1/*
2 * Mesa 3-D graphics library
3 * Version:  7.1
4 *
5 * Copyright (C) 1999-2008  Brian Paul   All Rights Reserved.
6 *
7 * Permission is hereby granted, free of charge, to any person obtaining a
8 * copy of this software and associated documentation files (the "Software"),
9 * to deal in the Software without restriction, including without limitation
10 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
11 * and/or sell copies of the Software, and to permit persons to whom the
12 * Software is furnished to do so, subject to the following conditions:
13 *
14 * The above copyright notice and this permission notice shall be included
15 * in all copies or substantial portions of the Software.
16 *
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
18 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
20 * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
21 * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
22 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
23 */
24
25/**
26 * \file glapi_entrypoint.c
27 *
28 * Arch-specific code for manipulating GL API entrypoints (dispatch stubs).
29 */
30
31
32#include "glapi/glapi_priv.h"
33#include "mapi/u_execmem.h"
34
35
36#ifdef USE_X86_ASM
37
38#if defined( GLX_USE_TLS )
39extern       GLubyte gl_dispatch_functions_start[];
40extern       GLubyte gl_dispatch_functions_end[];
41#else
42extern const GLubyte gl_dispatch_functions_start[];
43#endif
44
45#endif /* USE_X86_ASM */
46
47
48#if defined(DISPATCH_FUNCTION_SIZE)
49
50_glapi_proc
51get_entrypoint_address(unsigned int functionOffset)
52{
53   return (_glapi_proc) (gl_dispatch_functions_start
54                         + (DISPATCH_FUNCTION_SIZE * functionOffset));
55}
56
57#endif
58
59
60#if defined(USE_X86_ASM)
61
62/**
63 * Perform platform-specific GL API entry-point fixups.
64 */
65static void
66init_glapi_relocs( void )
67{
68#if defined(GLX_USE_TLS) && !defined(GLX_X86_READONLY_TEXT)
69    extern unsigned long _x86_get_dispatch(void);
70    char run_time_patch[] = {
71       0x65, 0xa1, 0, 0, 0, 0 /* movl %gs:0,%eax */
72    };
73    GLuint *offset = (GLuint *) &run_time_patch[2]; /* 32-bits for x86/32 */
74    const GLubyte * const get_disp = (const GLubyte *) run_time_patch;
75    GLubyte * curr_func = (GLubyte *) gl_dispatch_functions_start;
76
77    *offset = _x86_get_dispatch();
78    while ( curr_func != (GLubyte *) gl_dispatch_functions_end ) {
79	(void) memcpy( curr_func, get_disp, sizeof(run_time_patch));
80	curr_func += DISPATCH_FUNCTION_SIZE;
81    }
82#endif
83}
84
85
86/**
87 * Generate a dispatch function (entrypoint) which jumps through
88 * the given slot number (offset) in the current dispatch table.
89 * We need assembly language in order to accomplish this.
90 */
91_glapi_proc
92generate_entrypoint(unsigned int functionOffset)
93{
94   /* 32 is chosen as something of a magic offset.  For x86, the dispatch
95    * at offset 32 is the first one where the offset in the
96    * "jmp OFFSET*4(%eax)" can't be encoded in a single byte.
97    */
98   const GLubyte * const template_func = gl_dispatch_functions_start
99     + (DISPATCH_FUNCTION_SIZE * 32);
100   GLubyte * const code = (GLubyte *) u_execmem_alloc(DISPATCH_FUNCTION_SIZE);
101
102
103   if ( code != NULL ) {
104      (void) memcpy(code, template_func, DISPATCH_FUNCTION_SIZE);
105      fill_in_entrypoint_offset( (_glapi_proc) code, functionOffset );
106   }
107
108   return (_glapi_proc) code;
109}
110
111
112/**
113 * This function inserts a new dispatch offset into the assembly language
114 * stub that was generated with the preceeding function.
