Hi Laurent,
On Tue, Sep 19, 2017 at 04:52:29PM +0300, Laurent Pinchart wrote:
> Hi Sakari,
>
> Thank you for the patch.
>
> On Friday, 15 September 2017 17:17:13 EEST Sakari Ailus wrote:
> > Registering a notifier has required the knowledge of struct v4l2_device
> > for the reason that sub-devices generally are registered to the
> > v4l2_device (as well as the media device, also available through
> > v4l2_device).
> >
> > This information is not available for sub-device drivers at probe time.
> >
> > What this patch does is that it allows registering notifiers without
> > having v4l2_device around. Instead the sub-device pointer is stored in the
> > notifier. Once the sub-device of the driver that registered the notifier
> > is registered, the notifier will gain the knowledge of the v4l2_device,
> > and the binding of async sub-devices from the sub-device driver's notifier
> > may proceed.
> >
> > The root notifier's complete callback is only called when all sub-device
> > notifiers are completed.
>
> This is a bit hard to review, shouldn't it be split in two patches, one that
> refactors the functions, and another one that allows binding notifiers to
> subdevs ?
>
> > Signed-off-by: Sakari Ailus <sakari.ailus@xxxxxxxxxxxxxxx>
> > ---
> > drivers/media/v4l2-core/v4l2-async.c | 218 +++++++++++++++++++++++++++-----
> > include/media/v4l2-async.h | 16 ++-
> > 2 files changed, 203 insertions(+), 31 deletions(-)
> >
> > diff --git a/drivers/media/v4l2-core/v4l2-async.c
> > b/drivers/media/v4l2-core/v4l2-async.c index 4be2f16af051..52fe22b9b6b4
> > 100644
> > --- a/drivers/media/v4l2-core/v4l2-async.c
> > +++ b/drivers/media/v4l2-core/v4l2-async.c
> > @@ -53,6 +53,10 @@ static int v4l2_async_notifier_call_complete(struct
> > v4l2_async_notifier *n) return n->ops->complete(n);
> > }
> >
> > +static int v4l2_async_match_notify(struct v4l2_async_notifier *notifier,
> > + struct v4l2_subdev *sd,
> > + struct v4l2_async_subdev *asd);
>
> Forward declarations are often a sign that something is wrong :-/ If you
> really need to keep this I'd move it right before the function that needs it.
"Something being wrong" here is that we have a data structure (the graph)
and a portion of the graph is parsed at any given point of time; there is
no central location with the knowledge of each node in the graph. Therefore
the process is recursive: you only learn of new nodes to parse when you
have parsed something.
I can move the declaration closer to where it's used.
>
> > static bool match_i2c(struct v4l2_subdev *sd, struct v4l2_async_subdev
> > *asd) {
> > #if IS_ENABLED(CONFIG_I2C)
> > @@ -124,14 +128,127 @@ static struct v4l2_async_subdev
> > *v4l2_async_find_match( return NULL;
> > }
> >
> > +/* Find the sub-device notifier registered by a sub-device driver. */
> > +static struct v4l2_async_notifier *v4l2_async_find_subdev_notifier(
> > + struct v4l2_subdev *sd)
> > +{
> > + struct v4l2_async_notifier *n;
> > +
> > + list_for_each_entry(n, ¬ifier_list, list)
> > + if (n->sd == sd)
> > + return n;
> > +
> > + return NULL;
> > +}
> > +
> > +/* Return true if all sub-device notifiers are complete, false otherwise.
> > */
> > +static bool v4l2_async_subdev_notifiers_complete(
> > + struct v4l2_async_notifier *notifier)
> > +{
> > + struct v4l2_subdev *sd;
> > +
> > + if (!list_empty(¬ifier->waiting))
> > + return false;
> > +
> > + list_for_each_entry(sd, ¬ifier->done, async_list) {
> > + struct v4l2_async_notifier *subdev_notifier =
> > + v4l2_async_find_subdev_notifier(sd);
> > +
> > + if (subdev_notifier &&
> > + !v4l2_async_subdev_notifiers_complete(subdev_notifier))
> > + return false;
>
> This will loop forever if two subdevs add each other to their respective
> notifiers. We might not have any use case for that right now, but it's bound
> to happen, at least as a bug during development, and an infinite loop (with an
> additional stack overflow bonus) isn't very nice to debug.
Well, yes. If you have a driver bug then this is what could happen.
One option is to check whether an fwnode has already been associated with
an async subdev and fail if it is. I was originally thinking of adding that
but then ended up postponing that for later. I can add that to v14.
