Search
lxdream.org :: lxdream/src/sh4/sh4trans.c
lxdream 0.9.1
released Jun 29
Download Now
filename src/sh4/sh4trans.c
changeset 669:ab344e42bca9
prev619:0800a0137472
next711:4f0ba72e58fe
author nkeynes
date Mon May 12 10:00:13 2008 +0000 (15 years ago)
permissions -rw-r--r--
last change Cleanup most of the -Wall warnings (getting a bit sloppy...)
Convert FP code to use fixed banks rather than indirect pointer
(3-4% faster this way now)
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 <setjmp.h>
    21 #include "eventq.h"
    22 #include "syscall.h"
    23 #include "clock.h"
    24 #include "dreamcast.h"
    25 #include "sh4/sh4core.h"
    26 #include "sh4/sh4trans.h"
    27 #include "sh4/xltcache.h"
    30 static jmp_buf xlat_jmp_buf;
    31 static gboolean xlat_running = FALSE;
    33 gboolean sh4_xlat_is_running()
    34 {
    35     return xlat_running;
    36 }
    38 /**
    39  * Execute a timeslice using translated code only (ie translate/execute loop)
    40  */
    41 uint32_t sh4_xlat_run_slice( uint32_t nanosecs ) 
    42 {
    43     sh4r.slice_cycle = 0;
    45     if( sh4r.sh4_state != SH4_STATE_RUNNING ) {
    46 	sh4_sleep_run_slice(nanosecs);
    47     }
    49     switch( setjmp(xlat_jmp_buf) ) {
    50     case XLAT_EXIT_BREAKPOINT:
    51 	sh4_clear_breakpoint( sh4r.pc, BREAK_ONESHOT );
    52 	/* fallthrough */
    53     case XLAT_EXIT_HALT:
    54 	if( sh4r.sh4_state != SH4_STATE_STANDBY ) {
    55 	    TMU_run_slice( sh4r.slice_cycle );
    56 	    SCIF_run_slice( sh4r.slice_cycle );
    57 	    dreamcast_stop();
    58 	    return sh4r.slice_cycle;
    59 	}
    60     case XLAT_EXIT_SYSRESET:
    61 	dreamcast_reset();
    62 	break;
    63     case XLAT_EXIT_SLEEP:
    64 	sh4_sleep_run_slice(nanosecs);
    65         break;	
    66     }
    68     xlat_running = TRUE;
    69     void * (*code)() = NULL;
    70     while( sh4r.slice_cycle < nanosecs ) {
    71 	if( sh4r.event_pending <= sh4r.slice_cycle ) {
    72 	    if( sh4r.event_types & PENDING_EVENT ) {
    73 		event_execute();
    74 	    }
    75 	    /* Eventq execute may (quite likely) deliver an immediate IRQ */
    76 	    if( sh4r.event_types & PENDING_IRQ ) {
    77 		sh4_accept_interrupt();
    78 		code = NULL;
    79 	    }
    80 	}
    82 	if( code == NULL ) {
    83 	    if( sh4r.pc > 0xFFFFFF00 ) {
    84 		syscall_invoke( sh4r.pc );
    85 		sh4r.in_delay_slot = 0;
    86 		sh4r.pc = sh4r.pr;
    87 	    }
    89 	    code = xlat_get_code_by_vma( sh4r.pc );
    90 	    if( code == NULL ) {
    91 		code = sh4_translate_basic_block( sh4r.pc );
    92 	    }
    93 	}
    94 	code = code();
    95     }
    97     xlat_running = FALSE;
    98     sh4_starting = FALSE;
    99     sh4r.slice_cycle = nanosecs;
   100     if( sh4r.sh4_state != SH4_STATE_STANDBY ) {
   101 	TMU_run_slice( nanosecs );
   102 	SCIF_run_slice( nanosecs );
   103     }
   104     return nanosecs;
   105 }
   107 uint8_t *xlat_output;
   108 xlat_cache_block_t xlat_current_block;
   109 struct xlat_recovery_record xlat_recovery[MAX_RECOVERY_SIZE];
   110 uint32_t xlat_recovery_posn;
   112 void sh4_translate_add_recovery( uint32_t icount )
   113 {
   114     xlat_recovery[xlat_recovery_posn].xlat_offset = 
   115 	((uintptr_t)xlat_output) - ((uintptr_t)xlat_current_block->code);
   116     xlat_recovery[xlat_recovery_posn].sh4_icount = icount;
   117     xlat_recovery_posn++;
   118 }
   120 /**
   121  * Translate a linear basic block, ie all instructions from the start address
   122  * (inclusive) until the next branch/jump instruction or the end of the page
   123  * is reached.
