X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=tools%2Ftranslate.c;h=eeea0ac2ccf93df01bd481121fda69660dfefcc4;hb=b74c31e30301a2e5c03e2947a5017af28bf94e65;hp=4a0c10e02de3b53976404f7a106f4e7a863a0fc1;hpb=46411e6ce8749031ba7b19b9fd20c2a8c6968b22;p=ia32rtools.git diff --git a/tools/translate.c b/tools/translate.c index 4a0c10e..eeea0ac 100644 --- a/tools/translate.c +++ b/tools/translate.c @@ -26,6 +26,8 @@ static FILE *g_fhdr; exit(1); \ } while (0) +#include "masm_tools.h" + enum op_flags { OPF_RMD = (1 << 0), /* removed or optimized out */ OPF_DATA = (1 << 1), /* data processing - writes to dst opr */ @@ -38,7 +40,7 @@ enum op_flags { OPF_REPZ = (1 << 8), /* rep is repe/repz */ OPF_REPNZ = (1 << 9), /* rep is repne/repnz */ OPF_FARG = (1 << 10), /* push collected as func arg (no reuse) */ - OPF_EBP_S = (1 << 11), /* ebp used as scratch, not BP */ + OPF_EBP_S = (1 << 11), /* ebp used as scratch here, not BP */ }; enum op_op { @@ -124,6 +126,7 @@ struct parsed_opr { unsigned int type_from_var:1; // .. in header, sometimes wrong unsigned int size_mismatch:1; // type override differs from C unsigned int size_lt:1; // type override is larger than C + unsigned int had_ds:1; // had ds: prefix int reg; unsigned int val; char name[256]; @@ -145,7 +148,7 @@ struct parsed_op { }; // datap: -// OP_CALL - ptr to parsed_proto +// OP_CALL - parser proto hint (str), ptr to struct parsed_proto // (OPF_CC) - point to one of (OPF_FLAGS) that affects cc op struct parsed_equ { @@ -200,6 +203,7 @@ static int g_sp_frame; static int g_stack_frame_used; static int g_stack_fsz; static int g_ida_func_attr; +static int g_allow_regfunc; #define ferr(op_, fmt, ...) do { \ printf("error:%s:#%zd: '%s': " fmt, g_func, (op_) - ops, \ dump_op(op_), ##__VA_ARGS__); \ @@ -281,36 +285,6 @@ static int parse_reg(enum opr_lenmod *reg_lmod, const char *s) return -1; } -static unsigned long parse_number(const char *number) -{ - int len = strlen(number); - const char *p = number; - char *endp = NULL; - unsigned long ret; - int neg = 0; - int bad; - - if (*p == '-') { - neg = 1; - p++; - } - if (len > 1 && *p == '0') - p++; - if (number[len - 1] == 'h') { - ret = strtoul(p, &endp, 16); - bad = (*endp != 'h'); - } - else { - ret = strtoul(p, &endp, 10); - bad = (*endp != 0); - } - if (bad) - aerr("number parsing failed\n"); - if (neg) - ret = -ret; - return ret; -} - static int parse_indmode(char *name, int *regmask, int need_c_cvt) { enum opr_lenmod lmod; @@ -471,17 +445,17 @@ static int guess_lmod_from_c_type(enum opr_lenmod *lmod, const struct parsed_type *c_type) { static const char *dword_types[] = { - "int", "_DWORD", "UINT_PTR", - "DWORD", "HANDLE", "HWND", "HMODULE", + "int", "_DWORD", "UINT_PTR", "DWORD", "WPARAM", "LPARAM", "UINT", + "HIMC", }; static const char *word_types[] = { - "uint16_t", "int16_t", + "uint16_t", "int16_t", "_WORD", "unsigned __int16", "__int16", }; static const char *byte_types[] = { "uint8_t", "int8_t", "char", - "unsigned __int8", "__int8", "BYTE", + "unsigned __int8", "__int8", "BYTE", "_BYTE", "_UNKNOWN", }; const char *n; @@ -550,9 +524,10 @@ static int parse_operand(struct parsed_opr *opr, const struct parsed_proto *pp; enum opr_lenmod tmplmod; unsigned long number; + char buf[256]; int ret, len; int wordc_in; - char *tmp; + char *p; int i; if (w >= wordc) @@ -588,8 +563,10 @@ static int parse_operand(struct parsed_opr *opr, if (label != NULL) { opr->type = OPT_LABEL; - if (IS_START(label, "ds:")) + if (IS_START(label, "ds:")) { + opr->had_ds = 1; label += 3; + } strcpy(opr->name, label); return wordc; } @@ -617,7 +594,7 @@ static int parse_operand(struct parsed_opr *opr, return wordc; } if (IS(words[w], "(offset")) { - char *p = strchr(words[w + 1], ')'); + p = strchr(words[w + 1], ')'); if (p == NULL) aerr("parse of bracketed offset failed\n"); *p = 0; @@ -630,10 +607,11 @@ static int parse_operand(struct parsed_opr *opr, if (wordc_in != 1) aerr("parse_operand 1 word expected\n"); - tmp = words[w]; - if (IS_START(tmp, "ds:")) - tmp += 3; - strcpy(opr->name, tmp); + if (IS_START(words[w], "ds:")) { + opr->had_ds = 1; + memmove(words[w], words[w] + 3, strlen(words[w]) - 2); + } + strcpy(opr->name, words[w]); if (words[w][0] == '[') { opr->type = OPT_REGMEM; @@ -653,11 +631,13 @@ static int parse_operand(struct parsed_opr *opr, } else if (strchr(words[w], '[')) { // label[reg] form + p = strchr(words[w], '['); opr->type = OPT_REGMEM; - if (opr->lmod == OPLM_UNSPEC) - guess_lmod_from_name(opr); - parse_indmode(strchr(words[w], '['), regmask_indirect, 0); - return wordc; + parse_indmode(p, regmask_indirect, 0); + strncpy(buf, words[w], p - words[w]); + buf[p - words[w]] = 0; + pp = proto_parse(g_fhdr, buf, 1); + goto do_label; } else if (('0' <= words[w][0] && words[w][0] <= '9') || words[w][0] == '-') @@ -677,8 +657,9 @@ static int parse_operand(struct parsed_opr *opr, // most likely var in data segment opr->type = OPT_LABEL; + pp = proto_parse(g_fhdr, opr->name, 0); - pp = proto_parse(g_fhdr, opr->name); +do_label: if (pp != NULL) { if (pp->is_fptr || pp->is_func) { opr->lmod = OPLM_DWORD; @@ -994,6 +975,11 @@ static void parse_op(struct parsed_op *op, char words[16][256], int wordc) } break; + case OP_CALL: + // trashed regs must be explicitly detected later + op->regmask_dst = 0; + break; + default: break; } @@ -1461,17 +1447,32 @@ static void stack_frame_access(struct parsed_op *po, } } +static void check_func_pp(struct parsed_op *po, + const struct parsed_proto *pp, const char *pfx) +{ + char buf[256]; + + if (pp->argc_reg != 0) { + if (/*!g_allow_regfunc &&*/ !pp->is_fastcall) { + pp_print(buf, sizeof(buf), pp); + ferr(po, "%s: unexpected reg arg in icall: %s\n", pfx, buf); + } + if (pp->argc_stack > 0 && pp->argc_reg != 2) + ferr(po, "%s: %d reg arg(s) with %d stack arg(s)\n", + pfx, pp->argc_reg, pp->argc_stack); + } +} + static void check_label_read_ref(struct parsed_op *po, const char *name) { const struct parsed_proto *pp; - pp = proto_parse(g_fhdr, name); + pp = proto_parse(g_fhdr, name, 0); if (pp == NULL) ferr(po, "proto_parse failed for ref '%s'\n", name); - // currently we can take __cdecl and __stdcall - if (pp->is_func && pp->argc_reg != 0) - ferr(po, "reg-arg func reference?\n"); + if (pp->is_func) + check_func_pp(po, pp, "ref"); } static char *out_src_opr(char *buf, size_t buf_size, @@ -1577,8 +1578,11 @@ static char *out_src_opr(char *buf, size_t buf_size, ferr(po, "lea from const?\n"); printf_number(tmp1, sizeof(tmp1), popr->val); - snprintf(buf, buf_size, "%s%s", - simplify_cast_num(cast, popr->val), tmp1); + if (popr->val == 0 && strchr(cast, '*')) + snprintf(buf, buf_size, "NULL"); + else + snprintf(buf, buf_size, "%s%s", + simplify_cast_num(cast, popr->val), tmp1); break; default: @@ -1914,6 +1918,7 @@ static void op_set_clear_flag(struct parsed_op *po, static int scan_for_pop(int i, int opcnt, const char *reg, int magic, int depth, int *maxdepth, int do_flags) { + const struct parsed_proto *pp; struct parsed_op *po; int ret = 0; int j; @@ -1924,8 +1929,15 @@ static int scan_for_pop(int i, int opcnt, const char *reg, break; // already checked po->cc_scratch = magic; - if (po->flags & OPF_TAIL) + if (po->flags & OPF_TAIL) { + if (po->op == OP_CALL) { + pp = proto_parse(g_fhdr, po->operand[0].name, 0); + if (pp != NULL && pp->is_noreturn) + // no stack cleanup for noreturn + return ret; + } return -1; // deadend + } if ((po->flags & OPF_RMD) || (po->op == OP_PUSH && po->argnum != 0)) // arg push @@ -1934,15 +1946,13 @@ static int scan_for_pop(int i, int opcnt, const char *reg, if ((po->flags & OPF_JMP) && po->op != OP_CALL) { if (po->btj != NULL) { // jumptable - for (j = 0; j < po->btj->count - 1; j++) { + for (j = 0; j < po->btj->count; j++) { ret |= scan_for_pop(po->btj->d[j].bt_i, opcnt, reg, magic, depth, maxdepth, do_flags); if (ret < 0) return ret; // dead end } - // follow last jumptable entry - i = po->btj->d[j].bt_i - 1; - continue; + return ret; } if (po->bt_i < 0) { @@ -1966,24 +1976,26 @@ static int scan_for_pop(int i, int opcnt, const char *reg, && po->operand[0].type == OPT_REG && IS(po->operand[0].name, reg)) { - if (po->op == OP_PUSH) { + if (po->op == OP_PUSH && !(po->flags & OPF_FARG)) { depth++; if (depth > *maxdepth) *maxdepth = depth; if (do_flags) op_set_clear_flag(po, OPF_RSAVE, OPF_RMD); } - else if (depth == 0) { - if (do_flags) - op_set_clear_flag(po, OPF_RMD, OPF_RSAVE); - return 1; - } - else { - depth--; - if (depth < 0) // should not happen - ferr(po, "fail with depth\n"); - if (do_flags) - op_set_clear_flag(po, OPF_RSAVE, OPF_RMD); + else if (po->op == OP_POP) { + if (depth == 0) { + if (do_flags) + op_set_clear_flag(po, OPF_RMD, OPF_RSAVE); + return 1; + } + else { + depth--; + if (depth < 0) // should not happen + ferr(po, "fail with depth\n"); + if (do_flags) + op_set_clear_flag(po, OPF_RSAVE, OPF_RMD); + } } } } @@ -2028,14 +2040,26 @@ static int scan_for_pop_ret(int i, int opcnt, const char *reg, static int is_opr_modified(const struct parsed_opr *opr, const struct parsed_op *po) { + int mask; + if ((po->flags & OPF_RMD) || !