mmc: sdhci-of-arasan: Add support for eMMC5.1 on Xilinx Versal Net platform

Add support for eMMC5.1 on Xilinx Versal Net platform
- Add new compatible string(xlnx,versal-net-emmc).
- Add support for PHY which is part of Host Controller register space.
- Add DLL and Delay Chain mode support and corresponding tap delays for
all eMMC modes.
- Add Strobe select tap for HS400 mode.

Signed-off-by: Swati Agarwal <swati.agarwal@amd.com>
Co-developed-by: Sai Krishna Potthuri <sai.krishna.potthuri@amd.com>
Signed-off-by: Sai Krishna Potthuri <sai.krishna.potthuri@amd.com>
Acked-by: Adrian Hunter <adrian.hunter@intel.com>
Link: https://lore.kernel.org/r/20230403102551.3763054-3-sai.krishna.potthuri@amd.com
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>

Author: Swati Agarwal

drivers/mmc/host/sdhci-of-arasan.c

@@ -41,18 +41,53 @@
 #define VENDOR_ENHANCED_STROBE		BIT(0)
 
 #define PHY_CLK_TOO_SLOW_HZ		400000
+#define MIN_PHY_CLK_HZ			50000000
 
 #define SDHCI_ITAPDLY_CHGWIN		0x200
 #define SDHCI_ITAPDLY_ENABLE		0x100
 #define SDHCI_OTAPDLY_ENABLE		0x40
 
+#define PHY_CTRL_REG1			0x270
+#define PHY_CTRL_ITAPDLY_ENA_MASK	BIT(0)
+#define PHY_CTRL_ITAPDLY_SEL_MASK	GENMASK(5, 1)
+#define PHY_CTRL_ITAPDLY_SEL_SHIFT	1
+#define PHY_CTRL_ITAP_CHG_WIN_MASK	BIT(6)
+#define PHY_CTRL_OTAPDLY_ENA_MASK	BIT(8)
+#define PHY_CTRL_OTAPDLY_SEL_MASK	GENMASK(15, 12)
+#define PHY_CTRL_OTAPDLY_SEL_SHIFT	12
+#define PHY_CTRL_STRB_SEL_MASK		GENMASK(23, 16)
+#define PHY_CTRL_STRB_SEL_SHIFT		16
+#define PHY_CTRL_TEST_CTRL_MASK		GENMASK(31, 24)
+
+#define PHY_CTRL_REG2			0x274
+#define PHY_CTRL_EN_DLL_MASK		BIT(0)
+#define PHY_CTRL_DLL_RDY_MASK		BIT(1)
+#define PHY_CTRL_FREQ_SEL_MASK		GENMASK(6, 4)
+#define PHY_CTRL_FREQ_SEL_SHIFT		4
+#define PHY_CTRL_SEL_DLY_TX_MASK	BIT(16)
+#define PHY_CTRL_SEL_DLY_RX_MASK	BIT(17)
+#define FREQSEL_200M_170M		0x0
+#define FREQSEL_170M_140M	        0x1
+#define FREQSEL_140M_110M	        0x2
+#define FREQSEL_110M_80M	        0x3
+#define FREQSEL_80M_50M			0x4
+#define FREQSEL_275M_250M		0x5
+#define FREQSEL_250M_225M		0x6
+#define FREQSEL_225M_200M		0x7
+#define PHY_DLL_TIMEOUT_MS		100
+
 /* Default settings for ZynqMP Clock Phases */
 #define ZYNQMP_ICLK_PHASE {0, 63, 63, 0, 63,  0,   0, 183, 54,  0, 0}
 #define ZYNQMP_OCLK_PHASE {0, 72, 60, 0, 60, 72, 135, 48, 72, 135, 0}
 
 #define VERSAL_ICLK_PHASE {0, 132, 132, 0, 132, 0, 0, 162, 90, 0, 0}
 #define VERSAL_OCLK_PHASE {0,  60, 48, 0, 48, 72, 90, 36, 60, 90, 0}
 
