To be all honest, I never look at these metrics. I check per-base quality and adapter content. I do trimming if there are adapters detected. ChIP-seq is very noisy, this here might be some odd ligation artifacts, leftover adapter dimers that were sequenced or anything else. Much more important is to check alignment percentage, number of callable peaks and percent reads overlapping callable peaks. Also make a bigwig file (for example deeptools bamCoverage) and check peaks on the genome browser/IGV. That is the relevant QC.