Skip to contentSkip to conversation

rockchip: correctly set vop0 or vop1

PATCH Patrick Wildt 4 participants 8 messages Jun 7, 2020 – Jun 28, 2020

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

The EDP_LCDC_SEL bit has to be set correctly to select vop0 or vop1, but so far we have set it in both conditions, which is not correct.

Can someone verify this is the correct way round? vop1 -> set, vop0 -> clear?

drivers/video/rockchip/rk_edp.c +2 -1
index 92188be9275..000bd481408 100644
@@ -1062,7 +1062,8 @@ static int rk_edp_probe(struct udevice *dev)
 	rk_setreg(&priv->grf->soc_con12, 1 << 4);
 
 	/* select epd signal from vop0 or vop1 */
-	rk_setreg(&priv->grf->soc_con6, (vop_id == 1) ? (1 << 5) : (1 << 5));
+	rk_clrsetreg(&priv->grf->soc_con6, (1 << 5),
+	    (vop_id == 1) ? (1 << 5) : (0 << 5));
 
 	rockchip_edp_wait_hpd(priv);
 
#1 lore raw

Patrick Wildt <[email protected]> writes:

Hi,

Quoted 14 lines

The EDP_LCDC_SEL bit has to be set correctly to select vop0 or vop1, but so far we have set it in both conditions, which is not correct.

Can someone verify this is the correct way round? vop1 -> set, vop0 -> clear?

drivers/video/rockchip/rk_edp.c +0 -0
index 92188be9275..000bd481408 100644

drivers/video/rockchip/rk_edp.c static int rk_edp_probe(struct udevice *dev) · line 1066

Quoted 6 lines
drivers/video/rockchip/rk_edp.c +2 -1
@@ -1062,7 +1062,8 @@ static int rk_edp_probe(struct udevice *dev)
 	rk_setreg(&priv->grf->soc_con12, 1 << 4);
 
 	/* select epd signal from vop0 or vop1 */
-	rk_setreg(&priv->grf->soc_con6, (vop_id == 1) ? (1 << 5) : (1 << 5));
+	rk_clrsetreg(&priv->grf->soc_con6, (1 << 5),
+	    (vop_id == 1) ? (1 << 5) : (0 << 5));

While working on PBP EDP support, found this too but I'm not sure it's fine or not. For rk3399, my (not yet published) patch is doing:

+ if (vop_id == 0)
+ rk_clrreg(&priv->grf->soc_con20, (1 << 5));
+ else
+ rk_setreg(&priv->grf->soc_con20, (1 << 5));

I believe that the rk3288 may need similar treatment but I've yet to look@the rk3288 manual.

Arnaud

#2 lore raw
Quoted 33 lines · Arnaud Patard

Patrick Wildt <[email protected]> writes:

Hi,

Quoted 14 lines

The EDP_LCDC_SEL bit has to be set correctly to select vop0 or vop1, but so far we have set it in both conditions, which is not correct.

Can someone verify this is the correct way round? vop1 -> set, vop0 -> clear?

drivers/video/rockchip/rk_edp.c +0 -0
index 92188be9275..000bd481408 100644

drivers/video/rockchip/rk_edp.c static int rk_edp_probe(struct udevice *dev) · line 1066

Quoted 6 lines
drivers/video/rockchip/rk_edp.c +2 -1
@@ -1062,7 +1062,8 @@ static int rk_edp_probe(struct udevice *dev)
 	rk_setreg(&priv->grf->soc_con12, 1 << 4);
 
 	/* select epd signal from vop0 or vop1 */
-	rk_setreg(&priv->grf->soc_con6, (vop_id == 1) ? (1 << 5) : (1 << 5));
+	rk_clrsetreg(&priv->grf->soc_con6, (1 << 5),
+	    (vop_id == 1) ? (1 << 5) : (0 << 5));

While working on PBP EDP support, found this too but I'm not sure it's fine or not. For rk3399, my (not yet published) patch is doing:

(quoted diff) +4 -0
+       if (vop_id == 0)
+               rk_clrreg(&priv->grf->soc_con20, (1 << 5));
+       else
+               rk_setreg(&priv->grf->soc_con20, (1 << 5));
Quoted 4 lines

I believe that the rk3288 may need similar treatment but I've yet to look at the rk3288 manual.

