On 2020-02-27 13:59, Mathieu Poirier wrote:
> On Wed, Feb 19, 2020 at 06:57:45PM -0800, Siddharth Gupta wrote:
> > The SSR subdevice only adds callback for the unprepare event. Add
> > callbacks
> > for unprepare, start and prepare events. The client driver for a
> > particular
> > remoteproc might be interested in knowing the status of the remoteproc
> > while undergoing SSR, not just when the remoteproc has finished
> > shutting
> > down.
> >
> > Signed-off-by: Siddharth Gupta <sidgup@xxxxxxxxxxxxxx>
> > ---
> > drivers/remoteproc/qcom_common.c | 39
> > +++++++++++++++++++++++++++++++++++----
> > include/linux/remoteproc.h | 15 +++++++++++++++
> > 2 files changed, 50 insertions(+), 4 deletions(-)
> >
> > diff --git a/drivers/remoteproc/qcom_common.c
> > b/drivers/remoteproc/qcom_common.c
> > index 6714f27..6f04a5b 100644
> > --- a/drivers/remoteproc/qcom_common.c
> > +++ b/drivers/remoteproc/qcom_common.c
> > @@ -183,9 +183,9 @@ EXPORT_SYMBOL_GPL(qcom_remove_smd_subdev);
> > *
> > * Returns pointer to srcu notifier head on success, ERR_PTR on
> > failure.
> > *
> > - * This registers the @notify function as handler for restart
> > notifications. As
> > - * remote processors are stopped this function will be called, with
> > the rproc
> > - * pointer passed as a parameter.
> > + * This registers the @notify function as handler for
> > powerup/shutdown
> > + * notifications. This function will be invoked inside the
> > callbacks registered
> > + * for the ssr subdevice, with the rproc pointer passed as a
> > parameter.
> > */
> > void *qcom_register_ssr_notifier(struct rproc *rproc, struct
> > notifier_block *nb)
> > {
> > @@ -227,11 +227,39 @@ int qcom_unregister_ssr_notifier(void *notify,
> > struct notifier_block *nb)
> > }
> > EXPORT_SYMBOL_GPL(qcom_unregister_ssr_notifier);
> >
> > +static int ssr_notify_prepare(struct rproc_subdev *subdev)
> > +{
> > + struct qcom_rproc_ssr *ssr = to_ssr_subdev(subdev);
> > +
> > + srcu_notifier_call_chain(ssr->rproc_notif_list,
> > + RPROC_BEFORE_POWERUP, (void *)ssr->name);
> > + return 0;
> > +}
> > +
> > +static int ssr_notify_start(struct rproc_subdev *subdev)
> > +{
> > + struct qcom_rproc_ssr *ssr = to_ssr_subdev(subdev);
> > +
> > + srcu_notifier_call_chain(ssr->rproc_notif_list,
> > + RPROC_AFTER_POWERUP, (void *)ssr->name);
> > + return 0;
> > +}
> > +
> > +static void ssr_notify_stop(struct rproc_subdev *subdev, bool
> > crashed)
> > +{
> > + struct qcom_rproc_ssr *ssr = to_ssr_subdev(subdev);
> > +
> > + srcu_notifier_call_chain(ssr->rproc_notif_list,
> > + RPROC_BEFORE_SHUTDOWN, (void *)ssr->name);
> > +}
> > +
> > +
> > static void ssr_notify_unprepare(struct rproc_subdev *subdev)
> > {
> > struct qcom_rproc_ssr *ssr = to_ssr_subdev(subdev);
> >
> > - srcu_notifier_call_chain(ssr->rproc_notif_list, 0, (void
> > *)ssr->name);
> > + srcu_notifier_call_chain(ssr->rproc_notif_list,
> > + RPROC_AFTER_SHUTDOWN, (void *)ssr->name);
> > }
> >
> > /**
> > @@ -248,6 +276,9 @@ void qcom_add_ssr_subdev(struct rproc *rproc,
> > struct qcom_rproc_ssr *ssr,
> > {
> > ssr->name = ssr_name;
> > ssr->subdev.name = kstrdup("ssr_notifs", GFP_KERNEL);
> > + ssr->subdev.prepare = ssr_notify_prepare;
> > + ssr->subdev.start = ssr_notify_start;
> > + ssr->subdev.stop = ssr_notify_stop;
>
> Now that I have a better understanding of what this patchset is doing, I
> realise
> my comments in patch 04 won't work. To differentiate the subdevs of an
> rproc I
> suggest to wrap them in a generic structure with a type and an enum.
> That way
> you can differenciate between subdevices without having to add to the
> core.
Ok. I can try that.
>
> That being said, I don't understand what patches 5 and 6 are doing...
> Registering with the global ssr_notifiers allowed to gracefully shutdown
> all the
> MCUs in the system when one of them would go down. But now that we are
> using
> the notifier on a per MCU, I really don't see why each subdev couldn't
> implement
> the right prepare/start/stop functions.
>
> Am I missing something here?
We only want kernel clients to be notified when the Remoteproc they
are
interested
in changes state. For e.g. audio kernel driver should be notified when
audio
processor goes down but it does not care about any other remoteproc.
If you are suggesting that these kernel clients be added as subdevices
then
we will end up having many subdevices registered to each remoteproc.
So we
implemented a notifier chain per Remoteproc. This keeps the SSR
notifications as
the subdevice per remoteproc, and all interested clients can register
to it.