   124  * @return the address of the translated block
   125  * eg due to lack of buffer space.
   126  */
   127 void * sh4_translate_basic_block( sh4addr_t start )
   128 {
   129     sh4addr_t pc = start;
   130     sh4addr_t lastpc = (pc&0xFFFFF000)+0x1000;
   131     int done, i;
   132     xlat_current_block = xlat_start_block( start );
   133     xlat_output = (uint8_t *)xlat_current_block->code;
   134     xlat_recovery_posn = 0;
   135     uint8_t *eob = xlat_output + xlat_current_block->size;
   137     if( GET_ICACHE_END() < lastpc ) {
   138 	lastpc = GET_ICACHE_END();
   139     }
   141     sh4_translate_begin_block(pc);
   143     do {
   144 	/* check for breakpoints at this pc */
   145 	for( i=0; i<sh4_breakpoint_count; i++ ) {
   146 	    if( sh4_breakpoints[i].address == pc ) {
   147 		sh4_translate_emit_breakpoint(pc);
   148 		break;
   149 	    }
   150 	}
   151 	if( eob - xlat_output < MAX_INSTRUCTION_SIZE ) {
   152 	    uint8_t *oldstart = xlat_current_block->code;
   153 	    xlat_current_block = xlat_extend_block( xlat_output - oldstart + MAX_INSTRUCTION_SIZE );
   154 	    xlat_output = xlat_current_block->code + (xlat_output - oldstart);
   155 	    eob = xlat_current_block->code + xlat_current_block->size;
   156 	}
   157 	done = sh4_translate_instruction( pc ); 
   158 	assert( xlat_output <= eob );
   159 	pc += 2;
   160 	if ( pc >= lastpc ) {
   161 	    done = 2;
   162 	}
   163     } while( !done );
   164     pc += (done - 2);
   166     // Add end-of-block recovery for post-instruction checks
   167     sh4_translate_add_recovery( (pc - start)>>1 ); 
   169     int epilogue_size = sh4_translate_end_block_size();
   170     uint32_t recovery_size = sizeof(struct xlat_recovery_record)*xlat_recovery_posn;
   171     uint32_t finalsize = xlat_output - xlat_current_block->code + epilogue_size + recovery_size;
   172     if( eob - xlat_output < finalsize ) {
   173 	uint8_t *oldstart = xlat_current_block->code;
   174 	xlat_current_block = xlat_extend_block( finalsize );
   175 	xlat_output = xlat_current_block->code + (xlat_output - oldstart);
   176     }	
   177     sh4_translate_end_block(pc);
   179     /* Write the recovery records onto the end of the code block */
   180     memcpy( xlat_output, xlat_recovery, recovery_size);
   181     xlat_current_block->recover_table_offset = xlat_output - (uint8_t *)xlat_current_block->code;
   182     xlat_current_block->recover_table_size = xlat_recovery_posn;
   183     xlat_commit_block( finalsize, pc-start );
   184     return xlat_current_block->code;
   185 }
   187 /**
   188  * "Execute" the supplied recovery record. Currently this only updates
   189  * sh4r.pc and sh4r.slice_cycle according to the currently executing
   190  * instruction. In future this may be more sophisticated (ie will
   191  * call into generated code).
