4 * Translation cache management. This part is architecture independent.
6 * Copyright (c) 2005 Nathan Keynes.
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.
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.
19 #include <sys/types.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
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;
55 #ifdef XLAT_GENERATIONAL_CACHE
56 xlat_cache_block_t xlat_temp_cache;
57 xlat_cache_block_t xlat_temp_cache_ptr;
58 xlat_cache_block_t xlat_old_cache;
59 xlat_cache_block_t xlat_old_cache_ptr;
62 static void ***xlat_lut;
63 static gboolean xlat_initialized = FALSE;
65 void xlat_cache_init(void)
67 if( !xlat_initialized ) {
68 xlat_initialized = TRUE;
69 xlat_new_cache = mmap( NULL, XLAT_NEW_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_new_create_ptr = xlat_new_cache;
73 #ifdef XLAT_GENERATIONAL_CACHE
74 xlat_temp_cache = mmap( NULL, XLAT_TEMP_CACHE_SIZE, PROT_EXEC|PROT_READ|PROT_WRITE,
75 MAP_PRIVATE|MAP_ANON, -1, 0 );
76 xlat_old_cache = mmap( NULL, XLAT_OLD_CACHE_SIZE, PROT_EXEC|PROT_READ|PROT_WRITE,
77 MAP_PRIVATE|MAP_ANON, -1, 0 );
78 xlat_temp_cache_ptr = xlat_temp_cache;
79 xlat_old_cache_ptr = xlat_old_cache;
81 xlat_lut = mmap( NULL, XLAT_LUT_PAGES*sizeof(void *), PROT_READ|PROT_WRITE,
82 MAP_PRIVATE|MAP_ANON, -1, 0);
83 memset( xlat_lut, 0, XLAT_LUT_PAGES*sizeof(void *) );
89 * Reset the cache structure to its default state
91 void xlat_flush_cache()
93 xlat_cache_block_t tmp;
95 xlat_new_cache_ptr = xlat_new_cache;
96 xlat_new_cache_ptr->active = 0;
97 xlat_new_cache_ptr->size = XLAT_NEW_CACHE_SIZE - 2*sizeof(struct xlat_cache_block);
98 tmp = NEXT(xlat_new_cache_ptr);
101 #ifdef XLAT_GENERATIONAL_CACHE
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);
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);
115 for( i=0; i<XLAT_LUT_PAGES; i++ ) {
116 if( xlat_lut[i] != NULL ) {
117 memset( xlat_lut[i], 0, XLAT_LUT_PAGE_SIZE );
122 static void xlat_flush_page_by_lut( void **page )
125 for( i=0; i<XLAT_LUT_PAGE_ENTRIES; i++ ) {
126 if( IS_ENTRY_POINT(page[i]) ) {
127 XLAT_BLOCK_FOR_CODE(page[i])->active = 0;
133 void FASTCALL xlat_invalidate_word( sh4addr_t addr )
136 void **page = xlat_lut[XLAT_LUT_PAGE(addr)];
138 int entry = XLAT_LUT_ENTRY(addr);
139 if( page[entry] != NULL ) {
140 xlat_flush_page_by_lut(page);
146 void FASTCALL xlat_invalidate_long( sh4addr_t addr )
149 void **page = xlat_lut[XLAT_LUT_PAGE(addr)];
151 int entry = XLAT_LUT_ENTRY(addr);
152 if( page[entry] != NULL || page[entry+1] != NULL ) {
153 xlat_flush_page_by_lut(page);
159 void FASTCALL xlat_invalidate_block( sh4addr_t address, size_t size )
162 int entry_count = size >> 1; // words;
163 uint32_t page_no = XLAT_LUT_PAGE(address);
164 int entry = XLAT_LUT_ENTRY(address);
167 void **page = xlat_lut[page_no];
168 int page_entries = XLAT_LUT_PAGE_ENTRIES - entry;
169 if( entry_count < page_entries ) {
170 page_entries = entry_count;
173 if( page_entries == XLAT_LUT_PAGE_ENTRIES ) {
174 /* Overwriting the entire page anyway */
175 xlat_flush_page_by_lut(page);
177 for( i=entry; i<entry+page_entries; i++ ) {
178 if( page[i] != NULL ) {
179 xlat_flush_page_by_lut(page);
184 entry_count -= page_entries;
