spi: spi-zyqnmp-gqspi: Add tap delay and Versal platform support

Merge series from Amit Kumar Mahapatra <amit.kumar-mahapatra@amd.com>:

A bunch of improvements to the driver:

- Fix kernel-doc warnings in GQSPI driver.
- Avoid setting CPOL, CPHA & baud rate multiple times.
- Add Versal platform support in GQSPI driver.
- Add tap delay support in GQSPI driver.

Documentation/devicetree/bindings/spi/spi-zynqmp-qspi.yaml

@@ -14,7 +14,9 @@ allOf:
 
 properties:
   compatible:
-    const: xlnx,zynqmp-qspi-1.0
+    enum:
+      - xlnx,versal-qspi-1.0
+      - xlnx,zynqmp-qspi-1.0
 
   reg:
     maxItems: 2

drivers/firmware/xilinx/zynqmp.c

@@ -843,6 +843,13 @@ int zynqmp_pm_read_pggs(u32 index, u32 *value)
 }
 EXPORT_SYMBOL_GPL(zynqmp_pm_read_pggs);
 
+int zynqmp_pm_set_tapdelay_bypass(u32 index, u32 value)
+{
+	return zynqmp_pm_invoke_fn(PM_IOCTL, 0, IOCTL_SET_TAPDELAY_BYPASS,
+				   index, value, NULL);
+}
+EXPORT_SYMBOL_GPL(zynqmp_pm_set_tapdelay_bypass);
+
 /**
  * zynqmp_pm_set_boot_health_status() - PM API for setting healthy boot status
  * @value:	Status value to be written

drivers/spi/spi-zynqmp-gqspi.c

@@ -16,6 +16,7 @@
 #include <linux/module.h>
 #include <linux/of_irq.h>
 #include <linux/of_address.h>
+#include <linux/of_device.h>
 #include <linux/platform_device.h>
 #include <linux/pm_runtime.h>
 #include <linux/spi/spi.h>
@@ -34,6 +35,7 @@
 #define GQSPI_RXD_OFST			0x00000120
 #define GQSPI_TX_THRESHOLD_OFST		0x00000128
 #define GQSPI_RX_THRESHOLD_OFST		0x0000012C
+#define IOU_TAPDLY_BYPASS_OFST		0x0000003C
 #define GQSPI_LPBK_DLY_ADJ_OFST		0x00000138
 #define GQSPI_GEN_FIFO_OFST		0x00000140
 #define GQSPI_SEL_OFST			0x00000144
@@ -48,6 +50,7 @@
 #define GQSPI_QSPIDMA_DST_I_MASK_OFST	0x00000820
 #define GQSPI_QSPIDMA_DST_ADDR_OFST	0x00000800
 #define GQSPI_QSPIDMA_DST_ADDR_MSB_OFST 0x00000828
+#define GQSPI_DATA_DLY_ADJ_OFST         0x000001F8
 
 /* GQSPI register bit masks */
 #define GQSPI_SEL_MASK				0x00000001
@@ -136,11 +139,37 @@
 
 #define GQSPI_MAX_NUM_CS	2	/* Maximum number of chip selects */
 
+#define GQSPI_USE_DATA_DLY		0x1
+#define GQSPI_USE_DATA_DLY_SHIFT	31
+#define GQSPI_DATA_DLY_ADJ_VALUE	0x2
+#define GQSPI_DATA_DLY_ADJ_SHIFT	28
+#define GQSPI_LPBK_DLY_ADJ_DLY_1	0x1
+#define GQSPI_LPBK_DLY_ADJ_DLY_1_SHIFT	0x3
+#define TAP_DLY_BYPASS_LQSPI_RX_VALUE	0x1
+#define TAP_DLY_BYPASS_LQSPI_RX_SHIFT	0x2
+
+/* set to differentiate versal from zynqmp, 1=versal, 0=zynqmp */
+#define QSPI_QUIRK_HAS_TAPDELAY		BIT(0)
+
+#define GQSPI_FREQ_37_5MHZ	37500000
+#define GQSPI_FREQ_40MHZ	40000000
+#define GQSPI_FREQ_100MHZ	100000000
+#define GQSPI_FREQ_150MHZ	150000000
+
 #define SPI_AUTOSUSPEND_TIMEOUT		3000
 enum mode_type {GQSPI_MODE_IO, GQSPI_MODE_DMA};
 
