4 files changed, 898 insertions, 248 deletions

diff --git a/drivers/dma/shdma.c b/drivers/dma/shdma.c
index b75ce8b84c46..5d17e09cb625 100644
--- a/drivers/dma/shdma.c
+++ b/drivers/dma/shdma.c
@@ -24,8 +24,10 @@
 #include <linux/delay.h>
 #include <linux/dma-mapping.h>
 #include <linux/platform_device.h>
-#include <cpu/dma.h>
-#include <asm/dma-sh.h>
+#include <linux/pm_runtime.h>
+
+#include <asm/dmaengine.h>
+
 #include "shdma.h"
 
 /* DMA descriptor control */
@@ -38,30 +40,32 @@ enum sh_dmae_desc_status {
 };
 
 #define NR_DESCS_PER_CHANNEL 32
-/*
- * Define the default configuration for dual address memory-memory transfer.
- * The 0x400 value represents auto-request, external->external.
- *
- * And this driver set 4byte burst mode.
- * If you want to change mode, you need to change RS_DEFAULT of value.
- * (ex 1byte burst mode -> (RS_DUAL & ~TS_32)
- */
-#define RS_DEFAULT  (RS_DUAL)
+/* Default MEMCPY transfer size = 2^2 = 4 bytes */
+#define LOG2_DEFAULT_XFER_SIZE	2
 
 /* A bitmask with bits enough for enum sh_dmae_slave_chan_id */
 static unsigned long sh_dmae_slave_used[BITS_TO_LONGS(SHDMA_SLAVE_NUMBER)];
 
 static void sh_dmae_chan_ld_cleanup(struct sh_dmae_chan *sh_chan, bool all);
 
-#define SH_DMAC_CHAN_BASE(id) (dma_base_addr[id])
 static void sh_dmae_writel(struct sh_dmae_chan *sh_dc, u32 data, u32 reg)
 {
-	ctrl_outl(data, SH_DMAC_CHAN_BASE(sh_dc->id) + reg);
+	__raw_writel(data, sh_dc->base + reg / sizeof(u32));
 }
 
 static u32 sh_dmae_readl(struct sh_dmae_chan *sh_dc, u32 reg)
 {
-	return ctrl_inl(SH_DMAC_CHAN_BASE(sh_dc->id) + reg);
+	return __raw_readl(sh_dc->base + reg / sizeof(u32));
+}
+
+static u16 dmaor_read(struct sh_dmae_device *shdev)
+{
+	return __raw_readw(shdev->chan_reg + DMAOR / sizeof(u32));
+}
+
+static void dmaor_write(struct sh_dmae_device *shdev, u16 data)
+{
+	__raw_writew(data, shdev->chan_reg + DMAOR / sizeof(u32));
 }
 
 /*
@@ -69,24 +73,23 @@ static u32 sh_dmae_readl(struct sh_dmae_chan *sh_dc, u32 reg)
  *
  * SH7780 has two DMAOR register
  */
-static void sh_dmae_ctl_stop(int id)
+static void sh_dmae_ctl_stop(struct sh_dmae_device *shdev)
 {
-	unsigned short dmaor = dmaor_read_reg(id);
+	unsigned short dmaor = dmaor_read(shdev);
 
-	dmaor &= ~(DMAOR_NMIF | DMAOR_AE);
-	dmaor_write_reg(id, dmaor);
+	dmaor_write(shdev, dmaor & ~(DMAOR_NMIF | DMAOR_AE | DMAOR_DME));
 }
 
-static int sh_dmae_rst(int id)
+static int sh_dmae_rst(struct sh_dmae_device *shdev)
 {
 	unsigned short dmaor;
 
-	sh_dmae_ctl_stop(id);
-	dmaor = dmaor_read_reg(id) | DMAOR_INIT;
+	sh_dmae_ctl_stop(shdev);
+	dmaor = dmaor_read(shdev) | shdev->pdata->dmaor_init;
 
-	dmaor_write_reg(id, dmaor);
-	if (dmaor_read_reg(id) & (DMAOR_AE | DMAOR_NMIF)) {
-		pr_warning(KERN_ERR "dma-sh: Can't initialize DMAOR.\n");
+	dmaor_write(shdev, dmaor);
+	if (dmaor_read(shdev) & (DMAOR_AE | DMAOR_NMIF)) {
+		pr_warning("dma-sh: Can't initialize DMAOR.\n");
 		return -EINVAL;
 	}
 	return 0;
@@ -102,13 +105,36 @@ static bool dmae_is_busy(struct sh_dmae_chan *sh_chan)
 	return false; /* waiting */
 }
 
-static unsigned int ts_shift[] = TS_SHIFT;
-static inline unsigned int calc_xmit_shift(u32 chcr)
+static unsigned int calc_xmit_shift(struct sh_dmae_chan *sh_chan, u32 chcr)
 {
-	int cnt = ((chcr & CHCR_TS_LOW_MASK) >> CHCR_TS_LOW_SHIFT) |
-		((chcr & CHCR_TS_HIGH_MASK) >> CHCR_TS_HIGH_SHIFT);
+	struct sh_dmae_device *shdev = container_of(sh_chan->common.device,
+						struct sh_dmae_device, common);
+	struct sh_dmae_pdata *pdata = shdev->pdata;
+	int cnt = ((chcr & pdata->ts_low_mask) >> pdata->ts_low_shift) |
+		((chcr & pdata->ts_high_mask) >> pdata->ts_high_shift);
+
+	if (cnt >= pdata->ts_shift_num)
+		cnt = 0;
 
-	return ts_shift[cnt];
+	return pdata->ts_shift[cnt];
+}
+
+static u32 log2size_to_chcr(struct sh_dmae_chan *sh_chan, int l2size)
+{
+	struct sh_dmae_device *shdev = container_of(sh_chan->common.device,
+						struct sh_dmae_device, common);
+	struct sh_dmae_pdata *pdata = shdev->pdata;
+	int i;
+
+	for (i = 0; i < pdata->ts_shift_num; i++)
+		if (pdata->ts_shift[i] == l2size)
+			break;
+
+	if (i == pdata->ts_shift_num)
+		i = 0;
+
+	return ((i << pdata->ts_low_shift) & pdata->ts_low_mask) |
+		((i << pdata->ts_high_shift) & pdata->ts_high_mask);
 }
 
 static void dmae_set_reg(struct sh_dmae_chan *sh_chan, struct sh_dmae_regs *hw)
@@ -136,8 +162,13 @@ static void dmae_halt(struct sh_dmae_chan *sh_chan)
 
 static void dmae_init(struct sh_dmae_chan *sh_chan)
 {
-	u32 chcr = RS_DEFAULT; /* default is DUAL mode */
-	sh_chan->xmit_shift = calc_xmit_shift(chcr);
+	/*
+	 * Default configuration for dual address memory-memory transfer.
+	 * 0x400 represents auto-request.
+	 */
+	u32 chcr = DM_INC | SM_INC | 0x400 | log2size_to_chcr(sh_chan,
+						   LOG2_DEFAULT_XFER_SIZE);
+	sh_chan->xmit_shift = calc_xmit_shift(sh_chan, chcr);
 	sh_dmae_writel(sh_chan, chcr, CHCR);
 }
 
@@ -147,37 +178,26 @@ static int dmae_set_chcr(struct sh_dmae_chan *sh_chan, u32 val)
 	if (dmae_is_busy(sh_chan))
 		return -EBUSY;
 
-	sh_chan->xmit_shift = calc_xmit_shift(val);
+	sh_chan->xmit_shift = calc_xmit_shift(sh_chan, val);
 	sh_dmae_writel(sh_chan, val, CHCR);
 
 	return 0;
 }
 
-#define DMARS_SHIFT	8
-#define DMARS_CHAN_MSK	0x01
 static int dmae_set_dmars(struct sh_dmae_chan *sh_chan, u16 val)
 {
-	u32 addr;
-	int shift = 0;
+	struct sh_dmae_device *shdev = container_of(sh_chan->common.device,
+						struct sh_dmae_device, common);
+	struct sh_dmae_pdata *pdata = shdev->pdata;
+	struct sh_dmae_channel *chan_pdata = &pdata->channel[sh_chan->id];
+	u16 __iomem *addr = shdev->dmars + chan_pdata->dmars / sizeof(u16);
+	int shift = chan_pdata->dmars_bit;
 
 	if (dmae_is_busy(sh_chan))
 		return -EBUSY;
 
