Merge branch 'sa11x0-ir' into sa11x0
diff --git a/arch/arm/mach-sa1100/generic.c b/arch/arm/mach-sa1100/generic.c
index 2b33b45..1752686 100644
--- a/arch/arm/mach-sa1100/generic.c
+++ b/arch/arm/mach-sa1100/generic.c
@@ -14,6 +14,7 @@
 #include <linux/kernel.h>
 #include <linux/init.h>
 #include <linux/delay.h>
+#include <linux/dma-mapping.h>
 #include <linux/pm.h>
 #include <linux/cpufreq.h>
 #include <linux/ioport.h>
@@ -289,6 +290,29 @@
 	.id		= -1,
 };
 
+static struct resource sa11x0dma_resources[] = {
+	DEFINE_RES_MEM(__PREG(DDAR(0)), 6 * DMASp),
+	DEFINE_RES_IRQ(IRQ_DMA0),
+	DEFINE_RES_IRQ(IRQ_DMA1),
+	DEFINE_RES_IRQ(IRQ_DMA2),
+	DEFINE_RES_IRQ(IRQ_DMA3),
+	DEFINE_RES_IRQ(IRQ_DMA4),
+	DEFINE_RES_IRQ(IRQ_DMA5),
+};
+
+static u64 sa11x0dma_dma_mask = DMA_BIT_MASK(32);
+
+static struct platform_device sa11x0dma_device = {
+	.name		= "sa11x0-dma",
+	.id		= -1,
+	.dev = {
+		.dma_mask = &sa11x0dma_dma_mask,
+		.coherent_dma_mask = 0xffffffff,
+	},
+	.num_resources	= ARRAY_SIZE(sa11x0dma_resources),
+	.resource	= sa11x0dma_resources,
+};
+
 static struct platform_device *sa11x0_devices[] __initdata = {
 	&sa11x0udc_device,
 	&sa11x0uart1_device,
@@ -297,6 +321,7 @@
 	&sa11x0pcmcia_device,
 	&sa11x0fb_device,
 	&sa11x0rtc_device,
+	&sa11x0dma_device,
 };
 
 static int __init sa1100_init(void)
diff --git a/drivers/dma/Kconfig b/drivers/dma/Kconfig
index f1a2749..4a6c46de 100644
--- a/drivers/dma/Kconfig
+++ b/drivers/dma/Kconfig
@@ -252,6 +252,15 @@
 	help
 	  Enable support for the Cirrus Logic EP93xx M2P/M2M DMA controller.
 
+config DMA_SA11X0
+	tristate "SA-11x0 DMA support"
+	depends on ARCH_SA1100
+	select DMA_ENGINE
+	help
+	  Support the DMA engine found on Intel StrongARM SA-1100 and
+	  SA-1110 SoCs.  This DMA engine can only be used with on-chip
+	  devices.
+
 config DMA_ENGINE
 	bool
 
