Skip to contentSkip to conversation

rk3399 (Pinebook pro) EDP support

PATCH Superseded Arnaud Patard (Rtp) 4 participants 24 messages Oct 27, 2020 – Nov 13, 2020

Subsystems (4)

v2 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

v2 cover rk3399 (Pinebook pro) EDP support

This patchset add support for the rk3399 eDP. It has been tested on the pinebook pro and Google Kevin chromeos devices.

The changes have been written by studying the linux code, since I didn't find any manual for theses part of the RK3399 SoC.

For Kevin devices, this patchset is needed:
https://patchwork.ozlabs.org/project/uboot/list/?series=209553

On the linux kernel side, on recent kernels, it needs commit "pwm: rockchip: Keep enabled PWMs running while probing" otherwise the pinebook pro will freeze when probing the display.

Changes since v2:
- Add reset support for the VOP and eDP. This was needed and it appears to also solve the issue with warm reset.
- Fix a debug string in the vop driver
- Drop the patch "drivers/video/rockchip/rk_edp.c: Change clock rate".
- Address various comments

#1 lore raw

PATCH v2 1/10 drivers/video/rockchip/rk_vop.c: Use endpoint compatible string to find VOP mode

The current code is using an hard coded enum and the of node reg value of endpoint to find out if the endpoint is mipi/hdmi/lvds/edp/dp. The order is different between rk3288, rk3399 vop little, rk3399 vop big.

A possible solution would be to make sure that the rk3288.dtsi and rk3399.dtsi files have "expected" reg value or an other solution is to find the kind of endpoint by comparing the endpoint compatible value.

This patch is implementing the more flexible second solution.

arch/arm/include/asm/arch-rockchip/vop_rk3288.h +1 -14
Index: u-boot/arch/arm/include/asm/arch-rockchip/vop_rk3288.h
===================================================================
@@ -85,26 +85,13 @@ enum {
 	LB_RGB_1280X8 = 0x5
 };
 
-#if defined(CONFIG_ROCKCHIP_RK3399)
 enum vop_modes {
 	VOP_MODE_EDP = 0,
 	VOP_MODE_MIPI,
 	VOP_MODE_HDMI,
-	VOP_MODE_MIPI1,
-	VOP_MODE_DP,
-	VOP_MODE_NONE,
-};
-#else
-enum vop_modes {
-	VOP_MODE_EDP = 0,
-	VOP_MODE_HDMI,
 	VOP_MODE_LVDS,
-	VOP_MODE_MIPI,
-	VOP_MODE_NONE,
-	VOP_MODE_AUTO_DETECT,
-	VOP_MODE_UNKNOWN,
+	VOP_MODE_DP,
 };
-#endif
 
 /* VOP_VERSION_INFO */
 #define M_FPGA_VERSION (0xffff << 16)
drivers/video/rockchip/rk_vop.c +23 -2
Index: u-boot/drivers/video/rockchip/rk_vop.c
===================================================================
@@ -235,12 +235,11 @@ static int rk_display_init(struct udevic
 	struct clk clk;
 	enum video_log2_bpp l2bpp;
 	ofnode remote;
+	const char *compat;
 
 	debug("%s(%s, %lu, %s)\n", __func__,
 	      dev_read_name(dev), fbbase, ofnode_get_name(ep_node));
 
-	vop_id = ofnode_read_s32_default(ep_node, "reg", -1);
-	debug("vop_id=%d\n", vop_id);
 	ret = ofnode_read_u32(ep_node, "remote-endpoint", &remote_phandle);
 	if (ret)
 		return ret;
@@ -282,6 +281,28 @@ static int rk_display_init(struct udevic
 		if (disp)
 			break;
 	};
+	compat = ofnode_get_property(remote, "compatible", NULL);
+	if (!compat) {
+		debug("%s(%s): Failed to find compatible property\n",
+		      __func__, dev_read_name(dev));
+		return -EINVAL;
+	}
+	if (strstr(compat, "edp")) {
+		vop_id = VOP_MODE_EDP;
+	} else if (strstr(compat, "mipi")) {
+		vop_id = VOP_MODE_MIPI;
+	} else if (strstr(compat, "hdmi")) {
+		vop_id = VOP_MODE_HDMI;
+	} else if (strstr(compat, "cdn-dp")) {
+		vop_id = VOP_MODE_DP;
+	} else if (strstr(compat, "lvds")) {
+		vop_id = VOP_MODE_LVDS;
+	} else {
+		debug("%s(%s): Failed to find vop mode for %s\n",
+		      __func__, dev_read_name(dev), compat);
+		return -EINVAL;
+	}
+	debug("vop_id=%d\n", vop_id);
 
 	disp_uc_plat = dev_get_uclass_platdata(disp);
 	debug("Found device '%s', disp_uc_priv=%p\n", disp->name, disp_uc_plat);
#6 lore raw

PATCH v2 2/10 drivers/video/rockchip/rk_edp.c: Add rk3399 support

In reply to Arnaud Patard (Rtp)

According to linux commit "drm/rockchip: analogix_dp: add rk3399 eDP support" (82872e42bb1501dd9e60ca430f4bae45a469aa64), rk3288 and rk3399 eDP IPs are nearly the same, the difference is in the grf register (SOC_CON6 versus SOC_CON20). So, change the code to use the right register on each IP.