Arnaud

Yes, it does. If you look at the linux code, they have:

static const struct rockchip_dp_chip_data rk3399_edp = {
        .lcdsel_grf_reg = RK3399_GRF_SOC_CON20,
        .lcdsel_big = HIWORD_UPDATE(0, RK3399_EDP_LCDC_SEL),
        .lcdsel_lit = HIWORD_UPDATE(RK3399_EDP_LCDC_SEL, RK3399_EDP_LCDC_SEL),
        .chip_type = RK3399_EDP,
};
static const struct rockchip_dp_chip_data rk3288_dp = {
        .lcdsel_grf_reg = RK3288_GRF_SOC_CON6,
        .lcdsel_big = HIWORD_UPDATE(0, RK3288_EDP_LCDC_SEL),
        .lcdsel_lit = HIWORD_UPDATE(RK3288_EDP_LCDC_SEL, RK3288_EDP_LCDC_SEL),
        .chip_type = RK3288_DP,
};

which indicates that for rk3399 *and* rk3288 the bit has to be set to select "lit". Now your diff looks equivalent to mine, apart from using a different operation to achieve the same goal.

The linux code does

        ret = drm_of_encoder_active_endpoint_id(dp->dev->of_node, encoder);
        if (ret < 0)
                return;

        if (ret)
                val = dp->data->lcdsel_lit;
        else
                val = dp->data->lcdsel_big;

Assuming that endpoint_id essentiall returns vop id 0 or vop id 1, this would mean that vop1 -> lit -> set bit and vop0 -> big -> clr bit.

That said, my diff seems to be fine, and your RK3399 code as well. Do you agree?

Patrick

#3 lore raw
Quoted 70 lines · Patrick Wildt
Quoted 33 lines · Arnaud Patard

Patrick Wildt <[email protected]> writes:

Hi,

Quoted 14 lines

The EDP_LCDC_SEL bit has to be set correctly to select vop0 or vop1, but so far we have set it in both conditions, which is not correct.

Can someone verify this is the correct way round? vop1 -> set, vop0 -> clear?

drivers/video/rockchip/rk_edp.c +0 -0
index 92188be9275..000bd481408 100644

drivers/video/rockchip/rk_edp.c static int rk_edp_probe(struct udevice *dev) · line 1066

Quoted 6 lines
drivers/video/rockchip/rk_edp.c +2 -1
@@ -1062,7 +1062,8 @@ static int rk_edp_probe(struct udevice *dev)
 	rk_setreg(&priv->grf->soc_con12, 1 << 4);
 
 	/* select epd signal from vop0 or vop1 */
-	rk_setreg(&priv->grf->soc_con6, (vop_id == 1) ? (1 << 5) : (1 << 5));
+	rk_clrsetreg(&priv->grf->soc_con6, (1 << 5),
+	    (vop_id == 1) ? (1 << 5) : (0 << 5));

While working on PBP EDP support, found this too but I'm not sure it's fine or not. For rk3399, my (not yet published) patch is doing:

(quoted diff) +4 -0
+       if (vop_id == 0)
+               rk_clrreg(&priv->grf->soc_con20, (1 << 5));
+       else
+               rk_setreg(&priv->grf->soc_con20, (1 << 5));
Quoted 4 lines

I believe that the rk3288 may need similar treatment but I've yet to look at the rk3288 manual.

Arnaud

Yes, it does. If you look at the linux code, they have:

static const struct rockchip_dp_chip_data rk3399_edp = {
        .lcdsel_grf_reg = RK3399_GRF_SOC_CON20,
        .lcdsel_big = HIWORD_UPDATE(0, RK3399_EDP_LCDC_SEL),
        .lcdsel_lit = HIWORD_UPDATE(RK3399_EDP_LCDC_SEL, RK3399_EDP_LCDC_SEL),
        .chip_type = RK3399_EDP,
};
static const struct rockchip_dp_chip_data rk3288_dp = {
        .lcdsel_grf_reg = RK3288_GRF_SOC_CON6,
        .lcdsel_big = HIWORD_UPDATE(0, RK3288_EDP_LCDC_SEL),
        .lcdsel_lit = HIWORD_UPDATE(RK3288_EDP_LCDC_SEL, RK3288_EDP_LCDC_SEL),
        .chip_type = RK3288_DP,
};

which indicates that for rk3399 *and* rk3288 the bit has to be set to select "lit". Now your diff looks equivalent to mine, apart from using a different operation to achieve the same goal.

The linux code does

        ret = drm_of_encoder_active_endpoint_id(dp->dev->of_node, encoder);
        if (ret < 0)
                return;

        if (ret)
                val = dp->data->lcdsel_lit;
        else
                val = dp->data->lcdsel_big;
Quoted 5 lines

Assuming that endpoint_id essentiall returns vop id 0 or vop id 1, this would mean that vop1 -> lit -> set bit and vop0 -> big -> clr bit.

