All 32-bit variants of BPF_FETCH (add, and, or, xor, xchg, cmpxchg)
define a 32-bit subreg and thus have zext_dst set. Their encoding,
however, uses dst_reg field as a base register, which causes
opt_subreg_zext_lo32_rnd_hi32() to zero-extend said base register
instead of the one the insn really defines (r0 or src_reg).

Fix by properly choosing a register being defined, similar to how
check_atomic() already does that.
Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20210210204502.83429-1-iii@linux.ibm.com