/*
* ia32rtools
- * (C) notaz, 2013,2014
+ * (C) notaz, 2013-2015
*
* This work is licensed under the terms of 3-clause BSD license.
* See COPYING file in the top-level directory.
#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)
#include "masm_tools.h"
enum op_flags {
- OPF_RMD = (1 << 0), /* removed or optimized out */
+ OPF_RMD = (1 << 0), /* removed from code generation */
OPF_DATA = (1 << 1), /* data processing - writes to dst opr */
OPF_FLAGS = (1 << 2), /* sets flags */
OPF_JMP = (1 << 3), /* branch, call */
OPF_REP = (1 << 8), /* prefixed by rep */
OPF_REPZ = (1 << 9), /* rep is repe/repz */
OPF_REPNZ = (1 << 10), /* rep is repne/repnz */
- OPF_FARG = (1 << 11), /* push collected as func arg (no reuse) */
- OPF_EBP_S = (1 << 12), /* ebp used as scratch here, not BP */
- OPF_DF = (1 << 13), /* DF flag set */
- OPF_ATAIL = (1 << 14), /* tail call with reused arg frame */
- OPF_32BIT = (1 << 15), /* 32bit division */
- OPF_LOCK = (1 << 16), /* op has lock prefix */
- OPF_VAPUSH = (1 << 17), /* vararg ptr push (as call arg) */
+ OPF_FARG = (1 << 11), /* push collected as func arg */
+ OPF_FARGNR = (1 << 12), /* push collected as func arg (no reuse) */
+ OPF_EBP_S = (1 << 13), /* ebp used as scratch here, not BP */
+ OPF_DF = (1 << 14), /* DF flag set */
+ OPF_ATAIL = (1 << 15), /* tail call with reused arg frame */
+ OPF_32BIT = (1 << 16), /* 32bit division */
+ 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 {
// x87
// mmx
OP_EMMS,
+ // undefined
+ OP_UD2,
};
enum opr_type {
};
// datap:
-// OP_CALL - parser proto hint (str)
-// (OPF_CC) - point to one of (OPF_FLAGS) that affects cc op
-// OP_POP - point to OP_PUSH in push/pop pair
+// OP_CALL - parser proto hint (str)
+// (OPF_CC) - points to one of (OPF_FLAGS) that affects cc op
+// 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];
static struct parsed_op ops[MAX_OPS];
static struct parsed_equ *g_eqs;
static int g_eqcnt;
-static char g_labels[MAX_OPS][48];
+static char *g_labels[MAX_OPS];
static struct label_ref g_label_refs[MAX_OPS];
static const struct parsed_proto *g_func_pp;
static struct parsed_data *g_func_pd;
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;
#define ferr(op_, fmt, ...) do { \
printf("%s:%d: error: [%s] '%s': " fmt, asmfn, (op_)->asmln, g_func, \
printf("%s:%d: note: [%s] '%s': " fmt, asmfn, (op_)->asmln, g_func, \
dump_op(op_), ##__VA_ARGS__)
+#define ferr_assert(op_, cond) do { \
+ if (!(cond)) ferr(op_, "assertion '%s' failed on ln :%d\n", #cond, \
+ __LINE__); \
+} while (0)
+
const char *regs_r32[] = {
"eax", "ebx", "ecx", "edx", "esi", "edi", "ebp", "esp",
// not r32, but list here for easy parsing and printing
const struct parsed_type *c_type)
{
static const char *dword_types[] = {
- "int", "_DWORD", "UINT_PTR", "DWORD",
+ "uint32_t", "int", "_DWORD", "UINT_PTR", "DWORD",
"WPARAM", "LPARAM", "UINT", "__int32",
"LONG", "HIMC", "BOOL", "size_t",
"float",
// mmx
{ "emms", OP_EMMS, 0, 0, OPF_DATA },
{ "movq", OP_MOV, 2, 2, OPF_DATA },
+ // must be last
+ { "ud2", OP_UD2 },
};
static void parse_op(struct parsed_op *op, char words[16][256], int wordc)
break;
}
- if (i == ARRAY_SIZE(op_table))
- aerr("unhandled op: '%s'\n", words[0]);
+ if (i == ARRAY_SIZE(op_table)) {
+ if (!g_skip_func)
+ aerr("unhandled op: '%s'\n", words[0]);
+ i--; // OP_UD2
+ }
w++;
op->op = op_table[i].op;
op->regmask_src = op->regmask_dst = 0;
op->asmln = asmln;
+ if (op->op == OP_UD2)
+ return;
+
for (opr = 0; opr < op_table[i].maxopr; opr++) {
if (opr >= op_table[i].minopr && w >= wordc)
break;
&& op->operand[0].reg == op->operand[1].reg
&& IS(op->operand[0].name, op->operand[1].name)) // ! ah, al..
{
- op->flags |= OPF_RMD;
+ op->flags |= OPF_RMD | OPF_DONE;
op->regmask_src = op->regmask_dst = 0;
}
break;
char buf[16];
snprintf(buf, sizeof(buf), "%s+0", op->operand[0].name);
if (IS(buf, op->operand[1].name))
- op->flags |= OPF_RMD;
+ op->flags |= OPF_RMD | OPF_DONE;
}
break;
// fptrs must use 32bit args, callsite might have no information and
// lack a cast to smaller types, which results in incorrectly masked
// args passed (callee may assume masked args, it does on ARM)
- if (!pp->is_oslib) {
+ if (!pp->is_osinc) {
for (i = 0; i < pp->argc; i++) {
ret = guess_lmod_from_c_type(&tmp_lmod, &pp->arg[i].type);
if (ret && tmp_lmod != OPLM_DWORD)
|| ((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)
{
return -1; // deadend
}
- if ((po->flags & OPF_RMD)
+ if ((po->flags & (OPF_RMD|OPF_DONE))
|| (po->op == OP_PUSH && po->p_argnum != 0)) // arg push
continue;
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)
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);
&& po->operand[0].type == OPT_REG
&& IS(po->operand[0].name, reg))
{
- if (po->op == OP_PUSH && !(po->flags & OPF_FARG)) {
+ if (po->op == OP_PUSH && !(po->flags & OPF_FARGNR)) {
depth++;
if (depth > *maxdepth)
*maxdepth = depth;
continue;
for (j = i - 1; j >= 0; j--) {
- if (ops[j].flags & OPF_RMD)
+ if (ops[j].flags & (OPF_RMD|OPF_DONE))
continue;
if (ops[j].flags & OPF_JMP)
return -1;
- if (ops[j].op == OP_POP && ops[j].operand[0].type == OPT_REG
+ if (ops[j].op == OP_POP && ops[j].datap == NULL
+ && ops[j].operand[0].type == OPT_REG
&& IS(ops[j].operand[0].name, reg))
{
found = 1;
break;
}
- if (g_labels[j][0] != 0)
+ if (g_labels[j] != NULL)
return -1;
}
}
return found ? 0 : -1;
}
+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++) {
+ 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 (g_labels[j] != NULL)
+ is_multipath = 1;
+
+ if (po->op == OP_POP && !(po->flags & OPF_RMD))
+ {
+ 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;
+ }
+ }
+}
+
static void scan_propagate_df(int i, int opcnt)
{
struct parsed_op *po = &ops[i];
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
break;
if (po->op == OP_CLD) {
- po->flags |= OPF_RMD;
+ po->flags |= OPF_RMD | OPF_DONE;
return;
}
}
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) {
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)
{
}
ops[i].cc_scratch = magic;
- if (g_labels[i][0] != 0) {
+ if (g_labels[i] != NULL) {
*branched = 1;
lr = &g_label_refs[i];
static int scan_for_cdq_edx(int i)
{
while (i >= 0) {
- if (g_labels[i][0] != 0) {
+ if (g_labels[i] != NULL) {
if (g_label_refs[i].next != NULL)
return -1;
if (i > 0 && LAST_OP(i - 1)) {
static int scan_for_reg_clear(int i, int reg)
{
while (i >= 0) {
- if (g_labels[i][0] != 0) {
+ if (g_labels[i] != NULL) {
if (g_label_refs[i].next != NULL)
return -1;
if (i > 0 && LAST_OP(i - 1)) {
return -1;
}
+static void patch_esp_adjust(struct parsed_op *po, int adj)
+{
+ ferr_assert(po, po->op == OP_ADD);
+ ferr_assert(po, IS(opr_name(po, 0), "esp"));
+ ferr_assert(po, po->operand[1].type == OPT_CONST);
+
+ // this is a bit of a hack, but deals with use of
+ // single adj for multiple calls
+ po->operand[1].val -= adj;
+ po->flags |= OPF_RMD;
+ if (po->operand[1].val == 0)
+ po->flags |= OPF_DONE;
+ ferr_assert(po, (int)po->operand[1].val >= 0);
+}
+
// scan for positive, constant esp adjust
-static int scan_for_esp_adjust(int i, int opcnt, int *adj,
- int *multipath)
+// multipath case is preliminary
+static int scan_for_esp_adjust(int i, int opcnt,
+ int adj_expect, int *adj, int *is_multipath, int do_update)
{
struct parsed_op *po;
int first_pop = -1;
- *adj = *multipath = 0;
+ *adj = *is_multipath = 0;
- for (; i < opcnt; i++) {
- po = &ops[i];
+ for (; i < opcnt && *adj < adj_expect; i++) {
+ if (g_labels[i] != NULL)
+ *is_multipath = 1;
- if (g_labels[i][0] != 0)
- *multipath = 1;
+ po = &ops[i];
+ if (po->flags & OPF_DONE)
+ continue;
if (po->op == OP_ADD && po->operand[0].reg == xSP) {
if (po->operand[1].type != OPT_CONST)
*adj += po->operand[1].val;
if (*adj & 3)
ferr(&ops[i], "unaligned esp adjust: %x\n", *adj);
+ if (do_update) {
+ if (!*is_multipath)
+ patch_esp_adjust(po, adj_expect);
+ else
+ po->flags |= OPF_RMD;
+ }
return i;
}
- else if (po->op == OP_PUSH && !(po->flags & OPF_RMD)) {
+ else if (po->op == OP_PUSH) {
//if (first_pop == -1)
// first_pop = -2; // none
*adj -= lmod_bytes(po, po->operand[0].lmod);
}
- else if (po->op == OP_POP && !(po->flags & OPF_RMD)) {
- // seems like msvc only uses 'pop ecx' for stack realignment..
