[1/9] dt-bindings: mxsfb: Add compatible for i.MX8MP

Commit Message

Marek Vasut Feb. 28, 2022, 12:45 a.m. UTC

Add compatible string for i.MX8MP LCDIF variant. This is called LCDIFv3
and is completely different from the LCDIFv3 found in i.MX23 in that it
has a completely scrambled register layout compared to all previous LCDIF
variants. The new LCDIFv3 also supports 36bit address space. However,
except for the complete bit reshuffling, this is still LCDIF and it still
works like one.

Signed-off-by: Marek Vasut <[e-mail]>
Cc: Alexander Stein <[e-mail]>
Cc: Laurent Pinchart <[e-mail]>
Cc: Lucas Stach <[e-mail]>
Cc: Peng Fan <[e-mail]>
Cc: Rob Herring <[e-mail]>
Cc: Robby Cai <[e-mail]>
Cc: Sam Ravnborg <[e-mail]>
Cc: Stefan Agner <[e-mail]>
Cc: [e-mail]
---
 Documentation/devicetree/bindings/display/fsl,lcdif.yaml | 1 +
 1 file changed, 1 insertion(+)

Comments

Liu Ying Feb. 28, 2022, 6:37 a.m. UTC | #1

Hi Marek,

On Mon, 2022-02-28 at 01:45 +0100, Marek Vasut wrote:
> Add compatible string for i.MX8MP LCDIF variant. This is called LCDIFv3
> and is completely different from the LCDIFv3 found in i.MX23 in that it

In i.MX23 reference manual, there is no LCDIFv3 found, but only LCDIF.

> has a completely scrambled register layout compared to all previous LCDIF

It looks like no single register of i.MX8MP LCDIFv3 overlaps with
registers in other i.MX2x/6x/7x/8x LCDIFs. The LCDIFv3 block diagram is
totally different from the LCDIF block diagram, according to the SoC
reference manuals. LCDIFv3 supports SHADOW_EN bit to update horizontal
and vertical size of graphic, position of graphic on the panel, address
of graphic in memory and color formats or color palettes, which is not
supported by LCDIF and impacts display driver control mechanism
considerably. LCDIF supports DOTCLK interface, MPU interface and VSYNC
interface, while LCDIFv3 only supports parallel output as a counterpart
of the DOTCLK interface.

Generally speaking, LCDIFv3 is just a new display IP which happens to
have the word 'LCDIF' in its name.  Although both of LCDIFv3 and LCDIF
are display controllers for scanning out frames onto display devices, I
don't think they are in one family.

So, LCDIFv3 deserves a new separate dt-binding, IMO.

> variants. The new LCDIFv3 also supports 36bit address space. However,
> except for the complete bit reshuffling, this is still LCDIF and it still
> works like one.
> 
> Signed-off-by: Marek Vasut <[e-mail]>
> Cc: Alexander Stein <[e-mail]>
> Cc: Laurent Pinchart <[e-mail]>
> Cc: Lucas Stach <[e-mail]>
> Cc: Peng Fan <[e-mail]>
> Cc: Rob Herring <[e-mail]>
> Cc: Robby Cai <[e-mail]>
> Cc: Sam Ravnborg <[e-mail]>
> Cc: Stefan Agner <[e-mail]>
> Cc: [e-mail]
> ---
>  Documentation/devicetree/bindings/display/fsl,lcdif.yaml | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/Documentation/devicetree/bindings/display/fsl,lcdif.yaml b/Documentation/devicetree/bindings/display/fsl,lcdif.yaml
> index 900a56cae80e6..9831ab53a053d 100644
> --- a/Documentation/devicetree/bindings/display/fsl,lcdif.yaml
> +++ b/Documentation/devicetree/bindings/display/fsl,lcdif.yaml
> @@ -28,6 +28,7 @@ properties:
>                - fsl,imx7d-lcdif
>                - fsl,imx8mm-lcdif
>                - fsl,imx8mn-lcdif
> +              - fsl,imx8mp-lcdif

Even if LCDIFv3 is covered by this dt-binding(which is obviously not
the case), 'fsl,imx8mp-lcdif' should be after 'fsl,imx6x-lcdif' as an
enum, otherwise LCDIFv3 is compatible to LCDIF.

Regards,
Liu Ying

>                - fsl,imx8mq-lcdif
>            - const: fsl,imx6sx-lcdif
>

Laurent Pinchart Feb. 28, 2022, 6:40 a.m. UTC | #2

Hi Marek,

Thank you for the patch.

On Mon, Feb 28, 2022 at 01:45:57AM +0100, Marek Vasut wrote:
> Add compatible string for i.MX8MP LCDIF variant. This is called LCDIFv3
> and is completely different from the LCDIFv3 found in i.MX23 in that it
> has a completely scrambled register layout compared to all previous LCDIF
> variants. The new LCDIFv3 also supports 36bit address space. However,
> except for the complete bit reshuffling, this is still LCDIF and it still
> works like one.
> 
> Signed-off-by: Marek Vasut <[e-mail]>
> Cc: Alexander Stein <[e-mail]>
> Cc: Laurent Pinchart <[e-mail]>
> Cc: Lucas Stach <[e-mail]>
> Cc: Peng Fan <[e-mail]>
> Cc: Rob Herring <[e-mail]>
> Cc: Robby Cai <[e-mail]>
> Cc: Sam Ravnborg <[e-mail]>
> Cc: Stefan Agner <[e-mail]>
> Cc: [e-mail]
> ---
>  Documentation/devicetree/bindings/display/fsl,lcdif.yaml | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/Documentation/devicetree/bindings/display/fsl,lcdif.yaml b/Documentation/devicetree/bindings/display/fsl,lcdif.yaml
> index 900a56cae80e6..9831ab53a053d 100644
> --- a/Documentation/devicetree/bindings/display/fsl,lcdif.yaml
> +++ b/Documentation/devicetree/bindings/display/fsl,lcdif.yaml
> @@ -28,6 +28,7 @@ properties:
>                - fsl,imx7d-lcdif
>                - fsl,imx8mm-lcdif
>                - fsl,imx8mn-lcdif
> +              - fsl,imx8mp-lcdif

As the hardware isn't backward-compatible with any other version, I
think the new compatible string should go in the previous enum block,
not in this one. We don't want the imx6sx fallback.

>                - fsl,imx8mq-lcdif
>            - const: fsl,imx6sx-lcdif
>

Marek Vasut Feb. 28, 2022, 6:57 a.m. UTC | #3

On 2/28/22 07:37, Liu Ying wrote:
> Hi Marek,

Hi,

> On Mon, 2022-02-28 at 01:45 +0100, Marek Vasut wrote:
>> Add compatible string for i.MX8MP LCDIF variant. This is called LCDIFv3
>> and is completely different from the LCDIFv3 found in i.MX23 in that it
> 
> In i.MX23 reference manual, there is no LCDIFv3 found, but only LCDIF.

See i.MX23 HW_LCDIF_VERSION MAJOR=0x3 , that's LCDIF V3 . MX28 has LCDIF 
V4 .

>> has a completely scrambled register layout compared to all previous LCDIF
> 
> It looks like no single register of i.MX8MP LCDIFv3 overlaps with
> registers in other i.MX2x/6x/7x/8x LCDIFs. The LCDIFv3 block diagram is
> totally different from the LCDIF block diagram, according to the SoC
> reference manuals. LCDIFv3 supports SHADOW_EN bit to update horizontal
> and vertical size of graphic, position of graphic on the panel, address
> of graphic in memory and color formats or color palettes, which is not
> supported by LCDIF and impacts display driver control mechanism
> considerably. LCDIF supports DOTCLK interface, MPU interface and VSYNC
> interface, while LCDIFv3 only supports parallel output as a counterpart
> of the DOTCLK interface.
> 
> Generally speaking, LCDIFv3 is just a new display IP which happens to
> have the word 'LCDIF' in its name.  Although both of LCDIFv3 and LCDIF
> are display controllers for scanning out frames onto display devices, I
> don't think they are in one family.
> 
> So, LCDIFv3 deserves a new separate dt-binding, IMO.

It seems to me a lot of those bits just map to their previous 
equivalents in older LCDIF, others were dropped, so this is some sort of 
new LCDIF mutation, is it not ?

I am aware NXP has a separate driver in its downstream, but I'm not 
convinced the duplication of boilerplate code by introducing a separate 
driver for what looks like another LCDIF variant is the right approach.

>> variants. The new LCDIFv3 also supports 36bit address space. However,
>> except for the complete bit reshuffling, this is still LCDIF and it still
>> works like one.

[...]

Liu Ying Feb. 28, 2022, 8:18 a.m. UTC | #4

On Mon, 2022-02-28 at 07:57 +0100, Marek Vasut wrote:
> On 2/28/22 07:37, Liu Ying wrote:
> > Hi Marek,
> 
> Hi,
> 
> > On Mon, 2022-02-28 at 01:45 +0100, Marek Vasut wrote:
> > > Add compatible string for i.MX8MP LCDIF variant. This is called LCDIFv3
> > > and is completely different from the LCDIFv3 found in i.MX23 in that it
> > 
> > In i.MX23 reference manual, there is no LCDIFv3 found, but only LCDIF.
> 
> See i.MX23 HW_LCDIF_VERSION MAJOR=0x3 , that's LCDIF V3 . MX28 has LCDIF 
> V4 .

Ok, got it now. AFAIK, the SoC design team calls i.MX8MP display
controller as 'LCDIFv3'. Those in other SoCs are called 'LCDIF'.  There
is not even a register in i.MX8MP display controller to decribe the
version.

> 
> > > has a completely scrambled register layout compared to all previous LCDIF
> > 
> > It looks like no single register of i.MX8MP LCDIFv3 overlaps with
> > registers in other i.MX2x/6x/7x/8x LCDIFs. The LCDIFv3 block diagram is
> > totally different from the LCDIF block diagram, according to the SoC
> > reference manuals. LCDIFv3 supports SHADOW_EN bit to update horizontal
> > and vertical size of graphic, position of graphic on the panel, address
> > of graphic in memory and color formats or color palettes, which is not
> > supported by LCDIF and impacts display driver control mechanism
> > considerably. LCDIF supports DOTCLK interface, MPU interface and VSYNC
> > interface, while LCDIFv3 only supports parallel output as a counterpart
> > of the DOTCLK interface.
> > 
> > Generally speaking, LCDIFv3 is just a new display IP which happens to
> > have the word 'LCDIF' in its name.  Although both of LCDIFv3 and LCDIF
> > are display controllers for scanning out frames onto display devices, I
> > don't think they are in one family.
> > 
> > So, LCDIFv3 deserves a new separate dt-binding, IMO.
> 
> It seems to me a lot of those bits just map to their previous 
> equivalents in older LCDIF, others were dropped, so this is some sort of 
> new LCDIF mutation, is it not ?

I say 'LCDIFv3' and 'LCDIF' are totally two IPs, if I compare the names
of registers and the names of register bits . 

> 
> I am aware NXP has a separate driver in its downstream, but I'm not 
> convinced the duplication of boilerplate code by introducing a separate 
> driver for what looks like another LCDIF variant is the right approach.

Hmmm, given the two IPs, I think there should be separate drivers.
 With one single driver, there would be too many 'if/else' checks to
separate the logics for the IPs, just like Patch 9/9 does.  The
boilerplate code to do things like registering a drm device is
acceptable, IMO.

Aside from that, with separate drivers, we don't have to test too many
SoCs if we only want to touch either 'LCDIFv3' or 'LCDIF'. 

> 
> > > variants. The new LCDIFv3 also supports 36bit address space. However,
> > > except for the complete bit reshuffling, this is still LCDIF and it still
> > > works like one.
> 
> [...]

Marek Vasut Feb. 28, 2022, 3:34 p.m. UTC | #5

On 2/28/22 09:18, Liu Ying wrote:

Hi,

>>> On Mon, 2022-02-28 at 01:45 +0100, Marek Vasut wrote:
>>>> Add compatible string for i.MX8MP LCDIF variant. This is called LCDIFv3
>>>> and is completely different from the LCDIFv3 found in i.MX23 in that it
>>>
>>> In i.MX23 reference manual, there is no LCDIFv3 found, but only LCDIF.
>>
>> See i.MX23 HW_LCDIF_VERSION MAJOR=0x3 , that's LCDIF V3 . MX28 has LCDIF
>> V4 .
> 
> Ok, got it now. AFAIK, the SoC design team calls i.MX8MP display
> controller as 'LCDIFv3'. Those in other SoCs are called 'LCDIF'.  There
> is not even a register in i.MX8MP display controller to decribe the
> version.

We also don't have a version register on MX6SX and we call it LCDIF V6 
in the driver. The naming scheme is confusing.

