On 07/02/2020 17:14, David Sterba wrote: > On Thu, Feb 06, 2020 at 03:29:57PM +0000, Johannes Thumshirn wrote: >> On 06/02/2020 15:58, Christoph Hellwig wrote: >>> Also I just noticed don't even need the kmap/kunmap at all given that the >>> block device mapping is never in highmem. >>> >> >> This potentially touches more places, I'll cover that in a dedicated >> patchset. > > Are the kmap/kunmaps anywhere on the buffer_head call paths? I can't > find it anywhere, and given that the mapping does not contain highmem > pages we could rather avoid adding it from the beginning. There's at least one more in btrfs_read_disk_super(), but yes I can avoid them in the BH path and remove the ones I find in the next go.
