With an unreasonably small innodb_log_file_size, the page cleaner
thread would frequently acquire log_sys.flush_order_mutex and spend
a significant portion of CPU time spinning on that mutex when
determining the checkpoint LSN.