+/**
+ * struct qspi_platform_data - zynqmp qspi platform data structure
+ * @quirks:    Flags is used to identify the platform
+ */
+struct qspi_platform_data {
+	u32 quirks;
+};
+
 /**
  * struct zynqmp_qspi - Defines qspi driver instance
+ * @ctlr:		Pointer to the spi controller information
  * @regs:		Virtual address of the QSPI controller registers
  * @refclk:		Pointer to the peripheral clock
  * @pclk:		Pointer to the APB clock
@@ -157,6 +186,9 @@ enum mode_type {GQSPI_MODE_IO, GQSPI_MODE_DMA};
  * @genfifoentry:	Used for storing the genfifoentry instruction.
  * @mode:		Defines the mode in which QSPI is operating
  * @data_completion:	completion structure
+ * @op_lock:		Operational lock
+ * @speed_hz:          Current SPI bus clock speed in hz
+ * @has_tapdelay:	Used for tapdelay register available in qspi
  */
 struct zynqmp_qspi {
 	struct spi_controller *ctlr;
@@ -177,6 +209,8 @@ struct zynqmp_qspi {
 	enum mode_type mode;
 	struct completion data_completion;
 	struct mutex op_lock;
+	u32 speed_hz;
+	bool has_tapdelay;
 };
 
 /**
@@ -249,6 +283,56 @@ static void zynqmp_gqspi_selectslave(struct zynqmp_qspi *instanceptr,
 	}
 }
 
+/**
+ * zynqmp_qspi_set_tapdelay:   To configure qspi tap delays
+ * @xqspi:             Pointer to the zynqmp_qspi structure
+ * @baudrateval:       Buadrate to configure
+ */
+static void zynqmp_qspi_set_tapdelay(struct zynqmp_qspi *xqspi, u32 baudrateval)
+{
+	u32 tapdlybypass = 0, lpbkdlyadj = 0, datadlyadj = 0, clk_rate;
+	u32 reqhz = 0;
+
+	clk_rate = clk_get_rate(xqspi->refclk);
+	reqhz = (clk_rate / (GQSPI_BAUD_DIV_SHIFT << baudrateval));
+
+	if (!xqspi->has_tapdelay) {
+		if (reqhz <= GQSPI_FREQ_40MHZ) {
+			zynqmp_pm_set_tapdelay_bypass(PM_TAPDELAY_QSPI,
+						      PM_TAPDELAY_BYPASS_ENABLE);
+		} else if (reqhz <= GQSPI_FREQ_100MHZ) {
+			zynqmp_pm_set_tapdelay_bypass(PM_TAPDELAY_QSPI,
+						      PM_TAPDELAY_BYPASS_ENABLE);
+			lpbkdlyadj |= (GQSPI_LPBK_DLY_ADJ_USE_LPBK_MASK);
+			datadlyadj |= ((GQSPI_USE_DATA_DLY <<
+					GQSPI_USE_DATA_DLY_SHIFT)
+					| (GQSPI_DATA_DLY_ADJ_VALUE <<
+						GQSPI_DATA_DLY_ADJ_SHIFT));
+		} else if (reqhz <= GQSPI_FREQ_150MHZ) {
+			lpbkdlyadj |= GQSPI_LPBK_DLY_ADJ_USE_LPBK_MASK;
+		}
+	} else {
+		if (reqhz <= GQSPI_FREQ_37_5MHZ) {
+			tapdlybypass |= (TAP_DLY_BYPASS_LQSPI_RX_VALUE <<
+					TAP_DLY_BYPASS_LQSPI_RX_SHIFT);
+		} else if (reqhz <= GQSPI_FREQ_100MHZ) {
+			tapdlybypass |= (TAP_DLY_BYPASS_LQSPI_RX_VALUE <<
+					TAP_DLY_BYPASS_LQSPI_RX_SHIFT);
+			lpbkdlyadj |= (GQSPI_LPBK_DLY_ADJ_USE_LPBK_MASK);
+			datadlyadj |= (GQSPI_USE_DATA_DLY <<
+					GQSPI_USE_DATA_DLY_SHIFT);
+		} else if (reqhz <= GQSPI_FREQ_150MHZ) {
+			lpbkdlyadj |= (GQSPI_LPBK_DLY_ADJ_USE_LPBK_MASK
+				       | (GQSPI_LPBK_DLY_ADJ_DLY_1 <<
+					       GQSPI_LPBK_DLY_ADJ_DLY_1_SHIFT));
+		}
+		zynqmp_gqspi_write(xqspi,
+				   IOU_TAPDLY_BYPASS_OFST, tapdlybypass);
+	}
+	zynqmp_gqspi_write(xqspi, GQSPI_LPBK_DLY_ADJ_OFST, lpbkdlyadj);
+	zynqmp_gqspi_write(xqspi, GQSPI_DATA_DLY_ADJ_OFST, datadlyadj);
+}
+
 /**
  * zynqmp_qspi_init_hw - Initialize the hardware
  * @xqspi:	Pointer to the zynqmp_qspi structure
@@ -264,12 +348,15 @@ static void zynqmp_gqspi_selectslave(struct zynqmp_qspi *instanceptr,
  *	- Enable manual slave select
  *	- Enable manual start
  *	- Deselect all the chip select lines
- *	- Set the little endian mode of TX FIFO and
+ *	- Set the little endian mode of TX FIFO
+ *	- Set clock phase
+ *	- Set clock polarity and
  *	- Enable the QSPI controller
  */
 static void zynqmp_qspi_init_hw(struct zynqmp_qspi *xqspi)
 {
-	u32 config_reg;
+	u32 config_reg, baud_rate_val = 0;
+	ulong clk_rate;
 
