filename | src/sh4/xltcache.c |
changeset | 905:4c17ebd9ef5e |
prev | 901:32c5cf5e206f |
next | 922:8a8361264b1e |
author | nkeynes |
date | Wed Oct 29 23:51:58 2008 +0000 (15 years ago) |
permissions | -rw-r--r-- |
last change | Use regparam calling conventions for all functions called from translated code, along with a few other high-use functions. Can probably extend this to all functions, but as it is this is a nice performance boost |
view | annotate | diff | log | raw |
1 /**
2 * $Id$
3 *
4 * Translation cache management. This part is architecture independent.
5 *
6 * Copyright (c) 2005 Nathan Keynes.
7 *
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation; either version 2 of the License, or
11 * (at your option) any later version.
12 *
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
17 */
19 #include <sys/types.h>
20 #include <sys/mman.h>
21 #include <assert.h>
23 #include "dreamcast.h"
24 #include "sh4/sh4core.h"
25 #include "sh4/xltcache.h"
26 #include "x86dasm/x86dasm.h"
28 #define XLAT_LUT_PAGE_BITS 12
29 #define XLAT_LUT_TOTAL_BITS 28
30 #define XLAT_LUT_PAGE(addr) (((addr)>>13) & 0xFFFF)
31 #define XLAT_LUT_ENTRY(addr) (((addr)&0x1FFE) >> 1)
33 #define XLAT_LUT_PAGES (1<<(XLAT_LUT_TOTAL_BITS-XLAT_LUT_PAGE_BITS))
34 #define XLAT_LUT_PAGE_ENTRIES (1<<XLAT_LUT_PAGE_BITS)
35 #define XLAT_LUT_PAGE_SIZE (XLAT_LUT_PAGE_ENTRIES * sizeof(void *))
37 #define XLAT_LUT_ENTRY_EMPTY (void *)0
38 #define XLAT_LUT_ENTRY_USED (void *)1
40 #define NEXT(block) ( (xlat_cache_block_t)&((block)->code[(block)->size]))
41 #define IS_ENTRY_POINT(ent) (ent > XLAT_LUT_ENTRY_USED)
42 #define IS_ENTRY_USED(ent) (ent != XLAT_LUT_ENTRY_EMPTY)
44 #define MIN_BLOCK_SIZE 32
45 #define MIN_TOTAL_SIZE (sizeof(struct xlat_cache_block)+MIN_BLOCK_SIZE)
47 #define BLOCK_INACTIVE 0
48 #define BLOCK_ACTIVE 1
49 #define BLOCK_USED 2
51 xlat_cache_block_t xlat_new_cache;
52 xlat_cache_block_t xlat_new_cache_ptr;
53 xlat_cache_block_t xlat_new_create_ptr;
54 xlat_cache_block_t xlat_temp_cache;
55 xlat_cache_block_t xlat_temp_cache_ptr;
56 xlat_cache_block_t xlat_old_cache;
57 xlat_cache_block_t xlat_old_cache_ptr;
58 static void ***xlat_lut;
59 static gboolean xlat_initialized = FALSE;
61 void xlat_cache_init(void)
62 {
63 if( !xlat_initialized ) {
64 xlat_initialized = TRUE;
65 xlat_new_cache = mmap( NULL, XLAT_NEW_CACHE_SIZE, PROT_EXEC|PROT_READ|PROT_WRITE,
66 MAP_PRIVATE|MAP_ANON, -1, 0 );
67 xlat_temp_cache = mmap( NULL, XLAT_TEMP_CACHE_SIZE, PROT_EXEC|PROT_READ|PROT_WRITE,
68 MAP_PRIVATE|MAP_ANON, -1, 0 );
69 xlat_old_cache = mmap( NULL, XLAT_OLD_CACHE_SIZE, PROT_EXEC|PROT_READ|PROT_WRITE,
70 MAP_PRIVATE|MAP_ANON, -1, 0 );
71 xlat_new_cache_ptr = xlat_new_cache;
72 xlat_temp_cache_ptr = xlat_temp_cache;
73 xlat_old_cache_ptr = xlat_old_cache;
74 xlat_new_create_ptr = xlat_new_cache;
76 xlat_lut = mmap( NULL, XLAT_LUT_PAGES*sizeof(void *), PROT_READ|PROT_WRITE,
77 MAP_PRIVATE|MAP_ANON, -1, 0);
78 memset( xlat_lut, 0, XLAT_LUT_PAGES*sizeof(void *) );
79 }
80 xlat_flush_cache();
81 }
83 void xlat_print_free( FILE *out )
84 {
85 fprintf( out, "New space: %d\nTemp space: %d\nOld space: %d\n",
86 xlat_new_cache_ptr->size, xlat_temp_cache_ptr->size, xlat_old_cache_ptr->size );