(po->flags & OPF_DATA)) return 0; - if (opr->type == OPT_REG && po->operand[0].type == OPT_REG) { - if (po->regmask_dst & (1 << opr->reg)) - return 1; - else - return 0; + if (opr->type == OPT_REG) { + if (po->op == OP_CALL) { + mask = (1 << xAX) | (1 << xCX) | (1 << xDX); + if ((1 << opr->reg) & mask) + return 1; + else + return 0; + } + + if (po->operand[0].type == OPT_REG) { + if (po->regmask_dst & (1 << opr->reg)) + return 1; + else + return 0; + } } return IS(po->operand[0].name, opr->name); @@ -2043,8 +2067,9 @@ static int is_opr_modified(const struct parsed_opr *opr, // is any operand of parsed_op 'po_test' modified by parsed_op 'po'? static int is_any_opr_modified(const struct parsed_op *po_test, - const struct parsed_op *po) + const struct parsed_op *po, int c_mode) { + int mask; int i; if ((po->flags & OPF_RMD) || !(po->flags & OPF_DATA)) @@ -2058,9 +2083,12 @@ static int is_any_opr_modified(const struct parsed_op *po_test, // in reality, it can wreck any register, but in decompiled C // version it can only overwrite eax or edx:eax + mask = (1 << xAX) | (1 << xDX); + if (!c_mode) + mask |= 1 << xCX; + if (po->op == OP_CALL - && ((po_test->regmask_src | po_test->regmask_dst) - & ((1 << xAX)|(1 << xDX)))) + && ((po_test->regmask_src | po_test->regmask_dst) & mask)) return 1; for (i = 0; i < po_test->operand_cnt; i++) @@ -2071,13 +2099,14 @@ static int is_any_opr_modified(const struct parsed_op *po_test, } // scan for any po_test operand modification in range given -static int scan_for_mod(struct parsed_op *po_test, int i, int opcnt) +static int scan_for_mod(struct parsed_op *po_test, int i, int opcnt, + int c_mode) { if (po_test->operand_cnt == 1 && po_test->operand[0].type == OPT_CONST) return -1; for (; i < opcnt; i++) { - if (is_any_opr_modified(po_test, &ops[i])) + if (is_any_opr_modified(po_test, &ops[i], c_mode)) return i; } @@ -2164,8 +2193,6 @@ static int scan_for_cdq_edx(int i) if (ops[i].regmask_dst & (1 << xDX)) return -1; - if (g_labels[i][0] != 0) - return -1; } return -1; @@ -2193,8 +2220,6 @@ static int scan_for_reg_clear(int i, int reg) if (ops[i].regmask_dst & (1 << reg)) return -1; - if (g_labels[i][0] != 0) - return -1; } return -1; @@ -2203,6 +2228,7 @@ static int scan_for_reg_clear(int i, int reg) // scan for positive, constant esp adjust static int scan_for_esp_adjust(int i, int opcnt, int *adj) { + const struct parsed_proto *pp; struct parsed_op *po; int first_pop = -1; *adj = 0; @@ -2236,7 +2262,9 @@ static int scan_for_esp_adjust(int i, int opcnt, int *adj) break; if (po->operand[0].type != OPT_LABEL) break; - // TODO: should only allow combining __cdecl calls.. + pp = po->datap; + if (pp != NULL && pp->is_stdcall) + break; } } @@ -2284,9 +2312,11 @@ static void scan_fwd_set_flags(int i, int opcnt, int magic, int flags) } static const struct parsed_proto *try_recover_pp( - struct parsed_op *po, const struct parsed_opr *opr) + struct parsed_op *po, const struct parsed_opr *opr, int *search_instead) { const struct parsed_proto *pp = NULL; + char buf[256]; + char *p; // maybe an arg of g_func? if (opr->type == OPT_REGMEM && is_stack_access(po, opr)) @@ -2300,8 +2330,12 @@ static const struct parsed_proto *try_recover_pp( &offset, &stack_ra, NULL); if (ofs_reg[0] != 0) ferr(po, "offset reg on arg access?\n"); - if (offset <= stack_ra) - ferr(po, "stack var call unhandled yet\n"); + if (offset <= stack_ra) { + // search who set the stack var instead + if (search_instead != NULL) + *search_instead = 1; + return NULL; + } arg_i = (offset - stack_ra - 4) / 4; for (arg = arg_s = 0; arg < g_func_pp->argc; arg++) { @@ -2317,22 +2351,27 @@ static const struct parsed_proto *try_recover_pp( pp = g_func_pp->arg[arg].fptr; if (pp == NULL) ferr(po, "icall sa: arg%d is not a fptr?