/******************************************************************* * BuildPendingEventCheck * * Build a function that checks whether there are any * pending DPMI events. * * Stack layout: * * (sp+12) long eflags * (sp+6) long cs * (sp+2) long ip * (sp) word fs * * On entry to function, fs register points to a valid TEB. * On exit from function, stack will be popped. */ static void BuildPendingEventCheck(void) { /* Function header */ function_header( "DPMI_PendingEventCheck" ); /* Check for pending events. */ output( "\t.byte 0x64\n\ttestl $0xffffffff,(%d)\n", STRUCTOFFSET(TEB,GdiTebBatch) + STRUCTOFFSET(WINE_VM86_TEB_INFO,vm86_pending) ); output( "\tje %s\n", asm_name("DPMI_PendingEventCheck_Cleanup") ); output( "\t.byte 0x64\n\ttestl $0xffffffff,(%d)\n", STRUCTOFFSET(TEB,GdiTebBatch) + STRUCTOFFSET(WINE_VM86_TEB_INFO,dpmi_vif) ); output( "\tje %s\n", asm_name("DPMI_PendingEventCheck_Cleanup") ); /* Process pending events. */ output( "\tsti\n" ); /* Start cleanup. Restore fs register. */ output( "%s\n", asm_globl("DPMI_PendingEventCheck_Cleanup") ); output( "\tpopw %%fs\n" ); /* Return from function. */ output( "%s\n", asm_globl("DPMI_PendingEventCheck_Return") ); output( "\tiret\n" ); output_function_size( "DPMI_PendingEventCheck" ); }
/* output the import thunks of a Win32 module */ static void output_immediate_import_thunks(void) { int i, j, pos; int nb_imm = nb_imports - nb_delayed; static const char import_thunks[] = "__wine_spec_import_thunks"; if (!nb_imm) return; output( "\n/* immediate import thunks */\n\n" ); output( "\t.text\n" ); output( "\t.align %d\n", get_alignment(8) ); output( "%s:\n", asm_name(import_thunks)); for (i = pos = 0; i < nb_imports; i++) { if (dll_imports[i]->delay) continue; for (j = 0; j < dll_imports[i]->nb_imports; j++, pos += get_ptr_size()) { ORDDEF *odp = dll_imports[i]->imports[j]; output_import_thunk( odp->name ? odp->name : odp->export_name, ".L__wine_spec_import_data_ptrs", pos ); } pos += get_ptr_size(); } output_function_size( import_thunks ); }
/******************************************************************* * BuildRet16Func * * Build the return code for 16-bit callbacks */ static void BuildRet16Func(void) { function_header( "__wine_call_to_16_ret" ); /* Save %esp into %esi */ output( "\tmovl %%esp,%%esi\n" ); /* Restore 32-bit segment registers */ output( "\t.byte 0x2e\n\tmovl %s", asm_name("CallTo16_DataSelector") ); output( "-%s,%%edi\n", asm_name("__wine_call16_start") ); output( "\tmovw %%di,%%ds\n" ); output( "\tmovw %%di,%%es\n" ); output( "\t.byte 0x2e\n\tmov %s", asm_name("CallTo16_TebSelector") ); output( "-%s,%%fs\n", asm_name("__wine_call16_start") ); output( "\t.byte 0x64\n\tmov (%d),%%gs\n", GS_OFFSET ); /* Restore the 32-bit stack */ output( "\tmovw %%di,%%ss\n" ); output( "\t.byte 0x64\n\tmovl (%d),%%esp\n", STACKOFFSET ); /* Return to caller */ output( "\tlret\n" ); output_function_size( "__wine_call_to_16_ret" ); }
/* output import stubs for exported entry points that link to external symbols */ static void output_external_link_imports( DLLSPEC *spec ) { unsigned int i, pos; if (!ext_link_imports.count) return; /* nothing to do */ sort_names( &ext_link_imports ); /* get rid of duplicate names */ for (i = 1; i < ext_link_imports.count; i++) { if (!strcmp( ext_link_imports.names[i-1], ext_link_imports.names[i] )) remove_name( &ext_link_imports, i-- ); } output( "\n/* external link thunks */\n\n" ); output( "\t.data\n" ); output( "\t.align %d\n", get_alignment(get_ptr_size()) ); output( ".L__wine_spec_external_links:\n" ); for (i = 0; i < ext_link_imports.count; i++) output( "\t%s %s\n", get_asm_ptr_keyword(), asm_name(ext_link_imports.names[i]) ); output( "\n\t.text\n" ); output( "\t.align %d\n", get_alignment(get_ptr_size()) ); output( "%s:\n", asm_name("__wine_spec_external_link_thunks") ); for (i = pos = 0; i < ext_link_imports.count; i++) { char *buffer = strmake( "__wine_spec_ext_link_%s", ext_link_imports.names[i] ); output_import_thunk( buffer, ".L__wine_spec_external_links", pos ); free( buffer ); pos += get_ptr_size(); } output_function_size( "__wine_spec_external_link_thunks" ); }
/* output a single import thunk */ static void output_import_thunk( const char *name, const char *table, int pos ) { output( "\n\t.align %d\n", get_alignment(4) ); output( "\t%s\n", func_declaration(name) ); output( "%s\n", asm_globl(name) ); output_cfi( ".cfi_startproc" ); switch(target_cpu) { case CPU_x86: if (!UsePIC) { output( "\tjmp *(%s+%d)\n", table, pos ); } else { output( "\tcall %s\n", asm_name("__wine_spec_get_pc_thunk_eax") ); output( "1:\tjmp *%s+%d-1b(%%eax)\n", table, pos ); } break; case CPU_x86_64: output( "\tjmpq *%s+%d(%%rip)\n", table, pos ); break; case CPU_ARM: output( "\tldr IP,1f\n"); output( "\tldr PC,[PC,IP]\n" ); output( "1:\t.long %s+%u-(1b+4)\n", table, pos ); break; case CPU_ARM64: output( "\tadr x9, 1f\n" ); output( "\tldur x9, [x9, #0]\n" ); if (pos & 0xf000) output( "\tadd x9, x9, #%u\n", pos & 0xf000 ); if (pos & 0x0f00) output( "\tadd x9, x9, #%u\n", pos & 0x0f00 ); if (pos & 0x00f0) output( "\tadd x9, x9, #%u\n", pos & 0x00f0 ); if (pos & 0x000f) output( "\tadd x9, x9, #%u\n", pos & 0x000f ); output( "\tldur x9, [x9, #0]\n" ); output( "\tbr x9\n" ); output( "1:\t.quad %s\n", table ); break; case CPU_POWERPC: output( "\tmr %s, %s\n", ppc_reg(0), ppc_reg(31) ); if (target_platform == PLATFORM_APPLE) { output( "\tlis %s, ha16(%s+%d+32768)\n", ppc_reg(31), table, pos ); output( "\tla %s, lo16(%s+%d)(%s)\n", ppc_reg(31), table, pos, ppc_reg(31) ); } else { output( "\tlis %s, (%s+%d+32768)@h\n", ppc_reg(31), table, pos ); output( "\tla %s, (%s+%d)@l(%s)\n", ppc_reg(31), table, pos, ppc_reg(31) ); } output( "\tlwz %s, 0(%s)\n", ppc_reg(31), ppc_reg(31) ); output( "\tmtctr %s\n", ppc_reg(31) ); output( "\tmr %s, %s\n", ppc_reg(31), ppc_reg(0) ); output( "\tbctr\n" ); break; } output_cfi( ".cfi_endproc" ); output_function_size( name ); }
/******************************************************************* * BuildRelays16 * * Build all the 16-bit relay callbacks */ void BuildRelays16(void) { if (target_cpu != CPU_x86) { output( "/* File not used with this architecture. Do not edit! */\n\n" ); return; } /* File header */ output( "/* File generated automatically. Do not edit! */\n\n" ); output( "\t.text\n" ); output( "%s:\n\n", asm_name("__wine_spec_thunk_text_16") ); output( "%s\n", asm_globl("__wine_call16_start") ); /* Standard CallFrom16 routine */ BuildCallFrom16Core( FALSE, FALSE ); /* Register CallFrom16 routine */ BuildCallFrom16Core( TRUE, FALSE ); /* C16ThkSL CallFrom16 routine */ BuildCallFrom16Core( FALSE, TRUE ); /* Standard CallTo16 routine */ BuildCallTo16Core( 0 ); /* Register CallTo16 routine */ BuildCallTo16Core( 1 ); /* Standard CallTo16 return stub */ BuildRet16Func(); /* CBClientThunkSL routine */ BuildCallTo32CBClient( FALSE ); /* CBClientThunkSLEx routine */ BuildCallTo32CBClient( TRUE ); /* Pending DPMI events check stub */ BuildPendingEventCheck(); output( "%s\n", asm_globl("__wine_call16_end") ); output_function_size( "__wine_spec_thunk_text_16" ); /* Declare the return address and data selector variables */ output( "\n\t.data\n\t.align %d\n", get_alignment(4) ); output( "%s\n\t.long 0\n", asm_globl("CallTo16_DataSelector") ); output( "%s\n\t.long 0\n", asm_globl("CallTo16_TebSelector") ); if (UsePIC) output( "wine_ldt_copy_ptr:\t.long %s\n", asm_name("wine_ldt_copy") ); output_gnu_stack_note(); }
/* output the get_pc thunk if needed */ void output_get_pc_thunk(void) { if (target_cpu != CPU_x86) return; if (!UsePIC) return; output( "\n\t.text\n" ); output( "\t.align %d\n", get_alignment(4) ); output( "\t%s\n", func_declaration("__wine_spec_get_pc_thunk_eax") ); output( "%s:\n", asm_name("__wine_spec_get_pc_thunk_eax") ); output_cfi( ".cfi_startproc" ); output( "\tmovl (%%esp),%%eax\n" ); output( "\tret\n" ); output_cfi( ".cfi_endproc" ); output_function_size( "__wine_spec_get_pc_thunk_eax" ); }
/******************************************************************* * BuildRelays32 * * Build all the 32-bit relay callbacks */ void BuildRelays32(void) { if (target_cpu != CPU_x86) { output( "/* File not used with this architecture. Do not edit! */\n\n" ); return; } /* File header */ output( "/* File generated automatically. Do not edit! */\n\n" ); output( "\t.text\n" ); output( "%s:\n\n", asm_name("__wine_spec_thunk_text_32") ); /* 32-bit register entry point */ BuildCallFrom32Regs(); output_function_size( "__wine_spec_thunk_text_32" ); output_gnu_stack_note(); }
/******************************************************************* * output_asm_relays16 * * Build all the 16-bit relay callbacks */ void output_asm_relays16(void) { /* File header */ output( "\t.text\n" ); output( "%s:\n\n", asm_name("__wine_spec_thunk_text_16") ); output( "%s\n", asm_globl("__wine_call16_start") ); /* Standard CallFrom16 routine */ BuildCallFrom16Core( 0, 0 ); /* Register CallFrom16 routine */ BuildCallFrom16Core( 1, 0 ); /* C16ThkSL CallFrom16 routine */ BuildCallFrom16Core( 0, 1 ); /* Standard CallTo16 routine */ BuildCallTo16Core( 0 ); /* Register CallTo16 routine */ BuildCallTo16Core( 1 ); /* Standard CallTo16 return stub */ BuildRet16Func(); /* CBClientThunkSL routine */ BuildCallTo32CBClient( 0 ); /* CBClientThunkSLEx routine */ BuildCallTo32CBClient( 1 ); output( "%s\n", asm_globl("__wine_call16_end") ); output_function_size( "__wine_spec_thunk_text_16" ); /* Declare the return address and data selector variables */ output( "\n\t.data\n\t.align %d\n", get_alignment(4) ); output( "%s\n\t.long 0\n", asm_globl("CallTo16_DataSelector") ); output( "%s\n\t.long 0\n", asm_globl("CallTo16_TebSelector") ); }
/******************************************************************* * BuildCallTo16Core * * This routine builds the core routines used in 32->16 thunks: * * extern DWORD WINAPI wine_call_to_16( FARPROC16 target, DWORD cbArgs, PEXCEPTION_HANDLER handler ); * extern void WINAPI wine_call_to_16_regs( CONTEXT86 *context, DWORD cbArgs, PEXCEPTION_HANDLER handler ); * * These routines can be called directly from 32-bit code. * * All routines expect that the 16-bit stack contents (arguments) and the * return address (segptr to CallTo16_Ret) were already set up by the * caller; nb_args must contain the number of bytes to be conserved. The * 16-bit SS:SP will be set accordingly. * * All other registers are either taken from the CONTEXT86 structure * or else set to default values. The target routine address is either * given directly or taken from the CONTEXT86. */ static void BuildCallTo16Core( int reg_func ) { const char *name = reg_func ? "wine_call_to_16_regs" : "wine_call_to_16"; /* Function header */ function_header( name ); /* Function entry sequence */ output_cfi( ".cfi_startproc" ); output( "\tpushl %%ebp\n" ); output_cfi( ".cfi_adjust_cfa_offset 4" ); output_cfi( ".cfi_rel_offset %%ebp,0" ); output( "\tmovl %%esp, %%ebp\n" ); output_cfi( ".cfi_def_cfa_register %%ebp" ); /* Save the 32-bit registers */ output( "\tpushl %%ebx\n" ); output_cfi( ".cfi_rel_offset %%ebx,-4" ); output( "\tpushl %%esi\n" ); output_cfi( ".cfi_rel_offset %%esi,-8" ); output( "\tpushl %%edi\n" ); output_cfi( ".cfi_rel_offset %%edi,-12" ); output( "\t.byte 0x64\n\tmov %%gs,(%d)\n", GS_OFFSET ); /* Setup exception frame */ output( "\t.byte 0x64\n\tpushl (%d)\n", STACKOFFSET ); output( "\tpushl 16(%%ebp)\n" ); /* handler */ output( "\t.byte 0x64\n\tpushl (0)\n" ); output( "\t.byte 0x64\n\tmovl %%esp,(0)\n" ); /* Call the actual CallTo16 routine (simulate a lcall) */ output( "\tpushl %%cs\n" ); output( "\tcall .L%s\n", name ); /* Remove exception frame */ output( "\t.byte 0x64\n\tpopl (0)\n" ); output( "\taddl $4, %%esp\n" ); output( "\t.byte 0x64\n\tpopl (%d)\n", STACKOFFSET ); if ( !reg_func ) { /* Convert return value */ output( "\tandl $0xffff,%%eax\n" ); output( "\tshll $16,%%edx\n" ); output( "\torl %%edx,%%eax\n" ); } else { /* * Modify CONTEXT86 structure to contain new values * * NOTE: We restore only EAX, EBX, EDX, EDX, EBP, and ESP. * The segment registers as well as ESI and EDI should * not be modified by a well-behaved 16-bit routine in * any case. [If necessary, we could restore them as well, * at the cost of a somewhat less efficient return path.] */ output( "\tmovl 0x14(%%esp),%%edi\n" ); /* FIELD_OFFSET(STACK32FRAME,target) - FIELD_OFFSET(STACK32FRAME,edi) */ /* everything above edi has been popped already */ output( "\tmovl %%eax,0xb0(%%edi)\n"); /* Eax */ output( "\tmovl %%ebx,0xa4(%%edi)\n"); /* Ebx */ output( "\tmovl %%ecx,0xac(%%edi)\n"); /* Ecx */ output( "\tmovl %%edx,0xa8(%%edi)\n"); /* Edx */ output( "\tmovl %%ebp,0xb4(%%edi)\n"); /* Ebp */ output( "\tmovl %%esi,0xc4(%%edi)\n"); /* Esp */ /* The return glue code saved %esp into %esi */ } /* Restore the 32-bit registers */ output( "\tpopl %%edi\n" ); output_cfi( ".cfi_same_value %%edi" ); output( "\tpopl %%esi\n" ); output_cfi( ".cfi_same_value %%esi" ); output( "\tpopl %%ebx\n" ); output_cfi( ".cfi_same_value %%ebx" ); /* Function exit sequence */ output( "\tpopl %%ebp\n" ); output_cfi( ".cfi_def_cfa %%esp,4" ); output_cfi( ".cfi_same_value %%ebp" ); output( "\tret $12\n" ); output_cfi( ".cfi_endproc" ); /* Start of the actual CallTo16 routine */ output( ".L%s:\n", name ); /* Switch to the 16-bit stack */ output( "\tmovl %%esp,%%edx\n" ); output( "\t.byte 0x64\n\tmovw (%d),%%ss\n", STACKOFFSET + 2); output( "\t.byte 0x64\n\tmovw (%d),%%sp\n", STACKOFFSET ); output( "\t.byte 0x64\n\tmovl %%edx,(%d)\n", STACKOFFSET ); /* Make %bp point to the previous stackframe (built by CallFrom16) */ output( "\tmovzwl %%sp,%%ebp\n" ); output( "\tleal 0x2a(%%ebp),%%ebp\n"); /* FIELD_OFFSET(STACK16FRAME,bp) */ /* Add the specified offset to the new sp */ output( "\tsubw 0x2c(%%edx), %%sp\n"); /* FIELD_OFFSET(STACK32FRAME,nb_args) */ if (reg_func) { /* Push the called routine address */ output( "\tmovl 0x28(%%edx),%%edx\n"); /* FIELD_OFFSET(STACK32FRAME,target) */ output( "\tpushw 0xbc(%%edx)\n"); /* SegCs */ output( "\tpushw 0xb8(%%edx)\n"); /* Eip */ /* Get the registers */ output( "\tpushw 0x98(%%edx)\n"); /* SegDs */ output( "\tpushl 0x94(%%edx)\n"); /* SegEs */ output( "\tpopl %%es\n" ); output( "\tpushl 0x90(%%edx)\n"); /* SegFs */ output( "\tpopl %%fs\n" ); output( "\tpushl 0x8c(%%edx)\n"); /* SegGs */ output( "\tpopl %%gs\n" ); output( "\tmovl 0xb4(%%edx),%%ebp\n"); /* Ebp */ output( "\tmovl 0xa0(%%edx),%%esi\n"); /* Esi */ output( "\tmovl 0x9c(%%edx),%%edi\n"); /* Edi */ output( "\tmovl 0xb0(%%edx),%%eax\n"); /* Eax */ output( "\tmovl 0xa4(%%edx),%%ebx\n"); /* Ebx */ output( "\tmovl 0xac(%%edx),%%ecx\n"); /* Ecx */ output( "\tmovl 0xa8(%%edx),%%edx\n"); /* Edx */ /* Get the 16-bit ds */ output( "\tpopw %%ds\n" ); } else /* not a register function */ { /* Push the called routine address */ output( "\tpushl 0x28(%%edx)\n"); /* FIELD_OFFSET(STACK32FRAME,target) */ /* Set %fs and %gs to the value saved by the last CallFrom16 */ output( "\tpushw -22(%%ebp)\n" ); /* FIELD_OFFSET(STACK16FRAME,fs)-FIELD_OFFSET(STACK16FRAME,bp) */ output( "\tpopw %%fs\n" ); output( "\tpushw -20(%%ebp)\n" ); /* FIELD_OFFSET(STACK16FRAME,gs)-FIELD_OFFSET(STACK16FRAME,bp) */ output( "\tpopw %%gs\n" ); /* Set %ds and %es (and %ax just in case) equal to %ss */ output( "\tmovw %%ss,%%ax\n" ); output( "\tmovw %%ax,%%ds\n" ); output( "\tmovw %%ax,%%es\n" ); } /* Jump to the called routine */ output( "\t.byte 0x66\n" ); output( "\tlret\n" ); /* Function footer */ output_function_size( name ); }