- if (po->operand[0].type != OPT_REG || po->operand[0].reg != xCX)
- break;
- if (first_pop == -1 && *adj >= 0)
- first_pop = i;
+ else if (po->op == OP_POP) {
+ if (!(po->flags & OPF_DONE)) {
+ // seems like msvc only uses 'pop ecx' for stack realignment..
+ if (po->operand[0].type != OPT_REG || po->operand[0].reg != xCX)
+ break;
+ if (first_pop == -1 && *adj >= 0)
+ first_pop = i;
+ }
+ if (do_update && *adj >= 0) {
+ po->flags |= OPF_RMD;
+ if (!*is_multipath)
+ po->flags |= OPF_DONE;
+ }
+
*adj += lmod_bytes(po, po->operand[0].lmod);
}
else if (po->flags & (OPF_JMP|OPF_TAIL)) {
if (po->op == OP_JMP && po->btj == NULL) {
+ if (po->bt_i <= i)
+ break;
i = po->bt_i - 1;
continue;
}
break;
if (po->pp != NULL && po->pp->is_stdcall)
break;
+ // assume it's another cdecl call
}
}
int stack_ra = 0;
int offset = 0;
+ if (g_header_mode)
+ return NULL;
+
parse_stack_access(po, opr->name, ofs_reg,
&offset, &stack_ra, NULL, 0);
if (ofs_reg[0] != 0)
pp = proto_parse(g_fhdr, buf, g_quiet_pp);
}
else if (opr->type == OPT_OFFSET || opr->type == OPT_LABEL) {
- pp = proto_parse(g_fhdr, opr->name, 0);
- if (pp == NULL)
- ferr(po, "proto_parse failed for icall from '%s'\n", opr->name);
- check_func_pp(po, pp, "reg-fptr ref");
+ pp = proto_parse(g_fhdr, opr->name, g_quiet_pp);
+ if (pp == NULL) {
+ if (!g_header_mode)
+ ferr(po, "proto_parse failed for icall to '%s'\n", opr->name);
+ }
+ else
+ check_func_pp(po, pp, "reg-fptr ref");
}
return 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;
ops[i].cc_scratch = magic;
while (1) {
- if (g_labels[i][0] != 0) {
+ if (g_labels[i] != NULL) {
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;
}
*multi = 1;
}
- if (pp != NULL)
+ if (pp != NULL) {
*pp_found = pp;
+ *pp_i = po - ops;
+ }
}
-static const struct parsed_proto *resolve_icall(int i, int opcnt,
- int *multi_src)
+static void add_label_ref(struct label_ref *lr, int op_i)
{
- const struct parsed_proto *pp = NULL;
- int search_advice = 0;
-
- *multi_src = 0;
+ struct label_ref *lr_new;
- 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], &search_advice);
- if (!search_advice)
- break;
- // fallthrough
- default:
- scan_for_call_type(i, &ops[i].operand[0], i + opcnt * 9, &pp,
- multi_src);
- break;
+ if (lr->i == -1) {
+ lr->i = op_i;
+ return;
}
- return pp;
+ lr_new = calloc(1, sizeof(*lr_new));
+ lr_new->i = op_i;
+ lr_new->next = lr->next;
+ lr->next = lr_new;
}
-// find an instruction that changed opr before i op
-// *op_i must be set to -1 by 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,
- int magic, int *op_i, int *is_caller)
+static struct parsed_data *try_resolve_jumptab(int i, int opcnt)
{
- struct label_ref *lr;
- int ret = 0;
+ struct parsed_op *po = &ops[i];
+ struct parsed_data *pd;
+ char label[NAMELEN], *p;
+ int len, j, l;
- if (ops[i].cc_scratch == magic)
- return 0;
- ops[i].cc_scratch = magic;
+ p = strchr(po->operand[0].name, '[');
+ if (p == NULL)
+ return NULL;
- while (1) {
- if (g_labels[i][0] != 0) {
- lr = &g_label_refs[i];
- for (; lr != NULL; lr = lr->next) {
- check_i(&ops[i], lr->i);
- ret |= resolve_origin(lr->i, opr, magic, op_i, is_caller);
- }
- if (i > 0 && LAST_OP(i - 1))
- return ret;
- }
+ len = p - po->operand[0].name;
+ strncpy(label, po->operand[0].name, len);
+ label[len] = 0;
- i--;
- if (i < 0) {
- if (is_caller != NULL)
- *is_caller = 1;
- return -1;
+ 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 (ops[i].cc_scratch == magic)
- return 0;
- ops[i].cc_scratch = magic;
-
- if (!(ops[i].flags & OPF_DATA))
- continue;
- if (!is_opr_modified(opr, &ops[i]))
- continue;
+ if (pd->type != OPT_OFFSET)
+ ferr(po, "label '%s' with non-offset data?\n", label);
- if (*op_i >= 0) {
- if (*op_i == i)
- return 1;
- // XXX: could check if the other op does the same
- return -1;
+ // 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;
+ }
}
-
- *op_i = i;
- return 1;
}
+
+ return pd;
}
-static int try_resolve_const(int i, const struct parsed_opr *opr,
- int magic, unsigned int *val)
+static void clear_labels(int count)
{
- 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;
+ int i;
- *val = ops[i].operand[1].val;
- return 1;
+ for (i = 0; i < count; i++) {
+ if (g_labels[i] != NULL) {
+ free(g_labels[i]);
+ g_labels[i] = NULL;
+ }
}
-
- return -1;
}
-static int collect_call_args_r(struct parsed_op *po, int i,
- struct parsed_proto *pp, int *regmask, int *save_arg_vars,
- int *arg_grp, int arg, int magic, int need_op_saving, int may_reuse)
+static void resolve_branches_parse_calls(int opcnt)
{
- struct parsed_proto *pp_tmp;
- struct parsed_op *po_tmp;
- struct label_ref *lr;
- int need_to_save_current;
- int arg_grp_current = 0;
- int save_args_seen = 0;
- int save_args;
- int ret = 0;
- int reg;
- char buf[32];
- int j, k;
-
- if (i < 0) {
- ferr(po, "dead label encountered\n");
- return -1;
- }
-
- for (; arg < pp->argc; arg++)
- if (pp->arg[arg].reg == NULL)
- break;
- magic = (magic & 0xffffff) | (arg << 24);
+ 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 (j = i; j >= 0 && (arg < pp->argc || pp->is_unresolved); )
+ for (i = 0; i < opcnt; i++)
{
- if (((ops[j].cc_scratch ^ magic) & 0xffffff) == 0) {
- if (ops[j].cc_scratch != magic) {
- ferr(&ops[j], "arg collect hit same path with diff args for %s\n",
- pp->name);
- return -1;
- }
- // ok: have already been here
- return 0;
- }
- ops[j].cc_scratch = magic;
+ po = &ops[i];
+ po->bt_i = -1;
+ po->btj = NULL;
- if (g_labels[j][0] != 0 && g_label_refs[j].i != -1) {
- lr = &g_label_refs[j];
- if (lr->next != NULL)
- need_op_saving = 1;
+ 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)
+{
+ int ecx_push = 0, esp_sub = 0;
+ int found;
+ int i, j, l;
+
+ if (ops[0].op == OP_PUSH && IS(opr_name(&ops[0], 0), "ebp")
+ && ops[1].op == OP_MOV
+ && IS(opr_name(&ops[1], 0), "ebp")
+ && IS(opr_name(&ops[1], 1), "esp"))
+ {
+ g_bp_frame = 1;
+ ops[0].flags |= OPF_RMD | OPF_DONE;
+ ops[1].flags |= OPF_RMD | OPF_DONE;
+ i = 2;
+
+ if (ops[2].op == OP_SUB && IS(opr_name(&ops[2], 0), "esp")) {
+ g_stack_fsz = opr_const(&ops[2], 1);
+ ops[2].flags |= OPF_RMD | OPF_DONE;
+ i++;
+ }
+ else {
+ // another way msvc builds stack frame..
+ i = 2;
+ while (ops[i].op == OP_PUSH && IS(opr_name(&ops[i], 0), "ecx")) {
+ g_stack_fsz += 4;
+ ops[i].flags |= OPF_RMD | OPF_DONE;
+ ecx_push++;
+ i++;
+ }
+ // and another way..
+ if (i == 2 && ops[i].op == OP_MOV && ops[i].operand[0].reg == xAX
+ && ops[i].operand[1].type == OPT_CONST
+ && ops[i + 1].op == OP_CALL
+ && IS(opr_name(&ops[i + 1], 0), "__alloca_probe"))
+ {
+ g_stack_fsz += ops[i].operand[1].val;
+ ops[i].flags |= OPF_RMD | OPF_DONE;
+ i++;
+ ops[i].flags |= OPF_RMD | OPF_DONE;
+ i++;
+ }
+ }
+
+ found = 0;
+ do {
+ for (; i < opcnt; i++)
+ if (ops[i].flags & OPF_TAIL)
+ break;
+ j = i - 1;
+ if (i == opcnt && (ops[j].flags & OPF_JMP)) {
+ if (ops[j].bt_i != -1 || ops[j].btj != NULL)
+ break;
+ i--;
+ j--;
+ }
+
+ if ((ops[j].op == OP_POP && IS(opr_name(&ops[j], 0), "ebp"))
+ || ops[j].op == OP_LEAVE)
+ {
+ 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");
+
+ if (g_stack_fsz != 0) {
+ if (ops[j - 1].op == OP_MOV
+ && IS(opr_name(&ops[j - 1], 0), "esp")
+ && IS(opr_name(&ops[j - 1], 1), "ebp"))
+ {
+ ops[j - 1].flags |= OPF_RMD | OPF_DONE;
+ }
+ else if (ops[j].op != OP_LEAVE
+ && !(g_ida_func_attr & IDAFA_NORETURN))
+ {
+ ferr(&ops[j - 1], "esp restore expected\n");
+ }
+
+ if (ecx_push && ops[j - 2].op == OP_POP
+ && IS(opr_name(&ops[j - 2], 0), "ecx"))
+ {
+ ferr(&ops[j - 2], "unexpected ecx pop\n");
+ }
+ }
+
+ found = 1;
+ i++;
+ } while (i < opcnt);
+
+ if (!found)
+ ferr(ops, "missing ebp epilogue\n");
+ return;
+ }
+
+ // non-bp frame
+ i = 0;
+ while (ops[i].op == OP_PUSH && IS(opr_name(&ops[i], 0), "ecx")) {
+ ops[i].flags |= OPF_RMD | OPF_DONE;
+ g_stack_fsz += 4;
+ ecx_push++;
+ i++;
+ }
+
+ for (; i < opcnt; i++) {
+ if (ops[i].op == OP_PUSH || (ops[i].flags & (OPF_JMP|OPF_TAIL)))
+ break;
+ if (ops[i].op == OP_SUB && ops[i].operand[0].reg == xSP
+ && ops[i].operand[1].type == OPT_CONST)
+ {
+ g_stack_fsz = ops[i].operand[1].val;
+ ops[i].flags |= OPF_RMD | OPF_DONE;
+ esp_sub = 1;
+ break;
+ }
+ }
+
+ if (ecx_push && !esp_sub) {
+ // could actually be args for a call..