-	if (sh_chan->id & DMARS_CHAN_MSK)
-		shift = DMARS_SHIFT;
-
-	if (sh_chan->id < 6)
-		/* DMA0RS0 - DMA0RS2 */
-		addr = SH_DMARS_BASE0 + (sh_chan->id / 2) * 4;
-#ifdef SH_DMARS_BASE1
-	else if (sh_chan->id < 12)
-		/* DMA1RS0 - DMA1RS2 */
-		addr = SH_DMARS_BASE1 + ((sh_chan->id - 6) / 2) * 4;
-#endif
-	else
-		return -EINVAL;
-
-	ctrl_outw((val << shift) | (ctrl_inw(addr) & (0xFF00 >> shift)), addr);
+	__raw_writew((__raw_readw(addr) & (0xff00 >> shift)) | (val << shift),
+		     addr);
 
 	return 0;
 }
@@ -251,15 +271,15 @@ static struct sh_dmae_slave_config *sh_dmae_find_slave(
 	struct dma_device *dma_dev = sh_chan->common.device;
 	struct sh_dmae_device *shdev = container_of(dma_dev,
 					struct sh_dmae_device, common);
-	struct sh_dmae_pdata *pdata = &shdev->pdata;
+	struct sh_dmae_pdata *pdata = shdev->pdata;
 	int i;
 
 	if ((unsigned)slave_id >= SHDMA_SLAVE_NUMBER)
 		return NULL;
 
-	for (i = 0; i < pdata->config_num; i++)
-		if (pdata->config[i].slave_id == slave_id)
-			return pdata->config + i;
+	for (i = 0; i < pdata->slave_num; i++)
+		if (pdata->slave[i].slave_id == slave_id)
+			return pdata->slave + i;
 
 	return NULL;
 }
@@ -270,6 +290,8 @@ static int sh_dmae_alloc_chan_resources(struct dma_chan *chan)
 	struct sh_desc *desc;
 	struct sh_dmae_slave *param = chan->private;
 
+	pm_runtime_get_sync(sh_chan->dev);
+
 	/*
 	 * This relies on the guarantee from dmaengine that alloc_chan_resources
 	 * never runs concurrently with itself or free_chan_resources.
@@ -288,9 +310,8 @@ static int sh_dmae_alloc_chan_resources(struct dma_chan *chan)
 
 		dmae_set_dmars(sh_chan, cfg->mid_rid);
 		dmae_set_chcr(sh_chan, cfg->chcr);
-	} else {
-		if ((sh_dmae_readl(sh_chan, CHCR) & 0x700) != 0x400)
-			dmae_set_chcr(sh_chan, RS_DEFAULT);
+	} else if ((sh_dmae_readl(sh_chan, CHCR) & 0xf00) != 0x400) {
+		dmae_init(sh_chan);
 	}
 
 	spin_lock_bh(&sh_chan->desc_lock);
@@ -312,6 +333,9 @@ static int sh_dmae_alloc_chan_resources(struct dma_chan *chan)
 	}
 	spin_unlock_bh(&sh_chan->desc_lock);
 
+	if (!sh_chan->descs_allocated)
+		pm_runtime_put(sh_chan->dev);
+
 	return sh_chan->descs_allocated;
 }
 
@@ -323,6 +347,7 @@ static void sh_dmae_free_chan_resources(struct dma_chan *chan)
 	struct sh_dmae_chan *sh_chan = to_sh_chan(chan);
 	struct sh_desc *desc, *_desc;
 	LIST_HEAD(list);
+	int descs = sh_chan->descs_allocated;
 
 	dmae_halt(sh_chan);
 
@@ -343,6 +368,9 @@ static void sh_dmae_free_chan_resources(struct dma_chan *chan)
 
 	spin_unlock_bh(&sh_chan->desc_lock);
 
+	if (descs > 0)
+		pm_runtime_put(sh_chan->dev);
+
 	list_for_each_entry_safe(desc, _desc, &list, node)
 		kfree(desc);
 }
@@ -559,6 +587,19 @@ static void sh_dmae_terminate_all(struct dma_chan *chan)
 	if (!chan)
 		return;
 
+	dmae_halt(sh_chan);
+
+	spin_lock_bh(&sh_chan->desc_lock);
+	if (!list_empty(&sh_chan->ld_queue)) {
+		/* Record partial transfer */
+		struct sh_desc *desc = list_entry(sh_chan->ld_queue.next,
+						  struct sh_desc, node);
+		desc->partial = (desc->hw.tcr - sh_dmae_readl(sh_chan, TCR)) <<
+			sh_chan->xmit_shift;
+
+	}
+	spin_unlock_bh(&sh_chan->desc_lock);
+
 	sh_dmae_chan_ld_cleanup(sh_chan, true);
 }
 
@@ -661,7 +702,7 @@ static void sh_dmae_chan_ld_cleanup(struct sh_dmae_chan *sh_chan, bool all)
 
 static void sh_chan_xfer_ld_queue(struct sh_dmae_chan *sh_chan)
 {
-	struct sh_desc *sd;
+	struct sh_desc *desc;
 
 	spin_lock_bh(&sh_chan->desc_lock);
 	/* DMA work check */
@@ -671,10 +712,13 @@ static void sh_chan_xfer_ld_queue(struct sh_dmae_chan *sh_chan)
 	}
 
 	/* Find the first not transferred desciptor */
-	list_for_each_entry(sd, &sh_chan->ld_queue, node)
-		if (sd->mark == DESC_SUBMITTED) {
+	list_for_each_entry(desc, &sh_chan->ld_queue, node)
+		if (desc->mark == DESC_SUBMITTED) {
+			dev_dbg(sh_chan->dev, "Queue #%d to %d: %u@%x -> %x\n",
+				desc->async_tx.cookie, sh_chan->id,
+				desc->hw.tcr, desc->hw.sar, desc->hw.dar);
 			/* Get the ld start address from ld_queue */
-			dmae_set_reg(sh_chan, &sd->hw);
+			dmae_set_reg(sh_chan, &desc->hw);
 			dmae_start(sh_chan);
 			break;
 		}
@@ -696,6 +740,7 @@ static enum dma_status sh_dmae_is_complete(struct dma_chan *chan,
 	struct sh_dmae_chan *sh_chan = to_sh_chan(chan);
 	dma_cookie_t last_used;
 	dma_cookie_t last_complete;
+	enum dma_status status;
 
 	sh_dmae_chan_ld_cleanup(sh_chan, false);
 
@@ -709,7 +754,27 @@ static enum dma_status sh_dmae_is_complete(struct dma_chan *chan,
 	if (used)
 		*used = last_used;
 
-	return dma_async_is_complete(cookie, last_complete, last_used);
+	spin_lock_bh(&sh_chan->desc_lock);
+
+	status = dma_async_is_complete(cookie, last_complete, last_used);
+
+	/*
+	 * If we don't find cookie on the queue, it has been aborted and we have
+	 * to report error
+	 */
+	if (status != DMA_SUCCESS) {
+		struct sh_desc *desc;
+		status = DMA_ERROR;
+		list_for_each_entry(desc, &sh_chan->ld_queue, node)
+			if (desc->cookie == cookie) {
+				status = DMA_IN_PROGRESS;
+				break;
+			}
+	}
+
+	spin_unlock_bh(&sh_chan->desc_lock);
+
+	return status;
 }
 
 static irqreturn_t sh_dmae_interrupt(int irq, void *data)
@@ -732,40 +797,32 @@ static irqreturn_t sh_dmae_interrupt(int irq, void *data)
 #if defined(CONFIG_CPU_SH4)
 static irqreturn_t sh_dmae_err(int irq, void *data)
 {
-	int err = 0;
 	struct sh_dmae_device *shdev = (struct sh_dmae_device *)data;
+	int i;
 
