general: Improve documentation

README.md

@@ -67,8 +67,8 @@ failed |= fe_enc64(&cur, FE_XOR32rr, FE_AX, FE_AX);
 failed |= fe_enc64(&cur, FE_MOVZXr32m8, FE_CX, FE_MEM(FE_DI, 1, FE_AX, 0));
 // test ecx, ecx
 failed |= fe_enc64(&cur, FE_TEST32rr, FE_CX, FE_CX);
-// jz $
-// This will be replaced later FE_JMPL enforces use of longest offset
+// jnz $
+// This will be replaced later; FE_JMPL enforces use of longest offset
 uint8_t* fwd_jmp = cur;
 failed |= fe_enc64(&cur, FE_JNZ|FE_JMPL, (intptr_t) cur);
 uint8_t* loop_tgt = cur;
@@ -104,8 +104,9 @@ The API consists of one function to handle encode requests, as well as some macr
         - For offset operands, specify the target address.
 
 ## Known issues
-- The encoder doesn't support VEX encodings (yet).
 - The EVEX prefix (AVX-512) is not supported (yet).
+- MPX instructions are not supported.
+- HLE prefixes `xacquire`/`xrelease` are not supported by the encoder (yet).
 - Prefixes for indirect jumps and calls are not properly decoded, e.g. `notrack`, `bnd`.
 - Low test coverage. (Help needed.)
 - No Python API.

fadec-enc.h

@@ -22,16 +22,26 @@ typedef enum {
 
 typedef int64_t FeOp;
 
+/** Construct a memory operand. Unused parts can be set to 0 and will be
+ * ignored. FE_IP can be used as base register, in which case the offset is
+ * interpreted as the offset from the /current/ position -- the size of the
+ * encoded instruction will be subtracted during encoding. scale must be 1, 2,
+ * 4, or 8; but is ignored if  idx == 0. **/
 #define FE_MEM(base,sc,idx,off) (INT64_MIN | ((int64_t) ((base) & 0xfff) << 32) | ((int64_t) ((idx) & 0xfff) << 44) | ((int64_t) ((sc) & 0xf) << 56) | ((off) & 0xffffffff))
 
+/** Add segment override prefix. This may or may not generate prefixes for the
+ * ignored prefixes ES/CS/DS/SS in 64-bit mode. **/
 #define FE_SEG(seg) ((((seg) & 0x7) + 1) << 16)
+/** Do not use. **/
 #define FE_SEG_MASK 0x70000
+/** Overrides address size. **/
 #define FE_ADDR32 0x80000
 /** Used together with a RIP-relative (conditional) jump, this will force the
  * use of the encoding with the largest distance. Useful for reserving a jump
  * when the target offset is still unknown; if the jump is re-encoded later on,
  * FE_JMPL must be specified there, too, so that the encoding lengths match. **/
 #define FE_JMPL 0x100000
+/** Do not use. **/
 #define FE_MNEM_MASK 0xffff
 
 enum {
@@ -41,8 +51,22 @@ enum {
     FE_MNEM_MAX
 };
 
+/** Do not use. **/
 #define fe_enc64_1(buf, mnem, op0, op1, op2, op3, ...) fe_enc64_impl(buf, mnem, op0, op1, op2, op3)
+/** Encode a single instruction for 64-bit mode.
+ * \param buf Pointer to the buffer for instruction bytes, must have a size of
+ *        15 bytes. The pointer is advanced by the number of bytes used for
+ *        encoding the specified instruction.
+ * \param mnem Mnemonic, optionally or-ed with FE_SEG(), FE_ADDR32, or FE_JMPL.
+ * \param operands... Instruction operands. Immediate operands are passed as
+ *        plain value; register operands using the FeReg enum; memory operands
+ *        using FE_MEM(); and offset operands for RIP-relative jumps/calls are
+ *        specified as _address in buf_, e.g. (intptr_t) jmptgt, the address of
+ *        buf and the size of the encoded instruction are subtracted internally.
+ * \return Zero for success or a negative value in case of an error.
+ **/
 #define fe_enc64(buf, ...) fe_enc64_1(buf, __VA_ARGS__, 0, 0, 0, 0, 0)
+/** Do not use. **/
 int fe_enc64_impl(uint8_t** buf, uint64_t mnem, FeOp op0, FeOp op1, FeOp op2, FeOp op3);
 
 #ifdef __cplusplus

fadec.h

@@ -42,6 +42,7 @@ enum {
     FD_FLAG_64 = 1 << 7,
 };
 
+/** Operand types. **/
 typedef enum {
     FD_OT_NONE = 0,
     FD_OT_REG = 1,
@@ -186,8 +187,7 @@ const char* fdi_name(FdInstrType ty);
 #define FD_HAS_REP(instr) ((instr)->flags & FD_FLAG_REP)
 /** Indicates whether the instruction was encoded with a REPNZ prefix. **/
 #define FD_HAS_REPNZ(instr) ((instr)->flags & FD_FLAG_REPNZ)
-/** Indicates whether the instruction was encoded with a LOCK prefix. Note that
- * it is not checked whether the LOCK prefix is valid for the instruction. **/
+/** Indicates whether the instruction was encoded with a LOCK prefix. **/
 #define FD_HAS_LOCK(instr) ((instr)->flags & FD_FLAG_LOCK)
 /** Do not use. **/
 #define FD_IS64(instr) ((instr)->flags & FD_FLAG_64)
@@ -204,7 +204,7 @@ const char* fdi_name(FdInstrType ty);
 #define FD_OP_SIZE(instr,idx) ((instr)->operands[idx].size)
 /** Gets the accessed register index of a register operand. Note that /only/ the
  * index is returned, no further interpretation of the index (which depends on
- * the instruction type) is done. The register type can be fetches using
+ * the instruction type) is done. The register type can be fetched using
  * FD_OP_REG_TYPE, e.g. for distinguishing high-byte registers.
  * Only valid if  FD_OP_TYPE == FD_OT_REG  **/
 #define FD_OP_REG(instr,idx) ((FdReg) (instr)->operands[idx].reg)
@@ -236,7 +236,7 @@ const char* fdi_name(FdInstrType ty);
  * Only valid if  FD_OP_TYPE == FD_OT_MEM  **/
 #define FD_OP_DISP(instr,idx) ((int64_t) (instr)->disp)
 /** Gets the (sign-extended) encoded constant for an immediate operand.
- * Only valid if  FD_OP_TYPE == FD_OT_IMM  **/
+ * Only valid if  FD_OP_TYPE == FD_OT_IMM  or  FD_OP_TYPE == FD_OT_OFF  **/
 #define FD_OP_IMM(instr,idx) ((instr)->imm)
 
 #ifdef __cplusplus