/* output a single import thunk */ static void output_import_thunk( const char *name, const char *table, int pos ) { output( "\n\t.align %d\n", get_alignment(4) ); output( "\t%s\n", func_declaration(name) ); output( "%s\n", asm_globl(name) ); output_cfi( ".cfi_startproc" ); switch(target_cpu) { case CPU_x86: if (!UsePIC) { output( "\tjmp *(%s+%d)\n", table, pos ); } else { output( "\tcall %s\n", asm_name("__wine_spec_get_pc_thunk_eax") ); output( "1:\tjmp *%s+%d-1b(%%eax)\n", table, pos ); } break; case CPU_x86_64: output( "\tjmpq *%s+%d(%%rip)\n", table, pos ); break; case CPU_SPARC: if ( !UsePIC ) { output( "\tsethi %%hi(%s+%d), %%g1\n", table, pos ); output( "\tld [%%g1+%%lo(%s+%d)], %%g1\n", table, pos ); output( "\tjmp %%g1\n" ); output( "\tnop\n" ); } else { /* Hmpf. Stupid sparc assembler always interprets global variable names as GOT offsets, so we have to do it the long way ... */ output( "\tsave %%sp, -96, %%sp\n" ); output( "0:\tcall 1f\n" ); output( "\tnop\n" ); output( "1:\tsethi %%hi(%s+%d-0b), %%g1\n", table, pos ); output( "\tor %%g1, %%lo(%s+%d-0b), %%g1\n", table, pos ); output( "\tld [%%g1+%%o7], %%g1\n" ); output( "\tjmp %%g1\n" ); output( "\trestore\n" ); } break; case CPU_ARM: output( "\tldr IP,[PC,#0]\n"); output( "\tldr PC,[IP,#%d]\n", pos); output( "\t.long %s\n", table ); break; case CPU_ARM64: output( "\tadr x9, 1f\n" ); output( "\tldur x9, [x9, #0]\n" ); if (pos & 0xf000) output( "\tadd x9, x9, #%u\n", pos & 0xf000 ); if (pos & 0x0f00) output( "\tadd x9, x9, #%u\n", pos & 0x0f00 ); if (pos & 0x00f0) output( "\tadd x9, x9, #%u\n", pos & 0x00f0 ); if (pos & 0x000f) output( "\tadd x9, x9, #%u\n", pos & 0x000f ); output( "\tldur x9, [x9, #0]\n" ); output( "\tbr x9\n" ); output( "1:\t.quad %s\n", table ); break; case CPU_POWERPC: output( "\tmr %s, %s\n", ppc_reg(0), ppc_reg(31) ); if (target_platform == PLATFORM_APPLE) { output( "\tlis %s, ha16(%s+%d+32768)\n", ppc_reg(31), table, pos ); output( "\tla %s, lo16(%s+%d)(%s)\n", ppc_reg(31), table, pos, ppc_reg(31) ); } else { output( "\tlis %s, (%s+%d+32768)@h\n", ppc_reg(31), table, pos ); output( "\tla %s, (%s+%d)@l(%s)\n", ppc_reg(31), table, pos, ppc_reg(31) ); } output( "\tlwz %s, 0(%s)\n", ppc_reg(31), ppc_reg(31) ); output( "\tmtctr %s\n", ppc_reg(31) ); output( "\tmr %s, %s\n", ppc_reg(31), ppc_reg(0) ); output( "\tbctr\n" ); break; } output_cfi( ".cfi_endproc" ); output_function_size( name ); }
/******************************************************************* * output_stubs * * Output the functions for stub entry points */ void output_stubs( DLLSPEC *spec ) { const char *name, *exp_name; int i, count; if (!has_stubs( spec )) return; output( "\n/* stub functions */\n\n" ); output( "\t.text\n" ); for (i = count = 0; i < spec->nb_entry_points; i++) { ORDDEF *odp = &spec->entry_points[i]; if (odp->type != TYPE_STUB) continue; name = get_stub_name( odp, spec ); exp_name = odp->name ? odp->name : odp->export_name; output( "\t.align %d\n", get_alignment(4) ); output( "\t%s\n", func_declaration(name) ); output( "%s:\n", asm_name(name) ); output_cfi( ".cfi_startproc" ); switch (target_cpu) { case CPU_x86: /* flesh out the stub a bit to make safedisc happy */ output(" \tnop\n" ); output(" \tnop\n" ); output(" \tnop\n" ); output(" \tnop\n" ); output(" \tnop\n" ); output(" \tnop\n" ); output(" \tnop\n" ); output(" \tnop\n" ); output(" \tnop\n" ); output( "\tsubl $12,%%esp\n" ); output_cfi( ".cfi_adjust_cfa_offset 12" ); if (UsePIC) { output( "\tcall %s\n", asm_name("__wine_spec_get_pc_thunk_eax") ); output( "1:" ); if (exp_name) { output( "\tleal .L%s_string-1b(%%eax),%%ecx\n", name ); output( "\tmovl %%ecx,4(%%esp)\n" ); count++; } else output( "\tmovl $%d,4(%%esp)\n", odp->ordinal ); output( "\tleal .L__wine_spec_file_name-1b(%%eax),%%ecx\n" ); output( "\tmovl %%ecx,(%%esp)\n" ); } else { if (exp_name) { output( "\tmovl $.L%s_string,4(%%esp)\n", name ); count++; } else output( "\tmovl $%d,4(%%esp)\n", odp->ordinal ); output( "\tmovl $.L__wine_spec_file_name,(%%esp)\n" ); } output( "\tcall %s\n", asm_name("__wine_spec_unimplemented_stub") ); break; case CPU_x86_64: output( "\tsubq $8,%%rsp\n" ); output_cfi( ".cfi_adjust_cfa_offset 8" ); output( "\tleaq .L__wine_spec_file_name(%%rip),%%rdi\n" ); if (exp_name) { output( "leaq .L%s_string(%%rip),%%rsi\n", name ); count++; } else output( "\tmovq $%d,%%rsi\n", odp->ordinal ); output( "\tcall %s\n", asm_name("__wine_spec_unimplemented_stub") ); break; case CPU_ARM: output( "\tldr r0,[PC,#0]\n"); output( "\tmov PC,PC\n"); output( "\t.long .L__wine_spec_file_name\n" ); output( "\tldr r1,[PC,#0]\n"); output( "\tmov PC,PC\n"); if (exp_name) { output( "\t.long .L%s_string\n", name ); count++; } else output( "\t.long %d\n", odp->ordinal ); output( "\tbl %s\n", asm_name("__wine_spec_unimplemented_stub") ); break; default: assert(0); } output_cfi( ".cfi_endproc" ); output_function_size( name ); } if (count) { output( "\t%s\n", get_asm_string_section() ); for (i = 0; i < spec->nb_entry_points; i++) { ORDDEF *odp = &spec->entry_points[i]; if (odp->type != TYPE_STUB) continue; exp_name = odp->name ? odp->name : odp->export_name; if (exp_name) { name = get_stub_name( odp, spec ); output( ".L%s_string:\n", name ); output( "\t%s \"%s\"\n", get_asm_string_keyword(), exp_name ); } } } }
/******************************************************************* * BuildCallTo32CBClient * * Call a CBClient relay stub from 32-bit code (KERNEL.620). * * Since the relay stub is itself 32-bit, this should not be a problem; * unfortunately, the relay stubs are expected to switch back to a * 16-bit stack (and 16-bit code) after completion :-( * * This would conflict with our 16- vs. 32-bit stack handling, so * we simply switch *back* to our 32-bit stack before returning to * the caller ... * * The CBClient relay stub expects to be called with the following * 16-bit stack layout, and with ebp and ebx pointing into the 16-bit * stack at the designated places: * * ... * (ebp+14) original arguments to the callback routine * (ebp+10) far return address to original caller * (ebp+6) Thunklet target address * (ebp+2) Thunklet relay ID code * (ebp) BP (saved by CBClientGlueSL) * (ebp-2) SI (saved by CBClientGlueSL) * (ebp-4) DI (saved by CBClientGlueSL) * (ebp-6) DS (saved by CBClientGlueSL) * * ... buffer space used by the 16-bit side glue for temp copies * * (ebx+4) far return address to 16-bit side glue code * (ebx) saved 16-bit ss:sp (pointing to ebx+4) * * The 32-bit side glue code accesses both the original arguments (via ebp) * and the temporary copies prepared by the 16-bit side glue (via ebx). * After completion, the stub will load ss:sp from the buffer at ebx * and perform a far return to 16-bit code. * * To trick the relay stub into returning to us, we replace the 16-bit * return address to the glue code by a cs:ip pair pointing to our * return entry point (the original return address is saved first). * Our return stub thus called will then reload the 32-bit ss:esp and * return to 32-bit code (by using and ss:esp value that we have also * pushed onto the 16-bit stack before and a cs:eip values found at * that position on the 32-bit stack). The ss:esp to be restored is * found relative to the 16-bit stack pointer at: * * (ebx-4) ss (flat) * (ebx-8) sp (32-bit stack pointer) * * The second variant of this routine, CALL32_CBClientEx, which is used * to implement KERNEL.621, has to cope with yet another problem: Here, * the 32-bit side directly returns to the caller of the CBClient thunklet, * restoring registers saved by CBClientGlueSL and cleaning up the stack. * As we have to return to our 32-bit code first, we have to adapt the * layout of our temporary area so as to include values for the registers * that are to be restored, and later (in the implementation of KERNEL.621) * we *really* restore them. The return stub restores DS, DI, SI, and BP * from the stack, skips the next 8 bytes (CBClient relay code / target), * and then performs a lret NN, where NN is the number of arguments to be * removed. Thus, we prepare our temporary area as follows: * * (ebx+22) 16-bit cs (this segment) * (ebx+20) 16-bit ip ('16-bit' return entry point) * (ebx+16) 32-bit ss (flat) * (ebx+12) 32-bit sp (32-bit stack pointer) * (ebx+10) 16-bit bp (points to ebx+24) * (ebx+8) 16-bit si (ignored) * (ebx+6) 16-bit di (ignored) * (ebx+4) 16-bit ds (we actually use the flat DS here) * (ebx+2) 16-bit ss (16-bit stack segment) * (ebx+0) 16-bit sp (points to ebx+4) * * Note that we ensure that DS is not changed and remains the flat segment, * and the 32-bit stack pointer our own return stub needs fits just * perfectly into the 8 bytes that are skipped by the Windows stub. * One problem is that we have to determine the number of removed arguments, * as these have to be really removed in KERNEL.621. Thus, the BP value * that we place in the temporary area to be restored, contains the value * that SP would have if no arguments were removed. By