[PATCHv3,0/6] direct io dma alignment

Message

Keith Busch May 23, 2022, 9:01 p.m. UTC

From: Keith Busch <[e-mail]>

Here is the 3rd version enabling non-block aligned user space addresses
for O_DIRECT.

Changes since v2:

  Folded in improvements cleanin gup zone append pages (Christoph)

  Added documentation the exported attribute (Bart)

  Split bdev_dma_alignment() into its own patch (Christoph)

  Removed fs/ from implementing support for these address offsets for
  now

  Fixed up a couple places assuming SECTOR_SIZE multiple bv_len

On that last point, I searched through much of the code and found a few
other places that also assumed this bv_len size, but they didn't apply
to this change since they don't set 'dma_alignment'. This got me
thinking, though, should this be a new attribute, 'dio_alignment',
instead of using the pre-existing 'dma_alignment' for this purpose? Or
is it clear enough that drivers setting the existing attribute ought to
know what they're getting into?

Keith Busch (6):
  block/bio: remove duplicate append pages code
  block: export dma_alignment attribute
  block: introduce bdev_dma_alignment helper
  block/merge: count bytes instead of sectors
  block/bounce: count bytes instead of sectors
  block: relax direct io memory alignment

 Documentation/ABI/stable/sysfs-block |   9 ++
 block/bio.c                          | 118 +++++++++++++--------------
 block/blk-merge.c                    |  35 ++++----
 block/blk-sysfs.c                    |   7 ++
 block/bounce.c                       |   5 +-
 block/fops.c                         |  41 +++++++---
 include/linux/blkdev.h               |   5 ++
 7 files changed, 127 insertions(+), 93 deletions(-)