+ for (; i < opcnt; i++)
+ if (ops[i].op != OP_PUSH)
+ break;
+
+ if (ops[i].op == OP_CALL && ops[i].operand[0].type == OPT_LABEL) {
+ const struct parsed_proto *pp;
+ pp = proto_parse(g_fhdr, opr_name(&ops[i], 0), 1);
+ j = pp ? pp->argc_stack : 0;
+ while (i > 0 && j > 0) {
+ i--;
+ if (ops[i].op == OP_PUSH) {
+ ops[i].flags &= ~(OPF_RMD | OPF_DONE);
+ j--;
+ }
+ }
+ if (j != 0)
+ ferr(&ops[i], "unhandled prologue\n");
+
+ // recheck
+ i = g_stack_fsz = ecx_push = 0;
+ while (ops[i].op == OP_PUSH && IS(opr_name(&ops[i], 0), "ecx")) {
+ if (!(ops[i].flags & OPF_RMD))
+ break;
+ g_stack_fsz += 4;
+ ecx_push++;
+ i++;
+ }
+ }
+ }
+
+ found = 0;
+ if (ecx_push || esp_sub)
+ {
+ g_sp_frame = 1;
+
+ i++;
+ do {
+ for (; i < opcnt; i++)
+ if (ops[i].flags & OPF_TAIL)
+ break;
+ j = i - 1;
+ if (i == opcnt && (ops[j].flags & OPF_JMP)) {
+ if (ops[j].bt_i != -1 || ops[j].btj != NULL)
+ break;
+ i--;
+ j--;
+ }
+
+ if (ecx_push > 0) {
+ for (l = 0; l < ecx_push; l++) {
+ if (ops[j].op == OP_POP && IS(opr_name(&ops[j], 0), "ecx"))
+ /* pop ecx */;
+ else if (ops[j].op == OP_ADD
+ && IS(opr_name(&ops[j], 0), "esp")
+ && ops[j].operand[1].type == OPT_CONST)
+ {
+ /* add esp, N */
+ l += ops[j].operand[1].val / 4 - 1;
+ }
+ else
+ ferr(&ops[j], "'pop ecx' expected\n");
+
+ ops[j].flags |= OPF_RMD | OPF_DONE;
+ j--;
+ }
+ if (l != ecx_push)
+ ferr(&ops[j], "epilogue scan failed\n");
+
+ found = 1;
+ }
+
+ if (esp_sub) {
+ if (ops[j].op != OP_ADD
+ || !IS(opr_name(&ops[j], 0), "esp")
+ || ops[j].operand[1].type != OPT_CONST
+ || ops[j].operand[1].val != g_stack_fsz)
+ ferr(&ops[j], "'add esp' expected\n");
+
+ ops[j].flags |= OPF_RMD | OPF_DONE;
+ ops[j].operand[1].val = 0; // hack for stack arg scanner
+ found = 1;
+ }
+
+ i++;
+ } while (i < opcnt);
+
+ if (!found)
+ ferr(ops, "missing esp epilogue\n");
+ }
+}
+
+static const struct parsed_proto *resolve_icall(int i, int opcnt,
+ 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:
+ case OPT_LABEL:
+ case OPT_OFFSET:
+ 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,
+ pp_i, multi_src);
+ break;
+ }
+
+ return pp;
+}
+
+// find an instruction that changed opr before i op
+// *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,
+ int magic, int *op_i, int *is_caller)
+{
+ struct label_ref *lr;
+ int ret = 0;
+
+ 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_origin(lr->i, opr, magic, op_i, is_caller);
+ }
+ if (i > 0 && LAST_OP(i - 1))
+ return ret;
+ }
+
+ i--;
+ if (i < 0) {
+ if (is_caller != NULL)
+ *is_caller = 1;
+ return -1;
+ }
+
+ if (ops[i].cc_scratch == magic)
+ return ret;
+ ops[i].cc_scratch = magic;
+
+ if (!(ops[i].flags & OPF_DATA))
+ continue;
+ if (!is_opr_modified(opr, &ops[i]))
+ continue;
+
+ if (*op_i >= 0) {
+ if (*op_i == i)
+ return ret | 1;
+
+ // XXX: could check if the other op does the same
+ return -1;
+ }
+
+ *op_i = i;
+ return ret | 1;
+ }
+}
+
+// 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)
+{
+ struct label_ref *lr;
+ int ret = 0;
+
+ 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;
+}
+
+static struct parsed_proto *process_call_early(int i, int opcnt,
+ int *adj_i)
+{
+ struct parsed_op *po = &ops[i];
+ struct parsed_proto *pp;
+ int multipath = 0;
+ int adj = 0;
+ int ret;
+
+ pp = po->pp;
+ if (pp == NULL || pp->is_vararg || pp->argc_reg != 0)
+ // leave for later
+ return NULL;
+
+ // look for and make use of esp adjust
+ *adj_i = ret = -1;
+ if (!pp->is_stdcall && pp->argc_stack > 0)
+ ret = scan_for_esp_adjust(i + 1, opcnt,
+ pp->argc_stack * 4, &adj, &multipath, 0);
+ if (ret >= 0) {
+ if (pp->argc_stack > adj / 4)
+ return NULL;
+ if (multipath)
+ return NULL;
+ if (ops[ret].op == OP_POP && adj != 4)
+ return NULL;
+ }
+
+ *adj_i = ret;
+ return pp;
+}
+
+static struct parsed_proto *process_call(int i, int opcnt)
+{
+ struct parsed_op *po = &ops[i];
+ 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;
+
+ tmpname = opr_name(po, 0);
+ pp = po->pp;
+ if (pp == NULL)
+ {
+ // indirect call
+ 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);
+ pp = proto_clone(pp_c);
+ my_assert_not(pp, NULL);
+ if (multipath)
+ // not resolved just to single func
+ pp->is_fptr = 1;
+
+ switch (po->operand[0].type) {
+ 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;
+ break;
+ default:
+ break;
+ }
+ }
+ 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,
+ 32*4, &adj, &multipath, 0);
+ if (ret < 0 || adj < 0) {
+ if (!g_allow_regfunc)
+ ferr(po, "non-__cdecl indirect call unhandled yet\n");
+ pp->is_unresolved = 1;
+ adj = 0;
+ }
+ adj /= 4;
+ if (adj > ARRAY_SIZE(pp->arg))
+ ferr(po, "esp adjust too large: %d\n", adj);
+ pp->ret_type.name = strdup("int");
+ pp->argc = pp->argc_stack = adj;
+ for (arg = 0; arg < pp->argc; arg++)
+ pp->arg[arg].type.name = strdup("int");
+ }
+ po->pp = pp;
+ }
+
+ // look for and make use of esp adjust
+ multipath = 0;
+ ret = -1;
+ if (!pp->is_stdcall && pp->argc_stack > 0)
+ ret = scan_for_esp_adjust(i + 1, opcnt,
+ pp->argc_stack * 4, &adj, &multipath, 0);
+ if (ret >= 0) {
+ if (pp->is_vararg) {
+ if (adj / 4 < pp->argc_stack) {
+ fnote(po, "(this call)\n");
+ ferr(&ops[ret], "esp adjust is too small: %x < %x\n",
+ adj, pp->argc_stack * 4);
+ }
+ // modify pp to make it have varargs as normal args
+ arg = pp->argc;
+ pp->argc += adj / 4 - pp->argc_stack;
+ for (; arg < pp->argc; arg++) {
+ pp->arg[arg].type.name = strdup("int");
+ pp->argc_stack++;
+ }
+ if (pp->argc > ARRAY_SIZE(pp->arg))
+ ferr(po, "too many args for '%s'\n", tmpname);
+ }
+ if (pp->argc_stack > adj / 4) {
+ fnote(po, "(this call)\n");
+ ferr(&ops[ret], "stack tracking failed for '%s': %x %x\n",
+ tmpname, pp->argc_stack * 4, adj);
+ }
+
+ scan_for_esp_adjust(i + 1, opcnt,
+ pp->argc_stack * 4, &adj, &multipath, 1);
+ }
+ else if (pp->is_vararg)
+ ferr(po, "missing esp_adjust for vararg func '%s'\n",
+ pp->name);
+
+ return pp;
+}
+
+static int collect_call_args_early(struct parsed_op *po, int i,
+ struct parsed_proto *pp, int *regmask)
+{
+ int arg, ret;
+ int j;
+
+ for (arg = 0; arg < pp->argc; arg++)
+ if (pp->arg[arg].reg == NULL)
+ break;
+
+ // first see if it can be easily done
+ for (j = i; j > 0 && arg < pp->argc; )
+ {
+ if (g_labels[j] != NULL)
+ return -1;
+ j--;
+
+ if (ops[j].op == OP_CALL)
+ return -1;
+ else if (ops[j].op == OP_ADD && ops[j].operand[0].reg == xSP)
+ return -1;
+ else if (ops[j].op == OP_POP)
+ return -1;
+ else if (ops[j].flags & OPF_CJMP)
+ return -1;
+ else if (ops[j].op == OP_PUSH) {
+ if (ops[j].flags & (OPF_FARG|OPF_FARGNR))
+ return -1;
+ ret = scan_for_mod(&ops[j], j + 1, i, 1);
+ if (ret >= 0)
+ return -1;
+
+ if (pp->arg[arg].type.is_va_list)
+ return -1;
+
+ // next arg
+ for (arg++; arg < pp->argc; arg++)
+ if (pp->arg[arg].reg == NULL)
+ break;
+ }
+ }
+
+ if (arg < pp->argc)
+ return -1;
+
+ // now do it
