T0b1/fadec
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

decode: Encode trie node kind in 2 bits

Browse Source
This commit is contained in:
Alexis Engelke
2023-03-22 10:46:35 +01:00
parent e1084be859
commit dac2ff1987
2 changed files with 59 additions and 50 deletions
Download Patch File Download Diff File Expand all files Collapse all files

82
decode.c
View File

@@ -49,10 +49,8 @@ table_lookup(unsigned cur_idx, unsigned entry_idx) {
} }
static unsigned static unsigned
table_walk(unsigned cur_idx, unsigned entry_idx, unsigned* out_kind) { table_walk(unsigned table_entry, unsigned entry_idx) {
unsigned entry = table_lookup(cur_idx, entry_idx); return table_lookup(table_entry & ~0x3, entry_idx);
*out_kind = entry & ENTRY_MASK;
return (entry & ~ENTRY_MASK) >> 1;
} }
#define LOAD_LE_1(buf) ((uint64_t) *(uint8_t*) (buf)) #define LOAD_LE_1(buf) ((uint64_t) *(uint8_t*) (buf))
@@ -113,15 +111,14 @@ fd_decode(const uint8_t* buffer, size_t len_sz, int mode_int, uintptr_t address,
// Ensure that we can actually handle the decode request // Ensure that we can actually handle the decode request
DecodeMode mode; DecodeMode mode;
unsigned table_idx; unsigned table_root_idx;
unsigned kind = ENTRY_TABLE_ROOT;
switch (mode_int) switch (mode_int)
{ {
#if defined(FD_TABLE_OFFSET_32) #if defined(FD_TABLE_OFFSET_32)
case 32: table_idx = FD_TABLE_OFFSET_32; mode = DECODE_32; break; case 32: table_root_idx = FD_TABLE_OFFSET_32; mode = DECODE_32; break;
#endif #endif
#if defined(FD_TABLE_OFFSET_64) #if defined(FD_TABLE_OFFSET_64)
case 64: table_idx = FD_TABLE_OFFSET_64; mode = DECODE_64; break; case 64: table_root_idx = FD_TABLE_OFFSET_64; mode = DECODE_64; break;
#endif #endif
default: return FD_ERR_INTERNAL; default: return FD_ERR_INTERNAL;
} }
@@ -153,7 +150,7 @@ fd_decode(const uint8_t* buffer, size_t len_sz, int mode_int, uintptr_t address,
if (UNLIKELY(off >= len)) if (UNLIKELY(off >= len))
return FD_ERR_PARTIAL; return FD_ERR_PARTIAL;
uint8_t prefix = buffer[off]; uint8_t prefix = buffer[off];
table_entry = table_lookup(table_idx, prefix); table_entry = table_lookup(table_root_idx, prefix);
if (LIKELY(table_entry - 0xfff8 >= 8)) if (LIKELY(table_entry - 0xfff8 >= 8))
break; break;
prefixes[PF_REX] = 0; prefixes[PF_REX] = 0;
@@ -173,22 +170,22 @@ fd_decode(const uint8_t* buffer, size_t len_sz, int mode_int, uintptr_t address,
prefix_rep = prefixes[PF_REP]; prefix_rep = prefixes[PF_REP];
} }
kind = table_entry & ENTRY_MASK; // table_entry kinds: INSTR(0), T16(1), ESCAPE_A(2), ESCAPE_B(3)
table_idx = (table_entry & ~ENTRY_MASK) >> 1; if (LIKELY(!(table_entry & 2))) {
if (LIKELY(kind != 7)) {
off++; off++;
// Then, walk through ModR/M-encoded opcode extensions. // Then, walk through ModR/M-encoded opcode extensions.
if (kind == ENTRY_TABLE16 && LIKELY(off < len)) { if (table_entry & 1) {
if (UNLIKELY(off >= len))
return FD_ERR_PARTIAL;
unsigned isreg = (buffer[off] & 0xc0) == 0xc0 ? 8 : 0; unsigned isreg = (buffer[off] & 0xc0) == 0xc0 ? 8 : 0;
table_idx = table_walk(table_idx, ((buffer[off] >> 3) & 7) | isreg, &kind); table_entry = table_walk(table_entry, ((buffer[off] >> 3) & 7) | isreg);
if (kind == ENTRY_TABLE8E) // table_entry kinds: INSTR(0), T8E(1)
table_idx = table_walk(table_idx, buffer[off] & 7, &kind); if (table_entry & 1)
table_entry = table_walk(table_entry, buffer[off] & 7);
} }
if (UNLIKELY(kind != ENTRY_INSTR)) // table_entry kinds: INSTR(0)
return kind == 0 ? FD_ERR_UD : FD_ERR_PARTIAL;
goto direct; goto direct;
} }
@@ -221,7 +218,8 @@ fd_decode(const uint8_t* buffer, size_t len_sz, int mode_int, uintptr_t address,