-	/* IRQ Multi */
-	if (shdev->pdata.mode & SHDMA_MIX_IRQ) {
-		int __maybe_unused cnt = 0;
-		switch (irq) {
-#if defined(DMTE6_IRQ) && defined(DMAE1_IRQ)
-		case DMTE6_IRQ:
-			cnt++;
-#endif
-		case DMTE0_IRQ:
-			if (dmaor_read_reg(cnt) & (DMAOR_NMIF | DMAOR_AE)) {
-				disable_irq(irq);
-				return IRQ_HANDLED;
+	/* halt the dma controller */
+	sh_dmae_ctl_stop(shdev);
+
+	/* We cannot detect, which channel caused the error, have to reset all */
+	for (i = 0; i < SH_DMAC_MAX_CHANNELS; i++) {
+		struct sh_dmae_chan *sh_chan = shdev->chan[i];
+		if (sh_chan) {
+			struct sh_desc *desc;
+			/* Stop the channel */
+			dmae_halt(sh_chan);
+			/* Complete all  */
+			list_for_each_entry(desc, &sh_chan->ld_queue, node) {
+				struct dma_async_tx_descriptor *tx = &desc->async_tx;
+				desc->mark = DESC_IDLE;
+				if (tx->callback)
+					tx->callback(tx->callback_param);
 			}
-		default:
-			return IRQ_NONE;
+			list_splice_init(&sh_chan->ld_queue, &sh_chan->ld_free);
 		}
-	} else {
-		/* reset dma controller */
-		err = sh_dmae_rst(0);
-		if (err)
-			return err;
-#ifdef SH_DMAC_BASE1
-		if (shdev->pdata.mode & SHDMA_DMAOR1) {
-			err = sh_dmae_rst(1);
-			if (err)
-				return err;
-		}
-#endif
-		disable_irq(irq);
-		return IRQ_HANDLED;
 	}
+	sh_dmae_rst(shdev);
+
+	return IRQ_HANDLED;
 }
 #endif
 
@@ -796,19 +853,12 @@ static void dmae_do_tasklet(unsigned long data)
 	sh_dmae_chan_ld_cleanup(sh_chan, false);
 }
 
-static unsigned int get_dmae_irq(unsigned int id)
-{
-	unsigned int irq = 0;
-	if (id < ARRAY_SIZE(dmte_irq_map))
-		irq = dmte_irq_map[id];
-	return irq;
-}
-
-static int __devinit sh_dmae_chan_probe(struct sh_dmae_device *shdev, int id)
+static int __devinit sh_dmae_chan_probe(struct sh_dmae_device *shdev, int id,
+					int irq, unsigned long flags)
 {
 	int err;
-	unsigned int irq = get_dmae_irq(id);
-	unsigned long irqflags = IRQF_DISABLED;
+	struct sh_dmae_channel *chan_pdata = &shdev->pdata->channel[id];
+	struct platform_device *pdev = to_platform_device(shdev->common.dev);
 	struct sh_dmae_chan *new_sh_chan;
 
 	/* alloc channel */
@@ -819,8 +869,13 @@ static int __devinit sh_dmae_chan_probe(struct sh_dmae_device *shdev, int id)
 		return -ENOMEM;
 	}
 
+	/* copy struct dma_device */
+	new_sh_chan->common.device = &shdev->common;
+
 	new_sh_chan->dev = shdev->common.dev;
 	new_sh_chan->id = id;
+	new_sh_chan->irq = irq;
+	new_sh_chan->base = shdev->chan_reg + chan_pdata->offset / sizeof(u32);
 
 	/* Init DMA tasklet */
 	tasklet_init(&new_sh_chan->tasklet, dmae_do_tasklet,
@@ -835,29 +890,20 @@ static int __devinit sh_dmae_chan_probe(struct sh_dmae_device *shdev, int id)
 	INIT_LIST_HEAD(&new_sh_chan->ld_queue);
 	INIT_LIST_HEAD(&new_sh_chan->ld_free);
 
-	/* copy struct dma_device */
-	new_sh_chan->common.device = &shdev->common;
-
 	/* Add the channel to DMA device channel list */
 	list_add_tail(&new_sh_chan->common.device_node,
 			&shdev->common.channels);
 	shdev->common.chancnt++;
 
-	if (shdev->pdata.mode & SHDMA_MIX_IRQ) {
-		irqflags = IRQF_SHARED;
-#if defined(DMTE6_IRQ)
-		if (irq >= DMTE6_IRQ)
-			irq = DMTE6_IRQ;
-		else
-#endif
-			irq = DMTE0_IRQ;
-	}
-
-	snprintf(new_sh_chan->dev_id, sizeof(new_sh_chan->dev_id),
-		 "sh-dmae%d", new_sh_chan->id);
+	if (pdev->id >= 0)
+		snprintf(new_sh_chan->dev_id, sizeof(new_sh_chan->dev_id),
+			 "sh-dmae%d.%d", pdev->id, new_sh_chan->id);
+	else
+		snprintf(new_sh_chan->dev_id, sizeof(new_sh_chan->dev_id),
+			 "sh-dma%d", new_sh_chan->id);
 
 	/* set up channel irq */
-	err = request_irq(irq, &sh_dmae_interrupt, irqflags,
+	err = request_irq(irq, &sh_dmae_interrupt, flags,
 			  new_sh_chan->dev_id, new_sh_chan);
 	if (err) {
 		dev_err(shdev->common.dev, "DMA channel %d request_irq error "
@@ -881,12 +927,12 @@ static void sh_dmae_chan_remove(struct sh_dmae_device *shdev)
 
 	for (i = shdev->common.chancnt - 1 ; i >= 0 ; i--) {
 		if (shdev->chan[i]) {
-			struct sh_dmae_chan *shchan = shdev->chan[i];
-			if (!(shdev->pdata.mode & SHDMA_MIX_IRQ))
-				free_irq(dmte_irq_map[i], shchan);
+			struct sh_dmae_chan *sh_chan = shdev->chan[i];
 
-			list_del(&shchan->common.device_node);
-			kfree(shchan);
+			free_irq(sh_chan->irq, sh_chan);
+
+			list_del(&sh_chan->common.device_node);
+			kfree(sh_chan);
 			shdev->chan[i] = NULL;
 		}
 	}
@@ -895,47 +941,84 @@ static void sh_dmae_chan_remove(struct sh_dmae_device *shdev)
 
 static int __init sh_dmae_probe(struct platform_device *pdev)
 {
-	int err = 0, cnt, ecnt;
-	unsigned long irqflags = IRQF_DISABLED;
-#if defined(CONFIG_CPU_SH4)
-	int eirq[] = { DMAE0_IRQ,
-#if defined(DMAE1_IRQ)
-			DMAE1_IRQ
-#endif
-		};
-#endif
+	struct sh_dmae_pdata *pdata = pdev->dev.platform_data;
+	unsigned long irqflags = IRQF_DISABLED,
+		chan_flag[SH_DMAC_MAX_CHANNELS] = {};
+	int errirq, chan_irq[SH_DMAC_MAX_CHANNELS];
+	int err, i, irq_cnt = 0, irqres = 0;
 	struct sh_dmae_device *shdev;
+	struct resource *chan, *dmars, *errirq_res, *chanirq_res;
 
 	/* get platform data */
-	if (!pdev->dev.platform_data)
+	if (!pdata || !pdata->channel_num)
 		return -ENODEV;
 
+	chan = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+	/* DMARS area is optional, if absent, this controller cannot do slave DMA */
+	dmars = platform_get_resource(pdev, IORESOURCE_MEM, 1);
+	/*
+	 * IRQ resources:
+	 * 1. there always must be at least one IRQ IO-resource. On SH4 it is
+	 *    the error IRQ, in which case it is the only IRQ in this resource:
+	 *    start == end. If it is the only IRQ resource, all channels also
+	 *    use the same IRQ.
+	 * 2. DMA channel IRQ resources can be specified one per resource or in
+	 *    ranges (start != end)
+	 * 3. iff all events (channels and, optionally, error) on this
+	 *    controller use the same IRQ, only one IRQ resource can be
+	 *    specified, otherwise there must be one IRQ per channel, even if
+	 *    some of them are equal
+	 * 4. if all IRQs on this controller are equal or if some specific IRQs
+	 *    specify IORESOURCE_IRQ_SHAREABLE in their resources, they will be
+	 *    requested with the IRQF_SHARED flag
+	 */
+	errirq_res = platform_get_resource(pdev, IORESOURCE_IRQ, 0);
+	if (!chan || !errirq_res)
+		return -ENODEV;
+
+	if (!request_mem_region(chan->start, resource_size(chan), pdev->name)) {
+		dev_err(&pdev->dev, "DMAC register region already claimed\n");
+		return -EBUSY;
+	}
+
+	if (dmars && !request_mem_region(dmars->start, resource_size(dmars), pdev->name)) {
+		dev_err(&pdev->dev, "DMAC DMARS region already claimed\n");
+		err = -EBUSY;
+		goto ermrdmars;
+	}
+
+	err = -ENOMEM;
 	shdev = kzalloc(sizeof(struct sh_dmae_device), GFP_KERNEL);
 	if (!shdev) {
-		dev_err(&pdev->dev, "No enough memory\n");
-		return -ENOMEM;
+		dev_err(&pdev->dev, "Not enough memory\n");
+		goto ealloc;
+	}
+
+	shdev->chan_reg = ioremap(chan->start, resource_size(chan));
+	if (!shdev->chan_reg)
+		goto emapchan;
+	if (dmars) {
+		shdev->dmars = ioremap(dmars->start, resource_size(dmars));
+		if (!shdev->dmars)
+			goto emapdmars;
 	}
 
