OP_MOVS,
OP_CMPS,
OP_SCAS,
+ OP_RDTSC,
+ OP_CPUID,
OP_STD,
OP_CLD,
OP_RET,
enum sct_func_attr {
SCTFA_CLEAR_SF = (1 << 0), // clear stack frame
SCTFA_CLEAR_REGS = (1 << 1), // clear registers (mask)
- SCTFA_RM_REGS = (1 << 2), // don't emit regs
+ SCTFA_RM_REGS = (1 << 2), // don't emit regs (mask)
SCTFA_NOWARN = (1 << 3), // don't try to detect problems
+ SCTFA_ARGFRAME = (1 << 4), // copy all args to a struct, in order
};
enum x87_const {
static int g_skip_func;
static int g_allow_regfunc;
static int g_allow_user_icall;
+static int g_nowarn_reguse;
static int g_quiet_pp;
static int g_header_mode;
};
#define mxAX (1 << xAX)
+#define mxBX (1 << xBX)
#define mxCX (1 << xCX)
#define mxDX (1 << xDX)
#define mxSP (1 << xSP)
{ "repz", OPF_REP|OPF_REPZ },
{ "repne", OPF_REP|OPF_REPNZ },
{ "repnz", OPF_REP|OPF_REPNZ },
- { "lock", OPF_LOCK }, // ignored for now..
+ { "lock", OPF_LOCK },
};
#define OPF_CJMP_CC (OPF_JMP|OPF_CJMP|OPF_CC)
{ "scasb",OP_SCAS, 0, 0, OPF_DATA|OPF_FLAGS },
{ "scasw",OP_SCAS, 0, 0, OPF_DATA|OPF_FLAGS },
{ "scasd",OP_SCAS, 0, 0, OPF_DATA|OPF_FLAGS },
+ { "rdtsc",OP_RDTSC, 0, 0, OPF_DATA },
+ { "cpuid",OP_CPUID, 0, 0, OPF_DATA },
{ "std", OP_STD, 0, 0, OPF_DATA }, // special flag
{ "cld", OP_CLD, 0, 0, OPF_DATA },
{ "add", OP_ADD, 2, 2, OPF_DATA|OPF_FLAGS },
op->regmask_dst = op->regmask_src;
break;
+ case OP_RDTSC:
+ op->regmask_dst = mxAX | mxDX;
+ break;
+
+ case OP_CPUID:
+ // for now, ignore ecx dep for eax={4,7,b,d}
+ op->regmask_src = mxAX;
+ op->regmask_dst = mxAX | mxBX | mxCX | mxDX;
+ break;
+
case OP_LOOP:
op->regmask_dst = 1 << xCX;
// fallthrough
const char *prefix = "";
const char *bp_arg = NULL;
char ofs_reg[16] = { 0, };
+ char argname[8];
int i, arg_i, arg_s;
int unaligned = 0;
int stack_ra = 0;
snprintf(buf, buf_size, "%sap", cast);
return -1;
}
- ferr(po, "offset %d (%s,%d) doesn't map to any arg\n",
+ ferr(po, "offset 0x%x (%s,%d) doesn't map to any arg\n",
offset, bp_arg, arg_i);
}
if (ofs_reg[0] != 0)
popr->is_ptr = g_func_pp->arg[i].type.is_ptr;
retval = i;
+ snprintf(argname, sizeof(argname), "%sa%d",
+ g_sct_func_attr & SCTFA_ARGFRAME ? "af." : "", i + 1);
+
switch (popr->lmod)
{
case OPLM_BYTE:
if (is_lea)
ferr(po, "lea/byte to arg?\n");
if (is_src && (offset & 3) == 0)
- snprintf(buf, buf_size, "%sa%d",
- simplify_cast(cast, "(u8)"), i + 1);
+ snprintf(buf, buf_size, "%s%s",
+ simplify_cast(cast, "(u8)"), argname);
else
- snprintf(buf, buf_size, "%sBYTE%d(a%d)",
- cast, offset & 3, i + 1);
+ snprintf(buf, buf_size, "%sBYTE%d(%s)",
+ cast, offset & 3, argname);
break;
case OPLM_WORD:
if (!is_src) {
if (offset & 2)
ferr(po, "problematic arg store\n");
- snprintf(buf, buf_size, "%s((char *)&a%d + 1)",
- simplify_cast(cast, "*(u16 *)"), i + 1);
+ snprintf(buf, buf_size, "%s((char *)&%s + 1)",
+ simplify_cast(cast, "*(u16 *)"), argname);
}
else
ferr(po, "unaligned arg word load\n");
}
else if (is_src && (offset & 2) == 0)
- snprintf(buf, buf_size, "%sa%d",
- simplify_cast(cast, "(u16)"), i + 1);
+ snprintf(buf, buf_size, "%s%s",
+ simplify_cast(cast, "(u16)"), argname);
else
- snprintf(buf, buf_size, "%s%sWORD(a%d)",
- cast, (offset & 2) ? "HI" : "LO", i + 1);
