Re: [patch 13/13] signalfd/timerfd/asyncfd v5 - KAIO asyncfd support (example/maybe-broken) ...

[Davide Libenzi <davidel@xmailserver.org>]

On Wed, 14 Mar 2007, Benjamin LaHaise wrote:

> On Wed, Mar 14, 2007 at 04:24:54PM -0700, Davide Libenzi wrote:
> > Can you point me to a kernel path that ends up calling aio_complete() in a 
> > do-not-sleep mode?
> 
> If you remove that invariant, then it is very difficult for device drivers 
> and other code to make use of aio_complete().
> 
> > The offender I see is drivers/usb/gadget/inode.c that calls it with a 
> > spinlock held.
> 
> Which was from irq context last time I checked.
> 
> > The aio_run_iocb function seem to release/reacquire the lock before 
> > calling aio_complete().
> 
> That implies nothing -- aio_complete() has to acquire ctx_lock and cannot 
> be called holding the lock.  Sure, it could probably be split into 
> __aio_complete() and have aio_complete() wrap it acquiring the lock.

Yeah, of course. I do not plan revolutions. Just asking if it's a possible 
thing to do. I can mlock the userspace ring, if imposing that burden over 
aio_complete() is seen as too heavy.



- Davide