X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=tools%2Ftranslate.c;h=1fa3d95fdb686bb2129bfdba0479becf0d9e2727;hb=9ea60b8d585086fa64f7a8cd298ec1cd698ad56d;hp=5acdd11f817094ea1096e147a240805553dfdd38;hpb=c0de901581e6edfc9b0a18fe6bb89ee56da12804;p=ia32rtools.git diff --git a/tools/translate.c b/tools/translate.c index 5acdd11..1fa3d95 100644 --- a/tools/translate.c +++ b/tools/translate.c @@ -13,6 +13,7 @@ #include "my_assert.h" #include "my_str.h" +#include "common.h" #define ARRAY_SIZE(x) (sizeof(x) / sizeof(x[0])) #define IS(w, y) !strcmp(w, y) @@ -57,6 +58,7 @@ enum op_flags { OPF_LOCK = (1 << 17), /* op has lock prefix */ OPF_VAPUSH = (1 << 18), /* vararg ptr push (as call arg) */ OPF_DONE = (1 << 19), /* already fully handled by analysis */ + OPF_PPUSH = (1 << 20), /* part of complex push-pop graph */ }; enum op_op { @@ -113,7 +115,7 @@ enum op_op { // x87 // mmx OP_EMMS, - // mmx + // undefined OP_UD2, }; @@ -176,9 +178,10 @@ struct parsed_op { }; // datap: -// OP_CALL - parser proto hint (str) +// OP_CALL - parser proto hint (str) // (OPF_CC) - points to one of (OPF_FLAGS) that affects cc op -// OP_POP - points to OP_PUSH in push/pop pair +// OP_PUSH - points to OP_POP in complex push/pop graph +// OP_POP - points to OP_PUSH in simple push/pop pair struct parsed_equ { char name[64]; @@ -234,6 +237,7 @@ 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_skip_func; static int g_allow_regfunc; static int g_quiet_pp; static int g_header_mode; @@ -984,7 +988,8 @@ static void parse_op(struct parsed_op *op, char words[16][256], int wordc) } if (i == ARRAY_SIZE(op_table)) { - anote("unhandled op: '%s'\n", words[0]); + if (!g_skip_func) + aerr("unhandled op: '%s'\n", words[0]); i--; // OP_UD2 } w++; @@ -2137,6 +2142,10 @@ static void op_set_clear_flag(struct parsed_op *po, || ((ops[_i].flags & (OPF_JMP|OPF_CJMP|OPF_RMD)) == OPF_JMP \ && ops[_i].op != OP_CALL)) +#define check_i(po, i) \ + if ((i) < 0) \ + ferr(po, "bad " #i ": %d\n", i) + static int scan_for_pop(int i, int opcnt, const char *reg, int magic, int depth, int *maxdepth, int do_flags) { @@ -2169,6 +2178,7 @@ static int scan_for_pop(int i, int opcnt, const char *reg, if (po->btj != NULL) { // jumptable 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, reg, magic, depth, maxdepth, do_flags); if (ret < 0) @@ -2177,11 +2187,7 @@ static int scan_for_pop(int i, int opcnt, const char *reg, return ret; } - if (po->bt_i < 0) { - ferr(po, "dead branch\n"); - return -1; - } - + check_i(po, po->bt_i); if (po->flags & OPF_CJMP) { ret |= scan_for_pop(po->bt_i, opcnt, reg, magic, depth, maxdepth, do_flags); @@ -2259,22 +2265,44 @@ static int scan_for_pop_ret(int i, int opcnt, const char *reg, return found ? 0 : -1; } -static void scan_for_pop_const(int i, int opcnt) +static void scan_for_pop_const(int i, int opcnt, int *regmask_pp) { + struct parsed_op *po; + int is_multipath = 0; int j; for (j = i + 1; j < opcnt; j++) { - if ((ops[j].flags & (OPF_JMP|OPF_TAIL|OPF_RSAVE)) - || ops[j].op == OP_PUSH || g_labels[i] != NULL) + po = &ops[j]; + + if (po->op == OP_JMP && po->btj == NULL) { + ferr_assert(po, po->bt_i >= 0); + j = po->bt_i - 1; + continue; + } + + if ((po->flags & (OPF_JMP|OPF_TAIL|OPF_RSAVE)) + || po->op == OP_PUSH) { break; } - if (ops[j].op == OP_POP && !(ops[j].flags & (OPF_RMD|OPF_DONE))) + if (g_labels[j] != NULL) + is_multipath = 1; + + if (po->op == OP_POP && !(po->flags & OPF_RMD)) { - ops[i].flags |= OPF_RMD | OPF_DONE; - ops[j].flags |= OPF_DONE; - ops[j].datap = &ops[i]; + is_multipath |= !!(po->flags & OPF_PPUSH); + if (is_multipath) { + ops[i].flags |= OPF_PPUSH | OPF_DONE; + ops[i].datap = po; + po->flags |= OPF_PPUSH | OPF_DONE; + *regmask_pp |= 1 << po->operand[0].reg; + } + else { + ops[i].flags |= OPF_RMD | OPF_DONE; + po->flags |= OPF_DONE; + po->datap = &ops[i]; + } break; } } @@ -2297,16 +2325,14 @@ static void scan_propagate_df(int i, int opcnt) if (po->flags & OPF_JMP) { if (po->btj != NULL) { // jumptable - for (j = 0; j < po->btj->count; j++) + for (j = 0; j < po->btj->count; j++) { + check_i(po, po->btj->d[j].bt_i); scan_propagate_df(po->btj->d[j].bt_i, opcnt); + } return; } - if (po->bt_i < 0) { - ferr(po, "dead branch\n"); - return; - } - + check_i(po, po->bt_i); if (po->flags & OPF_CJMP) scan_propagate_df(po->bt_i, opcnt); else @@ -2326,13 +2352,57 @@ static void scan_propagate_df(int i, int opcnt) ferr(po, "missing DF clear?