>>>> has a completely scrambled register layout compared to all previous LCDIF
>>>
>>> It looks like no single register of i.MX8MP LCDIFv3 overlaps with
>>> registers in other i.MX2x/6x/7x/8x LCDIFs. The LCDIFv3 block diagram is
>>> totally different from the LCDIF block diagram, according to the SoC
>>> reference manuals. LCDIFv3 supports SHADOW_EN bit to update horizontal
>>> and vertical size of graphic, position of graphic on the panel, address
>>> of graphic in memory and color formats or color palettes, which is not
>>> supported by LCDIF and impacts display driver control mechanism
>>> considerably. LCDIF supports DOTCLK interface, MPU interface and VSYNC
>>> interface, while LCDIFv3 only supports parallel output as a counterpart
>>> of the DOTCLK interface.
>>>
>>> Generally speaking, LCDIFv3 is just a new display IP which happens to
>>> have the word 'LCDIF' in its name.  Although both of LCDIFv3 and LCDIF
>>> are display controllers for scanning out frames onto display devices, I
>>> don't think they are in one family.
>>>
>>> So, LCDIFv3 deserves a new separate dt-binding, IMO.
>>
>> It seems to me a lot of those bits just map to their previous
>> equivalents in older LCDIF, others were dropped, so this is some sort of
>> new LCDIF mutation, is it not ?
> 
> I say 'LCDIFv3' and 'LCDIF' are totally two IPs, if I compare the names
> of registers and the names of register bits .
> 
>>
>> I am aware NXP has a separate driver in its downstream, but I'm not
>> convinced the duplication of boilerplate code by introducing a separate
>> driver for what looks like another LCDIF variant is the right approach.
> 
> Hmmm, given the two IPs, I think there should be separate drivers.
>   With one single driver, there would be too many 'if/else' checks to
> separate the logics for the IPs, just like Patch 9/9 does.  The
> boilerplate code to do things like registering a drm device is
> acceptable, IMO.
> 
> Aside from that, with separate drivers, we don't have to test too many
> SoCs if we only want to touch either 'LCDIFv3' or 'LCDIF'.

But then, with two drivers, you also might miss fixes which get applied 
to one driver and not the other, eventually the two drivers will diverge 
and that's not good.

I might wait for opinion from the others whether this should be one or 
two drivers.

btw is there any plan to have LCDIFv4 or this LCDIFv3 in some other SoC 
than iMX8MP ?

Liu Ying March 1, 2022, 2:44 a.m. UTC | #6

On Mon, 2022-02-28 at 16:34 +0100, Marek Vasut wrote:
> On 2/28/22 09:18, Liu Ying wrote:
> 
> Hi,

Hi,

> 
> > > > On Mon, 2022-02-28 at 01:45 +0100, Marek Vasut wrote:
> > > > > Add compatible string for i.MX8MP LCDIF variant. This is called LCDIFv3
> > > > > and is completely different from the LCDIFv3 found in i.MX23 in that it
> > > > 
> > > > In i.MX23 reference manual, there is no LCDIFv3 found, but only LCDIF.
> > > 
> > > See i.MX23 HW_LCDIF_VERSION MAJOR=0x3 , that's LCDIF V3 . MX28 has LCDIF
> > > V4 .
> > 
> > Ok, got it now. AFAIK, the SoC design team calls i.MX8MP display
> > controller as 'LCDIFv3'. Those in other SoCs are called 'LCDIF'.  There
> > is not even a register in i.MX8MP display controller to decribe the
> > version.
> 
> We also don't have a version register on MX6SX and we call it LCDIF V6 
> in the driver. The naming scheme is confusing.

It looks ok for the current mxsfb drm driver to use its own version
tracking mechanism to distinguish kinda small difference across LCDIF
variants.  However, LCDIFv3 in i.MX8mp is a totally different IP, which
does not apply to the tracking mechanism.

> 
> > > > > has a completely scrambled register layout compared to all previous LCDIF
> > > > 
> > > > It looks like no single register of i.MX8MP LCDIFv3 overlaps with
> > > > registers in other i.MX2x/6x/7x/8x LCDIFs. The LCDIFv3 block diagram is
> > > > totally different from the LCDIF block diagram, according to the SoC
> > > > reference manuals. LCDIFv3 supports SHADOW_EN bit to update horizontal
> > > > and vertical size of graphic, position of graphic on the panel, address
> > > > of graphic in memory and color formats or color palettes, which is not
> > > > supported by LCDIF and impacts display driver control mechanism
> > > > considerably. LCDIF supports DOTCLK interface, MPU interface and VSYNC
> > > > interface, while LCDIFv3 only supports parallel output as a counterpart
> > > > of the DOTCLK interface.
> > > > 
> > > > Generally speaking, LCDIFv3 is just a new display IP which happens to
> > > > have the word 'LCDIF' in its name.  Although both of LCDIFv3 and LCDIF
> > > > are display controllers for scanning out frames onto display devices, I
> > > > don't think they are in one family.
> > > > 
> > > > So, LCDIFv3 deserves a new separate dt-binding, IMO.
> > > 
> > > It seems to me a lot of those bits just map to their previous
> > > equivalents in older LCDIF, others were dropped, so this is some sort of
> > > new LCDIF mutation, is it not ?
> > 
> > I say 'LCDIFv3' and 'LCDIF' are totally two IPs, if I compare the names
> > of registers and the names of register bits .
> > 
> > > I am aware NXP has a separate driver in its downstream, but I'm not
> > > convinced the duplication of boilerplate code by introducing a separate
> > > driver for what looks like another LCDIF variant is the right approach.
> > 
> > Hmmm, given the two IPs, I think there should be separate drivers.
> >   With one single driver, there would be too many 'if/else' checks to
> > separate the logics for the IPs, just like Patch 9/9 does.  The
> > boilerplate code to do things like registering a drm device is
> > acceptable, IMO.
> > 
> > Aside from that, with separate drivers, we don't have to test too many
> > SoCs if we only want to touch either 'LCDIFv3' or 'LCDIF'.
> 
> But then, with two drivers, you also might miss fixes which get applied 
> to one driver and not the other, eventually the two drivers will diverge 
> and that's not good.

Given the two totally different IPs, I don't see bugs of IP control
logics should be fixed for both drivers. Naturally, the two would
diverge due to different HWs. Looking at Patch 9/9, it basically
squashes code to control LCDIFv3 into the mxsfb drm driver with
'if/else' checks(barely no common control code), which is hard to
maintain and not able to achieve good scalability for both 'LCDIFv3'
and 'LCDIF'.

> 
> I might wait for opinion from the others whether this should be one or 
> two drivers.
> 
> btw is there any plan to have LCDIFv4 or this LCDIFv3 in some other SoC 
> than iMX8MP ?

I don't think this kind of information significantly impacts the
decision on two drivers or one.  And, I cannot share that until the
company unveils.

Lucas Stach March 1, 2022, 10:04 a.m. UTC | #7

Hi Marek, hi Liu,

Am Dienstag, dem 01.03.2022 um 10:44 +0800 schrieb Liu Ying:
> On Mon, 2022-02-28 at 16:34 +0100, Marek Vasut wrote:
> > On 2/28/22 09:18, Liu Ying wrote:
> > 
> > Hi,
> 
> Hi,
> 
> > 
> > > > > On Mon, 2022-02-28 at 01:45 +0100, Marek Vasut wrote:
> > > > > > Add compatible string for i.MX8MP LCDIF variant. This is called LCDIFv3
> > > > > > and is completely different from the LCDIFv3 found in i.MX23 in that it
> > > > > 
> > > > > In i.MX23 reference manual, there is no LCDIFv3 found, but only LCDIF.
> > > > 
> > > > See i.MX23 HW_LCDIF_VERSION MAJOR=0x3 , that's LCDIF V3 . MX28 has LCDIF
> > > > V4 .
> > > 
> > > Ok, got it now. AFAIK, the SoC design team calls i.MX8MP display
> > > controller as 'LCDIFv3'. Those in other SoCs are called 'LCDIF'.  There
> > > is not even a register in i.MX8MP display controller to decribe the
> > > version.
> > 
> > We also don't have a version register on MX6SX and we call it LCDIF V6 
> > in the driver. The naming scheme is confusing.
> 
> It looks ok for the current mxsfb drm driver to use its own version
> tracking mechanism to distinguish kinda small difference across LCDIF
> variants.  However, LCDIFv3 in i.MX8mp is a totally different IP, which
> does not apply to the tracking mechanism.
> 
> > 
> > > > > > has a completely scrambled register layout compared to all previous LCDIF
> > > > > 
> > > > > It looks like no single register of i.MX8MP LCDIFv3 overlaps with
> > > > > registers in other i.MX2x/6x/7x/8x LCDIFs. The LCDIFv3 block diagram is
> > > > > totally different from the LCDIF block diagram, according to the SoC
> > > > > reference manuals. LCDIFv3 supports SHADOW_EN bit to update horizontal
> > > > > and vertical size of graphic, position of graphic on the panel, address
> > > > > of graphic in memory and color formats or color palettes, which is not
> > > > > supported by LCDIF and impacts display driver control mechanism
> > > > > considerably. LCDIF supports DOTCLK interface, MPU interface and VSYNC
> > > > > interface, while LCDIFv3 only supports parallel output as a counterpart
> > > > > of the DOTCLK interface.
> > > > > 
> > > > > Generally speaking, LCDIFv3 is just a new display IP which happens to
> > > > > have the word 'LCDIF' in its name.  Although both of LCDIFv3 and LCDIF
> > > > > are display controllers for scanning out frames onto display devices, I
> > > > > don't think they are in one family.
> > > > > 
> > > > > So, LCDIFv3 deserves a new separate dt-binding, IMO.
> > > > 
> > > > It seems to me a lot of those bits just map to their previous
> > > > equivalents in older LCDIF, others were dropped, so this is some sort of
> > > > new LCDIF mutation, is it not ?
> > > 
> > > I say 'LCDIFv3' and 'LCDIF' are totally two IPs, if I compare the names
> > > of registers and the names of register bits .
> > > 
> > > > I am aware NXP has a separate driver in its downstream, but I'm not
> > > > convinced the duplication of boilerplate code by introducing a separate
> > > > driver for what looks like another LCDIF variant is the right approach.
> > > 
> > > Hmmm, given the two IPs, I think there should be separate drivers.
> > >   With one single driver, there would be too many 'if/else' checks to
> > > separate the logics for the IPs, just like Patch 9/9 does.  The
> > > boilerplate code to do things like registering a drm device is
> > > acceptable, IMO.
> > > 
> > > Aside from that, with separate drivers, we don't have to test too many
> > > SoCs if we only want to touch either 'LCDIFv3' or 'LCDIF'.
> > 
> > But then, with two drivers, you also might miss fixes which get applied 
> > to one driver and not the other, eventually the two drivers will diverge 
> > and that's not good.
> 
> Given the two totally different IPs, I don't see bugs of IP control
> logics should be fixed for both drivers. Naturally, the two would
> diverge due to different HWs. Looking at Patch 9/9, it basically
> squashes code to control LCDIFv3 into the mxsfb drm driver with
> 'if/else' checks(barely no common control code), which is hard to
> maintain and not able to achieve good scalability for both 'LCDIFv3'
> and 'LCDIF'.

I tend to agree with Liu here. Writing a DRM driver isn't that much
boilerplate anymore with all the helpers we have available in the
framework today.

The IP is so different from the currently supported LCDIF controllers
that I think trying to support this one in the existing driver actually
increases the chances to break something when modifying the driver in
the future. Not everyone is able to test all LCDIF versions. My vote is
on having a separate driver for the i.MX8MP LCDIF.

Regards,
Lucas

Marek Vasut March 1, 2022, 10:19 a.m. UTC | #8

On 3/1/22 11:04, Lucas Stach wrote:

Hi,

[...]

>> Given the two totally different IPs, I don't see bugs of IP control
>> logics should be fixed for both drivers. Naturally, the two would
>> diverge due to different HWs. Looking at Patch 9/9, it basically
>> squashes code to control LCDIFv3 into the mxsfb drm driver with
>> 'if/else' checks(barely no common control code), which is hard to
>> maintain and not able to achieve good scalability for both 'LCDIFv3'
>> and 'LCDIF'.
> 
> I tend to agree with Liu here. Writing a DRM driver isn't that much
> boilerplate anymore with all the helpers we have available in the
> framework today.

I did write a separate driver for this IP before I spent time merging 
them into single driver, that's when I realized a single driver is much 
better and discarded the separate driver idea.

> The IP is so different from the currently supported LCDIF controllers
> that I think trying to support this one in the existing driver actually
> increases the chances to break something when modifying the driver in
> the future. Not everyone is able to test all LCDIF versions. My vote is
> on having a separate driver for the i.MX8MP LCDIF.

