Re: [PATCH v8 3/9] firmware: arm_scmi: Add notification dispatch and delivery

[Lukasz Luba]

On 6/18/20 12:31 AM, Cristian Marussi wrote:
> On Mon, Jun 08, 2020 at 06:03:46PM +0100, Sudeep Holla wrote:
> > On Wed, May 20, 2020 at 09:11:12AM +0100, Cristian Marussi wrote:
> > > Add core SCMI Notifications dispatch and delivery support logic which is
> > > able, at first, to dispatch well-known received events from the RX ISR to
> > > the dedicated deferred worker, and then, from there, to final deliver the
> > > events to the registered users' callbacks.
> > >
> > > Dispatch and delivery is just added here, still not enabled.
> > >
> > > Reviewed-by: Jonathan Cameron <Jonathan.Cameron@xxxxxxxxxx>
> > > Signed-off-by: Cristian Marussi <cristian.marussi@xxxxxxx>
> > > ---
> > >   drivers/firmware/arm_scmi/notify.c | 354 ++++++++++++++++++++++++++++-
> > >   drivers/firmware/arm_scmi/notify.h |  10 +
> > >   2 files changed, 362 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/drivers/firmware/arm_scmi/notify.c b/drivers/firmware/arm_scmi/notify.c
> > > index 7cf61dbe2a8e..d582f71fde5b 100644
> > > --- a/drivers/firmware/arm_scmi/notify.c
> > > +++ b/drivers/firmware/arm_scmi/notify.c
> >
> > [...]
> >
> > > @@ -1085,6 +1422,12 @@ int scmi_notification_init(struct scmi_handle *handle)
> > >   	ni->gid = gid;
> > >   	ni->handle = handle;
> > >   
> > > +	ni->notify_wq = alloc_workqueue("scmi_notify",
> > > +					WQ_UNBOUND | WQ_FREEZABLE | WQ_SYSFS,
> > > +					0);
> >
> > What's the use of WQ_SYSFS for SCMI notifications ? Do we need it ?
>
> Lukasz asked for it, when we were talking about workqueues' priorities configurability.
> (not implemented in this series)

I confirm, I've asked if we can have a mechanism to control these
workqueues. They will be running concurrently with other CFS
tasks which could cause delays for them. They could also be scheduled
on a random core: big or little (depends on its utilization) but maybe
we would like to pin them explicitly to some cores, i.e
little only. We have also discussed a possible mechanism based on RT
threads (which could avoid CFS delays), but that would require a lot of
changes, so this flag here gives us some control.
But if you decide to remove this flag, we would probably find a solution
using uclamp or similar when needed.

Regards,
Lukasz

> Thanks
>
> Cristian
> > --
> > Regards,
> > Sudeep