io_uring: add napi busy polling support
Subsystems (3)
v5 cover io_uring: add napi busy polling support
This adds the napi busy polling support in io_uring.c. It adds a new napi_list to the io_ring_ctx structure. This list contains the list of napi_id's that are currently enabled for busy polling. This list is used to determine which napi id's enabled busy polling.
io-uring allows specifying two parameters:
- busy poll timeout and
- prefer busy poll to call of io_napi_busy_loop()
This sets the above parameters for the ring. The settings are passed with a new structure io_uring_napi.
There is also a corresponding liburing patch series, which enables this feature. The name of the series is "liburing: add add api for napi busy poll timeout". It also contains two programs to test the this.
Testing has shown that the round-trip times are reduced to 38us from 55us by enabling napi busy polling with a busy poll timeout of 100us. More detailled results are part of the commit message of the first patch.
Changes:
- V5:
- Refreshed to 6.1-rc6
- Use copy_from_user instead of memdup/kfree
- Removed the moving of napi_busy_poll_to
- Return -EINVAL if any of the reserved or padded fields are not 0.
- V4:
- Pass structure for napi config, instead of individual parameters
- V3:
- Refreshed to 6.1-rc5
- Added a new io-uring api for the prefer napi busy poll api and wire it to io_napi_busy_loop().
- Removed the unregister (implemented as register)
- Added more performance results to the first commit message.
- V2:
- Add missing defines if CONFIG_NET_RX_BUSY_POLL is not defined
- Changes signature of function io_napi_add_list to static inline if CONFIG_NET_RX_BUSY_POLL is not defined
- define some functions as static
Stefan Roesch (3):
io_uring: add napi busy polling support
io_uring: add api to set / get napi configuration.
io_uring: add api to set napi prefer busy poll
include/linux/io_uring_types.h | 8 + include/uapi/linux/io_uring.h | 12 ++ io_uring/io_uring.c | 302 +++++++++++++++++++++++++++++++++ io_uring/napi.h | 22 +++ io_uring/poll.c | 3 + io_uring/sqpoll.c | 10 ++ 6 files changed, 357 insertions(+) create mode 100644 io_uring/napi.h
base-commit: eb7081409f94a9a8608593d0fb63a1aa3d6f95d8
Signature · 1 line
2.30.2
PATCH v5 3/3 io_uring: add api to set napi prefer busy poll
This adds an api to register and unregister the napi prefer busy poll setting from liburing. To be able to use this functionality, the corresponding liburing patch is needed.
--- include/uapi/linux/io_uring.h | 3 ++- io_uring/io_uring.c | 6 +++++- 2 files changed, 7 insertions(+), 2 deletions(-)
include/uapi/linux/io_uring.h +2 -1
@@ -619,7 +619,8 @@ struct io_uring_buf_reg {/* argument for IORING_(UN)REGISTER_NAPI */ struct io_uring_napi {__u32 busy_poll_to; -__u32 pad; +__u8 prefer_busy_poll; +__u8 pad[3 ];__u64 resv; };
io_uring/io_uring.c +5 -1
@@ -4127,15 +4127,17 @@ static int io_register_napi(struct io_ring_ctx *ctx, void __user *arg)#ifdef CONFIG_NET_RX_BUSY_POLL const struct io_uring_napi curr = { .busy_poll_to = ctx->napi_busy_poll_to, + .prefer_busy_poll = ctx->napi_prefer_busy_poll };struct io_uring_napi napi;if (copy_from_user (&napi, arg,sizeof (napi)))return -EFAULT ; -if (napi.pad || napi.resv) +if (napi.pad[0 ] || napi.pad[1 ] || napi.pad[2 ] || napi.resv)return -EINVAL ;WRITE_ONCE (ctx->napi_busy_poll_to, napi.busy_poll_to); +WRITE_ONCE (ctx->napi_prefer_busy_poll, !!napi.prefer_busy_poll);if (copy_to_user (arg, &curr,sizeof (curr)))return -EFAULT ;
@@ -4151,12 +4153,14 @@ static int io_unregister_napi(struct io_ring_ctx *ctx, void __user *arg)#ifdef CONFIG_NET_RX_BUSY_POLL const struct io_uring_napi curr = { .busy_poll_to = ctx->napi_busy_poll_to, + .prefer_busy_poll = ctx->napi_prefer_busy_poll };if (copy_to_user (arg, &curr,sizeof (curr)))return -EFAULT ;WRITE_ONCE (ctx->napi_busy_poll_to,0 ); +WRITE_ONCE (ctx->napi_prefer_busy_poll, false);return 0 ;#else return -EINVAL ;
Signature · 1 line
2.30.2
PATCH v5 1/3 io_uring: add napi busy polling support
In reply to Stefan RoeschThis adds the napi busy polling support in io_uring.c. It adds a new napi_list to the io_ring_ctx structure. This list contains the list of napi_id's that are currently enabled for busy polling. The list is synchronized by the new napi_lock spin lock. The current default napi busy polling time is stored in napi_busy_poll_to. If napi busy polling is not enabled, the value is 0.
