nkeynes@539: /**
nkeynes@561:  * $Id$
nkeynes@539:  * 
nkeynes@539:  * Provides the implementation for the ia32 ABI (eg prologue, epilogue, and
nkeynes@539:  * calling conventions)
nkeynes@539:  *
nkeynes@539:  * Copyright (c) 2007 Nathan Keynes.
nkeynes@539:  *
nkeynes@539:  * This program is free software; you can redistribute it and/or modify
nkeynes@539:  * it under the terms of the GNU General Public License as published by
nkeynes@539:  * the Free Software Foundation; either version 2 of the License, or
nkeynes@539:  * (at your option) any later version.
nkeynes@539:  *
nkeynes@539:  * This program is distributed in the hope that it will be useful,
nkeynes@539:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
nkeynes@539:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
nkeynes@539:  * GNU General Public License for more details.
nkeynes@539:  */
nkeynes@539: 
nkeynes@539: #ifndef __lxdream_ia32abi_H
nkeynes@539: #define __lxdream_ia32abi_H 1
nkeynes@539: 
nkeynes@539: #define load_ptr( reg, ptr ) load_imm32( reg, (uint32_t)ptr );
nkeynes@539: 
nkeynes@539: /**
nkeynes@539:  * Note: clobbers EAX to make the indirect call - this isn't usually
nkeynes@539:  * a problem since the callee will usually clobber it anyway.
nkeynes@539:  */
nkeynes@539: #define CALL_FUNC0_SIZE 13
nkeynes@539: static inline void call_func0( void *ptr )
nkeynes@539: {
nkeynes@539:     int adj = (-sh4_x86.stack_posn)&0x0F;
nkeynes@539:     SUB_imm8s_r32( adj, R_ESP );
nkeynes@539:     load_imm32(R_EAX, (uint32_t)ptr);
nkeynes@539:     CALL_r32(R_EAX);
nkeynes@539:     ADD_imm8s_r32( adj, R_ESP );
nkeynes@539: }
nkeynes@539: 
nkeynes@539: #define CALL_FUNC1_SIZE 14
nkeynes@539: static inline void call_func1( void *ptr, int arg1 )
nkeynes@539: {
nkeynes@539:     int adj = (-4-sh4_x86.stack_posn)&0x0F;
nkeynes@539:     SUB_imm8s_r32( adj, R_ESP );
nkeynes@539:     PUSH_r32(arg1);
nkeynes@539:     load_imm32(R_EAX, (uint32_t)ptr);
nkeynes@539:     CALL_r32(R_EAX);
nkeynes@539:     ADD_imm8s_r32( adj+4, R_ESP );
nkeynes@539:     sh4_x86.stack_posn -= 4;
nkeynes@539: }
nkeynes@539: 
nkeynes@539: #define CALL_FUNC2_SIZE 15
nkeynes@539: static inline void call_func2( void *ptr, int arg1, int arg2 )
nkeynes@539: {
nkeynes@539:     int adj = (-8-sh4_x86.stack_posn)&0x0F;
nkeynes@539:     SUB_imm8s_r32( adj, R_ESP );
nkeynes@539:     PUSH_r32(arg2);
nkeynes@539:     PUSH_r32(arg1);
nkeynes@539:     load_imm32(R_EAX, (uint32_t)ptr);
nkeynes@539:     CALL_r32(R_EAX);
nkeynes@539:     ADD_imm8s_r32( adj+8, R_ESP );
nkeynes@539:     sh4_x86.stack_posn -= 8;
nkeynes@539: }
nkeynes@539: 
nkeynes@539: /**
nkeynes@539:  * Write a double (64-bit) value into memory, with the first word in arg2a, and
nkeynes@539:  * the second in arg2b
nkeynes@539:  * NB: 30 bytes
nkeynes@539:  */
nkeynes@539: #define MEM_WRITE_DOUBLE_SIZE 36
nkeynes@539: static inline void MEM_WRITE_DOUBLE( int addr, int arg2a, int arg2b )
nkeynes@539: {
nkeynes@539:     int adj = (-8-sh4_x86.stack_posn)&0x0F;