If you look at both controllers, it is clear it is still the LCDIF 
behind, even the CSC that is bolted on would suggest that.

I am also not happy when I look at the amount of duplication a separate 
driver would create, it will be some 50% of the code that would be just 
duplicated.

[...]

Lucas Stach March 1, 2022, 11:05 a.m. UTC | #9

Am Dienstag, dem 01.03.2022 um 11:19 +0100 schrieb Marek Vasut:
> On 3/1/22 11:04, Lucas Stach wrote:
> 
> Hi,
> 
> [...]
> 
> > > Given the two totally different IPs, I don't see bugs of IP control
> > > logics should be fixed for both drivers. Naturally, the two would
> > > diverge due to different HWs. Looking at Patch 9/9, it basically
> > > squashes code to control LCDIFv3 into the mxsfb drm driver with
> > > 'if/else' checks(barely no common control code), which is hard to
> > > maintain and not able to achieve good scalability for both 'LCDIFv3'
> > > and 'LCDIF'.
> > 
> > I tend to agree with Liu here. Writing a DRM driver isn't that much
> > boilerplate anymore with all the helpers we have available in the
> > framework today.
> 
> I did write a separate driver for this IP before I spent time merging 
> them into single driver, that's when I realized a single driver is much 
> better and discarded the separate driver idea.
> 
> > The IP is so different from the currently supported LCDIF controllers
> > that I think trying to support this one in the existing driver actually
> > increases the chances to break something when modifying the driver in
> > the future. Not everyone is able to test all LCDIF versions. My vote is
> > on having a separate driver for the i.MX8MP LCDIF.
> 
> If you look at both controllers, it is clear it is still the LCDIF 
> behind, even the CSC that is bolted on would suggest that.

Yes, but from a driver PoV what you care about is not really the
hardware blocks used to implement something, but the programming model,
i.e. the register interface exposed to software.

> 
> I am also not happy when I look at the amount of duplication a separate 
> driver would create, it will be some 50% of the code that would be just 
> duplicated.
> 
Yea, the duplicated code is still significant, as the HW itself is so
simple. However, if you find yourself in the situation where basically
every actual register access in the driver ends up being in a "if (some
HW rev) ... " clause, i still think it would be better to have a
separate driver, as the programming interface is just different.

Regards,
Lucas

Adam Ford March 1, 2022, 1:03 p.m. UTC | #10

On Tue, Mar 1, 2022 at 5:05 AM Lucas Stach <[e-mail]> wrote:
>
> Am Dienstag, dem 01.03.2022 um 11:19 +0100 schrieb Marek Vasut:
> > On 3/1/22 11:04, Lucas Stach wrote:
> >
> > Hi,
> >
> > [...]
> >
> > > > Given the two totally different IPs, I don't see bugs of IP control
> > > > logics should be fixed for both drivers. Naturally, the two would
> > > > diverge due to different HWs. Looking at Patch 9/9, it basically
> > > > squashes code to control LCDIFv3 into the mxsfb drm driver with
> > > > 'if/else' checks(barely no common control code), which is hard to
> > > > maintain and not able to achieve good scalability for both 'LCDIFv3'
> > > > and 'LCDIF'.
> > >
> > > I tend to agree with Liu here. Writing a DRM driver isn't that much
> > > boilerplate anymore with all the helpers we have available in the
> > > framework today.
> >
> > I did write a separate driver for this IP before I spent time merging
> > them into single driver, that's when I realized a single driver is much
> > better and discarded the separate driver idea.
> >
> > > The IP is so different from the currently supported LCDIF controllers
> > > that I think trying to support this one in the existing driver actually
> > > increases the chances to break something when modifying the driver in
> > > the future. Not everyone is able to test all LCDIF versions. My vote is
> > > on having a separate driver for the i.MX8MP LCDIF.
> >
> > If you look at both controllers, it is clear it is still the LCDIF
> > behind, even the CSC that is bolted on would suggest that.
>
> Yes, but from a driver PoV what you care about is not really the
> hardware blocks used to implement something, but the programming model,
> i.e. the register interface exposed to software.
>
> >
> > I am also not happy when I look at the amount of duplication a separate
> > driver would create, it will be some 50% of the code that would be just
> > duplicated.
> >
> Yea, the duplicated code is still significant, as the HW itself is so
> simple. However, if you find yourself in the situation where basically
> every actual register access in the driver ends up being in a "if (some
> HW rev) ... " clause, i still think it would be better to have a
> separate driver, as the programming interface is just different.

I tend to agree with Marek on this one.  We have an instance where the
blk-ctrl and the GPC driver between 8m, mini, nano, plus are close,
but different enough where each SoC has it's own set of tables and
some checks.   Lucas created the framework, and others adapted it for
various SoC's.  If there really is nearly 50% common code for the
LCDIF, why not either leave the driver as one or split the common code
into its own driver like lcdif-common and then have smaller drivers
that handle their specific variations.

adam
>
> Regards,
> Lucas
>
>

Lucas Stach March 1, 2022, 1:18 p.m. UTC | #11

Am Dienstag, dem 01.03.2022 um 07:03 -0600 schrieb Adam Ford:
> On Tue, Mar 1, 2022 at 5:05 AM Lucas Stach <[e-mail]> wrote:
> > 
> > Am Dienstag, dem 01.03.2022 um 11:19 +0100 schrieb Marek Vasut:
> > > On 3/1/22 11:04, Lucas Stach wrote:
> > > 
> > > Hi,
> > > 
> > > [...]
> > > 
> > > > > Given the two totally different IPs, I don't see bugs of IP control
> > > > > logics should be fixed for both drivers. Naturally, the two would
> > > > > diverge due to different HWs. Looking at Patch 9/9, it basically
> > > > > squashes code to control LCDIFv3 into the mxsfb drm driver with
> > > > > 'if/else' checks(barely no common control code), which is hard to
> > > > > maintain and not able to achieve good scalability for both 'LCDIFv3'
> > > > > and 'LCDIF'.
> > > > 
> > > > I tend to agree with Liu here. Writing a DRM driver isn't that much
> > > > boilerplate anymore with all the helpers we have available in the
> > > > framework today.
> > > 
> > > I did write a separate driver for this IP before I spent time merging
> > > them into single driver, that's when I realized a single driver is much
> > > better and discarded the separate driver idea.
> > > 
> > > > The IP is so different from the currently supported LCDIF controllers
> > > > that I think trying to support this one in the existing driver actually
> > > > increases the chances to break something when modifying the driver in
> > > > the future. Not everyone is able to test all LCDIF versions. My vote is
> > > > on having a separate driver for the i.MX8MP LCDIF.
> > > 
> > > If you look at both controllers, it is clear it is still the LCDIF
> > > behind, even the CSC that is bolted on would suggest that.
> > 
> > Yes, but from a driver PoV what you care about is not really the
> > hardware blocks used to implement something, but the programming model,
> > i.e. the register interface exposed to software.
> > 
> > > 
> > > I am also not happy when I look at the amount of duplication a separate
> > > driver would create, it will be some 50% of the code that would be just
> > > duplicated.
> > > 
> > Yea, the duplicated code is still significant, as the HW itself is so
> > simple. However, if you find yourself in the situation where basically
> > every actual register access in the driver ends up being in a "if (some
> > HW rev) ... " clause, i still think it would be better to have a
> > separate driver, as the programming interface is just different.
> 
> I tend to agree with Marek on this one.  We have an instance where the
> blk-ctrl and the GPC driver between 8m, mini, nano, plus are close,
> but different enough where each SoC has it's own set of tables and
> some checks.   Lucas created the framework, and others adapted it for
> various SoC's.  If there really is nearly 50% common code for the
> LCDIF, why not either leave the driver as one or split the common code
> into its own driver like lcdif-common and then have smaller drivers
> that handle their specific variations.

I don't know exactly how the standalone driver looks like, but I guess
the overlap is not really in any real HW specific parts, but the common
DRM boilerplate, so there isn't much point in creating a common lcdif
driver.

As you brought up the blk-ctrl as an example: I'm all for supporting
slightly different hardware in the same driver, as long as the HW
interface is close enough. But then I also opted for a separate 8MP
blk-ctrl driver for those blk-ctrls that differ significantly from the
others, as I think it would make the common driver unmaintainable
trying to support all the different variants in one driver.

Regards,
Lucas

Robby Cai March 1, 2022, 1:37 p.m. UTC | #12

>-----Original Message-----
>From: Lucas Stach <[e-mail]>
>Sent: 2022年3月1日 21:19
>To: Adam Ford <[e-mail]>
>Cc: Marek Vasut <[e-mail]>; Ying Liu (OSS) <[e-mail]>;
>dri-devel <[e-mail]>; devicetree
><[e-mail]>; Peng Fan <[e-mail]>; Alexander Stein
><[e-mail]>; Rob Herring <[e-mail]>;
>Laurent Pinchart <[e-mail]>; Sam Ravnborg
><[e-mail]>; Robby Cai <[e-mail]>
>Subject: [EXT] Re: [PATCH 1/9] dt-bindings: mxsfb: Add compatible for
>i.MX8MP
>
>Caution: EXT Email
>
>Am Dienstag, dem 01.03.2022 um 07:03 -0600 schrieb Adam Ford:
>> On Tue, Mar 1, 2022 at 5:05 AM Lucas Stach <[e-mail]>
>wrote:
>> >
>> > Am Dienstag, dem 01.03.2022 um 11:19 +0100 schrieb Marek Vasut:
>> > > On 3/1/22 11:04, Lucas Stach wrote:
>> > >
>> > > Hi,
>> > >
>> > > [...]
>> > >
>> > > > > Given the two totally different IPs, I don't see bugs of IP
>> > > > > control logics should be fixed for both drivers. Naturally,
>> > > > > the two would diverge due to different HWs. Looking at Patch
>> > > > > 9/9, it basically squashes code to control LCDIFv3 into the
>> > > > > mxsfb drm driver with 'if/else' checks(barely no common
>> > > > > control code), which is hard to maintain and not able to achieve good
>scalability for both 'LCDIFv3'
>> > > > > and 'LCDIF'.
>> > > >
>> > > > I tend to agree with Liu here. Writing a DRM driver isn't that
>> > > > much boilerplate anymore with all the helpers we have available
>> > > > in the framework today.
>> > >
>> > > I did write a separate driver for this IP before I spent time
>> > > merging them into single driver, that's when I realized a single
>> > > driver is much better and discarded the separate driver idea.
>> > >
>> > > > The IP is so different from the currently supported LCDIF
>> > > > controllers that I think trying to support this one in the
>> > > > existing driver actually increases the chances to break
>> > > > something when modifying the driver in the future. Not everyone
>> > > > is able to test all LCDIF versions. My vote is on having a separate driver
>for the i.MX8MP LCDIF.
>> > >
>> > > If you look at both controllers, it is clear it is still the LCDIF
>> > > behind, even the CSC that is bolted on would suggest that.
>> >
>> > Yes, but from a driver PoV what you care about is not really the
>> > hardware blocks used to implement something, but the programming
>> > model, i.e. the register interface exposed to software.
>> >
>> > >
>> > > I am also not happy when I look at the amount of duplication a
>> > > separate driver would create, it will be some 50% of the code that
>> > > would be just duplicated.
>> > >
>> > Yea, the duplicated code is still significant, as the HW itself is
>> > so simple. However, if you find yourself in the situation where
>> > basically every actual register access in the driver ends up being
>> > in a "if (some HW rev) ... " clause, i still think it would be
>> > better to have a separate driver, as the programming interface is just
>different.
>>
>> I tend to agree with Marek on this one.  We have an instance where the
>> blk-ctrl and the GPC driver between 8m, mini, nano, plus are close,
>> but different enough where each SoC has it's own set of tables and
>> some checks.   Lucas created the framework, and others adapted it for
>> various SoC's.  If there really is nearly 50% common code for the
>> LCDIF, why not either leave the driver as one or split the common code
>> into its own driver like lcdif-common and then have smaller drivers
>> that handle their specific variations.
>
>I don't know exactly how the standalone driver looks like, but I guess the
>overlap is not really in any real HW specific parts, but the common DRM
>boilerplate, so there isn't much point in creating a common lcdif driver.
>
>As you brought up the blk-ctrl as an example: I'm all for supporting slightly
>different hardware in the same driver, as long as the HW interface is close
>enough. But then I also opted for a separate 8MP blk-ctrl driver for those
>blk-ctrls that differ significantly from the others, as I think it would make the
>common driver unmaintainable trying to support all the different variants in
>one driver.
>
>Regards,
>Lucas

LCDIF on i.MX8MP is a different IP which is borrowed from non-iMX series, although it's also called 'LCDIF'.
We prefer not mix these two series of IPs in one driver for ease of maintenance and extension.