The clocks don't seem to be the same, the eDP clock is not at index 1 on rk3399, so don't try changing the clock at index 1 to rate 0 on rk3399.

drivers/video/rockchip/rk_edp.c +65 -20
Index: u-boot/drivers/video/rockchip/rk_edp.c
===================================================================
@@ -17,11 +17,10 @@
 #include <asm/gpio.h>
 #include <asm/io.h>
 #include <asm/arch-rockchip/clock.h>
+#include <asm/arch-rockchip/hardware.h>
 #include <asm/arch-rockchip/edp_rk3288.h>
 #include <asm/arch-rockchip/grf_rk3288.h>
-#include <asm/arch-rockchip/hardware.h>
-#include <dt-bindings/clock/rk3288-cru.h>
-#include <linux/delay.h>
+#include <asm/arch-rockchip/grf_rk3399.h>
 
 #define MAX_CR_LOOP 5
 #define MAX_EQ_LOOP 5
@@ -37,18 +36,42 @@ static const char * const pre_emph_names
 #define DP_VOLTAGE_MAX         DP_TRAIN_VOLTAGE_SWING_1200
 #define DP_PRE_EMPHASIS_MAX    DP_TRAIN_PRE_EMPHASIS_9_5
 
+#define RK3288_GRF_SOC_CON6	0x025c
+#define RK3288_GRF_SOC_CON12	0x0274
+#define RK3399_GRF_SOC_CON20	0x6250
+#define RK3399_GRF_SOC_CON25	0x6264
+
+enum rockchip_dp_types {
+	RK3288_DP = 0,
+	RK3399_EDP
+};
+
+struct rockchip_dp_data {
+	unsigned long reg_vop_big_little;
+	unsigned long reg_vop_big_little_sel;
+	unsigned long reg_ref_clk_sel;
+	unsigned long ref_clk_sel_bit;
+	enum rockchip_dp_types chip_type;
+};
+
 struct rk_edp_priv {
 	struct rk3288_edp *regs;
-	struct rk3288_grf *grf;
+	void *grf;
 	struct udevice *panel;
 	struct link_train link_train;
 	u8 train_set[4];
 };
 
-static void rk_edp_init_refclk(struct rk3288_edp *regs)
+static void rk_edp_init_refclk(struct rk3288_edp *regs, enum rockchip_dp_types chip_type)
 {
 	writel(SEL_24M, &regs->analog_ctl_2);
-	writel(REF_CLK_24M, &regs->pll_reg_1);
+	u32 reg;
+
+	reg = REF_CLK_24M;
+	if (chip_type == RK3288_DP)
+		reg ^= REF_CLK_MASK;
+	writel(reg, &regs->pll_reg_1);
+
 
 	writel(LDO_OUTPUT_V_SEL_145 | KVCO_DEFALUT | CHG_PUMP_CUR_SEL_5US |
 	       V2L_CUR_SEL_1MA, &regs->pll_reg_2);
@@ -1023,6 +1046,8 @@ static int rk_edp_probe(struct udevice *
 	struct display_plat *uc_plat = dev_get_uclass_platdata(dev);
 	struct rk_edp_priv *priv = dev_get_priv(dev);
 	struct rk3288_edp *regs = priv->regs;
+	struct rockchip_dp_data *edp_data = (struct rockchip_dp_data *)dev_get_driver_data(dev);
+
 	struct clk clk;
 	int ret;
 
@@ -1037,16 +1062,17 @@ static int rk_edp_probe(struct udevice *
 	int vop_id = uc_plat->source_id;
 	debug("%s, uc_plat=%p, vop_id=%u\n", __func__, uc_plat, vop_id);
 
-	ret = clk_get_by_index(dev, 1, &clk);
-	if (ret >= 0) {
-		ret = clk_set_rate(&clk, 0);
-		clk_free(&clk);
-	}
-	if (ret) {
-		debug("%s: Failed to set EDP clock: ret=%d\n", __func__, ret);
-		return ret;
+	if (edp_data->chip_type == RK3288_DP) {
+		ret = clk_get_by_index(dev, 1, &clk);
+		if (ret >= 0) {
+			ret = clk_set_rate(&clk, 0);
+			clk_free(&clk);
+		}
+		if (ret) {
+			debug("%s: Failed to set EDP clock: ret=%d\n", __func__, ret);
+			return ret;
+		}
 	}
-
 	ret = clk_get_by_index(uc_plat->src_dev, 0, &clk);
 	if (ret >= 0) {
 		ret = clk_set_rate(&clk, 192000000);
@@ -1059,15 +1085,17 @@ static int rk_edp_probe(struct udevice *
 	}
 
