On 2020-02-28 10:38, Mathieu Poirier wrote:
> On Thu, Feb 27, 2020 at 04:00:21PM -0800, rishabhb@xxxxxxxxxxxxxx
> wrote:
>> 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.
>
> It seems like I am missing information... Your are referring to
> "kernel
> clients" and as such I must assume some drivers that are not part of
> the
> remoteproc/rpmsg subsystems are calling qcom_register_ssr_notifier().
> I must
Yes these are not part of remoteproc framework and they will register
for notifications.
> also assume these drivers (or that functionality) are not yet upsream
> because
> all I can see calling qcom_register_ssr_notifier() is
> qcom_glink_ssr_probe().
Correct.These are not upstreamed.
>
> Speaking of which, what is the role of the qcom_glink_ssr_driver? Is
> the glink
> device that driver is handling the same as the glink device registed in
> adsp_probe() and q6v5_probe()?
glink ssr driver will send out notifications to remoteprocs that have
opened the
"glink_ssr" channel that some subsystem has gone down or booted up.
This
helps notify
neighboring subsystems about change in state of any other subsystem.