The NAPI_TIMEOUT is stored as a timeout to make sure that the time a napi entry is stored in the napi list is limited.
The busy poll timeout is also stored as part of the io_wait_queue. This is necessary as for sq polling the poll interval needs to be adjusted and the napi callback allows only to pass in one value.
This has been tested with two simple programs from the liburing library
repository: the napi client and the napi server program. The client sends a request, which has a timestamp in its payload and the server replies with the same payload. The client calculates the roundtrip time and stores it to calcualte the results.
The client is running on host1 and the server is running on host 2 (in the same rack). The measured times below are roundtrip times. They are average times over 5 runs each. Each run measures 1 million roundtrips.
no rx coal rx coal: frames=88,usecs=33
Default 57us 56us
client_poll=100us 47us 46us
server_poll=100us 51us 46us
client_poll=100us+ 40us 40us
server_poll=100us
client_poll=100us+ 41us 39us
server_poll=100us+
prefer napi busy poll on client
client_poll=100us+ 41us 39us
server_poll=100us+
prefer napi busy poll on server
client_poll=100us+ 41us 39us
server_poll=100us+
prefer napi busy poll on client + server
--- include/linux/io_uring_types.h | 8 ++ io_uring/io_uring.c | 245 +++++++++++++++++++++++++++++++++ io_uring/napi.h | 22 +++ io_uring/poll.c | 3 + io_uring/sqpoll.c | 10 ++ 5 files changed, 288 insertions(+) create mode 100644 io_uring/napi.h
include/linux/io_uring_types.h +8 -0
@@ -270,6 +270,14 @@ struct io_ring_ctx {struct xarray personalities;u32 pers_next; +#ifdef CONFIG_NET_RX_BUSY_POLL +struct list_head napi_list;/* track busy poll napi_id */ +spinlock_t napi_lock;/* napi_list lock */ + +unsigned int napi_busy_poll_to;/* napi busy poll default timeout */ +bool napi_prefer_busy_poll; +#endif +struct {/* * We cache a range of free CQEs we can use, once exhausted it
io_uring/io_uring.c +245 -0
299 lines not on this page · raw message
io_uring/napi.h new file +22 -0
@@ -0,0 +1,22 @@ +/* SPDX-License-Identifier: GPL-2.0 */ + +#ifndef IOU_NAPI_H +#define IOU_NAPI_H + +#include <linux/kernel.h> +#include <linux/io_uring.h> +#include <net/busy_poll.h> + +#ifdef CONFIG_NET_RX_BUSY_POLL + +void io_napi_add (struct file *file,struct io_ring_ctx *ctx); +bool io_napi_busy_loop (struct list_head *napi_list,bool prefer_busy_poll); + +#else + +static inline void io_napi_add (struct file *file,struct io_ring_ctx *ctx) +{ +} + +#endif +#endif
io_uring/poll.c +3 -0
@@ -15,6 +15,7 @@#include "io_uring.h" #include "refs.h" +#include "napi.h" #include "opdef.h" #include "kbuf.h" #include "poll.h"
@@ -259,6 +260,7 @@ static int io_poll_check_events(struct io_kiocb *req, bool *locked)io_req_set_res (req, mask,0 );return IOU_POLL_REMOVE_POLL_USE_RES ; } +io_napi_add (req->file, ctx); }else { ret =io_poll_issue (req, locked);if (ret ==IOU_STOP_MULTISHOT )
@@ -583,6 +585,7 @@ static int __io_arm_poll_handler(struct io_kiocb *req,__io_poll_execute (req, mask);return 0 ; } +io_napi_add (req->file, req->ctx);if (ipt->owning) {/*
io_uring/sqpoll.c +10 -0
@@ -15,6 +15,7 @@#include <uapi/linux/io_uring.h> #include "io_uring.h" +#include "napi.h" #include "sqpoll.h" #define IORING_SQPOLL_CAP_ENTRIES_VALUE 8
@@ -193,6 +194,15 @@ static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries) ret =io_submit_sqes (ctx, to_submit);mutex_unlock (&ctx->uring_lock); +#ifdef CONFIG_NET_RX_BUSY_POLL +spin_lock (&ctx->napi_lock); +if (!list_empty (&ctx->napi_list) && +READ_ONCE (ctx->napi_busy_poll_to) >0 && +io_napi_busy_loop (&ctx->napi_list, ctx->napi_prefer_busy_poll)) + ++ret; +spin_unlock (&ctx->napi_lock); +#endif +if (to_submit &&wq_has_sleeper (&ctx->sqo_sq_wait))wake_up (&ctx->sqo_sq_wait);if (creds)
Signature · 1 line
2.30.2
PATCH v5 2/3 io_uring: add api to set / get napi configuration.
In reply to Stefan RoeschThis adds an api to register the busy poll timeout from liburing. To be able to use this functionality, the corresponding liburing patch is needed.
--- include/uapi/linux/io_uring.h | 11 ++++++++ io_uring/io_uring.c | 53 +++++++++++++++++++++++++++++++++++ 2 files changed, 64 insertions(+)
include/uapi/linux/io_uring.h +11 -0
@@ -490,6 +490,10 @@ enum {/* register a range of fixed file slots for automatic slot allocation */ IORING_REGISTER_FILE_ALLOC_RANGE =25 , +/* set/clear busy poll settings */ +IORING_REGISTER_NAPI =26 , +IORING_UNREGISTER_NAPI =27 , +/* this goes last */ IORING_REGISTER_LAST };
@@ -612,6 +616,13 @@ struct io_uring_buf_reg {__u64 resv[3 ]; }; +/* argument for IORING_(UN)REGISTER_NAPI */ +struct io_uring_napi { +__u32 busy_poll_to; +__u32 pad; +__u64 resv; +}; +/* * io_uring_restriction->opcode values */
io_uring/io_uring.c +53 -0
65 lines not on this page · raw message
Signature · 1 line
2.30.2
Jens Axboe
maintainerQuoted 40 lines · Stefan Roesch
io_uring/io_uring.c +40
-0+/* +* io_napi_add() - Add napi id to the busy poll list +* @file: file pointer for socket +* @ctx: io-uring context +* +* Add the napi id of the socket to the napi busy poll list. +*/ +void io_napi_add (struct file *file,struct io_ring_ctx *ctx) +{ +unsigned int napi_id; +struct socket *sock; +struct sock *sk; +struct io_napi_entry *ne; + +if (!io_napi_busy_loop_on (ctx)) +return ; + + sock =sock_from_file (file); +if (!sock) +return ; + + sk = sock->sk; +if (!sk) +return ; + + napi_id =READ_ONCE (sk->sk_napi_id); + +/* Non-NAPI IDs can be rejected */ +if (napi_id <MIN_NAPI_ID ) +return ; + +spin_lock (&ctx->napi_lock); +list_for_each_entry (ne, &ctx->napi_list, list) { +if (ne->napi_id == napi_id) { + ne->timeout = jiffies +NAPI_TIMEOUT ; + goto out; + } + } + + ne =kmalloc (sizeof (*ne),GFP_NOWAIT );
io_uring/io_uring.c static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx, · line 2461
Quoted 10 lines
io_uring/io_uring.c +10
-0+if (!ne) + goto out; + + ne->napi_id = napi_id; + ne->timeout = jiffies +NAPI_TIMEOUT ; +list_add_tail (&ne->list, &ctx->napi_list); + +out: +spin_unlock (&ctx->napi_lock); +}
I think this all looks good now, just one minor comment on the above. Is the expectation here that we'll basically always add to the napi list? If so, then I think allocating 'ne' outside the spinlock would be a lot saner, and then just kfree() it for the unlikely case where we find a duplicate.
Signature · 1 line
Jens Axboe
Jens Axboe
maintainerRe: [PATCH v5 2/3] io_uring: add api to set / get napi configuration.