\n", arg + 1); - if (pp->argc_reg != 0) - ferr(po, "icall sa: reg arg in arg-call unhandled yet\n"); + check_func_pp(po, pp, "icall arg"); + } + else if (opr->type == OPT_REGMEM && strchr(opr->name + 1, '[')) { + // label[index] + p = strchr(opr->name + 1, '['); + memcpy(buf, opr->name, p - opr->name); + buf[p - opr->name] = 0; + pp = proto_parse(g_fhdr, buf, 0); } else if (opr->type == OPT_OFFSET || opr->type == OPT_LABEL) { - pp = proto_parse(g_fhdr, opr->name); + pp = proto_parse(g_fhdr, opr->name, 0); if (pp == NULL) ferr(po, "proto_parse failed for icall from '%s'\n", opr->name); - if (pp->argc_reg != 0) - ferr(po, "arg-call unhandled yet for '%s'\n", opr->name); + check_func_pp(po, pp, "reg-fptr ref"); } return pp; } -static void scan_for_call_type(int i, struct parsed_opr *opr, - int magic, const struct parsed_proto **pp_found) +static void scan_for_call_type(int i, const struct parsed_opr *opr, + int magic, const struct parsed_proto **pp_found, int *multi) { const struct parsed_proto *pp = NULL; struct parsed_op *po; @@ -2346,7 +2385,7 @@ static void scan_for_call_type(int i, struct parsed_opr *opr, if (g_labels[i][0] != 0) { lr = &g_label_refs[i]; for (; lr != NULL; lr = lr->next) - scan_for_call_type(lr->i, opr, magic, pp_found); + scan_for_call_type(lr->i, opr, magic, pp_found, multi); if (i > 0 && LAST_OP(i - 1)) return; } @@ -2386,46 +2425,54 @@ static void scan_for_call_type(int i, struct parsed_opr *opr, if (pp == NULL) ferr(po, "icall: arg%d (%s) is not a fptr?\n", i + 1, g_func_pp->arg[i].reg); - if (pp->argc_reg != 0) - ferr(po, "icall: reg arg in arg-call unhandled yet\n"); + check_func_pp(po, pp, "icall reg-arg"); } else - pp = try_recover_pp(po, opr); + pp = try_recover_pp(po, opr, NULL); - if (*pp_found != NULL && pp != NULL) { + if (*pp_found != NULL && pp != NULL && *pp_found != pp) { if (!IS((*pp_found)->ret_type.name, pp->ret_type.name) || (*pp_found)->is_stdcall != pp->is_stdcall + || (*pp_found)->is_fptr != pp->is_fptr || (*pp_found)->argc != pp->argc || (*pp_found)->argc_reg != pp->argc_reg || (*pp_found)->argc_stack != pp->argc_stack) { ferr(po, "icall: parsed_proto mismatch\n"); } + *multi = 1; } if (pp != NULL) *pp_found = pp; } -static const struct parsed_proto *resolve_call(int i, int opcnt) +static const struct parsed_proto *resolve_icall(int i, int opcnt, + int *multi_src) { const struct parsed_proto *pp = NULL; + int search_advice = 0; + + *multi_src = 0; switch (ops[i].operand[0].type) { case OPT_REGMEM: case OPT_LABEL: case OPT_OFFSET: - pp = try_recover_pp(&ops[i], &ops[i].operand[0]); - break; + pp = try_recover_pp(&ops[i], &ops[i].operand[0], &search_advice); + if (!search_advice) + break; + // fallthrough default: - scan_for_call_type(i, &ops[i].operand[0], i + opcnt * 9, &pp); + scan_for_call_type(i, &ops[i].operand[0], i + opcnt * 9, &pp, + multi_src); break; } return pp; } -static int collect_call_args(struct parsed_op *po, int i, - struct parsed_proto *pp, int *save_arg_vars, int arg, +static int collect_call_args_r(struct parsed_op *po, int i, + struct parsed_proto *pp, int *regmask, int *save_arg_vars, int arg, int magic, int need_op_saving, int may_reuse) { struct parsed_proto *pp_tmp; @@ -2444,7 +2491,7 @@ static int collect_call_args(struct parsed_op *po, int i, break; magic = (magic & 0xffffff) | (arg << 24); - for (j = i; j >= 0 && arg < pp->argc; ) + for (j = i; j >= 0 && (arg < pp->argc || pp->is_unresolved); ) { if (((ops[j].cc_scratch ^ magic) & 0xffffff) == 0) { if (ops[j].cc_scratch != magic) { @@ -2464,7 +2511,7 @@ static int collect_call_args(struct parsed_op *po, int i, for (; lr->next; lr = lr->next) { if ((ops[lr->i].flags & (OPF_JMP|OPF_CC)) != OPF_JMP) may_reuse = 1; - ret = collect_call_args(po, lr->i, pp, save_arg_vars, + ret = collect_call_args_r(po, lr->i, pp, regmask, save_arg_vars, arg, magic, need_op_saving, may_reuse); if (ret < 0) return ret; @@ -2478,7 +2525,7 @@ static int collect_call_args(struct parsed_op *po, int i, continue; } need_op_saving = 1; - ret = collect_call_args(po, lr->i, pp, save_arg_vars, + ret = collect_call_args_r(po, lr->i, pp, regmask, save_arg_vars, arg, magic, need_op_saving, may_reuse); if (ret < 0) return ret; @@ -2487,6 +2534,9 @@ static int collect_call_args(struct parsed_op *po, int i, if (ops[j].op == OP_CALL) { + if (pp->is_unresolved) + break; + pp_tmp = ops[j].datap; if (pp_tmp == NULL) ferr(po, "arg collect hit unparsed call '%s'\n", @@ -2496,22 +2546,34 @@ static int collect_call_args(struct parsed_op *po, int i, arg, pp->argc, opr_name(&ops[j], 0), pp_tmp->argc_stack); } else if (ops[j].op == OP_ADD && ops[j].operand[0].reg == xSP) { + if (pp->is_unresolved) + break; + ferr(po, "arg collect %d/%d hit esp adjust\n", arg, pp->argc); } else if (ops[j].op == OP_POP) { + if (pp->is_unresolved) + break; + ferr(po, "arg collect %d/%d hit pop\n", arg, pp->argc); } else if ((ops[j].flags & (OPF_JMP|OPF_CC)) == (OPF_JMP|OPF_CC)) { + if (pp->is_unresolved) + break; + may_reuse = 1; } else if (ops[j].op == OP_PUSH && !