+ for (arg = 0; arg < pp->argc; arg++)
+ if (pp->arg[arg].reg == NULL)
+ break;
+
+ for (j = i; j > 0 && arg < pp->argc; )
+ {
+ j--;
+
+ if (ops[j].op == OP_PUSH)
+ {
+ ops[j].p_argnext = -1;
+ ferr_assert(&ops[j], pp->arg[arg].datap == NULL);
+ pp->arg[arg].datap = &ops[j];
+
+ if (ops[j].operand[0].type == OPT_REG)
+ *regmask |= 1 << ops[j].operand[0].reg;
+
+ ops[j].flags |= OPF_RMD | OPF_DONE | OPF_FARGNR | OPF_FARG;
+ ops[j].flags &= ~OPF_RSAVE;
+
+ // next arg
+ for (arg++; arg < pp->argc; arg++)
+ if (pp->arg[arg].reg == NULL)
+ break;
+ }
+ }
+
+ return 0;
+}
+
+static int collect_call_args_r(struct parsed_op *po, int i,
+ struct parsed_proto *pp, int *regmask, int *save_arg_vars,
+ int *arg_grp, int arg, int magic, int need_op_saving, int may_reuse)
+{
+ struct parsed_proto *pp_tmp;
+ struct parsed_op *po_tmp;
+ struct label_ref *lr;
+ int need_to_save_current;
+ int arg_grp_current = 0;
+ int save_args_seen = 0;
+ int save_args;
+ int ret = 0;
+ int reg;
+ char buf[32];
+ int j, k;
+
+ if (i < 0) {
+ ferr(po, "dead label encountered\n");
+ return -1;
+ }
+
+ for (; arg < pp->argc; arg++)
+ if (pp->arg[arg].reg == NULL)
+ break;
+ magic = (magic & 0xffffff) | (arg << 24);
+
+ 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) {
+ ferr(&ops[j], "arg collect hit same path with diff args for %s\n",
+ pp->name);
+ return -1;
+ }
+ // ok: have already been here
+ return 0;
+ }
+ ops[j].cc_scratch = magic;
+
+ if (g_labels[j] != NULL && g_label_refs[j].i != -1) {
+ lr = &g_label_refs[j];
+ if (lr->next != NULL)
+ need_op_saving = 1;
for (; lr->next; lr = lr->next) {
check_i(&ops[j], lr->i);
if ((ops[lr->i].flags & (OPF_JMP|OPF_CJMP)) != OPF_JMP)
ferr(po, "arg collect %d/%d hit esp adjust of %d\n",
arg, pp->argc, ops[j].operand[1].val);
}
- else if (ops[j].op == OP_POP) {
+ else if (ops[j].op == OP_POP && !(ops[j].flags & OPF_DONE))
+ {
if (pp->is_unresolved)
break;
may_reuse = 1;
}
- else if (ops[j].op == OP_PUSH && !(ops[j].flags & OPF_FARG))
+ else if (ops[j].op == OP_PUSH
+ && !(ops[j].flags & (OPF_FARGNR|OPF_DONE)))
{
if (pp->is_unresolved && (ops[j].flags & OPF_RMD))
break;
// but that can't happen if we didn't branch, so they
// can be removed from future searches (handles nested calls)
if (!may_reuse)
- ops[j].flags |= OPF_FARG;
+ ops[j].flags |= OPF_FARGNR;
+ ops[j].flags |= OPF_FARG;
ops[j].flags &= ~OPF_RSAVE;
// check for __VALIST
if (!g_func_pp->is_vararg
|| strstr(ops[k].operand[1].name, buf))
{
- ops[k].flags |= OPF_RMD;
+ ops[k].flags |= OPF_RMD | OPF_DONE;
ops[j].flags |= OPF_RMD | OPF_VAPUSH;
save_args &= ~(1 << arg);
reg = -1;
ret = stack_frame_access(&ops[k], &ops[k].operand[1],
buf, sizeof(buf), ops[k].operand[1].name, "", 1, 0);
if (ret >= 0) {
- ops[k].flags |= OPF_RMD;
+ ops[k].flags |= OPF_RMD | OPF_DONE;
ops[j].flags |= OPF_RMD;
ops[j].p_argpass = ret + 1;
save_args &= ~(1 << arg);
save_arg_vars[arg_grp] |= save_arg_vars_current;
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;
-}
-
-// early check for tail call or branch back
-static int is_like_tailjmp(int j)
-{
- if (!(ops[j].flags & OPF_JMP))
- return 0;
-
- 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;
-
- return 0;
-}
-
-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;
-
- 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][0] && IS(g_labels[l], pd->d[j].u.label)) {
- add_label_ref(&g_label_refs[l], i);
- pd->d[j].bt_i = l;
- break;
- }
- }
+ 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 pd;
+ 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 output_std_flags(FILE *fout, struct parsed_op *po,
}
}
+enum {
+ OPP_FORCE_NORETURN = (1 << 0),
+ OPP_SIMPLE_ARGS = (1 << 1),
+ OPP_ALIGN = (1 << 2),
+};
+
static void output_pp_attrs(FILE *fout, const struct parsed_proto *pp,
- int is_noreturn)
+ int flags)
{
+ const char *cconv = "";
+
if (pp->is_fastcall)
- fprintf(fout, "__fastcall ");
+ cconv = "__fastcall ";
else if (pp->is_stdcall && pp->argc_reg == 0)
- fprintf(fout, "__stdcall ");
- if (pp->is_noreturn || is_noreturn)
+ cconv = "__stdcall ";
+
+ fprintf(fout, (flags & OPP_ALIGN) ? "%-16s" : "%s", cconv);
+
+ if (pp->is_noreturn || (flags & OPP_FORCE_NORETURN))
fprintf(fout, "noreturn ");
}
+static void output_pp(FILE *fout, const struct parsed_proto *pp,
+ int flags)
+{
+ int i;
+
+ fprintf(fout, (flags & OPP_ALIGN) ? "%-5s" : "%s ",
+ pp->ret_type.name);
+ if (pp->is_fptr)
+ fprintf(fout, "(");
+ output_pp_attrs(fout, pp, flags);
+ if (pp->is_fptr)
+ fprintf(fout, "*");
+ fprintf(fout, "%s", pp->name);
+ if (pp->is_fptr)
+ fprintf(fout, ")");
+
+ fprintf(fout, "(");
+ for (i = 0; i < pp->argc; i++) {
+ if (i > 0)
+ fprintf(fout, ", ");
+ if (pp->arg[i].fptr != NULL && !(flags & OPP_SIMPLE_ARGS)) {
+ // func pointer
+ output_pp(fout, pp->arg[i].fptr, 0);
+ }
+ else if (pp->arg[i].type.is_retreg) {
+ fprintf(fout, "u32 *r_%s", pp->arg[i].reg);
+ }
+ else {
+ fprintf(fout, "%s", pp->arg[i].type.name);
+ if (!pp->is_fptr)
+ fprintf(fout, " a%d", i + 1);
+ }
+ }
+ if (pp->is_vararg) {
+ if (i > 0)
+ fprintf(fout, ", ");
+ fprintf(fout, "...");
+ }
+ fprintf(fout, ")");
+}
+
+static int get_pp_arg_regmask(const struct parsed_proto *pp)
+{
+ int regmask = 0;
+ int i, reg;
+
+ for (i = 0; i < pp->argc; i++) {
+ if (pp->arg[i].reg != NULL) {
+ reg = char_array_i(regs_r32,
+ ARRAY_SIZE(regs_r32), pp->arg[i].reg);
+ if (reg < 0)
+ ferr(ops, "arg '%s' of func '%s' is not a reg?\n",
+ pp->arg[i].reg, pp->name);
+ regmask |= 1 << reg;
+ }
+ }
+
+ return regmask;
+}
+
static char *saved_arg_name(char *buf, size_t buf_size, int grp, int num)
{
char buf1[16];
return buf;
}
+static void gen_x_cleanup(int opcnt);
+
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;
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;
int no_output;
int i, j, l;
- int arg;
- int reg;
- int ret;
-
- g_bp_frame = g_sp_frame = g_stack_fsz = 0;
- g_stack_frame_used = 0;
-
- g_func_pp = proto_parse(fhdr, funcn, 0);
- if (g_func_pp == NULL)
- ferr(ops, "proto_parse failed for '%s'\n", funcn);
-
- for (i = 0; i < g_func_pp->argc; i++) {
- 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;
- }
- }
-
- // pass1:
- // - handle ebp/esp frame, remove ops related to it
- if (ops[0].op == OP_PUSH && IS(opr_name(&ops[0], 0), "ebp")
- && ops[1].op == OP_MOV
- && IS(opr_name(&ops[1], 0), "ebp")
- && IS(opr_name(&ops[1], 1), "esp"))
- {
- int ecx_push = 0;
-
- g_bp_frame = 1;
- ops[0].flags |= OPF_RMD;
- ops[1].flags |= OPF_RMD;
- i = 2;
-
- if (ops[2].op == OP_SUB && IS(opr_name(&ops[2], 0), "esp")) {
- g_stack_fsz = opr_const(&ops[2], 1);
- ops[2].flags |= OPF_RMD;
- i++;
- }
- else {
- // another way msvc builds stack frame..
- i = 2;
- while (ops[i].op == OP_PUSH && IS(opr_name(&ops[i], 0), "ecx")) {
- g_stack_fsz += 4;
- ops[i].flags |= OPF_RMD;
- ecx_push++;
- i++;
- }
- // and another way..
