Add plane funcs and helper funcs for ADE.

v6: None.
v5: None.
v4: None.
v3:
- A few cleanup.
v2:
- Remove abtraction layer.

Signed-off-by: Xinliang Liu <xinliang.liu@linaro.org>
---
  drivers/gpu/drm/hisilicon/kirin/kirin_drm_ade.c | 535 +++++++++++++++++++++++-
  1 file changed, 534 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/hisilicon/kirin/kirin_drm_ade.c b/drivers/gpu/drm/hisilicon/kirin/kirin_drm_ade.c
index bb93616dcf3d..aa2cf75cab39 100644
--- a/drivers/gpu/drm/hisilicon/kirin/kirin_drm_ade.c
+++ b/drivers/gpu/drm/hisilicon/kirin/kirin_drm_ade.c

@@ -27,13 +27,23 @@
  #include <drm/drm_crtc_helper.h>
  #include <drm/drm_atomic.h>
  #include <drm/drm_atomic_helper.h>
+#include <drm/drm_plane_helper.h>
+#include <drm/drm_gem_cma_helper.h>
+#include <drm/drm_fb_cma_helper.h>

  #include "kirin_drm_drv.h"
  #include "kirin_ade_reg.h"

+#define PRIMARY_CH	ADE_CH1 /* primary plane */
+#define OUT_OVLY	ADE_OVLY2 /* output overlay compositor */

+#define ADE_DEBUG	1
+
  #define to_ade_crtc(crtc) \
  	container_of(crtc, struct ade_crtc, base)

+#define to_ade_plane(plane) \
+	container_of(plane, struct ade_plane, base)
+
  struct ade_hw_ctx {
  	void __iomem  *base;
  	struct regmap *noc_regmap;

@@ -52,11 +62,76 @@ struct ade_crtc {
  	u32 out_format;
  };

+struct ade_plane {
+	struct drm_plane base;
+	void *ctx;
+	u8 ch; /* channel */
+};
+
  struct ade_data {
  	struct ade_crtc acrtc;
+	struct ade_plane aplane[ADE_CH_NUM];
  	struct ade_hw_ctx ctx;
  };

+/* ade-format info: */
+struct ade_format {
+	u32 pixel_format;
+	enum ade_fb_format ade_format;
+};
+
+static const struct ade_format ade_formats[] = {
+	/* 16bpp RGB: */
+	{ DRM_FORMAT_RGB565, ADE_RGB_565 },
+	{ DRM_FORMAT_BGR565, ADE_BGR_565 },
+	/* 24bpp RGB: */
+	{ DRM_FORMAT_RGB888, ADE_RGB_888 },
+	{ DRM_FORMAT_BGR888, ADE_BGR_888 },
+	/* 32bpp [A]RGB: */
+	{ DRM_FORMAT_XRGB8888, ADE_XRGB_8888 },
+	{ DRM_FORMAT_XBGR8888, ADE_XBGR_8888 },
+	{ DRM_FORMAT_RGBA8888, ADE_RGBA_8888 },
+	{ DRM_FORMAT_BGRA8888, ADE_BGRA_8888 },
+	{ DRM_FORMAT_ARGB8888, ADE_ARGB_8888 },
+	{ DRM_FORMAT_ABGR8888, ADE_ABGR_8888 },
+};
+
+static const u32 channel_formats1[] = {
+	/* channel 1,2,3,4 */
+	DRM_FORMAT_RGB565, DRM_FORMAT_BGR565, DRM_FORMAT_RGB888,
+	DRM_FORMAT_BGR888, DRM_FORMAT_XRGB8888, DRM_FORMAT_XBGR8888,
+	DRM_FORMAT_RGBA8888, DRM_FORMAT_BGRA8888, DRM_FORMAT_ARGB8888,
+	DRM_FORMAT_ABGR8888
+};
+
+u32 ade_get_channel_formats(u8 ch, const u32 **formats)
+{
+	switch (ch) {
+	case ADE_CH1:
+		*formats = channel_formats1;
+		return ARRAY_SIZE(channel_formats1);
+	default:
+		DRM_ERROR("no this channel %d\n", ch);
+		*formats = NULL;
+		return 0;
+	}
+}
+
+/* convert from fourcc format to ade format */
+static u32 ade_get_format(u32 pixel_format)
+{
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(ade_formats); i++)
+		if (ade_formats[i].pixel_format == pixel_format)
+			return ade_formats[i].ade_format;
+
+	/* not found */
+	DRM_ERROR("Not found pixel format!!fourcc_format= %d\n",
+		  pixel_format);
+	return ADE_FORMAT_NOT_SUPPORT;
+}
+
  static void ade_update_reload_bit(void __iomem *base, u32 bit_num, u32 val)
  {
  	u32 bit_ofst, reg_num;