That said, my diff seems to be fine, and your RK3399 code as well. Do you agree?

According to the code you've shown, it should be fine for rk3288 I guess but not for rk3399. Please note that it's grf soc_con6 register for rk3288 but grf soc_con20 for rk3399.

Arnaud

#4 lore raw
Quoted 75 lines · Arnaud Patard
Quoted 73 lines · Patrick Wildt
Quoted 33 lines · Arnaud Patard

Patrick Wildt <[email protected]> writes:

Hi,

Quoted 14 lines

The EDP_LCDC_SEL bit has to be set correctly to select vop0 or vop1, but so far we have set it in both conditions, which is not correct.

Can someone verify this is the correct way round? vop1 -> set, vop0 -> clear?

drivers/video/rockchip/rk_edp.c +0 -0
index 92188be9275..000bd481408 100644

drivers/video/rockchip/rk_edp.c static int rk_edp_probe(struct udevice *dev) · line 1066

Quoted 6 lines
drivers/video/rockchip/rk_edp.c +2 -1
@@ -1062,7 +1062,8 @@ static int rk_edp_probe(struct udevice *dev)
 	rk_setreg(&priv->grf->soc_con12, 1 << 4);
 
 	/* select epd signal from vop0 or vop1 */
-	rk_setreg(&priv->grf->soc_con6, (vop_id == 1) ? (1 << 5) : (1 << 5));
+	rk_clrsetreg(&priv->grf->soc_con6, (1 << 5),
+	    (vop_id == 1) ? (1 << 5) : (0 << 5));

While working on PBP EDP support, found this too but I'm not sure it's fine or not. For rk3399, my (not yet published) patch is doing:

(quoted diff) +4 -0
+       if (vop_id == 0)
+               rk_clrreg(&priv->grf->soc_con20, (1 << 5));
+       else
+               rk_setreg(&priv->grf->soc_con20, (1 << 5));
Quoted 4 lines

I believe that the rk3288 may need similar treatment but I've yet to look at the rk3288 manual.

Arnaud

Yes, it does. If you look at the linux code, they have:

static const struct rockchip_dp_chip_data rk3399_edp = {
        .lcdsel_grf_reg = RK3399_GRF_SOC_CON20,
        .lcdsel_big = HIWORD_UPDATE(0, RK3399_EDP_LCDC_SEL),
        .lcdsel_lit = HIWORD_UPDATE(RK3399_EDP_LCDC_SEL, RK3399_EDP_LCDC_SEL),
        .chip_type = RK3399_EDP,
};
static const struct rockchip_dp_chip_data rk3288_dp = {
        .lcdsel_grf_reg = RK3288_GRF_SOC_CON6,
        .lcdsel_big = HIWORD_UPDATE(0, RK3288_EDP_LCDC_SEL),
        .lcdsel_lit = HIWORD_UPDATE(RK3288_EDP_LCDC_SEL, RK3288_EDP_LCDC_SEL),
        .chip_type = RK3288_DP,
};

which indicates that for rk3399 *and* rk3288 the bit has to be set to select "lit". Now your diff looks equivalent to mine, apart from using a different operation to achieve the same goal.

The linux code does

        ret = drm_of_encoder_active_endpoint_id(dp->dev->of_node, encoder);
        if (ret < 0)
                return;

        if (ret)
                val = dp->data->lcdsel_lit;
        else
                val = dp->data->lcdsel_big;

Assuming that endpoint_id essentiall returns vop id 0 or vop id 1, this would mean that vop1 -> lit -> set bit and vop0 -> big -> clr bit.

Quoted 8 lines
Quoted 2 lines

That said, my diff seems to be fine, and your RK3399 code as well. Do you agree?

According to the code you've shown, it should be fine for rk3288 I guess but not for rk3399. Please note that it's grf soc_con6 register for rk3288 but grf soc_con20 for rk3399.

Arnaud

Exactly, which is why you had that if defined() in your diff, to compile one part of the code for RK3288, and the other for RK3399. :) The bit though happens to be the same.

#5 lore raw

+Andy Yan for this topic,

Hi Patrick and Arnaud,

??? I would like to leave this patch until the code fits for all the socs,

Thanks,

- Kever

Quoted 86 lines · Patrick Wildt
Quoted 73 lines · Arnaud Patard
Quoted 71 lines · Patrick Wildt
Quoted 32 lines · Arnaud Patard

Patrick Wildt <[email protected]> writes:

Hi,

Quoted 14 lines

The EDP_LCDC_SEL bit has to be set correctly to select vop0 or vop1, but so far we have set it in both conditions, which is not correct.