diff --git a/drivers/dma/Makefile b/drivers/dma/Makefile
index 009a222..86b795b 100644
--- a/drivers/dma/Makefile
+++ b/drivers/dma/Makefile
@@ -27,3 +27,4 @@
 obj-$(CONFIG_PCH_DMA) += pch_dma.o
 obj-$(CONFIG_AMBA_PL08X) += amba-pl08x.o
 obj-$(CONFIG_EP93XX_DMA) += ep93xx_dma.o
+obj-$(CONFIG_DMA_SA11X0) += sa11x0-dma.o
diff --git a/drivers/dma/sa11x0-dma.c b/drivers/dma/sa11x0-dma.c
new file mode 100644
index 0000000..16a6b48
--- /dev/null
+++ b/drivers/dma/sa11x0-dma.c
@@ -0,0 +1,1109 @@
+/*
+ * SA11x0 DMAengine support
+ *
+ * Copyright (C) 2012 Russell King
+ *   Derived in part from arch/arm/mach-sa1100/dma.c,
+ *   Copyright (C) 2000, 2001 by Nicolas Pitre
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+#include <linux/sched.h>
+#include <linux/device.h>
+#include <linux/dmaengine.h>
+#include <linux/init.h>
+#include <linux/interrupt.h>
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/platform_device.h>
+#include <linux/sa11x0-dma.h>
+#include <linux/slab.h>
+#include <linux/spinlock.h>
+
+#define NR_PHY_CHAN	6
+#define DMA_ALIGN	3
+#define DMA_MAX_SIZE	0x1fff
+#define DMA_CHUNK_SIZE	0x1000
+
+#define DMA_DDAR	0x00
+#define DMA_DCSR_S	0x04
+#define DMA_DCSR_C	0x08
+#define DMA_DCSR_R	0x0c
+#define DMA_DBSA	0x10
+#define DMA_DBTA	0x14
+#define DMA_DBSB	0x18
+#define DMA_DBTB	0x1c
+#define DMA_SIZE	0x20
+
+#define DCSR_RUN	(1 << 0)
+#define DCSR_IE		(1 << 1)
+#define DCSR_ERROR	(1 << 2)
+#define DCSR_DONEA	(1 << 3)
+#define DCSR_STRTA	(1 << 4)
+#define DCSR_DONEB	(1 << 5)
+#define DCSR_STRTB	(1 << 6)
+#define DCSR_BIU	(1 << 7)
+
+#define DDAR_RW		(1 << 0)	/* 0 = W, 1 = R */
+#define DDAR_E		(1 << 1)	/* 0 = LE, 1 = BE */
+#define DDAR_BS		(1 << 2)	/* 0 = BS4, 1 = BS8 */
+#define DDAR_DW		(1 << 3)	/* 0 = 8b, 1 = 16b */
+#define DDAR_Ser0UDCTr	(0x0 << 4)
+#define DDAR_Ser0UDCRc	(0x1 << 4)
+#define DDAR_Ser1SDLCTr	(0x2 << 4)
+#define DDAR_Ser1SDLCRc	(0x3 << 4)
+#define DDAR_Ser1UARTTr	(0x4 << 4)
+#define DDAR_Ser1UARTRc	(0x5 << 4)
+#define DDAR_Ser2ICPTr	(0x6 << 4)
+#define DDAR_Ser2ICPRc	(0x7 << 4)
+#define DDAR_Ser3UARTTr	(0x8 << 4)
+#define DDAR_Ser3UARTRc	(0x9 << 4)
+#define DDAR_Ser4MCP0Tr	(0xa << 4)
+#define DDAR_Ser4MCP0Rc	(0xb << 4)
+#define DDAR_Ser4MCP1Tr	(0xc << 4)
+#define DDAR_Ser4MCP1Rc	(0xd << 4)
+#define DDAR_Ser4SSPTr	(0xe << 4)
+#define DDAR_Ser4SSPRc	(0xf << 4)
+
+struct sa11x0_dma_sg {
+	u32			addr;
+	u32			len;
+};
+
+struct sa11x0_dma_desc {
+	struct dma_async_tx_descriptor tx;
+	u32			ddar;
+	size_t			size;
+
+	/* maybe protected by c->lock */
+	struct list_head	node;
+	unsigned		sglen;
+	struct sa11x0_dma_sg	sg[0];
+};
+
+struct sa11x0_dma_phy;
+
+struct sa11x0_dma_chan {
+	struct dma_chan		chan;
+	spinlock_t		lock;
+	dma_cookie_t		lc;
+
+	/* protected by c->lock */
+	struct sa11x0_dma_phy	*phy;
+	enum dma_status		status;
+	struct list_head	desc_submitted;
+	struct list_head	desc_issued;
+
+	/* protected by d->lock */
+	struct list_head	node;
+
+	u32			ddar;
+	const char		*name;
+};
+
+struct sa11x0_dma_phy {
+	void __iomem		*base;
+	struct sa11x0_dma_dev	*dev;
+	unsigned		num;
+
+	struct sa11x0_dma_chan	*vchan;
+
+	/* Protected by c->lock */
+	unsigned		sg_load;
+	struct sa11x0_dma_desc	*txd_load;
+	unsigned		sg_done;
+	struct sa11x0_dma_desc	*txd_done;
+#ifdef CONFIG_PM_SLEEP
+	u32			dbs[2];
+	u32			dbt[2];
+	u32			dcsr;
+#endif
+};
+
+struct sa11x0_dma_dev {
+	struct dma_device	slave;
+	void __iomem		*base;
+	spinlock_t		lock;
+	struct tasklet_struct	task;
+	struct list_head	chan_pending;
+	struct list_head	desc_complete;
+	struct sa11x0_dma_phy	phy[NR_PHY_CHAN];
+};
+
+static struct sa11x0_dma_chan *to_sa11x0_dma_chan(struct dma_chan *chan)
+{
+	return container_of(chan, struct sa11x0_dma_chan, chan);
+}
+
+static struct sa11x0_dma_dev *to_sa11x0_dma(struct dma_device *dmadev)
+{
+	return container_of(dmadev, struct sa11x0_dma_dev, slave);
+}
+
+static struct sa11x0_dma_desc *to_sa11x0_dma_tx(struct dma_async_tx_descriptor *tx)
+{
+	return container_of(tx, struct sa11x0_dma_desc, tx);
+}
+
+static struct sa11x0_dma_desc *sa11x0_dma_next_desc(struct sa11x0_dma_chan *c)
+{
+	if (list_empty(&c->desc_issued))
+		return NULL;
+
+	return list_first_entry(&c->desc_issued, struct sa11x0_dma_desc, node);
+}
+
+static void sa11x0_dma_start_desc(struct sa11x0_dma_phy *p, struct sa11x0_dma_desc *txd)
+{
+	list_del(&txd->node);
+	p->txd_load = txd;
+	p->sg_load = 0;
+
+	dev_vdbg(p->dev->slave.dev, "pchan %u: txd %p[%x]: starting: DDAR:%x\n",
+		p->num, txd, txd->tx.cookie, txd->ddar);
+}
+
+static void noinline sa11x0_dma_start_sg(struct sa11x0_dma_phy *p,
+	struct sa11x0_dma_chan *c)
+{
+	struct sa11x0_dma_desc *txd = p->txd_load;
+	struct sa11x0_dma_sg *sg;
+	void __iomem *base = p->base;
+	unsigned dbsx, dbtx;
+	u32 dcsr;
+
+	if (!txd)
+		return;
+
+	dcsr = readl_relaxed(base + DMA_DCSR_R);
+
+	/* Don't try to load the next transfer if both buffers are started */
+	if ((dcsr & (DCSR_STRTA | DCSR_STRTB)) == (DCSR_STRTA | DCSR_STRTB))
+		return;
+
+	if (p->sg_load == txd->sglen) {
+		struct sa11x0_dma_desc *txn = sa11x0_dma_next_desc(c);
+
+		/*
+		 * We have reached the end of the current descriptor.
+		 * Peek at the next descriptor, and if compatible with
+		 * the current, start processing it.
+		 */
+		if (txn && txn->ddar == txd->ddar) {
+			txd = txn;
+			sa11x0_dma_start_desc(p, txn);
+		} else {
+			p->txd_load = NULL;
+			return;
+		}
+	}
+
+	sg = &txd->sg[p->sg_load++];
+
+	/* Select buffer to load according to channel status */
+	if (((dcsr & (DCSR_BIU | DCSR_STRTB)) == (DCSR_BIU | DCSR_STRTB)) ||
+	    ((dcsr & (DCSR_BIU | DCSR_STRTA)) == 0)) {
+		dbsx = DMA_DBSA;
+		dbtx = DMA_DBTA;
+		dcsr = DCSR_STRTA | DCSR_IE | DCSR_RUN;
+	} else {
+		dbsx = DMA_DBSB;
+		dbtx = DMA_DBTB;
+		dcsr = DCSR_STRTB | DCSR_IE | DCSR_RUN;
+	}
+
+	writel_relaxed(sg->addr, base + dbsx);
+	writel_relaxed(sg->len, base + dbtx);
+	writel(dcsr, base + DMA_DCSR_S);
+
+	dev_dbg(p->dev->slave.dev, "pchan %u: load: DCSR:%02x DBS%c:%08x DBT%c:%08x\n",
+		p->num, dcsr,
+		'A' + (dbsx == DMA_DBSB), sg->addr,
+		'A' + (dbtx == DMA_DBTB), sg->len);
+}
+
+static void noinline sa11x0_dma_complete(struct sa11x0_dma_phy *p,
+	struct sa11x0_dma_chan *c)
+{
+	struct sa11x0_dma_desc *txd = p->txd_done;
+
+	if (++p->sg_done == txd->sglen) {
+		struct sa11x0_dma_dev *d = p->dev;
+
+		dev_vdbg(d->slave.dev, "pchan %u: txd %p[%x]: completed\n",
+			p->num, p->txd_done, p->txd_done->tx.cookie);
+
+		c->lc = txd->tx.cookie;
+
+		spin_lock(&d->lock);
+		list_add_tail(&txd->node, &d->desc_complete);
+		spin_unlock(&d->lock);
+
+		p->sg_done = 0;
+		p->txd_done = p->txd_load;
+
+		tasklet_schedule(&d->task);
+	}
+
+	sa11x0_dma_start_sg(p, c);
+}
+
+static irqreturn_t sa11x0_dma_irq(int irq, void *dev_id)
+{
+	struct sa11x0_dma_phy *p = dev_id;
+	struct sa11x0_dma_dev *d = p->dev;
+	struct sa11x0_dma_chan *c;
+	u32 dcsr;
+
+	dcsr = readl_relaxed(p->base + DMA_DCSR_R);
+	if (!(dcsr & (DCSR_ERROR | DCSR_DONEA | DCSR_DONEB)))
+		return IRQ_NONE;
+
+	/* Clear reported status bits */
+	writel_relaxed(dcsr & (DCSR_ERROR | DCSR_DONEA | DCSR_DONEB),
+		p->base + DMA_DCSR_C);
+
+	dev_dbg(d->slave.dev, "pchan %u: irq: DCSR:%02x\n", p->num, dcsr);
+
+	if (dcsr & DCSR_ERROR) {
+		dev_err(d->slave.dev, "pchan %u: error. DCSR:%02x DDAR:%08x DBSA:%08x DBTA:%08x DBSB:%08x DBTB:%08x\n",
+			p->num, dcsr,
+			readl_relaxed(p->base + DMA_DDAR),
+			readl_relaxed(p->base + DMA_DBSA),
+			readl_relaxed(p->base + DMA_DBTA),
+			readl_relaxed(p->base + DMA_DBSB),
+			readl_relaxed(p->base + DMA_DBTB));
+	}
+
+	c = p->vchan;
+	if (c) {
+		unsigned long flags;
+
+		spin_lock_irqsave(&c->lock, flags);
+		/*
+		 * Now that we're holding the lock, check that the vchan
+		 * really is associated with this pchan before touching the
+		 * hardware.  This should always succeed, because we won't
+		 * change p->vchan or c->phy while the channel is actively
+		 * transferring.
+		 */
+		if (c->phy == p) {
+			if (dcsr & DCSR_DONEA)
+				sa11x0_dma_complete(p, c);
+			if (dcsr & DCSR_DONEB)
+				sa11x0_dma_complete(p, c);
+		}
+		spin_unlock_irqrestore(&c->lock, flags);
+	}
+
+	return IRQ_HANDLED;
+}
+
+static void sa11x0_dma_start_txd(struct sa11x0_dma_chan *c)
+{
+	struct sa11x0_dma_desc *txd = sa11x0_dma_next_desc(c);
+
+	/* If the issued list is empty, we have no further txds to process */
+	if (txd) {
+		struct sa11x0_dma_phy *p = c->phy;
+
+		sa11x0_dma_start_desc(p, txd);
+		p->txd_done = txd;
+		p->sg_done = 0;
+
+		/* The channel should not have any transfers started */
+		WARN_ON(readl_relaxed(p->base + DMA_DCSR_R) &
+				      (DCSR_STRTA | DCSR_STRTB));
+
+		/* Clear the run and start bits before changing DDAR */
+		writel_relaxed(DCSR_RUN | DCSR_STRTA | DCSR_STRTB,
+			       p->base + DMA_DCSR_C);
+		writel_relaxed(txd->ddar, p->base + DMA_DDAR);
+
+		/* Try to start both buffers */
+		sa11x0_dma_start_sg(p, c);
+		sa11x0_dma_start_sg(p, c);
+	}
+}
+
+static void sa11x0_dma_tasklet(unsigned long arg)
+{
+	struct sa11x0_dma_dev *d = (struct sa11x0_dma_dev *)arg;
+	struct sa11x0_dma_phy *p;
+	struct sa11x0_dma_chan *c;
+	struct sa11x0_dma_desc *txd, *txn;
+	LIST_HEAD(head);
+	unsigned pch, pch_alloc = 0;
+
+	dev_dbg(d->slave.dev, "tasklet enter\n");
+
+	/* Get the completed tx descriptors */
+	spin_lock_irq(&d->lock);
+	list_splice_init(&d->desc_complete, &head);
+	spin_unlock_irq(&d->lock);
+
+	list_for_each_entry(txd, &head, node) {
+		c = to_sa11x0_dma_chan(txd->tx.chan);
+
+		dev_dbg(d->slave.dev, "vchan %p: txd %p[%x] completed\n",
+			c, txd, txd->tx.cookie);
+
+		spin_lock_irq(&c->lock);
+		p = c->phy;
+		if (p) {
+			if (!p->txd_done)
+				sa11x0_dma_start_txd(c);
+			if (!p->txd_done) {
+				/* No current txd associated with this channel */