(ops[j].flags & OPF_FARG)) { + if (pp->is_unresolved && (ops[j].flags & OPF_RMD)) + break; + pp->arg[arg].datap = &ops[j]; need_to_save_current = 0; if (!need_op_saving) { - ret = scan_for_mod(&ops[j], j + 1, i); + ret = scan_for_mod(&ops[j], j + 1, i, 1); need_to_save_current = (ret >= 0); } if (need_op_saving || need_to_save_current) { @@ -2534,22 +2596,73 @@ static int collect_call_args(struct parsed_op *po, int i, if (!may_reuse) ops[j].flags |= OPF_FARG; - // next arg - for (arg++; arg < pp->argc; arg++) - if (pp->arg[arg].reg == NULL) - break; + ops[j].flags &= ~OPF_RSAVE; + + arg++; + if (!pp->is_unresolved) { + // next arg + for (; arg < pp->argc; arg++) + if (pp->arg[arg].reg == NULL) + break; + } magic = (magic & 0xffffff) | (arg << 24); + + // tracking reg usage + if (ops[j].operand[0].type == OPT_REG) + *regmask |= 1 << ops[j].operand[0].reg; } } if (arg < pp->argc) { ferr(po, "arg collect failed for '%s': %d/%d\n", pp->name, arg, pp->argc); - ret = -1; + return -1; + } + + return arg; +} + +static int collect_call_args(struct parsed_op *po, int i, + struct parsed_proto *pp, int *regmask, int *save_arg_vars, + int magic) +{ + int ret; + int a; + + ret = collect_call_args_r(po, i, pp, regmask, save_arg_vars, + 0, magic, 0, 0); + 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"); } + return ret; } +static void pp_insert_reg_arg(struct parsed_proto *pp, const char *reg) +{ + int i; + + for (i = 0; i < pp->argc; i++) + if (pp->arg[i].reg == NULL) + break; + + if (pp->argc_stack) + memmove(&pp->arg[i + 1], &pp->arg[i], + sizeof(pp->arg[0]) * pp->argc_stack); + memset(&pp->arg[i], 0, sizeof(pp->arg[i])); + pp->arg[i].reg = strdup(reg); + pp->arg[i].type.name = strdup("int"); + pp->argc++; + pp->argc_reg++; +} + static void add_label_ref(struct label_ref *lr, int op_i) { struct label_ref *lr_new; @@ -2580,6 +2693,17 @@ static void output_std_flags(FILE *fout, struct parsed_op *po, } } +static void output_pp_attrs(FILE *fout, const struct parsed_proto *pp, + int is_noreturn) +{ + if (pp->is_fastcall) + fprintf(fout, "__fastcall "); + else if (pp->is_stdcall && pp->argc_reg == 0) + fprintf(fout, "__stdcall "); + if (pp->is_noreturn || is_noreturn) + fprintf(fout, "noreturn "); +} + static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt) { struct parsed_op *po, *delayed_flag_op = NULL, *tmp_op; @@ -2593,13 +2717,12 @@ static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt) int save_arg_vars = 0; int cmp_result_vars = 0; int need_mul_var = 0; - int have_func_ret = 0; - int have_normal_ret = 0; int had_decl = 0; int label_pending = 0; int regmask_save = 0; int regmask_arg = 0; int regmask_now = 0; + int regmask_init = 0; int regmask = 0; int pfomask = 0; int found = 0; @@ -2614,15 +2737,12 @@ static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt) g_bp_frame = g_sp_frame = g_stack_fsz = 0; g_stack_frame_used = 0; - g_func_pp = proto_parse(fhdr, funcn); + g_func_pp = proto_parse(fhdr, funcn, 0); if (g_func_pp == NULL) ferr(ops, "proto_parse failed for '%s'\n", funcn); fprintf(fout, "%s ", g_func_pp->ret_type.name); - if (g_func_pp->is_stdcall && g_func_pp->argc_reg == 0) - fprintf(fout, "__stdcall "); - if (g_ida_func_attr & IDAFA_NORETURN) - fprintf(fout, "noreturn "); + output_pp_attrs(fout, g_func_pp, g_ida_func_attr & IDAFA_NORETURN); fprintf(fout, "%s(", funcn); for (i = 0; i < g_func_pp->argc; i++) { @@ -2632,8 +2752,7 @@ static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt) // func pointer.. pp = g_func_pp->arg[i].fptr; fprintf(fout, "%s (", pp->ret_type.name); - if (pp->is_stdcall && pp->argc_reg == 0) - fprintf(fout, "__stdcall "); + output_pp_attrs(fout, pp, 0); fprintf(fout, "*a%d)(", i + 1); for (j = 0; j < pp->argc; j++) { if (j > 0) @@ -2647,6 +2766,13 @@ static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt) else { fprintf(fout, "%s a%d", g_func_pp->arg[i].type.name, i + 1); } + if (g_func_pp->arg[i].reg != NULL) { + reg = char_array_i(regs_r32, + ARRAY_SIZE(regs_r32), g_func_pp->arg[i].reg); + if (reg < 0) + ferr(ops, "arg '%s' is not a reg?