87 }
89 /**
90 * Reset the cache structure to its default state
91 */
92 void xlat_flush_cache()
93 {
94 xlat_cache_block_t tmp;
95 int i;
96 xlat_new_cache_ptr = xlat_new_cache;
97 xlat_new_cache_ptr->active = 0;
98 xlat_new_cache_ptr->size = XLAT_NEW_CACHE_SIZE - 2*sizeof(struct xlat_cache_block);
99 tmp = NEXT(xlat_new_cache_ptr);
100 tmp->active = 1;
101 tmp->size = 0;
102 xlat_temp_cache_ptr = xlat_temp_cache;
103 xlat_temp_cache_ptr->active = 0;
104 xlat_temp_cache_ptr->size = XLAT_TEMP_CACHE_SIZE - 2*sizeof(struct xlat_cache_block);
105 tmp = NEXT(xlat_temp_cache_ptr);
106 tmp->active = 1;
107 tmp->size = 0;
108 xlat_old_cache_ptr = xlat_old_cache;
109 xlat_old_cache_ptr->active = 0;
110 xlat_old_cache_ptr->size = XLAT_OLD_CACHE_SIZE - 2*sizeof(struct xlat_cache_block);
111 tmp = NEXT(xlat_old_cache_ptr);
112 tmp->active = 1;
113 tmp->size = 0;
114 for( i=0; i<XLAT_LUT_PAGES; i++ ) {
115 if( xlat_lut[i] != NULL ) {
116 memset( xlat_lut[i], 0, XLAT_LUT_PAGE_SIZE );
117 }
118 }
119 }
121 static void xlat_flush_page_by_lut( void **page )
122 {
123 int i;
124 for( i=0; i<XLAT_LUT_PAGE_ENTRIES; i++ ) {
125 if( IS_ENTRY_POINT(page[i]) ) {
126 XLAT_BLOCK_FOR_CODE(page[i])->active = 0;
127 }
128 page[i] = NULL;
129 }
130 }
132 void FASTCALL xlat_invalidate_word( sh4addr_t addr )
133 {
134 if( xlat_lut ) {
135 void **page = xlat_lut[XLAT_LUT_PAGE(addr)];
136 if( page != NULL ) {
137 int entry = XLAT_LUT_ENTRY(addr);
138 if( page[entry] != NULL ) {
139 xlat_flush_page_by_lut(page);
140 }
141 }
142 }
143 }
145 void FASTCALL xlat_invalidate_long( sh4addr_t addr )
146 {
147 if( xlat_lut ) {
148 void **page = xlat_lut[XLAT_LUT_PAGE(addr)];
149 if( page != NULL ) {
150 int entry = XLAT_LUT_ENTRY(addr);
151 if( page[entry] != NULL || page[entry+1] != NULL ) {
152 xlat_flush_page_by_lut(page);
153 }
154 }
155 }
156 }
158 void FASTCALL xlat_invalidate_block( sh4addr_t address, size_t size )
159 {
160 int i;
161 int entry_count = size >> 1; // words;
162 uint32_t page_no = XLAT_LUT_PAGE(address);
163 int entry = XLAT_LUT_ENTRY(address);
164 if( xlat_lut ) {
165 do {
166 void **page = xlat_lut[page_no];
167 int page_entries = XLAT_LUT_PAGE_ENTRIES - entry;
168 if( entry_count < page_entries ) {
169 page_entries = entry_count;
170 }
171 if( page != NULL ) {
172 if( page_entries == XLAT_LUT_PAGE_ENTRIES ) {
173 /* Overwriting the entire page anyway */
174 xlat_flush_page_by_lut(page);
175 } else {
176 for( i=entry; i<entry+page_entries; i++ ) {
177 if( page[i] != NULL ) {
178 xlat_flush_page_by_lut(page);
179 break;
180 }
181 }
182 }
183 entry_count -= page_entries;
184 }
185 page_no ++;
186 entry_count -= page_entries;
187 entry = 0;
188 } while( entry_count > 0 );
189 }
190 }
192 void FASTCALL xlat_flush_page( sh4addr_t address )
193 {
194 void **page = xlat_lut[XLAT_LUT_PAGE(address)];
195 if( page != NULL ) {
196 xlat_flush_page_by_lut(page);
197 }
198 }
200 void * FASTCALL xlat_get_code( sh4addr_t address )
201 {
202 void *result = NULL;
203 void **page = xlat_lut[XLAT_LUT_PAGE(address)];
204 if( page != NULL ) {
205 result = (void *)(((uintptr_t)(page[XLAT_LUT_ENTRY(address)])) & (~((uintptr_t)0x03)));
206 }
207 return result;
208 }