+				dev_dbg(d->slave.dev, "pchan %u: free\n", p->num);
+
+				/* Mark this channel free */
+				c->phy = NULL;
+				p->vchan = NULL;
+			}
+		}
+		spin_unlock_irq(&c->lock);
+	}
+
+	spin_lock_irq(&d->lock);
+	for (pch = 0; pch < NR_PHY_CHAN; pch++) {
+		p = &d->phy[pch];
+
+		if (p->vchan == NULL && !list_empty(&d->chan_pending)) {
+			c = list_first_entry(&d->chan_pending,
+				struct sa11x0_dma_chan, node);
+			list_del_init(&c->node);
+
+			pch_alloc |= 1 << pch;
+
+			/* Mark this channel allocated */
+			p->vchan = c;
+
+			dev_dbg(d->slave.dev, "pchan %u: alloc vchan %p\n", pch, c);
+		}
+	}
+	spin_unlock_irq(&d->lock);
+
+	for (pch = 0; pch < NR_PHY_CHAN; pch++) {
+		if (pch_alloc & (1 << pch)) {
+			p = &d->phy[pch];
+			c = p->vchan;
+
+			spin_lock_irq(&c->lock);
+			c->phy = p;
+
+			sa11x0_dma_start_txd(c);
+			spin_unlock_irq(&c->lock);
+		}
+	}
+
+	/* Now free the completed tx descriptor, and call their callbacks */
+	list_for_each_entry_safe(txd, txn, &head, node) {
+		dma_async_tx_callback callback = txd->tx.callback;
+		void *callback_param = txd->tx.callback_param;
+
+		dev_dbg(d->slave.dev, "txd %p[%x]: callback and free\n",
+			txd, txd->tx.cookie);
+
+		kfree(txd);
+
+		if (callback)
+			callback(callback_param);
+	}
+
+	dev_dbg(d->slave.dev, "tasklet exit\n");
+}
+
+
+static void sa11x0_dma_desc_free(struct sa11x0_dma_dev *d, struct list_head *head)
+{
+	struct sa11x0_dma_desc *txd, *txn;
+
+	list_for_each_entry_safe(txd, txn, head, node) {
+		dev_dbg(d->slave.dev, "txd %p: freeing\n", txd);
+		kfree(txd);
+	}
+}
+
+static int sa11x0_dma_alloc_chan_resources(struct dma_chan *chan)
+{
+	return 0;
+}
+
+static void sa11x0_dma_free_chan_resources(struct dma_chan *chan)
+{
+	struct sa11x0_dma_chan *c = to_sa11x0_dma_chan(chan);
+	struct sa11x0_dma_dev *d = to_sa11x0_dma(chan->device);
+	unsigned long flags;
+	LIST_HEAD(head);
+
+	spin_lock_irqsave(&c->lock, flags);
+	spin_lock(&d->lock);
+	list_del_init(&c->node);
+	spin_unlock(&d->lock);
+
+	list_splice_tail_init(&c->desc_submitted, &head);
+	list_splice_tail_init(&c->desc_issued, &head);
+	spin_unlock_irqrestore(&c->lock, flags);
+
+	sa11x0_dma_desc_free(d, &head);
+}
+
+static dma_addr_t sa11x0_dma_pos(struct sa11x0_dma_phy *p)
+{
+	unsigned reg;
+	u32 dcsr;
+
+	dcsr = readl_relaxed(p->base + DMA_DCSR_R);
+
+	if ((dcsr & (DCSR_BIU | DCSR_STRTA)) == DCSR_STRTA ||
+	    (dcsr & (DCSR_BIU | DCSR_STRTB)) == DCSR_BIU)
+		reg = DMA_DBSA;
+	else
+		reg = DMA_DBSB;
+
+	return readl_relaxed(p->base + reg);
+}
+
+static enum dma_status sa11x0_dma_tx_status(struct dma_chan *chan,
+	dma_cookie_t cookie, struct dma_tx_state *state)
+{
+	struct sa11x0_dma_chan *c = to_sa11x0_dma_chan(chan);
+	struct sa11x0_dma_dev *d = to_sa11x0_dma(chan->device);
+	struct sa11x0_dma_phy *p;
+	struct sa11x0_dma_desc *txd;
+	dma_cookie_t last_used, last_complete;
+	unsigned long flags;
+	enum dma_status ret;
+	size_t bytes = 0;
+
+	last_used = c->chan.cookie;
+	last_complete = c->lc;
+
+	ret = dma_async_is_complete(cookie, last_complete, last_used);
+	if (ret == DMA_SUCCESS) {
+		dma_set_tx_state(state, last_complete, last_used, 0);
+		return ret;
+	}
+
+	spin_lock_irqsave(&c->lock, flags);
+	p = c->phy;
+	ret = c->status;
+	if (p) {
+		dma_addr_t addr = sa11x0_dma_pos(p);
+
+		dev_vdbg(d->slave.dev, "tx_status: addr:%x\n", addr);
+
+		txd = p->txd_done;
+		if (txd) {
+			unsigned i;
+
+			for (i = 0; i < txd->sglen; i++) {
+				dev_vdbg(d->slave.dev, "tx_status: [%u] %x+%x\n",
+					i, txd->sg[i].addr, txd->sg[i].len);
+				if (addr >= txd->sg[i].addr &&
+				    addr < txd->sg[i].addr + txd->sg[i].len) {
+					unsigned len;
+
+					len = txd->sg[i].len -
+						(addr - txd->sg[i].addr);
+					dev_vdbg(d->slave.dev, "tx_status: [%u] +%x\n",
+						i, len);
+					bytes += len;
+					i++;
+					break;
+				}
+			}
+			for (; i < txd->sglen; i++) {
+				dev_vdbg(d->slave.dev, "tx_status: [%u] %x+%x ++\n",
+					i, txd->sg[i].addr, txd->sg[i].len);
+				bytes += txd->sg[i].len;
+			}
+		}
+		if (txd != p->txd_load && p->txd_load)
+			bytes += p->txd_load->size;
+	}
+	list_for_each_entry(txd, &c->desc_issued, node) {
+		bytes += txd->size;
+	}
+	spin_unlock_irqrestore(&c->lock, flags);
+
+	dma_set_tx_state(state, last_complete, last_used, bytes);
+
+	dev_vdbg(d->slave.dev, "tx_status: bytes 0x%zx\n", bytes);
+
+	return ret;
+}
+
+/*
+ * Move pending txds to the issued list, and re-init pending list.
+ * If not already pending, add this channel to the list of pending
+ * channels and trigger the tasklet to run.
+ */
+static void sa11x0_dma_issue_pending(struct dma_chan *chan)
+{
+	struct sa11x0_dma_chan *c = to_sa11x0_dma_chan(chan);
+	struct sa11x0_dma_dev *d = to_sa11x0_dma(chan->device);
+	unsigned long flags;
+
+	spin_lock_irqsave(&c->lock, flags);
+	list_splice_tail_init(&c->desc_submitted, &c->desc_issued);
+	if (!list_empty(&c->desc_issued)) {
+		spin_lock(&d->lock);
+		if (!c->phy && list_empty(&c->node)) {
+			list_add_tail(&c->node, &d->chan_pending);
+			tasklet_schedule(&d->task);
+			dev_dbg(d->slave.dev, "vchan %p: issued\n", c);
+		}
+		spin_unlock(&d->lock);
+	} else
+		dev_dbg(d->slave.dev, "vchan %p: nothing to issue\n", c);
+	spin_unlock_irqrestore(&c->lock, flags);
+}
+
+static dma_cookie_t sa11x0_dma_tx_submit(struct dma_async_tx_descriptor *tx)
+{
+	struct sa11x0_dma_chan *c = to_sa11x0_dma_chan(tx->chan);
+	struct sa11x0_dma_desc *txd = to_sa11x0_dma_tx(tx);
+	unsigned long flags;
+
+	spin_lock_irqsave(&c->lock, flags);
+	c->chan.cookie += 1;
+	if (c->chan.cookie < 0)
+		c->chan.cookie = 1;
+	txd->tx.cookie = c->chan.cookie;
+
+	list_add_tail(&txd->node, &c->desc_submitted);
+	spin_unlock_irqrestore(&c->lock, flags);
+
+	dev_dbg(tx->chan->device->dev, "vchan %p: txd %p[%x]: submitted\n",
+		c, txd, txd->tx.cookie);
+
+	return txd->tx.cookie;
+}
+
+static struct dma_async_tx_descriptor *sa11x0_dma_prep_slave_sg(
+	struct dma_chan *chan, struct scatterlist *sg, unsigned int sglen,
+	enum dma_transfer_direction dir, unsigned long flags)
+{
+	struct sa11x0_dma_chan *c = to_sa11x0_dma_chan(chan);
+	struct sa11x0_dma_desc *txd;
+	struct scatterlist *sgent;
+	unsigned i, j = sglen;
+	size_t size = 0;
+
+	/* SA11x0 channels can only operate in their native direction */
+	if (dir != (c->ddar & DDAR_RW ? DMA_DEV_TO_MEM : DMA_MEM_TO_DEV)) {
+		dev_err(chan->device->dev, "vchan %p: bad DMA direction: DDAR:%08x dir:%u\n",
+			c, c->ddar, dir);
+		return NULL;
+	}
+
+	/* Do not allow zero-sized txds */
+	if (sglen == 0)
+		return NULL;
+
+	for_each_sg(sg, sgent, sglen, i) {
+		dma_addr_t addr = sg_dma_address(sgent);
+		unsigned int len = sg_dma_len(sgent);
+
+		if (len > DMA_MAX_SIZE)
+			j += DIV_ROUND_UP(len, DMA_MAX_SIZE & ~DMA_ALIGN) - 1;
+		if (addr & DMA_ALIGN) {
+			dev_dbg(chan->device->dev, "vchan %p: bad buffer alignment: %08x\n",
+				c, addr);
+			return NULL;
+		}
+	}
+
+	txd = kzalloc(sizeof(*txd) + j * sizeof(txd->sg[0]), GFP_ATOMIC);
+	if (!txd) {
+		dev_dbg(chan->device->dev, "vchan %p: kzalloc failed\n", c);
+		return NULL;
+	}
+
+	j = 0;
+	for_each_sg(sg, sgent, sglen, i) {
+		dma_addr_t addr = sg_dma_address(sgent);
+		unsigned len = sg_dma_len(sgent);
+
+		size += len;
+
+		do {
+			unsigned tlen = len;
+
+			/*
+			 * Check whether the transfer will fit.  If not, try
+			 * to split the transfer up such that we end up with
+			 * equal chunks - but make sure that we preserve the
+			 * alignment.  This avoids small segments.
+			 */
+			if (tlen > DMA_MAX_SIZE) {
+				unsigned mult = DIV_ROUND_UP(tlen,
+					DMA_MAX_SIZE & ~DMA_ALIGN);
+
+				tlen = (tlen / mult) & ~DMA_ALIGN;
+			}
+
+			txd->sg[j].addr = addr;
+			txd->sg[j].len = tlen;
+
+			addr += tlen;
+			len -= tlen;
+			j++;
+		} while (len);
+	}
+
+	dma_async_tx_descriptor_init(&txd->tx, &c->chan);
+	txd->tx.flags = flags;
+	txd->tx.tx_submit = sa11x0_dma_tx_submit;
+	txd->ddar = c->ddar;
+	txd->size = size;
+	txd->sglen = j;
+
+	dev_dbg(chan->device->dev, "vchan %p: txd %p: size %u nr %u\n",
+		c, txd, txd->size, txd->sglen);
+
+	return &txd->tx;
+}
+
+static int sa11x0_dma_slave_config(struct sa11x0_dma_chan *c, struct dma_slave_config *cfg)
+{
+	u32 ddar = c->ddar & ((0xf << 4) | DDAR_RW);
+	dma_addr_t addr;
+	enum dma_slave_buswidth width;
+	u32 maxburst;
+
+	if (ddar & DDAR_RW) {
+		addr = cfg->src_addr;
+		width = cfg->src_addr_width;
+		maxburst = cfg->src_maxburst;
+	} else {
+		addr = cfg->dst_addr;
+		width = cfg->dst_addr_width;
+		maxburst = cfg->dst_maxburst;
+	}
+
+	if ((width != DMA_SLAVE_BUSWIDTH_1_BYTE &&
+	     width != DMA_SLAVE_BUSWIDTH_2_BYTES) ||
+	    (maxburst != 4 && maxburst != 8))
+		return -EINVAL;
+
+	if (width == DMA_SLAVE_BUSWIDTH_2_BYTES)
+		ddar |= DDAR_DW;
+	if (maxburst == 8)
+		ddar |= DDAR_BS;
+
+	dev_dbg(c->chan.device->dev, "vchan %p: dma_slave_config addr %x width %u burst %u\n",
+		c, addr, width, maxburst);
+
+	c->ddar = ddar | (addr & 0xf0000000) | (addr & 0x003ffffc) << 6;
+
+	return 0;
+}
+
+static int sa11x0_dma_control(struct dma_chan *chan, enum dma_ctrl_cmd cmd,
+	unsigned long arg)
+{
+	struct sa11x0_dma_chan *c = to_sa11x0_dma_chan(chan);
+	struct sa11x0_dma_dev *d = to_sa11x0_dma(chan->device);
+	struct sa11x0_dma_phy *p;
+	LIST_HEAD(head);
+	unsigned long flags;
+	int ret;
+
+	switch (cmd) {
+	case DMA_SLAVE_CONFIG:
+		return sa11x0_dma_slave_config(c, (struct dma_slave_config *)arg);
+
+	case DMA_TERMINATE_ALL:
+		dev_dbg(d->slave.dev, "vchan %p: terminate all\n", c);
+		/* Clear the tx descriptor lists */
+		spin_lock_irqsave(&c->lock, flags);
+		list_splice_tail_init(&c->desc_submitted, &head);
+		list_splice_tail_init(&c->desc_issued, &head);
+
+		p = c->phy;
+		if (p) {
+			struct sa11x0_dma_desc *txd, *txn;
+
