On Mon, May 15, 2023 at 08:49:54AM -0400, Vincent Fu wrote: > On 5/15/23 07:03, Ankit Kumar wrote: > > The NVM command set specification 1.0c supports 64 LBA formats. > > The first patch adds support for that. > > > > The io_uring_cmd ioengine assumes that logical block size is always > > power of 2. But with namespace formats where metadata is transferred > > at the end of logical block i.e. an extended logical block this is > > not true. > > This second patch adds the support for extended logical block sizes. > > This however doesn't include support for protection info and metadata > > transferred as separate buffer, we return error for those. > > > > Changes from v1: > > * Split the original patch into 2 as suggested by Vincent and Jens > > * Address all review comments from Vincent > > > > Ankit Kumar (2): > > engines/nvme: support for 64 LBA formats > > engines/io_uring: io_uring_cmd engine fixes > > > > engines/io_uring.c | 30 ++++++++++++++++++--- > > engines/nvme.c | 66 ++++++++++++++++++++++++++++++++++++++-------- > > engines/nvme.h | 6 ++--- > > 3 files changed, 84 insertions(+), 18 deletions(-) > > > > Applied with a change to the title of the second patch's commit message. > Thanks. > > Vincent > Thanks, I was wondering if extended LBA changes are required for t/io_uring, as its part of fast path, and any branching introduced because of lba_shift and lba_ext may have an impact on performace. Ankit