+ snprintf(buf, buf_size, "%s%sWORD(%s)",
+ cast, (offset & 2) ? "HI" : "LO", argname);
break;
case OPLM_DWORD:
if (offset & 3) {
unaligned = 1;
if (is_lea)
- snprintf(buf, buf_size, "(u32)&a%d + %d",
- i + 1, offset & 3);
+ snprintf(buf, buf_size, "(u32)&%s + %d",
+ argname, offset & 3);
else if (!is_src)
ferr(po, "unaligned arg store\n");
else {
// mov edx, [ebp+arg_4+2]; movsx ecx, dx
- snprintf(buf, buf_size, "%s(a%d >> %d)",
- prefix, i + 1, (offset & 3) * 8);
+ snprintf(buf, buf_size, "%s(%s >> %d)",
+ prefix, argname, (offset & 3) * 8);
}
}
else {
- snprintf(buf, buf_size, "%s%sa%d",
- prefix, is_lea ? "&" : "", i + 1);
+ snprintf(buf, buf_size, "%s%s%s",
+ prefix, is_lea ? "&" : "", argname);
}
break;
ferr_assert(po, !(offset & 7));
if (cast[0])
prefix = cast;
- snprintf(buf, buf_size, "%s%sa%d",
- prefix, is_lea ? "&" : "", i + 1);
+ snprintf(buf, buf_size, "%s%s%s",
+ prefix, is_lea ? "&" : "", argname);
break;
default:
// note: this skips over calls and rm'd stuff assuming they're handled
// so it's intended to use at one of final passes
+// exception: doesn't skip OPF_RSAVE stuff
static int scan_for_pop(int i, int opcnt, int magic, int reg,
- int depth, int seen_noreturn, int flags_set)
+ int depth, int seen_noreturn, int save_level, int flags_set)
{
struct parsed_op *po;
int relevant;
po->cc_scratch = magic;
if (po->flags & OPF_TAIL) {
- if (po->op == OP_CALL) {
- if (po->pp != NULL && po->pp->is_noreturn)
- seen_noreturn = 1;
- else
+ if (po->op == OP_CALL && po->pp != NULL && po->pp->is_noreturn) {
+ // msvc sometimes generates stack cleanup code after
+ // noreturn, set a flag and continue
+ seen_noreturn = 1;
+
+ // ... but stop if there is another path to next insn -
+ // if msvc skipped something stack tracking may mess up
+ if (i + 1 < opcnt && g_labels[i + 1] != NULL)
goto out;
}
else
goto out;
}
- if (po->flags & (OPF_RMD|OPF_DONE|OPF_FARG))
+ if (po->flags & OPF_FARG)
continue;
+ if (po->flags & (OPF_RMD|OPF_DONE)) {
+ if (!(po->flags & OPF_RSAVE))
+ continue;
+ // reprocess, there might be another push in some "parallel"
+ // path that took a pop what we should also take
+ }
if ((po->flags & OPF_JMP) && po->op != OP_CALL) {
if (po->btj != NULL) {
for (j = 0; j < po->btj->count; j++) {
check_i(po, po->btj->d[j].bt_i);
ret |= scan_for_pop(po->btj->d[j].bt_i, opcnt, magic, reg,
- depth, seen_noreturn, flags_set);
+ depth, seen_noreturn, save_level, flags_set);
if (ret < 0)
return ret; // dead end
}
check_i(po, po->bt_i);
if (po->flags & OPF_CJMP) {
ret |= scan_for_pop(po->bt_i, opcnt, magic, reg,
- depth, seen_noreturn, flags_set);
+ depth, seen_noreturn, save_level, flags_set);
if (ret < 0)
return ret; // dead end
}
}
else if (po->op == OP_POP) {
if (relevant && depth == 0) {
+ if (flags_set == 0 && save_level > 0) {
+ ret = scan_for_pop(i + 1, opcnt, magic, reg,
+ depth, seen_noreturn, save_level - 1, flags_set);
+ if (ret != 1)
+ // no pop for other levels, current one must be false
+ return -1;
+ }
po->flags |= flags_set;
return 1;
}
char buf[256];
char *p;
+ if (po->pp != NULL && (po->flags & OPF_DATA)) {
+ // hint given in asm
+ return po->pp;
+ }
+
// maybe an arg of g_func?
if (opr->type == OPT_REGMEM && is_stack_access(po, opr))
{
&& IS(po->operand[0].name, g_labels[l]))
{
if (l == i + 1 && po->op == OP_JMP) {
- // yet another alignment type..
- po->flags |= OPF_RMD|OPF_DONE;
+ // yet another alignment type...
+ po->flags |= OPF_RMD | OPF_DONE;
+ po->flags &= ~OPF_JMP;
+ po->op = OP_NOP;
break;
}
add_label_ref(&g_label_refs[l], i);
eliminate_seh_finally(opcnt);
}
+// check for prologue of many pushes and epilogue with pops
+static void check_simple_sequence(int opcnt, int *fsz)
+{
+ int found = 0;
+ int seq_len;
+ int seq_p;
+ int seq[4];
+ int reg;
+ int i, j;
+
+ for (i = 0; i < opcnt && i < ARRAY_SIZE(seq); i++) {
+ if (ops[i].op != OP_PUSH || ops[i].operand[0].type != OPT_REG)
+ break;
+ reg = ops[i].operand[0].reg;
+ if (reg != xBX && reg != xSI && reg != xDI && reg != xBP)
+ break;
+ for (j = 0; j < i; j++)
+ if (seq[j] == reg)
+ break;
+ if (j != i)
+ // probably something else is going on here
+ break;
+ seq[i] = reg;
+ }
+ seq_len = i;
+ if (seq_len == 0)
+ return;
+
+ for (; i < opcnt && seq_len > 0; i++) {
+ if (!(ops[i].flags & OPF_TAIL))
+ continue;
+
+ for (j = i - 1, seq_p = 0; j >= 0 && seq_p < seq_len; j--) {
+ if (ops[j].op != OP_POP || ops[j].operand[0].type != OPT_REG)
+ break;
+ if (ops[j].operand[0].reg != seq[seq_p])
+ break;
+ seq_p++;
+ }
+ found = seq_len = seq_p;
+ }
+ if (!found)
+ return;
+
+ for (i = 0; i < seq_len; i++)
+ ops[i].flags |= OPF_RMD | OPF_DONE | OPF_NOREGS;
+
+ for (; i < opcnt && seq_len > 0; i++) {
+ if (!(ops[i].flags & OPF_TAIL))
+ continue;
+
+ for (j = i - 1, seq_p = 0; j >= 0 && seq_p < seq_len; j--) {
+ ops[j].flags |= OPF_RMD | OPF_DONE | OPF_NOREGS;
+ seq_p++;
+ }
+ }
+
+ // unlike pushes after sub esp,
+ // IDA treats pushes like this as part of var area
+ *fsz += seq_len * 4;
+}
+
static int scan_prologue(int i, int opcnt, int *ecx_push, int *esp_sub)
{
- int j;
+ const char *name;
+ int j, len, ret;
for (; i < opcnt; i++)
if (!(ops[i].flags & OPF_DONE))
for (; i < opcnt; i++) {
if (i > 0 && g_labels[i] != NULL)
break;
- if (ops[i].op == OP_PUSH || (ops[i].flags & (OPF_JMP|OPF_TAIL)))
+ if (ops[i].flags & (OPF_JMP|OPF_TAIL))
+ break;
+ if (ops[i].flags & OPF_DONE)
+ continue;
+ if (ops[i].op == OP_PUSH)
break;
if (ops[i].op == OP_SUB && ops[i].operand[0].reg == xSP
&& ops[i].operand[1].type == OPT_CONST)
*esp_sub = 1;
break;
}
+ if (ops[i].op == OP_LEA && ops[i].operand[0].reg == xSP
+ && ops[i].operand[1].type == OPT_REGMEM
+ && IS_START(ops[i].operand[1].name, "esp-"))
+ {
+ name = ops[i].operand[1].name;
+ ret = sscanf(name, "esp-%x%n", &j, &len);
+ ferr_assert(&ops[i], ret == 1 && len == strlen(name));