@@ -89,7 +164,7 @@ static void ade_init(struct ade_hw_ctx *ctx)
  	/* clear overlay */
  	writel(0, base + ADE_OVLY1_TRANS_CFG);
  	writel(0, base + ADE_OVLY_CTL);
-	writel(0, base + ADE_OVLYX_CTL(ADE_OVLY2));
+	writel(0, base + ADE_OVLYX_CTL(OUT_OVLY));
  	/* clear reset and reload regs */
  	writel(MASK(32), base + ADE_SOFT_RST_SEL(0));
  	writel(MASK(32), base + ADE_SOFT_RST_SEL(1));

@@ -147,6 +222,10 @@ static void ade_ldi_set_mode(struct ade_crtc *acrtc,
  			  mode->clock * 1000, ret);
  	adj_mode->clock = clk_get_rate(ctx->ade_pix_clk) / 1000;

+	/* set overlay compositor output size */
+	writel(((width - 1) << OUTPUT_XSIZE_OFST) | (height - 1),
+	       base + ADE_OVLY_OUTPUT_SIZE(OUT_OVLY));
+
  	/* ctran6 setting */
  	writel(CTRAN_BYPASS_ON, base + ADE_CTRAN_DIS(ADE_CTRAN6));
  	 /* the configured value is actual value - 1 */

@@ -219,6 +298,10 @@ static void ade_display_enable(struct ade_crtc *acrtc)
  	void __iomem *base = ctx->base;
  	u32 out_fmt = acrtc->out_format;

+	/* enable output overlay compositor */
+	writel(ADE_ENABLE, base + ADE_OVLYX_CTL(OUT_OVLY));
+	ade_update_reload_bit(base, OVLY_OFST + OUT_OVLY, 0);
+
  	/* display source setting */
  	writel(DISP_SRC_OVLY2, base + ADE_DISP_SRC_CFG);

@@ -232,6 +315,97 @@ static void ade_display_enable(struct ade_crtc *acrtc)
  	writel(DSI_PCLK_ON, base + LDI_HDMI_DSI_GT);
  }