nkeynes@539:     SUB_imm8s_r32( adj, R_ESP );
nkeynes@539:     ADD_imm8s_r32( 4, addr );
nkeynes@539:     PUSH_r32(arg2b);
nkeynes@539:     PUSH_r32(addr);
nkeynes@539:     ADD_imm8s_r32( -4, addr );
nkeynes@539:     SUB_imm8s_r32( 8, R_ESP );
nkeynes@539:     PUSH_r32(arg2a);
nkeynes@539:     PUSH_r32(addr);
nkeynes@539:     load_imm32(R_EAX, (uint32_t)sh4_write_long);
nkeynes@539:     CALL_r32(R_EAX);
nkeynes@539:     ADD_imm8s_r32( 16, R_ESP );
nkeynes@539:     load_imm32(R_EAX, (uint32_t)sh4_write_long);
nkeynes@539:     CALL_r32(R_EAX);
nkeynes@539:     ADD_imm8s_r32( adj+8, R_ESP );
nkeynes@539:     sh4_x86.stack_posn -= 16;
nkeynes@539: }
nkeynes@539: 
nkeynes@539: /**
nkeynes@539:  * Read a double (64-bit) value from memory, writing the first word into arg2a
nkeynes@539:  * and the second into arg2b. The addr must not be in EAX
nkeynes@539:  * NB: 27 bytes
nkeynes@539:  */
nkeynes@539: #define MEM_READ_DOUBLE_SIZE 36
nkeynes@539: static inline void MEM_READ_DOUBLE( int addr, int arg2a, int arg2b )
nkeynes@539: {
nkeynes@539:     int adj = (-4-sh4_x86.stack_posn)&0x0F;
nkeynes@539:     int adj2 = (-8-sh4_x86.stack_posn)&0x0F;
nkeynes@539:     SUB_imm8s_r32( adj, R_ESP );
nkeynes@539:     PUSH_r32(addr);
nkeynes@539:     load_imm32(R_EAX, (uint32_t)sh4_read_long);
nkeynes@539:     CALL_r32(R_EAX);
nkeynes@571:     POP_r32(R_ECX);
nkeynes@539:     SUB_imm8s_r32( adj2-adj, R_ESP );
nkeynes@539:     PUSH_r32(R_EAX);
nkeynes@571:     ADD_imm8s_r32( 4, R_ECX );
nkeynes@571:     PUSH_r32(R_ECX);
nkeynes@539:     load_imm32(R_EAX, (uint32_t)sh4_read_long);
nkeynes@539:     CALL_r32(R_EAX);
nkeynes@539:     ADD_imm8s_r32( 4, R_ESP );
nkeynes@539:     MOV_r32_r32( R_EAX, arg2b );
nkeynes@539:     POP_r32(arg2a);
nkeynes@539:     ADD_imm8s_r32( adj2, R_ESP );
nkeynes@539:     sh4_x86.stack_posn -= 4;
nkeynes@539: }
nkeynes@539: 
nkeynes@539: /**
nkeynes@539:  * Emit the 'start of block' assembly. Sets up the stack frame and save
nkeynes@539:  * SI/DI as required
nkeynes@539:  */
nkeynes@539: void sh4_translate_begin_block( sh4addr_t pc ) 
nkeynes@539: {
nkeynes@539:     PUSH_r32(R_EBP);
nkeynes@539:     /* mov &sh4r, ebp */
nkeynes@539:     load_ptr( R_EBP, &sh4r );
nkeynes@539:     
nkeynes@539:     sh4_x86.in_delay_slot = FALSE;
nkeynes@539:     sh4_x86.priv_checked = FALSE;
nkeynes@539:     sh4_x86.fpuen_checked = FALSE;
nkeynes@539:     sh4_x86.branch_taken = FALSE;
nkeynes@539:     sh4_x86.backpatch_posn = 0;
nkeynes@571:     sh4_x86.recovery_posn = 0;
nkeynes@539:     sh4_x86.block_start_pc = pc;
nkeynes@539:     sh4_x86.tstate = TSTATE_NONE;
nkeynes@571:     sh4_x86.tlb_on = IS_MMU_ENABLED();
nkeynes@539:     sh4_x86.stack_posn = 8;
nkeynes@539: }
nkeynes@539: 
nkeynes@539: /**
nkeynes@539:  * Exit the block with sh4r.pc already written
nkeynes@539:  * Bytes: 15
nkeynes@539:  */
nkeynes@569: void exit_block_pcset( sh4addr_t pc )
nkeynes@539: {
nkeynes@539:     load_imm32( R_ECX, ((pc - sh4_x86.block_start_pc)>>1)*sh4_cpu_period ); // 5