\n"); } +// is operand 'opr' referenced by parsed_op 'po'? +static int is_opr_referenced(const struct parsed_opr *opr, + const struct parsed_op *po) +{ + int i, mask; + + if (opr->type == OPT_REG) { + mask = po->regmask_dst | po->regmask_src; + if (po->op == OP_CALL) + mask |= (1 << xAX) | (1 << xCX) | (1 << xDX); + if ((1 << opr->reg) & mask) + return 1; + else + return 0; + } + + for (i = 0; i < po->operand_cnt; i++) + if (IS(po->operand[0].name, opr->name)) + return 1; + + return 0; +} + +// is operand 'opr' read by parsed_op 'po'? +static int is_opr_read(const struct parsed_opr *opr, + const struct parsed_op *po) +{ + int mask; + + if (opr->type == OPT_REG) { + mask = po->regmask_src; + if (po->op == OP_CALL) + // assume worst case + mask |= (1 << xAX) | (1 << xCX) | (1 << xDX); + if ((1 << opr->reg) & mask) + return 1; + else + return 0; + } + + // yes I'm lazy + return 0; +} + // is operand 'opr' modified by parsed_op 'po'? 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)) + if (!(po->flags & OPF_DATA)) return 0; if (opr->type == OPT_REG) { @@ -2415,10 +2485,6 @@ static int scan_for_mod_opr0(struct parsed_op *po_test, return -1; } -#define check_i(po, i) \ - if ((i) < 0) \ - ferr(po, "bad " #i ": %d\n", i) - static int scan_for_flag_set(int i, int magic, int *branched, int *setters, int *setter_cnt) { @@ -2712,7 +2778,8 @@ static const struct parsed_proto *try_recover_pp( } static void scan_for_call_type(int i, const struct parsed_opr *opr, - int magic, const struct parsed_proto **pp_found, int *multi) + int magic, const struct parsed_proto **pp_found, int *pp_i, + int *multi) { const struct parsed_proto *pp = NULL; struct parsed_op *po; @@ -2725,7 +2792,7 @@ static void scan_for_call_type(int i, const struct parsed_opr *opr, lr = &g_label_refs[i]; for (; lr != NULL; lr = lr->next) { check_i(&ops[i], lr->i); - scan_for_call_type(lr->i, opr, magic, pp_found, multi); + scan_for_call_type(lr->i, opr, magic, pp_found, pp_i, multi); } if (i > 0 && LAST_OP(i - 1)) return; @@ -2790,24 +2857,176 @@ static void scan_for_call_type(int i, const struct parsed_opr *opr, } *multi = 1; } - if (pp != NULL) + if (pp != NULL) { *pp_found = pp; + *pp_i = po - ops; + } +} + +static void add_label_ref(struct label_ref *lr, int op_i) +{ + struct label_ref *lr_new; + + if (lr->i == -1) { + lr->i = op_i; + return; + } + + lr_new = calloc(1, sizeof(*lr_new)); + lr_new->i = op_i; + lr_new->next = lr->next; + lr->next = lr_new; } -// early check for tail call or branch back -static int is_like_tailjmp(int j) +static struct parsed_data *try_resolve_jumptab(int i, int opcnt) { - if (!(ops[j].flags & OPF_JMP)) - return 0; + struct parsed_op *po = &ops[i]; + struct parsed_data *pd; + char label[NAMELEN], *p; + int len, j, l; - if (ops[j].op == OP_JMP && !ops[j].operand[0].had_ds) - // probably local branch back.. - return 1; - if (ops[j].op == OP_CALL) - // probably noreturn call.. - return 1; + p = strchr(po->operand[0].name, '['); + if (p == NULL) + return NULL; - return 0; + len = p - po->operand[0].name; + strncpy(label, po->operand[0].name, len); + label[len] = 0; + + for (j = 0, pd = NULL; j < g_func_pd_cnt; j++) { + if (IS(g_func_pd[j].label, label)) { + pd = &g_func_pd[j]; + break; + } + } + if (pd == NULL) + //ferr(po, "label '%s' not parsed?\n", label); + return NULL; + + if (pd->type != OPT_OFFSET) + ferr(po, "label '%s' with non-offset data?\n", label); + + // find all labels, link + for (j = 0; j < pd->count; j++) { + for (l = 0; l < opcnt; l++) { + if (g_labels[l] != NULL && IS(g_labels[l], pd->d[j].u.label)) { + add_label_ref(&g_label_refs[l], i); + pd->d[j].bt_i = l; + break; + } + } + } + + return pd; +} + +static void clear_labels(int count) +{ + int i; + + for (i = 0; i < count; i++) { + if (g_labels[i] != NULL) { + free(g_labels[i]); + g_labels[i] = NULL; + } + } +} + +static void resolve_branches_parse_calls(int opcnt) +{ + const struct parsed_proto *pp_c; + struct parsed_proto *pp; + struct parsed_data *pd; + struct parsed_op *po; + const char *tmpname; + int i, l, ret; + + for (i = 0; i < opcnt; i++) + { + po = &ops[i]; + po->bt_i = -1; + po->btj = NULL; + + 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(g_fhdr, tmpname, g_header_mode); + if (!g_header_mode && pp_c == NULL) + ferr(po, "proto_parse failed for call '%s'\n", tmpname); + + if (pp_c != NULL) { + 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->pp = pp; + continue; + } + + if (!