openwrt/target/linux/mediatek/patches-5.15/213-spi-mediatek-add-mt7986-spi-support.patch

From 7d99750f96fc6904d54affebdc8c9b0bfae1e9e8 Mon Sep 17 00:00:00 2001
From: Sam Shih <sam.shih@mediatek.com>
Date: Sun, 17 Apr 2022 11:40:22 +0800
Subject: [PATCH] spi: mediatek: backport document and driver to support mt7986
 spi design

this patch add the support of ipm design and upgrade devicetree binding

The patch is comming from following threads
- https://lore.kernel.org/all/20220315032411.2826-1-leilk.liu@mediatek.com/
- https://lore.kernel.org/all/20220401071616.8874-1-leilk.liu@mediatek.com/

Signed-off-by: Sam Shih <sam.shih@mediatek.com>
---
 .../bindings/spi/mediatek,spi-mt65xx.yaml     | 111 ++++
 drivers/spi/spi-mt65xx.c                      | 509 ++++++++++++++++--
 2 files changed, 572 insertions(+), 48 deletions(-)
 create mode 100644 Documentation/devicetree/bindings/spi/mediatek,spi-mt65xx.yaml

--- /dev/null
+++ b/Documentation/devicetree/bindings/spi/mediatek,spi-mt65xx.yaml
@@ -0,0 +1,111 @@
+# SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause)
+%YAML 1.2
+---
+$id: http://devicetree.org/schemas/spi/mediatek,spi-mt65xx.yaml#
+$schema: http://devicetree.org/meta-schemas/core.yaml#
+
+title: SPI Bus controller for MediaTek ARM SoCs
+
+maintainers:
+  - Leilk Liu <leilk.liu@mediatek.com>
+
+allOf:
+  - $ref: "/schemas/spi/spi-controller.yaml#"
+
+properties:
+  compatible:
+    oneOf:
+      - items:
+          - enum:
+              - mediatek,mt7629-spi
+          - const: mediatek,mt7622-spi
+      - items:
+          - enum:
+              - mediatek,mt8516-spi
+          - const: mediatek,mt2712-spi
+      - items:
+          - enum:
+              - mediatek,mt6779-spi
+              - mediatek,mt8186-spi
+              - mediatek,mt8192-spi
+              - mediatek,mt8195-spi
+          - const: mediatek,mt6765-spi
+      - items:
+          - enum:
+              - mediatek,mt7986-spi-ipm
+          - const: mediatek,spi-ipm
+      - items:
+          - enum:
+              - mediatek,mt2701-spi
+              - mediatek,mt2712-spi
+              - mediatek,mt6589-spi
+              - mediatek,mt6765-spi
+              - mediatek,mt6893-spi
+              - mediatek,mt7622-spi
+              - mediatek,mt8135-spi
+              - mediatek,mt8173-spi
+              - mediatek,mt8183-spi
+
+  reg:
+    maxItems: 1
+
+  interrupts:
+    maxItems: 1
+
+  clocks:
+    minItems: 3
+    items:
+      - description: clock used for the parent clock
+      - description: clock used for the muxes clock
+      - description: clock used for the clock gate
+      - description: clock used for the AHB bus, this clock is optional
+
+  clock-names:
+    minItems: 3
+    items:
+      - const: parent-clk
+      - const: sel-clk
+      - const: spi-clk
+      - const: hclk
+
+  mediatek,pad-select:
+    $ref: /schemas/types.yaml#/definitions/uint32-array
+    minItems: 1
+    maxItems: 4
+    items:
+      enum: [0, 1, 2, 3]
+    description:
+      specify which pins group(ck/mi/mo/cs) spi controller used.
+      This is an array.
+
+required:
+  - compatible
+  - reg
+  - interrupts
+  - clocks
+  - clock-names
+  - '#address-cells'
+  - '#size-cells'
+
+unevaluatedProperties: false
+
+examples:
+  - |
+    #include <dt-bindings/clock/mt8173-clk.h>
+    #include <dt-bindings/gpio/gpio.h>
+    #include <dt-bindings/interrupt-controller/arm-gic.h>
+    #include <dt-bindings/interrupt-controller/irq.h>
+
+    spi@1100a000 {
+      compatible = "mediatek,mt8173-spi";
+      #address-cells = <1>;
+      #size-cells = <0>;
+      reg = <0x1100a000 0x1000>;
+      interrupts = <GIC_SPI 110 IRQ_TYPE_LEVEL_LOW>;
+      clocks = <&topckgen CLK_TOP_SYSPLL3_D2>,
+               <&topckgen CLK_TOP_SPI_SEL>,
+               <&pericfg CLK_PERI_SPI0>;
+      clock-names = "parent-clk", "sel-clk", "spi-clk";
+      cs-gpios = <&pio 105 GPIO_ACTIVE_LOW>, <&pio 72 GPIO_ACTIVE_LOW>;
+      mediatek,pad-select = <1>, <0>;
+    };
--- a/drivers/spi/spi-mt65xx.c
+++ b/drivers/spi/spi-mt65xx.c
@@ -12,11 +12,12 @@
 #include <linux/ioport.h>
 #include <linux/module.h>
 #include <linux/of.h>
-#include <linux/of_gpio.h>
+#include <linux/gpio/consumer.h>
 #include <linux/platform_device.h>
 #include <linux/platform_data/spi-mt65xx.h>
 #include <linux/pm_runtime.h>
 #include <linux/spi/spi.h>
+#include <linux/spi/spi-mem.h>
 #include <linux/dma-mapping.h>
 
