Re: [Qemu-devel] [PATCH for-2.2 00/10] AioContext cleanups and Win32 socket support

[Stefan Hajnoczi]

On Wed, Jul 09, 2014 at 11:53:00AM +0200, Paolo Bonzini wrote:
> This series simplifies heavily aio_poll by splitting it into three
> phases: prepare (aio_compute_timeout), poll, dispatch.  The resulting
> code shares more logic between aio_poll and the GSource wrappers,
> and makes it easier to add Win32 support for sockets.
> 
> Win32 support for sockets is a prerequisite for moving the NBD server
> into the BlockDriverState's attached AioContext.  It is done in the
> final patch, based on earlier work from Or Goshen (from Intel).
> I had to more or less rewrite it to fit the new framework, but you
> can see parts of Or's work, as well as traces of aio-posix.c and
> main-loop.c logic.
> 
> Tested with NBD boot under Wine.
> 
> Paolo
> 
> Paolo Bonzini (10):
>   AioContext: take bottom halves into account when computing aio_poll
>     timeout
>   aio-win32: Evaluate timers after handles
>   aio-win32: Factor out duplicate code into aio_dispatch_handlers
>   AioContext: run bottom halves after polling
>   AioContext: export and use aio_dispatch
>   test-aio: test timers on Windows too
>   aio-win32: add aio_set_dispatching optimization
>   AioContext: introduce aio_prepare
>   qemu-coroutine-io: fix for Win32
>   aio-win32: add support for sockets
> 
>  aio-posix.c         |  58 ++++--------
>  aio-win32.c         | 262 
> +++++++++++++++++++++++++++++++++++++++-------------
>  async.c             |  39 +++++---
>  block/Makefile.objs |   2 -
>  include/block/aio.h |  25 ++++-
>  nbd.c               |   2 +-
>  qemu-coroutine-io.c |   4 +-
>  tests/test-aio.c    |  48 +++-------
>  8 files changed, 277 insertions(+), 163 deletions(-)

I'm happy with this series except for my question about int vs int64_t
types for nanosecond time values.

pgp9TwB6IgL4i.pgp
Description: PGP signature