Re: [PATCH 1/2] V4L: remove clock name from v4l2_clk API

[Laurent Pinchart <laurent.pinchart@ideasonboard.com>]

Hi Josh,

On Monday 12 January 2015 17:14:33 Josh Wu wrote:
> On 1/9/2015 6:47 AM, Laurent Pinchart wrote:
> > On Thursday 08 January 2015 23:37:58 Guennadi Liakhovetski wrote:
> >> On Wed, 7 Jan 2015, Josh Wu wrote:
> >>> On 1/7/2015 6:17 AM, Guennadi Liakhovetski wrote:
> >>>> On Tue, 6 Jan 2015, Josh Wu wrote:
> >>>>> Hi, Guennadi
> >>>>> 
> >>>>> After look deep into this patch, I found you miss one line that should
> >>>>> be changed as well.
> >>>>> It's In function v4l2_clk_get(), there still has one line code called
> >>>>> v4l2_clk_find(dev_id, id).
> >>>>> You need to change it to v4l2_clk_find(dev_id, NULL) as well.
> >>>>> Otherwise the code that many sensor used: v4l2_clk_get(&client->dev,
> >>>>> "mclk") cannot acquired the "mclk" clock.
> >>>>> 
> >>>>> After above changes, this patch works for me.
> >>>> 
> >>>> I think you're right, in fact, since we now don't store CCF-based
> >>>> v4l2_clk wrappers on the list, this can be simplified even further,
> >>>> I'll update the patch. Did you only test this patch or both?
> >>> 
> >>> I tested both patches with Atmel-isi driver. For the 2/2 patch I applied
> >>> the modification Laurent suggested.
> >>> Those patches works for me.
> >>> 
> >>> The only concern is in ov2640 I still need to acquired two v4l2 clocks:
> >>>     "xvclk"  that will get the xvclk CCF clock directly.
> >>>     "mclk"  that make ISI driver call his clock_start()/stop() to
> >>>     enable/disable ISI's peripheral clock.
> >>> 
> >>> If I only get xvclk clock, then the camera capture will be failed with a
> >>> ISI timeout error.
> >> 
> >> No, this doesn't look right to me. The camera sensor has only one clock
> >> input, so, it should only request one clock. Where does the clock signal
> >> to the camera come from on your system?
> > 
> > That's correct, the sensor driver only has one clock input, so it should
> > just request the xvclk clock.
> > 
> >> If it comes from the ISI itself, you don't need to specify the clock in
> >> the DT, since the ISI doesn't produce a clock from DT. If you do want to
> >> have your clock consumer (ov2640) and the supplier (ISI) properly
> >> described in DT, you'll have to teach the ISI to register a CCF clock
> >> source, which then will be connected to from the ov2640. If you choose
> >> not to show your clock in the DT, you can just use v4l2_clk_get(dev,
> >> "xvclk") and it will be handled by v4l2_clk / soc-camera / isi-atmel.
> >> 
> >> If the closk to ov2640 is supplied by a separate clock source, then you
> >> v4l2_clk_get() will connect ov2640 to it directly and soc-camera will
> >> enable and disable it on power-on / -off as required.
> > 
> > The ISI has no way to supply a sensor clock, the clock is supplied by a
> > separate clock source.
> > 
> >> From your above description it looks like the clock to ov2640 is
> >> supplied by a separate source, but atmel-isi's .clock_start() /
> >> .clock_stop() functions still need to be called? By looking at those
> >> functions it looks like they turn on and off clocks, supplying the ISI
> >> itself... Instead of only turning on and off clocks, provided by the ISI
> >> to a camera sensor. If my understanding is right, then this is a bug in
> >> atmel-isi and it has to be fixed.
> > 
> > That's correct as well, the ISI driver needs to be fixed.
> 
> Thanks both of you for the details. Now I got it.
> Indeed, I need fix this in atmel-isi driver not in ov2640 driver.
> So I will send a new patch for this, which should move the ISI
> peripheral clock enable/disable() from clock_start/stop() to
> isi_camera_add_device/remove_device().

Shouldn't you move it to the start_streaming() and stop_streaming() functions 
instead ? An even better solution would be to use runtime PM to enable/disable 
the ISI clock in the runtime PM resume and suspend handlers, and call 
pm_runtime_get_sync() and pm_runtime_put() when you need the ISI to be 
operational.

-- 
Regards,

Laurent Pinchart