+#if ADE_DEBUG
+static void ade_rdma_dump_regs(void __iomem *base, u32 ch)
+{
+	u32 reg_ctrl, reg_addr, reg_size, reg_stride, reg_space, reg_en;
+	u32 val;
+
+	reg_ctrl = RD_CH_CTRL(ch);
+	reg_addr = RD_CH_ADDR(ch);
+	reg_size = RD_CH_SIZE(ch);
+	reg_stride = RD_CH_STRIDE(ch);
+	reg_space = RD_CH_SPACE(ch);
+	reg_en = RD_CH_EN(ch);
+
+	val = ade_read_reload_bit(base, RDMA_OFST + ch);
+	DRM_DEBUG_DRIVER("[rdma%d]: reload(%d)\n", ch + 1, val);
+	val = readl(base + reg_ctrl);
+	DRM_DEBUG_DRIVER("[rdma%d]: reg_ctrl(0x%08x)\n", ch + 1, val);
+	val = readl(base + reg_addr);
+	DRM_DEBUG_DRIVER("[rdma%d]: reg_addr(0x%08x)\n", ch + 1, val);
+	val = readl(base + reg_size);
+	DRM_DEBUG_DRIVER("[rdma%d]: reg_size(0x%08x)\n", ch + 1, val);
+	val = readl(base + reg_stride);
+	DRM_DEBUG_DRIVER("[rdma%d]: reg_stride(0x%08x)\n", ch + 1, val);
+	val = readl(base + reg_space);
+	DRM_DEBUG_DRIVER("[rdma%d]: reg_space(0x%08x)\n", ch + 1, val);
+	val = readl(base + reg_en);
+	DRM_DEBUG_DRIVER("[rdma%d]: reg_en(0x%08x)\n", ch + 1, val);
+}
+
+static void ade_clip_dump_regs(void __iomem *base, u32 ch)
+{
+	u32 val;
+
+	val = ade_read_reload_bit(base, CLIP_OFST + ch);
+	DRM_DEBUG_DRIVER("[clip%d]: reload(%d)\n", ch + 1, val);
+	val = readl(base + ADE_CLIP_DISABLE(ch));
+	DRM_DEBUG_DRIVER("[clip%d]: reg_clip_disable(0x%08x)\n", ch + 1, val);
+	val = readl(base + ADE_CLIP_SIZE0(ch));
+	DRM_DEBUG_DRIVER("[clip%d]: reg_clip_size0(0x%08x)\n", ch + 1, val);
+	val = readl(base + ADE_CLIP_SIZE1(ch));
+	DRM_DEBUG_DRIVER("[clip%d]: reg_clip_size1(0x%08x)\n", ch + 1, val);
+}
+
+static void ade_compositor_routing_dump_regs(void __iomem *base, u32 ch)
+{
+	u8 ovly_ch = 0; /* TODO: Only primary plane now */
+	u32 val;
+
+	val = readl(base + ADE_OVLY_CH_XY0(ovly_ch));
+	DRM_DEBUG_DRIVER("[overlay ch%d]: reg_ch_xy0(0x%08x)\n", ovly_ch, val);
+	val = readl(base + ADE_OVLY_CH_XY1(ovly_ch));
+	DRM_DEBUG_DRIVER("[overlay ch%d]: reg_ch_xy1(0x%08x)\n", ovly_ch, val);
+	val = readl(base + ADE_OVLY_CH_CTL(ovly_ch));
+	DRM_DEBUG_DRIVER("[overlay ch%d]: reg_ch_ctl(0x%08x)\n", ovly_ch, val);
+}
+
+static void ade_dump_overlay_compositor_regs(void __iomem *base, u32 comp)
+{
+	u32 val;
+
+	val = ade_read_reload_bit(base, OVLY_OFST + comp);
+	DRM_DEBUG_DRIVER("[overlay%d]: reload(%d)\n", comp + 1, val);
+	writel(ADE_ENABLE, base + ADE_OVLYX_CTL(comp));
+	DRM_DEBUG_DRIVER("[overlay%d]: reg_ctl(0x%08x)\n", comp + 1, val);
+	val = readl(base + ADE_OVLY_CTL);
+	DRM_DEBUG_DRIVER("ovly_ctl(0x%08x)\n", val);
+}
+
+static void ade_dump_regs(void __iomem *base)
+{
+	u32 i;
+
+	/* dump channel regs */
+	for (i = 0; i < ADE_CH_NUM; i++) {
+		/* dump rdma regs */
+		ade_rdma_dump_regs(base, i);
+
+		/* dump clip regs */
+		ade_clip_dump_regs(base, i);
+
+		/* dump compositor routing regs */
+		ade_compositor_routing_dump_regs(base, i);
+	}
+
+	/* dump overlay compositor regs */
+	ade_dump_overlay_compositor_regs(base, OUT_OVLY);
+}
+#else
+static void ade_dump_regs(void __iomem *base) { }
+#endif
+
  static void ade_crtc_enable(struct drm_crtc *crtc)
  {
  	struct ade_crtc *acrtc = to_ade_crtc(crtc);

@@ -249,6 +423,7 @@ static void ade_crtc_enable(struct drm_crtc *crtc)

  	ade_set_medianoc_qos(acrtc);
  	ade_display_enable(acrtc);
+	ade_dump_regs(ctx->base);
  	acrtc->enable = true;
  }

@@ -303,6 +478,7 @@ static void ade_crtc_atomic_flush(struct drm_crtc *crtc,

  	/* only crtc is eanbled regs take effect */
  	if (acrtc->enable) {
+		ade_dump_regs(base);
  		/* flush ade regitsters */
  		writel(ADE_ENABLE, base + ADE_EN);
  	}

@@ -359,6 +535,338 @@ static int ade_crtc_init(struct drm_device *dev, struct drm_crtc *crtc,
  	return 0;
  }