+ g_stack_fsz += j;
+ ops[i].flags |= OPF_RMD | OPF_DONE | OPF_NOREGS;
+ i++;
+ *esp_sub = 1;
+ break;
+ }
if (ops[i].op == OP_MOV && ops[i].operand[0].reg == xAX
&& ops[i].operand[1].type == OPT_CONST)
{
ops[j].flags |= OPF_RMD | OPF_DONE | OPF_NOREGS;
i = j + 1;
*esp_sub = 1;
+ break;
}
- break;
}
}
{
int ecx_push = 0, esp_sub = 0, pusha = 0;
int sandard_epilogue;
- int found;
+ int found, ret, len;
+ int push_fsz = 0;
int i, j, l;
if (g_seh_found == 2) {
}
// non-bp frame
+ check_simple_sequence(opcnt, &push_fsz);
i = scan_prologue(0, opcnt, &ecx_push, &esp_sub);
if (ecx_push && !esp_sub) {
for (; j >= 0; j--) {
if (ops[j].op != OP_MOV)
break;
- if (ops[j].operand[0].type != OPT_REGMEM)
- break;
- if (strstr(ops[j].operand[0].name, "arg_") == NULL)
- break;
+ if (ops[j].operand[0].type == OPT_REGMEM
+ && strstr(ops[j].operand[0].name, "arg_") != NULL)
+ continue;
+ if (ops[j].operand[0].type == OPT_REG)
+ continue; // assume arg-reg mov
+ break;
}
}
+ for (; j >= 0; j--) {
+ if ((ops[j].flags & (OPF_RMD | OPF_DONE | OPF_NOREGS)) !=
+ (OPF_RMD | OPF_DONE | OPF_NOREGS))
+ break;
+ }
+
if (ecx_push > 0 && !esp_sub) {
for (l = 0; l < ecx_push && j >= 0; l++) {
if (ops[j].op == OP_POP && IS(opr_name(&ops[j], 0), "ecx"))
}
if (esp_sub) {
- if (ops[j].op != OP_ADD
- || !IS(opr_name(&ops[j], 0), "esp")
- || ops[j].operand[1].type != OPT_CONST)
+ if (ops[j].op == OP_ADD
+ && IS(opr_name(&ops[j], 0), "esp")
+ && ops[j].operand[1].type == OPT_CONST)
{
- if (i < opcnt && ops[i].op == OP_CALL
- && ops[i].pp != NULL && ops[i].pp->is_noreturn)
- {
- // noreturn tailcall with no epilogue
- i++;
- found = 1;
- continue;
- }
- ferr(&ops[j], "'add esp' expected\n");
- }
-
- if (ops[j].operand[1].val < g_stack_fsz)
- ferr(&ops[j], "esp adj is too low (need %d)\n", g_stack_fsz);
+ if (ops[j].operand[1].val < g_stack_fsz)
+ ferr(&ops[j], "esp adj is too low (need %d)\n", g_stack_fsz);
- ops[j].operand[1].val -= g_stack_fsz; // for stack arg scanner
- if (ops[j].operand[1].val == 0)
+ ops[j].operand[1].val -= g_stack_fsz; // for stack arg scanner
+ if (ops[j].operand[1].val == 0)
+ ops[j].flags |= OPF_RMD | OPF_DONE | OPF_NOREGS;
+ found = 1;
+ }
+ else if (ops[j].op == OP_LEA && ops[j].operand[0].reg == xSP
+ && ops[j].operand[1].type == OPT_REGMEM
+ && IS_START(ops[j].operand[1].name, "esp+"))
+ {
+ const char *name = ops[j].operand[1].name;
+ ret = sscanf(name, "esp+%x%n", &l, &len);
+ ferr_assert(&ops[j], ret == 1 && len == strlen(name));
+ ferr_assert(&ops[j], l <= g_stack_fsz);
ops[j].flags |= OPF_RMD | OPF_DONE | OPF_NOREGS;
- found = 1;
+ found = 1;
+ }
+ else if (i < opcnt && ops[i].op == OP_CALL