+			dev_dbg(d->slave.dev, "pchan %u: terminating\n", p->num);
+			/* vchan is assigned to a pchan - stop the channel */
+			writel(DCSR_RUN | DCSR_IE |
+				DCSR_STRTA | DCSR_DONEA |
+				DCSR_STRTB | DCSR_DONEB,
+				p->base + DMA_DCSR_C);
+
+			list_for_each_entry_safe(txd, txn, &d->desc_complete, node)
+				if (txd->tx.chan == &c->chan)
+					list_move(&txd->node, &head);
+
+			if (p->txd_load) {
+				if (p->txd_load != p->txd_done)
+					list_add_tail(&p->txd_load->node, &head);
+				p->txd_load = NULL;
+			}
+			if (p->txd_done) {
+				list_add_tail(&p->txd_done->node, &head);
+				p->txd_done = NULL;
+			}
+			c->phy = NULL;
+			spin_lock(&d->lock);
+			p->vchan = NULL;
+			spin_unlock(&d->lock);
+			tasklet_schedule(&d->task);
+		}
+		spin_unlock_irqrestore(&c->lock, flags);
+		sa11x0_dma_desc_free(d, &head);
+		ret = 0;
+		break;
+
+	case DMA_PAUSE:
+		dev_dbg(d->slave.dev, "vchan %p: pause\n", c);
+		spin_lock_irqsave(&c->lock, flags);
+		if (c->status == DMA_IN_PROGRESS) {
+			c->status = DMA_PAUSED;
+
+			p = c->phy;
+			if (p) {
+				writel(DCSR_RUN | DCSR_IE, p->base + DMA_DCSR_C);
+			} else {
+				spin_lock(&d->lock);
+				list_del_init(&c->node);
+				spin_unlock(&d->lock);
+			}
+		}
+		spin_unlock_irqrestore(&c->lock, flags);
+		ret = 0;
+		break;
+
+	case DMA_RESUME:
+		dev_dbg(d->slave.dev, "vchan %p: resume\n", c);
+		spin_lock_irqsave(&c->lock, flags);
+		if (c->status == DMA_PAUSED) {
+			c->status = DMA_IN_PROGRESS;
+
+			p = c->phy;
+			if (p) {
+				writel(DCSR_RUN | DCSR_IE, p->base + DMA_DCSR_S);
+			} else if (!list_empty(&c->desc_issued)) {
+				spin_lock(&d->lock);
+				list_add_tail(&c->node, &d->chan_pending);
+				spin_unlock(&d->lock);
+			}
+		}
+		spin_unlock_irqrestore(&c->lock, flags);
+		ret = 0;
+		break;
+
+	default:
+		ret = -ENXIO;
+		break;
+	}
+
+	return ret;
+}
+
+struct sa11x0_dma_channel_desc {
+	u32 ddar;
+	const char *name;
+};
+
+#define CD(d1, d2) { .ddar = DDAR_##d1 | d2, .name = #d1 }
+static const struct sa11x0_dma_channel_desc chan_desc[] = {
+	CD(Ser0UDCTr, 0),
+	CD(Ser0UDCRc, DDAR_RW),
+	CD(Ser1SDLCTr, 0),
+	CD(Ser1SDLCRc, DDAR_RW),
+	CD(Ser1UARTTr, 0),
+	CD(Ser1UARTRc, DDAR_RW),
+	CD(Ser2ICPTr, 0),
+	CD(Ser2ICPRc, DDAR_RW),
+	CD(Ser3UARTTr, 0),
+	CD(Ser3UARTRc, DDAR_RW),
+	CD(Ser4MCP0Tr, 0),
+	CD(Ser4MCP0Rc, DDAR_RW),
+	CD(Ser4MCP1Tr, 0),
+	CD(Ser4MCP1Rc, DDAR_RW),
+	CD(Ser4SSPTr, 0),
+	CD(Ser4SSPRc, DDAR_RW),
+};
+
+static int __devinit sa11x0_dma_init_dmadev(struct dma_device *dmadev,
+	struct device *dev)
+{
+	unsigned i;
+
+	dmadev->chancnt = ARRAY_SIZE(chan_desc);
+	INIT_LIST_HEAD(&dmadev->channels);
+	dmadev->dev = dev;
+	dmadev->device_alloc_chan_resources = sa11x0_dma_alloc_chan_resources;
+	dmadev->device_free_chan_resources = sa11x0_dma_free_chan_resources;
+	dmadev->device_control = sa11x0_dma_control;
+	dmadev->device_tx_status = sa11x0_dma_tx_status;
+	dmadev->device_issue_pending = sa11x0_dma_issue_pending;
+
+	for (i = 0; i < dmadev->chancnt; i++) {
+		struct sa11x0_dma_chan *c;
+
+		c = kzalloc(sizeof(*c), GFP_KERNEL);
+		if (!c) {
+			dev_err(dev, "no memory for channel %u\n", i);
+			return -ENOMEM;
+		}
+
+		c->chan.device = dmadev;
+		c->status = DMA_IN_PROGRESS;
+		c->ddar = chan_desc[i].ddar;
+		c->name = chan_desc[i].name;
+		spin_lock_init(&c->lock);
+		INIT_LIST_HEAD(&c->desc_submitted);
+		INIT_LIST_HEAD(&c->desc_issued);
+		INIT_LIST_HEAD(&c->node);
+		list_add_tail(&c->chan.device_node, &dmadev->channels);
+	}
+
+	return dma_async_device_register(dmadev);
+}
+
+static int sa11x0_dma_request_irq(struct platform_device *pdev, int nr,
+	void *data)
+{
+	int irq = platform_get_irq(pdev, nr);
+
+	if (irq <= 0)
+		return -ENXIO;
+
+	return request_irq(irq, sa11x0_dma_irq, 0, dev_name(&pdev->dev), data);
+}
+
+static void sa11x0_dma_free_irq(struct platform_device *pdev, int nr,
+	void *data)
+{
+	int irq = platform_get_irq(pdev, nr);
+	if (irq > 0)
+		free_irq(irq, data);
+}
+
+static void sa11x0_dma_free_channels(struct dma_device *dmadev)
+{
+	struct sa11x0_dma_chan *c, *cn;
+
+	list_for_each_entry_safe(c, cn, &dmadev->channels, chan.device_node) {
+		list_del(&c->chan.device_node);
+		kfree(c);
+	}
+}
+
+static int __devinit sa11x0_dma_probe(struct platform_device *pdev)
+{
+	struct sa11x0_dma_dev *d;
+	struct resource *res;
+	unsigned i;
+	int ret;
+
+	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+	if (!res)
+		return -ENXIO;
+
+	d = kzalloc(sizeof(*d), GFP_KERNEL);
+	if (!d) {
+		ret = -ENOMEM;
+		goto err_alloc;
+	}
+
+	spin_lock_init(&d->lock);
+	INIT_LIST_HEAD(&d->chan_pending);
+	INIT_LIST_HEAD(&d->desc_complete);
+
+	d->base = ioremap(res->start, resource_size(res));
+	if (!d->base) {
+		ret = -ENOMEM;
+		goto err_ioremap;
+	}
+
+	tasklet_init(&d->task, sa11x0_dma_tasklet, (unsigned long)d);
+
+	for (i = 0; i < NR_PHY_CHAN; i++) {
+		struct sa11x0_dma_phy *p = &d->phy[i];
+
+		p->dev = d;
+		p->num = i;
+		p->base = d->base + i * DMA_SIZE;
+		writel_relaxed(DCSR_RUN | DCSR_IE | DCSR_ERROR |
+			DCSR_DONEA | DCSR_STRTA | DCSR_DONEB | DCSR_STRTB,
+			p->base + DMA_DCSR_C);
+		writel_relaxed(0, p->base + DMA_DDAR);
+
+		ret = sa11x0_dma_request_irq(pdev, i, p);
+		if (ret) {
+			while (i) {
+				i--;
+				sa11x0_dma_free_irq(pdev, i, &d->phy[i]);
+			}
+			goto err_irq;
+		}
+	}
+
+	dma_cap_set(DMA_SLAVE, d->slave.cap_mask);
+	d->slave.device_prep_slave_sg = sa11x0_dma_prep_slave_sg;
+	ret = sa11x0_dma_init_dmadev(&d->slave, &pdev->dev);
+	if (ret) {
+		dev_warn(d->slave.dev, "failed to register slave async device: %d\n",
+			ret);
+		goto err_slave_reg;
+	}
+
+	platform_set_drvdata(pdev, d);
+	return 0;
+
+ err_slave_reg:
+	sa11x0_dma_free_channels(&d->slave);
+	for (i = 0; i < NR_PHY_CHAN; i++)
+		sa11x0_dma_free_irq(pdev, i, &d->phy[i]);
+ err_irq:
+	tasklet_kill(&d->task);
+	iounmap(d->base);
+ err_ioremap:
+	kfree(d);
+ err_alloc:
+	return ret;
+}
+
+static int __devexit sa11x0_dma_remove(struct platform_device *pdev)
+{
+	struct sa11x0_dma_dev *d = platform_get_drvdata(pdev);
+	unsigned pch;
+
+	dma_async_device_unregister(&d->slave);
+
+	sa11x0_dma_free_channels(&d->slave);
+	for (pch = 0; pch < NR_PHY_CHAN; pch++)
+		sa11x0_dma_free_irq(pdev, pch, &d->phy[pch]);
+	tasklet_kill(&d->task);
+	iounmap(d->base);
+	kfree(d);
+
+	return 0;
+}
+
+#ifdef CONFIG_PM_SLEEP
+static int sa11x0_dma_suspend(struct device *dev)
+{
+	struct sa11x0_dma_dev *d = dev_get_drvdata(dev);
+	unsigned pch;
+
+	for (pch = 0; pch < NR_PHY_CHAN; pch++) {
+		struct sa11x0_dma_phy *p = &d->phy[pch];
+		u32 dcsr, saved_dcsr;
+
+		dcsr = saved_dcsr = readl_relaxed(p->base + DMA_DCSR_R);
+		if (dcsr & DCSR_RUN) {
+			writel(DCSR_RUN | DCSR_IE, p->base + DMA_DCSR_C);
+			dcsr = readl_relaxed(p->base + DMA_DCSR_R);
+		}
+
+		saved_dcsr &= DCSR_RUN | DCSR_IE;
+		if (dcsr & DCSR_BIU) {
+			p->dbs[0] = readl_relaxed(p->base + DMA_DBSB);
+			p->dbt[0] = readl_relaxed(p->base + DMA_DBTB);
+			p->dbs[1] = readl_relaxed(p->base + DMA_DBSA);
+			p->dbt[1] = readl_relaxed(p->base + DMA_DBTA);
+			saved_dcsr |= (dcsr & DCSR_STRTA ? DCSR_STRTB : 0) |
+				      (dcsr & DCSR_STRTB ? DCSR_STRTA : 0);
+		} else {
+			p->dbs[0] = readl_relaxed(p->base + DMA_DBSA);
+			p->dbt[0] = readl_relaxed(p->base + DMA_DBTA);
+			p->dbs[1] = readl_relaxed(p->base + DMA_DBSB);
+			p->dbt[1] = readl_relaxed(p->base + DMA_DBTB);
+			saved_dcsr |= dcsr & (DCSR_STRTA | DCSR_STRTB);
+		}
+		p->dcsr = saved_dcsr;
+
+		writel(DCSR_STRTA | DCSR_STRTB, p->base + DMA_DCSR_C);
+	}
+
+	return 0;
+}
+
+static int sa11x0_dma_resume(struct device *dev)
+{
+	struct sa11x0_dma_dev *d = dev_get_drvdata(dev);
+	unsigned pch;
+
+	for (pch = 0; pch < NR_PHY_CHAN; pch++) {
+		struct sa11x0_dma_phy *p = &d->phy[pch];
+		struct sa11x0_dma_desc *txd = NULL;
+		u32 dcsr = readl_relaxed(p->base + DMA_DCSR_R);
+
+		WARN_ON(dcsr & (DCSR_BIU | DCSR_STRTA | DCSR_STRTB | DCSR_RUN));
+
+		if (p->txd_done)
+			txd = p->txd_done;
+		else if (p->txd_load)
+			txd = p->txd_load;
+
+		if (!txd)
+			continue;
+
+		writel_relaxed(txd->ddar, p->base + DMA_DDAR);
+
+		writel_relaxed(p->dbs[0], p->base + DMA_DBSA);
+		writel_relaxed(p->dbt[0], p->base + DMA_DBTA);
+		writel_relaxed(p->dbs[1], p->base + DMA_DBSB);
+		writel_relaxed(p->dbt[1], p->base + DMA_DBTB);
+		writel_relaxed(p->dcsr, p->base + DMA_DCSR_S);
+	}
+
+	return 0;
+}
+#endif
+
+static const struct dev_pm_ops sa11x0_dma_pm_ops = {
+	.suspend_noirq = sa11x0_dma_suspend,
+	.resume_noirq = sa11x0_dma_resume,
+	.freeze_noirq = sa11x0_dma_suspend,
+	.thaw_noirq = sa11x0_dma_resume,
+	.poweroff_noirq = sa11x0_dma_suspend,
+	.restore_noirq = sa11x0_dma_resume,
+};
+
+static struct platform_driver sa11x0_dma_driver = {
+	.driver = {
+		.name	= "sa11x0-dma",
+		.owner	= THIS_MODULE,
+		.pm	= &sa11x0_dma_pm_ops,
+	},
+	.probe		= sa11x0_dma_probe,
+	.remove		= __devexit_p(sa11x0_dma_remove),
+};
+
+bool sa11x0_dma_filter_fn(struct dma_chan *chan, void *param)
+{
+	if (chan->device->dev->driver == &sa11x0_dma_driver.driver) {
+		struct sa11x0_dma_chan *c = to_sa11x0_dma_chan(chan);
+		const char *p = param;
+
+		return !strcmp(c->name, p);
+	}
+	return false;
+}
+EXPORT_SYMBOL(sa11x0_dma_filter_fn);
+
+static int __init sa11x0_dma_init(void)
+{
+	return platform_driver_register(&sa11x0_dma_driver);
+}
+subsys_initcall(sa11x0_dma_init);
+
+static void __exit sa11x0_dma_exit(void)
+{
+	platform_driver_unregister(&sa11x0_dma_driver);
+}
+module_exit(sa11x0_dma_exit);
+
+MODULE_AUTHOR("Russell King");
+MODULE_DESCRIPTION("SA-11x0 DMA driver");
+MODULE_LICENSE("GPL v2");
+MODULE_ALIAS("platform:sa11x0-dma");
diff --git a/drivers/net/irda/Kconfig b/drivers/net/irda/Kconfig
index e535137..4680478 100644
--- a/drivers/net/irda/Kconfig
+++ b/drivers/net/irda/Kconfig
@@ -356,7 +356,7 @@
 