- if (i == 2 && ops[i].op == OP_MOV && ops[i].operand[0].reg == xAX
- && ops[i].operand[1].type == OPT_CONST
- && ops[i + 1].op == OP_CALL
- && IS(opr_name(&ops[i + 1], 0), "__alloca_probe"))
- {
- g_stack_fsz += ops[i].operand[1].val;
- ops[i].flags |= OPF_RMD;
- i++;
- ops[i].flags |= OPF_RMD;
- i++;
- }
- }
-
- found = 0;
- do {
- for (; i < opcnt; i++)
- if (ops[i].op == OP_RET)
- break;
- j = i - 1;
- if (i == opcnt && (ops[j].flags & OPF_JMP)) {
- if (found && is_like_tailjmp(j))
- break;
- j--;
- }
-
- if ((ops[j].op == OP_POP && IS(opr_name(&ops[j], 0), "ebp"))
- || ops[j].op == OP_LEAVE)
- {
- ops[j].flags |= OPF_RMD;
- }
- else if (!(g_ida_func_attr & IDAFA_NORETURN))
- ferr(&ops[j], "'pop ebp' expected\n");
-
- if (g_stack_fsz != 0) {
- if (ops[j - 1].op == OP_MOV
- && IS(opr_name(&ops[j - 1], 0), "esp")
- && IS(opr_name(&ops[j - 1], 1), "ebp"))
- {
- ops[j - 1].flags |= OPF_RMD;
- }
- else if (ops[j].op != OP_LEAVE
- && !(g_ida_func_attr & IDAFA_NORETURN))
- {
- ferr(&ops[j - 1], "esp restore expected\n");
- }
-
- if (ecx_push && ops[j - 2].op == OP_POP
- && IS(opr_name(&ops[j - 2], 0), "ecx"))
- {
- ferr(&ops[j - 2], "unexpected ecx pop\n");
- }
- }
-
- found = 1;
- i++;
- } while (i < opcnt);
- }
- else {
- int ecx_push = 0, esp_sub = 0;
-
- i = 0;
- while (ops[i].op == OP_PUSH && IS(opr_name(&ops[i], 0), "ecx")) {
- ops[i].flags |= OPF_RMD;
- g_stack_fsz += 4;
- ecx_push++;
- i++;
- }
-
- for (; i < opcnt; i++) {
- if (ops[i].op == OP_PUSH || (ops[i].flags & (OPF_JMP|OPF_TAIL)))
- break;
- if (ops[i].op == OP_SUB && ops[i].operand[0].reg == xSP
- && ops[i].operand[1].type == OPT_CONST)
- {
- g_stack_fsz = ops[i].operand[1].val;
- ops[i].flags |= OPF_RMD;
- esp_sub = 1;
- break;
- }
- }
-
- found = 0;
- if (ecx_push || esp_sub)
- {
- g_sp_frame = 1;
-
- i++;
- do {
- for (; i < opcnt; i++)
- if (ops[i].op == OP_RET)
- break;
- j = i - 1;
- if (i == opcnt && (ops[j].flags & OPF_JMP)) {
- if (found && is_like_tailjmp(j))
- break;
- j--;
- }
-
- if (ecx_push > 0) {
- for (l = 0; l < ecx_push; l++) {
- if (ops[j].op != OP_POP
- || !IS(opr_name(&ops[j], 0), "ecx"))
- {
- ferr(&ops[j], "'pop ecx' expected\n");
- }
- ops[j].flags |= OPF_RMD;
- j--;
- }
-
- found = 1;
- }
-
- if (esp_sub) {
- if (ops[j].op != OP_ADD
- || !IS(opr_name(&ops[j], 0), "esp")
- || ops[j].operand[1].type != OPT_CONST
- || ops[j].operand[1].val != g_stack_fsz)
- ferr(&ops[j], "'add esp' expected\n");
- ops[j].flags |= OPF_RMD;
-
- found = 1;
- }
-
- i++;
- } while (i < 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)
- 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][0] && IS(po->operand[0].name, g_labels[l])) {
- if (l == i + 1 && po->op == OP_JMP) {
- // yet another alignment type..
- po->flags |= OPF_RMD;
- break;
- }
- add_label_ref(&g_label_refs[l], i);
- po->bt_i = l;
- break;
- }
- }
+ int arg;
+ int reg;
+ int ret;
- if (po->bt_i != -1 || (po->flags & OPF_RMD))
- continue;
+ g_bp_frame = g_sp_frame = g_stack_fsz = 0;
+ g_stack_frame_used = 0;
- if (po->operand[0].type == OPT_LABEL)
- // assume tail call
- goto tailcall;
+ g_func_pp = proto_parse(fhdr, funcn, 0);
+ if (g_func_pp == NULL)
+ ferr(ops, "proto_parse failed for '%s'\n", funcn);
- ferr(po, "unhandled branch\n");
+ regmask_arg = get_pp_arg_regmask(g_func_pp);
-tailcall:
- po->op = OP_CALL;
- po->flags |= OPF_TAIL;
- if (i > 0 && ops[i - 1].op == OP_POP)
- po->flags |= OPF_ATAIL;
- i--; // reprocess
- }
+ // 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);
// pass3:
// - remove dead labels
- // - process calls
for (i = 0; i < opcnt; i++)
{
- if (g_labels[i][0] != 0 && g_label_refs[i].i == -1)
- g_labels[i][0] = 0;
+ 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)
+ if (po->flags & (OPF_RMD|OPF_DONE))
continue;
if (po->op == OP_CALL)
{
- tmpname = opr_name(po, 0);
- pp = po->pp;
- if (pp == NULL)
- {
- // indirect call
- pp_c = resolve_icall(i, opcnt, &l);
- if (pp_c != NULL) {
- if (!pp_c->is_func && !pp_c->is_fptr)
- ferr(po, "call to non-func: %s\n", pp_c->name);
- pp = proto_clone(pp_c);
- my_assert_not(pp, NULL);
- if (l)
- // not resolved just to single func
- pp->is_fptr = 1;
-
- switch (po->operand[0].type) {
- case OPT_REG:
- // we resolved this call and no longer need the register
- po->regmask_src &= ~(1 << po->operand[0].reg);
- break;
- case OPT_REGMEM:
- pp->is_fptr = 1;
- break;
- default:
- break;
- }
- }
- 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, &l);
- 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: %d\n", j);
- pp->ret_type.name = strdup("int");
- pp->argc = pp->argc_stack = j;
- for (arg = 0; arg < pp->argc; arg++)
- pp->arg[arg].type.name = strdup("int");
- }
- po->pp = pp;
- }
-
- // look for and make use of esp adjust
- ret = -1;
- if (!pp->is_stdcall && pp->argc_stack > 0)
- ret = scan_for_esp_adjust(i + 1, opcnt, &j, &l);
- if (ret >= 0) {
- if (pp->is_vararg) {
- if (j / 4 < pp->argc_stack)
- ferr(po, "esp adjust is too small: %x < %x\n",
- j, pp->argc_stack * 4);
- // modify pp to make it have varargs as normal args
- arg = pp->argc;
- pp->argc += j / 4 - pp->argc_stack;
- for (; arg < pp->argc; arg++) {
- pp->arg[arg].type.name = strdup("int");
- pp->argc_stack++;
- }
- if (pp->argc > ARRAY_SIZE(pp->arg))
- ferr(po, "too many args for '%s'\n", tmpname);
- }
- if (pp->argc_stack != j / 4)
- ferr(po, "stack tracking failed for '%s': %x %x\n",
- tmpname, pp->argc_stack * 4, j);
-
- ops[ret].flags |= OPF_RMD;
- if (ops[ret].op == OP_POP && j > 4) {
- // deal with multi-pop stack adjust
- j = pp->argc_stack;
- while (ops[ret].op == OP_POP && j > 0 && ret < opcnt) {
- ops[ret].flags |= OPF_RMD;
- j--;
- ret++;
- }
- }
- else if (!l) {
- // a bit of a hack, but deals with use of
- // single adj for multiple calls
- ops[ret].operand[1].val -= j;
+ pp = process_call_early(i, opcnt, &j);
+ if (pp != NULL) {
+ if (!(po->flags & OPF_ATAIL))
+ // since we know the args, try to collect them
+ if (collect_call_args_early(po, i, pp, ®mask) != 0)
+ pp = NULL;
+ }
+
+ if (pp != NULL) {
+ if (j >= 0) {
+ // commit esp adjust
+ ops[j].flags |= OPF_RMD;
+ if (ops[j].op != OP_POP)
+ patch_esp_adjust(&ops[j], pp->argc_stack * 4);
+ else
+ ops[j].flags |= OPF_DONE;
}
+
+ if (strstr(pp->ret_type.name, "int64"))
+ need_tmp64 = 1;
+
+ po->flags |= OPF_DONE;
}
- else if (pp->is_vararg)
- ferr(po, "missing esp_adjust for vararg func '%s'\n",
- pp->name);
+ }
+ }
+
+ // pass5:
+ // - process calls
+ // - handle push <const>/pop pairs
+ for (i = 0; i < opcnt; i++)
+ {
+ po = &ops[i];
+ if (po->flags & (OPF_RMD|OPF_DONE))
+ continue;
+
+ if (po->op == OP_CALL && !(po->flags & OPF_DONE))
+ {
+ pp = process_call(i, opcnt);
if (!pp->is_unresolved && !(po->flags & OPF_ATAIL)) {
// since we know the args, collect them
if (strstr(pp->ret_type.name, "int64"))
need_tmp64 = 1;
}
+ 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, ®mask_pp);
}
- // pass4:
+ // pass6:
// - find POPs for PUSHes, rm both
// - scan for STD/CLD, propagate DF
// - scan for all used registers
// - find flag set ops for their users
// - do unreselved calls
// - declare indirect functions
- for (i = 0; i < opcnt; i++) {
+ for (i = 0; i < opcnt; i++)
+ {
po = &ops[i];
- if (po->flags & OPF_RMD)
+ if (po->flags & (OPF_RMD|OPF_DONE))
continue;
if (po->op == OP_PUSH && (po->flags & OPF_RSAVE)) {
regmask_save |= 1 << reg;
}
- if (po->op == OP_PUSH && po->p_argnum == 0
+ if (po->op == OP_PUSH && !(po->flags & OPF_FARG)
&& !(po->flags & OPF_RSAVE) && !g_func_pp->is_userstack)
{
if (po->operand[0].type == OPT_REG)
continue;
}
}
- else if (po->operand[0].type == OPT_CONST) {
- 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][0] != 0)
- {
- break;
- }
-
- if (!(ops[j].flags & OPF_RMD) && ops[j].op == OP_POP)
- {
- po->flags |= OPF_RMD;
- ops[j].datap = po;
- break;
- }
- }
- }
}
if (po->op == OP_STD) {
- po->flags |= OPF_DF | OPF_RMD;
+ po->flags |= OPF_DF | OPF_RMD | OPF_DONE;
scan_propagate_df(i + 1, opcnt);
}
need_tmp64 = 1;
}
else if (po->op == OP_CALL) {
+ // note: resolved non-reg calls are OPF_DONE already
pp = po->pp;
if (pp == NULL)
ferr(po, "NULL pp\n");
need_tmp64 = 1;
}
else if (po->op == OP_CLD)
- po->flags |= OPF_RMD;
+ po->flags |= OPF_RMD | OPF_DONE;
if (po->op == OP_RCL || po->op == OP_RCR || po->op == OP_XCHG) {
need_tmp_var = 1;
}
}
- // pass4:
+ // pass7:
// - confirm regmask_save, it might have been reduced
if (regmask_save != 0)
{
}
// the function itself
- fprintf(fout, "%s ", g_func_pp->ret_type.name);
- output_pp_attrs(fout, g_func_pp, g_ida_func_attr & IDAFA_NORETURN);
- fprintf(fout, "%s(", g_func_pp->name);
-
- for (i = 0; i < g_func_pp->argc; i++) {
- if (i > 0)
- fprintf(fout, ", ");
- if (g_func_pp->arg[i].fptr != NULL) {
- // func pointer..