 	/* grf_edp_ref_clk_sel: from internal 24MHz or 27MHz clock */
-	rk_setreg(&priv->grf->soc_con12, 1 << 4);
+	rk_setreg(priv->grf + edp_data->reg_ref_clk_sel,
+		  edp_data->ref_clk_sel_bit);
 
 	/* select epd signal from vop0 or vop1 */
-	rk_clrsetreg(&priv->grf->soc_con6, (1 << 5),
-	    (vop_id == 1) ? (1 << 5) : (0 << 5));
+	rk_clrsetreg(priv->grf + edp_data->reg_vop_big_little,
+		     edp_data->reg_vop_big_little_sel,
+		     (vop_id == 1) ? edp_data->reg_vop_big_little_sel : 0);
 
 	rockchip_edp_wait_hpd(priv);
 
-	rk_edp_init_refclk(regs);
+	rk_edp_init_refclk(regs, edp_data->chip_type);
 	rk_edp_init_interrupt(regs);
 	rk_edp_enable_sw_function(regs);
 	ret = rk_edp_init_analog_func(regs);
@@ -1083,8 +1111,25 @@ static const struct dm_display_ops dp_ro
 	.enable = rk_edp_enable,
 };
 
+static const struct rockchip_dp_data rk3399_edp = {
+	.reg_vop_big_little = RK3399_GRF_SOC_CON20,
+	.reg_vop_big_little_sel = BIT(5),
+	.reg_ref_clk_sel = RK3399_GRF_SOC_CON25,
+	.ref_clk_sel_bit = BIT(11),
+	.chip_type = RK3399_EDP,
+};
+
+static const struct rockchip_dp_data rk3288_dp = {
+	.reg_vop_big_little = RK3288_GRF_SOC_CON6,
+	.reg_vop_big_little_sel = BIT(5),
+	.reg_ref_clk_sel = RK3288_GRF_SOC_CON12,
+	.ref_clk_sel_bit = BIT(4),
+	.chip_type = RK3288_DP,
+};
+
 static const struct udevice_id rockchip_dp_ids[] = {
-	{ .compatible = "rockchip,rk3288-edp" },
+	{ .compatible = "rockchip,rk3288-edp", .data = (ulong)&rk3288_dp },
+	{ .compatible = "rockchip,rk3399-edp", .data = (ulong)&rk3399_edp },
 	{ }
 };
 
arch/arm/include/asm/arch-rockchip/edp_rk3288.h +3 -2
Index: u-boot/arch/arm/include/asm/arch-rockchip/edp_rk3288.h
===================================================================
@@ -232,8 +232,9 @@ check_member(rk3288_edp, pll_reg_5, 0xa0
 #define PD_CH0					(0x1 << 0)
 
 /* pll_reg_1 */
-#define REF_CLK_24M				(0x1 << 1)
-#define REF_CLK_27M				(0x0 << 1)
+#define REF_CLK_24M				(0x1 << 0)
+#define REF_CLK_27M				(0x0 << 0)
+#define REF_CLK_MASK				(0x1 << 0)
 
 /* line_map */
 #define LANE3_MAP_LOGIC_LANE_0			(0x0 << 6)
#8 lore raw

PATCH v2 3/10 drivers/video/rockchip/rk_edp.c: Change interrupt polarity configuration

In reply to Arnaud Patard (Rtp)

The linux code is setting polarity configuration to 3 but uboot code is setting it to 1. Change the configuration to match the linux configuration

arch/arm/include/asm/arch-rockchip/edp_rk3288.h +3 -1
Index: u-boot/arch/arm/include/asm/arch-rockchip/edp_rk3288.h
===================================================================
@@ -297,7 +297,9 @@ check_member(rk3288_edp, pll_reg_5, 0xa0
 
 /* int_ctl */
 #define SOFT_INT_CTRL				(0x1 << 2)
-#define INT_POL					(0x1 << 0)
+#define INT_POL1				(0x1 << 1)
+#define INT_POL0				(0x1 << 0)
+#define INT_POL					(INT_POL0 | INT_POL1)
 
 /* sys_ctl_1 */
 #define DET_STA					(0x1 << 2)
#10 lore raw

PATCH v2 4/10 drivers/video/rockchip/rk_vop.c: Reserve efi fb memory

In reply to Arnaud Patard (Rtp)

When booting with EFI and graphics, the memory used for framebuffer has to be reserved, otherwise it may leads to kernel memory overwrite.