 config SA1100_FIR
 	tristate "SA1100 Internal IR"
-	depends on ARCH_SA1100 && IRDA
+	depends on ARCH_SA1100 && IRDA && DMA_SA11X0
 
 config VIA_FIR
 	tristate "VIA VT8231/VT1211 SIR/MIR/FIR"
diff --git a/drivers/net/irda/sa1100_ir.c b/drivers/net/irda/sa1100_ir.c
index da27050..a0d1913 100644
--- a/drivers/net/irda/sa1100_ir.c
+++ b/drivers/net/irda/sa1100_ir.c
@@ -15,7 +15,7 @@
  *  This driver takes one kernel command line parameter, sa1100ir=, with
  *  the following options:
  *	max_rate:baudrate	- set the maximum baud rate
- *	power_leve:level	- set the transmitter power level
+ *	power_level:level	- set the transmitter power level
  *	tx_lpm:0|1		- set transmit low power mode
  */
 #include <linux/module.h>
@@ -30,13 +30,13 @@
 #include <linux/delay.h>
 #include <linux/platform_device.h>
 #include <linux/dma-mapping.h>
+#include <linux/dmaengine.h>
+#include <linux/sa11x0-dma.h>
 
 #include <net/irda/irda.h>
 #include <net/irda/wrapper.h>
 #include <net/irda/irda_device.h>
 
-#include <asm/irq.h>
-#include <mach/dma.h>
 #include <mach/hardware.h>
 #include <asm/mach/irda.h>
 
@@ -44,8 +44,15 @@
 static int tx_lpm;
 static int max_rate = 4000000;
 
+struct sa1100_buf {
+	struct device		*dev;
+	struct sk_buff		*skb;
+	struct scatterlist	sg;
+	struct dma_chan		*chan;
+	dma_cookie_t		cookie;
+};
+
 struct sa1100_irda {
-	unsigned char		hscr0;
 	unsigned char		utcr4;
 	unsigned char		power;
 	unsigned char		open;
@@ -53,12 +60,8 @@
 	int			speed;
 	int			newspeed;
 
-	struct sk_buff		*txskb;
-	struct sk_buff		*rxskb;
-	dma_addr_t		txbuf_dma;
-	dma_addr_t		rxbuf_dma;
-	dma_regs_t		*txdma;
-	dma_regs_t		*rxdma;
+	struct sa1100_buf	dma_rx;
+	struct sa1100_buf	dma_tx;
 
 	struct device		*dev;
 	struct irda_platform_data *pdata;
@@ -67,23 +70,103 @@
 
 	iobuff_t		tx_buff;
 	iobuff_t		rx_buff;
+
+	int (*tx_start)(struct sk_buff *, struct net_device *, struct sa1100_irda *);
+	irqreturn_t (*irq)(struct net_device *, struct sa1100_irda *);
 };
 
+static int sa1100_irda_set_speed(struct sa1100_irda *, int);
+
 #define IS_FIR(si)		((si)->speed >= 4000000)
 
 #define HPSIR_MAX_RXLEN		2047
 
+static struct dma_slave_config sa1100_irda_sir_tx = {
+	.direction	= DMA_TO_DEVICE,
+	.dst_addr	= __PREG(Ser2UTDR),
+	.dst_addr_width	= DMA_SLAVE_BUSWIDTH_1_BYTE,
+	.dst_maxburst	= 4,
+};
+
+static struct dma_slave_config sa1100_irda_fir_rx = {
+	.direction	= DMA_FROM_DEVICE,
+	.src_addr	= __PREG(Ser2HSDR),
+	.src_addr_width	= DMA_SLAVE_BUSWIDTH_1_BYTE,
+	.src_maxburst	= 8,
+};
+
+static struct dma_slave_config sa1100_irda_fir_tx = {
+	.direction	= DMA_TO_DEVICE,
+	.dst_addr	= __PREG(Ser2HSDR),
+	.dst_addr_width	= DMA_SLAVE_BUSWIDTH_1_BYTE,
+	.dst_maxburst	= 8,
+};
+
+static unsigned sa1100_irda_dma_xferred(struct sa1100_buf *buf)
+{
+	struct dma_chan *chan = buf->chan;
+	struct dma_tx_state state;
+	enum dma_status status;
+
+	status = chan->device->device_tx_status(chan, buf->cookie, &state);
+	if (status != DMA_PAUSED)
+		return 0;
+
+	return sg_dma_len(&buf->sg) - state.residue;
+}
+
+static int sa1100_irda_dma_request(struct device *dev, struct sa1100_buf *buf,
+	const char *name, struct dma_slave_config *cfg)
+{
+	dma_cap_mask_t m;
+	int ret;
+
+	dma_cap_zero(m);
+	dma_cap_set(DMA_SLAVE, m);
+
+	buf->chan = dma_request_channel(m, sa11x0_dma_filter_fn, (void *)name);
+	if (!buf->chan) {
+		dev_err(dev, "unable to request DMA channel for %s\n",
+			name);
+		return -ENOENT;
+	}
+
+	ret = dmaengine_slave_config(buf->chan, cfg);
+	if (ret)
+		dev_warn(dev, "DMA slave_config for %s returned %d\n",
+			name, ret);
+
+	buf->dev = buf->chan->device->dev;
+
+	return 0;
+}
+
+static void sa1100_irda_dma_start(struct sa1100_buf *buf,
+	enum dma_transfer_direction dir, dma_async_tx_callback cb, void *cb_p)
+{
+	struct dma_async_tx_descriptor *desc;
+	struct dma_chan *chan = buf->chan;
+
+	desc = chan->device->device_prep_slave_sg(chan, &buf->sg, 1, dir,
+			DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
+	if (desc) {
+		desc->callback = cb;
+		desc->callback_param = cb_p;
+		buf->cookie = dmaengine_submit(desc);
+		dma_async_issue_pending(chan);
+	}
+}
+
 /*
  * Allocate and map the receive buffer, unless it is already allocated.
  */
 static int sa1100_irda_rx_alloc(struct sa1100_irda *si)
 {
-	if (si->rxskb)
+	if (si->dma_rx.skb)
 		return 0;
 
-	si->rxskb = alloc_skb(HPSIR_MAX_RXLEN + 1, GFP_ATOMIC);
-
-	if (!si->rxskb) {
+	si->dma_rx.skb = alloc_skb(HPSIR_MAX_RXLEN + 1, GFP_ATOMIC);
+	if (!si->dma_rx.skb) {
 		printk(KERN_ERR "sa1100_ir: out of memory for RX SKB\n");
 		return -ENOMEM;
 	}
@@ -92,11 +175,14 @@
 	 * Align any IP headers that may be contained
 	 * within the frame.
 	 */
-	skb_reserve(si->rxskb, 1);
+	skb_reserve(si->dma_rx.skb, 1);
 
-	si->rxbuf_dma = dma_map_single(si->dev, si->rxskb->data,
-					HPSIR_MAX_RXLEN,
-					DMA_FROM_DEVICE);
+	sg_set_buf(&si->dma_rx.sg, si->dma_rx.skb->data, HPSIR_MAX_RXLEN);
+	if (dma_map_sg(si->dma_rx.dev, &si->dma_rx.sg, 1, DMA_FROM_DEVICE) == 0) {
+		dev_kfree_skb_any(si->dma_rx.skb);
+		return -ENOMEM;
+	}
+
 	return 0;
 }
 
@@ -106,7 +192,7 @@
  */
 static void sa1100_irda_rx_dma_start(struct sa1100_irda *si)
 {
-	if (!si->rxskb) {
+	if (!si->dma_rx.skb) {
 		printk(KERN_ERR "sa1100_ir: rx buffer went missing\n");
 		return;
 	}
@@ -114,14 +200,343 @@
 	/*
 	 * First empty receive FIFO
 	 */
-	Ser2HSCR0 = si->hscr0 | HSCR0_HSSP;
+	Ser2HSCR0 = HSCR0_HSSP;
 