 #define SPI_CFG0_REG                      0x0000
@@ -31,6 +32,7 @@
 #define SPI_CFG2_REG                      0x0028
 #define SPI_TX_SRC_REG_64                 0x002c
 #define SPI_RX_DST_REG_64                 0x0030
+#define SPI_CFG3_IPM_REG                  0x0040
 
 #define SPI_CFG0_SCK_HIGH_OFFSET          0
 #define SPI_CFG0_SCK_LOW_OFFSET           8
@@ -51,6 +53,7 @@
 #define SPI_CFG1_CS_IDLE_MASK             0xff
 #define SPI_CFG1_PACKET_LOOP_MASK         0xff00
 #define SPI_CFG1_PACKET_LENGTH_MASK       0x3ff0000
+#define SPI_CFG1_IPM_PACKET_LENGTH_MASK   GENMASK(31, 16)
 #define SPI_CFG2_SCK_HIGH_OFFSET          0
 #define SPI_CFG2_SCK_LOW_OFFSET           16
 
@@ -71,6 +74,24 @@
 #define SPI_CMD_TX_ENDIAN            BIT(15)
 #define SPI_CMD_FINISH_IE            BIT(16)
 #define SPI_CMD_PAUSE_IE             BIT(17)
+#define SPI_CMD_IPM_NONIDLE_MODE     BIT(19)
+#define SPI_CMD_IPM_SPIM_LOOP        BIT(21)
+#define SPI_CMD_IPM_GET_TICKDLY_OFFSET    22
+
+#define SPI_CMD_IPM_GET_TICKDLY_MASK	GENMASK(24, 22)
+
+#define PIN_MODE_CFG(x)	((x) / 2)
+
+#define SPI_CFG3_IPM_HALF_DUPLEX_DIR		BIT(2)
+#define SPI_CFG3_IPM_HALF_DUPLEX_EN		BIT(3)
+#define SPI_CFG3_IPM_XMODE_EN			BIT(4)
+#define SPI_CFG3_IPM_NODATA_FLAG		BIT(5)
+#define SPI_CFG3_IPM_CMD_BYTELEN_OFFSET		8
+#define SPI_CFG3_IPM_ADDR_BYTELEN_OFFSET	12
+
+#define SPI_CFG3_IPM_CMD_PIN_MODE_MASK		GENMASK(1, 0)
+#define SPI_CFG3_IPM_CMD_BYTELEN_MASK		GENMASK(11, 8)
+#define SPI_CFG3_IPM_ADDR_BYTELEN_MASK		GENMASK(15, 12)
 
 #define MT8173_SPI_MAX_PAD_SEL 3
 
@@ -81,6 +102,9 @@
 
 #define MTK_SPI_MAX_FIFO_SIZE 32U
 #define MTK_SPI_PACKET_SIZE 1024
+#define MTK_SPI_IPM_PACKET_SIZE SZ_64K
+#define MTK_SPI_IPM_PACKET_LOOP SZ_256
+
 #define MTK_SPI_32BITS_MASK  (0xffffffff)
 
 #define DMA_ADDR_EXT_BITS (36)
@@ -96,6 +120,8 @@ struct mtk_spi_compatible {
 	bool dma_ext;
 	/* some IC no need unprepare SPI clk */
 	bool no_need_unprepare;
+	/* IPM design adjust and extend register to support more features */
+	bool ipm_design;
 };
 
 struct mtk_spi {
@@ -103,7 +129,7 @@ struct mtk_spi {
 	u32 state;
 	int pad_num;
 	u32 *pad_sel;
-	struct clk *parent_clk, *sel_clk, *spi_clk;
+	struct clk *parent_clk, *sel_clk, *spi_clk, *spi_hclk;
 	struct spi_transfer *cur_transfer;
 	u32 xfer_len;
 	u32 num_xfered;
@@ -111,6 +137,11 @@ struct mtk_spi {
 	u32 tx_sgl_len, rx_sgl_len;
 	const struct mtk_spi_compatible *dev_comp;
 	u32 spi_clk_hz;
+	struct completion spimem_done;
+	bool use_spimem;
+	struct device *dev;
+	dma_addr_t tx_dma;
+	dma_addr_t rx_dma;
 };
 
 static const struct mtk_spi_compatible mtk_common_compat;
@@ -119,6 +150,12 @@ static const struct mtk_spi_compatible m
 	.must_tx = true,
 };
 
