io_uring: register single issuer task at creation
v2 cover io_uring: register single issuer task at creation
Registering the single issuer task from the first submit adds unnecesary complications to the API as well as the implementation. Where simply registering it at creation should not impose any barriers to getting the same performance wins. The only catch is users that might want to move the ring after creation but before submission. For these users allow them to create the ring with IORING_SETUP_R_DISABLED and then enable it on the submission task.
There is another problem in 6.1, with IORING_SETUP_DEFER_TASKRUN. That would like to check the submitter_task from unlocked contexts, which would be racy. If upfront the submitter_task is set at creation time it will simplify the logic there and probably increase performance (though this is unmeasured).
Patch 1 registers the task at creation of the io_uring, this works standalone in case you want to only merge this part for 6.0
Patch 2/3 cleans up the code from the old style
v2:
- add the IORING_SETUP_R_DISABLED logic
Dylan Yudaken (3):
io_uring: register single issuer task at creation
io_uring: simplify __io_uring_add_tctx_node
io_uring: remove io_register_submitter
io_uring/io_uring.c | 9 ++++++++- io_uring/tctx.c | 42 ++++++++++++++++++------------------------ io_uring/tctx.h | 6 ++++-- 3 files changed, 30 insertions(+), 27 deletions(-)
base-commit: f76349cf41451c5c42a99f18a9163377e4b364ff
Signature · 1 line
2.30.2
PATCH v2 1/3 io_uring: register single issuer task at creation
Instead of picking the task from the first submitter task, rather use the creator task or in the case of disabled (IORING_SETUP_R_DISABLED) the enabling task.
This approach allows a lot of simplification of the logic here. This removes init logic from the submission path, which can always be a bit confusing, but also removes the need for locking to write (or read) the submitter_task.
Users that want to move a ring before submitting can create the ring disabled and then enable it on the submitting task.
--- io_uring/io_uring.c | 7 +++++++ 1 file changed, 7 insertions(+)
io_uring/io_uring.c +7 -0
@@ -3357,6 +3357,10 @@ static __cold int io_uring_create(unsigned entries, struct io_uring_params *p, goto err; } +if (ctx->flags &IORING_SETUP_SINGLE_ISSUER + && !(ctx->flags &IORING_SETUP_R_DISABLED )) + ctx->submitter_task =get_task_struct (current); + file =io_uring_get_file (ctx);if (IS_ERR (file)) { ret =PTR_ERR (file);
@@ -3548,6 +3552,9 @@ static int io_register_enable_rings(struct io_ring_ctx *ctx)if (!(ctx->flags &IORING_SETUP_R_DISABLED ))return -EBADFD ; +if (ctx->flags &IORING_SETUP_SINGLE_ISSUER && !ctx->submitter_task) + ctx->submitter_task =get_task_struct (current); +if (ctx->restrictions.registered) ctx->restricted =1 ;
Signature · 1 line
2.30.2
Quoted 11 lines · Dylan Yudaken
Instead of picking the task from the first submitter task, rather use the creator task or in the case of disabled (
IORING_SETUP_R_DISABLED) the enabling task.This approach allows a lot of simplification of the logic here. This removes init logic from the submission path, which can always be a bit confusing, but also removes the need for locking to write (or read) the
submitter_task.Users that want to move a ring before submitting can create the ring disabled and then enable it on the submitting task.
I think Dylan briefly mentioned before that it might be a good idea to task limit registration as well. I can't think of a use case at the moment but I agree we may find some in the future.
io_uring/io_uring.c +3 -0
@@ -3706,6 +3706,9 @@ static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode,if (WARN_ON_ONCE (percpu_ref_is_dying (&ctx->refs)))return -ENXIO ; +if (ctx->submitter_task && ctx->submitter_task != current) +return -EEXIST ; +if (ctx->restricted) {if (opcode >=IORING_REGISTER_LAST )return -EINVAL ;
Signature · 1 line
Pavel Begunkov
Jens Axboe
maintainerQuoted 25 lines · Pavel Begunkov
Quoted 11 lines · Dylan Yudaken
Instead of picking the task from the first submitter task, rather use the creator task or in the case of disabled (
IORING_SETUP_R_DISABLED) the enabling task.This approach allows a lot of simplification of the logic here. This removes init logic from the submission path, which can always be a bit confusing, but also removes the need for locking to write (or read) the
submitter_task.Users that want to move a ring before submitting can create the ring disabled and then enable it on the submitting task.
I think Dylan briefly mentioned before that it might be a good idea to task limit registration as well. I can't think of a use case at the moment but I agree we may find some in the future.
io_uring/io_uring.c +0
-0@@ -3706,6 +3706,9 @@ static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, if (WARN_ON_ONCE(percpu_ref_is_dying(&ctx->refs))) return -ENXIO;
Quoted 7 lines
+ if (ctx->submitter_task&&ctx->submitter_task!= current) + return-EEXIST;
+
if (ctx->restricted) {
if (opcode >=IORING_REGISTER_LAST)
return-EINVAL;
Yes, I don't see any reason why not to enforce this for registration too. Don't think there's currently a need to do so, but it'd be easy to miss once we do add that. Let's queue that up for 6.1?
