Search
lxdream.org :: lxdream/src/sh4/sh4trans.c
lxdream 0.9.1
released Jun 29
Download Now
filename src/sh4/sh4trans.c
changeset 906:268ea359f884
prev905:4c17ebd9ef5e
next914:72abecf5a315
author nkeynes
date Thu Oct 30 05:50:21 2008 +0000 (15 years ago)
permissions -rw-r--r--
last change Fix x86-64 build (typos et al)
Remove Push/pop ebx - don't really need it and saves adding more target-specific asm
view annotate diff log raw
     1 /**
     2  * $Id$
     3  * 
     4  * SH4 translation core module. This part handles the non-target-specific
     5  * section of the translation.
     6  *
     7  * Copyright (c) 2005 Nathan Keynes.
     8  *
     9  * This program is free software; you can redistribute it and/or modify
    10  * it under the terms of the GNU General Public License as published by
    11  * the Free Software Foundation; either version 2 of the License, or
    12  * (at your option) any later version.
    13  *
    14  * This program is distributed in the hope that it will be useful,
    15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
    16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
    17  * GNU General Public License for more details.
    18  */
    19 #include <assert.h>
    20 #include "eventq.h"
    21 #include "syscall.h"
    22 #include "clock.h"
    23 #include "dreamcast.h"
    24 #include "sh4/sh4core.h"
    25 #include "sh4/sh4trans.h"
    26 #include "sh4/xltcache.h"
    29 /**
    30  * Execute a timeslice using translated code only (ie translate/execute loop)
    31  */
    32 uint32_t sh4_translate_run_slice( uint32_t nanosecs ) 
    33 {
    34     void * (*code)() = NULL;
    35     while( sh4r.slice_cycle < nanosecs ) {
    36         if( sh4r.event_pending <= sh4r.slice_cycle ) {
    37             if( sh4r.event_types & PENDING_EVENT ) {
    38                 event_execute();
    39             }
    40             /* Eventq execute may (quite likely) deliver an immediate IRQ */
    41             if( sh4r.event_types & PENDING_IRQ ) {
    42                 sh4_accept_interrupt();
    43                 code = NULL;
    44             }
    45         }
    47         if( code == NULL ) {
    48             if( sh4r.pc > 0xFFFFFF00 ) {
    49                 syscall_invoke( sh4r.pc );
    50                 sh4r.in_delay_slot = 0;
    51                 sh4r.pc = sh4r.pr;
    52             }
    54             code = xlat_get_code_by_vma( sh4r.pc );
    55             if( code == NULL || (sh4r.fpscr & (FPSCR_PR|FPSCR_SZ)) != XLAT_BLOCK_FPSCR(code) ) {
    56                 code = sh4_translate_basic_block( sh4r.pc );
    57             }
    58         }
    59         code = code();
    60     }
    61     return nanosecs;
    62 }
    64 uint8_t *xlat_output;
    65 xlat_cache_block_t xlat_current_block;
    66 struct xlat_recovery_record xlat_recovery[MAX_RECOVERY_SIZE];
    67 uint32_t xlat_recovery_posn;
    69 void sh4_translate_add_recovery( uint32_t icount )
    70 {
    71     xlat_recovery[xlat_recovery_posn].xlat_offset = 
    72         ((uintptr_t)xlat_output) - ((uintptr_t)xlat_current_block->code);
    73     xlat_recovery[xlat_recovery_posn].sh4_icount = icount;
    74     xlat_recovery_posn++;
    75 }
    77 /**
    78  * Translate a linear basic block, ie all instructions from the start address
    79  * (inclusive) until the next branch/jump instruction or the end of the page
    80  * is reached.
    81  * @return the address of the translated block
    82  * eg due to lack of buffer space.