+static const struct mtk_spi_compatible mtk_ipm_compat = {
+	.enhance_timing = true,
+	.dma_ext = true,
+	.ipm_design = true,
+};
+
 static const struct mtk_spi_compatible mt6765_compat = {
 	.need_pad_sel = true,
 	.must_tx = true,
@@ -160,6 +197,9 @@ static const struct mtk_chip_config mtk_
 };
 
 static const struct of_device_id mtk_spi_of_match[] = {
+	{ .compatible = "mediatek,spi-ipm",
+		.data = (void *)&mtk_ipm_compat,
+	},
 	{ .compatible = "mediatek,mt2701-spi",
 		.data = (void *)&mtk_common_compat,
 	},
@@ -278,12 +318,11 @@ static int mtk_spi_set_hw_cs_timing(stru
 	return 0;
 }
 
-static int mtk_spi_prepare_message(struct spi_master *master,
-				   struct spi_message *msg)
+static int mtk_spi_hw_init(struct spi_master *master,
+			   struct spi_device *spi)
 {
 	u16 cpha, cpol;
 	u32 reg_val;
-	struct spi_device *spi = msg->spi;
 	struct mtk_chip_config *chip_config = spi->controller_data;
 	struct mtk_spi *mdata = spi_master_get_devdata(master);
 
@@ -291,6 +330,15 @@ static int mtk_spi_prepare_message(struc
 	cpol = spi->mode & SPI_CPOL ? 1 : 0;
 
 	reg_val = readl(mdata->base + SPI_CMD_REG);
+	if (mdata->dev_comp->ipm_design) {
+		/* SPI transfer without idle time until packet length done */
+		reg_val |= SPI_CMD_IPM_NONIDLE_MODE;
+		if (spi->mode & SPI_LOOP)
+			reg_val |= SPI_CMD_IPM_SPIM_LOOP;
+		else
+			reg_val &= ~SPI_CMD_IPM_SPIM_LOOP;
+	}
+
 	if (cpha)
 		reg_val |= SPI_CMD_CPHA;
 	else
@@ -348,23 +396,39 @@ static int mtk_spi_prepare_message(struc
 		       mdata->base + SPI_PAD_SEL_REG);
 
 	/* tick delay */
-	reg_val = readl(mdata->base + SPI_CFG1_REG);
 	if (mdata->dev_comp->enhance_timing) {
-		reg_val &= ~SPI_CFG1_GET_TICK_DLY_MASK;
-		reg_val |= ((chip_config->tick_delay & 0x7)
-			    << SPI_CFG1_GET_TICK_DLY_OFFSET);
+		if (mdata->dev_comp->ipm_design) {
+			reg_val = readl(mdata->base + SPI_CMD_REG);
+			reg_val &= ~SPI_CMD_IPM_GET_TICKDLY_MASK;
+			reg_val |= ((chip_config->tick_delay & 0x7)
+				    << SPI_CMD_IPM_GET_TICKDLY_OFFSET);
+			writel(reg_val, mdata->base + SPI_CMD_REG);
+		} else {
+			reg_val = readl(mdata->base + SPI_CFG1_REG);
+			reg_val &= ~SPI_CFG1_GET_TICK_DLY_MASK;
+			reg_val |= ((chip_config->tick_delay & 0x7)
+				    << SPI_CFG1_GET_TICK_DLY_OFFSET);
+			writel(reg_val, mdata->base + SPI_CFG1_REG);
+		}
 	} else {
+		reg_val = readl(mdata->base + SPI_CFG1_REG);
 		reg_val &= ~SPI_CFG1_GET_TICK_DLY_MASK_V1;
 		reg_val |= ((chip_config->tick_delay & 0x3)
 			    << SPI_CFG1_GET_TICK_DLY_OFFSET_V1);
+		writel(reg_val, mdata->base + SPI_CFG1_REG);
 	}
-	writel(reg_val, mdata->base + SPI_CFG1_REG);
 
 	/* set hw cs timing */
 	mtk_spi_set_hw_cs_timing(spi);
 	return 0;
 }
 
+static int mtk_spi_prepare_message(struct spi_master *master,
+				   struct spi_message *msg)
+{
+	return mtk_spi_hw_init(master, msg->spi);
+}
+
 static void mtk_spi_set_cs(struct spi_device *spi, bool enable)
 {
 	u32 reg_val;
@@ -386,13 +450,13 @@ static void mtk_spi_set_cs(struct spi_de
 }
 
 static void mtk_spi_prepare_transfer(struct spi_master *master,
-				     struct spi_transfer *xfer)
+				     u32 speed_hz)
 {
 	u32 div, sck_time, reg_val;
 	struct mtk_spi *mdata = spi_master_get_devdata(master);
 
-	if (xfer->speed_hz < mdata->spi_clk_hz / 2)
-		div = DIV_ROUND_UP(mdata->spi_clk_hz, xfer->speed_hz);
+	if (speed_hz < mdata->spi_clk_hz / 2)
+		div = DIV_ROUND_UP(mdata->spi_clk_hz, speed_hz);
 	else
 		div = 1;
 
