Re: [Qemu-devel] [PATCH v1 00/14] port network layer onto glib

[liu ping fan <qemulist@gmail.com>]

On Wed, May 15, 2013 at 4:10 PM, Stefan Hajnoczi <stefanha@redhat.com> wrote:
> On Tue, May 07, 2013 at 01:46:48PM +0800, Liu Ping Fan wrote:
>> summary:
>>   patch1: GSource event abstraction
>>   patch2~6: port network backend to glib
>>   patch7~10: make network core re-entrant
>>   patch11~14:  port the slirp backend onto glib
>>
>> The slirp->lock's deadlock problem has been eliminated and works fine.
>> And other components seems more stable, so I change from RFCv- to v-
>>
>>
>>
>>
>> rfcv5->v1:
>>   1. re-arrange logic for net/socket.c
>>
>> rfcv4->rfcv5:
>>   1.use GList to reimplement EventsGSource
>>   2.make readable()/writable() return events which the backend is interested in
>>   3.fix the slirp->lock's potential deadlock issue
>>
>> rfcv3->rfcv4:
>>   1.separate GSource event to dedicated file
>>   2.integrated with net core re-entrant
>>   3.make slirp/  re-entrant
>>
>> rfcv2->rfcv3:
>>   1.drop hub and the frontend(virtio net)
>>   2.split the patch for NetClientSource
>>
>> rfcv1->rfcv2:
>>   1.NetClientState can associate with up to 2 GSource, for virtio net, one for tx, one for rx,
>>     so vq can run on different threads.
>>   2.make network front-end onto glib, currently virtio net dataplane
>>
>>
>>
>> Liu Ping Fan (14):
>>   util: introduce gsource event abstraction
>>   net: introduce bind_ctx to NetClientInfo
>>   net: port vde onto GSource
>>   net: port socket to GSource
>>   net: port tap onto GSource
>>   net: port tap-win32 onto GSource
>>   net: hub use lock to protect ports list
>>   net: introduce lock to protect NetQueue
>>   net: introduce lock to protect NetClientState's peer's access
>>   net: make netclient re-entrant with refcnt
>>   slirp: make timeout local
>>   slirp: make slirp event dispatch based on slirp instance, not global
>>   slirp: handle race condition
>>   slirp: use lock to protect the slirp_instances
>>
>>  hw/qdev-properties-system.c |   14 +
>>  include/net/net.h           |   12 +
>>  include/qemu/module.h       |    2 +
>>  main-loop.c                 |    4 -
>>  net/hub.c                   |   28 ++-
>>  net/net.c                   |  123 ++++++++-
>>  net/queue.c                 |   15 +-
>>  net/slirp.c                 |   35 +++-
>>  net/socket.c                |  194 +++++++++++---
>>  net/tap-win32.c             |   31 ++-
>>  net/tap.c                   |   64 ++++-
>>  net/vde.c                   |   31 ++-
>>  slirp/if.c                  |   57 +++-
>>  slirp/libslirp.h            |    7 +-
>>  slirp/main.h                |    3 +-
>>  slirp/mbuf.h                |    2 +
>>  slirp/slirp.c               |  670 ++++++++++++++++++++++---------------------
>>  slirp/slirp.h               |   11 +-
>>  slirp/socket.c              |    2 +
>>  slirp/socket.h              |    1 +
>>  stubs/slirp.c               |    8 -
>>  util/Makefile.objs          |    1 +
>>  util/event_gsource.c        |  157 ++++++++++
>>  util/event_gsource.h        |   49 ++++
>>  24 files changed, 1105 insertions(+), 416 deletions(-)
>>  create mode 100644 util/event_gsource.c
>>  create mode 100644 util/event_gsource.h
>
> Hi Ping Fan,
> Sorry for the slow review, I'm focussing on QEMU 1.5 at the moment but
> will look at this series soon.
>
Thank you very much :)
And as mentioned, please skip patch2~6: port network backend to glib.
I think mdroth method is better than mind, will port onto his.

Regards,
Pingfan
> Stefan