+ && ops[i].pp != NULL && ops[i].pp->is_noreturn)
+ {
+ // noreturn tailcall with no epilogue
+ found = 1;
+ }
+ else
+ ferr(&ops[j], "'add esp' expected\n");
}
i++;
if (!found)
ferr(ops, "missing esp epilogue\n");
}
+
+ if (g_stack_fsz != 0)
+ // see check_simple_sequence
+ g_stack_fsz += push_fsz;
}
// find an instruction that changed opr before i op
if (ret < 0)
return ret;
+ if (pp->is_unresolved) {
+ pp->argc += ret;
+ pp->argc_stack += ret;
+ for (a = 0; a < pp->argc; a++)
+ if (pp->arg[a].type.name == NULL)
+ pp->arg[a].type.name = strdup("int");
+ }
+
if (arg_grp != 0) {
// propagate arg_grp
for (a = 0; a < pp->argc; a++) {
}
}
- if (pp->is_unresolved) {
- pp->argc += ret;
- pp->argc_stack += ret;
- for (a = 0; a < pp->argc; a++)
- if (pp->arg[a].type.name == NULL)
- pp->arg[a].type.name = strdup("int");
- }
-
return ret;
}
&& !g_func_pp->is_userstack
&& po->operand[0].type == OPT_REG)
{
+ int save_level = 0;
+
reg = po->operand[0].reg;
ferr_assert(po, reg >= 0);
if (regmask_now & (1 << reg)) {
already_saved = regmask_save_now & (1 << reg);
flags_set = OPF_RSAVE | OPF_DONE;
+ save_level++;
}
- ret = scan_for_pop(i + 1, opcnt, i + opcnt * 3, reg, 0, 0, 0);
+ ret = scan_for_pop(i + 1, opcnt, i + opcnt * 3,
+ reg, 0, 0, save_level, 0);
if (ret == 1) {
scan_for_pop(i + 1, opcnt, i + opcnt * 4,
- reg, 0, 0, flags_set);
+ reg, 0, 0, save_level, flags_set);
}
else {
ret = scan_for_pop_ret(i + 1, opcnt, po->operand[0].reg, 0);
int need_double = 0;
int stack_align = 0;
int stack_fsz_adj = 0;
+ int lock_handled = 0;
int regmask_save = 0; // used regs saved/restored in this func
int regmask_arg; // regs from this function args (fastcall, etc)
int regmask_ret; // regs needed on ret
regmask_stack |= 1 << tmp_op->operand[0].reg;
}
- if (!((regmask_stack & (1 << xCX))
- && (regmask_stack & (1 << xDX))))
+ // quick dumb check for potential reg-args
+ for (j = i - 1; j >= 0 && ops[j].op == OP_MOV; j--)
+ if (ops[j].operand[0].type == OPT_REG)
+ regmask_stack &= ~(1 << ops[j].operand[0].reg);
+
+ if ((regmask_stack & (mxCX|mxDX)) != (mxCX|mxDX)
+ && ((regmask | regmask_arg) & (mxCX|mxDX)))
{
if (pp->argc_stack != 0
- || ((regmask | regmask_arg) & ((1 << xCX)|(1 << xDX))))
+ || ((regmask | regmask_arg) & (mxCX|mxDX)))
{
pp_insert_reg_arg(pp, "ecx");
pp->is_fastcall = 1;
regmask |= 1 << xCX;
}
if (pp->argc_stack != 0
- || ((regmask | regmask_arg) & (1 << xDX)))
+ || ((regmask | regmask_arg) & mxDX))
{
pp_insert_reg_arg(pp, "edx");
regmask_init |= 1 << xDX;
pp->is_stdcall = 1;
}
if (!(po->flags & OPF_TAIL)
- && !(g_sct_func_attr & SCTFA_NOWARN))
+ && !(g_sct_func_attr & SCTFA_NOWARN) && !g_nowarn_reguse)