@@ -423,12 +487,24 @@ static void mtk_spi_setup_packet(struct
 	u32 packet_size, packet_loop, reg_val;
 	struct mtk_spi *mdata = spi_master_get_devdata(master);
 
-	packet_size = min_t(u32, mdata->xfer_len, MTK_SPI_PACKET_SIZE);
+	if (mdata->dev_comp->ipm_design)
+		packet_size = min_t(u32,
+				    mdata->xfer_len,
+				    MTK_SPI_IPM_PACKET_SIZE);
+	else
+		packet_size = min_t(u32,
+				    mdata->xfer_len,
+				    MTK_SPI_PACKET_SIZE);
+
 	packet_loop = mdata->xfer_len / packet_size;
 
 	reg_val = readl(mdata->base + SPI_CFG1_REG);
-	reg_val &= ~(SPI_CFG1_PACKET_LENGTH_MASK | SPI_CFG1_PACKET_LOOP_MASK);
+	if (mdata->dev_comp->ipm_design)
+		reg_val &= ~SPI_CFG1_IPM_PACKET_LENGTH_MASK;
+	else
+		reg_val &= ~SPI_CFG1_PACKET_LENGTH_MASK;
 	reg_val |= (packet_size - 1) << SPI_CFG1_PACKET_LENGTH_OFFSET;
+	reg_val &= ~SPI_CFG1_PACKET_LOOP_MASK;
 	reg_val |= (packet_loop - 1) << SPI_CFG1_PACKET_LOOP_OFFSET;
 	writel(reg_val, mdata->base + SPI_CFG1_REG);
 }
@@ -523,7 +599,7 @@ static int mtk_spi_fifo_transfer(struct
 	mdata->cur_transfer = xfer;
 	mdata->xfer_len = min(MTK_SPI_MAX_FIFO_SIZE, xfer->len);
 	mdata->num_xfered = 0;
-	mtk_spi_prepare_transfer(master, xfer);
+	mtk_spi_prepare_transfer(master, xfer->speed_hz);
 	mtk_spi_setup_packet(master);
 
 	if (xfer->tx_buf) {
@@ -556,7 +632,7 @@ static int mtk_spi_dma_transfer(struct s
 	mdata->cur_transfer = xfer;
 	mdata->num_xfered = 0;
 
-	mtk_spi_prepare_transfer(master, xfer);
+	mtk_spi_prepare_transfer(master, xfer->speed_hz);
 
 	cmd = readl(mdata->base + SPI_CMD_REG);
 	if (xfer->tx_buf)
@@ -591,6 +667,19 @@ static int mtk_spi_transfer_one(struct s
 				struct spi_device *spi,
 				struct spi_transfer *xfer)
 {
+	struct mtk_spi *mdata = spi_master_get_devdata(spi->master);
+	u32 reg_val = 0;
+
+	/* prepare xfer direction and duplex mode */
+	if (mdata->dev_comp->ipm_design) {
+		if (!xfer->tx_buf || !xfer->rx_buf) {
+			reg_val |= SPI_CFG3_IPM_HALF_DUPLEX_EN;
+			if (xfer->rx_buf)
+				reg_val |= SPI_CFG3_IPM_HALF_DUPLEX_DIR;
+		}
+		writel(reg_val, mdata->base + SPI_CFG3_IPM_REG);
+	}
+
 	if (master->can_dma(master, spi, xfer))
 		return mtk_spi_dma_transfer(master, spi, xfer);
 	else
@@ -614,8 +703,9 @@ static int mtk_spi_setup(struct spi_devi
 	if (!spi->controller_data)
 		spi->controller_data = (void *)&mtk_default_chip_info;
 
-	if (mdata->dev_comp->need_pad_sel && gpio_is_valid(spi->cs_gpio))
-		gpio_direction_output(spi->cs_gpio, !(spi->mode & SPI_CS_HIGH));
+	if (mdata->dev_comp->need_pad_sel && spi->cs_gpiod)
+		/* CS de-asserted, gpiolib will handle inversion */
+		gpiod_direction_output(spi->cs_gpiod, 0);
 
 	return 0;
 }
@@ -633,6 +723,12 @@ static irqreturn_t mtk_spi_interrupt(int
 	else
 		mdata->state = MTK_SPI_IDLE;
 
+	/* SPI-MEM ops */
+	if (mdata->use_spimem) {
+		complete(&mdata->spimem_done);
+		return IRQ_HANDLED;
+	}
+
 	if (!master->can_dma(master, NULL, trans)) {
 		if (trans->rx_buf) {
 			cnt = mdata->xfer_len / 4;
@@ -718,6 +814,274 @@ static irqreturn_t mtk_spi_interrupt(int
 	return IRQ_HANDLED;
 }
 
