Re: [PATCH] drm/mxsfb: use bus_format to determine LCD bus width

From: Marek Vasut
Date: Wed Dec 14 2016 - 05:46:18 EST


On 12/14/2016 12:56 AM, Stefan Agner wrote:
> On 2016-12-08 20:24, Marek Vasut wrote:
>> On 12/09/2016 04:44 AM, Stefan Agner wrote:
>>> On 2016-12-08 15:33, Marek Vasut wrote:
>>>> On 12/08/2016 11:52 PM, Stefan Agner wrote:
>>>>> The LCD bus width does not need to align with the pixel format. The
>>>>> LCDIF controller automatically converts between pixel formats and
>>>>> bus width by padding or dropping LSBs.
>>>>>
>>>>> The DRM subsystem has the notion of bus_format which allows to
>>>>> determine what bus_formats are supported by the display. Choose the
>>>>> first available or fallback to 24 bit if none are available.
>>>>>
>>>>> Signed-off-by: Stefan Agner <stefan@xxxxxxxx>
>>>>> ---
>>>>> drivers/gpu/drm/mxsfb/mxsfb_crtc.c | 28 +++++++++++++++++++++++++---
>>>>> drivers/gpu/drm/mxsfb/mxsfb_regs.h | 1 +
>>>>> 2 files changed, 26 insertions(+), 3 deletions(-)
>>>>>
>>>>> diff --git a/drivers/gpu/drm/mxsfb/mxsfb_crtc.c b/drivers/gpu/drm/mxsfb/mxsfb_crtc.c
>>>>> index 4bcc8a3..00fa244 100644
>>>>> --- a/drivers/gpu/drm/mxsfb/mxsfb_crtc.c
>>>>> +++ b/drivers/gpu/drm/mxsfb/mxsfb_crtc.c
>>>>> @@ -65,13 +65,11 @@ static int mxsfb_set_pixel_fmt(struct mxsfb_drm_private *mxsfb)
>>>>
>>>> You should probably drop the WARNING: comment in this function too.
>>>>
>>>>> switch (format) {
>>>>> case DRM_FORMAT_RGB565:
>>>>> dev_dbg(drm->dev, "Setting up RGB565 mode\n");
>>>>> - ctrl |= CTRL_SET_BUS_WIDTH(STMLCDIF_16BIT);
>>>>> ctrl |= CTRL_SET_WORD_LENGTH(0);
>>>>> ctrl1 |= CTRL1_SET_BYTE_PACKAGING(0xf);
>>>>> break;
>>>>> case DRM_FORMAT_XRGB8888:
>>>>> dev_dbg(drm->dev, "Setting up XRGB8888 mode\n");
>>>>> - ctrl |= CTRL_SET_BUS_WIDTH(STMLCDIF_24BIT);
>>>>> ctrl |= CTRL_SET_WORD_LENGTH(3);
>>>>> /* Do not use packed pixels = one pixel per word instead. */
>>>>> ctrl1 |= CTRL1_SET_BYTE_PACKAGING(0x7);
>>>>> @@ -89,6 +87,9 @@ static int mxsfb_set_pixel_fmt(struct mxsfb_drm_private *mxsfb)
>>>>>
>>>>> static void mxsfb_enable_controller(struct mxsfb_drm_private *mxsfb)
>>>>> {
>>>>> + struct drm_crtc *crtc = &mxsfb->pipe.crtc;
>>>>> + struct drm_device *drm = crtc->dev;
>>>>> + u32 bus_format = MEDIA_BUS_FMT_RGB888_1X24;
>>>>
>>>> So why do you move the bus width configuration from
>>>> mxsfb_set_pixel_fmt() to here ? Is there any reason
>>>> for it?
>>>>
>>>
>>> To emphasize that it is not related to pixel format.
>>
>> Ah, can we then create some function, something like mxsfb_set_bus_fmt()
>> to be explicit about this bus (the wires coming out of the CPU) and
>> pixel (memory layout of the framebuffer) format difference ? I think
>> that'd help with readability.
>
> Sure we can, we just have to read the value of the register once more...
>
>>
>>> Also, if you have a
>>> controller with multiple framebuffers/layers, mxsfb_set_pixel_fmt would
>>> get called per layer...
>>
>> Which in this case, you don't and cannot have, right ?
>>
>
> I think the controller has basically support for one additional surface.
> They call it LCDIFx_AS (Alpha Surface). But those register have a
> slightly different layout, so I guess we can't really reuse
> mxsfb_set_pixel_fmt...

Oh, this is new since MX6 :) All right, good point.

>>> In a full DRM driver it probably would be part of a encoder function, I
>>> feel here it seems to map best to enable controller. It could probably
>>> also be in mxsfb_crtc_enable (or even mxsfb_crtc_mode_set_nofb I guess),
>>> but we do not touch LCDC_CTRL there...
>>
>> My feeling is it should be part of the mode_set_nofb, because we're
>> setting all the controller parameters there, so that should be all
>> kept in one place, which for a simple controller like this might be
>> the approach to take.
>>
>
> It kind of feels wrong in any CRTC callback, since it is a property of
> an encoder. But since we only have one CRTC and one encoder it really
> does not matter.
>
> My point was that we don't touch that register in mode_set_nofb. But
> when we anyway move the code in a separate function anyway, than we
> might as well call it from mxsfb_crtc_mode_set_nofb.

Given the above information, I kinda get your concern now.
Either way works for me then.

>>>>> u32 reg;
>>>>>
>>>>> if (mxsfb->clk_disp_axi)
>>>>> @@ -97,7 +98,28 @@ static void mxsfb_enable_controller(struct mxsfb_drm_private *mxsfb)
>>>>> mxsfb_enable_axi_clk(mxsfb);
>>>>>
>>>>> /* If it was disabled, re-enable the mode again */
>>>>> - writel(CTRL_DOTCLK_MODE, mxsfb->base + LCDC_CTRL + REG_SET);
>>>>> + reg = readl(mxsfb->base + LCDC_CTRL);
>>>>
>>>> Wouldn't it make more sense to cache the LCDC_CTRL content rather than
>>>> doing R-M-W on it ?
>>>>
>>>
>>> Not sure what you mean by cache? Isn't the variable reg counting as a
>>> cache?
>>
>> I mean caching the content of the register in struct mxsfb_drm_private,
>> so you always program content which you have under control into the
>> register. Heck, maybe I should've used regmap for this driver ...
>>
>
> The FSL DCU driver (used for Vybrid/LS1021a) which I currently maintain
> uses regmap. But it doesn't feel quite right either, the DRM subsystem
> actually holds current state already, just on a higher level. I first
> tried to use it for suspend and resume (before the DRM suspend resume
> helper were available), but it was messy, stuff got async between
> DRM/actual controller settings or have been written twice
> (unnecessarily).
>
>
> Quite often you actually have to touch a register from one subsystem
> only too (e.g. in DCU, the layer registers maps nicely with the DRM
> plane support). I guess this one register is a bit a catch all
> configuration register... Thinking about it, IMHO reading/writing that
> one register multiple times is really not a big deal...

Well, then we should probably go for RMW afterall.

--
Best regards,
Marek Vasut