebp),%%ecx\n" ); /* fetch number of args to copy */
833 output( "\tleal 4(,%%ecx,4),%%edx\n" ); /* add 4 for context arg */
834 output( "\tsubl %%edx,%%esp\n" );
835 output( "\tandl $~15,%%esp\n" );
836 output( "\tleal 20(%%ebp),%%esi\n" ); /* get %esp at time of call */
837 output( "\tmovl %%esp,%%edi\n" );
838 output( "\ttest %%ecx,%%ecx\n" );
839 output( "\tjz 1f\n" );
840 output( "\tcld\n" );
841 output( "\trep\n\tmovsl\n" ); /* copy args */
842 output( "1:\tleal %d(%%ebp),%%eax\n", -STACK_SPACE ); /* get addr of context struct */
843 output( "\tmovl %%eax,(%%edi)\n" ); /* and pass it as extra arg */
844 output( "\tmovl %%esi,%d(%%ebp)\n", 0xc4 /* Esp */ - STACK_SPACE );
846 /* Call the entry point */
848 output( "\tmovl 4(%%ebp),%%eax\n" ); /* get relay code addr */
849 output( "\taddl 12(%%ebp),%%eax\n" );
850 output( "\tcall *%%eax\n" );
851 output( "\tleal -%d(%%ebp),%%ecx\n", STACK_SPACE );
853 /* Restore the context structure */
855 output( "2:\tpushl 0x94(%%ecx)\n"); /* SegEs */
856 output( "\tpopl %%es\n" );
857 output( "\tpushl 0x90(%%ecx)\n"); /* SegFs */
858 output( "\tpopl %%fs\n" );
859 output( "\tpushl 0x8c(%%ecx)\n"); /* SegGs */
860 output( "\tpopl %%gs\n" );
862 output( "\tmovl 0x9c(%%ecx),%%edi\n"); /* Edi */
863 output( "\tmovl 0xa0(%%ecx),%%esi\n"); /* Esi */
864 output( "\tmovl 0xa8(%%ecx),%%edx\n"); /* Edx */
865 output( "\tmovl 0xa4(%%ecx),%%ebx\n"); /* Ebx */
866 output( "\tmovl 0xb0(%%ecx),%%eax\n"); /* Eax */
867 output( "\tmovl 0xb4(%%ecx),%%ebp\n"); /* Ebp */
869 output( "\tpushl 0xc8(%%ecx)\n"); /* SegSs */
870 output( "\tpopl %%ss\n" );
871 output( "\tmovl 0xc4(%%ecx),%%esp\n"); /* Esp */
873 output( "\tpushl 0xc0(%%ecx)\n"); /* EFlags */
874 output( "\tpushl 0xbc(%%ecx)\n"); /* SegCs */
875 output( "\tpushl 0xb8(%%ecx)\n"); /* Eip */
876 output( "\tpushl 0x98(%%ecx)\n"); /* SegDs */
877 output( "\tmovl 0xac(%%ecx),%%ecx\n"); /* Ecx */
879 output( "\tpopl %%ds\n" );
880 output( "\tiret\n" );
881 output_cfi( ".cfi_endproc" );
882 output_function_size( "__wine_call_from_regs" );
884 function_header( "__wine_restore_regs" );
885 output_cfi( ".cfi_startproc" );
886 output( "\tmovl 4(%%esp),%%ecx\n" );
887 output( "\tjmp 2b\n" );
888 output_cfi( ".cfi_endproc" );
889 output_function_size( "__wine_restore_regs" );
893 /*******************************************************************
894 * BuildPendingEventCheck
896 * Build a function that checks whether there are any
897 * pending DPMI events.
899 * Stack layout:
901 * (sp+12) long eflags
902 * (sp+6) long cs
903 * (sp+2) long ip
904 * (sp) word fs
906 * On entry to function, fs register points to a valid TEB.
907 * On exit from function, stack will be popped.
909 static void BuildPendingEventCheck(void)
911 /* Function header */
913 function_header( "DPMI_PendingEventCheck" );
915 /* Check for pending events. */
917 output( "\t.byte 0x64\n\ttestl $0xffffffff,(%d)\n", VM86_PENDING_OFFSET );
918 output( "\tje %s\n", asm_name("DPMI_PendingEventCheck_Cleanup") );
919 output( "\t.byte 0x64\n\ttestl $0xffffffff,(%d)\n", DPMI_VIF_OFFSET );
920 output( "\tje %s\n", asm_name("DPMI_PendingEventCheck_Cleanup") );
922 /* Process pending events. */
924 output( "\tsti\n" );
926 /* Start cleanup. Restore fs register. */
928 output( "%s\n", asm_globl("DPMI_PendingEventCheck_Cleanup") );
929 output( "\tpopw %%fs\n" );
931 /* Return from function. */
933 output( "%s\n", asm_globl("DPMI_PendingEventCheck_Return") );
934 output( "\tiret\n" );
936 output_function_size( "DPMI_PendingEventCheck" );
940 /*******************************************************************
941 * output_asm_relays16
943 * Build all the 16-bit relay callbacks
945 void output_asm_relays16(void)
947 /* File header */
949 output( "\t.text\n" );
950 output( "%s:\n\n", asm_name("__wine_spec_thunk_text_16") );
952 output( "%s\n", asm_globl("__wine_call16_start") );
954 /* Standard CallFrom16 routine */
955 BuildCallFrom16Core( 0, 0 );
957 /* Register CallFrom16 routine */
958 BuildCallFrom16Core( 1, 0 );
960 /* C16ThkSL CallFrom16 routine */
961 BuildCallFrom16Core( 0, 1 );
963 /* Standard CallTo16 routine */
964 BuildCallTo16Core( 0 );
966 /* Register CallTo16 routine */
967 BuildCallTo16Core( 1 );
969 /* Standard CallTo16 return stub */
970 BuildRet16Func();
972 /* CBClientThunkSL routine */
973 BuildCallTo32CBClient( 0 );
975 /* CBClientThunkSLEx routine */
976 BuildCallTo32CBClient( 1 );
978 /* Pending DPMI events check stub */
979 BuildPendingEventCheck();
981 output( "%s\n", asm_globl("__wine_call16_end") );
982 output_function_size( "__wine_spec_thunk_text_16" );
984 /* Declare the return address and data selector variables */
985 output( "\n\t.data\n\t.align %d\n", get_alignment(4) );
986 output( "%s\n\t.long 0\n", asm_globl("CallTo16_DataSelector") );
987 output( "%s\n\t.long 0\n", asm_globl("CallTo16_TebSelector") );
989 output( "\t.text\n" );
990 output( "%s:\n", asm_name("__wine_spec_thunk_text_32") );
991 build_call_from_regs_x86();
992 output_function_size( "__wine_spec_thunk_text_32" );
996 /*******************************************************************
997 * output_asm_relays
999 * Build all the assembly relay callbacks
1001 void output_asm_relays(void)
1003 switch (target_cpu)
1005 case CPU_x86:
1006 build_call_from_regs_x86();
1007 break;
1008 default:
1009 break;