[U-Boot] [RESEND PATCH 5/5] sunxi: video: add LCD support to DE2 driver
Vasily Khoruzhick
anarsoul at gmail.com
Thu Sep 21 05:33:36 UTC 2017
Hi,
I did few tests, see results inline.
On Tue, Sep 19, 2017 at 12:00 PM, Vasily Khoruzhick <anarsoul at gmail.com> wrote:
> On Tue, Sep 19, 2017 at 1:33 AM, Maxime Ripard
> <maxime.ripard at free-electrons.com> wrote:
>> On Mon, Sep 18, 2017 at 10:04:21PM -0700, Vasily Khoruzhick wrote:
>>> Extend DE2 driver with LCD support
>>
>> (All) your commit messages could use a bit more details.
>
> OK, will add in v2.
>
>> Here, for example, explaining the following things would help:
>> - Why are you creating yet another file
>
> Are you talking about any specific file? I guess adding another driver
> justifies creation of another file.
>
>> - What is the situation with old Allwinner SoCs that should share
>> the same code
>
> As far as I can tell, DE2 is present in H3, V3s, A64 and newer. LCD is
> supported in A64 only. I.e. hardware is not present
> in H3 or V3s
>
>> - What are the expected users
>
> Pinebook
>
>> - Which SoC / board have you tested it on
>
> A64 / Pinebook
>
>>
>> etc...
>>
>>> Signed-off-by: Vasily Khoruzhick <anarsoul at gmail.com>
>>> ---
>>> arch/arm/mach-sunxi/Kconfig | 2 +-
>>> drivers/video/sunxi/Makefile | 2 +-
>>> drivers/video/sunxi/sunxi_de2.c | 17 +++++
>>> drivers/video/sunxi/sunxi_lcd.c | 142 ++++++++++++++++++++++++++++++++++++++++
>>> 4 files changed, 161 insertions(+), 2 deletions(-)
>>> create mode 100644 drivers/video/sunxi/sunxi_lcd.c
>>>
>>> diff --git a/arch/arm/mach-sunxi/Kconfig b/arch/arm/mach-sunxi/Kconfig
>>> index 2309f59999..06d697e3a7 100644
>>> --- a/arch/arm/mach-sunxi/Kconfig
>>> +++ b/arch/arm/mach-sunxi/Kconfig
>>> @@ -680,7 +680,7 @@ config VIDEO_LCD_MODE
>>>
>>> config VIDEO_LCD_DCLK_PHASE
>>> int "LCD panel display clock phase"
>>> - depends on VIDEO
>>> + depends on VIDEO || DM_VIDEO
>>> default 1
>>> ---help---
>>> Select LCD panel display clock phase shift, range 0-3.
>>> diff --git a/drivers/video/sunxi/Makefile b/drivers/video/sunxi/Makefile
>>> index 0d64c2021f..8c91766c24 100644
>>> --- a/drivers/video/sunxi/Makefile
>>> +++ b/drivers/video/sunxi/Makefile
>>> @@ -6,4 +6,4 @@
>>> #
>>>
>>> obj-$(CONFIG_VIDEO_SUNXI) += sunxi_display.o lcdc.o tve_common.o ../videomodes.o
>>> -obj-$(CONFIG_VIDEO_DE2) += sunxi_de2.o sunxi_dw_hdmi.o lcdc.o ../dw_hdmi.o
>>> +obj-$(CONFIG_VIDEO_DE2) += sunxi_de2.o sunxi_dw_hdmi.o lcdc.o ../dw_hdmi.o sunxi_lcd.o
>>> diff --git a/drivers/video/sunxi/sunxi_de2.c b/drivers/video/sunxi/sunxi_de2.c
>>> index ee67764ac5..a838bbacd1 100644
>>> --- a/drivers/video/sunxi/sunxi_de2.c
>>> +++ b/drivers/video/sunxi/sunxi_de2.c
>>> @@ -232,6 +232,23 @@ static int sunxi_de2_probe(struct udevice *dev)
>>> if (!(gd->flags & GD_FLG_RELOC))
>>> return 0;
>>>
>>> + ret = uclass_find_device_by_name(UCLASS_DISPLAY,
>>> + "sunxi_lcd", &disp);
>>> + if (!ret) {
>>> + int mux;
>>> +
>>> + mux = 0;
>>> +
>>> + ret = sunxi_de2_init(dev, plat->base, VIDEO_BPP32, disp, mux,
>>> + false);
>>> + if (!ret) {
>>> + video_set_flush_dcache(dev, 1);
>>
>> Why do you need to flush the dcache here?
