Re: [RFC][WIP] DIO simplification and AIO-DIO stability

[Chris Mason <mason@suse.com>]

On Thursday 23 February 2006 02:29, Suparna Bhattacharya wrote:
> DIO code complexity and stability concerns were discussed way back during
> OLS and Kernel summit last year. Still, the lack of a solid alternative and
> motivation to subject oneself to the test of courage and delicate balance
> that fiddling with this code entails, has meant that gingerly applying
> fixes and bandaids as and when bugs are found, and moving on thereafter,
> continues to be the most palatable option.
>
> A recent AIO-DIO bug reported by Kenneth Chen, came very close
> to being the proverbial last straw for me. Hence, here is a rough attempt
> to put together a (currently WIP) draft towards DIO code simplication,
> based on suggestions that some of you have brought up at various times.
> Several details, e.g. range locking implementation still need to be fleshed
> out completely, ideas/comments/suggestions would be welcome.

I'm really in favor of this, and had actually started an implementation a 
while back.  At the time, I posted a different version that added yet another 
semaphore but simplified the rest of the locking (and held no locks during 
the dio/aio).

I'll try to dig up my original radix tagging code.  I'm not sure if I kept it, 
but it did pass Daniel's dio vs buffer io racing tests at the time.

-chris