comparing the actual * value of SP with this value in our return stub we can compute the number * of removed arguments. This is then returned to KERNEL.621. * * The stack layout of this function: * (ebp+20) nArgs pointer to variable receiving nr. of args (Ex only) * (ebp+16) esi pointer to caller's esi value * (ebp+12) arg ebp value to be set for relay stub * (ebp+8) func CBClient relay stub address * (ebp+4) ret addr * (ebp) ebp */ static void BuildCallTo32CBClient( int isEx ) { function_header( isEx ? "CALL32_CBClientEx" : "CALL32_CBClient" ); /* Entry code */ output_cfi( ".cfi_startproc" ); output( "\tpushl %%ebp\n" ); output_cfi( ".cfi_adjust_cfa_offset 4" ); output_cfi( ".cfi_rel_offset %%ebp,0" ); output( "\tmovl %%esp,%%ebp\n" ); output_cfi( ".cfi_def_cfa_register %%ebp" ); output( "\tpushl %%edi\n" ); output_cfi( ".cfi_rel_offset %%edi,-4" ); output( "\tpushl %%esi\n" ); output_cfi( ".cfi_rel_offset %%esi,-8" ); output( "\tpushl %%ebx\n" ); output_cfi( ".cfi_rel_offset %%ebx,-12" ); /* Get pointer to temporary area and save the 32-bit stack pointer */ output( "\tmovl 16(%%ebp), %%ebx\n" ); output( "\tleal -8(%%esp), %%eax\n" ); if ( !isEx ) output( "\tmovl %%eax, -8(%%ebx)\n" ); else output( "\tmovl %%eax, 12(%%ebx)\n" ); /* Set up registers and call CBClient relay stub (simulating a far call) */ output( "\tmovl 20(%%ebp), %%esi\n" ); output( "\tmovl (%%esi), %%esi\n" ); output( "\tmovl 8(%%ebp), %%eax\n" ); output( "\tmovl 12(%%ebp), %%ebp\n" ); output( "\tpushl %%cs\n" ); output( "\tcall *%%eax\n" ); /* Return new esi value to caller */ output( "\tmovl 32(%%esp), %%edi\n" ); output( "\tmovl %%esi, (%%edi)\n" ); /* Return argument size to caller */ if ( isEx ) { output( "\tmovl 36(%%esp), %%ebx\n" ); output( "\tmovl %%ebp, (%%ebx)\n" ); } /* Restore registers and return */ output( "\tpopl %%ebx\n" ); output_cfi( ".cfi_same_value %%ebx" ); output( "\tpopl %%esi\n" ); output_cfi( ".cfi_same_value %%esi" ); output( "\tpopl %%edi\n" ); output_cfi( ".cfi_same_value %%edi" ); output( "\tpopl %%ebp\n" ); output_cfi( ".cfi_def_cfa %%esp,4" ); output_cfi( ".cfi_same_value %%ebp" ); output( "\tret\n" ); output_cfi( ".cfi_endproc" ); output_function_size( isEx ? "CALL32_CBClientEx" : "CALL32_CBClient" ); /* '16-bit' return stub */ function_header( isEx ? "CALL32_CBClientEx_Ret" : "CALL32_CBClient_Ret" ); if ( !isEx ) { output( "\tmovzwl %%sp, %%ebx\n" ); output( "\tlssl %%ss:-16(%%ebx), %%esp\n" ); } else { output( "\tmovzwl %%bp, %%ebx\n" ); output( "\tsubw %%bp, %%sp\n" ); output( "\tmovzwl %%sp, %%ebp\n" ); output( "\tlssl %%ss:-12(%%ebx), %%esp\n" ); } output( "\tlret\n" ); output_function_size( isEx ? "CALL32_CBClientEx_Ret" : "CALL32_CBClient_Ret" ); }
/******************************************************************* * BuildCallFrom32Regs * * Build a 32-bit-to-Wine call-back function for a 'register' function. * 'args' is the number of dword arguments. * * Stack layout: * ... * (ebp+16) first arg * (ebp+12) ret addr to user code * (ebp+8) eax saved by relay code * (ebp+4) ret addr to relay code * (ebp+0) saved ebp * (ebp-128) buffer area to allow stack frame manipulation * (ebp-332) CONTEXT86 struct * (ebp-336) padding for stack alignment * (ebp-336-n) CONTEXT86 *argument * .... other arguments copied from (ebp+12) * * The entry point routine is called with a CONTEXT* extra argument, * following the normal args. In this context structure, EIP_reg * contains the return address to user code, and ESP_reg the stack * pointer on return (with the return address and arguments already * removed). */ static void BuildCallFrom32Regs(void) { static const int STACK_SPACE = 128 + sizeof(CONTEXT86); /* Function header */ function_header( "__wine_call_from_32_regs" ); /* Allocate some buffer space on the stack */ output( "\tpushl %%ebp\n" ); output( "\tmovl %%esp,%%ebp\n "); output( "\tleal -%d(%%esp), %%esp\n", STACK_SPACE + 4 /* for context arg */); /* Build the context structure */ output( "\tpushfl\n" ); output( "\tpopl %%eax\n" ); output( "\tmovl %%eax,%d(%%ebp)\n", CONTEXTOFFSET(EFlags) - STACK_SPACE ); output( "\tmovl 0(%%ebp),%%eax\n" ); output( "\tmovl %%eax,%d(%%ebp)\n", CONTEXTOFFSET(Ebp) - STACK_SPACE ); output( "\tmovl 8(%%ebp),%%eax\n" ); output( "\tmovl %%eax,%d(%%ebp)\n", CONTEXTOFFSET(Eax) - STACK_SPACE ); output( "\tmovl %%ebx,%d(%%ebp)\n", CONTEXTOFFSET(Ebx) - STACK_SPACE ); output( "\tmovl %%ecx,%d(%%ebp)\n", CONTEXTOFFSET(Ecx) - STACK_SPACE ); output( "\tmovl %%edx,%d(%%ebp)\n", CONTEXTOFFSET(Edx) - STACK_SPACE ); output( "\tmovl %%esi,%d(%%ebp)\n", CONTEXTOFFSET(Esi) - STACK_SPACE ); output( "\tmovl %%edi,%d(%%ebp)\n", CONTEXTOFFSET(Edi) - STACK_SPACE ); output( "\txorl %%eax,%%eax\n" ); output( "\tmovw %%cs,%%ax\n" ); output( "\tmovl %%eax,%d(%%ebp)\n", CONTEXTOFFSET(SegCs) - STACK_SPACE ); output( "\tmovw %%es,%%ax\n" ); output( "\tmovl %%eax,%d(%%ebp)\n", CONTEXTOFFSET(SegEs) - STACK_SPACE ); output( "\tmovw %%fs,%%ax\n" ); output( "\tmovl %%eax,%d(%%ebp)\n", CONTEXTOFFSET(SegFs) - STACK_SPACE ); output( "\tmovw %%gs,%%ax\n" ); output( "\tmovl %%eax,%d(%%ebp)\n", CONTEXTOFFSET(SegGs) - STACK_SPACE ); output( "\tmovw %%ss,%%ax\n" ); output( "\tmovl %%eax,%d(%%ebp)\n", CONTEXTOFFSET(SegSs) - STACK_SPACE ); output( "\tmovw %%ds,%%ax\n" ); output( "\tmovl %%eax,%d(%%ebp)\n", CONTEXTOFFSET(SegDs) - STACK_SPACE ); output( "\tmovw %%ax,%%es\n" ); /* set %es equal to %ds just in case */ output( "\tmovl $0x%x,%%eax\n", CONTEXT86_FULL ); output( "\tmovl %%eax,%d(%%ebp)\n", CONTEXTOFFSET(ContextFlags) - STACK_SPACE ); output( "\tmovl 12(%%ebp),%%eax\n" ); /* Get %eip at time of call */ output( "\tmovl %%eax,%d(%%ebp)\n", CONTEXTOFFSET(Eip) - STACK_SPACE ); /* Transfer the arguments */ output( "\tmovl 4(%%ebp),%%ebx\n" ); /* get relay code addr */ output( "\tmovzbl 4(%%ebx),%%ecx\n" ); /* fetch number of args to copy */ output( "\tsubl %%ecx,%%esp\n" ); output( "\tandl $~15,%%esp\n" ); output( "\tleal 16(%%ebp),%%esi\n" ); /* get %esp at time of call */ output( "\tmovl %%esp,%%edi\n" ); output( "\tshrl $2,%%ecx\n" ); output( "\tjz 1f\n" ); output( "\tcld\n" ); output( "\trep\n\tmovsl\n" ); /* copy args */ output( "1:\tleal %d(%%ebp),%%eax\n", -STACK_SPACE ); /* get addr of context struct */ output( "\tmovl %%eax,(%%edi)\n" ); /* and pass it as extra arg */ output( "\tmovzbl 5(%%ebx),%%eax\n" ); /* fetch number of args to remove */ output( "\tleal 16(%%ebp,%%eax),%%eax\n" ); output( "\tmovl %%eax,%d(%%ebp)\n", CONTEXTOFFSET(Esp) - STACK_SPACE ); /* Call the entry point */ output( "\taddl (%%ebx),%%ebx\n" ); output( "\tcall *%%ebx\n" ); output( "\tleal -%d(%%ebp),%%ecx\n", STACK_SPACE ); /* Restore the context structure */ output( "2:\tpushl %d(%%ecx)\n", CONTEXTOFFSET(SegEs) ); output( "\tpopl %%es\n" ); output( "\tpushl %d(%%ecx)\n", CONTEXTOFFSET(SegFs) ); output( "\tpopl %%fs\n" ); output( "\tpushl %d(%%ecx)\n", CONTEXTOFFSET(SegGs) ); output( "\tpopl %%gs\n" ); output( "\tmovl %d(%%ecx),%%edi\n", CONTEXTOFFSET(Edi) ); output( "\tmovl %d(%%ecx),%%esi\n", CONTEXTOFFSET(Esi) ); output( "\tmovl %d(%%ecx),%%edx\n", CONTEXTOFFSET(Edx) ); output( "\tmovl %d(%%ecx),%%ebx\n", CONTEXTOFFSET(Ebx) ); output( "\tmovl %d(%%ecx),%%eax\n", CONTEXTOFFSET(Eax) ); output( "\tmovl %d(%%ecx),%%ebp\n", CONTEXTOFFSET(Ebp) ); output( "\tpushl %d(%%ecx)\n", CONTEXTOFFSET(SegSs) ); output( "\tpopl %%ss\n" ); output( "\tmovl %d(%%ecx),%%esp\n", CONTEXTOFFSET(Esp) ); output( "\tpushl %d(%%ecx)\n", CONTEXTOFFSET(EFlags) ); output( "\tpushl %d(%%ecx)\n", CONTEXTOFFSET(SegCs) ); output( "\tpushl %d(%%ecx)\n", CONTEXTOFFSET(Eip) ); output( "\tpushl %d(%%ecx)\n", CONTEXTOFFSET(SegDs) ); output( "\tmovl %d(%%ecx),%%ecx\n", CONTEXTOFFSET(Ecx) ); output( "\tpopl %%ds\n" ); output( "\tiret\n" ); output_function_size( "__wine_call_from_32_regs" ); function_header( "__wine_call_from_32_restore_regs" ); output( "\tmovl 4(%%esp),%%ecx\n" ); output( "\tjmp 2b\n" ); output_function_size( "__wine_call_from_32_restore_regs" ); }