>
> > + }
> > +
> > + return true;
> > +}
> > +
> > +/* Get v4l2_device related to the notifier if one can be found. */
> > +static struct v4l2_device *v4l2_async_notifier_find_v4l2_dev(
> > + struct v4l2_async_notifier *notifier)
> > +{
> > + while (notifier->parent)
> > + notifier = notifier->parent;
> > +
> > + return notifier->v4l2_dev;
> > +}
> > +
> > +/* Test all async sub-devices in a notifier for a match. */
> > +static int v4l2_async_notifier_try_all_subdevs(
> > + struct v4l2_async_notifier *notifier)
> > +{
> > + struct v4l2_subdev *sd;
> > +
> > + if (!v4l2_async_notifier_find_v4l2_dev(notifier))
> > + return 0;
> > +
> > +again:
> > + list_for_each_entry(sd, &subdev_list, async_list) {
> > + struct v4l2_async_subdev *asd;
> > + int ret;
> > +
> > + asd = v4l2_async_find_match(notifier, sd);
> > + if (!asd)
> > + continue;
> > +
> > + ret = v4l2_async_match_notify(notifier, sd, asd);
> > + if (ret < 0)
> > + return ret;
> > +
> > + /*
> > + * v4l2_async_match_notify() may lead to registering a
> > + * new notifier and thus changing the async subdevs
> > + * list. In order to proceed safely from here, restart
> > + * parsing the list from the beginning.
> > + */
> > + goto again;
> > + }
> > +
> > + return 0;
> > +}
> > +
> > +/* Try completing a notifier. */
> > +static int v4l2_async_notifier_try_complete(
> > + struct v4l2_async_notifier *notifier)
> > +{
> > + do {
> > + int ret;
> > +
> > + /* Any local async sub-devices left? */
> > + if (!list_empty(¬ifier->waiting))
> > + return 0;
> > +
> > + /*
> > + * Any sub-device notifiers waiting for async subdevs
> > + * to be bound?
> > + */
> > + if (!v4l2_async_subdev_notifiers_complete(notifier))
> > + return 0;
> > +
> > + /* Proceed completing the notifier */
> > + ret = v4l2_async_notifier_call_complete(notifier);
> > + if (ret < 0)
> > + return ret;
> > +
> > + /*
> > + * Obtain notifier's parent. If there is one, repeat
> > + * the process, otherwise we're done here.
> > + */
> > + notifier = notifier->parent;
> > + } while (notifier);
> > +
> > + return 0;
> > +}
> > +
> > static int v4l2_async_match_notify(struct v4l2_async_notifier *notifier,
> > struct v4l2_subdev *sd,
> > struct v4l2_async_subdev *asd)
> > {
> > + struct v4l2_async_notifier *subdev_notifier;
> > int ret;
> >
> > - ret = v4l2_device_register_subdev(notifier->v4l2_dev, sd);
> > - if (ret < 0)
> > + ret = v4l2_device_register_subdev(
> > + v4l2_async_notifier_find_v4l2_dev(notifier), sd);
> > + if (ret)
> > return ret;
> >
> > ret = v4l2_async_notifier_call_bound(notifier, sd, asd);
> > @@ -148,10 +265,20 @@ static int v4l2_async_match_notify(struct
> > v4l2_async_notifier *notifier, /* Move from the global subdevice list to
> > notifier's done */
> > list_move(&sd->async_list, ¬ifier->done);
> >
> > - if (list_empty(¬ifier->waiting))
> > - return v4l2_async_notifier_call_complete(notifier);
> > + /*
> > + * See if the sub-device has a notifier. If it does, proceed
> > + * with checking for its async sub-devices.