187 entry_count -= page_entries;
189 } while( entry_count > 0 );
193 void FASTCALL xlat_flush_page( sh4addr_t address )
195 void **page = xlat_lut[XLAT_LUT_PAGE(address)];
197 xlat_flush_page_by_lut(page);
201 void * FASTCALL xlat_get_code( sh4addr_t address )
204 void **page = xlat_lut[XLAT_LUT_PAGE(address)];
206 result = (void *)(((uintptr_t)(page[XLAT_LUT_ENTRY(address)])) & (~((uintptr_t)0x03)));
211 xlat_recovery_record_t xlat_get_post_recovery( void *code, void *native_pc, gboolean with_terminal )
214 uintptr_t pc_offset = ((uint8_t *)native_pc) - ((uint8_t *)code);
215 xlat_cache_block_t block = XLAT_BLOCK_FOR_CODE(code);
216 uint32_t count = block->recover_table_size;
217 xlat_recovery_record_t records = (xlat_recovery_record_t)(&block->code[block->recover_table_offset]);
219 if( count > 0 && !with_terminal )
221 if( records[count-1].xlat_offset < pc_offset ) {
224 for( posn=count-1; posn > 0; posn-- ) {
225 if( records[posn-1].xlat_offset < pc_offset ) {
226 return &records[posn];
229 return &records[0]; // shouldn't happen
234 xlat_recovery_record_t xlat_get_pre_recovery( void *code, void *native_pc )
237 uintptr_t pc_offset = ((uint8_t *)native_pc) - ((uint8_t *)code);
238 xlat_cache_block_t block = XLAT_BLOCK_FOR_CODE(code);
239 uint32_t count = block->recover_table_size;
240 xlat_recovery_record_t records = (xlat_recovery_record_t)(&block->code[block->recover_table_offset]);
242 for( posn = 1; posn < count; posn++ ) {
243 if( records[posn].xlat_offset >= pc_offset ) {
244 return &records[posn-1];
247 return &records[count-1];
252 void ** FASTCALL xlat_get_lut_entry( sh4addr_t address )
254 void **page = xlat_lut[XLAT_LUT_PAGE(address)];
256 /* Add the LUT entry for the block */
258 xlat_lut[XLAT_LUT_PAGE(address)] = page =
259 mmap( NULL, XLAT_LUT_PAGE_SIZE, PROT_READ|PROT_WRITE,
260 MAP_PRIVATE|MAP_ANON, -1, 0 );
261 memset( page, 0, XLAT_LUT_PAGE_SIZE );
264 return &page[XLAT_LUT_ENTRY(address)];
269 uint32_t FASTCALL xlat_get_block_size( void *block )
271 xlat_cache_block_t xlt = (xlat_cache_block_t)(((char *)block)-sizeof(struct xlat_cache_block));
275 uint32_t FASTCALL xlat_get_code_size( void *block )
277 xlat_cache_block_t xlt = (xlat_cache_block_t)(((char *)block)-sizeof(struct xlat_cache_block));
278 if( xlt->recover_table_offset == 0 ) {
281 return xlt->recover_table_offset;
286 * Cut the specified block so that it has the given size, with the remaining data
287 * forming a new free block. If the free block would be less than the minimum size,
288 * the cut is not performed.
289 * @return the next block after the (possibly cut) block.
291 static inline xlat_cache_block_t xlat_cut_block( xlat_cache_block_t block, int cutsize )
293 cutsize = (cutsize + 3) & 0xFFFFFFFC; // force word alignment
294 assert( cutsize <= block->size );
295 if( block->size > cutsize + MIN_TOTAL_SIZE ) {
296 int oldsize = block->size;
297 block->size = cutsize;
298 xlat_cache_block_t next = NEXT(block);
300 next->size = oldsize - cutsize - sizeof(struct xlat_cache_block);
307 #ifdef XLAT_GENERATIONAL_CACHE
309 * Promote a block in temp space (or elsewhere for that matter) to old space.
311 * @param block to promote.
