Re: [Qemu-devel] [RFC] Replace posix-aio with custom thread pool

[Anthony Liguori <anthony@codemonkey.ws>]

Andrea Arcangeli wrote:
> On Fri, Dec 12, 2008 at 08:35:57AM -0600, Anthony Liguori wrote:
>   
>> I've been thinking about this, the problems I see are:
>>
>> 1) It's impossible to accept a file descriptor for a block device (possibly 
>> not a problem)
>>     
>
> What do you mean with accept? You mean to accept a tcp connection? How
> would a block device fd be related to accept(2)?
>   

I meant, if you wanted to pass a file descriptor as a raw device.  So:

qemu -hda raw:fd=4

Or something like that.  We don't support this today.

> Now that linux-aio is out of the picture for quite a long time for us,
> I guess it worth to wait preadv/pwritev and stick with that and
> reconsider linux-aio after they fix it... Waiting Gerd to post a full
> patch.
>
> But it's your call... I'm fine either ways. Clearly the os missing
> preadv/pwritev would need to be limited to 1 thread per fd (but 1
> thread per fd kind of breaks with the current _global_ list so I guess
> they'll be limited to just 1 thread otherwise it may be actually
> simpler to just open the file multiple times than to have a per-fd
> queue ;), not the end of the world for them.
>   

I think bouncing the iov and just using pread/pwrite may be our best 
bet.  It means memory allocation but we can cap it.  Since we're using 
threads, we just can force a thread to sleep until memory becomes 
available so it's actually pretty straight forward.

We can use libaio on older Linux's to simulate preadv/pwritev.  Use the 
proper syscalls on newer kernels, on BSDs, and bounce everything else.

Regards,

Anthony Liguori