[PATCH v3 05/11] drm/bridge: Add Analogix anx6345 support

[Vasily Khoruzhick]

On Fri, Feb 15, 2019 at 1:13 AM Andrzej Hajda <a.hajda at samsung.com> wrote:

Hi Andrzej,

Thanks for review!

> > +#include <linux/of_gpio.h>
> Do you need this header?

I'll drop it.

> > +#include <drm/drmP.h>
>
> drmP.h is/should be deprecated.

Same here

> > +struct anx6345_platform_data {
> > +     struct regulator *dvdd12;
> > +     struct regulator *dvdd25;
> > +     struct gpio_desc *gpiod_reset;
> > +};
>
> Why do you need this struct, why just do not embed it's fields directly
> into struct anx6345 ?

OK, I'll embed it into struct anx6345

> > +     if (WARN_ON(anx6345->powered))
> > +             return;
>
> It should not happen, you can remove this warn.

OK

> > +     if (pdata->dvdd12) {
>
> If regulators are required this will be never null.

Right, and regulator subsystem will return dummy regulator if it's
missing in dts.
I'll remove redundant checks.

> > +
> > +     if (pdata->dvdd25) {
>
> ditto

OK

> > +
> > +     if (anx6345->panel)
> > +             drm_panel_prepare(anx6345->panel);
>
> again, here and below: panel is never null, check can be removed.

That's not true, panel is optional. It can be DP connector, not a panel.

> > +
> > +     gpiod_set_value_cansleep(pdata->gpiod_reset, 0);
> > +     usleep_range(1000, 2000);
> > +
> > +     gpiod_set_value_cansleep(pdata->gpiod_reset, 1);
>
>
> Start/stop sequence seems odd regarding reset gpio:
>
> 1. In probe reset is set to low, in poweroff to high - incosistent.
>
> 2. If in case of disabled device reset should be 0, there is no point to
> set it again to 0 three lines above.
>
> 3. I suspect in dts reset gpio should be declared as active_low, and the
> logic in the driver should be reverted, in power off it should be set to
> high, in power on it should be lowered (logically).

OK, I'll look into it.

> > +err_poweroff:
> > +     DRM_ERROR("Failed DisplayPort transmitter initialization: %d\n", err);
>
> redundant message

OK, will drop.

> > +             DRM_ERROR("Get sink count failed %d\n", err);
>
> The rule of thumb I heard is that if you start message capitalized you
> should end with dot. Since I do not know if it is enforced in kernel I
> leave the decision up to you.

I grepped DRM_ERROR in driver/gpu/drm and they do exactly the same as here.
So I'll just keep it as is for consistency.

> > +static bool anx6345_bridge_mode_fixup(struct drm_bridge *bridge,
> > +                                   const struct drm_display_mode *mode,
> > +                                   struct drm_display_mode *adjusted_mode)
> > +{
> > +     if (mode->flags & DRM_MODE_FLAG_INTERLACE)
> > +             return false;
> > +
> > +     /* Max 1200p at 5.4 Ghz, one lane */
> > +     if (mode->clock > 154000)
> > +             return false;
>
> These checks should be in mode_valid callback.

OK

> > +     /* Map slave addresses of ANX6345 */
> > +     for (i = 0; i < I2C_NUM_ADDRESSES; i++) {
> > +             if (anx6345_i2c_addresses[i] >> 1 != client->addr)
> > +                     anx6345->i2c_clients[i] = i2c_new_dummy(client->adapter,
> > +                                             anx6345_i2c_addresses[i] >> 1);
> > +             else
> > +                     anx6345->i2c_clients[i] = client;
>
>
> I see this contredanse is copy/pasted from anx78*, but it looks quite
> complicated. As I understand there are two i2c addresses, why we cannot
> assume one address is for control interfaces and another  is dummy? It would
> simplify the code here and in other places.

Sorry, I don't get you, could you elaborate? Note that anx6345 uses
both addresses,
i2c_new_dummy() just registers new i2c device bound to a dummy driver and it's
supposed to be used for devices that consume more than one i2c address.

> > +     if (!found) {
> > +             DRM_ERROR("ANX%x (ver. %d) not supported by this driver\n",
> > +                       anx6345->chipid, version);
> > +             err = -ENODEV;
> > +             goto err_poweroff;
> > +     }
>
>
> As I see chip becomes powered forever, is it OK? Usually it should be
> powered only when pipeline starts, and powered-off after pipeline stops.

I'll look into how hard it would be to implement but personally I
think it's OK for now.
We can add more sophisticated power management once this driver is merged.