 	/* platform data */
-	memcpy(&shdev->pdata, pdev->dev.platform_data,
-			sizeof(struct sh_dmae_pdata));
+	shdev->pdata = pdata;
+
+	pm_runtime_enable(&pdev->dev);
+	pm_runtime_get_sync(&pdev->dev);
 
 	/* reset dma controller */
-	err = sh_dmae_rst(0);
+	err = sh_dmae_rst(shdev);
 	if (err)
 		goto rst_err;
 
-	/* SH7780/85/23 has DMAOR1 */
-	if (shdev->pdata.mode & SHDMA_DMAOR1) {
-		err = sh_dmae_rst(1);
-		if (err)
-			goto rst_err;
-	}
-
 	INIT_LIST_HEAD(&shdev->common.channels);
 
 	dma_cap_set(DMA_MEMCPY, shdev->common.cap_mask);
-	dma_cap_set(DMA_SLAVE, shdev->common.cap_mask);
+	if (dmars)
+		dma_cap_set(DMA_SLAVE, shdev->common.cap_mask);
 
 	shdev->common.device_alloc_chan_resources
 		= sh_dmae_alloc_chan_resources;
@@ -950,37 +1033,72 @@ static int __init sh_dmae_probe(struct platform_device *pdev)
 
 	shdev->common.dev = &pdev->dev;
 	/* Default transfer size of 32 bytes requires 32-byte alignment */
-	shdev->common.copy_align = 5;
+	shdev->common.copy_align = LOG2_DEFAULT_XFER_SIZE;
 
 #if defined(CONFIG_CPU_SH4)
-	/* Non Mix IRQ mode SH7722/SH7730 etc... */
-	if (shdev->pdata.mode & SHDMA_MIX_IRQ) {
+	chanirq_res = platform_get_resource(pdev, IORESOURCE_IRQ, 1);
+
+	if (!chanirq_res)
+		chanirq_res = errirq_res;
+	else
+		irqres++;
+
+	if (chanirq_res == errirq_res ||
+	    (errirq_res->flags & IORESOURCE_BITS) == IORESOURCE_IRQ_SHAREABLE)
 		irqflags = IRQF_SHARED;
-		eirq[0] = DMTE0_IRQ;
-#if defined(DMTE6_IRQ) && defined(DMAE1_IRQ)
-		eirq[1] = DMTE6_IRQ;
-#endif
+
+	errirq = errirq_res->start;
+
+	err = request_irq(errirq, sh_dmae_err, irqflags,
+			  "DMAC Address Error", shdev);
+	if (err) {
+		dev_err(&pdev->dev,
+			"DMA failed requesting irq #%d, error %d\n",
+			errirq, err);
+		goto eirq_err;
 	}
 
-	for (ecnt = 0 ; ecnt < ARRAY_SIZE(eirq); ecnt++) {
-		err = request_irq(eirq[ecnt], sh_dmae_err, irqflags,
-				  "DMAC Address Error", shdev);
-		if (err) {
-			dev_err(&pdev->dev, "DMA device request_irq"
-				"error (irq %d) with return %d\n",
-				eirq[ecnt], err);
-			goto eirq_err;
+#else
+	chanirq_res = errirq_res;
+#endif /* CONFIG_CPU_SH4 */
+
+	if (chanirq_res->start == chanirq_res->end &&
+	    !platform_get_resource(pdev, IORESOURCE_IRQ, 1)) {
+		/* Special case - all multiplexed */
+		for (; irq_cnt < pdata->channel_num; irq_cnt++) {
+			chan_irq[irq_cnt] = chanirq_res->start;
+			chan_flag[irq_cnt] = IRQF_SHARED;
 		}
+	} else {
+		do {
+			for (i = chanirq_res->start; i <= chanirq_res->end; i++) {
+				if ((errirq_res->flags & IORESOURCE_BITS) ==
+				    IORESOURCE_IRQ_SHAREABLE)
+					chan_flag[irq_cnt] = IRQF_SHARED;
+				else
+					chan_flag[irq_cnt] = IRQF_DISABLED;
+				dev_dbg(&pdev->dev,
+					"Found IRQ %d for channel %d\n",
+					i, irq_cnt);
+				chan_irq[irq_cnt++] = i;
+			}
+			chanirq_res = platform_get_resource(pdev,
+						IORESOURCE_IRQ, ++irqres);
+		} while (irq_cnt < pdata->channel_num && chanirq_res);
 	}
-#endif /* CONFIG_CPU_SH4 */
+
+	if (irq_cnt < pdata->channel_num)
+		goto eirqres;
 
 	/* Create DMA Channel */
-	for (cnt = 0 ; cnt < MAX_DMA_CHANNELS ; cnt++) {
-		err = sh_dmae_chan_probe(shdev, cnt);
+	for (i = 0; i < pdata->channel_num; i++) {
+		err = sh_dmae_chan_probe(shdev, i, chan_irq[i], chan_flag[i]);
 		if (err)
 			goto chan_probe_err;
 	}
 
+	pm_runtime_put(&pdev->dev);
+
 	platform_set_drvdata(pdev, shdev);
 	dma_async_device_register(&shdev->common);
 
@@ -988,13 +1106,24 @@ static int __init sh_dmae_probe(struct platform_device *pdev)
 
 chan_probe_err:
 	sh_dmae_chan_remove(shdev);
-
+eirqres:
+#if defined(CONFIG_CPU_SH4)
+	free_irq(errirq, shdev);
 eirq_err:
-	for (ecnt-- ; ecnt >= 0; ecnt--)
-		free_irq(eirq[ecnt], shdev);
-
+#endif
 rst_err:
+	pm_runtime_put(&pdev->dev);
+	if (dmars)
+		iounmap(shdev->dmars);
+emapdmars:
+	iounmap(shdev->chan_reg);
+emapchan:
 	kfree(shdev);
+ealloc:
+	if (dmars)
+		release_mem_region(dmars->start, resource_size(dmars));
+ermrdmars:
+	release_mem_region(chan->start, resource_size(chan));
 
 	return err;
 }
@@ -1002,36 +1131,39 @@ rst_err:
 static int __exit sh_dmae_remove(struct platform_device *pdev)
 {
 	struct sh_dmae_device *shdev = platform_get_drvdata(pdev);
+	struct resource *res;
+	int errirq = platform_get_irq(pdev, 0);
 
 	dma_async_device_unregister(&shdev->common);
 
-	if (shdev->pdata.mode & SHDMA_MIX_IRQ) {
-		free_irq(DMTE0_IRQ, shdev);
-#if defined(DMTE6_IRQ)
-		free_irq(DMTE6_IRQ, shdev);
-#endif
-	}
+	if (errirq > 0)
+		free_irq(errirq, shdev);
 
 	/* channel data remove */
 	sh_dmae_chan_remove(shdev);
 
-	if (!(shdev->pdata.mode & SHDMA_MIX_IRQ)) {
-		free_irq(DMAE0_IRQ, shdev);
-#if defined(DMAE1_IRQ)
-		free_irq(DMAE1_IRQ, shdev);
-#endif
-	}
+	pm_runtime_disable(&pdev->dev);
+
+	if (shdev->dmars)
+		iounmap(shdev->dmars);
+	iounmap(shdev->chan_reg);
+
 	kfree(shdev);
 
+	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+	if (res)
+		release_mem_region(res->start, resource_size(res));
+	res = platform_get_resource(pdev, IORESOURCE_MEM, 1);
+	if (res)
+		release_mem_region(res->start, resource_size(res));
+
 	return 0;
 }
 
 static void sh_dmae_shutdown(struct platform_device *pdev)
 {
 	struct sh_dmae_device *shdev = platform_get_drvdata(pdev);
-	sh_dmae_ctl_stop(0);
-	if (shdev->pdata.mode & SHDMA_DMAOR1)
-		sh_dmae_ctl_stop(1);
+	sh_dmae_ctl_stop(shdev);
 }
 
 static struct platform_driver sh_dmae_driver = {
diff --git a/drivers/dma/shdma.h b/drivers/dma/shdma.h
index 7e227f3c87c4..153609a1e96c 100644
--- a/drivers/dma/shdma.h
+++ b/drivers/dma/shdma.h
@@ -17,23 +17,9 @@
 #include <linux/interrupt.h>
 #include <linux/list.h>
 