Can someone verify this is the correct way round? vop1 -> set, vop0 -> clear?

drivers/video/rockchip/rk_edp.c +0 -0
index 92188be9275..000bd481408 100644

drivers/video/rockchip/rk_edp.c static int rk_edp_probe(struct udevice *dev) · line 1066

Quoted 6 lines
drivers/video/rockchip/rk_edp.c +2 -1
@@ -1062,7 +1062,8 @@ static int rk_edp_probe(struct udevice *dev)
  	rk_setreg(&priv->grf->soc_con12, 1 << 4);
 
  	/* select epd signal from vop0 or vop1 */
-	rk_setreg(&priv->grf->soc_con6, (vop_id == 1) ? (1 << 5) : (1 << 5));
+	rk_clrsetreg(&priv->grf->soc_con6, (1 << 5),
+	    (vop_id == 1) ? (1 << 5) : (0 << 5));

While working on PBP EDP support, found this too but I'm not sure it's fine or not. For rk3399, my (not yet published) patch is doing:

(quoted diff) +4 -0
+       if (vop_id == 0)
+               rk_clrreg(&priv->grf->soc_con20, (1 << 5));
+       else
+               rk_setreg(&priv->grf->soc_con20, (1 << 5));
Quoted 4 lines

I believe that the rk3288 may need similar treatment but I've yet to look at the rk3288 manual.

Arnaud

Yes, it does. If you look at the linux code, they have:

static const struct rockchip_dp_chip_data rk3399_edp = {
         .lcdsel_grf_reg = RK3399_GRF_SOC_CON20,
         .lcdsel_big = HIWORD_UPDATE(0, RK3399_EDP_LCDC_SEL),
         .lcdsel_lit = HIWORD_UPDATE(RK3399_EDP_LCDC_SEL, RK3399_EDP_LCDC_SEL),
         .chip_type = RK3399_EDP,
};
static const struct rockchip_dp_chip_data rk3288_dp = {
         .lcdsel_grf_reg = RK3288_GRF_SOC_CON6,
         .lcdsel_big = HIWORD_UPDATE(0, RK3288_EDP_LCDC_SEL),
         .lcdsel_lit = HIWORD_UPDATE(RK3288_EDP_LCDC_SEL, RK3288_EDP_LCDC_SEL),
         .chip_type = RK3288_DP,
};

which indicates that for rk3399 *and* rk3288 the bit has to be set to select "lit". Now your diff looks equivalent to mine, apart from using a different operation to achieve the same goal.

The linux code does

         ret = drm_of_encoder_active_endpoint_id(dp->dev->of_node, encoder);
         if (ret < 0)
                 return;

         if (ret)
                 val = dp->data->lcdsel_lit;
         else
                 val = dp->data->lcdsel_big;

Assuming that endpoint_id essentiall returns vop id 0 or vop id 1, this would mean that vop1 -> lit -> set bit and vop0 -> big -> clr bit.

Quoted 7 lines
Quoted 2 lines

That said, my diff seems to be fine, and your RK3399 code as well. Do you agree?

According to the code you've shown, it should be fine for rk3288 I guess but not for rk3399. Please note that it's grf soc_con6 register for rk3288 but grf soc_con20 for rk3399.

Arnaud

Exactly, which is why you had that if defined() in your diff, to compile one part of the code for RK3288, and the other for RK3399. :) The bit though happens to be the same.

#6 lore raw

Hi :

Quoted 4 lines · Kever Yang

+Andy Yan for this topic,

Hi Patrick and Arnaud,

Quoted 68 lines

??? I would like to leave this patch until the code fits for all the socs,

Thanks,

- Kever

Quoted 60 lines · Patrick Wildt
Quoted 59 lines · Arnaud Patard
Quoted 57 lines · Patrick Wildt
Quoted 34 lines · Arnaud Patard

Patrick Wildt <[email protected]> writes:

Hi,

Quoted 16 lines

The EDP_LCDC_SEL bit has to be set correctly to select vop0 or vop1, but so far we have set it in both conditions, which is not correct.