drivers/video/rockchip/rk_vop.c +9 -0
Index: u-boot/drivers/video/rockchip/rk_vop.c
===================================================================
@@ -20,6 +20,8 @@
 #include <asm/arch-rockchip/vop_rk3288.h>
 #include <dm/device-internal.h>
 #include <dm/uclass-internal.h>
+#include <efi.h>
+#include <efi_loader.h>
 #include <linux/bitops.h>
 #include <linux/err.h>
 #include <power/regulator.h>
@@ -394,6 +396,13 @@ int rk_vop_probe(struct udevice *dev)
 	if (!(gd->flags & GD_FLG_RELOC))
 		return 0;
 
+	plat->base = gd->bd->bi_dram[0].start + gd->bd->bi_dram[0].size - plat->size;
+
+#if defined(CONFIG_EFI_LOADER)
+	debug("Adding to EFI map %d @ %lx\n", plat->size, plat->base);
+	efi_add_memory_map(plat->base, plat->size, EFI_RESERVED_MEMORY_TYPE);
+#endif
+
 	priv->regs = (struct rk3288_vop *)dev_read_addr(dev);
 
 	/*
#12 lore raw

PATCH v2 5/10 rk3399-pinebook-pro-u-boot.dtsi: Enable edp

In reply to Arnaud Patard (Rtp)

- uboot rockchip edp code is looking for a rockchip,panel property for the edp dts node, so add it.

arch/arm/dts/rk3399-pinebook-pro-u-boot.dtsi +4 -0
Index: u-boot/arch/arm/dts/rk3399-pinebook-pro-u-boot.dtsi
===================================================================
@@ -49,3 +49,7 @@
 &vdd_log {
 	regulator-init-microvolt = <950000>;
 };
+
+&edp {
+	rockchip,panel = <&edp_panel>;
+};
#14 lore raw

PATCH v2 6/10 configs/pinebook-pro-rk3399_defconfig: enable SYS_USB_EVENT_POLL_VIA_INT_QUEUE

In reply to Arnaud Patard (Rtp)

The default configuration will use SYS_USB_EVENT_POLL for handling the usb keyboard and it makes the system really slow (eg slow keypress, loading kernel/initrd from grub-efi is taking ages).

Using CONFIG_SYS_USB_EVENT_POLL_VIA_INT_QUEUE seems to be improving things a lot, so use it.

configs/pinebook-pro-rk3399_defconfig +1 -0
Index: u-boot/configs/pinebook-pro-rk3399_defconfig
===================================================================
@@ -78,6 +78,7 @@ CONFIG_USB_OHCI_GENERIC=y
 CONFIG_USB_DWC3=y
 CONFIG_ROCKCHIP_USB2_PHY=y
 CONFIG_USB_KEYBOARD=y
+CONFIG_SYS_USB_EVENT_POLL_VIA_INT_QUEUE=y
 CONFIG_USB_HOST_ETHER=y
 CONFIG_USB_ETHER_ASIX=y
 CONFIG_USB_ETHER_RTL8152=y
#15 lore raw

PATCH v2 7/10 drivers/pwm/rk_pwm.c: Fix default polarity

In reply to Arnaud Patard (Rtp)

In the code, the default polarity is set to positive/positive, which is neither normal polarity or inverted polarity. It's only the hardware default. This leads to booting linux with wrong polarity setting.

Update the code to use PWM_DUTY_POSTIVE | PWM_INACTIVE_NEGATIVE by default instead.

drivers/pwm/rk_pwm.c +1 -1
Index: u-boot/drivers/pwm/rk_pwm.c
===================================================================
@@ -146,7 +146,7 @@ static int rk_pwm_probe(struct udevice *
 	priv->data = (struct rockchip_pwm_data *)dev_get_driver_data(dev);
 
 	if (priv->data->supports_polarity)
-		priv->conf_polarity = PWM_DUTY_POSTIVE | PWM_INACTIVE_POSTIVE;
+		priv->conf_polarity = PWM_DUTY_POSTIVE | PWM_INACTIVE_NEGATIVE;
 
 	return 0;
 }
#17 lore raw

PATCH v2 8/10 drivers/video/rockchip/rk_vop.c: Fix format of fbbase in debug string

In reply to Arnaud Patard (Rtp)

The debug string printing the device name, framebuffer address and of node is using %lu as format for the framebuffer address, which is not so nice. Change it to %lx.

drivers/video/rockchip/rk_vop.c +1 -1
Index: u-boot/drivers/video/rockchip/rk_vop.c
===================================================================
@@ -239,7 +239,7 @@ static int rk_display_init(struct udevic
 	ofnode remote;
 	const char *compat;
 
-	debug("%s(%s, %lu, %s)\n", __func__,
+	debug("%s(%s, %lx, %s)\n", __func__,
 	      dev_read_name(dev), fbbase, ofnode_get_name(ep_node));
 
 	ret = ofnode_read_u32(ep_node, "remote-endpoint", &remote_phandle);
#19 lore raw

PATCH v2 9/10 drivers/video/rockchip/rk_edp.c: Add missing reset support

In reply to Arnaud Patard (Rtp)