In reply to Stefan RoeschQuoted 6 lines · Stefan Roesch
io_uring/io_uring.c +6
-0+static int io_unregister_napi (struct io_ring_ctx *ctx,void __user *arg) +{ +#ifdef CONFIG_NET_RX_BUSY_POLL +const struct io_uring_napi curr = { + .busy_poll_to = ctx->napi_busy_poll_to, + };
io_uring/io_uring.c static __cold int io_register_iowq_max_workers(struct io_ring_ctx *ctx, · line 4164
Quoted 10 lines
io_uring/io_uring.c +10
-0+ +if (copy_to_user (arg, &curr,sizeof (curr))) +return -EFAULT ; + +WRITE_ONCE (ctx->napi_busy_poll_to,0 ); +return 0 ; +#else +return -EINVAL ; +#endif +}
Should probably check resv/pad here as well, maybe even the 'busy_poll_to' being zero?
Signature · 1 line
Jens Axboe
Jens Axboe
maintainerQuoted 49 lines · Jens Axboe
Quoted 48 lines · Stefan Roesch
io_uring/io_uring.c +48
-0+/* +* io_napi_add() - Add napi id to the busy poll list +* @file: file pointer for socket +* @ctx: io-uring context +* +* Add the napi id of the socket to the napi busy poll list. +*/ +void io_napi_add (struct file *file,struct io_ring_ctx *ctx) +{ +unsigned int napi_id; +struct socket *sock; +struct sock *sk; +struct io_napi_entry *ne; + +if (!io_napi_busy_loop_on (ctx)) +return ; + + sock =sock_from_file (file); +if (!sock) +return ; + + sk = sock->sk; +if (!sk) +return ; + + napi_id =READ_ONCE (sk->sk_napi_id); + +/* Non-NAPI IDs can be rejected */ +if (napi_id <MIN_NAPI_ID ) +return ; + +spin_lock (&ctx->napi_lock); +list_for_each_entry (ne, &ctx->napi_list, list) { +if (ne->napi_id == napi_id) { + ne->timeout = jiffies +NAPI_TIMEOUT ; + goto out; + } + } + + ne =kmalloc (sizeof (*ne),GFP_NOWAIT ); +if (!ne) + goto out; + + ne->napi_id = napi_id; + ne->timeout = jiffies +NAPI_TIMEOUT ; +list_add_tail (&ne->list, &ctx->napi_list); + +out:
Quoted 8 lines
Review
Quoted 2 lines
(quoted diff) +2
-0+ spin_unlock(&ctx->napi_lock); +}I think this all looks good now, just one minor comment on the above. Is the expectation here that we'll basically always add to the napi list? If so, then I think allocating 'ne' outside the spinlock would be a lot saner, and then just
kfree()it for the unlikely case where we find a duplicate.
After thinking about this a bit more, I don't think this is done in the most optimal fashion. If the list is longer than a few entries, this check (or check-alloc-insert) is pretty expensive and it'll add substantial overhead to the poll path for sockets if napi is enabled.
I think we should do something ala:
1) When arming poll AND napi has been enabled for the ring, then
alloc io_napi_entry upfront and store it in ->async_data.
2) Maintain the state in the io_napi_entry. If we're on the list, that can be checked with just list_empty(), for example. If not on the list, assign timeout and add.
3) Have regular request cleanup free it.
This could be combined with an alloc cache, I would not do that for the first iteration though.
This would make io_napi_add() cheap - no more list iteration, and no more allocations. And that is arguably the most important part, as that is called everytime the poll is woken up. Particularly for multishot that makes a big difference.
It's also designed much better imho, moving the more expensive bits to the setup side.
Signature · 1 line
Jens Axboe
Re: [PATCH v5 2/3] io_uring: add api to set / get napi configuration.
In reply to Jens AxboeQuoted 12 lines · Jens Axboe
Quoted 11 lines · Stefan Roesch
io_uring/io_uring.c +11
-0+static int io_unregister_napi (struct io_ring_ctx *ctx,void __user *arg) +{ +#ifdef CONFIG_NET_RX_BUSY_POLL +const struct io_uring_napi curr = { + .busy_poll_to = ctx->napi_busy_poll_to, + }; + +if (copy_to_user (arg, &curr,sizeof (curr))) +return -EFAULT ; + +WRITE_ONCE (ctx->napi_busy_poll_to,0 );
Quoted 8 lines
io_uring/io_uring.c static __cold int io_register_iowq_max_workers(struct io_ring_ctx *ctx, · line 4147
Quoted 5 lines
io_uring/io_uring.c +5
-0+return 0 ; +#else +return -EINVAL ; +#endif +}Should probably check resv/pad here as well, maybe even the '
busy_poll_to' being zero?