return FD_ERR_PARTIAL; return FD_ERR_PARTIAL;
if (UNLIKELY(mode == DECODE_32 && buffer[off + 1] < 0xc0)) { if (UNLIKELY(mode == DECODE_32 && buffer[off + 1] < 0xc0)) {
off++; off++;
table_idx = table_walk(table_idx, 0, &kind); table_entry = table_walk(table_entry, 0);
// table_entry kinds: INSTR(0)
goto direct; goto direct;
} }
@@ -297,46 +295,56 @@ fd_decode(const uint8_t* buffer, size_t len_sz, int mode_int, uintptr_t address,
} }
} }
table_idx = table_walk(table_idx, opcode_escape, &kind); table_entry = table_walk(table_entry, opcode_escape);
if (kind == ENTRY_TABLE256 && LIKELY(off < len)) // table_entry kinds: INSTR(0) [only for invalid], T256(2)
table_idx = table_walk(table_idx, buffer[off++], &kind); if (UNLIKELY(!table_entry))
return FD_ERR_UD;
if (UNLIKELY(off >= len))
return FD_ERR_PARTIAL;
table_entry = table_walk(table_entry, buffer[off++]);
// table_entry kinds: INSTR(0), T16(1), TVEX(2), TPREFIX(3)
// Handle mandatory prefixes (which behave like an opcode ext.). // Handle mandatory prefixes (which behave like an opcode ext.).
if (kind == ENTRY_TABLE_PREFIX) if ((table_entry & 3) == 3)
{ table_entry = table_walk(table_entry, mandatory_prefix);
table_idx = table_walk(table_idx, mandatory_prefix, &kind); // table_entry kinds: INSTR(0), T16(1), TVEX(2)
}
// Then, walk through ModR/M-encoded opcode extensions. // Then, walk through ModR/M-encoded opcode extensions.
if (kind == ENTRY_TABLE16 && LIKELY(off < len)) { if (table_entry & 1) {
if (UNLIKELY(off >= len))
return FD_ERR_PARTIAL;
unsigned isreg = (buffer[off] & 0xc0) == 0xc0 ? 8 : 0; unsigned isreg = (buffer[off] & 0xc0) == 0xc0 ? 8 : 0;
table_idx = table_walk(table_idx, ((buffer[off] >> 3) & 7) | isreg, &kind); table_entry = table_walk(table_entry, ((buffer[off] >> 3) & 7) | isreg);
if (kind == ENTRY_TABLE8E) // table_entry kinds: INSTR(0), T8E(1), TVEX(2)
table_idx = table_walk(table_idx, buffer[off] & 7, &kind); if (table_entry & 1)
table_entry = table_walk(table_entry, buffer[off] & 7);
} }
// table_entry kinds: INSTR(0), TVEX(2)
// For VEX prefix, we have to distinguish between VEX.W and VEX.L which may // For VEX prefix, we have to distinguish between VEX.W and VEX.L which may
// be part of the opcode. // be part of the opcode.
if (UNLIKELY(kind == ENTRY_TABLE_VEX)) if (UNLIKELY(table_entry & 2))
{ {
uint8_t index = 0; uint8_t index = 0;
index |= prefix_rex & PREFIX_REXW ? (1 << 0) : 0; index |= prefix_rex & PREFIX_REXW ? (1 << 0) : 0;
// When EVEX.L'L is the rounding mode, the instruction must not have // When EVEX.L'L is the rounding mode, the instruction must not have
// L'L constraints. // L'L constraints.
index |= vexl << 1; index |= vexl << 1;
table_idx = table_walk(table_idx, index, &kind); table_entry = table_walk(table_entry, index);
} }
// table_entry kinds: INSTR(0)
if (UNLIKELY(kind != ENTRY_INSTR)) direct:
return kind == 0 ? FD_ERR_UD : FD_ERR_PARTIAL; // table_entry kinds: INSTR(0)
if (UNLIKELY(!table_entry))
return FD_ERR_UD;
direct:;
static _Alignas(16) const struct InstrDesc descs[] = { static _Alignas(16) const struct InstrDesc descs[] = {
#define FD_DECODE_TABLE_DESCS #define FD_DECODE_TABLE_DESCS
#include <fadec-decode-private.inc> #include <fadec-decode-private.inc>
#undef FD_DECODE_TABLE_DESCS #undef FD_DECODE_TABLE_DESCS
}; };
const struct InstrDesc* desc = &descs[table_idx >> 2]; const struct InstrDesc* desc = &descs[table_entry >> 2];
instr->type = desc->type; instr->type = desc->type;
instr->addrsz = addr_size; instr->addrsz = addr_size;

