In computer systems, the built-in disk is essential for data storage as computers typically need to read and write data from the disk during operations. Some existing works have utilized the read/write characteristics of disks to infer sensitive information, such as browsing websites. However, previous research primarily focuses on hard disk drives (HDDs). With solid state drives (SSDs) gradually becoming the mainstream option for computer disks, these approaches have shown limitations. In this paper, we reveal a novel side-channel vulnerability targeting both HDDs and SSDs, named EMLogger. Specifically, we find that attackers can exploit the electromagnetic (EM) radiation leaked by the disk to detect ongoing activities on the computer. To enhance the strength of EM signals, we propose a sub-signal fusion method. Besides, we employ machine learning techniques for feature extraction and activity classification from the enhanced EM signals. Finally, we conduct real-world experiments on computers equipped with HDDs or SSDs. Our experimental results demonstrate that EMLogger achieves an accuracy of over 98% in inferring computer activities. Furthermore, the experiments validate the robustness of EMLogger at varying attack distances.