Hi I accepted the patches. I just changed the mempool sizes from 1 to MIN_IOS, so that the requests are not serialized when the system is out of memory.
Mikulas On Mon, 30 Mar 2026, Keith Busch wrote: > From: Keith Busch <[email protected]> > > This is the updated series for allowing arbitrary memory alignment, > enabling direct-io without bounce buffers for dm-crypt. > > The previous version was posted here: > > https://lore.kernel.org/dm-devel/[email protected]/ > > And is based on the current dm-7.1 branch, which has the necessary > dm-integrity updates that make patch 1 a no-op. > > Changes from v3: > > * Split off the generic dm-table queue limits update into its own prep > patch. > > * Squashed the rest into a single patch just so it doesn't introduce a > temporary commit point where the reported limits can't always be > executed. > > * Do not report a dma_alignment lower than the backing device's; this > is necessary as the incoming buffers are used directly for READ > rather than bounced like WRITE. > > * Fallback to a mempool if we can't allocate the scatterlist with > kmalloc and updated gfp_t flags accordingly. > > * Various cleanups. > > Keith Busch (2): > dm: initialize dma_alignment to 0 > dm-crypt: allow unaligned bio_vecs for direct io > > drivers/md/dm-crypt.c | 147 ++++++++++++++++++++++++++++++++++++------ > drivers/md/dm-table.c | 1 + > 2 files changed, 128 insertions(+), 20 deletions(-) > > -- > 2.52.0 >