\n", g_func_pp->arg[i].reg); + regmask_arg |= 1 << reg; + } } if (g_func_pp->is_vararg) { if (i > 0) @@ -2779,17 +2905,35 @@ static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt) continue; if (po->op == OP_CALL) { + pp = NULL; + if (po->operand[0].type == OPT_LABEL) { tmpname = opr_name(po, 0); if (IS_START(tmpname, "loc_")) ferr(po, "call to loc_*\n"); - pp_c = proto_parse(fhdr, tmpname); + pp_c = proto_parse(fhdr, tmpname, 0); if (pp_c == NULL) ferr(po, "proto_parse failed for call '%s'\n", tmpname); - if (pp_c->is_fptr && pp_c->argc_reg != 0) - ferr(po, "fptr call with reg arg\n"); + pp = proto_clone(pp_c); my_assert_not(pp, NULL); + } + else if (po->datap != NULL) { + pp = calloc(1, sizeof(*pp)); + my_assert_not(pp, NULL); + + ret = parse_protostr(po->datap, pp); + if (ret < 0) + ferr(po, "bad protostr supplied: %s\n", (char *)po->datap); + free(po->datap); + po->datap = NULL; + } + + if (pp != NULL) { + if (pp->is_fptr) + check_func_pp(po, pp, "fptr var call"); + if (pp->is_noreturn) + po->flags |= OPF_TAIL; po->datap = pp; } continue; @@ -2801,7 +2945,7 @@ static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt) if (po->operand[0].type == OPT_REGMEM) { char *p = strchr(po->operand[0].name, '['); if (p == NULL) - ferr(po, "unhandled indirect branch\n"); + goto tailcall; ret = p - po->operand[0].name; strncpy(buf1, po->operand[0].name, ret); buf1[ret] = 0; @@ -2875,18 +3019,28 @@ tailcall: if (pp == NULL) { // indirect call - pp_c = resolve_call(i, opcnt); - if (pp_c != NULL) + pp_c = resolve_icall(i, opcnt, &l); + if (pp_c != NULL) { pp = proto_clone(pp_c); + my_assert_not(pp, NULL); + if (l) + // not resolved just to single func + pp->is_fptr = 1; + } if (pp == NULL) { pp = calloc(1, sizeof(*pp)); my_assert_not(pp, NULL); + pp->is_fptr = 1; ret = scan_for_esp_adjust(i + 1, opcnt, &j); - if (ret < 0) - ferr(po, "non-__cdecl indirect call unhandled yet\n"); + if (ret < 0) { + if (!g_allow_regfunc) + ferr(po, "non-__cdecl indirect call unhandled yet\n"); + pp->is_unresolved = 1; + j = 0; + } j /= 4; if (j > ARRAY_SIZE(pp->arg)) - ferr(po, "esp adjust too large?\n"); + ferr(po, "esp adjust too large: %d\n", j); pp->ret_type.name = strdup("int"); pp->argc = pp->argc_stack = j; for (arg = 0; arg < pp->argc; arg++) @@ -2927,22 +3081,14 @@ tailcall: ferr(po, "missing esp_adjust for vararg func '%s'\n", pp->name); - // can't call functions with non-__cdecl callbacks yet - for (arg = 0; arg < pp->argc; arg++) { - if (pp->arg[arg].fptr != NULL) { - pp_tmp = pp->arg[arg].fptr; - if (pp_tmp->argc_reg != 0) - ferr(po, "'%s' has a callback with reg-args\n", tmpname); - } + if (!pp->is_unresolved) { + // since we know the args, collect them + collect_call_args(po, i, pp, ®mask, &save_arg_vars, + i + opcnt * 2); } - collect_call_args(po, i, pp, &save_arg_vars, - 0, i + opcnt * 2, 0, 0); - if (strstr(pp->ret_type.name, "int64")) need_mul_var = 1; - if (!(po->flags & OPF_TAIL) && !IS(pp->ret_type.name, "void")) - have_func_ret = 1; } } @@ -2950,6 +3096,7 @@ tailcall: // - find POPs for PUSHes, rm both // - scan for all used registers // - find flag set ops for their users + // - do unreselved calls // - declare indirect functions for (i = 0; i < opcnt; i++) { po = &ops[i]; @@ -3033,7 +3180,7 @@ tailcall: // if we can't because of operand modification, or if we // have math op, or branch, make it calculate flags explicitly if (tmp_op->op == OP_TEST || tmp_op->op == OP_CMP) { - if (branched || scan_for_mod(tmp_op, setters[j] + 1, i) >= 0) + if (branched || scan_for_mod(tmp_op, setters[j] + 1, i, 0) >= 0) pfomask = 1 << pfo; } else if (tmp_op->op == OP_CMPS) { @@ -3063,24 +3210,119 @@ tailcall: { need_mul_var = 1; } - else if (po->op == OP_CALL && po->operand[0].type != OPT_LABEL) { + else if (po->op == OP_CALL) { pp = po->datap; - my_assert_not(pp, NULL); - fprintf(fout, " %s (", pp->ret_type.name); - if (pp->is_stdcall && pp->argc_reg == 0) - fprintf(fout, "__stdcall "); - fprintf(fout, "*icall%d)(", i); - for (j = 0; j < pp->argc; j++) { - if (j > 0) - fprintf(fout, ", "); - fprintf(fout, "%s a%d", pp->arg[j].type.name, j + 1); + if (pp == NULL) + ferr(po, "NULL pp\n"); + + if (pp->is_unresolved) { + int regmask_stack = 0; + collect_call_args(po, i, pp, ®mask, &save_arg_vars, + i + opcnt * 2); + + // this is pretty rough guess: + // see ecx and edx were pushed (and not their saved versions) + for (arg = 0; arg < pp->argc; arg++) { + if (pp->arg[arg].reg != NULL) + continue; + + tmp_op = pp->arg[arg].datap; + if (tmp_op == NULL) + ferr(po, "parsed_op missing for arg%d\n", arg); + if (tmp_op->argnum == 0 && tmp_op->operand[0].type == OPT_REG) + regmask_stack |= 1 << tmp_op->operand[0].reg; + } + + if (!