In order to ensure that the eDP registers are in correct state, add missing support for the eDP reset lines found in the device-tree.

drivers/video/rockchip/rk_edp.c +22 -0
Index: u-boot/drivers/video/rockchip/rk_edp.c
===================================================================
@@ -8,11 +8,13 @@
 #include <clk.h>
 #include <display.h>
 #include <dm.h>
+#include <dm/device_compat.h>
 #include <edid.h>
 #include <log.h>
 #include <malloc.h>
 #include <panel.h>
 #include <regmap.h>
+#include <reset.h>
 #include <syscon.h>
 #include <asm/gpio.h>
 #include <asm/io.h>
@@ -1047,6 +1049,7 @@ static int rk_edp_probe(struct udevice *
 	struct rk_edp_priv *priv = dev_get_priv(dev);
 	struct rk3288_edp *regs = priv->regs;
 	struct rockchip_dp_data *edp_data = (struct rockchip_dp_data *)dev_get_driver_data(dev);
+	struct reset_ctl dp_rst;
 
 	struct clk clk;
 	int ret;
@@ -1059,6 +1062,25 @@ static int rk_edp_probe(struct udevice *
 		return ret;
 	}
 
+	ret = reset_get_by_name(dev, "dp", &dp_rst);
+	if (ret) {
+		dev_err(dev, "failed to get dp reset (ret=%d)\n", ret);
+		return ret;
+	}
+
+	ret = reset_assert(&dp_rst);
+	if (ret) {
+		dev_err(dev, "failed to assert dp reset (ret=%d)\n", ret);
+		return ret;
+	}
+	udelay(20);
+
+	ret = reset_deassert(&dp_rst);
+	if (ret) {
+		dev_err(dev, "failed to deassert dp reset (ret=%d)\n", ret);
+		return ret;
+	}
+
 	int vop_id = uc_plat->source_id;
 	debug("%s, uc_plat=%p, vop_id=%u\n", __func__, uc_plat, vop_id);
 
#21 lore raw

PATCH v2 10/10 drivers/video/rockchip/rk_vop.c: Add reset support

In reply to Arnaud Patard (Rtp)

In order to ensure that the VOP registers are in correct state, add missing support for the VOP reset lines found in the device-tree

drivers/video/rockchip/rk_vop.c +47 -3
Index: u-boot/drivers/video/rockchip/rk_vop.c
===================================================================
@@ -8,9 +8,11 @@
 #include <clk.h>
 #include <display.h>
 #include <dm.h>
+#include <dm/device_compat.h>
 #include <edid.h>
 #include <log.h>
 #include <regmap.h>
+#include <reset.h>
 #include <syscon.h>
 #include <video.h>
 #include <asm/gpio.h>
@@ -36,14 +38,16 @@ enum vop_pol {
 	DCLK_INVERT    = 3
 };
 
-static void rkvop_enable(struct rk3288_vop *regs, ulong fbbase,
+static void rkvop_enable(struct udevice *dev, struct rk3288_vop *regs, ulong fbbase,
 			 int fb_bits_per_pixel,
-			 const struct display_timing *edid)
+			 const struct display_timing *edid,
+			 struct reset_ctl *dclk_rst)
 {
 	u32 lb_mode;
 	u32 rgb_mode;
 	u32 hactive = edid->hactive.typ;
 	u32 vactive = edid->vactive.typ;
+	int ret;
 
 	writel(V_ACT_WIDTH(hactive - 1) | V_ACT_HEIGHT(vactive - 1),
 	       &regs->win0_act_info);
@@ -91,6 +95,18 @@ static void rkvop_enable(struct rk3288_v
 
 	writel(fbbase, &regs->win0_yrgb_mst);
 	writel(0x01, &regs->reg_cfg_done); /* enable reg config */
+
+	ret = reset_assert(dclk_rst);
+	if (ret) {
+		dev_warn(dev, "failed to assert dclk reset (ret=%d)\n", ret);
+		return;
+	}
+	udelay(20);
+
+	ret = reset_deassert(dclk_rst);
+	if (ret)
+		dev_warn(dev, "failed to deassert dclk reset (ret=%d)\n", ret);
+
 }
 
 static void rkvop_set_pin_polarity(struct udevice *dev,
@@ -238,6 +254,7 @@ static int rk_display_init(struct udevic
 	enum video_log2_bpp l2bpp;
 	ofnode remote;
 	const char *compat;
+	struct reset_ctl dclk_rst;
 
 	debug("%s(%s, %lx, %s)\n", __func__,
 	      dev_read_name(dev), fbbase, ofnode_get_name(ep_node));
@@ -354,7 +371,14 @@ static int rk_display_init(struct udevic
 	}
 
