Re: [Qemu-devel] [RFC 1/2] pci-dma-api-v1

[Anthony Liguori <anthony@codemonkey.ws>]

Andrea Arcangeli wrote:
> On Fri, Nov 28, 2008 at 07:59:13PM +0200, Blue Swirl wrote:
>   
>> I don't know, here's a pointer:
>> http://lists.gnu.org/archive/html/qemu-devel/2008-08/msg00092.html
>>     
>
> I'm in total agreement with it. The missing "proper vectored AIO
> operations" are bdrv_aio_readv/writev ;).
>
> I wonder how can possibly aio_readv/writev be missing in posix aio?
>   

Because it sucks.

> Unbelievable. It'd be totally trivial to add those to glibc, much
> easier infact than to pthread_create by hand, but how can we add a
> dependency on a certain glibc version?

I've already asked, Ulrich will not likely add this unless it first goes 
into the Posix standard.  It won't go into the Posix standard without an 
existing implementation.  Either way, we're talking about a multi-year 
process.

We need our own thread-pool based AIO implementation.

>  Ironically it'll be more
> user-friendly to add dependency on linux kernel-aio implementation
> that is already available for ages and it's guaranteed to run faster
> (or at least not slower).
>   

No, kernel-aio is more brain dead.  io_submit may block if the 
underlying file descriptor does not support asynchronous IO.  There is 
no way to detect this.  This means the only safe way to use io_submit is 
to call it from a thread, and to have a thread pool that can do multiple 
io_submits in parallel.

Yes, that's ridiculous.  There's a thread right now on 
linux-aio@vger.kernel.org discussing the future of linux-aio.  The 
prevailing wisdom seems to be that linux-aio does not have a future.

Regards,

Anthony Liguori