((regmask_stack & (1 << xCX)) + && (regmask_stack & (1 << xDX)))) + { + if (pp->argc_stack != 0 + || ((regmask | regmask_arg) & ((1 << xCX)|(1 << xDX)))) + { + pp_insert_reg_arg(pp, "ecx"); + pp->is_fastcall = 1; + regmask_init |= 1 << xCX; + regmask |= 1 << xCX; + } + if (pp->argc_stack != 0 + || ((regmask | regmask_arg) & (1 << xDX))) + { + pp_insert_reg_arg(pp, "edx"); + regmask_init |= 1 << xDX; + regmask |= 1 << xDX; + } + } + + // note: __cdecl doesn't fall into is_unresolved category + if (pp->argc_stack > 0) + pp->is_stdcall = 1; + } + + for (arg = 0; arg < pp->argc; arg++) { + if (pp->arg[arg].reg != NULL) { + reg = char_array_i(regs_r32, + ARRAY_SIZE(regs_r32), pp->arg[arg].reg); + if (reg < 0) + ferr(ops, "arg '%s' is not a reg?\n", pp->arg[arg].reg); + if (!(regmask & (1 << reg))) { + regmask_init |= 1 << reg; + regmask |= 1 << reg; + } + } + } + + if (pp->is_fptr) { + if (pp->name[0] != 0) { + memmove(pp->name + 2, pp->name, strlen(pp->name) + 1); + memcpy(pp->name, "i_", 2); + + // might be declared already + found = 0; + for (j = 0; j < i; j++) { + if (ops[j].op == OP_CALL && (pp_tmp = ops[j].datap)) { + if (pp_tmp->is_fptr && IS(pp->name, pp_tmp->name)) { + found = 1; + break; + } + } + } + if (found) + continue; + } + else + snprintf(pp->name, sizeof(pp->name), "icall%d", i); + + fprintf(fout, " %s (", pp->ret_type.name); + output_pp_attrs(fout, pp, 0); + fprintf(fout, "*%s)(", pp->name); + for (j = 0; j < pp->argc; j++) { + if (j > 0) + fprintf(fout, ", "); + fprintf(fout, "%s a%d", pp->arg[j].type.name, j + 1); + } + fprintf(fout, ");\n"); } - fprintf(fout, ");\n"); } - else if (po->op == OP_RET) - have_normal_ret = 1; + else if (po->op == OP_RET && !IS(g_func_pp->ret_type.name, "void")) + regmask |= 1 << xAX; + } + + // pass4: + // - confirm regmask_save, it might have been reduced + if (regmask_save != 0) + { + regmask_save = 0; + for (i = 0; i < opcnt; i++) { + po = &ops[i]; + if (po->flags & OPF_RMD) + continue; + + if (po->op == OP_PUSH && (po->flags & OPF_RSAVE)) + regmask_save |= 1 << po->operand[0].reg; + } } + // output LUTs/jumptables for (i = 0; i < g_func_pd_cnt; i++) { pd = &g_func_pd[i]; @@ -3120,31 +3362,26 @@ tailcall: if (g_func_pp->arg[i].reg != NULL) { reg = char_array_i(regs_r32, ARRAY_SIZE(regs_r32), g_func_pp->arg[i].reg); - if (reg < 0) - ferr(ops, "arg '%s' is not a reg?\n", g_func_pp->arg[i].reg); - - regmask_arg |= 1 << reg; - fprintf(fout, " u32 %s = (u32)a%d;\n", - g_func_pp->arg[i].reg, i + 1); + if (regmask & (1 << reg)) { + fprintf(fout, " u32 %s = (u32)a%d;\n", + g_func_pp->arg[i].reg, i + 1); + } + else + fprintf(fout, " // %s = a%d; // unused\n", + g_func_pp->arg[i].reg, i + 1); had_decl = 1; } } - // declare other regs - special case for eax - if (!((regmask | regmask_arg) & 1) - && (have_func_ret || have_normal_ret) - && !IS(g_func_pp->ret_type.name, "void")) - { - fprintf(fout, " u32 eax = 0;\n"); - had_decl = 1; - } - - regmask &= ~regmask_arg; - regmask &= ~(1 << xSP); - if (regmask) { + regmask_now = regmask & ~regmask_arg; + regmask_now &= ~(1 << xSP); + if (regmask_now) { for (reg = 0; reg < 8; reg++) { - if (regmask & (1 << reg)) { - fprintf(fout, " u32 %s;\n", regs_r32[reg]); + if (regmask_now & (1 << reg)) { + fprintf(fout, " u32 %s", regs_r32[reg]); + if (regmask_init & (1 << reg)) + fprintf(fout, " = 0"); + fprintf(fout, ";\n"); had_decl = 1; } } @@ -3632,11 +3869,10 @@ tailcall: case OP_CALL: assert_operand_cnt(1); pp = po->datap; - if (pp == NULL) - ferr(po, "NULL pp\n"); + my_assert_not(pp, NULL); - if (po->operand[0].type != OPT_LABEL) - fprintf(fout, " icall%d = %s;\n", i, + if (pp->is_fptr) + fprintf(fout, " %s = %s;\n", pp->name, out_src_opr(buf1, sizeof(buf1), po, &po->operand[0], "(void *)", 0)); @@ -3654,22 +3890,17 @@ tailcall: fprintf(fout, "(%s)", g_func_pp->ret_type.name); } } - else { + else if (regmask & (1 << xAX)) { fprintf(fout, "eax = "); if (pp->ret_type.is_ptr) fprintf(fout, "(u32)"); } } - if (po->operand[0].type != OPT_LABEL) { - fprintf(fout, "icall%d(", i); - } - else { - if (pp->name[0] == 0) - ferr(po, "missing pp->name\n"); - fprintf(fout, "%s%s(", pp->name, - pp->has_structarg ? "_sa" : ""); - } + if (pp->name[0] == 0) + ferr(po, "missing pp->name\n"); + fprintf(fout, "%s%s(", pp->name, + pp->has_structarg ? "_sa" : ""); for (arg = 0; arg < pp->argc; arg++) { if (arg > 0) @@ -3705,10 +3936,15 @@ tailcall: fprintf(fout, " eax = mul_tmp;"); } + if (pp->is_unresolved) { + snprintf(buf3, sizeof(buf3), " unresoved %dreg", + pp->argc_reg); + strcat(g_comment, buf3); + } + if (po->flags & OPF_TAIL) { - strcpy(g_comment, "tailcall"); ret = 0; - if (i == opcnt - 1) + if (i == opcnt - 1 || pp->is_noreturn) ret = 0; else if (IS(pp->ret_type.name, "void")) ret = 1; @@ -3717,12 +3953,19 @@ tailcall: // else already handled as 'return f()' if (ret) { - if (!IS(g_func_pp->ret_type.name, "void")) + if (!IS(g_func_pp->ret_type.name, "void")) { ferr(po, "int func -> void func tailcall?\n"); - fprintf(fout, "\n return;"); - strcpy(g_comment, "^ tailcall"); + } + else { + fprintf(fout, "\n return;"); + strcat(g_comment, " ^ tailcall"); + } } + else + strcat(g_comment, " tailcall"); } + if (pp->is_noreturn) + strcat(g_comment, " noreturn"); delayed_flag_op = NULL; last_arith_dst = NULL; break; @@ -3784,7 +4027,10 @@ tailcall: } if (g_comment[0] != 0) { - fprintf(fout, " // %s", g_comment); + char *p = g_comment; + while (my_isblank(*p)) + p++; + fprintf(fout, " // %s", p); g_comment[0] = 0; no_output = 0; } @@ -3803,7 +4049,7 @@ tailcall: // 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)) + if (is_any_opr_modified(delayed_flag_op, po, 0)) delayed_flag_op = NULL; } @@ -3990,6 +4236,7 @@ int main(int argc, char *argv[]) char line[256]; char words[20][256]; enum opr_lenmod lmod; + char *sctproto = NULL; int in_func = 0; int pending_endp = 0; int skip_func = 0; @@ -3997,20 +4244,24 @@ int main(int argc, char *argv[]) int eq_alloc; int verbose = 0; int arg_out; - int arg = 1; + int arg; int pi = 0; int i, j; int ret, len; char *p; int wordc; - if (argv[1] && IS(argv[1], "-v")) { - verbose = 1; - arg++; + for (arg = 1; arg < argc; arg++) { + if (IS(argv[arg], "-v")) + verbose = 1; + else if (IS(argv[arg], "-rf")) + g_allow_regfunc = 1; + else + break; } if (argc < arg + 3) { - printf("usage:\n%s [-v] <.c> <.asm> [rlist]*\n", + printf("usage:\n%s [-v] [-rf] <.c> <.asm> [rlist]*\n", argv[0]); return 1; } @@ -4046,8 +4297,11 @@ int main(int argc, char *argv[]) if (*p == 0 || *p == ';') continue; if (*p == '#') { - if (IS_START(p, "#if 0")) + if (IS_START(p, "#if 0") + || (g_allow_regfunc && IS_START(p, "#if NO_REGFUNC"))) + { skip_func = 1; + } else if (IS_START(p, "#endif")) skip_func = 0; continue; @@ -4207,11 +4461,16 @@ parse_words: aerr("too many words\n"); // alow asm patches in comments - if (*p == ';' && IS_START(p, "; sctpatch:")) { - p = sskip(p + 11); - if (*p == 0 || *p == ';') - continue; - goto parse_words; // lame + if (*p == ';') { + if (IS_START(p, "; sctpatch:")) { + p = sskip(p + 11); + if (*p == 0 || *p == ';') + continue; + goto parse_words; // lame + } + if (IS_START(p, "; sctproto:")) { + sctproto = strdup(p + 11); + } } if (wordc == 0) { @@ -4254,6 +4513,11 @@ do_pending_endp: i = 2; } else { + if (pd == NULL) { + if (verbose) + anote("skipping alignment byte?\n"); + continue; + } lmod = lmod_from_directive(words[0]); if (lmod != pd->lmod) aerr("lmod change? %d->%d\n", pd->lmod, lmod); @@ -4319,8 +4583,7 @@ do_pending_endp: aerr("proc '%s' while in_func '%s'?\n", words[0], g_func); p = words[0]; - if ((g_ida_func_attr & IDAFA_THUNK) - || bsearch(&p, rlist, rlist_len, sizeof(rlist[0]), cmpstringp)) + if (bsearch(&p, rlist, rlist_len, sizeof(rlist[0]), cmpstringp)) skip_func = 1; strcpy(g_func, words[0]); set_label(0, words[0]); @@ -4336,6 +4599,13 @@ do_pending_endp: aerr("endp '%s' while in_func '%s'?\n", words[0], g_func); + if ((g_ida_func_attr & IDAFA_THUNK) && pi == 1 + && ops[0].op == OP_JMP && ops[0].operand[0].had_ds) + { + // import jump + skip_func = 1; + } + if (!skip_func && func_chunks_used) { // start processing chunks struct chunk_item *ci, key = { g_func, 0 }; @@ -4386,7 +4656,8 @@ do_pending_endp: continue; } - if (wordc > 1 && IS(words[1], "=")) { + if (wordc > 1 && IS(words[1], "=")) + { if (wordc != 5) aerr("unhandled equ, wc=%d\n", wordc); if (g_eqcnt >= eq_alloc) { @@ -4420,6 +4691,12 @@ do_pending_endp: aerr("too many ops\n"); parse_op(&ops[pi], words, wordc); + + if (sctproto != NULL) { + if (ops[pi].op == OP_CALL) + ops[pi].datap = sctproto; + sctproto = NULL; + } pi++; }