    83  */
    84 void * sh4_translate_basic_block( sh4addr_t start )
    85 {
    86     sh4addr_t pc = start;
    87     sh4addr_t lastpc = (pc&0xFFFFF000)+0x1000;
    88     int done, i;
    89     xlat_current_block = xlat_start_block( start );
    90     xlat_output = (uint8_t *)xlat_current_block->code;
    91     xlat_recovery_posn = 0;
    92     uint8_t *eob = xlat_output + xlat_current_block->size;
    94     if( GET_ICACHE_END() < lastpc ) {
    95         lastpc = GET_ICACHE_END();
    96     }
    98     sh4_translate_begin_block(pc);
   100     do {
   101         /* check for breakpoints at this pc */
   102         for( i=0; i<sh4_breakpoint_count; i++ ) {
   103             if( sh4_breakpoints[i].address == pc ) {
   104                 sh4_translate_emit_breakpoint(pc);
   105                 break;
   106             }
   107         }
   108         if( eob - xlat_output < MAX_INSTRUCTION_SIZE ) {
   109             uint8_t *oldstart = xlat_current_block->code;
   110             xlat_current_block = xlat_extend_block( xlat_output - oldstart + MAX_INSTRUCTION_SIZE );
   111             xlat_output = xlat_current_block->code + (xlat_output - oldstart);
   112             eob = xlat_current_block->code + xlat_current_block->size;
   113         }
   114         done = sh4_translate_instruction( pc ); 
   115         assert( xlat_output <= eob );
   116         pc += 2;
   117         if ( pc >= lastpc ) {
   118             done = 2;
   119         }
   120     } while( !done );
   121     pc += (done - 2);
   123     // Add end-of-block recovery for post-instruction checks
   124     sh4_translate_add_recovery( (pc - start)>>1 ); 
   126     int epilogue_size = sh4_translate_end_block_size();
   127     uint32_t recovery_size = sizeof(struct xlat_recovery_record)*xlat_recovery_posn;
   128     uint32_t finalsize = (xlat_output - xlat_current_block->code) + epilogue_size + recovery_size;
   129     if( xlat_current_block->size < finalsize ) {
   130         uint8_t *oldstart = xlat_current_block->code;
   131         xlat_current_block = xlat_extend_block( finalsize );
   132         xlat_output = xlat_current_block->code + (xlat_output - oldstart);
   133     }	
   134     sh4_translate_end_block(pc);
   135     assert( xlat_output <= (xlat_current_block->code + xlat_current_block->size - recovery_size) );
   137     /* Write the recovery records onto the end of the code block */
   138     memcpy( xlat_output, xlat_recovery, recovery_size);
   139     xlat_current_block->recover_table_offset = xlat_output - (uint8_t *)xlat_current_block->code;
   140     xlat_current_block->recover_table_size = xlat_recovery_posn;
   141     xlat_current_block->fpscr = sh4r.fpscr & (FPSCR_PR|FPSCR_SZ);
   142     xlat_current_block->fpscr_mask = (FPSCR_PR|FPSCR_SZ);
   143     xlat_commit_block( finalsize, pc-start );
   144     return xlat_current_block->code;
   145 }
   147 /**
   148  * "Execute" the supplied recovery record. Currently this only updates
   149  * sh4r.pc and sh4r.slice_cycle according to the currently executing
   150  * instruction. In future this may be more sophisticated (ie will
   151  * call into generated code).
   152  */
   153 void sh4_translate_run_recovery( xlat_recovery_record_t recovery )
   154 {
   155     sh4r.slice_cycle += (recovery->sh4_icount * sh4_cpu_period);
   156     sh4r.pc += (recovery->sh4_icount<<1);
   157 }
   159 void sh4_translate_exit_recover( )
   160 {
   161     void *code = xlat_get_code_by_vma( sh4r.pc );
   162     if( code != NULL ) {
   163         uint32_t size = xlat_get_code_size( code );
   164         void *pc = xlat_get_native_pc( code, size );
   165         if( pc != NULL ) {
   166             // could be null if we're not actually running inside the translator
   167             xlat_recovery_record_t recover = xlat_get_post_recovery(code, pc, TRUE);
   168             if( recover != NULL ) {
   169                 // Can be null if there is no recovery necessary
   170                 sh4_translate_run_recovery(recover);
   171             }
   172         }
   173     }
   174 }
   176 void FASTCALL sh4_translate_breakpoint_hit(uint32_t pc)
   177 {
   178     if( sh4_starting && sh4r.slice_cycle == 0 && pc == sh4r.pc ) {
   179         return;
   180     }
   181     sh4_core_exit( CORE_EXIT_BREAKPOINT );
   182 }
   184 /**
   185  * Exit the current block at the end of the current instruction, flush the
   186  * translation cache (completely) and return control to sh4_xlat_run_slice.
   187  *
   188  * As a special case, if the current instruction is actually the last 
   189  * instruction in the block (ie it's in a delay slot), this function 
   190  * returns to allow normal completion of the translation block. Otherwise
   191  * this function never returns.
   192  *
   193  * Must only be invoked (indirectly) from within translated code.
   194  */
   195 gboolean sh4_translate_flush_cache()
   196 {
   197     void *code = xlat_get_code_by_vma( sh4r.pc );
   198     if( code != NULL ) {
   199         uint32_t size = xlat_get_code_size( code );
   200         void *pc = xlat_get_native_pc( code, size );
   201         assert( pc != NULL );
   203         xlat_recovery_record_t recover = xlat_get_post_recovery(code, pc, FALSE);
   204         if( recover != NULL ) {
   205             // Can be null if there is no recovery necessary
   206             sh4_translate_run_recovery(recover);
   207             xlat_flush_cache();
   208             return TRUE;
   209         } else {
   210             xlat_flush_cache();
   211             return FALSE;
   212         }
   213     }
   214 }
   216 void * FASTCALL xlat_get_code_by_vma( sh4vma_t vma )
   217 {
   218     void *result = NULL;
   220     if( IS_IN_ICACHE(vma) ) {
   221         return xlat_get_code( GET_ICACHE_PHYS(vma) );
   222     }
   224     if( vma > 0xFFFFFF00 ) {
   225         // lxdream hook
   226         return NULL;
   227     }
   229     if( !mmu_update_icache(vma) ) {
   230         // fault - off to the fault handler
   231         if( !mmu_update_icache(sh4r.pc) ) {
   232             // double fault - halt
   233             ERROR( "Double fault - halting" );
   234             sh4_core_exit(CORE_EXIT_HALT);
   235             return NULL;
   236         }
   237     }
   239     assert( IS_IN_ICACHE(sh4r.pc) );
   240     result = xlat_get_code( GET_ICACHE_PHYS(sh4r.pc) );
   241     return result;
   242 }
.