 	rkvop_mode_set(dev, &timing, vop_id);
-	rkvop_enable(regs, fbbase, 1 << l2bpp, &timing);
+
+	ret = reset_get_by_name(dev, "dclk", &dclk_rst);
+	if (ret) {
+		dev_err(dev, "failed to get dclk reset (ret=%d)\n", ret);
+		return ret;
+	}
+
+	rkvop_enable(dev, regs, fbbase, 1 << l2bpp, &timing, &dclk_rst);
 
 	ret = display_enable(disp, 1 << l2bpp, &timing);
 	if (ret)
@@ -391,11 +415,31 @@ int rk_vop_probe(struct udevice *dev)
 	struct rk_vop_priv *priv = dev_get_priv(dev);
 	int ret = 0;
 	ofnode port, node;
+	struct reset_ctl ahb_rst;
 
 	/* Before relocation we don't need to do anything */
 	if (!(gd->flags & GD_FLG_RELOC))
 		return 0;
 
+	ret = reset_get_by_name(dev, "ahb", &ahb_rst);
+	if (ret) {
+		dev_err(dev, "failed to get ahb reset (ret=%d)\n", ret);
+		return ret;
+	}
+
+	ret = reset_assert(&ahb_rst);
+	if (ret) {
+		dev_err(dev, "failed to assert ahb reset (ret=%d)\n", ret);
+	return ret;
+	}
+	udelay(20);
+
+	ret = reset_deassert(&ahb_rst);
+	if (ret) {
+		dev_err(dev, "failed to deassert ahb reset (ret=%d)\n", ret);
+		return ret;
+	}
+
 	plat->base = gd->bd->bi_dram[0].start + gd->bd->bi_dram[0].size - plat->size;
 
 #if defined(CONFIG_EFI_LOADER)
#13 lore raw

PATCH v2 5/10 rk3399-pinebook-pro-u-boot.dtsi: Enable edp

In reply to Arnaud Patard (Rtp)
Quoted 10 lines · Arnaud Patard

- uboot rockchip edp code is looking for a rockchip,panel property for the edp dts node, so add it.

Index: u-boot/arch/arm/dts/rk3399-pinebook-pro-u-boot.dtsi
===================================================================

arch/arm/dts/rk3399-pinebook-pro-u-boot.dtsi line 55

Quoted 7 lines
arch/arm/dts/rk3399-pinebook-pro-u-boot.dtsi +4 -0
@@ -49,3 +49,7 @@
 &vdd_log {
        regulator-init-microvolt = <950000>;
 };
+
+&edp {
+       rockchip,panel = <&edp_panel>;
+};

This should be in alphabetical order.

#23 lore raw

v2 cover rk3399 (Pinebook pro) EDP support

In reply to Arnaud Patard (Rtp)
Quoted 8 lines · Arnaud Patard (Rtp)

This patchset add support for the rk3399 eDP. It has been tested on the pinebook pro and Google Kevin chromeos devices.

The changes have been written by studying the linux code, since I didn't find any manual for theses part of the RK3399 SoC.

For Kevin devices, this patchset is needed:
https://patchwork.ozlabs.org/project/uboot/list/?series=209553

(Arnaud's patchset does not depend on any of mine, they are AFAIK independently applicable)

That series is not enough though -- gru-kevin isn't even upstream. Here's my git branch I tested (with success) on it including ugly hacks and patches not sent/sendable to the mailing lists yet:

https://github.com/alpernebbi/u-boot/commits/rk3399-gru-kevin/wip

Here's another branch that I'm keeping up-to-date for gru-bob with analogous changes, but I cannot test it myself:

    https://github.com/alpernebbi/u-boot/commits/rk3399-gru-bob-edp/wip

Those contain this version of this series now, and should make this series testable by people having one of the two chromebooks.

Quoted 11 lines

On the linux kernel side, on recent kernels, it needs commit "pwm: rockchip: Keep enabled PWMs running while probing" otherwise the pinebook pro will freeze when probing the display.

Changes since v2:
- Add reset support for the VOP and eDP. This was needed and it appears to also solve the issue with warm reset.
- Fix a debug string in the vop driver
- Drop the patch "drivers/video/rockchip/rk_edp.c: Change clock rate".
- Address various comments

#2 lore raw

PATCH v2 1/10 drivers/video/rockchip/rk_vop.c: Use endpoint compatible string to find VOP mode

In reply to Arnaud Patard (Rtp)
Quoted 4 lines · Arnaud Patard (Rtp)

The current code is using an hard coded enum and the of node reg value of endpoint to find out if the endpoint is mipi/hdmi/lvds/edp/dp. The order is different between rk3288, rk3399 vop little, rk3399 vop big.

Quoted 8 lines

A possible solution would be to make sure that the rk3288.dtsi and rk3399.dtsi files have "expected" reg value or an other solution is to find the kind of endpoint by comparing the endpoint compatible value.