nkeynes@539:     ADD_r32_sh4r( R_ECX, REG_OFFSET(slice_cycle) );    // 6
nkeynes@539:     load_spreg( R_EAX, REG_OFFSET(pc) );
nkeynes@570:     if( sh4_x86.tlb_on ) {
nkeynes@570: 	call_func1(xlat_get_code_by_vma,R_EAX);
nkeynes@570:     } else {
nkeynes@570: 	call_func1(xlat_get_code,R_EAX);
nkeynes@570:     }
nkeynes@539:     POP_r32(R_EBP);
nkeynes@539:     RET();
nkeynes@539: }
nkeynes@539: 
nkeynes@571: #define EXIT_BLOCK_SIZE(pc)  (24 + (IS_IN_ICACHE(pc)?5:CALL_FUNC1_SIZE))
nkeynes@571: 
nkeynes@571: 
nkeynes@539: /**
nkeynes@539:  * Exit the block to an absolute PC
nkeynes@539:  */
nkeynes@539: void exit_block( sh4addr_t pc, sh4addr_t endpc )
nkeynes@539: {
nkeynes@539:     load_imm32( R_ECX, pc );                            // 5
nkeynes@539:     store_spreg( R_ECX, REG_OFFSET(pc) );               // 3
nkeynes@571:     if( IS_IN_ICACHE(pc) ) {
nkeynes@571: 	MOV_moff32_EAX( xlat_get_lut_entry(GET_ICACHE_PHYS(pc)) ); // 5
nkeynes@571:     } else if( sh4_x86.tlb_on ) {
nkeynes@571: 	call_func1(xlat_get_code_by_vma,R_ECX);
nkeynes@571:     } else {
nkeynes@571: 	call_func1(xlat_get_code,R_ECX);
nkeynes@571:     }
nkeynes@571:     AND_imm8s_r32( 0xFC, R_EAX ); // 3
nkeynes@571:     load_imm32( R_ECX, ((endpc - sh4_x86.block_start_pc)>>1)*sh4_cpu_period ); // 5
nkeynes@571:     ADD_r32_sh4r( R_ECX, REG_OFFSET(slice_cycle) );     // 6
nkeynes@571:     POP_r32(R_EBP);
nkeynes@571:     RET();
nkeynes@571: }
nkeynes@571: 
nkeynes@571: #define EXIT_BLOCK_REL_SIZE(pc)  (27 + (IS_IN_ICACHE(pc)?5:CALL_FUNC1_SIZE))
nkeynes@571: 
nkeynes@571: /**
nkeynes@571:  * Exit the block to a relative PC
nkeynes@571:  */
nkeynes@571: void exit_block_rel( sh4addr_t pc, sh4addr_t endpc )
nkeynes@571: {
nkeynes@571:     load_imm32( R_ECX, pc - sh4_x86.block_start_pc );   // 5
nkeynes@571:     ADD_sh4r_r32( R_PC, R_ECX );
nkeynes@571:     store_spreg( R_ECX, REG_OFFSET(pc) );               // 3
nkeynes@571:     if( IS_IN_ICACHE(pc) ) {
nkeynes@571: 	MOV_moff32_EAX( xlat_get_lut_entry(GET_ICACHE_PHYS(pc)) ); // 5
nkeynes@571:     } else if( sh4_x86.tlb_on ) {
nkeynes@571: 	call_func1(xlat_get_code_by_vma,R_ECX);
nkeynes@571:     } else {
nkeynes@571: 	call_func1(xlat_get_code,R_ECX);
nkeynes@571:     }
nkeynes@539:     AND_imm8s_r32( 0xFC, R_EAX ); // 3
nkeynes@539:     load_imm32( R_ECX, ((endpc - sh4_x86.block_start_pc)>>1)*sh4_cpu_period ); // 5
nkeynes@539:     ADD_r32_sh4r( R_ECX, REG_OFFSET(slice_cycle) );     // 6
nkeynes@539:     POP_r32(R_EBP);
nkeynes@539:     RET();
nkeynes@539: }
nkeynes@539: 
nkeynes@539: /**
nkeynes@539:  * Write the block trailer (exception handling block)
nkeynes@539:  */
nkeynes@539: void sh4_translate_end_block( sh4addr_t pc ) {
nkeynes@539:     if( sh4_x86.branch_taken == FALSE ) {
nkeynes@539: 	// Didn't exit unconditionally already, so write the termination here
nkeynes@571: 	exit_block_rel( pc, pc );
nkeynes@539:     }
nkeynes@539:     if( sh4_x86.backpatch_posn != 0 ) {
nkeynes@559: 	unsigned int i;
nkeynes@559: 	// Raise exception
