Arm Exploring IO_uring For Graphics Drivers For Better Performance & Synchronization

[phoronix]

Phoronix: Arm Exploring IO_uring For Graphics Drivers For Better Performance & Synchronization

The IO_uring asynchronous I/O API for Linux is quite novel and has proven performance benefits. With time IO_uring has been adapted to other areas of the kernel like networking and now with a proposal raised by an Arm graphics driver engineer, it could potentially be adapted for use by Direct Rendering Manager (DRM) kernel graphics drivers...

https://www.phoronix.com/news/DRM-Graphics-Drivers-IO_uring

[davidbepo]

huh, that was definitely not in my bingo, but very neat

[Vermilion]

The PDF slides link is broken/missing

Edit: https://indico.freedesktop.org/event...ringforDRM.pdf

[daPramDeech]

pdf link https://indico.freedesktop.org/event...ringforDRM.pdf

[discordian]

unlike io drivers where blocks of data are just copied somewhere else, i consider this pretty interesting to access complex structures containing pointers from kernel-code that could be modified/corrupted by userspace at any time...
in comparison BPF has several sanity checks and the final (byte)-code is private to the kernel.

edit: watched the discussion, it seems more akin to passing filedescriptors to the queue. no big issues then... but i dont think that will help performance much?

[aviallon]

IO_uring go brrrr

[coder]

Does this mean we're now going to have a string of articles about some Axboe-type achieving ever greater numbers of draw calls per second?
🤩

[zexelon]

Given my personal experience with IO_uring on the file systems we had it going on... nothing but bad will happen here. We had to disable IO_uring in the end to keep the file system driver from crapping out in production.

Nothing ever corrupted (thankfully) but we could not get it run stably (past a few weeks at a time).

[Mark Rose]

Given my personal experience with IO_uring on the file systems we had it going on... nothing but bad will happen here. We had to disable IO_uring in the end to keep the file system driver from crapping out in production.

Nothing ever corrupted (thankfully) but we could not get it run stably (past a few weeks at a time).

Have you reported the bug?

[oleid]

Given my personal experience with IO_uring on the file systems we had it going on... nothing but bad will happen here. We had to disable IO_uring in the end to keep the file system driver from crapping out in production.

Nothing ever corrupted (thankfully) but we could not get it run stably (past a few weeks at a time).

Interesting. What programming language/library/framework combination did you use? Was it bare io_uring?

I was wondering because depending on what you're using, you don't have to program io_uring directly, just use something that uses io_uring under the hood. That is usually quite easy.