(po->flags & OPF_JMP) || po->op == OP_RET) + continue; + + if (po->operand[0].type == OPT_REGMEM) { + pd = try_resolve_jumptab(i, opcnt); + if (pd == NULL) + goto tailcall; + + po->btj = pd; + continue; + } + + for (l = 0; l < opcnt; l++) { + if (g_labels[l] != NULL + && 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; + break; + } + add_label_ref(&g_label_refs[l], i); + po->bt_i = l; + break; + } + } + + if (po->bt_i != -1 || (po->flags & OPF_RMD)) + continue; + + if (po->operand[0].type == OPT_LABEL) + // assume tail call + goto tailcall; + + ferr(po, "unhandled branch\n"); + +tailcall: + po->op = OP_CALL; + po->flags |= OPF_TAIL; + if (i > 0 && ops[i - 1].op == OP_POP) + po->flags |= OPF_ATAIL; + i--; // reprocess + } } static void scan_prologue_epilogue(int opcnt) @@ -2857,12 +3076,13 @@ static void scan_prologue_epilogue(int opcnt) found = 0; do { for (; i < opcnt; i++) - if (ops[i].op == OP_RET) + if (ops[i].flags & OPF_TAIL) break; j = i - 1; if (i == opcnt && (ops[j].flags & OPF_JMP)) { - if (found && is_like_tailjmp(j)) - break; + if (ops[j].bt_i != -1 || ops[j].btj != NULL) + break; + i--; j--; } @@ -2871,6 +3091,14 @@ static void scan_prologue_epilogue(int opcnt) { ops[j].flags |= OPF_RMD | OPF_DONE; } + else if (ops[i].op == OP_CALL && ops[i].pp != NULL + && ops[i].pp->is_noreturn) + { + // on noreturn, msvc sometimes cleans stack, sometimes not + i++; + found = 1; + continue; + } else if (!(g_ida_func_attr & IDAFA_NORETURN)) ferr(&ops[j], "'pop ebp' expected\n"); @@ -2898,6 +3126,8 @@ static void scan_prologue_epilogue(int opcnt) i++; } while (i < opcnt); + if (!found) + ferr(ops, "missing ebp epilogue\n"); return; } @@ -2963,12 +3193,13 @@ static void scan_prologue_epilogue(int opcnt) i++; do { for (; i < opcnt; i++) - if (ops[i].op == OP_RET) + if (ops[i].flags & OPF_TAIL) break; j = i - 1; if (i == opcnt && (ops[j].flags & OPF_JMP)) { - if (found && is_like_tailjmp(j)) - break; + if (ops[j].bt_i != -1 || ops[j].btj != NULL) + break; + i--; j--; } @@ -3009,16 +3240,20 @@ static void scan_prologue_epilogue(int opcnt) i++; } while (i < opcnt); + + if (!found) + ferr(ops, "missing esp epilogue\n"); } } static const struct parsed_proto *resolve_icall(int i, int opcnt, - int *multi_src) + int *pp_i, int *multi_src) { const struct parsed_proto *pp = NULL; int search_advice = 0; *multi_src = 0; + *pp_i = -1; switch (ops[i].operand[0].type) { case OPT_REGMEM: @@ -3030,7 +3265,7 @@ static const struct parsed_proto *resolve_icall(int i, int opcnt, // fallthrough default: scan_for_call_type(i, &ops[i].operand[0], i + opcnt * 9, &pp, - multi_src); + pp_i, multi_src); break; } @@ -3038,7 +3273,7 @@ static const struct parsed_proto *resolve_icall(int i, int opcnt, } // find an instruction that changed opr before i op -// *op_i must be set to -1 by caller +// *op_i must be set to -1 by the caller // *entry is set to 1 if one source is determined to be the caller // returns 1 if found, *op_i is then set to origin static int resolve_origin(int i, const struct parsed_opr *opr, @@ -3070,7 +3305,7 @@ static int resolve_origin(int i, const struct parsed_opr *opr, } if (ops[i].cc_scratch == magic) - return 0; + return ret; ops[i].cc_scratch = magic; if (!(ops[i].flags & OPF_DATA)) @@ -3080,31 +3315,136 @@ static int resolve_origin(int i, const struct parsed_opr *opr, if (*op_i >= 0) { if (*op_i == i) - return 1; + return ret | 1; + // XXX: could check if the other op does the same return -1; } *op_i = i; - return 1; + return ret | 1; } } -static int try_resolve_const(int i, const struct parsed_opr *opr, - int magic, unsigned int *val) +// find an instruction that previously referenced opr +// if multiple results are found - fail +// *op_i must be set to -1 by the caller +// returns 1 if found, *op_i is then set to referencer insn +static int resolve_last_ref(int i, const struct parsed_opr *opr, + int magic, int *op_i) { - int s_i = -1; - int ret; - - ret = resolve_origin(i, opr, magic, &s_i, NULL); - if (ret == 1) { - i = s_i; - if (ops[i].op != OP_MOV && ops[i].operand[1].type != OPT_CONST) - return -1; + struct label_ref *lr; + int ret = 0; - *val = ops[i].operand[1].val; - return 1; - } + if (ops[i].cc_scratch == magic) + return 0; + ops[i].cc_scratch = magic; + + while (1) { + if (g_labels[i] != NULL) { + lr = &g_label_refs[i]; + for (; lr != NULL; lr = lr->next) { + check_i(&ops[i], lr->i); + ret |= resolve_last_ref(lr->i, opr, magic, op_i); + } + if (i > 0 && LAST_OP(i - 1)) + return ret; + } + + i--; + if (i < 0) + return -1; + + if (ops[i].cc_scratch == magic) + return 0; + ops[i].cc_scratch = magic; + + if (!is_opr_referenced(opr, &ops[i])) + continue; + + if (*op_i >= 0) + return -1; + + *op_i = i; + return 1; + } +} + +// find next instruction that reads opr +// if multiple results are found - fail +// *op_i must be set to -1 by the caller +// returns 1 if found, *op_i is then set to referencer insn +static int find_next_read(int i, int opcnt, + const struct parsed_opr *opr, int magic, int *op_i) +{ + struct parsed_op *po; + int j, ret = 0; + + for (; i < opcnt; i++) + { + if (ops[i].cc_scratch == magic) + return 0; + ops[i].cc_scratch = magic; + + po = &ops[i]; + if ((po->flags & OPF_JMP) && po->op != OP_CALL) { + if (po->btj != NULL) { + // jumptable + for (j = 0; j < po->btj->count; j++) { + check_i(po, po->btj->d[j].bt_i); + ret |= find_next_read(po->btj->d[j].bt_i, opcnt, opr, + magic, op_i); + } + return ret; + } + + if (po->flags & OPF_RMD) + continue; + check_i(po, po->bt_i); + if (po->flags & OPF_CJMP) { + ret = find_next_read(po->bt_i, opcnt, opr, magic, op_i); + if (ret < 0) + return ret; + } + + i = po->bt_i - 1; + continue; + } + + if (!is_opr_read(opr, po)) { + if (is_opr_modified(opr, po)) + // it's overwritten + return 0; + if (po->flags & OPF_TAIL) + return 0; + continue; + } + + if (*op_i >= 0) + return -1; + + *op_i = i; + return 1; + } + + return 0; +} + +static int try_resolve_const(int i, const struct parsed_opr *opr, + int magic, unsigned int *val) +{ + int s_i = -1; + int ret; + + ret = resolve_origin(i, opr, magic, &s_i, NULL); + if (ret == 1) { + i = s_i; + if (ops[i].op != OP_MOV && ops[i].operand[1].type != OPT_CONST) + return -1; + + *val = ops[i].operand[1].val; + return 1; + } return -1; } @@ -3147,6 +3487,7 @@ static struct parsed_proto *process_call(int i, int opcnt) const struct parsed_proto *pp_c; struct parsed_proto *pp; const char *tmpname; + int call_i = -1, ref_i = -1; int adj = 0, multipath = 0; int ret, arg; @@ -3155,7 +3496,7 @@ static struct parsed_proto *process_call(int i, int opcnt) if (pp == NULL) { // indirect call - pp_c = resolve_icall(i, opcnt, &multipath); + pp_c = resolve_icall(i, opcnt, &call_i, &multipath); if (pp_c != NULL) { if (!pp_c->is_func && !pp_c->is_fptr) ferr(po, "call to non-func: %s\n", pp_c->name); @@ -3169,6 +3510,23 @@ static struct parsed_proto *process_call(int i, int opcnt) case OPT_REG: // we resolved this call and no longer need the register po->regmask_src &= ~(1 << po->operand[0].reg); + + if (!multipath && i != call_i && ops[call_i].op == OP_MOV + && ops[call_i].operand[1].type == OPT_LABEL) + { + // no other source users? + ret = resolve_last_ref(i, &po->operand[0], opcnt * 10, + &ref_i); + if (ret == 1 && call_i == ref_i) { + // and nothing uses it after us? + ref_i = -1; + ret = find_next_read(i + 1, opcnt, &po->operand[0], + opcnt * 11, &ref_i); + if (ret != 1) + // then also don't need the source mov + ops[call_i].flags |= OPF_RMD; + } + } break; case OPT_REGMEM: pp->is_fptr = 1; @@ -3617,75 +3975,6 @@ static void pp_insert_reg_arg(struct parsed_proto *pp, const char *reg) pp->argc_reg++; } -static void add_label_ref(struct label_ref *lr, int op_i) -{ - struct label_ref *lr_new; - - if (lr->i == -1) { - lr->i = op_i; - return; - } - - lr_new = calloc(1, sizeof(*lr_new)); - lr_new->i = op_i; - lr_new->next = lr->next; - lr->next = lr_new; -} - -static struct parsed_data *try_resolve_jumptab(int i, int opcnt) -{ - struct parsed_op *po = &ops[i]; - struct parsed_data *pd; - char label[NAMELEN], *p; - int len, j, l; - - p = strchr(po->operand[0].name, '['); - if (p == NULL) - return NULL; - - len = p - po->operand[0].name; - strncpy(label, po->operand[0].name, len); - label[len] = 0; - - for (j = 0, pd = NULL; j < g_func_pd_cnt; j++) { - if (IS(g_func_pd[j].label, label)) { - pd = &g_func_pd[j]; - break; - } - } - if (pd == NULL) - //ferr(po, "label '%s' not parsed?