 	/* Select the GQSPI mode */
 	zynqmp_gqspi_write(xqspi, GQSPI_SEL_OFST, GQSPI_SEL_MASK);
@@ -303,21 +390,37 @@ static void zynqmp_qspi_init_hw(struct zynqmp_qspi *xqspi)
 	config_reg |= GQSPI_CFG_WP_HOLD_MASK;
 	/* Clear pre-scalar by default */
 	config_reg &= ~GQSPI_CFG_BAUD_RATE_DIV_MASK;
-	/* CPHA 0 */
+	/* Set CPHA */
+	if (xqspi->ctlr->mode_bits & SPI_CPHA)
+		config_reg |= GQSPI_CFG_CLK_PHA_MASK;
+	else
 		config_reg &= ~GQSPI_CFG_CLK_PHA_MASK;
-	/* CPOL 0 */
+	/* Set CPOL */
+	if (xqspi->ctlr->mode_bits & SPI_CPOL)
+		config_reg |= GQSPI_CFG_CLK_POL_MASK;
+	else
 		config_reg &= ~GQSPI_CFG_CLK_POL_MASK;
+
+	/* Set the clock frequency */
+	clk_rate = clk_get_rate(xqspi->refclk);
+	while ((baud_rate_val < GQSPI_BAUD_DIV_MAX) &&
+	       (clk_rate /
+		(GQSPI_BAUD_DIV_SHIFT << baud_rate_val)) > xqspi->speed_hz)
+		baud_rate_val++;
+
+	config_reg &= ~GQSPI_CFG_BAUD_RATE_DIV_MASK;
+	config_reg |= (baud_rate_val << GQSPI_CFG_BAUD_RATE_DIV_SHIFT);
+
 	zynqmp_gqspi_write(xqspi, GQSPI_CONFIG_OFST, config_reg);
 
+	/* Set the tapdelay for clock frequency */
+	zynqmp_qspi_set_tapdelay(xqspi, baud_rate_val);
+
 	/* Clear the TX and RX FIFO */
 	zynqmp_gqspi_write(xqspi, GQSPI_FIFO_CTRL_OFST,
 			   GQSPI_FIFO_CTRL_RST_RX_FIFO_MASK |
 			   GQSPI_FIFO_CTRL_RST_TX_FIFO_MASK |
 			   GQSPI_FIFO_CTRL_RST_GEN_FIFO_MASK);
-	/* Set by default to allow for high frequencies */
-	zynqmp_gqspi_write(xqspi, GQSPI_LPBK_DLY_ADJ_OFST,
-			   zynqmp_gqspi_read(xqspi, GQSPI_LPBK_DLY_ADJ_OFST) |
-			   GQSPI_LPBK_DLY_ADJ_USE_LPBK_MASK);
 	/* Reset thresholds */
 	zynqmp_gqspi_write(xqspi, GQSPI_TX_THRESHOLD_OFST,
 			   GQSPI_TX_FIFO_THRESHOLD_RESET_VAL);
@@ -455,30 +558,30 @@ static int zynqmp_qspi_config_op(struct zynqmp_qspi *xqspi,
 				 struct spi_device *qspi)
 {
 	ulong clk_rate;
-	u32 config_reg, baud_rate_val = 0;
+	u32 config_reg, req_speed_hz, baud_rate_val = 0;
+
+	req_speed_hz = qspi->max_speed_hz;
+
+	if (xqspi->speed_hz != req_speed_hz) {
+		xqspi->speed_hz = req_speed_hz;
 
