1 files changed, 206 insertions, 38 deletions

diff --git a/drivers/spi/mxs_spi.c b/drivers/spi/mxs_spi.c
index 7859536a67..168dbe497e 100644
--- a/drivers/spi/mxs_spi.c
+++ b/drivers/spi/mxs_spi.c
@@ -31,17 +31,31 @@
 #include <asm/arch/clock.h>
 #include <asm/arch/imx-regs.h>
 #include <asm/arch/sys_proto.h>
+#include <asm/arch/dma.h>
 
 #define	MXS_SPI_MAX_TIMEOUT	1000000
 #define	MXS_SPI_PORT_OFFSET	0x2000
 #define MXS_SSP_CHIPSELECT_MASK		0x00300000
 #define MXS_SSP_CHIPSELECT_SHIFT	20
 
+#define MXSSSP_SMALL_TRANSFER	512
+
+/*
+ * CONFIG_MXS_SPI_DMA_ENABLE: Experimental mixed PIO/DMA support for MXS SPI
+ *                            host. Use with utmost caution!
+ *
+ *                            Enabling this is not yet recommended since this
+ *                            still doesn't support transfers to/from unaligned
+ *                            addresses. Therefore this driver will not work
+ *                            for example with saving environment. This is
+ *                            caused by DMA alignment constraints on MXS.
+ */
+
 struct mxs_spi_slave {
 	struct spi_slave	slave;
 	uint32_t		max_khz;
 	uint32_t		mode;
-	struct mx28_ssp_regs	*regs;
+	struct mxs_ssp_regs	*regs;
 };
 
 static inline struct mxs_spi_slave *to_mxs_slave(struct spi_slave *slave)
@@ -67,7 +81,7 @@ struct spi_slave *spi_setup_slave(unsigned int bus, unsigned int cs,
 {
 	struct mxs_spi_slave *mxs_slave;
 	uint32_t addr;
-	struct mx28_ssp_regs *ssp_regs;
+	struct mxs_ssp_regs *ssp_regs;
 	int reg;
 
 	if (!spi_cs_is_valid(bus, cs)) {
@@ -75,17 +89,20 @@ struct spi_slave *spi_setup_slave(unsigned int bus, unsigned int cs,
 		return NULL;
 	}
 
-	mxs_slave = malloc(sizeof(struct mxs_spi_slave));
+	mxs_slave = calloc(sizeof(struct mxs_spi_slave), 1);
 	if (!mxs_slave)
 		return NULL;
 
+	if (mxs_dma_init_channel(bus))
+		goto err_init;
+
 	addr = MXS_SSP0_BASE + (bus * MXS_SPI_PORT_OFFSET);
 
 	mxs_slave->slave.bus = bus;
 	mxs_slave->slave.cs = cs;
 	mxs_slave->max_khz = max_hz / 1000;
 	mxs_slave->mode = mode;
-	mxs_slave->regs = (struct mx28_ssp_regs *)addr;
+	mxs_slave->regs = (struct mxs_ssp_regs *)addr;
 	ssp_regs = mxs_slave->regs;
 
 	reg = readl(&ssp_regs->hw_ssp_ctrl0);
@@ -94,6 +111,10 @@ struct spi_slave *spi_setup_slave(unsigned int bus, unsigned int cs,
 
 	writel(reg, &ssp_regs->hw_ssp_ctrl0);
 	return &mxs_slave->slave;
+
+err_init:
+	free(mxs_slave);
+	return NULL;
 }
 
 void spi_free_slave(struct spi_slave *slave)
@@ -105,10 +126,10 @@ void spi_free_slave(struct spi_slave *slave)
 int spi_claim_bus(struct spi_slave *slave)
 {
 	struct mxs_spi_slave *mxs_slave = to_mxs_slave(slave);
-	struct mx28_ssp_regs *ssp_regs = mxs_slave->regs;
+	struct mxs_ssp_regs *ssp_regs = mxs_slave->regs;
 	uint32_t reg = 0;
 
-	mx28_reset_block(&ssp_regs->hw_ssp_ctrl0_reg);
+	mxs_reset_block(&ssp_regs->hw_ssp_ctrl0_reg);
 
 	writel(SSP_CTRL0_BUS_WIDTH_ONE_BIT, &ssp_regs->hw_ssp_ctrl0);
 
@@ -128,79 +149,63 @@ void spi_release_bus(struct spi_slave *slave)
 {
 }
 
-static void mxs_spi_start_xfer(struct mx28_ssp_regs *ssp_regs)
+static void mxs_spi_start_xfer(struct mxs_ssp_regs *ssp_regs)
 {
 	writel(SSP_CTRL0_LOCK_CS, &ssp_regs->hw_ssp_ctrl0_set);
 	writel(SSP_CTRL0_IGNORE_CRC, &ssp_regs->hw_ssp_ctrl0_clr);
 }
 
-static void mxs_spi_end_xfer(struct mx28_ssp_regs *ssp_regs)
+static void mxs_spi_end_xfer(struct mxs_ssp_regs *ssp_regs)
 {
 	writel(SSP_CTRL0_LOCK_CS, &ssp_regs->hw_ssp_ctrl0_clr);
 	writel(SSP_CTRL0_IGNORE_CRC, &ssp_regs->hw_ssp_ctrl0_set);
 }
 