\n", label); - return NULL; - - if (pd->type != OPT_OFFSET) - ferr(po, "label '%s' with non-offset data?\n", label); - - // find all labels, link - for (j = 0; j < pd->count; j++) { - for (l = 0; l < opcnt; l++) { - if (g_labels[l] != NULL && IS(g_labels[l], pd->d[j].u.label)) { - add_label_ref(&g_label_refs[l], i); - pd->d[j].bt_i = l; - break; - } - } - } - - return pd; -} - -static void clear_labels(int count) -{ - int i; - - for (i = 0; i < count; i++) { - if (g_labels[i] != NULL) { - free(g_labels[i]); - g_labels[i] = NULL; - } - } -} - static void output_std_flags(FILE *fout, struct parsed_op *po, int *pfomask, const char *dst_opr_text) { @@ -3802,10 +4091,8 @@ static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt) struct parsed_op *po, *delayed_flag_op = NULL, *tmp_op; struct parsed_opr *last_arith_dst = NULL; char buf1[256], buf2[256], buf3[256], cast[64]; - const struct parsed_proto *pp_c; struct parsed_proto *pp, *pp_tmp; struct parsed_data *pd; - const char *tmpname; unsigned int uval; int save_arg_vars[MAX_ARG_GRP] = { 0, }; int cond_vars = 0; @@ -3813,11 +4100,12 @@ static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt) int need_tmp64 = 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 regmask_save = 0; // regs saved/restored in this func + int regmask_arg = 0; // regs carrying function args (fastcall, etc) + int regmask_now; // temp + int regmask_init = 0; // regs that need zero initialization + int regmask_pp = 0; // regs used in complex push-pop graph + int regmask = 0; // used regs int pfomask = 0; int found = 0; int depth = 0; @@ -3837,110 +4125,28 @@ static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt) regmask_arg = get_pp_arg_regmask(g_func_pp); // pass1: - // - handle ebp/esp frame, remove ops related to it - scan_prologue_epilogue(opcnt); - - // pass2: - // - parse calls with labels // - resolve all branches - for (i = 0; i < opcnt; i++) - { - po = &ops[i]; - po->bt_i = -1; - po->btj = NULL; - - if (po->flags & (OPF_RMD|OPF_DONE)) - 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, 0); - if (pp_c == NULL) - ferr(po, "proto_parse failed for call '%s'\n", tmpname); - - 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->pp = pp; - continue; - } - - if (!(po->flags & OPF_JMP) || po->op == OP_RET) - continue; - - if (po->operand[0].type == OPT_REGMEM) { - pd = try_resolve_jumptab(i, opcnt); - if (pd == NULL) - goto tailcall; - - po->btj = pd; - continue; - } - - for (l = 0; l < opcnt; l++) { - if (g_labels[l] != NULL - && 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; - break; - } - add_label_ref(&g_label_refs[l], i); - po->bt_i = l; - break; - } - } - - if (po->bt_i != -1 || (po->flags & OPF_RMD)) - continue; - - if (po->operand[0].type == OPT_LABEL) - // assume tail call - goto tailcall; - - ferr(po, "unhandled branch\n"); + // - parse calls with labels + resolve_branches_parse_calls(opcnt); -tailcall: - po->op = OP_CALL; - po->flags |= OPF_TAIL; - if (i > 0 && ops[i - 1].op == OP_POP) - po->flags |= OPF_ATAIL; - i--; // reprocess - } + // pass2: + // - handle ebp/esp frame, remove ops related to it + scan_prologue_epilogue(opcnt); // pass3: // - remove dead labels - // - process trivial calls for (i = 0; i < opcnt; i++) { if (g_labels[i] != NULL && g_label_refs[i].i == -1) { free(g_labels[i]); g_labels[i] = NULL; } + } + // pass4: + // - process trivial calls + for (i = 0; i < opcnt; i++) + { po = &ops[i]; if (po->flags & (OPF_RMD|OPF_DONE)) continue; @@ -3973,7 +4179,7 @@ tailcall: } } - // pass4: + // pass5: // - process calls // - handle push /pop pairs for (i = 0; i < opcnt; i++) @@ -3997,10 +4203,10 @@ tailcall: } else if (po->op == OP_PUSH && !(po->flags & OPF_FARG) && !