Regards,
Robby

Marek Vasut March 2, 2022, 2:49 a.m. UTC | #13

On 3/1/22 14:37, Robby Cai wrote:

Hi,

[...]

>>> I tend to agree with Marek on this one.  We have an instance where the
>>> blk-ctrl and the GPC driver between 8m, mini, nano, plus are close,
>>> but different enough where each SoC has it's own set of tables and
>>> some checks.   Lucas created the framework, and others adapted it for
>>> various SoC's.  If there really is nearly 50% common code for the
>>> LCDIF, why not either leave the driver as one or split the common code
>>> into its own driver like lcdif-common and then have smaller drivers
>>> that handle their specific variations.
>>
>> I don't know exactly how the standalone driver looks like, but I guess the
>> overlap is not really in any real HW specific parts, but the common DRM
>> boilerplate, so there isn't much point in creating a common lcdif driver.
>>
>> As you brought up the blk-ctrl as an example: I'm all for supporting slightly
>> different hardware in the same driver, as long as the HW interface is close
>> enough. But then I also opted for a separate 8MP blk-ctrl driver for those
>> blk-ctrls that differ significantly from the others, as I think it would make the
>> common driver unmaintainable trying to support all the different variants in
>> one driver.
>>
>> Regards,
>> Lucas
> 
> LCDIF on i.MX8MP is a different IP which is borrowed from non-iMX series, although it's also called 'LCDIF'.
> We prefer not mix these two series of IPs in one driver for ease of maintenance and extension.

Where does the MX8MP LCDIF come from then, SGTL maybe ?

Marek Vasut March 2, 2022, 2:54 a.m. UTC | #14

On 3/1/22 14:18, Lucas Stach wrote:
> Am Dienstag, dem 01.03.2022 um 07:03 -0600 schrieb Adam Ford:
>> On Tue, Mar 1, 2022 at 5:05 AM Lucas Stach <[e-mail]> wrote:
>>>
>>> Am Dienstag, dem 01.03.2022 um 11:19 +0100 schrieb Marek Vasut:
>>>> On 3/1/22 11:04, Lucas Stach wrote:
>>>>
>>>> Hi,
>>>>
>>>> [...]
>>>>
>>>>>> Given the two totally different IPs, I don't see bugs of IP control
>>>>>> logics should be fixed for both drivers. Naturally, the two would
>>>>>> diverge due to different HWs. Looking at Patch 9/9, it basically
>>>>>> squashes code to control LCDIFv3 into the mxsfb drm driver with
>>>>>> 'if/else' checks(barely no common control code), which is hard to
>>>>>> maintain and not able to achieve good scalability for both 'LCDIFv3'
>>>>>> and 'LCDIF'.
>>>>>
>>>>> I tend to agree with Liu here. Writing a DRM driver isn't that much
>>>>> boilerplate anymore with all the helpers we have available in the
>>>>> framework today.
>>>>
>>>> I did write a separate driver for this IP before I spent time merging
>>>> them into single driver, that's when I realized a single driver is much
>>>> better and discarded the separate driver idea.
>>>>
>>>>> The IP is so different from the currently supported LCDIF controllers
>>>>> that I think trying to support this one in the existing driver actually
>>>>> increases the chances to break something when modifying the driver in
>>>>> the future. Not everyone is able to test all LCDIF versions. My vote is
>>>>> on having a separate driver for the i.MX8MP LCDIF.
>>>>
>>>> If you look at both controllers, it is clear it is still the LCDIF
>>>> behind, even the CSC that is bolted on would suggest that.
>>>
>>> Yes, but from a driver PoV what you care about is not really the
>>> hardware blocks used to implement something, but the programming model,
>>> i.e. the register interface exposed to software.
>>>
>>>>
>>>> I am also not happy when I look at the amount of duplication a separate
>>>> driver would create, it will be some 50% of the code that would be just
>>>> duplicated.
>>>>
>>> Yea, the duplicated code is still significant, as the HW itself is so
>>> simple. However, if you find yourself in the situation where basically
>>> every actual register access in the driver ends up being in a "if (some
>>> HW rev) ... " clause, i still think it would be better to have a
>>> separate driver, as the programming interface is just different.
>>
>> I tend to agree with Marek on this one.  We have an instance where the
>> blk-ctrl and the GPC driver between 8m, mini, nano, plus are close,
>> but different enough where each SoC has it's own set of tables and
>> some checks.   Lucas created the framework, and others adapted it for
>> various SoC's.  If there really is nearly 50% common code for the
>> LCDIF, why not either leave the driver as one or split the common code
>> into its own driver like lcdif-common and then have smaller drivers
>> that handle their specific variations.
> 
> I don't know exactly how the standalone driver looks like, but I guess
> the overlap is not really in any real HW specific parts, but the common
> DRM boilerplate, so there isn't much point in creating a common lcdif
> driver.

The mxsfb currently has 1280 LoC as of patch 8/9 of this series. Of 
that, there is some 400 LoC which are specific to old LCDIF and this 
patch adds 380 LoC for the new LCDIF. So that's 800 LoC or ~60% of 
shared boilerplate that would be duplicated .

> As you brought up the blk-ctrl as an example: I'm all for supporting
> slightly different hardware in the same driver, as long as the HW
> interface is close enough. But then I also opted for a separate 8MP
> blk-ctrl driver for those blk-ctrls that differ significantly from the
> others, as I think it would make the common driver unmaintainable
> trying to support all the different variants in one driver.

But then you also need to maintain two sets of boilerplate, they 
diverge, and that is not good.

Lucas Stach March 2, 2022, 9:23 a.m. UTC | #15

Am Mittwoch, dem 02.03.2022 um 03:54 +0100 schrieb Marek Vasut:
> On 3/1/22 14:18, Lucas Stach wrote:
> > Am Dienstag, dem 01.03.2022 um 07:03 -0600 schrieb Adam Ford:
> > > On Tue, Mar 1, 2022 at 5:05 AM Lucas Stach <[e-mail]> wrote:
> > > > 
> > > > Am Dienstag, dem 01.03.2022 um 11:19 +0100 schrieb Marek Vasut:
> > > > > On 3/1/22 11:04, Lucas Stach wrote:
> > > > > 
> > > > > Hi,
> > > > > 
> > > > > [...]
> > > > > 
> > > > > > > Given the two totally different IPs, I don't see bugs of IP control
> > > > > > > logics should be fixed for both drivers. Naturally, the two would
> > > > > > > diverge due to different HWs. Looking at Patch 9/9, it basically
> > > > > > > squashes code to control LCDIFv3 into the mxsfb drm driver with
> > > > > > > 'if/else' checks(barely no common control code), which is hard to
> > > > > > > maintain and not able to achieve good scalability for both 'LCDIFv3'
> > > > > > > and 'LCDIF'.
> > > > > > 
> > > > > > I tend to agree with Liu here. Writing a DRM driver isn't that much
> > > > > > boilerplate anymore with all the helpers we have available in the
> > > > > > framework today.
> > > > > 
> > > > > I did write a separate driver for this IP before I spent time merging
> > > > > them into single driver, that's when I realized a single driver is much
> > > > > better and discarded the separate driver idea.
> > > > > 
> > > > > > The IP is so different from the currently supported LCDIF controllers
> > > > > > that I think trying to support this one in the existing driver actually
> > > > > > increases the chances to break something when modifying the driver in
> > > > > > the future. Not everyone is able to test all LCDIF versions. My vote is
> > > > > > on having a separate driver for the i.MX8MP LCDIF.
> > > > > 
> > > > > If you look at both controllers, it is clear it is still the LCDIF
> > > > > behind, even the CSC that is bolted on would suggest that.
> > > > 
> > > > Yes, but from a driver PoV what you care about is not really the
> > > > hardware blocks used to implement something, but the programming model,
> > > > i.e. the register interface exposed to software.
> > > > 
> > > > > 
> > > > > I am also not happy when I look at the amount of duplication a separate
> > > > > driver would create, it will be some 50% of the code that would be just
> > > > > duplicated.
> > > > > 
> > > > Yea, the duplicated code is still significant, as the HW itself is so
> > > > simple. However, if you find yourself in the situation where basically
> > > > every actual register access in the driver ends up being in a "if (some
> > > > HW rev) ... " clause, i still think it would be better to have a
> > > > separate driver, as the programming interface is just different.
> > > 
> > > I tend to agree with Marek on this one.  We have an instance where the
> > > blk-ctrl and the GPC driver between 8m, mini, nano, plus are close,
> > > but different enough where each SoC has it's own set of tables and
> > > some checks.   Lucas created the framework, and others adapted it for
> > > various SoC's.  If there really is nearly 50% common code for the
> > > LCDIF, why not either leave the driver as one or split the common code
> > > into its own driver like lcdif-common and then have smaller drivers
> > > that handle their specific variations.
> > 
> > I don't know exactly how the standalone driver looks like, but I guess
> > the overlap is not really in any real HW specific parts, but the common
> > DRM boilerplate, so there isn't much point in creating a common lcdif
> > driver.
> 
> The mxsfb currently has 1280 LoC as of patch 8/9 of this series. Of 
> that, there is some 400 LoC which are specific to old LCDIF and this 
> patch adds 380 LoC for the new LCDIF. So that's 800 LoC or ~60% of 
> shared boilerplate that would be duplicated .

That is probably ignoring the fact that the 8MP LCDIF does not support
any overlays, so it could use the drm_simple_display_pipe
infrastructure to reduce the needed boilerplate.
> 
> > As you brought up the blk-ctrl as an example: I'm all for supporting
> > slightly different hardware in the same driver, as long as the HW
> > interface is close enough. But then I also opted for a separate 8MP
> > blk-ctrl driver for those blk-ctrls that differ significantly from the
> > others, as I think it would make the common driver unmaintainable
> > trying to support all the different variants in one driver.
> 
> But then you also need to maintain two sets of boilerplate, they 
> diverge, and that is not good.

I don't think that there is much chance for bugs going unfixed due to
divergence in the boilerplate, especially if you use the simple pipe
framework to handle most of that stuff for you, which gives you a lot
of code sharing with other simple DRM drivers.

Regards,
Lucas

Liu Ying March 2, 2022, 9:41 a.m. UTC | #16

On Wed, 2022-03-02 at 10:23 +0100, Lucas Stach wrote:
> Am Mittwoch, dem 02.03.2022 um 03:54 +0100 schrieb Marek Vasut:
> > On 3/1/22 14:18, Lucas Stach wrote:
> > > Am Dienstag, dem 01.03.2022 um 07:03 -0600 schrieb Adam Ford:
> > > > On Tue, Mar 1, 2022 at 5:05 AM Lucas Stach <[e-mail]> wrote:
> > > > > Am Dienstag, dem 01.03.2022 um 11:19 +0100 schrieb Marek Vasut:
> > > > > > On 3/1/22 11:04, Lucas Stach wrote:
> > > > > > 
> > > > > > Hi,
> > > > > > 
> > > > > > [...]
> > > > > > 
> > > > > > > > Given the two totally different IPs, I don't see bugs of IP control
> > > > > > > > logics should be fixed for both drivers. Naturally, the two would
> > > > > > > > diverge due to different HWs. Looking at Patch 9/9, it basically
> > > > > > > > squashes code to control LCDIFv3 into the mxsfb drm driver with
> > > > > > > > 'if/else' checks(barely no common control code), which is hard to
> > > > > > > > maintain and not able to achieve good scalability for both 'LCDIFv3'
> > > > > > > > and 'LCDIF'.
> > > > > > > 
> > > > > > > I tend to agree with Liu here. Writing a DRM driver isn't that much
> > > > > > > boilerplate anymore with all the helpers we have available in the
> > > > > > > framework today.
> > > > > > 
> > > > > > I did write a separate driver for this IP before I spent time merging
> > > > > > them into single driver, that's when I realized a single driver is much
> > > > > > better and discarded the separate driver idea.
> > > > > > 
> > > > > > > The IP is so different from the currently supported LCDIF controllers
> > > > > > > that I think trying to support this one in the existing driver actually
> > > > > > > increases the chances to break something when modifying the driver in
> > > > > > > the future. Not everyone is able to test all LCDIF versions. My vote is
> > > > > > > on having a separate driver for the i.MX8MP LCDIF.
> > > > > > 
> > > > > > If you look at both controllers, it is clear it is still the LCDIF
> > > > > > behind, even the CSC that is bolted on would suggest that.
> > > > > 
> > > > > Yes, but from a driver PoV what you care about is not really the
> > > > > hardware blocks used to implement something, but the programming model,
> > > > > i.e. the register interface exposed to software.
> > > > > 
> > > > > > I am also not happy when I look at the amount of duplication a separate
> > > > > > driver would create, it will be some 50% of the code that would be just
> > > > > > duplicated.
> > > > > > 
> > > > > Yea, the duplicated code is still significant, as the HW itself is so
> > > > > simple. However, if you find yourself in the situation where basically
> > > > > every actual register access in the driver ends up being in a "if (some
> > > > > HW rev) ... " clause, i still think it would be better to have a
> > > > > separate driver, as the programming interface is just different.
> > > > 
> > > > I tend to agree with Marek on this one.  We have an instance where the
> > > > blk-ctrl and the GPC driver between 8m, mini, nano, plus are close,
> > > > but different enough where each SoC has it's own set of tables and
> > > > some checks.   Lucas created the framework, and others adapted it for
> > > > various SoC's.  If there really is nearly 50% common code for the
> > > > LCDIF, why not either leave the driver as one or split the common code
> > > > into its own driver like lcdif-common and then have smaller drivers
> > > > that handle their specific variations.
> > > 
> > > I don't know exactly how the standalone driver looks like, but I guess
> > > the overlap is not really in any real HW specific parts, but the common
> > > DRM boilerplate, so there isn't much point in creating a common lcdif
> > > driver.
> > 
> > The mxsfb currently has 1280 LoC as of patch 8/9 of this series. Of 
> > that, there is some 400 LoC which are specific to old LCDIF and this 
> > patch adds 380 LoC for the new LCDIF. So that's 800 LoC or ~60% of 
> > shared boilerplate that would be duplicated .
> 
> That is probably ignoring the fact that the 8MP LCDIF does not support
> any overlays, so it could use the drm_simple_display_pipe
> infrastructure to reduce the needed boilerplate.

