Hi Thierry,

On Tuesday 26 November 2013 09:59:12 Thierry Reding wrote:

On Tue, Nov 26, 2013 at 02:54:41AM +0100, Laurent Pinchart wrote:

...

On Friday 22 November 2013 19:41:54 Thierry Reding wrote:

...

This binding specifies a set of common properties for display panels. It can be used as a basis by bindings for specific panels.

Bindings for three specific panels are provided to show how the simple panel binding can be used.

Signed-off-by: Thierry Reding treding@nvidia.com

This binding has already been discussed earlier. Both Laurent and Tomi seemed to be generally fine with this.

That's correct, I'm generally fine with your approach, but there's still one point I'd like to see addressed.

As I mentioned previously, I would like to avoid adding zillions of compatible properties to the driver, when we could use a single property in the DT bindings that would specify the timings instead. This would lower the amount of changes made to the simple panel driver, while keeping DT simple enough (at least in my opinion).

Specifying the full timings in every DT source file would be pretty verbose and could be error-prone. However, using a single property to specify one of the standard display timings, as orinally proposed by Hans de Goede (CC'ed) during a chat at the kernel summit would in my opinion be a good middle-ground solution.

Would you consider adding such a property to the simple panel bindings, and define a common compatible string ? Each panel should of course list its own compatibility string in addition to the common compatible string in case the need for extensions arises in the future.

My gripe with this is that it duplicates information. By definition a simple panel supports a limited number of display modes (typically only a single one), so once you know the compatible value (which needs to be there anyway) you can derive the mode from it. Adding a property that specifies the display mode is redundant.

But that could be said of pretty much every device :-) Let's take an example I came across today. The SPI DT bindings have a maximum clock speed property for every device. This is usually a property of the hardware, which could have been stored in the device drivers.

I don't particularly like expressing detailed device information in DT when the information is static for a given device and easily stored in the drivers. However, in this case, I believe the overhead of adding one mode property to DT is worth it, as there are zillion panel models in the wild. Going back to the SPI analogy, let's consider the m25p80 SPI flash memory driver. It supports many models, each of them implemented by a large number of manufacturers. The driver stores information for each model (such as the flash size), but the maximum SPI clock frequency is specified in DT as we don't want to have a list of all device models for all manufacturers in the driver (granted, one of the reasons to specify the max frequency in DT is that it can also be limited by the board, not only by the chip, but I believe my point remains valid).

Dave Airlie proposed something else, namely to keep a list of common modes within the driver and have each device reference that mode instead. I think that could work similarily well. It still requires the driver to be touched for each new panel, but the changes will be very small. They'll be more of the "add a new table entry" sort of change that we have in drivers like the 8250/16550-compatible serial. Most of that could probably be wrapped into a macro to make it concise.

That would be better, but not perfect :-) Given the very large number of panel models I would like to have a simple panel driver that doesn't need to be patched for every model. The odd cases can of course be handled in C, but the common case would really benefit from being handled in DT. Imagine how annoying it would be to update the USB mass storage driver with the VID:PID of every new USB flash device.

On Fri, Dec 06, 2013 at 02:58:00PM +0100, Thierry Reding wrote:

On Thu, Dec 05, 2013 at 12:45:17AM +0100, Laurent Pinchart wrote:

...

Hi Thierry,

On Tuesday 26 November 2013 09:59:12 Thierry Reding wrote:

...

On Tue, Nov 26, 2013 at 02:54:41AM +0100, Laurent Pinchart wrote:

...

On Friday 22 November 2013 19:41:54 Thierry Reding wrote:

...

This binding specifies a set of common properties for display panels. It can be used as a basis by bindings for specific panels.

Bindings for three specific panels are provided to show how the simple panel binding can be used.

Signed-off-by: Thierry Reding treding@nvidia.com

This binding has already been discussed earlier. Both Laurent and Tomi seemed to be generally fine with this.