 		/* Set the clock frequency */
-	/* If req_hz == 0, default to lowest speed */
+		/* If req_speed_hz == 0, default to lowest speed */
 		clk_rate = clk_get_rate(xqspi->refclk);
 
 		while ((baud_rate_val < GQSPI_BAUD_DIV_MAX) &&
 		       (clk_rate /
-		(GQSPI_BAUD_DIV_SHIFT << baud_rate_val)) > qspi->max_speed_hz)
+			(GQSPI_BAUD_DIV_SHIFT << baud_rate_val)) >
+		       req_speed_hz)
 			baud_rate_val++;
 
 		config_reg = zynqmp_gqspi_read(xqspi, GQSPI_CONFIG_OFST);
 
-	/* Set the QSPI clock phase and clock polarity */
-	config_reg &= (~GQSPI_CFG_CLK_PHA_MASK) & (~GQSPI_CFG_CLK_POL_MASK);
-
-	if (qspi->mode & SPI_CPHA)
-		config_reg |= GQSPI_CFG_CLK_PHA_MASK;
-	if (qspi->mode & SPI_CPOL)
-		config_reg |= GQSPI_CFG_CLK_POL_MASK;
-
 		config_reg &= ~GQSPI_CFG_BAUD_RATE_DIV_MASK;
 		config_reg |= (baud_rate_val << GQSPI_CFG_BAUD_RATE_DIV_SHIFT);
 		zynqmp_gqspi_write(xqspi, GQSPI_CONFIG_OFST, config_reg);
+		zynqmp_qspi_set_tapdelay(xqspi, baud_rate_val);
+	}
 	return 0;
 }
 
@@ -739,6 +842,8 @@ static irqreturn_t zynqmp_qspi_irq(int irq, void *dev_id)
 /**
  * zynqmp_qspi_setuprxdma - This function sets up the RX DMA operation
  * @xqspi:	xqspi is a pointer to the GQSPI instance.
+ *
+ * Return:	0 on success; error value otherwise.
  */
 static int zynqmp_qspi_setuprxdma(struct zynqmp_qspi *xqspi)
 {
@@ -823,6 +928,8 @@ static void zynqmp_qspi_write_op(struct zynqmp_qspi *xqspi, u8 tx_nbits,
  * @rx_nbits:	Receive buswidth.
  * @genfifoentry:	genfifoentry is pointer to the variable in which
  *			GENFIFO	mask is returned to calling function
+ *
+ * Return:	0 on success; error value otherwise.
  */
 static int zynqmp_qspi_read_op(struct zynqmp_qspi *xqspi, u8 rx_nbits,
 				u32 genfifoentry)
@@ -1087,6 +1194,16 @@ static const struct dev_pm_ops zynqmp_qspi_dev_pm_ops = {
 	SET_SYSTEM_SLEEP_PM_OPS(zynqmp_qspi_suspend, zynqmp_qspi_resume)
 };
 
+static const struct qspi_platform_data versal_qspi_def = {
+	.quirks = QSPI_QUIRK_HAS_TAPDELAY,
+};
+
+static const struct of_device_id zynqmp_qspi_of_match[] = {
+	{ .compatible = "xlnx,zynqmp-qspi-1.0"},
+	{ .compatible = "xlnx,versal-qspi-1.0", .data = &versal_qspi_def },
+	{ /* End of table */ }
+};
+
 static const struct spi_controller_mem_ops zynqmp_qspi_mem_ops = {
 	.exec_op = zynqmp_qspi_exec_op,
 };
@@ -1107,6 +1224,7 @@ static int zynqmp_qspi_probe(struct platform_device *pdev)
 	struct device *dev = &pdev->dev;
 	struct device_node *np = dev->of_node;
 	u32 num_cs;
+	const struct qspi_platform_data *p_data;
 
 	ctlr = spi_alloc_master(&pdev->dev, sizeof(*xqspi));
 	if (!ctlr)
@@ -1117,6 +1235,10 @@ static int zynqmp_qspi_probe(struct platform_device *pdev)
 	xqspi->ctlr = ctlr;
 	platform_set_drvdata(pdev, xqspi);
 