115 */
116void
117fill_in_entrypoint_offset(_glapi_proc entrypoint, unsigned int offset)
118{
119   GLubyte * const code = (GLubyte *) entrypoint;
120
121#if defined(GLX_USE_TLS)
122   *((unsigned int *)(code +  8)) = 4 * offset;
123#elif defined(THREADS)
124   *((unsigned int *)(code + 11)) = 4 * offset;
125   *((unsigned int *)(code + 22)) = 4 * offset;
126#else
127   *((unsigned int *)(code +  7)) = 4 * offset;
128#endif
129}
130
131
132#elif defined(USE_SPARC_ASM)
133
134extern void __glapi_sparc_icache_flush(unsigned int *);
135
136static void
137init_glapi_relocs( void )
138{
139#if defined(HAVE_PTHREAD) || defined(GLX_USE_TLS)
140    static const unsigned int template[] = {
141#ifdef GLX_USE_TLS
142	0x05000000, /* sethi %hi(_glapi_tls_Dispatch), %g2 */
143	0x8730e00a, /* srl %g3, 10, %g3 */
144	0x8410a000, /* or %g2, %lo(_glapi_tls_Dispatch), %g2 */
145#ifdef __arch64__
146	0xc259c002, /* ldx [%g7 + %g2], %g1 */
147	0xc2584003, /* ldx [%g1 + %g3], %g1 */
148#else
149	0xc201c002, /* ld [%g7 + %g2], %g1 */
150	0xc2004003, /* ld [%g1 + %g3], %g1 */
151#endif
152	0x81c04000, /* jmp %g1 */
153	0x01000000, /* nop  */
154#else
155#ifdef __arch64__
156	0x03000000, /* 64-bit 0x00 --> sethi %hh(_glapi_Dispatch), %g1 */
157	0x05000000, /* 64-bit 0x04 --> sethi %lm(_glapi_Dispatch), %g2 */
158	0x82106000, /* 64-bit 0x08 --> or %g1, %hm(_glapi_Dispatch), %g1 */
159	0x8730e00a, /* 64-bit 0x0c --> srl %g3, 10, %g3 */
160	0x83287020, /* 64-bit 0x10 --> sllx %g1, 32, %g1 */
161	0x82004002, /* 64-bit 0x14 --> add %g1, %g2, %g1 */
162	0xc2586000, /* 64-bit 0x18 --> ldx [%g1 + %lo(_glapi_Dispatch)], %g1 */
163#else
164	0x03000000, /* 32-bit 0x00 --> sethi %hi(_glapi_Dispatch), %g1 */
165	0x8730e00a, /* 32-bit 0x04 --> srl %g3, 10, %g3 */
166	0xc2006000, /* 32-bit 0x08 --> ld [%g1 + %lo(_glapi_Dispatch)], %g1 */
167#endif
168	0x80a06000, /*             --> cmp %g1, 0 */
169	0x02800005, /*             --> be +4*5 */
170	0x01000000, /*             -->  nop  */
171#ifdef __arch64__
172	0xc2584003, /* 64-bit      --> ldx [%g1 + %g3], %g1 */
173#else
174	0xc2004003, /* 32-bit      --> ld [%g1 + %g3], %g1 */
175#endif
176	0x81c04000, /*             --> jmp %g1 */
177	0x01000000, /*             --> nop  */
178#ifdef __arch64__
179	0x9de3bf80, /* 64-bit      --> save  %sp, -128, %sp */
180#else
181	0x9de3bfc0, /* 32-bit      --> save  %sp, -64, %sp */
182#endif
183	0xa0100003, /*             --> mov  %g3, %l0 */
184	0x40000000, /*             --> call _glapi_get_dispatch */
185	0x01000000, /*             -->  nop */
186	0x82100008, /*             --> mov %o0, %g1 */
187	0x86100010, /*             --> mov %l0, %g3 */
188	0x10bffff7, /*             --> ba -4*9 */
189	0x81e80000, /*             -->  restore  */
190#endif
191    };
192#ifdef GLX_USE_TLS
193    extern unsigned int __glapi_sparc_tls_stub;
194    extern unsigned long __glapi_sparc_get_dispatch(void);
195    unsigned int *code = &__glapi_sparc_tls_stub;
196    unsigned long dispatch = __glapi_sparc_get_dispatch();
197#else
198    extern unsigned int __glapi_sparc_pthread_stub;
199    unsigned int *code = &__glapi_sparc_pthread_stub;
200    unsigned long dispatch = (unsigned long) &_glapi_Dispatch;
201    unsigned long call_dest = (unsigned long ) &_glapi_get_dispatch;
202    int idx;
203#endif
204
205#ifdef GLX_USE_TLS
206    code[0] = template[0] | (dispatch >> 10);
207    code[1] = template[1];
208    __glapi_sparc_icache_flush(&code[0]);
209    code[2] = template[2] | (dispatch & 0x3ff);
210    code[3] = template[3];
211    __glapi_sparc_icache_flush(&code[2]);
212    code[4] = template[4];
213    code[5] = template[5];
214    __glapi_sparc_icache_flush(&code[4]);
215    code[6] = template[6];
216    __glapi_sparc_icache_flush(&code[6]);
217#else
218#if defined(__arch64__)
219    code[0] = template[0] | (dispatch >> (32 + 10));