- pp = g_func_pp->arg[i].fptr;
- fprintf(fout, "%s (", pp->ret_type.name);
- output_pp_attrs(fout, pp, 0);
- fprintf(fout, "*a%d)(", i + 1);
- for (j = 0; j < pp->argc; j++) {
- if (j > 0)
- fprintf(fout, ", ");
- if (pp->arg[j].fptr)
- ferr(ops, "nested fptr\n");
- fprintf(fout, "%s", pp->arg[j].type.name);
- }
- if (pp->is_vararg) {
- if (j > 0)
- fprintf(fout, ", ");
- fprintf(fout, "...");
- }
- fprintf(fout, ")");
- }
- else if (g_func_pp->arg[i].type.is_retreg) {
- fprintf(fout, "u32 *r_%s", g_func_pp->arg[i].reg);
- }
- else {
- fprintf(fout, "%s a%d", g_func_pp->arg[i].type.name, i + 1);
- }
- }
- if (g_func_pp->is_vararg) {
- if (i > 0)
- fprintf(fout, ", ");
- fprintf(fout, "...");
- }
-
- fprintf(fout, ")\n{\n");
+ ferr_assert(ops, !g_func_pp->is_fptr);
+ output_pp(fout, g_func_pp,
+ (g_ida_func_attr & IDAFA_NORETURN) ? OPP_FORCE_NORETURN : 0);
+ fprintf(fout, "\n{\n");
// declare indirect functions
for (i = 0; i < opcnt; i++) {
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, " ");
+ output_pp(fout, pp, OPP_SIMPLE_ARGS);
+ fprintf(fout, ";\n");
}
}
}
}
}
+ // declare normal registers
regmask_now = regmask & ~regmask_arg;
regmask_now &= ~(1 << xSP);
if (regmask_now & 0x00ff) {
}
}
+ // 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)) {
// output ops
for (i = 0; i < opcnt; i++)
{
- if (g_labels[i][0] != 0) {
+ if (g_labels[i] != NULL) {
fprintf(fout, "\n%s:\n", g_labels[i]);
label_pending = 1;
// 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
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;
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],
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
fprintf(fout, "}\n\n");
- // cleanup
+ gen_x_cleanup(opcnt);
+}
+
+static void gen_x_cleanup(int opcnt)
+{
+ int i;
+
for (i = 0; i < opcnt; i++) {
struct label_ref *lr, *lr_del;
int id;
int argc_stack;
int regmask_dep;
- int has_ret:3; // -1, 0, 1: unresolved, no, yes
+ int has_ret:3; // -1, 0, 1: unresolved, no, yes
unsigned int dep_resolved:1;
unsigned int is_stdcall:1;
struct func_proto_dep *dep_func;
int dep_func_cnt;
+ const struct parsed_proto *pp; // seed pp, if any
};
struct func_proto_dep {
static struct func_prototype *hg_fp;
static int hg_fp_cnt;
+static struct scanned_var {
+ char name[NAMELEN];
+ enum opr_lenmod lmod;
+ unsigned int is_seeded:1;
+ unsigned int is_c_str:1;
+ const struct parsed_proto *pp; // seed pp, if any
+} *hg_vars;
+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)
{
}
#endif
-static void gen_hdr(const char *funcn, int opcnt)
+// recursive register dep pass
+// - track saved regs (part 2)
+// - try to figure out arg-regs
+// - calculate reg deps
+static void gen_hdr_dep_pass(int i, int opcnt, unsigned char *cbits,
+ struct func_prototype *fp, int regmask_save, int regmask_dst,
+ int *regmask_dep, int *has_ret)
{
- struct func_prototype *fp;
struct func_proto_dep *dep;
- struct parsed_data *pd;
struct parsed_op *po;
- int regmask_save = 0;
- int regmask_dst = 0;
- int regmask_dep = 0;
- int max_bp_offset = 0;
- int has_ret = -1;
int from_caller = 0;
- int i, j, l, ret;
- int depth, reg;
-
- 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;
- hg_fp_cnt++;
+ int depth;
+ int j, l;
+ int reg;
+ int ret;
- // pass1:
- // - collect calls
- // - resolve all branches
- for (i = 0; i < opcnt; i++)
+ for (; i < opcnt; i++)
{
- po = &ops[i];
- po->bt_i = -1;
- po->btj = NULL;
-
- if (po->flags & OPF_RMD)
- continue;
-
- if (po->op == OP_CALL) {
- if (po->operand[0].type == OPT_LABEL)
- hg_fp_add_dep(fp, opr_name(po, 0));
-
- 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;
+ if (cbits[i >> 3] & (1 << (i & 7)))
+ return;
+ cbits[i >> 3] |= (1 << (i & 7));
- po->btj = pd;
- continue;
- }
+ po = &ops[i];
- for (l = 0; l < opcnt; l++) {
- if (g_labels[l][0] && IS(po->operand[0].name, g_labels[l])) {
- add_label_ref(&g_label_refs[l], i);
- po->bt_i = l;
- break;
+ 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);
+ 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 != -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
- }
-
- // pass2:
- // - remove dead labels
- for (i = 0; i < opcnt; i++)
- {
- if (g_labels[i][0] != 0 && g_label_refs[i].i == -1)
- g_labels[i][0] = 0;
- }
- // pass3:
- // - track saved regs
- // - try to figure out arg-regs
- for (i = 0; i < opcnt; i++)
- {
- po = &ops[i];
- if (po->flags & OPF_RMD)
+ 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);
+ }
+ else {
+ i = po->bt_i - 1;
+ }
continue;
+ }
- if (po->op == OP_PUSH && po->operand[0].type == OPT_REG)
+ if (po->flags & OPF_FARG)
+ /* (just calculate register deps) */;
+ else if (po->op == OP_PUSH && po->operand[0].type == OPT_REG)
{
reg = po->operand[0].reg;
if (reg < 0)
ferr(po, "reg not set for push?\n");
- depth = 0;
- ret = scan_for_pop(i + 1, opcnt,
- po->operand[0].name, i + opcnt * 1, 0, &depth, 0);
- if (ret == 1) {
+ if (po->flags & OPF_RSAVE) {
regmask_save |= 1 << reg;
- po->flags |= OPF_RMD;
- scan_for_pop(i + 1, opcnt,
- po->operand[0].name, i + opcnt * 2, 0, &depth, 1);
continue;
}
- ret = scan_for_pop_ret(i + 1, opcnt, po->operand[0].name, 0);
- if (ret == 0) {
+ if (po->flags & OPF_DONE)
+ continue;
+
+ depth = 0;
+ ret = scan_for_pop(i + 1, opcnt,
+ po->operand[0].name, i + opcnt * 2, 0, &depth, 0);
+ if (ret == 1) {
regmask_save |= 1 << reg;
po->flags |= OPF_RMD;
- scan_for_pop_ret(i + 1, opcnt, po->operand[0].name, OPF_RMD);
+ scan_for_pop(i + 1, opcnt,
+ po->operand[0].name, i + opcnt * 3, 0, &depth, 1);
continue;
}
}
- else if (po->op == OP_PUSH && po->operand[0].type == OPT_CONST) {
- 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][0] != 0)
- {
- break;
- }
-
- if (!(ops[j].flags & OPF_RMD) && ops[j].op == OP_POP)
- {
- po->flags |= OPF_RMD;
- ops[j].datap = po;
- break;
- }
- }
+ else if (po->flags & OPF_RMD)
continue;
- }
else if (po->op == OP_CALL) {
po->regmask_dst |= 1 << xAX;
dep->regmask_live = regmask_save | regmask_dst;
}
else if (po->op == OP_RET) {
- if (po->operand_cnt > 0)
+ if (po->operand_cnt > 0) {
fp->is_stdcall = 1;
+ if (fp->argc_stack >= 0
+ && fp->argc_stack != po->operand[0].val / 4)
+ ferr(po, "ret mismatch? (%d)\n", fp->argc_stack * 4);
+ fp->argc_stack = po->operand[0].val / 4;
+ }
}
- if (has_ret != 0 && (po->flags & OPF_TAIL)) {
+ // 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;
ret = 1;
opr.reg = xAX;
j = -1;
from_caller = 0;
- ret = resolve_origin(i, &opr, i + opcnt * 3, &j, &from_caller);
+ 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;
+ *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
- has_ret = 1;
+ *has_ret = 1;
}
else
- has_ret = 1;
+ *has_ret = 1;
}
}
- l = po->regmask_src & ~(regmask_save | regmask_dst);
+ l = regmask_save | regmask_dst;
+ if (g_bp_frame && !(po->flags & OPF_EBP_S))
+ l |= 1 << xBP;
+
+ l = po->regmask_src & ~l;
#if 0
if (l)
- fnote(po, "dep |= %04x, dst %04x, save %04x\n", l,
- regmask_dst, regmask_save);
+ fnote(po, "dep |= %04x, dst %04x, save %04x (f %x)\n",
