mirror of
https://mirrors.bfsu.edu.cn/git/linux.git
synced 2024-09-22 04:31:58 +08:00
377698d4ab
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). 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_DIO_CALLER_COMP, 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_CALLER_COMP 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_CALLER_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. -----BEGIN PGP SIGNATURE----- iQJEBAABCAAuFiEEwPw5LcreJtl1+l5K99NY+ylx4KYFAmTJllIQHGF4Ym9lQGtl cm5lbC5kawAKCRD301j7KXHgpiZTEADDnmmnxPGpZElCRSauiwuUoV1yHFnUytXK MUmCbV2Yf2Fuh4rpuf5JkQpmn/80NdSQDrUFD/1jJQ2bT4SPAE/WmjsenZf2mmmd DxKQ7vv+P4LgbU+tDuNbCnQsqeVSCUyxdVlMf43M1OxzCuxLnSmUQZZ/EbolS0+H ko6rrKFjpdWr4b+VeAnhyoBj4kQLNQG1ctij92RhO+Fgs1q7sGCXd1nQ3lrSXpHS T+bh2/vK8QPGwvJYvvaSAvTttUs3+X1fOlK4gb+tuiocpA08py9Qn4AA6iPbFnZR pE3vyYR0Yj2wzlqT4ucmTMIAFHpJNBk0wDeqv9ozzcoY862xk+bQQJf14XiSn8/4 WE/ipnRVb47xAI28UcrpBrrhdBje+hj0WCsKI7BoMrhrNWG9rOlK6FplEWyh/mWS Q0VdtHIlgTRZFa6597VIBBsv1QMCenNpCxZn3Kkg3BUkzhQrkW6ki7kd9JAm62dh VLu9vZPHufU52oDZ89EsHXQINow19X9zjhozsNv9nZFbPspt+Uk1n0aEO/sbxATI KPPQoztic6nVCOui1Sa8GVojZIefwioozcDdAbsq5gdir+45aq5BDFM354mtkZna oq8ioolKNVOnERpEizbLGzSd1YpLtJtaiO3MRrupNB6uESGdhv61SCickRSPOqMN U9p27/dvvg== =EnSA -----END PGP SIGNATURE----- Merge tag 'xfs-async-dio.6-2023-08-01' of git://git.kernel.dk/linux into iomap-6.6-mergeA Improve iomap/xfs async dio write performance 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). 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_DIO_CALLER_COMP, 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_CALLER_COMP 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_CALLER_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. * tag 'xfs-async-dio.6-2023-08-01' of git://git.kernel.dk/linux: iomap: support IOCB_DIO_CALLER_COMP io_uring/rw: add write support for IOCB_DIO_CALLER_COMP fs: add IOCB flags related to passing back dio completions iomap: add IOMAP_DIO_INLINE_COMP iomap: only set iocb->private for polled bio iomap: treat a write through cache the same as FUA iomap: use an unsigned type for IOMAP_DIO_* defines iomap: cleanup up iomap_dio_bio_end_io() Signed-off-by: Darrick J. Wong <djwong@kernel.org> |
||
|---|---|---|
| .. | ||
| advise.c | ||
| advise.h | ||
| alloc_cache.h | ||
| cancel.c | ||
| cancel.h | ||
| epoll.c | ||
| epoll.h | ||
| fdinfo.c | ||
| fdinfo.h | ||
| filetable.c | ||
| filetable.h | ||
| fs.c | ||
| fs.h | ||
| io_uring.c | ||
| io_uring.h | ||
| io-wq.c | ||
| io-wq.h | ||
| kbuf.c | ||
| kbuf.h | ||
| Makefile | ||
| msg_ring.c | ||
| msg_ring.h | ||
| net.c | ||
| net.h | ||
| nop.c | ||
| nop.h | ||
| notif.c | ||
| notif.h | ||
| opdef.c | ||
| opdef.h | ||
| openclose.c | ||
| openclose.h | ||
| poll.c | ||
| poll.h | ||
| refs.h | ||
| rsrc.c | ||
| rsrc.h | ||
| rw.c | ||
| rw.h | ||
| slist.h | ||
| splice.c | ||
| splice.h | ||
| sqpoll.c | ||
| sqpoll.h | ||
| statx.c | ||
| statx.h | ||
| sync.c | ||
| sync.h | ||
| tctx.c | ||
| tctx.h | ||
| timeout.c | ||
| timeout.h | ||
| uring_cmd.c | ||
| uring_cmd.h | ||
| xattr.c | ||
| xattr.h | ||