220    code[1] = template[1] | ((dispatch & 0xffffffff) >> 10);
221    __glapi_sparc_icache_flush(&code[0]);
222    code[2] = template[2] | ((dispatch >> 32) & 0x3ff);
223    code[3] = template[3];
224    __glapi_sparc_icache_flush(&code[2]);
225    code[4] = template[4];
226    code[5] = template[5];
227    __glapi_sparc_icache_flush(&code[4]);
228    code[6] = template[6] | (dispatch & 0x3ff);
229    idx = 7;
230#else
231    code[0] = template[0] | (dispatch >> 10);
232    code[1] = template[1];
233    __glapi_sparc_icache_flush(&code[0]);
234    code[2] = template[2] | (dispatch & 0x3ff);
235    idx = 3;
236#endif
237    code[idx + 0] = template[idx + 0];
238    __glapi_sparc_icache_flush(&code[idx - 1]);
239    code[idx + 1] = template[idx + 1];
240    code[idx + 2] = template[idx + 2];
241    __glapi_sparc_icache_flush(&code[idx + 1]);
242    code[idx + 3] = template[idx + 3];
243    code[idx + 4] = template[idx + 4];
244    __glapi_sparc_icache_flush(&code[idx + 3]);
245    code[idx + 5] = template[idx + 5];
246    code[idx + 6] = template[idx + 6];
247    __glapi_sparc_icache_flush(&code[idx + 5]);
248    code[idx + 7] = template[idx + 7];
249    code[idx + 8] = template[idx + 8] |
250	    (((call_dest - ((unsigned long) &code[idx + 8]))
251	      >> 2) & 0x3fffffff);
252    __glapi_sparc_icache_flush(&code[idx + 7]);
253    code[idx + 9] = template[idx + 9];
254    code[idx + 10] = template[idx + 10];
255    __glapi_sparc_icache_flush(&code[idx + 9]);
256    code[idx + 11] = template[idx + 11];
257    code[idx + 12] = template[idx + 12];
258    __glapi_sparc_icache_flush(&code[idx + 11]);
259    code[idx + 13] = template[idx + 13];
260    __glapi_sparc_icache_flush(&code[idx + 13]);
261#endif
262#endif
263}
264
265
266_glapi_proc
267generate_entrypoint(GLuint functionOffset)
268{
269#if defined(HAVE_PTHREAD) || defined(GLX_USE_TLS)
270   static const unsigned int template[] = {
271      0x07000000, /* sethi %hi(0), %g3 */
272      0x8210000f, /* mov  %o7, %g1 */
273      0x40000000, /* call */
274      0x9e100001, /* mov  %g1, %o7 */
275   };
276#ifdef GLX_USE_TLS
277   extern unsigned int __glapi_sparc_tls_stub;
278   unsigned long call_dest = (unsigned long ) &__glapi_sparc_tls_stub;
279#else
280   extern unsigned int __glapi_sparc_pthread_stub;
281   unsigned long call_dest = (unsigned long ) &__glapi_sparc_pthread_stub;
282#endif
283   unsigned int *code = (unsigned int *) u_execmem_alloc(sizeof(template));
284   if (code) {
285      code[0] = template[0] | (functionOffset & 0x3fffff);
286      code[1] = template[1];
287      __glapi_sparc_icache_flush(&code[0]);
288      code[2] = template[2] |
289         (((call_dest - ((unsigned long) &code[2]))
290	   >> 2) & 0x3fffffff);
291      code[3] = template[3];
292      __glapi_sparc_icache_flush(&code[2]);
293   }
294   return (_glapi_proc) code;
295#endif
296}
297
298
299void
300fill_in_entrypoint_offset(_glapi_proc entrypoint, GLuint offset)
301{
302   unsigned int *code = (unsigned int *) entrypoint;
303
304   code[0] &= ~0x3fffff;
305   code[0] |= (offset * sizeof(void *)) & 0x3fffff;
306   __glapi_sparc_icache_flush(&code[0]);
307}
308
309
310#else /* USE_*_ASM */
311
312static void
313init_glapi_relocs( void )
314{
315}
316
317
318_glapi_proc
319generate_entrypoint(GLuint functionOffset)
320{
321   (void) functionOffset;
322   return NULL;
323}
324
325
326void
327fill_in_entrypoint_offset(_glapi_proc entrypoint, GLuint offset)
328{
329   /* an unimplemented architecture */
330   (void) entrypoint;
331   (void) offset;
332}
333
334#endif /* USE_*_ASM */
335
336
337void
338init_glapi_relocs_once( void )
339{
340#if defined(HAVE_PTHREAD) || defined(GLX_USE_TLS)
341   static pthread_once_t once_control = PTHREAD_ONCE_INIT;
342   pthread_once( & once_control, init_glapi_relocs );
343#endif
344}
345