The drm_simple_display_pipe infrastructure is probably too simple for
i.MX8MP LCDIF, since it uses one only crtc for one drm device. i.MX8MP
embeds *three* LCDIF instances to support MIPI DSI, LVDS and HDMI
outputs respectively. To use that infrastructure means there would be
three dri cards in all. However, the three LCDIF instances can be
wrapped by the one drm device, which is not the boilerplate code in the
current mxsfb driver may handle.

Regards,
Liu Ying

> > > As you brought up the blk-ctrl as an example: I'm all for supporting
> > > slightly different hardware in the same driver, as long as the HW
> > > interface is close enough. But then I also opted for a separate 8MP
> > > blk-ctrl driver for those blk-ctrls that differ significantly from the
> > > others, as I think it would make the common driver unmaintainable
> > > trying to support all the different variants in one driver.
> > 
> > But then you also need to maintain two sets of boilerplate, they 
> > diverge, and that is not good.
> 
> I don't think that there is much chance for bugs going unfixed due to
> divergence in the boilerplate, especially if you use the simple pipe
> framework to handle most of that stuff for you, which gives you a lot
> of code sharing with other simple DRM drivers.
> 
> Regards,
> Lucas
> 
>

Lucas Stach March 2, 2022, 11:57 a.m. UTC | #17

Am Mittwoch, dem 02.03.2022 um 17:41 +0800 schrieb Liu Ying:
> On Wed, 2022-03-02 at 10:23 +0100, Lucas Stach wrote:
> > Am Mittwoch, dem 02.03.2022 um 03:54 +0100 schrieb Marek Vasut:
> > > On 3/1/22 14:18, Lucas Stach wrote:
> > > > Am Dienstag, dem 01.03.2022 um 07:03 -0600 schrieb Adam Ford:
> > > > > On Tue, Mar 1, 2022 at 5:05 AM Lucas Stach <[e-mail]> wrote:
> > > > > > Am Dienstag, dem 01.03.2022 um 11:19 +0100 schrieb Marek Vasut:
> > > > > > > On 3/1/22 11:04, Lucas Stach wrote:
> > > > > > > 
> > > > > > > Hi,
> > > > > > > 
> > > > > > > [...]
> > > > > > > 
> > > > > > > > > Given the two totally different IPs, I don't see bugs of IP control
> > > > > > > > > logics should be fixed for both drivers. Naturally, the two would
> > > > > > > > > diverge due to different HWs. Looking at Patch 9/9, it basically
> > > > > > > > > squashes code to control LCDIFv3 into the mxsfb drm driver with
> > > > > > > > > 'if/else' checks(barely no common control code), which is hard to
> > > > > > > > > maintain and not able to achieve good scalability for both 'LCDIFv3'
> > > > > > > > > and 'LCDIF'.
> > > > > > > > 
> > > > > > > > I tend to agree with Liu here. Writing a DRM driver isn't that much
> > > > > > > > boilerplate anymore with all the helpers we have available in the
> > > > > > > > framework today.
> > > > > > > 
> > > > > > > I did write a separate driver for this IP before I spent time merging
> > > > > > > them into single driver, that's when I realized a single driver is much
> > > > > > > better and discarded the separate driver idea.
> > > > > > > 
> > > > > > > > The IP is so different from the currently supported LCDIF controllers
> > > > > > > > that I think trying to support this one in the existing driver actually
> > > > > > > > increases the chances to break something when modifying the driver in
> > > > > > > > the future. Not everyone is able to test all LCDIF versions. My vote is
> > > > > > > > on having a separate driver for the i.MX8MP LCDIF.
> > > > > > > 
> > > > > > > If you look at both controllers, it is clear it is still the LCDIF
> > > > > > > behind, even the CSC that is bolted on would suggest that.
> > > > > > 
> > > > > > Yes, but from a driver PoV what you care about is not really the
> > > > > > hardware blocks used to implement something, but the programming model,
> > > > > > i.e. the register interface exposed to software.
> > > > > > 
> > > > > > > I am also not happy when I look at the amount of duplication a separate
> > > > > > > driver would create, it will be some 50% of the code that would be just
> > > > > > > duplicated.
> > > > > > > 
> > > > > > Yea, the duplicated code is still significant, as the HW itself is so
> > > > > > simple. However, if you find yourself in the situation where basically
> > > > > > every actual register access in the driver ends up being in a "if (some
> > > > > > HW rev) ... " clause, i still think it would be better to have a
> > > > > > separate driver, as the programming interface is just different.
> > > > > 
> > > > > I tend to agree with Marek on this one.  We have an instance where the
> > > > > blk-ctrl and the GPC driver between 8m, mini, nano, plus are close,
> > > > > but different enough where each SoC has it's own set of tables and
> > > > > some checks.   Lucas created the framework, and others adapted it for
> > > > > various SoC's.  If there really is nearly 50% common code for the
> > > > > LCDIF, why not either leave the driver as one or split the common code
> > > > > into its own driver like lcdif-common and then have smaller drivers
> > > > > that handle their specific variations.
> > > > 
> > > > I don't know exactly how the standalone driver looks like, but I guess
> > > > the overlap is not really in any real HW specific parts, but the common
> > > > DRM boilerplate, so there isn't much point in creating a common lcdif
> > > > driver.
> > > 
> > > The mxsfb currently has 1280 LoC as of patch 8/9 of this series. Of 
> > > that, there is some 400 LoC which are specific to old LCDIF and this 
> > > patch adds 380 LoC for the new LCDIF. So that's 800 LoC or ~60% of 
> > > shared boilerplate that would be duplicated .
> > 
> > That is probably ignoring the fact that the 8MP LCDIF does not support
> > any overlays, so it could use the drm_simple_display_pipe
> > infrastructure to reduce the needed boilerplate.
> 
> The drm_simple_display_pipe infrastructure is probably too simple for
> i.MX8MP LCDIF, since it uses one only crtc for one drm device. i.MX8MP
> embeds *three* LCDIF instances to support MIPI DSI, LVDS and HDMI
> outputs respectively. To use that infrastructure means there would be
> three dri cards in all. However, the three LCDIF instances can be
> wrapped by the one drm device, which is not the boilerplate code in the
> current mxsfb driver may handle.

While that may make things a little simpler for userspace, I'm not sure
if this is the right thing to do. It complicates the driver a lot,
especially if you want to get things like independent power management,
etc. right. It also creates a fake view for userspace, where is looks
like there might be some shared resources between the different display
paths, while in reality they are fully independent.

While we do something similar on the GPU side and collect all GPU cores
under a single DRM device, I'm not fully convinced that this was a good
decision. It now comes back to bite us when the SoC topologies get a
little more interesting and e.g. devices are behind different IOMMU
streams.

Regards,
Lucas

Robby Cai March 2, 2022, 1:14 p.m. UTC | #18

>-----Original Message-----
>From: Marek Vasut <[e-mail]>
>Sent: 2022年3月2日 10:50
>To: Robby Cai <[e-mail]>; Lucas Stach <[e-mail]>;
>Adam Ford <[e-mail]>
>Cc: Ying Liu (OSS) <[e-mail]>; dri-devel
><[e-mail]>; devicetree <[e-mail]>;
>Peng Fan <[e-mail]>; Alexander Stein
><[e-mail]>; Rob Herring <[e-mail]>;
>Laurent Pinchart <[e-mail]>; Sam Ravnborg
><[e-mail]>
>Subject: Re: [EXT] Re: [PATCH 1/9] dt-bindings: mxsfb: Add compatible for
>i.MX8MP
>
>Caution: EXT Email
>
>On 3/1/22 14:37, Robby Cai wrote:
>
>Hi,
>
>[...]
>
>>>> I tend to agree with Marek on this one.  We have an instance where
>>>> the blk-ctrl and the GPC driver between 8m, mini, nano, plus are
>>>> close, but different enough where each SoC has it's own set of tables and
>>>> some checks.   Lucas created the framework, and others adapted it for
>>>> various SoC's.  If there really is nearly 50% common code for the
>>>> LCDIF, why not either leave the driver as one or split the common
>>>> code into its own driver like lcdif-common and then have smaller
>>>> drivers that handle their specific variations.
>>>
>>> I don't know exactly how the standalone driver looks like, but I
>>> guess the overlap is not really in any real HW specific parts, but
>>> the common DRM boilerplate, so there isn't much point in creating a
>common lcdif driver.
>>>
>>> As you brought up the blk-ctrl as an example: I'm all for supporting
>>> slightly different hardware in the same driver, as long as the HW
>>> interface is close enough. But then I also opted for a separate 8MP
>>> blk-ctrl driver for those blk-ctrls that differ significantly from
>>> the others, as I think it would make the common driver unmaintainable
>>> trying to support all the different variants in one driver.
>>>
>>> Regards,
>>> Lucas
>>
>> LCDIF on i.MX8MP is a different IP which is borrowed from non-iMX series,
>although it's also called 'LCDIF'.
>> We prefer not mix these two series of IPs in one driver for ease of
>maintenance and extension.
>
>Where does the MX8MP LCDIF come from then, SGTL maybe ?

AFAIK, it's RT1170. You may have a check on RM [1]. Interestingly, this SoC has both eLCDIF and LCDIFv2, two IPs we are talking about.

