io_uring wants to relax the page alignment requirements for O_DIRECT, use the dma alignment for the underlying device instead. This poses a problem for btrfs because we expect to checksum a page and have a hard requirement that the page offset == 0. This may be solved with the subpage blocksize support, but will need to be revisited once that lands to verify that it's actually fixed. Jens has patches pending for this work and haven't been submitted upstream yet, but likely target is 5.13, so it's not a deep and burning issue.
io_uring wants to relax the page alignment requirements for O_DIRECT, use the dma alignment for the underlying device instead. This poses a problem for btrfs because we expect to checksum a page and have a hard requirement that the page offset == 0. This may be solved with the subpage blocksize support, but will need to be revisited once that lands to verify that it's actually fixed. Jens has patches pending for this work and haven't been submitted upstream yet, but likely target is 5.13, so it's not a deep and burning issue.