 	/*
 	 * Enable the DMA, receiver and receive interrupt.
 	 */
-	sa1100_clear_dma(si->rxdma);
-	sa1100_start_dma(si->rxdma, si->rxbuf_dma, HPSIR_MAX_RXLEN);
-	Ser2HSCR0 = si->hscr0 | HSCR0_HSSP | HSCR0_RXE;
+	dmaengine_terminate_all(si->dma_rx.chan);
+	sa1100_irda_dma_start(&si->dma_rx, DMA_DEV_TO_MEM, NULL, NULL);
+
+	Ser2HSCR0 = HSCR0_HSSP | HSCR0_RXE;
+}
+
+static void sa1100_irda_check_speed(struct sa1100_irda *si)
+{
+	if (si->newspeed) {
+		sa1100_irda_set_speed(si, si->newspeed);
+		si->newspeed = 0;
+	}
+}
+
+/*
+ * HP-SIR format support.
+ */
+static void sa1100_irda_sirtxdma_irq(void *id)
+{
+	struct net_device *dev = id;
+	struct sa1100_irda *si = netdev_priv(dev);
+
+	dma_unmap_sg(si->dma_tx.dev, &si->dma_tx.sg, 1, DMA_TO_DEVICE);
+	dev_kfree_skb(si->dma_tx.skb);
+	si->dma_tx.skb = NULL;
+
+	dev->stats.tx_packets++;
+	dev->stats.tx_bytes += sg_dma_len(&si->dma_tx.sg);
+
+	/* We need to ensure that the transmitter has finished. */
+	do
+		rmb();
+	while (Ser2UTSR1 & UTSR1_TBY);
+
+	/*
+	 * Ok, we've finished transmitting.  Now enable the receiver.
+	 * Sometimes we get a receive IRQ immediately after a transmit...
+	 */
+	Ser2UTSR0 = UTSR0_REB | UTSR0_RBB | UTSR0_RID;
+	Ser2UTCR3 = UTCR3_RIE | UTCR3_RXE | UTCR3_TXE;
+
+	sa1100_irda_check_speed(si);
+
+	/* I'm hungry! */
+	netif_wake_queue(dev);
+}
+
+static int sa1100_irda_sir_tx_start(struct sk_buff *skb, struct net_device *dev,
+	struct sa1100_irda *si)
+{
+	si->tx_buff.data = si->tx_buff.head;
+	si->tx_buff.len  = async_wrap_skb(skb, si->tx_buff.data,
+					  si->tx_buff.truesize);
+
+	si->dma_tx.skb = skb;
+	sg_set_buf(&si->dma_tx.sg, si->tx_buff.data, si->tx_buff.len);
+	if (dma_map_sg(si->dma_tx.dev, &si->dma_tx.sg, 1, DMA_TO_DEVICE) == 0) {
+		si->dma_tx.skb = NULL;
+		netif_wake_queue(dev);
+		dev->stats.tx_dropped++;
+		return NETDEV_TX_OK;
+	}
+
+	sa1100_irda_dma_start(&si->dma_tx, DMA_MEM_TO_DEV, sa1100_irda_sirtxdma_irq, dev);
+
+	/*
+	 * The mean turn-around time is enforced by XBOF padding,
+	 * so we don't have to do anything special here.
+	 */
+	Ser2UTCR3 = UTCR3_TXE;
+
+	return NETDEV_TX_OK;
+}
+
+static irqreturn_t sa1100_irda_sir_irq(struct net_device *dev, struct sa1100_irda *si)
+{
+	int status;
+
+	status = Ser2UTSR0;
+
+	/*
+	 * Deal with any receive errors first.  The bytes in error may be
+	 * the only bytes in the receive FIFO, so we do this first.
+	 */
+	while (status & UTSR0_EIF) {
+		int stat, data;
+
+		stat = Ser2UTSR1;
+		data = Ser2UTDR;
+
+		if (stat & (UTSR1_FRE | UTSR1_ROR)) {
+			dev->stats.rx_errors++;
+			if (stat & UTSR1_FRE)
+				dev->stats.rx_frame_errors++;
+			if (stat & UTSR1_ROR)
+				dev->stats.rx_fifo_errors++;
+		} else
+			async_unwrap_char(dev, &dev->stats, &si->rx_buff, data);
+
+		status = Ser2UTSR0;
+	}
+
+	/*
+	 * We must clear certain bits.
+	 */
+	Ser2UTSR0 = status & (UTSR0_RID | UTSR0_RBB | UTSR0_REB);
+
+	if (status & UTSR0_RFS) {
+		/*
+		 * There are at least 4 bytes in the FIFO.  Read 3 bytes
+		 * and leave the rest to the block below.
+		 */
+		async_unwrap_char(dev, &dev->stats, &si->rx_buff, Ser2UTDR);
+		async_unwrap_char(dev, &dev->stats, &si->rx_buff, Ser2UTDR);
+		async_unwrap_char(dev, &dev->stats, &si->rx_buff, Ser2UTDR);
+	}
+
+	if (status & (UTSR0_RFS | UTSR0_RID)) {
+		/*
+		 * Fifo contains more than 1 character.
+		 */
+		do {
+			async_unwrap_char(dev, &dev->stats, &si->rx_buff,
+					  Ser2UTDR);
+		} while (Ser2UTSR1 & UTSR1_RNE);
+
+	}
+
+	return IRQ_HANDLED;
+}
+
+/*
+ * FIR format support.
+ */
+static void sa1100_irda_firtxdma_irq(void *id)
+{
+	struct net_device *dev = id;
+	struct sa1100_irda *si = netdev_priv(dev);
+	struct sk_buff *skb;
+
+	/*
+	 * Wait for the transmission to complete.  Unfortunately,
+	 * the hardware doesn't give us an interrupt to indicate
+	 * "end of frame".
+	 */
+	do
+		rmb();
+	while (!(Ser2HSSR0 & HSSR0_TUR) || Ser2HSSR1 & HSSR1_TBY);
+
+	/*
+	 * Clear the transmit underrun bit.
+	 */
+	Ser2HSSR0 = HSSR0_TUR;
+
+	/*
+	 * Do we need to change speed?  Note that we're lazy
+	 * here - we don't free the old dma_rx.skb.  We don't need
+	 * to allocate a buffer either.
+	 */
+	sa1100_irda_check_speed(si);
+
+	/*
+	 * Start reception.  This disables the transmitter for
+	 * us.  This will be using the existing RX buffer.
+	 */
+	sa1100_irda_rx_dma_start(si);
+
+	/* Account and free the packet. */
+	skb = si->dma_tx.skb;
+	if (skb) {
+		dma_unmap_sg(si->dma_tx.dev, &si->dma_tx.sg, 1,
+			     DMA_TO_DEVICE);
+		dev->stats.tx_packets ++;
+		dev->stats.tx_bytes += skb->len;
+		dev_kfree_skb_irq(skb);
+		si->dma_tx.skb = NULL;
+	}
+
+	/*
+	 * Make sure that the TX queue is available for sending
+	 * (for retries).  TX has priority over RX at all times.
+	 */
+	netif_wake_queue(dev);
+}
+
+static int sa1100_irda_fir_tx_start(struct sk_buff *skb, struct net_device *dev,
+	struct sa1100_irda *si)
+{
+	int mtt = irda_get_mtt(skb);
+
+	si->dma_tx.skb = skb;
+	sg_set_buf(&si->dma_tx.sg, skb->data, skb->len);
+	if (dma_map_sg(si->dma_tx.dev, &si->dma_tx.sg, 1, DMA_TO_DEVICE) == 0) {
+		si->dma_tx.skb = NULL;
+		netif_wake_queue(dev);
+		dev->stats.tx_dropped++;
+		dev_kfree_skb(skb);
+		return NETDEV_TX_OK;
+	}
+
+	sa1100_irda_dma_start(&si->dma_tx, DMA_MEM_TO_DEV, sa1100_irda_firtxdma_irq, dev);
+
+	/*
+	 * If we have a mean turn-around time, impose the specified
+	 * specified delay.  We could shorten this by timing from
+	 * the point we received the packet.
+	 */
+	if (mtt)
+		udelay(mtt);
+
+	Ser2HSCR0 = HSCR0_HSSP | HSCR0_TXE;
+
+	return NETDEV_TX_OK;
+}
+
+static void sa1100_irda_fir_error(struct sa1100_irda *si, struct net_device *dev)
+{
+	struct sk_buff *skb = si->dma_rx.skb;
+	unsigned int len, stat, data;
+
+	if (!skb) {
+		printk(KERN_ERR "sa1100_ir: SKB is NULL!\n");
+		return;
+	}
+
+	/*
+	 * Get the current data position.
+	 */
+	len = sa1100_irda_dma_xferred(&si->dma_rx);
+	if (len > HPSIR_MAX_RXLEN)
+		len = HPSIR_MAX_RXLEN;
+	dma_unmap_sg(si->dma_rx.dev, &si->dma_rx.sg, 1, DMA_FROM_DEVICE);
+
+	do {
+		/*
+		 * Read Status, and then Data.
+		 */
+		stat = Ser2HSSR1;
+		rmb();
+		data = Ser2HSDR;
+
+		if (stat & (HSSR1_CRE | HSSR1_ROR)) {
+			dev->stats.rx_errors++;
+			if (stat & HSSR1_CRE)
+				dev->stats.rx_crc_errors++;
+			if (stat & HSSR1_ROR)
+				dev->stats.rx_frame_errors++;
+		} else
+			skb->data[len++] = data;
+
+		/*
+		 * If we hit the end of frame, there's
+		 * no point in continuing.
+		 */
+		if (stat & HSSR1_EOF)
+			break;
+	} while (Ser2HSSR0 & HSSR0_EIF);
+
+	if (stat & HSSR1_EOF) {
+		si->dma_rx.skb = NULL;
+
+		skb_put(skb, len);
+		skb->dev = dev;
+		skb_reset_mac_header(skb);
+		skb->protocol = htons(ETH_P_IRDA);
+		dev->stats.rx_packets++;
+		dev->stats.rx_bytes += len;
+
+		/*
+		 * Before we pass the buffer up, allocate a new one.
+		 */
+		sa1100_irda_rx_alloc(si);
+
+		netif_rx(skb);
+	} else {
+		/*
+		 * Remap the buffer - it was previously mapped, and we
+		 * hope that this succeeds.
+		 */
+		dma_map_sg(si->dma_rx.dev, &si->dma_rx.sg, 1, DMA_FROM_DEVICE);
+	}
+}
+
+/*
+ * We only have to handle RX events here; transmit events go via the TX
+ * DMA handler. We disable RX, process, and the restart RX.
+ */
+static irqreturn_t sa1100_irda_fir_irq(struct net_device *dev, struct sa1100_irda *si)
+{
+	/*
+	 * Stop RX DMA
+	 */
+	dmaengine_pause(si->dma_rx.chan);
+
+	/*
+	 * Framing error - we throw away the packet completely.
+	 * Clearing RXE flushes the error conditions and data
+	 * from the fifo.
+	 */
+	if (Ser2HSSR0 & (HSSR0_FRE | HSSR0_RAB)) {
+		dev->stats.rx_errors++;
+
+		if (Ser2HSSR0 & HSSR0_FRE)
+			dev->stats.rx_frame_errors++;
+
+		/*
+		 * Clear out the DMA...
+		 */
+		Ser2HSCR0 = HSCR0_HSSP;
+
+		/*
+		 * Clear selected status bits now, so we
+		 * don't miss them next time around.
+		 */
+		Ser2HSSR0 = HSSR0_FRE | HSSR0_RAB;
+	}
+
+	/*
+	 * Deal with any receive errors.  The any of the lowest
+	 * 8 bytes in the FIFO may contain an error.  We must read
+	 * them one by one.  The "error" could even be the end of
+	 * packet!
+	 */
+	if (Ser2HSSR0 & HSSR0_EIF)
+		sa1100_irda_fir_error(si, dev);
+
+	/*
+	 * No matter what happens, we must restart reception.
+	 */
+	sa1100_irda_rx_dma_start(si);
+
+	return IRQ_HANDLED;
 }
 
 /*
@@ -137,11 +552,12 @@
 	case 57600:	case 115200:
 		brd = 3686400 / (16 * speed) - 1;
 
-		/*
-		 * Stop the receive DMA.
-		 */
-		if (IS_FIR(si))
-			sa1100_stop_dma(si->rxdma);
+		/* Stop the receive DMA, and configure transmit. */
+		if (IS_FIR(si)) {
+			dmaengine_terminate_all(si->dma_rx.chan);
+			dmaengine_slave_config(si->dma_tx.chan,
+						&sa1100_irda_sir_tx);
+		}
 
 		local_irq_save(flags);
 
@@ -161,21 +577,27 @@
 			si->pdata->set_speed(si->dev, speed);
 
 		si->speed = speed;
+		si->tx_start = sa1100_irda_sir_tx_start;
+		si->irq = sa1100_irda_sir_irq;
 
 		local_irq_restore(flags);
 		ret = 0;
 		break;
 
 	case 4000000:
+		if (!IS_FIR(si))
+			dmaengine_slave_config(si->dma_tx.chan,
+						&sa1100_irda_fir_tx);
+
 		local_irq_save(flags);
 
-		si->hscr0 = 0;
-
 		Ser2HSSR0 = 0xff;
-		Ser2HSCR0 = si->hscr0 | HSCR0_HSSP;
+		Ser2HSCR0 = HSCR0_HSSP;
 		Ser2UTCR3 = 0;
 
 		si->speed = speed;
+		si->tx_start = sa1100_irda_fir_tx_start;
+		si->irq = sa1100_irda_fir_irq;
 
 		if (si->pdata->set_speed)
 			si->pdata->set_speed(si->dev, speed);
@@ -225,6 +647,86 @@
 	return ret;
 }
 
+static irqreturn_t sa1100_irda_irq(int irq, void *dev_id)
+{
+	struct net_device *dev = dev_id;
+	struct sa1100_irda *si = netdev_priv(dev);
+
+	return si->irq(dev, si);
+}
+
+static int sa1100_irda_hard_xmit(struct sk_buff *skb, struct net_device *dev)
+{
+	struct sa1100_irda *si = netdev_priv(dev);
+	int speed = irda_get_next_speed(skb);
+
+	/*
+	 * Does this packet contain a request to change the interface
+	 * speed?  If so, remember it until we complete the transmission
+	 * of this frame.
+	 */
+	if (speed != si->speed && speed != -1)
+		si->newspeed = speed;
+
+	/* If this is an empty frame, we can bypass a lot. */
+	if (skb->len == 0) {
+		sa1100_irda_check_speed(si);
+		dev_kfree_skb(skb);
+		return NETDEV_TX_OK;
+	}
+
+	netif_stop_queue(dev);
+
+	/* We must not already have a skb to transmit... */
+	BUG_ON(si->dma_tx.skb);
+
+	return si->tx_start(skb, dev, si);
+}
+
+static int
+sa1100_irda_ioctl(struct net_device *dev, struct ifreq *ifreq, int cmd)
+{
+	struct if_irda_req *rq = (struct if_irda_req *)ifreq;
+	struct sa1100_irda *si = netdev_priv(dev);
+	int ret = -EOPNOTSUPP;
+
+	switch (cmd) {
+	case SIOCSBANDWIDTH:
+		if (capable(CAP_NET_ADMIN)) {
+			/*
+			 * We are unable to set the speed if the
+			 * device is not running.
+			 */
+			if (si->open) {
+				ret = sa1100_irda_set_speed(si,
+						rq->ifr_baudrate);
+			} else {
+				printk("sa1100_irda_ioctl: SIOCSBANDWIDTH: !netif_running\n");
+				ret = 0;
+			}
+		}
+		break;
+
+	case SIOCSMEDIABUSY:
+		ret = -EPERM;
+		if (capable(CAP_NET_ADMIN)) {
+			irda_device_set_media_busy(dev, TRUE);
+			ret = 0;
+		}
+		break;
+
+	case SIOCGRECEIVING:
+		rq->ifr_receiving = IS_FIR(si) ? 0
+					: si->rx_buff.state != OUTSIDE_FRAME;
+		break;
+
+	default:
+		break;
+	}
+		
+	return ret;
+}
+
 static int sa1100_irda_startup(struct sa1100_irda *si)
 {
 	int ret;
@@ -277,8 +779,8 @@
 	/*
 	 * Stop all DMA activity.
 	 */
-	sa1100_stop_dma(si->rxdma);
-	sa1100_stop_dma(si->txdma);
+	dmaengine_terminate_all(si->dma_rx.chan);
+	dmaengine_terminate_all(si->dma_tx.chan);
 