This patch is implementing the more flexible second solution.

I didn't test individual patches, but non-pinebook-pro parts of this series as a whole (after some more patches of mine) get me a vidconsole on the rk3399-gru-kevin's display, so I'm going to send "Tested-by"s to the patches I had included in my branch:

#7 lore raw

PATCH v2 2/10 drivers/video/rockchip/rk_edp.c: Add rk3399 support

In reply to Arnaud Patard (Rtp)
Quoted 11 lines · Arnaud Patard (Rtp)

According to linux commit "drm/rockchip: analogix_dp: add rk3399 eDP support" (82872e42bb1501dd9e60ca430f4bae45a469aa64), rk3288 and rk3399 eDP IPs are nearly the same, the difference is in the grf register (SOC_CON6 versus SOC_CON20). So, change the code to use the right register on each IP.

The clocks don't seem to be the same, the eDP clock is not at index 1 on rk3399, so don't try changing the clock at index 1 to rate 0 on rk3399.

On a rk3399-gru-kevin, (see comment in patch 1):

#16 lore raw

PATCH v2 7/10 drivers/pwm/rk_pwm.c: Fix default polarity

In reply to Arnaud Patard (Rtp)
Quoted 9 lines · Arnaud Patard (Rtp)

In the code, the default polarity is set to positive/positive, which is neither normal polarity or inverted polarity. It's only the hardware default. This leads to booting linux with wrong polarity setting.

Update the code to use PWM_DUTY_POSTIVE | PWM_INACTIVE_NEGATIVE by default instead.

On a rk3399-gru-kevin, (see comment in patch 1):

#18 lore raw

PATCH v2 8/10 drivers/video/rockchip/rk_vop.c: Fix format of fbbase in debug string

In reply to Arnaud Patard (Rtp)
Quoted 5 lines · Arnaud Patard (Rtp)

The debug string printing the device name, framebuffer address and of node is using %lu as format for the framebuffer address, which is not so nice. Change it to %lx.

On a rk3399-gru-kevin, (see comment in patch 1):

#3 lore raw

PATCH v2 1/10 drivers/video/rockchip/rk_vop.c: Use endpoint compatible string to find VOP mode

In reply to Arnaud Patard (Rtp)

Hi Arnaud,

??? Thanks for your patch.

??? Please use module name as subject prefix instead of a file name,
eg. "rockchip: video: vop".

arch/arm/include/asm/arch-rockchip/vop_rk3288.h

Quoted 16 lines · Arnaud Patard (Rtp)

The current code is using an hard coded enum and the of node reg value of endpoint to find out if the endpoint is mipi/hdmi/lvds/edp/dp. The order is different between rk3288, rk3399 vop little, rk3399 vop big.

A possible solution would be to make sure that the rk3288.dtsi and rk3399.dtsi files have "expected" reg value or an other solution is to find the kind of endpoint by comparing the endpoint compatible value.

This patch is implementing the more flexible second solution.

Index: u-boot/arch/arm/include/asm/arch-rockchip/vop_rk3288.h
===================================================================

arch/arm/include/asm/arch-rockchip/vop_rk3288.h +0 -0

For the patch format, please make sure the u-boot is the root directory.

Thanks,

- Kever

Quoted 79 lines
arch/arm/include/asm/arch-rockchip/vop_rk3288.h +1 -14
@@ -85,26 +85,13 @@ enum {
  	LB_RGB_1280X8 = 0x5
  };
 
-#if defined(CONFIG_ROCKCHIP_RK3399)
  enum vop_modes {
  	VOP_MODE_EDP = 0,
  	VOP_MODE_MIPI,
  	VOP_MODE_HDMI,
-	VOP_MODE_MIPI1,
-	VOP_MODE_DP,
-	VOP_MODE_NONE,
-};
-#else
-enum vop_modes {
-	VOP_MODE_EDP = 0,
-	VOP_MODE_HDMI,
  	VOP_MODE_LVDS,
-	VOP_MODE_MIPI,
-	VOP_MODE_NONE,
-	VOP_MODE_AUTO_DETECT,
-	VOP_MODE_UNKNOWN,
+	VOP_MODE_DP,
  };
-#endif
 
  /* VOP_VERSION_INFO */
  #define M_FPGA_VERSION (0xffff << 16)

Index: u-boot/drivers/video/rockchip/rk_vop.c
===================================================================

drivers/video/rockchip/rk_vop.c +23 -2
@@ -235,12 +235,11 @@ static int rk_display_init(struct udevic
  	struct clk clk;
  	enum video_log2_bpp l2bpp;
  	ofnode remote;
+	const char *compat;
 
  	debug("%s(%s, %lu, %s)\n", __func__,
  	      dev_read_name(dev), fbbase, ofnode_get_name(ep_node));
 
