linux-stable/io_uring
Jens Axboe 7cc2a6eadc io_uring: add IORING_REGISTER_COPY_BUFFERS method
Buffers can get registered with io_uring, which allows to skip the
repeated pin_pages, unpin/unref pages for each O_DIRECT operation. This
reduces the overhead of O_DIRECT IO.

However, registrering buffers can take some time. Normally this isn't an
issue as it's done at initialization time (and hence less critical), but
for cases where rings can be created and destroyed as part of an IO
thread pool, registering the same buffers for multiple rings become a
more time sensitive proposition. As an example, let's say an application
has an IO memory pool of 500G. Initial registration takes:

Got 500 huge pages (each 1024MB)
Registered 500 pages in 409 msec

or about 0.4 seconds. If we go higher to 900 1GB huge pages being
registered:

Registered 900 pages in 738 msec

which is, as expected, a fully linear scaling.

Rather than have each ring pin/map/register the same buffer pool,
provide an io_uring_register(2) opcode to simply duplicate the buffers
that are registered with another ring. Adding the same 900GB of
registered buffers to the target ring can then be accomplished in:

Copied 900 pages in 17 usec

While timing differs a bit, this provides around a 25,000-40,000x
speedup for this use case.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
2024-09-12 10:14:15 -06:00
..
advise.c io_uring/advise: support 64-bit lengths 2024-06-16 14:54:55 -06:00
advise.h
alloc_cache.h io_uring/alloc_cache: switch to array based caching 2024-04-15 08:10:25 -06:00
cancel.c io_uring: fix warnings on shadow variables 2024-04-15 08:10:26 -06:00
cancel.h io_uring: fix cancellation overwriting req->flags 2024-06-13 19:25:28 -06:00
epoll.c io_uring: undeprecate epoll_ctl support 2023-05-26 20:22:41 -06:00
epoll.h
eventfd.c io_uring/eventfd: move refs to refcount_t 2024-09-08 16:43:57 -06:00
eventfd.h io_uring/eventfd: move eventfd handling to separate file 2024-06-16 14:54:55 -06:00
fdinfo.c io_uring: add napi busy settings to the fdinfo output 2024-08-25 08:27:00 -06:00
fdinfo.h
filetable.c io_uring/filetable: don't unnecessarily clear/reset bitmap 2024-05-08 08:27:45 -06:00
filetable.h io_uring: expand main struct io_kiocb flags to 64-bits 2024-02-08 13:27:03 -07:00
fs.c io_uring/fs: consider link->flags when getting path for LINKAT 2023-11-20 09:01:42 -07:00
fs.h
futex.c io_uring/alloc_cache: switch to array based caching 2024-04-15 08:10:25 -06:00
futex.h io_uring/alloc_cache: switch to array based caching 2024-04-15 08:10:25 -06:00
io-wq.c io_uring/io-wq: inherit cpuset of cgroup in io worker 2024-09-11 07:27:56 -06:00
io-wq.h io_uring/io-wq: make io_wq_work flags atomic 2024-06-16 14:54:55 -06:00
io_uring.c io_uring/kbuf: pass in 'len' argument for buffer commit 2024-08-29 08:44:51 -06:00
io_uring.h io_uring: add support for batch wait timeout 2024-08-25 08:27:01 -06:00
kbuf.c io_uring/kbuf: add support for incremental buffer consumption 2024-08-29 08:44:58 -06:00
kbuf.h io_uring/kbuf: add support for incremental buffer consumption 2024-08-29 08:44:58 -06:00
Makefile io_uring: add GCOV_PROFILE_URING Kconfig option 2024-08-30 10:52:02 -06:00
memmap.c io_uring: don't attempt to mmap larger than what the user asks for 2024-05-29 09:53:14 -06:00
memmap.h io_uring: move mapping/allocation helpers to a separate file 2024-04-15 08:10:26 -06:00
