Hi, iomap always punts async dio write completions to a workqueue, which has a cost in terms of efficiency (now you need an unrelated worker to process it) and latency (now you're bouncing a completion through an async worker, which is a classic slowdown scenario). Even for writes that should, in theory, be able to complete inline, if we race with truncate or need to invalidate pages post completion, we cannot sanely be in IRQ context as the locking types don't allow for that. io_uring handles IRQ completions via task_work, and for writes that don't need to do extra IO at completion time, we can safely complete them inline from that. This patchset adds IOCB_DEFER, which an IO issuer can set to inform the completion side that any extra work that needs doing for that completion can be punted to a safe task context. The iomap dio completion will happen in hard/soft irq context, and we need a saner context to process these completions. IOCB_DIO_DEFER is added, which can be set in a struct kiocb->ki_flags by the issuer. If the completion side of the iocb handling understands this flag, it can choose to set a kiocb->dio_complete() handler and just call ki_complete from IRQ context. The issuer must then ensure that this callback is processed from a task. io_uring punts IRQ completions to task_work already, so it's trivial wire it up to run more of the completion before posting a CQE. This is good for up to a 37% improvement in throughput/latency for low queue depth IO, patch 5 has the details. If we need to do real work at completion time, iomap will clear the IOMAP_DIO_DEFER_COMP flag. This work came about when Andres tested low queue depth dio writes for postgres and compared it to doing sync dio writes, showing that the async processing slows us down a lot. Dave, would appreciate your input on if the logic is right now in terms of when we can inline complete when DEFER is set! fs/iomap/direct-io.c | 102 +++++++++++++++++++++++++++++++++++-------- include/linux/fs.h | 30 ++++++++++++- io_uring/rw.c | 27 ++++++++++-- 3 files changed, 136 insertions(+), 23 deletions(-) Can also be found in a git branch here: https://git.kernel.dk/cgit/linux/log/?h=xfs-async-dio.3 Since v2: - Drop the poll specific bits. They end up folding under the last patch now anyway, and this avoids needing a weird "is bio still polled" or in_task() check. - Keep non-IOCB_DEFER writes in the workqueue. - Cleanup the iomap completion path first. - Add patch treating fua && has_fua the same as fua && !write_cache - Add explicit IOMAP_DIO_DEFER_COMP flag - Add comments -- Jens Axboe