{
// treat al write as overwrite to avoid many false positives
if (IS(pp->ret_type.name, "void") || pp->ret_type.is_float) {
need_double = 1;
break;
+ case OP_RDTSC:
case OPP_ALLSHL:
case OPP_ALLSHR:
need_tmp64 = 1;
default:
break;
}
+ }
+
+ // pass8: final adjustments
+ for (i = 0; i < opcnt; i++)
+ {
+ po = &ops[i];
+ if (po->flags & (OPF_RMD|OPF_DONE))
+ continue;
- // this might need it's own pass...
if (po->op != OP_FST && po->p_argnum > 0)
save_arg_vars[po->p_arggrp] |= 1 << (po->p_argnum - 1);
had_decl = 1;
}
+ if ((g_sct_func_attr & SCTFA_ARGFRAME) && g_func_pp->argc_stack) {
+ fprintf(fout, " struct { u32 ");
+ for (i = j = 0; i < g_func_pp->argc; i++) {
+ if (g_func_pp->arg[i].reg != NULL)
+ continue;
+ if (j++ != 0)
+ fprintf(fout, ", ");
+ fprintf(fout, "a%d", i + 1);
+ }
+ fprintf(fout, "; } af = {\n ");
+ for (i = j = 0; i < g_func_pp->argc; i++) {
+ if (g_func_pp->arg[i].reg != NULL)
+ continue;
+ if (j++ != 0)
+ fprintf(fout, ", ");
+ if (g_func_pp->arg[i].type.is_ptr)
+ fprintf(fout, "(u32)");
+ fprintf(fout, "a%d", i + 1);
+ }
+ fprintf(fout, "\n };\n");
+ }
+
if (g_func_pp->is_userstack) {
fprintf(fout, " u32 fake_sf[US_SZ_%s / 4];\n", g_func_pp->name);
fprintf(fout, " u32 *esp = &fake_sf[sizeof(fake_sf) / 4];\n");
if (po->flags & OPF_RMD)
continue;
+ lock_handled = 0;
no_output = 0;
#define assert_operand_cnt(n_) \
fprintf(fout, " for (; ecx != 0; ecx--, edi %c= %d)\n",
(po->flags & OPF_DF) ? '-' : '+',
lmod_bytes(po, po->operand[1].lmod));
- fprintf(fout, " %sedi = eax;",
+ fprintf(fout, " %sedi = eax;\n",
lmod_cast_u_ptr(po, po->operand[1].lmod));
- strcpy(g_comment, "rep stos");
+ fprintf(fout, " barrier();");
+ strcpy(g_comment, "^ rep stos");
}
else {
assert_operand_cnt(2);
" for (; ecx != 0; ecx--, edi %c= %d, esi %c= %d)\n",
l, j, l, j);
fprintf(fout,
- " %sedi = %sesi;", buf1, buf1);
- strcpy(g_comment, "rep movs");
+ " %sedi = %sesi;\n", buf1, buf1);
+ // this can overwrite many variables
+ fprintf(fout, " barrier();");
+ strcpy(g_comment, "^ rep movs");
}
else {
assert_operand_cnt(2);
delayed_flag_op = NULL;
break;
+ case OP_RDTSC:
+ fprintf(fout, " tmp64 = ext_rdtsc();\n");
+ fprintf(fout, " edx = tmp64 >> 32;\n");
+ fprintf(fout, " eax = tmp64;");
+ break;
+
+ case OP_CPUID:
+ fprintf(fout, " ext_cpuid(&eax, &ebx, &ecx, &edx);");
+ break;
+
// arithmetic w/flags
case OP_AND:
if (po->operand[1].type == OPT_CONST && !po->operand[1].val)
out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]);
if (po->operand[0].type == OPT_REG) {
+ ferr_assert(po, !(po->flags & OPF_LOCK));
strcpy(buf2, po->op == OP_INC ? "++" : "--");
