Re: [RFC 00/16] Discovery procedure refactoring

[Johan Hedberg <johan.hedberg@gmail.com>]

Hi André,

On Fri, Apr 29, 2011, Andre Guedes wrote:
> Hi all,
> 
> Today, discovery procedure is supported only in hciops. This refactoring has
> been done to provide easier implementation of discovery procedure in mgmtops.
> Lots of effort would be necessary to implement discovery procedure in mgmtops
> because its logic is spread out adapter layer and hciops layer.
> 
> The approach this patchset follows is moving all logic related to discovery
> procedure from adapter layer to hciops layer.
> 
> Future work will be:
>         - full support for discovery procedure (BR/EDR, LE-Only, BR/EDR/LE
>           devices) in kernel via management interface (today, only BR/EDR is
>           supported).
>         - name resolving support through mgmt interface (kernel + userspace)
> 
> Thanks,
> Guedes.
> 
> Anderson Briglia (1):
>   Implement mgmt start and stop discovery
> 
> Andre Guedes (15):
>   Add discovery callbacks to btd_adapter_ops
>   Replace inquiry/scanning calls by discovery calls
>   Add 'discov_state' field to struct dev_info
>   Code cleanup event.c
>   Remove Periodic Inquiry support in hciops
>   Change DiscoverSchedulerInterval default value
>   Add 'timeout' param to start_scanning callback
>   Refactoring adapter_set_state()
>   Remove 'suspend' param from stop_discovery()
>   Add extfeatures to struct dev_info
>   Implement start_discovery hciops callback
>   Remove obsolete code.
>   Implement stop_discovery hciops callback
>   Remove inquiry and scanning callbacks from btd_adapter_ops
>   Remove 'periodic' param from hciops_start_inquiry()
> 
>  plugins/hciops.c  |  373 ++++++++++++++++++++++++++++++++++++-----------------
>  plugins/mgmtops.c |   35 ++----
>  src/adapter.c     |  217 ++++++++++---------------------
>  src/adapter.h     |   19 +--
>  src/event.c       |   48 +-------
>  src/event.h       |    1 -
>  src/main.conf     |    4 +-
>  7 files changed, 345 insertions(+), 352 deletions(-)

All patches in this set have been pushed upstream. Thanks.

There seems to be at least one race condition that'd need fixing which I
can see with test-discovery: if the D-Bus client that requested
discovery exits like test-discovery does it can occur between sending
the HCI_Inquiry command but before the cmd_status or first inquiry event
arrives. In this case the adapter state doesn't seem to be yet updated
and so the ongoing inquiry doesn't get canceled even if it should. You
should be able to see this simply by running hcidump together with
test-discovery e.g. on your laptop.

Johan