Looking from the other side I usually view mito (chrM) mapping reads as contamination and remove them as they cause errors in MACS statistical analyses, such as FDR calculation. So I am guessing that genomic DNA will be a contaminant for your analysis, in which case I would first map reads to the entire sequence set (mito and nuclear) and then remove nuclear mapping reads. You should then be able to use MACS etc. If you use MACS you would need to alter the lambda model parameters as it looks at genomic regions 1kb and 10kb away from candidate regions.

I hope this is of some help.