fprintf(fout, " %s%s;", buf1, buf2);
}
+ else if (po->flags & OPF_LOCK) {
+ out_src_opr(buf2, sizeof(buf2), po, &po->operand[0], "", 1);
+ fprintf(fout, " __sync_fetch_and_%s((%s *)(%s), 1);",
+ po->op == OP_INC ? "add" : "sub",
+ lmod_type_u(po, po->operand[0].lmod), buf2);
+ strcat(g_comment, " lock");
+ lock_handled = 1;
+ }
else {
strcpy(buf2, po->op == OP_INC ? "+" : "-");
fprintf(fout, " %s %s= 1;", buf1, buf2);
if (pfomask != 0)
ferr(po, "missed flag calc, pfomask=%x\n", pfomask);
+ if ((po->flags & OPF_LOCK) && !lock_handled)
+ ferr(po, "unhandled lock\n");
+
// see is delayed flag stuff is still valid
if (delayed_flag_op != NULL && delayed_flag_op != po) {
if (is_any_opr_modified(delayed_flag_op, po, 0))
unsigned int dep_resolved:1;
unsigned int is_stdcall:1;
unsigned int eax_pass:1; // returns without touching eax
+ unsigned int ptr_taken:1; // pointer taken of this func
struct func_proto_dep *dep_func;
int dep_func_cnt;
const struct parsed_proto *pp; // seed pp, if any
unsigned int ret_dep:1; // return from this is caller's return
unsigned int has_ret:1; // found from eax use after return
unsigned int has_ret64:1;
+ unsigned int ptr_taken:1; // pointer taken, not a call
};
static struct func_prototype *hg_fp;
return NULL;
}
-static void hg_fp_add_dep(struct func_prototype *fp, const char *name)
+static void hg_fp_add_dep(struct func_prototype *fp, const char *name,
+ unsigned int ptr_taken)
{
+ struct func_proto_dep * dep;
+
// is it a dupe?
- if (hg_fp_find_dep(fp, name))
+ dep = hg_fp_find_dep(fp, name);
+ if (dep != NULL && dep->ptr_taken == ptr_taken)
return;
if ((fp->dep_func_cnt & 0xff) == 0) {
sizeof(fp->dep_func[0]) * 0x100);
}
fp->dep_func[fp->dep_func_cnt].name = strdup(name);
+ fp->dep_func[fp->dep_func_cnt].ptr_taken = ptr_taken;
fp->dep_func_cnt++;
}
if (po->flags & OPF_DONE)
continue;
- ret = scan_for_pop(i + 1, opcnt, i + opcnt * 2, reg, 0, 0, 0);
+ ret = scan_for_pop(i + 1, opcnt, i + opcnt * 2,
+ reg, 0, 0, 0, 0);
if (ret == 1) {
regmask_save |= 1 << reg;
po->flags |= OPF_RMD;
- scan_for_pop(i + 1, opcnt, i + opcnt * 3, reg, 0, 0, OPF_RMD);
+ scan_for_pop(i + 1, opcnt, i + opcnt * 3,
+ reg, 0, 0, 0, OPF_RMD);
continue;
}
}
struct func_prototype *fp;
struct func_proto_dep *dep;
struct parsed_op *po;
+ const char *tmpname;
int regmask_dummy = 0;
int regmask_dep;
int regmask_use;
// pass3:
// - remove dead labels
// - collect calls
+ // - collect function ptr refs
for (i = 0; i < opcnt; i++)
{
if (g_labels[i] != NULL && g_label_refs[i].i == -1) {
if (po->op == OP_CALL) {
if (po->operand[0].type == OPT_LABEL)
- hg_fp_add_dep(fp, opr_name(po, 0));
+ hg_fp_add_dep(fp, opr_name(po, 0), 0);