That's correct, I'm generally fine with your approach, but there's still one point I'd like to see addressed.

As I mentioned previously, I would like to avoid adding zillions of compatible properties to the driver, when we could use a single property in the DT bindings that would specify the timings instead. This would lower the amount of changes made to the simple panel driver, while keeping DT simple enough (at least in my opinion).

Specifying the full timings in every DT source file would be pretty verbose and could be error-prone. However, using a single property to specify one of the standard display timings, as orinally proposed by Hans de Goede (CC'ed) during a chat at the kernel summit would in my opinion be a good middle-ground solution.

Would you consider adding such a property to the simple panel bindings, and define a common compatible string ? Each panel should of course list its own compatibility string in addition to the common compatible string in case the need for extensions arises in the future.

My gripe with this is that it duplicates information. By definition a simple panel supports a limited number of display modes (typically only a single one), so once you know the compatible value (which needs to be there anyway) you can derive the mode from it. Adding a property that specifies the display mode is redundant.

But that could be said of pretty much every device :-) Let's take an example I came across today. The SPI DT bindings have a maximum clock speed property for every device. This is usually a property of the hardware, which could have been stored in the device drivers.

No it isn't. It's a board-specific property. It depends on all sorts of factors such as how long the connections are between SPI master and slave as well as what other electrical components happen to be in that path.

...

I don't particularly like expressing detailed device information in DT when the information is static for a given device and easily stored in the drivers. However, in this case, I believe the overhead of adding one mode property to DT is worth it, as there are zillion panel models in the wild. Going back to the SPI analogy, let's consider the m25p80 SPI flash memory driver. It supports many models, each of them implemented by a large number of manufacturers. The driver stores information for each model (such as the flash size), but the maximum SPI clock frequency is specified in DT as we don't want to have a list of all device models for all manufacturers in the driver (granted, one of the reasons to specify the max frequency in DT is that it can also be limited by the board, not only by the chip, but I believe my point remains valid).

On the contrary, it completely invalidates your point.

And in fact m25p80 is another perfect example to prove my point. Having display modes encoded within the panel driver and selected depending on the compatible string isn't anything other than the m25p_ids table.

When you want to support a new type of device, you need to add a new entry into the m25p_ids table, just like you would add a new display mode to the panel driver.

...

...

Dave Airlie proposed something else, namely to keep a list of common modes within the driver and have each device reference that mode instead. I think that could work similarily well. It still requires the driver to be touched for each new panel, but the changes will be very small. They'll be more of the "add a new table entry" sort of change that we have in drivers like the 8250/16550-compatible serial. Most of that could probably be wrapped into a macro to make it concise.

That would be better, but not perfect :-) Given the very large number of panel models I would like to have a simple panel driver that doesn't need to be patched for every model. The odd cases can of course be handled in C, but the common case would really benefit from being handled in DT.

I really don't think the number of panels supported in mainline will be overly big.

I for once have hardly seen a display twice in different projects.

Even if, I will gladly take patches that add them to the driver. And it isn't even a lot of work, you really just need to copy this stuff from the datasheet. And the larger the number of panels the kernel supports, the easier it will get to add new support because you may either just be able to reuse the compatible string because somebody else already added support for it, or you may be able to reuse an existing mode because your panel happens to match that mode.

...

Imagine how annoying it would be to update the USB mass storage driver with the VID:PID of every new USB flash device.

The electrical interface of USB is well defined and it is further well specified what USB mass storage means. In addition, USB has very good support for probing capabilities and characteristics. The equivalent for panels would be something like EDID. If the panel provided EDID, then of course you don't need to specify the mode, neither in DT nor in the driver.

What makes EDID useful is that it contains the modes. We don't have to maintain a database of all monitors available out there in the kernel. Still EDID data contains a vendor/model string which makes it possible to add quirks for a specific monitor should we have to.