27
parseinstrs.py
View File

@@ -323,16 +323,16 @@ class InstrDesc(NamedTuple):
return f"{{FDI_{mnem}, {enc[0]}, {enc[1]}, {enc[2]}}}" return f"{{FDI_{mnem}, {enc[0]}, {enc[1]}, {enc[2]}}}"
class EntryKind(Enum): class EntryKind(Enum):
NONE = 0 NONE = 0x00
PREFIX = 8 PREFIX = 0x10
ESCAPE = 7 INSTR = 0x20
INSTR = 1 WEAKINSTR = 0x30
WEAKINSTR = 9 TABLE16 = 0x01
TABLE256 = 2 TABLE8E = 0x11
TABLE16 = 3 ESCAPE = 0x02
TABLE8E = 4 TABLE256 = 0x12
TABLE_PREFIX = 5 TABLE_VEX = 0x22
TABLE_VEX = 6 TABLE_PREFIX = 0x03
TABLE_ROOT = -1 TABLE_ROOT = -1
@property @property
def is_table(self): def is_table(self):
@@ -561,7 +561,7 @@ class Trie:
entry = self.trie[entry_num] entry = self.trie[entry_num]
if not entry[entry_idx] or entry[entry_idx][0] == EntryKind.WEAKINSTR: if not entry[entry_idx] or entry[entry_idx][0] == EntryKind.WEAKINSTR:
kind = EntryKind.INSTR if not weak else EntryKind.WEAKINSTR kind = EntryKind.INSTR if not weak else EntryKind.WEAKINSTR
entry[entry_idx] = kind, descidx << 3 entry[entry_idx] = kind, descidx << 2
elif not weak: elif not weak:
raise Exception(f"redundant non-weak {opcode}") raise Exception(f"redundant non-weak {opcode}")
@@ -610,8 +610,8 @@ class Trie:
continue continue
for i, elem in enumerate(entry, start=off): for i, elem in enumerate(entry, start=off):
if elem is not None: if elem is not None:
value = (offsets[elem[1]] << 1) if elem[0].is_table else elem[1] value = offsets[elem[1]] if elem[0].is_table else elem[1]
data[i] = value | (elem[0].value & 7) data[i] = value | (elem[0].value & 3)
return tuple(data), [offsets[v] for _, v in self.trie[0]] return tuple(data), [offsets[v] for _, v in self.trie[0]]
@property @property
@@ -667,6 +667,7 @@ def decode_table(entries, args):
trie.add_prefix(rex, 0xfffe, i) trie.add_prefix(rex, 0xfffe, i)
mnems, descs, desc_map = set(), [], {} mnems, descs, desc_map = set(), [], {}
descs.append("{0}") # desc index zero is "invalid"
for weak, opcode, desc in entries: for weak, opcode, desc in entries:
ign66 = opcode.prefix in ("NP", "66", "F2", "F3") ign66 = opcode.prefix in ("NP", "66", "F2", "F3")
modrm = opcode.modreg or opcode.opcext modrm = opcode.modreg or opcode.opcext