>
> Copied from HDMI driver init. If it's not necessary why it's here for HDMI?
DE2 is not cache aware, so CPU should flush dcache after updating
framebuffer. If I remove this line,
dcache isn't flushed when framebuffer is updated, and thus picture is
a total mess (black background with some white stripes).
>>> + return 0;
>>> + }
>>> + }
>>> +
>>> + debug("%s: lcd display not found (ret=%d)\n", __func__, ret);
>>> +
>>> ret = uclass_find_device_by_name(UCLASS_DISPLAY,
>>> "sunxi_dw_hdmi", &disp);
>>> if (!ret) {
>>> diff --git a/drivers/video/sunxi/sunxi_lcd.c b/drivers/video/sunxi/sunxi_lcd.c
>>> new file mode 100644
>>> index 0000000000..154eb5835e
>>> --- /dev/null
>>> +++ b/drivers/video/sunxi/sunxi_lcd.c
>>> @@ -0,0 +1,142 @@
>>> +/*
>>> + * Allwinner LCD driver
>>> + *
>>> + * (C) Copyright 2017 Vasily Khoruzhick <anarsoul at gmail.com>
>>> + *
>>> + * SPDX-License-Identifier: GPL-2.0+
>>> + */
>>> +
>>> +#include <common.h>
>>> +#include <display.h>
>>> +#include <video_bridge.h>
>>> +#include <backlight.h>
>>> +#include <dm.h>
>>> +#include <edid.h>
>>> +#include <asm/io.h>
>>> +#include <asm/arch/clock.h>
>>> +#include <asm/arch/lcdc.h>
>>> +#include <asm/arch/gpio.h>
>>> +#include <asm/gpio.h>
>>> +
>>> +struct sunxi_lcd_priv {
>>> + struct display_timing timing;
>>> + int panel_bpp;
>>> +};
>>> +
>>> +static void sunxi_lcdc_config_pinmux(void)
>>> +{
>>> + int pin;
>>> + for (pin = SUNXI_GPD(0); pin <= SUNXI_GPD(21); pin++) {
>>> + sunxi_gpio_set_cfgpin(pin, SUNXI_GPD_LCD0);
>>> + sunxi_gpio_set_drv(pin, 3);
>>> + }
>>> +}
>>> +
>>> +static int sunxi_lcd_enable(struct udevice *dev, int bpp,
>>> + const struct display_timing *edid)
>>> +{
>>> + struct sunxi_ccm_reg * const ccm =
>>> + (struct sunxi_ccm_reg *)SUNXI_CCM_BASE;
>>> + struct sunxi_lcdc_reg * const lcdc =
>>> + (struct sunxi_lcdc_reg *)SUNXI_LCD0_BASE;
>>> + struct sunxi_lcd_priv *priv = dev_get_priv(dev);
>>> + struct udevice *backlight;
>>> + int clk_div, clk_double, ret;
>>> +
>>> + /* Reset off */
>>> + setbits_le32(&ccm->ahb_reset1_cfg, 1 << AHB_RESET_OFFSET_LCD0);
>>> +
>>> + /* Clock on */
>>> + setbits_le32(&ccm->ahb_gate1, 1 << AHB_GATE_OFFSET_LCD0);
>>
>> This has nothing to do with using a panel or not, it should be in
>> lcdc_init().
>
> Why? We don't need neither take it out of reset nor turn the clock on
> it if LCD is not used (e.g. HDMI-only case).
I'm leaving it here. It's not necessary for HDMI, and it doesn't work
without it for LCD.
>
>>> + lcdc_init(lcdc);
>>> + sunxi_lcdc_config_pinmux();
>>
>> This is already handled in sunxi_lcdc_tcon0_mode_set, why duplicate
>> it?
>
> Because the one that sunxi_lcdc_tcon0_mode_set() calls is
> DE1-specific. I don't want to split out that code that won't be used
> by DE2 driver.
