Message-ID: <20160313201131.GN11084@valkosipuli.retiisi.org.uk>
Date:	Sun, 13 Mar 2016 22:11:32 +0200
From:	Sakari Ailus <sakari.ailus@....fi>
To:	Shuah Khan <shuahkh@....samsung.com>
Cc:	mchehab@....samsung.com, tiwai@...e.com, clemens@...isch.de,
	hans.verkuil@...co.com, laurent.pinchart@...asonboard.com,
	sakari.ailus@...ux.intel.com, javier@....samsung.com,
	pawel@...iak.com, m.szyprowski@...sung.com,
	kyungmin.park@...sung.com, perex@...ex.cz, arnd@...db.de,
	dan.carpenter@...cle.com, tvboxspy@...il.com, crope@....fi,
	ruchandani.tina@...il.com, corbet@....net, chehabrafael@...il.com,
	k.kozlowski@...sung.com, stefanr@...6.in-berlin.de,
	inki.dae@...sung.com, jh1009.sung@...sung.com,
	elfring@...rs.sourceforge.net, prabhakar.csengg@...il.com,
	sw0312.kim@...sung.com, p.zabel@...gutronix.de,
	ricardo.ribalda@...il.com, labbott@...oraproject.org,
	pierre-louis.bossart@...ux.intel.com, ricard.wanderlof@...s.com,
	julian@...st.de, takamichiho@...il.com, dominic.sacre@....de,
	misterpib@...il.com, daniel@...que.org, gtmkramer@...all.nl,
	normalperson@...t.net, joe@...po.co.uk, linuxbugs@...tgam.net,
	johan@...ud.se, klock.android@...il.com, nenggun.kim@...sung.com,
	j.anaszewski@...sung.com, geliangtang@....com, albert@...tsing.nl,
	linux-kernel@...r.kernel.org, linux-media@...r.kernel.org,
	alsa-devel@...a-project.org
Subject: Re: [PATCH v3 06/22] media: Media Controller enable/disable source
 handler API

Hi Shuah,

On Thu, Mar 10, 2016 at 07:29:59AM -0700, Shuah Khan wrote:
> On 03/10/2016 12:35 AM, Sakari Ailus wrote:
> > Hi Shuah,
> > 
> > On Thu, Feb 11, 2016 at 04:41:22PM -0700, Shuah Khan wrote:
> >> Add new fields to struct media_device to add enable_source, and
> >> disable_source handlers, and source_priv to stash driver private
> >> data that is used to run these handlers. The enable_source handler
> >> finds source entity for the passed in entity and checks if it is
> >> available. When link is found, it activates it. Disable source
> >> handler deactivates the link.
> >>
> >> Bridge driver is expected to implement and set these handlers.
> >>
> >> Signed-off-by: Shuah Khan <shuahkh@....samsung.com>
> >> ---
> >>  include/media/media-device.h | 30 ++++++++++++++++++++++++++++++
> >>  1 file changed, 30 insertions(+)
> >>
> >> diff --git a/include/media/media-device.h b/include/media/media-device.h
> >> index 075a482..1a04644 100644
> >> --- a/include/media/media-device.h
> >> +++ b/include/media/media-device.h
> >> @@ -302,6 +302,11 @@ struct media_entity_notify {
> >>   * @entity_notify: List of registered entity_notify callbacks
> >>   * @lock:	Entities list lock
> >>   * @graph_mutex: Entities graph operation lock
> >> + *
> >> + * @source_priv: Driver Private data for enable/disable source handlers
> >> + * @enable_source: Enable Source Handler function pointer
> >> + * @disable_source: Disable Source Handler function pointer
> >> + *
> >>   * @link_notify: Link state change notification callback
> >>   *
> >>   * This structure represents an abstract high-level media device. It allows easy
> >> @@ -313,6 +318,26 @@ struct media_entity_notify {
> >>   *
> >>   * @model is a descriptive model name exported through sysfs. It doesn't have to
> >>   * be unique.
> >> + *
> >> + * @enable_source is a handler to find source entity for the
> >> + * sink entity  and activate the link between them if source
> >> + * entity is free. Drivers should call this handler before
> >> + * accessing the source.
> >> + *
> >> + * @disable_source is a handler to find source entity for the
> >> + * sink entity  and deactivate the link between them. Drivers
> >> + * should call this handler to release the source.
> >> + *
> > 
> > Is there a particular reason you're not simply (de)activating the link, but
> > instead add a new callback?
> 
> These two handlers are separate for a couple of reasons:
> 
> 1. Explicit and symmetric API is easier to use and maintain.
>    Similar what we do in other cases, register/unregister
>    get/put etc.

Link state is set explicitly (enabled or disabled). This is certainly not a
reason to create a redundant API for link setup.

> 2. This is more important. Disable handler makes sure the
>    owner is releasing the resource. Otherwise, when some
>    other application does enable, the owner could loose
>    the resource, if enable and disable are the same.
> 
>    e.g: Video app is holding the resource, DVB app does
>    enable. Disable handler makes sure Video/owner  is the one
>    that is asking to do the release.

Based on the later patches in this set, the enable_source() callback of the 
au0828 driver performs three things:

- Find the source entity,
- Enable the link from some au0828 entity to the source and
- Start the pipeline that begins from the I/O device node. The pipe object
  is embedded in struct video_device.

disable_source() undoes this in reverse order.

That's in the au0828 driver and rightly so.

Then it gets murkier. enable_source() and disable_source() callbacks
(through a few turns) will get called from v4l2-ioctl.c functions
v4l_querycap, v4l_s_fmt, v4l_s_frequency, v4l_s_std and v4l_querystd. This
is also performed in VB2 core vb2_core_streamon() function.

I certainly have no objections when it comes to blocking other processes
from setting the format when a process holding a file handle to a device has
e.g. set the format. The implementation is another matter.

What particularly does concern me in this patchset is:

- struct media_pipe is intended to be allocated by drivers embedded in
  another struct holding information the driver needs related to the
  pipeline. Moving this struct to struct video_device prevents this, which
  translates to v4l_enable_media_source() and v4l_disable_media_source()
  functions the patchset adds being specific to the au0828 driver. They
  should be part of that driver, and may not be part of the V4L2 core.
  struct media_pipe may not be added to struct video_device for the same
  reason.

- The IOCTL handlers in v4l2-ioctl.c already call driver-settable callbacks
  before this set. It looks like redundant callbacks are added there as
  ell. The same appears to be true for the VB2 callback. Could you use the
  existing callbacks instead of creating new ones?

As a short term solution, I propose moving the code to the au0828 driver.
Once it is there, we can see whether it could be made more generic if there
is a need for it elsewhere. I believe so. Support atomic pipeline
configuration and startup is a generic problem that requires a generic
solution: we should have a way to construct a pipeline and prevent other
users from messing with it before it's started. But as currently implemented
by this patchset, it is very specific to the au0828 driver and as such may
not be added to the V4L2 or the MC frameworks.

-- 
Best regards,

Sakari Ailus
e-mail: sakari.ailus@....fi	XMPP: sailus@...iisi.org.uk

Powered by blists - more mailing lists