313 static void xlat_promote_to_old_space( xlat_cache_block_t block )
315 int allocation = (int)-sizeof(struct xlat_cache_block);
316 int size = block->size;
317 xlat_cache_block_t curr = xlat_old_cache_ptr;
318 xlat_cache_block_t start_block = curr;
320 allocation += curr->size + sizeof(struct xlat_cache_block);
322 if( allocation > size ) {
325 if( curr->size == 0 ) { /* End-of-cache Sentinel */
326 /* Leave what we just released as free space and start again from the
329 start_block->active = 0;
330 start_block->size = allocation;
331 allocation = (int)-sizeof(struct xlat_cache_block);
332 start_block = curr = xlat_old_cache;
335 start_block->active = 1;
336 start_block->size = allocation;
337 start_block->lut_entry = block->lut_entry;
338 start_block->fpscr_mask = block->fpscr_mask;
339 start_block->fpscr = block->fpscr;
340 start_block->recover_table_offset = block->recover_table_offset;
341 start_block->recover_table_size = block->recover_table_size;
342 *block->lut_entry = &start_block->code;
343 memcpy( start_block->code, block->code, block->size );
344 xlat_old_cache_ptr = xlat_cut_block(start_block, size );
345 if( xlat_old_cache_ptr->size == 0 ) {
346 xlat_old_cache_ptr = xlat_old_cache;
351 * Similarly to the above method, promotes a block to temp space.
352 * TODO: Try to combine these - they're nearly identical
354 void xlat_promote_to_temp_space( xlat_cache_block_t block )
356 int size = block->size;
357 int allocation = (int)-sizeof(struct xlat_cache_block);
358 xlat_cache_block_t curr = xlat_temp_cache_ptr;
359 xlat_cache_block_t start_block = curr;
361 if( curr->active == BLOCK_USED ) {
362 xlat_promote_to_old_space( curr );
363 } else if( curr->active == BLOCK_ACTIVE ) {
364 // Active but not used, release block
365 *((uintptr_t *)curr->lut_entry) &= ((uintptr_t)0x03);
367 allocation += curr->size + sizeof(struct xlat_cache_block);
369 if( allocation > size ) {
372 if( curr->size == 0 ) { /* End-of-cache Sentinel */
373 /* Leave what we just released as free space and start again from the
376 start_block->active = 0;
377 start_block->size = allocation;
378 allocation = (int)-sizeof(struct xlat_cache_block);
379 start_block = curr = xlat_temp_cache;
382 start_block->active = 1;
383 start_block->size = allocation;
384 start_block->lut_entry = block->lut_entry;
385 start_block->fpscr_mask = block->fpscr_mask;
386 start_block->fpscr = block->fpscr;
387 start_block->recover_table_offset = block->recover_table_offset;
388 start_block->recover_table_size = block->recover_table_size;
389 *block->lut_entry = &start_block->code;
390 memcpy( start_block->code, block->code, block->size );
391 xlat_temp_cache_ptr = xlat_cut_block(start_block, size );
392 if( xlat_temp_cache_ptr->size == 0 ) {
393 xlat_temp_cache_ptr = xlat_temp_cache;
398 void xlat_promote_to_temp_space( xlat_cache_block_t block )
400 *block->lut_entry = 0;
405 * Returns the next block in the new cache list that can be written to by the
406 * translator. If the next block is active, it is evicted first.