+ l, regmask_dst, regmask_save, po->flags);
#endif
- regmask_dep |= l;
+ *regmask_dep |= l;
regmask_dst |= po->regmask_dst;
+
+ if (po->flags & OPF_TAIL)
+ return;
}
+}
- if (has_ret == -1 && (regmask_dep & (1 << xAX)))
- has_ret = 1;
+static void gen_hdr(const char *funcn, int opcnt)
+{
+ int save_arg_vars[MAX_ARG_GRP] = { 0, };
+ unsigned char cbits[MAX_OPS / 8];
+ const struct parsed_proto *pp_c;
+ struct parsed_proto *pp;
+ struct func_prototype *fp;
+ struct parsed_op *po;
+ int regmask_dummy = 0;
+ int regmask_dep;
+ int max_bp_offset = 0;
+ int has_ret;
+ int i, j, ret;
- for (i = 0; i < g_eqcnt; i++)
+ 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);
+
+ // pass3:
+ // - remove dead labels
+ // - collect 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;
+ }
+
+ po = &ops[i];
+ if (po->flags & (OPF_RMD|OPF_DONE))
+ continue;
+
+ if (po->op == OP_CALL) {
+ 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);
+ }
+ }
+
+ // pass4:
+ // - remove dead labels
+ // - handle push <const>/pop pairs
+ 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;
+ }
+
+ po = &ops[i];
+ if (po->flags & (OPF_RMD|OPF_DONE))
+ continue;
+
+ if (po->op == OP_PUSH && po->operand[0].type == OPT_CONST)
+ scan_for_pop_const(i, opcnt, ®mask_dummy);
+ }
+
+ // pass5:
+ // - process trivial calls
+ for (i = 0; i < opcnt; i++)
+ {
+ po = &ops[i];
+ if (po->flags & (OPF_RMD|OPF_DONE))
+ continue;
+
+ if (po->op == OP_CALL)
+ {
+ pp = process_call_early(i, opcnt, &j);
+ if (pp != NULL) {
+ if (!(po->flags & OPF_ATAIL))
+ // since we know the args, try to collect them
+ if (collect_call_args_early(po, i, pp, ®mask_dummy) != 0)
+ pp = NULL;
+ }
+
+ if (pp != NULL) {
+ if (j >= 0) {
+ // commit esp adjust
+ ops[j].flags |= OPF_RMD;
+ if (ops[j].op != OP_POP)
+ patch_esp_adjust(&ops[j], pp->argc_stack * 4);
+ else
+ ops[j].flags |= OPF_DONE;
+ }
+
+ po->flags |= OPF_DONE;
+ }
+ }
+ }
+
+ // pass6:
+ // - track saved regs (simple)
+ // - process calls
+ for (i = 0; i < opcnt; i++)
+ {
+ po = &ops[i];
+ if (po->flags & (OPF_RMD|OPF_DONE))
+ continue;
+
+ if (po->op == OP_PUSH && po->operand[0].type == OPT_REG)
+ {
+ ret = scan_for_pop_ret(i + 1, opcnt, po->operand[0].name, 0);
+ if (ret == 0) {
+ // regmask_save |= 1 << po->operand[0].reg; // do it later
+ po->flags |= OPF_RSAVE | OPF_RMD | OPF_DONE;
+ scan_for_pop_ret(i + 1, opcnt, po->operand[0].name, OPF_RMD);
+ }
+ }
+ else if (po->op == OP_CALL && !(po->flags & OPF_DONE))
+ {
+ pp = process_call(i, opcnt);
+
+ if (!pp->is_unresolved && !(po->flags & OPF_ATAIL)) {
+ // since we know the args, collect them
+ ret = collect_call_args(po, i, pp, ®mask_dummy, save_arg_vars,
+ i + opcnt * 1);
+ }
+ }
+ }
+
+ // pass7
+ memset(cbits, 0, sizeof(cbits));
+ regmask_dep = 0;
+ has_ret = -1;
+
+ gen_hdr_dep_pass(0, opcnt, cbits, fp, 0, 0, ®mask_dep, &has_ret);
+
+ // find unreachable code - must be fixed in IDA
+ for (i = 0; i < opcnt; i++)
+ {
+ if (cbits[i >> 3] & (1 << (i & 7)))
+ continue;
+
+ if (ops[i].op != OP_NOP)
+ ferr(&ops[i], "unreachable code\n");
+ }
+
+ 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;
+ }
- if (max_bp_offset > 0) {
+ if (fp->argc_stack < 0) {
max_bp_offset = (max_bp_offset + 3) & ~3;
- fp->argc_stack = max_bp_offset / 4 - 1;
- if (!(g_ida_func_attr & IDAFA_BP_FRAME))
+ fp->argc_stack = max_bp_offset / 4;
+ if ((g_ida_func_attr & IDAFA_BP_FRAME) && fp->argc_stack > 0)
fp->argc_stack--;
}
fp->regmask_dep = regmask_dep & ~(1 << xSP);
fp->has_ret = has_ret;
+#if 0
+ printf("// has_ret %d, regmask_dep %x\n",
+ fp->has_ret, fp->regmask_dep);
+ output_hdr_fp(stdout, fp, 1);
+ if (IS(funcn, "sub_10007F72")) exit(1);
+#endif
+
+ gen_x_cleanup(opcnt);
}
static void hg_fp_resolve_deps(struct func_prototype *fp)
{
struct func_prototype fp_s;
+ int dep;
int i;
// this thing is recursive, so mark first..
if (!fp->dep_func[i].proto->dep_resolved)
hg_fp_resolve_deps(fp->dep_func[i].proto);
- fp->regmask_dep |= ~fp->dep_func[i].regmask_live
- & fp->dep_func[i].proto->regmask_dep;
+ dep = ~fp->dep_func[i].regmask_live
+ & fp->dep_func[i].proto->regmask_dep;
+ fp->regmask_dep |= dep;
+ // 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;
}
}
}
-static void output_hdr(FILE *fout)
+static void output_hdr_fp(FILE *fout, const struct func_prototype *fp,
+ int count)
{
- struct func_prototype *fp;
- int had_usercall = 0;
+ const struct parsed_proto *pp;
+ char *p, namebuf[NAMELEN];
+ const char *name;
int regmask_dep;
int argc_stack;
- int i, j, arg;
-
- // resolve deps
- qsort(hg_fp, hg_fp_cnt, sizeof(hg_fp[0]), hg_fp_cmp_name);
- for (i = 0; i < hg_fp_cnt; i++)
- hg_fp_resolve_deps(&hg_fp[i]);
+ int j, arg;
- // note: messes up .proto ptr, don't use
- //qsort(hg_fp, hg_fp_cnt, sizeof(hg_fp[0]), hg_fp_cmp_id);
-
- for (i = 0; i < hg_fp_cnt; i++) {
- fp = &hg_fp[i];
+ for (; count > 0; count--, fp++) {
if (fp->has_ret == -1)
fprintf(fout, "// ret unresolved\n");
#if 0
fprintf(fout, "\n");
#endif
+ p = strchr(fp->name, '@');
+ if (p != NULL) {
+ memcpy(namebuf, fp->name, p - fp->name);
+ namebuf[p - fp->name] = 0;
+ name = namebuf;
+ }
+ else
+ name = fp->name;
+ if (name[0] == '_')
+ name++;
+
+ pp = proto_parse(g_fhdr, name, 1);
+ if (pp != NULL && pp->is_include)
+ continue;
+
+ if (fp->pp != NULL) {
+ // part of seed, output later
+ continue;
+ }
+
regmask_dep = fp->regmask_dep;
argc_stack = fp->argc_stack;
- fprintf(fout, fp->has_ret ? "int " : "void ");
+ fprintf(fout, "%-5s", fp->pp ? fp->pp->ret_type.name :
+ (fp->has_ret ? "int" : "void"));
if (regmask_dep && (fp->is_stdcall || argc_stack == 0)
&& (regmask_dep & ~((1 << xCX) | (1 << xDX))) == 0)
{
- fprintf(fout, "__fastcall ");
- if (had_usercall)
- fprintf(fout, " "); // align
+ fprintf(fout, " __fastcall ");
if (!(regmask_dep & (1 << xDX)) && argc_stack == 0)
argc_stack = 1;
else
}
else if (regmask_dep && !fp->is_stdcall) {
fprintf(fout, "/*__usercall*/ ");
- had_usercall = 1;
}
else if (regmask_dep) {
fprintf(fout, "/*__userpurge*/ ");
- had_usercall = 1;
}
else if (fp->is_stdcall)
- fprintf(fout, "__stdcall ");
+ fprintf(fout, " __stdcall ");
else
- fprintf(fout, "__cdecl ");
+ fprintf(fout, " __cdecl ");
- fprintf(fout, "%s(", fp->name);
+ fprintf(fout, "%s(", name);
arg = 0;
for (j = 0; j < xSP; j++) {
arg++;
if (arg != 1)
fprintf(fout, ", ");
- fprintf(fout, "int a%d/*<%s>*/", arg, regs_r32[j]);
+ if (fp->pp != NULL)
+ fprintf(fout, "%s", fp->pp->arg[arg - 1].type.name);
+ else
+ fprintf(fout, "int");
+ fprintf(fout, " a%d/*<%s>*/", arg, regs_r32[j]);
}
}
arg++;
if (arg != 1)
fprintf(fout, ", ");
- fprintf(fout, "int a%d", arg);
+ if (fp->pp != NULL) {
+ fprintf(fout, "%s", fp->pp->arg[arg - 1].type.name);
+ if (!fp->pp->arg[arg - 1].type.is_ptr)
+ fprintf(fout, " ");
+ }
+ else
+ fprintf(fout, "int ");
+ fprintf(fout, "a%d", arg);
}
fprintf(fout, ");\n");
}
}
+static void output_hdr(FILE *fout)
+{
+ static const char *lmod_c_names[] = {
+ [OPLM_UNSPEC] = "???",
+ [OPLM_BYTE] = "uint8_t",
+ [OPLM_WORD] = "uint16_t",
+ [OPLM_DWORD] = "uint32_t",