Jens, this function doesn't read from __user memory, it writes to __user memory.
@curr.resv and @curr.pad are on the kernel's stack. Both are already implicitly initialized to zero by the partial struct initializer.
Signature · 1 line
Ammar Faizi
Jens Axboe
maintainerRe: [PATCH v5 2/3] io_uring: add api to set / get napi configuration.
Quoted 22 lines · Ammar Faizi
Quoted 12 lines · Jens Axboe
Quoted 11 lines · Stefan Roesch
io_uring/io_uring.c +11
-0+static int io_unregister_napi (struct io_ring_ctx *ctx,void __user *arg) +{ +#ifdef CONFIG_NET_RX_BUSY_POLL + const struct io_uring_napi curr = { + .busy_poll_to = ctx->napi_busy_poll_to, + }; + + if (copy_to_user(arg, &curr, sizeof(curr))) + return -EFAULT; + + WRITE_ONCE(ctx->napi_busy_poll_to, 0);Quoted 8 lines
io_uring/io_uring.c static __cold int io_register_iowq_max_workers(struct io_ring_ctx *ctx, · line 4147
Quoted 5 lines
+ return 0;
io_uring/io_uring.c +4
-0+#else + return -EINVAL; +#endif +}Should probably check resv/pad here as well, maybe even the '
busy_poll_to' being zero?
Quoted 5 lines
Jens, this function doesn't read from
__usermemory, it writes to__usermemory.@curr.resv and @curr.pad are on the kernel's stack. Both are already implicitly initialized to zero by the partial struct initializer.
Oh yes, guess I totally missed that we don't care about the value at all (just zero the target) and copy back the old values.
Signature · 1 line
Jens Axboe
Re: [PATCH v5 2/3] io_uring: add api to set / get napi configuration.
In reply to Stefan RoeschQuoted 7 lines · Stefan Roesch
io_uring/io_uring.c +7
-0+static int io_unregister_napi (struct io_ring_ctx *ctx,void __user *arg) +{ +#ifdef CONFIG_NET_RX_BUSY_POLL +const struct io_uring_napi curr = { + .busy_poll_to = ctx->napi_busy_poll_to, + }; +
io_uring/io_uring.c static __cold int io_register_iowq_max_workers(struct io_ring_ctx *ctx, · line 4165
Quoted 10 lines
io_uring/io_uring.c +10
-0+if (copy_to_user (arg, &curr,sizeof (curr))) +return -EFAULT ; + +WRITE_ONCE (ctx->napi_busy_poll_to,0 ); +return 0 ; +#else +return -EINVAL ; +#endif +} +
I suggest allowing users to pass a NULL as the arg in case they don't want to care about the old values.
Something like:
io_uring_unregister_napi(ring, NULL);
What do you think?
Signature · 1 line
Ammar Faizi
Re: [PATCH v5 2/3] io_uring: add api to set / get napi configuration.
Quoted 21 lines · Ammar Faizi
Quoted 7 lines · Stefan Roesch
io_uring/io_uring.c +7
-0+static int io_unregister_napi (struct io_ring_ctx *ctx,void __user *arg) +{ +#ifdef CONFIG_NET_RX_BUSY_POLL +const struct io_uring_napi curr = { + .busy_poll_to = ctx->napi_busy_poll_to, + }; +io_uring/io_uring.c static __cold int io_register_iowq_max_workers(struct io_ring_ctx *ctx, · line 4165
Quoted 10 lines
io_uring/io_uring.c +10
-0+if (copy_to_user (arg, &curr,sizeof (curr))) +return -EFAULT ; + +WRITE_ONCE (ctx->napi_busy_poll_to,0 ); +return 0 ; +#else +return -EINVAL ; +#endif +} +I suggest allowing users to pass a NULL as the arg in case they don't want to care about the old values.
Quoted 5 lines
Something like:
io_uring_unregister_napi(ring, NULL);What do you think?
Sounds good, i can make that change in the next version.