408 xlat_cache_block_t xlat_start_block( sh4addr_t address )
410 if( xlat_new_cache_ptr->size == 0 ) {
411 xlat_new_cache_ptr = xlat_new_cache;
414 if( xlat_new_cache_ptr->active ) {
415 xlat_promote_to_temp_space( xlat_new_cache_ptr );
417 xlat_new_create_ptr = xlat_new_cache_ptr;
418 xlat_new_create_ptr->active = 1;
419 xlat_new_cache_ptr = NEXT(xlat_new_cache_ptr);
421 /* Add the LUT entry for the block */
422 if( xlat_lut[XLAT_LUT_PAGE(address)] == NULL ) {
423 xlat_lut[XLAT_LUT_PAGE(address)] =
424 mmap( NULL, XLAT_LUT_PAGE_SIZE, PROT_READ|PROT_WRITE,
425 MAP_PRIVATE|MAP_ANON, -1, 0 );
426 memset( xlat_lut[XLAT_LUT_PAGE(address)], 0, XLAT_LUT_PAGE_SIZE );
429 if( IS_ENTRY_POINT(xlat_lut[XLAT_LUT_PAGE(address)][XLAT_LUT_ENTRY(address)]) ) {
430 xlat_cache_block_t oldblock = XLAT_BLOCK_FOR_CODE(xlat_lut[XLAT_LUT_PAGE(address)][XLAT_LUT_ENTRY(address)]);
431 oldblock->active = 0;
434 xlat_lut[XLAT_LUT_PAGE(address)][XLAT_LUT_ENTRY(address)] =
435 &xlat_new_create_ptr->code;
436 xlat_new_create_ptr->lut_entry = xlat_lut[XLAT_LUT_PAGE(address)] + XLAT_LUT_ENTRY(address);
438 return xlat_new_create_ptr;
441 xlat_cache_block_t xlat_extend_block( uint32_t newSize )
443 while( xlat_new_create_ptr->size < newSize ) {
444 if( xlat_new_cache_ptr->size == 0 ) {
445 /* Migrate to the front of the cache to keep it contiguous */
446 xlat_new_create_ptr->active = 0;
447 sh4ptr_t olddata = xlat_new_create_ptr->code;
448 int oldsize = xlat_new_create_ptr->size;
449 int size = oldsize + MIN_BLOCK_SIZE; /* minimum expansion */
450 void **lut_entry = xlat_new_create_ptr->lut_entry;
451 int allocation = (int)-sizeof(struct xlat_cache_block);
452 xlat_new_cache_ptr = xlat_new_cache;
454 if( xlat_new_cache_ptr->active ) {
455 xlat_promote_to_temp_space( xlat_new_cache_ptr );
457 allocation += xlat_new_cache_ptr->size + sizeof(struct xlat_cache_block);
458 xlat_new_cache_ptr = NEXT(xlat_new_cache_ptr);
459 } while( allocation < size );
460 xlat_new_create_ptr = xlat_new_cache;
461 xlat_new_create_ptr->active = 1;
462 xlat_new_create_ptr->size = allocation;
463 xlat_new_create_ptr->lut_entry = lut_entry;
464 *lut_entry = &xlat_new_create_ptr->code;
465 memmove( xlat_new_create_ptr->code, olddata, oldsize );
467 if( xlat_new_cache_ptr->active ) {
468 xlat_promote_to_temp_space( xlat_new_cache_ptr );
470 xlat_new_create_ptr->size += xlat_new_cache_ptr->size + sizeof(struct xlat_cache_block);
471 xlat_new_cache_ptr = NEXT(xlat_new_cache_ptr);
474 return xlat_new_create_ptr;
478 void xlat_commit_block( uint32_t destsize, uint32_t srcsize )
480 void **ptr = xlat_new_create_ptr->lut_entry;
481 void **endptr = ptr + (srcsize>>2);
482 while( ptr < endptr ) {
484 *ptr = XLAT_LUT_ENTRY_USED;
489 xlat_new_cache_ptr = xlat_cut_block( xlat_new_create_ptr, destsize );
492 void xlat_delete_block( xlat_cache_block_t block )
495 *block->lut_entry = NULL;
498 void xlat_check_cache_integrity( xlat_cache_block_t cache, xlat_cache_block_t ptr, int size )
501 xlat_cache_block_t tail =
502 (xlat_cache_block_t)(((char *)cache) + size - sizeof(struct xlat_cache_block));
504 assert( tail->active == 1 );
505 assert( tail->size == 0 );
506 while( cache < tail ) {
507 assert( cache->active >= 0 && cache->active <= 2 );
508 assert( cache->size >= 0 && cache->size < size );
514 assert( cache == tail );
515 assert( foundptr == 1 || tail == ptr );
518 void xlat_check_integrity( )
520 xlat_check_cache_integrity( xlat_new_cache, xlat_new_cache_ptr, XLAT_NEW_CACHE_SIZE );
521 #ifdef XLAT_GENERATIONAL_CACHE
522 xlat_check_cache_integrity( xlat_temp_cache, xlat_temp_cache_ptr, XLAT_TEMP_CACHE_SIZE );
523 xlat_check_cache_integrity( xlat_old_cache, xlat_old_cache_ptr, XLAT_OLD_CACHE_SIZE );
.