+#define VERSAL_NET_EMMC_ICLK_PHASE {0, 0, 0, 0, 0, 0, 0, 0, 39, 0, 0}
+#define VERSAL_NET_EMMC_OCLK_PHASE {0, 113, 0, 0, 0, 0, 0, 0, 113, 79, 45}
+
+#define VERSAL_NET_PHY_CTRL_STRB90_STRB180_VAL		0X77
+
 /*
  * On some SoCs the syscon area has a feature where the upper 16-bits of
  * each 32-bit register act as a write mask for the lower 16-bits.  This allows
@@ -135,6 +170,7 @@ struct sdhci_arasan_clk_data {
  * @clk_ahb:		Pointer to the AHB clock
  * @phy:		Pointer to the generic phy
  * @is_phy_on:		True if the PHY is on; false if not.
+ * @internal_phy_reg:	True if the PHY is within the Host controller.
  * @has_cqe:		True if controller has command queuing engine.
  * @clk_data:		Struct for the Arasan Controller Clock Data.
  * @clk_ops:		Struct for the Arasan Controller Clock Operations.
@@ -147,6 +183,7 @@ struct sdhci_arasan_data {
 	struct clk	*clk_ahb;
 	struct phy	*phy;
 	bool		is_phy_on;
+	bool		internal_phy_reg;
 
 	bool		has_cqe;
 	struct sdhci_arasan_clk_data clk_data;
@@ -200,6 +237,61 @@ static const struct sdhci_arasan_soc_ctl_map intel_keembay_soc_ctl_map = {
 	.hiword_update = false,
 };
 
+static void sdhci_arasan_phy_set_delaychain(struct sdhci_host *host, bool enable)
+{
+	u32 reg;
+
+	reg = readl(host->ioaddr + PHY_CTRL_REG2);
+	if (enable)
+		reg |= (PHY_CTRL_SEL_DLY_TX_MASK | PHY_CTRL_SEL_DLY_RX_MASK);
+	else
+		reg &= ~(PHY_CTRL_SEL_DLY_TX_MASK | PHY_CTRL_SEL_DLY_RX_MASK);
+
+	writel(reg, host->ioaddr + PHY_CTRL_REG2);
+}
+
+static int sdhci_arasan_phy_set_dll(struct sdhci_host *host, bool enable)
+{
+	u32 reg;
+
+	reg = readl(host->ioaddr + PHY_CTRL_REG2);
+	if (enable)
+		reg |= PHY_CTRL_EN_DLL_MASK;
+	else
+		reg &= ~PHY_CTRL_EN_DLL_MASK;
+
+	writel(reg, host->ioaddr + PHY_CTRL_REG2);
+
+	if (!enable)
+		return 0;
+
+	return readl_relaxed_poll_timeout(host->ioaddr + PHY_CTRL_REG2, reg,
+					  (reg & PHY_CTRL_DLL_RDY_MASK), 10,
+					  1000 * PHY_DLL_TIMEOUT_MS);
+}
+
+static void sdhci_arasan_phy_dll_set_freq(struct sdhci_host *host, int clock)
+{
+	u32 reg, freq_sel, freq;
+
+	freq = DIV_ROUND_CLOSEST(clock, 1000000);
+	if (freq <= 200 && freq > 170)
+		freq_sel = FREQSEL_200M_170M;
+	else if (freq <= 170 && freq > 140)
+		freq_sel = FREQSEL_170M_140M;
+	else if (freq <= 140 && freq > 110)
+		freq_sel = FREQSEL_140M_110M;
+	else if (freq <= 110 && freq > 80)
+		freq_sel = FREQSEL_110M_80M;
+	else
+		freq_sel = FREQSEL_80M_50M;
+
+	reg = readl(host->ioaddr + PHY_CTRL_REG2);
+	reg &= ~PHY_CTRL_FREQ_SEL_MASK;
+	reg |= (freq_sel << PHY_CTRL_FREQ_SEL_SHIFT);
+	writel(reg, host->ioaddr + PHY_CTRL_REG2);
+}
+
 /**
  * sdhci_arasan_syscon_write - Write to a field in soc_ctl registers
  *
@@ -317,8 +409,21 @@ static void sdhci_arasan_set_clock(struct sdhci_host *host, unsigned int clock)
 	if (clk_data->set_clk_delays)
 		clk_data->set_clk_delays(host);
 
+	if (sdhci_arasan->internal_phy_reg && clock >= MIN_PHY_CLK_HZ) {
+		sdhci_writew(host, 0, SDHCI_CLOCK_CONTROL);
+		sdhci_arasan_phy_set_dll(host, 0);
+		sdhci_arasan_phy_set_delaychain(host, 0);
+		sdhci_arasan_phy_dll_set_freq(host, clock);
+	} else if (sdhci_arasan->internal_phy_reg) {
+		sdhci_writew(host, 0, SDHCI_CLOCK_CONTROL);
+		sdhci_arasan_phy_set_delaychain(host, 1);
+	}
+
 	sdhci_set_clock(host, clock);
 