/* output the delayed import table of a Win32 module */ static void output_delayed_imports( const DLLSPEC *spec ) { int i, j, mod; if (!nb_delayed) return; output( "\n/* delayed imports */\n\n" ); output( "\t.data\n" ); output( "\t.align %d\n", get_alignment(get_ptr_size()) ); output( "%s\n", asm_globl("__wine_spec_delay_imports") ); /* list of dlls */ for (i = j = mod = 0; i < nb_imports; i++) { if (!dll_imports[i]->delay) continue; output( "\t%s 0\n", get_asm_ptr_keyword() ); /* grAttrs */ output( "\t%s .L__wine_delay_name_%d\n", /* szName */ get_asm_ptr_keyword(), i ); output( "\t%s .L__wine_delay_modules+%d\n", /* phmod */ get_asm_ptr_keyword(), mod * get_ptr_size() ); output( "\t%s .L__wine_delay_IAT+%d\n", /* pIAT */ get_asm_ptr_keyword(), j * get_ptr_size() ); output( "\t%s .L__wine_delay_INT+%d\n", /* pINT */ get_asm_ptr_keyword(), j * get_ptr_size() ); output( "\t%s 0\n", get_asm_ptr_keyword() ); /* pBoundIAT */ output( "\t%s 0\n", get_asm_ptr_keyword() ); /* pUnloadIAT */ output( "\t%s 0\n", get_asm_ptr_keyword() ); /* dwTimeStamp */ j += dll_imports[i]->nb_imports; mod++; } output( "\t%s 0\n", get_asm_ptr_keyword() ); /* grAttrs */ output( "\t%s 0\n", get_asm_ptr_keyword() ); /* szName */ output( "\t%s 0\n", get_asm_ptr_keyword() ); /* phmod */ output( "\t%s 0\n", get_asm_ptr_keyword() ); /* pIAT */ output( "\t%s 0\n", get_asm_ptr_keyword() ); /* pINT */ output( "\t%s 0\n", get_asm_ptr_keyword() ); /* pBoundIAT */ output( "\t%s 0\n", get_asm_ptr_keyword() ); /* pUnloadIAT */ output( "\t%s 0\n", get_asm_ptr_keyword() ); /* dwTimeStamp */ output( "\n.L__wine_delay_IAT:\n" ); for (i = 0; i < nb_imports; i++) { if (!dll_imports[i]->delay) continue; for (j = 0; j < dll_imports[i]->nb_imports; j++) { ORDDEF *odp = dll_imports[i]->imports[j]; const char *name = odp->name ? odp->name : odp->export_name; output( "\t%s .L__wine_delay_imp_%d_%s\n", get_asm_ptr_keyword(), i, name ); } } output( "\n.L__wine_delay_INT:\n" ); for (i = 0; i < nb_imports; i++) { if (!dll_imports[i]->delay) continue; for (j = 0; j < dll_imports[i]->nb_imports; j++) { ORDDEF *odp = dll_imports[i]->imports[j]; if (!odp->name) output( "\t%s %d\n", get_asm_ptr_keyword(), odp->ordinal ); else output( "\t%s .L__wine_delay_data_%d_%s\n", get_asm_ptr_keyword(), i, odp->name ); } } output( "\n.L__wine_delay_modules:\n" ); for (i = 0; i < nb_imports; i++) { if (dll_imports[i]->delay) output( "\t%s 0\n", get_asm_ptr_keyword() ); } for (i = 0; i < nb_imports; i++) { if (!dll_imports[i]->delay) continue; output( ".L__wine_delay_name_%d:\n", i ); output( "\t%s \"%s\"\n", get_asm_string_keyword(), dll_imports[i]->spec->file_name ); } for (i = 0; i < nb_imports; i++) { if (!dll_imports[i]->delay) continue; for (j = 0; j < dll_imports[i]->nb_imports; j++) { ORDDEF *odp = dll_imports[i]->imports[j]; if (!odp->name) continue; output( ".L__wine_delay_data_%d_%s:\n", i, odp->name ); output( "\t%s \"%s\"\n", get_asm_string_keyword(), odp->name ); } } output_function_size( "__wine_spec_delay_imports" ); }
/******************************************************************* * BuildCallFrom16Core * * This routine builds the core routines used in 16->32 thunks: * CallFrom16Word, CallFrom16Long, CallFrom16Register, and CallFrom16Thunk. * * These routines are intended to be called via a far call (with 32-bit * operand size) from 16-bit code. The 16-bit code stub must push %bp, * the 32-bit entry point to be called, and the argument conversion * routine to be used (see stack layout below). * * The core routine completes the STACK16FRAME on the 16-bit stack and * switches to the 32-bit stack. Then, the argument conversion routine * is called; it gets passed the 32-bit entry point and a pointer to the * 16-bit arguments (on the 16-bit stack) as parameters. (You can either * use conversion routines automatically generated by BuildCallFrom16, * or write your own for special purposes.) * * The conversion routine must call the 32-bit entry point, passing it * the converted arguments, and return its return value to the core. * After the conversion routine has returned, the core switches back * to the 16-bit stack, converts the return value to the DX:AX format * (CallFrom16Long), and returns to the 16-bit call stub. All parameters, * including %bp, are popped off the stack. * * The 16-bit call stub now returns to the caller, popping the 16-bit * arguments if necessary (pascal calling convention). * * In the case of a 'register' function, CallFrom16Register fills a * CONTEXT86 structure with the values all registers had at the point * the first instruction of the 16-bit call stub was about to be * executed. A pointer to this CONTEXT86 is passed as third parameter * to the argument conversion routine, which typically passes it on * to the called 32-bit entry point. * * CallFrom16Thunk is a special variant used by the implementation of * the Win95 16->32 thunk functions C16ThkSL and C16ThkSL01 and is * implemented as follows: * On entry, the EBX register is set up to contain a flat pointer to the * 16-bit stack such that EBX+22 points to the first argument. * Then, the entry point is called, while EBP is set up to point * to the return address (on the 32-bit stack). * The called function returns with CX set to the number of bytes * to be popped of the caller's stack. * * Stack layout upon entry to the core routine (STACK16FRAME): * ... ... * (sp+24) word first 16-bit arg * (sp+22) word cs * (sp+20) word ip * (sp+18) word bp * (sp+14) long 32-bit entry point (reused for Win16 mutex recursion count) * (sp+12) word ip of actual entry point (necessary for relay debugging) * (sp+8) long relay (argument conversion) function entry point * (sp+4) long cs of 16-bit entry point * (sp) long ip of 16-bit entry point * * Added on the stack: * (sp-2) word saved gs * (sp-4) word saved fs * (sp-6) word saved es * (sp-8) word saved ds * (sp-12) long saved ebp * (sp-16) long saved ecx * (sp-20) long saved edx * (sp-24) long saved previous stack */ static void BuildCallFrom16Core( int reg_func, int thunk ) { /* Function header */ if (thunk) function_header( "__wine_call_from_16_thunk" ); else if (reg_func) function_header( "__wine_call_from_16_regs" ); else function_header( "__wine_call_from_16" ); /* Create STACK16FRAME (except STACK32FRAME link) */ output( "\tpushw %%gs\n" ); output( "\tpushw %%fs\n" ); output( "\tpushw %%es\n" ); output( "\tpushw %%ds\n" ); output( "\tpushl %%ebp\n" ); output( "\tpushl %%ecx\n" ); output( "\tpushl %%edx\n" ); /* Save original EFlags register */ if (reg_func) output( "\tpushfl\n" ); if ( UsePIC ) { output( "\tcall 1f\n" ); output( "1:\tpopl %%ecx\n" ); output( "\t.byte 0x2e\n\tmovl %s-1b(%%ecx),%%edx\n", asm_name("CallTo16_DataSelector") ); } else output( "\t.byte 0x2e\n\tmovl %s,%%edx\n", asm_name("CallTo16_DataSelector") ); /* Load 32-bit segment registers */ output( "\tmovw %%dx, %%ds\n" ); output( "\tmovw %%dx, %%es\n" ); if ( UsePIC ) output( "\tmovw %s-1b(%%ecx), %%fs\n", asm_name("CallTo16_TebSelector") ); else output( "\tmovw %s, %%fs\n", asm_name("CallTo16_TebSelector") ); output( "\t.byte 0x64\n\tmov (%d),%%gs\n", GS_OFFSET ); /* Translate STACK16FRAME base to flat offset in %edx */ output( "\tmovw %%ss, %%dx\n" ); output( "\tandl $0xfff8, %%edx\n" ); output( "\tshrl $1, %%edx\n" ); if (UsePIC) { output( "\taddl wine_ldt_copy_ptr-1b(%%ecx),%%edx\n" ); output( "\tmovl (%%edx), %%edx\n" ); } else output( "\tmovl %s(%%edx), %%edx\n", asm_name("wine_ldt_copy") ); output( "\tmovzwl %%sp, %%ebp\n" ); output( "\tleal %d(%%ebp,%%edx), %%edx\n", reg_func ? 