Re: [PATCHSET] Add support for simplified async direct-io

[Jens Axboe <axboe@fb.com>]

On 11/14/2016 11:11 AM, Christoph Hellwig wrote:
> On Mon, Nov 14, 2016 at 11:08:46AM -0700, Jens Axboe wrote:
>> It'd be cleaner to loop one level out, and avoid all that 'dio' stuff
>> instead. And then still retain the separate parts of the sync and async.
>> There's nothing to share there imho, and it just makes the code harder
>> to read.
>
> How do you avoid it for the async case?  We can only call ki_complete
> once all bios have finished, which means we need a tracking structure
> for it.  For the synchronous case we could in theory wait for the
> previous bio before sending the next, but there are plenty of RAID
> arrays that would prefer > 1MB I/O.  And we can pretty much reuse the
> async case for this anyway.

Just make the limit reasonable - with a order 1 alloc, we have 256 
pages, or 1MB of IO. Loop around that. We could play tricks with bio 
reuse and using the ref count for knowing when it'd be done.

-- 
Jens Axboe