-#define SH_DMA_TCR_MAX 0x00FFFFFF	/* 16MB */
-
-struct sh_dmae_regs {
-	u32 sar; /* SAR / source address */
-	u32 dar; /* DAR / destination address */
-	u32 tcr; /* TCR / transfer count */
-};
+#include <asm/dmaengine.h>
 
-struct sh_desc {
-	struct sh_dmae_regs hw;
-	struct list_head node;
-	struct dma_async_tx_descriptor async_tx;
-	enum dma_data_direction direction;
-	dma_cookie_t cookie;
-	int chunks;
-	int mark;
-};
+#define SH_DMA_TCR_MAX 0x00FFFFFF	/* 16MB */
 
 struct device;
 
@@ -47,14 +33,18 @@ struct sh_dmae_chan {
 	struct tasklet_struct tasklet;	/* Tasklet */
 	int descs_allocated;		/* desc count */
 	int xmit_shift;			/* log_2(bytes_per_xfer) */
+	int irq;
 	int id;				/* Raw id of this channel */
+	u32 __iomem *base;
 	char dev_id[16];		/* unique name per DMAC of channel */
 };
 
 struct sh_dmae_device {
 	struct dma_device common;
-	struct sh_dmae_chan *chan[MAX_DMA_CHANNELS];
-	struct sh_dmae_pdata pdata;
+	struct sh_dmae_chan *chan[SH_DMAC_MAX_CHANNELS];
+	struct sh_dmae_pdata *pdata;
+	u32 __iomem *chan_reg;
+	u16 __iomem *dmars;
 };
 
 #define to_sh_chan(chan) container_of(chan, struct sh_dmae_chan, common)
diff --git a/drivers/serial/Kconfig b/drivers/serial/Kconfig
index 746e07033dce..d6ff73395623 100644
--- a/drivers/serial/Kconfig
+++ b/drivers/serial/Kconfig
@@ -1009,6 +1009,10 @@ config SERIAL_SH_SCI_CONSOLE
 	depends on SERIAL_SH_SCI=y
 	select SERIAL_CORE_CONSOLE
 
+config SERIAL_SH_SCI_DMA
+	bool "DMA support"
+	depends on SERIAL_SH_SCI && SH_DMAE && EXPERIMENTAL
+
 config SERIAL_PNX8XXX
 	bool "Enable PNX8XXX SoCs' UART Support"
 	depends on MIPS && (SOC_PNX8550 || SOC_PNX833X)
diff --git a/drivers/serial/sh-sci.c b/drivers/serial/sh-sci.c
index 42f3333c4ad0..980f39449ee5 100644
--- a/drivers/serial/sh-sci.c
+++ b/drivers/serial/sh-sci.c
@@ -48,6 +48,9 @@
 #include <linux/ctype.h>
 #include <linux/err.h>
 #include <linux/list.h>
+#include <linux/dmaengine.h>
+#include <linux/scatterlist.h>
+#include <linux/timer.h>
 
 #ifdef CONFIG_SUPERH
 #include <asm/sh_bios.h>
@@ -84,6 +87,27 @@ struct sci_port {
 	struct clk		*dclk;
 
 	struct list_head	node;
+	struct dma_chan			*chan_tx;
+	struct dma_chan			*chan_rx;
+#ifdef CONFIG_SERIAL_SH_SCI_DMA
+	struct device			*dma_dev;
+	enum sh_dmae_slave_chan_id	slave_tx;
+	enum sh_dmae_slave_chan_id	slave_rx;
+	struct dma_async_tx_descriptor	*desc_tx;
+	struct dma_async_tx_descriptor	*desc_rx[2];
+	dma_cookie_t			cookie_tx;
+	dma_cookie_t			cookie_rx[2];
+	dma_cookie_t			active_rx;
+	struct scatterlist		sg_tx;
+	unsigned int			sg_len_tx;
+	struct scatterlist		sg_rx[2];
+	size_t				buf_len_rx;
+	struct sh_dmae_slave		param_tx;
+	struct sh_dmae_slave		param_rx;
+	struct work_struct		work_tx;
+	struct work_struct		work_rx;
+	struct timer_list		rx_timer;
+#endif
 };
 
 struct sh_sci_priv {
@@ -269,29 +293,44 @@ static inline void sci_init_pins(struct uart_port *port, unsigned int cflag)
     defined(CONFIG_CPU_SUBTYPE_SH7780) || \
     defined(CONFIG_CPU_SUBTYPE_SH7785) || \
     defined(CONFIG_CPU_SUBTYPE_SH7786)
-static inline int scif_txroom(struct uart_port *port)
+static int scif_txfill(struct uart_port *port)
 {
-	return SCIF_TXROOM_MAX - (sci_in(port, SCTFDR) & 0xff);
+	return sci_in(port, SCTFDR) & 0xff;
 }
 
-static inline int scif_rxroom(struct uart_port *port)
+static int scif_txroom(struct uart_port *port)
+{
+	return SCIF_TXROOM_MAX - scif_txfill(port);
+}
+
+static int scif_rxfill(struct uart_port *port)
 {
 	return sci_in(port, SCRFDR) & 0xff;
 }
 #elif defined(CONFIG_CPU_SUBTYPE_SH7763)
-static inline int scif_txroom(struct uart_port *port)
+static int scif_txfill(struct uart_port *port)
 {
-	if ((port->mapbase == 0xffe00000) ||
-	    (port->mapbase == 0xffe08000)) {
+	if (port->mapbase == 0xffe00000 ||
+	    port->mapbase == 0xffe08000)
 		/* SCIF0/1*/
-		return SCIF_TXROOM_MAX - (sci_in(port, SCTFDR) & 0xff);
-	} else {
+		return sci_in(port, SCTFDR) & 0xff;
+	else
 		/* SCIF2 */
-		return SCIF2_TXROOM_MAX - (sci_in(port, SCFDR) >> 8);
-	}
+		return sci_in(port, SCFDR) >> 8;
 }
 
-static inline int scif_rxroom(struct uart_port *port)
+static int scif_txroom(struct uart_port *port)
+{
+	if (port->mapbase == 0xffe00000 ||
+	    port->mapbase == 0xffe08000)
+		/* SCIF0/1*/
+		return SCIF_TXROOM_MAX - scif_txfill(port);
+	else
+		/* SCIF2 */
+		return SCIF2_TXROOM_MAX - scif_txfill(port);
+}
+
+static int scif_rxfill(struct uart_port *port)
 {
 	if ((port->mapbase == 0xffe00000) ||
 	    (port->mapbase == 0xffe08000)) {
@@ -303,23 +342,33 @@ static inline int scif_rxroom(struct uart_port *port)
 	}
 }
 #else
-static inline int scif_txroom(struct uart_port *port)
+static int scif_txfill(struct uart_port *port)
+{
+	return sci_in(port, SCFDR) >> 8;
+}
+
+static int scif_txroom(struct uart_port *port)
 {
-	return SCIF_TXROOM_MAX - (sci_in(port, SCFDR) >> 8);
+	return SCIF_TXROOM_MAX - scif_txfill(port);
 }
 
-static inline int scif_rxroom(struct uart_port *port)
+static int scif_rxfill(struct uart_port *port)
 {
 	return sci_in(port, SCFDR) & SCIF_RFDC_MASK;
 }
 #endif
 
-static inline int sci_txroom(struct uart_port *port)
+static int sci_txfill(struct uart_port *port)
 {
-	return (sci_in(port, SCxSR) & SCI_TDRE) != 0;
+	return !(sci_in(port, SCxSR) & SCI_TDRE);
 }
 
-static inline int sci_rxroom(struct uart_port *port)
+static int sci_txroom(struct uart_port *port)
+{
+	return !sci_txfill(port);
+}
+
+static int sci_rxfill(struct uart_port *port)
 {
 	return (sci_in(port, SCxSR) & SCxSR_RDxF(port)) != 0;
 }
@@ -406,9 +455,9 @@ static inline void sci_receive_chars(struct uart_port *port)
 
 	while (1) {
 		if (port->type == PORT_SCI)
-			count = sci_rxroom(port);
+			count = sci_rxfill(port);
 		else
-			count = scif_rxroom(port);
+			count = scif_rxfill(port);
 
 		/* Don't copy more bytes than there is room for in the buffer */
 		count = tty_buffer_request_room(tty, count);
@@ -453,10 +502,10 @@ static inline void sci_receive_chars(struct uart_port *port)
 				}
 