210 xlat_recovery_record_t xlat_get_post_recovery( void *code, void *native_pc, gboolean with_terminal )
211 {
212 if( code != NULL ) {
213 uintptr_t pc_offset = ((uint8_t *)native_pc) - ((uint8_t *)code);
214 xlat_cache_block_t block = XLAT_BLOCK_FOR_CODE(code);
215 uint32_t count = block->recover_table_size;
216 xlat_recovery_record_t records = (xlat_recovery_record_t)(&block->code[block->recover_table_offset]);
217 uint32_t posn;
218 if( count > 0 && !with_terminal )
219 count--;
220 if( records[count-1].xlat_offset < pc_offset ) {
221 return NULL;
222 }
223 for( posn=count-1; posn > 0; posn-- ) {
224 if( records[posn-1].xlat_offset < pc_offset ) {
225 return &records[posn];
226 }
227 }
228 return &records[0]; // shouldn't happen
229 }
230 return NULL;
231 }
233 xlat_recovery_record_t xlat_get_pre_recovery( void *code, void *native_pc )
234 {
235 if( code != NULL ) {
236 uintptr_t pc_offset = ((uint8_t *)native_pc) - ((uint8_t *)code);
237 xlat_cache_block_t block = XLAT_BLOCK_FOR_CODE(code);
238 uint32_t count = block->recover_table_size;
239 xlat_recovery_record_t records = (xlat_recovery_record_t)(&block->code[block->recover_table_offset]);
240 uint32_t posn;
241 for( posn = 1; posn < count; posn++ ) {
242 if( records[posn].xlat_offset >= pc_offset ) {
243 return &records[posn-1];
244 }
245 }
246 return &records[count-1];
247 }
248 return NULL;
249 }
251 void ** FASTCALL xlat_get_lut_entry( sh4addr_t address )
252 {
253 void **page = xlat_lut[XLAT_LUT_PAGE(address)];
255 /* Add the LUT entry for the block */
256 if( page == NULL ) {
257 xlat_lut[XLAT_LUT_PAGE(address)] = page =
258 mmap( NULL, XLAT_LUT_PAGE_SIZE, PROT_READ|PROT_WRITE,
259 MAP_PRIVATE|MAP_ANON, -1, 0 );
260 memset( page, 0, XLAT_LUT_PAGE_SIZE );
261 }
263 return &page[XLAT_LUT_ENTRY(address)];
264 }
268 uint32_t FASTCALL xlat_get_block_size( void *block )
269 {
270 xlat_cache_block_t xlt = (xlat_cache_block_t)(((char *)block)-sizeof(struct xlat_cache_block));
271 return xlt->size;
272 }
274 uint32_t FASTCALL xlat_get_code_size( void *block )
275 {
276 xlat_cache_block_t xlt = (xlat_cache_block_t)(((char *)block)-sizeof(struct xlat_cache_block));
277 if( xlt->recover_table_offset == 0 ) {
278 return xlt->size;
279 } else {
280 return xlt->recover_table_offset;
281 }
282 }
284 /**
285 * Cut the specified block so that it has the given size, with the remaining data
286 * forming a new free block. If the free block would be less than the minimum size,
287 * the cut is not performed.
288 * @return the next block after the (possibly cut) block.
289 */
290 static inline xlat_cache_block_t xlat_cut_block( xlat_cache_block_t block, int cutsize )
291 {
292 cutsize = (cutsize + 3) & 0xFFFFFFFC; // force word alignment
293 assert( cutsize <= block->size );
294 if( block->size > cutsize + MIN_TOTAL_SIZE ) {
295 int oldsize = block->size;
296 block->size = cutsize;
297 xlat_cache_block_t next = NEXT(block);
298 next->active = 0;
299 next->size = oldsize - cutsize - sizeof(struct xlat_cache_block);
300 return next;
301 } else {
302 return NEXT(block);
303 }
304 }
306 /**
307 * Promote a block in temp space (or elsewhere for that matter) to old space.
308 *
309 * @param block to promote.