+	if (sdhci_arasan->internal_phy_reg && clock >= MIN_PHY_CLK_HZ)
+		sdhci_arasan_phy_set_dll(host, 1);
+
 	if (sdhci_arasan->quirks & SDHCI_ARASAN_QUIRK_CLOCK_UNSTABLE)
 		/*
 		 * Some controllers immediately report SDHCI_CLOCK_INT_STABLE
@@ -881,6 +986,101 @@ static const struct clk_ops versal_sampleclk_ops = {
 	.set_phase = sdhci_versal_sampleclk_set_phase,
 };
 
+static int sdhci_versal_net_emmc_sdcardclk_set_phase(struct clk_hw *hw, int degrees)
+{
+	struct sdhci_arasan_clk_data *clk_data =
+		container_of(hw, struct sdhci_arasan_clk_data, sdcardclk_hw);
+	struct sdhci_arasan_data *sdhci_arasan =
+		container_of(clk_data, struct sdhci_arasan_data, clk_data);
+	struct sdhci_host *host = sdhci_arasan->host;
+	u8 tap_delay, tap_max = 0;
+
+	switch (host->timing) {
+	case MMC_TIMING_MMC_HS:
+	case MMC_TIMING_MMC_DDR52:
+		tap_max = 16;
+		break;
+	case MMC_TIMING_MMC_HS200:
+	case MMC_TIMING_MMC_HS400:
+		 /* For 200MHz clock, 32 Taps are available */
+		tap_max = 32;
+		break;
+	default:
+		break;
+	}
+
+	tap_delay = (degrees * tap_max) / 360;
+
+	/* Set the Clock Phase */
+	if (tap_delay) {
+		u32 regval;
+
+		regval = sdhci_readl(host, PHY_CTRL_REG1);
+		regval |= PHY_CTRL_OTAPDLY_ENA_MASK;
+		sdhci_writel(host, regval, PHY_CTRL_REG1);
+		regval &= ~PHY_CTRL_OTAPDLY_SEL_MASK;
+		regval |= tap_delay << PHY_CTRL_OTAPDLY_SEL_SHIFT;
+		sdhci_writel(host, regval, PHY_CTRL_REG1);
+	}
+
+	return 0;
+}
+
+static const struct clk_ops versal_net_sdcardclk_ops = {
+	.recalc_rate = sdhci_arasan_sdcardclk_recalc_rate,
+	.set_phase = sdhci_versal_net_emmc_sdcardclk_set_phase,
+};
+
+static int sdhci_versal_net_emmc_sampleclk_set_phase(struct clk_hw *hw, int degrees)
+{
+	struct sdhci_arasan_clk_data *clk_data =
+		container_of(hw, struct sdhci_arasan_clk_data, sampleclk_hw);
+	struct sdhci_arasan_data *sdhci_arasan =
+		container_of(clk_data, struct sdhci_arasan_data, clk_data);
+	struct sdhci_host *host = sdhci_arasan->host;
+	u8 tap_delay, tap_max = 0;
+	u32 regval;
+
+	switch (host->timing) {
+	case MMC_TIMING_MMC_HS:
+	case MMC_TIMING_MMC_DDR52:
+		tap_max = 32;
+		break;
+	case MMC_TIMING_MMC_HS400:
+		/* Strobe select tap point for strb90 and strb180 */
+		regval = sdhci_readl(host, PHY_CTRL_REG1);
+		regval &= ~PHY_CTRL_STRB_SEL_MASK;
+		regval |= VERSAL_NET_PHY_CTRL_STRB90_STRB180_VAL << PHY_CTRL_STRB_SEL_SHIFT;
+		sdhci_writel(host, regval, PHY_CTRL_REG1);
+		break;
+	default:
+		break;
+	}
+
+	tap_delay = (degrees * tap_max) / 360;
+
+	/* Set the Clock Phase */
+	if (tap_delay) {
+		regval = sdhci_readl(host, PHY_CTRL_REG1);
+		regval |= PHY_CTRL_ITAP_CHG_WIN_MASK;
+		sdhci_writel(host, regval, PHY_CTRL_REG1);
+		regval |= PHY_CTRL_ITAPDLY_ENA_MASK;
+		sdhci_writel(host, regval, PHY_CTRL_REG1);
+		regval &= ~PHY_CTRL_ITAPDLY_SEL_MASK;
+		regval |= tap_delay << PHY_CTRL_ITAPDLY_SEL_SHIFT;
+		sdhci_writel(host, regval, PHY_CTRL_REG1);
+		regval &= ~PHY_CTRL_ITAP_CHG_WIN_MASK;
+		sdhci_writel(host, regval, PHY_CTRL_REG1);
+	}
+
+	return 0;
+}
+
+static const struct clk_ops versal_net_sampleclk_ops = {
+	.recalc_rate = sdhci_arasan_sampleclk_recalc_rate,
+	.set_phase = sdhci_versal_net_emmc_sampleclk_set_phase,
+};
+
 static void arasan_zynqmp_dll_reset(struct sdhci_host *host, u32 deviceid)
 {
 	u16 clk;
@@ -1091,7 +1291,17 @@ static void arasan_dt_parse_clk_phases(struct device *dev,
 			clk_data->clk_phase_out[i] = versal_oclk_phase[i];
 		}
 	}
+	if (of_device_is_compatible(dev->of_node, "xlnx,versal-net-emmc")) {
+		u32 versal_net_iclk_phase[MMC_TIMING_MMC_HS400 + 1] =
+			VERSAL_NET_EMMC_ICLK_PHASE;
+		u32 versal_net_oclk_phase[MMC_TIMING_MMC_HS400 + 1] =
+			VERSAL_NET_EMMC_OCLK_PHASE;
 
