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

[Andrew Morton <akpm@osdl.org>]

Badari Pulavarty <pbadari@us.ibm.com> wrote:
>
> I am still trying to understand the whole proposal to give you better
>  feedback. But, my gut feeling is - its not going to be any more simpler
>  than what we have today :(
> 

Yes, that's my general reaction as well.  That code's solving a complex and
messy problem, so it got complex and messy.

Of course, a reimplementation might certainly end up faster, cleaner,
better.  A throw-away-and-reimplement exercise often has that result, but
mainly because on the second time the reimplementors understand the full
scope of the problem at the outset rather than at the end.  So this time
around, as you imply, we'd need to get a full problem description and set
of testcases collected.

That code does a _lot_ of stuff.  Fortunately, It's basically all in
direct-io.c and that file exports a single function.  So it's possible that
a reimplmentation could tick along alongside the existing implementation and
ideally, it's just a matter of changing one entry in each filesystem's a_ops.