310 */
311 static void xlat_promote_to_old_space( xlat_cache_block_t block )
312 {
313 int allocation = (int)-sizeof(struct xlat_cache_block);
314 int size = block->size;
315 xlat_cache_block_t curr = xlat_old_cache_ptr;
316 xlat_cache_block_t start_block = curr;
317 do {
318 allocation += curr->size + sizeof(struct xlat_cache_block);
319 curr = NEXT(curr);
320 if( allocation > size ) {
321 break; /* done */
322 }
323 if( curr->size == 0 ) { /* End-of-cache Sentinel */
324 /* Leave what we just released as free space and start again from the
325 * top of the cache
326 */
327 start_block->active = 0;
328 start_block->size = allocation;
329 allocation = (int)-sizeof(struct xlat_cache_block);
330 start_block = curr = xlat_old_cache;
331 }
332 } while(1);
333 start_block->active = 1;
334 start_block->size = allocation;
335 start_block->lut_entry = block->lut_entry;
336 start_block->fpscr_mask = block->fpscr_mask;
337 start_block->fpscr = block->fpscr;
338 start_block->recover_table_offset = block->recover_table_offset;
339 start_block->recover_table_size = block->recover_table_size;
340 *block->lut_entry = &start_block->code;
341 memcpy( start_block->code, block->code, block->size );
342 xlat_old_cache_ptr = xlat_cut_block(start_block, size );
343 if( xlat_old_cache_ptr->size == 0 ) {
344 xlat_old_cache_ptr = xlat_old_cache;
345 }
346 }
348 /**
349 * Similarly to the above method, promotes a block to temp space.
350 * TODO: Try to combine these - they're nearly identical
351 */
352 void xlat_promote_to_temp_space( xlat_cache_block_t block )
353 {
354 int size = block->size;
355 int allocation = (int)-sizeof(struct xlat_cache_block);
356 xlat_cache_block_t curr = xlat_temp_cache_ptr;
357 xlat_cache_block_t start_block = curr;
358 do {
359 if( curr->active == BLOCK_USED ) {
360 xlat_promote_to_old_space( curr );
361 } else if( curr->active == BLOCK_ACTIVE ) {
362 // Active but not used, release block
363 *((uintptr_t *)curr->lut_entry) &= ((uintptr_t)0x03);
364 }
365 allocation += curr->size + sizeof(struct xlat_cache_block);
366 curr = NEXT(curr);
367 if( allocation > size ) {
368 break; /* done */
369 }
370 if( curr->size == 0 ) { /* End-of-cache Sentinel */
371 /* Leave what we just released as free space and start again from the
372 * top of the cache
373 */
374 start_block->active = 0;
375 start_block->size = allocation;
376 allocation = (int)-sizeof(struct xlat_cache_block);
377 start_block = curr = xlat_temp_cache;
378 }
379 } while(1);
380 start_block->active = 1;
381 start_block->size = allocation;
382 start_block->lut_entry = block->lut_entry;
383 start_block->fpscr_mask = block->fpscr_mask;
384 start_block->fpscr = block->fpscr;
385 start_block->recover_table_offset = block->recover_table_offset;
386 start_block->recover_table_size = block->recover_table_size;
387 *block->lut_entry = &start_block->code;
388 memcpy( start_block->code, block->code, block->size );
389 xlat_temp_cache_ptr = xlat_cut_block(start_block, size );
390 if( xlat_temp_cache_ptr->size == 0 ) {
391 xlat_temp_cache_ptr = xlat_temp_cache;
392 }
394 }
396 /**
397 * Returns the next block in the new cache list that can be written to by the
398 * translator. If the next block is active, it is evicted first.