0 : -4 ); /* Get saved flags into %ecx */ if (reg_func) output( "\tpopl %%ecx\n" ); /* Get the 32-bit stack pointer from the TEB and complete STACK16FRAME */ output( "\t.byte 0x64\n\tmovl (%d), %%ebp\n", STACKOFFSET ); output( "\tpushl %%ebp\n" ); /* Switch stacks */ output( "\t.byte 0x64\n\tmovw %%ss, (%d)\n", STACKOFFSET + 2 ); output( "\t.byte 0x64\n\tmovw %%sp, (%d)\n", STACKOFFSET ); output( "\tpushl %%ds\n" ); output( "\tpopl %%ss\n" ); output( "\tmovl %%ebp, %%esp\n" ); output( "\taddl $0x20,%%ebp\n"); /* FIELD_OFFSET(STACK32FRAME,ebp) */ /* At this point: STACK16FRAME is completely set up DS, ES, SS: flat data segment FS: current TEB ESP: points to last STACK32FRAME EBP: points to ebp member of last STACK32FRAME EDX: points to current STACK16FRAME ECX: contains saved flags all other registers: unchanged */ /* Special case: C16ThkSL stub */ if ( thunk ) { /* Set up registers as expected and call thunk */ output( "\tleal 0x1a(%%edx),%%ebx\n" ); /* sizeof(STACK16FRAME)-22 */ output( "\tleal -4(%%esp), %%ebp\n" ); output( "\tcall *0x26(%%edx)\n"); /* FIELD_OFFSET(STACK16FRAME,entry_point) */ /* Switch stack back */ output( "\t.byte 0x64\n\tmovw (%d), %%ss\n", STACKOFFSET+2 ); output( "\t.byte 0x64\n\tmovzwl (%d), %%esp\n", STACKOFFSET ); output( "\t.byte 0x64\n\tpopl (%d)\n", STACKOFFSET ); /* Restore registers and return directly to caller */ output( "\taddl $8, %%esp\n" ); output( "\tpopl %%ebp\n" ); output( "\tpopw %%ds\n" ); output( "\tpopw %%es\n" ); output( "\tpopw %%fs\n" ); output( "\tpopw %%gs\n" ); output( "\taddl $20, %%esp\n" ); output( "\txorb %%ch, %%ch\n" ); output( "\tpopl %%ebx\n" ); output( "\taddw %%cx, %%sp\n" ); output( "\tpush %%ebx\n" ); output( "\t.byte 0x66\n" ); output( "\tlret\n" ); output_function_size( "__wine_call_from_16_thunk" ); return; } /* Build register CONTEXT */ if ( reg_func ) { output( "\tsubl $0x2cc,%%esp\n" ); /* sizeof(CONTEXT86) */ output( "\tmovl %%ecx,0xc0(%%esp)\n" ); /* EFlags */ output( "\tmovl %%eax,0xb0(%%esp)\n" ); /* Eax */ output( "\tmovl %%ebx,0xa4(%%esp)\n" ); /* Ebx */ output( "\tmovl %%esi,0xa0(%%esp)\n" ); /* Esi */ output( "\tmovl %%edi,0x9c(%%esp)\n" ); /* Edi */ output( "\tmovl 0x0c(%%edx),%%eax\n"); /* FIELD_OFFSET(STACK16FRAME,ebp) */ output( "\tmovl %%eax,0xb4(%%esp)\n" ); /* Ebp */ output( "\tmovl 0x08(%%edx),%%eax\n"); /* FIELD_OFFSET(STACK16FRAME,ecx) */ output( "\tmovl %%eax,0xac(%%esp)\n" ); /* Ecx */ output( "\tmovl 0x04(%%edx),%%eax\n"); /* FIELD_OFFSET(STACK16FRAME,edx) */ output( "\tmovl %%eax,0xa8(%%esp)\n" ); /* Edx */ output( "\tmovzwl 0x10(%%edx),%%eax\n"); /* FIELD_OFFSET(STACK16FRAME,ds) */ output( "\tmovl %%eax,0x98(%%esp)\n" ); /* SegDs */ output( "\tmovzwl 0x12(%%edx),%%eax\n"); /* FIELD_OFFSET(STACK16FRAME,es) */ output( "\tmovl %%eax,0x94(%%esp)\n" ); /* SegEs */ output( "\tmovzwl 0x14(%%edx),%%eax\n"); /* FIELD_OFFSET(STACK16FRAME,fs) */ output( "\tmovl %%eax,0x90(%%esp)\n" ); /* SegFs */ output( "\tmovzwl 0x16(%%edx),%%eax\n"); /* FIELD_OFFSET(STACK16FRAME,gs) */ output( "\tmovl %%eax,0x8c(%%esp)\n" ); /* SegGs */ output( "\tmovzwl 0x2e(%%edx),%%eax\n"); /* FIELD_OFFSET(STACK16FRAME,cs) */ output( "\tmovl %%eax,0xbc(%%esp)\n" ); /* SegCs */ output( "\tmovzwl 0x2c(%%edx),%%eax\n"); /* FIELD_OFFSET(STACK16FRAME,ip) */ output( "\tmovl %%eax,0xb8(%%esp)\n" ); /* Eip */ output( "\t.byte 0x64\n\tmovzwl (%d), %%eax\n", STACKOFFSET+2 ); output( "\tmovl %%eax,0xc8(%%esp)\n" ); /* SegSs */ output( "\t.byte 0x64\n\tmovzwl (%d), %%eax\n", STACKOFFSET ); output( "\taddl $0x2c,%%eax\n"); /* FIELD_OFFSET(STACK16FRAME,ip) */ output( "\tmovl %%eax,0xc4(%%esp)\n" ); /* Esp */ #if 0 output( "\tfsave 0x1c(%%esp)\n" ); /* FloatSave */ #endif /* Push address of CONTEXT86 structure -- popped by the relay routine */ output( "\tmovl %%esp,%%eax\n" ); output( "\tandl $~15,%%esp\n" ); output( "\tsubl $4,%%esp\n" ); output( "\tpushl %%eax\n" ); } else { output( "\tsubl $8,%%esp\n" ); output( "\tandl $~15,%%esp\n" ); output( "\taddl $8,%%esp\n" ); } /* Call relay routine (which will call the API entry point) */ output( "\tleal 0x30(%%edx),%%eax\n" ); /* sizeof(STACK16FRAME) */ output( "\tpushl %%eax\n" ); output( "\tpushl 0x26(%%edx)\n"); /* FIELD_OFFSET(STACK16FRAME,entry_point) */ output( "\tcall *0x20(%%edx)\n"); /* FIELD_OFFSET(STACK16FRAME,relay) */ if ( reg_func ) { output( "\tleal -748(%%ebp),%%ebx\n" ); /* sizeof(CONTEXT) + FIELD_OFFSET(STACK32FRAME,ebp) */ /* Switch stack back */ output( "\t.byte 0x64\n\tmovw (%d), %%ss\n", STACKOFFSET+2 ); output( "\t.byte 0x64\n\tmovzwl (%d), %%esp\n", STACKOFFSET ); output( "\t.byte 0x64\n\tpopl (%d)\n", STACKOFFSET ); /* Get return address to CallFrom16 stub */ output( "\taddw $0x14,%%sp\n" ); /* FIELD_OFFSET(STACK16FRAME,callfrom_ip)-4 */ output( "\tpopl %%eax\n" ); output( "\tpopl %%edx\n" ); /* Restore all registers from CONTEXT */ output( "\tmovw 0xc8(%%ebx),%%ss\n"); /* SegSs */ output( "\tmovl 0xc4(%%ebx),%%esp\n"); /* Esp */ output( "\taddl $4, %%esp\n" ); /* room for final return address */ output( "\tpushw 0xbc(%%ebx)\n"); /* SegCs */ output( "\tpushw 0xb8(%%ebx)\n"); /* Eip */ output( "\tpushl %%edx\n" ); output( "\tpushl %%eax\n" ); output( "\tpushl 0xc0(%%ebx)\n"); /* EFlags */ output( "\tpushl 0x98(%%ebx)\n"); /* SegDs */ output( "\tpushl 0x94(%%ebx)\n"); /* SegEs */ output( "\tpopl %%es\n" ); output( "\tpushl 0x90(%%ebx)\n"); /* SegFs */ output( "\tpopl %%fs\n" ); output( "\tpushl 0x8c(%%ebx)\n"); /* SegGs */ output( "\tpopl %%gs\n" ); output( "\tmovl 0xb4(%%ebx),%%ebp\n"); /* Ebp */ output( "\tmovl 0xa0(%%ebx),%%esi\n"); /* Esi */ output( "\tmovl 0x9c(%%ebx),%%edi\n"); /* Edi */ output( "\tmovl 0xb0(%%ebx),%%eax\n"); /* Eax */ output( "\tmovl 0xa8(%%ebx),%%edx\n"); /* Edx */ output( "\tmovl 0xac(%%ebx),%%ecx\n"); /* Ecx */ output( "\tmovl 0xa4(%%ebx),%%ebx\n"); /* Ebx */ output( "\tpopl %%ds\n" ); output( "\tpopfl\n" ); output( "\tlret\n" ); output_function_size( "__wine_call_from_16_regs" ); } else { /* Switch stack back */ output( "\t.byte 0x64\n\tmovw (%d), %%ss\n", STACKOFFSET+2 ); output( "\t.byte 0x64\n\tmovzwl (%d), %%esp\n", STACKOFFSET ); output( "\t.byte 0x64\n\tpopl (%d)\n", STACKOFFSET ); /* Restore registers */ output( "\tpopl %%edx\n" ); output( "\tpopl %%ecx\n" ); output( "\tpopl %%ebp\n" ); output( "\tpopw %%ds\n" ); output( "\tpopw %%es\n" ); output( "\tpopw %%fs\n" ); output( "\tpopw %%gs\n" ); /* Return to return stub which will return to caller */ output( "\tlret $12\n" ); output_function_size( "__wine_call_from_16" ); } }
/******************************************************************* * output_call16_function * * Build a 16-bit-to-Wine callback glue function. * * The generated routines are intended to be used as argument conversion * routines to be called by the CallFrom16... core. Thus, the prototypes of * the generated routines are (see also CallFrom16): * * extern WORD WINAPI __wine_spec_call16_C_xxx( FARPROC func, LPBYTE args ); * extern LONG WINAPI __wine_spec_call16_C_xxx( FARPROC func, LPBYTE args ); * extern void WINAPI __wine_spec_call16_C_xxx_regs( FARPROC func, LPBYTE args, CONTEXT86 *context ); * * where 'C' is the calling convention ('p' for pascal or 'c' for cdecl), * and each 'x' is an argument ('w'=word, 's'=signed word, 'l'=long, * 'p'=linear pointer, 't'=linear pointer to null-terminated string, * 'T'=segmented pointer to null-terminated string). * * The generated routines fetch the arguments from the 16-bit stack (pointed * to by 'args'); the offsets of the single argument values are computed * according to the calling convention and the argument types. Then, the * 32-bit entry point is called with these arguments. * * For register functions, the arguments (if present) are converted just * the same as for normal functions, but in addition the CONTEXT86 pointer * filled with the current register values is passed to the 32-bit routine. */ static void output_call16_function( ORDDEF *odp ) { char *name; int i, pos, stack_words; int argsize = get_function_argsize( odp ); int needs_ldt = (strpbrk( get_args_str( odp ), "pt" ) != NULL); name = strmake( ".L__wine_spec_call16_%s", get_relay_name(odp) ); output( "\t.align %d\n", get_alignment(4) ); output( "\t%s\n", func_declaration(name) ); output( "%s:\n", name ); output_cfi( ".cfi_startproc" ); output( "\tpushl %%ebp\n" ); output_cfi( ".cfi_adjust_cfa_offset 4" ); output_cfi( ".cfi_rel_offset %%ebp,0" ); output( "\tmovl %%esp,%%ebp\n" ); output_cfi( ".cfi_def_cfa_register %%ebp" ); stack_words = 2; if (needs_ldt) { output( "\tpushl %%esi\n" ); output_cfi( ".cfi_rel_offset %%esi,-4" ); stack_words++; if (UsePIC) { output( "\tcall %s\n", asm_name("__wine_spec_get_pc_thunk_eax") ); output( "1:\tmovl wine_ldt_copy_ptr-1b(%%eax),%%esi\n" ); } else output( "\tmovl $%s,%%esi\n", asm_name("_imp__wine_ldt_copy") ); } /* preserve 16-byte stack alignment */ stack_words += odp->u.func.nb_args; for (i = 0; i < odp->u.func.nb_args; i++) if (odp->u.func.args[i] == ARG_DOUBLE || odp->u.func.args[i] == ARG_INT64) stack_words++; if ((odp->flags & FLAG_REGISTER) || (odp->type == TYPE_VARARGS)) stack_words++; if (stack_words % 4) output( "\tsubl $%d,%%esp\n", 16 - 4 * (stack_words % 4) ); if (odp->u.func.nb_args || odp->type == TYPE_VARARGS) output( "\tmovl 12(%%ebp),%%ecx\n" ); /* args */ if (odp->flags & FLAG_REGISTER) { output( "\tpushl 16(%%ebp)\n" ); /* context */ } else if (odp->type == TYPE_VARARGS) { output( "\tleal %d(%%ecx),%%eax\n", argsize ); output( "\tpushl %%eax\n" ); /* va_list16 */ } pos = (odp->type == TYPE_PASCAL) ? 