Re: [PATCH/RFC v4 00/21] LED / flash API integration

From: Jacek Anaszewski
Date: Thu Aug 07 2014 - 04:21:23 EST


Hi Sakari,

On 08/06/2014 08:53 AM, Sakari Ailus wrote:
Hi Jacek,

On Fri, Jul 11, 2014 at 04:04:03PM +0200, Jacek Anaszewski wrote:
...
1) Who should register V4L2 Flash sub-device?

LED Flash Class devices, after introduction of the Flash Manager,
are not tightly coupled with any media controller. They are maintained
by the Flash Manager and made available for dynamic assignment to
any media system they are connected to through multiplexing devices.

In the proposed rough solution, when support for V4L2 Flash sub-devices
is enabled, there is a v4l2_device created for them to register in.
This however implies that V4L2 Flash device will not be available
in any media controller, which calls its existence into question.

Therefore I'd like to consult possible ways of solving this issue.
The option I see is implementing a mechanism for moving V4L2 Flash
sub-devices between media controllers. A V4L2 Flash sub-device
would initially be assigned to one media system in the relevant
device tree binding, but it could be dynamically reassigned to
the other one. However I'm not sure if media controller design
is prepared for dynamic modifications of its graph and how many
modifications in the existing drivers this solution would require.

Do you have a use case where you would need to strobe a flash from multiple
media devices at different times, or is this entirely theoretical? Typically
flash controllers are connected to a single source of hardware strobe (if
there's one) since the flash LEDs are in fact mounted next to a specific
camera sensor.

I took into account such arrangements in response to your message [1], where you were considering configurations like "one flash but two
cameras", "one camera and two flashes". And you also called for proposing generic solution.

One flash and two (or more) cameras case is easily conceivable -
You even mentioned stereo cameras. One camera and many flashes
arrangement might be useful in case of some professional devices which
might be designed so that they would be able to apply different scene
lighting. I haven't heard about such devices, but as you said
such a configuration isn't unthinkable.

If this is a real issue the way to solve it would be to have a single media
device instead of many.

I was considering adding media device, that would be a representation
of a flash manager, gathering all the registered flashes. Nonetheless,
finally I came to conclusion that a v4l2-device alone should suffice,
just to provide a Flash Manager representation allowing for v4l2-flash sub-devices to register in.
All the features provided by the media device are useless in case
of a set of V4L2 Flash sub-devices. They couldn't have any linkage
in such a device. The only benefit from having media device gathering
V4L2 Flash devices would be possibility of listing them.

2) Consequences of locking the Flash Manager during flash strobe

In case a LED Flash Class device depends on muxes involved in
routing the other LED Flash Class device's strobe signals,
the Flash Manager must be locked for the time of strobing
to prevent reconfiguration of the strobe signal routing
by the other device.

I wouldn't be concerned of this in particular. It's more important we do
actully have V4L2 flash API supported by LED flash drivers and that they do
implement the API correctly.

It's at least debatable whether you should try to prevent user space from
doing silly things or not. With complex devices it may relatively easily
lead to wrecking havoc with actual use cases which we certainly do not want.

In this case, if you just prevent changing the routing (do you have a use
case for it?) while strobing, someone else could still change the routing
just before you strobe.

Originally I started to implementing this so that strobe signal routing
was altered upon setting strobe source. With such an implementation the use case would be as follows:

1. Process 1 sets strobe source to external
2. Process 2 sets strobe source to software
3. Process 1 strobes the flash, unaware that strobe source setting has
been changed

To avoid such problems I changed the implementation so that the
routing is set in the led_flash_manager_setup_strobe function called
from led_set_flash_strobe and led_set_external_strobe functions.
led_flash_manager_setup_strobe sets strobe signal routing
and strobes the flash under lock and holds it for the flash timeout
period, which prevents spurious reconfiguration.

Nonetheless, I agree that trying to handle this problem is troublesome,
and would affect current V4L2 Flash SPI semantics. If you don't share
my concerns I am happy to leave this locking solution out :)

Best Regards,
Jacek Anaszewski

[1] http://www.spinics.net/lists/linux-leds/msg01617.html
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/