+static int mtk_spi_mem_adjust_op_size(struct spi_mem *mem,
+				      struct spi_mem_op *op)
+{
+	int opcode_len;
+
+	if (op->data.dir != SPI_MEM_NO_DATA) {
+		opcode_len = 1 + op->addr.nbytes + op->dummy.nbytes;
+		if (opcode_len + op->data.nbytes > MTK_SPI_IPM_PACKET_SIZE) {
+			op->data.nbytes = MTK_SPI_IPM_PACKET_SIZE - opcode_len;
+			/* force data buffer dma-aligned. */
+			op->data.nbytes -= op->data.nbytes % 4;
+		}
+	}
+
+	return 0;
+}
+
+static bool mtk_spi_mem_supports_op(struct spi_mem *mem,
+				    const struct spi_mem_op *op)
+{
+	if (!spi_mem_default_supports_op(mem, op))
+		return false;
+
+	if (op->addr.nbytes && op->dummy.nbytes &&
+	    op->addr.buswidth != op->dummy.buswidth)
+		return false;
+
+	if (op->addr.nbytes + op->dummy.nbytes > 16)
+		return false;
+
+	if (op->data.nbytes > MTK_SPI_IPM_PACKET_SIZE) {
+		if (op->data.nbytes / MTK_SPI_IPM_PACKET_SIZE >
+		    MTK_SPI_IPM_PACKET_LOOP ||
+		    op->data.nbytes % MTK_SPI_IPM_PACKET_SIZE != 0)
+			return false;
+	}
+
+	return true;
+}
+
+static void mtk_spi_mem_setup_dma_xfer(struct spi_master *master,
+				       const struct spi_mem_op *op)
+{
+	struct mtk_spi *mdata = spi_master_get_devdata(master);
+
+	writel((u32)(mdata->tx_dma & MTK_SPI_32BITS_MASK),
+	       mdata->base + SPI_TX_SRC_REG);
+#ifdef CONFIG_ARCH_DMA_ADDR_T_64BIT
+	if (mdata->dev_comp->dma_ext)
+		writel((u32)(mdata->tx_dma >> 32),
+		       mdata->base + SPI_TX_SRC_REG_64);
+#endif
+
+	if (op->data.dir == SPI_MEM_DATA_IN) {
+		writel((u32)(mdata->rx_dma & MTK_SPI_32BITS_MASK),
+		       mdata->base + SPI_RX_DST_REG);
+#ifdef CONFIG_ARCH_DMA_ADDR_T_64BIT
+		if (mdata->dev_comp->dma_ext)
+			writel((u32)(mdata->rx_dma >> 32),
+			       mdata->base + SPI_RX_DST_REG_64);
+#endif
+	}
+}
+
+static int mtk_spi_transfer_wait(struct spi_mem *mem,
+				 const struct spi_mem_op *op)
+{
+	struct mtk_spi *mdata = spi_master_get_devdata(mem->spi->master);
+	/*
+	 * For each byte we wait for 8 cycles of the SPI clock.
+	 * Since speed is defined in Hz and we want milliseconds,
+	 * so it should be 8 * 1000.
+	 */
+	u64 ms = 8000LL;
+
+	if (op->data.dir == SPI_MEM_NO_DATA)
+		ms *= 32; /* prevent we may get 0 for short transfers. */
+	else
+		ms *= op->data.nbytes;
+	ms = div_u64(ms, mem->spi->max_speed_hz);
+	ms += ms + 1000; /* 1s tolerance */
+
+	if (ms > UINT_MAX)
+		ms = UINT_MAX;
+
+	if (!wait_for_completion_timeout(&mdata->spimem_done,
+					 msecs_to_jiffies(ms))) {
+		dev_err(mdata->dev, "spi-mem transfer timeout\n");
+		return -ETIMEDOUT;
+	}
+
+	return 0;
+}
+
+static int mtk_spi_mem_exec_op(struct spi_mem *mem,
+			       const struct spi_mem_op *op)
+{
+	struct mtk_spi *mdata = spi_master_get_devdata(mem->spi->master);
+	u32 reg_val, nio, tx_size;
+	char *tx_tmp_buf, *rx_tmp_buf;
+	int ret = 0;
+
+	mdata->use_spimem = true;
+	reinit_completion(&mdata->spimem_done);
+
+	mtk_spi_reset(mdata);
+	mtk_spi_hw_init(mem->spi->master, mem->spi);
+	mtk_spi_prepare_transfer(mem->spi->master, mem->spi->max_speed_hz);
+
+	reg_val = readl(mdata->base + SPI_CFG3_IPM_REG);
+	/* opcode byte len */
+	reg_val &= ~SPI_CFG3_IPM_CMD_BYTELEN_MASK;
+	reg_val |= 1 << SPI_CFG3_IPM_CMD_BYTELEN_OFFSET;
+
+	/* addr & dummy byte len */
+	reg_val &= ~SPI_CFG3_IPM_ADDR_BYTELEN_MASK;
+	if (op->addr.nbytes || op->dummy.nbytes)
+		reg_val |= (op->addr.nbytes + op->dummy.nbytes) <<
+			    SPI_CFG3_IPM_ADDR_BYTELEN_OFFSET;
+
+	/* data byte len */
+	if (op->data.dir == SPI_MEM_NO_DATA) {
+		reg_val |= SPI_CFG3_IPM_NODATA_FLAG;
+		writel(0, mdata->base + SPI_CFG1_REG);
+	} else {
+		reg_val &= ~SPI_CFG3_IPM_NODATA_FLAG;
+		mdata->xfer_len = op->data.nbytes;
+		mtk_spi_setup_packet(mem->spi->master);
+	}
+
+	if (op->addr.nbytes || op->dummy.nbytes) {
+		if (op->addr.buswidth == 1 || op->dummy.buswidth == 1)
+			reg_val |= SPI_CFG3_IPM_XMODE_EN;