-int spi_xfer(struct spi_slave *slave, unsigned int bitlen,
-		const void *dout, void *din, unsigned long flags)
+static int mxs_spi_xfer_pio(struct mxs_spi_slave *slave,
+			char *data, int length, int write, unsigned long flags)
 {
-	struct mxs_spi_slave *mxs_slave = to_mxs_slave(slave);
-	struct mx28_ssp_regs *ssp_regs = mxs_slave->regs;
-	int len = bitlen / 8;
-	const char *tx = dout;
-	char *rx = din;
-	char dummy;
-
-	if (bitlen == 0) {
-		if (flags & SPI_XFER_END) {
-			rx = &dummy;
-			len = 1;
-		} else
-			return 0;
-	}
-
-	if (!rx && !tx)
-		return 0;
+	struct mxs_ssp_regs *ssp_regs = slave->regs;
 
 	if (flags & SPI_XFER_BEGIN)
 		mxs_spi_start_xfer(ssp_regs);
 
-	while (len--) {
+	while (length--) {
 		/* We transfer 1 byte */
 		writel(1, &ssp_regs->hw_ssp_xfer_size);
 
-		if ((flags & SPI_XFER_END) && !len)
+		if ((flags & SPI_XFER_END) && !length)
 			mxs_spi_end_xfer(ssp_regs);
 
-		if (tx)
+		if (write)
 			writel(SSP_CTRL0_READ, &ssp_regs->hw_ssp_ctrl0_clr);
 		else
 			writel(SSP_CTRL0_READ, &ssp_regs->hw_ssp_ctrl0_set);
 
 		writel(SSP_CTRL0_RUN, &ssp_regs->hw_ssp_ctrl0_set);
 
-		if (mx28_wait_mask_set(&ssp_regs->hw_ssp_ctrl0_reg,
+		if (mxs_wait_mask_set(&ssp_regs->hw_ssp_ctrl0_reg,
 			SSP_CTRL0_RUN, MXS_SPI_MAX_TIMEOUT)) {
 			printf("MXS SPI: Timeout waiting for start\n");
 			return -ETIMEDOUT;
 		}
 
-		if (tx)
-			writel(*tx++, &ssp_regs->hw_ssp_data);
+		if (write)
+			writel(*data++, &ssp_regs->hw_ssp_data);
 
 		writel(SSP_CTRL0_DATA_XFER, &ssp_regs->hw_ssp_ctrl0_set);
 
-		if (rx) {
-			if (mx28_wait_mask_clr(&ssp_regs->hw_ssp_status_reg,
+		if (!write) {
+			if (mxs_wait_mask_clr(&ssp_regs->hw_ssp_status_reg,
 				SSP_STATUS_FIFO_EMPTY, MXS_SPI_MAX_TIMEOUT)) {
 				printf("MXS SPI: Timeout waiting for data\n");
 				return -ETIMEDOUT;
 			}
 
-			*rx = readl(&ssp_regs->hw_ssp_data);
-			rx++;
+			*data = readl(&ssp_regs->hw_ssp_data);
+			data++;
 		}
 
-		if (mx28_wait_mask_clr(&ssp_regs->hw_ssp_ctrl0_reg,
+		if (mxs_wait_mask_clr(&ssp_regs->hw_ssp_ctrl0_reg,
 			SSP_CTRL0_RUN, MXS_SPI_MAX_TIMEOUT)) {
 			printf("MXS SPI: Timeout waiting for finish\n");
 			return -ETIMEDOUT;
@@ -209,3 +214,166 @@ int spi_xfer(struct spi_slave *slave, unsigned int bitlen,
 
