Hi Luiz,
On Thu, 2011-08-18 at 03:57 -0400, Luiz Augusto von Dentz wrote:
> Hi Peter,
>
> On Thu, Aug 18, 2011 at 3:37 AM, Peter Hurley <peter@xxxxxxxxxxxxxxxxxx> wrote:
> > Utilize already known tx buffer count values as parameters to
> > the tx scheduler, rather than determining the values on each
> > iteration of the tx scheduler.
> >
> > This is preparatory for merging transmission types (such as SCO
> > and ESCO) and also for handling tx buffer counters atomically.
> >
> > Signed-off-by: Peter Hurley <peter@xxxxxxxxxxxxxxxxxx>
... <snip> ...
> Nice work, it probably conflict with one of my patches to check if
> there is any connection, but I guess it is easy to rebase,
I saw that. I plan to merge these on top of your work when we get to
that point (assuming there's no objections to either proposals).
> btw it is
> probably a good idea to no initialize the variables in the declaration
> since with hci_conn_num check we may bail out without using them.
I will. The reason why I did that is to make the transition to atomic tx
buffer counts more obvious and less painful. The tasklet usage pattern
for atomic counters will look something like:
/* claim entire counter */
cnt = atomic_xchg(&hdev->acl_cnt, 0);
... schedule and send packets ...
/* add back in remaining unused count */
atomic_add(&hdev->acl_cnt, cnt);
Regards,
Peter Hurley
��.n��������+%������w��{.n�����{��� �^n�r������&��z�ޗ�zf���h���~����������_��+v���)ߣ�
