spi/rspi: add dmaengine support

This patch adds dmaengine supporting using sh_dma driver. The module
receives data by DMAC, it also needs TX DMAC to generate SPI's clocks.

Signed-off-by: Yoshihiro Shimoda <[e-mail]>
Signed-off-by: Grant Likely <[e-mail]>

drivers/spi/spi-rspi.c

···
 #include <linux/platform_device.h>
 #include <linux/io.h>
 #include <linux/clk.h>
+#include <linux/dmaengine.h>
+#include <linux/dma-mapping.h>
+#include <linux/sh_dma.h>
 #include <linux/spi/spi.h>
+#include <linux/spi/rspi.h>
 
 #define RSPI_SPCR		0x00
 #define RSPI_SSLP		0x01
···
 	spinlock_t lock;
 	struct clk *clk;
 	unsigned char spsr;
+
+	/* for dmaengine */
+	struct sh_dmae_slave dma_tx;
+	struct sh_dmae_slave dma_rx;
+	struct dma_chan *chan_tx;
+	struct dma_chan *chan_rx;
+	int irq;
+
+	unsigned dma_width_16bit:1;
+	unsigned dma_callbacked:1;
 };
 
 static void rspi_write8(struct rspi_data *rspi, u8 data, u16 offset)
···
 	return 0;
 }
 
-static int rspi_receive_pio(struct rspi_data *rspi, struct spi_message *mesg,
-			    struct spi_transfer *t)
+static void rspi_dma_complete(void *arg)
 {
-	int remain = t->len;
-	u8 *data;
+	struct rspi_data *rspi = arg;
+
+	rspi->dma_callbacked = 1;
+	wake_up_interruptible(&rspi->wait);
+}
+
+static int rspi_dma_map_sg(struct scatterlist *sg, void *buf, unsigned len,
+			   struct dma_chan *chan,
+			   enum dma_transfer_direction dir)
+{
+	sg_init_table(sg, 1);
+	sg_set_buf(sg, buf, len);
+	sg_dma_len(sg) = len;
+	return dma_map_sg(chan->device->dev, sg, 1, dir);
+}
+
+static void rspi_dma_unmap_sg(struct scatterlist *sg, struct dma_chan *chan,
+			      enum dma_transfer_direction dir)
+{
+	dma_unmap_sg(chan->device->dev, sg, 1, dir);
+}
+
+static void rspi_memory_to_8bit(void *buf, const void *data, unsigned len)
+{
+	u16 *dst = buf;
+	const u8 *src = data;
+
+	while (len) {
+		*dst++ = (u16)(*src++);
+		len--;
+	}
+}
+
+static void rspi_memory_from_8bit(void *buf, const void *data, unsigned len)
+{
+	u8 *dst = buf;
+	const u16 *src = data;
+
+	while (len) {
+		*dst++ = (u8)*src++;
+		len--;
+	}
+}
+
+static int rspi_send_dma(struct rspi_data *rspi, struct spi_transfer *t)
+{
+	struct scatterlist sg;
+	void *buf = NULL;
+	struct dma_async_tx_descriptor *desc;
+	unsigned len;
+	int ret = 0;
+
+	if (rspi->dma_width_16bit) {
+		/*
+		 * If DMAC bus width is 16-bit, the driver allocates a dummy
+		 * buffer. And, the driver converts original data into the
+		 * DMAC data as the following format:
+		 *  original data: 1st byte, 2nd byte ...
+		 *  DMAC data:     1st byte, dummy, 2nd byte, dummy ...
+		 */
+		len = t->len * 2;
+		buf = kmalloc(len, GFP_KERNEL);
+		if (!buf)
+			return -ENOMEM;
+		rspi_memory_to_8bit(buf, t->tx_buf, t->len);
+	} else {
+		len = t->len;
+		buf = (void *)t->tx_buf;
+	}
+
+	if (!rspi_dma_map_sg(&sg, buf, len, rspi->chan_tx, DMA_TO_DEVICE)) {
+		ret = -EFAULT;
+		goto end_nomap;
+	}
+	desc = dmaengine_prep_slave_sg(rspi->chan_tx, &sg, 1, DMA_TO_DEVICE,
+				       DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
+	if (!desc) {
+		ret = -EIO;
+		goto end;
+	}
+
+	/*
+	 * DMAC needs SPTIE, but if SPTIE is set, this IRQ routine will be
+	 * called. So, this driver disables the IRQ while DMA transfer.
+	 */
+	disable_irq(rspi->irq);
+
+	rspi_write8(rspi, rspi_read8(rspi, RSPI_SPCR) | SPCR_TXMD, RSPI_SPCR);
+	rspi_enable_irq(rspi, SPCR_SPTIE);
+	rspi->dma_callbacked = 0;
+
+	desc->callback = rspi_dma_complete;
+	desc->callback_param = rspi;
+	dmaengine_submit(desc);
+	dma_async_issue_pending(rspi->chan_tx);
+
+	ret = wait_event_interruptible_timeout(rspi->wait,
+					       rspi->dma_callbacked, HZ);
+	if (ret > 0 && rspi->dma_callbacked)
+		ret = 0;
+	else if (!ret)
+		ret = -ETIMEDOUT;
+	rspi_disable_irq(rspi, SPCR_SPTIE);
+
+	enable_irq(rspi->irq);
+
+end:
+	rspi_dma_unmap_sg(&sg, rspi->chan_tx, DMA_TO_DEVICE);
+end_nomap:
+	if (rspi->dma_width_16bit)
+		kfree(buf);
+
+	return ret;
+}
+
+static void rspi_receive_init(struct rspi_data *rspi)
+{
 	unsigned char spsr;
 
 	spsr = rspi_read8(rspi, RSPI_SPSR);
···
 	if (spsr & SPSR_OVRF)
 		rspi_write8(rspi, rspi_read8(rspi, RSPI_SPSR) & ~SPSR_OVRF,
 			    RSPI_SPCR);
+}
+
+static int rspi_receive_pio(struct rspi_data *rspi, struct spi_message *mesg,
+			    struct spi_transfer *t)
+{
+	int remain = t->len;
+	u8 *data;
+
+	rspi_receive_init(rspi);
 
 	data = (u8 *)t->rx_buf;
 	while (remain > 0) {
···
 	return 0;
 }
 
+static int rspi_receive_dma(struct rspi_data *rspi, struct spi_transfer *t)
+{
+	struct scatterlist sg, sg_dummy;
+	void *dummy = NULL, *rx_buf = NULL;
+	struct dma_async_tx_descriptor *desc, *desc_dummy;
+	unsigned len;
+	int ret = 0;
+
+	if (rspi->dma_width_16bit) {
+		/*
+		 * If DMAC bus width is 16-bit, the driver allocates a dummy
+		 * buffer. And, finally the driver converts the DMAC data into
+		 * actual data as the following format:
+		 *  DMAC data:   1st byte, dummy, 2nd byte, dummy ...
+		 *  actual data: 1st byte, 2nd byte ...
+		 */
+		len = t->len * 2;
+		rx_buf = kmalloc(len, GFP_KERNEL);
+		if (!rx_buf)
+			return -ENOMEM;
+	 } else {
+		len = t->len;
+		rx_buf = t->rx_buf;
+	}
+
+	/* prepare dummy transfer to generate SPI clocks */
+	dummy = kzalloc(len, GFP_KERNEL);
+	if (!dummy) {
+		ret = -ENOMEM;
+		goto end_nomap;
+	}
+	if (!rspi_dma_map_sg(&sg_dummy, dummy, len, rspi->chan_tx,
+			     DMA_TO_DEVICE)) {
+		ret = -EFAULT;
+		goto end_nomap;
+	}
+	desc_dummy = dmaengine_prep_slave_sg(rspi->chan_tx, &sg_dummy, 1,
+			DMA_TO_DEVICE, DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
+	if (!desc_dummy) {
+		ret = -EIO;
+		goto end_dummy_mapped;
+	}
+
+	/* prepare receive transfer */
+	if (!rspi_dma_map_sg(&sg, rx_buf, len, rspi->chan_rx,
+			     DMA_FROM_DEVICE)) {
+		ret = -EFAULT;
+		goto end_dummy_mapped;
+
+	}
+	desc = dmaengine_prep_slave_sg(rspi->chan_rx, &sg, 1, DMA_FROM_DEVICE,
+				       DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
+	if (!desc) {
+		ret = -EIO;
+		goto end;
+	}
+
+	rspi_receive_init(rspi);
+
+	/*
+	 * DMAC needs SPTIE, but if SPTIE is set, this IRQ routine will be
+	 * called. So, this driver disables the IRQ while DMA transfer.
+	 */
+	disable_irq(rspi->irq);
+
+	rspi_write8(rspi, rspi_read8(rspi, RSPI_SPCR) & ~SPCR_TXMD, RSPI_SPCR);
+	rspi_enable_irq(rspi, SPCR_SPTIE | SPCR_SPRIE);
+	rspi->dma_callbacked = 0;
+
+	desc->callback = rspi_dma_complete;
+	desc->callback_param = rspi;
+	dmaengine_submit(desc);
+	dma_async_issue_pending(rspi->chan_rx);
+
+	desc_dummy->callback = NULL;	/* No callback */
+	dmaengine_submit(desc_dummy);
+	dma_async_issue_pending(rspi->chan_tx);
+
+	ret = wait_event_interruptible_timeout(rspi->wait,
+					       rspi->dma_callbacked, HZ);
+	if (ret > 0 && rspi->dma_callbacked)
+		ret = 0;
+	else if (!ret)
+		ret = -ETIMEDOUT;
+	rspi_disable_irq(rspi, SPCR_SPTIE | SPCR_SPRIE);
+
+	enable_irq(rspi->irq);
+
+end:
+	rspi_dma_unmap_sg(&sg, rspi->chan_rx, DMA_FROM_DEVICE);
+end_dummy_mapped:
+	rspi_dma_unmap_sg(&sg_dummy, rspi->chan_tx, DMA_TO_DEVICE);
+end_nomap:
+	if (rspi->dma_width_16bit) {
+		if (!ret)
+			rspi_memory_from_8bit(t->rx_buf, rx_buf, t->len);
+		kfree(rx_buf);
+	}
+	kfree(dummy);
+
+	return ret;
+}
+
+static int rspi_is_dma(struct rspi_data *rspi, struct spi_transfer *t)
+{
+	if (t->tx_buf && rspi->chan_tx)
+		return 1;
+	/* If the module receives data by DMAC, it also needs TX DMAC */
+	if (t->rx_buf && rspi->chan_tx && rspi->chan_rx)
+		return 1;
+
+	return 0;
+}
+
 static void rspi_work(struct work_struct *work)
 {
 	struct rspi_data *rspi = container_of(work, struct rspi_data, ws);
···
 
 		list_for_each_entry(t, &mesg->transfers, transfer_list) {
 			if (t->tx_buf) {
-				ret = rspi_send_pio(rspi, mesg, t);
+				if (rspi_is_dma(rspi, t))
+					ret = rspi_send_dma(rspi, t);
+				else
+					ret = rspi_send_pio(rspi, mesg, t);
 				if (ret < 0)
 					goto error;
 			}
 			if (t->rx_buf) {
-				ret = rspi_receive_pio(rspi, mesg, t);
+				if (rspi_is_dma(rspi, t))
+					ret = rspi_receive_dma(rspi, t);
+				else
+					ret = rspi_receive_pio(rspi, mesg, t);
 				if (ret < 0)
 					goto error;
 			}
···
 	return ret;
 }
 
+static bool rspi_filter(struct dma_chan *chan, void *filter_param)
+{
+	chan->private = filter_param;
+	return true;
+}
+
+static void __devinit rspi_request_dma(struct rspi_data *rspi,
+				       struct platform_device *pdev)
+{
+	struct rspi_plat_data *rspi_pd = pdev->dev.platform_data;
+	dma_cap_mask_t mask;
+
+	if (!rspi_pd)
+		return;
+
+	rspi->dma_width_16bit = rspi_pd->dma_width_16bit;
+
+	/* If the module receives data by DMAC, it also needs TX DMAC */
+	if (rspi_pd->dma_rx_id && rspi_pd->dma_tx_id) {
+		dma_cap_zero(mask);
+		dma_cap_set(DMA_SLAVE, mask);
+		rspi->dma_rx.slave_id = rspi_pd->dma_rx_id;
+		rspi->chan_rx = dma_request_channel(mask, rspi_filter,
+						    &rspi->dma_rx);
+		if (rspi->chan_rx)
+			dev_info(&pdev->dev, "Use DMA when rx.\n");
+	}
+	if (rspi_pd->dma_tx_id) {
+		dma_cap_zero(mask);
+		dma_cap_set(DMA_SLAVE, mask);
+		rspi->dma_tx.slave_id = rspi_pd->dma_tx_id;
+		rspi->chan_tx = dma_request_channel(mask, rspi_filter,
+						    &rspi->dma_tx);
+		if (rspi->chan_tx)
+			dev_info(&pdev->dev, "Use DMA when tx\n");
+	}
+}
+
+static void __devexit rspi_release_dma(struct rspi_data *rspi)
+{
+	if (rspi->chan_tx)
+		dma_release_channel(rspi->chan_tx);
+	if (rspi->chan_rx)
+		dma_release_channel(rspi->chan_rx);
+}
+
 static int __devexit rspi_remove(struct platform_device *pdev)
 {
 	struct rspi_data *rspi = dev_get_drvdata(&pdev->dev);
 
 	spi_unregister_master(rspi->master);
+	rspi_release_dma(rspi);
 	free_irq(platform_get_irq(pdev, 0), rspi);
 	clk_put(rspi->clk);
 	iounmap(rspi->addr);
···
 		goto error3;
 	}
 
+	rspi->irq = irq;
+	rspi_request_dma(rspi, pdev);
+
 	ret = spi_register_master(master);
 	if (ret < 0) {
 		dev_err(&pdev->dev, "spi_register_master error.\n");
···
 	return 0;
 
 error4:
+	rspi_release_dma(rspi);
 	free_irq(irq, rspi);
 error3:
 	clk_put(rspi->clk);

include/linux/spi/rspi.h

···
+/*
+ * Renesas SPI driver
+ *
+ * Copyright (C) 2012  Renesas Solutions Corp.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
+ *
+ */
+
+#ifndef __LINUX_SPI_RENESAS_SPI_H__
+#define __LINUX_SPI_RENESAS_SPI_H__
+
+struct rspi_plat_data {
+	unsigned int dma_tx_id;
+	unsigned int dma_rx_id;
+
+	unsigned dma_width_16bit:1;	/* DMAC read/write width = 16-bit */
+};
+
+#endif