+		else
+			reg_val &= ~SPI_CFG3_IPM_XMODE_EN;
+	}
+
+	if (op->addr.buswidth == 2 ||
+	    op->dummy.buswidth == 2 ||
+	    op->data.buswidth == 2)
+		nio = 2;
+	else if (op->addr.buswidth == 4 ||
+		 op->dummy.buswidth == 4 ||
+		 op->data.buswidth == 4)
+		nio = 4;
+	else
+		nio = 1;
+
+	reg_val &= ~SPI_CFG3_IPM_CMD_PIN_MODE_MASK;
+	reg_val |= PIN_MODE_CFG(nio);
+
+	reg_val |= SPI_CFG3_IPM_HALF_DUPLEX_EN;
+	if (op->data.dir == SPI_MEM_DATA_IN)
+		reg_val |= SPI_CFG3_IPM_HALF_DUPLEX_DIR;
+	else
+		reg_val &= ~SPI_CFG3_IPM_HALF_DUPLEX_DIR;
+	writel(reg_val, mdata->base + SPI_CFG3_IPM_REG);
+
+	tx_size = 1 + op->addr.nbytes + op->dummy.nbytes;
+	if (op->data.dir == SPI_MEM_DATA_OUT)
+		tx_size += op->data.nbytes;
+
+	tx_size = max_t(u32, tx_size, 32);
+
+	tx_tmp_buf = kzalloc(tx_size, GFP_KERNEL | GFP_DMA);
+	if (!tx_tmp_buf) {
+		mdata->use_spimem = false;
+		return -ENOMEM;
+	}
+
+	tx_tmp_buf[0] = op->cmd.opcode;
+
+	if (op->addr.nbytes) {
+		int i;
+
+		for (i = 0; i < op->addr.nbytes; i++)
+			tx_tmp_buf[i + 1] = op->addr.val >>
+					(8 * (op->addr.nbytes - i - 1));
+	}
+
+	if (op->dummy.nbytes)
+		memset(tx_tmp_buf + op->addr.nbytes + 1,
+		       0xff,
+		       op->dummy.nbytes);
+
+	if (op->data.nbytes && op->data.dir == SPI_MEM_DATA_OUT)
+		memcpy(tx_tmp_buf + op->dummy.nbytes + op->addr.nbytes + 1,
+		       op->data.buf.out,
+		       op->data.nbytes);
+
+	mdata->tx_dma = dma_map_single(mdata->dev, tx_tmp_buf,
+				       tx_size, DMA_TO_DEVICE);
+	if (dma_mapping_error(mdata->dev, mdata->tx_dma)) {
+		ret = -ENOMEM;
+		goto err_exit;
+	}
+
+	if (op->data.dir == SPI_MEM_DATA_IN) {
+		if (!IS_ALIGNED((size_t)op->data.buf.in, 4)) {
+			rx_tmp_buf = kzalloc(op->data.nbytes,
+					     GFP_KERNEL | GFP_DMA);
+			if (!rx_tmp_buf) {
+				ret = -ENOMEM;
+				goto unmap_tx_dma;
+			}
+		} else {
+			rx_tmp_buf = op->data.buf.in;
+		}
+
+		mdata->rx_dma = dma_map_single(mdata->dev,
+					       rx_tmp_buf,
+					       op->data.nbytes,
+					       DMA_FROM_DEVICE);
+		if (dma_mapping_error(mdata->dev, mdata->rx_dma)) {
+			ret = -ENOMEM;
+			goto kfree_rx_tmp_buf;
+		}
+	}
+
+	reg_val = readl(mdata->base + SPI_CMD_REG);
+	reg_val |= SPI_CMD_TX_DMA;
+	if (op->data.dir == SPI_MEM_DATA_IN)
+		reg_val |= SPI_CMD_RX_DMA;
+	writel(reg_val, mdata->base + SPI_CMD_REG);
+
+	mtk_spi_mem_setup_dma_xfer(mem->spi->master, op);
+
+	mtk_spi_enable_transfer(mem->spi->master);
+
+	/* Wait for the interrupt. */
+	ret = mtk_spi_transfer_wait(mem, op);
+	if (ret)
+		goto unmap_rx_dma;
+
+	/* spi disable dma */
+	reg_val = readl(mdata->base + SPI_CMD_REG);
+	reg_val &= ~SPI_CMD_TX_DMA;
+	if (op->data.dir == SPI_MEM_DATA_IN)
+		reg_val &= ~SPI_CMD_RX_DMA;
+	writel(reg_val, mdata->base + SPI_CMD_REG);
+
+unmap_rx_dma:
+	if (op->data.dir == SPI_MEM_DATA_IN) {
+		dma_unmap_single(mdata->dev, mdata->rx_dma,
+				 op->data.nbytes, DMA_FROM_DEVICE);
+		if (!IS_ALIGNED((size_t)op->data.buf.in, 4))
+			memcpy(op->data.buf.in, rx_tmp_buf, op->data.nbytes);
+	}
+kfree_rx_tmp_buf:
+	if (op->data.dir == SPI_MEM_DATA_IN &&
+	    !IS_ALIGNED((size_t)op->data.buf.in, 4))
+		kfree(rx_tmp_buf);
+unmap_tx_dma:
+	dma_unmap_single(mdata->dev, mdata->tx_dma,
+			 tx_size, DMA_TO_DEVICE);
+err_exit:
+	kfree(tx_tmp_buf);
+	mdata->use_spimem = false;
+
+	return ret;
+}
+
+static const struct spi_controller_mem_ops mtk_spi_mem_ops = {
+	.adjust_op_size = mtk_spi_mem_adjust_op_size,
+	.supports_op = mtk_spi_mem_supports_op,
+	.exec_op = mtk_spi_mem_exec_op,
+};
+
 static int mtk_spi_probe(struct platform_device *pdev)
 {
 	struct spi_master *master;
@@ -741,6 +1105,7 @@ static int mtk_spi_probe(struct platform
 	master->can_dma = mtk_spi_can_dma;
 	master->setup = mtk_spi_setup;
 	master->set_cs_timing = mtk_spi_set_hw_cs_timing;
+	master->use_gpio_descriptors = true;
 