0 : argsize; for (i = odp->u.func.nb_args - 1; i >= 0; i--) { switch (odp->u.func.args[i]) { case ARG_WORD: if (odp->type != TYPE_PASCAL) pos -= 2; output( "\tmovzwl %d(%%ecx),%%eax\n", pos ); output( "\tpushl %%eax\n" ); if (odp->type == TYPE_PASCAL) pos += 2; break; case ARG_SWORD: if (odp->type != TYPE_PASCAL) pos -= 2; output( "\tmovswl %d(%%ecx),%%eax\n", pos ); output( "\tpushl %%eax\n" ); if (odp->type == TYPE_PASCAL) pos += 2; break; case ARG_INT64: case ARG_DOUBLE: if (odp->type != TYPE_PASCAL) pos -= 4; output( "\tpushl %d(%%ecx)\n", pos ); if (odp->type == TYPE_PASCAL) pos += 4; /* fall through */ case ARG_LONG: case ARG_FLOAT: case ARG_SEGPTR: case ARG_SEGSTR: if (odp->type != TYPE_PASCAL) pos -= 4; output( "\tpushl %d(%%ecx)\n", pos ); if (odp->type == TYPE_PASCAL) pos += 4; break; case ARG_PTR: case ARG_STR: case ARG_WSTR: case ARG_INT128: if (odp->type != TYPE_PASCAL) pos -= 4; output( "\tmovzwl %d(%%ecx),%%edx\n", pos + 2 ); /* sel */ output( "\tshr $3,%%edx\n" ); output( "\tmovzwl %d(%%ecx),%%eax\n", pos ); /* offset */ output( "\taddl (%%esi,%%edx,4),%%eax\n" ); output( "\tpushl %%eax\n" ); if (odp->type == TYPE_PASCAL) pos += 4; break; } } output( "\tcall *8(%%ebp)\n" ); if (needs_ldt) { output( "\tmovl -4(%%ebp),%%esi\n" ); output_cfi( ".cfi_same_value %%esi" ); } output( "\tleave\n" ); output_cfi( ".cfi_def_cfa %%esp,4" ); output_cfi( ".cfi_same_value %%ebp" ); output( "\tret\n" ); output_cfi( ".cfi_endproc" ); output_function_size( name ); free( name ); }
/* output the delayed import thunks of a Win32 module */ static void output_delayed_import_thunks( const DLLSPEC *spec ) { int i, idx, j, pos, extra_stack_storage = 0; static const char delayed_import_loaders[] = "__wine_spec_delayed_import_loaders"; static const char delayed_import_thunks[] = "__wine_spec_delayed_import_thunks"; if (!nb_delayed) return; output( "\n/* delayed import thunks */\n\n" ); output( "\t.text\n" ); output( "\t.align %d\n", get_alignment(8) ); output( "%s:\n", asm_name(delayed_import_loaders)); output( "\t%s\n", func_declaration("__wine_delay_load_asm") ); output( "%s:\n", asm_name("__wine_delay_load_asm") ); output_cfi( ".cfi_startproc" ); switch(target_cpu) { case CPU_x86: output( "\tpushl %%ecx\n" ); output_cfi( ".cfi_adjust_cfa_offset 4" ); output( "\tpushl %%edx\n" ); output_cfi( ".cfi_adjust_cfa_offset 4" ); output( "\tpushl %%eax\n" ); output_cfi( ".cfi_adjust_cfa_offset 4" ); output( "\tcall %s\n", asm_name("__wine_spec_delay_load") ); output_cfi( ".cfi_adjust_cfa_offset -4" ); output( "\tpopl %%edx\n" ); output_cfi( ".cfi_adjust_cfa_offset -4" ); output( "\tpopl %%ecx\n" ); output_cfi( ".cfi_adjust_cfa_offset -4" ); output( "\tjmp *%%eax\n" ); break; case CPU_x86_64: output( "\tsubq $88,%%rsp\n" ); output_cfi( ".cfi_adjust_cfa_offset 88" ); output( "\tmovq %%rdx,80(%%rsp)\n" ); output( "\tmovq %%rcx,72(%%rsp)\n" ); output( "\tmovq %%r8,64(%%rsp)\n" ); output( "\tmovq %%r9,56(%%rsp)\n" ); output( "\tmovq %%r10,48(%%rsp)\n" ); output( "\tmovq %%r11,40(%%rsp)\n" ); output( "\tmovq %%rax,%%rcx\n" ); output( "\tcall %s\n", asm_name("__wine_spec_delay_load") ); output( "\tmovq 40(%%rsp),%%r11\n" ); output( "\tmovq 48(%%rsp),%%r10\n" ); output( "\tmovq 56(%%rsp),%%r9\n" ); output( "\tmovq 64(%%rsp),%%r8\n" ); output( "\tmovq 72(%%rsp),%%rcx\n" ); output( "\tmovq 80(%%rsp),%%rdx\n" ); output( "\taddq $88,%%rsp\n" ); output_cfi( ".cfi_adjust_cfa_offset -88" ); output( "\tjmp *%%rax\n" ); break; case CPU_SPARC: output( "\tsave %%sp, -96, %%sp\n" ); output( "\tcall %s\n", asm_name("__wine_spec_delay_load") ); output( "\tmov %%g1, %%o0\n" ); output( "\tjmp %%o0\n" ); output( "\trestore\n" ); break; case CPU_ARM: output( "\tstmfd SP!, {r4-r10,FP,LR}\n" ); output( "\tmov LR,PC\n"); output( "\tadd LR,LR,#8\n"); output( "\tldr PC,[PC,#-4]\n"); output( "\t.long %s\n", asm_name("__wine_spec_delay_load") ); output( "\tmov IP,r0\n"); output( "\tldmfd SP!, {r4-r10,FP,LR}\n" ); output( "\tldmfd SP!, {r0-r3}\n" ); output( "\tmov PC,IP\n"); break; case CPU_ARM64: output( "\tstp x29, x30, [sp,#-16]!\n" ); output( "\tmov x29, sp\n" ); output( "\tadr x9, 1f\n" ); output( "\tldur x9, [x9, #0]\n" ); output( "\tblr x9\n" ); output( "\tmov x9, x0\n" ); output( "\tldp x29, x30, [sp],#16\n" ); output( "\tldp x0, x1, [sp,#16]\n" ); output( "\tldp x2, x3, [sp,#32]\n" ); output( "\tldp x4, x5, [sp,#48]\n" ); output( "\tldp x6, x7, [sp],#80\n" ); output( "\tbr x9\n" ); /* or "ret x9" */ output( "1:\t.quad %s\n", asm_name("__wine_spec_delay_load") ); break; case CPU_POWERPC: if (target_platform == PLATFORM_APPLE) extra_stack_storage = 56; /* Save all callee saved registers into a stackframe. */ output( "\tstwu %s, -%d(%s)\n",ppc_reg(1), 48+extra_stack_storage, ppc_reg(1)); output( "\tstw %s, %d(%s)\n", ppc_reg(3), 4+extra_stack_storage, ppc_reg(1)); output( "\tstw %s, %d(%s)\n", ppc_reg(4), 8+extra_stack_storage, ppc_reg(1)); output( "\tstw %s, %d(%s)\n", ppc_reg(5), 12+extra_stack_storage, ppc_reg(1)); output( "\tstw %s, %d(%s)\n", ppc_reg(6), 16+extra_stack_storage, ppc_reg(1)); output( "\tstw %s, %d(%s)\n", ppc_reg(7), 20+extra_stack_storage, ppc_reg(1)); output( "\tstw %s, %d(%s)\n", ppc_reg(8), 24+extra_stack_storage, ppc_reg(1)); output( "\tstw %s, %d(%s)\n", ppc_reg(9), 28+extra_stack_storage, ppc_reg(1)); output( "\tstw %s, %d(%s)\n", ppc_reg(10),32+extra_stack_storage, ppc_reg(1)); output( "\tstw %s, %d(%s)\n", ppc_reg(11),36+extra_stack_storage, ppc_reg(1)); output( "\tstw %s, %d(%s)\n", ppc_reg(12),40+extra_stack_storage, ppc_reg(1)); /* r0 -> r3 (arg1) */ output( "\tmr %s, %s\n", ppc_reg(3), ppc_reg(0)); /* save return address */ output( "\tmflr %s\n", ppc_reg(0)); output( "\tstw %s, %d(%s)\n", ppc_reg(0), 44+extra_stack_storage, ppc_reg(1)); /* Call the __wine_delay_load function, arg1 is arg1. */ output( "\tbl %s\n", asm_name("__wine_spec_delay_load") ); /* Load return value from call into ctr register */ output( "\tmtctr %s\n", ppc_reg(3)); /* restore all saved registers and drop stackframe. */ output( "\tlwz %s, %d(%s)\n", ppc_reg(3), 4+extra_stack_storage, ppc_reg(1)); output( "\tlwz %s, %d(%s)\n", ppc_reg(4), 8+extra_stack_storage, ppc_reg(1)); output( "\tlwz %s, %d(%s)\n", ppc_reg(5), 12+extra_stack_storage, ppc_reg(1)); output( "\tlwz %s, %d(%s)\n", ppc_reg(6), 16+extra_stack_storage, ppc_reg(1)); output( "\tlwz %s, %d(%s)\n", ppc_reg(7), 20+extra_stack_storage, ppc_reg(1)); output( "\tlwz %s, %d(%s)\n", ppc_reg(8), 24+extra_stack_storage, ppc_reg(1)); output( "\tlwz %s, %d(%s)\n", ppc_reg(9), 28+extra_stack_storage, ppc_reg(1)); output( "\tlwz %s, %d(%s)\n", ppc_reg(10),32+extra_stack_storage, ppc_reg(1)); output( "\tlwz %s, %d(%s)\n", ppc_reg(11),36+extra_stack_storage, ppc_reg(1)); output( "\tlwz %s, %d(%s)\n", ppc_reg(12),40+extra_stack_storage, ppc_reg(1)); /* Load return value from call into return register */ output( "\tlwz %s, %d(%s)\n", ppc_reg(0), 44+extra_stack_storage, ppc_reg(1)); output( "\tmtlr %s\n", ppc_reg(0)); output( "\taddi %s, %s, %d\n", ppc_reg(1), ppc_reg(1), 48+extra_stack_storage); /* branch to ctr register. */ output( "\tbctr\n"); break; } output_cfi( ".cfi_endproc" ); output_function_size( "__wine_delay_load_asm" ); output( "\n" ); for (i = idx = 0; i < nb_imports; i++) { if (!dll_imports[i]->delay) continue; for (j = 0; j < dll_imports[i]->nb_imports; j++) { ORDDEF *odp = dll_imports[i]->imports[j]; const char *name = odp->name ? odp->name : odp->export_name; output( ".L__wine_delay_imp_%d_%s:\n", i, name ); output_cfi( ".cfi_startproc" ); switch(target_cpu) { case CPU_x86: output( "\tmovl $%d, %%eax\n", (idx << 16) | j ); output( "\tjmp %s\n", asm_name("__wine_delay_load_asm") ); break; case CPU_x86_64: output( "\tmovq $%d,%%rax\n", (idx << 16) | j ); output( "\tjmp %s\n", asm_name("__wine_delay_load_asm") ); break; case CPU_SPARC: output( "\tset %d, %%g1\n", (idx << 16) | j ); output( "\tb,a %s\n", asm_name("__wine_delay_load_asm") ); output( "\tnop\n" ); break; case CPU_ARM: output( "\tstmfd SP!, {r0-r3}\n" ); output( "\tmov r0, #%d\n", idx ); output( "\tmov r1, #16384\n" ); output( "\tmul r1, r0, r1\n" ); output( "\tmov r0, r1\n" ); output( "\tmov r1, #4\n" ); output( "\tmul r1, r0, r1\n" ); output( "\tmov r0, r1\n" ); output( "\tadd r0, #%d\n", j ); output( "\tldr PC,[PC,#-4]\n"); output( "\t.long %s\n", asm_name("__wine_delay_load_asm") ); break; case CPU_ARM64: output( "\tstp x6, x7, [sp,#-80]!