+	p_data = of_device_get_match_data(&pdev->dev);
+	if (p_data && (p_data->quirks & QSPI_QUIRK_HAS_TAPDELAY))
+		xqspi->has_tapdelay = true;
+
 	xqspi->regs = devm_platform_ioremap_resource(pdev, 0);
 	if (IS_ERR(xqspi->regs)) {
 		ret = PTR_ERR(xqspi->regs);
@@ -1164,6 +1286,11 @@ static int zynqmp_qspi_probe(struct platform_device *pdev)
 		goto clk_dis_all;
 	}
 
+	ctlr->mode_bits = SPI_CPOL | SPI_CPHA | SPI_RX_DUAL | SPI_RX_QUAD |
+		SPI_TX_DUAL | SPI_TX_QUAD;
+	ctlr->max_speed_hz = clk_get_rate(xqspi->refclk) / 2;
+	xqspi->speed_hz = ctlr->max_speed_hz;
+
 	/* QSPI controller initializations */
 	zynqmp_qspi_init_hw(xqspi);
 
@@ -1199,10 +1326,7 @@ static int zynqmp_qspi_probe(struct platform_device *pdev)
 	ctlr->bits_per_word_mask = SPI_BPW_MASK(8);
 	ctlr->mem_ops = &zynqmp_qspi_mem_ops;
 	ctlr->setup = zynqmp_qspi_setup_op;
-	ctlr->max_speed_hz = clk_get_rate(xqspi->refclk) / 2;
 	ctlr->bits_per_word_mask = SPI_BPW_MASK(8);
-	ctlr->mode_bits = SPI_CPOL | SPI_CPHA | SPI_RX_DUAL | SPI_RX_QUAD |
-			    SPI_TX_DUAL | SPI_TX_QUAD;
 	ctlr->dev.of_node = np;
 	ctlr->auto_runtime_pm = true;
 
@@ -1253,11 +1377,6 @@ static int zynqmp_qspi_remove(struct platform_device *pdev)
 	return 0;
 }
 
-static const struct of_device_id zynqmp_qspi_of_match[] = {
-	{ .compatible = "xlnx,zynqmp-qspi-1.0", },
-	{ /* End of table */ }
-};
-
 MODULE_DEVICE_TABLE(of, zynqmp_qspi_of_match);
 
 static struct platform_driver zynqmp_qspi_driver = {

include/linux/firmware/xlnx-zynqmp.h

@@ -135,6 +135,7 @@ enum pm_ret_status {
 };
 
 enum pm_ioctl_id {
+	IOCTL_SET_TAPDELAY_BYPASS = 4,
 	IOCTL_SD_DLL_RESET = 6,
 	IOCTL_SET_SD_TAPDELAY = 7,
 	IOCTL_SET_PLL_FRAC_MODE = 8,
@@ -389,6 +390,18 @@ enum zynqmp_pm_shutdown_subtype {
 	ZYNQMP_PM_SHUTDOWN_SUBTYPE_SYSTEM = 2,
 };
 
+enum tap_delay_signal_type {
+	PM_TAPDELAY_NAND_DQS_IN = 0,
+	PM_TAPDELAY_NAND_DQS_OUT = 1,
+	PM_TAPDELAY_QSPI = 2,
+	PM_TAPDELAY_MAX = 3,
+};
+
+enum tap_delay_bypass_ctrl {
+	PM_TAPDELAY_BYPASS_DISABLE = 0,
+	PM_TAPDELAY_BYPASS_ENABLE = 1,
+};
+
 enum ospi_mux_select_type {
 	PM_OSPI_MUX_SEL_DMA = 0,
 	PM_OSPI_MUX_SEL_LINEAR = 1,
@@ -484,6 +497,7 @@ int zynqmp_pm_write_ggs(u32 index, u32 value);
 int zynqmp_pm_read_ggs(u32 index, u32 *value);
 int zynqmp_pm_write_pggs(u32 index, u32 value);
 int zynqmp_pm_read_pggs(u32 index, u32 *value);
+int zynqmp_pm_set_tapdelay_bypass(u32 index, u32 value);
 int zynqmp_pm_system_shutdown(const u32 type, const u32 subtype);
 int zynqmp_pm_set_boot_health_status(u32 value);
 int zynqmp_pm_pinctrl_request(const u32 pin);
@@ -696,6 +710,11 @@ static inline int zynqmp_pm_read_pggs(u32 index, u32 *value)
 	return -ENODEV;
 }
 
+static inline int zynqmp_pm_set_tapdelay_bypass(u32 index, u32 value)
+{
+	return -ENODEV;
+}
+
 static inline int zynqmp_pm_system_shutdown(const u32 type, const u32 subtype)
 {
 	return -ENODEV;