 				/* Store data and status */
-				if (status&SCxSR_FER(port)) {
+				if (status & SCxSR_FER(port)) {
 					flag = TTY_FRAME;
 					dev_notice(port->dev, "frame error\n");
-				} else if (status&SCxSR_PER(port)) {
+				} else if (status & SCxSR_PER(port)) {
 					flag = TTY_PARITY;
 					dev_notice(port->dev, "parity error\n");
 				} else
@@ -618,13 +667,39 @@ static inline int sci_handle_breaks(struct uart_port *port)
 	return copied;
 }
 
-static irqreturn_t sci_rx_interrupt(int irq, void *port)
+static irqreturn_t sci_rx_interrupt(int irq, void *ptr)
 {
+#ifdef CONFIG_SERIAL_SH_SCI_DMA
+	struct uart_port *port = ptr;
+	struct sci_port *s = to_sci_port(port);
+
+	if (s->chan_rx) {
+		unsigned long tout;
+		u16 scr = sci_in(port, SCSCR);
+		u16 ssr = sci_in(port, SCxSR);
+
+		/* Disable future Rx interrupts */
+		sci_out(port, SCSCR, scr & ~SCI_CTRL_FLAGS_RIE);
+		/* Clear current interrupt */
+		sci_out(port, SCxSR, ssr & ~(1 | SCxSR_RDxF(port)));
+		/* Calculate delay for 1.5 DMA buffers */
+		tout = (port->timeout - HZ / 50) * s->buf_len_rx * 3 /
+			port->fifosize / 2;
+		dev_dbg(port->dev, "Rx IRQ: setup timeout in %lu ms\n",
+			tout * 1000 / HZ);
+		if (tout < 2)
+			tout = 2;
+		mod_timer(&s->rx_timer, jiffies + tout);
+
+		return IRQ_HANDLED;
+	}
+#endif
+
 	/* I think sci_receive_chars has to be called irrespective
 	 * of whether the I_IXOFF is set, otherwise, how is the interrupt
 	 * to be disabled?
 	 */
-	sci_receive_chars(port);
+	sci_receive_chars(ptr);
 
 	return IRQ_HANDLED;
 }
@@ -680,6 +755,7 @@ static irqreturn_t sci_mpxed_interrupt(int irq, void *ptr)
 {
 	unsigned short ssr_status, scr_status, err_enabled;
 	struct uart_port *port = ptr;
+	struct sci_port *s = to_sci_port(port);
 	irqreturn_t ret = IRQ_NONE;
 
 	ssr_status = sci_in(port, SCxSR);
@@ -687,10 +763,15 @@ static irqreturn_t sci_mpxed_interrupt(int irq, void *ptr)
 	err_enabled = scr_status & (SCI_CTRL_FLAGS_REIE | SCI_CTRL_FLAGS_RIE);
 
 	/* Tx Interrupt */
-	if ((ssr_status & SCxSR_TDxE(port)) && (scr_status & SCI_CTRL_FLAGS_TIE))
+	if ((ssr_status & SCxSR_TDxE(port)) && (scr_status & SCI_CTRL_FLAGS_TIE) &&
+	    !s->chan_tx)
 		ret = sci_tx_interrupt(irq, ptr);
-	/* Rx Interrupt */
-	if ((ssr_status & SCxSR_RDxF(port)) && (scr_status & SCI_CTRL_FLAGS_RIE))
+	/*
+	 * Rx Interrupt: if we're using DMA, the DMA controller clears RDF /
+	 * DR flags
+	 */
+	if (((ssr_status & SCxSR_RDxF(port)) || s->chan_rx) &&
+	    (scr_status & SCI_CTRL_FLAGS_RIE))
 		ret = sci_rx_interrupt(irq, ptr);
 	/* Error Interrupt */
 	if ((ssr_status & SCxSR_ERRORS(port)) && err_enabled)
@@ -699,6 +780,10 @@ static irqreturn_t sci_mpxed_interrupt(int irq, void *ptr)
 	if ((ssr_status & SCxSR_BRK(port)) && err_enabled)
 		ret = sci_br_interrupt(irq, ptr);
 
+	WARN_ONCE(ret == IRQ_NONE,
+		  "%s: %d IRQ %d, status %x, control %x\n", __func__,
+		  irq, port->line, ssr_status, scr_status);
+
 	return ret;
 }
 
@@ -800,7 +885,9 @@ static void sci_free_irq(struct sci_port *port)
 static unsigned int sci_tx_empty(struct uart_port *port)
 {
 	unsigned short status = sci_in(port, SCxSR);
-	return status & SCxSR_TEND(port) ? TIOCSER_TEMT : 0;
+	unsigned short in_tx_fifo = scif_txfill(port);
+
+	return (status & SCxSR_TEND(port)) && !in_tx_fifo ? TIOCSER_TEMT : 0;
 }
 
 static void sci_set_mctrl(struct uart_port *port, unsigned int mctrl)
@@ -812,16 +899,297 @@ static void sci_set_mctrl(struct uart_port *port, unsigned int mctrl)
 
 static unsigned int sci_get_mctrl(struct uart_port *port)
 {
-	/* This routine is used for geting signals of: DTR, DCD, DSR, RI,
+	/* This routine is used for getting signals of: DTR, DCD, DSR, RI,
 	   and CTS/RTS */
 