 	/* Disable the port. */
 	Ser2UTCR3 = 0;
@@ -288,480 +790,6 @@
 		si->pdata->shutdown(si->dev);
 }
 
-#ifdef CONFIG_PM
-/*
- * Suspend the IrDA interface.
- */
-static int sa1100_irda_suspend(struct platform_device *pdev, pm_message_t state)
-{
-	struct net_device *dev = platform_get_drvdata(pdev);
-	struct sa1100_irda *si;
-
-	if (!dev)
-		return 0;
-
-	si = netdev_priv(dev);
-	if (si->open) {
-		/*
-		 * Stop the transmit queue
-		 */
-		netif_device_detach(dev);
-		disable_irq(dev->irq);
-		sa1100_irda_shutdown(si);
-		__sa1100_irda_set_power(si, 0);
-	}
-
-	return 0;
-}
-
-/*
- * Resume the IrDA interface.
- */
-static int sa1100_irda_resume(struct platform_device *pdev)
-{
-	struct net_device *dev = platform_get_drvdata(pdev);
-	struct sa1100_irda *si;
-
-	if (!dev)
-		return 0;
-
-	si = netdev_priv(dev);
-	if (si->open) {
-		/*
-		 * If we missed a speed change, initialise at the new speed
-		 * directly.  It is debatable whether this is actually
-		 * required, but in the interests of continuing from where
-		 * we left off it is desirable.  The converse argument is
-		 * that we should re-negotiate at 9600 baud again.
-		 */
-		if (si->newspeed) {
-			si->speed = si->newspeed;
-			si->newspeed = 0;
-		}
-
-		sa1100_irda_startup(si);
-		__sa1100_irda_set_power(si, si->power);
-		enable_irq(dev->irq);
-
-		/*
-		 * This automatically wakes up the queue
-		 */
-		netif_device_attach(dev);
-	}
-
-	return 0;
-}
-#else
-#define sa1100_irda_suspend	NULL
-#define sa1100_irda_resume	NULL
-#endif
-
-/*
- * HP-SIR format interrupt service routines.
- */
-static void sa1100_irda_hpsir_irq(struct net_device *dev)
-{
-	struct sa1100_irda *si = netdev_priv(dev);
-	int status;
-
-	status = Ser2UTSR0;
-
-	/*
-	 * Deal with any receive errors first.  The bytes in error may be
-	 * the only bytes in the receive FIFO, so we do this first.
-	 */
-	while (status & UTSR0_EIF) {
-		int stat, data;
-
-		stat = Ser2UTSR1;
-		data = Ser2UTDR;
-
-		if (stat & (UTSR1_FRE | UTSR1_ROR)) {
-			dev->stats.rx_errors++;
-			if (stat & UTSR1_FRE)
-				dev->stats.rx_frame_errors++;
-			if (stat & UTSR1_ROR)
-				dev->stats.rx_fifo_errors++;
-		} else
-			async_unwrap_char(dev, &dev->stats, &si->rx_buff, data);
-
-		status = Ser2UTSR0;
-	}
-
-	/*
-	 * We must clear certain bits.
-	 */
-	Ser2UTSR0 = status & (UTSR0_RID | UTSR0_RBB | UTSR0_REB);
-
-	if (status & UTSR0_RFS) {
-		/*
-		 * There are at least 4 bytes in the FIFO.  Read 3 bytes
-		 * and leave the rest to the block below.
-		 */
-		async_unwrap_char(dev, &dev->stats, &si->rx_buff, Ser2UTDR);
-		async_unwrap_char(dev, &dev->stats, &si->rx_buff, Ser2UTDR);
-		async_unwrap_char(dev, &dev->stats, &si->rx_buff, Ser2UTDR);
-	}
-
-	if (status & (UTSR0_RFS | UTSR0_RID)) {
-		/*
-		 * Fifo contains more than 1 character.
-		 */
-		do {
-			async_unwrap_char(dev, &dev->stats, &si->rx_buff,
-					  Ser2UTDR);
-		} while (Ser2UTSR1 & UTSR1_RNE);
-
-	}
-
-	if (status & UTSR0_TFS && si->tx_buff.len) {
-		/*
-		 * Transmitter FIFO is not full
-		 */
-		do {
-			Ser2UTDR = *si->tx_buff.data++;
-			si->tx_buff.len -= 1;
-		} while (Ser2UTSR1 & UTSR1_TNF && si->tx_buff.len);
-
-		if (si->tx_buff.len == 0) {
-			dev->stats.tx_packets++;
-			dev->stats.tx_bytes += si->tx_buff.data -
-					      si->tx_buff.head;
-
-			/*
-			 * We need to ensure that the transmitter has
-			 * finished.
-			 */
-			do
-				rmb();
-			while (Ser2UTSR1 & UTSR1_TBY);
-
-			/*
-			 * Ok, we've finished transmitting.  Now enable
-			 * the receiver.  Sometimes we get a receive IRQ
-			 * immediately after a transmit...
-			 */
-			Ser2UTSR0 = UTSR0_REB | UTSR0_RBB | UTSR0_RID;
-			Ser2UTCR3 = UTCR3_RIE | UTCR3_RXE | UTCR3_TXE;
-
-			if (si->newspeed) {
-				sa1100_irda_set_speed(si, si->newspeed);
-				si->newspeed = 0;
-			}
-
-			/* I'm hungry! */
-			netif_wake_queue(dev);
-		}
-	}
-}
-
-static void sa1100_irda_fir_error(struct sa1100_irda *si, struct net_device *dev)
-{
-	struct sk_buff *skb = si->rxskb;
-	dma_addr_t dma_addr;
-	unsigned int len, stat, data;
-
-	if (!skb) {
-		printk(KERN_ERR "sa1100_ir: SKB is NULL!\n");
-		return;
-	}
-
-	/*
-	 * Get the current data position.
-	 */
-	dma_addr = sa1100_get_dma_pos(si->rxdma);
-	len = dma_addr - si->rxbuf_dma;
-	if (len > HPSIR_MAX_RXLEN)
-		len = HPSIR_MAX_RXLEN;
-	dma_unmap_single(si->dev, si->rxbuf_dma, len, DMA_FROM_DEVICE);
-
-	do {
-		/*
-		 * Read Status, and then Data.
-		 */
-		stat = Ser2HSSR1;
-		rmb();
-		data = Ser2HSDR;
-
-		if (stat & (HSSR1_CRE | HSSR1_ROR)) {
-			dev->stats.rx_errors++;
-			if (stat & HSSR1_CRE)
-				dev->stats.rx_crc_errors++;
-			if (stat & HSSR1_ROR)
-				dev->stats.rx_frame_errors++;
-		} else
-			skb->data[len++] = data;
-
-		/*
-		 * If we hit the end of frame, there's
-		 * no point in continuing.
-		 */
-		if (stat & HSSR1_EOF)
-			break;
-	} while (Ser2HSSR0 & HSSR0_EIF);
-
-	if (stat & HSSR1_EOF) {
-		si->rxskb = NULL;
-
-		skb_put(skb, len);
-		skb->dev = dev;
-		skb_reset_mac_header(skb);
-		skb->protocol = htons(ETH_P_IRDA);
-		dev->stats.rx_packets++;
-		dev->stats.rx_bytes += len;
-
-		/*
-		 * Before we pass the buffer up, allocate a new one.
-		 */
-		sa1100_irda_rx_alloc(si);
-
-		netif_rx(skb);
-	} else {
-		/*
-		 * Remap the buffer.
-		 */
-		si->rxbuf_dma = dma_map_single(si->dev, si->rxskb->data,
-						HPSIR_MAX_RXLEN,
-						DMA_FROM_DEVICE);
-	}
-}
-
-/*
- * FIR format interrupt service routine.  We only have to
- * handle RX events; transmit events go via the TX DMA handler.
- *
- * No matter what, we disable RX, process, and the restart RX.
- */
-static void sa1100_irda_fir_irq(struct net_device *dev)
-{
-	struct sa1100_irda *si = netdev_priv(dev);
-
-	/*
-	 * Stop RX DMA
-	 */
-	sa1100_stop_dma(si->rxdma);
-
-	/*
-	 * Framing error - we throw away the packet completely.
-	 * Clearing RXE flushes the error conditions and data
-	 * from the fifo.
-	 */
-	if (Ser2HSSR0 & (HSSR0_FRE | HSSR0_RAB)) {
-		dev->stats.rx_errors++;
-
-		if (Ser2HSSR0 & HSSR0_FRE)
-			dev->stats.rx_frame_errors++;
-
-		/*
-		 * Clear out the DMA...
-		 */
-		Ser2HSCR0 = si->hscr0 | HSCR0_HSSP;
-
-		/*
-		 * Clear selected status bits now, so we
-		 * don't miss them next time around.
-		 */
-		Ser2HSSR0 = HSSR0_FRE | HSSR0_RAB;
-	}
-
-	/*
-	 * Deal with any receive errors.  The any of the lowest
-	 * 8 bytes in the FIFO may contain an error.  We must read
-	 * them one by one.  The "error" could even be the end of
-	 * packet!
-	 */
-	if (Ser2HSSR0 & HSSR0_EIF)
-		sa1100_irda_fir_error(si, dev);
-
-	/*
-	 * No matter what happens, we must restart reception.
-	 */
-	sa1100_irda_rx_dma_start(si);
-}
-
-static irqreturn_t sa1100_irda_irq(int irq, void *dev_id)
-{
-	struct net_device *dev = dev_id;
-	if (IS_FIR(((struct sa1100_irda *)netdev_priv(dev))))
-		sa1100_irda_fir_irq(dev);
-	else
-		sa1100_irda_hpsir_irq(dev);
-	return IRQ_HANDLED;
-}
-
-/*
- * TX DMA completion handler.
- */
-static void sa1100_irda_txdma_irq(void *id)
-{
-	struct net_device *dev = id;
-	struct sa1100_irda *si = netdev_priv(dev);
-	struct sk_buff *skb = si->txskb;
-
-	si->txskb = NULL;
-
-	/*
-	 * Wait for the transmission to complete.  Unfortunately,
-	 * the hardware doesn't give us an interrupt to indicate
-	 * "end of frame".
-	 */
-	do
-		rmb();
-	while (!(Ser2HSSR0 & HSSR0_TUR) || Ser2HSSR1 & HSSR1_TBY);
-
-	/*
-	 * Clear the transmit underrun bit.
-	 */
-	Ser2HSSR0 = HSSR0_TUR;
-
-	/*
-	 * Do we need to change speed?  Note that we're lazy
-	 * here - we don't free the old rxskb.  We don't need
-	 * to allocate a buffer either.
-	 */
-	if (si->newspeed) {
-		sa1100_irda_set_speed(si, si->newspeed);
-		si->newspeed = 0;
-	}
-
-	/*
-	 * Start reception.  This disables the transmitter for
-	 * us.  This will be using the existing RX buffer.
-	 */
-	sa1100_irda_rx_dma_start(si);
-
-	/*
-	 * Account and free the packet.
-	 */
-	if (skb) {
-		dma_unmap_single(si->dev, si->txbuf_dma, skb->len, DMA_TO_DEVICE);
-		dev->stats.tx_packets ++;
-		dev->stats.tx_bytes += skb->len;
-		dev_kfree_skb_irq(skb);
-	}
-
-	/*
-	 * Make sure that the TX queue is available for sending
-	 * (for retries).  TX has priority over RX at all times.
-	 */
-	netif_wake_queue(dev);
-}
-
-static int sa1100_irda_hard_xmit(struct sk_buff *skb, struct net_device *dev)
-{
-	struct sa1100_irda *si = netdev_priv(dev);
-	int speed = irda_get_next_speed(skb);
-
-	/*
-	 * Does this packet contain a request to change the interface
-	 * speed?  If so, remember it until we complete the transmission
-	 * of this frame.
-	 */
-	if (speed != si->speed && speed != -1)
-		si->newspeed = speed;
-
-	/*
-	 * If this is an empty frame, we can bypass a lot.
-	 */
-	if (skb->len == 0) {
-		if (si->newspeed) {
-			si->newspeed = 0;
-			sa1100_irda_set_speed(si, speed);
-		}
-		dev_kfree_skb(skb);
-		return NETDEV_TX_OK;
-	}
-
-	if (!IS_FIR(si)) {
-		netif_stop_queue(dev);
-
-		si->tx_buff.data = si->tx_buff.head;
-		si->tx_buff.len  = async_wrap_skb(skb, si->tx_buff.data,
-						  si->tx_buff.truesize);
-
-		/*
-		 * Set the transmit interrupt enable.  This will fire
-		 * off an interrupt immediately.  Note that we disable
-		 * the receiver so we won't get spurious characteres
-		 * received.
-		 */
-		Ser2UTCR3 = UTCR3_TIE | UTCR3_TXE;
-
-		dev_kfree_skb(skb);
-	} else {
-		int mtt = irda_get_mtt(skb);
-
-		/*
-		 * We must not be transmitting...
-		 */
-		BUG_ON(si->txskb);
-
-		netif_stop_queue(dev);
-
-		si->txskb = skb;
-		si->txbuf_dma = dma_map_single(si->dev, skb->data,
-					 skb->len, DMA_TO_DEVICE);
-
-		sa1100_start_dma(si->txdma, si->txbuf_dma, skb->len);
-
-		/*
-		 * If we have a mean turn-around time, impose the specified
-		 * specified delay.  We could shorten this by timing from
-		 * the point we received the packet.
-		 */
-		if (mtt)
-			udelay(mtt);
-
-		Ser2HSCR0 = si->hscr0 | HSCR0_HSSP | HSCR0_TXE;
-	}
-
-	return NETDEV_TX_OK;
-}
-
-static int
-sa1100_irda_ioctl(struct net_device *dev, struct ifreq *ifreq, int cmd)
-{
-	struct if_irda_req *rq = (struct if_irda_req *)ifreq;
-	struct sa1100_irda *si = netdev_priv(dev);
-	int ret = -EOPNOTSUPP;
-
-	switch (cmd) {
-	case SIOCSBANDWIDTH:
-		if (capable(CAP_NET_ADMIN)) {
-			/*
-			 * We are unable to set the speed if the
-			 * device is not running.
-			 */
-			if (si->open) {
-				ret = sa1100_irda_set_speed(si,
-						rq->ifr_baudrate);
-			} else {
-				printk("sa1100_irda_ioctl: SIOCSBANDWIDTH: !netif_running\n");
-				ret = 0;
-			}
-		}
-		break;
-
-	case SIOCSMEDIABUSY:
-		ret = -EPERM;
-		if (capable(CAP_NET_ADMIN)) {
-			irda_device_set_media_busy(dev, TRUE);
-			ret = 0;
-		}
-		break;
-
-	case SIOCGRECEIVING:
-		rq->ifr_receiving = IS_FIR(si) ? 0
-					: si->rx_buff.state != OUTSIDE_FRAME;
-		break;
-
-	default:
-		break;
-	}
-		
-	return ret;
-}
-
 static int sa1100_irda_start(struct net_device *dev)
 {
 	struct sa1100_irda *si = netdev_priv(dev);
@@ -769,26 +797,17 @@
 