Can someone verify this is the correct way round?? vop1 -> set, vop0 -> clear?

drivers/video/rockchip/rk_edp.c +0 -0
index 92188be9275..000bd481408 100644
@@ -1062,7 +1062,8 @@ static int rk_edp_probe(struct udevice *dev)
 ????? rk_setreg(&priv->grf->soc_con12, 1 << 4);
Quoted 5 lines

? ????? /* select epd signal from vop0 or vop1 */ -??? rk_setreg(&priv->grf->soc_con6, (vop_id == 1) ? (1 << 5) :
(1 << 5));
+??? rk_clrsetreg(&priv->grf->soc_con6, (1 << 5), +??????? (vop_id == 1) ? (1 << 5) : (0 << 5));

While working on PBP EDP support, found this too but I'm not sure it's
fine or not. For rk3399, my (not yet published) patch is doing:

+?????? if (vop_id == 0)
+?????????????? rk_clrreg(&priv->grf->soc_con20, (1 << 5)); +?????? else
+?????????????? rk_setreg(&priv->grf->soc_con20, (1 << 5));

Quoted 4 lines

I believe that the rk3288 may need similar treatment but I've yet to look at the rk3288 manual.

Arnaud

Yes, it does.? If you look at the linux code, they have:

static const struct rockchip_dp_chip_data rk3399_edp = { ???????? .lcdsel_grf_reg = RK3399_GRF_SOC_CON20, ???????? .lcdsel_big = HIWORD_UPDATE(0, RK3399_EDP_LCDC_SEL), ???????? .lcdsel_lit = HIWORD_UPDATE(RK3399_EDP_LCDC_SEL, RK3399_EDP_LCDC_SEL),
???????? .chip_type = RK3399_EDP,
};

static const struct rockchip_dp_chip_data rk3288_dp = { ???????? .lcdsel_grf_reg = RK3288_GRF_SOC_CON6, ???????? .lcdsel_big = HIWORD_UPDATE(0, RK3288_EDP_LCDC_SEL), ???????? .lcdsel_lit = HIWORD_UPDATE(RK3288_EDP_LCDC_SEL, RK3288_EDP_LCDC_SEL),
???????? .chip_type = RK3288_DP,
};

It's true that different soc have different grf register for selecting lcdc/vop, and so it is for other modules such as rockchip_gmac/pinctrl. The above code in linux kernel is a example for how? we handle this case.

Quoted 38 lines
Quoted 35 lines
Quoted 23 lines
Quoted 17 lines

which indicates that for rk3399 *and* rk3288 the bit has to be set to select "lit".? Now your diff looks equivalent to mine, apart from using
a different operation to achieve the same goal.

The linux code does

???????? ret = drm_of_encoder_active_endpoint_id(dp->dev->of_node, encoder);
???????? if (ret < 0)
???????????????? return;

???????? if (ret)
???????????????? val = dp->data->lcdsel_lit;
???????? else
???????????????? val = dp->data->lcdsel_big;

Quoted 6 lines

Assuming that endpoint_id essentiall returns vop id 0 or vop id 1, this
would mean that vop1 -> lit -> set bit and vop0 -> big -> clr bit.

That said, my diff seems to be fine, and your RK3399 code as well.? Do you agree?

Quoted 7 lines

According to the code you've shown, it should be fine for rk3288 I guess
but not for rk3399. Please note that it's grf soc_con6 register for rk3288
but grf soc_con20 for rk3399.

Arnaud

Exactly, which is why you had that if defined() in your diff, to compile one part of the code for RK3288, and the other for RK3399. :) The bit though happens to be the same.

Reviewed-by
#7 lore raw
In reply to Patrick Wildt
Quoted 8 lines · Patrick Wildt

The EDP_LCDC_SEL bit has to be set correctly to select vop0 or vop1, but so far we have set it in both conditions, which is not correct.

Can someone verify this is the correct way round? vop1 -> set, vop0 -> clear?

I will take this fix for rk3288 and later you can send support for rk3399.

Thanks,
- Kever

Quoted 17 lines
drivers/video/rockchip/rk_edp.c +2 -1
index 92188be9275..000bd481408 100644
@@ -1062,7 +1062,8 @@ static int rk_edp_probe(struct udevice *dev)
  	rk_setreg(&priv->grf->soc_con12, 1 << 4);
 
  	/* select epd signal from vop0 or vop1 */
-	rk_setreg(&priv->grf->soc_con6, (vop_id == 1) ? (1 << 5) : (1 << 5));
+	rk_clrsetreg(&priv->grf->soc_con6, (1 << 5),
+	    (vop_id == 1) ? (1 << 5) : (0 << 5));
 
  	rockchip_edp_wait_hpd(priv);