(no subject)

[Luis Gerhorst]

Hello Pavel,

I'm from a university and am searching for a project to work on in the
upcoming year. I am looking into allowing userspace to run multiple
system calls interleaved with application-specific logic using a single
context switch.

I noticed that you, Jens Axboe, and Carter Li discussed the possibility
of integrating eBPF into io_uring earlier this year [1, 2, 3]. Is there
any WIP on this topic?

If not I am considering to implement this. Besides the fact that AOT
eBPF is only supported for priviledged processes, are there any issues
you are aware of or reasons why this was not implemented yet?

Best,
Luis

[1] https://lore.kernel.org/io-uring/67b28e66-f2f8-99a1-dfd1-14f753d11f7a@gmail.com/
[2] https://lore.kernel.org/io-uring/8b3f182c-7c4b-da41-7ec8-bb4f22429ed1@kernel.dk/
[3] https://github.com/axboe/liburing/issues/58

Integrating eBPF into io-uring

[linux-kernel]

Forgot subject, sorry.

> On 6. Nov 2020, at 11:44, Luis Gerhorst <linux-kernel@luisgerhorst.de> wrote:
> 
> Hello Pavel,
> 
> I'm from a university and am searching for a project to work on in the
> upcoming year. I am looking into allowing userspace to run multiple
> system calls interleaved with application-specific logic using a single
> context switch.
> 
> I noticed that you, Jens Axboe, and Carter Li discussed the possibility
> of integrating eBPF into io_uring earlier this year [1, 2, 3]. Is there
> any WIP on this topic?
> 
> If not I am considering to implement this. Besides the fact that AOT
> eBPF is only supported for priviledged processes, are there any issues
> you are aware of or reasons why this was not implemented yet?
> 
> Best,
> Luis
> 
> [1] https://lore.kernel.org/io-uring/67b28e66-f2f8-99a1-dfd1-14f753d11f7a@gmail.com/
> [2] https://lore.kernel.org/io-uring/8b3f182c-7c4b-da41-7ec8-bb4f22429ed1@kernel.dk/
> [3] https://github.com/axboe/liburing/issues/58

Re:

[Pavel Begunkov]

On 06/11/2020 10:44, Luis Gerhorst wrote:
> Hello Pavel,
> 
> I'm from a university and am searching for a project to work on in the
> upcoming year. I am looking into allowing userspace to run multiple
> system calls interleaved with application-specific logic using a single
> context switch.
> 
> I noticed that you, Jens Axboe, and Carter Li discussed the possibility
> of integrating eBPF into io_uring earlier this year [1, 2, 3]. Is there
> any WIP on this topic?

To be honest, I've finally returned to it a week ago, just because got
more free time. I was implicitly patching/refactoring some bits keeping
this in mind but rather very lazily.

> If not I am considering to implement this. Besides the fact that AOT
> eBPF is only supported for priviledged processes, are there any issues
> you are aware of or reasons why this was not implemented yet?

All others I was anticipating are gone by now. I'd be really great to
think something out for non-privileged processes, but as you know that
doesn't hold us off.

> [1] https://lore.kernel.org/io-uring/67b28e66-f2f8-99a1-dfd1-14f753d11f7a@gmail.com/
> [2] https://lore.kernel.org/io-uring/8b3f182c-7c4b-da41-7ec8-bb4f22429ed1@kernel.dk/
> [3] https://github.com/axboe/liburing/issues/58
> 

-- 
Pavel Begunkov

Re:

[Ammar Faizi]

subscribe io-uring

Re:

[Jens Axboe]

On 2/24/20 10:37 AM, Jens Axboe wrote:
> Here's v3 of the poll async retry patchset. Changes since v2:
> 
> - Rebase on for-5.7/io_uring
> - Get rid of REQ_F_WORK bit
> - Improve the tracing additions
> - Fix linked_timeout case
> - Fully restore work from async task handler
> - Credentials now fixed
> - Fix task_works running from SQPOLL
> - Remove task cancellation stuff, we don't need it
> - fdinfo print improvements
> 
> I think this is getting pretty close to mergeable, I haven't found
> any issues with the test cases.

Gah, wrong directory, resending it. Ignore this thread.

-- 
Jens Axboe