+static void ade_rdma_set(void __iomem *base, struct drm_framebuffer *fb,
+			 u32 ch, u32 y, u32 in_h, u32 fmt)
+{
+	struct drm_gem_cma_object *obj = drm_fb_cma_get_gem_obj(fb, 0);
+	u32 reg_ctrl, reg_addr, reg_size, reg_stride, reg_space, reg_en;
+	u32 stride = fb->pitches[0];
+	u32 addr = (u32)obj->paddr + y * stride;
+
+	DRM_DEBUG_DRIVER("rdma%d: (y=%d, height=%d), stride=%d, paddr=0x%x\n",
+			 ch + 1, y, in_h, stride, (u32)obj->paddr);
+	DRM_DEBUG_DRIVER("addr=0x%x, fb:%dx%d, pixel_format=%d(%s)\n",
+			 addr, fb->width, fb->height, fmt,
+			 drm_get_format_name(fb->pixel_format));
+
+	/* get reg offset */
+	reg_ctrl = RD_CH_CTRL(ch);
+	reg_addr = RD_CH_ADDR(ch);
+	reg_size = RD_CH_SIZE(ch);
+	reg_stride = RD_CH_STRIDE(ch);
+	reg_space = RD_CH_SPACE(ch);
+	reg_en = RD_CH_EN(ch);
+
+	/*
+	 * TODO: set rotation
+	 */
+	writel((fmt << 16) & 0x1f0000, base + reg_ctrl);
+	writel(addr, base + reg_addr);
+	writel((in_h << 16) | stride, base + reg_size);
+	writel(stride, base + reg_stride);
+	writel(in_h * stride, base + reg_space);
+	writel(ADE_ENABLE, base + reg_en);
+	ade_update_reload_bit(base, RDMA_OFST + ch, 0);
+}
+
+static void ade_rdma_disable(void __iomem *base, u32 ch)
+{
+	u32 reg_en;
+
+	/* get reg offset */
+	reg_en = RD_CH_EN(ch);
+	writel(0, base + reg_en);
+	ade_update_reload_bit(base, RDMA_OFST + ch, 1);
+}
+
+static void ade_clip_set(void __iomem *base, u32 ch, u32 fb_w, u32 x,
+			 u32 in_w, u32 in_h)
+{
+	u32 disable_val;
+	u32 clip_left;
+	u32 clip_right;
+
+	/*
+	 * clip width, no need to clip height
+	 */
+	if (fb_w == in_w) { /* bypass */
+		disable_val = 1;
+		clip_left = 0;
+		clip_right = 0;
+	} else {
+		disable_val = 0;
+		clip_left = x;
+		clip_right = fb_w - (x + in_w) - 1;
+	}
+
+	DRM_DEBUG_DRIVER("clip%d: clip_left=%d, clip_right=%d\n",
+			 ch + 1, clip_left, clip_right);
+
+	writel(disable_val, base + ADE_CLIP_DISABLE(ch));
+	writel((fb_w - 1) << 16 | (in_h - 1), base + ADE_CLIP_SIZE0(ch));
+	writel(clip_left << 16 | clip_right, base + ADE_CLIP_SIZE1(ch));
+	ade_update_reload_bit(base, CLIP_OFST + ch, 0);
+}
+
+static void ade_clip_disable(void __iomem *base, u32 ch)
+{
+	writel(1, base + ADE_CLIP_DISABLE(ch));
+	ade_update_reload_bit(base, CLIP_OFST + ch, 1);
+}
+
+static bool has_Alpha_channel(int format)
+{
+	switch (format) {
+	case ADE_ARGB_8888:
+	case ADE_ABGR_8888:
+	case ADE_RGBA_8888:
+	case ADE_BGRA_8888:
+		return true;
+	default:
+		return false;
+	}
+}
+
+static void ade_get_blending_params(u32 fmt, u8 glb_alpha, u8 *alp_mode,
+				    u8 *alp_sel, u8 *under_alp_sel)
+{
+	bool has_alpha = has_Alpha_channel(fmt);
+
+	/*
+	 * get alp_mode
+	 */
+	if (has_alpha && glb_alpha < 255)
+		*alp_mode = ADE_ALP_PIXEL_AND_GLB;
+	else if (has_alpha)
+		*alp_mode = ADE_ALP_PIXEL;
+	else
+		*alp_mode = ADE_ALP_GLOBAL;
+
+	/*
+	 * get alp sel
+	 */
+	*alp_sel = ADE_ALP_MUL_COEFF_3; /* 1 */
+	*under_alp_sel = ADE_ALP_MUL_COEFF_2; /* 0 */
+}
+
+static void ade_compositor_routing_set(void __iomem *base, u8 ch,
+				       u32 x0, u32 y0,
+				       u32 in_w, u32 in_h, u32 fmt)
+{
+	u8 ovly_ch = 0; /* TODO: This is the zpos, only one plane now */