> > + */
> > + subdev_notifier = v4l2_async_find_subdev_notifier(sd);
> > + if (subdev_notifier && !subdev_notifier->parent) {
> > + subdev_notifier->parent = notifier;
> > + ret = v4l2_async_notifier_try_all_subdevs(subdev_notifier);
> > + if (ret)
> > + return ret;
> > + }
> >
> > - return 0;
> > + /* Try completing the notifier and its parent(s). */
> > + return v4l2_async_notifier_try_complete(notifier);
> > }
> >
> > static void v4l2_async_cleanup(struct v4l2_subdev *sd)
> > @@ -163,17 +290,15 @@ static void v4l2_async_cleanup(struct v4l2_subdev *sd)
> > sd->dev = NULL;
> > }
> >
> > -int v4l2_async_notifier_register(struct v4l2_device *v4l2_dev,
> > - struct v4l2_async_notifier *notifier)
> > +static int __v4l2_async_notifier_register(struct v4l2_async_notifier
> > *notifier) {
> > - struct v4l2_subdev *sd, *tmp;
> > struct v4l2_async_subdev *asd;
> > + int ret;
> > int i;
> >
> > - if (!v4l2_dev || notifier->num_subdevs > V4L2_MAX_SUBDEVS)
> > + if (notifier->num_subdevs > V4L2_MAX_SUBDEVS)
> > return -EINVAL;
> >
> > - notifier->v4l2_dev = v4l2_dev;
> > INIT_LIST_HEAD(¬ifier->waiting);
> > INIT_LIST_HEAD(¬ifier->done);
> >
> > @@ -200,18 +325,10 @@ int v4l2_async_notifier_register(struct v4l2_device
> > *v4l2_dev,
> >
> > mutex_lock(&list_lock);
> >
> > - list_for_each_entry_safe(sd, tmp, &subdev_list, async_list) {
> > - int ret;
> > -
> > - asd = v4l2_async_find_match(notifier, sd);
> > - if (!asd)
> > - continue;
> > -
> > - ret = v4l2_async_match_notify(notifier, sd, asd);
> > - if (ret < 0) {
> > - mutex_unlock(&list_lock);
> > - return ret;
> > - }
> > + ret = v4l2_async_notifier_try_all_subdevs(notifier);
> > + if (ret) {
> > + mutex_unlock(&list_lock);
> > + return ret;
> > }
> >
> > /* Keep also completed notifiers on the list */
> > @@ -221,29 +338,70 @@ int v4l2_async_notifier_register(struct v4l2_device
> > *v4l2_dev,
> >
> > return 0;
> > }
> > +
> > +int v4l2_async_notifier_register(struct v4l2_device *v4l2_dev,
> > + struct v4l2_async_notifier *notifier)
> > +{
> > + if (WARN_ON(!v4l2_dev || notifier->sd))
> > + return -EINVAL;
> > +
> > + notifier->v4l2_dev = v4l2_dev;
> > +
> > + return __v4l2_async_notifier_register(notifier);
> > +}
> > EXPORT_SYMBOL(v4l2_async_notifier_register);
> >
> > -void v4l2_async_notifier_unregister(struct v4l2_async_notifier *notifier)
> > +int v4l2_async_subdev_notifier_register(struct v4l2_subdev *sd,
> > + struct v4l2_async_notifier *notifier)
> > {
> > - struct v4l2_subdev *sd, *tmp;
> > + if (WARN_ON(!sd || notifier->v4l2_dev))
> > + return -EINVAL;
> >
> > - if (!notifier->v4l2_dev)
> > - return;
> > + notifier->sd = sd;
> >
> > - mutex_lock(&list_lock);
> > + return __v4l2_async_notifier_register(notifier);
> > +}
> > +EXPORT_SYMBOL(v4l2_async_subdev_notifier_register);
> >
> > - list_del(¬ifier->list);
> > +/* Unbind all sub-devices in the notifier tree. */
> > +static void v4l2_async_notifier_unbind_all_subdevs(
> > + struct v4l2_async_notifier *notifier)
> > +{
> > + struct v4l2_subdev *sd, *tmp;
> >
> > list_for_each_entry_safe(sd, tmp, ¬ifier->done, async_list) {
> > + struct v4l2_async_notifier *subdev_notifier =
> > + v4l2_async_find_subdev_notifier(sd);
> > +
> > + if (subdev_notifier)
> > + v4l2_async_notifier_unbind_all_subdevs(subdev_notifier);
> > +
> > v4l2_async_cleanup(sd);
> >
> > v4l2_async_notifier_call_unbind(notifier, sd, sd->asd);
> > - }
> >
> > - mutex_unlock(&list_lock);
> > + list_del(&sd->async_list);
> > + list_add(&sd->async_list, &subdev_list);
>
> How about list_move() ?
Yeah.
>
> This seems to be new code, and by the look of it, I wonder whether it doesn't
> belong in the reprobing removal patch.
This is not related to re-probing. Here we're moving an async sub-device
back to the global sub-device list when its notifier is going away.
>
> > + }
> >
> > + notifier->parent = NULL;
> > + notifier->sd = NULL;
> > notifier->v4l2_dev = NULL;
> > }
> > +
> > +void v4l2_async_notifier_unregister(struct v4l2_async_notifier *notifier)
> > +{
> > + if (!notifier->v4l2_dev && !notifier->sd)
> > + return;
> > +
> > + mutex_lock(&list_lock);
> > +
> > + v4l2_async_notifier_unbind_all_subdevs(notifier);
> > +
> > + list_del(¬ifier->list);
> > +
> > + mutex_unlock(&list_lock);
> > +}
> > EXPORT_SYMBOL(v4l2_async_notifier_unregister);
> >
> > void v4l2_async_notifier_release(struct v4l2_async_notifier *notifier)
>
--
Regards,
Sakari Ailus
e-mail: sakari.ailus@xxxxxx