msg_ring.c io_uring/msg_ring: fix uninitialized use of target_req->flags 2024-07-25 08:41:35 -06:00
msg_ring.h io_uring/msg_ring: add an alloc cache for io_kiocb entries 2024-06-24 08:39:55 -06:00
napi.c io_uring: user registered clockid for wait timeouts 2024-08-25 08:27:01 -06:00
napi.h io_uring/napi: postpone napi timeout adjustment 2024-08-25 08:27:01 -06:00
net.c io_uring/kbuf: pass in 'len' argument for buffer commit 2024-08-29 08:44:51 -06:00
net.h io_uring: Introduce IORING_OP_LISTEN 2024-06-19 07:57:21 -06:00
nop.c io_uring: support to inject result for NOP 2024-05-10 06:09:45 -06:00
nop.h
notif.c io_uring/notif: disable LAZY_WAKE for linked notifs 2024-04-30 13:06:27 -06:00
notif.h io_uring/notif: implement notification stacking 2024-04-22 19:31:18 -06:00
opdef.c io_uring: Fix probe of disabled operations 2024-06-19 08:58:00 -06:00
opdef.h io_uring: Fix probe of disabled operations 2024-06-19 08:58:00 -06:00
openclose.c io_uring: enable audit and restrict cred override for IORING_OP_FIXED_FD_INSTALL 2024-01-23 15:25:14 -07:00
openclose.h io_uring/openclose: add support for IORING_OP_FIXED_FD_INSTALL 2023-12-12 07:42:57 -07:00
poll.c io_uring: keep multishot request NAPI timeout current 2024-07-30 06:18:58 -06:00
poll.h io_uring/poll: shrink alloc cache size to 32 2024-04-15 08:10:25 -06:00
refs.h io_uring: kill dead code in io_req_complete_post 2024-04-15 08:10:26 -06:00
register.c io_uring: add IORING_REGISTER_COPY_BUFFERS method 2024-09-12 10:14:15 -06:00
register.h io_uring/register: provide helper to get io_ring_ctx from 'fd' 2024-09-12 10:14:05 -06:00
rsrc.c io_uring: add IORING_REGISTER_COPY_BUFFERS method 2024-09-12 10:14:15 -06:00
rsrc.h io_uring: add IORING_REGISTER_COPY_BUFFERS method 2024-09-12 10:14:15 -06:00
rw.c io_uring/rw: drop -EOPNOTSUPP check in __io_complete_rw_common() 2024-09-10 09:34:44 -06:00
rw.h io_uring/alloc_cache: switch to array based caching 2024-04-15 08:10:25 -06:00
slist.h io_uring: silence variable ‘prev’ set but not used warning 2023-03-09 10:10:58 -07:00
splice.c splice: return type ssize_t from all helpers 2023-12-12 16:19:59 +01:00
splice.h
sqpoll.c io_uring/sqpoll: do not allow pinning outside of cpuset 2024-09-09 09:09:08 -06:00
sqpoll.h io_uring/sqpoll: statistics of the true utilization of sq threads 2024-03-01 06:28:19 -07:00
statx.c vfs: retire user_path_at_empty and drop empty arg from getname_flags 2024-06-05 17:03:57 +02:00
statx.h
sync.c io_uring: for requests that require async, force it 2023-01-29 15:18:26 -07:00
sync.h
tctx.c io_uring: Add io_uring_setup flag to pre-register ring fd and never install it 2023-05-16 08:06:00 -06:00
tctx.h io_uring: simplify __io_uring_add_tctx_node 2022-10-07 12:25:30 -06:00
timeout.c io_uring: fix io_match_task must_hold 2024-07-24 08:01:49 -06:00
timeout.h io_uring: remove unused return from io_disarm_next 2022-09-21 13:15:01 -06:00
truncate.c io_uring: add support for ftruncate 2024-02-09 09:04:39 -07:00
truncate.h io_uring: add support for ftruncate 2024-02-09 09:04:39 -07:00
uring_cmd.c io_uring: simplify io_uring_cmd return 2024-07-24 08:01:49 -06:00
uring_cmd.h io_uring/alloc_cache: switch to array based caching 2024-04-15 08:10:25 -06:00
waitid.c io_uring: remove struct io_tw_state::locked 2024-04-15 08:10:24 -06:00
waitid.h io_uring: add IORING_OP_WAITID support 2023-09-21 12:04:45 -06:00
xattr.c vfs: retire user_path_at_empty and drop empty arg from getname_flags 2024-06-05 17:03:57 +02:00
xattr.h