-	vop_id = ofnode_read_s32_default(ep_node, "reg", -1);
-	debug("vop_id=%d\n", vop_id);
  	ret = ofnode_read_u32(ep_node, "remote-endpoint", &remote_phandle);
  	if (ret)
  		return ret;
@@ -282,6 +281,28 @@ static int rk_display_init(struct udevic
  		if (disp)
  			break;
  	};
+	compat = ofnode_get_property(remote, "compatible", NULL);
+	if (!compat) {
+		debug("%s(%s): Failed to find compatible property\n",
+		      __func__, dev_read_name(dev));
+		return -EINVAL;
+	}
+	if (strstr(compat, "edp")) {
+		vop_id = VOP_MODE_EDP;
+	} else if (strstr(compat, "mipi")) {
+		vop_id = VOP_MODE_MIPI;
+	} else if (strstr(compat, "hdmi")) {
+		vop_id = VOP_MODE_HDMI;
+	} else if (strstr(compat, "cdn-dp")) {
+		vop_id = VOP_MODE_DP;
+	} else if (strstr(compat, "lvds")) {
+		vop_id = VOP_MODE_LVDS;
+	} else {
+		debug("%s(%s): Failed to find vop mode for %s\n",
+		      __func__, dev_read_name(dev), compat);
+		return -EINVAL;
+	}
+	debug("vop_id=%d\n", vop_id);
 
  	disp_uc_plat = dev_get_uclass_platdata(disp);
  	debug("Found device '%s', disp_uc_priv=%p\n", disp->name, disp_uc_plat);
#4 lore raw

PATCH v2 1/10 drivers/video/rockchip/rk_vop.c: Use endpoint compatible string to find VOP mode

Quoted 7 lines · Kever Yang

Hi Arnaud,

??? Thanks for your patch.

??? Please use module name as subject prefix instead of a file name, eg. "rockchip: video: vop".

ok. will fix.

Quoted 11 lines · Arnaud Patard (Rtp)
Quoted 10 lines

The current code is using an hard coded enum and the of node reg value of endpoint to find out if the endpoint is mipi/hdmi/lvds/edp/dp. The order is different between rk3288, rk3399 vop little, rk3399 vop big.

A possible solution would be to make sure that the rk3288.dtsi and rk3399.dtsi files have "expected" reg value or an other solution is to find the kind of endpoint by comparing the endpoint compatible value.

This patch is implementing the more flexible second solution.

Quoted 9 lines

arch/arm/include/asm/arch-rockchip/vop_rk3288.h

Quoted 6 lines

Index: u-boot/arch/arm/include/asm/arch-rockchip/vop_rk3288.h
===================================================================

arch/arm/include/asm/arch-rockchip/vop_rk3288.h +0 -0

For the patch format, please make sure the u-boot is the root directory.

Can you please explain this point ? It's a valid patch which can be applied with patch -p1. I've been generating patches with quilt for ages and patches got merged in various projects including in the kernel even before git, so I fail to get what's wrong with it.
Or are you telling me that uboot is accepting only patches generated with git ?

Arnaud

#5 lore raw

PATCH v2 1/10 drivers/video/rockchip/rk_vop.c: Use endpoint compatible string to find VOP mode

Quoted 30 lines · Arnaud Patard (Rtp)
Quoted 7 lines · Kever Yang

Hi Arnaud,

??? Thanks for your patch.

??? Please use module name as subject prefix instead of a file name,
eg. "rockchip: video: vop".

ok. will fix.

Quoted 11 lines · Arnaud Patard (Rtp)
Quoted 10 lines

The current code is using an hard coded enum and the of node reg value of endpoint to find out if the endpoint is mipi/hdmi/lvds/edp/dp. The order is different between rk3288, rk3399 vop little, rk3399 vop big.

A possible solution would be to make sure that the rk3288.dtsi and rk3399.dtsi files have "expected" reg value or an other solution is to find the kind of endpoint by comparing the endpoint compatible value.

This patch is implementing the more flexible second solution.

Quoted 8 lines

arch/arm/include/asm/arch-rockchip/vop_rk3288.h

Quoted 6 lines

Index: u-boot/arch/arm/include/asm/arch-rockchip/vop_rk3288.h
===================================================================

arch/arm/include/asm/arch-rockchip/vop_rk3288.h +0 -0

For the patch format, please make sure the u-boot is the root directory.

Quoted 5 lines

Can you please explain this point ? It's a valid patch which can be applied with patch -p1. I've been generating patches with quilt for ages and patches got merged in various projects including in the kernel even before git, so I fail to get what's wrong with it.
Or are you telling me that uboot is accepting only patches generated with git ?

Sorry, I though this patch may not able to apply with git am, but after try with it,

this patch can be apply directly, them I think you don't need to update this.

Thanks,

- Kever

Quoted 4 lines

Arnaud