+ [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++)
+ hg_fp_resolve_deps(&hg_fp[i]);
+
+ // note: messes up .proto ptr, don't use
+ //qsort(hg_fp, hg_fp_cnt, sizeof(hg_fp[0]), hg_fp_cmp_id);
+
+ // output variables
+ for (i = 0; i < hg_var_cnt; i++) {
+ var = &hg_vars[i];
+
+ if (var->pp != NULL)
+ // part of seed
+ continue;
+ else if (var->is_c_str)
+ fprintf(fout, "extern %-8s %s[];", "char", var->name);
+ else
+ fprintf(fout, "extern %-8s %s;",
+ lmod_c_names[var->lmod], var->name);
+
+ if (var->is_seeded)
+ fprintf(fout, " // seeded");
+ fprintf(fout, "\n");
+ }
+
+ fprintf(fout, "\n");
+
+ // output function prototypes
+ output_hdr_fp(fout, hg_fp, hg_fp_cnt);
+
+ // seed passthrough
+ fprintf(fout, "\n// - seed -\n");
+
+ rewind(g_fhdr);
+ while (fgets(line, sizeof(line), g_fhdr))
+ fwrite(line, 1, strlen(line), fout);
+}
+
+// '=' needs special treatment
+// also ' quote
+static char *next_word_s(char *w, size_t wsize, char *s)
+{
+ size_t i;
+
+ s = sskip(s);
+
+ i = 0;
+ if (*s == '\'') {
+ w[0] = s[0];
+ for (i = 1; i < wsize - 1; i++) {
+ if (s[i] == 0) {
+ printf("warning: missing closing quote: \"%s\"\n", s);
+ break;
+ }
+ if (s[i] == '\'')
+ break;
+ w[i] = s[i];
+ }
+ }
+
+ for (; i < wsize - 1; i++) {
+ if (s[i] == 0 || my_isblank(s[i]) || (s[i] == '=' && i > 0))
+ break;
+ w[i] = s[i];
+ }
+ w[i] = 0;
+
+ if (s[i] != 0 && !my_isblank(s[i]) && s[i] != '=')
+ printf("warning: '%s' truncated\n", w);
+
+ return s + i;
+}
+
+static void scan_variables(FILE *fasm)
+{
+ struct scanned_var *var;
+ char line[256] = { 0, };
+ char words[3][256];
+ char *p = NULL;
+ int wordc;
+ int l;
+
+ while (!feof(fasm))
+ {
+ // skip to next data section
+ while (my_fgets(line, sizeof(line), fasm))
+ {
+ asmln++;
+
+ p = sskip(line);
+ if (*p == 0 || *p == ';')
+ continue;
+
+ p = sskip(next_word_s(words[0], sizeof(words[0]), p));
+ if (*p == 0 || *p == ';')
+ continue;
+
+ if (*p != 's' || !IS_START(p, "segment para public"))
+ continue;
+
+ break;
+ }
+
+ if (p == NULL || !IS_START(p, "segment para public"))
+ break;
+ p = sskip(p + 19);
+
+ if (!IS_START(p, "'DATA'"))
+ continue;
+
+ // now process it
+ while (my_fgets(line, sizeof(line), fasm))
+ {
+ asmln++;
+
+ p = line;
+ if (my_isblank(*p))
+ continue;
+
+ p = sskip(p);
+ if (*p == 0 || *p == ';')
+ continue;
+
+ for (wordc = 0; wordc < ARRAY_SIZE(words); wordc++) {
+ words[wordc][0] = 0;
+ p = sskip(next_word_s(words[wordc], sizeof(words[0]), p));
+ if (*p == 0 || *p == ';') {
+ wordc++;
+ break;
+ }
+ }
+
+ if (wordc == 2 && IS(words[1], "ends"))
+ break;
+ 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));
+ my_assert_not(hg_vars, NULL);
+ memset(hg_vars + hg_var_cnt, 0, sizeof(hg_vars[0]) * 0x100);
+ }
+
+ var = &hg_vars[hg_var_cnt++];
+ snprintf(var->name, sizeof(var->name), "%s", words[0]);
+
+ // maybe already in seed header?
+ var->pp = proto_parse(g_fhdr, var->name, 1);
+ if (var->pp != NULL) {
+ if (var->pp->is_fptr) {
+ var->lmod = OPLM_DWORD;
+ //var->is_ptr = 1;
+ }
+ else if (var->pp->is_func)
+ aerr("func?\n");
+ else if (!guess_lmod_from_c_type(&var->lmod, &var->pp->type))
+ aerr("unhandled C type '%s' for '%s'\n",
+ var->pp->type.name, var->name);
+
+ var->is_seeded = 1;
+ continue;
+ }
+
+ if (IS(words[1], "dd"))
+ var->lmod = OPLM_DWORD;
+ else if (IS(words[1], "dw"))
+ var->lmod = OPLM_WORD;
+ else if (IS(words[1], "db")) {
+ var->lmod = OPLM_BYTE;
+ if (wordc >= 3 && (l = strlen(words[2])) > 4) {
+ if (words[2][0] == '\'' && IS(words[2] + l - 2, ",0"))
+ var->is_c_str = 1;
+ }
+ }
+ else if (IS(words[1], "dq"))
+ var->lmod = OPLM_QWORD;
+ //else if (IS(words[1], "dt"))
+ else
+ aerr("type '%s' not known\n", words[1]);
+ }
+ }
+
+ rewind(fasm);
+ asmln = 0;
+}
+
static void set_label(int i, const char *name)
{
const char *p;
if (p != NULL)
len = p - name;
- if (len > sizeof(g_labels[0]) - 1)
- aerr("label '%s' too long: %d\n", name, len);
- if (g_labels[i][0] != 0 && !IS_START(g_labels[i], "algn_"))
+ if (g_labels[i] != NULL && !IS_START(g_labels[i], "algn_"))
aerr("dupe label '%s' vs '%s'?\n", name, g_labels[i]);
+ g_labels[i] = realloc(g_labels[i], len + 1);
+ my_assert_not(g_labels[i], NULL);
memcpy(g_labels[i], name, len);
g_labels[i][len] = 0;
}
-// '=' needs special treatment..
-static char *next_word_s(char *w, size_t wsize, char *s)
-{
- size_t i;
-
- s = sskip(s);
-
- for (i = 0; i < wsize - 1; i++) {
- if (s[i] == 0 || my_isblank(s[i]) || (s[i] == '=' && i > 0))
- break;
- w[i] = s[i];
- }
- w[i] = 0;
-
- if (s[i] != 0 && !my_isblank(s[i]) && s[i] != '=')
- printf("warning: '%s' truncated\n", w);
-
- return s + i;
-}
-
struct chunk_item {
char *name;
long fptr;
oldpos = ftell(fasm);
oldasmln = asmln;
- while (fgets(line, sizeof(line), fasm))
+ while (my_fgets(line, sizeof(line), fasm))
{
wordc = 0;
asmln++;
int pending_endp = 0;
int skip_func = 0;
int skip_warned = 0;
- int header_mode = 0;
int eq_alloc;
int verbose = 0;
int multi_seg = 0;
else if (IS(argv[arg], "-m"))
multi_seg = 1;
else if (IS(argv[arg], "-hdr"))
- header_mode = g_quiet_pp = 1;
+ g_header_mode = g_quiet_pp = g_allow_regfunc = 1;
else
break;
}
if (argc < arg + 3) {
- printf("usage:\n%s [-v] [-rf] [-m] <.c> <.asm> <hdrf> [rlist]*\n"
- "%s -hdr <.h> <.asm> <seed_hdrf> [rlist]*\n",
+ printf("usage:\n%s [-v] [-rf] [-m] <.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"
+ " -m - allow multiple .text sections\n"
+ "[rlist] is a file with function names to skip,"
+ " one per line\n",
argv[0], argv[0]);
return 1;
}
frlist = fopen(argv[arg], "r");
my_assert_not(frlist, NULL);
- while (fgets(line, sizeof(line), frlist)) {
+ while (my_fgets(line, sizeof(line), frlist)) {
p = sskip(line);
if (*p == 0 || *p == ';')
continue;
g_label_refs[i].next = NULL;
}
- while (fgets(line, sizeof(line), fasm))
+ if (g_header_mode)
+ scan_variables(fasm);
+
+ while (my_fgets(line, sizeof(line), fasm))
{
wordc = 0;
asmln++;
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)))
{
continue;
}
- if (in_func && !skip_func) {
- if (header_mode)
+ if (in_func && !g_skip_func) {
+ if (g_header_mode)
gen_hdr(g_func, pi);
else
gen_func(fout, g_fhdr, g_func, pi);
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;
if (pi != 0) {
memset(&ops, 0, pi * sizeof(ops[0]));
- memset(g_labels, 0, pi * sizeof(g_labels[0]));
+ clear_labels(pi);
pi = 0;
}
g_eqcnt = 0;
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;
&& 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 };
}
// scan for next text segment
- while (fgets(line, sizeof(line), fasm)) {
+ while (my_fgets(line, sizeof(line), fasm)) {
asmln++;
p = sskip(line);
if (*p == 0 || *p == ';')
continue;
}
- if (!in_func || skip_func) {
- if (!skip_warned && !skip_func && g_labels[pi][0] != 0) {
+ 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;
}
- g_labels[pi][0] = 0;
+ free(g_labels[pi]);
+ g_labels[pi] = NULL;
continue;
}
pi++;
}
- if (header_mode)
+ if (g_header_mode)
output_hdr(fout);
fclose(fout);
notaz.gp2x.de / ia32rtools.git / blobdiff