 	of_id = of_match_node(mtk_spi_of_match, pdev->dev.of_node);
 	if (!of_id) {
@@ -757,6 +1122,14 @@ static int mtk_spi_probe(struct platform
 
 	if (mdata->dev_comp->must_tx)
 		master->flags = SPI_MASTER_MUST_TX;
+	if (mdata->dev_comp->ipm_design)
+		master->mode_bits |= SPI_LOOP;
+
+	if (mdata->dev_comp->ipm_design) {
+		mdata->dev = &pdev->dev;
+		master->mem_ops = &mtk_spi_mem_ops;
+		init_completion(&mdata->spimem_done);
+	}
 
 	if (mdata->dev_comp->need_pad_sel) {
 		mdata->pad_num = of_property_count_u32_elems(
@@ -833,25 +1206,40 @@ static int mtk_spi_probe(struct platform
 		goto err_put_master;
 	}
 
+	mdata->spi_hclk = devm_clk_get_optional(&pdev->dev, "hclk");
+	if (IS_ERR(mdata->spi_hclk)) {
+		ret = PTR_ERR(mdata->spi_hclk);
+		dev_err(&pdev->dev, "failed to get hclk: %d\n", ret);
+		goto err_put_master;
+	}
+
+	ret = clk_prepare_enable(mdata->spi_hclk);
+	if (ret < 0) {
+		dev_err(&pdev->dev, "failed to enable hclk (%d)\n", ret);
+		goto err_put_master;
+	}
+
 	ret = clk_prepare_enable(mdata->spi_clk);
 	if (ret < 0) {
 		dev_err(&pdev->dev, "failed to enable spi_clk (%d)\n", ret);
-		goto err_put_master;
+		goto err_disable_spi_hclk;
 	}
 
 	ret = clk_set_parent(mdata->sel_clk, mdata->parent_clk);
 	if (ret < 0) {
 		dev_err(&pdev->dev, "failed to clk_set_parent (%d)\n", ret);
-		clk_disable_unprepare(mdata->spi_clk);
-		goto err_put_master;
+		goto err_disable_spi_clk;
 	}
 
 	mdata->spi_clk_hz = clk_get_rate(mdata->spi_clk);
 
-	if (mdata->dev_comp->no_need_unprepare)
+	if (mdata->dev_comp->no_need_unprepare) {
 		clk_disable(mdata->spi_clk);
-	else
+		clk_disable(mdata->spi_hclk);
+	} else {
 		clk_disable_unprepare(mdata->spi_clk);
+		clk_disable_unprepare(mdata->spi_hclk);
+	}
 
 	pm_runtime_enable(&pdev->dev);
 
@@ -864,25 +1252,12 @@ static int mtk_spi_probe(struct platform
 			goto err_disable_runtime_pm;
 		}
 
