On 6/6/19 12:04 AM, Christoph Hellwig wrote:
On Wed, Jun 05, 2019 at 11:47:23PM -0700, Sagi Grimberg wrote:
Not sure I understand.
max_segment_size and virt_boundary_mask are related how?
virt_boundary_devices has hardware segment size of a single page (device
page as identified by the boundary, not necessarily linux PAGE_SIZE).
So we don't need a max_segment_size in the Linux size, as any amount of
hardware segments fitting the virt boundary can be merged into a 'Linux
segment'. Because of the accouting for the merges is hard and was
partially broken anyway we've now dropped the accounting and force the
max_segment_size to be unlimited in the block layer if a device sets
the virt_boundary_mask.
Hi Christoph,
What if that accounting would ever be implemented in the block layer?
Would that mean that v2 of this patch becomes incorrect and has to be
changed back into v1 of this patch? Is v1 incorrect today?
Thank you,
Bart.