399 */
400 xlat_cache_block_t xlat_start_block( sh4addr_t address )
401 {
402 if( xlat_new_cache_ptr->size == 0 ) {
403 xlat_new_cache_ptr = xlat_new_cache;
404 }
406 if( xlat_new_cache_ptr->active ) {
407 xlat_promote_to_temp_space( xlat_new_cache_ptr );
408 }
409 xlat_new_create_ptr = xlat_new_cache_ptr;
410 xlat_new_create_ptr->active = 1;
411 xlat_new_cache_ptr = NEXT(xlat_new_cache_ptr);
413 /* Add the LUT entry for the block */
414 if( xlat_lut[XLAT_LUT_PAGE(address)] == NULL ) {
415 xlat_lut[XLAT_LUT_PAGE(address)] =
416 mmap( NULL, XLAT_LUT_PAGE_SIZE, PROT_READ|PROT_WRITE,
417 MAP_PRIVATE|MAP_ANON, -1, 0 );
418 memset( xlat_lut[XLAT_LUT_PAGE(address)], 0, XLAT_LUT_PAGE_SIZE );
419 }
421 if( IS_ENTRY_POINT(xlat_lut[XLAT_LUT_PAGE(address)][XLAT_LUT_ENTRY(address)]) ) {
422 xlat_cache_block_t oldblock = XLAT_BLOCK_FOR_CODE(xlat_lut[XLAT_LUT_PAGE(address)][XLAT_LUT_ENTRY(address)]);
423 oldblock->active = 0;
424 }
426 xlat_lut[XLAT_LUT_PAGE(address)][XLAT_LUT_ENTRY(address)] =
427 &xlat_new_create_ptr->code;
428 xlat_new_create_ptr->lut_entry = xlat_lut[XLAT_LUT_PAGE(address)] + XLAT_LUT_ENTRY(address);
430 return xlat_new_create_ptr;
431 }
433 xlat_cache_block_t xlat_extend_block( uint32_t newSize )
434 {
435 while( xlat_new_create_ptr->size < newSize ) {
436 if( xlat_new_cache_ptr->size == 0 ) {
437 /* Migrate to the front of the cache to keep it contiguous */
438 xlat_new_create_ptr->active = 0;
439 sh4ptr_t olddata = xlat_new_create_ptr->code;
440 int oldsize = xlat_new_create_ptr->size;
441 int size = oldsize + MIN_BLOCK_SIZE; /* minimum expansion */
442 void **lut_entry = xlat_new_create_ptr->lut_entry;
443 int allocation = (int)-sizeof(struct xlat_cache_block);
444 xlat_new_cache_ptr = xlat_new_cache;
445 do {
446 if( xlat_new_cache_ptr->active ) {
447 xlat_promote_to_temp_space( xlat_new_cache_ptr );
448 }
449 allocation += xlat_new_cache_ptr->size + sizeof(struct xlat_cache_block);
450 xlat_new_cache_ptr = NEXT(xlat_new_cache_ptr);
451 } while( allocation < size );
452 xlat_new_create_ptr = xlat_new_cache;
453 xlat_new_create_ptr->active = 1;
454 xlat_new_create_ptr->size = allocation;
455 xlat_new_create_ptr->lut_entry = lut_entry;
456 *lut_entry = &xlat_new_create_ptr->code;
457 memmove( xlat_new_create_ptr->code, olddata, oldsize );
458 } else {
459 if( xlat_new_cache_ptr->active ) {
460 xlat_promote_to_temp_space( xlat_new_cache_ptr );
461 }
462 xlat_new_create_ptr->size += xlat_new_cache_ptr->size + sizeof(struct xlat_cache_block);
463 xlat_new_cache_ptr = NEXT(xlat_new_cache_ptr);
464 }
465 }
466 return xlat_new_create_ptr;
468 }
470 void xlat_commit_block( uint32_t destsize, uint32_t srcsize )
471 {
472 void **ptr = xlat_new_create_ptr->lut_entry;
473 void **endptr = ptr + (srcsize>>2);
474 while( ptr < endptr ) {
475 if( *ptr == NULL ) {
476 *ptr = XLAT_LUT_ENTRY_USED;
477 }
478 ptr++;
479 }
481 xlat_new_cache_ptr = xlat_cut_block( xlat_new_create_ptr, destsize );
482 }
484 void xlat_delete_block( xlat_cache_block_t block )
485 {
486 block->active = 0;
487 *block->lut_entry = NULL;
488 }
490 void xlat_check_cache_integrity( xlat_cache_block_t cache, xlat_cache_block_t ptr, int size )
491 {
492 int foundptr = 0;
493 xlat_cache_block_t tail =
494 (xlat_cache_block_t)(((char *)cache) + size - sizeof(struct xlat_cache_block));
496 assert( tail->active == 1 );
497 assert( tail->size == 0 );
498 while( cache < tail ) {
499 assert( cache->active >= 0 && cache->active <= 2 );
500 assert( cache->size >= 0 && cache->size < size );
501 if( cache == ptr ) {
502 foundptr = 1;
503 }
504 cache = NEXT(cache);
505 }
506 assert( cache == tail );
507 assert( foundptr == 1 || tail == ptr );
508 }
510 void xlat_check_integrity( )
511 {
512 xlat_check_cache_integrity( xlat_new_cache, xlat_new_cache_ptr, XLAT_NEW_CACHE_SIZE );
513 xlat_check_cache_integrity( xlat_temp_cache, xlat_temp_cache_ptr, XLAT_TEMP_CACHE_SIZE );
514 xlat_check_cache_integrity( xlat_old_cache, xlat_old_cache_ptr, XLAT_OLD_CACHE_SIZE );
515 }
.