Hi António,
for the "basic" number of mapped reads, I would simply use Log.final.out entries:
Uniquely mapped reads number + Number of reads mapped to multiple loci + Number of reads mapped to too many loci
This number will include some of the chimeric/circular reads - those which can be mapped with short enough soft-clipping.
The number of chimeric reads should be very small compared to the total number of reads, so, in principle, it will not affect the normalization factor significantly.
However, if you want to be rigorous, to avoid double counting you would need to extract read names from Aligned.out.sam and Chimeric.out.sam, and count the unique names, e.g.:
cut -f1 Aligned.out.sam Chimeric.out.sam | grep -v "^@" | sort | uniq | wc
Cheers
Alex