(po->flags & OPF_RSAVE) && po->operand[0].type == OPT_CONST) - scan_for_pop_const(i, opcnt); + scan_for_pop_const(i, opcnt, ®mask_pp); } - // pass5: + // pass6: // - find POPs for PUSHes, rm both // - scan for STD/CLD, propagate DF // - scan for all used registers @@ -4245,7 +4451,7 @@ tailcall: } } - // pass6: + // pass7: // - confirm regmask_save, it might have been reduced if (regmask_save != 0) { @@ -4383,6 +4589,7 @@ tailcall: } } + // declare normal registers regmask_now = regmask & ~regmask_arg; regmask_now &= ~(1 << xSP); if (regmask_now & 0x00ff) { @@ -4429,6 +4636,16 @@ tailcall: } } + // declare push-pop temporaries + if (regmask_pp) { + for (reg = 0; reg < 8; reg++) { + if (regmask_pp & (1 << reg)) { + fprintf(fout, " u32 pp_%s;\n", regs_r32[reg]); + had_decl = 1; + } + } + } + if (cond_vars) { for (i = 0; i < 8; i++) { if (cond_vars & (1 << i)) { @@ -4766,6 +4983,19 @@ tailcall: // arithmetic w/flags case OP_AND: + if (po->operand[1].type == OPT_CONST && !po->operand[1].val) { + // deal with complex dst clear + assert_operand_cnt(2); + fprintf(fout, " %s = %s;", + out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]), + out_src_opr(buf2, sizeof(buf2), po, &po->operand[1], + default_cast_to(buf3, sizeof(buf3), &po->operand[0]), 0)); + output_std_flags(fout, po, &pfomask, buf1); + last_arith_dst = &po->operand[0]; + delayed_flag_op = NULL; + break; + } + // fallthrough case OP_OR: propagate_lmod(po, &po->operand[0], &po->operand[1]); // fallthrough @@ -5368,6 +5598,13 @@ tailcall: fprintf(fout, " s_%s = %s;", buf1, buf1); break; } + else if (po->flags & OPF_PPUSH) { + tmp_op = po->datap; + ferr_assert(po, tmp_op != NULL); + out_dst_opr(buf2, sizeof(buf2), po, &tmp_op->operand[0]); + fprintf(fout, " pp_%s = %s;", buf2, buf1); + break; + } else if (g_func_pp->is_userstack) { fprintf(fout, " *(--esp) = %s;", buf1); break; @@ -5378,15 +5615,20 @@ tailcall: break; case OP_POP: + out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]); if (po->flags & OPF_RSAVE) { - out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]); fprintf(fout, " %s = s_%s;", buf1, buf1); break; } + else if (po->flags & OPF_PPUSH) { + // push/pop graph + ferr_assert(po, po->datap == NULL); + fprintf(fout, " %s = pp_%s;", buf1, buf1); + break; + } else if (po->datap != NULL) { // push/pop pair tmp_op = po->datap; - out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]); fprintf(fout, " %s = %s;", buf1, out_src_opr(buf2, sizeof(buf2), tmp_op, &tmp_op->operand[0], @@ -5394,8 +5636,7 @@ tailcall: break; } else if (g_func_pp->is_userstack) { - fprintf(fout, " %s = *esp++;", - out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0])); + fprintf(fout, " %s = *esp++;", buf1); break; } else @@ -5528,6 +5769,25 @@ static int hg_var_cnt; static void output_hdr_fp(FILE *fout, const struct func_prototype *fp, int count); +struct func_prototype *hg_fp_add(const char *funcn) +{ + struct func_prototype *fp; + + if ((hg_fp_cnt & 0xff) == 0) { + hg_fp = realloc(hg_fp, sizeof(hg_fp[0]) * (hg_fp_cnt + 0x100)); + my_assert_not(hg_fp, NULL); + memset(hg_fp + hg_fp_cnt, 0, sizeof(hg_fp[0]) * 0x100); + } + + fp = &hg_fp[hg_fp_cnt]; + snprintf(fp->name, sizeof(fp->name), "%s", funcn); + fp->id = hg_fp_cnt; + fp->argc_stack = -1; + hg_fp_cnt++; + + return fp; +} + static struct func_proto_dep *hg_fp_find_dep(struct func_prototype *fp, const char *name) { @@ -5599,17 +5859,14 @@ static void gen_hdr_dep_pass(int i, int opcnt, unsigned char *cbits, if (po->btj != NULL) { // jumptable for (j = 0; j < po->btj->count; j++) { + check_i(po, po->btj->d[j].bt_i); gen_hdr_dep_pass(po->btj->d[j].bt_i, opcnt, cbits, fp, regmask_save, regmask_dst, regmask_dep, has_ret); } return; } - if (po->bt_i < 0) { - ferr(po, "dead branch\n"); - return; - } - + check_i(po, po->bt_i); if (po->flags & OPF_CJMP) { gen_hdr_dep_pass(po->bt_i, opcnt, cbits, fp, regmask_save, regmask_dst, regmask_dep, has_ret); @@ -5665,6 +5922,8 @@ static void gen_hdr_dep_pass(int i, int opcnt, unsigned char *cbits, } } + // if has_ret is 0, there is uninitialized eax path, + // which means it's most likely void func if (*has_ret != 0 && (po->flags & OPF_TAIL)) { if (po->op == OP_CALL) { j = i; @@ -5679,13 +5938,13 @@ static void gen_hdr_dep_pass(int i, int opcnt, unsigned char *cbits, ret = resolve_origin(i, &opr, i + opcnt * 4, &j, &from_caller); } - if (ret == -1 && from_caller) { + if (ret != 1 && from_caller) { // unresolved eax - probably void func *has_ret = 0; } else { - if (ops[j].op == OP_CALL) { - dep = hg_fp_find_dep(fp, po->operand[0].name); + if (j >= 0 && ops[j].op == OP_CALL) { + dep = hg_fp_find_dep(fp, ops[j].operand[0].name); if (dep != NULL) dep->ret_dep = 1; else @@ -5721,124 +5980,55 @@ static void gen_hdr(const char *funcn, int opcnt) const struct parsed_proto *pp_c; struct parsed_proto *pp; struct func_prototype *fp; - struct parsed_data *pd; struct parsed_op *po; - const