sync causing a flush with no data
I have noticed that when entering suspend-to-ram, the kernel syncs the filesystem, and ext4 issues a flush even though nothing has been written, causing a disk that has been suspended with runtime pm to wake up for no reason.
It looks like ext4_sync_fs() calls jbd2_trans_will_send_data_barrier(), which returns 0, causing a call to blkdev_issue_flush(). Shouldn't this return a 1 if the transaction is empty?
Quoted 3 lines · Phillip Susi
It looks like
ext4_sync_fs()callsjbd2_trans_will_send_data_barrier(), which returns 0, causing a call toblkdev_issue_flush(). Shouldn't this return a 1 if the transaction is empty?
Tracing a little more, it appears that jbd2_trans_will_send_data_barrier() is returning 0 because the transaction is already committed. How can this be? If a transaction has been committed, shouldn't a new one be opened? If not, and the transaction indeed has already been committed, then why is that a reason for the fs to issue another barrier? If it has already been committed, then the barrier should already have been issued shouldn't it?