nkeynes@539: 	uint8_t *end_ptr = xlat_output;
nkeynes@571: 	MOV_r32_r32( R_EDX, R_ECX );
nkeynes@539: 	ADD_r32_r32( R_EDX, R_ECX );
nkeynes@571: 	ADD_r32_sh4r( R_ECX, R_PC );
nkeynes@539: 	MOV_moff32_EAX( &sh4_cpu_period );
nkeynes@539: 	MUL_r32( R_EDX );
nkeynes@539: 	ADD_r32_sh4r( R_EAX, REG_OFFSET(slice_cycle) );
nkeynes@539: 
nkeynes@559:         POP_r32(R_EDX);
nkeynes@559:         call_func1( sh4_raise_exception, R_EDX );
nkeynes@571: 	load_spreg( R_EAX, R_PC );
nkeynes@570: 	if( sh4_x86.tlb_on ) {
nkeynes@570: 	    call_func1(xlat_get_code_by_vma,R_EAX);
nkeynes@570: 	} else {
nkeynes@570: 	    call_func1(xlat_get_code,R_EAX);
nkeynes@570: 	}
nkeynes@539: 	POP_r32(R_EBP);
nkeynes@539: 	RET();
nkeynes@539: 
nkeynes@559: 	// Exception already raised - just cleanup
nkeynes@559: 	uint8_t *preexc_ptr = xlat_output;
nkeynes@571: 	MOV_r32_r32( R_EDX, R_ECX );
nkeynes@559: 	ADD_r32_r32( R_EDX, R_ECX );
nkeynes@571: 	ADD_r32_sh4r( R_ECX, R_SPC );
nkeynes@559: 	MOV_moff32_EAX( &sh4_cpu_period );
nkeynes@559: 	MUL_r32( R_EDX );
nkeynes@559: 	ADD_r32_sh4r( R_EAX, REG_OFFSET(slice_cycle) );
nkeynes@571: 	load_spreg( R_EAX, R_PC );
nkeynes@570: 	if( sh4_x86.tlb_on ) {
nkeynes@570: 	    call_func1(xlat_get_code_by_vma,R_EAX);
nkeynes@570: 	} else {
nkeynes@570: 	    call_func1(xlat_get_code,R_EAX);
nkeynes@570: 	}
nkeynes@559: 	POP_r32(R_EBP);
nkeynes@559: 	RET();
nkeynes@559: 
nkeynes@559: 	for( i=0; i< sh4_x86.backpatch_posn; i++ ) {
nkeynes@559: 	    *sh4_x86.backpatch_list[i].fixup_addr =
nkeynes@559: 		xlat_output - ((uint8_t *)sh4_x86.backpatch_list[i].fixup_addr) - 4;
nkeynes@559: 	    if( sh4_x86.backpatch_list[i].exc_code == -1 ) {
nkeynes@559: 		load_imm32( R_EDX, sh4_x86.backpatch_list[i].fixup_icount );
nkeynes@559: 		int rel = preexc_ptr - xlat_output;
nkeynes@559: 		JMP_rel(rel);
nkeynes@559: 	    } else {
nkeynes@559: 		PUSH_imm32( sh4_x86.backpatch_list[i].exc_code );
nkeynes@559: 		load_imm32( R_EDX, sh4_x86.backpatch_list[i].fixup_icount );
nkeynes@559: 		int rel = end_ptr - xlat_output;
nkeynes@559: 		JMP_rel(rel);
nkeynes@559: 	    }
nkeynes@559: 	}
nkeynes@539:     }
nkeynes@539: }
nkeynes@539: 
nkeynes@577: void *xlat_get_native_pc()
nkeynes@577: {
nkeynes@577:     void *result = NULL;
nkeynes@577:     asm(
nkeynes@577: 	"mov %%ebp, %%eax\n\t"
nkeynes@577: 	"mov $0x8, %%ecx\n\t"
nkeynes@577: 	"mov %1, %%edx\n"
nkeynes@577: "frame_loop: test %%eax, %%eax\n\t"
nkeynes@577: 	"je frame_not_found\n\t"
nkeynes@577: 	"cmp (%%eax), %%edx\n\t"
nkeynes@577: 	"je frame_found\n\t"
nkeynes@577: 	"sub $0x1, %%ecx\n\t"
nkeynes@577: 	"je frame_not_found\n\t"
nkeynes@577: 	"movl (%%eax), %%eax\n\t"
nkeynes@577: 	"jmp frame_loop\n"
nkeynes@577: "frame_found: movl 0x4(%%eax), %0\n"
nkeynes@577: "frame_not_found:"
nkeynes@577: 	: "=r" (result)
nkeynes@577: 	: "r" (&sh4r)
nkeynes@577: 	: "eax", "ecx", "edx" );
nkeynes@577:     return result;
nkeynes@577: }
nkeynes@577: 
nkeynes@577: 
nkeynes@539: #endif
nkeynes@539: 
nkeynes@539: