Submitting IO events with IORING_FEAT_NODROP and overflown CQ ring

Submitting IO events with IORING_FEAT_NODROP and overflown CQ ring

[Alexander Babayants]

Hi!

Per io_uring_setup(2):

> IORING_FEAT_NODROP
...
> If this overflow condition is entered,
> attempting to submit more IO will fail with the -EBUSY
> error value, if it can't flush the overflown events to the
> CQ ring. If this happens, the application must reap events
> from the CQ ring and attempt the submit again.

But I'm not getting -EBUSY on a 6.2 kernel when submitting to the ring
with a full CQ ring and an overflow flag set. I'm not an expert in
uring code, but it seems the error was removed in commit
1b346e4aa8e79227391ffd6b7c6ee5acf0fa8bfc. Could you please check if
I'm right and if that change was intentional?

-- 
Regards,
Alexander Babayants.

Re: Submitting IO events with IORING_FEAT_NODROP and overflown CQ ring

[Jens Axboe]

On 5/3/24 5:21 PM, Alexander Babayants wrote:
> Hi!
> 
> Per io_uring_setup(2):
> 
>> IORING_FEAT_NODROP
> ...
>> If this overflow condition is entered,
>> attempting to submit more IO will fail with the -EBUSY
>> error value, if it can't flush the overflown events to the
>> CQ ring. If this happens, the application must reap events
>> from the CQ ring and attempt the submit again.
> 
> But I'm not getting -EBUSY on a 6.2 kernel when submitting to the ring
> with a full CQ ring and an overflow flag set. I'm not an expert in
> uring code, but it seems the error was removed in commit
> 1b346e4aa8e79227391ffd6b7c6ee5acf0fa8bfc. Could you please check if
> I'm right and if that change was intentional?

This is correct and intentional. The man page should get updated with a
note on the current behavior.

-- 
Jens Axboe