\n" ); output( "\tstp x4, x5, [sp,#48]\n" ); output( "\tstp x2, x3, [sp,#32]\n" ); output( "\tstp x0, x1, [sp,#16]\n" ); output( "\tmov x0, #%d\n", idx ); output( "\tmov x1, #16384\n" ); output( "\tmul x1, x0, x1\n" ); output( "\tmov x0, x1\n" ); output( "\tmov x1, #4\n" ); output( "\tmul x1, x0, x1\n" ); output( "\tmov x0, x1\n" ); output( "\tadd x0, x0, #%d\n", j ); output( "\tadr x9, 1f\n" ); output( "\tldur x9, [x9, #0]\n" ); output( "\tbr x9\n" ); output( "1:\t.quad %s\n", asm_name("__wine_delay_load_asm") ); break; case CPU_POWERPC: switch(target_platform) { case PLATFORM_APPLE: /* On Darwin we can use r0 and r2 */ /* Upper part in r2 */ output( "\tlis %s, %d\n", ppc_reg(2), idx); /* Lower part + r2 -> r0, Note we can't use r0 directly */ output( "\taddi %s, %s, %d\n", ppc_reg(0), ppc_reg(2), j); output( "\tb %s\n", asm_name("__wine_delay_load_asm") ); break; default: /* On linux we can't use r2 since r2 is not a scratch register (hold the TOC) */ /* Save r13 on the stack */ output( "\taddi %s, %s, -0x4\n", ppc_reg(1), ppc_reg(1)); output( "\tstw %s, 0(%s)\n", ppc_reg(13), ppc_reg(1)); /* Upper part in r13 */ output( "\tlis %s, %d\n", ppc_reg(13), idx); /* Lower part + r13 -> r0, Note we can't use r0 directly */ output( "\taddi %s, %s, %d\n", ppc_reg(0), ppc_reg(13), j); /* Restore r13 */ output( "\tstw %s, 0(%s)\n", ppc_reg(13), ppc_reg(1)); output( "\taddic %s, %s, 0x4\n", ppc_reg(1), ppc_reg(1)); output( "\tb %s\n", asm_name("__wine_delay_load_asm") ); break; } break; } output_cfi( ".cfi_endproc" ); } idx++; } output_function_size( delayed_import_loaders ); output( "\n\t.align %d\n", get_alignment(get_ptr_size()) ); output( "%s:\n", asm_name(delayed_import_thunks)); for (i = pos = 0; i < nb_imports; i++) { if (!dll_imports[i]->delay) continue; for (j = 0; j < dll_imports[i]->nb_imports; j++, pos += get_ptr_size()) { ORDDEF *odp = dll_imports[i]->imports[j]; output_import_thunk( odp->name ? odp->name : odp->export_name, ".L__wine_delay_IAT", pos ); } } output_function_size( delayed_import_thunks ); }
/******************************************************************* * BuildCallTo16Core * * This routine builds the core routines used in 32->16 thunks: * * extern DWORD WINAPI wine_call_to_16( FARPROC16 target, DWORD cbArgs, PEXCEPTION_HANDLER handler ); * extern void WINAPI wine_call_to_16_regs( CONTEXT86 *context, DWORD cbArgs, PEXCEPTION_HANDLER handler ); * * These routines can be called directly from 32-bit code. * * All routines expect that the 16-bit stack contents (arguments) and the * return address (segptr to CallTo16_Ret) were already set up by the * caller; nb_args must contain the number of bytes to be conserved. The * 16-bit SS:SP will be set accordingly. * * All other registers are either taken from the CONTEXT86 structure * or else set to default values. The target routine address is either * given directly or taken from the CONTEXT86. */ static void BuildCallTo16Core( int reg_func ) { const char *name = reg_func ? "wine_call_to_16_regs" : "wine_call_to_16"; /* Function header */ function_header( name ); /* Function entry sequence */ output( "\tpushl %%ebp\n" ); output( "\tmovl %%esp, %%ebp\n" ); /* Save the 32-bit registers */ output( "\tpushl %%ebx\n" ); output( "\tpushl %%esi\n" ); output( "\tpushl %%edi\n" ); output( "\t.byte 0x64\n\tmov %%gs,(%d)\n", GS_OFFSET ); /* Setup exception frame */ output( "\t.byte 0x64\n\tpushl (%d)\n", STACKOFFSET ); output( "\tpushl 16(%%ebp)\n" ); /* handler */ output( "\t.byte 0x64\n\tpushl (0)\n" ); output( "\t.byte 0x64\n\tmovl %%esp,(0)\n" ); /* Call the actual CallTo16 routine (simulate a lcall) */ output( "\tpushl %%cs\n" ); output( "\tcall .L%s\n", name ); /* Remove exception frame */ output( "\t.byte 0x64\n\tpopl (0)\n" ); output( "\taddl $4, %%esp\n" ); output( "\t.byte 0x64\n\tpopl (%d)\n", STACKOFFSET ); if ( !reg_func ) { /* Convert return value */ output( "\tandl $0xffff,%%eax\n" ); output( "\tshll $16,%%edx\n" ); output( "\torl %%edx,%%eax\n" ); } else { /* * Modify CONTEXT86 structure to contain new values * * NOTE: We restore only EAX, EBX, EDX, EDX, EBP, and ESP. * The segment registers as well as ESI and EDI should * not be modified by a well-behaved 16-bit routine in * any case. [If necessary, we could restore them as well, * at the cost of a somewhat less efficient return path.] */ output( "\tmovl %d(%%esp), %%edi\n", STACK32OFFSET(target) - STACK32OFFSET(edi)); /* everything above edi has been popped already */ output( "\tmovl %%eax, %d(%%edi)\n", CONTEXTOFFSET(Eax) ); output( "\tmovl %%ebx, %d(%%edi)\n", CONTEXTOFFSET(Ebx) ); output( "\tmovl %%ecx, %d(%%edi)\n", CONTEXTOFFSET(Ecx) ); output( "\tmovl %%edx, %d(%%edi)\n", CONTEXTOFFSET(Edx) ); output( "\tmovl %%ebp, %d(%%edi)\n", CONTEXTOFFSET(Ebp) ); output( "\tmovl %%esi, %d(%%edi)\n", CONTEXTOFFSET(Esp) ); /* The return glue code saved %esp into %esi */ } /* Restore the 32-bit registers */ output( "\tpopl %%edi\n" ); output( "\tpopl %%esi\n" ); output( "\tpopl %%ebx\n" ); /* Function exit sequence */ output( "\tpopl %%ebp\n" ); output( "\tret $12\n" ); /* Start of the actual CallTo16 routine */ output( ".L%s:\n", name ); /* Switch to the 16-bit stack */ output( "\tmovl %%esp,%%edx\n" ); output( "\t.byte 0x64\n\tmovw (%d),%%ss\n", STACKOFFSET + 2); output( "\t.byte 0x64\n\tmovw (%d),%%sp\n", STACKOFFSET ); output( "\t.byte 0x64\n\tmovl %%edx,(%d)\n", STACKOFFSET ); /* Make %bp point to the previous stackframe (built by CallFrom16) */ output( "\tmovzwl %%sp,%%ebp\n" ); output( "\tleal %d(%%ebp),%%ebp\n", STACK16OFFSET(bp) ); /* Add the specified offset to the new sp */ output( "\tsubw %d(%%edx), %%sp\n", STACK32OFFSET(nb_args) ); if (reg_func) { /* Push the called routine address */ output( "\tmovl %d(%%edx),%%edx\n", STACK32OFFSET(target) ); output( "\tpushw %d(%%edx)\n", CONTEXTOFFSET(SegCs) ); output( "\tpushw %d(%%edx)\n", CONTEXTOFFSET(Eip) ); /* Get the registers */ output( "\tpushw %d(%%edx)\n", CONTEXTOFFSET(SegDs) ); output( "\tpushl %d(%%edx)\n", CONTEXTOFFSET(SegEs) ); output( "\tpopl %%es\n" ); output( "\tpushl %d(%%edx)\n", CONTEXTOFFSET(SegFs) ); output( "\tpopl %%fs\n" ); output( "\tpushl %d(%%edx)\n", CONTEXTOFFSET(SegGs) ); output( "\tpopl %%gs\n" ); output( "\tmovl %d(%%edx),%%ebp\n", CONTEXTOFFSET(Ebp) ); output( "\tmovl %d(%%edx),%%esi\n", CONTEXTOFFSET(Esi) ); output( "\tmovl %d(%%edx),%%edi\n", CONTEXTOFFSET(Edi) ); output( "\tmovl %d(%%edx),%%eax\n", CONTEXTOFFSET(Eax) ); output( "\tmovl %d(%%edx),%%ebx\n", CONTEXTOFFSET(Ebx) ); output( "\tmovl %d(%%edx),%%ecx\n", CONTEXTOFFSET(Ecx) ); output( "\tmovl %d(%%edx),%%edx\n", CONTEXTOFFSET(Edx) ); /* Get the 16-bit ds */ output( "\tpopw %%ds\n" ); } else /* not a register function */ { /* Push the called routine address */ output( "\tpushl %d(%%edx)\n", STACK32OFFSET(target) ); /* Set %fs and %gs to the value saved by the last CallFrom16 */ output( "\tpushw %d(%%ebp)\n", STACK16OFFSET(fs)-STACK16OFFSET(bp) ); output( "\tpopw %%fs\n" ); output( "\tpushw %d(%%ebp)\n", STACK16OFFSET(gs)-STACK16OFFSET(bp) ); output( "\tpopw %%gs\n" ); /* Set %ds and %es (and %ax just in case) equal to %ss */ output( "\tmovw %%ss,%%ax\n" ); output( "\tmovw %%ax,%%ds\n" ); output( "\tmovw %%ax,%%es\n" ); } /* Jump to the called routine */ output( "\t.byte 0x66\n" ); output( "\tlret\n" ); /* Function footer */ output_function_size( name ); }