Signature · 1 line
Jens Axboe
Quoted 27 lines · Jens Axboe
Quoted 26 lines · Pavel Begunkov
Quoted 11 lines · Dylan Yudaken
Instead of picking the task from the first submitter task, rather use the creator task or in the case of disabled (
IORING_SETUP_R_DISABLED) the enabling task.This approach allows a lot of simplification of the logic here. This removes init logic from the submission path, which can always be a bit confusing, but also removes the need for locking to write (or read) the
submitter_task.Users that want to move a ring before submitting can create the ring disabled and then enable it on the submitting task.
I think Dylan briefly mentioned before that it might be a good idea to task limit registration as well. I can't think of a use case at the moment but I agree we may find some in the future.
io_uring/io_uring.c +0
-0@@ -3706,6 +3706,9 @@ static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, if (WARN_ON_ONCE(percpu_ref_is_dying(&ctx->refs))) return -ENXIO;
Quoted 10 lines
io_uring/io_uring.c static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, · line 3714
Quoted 6 lines
io_uring/io_uring.c +3
-0+ if (ctx->submitter_task && ctx->submitter_task != current) + return -EEXIST; + if (ctx->restricted) { if (opcode >= IORING_REGISTER_LAST) return -EINVAL;Yes, I don't see any reason why not to enforce this for registration too. Don't think there's currently a need to do so, but it'd be easy to miss once we do add that. Let's queue that up for 6.1?
6.1 + stable sounds ok, I don't have an opinion on how to how to merge it.
Signature · 1 line
Pavel Begunkov
Jens Axboe
maintainerQuoted 38 lines · Pavel Begunkov
Quoted 28 lines · Jens Axboe
Quoted 27 lines · Pavel Begunkov
Quoted 11 lines · Dylan Yudaken
Instead of picking the task from the first submitter task, rather use the creator task or in the case of disabled (
IORING_SETUP_R_DISABLED) the enabling task.This approach allows a lot of simplification of the logic here. This removes init logic from the submission path, which can always be a bit confusing, but also removes the need for locking to write (or read) the
submitter_task.Users that want to move a ring before submitting can create the ring disabled and then enable it on the submitting task.
I think Dylan briefly mentioned before that it might be a good idea to task limit registration as well. I can't think of a use case at the moment but I agree we may find some in the future.
io_uring/io_uring.c +1
-0@@ -3706,6 +3706,9 @@ static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, if (WARN_ON_ONCE(percpu_ref_is_dying(&ctx->refs))) return -ENXIO; + if (ctx->submitter_task && ctx->submitter_task != current) + return -EEXIST;Quoted 8 lines
io_uring/io_uring.c static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, · line 3714
Quoted 4 lines
io_uring/io_uring.c +1
-0+ if (ctx->restricted) { if (opcode >= IORING_REGISTER_LAST) return -EINVAL;Yes, I don't see any reason why not to enforce this for registration too. Don't think there's currently a need to do so, but it'd be easy to miss once we do add that. Let's queue that up for 6.1?
Quoted 2 lines
6.1 + stable sounds ok, I don't have an opinion on how to how to merge it.
That's the plan. If you can just send it out as a separate commit, I'll stage it up behind the two others from Dylan.
Signature · 1 line
Jens Axboe
PATCH v2 2/3 io_uring: simplify __io_uring_add_tctx_node
In reply to Dylan YudakenRemove submitter parameter from __io_uring_add_tctx_node.
It was only called from one place, and we can do that logic in that one place.