   192  */
   193 void sh4_translate_run_recovery( xlat_recovery_record_t recovery )
   194 {
   195     sh4r.slice_cycle += (recovery->sh4_icount * sh4_cpu_period);
   196     sh4r.pc += (recovery->sh4_icount<<1);
   197 }
   199 void sh4_translate_unwind_stack( gboolean abort_after, unwind_thunk_t thunk )
   200 {
   201     void *pc = xlat_get_native_pc();
   203     assert( pc != NULL );
   204     void *code = xlat_get_code( sh4r.pc );
   205     xlat_recovery_record_t recover = xlat_get_recovery(code, pc, TRUE);
   206     if( recover != NULL ) {
   207 	// Can be null if there is no recovery necessary
   208 	sh4_translate_run_recovery(recover);
   209     }
   210     if( thunk != NULL ) {
   211 	thunk();
   212     }
   213     // finally longjmp back into sh4_xlat_run_slice
   214     xlat_running = FALSE;
   215     longjmp(xlat_jmp_buf, XLAT_EXIT_CONTINUE);
   216 } 
   218 void sh4_translate_exit( int exit_code )
   219 {
   220     void *pc = xlat_get_native_pc();
   221     if( pc != NULL ) {
   222 	// could be null if we're not actually running inside the translator
   223 	void *code = xlat_get_code( sh4r.pc );
   224 	xlat_recovery_record_t recover = xlat_get_recovery(code, pc, TRUE);
   225 	if( recover != NULL ) {
   226 	    // Can be null if there is no recovery necessary
   227 	    sh4_translate_run_recovery(recover);
   228 	}
   229     }
   230     // finally longjmp back into sh4_xlat_run_slice
   231     xlat_running = FALSE;
   232     longjmp(xlat_jmp_buf, exit_code);
   233 }
   235 void sh4_translate_breakpoint_hit(uint32_t pc)
   236 {
   237     if( sh4_starting && sh4r.slice_cycle == 0 && pc == sh4r.pc ) {
   238 	return;
   239     }
   240     sh4_translate_exit( XLAT_EXIT_BREAKPOINT );
   241 }
   243 /**
   244  * Exit the current block at the end of the current instruction, flush the
   245  * translation cache (completely) and return control to sh4_xlat_run_slice.
   246  *
   247  * As a special case, if the current instruction is actually the last 
   248  * instruction in the block (ie it's in a delay slot), this function 
   249  * returns to allow normal completion of the translation block. Otherwise
   250  * this function never returns.
   251  *
   252  * Must only be invoked (indirectly) from within translated code.
   253  */
   254 void sh4_translate_flush_cache()
   255 {
   256     void *pc = xlat_get_native_pc();
   257     assert( pc != NULL );
   259     void *code = xlat_get_code( sh4r.pc );
   260     xlat_recovery_record_t recover = xlat_get_recovery(code, pc, TRUE);
   261     if( recover != NULL ) {
   262 	// Can be null if there is no recovery necessary
   263 	sh4_translate_run_recovery(recover);
   264 	xlat_flush_cache();
   265 	xlat_running = FALSE;
   266 	longjmp(xlat_jmp_buf, XLAT_EXIT_CONTINUE);
   267     } else {
   268 	xlat_flush_cache();
   269 	return;
   270     }
   271 }
   273 void *xlat_get_code_by_vma( sh4vma_t vma )
   274 {
   275     void *result = NULL;
   277     if( IS_IN_ICACHE(vma) ) {
   278 	return xlat_get_code( GET_ICACHE_PHYS(vma) );
   279     }
   281     if( vma > 0xFFFFFF00 ) {
   282 	// lxdream hook
   283 	return NULL;
   284     }
   286     if( !mmu_update_icache(vma) ) {
   287 	// fault - off to the fault handler
   288 	if( !mmu_update_icache(sh4r.pc) ) {
   289 	    // double fault - halt
   290 	    ERROR( "Double fault - halting" );
   291 	    dreamcast_stop();
   292 	    return NULL;
   293 	}
   294     }
   296     assert( IS_IN_ICACHE(sh4r.pc) );
   297     result = xlat_get_code( GET_ICACHE_PHYS(sh4r.pc) );
   298     return result;
   299 }
.