[1] https://www.nxp.com/webapp/Download?colCode=IMXRT1170RM

Regards,
Robby

Liu Ying March 3, 2022, 2:54 a.m. UTC | #19

On Wed, 2022-03-02 at 12:57 +0100, Lucas Stach wrote:
> Am Mittwoch, dem 02.03.2022 um 17:41 +0800 schrieb Liu Ying:
> > On Wed, 2022-03-02 at 10:23 +0100, Lucas Stach wrote:
> > > Am Mittwoch, dem 02.03.2022 um 03:54 +0100 schrieb Marek Vasut:
> > > > On 3/1/22 14:18, Lucas Stach wrote:
> > > > > Am Dienstag, dem 01.03.2022 um 07:03 -0600 schrieb Adam Ford:
> > > > > > On Tue, Mar 1, 2022 at 5:05 AM Lucas Stach <[e-mail]> wrote:
> > > > > > > Am Dienstag, dem 01.03.2022 um 11:19 +0100 schrieb Marek Vasut:
> > > > > > > > On 3/1/22 11:04, Lucas Stach wrote:
> > > > > > > > 
> > > > > > > > Hi,
> > > > > > > > 
> > > > > > > > [...]
> > > > > > > > 
> > > > > > > > > > Given the two totally different IPs, I don't see bugs of IP control
> > > > > > > > > > logics should be fixed for both drivers. Naturally, the two would
> > > > > > > > > > diverge due to different HWs. Looking at Patch 9/9, it basically
> > > > > > > > > > squashes code to control LCDIFv3 into the mxsfb drm driver with
> > > > > > > > > > 'if/else' checks(barely no common control code), which is hard to
> > > > > > > > > > maintain and not able to achieve good scalability for both 'LCDIFv3'
> > > > > > > > > > and 'LCDIF'.
> > > > > > > > > 
> > > > > > > > > I tend to agree with Liu here. Writing a DRM driver isn't that much
> > > > > > > > > boilerplate anymore with all the helpers we have available in the
> > > > > > > > > framework today.
> > > > > > > > 
> > > > > > > > I did write a separate driver for this IP before I spent time merging
> > > > > > > > them into single driver, that's when I realized a single driver is much
> > > > > > > > better and discarded the separate driver idea.
> > > > > > > > 
> > > > > > > > > The IP is so different from the currently supported LCDIF controllers
> > > > > > > > > that I think trying to support this one in the existing driver actually
> > > > > > > > > increases the chances to break something when modifying the driver in
> > > > > > > > > the future. Not everyone is able to test all LCDIF versions. My vote is
> > > > > > > > > on having a separate driver for the i.MX8MP LCDIF.
> > > > > > > > 
> > > > > > > > If you look at both controllers, it is clear it is still the LCDIF
> > > > > > > > behind, even the CSC that is bolted on would suggest that.
> > > > > > > 
> > > > > > > Yes, but from a driver PoV what you care about is not really the
> > > > > > > hardware blocks used to implement something, but the programming model,
> > > > > > > i.e. the register interface exposed to software.
> > > > > > > 
> > > > > > > > I am also not happy when I look at the amount of duplication a separate
> > > > > > > > driver would create, it will be some 50% of the code that would be just
> > > > > > > > duplicated.
> > > > > > > > 
> > > > > > > Yea, the duplicated code is still significant, as the HW itself is so
> > > > > > > simple. However, if you find yourself in the situation where basically
> > > > > > > every actual register access in the driver ends up being in a "if (some
> > > > > > > HW rev) ... " clause, i still think it would be better to have a
> > > > > > > separate driver, as the programming interface is just different.
> > > > > > 
> > > > > > I tend to agree with Marek on this one.  We have an instance where the
> > > > > > blk-ctrl and the GPC driver between 8m, mini, nano, plus are close,
> > > > > > but different enough where each SoC has it's own set of tables and
> > > > > > some checks.   Lucas created the framework, and others adapted it for
> > > > > > various SoC's.  If there really is nearly 50% common code for the
> > > > > > LCDIF, why not either leave the driver as one or split the common code
> > > > > > into its own driver like lcdif-common and then have smaller drivers
> > > > > > that handle their specific variations.
> > > > > 
> > > > > I don't know exactly how the standalone driver looks like, but I guess
> > > > > the overlap is not really in any real HW specific parts, but the common
> > > > > DRM boilerplate, so there isn't much point in creating a common lcdif
> > > > > driver.
> > > > 
> > > > The mxsfb currently has 1280 LoC as of patch 8/9 of this series. Of 
> > > > that, there is some 400 LoC which are specific to old LCDIF and this 
> > > > patch adds 380 LoC for the new LCDIF. So that's 800 LoC or ~60% of 
> > > > shared boilerplate that would be duplicated .
> > > 
> > > That is probably ignoring the fact that the 8MP LCDIF does not support
> > > any overlays, so it could use the drm_simple_display_pipe
> > > infrastructure to reduce the needed boilerplate.
> > 
> > The drm_simple_display_pipe infrastructure is probably too simple for
> > i.MX8MP LCDIF, since it uses one only crtc for one drm device. i.MX8MP
> > embeds *three* LCDIF instances to support MIPI DSI, LVDS and HDMI
> > outputs respectively. To use that infrastructure means there would be
> > three dri cards in all. However, the three LCDIF instances can be
> > wrapped by the one drm device, which is not the boilerplate code in the
> > current mxsfb driver may handle.
> 
> While that may make things a little simpler for userspace, I'm not sure
> if this is the right thing to do. It complicates the driver a lot,
> especially if you want to get things like independent power management,
> etc. right. It also creates a fake view for userspace, where is looks
> like there might be some shared resources between the different display
> paths, while in reality they are fully independent.

Trade-off will be made between one drm device and three. My first
impression of using the drm_simple_display_pipe infrastructure is that
it's too simple and less flexible/scalable, because SoC designer will
be likely to add muxes between CRTCs and encoders/bridges or overlay
plane(s) in next generations of SoCs(SW developers don't seem have good
reasons to suggest not to do that).  Another concern is that whether
the userspace may use the three drm devices well or not. 

A few more points:
1) With one drm device, userspace may use drm lease APIs to control
those independant pipes with drm masters(not sure about the userspace
maturity).
2) Code to gather all LCDIFs as one drm device has chance to be created
as helpers once there are similar use cases in other drivers(maybe,
there is/are already).
3) Power management doesn't seem to be a problem, since each LCDIF has
it's own struct device which can be used to do runtime PM at some
drm_crtc_helper_funcs callbacks.
4) Regarding the fake view of shared resources, atomic check can handle
that, so it doesn't seem to be a big problem, either.

> 
> While we do something similar on the GPU side and collect all GPU cores
> under a single DRM device, I'm not fully convinced that this was a good
> decision. It now comes back to bite us when the SoC topologies get a
> little more interesting and e.g. devices are behind different IOMMU
> streams.

Right, SoC topologies may change, like the aforementioned muxes.
Generally speaking, I think one drm device is more flexible and
scalable than three.

Regards,
Liu Ying

Marek Vasut March 3, 2022, 3:14 a.m. UTC | #20

On 3/2/22 10:23, Lucas Stach wrote:

[...]

>>>> I tend to agree with Marek on this one.  We have an instance where the
>>>> blk-ctrl and the GPC driver between 8m, mini, nano, plus are close,
>>>> but different enough where each SoC has it's own set of tables and
>>>> some checks.   Lucas created the framework, and others adapted it for
>>>> various SoC's.  If there really is nearly 50% common code for the
>>>> LCDIF, why not either leave the driver as one or split the common code
>>>> into its own driver like lcdif-common and then have smaller drivers
>>>> that handle their specific variations.
>>>
>>> I don't know exactly how the standalone driver looks like, but I guess
>>> the overlap is not really in any real HW specific parts, but the common
>>> DRM boilerplate, so there isn't much point in creating a common lcdif
>>> driver.
>>
>> The mxsfb currently has 1280 LoC as of patch 8/9 of this series. Of
>> that, there is some 400 LoC which are specific to old LCDIF and this
>> patch adds 380 LoC for the new LCDIF. So that's 800 LoC or ~60% of
>> shared boilerplate that would be duplicated .
> 
> That is probably ignoring the fact that the 8MP LCDIF does not support
> any overlays, so it could use the drm_simple_display_pipe
> infrastructure to reduce the needed boilerplate.

It seems the IMXRT1070 LCDIF v2 (heh ...) does support overlays, so no, 
the mxsfb and hypothetical lcdif drivers would look really very similar.

>>> As you brought up the blk-ctrl as an example: I'm all for supporting
>>> slightly different hardware in the same driver, as long as the HW
>>> interface is close enough. But then I also opted for a separate 8MP
>>> blk-ctrl driver for those blk-ctrls that differ significantly from the
>>> others, as I think it would make the common driver unmaintainable
>>> trying to support all the different variants in one driver.
>>
>> But then you also need to maintain two sets of boilerplate, they
>> diverge, and that is not good.
> 
> I don't think that there is much chance for bugs going unfixed due to
> divergence in the boilerplate, especially if you use the simple pipe
> framework to handle most of that stuff for you, which gives you a lot
> of code sharing with other simple DRM drivers.

But I can not use the simple pipe because overlays, see imxrt1070 .

[...]

We can always split the drivers later if this becomes unmaintainable 
too, no ?

Marek Vasut March 3, 2022, 3:16 a.m. UTC | #21

On 3/2/22 14:14, Robby Cai wrote:

Hi

[...]

>>> LCDIF on i.MX8MP is a different IP which is borrowed from non-iMX series,
>> although it's also called 'LCDIF'.
>>> We prefer not mix these two series of IPs in one driver for ease of
>> maintenance and extension.
>>
>> Where does the MX8MP LCDIF come from then, SGTL maybe ?
> 
> AFAIK, it's RT1170. You may have a check on RM [1]. Interestingly, this SoC has both eLCDIF and LCDIFv2, two IPs we are talking about.
> 
> [1] https://www.nxp.com/webapp/Download?colCode=IMXRT1170RM

That's interesting, I wasn't aware of this one. So that's MX8MP LCDIF + 
overlays. Did the LCDIF v2 in iMXRT1070 have any predecessor too ?

Marek Vasut March 3, 2022, 4:14 a.m. UTC | #22

On 3/3/22 03:54, Liu Ying wrote:
> On Wed, 2022-03-02 at 12:57 +0100, Lucas Stach wrote:
>> Am Mittwoch, dem 02.03.2022 um 17:41 +0800 schrieb Liu Ying:
>>> On Wed, 2022-03-02 at 10:23 +0100, Lucas Stach wrote:
>>>> Am Mittwoch, dem 02.03.2022 um 03:54 +0100 schrieb Marek Vasut:
>>>>> On 3/1/22 14:18, Lucas Stach wrote:
>>>>>> Am Dienstag, dem 01.03.2022 um 07:03 -0600 schrieb Adam Ford:
>>>>>>> On Tue, Mar 1, 2022 at 5:05 AM Lucas Stach <[e-mail]> wrote:
>>>>>>>> Am Dienstag, dem 01.03.2022 um 11:19 +0100 schrieb Marek Vasut:
>>>>>>>>> On 3/1/22 11:04, Lucas Stach wrote:
>>>>>>>>>
>>>>>>>>> Hi,
>>>>>>>>>
>>>>>>>>> [...]
>>>>>>>>>
>>>>>>>>>>> Given the two totally different IPs, I don't see bugs of IP control
>>>>>>>>>>> logics should be fixed for both drivers. Naturally, the two would
>>>>>>>>>>> diverge due to different HWs. Looking at Patch 9/9, it basically
>>>>>>>>>>> squashes code to control LCDIFv3 into the mxsfb drm driver with
>>>>>>>>>>> 'if/else' checks(barely no common control code), which is hard to
>>>>>>>>>>> maintain and not able to achieve good scalability for both 'LCDIFv3'
>>>>>>>>>>> and 'LCDIF'.
>>>>>>>>>>
>>>>>>>>>> I tend to agree with Liu here. Writing a DRM driver isn't that much
>>>>>>>>>> boilerplate anymore with all the helpers we have available in the
>>>>>>>>>> framework today.
>>>>>>>>>
>>>>>>>>> I did write a separate driver for this IP before I spent time merging
>>>>>>>>> them into single driver, that's when I realized a single driver is much
>>>>>>>>> better and discarded the separate driver idea.
>>>>>>>>>
>>>>>>>>>> The IP is so different from the currently supported LCDIF controllers
>>>>>>>>>> that I think trying to support this one in the existing driver actually
>>>>>>>>>> increases the chances to break something when modifying the driver in
>>>>>>>>>> the future. Not everyone is able to test all LCDIF versions. My vote is
>>>>>>>>>> on having a separate driver for the i.MX8MP LCDIF.
>>>>>>>>>
>>>>>>>>> If you look at both controllers, it is clear it is still the LCDIF
>>>>>>>>> behind, even the CSC that is bolted on would suggest that.
>>>>>>>>
>>>>>>>> Yes, but from a driver PoV what you care about is not really the
>>>>>>>> hardware blocks used to implement something, but the programming model,
>>>>>>>> i.e. the register interface exposed to software.
>>>>>>>>
>>>>>>>>> I am also not happy when I look at the amount of duplication a separate
>>>>>>>>> driver would create, it will be some 50% of the code that would be just
>>>>>>>>> duplicated.
>>>>>>>>>
>>>>>>>> Yea, the duplicated code is still significant, as the HW itself is so
>>>>>>>> simple. However, if you find yourself in the situation where basically
>>>>>>>> every actual register access in the driver ends up being in a "if (some
>>>>>>>> HW rev) ... " clause, i still think it would be better to have a
>>>>>>>> separate driver, as the programming interface is just different.
>>>>>>>
>>>>>>> I tend to agree with Marek on this one.  We have an instance where the
>>>>>>> blk-ctrl and the GPC driver between 8m, mini, nano, plus are close,
>>>>>>> but different enough where each SoC has it's own set of tables and
>>>>>>> some checks.   Lucas created the framework, and others adapted it for
>>>>>>> various SoC's.  If there really is nearly 50% common code for the
>>>>>>> LCDIF, why not either leave the driver as one or split the common code
>>>>>>> into its own driver like lcdif-common and then have smaller drivers
>>>>>>> that handle their specific variations.
>>>>>>
>>>>>> I don't know exactly how the standalone driver looks like, but I guess
>>>>>> the overlap is not really in any real HW specific parts, but the common
>>>>>> DRM boilerplate, so there isn't much point in creating a common lcdif
>>>>>> driver.
>>>>>
>>>>> The mxsfb currently has 1280 LoC as of patch 8/9 of this series. Of
>>>>> that, there is some 400 LoC which are specific to old LCDIF and this
>>>>> patch adds 380 LoC for the new LCDIF. So that's 800 LoC or ~60% of
>>>>> shared boilerplate that would be duplicated .
>>>>
>>>> That is probably ignoring the fact that the 8MP LCDIF does not support
>>>> any overlays, so it could use the drm_simple_display_pipe
>>>> infrastructure to reduce the needed boilerplate.
>>>
>>> The drm_simple_display_pipe infrastructure is probably too simple for
>>> i.MX8MP LCDIF, since it uses one only crtc for one drm device. i.MX8MP
>>> embeds *three* LCDIF instances to support MIPI DSI, LVDS and HDMI
>>> outputs respectively. To use that infrastructure means there would be
>>> three dri cards in all. However, the three LCDIF instances can be
>>> wrapped by the one drm device, which is not the boilerplate code in the
>>> current mxsfb driver may handle.
>>
>> While that may make things a little simpler for userspace, I'm not sure
>> if this is the right thing to do. It complicates the driver a lot,
>> especially if you want to get things like independent power management,
>> etc. right. It also creates a fake view for userspace, where is looks
>> like there might be some shared resources between the different display
>> paths, while in reality they are fully independent.
> 
> Trade-off will be made between one drm device and three. My first
> impression of using the drm_simple_display_pipe infrastructure is that
> it's too simple and less flexible/scalable, because SoC designer will
> be likely to add muxes between CRTCs and encoders/bridges or overlay
> plane(s) in next generations of SoCs(SW developers don't seem have good
> reasons to suggest not to do that).  Another concern is that whether
> the userspace may use the three drm devices well or not.
> 
> A few more points:
> 1) With one drm device, userspace may use drm lease APIs to control
> those independant pipes with drm masters(not sure about the userspace
> maturity).
> 2) Code to gather all LCDIFs as one drm device has chance to be created
> as helpers once there are similar use cases in other drivers(maybe,
> there is/are already).
> 3) Power management doesn't seem to be a problem, since each LCDIF has
> it's own struct device which can be used to do runtime PM at some
> drm_crtc_helper_funcs callbacks.
> 4) Regarding the fake view of shared resources, atomic check can handle
> that, so it doesn't seem to be a big problem, either.
> 
>>
>> While we do something similar on the GPU side and collect all GPU cores
>> under a single DRM device, I'm not fully convinced that this was a good
>> decision. It now comes back to bite us when the SoC topologies get a
>> little more interesting and e.g. devices are behind different IOMMU
>> streams.
> 
> Right, SoC topologies may change, like the aforementioned muxes.
> Generally speaking, I think one drm device is more flexible and
> scalable than three.