>
>>> + lcdc_pll_set(ccm, 0, edid->pixelclock.typ / 1000,
>>> + &clk_div, &clk_double);
>>> + lcdc_tcon0_mode_set(lcdc, edid, clk_div, false,
>>> + priv->panel_bpp, CONFIG_VIDEO_LCD_DCLK_PHASE);
>>> + lcdc_enable(lcdc, priv->panel_bpp);
>>> +
>>> + ret = uclass_get_device(UCLASS_PANEL_BACKLIGHT, 0, &backlight);
>>> + if (!ret)
>>> + backlight_enable(backlight);
>>> +
>>> + return 0;
>>> +}
>>> +
>>> +static int sunxi_lcd_read_timing(struct udevice *dev,
>>> + struct display_timing *timing)
>>> +{
>>> + struct sunxi_lcd_priv *priv = dev_get_priv(dev);
>>> + memcpy(timing, &priv->timing, sizeof(struct display_timing));
>>> +
>>> + return 0;
>>> +}
>>> +
>>> +static int sunxi_lcd_probe(struct udevice *dev)
>>> +{
>>> + struct udevice *cdev;
>>> + struct sunxi_lcd_priv *priv = dev_get_priv(dev);
>>> + int ret;
>>> +
>>> + /* make sure that clock is active */
>>> + clock_set_pll10(432000000);
>>
>> Why do you need it active, and why at that rate?
>
> Will check.
Not necessary, will drop in v2.
>>> +
>>> +#ifdef CONFIG_VIDEO_BRIDGE
>>> + /* Try to get timings from bridge first */
>>> + ret = uclass_get_device(UCLASS_VIDEO_BRIDGE, 0, &cdev);
>>> + if (!ret) {
>>> + u8 edid[EDID_SIZE];
>>> + int channel_bpp;
>>> +
>>> + ret = video_bridge_attach(cdev);
>>> + if (ret) {
>>> + debug("video bridge attach failed: %d\n", ret);
>>> + return ret;
>>> + }
>>> + ret = video_bridge_read_edid(cdev, edid, EDID_SIZE);
>>> + if (ret <= 0) {
>>> + debug("video bridge failed to read edid: %d\n", ret);
>>> + return ret ? ret : -EIO;
>>> + }
>>> + ret = edid_get_timing(edid, ret, &priv->timing, &channel_bpp);
>>> + priv->panel_bpp = channel_bpp * 3;
>>> + return ret;
>>> + }
>>> + debug("video bridge not found: %d\n", ret);
>>> +#endif
>>> + /* Fallback to timings from DT if there's no bridge */
>>> + ret = uclass_get_device(UCLASS_PANEL, 0, &cdev);
>>> + if (ret) {
>>> + debug("video panel not found: %d\n", ret);
>>> + return ret;
>>> + }
>>> +
>>> + if (fdtdec_decode_display_timing(gd->fdt_blob, dev_of_offset(cdev),
>>> + 0, &priv->timing)) {
>>> + debug("%s: Failed to decode display timing\n", __func__);
>>> + return -EINVAL;
>>> + }
>>
>> How does that work with simple-panel style bindings that are most of
>> the panels merged these days?
>
> It should just work, but I haven't even tested this part of the code -
> I don't have a panel with parallel interface nor A64 board to connect
> it to.
> Do you want me to drop it and make this driver dependent on CONFIG_VIDEO_BRIDGE
I tested missing bridge codepath as Jerjej suggested and it works
fine. I extended it with reading panel bpp from fdt to make it even
better.
>>> + priv->panel_bpp = 16;
>>> +
>>> + return 0;
>>> +}
>>> +
>>> +static const struct dm_display_ops sunxi_lcd_ops = {
>>> + .read_timing = sunxi_lcd_read_timing,
>>> + .enable = sunxi_lcd_enable,
>>> +};
>>> +
>>> +U_BOOT_DRIVER(sunxi_lcd) = {
>>> + .name = "sunxi_lcd",
>>> + .id = UCLASS_DISPLAY,
>>> + .ops = &sunxi_lcd_ops,
>>> + .probe = sunxi_lcd_probe,
>>> + .priv_auto_alloc_size = sizeof(struct sunxi_lcd_priv),
>>> +};
>>> +
>>> +#ifdef CONFIG_MACH_SUN50I
>>
>> Why do you restrict it to the A64 here?
>
> Because the hardware is present in A64 only.
>
>> Thanks,
>> Maxime
>>
>> --
>> Maxime Ripard, Free Electrons
>> Embedded Linux and Kernel engineering
>> http://free-electrons.com
More information about the U-Boot
mailing list