char *tmpname; int regmask_dummy = 0; int regmask_dep; int max_bp_offset = 0; int has_ret; - int i, j, l, ret; - - if ((hg_fp_cnt & 0xff) == 0) { - hg_fp = realloc(hg_fp, sizeof(hg_fp[0]) * (hg_fp_cnt + 0x100)); - my_assert_not(hg_fp, NULL); - memset(hg_fp + hg_fp_cnt, 0, sizeof(hg_fp[0]) * 0x100); - } + int i, j, ret; - fp = &hg_fp[hg_fp_cnt]; - snprintf(fp->name, sizeof(fp->name), "%s", funcn); - fp->id = hg_fp_cnt; - fp->argc_stack = -1; - hg_fp_cnt++; - - // perhaps already in seed header? - fp->pp = proto_parse(g_fhdr, funcn, 1); - if (fp->pp != NULL) { - fp->argc_stack = fp->pp->argc_stack; - fp->is_stdcall = fp->pp->is_stdcall; - fp->regmask_dep = get_pp_arg_regmask(fp->pp); - fp->has_ret = !IS(fp->pp->ret_type.name, "void"); + pp_c = proto_parse(g_fhdr, funcn, 1); + if (pp_c != NULL) + // already in seed, will add to hg_fp later return; - } + + fp = hg_fp_add(funcn); g_bp_frame = g_sp_frame = g_stack_fsz = 0; g_stack_frame_used = 0; // pass1: + // - resolve all branches + // - parse calls with labels + resolve_branches_parse_calls(opcnt); + + // pass2: // - handle ebp/esp frame, remove ops related to it scan_prologue_epilogue(opcnt); - // pass2: + // pass3: + // - remove dead labels // - collect calls - // - resolve all branches for (i = 0; i < opcnt; i++) { - po = &ops[i]; - po->bt_i = -1; - po->btj = NULL; + if (g_labels[i] != NULL && g_label_refs[i].i == -1) { + free(g_labels[i]); + g_labels[i] = NULL; + } + po = &ops[i]; if (po->flags & (OPF_RMD|OPF_DONE)) continue; if (po->op == OP_CALL) { - tmpname = opr_name(po, 0); - pp = NULL; - if (po->operand[0].type == OPT_LABEL) { - hg_fp_add_dep(fp, tmpname); - - // perhaps a call to already known func? - pp_c = proto_parse(g_fhdr, tmpname, 1); - if (pp_c != NULL) - pp = proto_clone(pp_c); - } - 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 && pp->is_noreturn) - po->flags |= OPF_TAIL; - - po->pp = pp; - continue; - } - - if (!(po->flags & OPF_JMP) || po->op == OP_RET) - continue; - - if (po->operand[0].type == OPT_REGMEM) { - pd = try_resolve_jumptab(i, opcnt); - if (pd == NULL) - goto tailcall; - - po->btj = pd; - continue; - } - - for (l = 0; l < opcnt; l++) { - if (g_labels[l] != NULL - && IS(po->operand[0].name, g_labels[l])) - { - add_label_ref(&g_label_refs[l], i); - po->bt_i = l; - break; - } + if (po->operand[0].type == OPT_LABEL) + hg_fp_add_dep(fp, opr_name(po, 0)); + else if (po->pp != NULL) + hg_fp_add_dep(fp, po->pp->name); } - - if (po->bt_i != -1 || (po->flags & OPF_RMD)) - continue; - - if (po->operand[0].type == OPT_LABEL) - // assume tail call - goto tailcall; - - ferr(po, "unhandled branch\n"); - -tailcall: - po->op = OP_CALL; - po->flags |= OPF_TAIL; - if (i > 0 && ops[i - 1].op == OP_POP) - po->flags |= OPF_ATAIL; - i--; // reprocess } - // pass3: + // pass4: // - remove dead labels // - handle push /pop pairs for (i = 0; i < opcnt; i++) @@ -5853,10 +6043,10 @@ tailcall: continue; if (po->op == OP_PUSH && po->operand[0].type == OPT_CONST) - scan_for_pop_const(i, opcnt); + scan_for_pop_const(i, opcnt, ®mask_dummy); } - // pass4: + // pass5: // - process trivial calls for (i = 0; i < opcnt; i++) { @@ -5889,7 +6079,7 @@ tailcall: } } - // pass5: + // pass6: // - track saved regs (simple) // - process calls for (i = 0; i < opcnt; i++) @@ -5919,7 +6109,7 @@ tailcall: } } - // pass6 + // pass7 memset(cbits, 0, sizeof(cbits)); regmask_dep = 0; has_ret = -1; @@ -5936,9 +6126,6 @@ tailcall: ferr(&ops[i], "unreachable code\n"); } - if (has_ret == -1 && (regmask_dep & (1 << xAX))) - has_ret = 1; - for (i = 0; i < g_eqcnt; i++) { if (g_eqs[i].offset > max_bp_offset && g_eqs[i].offset < 4*32) max_bp_offset = g_eqs[i].offset; @@ -5957,7 +6144,7 @@ tailcall: printf("// has_ret %d, regmask_dep %x\n", fp->has_ret, fp->regmask_dep); output_hdr_fp(stdout, fp, 1); - if (IS(funcn, "sub_100073FD")) exit(1); + if (IS(funcn, "sub_10007F72")) exit(1); #endif gen_x_cleanup(opcnt); @@ -5986,7 +6173,7 @@ static void hg_fp_resolve_deps(struct func_prototype *fp) // printf("dep %s %s |= %x\n", fp->name, // fp->dep_func[i].name, dep); - if (fp->has_ret == -1) + if (fp->has_ret == -1 && fp->dep_func[i].ret_dep) fp->has_ret = fp->dep_func[i].proto->has_ret; } } @@ -6032,17 +6219,8 @@ static void output_hdr_fp(FILE *fout, const struct func_prototype *fp, continue; if (fp->pp != NULL) { - // prefer fp for common style, - // only use output_pp if args are complex - for (j = 0; j < fp->pp->argc; j++) { - if (fp->pp->arg[j].fptr != NULL) - break; - } - if (j != fp->pp->argc) { - output_pp(fout, fp->pp, OPP_ALIGN); - fprintf(fout, ";\n"); - continue; - } + // part of seed, output later + continue; } regmask_dep = fp->regmask_dep; @@ -6115,9 +6293,25 @@ static void output_hdr(FILE *fout) [OPLM_QWORD] = "uint64_t", }; const struct scanned_var *var; + struct func_prototype *fp; char line[256] = { 0, }; + char name[256]; int i; + // add stuff from headers + for (i = 0; i < pp_cache_size; i++) { + if (pp_cache[i].is_cinc && !pp_cache[i].is_stdcall) + snprintf(name, sizeof(name), "_%s", pp_cache[i].name); + else + snprintf(name, sizeof(name), "%s", pp_cache[i].name); + fp = hg_fp_add(name); + fp->pp = &pp_cache[i]; + fp->argc_stack = fp->pp->argc_stack; + fp->is_stdcall = fp->pp->is_stdcall; + fp->regmask_dep = get_pp_arg_regmask(fp->pp); + fp->has_ret = !IS(fp->pp->ret_type.name, "void"); + } + // resolve deps qsort(hg_fp, hg_fp_cnt, sizeof(hg_fp[0]), hg_fp_cmp_name); for (i = 0; i < hg_fp_cnt; i++) @@ -6130,11 +6324,9 @@ static void output_hdr(FILE *fout) for (i = 0; i < hg_var_cnt; i++) { var = &hg_vars[i]; - if (var->pp != NULL && var->pp->is_fptr) { - fprintf(fout, "extern "); - output_pp(fout, var->pp, 0); - fprintf(fout, ";"); - } + if (var->pp != NULL) + // part of seed + continue; else if (var->is_c_str) fprintf(fout, "extern %-8s %s[];", "char", var->name); else @@ -6151,38 +6343,12 @@ static void output_hdr(FILE *fout) // output function prototypes output_hdr_fp(fout, hg_fp, hg_fp_cnt); - // include passthrough - fprintf(fout, "\n// for translate\n"); + // seed passthrough + fprintf(fout, "\n// - seed -\n"); rewind(g_fhdr); - while (fgets(line, sizeof(line), g_fhdr)) { - if (IS_START(line, "//#")) - fwrite(line, 1, strlen(line), fout); - } -} - -// read a line, truncating it if it doesn't fit -static char *my_fgets(char *s, size_t size, FILE *stream) -{ - char *ret, *ret2; - char buf[64]; - int p; - - p = size - 2; - if (p >= 0) - s[p] = 0; - - ret = fgets(s, size, stream); - if (ret != NULL && p >= 0 && s[p] != 0 && s[p] != '\n') { - p = sizeof(buf) - 2; - do { - buf[p] = 0; - ret2 = fgets(buf, sizeof(buf), stream); - } - while (ret2 != NULL && buf[p] != 0 && buf[p] != '\n'); - } - - return ret; + while (fgets(line, sizeof(line), g_fhdr)) + fwrite(line, 1, strlen(line), fout); } // '=' needs special treatment @@ -6284,6 +6450,11 @@ static void scan_variables(FILE *fasm) if (wordc < 2) continue; + if (IS_START(words[0], "__IMPORT_DESCRIPTOR_")) { + // when this starts, we don't need anything from this section + break; + } + if ((hg_var_cnt & 0xff) == 0) { hg_vars = realloc(hg_vars, sizeof(hg_vars[0]) * (hg_var_cnt + 0x100)); @@ -6741,7 +6912,7 @@ parse_words: do_pending_endp: // do delayed endp processing to collect switch jumptables if (pending_endp) { - if (in_func && !skip_func && !end && wordc >= 2 + if (in_func && !g_skip_func && !end && wordc >= 2 && ((words[0][0] == 'd' && words[0][2] == 0) || (words[1][0] == 'd' && words[1][2] == 0))) { @@ -6796,7 +6967,7 @@ do_pending_endp: continue; } - if (in_func && !skip_func) { + if (in_func && !g_skip_func) { if (g_header_mode) gen_hdr(g_func, pi); else @@ -6807,7 +6978,7 @@ do_pending_endp: in_func = 0; g_ida_func_attr = 0; skip_warned = 0; - skip_func = 0; + g_skip_func = 0; g_func[0] = 0; func_chunks_used = 0; func_chunk_i = -1; @@ -6841,7 +7012,7 @@ do_pending_endp: words[0], g_func); p = words[0]; if (bsearch(&p, rlist, rlist_len, sizeof(rlist[0]), cmpstringp)) - skip_func = 1; + g_skip_func = 1; strcpy(g_func, words[0]); set_label(0, words[0]); in_func = 1; @@ -6860,10 +7031,10 @@ do_pending_endp: && ops[0].op == OP_JMP && ops[0].operand[0].had_ds) { // import jump - skip_func = 1; + g_skip_func = 1; } - if (!skip_func && func_chunks_used) { + if (!g_skip_func && func_chunks_used) { // start processing chunks struct chunk_item *ci, key = { g_func, 0 }; @@ -6922,8 +7093,8 @@ do_pending_endp: continue; } - if (!in_func || skip_func) { - if (!skip_warned && !skip_func && g_labels[pi] != NULL) { + if (!in_func || g_skip_func) { + if (!skip_warned && !g_skip_func && g_labels[pi] != NULL) { if (verbose) anote("skipping from '%s'\n", g_labels[pi]); skip_warned = 1;