+		for (i = 0; i <= MMC_TIMING_MMC_HS400; i++) {
+			clk_data->clk_phase_in[i] = versal_net_iclk_phase[i];
+			clk_data->clk_phase_out[i] = versal_net_oclk_phase[i];
+		}
+	}
 	arasan_dt_read_clk_phase(dev, clk_data, MMC_TIMING_LEGACY,
 				 "clk-phase-legacy");
 	arasan_dt_read_clk_phase(dev, clk_data, MMC_TIMING_MMC_HS,
@@ -1203,6 +1413,14 @@ static const struct sdhci_pltfm_data sdhci_arasan_zynqmp_pdata = {
 			SDHCI_QUIRK2_STOP_WITH_TC,
 };
 
+static const struct sdhci_pltfm_data sdhci_arasan_versal_net_pdata = {
+	.ops = &sdhci_arasan_ops,
+	.quirks2 = SDHCI_QUIRK2_PRESET_VALUE_BROKEN |
+			SDHCI_QUIRK2_CLOCK_DIV_ZERO_BROKEN |
+			SDHCI_QUIRK2_STOP_WITH_TC |
+			SDHCI_QUIRK2_CAPS_BIT63_FOR_HS400,
+};
+
 static const struct sdhci_arasan_clk_ops zynqmp_clk_ops = {
 	.sdcardclk_ops = &zynqmp_sdcardclk_ops,
 	.sampleclk_ops = &zynqmp_sampleclk_ops,
@@ -1223,6 +1441,16 @@ static struct sdhci_arasan_of_data sdhci_arasan_versal_data = {
 	.clk_ops = &versal_clk_ops,
 };
 
+static const struct sdhci_arasan_clk_ops versal_net_clk_ops = {
+	.sdcardclk_ops = &versal_net_sdcardclk_ops,
+	.sampleclk_ops = &versal_net_sampleclk_ops,
+};
+
+static struct sdhci_arasan_of_data sdhci_arasan_versal_net_data = {
+	.pdata = &sdhci_arasan_versal_net_pdata,
+	.clk_ops = &versal_net_clk_ops,
+};
+
 static struct sdhci_arasan_of_data intel_keembay_emmc_data = {
 	.soc_ctl_map = &intel_keembay_soc_ctl_map,
 	.pdata = &sdhci_keembay_emmc_pdata,
@@ -1288,6 +1516,10 @@ static const struct of_device_id sdhci_arasan_of_match[] = {
 		.compatible = "xlnx,versal-8.9a",
 		.data = &sdhci_arasan_versal_data,
 	},
+	{
+		.compatible = "xlnx,versal-net-emmc",
+		.data = &sdhci_arasan_versal_net_data,
+	},
 	{ /* sentinel */ }
 };
 MODULE_DEVICE_TABLE(of, sdhci_arasan_of_match);
@@ -1761,6 +1993,9 @@ static int sdhci_arasan_probe(struct platform_device *pdev)
 			host->mmc->caps2 |= MMC_CAP2_CQE_DCMD;
 	}
 
+	if (of_device_is_compatible(np, "xlnx,versal-net-emmc"))
+		sdhci_arasan->internal_phy_reg = true;
+
 	ret = sdhci_arasan_add_host(sdhci_arasan);
 	if (ret)
 		goto err_add_host;