Hey Maxime, On Sat, 4 Apr 2020 at 11:34, Maxime Ripard <maxime@xxxxxxxxxx> wrote: > > Hi, > > On Sat, Apr 04, 2020 at 02:27:36AM +0300, Sakari Ailus wrote: > > Hi Robert, > > > > On Thu, Apr 02, 2020 at 12:10:00PM +0200, Robert Foss wrote: > > > Hey Maxime, > > > > > > On Wed, 1 Apr 2020 at 10:07, Maxime Ripard <maxime@xxxxxxxxxx> wrote: > > > > > > > > Hi, > > > > > > > > On Tue, Mar 31, 2020 at 03:33:44PM +0200, Robert Foss wrote: > > > > > From: Dongchun Zhu <dongchun.zhu@xxxxxxxxxxxx> > > > > > > > > > > This patch adds documentation of device tree in YAML schema for the > > > > > OV8856 CMOS image sensor. > > > > > > > > > > Signed-off-by: Dongchun Zhu <dongchun.zhu@xxxxxxxxxxxx> > > > > > Signed-off-by: Robert Foss <robert.foss@xxxxxxxxxx> > > > > > --- > > > > > > > > > > - Changes since v5: > > > > > * Add assigned-clocks and assigned-clock-rates > > > > > * robher: dt-schema errors > > > > > > > > > > - Changes since v4: > > > > > * Fabio: Change reset-gpio to GPIO_ACTIVE_LOW, explain in description > > > > > * Add clock-lanes property to example > > > > > * robher: Fix syntax error in devicetree example > > > > > > > > > > - Changes since v3: > > > > > * robher: Fix syntax error > > > > > * robher: Removed maxItems > > > > > * Fixes yaml 'make dt-binding-check' errors > > > > > > > > > > - Changes since v2: > > > > > Fixes comments from from Andy, Tomasz, Sakari, Rob. > > > > > * Convert text documentation to YAML schema. > > > > > > > > > > - Changes since v1: > > > > > Fixes comments from Sakari, Tomasz > > > > > * Add clock-frequency and link-frequencies in DT > > > > > > > > > > .../devicetree/bindings/media/i2c/ov8856.yaml | 150 ++++++++++++++++++ > > > > > MAINTAINERS | 1 + > > > > > 2 files changed, 151 insertions(+) > > > > > create mode 100644 Documentation/devicetree/bindings/media/i2c/ov8856.yaml > > > > > > > > > > diff --git a/Documentation/devicetree/bindings/media/i2c/ov8856.yaml b/Documentation/devicetree/bindings/media/i2c/ov8856.yaml > > > > > new file mode 100644 > > > > > index 000000000000..beeddfbb8709 > > > > > --- /dev/null > > > > > +++ b/Documentation/devicetree/bindings/media/i2c/ov8856.yaml > > > > > @@ -0,0 +1,150 @@ > > > > > +# SPDX-License-Identifier: (GPL-2.0 OR BSD-2-Clause) > > > > > +# Copyright (c) 2019 MediaTek Inc. > > > > > +%YAML 1.2 > > > > > +--- > > > > > +$id: http://devicetree.org/schemas/media/i2c/ov8856.yaml# > > > > > +$schema: http://devicetree.org/meta-schemas/core.yaml# > > > > > + > > > > > +title: Omnivision OV8856 CMOS Sensor Device Tree Bindings > > > > > + > > > > > +maintainers: > > > > > + - Ben Kao <ben.kao@xxxxxxxxx> > > > > > + - Dongchun Zhu <dongchun.zhu@xxxxxxxxxxxx> > > > > > + > > > > > +description: |- > > > > > + The Omnivision OV8856 is a high performance, 1/4-inch, 8 megapixel, CMOS > > > > > + image sensor that delivers 3264x2448 at 30fps. It provides full-frame, > > > > > + sub-sampled, and windowed 10-bit MIPI images in various formats via the > > > > > + Serial Camera Control Bus (SCCB) interface. This chip is programmable > > > > > + through I2C and two-wire SCCB. The sensor output is available via CSI-2 > > > > > + serial data output (up to 4-lane). > > > > > + > > > > > +properties: > > > > > + compatible: > > > > > + const: ovti,ov8856 > > > > > + > > > > > + reg: > > > > > + maxItems: 1 > > > > > + > > > > > + clocks: > > > > > + maxItems: 1 > > > > > + > > > > > + clock-names: > > > > > + description: > > > > > + Input clock for the sensor. > > > > > + items: > > > > > + - const: xvclk > > > > > + > > > > > + clock-frequency: > > > > > + description: > > > > > + Frequency of the xvclk clock in Hertz. > > > > > > > > We also had that discussion recently for another omnivision sensor > > > > (ov5645 iirc), but what is clock-frequency useful for? > > > > > > > > It seems that the sensor is passed in clocks, so if you need to > > > > retrieve the clock rate you should use the clock API instead. > > > > > > > > Looking at the driver, it looks like it first retrieves the clock, set > > > > it to clock-frequency, and then checks that this is OV8856_XVCLK_19_2 > > > > (19.2 MHz). > > > > > > As far as I understand it, 19.2MHz is requirement for the sensor mode > > > that currently defaults to. Some modes require higher clock speeds > > > than this however. > > > > It's very system specific. Either way, bindings should not assume a > > particular driver implementation. > > > > > > > > > > > > > The datasheet says that the sensor can have any frequency in the 6 - > > > > 27 MHz range, so this is a driver limitation and should be set in the > > > > driver using the clock API, and you can always bail out if it doesn't > > > > provide a rate that is not acceptable for the drivers assumption. > > > > > > > > In any case, you don't need clock-frequency here... > > > > > > So your suggestion is that we remove all clocks-rate properties, and > > > replace the clk_get_rate() calls in the driver with clk_set_rate() > > > calls for the desired frequencies? > > > > The driver shouldn't set the rate here unless it gets it from DT (but that > > was not the intention). So the driver should get the frequency instead. > > I'm actually saying the opposite :) > > Like you were saying, the binding (or DT, for that matter) shouldn't > assume a particular driver implementation. > > So one corollary is that if the driver has some restrictions in Linux, > it shouldn't be part of the binding, right? > > This binding uses multiple clock properties but as far as I can see, > the driver retrieves a clock using clocks and makes sure that its rate > match its limitation of 19.2MHz using clock-frequency (which is > redundant on a clk_get_rate on the clocks provided earlier). > > I'm suspecting that the parent clock on multiple SoCs can be > configured and is not a fixed rate crystal, so assigned-clocks-rate is > here just to make sure we set the frequency at the one being checked > in the driver's probe so that it all works. > > But that 19.2MHz is not a limitation of the device itself, it's a > limitation of our implementation, so we can instead implement > something equivalent in Linux using a clk_set_rate to 19.2MHz (to make > sure that our parent clock is configured at the right rate) and the > clk_get_rate and compare that to 19.2MHz (to make sure that it's not > been rounded too far apart from the frequency we expect). > > This is doing exactly the same thing, except that we don't encode our > implementation limitations in the DT, but in the driver instead. > Thanks for taking the time to explain this. I'll spin a new revision that moves the clock rate handling into the driver.