So I really vote for putting the mode data into dt for the 98% case and an exact vendor/model for the remaining quirks. This is simple anyway since we already have a binding for display modelines. Supporting this in the panel driver is trivial.

Sascha

On Tue, Nov 26, 2013 at 10:01:54AM +0800, Daniel Kurtz wrote:

On Sat, Nov 23, 2013 at 2:41 AM, Thierry Reding thierry.reding@gmail.com wrote:

...

This binding specifies a set of common properties for display panels. It can be used as a basis by bindings for specific panels.

Bindings for three specific panels are provided to show how the simple panel binding can be used.

[...]

...

+Example:

•   panel: panel {
  

•           compatible = "cptt,claa101wb01";
  

•           ddc-i2c-bus = <&panelddc>;
  

•           power-supply = <&vdd_pnl_reg>;
  

•           enable-gpios = <&gpio 90 0>;
  

•           backlight = <&backlight>;
  

•   };
  

Pardon the question if this has been discussed before, but, would this also be a good binding in which to store a panel's physical dimensions (mmWidth, mmHeight)? Or is there another binding more appropriate?

This binding doesn't define those properties because they are implied by the compatible string. The corresponding driver defines a structure like this:

struct panel_desc { const struct drm_display_mode *modes; unsigned int num_modes;

struct { unsigned int width; unsigned int height; } size; };

The panel_desc.size structure contains the physical panel dimensions, specified in mm. With that, a panel will be supported by defining something like this:

static const struct drm_display_mode foo_mode = { ... };

static const struct panel_desc foo = { .modes = &foo_mode, .num_modes = 1, .size { .width = 223, .height = 125, }, };

static const struct of_device_id platform_of_match[] = { ... { .compatible = "vendor,foo", .data = &foo, }, ... };

I assume you have a specific use-case in mind. Does this provide what you need? The driver patches were posted earlier[0][1]. They have changed slightly since then to address review comments, but nothing significant.

Thierry

[0]: http://www.spinics.net/lists/devicetree/msg08498.html [1]: http://www.spinics.net/lists/devicetree/msg08499.html

On 12/11/2013 11:16 PM, Tomi Valkeinen wrote:

• PGP Signed by an unknown key

On 2013-11-22 20:41, Thierry Reding wrote:

...

+Example:

• panel: panel {

• compatible = "cptt,claa101wb01";
  

• ddc-i2c-bus = <&panelddc>;
  

• power-supply = <&vdd_pnl_reg>;
  

• enable-gpios = <&gpio 90 0>;
  

• backlight = <&backlight>;
  

• };

I'm somewhat torn with this, as I agree with Thierry that it's correct to have a panel database in the driver, but, on the other hand, it does seem impractical.

In my experience, there are lots of panels out there, and each board I have has a different one. So, while just a gut feeling, we could end up with lots of panel, each used only on one board.

With a quick thought, things would work fine if we just added the videomode data to the DT data, instead of a driver database, as Laurent suggested.

However... I don't think the panels are usually as simple as that. With the panels I've worked with, the driver has to know things like:

• Does the power supply need to be enabled before the enable gpio, and

if so, how long before? And the same for power off.

• Does the video stream need to be enabled before the enable gpio, and

if so, how long before? And the same for power off.

• Is the gpio enable, power down, or reset? If reset, what are the timings.

Where will those be defined? This goes back to the power sequence stuff again... (Was the power sequences series forgotten?)

Sorry for the very late reply - power sequences are forgotten for now, but I don't mind reviving them if a clear use-case emerges. The main point of power seqs (at least in my mind) was to be able to define them in the DT to avoid things like the panel DB you mention. This idea has been dismissed for good reasons, and anyway in the case of the panel it is clear that this is not what we want to do.

Now if we make a power sequences framework without DT support, we will end up with something that would still require a panel database, and can anyway be expressed by functions. The gain would be automated error-handling, and reduced footprint for power sequences. Not sure that's enough to justify resurrecting the power seqs.

Alex.