/* 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 );
}
/* 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" );
}
/* 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 );
}
/* 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" );
}
/*******************************************************************
* 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_relay_debug
*
* Output entry points for relay debugging
*/
static void output_relay_debug( DLLSPEC *spec )
{
int i, j;
unsigned int pos, args, flags;
/* first the table of entry point offsets */
output( "\t%s\n", get_asm_rodata_section() );
output( "\t.align %d\n", get_alignment(4) );
output( ".L__wine_spec_relay_entry_point_offsets:\n" );
for (i = spec->base; i <= spec->limit; i++)
{
ORDDEF *odp = spec->ordinals[i];
if (needs_relay( odp ))
output( "\t.long .L__wine_spec_relay_entry_point_%d-__wine_spec_relay_entry_points\n", i );
else
output( "\t.long 0\n" );
}
/* then the table of argument types */
output( "\t.align %d\n", get_alignment(4) );
output( ".L__wine_spec_relay_arg_types:\n" );
for (i = spec->base; i <= spec->limit; i++)
{
ORDDEF *odp = spec->ordinals[i];
unsigned int mask = 0;
if (needs_relay( odp ))
{
for (j = pos = 0; pos < 16 && j < odp->u.func.nb_args; j++)
{
switch (odp->u.func.args[j])
{
case ARG_STR: mask |= 1 << (2 * pos++); break;
case ARG_WSTR: mask |= 2 << (2 * pos++); break;
case ARG_INT64:
case ARG_DOUBLE: pos += 8 / get_ptr_size(); break;
case ARG_INT128: pos += (target_cpu == CPU_x86) ? 4 : 1; break;
default: pos++; break;
}
}
}
output( "\t.long 0x%08x\n", mask );
}
/* then the relay thunks */
output( "\t.text\n" );
output( "__wine_spec_relay_entry_points:\n" );
output( "\tnop\n" ); /* to avoid 0 offset */
for (i = spec->base; i <= spec->limit; i++)
{
ORDDEF *odp = spec->ordinals[i];
if (!needs_relay( odp )) continue;
output( "\t.align %d\n", get_alignment(4) );
output( ".L__wine_spec_relay_entry_point_%d:\n", i );
output_cfi( ".cfi_startproc" );
args = get_args_size(odp) / get_ptr_size();
flags = 0;
switch (target_cpu)
{
case CPU_x86:
if (odp->type == TYPE_THISCALL) /* add the this pointer */
{
output( "\tpopl %%eax\n" );
output( "\tpushl %%ecx\n" );
output( "\tpushl %%eax\n" );
flags |= 2;
}
if (odp->flags & FLAG_REGISTER)
output( "\tpushl %%eax\n" );
else
output( "\tpushl %%esp\n" );
output_cfi( ".cfi_adjust_cfa_offset 4" );
if (odp->flags & FLAG_RET64) flags |= 1;
output( "\tpushl $%u\n", (flags << 24) | (args << 16) | (i - spec->base) );
output_cfi( ".cfi_adjust_cfa_offset 4" );
if (UsePIC)
{
output( "\tcall %s\n", asm_name("__wine_spec_get_pc_thunk_eax") );
output( "1:\tleal .L__wine_spec_relay_descr-1b(%%eax),%%eax\n" );
}
else output( "\tmovl $.L__wine_spec_relay_descr,%%eax\n" );
output( "\tpushl %%eax\n" );
output_cfi( ".cfi_adjust_cfa_offset 4" );
if (odp->flags & FLAG_REGISTER)
{
output( "\tcall *8(%%eax)\n" );
}
else
{
output( "\tcall *4(%%eax)\n" );
output_cfi( ".cfi_adjust_cfa_offset -12" );
if (odp->type == TYPE_STDCALL || odp->type == TYPE_THISCALL)
output( "\tret $%u\n", args * get_ptr_size() );
else
output( "\tret\n" );
}
break;
case CPU_x86_64:
output( "\tsubq $40,%%rsp\n" );
output_cfi( ".cfi_adjust_cfa_offset 40" );
switch (args)
{
default: output( "\tmovq %%%s,72(%%rsp)\n", is_float_arg( odp, 3 ) ? "xmm3" : "r9" );
/* fall through */
case 3: output( "\tmovq %%%s,64(%%rsp)\n", is_float_arg( odp, 2 ) ? "xmm2" : "r8" );
/* fall through */
case 2: output( "\tmovq %%%s,56(%%rsp)\n", is_float_arg( odp, 1 ) ? "xmm1" : "rdx" );
/* fall through */
case 1: output( "\tmovq %%%s,48(%%rsp)\n", is_float_arg( odp, 0 ) ? "xmm0" : "rcx" );
/* fall through */
case 0: break;
}
output( "\tleaq 40(%%rsp),%%r8\n" );
output( "\tmovq $%u,%%rdx\n", (flags << 24) | (args << 16) | (i - spec->base) );
output( "\tleaq .L__wine_spec_relay_descr(%%rip),%%rcx\n" );
output( "\tcallq *8(%%rcx)\n" );
output( "\taddq $40,%%rsp\n" );
output_cfi( ".cfi_adjust_cfa_offset -40" );
output( "\tret\n" );
break;
default:
assert(0);
}
output_cfi( ".cfi_endproc" );
}
}
/*******************************************************************
* 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 );
}