 	return TIOCM_DTR | TIOCM_RTS | TIOCM_DSR;
 }
 
+#ifdef CONFIG_SERIAL_SH_SCI_DMA
+static void sci_dma_tx_complete(void *arg)
+{
+	struct sci_port *s = arg;
+	struct uart_port *port = &s->port;
+	struct circ_buf *xmit = &port->state->xmit;
+	unsigned long flags;
+
+	dev_dbg(port->dev, "%s(%d)\n", __func__, port->line);
+
+	spin_lock_irqsave(&port->lock, flags);
+
+	xmit->tail += s->sg_tx.length;
+	xmit->tail &= UART_XMIT_SIZE - 1;
+
+	port->icount.tx += s->sg_tx.length;
+
+	async_tx_ack(s->desc_tx);
+	s->cookie_tx = -EINVAL;
+	s->desc_tx = NULL;
+
+	spin_unlock_irqrestore(&port->lock, flags);
+
+	if (uart_circ_chars_pending(xmit) < WAKEUP_CHARS)
+		uart_write_wakeup(port);
+
+	if (uart_circ_chars_pending(xmit))
+		schedule_work(&s->work_tx);
+}
+
+/* Locking: called with port lock held */
+static int sci_dma_rx_push(struct sci_port *s, struct tty_struct *tty,
+			   size_t count)
+{
+	struct uart_port *port = &s->port;
+	int i, active, room;
+
+	room = tty_buffer_request_room(tty, count);
+
+	if (s->active_rx == s->cookie_rx[0]) {
+		active = 0;
+	} else if (s->active_rx == s->cookie_rx[1]) {
+		active = 1;
+	} else {
+		dev_err(port->dev, "cookie %d not found!\n", s->active_rx);
+		return 0;
+	}
+
+	if (room < count)
+		dev_warn(port->dev, "Rx overrun: dropping %u bytes\n",
+			 count - room);
+	if (!room)
+		return room;
+
+	for (i = 0; i < room; i++)
+		tty_insert_flip_char(tty, ((u8 *)sg_virt(&s->sg_rx[active]))[i],
+				     TTY_NORMAL);
+
+	port->icount.rx += room;
+
+	return room;
+}
+
+static void sci_dma_rx_complete(void *arg)
+{
+	struct sci_port *s = arg;
+	struct uart_port *port = &s->port;
+	struct tty_struct *tty = port->state->port.tty;
+	unsigned long flags;
+	int count;
+
+	dev_dbg(port->dev, "%s(%d)\n", __func__, port->line);
+
+	spin_lock_irqsave(&port->lock, flags);
+
+	count = sci_dma_rx_push(s, tty, s->buf_len_rx);
+
+	mod_timer(&s->rx_timer, jiffies + msecs_to_jiffies(5));
+
+	spin_unlock_irqrestore(&port->lock, flags);
+
+	if (count)
+		tty_flip_buffer_push(tty);
+
+	schedule_work(&s->work_rx);
+}
+
+static void sci_start_rx(struct uart_port *port);
+static void sci_start_tx(struct uart_port *port);
+
+static void sci_rx_dma_release(struct sci_port *s, bool enable_pio)
+{
+	struct dma_chan *chan = s->chan_rx;
+	struct uart_port *port = &s->port;
+
+	s->chan_rx = NULL;
+	s->cookie_rx[0] = s->cookie_rx[1] = -EINVAL;
+	dma_release_channel(chan);
+	dma_free_coherent(port->dev, s->buf_len_rx * 2,
+			  sg_virt(&s->sg_rx[0]), sg_dma_address(&s->sg_rx[0]));
+	if (enable_pio)
+		sci_start_rx(port);
+}
+
+static void sci_tx_dma_release(struct sci_port *s, bool enable_pio)
+{
+	struct dma_chan *chan = s->chan_tx;
+	struct uart_port *port = &s->port;
+
+	s->chan_tx = NULL;
+	s->cookie_tx = -EINVAL;
+	dma_release_channel(chan);
+	if (enable_pio)
+		sci_start_tx(port);
+}
+
+static void sci_submit_rx(struct sci_port *s)
+{
+	struct dma_chan *chan = s->chan_rx;
+	int i;
+
+	for (i = 0; i < 2; i++) {
+		struct scatterlist *sg = &s->sg_rx[i];
+		struct dma_async_tx_descriptor *desc;
+
+		desc = chan->device->device_prep_slave_sg(chan,
+			sg, 1, DMA_FROM_DEVICE, DMA_PREP_INTERRUPT);
+
+		if (desc) {
+			s->desc_rx[i] = desc;
+			desc->callback = sci_dma_rx_complete;
+			desc->callback_param = s;
+			s->cookie_rx[i] = desc->tx_submit(desc);
+		}
+
+		if (!desc || s->cookie_rx[i] < 0) {
+			if (i) {
+				async_tx_ack(s->desc_rx[0]);
+				s->cookie_rx[0] = -EINVAL;
+			}
+			if (desc) {
+				async_tx_ack(desc);
+				s->cookie_rx[i] = -EINVAL;
+			}
+			dev_warn(s->port.dev,
+				 "failed to re-start DMA, using PIO\n");
+			sci_rx_dma_release(s, true);
+			return;
+		}
+	}
+
+	s->active_rx = s->cookie_rx[0];
+
+	dma_async_issue_pending(chan);
+}
+
+static void work_fn_rx(struct work_struct *work)
+{
+	struct sci_port *s = container_of(work, struct sci_port, work_rx);
+	struct uart_port *port = &s->port;
+	struct dma_async_tx_descriptor *desc;
+	int new;
+
+	if (s->active_rx == s->cookie_rx[0]) {
+		new = 0;
+	} else if (s->active_rx == s->cookie_rx[1]) {
+		new = 1;
+	} else {
+		dev_err(port->dev, "cookie %d not found!\n", s->active_rx);
+		return;
+	}
+	desc = s->desc_rx[new];
+
+	if (dma_async_is_tx_complete(s->chan_rx, s->active_rx, NULL, NULL) !=
+	    DMA_SUCCESS) {
+		/* Handle incomplete DMA receive */
+		struct tty_struct *tty = port->state->port.tty;
+		struct dma_chan *chan = s->chan_rx;
+		struct sh_desc *sh_desc = container_of(desc, struct sh_desc,
+						       async_tx);
+		unsigned long flags;
+		int count;
+
+		chan->device->device_terminate_all(chan);
+		dev_dbg(port->dev, "Read %u bytes with cookie %d\n",
+			sh_desc->partial, sh_desc->cookie);
+
+		spin_lock_irqsave(&port->lock, flags);
+		count = sci_dma_rx_push(s, tty, sh_desc->partial);
+		spin_unlock_irqrestore(&port->lock, flags);
+
+		if (count)
+			tty_flip_buffer_push(tty);
+
+		sci_submit_rx(s);
+
+		return;
+	}
+
+	s->cookie_rx[new] = desc->tx_submit(desc);
+	if (s->cookie_rx[new] < 0) {
+		dev_warn(port->dev, "Failed submitting Rx DMA descriptor\n");
+		sci_rx_dma_release(s, true);
+		return;
+	}
+
+	dev_dbg(port->dev, "%s: cookie %d #%d\n", __func__,
+		s->cookie_rx[new], new);
+
+	s->active_rx = s->cookie_rx[!new];
+}
+
+static void work_fn_tx(struct work_struct *work)
+{
+	struct sci_port *s = container_of(work, struct sci_port, work_tx);
+	struct dma_async_tx_descriptor *desc;
+	struct dma_chan *chan = s->chan_tx;
+	struct uart_port *port = &s->port;
+	struct circ_buf *xmit = &port->state->xmit;
+	struct scatterlist *sg = &s->sg_tx;
+
+	/*
+	 * DMA is idle now.
+	 * Port xmit buffer is already mapped, and it is one page... Just adjust
+	 * offsets and lengths. Since it is a circular buffer, we have to
+	 * transmit till the end, and then the rest. Take the port lock to get a
+	 * consistent xmit buffer state.
+	 */
+	spin_lock_irq(&port->lock);
+	sg->offset = xmit->tail & (UART_XMIT_SIZE - 1);
+	sg->dma_address = (sg_dma_address(sg) & ~(UART_XMIT_SIZE - 1)) +
+		sg->offset;
+	sg->length = min((int)CIRC_CNT(xmit->head, xmit->tail, UART_XMIT_SIZE),
+		CIRC_CNT_TO_END(xmit->head, xmit->tail, UART_XMIT_SIZE));
+	sg->dma_length = sg->length;
+	spin_unlock_irq(&port->lock);
+
+	BUG_ON(!sg->length);
+
+	desc = chan->device->device_prep_slave_sg(chan,
+			sg, s->sg_len_tx, DMA_TO_DEVICE,
+			DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
+	if (!desc) {
+		/* switch to PIO */
+		sci_tx_dma_release(s, true);
+		return;
+	}
+
+	dma_sync_sg_for_device(port->dev, sg, 1, DMA_TO_DEVICE);
+
+	spin_lock_irq(&port->lock);
+	s->desc_tx = desc;
+	desc->callback = sci_dma_tx_complete;
+	desc->callback_param = s;
+	spin_unlock_irq(&port->lock);
+	s->cookie_tx = desc->tx_submit(desc);
+	if (s->cookie_tx < 0) {
+		dev_warn(port->dev, "Failed submitting Tx DMA descriptor\n");
+		/* switch to PIO */
+		sci_tx_dma_release(s, true);
+		return;
+	}
+
+	dev_dbg(port->dev, "%s: %p: %d...%d, cookie %d\n", __func__,
+		xmit->buf, xmit->tail, xmit->head, s->cookie_tx);
+
+	dma_async_issue_pending(chan);
+}
+#endif
+
 static void sci_start_tx(struct uart_port *port)
 {
 	unsigned short ctrl;
 
+#ifdef CONFIG_SERIAL_SH_SCI_DMA
+	struct sci_port *s = to_sci_port(port);
+
+	if (s->chan_tx) {
+		if (!uart_circ_empty(&s->port.state->xmit) && s->cookie_tx < 0)
+			schedule_work(&s->work_tx);
+
+		return;
+	}
+#endif
+
 	/* Set TIE (Transmit Interrupt Enable) bit in SCSCR */
 	ctrl = sci_in(port, SCSCR);
 	ctrl |= SCI_CTRL_FLAGS_TIE;
@@ -838,13 +1206,12 @@ static void sci_stop_tx(struct uart_port *port)
 	sci_out(port, SCSCR, ctrl);
 }
 
-static void sci_start_rx(struct uart_port *port, unsigned int tty_start)
+static void sci_start_rx(struct uart_port *port)
 {
-	unsigned short ctrl;
+	unsigned short ctrl = SCI_CTRL_FLAGS_RIE | SCI_CTRL_FLAGS_REIE;
 
 	/* Set RIE (Receive Interrupt Enable) bit in SCSCR */
-	ctrl = sci_in(port, SCSCR);
-	ctrl |= SCI_CTRL_FLAGS_RIE | SCI_CTRL_FLAGS_REIE;
+	ctrl |= sci_in(port, SCSCR);
 	sci_out(port, SCSCR, ctrl);
 }
 
@@ -868,16 +1235,154 @@ static void sci_break_ctl(struct uart_port *port, int break_state)
 	/* Nothing here yet .. */
 }
 