 	si->speed = 9600;
 
-	err = request_irq(dev->irq, sa1100_irda_irq, 0, dev->name, dev);
-	if (err)
-		goto err_irq;
-
-	err = sa1100_request_dma(DMA_Ser2HSSPRd, "IrDA receive",
-				 NULL, NULL, &si->rxdma);
+	err = sa1100_irda_dma_request(si->dev, &si->dma_rx, "Ser2ICPRc",
+				&sa1100_irda_fir_rx);
 	if (err)
 		goto err_rx_dma;
 
-	err = sa1100_request_dma(DMA_Ser2HSSPWr, "IrDA transmit",
-				 sa1100_irda_txdma_irq, dev, &si->txdma);
+	err = sa1100_irda_dma_request(si->dev, &si->dma_tx, "Ser2ICPTr",
+				&sa1100_irda_sir_tx);
 	if (err)
 		goto err_tx_dma;
 
 	/*
-	 * The interrupt must remain disabled for now.
-	 */
-	disable_irq(dev->irq);
-
-	/*
 	 * Setup the serial port for the specified speed.
 	 */
 	err = sa1100_irda_startup(si);
@@ -803,44 +822,60 @@
 	if (!si->irlap)
 		goto err_irlap;
 
+	err = request_irq(dev->irq, sa1100_irda_irq, 0, dev->name, dev);
+	if (err)
+		goto err_irq;
+
 	/*
 	 * Now enable the interrupt and start the queue
 	 */
 	si->open = 1;
 	sa1100_set_power(si, power_level); /* low power mode */
-	enable_irq(dev->irq);
+
 	netif_start_queue(dev);
 	return 0;
 
+err_irq:
+	irlap_close(si->irlap);
 err_irlap:
 	si->open = 0;
 	sa1100_irda_shutdown(si);
 err_startup:
-	sa1100_free_dma(si->txdma);
+	dma_release_channel(si->dma_tx.chan);
 err_tx_dma:
-	sa1100_free_dma(si->rxdma);
+	dma_release_channel(si->dma_rx.chan);
 err_rx_dma:
-	free_irq(dev->irq, dev);
-err_irq:
 	return err;
 }
 
 static int sa1100_irda_stop(struct net_device *dev)
 {
 	struct sa1100_irda *si = netdev_priv(dev);
+	struct sk_buff *skb;
 
-	disable_irq(dev->irq);
+	netif_stop_queue(dev);
+
+	si->open = 0;
 	sa1100_irda_shutdown(si);
 
 	/*
-	 * If we have been doing DMA receive, make sure we
+	 * If we have been doing any DMA activity, make sure we
 	 * tidy that up cleanly.
 	 */
-	if (si->rxskb) {
-		dma_unmap_single(si->dev, si->rxbuf_dma, HPSIR_MAX_RXLEN,
-				 DMA_FROM_DEVICE);
-		dev_kfree_skb(si->rxskb);
-		si->rxskb = NULL;
+	skb = si->dma_rx.skb;
+	if (skb) {
+		dma_unmap_sg(si->dma_rx.dev, &si->dma_rx.sg, 1,
+			     DMA_FROM_DEVICE);
+		dev_kfree_skb(skb);
+		si->dma_rx.skb = NULL;
+	}
+
+	skb = si->dma_tx.skb;
+	if (skb) {
+		dma_unmap_sg(si->dma_tx.dev, &si->dma_tx.sg, 1,
+			     DMA_TO_DEVICE);
+		dev_kfree_skb(skb);
+		si->dma_tx.skb = NULL;
 	}
 
 	/* Stop IrLAP */
@@ -849,14 +884,11 @@
 		si->irlap = NULL;
 	}
 
-	netif_stop_queue(dev);
-	si->open = 0;
-
 	/*
 	 * Free resources
 	 */
-	sa1100_free_dma(si->txdma);
-	sa1100_free_dma(si->rxdma);
+	dma_release_channel(si->dma_tx.chan);
+	dma_release_channel(si->dma_rx.chan);
 	free_irq(dev->irq, dev);
 
 	sa1100_set_power(si, 0);
@@ -888,11 +920,15 @@
 	struct net_device *dev;
 	struct sa1100_irda *si;
 	unsigned int baudrate_mask;
-	int err;
+	int err, irq;
 
 	if (!pdev->dev.platform_data)
 		return -EINVAL;
 
+	irq = platform_get_irq(pdev, 0);
+	if (irq <= 0)
+		return irq < 0 ? irq : -ENXIO;
+
 	err = request_mem_region(__PREG(Ser2UTCR0), 0x24, "IrDA") ? 0 : -EBUSY;
 	if (err)
 		goto err_mem_1;
@@ -907,22 +943,27 @@
 	if (!dev)
 		goto err_mem_4;
 
+	SET_NETDEV_DEV(dev, &pdev->dev);
+
 	si = netdev_priv(dev);
 	si->dev = &pdev->dev;
 	si->pdata = pdev->dev.platform_data;
 
+	sg_init_table(&si->dma_rx.sg, 1);
+	sg_init_table(&si->dma_tx.sg, 1);
+
 	/*
 	 * Initialise the HP-SIR buffers
 	 */
 	err = sa1100_irda_init_iobuf(&si->rx_buff, 14384);
 	if (err)
 		goto err_mem_5;
-	err = sa1100_irda_init_iobuf(&si->tx_buff, 4000);
+	err = sa1100_irda_init_iobuf(&si->tx_buff, IRDA_SIR_MAX_FRAME);
 	if (err)
 		goto err_mem_5;
 
 	dev->netdev_ops	= &sa1100_irda_netdev_ops;
-	dev->irq	= IRQ_Ser2ICP;
+	dev->irq	= irq;
 
 	irda_init_max_qos_capabilies(&si->qos);
 
@@ -996,6 +1037,74 @@
 	return 0;
 }
 
+#ifdef CONFIG_PM
+/*
+ * Suspend the IrDA interface.
+ */
+static int sa1100_irda_suspend(struct platform_device *pdev, pm_message_t state)
+{
+	struct net_device *dev = platform_get_drvdata(pdev);
+	struct sa1100_irda *si;
+
+	if (!dev)
+		return 0;
+
+	si = netdev_priv(dev);
+	if (si->open) {
+		/*
+		 * Stop the transmit queue
+		 */
+		netif_device_detach(dev);
+		disable_irq(dev->irq);
+		sa1100_irda_shutdown(si);
+		__sa1100_irda_set_power(si, 0);
+	}
+
+	return 0;
+}
+
+/*
+ * Resume the IrDA interface.
+ */
+static int sa1100_irda_resume(struct platform_device *pdev)
+{
+	struct net_device *dev = platform_get_drvdata(pdev);
+	struct sa1100_irda *si;
+
+	if (!dev)
+		return 0;
+
+	si = netdev_priv(dev);
+	if (si->open) {
+		/*
+		 * If we missed a speed change, initialise at the new speed
+		 * directly.  It is debatable whether this is actually
+		 * required, but in the interests of continuing from where
+		 * we left off it is desirable.  The converse argument is
+		 * that we should re-negotiate at 9600 baud again.
+		 */
+		if (si->newspeed) {
+			si->speed = si->newspeed;
+			si->newspeed = 0;
+		}
+
+		sa1100_irda_startup(si);
+		__sa1100_irda_set_power(si, si->power);
+		enable_irq(dev->irq);
+
+		/*
+		 * This automatically wakes up the queue
+		 */
+		netif_device_attach(dev);
+	}
+
+	return 0;
+}
+#else
+#define sa1100_irda_suspend	NULL
+#define sa1100_irda_resume	NULL
+#endif
+
 static struct platform_driver sa1100ir_driver = {
 	.probe		= sa1100_irda_probe,
 	.remove		= sa1100_irda_remove,
diff --git a/include/linux/sa11x0-dma.h b/include/linux/sa11x0-dma.h
new file mode 100644
index 0000000..65839a5
--- /dev/null
+++ b/include/linux/sa11x0-dma.h
@@ -0,0 +1,24 @@
+/*
+ * SA11x0 DMA Engine support
+ *
+ * Copyright (C) 2012 Russell King
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+#ifndef __LINUX_SA11X0_DMA_H
+#define __LINUX_SA11X0_DMA_H
+
+struct dma_chan;
+
+#if defined(CONFIG_DMA_SA11X0) || defined(CONFIG_DMA_SA11X0_MODULE)
+bool sa11x0_dma_filter_fn(struct dma_chan *, void *);
+#else
+static inline bool sa11x0_dma_filter_fn(struct dma_chan *c, void *d)
+{
+	return false;
+}
+#endif
+
+#endif