--- io_uring/io_uring.c | 2 +- io_uring/tctx.c | 30 ++++++++++++++++++++---------- io_uring/tctx.h | 6 ++++-- 3 files changed, 25 insertions(+), 13 deletions(-)
io_uring/io_uring.c +1 -1
@@ -3183,7 +3183,7 @@ static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file)if (fd <0 )return fd; - ret =__io_uring_add_tctx_node (ctx, false); + ret =__io_uring_add_tctx_node (ctx);if (ret) {put_unused_fd (fd);return ret;
io_uring/tctx.c +20 -10
@@ -105,18 +105,12 @@ static int io_register_submitter(struct io_ring_ctx *ctx)return ret; } -int __io_uring_add_tctx_node (struct io_ring_ctx *ctx,bool submitter) +int __io_uring_add_tctx_node (struct io_ring_ctx *ctx) {struct io_uring_task *tctx = current->io_uring;struct io_tctx_node *node;int ret; -if ((ctx->flags &IORING_SETUP_SINGLE_ISSUER ) && submitter) { - ret =io_register_submitter (ctx); -if (ret) -return ret; - } -if (unlikely (!tctx)) { ret =io_uring_alloc_task_context (current, ctx);if (unlikely (ret))
@@ -150,8 +144,24 @@ int __io_uring_add_tctx_node(struct io_ring_ctx *ctx, bool submitter)list_add (&node->ctx_node, &ctx->tctx_list);mutex_unlock (&ctx->uring_lock); } -if (submitter) - tctx->last = ctx; +return 0 ; +} + +int __io_uring_add_tctx_node_from_submit (struct io_ring_ctx *ctx) +{ +int ret; + +if (ctx->flags &IORING_SETUP_SINGLE_ISSUER ) { + ret =io_register_submitter (ctx); +if (ret) +return ret; + } + + ret =__io_uring_add_tctx_node (ctx); +if (ret) +return ret; + + current->io_uring->last = ctx;return 0 ; }
@@ -259,7 +269,7 @@ int io_ringfd_register(struct io_ring_ctx *ctx, void __user *__arg,return -EINVAL ;mutex_unlock (&ctx->uring_lock); - ret =__io_uring_add_tctx_node (ctx, false); + ret =__io_uring_add_tctx_node (ctx);mutex_lock (&ctx->uring_lock);if (ret)return ret;
io_uring/tctx.h +4 -2
@@ -9,7 +9,8 @@ struct io_tctx_node {int io_uring_alloc_task_context (struct task_struct *task,struct io_ring_ctx *ctx);void io_uring_del_tctx_node (unsigned long index); -int __io_uring_add_tctx_node (struct io_ring_ctx *ctx,bool submitter); +int __io_uring_add_tctx_node (struct io_ring_ctx *ctx); +int __io_uring_add_tctx_node_from_submit (struct io_ring_ctx *ctx);void io_uring_clean_tctx (struct io_uring_task *tctx);void io_uring_unreg_ringfd (void );
@@ -27,5 +28,6 @@ static inline int io_uring_add_tctx_node(struct io_ring_ctx *ctx)if (likely (tctx && tctx->last == ctx))return 0 ; -return __io_uring_add_tctx_node (ctx, true); + +return __io_uring_add_tctx_node_from_submit (ctx); }
Signature · 1 line
2.30.2
PATCH v2 3/3 io_uring: remove io_register_submitter
In reply to Dylan Yudakenthis is no longer needed, as submitter_task is set at creation time.
--- io_uring/tctx.c | 22 +++------------------- 1 file changed, 3 insertions(+), 19 deletions(-)
io_uring/tctx.c +3 -19
@@ -91,20 +91,6 @@ __cold int io_uring_alloc_task_context(struct task_struct *task,return 0 ; } -static int io_register_submitter (struct io_ring_ctx *ctx) -{ -int ret =0 ; - -mutex_lock (&ctx->uring_lock); -if (!ctx->submitter_task) - ctx->submitter_task =get_task_struct (current); -else if (ctx->submitter_task != current) - ret = -EEXIST ; -mutex_unlock (&ctx->uring_lock); - -return ret; -} -int __io_uring_add_tctx_node (struct io_ring_ctx *ctx) {struct io_uring_task *tctx = current->io_uring;
@@ -151,11 +137,9 @@ int __io_uring_add_tctx_node_from_submit(struct io_ring_ctx *ctx) {int ret; -if (ctx->flags &IORING_SETUP_SINGLE_ISSUER ) { - ret =io_register_submitter (ctx); -if (ret) -return ret; - } +if (ctx->flags &IORING_SETUP_SINGLE_ISSUER + && ctx->submitter_task != current) +return -EEXIST ; ret =__io_uring_add_tctx_node (ctx);if (ret)
Signature · 1 line
2.30.2
Jens Axboe
maintainerQuoted 14 lines · Dylan Yudaken
Registering the single issuer task from the first submit adds unnecesary complications to the API as well as the implementation. Where simply registering it at creation should not impose any barriers to getting the same performance wins. The only catch is users that might want to move the ring after creation but before submission. For these users allow them to create the ring with
IORING_SETUP_R_DISABLEDand then enable it on the submission task.There is another problem in 6.1, with
IORING_SETUP_DEFER_TASKRUN. That would like to check thesubmitter_taskfrom unlocked contexts, which would be racy. If upfront thesubmitter_taskis set at creation time it will simplify the logic there and probably increase performance (though this is unmeasured).
Quoted 4 lines
Patch 1 registers the task at creation of the
io_uring, this works standalone in case you want to only merge this part for 6.0Patch 2/3 cleans up the code from the old style
Thanks, I like 1/3 a lot better now. Will provide applications with an easy path to use SINGLE_ISSUER, even if they currently setup the ring from a different thread/task than they end up using it from.
I've updated the 6.0 and 6.1 repos to reflect this.
Signature · 1 line
Jens Axboe