Re: [Qemu-devel] [RFC PATCH 0/2] port network layer onto glib

[Paolo Bonzini]

> 1) It has no facility for timer events

Yup, it's on the todo list.

> 2) It's tied to file descriptors (only a problem for win32)

The other way round: it's not tied to file descriptors for win32,
which is already a problem for e.g. networked backends.  main-loop.c
has the code that is needed, but it's low on the todo list.

Note that tap-win32 for example doesn't need this.

> 3) The fd support is tied to read/write without an extensibility
> mechanism for other fd events

Yes.  Though internally it uses g_poll, so it's "just" a matter
of defining the right API.
 
> 4) It's got no mechanism to interact with signals

signalfd?  (GSource doesn't have any such mechanism directly).

> 5) It's not obvious how we would integrate with polling-based
> callbacks like we have in the character layer and networking layer.

Isn't io_flush one such mechanism?  Right now it applies to both
io_read and io_write, but really it is never used for io_write.

Also, this and (3) might be the same problem.

> So I agree it's simple but I don't think it can reasonably stay simple.
> I think if we added all of the above, the best we would expect to end
> up with is something that looked like glib.
>
> As it stands, the lack of (5) would make it extremely difficult to
> convert the networking layer.

Quite possible, I've never looked very much at the networking layer.

Paolo

> Regards,
> 
> Anthony Liguori
> 
> >
> >>> and AioContext's code is vastly simpler than GMainLoop's.
> >> 
> >> For now.
> >
> > Fair enough. :)
> >
> >>> AioContext is also documented and unit tested, with tests
> >>> for both standalone and GSource operation.  Unit tests for
> >>> AioContext
> >>> users are trivial to write, we have one in test-thread-pool.
> >>>
> >>>> Did you have a specific concern with using glib vs. AioContext?
> >>>>  Is it
> >>>> about reusing code in the block layer where AioContext is
> >>>> required?
> >>>
> >>> In the short term yes, code duplication is a concern.  We already
> >>> have
> >>> two implementation of virtio.
> >> 
> >> I share your concern but in the opposite direction.  We have three
> >> main
> >> loops today.
> >
> > Yes, and two of them (main-loop.c/qemu-timer.c and async.c) can be
> > merged.
> >
> >>> I would like the dataplane virtio code to
> >>> grow everything else that needs to be in all dataplane-style
> >>> devices
> >>> (for example, things such as setting up the guest<->host
> >>> notifiers), and
> >>> the hw/virtio.c API implemented on top of it (or dead
> >>> altogether).
> >>> Usage of AioContext is pretty much forced by the block layer.
> >> 
> >> I don't think that AioContext is the right answer because it makes
> >> it
> >> too easy to shoot yourself in the foot.
> >
> > See above, if nesting is the problem it's gone.
> >
> > Paolo
>