-		if (!master->cs_gpios && master->num_chipselect > 1) {
+		if (!master->cs_gpiods && master->num_chipselect > 1) {
 			dev_err(&pdev->dev,
 				"cs_gpios not specified and num_chipselect > 1\n");
 			ret = -EINVAL;
 			goto err_disable_runtime_pm;
 		}
-
-		if (master->cs_gpios) {
-			for (i = 0; i < master->num_chipselect; i++) {
-				ret = devm_gpio_request(&pdev->dev,
-							master->cs_gpios[i],
-							dev_name(&pdev->dev));
-				if (ret) {
-					dev_err(&pdev->dev,
-						"can't get CS GPIO %i\n", i);
-					goto err_disable_runtime_pm;
-				}
-			}
-		}
 	}
 
 	if (mdata->dev_comp->dma_ext)
@@ -904,6 +1279,10 @@ static int mtk_spi_probe(struct platform
 
 err_disable_runtime_pm:
 	pm_runtime_disable(&pdev->dev);
+err_disable_spi_clk:
+	clk_disable_unprepare(mdata->spi_clk);
+err_disable_spi_hclk:
+	clk_disable_unprepare(mdata->spi_hclk);
 err_put_master:
 	spi_master_put(master);
 
@@ -922,8 +1301,10 @@ static int mtk_spi_remove(struct platfor
 
 	mtk_spi_reset(mdata);
 
-	if (mdata->dev_comp->no_need_unprepare)
+	if (mdata->dev_comp->no_need_unprepare) {
 		clk_unprepare(mdata->spi_clk);
+		clk_unprepare(mdata->spi_hclk);
+	}
 
 	pm_runtime_put_noidle(&pdev->dev);
 	pm_runtime_disable(&pdev->dev);
@@ -942,8 +1323,10 @@ static int mtk_spi_suspend(struct device
 	if (ret)
 		return ret;
 
-	if (!pm_runtime_suspended(dev))
+	if (!pm_runtime_suspended(dev)) {
 		clk_disable_unprepare(mdata->spi_clk);
+		clk_disable_unprepare(mdata->spi_hclk);
+	}
 
 	return ret;
 }
@@ -960,11 +1343,20 @@ static int mtk_spi_resume(struct device
 			dev_err(dev, "failed to enable spi_clk (%d)\n", ret);
 			return ret;
 		}
+
+		ret = clk_prepare_enable(mdata->spi_hclk);
+		if (ret < 0) {
+			dev_err(dev, "failed to enable spi_hclk (%d)\n", ret);
+			clk_disable_unprepare(mdata->spi_clk);
+			return ret;
+		}
 	}
 
 	ret = spi_master_resume(master);
-	if (ret < 0)
+	if (ret < 0) {
 		clk_disable_unprepare(mdata->spi_clk);
+		clk_disable_unprepare(mdata->spi_hclk);
+	}
 
 	return ret;
 }
@@ -976,10 +1368,13 @@ static int mtk_spi_runtime_suspend(struc
 	struct spi_master *master = dev_get_drvdata(dev);
 	struct mtk_spi *mdata = spi_master_get_devdata(master);
 
-	if (mdata->dev_comp->no_need_unprepare)
+	if (mdata->dev_comp->no_need_unprepare) {
 		clk_disable(mdata->spi_clk);
-	else
+		clk_disable(mdata->spi_hclk);
+	} else {
 		clk_disable_unprepare(mdata->spi_clk);
+		clk_disable_unprepare(mdata->spi_hclk);
+	}
 
 	return 0;
 }
@@ -990,13 +1385,31 @@ static int mtk_spi_runtime_resume(struct
 	struct mtk_spi *mdata = spi_master_get_devdata(master);
 	int ret;
 
-	if (mdata->dev_comp->no_need_unprepare)
+	if (mdata->dev_comp->no_need_unprepare) {
 		ret = clk_enable(mdata->spi_clk);
-	else
+		if (ret < 0) {
+			dev_err(dev, "failed to enable spi_clk (%d)\n", ret);
+			return ret;
+		}
+		ret = clk_enable(mdata->spi_hclk);
+		if (ret < 0) {
+			dev_err(dev, "failed to enable spi_hclk (%d)\n", ret);
+			clk_disable(mdata->spi_clk);
+			return ret;
+		}
+	} else {
 		ret = clk_prepare_enable(mdata->spi_clk);
-	if (ret < 0) {
-		dev_err(dev, "failed to enable spi_clk (%d)\n", ret);
-		return ret;
+		if (ret < 0) {
+			dev_err(dev, "failed to prepare_enable spi_clk (%d)\n", ret);
+			return ret;
+		}
+
+		ret = clk_prepare_enable(mdata->spi_hclk);
+		if (ret < 0) {
+			dev_err(dev, "failed to prepare_enable spi_hclk (%d)\n", ret);
+			clk_disable_unprepare(mdata->spi_clk);
+			return ret;
+		}
 	}
 
 	return 0;