I agree with Lucas on one driver instance - one IP instance. Each IP 
instance is separate, so it should have separate driver instance bound 
to it.

Lucas Stach March 3, 2022, 8:19 a.m. UTC | #23

Am Donnerstag, dem 03.03.2022 um 10:54 +0800 schrieb Liu Ying:
> On Wed, 2022-03-02 at 12:57 +0100, Lucas Stach wrote:
> > Am Mittwoch, dem 02.03.2022 um 17:41 +0800 schrieb Liu Ying:
> > > On Wed, 2022-03-02 at 10:23 +0100, Lucas Stach wrote:
> > > > Am Mittwoch, dem 02.03.2022 um 03:54 +0100 schrieb Marek Vasut:
> > > > > On 3/1/22 14:18, Lucas Stach wrote:
> > > > > > Am Dienstag, dem 01.03.2022 um 07:03 -0600 schrieb Adam Ford:
> > > > > > > On Tue, Mar 1, 2022 at 5:05 AM Lucas Stach <[e-mail]> wrote:
> > > > > > > > Am Dienstag, dem 01.03.2022 um 11:19 +0100 schrieb Marek Vasut:
> > > > > > > > > On 3/1/22 11:04, Lucas Stach wrote:
> > > > > > > > > 
> > > > > > > > > Hi,
> > > > > > > > > 
> > > > > > > > > [...]
> > > > > > > > > 
> > > > > > > > > > > Given the two totally different IPs, I don't see bugs of IP control
> > > > > > > > > > > logics should be fixed for both drivers. Naturally, the two would
> > > > > > > > > > > diverge due to different HWs. Looking at Patch 9/9, it basically
> > > > > > > > > > > squashes code to control LCDIFv3 into the mxsfb drm driver with
> > > > > > > > > > > 'if/else' checks(barely no common control code), which is hard to
> > > > > > > > > > > maintain and not able to achieve good scalability for both 'LCDIFv3'
> > > > > > > > > > > and 'LCDIF'.
> > > > > > > > > > 
> > > > > > > > > > I tend to agree with Liu here. Writing a DRM driver isn't that much
> > > > > > > > > > boilerplate anymore with all the helpers we have available in the
> > > > > > > > > > framework today.
> > > > > > > > > 
> > > > > > > > > I did write a separate driver for this IP before I spent time merging
> > > > > > > > > them into single driver, that's when I realized a single driver is much
> > > > > > > > > better and discarded the separate driver idea.
> > > > > > > > > 
> > > > > > > > > > The IP is so different from the currently supported LCDIF controllers
> > > > > > > > > > that I think trying to support this one in the existing driver actually
> > > > > > > > > > increases the chances to break something when modifying the driver in
> > > > > > > > > > the future. Not everyone is able to test all LCDIF versions. My vote is
> > > > > > > > > > on having a separate driver for the i.MX8MP LCDIF.
> > > > > > > > > 
> > > > > > > > > If you look at both controllers, it is clear it is still the LCDIF
> > > > > > > > > behind, even the CSC that is bolted on would suggest that.
> > > > > > > > 
> > > > > > > > Yes, but from a driver PoV what you care about is not really the
> > > > > > > > hardware blocks used to implement something, but the programming model,
> > > > > > > > i.e. the register interface exposed to software.
> > > > > > > > 
> > > > > > > > > I am also not happy when I look at the amount of duplication a separate
> > > > > > > > > driver would create, it will be some 50% of the code that would be just
> > > > > > > > > duplicated.
> > > > > > > > > 
> > > > > > > > Yea, the duplicated code is still significant, as the HW itself is so
> > > > > > > > simple. However, if you find yourself in the situation where basically
> > > > > > > > every actual register access in the driver ends up being in a "if (some
> > > > > > > > HW rev) ... " clause, i still think it would be better to have a
> > > > > > > > separate driver, as the programming interface is just different.
> > > > > > > 
> > > > > > > I tend to agree with Marek on this one.  We have an instance where the
> > > > > > > blk-ctrl and the GPC driver between 8m, mini, nano, plus are close,
> > > > > > > but different enough where each SoC has it's own set of tables and
> > > > > > > some checks.   Lucas created the framework, and others adapted it for
> > > > > > > various SoC's.  If there really is nearly 50% common code for the
> > > > > > > LCDIF, why not either leave the driver as one or split the common code
> > > > > > > into its own driver like lcdif-common and then have smaller drivers
> > > > > > > that handle their specific variations.
> > > > > > 
> > > > > > I don't know exactly how the standalone driver looks like, but I guess
> > > > > > the overlap is not really in any real HW specific parts, but the common
> > > > > > DRM boilerplate, so there isn't much point in creating a common lcdif
> > > > > > driver.
> > > > > 
> > > > > The mxsfb currently has 1280 LoC as of patch 8/9 of this series. Of 
> > > > > that, there is some 400 LoC which are specific to old LCDIF and this 
> > > > > patch adds 380 LoC for the new LCDIF. So that's 800 LoC or ~60% of 
> > > > > shared boilerplate that would be duplicated .
> > > > 
> > > > That is probably ignoring the fact that the 8MP LCDIF does not support
> > > > any overlays, so it could use the drm_simple_display_pipe
> > > > infrastructure to reduce the needed boilerplate.
> > > 
> > > The drm_simple_display_pipe infrastructure is probably too simple for
> > > i.MX8MP LCDIF, since it uses one only crtc for one drm device. i.MX8MP
> > > embeds *three* LCDIF instances to support MIPI DSI, LVDS and HDMI
> > > outputs respectively. To use that infrastructure means there would be
> > > three dri cards in all. However, the three LCDIF instances can be
> > > wrapped by the one drm device, which is not the boilerplate code in the
> > > current mxsfb driver may handle.
> > 
> > While that may make things a little simpler for userspace, I'm not sure
> > if this is the right thing to do. It complicates the driver a lot,
> > especially if you want to get things like independent power management,
> > etc. right. It also creates a fake view for userspace, where is looks
> > like there might be some shared resources between the different display
> > paths, while in reality they are fully independent.
> 
> Trade-off will be made between one drm device and three. My first
> impression of using the drm_simple_display_pipe infrastructure is that
> it's too simple and less flexible/scalable, because SoC designer will
> be likely to add muxes between CRTCs and encoders/bridges or overlay
> plane(s) in next generations of SoCs(SW developers don't seem have good
> reasons to suggest not to do that).  Another concern is that whether
> the userspace may use the three drm devices well or not. 
> 
> A few more points:
> 1) With one drm device, userspace may use drm lease APIs to control
> those independant pipes with drm masters(not sure about the userspace
> maturity).

I'm not sure why you are so keen on using DRM leases in your
downstream. Actually this argument is a argument in _favor_ of
independent DRM devices: you don't need to deal with leases when every
userspace component can just exclusively use a DRM device.

> 2) Code to gather all LCDIFs as one drm device has chance to be created
> as helpers once there are similar use cases in other drivers(maybe,
> there is/are already).

We already gather the GPU cores in etnaviv and as I said this decision
proves to add complications in the long run. For example prime import
with the DRM helpers is currently bound to the DRM device, so if your
actual HW devices backing the DRM device have differing DMA constraints
things get really messy.

> 3) Power management doesn't seem to be a problem, since each LCDIF has
> it's own struct device which can be used to do runtime PM at some
> drm_crtc_helper_funcs callbacks.

It's not a big problem, but it adds complexity that wouldn't be there
if you have a simple 1 IP instance <-> 1 DRM device mapping.

> 4) Regarding the fake view of shared resources, atomic check can handle
> that, so it doesn't seem to be a big problem, either.

Sure, there isn't even anything to handle, as the pipes are truly
independent.

> 
> > 
> > While we do something similar on the GPU side and collect all GPU cores
> > under a single DRM device, I'm not fully convinced that this was a good
> > decision. It now comes back to bite us when the SoC topologies get a
> > little more interesting and e.g. devices are behind different IOMMU
> > streams.
> 
> Right, SoC topologies may change, like the aforementioned muxes.
> Generally speaking, I think one drm device is more flexible and
> scalable than three.
> 
> Regards,
> Liu Ying
> 
>

Lucas Stach March 3, 2022, 8:21 a.m. UTC | #24

Am Donnerstag, dem 03.03.2022 um 04:14 +0100 schrieb Marek Vasut:
> On 3/2/22 10:23, Lucas Stach wrote:
> 
> [...]
> 
> > > > > I tend to agree with Marek on this one.  We have an instance where the
> > > > > blk-ctrl and the GPC driver between 8m, mini, nano, plus are close,
> > > > > but different enough where each SoC has it's own set of tables and
> > > > > some checks.   Lucas created the framework, and others adapted it for
> > > > > various SoC's.  If there really is nearly 50% common code for the
> > > > > LCDIF, why not either leave the driver as one or split the common code
> > > > > into its own driver like lcdif-common and then have smaller drivers
> > > > > that handle their specific variations.
> > > > 
> > > > I don't know exactly how the standalone driver looks like, but I guess
> > > > the overlap is not really in any real HW specific parts, but the common
> > > > DRM boilerplate, so there isn't much point in creating a common lcdif
> > > > driver.
> > > 
> > > The mxsfb currently has 1280 LoC as of patch 8/9 of this series. Of
> > > that, there is some 400 LoC which are specific to old LCDIF and this
> > > patch adds 380 LoC for the new LCDIF. So that's 800 LoC or ~60% of
> > > shared boilerplate that would be duplicated .
> > 
> > That is probably ignoring the fact that the 8MP LCDIF does not support
> > any overlays, so it could use the drm_simple_display_pipe
> > infrastructure to reduce the needed boilerplate.
> 
> It seems the IMXRT1070 LCDIF v2 (heh ...) does support overlays, so no, 
> the mxsfb and hypothetical lcdif drivers would look really very similar.
> 
> > > > As you brought up the blk-ctrl as an example: I'm all for supporting
> > > > slightly different hardware in the same driver, as long as the HW
> > > > interface is close enough. But then I also opted for a separate 8MP
> > > > blk-ctrl driver for those blk-ctrls that differ significantly from the
> > > > others, as I think it would make the common driver unmaintainable
> > > > trying to support all the different variants in one driver.
> > > 
> > > But then you also need to maintain two sets of boilerplate, they
> > > diverge, and that is not good.
> > 
> > I don't think that there is much chance for bugs going unfixed due to
> > divergence in the boilerplate, especially if you use the simple pipe
> > framework to handle most of that stuff for you, which gives you a lot
> > of code sharing with other simple DRM drivers.
> 
> But I can not use the simple pipe because overlays, see imxrt1070 .
> 
> [...]
> 
> We can always split the drivers later if this becomes unmaintainable 
> too, no ?