else if (po->pp != NULL)
- hg_fp_add_dep(fp, po->pp->name);
+ hg_fp_add_dep(fp, po->pp->name, 0);
+ }
+ else if (po->op == OP_MOV && po->operand[1].type == OPT_OFFSET) {
+ tmpname = opr_name(po, 1);
+ if (IS_START(tmpname, "p_") || IS_START(tmpname, "sub_"))
+ hg_fp_add_dep(fp, tmpname, 1);
+ }
+ else if (po->op == OP_PUSH && po->operand[0].type == OPT_OFFSET) {
+ tmpname = opr_name(po, 0);
+ if (IS_START(tmpname, "p_") || IS_START(tmpname, "sub_"))
+ hg_fp_add_dep(fp, tmpname, 1);
}
}
dep->proto = bsearch(&fp_s, hg_fp, hg_fp_cnt,
sizeof(hg_fp[0]), hg_fp_cmp_name);
if (dep->proto != NULL) {
+ if (dep->ptr_taken) {
+ dep->proto->ptr_taken = 1;
+ continue;
+ }
+
if (!dep->proto->dep_resolved)
hg_fp_resolve_deps(dep->proto);
strcpy(fp_s.name, hg_refs[i]);
fp = bsearch(&fp_s, hg_fp, hg_fp_cnt,
sizeof(hg_fp[0]), hg_fp_cmp_name);
- if (fp == NULL)
- continue;
-
- if (fp->argc_stack != 0 && (fp->regmask_dep & (mxCX | mxDX)))
- fp->regmask_dep |= mxCX | mxDX;
+ if (fp != NULL)
+ fp->ptr_taken = 1;
}
}
regmask_dep = fp->regmask_dep;
argc_normal = fp->argc_stack;
+ if (fp->ptr_taken && regmask_dep
+ && (regmask_dep & ~(mxCX|mxDX)) == 0)
+ {
+ if ((regmask_dep & mxDX) || fp->argc_stack > 0)
+ regmask_dep |= mxCX | mxDX;
+ }
fprintf(fout, "%-5s",
fp->pp ? fp->pp->ret_type.name :
g_allow_regfunc = 1;
else if (IS(argv[arg], "-uc"))
g_allow_user_icall = 1;
+ else if (IS(argv[arg], "-wu"))
+ g_nowarn_reguse = 1;
else if (IS(argv[arg], "-m"))
multi_seg = 1;
else if (IS(argv[arg], "-hdr"))
}
if (argc < arg + 3) {
- printf("usage:\n%s [-v] [-rf] [-m] <.c> <.asm> <hdr.h> [rlist]*\n"
+ printf("usage:\n%s [options] <.c> <.asm> <hdr.h> [rlist]*\n"
"%s -hdr <out.h> <.asm> <seed.h> [rlist]*\n"
"options:\n"
" -hdr - header generation mode\n"
" -rf - allow unannotated indirect calls\n"
" -uc - allow ind. calls/refs to __usercall\n"
" -m - allow multiple .text sections\n"
+ " -wu - don't warn about bad reg use\n"
"[rlist] is a file with function names to skip,"
" one per line\n",
argv[0], argv[0]);
"clear_regmask",
"rm_regmask",
"nowarn",
+ "argframe",
};
// parse manual attribute-list comment
continue;
goto parse_words; // lame
}
- if (IS_START(p, "; sctproto:")) {
- sctproto = strdup(p + 11);
- }
else if (IS_START(p, "; sctend")) {
end = 1;
if (!pending_endp)
break;
}
+ else if (g_skip_func)
+ /* ignore remaining attrs */;
+ else if (IS_START(p, "; sctproto:")) {
+ sctproto = strdup(p + 11);
+ }
else if (IS_START(p, "; sctskip_start")) {
- if (in_func && !g_skip_func) {
+ if (in_func) {
if (!skip_code) {
ops[pi].op = OPP_ABORT;
ops[pi].asmln = asmln;
notaz.gp2x.de / ia32rtools.git / blobdiff