+#ifdef CONFIG_SERIAL_SH_SCI_DMA
+static bool filter(struct dma_chan *chan, void *slave)
+{
+	struct sh_dmae_slave *param = slave;
+
+	dev_dbg(chan->device->dev, "%s: slave ID %d\n", __func__,
+		param->slave_id);
+
+	if (param->dma_dev == chan->device->dev) {
+		chan->private = param;
+		return true;
+	} else {
+		return false;
+	}
+}
+
+static void rx_timer_fn(unsigned long arg)
+{
+	struct sci_port *s = (struct sci_port *)arg;
+	struct uart_port *port = &s->port;
+
+	u16 scr = sci_in(port, SCSCR);
+	sci_out(port, SCSCR, scr | SCI_CTRL_FLAGS_RIE);
+	dev_dbg(port->dev, "DMA Rx timed out\n");
+	schedule_work(&s->work_rx);
+}
+
+static void sci_request_dma(struct uart_port *port)
+{
+	struct sci_port *s = to_sci_port(port);
+	struct sh_dmae_slave *param;
+	struct dma_chan *chan;
+	dma_cap_mask_t mask;
+	int nent;
+
+	dev_dbg(port->dev, "%s: port %d DMA %p\n", __func__,
+		port->line, s->dma_dev);
+
+	if (!s->dma_dev)
+		return;
+
+	dma_cap_zero(mask);
+	dma_cap_set(DMA_SLAVE, mask);
+
+	param = &s->param_tx;
+
+	/* Slave ID, e.g., SHDMA_SLAVE_SCIF0_TX */
+	param->slave_id = s->slave_tx;
+	param->dma_dev = s->dma_dev;
+
+	s->cookie_tx = -EINVAL;
+	chan = dma_request_channel(mask, filter, param);
+	dev_dbg(port->dev, "%s: TX: got channel %p\n", __func__, chan);
+	if (chan) {
+		s->chan_tx = chan;
+		sg_init_table(&s->sg_tx, 1);
+		/* UART circular tx buffer is an aligned page. */
+		BUG_ON((int)port->state->xmit.buf & ~PAGE_MASK);
+		sg_set_page(&s->sg_tx, virt_to_page(port->state->xmit.buf),
+			    UART_XMIT_SIZE, (int)port->state->xmit.buf & ~PAGE_MASK);
+		nent = dma_map_sg(port->dev, &s->sg_tx, 1, DMA_TO_DEVICE);
+		if (!nent)
+			sci_tx_dma_release(s, false);
+		else
+			dev_dbg(port->dev, "%s: mapped %d@%p to %x\n", __func__,
+				sg_dma_len(&s->sg_tx),
+				port->state->xmit.buf, sg_dma_address(&s->sg_tx));
+
+		s->sg_len_tx = nent;
+
+		INIT_WORK(&s->work_tx, work_fn_tx);
+	}
+
+	param = &s->param_rx;
+
+	/* Slave ID, e.g., SHDMA_SLAVE_SCIF0_RX */
+	param->slave_id = s->slave_rx;
+	param->dma_dev = s->dma_dev;
+
+	chan = dma_request_channel(mask, filter, param);
+	dev_dbg(port->dev, "%s: RX: got channel %p\n", __func__, chan);
+	if (chan) {
+		dma_addr_t dma[2];
+		void *buf[2];
+		int i;
+
+		s->chan_rx = chan;
+
+		s->buf_len_rx = 2 * max(16, (int)port->fifosize);
+		buf[0] = dma_alloc_coherent(port->dev, s->buf_len_rx * 2,
+					    &dma[0], GFP_KERNEL);
+
+		if (!buf[0]) {
+			dev_warn(port->dev,
+				 "failed to allocate dma buffer, using PIO\n");
+			sci_rx_dma_release(s, true);
+			return;
+		}
+
+		buf[1] = buf[0] + s->buf_len_rx;
+		dma[1] = dma[0] + s->buf_len_rx;
+
+		for (i = 0; i < 2; i++) {
+			struct scatterlist *sg = &s->sg_rx[i];
+
+			sg_init_table(sg, 1);
+			sg_set_page(sg, virt_to_page(buf[i]), s->buf_len_rx,
+				    (int)buf[i] & ~PAGE_MASK);
+			sg->dma_address = dma[i];
+			sg->dma_length = sg->length;
+		}
+
+		INIT_WORK(&s->work_rx, work_fn_rx);
+		setup_timer(&s->rx_timer, rx_timer_fn, (unsigned long)s);
+
+		sci_submit_rx(s);
+	}
+}
+
+static void sci_free_dma(struct uart_port *port)
+{
+	struct sci_port *s = to_sci_port(port);
+
+	if (!s->dma_dev)
+		return;
+
+	if (s->chan_tx)
+		sci_tx_dma_release(s, false);
+	if (s->chan_rx)
+		sci_rx_dma_release(s, false);
+}
+#endif
+
 static int sci_startup(struct uart_port *port)
 {
 	struct sci_port *s = to_sci_port(port);
 
+	dev_dbg(port->dev, "%s(%d)\n", __func__, port->line);
+
 	if (s->enable)
 		s->enable(port);
 
 	sci_request_irq(s);
+#ifdef CONFIG_SERIAL_SH_SCI_DMA
+	sci_request_dma(port);
+#endif
 	sci_start_tx(port);
-	sci_start_rx(port, 1);
+	sci_start_rx(port);
 
 	return 0;
 }
@@ -886,8 +1391,13 @@ static void sci_shutdown(struct uart_port *port)
 {
 	struct sci_port *s = to_sci_port(port);
 
+	dev_dbg(port->dev, "%s(%d)\n", __func__, port->line);
+
 	sci_stop_rx(port);
 	sci_stop_tx(port);
+#ifdef CONFIG_SERIAL_SH_SCI_DMA
+	sci_free_dma(port);
+#endif
 	sci_free_irq(s);
 
 	if (s->disable)
@@ -937,6 +1447,9 @@ static void sci_set_termios(struct uart_port *port, struct ktermios *termios,
 
 	sci_out(port, SCSMR, smr_val);
 
+	dev_dbg(port->dev, "%s: SMR %x, t %x, SCSCR %x\n", __func__, smr_val, t,
+		SCSCR_INIT(port));
+
 	if (t > 0) {
 		if (t >= 256) {
 			sci_out(port, SCSMR, (sci_in(port, SCSMR) & ~3) | 1);
@@ -954,7 +1467,7 @@ static void sci_set_termios(struct uart_port *port, struct ktermios *termios,
 	sci_out(port, SCSCR, SCSCR_INIT(port));
 
 	if ((termios->c_cflag & CREAD) != 0)
-		sci_start_rx(port, 0);
+		sci_start_rx(port);
 }
 
 static const char *sci_type(struct uart_port *port)
@@ -1049,19 +1562,21 @@ static void __devinit sci_init_single(struct platform_device *dev,
 				      unsigned int index,
 				      struct plat_sci_port *p)
 {
-	sci_port->port.ops	= &sci_uart_ops;
-	sci_port->port.iotype	= UPIO_MEM;
-	sci_port->port.line	= index;
+	struct uart_port *port = &sci_port->port;
+
+	port->ops	= &sci_uart_ops;
+	port->iotype	= UPIO_MEM;
+	port->line	= index;
 
 	switch (p->type) {
 	case PORT_SCIFA:
-		sci_port->port.fifosize = 64;
+		port->fifosize = 64;
 		break;
 	case PORT_SCIF:
-		sci_port->port.fifosize = 16;
+		port->fifosize = 16;
 		break;
 	default:
-		sci_port->port.fifosize = 1;
+		port->fifosize = 1;
 		break;
 	}
 
@@ -1070,19 +1585,28 @@ static void __devinit sci_init_single(struct platform_device *dev,
 		sci_port->dclk = clk_get(&dev->dev, "peripheral_clk");
 		sci_port->enable = sci_clk_enable;
 		sci_port->disable = sci_clk_disable;
-		sci_port->port.dev = &dev->dev;
+		port->dev = &dev->dev;
 	}
 
 	sci_port->break_timer.data = (unsigned long)sci_port;
 	sci_port->break_timer.function = sci_break_timer;
 	init_timer(&sci_port->break_timer);
 
-	sci_port->port.mapbase	= p->mapbase;
-	sci_port->port.membase	= p->membase;
+	port->mapbase	= p->mapbase;
+	port->membase	= p->membase;
 
-	sci_port->port.irq	= p->irqs[SCIx_TXI_IRQ];
-	sci_port->port.flags	= p->flags;
-	sci_port->type		= sci_port->port.type = p->type;
+	port->irq	= p->irqs[SCIx_TXI_IRQ];
+	port->flags	= p->flags;
+	sci_port->type	= port->type = p->type;
+
+#ifdef CONFIG_SERIAL_SH_SCI_DMA
+	sci_port->dma_dev	= p->dma_dev;
+	sci_port->slave_tx	= p->dma_slave_tx;
+	sci_port->slave_rx	= p->dma_slave_rx;
+
+	dev_dbg(port->dev, "%s: DMA device %p, tx %d, rx %d\n", __func__,
+		p->dma_dev, p->dma_slave_tx, p->dma_slave_rx);
+#endif
 
 	memcpy(&sci_port->irqs, &p->irqs, sizeof(p->irqs));
 }