 	return 0;
 }
+
+static int mxs_spi_xfer_dma(struct mxs_spi_slave *slave,
+			char *data, int length, int write, unsigned long flags)
+{
+	const int xfer_max_sz = 0xff00;
+	const int desc_count = DIV_ROUND_UP(length, xfer_max_sz) + 1;
+	struct mxs_ssp_regs *ssp_regs = slave->regs;
+	struct mxs_dma_desc *dp;
+	uint32_t ctrl0;
+	uint32_t cache_data_count;
+	int dmach;
+	int tl;
+
+	ALLOC_CACHE_ALIGN_BUFFER(struct mxs_dma_desc, desc, desc_count);
+
+	memset(desc, 0, sizeof(struct mxs_dma_desc) * desc_count);
+
+	ctrl0 = readl(&ssp_regs->hw_ssp_ctrl0);
+	ctrl0 |= SSP_CTRL0_DATA_XFER;
+
+	if (flags & SPI_XFER_BEGIN)
+		ctrl0 |= SSP_CTRL0_LOCK_CS;
+	if (!write)
+		ctrl0 |= SSP_CTRL0_READ;
+
+	writel(length, &ssp_regs->hw_ssp_xfer_size);
+
+	if (length % ARCH_DMA_MINALIGN)
+		cache_data_count = roundup(length, ARCH_DMA_MINALIGN);
+	else
+		cache_data_count = length;
+
+	if (write)
+		/* Flush data to DRAM so DMA can pick them up */
+		flush_dcache_range((uint32_t)data,
+			(uint32_t)(data + cache_data_count));
+
+	dmach = MXS_DMA_CHANNEL_AHB_APBH_SSP0 + slave->slave.bus;
+
+	dp = desc;
+	while (length) {
+		dp->address = (dma_addr_t)dp;
+		dp->cmd.address = (dma_addr_t)data;
+
+		/*
+		 * This is correct, even though it does indeed look insane.
+		 * I hereby have to, wholeheartedly, thank Freescale Inc.,
+		 * for always inventing insane hardware and keeping me busy
+		 * and employed ;-)
+		 */
+		if (write)
+			dp->cmd.data = MXS_DMA_DESC_COMMAND_DMA_READ;
+		else
+			dp->cmd.data = MXS_DMA_DESC_COMMAND_DMA_WRITE;
+
+		/*
+		 * The DMA controller can transfer large chunks (64kB) at
+		 * time by setting the transfer length to 0. Setting tl to
+		 * 0x10000 will overflow below and make .data contain 0.
+		 * Otherwise, 0xff00 is the transfer maximum.
+		 */
+		if (length >= 0x10000)
+			tl = 0x10000;
+		else
+			tl = min(length, xfer_max_sz);
+
+		dp->cmd.data |=
+			(tl << MXS_DMA_DESC_BYTES_OFFSET) |
+			(1 << MXS_DMA_DESC_PIO_WORDS_OFFSET) |
+			MXS_DMA_DESC_HALT_ON_TERMINATE |
+			MXS_DMA_DESC_TERMINATE_FLUSH;
+		dp->cmd.pio_words[0] = ctrl0;
+
+		data += tl;
+		length -= tl;
+
+		mxs_dma_desc_append(dmach, dp);
+
+		dp++;
+	}
+
+	dp->address = (dma_addr_t)dp;
+	dp->cmd.address = (dma_addr_t)0;
+	dp->cmd.data = MXS_DMA_DESC_COMMAND_NO_DMAXFER |
+			(1 << MXS_DMA_DESC_PIO_WORDS_OFFSET) |
+			MXS_DMA_DESC_IRQ | MXS_DMA_DESC_DEC_SEM;
+	if (flags & SPI_XFER_END) {
+		ctrl0 &= ~SSP_CTRL0_LOCK_CS;
+		dp->cmd.pio_words[0] = ctrl0 | SSP_CTRL0_IGNORE_CRC;
+	}
+	mxs_dma_desc_append(dmach, dp);
+
+	if (mxs_dma_go(dmach))
+		return -EINVAL;
+
+	/* The data arrived into DRAM, invalidate cache over them */
+	if (!write) {
+		invalidate_dcache_range((uint32_t)data,
+			(uint32_t)(data + cache_data_count));
+	}
+
+	return 0;
+}
+
+int spi_xfer(struct spi_slave *slave, unsigned int bitlen,
+		const void *dout, void *din, unsigned long flags)
+{
+	struct mxs_spi_slave *mxs_slave = to_mxs_slave(slave);
+	struct mxs_ssp_regs *ssp_regs = mxs_slave->regs;
+	int len = bitlen / 8;
+	char dummy;
+	int write = 0;
+	char *data = NULL;
+
+#ifdef CONFIG_MXS_SPI_DMA_ENABLE
+	int dma = 1;
+#else
+	int dma = 0;
+#endif
+
+	if (bitlen == 0) {
+		if (flags & SPI_XFER_END) {
+			din = (void *)&dummy;
+			len = 1;
+		} else
+			return 0;
+	}
+
+	/* Half-duplex only */
+	if (din && dout)
+		return -EINVAL;
+	/* No data */
+	if (!din && !dout)
+		return 0;
+
+	if (dout) {
+		data = (char *)dout;
+		write = 1;
+	} else if (din) {
+		data = (char *)din;
+		write = 0;
+	}
+
+	/*
+	 * Check for alignment, if the buffer is aligned, do DMA transfer,
+	 * PIO otherwise. This is a temporary workaround until proper bounce
+	 * buffer is in place.
+	 */
+	if (dma) {
+		if (((uint32_t)data) & (ARCH_DMA_MINALIGN - 1))
+			dma = 0;
+		if (((uint32_t)len) & (ARCH_DMA_MINALIGN - 1))
+			dma = 0;
+	}
+
+	if (!dma || (len < MXSSSP_SMALL_TRANSFER)) {
+		writel(SSP_CTRL1_DMA_ENABLE, &ssp_regs->hw_ssp_ctrl1_clr);
+		return mxs_spi_xfer_pio(mxs_slave, data, len, write, flags);
+	} else {
+		writel(SSP_CTRL1_DMA_ENABLE, &ssp_regs->hw_ssp_ctrl1_set);
+		return mxs_spi_xfer_dma(mxs_slave, data, len, write, flags);
+	}
+}