Changes to v2.1:
  - support FRED
  - support APX
  - support Trigger Tracing (PTTT)
  - support resync on error
  - move insn decoder on top of block decoder
  - bug fixes

Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>

XED v2025.11.23 removed xed_decoded_inst_get_byte().  Take the bytes from
the original input to xed_decode().

Fixes #119.

Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>

Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>

Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>

Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>

Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>

Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>

Any error in pt_sb_pevent_init() after loading the trace buffer results in
a call to pt_sb_pevent_dtor() to undo any partial initialization.

pt_sb_pevent_dtor() incorrectly frees not just any memory allocated inside
the pevent private data, but also the private data object itself.

This results in a double-free when the private data object is later freed
again in pt_sb_alloc_pevent_decoder().

We do want to allow allocating the private data as part of another object,
so we cannot allow its dtor() to free the object itself.  Clarify that in
the dtor() comment.

Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>

When decoding an instruction truncated after a REX2 prefix, the prefix
decoder will read the next byte without checking the size.

In almost all cases, this will be another byte in the struct pt_insn's raw
array, which has been zero-initialized in most flows.  Initialize it on
the flows that used an uninitialized pt_insn.

If the REX2 prefix is preceded by other prefixes such that REX2 is at the
very end of the raw array, this will be the struct pt_insn's size field,
which has been initialized to the number of valid bytes in the raw array,
i.e. 15.

Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>