Not if you want to keep the same userspace running. As userspace has
some ties to the DRM driver name, e.g. for finding the right GBM
implementation, splitting the driver later on would be a UABI break.

Regards,
Lucas

Liu Ying March 3, 2022, 9:14 a.m. UTC | #25

On Thu, 2022-03-03 at 09:19 +0100, Lucas Stach wrote:
> Am Donnerstag, dem 03.03.2022 um 10:54 +0800 schrieb Liu Ying:
> > On Wed, 2022-03-02 at 12:57 +0100, Lucas Stach wrote:
> > > Am Mittwoch, dem 02.03.2022 um 17:41 +0800 schrieb Liu Ying:
> > > > On Wed, 2022-03-02 at 10:23 +0100, Lucas Stach wrote:
> > > > > Am Mittwoch, dem 02.03.2022 um 03:54 +0100 schrieb Marek Vasut:
> > > > > > On 3/1/22 14:18, Lucas Stach wrote:
> > > > > > > Am Dienstag, dem 01.03.2022 um 07:03 -0600 schrieb Adam Ford:
> > > > > > > > On Tue, Mar 1, 2022 at 5:05 AM Lucas Stach <[e-mail]> wrote:
> > > > > > > > > Am Dienstag, dem 01.03.2022 um 11:19 +0100 schrieb Marek Vasut:
> > > > > > > > > > On 3/1/22 11:04, Lucas Stach wrote:
> > > > > > > > > > 
> > > > > > > > > > Hi,
> > > > > > > > > > 
> > > > > > > > > > [...]
> > > > > > > > > > 
> > > > > > > > > > > > Given the two totally different IPs, I don't see bugs of IP control
> > > > > > > > > > > > logics should be fixed for both drivers. Naturally, the two would
> > > > > > > > > > > > diverge due to different HWs. Looking at Patch 9/9, it basically
> > > > > > > > > > > > squashes code to control LCDIFv3 into the mxsfb drm driver with
> > > > > > > > > > > > 'if/else' checks(barely no common control code), which is hard to
> > > > > > > > > > > > maintain and not able to achieve good scalability for both 'LCDIFv3'
> > > > > > > > > > > > and 'LCDIF'.
> > > > > > > > > > > 
> > > > > > > > > > > I tend to agree with Liu here. Writing a DRM driver isn't that much
> > > > > > > > > > > boilerplate anymore with all the helpers we have available in the
> > > > > > > > > > > framework today.
> > > > > > > > > > 
> > > > > > > > > > I did write a separate driver for this IP before I spent time merging
> > > > > > > > > > them into single driver, that's when I realized a single driver is much
> > > > > > > > > > better and discarded the separate driver idea.
> > > > > > > > > > 
> > > > > > > > > > > The IP is so different from the currently supported LCDIF controllers
> > > > > > > > > > > that I think trying to support this one in the existing driver actually
> > > > > > > > > > > increases the chances to break something when modifying the driver in
> > > > > > > > > > > the future. Not everyone is able to test all LCDIF versions. My vote is
> > > > > > > > > > > on having a separate driver for the i.MX8MP LCDIF.
> > > > > > > > > > 
> > > > > > > > > > If you look at both controllers, it is clear it is still the LCDIF
> > > > > > > > > > behind, even the CSC that is bolted on would suggest that.
> > > > > > > > > 
> > > > > > > > > Yes, but from a driver PoV what you care about is not really the
> > > > > > > > > hardware blocks used to implement something, but the programming model,
> > > > > > > > > i.e. the register interface exposed to software.
> > > > > > > > > 
> > > > > > > > > > I am also not happy when I look at the amount of duplication a separate
> > > > > > > > > > driver would create, it will be some 50% of the code that would be just
> > > > > > > > > > duplicated.
> > > > > > > > > > 
> > > > > > > > > Yea, the duplicated code is still significant, as the HW itself is so
> > > > > > > > > simple. However, if you find yourself in the situation where basically
> > > > > > > > > every actual register access in the driver ends up being in a "if (some
> > > > > > > > > HW rev) ... " clause, i still think it would be better to have a
> > > > > > > > > separate driver, as the programming interface is just different.
> > > > > > > > 
> > > > > > > > I tend to agree with Marek on this one.  We have an instance where the
> > > > > > > > blk-ctrl and the GPC driver between 8m, mini, nano, plus are close,
> > > > > > > > but different enough where each SoC has it's own set of tables and
> > > > > > > > some checks.   Lucas created the framework, and others adapted it for
> > > > > > > > various SoC's.  If there really is nearly 50% common code for the
> > > > > > > > LCDIF, why not either leave the driver as one or split the common code
> > > > > > > > into its own driver like lcdif-common and then have smaller drivers
> > > > > > > > that handle their specific variations.
> > > > > > > 
> > > > > > > I don't know exactly how the standalone driver looks like, but I guess
> > > > > > > the overlap is not really in any real HW specific parts, but the common
> > > > > > > DRM boilerplate, so there isn't much point in creating a common lcdif
> > > > > > > driver.
> > > > > > 
> > > > > > The mxsfb currently has 1280 LoC as of patch 8/9 of this series. Of 
> > > > > > that, there is some 400 LoC which are specific to old LCDIF and this 
> > > > > > patch adds 380 LoC for the new LCDIF. So that's 800 LoC or ~60% of 
> > > > > > shared boilerplate that would be duplicated .
> > > > > 
> > > > > That is probably ignoring the fact that the 8MP LCDIF does not support
> > > > > any overlays, so it could use the drm_simple_display_pipe
> > > > > infrastructure to reduce the needed boilerplate.
> > > > 
> > > > The drm_simple_display_pipe infrastructure is probably too simple for
> > > > i.MX8MP LCDIF, since it uses one only crtc for one drm device. i.MX8MP
> > > > embeds *three* LCDIF instances to support MIPI DSI, LVDS and HDMI
> > > > outputs respectively. To use that infrastructure means there would be
> > > > three dri cards in all. However, the three LCDIF instances can be
> > > > wrapped by the one drm device, which is not the boilerplate code in the
> > > > current mxsfb driver may handle.
> > > 
> > > While that may make things a little simpler for userspace, I'm not sure
> > > if this is the right thing to do. It complicates the driver a lot,
> > > especially if you want to get things like independent power management,
> > > etc. right. It also creates a fake view for userspace, where is looks
> > > like there might be some shared resources between the different display
> > > paths, while in reality they are fully independent.
> > 
> > Trade-off will be made between one drm device and three. My first
> > impression of using the drm_simple_display_pipe infrastructure is that
> > it's too simple and less flexible/scalable, because SoC designer will
> > be likely to add muxes between CRTCs and encoders/bridges or overlay
> > plane(s) in next generations of SoCs(SW developers don't seem have good
> > reasons to suggest not to do that).  Another concern is that whether
> > the userspace may use the three drm devices well or not. 
> > 
> > A few more points:
> > 1) With one drm device, userspace may use drm lease APIs to control
> > those independant pipes with drm masters(not sure about the userspace
> > maturity).
> 
> I'm not sure why you are so keen on using DRM leases in your
> downstream. Actually this argument is a argument in _favor_ of
> independent DRM devices: you don't need to deal with leases when every
> userspace component can just exclusively use a DRM device.

Userspace may choose to use drm lease APIs to talk to kernel. It's a
feature kinda nice to have. Imagine that an user has already written an
application which uses the APIs. 

> 
> > 2) Code to gather all LCDIFs as one drm device has chance to be created
> > as helpers once there are similar use cases in other drivers(maybe,
> > there is/are already).
> 
> We already gather the GPU cores in etnaviv and as I said this decision
> proves to add complications in the long run. For example prime import
> with the DRM helpers is currently bound to the DRM device, so if your
> actual HW devices backing the DRM device have differing DMA constraints
> things get really messy.

LCDIFs in one SoC are very likely symmetric from the embodying system
PoV. So, maybe, that's not a big problem?

> 
> > 3) Power management doesn't seem to be a problem, since each LCDIF has
> > it's own struct device which can be used to do runtime PM at some
> > drm_crtc_helper_funcs callbacks.
> 
> It's not a big problem, but it adds complexity that wouldn't be there
> if you have a simple 1 IP instance <-> 1 DRM device mapping.

Just a little bit more complexity, I think. I tend to take that to
achieve better flexibilty and scalabilty(to support potential muxes).
It's a trade-off as I mentioned - I tend to choose one drm device, but
_no_ strong opinion on three.  Once the muxes become real, it looks
like we have to use the 'one drm device' solution.

> 
> > 4) Regarding the fake view of shared resources, atomic check can handle
> > that, so it doesn't seem to be a big problem, either.
> 
> Sure, there isn't even anything to handle, as the pipes are truly
> independent.
> 

Yes, but it's still a trade-off.

Marek Vasut March 4, 2022, 4:50 a.m. UTC | #26

On 3/3/22 09:21, Lucas Stach wrote:
> Am Donnerstag, dem 03.03.2022 um 04:14 +0100 schrieb Marek Vasut:
>> On 3/2/22 10:23, Lucas Stach wrote:
>>
>> [...]
>>
>>>>>> I tend to agree with Marek on this one.  We have an instance where the
>>>>>> blk-ctrl and the GPC driver between 8m, mini, nano, plus are close,
>>>>>> but different enough where each SoC has it's own set of tables and
>>>>>> some checks.   Lucas created the framework, and others adapted it for
>>>>>> various SoC's.  If there really is nearly 50% common code for the
>>>>>> LCDIF, why not either leave the driver as one or split the common code
>>>>>> into its own driver like lcdif-common and then have smaller drivers
>>>>>> that handle their specific variations.
>>>>>
>>>>> I don't know exactly how the standalone driver looks like, but I guess
>>>>> the overlap is not really in any real HW specific parts, but the common
>>>>> DRM boilerplate, so there isn't much point in creating a common lcdif
>>>>> driver.
>>>>
>>>> The mxsfb currently has 1280 LoC as of patch 8/9 of this series. Of
>>>> that, there is some 400 LoC which are specific to old LCDIF and this
>>>> patch adds 380 LoC for the new LCDIF. So that's 800 LoC or ~60% of
>>>> shared boilerplate that would be duplicated .
>>>
>>> That is probably ignoring the fact that the 8MP LCDIF does not support
>>> any overlays, so it could use the drm_simple_display_pipe
>>> infrastructure to reduce the needed boilerplate.
>>
>> It seems the IMXRT1070 LCDIF v2 (heh ...) does support overlays, so no,
>> the mxsfb and hypothetical lcdif drivers would look really very similar.
>>
>>>>> As you brought up the blk-ctrl as an example: I'm all for supporting
>>>>> slightly different hardware in the same driver, as long as the HW
>>>>> interface is close enough. But then I also opted for a separate 8MP
>>>>> blk-ctrl driver for those blk-ctrls that differ significantly from the
>>>>> others, as I think it would make the common driver unmaintainable
>>>>> trying to support all the different variants in one driver.
>>>>
>>>> But then you also need to maintain two sets of boilerplate, they
>>>> diverge, and that is not good.
>>>
>>> I don't think that there is much chance for bugs going unfixed due to
>>> divergence in the boilerplate, especially if you use the simple pipe
>>> framework to handle most of that stuff for you, which gives you a lot
>>> of code sharing with other simple DRM drivers.
>>
>> But I can not use the simple pipe because overlays, see imxrt1070 .
>>
>> [...]
>>
>> We can always split the drivers later if this becomes unmaintainable
>> too, no ?
> 
> Not if you want to keep the same userspace running. As userspace has
> some ties to the DRM driver name, e.g. for finding the right GBM
> implementation, splitting the driver later on would be a UABI break.

Hum, so what other options do we have left ? Duplicate 60% of the driver 
right away ?

Patch

diff --git a/Documentation/devicetree/bindings/display/fsl,lcdif.yaml b/Documentation/devicetree/bindings/display/fsl,lcdif.yaml
index 900a56cae80e6..9831ab53a053d 100644
--- a/Documentation/devicetree/bindings/display/fsl,lcdif.yaml
+++ b/Documentation/devicetree/bindings/display/fsl,lcdif.yaml
@@ -28,6 +28,7 @@  properties:
               - fsl,imx7d-lcdif
               - fsl,imx8mm-lcdif
               - fsl,imx8mn-lcdif
+              - fsl,imx8mp-lcdif
               - fsl,imx8mq-lcdif
           - const: fsl,imx6sx-lcdif