ioctl() for io_uring

Posted Feb 5, 2021 14:12 UTC (Fri) by rhdxmr (guest, #44404)
In reply to: ioctl() for io_uring by johill
Parent article: ioctl() for io_uring

That would be awesome. io_uring supplies data to BPF program and all process is done inside kernel using BPF.
Without any syscall and copying buffer to userspace, all IO related tasks can be achieved.

ioctl() for io_uring

Posted Feb 6, 2021 20:14 UTC (Sat) by matthias (subscriber, #94967) [Link] (2 responses)

That is still not thought to the end. Clearly, BPF needs to be able to issue new io_uring calls. After all, the calls to be made might depend on the processing of the data. We should definitely get rid of all those unnecessary context switches and abandon userspace altogether ;)

ioctl() for io_uring

Posted Feb 8, 2021 8:05 UTC (Mon) by johill (subscriber, #25196) [Link] (1 responses)

Isn't Linux a kind of BPF-(micro-?)kernel with legacy userspace support? ;-)

ioctl() for io_uring

Posted Feb 8, 2021 9:31 UTC (Mon) by Sesse (subscriber, #53779) [Link]

People keep bringing up this meme, but I honestly don't understand what all the fuzz is about. I keep looking for uses for eBPF, but I really can't find them—there are some admin tools to show me slow I/Os that I've used maybe once, and that's it. Is this really such a revolution?

io_uring, on the other hand, I can perfectly well see lots of uses for (and I have used it myself).